On Tue, Aug 06, 2019 at 02:22:13PM +0800, Chen-Yu Tsai wrote:
On Thu, Aug 1, 2019 at 1:32 PM Jernej Škrabec jernej.skrabec@gmail.com 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 Škrabec 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@64studio.com
wrote:
On Tue, 4 Jun 2019 at 09:43, Code Kipper codekipper@gmail.com wrote: > On Tue, 4 Jun 2019 at 09:58, Maxime Ripard > maxime.ripard@bootlin.com
wrote:
> > On Mon, Jun 03, 2019 at 07:47:32PM +0200, codekipper@gmail.com
wrote:
> > > From: Marcus Cooper codekipper@gmail.com > > > > > > 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@gmail.com > > > > 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