Re: [PATCH v2 1/3] ALSA: pcm: add snd_pcm_period_elapsed() variant without acquiring lock of PCM substream

[Takashi Iwai <tiwai@suse.de>]

On Fri, 11 Jun 2021 09:07:57 +0200,
Takashi Sakamoto wrote:
> 
> On Fri, Jun 11, 2021 at 08:47:59AM +0200, Takashi Iwai wrote:
> > On Fri, 11 Jun 2021 05:38:16 +0200,
> > Takashi Sakamoto wrote:
> > > 
> > > Hi,
> > > 
> > > On Thu, Jun 10, 2021 at 01:03:19PM +0200, Takashi Iwai wrote:
> > > > On Thu, 10 Jun 2021 12:12:43 +0200, Takashi Sakamoto wrote:
> > > > > 
> > > > > On Thu, Jun 10, 2021 at 10:36:57AM +0200, Takashi Iwai wrote:
> > > > > > Again, my *only* point is about the sleep.  You addition was:
> > > > > > 
> > > > > > + * Context: Any context in which lock of PCM substream is already acquired. This function may not
> > > > > > + * sleep.
> > > > > > 
> > > > > > where "This function may not sleep" is stated incorrectly.
> > > > > 
> > > > > Hm. Would I request you to show the detail case that the call of function
> > > > > (snd_pcm_period_elapsed_under_stream_lock()) goes sleep except for
> > > > > driver-side implementation of snd_pcm_ops.{pointer, trigger,
> > > > > get_time_info}? At least, in callgraph I find no function call to
> > > > > yield...
> > > > 
> > > > True.  But the fact that those callbacks may sleep means that the
> > > > function would go sleeping after all.
> > > 
> > > Thanks. After all, our discussion comes from the ambiguity that what
> > > has responsibility at yielding processor under the lock. I think it helpful
> > > to describe devide responsibilities about the yielding. I'm glad for you
> > > to review patch below:
> > 
> > Well, I don't think it's worth to mention "ALSA core may not sleep".
> > It's just casually so for now, but it doesn't mean that this will be
> > guaranteed in future.  After all, this function call may sleep in
> > the nonatomic mode (that's the very reason for that mode!), and the
> > caller has to be prepared for that, no matter whether you do sleep in
> > the callbacks or not.
> 
> I have an opinion that we should guarantee it as long as maintaining
> existent in-kernel drivers, which call it in hw/sw IRQ context. This is
> not the issue 'casually so for now'.

It *is* casually so for now, and I see no big merit for the ALSA core
about such a limitation.  The PCM core might need to introduce another
lock in future for some reason, and that'll be a mutex in nonatomic
mode.  If we guarantee the current behavior, it would become
impossible.  After all, the preempt is still allowed even if there is
no sleeper in snd_pcm_period*() itself.

For atomic mode, it's under the stream spin lock, so it's clearly no
sleep / no preempt.
For non-atomic mode, it's under the stream mutex lock, and that's
all.  There should be no other restriction there.

We don't want to choke ourselves unnecessarily.


thanks,

Takashi

> 
> If you had a plan to rewrite or drop the drivers near future, you could say
> it.
> 
> > > ======== 8< --------
> > > 
> > > >From 98e1b8332a95935ae875c637d3ddc27e68689aa0 Mon Sep 17 00:00:00 2001
> > > From: Takashi Sakamoto <o-takashi@sakamocchi.jp>
> > > Date: Fri, 11 Jun 2021 11:03:46 +0900
> > > Subject: [PATCH] ALSA: pcm: add context section for documentation about
> > >  period-elapsed kernel APIs
> > > 
> > > This commit fulfils documentation of period-elapsed kernel APIs with their
> > > context section.
> > > 
> > > Signed-off-by: Takashi Sakamoto <o-takashi@sakamocchi.jp>
> > > ---
> > >  sound/core/pcm_lib.c | 8 ++++++++
> > >  1 file changed, 8 insertions(+)
> > > 
> > > diff --git a/sound/core/pcm_lib.c b/sound/core/pcm_lib.c
> > > index 7d5883432085..5d28d63a3216 100644
> > > --- a/sound/core/pcm_lib.c
> > > +++ b/sound/core/pcm_lib.c
> > > @@ -1803,6 +1803,10 @@ EXPORT_SYMBOL(snd_pcm_lib_ioctl);
> > >   * - .get_time_info - to retrieve audio time stamp if needed.
> > >   *
> > >   * Even if more than one periods have elapsed since the last call, you have to call this only once.
> > > + *
> > > + * Context: Any context in which lock of PCM substream is already acquired. The function may not
> > > + * sleep by ALSA PCM core. The function may sleep in the above callbacks by driver which should
> > > + * configures PCM device for it (@snd_pcm.nonatomic).
> > >   */
> > >  void snd_pcm_period_elapsed_under_stream_lock(struct snd_pcm_substream *substream)
> > >  {
> > > @@ -1836,6 +1840,10 @@ EXPORT_SYMBOL(snd_pcm_period_elapsed_under_stream_lock);
> > >   * It's typically called by any type of IRQ handler when hardware IRQ occurs to notify event that
> > >   * the batch of audio data frames as the same size as the period of buffer is already processed in
> > >   * audio data transmission.
> > > + *
> > > + * Context: Any context in which lock of PCM substream is not acquired yet. It depends on
> > > + * configuration of PCM device (@snd_pcm.nonatomic) by driver whether the function may or may not
> > > + * sleep by operating lock of PCM substream.
> > >   */
> > >  void snd_pcm_period_elapsed(struct snd_pcm_substream *substream)
> > >  {
> > > -- 
> > > 2.27.0
> > > 
> > > ======== 8< --------
> > > 
> > > Thanks
> > > 
> > > Takashi Sakamoto
> 
> 
> Regards
> 
> Takashi Sakamoto
>