[alsa-devel] DMix period_size should always be half of buffer size?
This is a followup to my earlier question about dmix not being able to mix small files. I investigated it more and here is what I found.
Again background info. I have a 64KB as kernel buffer and a 64KB circular buffer for my device. In a poll thread I copy the data obtained from ALSA lib from the kernel buffer to my device circular buffer by keeping track of appl_ptr and hw_ptr(dmix only). Everything was working fine except when I was trying to mix files smaller than 64KB(my buffer size), the small file was not getting mixed through the dmix plugin. Anything over 64KB was getting mixed just fine. So I set about reducing my period size from 32KB to 4KB while maintaining my buffer size at 64KB.
When I tried to do this my hw_ptr never gets incremented from 0. Hence my driver is stuck in a loop waiting for bytes to copy. This is only observed when I am using dmix plugin and NOT when using hw or plughw. When my period size is half the size of my buffer everything works fine with dmix (I verified this by reducing my buffer size to 8KB and making my period size 4KB. hw_ptr gets incremented then but I get buffer problems). It seems like dmix is forcing me to keep my period size half the size of my buffer. hw and plughw devices do not seem to do that.
Is this a limitation of dmix. Could anyone comment on the dmix limitation that I am experiencing.
Thanks a lot of all your help.
Prabhu.
On Thu, 3 May 2007, Prabhu Sivaraja wrote:
This is a followup to my earlier question about dmix not being able to mix small files. I investigated it more and here is what I found.
Again background info. I have a 64KB as kernel buffer and a 64KB circular buffer for my device. In a poll thread I copy the data obtained from ALSA lib from the kernel buffer to my device circular buffer by keeping track of appl_ptr and hw_ptr(dmix only). Everything was working fine except when I was trying to mix files smaller than 64KB(my buffer size), the small file was not getting mixed through the dmix plugin. Anything over 64KB was getting mixed just fine. So I set about reducing my period size from 32KB to 4KB while maintaining my buffer size at 64KB.
When I tried to do this my hw_ptr never gets incremented from 0. Hence my driver is stuck in a loop waiting for bytes to copy. This is only observed when I am using dmix plugin and NOT when using hw or plughw. When my period size is half the size of my buffer everything works fine with dmix (I verified this by reducing my buffer size to 8KB and making my period size 4KB. hw_ptr gets incremented then but I get buffer problems). It seems like dmix is forcing me to keep my period size half the size of my buffer. hw and plughw devices do not seem to do that.
Is this a limitation of dmix. Could anyone comment on the dmix limitation that I am experiencing.
No, dmix works with more periods well. Send us debug info from /proc/asound/card0/pcm0p (when the stream is opened) and check, if your driver notifies higher PCM level with period_elapsed() and if your position callback returns right value when the stream is started.
Jaroslav
----- Jaroslav Kysela perex@suse.cz Linux Kernel Sound Maintainer ALSA Project, SUSE Labs
Thanks for your help Jaroslav. I am attaching the debug files from /proc/asound/card0/pcm0p/sub0 that I captured while playing a wav file through aplay. I have attached both instances with 64KB bufsize/32KB periodsize(works) and 64KB bufsize/4KB periodsize(does NOT work).
Here are the weird things I noticed in the attached debug captures. Maybe it would make more sense to you. 1. If you take a look at the hw_params file you would see that period and buffer size reported are 1/4th of what I have declared in the driver. Why is this? Seems weird.
2. Status file clearly shows the hw_ptr has not moved in the 4KB period size. I captured these files few secs after starting aplay. hw_ptr never increments from 0 when period size size is NOT half the buf size. About calling period_elapsed( ) like you mentioned in your mail, dont you think it is irrelevant because the hw_ptr never moves to begin with? If it moved for the first period and then never moved I could see how calling period_elapsed( ) would matter but in this case where it never moved from the beginning.
Regards Prabhu.
On 5/4/07, Jaroslav Kysela perex@suse.cz wrote:
On Thu, 3 May 2007, Prabhu Sivaraja wrote:
This is a followup to my earlier question about dmix not being able to
mix
small files. I investigated it more and here is what I found.
Again background info. I have a 64KB as kernel buffer and a 64KB circular buffer for my device.
In
a poll thread I copy the data obtained from ALSA lib from the kernel
buffer
to my device circular buffer by keeping track of appl_ptr and
hw_ptr(dmix
only). Everything was working fine except when I was trying to mix files smaller than 64KB(my buffer size), the small file was not getting mixed through the dmix plugin. Anything over 64KB was getting mixed just fine.
So
I set about reducing my period size from 32KB to 4KB while maintaining
my
buffer size at 64KB.
When I tried to do this my hw_ptr never gets incremented from 0. Hence
my
driver is stuck in a loop waiting for bytes to copy. This is only
observed
when I am using dmix plugin and NOT when using hw or plughw. When my
period
size is half the size of my buffer everything works fine with dmix (I verified this by reducing my buffer size to 8KB and making my period
size
4KB. hw_ptr gets incremented then but I get buffer problems). It seems
like
dmix is forcing me to keep my period size half the size of my buffer. hw
and
plughw devices do not seem to do that.
Is this a limitation of dmix. Could anyone comment on the dmix
limitation
that I am experiencing.
No, dmix works with more periods well. Send us debug info from /proc/asound/card0/pcm0p (when the stream is opened) and check, if your driver notifies higher PCM level with period_elapsed() and if your position callback returns right value when the stream is started.
Jaroslav
Jaroslav Kysela perex@suse.cz Linux Kernel Sound Maintainer ALSA Project, SUSE Labs
participants (2)
-
Jaroslav Kysela -
Prabhu Sivaraja