arm64: atomics: prefetch the destination word for write prior to stxr

The cost of changing a cacheline from shared to exclusive state can be
significant, especially when this is triggered by an exclusive store,
since it may result in having to retry the transaction.

This patch makes use of prfm to prefetch cachelines for write prior to
ldxr/stxr loops when using the ll/sc atomic routines.

Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>

4 files changed, 21 insertions, 0 deletions

diff --git a/arch/arm64/include/asm/atomic_ll_sc.h b/arch/arm64/include/asm/atomic_ll_sc.h
index 5a9fb37..50d6abd 100644
--- a/arch/arm64/include/asm/atomic_ll_sc.h
+++ b/arch/arm64/include/asm/atomic_ll_sc.h
@@ -45,6 +45,7 @@ __LL_SC_PREFIX(atomic_##op(int i, atomic_t *v))				\
 	int result;							\
 									\
 	asm volatile("// atomic_" #op "\n"				\
+"	prfm	pstl1strm, %2\n"					\
 "1:	ldxr	%w0, %2\n"						\
 "	" #asm_op "	%w0, %w0, %w3\n"				\
 "	stxr	%w1, %w0, %2\n"						\
@@ -62,6 +63,7 @@ __LL_SC_PREFIX(atomic_##op##_return(int i, atomic_t *v))		\
 	int result;							\
 									\
 	asm volatile("// atomic_" #op "_return\n"			\
+"	prfm	pstl1strm, %2\n"					\
 "1:	ldxr	%w0, %2\n"						\
 "	" #asm_op "	%w0, %w0, %w3\n"				\
 "	stlxr	%w1, %w0, %2\n"						\
@@ -98,6 +100,7 @@ __LL_SC_PREFIX(atomic_cmpxchg(atomic_t *ptr, int old, int new))
 	int oldval;
 
 	asm volatile("// atomic_cmpxchg\n"
+"	prfm	pstl1strm, %2\n"
 "1:	ldxr	%w1, %2\n"
 "	eor	%w0, %w1, %w3\n"
 "	cbnz	%w0, 2f\n"
@@ -121,6 +124,7 @@ __LL_SC_PREFIX(atomic64_##op(long i, atomic64_t *v))			\
 	unsigned long tmp;						\
 									\
 	asm volatile("// atomic64_" #op "\n"				\
+"	prfm	pstl1strm, %2\n"					\
 "1:	ldxr	%0, %2\n"						\
 "	" #asm_op "	%0, %0, %3\n"					\
 "	stxr	%w1, %0, %2\n"						\
@@ -138,6 +142,7 @@ __LL_SC_PREFIX(atomic64_##op##_return(long i, atomic64_t *v))		\
 	unsigned long tmp;						\
 									\
 	asm volatile("// atomic64_" #op "_return\n"			\
+"	prfm	pstl1strm, %2\n"					\
 "1:	ldxr	%0, %2\n"						\
 "	" #asm_op "	%0, %0, %3\n"					\
 "	stlxr	%w1, %0, %2\n"						\
@@ -174,6 +179,7 @@ __LL_SC_PREFIX(atomic64_cmpxchg(atomic64_t *ptr, long old, long new))
 	unsigned long res;
 
 	asm volatile("// atomic64_cmpxchg\n"
+"	prfm	pstl1strm, %2\n"
 "1:	ldxr	%1, %2\n"
 "	eor	%0, %1, %3\n"
 "	cbnz	%w0, 2f\n"
@@ -196,6 +202,7 @@ __LL_SC_PREFIX(atomic64_dec_if_positive(atomic64_t *v))
 	unsigned long tmp;
 
 	asm volatile("// atomic64_dec_if_positive\n"
+"	prfm	pstl1strm, %2\n"
 "1:	ldxr	%0, %2\n"
 "	subs	%0, %0, #1\n"
 "	b.mi	2f\n"
@@ -220,6 +227,7 @@ __LL_SC_PREFIX(__cmpxchg_case_##name(volatile void *ptr,		\
 	unsigned long tmp, oldval;					\
 									\
 	asm volatile(							\
+	"	prfm	pstl1strm, %2\n"				\
 	"1:	ldxr" #sz "\t%" #w "[oldval], %[v]\n"			\
 	"	eor	%" #w "[tmp], %" #w "[oldval], %" #w "[old]\n"	\
 	"	cbnz	%" #w "[tmp], 2f\n"				\
@@ -259,6 +267,7 @@ __LL_SC_PREFIX(__cmpxchg_double##name(unsigned long old1,		\
 	unsigned long tmp, ret;						\
 									\
 	asm volatile("// __cmpxchg_double" #name "\n"			\
+	"	prfm	pstl1strm, %2\n"				\
 	"1:	ldxp	%0, %1, %2\n"					\
 	"	eor	%0, %0, %3\n"					\
 	"	eor	%1, %1, %4\n"					\
diff --git a/arch/arm64/include/asm/cmpxchg.h b/arch/arm64/include/asm/cmpxchg.h
index f702126..7bfda09 100644
--- a/arch/arm64/include/asm/cmpxchg.h
+++ b/arch/arm64/include/asm/cmpxchg.h
@@ -33,12 +33,14 @@ static inline unsigned long __xchg(unsigned long x, volatile void *ptr, int size
 	case 1:
 		asm volatile(ARM64_LSE_ATOMIC_INSN(
 		/* LL/SC */
+		"	prfm	pstl1strm, %2\n"
 		"1:	ldxrb	%w0, %2\n"
 		"	stlxrb	%w1, %w3, %2\n"
 		"	cbnz	%w1, 1b\n"
 		"	dmb	ish",
 		/* LSE atomics */
 		"	nop\n"
+		"	nop\n"
 		"	swpalb	%w3, %w0, %2\n"
 		"	nop\n"
 		"	nop")
@@ -49,12 +51,14 @@ static inline unsigned long __xchg(unsigned long x, volatile void *ptr, int size
 	case 2:
 		asm volatile(ARM64_LSE_ATOMIC_INSN(
 		/* LL/SC */
+		"	prfm	pstl1strm, %2\n"
 		"1:	ldxrh	%w0, %2\n"
 		"	stlxrh	%w1, %w3, %2\n"
 		"	cbnz	%w1, 1b\n"
 		"	dmb	ish",
 		/* LSE atomics */
 		"	nop\n"
+		"	nop\n"
 		"	swpalh	%w3, %w0, %2\n"
 		"	nop\n"
 		"	nop")
@@ -65,12 +69,14 @@ static inline unsigned long __xchg(unsigned long x, volatile void *ptr, int size
 	case 4:
 		asm volatile(ARM64_LSE_ATOMIC_INSN(
 		/* LL/SC */
+		"	prfm	pstl1strm, %2\n"
 		"1:	ldxr	%w0, %2\n"
 		"	stlxr	%w1, %w3, %2\n"
 		"	cbnz	%w1, 1b\n"
 		"	dmb	ish",
 		/* LSE atomics */
 		"	nop\n"
+		"	nop\n"
 		"	swpal	%w3, %w0, %2\n"
 		"	nop\n"
 		"	nop")
@@ -81,12 +87,14 @@ static inline unsigned long __xchg(unsigned long x, volatile void *ptr, int size
 	case 8:
 		asm volatile(ARM64_LSE_ATOMIC_INSN(
 		/* LL/SC */
+		"	prfm	pstl1strm, %2\n"
 		"1:	ldxr	%0, %2\n"
 		"	stlxr	%w1, %3, %2\n"
 		"	cbnz	%w1, 1b\n"
 		"	dmb	ish",
 		/* LSE atomics */
 		"	nop\n"
+		"	nop\n"
 		"	swpal	%3, %0, %2\n"
 		"	nop\n"
 		"	nop")
diff --git a/arch/arm64/include/asm/futex.h b/arch/arm64/include/asm/futex.h
index 775e85b..007a69f 100644
--- a/arch/arm64/include/asm/futex.h
+++ b/arch/arm64/include/asm/futex.h
@@ -30,6 +30,7 @@
 	asm volatile(							\
 	ALTERNATIVE("nop", SET_PSTATE_PAN(0), ARM64_HAS_PAN,		\
 		    CONFIG_ARM64_PAN)					\
+"	prfm	pstl1strm, %2\n"					\
 "1:	ldxr	%w1, %2\n"						\
 	insn "\n"							\
 "2:	stlxr	%w3, %w0, %2\n"						\
@@ -120,6 +121,7 @@ futex_atomic_cmpxchg_inatomic(u32 *uval, u32 __user *uaddr,
 		return -EFAULT;
 
 	asm volatile("// futex_atomic_cmpxchg_inatomic\n"
+"	prfm	pstl1strm, %2\n"
 "1:	ldxr	%w1, %2\n"
 "	sub	%w3, %w1, %w4\n"
 "	cbnz	%w3, 3f\n"
diff --git a/arch/arm64/lib/bitops.S b/arch/arm64/lib/bitops.S
index bc18457..43ac736 100644
--- a/arch/arm64/lib/bitops.S
+++ b/arch/arm64/lib/bitops.S
@@ -31,6 +31,7 @@ ENTRY(	\name	)
 	eor	w0, w0, w3		// Clear low bits
 	mov	x2, #1
 	add	x1, x1, x0, lsr #3	// Get word offset
+alt_lse "	prfm	pstl1strm, [x1]",	"nop"
 	lsl	x3, x2, x3		// Create mask
 
 alt_lse	"1:	ldxr	x2, [x1]",		"\lse	x3, [x1]"
@@ -48,6 +49,7 @@ ENTRY(	\name	)
 	eor	w0, w0, w3		// Clear low bits
 	mov	x2, #1
 	add	x1, x1, x0, lsr #3	// Get word offset
+alt_lse "	prfm	pstl1strm, [x1]",	"nop"
 	lsl	x4, x2, x3		// Create mask
 
 alt_lse	"1:	ldxr	x2, [x1]",		"\lse	x4, x2, [x1]"