[alsa-devel] Qustions on dmix - aplay - hw_ptr - appl_ptr

[Takashi Iwai]

At Tue, 17 Apr 2007 10:12:07 -0700,
Matthew Foster wrote:
> 
> I am a driver developer that is new to ALSA. I have read the tutorial as an
> example on how to write an ALSA driver. I appreciate the documentation you
> have provided online. I was wondering if you might have time to possibly
> answer a question. I am having trouble understanding things in the
> memory/buffer management area. 
> 
>  I am not writing a driver for a PCI device, I am writing a driver that acts
> as an interface from the ALSA to a render device. So I essentially have a
> circular buffer I am transferring the ALSA data to.  I set up the ALSA
> buffer like this:
> 
> snd_pcm_lib_preallocate_pages_for_all(pcm, SNDRV_DMA_TYPE_CONTINUOUS,
> snd_dma_continuous_data(GFP_KERNEL),
> ALSA_BUF_SIZE, ALSA_BUF_SIZE)) < 0);
> 
> And later on in my driver after open is called I have a buffer polling loop
> that does the period notification and constantly monitors my render (ouput)
> buffer. I have my asound.conf setup like this:
> 
>     pcm.mix {
>         type dmix
>         ipc_key 1024
>         slave {
>         pcm "hw:0,0"
>             rate 48000
>         }
>     }
> 
>     pcm.!default {
>         type plug
>         slave.pcm "mix"
>     }
> 
> This is to enable the dmix plugin so I can run aplay more than once and have
> the two streams mixed in.  
> 
> The problem I am having is that it seems that when I run aplay like this:
> "aplay close.wav" I only see this pointer moving:
> 
> substream->runtime->status->hw_ptrbir
> 
> And if I run aplay like this: "aplay -D plughw:0,0 close.wav" I only see
> this pointer moving:
> 
> substream->runtime->control->appl_ptr
> 
> This seems odd to me and currently in my code I just check which pointer
> moves and use that to calculate how much data I need to write to my render
> device. This is so I can support both use cases.

The behavior of dmix is somehow special.  It mmaps a buffer to
multiple instances and runs continuously without XRUN check as long as
clients are connected.  Dmix assumes the driver running a la DMA.
That is, hw_ptr can be constantly updated in a ring buffer.  Since it
disables the XRUN check by adjusting stop_threshold, appl_ptr is
meaningless here. 

Thus, for supporting dmix on your device, you may need to ignore
appl_ptr and simply transfer the period-size chunk at each time.

Currently, the support of non-DMA style hardwares is poor on ALSA,
which I'd like to improve in near future.  At least, we should provide
a proper framework.

> Another major problem is that when I play a short wav file 25KB or so and
> use the dmix "aplay close.wav" it randomly doesn't play at sometimes. Also
> if I run "aplay close.wav" and then very quickly run "aplay close.wav" again
> at the same prompt I get corruption of the data which produces screeching.
> In the case where I run aplay one right after the other I notice that the
> hw_ptr moves differently (less) than if I run it, wait, run again. 

Possibly depending on the behavior above?

> I am also noticing also that when I do the aplay over and over quickly that
> these values change (get lower) in my snd_pcm_prepare callback:
> 
>   chip_data->buf_size        = snd_pcm_lib_buffer_bytes(substream);
> 
>   chip_data->period_buf_size = snd_pcm_lib_period_bytes(substream);
> 
> It seems like those value should remain constant, and I use those values to
> calculate later in my polling thread to notify of a period lapse. It's
> almost like the runtime is telling me the period has changed and the buffer
> size has changed. But this only happens in this case with a short wav file
> and running aplay over and over quickly. 

This sounds odd.  Do you call snd_pcm_lib_malloc() in hw_params
callback? 


Takashi