Skip to contentSkip to conversation

locking/pvqspinlock/x86: Remove redundant CMP after CMPXCHG in __raw_callee_save___pv_queued_spin_unlock()

PATCH Applied Uros Bizjak 2 participants 3 messages Apr 12, 2024 x86

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 locking/pvqspinlock/x86: Remove redundant CMP after CMPXCHG in __raw_callee_save___pv_queued_spin_unlock()

x86 CMPXCHG instruction returns success in the ZF flag. Remove redundant CMP instruction after CMPXCHG that performs the same check.

Also update the function comment to mention the modern version of the equivalent C code.

Cc Peter Zijlstra <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Will Deacon <[email protected]>
Cc Waiman Long <[email protected]>
Cc Boqun Feng <[email protected]>
---
 arch/x86/include/asm/qspinlock_paravirt.h | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)
arch/x86/include/asm/qspinlock_paravirt.h +2 -3
index ef9697f20129..466af57b8ed6 100644
@@ -25,9 +25,9 @@ __PV_CALLEE_SAVE_REGS_THUNK(__pv_queued_spin_unlock_slowpath, ".spinlock.text");
  *
  * void __lockfunc __pv_queued_spin_unlock(struct qspinlock *lock)
  * {
- *	u8 lockval = cmpxchg(&lock->locked, _Q_LOCKED_VAL, 0);
+ *	u8 lockval = _Q_LOCKED_VAL;
  *
- *	if (likely(lockval == _Q_LOCKED_VAL))
+ *	if (try_cmpxchg(&lock->locked, &lockval, 0))
  *		return;
  *	pv_queued_spin_unlock_slowpath(lock, lockval);
  * }
@@ -43,7 +43,6 @@ __PV_CALLEE_SAVE_REGS_THUNK(__pv_queued_spin_unlock_slowpath, ".spinlock.text");
 	"mov   $0x1,%eax\n\t"						\
 	"xor   %edx,%edx\n\t"						\
 	LOCK_PREFIX "cmpxchg %dl,(%rdi)\n\t"				\
-	"cmp   $0x1,%al\n\t"						\
 	"jne   .slowpath\n\t"						\
 	"pop   %rdx\n\t"						\
 	FRAME_END							\
Signature · 1 line
2.44.0
merge bot Applied
#1 lore raw
The following commit has been merged into the locking/core branch of tip:

The following commit has been merged into the locking/core branch of tip:

Commit-ID:     8793f85688fee3c6640f1cc3c9feab54c0910fc1
Gitweb:        https://git.kernel.org/tip/8793f85688fee3c6640f1cc3c9feab54c0910fc1
Author:        Uros Bizjak <[email protected]>
AuthorDate:    Fri, 12 Apr 2024 10:38:53 +02:00
Committer:     Ingo Molnar <[email protected]>
CommitterDate: Fri, 12 Apr 2024 10:56:45 +02:00

locking/pvqspinlock/x86: Remove redundant CMP after CMPXCHG in __raw_callee_save___pv_queued_spin_unlock()

x86 CMPXCHG instruction returns success in the ZF flag. Remove redundant CMP instruction after CMPXCHG that performs the same check.

Also update the function comment to mention the modern version of the equivalent C code.

---
 arch/x86/include/asm/qspinlock_paravirt.h | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)
arch/x86/include/asm/qspinlock_paravirt.h +2 -3
index ef9697f..466af57 100644
@@ -25,9 +25,9 @@ __PV_CALLEE_SAVE_REGS_THUNK(__pv_queued_spin_unlock_slowpath, ".spinlock.text");
  *
  * void __lockfunc __pv_queued_spin_unlock(struct qspinlock *lock)
  * {
- *	u8 lockval = cmpxchg(&lock->locked, _Q_LOCKED_VAL, 0);
+ *	u8 lockval = _Q_LOCKED_VAL;
  *
- *	if (likely(lockval == _Q_LOCKED_VAL))
+ *	if (try_cmpxchg(&lock->locked, &lockval, 0))
  *		return;
  *	pv_queued_spin_unlock_slowpath(lock, lockval);
  * }
@@ -43,7 +43,6 @@ __PV_CALLEE_SAVE_REGS_THUNK(__pv_queued_spin_unlock_slowpath, ".spinlock.text");
 	"mov   $0x1,%eax\n\t"						\
 	"xor   %edx,%edx\n\t"						\
 	LOCK_PREFIX "cmpxchg %dl,(%rdi)\n\t"				\
-	"cmp   $0x1,%al\n\t"						\
 	"jne   .slowpath\n\t"						\
 	"pop   %rdx\n\t"						\
 	FRAME_END							\
merge bot Applied
#2 lore raw
In reply to Uros Bizjak
The following commit has been merged into the locking/core branch of tip:

The following commit has been merged into the locking/core branch of tip:

Commit-ID:     91095666125a666c8f20c2323b742c53165c0325
Gitweb:        https://git.kernel.org/tip/91095666125a666c8f20c2323b742c53165c0325
Author:        Uros Bizjak <[email protected]>
AuthorDate:    Fri, 12 Apr 2024 10:38:53 +02:00
Committer:     Ingo Molnar <[email protected]>
CommitterDate: Fri, 12 Apr 2024 11:42:39 +02:00

locking/pvqspinlock/x86: Remove redundant CMP after CMPXCHG in __raw_callee_save___pv_queued_spin_unlock()

x86 CMPXCHG instruction returns success in the ZF flag. Remove redundant CMP instruction after CMPXCHG that performs the same check.

Also update the function comment to mention the modern version of the equivalent C code.

---
 arch/x86/include/asm/qspinlock_paravirt.h | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)
arch/x86/include/asm/qspinlock_paravirt.h +2 -3
index ef9697f..466af57 100644
@@ -25,9 +25,9 @@ __PV_CALLEE_SAVE_REGS_THUNK(__pv_queued_spin_unlock_slowpath, ".spinlock.text");
  *
  * void __lockfunc __pv_queued_spin_unlock(struct qspinlock *lock)
  * {
- *	u8 lockval = cmpxchg(&lock->locked, _Q_LOCKED_VAL, 0);
+ *	u8 lockval = _Q_LOCKED_VAL;
  *
- *	if (likely(lockval == _Q_LOCKED_VAL))
+ *	if (try_cmpxchg(&lock->locked, &lockval, 0))
  *		return;
  *	pv_queued_spin_unlock_slowpath(lock, lockval);
  * }
@@ -43,7 +43,6 @@ __PV_CALLEE_SAVE_REGS_THUNK(__pv_queued_spin_unlock_slowpath, ".spinlock.text");
 	"mov   $0x1,%eax\n\t"						\
 	"xor   %edx,%edx\n\t"						\
 	LOCK_PREFIX "cmpxchg %dl,(%rdi)\n\t"				\
-	"cmp   $0x1,%al\n\t"						\
 	"jne   .slowpath\n\t"						\
 	"pop   %rdx\n\t"						\
 	FRAME_END							\