On Tue, 15 Mar 2022 14:15:20 +0100,
Arnaud POULIQUEN wrote:
>
> Hello,
>
>
> ST Restricted
>
> > -----Original Message-----
> > From: Takashi Iwai <[email protected]>
> > Sent: mardi 15 mars 2022 11:28
> > To: Daniel Palmer <[email protected]>
> > Cc: [email protected]; [email protected]; Arnaud POULIQUEN
> > <[email protected]>; [email protected]; linux-
> > [email protected]
> > Subject: Re: [PATCH] ASoC: sti: sti_uniperif: Remove driver
> >
> > On Tue, 15 Mar 2022 10:13:19 +0100,
> > Daniel Palmer wrote:
> > >
> > > This driver seems to be in the "only good for attracting bot generated
> > > patches" phase of it's life.
> > >
> > > It doesn't seem like anyone actually tested the patches that have
> > > been applied in the last few years as uni_reader_irq_handler()
> > > had a dead lock added to it (it locks the stream, then calls
> > > snd_pcm_stop_xrun() which will also lock the stream).
> >
> > Mea culpa, that was an obvious deadlock I overlooked in the patch
> > series.
> >
> > > Seems best just to remove it.
> > >
> > > Signed-off-by: Daniel Palmer <[email protected]>
> > > ---
> > > I've never used this driver, don't have the hardware etc.
> > > I just noticed that this looks broken when debugging my
> > > own driver that uses snd_pcm_stop_xrun() and was looking
> > > at other users to see if I was using it wrong and noticed
> > > this was the only place that locked the stream before
> > > calling snd_pcm_stop_xrun().
> > >
> > > There are probably some other bits of the driver that
> > > should be removed but I didn't look that hard.
> > >
> > > TL;DR; This driver seems broken, seems like nobody uses
> > > it. Maybe it should be deleted?
> >
> > Yeah, that looks dead.
> >
>
> The platform is still used for instance:
> https://lore.kernel.org/all/[email protected]/
>
> So please do not remove the driver
Ah, it's always good to see a vital sign!
> The issue has not been detected because it is related to an error that
> occurs only when we reach the limit of the platform, with application
> that stop the stream at same time.
> So almost no chance to occur.
>
> > OTOH, if anyone really wants to keep the stuff, please revert the
> > commit dc865fb9e7c2251c9585ff6a7bf185d499db13e4.
>
> Yes reverting the commit is one solution.
> The other is to clean-up the snd_pcm_stream_lock/ snd_pcm_stream_unlock in the
> Handler.
That would work, but maybe it's safer to keep that lock, as the state
change isn't protected by irq_lock but only implicitly by stream lock
in start/stop callbacks.
thanks,
Takashi
ST Restricted
> -----Original Message-----
> From: Takashi Iwai <[email protected]>
> Sent: mardi 15 mars 2022 15:35
> To: Arnaud POULIQUEN <[email protected]>
> Cc: Daniel Palmer <[email protected]>; [email protected]; [email protected];
> [email protected]; [email protected]
> Subject: Re: [PATCH] ASoC: sti: sti_uniperif: Remove driver
>
> On Tue, 15 Mar 2022 14:15:20 +0100,
> Arnaud POULIQUEN wrote:
> >
> > Hello,
> >
> >
> > ST Restricted
> >
> > > -----Original Message-----
> > > From: Takashi Iwai <[email protected]>
> > > Sent: mardi 15 mars 2022 11:28
> > > To: Daniel Palmer <[email protected]>
> > > Cc: [email protected]; [email protected]; Arnaud POULIQUEN
> > > <[email protected]>; [email protected]; linux-
> > > [email protected]
> > > Subject: Re: [PATCH] ASoC: sti: sti_uniperif: Remove driver
> > >
> > > On Tue, 15 Mar 2022 10:13:19 +0100,
> > > Daniel Palmer wrote:
> > > >
> > > > This driver seems to be in the "only good for attracting bot
> > > > generated patches" phase of it's life.
> > > >
> > > > It doesn't seem like anyone actually tested the patches that have
> > > > been applied in the last few years as uni_reader_irq_handler() had
> > > > a dead lock added to it (it locks the stream, then calls
> > > > snd_pcm_stop_xrun() which will also lock the stream).
> > >
> > > Mea culpa, that was an obvious deadlock I overlooked in the patch
> > > series.
> > >
> > > > Seems best just to remove it.
> > > >
> > > > Signed-off-by: Daniel Palmer <[email protected]>
> > > > ---
> > > > I've never used this driver, don't have the hardware etc.
> > > > I just noticed that this looks broken when debugging my own
> > > > driver that uses snd_pcm_stop_xrun() and was looking at other
> > > > users to see if I was using it wrong and noticed this was the
> > > > only place that locked the stream before calling
> > > > snd_pcm_stop_xrun().
> > > >
> > > > There are probably some other bits of the driver that should be
> > > > removed but I didn't look that hard.
> > > >
> > > > TL;DR; This driver seems broken, seems like nobody uses it.
> > > > Maybe it should be deleted?
> > >
> > > Yeah, that looks dead.
> > >
> >
> > The platform is still used for instance:
> > https://lore.kernel.org/all/1d95209f-9cb4-47a3-2696-7a93df7cdc05@foss.
> > st.com/
> >
> > So please do not remove the driver
>
> Ah, it's always good to see a vital sign!
>
> > The issue has not been detected because it is related to an error that
> > occurs only when we reach the limit of the platform, with application
> > that stop the stream at same time.
> > So almost no chance to occur.
> >
> > > OTOH, if anyone really wants to keep the stuff, please revert the
> > > commit dc865fb9e7c2251c9585ff6a7bf185d499db13e4.
> >
> > Yes reverting the commit is one solution.
> > The other is to clean-up the snd_pcm_stream_lock/
> > snd_pcm_stream_unlock in the Handler.
>
> That would work, but maybe it's safer to keep that lock, as the state change
> isn't protected by irq_lock but only implicitly by stream lock in start/stop
> callbacks.
You are right, trying to use the snd_pcm_stop_xrun needs deeper update
that could introduce regression.
It seems wiser to revert your commit dc865fb9e7c2 as you propose.
Thanks,
Arnaud
>
>
> thanks,
>
> Takashi