21.09.2014 15:53, Clemens Ladisch wrote:
> Raymond Yau wrote:
>> Does snd-oxygen provide this position with granularity which is less than
>> the minimum period size ?
>
> Yes.  Its DMA controller uses a burst size of 32 bytes.  The pointer registers
> are documented as reporting the position with 4-byte accuracy, but there were
> problems when the period/buffer sizes were not aligned to 32 bytes.  I've set
> the minimum period size to 64 bytes, just to be safe.

Well, my friend has tested the card anyway, and I am afraid that the 
situation is in fact more complex than documented. Look how fast (8 
samples per ioctl in a busy loop!) the card eats the first 256 samples. 
This is very suspicious, and I am not really sure whether we can 
actually perform a rewind over these 256 samples.

The "usual" step of the pointer updates (via snd_pcm_avail) is indeed 8 
samples = 32 bytes with some fine-grained updates sometimes showing up 
in the middle, which matches the documented burst size.

-- 
Alexander E. Patrakov