Re: [linux-sunxi] [PATCH 05/16] ASoc: sun4i-i2s: Add 20 and 24 bit support

[Charles Keepax <ckeepax@opensource.cirrus.com>]

On Thu, Sep 03, 2020 at 09:40:23AM +0200, Maxime Ripard wrote:
> On Wed, Sep 02, 2020 at 09:22:33PM -0500, Samuel Holland wrote:
> > On 9/2/20 1:10 PM, Jernej Škrabec wrote:
> > > Hi Samuel!
> > > 
> > > Dne petek, 10. julij 2020 ob 07:44:51 CEST je Samuel Holland napisal(a):
> > >> On 7/4/20 6:38 AM, Clément Péron wrote:
> > >>> From: Marcus Cooper <codekipper@gmail.com>
> > >>>
> > >>> Extend the functionality of the driver to include support of 20 and
> > >>> 24 bits per sample.
> > >>>
> > >>> Signed-off-by: Marcus Cooper <codekipper@gmail.com>
> > >>> Signed-off-by: Clément Péron <peron.clem@gmail.com>
> > >>> ---
> > >>>
> > >>>  sound/soc/sunxi/sun4i-i2s.c | 11 +++++++++--
> > >>>  1 file changed, 9 insertions(+), 2 deletions(-)
> > >>>
> > >>> diff --git a/sound/soc/sunxi/sun4i-i2s.c b/sound/soc/sunxi/sun4i-i2s.c
> > >>> index f78167e152ce..bc7f9343bc7a 100644
> > >>> --- a/sound/soc/sunxi/sun4i-i2s.c
> > >>> +++ b/sound/soc/sunxi/sun4i-i2s.c
> > >>> @@ -577,6 +577,9 @@ static int sun4i_i2s_hw_params(struct
> > >>> snd_pcm_substream *substream,> 
> > >>>  	case 16:
> > >>>  		width = DMA_SLAVE_BUSWIDTH_2_BYTES;
> > >>>  		break;
> > >>>
> > >>> +	case 32:
> > >>> +		width = DMA_SLAVE_BUSWIDTH_4_BYTES;
> > >>> +		break;
> > >>
> > >> This breaks the sun4i variants, because sun4i_i2s_get_wss returns 4 for a 32
> > >> bit width, but it needs to return 3.
> > > 
> > > I'm not sure what has WSS with physical width and DMA?
> > 
> > This is the change where creating a S24_LE stream no longer fails with -EINVAL.
> > So this is the change where userspace stops downsampling 24-bit audio sources.
> > So this is the change where playback of 24-bit audio sources breaks, because WSS
> > is programmed wrong.
> > 
> > >> As a side note, I wonder why we use the physical width (the spacing between
> > >> samples in RAM) to drive the slot width. S24_LE takes up 4 bytes per sample
> > >> in RAM, which we need for DMA. But I don't see why we would want to
> > >> transmit the padding over the wire. I would expect it to be transmitted the
> > >> same as S24_3LE (which has no padding). It did not matter before, because
> > >> the only supported format had no padding.
> > > 
> > > Allwinner DMA engines support only 1, 2, 4 and sometimes 8 bytes for bus 
> > > width, so if sample is 24 bits in size, we have no other way but to transmit 
> > > padding too.
> > 
> > I understand why we do 4 byte DMA from RAM <=> I2S FIFO; that was not my
> > question. I'm referring to the actual wire format (FIFO <=> PCM_DIN/DOUT). The
> > sample is already truncated from 32 bits to 24 bits in the FIFO -- that's what
> > TXIM and RXOM in FIFO_CTRL control.
> > 
> > If a sample is 24 bits wide, why would we send 32 BCLKs for every LRCK? I would
> > expect the slot width to match the sample resolution by default. But yet we have
> > this code in the driver:
> > 
> >     unsigned int word_size = params_width(params);
> >     unsigned int slot_width = params_physical_width(params);
> > 
> > I think slot_width should be the same as word_size, and I suggest changing it
> > before adding 20/24-bit support.
> 
> Generally speaking, the slot width doesn't necessarily match the
> physical width. With TDM for example you may very well have slots
> larger than their samples.
> 
> That being said, S24 is explicitly a format where you send a sample of
> 24 bits in a 32-bit word (in the lowest three bytes, little endian)
> 
> See:
> https://elixir.bootlin.com/linux/v5.9-rc3/source/sound/core/pcm_misc.c#L75
> https://mailman.alsa-project.org/pipermail/alsa-devel/2013-April/061073.html
> 
> 24 bits of data over three bytes like you suggest is S24_3LE
> 

My understanding is physical_width refers to the in memory
representation, but shouldn't be used to control the slot width
on the bus. If not specified otherwise (say through the set_tdm
callback), and if the appropriate BCLK is supported, then the slot
should be just large enough to hold the data.

Thanks,
Charles