[PATCH v2 0/2] locking/atomic/x86: Introduce arch_try_cmpxchg64

[PATCH v2 0/2] locking/atomic/x86: Introduce arch_try_cmpxchg64

[Uros Bizjak]

The followign patchset introduces generic support for try_cmpxchg64
and introduces arch_try_cmpxchg64 for 64-bit and 32-bit targets.
On 64-bit targets, the generated assembly improves from:

  ab:	89 c8                	mov    %ecx,%eax
  ad:	48 89 4c 24 60       	mov    %rcx,0x60(%rsp)
  b2:	83 e0 fd             	and    $0xfffffffd,%eax
  b5:	89 54 24 64          	mov    %edx,0x64(%rsp)
  b9:	88 44 24 60          	mov    %al,0x60(%rsp)
  bd:	48 89 c8             	mov    %rcx,%rax
  c0:	c6 44 24 62 f2       	movb   $0xf2,0x62(%rsp)
  c5:	48 8b 74 24 60       	mov    0x60(%rsp),%rsi
  ca:	f0 49 0f b1 34 24    	lock cmpxchg %rsi,(%r12)
  d0:	48 39 c1             	cmp    %rax,%rcx
  d3:	75 cf                	jne    a4 <t+0xa4>

to:

  b3:	89 c2                	mov    %eax,%edx
  b5:	48 89 44 24 60       	mov    %rax,0x60(%rsp)
  ba:	83 e2 fd             	and    $0xfffffffd,%edx
  bd:	89 4c 24 64          	mov    %ecx,0x64(%rsp)
  c1:	88 54 24 60          	mov    %dl,0x60(%rsp)
  c5:	c6 44 24 62 f2       	movb   $0xf2,0x62(%rsp)
  ca:	48 8b 54 24 60       	mov    0x60(%rsp),%rdx
  cf:	f0 48 0f b1 13       	lock cmpxchg %rdx,(%rbx)
  d4:	75 d5                	jne    ab <t+0xab>

where a move and a compare after cmpxchg is saved.  The improvements
for 32-bit targets are even more noticeable, because dual-word compare
after cmpxchg8b gets eliminated.

Changes since v1:
 * Implement full support for try_cmpxchg64{,_acquire,_release,_relaxed}
and their falbacks involving cmpxchg64.
 * Split patch to generic and target-dependant part.
 * Modernize __try_cmpxchg64 asm template with symbolic operand name.
 * Use generic fallback when arch_try_cmpxchg64 is not defined.

Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Will Deacon <will@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Boqun Feng <boqun.feng@gmail.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: "Paul E. McKenney" <paulmck@kernel.org>
Cc: Marco Elver <elver@google.com>

Uros Bizjak (2):
  locking/atomic: Add generic try_cmpxchg64 support
  locking/atomic/x86: Introduce arch_try_cmpxchg64

 arch/x86/include/asm/cmpxchg_32.h           | 21 ++++++
 arch/x86/include/asm/cmpxchg_64.h           |  6 ++
 include/linux/atomic/atomic-arch-fallback.h | 72 ++++++++++++++++++++-
 include/linux/atomic/atomic-instrumented.h  | 40 +++++++++++-
 scripts/atomic/gen-atomic-fallback.sh       | 31 +++++----
 scripts/atomic/gen-atomic-instrumented.sh   |  2 +-
 6 files changed, 156 insertions(+), 16 deletions(-)

-- 
2.35.1

[PATCH v2 1/2] locking/atomic: Add generic try_cmpxchg64 support

[Uros Bizjak]

Add generic support for try_cmpxchg64{,_acquire,_release,_relaxed}
and their falbacks involving cmpxchg64.

Signed-off-by: Uros Bizjak <ubizjak@gmail.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Will Deacon <will@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Boqun Feng <boqun.feng@gmail.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: "Paul E. McKenney" <paulmck@kernel.org>
Cc: Marco Elver <elver@google.com>
---
 include/linux/atomic/atomic-arch-fallback.h | 72 ++++++++++++++++++++-
 include/linux/atomic/atomic-instrumented.h  | 40 +++++++++++-
 scripts/atomic/gen-atomic-fallback.sh       | 31 +++++----
 scripts/atomic/gen-atomic-instrumented.sh   |  2 +-
 4 files changed, 129 insertions(+), 16 deletions(-)

diff --git a/include/linux/atomic/atomic-arch-fallback.h b/include/linux/atomic/atomic-arch-fallback.h
index 6db58d180866..77bc5522e61c 100644
--- a/include/linux/atomic/atomic-arch-fallback.h
+++ b/include/linux/atomic/atomic-arch-fallback.h
@@ -147,6 +147,76 @@
 
 #endif /* arch_try_cmpxchg_relaxed */
 
+#ifndef arch_try_cmpxchg64_relaxed
+#ifdef arch_try_cmpxchg64
+#define arch_try_cmpxchg64_acquire arch_try_cmpxchg64
+#define arch_try_cmpxchg64_release arch_try_cmpxchg64
+#define arch_try_cmpxchg64_relaxed arch_try_cmpxchg64
+#endif /* arch_try_cmpxchg64 */
+
+#ifndef arch_try_cmpxchg64
+#define arch_try_cmpxchg64(_ptr, _oldp, _new) \
+({ \
+	typeof(*(_ptr)) *___op = (_oldp), ___o = *___op, ___r; \
+	___r = arch_cmpxchg64((_ptr), ___o, (_new)); \
+	if (unlikely(___r != ___o)) \
+		*___op = ___r; \
+	likely(___r == ___o); \
+})
+#endif /* arch_try_cmpxchg64 */
+
+#ifndef arch_try_cmpxchg64_acquire
+#define arch_try_cmpxchg64_acquire(_ptr, _oldp, _new) \
+({ \
+	typeof(*(_ptr)) *___op = (_oldp), ___o = *___op, ___r; \
+	___r = arch_cmpxchg64_acquire((_ptr), ___o, (_new)); \
+	if (unlikely(___r != ___o)) \
+		*___op = ___r; \
+	likely(___r == ___o); \
+})
+#endif /* arch_try_cmpxchg64_acquire */
+
+#ifndef arch_try_cmpxchg64_release
+#define arch_try_cmpxchg64_release(_ptr, _oldp, _new) \
+({ \
+	typeof(*(_ptr)) *___op = (_oldp), ___o = *___op, ___r; \
+	___r = arch_cmpxchg64_release((_ptr), ___o, (_new)); \
+	if (unlikely(___r != ___o)) \
+		*___op = ___r; \
+	likely(___r == ___o); \
+})
+#endif /* arch_try_cmpxchg64_release */
+
+#ifndef arch_try_cmpxchg64_relaxed
+#define arch_try_cmpxchg64_relaxed(_ptr, _oldp, _new) \
+({ \
+	typeof(*(_ptr)) *___op = (_oldp), ___o = *___op, ___r; \
+	___r = arch_cmpxchg64_relaxed((_ptr), ___o, (_new)); \
+	if (unlikely(___r != ___o)) \
+		*___op = ___r; \
+	likely(___r == ___o); \
+})
+#endif /* arch_try_cmpxchg64_relaxed */
+
+#else /* arch_try_cmpxchg64_relaxed */
+
+#ifndef arch_try_cmpxchg64_acquire
+#define arch_try_cmpxchg64_acquire(...) \
+	__atomic_op_acquire(arch_try_cmpxchg64, __VA_ARGS__)
+#endif
+
+#ifndef arch_try_cmpxchg64_release
+#define arch_try_cmpxchg64_release(...) \
+	__atomic_op_release(arch_try_cmpxchg64, __VA_ARGS__)
+#endif
+
+#ifndef arch_try_cmpxchg64
+#define arch_try_cmpxchg64(...) \
+	__atomic_op_fence(arch_try_cmpxchg64, __VA_ARGS__)
+#endif
+
+#endif /* arch_try_cmpxchg64_relaxed */
+
 #ifndef arch_atomic_read_acquire
 static __always_inline int
 arch_atomic_read_acquire(const atomic_t *v)
@@ -2386,4 +2456,4 @@ arch_atomic64_dec_if_positive(atomic64_t *v)
 #endif
 
 #endif /* _LINUX_ATOMIC_FALLBACK_H */
-// 8e2cc06bc0d2c0967d2f8424762bd48555ee40ae
+// b5e87bdd5ede61470c29f7a7e4de781af3770f09
diff --git a/include/linux/atomic/atomic-instrumented.h b/include/linux/atomic/atomic-instrumented.h
index 5d69b143c28e..7a139ec030b0 100644
--- a/include/linux/atomic/atomic-instrumented.h
+++ b/include/linux/atomic/atomic-instrumented.h
@@ -2006,6 +2006,44 @@ atomic_long_dec_if_positive(atomic_long_t *v)
 	arch_try_cmpxchg_relaxed(__ai_ptr, __ai_oldp, __VA_ARGS__); \
 })
 
+#define try_cmpxchg64(ptr, oldp, ...) \
+({ \
+	typeof(ptr) __ai_ptr = (ptr); \
+	typeof(oldp) __ai_oldp = (oldp); \
+	kcsan_mb(); \
+	instrument_atomic_write(__ai_ptr, sizeof(*__ai_ptr)); \
+	instrument_atomic_write(__ai_oldp, sizeof(*__ai_oldp)); \
+	arch_try_cmpxchg64(__ai_ptr, __ai_oldp, __VA_ARGS__); \
+})
+
+#define try_cmpxchg64_acquire(ptr, oldp, ...) \
+({ \
+	typeof(ptr) __ai_ptr = (ptr); \
+	typeof(oldp) __ai_oldp = (oldp); \
+	instrument_atomic_write(__ai_ptr, sizeof(*__ai_ptr)); \
+	instrument_atomic_write(__ai_oldp, sizeof(*__ai_oldp)); \
+	arch_try_cmpxchg64_acquire(__ai_ptr, __ai_oldp, __VA_ARGS__); \
+})
+
+#define try_cmpxchg64_release(ptr, oldp, ...) \
+({ \
+	typeof(ptr) __ai_ptr = (ptr); \
+	typeof(oldp) __ai_oldp = (oldp); \
+	kcsan_release(); \
+	instrument_atomic_write(__ai_ptr, sizeof(*__ai_ptr)); \
+	instrument_atomic_write(__ai_oldp, sizeof(*__ai_oldp)); \
+	arch_try_cmpxchg64_release(__ai_ptr, __ai_oldp, __VA_ARGS__); \
+})
+
+#define try_cmpxchg64_relaxed(ptr, oldp, ...) \
+({ \
+	typeof(ptr) __ai_ptr = (ptr); \
+	typeof(oldp) __ai_oldp = (oldp); \
+	instrument_atomic_write(__ai_ptr, sizeof(*__ai_ptr)); \
+	instrument_atomic_write(__ai_oldp, sizeof(*__ai_oldp)); \
+	arch_try_cmpxchg64_relaxed(__ai_ptr, __ai_oldp, __VA_ARGS__); \
+})
+
 #define cmpxchg_local(ptr, ...) \
 ({ \
 	typeof(ptr) __ai_ptr = (ptr); \
@@ -2045,4 +2083,4 @@ atomic_long_dec_if_positive(atomic_long_t *v)
 })
 
 #endif /* _LINUX_ATOMIC_INSTRUMENTED_H */
-// 87c974b93032afd42143613434d1a7788fa598f9
+// 764f741eb77a7ad565dc8d99ce2837d5542e8aee
diff --git a/scripts/atomic/gen-atomic-fallback.sh b/scripts/atomic/gen-atomic-fallback.sh
index 8e2da71f1d5f..3a07695e3c89 100755
--- a/scripts/atomic/gen-atomic-fallback.sh
+++ b/scripts/atomic/gen-atomic-fallback.sh
@@ -164,41 +164,44 @@ gen_xchg_fallbacks()
 
 gen_try_cmpxchg_fallback()
 {
+	local cmpxchg="$1"; shift;
 	local order="$1"; shift;
 
 cat <<EOF
-#ifndef arch_try_cmpxchg${order}
-#define arch_try_cmpxchg${order}(_ptr, _oldp, _new) \\
+#ifndef arch_try_${cmpxchg}${order}
+#define arch_try_${cmpxchg}${order}(_ptr, _oldp, _new) \\
 ({ \\
 	typeof(*(_ptr)) *___op = (_oldp), ___o = *___op, ___r; \\
-	___r = arch_cmpxchg${order}((_ptr), ___o, (_new)); \\
+	___r = arch_${cmpxchg}${order}((_ptr), ___o, (_new)); \\
 	if (unlikely(___r != ___o)) \\
 		*___op = ___r; \\
 	likely(___r == ___o); \\
 })
-#endif /* arch_try_cmpxchg${order} */
+#endif /* arch_try_${cmpxchg}${order} */
 
 EOF
 }
 
 gen_try_cmpxchg_fallbacks()
 {
-	printf "#ifndef arch_try_cmpxchg_relaxed\n"
-	printf "#ifdef arch_try_cmpxchg\n"
+	local cmpxchg="$1"; shift;
 
-	gen_basic_fallbacks "arch_try_cmpxchg"
+	printf "#ifndef arch_try_${cmpxchg}_relaxed\n"
+	printf "#ifdef arch_try_${cmpxchg}\n"
 
-	printf "#endif /* arch_try_cmpxchg */\n\n"
+	gen_basic_fallbacks "arch_try_${cmpxchg}"
+
+	printf "#endif /* arch_try_${cmpxchg} */\n\n"
 
 	for order in "" "_acquire" "_release" "_relaxed"; do
-		gen_try_cmpxchg_fallback "${order}"
+		gen_try_cmpxchg_fallback "${cmpxchg}" "${order}"
 	done
 
-	printf "#else /* arch_try_cmpxchg_relaxed */\n"
+	printf "#else /* arch_try_${cmpxchg}_relaxed */\n"
 
-	gen_order_fallbacks "arch_try_cmpxchg"
+	gen_order_fallbacks "arch_try_${cmpxchg}"
 
-	printf "#endif /* arch_try_cmpxchg_relaxed */\n\n"
+	printf "#endif /* arch_try_${cmpxchg}_relaxed */\n\n"
 }
 
 cat << EOF
@@ -218,7 +221,9 @@ for xchg in "arch_xchg" "arch_cmpxchg" "arch_cmpxchg64"; do
 	gen_xchg_fallbacks "${xchg}"
 done
 
-gen_try_cmpxchg_fallbacks
+for cmpxchg in "cmpxchg" "cmpxchg64"; do
+	gen_try_cmpxchg_fallbacks "${cmpxchg}"
+done
 
 grep '^[a-z]' "$1" | while read name meta args; do
 	gen_proto "${meta}" "${name}" "atomic" "int" ${args}
diff --git a/scripts/atomic/gen-atomic-instrumented.sh b/scripts/atomic/gen-atomic-instrumented.sh
index 68f902731d01..77c06526a574 100755
--- a/scripts/atomic/gen-atomic-instrumented.sh
+++ b/scripts/atomic/gen-atomic-instrumented.sh
@@ -166,7 +166,7 @@ grep '^[a-z]' "$1" | while read name meta args; do
 done
 
 
-for xchg in "xchg" "cmpxchg" "cmpxchg64" "try_cmpxchg"; do
+for xchg in "xchg" "cmpxchg" "cmpxchg64" "try_cmpxchg" "try_cmpxchg64"; do
 	for order in "" "_acquire" "_release" "_relaxed"; do
 		gen_xchg "${xchg}" "${order}" ""
 		printf "\n"
-- 
2.35.1

[PATCH v2 2/2] locking/atomic/x86: Introduce arch_try_cmpxchg64

[Uros Bizjak]

Introduce arch_try_cmpxchg64 for 64-bit and 32-bit targets to improve
code using cmpxchg64.  On 64-bit targets, the generated assembly improves
from:

  ab:	89 c8                	mov    %ecx,%eax
  ad:	48 89 4c 24 60       	mov    %rcx,0x60(%rsp)
  b2:	83 e0 fd             	and    $0xfffffffd,%eax
  b5:	89 54 24 64          	mov    %edx,0x64(%rsp)
  b9:	88 44 24 60          	mov    %al,0x60(%rsp)
  bd:	48 89 c8             	mov    %rcx,%rax
  c0:	c6 44 24 62 f2       	movb   $0xf2,0x62(%rsp)
  c5:	48 8b 74 24 60       	mov    0x60(%rsp),%rsi
  ca:	f0 49 0f b1 34 24    	lock cmpxchg %rsi,(%r12)
  d0:	48 39 c1             	cmp    %rax,%rcx
  d3:	75 cf                	jne    a4 <t+0xa4>

to:

  b3:	89 c2                	mov    %eax,%edx
  b5:	48 89 44 24 60       	mov    %rax,0x60(%rsp)
  ba:	83 e2 fd             	and    $0xfffffffd,%edx
  bd:	89 4c 24 64          	mov    %ecx,0x64(%rsp)
  c1:	88 54 24 60          	mov    %dl,0x60(%rsp)
  c5:	c6 44 24 62 f2       	movb   $0xf2,0x62(%rsp)
  ca:	48 8b 54 24 60       	mov    0x60(%rsp),%rdx
  cf:	f0 48 0f b1 13       	lock cmpxchg %rdx,(%rbx)
  d4:	75 d5                	jne    ab <t+0xab>

where a move and a compare after cmpxchg is saved.  The improvements
for 32-bit targets are even more noticeable, because dual-word compare
after cmpxchg8b gets eliminated.

Signed-off-by: Uros Bizjak <ubizjak@gmail.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Will Deacon <will@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Boqun Feng <boqun.feng@gmail.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: "Paul E. McKenney" <paulmck@kernel.org>
Cc: Marco Elver <elver@google.com>
---
 arch/x86/include/asm/cmpxchg_32.h | 21 +++++++++++++++++++++
 arch/x86/include/asm/cmpxchg_64.h |  6 ++++++
 2 files changed, 27 insertions(+)

diff --git a/arch/x86/include/asm/cmpxchg_32.h b/arch/x86/include/asm/cmpxchg_32.h
index 0a7fe0321613..16a604cf24d4 100644
--- a/arch/x86/include/asm/cmpxchg_32.h
+++ b/arch/x86/include/asm/cmpxchg_32.h
@@ -42,6 +42,9 @@ static inline void set_64bit(volatile u64 *ptr, u64 value)
 #define arch_cmpxchg64_local(ptr, o, n)					\
 	((__typeof__(*(ptr)))__cmpxchg64_local((ptr), (unsigned long long)(o), \
 					       (unsigned long long)(n)))
+#define arch_try_cmpxchg64(ptr, po, n)					\
+	((__typeof__(*(ptr)))__try_cmpxchg64((ptr), (unsigned long long *)(po), \
+					     (unsigned long long)(n)))
 #endif
 
 static inline u64 __cmpxchg64(volatile u64 *ptr, u64 old, u64 new)
@@ -70,6 +73,24 @@ static inline u64 __cmpxchg64_local(volatile u64 *ptr, u64 old, u64 new)
 	return prev;
 }
 
+static inline bool __try_cmpxchg64(volatile u64 *ptr, u64 *pold, u64 new)
+{
+	bool success;
+	u64 old = *pold;
+	asm volatile(LOCK_PREFIX "cmpxchg8b %[ptr]"
+		     CC_SET(z)
+		     : CC_OUT(z) (success),
+		       [ptr] "+m" (*ptr),
+		       "+A" (old)
+		     : "b" ((u32)new),
+		       "c" ((u32)(new >> 32))
+		     : "memory");
+
+	if (unlikely(!success))
+		*pold = old;
+	return success;
+}
+
 #ifndef CONFIG_X86_CMPXCHG64
 /*
  * Building a kernel capable running on 80386 and 80486. It may be necessary
diff --git a/arch/x86/include/asm/cmpxchg_64.h b/arch/x86/include/asm/cmpxchg_64.h
index 072e5459fe2f..250187ac8248 100644
--- a/arch/x86/include/asm/cmpxchg_64.h
+++ b/arch/x86/include/asm/cmpxchg_64.h
@@ -19,6 +19,12 @@ static inline void set_64bit(volatile u64 *ptr, u64 val)
 	arch_cmpxchg_local((ptr), (o), (n));				\
 })
 
+#define arch_try_cmpxchg64(ptr, po, n)					\
+({									\
+	BUILD_BUG_ON(sizeof(*(ptr)) != 8);				\
+	arch_try_cmpxchg((ptr), (po), (n));				\
+})
+
 #define system_has_cmpxchg_double() boot_cpu_has(X86_FEATURE_CX16)
 
 #endif /* _ASM_X86_CMPXCHG_64_H */
-- 
2.35.1

Re: [PATCH v2 2/2] locking/atomic/x86: Introduce arch_try_cmpxchg64

[Peter Zijlstra]

On Fri, May 13, 2022 at 05:30:09PM +0200, Uros Bizjak wrote:


> +#define arch_try_cmpxchg64(ptr, po, n)					\
> +	((__typeof__(*(ptr)))__try_cmpxchg64((ptr), (unsigned long long *)(po), \
> +					     (unsigned long long)(n)))

That doesn't look right (unless it's so late I really can't read
anymore, in which case ignore me and I'll try again on monday). But the
return value of try_cmpxchg is bool, not typeof(*ptr).

Re: [PATCH v2 2/2] locking/atomic/x86: Introduce arch_try_cmpxchg64

[Uros Bizjak]

On Sat, May 14, 2022 at 12:18 AM Peter Zijlstra <peterz@infradead.org> wrote:
>
> On Fri, May 13, 2022 at 05:30:09PM +0200, Uros Bizjak wrote:
>
>
> > +#define arch_try_cmpxchg64(ptr, po, n)                                       \
> > +     ((__typeof__(*(ptr)))__try_cmpxchg64((ptr), (unsigned long long *)(po), \
> > +                                          (unsigned long long)(n)))
>
> That doesn't look right (unless it's so late I really can't read
> anymore, in which case ignore me and I'll try again on monday). But the
> return value of try_cmpxchg is bool, not typeof(*ptr).

No, you are right, I was too eager when copying the code from the
above arch_cmpxchg64 definition. Unfortunately, although the cast is
benign and the compiler figures out that the cast is unnecessary, it
doesn't warn here...

Actually, since __try_cmpxchg64 is already bool, we don't need any
cast here, and the definition can be substantially simplified.

Thank you for another pair of eyes - I did eyeball this code
extensively, but the issue slipped through somehow.

Patch v3 is in the works.

Thanks,
Uros.