Use atomic_try_cmpxchg_acquire(*ptr, &old, new) instead of
atomic_cmpxchg_acquire(*ptr, old, new) == old in kexec_trylock().
x86 CMPXCHG instruction returns success in ZF flag, so
this change saves a compare after cmpxchg.

Signed-off-by: Uros Bizjak <ubizjak@gmail.com>
Cc: Eric Biederman <ebiederm@xmission.com>
---
 kernel/kexec_internal.h | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/kernel/kexec_internal.h b/kernel/kexec_internal.h
index 2595defe8c0d..d35d9792402d 100644
--- a/kernel/kexec_internal.h
+++ b/kernel/kexec_internal.h
@@ -23,7 +23,8 @@ int kimage_is_destination_range(struct kimage *image,
 extern atomic_t __kexec_lock;
 static inline bool kexec_trylock(void)
 {
-	return atomic_cmpxchg_acquire(&__kexec_lock, 0, 1) == 0;
+	int old = 0;
+	return atomic_try_cmpxchg_acquire(&__kexec_lock, &old, 1);
 }
 static inline void kexec_unlock(void)
 {
-- 
2.42.0

On 07/19/24 at 12:38pm, Uros Bizjak wrote:
> Use atomic_try_cmpxchg_acquire(*ptr, &old, new) instead of
> atomic_cmpxchg_acquire(*ptr, old, new) == old in kexec_trylock().
> x86 CMPXCHG instruction returns success in ZF flag, so
> this change saves a compare after cmpxchg.

Seems it can simplify code even though on non-x86 arch, should we
replace atomic_try_cmpxchg_acquire() with atomic_try_cmpxchg_acquire()
in all similar places?

For this one,

Acked-by: Baoquan He <bhe@redhat.com>

> 
> Signed-off-by: Uros Bizjak <ubizjak@gmail.com>
> Cc: Eric Biederman <ebiederm@xmission.com>
> ---
>  kernel/kexec_internal.h | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/kernel/kexec_internal.h b/kernel/kexec_internal.h
> index 2595defe8c0d..d35d9792402d 100644
> --- a/kernel/kexec_internal.h
> +++ b/kernel/kexec_internal.h
> @@ -23,7 +23,8 @@ int kimage_is_destination_range(struct kimage *image,
>  extern atomic_t __kexec_lock;
>  static inline bool kexec_trylock(void)
>  {
> -	return atomic_cmpxchg_acquire(&__kexec_lock, 0, 1) == 0;
> +	int old = 0;
> +	return atomic_try_cmpxchg_acquire(&__kexec_lock, &old, 1);
>  }
>  static inline void kexec_unlock(void)
>  {
> -- 
> 2.42.0
> 
> 
> _______________________________________________
> kexec mailing list
> kexec@lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/kexec
>

On Mon, Jul 22, 2024 at 5:09 AM Baoquan He <bhe@redhat.com> wrote:
>
> On 07/19/24 at 12:38pm, Uros Bizjak wrote:
> > Use atomic_try_cmpxchg_acquire(*ptr, &old, new) instead of
> > atomic_cmpxchg_acquire(*ptr, old, new) == old in kexec_trylock().
> > x86 CMPXCHG instruction returns success in ZF flag, so
> > this change saves a compare after cmpxchg.
>
> Seems it can simplify code even though on non-x86 arch, should we
> replace atomic_try_cmpxchg_acquire() with atomic_try_cmpxchg_acquire()
> in all similar places?

Yes, the change is beneficial also for non-x86 architectures, please
see analysis at thread [1]. I've been looking through the kernel
sources for these places for quite some time, and I believe I have
changed most of the places. The change is relatively straightforward,
and immediately results in a better code.

[1] https://lore.kernel.org/lkml/871qwgmqws.fsf@mpe.ellerman.id.au/

>
> For this one,
>
> Acked-by: Baoquan He <bhe@redhat.com>

Thanks,
Uros.

On 07/22/24 at 10:53am, Uros Bizjak wrote:
> On Mon, Jul 22, 2024 at 5:09 AM Baoquan He <bhe@redhat.com> wrote:
> >
> > On 07/19/24 at 12:38pm, Uros Bizjak wrote:
> > > Use atomic_try_cmpxchg_acquire(*ptr, &old, new) instead of
> > > atomic_cmpxchg_acquire(*ptr, old, new) == old in kexec_trylock().
> > > x86 CMPXCHG instruction returns success in ZF flag, so
> > > this change saves a compare after cmpxchg.
> >
> > Seems it can simplify code even though on non-x86 arch, should we
> > replace atomic_try_cmpxchg_acquire() with atomic_try_cmpxchg_acquire()
> > in all similar places?
> 
> Yes, the change is beneficial also for non-x86 architectures, please
> see analysis at thread [1]. I've been looking through the kernel
> sources for these places for quite some time, and I believe I have
> changed most of the places. The change is relatively straightforward,
> and immediately results in a better code.
> 
> [1] https://lore.kernel.org/lkml/871qwgmqws.fsf@mpe.ellerman.id.au/

Good to know, thanks for telling.