[Sound-open-firmware] [RFC/RFT] dai: Make system stable by moving DAI buffer pointer update location

[Liam Girdwood]

On Fri, 2018-03-30 at 09:42 -0700, Ranjani Sridharan wrote:
> On Sat, 2018-03-31 at 00:31 +0800, Keyon Jie wrote:
> > From: Wu Zhigang <zhigang.wu at linux.intel.com>
> > 
> > Move the DAI's buffer pointer update location from dai_dma_cb()
> > to dai_copy(). This will make pipeline more stable. The xrun
> > possibility
> > will be lower.Even in xrun it will recover, it will not have dsp
> > oops.
> > DAI's buffer pointer will be updated in the pipeline copy sequence
> > as other components did. Then the DMA will not update the buffer
> > pointer
> > by itself. in HOST--->VOL--->DAI pipeline, when DMA finished, it has
> > to
> > reschedule the pipeline to let the pointer update. this could avoid
> > the xrun when adjust the read and write pointers by different
> > componets.
> 
> I may be wrong here but how do we make sure we are always in sync with
> the clock rate in this case? 

Pipeline scheduling can either based on DMA IRQ or an internal timer. In this
case the r/w pointer updates are aligned with the rest of the pipeline updates
and are scheduled by the DMA IRQ.

Liam