Re: regression/bisected/6.9 commit 587d67fd929ad89801bcc429675bda90d53f6592 decrease 30% of gaming performance

[Takashi Iwai]

On Sun, 31 Mar 2024 22:21:20 +0200,
Mikhail Gavrilov wrote:
> 
> Hi,
> 
> I spotted that FPS was around 122 [1] in the Shadow of the Tomb Raider
> benchmark at commit f6cef5f8c37f but after moving to commit
> 4ae3dc83b047 it decreased to 84 [2].
> 
> I bisected it and the first bad commit was 587d67fd929a.
> Author: Takashi Iwai <tiwai@xxxxxxx>
> Date:   Fri Mar 15 11:14:42 2024 +0100
> 
>     ALSA: timer: Fix missing irq-disable at closing
> 
>     The conversion to guard macro dropped the irq-disablement at closing
>     mistakenly, which may lead to a race.  Fix it.
> 
>     Fixes: beb45974dd49 ("ALSA: timer: Use guard() for locking")
>     Reported-by: syzbot+28c1a5a5b041a754b947@xxxxxxxxxxxxxxxxxxxxxxxxx
>     Closes: http://lore.kernel.org/r/0000000000000b9a510613b0145f@xxxxxxxxxx
>     Message-ID: <20240315101447.18395-1-tiwai@xxxxxxx>
>     Signed-off-by: Takashi Iwai <tiwai@xxxxxxx>
> 
>  sound/core/timer.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> To make sure of this I builded kernel at commit 712e14250dd2 with
> revert 587d67fd929a and the benchmarking result was returned to best
> observed by me.
> 
> [1] good results - https://postimg.cc/G4NPHMyk
> [2] bad results - https://postimg.cc/1n0D7sYH
> 
> I attached here my build .config and kernel log.
> Is it possible to find a better approach than 587d67fd929a?

Is it a regression against the performance on 6.8?

If so, what happens when you revert both this (587d67fd929a) and
beb45974dd49?  That should make the timer code back to 6.8 code.


Takashi