[LAD] 2.6.20 scheduler and buffer underruns

[Darren Gibbs]

So the story is more interesting now... with SCHED_FIFO I'm seeing  
occational delays of about 60 sample frames (@44K) when there is very  
little other CPU activity.  It still looks like a buffer underrun in  
that the audio data is contiguous on both sides of the delay, so I'm  
not dropping a buffer.  When I load the CPU with another task doing a  
while(1) loop, the glitches go away.  Hmmm....

On Jul 11, 2007, at 1:57 PM, Darren Gibbs wrote:

> Thanks, cool... that solves the glitching that scales with CPU  
> load, I still have more rare glitches (every few seconds) that look  
> like buffer underruns.  I'm suspect this is because some other  
> kernel entity has interrupts turned off when the DMA interrupt  
> needs to be serviced to switch buffers.  Are there any clever tools  
> for figuring out who might be doing this?
>
>
> On Jul 11, 2007, at 12:33 PM, Lee Revell wrote:
>
>> On 7/11/07, Darren Gibbs <tsquank at yahoo.com> wrote:
>>> We're doing an ARM-based embedded device, which right now is running
>>> vanilla 2.6.20.  For the sake of simplicity we wrote an OSS driver
>>> that's simply double-buffering and writing to the DAC via I2S.  We
>>> have a buffer underrun problem that is directly proportional to CPU
>>> load... no glitches when simply cat-ing a file to /dev/dsp, but lots
>>> of glitches when other things are happening on the system.  Can
>>> anyone suggest tools/techniques/patches for improving the situation?
>>
>> Run the audio playback app with SCHED_FIFO priority.
>>
>> Lee