From cdd6c482c9ff9c55475ee7392ec8f672eddb7be6 Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Mon, 21 Sep 2009 12:02:48 +0200 Subject: perf: Do the big rename: Performance Counters -> Performance Events Bye-bye Performance Counters, welcome Performance Events! In the past few months the perfcounters subsystem has grown out its initial role of counting hardware events, and has become (and is becoming) a much broader generic event enumeration, reporting, logging, monitoring, analysis facility. Naming its core object 'perf_counter' and naming the subsystem 'perfcounters' has become more and more of a misnomer. With pending code like hw-breakpoints support the 'counter' name is less and less appropriate. All in one, we've decided to rename the subsystem to 'performance events' and to propagate this rename through all fields, variables and API names. (in an ABI compatible fashion) The word 'event' is also a bit shorter than 'counter' - which makes it slightly more convenient to write/handle as well. Thanks goes to Stephane Eranian who first observed this misnomer and suggested a rename. User-space tooling and ABI compatibility is not affected - this patch should be function-invariant. (Also, defconfigs were not touched to keep the size down.) This patch has been generated via the following script: FILES=$(find * -type f | grep -vE 'oprofile|[^K]config') sed -i \ -e 's/PERF_EVENT_/PERF_RECORD_/g' \ -e 's/PERF_COUNTER/PERF_EVENT/g' \ -e 's/perf_counter/perf_event/g' \ -e 's/nb_counters/nb_events/g' \ -e 's/swcounter/swevent/g' \ -e 's/tpcounter_event/tp_event/g' \ $FILES for N in $(find . -name perf_counter.[ch]); do M=$(echo $N | sed 's/perf_counter/perf_event/g') mv $N $M done FILES=$(find . -name perf_event.*) sed -i \ -e 's/COUNTER_MASK/REG_MASK/g' \ -e 's/COUNTER/EVENT/g' \ -e 's/\/event_id/g' \ -e 's/counter/event/g' \ -e 's/Counter/Event/g' \ $FILES ... to keep it as correct as possible. This script can also be used by anyone who has pending perfcounters patches - it converts a Linux kernel tree over to the new naming. We tried to time this change to the point in time where the amount of pending patches is the smallest: the end of the merge window. Namespace clashes were fixed up in a preparatory patch - and some stylistic fallout will be fixed up in a subsequent patch. ( NOTE: 'counters' are still the proper terminology when we deal with hardware registers - and these sed scripts are a bit over-eager in renaming them. I've undone some of that, but in case there's something left where 'counter' would be better than 'event' we can undo that on an individual basis instead of touching an otherwise nicely automated patch. ) Suggested-by: Stephane Eranian Acked-by: Peter Zijlstra Acked-by: Paul Mackerras Reviewed-by: Arjan van de Ven Cc: Mike Galbraith Cc: Arnaldo Carvalho de Melo Cc: Frederic Weisbecker Cc: Steven Rostedt Cc: Benjamin Herrenschmidt Cc: David Howells Cc: Kyle McMartin Cc: Martin Schwidefsky Cc: "David S. Miller" Cc: Thomas Gleixner Cc: "H. Peter Anvin" Cc: LKML-Reference: Signed-off-by: Ingo Molnar diff --git a/arch/arm/include/asm/unistd.h b/arch/arm/include/asm/unistd.h index 9122c9e..89f7ead 100644 --- a/arch/arm/include/asm/unistd.h +++ b/arch/arm/include/asm/unistd.h @@ -390,7 +390,7 @@ #define __NR_preadv (__NR_SYSCALL_BASE+361) #define __NR_pwritev (__NR_SYSCALL_BASE+362) #define __NR_rt_tgsigqueueinfo (__NR_SYSCALL_BASE+363) -#define __NR_perf_counter_open (__NR_SYSCALL_BASE+364) +#define __NR_perf_event_open (__NR_SYSCALL_BASE+364) /* * The following SWIs are ARM private. diff --git a/arch/arm/kernel/calls.S b/arch/arm/kernel/calls.S index ecfa989..fafce1b 100644 --- a/arch/arm/kernel/calls.S +++ b/arch/arm/kernel/calls.S @@ -373,7 +373,7 @@ CALL(sys_preadv) CALL(sys_pwritev) CALL(sys_rt_tgsigqueueinfo) - CALL(sys_perf_counter_open) + CALL(sys_perf_event_open) #ifndef syscalls_counted .equ syscalls_padding, ((NR_syscalls + 3) & ~3) - NR_syscalls #define syscalls_counted diff --git a/arch/blackfin/include/asm/unistd.h b/arch/blackfin/include/asm/unistd.h index c8e7ee4..02b1529 100644 --- a/arch/blackfin/include/asm/unistd.h +++ b/arch/blackfin/include/asm/unistd.h @@ -381,7 +381,7 @@ #define __NR_preadv 366 #define __NR_pwritev 367 #define __NR_rt_tgsigqueueinfo 368 -#define __NR_perf_counter_open 369 +#define __NR_perf_event_open 369 #define __NR_syscall 370 #define NR_syscalls __NR_syscall diff --git a/arch/blackfin/mach-common/entry.S b/arch/blackfin/mach-common/entry.S index 01af24c..1e7cac2 100644 --- a/arch/blackfin/mach-common/entry.S +++ b/arch/blackfin/mach-common/entry.S @@ -1620,7 +1620,7 @@ ENTRY(_sys_call_table) .long _sys_preadv .long _sys_pwritev .long _sys_rt_tgsigqueueinfo - .long _sys_perf_counter_open + .long _sys_perf_event_open .rept NR_syscalls-(.-_sys_call_table)/4 .long _sys_ni_syscall diff --git a/arch/frv/Kconfig b/arch/frv/Kconfig index b86e19c..4b5830b 100644 --- a/arch/frv/Kconfig +++ b/arch/frv/Kconfig @@ -7,7 +7,7 @@ config FRV default y select HAVE_IDE select HAVE_ARCH_TRACEHOOK - select HAVE_PERF_COUNTERS + select HAVE_PERF_EVENTS config ZONE_DMA bool diff --git a/arch/frv/include/asm/perf_counter.h b/arch/frv/include/asm/perf_counter.h deleted file mode 100644 index ccf726e..0000000 --- a/arch/frv/include/asm/perf_counter.h +++ /dev/null @@ -1,17 +0,0 @@ -/* FRV performance counter support - * - * Copyright (C) 2009 Red Hat, Inc. All Rights Reserved. - * Written by David Howells (dhowells@redhat.com) - * - * This program is free software; you can redistribute it and/or - * modify it under the terms of the GNU General Public Licence - * as published by the Free Software Foundation; either version - * 2 of the Licence, or (at your option) any later version. - */ - -#ifndef _ASM_PERF_COUNTER_H -#define _ASM_PERF_COUNTER_H - -#define PERF_COUNTER_INDEX_OFFSET 0 - -#endif /* _ASM_PERF_COUNTER_H */ diff --git a/arch/frv/include/asm/perf_event.h b/arch/frv/include/asm/perf_event.h new file mode 100644 index 0000000..a69e015 --- /dev/null +++ b/arch/frv/include/asm/perf_event.h @@ -0,0 +1,17 @@ +/* FRV performance event support + * + * Copyright (C) 2009 Red Hat, Inc. All Rights Reserved. + * Written by David Howells (dhowells@redhat.com) + * + * This program is free software; you can redistribute it and/or + * modify it under the terms of the GNU General Public Licence + * as published by the Free Software Foundation; either version + * 2 of the Licence, or (at your option) any later version. + */ + +#ifndef _ASM_PERF_EVENT_H +#define _ASM_PERF_EVENT_H + +#define PERF_EVENT_INDEX_OFFSET 0 + +#endif /* _ASM_PERF_EVENT_H */ diff --git a/arch/frv/include/asm/unistd.h b/arch/frv/include/asm/unistd.h index 4a8fb42..be6ef0f 100644 --- a/arch/frv/include/asm/unistd.h +++ b/arch/frv/include/asm/unistd.h @@ -342,7 +342,7 @@ #define __NR_preadv 333 #define __NR_pwritev 334 #define __NR_rt_tgsigqueueinfo 335 -#define __NR_perf_counter_open 336 +#define __NR_perf_event_open 336 #ifdef __KERNEL__ diff --git a/arch/frv/kernel/entry.S b/arch/frv/kernel/entry.S index fde1e44..189397e 100644 --- a/arch/frv/kernel/entry.S +++ b/arch/frv/kernel/entry.S @@ -1525,6 +1525,6 @@ sys_call_table: .long sys_preadv .long sys_pwritev .long sys_rt_tgsigqueueinfo /* 335 */ - .long sys_perf_counter_open + .long sys_perf_event_open syscall_table_size = (. - sys_call_table) diff --git a/arch/frv/lib/Makefile b/arch/frv/lib/Makefile index 0a37721..f470975 100644 --- a/arch/frv/lib/Makefile +++ b/arch/frv/lib/Makefile @@ -5,4 +5,4 @@ lib-y := \ __ashldi3.o __lshrdi3.o __muldi3.o __ashrdi3.o __negdi2.o __ucmpdi2.o \ checksum.o memcpy.o memset.o atomic-ops.o atomic64-ops.o \ - outsl_ns.o outsl_sw.o insl_ns.o insl_sw.o cache.o perf_counter.o + outsl_ns.o outsl_sw.o insl_ns.o insl_sw.o cache.o perf_event.o diff --git a/arch/frv/lib/perf_counter.c b/arch/frv/lib/perf_counter.c deleted file mode 100644 index 2000fee..0000000 --- a/arch/frv/lib/perf_counter.c +++ /dev/null @@ -1,19 +0,0 @@ -/* Performance counter handling - * - * Copyright (C) 2009 Red Hat, Inc. All Rights Reserved. - * Written by David Howells (dhowells@redhat.com) - * - * This program is free software; you can redistribute it and/or - * modify it under the terms of the GNU General Public Licence - * as published by the Free Software Foundation; either version - * 2 of the Licence, or (at your option) any later version. - */ - -#include - -/* - * mark the performance counter as pending - */ -void set_perf_counter_pending(void) -{ -} diff --git a/arch/frv/lib/perf_event.c b/arch/frv/lib/perf_event.c new file mode 100644 index 0000000..9ac5acf --- /dev/null +++ b/arch/frv/lib/perf_event.c @@ -0,0 +1,19 @@ +/* Performance event handling + * + * Copyright (C) 2009 Red Hat, Inc. All Rights Reserved. + * Written by David Howells (dhowells@redhat.com) + * + * This program is free software; you can redistribute it and/or + * modify it under the terms of the GNU General Public Licence + * as published by the Free Software Foundation; either version + * 2 of the Licence, or (at your option) any later version. + */ + +#include + +/* + * mark the performance event as pending + */ +void set_perf_event_pending(void) +{ +} diff --git a/arch/m68k/include/asm/unistd.h b/arch/m68k/include/asm/unistd.h index 946d869..48b87f5 100644 --- a/arch/m68k/include/asm/unistd.h +++ b/arch/m68k/include/asm/unistd.h @@ -335,7 +335,7 @@ #define __NR_preadv 329 #define __NR_pwritev 330 #define __NR_rt_tgsigqueueinfo 331 -#define __NR_perf_counter_open 332 +#define __NR_perf_event_open 332 #ifdef __KERNEL__ diff --git a/arch/m68k/kernel/entry.S b/arch/m68k/kernel/entry.S index 922f52e..c5b3363 100644 --- a/arch/m68k/kernel/entry.S +++ b/arch/m68k/kernel/entry.S @@ -756,5 +756,5 @@ sys_call_table: .long sys_preadv .long sys_pwritev /* 330 */ .long sys_rt_tgsigqueueinfo - .long sys_perf_counter_open + .long sys_perf_event_open diff --git a/arch/m68knommu/kernel/syscalltable.S b/arch/m68knommu/kernel/syscalltable.S index 0ae123e..23535cc 100644 --- a/arch/m68knommu/kernel/syscalltable.S +++ b/arch/m68knommu/kernel/syscalltable.S @@ -350,7 +350,7 @@ ENTRY(sys_call_table) .long sys_preadv .long sys_pwritev /* 330 */ .long sys_rt_tgsigqueueinfo - .long sys_perf_counter_open + .long sys_perf_event_open .rept NR_syscalls-(.-sys_call_table)/4 .long sys_ni_syscall diff --git a/arch/microblaze/include/asm/unistd.h b/arch/microblaze/include/asm/unistd.h index 0b85232..cb05a07 100644 --- a/arch/microblaze/include/asm/unistd.h +++ b/arch/microblaze/include/asm/unistd.h @@ -381,7 +381,7 @@ #define __NR_preadv 363 /* new */ #define __NR_pwritev 364 /* new */ #define __NR_rt_tgsigqueueinfo 365 /* new */ -#define __NR_perf_counter_open 366 /* new */ +#define __NR_perf_event_open 366 /* new */ #define __NR_syscalls 367 diff --git a/arch/microblaze/kernel/syscall_table.S b/arch/microblaze/kernel/syscall_table.S index 4572160..ecec191 100644 --- a/arch/microblaze/kernel/syscall_table.S +++ b/arch/microblaze/kernel/syscall_table.S @@ -370,4 +370,4 @@ ENTRY(sys_call_table) .long sys_ni_syscall .long sys_ni_syscall .long sys_rt_tgsigqueueinfo /* 365 */ - .long sys_perf_counter_open + .long sys_perf_event_open diff --git a/arch/mips/include/asm/unistd.h b/arch/mips/include/asm/unistd.h index e753a77..8c9dfa9 100644 --- a/arch/mips/include/asm/unistd.h +++ b/arch/mips/include/asm/unistd.h @@ -353,7 +353,7 @@ #define __NR_preadv (__NR_Linux + 330) #define __NR_pwritev (__NR_Linux + 331) #define __NR_rt_tgsigqueueinfo (__NR_Linux + 332) -#define __NR_perf_counter_open (__NR_Linux + 333) +#define __NR_perf_event_open (__NR_Linux + 333) #define __NR_accept4 (__NR_Linux + 334) /* @@ -664,7 +664,7 @@ #define __NR_preadv (__NR_Linux + 289) #define __NR_pwritev (__NR_Linux + 290) #define __NR_rt_tgsigqueueinfo (__NR_Linux + 291) -#define __NR_perf_counter_open (__NR_Linux + 292) +#define __NR_perf_event_open (__NR_Linux + 292) #define __NR_accept4 (__NR_Linux + 293) /* @@ -979,7 +979,7 @@ #define __NR_preadv (__NR_Linux + 293) #define __NR_pwritev (__NR_Linux + 294) #define __NR_rt_tgsigqueueinfo (__NR_Linux + 295) -#define __NR_perf_counter_open (__NR_Linux + 296) +#define __NR_perf_event_open (__NR_Linux + 296) #define __NR_accept4 (__NR_Linux + 297) /* diff --git a/arch/mips/kernel/scall32-o32.S b/arch/mips/kernel/scall32-o32.S index 7c2de4f..fd2a9bb 100644 --- a/arch/mips/kernel/scall32-o32.S +++ b/arch/mips/kernel/scall32-o32.S @@ -581,7 +581,7 @@ einval: li v0, -ENOSYS sys sys_preadv 6 /* 4330 */ sys sys_pwritev 6 sys sys_rt_tgsigqueueinfo 4 - sys sys_perf_counter_open 5 + sys sys_perf_event_open 5 sys sys_accept4 4 .endm diff --git a/arch/mips/kernel/scall64-64.S b/arch/mips/kernel/scall64-64.S index b97b993..18bf7f3 100644 --- a/arch/mips/kernel/scall64-64.S +++ b/arch/mips/kernel/scall64-64.S @@ -418,6 +418,6 @@ sys_call_table: PTR sys_preadv PTR sys_pwritev /* 5390 */ PTR sys_rt_tgsigqueueinfo - PTR sys_perf_counter_open + PTR sys_perf_event_open PTR sys_accept4 .size sys_call_table,.-sys_call_table diff --git a/arch/mips/kernel/scall64-n32.S b/arch/mips/kernel/scall64-n32.S index 1a6ae12..6ebc079 100644 --- a/arch/mips/kernel/scall64-n32.S +++ b/arch/mips/kernel/scall64-n32.S @@ -416,6 +416,6 @@ EXPORT(sysn32_call_table) PTR sys_preadv PTR sys_pwritev PTR compat_sys_rt_tgsigqueueinfo /* 5295 */ - PTR sys_perf_counter_open + PTR sys_perf_event_open PTR sys_accept4 .size sysn32_call_table,.-sysn32_call_table diff --git a/arch/mips/kernel/scall64-o32.S b/arch/mips/kernel/scall64-o32.S index cd31087..9bbf977 100644 --- a/arch/mips/kernel/scall64-o32.S +++ b/arch/mips/kernel/scall64-o32.S @@ -536,6 +536,6 @@ sys_call_table: PTR compat_sys_preadv /* 4330 */ PTR compat_sys_pwritev PTR compat_sys_rt_tgsigqueueinfo - PTR sys_perf_counter_open + PTR sys_perf_event_open PTR sys_accept4 .size sys_call_table,.-sys_call_table diff --git a/arch/mn10300/include/asm/unistd.h b/arch/mn10300/include/asm/unistd.h index fad6861..2a98393 100644 --- a/arch/mn10300/include/asm/unistd.h +++ b/arch/mn10300/include/asm/unistd.h @@ -347,7 +347,7 @@ #define __NR_preadv 334 #define __NR_pwritev 335 #define __NR_rt_tgsigqueueinfo 336 -#define __NR_perf_counter_open 337 +#define __NR_perf_event_open 337 #ifdef __KERNEL__ diff --git a/arch/mn10300/kernel/entry.S b/arch/mn10300/kernel/entry.S index e0d2563..a94e7ea 100644 --- a/arch/mn10300/kernel/entry.S +++ b/arch/mn10300/kernel/entry.S @@ -723,7 +723,7 @@ ENTRY(sys_call_table) .long sys_preadv .long sys_pwritev /* 335 */ .long sys_rt_tgsigqueueinfo - .long sys_perf_counter_open + .long sys_perf_event_open nr_syscalls=(.-sys_call_table)/4 diff --git a/arch/parisc/Kconfig b/arch/parisc/Kconfig index 06f8d5b..f388dc6 100644 --- a/arch/parisc/Kconfig +++ b/arch/parisc/Kconfig @@ -16,7 +16,7 @@ config PARISC select RTC_DRV_GENERIC select INIT_ALL_POSSIBLE select BUG - select HAVE_PERF_COUNTERS + select HAVE_PERF_EVENTS select GENERIC_ATOMIC64 if !64BIT help The PA-RISC microprocessor is designed by Hewlett-Packard and used diff --git a/arch/parisc/include/asm/perf_counter.h b/arch/parisc/include/asm/perf_counter.h deleted file mode 100644 index dc9e829..0000000 --- a/arch/parisc/include/asm/perf_counter.h +++ /dev/null @@ -1,7 +0,0 @@ -#ifndef __ASM_PARISC_PERF_COUNTER_H -#define __ASM_PARISC_PERF_COUNTER_H - -/* parisc only supports software counters through this interface. */ -static inline void set_perf_counter_pending(void) { } - -#endif /* __ASM_PARISC_PERF_COUNTER_H */ diff --git a/arch/parisc/include/asm/perf_event.h b/arch/parisc/include/asm/perf_event.h new file mode 100644 index 0000000..cc146427 --- /dev/null +++ b/arch/parisc/include/asm/perf_event.h @@ -0,0 +1,7 @@ +#ifndef __ASM_PARISC_PERF_EVENT_H +#define __ASM_PARISC_PERF_EVENT_H + +/* parisc only supports software events through this interface. */ +static inline void set_perf_event_pending(void) { } + +#endif /* __ASM_PARISC_PERF_EVENT_H */ diff --git a/arch/parisc/include/asm/unistd.h b/arch/parisc/include/asm/unistd.h index f3d3b8b..cda1583 100644 --- a/arch/parisc/include/asm/unistd.h +++ b/arch/parisc/include/asm/unistd.h @@ -810,9 +810,9 @@ #define __NR_preadv (__NR_Linux + 315) #define __NR_pwritev (__NR_Linux + 316) #define __NR_rt_tgsigqueueinfo (__NR_Linux + 317) -#define __NR_perf_counter_open (__NR_Linux + 318) +#define __NR_perf_event_open (__NR_Linux + 318) -#define __NR_Linux_syscalls (__NR_perf_counter_open + 1) +#define __NR_Linux_syscalls (__NR_perf_event_open + 1) #define __IGNORE_select /* newselect */ diff --git a/arch/parisc/kernel/syscall_table.S b/arch/parisc/kernel/syscall_table.S index cf145eb..843f423 100644 --- a/arch/parisc/kernel/syscall_table.S +++ b/arch/parisc/kernel/syscall_table.S @@ -416,7 +416,7 @@ ENTRY_COMP(preadv) /* 315 */ ENTRY_COMP(pwritev) ENTRY_COMP(rt_tgsigqueueinfo) - ENTRY_SAME(perf_counter_open) + ENTRY_SAME(perf_event_open) /* Nothing yet */ diff --git a/arch/powerpc/Kconfig b/arch/powerpc/Kconfig index 8250902..4fd4790 100644 --- a/arch/powerpc/Kconfig +++ b/arch/powerpc/Kconfig @@ -129,7 +129,7 @@ config PPC select HAVE_OPROFILE select HAVE_SYSCALL_WRAPPERS if PPC64 select GENERIC_ATOMIC64 if PPC32 - select HAVE_PERF_COUNTERS + select HAVE_PERF_EVENTS config EARLY_PRINTK bool diff --git a/arch/powerpc/include/asm/hw_irq.h b/arch/powerpc/include/asm/hw_irq.h index e73d554..abbc2aa 100644 --- a/arch/powerpc/include/asm/hw_irq.h +++ b/arch/powerpc/include/asm/hw_irq.h @@ -135,43 +135,43 @@ static inline int irqs_disabled_flags(unsigned long flags) */ struct irq_chip; -#ifdef CONFIG_PERF_COUNTERS +#ifdef CONFIG_PERF_EVENTS #ifdef CONFIG_PPC64 -static inline unsigned long test_perf_counter_pending(void) +static inline unsigned long test_perf_event_pending(void) { unsigned long x; asm volatile("lbz %0,%1(13)" : "=r" (x) - : "i" (offsetof(struct paca_struct, perf_counter_pending))); + : "i" (offsetof(struct paca_struct, perf_event_pending))); return x; } -static inline void set_perf_counter_pending(void) +static inline void set_perf_event_pending(void) { asm volatile("stb %0,%1(13)" : : "r" (1), - "i" (offsetof(struct paca_struct, perf_counter_pending))); + "i" (offsetof(struct paca_struct, perf_event_pending))); } -static inline void clear_perf_counter_pending(void) +static inline void clear_perf_event_pending(void) { asm volatile("stb %0,%1(13)" : : "r" (0), - "i" (offsetof(struct paca_struct, perf_counter_pending))); + "i" (offsetof(struct paca_struct, perf_event_pending))); } #endif /* CONFIG_PPC64 */ -#else /* CONFIG_PERF_COUNTERS */ +#else /* CONFIG_PERF_EVENTS */ -static inline unsigned long test_perf_counter_pending(void) +static inline unsigned long test_perf_event_pending(void) { return 0; } -static inline void clear_perf_counter_pending(void) {} -#endif /* CONFIG_PERF_COUNTERS */ +static inline void clear_perf_event_pending(void) {} +#endif /* CONFIG_PERF_EVENTS */ #endif /* __KERNEL__ */ #endif /* _ASM_POWERPC_HW_IRQ_H */ diff --git a/arch/powerpc/include/asm/paca.h b/arch/powerpc/include/asm/paca.h index b634456..154f405 100644 --- a/arch/powerpc/include/asm/paca.h +++ b/arch/powerpc/include/asm/paca.h @@ -122,7 +122,7 @@ struct paca_struct { u8 soft_enabled; /* irq soft-enable flag */ u8 hard_enabled; /* set if irqs are enabled in MSR */ u8 io_sync; /* writel() needs spin_unlock sync */ - u8 perf_counter_pending; /* PM interrupt while soft-disabled */ + u8 perf_event_pending; /* PM interrupt while soft-disabled */ /* Stuff for accurate time accounting */ u64 user_time; /* accumulated usermode TB ticks */ diff --git a/arch/powerpc/include/asm/perf_counter.h b/arch/powerpc/include/asm/perf_counter.h deleted file mode 100644 index 0ea0639..0000000 --- a/arch/powerpc/include/asm/perf_counter.h +++ /dev/null @@ -1,110 +0,0 @@ -/* - * Performance counter support - PowerPC-specific definitions. - * - * Copyright 2008-2009 Paul Mackerras, IBM Corporation. - * - * This program is free software; you can redistribute it and/or - * modify it under the terms of the GNU General Public License - * as published by the Free Software Foundation; either version - * 2 of the License, or (at your option) any later version. - */ -#include - -#include - -#define MAX_HWCOUNTERS 8 -#define MAX_EVENT_ALTERNATIVES 8 -#define MAX_LIMITED_HWCOUNTERS 2 - -/* - * This struct provides the constants and functions needed to - * describe the PMU on a particular POWER-family CPU. - */ -struct power_pmu { - const char *name; - int n_counter; - int max_alternatives; - unsigned long add_fields; - unsigned long test_adder; - int (*compute_mmcr)(u64 events[], int n_ev, - unsigned int hwc[], unsigned long mmcr[]); - int (*get_constraint)(u64 event, unsigned long *mskp, - unsigned long *valp); - int (*get_alternatives)(u64 event, unsigned int flags, - u64 alt[]); - void (*disable_pmc)(unsigned int pmc, unsigned long mmcr[]); - int (*limited_pmc_event)(u64 event); - u32 flags; - int n_generic; - int *generic_events; - int (*cache_events)[PERF_COUNT_HW_CACHE_MAX] - [PERF_COUNT_HW_CACHE_OP_MAX] - [PERF_COUNT_HW_CACHE_RESULT_MAX]; -}; - -/* - * Values for power_pmu.flags - */ -#define PPMU_LIMITED_PMC5_6 1 /* PMC5/6 have limited function */ -#define PPMU_ALT_SIPR 2 /* uses alternate posn for SIPR/HV */ - -/* - * Values for flags to get_alternatives() - */ -#define PPMU_LIMITED_PMC_OK 1 /* can put this on a limited PMC */ -#define PPMU_LIMITED_PMC_REQD 2 /* have to put this on a limited PMC */ -#define PPMU_ONLY_COUNT_RUN 4 /* only counting in run state */ - -extern int register_power_pmu(struct power_pmu *); - -struct pt_regs; -extern unsigned long perf_misc_flags(struct pt_regs *regs); -extern unsigned long perf_instruction_pointer(struct pt_regs *regs); - -#define PERF_COUNTER_INDEX_OFFSET 1 - -/* - * Only override the default definitions in include/linux/perf_counter.h - * if we have hardware PMU support. - */ -#ifdef CONFIG_PPC_PERF_CTRS -#define perf_misc_flags(regs) perf_misc_flags(regs) -#endif - -/* - * The power_pmu.get_constraint function returns a 32/64-bit value and - * a 32/64-bit mask that express the constraints between this event and - * other events. - * - * The value and mask are divided up into (non-overlapping) bitfields - * of three different types: - * - * Select field: this expresses the constraint that some set of bits - * in MMCR* needs to be set to a specific value for this event. For a - * select field, the mask contains 1s in every bit of the field, and - * the value contains a unique value for each possible setting of the - * MMCR* bits. The constraint checking code will ensure that two events - * that set the same field in their masks have the same value in their - * value dwords. - * - * Add field: this expresses the constraint that there can be at most - * N events in a particular class. A field of k bits can be used for - * N <= 2^(k-1) - 1. The mask has the most significant bit of the field - * set (and the other bits 0), and the value has only the least significant - * bit of the field set. In addition, the 'add_fields' and 'test_adder' - * in the struct power_pmu for this processor come into play. The - * add_fields value contains 1 in the LSB of the field, and the - * test_adder contains 2^(k-1) - 1 - N in the field. - * - * NAND field: this expresses the constraint that you may not have events - * in all of a set of classes. (For example, on PPC970, you can't select - * events from the FPU, ISU and IDU simultaneously, although any two are - * possible.) For N classes, the field is N+1 bits wide, and each class - * is assigned one bit from the least-significant N bits. The mask has - * only the most-significant bit set, and the value has only the bit - * for the event's class set. The test_adder has the least significant - * bit set in the field. - * - * If an event is not subject to the constraint expressed by a particular - * field, then it will have 0 in both the mask and value for that field. - */ diff --git a/arch/powerpc/include/asm/perf_event.h b/arch/powerpc/include/asm/perf_event.h new file mode 100644 index 0000000..2499aaa --- /dev/null +++ b/arch/powerpc/include/asm/perf_event.h @@ -0,0 +1,110 @@ +/* + * Performance event support - PowerPC-specific definitions. + * + * Copyright 2008-2009 Paul Mackerras, IBM Corporation. + * + * This program is free software; you can redistribute it and/or + * modify it under the terms of the GNU General Public License + * as published by the Free Software Foundation; either version + * 2 of the License, or (at your option) any later version. + */ +#include + +#include + +#define MAX_HWEVENTS 8 +#define MAX_EVENT_ALTERNATIVES 8 +#define MAX_LIMITED_HWEVENTS 2 + +/* + * This struct provides the constants and functions needed to + * describe the PMU on a particular POWER-family CPU. + */ +struct power_pmu { + const char *name; + int n_event; + int max_alternatives; + unsigned long add_fields; + unsigned long test_adder; + int (*compute_mmcr)(u64 events[], int n_ev, + unsigned int hwc[], unsigned long mmcr[]); + int (*get_constraint)(u64 event_id, unsigned long *mskp, + unsigned long *valp); + int (*get_alternatives)(u64 event_id, unsigned int flags, + u64 alt[]); + void (*disable_pmc)(unsigned int pmc, unsigned long mmcr[]); + int (*limited_pmc_event)(u64 event_id); + u32 flags; + int n_generic; + int *generic_events; + int (*cache_events)[PERF_COUNT_HW_CACHE_MAX] + [PERF_COUNT_HW_CACHE_OP_MAX] + [PERF_COUNT_HW_CACHE_RESULT_MAX]; +}; + +/* + * Values for power_pmu.flags + */ +#define PPMU_LIMITED_PMC5_6 1 /* PMC5/6 have limited function */ +#define PPMU_ALT_SIPR 2 /* uses alternate posn for SIPR/HV */ + +/* + * Values for flags to get_alternatives() + */ +#define PPMU_LIMITED_PMC_OK 1 /* can put this on a limited PMC */ +#define PPMU_LIMITED_PMC_REQD 2 /* have to put this on a limited PMC */ +#define PPMU_ONLY_COUNT_RUN 4 /* only counting in run state */ + +extern int register_power_pmu(struct power_pmu *); + +struct pt_regs; +extern unsigned long perf_misc_flags(struct pt_regs *regs); +extern unsigned long perf_instruction_pointer(struct pt_regs *regs); + +#define PERF_EVENT_INDEX_OFFSET 1 + +/* + * Only override the default definitions in include/linux/perf_event.h + * if we have hardware PMU support. + */ +#ifdef CONFIG_PPC_PERF_CTRS +#define perf_misc_flags(regs) perf_misc_flags(regs) +#endif + +/* + * The power_pmu.get_constraint function returns a 32/64-bit value and + * a 32/64-bit mask that express the constraints between this event_id and + * other events. + * + * The value and mask are divided up into (non-overlapping) bitfields + * of three different types: + * + * Select field: this expresses the constraint that some set of bits + * in MMCR* needs to be set to a specific value for this event_id. For a + * select field, the mask contains 1s in every bit of the field, and + * the value contains a unique value for each possible setting of the + * MMCR* bits. The constraint checking code will ensure that two events + * that set the same field in their masks have the same value in their + * value dwords. + * + * Add field: this expresses the constraint that there can be at most + * N events in a particular class. A field of k bits can be used for + * N <= 2^(k-1) - 1. The mask has the most significant bit of the field + * set (and the other bits 0), and the value has only the least significant + * bit of the field set. In addition, the 'add_fields' and 'test_adder' + * in the struct power_pmu for this processor come into play. The + * add_fields value contains 1 in the LSB of the field, and the + * test_adder contains 2^(k-1) - 1 - N in the field. + * + * NAND field: this expresses the constraint that you may not have events + * in all of a set of classes. (For example, on PPC970, you can't select + * events from the FPU, ISU and IDU simultaneously, although any two are + * possible.) For N classes, the field is N+1 bits wide, and each class + * is assigned one bit from the least-significant N bits. The mask has + * only the most-significant bit set, and the value has only the bit + * for the event_id's class set. The test_adder has the least significant + * bit set in the field. + * + * If an event_id is not subject to the constraint expressed by a particular + * field, then it will have 0 in both the mask and value for that field. + */ diff --git a/arch/powerpc/include/asm/systbl.h b/arch/powerpc/include/asm/systbl.h index ed24bd9..c7d671a 100644 --- a/arch/powerpc/include/asm/systbl.h +++ b/arch/powerpc/include/asm/systbl.h @@ -322,7 +322,7 @@ SYSCALL_SPU(epoll_create1) SYSCALL_SPU(dup3) SYSCALL_SPU(pipe2) SYSCALL(inotify_init1) -SYSCALL_SPU(perf_counter_open) +SYSCALL_SPU(perf_event_open) COMPAT_SYS_SPU(preadv) COMPAT_SYS_SPU(pwritev) COMPAT_SYS(rt_tgsigqueueinfo) diff --git a/arch/powerpc/include/asm/unistd.h b/arch/powerpc/include/asm/unistd.h index cef080b..f6ca761 100644 --- a/arch/powerpc/include/asm/unistd.h +++ b/arch/powerpc/include/asm/unistd.h @@ -341,7 +341,7 @@ #define __NR_dup3 316 #define __NR_pipe2 317 #define __NR_inotify_init1 318 -#define __NR_perf_counter_open 319 +#define __NR_perf_event_open 319 #define __NR_preadv 320 #define __NR_pwritev 321 #define __NR_rt_tgsigqueueinfo 322 diff --git a/arch/powerpc/kernel/Makefile b/arch/powerpc/kernel/Makefile index 569f79c..b23664a 100644 --- a/arch/powerpc/kernel/Makefile +++ b/arch/powerpc/kernel/Makefile @@ -97,7 +97,7 @@ obj64-$(CONFIG_AUDIT) += compat_audit.o obj-$(CONFIG_DYNAMIC_FTRACE) += ftrace.o obj-$(CONFIG_FUNCTION_GRAPH_TRACER) += ftrace.o -obj-$(CONFIG_PPC_PERF_CTRS) += perf_counter.o perf_callchain.o +obj-$(CONFIG_PPC_PERF_CTRS) += perf_event.o perf_callchain.o obj64-$(CONFIG_PPC_PERF_CTRS) += power4-pmu.o ppc970-pmu.o power5-pmu.o \ power5+-pmu.o power6-pmu.o power7-pmu.o obj32-$(CONFIG_PPC_PERF_CTRS) += mpc7450-pmu.o diff --git a/arch/powerpc/kernel/asm-offsets.c b/arch/powerpc/kernel/asm-offsets.c index f0df285..0812b0f 100644 --- a/arch/powerpc/kernel/asm-offsets.c +++ b/arch/powerpc/kernel/asm-offsets.c @@ -133,7 +133,7 @@ int main(void) DEFINE(PACAKMSR, offsetof(struct paca_struct, kernel_msr)); DEFINE(PACASOFTIRQEN, offsetof(struct paca_struct, soft_enabled)); DEFINE(PACAHARDIRQEN, offsetof(struct paca_struct, hard_enabled)); - DEFINE(PACAPERFPEND, offsetof(struct paca_struct, perf_counter_pending)); + DEFINE(PACAPERFPEND, offsetof(struct paca_struct, perf_event_pending)); DEFINE(PACACONTEXTID, offsetof(struct paca_struct, context.id)); #ifdef CONFIG_PPC_MM_SLICES DEFINE(PACALOWSLICESPSIZE, offsetof(struct paca_struct, diff --git a/arch/powerpc/kernel/entry_64.S b/arch/powerpc/kernel/entry_64.S index 66bcda3..900e0ee 100644 --- a/arch/powerpc/kernel/entry_64.S +++ b/arch/powerpc/kernel/entry_64.S @@ -556,14 +556,14 @@ ALT_FW_FTR_SECTION_END_IFCLR(FW_FEATURE_ISERIES) 2: TRACE_AND_RESTORE_IRQ(r5); -#ifdef CONFIG_PERF_COUNTERS - /* check paca->perf_counter_pending if we're enabling ints */ +#ifdef CONFIG_PERF_EVENTS + /* check paca->perf_event_pending if we're enabling ints */ lbz r3,PACAPERFPEND(r13) and. r3,r3,r5 beq 27f - bl .perf_counter_do_pending + bl .perf_event_do_pending 27: -#endif /* CONFIG_PERF_COUNTERS */ +#endif /* CONFIG_PERF_EVENTS */ /* extract EE bit and use it to restore paca->hard_enabled */ ld r3,_MSR(r1) diff --git a/arch/powerpc/kernel/irq.c b/arch/powerpc/kernel/irq.c index f7f376e..e5d1211 100644 --- a/arch/powerpc/kernel/irq.c +++ b/arch/powerpc/kernel/irq.c @@ -53,7 +53,7 @@ #include #include #include -#include +#include #include #include @@ -138,9 +138,9 @@ notrace void raw_local_irq_restore(unsigned long en) } #endif /* CONFIG_PPC_STD_MMU_64 */ - if (test_perf_counter_pending()) { - clear_perf_counter_pending(); - perf_counter_do_pending(); + if (test_perf_event_pending()) { + clear_perf_event_pending(); + perf_event_do_pending(); } /* diff --git a/arch/powerpc/kernel/mpc7450-pmu.c b/arch/powerpc/kernel/mpc7450-pmu.c index cc466d0..09d7202 100644 --- a/arch/powerpc/kernel/mpc7450-pmu.c +++ b/arch/powerpc/kernel/mpc7450-pmu.c @@ -9,7 +9,7 @@ * 2 of the License, or (at your option) any later version. */ #include -#include +#include #include #include diff --git a/arch/powerpc/kernel/perf_callchain.c b/arch/powerpc/kernel/perf_callchain.c index f74b62c..0a03cf7 100644 --- a/arch/powerpc/kernel/perf_callchain.c +++ b/arch/powerpc/kernel/perf_callchain.c @@ -10,7 +10,7 @@ */ #include #include -#include +#include #include #include #include diff --git a/arch/powerpc/kernel/perf_counter.c b/arch/powerpc/kernel/perf_counter.c deleted file mode 100644 index 5ccf9bc..0000000 --- a/arch/powerpc/kernel/perf_counter.c +++ /dev/null @@ -1,1315 +0,0 @@ -/* - * Performance counter support - powerpc architecture code - * - * Copyright 2008-2009 Paul Mackerras, IBM Corporation. - * - * This program is free software; you can redistribute it and/or - * modify it under the terms of the GNU General Public License - * as published by the Free Software Foundation; either version - * 2 of the License, or (at your option) any later version. - */ -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include - -struct cpu_hw_counters { - int n_counters; - int n_percpu; - int disabled; - int n_added; - int n_limited; - u8 pmcs_enabled; - struct perf_counter *counter[MAX_HWCOUNTERS]; - u64 events[MAX_HWCOUNTERS]; - unsigned int flags[MAX_HWCOUNTERS]; - unsigned long mmcr[3]; - struct perf_counter *limited_counter[MAX_LIMITED_HWCOUNTERS]; - u8 limited_hwidx[MAX_LIMITED_HWCOUNTERS]; - u64 alternatives[MAX_HWCOUNTERS][MAX_EVENT_ALTERNATIVES]; - unsigned long amasks[MAX_HWCOUNTERS][MAX_EVENT_ALTERNATIVES]; - unsigned long avalues[MAX_HWCOUNTERS][MAX_EVENT_ALTERNATIVES]; -}; -DEFINE_PER_CPU(struct cpu_hw_counters, cpu_hw_counters); - -struct power_pmu *ppmu; - -/* - * Normally, to ignore kernel events we set the FCS (freeze counters - * in supervisor mode) bit in MMCR0, but if the kernel runs with the - * hypervisor bit set in the MSR, or if we are running on a processor - * where the hypervisor bit is forced to 1 (as on Apple G5 processors), - * then we need to use the FCHV bit to ignore kernel events. - */ -static unsigned int freeze_counters_kernel = MMCR0_FCS; - -/* - * 32-bit doesn't have MMCRA but does have an MMCR2, - * and a few other names are different. - */ -#ifdef CONFIG_PPC32 - -#define MMCR0_FCHV 0 -#define MMCR0_PMCjCE MMCR0_PMCnCE - -#define SPRN_MMCRA SPRN_MMCR2 -#define MMCRA_SAMPLE_ENABLE 0 - -static inline unsigned long perf_ip_adjust(struct pt_regs *regs) -{ - return 0; -} -static inline void perf_get_data_addr(struct pt_regs *regs, u64 *addrp) { } -static inline u32 perf_get_misc_flags(struct pt_regs *regs) -{ - return 0; -} -static inline void perf_read_regs(struct pt_regs *regs) { } -static inline int perf_intr_is_nmi(struct pt_regs *regs) -{ - return 0; -} - -#endif /* CONFIG_PPC32 */ - -/* - * Things that are specific to 64-bit implementations. - */ -#ifdef CONFIG_PPC64 - -static inline unsigned long perf_ip_adjust(struct pt_regs *regs) -{ - unsigned long mmcra = regs->dsisr; - - if ((mmcra & MMCRA_SAMPLE_ENABLE) && !(ppmu->flags & PPMU_ALT_SIPR)) { - unsigned long slot = (mmcra & MMCRA_SLOT) >> MMCRA_SLOT_SHIFT; - if (slot > 1) - return 4 * (slot - 1); - } - return 0; -} - -/* - * The user wants a data address recorded. - * If we're not doing instruction sampling, give them the SDAR - * (sampled data address). If we are doing instruction sampling, then - * only give them the SDAR if it corresponds to the instruction - * pointed to by SIAR; this is indicated by the [POWER6_]MMCRA_SDSYNC - * bit in MMCRA. - */ -static inline void perf_get_data_addr(struct pt_regs *regs, u64 *addrp) -{ - unsigned long mmcra = regs->dsisr; - unsigned long sdsync = (ppmu->flags & PPMU_ALT_SIPR) ? - POWER6_MMCRA_SDSYNC : MMCRA_SDSYNC; - - if (!(mmcra & MMCRA_SAMPLE_ENABLE) || (mmcra & sdsync)) - *addrp = mfspr(SPRN_SDAR); -} - -static inline u32 perf_get_misc_flags(struct pt_regs *regs) -{ - unsigned long mmcra = regs->dsisr; - - if (TRAP(regs) != 0xf00) - return 0; /* not a PMU interrupt */ - - if (ppmu->flags & PPMU_ALT_SIPR) { - if (mmcra & POWER6_MMCRA_SIHV) - return PERF_EVENT_MISC_HYPERVISOR; - return (mmcra & POWER6_MMCRA_SIPR) ? - PERF_EVENT_MISC_USER : PERF_EVENT_MISC_KERNEL; - } - if (mmcra & MMCRA_SIHV) - return PERF_EVENT_MISC_HYPERVISOR; - return (mmcra & MMCRA_SIPR) ? PERF_EVENT_MISC_USER : - PERF_EVENT_MISC_KERNEL; -} - -/* - * Overload regs->dsisr to store MMCRA so we only need to read it once - * on each interrupt. - */ -static inline void perf_read_regs(struct pt_regs *regs) -{ - regs->dsisr = mfspr(SPRN_MMCRA); -} - -/* - * If interrupts were soft-disabled when a PMU interrupt occurs, treat - * it as an NMI. - */ -static inline int perf_intr_is_nmi(struct pt_regs *regs) -{ - return !regs->softe; -} - -#endif /* CONFIG_PPC64 */ - -static void perf_counter_interrupt(struct pt_regs *regs); - -void perf_counter_print_debug(void) -{ -} - -/* - * Read one performance monitor counter (PMC). - */ -static unsigned long read_pmc(int idx) -{ - unsigned long val; - - switch (idx) { - case 1: - val = mfspr(SPRN_PMC1); - break; - case 2: - val = mfspr(SPRN_PMC2); - break; - case 3: - val = mfspr(SPRN_PMC3); - break; - case 4: - val = mfspr(SPRN_PMC4); - break; - case 5: - val = mfspr(SPRN_PMC5); - break; - case 6: - val = mfspr(SPRN_PMC6); - break; -#ifdef CONFIG_PPC64 - case 7: - val = mfspr(SPRN_PMC7); - break; - case 8: - val = mfspr(SPRN_PMC8); - break; -#endif /* CONFIG_PPC64 */ - default: - printk(KERN_ERR "oops trying to read PMC%d\n", idx); - val = 0; - } - return val; -} - -/* - * Write one PMC. - */ -static void write_pmc(int idx, unsigned long val) -{ - switch (idx) { - case 1: - mtspr(SPRN_PMC1, val); - break; - case 2: - mtspr(SPRN_PMC2, val); - break; - case 3: - mtspr(SPRN_PMC3, val); - break; - case 4: - mtspr(SPRN_PMC4, val); - break; - case 5: - mtspr(SPRN_PMC5, val); - break; - case 6: - mtspr(SPRN_PMC6, val); - break; -#ifdef CONFIG_PPC64 - case 7: - mtspr(SPRN_PMC7, val); - break; - case 8: - mtspr(SPRN_PMC8, val); - break; -#endif /* CONFIG_PPC64 */ - default: - printk(KERN_ERR "oops trying to write PMC%d\n", idx); - } -} - -/* - * Check if a set of events can all go on the PMU at once. - * If they can't, this will look at alternative codes for the events - * and see if any combination of alternative codes is feasible. - * The feasible set is returned in event[]. - */ -static int power_check_constraints(struct cpu_hw_counters *cpuhw, - u64 event[], unsigned int cflags[], - int n_ev) -{ - unsigned long mask, value, nv; - unsigned long smasks[MAX_HWCOUNTERS], svalues[MAX_HWCOUNTERS]; - int n_alt[MAX_HWCOUNTERS], choice[MAX_HWCOUNTERS]; - int i, j; - unsigned long addf = ppmu->add_fields; - unsigned long tadd = ppmu->test_adder; - - if (n_ev > ppmu->n_counter) - return -1; - - /* First see if the events will go on as-is */ - for (i = 0; i < n_ev; ++i) { - if ((cflags[i] & PPMU_LIMITED_PMC_REQD) - && !ppmu->limited_pmc_event(event[i])) { - ppmu->get_alternatives(event[i], cflags[i], - cpuhw->alternatives[i]); - event[i] = cpuhw->alternatives[i][0]; - } - if (ppmu->get_constraint(event[i], &cpuhw->amasks[i][0], - &cpuhw->avalues[i][0])) - return -1; - } - value = mask = 0; - for (i = 0; i < n_ev; ++i) { - nv = (value | cpuhw->avalues[i][0]) + - (value & cpuhw->avalues[i][0] & addf); - if ((((nv + tadd) ^ value) & mask) != 0 || - (((nv + tadd) ^ cpuhw->avalues[i][0]) & - cpuhw->amasks[i][0]) != 0) - break; - value = nv; - mask |= cpuhw->amasks[i][0]; - } - if (i == n_ev) - return 0; /* all OK */ - - /* doesn't work, gather alternatives... */ - if (!ppmu->get_alternatives) - return -1; - for (i = 0; i < n_ev; ++i) { - choice[i] = 0; - n_alt[i] = ppmu->get_alternatives(event[i], cflags[i], - cpuhw->alternatives[i]); - for (j = 1; j < n_alt[i]; ++j) - ppmu->get_constraint(cpuhw->alternatives[i][j], - &cpuhw->amasks[i][j], - &cpuhw->avalues[i][j]); - } - - /* enumerate all possibilities and see if any will work */ - i = 0; - j = -1; - value = mask = nv = 0; - while (i < n_ev) { - if (j >= 0) { - /* we're backtracking, restore context */ - value = svalues[i]; - mask = smasks[i]; - j = choice[i]; - } - /* - * See if any alternative k for event i, - * where k > j, will satisfy the constraints. - */ - while (++j < n_alt[i]) { - nv = (value | cpuhw->avalues[i][j]) + - (value & cpuhw->avalues[i][j] & addf); - if ((((nv + tadd) ^ value) & mask) == 0 && - (((nv + tadd) ^ cpuhw->avalues[i][j]) - & cpuhw->amasks[i][j]) == 0) - break; - } - if (j >= n_alt[i]) { - /* - * No feasible alternative, backtrack - * to event i-1 and continue enumerating its - * alternatives from where we got up to. - */ - if (--i < 0) - return -1; - } else { - /* - * Found a feasible alternative for event i, - * remember where we got up to with this event, - * go on to the next event, and start with - * the first alternative for it. - */ - choice[i] = j; - svalues[i] = value; - smasks[i] = mask; - value = nv; - mask |= cpuhw->amasks[i][j]; - ++i; - j = -1; - } - } - - /* OK, we have a feasible combination, tell the caller the solution */ - for (i = 0; i < n_ev; ++i) - event[i] = cpuhw->alternatives[i][choice[i]]; - return 0; -} - -/* - * Check if newly-added counters have consistent settings for - * exclude_{user,kernel,hv} with each other and any previously - * added counters. - */ -static int check_excludes(struct perf_counter **ctrs, unsigned int cflags[], - int n_prev, int n_new) -{ - int eu = 0, ek = 0, eh = 0; - int i, n, first; - struct perf_counter *counter; - - n = n_prev + n_new; - if (n <= 1) - return 0; - - first = 1; - for (i = 0; i < n; ++i) { - if (cflags[i] & PPMU_LIMITED_PMC_OK) { - cflags[i] &= ~PPMU_LIMITED_PMC_REQD; - continue; - } - counter = ctrs[i]; - if (first) { - eu = counter->attr.exclude_user; - ek = counter->attr.exclude_kernel; - eh = counter->attr.exclude_hv; - first = 0; - } else if (counter->attr.exclude_user != eu || - counter->attr.exclude_kernel != ek || - counter->attr.exclude_hv != eh) { - return -EAGAIN; - } - } - - if (eu || ek || eh) - for (i = 0; i < n; ++i) - if (cflags[i] & PPMU_LIMITED_PMC_OK) - cflags[i] |= PPMU_LIMITED_PMC_REQD; - - return 0; -} - -static void power_pmu_read(struct perf_counter *counter) -{ - s64 val, delta, prev; - - if (!counter->hw.idx) - return; - /* - * Performance monitor interrupts come even when interrupts - * are soft-disabled, as long as interrupts are hard-enabled. - * Therefore we treat them like NMIs. - */ - do { - prev = atomic64_read(&counter->hw.prev_count); - barrier(); - val = read_pmc(counter->hw.idx); - } while (atomic64_cmpxchg(&counter->hw.prev_count, prev, val) != prev); - - /* The counters are only 32 bits wide */ - delta = (val - prev) & 0xfffffffful; - atomic64_add(delta, &counter->count); - atomic64_sub(delta, &counter->hw.period_left); -} - -/* - * On some machines, PMC5 and PMC6 can't be written, don't respect - * the freeze conditions, and don't generate interrupts. This tells - * us if `counter' is using such a PMC. - */ -static int is_limited_pmc(int pmcnum) -{ - return (ppmu->flags & PPMU_LIMITED_PMC5_6) - && (pmcnum == 5 || pmcnum == 6); -} - -static void freeze_limited_counters(struct cpu_hw_counters *cpuhw, - unsigned long pmc5, unsigned long pmc6) -{ - struct perf_counter *counter; - u64 val, prev, delta; - int i; - - for (i = 0; i < cpuhw->n_limited; ++i) { - counter = cpuhw->limited_counter[i]; - if (!counter->hw.idx) - continue; - val = (counter->hw.idx == 5) ? pmc5 : pmc6; - prev = atomic64_read(&counter->hw.prev_count); - counter->hw.idx = 0; - delta = (val - prev) & 0xfffffffful; - atomic64_add(delta, &counter->count); - } -} - -static void thaw_limited_counters(struct cpu_hw_counters *cpuhw, - unsigned long pmc5, unsigned long pmc6) -{ - struct perf_counter *counter; - u64 val; - int i; - - for (i = 0; i < cpuhw->n_limited; ++i) { - counter = cpuhw->limited_counter[i]; - counter->hw.idx = cpuhw->limited_hwidx[i]; - val = (counter->hw.idx == 5) ? pmc5 : pmc6; - atomic64_set(&counter->hw.prev_count, val); - perf_counter_update_userpage(counter); - } -} - -/* - * Since limited counters don't respect the freeze conditions, we - * have to read them immediately after freezing or unfreezing the - * other counters. We try to keep the values from the limited - * counters as consistent as possible by keeping the delay (in - * cycles and instructions) between freezing/unfreezing and reading - * the limited counters as small and consistent as possible. - * Therefore, if any limited counters are in use, we read them - * both, and always in the same order, to minimize variability, - * and do it inside the same asm that writes MMCR0. - */ -static void write_mmcr0(struct cpu_hw_counters *cpuhw, unsigned long mmcr0) -{ - unsigned long pmc5, pmc6; - - if (!cpuhw->n_limited) { - mtspr(SPRN_MMCR0, mmcr0); - return; - } - - /* - * Write MMCR0, then read PMC5 and PMC6 immediately. - * To ensure we don't get a performance monitor interrupt - * between writing MMCR0 and freezing/thawing the limited - * counters, we first write MMCR0 with the counter overflow - * interrupt enable bits turned off. - */ - asm volatile("mtspr %3,%2; mfspr %0,%4; mfspr %1,%5" - : "=&r" (pmc5), "=&r" (pmc6) - : "r" (mmcr0 & ~(MMCR0_PMC1CE | MMCR0_PMCjCE)), - "i" (SPRN_MMCR0), - "i" (SPRN_PMC5), "i" (SPRN_PMC6)); - - if (mmcr0 & MMCR0_FC) - freeze_limited_counters(cpuhw, pmc5, pmc6); - else - thaw_limited_counters(cpuhw, pmc5, pmc6); - - /* - * Write the full MMCR0 including the counter overflow interrupt - * enable bits, if necessary. - */ - if (mmcr0 & (MMCR0_PMC1CE | MMCR0_PMCjCE)) - mtspr(SPRN_MMCR0, mmcr0); -} - -/* - * Disable all counters to prevent PMU interrupts and to allow - * counters to be added or removed. - */ -void hw_perf_disable(void) -{ - struct cpu_hw_counters *cpuhw; - unsigned long flags; - - if (!ppmu) - return; - local_irq_save(flags); - cpuhw = &__get_cpu_var(cpu_hw_counters); - - if (!cpuhw->disabled) { - cpuhw->disabled = 1; - cpuhw->n_added = 0; - - /* - * Check if we ever enabled the PMU on this cpu. - */ - if (!cpuhw->pmcs_enabled) { - ppc_enable_pmcs(); - cpuhw->pmcs_enabled = 1; - } - - /* - * Disable instruction sampling if it was enabled - */ - if (cpuhw->mmcr[2] & MMCRA_SAMPLE_ENABLE) { - mtspr(SPRN_MMCRA, - cpuhw->mmcr[2] & ~MMCRA_SAMPLE_ENABLE); - mb(); - } - - /* - * Set the 'freeze counters' bit. - * The barrier is to make sure the mtspr has been - * executed and the PMU has frozen the counters - * before we return. - */ - write_mmcr0(cpuhw, mfspr(SPRN_MMCR0) | MMCR0_FC); - mb(); - } - local_irq_restore(flags); -} - -/* - * Re-enable all counters if disable == 0. - * If we were previously disabled and counters were added, then - * put the new config on the PMU. - */ -void hw_perf_enable(void) -{ - struct perf_counter *counter; - struct cpu_hw_counters *cpuhw; - unsigned long flags; - long i; - unsigned long val; - s64 left; - unsigned int hwc_index[MAX_HWCOUNTERS]; - int n_lim; - int idx; - - if (!ppmu) - return; - local_irq_save(flags); - cpuhw = &__get_cpu_var(cpu_hw_counters); - if (!cpuhw->disabled) { - local_irq_restore(flags); - return; - } - cpuhw->disabled = 0; - - /* - * If we didn't change anything, or only removed counters, - * no need to recalculate MMCR* settings and reset the PMCs. - * Just reenable the PMU with the current MMCR* settings - * (possibly updated for removal of counters). - */ - if (!cpuhw->n_added) { - mtspr(SPRN_MMCRA, cpuhw->mmcr[2] & ~MMCRA_SAMPLE_ENABLE); - mtspr(SPRN_MMCR1, cpuhw->mmcr[1]); - if (cpuhw->n_counters == 0) - ppc_set_pmu_inuse(0); - goto out_enable; - } - - /* - * Compute MMCR* values for the new set of counters - */ - if (ppmu->compute_mmcr(cpuhw->events, cpuhw->n_counters, hwc_index, - cpuhw->mmcr)) { - /* shouldn't ever get here */ - printk(KERN_ERR "oops compute_mmcr failed\n"); - goto out; - } - - /* - * Add in MMCR0 freeze bits corresponding to the - * attr.exclude_* bits for the first counter. - * We have already checked that all counters have the - * same values for these bits as the first counter. - */ - counter = cpuhw->counter[0]; - if (counter->attr.exclude_user) - cpuhw->mmcr[0] |= MMCR0_FCP; - if (counter->attr.exclude_kernel) - cpuhw->mmcr[0] |= freeze_counters_kernel; - if (counter->attr.exclude_hv) - cpuhw->mmcr[0] |= MMCR0_FCHV; - - /* - * Write the new configuration to MMCR* with the freeze - * bit set and set the hardware counters to their initial values. - * Then unfreeze the counters. - */ - ppc_set_pmu_inuse(1); - mtspr(SPRN_MMCRA, cpuhw->mmcr[2] & ~MMCRA_SAMPLE_ENABLE); - mtspr(SPRN_MMCR1, cpuhw->mmcr[1]); - mtspr(SPRN_MMCR0, (cpuhw->mmcr[0] & ~(MMCR0_PMC1CE | MMCR0_PMCjCE)) - | MMCR0_FC); - - /* - * Read off any pre-existing counters that need to move - * to another PMC. - */ - for (i = 0; i < cpuhw->n_counters; ++i) { - counter = cpuhw->counter[i]; - if (counter->hw.idx && counter->hw.idx != hwc_index[i] + 1) { - power_pmu_read(counter); - write_pmc(counter->hw.idx, 0); - counter->hw.idx = 0; - } - } - - /* - * Initialize the PMCs for all the new and moved counters. - */ - cpuhw->n_limited = n_lim = 0; - for (i = 0; i < cpuhw->n_counters; ++i) { - counter = cpuhw->counter[i]; - if (counter->hw.idx) - continue; - idx = hwc_index[i] + 1; - if (is_limited_pmc(idx)) { - cpuhw->limited_counter[n_lim] = counter; - cpuhw->limited_hwidx[n_lim] = idx; - ++n_lim; - continue; - } - val = 0; - if (counter->hw.sample_period) { - left = atomic64_read(&counter->hw.period_left); - if (left < 0x80000000L) - val = 0x80000000L - left; - } - atomic64_set(&counter->hw.prev_count, val); - counter->hw.idx = idx; - write_pmc(idx, val); - perf_counter_update_userpage(counter); - } - cpuhw->n_limited = n_lim; - cpuhw->mmcr[0] |= MMCR0_PMXE | MMCR0_FCECE; - - out_enable: - mb(); - write_mmcr0(cpuhw, cpuhw->mmcr[0]); - - /* - * Enable instruction sampling if necessary - */ - if (cpuhw->mmcr[2] & MMCRA_SAMPLE_ENABLE) { - mb(); - mtspr(SPRN_MMCRA, cpuhw->mmcr[2]); - } - - out: - local_irq_restore(flags); -} - -static int collect_events(struct perf_counter *group, int max_count, - struct perf_counter *ctrs[], u64 *events, - unsigned int *flags) -{ - int n = 0; - struct perf_counter *counter; - - if (!is_software_counter(group)) { - if (n >= max_count) - return -1; - ctrs[n] = group; - flags[n] = group->hw.counter_base; - events[n++] = group->hw.config; - } - list_for_each_entry(counter, &group->sibling_list, list_entry) { - if (!is_software_counter(counter) && - counter->state != PERF_COUNTER_STATE_OFF) { - if (n >= max_count) - return -1; - ctrs[n] = counter; - flags[n] = counter->hw.counter_base; - events[n++] = counter->hw.config; - } - } - return n; -} - -static void counter_sched_in(struct perf_counter *counter, int cpu) -{ - counter->state = PERF_COUNTER_STATE_ACTIVE; - counter->oncpu = cpu; - counter->tstamp_running += counter->ctx->time - counter->tstamp_stopped; - if (is_software_counter(counter)) - counter->pmu->enable(counter); -} - -/* - * Called to enable a whole group of counters. - * Returns 1 if the group was enabled, or -EAGAIN if it could not be. - * Assumes the caller has disabled interrupts and has - * frozen the PMU with hw_perf_save_disable. - */ -int hw_perf_group_sched_in(struct perf_counter *group_leader, - struct perf_cpu_context *cpuctx, - struct perf_counter_context *ctx, int cpu) -{ - struct cpu_hw_counters *cpuhw; - long i, n, n0; - struct perf_counter *sub; - - if (!ppmu) - return 0; - cpuhw = &__get_cpu_var(cpu_hw_counters); - n0 = cpuhw->n_counters; - n = collect_events(group_leader, ppmu->n_counter - n0, - &cpuhw->counter[n0], &cpuhw->events[n0], - &cpuhw->flags[n0]); - if (n < 0) - return -EAGAIN; - if (check_excludes(cpuhw->counter, cpuhw->flags, n0, n)) - return -EAGAIN; - i = power_check_constraints(cpuhw, cpuhw->events, cpuhw->flags, n + n0); - if (i < 0) - return -EAGAIN; - cpuhw->n_counters = n0 + n; - cpuhw->n_added += n; - - /* - * OK, this group can go on; update counter states etc., - * and enable any software counters - */ - for (i = n0; i < n0 + n; ++i) - cpuhw->counter[i]->hw.config = cpuhw->events[i]; - cpuctx->active_oncpu += n; - n = 1; - counter_sched_in(group_leader, cpu); - list_for_each_entry(sub, &group_leader->sibling_list, list_entry) { - if (sub->state != PERF_COUNTER_STATE_OFF) { - counter_sched_in(sub, cpu); - ++n; - } - } - ctx->nr_active += n; - - return 1; -} - -/* - * Add a counter to the PMU. - * If all counters are not already frozen, then we disable and - * re-enable the PMU in order to get hw_perf_enable to do the - * actual work of reconfiguring the PMU. - */ -static int power_pmu_enable(struct perf_counter *counter) -{ - struct cpu_hw_counters *cpuhw; - unsigned long flags; - int n0; - int ret = -EAGAIN; - - local_irq_save(flags); - perf_disable(); - - /* - * Add the counter to the list (if there is room) - * and check whether the total set is still feasible. - */ - cpuhw = &__get_cpu_var(cpu_hw_counters); - n0 = cpuhw->n_counters; - if (n0 >= ppmu->n_counter) - goto out; - cpuhw->counter[n0] = counter; - cpuhw->events[n0] = counter->hw.config; - cpuhw->flags[n0] = counter->hw.counter_base; - if (check_excludes(cpuhw->counter, cpuhw->flags, n0, 1)) - goto out; - if (power_check_constraints(cpuhw, cpuhw->events, cpuhw->flags, n0 + 1)) - goto out; - - counter->hw.config = cpuhw->events[n0]; - ++cpuhw->n_counters; - ++cpuhw->n_added; - - ret = 0; - out: - perf_enable(); - local_irq_restore(flags); - return ret; -} - -/* - * Remove a counter from the PMU. - */ -static void power_pmu_disable(struct perf_counter *counter) -{ - struct cpu_hw_counters *cpuhw; - long i; - unsigned long flags; - - local_irq_save(flags); - perf_disable(); - - power_pmu_read(counter); - - cpuhw = &__get_cpu_var(cpu_hw_counters); - for (i = 0; i < cpuhw->n_counters; ++i) { - if (counter == cpuhw->counter[i]) { - while (++i < cpuhw->n_counters) - cpuhw->counter[i-1] = cpuhw->counter[i]; - --cpuhw->n_counters; - ppmu->disable_pmc(counter->hw.idx - 1, cpuhw->mmcr); - if (counter->hw.idx) { - write_pmc(counter->hw.idx, 0); - counter->hw.idx = 0; - } - perf_counter_update_userpage(counter); - break; - } - } - for (i = 0; i < cpuhw->n_limited; ++i) - if (counter == cpuhw->limited_counter[i]) - break; - if (i < cpuhw->n_limited) { - while (++i < cpuhw->n_limited) { - cpuhw->limited_counter[i-1] = cpuhw->limited_counter[i]; - cpuhw->limited_hwidx[i-1] = cpuhw->limited_hwidx[i]; - } - --cpuhw->n_limited; - } - if (cpuhw->n_counters == 0) { - /* disable exceptions if no counters are running */ - cpuhw->mmcr[0] &= ~(MMCR0_PMXE | MMCR0_FCECE); - } - - perf_enable(); - local_irq_restore(flags); -} - -/* - * Re-enable interrupts on a counter after they were throttled - * because they were coming too fast. - */ -static void power_pmu_unthrottle(struct perf_counter *counter) -{ - s64 val, left; - unsigned long flags; - - if (!counter->hw.idx || !counter->hw.sample_period) - return; - local_irq_save(flags); - perf_disable(); - power_pmu_read(counter); - left = counter->hw.sample_period; - counter->hw.last_period = left; - val = 0; - if (left < 0x80000000L) - val = 0x80000000L - left; - write_pmc(counter->hw.idx, val); - atomic64_set(&counter->hw.prev_count, val); - atomic64_set(&counter->hw.period_left, left); - perf_counter_update_userpage(counter); - perf_enable(); - local_irq_restore(flags); -} - -struct pmu power_pmu = { - .enable = power_pmu_enable, - .disable = power_pmu_disable, - .read = power_pmu_read, - .unthrottle = power_pmu_unthrottle, -}; - -/* - * Return 1 if we might be able to put counter on a limited PMC, - * or 0 if not. - * A counter can only go on a limited PMC if it counts something - * that a limited PMC can count, doesn't require interrupts, and - * doesn't exclude any processor mode. - */ -static int can_go_on_limited_pmc(struct perf_counter *counter, u64 ev, - unsigned int flags) -{ - int n; - u64 alt[MAX_EVENT_ALTERNATIVES]; - - if (counter->attr.exclude_user - || counter->attr.exclude_kernel - || counter->attr.exclude_hv - || counter->attr.sample_period) - return 0; - - if (ppmu->limited_pmc_event(ev)) - return 1; - - /* - * The requested event isn't on a limited PMC already; - * see if any alternative code goes on a limited PMC. - */ - if (!ppmu->get_alternatives) - return 0; - - flags |= PPMU_LIMITED_PMC_OK | PPMU_LIMITED_PMC_REQD; - n = ppmu->get_alternatives(ev, flags, alt); - - return n > 0; -} - -/* - * Find an alternative event that goes on a normal PMC, if possible, - * and return the event code, or 0 if there is no such alternative. - * (Note: event code 0 is "don't count" on all machines.) - */ -static u64 normal_pmc_alternative(u64 ev, unsigned long flags) -{ - u64 alt[MAX_EVENT_ALTERNATIVES]; - int n; - - flags &= ~(PPMU_LIMITED_PMC_OK | PPMU_LIMITED_PMC_REQD); - n = ppmu->get_alternatives(ev, flags, alt); - if (!n) - return 0; - return alt[0]; -} - -/* Number of perf_counters counting hardware events */ -static atomic_t num_counters; -/* Used to avoid races in calling reserve/release_pmc_hardware */ -static DEFINE_MUTEX(pmc_reserve_mutex); - -/* - * Release the PMU if this is the last perf_counter. - */ -static void hw_perf_counter_destroy(struct perf_counter *counter) -{ - if (!atomic_add_unless(&num_counters, -1, 1)) { - mutex_lock(&pmc_reserve_mutex); - if (atomic_dec_return(&num_counters) == 0) - release_pmc_hardware(); - mutex_unlock(&pmc_reserve_mutex); - } -} - -/* - * Translate a generic cache event config to a raw event code. - */ -static int hw_perf_cache_event(u64 config, u64 *eventp) -{ - unsigned long type, op, result; - int ev; - - if (!ppmu->cache_events) - return -EINVAL; - - /* unpack config */ - type = config & 0xff; - op = (config >> 8) & 0xff; - result = (config >> 16) & 0xff; - - if (type >= PERF_COUNT_HW_CACHE_MAX || - op >= PERF_COUNT_HW_CACHE_OP_MAX || - result >= PERF_COUNT_HW_CACHE_RESULT_MAX) - return -EINVAL; - - ev = (*ppmu->cache_events)[type][op][result]; - if (ev == 0) - return -EOPNOTSUPP; - if (ev == -1) - return -EINVAL; - *eventp = ev; - return 0; -} - -const struct pmu *hw_perf_counter_init(struct perf_counter *counter) -{ - u64 ev; - unsigned long flags; - struct perf_counter *ctrs[MAX_HWCOUNTERS]; - u64 events[MAX_HWCOUNTERS]; - unsigned int cflags[MAX_HWCOUNTERS]; - int n; - int err; - struct cpu_hw_counters *cpuhw; - - if (!ppmu) - return ERR_PTR(-ENXIO); - switch (counter->attr.type) { - case PERF_TYPE_HARDWARE: - ev = counter->attr.config; - if (ev >= ppmu->n_generic || ppmu->generic_events[ev] == 0) - return ERR_PTR(-EOPNOTSUPP); - ev = ppmu->generic_events[ev]; - break; - case PERF_TYPE_HW_CACHE: - err = hw_perf_cache_event(counter->attr.config, &ev); - if (err) - return ERR_PTR(err); - break; - case PERF_TYPE_RAW: - ev = counter->attr.config; - break; - default: - return ERR_PTR(-EINVAL); - } - counter->hw.config_base = ev; - counter->hw.idx = 0; - - /* - * If we are not running on a hypervisor, force the - * exclude_hv bit to 0 so that we don't care what - * the user set it to. - */ - if (!firmware_has_feature(FW_FEATURE_LPAR)) - counter->attr.exclude_hv = 0; - - /* - * If this is a per-task counter, then we can use - * PM_RUN_* events interchangeably with their non RUN_* - * equivalents, e.g. PM_RUN_CYC instead of PM_CYC. - * XXX we should check if the task is an idle task. - */ - flags = 0; - if (counter->ctx->task) - flags |= PPMU_ONLY_COUNT_RUN; - - /* - * If this machine has limited counters, check whether this - * event could go on a limited counter. - */ - if (ppmu->flags & PPMU_LIMITED_PMC5_6) { - if (can_go_on_limited_pmc(counter, ev, flags)) { - flags |= PPMU_LIMITED_PMC_OK; - } else if (ppmu->limited_pmc_event(ev)) { - /* - * The requested event is on a limited PMC, - * but we can't use a limited PMC; see if any - * alternative goes on a normal PMC. - */ - ev = normal_pmc_alternative(ev, flags); - if (!ev) - return ERR_PTR(-EINVAL); - } - } - - /* - * If this is in a group, check if it can go on with all the - * other hardware counters in the group. We assume the counter - * hasn't been linked into its leader's sibling list at this point. - */ - n = 0; - if (counter->group_leader != counter) { - n = collect_events(counter->group_leader, ppmu->n_counter - 1, - ctrs, events, cflags); - if (n < 0) - return ERR_PTR(-EINVAL); - } - events[n] = ev; - ctrs[n] = counter; - cflags[n] = flags; - if (check_excludes(ctrs, cflags, n, 1)) - return ERR_PTR(-EINVAL); - - cpuhw = &get_cpu_var(cpu_hw_counters); - err = power_check_constraints(cpuhw, events, cflags, n + 1); - put_cpu_var(cpu_hw_counters); - if (err) - return ERR_PTR(-EINVAL); - - counter->hw.config = events[n]; - counter->hw.counter_base = cflags[n]; - counter->hw.last_period = counter->hw.sample_period; - atomic64_set(&counter->hw.period_left, counter->hw.last_period); - - /* - * See if we need to reserve the PMU. - * If no counters are currently in use, then we have to take a - * mutex to ensure that we don't race with another task doing - * reserve_pmc_hardware or release_pmc_hardware. - */ - err = 0; - if (!atomic_inc_not_zero(&num_counters)) { - mutex_lock(&pmc_reserve_mutex); - if (atomic_read(&num_counters) == 0 && - reserve_pmc_hardware(perf_counter_interrupt)) - err = -EBUSY; - else - atomic_inc(&num_counters); - mutex_unlock(&pmc_reserve_mutex); - } - counter->destroy = hw_perf_counter_destroy; - - if (err) - return ERR_PTR(err); - return &power_pmu; -} - -/* - * A counter has overflowed; update its count and record - * things if requested. Note that interrupts are hard-disabled - * here so there is no possibility of being interrupted. - */ -static void record_and_restart(struct perf_counter *counter, unsigned long val, - struct pt_regs *regs, int nmi) -{ - u64 period = counter->hw.sample_period; - s64 prev, delta, left; - int record = 0; - - /* we don't have to worry about interrupts here */ - prev = atomic64_read(&counter->hw.prev_count); - delta = (val - prev) & 0xfffffffful; - atomic64_add(delta, &counter->count); - - /* - * See if the total period for this counter has expired, - * and update for the next period. - */ - val = 0; - left = atomic64_read(&counter->hw.period_left) - delta; - if (period) { - if (left <= 0) { - left += period; - if (left <= 0) - left = period; - record = 1; - } - if (left < 0x80000000LL) - val = 0x80000000LL - left; - } - - /* - * Finally record data if requested. - */ - if (record) { - struct perf_sample_data data = { - .addr = 0, - .period = counter->hw.last_period, - }; - - if (counter->attr.sample_type & PERF_SAMPLE_ADDR) - perf_get_data_addr(regs, &data.addr); - - if (perf_counter_overflow(counter, nmi, &data, regs)) { - /* - * Interrupts are coming too fast - throttle them - * by setting the counter to 0, so it will be - * at least 2^30 cycles until the next interrupt - * (assuming each counter counts at most 2 counts - * per cycle). - */ - val = 0; - left = ~0ULL >> 1; - } - } - - write_pmc(counter->hw.idx, val); - atomic64_set(&counter->hw.prev_count, val); - atomic64_set(&counter->hw.period_left, left); - perf_counter_update_userpage(counter); -} - -/* - * Called from generic code to get the misc flags (i.e. processor mode) - * for an event. - */ -unsigned long perf_misc_flags(struct pt_regs *regs) -{ - u32 flags = perf_get_misc_flags(regs); - - if (flags) - return flags; - return user_mode(regs) ? PERF_EVENT_MISC_USER : - PERF_EVENT_MISC_KERNEL; -} - -/* - * Called from generic code to get the instruction pointer - * for an event. - */ -unsigned long perf_instruction_pointer(struct pt_regs *regs) -{ - unsigned long ip; - - if (TRAP(regs) != 0xf00) - return regs->nip; /* not a PMU interrupt */ - - ip = mfspr(SPRN_SIAR) + perf_ip_adjust(regs); - return ip; -} - -/* - * Performance monitor interrupt stuff - */ -static void perf_counter_interrupt(struct pt_regs *regs) -{ - int i; - struct cpu_hw_counters *cpuhw = &__get_cpu_var(cpu_hw_counters); - struct perf_counter *counter; - unsigned long val; - int found = 0; - int nmi; - - if (cpuhw->n_limited) - freeze_limited_counters(cpuhw, mfspr(SPRN_PMC5), - mfspr(SPRN_PMC6)); - - perf_read_regs(regs); - - nmi = perf_intr_is_nmi(regs); - if (nmi) - nmi_enter(); - else - irq_enter(); - - for (i = 0; i < cpuhw->n_counters; ++i) { - counter = cpuhw->counter[i]; - if (!counter->hw.idx || is_limited_pmc(counter->hw.idx)) - continue; - val = read_pmc(counter->hw.idx); - if ((int)val < 0) { - /* counter has overflowed */ - found = 1; - record_and_restart(counter, val, regs, nmi); - } - } - - /* - * In case we didn't find and reset the counter that caused - * the interrupt, scan all counters and reset any that are - * negative, to avoid getting continual interrupts. - * Any that we processed in the previous loop will not be negative. - */ - if (!found) { - for (i = 0; i < ppmu->n_counter; ++i) { - if (is_limited_pmc(i + 1)) - continue; - val = read_pmc(i + 1); - if ((int)val < 0) - write_pmc(i + 1, 0); - } - } - - /* - * Reset MMCR0 to its normal value. This will set PMXE and - * clear FC (freeze counters) and PMAO (perf mon alert occurred) - * and thus allow interrupts to occur again. - * XXX might want to use MSR.PM to keep the counters frozen until - * we get back out of this interrupt. - */ - write_mmcr0(cpuhw, cpuhw->mmcr[0]); - - if (nmi) - nmi_exit(); - else - irq_exit(); -} - -void hw_perf_counter_setup(int cpu) -{ - struct cpu_hw_counters *cpuhw = &per_cpu(cpu_hw_counters, cpu); - - if (!ppmu) - return; - memset(cpuhw, 0, sizeof(*cpuhw)); - cpuhw->mmcr[0] = MMCR0_FC; -} - -int register_power_pmu(struct power_pmu *pmu) -{ - if (ppmu) - return -EBUSY; /* something's already registered */ - - ppmu = pmu; - pr_info("%s performance monitor hardware support registered\n", - pmu->name); - -#ifdef MSR_HV - /* - * Use FCHV to ignore kernel events if MSR.HV is set. - */ - if (mfmsr() & MSR_HV) - freeze_counters_kernel = MMCR0_FCHV; -#endif /* CONFIG_PPC64 */ - - return 0; -} diff --git a/arch/powerpc/kernel/perf_event.c b/arch/powerpc/kernel/perf_event.c new file mode 100644 index 0000000..c98321f --- /dev/null +++ b/arch/powerpc/kernel/perf_event.c @@ -0,0 +1,1315 @@ +/* + * Performance event support - powerpc architecture code + * + * Copyright 2008-2009 Paul Mackerras, IBM Corporation. + * + * This program is free software; you can redistribute it and/or + * modify it under the terms of the GNU General Public License + * as published by the Free Software Foundation; either version + * 2 of the License, or (at your option) any later version. + */ +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +struct cpu_hw_events { + int n_events; + int n_percpu; + int disabled; + int n_added; + int n_limited; + u8 pmcs_enabled; + struct perf_event *event[MAX_HWEVENTS]; + u64 events[MAX_HWEVENTS]; + unsigned int flags[MAX_HWEVENTS]; + unsigned long mmcr[3]; + struct perf_event *limited_event[MAX_LIMITED_HWEVENTS]; + u8 limited_hwidx[MAX_LIMITED_HWEVENTS]; + u64 alternatives[MAX_HWEVENTS][MAX_EVENT_ALTERNATIVES]; + unsigned long amasks[MAX_HWEVENTS][MAX_EVENT_ALTERNATIVES]; + unsigned long avalues[MAX_HWEVENTS][MAX_EVENT_ALTERNATIVES]; +}; +DEFINE_PER_CPU(struct cpu_hw_events, cpu_hw_events); + +struct power_pmu *ppmu; + +/* + * Normally, to ignore kernel events we set the FCS (freeze events + * in supervisor mode) bit in MMCR0, but if the kernel runs with the + * hypervisor bit set in the MSR, or if we are running on a processor + * where the hypervisor bit is forced to 1 (as on Apple G5 processors), + * then we need to use the FCHV bit to ignore kernel events. + */ +static unsigned int freeze_events_kernel = MMCR0_FCS; + +/* + * 32-bit doesn't have MMCRA but does have an MMCR2, + * and a few other names are different. + */ +#ifdef CONFIG_PPC32 + +#define MMCR0_FCHV 0 +#define MMCR0_PMCjCE MMCR0_PMCnCE + +#define SPRN_MMCRA SPRN_MMCR2 +#define MMCRA_SAMPLE_ENABLE 0 + +static inline unsigned long perf_ip_adjust(struct pt_regs *regs) +{ + return 0; +} +static inline void perf_get_data_addr(struct pt_regs *regs, u64 *addrp) { } +static inline u32 perf_get_misc_flags(struct pt_regs *regs) +{ + return 0; +} +static inline void perf_read_regs(struct pt_regs *regs) { } +static inline int perf_intr_is_nmi(struct pt_regs *regs) +{ + return 0; +} + +#endif /* CONFIG_PPC32 */ + +/* + * Things that are specific to 64-bit implementations. + */ +#ifdef CONFIG_PPC64 + +static inline unsigned long perf_ip_adjust(struct pt_regs *regs) +{ + unsigned long mmcra = regs->dsisr; + + if ((mmcra & MMCRA_SAMPLE_ENABLE) && !(ppmu->flags & PPMU_ALT_SIPR)) { + unsigned long slot = (mmcra & MMCRA_SLOT) >> MMCRA_SLOT_SHIFT; + if (slot > 1) + return 4 * (slot - 1); + } + return 0; +} + +/* + * The user wants a data address recorded. + * If we're not doing instruction sampling, give them the SDAR + * (sampled data address). If we are doing instruction sampling, then + * only give them the SDAR if it corresponds to the instruction + * pointed to by SIAR; this is indicated by the [POWER6_]MMCRA_SDSYNC + * bit in MMCRA. + */ +static inline void perf_get_data_addr(struct pt_regs *regs, u64 *addrp) +{ + unsigned long mmcra = regs->dsisr; + unsigned long sdsync = (ppmu->flags & PPMU_ALT_SIPR) ? + POWER6_MMCRA_SDSYNC : MMCRA_SDSYNC; + + if (!(mmcra & MMCRA_SAMPLE_ENABLE) || (mmcra & sdsync)) + *addrp = mfspr(SPRN_SDAR); +} + +static inline u32 perf_get_misc_flags(struct pt_regs *regs) +{ + unsigned long mmcra = regs->dsisr; + + if (TRAP(regs) != 0xf00) + return 0; /* not a PMU interrupt */ + + if (ppmu->flags & PPMU_ALT_SIPR) { + if (mmcra & POWER6_MMCRA_SIHV) + return PERF_RECORD_MISC_HYPERVISOR; + return (mmcra & POWER6_MMCRA_SIPR) ? + PERF_RECORD_MISC_USER : PERF_RECORD_MISC_KERNEL; + } + if (mmcra & MMCRA_SIHV) + return PERF_RECORD_MISC_HYPERVISOR; + return (mmcra & MMCRA_SIPR) ? PERF_RECORD_MISC_USER : + PERF_RECORD_MISC_KERNEL; +} + +/* + * Overload regs->dsisr to store MMCRA so we only need to read it once + * on each interrupt. + */ +static inline void perf_read_regs(struct pt_regs *regs) +{ + regs->dsisr = mfspr(SPRN_MMCRA); +} + +/* + * If interrupts were soft-disabled when a PMU interrupt occurs, treat + * it as an NMI. + */ +static inline int perf_intr_is_nmi(struct pt_regs *regs) +{ + return !regs->softe; +} + +#endif /* CONFIG_PPC64 */ + +static void perf_event_interrupt(struct pt_regs *regs); + +void perf_event_print_debug(void) +{ +} + +/* + * Read one performance monitor event (PMC). + */ +static unsigned long read_pmc(int idx) +{ + unsigned long val; + + switch (idx) { + case 1: + val = mfspr(SPRN_PMC1); + break; + case 2: + val = mfspr(SPRN_PMC2); + break; + case 3: + val = mfspr(SPRN_PMC3); + break; + case 4: + val = mfspr(SPRN_PMC4); + break; + case 5: + val = mfspr(SPRN_PMC5); + break; + case 6: + val = mfspr(SPRN_PMC6); + break; +#ifdef CONFIG_PPC64 + case 7: + val = mfspr(SPRN_PMC7); + break; + case 8: + val = mfspr(SPRN_PMC8); + break; +#endif /* CONFIG_PPC64 */ + default: + printk(KERN_ERR "oops trying to read PMC%d\n", idx); + val = 0; + } + return val; +} + +/* + * Write one PMC. + */ +static void write_pmc(int idx, unsigned long val) +{ + switch (idx) { + case 1: + mtspr(SPRN_PMC1, val); + break; + case 2: + mtspr(SPRN_PMC2, val); + break; + case 3: + mtspr(SPRN_PMC3, val); + break; + case 4: + mtspr(SPRN_PMC4, val); + break; + case 5: + mtspr(SPRN_PMC5, val); + break; + case 6: + mtspr(SPRN_PMC6, val); + break; +#ifdef CONFIG_PPC64 + case 7: + mtspr(SPRN_PMC7, val); + break; + case 8: + mtspr(SPRN_PMC8, val); + break; +#endif /* CONFIG_PPC64 */ + default: + printk(KERN_ERR "oops trying to write PMC%d\n", idx); + } +} + +/* + * Check if a set of events can all go on the PMU at once. + * If they can't, this will look at alternative codes for the events + * and see if any combination of alternative codes is feasible. + * The feasible set is returned in event_id[]. + */ +static int power_check_constraints(struct cpu_hw_events *cpuhw, + u64 event_id[], unsigned int cflags[], + int n_ev) +{ + unsigned long mask, value, nv; + unsigned long smasks[MAX_HWEVENTS], svalues[MAX_HWEVENTS]; + int n_alt[MAX_HWEVENTS], choice[MAX_HWEVENTS]; + int i, j; + unsigned long addf = ppmu->add_fields; + unsigned long tadd = ppmu->test_adder; + + if (n_ev > ppmu->n_event) + return -1; + + /* First see if the events will go on as-is */ + for (i = 0; i < n_ev; ++i) { + if ((cflags[i] & PPMU_LIMITED_PMC_REQD) + && !ppmu->limited_pmc_event(event_id[i])) { + ppmu->get_alternatives(event_id[i], cflags[i], + cpuhw->alternatives[i]); + event_id[i] = cpuhw->alternatives[i][0]; + } + if (ppmu->get_constraint(event_id[i], &cpuhw->amasks[i][0], + &cpuhw->avalues[i][0])) + return -1; + } + value = mask = 0; + for (i = 0; i < n_ev; ++i) { + nv = (value | cpuhw->avalues[i][0]) + + (value & cpuhw->avalues[i][0] & addf); + if ((((nv + tadd) ^ value) & mask) != 0 || + (((nv + tadd) ^ cpuhw->avalues[i][0]) & + cpuhw->amasks[i][0]) != 0) + break; + value = nv; + mask |= cpuhw->amasks[i][0]; + } + if (i == n_ev) + return 0; /* all OK */ + + /* doesn't work, gather alternatives... */ + if (!ppmu->get_alternatives) + return -1; + for (i = 0; i < n_ev; ++i) { + choice[i] = 0; + n_alt[i] = ppmu->get_alternatives(event_id[i], cflags[i], + cpuhw->alternatives[i]); + for (j = 1; j < n_alt[i]; ++j) + ppmu->get_constraint(cpuhw->alternatives[i][j], + &cpuhw->amasks[i][j], + &cpuhw->avalues[i][j]); + } + + /* enumerate all possibilities and see if any will work */ + i = 0; + j = -1; + value = mask = nv = 0; + while (i < n_ev) { + if (j >= 0) { + /* we're backtracking, restore context */ + value = svalues[i]; + mask = smasks[i]; + j = choice[i]; + } + /* + * See if any alternative k for event_id i, + * where k > j, will satisfy the constraints. + */ + while (++j < n_alt[i]) { + nv = (value | cpuhw->avalues[i][j]) + + (value & cpuhw->avalues[i][j] & addf); + if ((((nv + tadd) ^ value) & mask) == 0 && + (((nv + tadd) ^ cpuhw->avalues[i][j]) + & cpuhw->amasks[i][j]) == 0) + break; + } + if (j >= n_alt[i]) { + /* + * No feasible alternative, backtrack + * to event_id i-1 and continue enumerating its + * alternatives from where we got up to. + */ + if (--i < 0) + return -1; + } else { + /* + * Found a feasible alternative for event_id i, + * remember where we got up to with this event_id, + * go on to the next event_id, and start with + * the first alternative for it. + */ + choice[i] = j; + svalues[i] = value; + smasks[i] = mask; + value = nv; + mask |= cpuhw->amasks[i][j]; + ++i; + j = -1; + } + } + + /* OK, we have a feasible combination, tell the caller the solution */ + for (i = 0; i < n_ev; ++i) + event_id[i] = cpuhw->alternatives[i][choice[i]]; + return 0; +} + +/* + * Check if newly-added events have consistent settings for + * exclude_{user,kernel,hv} with each other and any previously + * added events. + */ +static int check_excludes(struct perf_event **ctrs, unsigned int cflags[], + int n_prev, int n_new) +{ + int eu = 0, ek = 0, eh = 0; + int i, n, first; + struct perf_event *event; + + n = n_prev + n_new; + if (n <= 1) + return 0; + + first = 1; + for (i = 0; i < n; ++i) { + if (cflags[i] & PPMU_LIMITED_PMC_OK) { + cflags[i] &= ~PPMU_LIMITED_PMC_REQD; + continue; + } + event = ctrs[i]; + if (first) { + eu = event->attr.exclude_user; + ek = event->attr.exclude_kernel; + eh = event->attr.exclude_hv; + first = 0; + } else if (event->attr.exclude_user != eu || + event->attr.exclude_kernel != ek || + event->attr.exclude_hv != eh) { + return -EAGAIN; + } + } + + if (eu || ek || eh) + for (i = 0; i < n; ++i) + if (cflags[i] & PPMU_LIMITED_PMC_OK) + cflags[i] |= PPMU_LIMITED_PMC_REQD; + + return 0; +} + +static void power_pmu_read(struct perf_event *event) +{ + s64 val, delta, prev; + + if (!event->hw.idx) + return; + /* + * Performance monitor interrupts come even when interrupts + * are soft-disabled, as long as interrupts are hard-enabled. + * Therefore we treat them like NMIs. + */ + do { + prev = atomic64_read(&event->hw.prev_count); + barrier(); + val = read_pmc(event->hw.idx); + } while (atomic64_cmpxchg(&event->hw.prev_count, prev, val) != prev); + + /* The events are only 32 bits wide */ + delta = (val - prev) & 0xfffffffful; + atomic64_add(delta, &event->count); + atomic64_sub(delta, &event->hw.period_left); +} + +/* + * On some machines, PMC5 and PMC6 can't be written, don't respect + * the freeze conditions, and don't generate interrupts. This tells + * us if `event' is using such a PMC. + */ +static int is_limited_pmc(int pmcnum) +{ + return (ppmu->flags & PPMU_LIMITED_PMC5_6) + && (pmcnum == 5 || pmcnum == 6); +} + +static void freeze_limited_events(struct cpu_hw_events *cpuhw, + unsigned long pmc5, unsigned long pmc6) +{ + struct perf_event *event; + u64 val, prev, delta; + int i; + + for (i = 0; i < cpuhw->n_limited; ++i) { + event = cpuhw->limited_event[i]; + if (!event->hw.idx) + continue; + val = (event->hw.idx == 5) ? pmc5 : pmc6; + prev = atomic64_read(&event->hw.prev_count); + event->hw.idx = 0; + delta = (val - prev) & 0xfffffffful; + atomic64_add(delta, &event->count); + } +} + +static void thaw_limited_events(struct cpu_hw_events *cpuhw, + unsigned long pmc5, unsigned long pmc6) +{ + struct perf_event *event; + u64 val; + int i; + + for (i = 0; i < cpuhw->n_limited; ++i) { + event = cpuhw->limited_event[i]; + event->hw.idx = cpuhw->limited_hwidx[i]; + val = (event->hw.idx == 5) ? pmc5 : pmc6; + atomic64_set(&event->hw.prev_count, val); + perf_event_update_userpage(event); + } +} + +/* + * Since limited events don't respect the freeze conditions, we + * have to read them immediately after freezing or unfreezing the + * other events. We try to keep the values from the limited + * events as consistent as possible by keeping the delay (in + * cycles and instructions) between freezing/unfreezing and reading + * the limited events as small and consistent as possible. + * Therefore, if any limited events are in use, we read them + * both, and always in the same order, to minimize variability, + * and do it inside the same asm that writes MMCR0. + */ +static void write_mmcr0(struct cpu_hw_events *cpuhw, unsigned long mmcr0) +{ + unsigned long pmc5, pmc6; + + if (!cpuhw->n_limited) { + mtspr(SPRN_MMCR0, mmcr0); + return; + } + + /* + * Write MMCR0, then read PMC5 and PMC6 immediately. + * To ensure we don't get a performance monitor interrupt + * between writing MMCR0 and freezing/thawing the limited + * events, we first write MMCR0 with the event overflow + * interrupt enable bits turned off. + */ + asm volatile("mtspr %3,%2; mfspr %0,%4; mfspr %1,%5" + : "=&r" (pmc5), "=&r" (pmc6) + : "r" (mmcr0 & ~(MMCR0_PMC1CE | MMCR0_PMCjCE)), + "i" (SPRN_MMCR0), + "i" (SPRN_PMC5), "i" (SPRN_PMC6)); + + if (mmcr0 & MMCR0_FC) + freeze_limited_events(cpuhw, pmc5, pmc6); + else + thaw_limited_events(cpuhw, pmc5, pmc6); + + /* + * Write the full MMCR0 including the event overflow interrupt + * enable bits, if necessary. + */ + if (mmcr0 & (MMCR0_PMC1CE | MMCR0_PMCjCE)) + mtspr(SPRN_MMCR0, mmcr0); +} + +/* + * Disable all events to prevent PMU interrupts and to allow + * events to be added or removed. + */ +void hw_perf_disable(void) +{ + struct cpu_hw_events *cpuhw; + unsigned long flags; + + if (!ppmu) + return; + local_irq_save(flags); + cpuhw = &__get_cpu_var(cpu_hw_events); + + if (!cpuhw->disabled) { + cpuhw->disabled = 1; + cpuhw->n_added = 0; + + /* + * Check if we ever enabled the PMU on this cpu. + */ + if (!cpuhw->pmcs_enabled) { + ppc_enable_pmcs(); + cpuhw->pmcs_enabled = 1; + } + + /* + * Disable instruction sampling if it was enabled + */ + if (cpuhw->mmcr[2] & MMCRA_SAMPLE_ENABLE) { + mtspr(SPRN_MMCRA, + cpuhw->mmcr[2] & ~MMCRA_SAMPLE_ENABLE); + mb(); + } + + /* + * Set the 'freeze events' bit. + * The barrier is to make sure the mtspr has been + * executed and the PMU has frozen the events + * before we return. + */ + write_mmcr0(cpuhw, mfspr(SPRN_MMCR0) | MMCR0_FC); + mb(); + } + local_irq_restore(flags); +} + +/* + * Re-enable all events if disable == 0. + * If we were previously disabled and events were added, then + * put the new config on the PMU. + */ +void hw_perf_enable(void) +{ + struct perf_event *event; + struct cpu_hw_events *cpuhw; + unsigned long flags; + long i; + unsigned long val; + s64 left; + unsigned int hwc_index[MAX_HWEVENTS]; + int n_lim; + int idx; + + if (!ppmu) + return; + local_irq_save(flags); + cpuhw = &__get_cpu_var(cpu_hw_events); + if (!cpuhw->disabled) { + local_irq_restore(flags); + return; + } + cpuhw->disabled = 0; + + /* + * If we didn't change anything, or only removed events, + * no need to recalculate MMCR* settings and reset the PMCs. + * Just reenable the PMU with the current MMCR* settings + * (possibly updated for removal of events). + */ + if (!cpuhw->n_added) { + mtspr(SPRN_MMCRA, cpuhw->mmcr[2] & ~MMCRA_SAMPLE_ENABLE); + mtspr(SPRN_MMCR1, cpuhw->mmcr[1]); + if (cpuhw->n_events == 0) + ppc_set_pmu_inuse(0); + goto out_enable; + } + + /* + * Compute MMCR* values for the new set of events + */ + if (ppmu->compute_mmcr(cpuhw->events, cpuhw->n_events, hwc_index, + cpuhw->mmcr)) { + /* shouldn't ever get here */ + printk(KERN_ERR "oops compute_mmcr failed\n"); + goto out; + } + + /* + * Add in MMCR0 freeze bits corresponding to the + * attr.exclude_* bits for the first event. + * We have already checked that all events have the + * same values for these bits as the first event. + */ + event = cpuhw->event[0]; + if (event->attr.exclude_user) + cpuhw->mmcr[0] |= MMCR0_FCP; + if (event->attr.exclude_kernel) + cpuhw->mmcr[0] |= freeze_events_kernel; + if (event->attr.exclude_hv) + cpuhw->mmcr[0] |= MMCR0_FCHV; + + /* + * Write the new configuration to MMCR* with the freeze + * bit set and set the hardware events to their initial values. + * Then unfreeze the events. + */ + ppc_set_pmu_inuse(1); + mtspr(SPRN_MMCRA, cpuhw->mmcr[2] & ~MMCRA_SAMPLE_ENABLE); + mtspr(SPRN_MMCR1, cpuhw->mmcr[1]); + mtspr(SPRN_MMCR0, (cpuhw->mmcr[0] & ~(MMCR0_PMC1CE | MMCR0_PMCjCE)) + | MMCR0_FC); + + /* + * Read off any pre-existing events that need to move + * to another PMC. + */ + for (i = 0; i < cpuhw->n_events; ++i) { + event = cpuhw->event[i]; + if (event->hw.idx && event->hw.idx != hwc_index[i] + 1) { + power_pmu_read(event); + write_pmc(event->hw.idx, 0); + event->hw.idx = 0; + } + } + + /* + * Initialize the PMCs for all the new and moved events. + */ + cpuhw->n_limited = n_lim = 0; + for (i = 0; i < cpuhw->n_events; ++i) { + event = cpuhw->event[i]; + if (event->hw.idx) + continue; + idx = hwc_index[i] + 1; + if (is_limited_pmc(idx)) { + cpuhw->limited_event[n_lim] = event; + cpuhw->limited_hwidx[n_lim] = idx; + ++n_lim; + continue; + } + val = 0; + if (event->hw.sample_period) { + left = atomic64_read(&event->hw.period_left); + if (left < 0x80000000L) + val = 0x80000000L - left; + } + atomic64_set(&event->hw.prev_count, val); + event->hw.idx = idx; + write_pmc(idx, val); + perf_event_update_userpage(event); + } + cpuhw->n_limited = n_lim; + cpuhw->mmcr[0] |= MMCR0_PMXE | MMCR0_FCECE; + + out_enable: + mb(); + write_mmcr0(cpuhw, cpuhw->mmcr[0]); + + /* + * Enable instruction sampling if necessary + */ + if (cpuhw->mmcr[2] & MMCRA_SAMPLE_ENABLE) { + mb(); + mtspr(SPRN_MMCRA, cpuhw->mmcr[2]); + } + + out: + local_irq_restore(flags); +} + +static int collect_events(struct perf_event *group, int max_count, + struct perf_event *ctrs[], u64 *events, + unsigned int *flags) +{ + int n = 0; + struct perf_event *event; + + if (!is_software_event(group)) { + if (n >= max_count) + return -1; + ctrs[n] = group; + flags[n] = group->hw.event_base; + events[n++] = group->hw.config; + } + list_for_each_entry(event, &group->sibling_list, list_entry) { + if (!is_software_event(event) && + event->state != PERF_EVENT_STATE_OFF) { + if (n >= max_count) + return -1; + ctrs[n] = event; + flags[n] = event->hw.event_base; + events[n++] = event->hw.config; + } + } + return n; +} + +static void event_sched_in(struct perf_event *event, int cpu) +{ + event->state = PERF_EVENT_STATE_ACTIVE; + event->oncpu = cpu; + event->tstamp_running += event->ctx->time - event->tstamp_stopped; + if (is_software_event(event)) + event->pmu->enable(event); +} + +/* + * Called to enable a whole group of events. + * Returns 1 if the group was enabled, or -EAGAIN if it could not be. + * Assumes the caller has disabled interrupts and has + * frozen the PMU with hw_perf_save_disable. + */ +int hw_perf_group_sched_in(struct perf_event *group_leader, + struct perf_cpu_context *cpuctx, + struct perf_event_context *ctx, int cpu) +{ + struct cpu_hw_events *cpuhw; + long i, n, n0; + struct perf_event *sub; + + if (!ppmu) + return 0; + cpuhw = &__get_cpu_var(cpu_hw_events); + n0 = cpuhw->n_events; + n = collect_events(group_leader, ppmu->n_event - n0, + &cpuhw->event[n0], &cpuhw->events[n0], + &cpuhw->flags[n0]); + if (n < 0) + return -EAGAIN; + if (check_excludes(cpuhw->event, cpuhw->flags, n0, n)) + return -EAGAIN; + i = power_check_constraints(cpuhw, cpuhw->events, cpuhw->flags, n + n0); + if (i < 0) + return -EAGAIN; + cpuhw->n_events = n0 + n; + cpuhw->n_added += n; + + /* + * OK, this group can go on; update event states etc., + * and enable any software events + */ + for (i = n0; i < n0 + n; ++i) + cpuhw->event[i]->hw.config = cpuhw->events[i]; + cpuctx->active_oncpu += n; + n = 1; + event_sched_in(group_leader, cpu); + list_for_each_entry(sub, &group_leader->sibling_list, list_entry) { + if (sub->state != PERF_EVENT_STATE_OFF) { + event_sched_in(sub, cpu); + ++n; + } + } + ctx->nr_active += n; + + return 1; +} + +/* + * Add a event to the PMU. + * If all events are not already frozen, then we disable and + * re-enable the PMU in order to get hw_perf_enable to do the + * actual work of reconfiguring the PMU. + */ +static int power_pmu_enable(struct perf_event *event) +{ + struct cpu_hw_events *cpuhw; + unsigned long flags; + int n0; + int ret = -EAGAIN; + + local_irq_save(flags); + perf_disable(); + + /* + * Add the event to the list (if there is room) + * and check whether the total set is still feasible. + */ + cpuhw = &__get_cpu_var(cpu_hw_events); + n0 = cpuhw->n_events; + if (n0 >= ppmu->n_event) + goto out; + cpuhw->event[n0] = event; + cpuhw->events[n0] = event->hw.config; + cpuhw->flags[n0] = event->hw.event_base; + if (check_excludes(cpuhw->event, cpuhw->flags, n0, 1)) + goto out; + if (power_check_constraints(cpuhw, cpuhw->events, cpuhw->flags, n0 + 1)) + goto out; + + event->hw.config = cpuhw->events[n0]; + ++cpuhw->n_events; + ++cpuhw->n_added; + + ret = 0; + out: + perf_enable(); + local_irq_restore(flags); + return ret; +} + +/* + * Remove a event from the PMU. + */ +static void power_pmu_disable(struct perf_event *event) +{ + struct cpu_hw_events *cpuhw; + long i; + unsigned long flags; + + local_irq_save(flags); + perf_disable(); + + power_pmu_read(event); + + cpuhw = &__get_cpu_var(cpu_hw_events); + for (i = 0; i < cpuhw->n_events; ++i) { + if (event == cpuhw->event[i]) { + while (++i < cpuhw->n_events) + cpuhw->event[i-1] = cpuhw->event[i]; + --cpuhw->n_events; + ppmu->disable_pmc(event->hw.idx - 1, cpuhw->mmcr); + if (event->hw.idx) { + write_pmc(event->hw.idx, 0); + event->hw.idx = 0; + } + perf_event_update_userpage(event); + break; + } + } + for (i = 0; i < cpuhw->n_limited; ++i) + if (event == cpuhw->limited_event[i]) + break; + if (i < cpuhw->n_limited) { + while (++i < cpuhw->n_limited) { + cpuhw->limited_event[i-1] = cpuhw->limited_event[i]; + cpuhw->limited_hwidx[i-1] = cpuhw->limited_hwidx[i]; + } + --cpuhw->n_limited; + } + if (cpuhw->n_events == 0) { + /* disable exceptions if no events are running */ + cpuhw->mmcr[0] &= ~(MMCR0_PMXE | MMCR0_FCECE); + } + + perf_enable(); + local_irq_restore(flags); +} + +/* + * Re-enable interrupts on a event after they were throttled + * because they were coming too fast. + */ +static void power_pmu_unthrottle(struct perf_event *event) +{ + s64 val, left; + unsigned long flags; + + if (!event->hw.idx || !event->hw.sample_period) + return; + local_irq_save(flags); + perf_disable(); + power_pmu_read(event); + left = event->hw.sample_period; + event->hw.last_period = left; + val = 0; + if (left < 0x80000000L) + val = 0x80000000L - left; + write_pmc(event->hw.idx, val); + atomic64_set(&event->hw.prev_count, val); + atomic64_set(&event->hw.period_left, left); + perf_event_update_userpage(event); + perf_enable(); + local_irq_restore(flags); +} + +struct pmu power_pmu = { + .enable = power_pmu_enable, + .disable = power_pmu_disable, + .read = power_pmu_read, + .unthrottle = power_pmu_unthrottle, +}; + +/* + * Return 1 if we might be able to put event on a limited PMC, + * or 0 if not. + * A event can only go on a limited PMC if it counts something + * that a limited PMC can count, doesn't require interrupts, and + * doesn't exclude any processor mode. + */ +static int can_go_on_limited_pmc(struct perf_event *event, u64 ev, + unsigned int flags) +{ + int n; + u64 alt[MAX_EVENT_ALTERNATIVES]; + + if (event->attr.exclude_user + || event->attr.exclude_kernel + || event->attr.exclude_hv + || event->attr.sample_period) + return 0; + + if (ppmu->limited_pmc_event(ev)) + return 1; + + /* + * The requested event_id isn't on a limited PMC already; + * see if any alternative code goes on a limited PMC. + */ + if (!ppmu->get_alternatives) + return 0; + + flags |= PPMU_LIMITED_PMC_OK | PPMU_LIMITED_PMC_REQD; + n = ppmu->get_alternatives(ev, flags, alt); + + return n > 0; +} + +/* + * Find an alternative event_id that goes on a normal PMC, if possible, + * and return the event_id code, or 0 if there is no such alternative. + * (Note: event_id code 0 is "don't count" on all machines.) + */ +static u64 normal_pmc_alternative(u64 ev, unsigned long flags) +{ + u64 alt[MAX_EVENT_ALTERNATIVES]; + int n; + + flags &= ~(PPMU_LIMITED_PMC_OK | PPMU_LIMITED_PMC_REQD); + n = ppmu->get_alternatives(ev, flags, alt); + if (!n) + return 0; + return alt[0]; +} + +/* Number of perf_events counting hardware events */ +static atomic_t num_events; +/* Used to avoid races in calling reserve/release_pmc_hardware */ +static DEFINE_MUTEX(pmc_reserve_mutex); + +/* + * Release the PMU if this is the last perf_event. + */ +static void hw_perf_event_destroy(struct perf_event *event) +{ + if (!atomic_add_unless(&num_events, -1, 1)) { + mutex_lock(&pmc_reserve_mutex); + if (atomic_dec_return(&num_events) == 0) + release_pmc_hardware(); + mutex_unlock(&pmc_reserve_mutex); + } +} + +/* + * Translate a generic cache event_id config to a raw event_id code. + */ +static int hw_perf_cache_event(u64 config, u64 *eventp) +{ + unsigned long type, op, result; + int ev; + + if (!ppmu->cache_events) + return -EINVAL; + + /* unpack config */ + type = config & 0xff; + op = (config >> 8) & 0xff; + result = (config >> 16) & 0xff; + + if (type >= PERF_COUNT_HW_CACHE_MAX || + op >= PERF_COUNT_HW_CACHE_OP_MAX || + result >= PERF_COUNT_HW_CACHE_RESULT_MAX) + return -EINVAL; + + ev = (*ppmu->cache_events)[type][op][result]; + if (ev == 0) + return -EOPNOTSUPP; + if (ev == -1) + return -EINVAL; + *eventp = ev; + return 0; +} + +const struct pmu *hw_perf_event_init(struct perf_event *event) +{ + u64 ev; + unsigned long flags; + struct perf_event *ctrs[MAX_HWEVENTS]; + u64 events[MAX_HWEVENTS]; + unsigned int cflags[MAX_HWEVENTS]; + int n; + int err; + struct cpu_hw_events *cpuhw; + + if (!ppmu) + return ERR_PTR(-ENXIO); + switch (event->attr.type) { + case PERF_TYPE_HARDWARE: + ev = event->attr.config; + if (ev >= ppmu->n_generic || ppmu->generic_events[ev] == 0) + return ERR_PTR(-EOPNOTSUPP); + ev = ppmu->generic_events[ev]; + break; + case PERF_TYPE_HW_CACHE: + err = hw_perf_cache_event(event->attr.config, &ev); + if (err) + return ERR_PTR(err); + break; + case PERF_TYPE_RAW: + ev = event->attr.config; + break; + default: + return ERR_PTR(-EINVAL); + } + event->hw.config_base = ev; + event->hw.idx = 0; + + /* + * If we are not running on a hypervisor, force the + * exclude_hv bit to 0 so that we don't care what + * the user set it to. + */ + if (!firmware_has_feature(FW_FEATURE_LPAR)) + event->attr.exclude_hv = 0; + + /* + * If this is a per-task event, then we can use + * PM_RUN_* events interchangeably with their non RUN_* + * equivalents, e.g. PM_RUN_CYC instead of PM_CYC. + * XXX we should check if the task is an idle task. + */ + flags = 0; + if (event->ctx->task) + flags |= PPMU_ONLY_COUNT_RUN; + + /* + * If this machine has limited events, check whether this + * event_id could go on a limited event. + */ + if (ppmu->flags & PPMU_LIMITED_PMC5_6) { + if (can_go_on_limited_pmc(event, ev, flags)) { + flags |= PPMU_LIMITED_PMC_OK; + } else if (ppmu->limited_pmc_event(ev)) { + /* + * The requested event_id is on a limited PMC, + * but we can't use a limited PMC; see if any + * alternative goes on a normal PMC. + */ + ev = normal_pmc_alternative(ev, flags); + if (!ev) + return ERR_PTR(-EINVAL); + } + } + + /* + * If this is in a group, check if it can go on with all the + * other hardware events in the group. We assume the event + * hasn't been linked into its leader's sibling list at this point. + */ + n = 0; + if (event->group_leader != event) { + n = collect_events(event->group_leader, ppmu->n_event - 1, + ctrs, events, cflags); + if (n < 0) + return ERR_PTR(-EINVAL); + } + events[n] = ev; + ctrs[n] = event; + cflags[n] = flags; + if (check_excludes(ctrs, cflags, n, 1)) + return ERR_PTR(-EINVAL); + + cpuhw = &get_cpu_var(cpu_hw_events); + err = power_check_constraints(cpuhw, events, cflags, n + 1); + put_cpu_var(cpu_hw_events); + if (err) + return ERR_PTR(-EINVAL); + + event->hw.config = events[n]; + event->hw.event_base = cflags[n]; + event->hw.last_period = event->hw.sample_period; + atomic64_set(&event->hw.period_left, event->hw.last_period); + + /* + * See if we need to reserve the PMU. + * If no events are currently in use, then we have to take a + * mutex to ensure that we don't race with another task doing + * reserve_pmc_hardware or release_pmc_hardware. + */ + err = 0; + if (!atomic_inc_not_zero(&num_events)) { + mutex_lock(&pmc_reserve_mutex); + if (atomic_read(&num_events) == 0 && + reserve_pmc_hardware(perf_event_interrupt)) + err = -EBUSY; + else + atomic_inc(&num_events); + mutex_unlock(&pmc_reserve_mutex); + } + event->destroy = hw_perf_event_destroy; + + if (err) + return ERR_PTR(err); + return &power_pmu; +} + +/* + * A event has overflowed; update its count and record + * things if requested. Note that interrupts are hard-disabled + * here so there is no possibility of being interrupted. + */ +static void record_and_restart(struct perf_event *event, unsigned long val, + struct pt_regs *regs, int nmi) +{ + u64 period = event->hw.sample_period; + s64 prev, delta, left; + int record = 0; + + /* we don't have to worry about interrupts here */ + prev = atomic64_read(&event->hw.prev_count); + delta = (val - prev) & 0xfffffffful; + atomic64_add(delta, &event->count); + + /* + * See if the total period for this event has expired, + * and update for the next period. + */ + val = 0; + left = atomic64_read(&event->hw.period_left) - delta; + if (period) { + if (left <= 0) { + left += period; + if (left <= 0) + left = period; + record = 1; + } + if (left < 0x80000000LL) + val = 0x80000000LL - left; + } + + /* + * Finally record data if requested. + */ + if (record) { + struct perf_sample_data data = { + .addr = 0, + .period = event->hw.last_period, + }; + + if (event->attr.sample_type & PERF_SAMPLE_ADDR) + perf_get_data_addr(regs, &data.addr); + + if (perf_event_overflow(event, nmi, &data, regs)) { + /* + * Interrupts are coming too fast - throttle them + * by setting the event to 0, so it will be + * at least 2^30 cycles until the next interrupt + * (assuming each event counts at most 2 counts + * per cycle). + */ + val = 0; + left = ~0ULL >> 1; + } + } + + write_pmc(event->hw.idx, val); + atomic64_set(&event->hw.prev_count, val); + atomic64_set(&event->hw.period_left, left); + perf_event_update_userpage(event); +} + +/* + * Called from generic code to get the misc flags (i.e. processor mode) + * for an event_id. + */ +unsigned long perf_misc_flags(struct pt_regs *regs) +{ + u32 flags = perf_get_misc_flags(regs); + + if (flags) + return flags; + return user_mode(regs) ? PERF_RECORD_MISC_USER : + PERF_RECORD_MISC_KERNEL; +} + +/* + * Called from generic code to get the instruction pointer + * for an event_id. + */ +unsigned long perf_instruction_pointer(struct pt_regs *regs) +{ + unsigned long ip; + + if (TRAP(regs) != 0xf00) + return regs->nip; /* not a PMU interrupt */ + + ip = mfspr(SPRN_SIAR) + perf_ip_adjust(regs); + return ip; +} + +/* + * Performance monitor interrupt stuff + */ +static void perf_event_interrupt(struct pt_regs *regs) +{ + int i; + struct cpu_hw_events *cpuhw = &__get_cpu_var(cpu_hw_events); + struct perf_event *event; + unsigned long val; + int found = 0; + int nmi; + + if (cpuhw->n_limited) + freeze_limited_events(cpuhw, mfspr(SPRN_PMC5), + mfspr(SPRN_PMC6)); + + perf_read_regs(regs); + + nmi = perf_intr_is_nmi(regs); + if (nmi) + nmi_enter(); + else + irq_enter(); + + for (i = 0; i < cpuhw->n_events; ++i) { + event = cpuhw->event[i]; + if (!event->hw.idx || is_limited_pmc(event->hw.idx)) + continue; + val = read_pmc(event->hw.idx); + if ((int)val < 0) { + /* event has overflowed */ + found = 1; + record_and_restart(event, val, regs, nmi); + } + } + + /* + * In case we didn't find and reset the event that caused + * the interrupt, scan all events and reset any that are + * negative, to avoid getting continual interrupts. + * Any that we processed in the previous loop will not be negative. + */ + if (!found) { + for (i = 0; i < ppmu->n_event; ++i) { + if (is_limited_pmc(i + 1)) + continue; + val = read_pmc(i + 1); + if ((int)val < 0) + write_pmc(i + 1, 0); + } + } + + /* + * Reset MMCR0 to its normal value. This will set PMXE and + * clear FC (freeze events) and PMAO (perf mon alert occurred) + * and thus allow interrupts to occur again. + * XXX might want to use MSR.PM to keep the events frozen until + * we get back out of this interrupt. + */ + write_mmcr0(cpuhw, cpuhw->mmcr[0]); + + if (nmi) + nmi_exit(); + else + irq_exit(); +} + +void hw_perf_event_setup(int cpu) +{ + struct cpu_hw_events *cpuhw = &per_cpu(cpu_hw_events, cpu); + + if (!ppmu) + return; + memset(cpuhw, 0, sizeof(*cpuhw)); + cpuhw->mmcr[0] = MMCR0_FC; +} + +int register_power_pmu(struct power_pmu *pmu) +{ + if (ppmu) + return -EBUSY; /* something's already registered */ + + ppmu = pmu; + pr_info("%s performance monitor hardware support registered\n", + pmu->name); + +#ifdef MSR_HV + /* + * Use FCHV to ignore kernel events if MSR.HV is set. + */ + if (mfmsr() & MSR_HV) + freeze_events_kernel = MMCR0_FCHV; +#endif /* CONFIG_PPC64 */ + + return 0; +} diff --git a/arch/powerpc/kernel/power4-pmu.c b/arch/powerpc/kernel/power4-pmu.c index 3c90a3d..2a361cd 100644 --- a/arch/powerpc/kernel/power4-pmu.c +++ b/arch/powerpc/kernel/power4-pmu.c @@ -9,7 +9,7 @@ * 2 of the License, or (at your option) any later version. */ #include -#include +#include #include #include #include diff --git a/arch/powerpc/kernel/power5+-pmu.c b/arch/powerpc/kernel/power5+-pmu.c index 31918af..0f4c1c7 100644 --- a/arch/powerpc/kernel/power5+-pmu.c +++ b/arch/powerpc/kernel/power5+-pmu.c @@ -9,7 +9,7 @@ * 2 of the License, or (at your option) any later version. */ #include -#include +#include #include #include #include diff --git a/arch/powerpc/kernel/power5-pmu.c b/arch/powerpc/kernel/power5-pmu.c index 867f6f6..c351b3a 100644 --- a/arch/powerpc/kernel/power5-pmu.c +++ b/arch/powerpc/kernel/power5-pmu.c @@ -9,7 +9,7 @@ * 2 of the License, or (at your option) any later version. */ #include -#include +#include #include #include #include diff --git a/arch/powerpc/kernel/power6-pmu.c b/arch/powerpc/kernel/power6-pmu.c index fa21890..ca399ba 100644 --- a/arch/powerpc/kernel/power6-pmu.c +++ b/arch/powerpc/kernel/power6-pmu.c @@ -9,7 +9,7 @@ * 2 of the License, or (at your option) any later version. */ #include -#include +#include #include #include #include diff --git a/arch/powerpc/kernel/power7-pmu.c b/arch/powerpc/kernel/power7-pmu.c index 018d094..28a4daa 100644 --- a/arch/powerpc/kernel/power7-pmu.c +++ b/arch/powerpc/kernel/power7-pmu.c @@ -9,7 +9,7 @@ * 2 of the License, or (at your option) any later version. */ #include -#include +#include #include #include #include diff --git a/arch/powerpc/kernel/ppc970-pmu.c b/arch/powerpc/kernel/ppc970-pmu.c index 75dccb7..4795744 100644 --- a/arch/powerpc/kernel/ppc970-pmu.c +++ b/arch/powerpc/kernel/ppc970-pmu.c @@ -9,7 +9,7 @@ * 2 of the License, or (at your option) any later version. */ #include -#include +#include #include #include diff --git a/arch/powerpc/kernel/time.c b/arch/powerpc/kernel/time.c index 465e498..df45a74 100644 --- a/arch/powerpc/kernel/time.c +++ b/arch/powerpc/kernel/time.c @@ -53,7 +53,7 @@ #include #include #include -#include +#include #include #include @@ -527,25 +527,25 @@ void __init iSeries_time_init_early(void) } #endif /* CONFIG_PPC_ISERIES */ -#if defined(CONFIG_PERF_COUNTERS) && defined(CONFIG_PPC32) -DEFINE_PER_CPU(u8, perf_counter_pending); +#if defined(CONFIG_PERF_EVENTS) && defined(CONFIG_PPC32) +DEFINE_PER_CPU(u8, perf_event_pending); -void set_perf_counter_pending(void) +void set_perf_event_pending(void) { - get_cpu_var(perf_counter_pending) = 1; + get_cpu_var(perf_event_pending) = 1; set_dec(1); - put_cpu_var(perf_counter_pending); + put_cpu_var(perf_event_pending); } -#define test_perf_counter_pending() __get_cpu_var(perf_counter_pending) -#define clear_perf_counter_pending() __get_cpu_var(perf_counter_pending) = 0 +#define test_perf_event_pending() __get_cpu_var(perf_event_pending) +#define clear_perf_event_pending() __get_cpu_var(perf_event_pending) = 0 -#else /* CONFIG_PERF_COUNTERS && CONFIG_PPC32 */ +#else /* CONFIG_PERF_EVENTS && CONFIG_PPC32 */ -#define test_perf_counter_pending() 0 -#define clear_perf_counter_pending() +#define test_perf_event_pending() 0 +#define clear_perf_event_pending() -#endif /* CONFIG_PERF_COUNTERS && CONFIG_PPC32 */ +#endif /* CONFIG_PERF_EVENTS && CONFIG_PPC32 */ /* * For iSeries shared processors, we have to let the hypervisor @@ -573,9 +573,9 @@ void timer_interrupt(struct pt_regs * regs) set_dec(DECREMENTER_MAX); #ifdef CONFIG_PPC32 - if (test_perf_counter_pending()) { - clear_perf_counter_pending(); - perf_counter_do_pending(); + if (test_perf_event_pending()) { + clear_perf_event_pending(); + perf_event_do_pending(); } if (atomic_read(&ppc_n_lost_interrupts) != 0) do_IRQ(regs); diff --git a/arch/powerpc/mm/fault.c b/arch/powerpc/mm/fault.c index 830bef0..e7dae82 100644 --- a/arch/powerpc/mm/fault.c +++ b/arch/powerpc/mm/fault.c @@ -29,7 +29,7 @@ #include #include #include -#include +#include #include #include @@ -171,7 +171,7 @@ int __kprobes do_page_fault(struct pt_regs *regs, unsigned long address, die("Weird page fault", regs, SIGSEGV); } - perf_swcounter_event(PERF_COUNT_SW_PAGE_FAULTS, 1, 0, regs, address); + perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, 0, regs, address); /* When running in the kernel we expect faults to occur only to * addresses in user space. All other faults represent errors in the @@ -312,7 +312,7 @@ good_area: } if (ret & VM_FAULT_MAJOR) { current->maj_flt++; - perf_swcounter_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, 0, + perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, 0, regs, address); #ifdef CONFIG_PPC_SMLPAR if (firmware_has_feature(FW_FEATURE_CMO)) { @@ -323,7 +323,7 @@ good_area: #endif } else { current->min_flt++; - perf_swcounter_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, 0, + perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, 0, regs, address); } up_read(&mm->mmap_sem); diff --git a/arch/powerpc/platforms/Kconfig.cputype b/arch/powerpc/platforms/Kconfig.cputype index 9efc8bd..e382cae 100644 --- a/arch/powerpc/platforms/Kconfig.cputype +++ b/arch/powerpc/platforms/Kconfig.cputype @@ -280,9 +280,9 @@ config PPC_HAVE_PMU_SUPPORT config PPC_PERF_CTRS def_bool y - depends on PERF_COUNTERS && PPC_HAVE_PMU_SUPPORT + depends on PERF_EVENTS && PPC_HAVE_PMU_SUPPORT help - This enables the powerpc-specific perf_counter back-end. + This enables the powerpc-specific perf_event back-end. config SMP depends on PPC_BOOK3S || PPC_BOOK3E || FSL_BOOKE diff --git a/arch/s390/Kconfig b/arch/s390/Kconfig index 1c866ef..43c0aca 100644 --- a/arch/s390/Kconfig +++ b/arch/s390/Kconfig @@ -94,7 +94,7 @@ config S390 select HAVE_KVM if 64BIT select HAVE_ARCH_TRACEHOOK select INIT_ALL_POSSIBLE - select HAVE_PERF_COUNTERS + select HAVE_PERF_EVENTS config SCHED_OMIT_FRAME_POINTER bool diff --git a/arch/s390/include/asm/perf_counter.h b/arch/s390/include/asm/perf_counter.h deleted file mode 100644 index 7015188..0000000 --- a/arch/s390/include/asm/perf_counter.h +++ /dev/null @@ -1,10 +0,0 @@ -/* - * Performance counter support - s390 specific definitions. - * - * Copyright 2009 Martin Schwidefsky, IBM Corporation. - */ - -static inline void set_perf_counter_pending(void) {} -static inline void clear_perf_counter_pending(void) {} - -#define PERF_COUNTER_INDEX_OFFSET 0 diff --git a/arch/s390/include/asm/perf_event.h b/arch/s390/include/asm/perf_event.h new file mode 100644 index 0000000..3840cbe --- /dev/null +++ b/arch/s390/include/asm/perf_event.h @@ -0,0 +1,10 @@ +/* + * Performance event support - s390 specific definitions. + * + * Copyright 2009 Martin Schwidefsky, IBM Corporation. + */ + +static inline void set_perf_event_pending(void) {} +static inline void clear_perf_event_pending(void) {} + +#define PERF_EVENT_INDEX_OFFSET 0 diff --git a/arch/s390/include/asm/unistd.h b/arch/s390/include/asm/unistd.h index c80602d..cb5232d 100644 --- a/arch/s390/include/asm/unistd.h +++ b/arch/s390/include/asm/unistd.h @@ -268,7 +268,7 @@ #define __NR_preadv 328 #define __NR_pwritev 329 #define __NR_rt_tgsigqueueinfo 330 -#define __NR_perf_counter_open 331 +#define __NR_perf_event_open 331 #define NR_syscalls 332 /* diff --git a/arch/s390/kernel/compat_wrapper.S b/arch/s390/kernel/compat_wrapper.S index 88a83366..6247900 100644 --- a/arch/s390/kernel/compat_wrapper.S +++ b/arch/s390/kernel/compat_wrapper.S @@ -1832,11 +1832,11 @@ compat_sys_rt_tgsigqueueinfo_wrapper: llgtr %r5,%r5 # struct compat_siginfo * jg compat_sys_rt_tgsigqueueinfo_wrapper # branch to system call - .globl sys_perf_counter_open_wrapper -sys_perf_counter_open_wrapper: - llgtr %r2,%r2 # const struct perf_counter_attr * + .globl sys_perf_event_open_wrapper +sys_perf_event_open_wrapper: + llgtr %r2,%r2 # const struct perf_event_attr * lgfr %r3,%r3 # pid_t lgfr %r4,%r4 # int lgfr %r5,%r5 # int llgfr %r6,%r6 # unsigned long - jg sys_perf_counter_open # branch to system call + jg sys_perf_event_open # branch to system call diff --git a/arch/s390/kernel/syscalls.S b/arch/s390/kernel/syscalls.S index ad1acd2..0b50836 100644 --- a/arch/s390/kernel/syscalls.S +++ b/arch/s390/kernel/syscalls.S @@ -339,4 +339,4 @@ SYSCALL(sys_epoll_create1,sys_epoll_create1,sys_epoll_create1_wrapper) SYSCALL(sys_preadv,sys_preadv,compat_sys_preadv_wrapper) SYSCALL(sys_pwritev,sys_pwritev,compat_sys_pwritev_wrapper) SYSCALL(sys_rt_tgsigqueueinfo,sys_rt_tgsigqueueinfo,compat_sys_rt_tgsigqueueinfo_wrapper) /* 330 */ -SYSCALL(sys_perf_counter_open,sys_perf_counter_open,sys_perf_counter_open_wrapper) +SYSCALL(sys_perf_event_open,sys_perf_event_open,sys_perf_event_open_wrapper) diff --git a/arch/s390/mm/fault.c b/arch/s390/mm/fault.c index 1abbadd..6d50746 100644 --- a/arch/s390/mm/fault.c +++ b/arch/s390/mm/fault.c @@ -10,7 +10,7 @@ * Copyright (C) 1995 Linus Torvalds */ -#include +#include #include #include #include @@ -306,7 +306,7 @@ do_exception(struct pt_regs *regs, unsigned long error_code, int write) * interrupts again and then search the VMAs */ local_irq_enable(); - perf_swcounter_event(PERF_COUNT_SW_PAGE_FAULTS, 1, 0, regs, address); + perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, 0, regs, address); down_read(&mm->mmap_sem); si_code = SEGV_MAPERR; @@ -366,11 +366,11 @@ good_area: } if (fault & VM_FAULT_MAJOR) { tsk->maj_flt++; - perf_swcounter_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, 0, + perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, 0, regs, address); } else { tsk->min_flt++; - perf_swcounter_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, 0, + perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, 0, regs, address); } up_read(&mm->mmap_sem); diff --git a/arch/sh/Kconfig b/arch/sh/Kconfig index 4df3570..b940424 100644 --- a/arch/sh/Kconfig +++ b/arch/sh/Kconfig @@ -16,7 +16,7 @@ config SUPERH select HAVE_IOREMAP_PROT if MMU select HAVE_ARCH_TRACEHOOK select HAVE_DMA_API_DEBUG - select HAVE_PERF_COUNTERS + select HAVE_PERF_EVENTS select HAVE_KERNEL_GZIP select HAVE_KERNEL_BZIP2 select HAVE_KERNEL_LZMA diff --git a/arch/sh/include/asm/perf_counter.h b/arch/sh/include/asm/perf_counter.h deleted file mode 100644 index d8e6bb9..0000000 --- a/arch/sh/include/asm/perf_counter.h +++ /dev/null @@ -1,9 +0,0 @@ -#ifndef __ASM_SH_PERF_COUNTER_H -#define __ASM_SH_PERF_COUNTER_H - -/* SH only supports software counters through this interface. */ -static inline void set_perf_counter_pending(void) {} - -#define PERF_COUNTER_INDEX_OFFSET 0 - -#endif /* __ASM_SH_PERF_COUNTER_H */ diff --git a/arch/sh/include/asm/perf_event.h b/arch/sh/include/asm/perf_event.h new file mode 100644 index 0000000..11a3022 --- /dev/null +++ b/arch/sh/include/asm/perf_event.h @@ -0,0 +1,9 @@ +#ifndef __ASM_SH_PERF_EVENT_H +#define __ASM_SH_PERF_EVENT_H + +/* SH only supports software events through this interface. */ +static inline void set_perf_event_pending(void) {} + +#define PERF_EVENT_INDEX_OFFSET 0 + +#endif /* __ASM_SH_PERF_EVENT_H */ diff --git a/arch/sh/include/asm/unistd_32.h b/arch/sh/include/asm/unistd_32.h index 925dd40..f3fd1b9 100644 --- a/arch/sh/include/asm/unistd_32.h +++ b/arch/sh/include/asm/unistd_32.h @@ -344,7 +344,7 @@ #define __NR_preadv 333 #define __NR_pwritev 334 #define __NR_rt_tgsigqueueinfo 335 -#define __NR_perf_counter_open 336 +#define __NR_perf_event_open 336 #define NR_syscalls 337 diff --git a/arch/sh/include/asm/unistd_64.h b/arch/sh/include/asm/unistd_64.h index 2b84bc9..343ce8f 100644 --- a/arch/sh/include/asm/unistd_64.h +++ b/arch/sh/include/asm/unistd_64.h @@ -384,7 +384,7 @@ #define __NR_preadv 361 #define __NR_pwritev 362 #define __NR_rt_tgsigqueueinfo 363 -#define __NR_perf_counter_open 364 +#define __NR_perf_event_open 364 #ifdef __KERNEL__ diff --git a/arch/sh/kernel/syscalls_32.S b/arch/sh/kernel/syscalls_32.S index 16ba225..19fd11d 100644 --- a/arch/sh/kernel/syscalls_32.S +++ b/arch/sh/kernel/syscalls_32.S @@ -352,4 +352,4 @@ ENTRY(sys_call_table) .long sys_preadv .long sys_pwritev .long sys_rt_tgsigqueueinfo /* 335 */ - .long sys_perf_counter_open + .long sys_perf_event_open diff --git a/arch/sh/kernel/syscalls_64.S b/arch/sh/kernel/syscalls_64.S index af6fb74..5bfde6c 100644 --- a/arch/sh/kernel/syscalls_64.S +++ b/arch/sh/kernel/syscalls_64.S @@ -390,4 +390,4 @@ sys_call_table: .long sys_preadv .long sys_pwritev .long sys_rt_tgsigqueueinfo - .long sys_perf_counter_open + .long sys_perf_event_open diff --git a/arch/sh/mm/fault_32.c b/arch/sh/mm/fault_32.c index 781b413..4753010 100644 --- a/arch/sh/mm/fault_32.c +++ b/arch/sh/mm/fault_32.c @@ -15,7 +15,7 @@ #include #include #include -#include +#include #include #include #include @@ -157,7 +157,7 @@ asmlinkage void __kprobes do_page_fault(struct pt_regs *regs, if ((regs->sr & SR_IMASK) != SR_IMASK) local_irq_enable(); - perf_swcounter_event(PERF_COUNT_SW_PAGE_FAULTS, 1, 0, regs, address); + perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, 0, regs, address); /* * If we're in an interrupt, have no user context or are running @@ -208,11 +208,11 @@ survive: } if (fault & VM_FAULT_MAJOR) { tsk->maj_flt++; - perf_swcounter_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, 0, + perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, 0, regs, address); } else { tsk->min_flt++; - perf_swcounter_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, 0, + perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, 0, regs, address); } diff --git a/arch/sh/mm/tlbflush_64.c b/arch/sh/mm/tlbflush_64.c index 2dcc485..de0b0e88 100644 --- a/arch/sh/mm/tlbflush_64.c +++ b/arch/sh/mm/tlbflush_64.c @@ -20,7 +20,7 @@ #include #include #include -#include +#include #include #include #include @@ -116,7 +116,7 @@ asmlinkage void do_page_fault(struct pt_regs *regs, unsigned long writeaccess, /* Not an IO address, so reenable interrupts */ local_irq_enable(); - perf_swcounter_event(PERF_COUNT_SW_PAGE_FAULTS, 1, 0, regs, address); + perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, 0, regs, address); /* * If we're in an interrupt or have no user @@ -201,11 +201,11 @@ survive: if (fault & VM_FAULT_MAJOR) { tsk->maj_flt++; - perf_swcounter_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, 0, + perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, 0, regs, address); } else { tsk->min_flt++; - perf_swcounter_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, 0, + perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, 0, regs, address); } diff --git a/arch/sparc/Kconfig b/arch/sparc/Kconfig index 86b8234..97fca46 100644 --- a/arch/sparc/Kconfig +++ b/arch/sparc/Kconfig @@ -25,7 +25,7 @@ config SPARC select ARCH_WANT_OPTIONAL_GPIOLIB select RTC_CLASS select RTC_DRV_M48T59 - select HAVE_PERF_COUNTERS + select HAVE_PERF_EVENTS select HAVE_DMA_ATTRS select HAVE_DMA_API_DEBUG @@ -47,7 +47,7 @@ config SPARC64 select RTC_DRV_BQ4802 select RTC_DRV_SUN4V select RTC_DRV_STARFIRE - select HAVE_PERF_COUNTERS + select HAVE_PERF_EVENTS config ARCH_DEFCONFIG string diff --git a/arch/sparc/include/asm/perf_counter.h b/arch/sparc/include/asm/perf_counter.h deleted file mode 100644 index 5d7a8ca..0000000 --- a/arch/sparc/include/asm/perf_counter.h +++ /dev/null @@ -1,14 +0,0 @@ -#ifndef __ASM_SPARC_PERF_COUNTER_H -#define __ASM_SPARC_PERF_COUNTER_H - -extern void set_perf_counter_pending(void); - -#define PERF_COUNTER_INDEX_OFFSET 0 - -#ifdef CONFIG_PERF_COUNTERS -extern void init_hw_perf_counters(void); -#else -static inline void init_hw_perf_counters(void) { } -#endif - -#endif diff --git a/arch/sparc/include/asm/perf_event.h b/arch/sparc/include/asm/perf_event.h new file mode 100644 index 0000000..7e26698 --- /dev/null +++ b/arch/sparc/include/asm/perf_event.h @@ -0,0 +1,14 @@ +#ifndef __ASM_SPARC_PERF_EVENT_H +#define __ASM_SPARC_PERF_EVENT_H + +extern void set_perf_event_pending(void); + +#define PERF_EVENT_INDEX_OFFSET 0 + +#ifdef CONFIG_PERF_EVENTS +extern void init_hw_perf_events(void); +#else +static inline void init_hw_perf_events(void) { } +#endif + +#endif diff --git a/arch/sparc/include/asm/unistd.h b/arch/sparc/include/asm/unistd.h index 706df66..42f2316 100644 --- a/arch/sparc/include/asm/unistd.h +++ b/arch/sparc/include/asm/unistd.h @@ -395,7 +395,7 @@ #define __NR_preadv 324 #define __NR_pwritev 325 #define __NR_rt_tgsigqueueinfo 326 -#define __NR_perf_counter_open 327 +#define __NR_perf_event_open 327 #define NR_SYSCALLS 328 diff --git a/arch/sparc/kernel/Makefile b/arch/sparc/kernel/Makefile index 247cc62..3a048fa 100644 --- a/arch/sparc/kernel/Makefile +++ b/arch/sparc/kernel/Makefile @@ -104,5 +104,5 @@ obj-$(CONFIG_AUDIT) += audit.o audit--$(CONFIG_AUDIT) := compat_audit.o obj-$(CONFIG_COMPAT) += $(audit--y) -pc--$(CONFIG_PERF_COUNTERS) := perf_counter.o +pc--$(CONFIG_PERF_EVENTS) := perf_event.o obj-$(CONFIG_SPARC64) += $(pc--y) diff --git a/arch/sparc/kernel/nmi.c b/arch/sparc/kernel/nmi.c index 378eb53..b129611 100644 --- a/arch/sparc/kernel/nmi.c +++ b/arch/sparc/kernel/nmi.c @@ -19,7 +19,7 @@ #include #include -#include +#include #include #include #include @@ -265,7 +265,7 @@ int __init nmi_init(void) } } if (!err) - init_hw_perf_counters(); + init_hw_perf_events(); return err; } diff --git a/arch/sparc/kernel/pcr.c b/arch/sparc/kernel/pcr.c index 68ff001..2d94e7a 100644 --- a/arch/sparc/kernel/pcr.c +++ b/arch/sparc/kernel/pcr.c @@ -7,7 +7,7 @@ #include #include -#include +#include #include #include @@ -15,7 +15,7 @@ /* This code is shared between various users of the performance * counters. Users will be oprofile, pseudo-NMI watchdog, and the - * perf_counter support layer. + * perf_event support layer. */ #define PCR_SUN4U_ENABLE (PCR_PIC_PRIV | PCR_STRACE | PCR_UTRACE) @@ -42,14 +42,14 @@ void deferred_pcr_work_irq(int irq, struct pt_regs *regs) old_regs = set_irq_regs(regs); irq_enter(); -#ifdef CONFIG_PERF_COUNTERS - perf_counter_do_pending(); +#ifdef CONFIG_PERF_EVENTS + perf_event_do_pending(); #endif irq_exit(); set_irq_regs(old_regs); } -void set_perf_counter_pending(void) +void set_perf_event_pending(void) { set_softint(1 << PIL_DEFERRED_PCR_WORK); } diff --git a/arch/sparc/kernel/perf_counter.c b/arch/sparc/kernel/perf_counter.c deleted file mode 100644 index b1265ce..0000000 --- a/arch/sparc/kernel/perf_counter.c +++ /dev/null @@ -1,556 +0,0 @@ -/* Performance counter support for sparc64. - * - * Copyright (C) 2009 David S. Miller - * - * This code is based almost entirely upon the x86 perf counter - * code, which is: - * - * Copyright (C) 2008 Thomas Gleixner - * Copyright (C) 2008-2009 Red Hat, Inc., Ingo Molnar - * Copyright (C) 2009 Jaswinder Singh Rajput - * Copyright (C) 2009 Advanced Micro Devices, Inc., Robert Richter - * Copyright (C) 2008-2009 Red Hat, Inc., Peter Zijlstra - */ - -#include -#include -#include -#include -#include - -#include -#include -#include -#include - -/* Sparc64 chips have two performance counters, 32-bits each, with - * overflow interrupts generated on transition from 0xffffffff to 0. - * The counters are accessed in one go using a 64-bit register. - * - * Both counters are controlled using a single control register. The - * only way to stop all sampling is to clear all of the context (user, - * supervisor, hypervisor) sampling enable bits. But these bits apply - * to both counters, thus the two counters can't be enabled/disabled - * individually. - * - * The control register has two event fields, one for each of the two - * counters. It's thus nearly impossible to have one counter going - * while keeping the other one stopped. Therefore it is possible to - * get overflow interrupts for counters not currently "in use" and - * that condition must be checked in the overflow interrupt handler. - * - * So we use a hack, in that we program inactive counters with the - * "sw_count0" and "sw_count1" events. These count how many times - * the instruction "sethi %hi(0xfc000), %g0" is executed. It's an - * unusual way to encode a NOP and therefore will not trigger in - * normal code. - */ - -#define MAX_HWCOUNTERS 2 -#define MAX_PERIOD ((1UL << 32) - 1) - -#define PIC_UPPER_INDEX 0 -#define PIC_LOWER_INDEX 1 - -struct cpu_hw_counters { - struct perf_counter *counters[MAX_HWCOUNTERS]; - unsigned long used_mask[BITS_TO_LONGS(MAX_HWCOUNTERS)]; - unsigned long active_mask[BITS_TO_LONGS(MAX_HWCOUNTERS)]; - int enabled; -}; -DEFINE_PER_CPU(struct cpu_hw_counters, cpu_hw_counters) = { .enabled = 1, }; - -struct perf_event_map { - u16 encoding; - u8 pic_mask; -#define PIC_NONE 0x00 -#define PIC_UPPER 0x01 -#define PIC_LOWER 0x02 -}; - -struct sparc_pmu { - const struct perf_event_map *(*event_map)(int); - int max_events; - int upper_shift; - int lower_shift; - int event_mask; - int hv_bit; - int irq_bit; - int upper_nop; - int lower_nop; -}; - -static const struct perf_event_map ultra3i_perfmon_event_map[] = { - [PERF_COUNT_HW_CPU_CYCLES] = { 0x0000, PIC_UPPER | PIC_LOWER }, - [PERF_COUNT_HW_INSTRUCTIONS] = { 0x0001, PIC_UPPER | PIC_LOWER }, - [PERF_COUNT_HW_CACHE_REFERENCES] = { 0x0009, PIC_LOWER }, - [PERF_COUNT_HW_CACHE_MISSES] = { 0x0009, PIC_UPPER }, -}; - -static const struct perf_event_map *ultra3i_event_map(int event) -{ - return &ultra3i_perfmon_event_map[event]; -} - -static const struct sparc_pmu ultra3i_pmu = { - .event_map = ultra3i_event_map, - .max_events = ARRAY_SIZE(ultra3i_perfmon_event_map), - .upper_shift = 11, - .lower_shift = 4, - .event_mask = 0x3f, - .upper_nop = 0x1c, - .lower_nop = 0x14, -}; - -static const struct perf_event_map niagara2_perfmon_event_map[] = { - [PERF_COUNT_HW_CPU_CYCLES] = { 0x02ff, PIC_UPPER | PIC_LOWER }, - [PERF_COUNT_HW_INSTRUCTIONS] = { 0x02ff, PIC_UPPER | PIC_LOWER }, - [PERF_COUNT_HW_CACHE_REFERENCES] = { 0x0208, PIC_UPPER | PIC_LOWER }, - [PERF_COUNT_HW_CACHE_MISSES] = { 0x0302, PIC_UPPER | PIC_LOWER }, - [PERF_COUNT_HW_BRANCH_INSTRUCTIONS] = { 0x0201, PIC_UPPER | PIC_LOWER }, - [PERF_COUNT_HW_BRANCH_MISSES] = { 0x0202, PIC_UPPER | PIC_LOWER }, -}; - -static const struct perf_event_map *niagara2_event_map(int event) -{ - return &niagara2_perfmon_event_map[event]; -} - -static const struct sparc_pmu niagara2_pmu = { - .event_map = niagara2_event_map, - .max_events = ARRAY_SIZE(niagara2_perfmon_event_map), - .upper_shift = 19, - .lower_shift = 6, - .event_mask = 0xfff, - .hv_bit = 0x8, - .irq_bit = 0x03, - .upper_nop = 0x220, - .lower_nop = 0x220, -}; - -static const struct sparc_pmu *sparc_pmu __read_mostly; - -static u64 event_encoding(u64 event, int idx) -{ - if (idx == PIC_UPPER_INDEX) - event <<= sparc_pmu->upper_shift; - else - event <<= sparc_pmu->lower_shift; - return event; -} - -static u64 mask_for_index(int idx) -{ - return event_encoding(sparc_pmu->event_mask, idx); -} - -static u64 nop_for_index(int idx) -{ - return event_encoding(idx == PIC_UPPER_INDEX ? - sparc_pmu->upper_nop : - sparc_pmu->lower_nop, idx); -} - -static inline void sparc_pmu_enable_counter(struct hw_perf_counter *hwc, - int idx) -{ - u64 val, mask = mask_for_index(idx); - - val = pcr_ops->read(); - pcr_ops->write((val & ~mask) | hwc->config); -} - -static inline void sparc_pmu_disable_counter(struct hw_perf_counter *hwc, - int idx) -{ - u64 mask = mask_for_index(idx); - u64 nop = nop_for_index(idx); - u64 val = pcr_ops->read(); - - pcr_ops->write((val & ~mask) | nop); -} - -void hw_perf_enable(void) -{ - struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters); - u64 val; - int i; - - if (cpuc->enabled) - return; - - cpuc->enabled = 1; - barrier(); - - val = pcr_ops->read(); - - for (i = 0; i < MAX_HWCOUNTERS; i++) { - struct perf_counter *cp = cpuc->counters[i]; - struct hw_perf_counter *hwc; - - if (!cp) - continue; - hwc = &cp->hw; - val |= hwc->config_base; - } - - pcr_ops->write(val); -} - -void hw_perf_disable(void) -{ - struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters); - u64 val; - - if (!cpuc->enabled) - return; - - cpuc->enabled = 0; - - val = pcr_ops->read(); - val &= ~(PCR_UTRACE | PCR_STRACE | - sparc_pmu->hv_bit | sparc_pmu->irq_bit); - pcr_ops->write(val); -} - -static u32 read_pmc(int idx) -{ - u64 val; - - read_pic(val); - if (idx == PIC_UPPER_INDEX) - val >>= 32; - - return val & 0xffffffff; -} - -static void write_pmc(int idx, u64 val) -{ - u64 shift, mask, pic; - - shift = 0; - if (idx == PIC_UPPER_INDEX) - shift = 32; - - mask = ((u64) 0xffffffff) << shift; - val <<= shift; - - read_pic(pic); - pic &= ~mask; - pic |= val; - write_pic(pic); -} - -static int sparc_perf_counter_set_period(struct perf_counter *counter, - struct hw_perf_counter *hwc, int idx) -{ - s64 left = atomic64_read(&hwc->period_left); - s64 period = hwc->sample_period; - int ret = 0; - - if (unlikely(left <= -period)) { - left = period; - atomic64_set(&hwc->period_left, left); - hwc->last_period = period; - ret = 1; - } - - if (unlikely(left <= 0)) { - left += period; - atomic64_set(&hwc->period_left, left); - hwc->last_period = period; - ret = 1; - } - if (left > MAX_PERIOD) - left = MAX_PERIOD; - - atomic64_set(&hwc->prev_count, (u64)-left); - - write_pmc(idx, (u64)(-left) & 0xffffffff); - - perf_counter_update_userpage(counter); - - return ret; -} - -static int sparc_pmu_enable(struct perf_counter *counter) -{ - struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters); - struct hw_perf_counter *hwc = &counter->hw; - int idx = hwc->idx; - - if (test_and_set_bit(idx, cpuc->used_mask)) - return -EAGAIN; - - sparc_pmu_disable_counter(hwc, idx); - - cpuc->counters[idx] = counter; - set_bit(idx, cpuc->active_mask); - - sparc_perf_counter_set_period(counter, hwc, idx); - sparc_pmu_enable_counter(hwc, idx); - perf_counter_update_userpage(counter); - return 0; -} - -static u64 sparc_perf_counter_update(struct perf_counter *counter, - struct hw_perf_counter *hwc, int idx) -{ - int shift = 64 - 32; - u64 prev_raw_count, new_raw_count; - s64 delta; - -again: - prev_raw_count = atomic64_read(&hwc->prev_count); - new_raw_count = read_pmc(idx); - - if (atomic64_cmpxchg(&hwc->prev_count, prev_raw_count, - new_raw_count) != prev_raw_count) - goto again; - - delta = (new_raw_count << shift) - (prev_raw_count << shift); - delta >>= shift; - - atomic64_add(delta, &counter->count); - atomic64_sub(delta, &hwc->period_left); - - return new_raw_count; -} - -static void sparc_pmu_disable(struct perf_counter *counter) -{ - struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters); - struct hw_perf_counter *hwc = &counter->hw; - int idx = hwc->idx; - - clear_bit(idx, cpuc->active_mask); - sparc_pmu_disable_counter(hwc, idx); - - barrier(); - - sparc_perf_counter_update(counter, hwc, idx); - cpuc->counters[idx] = NULL; - clear_bit(idx, cpuc->used_mask); - - perf_counter_update_userpage(counter); -} - -static void sparc_pmu_read(struct perf_counter *counter) -{ - struct hw_perf_counter *hwc = &counter->hw; - sparc_perf_counter_update(counter, hwc, hwc->idx); -} - -static void sparc_pmu_unthrottle(struct perf_counter *counter) -{ - struct hw_perf_counter *hwc = &counter->hw; - sparc_pmu_enable_counter(hwc, hwc->idx); -} - -static atomic_t active_counters = ATOMIC_INIT(0); -static DEFINE_MUTEX(pmc_grab_mutex); - -void perf_counter_grab_pmc(void) -{ - if (atomic_inc_not_zero(&active_counters)) - return; - - mutex_lock(&pmc_grab_mutex); - if (atomic_read(&active_counters) == 0) { - if (atomic_read(&nmi_active) > 0) { - on_each_cpu(stop_nmi_watchdog, NULL, 1); - BUG_ON(atomic_read(&nmi_active) != 0); - } - atomic_inc(&active_counters); - } - mutex_unlock(&pmc_grab_mutex); -} - -void perf_counter_release_pmc(void) -{ - if (atomic_dec_and_mutex_lock(&active_counters, &pmc_grab_mutex)) { - if (atomic_read(&nmi_active) == 0) - on_each_cpu(start_nmi_watchdog, NULL, 1); - mutex_unlock(&pmc_grab_mutex); - } -} - -static void hw_perf_counter_destroy(struct perf_counter *counter) -{ - perf_counter_release_pmc(); -} - -static int __hw_perf_counter_init(struct perf_counter *counter) -{ - struct perf_counter_attr *attr = &counter->attr; - struct hw_perf_counter *hwc = &counter->hw; - const struct perf_event_map *pmap; - u64 enc; - - if (atomic_read(&nmi_active) < 0) - return -ENODEV; - - if (attr->type != PERF_TYPE_HARDWARE) - return -EOPNOTSUPP; - - if (attr->config >= sparc_pmu->max_events) - return -EINVAL; - - perf_counter_grab_pmc(); - counter->destroy = hw_perf_counter_destroy; - - /* We save the enable bits in the config_base. So to - * turn off sampling just write 'config', and to enable - * things write 'config | config_base'. - */ - hwc->config_base = sparc_pmu->irq_bit; - if (!attr->exclude_user) - hwc->config_base |= PCR_UTRACE; - if (!attr->exclude_kernel) - hwc->config_base |= PCR_STRACE; - if (!attr->exclude_hv) - hwc->config_base |= sparc_pmu->hv_bit; - - if (!hwc->sample_period) { - hwc->sample_period = MAX_PERIOD; - hwc->last_period = hwc->sample_period; - atomic64_set(&hwc->period_left, hwc->sample_period); - } - - pmap = sparc_pmu->event_map(attr->config); - - enc = pmap->encoding; - if (pmap->pic_mask & PIC_UPPER) { - hwc->idx = PIC_UPPER_INDEX; - enc <<= sparc_pmu->upper_shift; - } else { - hwc->idx = PIC_LOWER_INDEX; - enc <<= sparc_pmu->lower_shift; - } - - hwc->config |= enc; - return 0; -} - -static const struct pmu pmu = { - .enable = sparc_pmu_enable, - .disable = sparc_pmu_disable, - .read = sparc_pmu_read, - .unthrottle = sparc_pmu_unthrottle, -}; - -const struct pmu *hw_perf_counter_init(struct perf_counter *counter) -{ - int err = __hw_perf_counter_init(counter); - - if (err) - return ERR_PTR(err); - return &pmu; -} - -void perf_counter_print_debug(void) -{ - unsigned long flags; - u64 pcr, pic; - int cpu; - - if (!sparc_pmu) - return; - - local_irq_save(flags); - - cpu = smp_processor_id(); - - pcr = pcr_ops->read(); - read_pic(pic); - - pr_info("\n"); - pr_info("CPU#%d: PCR[%016llx] PIC[%016llx]\n", - cpu, pcr, pic); - - local_irq_restore(flags); -} - -static int __kprobes perf_counter_nmi_handler(struct notifier_block *self, - unsigned long cmd, void *__args) -{ - struct die_args *args = __args; - struct perf_sample_data data; - struct cpu_hw_counters *cpuc; - struct pt_regs *regs; - int idx; - - if (!atomic_read(&active_counters)) - return NOTIFY_DONE; - - switch (cmd) { - case DIE_NMI: - break; - - default: - return NOTIFY_DONE; - } - - regs = args->regs; - - data.addr = 0; - - cpuc = &__get_cpu_var(cpu_hw_counters); - for (idx = 0; idx < MAX_HWCOUNTERS; idx++) { - struct perf_counter *counter = cpuc->counters[idx]; - struct hw_perf_counter *hwc; - u64 val; - - if (!test_bit(idx, cpuc->active_mask)) - continue; - hwc = &counter->hw; - val = sparc_perf_counter_update(counter, hwc, idx); - if (val & (1ULL << 31)) - continue; - - data.period = counter->hw.last_period; - if (!sparc_perf_counter_set_period(counter, hwc, idx)) - continue; - - if (perf_counter_overflow(counter, 1, &data, regs)) - sparc_pmu_disable_counter(hwc, idx); - } - - return NOTIFY_STOP; -} - -static __read_mostly struct notifier_block perf_counter_nmi_notifier = { - .notifier_call = perf_counter_nmi_handler, -}; - -static bool __init supported_pmu(void) -{ - if (!strcmp(sparc_pmu_type, "ultra3i")) { - sparc_pmu = &ultra3i_pmu; - return true; - } - if (!strcmp(sparc_pmu_type, "niagara2")) { - sparc_pmu = &niagara2_pmu; - return true; - } - return false; -} - -void __init init_hw_perf_counters(void) -{ - pr_info("Performance counters: "); - - if (!supported_pmu()) { - pr_cont("No support for PMU type '%s'\n", sparc_pmu_type); - return; - } - - pr_cont("Supported PMU type is '%s'\n", sparc_pmu_type); - - /* All sparc64 PMUs currently have 2 counters. But this simple - * driver only supports one active counter at a time. - */ - perf_max_counters = 1; - - register_die_notifier(&perf_counter_nmi_notifier); -} diff --git a/arch/sparc/kernel/perf_event.c b/arch/sparc/kernel/perf_event.c new file mode 100644 index 0000000..2d6a1b1 --- /dev/null +++ b/arch/sparc/kernel/perf_event.c @@ -0,0 +1,556 @@ +/* Performance event support for sparc64. + * + * Copyright (C) 2009 David S. Miller + * + * This code is based almost entirely upon the x86 perf event + * code, which is: + * + * Copyright (C) 2008 Thomas Gleixner + * Copyright (C) 2008-2009 Red Hat, Inc., Ingo Molnar + * Copyright (C) 2009 Jaswinder Singh Rajput + * Copyright (C) 2009 Advanced Micro Devices, Inc., Robert Richter + * Copyright (C) 2008-2009 Red Hat, Inc., Peter Zijlstra + */ + +#include +#include +#include +#include +#include + +#include +#include +#include +#include + +/* Sparc64 chips have two performance counters, 32-bits each, with + * overflow interrupts generated on transition from 0xffffffff to 0. + * The counters are accessed in one go using a 64-bit register. + * + * Both counters are controlled using a single control register. The + * only way to stop all sampling is to clear all of the context (user, + * supervisor, hypervisor) sampling enable bits. But these bits apply + * to both counters, thus the two counters can't be enabled/disabled + * individually. + * + * The control register has two event fields, one for each of the two + * counters. It's thus nearly impossible to have one counter going + * while keeping the other one stopped. Therefore it is possible to + * get overflow interrupts for counters not currently "in use" and + * that condition must be checked in the overflow interrupt handler. + * + * So we use a hack, in that we program inactive counters with the + * "sw_count0" and "sw_count1" events. These count how many times + * the instruction "sethi %hi(0xfc000), %g0" is executed. It's an + * unusual way to encode a NOP and therefore will not trigger in + * normal code. + */ + +#define MAX_HWEVENTS 2 +#define MAX_PERIOD ((1UL << 32) - 1) + +#define PIC_UPPER_INDEX 0 +#define PIC_LOWER_INDEX 1 + +struct cpu_hw_events { + struct perf_event *events[MAX_HWEVENTS]; + unsigned long used_mask[BITS_TO_LONGS(MAX_HWEVENTS)]; + unsigned long active_mask[BITS_TO_LONGS(MAX_HWEVENTS)]; + int enabled; +}; +DEFINE_PER_CPU(struct cpu_hw_events, cpu_hw_events) = { .enabled = 1, }; + +struct perf_event_map { + u16 encoding; + u8 pic_mask; +#define PIC_NONE 0x00 +#define PIC_UPPER 0x01 +#define PIC_LOWER 0x02 +}; + +struct sparc_pmu { + const struct perf_event_map *(*event_map)(int); + int max_events; + int upper_shift; + int lower_shift; + int event_mask; + int hv_bit; + int irq_bit; + int upper_nop; + int lower_nop; +}; + +static const struct perf_event_map ultra3i_perfmon_event_map[] = { + [PERF_COUNT_HW_CPU_CYCLES] = { 0x0000, PIC_UPPER | PIC_LOWER }, + [PERF_COUNT_HW_INSTRUCTIONS] = { 0x0001, PIC_UPPER | PIC_LOWER }, + [PERF_COUNT_HW_CACHE_REFERENCES] = { 0x0009, PIC_LOWER }, + [PERF_COUNT_HW_CACHE_MISSES] = { 0x0009, PIC_UPPER }, +}; + +static const struct perf_event_map *ultra3i_event_map(int event_id) +{ + return &ultra3i_perfmon_event_map[event_id]; +} + +static const struct sparc_pmu ultra3i_pmu = { + .event_map = ultra3i_event_map, + .max_events = ARRAY_SIZE(ultra3i_perfmon_event_map), + .upper_shift = 11, + .lower_shift = 4, + .event_mask = 0x3f, + .upper_nop = 0x1c, + .lower_nop = 0x14, +}; + +static const struct perf_event_map niagara2_perfmon_event_map[] = { + [PERF_COUNT_HW_CPU_CYCLES] = { 0x02ff, PIC_UPPER | PIC_LOWER }, + [PERF_COUNT_HW_INSTRUCTIONS] = { 0x02ff, PIC_UPPER | PIC_LOWER }, + [PERF_COUNT_HW_CACHE_REFERENCES] = { 0x0208, PIC_UPPER | PIC_LOWER }, + [PERF_COUNT_HW_CACHE_MISSES] = { 0x0302, PIC_UPPER | PIC_LOWER }, + [PERF_COUNT_HW_BRANCH_INSTRUCTIONS] = { 0x0201, PIC_UPPER | PIC_LOWER }, + [PERF_COUNT_HW_BRANCH_MISSES] = { 0x0202, PIC_UPPER | PIC_LOWER }, +}; + +static const struct perf_event_map *niagara2_event_map(int event_id) +{ + return &niagara2_perfmon_event_map[event_id]; +} + +static const struct sparc_pmu niagara2_pmu = { + .event_map = niagara2_event_map, + .max_events = ARRAY_SIZE(niagara2_perfmon_event_map), + .upper_shift = 19, + .lower_shift = 6, + .event_mask = 0xfff, + .hv_bit = 0x8, + .irq_bit = 0x03, + .upper_nop = 0x220, + .lower_nop = 0x220, +}; + +static const struct sparc_pmu *sparc_pmu __read_mostly; + +static u64 event_encoding(u64 event_id, int idx) +{ + if (idx == PIC_UPPER_INDEX) + event_id <<= sparc_pmu->upper_shift; + else + event_id <<= sparc_pmu->lower_shift; + return event_id; +} + +static u64 mask_for_index(int idx) +{ + return event_encoding(sparc_pmu->event_mask, idx); +} + +static u64 nop_for_index(int idx) +{ + return event_encoding(idx == PIC_UPPER_INDEX ? + sparc_pmu->upper_nop : + sparc_pmu->lower_nop, idx); +} + +static inline void sparc_pmu_enable_event(struct hw_perf_event *hwc, + int idx) +{ + u64 val, mask = mask_for_index(idx); + + val = pcr_ops->read(); + pcr_ops->write((val & ~mask) | hwc->config); +} + +static inline void sparc_pmu_disable_event(struct hw_perf_event *hwc, + int idx) +{ + u64 mask = mask_for_index(idx); + u64 nop = nop_for_index(idx); + u64 val = pcr_ops->read(); + + pcr_ops->write((val & ~mask) | nop); +} + +void hw_perf_enable(void) +{ + struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events); + u64 val; + int i; + + if (cpuc->enabled) + return; + + cpuc->enabled = 1; + barrier(); + + val = pcr_ops->read(); + + for (i = 0; i < MAX_HWEVENTS; i++) { + struct perf_event *cp = cpuc->events[i]; + struct hw_perf_event *hwc; + + if (!cp) + continue; + hwc = &cp->hw; + val |= hwc->config_base; + } + + pcr_ops->write(val); +} + +void hw_perf_disable(void) +{ + struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events); + u64 val; + + if (!cpuc->enabled) + return; + + cpuc->enabled = 0; + + val = pcr_ops->read(); + val &= ~(PCR_UTRACE | PCR_STRACE | + sparc_pmu->hv_bit | sparc_pmu->irq_bit); + pcr_ops->write(val); +} + +static u32 read_pmc(int idx) +{ + u64 val; + + read_pic(val); + if (idx == PIC_UPPER_INDEX) + val >>= 32; + + return val & 0xffffffff; +} + +static void write_pmc(int idx, u64 val) +{ + u64 shift, mask, pic; + + shift = 0; + if (idx == PIC_UPPER_INDEX) + shift = 32; + + mask = ((u64) 0xffffffff) << shift; + val <<= shift; + + read_pic(pic); + pic &= ~mask; + pic |= val; + write_pic(pic); +} + +static int sparc_perf_event_set_period(struct perf_event *event, + struct hw_perf_event *hwc, int idx) +{ + s64 left = atomic64_read(&hwc->period_left); + s64 period = hwc->sample_period; + int ret = 0; + + if (unlikely(left <= -period)) { + left = period; + atomic64_set(&hwc->period_left, left); + hwc->last_period = period; + ret = 1; + } + + if (unlikely(left <= 0)) { + left += period; + atomic64_set(&hwc->period_left, left); + hwc->last_period = period; + ret = 1; + } + if (left > MAX_PERIOD) + left = MAX_PERIOD; + + atomic64_set(&hwc->prev_count, (u64)-left); + + write_pmc(idx, (u64)(-left) & 0xffffffff); + + perf_event_update_userpage(event); + + return ret; +} + +static int sparc_pmu_enable(struct perf_event *event) +{ + struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events); + struct hw_perf_event *hwc = &event->hw; + int idx = hwc->idx; + + if (test_and_set_bit(idx, cpuc->used_mask)) + return -EAGAIN; + + sparc_pmu_disable_event(hwc, idx); + + cpuc->events[idx] = event; + set_bit(idx, cpuc->active_mask); + + sparc_perf_event_set_period(event, hwc, idx); + sparc_pmu_enable_event(hwc, idx); + perf_event_update_userpage(event); + return 0; +} + +static u64 sparc_perf_event_update(struct perf_event *event, + struct hw_perf_event *hwc, int idx) +{ + int shift = 64 - 32; + u64 prev_raw_count, new_raw_count; + s64 delta; + +again: + prev_raw_count = atomic64_read(&hwc->prev_count); + new_raw_count = read_pmc(idx); + + if (atomic64_cmpxchg(&hwc->prev_count, prev_raw_count, + new_raw_count) != prev_raw_count) + goto again; + + delta = (new_raw_count << shift) - (prev_raw_count << shift); + delta >>= shift; + + atomic64_add(delta, &event->count); + atomic64_sub(delta, &hwc->period_left); + + return new_raw_count; +} + +static void sparc_pmu_disable(struct perf_event *event) +{ + struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events); + struct hw_perf_event *hwc = &event->hw; + int idx = hwc->idx; + + clear_bit(idx, cpuc->active_mask); + sparc_pmu_disable_event(hwc, idx); + + barrier(); + + sparc_perf_event_update(event, hwc, idx); + cpuc->events[idx] = NULL; + clear_bit(idx, cpuc->used_mask); + + perf_event_update_userpage(event); +} + +static void sparc_pmu_read(struct perf_event *event) +{ + struct hw_perf_event *hwc = &event->hw; + sparc_perf_event_update(event, hwc, hwc->idx); +} + +static void sparc_pmu_unthrottle(struct perf_event *event) +{ + struct hw_perf_event *hwc = &event->hw; + sparc_pmu_enable_event(hwc, hwc->idx); +} + +static atomic_t active_events = ATOMIC_INIT(0); +static DEFINE_MUTEX(pmc_grab_mutex); + +void perf_event_grab_pmc(void) +{ + if (atomic_inc_not_zero(&active_events)) + return; + + mutex_lock(&pmc_grab_mutex); + if (atomic_read(&active_events) == 0) { + if (atomic_read(&nmi_active) > 0) { + on_each_cpu(stop_nmi_watchdog, NULL, 1); + BUG_ON(atomic_read(&nmi_active) != 0); + } + atomic_inc(&active_events); + } + mutex_unlock(&pmc_grab_mutex); +} + +void perf_event_release_pmc(void) +{ + if (atomic_dec_and_mutex_lock(&active_events, &pmc_grab_mutex)) { + if (atomic_read(&nmi_active) == 0) + on_each_cpu(start_nmi_watchdog, NULL, 1); + mutex_unlock(&pmc_grab_mutex); + } +} + +static void hw_perf_event_destroy(struct perf_event *event) +{ + perf_event_release_pmc(); +} + +static int __hw_perf_event_init(struct perf_event *event) +{ + struct perf_event_attr *attr = &event->attr; + struct hw_perf_event *hwc = &event->hw; + const struct perf_event_map *pmap; + u64 enc; + + if (atomic_read(&nmi_active) < 0) + return -ENODEV; + + if (attr->type != PERF_TYPE_HARDWARE) + return -EOPNOTSUPP; + + if (attr->config >= sparc_pmu->max_events) + return -EINVAL; + + perf_event_grab_pmc(); + event->destroy = hw_perf_event_destroy; + + /* We save the enable bits in the config_base. So to + * turn off sampling just write 'config', and to enable + * things write 'config | config_base'. + */ + hwc->config_base = sparc_pmu->irq_bit; + if (!attr->exclude_user) + hwc->config_base |= PCR_UTRACE; + if (!attr->exclude_kernel) + hwc->config_base |= PCR_STRACE; + if (!attr->exclude_hv) + hwc->config_base |= sparc_pmu->hv_bit; + + if (!hwc->sample_period) { + hwc->sample_period = MAX_PERIOD; + hwc->last_period = hwc->sample_period; + atomic64_set(&hwc->period_left, hwc->sample_period); + } + + pmap = sparc_pmu->event_map(attr->config); + + enc = pmap->encoding; + if (pmap->pic_mask & PIC_UPPER) { + hwc->idx = PIC_UPPER_INDEX; + enc <<= sparc_pmu->upper_shift; + } else { + hwc->idx = PIC_LOWER_INDEX; + enc <<= sparc_pmu->lower_shift; + } + + hwc->config |= enc; + return 0; +} + +static const struct pmu pmu = { + .enable = sparc_pmu_enable, + .disable = sparc_pmu_disable, + .read = sparc_pmu_read, + .unthrottle = sparc_pmu_unthrottle, +}; + +const struct pmu *hw_perf_event_init(struct perf_event *event) +{ + int err = __hw_perf_event_init(event); + + if (err) + return ERR_PTR(err); + return &pmu; +} + +void perf_event_print_debug(void) +{ + unsigned long flags; + u64 pcr, pic; + int cpu; + + if (!sparc_pmu) + return; + + local_irq_save(flags); + + cpu = smp_processor_id(); + + pcr = pcr_ops->read(); + read_pic(pic); + + pr_info("\n"); + pr_info("CPU#%d: PCR[%016llx] PIC[%016llx]\n", + cpu, pcr, pic); + + local_irq_restore(flags); +} + +static int __kprobes perf_event_nmi_handler(struct notifier_block *self, + unsigned long cmd, void *__args) +{ + struct die_args *args = __args; + struct perf_sample_data data; + struct cpu_hw_events *cpuc; + struct pt_regs *regs; + int idx; + + if (!atomic_read(&active_events)) + return NOTIFY_DONE; + + switch (cmd) { + case DIE_NMI: + break; + + default: + return NOTIFY_DONE; + } + + regs = args->regs; + + data.addr = 0; + + cpuc = &__get_cpu_var(cpu_hw_events); + for (idx = 0; idx < MAX_HWEVENTS; idx++) { + struct perf_event *event = cpuc->events[idx]; + struct hw_perf_event *hwc; + u64 val; + + if (!test_bit(idx, cpuc->active_mask)) + continue; + hwc = &event->hw; + val = sparc_perf_event_update(event, hwc, idx); + if (val & (1ULL << 31)) + continue; + + data.period = event->hw.last_period; + if (!sparc_perf_event_set_period(event, hwc, idx)) + continue; + + if (perf_event_overflow(event, 1, &data, regs)) + sparc_pmu_disable_event(hwc, idx); + } + + return NOTIFY_STOP; +} + +static __read_mostly struct notifier_block perf_event_nmi_notifier = { + .notifier_call = perf_event_nmi_handler, +}; + +static bool __init supported_pmu(void) +{ + if (!strcmp(sparc_pmu_type, "ultra3i")) { + sparc_pmu = &ultra3i_pmu; + return true; + } + if (!strcmp(sparc_pmu_type, "niagara2")) { + sparc_pmu = &niagara2_pmu; + return true; + } + return false; +} + +void __init init_hw_perf_events(void) +{ + pr_info("Performance events: "); + + if (!supported_pmu()) { + pr_cont("No support for PMU type '%s'\n", sparc_pmu_type); + return; + } + + pr_cont("Supported PMU type is '%s'\n", sparc_pmu_type); + + /* All sparc64 PMUs currently have 2 events. But this simple + * driver only supports one active event at a time. + */ + perf_max_events = 1; + + register_die_notifier(&perf_event_nmi_notifier); +} diff --git a/arch/sparc/kernel/systbls_32.S b/arch/sparc/kernel/systbls_32.S index 0418157..0f1658d 100644 --- a/arch/sparc/kernel/systbls_32.S +++ b/arch/sparc/kernel/systbls_32.S @@ -82,5 +82,5 @@ sys_call_table: /*310*/ .long sys_utimensat, sys_signalfd, sys_timerfd_create, sys_eventfd, sys_fallocate /*315*/ .long sys_timerfd_settime, sys_timerfd_gettime, sys_signalfd4, sys_eventfd2, sys_epoll_create1 /*320*/ .long sys_dup3, sys_pipe2, sys_inotify_init1, sys_accept4, sys_preadv -/*325*/ .long sys_pwritev, sys_rt_tgsigqueueinfo, sys_perf_counter_open +/*325*/ .long sys_pwritev, sys_rt_tgsigqueueinfo, sys_perf_event_open diff --git a/arch/sparc/kernel/systbls_64.S b/arch/sparc/kernel/systbls_64.S index 91b06b7..009825f 100644 --- a/arch/sparc/kernel/systbls_64.S +++ b/arch/sparc/kernel/systbls_64.S @@ -83,7 +83,7 @@ sys_call_table32: /*310*/ .word compat_sys_utimensat, compat_sys_signalfd, sys_timerfd_create, sys_eventfd, compat_sys_fallocate .word compat_sys_timerfd_settime, compat_sys_timerfd_gettime, compat_sys_signalfd4, sys_eventfd2, sys_epoll_create1 /*320*/ .word sys_dup3, sys_pipe2, sys_inotify_init1, sys_accept4, compat_sys_preadv - .word compat_sys_pwritev, compat_sys_rt_tgsigqueueinfo, sys_perf_counter_open + .word compat_sys_pwritev, compat_sys_rt_tgsigqueueinfo, sys_perf_event_open #endif /* CONFIG_COMPAT */ @@ -158,4 +158,4 @@ sys_call_table: /*310*/ .word sys_utimensat, sys_signalfd, sys_timerfd_create, sys_eventfd, sys_fallocate .word sys_timerfd_settime, sys_timerfd_gettime, sys_signalfd4, sys_eventfd2, sys_epoll_create1 /*320*/ .word sys_dup3, sys_pipe2, sys_inotify_init1, sys_accept4, sys_preadv - .word sys_pwritev, sys_rt_tgsigqueueinfo, sys_perf_counter_open + .word sys_pwritev, sys_rt_tgsigqueueinfo, sys_perf_event_open diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig index 51c5901..e4ff5d1 100644 --- a/arch/x86/Kconfig +++ b/arch/x86/Kconfig @@ -24,7 +24,7 @@ config X86 select HAVE_UNSTABLE_SCHED_CLOCK select HAVE_IDE select HAVE_OPROFILE - select HAVE_PERF_COUNTERS if (!M386 && !M486) + select HAVE_PERF_EVENTS if (!M386 && !M486) select HAVE_IOREMAP_PROT select HAVE_KPROBES select ARCH_WANT_OPTIONAL_GPIOLIB diff --git a/arch/x86/ia32/ia32entry.S b/arch/x86/ia32/ia32entry.S index ba331bf..74619c4 100644 --- a/arch/x86/ia32/ia32entry.S +++ b/arch/x86/ia32/ia32entry.S @@ -831,5 +831,5 @@ ia32_sys_call_table: .quad compat_sys_preadv .quad compat_sys_pwritev .quad compat_sys_rt_tgsigqueueinfo /* 335 */ - .quad sys_perf_counter_open + .quad sys_perf_event_open ia32_syscall_end: diff --git a/arch/x86/include/asm/entry_arch.h b/arch/x86/include/asm/entry_arch.h index 5e3f204..f5693c8 100644 --- a/arch/x86/include/asm/entry_arch.h +++ b/arch/x86/include/asm/entry_arch.h @@ -49,7 +49,7 @@ BUILD_INTERRUPT(apic_timer_interrupt,LOCAL_TIMER_VECTOR) BUILD_INTERRUPT(error_interrupt,ERROR_APIC_VECTOR) BUILD_INTERRUPT(spurious_interrupt,SPURIOUS_APIC_VECTOR) -#ifdef CONFIG_PERF_COUNTERS +#ifdef CONFIG_PERF_EVENTS BUILD_INTERRUPT(perf_pending_interrupt, LOCAL_PENDING_VECTOR) #endif diff --git a/arch/x86/include/asm/perf_counter.h b/arch/x86/include/asm/perf_counter.h deleted file mode 100644 index e7b7c93..0000000 --- a/arch/x86/include/asm/perf_counter.h +++ /dev/null @@ -1,108 +0,0 @@ -#ifndef _ASM_X86_PERF_COUNTER_H -#define _ASM_X86_PERF_COUNTER_H - -/* - * Performance counter hw details: - */ - -#define X86_PMC_MAX_GENERIC 8 -#define X86_PMC_MAX_FIXED 3 - -#define X86_PMC_IDX_GENERIC 0 -#define X86_PMC_IDX_FIXED 32 -#define X86_PMC_IDX_MAX 64 - -#define MSR_ARCH_PERFMON_PERFCTR0 0xc1 -#define MSR_ARCH_PERFMON_PERFCTR1 0xc2 - -#define MSR_ARCH_PERFMON_EVENTSEL0 0x186 -#define MSR_ARCH_PERFMON_EVENTSEL1 0x187 - -#define ARCH_PERFMON_EVENTSEL0_ENABLE (1 << 22) -#define ARCH_PERFMON_EVENTSEL_INT (1 << 20) -#define ARCH_PERFMON_EVENTSEL_OS (1 << 17) -#define ARCH_PERFMON_EVENTSEL_USR (1 << 16) - -/* - * Includes eventsel and unit mask as well: - */ -#define ARCH_PERFMON_EVENT_MASK 0xffff - -#define ARCH_PERFMON_UNHALTED_CORE_CYCLES_SEL 0x3c -#define ARCH_PERFMON_UNHALTED_CORE_CYCLES_UMASK (0x00 << 8) -#define ARCH_PERFMON_UNHALTED_CORE_CYCLES_INDEX 0 -#define ARCH_PERFMON_UNHALTED_CORE_CYCLES_PRESENT \ - (1 << (ARCH_PERFMON_UNHALTED_CORE_CYCLES_INDEX)) - -#define ARCH_PERFMON_BRANCH_MISSES_RETIRED 6 - -/* - * Intel "Architectural Performance Monitoring" CPUID - * detection/enumeration details: - */ -union cpuid10_eax { - struct { - unsigned int version_id:8; - unsigned int num_counters:8; - unsigned int bit_width:8; - unsigned int mask_length:8; - } split; - unsigned int full; -}; - -union cpuid10_edx { - struct { - unsigned int num_counters_fixed:4; - unsigned int reserved:28; - } split; - unsigned int full; -}; - - -/* - * Fixed-purpose performance counters: - */ - -/* - * All 3 fixed-mode PMCs are configured via this single MSR: - */ -#define MSR_ARCH_PERFMON_FIXED_CTR_CTRL 0x38d - -/* - * The counts are available in three separate MSRs: - */ - -/* Instr_Retired.Any: */ -#define MSR_ARCH_PERFMON_FIXED_CTR0 0x309 -#define X86_PMC_IDX_FIXED_INSTRUCTIONS (X86_PMC_IDX_FIXED + 0) - -/* CPU_CLK_Unhalted.Core: */ -#define MSR_ARCH_PERFMON_FIXED_CTR1 0x30a -#define X86_PMC_IDX_FIXED_CPU_CYCLES (X86_PMC_IDX_FIXED + 1) - -/* CPU_CLK_Unhalted.Ref: */ -#define MSR_ARCH_PERFMON_FIXED_CTR2 0x30b -#define X86_PMC_IDX_FIXED_BUS_CYCLES (X86_PMC_IDX_FIXED + 2) - -/* - * We model BTS tracing as another fixed-mode PMC. - * - * We choose a value in the middle of the fixed counter range, since lower - * values are used by actual fixed counters and higher values are used - * to indicate other overflow conditions in the PERF_GLOBAL_STATUS msr. - */ -#define X86_PMC_IDX_FIXED_BTS (X86_PMC_IDX_FIXED + 16) - - -#ifdef CONFIG_PERF_COUNTERS -extern void init_hw_perf_counters(void); -extern void perf_counters_lapic_init(void); - -#define PERF_COUNTER_INDEX_OFFSET 0 - -#else -static inline void init_hw_perf_counters(void) { } -static inline void perf_counters_lapic_init(void) { } -#endif - -#endif /* _ASM_X86_PERF_COUNTER_H */ diff --git a/arch/x86/include/asm/perf_event.h b/arch/x86/include/asm/perf_event.h new file mode 100644 index 0000000..ad7ce3f --- /dev/null +++ b/arch/x86/include/asm/perf_event.h @@ -0,0 +1,108 @@ +#ifndef _ASM_X86_PERF_EVENT_H +#define _ASM_X86_PERF_EVENT_H + +/* + * Performance event hw details: + */ + +#define X86_PMC_MAX_GENERIC 8 +#define X86_PMC_MAX_FIXED 3 + +#define X86_PMC_IDX_GENERIC 0 +#define X86_PMC_IDX_FIXED 32 +#define X86_PMC_IDX_MAX 64 + +#define MSR_ARCH_PERFMON_PERFCTR0 0xc1 +#define MSR_ARCH_PERFMON_PERFCTR1 0xc2 + +#define MSR_ARCH_PERFMON_EVENTSEL0 0x186 +#define MSR_ARCH_PERFMON_EVENTSEL1 0x187 + +#define ARCH_PERFMON_EVENTSEL0_ENABLE (1 << 22) +#define ARCH_PERFMON_EVENTSEL_INT (1 << 20) +#define ARCH_PERFMON_EVENTSEL_OS (1 << 17) +#define ARCH_PERFMON_EVENTSEL_USR (1 << 16) + +/* + * Includes eventsel and unit mask as well: + */ +#define ARCH_PERFMON_EVENT_MASK 0xffff + +#define ARCH_PERFMON_UNHALTED_CORE_CYCLES_SEL 0x3c +#define ARCH_PERFMON_UNHALTED_CORE_CYCLES_UMASK (0x00 << 8) +#define ARCH_PERFMON_UNHALTED_CORE_CYCLES_INDEX 0 +#define ARCH_PERFMON_UNHALTED_CORE_CYCLES_PRESENT \ + (1 << (ARCH_PERFMON_UNHALTED_CORE_CYCLES_INDEX)) + +#define ARCH_PERFMON_BRANCH_MISSES_RETIRED 6 + +/* + * Intel "Architectural Performance Monitoring" CPUID + * detection/enumeration details: + */ +union cpuid10_eax { + struct { + unsigned int version_id:8; + unsigned int num_events:8; + unsigned int bit_width:8; + unsigned int mask_length:8; + } split; + unsigned int full; +}; + +union cpuid10_edx { + struct { + unsigned int num_events_fixed:4; + unsigned int reserved:28; + } split; + unsigned int full; +}; + + +/* + * Fixed-purpose performance events: + */ + +/* + * All 3 fixed-mode PMCs are configured via this single MSR: + */ +#define MSR_ARCH_PERFMON_FIXED_CTR_CTRL 0x38d + +/* + * The counts are available in three separate MSRs: + */ + +/* Instr_Retired.Any: */ +#define MSR_ARCH_PERFMON_FIXED_CTR0 0x309 +#define X86_PMC_IDX_FIXED_INSTRUCTIONS (X86_PMC_IDX_FIXED + 0) + +/* CPU_CLK_Unhalted.Core: */ +#define MSR_ARCH_PERFMON_FIXED_CTR1 0x30a +#define X86_PMC_IDX_FIXED_CPU_CYCLES (X86_PMC_IDX_FIXED + 1) + +/* CPU_CLK_Unhalted.Ref: */ +#define MSR_ARCH_PERFMON_FIXED_CTR2 0x30b +#define X86_PMC_IDX_FIXED_BUS_CYCLES (X86_PMC_IDX_FIXED + 2) + +/* + * We model BTS tracing as another fixed-mode PMC. + * + * We choose a value in the middle of the fixed event range, since lower + * values are used by actual fixed events and higher values are used + * to indicate other overflow conditions in the PERF_GLOBAL_STATUS msr. + */ +#define X86_PMC_IDX_FIXED_BTS (X86_PMC_IDX_FIXED + 16) + + +#ifdef CONFIG_PERF_EVENTS +extern void init_hw_perf_events(void); +extern void perf_events_lapic_init(void); + +#define PERF_EVENT_INDEX_OFFSET 0 + +#else +static inline void init_hw_perf_events(void) { } +static inline void perf_events_lapic_init(void) { } +#endif + +#endif /* _ASM_X86_PERF_EVENT_H */ diff --git a/arch/x86/include/asm/unistd_32.h b/arch/x86/include/asm/unistd_32.h index 8deaada..6fb3c20 100644 --- a/arch/x86/include/asm/unistd_32.h +++ b/arch/x86/include/asm/unistd_32.h @@ -341,7 +341,7 @@ #define __NR_preadv 333 #define __NR_pwritev 334 #define __NR_rt_tgsigqueueinfo 335 -#define __NR_perf_counter_open 336 +#define __NR_perf_event_open 336 #ifdef __KERNEL__ diff --git a/arch/x86/include/asm/unistd_64.h b/arch/x86/include/asm/unistd_64.h index b9f3c60..8d3ad0a 100644 --- a/arch/x86/include/asm/unistd_64.h +++ b/arch/x86/include/asm/unistd_64.h @@ -659,8 +659,8 @@ __SYSCALL(__NR_preadv, sys_preadv) __SYSCALL(__NR_pwritev, sys_pwritev) #define __NR_rt_tgsigqueueinfo 297 __SYSCALL(__NR_rt_tgsigqueueinfo, sys_rt_tgsigqueueinfo) -#define __NR_perf_counter_open 298 -__SYSCALL(__NR_perf_counter_open, sys_perf_counter_open) +#define __NR_perf_event_open 298 +__SYSCALL(__NR_perf_event_open, sys_perf_event_open) #ifndef __NO_STUBS #define __ARCH_WANT_OLD_READDIR diff --git a/arch/x86/kernel/apic/apic.c b/arch/x86/kernel/apic/apic.c index a34601f..754174d 100644 --- a/arch/x86/kernel/apic/apic.c +++ b/arch/x86/kernel/apic/apic.c @@ -14,7 +14,7 @@ * Mikael Pettersson : PM converted to driver model. */ -#include +#include #include #include #include @@ -35,7 +35,7 @@ #include #include -#include +#include #include #include #include @@ -1189,7 +1189,7 @@ void __cpuinit setup_local_APIC(void) apic_write(APIC_ESR, 0); } #endif - perf_counters_lapic_init(); + perf_events_lapic_init(); preempt_disable(); diff --git a/arch/x86/kernel/cpu/Makefile b/arch/x86/kernel/cpu/Makefile index 8dd3063..68537e9 100644 --- a/arch/x86/kernel/cpu/Makefile +++ b/arch/x86/kernel/cpu/Makefile @@ -27,7 +27,7 @@ obj-$(CONFIG_CPU_SUP_CENTAUR) += centaur.o obj-$(CONFIG_CPU_SUP_TRANSMETA_32) += transmeta.o obj-$(CONFIG_CPU_SUP_UMC_32) += umc.o -obj-$(CONFIG_PERF_COUNTERS) += perf_counter.o +obj-$(CONFIG_PERF_EVENTS) += perf_event.o obj-$(CONFIG_X86_MCE) += mcheck/ obj-$(CONFIG_MTRR) += mtrr/ diff --git a/arch/x86/kernel/cpu/common.c b/arch/x86/kernel/cpu/common.c index 2fea97e..cc25c2b 100644 --- a/arch/x86/kernel/cpu/common.c +++ b/arch/x86/kernel/cpu/common.c @@ -13,7 +13,7 @@ #include #include -#include +#include #include #include #include @@ -869,7 +869,7 @@ void __init identify_boot_cpu(void) #else vgetcpu_set_mode(); #endif - init_hw_perf_counters(); + init_hw_perf_events(); } void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c) diff --git a/arch/x86/kernel/cpu/perf_counter.c b/arch/x86/kernel/cpu/perf_counter.c deleted file mode 100644 index b1f1156..0000000 --- a/arch/x86/kernel/cpu/perf_counter.c +++ /dev/null @@ -1,2298 +0,0 @@ -/* - * Performance counter x86 architecture code - * - * Copyright (C) 2008 Thomas Gleixner - * Copyright (C) 2008-2009 Red Hat, Inc., Ingo Molnar - * Copyright (C) 2009 Jaswinder Singh Rajput - * Copyright (C) 2009 Advanced Micro Devices, Inc., Robert Richter - * Copyright (C) 2008-2009 Red Hat, Inc., Peter Zijlstra - * Copyright (C) 2009 Intel Corporation, - * - * For licencing details see kernel-base/COPYING - */ - -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include - -#include -#include -#include - -static u64 perf_counter_mask __read_mostly; - -/* The maximal number of PEBS counters: */ -#define MAX_PEBS_COUNTERS 4 - -/* The size of a BTS record in bytes: */ -#define BTS_RECORD_SIZE 24 - -/* The size of a per-cpu BTS buffer in bytes: */ -#define BTS_BUFFER_SIZE (BTS_RECORD_SIZE * 2048) - -/* The BTS overflow threshold in bytes from the end of the buffer: */ -#define BTS_OVFL_TH (BTS_RECORD_SIZE * 128) - - -/* - * Bits in the debugctlmsr controlling branch tracing. - */ -#define X86_DEBUGCTL_TR (1 << 6) -#define X86_DEBUGCTL_BTS (1 << 7) -#define X86_DEBUGCTL_BTINT (1 << 8) -#define X86_DEBUGCTL_BTS_OFF_OS (1 << 9) -#define X86_DEBUGCTL_BTS_OFF_USR (1 << 10) - -/* - * A debug store configuration. - * - * We only support architectures that use 64bit fields. - */ -struct debug_store { - u64 bts_buffer_base; - u64 bts_index; - u64 bts_absolute_maximum; - u64 bts_interrupt_threshold; - u64 pebs_buffer_base; - u64 pebs_index; - u64 pebs_absolute_maximum; - u64 pebs_interrupt_threshold; - u64 pebs_counter_reset[MAX_PEBS_COUNTERS]; -}; - -struct cpu_hw_counters { - struct perf_counter *counters[X86_PMC_IDX_MAX]; - unsigned long used_mask[BITS_TO_LONGS(X86_PMC_IDX_MAX)]; - unsigned long active_mask[BITS_TO_LONGS(X86_PMC_IDX_MAX)]; - unsigned long interrupts; - int enabled; - struct debug_store *ds; -}; - -/* - * struct x86_pmu - generic x86 pmu - */ -struct x86_pmu { - const char *name; - int version; - int (*handle_irq)(struct pt_regs *); - void (*disable_all)(void); - void (*enable_all)(void); - void (*enable)(struct hw_perf_counter *, int); - void (*disable)(struct hw_perf_counter *, int); - unsigned eventsel; - unsigned perfctr; - u64 (*event_map)(int); - u64 (*raw_event)(u64); - int max_events; - int num_counters; - int num_counters_fixed; - int counter_bits; - u64 counter_mask; - int apic; - u64 max_period; - u64 intel_ctrl; - void (*enable_bts)(u64 config); - void (*disable_bts)(void); -}; - -static struct x86_pmu x86_pmu __read_mostly; - -static DEFINE_PER_CPU(struct cpu_hw_counters, cpu_hw_counters) = { - .enabled = 1, -}; - -/* - * Not sure about some of these - */ -static const u64 p6_perfmon_event_map[] = -{ - [PERF_COUNT_HW_CPU_CYCLES] = 0x0079, - [PERF_COUNT_HW_INSTRUCTIONS] = 0x00c0, - [PERF_COUNT_HW_CACHE_REFERENCES] = 0x0f2e, - [PERF_COUNT_HW_CACHE_MISSES] = 0x012e, - [PERF_COUNT_HW_BRANCH_INSTRUCTIONS] = 0x00c4, - [PERF_COUNT_HW_BRANCH_MISSES] = 0x00c5, - [PERF_COUNT_HW_BUS_CYCLES] = 0x0062, -}; - -static u64 p6_pmu_event_map(int hw_event) -{ - return p6_perfmon_event_map[hw_event]; -} - -/* - * Counter setting that is specified not to count anything. - * We use this to effectively disable a counter. - * - * L2_RQSTS with 0 MESI unit mask. - */ -#define P6_NOP_COUNTER 0x0000002EULL - -static u64 p6_pmu_raw_event(u64 hw_event) -{ -#define P6_EVNTSEL_EVENT_MASK 0x000000FFULL -#define P6_EVNTSEL_UNIT_MASK 0x0000FF00ULL -#define P6_EVNTSEL_EDGE_MASK 0x00040000ULL -#define P6_EVNTSEL_INV_MASK 0x00800000ULL -#define P6_EVNTSEL_COUNTER_MASK 0xFF000000ULL - -#define P6_EVNTSEL_MASK \ - (P6_EVNTSEL_EVENT_MASK | \ - P6_EVNTSEL_UNIT_MASK | \ - P6_EVNTSEL_EDGE_MASK | \ - P6_EVNTSEL_INV_MASK | \ - P6_EVNTSEL_COUNTER_MASK) - - return hw_event & P6_EVNTSEL_MASK; -} - - -/* - * Intel PerfMon v3. Used on Core2 and later. - */ -static const u64 intel_perfmon_event_map[] = -{ - [PERF_COUNT_HW_CPU_CYCLES] = 0x003c, - [PERF_COUNT_HW_INSTRUCTIONS] = 0x00c0, - [PERF_COUNT_HW_CACHE_REFERENCES] = 0x4f2e, - [PERF_COUNT_HW_CACHE_MISSES] = 0x412e, - [PERF_COUNT_HW_BRANCH_INSTRUCTIONS] = 0x00c4, - [PERF_COUNT_HW_BRANCH_MISSES] = 0x00c5, - [PERF_COUNT_HW_BUS_CYCLES] = 0x013c, -}; - -static u64 intel_pmu_event_map(int hw_event) -{ - return intel_perfmon_event_map[hw_event]; -} - -/* - * Generalized hw caching related hw_event table, filled - * in on a per model basis. A value of 0 means - * 'not supported', -1 means 'hw_event makes no sense on - * this CPU', any other value means the raw hw_event - * ID. - */ - -#define C(x) PERF_COUNT_HW_CACHE_##x - -static u64 __read_mostly hw_cache_event_ids - [PERF_COUNT_HW_CACHE_MAX] - [PERF_COUNT_HW_CACHE_OP_MAX] - [PERF_COUNT_HW_CACHE_RESULT_MAX]; - -static const u64 nehalem_hw_cache_event_ids - [PERF_COUNT_HW_CACHE_MAX] - [PERF_COUNT_HW_CACHE_OP_MAX] - [PERF_COUNT_HW_CACHE_RESULT_MAX] = -{ - [ C(L1D) ] = { - [ C(OP_READ) ] = { - [ C(RESULT_ACCESS) ] = 0x0f40, /* L1D_CACHE_LD.MESI */ - [ C(RESULT_MISS) ] = 0x0140, /* L1D_CACHE_LD.I_STATE */ - }, - [ C(OP_WRITE) ] = { - [ C(RESULT_ACCESS) ] = 0x0f41, /* L1D_CACHE_ST.MESI */ - [ C(RESULT_MISS) ] = 0x0141, /* L1D_CACHE_ST.I_STATE */ - }, - [ C(OP_PREFETCH) ] = { - [ C(RESULT_ACCESS) ] = 0x014e, /* L1D_PREFETCH.REQUESTS */ - [ C(RESULT_MISS) ] = 0x024e, /* L1D_PREFETCH.MISS */ - }, - }, - [ C(L1I ) ] = { - [ C(OP_READ) ] = { - [ C(RESULT_ACCESS) ] = 0x0380, /* L1I.READS */ - [ C(RESULT_MISS) ] = 0x0280, /* L1I.MISSES */ - }, - [ C(OP_WRITE) ] = { - [ C(RESULT_ACCESS) ] = -1, - [ C(RESULT_MISS) ] = -1, - }, - [ C(OP_PREFETCH) ] = { - [ C(RESULT_ACCESS) ] = 0x0, - [ C(RESULT_MISS) ] = 0x0, - }, - }, - [ C(LL ) ] = { - [ C(OP_READ) ] = { - [ C(RESULT_ACCESS) ] = 0x0324, /* L2_RQSTS.LOADS */ - [ C(RESULT_MISS) ] = 0x0224, /* L2_RQSTS.LD_MISS */ - }, - [ C(OP_WRITE) ] = { - [ C(RESULT_ACCESS) ] = 0x0c24, /* L2_RQSTS.RFOS */ - [ C(RESULT_MISS) ] = 0x0824, /* L2_RQSTS.RFO_MISS */ - }, - [ C(OP_PREFETCH) ] = { - [ C(RESULT_ACCESS) ] = 0x4f2e, /* LLC Reference */ - [ C(RESULT_MISS) ] = 0x412e, /* LLC Misses */ - }, - }, - [ C(DTLB) ] = { - [ C(OP_READ) ] = { - [ C(RESULT_ACCESS) ] = 0x0f40, /* L1D_CACHE_LD.MESI (alias) */ - [ C(RESULT_MISS) ] = 0x0108, /* DTLB_LOAD_MISSES.ANY */ - }, - [ C(OP_WRITE) ] = { - [ C(RESULT_ACCESS) ] = 0x0f41, /* L1D_CACHE_ST.MESI (alias) */ - [ C(RESULT_MISS) ] = 0x010c, /* MEM_STORE_RETIRED.DTLB_MISS */ - }, - [ C(OP_PREFETCH) ] = { - [ C(RESULT_ACCESS) ] = 0x0, - [ C(RESULT_MISS) ] = 0x0, - }, - }, - [ C(ITLB) ] = { - [ C(OP_READ) ] = { - [ C(RESULT_ACCESS) ] = 0x01c0, /* INST_RETIRED.ANY_P */ - [ C(RESULT_MISS) ] = 0x20c8, /* ITLB_MISS_RETIRED */ - }, - [ C(OP_WRITE) ] = { - [ C(RESULT_ACCESS) ] = -1, - [ C(RESULT_MISS) ] = -1, - }, - [ C(OP_PREFETCH) ] = { - [ C(RESULT_ACCESS) ] = -1, - [ C(RESULT_MISS) ] = -1, - }, - }, - [ C(BPU ) ] = { - [ C(OP_READ) ] = { - [ C(RESULT_ACCESS) ] = 0x00c4, /* BR_INST_RETIRED.ALL_BRANCHES */ - [ C(RESULT_MISS) ] = 0x03e8, /* BPU_CLEARS.ANY */ - }, - [ C(OP_WRITE) ] = { - [ C(RESULT_ACCESS) ] = -1, - [ C(RESULT_MISS) ] = -1, - }, - [ C(OP_PREFETCH) ] = { - [ C(RESULT_ACCESS) ] = -1, - [ C(RESULT_MISS) ] = -1, - }, - }, -}; - -static const u64 core2_hw_cache_event_ids - [PERF_COUNT_HW_CACHE_MAX] - [PERF_COUNT_HW_CACHE_OP_MAX] - [PERF_COUNT_HW_CACHE_RESULT_MAX] = -{ - [ C(L1D) ] = { - [ C(OP_READ) ] = { - [ C(RESULT_ACCESS) ] = 0x0f40, /* L1D_CACHE_LD.MESI */ - [ C(RESULT_MISS) ] = 0x0140, /* L1D_CACHE_LD.I_STATE */ - }, - [ C(OP_WRITE) ] = { - [ C(RESULT_ACCESS) ] = 0x0f41, /* L1D_CACHE_ST.MESI */ - [ C(RESULT_MISS) ] = 0x0141, /* L1D_CACHE_ST.I_STATE */ - }, - [ C(OP_PREFETCH) ] = { - [ C(RESULT_ACCESS) ] = 0x104e, /* L1D_PREFETCH.REQUESTS */ - [ C(RESULT_MISS) ] = 0, - }, - }, - [ C(L1I ) ] = { - [ C(OP_READ) ] = { - [ C(RESULT_ACCESS) ] = 0x0080, /* L1I.READS */ - [ C(RESULT_MISS) ] = 0x0081, /* L1I.MISSES */ - }, - [ C(OP_WRITE) ] = { - [ C(RESULT_ACCESS) ] = -1, - [ C(RESULT_MISS) ] = -1, - }, - [ C(OP_PREFETCH) ] = { - [ C(RESULT_ACCESS) ] = 0, - [ C(RESULT_MISS) ] = 0, - }, - }, - [ C(LL ) ] = { - [ C(OP_READ) ] = { - [ C(RESULT_ACCESS) ] = 0x4f29, /* L2_LD.MESI */ - [ C(RESULT_MISS) ] = 0x4129, /* L2_LD.ISTATE */ - }, - [ C(OP_WRITE) ] = { - [ C(RESULT_ACCESS) ] = 0x4f2A, /* L2_ST.MESI */ - [ C(RESULT_MISS) ] = 0x412A, /* L2_ST.ISTATE */ - }, - [ C(OP_PREFETCH) ] = { - [ C(RESULT_ACCESS) ] = 0, - [ C(RESULT_MISS) ] = 0, - }, - }, - [ C(DTLB) ] = { - [ C(OP_READ) ] = { - [ C(RESULT_ACCESS) ] = 0x0f40, /* L1D_CACHE_LD.MESI (alias) */ - [ C(RESULT_MISS) ] = 0x0208, /* DTLB_MISSES.MISS_LD */ - }, - [ C(OP_WRITE) ] = { - [ C(RESULT_ACCESS) ] = 0x0f41, /* L1D_CACHE_ST.MESI (alias) */ - [ C(RESULT_MISS) ] = 0x0808, /* DTLB_MISSES.MISS_ST */ - }, - [ C(OP_PREFETCH) ] = { - [ C(RESULT_ACCESS) ] = 0, - [ C(RESULT_MISS) ] = 0, - }, - }, - [ C(ITLB) ] = { - [ C(OP_READ) ] = { - [ C(RESULT_ACCESS) ] = 0x00c0, /* INST_RETIRED.ANY_P */ - [ C(RESULT_MISS) ] = 0x1282, /* ITLBMISSES */ - }, - [ C(OP_WRITE) ] = { - [ C(RESULT_ACCESS) ] = -1, - [ C(RESULT_MISS) ] = -1, - }, - [ C(OP_PREFETCH) ] = { - [ C(RESULT_ACCESS) ] = -1, - [ C(RESULT_MISS) ] = -1, - }, - }, - [ C(BPU ) ] = { - [ C(OP_READ) ] = { - [ C(RESULT_ACCESS) ] = 0x00c4, /* BR_INST_RETIRED.ANY */ - [ C(RESULT_MISS) ] = 0x00c5, /* BP_INST_RETIRED.MISPRED */ - }, - [ C(OP_WRITE) ] = { - [ C(RESULT_ACCESS) ] = -1, - [ C(RESULT_MISS) ] = -1, - }, - [ C(OP_PREFETCH) ] = { - [ C(RESULT_ACCESS) ] = -1, - [ C(RESULT_MISS) ] = -1, - }, - }, -}; - -static const u64 atom_hw_cache_event_ids - [PERF_COUNT_HW_CACHE_MAX] - [PERF_COUNT_HW_CACHE_OP_MAX] - [PERF_COUNT_HW_CACHE_RESULT_MAX] = -{ - [ C(L1D) ] = { - [ C(OP_READ) ] = { - [ C(RESULT_ACCESS) ] = 0x2140, /* L1D_CACHE.LD */ - [ C(RESULT_MISS) ] = 0, - }, - [ C(OP_WRITE) ] = { - [ C(RESULT_ACCESS) ] = 0x2240, /* L1D_CACHE.ST */ - [ C(RESULT_MISS) ] = 0, - }, - [ C(OP_PREFETCH) ] = { - [ C(RESULT_ACCESS) ] = 0x0, - [ C(RESULT_MISS) ] = 0, - }, - }, - [ C(L1I ) ] = { - [ C(OP_READ) ] = { - [ C(RESULT_ACCESS) ] = 0x0380, /* L1I.READS */ - [ C(RESULT_MISS) ] = 0x0280, /* L1I.MISSES */ - }, - [ C(OP_WRITE) ] = { - [ C(RESULT_ACCESS) ] = -1, - [ C(RESULT_MISS) ] = -1, - }, - [ C(OP_PREFETCH) ] = { - [ C(RESULT_ACCESS) ] = 0, - [ C(RESULT_MISS) ] = 0, - }, - }, - [ C(LL ) ] = { - [ C(OP_READ) ] = { - [ C(RESULT_ACCESS) ] = 0x4f29, /* L2_LD.MESI */ - [ C(RESULT_MISS) ] = 0x4129, /* L2_LD.ISTATE */ - }, - [ C(OP_WRITE) ] = { - [ C(RESULT_ACCESS) ] = 0x4f2A, /* L2_ST.MESI */ - [ C(RESULT_MISS) ] = 0x412A, /* L2_ST.ISTATE */ - }, - [ C(OP_PREFETCH) ] = { - [ C(RESULT_ACCESS) ] = 0, - [ C(RESULT_MISS) ] = 0, - }, - }, - [ C(DTLB) ] = { - [ C(OP_READ) ] = { - [ C(RESULT_ACCESS) ] = 0x2140, /* L1D_CACHE_LD.MESI (alias) */ - [ C(RESULT_MISS) ] = 0x0508, /* DTLB_MISSES.MISS_LD */ - }, - [ C(OP_WRITE) ] = { - [ C(RESULT_ACCESS) ] = 0x2240, /* L1D_CACHE_ST.MESI (alias) */ - [ C(RESULT_MISS) ] = 0x0608, /* DTLB_MISSES.MISS_ST */ - }, - [ C(OP_PREFETCH) ] = { - [ C(RESULT_ACCESS) ] = 0, - [ C(RESULT_MISS) ] = 0, - }, - }, - [ C(ITLB) ] = { - [ C(OP_READ) ] = { - [ C(RESULT_ACCESS) ] = 0x00c0, /* INST_RETIRED.ANY_P */ - [ C(RESULT_MISS) ] = 0x0282, /* ITLB.MISSES */ - }, - [ C(OP_WRITE) ] = { - [ C(RESULT_ACCESS) ] = -1, - [ C(RESULT_MISS) ] = -1, - }, - [ C(OP_PREFETCH) ] = { - [ C(RESULT_ACCESS) ] = -1, - [ C(RESULT_MISS) ] = -1, - }, - }, - [ C(BPU ) ] = { - [ C(OP_READ) ] = { - [ C(RESULT_ACCESS) ] = 0x00c4, /* BR_INST_RETIRED.ANY */ - [ C(RESULT_MISS) ] = 0x00c5, /* BP_INST_RETIRED.MISPRED */ - }, - [ C(OP_WRITE) ] = { - [ C(RESULT_ACCESS) ] = -1, - [ C(RESULT_MISS) ] = -1, - }, - [ C(OP_PREFETCH) ] = { - [ C(RESULT_ACCESS) ] = -1, - [ C(RESULT_MISS) ] = -1, - }, - }, -}; - -static u64 intel_pmu_raw_event(u64 hw_event) -{ -#define CORE_EVNTSEL_EVENT_MASK 0x000000FFULL -#define CORE_EVNTSEL_UNIT_MASK 0x0000FF00ULL -#define CORE_EVNTSEL_EDGE_MASK 0x00040000ULL -#define CORE_EVNTSEL_INV_MASK 0x00800000ULL -#define CORE_EVNTSEL_COUNTER_MASK 0xFF000000ULL - -#define CORE_EVNTSEL_MASK \ - (CORE_EVNTSEL_EVENT_MASK | \ - CORE_EVNTSEL_UNIT_MASK | \ - CORE_EVNTSEL_EDGE_MASK | \ - CORE_EVNTSEL_INV_MASK | \ - CORE_EVNTSEL_COUNTER_MASK) - - return hw_event & CORE_EVNTSEL_MASK; -} - -static const u64 amd_hw_cache_event_ids - [PERF_COUNT_HW_CACHE_MAX] - [PERF_COUNT_HW_CACHE_OP_MAX] - [PERF_COUNT_HW_CACHE_RESULT_MAX] = -{ - [ C(L1D) ] = { - [ C(OP_READ) ] = { - [ C(RESULT_ACCESS) ] = 0x0040, /* Data Cache Accesses */ - [ C(RESULT_MISS) ] = 0x0041, /* Data Cache Misses */ - }, - [ C(OP_WRITE) ] = { - [ C(RESULT_ACCESS) ] = 0x0142, /* Data Cache Refills :system */ - [ C(RESULT_MISS) ] = 0, - }, - [ C(OP_PREFETCH) ] = { - [ C(RESULT_ACCESS) ] = 0x0267, /* Data Prefetcher :attempts */ - [ C(RESULT_MISS) ] = 0x0167, /* Data Prefetcher :cancelled */ - }, - }, - [ C(L1I ) ] = { - [ C(OP_READ) ] = { - [ C(RESULT_ACCESS) ] = 0x0080, /* Instruction cache fetches */ - [ C(RESULT_MISS) ] = 0x0081, /* Instruction cache misses */ - }, - [ C(OP_WRITE) ] = { - [ C(RESULT_ACCESS) ] = -1, - [ C(RESULT_MISS) ] = -1, - }, - [ C(OP_PREFETCH) ] = { - [ C(RESULT_ACCESS) ] = 0x014B, /* Prefetch Instructions :Load */ - [ C(RESULT_MISS) ] = 0, - }, - }, - [ C(LL ) ] = { - [ C(OP_READ) ] = { - [ C(RESULT_ACCESS) ] = 0x037D, /* Requests to L2 Cache :IC+DC */ - [ C(RESULT_MISS) ] = 0x037E, /* L2 Cache Misses : IC+DC */ - }, - [ C(OP_WRITE) ] = { - [ C(RESULT_ACCESS) ] = 0x017F, /* L2 Fill/Writeback */ - [ C(RESULT_MISS) ] = 0, - }, - [ C(OP_PREFETCH) ] = { - [ C(RESULT_ACCESS) ] = 0, - [ C(RESULT_MISS) ] = 0, - }, - }, - [ C(DTLB) ] = { - [ C(OP_READ) ] = { - [ C(RESULT_ACCESS) ] = 0x0040, /* Data Cache Accesses */ - [ C(RESULT_MISS) ] = 0x0046, /* L1 DTLB and L2 DLTB Miss */ - }, - [ C(OP_WRITE) ] = { - [ C(RESULT_ACCESS) ] = 0, - [ C(RESULT_MISS) ] = 0, - }, - [ C(OP_PREFETCH) ] = { - [ C(RESULT_ACCESS) ] = 0, - [ C(RESULT_MISS) ] = 0, - }, - }, - [ C(ITLB) ] = { - [ C(OP_READ) ] = { - [ C(RESULT_ACCESS) ] = 0x0080, /* Instruction fecthes */ - [ C(RESULT_MISS) ] = 0x0085, /* Instr. fetch ITLB misses */ - }, - [ C(OP_WRITE) ] = { - [ C(RESULT_ACCESS) ] = -1, - [ C(RESULT_MISS) ] = -1, - }, - [ C(OP_PREFETCH) ] = { - [ C(RESULT_ACCESS) ] = -1, - [ C(RESULT_MISS) ] = -1, - }, - }, - [ C(BPU ) ] = { - [ C(OP_READ) ] = { - [ C(RESULT_ACCESS) ] = 0x00c2, /* Retired Branch Instr. */ - [ C(RESULT_MISS) ] = 0x00c3, /* Retired Mispredicted BI */ - }, - [ C(OP_WRITE) ] = { - [ C(RESULT_ACCESS) ] = -1, - [ C(RESULT_MISS) ] = -1, - }, - [ C(OP_PREFETCH) ] = { - [ C(RESULT_ACCESS) ] = -1, - [ C(RESULT_MISS) ] = -1, - }, - }, -}; - -/* - * AMD Performance Monitor K7 and later. - */ -static const u64 amd_perfmon_event_map[] = -{ - [PERF_COUNT_HW_CPU_CYCLES] = 0x0076, - [PERF_COUNT_HW_INSTRUCTIONS] = 0x00c0, - [PERF_COUNT_HW_CACHE_REFERENCES] = 0x0080, - [PERF_COUNT_HW_CACHE_MISSES] = 0x0081, - [PERF_COUNT_HW_BRANCH_INSTRUCTIONS] = 0x00c4, - [PERF_COUNT_HW_BRANCH_MISSES] = 0x00c5, -}; - -static u64 amd_pmu_event_map(int hw_event) -{ - return amd_perfmon_event_map[hw_event]; -} - -static u64 amd_pmu_raw_event(u64 hw_event) -{ -#define K7_EVNTSEL_EVENT_MASK 0x7000000FFULL -#define K7_EVNTSEL_UNIT_MASK 0x00000FF00ULL -#define K7_EVNTSEL_EDGE_MASK 0x000040000ULL -#define K7_EVNTSEL_INV_MASK 0x000800000ULL -#define K7_EVNTSEL_COUNTER_MASK 0x0FF000000ULL - -#define K7_EVNTSEL_MASK \ - (K7_EVNTSEL_EVENT_MASK | \ - K7_EVNTSEL_UNIT_MASK | \ - K7_EVNTSEL_EDGE_MASK | \ - K7_EVNTSEL_INV_MASK | \ - K7_EVNTSEL_COUNTER_MASK) - - return hw_event & K7_EVNTSEL_MASK; -} - -/* - * Propagate counter elapsed time into the generic counter. - * Can only be executed on the CPU where the counter is active. - * Returns the delta events processed. - */ -static u64 -x86_perf_counter_update(struct perf_counter *counter, - struct hw_perf_counter *hwc, int idx) -{ - int shift = 64 - x86_pmu.counter_bits; - u64 prev_raw_count, new_raw_count; - s64 delta; - - if (idx == X86_PMC_IDX_FIXED_BTS) - return 0; - - /* - * Careful: an NMI might modify the previous counter value. - * - * Our tactic to handle this is to first atomically read and - * exchange a new raw count - then add that new-prev delta - * count to the generic counter atomically: - */ -again: - prev_raw_count = atomic64_read(&hwc->prev_count); - rdmsrl(hwc->counter_base + idx, new_raw_count); - - if (atomic64_cmpxchg(&hwc->prev_count, prev_raw_count, - new_raw_count) != prev_raw_count) - goto again; - - /* - * Now we have the new raw value and have updated the prev - * timestamp already. We can now calculate the elapsed delta - * (counter-)time and add that to the generic counter. - * - * Careful, not all hw sign-extends above the physical width - * of the count. - */ - delta = (new_raw_count << shift) - (prev_raw_count << shift); - delta >>= shift; - - atomic64_add(delta, &counter->count); - atomic64_sub(delta, &hwc->period_left); - - return new_raw_count; -} - -static atomic_t active_counters; -static DEFINE_MUTEX(pmc_reserve_mutex); - -static bool reserve_pmc_hardware(void) -{ -#ifdef CONFIG_X86_LOCAL_APIC - int i; - - if (nmi_watchdog == NMI_LOCAL_APIC) - disable_lapic_nmi_watchdog(); - - for (i = 0; i < x86_pmu.num_counters; i++) { - if (!reserve_perfctr_nmi(x86_pmu.perfctr + i)) - goto perfctr_fail; - } - - for (i = 0; i < x86_pmu.num_counters; i++) { - if (!reserve_evntsel_nmi(x86_pmu.eventsel + i)) - goto eventsel_fail; - } -#endif - - return true; - -#ifdef CONFIG_X86_LOCAL_APIC -eventsel_fail: - for (i--; i >= 0; i--) - release_evntsel_nmi(x86_pmu.eventsel + i); - - i = x86_pmu.num_counters; - -perfctr_fail: - for (i--; i >= 0; i--) - release_perfctr_nmi(x86_pmu.perfctr + i); - - if (nmi_watchdog == NMI_LOCAL_APIC) - enable_lapic_nmi_watchdog(); - - return false; -#endif -} - -static void release_pmc_hardware(void) -{ -#ifdef CONFIG_X86_LOCAL_APIC - int i; - - for (i = 0; i < x86_pmu.num_counters; i++) { - release_perfctr_nmi(x86_pmu.perfctr + i); - release_evntsel_nmi(x86_pmu.eventsel + i); - } - - if (nmi_watchdog == NMI_LOCAL_APIC) - enable_lapic_nmi_watchdog(); -#endif -} - -static inline bool bts_available(void) -{ - return x86_pmu.enable_bts != NULL; -} - -static inline void init_debug_store_on_cpu(int cpu) -{ - struct debug_store *ds = per_cpu(cpu_hw_counters, cpu).ds; - - if (!ds) - return; - - wrmsr_on_cpu(cpu, MSR_IA32_DS_AREA, - (u32)((u64)(unsigned long)ds), - (u32)((u64)(unsigned long)ds >> 32)); -} - -static inline void fini_debug_store_on_cpu(int cpu) -{ - if (!per_cpu(cpu_hw_counters, cpu).ds) - return; - - wrmsr_on_cpu(cpu, MSR_IA32_DS_AREA, 0, 0); -} - -static void release_bts_hardware(void) -{ - int cpu; - - if (!bts_available()) - return; - - get_online_cpus(); - - for_each_online_cpu(cpu) - fini_debug_store_on_cpu(cpu); - - for_each_possible_cpu(cpu) { - struct debug_store *ds = per_cpu(cpu_hw_counters, cpu).ds; - - if (!ds) - continue; - - per_cpu(cpu_hw_counters, cpu).ds = NULL; - - kfree((void *)(unsigned long)ds->bts_buffer_base); - kfree(ds); - } - - put_online_cpus(); -} - -static int reserve_bts_hardware(void) -{ - int cpu, err = 0; - - if (!bts_available()) - return 0; - - get_online_cpus(); - - for_each_possible_cpu(cpu) { - struct debug_store *ds; - void *buffer; - - err = -ENOMEM; - buffer = kzalloc(BTS_BUFFER_SIZE, GFP_KERNEL); - if (unlikely(!buffer)) - break; - - ds = kzalloc(sizeof(*ds), GFP_KERNEL); - if (unlikely(!ds)) { - kfree(buffer); - break; - } - - ds->bts_buffer_base = (u64)(unsigned long)buffer; - ds->bts_index = ds->bts_buffer_base; - ds->bts_absolute_maximum = - ds->bts_buffer_base + BTS_BUFFER_SIZE; - ds->bts_interrupt_threshold = - ds->bts_absolute_maximum - BTS_OVFL_TH; - - per_cpu(cpu_hw_counters, cpu).ds = ds; - err = 0; - } - - if (err) - release_bts_hardware(); - else { - for_each_online_cpu(cpu) - init_debug_store_on_cpu(cpu); - } - - put_online_cpus(); - - return err; -} - -static void hw_perf_counter_destroy(struct perf_counter *counter) -{ - if (atomic_dec_and_mutex_lock(&active_counters, &pmc_reserve_mutex)) { - release_pmc_hardware(); - release_bts_hardware(); - mutex_unlock(&pmc_reserve_mutex); - } -} - -static inline int x86_pmu_initialized(void) -{ - return x86_pmu.handle_irq != NULL; -} - -static inline int -set_ext_hw_attr(struct hw_perf_counter *hwc, struct perf_counter_attr *attr) -{ - unsigned int cache_type, cache_op, cache_result; - u64 config, val; - - config = attr->config; - - cache_type = (config >> 0) & 0xff; - if (cache_type >= PERF_COUNT_HW_CACHE_MAX) - return -EINVAL; - - cache_op = (config >> 8) & 0xff; - if (cache_op >= PERF_COUNT_HW_CACHE_OP_MAX) - return -EINVAL; - - cache_result = (config >> 16) & 0xff; - if (cache_result >= PERF_COUNT_HW_CACHE_RESULT_MAX) - return -EINVAL; - - val = hw_cache_event_ids[cache_type][cache_op][cache_result]; - - if (val == 0) - return -ENOENT; - - if (val == -1) - return -EINVAL; - - hwc->config |= val; - - return 0; -} - -static void intel_pmu_enable_bts(u64 config) -{ - unsigned long debugctlmsr; - - debugctlmsr = get_debugctlmsr(); - - debugctlmsr |= X86_DEBUGCTL_TR; - debugctlmsr |= X86_DEBUGCTL_BTS; - debugctlmsr |= X86_DEBUGCTL_BTINT; - - if (!(config & ARCH_PERFMON_EVENTSEL_OS)) - debugctlmsr |= X86_DEBUGCTL_BTS_OFF_OS; - - if (!(config & ARCH_PERFMON_EVENTSEL_USR)) - debugctlmsr |= X86_DEBUGCTL_BTS_OFF_USR; - - update_debugctlmsr(debugctlmsr); -} - -static void intel_pmu_disable_bts(void) -{ - struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters); - unsigned long debugctlmsr; - - if (!cpuc->ds) - return; - - debugctlmsr = get_debugctlmsr(); - - debugctlmsr &= - ~(X86_DEBUGCTL_TR | X86_DEBUGCTL_BTS | X86_DEBUGCTL_BTINT | - X86_DEBUGCTL_BTS_OFF_OS | X86_DEBUGCTL_BTS_OFF_USR); - - update_debugctlmsr(debugctlmsr); -} - -/* - * Setup the hardware configuration for a given attr_type - */ -static int __hw_perf_counter_init(struct perf_counter *counter) -{ - struct perf_counter_attr *attr = &counter->attr; - struct hw_perf_counter *hwc = &counter->hw; - u64 config; - int err; - - if (!x86_pmu_initialized()) - return -ENODEV; - - err = 0; - if (!atomic_inc_not_zero(&active_counters)) { - mutex_lock(&pmc_reserve_mutex); - if (atomic_read(&active_counters) == 0) { - if (!reserve_pmc_hardware()) - err = -EBUSY; - else - err = reserve_bts_hardware(); - } - if (!err) - atomic_inc(&active_counters); - mutex_unlock(&pmc_reserve_mutex); - } - if (err) - return err; - - counter->destroy = hw_perf_counter_destroy; - - /* - * Generate PMC IRQs: - * (keep 'enabled' bit clear for now) - */ - hwc->config = ARCH_PERFMON_EVENTSEL_INT; - - /* - * Count user and OS events unless requested not to. - */ - if (!attr->exclude_user) - hwc->config |= ARCH_PERFMON_EVENTSEL_USR; - if (!attr->exclude_kernel) - hwc->config |= ARCH_PERFMON_EVENTSEL_OS; - - if (!hwc->sample_period) { - hwc->sample_period = x86_pmu.max_period; - hwc->last_period = hwc->sample_period; - atomic64_set(&hwc->period_left, hwc->sample_period); - } else { - /* - * If we have a PMU initialized but no APIC - * interrupts, we cannot sample hardware - * counters (user-space has to fall back and - * sample via a hrtimer based software counter): - */ - if (!x86_pmu.apic) - return -EOPNOTSUPP; - } - - /* - * Raw hw_event type provide the config in the hw_event structure - */ - if (attr->type == PERF_TYPE_RAW) { - hwc->config |= x86_pmu.raw_event(attr->config); - return 0; - } - - if (attr->type == PERF_TYPE_HW_CACHE) - return set_ext_hw_attr(hwc, attr); - - if (attr->config >= x86_pmu.max_events) - return -EINVAL; - - /* - * The generic map: - */ - config = x86_pmu.event_map(attr->config); - - if (config == 0) - return -ENOENT; - - if (config == -1LL) - return -EINVAL; - - /* - * Branch tracing: - */ - if ((attr->config == PERF_COUNT_HW_BRANCH_INSTRUCTIONS) && - (hwc->sample_period == 1)) { - /* BTS is not supported by this architecture. */ - if (!bts_available()) - return -EOPNOTSUPP; - - /* BTS is currently only allowed for user-mode. */ - if (hwc->config & ARCH_PERFMON_EVENTSEL_OS) - return -EOPNOTSUPP; - } - - hwc->config |= config; - - return 0; -} - -static void p6_pmu_disable_all(void) -{ - struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters); - u64 val; - - if (!cpuc->enabled) - return; - - cpuc->enabled = 0; - barrier(); - - /* p6 only has one enable register */ - rdmsrl(MSR_P6_EVNTSEL0, val); - val &= ~ARCH_PERFMON_EVENTSEL0_ENABLE; - wrmsrl(MSR_P6_EVNTSEL0, val); -} - -static void intel_pmu_disable_all(void) -{ - struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters); - - if (!cpuc->enabled) - return; - - cpuc->enabled = 0; - barrier(); - - wrmsrl(MSR_CORE_PERF_GLOBAL_CTRL, 0); - - if (test_bit(X86_PMC_IDX_FIXED_BTS, cpuc->active_mask)) - intel_pmu_disable_bts(); -} - -static void amd_pmu_disable_all(void) -{ - struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters); - int idx; - - if (!cpuc->enabled) - return; - - cpuc->enabled = 0; - /* - * ensure we write the disable before we start disabling the - * counters proper, so that amd_pmu_enable_counter() does the - * right thing. - */ - barrier(); - - for (idx = 0; idx < x86_pmu.num_counters; idx++) { - u64 val; - - if (!test_bit(idx, cpuc->active_mask)) - continue; - rdmsrl(MSR_K7_EVNTSEL0 + idx, val); - if (!(val & ARCH_PERFMON_EVENTSEL0_ENABLE)) - continue; - val &= ~ARCH_PERFMON_EVENTSEL0_ENABLE; - wrmsrl(MSR_K7_EVNTSEL0 + idx, val); - } -} - -void hw_perf_disable(void) -{ - if (!x86_pmu_initialized()) - return; - return x86_pmu.disable_all(); -} - -static void p6_pmu_enable_all(void) -{ - struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters); - unsigned long val; - - if (cpuc->enabled) - return; - - cpuc->enabled = 1; - barrier(); - - /* p6 only has one enable register */ - rdmsrl(MSR_P6_EVNTSEL0, val); - val |= ARCH_PERFMON_EVENTSEL0_ENABLE; - wrmsrl(MSR_P6_EVNTSEL0, val); -} - -static void intel_pmu_enable_all(void) -{ - struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters); - - if (cpuc->enabled) - return; - - cpuc->enabled = 1; - barrier(); - - wrmsrl(MSR_CORE_PERF_GLOBAL_CTRL, x86_pmu.intel_ctrl); - - if (test_bit(X86_PMC_IDX_FIXED_BTS, cpuc->active_mask)) { - struct perf_counter *counter = - cpuc->counters[X86_PMC_IDX_FIXED_BTS]; - - if (WARN_ON_ONCE(!counter)) - return; - - intel_pmu_enable_bts(counter->hw.config); - } -} - -static void amd_pmu_enable_all(void) -{ - struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters); - int idx; - - if (cpuc->enabled) - return; - - cpuc->enabled = 1; - barrier(); - - for (idx = 0; idx < x86_pmu.num_counters; idx++) { - struct perf_counter *counter = cpuc->counters[idx]; - u64 val; - - if (!test_bit(idx, cpuc->active_mask)) - continue; - - val = counter->hw.config; - val |= ARCH_PERFMON_EVENTSEL0_ENABLE; - wrmsrl(MSR_K7_EVNTSEL0 + idx, val); - } -} - -void hw_perf_enable(void) -{ - if (!x86_pmu_initialized()) - return; - x86_pmu.enable_all(); -} - -static inline u64 intel_pmu_get_status(void) -{ - u64 status; - - rdmsrl(MSR_CORE_PERF_GLOBAL_STATUS, status); - - return status; -} - -static inline void intel_pmu_ack_status(u64 ack) -{ - wrmsrl(MSR_CORE_PERF_GLOBAL_OVF_CTRL, ack); -} - -static inline void x86_pmu_enable_counter(struct hw_perf_counter *hwc, int idx) -{ - (void)checking_wrmsrl(hwc->config_base + idx, - hwc->config | ARCH_PERFMON_EVENTSEL0_ENABLE); -} - -static inline void x86_pmu_disable_counter(struct hw_perf_counter *hwc, int idx) -{ - (void)checking_wrmsrl(hwc->config_base + idx, hwc->config); -} - -static inline void -intel_pmu_disable_fixed(struct hw_perf_counter *hwc, int __idx) -{ - int idx = __idx - X86_PMC_IDX_FIXED; - u64 ctrl_val, mask; - - mask = 0xfULL << (idx * 4); - - rdmsrl(hwc->config_base, ctrl_val); - ctrl_val &= ~mask; - (void)checking_wrmsrl(hwc->config_base, ctrl_val); -} - -static inline void -p6_pmu_disable_counter(struct hw_perf_counter *hwc, int idx) -{ - struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters); - u64 val = P6_NOP_COUNTER; - - if (cpuc->enabled) - val |= ARCH_PERFMON_EVENTSEL0_ENABLE; - - (void)checking_wrmsrl(hwc->config_base + idx, val); -} - -static inline void -intel_pmu_disable_counter(struct hw_perf_counter *hwc, int idx) -{ - if (unlikely(idx == X86_PMC_IDX_FIXED_BTS)) { - intel_pmu_disable_bts(); - return; - } - - if (unlikely(hwc->config_base == MSR_ARCH_PERFMON_FIXED_CTR_CTRL)) { - intel_pmu_disable_fixed(hwc, idx); - return; - } - - x86_pmu_disable_counter(hwc, idx); -} - -static inline void -amd_pmu_disable_counter(struct hw_perf_counter *hwc, int idx) -{ - x86_pmu_disable_counter(hwc, idx); -} - -static DEFINE_PER_CPU(u64 [X86_PMC_IDX_MAX], pmc_prev_left); - -/* - * Set the next IRQ period, based on the hwc->period_left value. - * To be called with the counter disabled in hw: - */ -static int -x86_perf_counter_set_period(struct perf_counter *counter, - struct hw_perf_counter *hwc, int idx) -{ - s64 left = atomic64_read(&hwc->period_left); - s64 period = hwc->sample_period; - int err, ret = 0; - - if (idx == X86_PMC_IDX_FIXED_BTS) - return 0; - - /* - * If we are way outside a reasoable range then just skip forward: - */ - if (unlikely(left <= -period)) { - left = period; - atomic64_set(&hwc->period_left, left); - hwc->last_period = period; - ret = 1; - } - - if (unlikely(left <= 0)) { - left += period; - atomic64_set(&hwc->period_left, left); - hwc->last_period = period; - ret = 1; - } - /* - * Quirk: certain CPUs dont like it if just 1 hw_event is left: - */ - if (unlikely(left < 2)) - left = 2; - - if (left > x86_pmu.max_period) - left = x86_pmu.max_period; - - per_cpu(pmc_prev_left[idx], smp_processor_id()) = left; - - /* - * The hw counter starts counting from this counter offset, - * mark it to be able to extra future deltas: - */ - atomic64_set(&hwc->prev_count, (u64)-left); - - err = checking_wrmsrl(hwc->counter_base + idx, - (u64)(-left) & x86_pmu.counter_mask); - - perf_counter_update_userpage(counter); - - return ret; -} - -static inline void -intel_pmu_enable_fixed(struct hw_perf_counter *hwc, int __idx) -{ - int idx = __idx - X86_PMC_IDX_FIXED; - u64 ctrl_val, bits, mask; - int err; - - /* - * Enable IRQ generation (0x8), - * and enable ring-3 counting (0x2) and ring-0 counting (0x1) - * if requested: - */ - bits = 0x8ULL; - if (hwc->config & ARCH_PERFMON_EVENTSEL_USR) - bits |= 0x2; - if (hwc->config & ARCH_PERFMON_EVENTSEL_OS) - bits |= 0x1; - bits <<= (idx * 4); - mask = 0xfULL << (idx * 4); - - rdmsrl(hwc->config_base, ctrl_val); - ctrl_val &= ~mask; - ctrl_val |= bits; - err = checking_wrmsrl(hwc->config_base, ctrl_val); -} - -static void p6_pmu_enable_counter(struct hw_perf_counter *hwc, int idx) -{ - struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters); - u64 val; - - val = hwc->config; - if (cpuc->enabled) - val |= ARCH_PERFMON_EVENTSEL0_ENABLE; - - (void)checking_wrmsrl(hwc->config_base + idx, val); -} - - -static void intel_pmu_enable_counter(struct hw_perf_counter *hwc, int idx) -{ - if (unlikely(idx == X86_PMC_IDX_FIXED_BTS)) { - if (!__get_cpu_var(cpu_hw_counters).enabled) - return; - - intel_pmu_enable_bts(hwc->config); - return; - } - - if (unlikely(hwc->config_base == MSR_ARCH_PERFMON_FIXED_CTR_CTRL)) { - intel_pmu_enable_fixed(hwc, idx); - return; - } - - x86_pmu_enable_counter(hwc, idx); -} - -static void amd_pmu_enable_counter(struct hw_perf_counter *hwc, int idx) -{ - struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters); - - if (cpuc->enabled) - x86_pmu_enable_counter(hwc, idx); -} - -static int -fixed_mode_idx(struct perf_counter *counter, struct hw_perf_counter *hwc) -{ - unsigned int hw_event; - - hw_event = hwc->config & ARCH_PERFMON_EVENT_MASK; - - if (unlikely((hw_event == - x86_pmu.event_map(PERF_COUNT_HW_BRANCH_INSTRUCTIONS)) && - (hwc->sample_period == 1))) - return X86_PMC_IDX_FIXED_BTS; - - if (!x86_pmu.num_counters_fixed) - return -1; - - if (unlikely(hw_event == x86_pmu.event_map(PERF_COUNT_HW_INSTRUCTIONS))) - return X86_PMC_IDX_FIXED_INSTRUCTIONS; - if (unlikely(hw_event == x86_pmu.event_map(PERF_COUNT_HW_CPU_CYCLES))) - return X86_PMC_IDX_FIXED_CPU_CYCLES; - if (unlikely(hw_event == x86_pmu.event_map(PERF_COUNT_HW_BUS_CYCLES))) - return X86_PMC_IDX_FIXED_BUS_CYCLES; - - return -1; -} - -/* - * Find a PMC slot for the freshly enabled / scheduled in counter: - */ -static int x86_pmu_enable(struct perf_counter *counter) -{ - struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters); - struct hw_perf_counter *hwc = &counter->hw; - int idx; - - idx = fixed_mode_idx(counter, hwc); - if (idx == X86_PMC_IDX_FIXED_BTS) { - /* BTS is already occupied. */ - if (test_and_set_bit(idx, cpuc->used_mask)) - return -EAGAIN; - - hwc->config_base = 0; - hwc->counter_base = 0; - hwc->idx = idx; - } else if (idx >= 0) { - /* - * Try to get the fixed counter, if that is already taken - * then try to get a generic counter: - */ - if (test_and_set_bit(idx, cpuc->used_mask)) - goto try_generic; - - hwc->config_base = MSR_ARCH_PERFMON_FIXED_CTR_CTRL; - /* - * We set it so that counter_base + idx in wrmsr/rdmsr maps to - * MSR_ARCH_PERFMON_FIXED_CTR0 ... CTR2: - */ - hwc->counter_base = - MSR_ARCH_PERFMON_FIXED_CTR0 - X86_PMC_IDX_FIXED; - hwc->idx = idx; - } else { - idx = hwc->idx; - /* Try to get the previous generic counter again */ - if (test_and_set_bit(idx, cpuc->used_mask)) { -try_generic: - idx = find_first_zero_bit(cpuc->used_mask, - x86_pmu.num_counters); - if (idx == x86_pmu.num_counters) - return -EAGAIN; - - set_bit(idx, cpuc->used_mask); - hwc->idx = idx; - } - hwc->config_base = x86_pmu.eventsel; - hwc->counter_base = x86_pmu.perfctr; - } - - perf_counters_lapic_init(); - - x86_pmu.disable(hwc, idx); - - cpuc->counters[idx] = counter; - set_bit(idx, cpuc->active_mask); - - x86_perf_counter_set_period(counter, hwc, idx); - x86_pmu.enable(hwc, idx); - - perf_counter_update_userpage(counter); - - return 0; -} - -static void x86_pmu_unthrottle(struct perf_counter *counter) -{ - struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters); - struct hw_perf_counter *hwc = &counter->hw; - - if (WARN_ON_ONCE(hwc->idx >= X86_PMC_IDX_MAX || - cpuc->counters[hwc->idx] != counter)) - return; - - x86_pmu.enable(hwc, hwc->idx); -} - -void perf_counter_print_debug(void) -{ - u64 ctrl, status, overflow, pmc_ctrl, pmc_count, prev_left, fixed; - struct cpu_hw_counters *cpuc; - unsigned long flags; - int cpu, idx; - - if (!x86_pmu.num_counters) - return; - - local_irq_save(flags); - - cpu = smp_processor_id(); - cpuc = &per_cpu(cpu_hw_counters, cpu); - - if (x86_pmu.version >= 2) { - rdmsrl(MSR_CORE_PERF_GLOBAL_CTRL, ctrl); - rdmsrl(MSR_CORE_PERF_GLOBAL_STATUS, status); - rdmsrl(MSR_CORE_PERF_GLOBAL_OVF_CTRL, overflow); - rdmsrl(MSR_ARCH_PERFMON_FIXED_CTR_CTRL, fixed); - - pr_info("\n"); - pr_info("CPU#%d: ctrl: %016llx\n", cpu, ctrl); - pr_info("CPU#%d: status: %016llx\n", cpu, status); - pr_info("CPU#%d: overflow: %016llx\n", cpu, overflow); - pr_info("CPU#%d: fixed: %016llx\n", cpu, fixed); - } - pr_info("CPU#%d: used: %016llx\n", cpu, *(u64 *)cpuc->used_mask); - - for (idx = 0; idx < x86_pmu.num_counters; idx++) { - rdmsrl(x86_pmu.eventsel + idx, pmc_ctrl); - rdmsrl(x86_pmu.perfctr + idx, pmc_count); - - prev_left = per_cpu(pmc_prev_left[idx], cpu); - - pr_info("CPU#%d: gen-PMC%d ctrl: %016llx\n", - cpu, idx, pmc_ctrl); - pr_info("CPU#%d: gen-PMC%d count: %016llx\n", - cpu, idx, pmc_count); - pr_info("CPU#%d: gen-PMC%d left: %016llx\n", - cpu, idx, prev_left); - } - for (idx = 0; idx < x86_pmu.num_counters_fixed; idx++) { - rdmsrl(MSR_ARCH_PERFMON_FIXED_CTR0 + idx, pmc_count); - - pr_info("CPU#%d: fixed-PMC%d count: %016llx\n", - cpu, idx, pmc_count); - } - local_irq_restore(flags); -} - -static void intel_pmu_drain_bts_buffer(struct cpu_hw_counters *cpuc) -{ - struct debug_store *ds = cpuc->ds; - struct bts_record { - u64 from; - u64 to; - u64 flags; - }; - struct perf_counter *counter = cpuc->counters[X86_PMC_IDX_FIXED_BTS]; - struct bts_record *at, *top; - struct perf_output_handle handle; - struct perf_event_header header; - struct perf_sample_data data; - struct pt_regs regs; - - if (!counter) - return; - - if (!ds) - return; - - at = (struct bts_record *)(unsigned long)ds->bts_buffer_base; - top = (struct bts_record *)(unsigned long)ds->bts_index; - - if (top <= at) - return; - - ds->bts_index = ds->bts_buffer_base; - - - data.period = counter->hw.last_period; - data.addr = 0; - regs.ip = 0; - - /* - * Prepare a generic sample, i.e. fill in the invariant fields. - * We will overwrite the from and to address before we output - * the sample. - */ - perf_prepare_sample(&header, &data, counter, ®s); - - if (perf_output_begin(&handle, counter, - header.size * (top - at), 1, 1)) - return; - - for (; at < top; at++) { - data.ip = at->from; - data.addr = at->to; - - perf_output_sample(&handle, &header, &data, counter); - } - - perf_output_end(&handle); - - /* There's new data available. */ - counter->hw.interrupts++; - counter->pending_kill = POLL_IN; -} - -static void x86_pmu_disable(struct perf_counter *counter) -{ - struct cpu_hw_counters *cpuc = &__get_cpu_var(cpu_hw_counters); - struct hw_perf_counter *hwc = &counter->hw; - int idx = hwc->idx; - - /* - * Must be done before we disable, otherwise the nmi handler - * could reenable again: - */ - clear_bit(idx, cpuc->active_mask); - x86_pmu.disable(hwc, idx); - - /* - * Make sure the cleared pointer becomes visible before we - * (potentially) free the counter: - */ - barrier(); - - /* - * Drain the remaining delta count out of a counter - * that we are disabling: - */ - x86_perf_counter_update(counter, hwc, idx); - - /* Drain the remaining BTS records. */ - if (unlikely(idx == X86_PMC_IDX_FIXED_BTS)) - intel_pmu_drain_bts_buffer(cpuc); - - cpuc->counters[idx] = NULL; - clear_bit(idx, cpuc->used_mask); - - perf_counter_update_userpage(counter); -} - -/* - * Save and restart an expired counter. Called by NMI contexts, - * so it has to be careful about preempting normal counter ops: - */ -static int intel_pmu_save_and_restart(struct perf_counter *counter) -{ - struct hw_perf_counter *hwc = &counter->hw; - int idx = hwc->idx; - int ret; - - x86_perf_counter_update(counter, hwc, idx); - ret = x86_perf_counter_set_period(counter, hwc, idx); - - if (counter->state == PERF_COUNTER_STATE_ACTIVE) - intel_pmu_enable_counter(hwc, idx); - - return ret; -} - -static void intel_pmu_reset(void) -{ - struct debug_store *ds = __get_cpu_var(cpu_hw_counters).ds; - unsigned long flags; - int idx; - - if (!x86_pmu.num_counters) - return; - - local_irq_save(flags); - - printk("clearing PMU state on CPU#%d\n", smp_processor_id()); - - for (idx = 0; idx < x86_pmu.num_counters; idx++) { - checking_wrmsrl(x86_pmu.eventsel + idx, 0ull); - checking_wrmsrl(x86_pmu.perfctr + idx, 0ull); - } - for (idx = 0; idx < x86_pmu.num_counters_fixed; idx++) { - checking_wrmsrl(MSR_ARCH_PERFMON_FIXED_CTR0 + idx, 0ull); - } - if (ds) - ds->bts_index = ds->bts_buffer_base; - - local_irq_restore(flags); -} - -static int p6_pmu_handle_irq(struct pt_regs *regs) -{ - struct perf_sample_data data; - struct cpu_hw_counters *cpuc; - struct perf_counter *counter; - struct hw_perf_counter *hwc; - int idx, handled = 0; - u64 val; - - data.addr = 0; - - cpuc = &__get_cpu_var(cpu_hw_counters); - - for (idx = 0; idx < x86_pmu.num_counters; idx++) { - if (!test_bit(idx, cpuc->active_mask)) - continue; - - counter = cpuc->counters[idx]; - hwc = &counter->hw; - - val = x86_perf_counter_update(counter, hwc, idx); - if (val & (1ULL << (x86_pmu.counter_bits - 1))) - continue; - - /* - * counter overflow - */ - handled = 1; - data.period = counter->hw.last_period; - - if (!x86_perf_counter_set_period(counter, hwc, idx)) - continue; - - if (perf_counter_overflow(counter, 1, &data, regs)) - p6_pmu_disable_counter(hwc, idx); - } - - if (handled) - inc_irq_stat(apic_perf_irqs); - - return handled; -} - -/* - * This handler is triggered by the local APIC, so the APIC IRQ handling - * rules apply: - */ -static int intel_pmu_handle_irq(struct pt_regs *regs) -{ - struct perf_sample_data data; - struct cpu_hw_counters *cpuc; - int bit, loops; - u64 ack, status; - - data.addr = 0; - - cpuc = &__get_cpu_var(cpu_hw_counters); - - perf_disable(); - intel_pmu_drain_bts_buffer(cpuc); - status = intel_pmu_get_status(); - if (!status) { - perf_enable(); - return 0; - } - - loops = 0; -again: - if (++loops > 100) { - WARN_ONCE(1, "perfcounters: irq loop stuck!\n"); - perf_counter_print_debug(); - intel_pmu_reset(); - perf_enable(); - return 1; - } - - inc_irq_stat(apic_perf_irqs); - ack = status; - for_each_bit(bit, (unsigned long *)&status, X86_PMC_IDX_MAX) { - struct perf_counter *counter = cpuc->counters[bit]; - - clear_bit(bit, (unsigned long *) &status); - if (!test_bit(bit, cpuc->active_mask)) - continue; - - if (!intel_pmu_save_and_restart(counter)) - continue; - - data.period = counter->hw.last_period; - - if (perf_counter_overflow(counter, 1, &data, regs)) - intel_pmu_disable_counter(&counter->hw, bit); - } - - intel_pmu_ack_status(ack); - - /* - * Repeat if there is more work to be done: - */ - status = intel_pmu_get_status(); - if (status) - goto again; - - perf_enable(); - - return 1; -} - -static int amd_pmu_handle_irq(struct pt_regs *regs) -{ - struct perf_sample_data data; - struct cpu_hw_counters *cpuc; - struct perf_counter *counter; - struct hw_perf_counter *hwc; - int idx, handled = 0; - u64 val; - - data.addr = 0; - - cpuc = &__get_cpu_var(cpu_hw_counters); - - for (idx = 0; idx < x86_pmu.num_counters; idx++) { - if (!test_bit(idx, cpuc->active_mask)) - continue; - - counter = cpuc->counters[idx]; - hwc = &counter->hw; - - val = x86_perf_counter_update(counter, hwc, idx); - if (val & (1ULL << (x86_pmu.counter_bits - 1))) - continue; - - /* - * counter overflow - */ - handled = 1; - data.period = counter->hw.last_period; - - if (!x86_perf_counter_set_period(counter, hwc, idx)) - continue; - - if (perf_counter_overflow(counter, 1, &data, regs)) - amd_pmu_disable_counter(hwc, idx); - } - - if (handled) - inc_irq_stat(apic_perf_irqs); - - return handled; -} - -void smp_perf_pending_interrupt(struct pt_regs *regs) -{ - irq_enter(); - ack_APIC_irq(); - inc_irq_stat(apic_pending_irqs); - perf_counter_do_pending(); - irq_exit(); -} - -void set_perf_counter_pending(void) -{ -#ifdef CONFIG_X86_LOCAL_APIC - apic->send_IPI_self(LOCAL_PENDING_VECTOR); -#endif -} - -void perf_counters_lapic_init(void) -{ -#ifdef CONFIG_X86_LOCAL_APIC - if (!x86_pmu.apic || !x86_pmu_initialized()) - return; - - /* - * Always use NMI for PMU - */ - apic_write(APIC_LVTPC, APIC_DM_NMI); -#endif -} - -static int __kprobes -perf_counter_nmi_handler(struct notifier_block *self, - unsigned long cmd, void *__args) -{ - struct die_args *args = __args; - struct pt_regs *regs; - - if (!atomic_read(&active_counters)) - return NOTIFY_DONE; - - switch (cmd) { - case DIE_NMI: - case DIE_NMI_IPI: - break; - - default: - return NOTIFY_DONE; - } - - regs = args->regs; - -#ifdef CONFIG_X86_LOCAL_APIC - apic_write(APIC_LVTPC, APIC_DM_NMI); -#endif - /* - * Can't rely on the handled return value to say it was our NMI, two - * counters could trigger 'simultaneously' raising two back-to-back NMIs. - * - * If the first NMI handles both, the latter will be empty and daze - * the CPU. - */ - x86_pmu.handle_irq(regs); - - return NOTIFY_STOP; -} - -static __read_mostly struct notifier_block perf_counter_nmi_notifier = { - .notifier_call = perf_counter_nmi_handler, - .next = NULL, - .priority = 1 -}; - -static struct x86_pmu p6_pmu = { - .name = "p6", - .handle_irq = p6_pmu_handle_irq, - .disable_all = p6_pmu_disable_all, - .enable_all = p6_pmu_enable_all, - .enable = p6_pmu_enable_counter, - .disable = p6_pmu_disable_counter, - .eventsel = MSR_P6_EVNTSEL0, - .perfctr = MSR_P6_PERFCTR0, - .event_map = p6_pmu_event_map, - .raw_event = p6_pmu_raw_event, - .max_events = ARRAY_SIZE(p6_perfmon_event_map), - .apic = 1, - .max_period = (1ULL << 31) - 1, - .version = 0, - .num_counters = 2, - /* - * Counters have 40 bits implemented. However they are designed such - * that bits [32-39] are sign extensions of bit 31. As such the - * effective width of a counter for P6-like PMU is 32 bits only. - * - * See IA-32 Intel Architecture Software developer manual Vol 3B - */ - .counter_bits = 32, - .counter_mask = (1ULL << 32) - 1, -}; - -static struct x86_pmu intel_pmu = { - .name = "Intel", - .handle_irq = intel_pmu_handle_irq, - .disable_all = intel_pmu_disable_all, - .enable_all = intel_pmu_enable_all, - .enable = intel_pmu_enable_counter, - .disable = intel_pmu_disable_counter, - .eventsel = MSR_ARCH_PERFMON_EVENTSEL0, - .perfctr = MSR_ARCH_PERFMON_PERFCTR0, - .event_map = intel_pmu_event_map, - .raw_event = intel_pmu_raw_event, - .max_events = ARRAY_SIZE(intel_perfmon_event_map), - .apic = 1, - /* - * Intel PMCs cannot be accessed sanely above 32 bit width, - * so we install an artificial 1<<31 period regardless of - * the generic counter period: - */ - .max_period = (1ULL << 31) - 1, - .enable_bts = intel_pmu_enable_bts, - .disable_bts = intel_pmu_disable_bts, -}; - -static struct x86_pmu amd_pmu = { - .name = "AMD", - .handle_irq = amd_pmu_handle_irq, - .disable_all = amd_pmu_disable_all, - .enable_all = amd_pmu_enable_all, - .enable = amd_pmu_enable_counter, - .disable = amd_pmu_disable_counter, - .eventsel = MSR_K7_EVNTSEL0, - .perfctr = MSR_K7_PERFCTR0, - .event_map = amd_pmu_event_map, - .raw_event = amd_pmu_raw_event, - .max_events = ARRAY_SIZE(amd_perfmon_event_map), - .num_counters = 4, - .counter_bits = 48, - .counter_mask = (1ULL << 48) - 1, - .apic = 1, - /* use highest bit to detect overflow */ - .max_period = (1ULL << 47) - 1, -}; - -static int p6_pmu_init(void) -{ - switch (boot_cpu_data.x86_model) { - case 1: - case 3: /* Pentium Pro */ - case 5: - case 6: /* Pentium II */ - case 7: - case 8: - case 11: /* Pentium III */ - break; - case 9: - case 13: - /* Pentium M */ - break; - default: - pr_cont("unsupported p6 CPU model %d ", - boot_cpu_data.x86_model); - return -ENODEV; - } - - x86_pmu = p6_pmu; - - if (!cpu_has_apic) { - pr_info("no APIC, boot with the \"lapic\" boot parameter to force-enable it.\n"); - pr_info("no hardware sampling interrupt available.\n"); - x86_pmu.apic = 0; - } - - return 0; -} - -static int intel_pmu_init(void) -{ - union cpuid10_edx edx; - union cpuid10_eax eax; - unsigned int unused; - unsigned int ebx; - int version; - - if (!cpu_has(&boot_cpu_data, X86_FEATURE_ARCH_PERFMON)) { - /* check for P6 processor family */ - if (boot_cpu_data.x86 == 6) { - return p6_pmu_init(); - } else { - return -ENODEV; - } - } - - /* - * Check whether the Architectural PerfMon supports - * Branch Misses Retired hw_event or not. - */ - cpuid(10, &eax.full, &ebx, &unused, &edx.full); - if (eax.split.mask_length <= ARCH_PERFMON_BRANCH_MISSES_RETIRED) - return -ENODEV; - - version = eax.split.version_id; - if (version < 2) - return -ENODEV; - - x86_pmu = intel_pmu; - x86_pmu.version = version; - x86_pmu.num_counters = eax.split.num_counters; - x86_pmu.counter_bits = eax.split.bit_width; - x86_pmu.counter_mask = (1ULL << eax.split.bit_width) - 1; - - /* - * Quirk: v2 perfmon does not report fixed-purpose counters, so - * assume at least 3 counters: - */ - x86_pmu.num_counters_fixed = max((int)edx.split.num_counters_fixed, 3); - - /* - * Install the hw-cache-events table: - */ - switch (boot_cpu_data.x86_model) { - case 15: /* original 65 nm celeron/pentium/core2/xeon, "Merom"/"Conroe" */ - case 22: /* single-core 65 nm celeron/core2solo "Merom-L"/"Conroe-L" */ - case 23: /* current 45 nm celeron/core2/xeon "Penryn"/"Wolfdale" */ - case 29: /* six-core 45 nm xeon "Dunnington" */ - memcpy(hw_cache_event_ids, core2_hw_cache_event_ids, - sizeof(hw_cache_event_ids)); - - pr_cont("Core2 events, "); - break; - default: - case 26: - memcpy(hw_cache_event_ids, nehalem_hw_cache_event_ids, - sizeof(hw_cache_event_ids)); - - pr_cont("Nehalem/Corei7 events, "); - break; - case 28: - memcpy(hw_cache_event_ids, atom_hw_cache_event_ids, - sizeof(hw_cache_event_ids)); - - pr_cont("Atom events, "); - break; - } - return 0; -} - -static int amd_pmu_init(void) -{ - /* Performance-monitoring supported from K7 and later: */ - if (boot_cpu_data.x86 < 6) - return -ENODEV; - - x86_pmu = amd_pmu; - - /* Events are common for all AMDs */ - memcpy(hw_cache_event_ids, amd_hw_cache_event_ids, - sizeof(hw_cache_event_ids)); - - return 0; -} - -void __init init_hw_perf_counters(void) -{ - int err; - - pr_info("Performance Counters: "); - - switch (boot_cpu_data.x86_vendor) { - case X86_VENDOR_INTEL: - err = intel_pmu_init(); - break; - case X86_VENDOR_AMD: - err = amd_pmu_init(); - break; - default: - return; - } - if (err != 0) { - pr_cont("no PMU driver, software counters only.\n"); - return; - } - - pr_cont("%s PMU driver.\n", x86_pmu.name); - - if (x86_pmu.num_counters > X86_PMC_MAX_GENERIC) { - WARN(1, KERN_ERR "hw perf counters %d > max(%d), clipping!", - x86_pmu.num_counters, X86_PMC_MAX_GENERIC); - x86_pmu.num_counters = X86_PMC_MAX_GENERIC; - } - perf_counter_mask = (1 << x86_pmu.num_counters) - 1; - perf_max_counters = x86_pmu.num_counters; - - if (x86_pmu.num_counters_fixed > X86_PMC_MAX_FIXED) { - WARN(1, KERN_ERR "hw perf counters fixed %d > max(%d), clipping!", - x86_pmu.num_counters_fixed, X86_PMC_MAX_FIXED); - x86_pmu.num_counters_fixed = X86_PMC_MAX_FIXED; - } - - perf_counter_mask |= - ((1LL << x86_pmu.num_counters_fixed)-1) << X86_PMC_IDX_FIXED; - x86_pmu.intel_ctrl = perf_counter_mask; - - perf_counters_lapic_init(); - register_die_notifier(&perf_counter_nmi_notifier); - - pr_info("... version: %d\n", x86_pmu.version); - pr_info("... bit width: %d\n", x86_pmu.counter_bits); - pr_info("... generic counters: %d\n", x86_pmu.num_counters); - pr_info("... value mask: %016Lx\n", x86_pmu.counter_mask); - pr_info("... max period: %016Lx\n", x86_pmu.max_period); - pr_info("... fixed-purpose counters: %d\n", x86_pmu.num_counters_fixed); - pr_info("... counter mask: %016Lx\n", perf_counter_mask); -} - -static inline void x86_pmu_read(struct perf_counter *counter) -{ - x86_perf_counter_update(counter, &counter->hw, counter->hw.idx); -} - -static const struct pmu pmu = { - .enable = x86_pmu_enable, - .disable = x86_pmu_disable, - .read = x86_pmu_read, - .unthrottle = x86_pmu_unthrottle, -}; - -const struct pmu *hw_perf_counter_init(struct perf_counter *counter) -{ - int err; - - err = __hw_perf_counter_init(counter); - if (err) { - if (counter->destroy) - counter->destroy(counter); - return ERR_PTR(err); - } - - return &pmu; -} - -/* - * callchain support - */ - -static inline -void callchain_store(struct perf_callchain_entry *entry, u64 ip) -{ - if (entry->nr < PERF_MAX_STACK_DEPTH) - entry->ip[entry->nr++] = ip; -} - -static DEFINE_PER_CPU(struct perf_callchain_entry, pmc_irq_entry); -static DEFINE_PER_CPU(struct perf_callchain_entry, pmc_nmi_entry); -static DEFINE_PER_CPU(int, in_nmi_frame); - - -static void -backtrace_warning_symbol(void *data, char *msg, unsigned long symbol) -{ - /* Ignore warnings */ -} - -static void backtrace_warning(void *data, char *msg) -{ - /* Ignore warnings */ -} - -static int backtrace_stack(void *data, char *name) -{ - per_cpu(in_nmi_frame, smp_processor_id()) = - x86_is_stack_id(NMI_STACK, name); - - return 0; -} - -static void backtrace_address(void *data, unsigned long addr, int reliable) -{ - struct perf_callchain_entry *entry = data; - - if (per_cpu(in_nmi_frame, smp_processor_id())) - return; - - if (reliable) - callchain_store(entry, addr); -} - -static const struct stacktrace_ops backtrace_ops = { - .warning = backtrace_warning, - .warning_symbol = backtrace_warning_symbol, - .stack = backtrace_stack, - .address = backtrace_address, -}; - -#include "../dumpstack.h" - -static void -perf_callchain_kernel(struct pt_regs *regs, struct perf_callchain_entry *entry) -{ - callchain_store(entry, PERF_CONTEXT_KERNEL); - callchain_store(entry, regs->ip); - - dump_trace(NULL, regs, NULL, 0, &backtrace_ops, entry); -} - -/* - * best effort, GUP based copy_from_user() that assumes IRQ or NMI context - */ -static unsigned long -copy_from_user_nmi(void *to, const void __user *from, unsigned long n) -{ - unsigned long offset, addr = (unsigned long)from; - int type = in_nmi() ? KM_NMI : KM_IRQ0; - unsigned long size, len = 0; - struct page *page; - void *map; - int ret; - - do { - ret = __get_user_pages_fast(addr, 1, 0, &page); - if (!ret) - break; - - offset = addr & (PAGE_SIZE - 1); - size = min(PAGE_SIZE - offset, n - len); - - map = kmap_atomic(page, type); - memcpy(to, map+offset, size); - kunmap_atomic(map, type); - put_page(page); - - len += size; - to += size; - addr += size; - - } while (len < n); - - return len; -} - -static int copy_stack_frame(const void __user *fp, struct stack_frame *frame) -{ - unsigned long bytes; - - bytes = copy_from_user_nmi(frame, fp, sizeof(*frame)); - - return bytes == sizeof(*frame); -} - -static void -perf_callchain_user(struct pt_regs *regs, struct perf_callchain_entry *entry) -{ - struct stack_frame frame; - const void __user *fp; - - if (!user_mode(regs)) - regs = task_pt_regs(current); - - fp = (void __user *)regs->bp; - - callchain_store(entry, PERF_CONTEXT_USER); - callchain_store(entry, regs->ip); - - while (entry->nr < PERF_MAX_STACK_DEPTH) { - frame.next_frame = NULL; - frame.return_address = 0; - - if (!copy_stack_frame(fp, &frame)) - break; - - if ((unsigned long)fp < regs->sp) - break; - - callchain_store(entry, frame.return_address); - fp = frame.next_frame; - } -} - -static void -perf_do_callchain(struct pt_regs *regs, struct perf_callchain_entry *entry) -{ - int is_user; - - if (!regs) - return; - - is_user = user_mode(regs); - - if (!current || current->pid == 0) - return; - - if (is_user && current->state != TASK_RUNNING) - return; - - if (!is_user) - perf_callchain_kernel(regs, entry); - - if (current->mm) - perf_callchain_user(regs, entry); -} - -struct perf_callchain_entry *perf_callchain(struct pt_regs *regs) -{ - struct perf_callchain_entry *entry; - - if (in_nmi()) - entry = &__get_cpu_var(pmc_nmi_entry); - else - entry = &__get_cpu_var(pmc_irq_entry); - - entry->nr = 0; - - perf_do_callchain(regs, entry); - - return entry; -} - -void hw_perf_counter_setup_online(int cpu) -{ - init_debug_store_on_cpu(cpu); -} diff --git a/arch/x86/kernel/cpu/perf_event.c b/arch/x86/kernel/cpu/perf_event.c new file mode 100644 index 0000000..0d03629 --- /dev/null +++ b/arch/x86/kernel/cpu/perf_event.c @@ -0,0 +1,2298 @@ +/* + * Performance events x86 architecture code + * + * Copyright (C) 2008 Thomas Gleixner + * Copyright (C) 2008-2009 Red Hat, Inc., Ingo Molnar + * Copyright (C) 2009 Jaswinder Singh Rajput + * Copyright (C) 2009 Advanced Micro Devices, Inc., Robert Richter + * Copyright (C) 2008-2009 Red Hat, Inc., Peter Zijlstra + * Copyright (C) 2009 Intel Corporation, + * + * For licencing details see kernel-base/COPYING + */ + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include +#include +#include + +static u64 perf_event_mask __read_mostly; + +/* The maximal number of PEBS events: */ +#define MAX_PEBS_EVENTS 4 + +/* The size of a BTS record in bytes: */ +#define BTS_RECORD_SIZE 24 + +/* The size of a per-cpu BTS buffer in bytes: */ +#define BTS_BUFFER_SIZE (BTS_RECORD_SIZE * 2048) + +/* The BTS overflow threshold in bytes from the end of the buffer: */ +#define BTS_OVFL_TH (BTS_RECORD_SIZE * 128) + + +/* + * Bits in the debugctlmsr controlling branch tracing. + */ +#define X86_DEBUGCTL_TR (1 << 6) +#define X86_DEBUGCTL_BTS (1 << 7) +#define X86_DEBUGCTL_BTINT (1 << 8) +#define X86_DEBUGCTL_BTS_OFF_OS (1 << 9) +#define X86_DEBUGCTL_BTS_OFF_USR (1 << 10) + +/* + * A debug store configuration. + * + * We only support architectures that use 64bit fields. + */ +struct debug_store { + u64 bts_buffer_base; + u64 bts_index; + u64 bts_absolute_maximum; + u64 bts_interrupt_threshold; + u64 pebs_buffer_base; + u64 pebs_index; + u64 pebs_absolute_maximum; + u64 pebs_interrupt_threshold; + u64 pebs_event_reset[MAX_PEBS_EVENTS]; +}; + +struct cpu_hw_events { + struct perf_event *events[X86_PMC_IDX_MAX]; + unsigned long used_mask[BITS_TO_LONGS(X86_PMC_IDX_MAX)]; + unsigned long active_mask[BITS_TO_LONGS(X86_PMC_IDX_MAX)]; + unsigned long interrupts; + int enabled; + struct debug_store *ds; +}; + +/* + * struct x86_pmu - generic x86 pmu + */ +struct x86_pmu { + const char *name; + int version; + int (*handle_irq)(struct pt_regs *); + void (*disable_all)(void); + void (*enable_all)(void); + void (*enable)(struct hw_perf_event *, int); + void (*disable)(struct hw_perf_event *, int); + unsigned eventsel; + unsigned perfctr; + u64 (*event_map)(int); + u64 (*raw_event)(u64); + int max_events; + int num_events; + int num_events_fixed; + int event_bits; + u64 event_mask; + int apic; + u64 max_period; + u64 intel_ctrl; + void (*enable_bts)(u64 config); + void (*disable_bts)(void); +}; + +static struct x86_pmu x86_pmu __read_mostly; + +static DEFINE_PER_CPU(struct cpu_hw_events, cpu_hw_events) = { + .enabled = 1, +}; + +/* + * Not sure about some of these + */ +static const u64 p6_perfmon_event_map[] = +{ + [PERF_COUNT_HW_CPU_CYCLES] = 0x0079, + [PERF_COUNT_HW_INSTRUCTIONS] = 0x00c0, + [PERF_COUNT_HW_CACHE_REFERENCES] = 0x0f2e, + [PERF_COUNT_HW_CACHE_MISSES] = 0x012e, + [PERF_COUNT_HW_BRANCH_INSTRUCTIONS] = 0x00c4, + [PERF_COUNT_HW_BRANCH_MISSES] = 0x00c5, + [PERF_COUNT_HW_BUS_CYCLES] = 0x0062, +}; + +static u64 p6_pmu_event_map(int hw_event) +{ + return p6_perfmon_event_map[hw_event]; +} + +/* + * Event setting that is specified not to count anything. + * We use this to effectively disable a counter. + * + * L2_RQSTS with 0 MESI unit mask. + */ +#define P6_NOP_EVENT 0x0000002EULL + +static u64 p6_pmu_raw_event(u64 hw_event) +{ +#define P6_EVNTSEL_EVENT_MASK 0x000000FFULL +#define P6_EVNTSEL_UNIT_MASK 0x0000FF00ULL +#define P6_EVNTSEL_EDGE_MASK 0x00040000ULL +#define P6_EVNTSEL_INV_MASK 0x00800000ULL +#define P6_EVNTSEL_REG_MASK 0xFF000000ULL + +#define P6_EVNTSEL_MASK \ + (P6_EVNTSEL_EVENT_MASK | \ + P6_EVNTSEL_UNIT_MASK | \ + P6_EVNTSEL_EDGE_MASK | \ + P6_EVNTSEL_INV_MASK | \ + P6_EVNTSEL_REG_MASK) + + return hw_event & P6_EVNTSEL_MASK; +} + + +/* + * Intel PerfMon v3. Used on Core2 and later. + */ +static const u64 intel_perfmon_event_map[] = +{ + [PERF_COUNT_HW_CPU_CYCLES] = 0x003c, + [PERF_COUNT_HW_INSTRUCTIONS] = 0x00c0, + [PERF_COUNT_HW_CACHE_REFERENCES] = 0x4f2e, + [PERF_COUNT_HW_CACHE_MISSES] = 0x412e, + [PERF_COUNT_HW_BRANCH_INSTRUCTIONS] = 0x00c4, + [PERF_COUNT_HW_BRANCH_MISSES] = 0x00c5, + [PERF_COUNT_HW_BUS_CYCLES] = 0x013c, +}; + +static u64 intel_pmu_event_map(int hw_event) +{ + return intel_perfmon_event_map[hw_event]; +} + +/* + * Generalized hw caching related hw_event table, filled + * in on a per model basis. A value of 0 means + * 'not supported', -1 means 'hw_event makes no sense on + * this CPU', any other value means the raw hw_event + * ID. + */ + +#define C(x) PERF_COUNT_HW_CACHE_##x + +static u64 __read_mostly hw_cache_event_ids + [PERF_COUNT_HW_CACHE_MAX] + [PERF_COUNT_HW_CACHE_OP_MAX] + [PERF_COUNT_HW_CACHE_RESULT_MAX]; + +static const u64 nehalem_hw_cache_event_ids + [PERF_COUNT_HW_CACHE_MAX] + [PERF_COUNT_HW_CACHE_OP_MAX] + [PERF_COUNT_HW_CACHE_RESULT_MAX] = +{ + [ C(L1D) ] = { + [ C(OP_READ) ] = { + [ C(RESULT_ACCESS) ] = 0x0f40, /* L1D_CACHE_LD.MESI */ + [ C(RESULT_MISS) ] = 0x0140, /* L1D_CACHE_LD.I_STATE */ + }, + [ C(OP_WRITE) ] = { + [ C(RESULT_ACCESS) ] = 0x0f41, /* L1D_CACHE_ST.MESI */ + [ C(RESULT_MISS) ] = 0x0141, /* L1D_CACHE_ST.I_STATE */ + }, + [ C(OP_PREFETCH) ] = { + [ C(RESULT_ACCESS) ] = 0x014e, /* L1D_PREFETCH.REQUESTS */ + [ C(RESULT_MISS) ] = 0x024e, /* L1D_PREFETCH.MISS */ + }, + }, + [ C(L1I ) ] = { + [ C(OP_READ) ] = { + [ C(RESULT_ACCESS) ] = 0x0380, /* L1I.READS */ + [ C(RESULT_MISS) ] = 0x0280, /* L1I.MISSES */ + }, + [ C(OP_WRITE) ] = { + [ C(RESULT_ACCESS) ] = -1, + [ C(RESULT_MISS) ] = -1, + }, + [ C(OP_PREFETCH) ] = { + [ C(RESULT_ACCESS) ] = 0x0, + [ C(RESULT_MISS) ] = 0x0, + }, + }, + [ C(LL ) ] = { + [ C(OP_READ) ] = { + [ C(RESULT_ACCESS) ] = 0x0324, /* L2_RQSTS.LOADS */ + [ C(RESULT_MISS) ] = 0x0224, /* L2_RQSTS.LD_MISS */ + }, + [ C(OP_WRITE) ] = { + [ C(RESULT_ACCESS) ] = 0x0c24, /* L2_RQSTS.RFOS */ + [ C(RESULT_MISS) ] = 0x0824, /* L2_RQSTS.RFO_MISS */ + }, + [ C(OP_PREFETCH) ] = { + [ C(RESULT_ACCESS) ] = 0x4f2e, /* LLC Reference */ + [ C(RESULT_MISS) ] = 0x412e, /* LLC Misses */ + }, + }, + [ C(DTLB) ] = { + [ C(OP_READ) ] = { + [ C(RESULT_ACCESS) ] = 0x0f40, /* L1D_CACHE_LD.MESI (alias) */ + [ C(RESULT_MISS) ] = 0x0108, /* DTLB_LOAD_MISSES.ANY */ + }, + [ C(OP_WRITE) ] = { + [ C(RESULT_ACCESS) ] = 0x0f41, /* L1D_CACHE_ST.MESI (alias) */ + [ C(RESULT_MISS) ] = 0x010c, /* MEM_STORE_RETIRED.DTLB_MISS */ + }, + [ C(OP_PREFETCH) ] = { + [ C(RESULT_ACCESS) ] = 0x0, + [ C(RESULT_MISS) ] = 0x0, + }, + }, + [ C(ITLB) ] = { + [ C(OP_READ) ] = { + [ C(RESULT_ACCESS) ] = 0x01c0, /* INST_RETIRED.ANY_P */ + [ C(RESULT_MISS) ] = 0x20c8, /* ITLB_MISS_RETIRED */ + }, + [ C(OP_WRITE) ] = { + [ C(RESULT_ACCESS) ] = -1, + [ C(RESULT_MISS) ] = -1, + }, + [ C(OP_PREFETCH) ] = { + [ C(RESULT_ACCESS) ] = -1, + [ C(RESULT_MISS) ] = -1, + }, + }, + [ C(BPU ) ] = { + [ C(OP_READ) ] = { + [ C(RESULT_ACCESS) ] = 0x00c4, /* BR_INST_RETIRED.ALL_BRANCHES */ + [ C(RESULT_MISS) ] = 0x03e8, /* BPU_CLEARS.ANY */ + }, + [ C(OP_WRITE) ] = { + [ C(RESULT_ACCESS) ] = -1, + [ C(RESULT_MISS) ] = -1, + }, + [ C(OP_PREFETCH) ] = { + [ C(RESULT_ACCESS) ] = -1, + [ C(RESULT_MISS) ] = -1, + }, + }, +}; + +static const u64 core2_hw_cache_event_ids + [PERF_COUNT_HW_CACHE_MAX] + [PERF_COUNT_HW_CACHE_OP_MAX] + [PERF_COUNT_HW_CACHE_RESULT_MAX] = +{ + [ C(L1D) ] = { + [ C(OP_READ) ] = { + [ C(RESULT_ACCESS) ] = 0x0f40, /* L1D_CACHE_LD.MESI */ + [ C(RESULT_MISS) ] = 0x0140, /* L1D_CACHE_LD.I_STATE */ + }, + [ C(OP_WRITE) ] = { + [ C(RESULT_ACCESS) ] = 0x0f41, /* L1D_CACHE_ST.MESI */ + [ C(RESULT_MISS) ] = 0x0141, /* L1D_CACHE_ST.I_STATE */ + }, + [ C(OP_PREFETCH) ] = { + [ C(RESULT_ACCESS) ] = 0x104e, /* L1D_PREFETCH.REQUESTS */ + [ C(RESULT_MISS) ] = 0, + }, + }, + [ C(L1I ) ] = { + [ C(OP_READ) ] = { + [ C(RESULT_ACCESS) ] = 0x0080, /* L1I.READS */ + [ C(RESULT_MISS) ] = 0x0081, /* L1I.MISSES */ + }, + [ C(OP_WRITE) ] = { + [ C(RESULT_ACCESS) ] = -1, + [ C(RESULT_MISS) ] = -1, + }, + [ C(OP_PREFETCH) ] = { + [ C(RESULT_ACCESS) ] = 0, + [ C(RESULT_MISS) ] = 0, + }, + }, + [ C(LL ) ] = { + [ C(OP_READ) ] = { + [ C(RESULT_ACCESS) ] = 0x4f29, /* L2_LD.MESI */ + [ C(RESULT_MISS) ] = 0x4129, /* L2_LD.ISTATE */ + }, + [ C(OP_WRITE) ] = { + [ C(RESULT_ACCESS) ] = 0x4f2A, /* L2_ST.MESI */ + [ C(RESULT_MISS) ] = 0x412A, /* L2_ST.ISTATE */ + }, + [ C(OP_PREFETCH) ] = { + [ C(RESULT_ACCESS) ] = 0, + [ C(RESULT_MISS) ] = 0, + }, + }, + [ C(DTLB) ] = { + [ C(OP_READ) ] = { + [ C(RESULT_ACCESS) ] = 0x0f40, /* L1D_CACHE_LD.MESI (alias) */ + [ C(RESULT_MISS) ] = 0x0208, /* DTLB_MISSES.MISS_LD */ + }, + [ C(OP_WRITE) ] = { + [ C(RESULT_ACCESS) ] = 0x0f41, /* L1D_CACHE_ST.MESI (alias) */ + [ C(RESULT_MISS) ] = 0x0808, /* DTLB_MISSES.MISS_ST */ + }, + [ C(OP_PREFETCH) ] = { + [ C(RESULT_ACCESS) ] = 0, + [ C(RESULT_MISS) ] = 0, + }, + }, + [ C(ITLB) ] = { + [ C(OP_READ) ] = { + [ C(RESULT_ACCESS) ] = 0x00c0, /* INST_RETIRED.ANY_P */ + [ C(RESULT_MISS) ] = 0x1282, /* ITLBMISSES */ + }, + [ C(OP_WRITE) ] = { + [ C(RESULT_ACCESS) ] = -1, + [ C(RESULT_MISS) ] = -1, + }, + [ C(OP_PREFETCH) ] = { + [ C(RESULT_ACCESS) ] = -1, + [ C(RESULT_MISS) ] = -1, + }, + }, + [ C(BPU ) ] = { + [ C(OP_READ) ] = { + [ C(RESULT_ACCESS) ] = 0x00c4, /* BR_INST_RETIRED.ANY */ + [ C(RESULT_MISS) ] = 0x00c5, /* BP_INST_RETIRED.MISPRED */ + }, + [ C(OP_WRITE) ] = { + [ C(RESULT_ACCESS) ] = -1, + [ C(RESULT_MISS) ] = -1, + }, + [ C(OP_PREFETCH) ] = { + [ C(RESULT_ACCESS) ] = -1, + [ C(RESULT_MISS) ] = -1, + }, + }, +}; + +static const u64 atom_hw_cache_event_ids + [PERF_COUNT_HW_CACHE_MAX] + [PERF_COUNT_HW_CACHE_OP_MAX] + [PERF_COUNT_HW_CACHE_RESULT_MAX] = +{ + [ C(L1D) ] = { + [ C(OP_READ) ] = { + [ C(RESULT_ACCESS) ] = 0x2140, /* L1D_CACHE.LD */ + [ C(RESULT_MISS) ] = 0, + }, + [ C(OP_WRITE) ] = { + [ C(RESULT_ACCESS) ] = 0x2240, /* L1D_CACHE.ST */ + [ C(RESULT_MISS) ] = 0, + }, + [ C(OP_PREFETCH) ] = { + [ C(RESULT_ACCESS) ] = 0x0, + [ C(RESULT_MISS) ] = 0, + }, + }, + [ C(L1I ) ] = { + [ C(OP_READ) ] = { + [ C(RESULT_ACCESS) ] = 0x0380, /* L1I.READS */ + [ C(RESULT_MISS) ] = 0x0280, /* L1I.MISSES */ + }, + [ C(OP_WRITE) ] = { + [ C(RESULT_ACCESS) ] = -1, + [ C(RESULT_MISS) ] = -1, + }, + [ C(OP_PREFETCH) ] = { + [ C(RESULT_ACCESS) ] = 0, + [ C(RESULT_MISS) ] = 0, + }, + }, + [ C(LL ) ] = { + [ C(OP_READ) ] = { + [ C(RESULT_ACCESS) ] = 0x4f29, /* L2_LD.MESI */ + [ C(RESULT_MISS) ] = 0x4129, /* L2_LD.ISTATE */ + }, + [ C(OP_WRITE) ] = { + [ C(RESULT_ACCESS) ] = 0x4f2A, /* L2_ST.MESI */ + [ C(RESULT_MISS) ] = 0x412A, /* L2_ST.ISTATE */ + }, + [ C(OP_PREFETCH) ] = { + [ C(RESULT_ACCESS) ] = 0, + [ C(RESULT_MISS) ] = 0, + }, + }, + [ C(DTLB) ] = { + [ C(OP_READ) ] = { + [ C(RESULT_ACCESS) ] = 0x2140, /* L1D_CACHE_LD.MESI (alias) */ + [ C(RESULT_MISS) ] = 0x0508, /* DTLB_MISSES.MISS_LD */ + }, + [ C(OP_WRITE) ] = { + [ C(RESULT_ACCESS) ] = 0x2240, /* L1D_CACHE_ST.MESI (alias) */ + [ C(RESULT_MISS) ] = 0x0608, /* DTLB_MISSES.MISS_ST */ + }, + [ C(OP_PREFETCH) ] = { + [ C(RESULT_ACCESS) ] = 0, + [ C(RESULT_MISS) ] = 0, + }, + }, + [ C(ITLB) ] = { + [ C(OP_READ) ] = { + [ C(RESULT_ACCESS) ] = 0x00c0, /* INST_RETIRED.ANY_P */ + [ C(RESULT_MISS) ] = 0x0282, /* ITLB.MISSES */ + }, + [ C(OP_WRITE) ] = { + [ C(RESULT_ACCESS) ] = -1, + [ C(RESULT_MISS) ] = -1, + }, + [ C(OP_PREFETCH) ] = { + [ C(RESULT_ACCESS) ] = -1, + [ C(RESULT_MISS) ] = -1, + }, + }, + [ C(BPU ) ] = { + [ C(OP_READ) ] = { + [ C(RESULT_ACCESS) ] = 0x00c4, /* BR_INST_RETIRED.ANY */ + [ C(RESULT_MISS) ] = 0x00c5, /* BP_INST_RETIRED.MISPRED */ + }, + [ C(OP_WRITE) ] = { + [ C(RESULT_ACCESS) ] = -1, + [ C(RESULT_MISS) ] = -1, + }, + [ C(OP_PREFETCH) ] = { + [ C(RESULT_ACCESS) ] = -1, + [ C(RESULT_MISS) ] = -1, + }, + }, +}; + +static u64 intel_pmu_raw_event(u64 hw_event) +{ +#define CORE_EVNTSEL_EVENT_MASK 0x000000FFULL +#define CORE_EVNTSEL_UNIT_MASK 0x0000FF00ULL +#define CORE_EVNTSEL_EDGE_MASK 0x00040000ULL +#define CORE_EVNTSEL_INV_MASK 0x00800000ULL +#define CORE_EVNTSEL_REG_MASK 0xFF000000ULL + +#define CORE_EVNTSEL_MASK \ + (CORE_EVNTSEL_EVENT_MASK | \ + CORE_EVNTSEL_UNIT_MASK | \ + CORE_EVNTSEL_EDGE_MASK | \ + CORE_EVNTSEL_INV_MASK | \ + CORE_EVNTSEL_REG_MASK) + + return hw_event & CORE_EVNTSEL_MASK; +} + +static const u64 amd_hw_cache_event_ids + [PERF_COUNT_HW_CACHE_MAX] + [PERF_COUNT_HW_CACHE_OP_MAX] + [PERF_COUNT_HW_CACHE_RESULT_MAX] = +{ + [ C(L1D) ] = { + [ C(OP_READ) ] = { + [ C(RESULT_ACCESS) ] = 0x0040, /* Data Cache Accesses */ + [ C(RESULT_MISS) ] = 0x0041, /* Data Cache Misses */ + }, + [ C(OP_WRITE) ] = { + [ C(RESULT_ACCESS) ] = 0x0142, /* Data Cache Refills :system */ + [ C(RESULT_MISS) ] = 0, + }, + [ C(OP_PREFETCH) ] = { + [ C(RESULT_ACCESS) ] = 0x0267, /* Data Prefetcher :attempts */ + [ C(RESULT_MISS) ] = 0x0167, /* Data Prefetcher :cancelled */ + }, + }, + [ C(L1I ) ] = { + [ C(OP_READ) ] = { + [ C(RESULT_ACCESS) ] = 0x0080, /* Instruction cache fetches */ + [ C(RESULT_MISS) ] = 0x0081, /* Instruction cache misses */ + }, + [ C(OP_WRITE) ] = { + [ C(RESULT_ACCESS) ] = -1, + [ C(RESULT_MISS) ] = -1, + }, + [ C(OP_PREFETCH) ] = { + [ C(RESULT_ACCESS) ] = 0x014B, /* Prefetch Instructions :Load */ + [ C(RESULT_MISS) ] = 0, + }, + }, + [ C(LL ) ] = { + [ C(OP_READ) ] = { + [ C(RESULT_ACCESS) ] = 0x037D, /* Requests to L2 Cache :IC+DC */ + [ C(RESULT_MISS) ] = 0x037E, /* L2 Cache Misses : IC+DC */ + }, + [ C(OP_WRITE) ] = { + [ C(RESULT_ACCESS) ] = 0x017F, /* L2 Fill/Writeback */ + [ C(RESULT_MISS) ] = 0, + }, + [ C(OP_PREFETCH) ] = { + [ C(RESULT_ACCESS) ] = 0, + [ C(RESULT_MISS) ] = 0, + }, + }, + [ C(DTLB) ] = { + [ C(OP_READ) ] = { + [ C(RESULT_ACCESS) ] = 0x0040, /* Data Cache Accesses */ + [ C(RESULT_MISS) ] = 0x0046, /* L1 DTLB and L2 DLTB Miss */ + }, + [ C(OP_WRITE) ] = { + [ C(RESULT_ACCESS) ] = 0, + [ C(RESULT_MISS) ] = 0, + }, + [ C(OP_PREFETCH) ] = { + [ C(RESULT_ACCESS) ] = 0, + [ C(RESULT_MISS) ] = 0, + }, + }, + [ C(ITLB) ] = { + [ C(OP_READ) ] = { + [ C(RESULT_ACCESS) ] = 0x0080, /* Instruction fecthes */ + [ C(RESULT_MISS) ] = 0x0085, /* Instr. fetch ITLB misses */ + }, + [ C(OP_WRITE) ] = { + [ C(RESULT_ACCESS) ] = -1, + [ C(RESULT_MISS) ] = -1, + }, + [ C(OP_PREFETCH) ] = { + [ C(RESULT_ACCESS) ] = -1, + [ C(RESULT_MISS) ] = -1, + }, + }, + [ C(BPU ) ] = { + [ C(OP_READ) ] = { + [ C(RESULT_ACCESS) ] = 0x00c2, /* Retired Branch Instr. */ + [ C(RESULT_MISS) ] = 0x00c3, /* Retired Mispredicted BI */ + }, + [ C(OP_WRITE) ] = { + [ C(RESULT_ACCESS) ] = -1, + [ C(RESULT_MISS) ] = -1, + }, + [ C(OP_PREFETCH) ] = { + [ C(RESULT_ACCESS) ] = -1, + [ C(RESULT_MISS) ] = -1, + }, + }, +}; + +/* + * AMD Performance Monitor K7 and later. + */ +static const u64 amd_perfmon_event_map[] = +{ + [PERF_COUNT_HW_CPU_CYCLES] = 0x0076, + [PERF_COUNT_HW_INSTRUCTIONS] = 0x00c0, + [PERF_COUNT_HW_CACHE_REFERENCES] = 0x0080, + [PERF_COUNT_HW_CACHE_MISSES] = 0x0081, + [PERF_COUNT_HW_BRANCH_INSTRUCTIONS] = 0x00c4, + [PERF_COUNT_HW_BRANCH_MISSES] = 0x00c5, +}; + +static u64 amd_pmu_event_map(int hw_event) +{ + return amd_perfmon_event_map[hw_event]; +} + +static u64 amd_pmu_raw_event(u64 hw_event) +{ +#define K7_EVNTSEL_EVENT_MASK 0x7000000FFULL +#define K7_EVNTSEL_UNIT_MASK 0x00000FF00ULL +#define K7_EVNTSEL_EDGE_MASK 0x000040000ULL +#define K7_EVNTSEL_INV_MASK 0x000800000ULL +#define K7_EVNTSEL_REG_MASK 0x0FF000000ULL + +#define K7_EVNTSEL_MASK \ + (K7_EVNTSEL_EVENT_MASK | \ + K7_EVNTSEL_UNIT_MASK | \ + K7_EVNTSEL_EDGE_MASK | \ + K7_EVNTSEL_INV_MASK | \ + K7_EVNTSEL_REG_MASK) + + return hw_event & K7_EVNTSEL_MASK; +} + +/* + * Propagate event elapsed time into the generic event. + * Can only be executed on the CPU where the event is active. + * Returns the delta events processed. + */ +static u64 +x86_perf_event_update(struct perf_event *event, + struct hw_perf_event *hwc, int idx) +{ + int shift = 64 - x86_pmu.event_bits; + u64 prev_raw_count, new_raw_count; + s64 delta; + + if (idx == X86_PMC_IDX_FIXED_BTS) + return 0; + + /* + * Careful: an NMI might modify the previous event value. + * + * Our tactic to handle this is to first atomically read and + * exchange a new raw count - then add that new-prev delta + * count to the generic event atomically: + */ +again: + prev_raw_count = atomic64_read(&hwc->prev_count); + rdmsrl(hwc->event_base + idx, new_raw_count); + + if (atomic64_cmpxchg(&hwc->prev_count, prev_raw_count, + new_raw_count) != prev_raw_count) + goto again; + + /* + * Now we have the new raw value and have updated the prev + * timestamp already. We can now calculate the elapsed delta + * (event-)time and add that to the generic event. + * + * Careful, not all hw sign-extends above the physical width + * of the count. + */ + delta = (new_raw_count << shift) - (prev_raw_count << shift); + delta >>= shift; + + atomic64_add(delta, &event->count); + atomic64_sub(delta, &hwc->period_left); + + return new_raw_count; +} + +static atomic_t active_events; +static DEFINE_MUTEX(pmc_reserve_mutex); + +static bool reserve_pmc_hardware(void) +{ +#ifdef CONFIG_X86_LOCAL_APIC + int i; + + if (nmi_watchdog == NMI_LOCAL_APIC) + disable_lapic_nmi_watchdog(); + + for (i = 0; i < x86_pmu.num_events; i++) { + if (!reserve_perfctr_nmi(x86_pmu.perfctr + i)) + goto perfctr_fail; + } + + for (i = 0; i < x86_pmu.num_events; i++) { + if (!reserve_evntsel_nmi(x86_pmu.eventsel + i)) + goto eventsel_fail; + } +#endif + + return true; + +#ifdef CONFIG_X86_LOCAL_APIC +eventsel_fail: + for (i--; i >= 0; i--) + release_evntsel_nmi(x86_pmu.eventsel + i); + + i = x86_pmu.num_events; + +perfctr_fail: + for (i--; i >= 0; i--) + release_perfctr_nmi(x86_pmu.perfctr + i); + + if (nmi_watchdog == NMI_LOCAL_APIC) + enable_lapic_nmi_watchdog(); + + return false; +#endif +} + +static void release_pmc_hardware(void) +{ +#ifdef CONFIG_X86_LOCAL_APIC + int i; + + for (i = 0; i < x86_pmu.num_events; i++) { + release_perfctr_nmi(x86_pmu.perfctr + i); + release_evntsel_nmi(x86_pmu.eventsel + i); + } + + if (nmi_watchdog == NMI_LOCAL_APIC) + enable_lapic_nmi_watchdog(); +#endif +} + +static inline bool bts_available(void) +{ + return x86_pmu.enable_bts != NULL; +} + +static inline void init_debug_store_on_cpu(int cpu) +{ + struct debug_store *ds = per_cpu(cpu_hw_events, cpu).ds; + + if (!ds) + return; + + wrmsr_on_cpu(cpu, MSR_IA32_DS_AREA, + (u32)((u64)(unsigned long)ds), + (u32)((u64)(unsigned long)ds >> 32)); +} + +static inline void fini_debug_store_on_cpu(int cpu) +{ + if (!per_cpu(cpu_hw_events, cpu).ds) + return; + + wrmsr_on_cpu(cpu, MSR_IA32_DS_AREA, 0, 0); +} + +static void release_bts_hardware(void) +{ + int cpu; + + if (!bts_available()) + return; + + get_online_cpus(); + + for_each_online_cpu(cpu) + fini_debug_store_on_cpu(cpu); + + for_each_possible_cpu(cpu) { + struct debug_store *ds = per_cpu(cpu_hw_events, cpu).ds; + + if (!ds) + continue; + + per_cpu(cpu_hw_events, cpu).ds = NULL; + + kfree((void *)(unsigned long)ds->bts_buffer_base); + kfree(ds); + } + + put_online_cpus(); +} + +static int reserve_bts_hardware(void) +{ + int cpu, err = 0; + + if (!bts_available()) + return 0; + + get_online_cpus(); + + for_each_possible_cpu(cpu) { + struct debug_store *ds; + void *buffer; + + err = -ENOMEM; + buffer = kzalloc(BTS_BUFFER_SIZE, GFP_KERNEL); + if (unlikely(!buffer)) + break; + + ds = kzalloc(sizeof(*ds), GFP_KERNEL); + if (unlikely(!ds)) { + kfree(buffer); + break; + } + + ds->bts_buffer_base = (u64)(unsigned long)buffer; + ds->bts_index = ds->bts_buffer_base; + ds->bts_absolute_maximum = + ds->bts_buffer_base + BTS_BUFFER_SIZE; + ds->bts_interrupt_threshold = + ds->bts_absolute_maximum - BTS_OVFL_TH; + + per_cpu(cpu_hw_events, cpu).ds = ds; + err = 0; + } + + if (err) + release_bts_hardware(); + else { + for_each_online_cpu(cpu) + init_debug_store_on_cpu(cpu); + } + + put_online_cpus(); + + return err; +} + +static void hw_perf_event_destroy(struct perf_event *event) +{ + if (atomic_dec_and_mutex_lock(&active_events, &pmc_reserve_mutex)) { + release_pmc_hardware(); + release_bts_hardware(); + mutex_unlock(&pmc_reserve_mutex); + } +} + +static inline int x86_pmu_initialized(void) +{ + return x86_pmu.handle_irq != NULL; +} + +static inline int +set_ext_hw_attr(struct hw_perf_event *hwc, struct perf_event_attr *attr) +{ + unsigned int cache_type, cache_op, cache_result; + u64 config, val; + + config = attr->config; + + cache_type = (config >> 0) & 0xff; + if (cache_type >= PERF_COUNT_HW_CACHE_MAX) + return -EINVAL; + + cache_op = (config >> 8) & 0xff; + if (cache_op >= PERF_COUNT_HW_CACHE_OP_MAX) + return -EINVAL; + + cache_result = (config >> 16) & 0xff; + if (cache_result >= PERF_COUNT_HW_CACHE_RESULT_MAX) + return -EINVAL; + + val = hw_cache_event_ids[cache_type][cache_op][cache_result]; + + if (val == 0) + return -ENOENT; + + if (val == -1) + return -EINVAL; + + hwc->config |= val; + + return 0; +} + +static void intel_pmu_enable_bts(u64 config) +{ + unsigned long debugctlmsr; + + debugctlmsr = get_debugctlmsr(); + + debugctlmsr |= X86_DEBUGCTL_TR; + debugctlmsr |= X86_DEBUGCTL_BTS; + debugctlmsr |= X86_DEBUGCTL_BTINT; + + if (!(config & ARCH_PERFMON_EVENTSEL_OS)) + debugctlmsr |= X86_DEBUGCTL_BTS_OFF_OS; + + if (!(config & ARCH_PERFMON_EVENTSEL_USR)) + debugctlmsr |= X86_DEBUGCTL_BTS_OFF_USR; + + update_debugctlmsr(debugctlmsr); +} + +static void intel_pmu_disable_bts(void) +{ + struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events); + unsigned long debugctlmsr; + + if (!cpuc->ds) + return; + + debugctlmsr = get_debugctlmsr(); + + debugctlmsr &= + ~(X86_DEBUGCTL_TR | X86_DEBUGCTL_BTS | X86_DEBUGCTL_BTINT | + X86_DEBUGCTL_BTS_OFF_OS | X86_DEBUGCTL_BTS_OFF_USR); + + update_debugctlmsr(debugctlmsr); +} + +/* + * Setup the hardware configuration for a given attr_type + */ +static int __hw_perf_event_init(struct perf_event *event) +{ + struct perf_event_attr *attr = &event->attr; + struct hw_perf_event *hwc = &event->hw; + u64 config; + int err; + + if (!x86_pmu_initialized()) + return -ENODEV; + + err = 0; + if (!atomic_inc_not_zero(&active_events)) { + mutex_lock(&pmc_reserve_mutex); + if (atomic_read(&active_events) == 0) { + if (!reserve_pmc_hardware()) + err = -EBUSY; + else + err = reserve_bts_hardware(); + } + if (!err) + atomic_inc(&active_events); + mutex_unlock(&pmc_reserve_mutex); + } + if (err) + return err; + + event->destroy = hw_perf_event_destroy; + + /* + * Generate PMC IRQs: + * (keep 'enabled' bit clear for now) + */ + hwc->config = ARCH_PERFMON_EVENTSEL_INT; + + /* + * Count user and OS events unless requested not to. + */ + if (!attr->exclude_user) + hwc->config |= ARCH_PERFMON_EVENTSEL_USR; + if (!attr->exclude_kernel) + hwc->config |= ARCH_PERFMON_EVENTSEL_OS; + + if (!hwc->sample_period) { + hwc->sample_period = x86_pmu.max_period; + hwc->last_period = hwc->sample_period; + atomic64_set(&hwc->period_left, hwc->sample_period); + } else { + /* + * If we have a PMU initialized but no APIC + * interrupts, we cannot sample hardware + * events (user-space has to fall back and + * sample via a hrtimer based software event): + */ + if (!x86_pmu.apic) + return -EOPNOTSUPP; + } + + /* + * Raw hw_event type provide the config in the hw_event structure + */ + if (attr->type == PERF_TYPE_RAW) { + hwc->config |= x86_pmu.raw_event(attr->config); + return 0; + } + + if (attr->type == PERF_TYPE_HW_CACHE) + return set_ext_hw_attr(hwc, attr); + + if (attr->config >= x86_pmu.max_events) + return -EINVAL; + + /* + * The generic map: + */ + config = x86_pmu.event_map(attr->config); + + if (config == 0) + return -ENOENT; + + if (config == -1LL) + return -EINVAL; + + /* + * Branch tracing: + */ + if ((attr->config == PERF_COUNT_HW_BRANCH_INSTRUCTIONS) && + (hwc->sample_period == 1)) { + /* BTS is not supported by this architecture. */ + if (!bts_available()) + return -EOPNOTSUPP; + + /* BTS is currently only allowed for user-mode. */ + if (hwc->config & ARCH_PERFMON_EVENTSEL_OS) + return -EOPNOTSUPP; + } + + hwc->config |= config; + + return 0; +} + +static void p6_pmu_disable_all(void) +{ + struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events); + u64 val; + + if (!cpuc->enabled) + return; + + cpuc->enabled = 0; + barrier(); + + /* p6 only has one enable register */ + rdmsrl(MSR_P6_EVNTSEL0, val); + val &= ~ARCH_PERFMON_EVENTSEL0_ENABLE; + wrmsrl(MSR_P6_EVNTSEL0, val); +} + +static void intel_pmu_disable_all(void) +{ + struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events); + + if (!cpuc->enabled) + return; + + cpuc->enabled = 0; + barrier(); + + wrmsrl(MSR_CORE_PERF_GLOBAL_CTRL, 0); + + if (test_bit(X86_PMC_IDX_FIXED_BTS, cpuc->active_mask)) + intel_pmu_disable_bts(); +} + +static void amd_pmu_disable_all(void) +{ + struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events); + int idx; + + if (!cpuc->enabled) + return; + + cpuc->enabled = 0; + /* + * ensure we write the disable before we start disabling the + * events proper, so that amd_pmu_enable_event() does the + * right thing. + */ + barrier(); + + for (idx = 0; idx < x86_pmu.num_events; idx++) { + u64 val; + + if (!test_bit(idx, cpuc->active_mask)) + continue; + rdmsrl(MSR_K7_EVNTSEL0 + idx, val); + if (!(val & ARCH_PERFMON_EVENTSEL0_ENABLE)) + continue; + val &= ~ARCH_PERFMON_EVENTSEL0_ENABLE; + wrmsrl(MSR_K7_EVNTSEL0 + idx, val); + } +} + +void hw_perf_disable(void) +{ + if (!x86_pmu_initialized()) + return; + return x86_pmu.disable_all(); +} + +static void p6_pmu_enable_all(void) +{ + struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events); + unsigned long val; + + if (cpuc->enabled) + return; + + cpuc->enabled = 1; + barrier(); + + /* p6 only has one enable register */ + rdmsrl(MSR_P6_EVNTSEL0, val); + val |= ARCH_PERFMON_EVENTSEL0_ENABLE; + wrmsrl(MSR_P6_EVNTSEL0, val); +} + +static void intel_pmu_enable_all(void) +{ + struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events); + + if (cpuc->enabled) + return; + + cpuc->enabled = 1; + barrier(); + + wrmsrl(MSR_CORE_PERF_GLOBAL_CTRL, x86_pmu.intel_ctrl); + + if (test_bit(X86_PMC_IDX_FIXED_BTS, cpuc->active_mask)) { + struct perf_event *event = + cpuc->events[X86_PMC_IDX_FIXED_BTS]; + + if (WARN_ON_ONCE(!event)) + return; + + intel_pmu_enable_bts(event->hw.config); + } +} + +static void amd_pmu_enable_all(void) +{ + struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events); + int idx; + + if (cpuc->enabled) + return; + + cpuc->enabled = 1; + barrier(); + + for (idx = 0; idx < x86_pmu.num_events; idx++) { + struct perf_event *event = cpuc->events[idx]; + u64 val; + + if (!test_bit(idx, cpuc->active_mask)) + continue; + + val = event->hw.config; + val |= ARCH_PERFMON_EVENTSEL0_ENABLE; + wrmsrl(MSR_K7_EVNTSEL0 + idx, val); + } +} + +void hw_perf_enable(void) +{ + if (!x86_pmu_initialized()) + return; + x86_pmu.enable_all(); +} + +static inline u64 intel_pmu_get_status(void) +{ + u64 status; + + rdmsrl(MSR_CORE_PERF_GLOBAL_STATUS, status); + + return status; +} + +static inline void intel_pmu_ack_status(u64 ack) +{ + wrmsrl(MSR_CORE_PERF_GLOBAL_OVF_CTRL, ack); +} + +static inline void x86_pmu_enable_event(struct hw_perf_event *hwc, int idx) +{ + (void)checking_wrmsrl(hwc->config_base + idx, + hwc->config | ARCH_PERFMON_EVENTSEL0_ENABLE); +} + +static inline void x86_pmu_disable_event(struct hw_perf_event *hwc, int idx) +{ + (void)checking_wrmsrl(hwc->config_base + idx, hwc->config); +} + +static inline void +intel_pmu_disable_fixed(struct hw_perf_event *hwc, int __idx) +{ + int idx = __idx - X86_PMC_IDX_FIXED; + u64 ctrl_val, mask; + + mask = 0xfULL << (idx * 4); + + rdmsrl(hwc->config_base, ctrl_val); + ctrl_val &= ~mask; + (void)checking_wrmsrl(hwc->config_base, ctrl_val); +} + +static inline void +p6_pmu_disable_event(struct hw_perf_event *hwc, int idx) +{ + struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events); + u64 val = P6_NOP_EVENT; + + if (cpuc->enabled) + val |= ARCH_PERFMON_EVENTSEL0_ENABLE; + + (void)checking_wrmsrl(hwc->config_base + idx, val); +} + +static inline void +intel_pmu_disable_event(struct hw_perf_event *hwc, int idx) +{ + if (unlikely(idx == X86_PMC_IDX_FIXED_BTS)) { + intel_pmu_disable_bts(); + return; + } + + if (unlikely(hwc->config_base == MSR_ARCH_PERFMON_FIXED_CTR_CTRL)) { + intel_pmu_disable_fixed(hwc, idx); + return; + } + + x86_pmu_disable_event(hwc, idx); +} + +static inline void +amd_pmu_disable_event(struct hw_perf_event *hwc, int idx) +{ + x86_pmu_disable_event(hwc, idx); +} + +static DEFINE_PER_CPU(u64 [X86_PMC_IDX_MAX], pmc_prev_left); + +/* + * Set the next IRQ period, based on the hwc->period_left value. + * To be called with the event disabled in hw: + */ +static int +x86_perf_event_set_period(struct perf_event *event, + struct hw_perf_event *hwc, int idx) +{ + s64 left = atomic64_read(&hwc->period_left); + s64 period = hwc->sample_period; + int err, ret = 0; + + if (idx == X86_PMC_IDX_FIXED_BTS) + return 0; + + /* + * If we are way outside a reasoable range then just skip forward: + */ + if (unlikely(left <= -period)) { + left = period; + atomic64_set(&hwc->period_left, left); + hwc->last_period = period; + ret = 1; + } + + if (unlikely(left <= 0)) { + left += period; + atomic64_set(&hwc->period_left, left); + hwc->last_period = period; + ret = 1; + } + /* + * Quirk: certain CPUs dont like it if just 1 hw_event is left: + */ + if (unlikely(left < 2)) + left = 2; + + if (left > x86_pmu.max_period) + left = x86_pmu.max_period; + + per_cpu(pmc_prev_left[idx], smp_processor_id()) = left; + + /* + * The hw event starts counting from this event offset, + * mark it to be able to extra future deltas: + */ + atomic64_set(&hwc->prev_count, (u64)-left); + + err = checking_wrmsrl(hwc->event_base + idx, + (u64)(-left) & x86_pmu.event_mask); + + perf_event_update_userpage(event); + + return ret; +} + +static inline void +intel_pmu_enable_fixed(struct hw_perf_event *hwc, int __idx) +{ + int idx = __idx - X86_PMC_IDX_FIXED; + u64 ctrl_val, bits, mask; + int err; + + /* + * Enable IRQ generation (0x8), + * and enable ring-3 counting (0x2) and ring-0 counting (0x1) + * if requested: + */ + bits = 0x8ULL; + if (hwc->config & ARCH_PERFMON_EVENTSEL_USR) + bits |= 0x2; + if (hwc->config & ARCH_PERFMON_EVENTSEL_OS) + bits |= 0x1; + bits <<= (idx * 4); + mask = 0xfULL << (idx * 4); + + rdmsrl(hwc->config_base, ctrl_val); + ctrl_val &= ~mask; + ctrl_val |= bits; + err = checking_wrmsrl(hwc->config_base, ctrl_val); +} + +static void p6_pmu_enable_event(struct hw_perf_event *hwc, int idx) +{ + struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events); + u64 val; + + val = hwc->config; + if (cpuc->enabled) + val |= ARCH_PERFMON_EVENTSEL0_ENABLE; + + (void)checking_wrmsrl(hwc->config_base + idx, val); +} + + +static void intel_pmu_enable_event(struct hw_perf_event *hwc, int idx) +{ + if (unlikely(idx == X86_PMC_IDX_FIXED_BTS)) { + if (!__get_cpu_var(cpu_hw_events).enabled) + return; + + intel_pmu_enable_bts(hwc->config); + return; + } + + if (unlikely(hwc->config_base == MSR_ARCH_PERFMON_FIXED_CTR_CTRL)) { + intel_pmu_enable_fixed(hwc, idx); + return; + } + + x86_pmu_enable_event(hwc, idx); +} + +static void amd_pmu_enable_event(struct hw_perf_event *hwc, int idx) +{ + struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events); + + if (cpuc->enabled) + x86_pmu_enable_event(hwc, idx); +} + +static int +fixed_mode_idx(struct perf_event *event, struct hw_perf_event *hwc) +{ + unsigned int hw_event; + + hw_event = hwc->config & ARCH_PERFMON_EVENT_MASK; + + if (unlikely((hw_event == + x86_pmu.event_map(PERF_COUNT_HW_BRANCH_INSTRUCTIONS)) && + (hwc->sample_period == 1))) + return X86_PMC_IDX_FIXED_BTS; + + if (!x86_pmu.num_events_fixed) + return -1; + + if (unlikely(hw_event == x86_pmu.event_map(PERF_COUNT_HW_INSTRUCTIONS))) + return X86_PMC_IDX_FIXED_INSTRUCTIONS; + if (unlikely(hw_event == x86_pmu.event_map(PERF_COUNT_HW_CPU_CYCLES))) + return X86_PMC_IDX_FIXED_CPU_CYCLES; + if (unlikely(hw_event == x86_pmu.event_map(PERF_COUNT_HW_BUS_CYCLES))) + return X86_PMC_IDX_FIXED_BUS_CYCLES; + + return -1; +} + +/* + * Find a PMC slot for the freshly enabled / scheduled in event: + */ +static int x86_pmu_enable(struct perf_event *event) +{ + struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events); + struct hw_perf_event *hwc = &event->hw; + int idx; + + idx = fixed_mode_idx(event, hwc); + if (idx == X86_PMC_IDX_FIXED_BTS) { + /* BTS is already occupied. */ + if (test_and_set_bit(idx, cpuc->used_mask)) + return -EAGAIN; + + hwc->config_base = 0; + hwc->event_base = 0; + hwc->idx = idx; + } else if (idx >= 0) { + /* + * Try to get the fixed event, if that is already taken + * then try to get a generic event: + */ + if (test_and_set_bit(idx, cpuc->used_mask)) + goto try_generic; + + hwc->config_base = MSR_ARCH_PERFMON_FIXED_CTR_CTRL; + /* + * We set it so that event_base + idx in wrmsr/rdmsr maps to + * MSR_ARCH_PERFMON_FIXED_CTR0 ... CTR2: + */ + hwc->event_base = + MSR_ARCH_PERFMON_FIXED_CTR0 - X86_PMC_IDX_FIXED; + hwc->idx = idx; + } else { + idx = hwc->idx; + /* Try to get the previous generic event again */ + if (test_and_set_bit(idx, cpuc->used_mask)) { +try_generic: + idx = find_first_zero_bit(cpuc->used_mask, + x86_pmu.num_events); + if (idx == x86_pmu.num_events) + return -EAGAIN; + + set_bit(idx, cpuc->used_mask); + hwc->idx = idx; + } + hwc->config_base = x86_pmu.eventsel; + hwc->event_base = x86_pmu.perfctr; + } + + perf_events_lapic_init(); + + x86_pmu.disable(hwc, idx); + + cpuc->events[idx] = event; + set_bit(idx, cpuc->active_mask); + + x86_perf_event_set_period(event, hwc, idx); + x86_pmu.enable(hwc, idx); + + perf_event_update_userpage(event); + + return 0; +} + +static void x86_pmu_unthrottle(struct perf_event *event) +{ + struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events); + struct hw_perf_event *hwc = &event->hw; + + if (WARN_ON_ONCE(hwc->idx >= X86_PMC_IDX_MAX || + cpuc->events[hwc->idx] != event)) + return; + + x86_pmu.enable(hwc, hwc->idx); +} + +void perf_event_print_debug(void) +{ + u64 ctrl, status, overflow, pmc_ctrl, pmc_count, prev_left, fixed; + struct cpu_hw_events *cpuc; + unsigned long flags; + int cpu, idx; + + if (!x86_pmu.num_events) + return; + + local_irq_save(flags); + + cpu = smp_processor_id(); + cpuc = &per_cpu(cpu_hw_events, cpu); + + if (x86_pmu.version >= 2) { + rdmsrl(MSR_CORE_PERF_GLOBAL_CTRL, ctrl); + rdmsrl(MSR_CORE_PERF_GLOBAL_STATUS, status); + rdmsrl(MSR_CORE_PERF_GLOBAL_OVF_CTRL, overflow); + rdmsrl(MSR_ARCH_PERFMON_FIXED_CTR_CTRL, fixed); + + pr_info("\n"); + pr_info("CPU#%d: ctrl: %016llx\n", cpu, ctrl); + pr_info("CPU#%d: status: %016llx\n", cpu, status); + pr_info("CPU#%d: overflow: %016llx\n", cpu, overflow); + pr_info("CPU#%d: fixed: %016llx\n", cpu, fixed); + } + pr_info("CPU#%d: used: %016llx\n", cpu, *(u64 *)cpuc->used_mask); + + for (idx = 0; idx < x86_pmu.num_events; idx++) { + rdmsrl(x86_pmu.eventsel + idx, pmc_ctrl); + rdmsrl(x86_pmu.perfctr + idx, pmc_count); + + prev_left = per_cpu(pmc_prev_left[idx], cpu); + + pr_info("CPU#%d: gen-PMC%d ctrl: %016llx\n", + cpu, idx, pmc_ctrl); + pr_info("CPU#%d: gen-PMC%d count: %016llx\n", + cpu, idx, pmc_count); + pr_info("CPU#%d: gen-PMC%d left: %016llx\n", + cpu, idx, prev_left); + } + for (idx = 0; idx < x86_pmu.num_events_fixed; idx++) { + rdmsrl(MSR_ARCH_PERFMON_FIXED_CTR0 + idx, pmc_count); + + pr_info("CPU#%d: fixed-PMC%d count: %016llx\n", + cpu, idx, pmc_count); + } + local_irq_restore(flags); +} + +static void intel_pmu_drain_bts_buffer(struct cpu_hw_events *cpuc) +{ + struct debug_store *ds = cpuc->ds; + struct bts_record { + u64 from; + u64 to; + u64 flags; + }; + struct perf_event *event = cpuc->events[X86_PMC_IDX_FIXED_BTS]; + struct bts_record *at, *top; + struct perf_output_handle handle; + struct perf_event_header header; + struct perf_sample_data data; + struct pt_regs regs; + + if (!event) + return; + + if (!ds) + return; + + at = (struct bts_record *)(unsigned long)ds->bts_buffer_base; + top = (struct bts_record *)(unsigned long)ds->bts_index; + + if (top <= at) + return; + + ds->bts_index = ds->bts_buffer_base; + + + data.period = event->hw.last_period; + data.addr = 0; + regs.ip = 0; + + /* + * Prepare a generic sample, i.e. fill in the invariant fields. + * We will overwrite the from and to address before we output + * the sample. + */ + perf_prepare_sample(&header, &data, event, ®s); + + if (perf_output_begin(&handle, event, + header.size * (top - at), 1, 1)) + return; + + for (; at < top; at++) { + data.ip = at->from; + data.addr = at->to; + + perf_output_sample(&handle, &header, &data, event); + } + + perf_output_end(&handle); + + /* There's new data available. */ + event->hw.interrupts++; + event->pending_kill = POLL_IN; +} + +static void x86_pmu_disable(struct perf_event *event) +{ + struct cpu_hw_events *cpuc = &__get_cpu_var(cpu_hw_events); + struct hw_perf_event *hwc = &event->hw; + int idx = hwc->idx; + + /* + * Must be done before we disable, otherwise the nmi handler + * could reenable again: + */ + clear_bit(idx, cpuc->active_mask); + x86_pmu.disable(hwc, idx); + + /* + * Make sure the cleared pointer becomes visible before we + * (potentially) free the event: + */ + barrier(); + + /* + * Drain the remaining delta count out of a event + * that we are disabling: + */ + x86_perf_event_update(event, hwc, idx); + + /* Drain the remaining BTS records. */ + if (unlikely(idx == X86_PMC_IDX_FIXED_BTS)) + intel_pmu_drain_bts_buffer(cpuc); + + cpuc->events[idx] = NULL; + clear_bit(idx, cpuc->used_mask); + + perf_event_update_userpage(event); +} + +/* + * Save and restart an expired event. Called by NMI contexts, + * so it has to be careful about preempting normal event ops: + */ +static int intel_pmu_save_and_restart(struct perf_event *event) +{ + struct hw_perf_event *hwc = &event->hw; + int idx = hwc->idx; + int ret; + + x86_perf_event_update(event, hwc, idx); + ret = x86_perf_event_set_period(event, hwc, idx); + + if (event->state == PERF_EVENT_STATE_ACTIVE) + intel_pmu_enable_event(hwc, idx); + + return ret; +} + +static void intel_pmu_reset(void) +{ + struct debug_store *ds = __get_cpu_var(cpu_hw_events).ds; + unsigned long flags; + int idx; + + if (!x86_pmu.num_events) + return; + + local_irq_save(flags); + + printk("clearing PMU state on CPU#%d\n", smp_processor_id()); + + for (idx = 0; idx < x86_pmu.num_events; idx++) { + checking_wrmsrl(x86_pmu.eventsel + idx, 0ull); + checking_wrmsrl(x86_pmu.perfctr + idx, 0ull); + } + for (idx = 0; idx < x86_pmu.num_events_fixed; idx++) { + checking_wrmsrl(MSR_ARCH_PERFMON_FIXED_CTR0 + idx, 0ull); + } + if (ds) + ds->bts_index = ds->bts_buffer_base; + + local_irq_restore(flags); +} + +static int p6_pmu_handle_irq(struct pt_regs *regs) +{ + struct perf_sample_data data; + struct cpu_hw_events *cpuc; + struct perf_event *event; + struct hw_perf_event *hwc; + int idx, handled = 0; + u64 val; + + data.addr = 0; + + cpuc = &__get_cpu_var(cpu_hw_events); + + for (idx = 0; idx < x86_pmu.num_events; idx++) { + if (!test_bit(idx, cpuc->active_mask)) + continue; + + event = cpuc->events[idx]; + hwc = &event->hw; + + val = x86_perf_event_update(event, hwc, idx); + if (val & (1ULL << (x86_pmu.event_bits - 1))) + continue; + + /* + * event overflow + */ + handled = 1; + data.period = event->hw.last_period; + + if (!x86_perf_event_set_period(event, hwc, idx)) + continue; + + if (perf_event_overflow(event, 1, &data, regs)) + p6_pmu_disable_event(hwc, idx); + } + + if (handled) + inc_irq_stat(apic_perf_irqs); + + return handled; +} + +/* + * This handler is triggered by the local APIC, so the APIC IRQ handling + * rules apply: + */ +static int intel_pmu_handle_irq(struct pt_regs *regs) +{ + struct perf_sample_data data; + struct cpu_hw_events *cpuc; + int bit, loops; + u64 ack, status; + + data.addr = 0; + + cpuc = &__get_cpu_var(cpu_hw_events); + + perf_disable(); + intel_pmu_drain_bts_buffer(cpuc); + status = intel_pmu_get_status(); + if (!status) { + perf_enable(); + return 0; + } + + loops = 0; +again: + if (++loops > 100) { + WARN_ONCE(1, "perfevents: irq loop stuck!\n"); + perf_event_print_debug(); + intel_pmu_reset(); + perf_enable(); + return 1; + } + + inc_irq_stat(apic_perf_irqs); + ack = status; + for_each_bit(bit, (unsigned long *)&status, X86_PMC_IDX_MAX) { + struct perf_event *event = cpuc->events[bit]; + + clear_bit(bit, (unsigned long *) &status); + if (!test_bit(bit, cpuc->active_mask)) + continue; + + if (!intel_pmu_save_and_restart(event)) + continue; + + data.period = event->hw.last_period; + + if (perf_event_overflow(event, 1, &data, regs)) + intel_pmu_disable_event(&event->hw, bit); + } + + intel_pmu_ack_status(ack); + + /* + * Repeat if there is more work to be done: + */ + status = intel_pmu_get_status(); + if (status) + goto again; + + perf_enable(); + + return 1; +} + +static int amd_pmu_handle_irq(struct pt_regs *regs) +{ + struct perf_sample_data data; + struct cpu_hw_events *cpuc; + struct perf_event *event; + struct hw_perf_event *hwc; + int idx, handled = 0; + u64 val; + + data.addr = 0; + + cpuc = &__get_cpu_var(cpu_hw_events); + + for (idx = 0; idx < x86_pmu.num_events; idx++) { + if (!test_bit(idx, cpuc->active_mask)) + continue; + + event = cpuc->events[idx]; + hwc = &event->hw; + + val = x86_perf_event_update(event, hwc, idx); + if (val & (1ULL << (x86_pmu.event_bits - 1))) + continue; + + /* + * event overflow + */ + handled = 1; + data.period = event->hw.last_period; + + if (!x86_perf_event_set_period(event, hwc, idx)) + continue; + + if (perf_event_overflow(event, 1, &data, regs)) + amd_pmu_disable_event(hwc, idx); + } + + if (handled) + inc_irq_stat(apic_perf_irqs); + + return handled; +} + +void smp_perf_pending_interrupt(struct pt_regs *regs) +{ + irq_enter(); + ack_APIC_irq(); + inc_irq_stat(apic_pending_irqs); + perf_event_do_pending(); + irq_exit(); +} + +void set_perf_event_pending(void) +{ +#ifdef CONFIG_X86_LOCAL_APIC + apic->send_IPI_self(LOCAL_PENDING_VECTOR); +#endif +} + +void perf_events_lapic_init(void) +{ +#ifdef CONFIG_X86_LOCAL_APIC + if (!x86_pmu.apic || !x86_pmu_initialized()) + return; + + /* + * Always use NMI for PMU + */ + apic_write(APIC_LVTPC, APIC_DM_NMI); +#endif +} + +static int __kprobes +perf_event_nmi_handler(struct notifier_block *self, + unsigned long cmd, void *__args) +{ + struct die_args *args = __args; + struct pt_regs *regs; + + if (!atomic_read(&active_events)) + return NOTIFY_DONE; + + switch (cmd) { + case DIE_NMI: + case DIE_NMI_IPI: + break; + + default: + return NOTIFY_DONE; + } + + regs = args->regs; + +#ifdef CONFIG_X86_LOCAL_APIC + apic_write(APIC_LVTPC, APIC_DM_NMI); +#endif + /* + * Can't rely on the handled return value to say it was our NMI, two + * events could trigger 'simultaneously' raising two back-to-back NMIs. + * + * If the first NMI handles both, the latter will be empty and daze + * the CPU. + */ + x86_pmu.handle_irq(regs); + + return NOTIFY_STOP; +} + +static __read_mostly struct notifier_block perf_event_nmi_notifier = { + .notifier_call = perf_event_nmi_handler, + .next = NULL, + .priority = 1 +}; + +static struct x86_pmu p6_pmu = { + .name = "p6", + .handle_irq = p6_pmu_handle_irq, + .disable_all = p6_pmu_disable_all, + .enable_all = p6_pmu_enable_all, + .enable = p6_pmu_enable_event, + .disable = p6_pmu_disable_event, + .eventsel = MSR_P6_EVNTSEL0, + .perfctr = MSR_P6_PERFCTR0, + .event_map = p6_pmu_event_map, + .raw_event = p6_pmu_raw_event, + .max_events = ARRAY_SIZE(p6_perfmon_event_map), + .apic = 1, + .max_period = (1ULL << 31) - 1, + .version = 0, + .num_events = 2, + /* + * Events have 40 bits implemented. However they are designed such + * that bits [32-39] are sign extensions of bit 31. As such the + * effective width of a event for P6-like PMU is 32 bits only. + * + * See IA-32 Intel Architecture Software developer manual Vol 3B + */ + .event_bits = 32, + .event_mask = (1ULL << 32) - 1, +}; + +static struct x86_pmu intel_pmu = { + .name = "Intel", + .handle_irq = intel_pmu_handle_irq, + .disable_all = intel_pmu_disable_all, + .enable_all = intel_pmu_enable_all, + .enable = intel_pmu_enable_event, + .disable = intel_pmu_disable_event, + .eventsel = MSR_ARCH_PERFMON_EVENTSEL0, + .perfctr = MSR_ARCH_PERFMON_PERFCTR0, + .event_map = intel_pmu_event_map, + .raw_event = intel_pmu_raw_event, + .max_events = ARRAY_SIZE(intel_perfmon_event_map), + .apic = 1, + /* + * Intel PMCs cannot be accessed sanely above 32 bit width, + * so we install an artificial 1<<31 period regardless of + * the generic event period: + */ + .max_period = (1ULL << 31) - 1, + .enable_bts = intel_pmu_enable_bts, + .disable_bts = intel_pmu_disable_bts, +}; + +static struct x86_pmu amd_pmu = { + .name = "AMD", + .handle_irq = amd_pmu_handle_irq, + .disable_all = amd_pmu_disable_all, + .enable_all = amd_pmu_enable_all, + .enable = amd_pmu_enable_event, + .disable = amd_pmu_disable_event, + .eventsel = MSR_K7_EVNTSEL0, + .perfctr = MSR_K7_PERFCTR0, + .event_map = amd_pmu_event_map, + .raw_event = amd_pmu_raw_event, + .max_events = ARRAY_SIZE(amd_perfmon_event_map), + .num_events = 4, + .event_bits = 48, + .event_mask = (1ULL << 48) - 1, + .apic = 1, + /* use highest bit to detect overflow */ + .max_period = (1ULL << 47) - 1, +}; + +static int p6_pmu_init(void) +{ + switch (boot_cpu_data.x86_model) { + case 1: + case 3: /* Pentium Pro */ + case 5: + case 6: /* Pentium II */ + case 7: + case 8: + case 11: /* Pentium III */ + break; + case 9: + case 13: + /* Pentium M */ + break; + default: + pr_cont("unsupported p6 CPU model %d ", + boot_cpu_data.x86_model); + return -ENODEV; + } + + x86_pmu = p6_pmu; + + if (!cpu_has_apic) { + pr_info("no APIC, boot with the \"lapic\" boot parameter to force-enable it.\n"); + pr_info("no hardware sampling interrupt available.\n"); + x86_pmu.apic = 0; + } + + return 0; +} + +static int intel_pmu_init(void) +{ + union cpuid10_edx edx; + union cpuid10_eax eax; + unsigned int unused; + unsigned int ebx; + int version; + + if (!cpu_has(&boot_cpu_data, X86_FEATURE_ARCH_PERFMON)) { + /* check for P6 processor family */ + if (boot_cpu_data.x86 == 6) { + return p6_pmu_init(); + } else { + return -ENODEV; + } + } + + /* + * Check whether the Architectural PerfMon supports + * Branch Misses Retired hw_event or not. + */ + cpuid(10, &eax.full, &ebx, &unused, &edx.full); + if (eax.split.mask_length <= ARCH_PERFMON_BRANCH_MISSES_RETIRED) + return -ENODEV; + + version = eax.split.version_id; + if (version < 2) + return -ENODEV; + + x86_pmu = intel_pmu; + x86_pmu.version = version; + x86_pmu.num_events = eax.split.num_events; + x86_pmu.event_bits = eax.split.bit_width; + x86_pmu.event_mask = (1ULL << eax.split.bit_width) - 1; + + /* + * Quirk: v2 perfmon does not report fixed-purpose events, so + * assume at least 3 events: + */ + x86_pmu.num_events_fixed = max((int)edx.split.num_events_fixed, 3); + + /* + * Install the hw-cache-events table: + */ + switch (boot_cpu_data.x86_model) { + case 15: /* original 65 nm celeron/pentium/core2/xeon, "Merom"/"Conroe" */ + case 22: /* single-core 65 nm celeron/core2solo "Merom-L"/"Conroe-L" */ + case 23: /* current 45 nm celeron/core2/xeon "Penryn"/"Wolfdale" */ + case 29: /* six-core 45 nm xeon "Dunnington" */ + memcpy(hw_cache_event_ids, core2_hw_cache_event_ids, + sizeof(hw_cache_event_ids)); + + pr_cont("Core2 events, "); + break; + default: + case 26: + memcpy(hw_cache_event_ids, nehalem_hw_cache_event_ids, + sizeof(hw_cache_event_ids)); + + pr_cont("Nehalem/Corei7 events, "); + break; + case 28: + memcpy(hw_cache_event_ids, atom_hw_cache_event_ids, + sizeof(hw_cache_event_ids)); + + pr_cont("Atom events, "); + break; + } + return 0; +} + +static int amd_pmu_init(void) +{ + /* Performance-monitoring supported from K7 and later: */ + if (boot_cpu_data.x86 < 6) + return -ENODEV; + + x86_pmu = amd_pmu; + + /* Events are common for all AMDs */ + memcpy(hw_cache_event_ids, amd_hw_cache_event_ids, + sizeof(hw_cache_event_ids)); + + return 0; +} + +void __init init_hw_perf_events(void) +{ + int err; + + pr_info("Performance Events: "); + + switch (boot_cpu_data.x86_vendor) { + case X86_VENDOR_INTEL: + err = intel_pmu_init(); + break; + case X86_VENDOR_AMD: + err = amd_pmu_init(); + break; + default: + return; + } + if (err != 0) { + pr_cont("no PMU driver, software events only.\n"); + return; + } + + pr_cont("%s PMU driver.\n", x86_pmu.name); + + if (x86_pmu.num_events > X86_PMC_MAX_GENERIC) { + WARN(1, KERN_ERR "hw perf events %d > max(%d), clipping!", + x86_pmu.num_events, X86_PMC_MAX_GENERIC); + x86_pmu.num_events = X86_PMC_MAX_GENERIC; + } + perf_event_mask = (1 << x86_pmu.num_events) - 1; + perf_max_events = x86_pmu.num_events; + + if (x86_pmu.num_events_fixed > X86_PMC_MAX_FIXED) { + WARN(1, KERN_ERR "hw perf events fixed %d > max(%d), clipping!", + x86_pmu.num_events_fixed, X86_PMC_MAX_FIXED); + x86_pmu.num_events_fixed = X86_PMC_MAX_FIXED; + } + + perf_event_mask |= + ((1LL << x86_pmu.num_events_fixed)-1) << X86_PMC_IDX_FIXED; + x86_pmu.intel_ctrl = perf_event_mask; + + perf_events_lapic_init(); + register_die_notifier(&perf_event_nmi_notifier); + + pr_info("... version: %d\n", x86_pmu.version); + pr_info("... bit width: %d\n", x86_pmu.event_bits); + pr_info("... generic events: %d\n", x86_pmu.num_events); + pr_info("... value mask: %016Lx\n", x86_pmu.event_mask); + pr_info("... max period: %016Lx\n", x86_pmu.max_period); + pr_info("... fixed-purpose events: %d\n", x86_pmu.num_events_fixed); + pr_info("... event mask: %016Lx\n", perf_event_mask); +} + +static inline void x86_pmu_read(struct perf_event *event) +{ + x86_perf_event_update(event, &event->hw, event->hw.idx); +} + +static const struct pmu pmu = { + .enable = x86_pmu_enable, + .disable = x86_pmu_disable, + .read = x86_pmu_read, + .unthrottle = x86_pmu_unthrottle, +}; + +const struct pmu *hw_perf_event_init(struct perf_event *event) +{ + int err; + + err = __hw_perf_event_init(event); + if (err) { + if (event->destroy) + event->destroy(event); + return ERR_PTR(err); + } + + return &pmu; +} + +/* + * callchain support + */ + +static inline +void callchain_store(struct perf_callchain_entry *entry, u64 ip) +{ + if (entry->nr < PERF_MAX_STACK_DEPTH) + entry->ip[entry->nr++] = ip; +} + +static DEFINE_PER_CPU(struct perf_callchain_entry, pmc_irq_entry); +static DEFINE_PER_CPU(struct perf_callchain_entry, pmc_nmi_entry); +static DEFINE_PER_CPU(int, in_nmi_frame); + + +static void +backtrace_warning_symbol(void *data, char *msg, unsigned long symbol) +{ + /* Ignore warnings */ +} + +static void backtrace_warning(void *data, char *msg) +{ + /* Ignore warnings */ +} + +static int backtrace_stack(void *data, char *name) +{ + per_cpu(in_nmi_frame, smp_processor_id()) = + x86_is_stack_id(NMI_STACK, name); + + return 0; +} + +static void backtrace_address(void *data, unsigned long addr, int reliable) +{ + struct perf_callchain_entry *entry = data; + + if (per_cpu(in_nmi_frame, smp_processor_id())) + return; + + if (reliable) + callchain_store(entry, addr); +} + +static const struct stacktrace_ops backtrace_ops = { + .warning = backtrace_warning, + .warning_symbol = backtrace_warning_symbol, + .stack = backtrace_stack, + .address = backtrace_address, +}; + +#include "../dumpstack.h" + +static void +perf_callchain_kernel(struct pt_regs *regs, struct perf_callchain_entry *entry) +{ + callchain_store(entry, PERF_CONTEXT_KERNEL); + callchain_store(entry, regs->ip); + + dump_trace(NULL, regs, NULL, 0, &backtrace_ops, entry); +} + +/* + * best effort, GUP based copy_from_user() that assumes IRQ or NMI context + */ +static unsigned long +copy_from_user_nmi(void *to, const void __user *from, unsigned long n) +{ + unsigned long offset, addr = (unsigned long)from; + int type = in_nmi() ? KM_NMI : KM_IRQ0; + unsigned long size, len = 0; + struct page *page; + void *map; + int ret; + + do { + ret = __get_user_pages_fast(addr, 1, 0, &page); + if (!ret) + break; + + offset = addr & (PAGE_SIZE - 1); + size = min(PAGE_SIZE - offset, n - len); + + map = kmap_atomic(page, type); + memcpy(to, map+offset, size); + kunmap_atomic(map, type); + put_page(page); + + len += size; + to += size; + addr += size; + + } while (len < n); + + return len; +} + +static int copy_stack_frame(const void __user *fp, struct stack_frame *frame) +{ + unsigned long bytes; + + bytes = copy_from_user_nmi(frame, fp, sizeof(*frame)); + + return bytes == sizeof(*frame); +} + +static void +perf_callchain_user(struct pt_regs *regs, struct perf_callchain_entry *entry) +{ + struct stack_frame frame; + const void __user *fp; + + if (!user_mode(regs)) + regs = task_pt_regs(current); + + fp = (void __user *)regs->bp; + + callchain_store(entry, PERF_CONTEXT_USER); + callchain_store(entry, regs->ip); + + while (entry->nr < PERF_MAX_STACK_DEPTH) { + frame.next_frame = NULL; + frame.return_address = 0; + + if (!copy_stack_frame(fp, &frame)) + break; + + if ((unsigned long)fp < regs->sp) + break; + + callchain_store(entry, frame.return_address); + fp = frame.next_frame; + } +} + +static void +perf_do_callchain(struct pt_regs *regs, struct perf_callchain_entry *entry) +{ + int is_user; + + if (!regs) + return; + + is_user = user_mode(regs); + + if (!current || current->pid == 0) + return; + + if (is_user && current->state != TASK_RUNNING) + return; + + if (!is_user) + perf_callchain_kernel(regs, entry); + + if (current->mm) + perf_callchain_user(regs, entry); +} + +struct perf_callchain_entry *perf_callchain(struct pt_regs *regs) +{ + struct perf_callchain_entry *entry; + + if (in_nmi()) + entry = &__get_cpu_var(pmc_nmi_entry); + else + entry = &__get_cpu_var(pmc_irq_entry); + + entry->nr = 0; + + perf_do_callchain(regs, entry); + + return entry; +} + +void hw_perf_event_setup_online(int cpu) +{ + init_debug_store_on_cpu(cpu); +} diff --git a/arch/x86/kernel/cpu/perfctr-watchdog.c b/arch/x86/kernel/cpu/perfctr-watchdog.c index 392bea4..fab786f 100644 --- a/arch/x86/kernel/cpu/perfctr-watchdog.c +++ b/arch/x86/kernel/cpu/perfctr-watchdog.c @@ -20,7 +20,7 @@ #include #include -#include +#include struct nmi_watchdog_ctlblk { unsigned int cccr_msr; diff --git a/arch/x86/kernel/entry_64.S b/arch/x86/kernel/entry_64.S index d59fe32..681c3fd 100644 --- a/arch/x86/kernel/entry_64.S +++ b/arch/x86/kernel/entry_64.S @@ -1021,7 +1021,7 @@ apicinterrupt ERROR_APIC_VECTOR \ apicinterrupt SPURIOUS_APIC_VECTOR \ spurious_interrupt smp_spurious_interrupt -#ifdef CONFIG_PERF_COUNTERS +#ifdef CONFIG_PERF_EVENTS apicinterrupt LOCAL_PENDING_VECTOR \ perf_pending_interrupt smp_perf_pending_interrupt #endif diff --git a/arch/x86/kernel/irqinit.c b/arch/x86/kernel/irqinit.c index 3008831..40f3077 100644 --- a/arch/x86/kernel/irqinit.c +++ b/arch/x86/kernel/irqinit.c @@ -208,7 +208,7 @@ static void __init apic_intr_init(void) alloc_intr_gate(ERROR_APIC_VECTOR, error_interrupt); /* Performance monitoring interrupts: */ -# ifdef CONFIG_PERF_COUNTERS +# ifdef CONFIG_PERF_EVENTS alloc_intr_gate(LOCAL_PENDING_VECTOR, perf_pending_interrupt); # endif diff --git a/arch/x86/kernel/syscall_table_32.S b/arch/x86/kernel/syscall_table_32.S index d51321d..0157cd2 100644 --- a/arch/x86/kernel/syscall_table_32.S +++ b/arch/x86/kernel/syscall_table_32.S @@ -335,4 +335,4 @@ ENTRY(sys_call_table) .long sys_preadv .long sys_pwritev .long sys_rt_tgsigqueueinfo /* 335 */ - .long sys_perf_counter_open + .long sys_perf_event_open diff --git a/arch/x86/mm/fault.c b/arch/x86/mm/fault.c index 775a020..82728f2 100644 --- a/arch/x86/mm/fault.c +++ b/arch/x86/mm/fault.c @@ -10,7 +10,7 @@ #include /* max_low_pfn */ #include /* __kprobes, ... */ #include /* kmmio_handler, ... */ -#include /* perf_swcounter_event */ +#include /* perf_sw_event */ #include /* dotraplinkage, ... */ #include /* pgd_*(), ... */ @@ -1017,7 +1017,7 @@ do_page_fault(struct pt_regs *regs, unsigned long error_code) if (unlikely(error_code & PF_RSVD)) pgtable_bad(regs, error_code, address); - perf_swcounter_event(PERF_COUNT_SW_PAGE_FAULTS, 1, 0, regs, address); + perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, 0, regs, address); /* * If we're in an interrupt, have no user context or are running @@ -1114,11 +1114,11 @@ good_area: if (fault & VM_FAULT_MAJOR) { tsk->maj_flt++; - perf_swcounter_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, 0, + perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, 0, regs, address); } else { tsk->min_flt++; - perf_swcounter_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, 0, + perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, 0, regs, address); } diff --git a/arch/x86/oprofile/op_model_ppro.c b/arch/x86/oprofile/op_model_ppro.c index 4899215..8eb0587 100644 --- a/arch/x86/oprofile/op_model_ppro.c +++ b/arch/x86/oprofile/op_model_ppro.c @@ -234,11 +234,11 @@ static void arch_perfmon_setup_counters(void) if (eax.split.version_id == 0 && current_cpu_data.x86 == 6 && current_cpu_data.x86_model == 15) { eax.split.version_id = 2; - eax.split.num_counters = 2; + eax.split.num_events = 2; eax.split.bit_width = 40; } - num_counters = eax.split.num_counters; + num_counters = eax.split.num_events; op_arch_perfmon_spec.num_counters = num_counters; op_arch_perfmon_spec.num_controls = num_counters; diff --git a/arch/x86/oprofile/op_x86_model.h b/arch/x86/oprofile/op_x86_model.h index b837761..7b8e75d 100644 --- a/arch/x86/oprofile/op_x86_model.h +++ b/arch/x86/oprofile/op_x86_model.h @@ -13,7 +13,7 @@ #define OP_X86_MODEL_H #include -#include +#include struct op_msr { unsigned long addr; diff --git a/drivers/char/sysrq.c b/drivers/char/sysrq.c index 50eecfe..44203ff 100644 --- a/drivers/char/sysrq.c +++ b/drivers/char/sysrq.c @@ -26,7 +26,7 @@ #include #include #include -#include +#include #include #include #include @@ -252,7 +252,7 @@ static void sysrq_handle_showregs(int key, struct tty_struct *tty) struct pt_regs *regs = get_irq_regs(); if (regs) show_regs(regs); - perf_counter_print_debug(); + perf_event_print_debug(); } static struct sysrq_key_op sysrq_showregs_op = { .handler = sysrq_handle_showregs, diff --git a/fs/exec.c b/fs/exec.c index 172ceb6..434dba7 100644 --- a/fs/exec.c +++ b/fs/exec.c @@ -33,7 +33,7 @@ #include #include #include -#include +#include #include #include #include @@ -923,7 +923,7 @@ void set_task_comm(struct task_struct *tsk, char *buf) task_lock(tsk); strlcpy(tsk->comm, buf, sizeof(tsk->comm)); task_unlock(tsk); - perf_counter_comm(tsk); + perf_event_comm(tsk); } int flush_old_exec(struct linux_binprm * bprm) @@ -997,7 +997,7 @@ int flush_old_exec(struct linux_binprm * bprm) * security domain: */ if (!get_dumpable(current->mm)) - perf_counter_exit_task(current); + perf_event_exit_task(current); /* An exec changes our domain. We are no longer part of the thread group */ diff --git a/include/asm-generic/unistd.h b/include/asm-generic/unistd.h index 1125e5a..d76b66a 100644 --- a/include/asm-generic/unistd.h +++ b/include/asm-generic/unistd.h @@ -620,8 +620,8 @@ __SYSCALL(__NR_move_pages, sys_move_pages) #define __NR_rt_tgsigqueueinfo 240 __SYSCALL(__NR_rt_tgsigqueueinfo, sys_rt_tgsigqueueinfo) -#define __NR_perf_counter_open 241 -__SYSCALL(__NR_perf_counter_open, sys_perf_counter_open) +#define __NR_perf_event_open 241 +__SYSCALL(__NR_perf_event_open, sys_perf_event_open) #undef __NR_syscalls #define __NR_syscalls 242 diff --git a/include/linux/init_task.h b/include/linux/init_task.h index 9e7f2e8..21a6f5d 100644 --- a/include/linux/init_task.h +++ b/include/linux/init_task.h @@ -106,13 +106,13 @@ extern struct group_info init_groups; extern struct cred init_cred; -#ifdef CONFIG_PERF_COUNTERS -# define INIT_PERF_COUNTERS(tsk) \ - .perf_counter_mutex = \ - __MUTEX_INITIALIZER(tsk.perf_counter_mutex), \ - .perf_counter_list = LIST_HEAD_INIT(tsk.perf_counter_list), +#ifdef CONFIG_PERF_EVENTS +# define INIT_PERF_EVENTS(tsk) \ + .perf_event_mutex = \ + __MUTEX_INITIALIZER(tsk.perf_event_mutex), \ + .perf_event_list = LIST_HEAD_INIT(tsk.perf_event_list), #else -# define INIT_PERF_COUNTERS(tsk) +# define INIT_PERF_EVENTS(tsk) #endif /* @@ -178,7 +178,7 @@ extern struct cred init_cred; }, \ .dirties = INIT_PROP_LOCAL_SINGLE(dirties), \ INIT_IDS \ - INIT_PERF_COUNTERS(tsk) \ + INIT_PERF_EVENTS(tsk) \ INIT_TRACE_IRQFLAGS \ INIT_LOCKDEP \ INIT_FTRACE_GRAPH \ diff --git a/include/linux/perf_counter.h b/include/linux/perf_counter.h deleted file mode 100644 index f648627..0000000 --- a/include/linux/perf_counter.h +++ /dev/null @@ -1,858 +0,0 @@ -/* - * Performance counters: - * - * Copyright (C) 2008-2009, Thomas Gleixner - * Copyright (C) 2008-2009, Red Hat, Inc., Ingo Molnar - * Copyright (C) 2008-2009, Red Hat, Inc., Peter Zijlstra - * - * Data type definitions, declarations, prototypes. - * - * Started by: Thomas Gleixner and Ingo Molnar - * - * For licencing details see kernel-base/COPYING - */ -#ifndef _LINUX_PERF_COUNTER_H -#define _LINUX_PERF_COUNTER_H - -#include -#include -#include - -/* - * User-space ABI bits: - */ - -/* - * attr.type - */ -enum perf_type_id { - PERF_TYPE_HARDWARE = 0, - PERF_TYPE_SOFTWARE = 1, - PERF_TYPE_TRACEPOINT = 2, - PERF_TYPE_HW_CACHE = 3, - PERF_TYPE_RAW = 4, - - PERF_TYPE_MAX, /* non-ABI */ -}; - -/* - * Generalized performance counter event types, used by the - * attr.event_id parameter of the sys_perf_counter_open() - * syscall: - */ -enum perf_hw_id { - /* - * Common hardware events, generalized by the kernel: - */ - PERF_COUNT_HW_CPU_CYCLES = 0, - PERF_COUNT_HW_INSTRUCTIONS = 1, - PERF_COUNT_HW_CACHE_REFERENCES = 2, - PERF_COUNT_HW_CACHE_MISSES = 3, - PERF_COUNT_HW_BRANCH_INSTRUCTIONS = 4, - PERF_COUNT_HW_BRANCH_MISSES = 5, - PERF_COUNT_HW_BUS_CYCLES = 6, - - PERF_COUNT_HW_MAX, /* non-ABI */ -}; - -/* - * Generalized hardware cache counters: - * - * { L1-D, L1-I, LLC, ITLB, DTLB, BPU } x - * { read, write, prefetch } x - * { accesses, misses } - */ -enum perf_hw_cache_id { - PERF_COUNT_HW_CACHE_L1D = 0, - PERF_COUNT_HW_CACHE_L1I = 1, - PERF_COUNT_HW_CACHE_LL = 2, - PERF_COUNT_HW_CACHE_DTLB = 3, - PERF_COUNT_HW_CACHE_ITLB = 4, - PERF_COUNT_HW_CACHE_BPU = 5, - - PERF_COUNT_HW_CACHE_MAX, /* non-ABI */ -}; - -enum perf_hw_cache_op_id { - PERF_COUNT_HW_CACHE_OP_READ = 0, - PERF_COUNT_HW_CACHE_OP_WRITE = 1, - PERF_COUNT_HW_CACHE_OP_PREFETCH = 2, - - PERF_COUNT_HW_CACHE_OP_MAX, /* non-ABI */ -}; - -enum perf_hw_cache_op_result_id { - PERF_COUNT_HW_CACHE_RESULT_ACCESS = 0, - PERF_COUNT_HW_CACHE_RESULT_MISS = 1, - - PERF_COUNT_HW_CACHE_RESULT_MAX, /* non-ABI */ -}; - -/* - * Special "software" counters provided by the kernel, even if the hardware - * does not support performance counters. These counters measure various - * physical and sw events of the kernel (and allow the profiling of them as - * well): - */ -enum perf_sw_ids { - PERF_COUNT_SW_CPU_CLOCK = 0, - PERF_COUNT_SW_TASK_CLOCK = 1, - PERF_COUNT_SW_PAGE_FAULTS = 2, - PERF_COUNT_SW_CONTEXT_SWITCHES = 3, - PERF_COUNT_SW_CPU_MIGRATIONS = 4, - PERF_COUNT_SW_PAGE_FAULTS_MIN = 5, - PERF_COUNT_SW_PAGE_FAULTS_MAJ = 6, - - PERF_COUNT_SW_MAX, /* non-ABI */ -}; - -/* - * Bits that can be set in attr.sample_type to request information - * in the overflow packets. - */ -enum perf_counter_sample_format { - PERF_SAMPLE_IP = 1U << 0, - PERF_SAMPLE_TID = 1U << 1, - PERF_SAMPLE_TIME = 1U << 2, - PERF_SAMPLE_ADDR = 1U << 3, - PERF_SAMPLE_READ = 1U << 4, - PERF_SAMPLE_CALLCHAIN = 1U << 5, - PERF_SAMPLE_ID = 1U << 6, - PERF_SAMPLE_CPU = 1U << 7, - PERF_SAMPLE_PERIOD = 1U << 8, - PERF_SAMPLE_STREAM_ID = 1U << 9, - PERF_SAMPLE_RAW = 1U << 10, - - PERF_SAMPLE_MAX = 1U << 11, /* non-ABI */ -}; - -/* - * The format of the data returned by read() on a perf counter fd, - * as specified by attr.read_format: - * - * struct read_format { - * { u64 value; - * { u64 time_enabled; } && PERF_FORMAT_ENABLED - * { u64 time_running; } && PERF_FORMAT_RUNNING - * { u64 id; } && PERF_FORMAT_ID - * } && !PERF_FORMAT_GROUP - * - * { u64 nr; - * { u64 time_enabled; } && PERF_FORMAT_ENABLED - * { u64 time_running; } && PERF_FORMAT_RUNNING - * { u64 value; - * { u64 id; } && PERF_FORMAT_ID - * } cntr[nr]; - * } && PERF_FORMAT_GROUP - * }; - */ -enum perf_counter_read_format { - PERF_FORMAT_TOTAL_TIME_ENABLED = 1U << 0, - PERF_FORMAT_TOTAL_TIME_RUNNING = 1U << 1, - PERF_FORMAT_ID = 1U << 2, - PERF_FORMAT_GROUP = 1U << 3, - - PERF_FORMAT_MAX = 1U << 4, /* non-ABI */ -}; - -#define PERF_ATTR_SIZE_VER0 64 /* sizeof first published struct */ - -/* - * Hardware event to monitor via a performance monitoring counter: - */ -struct perf_counter_attr { - - /* - * Major type: hardware/software/tracepoint/etc. - */ - __u32 type; - - /* - * Size of the attr structure, for fwd/bwd compat. - */ - __u32 size; - - /* - * Type specific configuration information. - */ - __u64 config; - - union { - __u64 sample_period; - __u64 sample_freq; - }; - - __u64 sample_type; - __u64 read_format; - - __u64 disabled : 1, /* off by default */ - inherit : 1, /* children inherit it */ - pinned : 1, /* must always be on PMU */ - exclusive : 1, /* only group on PMU */ - exclude_user : 1, /* don't count user */ - exclude_kernel : 1, /* ditto kernel */ - exclude_hv : 1, /* ditto hypervisor */ - exclude_idle : 1, /* don't count when idle */ - mmap : 1, /* include mmap data */ - comm : 1, /* include comm data */ - freq : 1, /* use freq, not period */ - inherit_stat : 1, /* per task counts */ - enable_on_exec : 1, /* next exec enables */ - task : 1, /* trace fork/exit */ - watermark : 1, /* wakeup_watermark */ - - __reserved_1 : 49; - - union { - __u32 wakeup_events; /* wakeup every n events */ - __u32 wakeup_watermark; /* bytes before wakeup */ - }; - __u32 __reserved_2; - - __u64 __reserved_3; -}; - -/* - * Ioctls that can be done on a perf counter fd: - */ -#define PERF_COUNTER_IOC_ENABLE _IO ('$', 0) -#define PERF_COUNTER_IOC_DISABLE _IO ('$', 1) -#define PERF_COUNTER_IOC_REFRESH _IO ('$', 2) -#define PERF_COUNTER_IOC_RESET _IO ('$', 3) -#define PERF_COUNTER_IOC_PERIOD _IOW('$', 4, u64) -#define PERF_COUNTER_IOC_SET_OUTPUT _IO ('$', 5) - -enum perf_counter_ioc_flags { - PERF_IOC_FLAG_GROUP = 1U << 0, -}; - -/* - * Structure of the page that can be mapped via mmap - */ -struct perf_counter_mmap_page { - __u32 version; /* version number of this structure */ - __u32 compat_version; /* lowest version this is compat with */ - - /* - * Bits needed to read the hw counters in user-space. - * - * u32 seq; - * s64 count; - * - * do { - * seq = pc->lock; - * - * barrier() - * if (pc->index) { - * count = pmc_read(pc->index - 1); - * count += pc->offset; - * } else - * goto regular_read; - * - * barrier(); - * } while (pc->lock != seq); - * - * NOTE: for obvious reason this only works on self-monitoring - * processes. - */ - __u32 lock; /* seqlock for synchronization */ - __u32 index; /* hardware counter identifier */ - __s64 offset; /* add to hardware counter value */ - __u64 time_enabled; /* time counter active */ - __u64 time_running; /* time counter on cpu */ - - /* - * Hole for extension of the self monitor capabilities - */ - - __u64 __reserved[123]; /* align to 1k */ - - /* - * Control data for the mmap() data buffer. - * - * User-space reading the @data_head value should issue an rmb(), on - * SMP capable platforms, after reading this value -- see - * perf_counter_wakeup(). - * - * When the mapping is PROT_WRITE the @data_tail value should be - * written by userspace to reflect the last read data. In this case - * the kernel will not over-write unread data. - */ - __u64 data_head; /* head in the data section */ - __u64 data_tail; /* user-space written tail */ -}; - -#define PERF_EVENT_MISC_CPUMODE_MASK (3 << 0) -#define PERF_EVENT_MISC_CPUMODE_UNKNOWN (0 << 0) -#define PERF_EVENT_MISC_KERNEL (1 << 0) -#define PERF_EVENT_MISC_USER (2 << 0) -#define PERF_EVENT_MISC_HYPERVISOR (3 << 0) - -struct perf_event_header { - __u32 type; - __u16 misc; - __u16 size; -}; - -enum perf_event_type { - - /* - * The MMAP events record the PROT_EXEC mappings so that we can - * correlate userspace IPs to code. They have the following structure: - * - * struct { - * struct perf_event_header header; - * - * u32 pid, tid; - * u64 addr; - * u64 len; - * u64 pgoff; - * char filename[]; - * }; - */ - PERF_EVENT_MMAP = 1, - - /* - * struct { - * struct perf_event_header header; - * u64 id; - * u64 lost; - * }; - */ - PERF_EVENT_LOST = 2, - - /* - * struct { - * struct perf_event_header header; - * - * u32 pid, tid; - * char comm[]; - * }; - */ - PERF_EVENT_COMM = 3, - - /* - * struct { - * struct perf_event_header header; - * u32 pid, ppid; - * u32 tid, ptid; - * u64 time; - * }; - */ - PERF_EVENT_EXIT = 4, - - /* - * struct { - * struct perf_event_header header; - * u64 time; - * u64 id; - * u64 stream_id; - * }; - */ - PERF_EVENT_THROTTLE = 5, - PERF_EVENT_UNTHROTTLE = 6, - - /* - * struct { - * struct perf_event_header header; - * u32 pid, ppid; - * u32 tid, ptid; - * { u64 time; } && PERF_SAMPLE_TIME - * }; - */ - PERF_EVENT_FORK = 7, - - /* - * struct { - * struct perf_event_header header; - * u32 pid, tid; - * - * struct read_format values; - * }; - */ - PERF_EVENT_READ = 8, - - /* - * struct { - * struct perf_event_header header; - * - * { u64 ip; } && PERF_SAMPLE_IP - * { u32 pid, tid; } && PERF_SAMPLE_TID - * { u64 time; } && PERF_SAMPLE_TIME - * { u64 addr; } && PERF_SAMPLE_ADDR - * { u64 id; } && PERF_SAMPLE_ID - * { u64 stream_id;} && PERF_SAMPLE_STREAM_ID - * { u32 cpu, res; } && PERF_SAMPLE_CPU - * { u64 period; } && PERF_SAMPLE_PERIOD - * - * { struct read_format values; } && PERF_SAMPLE_READ - * - * { u64 nr, - * u64 ips[nr]; } && PERF_SAMPLE_CALLCHAIN - * - * # - * # The RAW record below is opaque data wrt the ABI - * # - * # That is, the ABI doesn't make any promises wrt to - * # the stability of its content, it may vary depending - * # on event, hardware, kernel version and phase of - * # the moon. - * # - * # In other words, PERF_SAMPLE_RAW contents are not an ABI. - * # - * - * { u32 size; - * char data[size];}&& PERF_SAMPLE_RAW - * }; - */ - PERF_EVENT_SAMPLE = 9, - - PERF_EVENT_MAX, /* non-ABI */ -}; - -enum perf_callchain_context { - PERF_CONTEXT_HV = (__u64)-32, - PERF_CONTEXT_KERNEL = (__u64)-128, - PERF_CONTEXT_USER = (__u64)-512, - - PERF_CONTEXT_GUEST = (__u64)-2048, - PERF_CONTEXT_GUEST_KERNEL = (__u64)-2176, - PERF_CONTEXT_GUEST_USER = (__u64)-2560, - - PERF_CONTEXT_MAX = (__u64)-4095, -}; - -#define PERF_FLAG_FD_NO_GROUP (1U << 0) -#define PERF_FLAG_FD_OUTPUT (1U << 1) - -#ifdef __KERNEL__ -/* - * Kernel-internal data types and definitions: - */ - -#ifdef CONFIG_PERF_COUNTERS -# include -#endif - -#include -#include -#include -#include -#include -#include -#include -#include -#include - -#define PERF_MAX_STACK_DEPTH 255 - -struct perf_callchain_entry { - __u64 nr; - __u64 ip[PERF_MAX_STACK_DEPTH]; -}; - -struct perf_raw_record { - u32 size; - void *data; -}; - -struct task_struct; - -/** - * struct hw_perf_counter - performance counter hardware details: - */ -struct hw_perf_counter { -#ifdef CONFIG_PERF_COUNTERS - union { - struct { /* hardware */ - u64 config; - unsigned long config_base; - unsigned long counter_base; - int idx; - }; - union { /* software */ - atomic64_t count; - struct hrtimer hrtimer; - }; - }; - atomic64_t prev_count; - u64 sample_period; - u64 last_period; - atomic64_t period_left; - u64 interrupts; - - u64 freq_count; - u64 freq_interrupts; - u64 freq_stamp; -#endif -}; - -struct perf_counter; - -/** - * struct pmu - generic performance monitoring unit - */ -struct pmu { - int (*enable) (struct perf_counter *counter); - void (*disable) (struct perf_counter *counter); - void (*read) (struct perf_counter *counter); - void (*unthrottle) (struct perf_counter *counter); -}; - -/** - * enum perf_counter_active_state - the states of a counter - */ -enum perf_counter_active_state { - PERF_COUNTER_STATE_ERROR = -2, - PERF_COUNTER_STATE_OFF = -1, - PERF_COUNTER_STATE_INACTIVE = 0, - PERF_COUNTER_STATE_ACTIVE = 1, -}; - -struct file; - -struct perf_mmap_data { - struct rcu_head rcu_head; - int nr_pages; /* nr of data pages */ - int writable; /* are we writable */ - int nr_locked; /* nr pages mlocked */ - - atomic_t poll; /* POLL_ for wakeups */ - atomic_t events; /* event limit */ - - atomic_long_t head; /* write position */ - atomic_long_t done_head; /* completed head */ - - atomic_t lock; /* concurrent writes */ - atomic_t wakeup; /* needs a wakeup */ - atomic_t lost; /* nr records lost */ - - long watermark; /* wakeup watermark */ - - struct perf_counter_mmap_page *user_page; - void *data_pages[0]; -}; - -struct perf_pending_entry { - struct perf_pending_entry *next; - void (*func)(struct perf_pending_entry *); -}; - -/** - * struct perf_counter - performance counter kernel representation: - */ -struct perf_counter { -#ifdef CONFIG_PERF_COUNTERS - struct list_head group_entry; - struct list_head event_entry; - struct list_head sibling_list; - int nr_siblings; - struct perf_counter *group_leader; - struct perf_counter *output; - const struct pmu *pmu; - - enum perf_counter_active_state state; - atomic64_t count; - - /* - * These are the total time in nanoseconds that the counter - * has been enabled (i.e. eligible to run, and the task has - * been scheduled in, if this is a per-task counter) - * and running (scheduled onto the CPU), respectively. - * - * They are computed from tstamp_enabled, tstamp_running and - * tstamp_stopped when the counter is in INACTIVE or ACTIVE state. - */ - u64 total_time_enabled; - u64 total_time_running; - - /* - * These are timestamps used for computing total_time_enabled - * and total_time_running when the counter is in INACTIVE or - * ACTIVE state, measured in nanoseconds from an arbitrary point - * in time. - * tstamp_enabled: the notional time when the counter was enabled - * tstamp_running: the notional time when the counter was scheduled on - * tstamp_stopped: in INACTIVE state, the notional time when the - * counter was scheduled off. - */ - u64 tstamp_enabled; - u64 tstamp_running; - u64 tstamp_stopped; - - struct perf_counter_attr attr; - struct hw_perf_counter hw; - - struct perf_counter_context *ctx; - struct file *filp; - - /* - * These accumulate total time (in nanoseconds) that children - * counters have been enabled and running, respectively. - */ - atomic64_t child_total_time_enabled; - atomic64_t child_total_time_running; - - /* - * Protect attach/detach and child_list: - */ - struct mutex child_mutex; - struct list_head child_list; - struct perf_counter *parent; - - int oncpu; - int cpu; - - struct list_head owner_entry; - struct task_struct *owner; - - /* mmap bits */ - struct mutex mmap_mutex; - atomic_t mmap_count; - struct perf_mmap_data *data; - - /* poll related */ - wait_queue_head_t waitq; - struct fasync_struct *fasync; - - /* delayed work for NMIs and such */ - int pending_wakeup; - int pending_kill; - int pending_disable; - struct perf_pending_entry pending; - - atomic_t event_limit; - - void (*destroy)(struct perf_counter *); - struct rcu_head rcu_head; - - struct pid_namespace *ns; - u64 id; -#endif -}; - -/** - * struct perf_counter_context - counter context structure - * - * Used as a container for task counters and CPU counters as well: - */ -struct perf_counter_context { - /* - * Protect the states of the counters in the list, - * nr_active, and the list: - */ - spinlock_t lock; - /* - * Protect the list of counters. Locking either mutex or lock - * is sufficient to ensure the list doesn't change; to change - * the list you need to lock both the mutex and the spinlock. - */ - struct mutex mutex; - - struct list_head group_list; - struct list_head event_list; - int nr_counters; - int nr_active; - int is_active; - int nr_stat; - atomic_t refcount; - struct task_struct *task; - - /* - * Context clock, runs when context enabled. - */ - u64 time; - u64 timestamp; - - /* - * These fields let us detect when two contexts have both - * been cloned (inherited) from a common ancestor. - */ - struct perf_counter_context *parent_ctx; - u64 parent_gen; - u64 generation; - int pin_count; - struct rcu_head rcu_head; -}; - -/** - * struct perf_counter_cpu_context - per cpu counter context structure - */ -struct perf_cpu_context { - struct perf_counter_context ctx; - struct perf_counter_context *task_ctx; - int active_oncpu; - int max_pertask; - int exclusive; - - /* - * Recursion avoidance: - * - * task, softirq, irq, nmi context - */ - int recursion[4]; -}; - -struct perf_output_handle { - struct perf_counter *counter; - struct perf_mmap_data *data; - unsigned long head; - unsigned long offset; - int nmi; - int sample; - int locked; - unsigned long flags; -}; - -#ifdef CONFIG_PERF_COUNTERS - -/* - * Set by architecture code: - */ -extern int perf_max_counters; - -extern const struct pmu *hw_perf_counter_init(struct perf_counter *counter); - -extern void perf_counter_task_sched_in(struct task_struct *task, int cpu); -extern void perf_counter_task_sched_out(struct task_struct *task, - struct task_struct *next, int cpu); -extern void perf_counter_task_tick(struct task_struct *task, int cpu); -extern int perf_counter_init_task(struct task_struct *child); -extern void perf_counter_exit_task(struct task_struct *child); -extern void perf_counter_free_task(struct task_struct *task); -extern void set_perf_counter_pending(void); -extern void perf_counter_do_pending(void); -extern void perf_counter_print_debug(void); -extern void __perf_disable(void); -extern bool __perf_enable(void); -extern void perf_disable(void); -extern void perf_enable(void); -extern int perf_counter_task_disable(void); -extern int perf_counter_task_enable(void); -extern int hw_perf_group_sched_in(struct perf_counter *group_leader, - struct perf_cpu_context *cpuctx, - struct perf_counter_context *ctx, int cpu); -extern void perf_counter_update_userpage(struct perf_counter *counter); - -struct perf_sample_data { - u64 type; - - u64 ip; - struct { - u32 pid; - u32 tid; - } tid_entry; - u64 time; - u64 addr; - u64 id; - u64 stream_id; - struct { - u32 cpu; - u32 reserved; - } cpu_entry; - u64 period; - struct perf_callchain_entry *callchain; - struct perf_raw_record *raw; -}; - -extern void perf_output_sample(struct perf_output_handle *handle, - struct perf_event_header *header, - struct perf_sample_data *data, - struct perf_counter *counter); -extern void perf_prepare_sample(struct perf_event_header *header, - struct perf_sample_data *data, - struct perf_counter *counter, - struct pt_regs *regs); - -extern int perf_counter_overflow(struct perf_counter *counter, int nmi, - struct perf_sample_data *data, - struct pt_regs *regs); - -/* - * Return 1 for a software counter, 0 for a hardware counter - */ -static inline int is_software_counter(struct perf_counter *counter) -{ - return (counter->attr.type != PERF_TYPE_RAW) && - (counter->attr.type != PERF_TYPE_HARDWARE) && - (counter->attr.type != PERF_TYPE_HW_CACHE); -} - -extern atomic_t perf_swcounter_enabled[PERF_COUNT_SW_MAX]; - -extern void __perf_swcounter_event(u32, u64, int, struct pt_regs *, u64); - -static inline void -perf_swcounter_event(u32 event, u64 nr, int nmi, struct pt_regs *regs, u64 addr) -{ - if (atomic_read(&perf_swcounter_enabled[event])) - __perf_swcounter_event(event, nr, nmi, regs, addr); -} - -extern void __perf_counter_mmap(struct vm_area_struct *vma); - -static inline void perf_counter_mmap(struct vm_area_struct *vma) -{ - if (vma->vm_flags & VM_EXEC) - __perf_counter_mmap(vma); -} - -extern void perf_counter_comm(struct task_struct *tsk); -extern void perf_counter_fork(struct task_struct *tsk); - -extern struct perf_callchain_entry *perf_callchain(struct pt_regs *regs); - -extern int sysctl_perf_counter_paranoid; -extern int sysctl_perf_counter_mlock; -extern int sysctl_perf_counter_sample_rate; - -extern void perf_counter_init(void); -extern void perf_tpcounter_event(int event_id, u64 addr, u64 count, - void *record, int entry_size); - -#ifndef perf_misc_flags -#define perf_misc_flags(regs) (user_mode(regs) ? PERF_EVENT_MISC_USER : \ - PERF_EVENT_MISC_KERNEL) -#define perf_instruction_pointer(regs) instruction_pointer(regs) -#endif - -extern int perf_output_begin(struct perf_output_handle *handle, - struct perf_counter *counter, unsigned int size, - int nmi, int sample); -extern void perf_output_end(struct perf_output_handle *handle); -extern void perf_output_copy(struct perf_output_handle *handle, - const void *buf, unsigned int len); -#else -static inline void -perf_counter_task_sched_in(struct task_struct *task, int cpu) { } -static inline void -perf_counter_task_sched_out(struct task_struct *task, - struct task_struct *next, int cpu) { } -static inline void -perf_counter_task_tick(struct task_struct *task, int cpu) { } -static inline int perf_counter_init_task(struct task_struct *child) { return 0; } -static inline void perf_counter_exit_task(struct task_struct *child) { } -static inline void perf_counter_free_task(struct task_struct *task) { } -static inline void perf_counter_do_pending(void) { } -static inline void perf_counter_print_debug(void) { } -static inline void perf_disable(void) { } -static inline void perf_enable(void) { } -static inline int perf_counter_task_disable(void) { return -EINVAL; } -static inline int perf_counter_task_enable(void) { return -EINVAL; } - -static inline void -perf_swcounter_event(u32 event, u64 nr, int nmi, - struct pt_regs *regs, u64 addr) { } - -static inline void perf_counter_mmap(struct vm_area_struct *vma) { } -static inline void perf_counter_comm(struct task_struct *tsk) { } -static inline void perf_counter_fork(struct task_struct *tsk) { } -static inline void perf_counter_init(void) { } - -#endif - -#define perf_output_put(handle, x) \ - perf_output_copy((handle), &(x), sizeof(x)) - -#endif /* __KERNEL__ */ -#endif /* _LINUX_PERF_COUNTER_H */ diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h new file mode 100644 index 0000000..ae9d9ed --- /dev/null +++ b/include/linux/perf_event.h @@ -0,0 +1,858 @@ +/* + * Performance events: + * + * Copyright (C) 2008-2009, Thomas Gleixner + * Copyright (C) 2008-2009, Red Hat, Inc., Ingo Molnar + * Copyright (C) 2008-2009, Red Hat, Inc., Peter Zijlstra + * + * Data type definitions, declarations, prototypes. + * + * Started by: Thomas Gleixner and Ingo Molnar + * + * For licencing details see kernel-base/COPYING + */ +#ifndef _LINUX_PERF_EVENT_H +#define _LINUX_PERF_EVENT_H + +#include +#include +#include + +/* + * User-space ABI bits: + */ + +/* + * attr.type + */ +enum perf_type_id { + PERF_TYPE_HARDWARE = 0, + PERF_TYPE_SOFTWARE = 1, + PERF_TYPE_TRACEPOINT = 2, + PERF_TYPE_HW_CACHE = 3, + PERF_TYPE_RAW = 4, + + PERF_TYPE_MAX, /* non-ABI */ +}; + +/* + * Generalized performance event event_id types, used by the + * attr.event_id parameter of the sys_perf_event_open() + * syscall: + */ +enum perf_hw_id { + /* + * Common hardware events, generalized by the kernel: + */ + PERF_COUNT_HW_CPU_CYCLES = 0, + PERF_COUNT_HW_INSTRUCTIONS = 1, + PERF_COUNT_HW_CACHE_REFERENCES = 2, + PERF_COUNT_HW_CACHE_MISSES = 3, + PERF_COUNT_HW_BRANCH_INSTRUCTIONS = 4, + PERF_COUNT_HW_BRANCH_MISSES = 5, + PERF_COUNT_HW_BUS_CYCLES = 6, + + PERF_COUNT_HW_MAX, /* non-ABI */ +}; + +/* + * Generalized hardware cache events: + * + * { L1-D, L1-I, LLC, ITLB, DTLB, BPU } x + * { read, write, prefetch } x + * { accesses, misses } + */ +enum perf_hw_cache_id { + PERF_COUNT_HW_CACHE_L1D = 0, + PERF_COUNT_HW_CACHE_L1I = 1, + PERF_COUNT_HW_CACHE_LL = 2, + PERF_COUNT_HW_CACHE_DTLB = 3, + PERF_COUNT_HW_CACHE_ITLB = 4, + PERF_COUNT_HW_CACHE_BPU = 5, + + PERF_COUNT_HW_CACHE_MAX, /* non-ABI */ +}; + +enum perf_hw_cache_op_id { + PERF_COUNT_HW_CACHE_OP_READ = 0, + PERF_COUNT_HW_CACHE_OP_WRITE = 1, + PERF_COUNT_HW_CACHE_OP_PREFETCH = 2, + + PERF_COUNT_HW_CACHE_OP_MAX, /* non-ABI */ +}; + +enum perf_hw_cache_op_result_id { + PERF_COUNT_HW_CACHE_RESULT_ACCESS = 0, + PERF_COUNT_HW_CACHE_RESULT_MISS = 1, + + PERF_COUNT_HW_CACHE_RESULT_MAX, /* non-ABI */ +}; + +/* + * Special "software" events provided by the kernel, even if the hardware + * does not support performance events. These events measure various + * physical and sw events of the kernel (and allow the profiling of them as + * well): + */ +enum perf_sw_ids { + PERF_COUNT_SW_CPU_CLOCK = 0, + PERF_COUNT_SW_TASK_CLOCK = 1, + PERF_COUNT_SW_PAGE_FAULTS = 2, + PERF_COUNT_SW_CONTEXT_SWITCHES = 3, + PERF_COUNT_SW_CPU_MIGRATIONS = 4, + PERF_COUNT_SW_PAGE_FAULTS_MIN = 5, + PERF_COUNT_SW_PAGE_FAULTS_MAJ = 6, + + PERF_COUNT_SW_MAX, /* non-ABI */ +}; + +/* + * Bits that can be set in attr.sample_type to request information + * in the overflow packets. + */ +enum perf_event_sample_format { + PERF_SAMPLE_IP = 1U << 0, + PERF_SAMPLE_TID = 1U << 1, + PERF_SAMPLE_TIME = 1U << 2, + PERF_SAMPLE_ADDR = 1U << 3, + PERF_SAMPLE_READ = 1U << 4, + PERF_SAMPLE_CALLCHAIN = 1U << 5, + PERF_SAMPLE_ID = 1U << 6, + PERF_SAMPLE_CPU = 1U << 7, + PERF_SAMPLE_PERIOD = 1U << 8, + PERF_SAMPLE_STREAM_ID = 1U << 9, + PERF_SAMPLE_RAW = 1U << 10, + + PERF_SAMPLE_MAX = 1U << 11, /* non-ABI */ +}; + +/* + * The format of the data returned by read() on a perf event fd, + * as specified by attr.read_format: + * + * struct read_format { + * { u64 value; + * { u64 time_enabled; } && PERF_FORMAT_ENABLED + * { u64 time_running; } && PERF_FORMAT_RUNNING + * { u64 id; } && PERF_FORMAT_ID + * } && !PERF_FORMAT_GROUP + * + * { u64 nr; + * { u64 time_enabled; } && PERF_FORMAT_ENABLED + * { u64 time_running; } && PERF_FORMAT_RUNNING + * { u64 value; + * { u64 id; } && PERF_FORMAT_ID + * } cntr[nr]; + * } && PERF_FORMAT_GROUP + * }; + */ +enum perf_event_read_format { + PERF_FORMAT_TOTAL_TIME_ENABLED = 1U << 0, + PERF_FORMAT_TOTAL_TIME_RUNNING = 1U << 1, + PERF_FORMAT_ID = 1U << 2, + PERF_FORMAT_GROUP = 1U << 3, + + PERF_FORMAT_MAX = 1U << 4, /* non-ABI */ +}; + +#define PERF_ATTR_SIZE_VER0 64 /* sizeof first published struct */ + +/* + * Hardware event_id to monitor via a performance monitoring event: + */ +struct perf_event_attr { + + /* + * Major type: hardware/software/tracepoint/etc. + */ + __u32 type; + + /* + * Size of the attr structure, for fwd/bwd compat. + */ + __u32 size; + + /* + * Type specific configuration information. + */ + __u64 config; + + union { + __u64 sample_period; + __u64 sample_freq; + }; + + __u64 sample_type; + __u64 read_format; + + __u64 disabled : 1, /* off by default */ + inherit : 1, /* children inherit it */ + pinned : 1, /* must always be on PMU */ + exclusive : 1, /* only group on PMU */ + exclude_user : 1, /* don't count user */ + exclude_kernel : 1, /* ditto kernel */ + exclude_hv : 1, /* ditto hypervisor */ + exclude_idle : 1, /* don't count when idle */ + mmap : 1, /* include mmap data */ + comm : 1, /* include comm data */ + freq : 1, /* use freq, not period */ + inherit_stat : 1, /* per task counts */ + enable_on_exec : 1, /* next exec enables */ + task : 1, /* trace fork/exit */ + watermark : 1, /* wakeup_watermark */ + + __reserved_1 : 49; + + union { + __u32 wakeup_events; /* wakeup every n events */ + __u32 wakeup_watermark; /* bytes before wakeup */ + }; + __u32 __reserved_2; + + __u64 __reserved_3; +}; + +/* + * Ioctls that can be done on a perf event fd: + */ +#define PERF_EVENT_IOC_ENABLE _IO ('$', 0) +#define PERF_EVENT_IOC_DISABLE _IO ('$', 1) +#define PERF_EVENT_IOC_REFRESH _IO ('$', 2) +#define PERF_EVENT_IOC_RESET _IO ('$', 3) +#define PERF_EVENT_IOC_PERIOD _IOW('$', 4, u64) +#define PERF_EVENT_IOC_SET_OUTPUT _IO ('$', 5) + +enum perf_event_ioc_flags { + PERF_IOC_FLAG_GROUP = 1U << 0, +}; + +/* + * Structure of the page that can be mapped via mmap + */ +struct perf_event_mmap_page { + __u32 version; /* version number of this structure */ + __u32 compat_version; /* lowest version this is compat with */ + + /* + * Bits needed to read the hw events in user-space. + * + * u32 seq; + * s64 count; + * + * do { + * seq = pc->lock; + * + * barrier() + * if (pc->index) { + * count = pmc_read(pc->index - 1); + * count += pc->offset; + * } else + * goto regular_read; + * + * barrier(); + * } while (pc->lock != seq); + * + * NOTE: for obvious reason this only works on self-monitoring + * processes. + */ + __u32 lock; /* seqlock for synchronization */ + __u32 index; /* hardware event identifier */ + __s64 offset; /* add to hardware event value */ + __u64 time_enabled; /* time event active */ + __u64 time_running; /* time event on cpu */ + + /* + * Hole for extension of the self monitor capabilities + */ + + __u64 __reserved[123]; /* align to 1k */ + + /* + * Control data for the mmap() data buffer. + * + * User-space reading the @data_head value should issue an rmb(), on + * SMP capable platforms, after reading this value -- see + * perf_event_wakeup(). + * + * When the mapping is PROT_WRITE the @data_tail value should be + * written by userspace to reflect the last read data. In this case + * the kernel will not over-write unread data. + */ + __u64 data_head; /* head in the data section */ + __u64 data_tail; /* user-space written tail */ +}; + +#define PERF_RECORD_MISC_CPUMODE_MASK (3 << 0) +#define PERF_RECORD_MISC_CPUMODE_UNKNOWN (0 << 0) +#define PERF_RECORD_MISC_KERNEL (1 << 0) +#define PERF_RECORD_MISC_USER (2 << 0) +#define PERF_RECORD_MISC_HYPERVISOR (3 << 0) + +struct perf_event_header { + __u32 type; + __u16 misc; + __u16 size; +}; + +enum perf_event_type { + + /* + * The MMAP events record the PROT_EXEC mappings so that we can + * correlate userspace IPs to code. They have the following structure: + * + * struct { + * struct perf_event_header header; + * + * u32 pid, tid; + * u64 addr; + * u64 len; + * u64 pgoff; + * char filename[]; + * }; + */ + PERF_RECORD_MMAP = 1, + + /* + * struct { + * struct perf_event_header header; + * u64 id; + * u64 lost; + * }; + */ + PERF_RECORD_LOST = 2, + + /* + * struct { + * struct perf_event_header header; + * + * u32 pid, tid; + * char comm[]; + * }; + */ + PERF_RECORD_COMM = 3, + + /* + * struct { + * struct perf_event_header header; + * u32 pid, ppid; + * u32 tid, ptid; + * u64 time; + * }; + */ + PERF_RECORD_EXIT = 4, + + /* + * struct { + * struct perf_event_header header; + * u64 time; + * u64 id; + * u64 stream_id; + * }; + */ + PERF_RECORD_THROTTLE = 5, + PERF_RECORD_UNTHROTTLE = 6, + + /* + * struct { + * struct perf_event_header header; + * u32 pid, ppid; + * u32 tid, ptid; + * { u64 time; } && PERF_SAMPLE_TIME + * }; + */ + PERF_RECORD_FORK = 7, + + /* + * struct { + * struct perf_event_header header; + * u32 pid, tid; + * + * struct read_format values; + * }; + */ + PERF_RECORD_READ = 8, + + /* + * struct { + * struct perf_event_header header; + * + * { u64 ip; } && PERF_SAMPLE_IP + * { u32 pid, tid; } && PERF_SAMPLE_TID + * { u64 time; } && PERF_SAMPLE_TIME + * { u64 addr; } && PERF_SAMPLE_ADDR + * { u64 id; } && PERF_SAMPLE_ID + * { u64 stream_id;} && PERF_SAMPLE_STREAM_ID + * { u32 cpu, res; } && PERF_SAMPLE_CPU + * { u64 period; } && PERF_SAMPLE_PERIOD + * + * { struct read_format values; } && PERF_SAMPLE_READ + * + * { u64 nr, + * u64 ips[nr]; } && PERF_SAMPLE_CALLCHAIN + * + * # + * # The RAW record below is opaque data wrt the ABI + * # + * # That is, the ABI doesn't make any promises wrt to + * # the stability of its content, it may vary depending + * # on event_id, hardware, kernel version and phase of + * # the moon. + * # + * # In other words, PERF_SAMPLE_RAW contents are not an ABI. + * # + * + * { u32 size; + * char data[size];}&& PERF_SAMPLE_RAW + * }; + */ + PERF_RECORD_SAMPLE = 9, + + PERF_RECORD_MAX, /* non-ABI */ +}; + +enum perf_callchain_context { + PERF_CONTEXT_HV = (__u64)-32, + PERF_CONTEXT_KERNEL = (__u64)-128, + PERF_CONTEXT_USER = (__u64)-512, + + PERF_CONTEXT_GUEST = (__u64)-2048, + PERF_CONTEXT_GUEST_KERNEL = (__u64)-2176, + PERF_CONTEXT_GUEST_USER = (__u64)-2560, + + PERF_CONTEXT_MAX = (__u64)-4095, +}; + +#define PERF_FLAG_FD_NO_GROUP (1U << 0) +#define PERF_FLAG_FD_OUTPUT (1U << 1) + +#ifdef __KERNEL__ +/* + * Kernel-internal data types and definitions: + */ + +#ifdef CONFIG_PERF_EVENTS +# include +#endif + +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#define PERF_MAX_STACK_DEPTH 255 + +struct perf_callchain_entry { + __u64 nr; + __u64 ip[PERF_MAX_STACK_DEPTH]; +}; + +struct perf_raw_record { + u32 size; + void *data; +}; + +struct task_struct; + +/** + * struct hw_perf_event - performance event hardware details: + */ +struct hw_perf_event { +#ifdef CONFIG_PERF_EVENTS + union { + struct { /* hardware */ + u64 config; + unsigned long config_base; + unsigned long event_base; + int idx; + }; + union { /* software */ + atomic64_t count; + struct hrtimer hrtimer; + }; + }; + atomic64_t prev_count; + u64 sample_period; + u64 last_period; + atomic64_t period_left; + u64 interrupts; + + u64 freq_count; + u64 freq_interrupts; + u64 freq_stamp; +#endif +}; + +struct perf_event; + +/** + * struct pmu - generic performance monitoring unit + */ +struct pmu { + int (*enable) (struct perf_event *event); + void (*disable) (struct perf_event *event); + void (*read) (struct perf_event *event); + void (*unthrottle) (struct perf_event *event); +}; + +/** + * enum perf_event_active_state - the states of a event + */ +enum perf_event_active_state { + PERF_EVENT_STATE_ERROR = -2, + PERF_EVENT_STATE_OFF = -1, + PERF_EVENT_STATE_INACTIVE = 0, + PERF_EVENT_STATE_ACTIVE = 1, +}; + +struct file; + +struct perf_mmap_data { + struct rcu_head rcu_head; + int nr_pages; /* nr of data pages */ + int writable; /* are we writable */ + int nr_locked; /* nr pages mlocked */ + + atomic_t poll; /* POLL_ for wakeups */ + atomic_t events; /* event_id limit */ + + atomic_long_t head; /* write position */ + atomic_long_t done_head; /* completed head */ + + atomic_t lock; /* concurrent writes */ + atomic_t wakeup; /* needs a wakeup */ + atomic_t lost; /* nr records lost */ + + long watermark; /* wakeup watermark */ + + struct perf_event_mmap_page *user_page; + void *data_pages[0]; +}; + +struct perf_pending_entry { + struct perf_pending_entry *next; + void (*func)(struct perf_pending_entry *); +}; + +/** + * struct perf_event - performance event kernel representation: + */ +struct perf_event { +#ifdef CONFIG_PERF_EVENTS + struct list_head group_entry; + struct list_head event_entry; + struct list_head sibling_list; + int nr_siblings; + struct perf_event *group_leader; + struct perf_event *output; + const struct pmu *pmu; + + enum perf_event_active_state state; + atomic64_t count; + + /* + * These are the total time in nanoseconds that the event + * has been enabled (i.e. eligible to run, and the task has + * been scheduled in, if this is a per-task event) + * and running (scheduled onto the CPU), respectively. + * + * They are computed from tstamp_enabled, tstamp_running and + * tstamp_stopped when the event is in INACTIVE or ACTIVE state. + */ + u64 total_time_enabled; + u64 total_time_running; + + /* + * These are timestamps used for computing total_time_enabled + * and total_time_running when the event is in INACTIVE or + * ACTIVE state, measured in nanoseconds from an arbitrary point + * in time. + * tstamp_enabled: the notional time when the event was enabled + * tstamp_running: the notional time when the event was scheduled on + * tstamp_stopped: in INACTIVE state, the notional time when the + * event was scheduled off. + */ + u64 tstamp_enabled; + u64 tstamp_running; + u64 tstamp_stopped; + + struct perf_event_attr attr; + struct hw_perf_event hw; + + struct perf_event_context *ctx; + struct file *filp; + + /* + * These accumulate total time (in nanoseconds) that children + * events have been enabled and running, respectively. + */ + atomic64_t child_total_time_enabled; + atomic64_t child_total_time_running; + + /* + * Protect attach/detach and child_list: + */ + struct mutex child_mutex; + struct list_head child_list; + struct perf_event *parent; + + int oncpu; + int cpu; + + struct list_head owner_entry; + struct task_struct *owner; + + /* mmap bits */ + struct mutex mmap_mutex; + atomic_t mmap_count; + struct perf_mmap_data *data; + + /* poll related */ + wait_queue_head_t waitq; + struct fasync_struct *fasync; + + /* delayed work for NMIs and such */ + int pending_wakeup; + int pending_kill; + int pending_disable; + struct perf_pending_entry pending; + + atomic_t event_limit; + + void (*destroy)(struct perf_event *); + struct rcu_head rcu_head; + + struct pid_namespace *ns; + u64 id; +#endif +}; + +/** + * struct perf_event_context - event context structure + * + * Used as a container for task events and CPU events as well: + */ +struct perf_event_context { + /* + * Protect the states of the events in the list, + * nr_active, and the list: + */ + spinlock_t lock; + /* + * Protect the list of events. Locking either mutex or lock + * is sufficient to ensure the list doesn't change; to change + * the list you need to lock both the mutex and the spinlock. + */ + struct mutex mutex; + + struct list_head group_list; + struct list_head event_list; + int nr_events; + int nr_active; + int is_active; + int nr_stat; + atomic_t refcount; + struct task_struct *task; + + /* + * Context clock, runs when context enabled. + */ + u64 time; + u64 timestamp; + + /* + * These fields let us detect when two contexts have both + * been cloned (inherited) from a common ancestor. + */ + struct perf_event_context *parent_ctx; + u64 parent_gen; + u64 generation; + int pin_count; + struct rcu_head rcu_head; +}; + +/** + * struct perf_event_cpu_context - per cpu event context structure + */ +struct perf_cpu_context { + struct perf_event_context ctx; + struct perf_event_context *task_ctx; + int active_oncpu; + int max_pertask; + int exclusive; + + /* + * Recursion avoidance: + * + * task, softirq, irq, nmi context + */ + int recursion[4]; +}; + +struct perf_output_handle { + struct perf_event *event; + struct perf_mmap_data *data; + unsigned long head; + unsigned long offset; + int nmi; + int sample; + int locked; + unsigned long flags; +}; + +#ifdef CONFIG_PERF_EVENTS + +/* + * Set by architecture code: + */ +extern int perf_max_events; + +extern const struct pmu *hw_perf_event_init(struct perf_event *event); + +extern void perf_event_task_sched_in(struct task_struct *task, int cpu); +extern void perf_event_task_sched_out(struct task_struct *task, + struct task_struct *next, int cpu); +extern void perf_event_task_tick(struct task_struct *task, int cpu); +extern int perf_event_init_task(struct task_struct *child); +extern void perf_event_exit_task(struct task_struct *child); +extern void perf_event_free_task(struct task_struct *task); +extern void set_perf_event_pending(void); +extern void perf_event_do_pending(void); +extern void perf_event_print_debug(void); +extern void __perf_disable(void); +extern bool __perf_enable(void); +extern void perf_disable(void); +extern void perf_enable(void); +extern int perf_event_task_disable(void); +extern int perf_event_task_enable(void); +extern int hw_perf_group_sched_in(struct perf_event *group_leader, + struct perf_cpu_context *cpuctx, + struct perf_event_context *ctx, int cpu); +extern void perf_event_update_userpage(struct perf_event *event); + +struct perf_sample_data { + u64 type; + + u64 ip; + struct { + u32 pid; + u32 tid; + } tid_entry; + u64 time; + u64 addr; + u64 id; + u64 stream_id; + struct { + u32 cpu; + u32 reserved; + } cpu_entry; + u64 period; + struct perf_callchain_entry *callchain; + struct perf_raw_record *raw; +}; + +extern void perf_output_sample(struct perf_output_handle *handle, + struct perf_event_header *header, + struct perf_sample_data *data, + struct perf_event *event); +extern void perf_prepare_sample(struct perf_event_header *header, + struct perf_sample_data *data, + struct perf_event *event, + struct pt_regs *regs); + +extern int perf_event_overflow(struct perf_event *event, int nmi, + struct perf_sample_data *data, + struct pt_regs *regs); + +/* + * Return 1 for a software event, 0 for a hardware event + */ +static inline int is_software_event(struct perf_event *event) +{ + return (event->attr.type != PERF_TYPE_RAW) && + (event->attr.type != PERF_TYPE_HARDWARE) && + (event->attr.type != PERF_TYPE_HW_CACHE); +} + +extern atomic_t perf_swevent_enabled[PERF_COUNT_SW_MAX]; + +extern void __perf_sw_event(u32, u64, int, struct pt_regs *, u64); + +static inline void +perf_sw_event(u32 event_id, u64 nr, int nmi, struct pt_regs *regs, u64 addr) +{ + if (atomic_read(&perf_swevent_enabled[event_id])) + __perf_sw_event(event_id, nr, nmi, regs, addr); +} + +extern void __perf_event_mmap(struct vm_area_struct *vma); + +static inline void perf_event_mmap(struct vm_area_struct *vma) +{ + if (vma->vm_flags & VM_EXEC) + __perf_event_mmap(vma); +} + +extern void perf_event_comm(struct task_struct *tsk); +extern void perf_event_fork(struct task_struct *tsk); + +extern struct perf_callchain_entry *perf_callchain(struct pt_regs *regs); + +extern int sysctl_perf_event_paranoid; +extern int sysctl_perf_event_mlock; +extern int sysctl_perf_event_sample_rate; + +extern void perf_event_init(void); +extern void perf_tp_event(int event_id, u64 addr, u64 count, + void *record, int entry_size); + +#ifndef perf_misc_flags +#define perf_misc_flags(regs) (user_mode(regs) ? PERF_RECORD_MISC_USER : \ + PERF_RECORD_MISC_KERNEL) +#define perf_instruction_pointer(regs) instruction_pointer(regs) +#endif + +extern int perf_output_begin(struct perf_output_handle *handle, + struct perf_event *event, unsigned int size, + int nmi, int sample); +extern void perf_output_end(struct perf_output_handle *handle); +extern void perf_output_copy(struct perf_output_handle *handle, + const void *buf, unsigned int len); +#else +static inline void +perf_event_task_sched_in(struct task_struct *task, int cpu) { } +static inline void +perf_event_task_sched_out(struct task_struct *task, + struct task_struct *next, int cpu) { } +static inline void +perf_event_task_tick(struct task_struct *task, int cpu) { } +static inline int perf_event_init_task(struct task_struct *child) { return 0; } +static inline void perf_event_exit_task(struct task_struct *child) { } +static inline void perf_event_free_task(struct task_struct *task) { } +static inline void perf_event_do_pending(void) { } +static inline void perf_event_print_debug(void) { } +static inline void perf_disable(void) { } +static inline void perf_enable(void) { } +static inline int perf_event_task_disable(void) { return -EINVAL; } +static inline int perf_event_task_enable(void) { return -EINVAL; } + +static inline void +perf_sw_event(u32 event_id, u64 nr, int nmi, + struct pt_regs *regs, u64 addr) { } + +static inline void perf_event_mmap(struct vm_area_struct *vma) { } +static inline void perf_event_comm(struct task_struct *tsk) { } +static inline void perf_event_fork(struct task_struct *tsk) { } +static inline void perf_event_init(void) { } + +#endif + +#define perf_output_put(handle, x) \ + perf_output_copy((handle), &(x), sizeof(x)) + +#endif /* __KERNEL__ */ +#endif /* _LINUX_PERF_EVENT_H */ diff --git a/include/linux/prctl.h b/include/linux/prctl.h index b00df4c..07bff66 100644 --- a/include/linux/prctl.h +++ b/include/linux/prctl.h @@ -85,7 +85,7 @@ #define PR_SET_TIMERSLACK 29 #define PR_GET_TIMERSLACK 30 -#define PR_TASK_PERF_COUNTERS_DISABLE 31 -#define PR_TASK_PERF_COUNTERS_ENABLE 32 +#define PR_TASK_PERF_EVENTS_DISABLE 31 +#define PR_TASK_PERF_EVENTS_ENABLE 32 #endif /* _LINUX_PRCTL_H */ diff --git a/include/linux/sched.h b/include/linux/sched.h index 8af3d24..8b265a8 100644 --- a/include/linux/sched.h +++ b/include/linux/sched.h @@ -100,7 +100,7 @@ struct robust_list_head; struct bio; struct fs_struct; struct bts_context; -struct perf_counter_context; +struct perf_event_context; /* * List of flags we want to share for kernel threads, @@ -701,7 +701,7 @@ struct user_struct { #endif #endif -#ifdef CONFIG_PERF_COUNTERS +#ifdef CONFIG_PERF_EVENTS atomic_long_t locked_vm; #endif }; @@ -1449,10 +1449,10 @@ struct task_struct { struct list_head pi_state_list; struct futex_pi_state *pi_state_cache; #endif -#ifdef CONFIG_PERF_COUNTERS - struct perf_counter_context *perf_counter_ctxp; - struct mutex perf_counter_mutex; - struct list_head perf_counter_list; +#ifdef CONFIG_PERF_EVENTS + struct perf_event_context *perf_event_ctxp; + struct mutex perf_event_mutex; + struct list_head perf_event_list; #endif #ifdef CONFIG_NUMA struct mempolicy *mempolicy; /* Protected by alloc_lock */ diff --git a/include/linux/syscalls.h b/include/linux/syscalls.h index a8e3782..02f19f9 100644 --- a/include/linux/syscalls.h +++ b/include/linux/syscalls.h @@ -55,7 +55,7 @@ struct compat_timeval; struct robust_list_head; struct getcpu_cache; struct old_linux_dirent; -struct perf_counter_attr; +struct perf_event_attr; #include #include @@ -885,7 +885,7 @@ asmlinkage long sys_ppoll(struct pollfd __user *, unsigned int, int kernel_execve(const char *filename, char *const argv[], char *const envp[]); -asmlinkage long sys_perf_counter_open( - struct perf_counter_attr __user *attr_uptr, +asmlinkage long sys_perf_event_open( + struct perf_event_attr __user *attr_uptr, pid_t pid, int cpu, int group_fd, unsigned long flags); #endif diff --git a/include/trace/ftrace.h b/include/trace/ftrace.h index 72a3b43..ec91e78 100644 --- a/include/trace/ftrace.h +++ b/include/trace/ftrace.h @@ -378,7 +378,7 @@ static inline int ftrace_get_offsets_##call( \ #ifdef CONFIG_EVENT_PROFILE /* - * Generate the functions needed for tracepoint perf_counter support. + * Generate the functions needed for tracepoint perf_event support. * * NOTE: The insertion profile callback (ftrace_profile_) is defined later * @@ -656,7 +656,7 @@ __attribute__((section("_ftrace_events"))) event_##call = { \ * { * struct ftrace_data_offsets_ __maybe_unused __data_offsets; * struct ftrace_event_call *event_call = &event_; - * extern void perf_tpcounter_event(int, u64, u64, void *, int); + * extern void perf_tp_event(int, u64, u64, void *, int); * struct ftrace_raw_##call *entry; * u64 __addr = 0, __count = 1; * unsigned long irq_flags; @@ -691,7 +691,7 @@ __attribute__((section("_ftrace_events"))) event_##call = { \ * * <- affect our values * - * perf_tpcounter_event(event_call->id, __addr, __count, entry, + * perf_tp_event(event_call->id, __addr, __count, entry, * __entry_size); <- submit them to perf counter * } while (0); * @@ -712,7 +712,7 @@ static void ftrace_profile_##call(proto) \ { \ struct ftrace_data_offsets_##call __maybe_unused __data_offsets;\ struct ftrace_event_call *event_call = &event_##call; \ - extern void perf_tpcounter_event(int, u64, u64, void *, int); \ + extern void perf_tp_event(int, u64, u64, void *, int); \ struct ftrace_raw_##call *entry; \ u64 __addr = 0, __count = 1; \ unsigned long irq_flags; \ @@ -742,7 +742,7 @@ static void ftrace_profile_##call(proto) \ \ { assign; } \ \ - perf_tpcounter_event(event_call->id, __addr, __count, entry,\ + perf_tp_event(event_call->id, __addr, __count, entry,\ __entry_size); \ } while (0); \ \ diff --git a/init/Kconfig b/init/Kconfig index 8e8b76d..cfdf5c3 100644 --- a/init/Kconfig +++ b/init/Kconfig @@ -915,17 +915,17 @@ config AIO by some high performance threaded applications. Disabling this option saves about 7k. -config HAVE_PERF_COUNTERS +config HAVE_PERF_EVENTS bool help See tools/perf/design.txt for details. menu "Performance Counters" -config PERF_COUNTERS +config PERF_EVENTS bool "Kernel Performance Counters" default y if PROFILING - depends on HAVE_PERF_COUNTERS + depends on HAVE_PERF_EVENTS select ANON_INODES help Enable kernel support for performance counter hardware. @@ -947,7 +947,7 @@ config PERF_COUNTERS config EVENT_PROFILE bool "Tracepoint profiling sources" - depends on PERF_COUNTERS && EVENT_TRACING + depends on PERF_EVENTS && EVENT_TRACING default y help Allow the use of tracepoints as software performance counters. diff --git a/kernel/Makefile b/kernel/Makefile index 3d9c7e2..e26a546 100644 --- a/kernel/Makefile +++ b/kernel/Makefile @@ -96,7 +96,7 @@ obj-$(CONFIG_X86_DS) += trace/ obj-$(CONFIG_RING_BUFFER) += trace/ obj-$(CONFIG_SMP) += sched_cpupri.o obj-$(CONFIG_SLOW_WORK) += slow-work.o -obj-$(CONFIG_PERF_COUNTERS) += perf_counter.o +obj-$(CONFIG_PERF_EVENTS) += perf_event.o ifneq ($(CONFIG_SCHED_OMIT_FRAME_POINTER),y) # According to Alan Modra , the -fno-omit-frame-pointer is diff --git a/kernel/exit.c b/kernel/exit.c index ae5d866..e47ee8a 100644 --- a/kernel/exit.c +++ b/kernel/exit.c @@ -47,7 +47,7 @@ #include #include #include -#include +#include #include #include @@ -154,8 +154,8 @@ static void delayed_put_task_struct(struct rcu_head *rhp) { struct task_struct *tsk = container_of(rhp, struct task_struct, rcu); -#ifdef CONFIG_PERF_COUNTERS - WARN_ON_ONCE(tsk->perf_counter_ctxp); +#ifdef CONFIG_PERF_EVENTS + WARN_ON_ONCE(tsk->perf_event_ctxp); #endif trace_sched_process_free(tsk); put_task_struct(tsk); @@ -981,7 +981,7 @@ NORET_TYPE void do_exit(long code) * Flush inherited counters to the parent - before the parent * gets woken up by child-exit notifications. */ - perf_counter_exit_task(tsk); + perf_event_exit_task(tsk); exit_notify(tsk, group_dead); #ifdef CONFIG_NUMA diff --git a/kernel/fork.c b/kernel/fork.c index bfee931..2cebfb2 100644 --- a/kernel/fork.c +++ b/kernel/fork.c @@ -61,7 +61,7 @@ #include #include #include -#include +#include #include #include @@ -1078,7 +1078,7 @@ static struct task_struct *copy_process(unsigned long clone_flags, /* Perform scheduler related setup. Assign this task to a CPU. */ sched_fork(p, clone_flags); - retval = perf_counter_init_task(p); + retval = perf_event_init_task(p); if (retval) goto bad_fork_cleanup_policy; @@ -1253,7 +1253,7 @@ static struct task_struct *copy_process(unsigned long clone_flags, write_unlock_irq(&tasklist_lock); proc_fork_connector(p); cgroup_post_fork(p); - perf_counter_fork(p); + perf_event_fork(p); return p; bad_fork_free_pid: @@ -1280,7 +1280,7 @@ bad_fork_cleanup_semundo: bad_fork_cleanup_audit: audit_free(p); bad_fork_cleanup_policy: - perf_counter_free_task(p); + perf_event_free_task(p); #ifdef CONFIG_NUMA mpol_put(p->mempolicy); bad_fork_cleanup_cgroup: diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c deleted file mode 100644 index 62de0db..0000000 --- a/kernel/perf_counter.c +++ /dev/null @@ -1,5000 +0,0 @@ -/* - * Performance counter core code - * - * Copyright (C) 2008 Thomas Gleixner - * Copyright (C) 2008-2009 Red Hat, Inc., Ingo Molnar - * Copyright (C) 2008-2009 Red Hat, Inc., Peter Zijlstra - * Copyright © 2009 Paul Mackerras, IBM Corp. - * - * For licensing details see kernel-base/COPYING - */ - -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include - -#include - -/* - * Each CPU has a list of per CPU counters: - */ -DEFINE_PER_CPU(struct perf_cpu_context, perf_cpu_context); - -int perf_max_counters __read_mostly = 1; -static int perf_reserved_percpu __read_mostly; -static int perf_overcommit __read_mostly = 1; - -static atomic_t nr_counters __read_mostly; -static atomic_t nr_mmap_counters __read_mostly; -static atomic_t nr_comm_counters __read_mostly; -static atomic_t nr_task_counters __read_mostly; - -/* - * perf counter paranoia level: - * -1 - not paranoid at all - * 0 - disallow raw tracepoint access for unpriv - * 1 - disallow cpu counters for unpriv - * 2 - disallow kernel profiling for unpriv - */ -int sysctl_perf_counter_paranoid __read_mostly = 1; - -static inline bool perf_paranoid_tracepoint_raw(void) -{ - return sysctl_perf_counter_paranoid > -1; -} - -static inline bool perf_paranoid_cpu(void) -{ - return sysctl_perf_counter_paranoid > 0; -} - -static inline bool perf_paranoid_kernel(void) -{ - return sysctl_perf_counter_paranoid > 1; -} - -int sysctl_perf_counter_mlock __read_mostly = 512; /* 'free' kb per user */ - -/* - * max perf counter sample rate - */ -int sysctl_perf_counter_sample_rate __read_mostly = 100000; - -static atomic64_t perf_counter_id; - -/* - * Lock for (sysadmin-configurable) counter reservations: - */ -static DEFINE_SPINLOCK(perf_resource_lock); - -/* - * Architecture provided APIs - weak aliases: - */ -extern __weak const struct pmu *hw_perf_counter_init(struct perf_counter *counter) -{ - return NULL; -} - -void __weak hw_perf_disable(void) { barrier(); } -void __weak hw_perf_enable(void) { barrier(); } - -void __weak hw_perf_counter_setup(int cpu) { barrier(); } -void __weak hw_perf_counter_setup_online(int cpu) { barrier(); } - -int __weak -hw_perf_group_sched_in(struct perf_counter *group_leader, - struct perf_cpu_context *cpuctx, - struct perf_counter_context *ctx, int cpu) -{ - return 0; -} - -void __weak perf_counter_print_debug(void) { } - -static DEFINE_PER_CPU(int, perf_disable_count); - -void __perf_disable(void) -{ - __get_cpu_var(perf_disable_count)++; -} - -bool __perf_enable(void) -{ - return !--__get_cpu_var(perf_disable_count); -} - -void perf_disable(void) -{ - __perf_disable(); - hw_perf_disable(); -} - -void perf_enable(void) -{ - if (__perf_enable()) - hw_perf_enable(); -} - -static void get_ctx(struct perf_counter_context *ctx) -{ - WARN_ON(!atomic_inc_not_zero(&ctx->refcount)); -} - -static void free_ctx(struct rcu_head *head) -{ - struct perf_counter_context *ctx; - - ctx = container_of(head, struct perf_counter_context, rcu_head); - kfree(ctx); -} - -static void put_ctx(struct perf_counter_context *ctx) -{ - if (atomic_dec_and_test(&ctx->refcount)) { - if (ctx->parent_ctx) - put_ctx(ctx->parent_ctx); - if (ctx->task) - put_task_struct(ctx->task); - call_rcu(&ctx->rcu_head, free_ctx); - } -} - -static void unclone_ctx(struct perf_counter_context *ctx) -{ - if (ctx->parent_ctx) { - put_ctx(ctx->parent_ctx); - ctx->parent_ctx = NULL; - } -} - -/* - * If we inherit counters we want to return the parent counter id - * to userspace. - */ -static u64 primary_counter_id(struct perf_counter *counter) -{ - u64 id = counter->id; - - if (counter->parent) - id = counter->parent->id; - - return id; -} - -/* - * Get the perf_counter_context for a task and lock it. - * This has to cope with with the fact that until it is locked, - * the context could get moved to another task. - */ -static struct perf_counter_context * -perf_lock_task_context(struct task_struct *task, unsigned long *flags) -{ - struct perf_counter_context *ctx; - - rcu_read_lock(); - retry: - ctx = rcu_dereference(task->perf_counter_ctxp); - if (ctx) { - /* - * If this context is a clone of another, it might - * get swapped for another underneath us by - * perf_counter_task_sched_out, though the - * rcu_read_lock() protects us from any context - * getting freed. Lock the context and check if it - * got swapped before we could get the lock, and retry - * if so. If we locked the right context, then it - * can't get swapped on us any more. - */ - spin_lock_irqsave(&ctx->lock, *flags); - if (ctx != rcu_dereference(task->perf_counter_ctxp)) { - spin_unlock_irqrestore(&ctx->lock, *flags); - goto retry; - } - - if (!atomic_inc_not_zero(&ctx->refcount)) { - spin_unlock_irqrestore(&ctx->lock, *flags); - ctx = NULL; - } - } - rcu_read_unlock(); - return ctx; -} - -/* - * Get the context for a task and increment its pin_count so it - * can't get swapped to another task. This also increments its - * reference count so that the context can't get freed. - */ -static struct perf_counter_context *perf_pin_task_context(struct task_struct *task) -{ - struct perf_counter_context *ctx; - unsigned long flags; - - ctx = perf_lock_task_context(task, &flags); - if (ctx) { - ++ctx->pin_count; - spin_unlock_irqrestore(&ctx->lock, flags); - } - return ctx; -} - -static void perf_unpin_context(struct perf_counter_context *ctx) -{ - unsigned long flags; - - spin_lock_irqsave(&ctx->lock, flags); - --ctx->pin_count; - spin_unlock_irqrestore(&ctx->lock, flags); - put_ctx(ctx); -} - -/* - * Add a counter from the lists for its context. - * Must be called with ctx->mutex and ctx->lock held. - */ -static void -list_add_counter(struct perf_counter *counter, struct perf_counter_context *ctx) -{ - struct perf_counter *group_leader = counter->group_leader; - - /* - * Depending on whether it is a standalone or sibling counter, - * add it straight to the context's counter list, or to the group - * leader's sibling list: - */ - if (group_leader == counter) - list_add_tail(&counter->group_entry, &ctx->group_list); - else { - list_add_tail(&counter->group_entry, &group_leader->sibling_list); - group_leader->nr_siblings++; - } - - list_add_rcu(&counter->event_entry, &ctx->event_list); - ctx->nr_counters++; - if (counter->attr.inherit_stat) - ctx->nr_stat++; -} - -/* - * Remove a counter from the lists for its context. - * Must be called with ctx->mutex and ctx->lock held. - */ -static void -list_del_counter(struct perf_counter *counter, struct perf_counter_context *ctx) -{ - struct perf_counter *sibling, *tmp; - - if (list_empty(&counter->group_entry)) - return; - ctx->nr_counters--; - if (counter->attr.inherit_stat) - ctx->nr_stat--; - - list_del_init(&counter->group_entry); - list_del_rcu(&counter->event_entry); - - if (counter->group_leader != counter) - counter->group_leader->nr_siblings--; - - /* - * If this was a group counter with sibling counters then - * upgrade the siblings to singleton counters by adding them - * to the context list directly: - */ - list_for_each_entry_safe(sibling, tmp, &counter->sibling_list, group_entry) { - - list_move_tail(&sibling->group_entry, &ctx->group_list); - sibling->group_leader = sibling; - } -} - -static void -counter_sched_out(struct perf_counter *counter, - struct perf_cpu_context *cpuctx, - struct perf_counter_context *ctx) -{ - if (counter->state != PERF_COUNTER_STATE_ACTIVE) - return; - - counter->state = PERF_COUNTER_STATE_INACTIVE; - if (counter->pending_disable) { - counter->pending_disable = 0; - counter->state = PERF_COUNTER_STATE_OFF; - } - counter->tstamp_stopped = ctx->time; - counter->pmu->disable(counter); - counter->oncpu = -1; - - if (!is_software_counter(counter)) - cpuctx->active_oncpu--; - ctx->nr_active--; - if (counter->attr.exclusive || !cpuctx->active_oncpu) - cpuctx->exclusive = 0; -} - -static void -group_sched_out(struct perf_counter *group_counter, - struct perf_cpu_context *cpuctx, - struct perf_counter_context *ctx) -{ - struct perf_counter *counter; - - if (group_counter->state != PERF_COUNTER_STATE_ACTIVE) - return; - - counter_sched_out(group_counter, cpuctx, ctx); - - /* - * Schedule out siblings (if any): - */ - list_for_each_entry(counter, &group_counter->sibling_list, group_entry) - counter_sched_out(counter, cpuctx, ctx); - - if (group_counter->attr.exclusive) - cpuctx->exclusive = 0; -} - -/* - * Cross CPU call to remove a performance counter - * - * We disable the counter on the hardware level first. After that we - * remove it from the context list. - */ -static void __perf_counter_remove_from_context(void *info) -{ - struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); - struct perf_counter *counter = info; - struct perf_counter_context *ctx = counter->ctx; - - /* - * If this is a task context, we need to check whether it is - * the current task context of this cpu. If not it has been - * scheduled out before the smp call arrived. - */ - if (ctx->task && cpuctx->task_ctx != ctx) - return; - - spin_lock(&ctx->lock); - /* - * Protect the list operation against NMI by disabling the - * counters on a global level. - */ - perf_disable(); - - counter_sched_out(counter, cpuctx, ctx); - - list_del_counter(counter, ctx); - - if (!ctx->task) { - /* - * Allow more per task counters with respect to the - * reservation: - */ - cpuctx->max_pertask = - min(perf_max_counters - ctx->nr_counters, - perf_max_counters - perf_reserved_percpu); - } - - perf_enable(); - spin_unlock(&ctx->lock); -} - - -/* - * Remove the counter from a task's (or a CPU's) list of counters. - * - * Must be called with ctx->mutex held. - * - * CPU counters are removed with a smp call. For task counters we only - * call when the task is on a CPU. - * - * If counter->ctx is a cloned context, callers must make sure that - * every task struct that counter->ctx->task could possibly point to - * remains valid. This is OK when called from perf_release since - * that only calls us on the top-level context, which can't be a clone. - * When called from perf_counter_exit_task, it's OK because the - * context has been detached from its task. - */ -static void perf_counter_remove_from_context(struct perf_counter *counter) -{ - struct perf_counter_context *ctx = counter->ctx; - struct task_struct *task = ctx->task; - - if (!task) { - /* - * Per cpu counters are removed via an smp call and - * the removal is always sucessful. - */ - smp_call_function_single(counter->cpu, - __perf_counter_remove_from_context, - counter, 1); - return; - } - -retry: - task_oncpu_function_call(task, __perf_counter_remove_from_context, - counter); - - spin_lock_irq(&ctx->lock); - /* - * If the context is active we need to retry the smp call. - */ - if (ctx->nr_active && !list_empty(&counter->group_entry)) { - spin_unlock_irq(&ctx->lock); - goto retry; - } - - /* - * The lock prevents that this context is scheduled in so we - * can remove the counter safely, if the call above did not - * succeed. - */ - if (!list_empty(&counter->group_entry)) { - list_del_counter(counter, ctx); - } - spin_unlock_irq(&ctx->lock); -} - -static inline u64 perf_clock(void) -{ - return cpu_clock(smp_processor_id()); -} - -/* - * Update the record of the current time in a context. - */ -static void update_context_time(struct perf_counter_context *ctx) -{ - u64 now = perf_clock(); - - ctx->time += now - ctx->timestamp; - ctx->timestamp = now; -} - -/* - * Update the total_time_enabled and total_time_running fields for a counter. - */ -static void update_counter_times(struct perf_counter *counter) -{ - struct perf_counter_context *ctx = counter->ctx; - u64 run_end; - - if (counter->state < PERF_COUNTER_STATE_INACTIVE || - counter->group_leader->state < PERF_COUNTER_STATE_INACTIVE) - return; - - counter->total_time_enabled = ctx->time - counter->tstamp_enabled; - - if (counter->state == PERF_COUNTER_STATE_INACTIVE) - run_end = counter->tstamp_stopped; - else - run_end = ctx->time; - - counter->total_time_running = run_end - counter->tstamp_running; -} - -/* - * Update total_time_enabled and total_time_running for all counters in a group. - */ -static void update_group_times(struct perf_counter *leader) -{ - struct perf_counter *counter; - - update_counter_times(leader); - list_for_each_entry(counter, &leader->sibling_list, group_entry) - update_counter_times(counter); -} - -/* - * Cross CPU call to disable a performance counter - */ -static void __perf_counter_disable(void *info) -{ - struct perf_counter *counter = info; - struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); - struct perf_counter_context *ctx = counter->ctx; - - /* - * If this is a per-task counter, need to check whether this - * counter's task is the current task on this cpu. - */ - if (ctx->task && cpuctx->task_ctx != ctx) - return; - - spin_lock(&ctx->lock); - - /* - * If the counter is on, turn it off. - * If it is in error state, leave it in error state. - */ - if (counter->state >= PERF_COUNTER_STATE_INACTIVE) { - update_context_time(ctx); - update_group_times(counter); - if (counter == counter->group_leader) - group_sched_out(counter, cpuctx, ctx); - else - counter_sched_out(counter, cpuctx, ctx); - counter->state = PERF_COUNTER_STATE_OFF; - } - - spin_unlock(&ctx->lock); -} - -/* - * Disable a counter. - * - * If counter->ctx is a cloned context, callers must make sure that - * every task struct that counter->ctx->task could possibly point to - * remains valid. This condition is satisifed when called through - * perf_counter_for_each_child or perf_counter_for_each because they - * hold the top-level counter's child_mutex, so any descendant that - * goes to exit will block in sync_child_counter. - * When called from perf_pending_counter it's OK because counter->ctx - * is the current context on this CPU and preemption is disabled, - * hence we can't get into perf_counter_task_sched_out for this context. - */ -static void perf_counter_disable(struct perf_counter *counter) -{ - struct perf_counter_context *ctx = counter->ctx; - struct task_struct *task = ctx->task; - - if (!task) { - /* - * Disable the counter on the cpu that it's on - */ - smp_call_function_single(counter->cpu, __perf_counter_disable, - counter, 1); - return; - } - - retry: - task_oncpu_function_call(task, __perf_counter_disable, counter); - - spin_lock_irq(&ctx->lock); - /* - * If the counter is still active, we need to retry the cross-call. - */ - if (counter->state == PERF_COUNTER_STATE_ACTIVE) { - spin_unlock_irq(&ctx->lock); - goto retry; - } - - /* - * Since we have the lock this context can't be scheduled - * in, so we can change the state safely. - */ - if (counter->state == PERF_COUNTER_STATE_INACTIVE) { - update_group_times(counter); - counter->state = PERF_COUNTER_STATE_OFF; - } - - spin_unlock_irq(&ctx->lock); -} - -static int -counter_sched_in(struct perf_counter *counter, - struct perf_cpu_context *cpuctx, - struct perf_counter_context *ctx, - int cpu) -{ - if (counter->state <= PERF_COUNTER_STATE_OFF) - return 0; - - counter->state = PERF_COUNTER_STATE_ACTIVE; - counter->oncpu = cpu; /* TODO: put 'cpu' into cpuctx->cpu */ - /* - * The new state must be visible before we turn it on in the hardware: - */ - smp_wmb(); - - if (counter->pmu->enable(counter)) { - counter->state = PERF_COUNTER_STATE_INACTIVE; - counter->oncpu = -1; - return -EAGAIN; - } - - counter->tstamp_running += ctx->time - counter->tstamp_stopped; - - if (!is_software_counter(counter)) - cpuctx->active_oncpu++; - ctx->nr_active++; - - if (counter->attr.exclusive) - cpuctx->exclusive = 1; - - return 0; -} - -static int -group_sched_in(struct perf_counter *group_counter, - struct perf_cpu_context *cpuctx, - struct perf_counter_context *ctx, - int cpu) -{ - struct perf_counter *counter, *partial_group; - int ret; - - if (group_counter->state == PERF_COUNTER_STATE_OFF) - return 0; - - ret = hw_perf_group_sched_in(group_counter, cpuctx, ctx, cpu); - if (ret) - return ret < 0 ? ret : 0; - - if (counter_sched_in(group_counter, cpuctx, ctx, cpu)) - return -EAGAIN; - - /* - * Schedule in siblings as one group (if any): - */ - list_for_each_entry(counter, &group_counter->sibling_list, group_entry) { - if (counter_sched_in(counter, cpuctx, ctx, cpu)) { - partial_group = counter; - goto group_error; - } - } - - return 0; - -group_error: - /* - * Groups can be scheduled in as one unit only, so undo any - * partial group before returning: - */ - list_for_each_entry(counter, &group_counter->sibling_list, group_entry) { - if (counter == partial_group) - break; - counter_sched_out(counter, cpuctx, ctx); - } - counter_sched_out(group_counter, cpuctx, ctx); - - return -EAGAIN; -} - -/* - * Return 1 for a group consisting entirely of software counters, - * 0 if the group contains any hardware counters. - */ -static int is_software_only_group(struct perf_counter *leader) -{ - struct perf_counter *counter; - - if (!is_software_counter(leader)) - return 0; - - list_for_each_entry(counter, &leader->sibling_list, group_entry) - if (!is_software_counter(counter)) - return 0; - - return 1; -} - -/* - * Work out whether we can put this counter group on the CPU now. - */ -static int group_can_go_on(struct perf_counter *counter, - struct perf_cpu_context *cpuctx, - int can_add_hw) -{ - /* - * Groups consisting entirely of software counters can always go on. - */ - if (is_software_only_group(counter)) - return 1; - /* - * If an exclusive group is already on, no other hardware - * counters can go on. - */ - if (cpuctx->exclusive) - return 0; - /* - * If this group is exclusive and there are already - * counters on the CPU, it can't go on. - */ - if (counter->attr.exclusive && cpuctx->active_oncpu) - return 0; - /* - * Otherwise, try to add it if all previous groups were able - * to go on. - */ - return can_add_hw; -} - -static void add_counter_to_ctx(struct perf_counter *counter, - struct perf_counter_context *ctx) -{ - list_add_counter(counter, ctx); - counter->tstamp_enabled = ctx->time; - counter->tstamp_running = ctx->time; - counter->tstamp_stopped = ctx->time; -} - -/* - * Cross CPU call to install and enable a performance counter - * - * Must be called with ctx->mutex held - */ -static void __perf_install_in_context(void *info) -{ - struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); - struct perf_counter *counter = info; - struct perf_counter_context *ctx = counter->ctx; - struct perf_counter *leader = counter->group_leader; - int cpu = smp_processor_id(); - int err; - - /* - * If this is a task context, we need to check whether it is - * the current task context of this cpu. If not it has been - * scheduled out before the smp call arrived. - * Or possibly this is the right context but it isn't - * on this cpu because it had no counters. - */ - if (ctx->task && cpuctx->task_ctx != ctx) { - if (cpuctx->task_ctx || ctx->task != current) - return; - cpuctx->task_ctx = ctx; - } - - spin_lock(&ctx->lock); - ctx->is_active = 1; - update_context_time(ctx); - - /* - * Protect the list operation against NMI by disabling the - * counters on a global level. NOP for non NMI based counters. - */ - perf_disable(); - - add_counter_to_ctx(counter, ctx); - - /* - * Don't put the counter on if it is disabled or if - * it is in a group and the group isn't on. - */ - if (counter->state != PERF_COUNTER_STATE_INACTIVE || - (leader != counter && leader->state != PERF_COUNTER_STATE_ACTIVE)) - goto unlock; - - /* - * An exclusive counter can't go on if there are already active - * hardware counters, and no hardware counter can go on if there - * is already an exclusive counter on. - */ - if (!group_can_go_on(counter, cpuctx, 1)) - err = -EEXIST; - else - err = counter_sched_in(counter, cpuctx, ctx, cpu); - - if (err) { - /* - * This counter couldn't go on. If it is in a group - * then we have to pull the whole group off. - * If the counter group is pinned then put it in error state. - */ - if (leader != counter) - group_sched_out(leader, cpuctx, ctx); - if (leader->attr.pinned) { - update_group_times(leader); - leader->state = PERF_COUNTER_STATE_ERROR; - } - } - - if (!err && !ctx->task && cpuctx->max_pertask) - cpuctx->max_pertask--; - - unlock: - perf_enable(); - - spin_unlock(&ctx->lock); -} - -/* - * Attach a performance counter to a context - * - * First we add the counter to the list with the hardware enable bit - * in counter->hw_config cleared. - * - * If the counter is attached to a task which is on a CPU we use a smp - * call to enable it in the task context. The task might have been - * scheduled away, but we check this in the smp call again. - * - * Must be called with ctx->mutex held. - */ -static void -perf_install_in_context(struct perf_counter_context *ctx, - struct perf_counter *counter, - int cpu) -{ - struct task_struct *task = ctx->task; - - if (!task) { - /* - * Per cpu counters are installed via an smp call and - * the install is always sucessful. - */ - smp_call_function_single(cpu, __perf_install_in_context, - counter, 1); - return; - } - -retry: - task_oncpu_function_call(task, __perf_install_in_context, - counter); - - spin_lock_irq(&ctx->lock); - /* - * we need to retry the smp call. - */ - if (ctx->is_active && list_empty(&counter->group_entry)) { - spin_unlock_irq(&ctx->lock); - goto retry; - } - - /* - * The lock prevents that this context is scheduled in so we - * can add the counter safely, if it the call above did not - * succeed. - */ - if (list_empty(&counter->group_entry)) - add_counter_to_ctx(counter, ctx); - spin_unlock_irq(&ctx->lock); -} - -/* - * Put a counter into inactive state and update time fields. - * Enabling the leader of a group effectively enables all - * the group members that aren't explicitly disabled, so we - * have to update their ->tstamp_enabled also. - * Note: this works for group members as well as group leaders - * since the non-leader members' sibling_lists will be empty. - */ -static void __perf_counter_mark_enabled(struct perf_counter *counter, - struct perf_counter_context *ctx) -{ - struct perf_counter *sub; - - counter->state = PERF_COUNTER_STATE_INACTIVE; - counter->tstamp_enabled = ctx->time - counter->total_time_enabled; - list_for_each_entry(sub, &counter->sibling_list, group_entry) - if (sub->state >= PERF_COUNTER_STATE_INACTIVE) - sub->tstamp_enabled = - ctx->time - sub->total_time_enabled; -} - -/* - * Cross CPU call to enable a performance counter - */ -static void __perf_counter_enable(void *info) -{ - struct perf_counter *counter = info; - struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); - struct perf_counter_context *ctx = counter->ctx; - struct perf_counter *leader = counter->group_leader; - int err; - - /* - * If this is a per-task counter, need to check whether this - * counter's task is the current task on this cpu. - */ - if (ctx->task && cpuctx->task_ctx != ctx) { - if (cpuctx->task_ctx || ctx->task != current) - return; - cpuctx->task_ctx = ctx; - } - - spin_lock(&ctx->lock); - ctx->is_active = 1; - update_context_time(ctx); - - if (counter->state >= PERF_COUNTER_STATE_INACTIVE) - goto unlock; - __perf_counter_mark_enabled(counter, ctx); - - /* - * If the counter is in a group and isn't the group leader, - * then don't put it on unless the group is on. - */ - if (leader != counter && leader->state != PERF_COUNTER_STATE_ACTIVE) - goto unlock; - - if (!group_can_go_on(counter, cpuctx, 1)) { - err = -EEXIST; - } else { - perf_disable(); - if (counter == leader) - err = group_sched_in(counter, cpuctx, ctx, - smp_processor_id()); - else - err = counter_sched_in(counter, cpuctx, ctx, - smp_processor_id()); - perf_enable(); - } - - if (err) { - /* - * If this counter can't go on and it's part of a - * group, then the whole group has to come off. - */ - if (leader != counter) - group_sched_out(leader, cpuctx, ctx); - if (leader->attr.pinned) { - update_group_times(leader); - leader->state = PERF_COUNTER_STATE_ERROR; - } - } - - unlock: - spin_unlock(&ctx->lock); -} - -/* - * Enable a counter. - * - * If counter->ctx is a cloned context, callers must make sure that - * every task struct that counter->ctx->task could possibly point to - * remains valid. This condition is satisfied when called through - * perf_counter_for_each_child or perf_counter_for_each as described - * for perf_counter_disable. - */ -static void perf_counter_enable(struct perf_counter *counter) -{ - struct perf_counter_context *ctx = counter->ctx; - struct task_struct *task = ctx->task; - - if (!task) { - /* - * Enable the counter on the cpu that it's on - */ - smp_call_function_single(counter->cpu, __perf_counter_enable, - counter, 1); - return; - } - - spin_lock_irq(&ctx->lock); - if (counter->state >= PERF_COUNTER_STATE_INACTIVE) - goto out; - - /* - * If the counter is in error state, clear that first. - * That way, if we see the counter in error state below, we - * know that it has gone back into error state, as distinct - * from the task having been scheduled away before the - * cross-call arrived. - */ - if (counter->state == PERF_COUNTER_STATE_ERROR) - counter->state = PERF_COUNTER_STATE_OFF; - - retry: - spin_unlock_irq(&ctx->lock); - task_oncpu_function_call(task, __perf_counter_enable, counter); - - spin_lock_irq(&ctx->lock); - - /* - * If the context is active and the counter is still off, - * we need to retry the cross-call. - */ - if (ctx->is_active && counter->state == PERF_COUNTER_STATE_OFF) - goto retry; - - /* - * Since we have the lock this context can't be scheduled - * in, so we can change the state safely. - */ - if (counter->state == PERF_COUNTER_STATE_OFF) - __perf_counter_mark_enabled(counter, ctx); - - out: - spin_unlock_irq(&ctx->lock); -} - -static int perf_counter_refresh(struct perf_counter *counter, int refresh) -{ - /* - * not supported on inherited counters - */ - if (counter->attr.inherit) - return -EINVAL; - - atomic_add(refresh, &counter->event_limit); - perf_counter_enable(counter); - - return 0; -} - -void __perf_counter_sched_out(struct perf_counter_context *ctx, - struct perf_cpu_context *cpuctx) -{ - struct perf_counter *counter; - - spin_lock(&ctx->lock); - ctx->is_active = 0; - if (likely(!ctx->nr_counters)) - goto out; - update_context_time(ctx); - - perf_disable(); - if (ctx->nr_active) { - list_for_each_entry(counter, &ctx->group_list, group_entry) { - if (counter != counter->group_leader) - counter_sched_out(counter, cpuctx, ctx); - else - group_sched_out(counter, cpuctx, ctx); - } - } - perf_enable(); - out: - spin_unlock(&ctx->lock); -} - -/* - * Test whether two contexts are equivalent, i.e. whether they - * have both been cloned from the same version of the same context - * and they both have the same number of enabled counters. - * If the number of enabled counters is the same, then the set - * of enabled counters should be the same, because these are both - * inherited contexts, therefore we can't access individual counters - * in them directly with an fd; we can only enable/disable all - * counters via prctl, or enable/disable all counters in a family - * via ioctl, which will have the same effect on both contexts. - */ -static int context_equiv(struct perf_counter_context *ctx1, - struct perf_counter_context *ctx2) -{ - return ctx1->parent_ctx && ctx1->parent_ctx == ctx2->parent_ctx - && ctx1->parent_gen == ctx2->parent_gen - && !ctx1->pin_count && !ctx2->pin_count; -} - -static void __perf_counter_read(void *counter); - -static void __perf_counter_sync_stat(struct perf_counter *counter, - struct perf_counter *next_counter) -{ - u64 value; - - if (!counter->attr.inherit_stat) - return; - - /* - * Update the counter value, we cannot use perf_counter_read() - * because we're in the middle of a context switch and have IRQs - * disabled, which upsets smp_call_function_single(), however - * we know the counter must be on the current CPU, therefore we - * don't need to use it. - */ - switch (counter->state) { - case PERF_COUNTER_STATE_ACTIVE: - __perf_counter_read(counter); - break; - - case PERF_COUNTER_STATE_INACTIVE: - update_counter_times(counter); - break; - - default: - break; - } - - /* - * In order to keep per-task stats reliable we need to flip the counter - * values when we flip the contexts. - */ - value = atomic64_read(&next_counter->count); - value = atomic64_xchg(&counter->count, value); - atomic64_set(&next_counter->count, value); - - swap(counter->total_time_enabled, next_counter->total_time_enabled); - swap(counter->total_time_running, next_counter->total_time_running); - - /* - * Since we swizzled the values, update the user visible data too. - */ - perf_counter_update_userpage(counter); - perf_counter_update_userpage(next_counter); -} - -#define list_next_entry(pos, member) \ - list_entry(pos->member.next, typeof(*pos), member) - -static void perf_counter_sync_stat(struct perf_counter_context *ctx, - struct perf_counter_context *next_ctx) -{ - struct perf_counter *counter, *next_counter; - - if (!ctx->nr_stat) - return; - - counter = list_first_entry(&ctx->event_list, - struct perf_counter, event_entry); - - next_counter = list_first_entry(&next_ctx->event_list, - struct perf_counter, event_entry); - - while (&counter->event_entry != &ctx->event_list && - &next_counter->event_entry != &next_ctx->event_list) { - - __perf_counter_sync_stat(counter, next_counter); - - counter = list_next_entry(counter, event_entry); - next_counter = list_next_entry(next_counter, event_entry); - } -} - -/* - * Called from scheduler to remove the counters of the current task, - * with interrupts disabled. - * - * We stop each counter and update the counter value in counter->count. - * - * This does not protect us against NMI, but disable() - * sets the disabled bit in the control field of counter _before_ - * accessing the counter control register. If a NMI hits, then it will - * not restart the counter. - */ -void perf_counter_task_sched_out(struct task_struct *task, - struct task_struct *next, int cpu) -{ - struct perf_cpu_context *cpuctx = &per_cpu(perf_cpu_context, cpu); - struct perf_counter_context *ctx = task->perf_counter_ctxp; - struct perf_counter_context *next_ctx; - struct perf_counter_context *parent; - struct pt_regs *regs; - int do_switch = 1; - - regs = task_pt_regs(task); - perf_swcounter_event(PERF_COUNT_SW_CONTEXT_SWITCHES, 1, 1, regs, 0); - - if (likely(!ctx || !cpuctx->task_ctx)) - return; - - update_context_time(ctx); - - rcu_read_lock(); - parent = rcu_dereference(ctx->parent_ctx); - next_ctx = next->perf_counter_ctxp; - if (parent && next_ctx && - rcu_dereference(next_ctx->parent_ctx) == parent) { - /* - * Looks like the two contexts are clones, so we might be - * able to optimize the context switch. We lock both - * contexts and check that they are clones under the - * lock (including re-checking that neither has been - * uncloned in the meantime). It doesn't matter which - * order we take the locks because no other cpu could - * be trying to lock both of these tasks. - */ - spin_lock(&ctx->lock); - spin_lock_nested(&next_ctx->lock, SINGLE_DEPTH_NESTING); - if (context_equiv(ctx, next_ctx)) { - /* - * XXX do we need a memory barrier of sorts - * wrt to rcu_dereference() of perf_counter_ctxp - */ - task->perf_counter_ctxp = next_ctx; - next->perf_counter_ctxp = ctx; - ctx->task = next; - next_ctx->task = task; - do_switch = 0; - - perf_counter_sync_stat(ctx, next_ctx); - } - spin_unlock(&next_ctx->lock); - spin_unlock(&ctx->lock); - } - rcu_read_unlock(); - - if (do_switch) { - __perf_counter_sched_out(ctx, cpuctx); - cpuctx->task_ctx = NULL; - } -} - -/* - * Called with IRQs disabled - */ -static void __perf_counter_task_sched_out(struct perf_counter_context *ctx) -{ - struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); - - if (!cpuctx->task_ctx) - return; - - if (WARN_ON_ONCE(ctx != cpuctx->task_ctx)) - return; - - __perf_counter_sched_out(ctx, cpuctx); - cpuctx->task_ctx = NULL; -} - -/* - * Called with IRQs disabled - */ -static void perf_counter_cpu_sched_out(struct perf_cpu_context *cpuctx) -{ - __perf_counter_sched_out(&cpuctx->ctx, cpuctx); -} - -static void -__perf_counter_sched_in(struct perf_counter_context *ctx, - struct perf_cpu_context *cpuctx, int cpu) -{ - struct perf_counter *counter; - int can_add_hw = 1; - - spin_lock(&ctx->lock); - ctx->is_active = 1; - if (likely(!ctx->nr_counters)) - goto out; - - ctx->timestamp = perf_clock(); - - perf_disable(); - - /* - * First go through the list and put on any pinned groups - * in order to give them the best chance of going on. - */ - list_for_each_entry(counter, &ctx->group_list, group_entry) { - if (counter->state <= PERF_COUNTER_STATE_OFF || - !counter->attr.pinned) - continue; - if (counter->cpu != -1 && counter->cpu != cpu) - continue; - - if (counter != counter->group_leader) - counter_sched_in(counter, cpuctx, ctx, cpu); - else { - if (group_can_go_on(counter, cpuctx, 1)) - group_sched_in(counter, cpuctx, ctx, cpu); - } - - /* - * If this pinned group hasn't been scheduled, - * put it in error state. - */ - if (counter->state == PERF_COUNTER_STATE_INACTIVE) { - update_group_times(counter); - counter->state = PERF_COUNTER_STATE_ERROR; - } - } - - list_for_each_entry(counter, &ctx->group_list, group_entry) { - /* - * Ignore counters in OFF or ERROR state, and - * ignore pinned counters since we did them already. - */ - if (counter->state <= PERF_COUNTER_STATE_OFF || - counter->attr.pinned) - continue; - - /* - * Listen to the 'cpu' scheduling filter constraint - * of counters: - */ - if (counter->cpu != -1 && counter->cpu != cpu) - continue; - - if (counter != counter->group_leader) { - if (counter_sched_in(counter, cpuctx, ctx, cpu)) - can_add_hw = 0; - } else { - if (group_can_go_on(counter, cpuctx, can_add_hw)) { - if (group_sched_in(counter, cpuctx, ctx, cpu)) - can_add_hw = 0; - } - } - } - perf_enable(); - out: - spin_unlock(&ctx->lock); -} - -/* - * Called from scheduler to add the counters of the current task - * with interrupts disabled. - * - * We restore the counter value and then enable it. - * - * This does not protect us against NMI, but enable() - * sets the enabled bit in the control field of counter _before_ - * accessing the counter control register. If a NMI hits, then it will - * keep the counter running. - */ -void perf_counter_task_sched_in(struct task_struct *task, int cpu) -{ - struct perf_cpu_context *cpuctx = &per_cpu(perf_cpu_context, cpu); - struct perf_counter_context *ctx = task->perf_counter_ctxp; - - if (likely(!ctx)) - return; - if (cpuctx->task_ctx == ctx) - return; - __perf_counter_sched_in(ctx, cpuctx, cpu); - cpuctx->task_ctx = ctx; -} - -static void perf_counter_cpu_sched_in(struct perf_cpu_context *cpuctx, int cpu) -{ - struct perf_counter_context *ctx = &cpuctx->ctx; - - __perf_counter_sched_in(ctx, cpuctx, cpu); -} - -#define MAX_INTERRUPTS (~0ULL) - -static void perf_log_throttle(struct perf_counter *counter, int enable); - -static void perf_adjust_period(struct perf_counter *counter, u64 events) -{ - struct hw_perf_counter *hwc = &counter->hw; - u64 period, sample_period; - s64 delta; - - events *= hwc->sample_period; - period = div64_u64(events, counter->attr.sample_freq); - - delta = (s64)(period - hwc->sample_period); - delta = (delta + 7) / 8; /* low pass filter */ - - sample_period = hwc->sample_period + delta; - - if (!sample_period) - sample_period = 1; - - hwc->sample_period = sample_period; -} - -static void perf_ctx_adjust_freq(struct perf_counter_context *ctx) -{ - struct perf_counter *counter; - struct hw_perf_counter *hwc; - u64 interrupts, freq; - - spin_lock(&ctx->lock); - list_for_each_entry(counter, &ctx->group_list, group_entry) { - if (counter->state != PERF_COUNTER_STATE_ACTIVE) - continue; - - hwc = &counter->hw; - - interrupts = hwc->interrupts; - hwc->interrupts = 0; - - /* - * unthrottle counters on the tick - */ - if (interrupts == MAX_INTERRUPTS) { - perf_log_throttle(counter, 1); - counter->pmu->unthrottle(counter); - interrupts = 2*sysctl_perf_counter_sample_rate/HZ; - } - - if (!counter->attr.freq || !counter->attr.sample_freq) - continue; - - /* - * if the specified freq < HZ then we need to skip ticks - */ - if (counter->attr.sample_freq < HZ) { - freq = counter->attr.sample_freq; - - hwc->freq_count += freq; - hwc->freq_interrupts += interrupts; - - if (hwc->freq_count < HZ) - continue; - - interrupts = hwc->freq_interrupts; - hwc->freq_interrupts = 0; - hwc->freq_count -= HZ; - } else - freq = HZ; - - perf_adjust_period(counter, freq * interrupts); - - /* - * In order to avoid being stalled by an (accidental) huge - * sample period, force reset the sample period if we didn't - * get any events in this freq period. - */ - if (!interrupts) { - perf_disable(); - counter->pmu->disable(counter); - atomic64_set(&hwc->period_left, 0); - counter->pmu->enable(counter); - perf_enable(); - } - } - spin_unlock(&ctx->lock); -} - -/* - * Round-robin a context's counters: - */ -static void rotate_ctx(struct perf_counter_context *ctx) -{ - struct perf_counter *counter; - - if (!ctx->nr_counters) - return; - - spin_lock(&ctx->lock); - /* - * Rotate the first entry last (works just fine for group counters too): - */ - perf_disable(); - list_for_each_entry(counter, &ctx->group_list, group_entry) { - list_move_tail(&counter->group_entry, &ctx->group_list); - break; - } - perf_enable(); - - spin_unlock(&ctx->lock); -} - -void perf_counter_task_tick(struct task_struct *curr, int cpu) -{ - struct perf_cpu_context *cpuctx; - struct perf_counter_context *ctx; - - if (!atomic_read(&nr_counters)) - return; - - cpuctx = &per_cpu(perf_cpu_context, cpu); - ctx = curr->perf_counter_ctxp; - - perf_ctx_adjust_freq(&cpuctx->ctx); - if (ctx) - perf_ctx_adjust_freq(ctx); - - perf_counter_cpu_sched_out(cpuctx); - if (ctx) - __perf_counter_task_sched_out(ctx); - - rotate_ctx(&cpuctx->ctx); - if (ctx) - rotate_ctx(ctx); - - perf_counter_cpu_sched_in(cpuctx, cpu); - if (ctx) - perf_counter_task_sched_in(curr, cpu); -} - -/* - * Enable all of a task's counters that have been marked enable-on-exec. - * This expects task == current. - */ -static void perf_counter_enable_on_exec(struct task_struct *task) -{ - struct perf_counter_context *ctx; - struct perf_counter *counter; - unsigned long flags; - int enabled = 0; - - local_irq_save(flags); - ctx = task->perf_counter_ctxp; - if (!ctx || !ctx->nr_counters) - goto out; - - __perf_counter_task_sched_out(ctx); - - spin_lock(&ctx->lock); - - list_for_each_entry(counter, &ctx->group_list, group_entry) { - if (!counter->attr.enable_on_exec) - continue; - counter->attr.enable_on_exec = 0; - if (counter->state >= PERF_COUNTER_STATE_INACTIVE) - continue; - __perf_counter_mark_enabled(counter, ctx); - enabled = 1; - } - - /* - * Unclone this context if we enabled any counter. - */ - if (enabled) - unclone_ctx(ctx); - - spin_unlock(&ctx->lock); - - perf_counter_task_sched_in(task, smp_processor_id()); - out: - local_irq_restore(flags); -} - -/* - * Cross CPU call to read the hardware counter - */ -static void __perf_counter_read(void *info) -{ - struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); - struct perf_counter *counter = info; - struct perf_counter_context *ctx = counter->ctx; - unsigned long flags; - - /* - * If this is a task context, we need to check whether it is - * the current task context of this cpu. If not it has been - * scheduled out before the smp call arrived. In that case - * counter->count would have been updated to a recent sample - * when the counter was scheduled out. - */ - if (ctx->task && cpuctx->task_ctx != ctx) - return; - - local_irq_save(flags); - if (ctx->is_active) - update_context_time(ctx); - counter->pmu->read(counter); - update_counter_times(counter); - local_irq_restore(flags); -} - -static u64 perf_counter_read(struct perf_counter *counter) -{ - /* - * If counter is enabled and currently active on a CPU, update the - * value in the counter structure: - */ - if (counter->state == PERF_COUNTER_STATE_ACTIVE) { - smp_call_function_single(counter->oncpu, - __perf_counter_read, counter, 1); - } else if (counter->state == PERF_COUNTER_STATE_INACTIVE) { - update_counter_times(counter); - } - - return atomic64_read(&counter->count); -} - -/* - * Initialize the perf_counter context in a task_struct: - */ -static void -__perf_counter_init_context(struct perf_counter_context *ctx, - struct task_struct *task) -{ - memset(ctx, 0, sizeof(*ctx)); - spin_lock_init(&ctx->lock); - mutex_init(&ctx->mutex); - INIT_LIST_HEAD(&ctx->group_list); - INIT_LIST_HEAD(&ctx->event_list); - atomic_set(&ctx->refcount, 1); - ctx->task = task; -} - -static struct perf_counter_context *find_get_context(pid_t pid, int cpu) -{ - struct perf_counter_context *ctx; - struct perf_cpu_context *cpuctx; - struct task_struct *task; - unsigned long flags; - int err; - - /* - * If cpu is not a wildcard then this is a percpu counter: - */ - if (cpu != -1) { - /* Must be root to operate on a CPU counter: */ - if (perf_paranoid_cpu() && !capable(CAP_SYS_ADMIN)) - return ERR_PTR(-EACCES); - - if (cpu < 0 || cpu > num_possible_cpus()) - return ERR_PTR(-EINVAL); - - /* - * We could be clever and allow to attach a counter to an - * offline CPU and activate it when the CPU comes up, but - * that's for later. - */ - if (!cpu_isset(cpu, cpu_online_map)) - return ERR_PTR(-ENODEV); - - cpuctx = &per_cpu(perf_cpu_context, cpu); - ctx = &cpuctx->ctx; - get_ctx(ctx); - - return ctx; - } - - rcu_read_lock(); - if (!pid) - task = current; - else - task = find_task_by_vpid(pid); - if (task) - get_task_struct(task); - rcu_read_unlock(); - - if (!task) - return ERR_PTR(-ESRCH); - - /* - * Can't attach counters to a dying task. - */ - err = -ESRCH; - if (task->flags & PF_EXITING) - goto errout; - - /* Reuse ptrace permission checks for now. */ - err = -EACCES; - if (!ptrace_may_access(task, PTRACE_MODE_READ)) - goto errout; - - retry: - ctx = perf_lock_task_context(task, &flags); - if (ctx) { - unclone_ctx(ctx); - spin_unlock_irqrestore(&ctx->lock, flags); - } - - if (!ctx) { - ctx = kmalloc(sizeof(struct perf_counter_context), GFP_KERNEL); - err = -ENOMEM; - if (!ctx) - goto errout; - __perf_counter_init_context(ctx, task); - get_ctx(ctx); - if (cmpxchg(&task->perf_counter_ctxp, NULL, ctx)) { - /* - * We raced with some other task; use - * the context they set. - */ - kfree(ctx); - goto retry; - } - get_task_struct(task); - } - - put_task_struct(task); - return ctx; - - errout: - put_task_struct(task); - return ERR_PTR(err); -} - -static void free_counter_rcu(struct rcu_head *head) -{ - struct perf_counter *counter; - - counter = container_of(head, struct perf_counter, rcu_head); - if (counter->ns) - put_pid_ns(counter->ns); - kfree(counter); -} - -static void perf_pending_sync(struct perf_counter *counter); - -static void free_counter(struct perf_counter *counter) -{ - perf_pending_sync(counter); - - if (!counter->parent) { - atomic_dec(&nr_counters); - if (counter->attr.mmap) - atomic_dec(&nr_mmap_counters); - if (counter->attr.comm) - atomic_dec(&nr_comm_counters); - if (counter->attr.task) - atomic_dec(&nr_task_counters); - } - - if (counter->output) { - fput(counter->output->filp); - counter->output = NULL; - } - - if (counter->destroy) - counter->destroy(counter); - - put_ctx(counter->ctx); - call_rcu(&counter->rcu_head, free_counter_rcu); -} - -/* - * Called when the last reference to the file is gone. - */ -static int perf_release(struct inode *inode, struct file *file) -{ - struct perf_counter *counter = file->private_data; - struct perf_counter_context *ctx = counter->ctx; - - file->private_data = NULL; - - WARN_ON_ONCE(ctx->parent_ctx); - mutex_lock(&ctx->mutex); - perf_counter_remove_from_context(counter); - mutex_unlock(&ctx->mutex); - - mutex_lock(&counter->owner->perf_counter_mutex); - list_del_init(&counter->owner_entry); - mutex_unlock(&counter->owner->perf_counter_mutex); - put_task_struct(counter->owner); - - free_counter(counter); - - return 0; -} - -static int perf_counter_read_size(struct perf_counter *counter) -{ - int entry = sizeof(u64); /* value */ - int size = 0; - int nr = 1; - - if (counter->attr.read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) - size += sizeof(u64); - - if (counter->attr.read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) - size += sizeof(u64); - - if (counter->attr.read_format & PERF_FORMAT_ID) - entry += sizeof(u64); - - if (counter->attr.read_format & PERF_FORMAT_GROUP) { - nr += counter->group_leader->nr_siblings; - size += sizeof(u64); - } - - size += entry * nr; - - return size; -} - -static u64 perf_counter_read_value(struct perf_counter *counter) -{ - struct perf_counter *child; - u64 total = 0; - - total += perf_counter_read(counter); - list_for_each_entry(child, &counter->child_list, child_list) - total += perf_counter_read(child); - - return total; -} - -static int perf_counter_read_entry(struct perf_counter *counter, - u64 read_format, char __user *buf) -{ - int n = 0, count = 0; - u64 values[2]; - - values[n++] = perf_counter_read_value(counter); - if (read_format & PERF_FORMAT_ID) - values[n++] = primary_counter_id(counter); - - count = n * sizeof(u64); - - if (copy_to_user(buf, values, count)) - return -EFAULT; - - return count; -} - -static int perf_counter_read_group(struct perf_counter *counter, - u64 read_format, char __user *buf) -{ - struct perf_counter *leader = counter->group_leader, *sub; - int n = 0, size = 0, err = -EFAULT; - u64 values[3]; - - values[n++] = 1 + leader->nr_siblings; - if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) { - values[n++] = leader->total_time_enabled + - atomic64_read(&leader->child_total_time_enabled); - } - if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) { - values[n++] = leader->total_time_running + - atomic64_read(&leader->child_total_time_running); - } - - size = n * sizeof(u64); - - if (copy_to_user(buf, values, size)) - return -EFAULT; - - err = perf_counter_read_entry(leader, read_format, buf + size); - if (err < 0) - return err; - - size += err; - - list_for_each_entry(sub, &leader->sibling_list, group_entry) { - err = perf_counter_read_entry(sub, read_format, - buf + size); - if (err < 0) - return err; - - size += err; - } - - return size; -} - -static int perf_counter_read_one(struct perf_counter *counter, - u64 read_format, char __user *buf) -{ - u64 values[4]; - int n = 0; - - values[n++] = perf_counter_read_value(counter); - if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) { - values[n++] = counter->total_time_enabled + - atomic64_read(&counter->child_total_time_enabled); - } - if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) { - values[n++] = counter->total_time_running + - atomic64_read(&counter->child_total_time_running); - } - if (read_format & PERF_FORMAT_ID) - values[n++] = primary_counter_id(counter); - - if (copy_to_user(buf, values, n * sizeof(u64))) - return -EFAULT; - - return n * sizeof(u64); -} - -/* - * Read the performance counter - simple non blocking version for now - */ -static ssize_t -perf_read_hw(struct perf_counter *counter, char __user *buf, size_t count) -{ - u64 read_format = counter->attr.read_format; - int ret; - - /* - * Return end-of-file for a read on a counter that is in - * error state (i.e. because it was pinned but it couldn't be - * scheduled on to the CPU at some point). - */ - if (counter->state == PERF_COUNTER_STATE_ERROR) - return 0; - - if (count < perf_counter_read_size(counter)) - return -ENOSPC; - - WARN_ON_ONCE(counter->ctx->parent_ctx); - mutex_lock(&counter->child_mutex); - if (read_format & PERF_FORMAT_GROUP) - ret = perf_counter_read_group(counter, read_format, buf); - else - ret = perf_counter_read_one(counter, read_format, buf); - mutex_unlock(&counter->child_mutex); - - return ret; -} - -static ssize_t -perf_read(struct file *file, char __user *buf, size_t count, loff_t *ppos) -{ - struct perf_counter *counter = file->private_data; - - return perf_read_hw(counter, buf, count); -} - -static unsigned int perf_poll(struct file *file, poll_table *wait) -{ - struct perf_counter *counter = file->private_data; - struct perf_mmap_data *data; - unsigned int events = POLL_HUP; - - rcu_read_lock(); - data = rcu_dereference(counter->data); - if (data) - events = atomic_xchg(&data->poll, 0); - rcu_read_unlock(); - - poll_wait(file, &counter->waitq, wait); - - return events; -} - -static void perf_counter_reset(struct perf_counter *counter) -{ - (void)perf_counter_read(counter); - atomic64_set(&counter->count, 0); - perf_counter_update_userpage(counter); -} - -/* - * Holding the top-level counter's child_mutex means that any - * descendant process that has inherited this counter will block - * in sync_child_counter if it goes to exit, thus satisfying the - * task existence requirements of perf_counter_enable/disable. - */ -static void perf_counter_for_each_child(struct perf_counter *counter, - void (*func)(struct perf_counter *)) -{ - struct perf_counter *child; - - WARN_ON_ONCE(counter->ctx->parent_ctx); - mutex_lock(&counter->child_mutex); - func(counter); - list_for_each_entry(child, &counter->child_list, child_list) - func(child); - mutex_unlock(&counter->child_mutex); -} - -static void perf_counter_for_each(struct perf_counter *counter, - void (*func)(struct perf_counter *)) -{ - struct perf_counter_context *ctx = counter->ctx; - struct perf_counter *sibling; - - WARN_ON_ONCE(ctx->parent_ctx); - mutex_lock(&ctx->mutex); - counter = counter->group_leader; - - perf_counter_for_each_child(counter, func); - func(counter); - list_for_each_entry(sibling, &counter->sibling_list, group_entry) - perf_counter_for_each_child(counter, func); - mutex_unlock(&ctx->mutex); -} - -static int perf_counter_period(struct perf_counter *counter, u64 __user *arg) -{ - struct perf_counter_context *ctx = counter->ctx; - unsigned long size; - int ret = 0; - u64 value; - - if (!counter->attr.sample_period) - return -EINVAL; - - size = copy_from_user(&value, arg, sizeof(value)); - if (size != sizeof(value)) - return -EFAULT; - - if (!value) - return -EINVAL; - - spin_lock_irq(&ctx->lock); - if (counter->attr.freq) { - if (value > sysctl_perf_counter_sample_rate) { - ret = -EINVAL; - goto unlock; - } - - counter->attr.sample_freq = value; - } else { - counter->attr.sample_period = value; - counter->hw.sample_period = value; - } -unlock: - spin_unlock_irq(&ctx->lock); - - return ret; -} - -int perf_counter_set_output(struct perf_counter *counter, int output_fd); - -static long perf_ioctl(struct file *file, unsigned int cmd, unsigned long arg) -{ - struct perf_counter *counter = file->private_data; - void (*func)(struct perf_counter *); - u32 flags = arg; - - switch (cmd) { - case PERF_COUNTER_IOC_ENABLE: - func = perf_counter_enable; - break; - case PERF_COUNTER_IOC_DISABLE: - func = perf_counter_disable; - break; - case PERF_COUNTER_IOC_RESET: - func = perf_counter_reset; - break; - - case PERF_COUNTER_IOC_REFRESH: - return perf_counter_refresh(counter, arg); - - case PERF_COUNTER_IOC_PERIOD: - return perf_counter_period(counter, (u64 __user *)arg); - - case PERF_COUNTER_IOC_SET_OUTPUT: - return perf_counter_set_output(counter, arg); - - default: - return -ENOTTY; - } - - if (flags & PERF_IOC_FLAG_GROUP) - perf_counter_for_each(counter, func); - else - perf_counter_for_each_child(counter, func); - - return 0; -} - -int perf_counter_task_enable(void) -{ - struct perf_counter *counter; - - mutex_lock(¤t->perf_counter_mutex); - list_for_each_entry(counter, ¤t->perf_counter_list, owner_entry) - perf_counter_for_each_child(counter, perf_counter_enable); - mutex_unlock(¤t->perf_counter_mutex); - - return 0; -} - -int perf_counter_task_disable(void) -{ - struct perf_counter *counter; - - mutex_lock(¤t->perf_counter_mutex); - list_for_each_entry(counter, ¤t->perf_counter_list, owner_entry) - perf_counter_for_each_child(counter, perf_counter_disable); - mutex_unlock(¤t->perf_counter_mutex); - - return 0; -} - -#ifndef PERF_COUNTER_INDEX_OFFSET -# define PERF_COUNTER_INDEX_OFFSET 0 -#endif - -static int perf_counter_index(struct perf_counter *counter) -{ - if (counter->state != PERF_COUNTER_STATE_ACTIVE) - return 0; - - return counter->hw.idx + 1 - PERF_COUNTER_INDEX_OFFSET; -} - -/* - * Callers need to ensure there can be no nesting of this function, otherwise - * the seqlock logic goes bad. We can not serialize this because the arch - * code calls this from NMI context. - */ -void perf_counter_update_userpage(struct perf_counter *counter) -{ - struct perf_counter_mmap_page *userpg; - struct perf_mmap_data *data; - - rcu_read_lock(); - data = rcu_dereference(counter->data); - if (!data) - goto unlock; - - userpg = data->user_page; - - /* - * Disable preemption so as to not let the corresponding user-space - * spin too long if we get preempted. - */ - preempt_disable(); - ++userpg->lock; - barrier(); - userpg->index = perf_counter_index(counter); - userpg->offset = atomic64_read(&counter->count); - if (counter->state == PERF_COUNTER_STATE_ACTIVE) - userpg->offset -= atomic64_read(&counter->hw.prev_count); - - userpg->time_enabled = counter->total_time_enabled + - atomic64_read(&counter->child_total_time_enabled); - - userpg->time_running = counter->total_time_running + - atomic64_read(&counter->child_total_time_running); - - barrier(); - ++userpg->lock; - preempt_enable(); -unlock: - rcu_read_unlock(); -} - -static int perf_mmap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) -{ - struct perf_counter *counter = vma->vm_file->private_data; - struct perf_mmap_data *data; - int ret = VM_FAULT_SIGBUS; - - if (vmf->flags & FAULT_FLAG_MKWRITE) { - if (vmf->pgoff == 0) - ret = 0; - return ret; - } - - rcu_read_lock(); - data = rcu_dereference(counter->data); - if (!data) - goto unlock; - - if (vmf->pgoff == 0) { - vmf->page = virt_to_page(data->user_page); - } else { - int nr = vmf->pgoff - 1; - - if ((unsigned)nr > data->nr_pages) - goto unlock; - - if (vmf->flags & FAULT_FLAG_WRITE) - goto unlock; - - vmf->page = virt_to_page(data->data_pages[nr]); - } - - get_page(vmf->page); - vmf->page->mapping = vma->vm_file->f_mapping; - vmf->page->index = vmf->pgoff; - - ret = 0; -unlock: - rcu_read_unlock(); - - return ret; -} - -static int perf_mmap_data_alloc(struct perf_counter *counter, int nr_pages) -{ - struct perf_mmap_data *data; - unsigned long size; - int i; - - WARN_ON(atomic_read(&counter->mmap_count)); - - size = sizeof(struct perf_mmap_data); - size += nr_pages * sizeof(void *); - - data = kzalloc(size, GFP_KERNEL); - if (!data) - goto fail; - - data->user_page = (void *)get_zeroed_page(GFP_KERNEL); - if (!data->user_page) - goto fail_user_page; - - for (i = 0; i < nr_pages; i++) { - data->data_pages[i] = (void *)get_zeroed_page(GFP_KERNEL); - if (!data->data_pages[i]) - goto fail_data_pages; - } - - data->nr_pages = nr_pages; - atomic_set(&data->lock, -1); - - if (counter->attr.watermark) { - data->watermark = min_t(long, PAGE_SIZE * nr_pages, - counter->attr.wakeup_watermark); - } - if (!data->watermark) - data->watermark = max(PAGE_SIZE, PAGE_SIZE * nr_pages / 4); - - rcu_assign_pointer(counter->data, data); - - return 0; - -fail_data_pages: - for (i--; i >= 0; i--) - free_page((unsigned long)data->data_pages[i]); - - free_page((unsigned long)data->user_page); - -fail_user_page: - kfree(data); - -fail: - return -ENOMEM; -} - -static void perf_mmap_free_page(unsigned long addr) -{ - struct page *page = virt_to_page((void *)addr); - - page->mapping = NULL; - __free_page(page); -} - -static void __perf_mmap_data_free(struct rcu_head *rcu_head) -{ - struct perf_mmap_data *data; - int i; - - data = container_of(rcu_head, struct perf_mmap_data, rcu_head); - - perf_mmap_free_page((unsigned long)data->user_page); - for (i = 0; i < data->nr_pages; i++) - perf_mmap_free_page((unsigned long)data->data_pages[i]); - - kfree(data); -} - -static void perf_mmap_data_free(struct perf_counter *counter) -{ - struct perf_mmap_data *data = counter->data; - - WARN_ON(atomic_read(&counter->mmap_count)); - - rcu_assign_pointer(counter->data, NULL); - call_rcu(&data->rcu_head, __perf_mmap_data_free); -} - -static void perf_mmap_open(struct vm_area_struct *vma) -{ - struct perf_counter *counter = vma->vm_file->private_data; - - atomic_inc(&counter->mmap_count); -} - -static void perf_mmap_close(struct vm_area_struct *vma) -{ - struct perf_counter *counter = vma->vm_file->private_data; - - WARN_ON_ONCE(counter->ctx->parent_ctx); - if (atomic_dec_and_mutex_lock(&counter->mmap_count, &counter->mmap_mutex)) { - struct user_struct *user = current_user(); - - atomic_long_sub(counter->data->nr_pages + 1, &user->locked_vm); - vma->vm_mm->locked_vm -= counter->data->nr_locked; - perf_mmap_data_free(counter); - mutex_unlock(&counter->mmap_mutex); - } -} - -static struct vm_operations_struct perf_mmap_vmops = { - .open = perf_mmap_open, - .close = perf_mmap_close, - .fault = perf_mmap_fault, - .page_mkwrite = perf_mmap_fault, -}; - -static int perf_mmap(struct file *file, struct vm_area_struct *vma) -{ - struct perf_counter *counter = file->private_data; - unsigned long user_locked, user_lock_limit; - struct user_struct *user = current_user(); - unsigned long locked, lock_limit; - unsigned long vma_size; - unsigned long nr_pages; - long user_extra, extra; - int ret = 0; - - if (!(vma->vm_flags & VM_SHARED)) - return -EINVAL; - - vma_size = vma->vm_end - vma->vm_start; - nr_pages = (vma_size / PAGE_SIZE) - 1; - - /* - * If we have data pages ensure they're a power-of-two number, so we - * can do bitmasks instead of modulo. - */ - if (nr_pages != 0 && !is_power_of_2(nr_pages)) - return -EINVAL; - - if (vma_size != PAGE_SIZE * (1 + nr_pages)) - return -EINVAL; - - if (vma->vm_pgoff != 0) - return -EINVAL; - - WARN_ON_ONCE(counter->ctx->parent_ctx); - mutex_lock(&counter->mmap_mutex); - if (counter->output) { - ret = -EINVAL; - goto unlock; - } - - if (atomic_inc_not_zero(&counter->mmap_count)) { - if (nr_pages != counter->data->nr_pages) - ret = -EINVAL; - goto unlock; - } - - user_extra = nr_pages + 1; - user_lock_limit = sysctl_perf_counter_mlock >> (PAGE_SHIFT - 10); - - /* - * Increase the limit linearly with more CPUs: - */ - user_lock_limit *= num_online_cpus(); - - user_locked = atomic_long_read(&user->locked_vm) + user_extra; - - extra = 0; - if (user_locked > user_lock_limit) - extra = user_locked - user_lock_limit; - - lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur; - lock_limit >>= PAGE_SHIFT; - locked = vma->vm_mm->locked_vm + extra; - - if ((locked > lock_limit) && perf_paranoid_tracepoint_raw() && - !capable(CAP_IPC_LOCK)) { - ret = -EPERM; - goto unlock; - } - - WARN_ON(counter->data); - ret = perf_mmap_data_alloc(counter, nr_pages); - if (ret) - goto unlock; - - atomic_set(&counter->mmap_count, 1); - atomic_long_add(user_extra, &user->locked_vm); - vma->vm_mm->locked_vm += extra; - counter->data->nr_locked = extra; - if (vma->vm_flags & VM_WRITE) - counter->data->writable = 1; - -unlock: - mutex_unlock(&counter->mmap_mutex); - - vma->vm_flags |= VM_RESERVED; - vma->vm_ops = &perf_mmap_vmops; - - return ret; -} - -static int perf_fasync(int fd, struct file *filp, int on) -{ - struct inode *inode = filp->f_path.dentry->d_inode; - struct perf_counter *counter = filp->private_data; - int retval; - - mutex_lock(&inode->i_mutex); - retval = fasync_helper(fd, filp, on, &counter->fasync); - mutex_unlock(&inode->i_mutex); - - if (retval < 0) - return retval; - - return 0; -} - -static const struct file_operations perf_fops = { - .release = perf_release, - .read = perf_read, - .poll = perf_poll, - .unlocked_ioctl = perf_ioctl, - .compat_ioctl = perf_ioctl, - .mmap = perf_mmap, - .fasync = perf_fasync, -}; - -/* - * Perf counter wakeup - * - * If there's data, ensure we set the poll() state and publish everything - * to user-space before waking everybody up. - */ - -void perf_counter_wakeup(struct perf_counter *counter) -{ - wake_up_all(&counter->waitq); - - if (counter->pending_kill) { - kill_fasync(&counter->fasync, SIGIO, counter->pending_kill); - counter->pending_kill = 0; - } -} - -/* - * Pending wakeups - * - * Handle the case where we need to wakeup up from NMI (or rq->lock) context. - * - * The NMI bit means we cannot possibly take locks. Therefore, maintain a - * single linked list and use cmpxchg() to add entries lockless. - */ - -static void perf_pending_counter(struct perf_pending_entry *entry) -{ - struct perf_counter *counter = container_of(entry, - struct perf_counter, pending); - - if (counter->pending_disable) { - counter->pending_disable = 0; - __perf_counter_disable(counter); - } - - if (counter->pending_wakeup) { - counter->pending_wakeup = 0; - perf_counter_wakeup(counter); - } -} - -#define PENDING_TAIL ((struct perf_pending_entry *)-1UL) - -static DEFINE_PER_CPU(struct perf_pending_entry *, perf_pending_head) = { - PENDING_TAIL, -}; - -static void perf_pending_queue(struct perf_pending_entry *entry, - void (*func)(struct perf_pending_entry *)) -{ - struct perf_pending_entry **head; - - if (cmpxchg(&entry->next, NULL, PENDING_TAIL) != NULL) - return; - - entry->func = func; - - head = &get_cpu_var(perf_pending_head); - - do { - entry->next = *head; - } while (cmpxchg(head, entry->next, entry) != entry->next); - - set_perf_counter_pending(); - - put_cpu_var(perf_pending_head); -} - -static int __perf_pending_run(void) -{ - struct perf_pending_entry *list; - int nr = 0; - - list = xchg(&__get_cpu_var(perf_pending_head), PENDING_TAIL); - while (list != PENDING_TAIL) { - void (*func)(struct perf_pending_entry *); - struct perf_pending_entry *entry = list; - - list = list->next; - - func = entry->func; - entry->next = NULL; - /* - * Ensure we observe the unqueue before we issue the wakeup, - * so that we won't be waiting forever. - * -- see perf_not_pending(). - */ - smp_wmb(); - - func(entry); - nr++; - } - - return nr; -} - -static inline int perf_not_pending(struct perf_counter *counter) -{ - /* - * If we flush on whatever cpu we run, there is a chance we don't - * need to wait. - */ - get_cpu(); - __perf_pending_run(); - put_cpu(); - - /* - * Ensure we see the proper queue state before going to sleep - * so that we do not miss the wakeup. -- see perf_pending_handle() - */ - smp_rmb(); - return counter->pending.next == NULL; -} - -static void perf_pending_sync(struct perf_counter *counter) -{ - wait_event(counter->waitq, perf_not_pending(counter)); -} - -void perf_counter_do_pending(void) -{ - __perf_pending_run(); -} - -/* - * Callchain support -- arch specific - */ - -__weak struct perf_callchain_entry *perf_callchain(struct pt_regs *regs) -{ - return NULL; -} - -/* - * Output - */ -static bool perf_output_space(struct perf_mmap_data *data, unsigned long tail, - unsigned long offset, unsigned long head) -{ - unsigned long mask; - - if (!data->writable) - return true; - - mask = (data->nr_pages << PAGE_SHIFT) - 1; - - offset = (offset - tail) & mask; - head = (head - tail) & mask; - - if ((int)(head - offset) < 0) - return false; - - return true; -} - -static void perf_output_wakeup(struct perf_output_handle *handle) -{ - atomic_set(&handle->data->poll, POLL_IN); - - if (handle->nmi) { - handle->counter->pending_wakeup = 1; - perf_pending_queue(&handle->counter->pending, - perf_pending_counter); - } else - perf_counter_wakeup(handle->counter); -} - -/* - * Curious locking construct. - * - * We need to ensure a later event doesn't publish a head when a former - * event isn't done writing. However since we need to deal with NMIs we - * cannot fully serialize things. - * - * What we do is serialize between CPUs so we only have to deal with NMI - * nesting on a single CPU. - * - * We only publish the head (and generate a wakeup) when the outer-most - * event completes. - */ -static void perf_output_lock(struct perf_output_handle *handle) -{ - struct perf_mmap_data *data = handle->data; - int cpu; - - handle->locked = 0; - - local_irq_save(handle->flags); - cpu = smp_processor_id(); - - if (in_nmi() && atomic_read(&data->lock) == cpu) - return; - - while (atomic_cmpxchg(&data->lock, -1, cpu) != -1) - cpu_relax(); - - handle->locked = 1; -} - -static void perf_output_unlock(struct perf_output_handle *handle) -{ - struct perf_mmap_data *data = handle->data; - unsigned long head; - int cpu; - - data->done_head = data->head; - - if (!handle->locked) - goto out; - -again: - /* - * The xchg implies a full barrier that ensures all writes are done - * before we publish the new head, matched by a rmb() in userspace when - * reading this position. - */ - while ((head = atomic_long_xchg(&data->done_head, 0))) - data->user_page->data_head = head; - - /* - * NMI can happen here, which means we can miss a done_head update. - */ - - cpu = atomic_xchg(&data->lock, -1); - WARN_ON_ONCE(cpu != smp_processor_id()); - - /* - * Therefore we have to validate we did not indeed do so. - */ - if (unlikely(atomic_long_read(&data->done_head))) { - /* - * Since we had it locked, we can lock it again. - */ - while (atomic_cmpxchg(&data->lock, -1, cpu) != -1) - cpu_relax(); - - goto again; - } - - if (atomic_xchg(&data->wakeup, 0)) - perf_output_wakeup(handle); -out: - local_irq_restore(handle->flags); -} - -void perf_output_copy(struct perf_output_handle *handle, - const void *buf, unsigned int len) -{ - unsigned int pages_mask; - unsigned int offset; - unsigned int size; - void **pages; - - offset = handle->offset; - pages_mask = handle->data->nr_pages - 1; - pages = handle->data->data_pages; - - do { - unsigned int page_offset; - int nr; - - nr = (offset >> PAGE_SHIFT) & pages_mask; - page_offset = offset & (PAGE_SIZE - 1); - size = min_t(unsigned int, PAGE_SIZE - page_offset, len); - - memcpy(pages[nr] + page_offset, buf, size); - - len -= size; - buf += size; - offset += size; - } while (len); - - handle->offset = offset; - - /* - * Check we didn't copy past our reservation window, taking the - * possible unsigned int wrap into account. - */ - WARN_ON_ONCE(((long)(handle->head - handle->offset)) < 0); -} - -int perf_output_begin(struct perf_output_handle *handle, - struct perf_counter *counter, unsigned int size, - int nmi, int sample) -{ - struct perf_counter *output_counter; - struct perf_mmap_data *data; - unsigned long tail, offset, head; - int have_lost; - struct { - struct perf_event_header header; - u64 id; - u64 lost; - } lost_event; - - rcu_read_lock(); - /* - * For inherited counters we send all the output towards the parent. - */ - if (counter->parent) - counter = counter->parent; - - output_counter = rcu_dereference(counter->output); - if (output_counter) - counter = output_counter; - - data = rcu_dereference(counter->data); - if (!data) - goto out; - - handle->data = data; - handle->counter = counter; - handle->nmi = nmi; - handle->sample = sample; - - if (!data->nr_pages) - goto fail; - - have_lost = atomic_read(&data->lost); - if (have_lost) - size += sizeof(lost_event); - - perf_output_lock(handle); - - do { - /* - * Userspace could choose to issue a mb() before updating the - * tail pointer. So that all reads will be completed before the - * write is issued. - */ - tail = ACCESS_ONCE(data->user_page->data_tail); - smp_rmb(); - offset = head = atomic_long_read(&data->head); - head += size; - if (unlikely(!perf_output_space(data, tail, offset, head))) - goto fail; - } while (atomic_long_cmpxchg(&data->head, offset, head) != offset); - - handle->offset = offset; - handle->head = head; - - if (head - tail > data->watermark) - atomic_set(&data->wakeup, 1); - - if (have_lost) { - lost_event.header.type = PERF_EVENT_LOST; - lost_event.header.misc = 0; - lost_event.header.size = sizeof(lost_event); - lost_event.id = counter->id; - lost_event.lost = atomic_xchg(&data->lost, 0); - - perf_output_put(handle, lost_event); - } - - return 0; - -fail: - atomic_inc(&data->lost); - perf_output_unlock(handle); -out: - rcu_read_unlock(); - - return -ENOSPC; -} - -void perf_output_end(struct perf_output_handle *handle) -{ - struct perf_counter *counter = handle->counter; - struct perf_mmap_data *data = handle->data; - - int wakeup_events = counter->attr.wakeup_events; - - if (handle->sample && wakeup_events) { - int events = atomic_inc_return(&data->events); - if (events >= wakeup_events) { - atomic_sub(wakeup_events, &data->events); - atomic_set(&data->wakeup, 1); - } - } - - perf_output_unlock(handle); - rcu_read_unlock(); -} - -static u32 perf_counter_pid(struct perf_counter *counter, struct task_struct *p) -{ - /* - * only top level counters have the pid namespace they were created in - */ - if (counter->parent) - counter = counter->parent; - - return task_tgid_nr_ns(p, counter->ns); -} - -static u32 perf_counter_tid(struct perf_counter *counter, struct task_struct *p) -{ - /* - * only top level counters have the pid namespace they were created in - */ - if (counter->parent) - counter = counter->parent; - - return task_pid_nr_ns(p, counter->ns); -} - -static void perf_output_read_one(struct perf_output_handle *handle, - struct perf_counter *counter) -{ - u64 read_format = counter->attr.read_format; - u64 values[4]; - int n = 0; - - values[n++] = atomic64_read(&counter->count); - if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) { - values[n++] = counter->total_time_enabled + - atomic64_read(&counter->child_total_time_enabled); - } - if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) { - values[n++] = counter->total_time_running + - atomic64_read(&counter->child_total_time_running); - } - if (read_format & PERF_FORMAT_ID) - values[n++] = primary_counter_id(counter); - - perf_output_copy(handle, values, n * sizeof(u64)); -} - -/* - * XXX PERF_FORMAT_GROUP vs inherited counters seems difficult. - */ -static void perf_output_read_group(struct perf_output_handle *handle, - struct perf_counter *counter) -{ - struct perf_counter *leader = counter->group_leader, *sub; - u64 read_format = counter->attr.read_format; - u64 values[5]; - int n = 0; - - values[n++] = 1 + leader->nr_siblings; - - if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) - values[n++] = leader->total_time_enabled; - - if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) - values[n++] = leader->total_time_running; - - if (leader != counter) - leader->pmu->read(leader); - - values[n++] = atomic64_read(&leader->count); - if (read_format & PERF_FORMAT_ID) - values[n++] = primary_counter_id(leader); - - perf_output_copy(handle, values, n * sizeof(u64)); - - list_for_each_entry(sub, &leader->sibling_list, group_entry) { - n = 0; - - if (sub != counter) - sub->pmu->read(sub); - - values[n++] = atomic64_read(&sub->count); - if (read_format & PERF_FORMAT_ID) - values[n++] = primary_counter_id(sub); - - perf_output_copy(handle, values, n * sizeof(u64)); - } -} - -static void perf_output_read(struct perf_output_handle *handle, - struct perf_counter *counter) -{ - if (counter->attr.read_format & PERF_FORMAT_GROUP) - perf_output_read_group(handle, counter); - else - perf_output_read_one(handle, counter); -} - -void perf_output_sample(struct perf_output_handle *handle, - struct perf_event_header *header, - struct perf_sample_data *data, - struct perf_counter *counter) -{ - u64 sample_type = data->type; - - perf_output_put(handle, *header); - - if (sample_type & PERF_SAMPLE_IP) - perf_output_put(handle, data->ip); - - if (sample_type & PERF_SAMPLE_TID) - perf_output_put(handle, data->tid_entry); - - if (sample_type & PERF_SAMPLE_TIME) - perf_output_put(handle, data->time); - - if (sample_type & PERF_SAMPLE_ADDR) - perf_output_put(handle, data->addr); - - if (sample_type & PERF_SAMPLE_ID) - perf_output_put(handle, data->id); - - if (sample_type & PERF_SAMPLE_STREAM_ID) - perf_output_put(handle, data->stream_id); - - if (sample_type & PERF_SAMPLE_CPU) - perf_output_put(handle, data->cpu_entry); - - if (sample_type & PERF_SAMPLE_PERIOD) - perf_output_put(handle, data->period); - - if (sample_type & PERF_SAMPLE_READ) - perf_output_read(handle, counter); - - if (sample_type & PERF_SAMPLE_CALLCHAIN) { - if (data->callchain) { - int size = 1; - - if (data->callchain) - size += data->callchain->nr; - - size *= sizeof(u64); - - perf_output_copy(handle, data->callchain, size); - } else { - u64 nr = 0; - perf_output_put(handle, nr); - } - } - - if (sample_type & PERF_SAMPLE_RAW) { - if (data->raw) { - perf_output_put(handle, data->raw->size); - perf_output_copy(handle, data->raw->data, - data->raw->size); - } else { - struct { - u32 size; - u32 data; - } raw = { - .size = sizeof(u32), - .data = 0, - }; - perf_output_put(handle, raw); - } - } -} - -void perf_prepare_sample(struct perf_event_header *header, - struct perf_sample_data *data, - struct perf_counter *counter, - struct pt_regs *regs) -{ - u64 sample_type = counter->attr.sample_type; - - data->type = sample_type; - - header->type = PERF_EVENT_SAMPLE; - header->size = sizeof(*header); - - header->misc = 0; - header->misc |= perf_misc_flags(regs); - - if (sample_type & PERF_SAMPLE_IP) { - data->ip = perf_instruction_pointer(regs); - - header->size += sizeof(data->ip); - } - - if (sample_type & PERF_SAMPLE_TID) { - /* namespace issues */ - data->tid_entry.pid = perf_counter_pid(counter, current); - data->tid_entry.tid = perf_counter_tid(counter, current); - - header->size += sizeof(data->tid_entry); - } - - if (sample_type & PERF_SAMPLE_TIME) { - data->time = perf_clock(); - - header->size += sizeof(data->time); - } - - if (sample_type & PERF_SAMPLE_ADDR) - header->size += sizeof(data->addr); - - if (sample_type & PERF_SAMPLE_ID) { - data->id = primary_counter_id(counter); - - header->size += sizeof(data->id); - } - - if (sample_type & PERF_SAMPLE_STREAM_ID) { - data->stream_id = counter->id; - - header->size += sizeof(data->stream_id); - } - - if (sample_type & PERF_SAMPLE_CPU) { - data->cpu_entry.cpu = raw_smp_processor_id(); - data->cpu_entry.reserved = 0; - - header->size += sizeof(data->cpu_entry); - } - - if (sample_type & PERF_SAMPLE_PERIOD) - header->size += sizeof(data->period); - - if (sample_type & PERF_SAMPLE_READ) - header->size += perf_counter_read_size(counter); - - if (sample_type & PERF_SAMPLE_CALLCHAIN) { - int size = 1; - - data->callchain = perf_callchain(regs); - - if (data->callchain) - size += data->callchain->nr; - - header->size += size * sizeof(u64); - } - - if (sample_type & PERF_SAMPLE_RAW) { - int size = sizeof(u32); - - if (data->raw) - size += data->raw->size; - else - size += sizeof(u32); - - WARN_ON_ONCE(size & (sizeof(u64)-1)); - header->size += size; - } -} - -static void perf_counter_output(struct perf_counter *counter, int nmi, - struct perf_sample_data *data, - struct pt_regs *regs) -{ - struct perf_output_handle handle; - struct perf_event_header header; - - perf_prepare_sample(&header, data, counter, regs); - - if (perf_output_begin(&handle, counter, header.size, nmi, 1)) - return; - - perf_output_sample(&handle, &header, data, counter); - - perf_output_end(&handle); -} - -/* - * read event - */ - -struct perf_read_event { - struct perf_event_header header; - - u32 pid; - u32 tid; -}; - -static void -perf_counter_read_event(struct perf_counter *counter, - struct task_struct *task) -{ - struct perf_output_handle handle; - struct perf_read_event read_event = { - .header = { - .type = PERF_EVENT_READ, - .misc = 0, - .size = sizeof(read_event) + perf_counter_read_size(counter), - }, - .pid = perf_counter_pid(counter, task), - .tid = perf_counter_tid(counter, task), - }; - int ret; - - ret = perf_output_begin(&handle, counter, read_event.header.size, 0, 0); - if (ret) - return; - - perf_output_put(&handle, read_event); - perf_output_read(&handle, counter); - - perf_output_end(&handle); -} - -/* - * task tracking -- fork/exit - * - * enabled by: attr.comm | attr.mmap | attr.task - */ - -struct perf_task_event { - struct task_struct *task; - struct perf_counter_context *task_ctx; - - struct { - struct perf_event_header header; - - u32 pid; - u32 ppid; - u32 tid; - u32 ptid; - u64 time; - } event; -}; - -static void perf_counter_task_output(struct perf_counter *counter, - struct perf_task_event *task_event) -{ - struct perf_output_handle handle; - int size; - struct task_struct *task = task_event->task; - int ret; - - size = task_event->event.header.size; - ret = perf_output_begin(&handle, counter, size, 0, 0); - - if (ret) - return; - - task_event->event.pid = perf_counter_pid(counter, task); - task_event->event.ppid = perf_counter_pid(counter, current); - - task_event->event.tid = perf_counter_tid(counter, task); - task_event->event.ptid = perf_counter_tid(counter, current); - - task_event->event.time = perf_clock(); - - perf_output_put(&handle, task_event->event); - - perf_output_end(&handle); -} - -static int perf_counter_task_match(struct perf_counter *counter) -{ - if (counter->attr.comm || counter->attr.mmap || counter->attr.task) - return 1; - - return 0; -} - -static void perf_counter_task_ctx(struct perf_counter_context *ctx, - struct perf_task_event *task_event) -{ - struct perf_counter *counter; - - if (system_state != SYSTEM_RUNNING || list_empty(&ctx->event_list)) - return; - - rcu_read_lock(); - list_for_each_entry_rcu(counter, &ctx->event_list, event_entry) { - if (perf_counter_task_match(counter)) - perf_counter_task_output(counter, task_event); - } - rcu_read_unlock(); -} - -static void perf_counter_task_event(struct perf_task_event *task_event) -{ - struct perf_cpu_context *cpuctx; - struct perf_counter_context *ctx = task_event->task_ctx; - - cpuctx = &get_cpu_var(perf_cpu_context); - perf_counter_task_ctx(&cpuctx->ctx, task_event); - put_cpu_var(perf_cpu_context); - - rcu_read_lock(); - if (!ctx) - ctx = rcu_dereference(task_event->task->perf_counter_ctxp); - if (ctx) - perf_counter_task_ctx(ctx, task_event); - rcu_read_unlock(); -} - -static void perf_counter_task(struct task_struct *task, - struct perf_counter_context *task_ctx, - int new) -{ - struct perf_task_event task_event; - - if (!atomic_read(&nr_comm_counters) && - !atomic_read(&nr_mmap_counters) && - !atomic_read(&nr_task_counters)) - return; - - task_event = (struct perf_task_event){ - .task = task, - .task_ctx = task_ctx, - .event = { - .header = { - .type = new ? PERF_EVENT_FORK : PERF_EVENT_EXIT, - .misc = 0, - .size = sizeof(task_event.event), - }, - /* .pid */ - /* .ppid */ - /* .tid */ - /* .ptid */ - }, - }; - - perf_counter_task_event(&task_event); -} - -void perf_counter_fork(struct task_struct *task) -{ - perf_counter_task(task, NULL, 1); -} - -/* - * comm tracking - */ - -struct perf_comm_event { - struct task_struct *task; - char *comm; - int comm_size; - - struct { - struct perf_event_header header; - - u32 pid; - u32 tid; - } event; -}; - -static void perf_counter_comm_output(struct perf_counter *counter, - struct perf_comm_event *comm_event) -{ - struct perf_output_handle handle; - int size = comm_event->event.header.size; - int ret = perf_output_begin(&handle, counter, size, 0, 0); - - if (ret) - return; - - comm_event->event.pid = perf_counter_pid(counter, comm_event->task); - comm_event->event.tid = perf_counter_tid(counter, comm_event->task); - - perf_output_put(&handle, comm_event->event); - perf_output_copy(&handle, comm_event->comm, - comm_event->comm_size); - perf_output_end(&handle); -} - -static int perf_counter_comm_match(struct perf_counter *counter) -{ - if (counter->attr.comm) - return 1; - - return 0; -} - -static void perf_counter_comm_ctx(struct perf_counter_context *ctx, - struct perf_comm_event *comm_event) -{ - struct perf_counter *counter; - - if (system_state != SYSTEM_RUNNING || list_empty(&ctx->event_list)) - return; - - rcu_read_lock(); - list_for_each_entry_rcu(counter, &ctx->event_list, event_entry) { - if (perf_counter_comm_match(counter)) - perf_counter_comm_output(counter, comm_event); - } - rcu_read_unlock(); -} - -static void perf_counter_comm_event(struct perf_comm_event *comm_event) -{ - struct perf_cpu_context *cpuctx; - struct perf_counter_context *ctx; - unsigned int size; - char comm[TASK_COMM_LEN]; - - memset(comm, 0, sizeof(comm)); - strncpy(comm, comm_event->task->comm, sizeof(comm)); - size = ALIGN(strlen(comm)+1, sizeof(u64)); - - comm_event->comm = comm; - comm_event->comm_size = size; - - comm_event->event.header.size = sizeof(comm_event->event) + size; - - cpuctx = &get_cpu_var(perf_cpu_context); - perf_counter_comm_ctx(&cpuctx->ctx, comm_event); - put_cpu_var(perf_cpu_context); - - rcu_read_lock(); - /* - * doesn't really matter which of the child contexts the - * events ends up in. - */ - ctx = rcu_dereference(current->perf_counter_ctxp); - if (ctx) - perf_counter_comm_ctx(ctx, comm_event); - rcu_read_unlock(); -} - -void perf_counter_comm(struct task_struct *task) -{ - struct perf_comm_event comm_event; - - if (task->perf_counter_ctxp) - perf_counter_enable_on_exec(task); - - if (!atomic_read(&nr_comm_counters)) - return; - - comm_event = (struct perf_comm_event){ - .task = task, - /* .comm */ - /* .comm_size */ - .event = { - .header = { - .type = PERF_EVENT_COMM, - .misc = 0, - /* .size */ - }, - /* .pid */ - /* .tid */ - }, - }; - - perf_counter_comm_event(&comm_event); -} - -/* - * mmap tracking - */ - -struct perf_mmap_event { - struct vm_area_struct *vma; - - const char *file_name; - int file_size; - - struct { - struct perf_event_header header; - - u32 pid; - u32 tid; - u64 start; - u64 len; - u64 pgoff; - } event; -}; - -static void perf_counter_mmap_output(struct perf_counter *counter, - struct perf_mmap_event *mmap_event) -{ - struct perf_output_handle handle; - int size = mmap_event->event.header.size; - int ret = perf_output_begin(&handle, counter, size, 0, 0); - - if (ret) - return; - - mmap_event->event.pid = perf_counter_pid(counter, current); - mmap_event->event.tid = perf_counter_tid(counter, current); - - perf_output_put(&handle, mmap_event->event); - perf_output_copy(&handle, mmap_event->file_name, - mmap_event->file_size); - perf_output_end(&handle); -} - -static int perf_counter_mmap_match(struct perf_counter *counter, - struct perf_mmap_event *mmap_event) -{ - if (counter->attr.mmap) - return 1; - - return 0; -} - -static void perf_counter_mmap_ctx(struct perf_counter_context *ctx, - struct perf_mmap_event *mmap_event) -{ - struct perf_counter *counter; - - if (system_state != SYSTEM_RUNNING || list_empty(&ctx->event_list)) - return; - - rcu_read_lock(); - list_for_each_entry_rcu(counter, &ctx->event_list, event_entry) { - if (perf_counter_mmap_match(counter, mmap_event)) - perf_counter_mmap_output(counter, mmap_event); - } - rcu_read_unlock(); -} - -static void perf_counter_mmap_event(struct perf_mmap_event *mmap_event) -{ - struct perf_cpu_context *cpuctx; - struct perf_counter_context *ctx; - struct vm_area_struct *vma = mmap_event->vma; - struct file *file = vma->vm_file; - unsigned int size; - char tmp[16]; - char *buf = NULL; - const char *name; - - memset(tmp, 0, sizeof(tmp)); - - if (file) { - /* - * d_path works from the end of the buffer backwards, so we - * need to add enough zero bytes after the string to handle - * the 64bit alignment we do later. - */ - buf = kzalloc(PATH_MAX + sizeof(u64), GFP_KERNEL); - if (!buf) { - name = strncpy(tmp, "//enomem", sizeof(tmp)); - goto got_name; - } - name = d_path(&file->f_path, buf, PATH_MAX); - if (IS_ERR(name)) { - name = strncpy(tmp, "//toolong", sizeof(tmp)); - goto got_name; - } - } else { - if (arch_vma_name(mmap_event->vma)) { - name = strncpy(tmp, arch_vma_name(mmap_event->vma), - sizeof(tmp)); - goto got_name; - } - - if (!vma->vm_mm) { - name = strncpy(tmp, "[vdso]", sizeof(tmp)); - goto got_name; - } - - name = strncpy(tmp, "//anon", sizeof(tmp)); - goto got_name; - } - -got_name: - size = ALIGN(strlen(name)+1, sizeof(u64)); - - mmap_event->file_name = name; - mmap_event->file_size = size; - - mmap_event->event.header.size = sizeof(mmap_event->event) + size; - - cpuctx = &get_cpu_var(perf_cpu_context); - perf_counter_mmap_ctx(&cpuctx->ctx, mmap_event); - put_cpu_var(perf_cpu_context); - - rcu_read_lock(); - /* - * doesn't really matter which of the child contexts the - * events ends up in. - */ - ctx = rcu_dereference(current->perf_counter_ctxp); - if (ctx) - perf_counter_mmap_ctx(ctx, mmap_event); - rcu_read_unlock(); - - kfree(buf); -} - -void __perf_counter_mmap(struct vm_area_struct *vma) -{ - struct perf_mmap_event mmap_event; - - if (!atomic_read(&nr_mmap_counters)) - return; - - mmap_event = (struct perf_mmap_event){ - .vma = vma, - /* .file_name */ - /* .file_size */ - .event = { - .header = { - .type = PERF_EVENT_MMAP, - .misc = 0, - /* .size */ - }, - /* .pid */ - /* .tid */ - .start = vma->vm_start, - .len = vma->vm_end - vma->vm_start, - .pgoff = vma->vm_pgoff, - }, - }; - - perf_counter_mmap_event(&mmap_event); -} - -/* - * IRQ throttle logging - */ - -static void perf_log_throttle(struct perf_counter *counter, int enable) -{ - struct perf_output_handle handle; - int ret; - - struct { - struct perf_event_header header; - u64 time; - u64 id; - u64 stream_id; - } throttle_event = { - .header = { - .type = PERF_EVENT_THROTTLE, - .misc = 0, - .size = sizeof(throttle_event), - }, - .time = perf_clock(), - .id = primary_counter_id(counter), - .stream_id = counter->id, - }; - - if (enable) - throttle_event.header.type = PERF_EVENT_UNTHROTTLE; - - ret = perf_output_begin(&handle, counter, sizeof(throttle_event), 1, 0); - if (ret) - return; - - perf_output_put(&handle, throttle_event); - perf_output_end(&handle); -} - -/* - * Generic counter overflow handling, sampling. - */ - -static int __perf_counter_overflow(struct perf_counter *counter, int nmi, - int throttle, struct perf_sample_data *data, - struct pt_regs *regs) -{ - int events = atomic_read(&counter->event_limit); - struct hw_perf_counter *hwc = &counter->hw; - int ret = 0; - - throttle = (throttle && counter->pmu->unthrottle != NULL); - - if (!throttle) { - hwc->interrupts++; - } else { - if (hwc->interrupts != MAX_INTERRUPTS) { - hwc->interrupts++; - if (HZ * hwc->interrupts > - (u64)sysctl_perf_counter_sample_rate) { - hwc->interrupts = MAX_INTERRUPTS; - perf_log_throttle(counter, 0); - ret = 1; - } - } else { - /* - * Keep re-disabling counters even though on the previous - * pass we disabled it - just in case we raced with a - * sched-in and the counter got enabled again: - */ - ret = 1; - } - } - - if (counter->attr.freq) { - u64 now = perf_clock(); - s64 delta = now - hwc->freq_stamp; - - hwc->freq_stamp = now; - - if (delta > 0 && delta < TICK_NSEC) - perf_adjust_period(counter, NSEC_PER_SEC / (int)delta); - } - - /* - * XXX event_limit might not quite work as expected on inherited - * counters - */ - - counter->pending_kill = POLL_IN; - if (events && atomic_dec_and_test(&counter->event_limit)) { - ret = 1; - counter->pending_kill = POLL_HUP; - if (nmi) { - counter->pending_disable = 1; - perf_pending_queue(&counter->pending, - perf_pending_counter); - } else - perf_counter_disable(counter); - } - - perf_counter_output(counter, nmi, data, regs); - return ret; -} - -int perf_counter_overflow(struct perf_counter *counter, int nmi, - struct perf_sample_data *data, - struct pt_regs *regs) -{ - return __perf_counter_overflow(counter, nmi, 1, data, regs); -} - -/* - * Generic software counter infrastructure - */ - -/* - * We directly increment counter->count and keep a second value in - * counter->hw.period_left to count intervals. This period counter - * is kept in the range [-sample_period, 0] so that we can use the - * sign as trigger. - */ - -static u64 perf_swcounter_set_period(struct perf_counter *counter) -{ - struct hw_perf_counter *hwc = &counter->hw; - u64 period = hwc->last_period; - u64 nr, offset; - s64 old, val; - - hwc->last_period = hwc->sample_period; - -again: - old = val = atomic64_read(&hwc->period_left); - if (val < 0) - return 0; - - nr = div64_u64(period + val, period); - offset = nr * period; - val -= offset; - if (atomic64_cmpxchg(&hwc->period_left, old, val) != old) - goto again; - - return nr; -} - -static void perf_swcounter_overflow(struct perf_counter *counter, - int nmi, struct perf_sample_data *data, - struct pt_regs *regs) -{ - struct hw_perf_counter *hwc = &counter->hw; - int throttle = 0; - u64 overflow; - - data->period = counter->hw.last_period; - overflow = perf_swcounter_set_period(counter); - - if (hwc->interrupts == MAX_INTERRUPTS) - return; - - for (; overflow; overflow--) { - if (__perf_counter_overflow(counter, nmi, throttle, - data, regs)) { - /* - * We inhibit the overflow from happening when - * hwc->interrupts == MAX_INTERRUPTS. - */ - break; - } - throttle = 1; - } -} - -static void perf_swcounter_unthrottle(struct perf_counter *counter) -{ - /* - * Nothing to do, we already reset hwc->interrupts. - */ -} - -static void perf_swcounter_add(struct perf_counter *counter, u64 nr, - int nmi, struct perf_sample_data *data, - struct pt_regs *regs) -{ - struct hw_perf_counter *hwc = &counter->hw; - - atomic64_add(nr, &counter->count); - - if (!hwc->sample_period) - return; - - if (!regs) - return; - - if (!atomic64_add_negative(nr, &hwc->period_left)) - perf_swcounter_overflow(counter, nmi, data, regs); -} - -static int perf_swcounter_is_counting(struct perf_counter *counter) -{ - /* - * The counter is active, we're good! - */ - if (counter->state == PERF_COUNTER_STATE_ACTIVE) - return 1; - - /* - * The counter is off/error, not counting. - */ - if (counter->state != PERF_COUNTER_STATE_INACTIVE) - return 0; - - /* - * The counter is inactive, if the context is active - * we're part of a group that didn't make it on the 'pmu', - * not counting. - */ - if (counter->ctx->is_active) - return 0; - - /* - * We're inactive and the context is too, this means the - * task is scheduled out, we're counting events that happen - * to us, like migration events. - */ - return 1; -} - -static int perf_swcounter_match(struct perf_counter *counter, - enum perf_type_id type, - u32 event_id, struct pt_regs *regs) -{ - if (!perf_swcounter_is_counting(counter)) - return 0; - - if (counter->attr.type != type) - return 0; - if (counter->attr.config != event_id) - return 0; - - if (regs) { - if (counter->attr.exclude_user && user_mode(regs)) - return 0; - - if (counter->attr.exclude_kernel && !user_mode(regs)) - return 0; - } - - return 1; -} - -static void perf_swcounter_ctx_event(struct perf_counter_context *ctx, - enum perf_type_id type, - u32 event_id, u64 nr, int nmi, - struct perf_sample_data *data, - struct pt_regs *regs) -{ - struct perf_counter *counter; - - if (system_state != SYSTEM_RUNNING || list_empty(&ctx->event_list)) - return; - - rcu_read_lock(); - list_for_each_entry_rcu(counter, &ctx->event_list, event_entry) { - if (perf_swcounter_match(counter, type, event_id, regs)) - perf_swcounter_add(counter, nr, nmi, data, regs); - } - rcu_read_unlock(); -} - -static int *perf_swcounter_recursion_context(struct perf_cpu_context *cpuctx) -{ - if (in_nmi()) - return &cpuctx->recursion[3]; - - if (in_irq()) - return &cpuctx->recursion[2]; - - if (in_softirq()) - return &cpuctx->recursion[1]; - - return &cpuctx->recursion[0]; -} - -static void do_perf_swcounter_event(enum perf_type_id type, u32 event, - u64 nr, int nmi, - struct perf_sample_data *data, - struct pt_regs *regs) -{ - struct perf_cpu_context *cpuctx = &get_cpu_var(perf_cpu_context); - int *recursion = perf_swcounter_recursion_context(cpuctx); - struct perf_counter_context *ctx; - - if (*recursion) - goto out; - - (*recursion)++; - barrier(); - - perf_swcounter_ctx_event(&cpuctx->ctx, type, event, - nr, nmi, data, regs); - rcu_read_lock(); - /* - * doesn't really matter which of the child contexts the - * events ends up in. - */ - ctx = rcu_dereference(current->perf_counter_ctxp); - if (ctx) - perf_swcounter_ctx_event(ctx, type, event, nr, nmi, data, regs); - rcu_read_unlock(); - - barrier(); - (*recursion)--; - -out: - put_cpu_var(perf_cpu_context); -} - -void __perf_swcounter_event(u32 event, u64 nr, int nmi, - struct pt_regs *regs, u64 addr) -{ - struct perf_sample_data data = { - .addr = addr, - }; - - do_perf_swcounter_event(PERF_TYPE_SOFTWARE, event, nr, nmi, - &data, regs); -} - -static void perf_swcounter_read(struct perf_counter *counter) -{ -} - -static int perf_swcounter_enable(struct perf_counter *counter) -{ - struct hw_perf_counter *hwc = &counter->hw; - - if (hwc->sample_period) { - hwc->last_period = hwc->sample_period; - perf_swcounter_set_period(counter); - } - return 0; -} - -static void perf_swcounter_disable(struct perf_counter *counter) -{ -} - -static const struct pmu perf_ops_generic = { - .enable = perf_swcounter_enable, - .disable = perf_swcounter_disable, - .read = perf_swcounter_read, - .unthrottle = perf_swcounter_unthrottle, -}; - -/* - * hrtimer based swcounter callback - */ - -static enum hrtimer_restart perf_swcounter_hrtimer(struct hrtimer *hrtimer) -{ - enum hrtimer_restart ret = HRTIMER_RESTART; - struct perf_sample_data data; - struct pt_regs *regs; - struct perf_counter *counter; - u64 period; - - counter = container_of(hrtimer, struct perf_counter, hw.hrtimer); - counter->pmu->read(counter); - - data.addr = 0; - regs = get_irq_regs(); - /* - * In case we exclude kernel IPs or are somehow not in interrupt - * context, provide the next best thing, the user IP. - */ - if ((counter->attr.exclude_kernel || !regs) && - !counter->attr.exclude_user) - regs = task_pt_regs(current); - - if (regs) { - if (perf_counter_overflow(counter, 0, &data, regs)) - ret = HRTIMER_NORESTART; - } - - period = max_t(u64, 10000, counter->hw.sample_period); - hrtimer_forward_now(hrtimer, ns_to_ktime(period)); - - return ret; -} - -/* - * Software counter: cpu wall time clock - */ - -static void cpu_clock_perf_counter_update(struct perf_counter *counter) -{ - int cpu = raw_smp_processor_id(); - s64 prev; - u64 now; - - now = cpu_clock(cpu); - prev = atomic64_read(&counter->hw.prev_count); - atomic64_set(&counter->hw.prev_count, now); - atomic64_add(now - prev, &counter->count); -} - -static int cpu_clock_perf_counter_enable(struct perf_counter *counter) -{ - struct hw_perf_counter *hwc = &counter->hw; - int cpu = raw_smp_processor_id(); - - atomic64_set(&hwc->prev_count, cpu_clock(cpu)); - hrtimer_init(&hwc->hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); - hwc->hrtimer.function = perf_swcounter_hrtimer; - if (hwc->sample_period) { - u64 period = max_t(u64, 10000, hwc->sample_period); - __hrtimer_start_range_ns(&hwc->hrtimer, - ns_to_ktime(period), 0, - HRTIMER_MODE_REL, 0); - } - - return 0; -} - -static void cpu_clock_perf_counter_disable(struct perf_counter *counter) -{ - if (counter->hw.sample_period) - hrtimer_cancel(&counter->hw.hrtimer); - cpu_clock_perf_counter_update(counter); -} - -static void cpu_clock_perf_counter_read(struct perf_counter *counter) -{ - cpu_clock_perf_counter_update(counter); -} - -static const struct pmu perf_ops_cpu_clock = { - .enable = cpu_clock_perf_counter_enable, - .disable = cpu_clock_perf_counter_disable, - .read = cpu_clock_perf_counter_read, -}; - -/* - * Software counter: task time clock - */ - -static void task_clock_perf_counter_update(struct perf_counter *counter, u64 now) -{ - u64 prev; - s64 delta; - - prev = atomic64_xchg(&counter->hw.prev_count, now); - delta = now - prev; - atomic64_add(delta, &counter->count); -} - -static int task_clock_perf_counter_enable(struct perf_counter *counter) -{ - struct hw_perf_counter *hwc = &counter->hw; - u64 now; - - now = counter->ctx->time; - - atomic64_set(&hwc->prev_count, now); - hrtimer_init(&hwc->hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); - hwc->hrtimer.function = perf_swcounter_hrtimer; - if (hwc->sample_period) { - u64 period = max_t(u64, 10000, hwc->sample_period); - __hrtimer_start_range_ns(&hwc->hrtimer, - ns_to_ktime(period), 0, - HRTIMER_MODE_REL, 0); - } - - return 0; -} - -static void task_clock_perf_counter_disable(struct perf_counter *counter) -{ - if (counter->hw.sample_period) - hrtimer_cancel(&counter->hw.hrtimer); - task_clock_perf_counter_update(counter, counter->ctx->time); - -} - -static void task_clock_perf_counter_read(struct perf_counter *counter) -{ - u64 time; - - if (!in_nmi()) { - update_context_time(counter->ctx); - time = counter->ctx->time; - } else { - u64 now = perf_clock(); - u64 delta = now - counter->ctx->timestamp; - time = counter->ctx->time + delta; - } - - task_clock_perf_counter_update(counter, time); -} - -static const struct pmu perf_ops_task_clock = { - .enable = task_clock_perf_counter_enable, - .disable = task_clock_perf_counter_disable, - .read = task_clock_perf_counter_read, -}; - -#ifdef CONFIG_EVENT_PROFILE -void perf_tpcounter_event(int event_id, u64 addr, u64 count, void *record, - int entry_size) -{ - struct perf_raw_record raw = { - .size = entry_size, - .data = record, - }; - - struct perf_sample_data data = { - .addr = addr, - .raw = &raw, - }; - - struct pt_regs *regs = get_irq_regs(); - - if (!regs) - regs = task_pt_regs(current); - - do_perf_swcounter_event(PERF_TYPE_TRACEPOINT, event_id, count, 1, - &data, regs); -} -EXPORT_SYMBOL_GPL(perf_tpcounter_event); - -extern int ftrace_profile_enable(int); -extern void ftrace_profile_disable(int); - -static void tp_perf_counter_destroy(struct perf_counter *counter) -{ - ftrace_profile_disable(counter->attr.config); -} - -static const struct pmu *tp_perf_counter_init(struct perf_counter *counter) -{ - /* - * Raw tracepoint data is a severe data leak, only allow root to - * have these. - */ - if ((counter->attr.sample_type & PERF_SAMPLE_RAW) && - perf_paranoid_tracepoint_raw() && - !capable(CAP_SYS_ADMIN)) - return ERR_PTR(-EPERM); - - if (ftrace_profile_enable(counter->attr.config)) - return NULL; - - counter->destroy = tp_perf_counter_destroy; - - return &perf_ops_generic; -} -#else -static const struct pmu *tp_perf_counter_init(struct perf_counter *counter) -{ - return NULL; -} -#endif - -atomic_t perf_swcounter_enabled[PERF_COUNT_SW_MAX]; - -static void sw_perf_counter_destroy(struct perf_counter *counter) -{ - u64 event_id = counter->attr.config; - - WARN_ON(counter->parent); - - atomic_dec(&perf_swcounter_enabled[event_id]); -} - -static const struct pmu *sw_perf_counter_init(struct perf_counter *counter) -{ - const struct pmu *pmu = NULL; - u64 event_id = counter->attr.config; - - /* - * Software counters (currently) can't in general distinguish - * between user, kernel and hypervisor events. - * However, context switches and cpu migrations are considered - * to be kernel events, and page faults are never hypervisor - * events. - */ - switch (event_id) { - case PERF_COUNT_SW_CPU_CLOCK: - pmu = &perf_ops_cpu_clock; - - break; - case PERF_COUNT_SW_TASK_CLOCK: - /* - * If the user instantiates this as a per-cpu counter, - * use the cpu_clock counter instead. - */ - if (counter->ctx->task) - pmu = &perf_ops_task_clock; - else - pmu = &perf_ops_cpu_clock; - - break; - case PERF_COUNT_SW_PAGE_FAULTS: - case PERF_COUNT_SW_PAGE_FAULTS_MIN: - case PERF_COUNT_SW_PAGE_FAULTS_MAJ: - case PERF_COUNT_SW_CONTEXT_SWITCHES: - case PERF_COUNT_SW_CPU_MIGRATIONS: - if (!counter->parent) { - atomic_inc(&perf_swcounter_enabled[event_id]); - counter->destroy = sw_perf_counter_destroy; - } - pmu = &perf_ops_generic; - break; - } - - return pmu; -} - -/* - * Allocate and initialize a counter structure - */ -static struct perf_counter * -perf_counter_alloc(struct perf_counter_attr *attr, - int cpu, - struct perf_counter_context *ctx, - struct perf_counter *group_leader, - struct perf_counter *parent_counter, - gfp_t gfpflags) -{ - const struct pmu *pmu; - struct perf_counter *counter; - struct hw_perf_counter *hwc; - long err; - - counter = kzalloc(sizeof(*counter), gfpflags); - if (!counter) - return ERR_PTR(-ENOMEM); - - /* - * Single counters are their own group leaders, with an - * empty sibling list: - */ - if (!group_leader) - group_leader = counter; - - mutex_init(&counter->child_mutex); - INIT_LIST_HEAD(&counter->child_list); - - INIT_LIST_HEAD(&counter->group_entry); - INIT_LIST_HEAD(&counter->event_entry); - INIT_LIST_HEAD(&counter->sibling_list); - init_waitqueue_head(&counter->waitq); - - mutex_init(&counter->mmap_mutex); - - counter->cpu = cpu; - counter->attr = *attr; - counter->group_leader = group_leader; - counter->pmu = NULL; - counter->ctx = ctx; - counter->oncpu = -1; - - counter->parent = parent_counter; - - counter->ns = get_pid_ns(current->nsproxy->pid_ns); - counter->id = atomic64_inc_return(&perf_counter_id); - - counter->state = PERF_COUNTER_STATE_INACTIVE; - - if (attr->disabled) - counter->state = PERF_COUNTER_STATE_OFF; - - pmu = NULL; - - hwc = &counter->hw; - hwc->sample_period = attr->sample_period; - if (attr->freq && attr->sample_freq) - hwc->sample_period = 1; - hwc->last_period = hwc->sample_period; - - atomic64_set(&hwc->period_left, hwc->sample_period); - - /* - * we currently do not support PERF_FORMAT_GROUP on inherited counters - */ - if (attr->inherit && (attr->read_format & PERF_FORMAT_GROUP)) - goto done; - - switch (attr->type) { - case PERF_TYPE_RAW: - case PERF_TYPE_HARDWARE: - case PERF_TYPE_HW_CACHE: - pmu = hw_perf_counter_init(counter); - break; - - case PERF_TYPE_SOFTWARE: - pmu = sw_perf_counter_init(counter); - break; - - case PERF_TYPE_TRACEPOINT: - pmu = tp_perf_counter_init(counter); - break; - - default: - break; - } -done: - err = 0; - if (!pmu) - err = -EINVAL; - else if (IS_ERR(pmu)) - err = PTR_ERR(pmu); - - if (err) { - if (counter->ns) - put_pid_ns(counter->ns); - kfree(counter); - return ERR_PTR(err); - } - - counter->pmu = pmu; - - if (!counter->parent) { - atomic_inc(&nr_counters); - if (counter->attr.mmap) - atomic_inc(&nr_mmap_counters); - if (counter->attr.comm) - atomic_inc(&nr_comm_counters); - if (counter->attr.task) - atomic_inc(&nr_task_counters); - } - - return counter; -} - -static int perf_copy_attr(struct perf_counter_attr __user *uattr, - struct perf_counter_attr *attr) -{ - u32 size; - int ret; - - if (!access_ok(VERIFY_WRITE, uattr, PERF_ATTR_SIZE_VER0)) - return -EFAULT; - - /* - * zero the full structure, so that a short copy will be nice. - */ - memset(attr, 0, sizeof(*attr)); - - ret = get_user(size, &uattr->size); - if (ret) - return ret; - - if (size > PAGE_SIZE) /* silly large */ - goto err_size; - - if (!size) /* abi compat */ - size = PERF_ATTR_SIZE_VER0; - - if (size < PERF_ATTR_SIZE_VER0) - goto err_size; - - /* - * If we're handed a bigger struct than we know of, - * ensure all the unknown bits are 0 - i.e. new - * user-space does not rely on any kernel feature - * extensions we dont know about yet. - */ - if (size > sizeof(*attr)) { - unsigned char __user *addr; - unsigned char __user *end; - unsigned char val; - - addr = (void __user *)uattr + sizeof(*attr); - end = (void __user *)uattr + size; - - for (; addr < end; addr++) { - ret = get_user(val, addr); - if (ret) - return ret; - if (val) - goto err_size; - } - size = sizeof(*attr); - } - - ret = copy_from_user(attr, uattr, size); - if (ret) - return -EFAULT; - - /* - * If the type exists, the corresponding creation will verify - * the attr->config. - */ - if (attr->type >= PERF_TYPE_MAX) - return -EINVAL; - - if (attr->__reserved_1 || attr->__reserved_2 || attr->__reserved_3) - return -EINVAL; - - if (attr->sample_type & ~(PERF_SAMPLE_MAX-1)) - return -EINVAL; - - if (attr->read_format & ~(PERF_FORMAT_MAX-1)) - return -EINVAL; - -out: - return ret; - -err_size: - put_user(sizeof(*attr), &uattr->size); - ret = -E2BIG; - goto out; -} - -int perf_counter_set_output(struct perf_counter *counter, int output_fd) -{ - struct perf_counter *output_counter = NULL; - struct file *output_file = NULL; - struct perf_counter *old_output; - int fput_needed = 0; - int ret = -EINVAL; - - if (!output_fd) - goto set; - - output_file = fget_light(output_fd, &fput_needed); - if (!output_file) - return -EBADF; - - if (output_file->f_op != &perf_fops) - goto out; - - output_counter = output_file->private_data; - - /* Don't chain output fds */ - if (output_counter->output) - goto out; - - /* Don't set an output fd when we already have an output channel */ - if (counter->data) - goto out; - - atomic_long_inc(&output_file->f_count); - -set: - mutex_lock(&counter->mmap_mutex); - old_output = counter->output; - rcu_assign_pointer(counter->output, output_counter); - mutex_unlock(&counter->mmap_mutex); - - if (old_output) { - /* - * we need to make sure no existing perf_output_*() - * is still referencing this counter. - */ - synchronize_rcu(); - fput(old_output->filp); - } - - ret = 0; -out: - fput_light(output_file, fput_needed); - return ret; -} - -/** - * sys_perf_counter_open - open a performance counter, associate it to a task/cpu - * - * @attr_uptr: event type attributes for monitoring/sampling - * @pid: target pid - * @cpu: target cpu - * @group_fd: group leader counter fd - */ -SYSCALL_DEFINE5(perf_counter_open, - struct perf_counter_attr __user *, attr_uptr, - pid_t, pid, int, cpu, int, group_fd, unsigned long, flags) -{ - struct perf_counter *counter, *group_leader; - struct perf_counter_attr attr; - struct perf_counter_context *ctx; - struct file *counter_file = NULL; - struct file *group_file = NULL; - int fput_needed = 0; - int fput_needed2 = 0; - int err; - - /* for future expandability... */ - if (flags & ~(PERF_FLAG_FD_NO_GROUP | PERF_FLAG_FD_OUTPUT)) - return -EINVAL; - - err = perf_copy_attr(attr_uptr, &attr); - if (err) - return err; - - if (!attr.exclude_kernel) { - if (perf_paranoid_kernel() && !capable(CAP_SYS_ADMIN)) - return -EACCES; - } - - if (attr.freq) { - if (attr.sample_freq > sysctl_perf_counter_sample_rate) - return -EINVAL; - } - - /* - * Get the target context (task or percpu): - */ - ctx = find_get_context(pid, cpu); - if (IS_ERR(ctx)) - return PTR_ERR(ctx); - - /* - * Look up the group leader (we will attach this counter to it): - */ - group_leader = NULL; - if (group_fd != -1 && !(flags & PERF_FLAG_FD_NO_GROUP)) { - err = -EINVAL; - group_file = fget_light(group_fd, &fput_needed); - if (!group_file) - goto err_put_context; - if (group_file->f_op != &perf_fops) - goto err_put_context; - - group_leader = group_file->private_data; - /* - * Do not allow a recursive hierarchy (this new sibling - * becoming part of another group-sibling): - */ - if (group_leader->group_leader != group_leader) - goto err_put_context; - /* - * Do not allow to attach to a group in a different - * task or CPU context: - */ - if (group_leader->ctx != ctx) - goto err_put_context; - /* - * Only a group leader can be exclusive or pinned - */ - if (attr.exclusive || attr.pinned) - goto err_put_context; - } - - counter = perf_counter_alloc(&attr, cpu, ctx, group_leader, - NULL, GFP_KERNEL); - err = PTR_ERR(counter); - if (IS_ERR(counter)) - goto err_put_context; - - err = anon_inode_getfd("[perf_counter]", &perf_fops, counter, 0); - if (err < 0) - goto err_free_put_context; - - counter_file = fget_light(err, &fput_needed2); - if (!counter_file) - goto err_free_put_context; - - if (flags & PERF_FLAG_FD_OUTPUT) { - err = perf_counter_set_output(counter, group_fd); - if (err) - goto err_fput_free_put_context; - } - - counter->filp = counter_file; - WARN_ON_ONCE(ctx->parent_ctx); - mutex_lock(&ctx->mutex); - perf_install_in_context(ctx, counter, cpu); - ++ctx->generation; - mutex_unlock(&ctx->mutex); - - counter->owner = current; - get_task_struct(current); - mutex_lock(¤t->perf_counter_mutex); - list_add_tail(&counter->owner_entry, ¤t->perf_counter_list); - mutex_unlock(¤t->perf_counter_mutex); - -err_fput_free_put_context: - fput_light(counter_file, fput_needed2); - -err_free_put_context: - if (err < 0) - kfree(counter); - -err_put_context: - if (err < 0) - put_ctx(ctx); - - fput_light(group_file, fput_needed); - - return err; -} - -/* - * inherit a counter from parent task to child task: - */ -static struct perf_counter * -inherit_counter(struct perf_counter *parent_counter, - struct task_struct *parent, - struct perf_counter_context *parent_ctx, - struct task_struct *child, - struct perf_counter *group_leader, - struct perf_counter_context *child_ctx) -{ - struct perf_counter *child_counter; - - /* - * Instead of creating recursive hierarchies of counters, - * we link inherited counters back to the original parent, - * which has a filp for sure, which we use as the reference - * count: - */ - if (parent_counter->parent) - parent_counter = parent_counter->parent; - - child_counter = perf_counter_alloc(&parent_counter->attr, - parent_counter->cpu, child_ctx, - group_leader, parent_counter, - GFP_KERNEL); - if (IS_ERR(child_counter)) - return child_counter; - get_ctx(child_ctx); - - /* - * Make the child state follow the state of the parent counter, - * not its attr.disabled bit. We hold the parent's mutex, - * so we won't race with perf_counter_{en, dis}able_family. - */ - if (parent_counter->state >= PERF_COUNTER_STATE_INACTIVE) - child_counter->state = PERF_COUNTER_STATE_INACTIVE; - else - child_counter->state = PERF_COUNTER_STATE_OFF; - - if (parent_counter->attr.freq) - child_counter->hw.sample_period = parent_counter->hw.sample_period; - - /* - * Link it up in the child's context: - */ - add_counter_to_ctx(child_counter, child_ctx); - - /* - * Get a reference to the parent filp - we will fput it - * when the child counter exits. This is safe to do because - * we are in the parent and we know that the filp still - * exists and has a nonzero count: - */ - atomic_long_inc(&parent_counter->filp->f_count); - - /* - * Link this into the parent counter's child list - */ - WARN_ON_ONCE(parent_counter->ctx->parent_ctx); - mutex_lock(&parent_counter->child_mutex); - list_add_tail(&child_counter->child_list, &parent_counter->child_list); - mutex_unlock(&parent_counter->child_mutex); - - return child_counter; -} - -static int inherit_group(struct perf_counter *parent_counter, - struct task_struct *parent, - struct perf_counter_context *parent_ctx, - struct task_struct *child, - struct perf_counter_context *child_ctx) -{ - struct perf_counter *leader; - struct perf_counter *sub; - struct perf_counter *child_ctr; - - leader = inherit_counter(parent_counter, parent, parent_ctx, - child, NULL, child_ctx); - if (IS_ERR(leader)) - return PTR_ERR(leader); - list_for_each_entry(sub, &parent_counter->sibling_list, group_entry) { - child_ctr = inherit_counter(sub, parent, parent_ctx, - child, leader, child_ctx); - if (IS_ERR(child_ctr)) - return PTR_ERR(child_ctr); - } - return 0; -} - -static void sync_child_counter(struct perf_counter *child_counter, - struct task_struct *child) -{ - struct perf_counter *parent_counter = child_counter->parent; - u64 child_val; - - if (child_counter->attr.inherit_stat) - perf_counter_read_event(child_counter, child); - - child_val = atomic64_read(&child_counter->count); - - /* - * Add back the child's count to the parent's count: - */ - atomic64_add(child_val, &parent_counter->count); - atomic64_add(child_counter->total_time_enabled, - &parent_counter->child_total_time_enabled); - atomic64_add(child_counter->total_time_running, - &parent_counter->child_total_time_running); - - /* - * Remove this counter from the parent's list - */ - WARN_ON_ONCE(parent_counter->ctx->parent_ctx); - mutex_lock(&parent_counter->child_mutex); - list_del_init(&child_counter->child_list); - mutex_unlock(&parent_counter->child_mutex); - - /* - * Release the parent counter, if this was the last - * reference to it. - */ - fput(parent_counter->filp); -} - -static void -__perf_counter_exit_task(struct perf_counter *child_counter, - struct perf_counter_context *child_ctx, - struct task_struct *child) -{ - struct perf_counter *parent_counter; - - update_counter_times(child_counter); - perf_counter_remove_from_context(child_counter); - - parent_counter = child_counter->parent; - /* - * It can happen that parent exits first, and has counters - * that are still around due to the child reference. These - * counters need to be zapped - but otherwise linger. - */ - if (parent_counter) { - sync_child_counter(child_counter, child); - free_counter(child_counter); - } -} - -/* - * When a child task exits, feed back counter values to parent counters. - */ -void perf_counter_exit_task(struct task_struct *child) -{ - struct perf_counter *child_counter, *tmp; - struct perf_counter_context *child_ctx; - unsigned long flags; - - if (likely(!child->perf_counter_ctxp)) { - perf_counter_task(child, NULL, 0); - return; - } - - local_irq_save(flags); - /* - * We can't reschedule here because interrupts are disabled, - * and either child is current or it is a task that can't be - * scheduled, so we are now safe from rescheduling changing - * our context. - */ - child_ctx = child->perf_counter_ctxp; - __perf_counter_task_sched_out(child_ctx); - - /* - * Take the context lock here so that if find_get_context is - * reading child->perf_counter_ctxp, we wait until it has - * incremented the context's refcount before we do put_ctx below. - */ - spin_lock(&child_ctx->lock); - child->perf_counter_ctxp = NULL; - /* - * If this context is a clone; unclone it so it can't get - * swapped to another process while we're removing all - * the counters from it. - */ - unclone_ctx(child_ctx); - spin_unlock_irqrestore(&child_ctx->lock, flags); - - /* - * Report the task dead after unscheduling the counters so that we - * won't get any samples after PERF_EVENT_EXIT. We can however still - * get a few PERF_EVENT_READ events. - */ - perf_counter_task(child, child_ctx, 0); - - /* - * We can recurse on the same lock type through: - * - * __perf_counter_exit_task() - * sync_child_counter() - * fput(parent_counter->filp) - * perf_release() - * mutex_lock(&ctx->mutex) - * - * But since its the parent context it won't be the same instance. - */ - mutex_lock_nested(&child_ctx->mutex, SINGLE_DEPTH_NESTING); - -again: - list_for_each_entry_safe(child_counter, tmp, &child_ctx->group_list, - group_entry) - __perf_counter_exit_task(child_counter, child_ctx, child); - - /* - * If the last counter was a group counter, it will have appended all - * its siblings to the list, but we obtained 'tmp' before that which - * will still point to the list head terminating the iteration. - */ - if (!list_empty(&child_ctx->group_list)) - goto again; - - mutex_unlock(&child_ctx->mutex); - - put_ctx(child_ctx); -} - -/* - * free an unexposed, unused context as created by inheritance by - * init_task below, used by fork() in case of fail. - */ -void perf_counter_free_task(struct task_struct *task) -{ - struct perf_counter_context *ctx = task->perf_counter_ctxp; - struct perf_counter *counter, *tmp; - - if (!ctx) - return; - - mutex_lock(&ctx->mutex); -again: - list_for_each_entry_safe(counter, tmp, &ctx->group_list, group_entry) { - struct perf_counter *parent = counter->parent; - - if (WARN_ON_ONCE(!parent)) - continue; - - mutex_lock(&parent->child_mutex); - list_del_init(&counter->child_list); - mutex_unlock(&parent->child_mutex); - - fput(parent->filp); - - list_del_counter(counter, ctx); - free_counter(counter); - } - - if (!list_empty(&ctx->group_list)) - goto again; - - mutex_unlock(&ctx->mutex); - - put_ctx(ctx); -} - -/* - * Initialize the perf_counter context in task_struct - */ -int perf_counter_init_task(struct task_struct *child) -{ - struct perf_counter_context *child_ctx, *parent_ctx; - struct perf_counter_context *cloned_ctx; - struct perf_counter *counter; - struct task_struct *parent = current; - int inherited_all = 1; - int ret = 0; - - child->perf_counter_ctxp = NULL; - - mutex_init(&child->perf_counter_mutex); - INIT_LIST_HEAD(&child->perf_counter_list); - - if (likely(!parent->perf_counter_ctxp)) - return 0; - - /* - * This is executed from the parent task context, so inherit - * counters that have been marked for cloning. - * First allocate and initialize a context for the child. - */ - - child_ctx = kmalloc(sizeof(struct perf_counter_context), GFP_KERNEL); - if (!child_ctx) - return -ENOMEM; - - __perf_counter_init_context(child_ctx, child); - child->perf_counter_ctxp = child_ctx; - get_task_struct(child); - - /* - * If the parent's context is a clone, pin it so it won't get - * swapped under us. - */ - parent_ctx = perf_pin_task_context(parent); - - /* - * No need to check if parent_ctx != NULL here; since we saw - * it non-NULL earlier, the only reason for it to become NULL - * is if we exit, and since we're currently in the middle of - * a fork we can't be exiting at the same time. - */ - - /* - * Lock the parent list. No need to lock the child - not PID - * hashed yet and not running, so nobody can access it. - */ - mutex_lock(&parent_ctx->mutex); - - /* - * We dont have to disable NMIs - we are only looking at - * the list, not manipulating it: - */ - list_for_each_entry_rcu(counter, &parent_ctx->event_list, event_entry) { - if (counter != counter->group_leader) - continue; - - if (!counter->attr.inherit) { - inherited_all = 0; - continue; - } - - ret = inherit_group(counter, parent, parent_ctx, - child, child_ctx); - if (ret) { - inherited_all = 0; - break; - } - } - - if (inherited_all) { - /* - * Mark the child context as a clone of the parent - * context, or of whatever the parent is a clone of. - * Note that if the parent is a clone, it could get - * uncloned at any point, but that doesn't matter - * because the list of counters and the generation - * count can't have changed since we took the mutex. - */ - cloned_ctx = rcu_dereference(parent_ctx->parent_ctx); - if (cloned_ctx) { - child_ctx->parent_ctx = cloned_ctx; - child_ctx->parent_gen = parent_ctx->parent_gen; - } else { - child_ctx->parent_ctx = parent_ctx; - child_ctx->parent_gen = parent_ctx->generation; - } - get_ctx(child_ctx->parent_ctx); - } - - mutex_unlock(&parent_ctx->mutex); - - perf_unpin_context(parent_ctx); - - return ret; -} - -static void __cpuinit perf_counter_init_cpu(int cpu) -{ - struct perf_cpu_context *cpuctx; - - cpuctx = &per_cpu(perf_cpu_context, cpu); - __perf_counter_init_context(&cpuctx->ctx, NULL); - - spin_lock(&perf_resource_lock); - cpuctx->max_pertask = perf_max_counters - perf_reserved_percpu; - spin_unlock(&perf_resource_lock); - - hw_perf_counter_setup(cpu); -} - -#ifdef CONFIG_HOTPLUG_CPU -static void __perf_counter_exit_cpu(void *info) -{ - struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); - struct perf_counter_context *ctx = &cpuctx->ctx; - struct perf_counter *counter, *tmp; - - list_for_each_entry_safe(counter, tmp, &ctx->group_list, group_entry) - __perf_counter_remove_from_context(counter); -} -static void perf_counter_exit_cpu(int cpu) -{ - struct perf_cpu_context *cpuctx = &per_cpu(perf_cpu_context, cpu); - struct perf_counter_context *ctx = &cpuctx->ctx; - - mutex_lock(&ctx->mutex); - smp_call_function_single(cpu, __perf_counter_exit_cpu, NULL, 1); - mutex_unlock(&ctx->mutex); -} -#else -static inline void perf_counter_exit_cpu(int cpu) { } -#endif - -static int __cpuinit -perf_cpu_notify(struct notifier_block *self, unsigned long action, void *hcpu) -{ - unsigned int cpu = (long)hcpu; - - switch (action) { - - case CPU_UP_PREPARE: - case CPU_UP_PREPARE_FROZEN: - perf_counter_init_cpu(cpu); - break; - - case CPU_ONLINE: - case CPU_ONLINE_FROZEN: - hw_perf_counter_setup_online(cpu); - break; - - case CPU_DOWN_PREPARE: - case CPU_DOWN_PREPARE_FROZEN: - perf_counter_exit_cpu(cpu); - break; - - default: - break; - } - - return NOTIFY_OK; -} - -/* - * This has to have a higher priority than migration_notifier in sched.c. - */ -static struct notifier_block __cpuinitdata perf_cpu_nb = { - .notifier_call = perf_cpu_notify, - .priority = 20, -}; - -void __init perf_counter_init(void) -{ - perf_cpu_notify(&perf_cpu_nb, (unsigned long)CPU_UP_PREPARE, - (void *)(long)smp_processor_id()); - perf_cpu_notify(&perf_cpu_nb, (unsigned long)CPU_ONLINE, - (void *)(long)smp_processor_id()); - register_cpu_notifier(&perf_cpu_nb); -} - -static ssize_t perf_show_reserve_percpu(struct sysdev_class *class, char *buf) -{ - return sprintf(buf, "%d\n", perf_reserved_percpu); -} - -static ssize_t -perf_set_reserve_percpu(struct sysdev_class *class, - const char *buf, - size_t count) -{ - struct perf_cpu_context *cpuctx; - unsigned long val; - int err, cpu, mpt; - - err = strict_strtoul(buf, 10, &val); - if (err) - return err; - if (val > perf_max_counters) - return -EINVAL; - - spin_lock(&perf_resource_lock); - perf_reserved_percpu = val; - for_each_online_cpu(cpu) { - cpuctx = &per_cpu(perf_cpu_context, cpu); - spin_lock_irq(&cpuctx->ctx.lock); - mpt = min(perf_max_counters - cpuctx->ctx.nr_counters, - perf_max_counters - perf_reserved_percpu); - cpuctx->max_pertask = mpt; - spin_unlock_irq(&cpuctx->ctx.lock); - } - spin_unlock(&perf_resource_lock); - - return count; -} - -static ssize_t perf_show_overcommit(struct sysdev_class *class, char *buf) -{ - return sprintf(buf, "%d\n", perf_overcommit); -} - -static ssize_t -perf_set_overcommit(struct sysdev_class *class, const char *buf, size_t count) -{ - unsigned long val; - int err; - - err = strict_strtoul(buf, 10, &val); - if (err) - return err; - if (val > 1) - return -EINVAL; - - spin_lock(&perf_resource_lock); - perf_overcommit = val; - spin_unlock(&perf_resource_lock); - - return count; -} - -static SYSDEV_CLASS_ATTR( - reserve_percpu, - 0644, - perf_show_reserve_percpu, - perf_set_reserve_percpu - ); - -static SYSDEV_CLASS_ATTR( - overcommit, - 0644, - perf_show_overcommit, - perf_set_overcommit - ); - -static struct attribute *perfclass_attrs[] = { - &attr_reserve_percpu.attr, - &attr_overcommit.attr, - NULL -}; - -static struct attribute_group perfclass_attr_group = { - .attrs = perfclass_attrs, - .name = "perf_counters", -}; - -static int __init perf_counter_sysfs_init(void) -{ - return sysfs_create_group(&cpu_sysdev_class.kset.kobj, - &perfclass_attr_group); -} -device_initcall(perf_counter_sysfs_init); diff --git a/kernel/perf_event.c b/kernel/perf_event.c new file mode 100644 index 0000000..6e8b99a --- /dev/null +++ b/kernel/perf_event.c @@ -0,0 +1,5000 @@ +/* + * Performance event core code + * + * Copyright (C) 2008 Thomas Gleixner + * Copyright (C) 2008-2009 Red Hat, Inc., Ingo Molnar + * Copyright (C) 2008-2009 Red Hat, Inc., Peter Zijlstra + * Copyright © 2009 Paul Mackerras, IBM Corp. + * + * For licensing details see kernel-base/COPYING + */ + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include + +/* + * Each CPU has a list of per CPU events: + */ +DEFINE_PER_CPU(struct perf_cpu_context, perf_cpu_context); + +int perf_max_events __read_mostly = 1; +static int perf_reserved_percpu __read_mostly; +static int perf_overcommit __read_mostly = 1; + +static atomic_t nr_events __read_mostly; +static atomic_t nr_mmap_events __read_mostly; +static atomic_t nr_comm_events __read_mostly; +static atomic_t nr_task_events __read_mostly; + +/* + * perf event paranoia level: + * -1 - not paranoid at all + * 0 - disallow raw tracepoint access for unpriv + * 1 - disallow cpu events for unpriv + * 2 - disallow kernel profiling for unpriv + */ +int sysctl_perf_event_paranoid __read_mostly = 1; + +static inline bool perf_paranoid_tracepoint_raw(void) +{ + return sysctl_perf_event_paranoid > -1; +} + +static inline bool perf_paranoid_cpu(void) +{ + return sysctl_perf_event_paranoid > 0; +} + +static inline bool perf_paranoid_kernel(void) +{ + return sysctl_perf_event_paranoid > 1; +} + +int sysctl_perf_event_mlock __read_mostly = 512; /* 'free' kb per user */ + +/* + * max perf event sample rate + */ +int sysctl_perf_event_sample_rate __read_mostly = 100000; + +static atomic64_t perf_event_id; + +/* + * Lock for (sysadmin-configurable) event reservations: + */ +static DEFINE_SPINLOCK(perf_resource_lock); + +/* + * Architecture provided APIs - weak aliases: + */ +extern __weak const struct pmu *hw_perf_event_init(struct perf_event *event) +{ + return NULL; +} + +void __weak hw_perf_disable(void) { barrier(); } +void __weak hw_perf_enable(void) { barrier(); } + +void __weak hw_perf_event_setup(int cpu) { barrier(); } +void __weak hw_perf_event_setup_online(int cpu) { barrier(); } + +int __weak +hw_perf_group_sched_in(struct perf_event *group_leader, + struct perf_cpu_context *cpuctx, + struct perf_event_context *ctx, int cpu) +{ + return 0; +} + +void __weak perf_event_print_debug(void) { } + +static DEFINE_PER_CPU(int, perf_disable_count); + +void __perf_disable(void) +{ + __get_cpu_var(perf_disable_count)++; +} + +bool __perf_enable(void) +{ + return !--__get_cpu_var(perf_disable_count); +} + +void perf_disable(void) +{ + __perf_disable(); + hw_perf_disable(); +} + +void perf_enable(void) +{ + if (__perf_enable()) + hw_perf_enable(); +} + +static void get_ctx(struct perf_event_context *ctx) +{ + WARN_ON(!atomic_inc_not_zero(&ctx->refcount)); +} + +static void free_ctx(struct rcu_head *head) +{ + struct perf_event_context *ctx; + + ctx = container_of(head, struct perf_event_context, rcu_head); + kfree(ctx); +} + +static void put_ctx(struct perf_event_context *ctx) +{ + if (atomic_dec_and_test(&ctx->refcount)) { + if (ctx->parent_ctx) + put_ctx(ctx->parent_ctx); + if (ctx->task) + put_task_struct(ctx->task); + call_rcu(&ctx->rcu_head, free_ctx); + } +} + +static void unclone_ctx(struct perf_event_context *ctx) +{ + if (ctx->parent_ctx) { + put_ctx(ctx->parent_ctx); + ctx->parent_ctx = NULL; + } +} + +/* + * If we inherit events we want to return the parent event id + * to userspace. + */ +static u64 primary_event_id(struct perf_event *event) +{ + u64 id = event->id; + + if (event->parent) + id = event->parent->id; + + return id; +} + +/* + * Get the perf_event_context for a task and lock it. + * This has to cope with with the fact that until it is locked, + * the context could get moved to another task. + */ +static struct perf_event_context * +perf_lock_task_context(struct task_struct *task, unsigned long *flags) +{ + struct perf_event_context *ctx; + + rcu_read_lock(); + retry: + ctx = rcu_dereference(task->perf_event_ctxp); + if (ctx) { + /* + * If this context is a clone of another, it might + * get swapped for another underneath us by + * perf_event_task_sched_out, though the + * rcu_read_lock() protects us from any context + * getting freed. Lock the context and check if it + * got swapped before we could get the lock, and retry + * if so. If we locked the right context, then it + * can't get swapped on us any more. + */ + spin_lock_irqsave(&ctx->lock, *flags); + if (ctx != rcu_dereference(task->perf_event_ctxp)) { + spin_unlock_irqrestore(&ctx->lock, *flags); + goto retry; + } + + if (!atomic_inc_not_zero(&ctx->refcount)) { + spin_unlock_irqrestore(&ctx->lock, *flags); + ctx = NULL; + } + } + rcu_read_unlock(); + return ctx; +} + +/* + * Get the context for a task and increment its pin_count so it + * can't get swapped to another task. This also increments its + * reference count so that the context can't get freed. + */ +static struct perf_event_context *perf_pin_task_context(struct task_struct *task) +{ + struct perf_event_context *ctx; + unsigned long flags; + + ctx = perf_lock_task_context(task, &flags); + if (ctx) { + ++ctx->pin_count; + spin_unlock_irqrestore(&ctx->lock, flags); + } + return ctx; +} + +static void perf_unpin_context(struct perf_event_context *ctx) +{ + unsigned long flags; + + spin_lock_irqsave(&ctx->lock, flags); + --ctx->pin_count; + spin_unlock_irqrestore(&ctx->lock, flags); + put_ctx(ctx); +} + +/* + * Add a event from the lists for its context. + * Must be called with ctx->mutex and ctx->lock held. + */ +static void +list_add_event(struct perf_event *event, struct perf_event_context *ctx) +{ + struct perf_event *group_leader = event->group_leader; + + /* + * Depending on whether it is a standalone or sibling event, + * add it straight to the context's event list, or to the group + * leader's sibling list: + */ + if (group_leader == event) + list_add_tail(&event->group_entry, &ctx->group_list); + else { + list_add_tail(&event->group_entry, &group_leader->sibling_list); + group_leader->nr_siblings++; + } + + list_add_rcu(&event->event_entry, &ctx->event_list); + ctx->nr_events++; + if (event->attr.inherit_stat) + ctx->nr_stat++; +} + +/* + * Remove a event from the lists for its context. + * Must be called with ctx->mutex and ctx->lock held. + */ +static void +list_del_event(struct perf_event *event, struct perf_event_context *ctx) +{ + struct perf_event *sibling, *tmp; + + if (list_empty(&event->group_entry)) + return; + ctx->nr_events--; + if (event->attr.inherit_stat) + ctx->nr_stat--; + + list_del_init(&event->group_entry); + list_del_rcu(&event->event_entry); + + if (event->group_leader != event) + event->group_leader->nr_siblings--; + + /* + * If this was a group event with sibling events then + * upgrade the siblings to singleton events by adding them + * to the context list directly: + */ + list_for_each_entry_safe(sibling, tmp, &event->sibling_list, group_entry) { + + list_move_tail(&sibling->group_entry, &ctx->group_list); + sibling->group_leader = sibling; + } +} + +static void +event_sched_out(struct perf_event *event, + struct perf_cpu_context *cpuctx, + struct perf_event_context *ctx) +{ + if (event->state != PERF_EVENT_STATE_ACTIVE) + return; + + event->state = PERF_EVENT_STATE_INACTIVE; + if (event->pending_disable) { + event->pending_disable = 0; + event->state = PERF_EVENT_STATE_OFF; + } + event->tstamp_stopped = ctx->time; + event->pmu->disable(event); + event->oncpu = -1; + + if (!is_software_event(event)) + cpuctx->active_oncpu--; + ctx->nr_active--; + if (event->attr.exclusive || !cpuctx->active_oncpu) + cpuctx->exclusive = 0; +} + +static void +group_sched_out(struct perf_event *group_event, + struct perf_cpu_context *cpuctx, + struct perf_event_context *ctx) +{ + struct perf_event *event; + + if (group_event->state != PERF_EVENT_STATE_ACTIVE) + return; + + event_sched_out(group_event, cpuctx, ctx); + + /* + * Schedule out siblings (if any): + */ + list_for_each_entry(event, &group_event->sibling_list, group_entry) + event_sched_out(event, cpuctx, ctx); + + if (group_event->attr.exclusive) + cpuctx->exclusive = 0; +} + +/* + * Cross CPU call to remove a performance event + * + * We disable the event on the hardware level first. After that we + * remove it from the context list. + */ +static void __perf_event_remove_from_context(void *info) +{ + struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); + struct perf_event *event = info; + struct perf_event_context *ctx = event->ctx; + + /* + * If this is a task context, we need to check whether it is + * the current task context of this cpu. If not it has been + * scheduled out before the smp call arrived. + */ + if (ctx->task && cpuctx->task_ctx != ctx) + return; + + spin_lock(&ctx->lock); + /* + * Protect the list operation against NMI by disabling the + * events on a global level. + */ + perf_disable(); + + event_sched_out(event, cpuctx, ctx); + + list_del_event(event, ctx); + + if (!ctx->task) { + /* + * Allow more per task events with respect to the + * reservation: + */ + cpuctx->max_pertask = + min(perf_max_events - ctx->nr_events, + perf_max_events - perf_reserved_percpu); + } + + perf_enable(); + spin_unlock(&ctx->lock); +} + + +/* + * Remove the event from a task's (or a CPU's) list of events. + * + * Must be called with ctx->mutex held. + * + * CPU events are removed with a smp call. For task events we only + * call when the task is on a CPU. + * + * If event->ctx is a cloned context, callers must make sure that + * every task struct that event->ctx->task could possibly point to + * remains valid. This is OK when called from perf_release since + * that only calls us on the top-level context, which can't be a clone. + * When called from perf_event_exit_task, it's OK because the + * context has been detached from its task. + */ +static void perf_event_remove_from_context(struct perf_event *event) +{ + struct perf_event_context *ctx = event->ctx; + struct task_struct *task = ctx->task; + + if (!task) { + /* + * Per cpu events are removed via an smp call and + * the removal is always sucessful. + */ + smp_call_function_single(event->cpu, + __perf_event_remove_from_context, + event, 1); + return; + } + +retry: + task_oncpu_function_call(task, __perf_event_remove_from_context, + event); + + spin_lock_irq(&ctx->lock); + /* + * If the context is active we need to retry the smp call. + */ + if (ctx->nr_active && !list_empty(&event->group_entry)) { + spin_unlock_irq(&ctx->lock); + goto retry; + } + + /* + * The lock prevents that this context is scheduled in so we + * can remove the event safely, if the call above did not + * succeed. + */ + if (!list_empty(&event->group_entry)) { + list_del_event(event, ctx); + } + spin_unlock_irq(&ctx->lock); +} + +static inline u64 perf_clock(void) +{ + return cpu_clock(smp_processor_id()); +} + +/* + * Update the record of the current time in a context. + */ +static void update_context_time(struct perf_event_context *ctx) +{ + u64 now = perf_clock(); + + ctx->time += now - ctx->timestamp; + ctx->timestamp = now; +} + +/* + * Update the total_time_enabled and total_time_running fields for a event. + */ +static void update_event_times(struct perf_event *event) +{ + struct perf_event_context *ctx = event->ctx; + u64 run_end; + + if (event->state < PERF_EVENT_STATE_INACTIVE || + event->group_leader->state < PERF_EVENT_STATE_INACTIVE) + return; + + event->total_time_enabled = ctx->time - event->tstamp_enabled; + + if (event->state == PERF_EVENT_STATE_INACTIVE) + run_end = event->tstamp_stopped; + else + run_end = ctx->time; + + event->total_time_running = run_end - event->tstamp_running; +} + +/* + * Update total_time_enabled and total_time_running for all events in a group. + */ +static void update_group_times(struct perf_event *leader) +{ + struct perf_event *event; + + update_event_times(leader); + list_for_each_entry(event, &leader->sibling_list, group_entry) + update_event_times(event); +} + +/* + * Cross CPU call to disable a performance event + */ +static void __perf_event_disable(void *info) +{ + struct perf_event *event = info; + struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); + struct perf_event_context *ctx = event->ctx; + + /* + * If this is a per-task event, need to check whether this + * event's task is the current task on this cpu. + */ + if (ctx->task && cpuctx->task_ctx != ctx) + return; + + spin_lock(&ctx->lock); + + /* + * If the event is on, turn it off. + * If it is in error state, leave it in error state. + */ + if (event->state >= PERF_EVENT_STATE_INACTIVE) { + update_context_time(ctx); + update_group_times(event); + if (event == event->group_leader) + group_sched_out(event, cpuctx, ctx); + else + event_sched_out(event, cpuctx, ctx); + event->state = PERF_EVENT_STATE_OFF; + } + + spin_unlock(&ctx->lock); +} + +/* + * Disable a event. + * + * If event->ctx is a cloned context, callers must make sure that + * every task struct that event->ctx->task could possibly point to + * remains valid. This condition is satisifed when called through + * perf_event_for_each_child or perf_event_for_each because they + * hold the top-level event's child_mutex, so any descendant that + * goes to exit will block in sync_child_event. + * When called from perf_pending_event it's OK because event->ctx + * is the current context on this CPU and preemption is disabled, + * hence we can't get into perf_event_task_sched_out for this context. + */ +static void perf_event_disable(struct perf_event *event) +{ + struct perf_event_context *ctx = event->ctx; + struct task_struct *task = ctx->task; + + if (!task) { + /* + * Disable the event on the cpu that it's on + */ + smp_call_function_single(event->cpu, __perf_event_disable, + event, 1); + return; + } + + retry: + task_oncpu_function_call(task, __perf_event_disable, event); + + spin_lock_irq(&ctx->lock); + /* + * If the event is still active, we need to retry the cross-call. + */ + if (event->state == PERF_EVENT_STATE_ACTIVE) { + spin_unlock_irq(&ctx->lock); + goto retry; + } + + /* + * Since we have the lock this context can't be scheduled + * in, so we can change the state safely. + */ + if (event->state == PERF_EVENT_STATE_INACTIVE) { + update_group_times(event); + event->state = PERF_EVENT_STATE_OFF; + } + + spin_unlock_irq(&ctx->lock); +} + +static int +event_sched_in(struct perf_event *event, + struct perf_cpu_context *cpuctx, + struct perf_event_context *ctx, + int cpu) +{ + if (event->state <= PERF_EVENT_STATE_OFF) + return 0; + + event->state = PERF_EVENT_STATE_ACTIVE; + event->oncpu = cpu; /* TODO: put 'cpu' into cpuctx->cpu */ + /* + * The new state must be visible before we turn it on in the hardware: + */ + smp_wmb(); + + if (event->pmu->enable(event)) { + event->state = PERF_EVENT_STATE_INACTIVE; + event->oncpu = -1; + return -EAGAIN; + } + + event->tstamp_running += ctx->time - event->tstamp_stopped; + + if (!is_software_event(event)) + cpuctx->active_oncpu++; + ctx->nr_active++; + + if (event->attr.exclusive) + cpuctx->exclusive = 1; + + return 0; +} + +static int +group_sched_in(struct perf_event *group_event, + struct perf_cpu_context *cpuctx, + struct perf_event_context *ctx, + int cpu) +{ + struct perf_event *event, *partial_group; + int ret; + + if (group_event->state == PERF_EVENT_STATE_OFF) + return 0; + + ret = hw_perf_group_sched_in(group_event, cpuctx, ctx, cpu); + if (ret) + return ret < 0 ? ret : 0; + + if (event_sched_in(group_event, cpuctx, ctx, cpu)) + return -EAGAIN; + + /* + * Schedule in siblings as one group (if any): + */ + list_for_each_entry(event, &group_event->sibling_list, group_entry) { + if (event_sched_in(event, cpuctx, ctx, cpu)) { + partial_group = event; + goto group_error; + } + } + + return 0; + +group_error: + /* + * Groups can be scheduled in as one unit only, so undo any + * partial group before returning: + */ + list_for_each_entry(event, &group_event->sibling_list, group_entry) { + if (event == partial_group) + break; + event_sched_out(event, cpuctx, ctx); + } + event_sched_out(group_event, cpuctx, ctx); + + return -EAGAIN; +} + +/* + * Return 1 for a group consisting entirely of software events, + * 0 if the group contains any hardware events. + */ +static int is_software_only_group(struct perf_event *leader) +{ + struct perf_event *event; + + if (!is_software_event(leader)) + return 0; + + list_for_each_entry(event, &leader->sibling_list, group_entry) + if (!is_software_event(event)) + return 0; + + return 1; +} + +/* + * Work out whether we can put this event group on the CPU now. + */ +static int group_can_go_on(struct perf_event *event, + struct perf_cpu_context *cpuctx, + int can_add_hw) +{ + /* + * Groups consisting entirely of software events can always go on. + */ + if (is_software_only_group(event)) + return 1; + /* + * If an exclusive group is already on, no other hardware + * events can go on. + */ + if (cpuctx->exclusive) + return 0; + /* + * If this group is exclusive and there are already + * events on the CPU, it can't go on. + */ + if (event->attr.exclusive && cpuctx->active_oncpu) + return 0; + /* + * Otherwise, try to add it if all previous groups were able + * to go on. + */ + return can_add_hw; +} + +static void add_event_to_ctx(struct perf_event *event, + struct perf_event_context *ctx) +{ + list_add_event(event, ctx); + event->tstamp_enabled = ctx->time; + event->tstamp_running = ctx->time; + event->tstamp_stopped = ctx->time; +} + +/* + * Cross CPU call to install and enable a performance event + * + * Must be called with ctx->mutex held + */ +static void __perf_install_in_context(void *info) +{ + struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); + struct perf_event *event = info; + struct perf_event_context *ctx = event->ctx; + struct perf_event *leader = event->group_leader; + int cpu = smp_processor_id(); + int err; + + /* + * If this is a task context, we need to check whether it is + * the current task context of this cpu. If not it has been + * scheduled out before the smp call arrived. + * Or possibly this is the right context but it isn't + * on this cpu because it had no events. + */ + if (ctx->task && cpuctx->task_ctx != ctx) { + if (cpuctx->task_ctx || ctx->task != current) + return; + cpuctx->task_ctx = ctx; + } + + spin_lock(&ctx->lock); + ctx->is_active = 1; + update_context_time(ctx); + + /* + * Protect the list operation against NMI by disabling the + * events on a global level. NOP for non NMI based events. + */ + perf_disable(); + + add_event_to_ctx(event, ctx); + + /* + * Don't put the event on if it is disabled or if + * it is in a group and the group isn't on. + */ + if (event->state != PERF_EVENT_STATE_INACTIVE || + (leader != event && leader->state != PERF_EVENT_STATE_ACTIVE)) + goto unlock; + + /* + * An exclusive event can't go on if there are already active + * hardware events, and no hardware event can go on if there + * is already an exclusive event on. + */ + if (!group_can_go_on(event, cpuctx, 1)) + err = -EEXIST; + else + err = event_sched_in(event, cpuctx, ctx, cpu); + + if (err) { + /* + * This event couldn't go on. If it is in a group + * then we have to pull the whole group off. + * If the event group is pinned then put it in error state. + */ + if (leader != event) + group_sched_out(leader, cpuctx, ctx); + if (leader->attr.pinned) { + update_group_times(leader); + leader->state = PERF_EVENT_STATE_ERROR; + } + } + + if (!err && !ctx->task && cpuctx->max_pertask) + cpuctx->max_pertask--; + + unlock: + perf_enable(); + + spin_unlock(&ctx->lock); +} + +/* + * Attach a performance event to a context + * + * First we add the event to the list with the hardware enable bit + * in event->hw_config cleared. + * + * If the event is attached to a task which is on a CPU we use a smp + * call to enable it in the task context. The task might have been + * scheduled away, but we check this in the smp call again. + * + * Must be called with ctx->mutex held. + */ +static void +perf_install_in_context(struct perf_event_context *ctx, + struct perf_event *event, + int cpu) +{ + struct task_struct *task = ctx->task; + + if (!task) { + /* + * Per cpu events are installed via an smp call and + * the install is always sucessful. + */ + smp_call_function_single(cpu, __perf_install_in_context, + event, 1); + return; + } + +retry: + task_oncpu_function_call(task, __perf_install_in_context, + event); + + spin_lock_irq(&ctx->lock); + /* + * we need to retry the smp call. + */ + if (ctx->is_active && list_empty(&event->group_entry)) { + spin_unlock_irq(&ctx->lock); + goto retry; + } + + /* + * The lock prevents that this context is scheduled in so we + * can add the event safely, if it the call above did not + * succeed. + */ + if (list_empty(&event->group_entry)) + add_event_to_ctx(event, ctx); + spin_unlock_irq(&ctx->lock); +} + +/* + * Put a event into inactive state and update time fields. + * Enabling the leader of a group effectively enables all + * the group members that aren't explicitly disabled, so we + * have to update their ->tstamp_enabled also. + * Note: this works for group members as well as group leaders + * since the non-leader members' sibling_lists will be empty. + */ +static void __perf_event_mark_enabled(struct perf_event *event, + struct perf_event_context *ctx) +{ + struct perf_event *sub; + + event->state = PERF_EVENT_STATE_INACTIVE; + event->tstamp_enabled = ctx->time - event->total_time_enabled; + list_for_each_entry(sub, &event->sibling_list, group_entry) + if (sub->state >= PERF_EVENT_STATE_INACTIVE) + sub->tstamp_enabled = + ctx->time - sub->total_time_enabled; +} + +/* + * Cross CPU call to enable a performance event + */ +static void __perf_event_enable(void *info) +{ + struct perf_event *event = info; + struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); + struct perf_event_context *ctx = event->ctx; + struct perf_event *leader = event->group_leader; + int err; + + /* + * If this is a per-task event, need to check whether this + * event's task is the current task on this cpu. + */ + if (ctx->task && cpuctx->task_ctx != ctx) { + if (cpuctx->task_ctx || ctx->task != current) + return; + cpuctx->task_ctx = ctx; + } + + spin_lock(&ctx->lock); + ctx->is_active = 1; + update_context_time(ctx); + + if (event->state >= PERF_EVENT_STATE_INACTIVE) + goto unlock; + __perf_event_mark_enabled(event, ctx); + + /* + * If the event is in a group and isn't the group leader, + * then don't put it on unless the group is on. + */ + if (leader != event && leader->state != PERF_EVENT_STATE_ACTIVE) + goto unlock; + + if (!group_can_go_on(event, cpuctx, 1)) { + err = -EEXIST; + } else { + perf_disable(); + if (event == leader) + err = group_sched_in(event, cpuctx, ctx, + smp_processor_id()); + else + err = event_sched_in(event, cpuctx, ctx, + smp_processor_id()); + perf_enable(); + } + + if (err) { + /* + * If this event can't go on and it's part of a + * group, then the whole group has to come off. + */ + if (leader != event) + group_sched_out(leader, cpuctx, ctx); + if (leader->attr.pinned) { + update_group_times(leader); + leader->state = PERF_EVENT_STATE_ERROR; + } + } + + unlock: + spin_unlock(&ctx->lock); +} + +/* + * Enable a event. + * + * If event->ctx is a cloned context, callers must make sure that + * every task struct that event->ctx->task could possibly point to + * remains valid. This condition is satisfied when called through + * perf_event_for_each_child or perf_event_for_each as described + * for perf_event_disable. + */ +static void perf_event_enable(struct perf_event *event) +{ + struct perf_event_context *ctx = event->ctx; + struct task_struct *task = ctx->task; + + if (!task) { + /* + * Enable the event on the cpu that it's on + */ + smp_call_function_single(event->cpu, __perf_event_enable, + event, 1); + return; + } + + spin_lock_irq(&ctx->lock); + if (event->state >= PERF_EVENT_STATE_INACTIVE) + goto out; + + /* + * If the event is in error state, clear that first. + * That way, if we see the event in error state below, we + * know that it has gone back into error state, as distinct + * from the task having been scheduled away before the + * cross-call arrived. + */ + if (event->state == PERF_EVENT_STATE_ERROR) + event->state = PERF_EVENT_STATE_OFF; + + retry: + spin_unlock_irq(&ctx->lock); + task_oncpu_function_call(task, __perf_event_enable, event); + + spin_lock_irq(&ctx->lock); + + /* + * If the context is active and the event is still off, + * we need to retry the cross-call. + */ + if (ctx->is_active && event->state == PERF_EVENT_STATE_OFF) + goto retry; + + /* + * Since we have the lock this context can't be scheduled + * in, so we can change the state safely. + */ + if (event->state == PERF_EVENT_STATE_OFF) + __perf_event_mark_enabled(event, ctx); + + out: + spin_unlock_irq(&ctx->lock); +} + +static int perf_event_refresh(struct perf_event *event, int refresh) +{ + /* + * not supported on inherited events + */ + if (event->attr.inherit) + return -EINVAL; + + atomic_add(refresh, &event->event_limit); + perf_event_enable(event); + + return 0; +} + +void __perf_event_sched_out(struct perf_event_context *ctx, + struct perf_cpu_context *cpuctx) +{ + struct perf_event *event; + + spin_lock(&ctx->lock); + ctx->is_active = 0; + if (likely(!ctx->nr_events)) + goto out; + update_context_time(ctx); + + perf_disable(); + if (ctx->nr_active) { + list_for_each_entry(event, &ctx->group_list, group_entry) { + if (event != event->group_leader) + event_sched_out(event, cpuctx, ctx); + else + group_sched_out(event, cpuctx, ctx); + } + } + perf_enable(); + out: + spin_unlock(&ctx->lock); +} + +/* + * Test whether two contexts are equivalent, i.e. whether they + * have both been cloned from the same version of the same context + * and they both have the same number of enabled events. + * If the number of enabled events is the same, then the set + * of enabled events should be the same, because these are both + * inherited contexts, therefore we can't access individual events + * in them directly with an fd; we can only enable/disable all + * events via prctl, or enable/disable all events in a family + * via ioctl, which will have the same effect on both contexts. + */ +static int context_equiv(struct perf_event_context *ctx1, + struct perf_event_context *ctx2) +{ + return ctx1->parent_ctx && ctx1->parent_ctx == ctx2->parent_ctx + && ctx1->parent_gen == ctx2->parent_gen + && !ctx1->pin_count && !ctx2->pin_count; +} + +static void __perf_event_read(void *event); + +static void __perf_event_sync_stat(struct perf_event *event, + struct perf_event *next_event) +{ + u64 value; + + if (!event->attr.inherit_stat) + return; + + /* + * Update the event value, we cannot use perf_event_read() + * because we're in the middle of a context switch and have IRQs + * disabled, which upsets smp_call_function_single(), however + * we know the event must be on the current CPU, therefore we + * don't need to use it. + */ + switch (event->state) { + case PERF_EVENT_STATE_ACTIVE: + __perf_event_read(event); + break; + + case PERF_EVENT_STATE_INACTIVE: + update_event_times(event); + break; + + default: + break; + } + + /* + * In order to keep per-task stats reliable we need to flip the event + * values when we flip the contexts. + */ + value = atomic64_read(&next_event->count); + value = atomic64_xchg(&event->count, value); + atomic64_set(&next_event->count, value); + + swap(event->total_time_enabled, next_event->total_time_enabled); + swap(event->total_time_running, next_event->total_time_running); + + /* + * Since we swizzled the values, update the user visible data too. + */ + perf_event_update_userpage(event); + perf_event_update_userpage(next_event); +} + +#define list_next_entry(pos, member) \ + list_entry(pos->member.next, typeof(*pos), member) + +static void perf_event_sync_stat(struct perf_event_context *ctx, + struct perf_event_context *next_ctx) +{ + struct perf_event *event, *next_event; + + if (!ctx->nr_stat) + return; + + event = list_first_entry(&ctx->event_list, + struct perf_event, event_entry); + + next_event = list_first_entry(&next_ctx->event_list, + struct perf_event, event_entry); + + while (&event->event_entry != &ctx->event_list && + &next_event->event_entry != &next_ctx->event_list) { + + __perf_event_sync_stat(event, next_event); + + event = list_next_entry(event, event_entry); + next_event = list_next_entry(next_event, event_entry); + } +} + +/* + * Called from scheduler to remove the events of the current task, + * with interrupts disabled. + * + * We stop each event and update the event value in event->count. + * + * This does not protect us against NMI, but disable() + * sets the disabled bit in the control field of event _before_ + * accessing the event control register. If a NMI hits, then it will + * not restart the event. + */ +void perf_event_task_sched_out(struct task_struct *task, + struct task_struct *next, int cpu) +{ + struct perf_cpu_context *cpuctx = &per_cpu(perf_cpu_context, cpu); + struct perf_event_context *ctx = task->perf_event_ctxp; + struct perf_event_context *next_ctx; + struct perf_event_context *parent; + struct pt_regs *regs; + int do_switch = 1; + + regs = task_pt_regs(task); + perf_sw_event(PERF_COUNT_SW_CONTEXT_SWITCHES, 1, 1, regs, 0); + + if (likely(!ctx || !cpuctx->task_ctx)) + return; + + update_context_time(ctx); + + rcu_read_lock(); + parent = rcu_dereference(ctx->parent_ctx); + next_ctx = next->perf_event_ctxp; + if (parent && next_ctx && + rcu_dereference(next_ctx->parent_ctx) == parent) { + /* + * Looks like the two contexts are clones, so we might be + * able to optimize the context switch. We lock both + * contexts and check that they are clones under the + * lock (including re-checking that neither has been + * uncloned in the meantime). It doesn't matter which + * order we take the locks because no other cpu could + * be trying to lock both of these tasks. + */ + spin_lock(&ctx->lock); + spin_lock_nested(&next_ctx->lock, SINGLE_DEPTH_NESTING); + if (context_equiv(ctx, next_ctx)) { + /* + * XXX do we need a memory barrier of sorts + * wrt to rcu_dereference() of perf_event_ctxp + */ + task->perf_event_ctxp = next_ctx; + next->perf_event_ctxp = ctx; + ctx->task = next; + next_ctx->task = task; + do_switch = 0; + + perf_event_sync_stat(ctx, next_ctx); + } + spin_unlock(&next_ctx->lock); + spin_unlock(&ctx->lock); + } + rcu_read_unlock(); + + if (do_switch) { + __perf_event_sched_out(ctx, cpuctx); + cpuctx->task_ctx = NULL; + } +} + +/* + * Called with IRQs disabled + */ +static void __perf_event_task_sched_out(struct perf_event_context *ctx) +{ + struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); + + if (!cpuctx->task_ctx) + return; + + if (WARN_ON_ONCE(ctx != cpuctx->task_ctx)) + return; + + __perf_event_sched_out(ctx, cpuctx); + cpuctx->task_ctx = NULL; +} + +/* + * Called with IRQs disabled + */ +static void perf_event_cpu_sched_out(struct perf_cpu_context *cpuctx) +{ + __perf_event_sched_out(&cpuctx->ctx, cpuctx); +} + +static void +__perf_event_sched_in(struct perf_event_context *ctx, + struct perf_cpu_context *cpuctx, int cpu) +{ + struct perf_event *event; + int can_add_hw = 1; + + spin_lock(&ctx->lock); + ctx->is_active = 1; + if (likely(!ctx->nr_events)) + goto out; + + ctx->timestamp = perf_clock(); + + perf_disable(); + + /* + * First go through the list and put on any pinned groups + * in order to give them the best chance of going on. + */ + list_for_each_entry(event, &ctx->group_list, group_entry) { + if (event->state <= PERF_EVENT_STATE_OFF || + !event->attr.pinned) + continue; + if (event->cpu != -1 && event->cpu != cpu) + continue; + + if (event != event->group_leader) + event_sched_in(event, cpuctx, ctx, cpu); + else { + if (group_can_go_on(event, cpuctx, 1)) + group_sched_in(event, cpuctx, ctx, cpu); + } + + /* + * If this pinned group hasn't been scheduled, + * put it in error state. + */ + if (event->state == PERF_EVENT_STATE_INACTIVE) { + update_group_times(event); + event->state = PERF_EVENT_STATE_ERROR; + } + } + + list_for_each_entry(event, &ctx->group_list, group_entry) { + /* + * Ignore events in OFF or ERROR state, and + * ignore pinned events since we did them already. + */ + if (event->state <= PERF_EVENT_STATE_OFF || + event->attr.pinned) + continue; + + /* + * Listen to the 'cpu' scheduling filter constraint + * of events: + */ + if (event->cpu != -1 && event->cpu != cpu) + continue; + + if (event != event->group_leader) { + if (event_sched_in(event, cpuctx, ctx, cpu)) + can_add_hw = 0; + } else { + if (group_can_go_on(event, cpuctx, can_add_hw)) { + if (group_sched_in(event, cpuctx, ctx, cpu)) + can_add_hw = 0; + } + } + } + perf_enable(); + out: + spin_unlock(&ctx->lock); +} + +/* + * Called from scheduler to add the events of the current task + * with interrupts disabled. + * + * We restore the event value and then enable it. + * + * This does not protect us against NMI, but enable() + * sets the enabled bit in the control field of event _before_ + * accessing the event control register. If a NMI hits, then it will + * keep the event running. + */ +void perf_event_task_sched_in(struct task_struct *task, int cpu) +{ + struct perf_cpu_context *cpuctx = &per_cpu(perf_cpu_context, cpu); + struct perf_event_context *ctx = task->perf_event_ctxp; + + if (likely(!ctx)) + return; + if (cpuctx->task_ctx == ctx) + return; + __perf_event_sched_in(ctx, cpuctx, cpu); + cpuctx->task_ctx = ctx; +} + +static void perf_event_cpu_sched_in(struct perf_cpu_context *cpuctx, int cpu) +{ + struct perf_event_context *ctx = &cpuctx->ctx; + + __perf_event_sched_in(ctx, cpuctx, cpu); +} + +#define MAX_INTERRUPTS (~0ULL) + +static void perf_log_throttle(struct perf_event *event, int enable); + +static void perf_adjust_period(struct perf_event *event, u64 events) +{ + struct hw_perf_event *hwc = &event->hw; + u64 period, sample_period; + s64 delta; + + events *= hwc->sample_period; + period = div64_u64(events, event->attr.sample_freq); + + delta = (s64)(period - hwc->sample_period); + delta = (delta + 7) / 8; /* low pass filter */ + + sample_period = hwc->sample_period + delta; + + if (!sample_period) + sample_period = 1; + + hwc->sample_period = sample_period; +} + +static void perf_ctx_adjust_freq(struct perf_event_context *ctx) +{ + struct perf_event *event; + struct hw_perf_event *hwc; + u64 interrupts, freq; + + spin_lock(&ctx->lock); + list_for_each_entry(event, &ctx->group_list, group_entry) { + if (event->state != PERF_EVENT_STATE_ACTIVE) + continue; + + hwc = &event->hw; + + interrupts = hwc->interrupts; + hwc->interrupts = 0; + + /* + * unthrottle events on the tick + */ + if (interrupts == MAX_INTERRUPTS) { + perf_log_throttle(event, 1); + event->pmu->unthrottle(event); + interrupts = 2*sysctl_perf_event_sample_rate/HZ; + } + + if (!event->attr.freq || !event->attr.sample_freq) + continue; + + /* + * if the specified freq < HZ then we need to skip ticks + */ + if (event->attr.sample_freq < HZ) { + freq = event->attr.sample_freq; + + hwc->freq_count += freq; + hwc->freq_interrupts += interrupts; + + if (hwc->freq_count < HZ) + continue; + + interrupts = hwc->freq_interrupts; + hwc->freq_interrupts = 0; + hwc->freq_count -= HZ; + } else + freq = HZ; + + perf_adjust_period(event, freq * interrupts); + + /* + * In order to avoid being stalled by an (accidental) huge + * sample period, force reset the sample period if we didn't + * get any events in this freq period. + */ + if (!interrupts) { + perf_disable(); + event->pmu->disable(event); + atomic64_set(&hwc->period_left, 0); + event->pmu->enable(event); + perf_enable(); + } + } + spin_unlock(&ctx->lock); +} + +/* + * Round-robin a context's events: + */ +static void rotate_ctx(struct perf_event_context *ctx) +{ + struct perf_event *event; + + if (!ctx->nr_events) + return; + + spin_lock(&ctx->lock); + /* + * Rotate the first entry last (works just fine for group events too): + */ + perf_disable(); + list_for_each_entry(event, &ctx->group_list, group_entry) { + list_move_tail(&event->group_entry, &ctx->group_list); + break; + } + perf_enable(); + + spin_unlock(&ctx->lock); +} + +void perf_event_task_tick(struct task_struct *curr, int cpu) +{ + struct perf_cpu_context *cpuctx; + struct perf_event_context *ctx; + + if (!atomic_read(&nr_events)) + return; + + cpuctx = &per_cpu(perf_cpu_context, cpu); + ctx = curr->perf_event_ctxp; + + perf_ctx_adjust_freq(&cpuctx->ctx); + if (ctx) + perf_ctx_adjust_freq(ctx); + + perf_event_cpu_sched_out(cpuctx); + if (ctx) + __perf_event_task_sched_out(ctx); + + rotate_ctx(&cpuctx->ctx); + if (ctx) + rotate_ctx(ctx); + + perf_event_cpu_sched_in(cpuctx, cpu); + if (ctx) + perf_event_task_sched_in(curr, cpu); +} + +/* + * Enable all of a task's events that have been marked enable-on-exec. + * This expects task == current. + */ +static void perf_event_enable_on_exec(struct task_struct *task) +{ + struct perf_event_context *ctx; + struct perf_event *event; + unsigned long flags; + int enabled = 0; + + local_irq_save(flags); + ctx = task->perf_event_ctxp; + if (!ctx || !ctx->nr_events) + goto out; + + __perf_event_task_sched_out(ctx); + + spin_lock(&ctx->lock); + + list_for_each_entry(event, &ctx->group_list, group_entry) { + if (!event->attr.enable_on_exec) + continue; + event->attr.enable_on_exec = 0; + if (event->state >= PERF_EVENT_STATE_INACTIVE) + continue; + __perf_event_mark_enabled(event, ctx); + enabled = 1; + } + + /* + * Unclone this context if we enabled any event. + */ + if (enabled) + unclone_ctx(ctx); + + spin_unlock(&ctx->lock); + + perf_event_task_sched_in(task, smp_processor_id()); + out: + local_irq_restore(flags); +} + +/* + * Cross CPU call to read the hardware event + */ +static void __perf_event_read(void *info) +{ + struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); + struct perf_event *event = info; + struct perf_event_context *ctx = event->ctx; + unsigned long flags; + + /* + * If this is a task context, we need to check whether it is + * the current task context of this cpu. If not it has been + * scheduled out before the smp call arrived. In that case + * event->count would have been updated to a recent sample + * when the event was scheduled out. + */ + if (ctx->task && cpuctx->task_ctx != ctx) + return; + + local_irq_save(flags); + if (ctx->is_active) + update_context_time(ctx); + event->pmu->read(event); + update_event_times(event); + local_irq_restore(flags); +} + +static u64 perf_event_read(struct perf_event *event) +{ + /* + * If event is enabled and currently active on a CPU, update the + * value in the event structure: + */ + if (event->state == PERF_EVENT_STATE_ACTIVE) { + smp_call_function_single(event->oncpu, + __perf_event_read, event, 1); + } else if (event->state == PERF_EVENT_STATE_INACTIVE) { + update_event_times(event); + } + + return atomic64_read(&event->count); +} + +/* + * Initialize the perf_event context in a task_struct: + */ +static void +__perf_event_init_context(struct perf_event_context *ctx, + struct task_struct *task) +{ + memset(ctx, 0, sizeof(*ctx)); + spin_lock_init(&ctx->lock); + mutex_init(&ctx->mutex); + INIT_LIST_HEAD(&ctx->group_list); + INIT_LIST_HEAD(&ctx->event_list); + atomic_set(&ctx->refcount, 1); + ctx->task = task; +} + +static struct perf_event_context *find_get_context(pid_t pid, int cpu) +{ + struct perf_event_context *ctx; + struct perf_cpu_context *cpuctx; + struct task_struct *task; + unsigned long flags; + int err; + + /* + * If cpu is not a wildcard then this is a percpu event: + */ + if (cpu != -1) { + /* Must be root to operate on a CPU event: */ + if (perf_paranoid_cpu() && !capable(CAP_SYS_ADMIN)) + return ERR_PTR(-EACCES); + + if (cpu < 0 || cpu > num_possible_cpus()) + return ERR_PTR(-EINVAL); + + /* + * We could be clever and allow to attach a event to an + * offline CPU and activate it when the CPU comes up, but + * that's for later. + */ + if (!cpu_isset(cpu, cpu_online_map)) + return ERR_PTR(-ENODEV); + + cpuctx = &per_cpu(perf_cpu_context, cpu); + ctx = &cpuctx->ctx; + get_ctx(ctx); + + return ctx; + } + + rcu_read_lock(); + if (!pid) + task = current; + else + task = find_task_by_vpid(pid); + if (task) + get_task_struct(task); + rcu_read_unlock(); + + if (!task) + return ERR_PTR(-ESRCH); + + /* + * Can't attach events to a dying task. + */ + err = -ESRCH; + if (task->flags & PF_EXITING) + goto errout; + + /* Reuse ptrace permission checks for now. */ + err = -EACCES; + if (!ptrace_may_access(task, PTRACE_MODE_READ)) + goto errout; + + retry: + ctx = perf_lock_task_context(task, &flags); + if (ctx) { + unclone_ctx(ctx); + spin_unlock_irqrestore(&ctx->lock, flags); + } + + if (!ctx) { + ctx = kmalloc(sizeof(struct perf_event_context), GFP_KERNEL); + err = -ENOMEM; + if (!ctx) + goto errout; + __perf_event_init_context(ctx, task); + get_ctx(ctx); + if (cmpxchg(&task->perf_event_ctxp, NULL, ctx)) { + /* + * We raced with some other task; use + * the context they set. + */ + kfree(ctx); + goto retry; + } + get_task_struct(task); + } + + put_task_struct(task); + return ctx; + + errout: + put_task_struct(task); + return ERR_PTR(err); +} + +static void free_event_rcu(struct rcu_head *head) +{ + struct perf_event *event; + + event = container_of(head, struct perf_event, rcu_head); + if (event->ns) + put_pid_ns(event->ns); + kfree(event); +} + +static void perf_pending_sync(struct perf_event *event); + +static void free_event(struct perf_event *event) +{ + perf_pending_sync(event); + + if (!event->parent) { + atomic_dec(&nr_events); + if (event->attr.mmap) + atomic_dec(&nr_mmap_events); + if (event->attr.comm) + atomic_dec(&nr_comm_events); + if (event->attr.task) + atomic_dec(&nr_task_events); + } + + if (event->output) { + fput(event->output->filp); + event->output = NULL; + } + + if (event->destroy) + event->destroy(event); + + put_ctx(event->ctx); + call_rcu(&event->rcu_head, free_event_rcu); +} + +/* + * Called when the last reference to the file is gone. + */ +static int perf_release(struct inode *inode, struct file *file) +{ + struct perf_event *event = file->private_data; + struct perf_event_context *ctx = event->ctx; + + file->private_data = NULL; + + WARN_ON_ONCE(ctx->parent_ctx); + mutex_lock(&ctx->mutex); + perf_event_remove_from_context(event); + mutex_unlock(&ctx->mutex); + + mutex_lock(&event->owner->perf_event_mutex); + list_del_init(&event->owner_entry); + mutex_unlock(&event->owner->perf_event_mutex); + put_task_struct(event->owner); + + free_event(event); + + return 0; +} + +static int perf_event_read_size(struct perf_event *event) +{ + int entry = sizeof(u64); /* value */ + int size = 0; + int nr = 1; + + if (event->attr.read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) + size += sizeof(u64); + + if (event->attr.read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) + size += sizeof(u64); + + if (event->attr.read_format & PERF_FORMAT_ID) + entry += sizeof(u64); + + if (event->attr.read_format & PERF_FORMAT_GROUP) { + nr += event->group_leader->nr_siblings; + size += sizeof(u64); + } + + size += entry * nr; + + return size; +} + +static u64 perf_event_read_value(struct perf_event *event) +{ + struct perf_event *child; + u64 total = 0; + + total += perf_event_read(event); + list_for_each_entry(child, &event->child_list, child_list) + total += perf_event_read(child); + + return total; +} + +static int perf_event_read_entry(struct perf_event *event, + u64 read_format, char __user *buf) +{ + int n = 0, count = 0; + u64 values[2]; + + values[n++] = perf_event_read_value(event); + if (read_format & PERF_FORMAT_ID) + values[n++] = primary_event_id(event); + + count = n * sizeof(u64); + + if (copy_to_user(buf, values, count)) + return -EFAULT; + + return count; +} + +static int perf_event_read_group(struct perf_event *event, + u64 read_format, char __user *buf) +{ + struct perf_event *leader = event->group_leader, *sub; + int n = 0, size = 0, err = -EFAULT; + u64 values[3]; + + values[n++] = 1 + leader->nr_siblings; + if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) { + values[n++] = leader->total_time_enabled + + atomic64_read(&leader->child_total_time_enabled); + } + if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) { + values[n++] = leader->total_time_running + + atomic64_read(&leader->child_total_time_running); + } + + size = n * sizeof(u64); + + if (copy_to_user(buf, values, size)) + return -EFAULT; + + err = perf_event_read_entry(leader, read_format, buf + size); + if (err < 0) + return err; + + size += err; + + list_for_each_entry(sub, &leader->sibling_list, group_entry) { + err = perf_event_read_entry(sub, read_format, + buf + size); + if (err < 0) + return err; + + size += err; + } + + return size; +} + +static int perf_event_read_one(struct perf_event *event, + u64 read_format, char __user *buf) +{ + u64 values[4]; + int n = 0; + + values[n++] = perf_event_read_value(event); + if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) { + values[n++] = event->total_time_enabled + + atomic64_read(&event->child_total_time_enabled); + } + if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) { + values[n++] = event->total_time_running + + atomic64_read(&event->child_total_time_running); + } + if (read_format & PERF_FORMAT_ID) + values[n++] = primary_event_id(event); + + if (copy_to_user(buf, values, n * sizeof(u64))) + return -EFAULT; + + return n * sizeof(u64); +} + +/* + * Read the performance event - simple non blocking version for now + */ +static ssize_t +perf_read_hw(struct perf_event *event, char __user *buf, size_t count) +{ + u64 read_format = event->attr.read_format; + int ret; + + /* + * Return end-of-file for a read on a event that is in + * error state (i.e. because it was pinned but it couldn't be + * scheduled on to the CPU at some point). + */ + if (event->state == PERF_EVENT_STATE_ERROR) + return 0; + + if (count < perf_event_read_size(event)) + return -ENOSPC; + + WARN_ON_ONCE(event->ctx->parent_ctx); + mutex_lock(&event->child_mutex); + if (read_format & PERF_FORMAT_GROUP) + ret = perf_event_read_group(event, read_format, buf); + else + ret = perf_event_read_one(event, read_format, buf); + mutex_unlock(&event->child_mutex); + + return ret; +} + +static ssize_t +perf_read(struct file *file, char __user *buf, size_t count, loff_t *ppos) +{ + struct perf_event *event = file->private_data; + + return perf_read_hw(event, buf, count); +} + +static unsigned int perf_poll(struct file *file, poll_table *wait) +{ + struct perf_event *event = file->private_data; + struct perf_mmap_data *data; + unsigned int events = POLL_HUP; + + rcu_read_lock(); + data = rcu_dereference(event->data); + if (data) + events = atomic_xchg(&data->poll, 0); + rcu_read_unlock(); + + poll_wait(file, &event->waitq, wait); + + return events; +} + +static void perf_event_reset(struct perf_event *event) +{ + (void)perf_event_read(event); + atomic64_set(&event->count, 0); + perf_event_update_userpage(event); +} + +/* + * Holding the top-level event's child_mutex means that any + * descendant process that has inherited this event will block + * in sync_child_event if it goes to exit, thus satisfying the + * task existence requirements of perf_event_enable/disable. + */ +static void perf_event_for_each_child(struct perf_event *event, + void (*func)(struct perf_event *)) +{ + struct perf_event *child; + + WARN_ON_ONCE(event->ctx->parent_ctx); + mutex_lock(&event->child_mutex); + func(event); + list_for_each_entry(child, &event->child_list, child_list) + func(child); + mutex_unlock(&event->child_mutex); +} + +static void perf_event_for_each(struct perf_event *event, + void (*func)(struct perf_event *)) +{ + struct perf_event_context *ctx = event->ctx; + struct perf_event *sibling; + + WARN_ON_ONCE(ctx->parent_ctx); + mutex_lock(&ctx->mutex); + event = event->group_leader; + + perf_event_for_each_child(event, func); + func(event); + list_for_each_entry(sibling, &event->sibling_list, group_entry) + perf_event_for_each_child(event, func); + mutex_unlock(&ctx->mutex); +} + +static int perf_event_period(struct perf_event *event, u64 __user *arg) +{ + struct perf_event_context *ctx = event->ctx; + unsigned long size; + int ret = 0; + u64 value; + + if (!event->attr.sample_period) + return -EINVAL; + + size = copy_from_user(&value, arg, sizeof(value)); + if (size != sizeof(value)) + return -EFAULT; + + if (!value) + return -EINVAL; + + spin_lock_irq(&ctx->lock); + if (event->attr.freq) { + if (value > sysctl_perf_event_sample_rate) { + ret = -EINVAL; + goto unlock; + } + + event->attr.sample_freq = value; + } else { + event->attr.sample_period = value; + event->hw.sample_period = value; + } +unlock: + spin_unlock_irq(&ctx->lock); + + return ret; +} + +int perf_event_set_output(struct perf_event *event, int output_fd); + +static long perf_ioctl(struct file *file, unsigned int cmd, unsigned long arg) +{ + struct perf_event *event = file->private_data; + void (*func)(struct perf_event *); + u32 flags = arg; + + switch (cmd) { + case PERF_EVENT_IOC_ENABLE: + func = perf_event_enable; + break; + case PERF_EVENT_IOC_DISABLE: + func = perf_event_disable; + break; + case PERF_EVENT_IOC_RESET: + func = perf_event_reset; + break; + + case PERF_EVENT_IOC_REFRESH: + return perf_event_refresh(event, arg); + + case PERF_EVENT_IOC_PERIOD: + return perf_event_period(event, (u64 __user *)arg); + + case PERF_EVENT_IOC_SET_OUTPUT: + return perf_event_set_output(event, arg); + + default: + return -ENOTTY; + } + + if (flags & PERF_IOC_FLAG_GROUP) + perf_event_for_each(event, func); + else + perf_event_for_each_child(event, func); + + return 0; +} + +int perf_event_task_enable(void) +{ + struct perf_event *event; + + mutex_lock(¤t->perf_event_mutex); + list_for_each_entry(event, ¤t->perf_event_list, owner_entry) + perf_event_for_each_child(event, perf_event_enable); + mutex_unlock(¤t->perf_event_mutex); + + return 0; +} + +int perf_event_task_disable(void) +{ + struct perf_event *event; + + mutex_lock(¤t->perf_event_mutex); + list_for_each_entry(event, ¤t->perf_event_list, owner_entry) + perf_event_for_each_child(event, perf_event_disable); + mutex_unlock(¤t->perf_event_mutex); + + return 0; +} + +#ifndef PERF_EVENT_INDEX_OFFSET +# define PERF_EVENT_INDEX_OFFSET 0 +#endif + +static int perf_event_index(struct perf_event *event) +{ + if (event->state != PERF_EVENT_STATE_ACTIVE) + return 0; + + return event->hw.idx + 1 - PERF_EVENT_INDEX_OFFSET; +} + +/* + * Callers need to ensure there can be no nesting of this function, otherwise + * the seqlock logic goes bad. We can not serialize this because the arch + * code calls this from NMI context. + */ +void perf_event_update_userpage(struct perf_event *event) +{ + struct perf_event_mmap_page *userpg; + struct perf_mmap_data *data; + + rcu_read_lock(); + data = rcu_dereference(event->data); + if (!data) + goto unlock; + + userpg = data->user_page; + + /* + * Disable preemption so as to not let the corresponding user-space + * spin too long if we get preempted. + */ + preempt_disable(); + ++userpg->lock; + barrier(); + userpg->index = perf_event_index(event); + userpg->offset = atomic64_read(&event->count); + if (event->state == PERF_EVENT_STATE_ACTIVE) + userpg->offset -= atomic64_read(&event->hw.prev_count); + + userpg->time_enabled = event->total_time_enabled + + atomic64_read(&event->child_total_time_enabled); + + userpg->time_running = event->total_time_running + + atomic64_read(&event->child_total_time_running); + + barrier(); + ++userpg->lock; + preempt_enable(); +unlock: + rcu_read_unlock(); +} + +static int perf_mmap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) +{ + struct perf_event *event = vma->vm_file->private_data; + struct perf_mmap_data *data; + int ret = VM_FAULT_SIGBUS; + + if (vmf->flags & FAULT_FLAG_MKWRITE) { + if (vmf->pgoff == 0) + ret = 0; + return ret; + } + + rcu_read_lock(); + data = rcu_dereference(event->data); + if (!data) + goto unlock; + + if (vmf->pgoff == 0) { + vmf->page = virt_to_page(data->user_page); + } else { + int nr = vmf->pgoff - 1; + + if ((unsigned)nr > data->nr_pages) + goto unlock; + + if (vmf->flags & FAULT_FLAG_WRITE) + goto unlock; + + vmf->page = virt_to_page(data->data_pages[nr]); + } + + get_page(vmf->page); + vmf->page->mapping = vma->vm_file->f_mapping; + vmf->page->index = vmf->pgoff; + + ret = 0; +unlock: + rcu_read_unlock(); + + return ret; +} + +static int perf_mmap_data_alloc(struct perf_event *event, int nr_pages) +{ + struct perf_mmap_data *data; + unsigned long size; + int i; + + WARN_ON(atomic_read(&event->mmap_count)); + + size = sizeof(struct perf_mmap_data); + size += nr_pages * sizeof(void *); + + data = kzalloc(size, GFP_KERNEL); + if (!data) + goto fail; + + data->user_page = (void *)get_zeroed_page(GFP_KERNEL); + if (!data->user_page) + goto fail_user_page; + + for (i = 0; i < nr_pages; i++) { + data->data_pages[i] = (void *)get_zeroed_page(GFP_KERNEL); + if (!data->data_pages[i]) + goto fail_data_pages; + } + + data->nr_pages = nr_pages; + atomic_set(&data->lock, -1); + + if (event->attr.watermark) { + data->watermark = min_t(long, PAGE_SIZE * nr_pages, + event->attr.wakeup_watermark); + } + if (!data->watermark) + data->watermark = max(PAGE_SIZE, PAGE_SIZE * nr_pages / 4); + + rcu_assign_pointer(event->data, data); + + return 0; + +fail_data_pages: + for (i--; i >= 0; i--) + free_page((unsigned long)data->data_pages[i]); + + free_page((unsigned long)data->user_page); + +fail_user_page: + kfree(data); + +fail: + return -ENOMEM; +} + +static void perf_mmap_free_page(unsigned long addr) +{ + struct page *page = virt_to_page((void *)addr); + + page->mapping = NULL; + __free_page(page); +} + +static void __perf_mmap_data_free(struct rcu_head *rcu_head) +{ + struct perf_mmap_data *data; + int i; + + data = container_of(rcu_head, struct perf_mmap_data, rcu_head); + + perf_mmap_free_page((unsigned long)data->user_page); + for (i = 0; i < data->nr_pages; i++) + perf_mmap_free_page((unsigned long)data->data_pages[i]); + + kfree(data); +} + +static void perf_mmap_data_free(struct perf_event *event) +{ + struct perf_mmap_data *data = event->data; + + WARN_ON(atomic_read(&event->mmap_count)); + + rcu_assign_pointer(event->data, NULL); + call_rcu(&data->rcu_head, __perf_mmap_data_free); +} + +static void perf_mmap_open(struct vm_area_struct *vma) +{ + struct perf_event *event = vma->vm_file->private_data; + + atomic_inc(&event->mmap_count); +} + +static void perf_mmap_close(struct vm_area_struct *vma) +{ + struct perf_event *event = vma->vm_file->private_data; + + WARN_ON_ONCE(event->ctx->parent_ctx); + if (atomic_dec_and_mutex_lock(&event->mmap_count, &event->mmap_mutex)) { + struct user_struct *user = current_user(); + + atomic_long_sub(event->data->nr_pages + 1, &user->locked_vm); + vma->vm_mm->locked_vm -= event->data->nr_locked; + perf_mmap_data_free(event); + mutex_unlock(&event->mmap_mutex); + } +} + +static struct vm_operations_struct perf_mmap_vmops = { + .open = perf_mmap_open, + .close = perf_mmap_close, + .fault = perf_mmap_fault, + .page_mkwrite = perf_mmap_fault, +}; + +static int perf_mmap(struct file *file, struct vm_area_struct *vma) +{ + struct perf_event *event = file->private_data; + unsigned long user_locked, user_lock_limit; + struct user_struct *user = current_user(); + unsigned long locked, lock_limit; + unsigned long vma_size; + unsigned long nr_pages; + long user_extra, extra; + int ret = 0; + + if (!(vma->vm_flags & VM_SHARED)) + return -EINVAL; + + vma_size = vma->vm_end - vma->vm_start; + nr_pages = (vma_size / PAGE_SIZE) - 1; + + /* + * If we have data pages ensure they're a power-of-two number, so we + * can do bitmasks instead of modulo. + */ + if (nr_pages != 0 && !is_power_of_2(nr_pages)) + return -EINVAL; + + if (vma_size != PAGE_SIZE * (1 + nr_pages)) + return -EINVAL; + + if (vma->vm_pgoff != 0) + return -EINVAL; + + WARN_ON_ONCE(event->ctx->parent_ctx); + mutex_lock(&event->mmap_mutex); + if (event->output) { + ret = -EINVAL; + goto unlock; + } + + if (atomic_inc_not_zero(&event->mmap_count)) { + if (nr_pages != event->data->nr_pages) + ret = -EINVAL; + goto unlock; + } + + user_extra = nr_pages + 1; + user_lock_limit = sysctl_perf_event_mlock >> (PAGE_SHIFT - 10); + + /* + * Increase the limit linearly with more CPUs: + */ + user_lock_limit *= num_online_cpus(); + + user_locked = atomic_long_read(&user->locked_vm) + user_extra; + + extra = 0; + if (user_locked > user_lock_limit) + extra = user_locked - user_lock_limit; + + lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur; + lock_limit >>= PAGE_SHIFT; + locked = vma->vm_mm->locked_vm + extra; + + if ((locked > lock_limit) && perf_paranoid_tracepoint_raw() && + !capable(CAP_IPC_LOCK)) { + ret = -EPERM; + goto unlock; + } + + WARN_ON(event->data); + ret = perf_mmap_data_alloc(event, nr_pages); + if (ret) + goto unlock; + + atomic_set(&event->mmap_count, 1); + atomic_long_add(user_extra, &user->locked_vm); + vma->vm_mm->locked_vm += extra; + event->data->nr_locked = extra; + if (vma->vm_flags & VM_WRITE) + event->data->writable = 1; + +unlock: + mutex_unlock(&event->mmap_mutex); + + vma->vm_flags |= VM_RESERVED; + vma->vm_ops = &perf_mmap_vmops; + + return ret; +} + +static int perf_fasync(int fd, struct file *filp, int on) +{ + struct inode *inode = filp->f_path.dentry->d_inode; + struct perf_event *event = filp->private_data; + int retval; + + mutex_lock(&inode->i_mutex); + retval = fasync_helper(fd, filp, on, &event->fasync); + mutex_unlock(&inode->i_mutex); + + if (retval < 0) + return retval; + + return 0; +} + +static const struct file_operations perf_fops = { + .release = perf_release, + .read = perf_read, + .poll = perf_poll, + .unlocked_ioctl = perf_ioctl, + .compat_ioctl = perf_ioctl, + .mmap = perf_mmap, + .fasync = perf_fasync, +}; + +/* + * Perf event wakeup + * + * If there's data, ensure we set the poll() state and publish everything + * to user-space before waking everybody up. + */ + +void perf_event_wakeup(struct perf_event *event) +{ + wake_up_all(&event->waitq); + + if (event->pending_kill) { + kill_fasync(&event->fasync, SIGIO, event->pending_kill); + event->pending_kill = 0; + } +} + +/* + * Pending wakeups + * + * Handle the case where we need to wakeup up from NMI (or rq->lock) context. + * + * The NMI bit means we cannot possibly take locks. Therefore, maintain a + * single linked list and use cmpxchg() to add entries lockless. + */ + +static void perf_pending_event(struct perf_pending_entry *entry) +{ + struct perf_event *event = container_of(entry, + struct perf_event, pending); + + if (event->pending_disable) { + event->pending_disable = 0; + __perf_event_disable(event); + } + + if (event->pending_wakeup) { + event->pending_wakeup = 0; + perf_event_wakeup(event); + } +} + +#define PENDING_TAIL ((struct perf_pending_entry *)-1UL) + +static DEFINE_PER_CPU(struct perf_pending_entry *, perf_pending_head) = { + PENDING_TAIL, +}; + +static void perf_pending_queue(struct perf_pending_entry *entry, + void (*func)(struct perf_pending_entry *)) +{ + struct perf_pending_entry **head; + + if (cmpxchg(&entry->next, NULL, PENDING_TAIL) != NULL) + return; + + entry->func = func; + + head = &get_cpu_var(perf_pending_head); + + do { + entry->next = *head; + } while (cmpxchg(head, entry->next, entry) != entry->next); + + set_perf_event_pending(); + + put_cpu_var(perf_pending_head); +} + +static int __perf_pending_run(void) +{ + struct perf_pending_entry *list; + int nr = 0; + + list = xchg(&__get_cpu_var(perf_pending_head), PENDING_TAIL); + while (list != PENDING_TAIL) { + void (*func)(struct perf_pending_entry *); + struct perf_pending_entry *entry = list; + + list = list->next; + + func = entry->func; + entry->next = NULL; + /* + * Ensure we observe the unqueue before we issue the wakeup, + * so that we won't be waiting forever. + * -- see perf_not_pending(). + */ + smp_wmb(); + + func(entry); + nr++; + } + + return nr; +} + +static inline int perf_not_pending(struct perf_event *event) +{ + /* + * If we flush on whatever cpu we run, there is a chance we don't + * need to wait. + */ + get_cpu(); + __perf_pending_run(); + put_cpu(); + + /* + * Ensure we see the proper queue state before going to sleep + * so that we do not miss the wakeup. -- see perf_pending_handle() + */ + smp_rmb(); + return event->pending.next == NULL; +} + +static void perf_pending_sync(struct perf_event *event) +{ + wait_event(event->waitq, perf_not_pending(event)); +} + +void perf_event_do_pending(void) +{ + __perf_pending_run(); +} + +/* + * Callchain support -- arch specific + */ + +__weak struct perf_callchain_entry *perf_callchain(struct pt_regs *regs) +{ + return NULL; +} + +/* + * Output + */ +static bool perf_output_space(struct perf_mmap_data *data, unsigned long tail, + unsigned long offset, unsigned long head) +{ + unsigned long mask; + + if (!data->writable) + return true; + + mask = (data->nr_pages << PAGE_SHIFT) - 1; + + offset = (offset - tail) & mask; + head = (head - tail) & mask; + + if ((int)(head - offset) < 0) + return false; + + return true; +} + +static void perf_output_wakeup(struct perf_output_handle *handle) +{ + atomic_set(&handle->data->poll, POLL_IN); + + if (handle->nmi) { + handle->event->pending_wakeup = 1; + perf_pending_queue(&handle->event->pending, + perf_pending_event); + } else + perf_event_wakeup(handle->event); +} + +/* + * Curious locking construct. + * + * We need to ensure a later event_id doesn't publish a head when a former + * event_id isn't done writing. However since we need to deal with NMIs we + * cannot fully serialize things. + * + * What we do is serialize between CPUs so we only have to deal with NMI + * nesting on a single CPU. + * + * We only publish the head (and generate a wakeup) when the outer-most + * event_id completes. + */ +static void perf_output_lock(struct perf_output_handle *handle) +{ + struct perf_mmap_data *data = handle->data; + int cpu; + + handle->locked = 0; + + local_irq_save(handle->flags); + cpu = smp_processor_id(); + + if (in_nmi() && atomic_read(&data->lock) == cpu) + return; + + while (atomic_cmpxchg(&data->lock, -1, cpu) != -1) + cpu_relax(); + + handle->locked = 1; +} + +static void perf_output_unlock(struct perf_output_handle *handle) +{ + struct perf_mmap_data *data = handle->data; + unsigned long head; + int cpu; + + data->done_head = data->head; + + if (!handle->locked) + goto out; + +again: + /* + * The xchg implies a full barrier that ensures all writes are done + * before we publish the new head, matched by a rmb() in userspace when + * reading this position. + */ + while ((head = atomic_long_xchg(&data->done_head, 0))) + data->user_page->data_head = head; + + /* + * NMI can happen here, which means we can miss a done_head update. + */ + + cpu = atomic_xchg(&data->lock, -1); + WARN_ON_ONCE(cpu != smp_processor_id()); + + /* + * Therefore we have to validate we did not indeed do so. + */ + if (unlikely(atomic_long_read(&data->done_head))) { + /* + * Since we had it locked, we can lock it again. + */ + while (atomic_cmpxchg(&data->lock, -1, cpu) != -1) + cpu_relax(); + + goto again; + } + + if (atomic_xchg(&data->wakeup, 0)) + perf_output_wakeup(handle); +out: + local_irq_restore(handle->flags); +} + +void perf_output_copy(struct perf_output_handle *handle, + const void *buf, unsigned int len) +{ + unsigned int pages_mask; + unsigned int offset; + unsigned int size; + void **pages; + + offset = handle->offset; + pages_mask = handle->data->nr_pages - 1; + pages = handle->data->data_pages; + + do { + unsigned int page_offset; + int nr; + + nr = (offset >> PAGE_SHIFT) & pages_mask; + page_offset = offset & (PAGE_SIZE - 1); + size = min_t(unsigned int, PAGE_SIZE - page_offset, len); + + memcpy(pages[nr] + page_offset, buf, size); + + len -= size; + buf += size; + offset += size; + } while (len); + + handle->offset = offset; + + /* + * Check we didn't copy past our reservation window, taking the + * possible unsigned int wrap into account. + */ + WARN_ON_ONCE(((long)(handle->head - handle->offset)) < 0); +} + +int perf_output_begin(struct perf_output_handle *handle, + struct perf_event *event, unsigned int size, + int nmi, int sample) +{ + struct perf_event *output_event; + struct perf_mmap_data *data; + unsigned long tail, offset, head; + int have_lost; + struct { + struct perf_event_header header; + u64 id; + u64 lost; + } lost_event; + + rcu_read_lock(); + /* + * For inherited events we send all the output towards the parent. + */ + if (event->parent) + event = event->parent; + + output_event = rcu_dereference(event->output); + if (output_event) + event = output_event; + + data = rcu_dereference(event->data); + if (!data) + goto out; + + handle->data = data; + handle->event = event; + handle->nmi = nmi; + handle->sample = sample; + + if (!data->nr_pages) + goto fail; + + have_lost = atomic_read(&data->lost); + if (have_lost) + size += sizeof(lost_event); + + perf_output_lock(handle); + + do { + /* + * Userspace could choose to issue a mb() before updating the + * tail pointer. So that all reads will be completed before the + * write is issued. + */ + tail = ACCESS_ONCE(data->user_page->data_tail); + smp_rmb(); + offset = head = atomic_long_read(&data->head); + head += size; + if (unlikely(!perf_output_space(data, tail, offset, head))) + goto fail; + } while (atomic_long_cmpxchg(&data->head, offset, head) != offset); + + handle->offset = offset; + handle->head = head; + + if (head - tail > data->watermark) + atomic_set(&data->wakeup, 1); + + if (have_lost) { + lost_event.header.type = PERF_RECORD_LOST; + lost_event.header.misc = 0; + lost_event.header.size = sizeof(lost_event); + lost_event.id = event->id; + lost_event.lost = atomic_xchg(&data->lost, 0); + + perf_output_put(handle, lost_event); + } + + return 0; + +fail: + atomic_inc(&data->lost); + perf_output_unlock(handle); +out: + rcu_read_unlock(); + + return -ENOSPC; +} + +void perf_output_end(struct perf_output_handle *handle) +{ + struct perf_event *event = handle->event; + struct perf_mmap_data *data = handle->data; + + int wakeup_events = event->attr.wakeup_events; + + if (handle->sample && wakeup_events) { + int events = atomic_inc_return(&data->events); + if (events >= wakeup_events) { + atomic_sub(wakeup_events, &data->events); + atomic_set(&data->wakeup, 1); + } + } + + perf_output_unlock(handle); + rcu_read_unlock(); +} + +static u32 perf_event_pid(struct perf_event *event, struct task_struct *p) +{ + /* + * only top level events have the pid namespace they were created in + */ + if (event->parent) + event = event->parent; + + return task_tgid_nr_ns(p, event->ns); +} + +static u32 perf_event_tid(struct perf_event *event, struct task_struct *p) +{ + /* + * only top level events have the pid namespace they were created in + */ + if (event->parent) + event = event->parent; + + return task_pid_nr_ns(p, event->ns); +} + +static void perf_output_read_one(struct perf_output_handle *handle, + struct perf_event *event) +{ + u64 read_format = event->attr.read_format; + u64 values[4]; + int n = 0; + + values[n++] = atomic64_read(&event->count); + if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) { + values[n++] = event->total_time_enabled + + atomic64_read(&event->child_total_time_enabled); + } + if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) { + values[n++] = event->total_time_running + + atomic64_read(&event->child_total_time_running); + } + if (read_format & PERF_FORMAT_ID) + values[n++] = primary_event_id(event); + + perf_output_copy(handle, values, n * sizeof(u64)); +} + +/* + * XXX PERF_FORMAT_GROUP vs inherited events seems difficult. + */ +static void perf_output_read_group(struct perf_output_handle *handle, + struct perf_event *event) +{ + struct perf_event *leader = event->group_leader, *sub; + u64 read_format = event->attr.read_format; + u64 values[5]; + int n = 0; + + values[n++] = 1 + leader->nr_siblings; + + if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) + values[n++] = leader->total_time_enabled; + + if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) + values[n++] = leader->total_time_running; + + if (leader != event) + leader->pmu->read(leader); + + values[n++] = atomic64_read(&leader->count); + if (read_format & PERF_FORMAT_ID) + values[n++] = primary_event_id(leader); + + perf_output_copy(handle, values, n * sizeof(u64)); + + list_for_each_entry(sub, &leader->sibling_list, group_entry) { + n = 0; + + if (sub != event) + sub->pmu->read(sub); + + values[n++] = atomic64_read(&sub->count); + if (read_format & PERF_FORMAT_ID) + values[n++] = primary_event_id(sub); + + perf_output_copy(handle, values, n * sizeof(u64)); + } +} + +static void perf_output_read(struct perf_output_handle *handle, + struct perf_event *event) +{ + if (event->attr.read_format & PERF_FORMAT_GROUP) + perf_output_read_group(handle, event); + else + perf_output_read_one(handle, event); +} + +void perf_output_sample(struct perf_output_handle *handle, + struct perf_event_header *header, + struct perf_sample_data *data, + struct perf_event *event) +{ + u64 sample_type = data->type; + + perf_output_put(handle, *header); + + if (sample_type & PERF_SAMPLE_IP) + perf_output_put(handle, data->ip); + + if (sample_type & PERF_SAMPLE_TID) + perf_output_put(handle, data->tid_entry); + + if (sample_type & PERF_SAMPLE_TIME) + perf_output_put(handle, data->time); + + if (sample_type & PERF_SAMPLE_ADDR) + perf_output_put(handle, data->addr); + + if (sample_type & PERF_SAMPLE_ID) + perf_output_put(handle, data->id); + + if (sample_type & PERF_SAMPLE_STREAM_ID) + perf_output_put(handle, data->stream_id); + + if (sample_type & PERF_SAMPLE_CPU) + perf_output_put(handle, data->cpu_entry); + + if (sample_type & PERF_SAMPLE_PERIOD) + perf_output_put(handle, data->period); + + if (sample_type & PERF_SAMPLE_READ) + perf_output_read(handle, event); + + if (sample_type & PERF_SAMPLE_CALLCHAIN) { + if (data->callchain) { + int size = 1; + + if (data->callchain) + size += data->callchain->nr; + + size *= sizeof(u64); + + perf_output_copy(handle, data->callchain, size); + } else { + u64 nr = 0; + perf_output_put(handle, nr); + } + } + + if (sample_type & PERF_SAMPLE_RAW) { + if (data->raw) { + perf_output_put(handle, data->raw->size); + perf_output_copy(handle, data->raw->data, + data->raw->size); + } else { + struct { + u32 size; + u32 data; + } raw = { + .size = sizeof(u32), + .data = 0, + }; + perf_output_put(handle, raw); + } + } +} + +void perf_prepare_sample(struct perf_event_header *header, + struct perf_sample_data *data, + struct perf_event *event, + struct pt_regs *regs) +{ + u64 sample_type = event->attr.sample_type; + + data->type = sample_type; + + header->type = PERF_RECORD_SAMPLE; + header->size = sizeof(*header); + + header->misc = 0; + header->misc |= perf_misc_flags(regs); + + if (sample_type & PERF_SAMPLE_IP) { + data->ip = perf_instruction_pointer(regs); + + header->size += sizeof(data->ip); + } + + if (sample_type & PERF_SAMPLE_TID) { + /* namespace issues */ + data->tid_entry.pid = perf_event_pid(event, current); + data->tid_entry.tid = perf_event_tid(event, current); + + header->size += sizeof(data->tid_entry); + } + + if (sample_type & PERF_SAMPLE_TIME) { + data->time = perf_clock(); + + header->size += sizeof(data->time); + } + + if (sample_type & PERF_SAMPLE_ADDR) + header->size += sizeof(data->addr); + + if (sample_type & PERF_SAMPLE_ID) { + data->id = primary_event_id(event); + + header->size += sizeof(data->id); + } + + if (sample_type & PERF_SAMPLE_STREAM_ID) { + data->stream_id = event->id; + + header->size += sizeof(data->stream_id); + } + + if (sample_type & PERF_SAMPLE_CPU) { + data->cpu_entry.cpu = raw_smp_processor_id(); + data->cpu_entry.reserved = 0; + + header->size += sizeof(data->cpu_entry); + } + + if (sample_type & PERF_SAMPLE_PERIOD) + header->size += sizeof(data->period); + + if (sample_type & PERF_SAMPLE_READ) + header->size += perf_event_read_size(event); + + if (sample_type & PERF_SAMPLE_CALLCHAIN) { + int size = 1; + + data->callchain = perf_callchain(regs); + + if (data->callchain) + size += data->callchain->nr; + + header->size += size * sizeof(u64); + } + + if (sample_type & PERF_SAMPLE_RAW) { + int size = sizeof(u32); + + if (data->raw) + size += data->raw->size; + else + size += sizeof(u32); + + WARN_ON_ONCE(size & (sizeof(u64)-1)); + header->size += size; + } +} + +static void perf_event_output(struct perf_event *event, int nmi, + struct perf_sample_data *data, + struct pt_regs *regs) +{ + struct perf_output_handle handle; + struct perf_event_header header; + + perf_prepare_sample(&header, data, event, regs); + + if (perf_output_begin(&handle, event, header.size, nmi, 1)) + return; + + perf_output_sample(&handle, &header, data, event); + + perf_output_end(&handle); +} + +/* + * read event_id + */ + +struct perf_read_event { + struct perf_event_header header; + + u32 pid; + u32 tid; +}; + +static void +perf_event_read_event(struct perf_event *event, + struct task_struct *task) +{ + struct perf_output_handle handle; + struct perf_read_event read_event = { + .header = { + .type = PERF_RECORD_READ, + .misc = 0, + .size = sizeof(read_event) + perf_event_read_size(event), + }, + .pid = perf_event_pid(event, task), + .tid = perf_event_tid(event, task), + }; + int ret; + + ret = perf_output_begin(&handle, event, read_event.header.size, 0, 0); + if (ret) + return; + + perf_output_put(&handle, read_event); + perf_output_read(&handle, event); + + perf_output_end(&handle); +} + +/* + * task tracking -- fork/exit + * + * enabled by: attr.comm | attr.mmap | attr.task + */ + +struct perf_task_event { + struct task_struct *task; + struct perf_event_context *task_ctx; + + struct { + struct perf_event_header header; + + u32 pid; + u32 ppid; + u32 tid; + u32 ptid; + u64 time; + } event_id; +}; + +static void perf_event_task_output(struct perf_event *event, + struct perf_task_event *task_event) +{ + struct perf_output_handle handle; + int size; + struct task_struct *task = task_event->task; + int ret; + + size = task_event->event_id.header.size; + ret = perf_output_begin(&handle, event, size, 0, 0); + + if (ret) + return; + + task_event->event_id.pid = perf_event_pid(event, task); + task_event->event_id.ppid = perf_event_pid(event, current); + + task_event->event_id.tid = perf_event_tid(event, task); + task_event->event_id.ptid = perf_event_tid(event, current); + + task_event->event_id.time = perf_clock(); + + perf_output_put(&handle, task_event->event_id); + + perf_output_end(&handle); +} + +static int perf_event_task_match(struct perf_event *event) +{ + if (event->attr.comm || event->attr.mmap || event->attr.task) + return 1; + + return 0; +} + +static void perf_event_task_ctx(struct perf_event_context *ctx, + struct perf_task_event *task_event) +{ + struct perf_event *event; + + if (system_state != SYSTEM_RUNNING || list_empty(&ctx->event_list)) + return; + + rcu_read_lock(); + list_for_each_entry_rcu(event, &ctx->event_list, event_entry) { + if (perf_event_task_match(event)) + perf_event_task_output(event, task_event); + } + rcu_read_unlock(); +} + +static void perf_event_task_event(struct perf_task_event *task_event) +{ + struct perf_cpu_context *cpuctx; + struct perf_event_context *ctx = task_event->task_ctx; + + cpuctx = &get_cpu_var(perf_cpu_context); + perf_event_task_ctx(&cpuctx->ctx, task_event); + put_cpu_var(perf_cpu_context); + + rcu_read_lock(); + if (!ctx) + ctx = rcu_dereference(task_event->task->perf_event_ctxp); + if (ctx) + perf_event_task_ctx(ctx, task_event); + rcu_read_unlock(); +} + +static void perf_event_task(struct task_struct *task, + struct perf_event_context *task_ctx, + int new) +{ + struct perf_task_event task_event; + + if (!atomic_read(&nr_comm_events) && + !atomic_read(&nr_mmap_events) && + !atomic_read(&nr_task_events)) + return; + + task_event = (struct perf_task_event){ + .task = task, + .task_ctx = task_ctx, + .event_id = { + .header = { + .type = new ? PERF_RECORD_FORK : PERF_RECORD_EXIT, + .misc = 0, + .size = sizeof(task_event.event_id), + }, + /* .pid */ + /* .ppid */ + /* .tid */ + /* .ptid */ + }, + }; + + perf_event_task_event(&task_event); +} + +void perf_event_fork(struct task_struct *task) +{ + perf_event_task(task, NULL, 1); +} + +/* + * comm tracking + */ + +struct perf_comm_event { + struct task_struct *task; + char *comm; + int comm_size; + + struct { + struct perf_event_header header; + + u32 pid; + u32 tid; + } event_id; +}; + +static void perf_event_comm_output(struct perf_event *event, + struct perf_comm_event *comm_event) +{ + struct perf_output_handle handle; + int size = comm_event->event_id.header.size; + int ret = perf_output_begin(&handle, event, size, 0, 0); + + if (ret) + return; + + comm_event->event_id.pid = perf_event_pid(event, comm_event->task); + comm_event->event_id.tid = perf_event_tid(event, comm_event->task); + + perf_output_put(&handle, comm_event->event_id); + perf_output_copy(&handle, comm_event->comm, + comm_event->comm_size); + perf_output_end(&handle); +} + +static int perf_event_comm_match(struct perf_event *event) +{ + if (event->attr.comm) + return 1; + + return 0; +} + +static void perf_event_comm_ctx(struct perf_event_context *ctx, + struct perf_comm_event *comm_event) +{ + struct perf_event *event; + + if (system_state != SYSTEM_RUNNING || list_empty(&ctx->event_list)) + return; + + rcu_read_lock(); + list_for_each_entry_rcu(event, &ctx->event_list, event_entry) { + if (perf_event_comm_match(event)) + perf_event_comm_output(event, comm_event); + } + rcu_read_unlock(); +} + +static void perf_event_comm_event(struct perf_comm_event *comm_event) +{ + struct perf_cpu_context *cpuctx; + struct perf_event_context *ctx; + unsigned int size; + char comm[TASK_COMM_LEN]; + + memset(comm, 0, sizeof(comm)); + strncpy(comm, comm_event->task->comm, sizeof(comm)); + size = ALIGN(strlen(comm)+1, sizeof(u64)); + + comm_event->comm = comm; + comm_event->comm_size = size; + + comm_event->event_id.header.size = sizeof(comm_event->event_id) + size; + + cpuctx = &get_cpu_var(perf_cpu_context); + perf_event_comm_ctx(&cpuctx->ctx, comm_event); + put_cpu_var(perf_cpu_context); + + rcu_read_lock(); + /* + * doesn't really matter which of the child contexts the + * events ends up in. + */ + ctx = rcu_dereference(current->perf_event_ctxp); + if (ctx) + perf_event_comm_ctx(ctx, comm_event); + rcu_read_unlock(); +} + +void perf_event_comm(struct task_struct *task) +{ + struct perf_comm_event comm_event; + + if (task->perf_event_ctxp) + perf_event_enable_on_exec(task); + + if (!atomic_read(&nr_comm_events)) + return; + + comm_event = (struct perf_comm_event){ + .task = task, + /* .comm */ + /* .comm_size */ + .event_id = { + .header = { + .type = PERF_RECORD_COMM, + .misc = 0, + /* .size */ + }, + /* .pid */ + /* .tid */ + }, + }; + + perf_event_comm_event(&comm_event); +} + +/* + * mmap tracking + */ + +struct perf_mmap_event { + struct vm_area_struct *vma; + + const char *file_name; + int file_size; + + struct { + struct perf_event_header header; + + u32 pid; + u32 tid; + u64 start; + u64 len; + u64 pgoff; + } event_id; +}; + +static void perf_event_mmap_output(struct perf_event *event, + struct perf_mmap_event *mmap_event) +{ + struct perf_output_handle handle; + int size = mmap_event->event_id.header.size; + int ret = perf_output_begin(&handle, event, size, 0, 0); + + if (ret) + return; + + mmap_event->event_id.pid = perf_event_pid(event, current); + mmap_event->event_id.tid = perf_event_tid(event, current); + + perf_output_put(&handle, mmap_event->event_id); + perf_output_copy(&handle, mmap_event->file_name, + mmap_event->file_size); + perf_output_end(&handle); +} + +static int perf_event_mmap_match(struct perf_event *event, + struct perf_mmap_event *mmap_event) +{ + if (event->attr.mmap) + return 1; + + return 0; +} + +static void perf_event_mmap_ctx(struct perf_event_context *ctx, + struct perf_mmap_event *mmap_event) +{ + struct perf_event *event; + + if (system_state != SYSTEM_RUNNING || list_empty(&ctx->event_list)) + return; + + rcu_read_lock(); + list_for_each_entry_rcu(event, &ctx->event_list, event_entry) { + if (perf_event_mmap_match(event, mmap_event)) + perf_event_mmap_output(event, mmap_event); + } + rcu_read_unlock(); +} + +static void perf_event_mmap_event(struct perf_mmap_event *mmap_event) +{ + struct perf_cpu_context *cpuctx; + struct perf_event_context *ctx; + struct vm_area_struct *vma = mmap_event->vma; + struct file *file = vma->vm_file; + unsigned int size; + char tmp[16]; + char *buf = NULL; + const char *name; + + memset(tmp, 0, sizeof(tmp)); + + if (file) { + /* + * d_path works from the end of the buffer backwards, so we + * need to add enough zero bytes after the string to handle + * the 64bit alignment we do later. + */ + buf = kzalloc(PATH_MAX + sizeof(u64), GFP_KERNEL); + if (!buf) { + name = strncpy(tmp, "//enomem", sizeof(tmp)); + goto got_name; + } + name = d_path(&file->f_path, buf, PATH_MAX); + if (IS_ERR(name)) { + name = strncpy(tmp, "//toolong", sizeof(tmp)); + goto got_name; + } + } else { + if (arch_vma_name(mmap_event->vma)) { + name = strncpy(tmp, arch_vma_name(mmap_event->vma), + sizeof(tmp)); + goto got_name; + } + + if (!vma->vm_mm) { + name = strncpy(tmp, "[vdso]", sizeof(tmp)); + goto got_name; + } + + name = strncpy(tmp, "//anon", sizeof(tmp)); + goto got_name; + } + +got_name: + size = ALIGN(strlen(name)+1, sizeof(u64)); + + mmap_event->file_name = name; + mmap_event->file_size = size; + + mmap_event->event_id.header.size = sizeof(mmap_event->event_id) + size; + + cpuctx = &get_cpu_var(perf_cpu_context); + perf_event_mmap_ctx(&cpuctx->ctx, mmap_event); + put_cpu_var(perf_cpu_context); + + rcu_read_lock(); + /* + * doesn't really matter which of the child contexts the + * events ends up in. + */ + ctx = rcu_dereference(current->perf_event_ctxp); + if (ctx) + perf_event_mmap_ctx(ctx, mmap_event); + rcu_read_unlock(); + + kfree(buf); +} + +void __perf_event_mmap(struct vm_area_struct *vma) +{ + struct perf_mmap_event mmap_event; + + if (!atomic_read(&nr_mmap_events)) + return; + + mmap_event = (struct perf_mmap_event){ + .vma = vma, + /* .file_name */ + /* .file_size */ + .event_id = { + .header = { + .type = PERF_RECORD_MMAP, + .misc = 0, + /* .size */ + }, + /* .pid */ + /* .tid */ + .start = vma->vm_start, + .len = vma->vm_end - vma->vm_start, + .pgoff = vma->vm_pgoff, + }, + }; + + perf_event_mmap_event(&mmap_event); +} + +/* + * IRQ throttle logging + */ + +static void perf_log_throttle(struct perf_event *event, int enable) +{ + struct perf_output_handle handle; + int ret; + + struct { + struct perf_event_header header; + u64 time; + u64 id; + u64 stream_id; + } throttle_event = { + .header = { + .type = PERF_RECORD_THROTTLE, + .misc = 0, + .size = sizeof(throttle_event), + }, + .time = perf_clock(), + .id = primary_event_id(event), + .stream_id = event->id, + }; + + if (enable) + throttle_event.header.type = PERF_RECORD_UNTHROTTLE; + + ret = perf_output_begin(&handle, event, sizeof(throttle_event), 1, 0); + if (ret) + return; + + perf_output_put(&handle, throttle_event); + perf_output_end(&handle); +} + +/* + * Generic event overflow handling, sampling. + */ + +static int __perf_event_overflow(struct perf_event *event, int nmi, + int throttle, struct perf_sample_data *data, + struct pt_regs *regs) +{ + int events = atomic_read(&event->event_limit); + struct hw_perf_event *hwc = &event->hw; + int ret = 0; + + throttle = (throttle && event->pmu->unthrottle != NULL); + + if (!throttle) { + hwc->interrupts++; + } else { + if (hwc->interrupts != MAX_INTERRUPTS) { + hwc->interrupts++; + if (HZ * hwc->interrupts > + (u64)sysctl_perf_event_sample_rate) { + hwc->interrupts = MAX_INTERRUPTS; + perf_log_throttle(event, 0); + ret = 1; + } + } else { + /* + * Keep re-disabling events even though on the previous + * pass we disabled it - just in case we raced with a + * sched-in and the event got enabled again: + */ + ret = 1; + } + } + + if (event->attr.freq) { + u64 now = perf_clock(); + s64 delta = now - hwc->freq_stamp; + + hwc->freq_stamp = now; + + if (delta > 0 && delta < TICK_NSEC) + perf_adjust_period(event, NSEC_PER_SEC / (int)delta); + } + + /* + * XXX event_limit might not quite work as expected on inherited + * events + */ + + event->pending_kill = POLL_IN; + if (events && atomic_dec_and_test(&event->event_limit)) { + ret = 1; + event->pending_kill = POLL_HUP; + if (nmi) { + event->pending_disable = 1; + perf_pending_queue(&event->pending, + perf_pending_event); + } else + perf_event_disable(event); + } + + perf_event_output(event, nmi, data, regs); + return ret; +} + +int perf_event_overflow(struct perf_event *event, int nmi, + struct perf_sample_data *data, + struct pt_regs *regs) +{ + return __perf_event_overflow(event, nmi, 1, data, regs); +} + +/* + * Generic software event infrastructure + */ + +/* + * We directly increment event->count and keep a second value in + * event->hw.period_left to count intervals. This period event + * is kept in the range [-sample_period, 0] so that we can use the + * sign as trigger. + */ + +static u64 perf_swevent_set_period(struct perf_event *event) +{ + struct hw_perf_event *hwc = &event->hw; + u64 period = hwc->last_period; + u64 nr, offset; + s64 old, val; + + hwc->last_period = hwc->sample_period; + +again: + old = val = atomic64_read(&hwc->period_left); + if (val < 0) + return 0; + + nr = div64_u64(period + val, period); + offset = nr * period; + val -= offset; + if (atomic64_cmpxchg(&hwc->period_left, old, val) != old) + goto again; + + return nr; +} + +static void perf_swevent_overflow(struct perf_event *event, + int nmi, struct perf_sample_data *data, + struct pt_regs *regs) +{ + struct hw_perf_event *hwc = &event->hw; + int throttle = 0; + u64 overflow; + + data->period = event->hw.last_period; + overflow = perf_swevent_set_period(event); + + if (hwc->interrupts == MAX_INTERRUPTS) + return; + + for (; overflow; overflow--) { + if (__perf_event_overflow(event, nmi, throttle, + data, regs)) { + /* + * We inhibit the overflow from happening when + * hwc->interrupts == MAX_INTERRUPTS. + */ + break; + } + throttle = 1; + } +} + +static void perf_swevent_unthrottle(struct perf_event *event) +{ + /* + * Nothing to do, we already reset hwc->interrupts. + */ +} + +static void perf_swevent_add(struct perf_event *event, u64 nr, + int nmi, struct perf_sample_data *data, + struct pt_regs *regs) +{ + struct hw_perf_event *hwc = &event->hw; + + atomic64_add(nr, &event->count); + + if (!hwc->sample_period) + return; + + if (!regs) + return; + + if (!atomic64_add_negative(nr, &hwc->period_left)) + perf_swevent_overflow(event, nmi, data, regs); +} + +static int perf_swevent_is_counting(struct perf_event *event) +{ + /* + * The event is active, we're good! + */ + if (event->state == PERF_EVENT_STATE_ACTIVE) + return 1; + + /* + * The event is off/error, not counting. + */ + if (event->state != PERF_EVENT_STATE_INACTIVE) + return 0; + + /* + * The event is inactive, if the context is active + * we're part of a group that didn't make it on the 'pmu', + * not counting. + */ + if (event->ctx->is_active) + return 0; + + /* + * We're inactive and the context is too, this means the + * task is scheduled out, we're counting events that happen + * to us, like migration events. + */ + return 1; +} + +static int perf_swevent_match(struct perf_event *event, + enum perf_type_id type, + u32 event_id, struct pt_regs *regs) +{ + if (!perf_swevent_is_counting(event)) + return 0; + + if (event->attr.type != type) + return 0; + if (event->attr.config != event_id) + return 0; + + if (regs) { + if (event->attr.exclude_user && user_mode(regs)) + return 0; + + if (event->attr.exclude_kernel && !user_mode(regs)) + return 0; + } + + return 1; +} + +static void perf_swevent_ctx_event(struct perf_event_context *ctx, + enum perf_type_id type, + u32 event_id, u64 nr, int nmi, + struct perf_sample_data *data, + struct pt_regs *regs) +{ + struct perf_event *event; + + if (system_state != SYSTEM_RUNNING || list_empty(&ctx->event_list)) + return; + + rcu_read_lock(); + list_for_each_entry_rcu(event, &ctx->event_list, event_entry) { + if (perf_swevent_match(event, type, event_id, regs)) + perf_swevent_add(event, nr, nmi, data, regs); + } + rcu_read_unlock(); +} + +static int *perf_swevent_recursion_context(struct perf_cpu_context *cpuctx) +{ + if (in_nmi()) + return &cpuctx->recursion[3]; + + if (in_irq()) + return &cpuctx->recursion[2]; + + if (in_softirq()) + return &cpuctx->recursion[1]; + + return &cpuctx->recursion[0]; +} + +static void do_perf_sw_event(enum perf_type_id type, u32 event_id, + u64 nr, int nmi, + struct perf_sample_data *data, + struct pt_regs *regs) +{ + struct perf_cpu_context *cpuctx = &get_cpu_var(perf_cpu_context); + int *recursion = perf_swevent_recursion_context(cpuctx); + struct perf_event_context *ctx; + + if (*recursion) + goto out; + + (*recursion)++; + barrier(); + + perf_swevent_ctx_event(&cpuctx->ctx, type, event_id, + nr, nmi, data, regs); + rcu_read_lock(); + /* + * doesn't really matter which of the child contexts the + * events ends up in. + */ + ctx = rcu_dereference(current->perf_event_ctxp); + if (ctx) + perf_swevent_ctx_event(ctx, type, event_id, nr, nmi, data, regs); + rcu_read_unlock(); + + barrier(); + (*recursion)--; + +out: + put_cpu_var(perf_cpu_context); +} + +void __perf_sw_event(u32 event_id, u64 nr, int nmi, + struct pt_regs *regs, u64 addr) +{ + struct perf_sample_data data = { + .addr = addr, + }; + + do_perf_sw_event(PERF_TYPE_SOFTWARE, event_id, nr, nmi, + &data, regs); +} + +static void perf_swevent_read(struct perf_event *event) +{ +} + +static int perf_swevent_enable(struct perf_event *event) +{ + struct hw_perf_event *hwc = &event->hw; + + if (hwc->sample_period) { + hwc->last_period = hwc->sample_period; + perf_swevent_set_period(event); + } + return 0; +} + +static void perf_swevent_disable(struct perf_event *event) +{ +} + +static const struct pmu perf_ops_generic = { + .enable = perf_swevent_enable, + .disable = perf_swevent_disable, + .read = perf_swevent_read, + .unthrottle = perf_swevent_unthrottle, +}; + +/* + * hrtimer based swevent callback + */ + +static enum hrtimer_restart perf_swevent_hrtimer(struct hrtimer *hrtimer) +{ + enum hrtimer_restart ret = HRTIMER_RESTART; + struct perf_sample_data data; + struct pt_regs *regs; + struct perf_event *event; + u64 period; + + event = container_of(hrtimer, struct perf_event, hw.hrtimer); + event->pmu->read(event); + + data.addr = 0; + regs = get_irq_regs(); + /* + * In case we exclude kernel IPs or are somehow not in interrupt + * context, provide the next best thing, the user IP. + */ + if ((event->attr.exclude_kernel || !regs) && + !event->attr.exclude_user) + regs = task_pt_regs(current); + + if (regs) { + if (perf_event_overflow(event, 0, &data, regs)) + ret = HRTIMER_NORESTART; + } + + period = max_t(u64, 10000, event->hw.sample_period); + hrtimer_forward_now(hrtimer, ns_to_ktime(period)); + + return ret; +} + +/* + * Software event: cpu wall time clock + */ + +static void cpu_clock_perf_event_update(struct perf_event *event) +{ + int cpu = raw_smp_processor_id(); + s64 prev; + u64 now; + + now = cpu_clock(cpu); + prev = atomic64_read(&event->hw.prev_count); + atomic64_set(&event->hw.prev_count, now); + atomic64_add(now - prev, &event->count); +} + +static int cpu_clock_perf_event_enable(struct perf_event *event) +{ + struct hw_perf_event *hwc = &event->hw; + int cpu = raw_smp_processor_id(); + + atomic64_set(&hwc->prev_count, cpu_clock(cpu)); + hrtimer_init(&hwc->hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); + hwc->hrtimer.function = perf_swevent_hrtimer; + if (hwc->sample_period) { + u64 period = max_t(u64, 10000, hwc->sample_period); + __hrtimer_start_range_ns(&hwc->hrtimer, + ns_to_ktime(period), 0, + HRTIMER_MODE_REL, 0); + } + + return 0; +} + +static void cpu_clock_perf_event_disable(struct perf_event *event) +{ + if (event->hw.sample_period) + hrtimer_cancel(&event->hw.hrtimer); + cpu_clock_perf_event_update(event); +} + +static void cpu_clock_perf_event_read(struct perf_event *event) +{ + cpu_clock_perf_event_update(event); +} + +static const struct pmu perf_ops_cpu_clock = { + .enable = cpu_clock_perf_event_enable, + .disable = cpu_clock_perf_event_disable, + .read = cpu_clock_perf_event_read, +}; + +/* + * Software event: task time clock + */ + +static void task_clock_perf_event_update(struct perf_event *event, u64 now) +{ + u64 prev; + s64 delta; + + prev = atomic64_xchg(&event->hw.prev_count, now); + delta = now - prev; + atomic64_add(delta, &event->count); +} + +static int task_clock_perf_event_enable(struct perf_event *event) +{ + struct hw_perf_event *hwc = &event->hw; + u64 now; + + now = event->ctx->time; + + atomic64_set(&hwc->prev_count, now); + hrtimer_init(&hwc->hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); + hwc->hrtimer.function = perf_swevent_hrtimer; + if (hwc->sample_period) { + u64 period = max_t(u64, 10000, hwc->sample_period); + __hrtimer_start_range_ns(&hwc->hrtimer, + ns_to_ktime(period), 0, + HRTIMER_MODE_REL, 0); + } + + return 0; +} + +static void task_clock_perf_event_disable(struct perf_event *event) +{ + if (event->hw.sample_period) + hrtimer_cancel(&event->hw.hrtimer); + task_clock_perf_event_update(event, event->ctx->time); + +} + +static void task_clock_perf_event_read(struct perf_event *event) +{ + u64 time; + + if (!in_nmi()) { + update_context_time(event->ctx); + time = event->ctx->time; + } else { + u64 now = perf_clock(); + u64 delta = now - event->ctx->timestamp; + time = event->ctx->time + delta; + } + + task_clock_perf_event_update(event, time); +} + +static const struct pmu perf_ops_task_clock = { + .enable = task_clock_perf_event_enable, + .disable = task_clock_perf_event_disable, + .read = task_clock_perf_event_read, +}; + +#ifdef CONFIG_EVENT_PROFILE +void perf_tp_event(int event_id, u64 addr, u64 count, void *record, + int entry_size) +{ + struct perf_raw_record raw = { + .size = entry_size, + .data = record, + }; + + struct perf_sample_data data = { + .addr = addr, + .raw = &raw, + }; + + struct pt_regs *regs = get_irq_regs(); + + if (!regs) + regs = task_pt_regs(current); + + do_perf_sw_event(PERF_TYPE_TRACEPOINT, event_id, count, 1, + &data, regs); +} +EXPORT_SYMBOL_GPL(perf_tp_event); + +extern int ftrace_profile_enable(int); +extern void ftrace_profile_disable(int); + +static void tp_perf_event_destroy(struct perf_event *event) +{ + ftrace_profile_disable(event->attr.config); +} + +static const struct pmu *tp_perf_event_init(struct perf_event *event) +{ + /* + * Raw tracepoint data is a severe data leak, only allow root to + * have these. + */ + if ((event->attr.sample_type & PERF_SAMPLE_RAW) && + perf_paranoid_tracepoint_raw() && + !capable(CAP_SYS_ADMIN)) + return ERR_PTR(-EPERM); + + if (ftrace_profile_enable(event->attr.config)) + return NULL; + + event->destroy = tp_perf_event_destroy; + + return &perf_ops_generic; +} +#else +static const struct pmu *tp_perf_event_init(struct perf_event *event) +{ + return NULL; +} +#endif + +atomic_t perf_swevent_enabled[PERF_COUNT_SW_MAX]; + +static void sw_perf_event_destroy(struct perf_event *event) +{ + u64 event_id = event->attr.config; + + WARN_ON(event->parent); + + atomic_dec(&perf_swevent_enabled[event_id]); +} + +static const struct pmu *sw_perf_event_init(struct perf_event *event) +{ + const struct pmu *pmu = NULL; + u64 event_id = event->attr.config; + + /* + * Software events (currently) can't in general distinguish + * between user, kernel and hypervisor events. + * However, context switches and cpu migrations are considered + * to be kernel events, and page faults are never hypervisor + * events. + */ + switch (event_id) { + case PERF_COUNT_SW_CPU_CLOCK: + pmu = &perf_ops_cpu_clock; + + break; + case PERF_COUNT_SW_TASK_CLOCK: + /* + * If the user instantiates this as a per-cpu event, + * use the cpu_clock event instead. + */ + if (event->ctx->task) + pmu = &perf_ops_task_clock; + else + pmu = &perf_ops_cpu_clock; + + break; + case PERF_COUNT_SW_PAGE_FAULTS: + case PERF_COUNT_SW_PAGE_FAULTS_MIN: + case PERF_COUNT_SW_PAGE_FAULTS_MAJ: + case PERF_COUNT_SW_CONTEXT_SWITCHES: + case PERF_COUNT_SW_CPU_MIGRATIONS: + if (!event->parent) { + atomic_inc(&perf_swevent_enabled[event_id]); + event->destroy = sw_perf_event_destroy; + } + pmu = &perf_ops_generic; + break; + } + + return pmu; +} + +/* + * Allocate and initialize a event structure + */ +static struct perf_event * +perf_event_alloc(struct perf_event_attr *attr, + int cpu, + struct perf_event_context *ctx, + struct perf_event *group_leader, + struct perf_event *parent_event, + gfp_t gfpflags) +{ + const struct pmu *pmu; + struct perf_event *event; + struct hw_perf_event *hwc; + long err; + + event = kzalloc(sizeof(*event), gfpflags); + if (!event) + return ERR_PTR(-ENOMEM); + + /* + * Single events are their own group leaders, with an + * empty sibling list: + */ + if (!group_leader) + group_leader = event; + + mutex_init(&event->child_mutex); + INIT_LIST_HEAD(&event->child_list); + + INIT_LIST_HEAD(&event->group_entry); + INIT_LIST_HEAD(&event->event_entry); + INIT_LIST_HEAD(&event->sibling_list); + init_waitqueue_head(&event->waitq); + + mutex_init(&event->mmap_mutex); + + event->cpu = cpu; + event->attr = *attr; + event->group_leader = group_leader; + event->pmu = NULL; + event->ctx = ctx; + event->oncpu = -1; + + event->parent = parent_event; + + event->ns = get_pid_ns(current->nsproxy->pid_ns); + event->id = atomic64_inc_return(&perf_event_id); + + event->state = PERF_EVENT_STATE_INACTIVE; + + if (attr->disabled) + event->state = PERF_EVENT_STATE_OFF; + + pmu = NULL; + + hwc = &event->hw; + hwc->sample_period = attr->sample_period; + if (attr->freq && attr->sample_freq) + hwc->sample_period = 1; + hwc->last_period = hwc->sample_period; + + atomic64_set(&hwc->period_left, hwc->sample_period); + + /* + * we currently do not support PERF_FORMAT_GROUP on inherited events + */ + if (attr->inherit && (attr->read_format & PERF_FORMAT_GROUP)) + goto done; + + switch (attr->type) { + case PERF_TYPE_RAW: + case PERF_TYPE_HARDWARE: + case PERF_TYPE_HW_CACHE: + pmu = hw_perf_event_init(event); + break; + + case PERF_TYPE_SOFTWARE: + pmu = sw_perf_event_init(event); + break; + + case PERF_TYPE_TRACEPOINT: + pmu = tp_perf_event_init(event); + break; + + default: + break; + } +done: + err = 0; + if (!pmu) + err = -EINVAL; + else if (IS_ERR(pmu)) + err = PTR_ERR(pmu); + + if (err) { + if (event->ns) + put_pid_ns(event->ns); + kfree(event); + return ERR_PTR(err); + } + + event->pmu = pmu; + + if (!event->parent) { + atomic_inc(&nr_events); + if (event->attr.mmap) + atomic_inc(&nr_mmap_events); + if (event->attr.comm) + atomic_inc(&nr_comm_events); + if (event->attr.task) + atomic_inc(&nr_task_events); + } + + return event; +} + +static int perf_copy_attr(struct perf_event_attr __user *uattr, + struct perf_event_attr *attr) +{ + u32 size; + int ret; + + if (!access_ok(VERIFY_WRITE, uattr, PERF_ATTR_SIZE_VER0)) + return -EFAULT; + + /* + * zero the full structure, so that a short copy will be nice. + */ + memset(attr, 0, sizeof(*attr)); + + ret = get_user(size, &uattr->size); + if (ret) + return ret; + + if (size > PAGE_SIZE) /* silly large */ + goto err_size; + + if (!size) /* abi compat */ + size = PERF_ATTR_SIZE_VER0; + + if (size < PERF_ATTR_SIZE_VER0) + goto err_size; + + /* + * If we're handed a bigger struct than we know of, + * ensure all the unknown bits are 0 - i.e. new + * user-space does not rely on any kernel feature + * extensions we dont know about yet. + */ + if (size > sizeof(*attr)) { + unsigned char __user *addr; + unsigned char __user *end; + unsigned char val; + + addr = (void __user *)uattr + sizeof(*attr); + end = (void __user *)uattr + size; + + for (; addr < end; addr++) { + ret = get_user(val, addr); + if (ret) + return ret; + if (val) + goto err_size; + } + size = sizeof(*attr); + } + + ret = copy_from_user(attr, uattr, size); + if (ret) + return -EFAULT; + + /* + * If the type exists, the corresponding creation will verify + * the attr->config. + */ + if (attr->type >= PERF_TYPE_MAX) + return -EINVAL; + + if (attr->__reserved_1 || attr->__reserved_2 || attr->__reserved_3) + return -EINVAL; + + if (attr->sample_type & ~(PERF_SAMPLE_MAX-1)) + return -EINVAL; + + if (attr->read_format & ~(PERF_FORMAT_MAX-1)) + return -EINVAL; + +out: + return ret; + +err_size: + put_user(sizeof(*attr), &uattr->size); + ret = -E2BIG; + goto out; +} + +int perf_event_set_output(struct perf_event *event, int output_fd) +{ + struct perf_event *output_event = NULL; + struct file *output_file = NULL; + struct perf_event *old_output; + int fput_needed = 0; + int ret = -EINVAL; + + if (!output_fd) + goto set; + + output_file = fget_light(output_fd, &fput_needed); + if (!output_file) + return -EBADF; + + if (output_file->f_op != &perf_fops) + goto out; + + output_event = output_file->private_data; + + /* Don't chain output fds */ + if (output_event->output) + goto out; + + /* Don't set an output fd when we already have an output channel */ + if (event->data) + goto out; + + atomic_long_inc(&output_file->f_count); + +set: + mutex_lock(&event->mmap_mutex); + old_output = event->output; + rcu_assign_pointer(event->output, output_event); + mutex_unlock(&event->mmap_mutex); + + if (old_output) { + /* + * we need to make sure no existing perf_output_*() + * is still referencing this event. + */ + synchronize_rcu(); + fput(old_output->filp); + } + + ret = 0; +out: + fput_light(output_file, fput_needed); + return ret; +} + +/** + * sys_perf_event_open - open a performance event, associate it to a task/cpu + * + * @attr_uptr: event_id type attributes for monitoring/sampling + * @pid: target pid + * @cpu: target cpu + * @group_fd: group leader event fd + */ +SYSCALL_DEFINE5(perf_event_open, + struct perf_event_attr __user *, attr_uptr, + pid_t, pid, int, cpu, int, group_fd, unsigned long, flags) +{ + struct perf_event *event, *group_leader; + struct perf_event_attr attr; + struct perf_event_context *ctx; + struct file *event_file = NULL; + struct file *group_file = NULL; + int fput_needed = 0; + int fput_needed2 = 0; + int err; + + /* for future expandability... */ + if (flags & ~(PERF_FLAG_FD_NO_GROUP | PERF_FLAG_FD_OUTPUT)) + return -EINVAL; + + err = perf_copy_attr(attr_uptr, &attr); + if (err) + return err; + + if (!attr.exclude_kernel) { + if (perf_paranoid_kernel() && !capable(CAP_SYS_ADMIN)) + return -EACCES; + } + + if (attr.freq) { + if (attr.sample_freq > sysctl_perf_event_sample_rate) + return -EINVAL; + } + + /* + * Get the target context (task or percpu): + */ + ctx = find_get_context(pid, cpu); + if (IS_ERR(ctx)) + return PTR_ERR(ctx); + + /* + * Look up the group leader (we will attach this event to it): + */ + group_leader = NULL; + if (group_fd != -1 && !(flags & PERF_FLAG_FD_NO_GROUP)) { + err = -EINVAL; + group_file = fget_light(group_fd, &fput_needed); + if (!group_file) + goto err_put_context; + if (group_file->f_op != &perf_fops) + goto err_put_context; + + group_leader = group_file->private_data; + /* + * Do not allow a recursive hierarchy (this new sibling + * becoming part of another group-sibling): + */ + if (group_leader->group_leader != group_leader) + goto err_put_context; + /* + * Do not allow to attach to a group in a different + * task or CPU context: + */ + if (group_leader->ctx != ctx) + goto err_put_context; + /* + * Only a group leader can be exclusive or pinned + */ + if (attr.exclusive || attr.pinned) + goto err_put_context; + } + + event = perf_event_alloc(&attr, cpu, ctx, group_leader, + NULL, GFP_KERNEL); + err = PTR_ERR(event); + if (IS_ERR(event)) + goto err_put_context; + + err = anon_inode_getfd("[perf_event]", &perf_fops, event, 0); + if (err < 0) + goto err_free_put_context; + + event_file = fget_light(err, &fput_needed2); + if (!event_file) + goto err_free_put_context; + + if (flags & PERF_FLAG_FD_OUTPUT) { + err = perf_event_set_output(event, group_fd); + if (err) + goto err_fput_free_put_context; + } + + event->filp = event_file; + WARN_ON_ONCE(ctx->parent_ctx); + mutex_lock(&ctx->mutex); + perf_install_in_context(ctx, event, cpu); + ++ctx->generation; + mutex_unlock(&ctx->mutex); + + event->owner = current; + get_task_struct(current); + mutex_lock(¤t->perf_event_mutex); + list_add_tail(&event->owner_entry, ¤t->perf_event_list); + mutex_unlock(¤t->perf_event_mutex); + +err_fput_free_put_context: + fput_light(event_file, fput_needed2); + +err_free_put_context: + if (err < 0) + kfree(event); + +err_put_context: + if (err < 0) + put_ctx(ctx); + + fput_light(group_file, fput_needed); + + return err; +} + +/* + * inherit a event from parent task to child task: + */ +static struct perf_event * +inherit_event(struct perf_event *parent_event, + struct task_struct *parent, + struct perf_event_context *parent_ctx, + struct task_struct *child, + struct perf_event *group_leader, + struct perf_event_context *child_ctx) +{ + struct perf_event *child_event; + + /* + * Instead of creating recursive hierarchies of events, + * we link inherited events back to the original parent, + * which has a filp for sure, which we use as the reference + * count: + */ + if (parent_event->parent) + parent_event = parent_event->parent; + + child_event = perf_event_alloc(&parent_event->attr, + parent_event->cpu, child_ctx, + group_leader, parent_event, + GFP_KERNEL); + if (IS_ERR(child_event)) + return child_event; + get_ctx(child_ctx); + + /* + * Make the child state follow the state of the parent event, + * not its attr.disabled bit. We hold the parent's mutex, + * so we won't race with perf_event_{en, dis}able_family. + */ + if (parent_event->state >= PERF_EVENT_STATE_INACTIVE) + child_event->state = PERF_EVENT_STATE_INACTIVE; + else + child_event->state = PERF_EVENT_STATE_OFF; + + if (parent_event->attr.freq) + child_event->hw.sample_period = parent_event->hw.sample_period; + + /* + * Link it up in the child's context: + */ + add_event_to_ctx(child_event, child_ctx); + + /* + * Get a reference to the parent filp - we will fput it + * when the child event exits. This is safe to do because + * we are in the parent and we know that the filp still + * exists and has a nonzero count: + */ + atomic_long_inc(&parent_event->filp->f_count); + + /* + * Link this into the parent event's child list + */ + WARN_ON_ONCE(parent_event->ctx->parent_ctx); + mutex_lock(&parent_event->child_mutex); + list_add_tail(&child_event->child_list, &parent_event->child_list); + mutex_unlock(&parent_event->child_mutex); + + return child_event; +} + +static int inherit_group(struct perf_event *parent_event, + struct task_struct *parent, + struct perf_event_context *parent_ctx, + struct task_struct *child, + struct perf_event_context *child_ctx) +{ + struct perf_event *leader; + struct perf_event *sub; + struct perf_event *child_ctr; + + leader = inherit_event(parent_event, parent, parent_ctx, + child, NULL, child_ctx); + if (IS_ERR(leader)) + return PTR_ERR(leader); + list_for_each_entry(sub, &parent_event->sibling_list, group_entry) { + child_ctr = inherit_event(sub, parent, parent_ctx, + child, leader, child_ctx); + if (IS_ERR(child_ctr)) + return PTR_ERR(child_ctr); + } + return 0; +} + +static void sync_child_event(struct perf_event *child_event, + struct task_struct *child) +{ + struct perf_event *parent_event = child_event->parent; + u64 child_val; + + if (child_event->attr.inherit_stat) + perf_event_read_event(child_event, child); + + child_val = atomic64_read(&child_event->count); + + /* + * Add back the child's count to the parent's count: + */ + atomic64_add(child_val, &parent_event->count); + atomic64_add(child_event->total_time_enabled, + &parent_event->child_total_time_enabled); + atomic64_add(child_event->total_time_running, + &parent_event->child_total_time_running); + + /* + * Remove this event from the parent's list + */ + WARN_ON_ONCE(parent_event->ctx->parent_ctx); + mutex_lock(&parent_event->child_mutex); + list_del_init(&child_event->child_list); + mutex_unlock(&parent_event->child_mutex); + + /* + * Release the parent event, if this was the last + * reference to it. + */ + fput(parent_event->filp); +} + +static void +__perf_event_exit_task(struct perf_event *child_event, + struct perf_event_context *child_ctx, + struct task_struct *child) +{ + struct perf_event *parent_event; + + update_event_times(child_event); + perf_event_remove_from_context(child_event); + + parent_event = child_event->parent; + /* + * It can happen that parent exits first, and has events + * that are still around due to the child reference. These + * events need to be zapped - but otherwise linger. + */ + if (parent_event) { + sync_child_event(child_event, child); + free_event(child_event); + } +} + +/* + * When a child task exits, feed back event values to parent events. + */ +void perf_event_exit_task(struct task_struct *child) +{ + struct perf_event *child_event, *tmp; + struct perf_event_context *child_ctx; + unsigned long flags; + + if (likely(!child->perf_event_ctxp)) { + perf_event_task(child, NULL, 0); + return; + } + + local_irq_save(flags); + /* + * We can't reschedule here because interrupts are disabled, + * and either child is current or it is a task that can't be + * scheduled, so we are now safe from rescheduling changing + * our context. + */ + child_ctx = child->perf_event_ctxp; + __perf_event_task_sched_out(child_ctx); + + /* + * Take the context lock here so that if find_get_context is + * reading child->perf_event_ctxp, we wait until it has + * incremented the context's refcount before we do put_ctx below. + */ + spin_lock(&child_ctx->lock); + child->perf_event_ctxp = NULL; + /* + * If this context is a clone; unclone it so it can't get + * swapped to another process while we're removing all + * the events from it. + */ + unclone_ctx(child_ctx); + spin_unlock_irqrestore(&child_ctx->lock, flags); + + /* + * Report the task dead after unscheduling the events so that we + * won't get any samples after PERF_RECORD_EXIT. We can however still + * get a few PERF_RECORD_READ events. + */ + perf_event_task(child, child_ctx, 0); + + /* + * We can recurse on the same lock type through: + * + * __perf_event_exit_task() + * sync_child_event() + * fput(parent_event->filp) + * perf_release() + * mutex_lock(&ctx->mutex) + * + * But since its the parent context it won't be the same instance. + */ + mutex_lock_nested(&child_ctx->mutex, SINGLE_DEPTH_NESTING); + +again: + list_for_each_entry_safe(child_event, tmp, &child_ctx->group_list, + group_entry) + __perf_event_exit_task(child_event, child_ctx, child); + + /* + * If the last event was a group event, it will have appended all + * its siblings to the list, but we obtained 'tmp' before that which + * will still point to the list head terminating the iteration. + */ + if (!list_empty(&child_ctx->group_list)) + goto again; + + mutex_unlock(&child_ctx->mutex); + + put_ctx(child_ctx); +} + +/* + * free an unexposed, unused context as created by inheritance by + * init_task below, used by fork() in case of fail. + */ +void perf_event_free_task(struct task_struct *task) +{ + struct perf_event_context *ctx = task->perf_event_ctxp; + struct perf_event *event, *tmp; + + if (!ctx) + return; + + mutex_lock(&ctx->mutex); +again: + list_for_each_entry_safe(event, tmp, &ctx->group_list, group_entry) { + struct perf_event *parent = event->parent; + + if (WARN_ON_ONCE(!parent)) + continue; + + mutex_lock(&parent->child_mutex); + list_del_init(&event->child_list); + mutex_unlock(&parent->child_mutex); + + fput(parent->filp); + + list_del_event(event, ctx); + free_event(event); + } + + if (!list_empty(&ctx->group_list)) + goto again; + + mutex_unlock(&ctx->mutex); + + put_ctx(ctx); +} + +/* + * Initialize the perf_event context in task_struct + */ +int perf_event_init_task(struct task_struct *child) +{ + struct perf_event_context *child_ctx, *parent_ctx; + struct perf_event_context *cloned_ctx; + struct perf_event *event; + struct task_struct *parent = current; + int inherited_all = 1; + int ret = 0; + + child->perf_event_ctxp = NULL; + + mutex_init(&child->perf_event_mutex); + INIT_LIST_HEAD(&child->perf_event_list); + + if (likely(!parent->perf_event_ctxp)) + return 0; + + /* + * This is executed from the parent task context, so inherit + * events that have been marked for cloning. + * First allocate and initialize a context for the child. + */ + + child_ctx = kmalloc(sizeof(struct perf_event_context), GFP_KERNEL); + if (!child_ctx) + return -ENOMEM; + + __perf_event_init_context(child_ctx, child); + child->perf_event_ctxp = child_ctx; + get_task_struct(child); + + /* + * If the parent's context is a clone, pin it so it won't get + * swapped under us. + */ + parent_ctx = perf_pin_task_context(parent); + + /* + * No need to check if parent_ctx != NULL here; since we saw + * it non-NULL earlier, the only reason for it to become NULL + * is if we exit, and since we're currently in the middle of + * a fork we can't be exiting at the same time. + */ + + /* + * Lock the parent list. No need to lock the child - not PID + * hashed yet and not running, so nobody can access it. + */ + mutex_lock(&parent_ctx->mutex); + + /* + * We dont have to disable NMIs - we are only looking at + * the list, not manipulating it: + */ + list_for_each_entry_rcu(event, &parent_ctx->event_list, event_entry) { + if (event != event->group_leader) + continue; + + if (!event->attr.inherit) { + inherited_all = 0; + continue; + } + + ret = inherit_group(event, parent, parent_ctx, + child, child_ctx); + if (ret) { + inherited_all = 0; + break; + } + } + + if (inherited_all) { + /* + * Mark the child context as a clone of the parent + * context, or of whatever the parent is a clone of. + * Note that if the parent is a clone, it could get + * uncloned at any point, but that doesn't matter + * because the list of events and the generation + * count can't have changed since we took the mutex. + */ + cloned_ctx = rcu_dereference(parent_ctx->parent_ctx); + if (cloned_ctx) { + child_ctx->parent_ctx = cloned_ctx; + child_ctx->parent_gen = parent_ctx->parent_gen; + } else { + child_ctx->parent_ctx = parent_ctx; + child_ctx->parent_gen = parent_ctx->generation; + } + get_ctx(child_ctx->parent_ctx); + } + + mutex_unlock(&parent_ctx->mutex); + + perf_unpin_context(parent_ctx); + + return ret; +} + +static void __cpuinit perf_event_init_cpu(int cpu) +{ + struct perf_cpu_context *cpuctx; + + cpuctx = &per_cpu(perf_cpu_context, cpu); + __perf_event_init_context(&cpuctx->ctx, NULL); + + spin_lock(&perf_resource_lock); + cpuctx->max_pertask = perf_max_events - perf_reserved_percpu; + spin_unlock(&perf_resource_lock); + + hw_perf_event_setup(cpu); +} + +#ifdef CONFIG_HOTPLUG_CPU +static void __perf_event_exit_cpu(void *info) +{ + struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); + struct perf_event_context *ctx = &cpuctx->ctx; + struct perf_event *event, *tmp; + + list_for_each_entry_safe(event, tmp, &ctx->group_list, group_entry) + __perf_event_remove_from_context(event); +} +static void perf_event_exit_cpu(int cpu) +{ + struct perf_cpu_context *cpuctx = &per_cpu(perf_cpu_context, cpu); + struct perf_event_context *ctx = &cpuctx->ctx; + + mutex_lock(&ctx->mutex); + smp_call_function_single(cpu, __perf_event_exit_cpu, NULL, 1); + mutex_unlock(&ctx->mutex); +} +#else +static inline void perf_event_exit_cpu(int cpu) { } +#endif + +static int __cpuinit +perf_cpu_notify(struct notifier_block *self, unsigned long action, void *hcpu) +{ + unsigned int cpu = (long)hcpu; + + switch (action) { + + case CPU_UP_PREPARE: + case CPU_UP_PREPARE_FROZEN: + perf_event_init_cpu(cpu); + break; + + case CPU_ONLINE: + case CPU_ONLINE_FROZEN: + hw_perf_event_setup_online(cpu); + break; + + case CPU_DOWN_PREPARE: + case CPU_DOWN_PREPARE_FROZEN: + perf_event_exit_cpu(cpu); + break; + + default: + break; + } + + return NOTIFY_OK; +} + +/* + * This has to have a higher priority than migration_notifier in sched.c. + */ +static struct notifier_block __cpuinitdata perf_cpu_nb = { + .notifier_call = perf_cpu_notify, + .priority = 20, +}; + +void __init perf_event_init(void) +{ + perf_cpu_notify(&perf_cpu_nb, (unsigned long)CPU_UP_PREPARE, + (void *)(long)smp_processor_id()); + perf_cpu_notify(&perf_cpu_nb, (unsigned long)CPU_ONLINE, + (void *)(long)smp_processor_id()); + register_cpu_notifier(&perf_cpu_nb); +} + +static ssize_t perf_show_reserve_percpu(struct sysdev_class *class, char *buf) +{ + return sprintf(buf, "%d\n", perf_reserved_percpu); +} + +static ssize_t +perf_set_reserve_percpu(struct sysdev_class *class, + const char *buf, + size_t count) +{ + struct perf_cpu_context *cpuctx; + unsigned long val; + int err, cpu, mpt; + + err = strict_strtoul(buf, 10, &val); + if (err) + return err; + if (val > perf_max_events) + return -EINVAL; + + spin_lock(&perf_resource_lock); + perf_reserved_percpu = val; + for_each_online_cpu(cpu) { + cpuctx = &per_cpu(perf_cpu_context, cpu); + spin_lock_irq(&cpuctx->ctx.lock); + mpt = min(perf_max_events - cpuctx->ctx.nr_events, + perf_max_events - perf_reserved_percpu); + cpuctx->max_pertask = mpt; + spin_unlock_irq(&cpuctx->ctx.lock); + } + spin_unlock(&perf_resource_lock); + + return count; +} + +static ssize_t perf_show_overcommit(struct sysdev_class *class, char *buf) +{ + return sprintf(buf, "%d\n", perf_overcommit); +} + +static ssize_t +perf_set_overcommit(struct sysdev_class *class, const char *buf, size_t count) +{ + unsigned long val; + int err; + + err = strict_strtoul(buf, 10, &val); + if (err) + return err; + if (val > 1) + return -EINVAL; + + spin_lock(&perf_resource_lock); + perf_overcommit = val; + spin_unlock(&perf_resource_lock); + + return count; +} + +static SYSDEV_CLASS_ATTR( + reserve_percpu, + 0644, + perf_show_reserve_percpu, + perf_set_reserve_percpu + ); + +static SYSDEV_CLASS_ATTR( + overcommit, + 0644, + perf_show_overcommit, + perf_set_overcommit + ); + +static struct attribute *perfclass_attrs[] = { + &attr_reserve_percpu.attr, + &attr_overcommit.attr, + NULL +}; + +static struct attribute_group perfclass_attr_group = { + .attrs = perfclass_attrs, + .name = "perf_events", +}; + +static int __init perf_event_sysfs_init(void) +{ + return sysfs_create_group(&cpu_sysdev_class.kset.kobj, + &perfclass_attr_group); +} +device_initcall(perf_event_sysfs_init); diff --git a/kernel/sched.c b/kernel/sched.c index faf4d46..291c8d2 100644 --- a/kernel/sched.c +++ b/kernel/sched.c @@ -39,7 +39,7 @@ #include #include #include -#include +#include #include #include #include @@ -2059,7 +2059,7 @@ void set_task_cpu(struct task_struct *p, unsigned int new_cpu) if (task_hot(p, old_rq->clock, NULL)) schedstat_inc(p, se.nr_forced2_migrations); #endif - perf_swcounter_event(PERF_COUNT_SW_CPU_MIGRATIONS, + perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, 1, NULL, 0); } p->se.vruntime -= old_cfsrq->min_vruntime - @@ -2724,7 +2724,7 @@ static void finish_task_switch(struct rq *rq, struct task_struct *prev) */ prev_state = prev->state; finish_arch_switch(prev); - perf_counter_task_sched_in(current, cpu_of(rq)); + perf_event_task_sched_in(current, cpu_of(rq)); finish_lock_switch(rq, prev); fire_sched_in_preempt_notifiers(current); @@ -5199,7 +5199,7 @@ void scheduler_tick(void) curr->sched_class->task_tick(rq, curr, 0); spin_unlock(&rq->lock); - perf_counter_task_tick(curr, cpu); + perf_event_task_tick(curr, cpu); #ifdef CONFIG_SMP rq->idle_at_tick = idle_cpu(cpu); @@ -5415,7 +5415,7 @@ need_resched_nonpreemptible: if (likely(prev != next)) { sched_info_switch(prev, next); - perf_counter_task_sched_out(prev, next, cpu); + perf_event_task_sched_out(prev, next, cpu); rq->nr_switches++; rq->curr = next; @@ -7692,7 +7692,7 @@ migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) /* * Register at high priority so that task migration (migrate_all_tasks) * happens before everything else. This has to be lower priority than - * the notifier in the perf_counter subsystem, though. + * the notifier in the perf_event subsystem, though. */ static struct notifier_block __cpuinitdata migration_notifier = { .notifier_call = migration_call, @@ -9549,7 +9549,7 @@ void __init sched_init(void) alloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); #endif /* SMP */ - perf_counter_init(); + perf_event_init(); scheduler_running = 1; } diff --git a/kernel/sys.c b/kernel/sys.c index b3f1097..ea5c3bc 100644 --- a/kernel/sys.c +++ b/kernel/sys.c @@ -14,7 +14,7 @@ #include #include #include -#include +#include #include #include #include @@ -1511,11 +1511,11 @@ SYSCALL_DEFINE5(prctl, int, option, unsigned long, arg2, unsigned long, arg3, case PR_SET_TSC: error = SET_TSC_CTL(arg2); break; - case PR_TASK_PERF_COUNTERS_DISABLE: - error = perf_counter_task_disable(); + case PR_TASK_PERF_EVENTS_DISABLE: + error = perf_event_task_disable(); break; - case PR_TASK_PERF_COUNTERS_ENABLE: - error = perf_counter_task_enable(); + case PR_TASK_PERF_EVENTS_ENABLE: + error = perf_event_task_enable(); break; case PR_GET_TIMERSLACK: error = current->timer_slack_ns; diff --git a/kernel/sys_ni.c b/kernel/sys_ni.c index 68320f6..515bc23 100644 --- a/kernel/sys_ni.c +++ b/kernel/sys_ni.c @@ -177,4 +177,4 @@ cond_syscall(sys_eventfd); cond_syscall(sys_eventfd2); /* performance counters: */ -cond_syscall(sys_perf_counter_open); +cond_syscall(sys_perf_event_open); diff --git a/kernel/sysctl.c b/kernel/sysctl.c index 1a631ba..6ba49c7 100644 --- a/kernel/sysctl.c +++ b/kernel/sysctl.c @@ -50,7 +50,7 @@ #include #include #include -#include +#include #include #include @@ -964,28 +964,28 @@ static struct ctl_table kern_table[] = { .child = slow_work_sysctls, }, #endif -#ifdef CONFIG_PERF_COUNTERS +#ifdef CONFIG_PERF_EVENTS { .ctl_name = CTL_UNNUMBERED, - .procname = "perf_counter_paranoid", - .data = &sysctl_perf_counter_paranoid, - .maxlen = sizeof(sysctl_perf_counter_paranoid), + .procname = "perf_event_paranoid", + .data = &sysctl_perf_event_paranoid, + .maxlen = sizeof(sysctl_perf_event_paranoid), .mode = 0644, .proc_handler = &proc_dointvec, }, { .ctl_name = CTL_UNNUMBERED, - .procname = "perf_counter_mlock_kb", - .data = &sysctl_perf_counter_mlock, - .maxlen = sizeof(sysctl_perf_counter_mlock), + .procname = "perf_event_mlock_kb", + .data = &sysctl_perf_event_mlock, + .maxlen = sizeof(sysctl_perf_event_mlock), .mode = 0644, .proc_handler = &proc_dointvec, }, { .ctl_name = CTL_UNNUMBERED, - .procname = "perf_counter_max_sample_rate", - .data = &sysctl_perf_counter_sample_rate, - .maxlen = sizeof(sysctl_perf_counter_sample_rate), + .procname = "perf_event_max_sample_rate", + .data = &sysctl_perf_event_sample_rate, + .maxlen = sizeof(sysctl_perf_event_sample_rate), .mode = 0644, .proc_handler = &proc_dointvec, }, diff --git a/kernel/timer.c b/kernel/timer.c index bbb5107..811e5c3 100644 --- a/kernel/timer.c +++ b/kernel/timer.c @@ -37,7 +37,7 @@ #include #include #include -#include +#include #include #include @@ -1187,7 +1187,7 @@ static void run_timer_softirq(struct softirq_action *h) { struct tvec_base *base = __get_cpu_var(tvec_bases); - perf_counter_do_pending(); + perf_event_do_pending(); hrtimer_run_pending(); diff --git a/kernel/trace/trace_syscalls.c b/kernel/trace/trace_syscalls.c index 8712ce3..233f348 100644 --- a/kernel/trace/trace_syscalls.c +++ b/kernel/trace/trace_syscalls.c @@ -2,7 +2,7 @@ #include #include #include -#include +#include #include #include "trace_output.h" @@ -414,7 +414,7 @@ static void prof_syscall_enter(struct pt_regs *regs, long id) rec->nr = syscall_nr; syscall_get_arguments(current, regs, 0, sys_data->nb_args, (unsigned long *)&rec->args); - perf_tpcounter_event(sys_data->enter_id, 0, 1, rec, size); + perf_tp_event(sys_data->enter_id, 0, 1, rec, size); } while(0); } @@ -476,7 +476,7 @@ static void prof_syscall_exit(struct pt_regs *regs, long ret) rec.nr = syscall_nr; rec.ret = syscall_get_return_value(current, regs); - perf_tpcounter_event(sys_data->exit_id, 0, 1, &rec, sizeof(rec)); + perf_tp_event(sys_data->exit_id, 0, 1, &rec, sizeof(rec)); } int reg_prof_syscall_exit(char *name) diff --git a/mm/mmap.c b/mm/mmap.c index 26892e3..376492e 100644 --- a/mm/mmap.c +++ b/mm/mmap.c @@ -28,7 +28,7 @@ #include #include #include -#include +#include #include #include @@ -1220,7 +1220,7 @@ munmap_back: if (correct_wcount) atomic_inc(&inode->i_writecount); out: - perf_counter_mmap(vma); + perf_event_mmap(vma); mm->total_vm += len >> PAGE_SHIFT; vm_stat_account(mm, vm_flags, file, len >> PAGE_SHIFT); @@ -2308,7 +2308,7 @@ int install_special_mapping(struct mm_struct *mm, mm->total_vm += len >> PAGE_SHIFT; - perf_counter_mmap(vma); + perf_event_mmap(vma); return 0; } diff --git a/mm/mprotect.c b/mm/mprotect.c index d80311b..8bc969d 100644 --- a/mm/mprotect.c +++ b/mm/mprotect.c @@ -23,7 +23,7 @@ #include #include #include -#include +#include #include #include #include @@ -300,7 +300,7 @@ SYSCALL_DEFINE3(mprotect, unsigned long, start, size_t, len, error = mprotect_fixup(vma, &prev, nstart, tmp, newflags); if (error) goto out; - perf_counter_mmap(vma); + perf_event_mmap(vma); nstart = tmp; if (nstart < prev->vm_end) diff --git a/tools/perf/Makefile b/tools/perf/Makefile index 0aba8b6..b5f1953 100644 --- a/tools/perf/Makefile +++ b/tools/perf/Makefile @@ -318,7 +318,7 @@ export PERL_PATH LIB_FILE=libperf.a -LIB_H += ../../include/linux/perf_counter.h +LIB_H += ../../include/linux/perf_event.h LIB_H += ../../include/linux/rbtree.h LIB_H += ../../include/linux/list.h LIB_H += util/include/linux/list.h diff --git a/tools/perf/builtin-annotate.c b/tools/perf/builtin-annotate.c index 043d85b..1ec7416 100644 --- a/tools/perf/builtin-annotate.c +++ b/tools/perf/builtin-annotate.c @@ -505,7 +505,7 @@ process_sample_event(event_t *event, unsigned long offset, unsigned long head) return -1; } - if (event->header.misc & PERF_EVENT_MISC_KERNEL) { + if (event->header.misc & PERF_RECORD_MISC_KERNEL) { show = SHOW_KERNEL; level = 'k'; @@ -513,7 +513,7 @@ process_sample_event(event_t *event, unsigned long offset, unsigned long head) dump_printf(" ...... dso: %s\n", dso->name); - } else if (event->header.misc & PERF_EVENT_MISC_USER) { + } else if (event->header.misc & PERF_RECORD_MISC_USER) { show = SHOW_USER; level = '.'; @@ -565,7 +565,7 @@ process_mmap_event(event_t *event, unsigned long offset, unsigned long head) thread = threads__findnew(event->mmap.pid, &threads, &last_match); - dump_printf("%p [%p]: PERF_EVENT_MMAP %d: [%p(%p) @ %p]: %s\n", + dump_printf("%p [%p]: PERF_RECORD_MMAP %d: [%p(%p) @ %p]: %s\n", (void *)(offset + head), (void *)(long)(event->header.size), event->mmap.pid, @@ -575,7 +575,7 @@ process_mmap_event(event_t *event, unsigned long offset, unsigned long head) event->mmap.filename); if (thread == NULL || map == NULL) { - dump_printf("problem processing PERF_EVENT_MMAP, skipping event.\n"); + dump_printf("problem processing PERF_RECORD_MMAP, skipping event.\n"); return 0; } @@ -591,14 +591,14 @@ process_comm_event(event_t *event, unsigned long offset, unsigned long head) struct thread *thread; thread = threads__findnew(event->comm.pid, &threads, &last_match); - dump_printf("%p [%p]: PERF_EVENT_COMM: %s:%d\n", + dump_printf("%p [%p]: PERF_RECORD_COMM: %s:%d\n", (void *)(offset + head), (void *)(long)(event->header.size), event->comm.comm, event->comm.pid); if (thread == NULL || thread__set_comm(thread, event->comm.comm)) { - dump_printf("problem processing PERF_EVENT_COMM, skipping event.\n"); + dump_printf("problem processing PERF_RECORD_COMM, skipping event.\n"); return -1; } total_comm++; @@ -614,7 +614,7 @@ process_fork_event(event_t *event, unsigned long offset, unsigned long head) thread = threads__findnew(event->fork.pid, &threads, &last_match); parent = threads__findnew(event->fork.ppid, &threads, &last_match); - dump_printf("%p [%p]: PERF_EVENT_FORK: %d:%d\n", + dump_printf("%p [%p]: PERF_RECORD_FORK: %d:%d\n", (void *)(offset + head), (void *)(long)(event->header.size), event->fork.pid, event->fork.ppid); @@ -627,7 +627,7 @@ process_fork_event(event_t *event, unsigned long offset, unsigned long head) return 0; if (!thread || !parent || thread__fork(thread, parent)) { - dump_printf("problem processing PERF_EVENT_FORK, skipping event.\n"); + dump_printf("problem processing PERF_RECORD_FORK, skipping event.\n"); return -1; } total_fork++; @@ -639,23 +639,23 @@ static int process_event(event_t *event, unsigned long offset, unsigned long head) { switch (event->header.type) { - case PERF_EVENT_SAMPLE: + case PERF_RECORD_SAMPLE: return process_sample_event(event, offset, head); - case PERF_EVENT_MMAP: + case PERF_RECORD_MMAP: return process_mmap_event(event, offset, head); - case PERF_EVENT_COMM: + case PERF_RECORD_COMM: return process_comm_event(event, offset, head); - case PERF_EVENT_FORK: + case PERF_RECORD_FORK: return process_fork_event(event, offset, head); /* * We dont process them right now but they are fine: */ - case PERF_EVENT_THROTTLE: - case PERF_EVENT_UNTHROTTLE: + case PERF_RECORD_THROTTLE: + case PERF_RECORD_UNTHROTTLE: return 0; default: diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c index 2459e5a..a5a050a 100644 --- a/tools/perf/builtin-record.c +++ b/tools/perf/builtin-record.c @@ -77,7 +77,7 @@ static struct mmap_data mmap_array[MAX_NR_CPUS][MAX_COUNTERS]; static unsigned long mmap_read_head(struct mmap_data *md) { - struct perf_counter_mmap_page *pc = md->base; + struct perf_event_mmap_page *pc = md->base; long head; head = pc->data_head; @@ -88,7 +88,7 @@ static unsigned long mmap_read_head(struct mmap_data *md) static void mmap_write_tail(struct mmap_data *md, unsigned long tail) { - struct perf_counter_mmap_page *pc = md->base; + struct perf_event_mmap_page *pc = md->base; /* * ensure all reads are done before we write the tail out. @@ -233,7 +233,7 @@ static pid_t pid_synthesize_comm_event(pid_t pid, int full) } } - comm_ev.header.type = PERF_EVENT_COMM; + comm_ev.header.type = PERF_RECORD_COMM; size = ALIGN(size, sizeof(u64)); comm_ev.header.size = sizeof(comm_ev) - (sizeof(comm_ev.comm) - size); @@ -288,7 +288,7 @@ static void pid_synthesize_mmap_samples(pid_t pid, pid_t tgid) while (1) { char bf[BUFSIZ], *pbf = bf; struct mmap_event mmap_ev = { - .header = { .type = PERF_EVENT_MMAP }, + .header = { .type = PERF_RECORD_MMAP }, }; int n; size_t size; @@ -355,7 +355,7 @@ static void synthesize_all(void) static int group_fd; -static struct perf_header_attr *get_header_attr(struct perf_counter_attr *a, int nr) +static struct perf_header_attr *get_header_attr(struct perf_event_attr *a, int nr) { struct perf_header_attr *h_attr; @@ -371,7 +371,7 @@ static struct perf_header_attr *get_header_attr(struct perf_counter_attr *a, int static void create_counter(int counter, int cpu, pid_t pid) { - struct perf_counter_attr *attr = attrs + counter; + struct perf_event_attr *attr = attrs + counter; struct perf_header_attr *h_attr; int track = !counter; /* only the first counter needs these */ struct { @@ -417,7 +417,7 @@ static void create_counter(int counter, int cpu, pid_t pid) attr->disabled = 1; try_again: - fd[nr_cpu][counter] = sys_perf_counter_open(attr, pid, cpu, group_fd, 0); + fd[nr_cpu][counter] = sys_perf_event_open(attr, pid, cpu, group_fd, 0); if (fd[nr_cpu][counter] < 0) { int err = errno; @@ -444,7 +444,7 @@ try_again: printf("\n"); error("perfcounter syscall returned with %d (%s)\n", fd[nr_cpu][counter], strerror(err)); - die("No CONFIG_PERF_COUNTERS=y kernel support configured?\n"); + die("No CONFIG_PERF_EVENTS=y kernel support configured?\n"); exit(-1); } @@ -478,7 +478,7 @@ try_again: if (multiplex && fd[nr_cpu][counter] != multiplex_fd) { int ret; - ret = ioctl(fd[nr_cpu][counter], PERF_COUNTER_IOC_SET_OUTPUT, multiplex_fd); + ret = ioctl(fd[nr_cpu][counter], PERF_EVENT_IOC_SET_OUTPUT, multiplex_fd); assert(ret != -1); } else { event_array[nr_poll].fd = fd[nr_cpu][counter]; @@ -496,7 +496,7 @@ try_again: } } - ioctl(fd[nr_cpu][counter], PERF_COUNTER_IOC_ENABLE); + ioctl(fd[nr_cpu][counter], PERF_EVENT_IOC_ENABLE); } static void open_counters(int cpu, pid_t pid) @@ -642,7 +642,7 @@ static int __cmd_record(int argc, const char **argv) if (done) { for (i = 0; i < nr_cpu; i++) { for (counter = 0; counter < nr_counters; counter++) - ioctl(fd[i][counter], PERF_COUNTER_IOC_DISABLE); + ioctl(fd[i][counter], PERF_EVENT_IOC_DISABLE); } } } diff --git a/tools/perf/builtin-report.c b/tools/perf/builtin-report.c index cdf9a8d..19669c2 100644 --- a/tools/perf/builtin-report.c +++ b/tools/perf/builtin-report.c @@ -1121,7 +1121,7 @@ process_sample_event(event_t *event, unsigned long offset, unsigned long head) more_data += sizeof(u64); } - dump_printf("%p [%p]: PERF_EVENT_SAMPLE (IP, %d): %d/%d: %p period: %Ld\n", + dump_printf("%p [%p]: PERF_RECORD_SAMPLE (IP, %d): %d/%d: %p period: %Ld\n", (void *)(offset + head), (void *)(long)(event->header.size), event->header.misc, @@ -1158,9 +1158,9 @@ process_sample_event(event_t *event, unsigned long offset, unsigned long head) if (comm_list && !strlist__has_entry(comm_list, thread->comm)) return 0; - cpumode = event->header.misc & PERF_EVENT_MISC_CPUMODE_MASK; + cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK; - if (cpumode == PERF_EVENT_MISC_KERNEL) { + if (cpumode == PERF_RECORD_MISC_KERNEL) { show = SHOW_KERNEL; level = 'k'; @@ -1168,7 +1168,7 @@ process_sample_event(event_t *event, unsigned long offset, unsigned long head) dump_printf(" ...... dso: %s\n", dso->name); - } else if (cpumode == PERF_EVENT_MISC_USER) { + } else if (cpumode == PERF_RECORD_MISC_USER) { show = SHOW_USER; level = '.'; @@ -1210,7 +1210,7 @@ process_mmap_event(event_t *event, unsigned long offset, unsigned long head) thread = threads__findnew(event->mmap.pid, &threads, &last_match); - dump_printf("%p [%p]: PERF_EVENT_MMAP %d/%d: [%p(%p) @ %p]: %s\n", + dump_printf("%p [%p]: PERF_RECORD_MMAP %d/%d: [%p(%p) @ %p]: %s\n", (void *)(offset + head), (void *)(long)(event->header.size), event->mmap.pid, @@ -1221,7 +1221,7 @@ process_mmap_event(event_t *event, unsigned long offset, unsigned long head) event->mmap.filename); if (thread == NULL || map == NULL) { - dump_printf("problem processing PERF_EVENT_MMAP, skipping event.\n"); + dump_printf("problem processing PERF_RECORD_MMAP, skipping event.\n"); return 0; } @@ -1238,14 +1238,14 @@ process_comm_event(event_t *event, unsigned long offset, unsigned long head) thread = threads__findnew(event->comm.pid, &threads, &last_match); - dump_printf("%p [%p]: PERF_EVENT_COMM: %s:%d\n", + dump_printf("%p [%p]: PERF_RECORD_COMM: %s:%d\n", (void *)(offset + head), (void *)(long)(event->header.size), event->comm.comm, event->comm.pid); if (thread == NULL || thread__set_comm_adjust(thread, event->comm.comm)) { - dump_printf("problem processing PERF_EVENT_COMM, skipping event.\n"); + dump_printf("problem processing PERF_RECORD_COMM, skipping event.\n"); return -1; } total_comm++; @@ -1262,10 +1262,10 @@ process_task_event(event_t *event, unsigned long offset, unsigned long head) thread = threads__findnew(event->fork.pid, &threads, &last_match); parent = threads__findnew(event->fork.ppid, &threads, &last_match); - dump_printf("%p [%p]: PERF_EVENT_%s: (%d:%d):(%d:%d)\n", + dump_printf("%p [%p]: PERF_RECORD_%s: (%d:%d):(%d:%d)\n", (void *)(offset + head), (void *)(long)(event->header.size), - event->header.type == PERF_EVENT_FORK ? "FORK" : "EXIT", + event->header.type == PERF_RECORD_FORK ? "FORK" : "EXIT", event->fork.pid, event->fork.tid, event->fork.ppid, event->fork.ptid); @@ -1276,11 +1276,11 @@ process_task_event(event_t *event, unsigned long offset, unsigned long head) if (thread == parent) return 0; - if (event->header.type == PERF_EVENT_EXIT) + if (event->header.type == PERF_RECORD_EXIT) return 0; if (!thread || !parent || thread__fork(thread, parent)) { - dump_printf("problem processing PERF_EVENT_FORK, skipping event.\n"); + dump_printf("problem processing PERF_RECORD_FORK, skipping event.\n"); return -1; } total_fork++; @@ -1291,7 +1291,7 @@ process_task_event(event_t *event, unsigned long offset, unsigned long head) static int process_lost_event(event_t *event, unsigned long offset, unsigned long head) { - dump_printf("%p [%p]: PERF_EVENT_LOST: id:%Ld: lost:%Ld\n", + dump_printf("%p [%p]: PERF_RECORD_LOST: id:%Ld: lost:%Ld\n", (void *)(offset + head), (void *)(long)(event->header.size), event->lost.id, @@ -1305,7 +1305,7 @@ process_lost_event(event_t *event, unsigned long offset, unsigned long head) static int process_read_event(event_t *event, unsigned long offset, unsigned long head) { - struct perf_counter_attr *attr; + struct perf_event_attr *attr; attr = perf_header__find_attr(event->read.id, header); @@ -1319,7 +1319,7 @@ process_read_event(event_t *event, unsigned long offset, unsigned long head) event->read.value); } - dump_printf("%p [%p]: PERF_EVENT_READ: %d %d %s %Lu\n", + dump_printf("%p [%p]: PERF_RECORD_READ: %d %d %s %Lu\n", (void *)(offset + head), (void *)(long)(event->header.size), event->read.pid, @@ -1337,31 +1337,31 @@ process_event(event_t *event, unsigned long offset, unsigned long head) trace_event(event); switch (event->header.type) { - case PERF_EVENT_SAMPLE: + case PERF_RECORD_SAMPLE: return process_sample_event(event, offset, head); - case PERF_EVENT_MMAP: + case PERF_RECORD_MMAP: return process_mmap_event(event, offset, head); - case PERF_EVENT_COMM: + case PERF_RECORD_COMM: return process_comm_event(event, offset, head); - case PERF_EVENT_FORK: - case PERF_EVENT_EXIT: + case PERF_RECORD_FORK: + case PERF_RECORD_EXIT: return process_task_event(event, offset, head); - case PERF_EVENT_LOST: + case PERF_RECORD_LOST: return process_lost_event(event, offset, head); - case PERF_EVENT_READ: + case PERF_RECORD_READ: return process_read_event(event, offset, head); /* * We dont process them right now but they are fine: */ - case PERF_EVENT_THROTTLE: - case PERF_EVENT_UNTHROTTLE: + case PERF_RECORD_THROTTLE: + case PERF_RECORD_UNTHROTTLE: return 0; default: diff --git a/tools/perf/builtin-sched.c b/tools/perf/builtin-sched.c index 275d79c..ea9c15c 100644 --- a/tools/perf/builtin-sched.c +++ b/tools/perf/builtin-sched.c @@ -1573,7 +1573,7 @@ process_sample_event(event_t *event, unsigned long offset, unsigned long head) more_data += sizeof(u64); } - dump_printf("%p [%p]: PERF_EVENT_SAMPLE (IP, %d): %d/%d: %p period: %Ld\n", + dump_printf("%p [%p]: PERF_RECORD_SAMPLE (IP, %d): %d/%d: %p period: %Ld\n", (void *)(offset + head), (void *)(long)(event->header.size), event->header.misc, @@ -1589,9 +1589,9 @@ process_sample_event(event_t *event, unsigned long offset, unsigned long head) return -1; } - cpumode = event->header.misc & PERF_EVENT_MISC_CPUMODE_MASK; + cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK; - if (cpumode == PERF_EVENT_MISC_KERNEL) { + if (cpumode == PERF_RECORD_MISC_KERNEL) { show = SHOW_KERNEL; level = 'k'; @@ -1599,7 +1599,7 @@ process_sample_event(event_t *event, unsigned long offset, unsigned long head) dump_printf(" ...... dso: %s\n", dso->name); - } else if (cpumode == PERF_EVENT_MISC_USER) { + } else if (cpumode == PERF_RECORD_MISC_USER) { show = SHOW_USER; level = '.'; @@ -1626,23 +1626,23 @@ process_event(event_t *event, unsigned long offset, unsigned long head) nr_events++; switch (event->header.type) { - case PERF_EVENT_MMAP: + case PERF_RECORD_MMAP: return 0; - case PERF_EVENT_LOST: + case PERF_RECORD_LOST: nr_lost_chunks++; nr_lost_events += event->lost.lost; return 0; - case PERF_EVENT_COMM: + case PERF_RECORD_COMM: return process_comm_event(event, offset, head); - case PERF_EVENT_EXIT ... PERF_EVENT_READ: + case PERF_RECORD_EXIT ... PERF_RECORD_READ: return 0; - case PERF_EVENT_SAMPLE: + case PERF_RECORD_SAMPLE: return process_sample_event(event, offset, head); - case PERF_EVENT_MAX: + case PERF_RECORD_MAX: default: return -1; } diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c index 61b8282..16af2d8 100644 --- a/tools/perf/builtin-stat.c +++ b/tools/perf/builtin-stat.c @@ -48,7 +48,7 @@ #include #include -static struct perf_counter_attr default_attrs[] = { +static struct perf_event_attr default_attrs[] = { { .type = PERF_TYPE_SOFTWARE, .config = PERF_COUNT_SW_TASK_CLOCK }, { .type = PERF_TYPE_SOFTWARE, .config = PERF_COUNT_SW_CONTEXT_SWITCHES}, @@ -130,11 +130,11 @@ struct stats runtime_cycles_stats; attrs[counter].config == PERF_COUNT_##c) #define ERR_PERF_OPEN \ -"Error: counter %d, sys_perf_counter_open() syscall returned with %d (%s)\n" +"Error: counter %d, sys_perf_event_open() syscall returned with %d (%s)\n" static void create_perf_stat_counter(int counter, int pid) { - struct perf_counter_attr *attr = attrs + counter; + struct perf_event_attr *attr = attrs + counter; if (scale) attr->read_format = PERF_FORMAT_TOTAL_TIME_ENABLED | @@ -144,7 +144,7 @@ static void create_perf_stat_counter(int counter, int pid) unsigned int cpu; for (cpu = 0; cpu < nr_cpus; cpu++) { - fd[cpu][counter] = sys_perf_counter_open(attr, -1, cpu, -1, 0); + fd[cpu][counter] = sys_perf_event_open(attr, -1, cpu, -1, 0); if (fd[cpu][counter] < 0 && verbose) fprintf(stderr, ERR_PERF_OPEN, counter, fd[cpu][counter], strerror(errno)); @@ -154,7 +154,7 @@ static void create_perf_stat_counter(int counter, int pid) attr->disabled = 1; attr->enable_on_exec = 1; - fd[0][counter] = sys_perf_counter_open(attr, pid, -1, -1, 0); + fd[0][counter] = sys_perf_event_open(attr, pid, -1, -1, 0); if (fd[0][counter] < 0 && verbose) fprintf(stderr, ERR_PERF_OPEN, counter, fd[0][counter], strerror(errno)); diff --git a/tools/perf/builtin-timechart.c b/tools/perf/builtin-timechart.c index 6004063..4405681 100644 --- a/tools/perf/builtin-timechart.c +++ b/tools/perf/builtin-timechart.c @@ -937,21 +937,21 @@ process_event(event_t *event) switch (event->header.type) { - case PERF_EVENT_COMM: + case PERF_RECORD_COMM: return process_comm_event(event); - case PERF_EVENT_FORK: + case PERF_RECORD_FORK: return process_fork_event(event); - case PERF_EVENT_EXIT: + case PERF_RECORD_EXIT: return process_exit_event(event); - case PERF_EVENT_SAMPLE: + case PERF_RECORD_SAMPLE: return queue_sample_event(event); /* * We dont process them right now but they are fine: */ - case PERF_EVENT_MMAP: - case PERF_EVENT_THROTTLE: - case PERF_EVENT_UNTHROTTLE: + case PERF_RECORD_MMAP: + case PERF_RECORD_THROTTLE: + case PERF_RECORD_UNTHROTTLE: return 0; default: diff --git a/tools/perf/builtin-top.c b/tools/perf/builtin-top.c index 4002ccb..1ca8889 100644 --- a/tools/perf/builtin-top.c +++ b/tools/perf/builtin-top.c @@ -901,7 +901,7 @@ struct mmap_data { static unsigned int mmap_read_head(struct mmap_data *md) { - struct perf_counter_mmap_page *pc = md->base; + struct perf_event_mmap_page *pc = md->base; int head; head = pc->data_head; @@ -977,9 +977,9 @@ static void mmap_read_counter(struct mmap_data *md) old += size; - if (event->header.type == PERF_EVENT_SAMPLE) { + if (event->header.type == PERF_RECORD_SAMPLE) { int user = - (event->header.misc & PERF_EVENT_MISC_CPUMODE_MASK) == PERF_EVENT_MISC_USER; + (event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK) == PERF_RECORD_MISC_USER; process_event(event->ip.ip, md->counter, user); } } @@ -1005,7 +1005,7 @@ int group_fd; static void start_counter(int i, int counter) { - struct perf_counter_attr *attr; + struct perf_event_attr *attr; int cpu; cpu = profile_cpu; @@ -1019,7 +1019,7 @@ static void start_counter(int i, int counter) attr->inherit = (cpu < 0) && inherit; try_again: - fd[i][counter] = sys_perf_counter_open(attr, target_pid, cpu, group_fd, 0); + fd[i][counter] = sys_perf_event_open(attr, target_pid, cpu, group_fd, 0); if (fd[i][counter] < 0) { int err = errno; @@ -1044,7 +1044,7 @@ try_again: printf("\n"); error("perfcounter syscall returned with %d (%s)\n", fd[i][counter], strerror(err)); - die("No CONFIG_PERF_COUNTERS=y kernel support configured?\n"); + die("No CONFIG_PERF_EVENTS=y kernel support configured?\n"); exit(-1); } assert(fd[i][counter] >= 0); diff --git a/tools/perf/builtin-trace.c b/tools/perf/builtin-trace.c index 914ab36..e9d256e 100644 --- a/tools/perf/builtin-trace.c +++ b/tools/perf/builtin-trace.c @@ -35,14 +35,14 @@ process_comm_event(event_t *event, unsigned long offset, unsigned long head) thread = threads__findnew(event->comm.pid, &threads, &last_match); - dump_printf("%p [%p]: PERF_EVENT_COMM: %s:%d\n", + dump_printf("%p [%p]: PERF_RECORD_COMM: %s:%d\n", (void *)(offset + head), (void *)(long)(event->header.size), event->comm.comm, event->comm.pid); if (thread == NULL || thread__set_comm(thread, event->comm.comm)) { - dump_printf("problem processing PERF_EVENT_COMM, skipping event.\n"); + dump_printf("problem processing PERF_RECORD_COMM, skipping event.\n"); return -1; } total_comm++; @@ -82,7 +82,7 @@ process_sample_event(event_t *event, unsigned long offset, unsigned long head) more_data += sizeof(u64); } - dump_printf("%p [%p]: PERF_EVENT_SAMPLE (IP, %d): %d/%d: %p period: %Ld\n", + dump_printf("%p [%p]: PERF_RECORD_SAMPLE (IP, %d): %d/%d: %p period: %Ld\n", (void *)(offset + head), (void *)(long)(event->header.size), event->header.misc, @@ -98,9 +98,9 @@ process_sample_event(event_t *event, unsigned long offset, unsigned long head) return -1; } - cpumode = event->header.misc & PERF_EVENT_MISC_CPUMODE_MASK; + cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK; - if (cpumode == PERF_EVENT_MISC_KERNEL) { + if (cpumode == PERF_RECORD_MISC_KERNEL) { show = SHOW_KERNEL; level = 'k'; @@ -108,7 +108,7 @@ process_sample_event(event_t *event, unsigned long offset, unsigned long head) dump_printf(" ...... dso: %s\n", dso->name); - } else if (cpumode == PERF_EVENT_MISC_USER) { + } else if (cpumode == PERF_RECORD_MISC_USER) { show = SHOW_USER; level = '.'; @@ -146,19 +146,19 @@ process_event(event_t *event, unsigned long offset, unsigned long head) trace_event(event); switch (event->header.type) { - case PERF_EVENT_MMAP ... PERF_EVENT_LOST: + case PERF_RECORD_MMAP ... PERF_RECORD_LOST: return 0; - case PERF_EVENT_COMM: + case PERF_RECORD_COMM: return process_comm_event(event, offset, head); - case PERF_EVENT_EXIT ... PERF_EVENT_READ: + case PERF_RECORD_EXIT ... PERF_RECORD_READ: return 0; - case PERF_EVENT_SAMPLE: + case PERF_RECORD_SAMPLE: return process_sample_event(event, offset, head); - case PERF_EVENT_MAX: + case PERF_RECORD_MAX: default: return -1; } diff --git a/tools/perf/design.txt b/tools/perf/design.txt index f71e0d2..f1946d1 100644 --- a/tools/perf/design.txt +++ b/tools/perf/design.txt @@ -18,10 +18,10 @@ underlying hardware counters. Performance counters are accessed via special file descriptors. There's one file descriptor per virtual counter used. -The special file descriptor is opened via the perf_counter_open() +The special file descriptor is opened via the perf_event_open() system call: - int sys_perf_counter_open(struct perf_counter_hw_event *hw_event_uptr, + int sys_perf_event_open(struct perf_event_hw_event *hw_event_uptr, pid_t pid, int cpu, int group_fd, unsigned long flags); @@ -32,9 +32,9 @@ can be used to set the blocking mode, etc. Multiple counters can be kept open at a time, and the counters can be poll()ed. -When creating a new counter fd, 'perf_counter_hw_event' is: +When creating a new counter fd, 'perf_event_hw_event' is: -struct perf_counter_hw_event { +struct perf_event_hw_event { /* * The MSB of the config word signifies if the rest contains cpu * specific (raw) counter configuration data, if unset, the next @@ -93,7 +93,7 @@ specified by 'event_id': /* * Generalized performance counter event types, used by the hw_event.event_id - * parameter of the sys_perf_counter_open() syscall: + * parameter of the sys_perf_event_open() syscall: */ enum hw_event_ids { /* @@ -159,7 +159,7 @@ in size. * reads on the counter should return the indicated quantities, * in increasing order of bit value, after the counter value. */ -enum perf_counter_read_format { +enum perf_event_read_format { PERF_FORMAT_TOTAL_TIME_ENABLED = 1, PERF_FORMAT_TOTAL_TIME_RUNNING = 2, }; @@ -178,7 +178,7 @@ interrupt: * Bits that can be set in hw_event.record_type to request information * in the overflow packets. */ -enum perf_counter_record_format { +enum perf_event_record_format { PERF_RECORD_IP = 1U << 0, PERF_RECORD_TID = 1U << 1, PERF_RECORD_TIME = 1U << 2, @@ -228,7 +228,7 @@ these events are recorded in the ring-buffer (see below). The 'comm' bit allows tracking of process comm data on process creation. This too is recorded in the ring-buffer (see below). -The 'pid' parameter to the perf_counter_open() system call allows the +The 'pid' parameter to the perf_event_open() system call allows the counter to be specific to a task: pid == 0: if the pid parameter is zero, the counter is attached to the @@ -258,7 +258,7 @@ The 'flags' parameter is currently unused and must be zero. The 'group_fd' parameter allows counter "groups" to be set up. A counter group has one counter which is the group "leader". The leader -is created first, with group_fd = -1 in the perf_counter_open call +is created first, with group_fd = -1 in the perf_event_open call that creates it. The rest of the group members are created subsequently, with group_fd giving the fd of the group leader. (A single counter on its own is created with group_fd = -1 and is @@ -277,13 +277,13 @@ tracking are logged into a ring-buffer. This ring-buffer is created and accessed through mmap(). The mmap size should be 1+2^n pages, where the first page is a meta-data page -(struct perf_counter_mmap_page) that contains various bits of information such +(struct perf_event_mmap_page) that contains various bits of information such as where the ring-buffer head is. /* * Structure of the page that can be mapped via mmap */ -struct perf_counter_mmap_page { +struct perf_event_mmap_page { __u32 version; /* version number of this structure */ __u32 compat_version; /* lowest version this is compat with */ @@ -317,7 +317,7 @@ struct perf_counter_mmap_page { * Control data for the mmap() data buffer. * * User-space reading this value should issue an rmb(), on SMP capable - * platforms, after reading this value -- see perf_counter_wakeup(). + * platforms, after reading this value -- see perf_event_wakeup(). */ __u32 data_head; /* head in the data section */ }; @@ -327,9 +327,9 @@ NOTE: the hw-counter userspace bits are arch specific and are currently only The following 2^n pages are the ring-buffer which contains events of the form: -#define PERF_EVENT_MISC_KERNEL (1 << 0) -#define PERF_EVENT_MISC_USER (1 << 1) -#define PERF_EVENT_MISC_OVERFLOW (1 << 2) +#define PERF_RECORD_MISC_KERNEL (1 << 0) +#define PERF_RECORD_MISC_USER (1 << 1) +#define PERF_RECORD_MISC_OVERFLOW (1 << 2) struct perf_event_header { __u32 type; @@ -353,8 +353,8 @@ enum perf_event_type { * char filename[]; * }; */ - PERF_EVENT_MMAP = 1, - PERF_EVENT_MUNMAP = 2, + PERF_RECORD_MMAP = 1, + PERF_RECORD_MUNMAP = 2, /* * struct { @@ -364,10 +364,10 @@ enum perf_event_type { * char comm[]; * }; */ - PERF_EVENT_COMM = 3, + PERF_RECORD_COMM = 3, /* - * When header.misc & PERF_EVENT_MISC_OVERFLOW the event_type field + * When header.misc & PERF_RECORD_MISC_OVERFLOW the event_type field * will be PERF_RECORD_* * * struct { @@ -397,7 +397,7 @@ Notification of new events is possible through poll()/select()/epoll() and fcntl() managing signals. Normally a notification is generated for every page filled, however one can -additionally set perf_counter_hw_event.wakeup_events to generate one every +additionally set perf_event_hw_event.wakeup_events to generate one every so many counter overflow events. Future work will include a splice() interface to the ring-buffer. @@ -409,11 +409,11 @@ events but does continue to exist and maintain its count value. An individual counter or counter group can be enabled with - ioctl(fd, PERF_COUNTER_IOC_ENABLE); + ioctl(fd, PERF_EVENT_IOC_ENABLE); or disabled with - ioctl(fd, PERF_COUNTER_IOC_DISABLE); + ioctl(fd, PERF_EVENT_IOC_DISABLE); Enabling or disabling the leader of a group enables or disables the whole group; that is, while the group leader is disabled, none of the @@ -424,16 +424,16 @@ other counter. Additionally, non-inherited overflow counters can use - ioctl(fd, PERF_COUNTER_IOC_REFRESH, nr); + ioctl(fd, PERF_EVENT_IOC_REFRESH, nr); to enable a counter for 'nr' events, after which it gets disabled again. A process can enable or disable all the counter groups that are attached to it, using prctl: - prctl(PR_TASK_PERF_COUNTERS_ENABLE); + prctl(PR_TASK_PERF_EVENTS_ENABLE); - prctl(PR_TASK_PERF_COUNTERS_DISABLE); + prctl(PR_TASK_PERF_EVENTS_DISABLE); This applies to all counters on the current process, whether created by this process or by another, and doesn't affect any counters that @@ -447,11 +447,11 @@ Arch requirements If your architecture does not have hardware performance metrics, you can still use the generic software counters based on hrtimers for sampling. -So to start with, in order to add HAVE_PERF_COUNTERS to your Kconfig, you +So to start with, in order to add HAVE_PERF_EVENTS to your Kconfig, you will need at least this: - - asm/perf_counter.h - a basic stub will suffice at first + - asm/perf_event.h - a basic stub will suffice at first - support for atomic64 types (and associated helper functions) - - set_perf_counter_pending() implemented + - set_perf_event_pending() implemented If your architecture does have hardware capabilities, you can override the -weak stub hw_perf_counter_init() to register hardware counters. +weak stub hw_perf_event_init() to register hardware counters. diff --git a/tools/perf/perf.h b/tools/perf/perf.h index 2abeb20..8cc4623 100644 --- a/tools/perf/perf.h +++ b/tools/perf/perf.h @@ -52,15 +52,15 @@ #include #include -#include "../../include/linux/perf_counter.h" +#include "../../include/linux/perf_event.h" #include "util/types.h" /* - * prctl(PR_TASK_PERF_COUNTERS_DISABLE) will (cheaply) disable all + * prctl(PR_TASK_PERF_EVENTS_DISABLE) will (cheaply) disable all * counters in the current task. */ -#define PR_TASK_PERF_COUNTERS_DISABLE 31 -#define PR_TASK_PERF_COUNTERS_ENABLE 32 +#define PR_TASK_PERF_EVENTS_DISABLE 31 +#define PR_TASK_PERF_EVENTS_ENABLE 32 #ifndef NSEC_PER_SEC # define NSEC_PER_SEC 1000000000ULL @@ -90,12 +90,12 @@ static inline unsigned long long rdclock(void) _min1 < _min2 ? _min1 : _min2; }) static inline int -sys_perf_counter_open(struct perf_counter_attr *attr, +sys_perf_event_open(struct perf_event_attr *attr, pid_t pid, int cpu, int group_fd, unsigned long flags) { attr->size = sizeof(*attr); - return syscall(__NR_perf_counter_open, attr, pid, cpu, + return syscall(__NR_perf_event_open, attr, pid, cpu, group_fd, flags); } diff --git a/tools/perf/util/event.h b/tools/perf/util/event.h index 018d414..2c9c26d6 100644 --- a/tools/perf/util/event.h +++ b/tools/perf/util/event.h @@ -1,5 +1,5 @@ -#ifndef __PERF_EVENT_H -#define __PERF_EVENT_H +#ifndef __PERF_RECORD_H +#define __PERF_RECORD_H #include "../perf.h" #include "util.h" #include diff --git a/tools/perf/util/header.c b/tools/perf/util/header.c index bb4fca3..e306857 100644 --- a/tools/perf/util/header.c +++ b/tools/perf/util/header.c @@ -9,7 +9,7 @@ /* * Create new perf.data header attribute: */ -struct perf_header_attr *perf_header_attr__new(struct perf_counter_attr *attr) +struct perf_header_attr *perf_header_attr__new(struct perf_event_attr *attr) { struct perf_header_attr *self = malloc(sizeof(*self)); @@ -134,7 +134,7 @@ struct perf_file_section { }; struct perf_file_attr { - struct perf_counter_attr attr; + struct perf_event_attr attr; struct perf_file_section ids; }; @@ -320,7 +320,7 @@ u64 perf_header__sample_type(struct perf_header *header) return type; } -struct perf_counter_attr * +struct perf_event_attr * perf_header__find_attr(u64 id, struct perf_header *header) { int i; diff --git a/tools/perf/util/header.h b/tools/perf/util/header.h index 7b0e84a..a0761bc 100644 --- a/tools/perf/util/header.h +++ b/tools/perf/util/header.h @@ -1,12 +1,12 @@ #ifndef _PERF_HEADER_H #define _PERF_HEADER_H -#include "../../../include/linux/perf_counter.h" +#include "../../../include/linux/perf_event.h" #include #include "types.h" struct perf_header_attr { - struct perf_counter_attr attr; + struct perf_event_attr attr; int ids, size; u64 *id; off_t id_offset; @@ -34,11 +34,11 @@ char *perf_header__find_event(u64 id); struct perf_header_attr * -perf_header_attr__new(struct perf_counter_attr *attr); +perf_header_attr__new(struct perf_event_attr *attr); void perf_header_attr__add_id(struct perf_header_attr *self, u64 id); u64 perf_header__sample_type(struct perf_header *header); -struct perf_counter_attr * +struct perf_event_attr * perf_header__find_attr(u64 id, struct perf_header *header); diff --git a/tools/perf/util/parse-events.c b/tools/perf/util/parse-events.c index 89172fd..13ab4b8 100644 --- a/tools/perf/util/parse-events.c +++ b/tools/perf/util/parse-events.c @@ -10,7 +10,7 @@ int nr_counters; -struct perf_counter_attr attrs[MAX_COUNTERS]; +struct perf_event_attr attrs[MAX_COUNTERS]; struct event_symbol { u8 type; @@ -48,13 +48,13 @@ static struct event_symbol event_symbols[] = { { CSW(CPU_MIGRATIONS), "cpu-migrations", "migrations" }, }; -#define __PERF_COUNTER_FIELD(config, name) \ - ((config & PERF_COUNTER_##name##_MASK) >> PERF_COUNTER_##name##_SHIFT) +#define __PERF_EVENT_FIELD(config, name) \ + ((config & PERF_EVENT_##name##_MASK) >> PERF_EVENT_##name##_SHIFT) -#define PERF_COUNTER_RAW(config) __PERF_COUNTER_FIELD(config, RAW) -#define PERF_COUNTER_CONFIG(config) __PERF_COUNTER_FIELD(config, CONFIG) -#define PERF_COUNTER_TYPE(config) __PERF_COUNTER_FIELD(config, TYPE) -#define PERF_COUNTER_ID(config) __PERF_COUNTER_FIELD(config, EVENT) +#define PERF_EVENT_RAW(config) __PERF_EVENT_FIELD(config, RAW) +#define PERF_EVENT_CONFIG(config) __PERF_EVENT_FIELD(config, CONFIG) +#define PERF_EVENT_TYPE(config) __PERF_EVENT_FIELD(config, TYPE) +#define PERF_EVENT_ID(config) __PERF_EVENT_FIELD(config, EVENT) static const char *hw_event_names[] = { "cycles", @@ -352,7 +352,7 @@ static int parse_aliases(const char **str, const char *names[][MAX_ALIASES], int } static enum event_result -parse_generic_hw_event(const char **str, struct perf_counter_attr *attr) +parse_generic_hw_event(const char **str, struct perf_event_attr *attr) { const char *s = *str; int cache_type = -1, cache_op = -1, cache_result = -1; @@ -417,7 +417,7 @@ parse_single_tracepoint_event(char *sys_name, const char *evt_name, unsigned int evt_length, char *flags, - struct perf_counter_attr *attr, + struct perf_event_attr *attr, const char **strp) { char evt_path[MAXPATHLEN]; @@ -505,7 +505,7 @@ parse_subsystem_tracepoint_event(char *sys_name, char *flags) static enum event_result parse_tracepoint_event(const char **strp, - struct perf_counter_attr *attr) + struct perf_event_attr *attr) { const char *evt_name; char *flags; @@ -563,7 +563,7 @@ static int check_events(const char *str, unsigned int i) } static enum event_result -parse_symbolic_event(const char **strp, struct perf_counter_attr *attr) +parse_symbolic_event(const char **strp, struct perf_event_attr *attr) { const char *str = *strp; unsigned int i; @@ -582,7 +582,7 @@ parse_symbolic_event(const char **strp, struct perf_counter_attr *attr) } static enum event_result -parse_raw_event(const char **strp, struct perf_counter_attr *attr) +parse_raw_event(const char **strp, struct perf_event_attr *attr) { const char *str = *strp; u64 config; @@ -601,7 +601,7 @@ parse_raw_event(const char **strp, struct perf_counter_attr *attr) } static enum event_result -parse_numeric_event(const char **strp, struct perf_counter_attr *attr) +parse_numeric_event(const char **strp, struct perf_event_attr *attr) { const char *str = *strp; char *endp; @@ -623,7 +623,7 @@ parse_numeric_event(const char **strp, struct perf_counter_attr *attr) } static enum event_result -parse_event_modifier(const char **strp, struct perf_counter_attr *attr) +parse_event_modifier(const char **strp, struct perf_event_attr *attr) { const char *str = *strp; int eu = 1, ek = 1, eh = 1; @@ -656,7 +656,7 @@ parse_event_modifier(const char **strp, struct perf_counter_attr *attr) * Symbolic names are (almost) exactly matched. */ static enum event_result -parse_event_symbols(const char **str, struct perf_counter_attr *attr) +parse_event_symbols(const char **str, struct perf_event_attr *attr) { enum event_result ret; @@ -711,7 +711,7 @@ static void store_event_type(const char *orgname) int parse_events(const struct option *opt __used, const char *str, int unset __used) { - struct perf_counter_attr attr; + struct perf_event_attr attr; enum event_result ret; if (strchr(str, ':')) diff --git a/tools/perf/util/parse-events.h b/tools/perf/util/parse-events.h index 60704c1..30c6081 100644 --- a/tools/perf/util/parse-events.h +++ b/tools/perf/util/parse-events.h @@ -16,7 +16,7 @@ extern struct tracepoint_path *tracepoint_id_to_path(u64 config); extern int nr_counters; -extern struct perf_counter_attr attrs[MAX_COUNTERS]; +extern struct perf_event_attr attrs[MAX_COUNTERS]; extern const char *event_name(int ctr); extern const char *__event_name(int type, u64 config); diff --git a/tools/perf/util/trace-event-info.c b/tools/perf/util/trace-event-info.c index 1fd824c..af4b057 100644 --- a/tools/perf/util/trace-event-info.c +++ b/tools/perf/util/trace-event-info.c @@ -480,12 +480,12 @@ out: } static struct tracepoint_path * -get_tracepoints_path(struct perf_counter_attr *pattrs, int nb_counters) +get_tracepoints_path(struct perf_event_attr *pattrs, int nb_events) { struct tracepoint_path path, *ppath = &path; int i; - for (i = 0; i < nb_counters; i++) { + for (i = 0; i < nb_events; i++) { if (pattrs[i].type != PERF_TYPE_TRACEPOINT) continue; ppath->next = tracepoint_id_to_path(pattrs[i].config); @@ -496,7 +496,7 @@ get_tracepoints_path(struct perf_counter_attr *pattrs, int nb_counters) return path.next; } -void read_tracing_data(struct perf_counter_attr *pattrs, int nb_counters) +void read_tracing_data(struct perf_event_attr *pattrs, int nb_events) { char buf[BUFSIZ]; struct tracepoint_path *tps; @@ -530,7 +530,7 @@ void read_tracing_data(struct perf_counter_attr *pattrs, int nb_counters) page_size = getpagesize(); write_or_die(&page_size, 4); - tps = get_tracepoints_path(pattrs, nb_counters); + tps = get_tracepoints_path(pattrs, nb_events); read_header_files(); read_ftrace_files(tps); diff --git a/tools/perf/util/trace-event.h b/tools/perf/util/trace-event.h index d35ebf1..693f815 100644 --- a/tools/perf/util/trace-event.h +++ b/tools/perf/util/trace-event.h @@ -240,6 +240,6 @@ unsigned long long raw_field_value(struct event *event, const char *name, void *data); void *raw_field_ptr(struct event *event, const char *name, void *data); -void read_tracing_data(struct perf_counter_attr *pattrs, int nb_counters); +void read_tracing_data(struct perf_event_attr *pattrs, int nb_events); #endif /* _TRACE_EVENTS_H */ -- cgit v0.10.2