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

[Takashi Iwai]

On Wed, 24 Apr 2024 15:33:10 +0200,
Takashi Iwai wrote:
> 
> On Wed, 24 Apr 2024 15:21:26 +0200,
> Mikhail Gavrilov wrote:
> > 
> > On Thu, Apr 18, 2024 at 11:22 AM Mikhail Gavrilov
> > <mikhail.v.gavrilov@xxxxxxxxx> wrote:
> > >
> > > On Tue, Apr 2, 2024 at 12:29 AM <mikhail.v.gavrilov@xxxxxxxxx> wrote:
> > > >
> > > > On Mon, 2024-04-01 at 21:50 +0500, Mikhail Gavrilov wrote:
> > > > > On Mon, Apr 1, 2024 at 7:48 PM Takashi Iwai <tiwai@xxxxxxx> wrote:
> > > > > >
> > > > > >
> > > > > > Then it shouldn't be dmix/dsnoop.  You'd better try profiling
> > > > > > what's
> > > > > > going on there.  e.g. try to get perf results with and without the
> > > > > > commit.
> > > > >
> > > > >
> > > > > Here is original perf files:
> > > > > [1] perf-with-revert-587d67fd929a.data -
> > > > > https://mega.nz/file/EpZmXbgL#vEnxuODoB__jM6TwV6XbwC_TWMMqvT7qaWkJU22JY48
> > > > > [2] perf-with-revert-587d67fd929a-beb45974dd49.data -
> > > > > https://mega.nz/file/0lI31CDA#hKsv6vX7t5u-Sx1_p2E7-Y32-z5VFPlW2Y793oM-JRU
> > > > >
> > >
> > > Excuse me. Can you have time to look in my perf files?
> > >
> > 
> > This is a friendly reminder
> 
> Honestly speaking, it's your role to analyze what's going on.
> Such a performance issue is pretty much dependent on the installed
> system, so giving a perf data to me won't help much, unfortunately.

That said, maybe the first thing you can try would be to check who is
actually calling the corresponding function
(snd_timer_close_locked()).  Put a debug print or a tracing hook to
watch out for figuring out.  If the commit was really relevant, it
must be called very frequently and concurrently, and I don't know
really who does it except for dmix/dsnoop.


Takashi