Re: [linux-sunxi] Re: [PATCH v4 6/9] ASoC: sun4i-i2s: Add multi-lane functionality

[Maxime Ripard]

On Tue, Aug 06, 2019 at 02:22:13PM +0800, Chen-Yu Tsai wrote:
> On Thu, Aug 1, 2019 at 1:32 PM Jernej Åkrabec <jernej.skrabec@xxxxxxxxx> wrote:
> >
> > Dne sreda, 31. julij 2019 ob 14:29:53 CEST je Maxime Ripard napisal(a):
> > > On Tue, Jul 30, 2019 at 07:57:10PM +0200, Jernej Åkrabec wrote:
> > > > Dne torek, 04. junij 2019 ob 11:38:44 CEST je Code Kipper napisal(a):
> > > > > On Tue, 4 Jun 2019 at 11:02, Christopher Obbard <chris@xxxxxxxxxxxx>
> > wrote:
> > > > > > On Tue, 4 Jun 2019 at 09:43, Code Kipper <codekipper@xxxxxxxxx> wrote:
> > > > > > > On Tue, 4 Jun 2019 at 09:58, Maxime Ripard
> > > > > > > <maxime.ripard@xxxxxxxxxxx>
> > > >
> > > > wrote:
> > > > > > > > On Mon, Jun 03, 2019 at 07:47:32PM +0200, codekipper@xxxxxxxxx
> > wrote:
> > > > > > > > > From: Marcus Cooper <codekipper@xxxxxxxxx>
> > > > > > > > >
> > > > > > > > > The i2s block supports multi-lane i2s output however this
> > > > > > > > > functionality
> > > > > > > > > is only possible in earlier SoCs where the pins are exposed and
> > > > > > > > > for
> > > > > > > > > the i2s block used for HDMI audio on the later SoCs.
> > > > > > > > >
> > > > > > > > > To enable this functionality, an optional property has been
> > > > > > > > > added to
> > > > > > > > > the bindings.
> > > > > > > > >
> > > > > > > > > Signed-off-by: Marcus Cooper <codekipper@xxxxxxxxx>
> > > > > > > >
> > > > > > > > I'd like to have Mark's input on this, but I'm really worried
> > > > > > > > about
> > > > > > > > the interaction with the proper TDM support.
> > > > > > > >
> > > > > > > > Our fundamental issue is that the controller can have up to 8
> > > > > > > > channels, but either on 4 lines (instead of 1), or 8 channels on 1
> > > > > > > > (like proper TDM) (or any combination between the two, but that
> > > > > > > > should
> > > > > > > > be pretty rare).
> > > > > > >
> > > > > > > I understand...maybe the TDM needs to be extended to support this to
> > > > > > > consider channel mapping and multiple transfer lines. I was thinking
> > > > > > > about the later when someone was requesting support on IIRC a while
> > > > > > > ago, I thought masking might of been a solution. These can wait as
> > > > > > > the
> > > > > > > only consumer at the moment is LibreELEC and we can patch it there.
> > > > > >
> > > > > > Hi Marcus,
> > > > > >
> > > > > > FWIW, the TI McASP driver has support for TDM & (i think?) multiple
> > > > > > transfer lines which are called serializers.
> > > > > > Maybe this can help with inspiration?
> > > > > > see
> > > > > > https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tre
> > > > > > e/s
> > > > > > ound/soc/ti/davinci-mcasp.c sample DTS:
> > > > > >
> > > > > > &mcasp0 {
> > > > > >
> > > > > >     #sound-dai-cells = <0>;
> > > > > >     status = "okay";
> > > > > >     pinctrl-names = "default";
> > > > > >     pinctrl-0 = <&mcasp0_pins>;
> > > > > >
> > > > > >     op-mode = <0>;
> > > > > >     tdm-slots = <8>;
> > > > > >     serial-dir = <
> > > > > >
> > > > > >         2 0 1 0
> > > > > >         0 0 0 0
> > > > > >         0 0 0 0
> > > > > >         0 0 0 0
> > > > > >     >
> > > > > >     >;
> > > > > >
> > > > > >     tx-num-evt = <1>;
> > > > > >     rx-num-evt = <1>;
> > > > > >
> > > > > > };
> > > > > >
> > > > > > Cheers!
> > > > >
> > > > > Thanks, this looks good.
> > > >
> > > > I would really like to see this issue resolved, because HDMI audio support
> > > > in mainline Linux for those SoCs is long overdue.
> > > >
> > > > However, there is a possibility to still add HDMI audio suport (stereo
> > > > only) even if this issue is not completely solved. If we agree that
> > > > configuration of channels would be solved with additional property as
> > > > Christopher suggested, support for >2 channels can be left for a later
> > > > time when support for that property would be implemented. Currently,
> > > > stereo HDMI audio support can be added with a few patches.
> > > >
> > > > I think all I2S cores are really the same, no matter if internally
> > > > connected to HDMI controller or routed to pins, so it would make sense to
> > > > use same compatible for all of them. It's just that those I2S cores which
> > > > are routed to pins can use only one lane and >2 channels can be used only
> > > > in TDM mode of operation, if I understand this correctly.
> > > >
> > > > New property would have to be optional, so it's omission would result in
> > > > same channel configuration as it is already present, to preserve
> > > > compatibility with old device trees. And this mode is already sufficient
> > > > for stereo HDMI audio support.
> > >
> > > Yeah, it looks like a good plan.
> > >
> > > > Side note: HDMI audio with current sun4i-i2s driver has a delay (about a
> > > > second), supposedly because DW HDMI controller automatically generates CTS
> > > > value based on I2S clock (auto CTS value generation is enabled per
> > > > DesignWare recomendation for DW HDMI I2S interface).
> > >
> > > Is that a constant delay during the audio playback, or only at startup?
> >
> > I think it's just at startup, e.g. if you're watching movie, audio is in sync,
> > it's just that first second or so is silent.
> >
> > >
> > > > I2S driver from BSP Linux solves that by having I2S clock output
> > > > enabled all the time. Should this be flagged with some additional
> > > > property in DT?
> > >
> > > I'd say that if the codec has that requirement, then it should be
> > > between the codec and the DAI, the DT doesn't really have anything to
> > > do with this.
> >
> > Ok, but how to communicate that fact to I2S driver then? BSP driver solves
> > that by using different compatible, but as I said before, I2S cores are not
> > really different, so this seems wrong.
>
> Maybe we could make the DW-HDMI I2S driver require the I2S clock be on all
> the time? You wouldn't need any changes to the DT.

That's an option, but I'd really like to avoid it if possible.

I guess we could also just add a delay in the powerup path in the HDMI
case? Would it work?

maxime

--
Maxime Ripard, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com

Attachment: signature.asc
Description: PGP signature