radeon multi-channel lpcm

radeon multi-channel lpcm

[Pierre Ossman]

[-- Attachment #1.1: Type: text/plain, Size: 1162 bytes --]

Hi,

I figured I'd have a go at finishing up support for multi-channel lcpm
over HDMI on radeon cards. I read a post from the AMD guys stating that
the GPU portions were already in place, and it was the audio pieces
that were missing. Unfortunately I was unable to figure out exactly what
it is that remains to be done.

I poked around in patch_hdmi.c, and concluded that the generic code is
what's being used for modern radeon cards. The card exposes two widgets
(and output widget and a pin complex). But both widgets indicate
through their capability registers that they only support stereo.

I don't know enough about HDA to have any clue where to go from here.
Should the capability field be ignored for HDMI? Or does it need some
coaxing to allow more than stereo? The ELD stuff isn't properly updated
from the radeon driver, but the code seems to suggest it is just for
hints, and not critical.

Rgds
-- 
     -- Pierre Ossman

  WARNING: This correspondence is being monitored by FRA, a
  Swedish intelligence agency. Make sure your server uses
  encryption for SMTP traffic and consider using PGP for
  end-to-end encryption.

[-- Attachment #1.2: signature.asc --]
[-- Type: application/pgp-signature, Size: 230 bytes --]

[-- Attachment #2: Type: text/plain, Size: 0 bytes --]

Re: radeon multi-channel lpcm

[Pierre Ossman]

[-- Attachment #1.1: Type: text/plain, Size: 1441 bytes --]

Noone with any insight? I'm ready to do some leg work, but I couldn't
find anything in either the HDA specification, or the AMD
documentation. :/

On Fri, 31 Aug 2012 22:34:47 +0200
Pierre Ossman <pierre-list@ossman.eu> wrote:

> Hi,
> 
> I figured I'd have a go at finishing up support for multi-channel lcpm
> over HDMI on radeon cards. I read a post from the AMD guys stating that
> the GPU portions were already in place, and it was the audio pieces
> that were missing. Unfortunately I was unable to figure out exactly what
> it is that remains to be done.
> 
> I poked around in patch_hdmi.c, and concluded that the generic code is
> what's being used for modern radeon cards. The card exposes two widgets
> (and output widget and a pin complex). But both widgets indicate
> through their capability registers that they only support stereo.
> 
> I don't know enough about HDA to have any clue where to go from here.
> Should the capability field be ignored for HDMI? Or does it need some
> coaxing to allow more than stereo? The ELD stuff isn't properly updated
> from the radeon driver, but the code seems to suggest it is just for
> hints, and not critical.
> 
> Rgds


-- 
     -- Pierre Ossman

  WARNING: This correspondence is being monitored by FRA, a
  Swedish intelligence agency. Make sure your server uses
  encryption for SMTP traffic and consider using PGP for
  end-to-end encryption.

[-- Attachment #1.2: signature.asc --]
[-- Type: application/pgp-signature, Size: 230 bytes --]

[-- Attachment #2: Type: text/plain, Size: 0 bytes --]

Re: radeon multi-channel lpcm

[Takashi Iwai]

At Wed, 5 Sep 2012 18:19:23 +0200,
Pierre Ossman wrote:
> 
> Noone with any insight? I'm ready to do some leg work, but I couldn't
> find anything in either the HDA specification, or the AMD
> documentation. :/

Well, HD-audio HDMI codec driver just extracts the available channel
setup from the given ELD, and the ELD data is sent from the graphics
driver.  So, if anything missing in ELD, it's supposed to be in the
graphics driver side.  In other words, if ELD doesn't give 6 channels,
HD-audio codec won't follow that channel setup, too.


Takashi

> 
> On Fri, 31 Aug 2012 22:34:47 +0200
> Pierre Ossman <pierre-list@ossman.eu> wrote:
> 
> > Hi,
> > 
> > I figured I'd have a go at finishing up support for multi-channel lcpm
> > over HDMI on radeon cards. I read a post from the AMD guys stating that
> > the GPU portions were already in place, and it was the audio pieces
> > that were missing. Unfortunately I was unable to figure out exactly what
> > it is that remains to be done.
> > 
> > I poked around in patch_hdmi.c, and concluded that the generic code is
> > what's being used for modern radeon cards. The card exposes two widgets
> > (and output widget and a pin complex). But both widgets indicate
> > through their capability registers that they only support stereo.
> > 
> > I don't know enough about HDA to have any clue where to go from here.
> > Should the capability field be ignored for HDMI? Or does it need some
> > coaxing to allow more than stereo? The ELD stuff isn't properly updated
> > from the radeon driver, but the code seems to suggest it is just for
> > hints, and not critical.
> > 
> > Rgds
> 
> 
> -- 
>      -- Pierre Ossman
> 
>   WARNING: This correspondence is being monitored by FRA, a
>   Swedish intelligence agency. Make sure your server uses
>   encryption for SMTP traffic and consider using PGP for
>   end-to-end encryption.
> [1.2 signature.asc <application/pgp-signature (7bit)>]
> 
> [2  <text/plain; us-ascii (7bit)>]
> _______________________________________________
> Alsa-devel mailing list
> Alsa-devel@alsa-project.org
> http://mailman.alsa-project.org/mailman/listinfo/alsa-devel

Re: radeon multi-channel lpcm

[Pierre Ossman]

[-- Attachment #1.1: Type: text/plain, Size: 1730 bytes --]

On Wed, 05 Sep 2012 18:39:01 +0200
Takashi Iwai <tiwai@suse.de> wrote:

> At Wed, 5 Sep 2012 18:19:23 +0200,
> Pierre Ossman wrote:
> > 
> > Noone with any insight? I'm ready to do some leg work, but I couldn't
> > find anything in either the HDA specification, or the AMD
> > documentation. :/
> 
> Well, HD-audio HDMI codec driver just extracts the available channel
> setup from the given ELD, and the ELD data is sent from the graphics
> driver.  So, if anything missing in ELD, it's supposed to be in the
> graphics driver side.  In other words, if ELD doesn't give 6 channels,
> HD-audio codec won't follow that channel setup, too.
> 

I see. I got a bit confused by this code though:

	/* Restrict capabilities by ELD if this isn't disabled */
	if (!static_hdmi_pcm && eld->eld_valid) {

Which made me think that the ELD was just used to make sure you didn't
send something the other end couldn't handle. IOW, eld_valid == false
would give you all features the hw was capable off.


Some follow up questions, to confirm I haven't confused things:

 - The ELD really needs to go into the hardware, not just end up in the
   hands of the audio driver? IOW, we cannot override this?

 - The HDA interface only defines reading the ELD, so writing it would
   have to be done by the graphics driver using some GPU interface?

If so, I need to poke the AMD guys as I couldn't find any information
in the public documentation about writing the ELD.

Thanks
-- 
     -- Pierre Ossman

  WARNING: This correspondence is being monitored by FRA, a
  Swedish intelligence agency. Make sure your server uses
  encryption for SMTP traffic and consider using PGP for
  end-to-end encryption.

[-- Attachment #1.2: signature.asc --]
[-- Type: application/pgp-signature, Size: 230 bytes --]

[-- Attachment #2: Type: text/plain, Size: 0 bytes --]

Re: radeon multi-channel lpcm

[Takashi Iwai]

At Wed, 5 Sep 2012 19:22:39 +0200,
Pierre Ossman wrote:
> 
> On Wed, 05 Sep 2012 18:39:01 +0200
> Takashi Iwai <tiwai@suse.de> wrote:
> 
> > At Wed, 5 Sep 2012 18:19:23 +0200,
> > Pierre Ossman wrote:
> > > 
> > > Noone with any insight? I'm ready to do some leg work, but I couldn't
> > > find anything in either the HDA specification, or the AMD
> > > documentation. :/
> > 
> > Well, HD-audio HDMI codec driver just extracts the available channel
> > setup from the given ELD, and the ELD data is sent from the graphics
> > driver.  So, if anything missing in ELD, it's supposed to be in the
> > graphics driver side.  In other words, if ELD doesn't give 6 channels,
> > HD-audio codec won't follow that channel setup, too.
> > 
> 
> I see. I got a bit confused by this code though:
> 
> 	/* Restrict capabilities by ELD if this isn't disabled */
> 	if (!static_hdmi_pcm && eld->eld_valid) {
> 
> Which made me think that the ELD was just used to make sure you didn't
> send something the other end couldn't handle. IOW, eld_valid == false
> would give you all features the hw was capable off.
> 
> 
> Some follow up questions, to confirm I haven't confused things:
> 
>  - The ELD really needs to go into the hardware, not just end up in the
>    hands of the audio driver? IOW, we cannot override this?

Usually ELD won't go to the hardware as is but translated in a
different way in the graphics side.

>  - The HDA interface only defines reading the ELD, so writing it would
>    have to be done by the graphics driver using some GPU interface?

Yes.

> If so, I need to poke the AMD guys as I couldn't find any information
> in the public documentation about writing the ELD.

It's not about "writing" ELD.  If you got no valid ELD, it is the
graphics driver who didn't send it.  The graphics hardware certainly
sends some audio information, and the graphics driver translates into
an ELD and notifies to the audio codec.  This part is missing or
wrongly done, supposedly.


Takashi

> 
> Thanks
> -- 
>      -- Pierre Ossman
> 
>   WARNING: This correspondence is being monitored by FRA, a
>   Swedish intelligence agency. Make sure your server uses
>   encryption for SMTP traffic and consider using PGP for
>   end-to-end encryption.

Re: radeon multi-channel lpcm

[Pierre Ossman]

[-- Attachment #1.1: Type: text/plain, Size: 2354 bytes --]

On Thu, 06 Sep 2012 07:59:52 +0200
Takashi Iwai <tiwai@suse.de> wrote:

> At Wed, 5 Sep 2012 19:22:39 +0200,
> Pierre Ossman wrote:
> > 
> > Some follow up questions, to confirm I haven't confused things:
> > 
> >  - The ELD really needs to go into the hardware, not just end up in the
> >    hands of the audio driver? IOW, we cannot override this?
> 
> Usually ELD won't go to the hardware as is but translated in a
> different way in the graphics side.
> 
> >  - The HDA interface only defines reading the ELD, so writing it would
> >    have to be done by the graphics driver using some GPU interface?
> 
> Yes.
> 
> > If so, I need to poke the AMD guys as I couldn't find any information
> > in the public documentation about writing the ELD.
> 
> It's not about "writing" ELD.  If you got no valid ELD, it is the
> graphics driver who didn't send it.  The graphics hardware certainly
> sends some audio information, and the graphics driver translates into
> an ELD and notifies to the audio codec.  This part is missing or
> wrongly done, supposedly.
> 

I've studied the Intel driver to get some feel for it, and it seems to
work like this:

 1) The graphics driver gets the EDID data from the GPU.

 2) The kernel constructs the ELD structure from the EDID.

 3) The graphics driver writes the ELD to some GPU registers.

 4) The audio driver reads the ELD via the HDA interface.

But if you're saying that the hardware doesn't really need the ELD,
then steps 3-4 could be replaced by a pure software channel. This would
be convenient as it is currently unknown how to do 3) on radeon.

(2 is currently missing in the radeon driver, but it is trivial to add
AFAICT)

I am a bit confused though. If the hardware doesn't actually look at
the ELD, why are the widgets caps not indicating support for 8
channels? Isn't that coming straight from the hardware?

And static_hdmi_pcm is described as "Don't restrict PCM parameters per
ELD info", which doesn't seem to be the case as I cannot get more than
2 channels no matter the setting of that parameter.

Rgds
-- 
     -- Pierre Ossman

  WARNING: This correspondence is being monitored by FRA, a
  Swedish intelligence agency. Make sure your server uses
  encryption for SMTP traffic and consider using PGP for
  end-to-end encryption.

[-- Attachment #1.2: signature.asc --]
[-- Type: application/pgp-signature, Size: 230 bytes --]

[-- Attachment #2: Type: text/plain, Size: 0 bytes --]

Re: radeon multi-channel lpcm

[Takashi Iwai]

At Thu, 6 Sep 2012 09:09:00 +0200,
Pierre Ossman wrote:
> 
> On Thu, 06 Sep 2012 07:59:52 +0200
> Takashi Iwai <tiwai@suse.de> wrote:
> 
> > At Wed, 5 Sep 2012 19:22:39 +0200,
> > Pierre Ossman wrote:
> > > 
> > > Some follow up questions, to confirm I haven't confused things:
> > > 
> > >  - The ELD really needs to go into the hardware, not just end up in the
> > >    hands of the audio driver? IOW, we cannot override this?
> > 
> > Usually ELD won't go to the hardware as is but translated in a
> > different way in the graphics side.
> > 
> > >  - The HDA interface only defines reading the ELD, so writing it would
> > >    have to be done by the graphics driver using some GPU interface?
> > 
> > Yes.
> > 
> > > If so, I need to poke the AMD guys as I couldn't find any information
> > > in the public documentation about writing the ELD.
> > 
> > It's not about "writing" ELD.  If you got no valid ELD, it is the
> > graphics driver who didn't send it.  The graphics hardware certainly
> > sends some audio information, and the graphics driver translates into
> > an ELD and notifies to the audio codec.  This part is missing or
> > wrongly done, supposedly.
> > 
> 
> I've studied the Intel driver to get some feel for it, and it seems to
> work like this:
> 
>  1) The graphics driver gets the EDID data from the GPU.
> 
>  2) The kernel constructs the ELD structure from the EDID.
> 
>  3) The graphics driver writes the ELD to some GPU registers.
> 
>  4) The audio driver reads the ELD via the HDA interface.
> 
> But if you're saying that the hardware doesn't really need the ELD,
> then steps 3-4 could be replaced by a pure software channel.

No, what I meant is about the audio codec perspective.  From the audio
codec, it has no direct hardware access to GPU that actually handles
the ELD register things.  It always communicate via HD-audio verbs to
send/receive DIP bytes so the actual work is purely in the graphics
driver side.

> This would
> be convenient as it is currently unknown how to do 3) on radeon.
> 
> (2 is currently missing in the radeon driver, but it is trivial to add
> AFAICT)

Yeah, that's the big missing piece.


Takashi

Re: radeon multi-channel lpcm

[Pierre Ossman]

[-- Attachment #1.1: Type: text/plain, Size: 1748 bytes --]

On Thu, 06 Sep 2012 09:28:52 +0200
Takashi Iwai <tiwai@suse.de> wrote:

> At Thu, 6 Sep 2012 09:09:00 +0200,
> Pierre Ossman wrote:
> > 
> > But if you're saying that the hardware doesn't really need the ELD,
> > then steps 3-4 could be replaced by a pure software channel.
> 
> No, what I meant is about the audio codec perspective.  From the audio
> codec, it has no direct hardware access to GPU that actually handles
> the ELD register things.  It always communicate via HD-audio verbs to
> send/receive DIP bytes so the actual work is purely in the graphics
> driver side.
> 

I assume "codec" here is in the HDA sense, as in the hardware hiding
somewhere behind the HDA interface.

In the Intel case, I'd say the audio hardware as a whole does have some
back channel to the GPU since a write to a GPU register somehow results
in data in the audio hardware. But yes, it would be a purely graphics
driver thing as it has both the ELD, and access to the registers you
can write to.

I'm working on the assumption that the radeon hardware works in similar
fashion as the Intel ones.

> > This would
> > be convenient as it is currently unknown how to do 3) on radeon.
> > 
> > (2 is currently missing in the radeon driver, but it is trivial to add
> > AFAICT)
> 
> Yeah, that's the big missing piece.
> 

So I'll go poke the AMD people, as I cannot find anything in the public
docs about ELD registers. Hopefully it's not something that's still
classified... :/

Thanks
-- 
     -- Pierre Ossman

  WARNING: This correspondence is being monitored by FRA, a
  Swedish intelligence agency. Make sure your server uses
  encryption for SMTP traffic and consider using PGP for
  end-to-end encryption.

[-- Attachment #1.2: signature.asc --]
[-- Type: application/pgp-signature, Size: 230 bytes --]

[-- Attachment #2: Type: text/plain, Size: 0 bytes --]