Underrun problem and some basic queries

Underrun problem and some basic queries

[ANISH KUMAR]

Hi Takashi et al,
 
This post of mine is just a reminder about my earlier unanswered questions
which i raised about underrun.
http://mailman.alsa-project.org/pipermail/alsa-devel/2010-January/024286.html

I guess you were on holiday and i think this is a bit long list of questions which
no body cared to answer.This time please answer them as this would help all the
new entrants to understand ALSA.

Best Regards,
anish

Underrun problem and some basic queries

[ANISH KUMAR]

Dear ALSA Team,
I have some queries.Please answer them.(previous post please ignore as they couldn't appear because of encoding problem)
I am getting underrun problem whenever the playback is about to get finished.
happening in snd_pcm_update_hw_ptr_post function:
if (avail >= stop_threshold) {
        if (substream->runtime->status->state == SNDRV_PCM_STATE_DRAINING)
            snd_pcm_drain_done(substream);
        else
            xrun(substream); //here it is happening..........
        return -EPIPE;
}
1. As i understand if i dont call drain function this will happen.Ismy understanding correct?Because if am calling drain function then thisproblem is not happening.If the application doesn't call drain functionthen xrun is the only way to stop ALSA or does ALSA detect that thereare no more samples to play and automatically stop the play(if yes,canyou point out the source code)?
2. snd_pcm_playback_avail() returns the number of samples that stillneeds to be played?Below is the diagram which is my understanding.
        
    Ring buffer (DMA buffer created using preallocated buffer during system initalisation)
-----------------------------------------------------------------
|          |          |          |          |          |          |
-----------------------------------------------------------------
    ^  ^            
    |   |
hw_ptr appl_ptr
After first write call:   
-----------------------------------------------------------------
|          |          |          |          |          |          |
-----------------------------------------------------------------
     ^                               ^            
     |                                |
 appl_ptr                     hw_ptr 
So my understanding is once wait_for_avail_min returns insnd_pcm_lib_write1 function.hw_ptr is increased by some amount(periodsize) but appl_ptr will increase only when wait_for_avail_minreturns.And difference of (hw_ptr+runtime->buffersize-appl_ptr) will give us unplayed samples in the case of playback?
Is this whole ring buffer is filled after the snd_pcm_lib_write_transfer function is returned?
3.In some rare cases (playback case) i have seen thatwait_for_avail_min doesn't return and driver pointer function keeps ongetting called causing hw_ptr to increase.This causessnd_pcm_playback_avail() function to return a value less than avail_min(as hw_ptr is increasing and appl_ptr is stagnant as it is stuck insnd_pcm_lib_write1 function) which inturn causes wake-up fromsnd_pcm_update_hw_ptr_post() not to get called.When this type ofscenario happens?Is it a normal scenario if yes,then how will it willever come out from wait_for_avail_min?
According to my understanding appl_ptr will increase only when we return from wait_for_avail_min?
4.how does alsa know that it is running short of data?As iunderstand that it does not free the memory and data is onlyoverwritten.So how does it distinguish between old and new data?
5.Burst of noise due to stale data problem reported by Jon Smirl and for which Mr Jaroslav has suggested this solution http://mailman.alsa-project.org/pipermail/alsa-devel/2009-May/017119.html.Is it implemented?
 
Best Regards..

Re: Underrun problem and some basic queries

[Jaroslav Kysela]

On Mon, 4 Jan 2010, ANISH KUMAR wrote:

>
> Dear ALSA Team,
> I have some queries.Please answer them.(previous post please ignore as they couldn't appear because of encoding problem)
> I am getting underrun problem whenever the playback is about to get finished.
> happening in snd_pcm_update_hw_ptr_post function:
> if (avail >= stop_threshold) {
>        if (substream->runtime->status->state == SNDRV_PCM_STATE_DRAINING)
>            snd_pcm_drain_done(substream);
>        else
>            xrun(substream); //here it is happening..........
>        return -EPIPE; }
>
> 1. As i understand if i dont call drain function 
> this will happen.Ismy understanding correct?Because if am calling drain 
> function then thisproblem is not happening.If the application doesn't 
> call drain functionthen xrun is the only way to stop ALSA or does ALSA 
> detect that thereare no more samples to play and automatically stop the 
> play(if yes,canyou point out the source code)?

You showed the stop point in above if (avail >= threshold) condition.

> 2. snd_pcm_playback_avail() returns the number of samples that 
> stillneeds to be played?Below is the diagram which is my understanding.

No, it's the room for new samples to fill the ring buffer to full.

>    Ring buffer (DMA buffer created using preallocated buffer during system initalisation)
> -----------------------------------------------------------------
> |          |          |          |          |          |          |
> -----------------------------------------------------------------
>    ^  ^
>    |   |
> hw_ptr appl_ptr
> After first write call:
> -----------------------------------------------------------------
> |          |          |          |          |          |          |
> -----------------------------------------------------------------
>     ^                               ^
>     |                                |
> appl_ptr                     hw_ptr

If hw_ptr > appl_ptr then underrun occurs.

> So my understanding is once wait_for_avail_min returns 
> insnd_pcm_lib_write1 function.hw_ptr is increased by some 
> amount(periodsize) but appl_ptr will increase only when 
> wait_for_avail_minreturns.And difference of 
> (hw_ptr+runtime->buffersize-appl_ptr) will give us unplayed samples in 
> the case of playback?

No.

> Is this whole ring buffer is filled after the 
> snd_pcm_lib_write_transfer function is returned?

Not necessary. It depends how much samples an application feeds in.

> 3.In some rare cases (playback case) i have seen thatwait_for_avail_min 
> doesn't return and driver pointer function keeps ongetting called 
> causing hw_ptr to increase.This causessnd_pcm_playback_avail() function 
> to return a value less than avail_min(as hw_ptr is increasing and 
> appl_ptr is stagnant as it is stuck insnd_pcm_lib_write1 function) which 
> inturn causes wake-up fromsnd_pcm_update_hw_ptr_post() not to get 
> called.When this type ofscenario happens?Is it a normal scenario if 
> yes,then how will it willever come out from wait_for_avail_min? 
> According to my understanding appl_ptr will increase only when we return 
> from wait_for_avail_min?

I'm not sure if you're not mentioning a problem solved in:

http://git.alsa-project.org/?p=alsa-kernel.git;a=commitdiff;h=c91a988dc6551c66418690e36b2a23cdb0255da8

> 4.how does alsa know that it is running short 
> of data?As iunderstand that it does not free the memory and data is 
> onlyoverwritten.So how does it distinguish between old and new data?

Everything in ALSA PCM is about comparing hw_ptr and appl_ptr. So 
basically, the ring buffer size is maximum time for application to fill 
and for driver to eat samples. If this time is surpased, wrong things 
happen.

> 5.Burst of noise due to stale data problem reported by Jon Smirl and for 
> which Mr Jaroslav has suggested this solution 
> http://mailman.alsa-project.org/pipermail/alsa-devel/2009-May/017119.html.Is 
> it implemented?

Nope. Patches are welcome.

 						Jaroslav

-----
Jaroslav Kysela <perex@perex.cz>
Linux Kernel Sound Maintainer
ALSA Project, Red Hat, Inc.