Question on wait_for_avail function

Question on wait_for_avail function

[gsantosh]

Hi Liam,

Consider an use case
1) Write is blocked as there is no space to write the data,
after the return from snd_pcm_playback_avail()function in wait_for_avail
2) HAL issues pause in parallel to the write
I see that the following function snd_pcm_post_pause unblock the write
		wake_up(&runtime->sleep);
		wake_up(&runtime->tsleep);
But i do not see any case handling for the pause state in the
wait_for_avail function, it is not coming out of the for loop as there
will not be any change in the buffer states.

Please explain what is the intention behind the wake in pause callback.

Want to understand why there is no handling of pause in wait_for_avail
function.

and also is there any restriction from ALSA framework that we should not
issue pause while write is blocked or going on.

Regards,
Santosh M G.

Re: Question on wait_for_avail function

[Clemens Ladisch]

gsantosh@codeaurora.org wrote:
> 2) HAL issues pause in parallel to the write

This is illegal.
<http://www.alsa-project.org/main/index.php/SMP_Design> says:
| The use of returned handles [such as snd_pcm_t*] must be serialized in
| the application using own locking scheme.

If you want to issue a pause while waiting for data to be written, use
non-blocking writes.


Regards,
Clemens

Re: Question on wait_for_avail function

[Patrick Lai]

+ << Jaroslav >>
On 6/5/2013 2:17 AM, Clemens Ladisch wrote:
> gsantosh@codeaurora.org wrote:
>> 2) HAL issues pause in parallel to the write
>
> This is illegal.
> <http://www.alsa-project.org/main/index.php/SMP_Design> says:
> | The use of returned handles [such as snd_pcm_t*] must be serialized in
> | the application using own locking scheme.
>
> If you want to issue a pause while waiting for data to be written, use
> non-blocking writes.
>

Based on my understanding of SMP_Design and response above, pause and 
write to same
PCM device are expected to be serialized. However, my interpretation of
following patch is that pause and stop are expected to wakes up process
waiting in pcm_lib.c wait_for_avail(). It seems to give me the
indication that multiple threads acting on same PCM device is expected.

ALSA: pcm_core: Fix wake_up() optimization

This change fixes the "ALSA: pcm_lib - optimize wake_up() calls for PCM
I/O" commit. New sleeping queue is introduced to separate user space
and kernel space wake_ups. runtime->nowake is renamed to twake
(transfer wake).

Jaroslav: Can you share your thought as you are author of this patch?

Thanks
Patrick
>
> Regards,
> Clemens
> _______________________________________________
> Alsa-devel mailing list
> Alsa-devel@alsa-project.org
> http://mailman.alsa-project.org/mailman/listinfo/alsa-devel
>


-- 
Sent by an employee of the Qualcomm Innovation Center, Inc.
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum.

Re: Question on wait_for_avail function

[Takashi Iwai]

At Wed, 05 Jun 2013 23:25:15 -0700,
Patrick Lai wrote:
> 
> + << Jaroslav >>
> On 6/5/2013 2:17 AM, Clemens Ladisch wrote:
> > gsantosh@codeaurora.org wrote:
> >> 2) HAL issues pause in parallel to the write
> >
> > This is illegal.
> > <http://www.alsa-project.org/main/index.php/SMP_Design> says:
> > | The use of returned handles [such as snd_pcm_t*] must be serialized in
> > | the application using own locking scheme.
> >
> > If you want to issue a pause while waiting for data to be written, use
> > non-blocking writes.
> >
> 
> Based on my understanding of SMP_Design and response above, pause and 
> write to same
> PCM device are expected to be serialized. However, my interpretation of
> following patch is that pause and stop are expected to wakes up process
> waiting in pcm_lib.c wait_for_avail(). It seems to give me the
> indication that multiple threads acting on same PCM device is expected.
> 
> ALSA: pcm_core: Fix wake_up() optimization
> 
> This change fixes the "ALSA: pcm_lib - optimize wake_up() calls for PCM
> I/O" commit. New sleeping queue is introduced to separate user space
> and kernel space wake_ups. runtime->nowake is renamed to twake
> (transfer wake).
> 
> Jaroslav: Can you share your thought as you are author of this patch?

I don't understand what's the problem in this thread...

The behavior of PAUSED stream in wait_for_avail() is clear.  If it's
woken up by some events, it simply goes to sleep again without exiting
the loop.


Takashi

Re: Question on wait_for_avail function

[gsantosh]

> At Wed, 05 Jun 2013 23:25:15 -0700,
> Patrick Lai wrote:
>>
>> + << Jaroslav >>
>> On 6/5/2013 2:17 AM, Clemens Ladisch wrote:
>> > gsantosh@codeaurora.org wrote:
>> >> 2) HAL issues pause in parallel to the write
>> >
>> > This is illegal.
>> > <http://www.alsa-project.org/main/index.php/SMP_Design> says:
>> > | The use of returned handles [such as snd_pcm_t*] must be serialized
>> in
>> > | the application using own locking scheme.
>> >
>> > If you want to issue a pause while waiting for data to be written, use
>> > non-blocking writes.
>> >
>>
>> Based on my understanding of SMP_Design and response above, pause and
>> write to same
>> PCM device are expected to be serialized. However, my interpretation of
>> following patch is that pause and stop are expected to wakes up process
>> waiting in pcm_lib.c wait_for_avail(). It seems to give me the
>> indication that multiple threads acting on same PCM device is expected.
>>
>> ALSA: pcm_core: Fix wake_up() optimization
>>
>> This change fixes the "ALSA: pcm_lib - optimize wake_up() calls for PCM
>> I/O" commit. New sleeping queue is introduced to separate user space
>> and kernel space wake_ups. runtime->nowake is renamed to twake
>> (transfer wake).
>>
>> Jaroslav: Can you share your thought as you are author of this patch?
>
> I don't understand what's the problem in this thread...
>
> The behavior of PAUSED stream in wait_for_avail() is clear.  If it's
> woken up by some events, it simply goes to sleep again without exiting
> the loop.

We are facing the following issue and want to understand how ALSA is
implemented pause, prepare and write if used asynchronously.

for the every seek operation we issue pause and prepare, till now in all
the system the issue of (pause + prepare) commands and write are
serialized.
we are trying to decouple the write and other operation (pause and
prepare) to issue in commands asynchronously.

Assume the case below, write is blocked in wait_for_avail as there is no
space for the buffer, in between user issued seek on the same session this
triggers pause and prepare operations.

pause command will unblock the thread blocked in wait_for_avail, this
unblocked thread once again check if buffer avail and if buffer is not
avail it goes for wait in loop.
after the pause operation is done, HAL issues prepare in prepare we reset
the buffer but never unblock the thread waiting in the wait_for_avail.
this results in write thread blocked forever and it never gets unblocked
as a result the session is blocked.

please help us to know why the implementation of the pause and prepare is
this way and are we allowed to do the pause, prepare and write in
asynchronously on the same session in copy interface.




>
>
> Takashi
> _______________________________________________
> Alsa-devel mailing list
> Alsa-devel@alsa-project.org
> http://mailman.alsa-project.org/mailman/listinfo/alsa-devel
>

Re: Question on wait_for_avail function

[Takashi Iwai]

At Thu, 6 Jun 2013 18:29:35 -0000,
gsantosh@codeaurora.org wrote:
> 
> > At Wed, 05 Jun 2013 23:25:15 -0700,
> > Patrick Lai wrote:
> >>
> >> + << Jaroslav >>
> >> On 6/5/2013 2:17 AM, Clemens Ladisch wrote:
> >> > gsantosh@codeaurora.org wrote:
> >> >> 2) HAL issues pause in parallel to the write
> >> >
> >> > This is illegal.
> >> > <http://www.alsa-project.org/main/index.php/SMP_Design> says:
> >> > | The use of returned handles [such as snd_pcm_t*] must be serialized
> >> in
> >> > | the application using own locking scheme.
> >> >
> >> > If you want to issue a pause while waiting for data to be written, use
> >> > non-blocking writes.
> >> >
> >>
> >> Based on my understanding of SMP_Design and response above, pause and
> >> write to same
> >> PCM device are expected to be serialized. However, my interpretation of
> >> following patch is that pause and stop are expected to wakes up process
> >> waiting in pcm_lib.c wait_for_avail(). It seems to give me the
> >> indication that multiple threads acting on same PCM device is expected.
> >>
> >> ALSA: pcm_core: Fix wake_up() optimization
> >>
> >> This change fixes the "ALSA: pcm_lib - optimize wake_up() calls for PCM
> >> I/O" commit. New sleeping queue is introduced to separate user space
> >> and kernel space wake_ups. runtime->nowake is renamed to twake
> >> (transfer wake).
> >>
> >> Jaroslav: Can you share your thought as you are author of this patch?
> >
> > I don't understand what's the problem in this thread...
> >
> > The behavior of PAUSED stream in wait_for_avail() is clear.  If it's
> > woken up by some events, it simply goes to sleep again without exiting
> > the loop.
> 
> We are facing the following issue and want to understand how ALSA is
> implemented pause, prepare and write if used asynchronously.
> 
> for the every seek operation we issue pause and prepare, till now in all
> the system the issue of (pause + prepare) commands and write are
> serialized.
> we are trying to decouple the write and other operation (pause and
> prepare) to issue in commands asynchronously.
> 
> Assume the case below, write is blocked in wait_for_avail as there is no
> space for the buffer, in between user issued seek on the same session this
> triggers pause and prepare operations.
> 
> pause command will unblock the thread blocked in wait_for_avail, this
> unblocked thread once again check if buffer avail and if buffer is not
> avail it goes for wait in loop.
> after the pause operation is done, HAL issues prepare in prepare we reset
> the buffer but never unblock the thread waiting in the wait_for_avail.
> this results in write thread blocked forever and it never gets unblocked
> as a result the session is blocked.

What do you mean "prepare" in the context above?  Moving to
SND_PCM_STATE_PREPARE?  If so, you can't.  For preparing the stream,
you must stop the stream once.  So, you have to issue snd_pcm_drop(),
so that stream goes to SND_PCM_STATE_SETUP, then you can do prepare,
and restart.

In other words, from the PAUSED state, it can change to only either
RUNNING (via PAUSE_RESUME trigger) or SETUP (via snd_pcm_drop) state.
(There are other possible states like DRAINING, SUSPENDED or
 DISCONNECTED, but let's ignore now. )


Takashi

> please help us to know why the implementation of the pause and prepare is
> this way and are we allowed to do the pause, prepare and write in
> asynchronously on the same session in copy interface.
> 
> 
> 
> 
> >
> >
> > Takashi
> > _______________________________________________
> > Alsa-devel mailing list
> > Alsa-devel@alsa-project.org
> > http://mailman.alsa-project.org/mailman/listinfo/alsa-devel
> >
> 
> 
> _______________________________________________
> Alsa-devel mailing list
> Alsa-devel@alsa-project.org
> http://mailman.alsa-project.org/mailman/listinfo/alsa-devel
>