[RFC PATCH 2/4] ASoC: soc-generic-dmaengine-pcm: Add custom prepare and submit function

[Sit, Michael Wei Hong]

> -----Original Message-----
> From: Sia, Jee Heng <jee.heng.sia at intel.com>
> Sent: Tuesday, 8 December, 2020 11:21 AM
> To: Pierre-Louis Bossart <pierre-louis.bossart at linux.intel.com>;
> Lars-Peter Clausen <lars at metafoo.de>; Sit, Michael Wei Hong
> <michael.wei.hong.sit at intel.com>; Shevchenko, Andriy
> <andriy.shevchenko at intel.com>
> Cc: Rojewski, Cezary <cezary.rojewski at intel.com>; alsa-
> devel at alsa-project.org; tiwai at suse.com;
> liam.r.girdwood at linux.intel.com; vkoul at kernel.org;
> broonie at kernel.org
> Subject: RE: [RFC PATCH 2/4] ASoC: soc-generic-dmaengine-pcm:
> Add custom prepare and submit function
> 
> 
> 
> > -----Original Message-----
> > From: Pierre-Louis Bossart <pierre-
> louis.bossart at linux.intel.com>
> > Sent: 07 December 2020 11:36 PM
> > To: Lars-Peter Clausen <lars at metafoo.de>; Sit, Michael Wei
> Hong
> > <michael.wei.hong.sit at intel.com>; Sia, Jee Heng
> > <jee.heng.sia at intel.com>; Shevchenko, Andriy
> > <andriy.shevchenko at intel.com>
> > Cc: Rojewski, Cezary <cezary.rojewski at intel.com>; alsa-
> devel at alsa-
> > project.org; tiwai at suse.com; liam.r.girdwood at linux.intel.com;
> > vkoul at kernel.org; broonie at kernel.org
> > Subject: Re: [RFC PATCH 2/4] ASoC: soc-generic-dmaengine-
> pcm: Add
> > custom prepare and submit function
> >
> >
> >
> >
> > > If you really want to limit your period size you need to install a
> > > range constraint on the
> SNDRV_PCM_HW_PARAM_PERIOD_SIZE
> > parameter.
> > >
> > > But I'd highly recommend against it and just split the transfer
> into
> > > multiple segments in the DMA driver. Needlessly limiting the
> period
> > > size will increase the number of interrupts during audio
> > > playback/recording and hurt the power efficiency of your
> system.
> >
> > Yes that was also an objection from me, the fix should be in the
> DMA
> > level. The 1024 block limitation would mean restricting the
> period
> > size to be at most 5.3 or 10.6ms (16 and 32-bit cases). That's way
> to small.
> [>>] Seems like complexity increases if splitting the segments in
> ASoC. This is not a framework issue nor architecture issue.
> If introducing new API to DMAENGINE to constraint the number
> of items is not recommended, then lets split the segments in
> DMA driver.

With the increased complexity of introducing new APIs can we move the segment splitting to the DMA driver?
Anymore concerns with doing so?