Re: [PATCH] drm/i915/pmu: Use local64_try_cmpxchg in i915_pmu_event_read

[Jani Nikula]

On Mon, 03 Jul 2023, Uros Bizjak <ubizjak@xxxxxxxxx> wrote:
> Use local64_try_cmpxchg instead of local64_cmpxchg (*ptr, old, new) == old
> in i915_pmu_event_read.  x86 CMPXCHG instruction returns success in ZF flag,
> so this change saves a compare after cmpxchg (and related move instruction
> in front of cmpxchg).
>
> Also, try_cmpxchg implicitly assigns old *ptr value to "old" when cmpxchg
> fails. There is no need to re-read the value in the loop.
>
> No functional change intended.
>
> Cc: Jani Nikula <jani.nikula@xxxxxxxxxxxxxxx>
> Cc: Joonas Lahtinen <joonas.lahtinen@xxxxxxxxxxxxxxx>
> Cc: Rodrigo Vivi <rodrigo.vivi@xxxxxxxxx>
> Cc: Tvrtko Ursulin <tvrtko.ursulin@xxxxxxxxxxxxxxx>
> Cc: David Airlie <airlied@xxxxxxxxx>
> Cc: Daniel Vetter <daniel@xxxxxxxx>
> Signed-off-by: Uros Bizjak <ubizjak@xxxxxxxxx>
> ---
>  drivers/gpu/drm/i915/i915_pmu.c | 9 ++++-----
>  1 file changed, 4 insertions(+), 5 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/i915_pmu.c b/drivers/gpu/drm/i915/i915_pmu.c
> index d35973b41186..108b675088ba 100644
> --- a/drivers/gpu/drm/i915/i915_pmu.c
> +++ b/drivers/gpu/drm/i915/i915_pmu.c
> @@ -696,12 +696,11 @@ static void i915_pmu_event_read(struct perf_event *event)
>  		event->hw.state = PERF_HES_STOPPED;
>  		return;
>  	}
> -again:
> -	prev = local64_read(&hwc->prev_count);
> -	new = __i915_pmu_event_read(event);
>  
> -	if (local64_cmpxchg(&hwc->prev_count, prev, new) != prev)
> -		goto again;
> +	prev = local64_read(&hwc->prev_count);
> +	do {
> +		new = __i915_pmu_event_read(event);
> +	} while (!local64_try_cmpxchg(&hwc->prev_count, &prev, new));

You could save everyone a lot of time by actually documenting what these
functions do. Assume you don't know what local64_try_cmpxchg() does, and
see how many calls you have to go through to figure it out.

Because the next time I encounter this code or a patch like this, I'm
probably going to have to do that again.

To me, the old one was more readable. The optimization is meaningless to
me if it's not quantified but reduces readability.


BR,
Jani.


>  
>  	local64_add(new - prev, &event->count);
>  }

-- 
Jani Nikula, Intel Open Source Graphics Center