From f06c38103ea9dbca27c3f4d77f444ddefb5477cd Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Mon, 12 May 2008 21:20:47 +0200 Subject: ftrace: add sysprof plugin very first baby version. Signed-off-by: Ingo Molnar Signed-off-by: Thomas Gleixner diff --git a/kernel/trace/Kconfig b/kernel/trace/Kconfig index 5c2295b..e101c9a 100644 --- a/kernel/trace/Kconfig +++ b/kernel/trace/Kconfig @@ -75,6 +75,14 @@ config PREEMPT_TRACER enabled. This option and the irqs-off timing option can be used together or separately.) +config SYSPROF_TRACER + bool "Sysprof Tracer" + depends on DEBUG_KERNEL + select TRACING + help + This tracer provides the trace needed by the 'Sysprof' userspace + tool. + config SCHED_TRACER bool "Scheduling Latency Tracer" depends on HAVE_FTRACE diff --git a/kernel/trace/Makefile b/kernel/trace/Makefile index d9efbbf..7aec123 100644 --- a/kernel/trace/Makefile +++ b/kernel/trace/Makefile @@ -14,6 +14,7 @@ obj-$(CONFIG_FTRACE) += libftrace.o obj-$(CONFIG_TRACING) += trace.o obj-$(CONFIG_CONTEXT_SWITCH_TRACER) += trace_sched_switch.o +obj-$(CONFIG_SYSPROF_TRACER) += trace_sysprof.o obj-$(CONFIG_FTRACE) += trace_functions.o obj-$(CONFIG_IRQSOFF_TRACER) += trace_irqsoff.o obj-$(CONFIG_PREEMPT_TRACER) += trace_irqsoff.o diff --git a/kernel/trace/trace_sysprof.c b/kernel/trace/trace_sysprof.c new file mode 100644 index 0000000..6c139bc --- /dev/null +++ b/kernel/trace/trace_sysprof.c @@ -0,0 +1,80 @@ +/* + * trace stack traces + * + * Copyright (C) 2007 Steven Rostedt + * Copyright (C) 2008 Ingo Molnar + * + */ +#include +#include +#include +#include +#include +#include +#include + +#include "trace.h" + +static struct trace_array *ctx_trace; +static int __read_mostly tracer_enabled; + +static notrace void stack_reset(struct trace_array *tr) +{ + int cpu; + + tr->time_start = ftrace_now(tr->cpu); + + for_each_online_cpu(cpu) + tracing_reset(tr->data[cpu]); +} + +static notrace void start_stack_trace(struct trace_array *tr) +{ + stack_reset(tr); + tracer_enabled = 1; +} + +static notrace void stop_stack_trace(struct trace_array *tr) +{ + tracer_enabled = 0; +} + +static notrace void stack_trace_init(struct trace_array *tr) +{ + ctx_trace = tr; + + if (tr->ctrl) + start_stack_trace(tr); +} + +static notrace void stack_trace_reset(struct trace_array *tr) +{ + if (tr->ctrl) + stop_stack_trace(tr); +} + +static void stack_trace_ctrl_update(struct trace_array *tr) +{ + /* When starting a new trace, reset the buffers */ + if (tr->ctrl) + start_stack_trace(tr); + else + stop_stack_trace(tr); +} + +static struct tracer stack_trace __read_mostly = +{ + .name = "sysprof", + .init = stack_trace_init, + .reset = stack_trace_reset, + .ctrl_update = stack_trace_ctrl_update, +#ifdef CONFIG_FTRACE_SELFTEST + .selftest = trace_selftest_startup_stack, +#endif +}; + +__init static int init_stack_trace(void) +{ + return register_tracer(&stack_trace); +} +device_initcall(init_stack_trace); -- cgit v0.10.2 From 0075fa80305f3231a2d5df97b00d7f55a48ea27e Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Mon, 12 May 2008 21:20:47 +0200 Subject: ftrace: extend sysprof plugin add per CPU hrtimers. Signed-off-by: Ingo Molnar Signed-off-by: Thomas Gleixner diff --git a/kernel/trace/trace_sysprof.c b/kernel/trace/trace_sysprof.c index 6c139bc..ba55b87 100644 --- a/kernel/trace/trace_sysprof.c +++ b/kernel/trace/trace_sysprof.c @@ -5,19 +5,76 @@ * Copyright (C) 2008 Ingo Molnar * */ -#include -#include -#include #include +#include +#include #include -#include #include +#include +#include #include "trace.h" static struct trace_array *ctx_trace; static int __read_mostly tracer_enabled; +static const unsigned long sample_period = 1000000; + +/* + * Per CPU hrtimers that do the profiling: + */ +static DEFINE_PER_CPU(struct hrtimer, stack_trace_hrtimer); + +static enum hrtimer_restart stack_trace_timer_fn(struct hrtimer *hrtimer) +{ + /* trace here */ + panic_timeout++; + + hrtimer_forward_now(hrtimer, ns_to_ktime(sample_period)); + + return HRTIMER_RESTART; +} + +static void start_stack_timer(int cpu) +{ + struct hrtimer *hrtimer = &per_cpu(stack_trace_hrtimer, cpu); + + hrtimer_init(hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); + hrtimer->function = stack_trace_timer_fn; + hrtimer->cb_mode = HRTIMER_CB_IRQSAFE_NO_SOFTIRQ; + + hrtimer_start(hrtimer, ns_to_ktime(sample_period), HRTIMER_MODE_REL); +} + +static void start_stack_timers(void) +{ + cpumask_t saved_mask = current->cpus_allowed; + int cpu; + + for_each_online_cpu(cpu) { + set_cpus_allowed_ptr(current, &cpumask_of_cpu(cpu)); + start_stack_timer(cpu); + printk("started timer on cpu%d\n", cpu); + } + set_cpus_allowed_ptr(current, &saved_mask); +} + +static void stop_stack_timer(int cpu) +{ + struct hrtimer *hrtimer = &per_cpu(stack_trace_hrtimer, cpu); + + hrtimer_cancel(hrtimer); + printk("cancelled timer on cpu%d\n", cpu); +} + +static void stop_stack_timers(void) +{ + int cpu; + + for_each_online_cpu(cpu) + stop_stack_timer(cpu); +} + static notrace void stack_reset(struct trace_array *tr) { int cpu; @@ -31,11 +88,13 @@ static notrace void stack_reset(struct trace_array *tr) static notrace void start_stack_trace(struct trace_array *tr) { stack_reset(tr); + start_stack_timers(); tracer_enabled = 1; } static notrace void stop_stack_trace(struct trace_array *tr) { + stop_stack_timers(); tracer_enabled = 0; } -- cgit v0.10.2 From 56a08bdcff20f0022bd9160c1093e56f763499aa Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Mon, 12 May 2008 21:20:47 +0200 Subject: ftrace: extend sysprof plugin some more Signed-off-by: Ingo Molnar Signed-off-by: Thomas Gleixner diff --git a/kernel/trace/trace_sysprof.c b/kernel/trace/trace_sysprof.c index ba55b87..b1137c1 100644 --- a/kernel/trace/trace_sysprof.c +++ b/kernel/trace/trace_sysprof.c @@ -3,7 +3,7 @@ * * Copyright (C) 2007 Steven Rostedt * Copyright (C) 2008 Ingo Molnar - * + * Copyright (C) 2004, 2005, Soeren Sandmann */ #include #include @@ -11,13 +11,17 @@ #include #include #include +#include #include #include "trace.h" -static struct trace_array *ctx_trace; +static struct trace_array *sysprof_trace; static int __read_mostly tracer_enabled; +/* + * 10 msecs for now: + */ static const unsigned long sample_period = 1000000; /* @@ -25,10 +29,78 @@ static const unsigned long sample_period = 1000000; */ static DEFINE_PER_CPU(struct hrtimer, stack_trace_hrtimer); +struct stack_frame { + const void __user *next_fp; + unsigned long return_address; +}; + +static int copy_stack_frame(const void __user *fp, struct stack_frame *frame) +{ + if (!access_ok(VERIFY_READ, fp, sizeof(*frame))) + return 0; + + if (__copy_from_user_inatomic(frame, frame_pointer, sizeof(*frame))) + return 0; + + return 1; +} + +#define SYSPROF_MAX_ADDRESSES 512 + +static void timer_notify(struct pt_regs *regs, int cpu) +{ + const void __user *frame_pointer; + struct trace_array_cpu *data; + struct stack_frame frame; + struct trace_array *tr; + int is_user; + int i; + + if (!regs) + return; + + tr = sysprof_trace; + data = tr->data[cpu]; + is_user = user_mode(regs); + + if (!current || current->pid == 0) + return; + + if (is_user && current->state != TASK_RUNNING) + return; + + if (!is_user) { + /* kernel */ + ftrace(tr, data, current->pid, 1, 0); + return; + + } + + trace_special(tr, data, 0, current->pid, regs->ip); + + frame_pointer = (void __user *)regs->bp; + + for (i = 0; i < SYSPROF_MAX_ADDRESSES; i++) { + if (!copy_stack_frame(frame_pointer, &frame)) + break; + if ((unsigned long)frame_pointer < regs->sp) + break; + + trace_special(tr, data, 1, frame.return_address, + (unsigned long)frame_pointer); + frame_pointer = frame.next_fp; + } + + trace_special(tr, data, 2, current->pid, i); + + if (i == SYSPROF_MAX_ADDRESSES) + trace_special(tr, data, -1, -1, -1); +} + static enum hrtimer_restart stack_trace_timer_fn(struct hrtimer *hrtimer) { /* trace here */ - panic_timeout++; + timer_notify(get_irq_regs(), smp_processor_id()); hrtimer_forward_now(hrtimer, ns_to_ktime(sample_period)); @@ -100,7 +172,7 @@ static notrace void stop_stack_trace(struct trace_array *tr) static notrace void stack_trace_init(struct trace_array *tr) { - ctx_trace = tr; + sysprof_trace = tr; if (tr->ctrl) start_stack_trace(tr); -- cgit v0.10.2 From a6dd24f8d00cbccb560b19a723e6fb9bdfb20799 Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Mon, 12 May 2008 21:20:47 +0200 Subject: ftrace: sysprof-plugin, add self-tests Signed-off-by: Ingo Molnar Signed-off-by: Thomas Gleixner diff --git a/kernel/trace/trace.h b/kernel/trace/trace.h index c460e85..b2198bc 100644 --- a/kernel/trace/trace.h +++ b/kernel/trace/trace.h @@ -280,6 +280,10 @@ extern int trace_selftest_startup_wakeup(struct tracer *trace, extern int trace_selftest_startup_sched_switch(struct tracer *trace, struct trace_array *tr); #endif +#ifdef CONFIG_SYSPROF_TRACER +extern int trace_selftest_startup_sysprof(struct tracer *trace, + struct trace_array *tr); +#endif #endif /* CONFIG_FTRACE_STARTUP_TEST */ extern void *head_page(struct trace_array_cpu *data); diff --git a/kernel/trace/trace_selftest.c b/kernel/trace/trace_selftest.c index 3877dd9..033a6fb 100644 --- a/kernel/trace/trace_selftest.c +++ b/kernel/trace/trace_selftest.c @@ -537,3 +537,31 @@ trace_selftest_startup_sched_switch(struct tracer *trace, struct trace_array *tr return ret; } #endif /* CONFIG_CONTEXT_SWITCH_TRACER */ + +#ifdef CONFIG_SYSPROF_TRACER +int +trace_selftest_startup_sysprof(struct tracer *trace, struct trace_array *tr) +{ + unsigned long count; + int ret; + + /* start the tracing */ + tr->ctrl = 1; + trace->init(tr); + /* Sleep for a 1/10 of a second */ + msleep(100); + /* stop the tracing. */ + tr->ctrl = 0; + trace->ctrl_update(tr); + /* check the trace buffer */ + ret = trace_test_buffer(tr, &count); + trace->reset(tr); + + if (!ret && !count) { + printk(KERN_CONT ".. no entries found .."); + ret = -1; + } + + return ret; +} +#endif /* CONFIG_SYSPROF_TRACER */ diff --git a/kernel/trace/trace_sysprof.c b/kernel/trace/trace_sysprof.c index b1137c1..b78f12f 100644 --- a/kernel/trace/trace_sysprof.c +++ b/kernel/trace/trace_sysprof.c @@ -126,7 +126,7 @@ static void start_stack_timers(void) for_each_online_cpu(cpu) { set_cpus_allowed_ptr(current, &cpumask_of_cpu(cpu)); start_stack_timer(cpu); - printk("started timer on cpu%d\n", cpu); + printk(KERN_INFO "started sysprof timer on cpu%d\n", cpu); } set_cpus_allowed_ptr(current, &saved_mask); } @@ -136,7 +136,7 @@ static void stop_stack_timer(int cpu) struct hrtimer *hrtimer = &per_cpu(stack_trace_hrtimer, cpu); hrtimer_cancel(hrtimer); - printk("cancelled timer on cpu%d\n", cpu); + printk(KERN_INFO "cancelled sysprof timer on cpu%d\n", cpu); } static void stop_stack_timers(void) @@ -200,7 +200,7 @@ static struct tracer stack_trace __read_mostly = .reset = stack_trace_reset, .ctrl_update = stack_trace_ctrl_update, #ifdef CONFIG_FTRACE_SELFTEST - .selftest = trace_selftest_startup_stack, + .selftest = trace_selftest_startup_sysprof, #endif }; -- cgit v0.10.2 From 842af315e8b0adad58fc642eaa5e6f53525e0534 Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Mon, 12 May 2008 21:20:47 +0200 Subject: ftrace: sysprof plugin improvement add sample maximum depth. Signed-off-by: Ingo Molnar Signed-off-by: Thomas Gleixner diff --git a/kernel/trace/trace_sysprof.c b/kernel/trace/trace_sysprof.c index b78f12f..7f6fccc 100644 --- a/kernel/trace/trace_sysprof.c +++ b/kernel/trace/trace_sysprof.c @@ -23,6 +23,7 @@ static int __read_mostly tracer_enabled; * 10 msecs for now: */ static const unsigned long sample_period = 1000000; +static const unsigned int sample_max_depth = 512; /* * Per CPU hrtimers that do the profiling: @@ -45,8 +46,6 @@ static int copy_stack_frame(const void __user *fp, struct stack_frame *frame) return 1; } -#define SYSPROF_MAX_ADDRESSES 512 - static void timer_notify(struct pt_regs *regs, int cpu) { const void __user *frame_pointer; @@ -80,7 +79,7 @@ static void timer_notify(struct pt_regs *regs, int cpu) frame_pointer = (void __user *)regs->bp; - for (i = 0; i < SYSPROF_MAX_ADDRESSES; i++) { + for (i = 0; i < sample_max_depth; i++) { if (!copy_stack_frame(frame_pointer, &frame)) break; if ((unsigned long)frame_pointer < regs->sp) @@ -93,7 +92,7 @@ static void timer_notify(struct pt_regs *regs, int cpu) trace_special(tr, data, 2, current->pid, i); - if (i == SYSPROF_MAX_ADDRESSES) + if (i == sample_max_depth) trace_special(tr, data, -1, -1, -1); } @@ -126,7 +125,6 @@ static void start_stack_timers(void) for_each_online_cpu(cpu) { set_cpus_allowed_ptr(current, &cpumask_of_cpu(cpu)); start_stack_timer(cpu); - printk(KERN_INFO "started sysprof timer on cpu%d\n", cpu); } set_cpus_allowed_ptr(current, &saved_mask); } @@ -136,7 +134,6 @@ static void stop_stack_timer(int cpu) struct hrtimer *hrtimer = &per_cpu(stack_trace_hrtimer, cpu); hrtimer_cancel(hrtimer); - printk(KERN_INFO "cancelled sysprof timer on cpu%d\n", cpu); } static void stop_stack_timers(void) -- cgit v0.10.2 From ef4ab15ff34fd9c65e92bee70f58e7179da881c5 Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Mon, 12 May 2008 21:20:48 +0200 Subject: ftrace: make sysprof dependent on x86 for now that's the only tested platform for now. If there's interest we can make it generic easily. Signed-off-by: Ingo Molnar Signed-off-by: Thomas Gleixner diff --git a/kernel/trace/Kconfig b/kernel/trace/Kconfig index e101c9a..9b49526 100644 --- a/kernel/trace/Kconfig +++ b/kernel/trace/Kconfig @@ -77,7 +77,7 @@ config PREEMPT_TRACER config SYSPROF_TRACER bool "Sysprof Tracer" - depends on DEBUG_KERNEL + depends on DEBUG_KERNEL && X86 select TRACING help This tracer provides the trace needed by the 'Sysprof' userspace diff --git a/kernel/trace/trace_selftest.c b/kernel/trace/trace_selftest.c index 033a6fb..5588ecc 100644 --- a/kernel/trace/trace_selftest.c +++ b/kernel/trace/trace_selftest.c @@ -557,11 +557,6 @@ trace_selftest_startup_sysprof(struct tracer *trace, struct trace_array *tr) ret = trace_test_buffer(tr, &count); trace->reset(tr); - if (!ret && !count) { - printk(KERN_CONT ".. no entries found .."); - ret = -1; - } - return ret; } #endif /* CONFIG_SYSPROF_TRACER */ -- cgit v0.10.2 From 9f6b4e3f4a24f2590f1c96f117fc45fbea9b0fa4 Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Mon, 12 May 2008 21:20:48 +0200 Subject: ftrace: sysprof fix Signed-off-by: Ingo Molnar Signed-off-by: Thomas Gleixner diff --git a/kernel/trace/trace_sysprof.c b/kernel/trace/trace_sysprof.c index 7f6fccc..f9a09fe 100644 --- a/kernel/trace/trace_sysprof.c +++ b/kernel/trace/trace_sysprof.c @@ -37,21 +37,26 @@ struct stack_frame { static int copy_stack_frame(const void __user *fp, struct stack_frame *frame) { + int ret; + if (!access_ok(VERIFY_READ, fp, sizeof(*frame))) return 0; - if (__copy_from_user_inatomic(frame, frame_pointer, sizeof(*frame))) - return 0; + ret = 1; + pagefault_disable(); + if (__copy_from_user_inatomic(frame, fp, sizeof(*frame))) + ret = 0; + pagefault_enable(); - return 1; + return ret; } static void timer_notify(struct pt_regs *regs, int cpu) { - const void __user *frame_pointer; struct trace_array_cpu *data; struct stack_frame frame; struct trace_array *tr; + const void __user *fp; int is_user; int i; @@ -77,21 +82,26 @@ static void timer_notify(struct pt_regs *regs, int cpu) trace_special(tr, data, 0, current->pid, regs->ip); - frame_pointer = (void __user *)regs->bp; + fp = (void __user *)regs->bp; for (i = 0; i < sample_max_depth; i++) { - if (!copy_stack_frame(frame_pointer, &frame)) + frame.next_fp = 0; + frame.return_address = 0; + if (!copy_stack_frame(fp, &frame)) break; - if ((unsigned long)frame_pointer < regs->sp) + if ((unsigned long)fp < regs->sp) break; trace_special(tr, data, 1, frame.return_address, - (unsigned long)frame_pointer); - frame_pointer = frame.next_fp; + (unsigned long)fp); + fp = frame.next_fp; } trace_special(tr, data, 2, current->pid, i); + /* + * Special trace entry if we overflow the max depth: + */ if (i == sample_max_depth) trace_special(tr, data, -1, -1, -1); } -- cgit v0.10.2 From d618b3e6e50970a6248ac857653fdd49bcd3c045 Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Mon, 12 May 2008 21:20:49 +0200 Subject: ftrace: sysprof updates make the sample period configurable. Signed-off-by: Ingo Molnar Signed-off-by: Thomas Gleixner diff --git a/kernel/trace/trace.c b/kernel/trace/trace.c index 3271916..95b7c48 100644 --- a/kernel/trace/trace.c +++ b/kernel/trace/trace.c @@ -2800,6 +2800,9 @@ static __init void tracer_init_debugfs(void) pr_warning("Could not create debugfs " "'dyn_ftrace_total_info' entry\n"); #endif +#ifdef CONFIG_SYSPROF_TRACER + init_tracer_sysprof_debugfs(d_tracer); +#endif } static int trace_alloc_page(void) diff --git a/kernel/trace/trace.h b/kernel/trace/trace.h index b2198bc..b7f85d9 100644 --- a/kernel/trace/trace.h +++ b/kernel/trace/trace.h @@ -188,6 +188,8 @@ struct trace_iterator { void tracing_reset(struct trace_array_cpu *data); int tracing_open_generic(struct inode *inode, struct file *filp); struct dentry *tracing_init_dentry(void); +void init_tracer_sysprof_debugfs(struct dentry *d_tracer); + void ftrace(struct trace_array *tr, struct trace_array_cpu *data, unsigned long ip, diff --git a/kernel/trace/trace_sysprof.c b/kernel/trace/trace_sysprof.c index f9a09fe..1940623 100644 --- a/kernel/trace/trace_sysprof.c +++ b/kernel/trace/trace_sysprof.c @@ -20,11 +20,12 @@ static struct trace_array *sysprof_trace; static int __read_mostly tracer_enabled; /* - * 10 msecs for now: + * 1 msec sample interval by default: */ -static const unsigned long sample_period = 1000000; +static unsigned long sample_period = 1000000; static const unsigned int sample_max_depth = 512; +static DEFINE_MUTEX(sample_timer_lock); /* * Per CPU hrtimers that do the profiling: */ @@ -166,15 +167,19 @@ static notrace void stack_reset(struct trace_array *tr) static notrace void start_stack_trace(struct trace_array *tr) { + mutex_lock(&sample_timer_lock); stack_reset(tr); start_stack_timers(); tracer_enabled = 1; + mutex_unlock(&sample_timer_lock); } static notrace void stop_stack_trace(struct trace_array *tr) { + mutex_lock(&sample_timer_lock); stop_stack_timers(); tracer_enabled = 0; + mutex_unlock(&sample_timer_lock); } static notrace void stack_trace_init(struct trace_array *tr) @@ -216,3 +221,64 @@ __init static int init_stack_trace(void) return register_tracer(&stack_trace); } device_initcall(init_stack_trace); + +#define MAX_LONG_DIGITS 22 + +static ssize_t +sysprof_sample_read(struct file *filp, char __user *ubuf, + size_t cnt, loff_t *ppos) +{ + char buf[MAX_LONG_DIGITS]; + int r; + + r = sprintf(buf, "%ld\n", nsecs_to_usecs(sample_period)); + + return simple_read_from_buffer(ubuf, cnt, ppos, buf, r); +} + +static ssize_t +sysprof_sample_write(struct file *filp, const char __user *ubuf, + size_t cnt, loff_t *ppos) +{ + char buf[MAX_LONG_DIGITS]; + unsigned long val; + + if (cnt > MAX_LONG_DIGITS-1) + cnt = MAX_LONG_DIGITS-1; + + if (copy_from_user(&buf, ubuf, cnt)) + return -EFAULT; + + buf[cnt] = 0; + + val = simple_strtoul(buf, NULL, 10); + /* + * Enforce a minimum sample period of 100 usecs: + */ + if (val < 100) + val = 100; + + mutex_lock(&sample_timer_lock); + stop_stack_timers(); + sample_period = val * 1000; + start_stack_timers(); + mutex_unlock(&sample_timer_lock); + + return cnt; +} + +static struct file_operations sysprof_sample_fops = { + .read = sysprof_sample_read, + .write = sysprof_sample_write, +}; + +void init_tracer_sysprof_debugfs(struct dentry *d_tracer) +{ + struct dentry *entry; + + entry = debugfs_create_file("sysprof_sample_period", 0644, + d_tracer, NULL, &sysprof_sample_fops); + if (entry) + return; + pr_warning("Could not create debugfs 'dyn_ftrace_total_info' entry\n"); +} -- cgit v0.10.2 From ada6b835067dc022f11cdae1c313a3710d3d977c Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Fri, 23 May 2008 23:50:41 +0200 Subject: ftrace: remove notrace Remove the notrace annotations. The build logic takes care of that. Signed-off-by: Thomas Gleixner diff --git a/kernel/trace/trace_sysprof.c b/kernel/trace/trace_sysprof.c index 1940623..3b1e4ba 100644 --- a/kernel/trace/trace_sysprof.c +++ b/kernel/trace/trace_sysprof.c @@ -155,7 +155,7 @@ static void stop_stack_timers(void) stop_stack_timer(cpu); } -static notrace void stack_reset(struct trace_array *tr) +static void stack_reset(struct trace_array *tr) { int cpu; @@ -165,7 +165,7 @@ static notrace void stack_reset(struct trace_array *tr) tracing_reset(tr->data[cpu]); } -static notrace void start_stack_trace(struct trace_array *tr) +static void start_stack_trace(struct trace_array *tr) { mutex_lock(&sample_timer_lock); stack_reset(tr); @@ -174,7 +174,7 @@ static notrace void start_stack_trace(struct trace_array *tr) mutex_unlock(&sample_timer_lock); } -static notrace void stop_stack_trace(struct trace_array *tr) +static void stop_stack_trace(struct trace_array *tr) { mutex_lock(&sample_timer_lock); stop_stack_timers(); @@ -182,7 +182,7 @@ static notrace void stop_stack_trace(struct trace_array *tr) mutex_unlock(&sample_timer_lock); } -static notrace void stack_trace_init(struct trace_array *tr) +static void stack_trace_init(struct trace_array *tr) { sysprof_trace = tr; @@ -190,7 +190,7 @@ static notrace void stack_trace_init(struct trace_array *tr) start_stack_trace(tr); } -static notrace void stack_trace_reset(struct trace_array *tr) +static void stack_trace_reset(struct trace_array *tr) { if (tr->ctrl) stop_stack_trace(tr); -- cgit v0.10.2 From 9caee613d3b860ae81b79370eeae9ac967c07536 Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Fri, 23 May 2008 23:55:54 +0200 Subject: ftrace: fix __trace_special() Signed-off-by: Thomas Gleixner diff --git a/kernel/trace/trace_sysprof.c b/kernel/trace/trace_sysprof.c index 3b1e4ba..76dd953 100644 --- a/kernel/trace/trace_sysprof.c +++ b/kernel/trace/trace_sysprof.c @@ -81,7 +81,7 @@ static void timer_notify(struct pt_regs *regs, int cpu) } - trace_special(tr, data, 0, current->pid, regs->ip); + __trace_special(tr, data, 0, current->pid, regs->ip); fp = (void __user *)regs->bp; @@ -93,18 +93,18 @@ static void timer_notify(struct pt_regs *regs, int cpu) if ((unsigned long)fp < regs->sp) break; - trace_special(tr, data, 1, frame.return_address, + __trace_special(tr, data, 1, frame.return_address, (unsigned long)fp); fp = frame.next_fp; } - trace_special(tr, data, 2, current->pid, i); + __trace_special(tr, data, 2, current->pid, i); /* * Special trace entry if we overflow the max depth: */ if (i == sample_max_depth) - trace_special(tr, data, -1, -1, -1); + __trace_special(tr, data, -1, -1, -1); } static enum hrtimer_restart stack_trace_timer_fn(struct hrtimer *hrtimer) -- cgit v0.10.2 From 5fc4511c756860149b81aead6eca5bdf5c438ea7 Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Fri, 23 May 2008 23:58:21 +0200 Subject: ftrace: make it more available in the Kconfig Signed-off-by: Thomas Gleixner diff --git a/kernel/trace/Kconfig b/kernel/trace/Kconfig index 9b49526..e101c9a 100644 --- a/kernel/trace/Kconfig +++ b/kernel/trace/Kconfig @@ -77,7 +77,7 @@ config PREEMPT_TRACER config SYSPROF_TRACER bool "Sysprof Tracer" - depends on DEBUG_KERNEL && X86 + depends on DEBUG_KERNEL select TRACING help This tracer provides the trace needed by the 'Sysprof' userspace -- cgit v0.10.2 From cd2134b1dda92fd450e6a1e12b1c7960dd6a2178 Mon Sep 17 00:00:00 2001 From: Soeren Sandmann Pedersen Date: Mon, 12 May 2008 21:20:54 +0200 Subject: sysprof: kernel trace add kernel backtracing to the sysprof tracer. change the format of the data, so that type=0 means beginning of stack trace, 1 means kernel address, 2 means user address, and 3 means end of trace. EIP addresses are no longer distinguished from return addresses, mostly because sysprof userspace doesn't make use of it. It may be worthwhile adding this back in though, just in case it becomes interesting. Signed-off-by: Ingo Molnar Signed-off-by: Thomas Gleixner diff --git a/kernel/trace/trace_sysprof.c b/kernel/trace/trace_sysprof.c index 76dd953..ebcb66d 100644 --- a/kernel/trace/trace_sysprof.c +++ b/kernel/trace/trace_sysprof.c @@ -14,6 +14,8 @@ #include #include +#include + #include "trace.h" static struct trace_array *sysprof_trace; @@ -52,6 +54,77 @@ static int copy_stack_frame(const void __user *fp, struct stack_frame *frame) return ret; } +struct backtrace_info { + struct trace_array_cpu *data; + struct trace_array *tr; + int pos; +}; + +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) +{ + /* Don't bother with IRQ stacks for now */ + return -1; +} + +static void backtrace_address(void *data, unsigned long addr, int reliable) +{ + struct backtrace_info *info = data; + + if (info->pos < sample_max_depth && reliable) { + __trace_special(info->tr, info->data, 1, addr, 0); + + info->pos++; + } +} + +const static struct stacktrace_ops backtrace_ops = { + .warning = backtrace_warning, + .warning_symbol = backtrace_warning_symbol, + .stack = backtrace_stack, + .address = backtrace_address, +}; + +static struct pt_regs * +trace_kernel(struct pt_regs *regs, struct trace_array *tr, + struct trace_array_cpu *data) +{ + struct backtrace_info info; + unsigned long bp; + char *user_stack; + char *stack; + + info.tr = tr; + info.data = data; + info.pos = 1; + + __trace_special(info.tr, info.data, 1, regs->ip, 0); + + stack = ((char *)regs + sizeof(struct pt_regs)); +#ifdef CONFIG_FRAME_POINTER + bp = regs->bp; +#else + bp = 0; +#endif + + dump_trace(NULL, regs, (void *)stack, bp, &backtrace_ops, &info); + + /* Now trace the user stack */ + user_stack = ((char *)current->thread.sp0 - sizeof(struct pt_regs)); + + return (struct pt_regs *)user_stack; +} + static void timer_notify(struct pt_regs *regs, int cpu) { struct trace_array_cpu *data; @@ -74,17 +147,15 @@ static void timer_notify(struct pt_regs *regs, int cpu) if (is_user && current->state != TASK_RUNNING) return; - if (!is_user) { - /* kernel */ - ftrace(tr, data, current->pid, 1, 0); - return; + __trace_special(tr, data, 0, 0, current->pid); - } - - __trace_special(tr, data, 0, current->pid, regs->ip); + if (!is_user) + regs = trace_kernel(regs, tr, data); fp = (void __user *)regs->bp; + __trace_special(tr, data, 2, regs->ip, 0); + for (i = 0; i < sample_max_depth; i++) { frame.next_fp = 0; frame.return_address = 0; @@ -93,12 +164,12 @@ static void timer_notify(struct pt_regs *regs, int cpu) if ((unsigned long)fp < regs->sp) break; - __trace_special(tr, data, 1, frame.return_address, + __trace_special(tr, data, 2, frame.return_address, (unsigned long)fp); fp = frame.next_fp; } - __trace_special(tr, data, 2, current->pid, i); + __trace_special(tr, data, 3, current->pid, i); /* * Special trace entry if we overflow the max depth: -- cgit v0.10.2 From 8a9e94c1fbfdac45a3b6811b880777c4116aa309 Mon Sep 17 00:00:00 2001 From: Ingo Molnar Date: Mon, 12 May 2008 21:20:54 +0200 Subject: sysprof: update copyrights Signed-off-by: Ingo Molnar Signed-off-by: Thomas Gleixner diff --git a/kernel/trace/trace_sysprof.c b/kernel/trace/trace_sysprof.c index ebcb66d..fe23d6d 100644 --- a/kernel/trace/trace_sysprof.c +++ b/kernel/trace/trace_sysprof.c @@ -1,9 +1,9 @@ /* * trace stack traces * + * Copyright (C) 2004-2008, Soeren Sandmann * Copyright (C) 2007 Steven Rostedt * Copyright (C) 2008 Ingo Molnar - * Copyright (C) 2004, 2005, Soeren Sandmann */ #include #include -- cgit v0.10.2 From cf3271a73b612a03da00681ecd9bfefab37c74c9 Mon Sep 17 00:00:00 2001 From: Soeren Sandmann Date: Mon, 12 May 2008 05:28:50 +0200 Subject: ftrace/sysprof: don't trace the user stack if we are a kernel thread. Check that current->mm is non-NULL before attempting to trace the user stack. Also take depth of the kernel stack into account when comparing against sample_max_depth. Signed-off-by: Ingo Molnar Signed-off-by: Thomas Gleixner diff --git a/kernel/trace/trace_sysprof.c b/kernel/trace/trace_sysprof.c index fe23d6d..2301e1e 100644 --- a/kernel/trace/trace_sysprof.c +++ b/kernel/trace/trace_sysprof.c @@ -95,13 +95,12 @@ const static struct stacktrace_ops backtrace_ops = { .address = backtrace_address, }; -static struct pt_regs * +static int trace_kernel(struct pt_regs *regs, struct trace_array *tr, struct trace_array_cpu *data) { struct backtrace_info info; unsigned long bp; - char *user_stack; char *stack; info.tr = tr; @@ -119,10 +118,7 @@ trace_kernel(struct pt_regs *regs, struct trace_array *tr, dump_trace(NULL, regs, (void *)stack, bp, &backtrace_ops, &info); - /* Now trace the user stack */ - user_stack = ((char *)current->thread.sp0 - sizeof(struct pt_regs)); - - return (struct pt_regs *)user_stack; + return info.pos; } static void timer_notify(struct pt_regs *regs, int cpu) @@ -150,32 +146,44 @@ static void timer_notify(struct pt_regs *regs, int cpu) __trace_special(tr, data, 0, 0, current->pid); if (!is_user) - regs = trace_kernel(regs, tr, data); + i = trace_kernel(regs, tr, data); + else + i = 0; - fp = (void __user *)regs->bp; + /* + * Trace user stack if we are not a kernel thread + */ + if (current->mm && i < sample_max_depth) { + regs = (struct pt_regs *)current->thread.sp0 - 1; - __trace_special(tr, data, 2, regs->ip, 0); + fp = (void __user *)regs->bp; - for (i = 0; i < sample_max_depth; i++) { - frame.next_fp = 0; - frame.return_address = 0; - if (!copy_stack_frame(fp, &frame)) - break; - if ((unsigned long)fp < regs->sp) - break; + __trace_special(tr, data, 2, regs->ip, 0); - __trace_special(tr, data, 2, frame.return_address, - (unsigned long)fp); - fp = frame.next_fp; - } + while (i < sample_max_depth) { + frame.next_fp = 0; + frame.return_address = 0; + if (!copy_stack_frame(fp, &frame)) + break; + if ((unsigned long)fp < regs->sp) + break; - __trace_special(tr, data, 3, current->pid, i); + __trace_special(tr, data, 2, frame.return_address, + (unsigned long)fp); + fp = frame.next_fp; + + i++; + } + + } /* * Special trace entry if we overflow the max depth: */ if (i == sample_max_depth) __trace_special(tr, data, -1, -1, -1); + + __trace_special(tr, data, 3, current->pid, i); } static enum hrtimer_restart stack_trace_timer_fn(struct hrtimer *hrtimer) -- cgit v0.10.2 From 4d2df795f0c3eb91f97a666f47716121a2f166ed Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Sat, 24 May 2008 15:00:46 +0200 Subject: sysprof: make it depend on X86 Signed-off-by: Thomas Gleixner diff --git a/kernel/trace/Kconfig b/kernel/trace/Kconfig index e101c9a..263e9e6 100644 --- a/kernel/trace/Kconfig +++ b/kernel/trace/Kconfig @@ -77,7 +77,7 @@ config PREEMPT_TRACER config SYSPROF_TRACER bool "Sysprof Tracer" - depends on DEBUG_KERNEL + depends on X86 select TRACING help This tracer provides the trace needed by the 'Sysprof' userspace -- cgit v0.10.2