* Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
> On Thu, 11 Apr 2024 at 06:33, tip-bot2 for Uros Bizjak
> <tip-bot2@xxxxxxxxxxxxx> wrote:
> >
> > Use try_cmpxchg_acquire(*ptr, &old, new) instead of
> > cmpxchg_relaxed(*ptr, old, new) == old in trylock_clear_pending().
>
> The above commit message is horribly confusing and wrong.
>
> I was going "that's not right", because it says "use acquire instead
> of relaxed" memory ordering, and then goes on to say "No functional
> change intended".
>
> But it turns out the *code* was always acquire, and it's only the
> commit message that is wrong, presumably due to a bit too much
> cut-and-paste.
Yeah, the replacement is cmpxchg_acquire() => try_cmpxchg_acquire(), with
no change to memory ordering.
> But please fix the commit message, and use the right memory ordering
> in the explanations too.
Done, find below the new patch, which a hopefully better commit message.
I also added your Reviewed-by tag optimistically :)
Thanks,
Ingo
===========================>
From: Uros Bizjak <ubizjak@xxxxxxxxx>
Date: Mon, 25 Mar 2024 15:09:32 +0100
Subject: [PATCH] locking/pvqspinlock: Use try_cmpxchg_acquire() in trylock_clear_pending()
Replace this pattern in trylock_clear_pending():
cmpxchg_acquire(*ptr, old, new) == old
.. with the simpler and faster:
try_cmpxchg_acquire(*ptr, &old, new)
The x86 CMPXCHG instruction returns success in the ZF flag, so this change
saves a compare after the CMPXCHG.
Also change the return type of the function to bool and streamline
the control flow in the _Q_PENDING_BITS == 8 variant a bit.
No functional change intended.
Signed-off-by: Uros Bizjak <ubizjak@xxxxxxxxx>
Signed-off-by: Ingo Molnar <mingo@xxxxxxxxxx>
Reviewed-by: Waiman Long <longman@xxxxxxxxxx>
Reviewed-by: Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx>
Link: https://lore.kernel.org/r/20240325140943.815051-1-ubizjak@xxxxxxxxx
---
kernel/locking/qspinlock_paravirt.h | 31 +++++++++++++------------------
1 file changed, 13 insertions(+), 18 deletions(-)
diff --git a/kernel/locking/qspinlock_paravirt.h b/kernel/locking/qspinlock_paravirt.h
index 169950fe1aad..77ba80bd95f9 100644
--- a/kernel/locking/qspinlock_paravirt.h
+++ b/kernel/locking/qspinlock_paravirt.h
@@ -116,11 +116,12 @@ static __always_inline void set_pending(struct qspinlock *lock)
* barrier. Therefore, an atomic cmpxchg_acquire() is used to acquire the
* lock just to be sure that it will get it.
*/
-static __always_inline int trylock_clear_pending(struct qspinlock *lock)
+static __always_inline bool trylock_clear_pending(struct qspinlock *lock)
{
+ u16 old = _Q_PENDING_VAL;
+
return !READ_ONCE(lock->locked) &&
- (cmpxchg_acquire(&lock->locked_pending, _Q_PENDING_VAL,
- _Q_LOCKED_VAL) == _Q_PENDING_VAL);
+ try_cmpxchg_acquire(&lock->locked_pending, &old, _Q_LOCKED_VAL);
}
#else /* _Q_PENDING_BITS == 8 */
static __always_inline void set_pending(struct qspinlock *lock)
@@ -128,27 +129,21 @@ static __always_inline void set_pending(struct qspinlock *lock)
atomic_or(_Q_PENDING_VAL, &lock->val);
}
-static __always_inline int trylock_clear_pending(struct qspinlock *lock)
+static __always_inline bool trylock_clear_pending(struct qspinlock *lock)
{
- int val = atomic_read(&lock->val);
-
- for (;;) {
- int old, new;
-
- if (val & _Q_LOCKED_MASK)
- break;
+ int old, new;
+ old = atomic_read(&lock->val);
+ do {
+ if (old & _Q_LOCKED_MASK)
+ return false;
/*
* Try to clear pending bit & set locked bit
*/
- old = val;
- new = (val & ~_Q_PENDING_MASK) | _Q_LOCKED_VAL;
- val = atomic_cmpxchg_acquire(&lock->val, old, new);
+ new = (old & ~_Q_PENDING_MASK) | _Q_LOCKED_VAL;
+ } while (!atomic_try_cmpxchg_acquire (&lock->val, &old, new));
- if (val == old)
- return 1;
- }
- return 0;
+ return true;
}
#endif /* _Q_PENDING_BITS == 8 */
|