[alsa-devel] [Alsa-user] hdmi sound rerouted?

[Takashi Iwai]

At Mon, 30 Sep 2013 02:23:22 +0000,
Deucher, Alexander wrote:
> 
> > -----Original Message-----
> > From: Clemens Ladisch [mailto:cladisch at googlemail.com]
> > Sent: Saturday, September 28, 2013 6:33 PM
> > To: Dagg Stompler
> > Cc: alsa-user at lists.sourceforge.net; Deucher, Alexander; alsa-devel at alsa-
> > project.org
> > Subject: Re: [Alsa-user] [alsa-devel] hdmi sound rerouted?
> > 
> > Dagg Stompler wrote:
> > >> daggs at gmx.com wrote:
> > >>> card 0: PCH [HDA Intel PCH], device 3: HDMI 0 [HDMI 0]
> > >>> card 1: Generic [HD-Audio Generic], device 3: HDMI 0 [HDMI 0]
> > >>> card 2: HDMI [HDA ATI HDMI], device 3: HDMI 0 [HDMI 0]
> > >>
> > >> The second one is called "Generic" because your kernel does not (yet)
> > >> know the hardware.
> > >>
> > >>> 01:00.1 Audio device: Advanced Micro Devices, Inc. [AMD/ATI] Cedar
> > HDMI Audio [Radeon HD 5400/6300 Series]
> > >>> 06:00.1 Audio device: Advanced Micro Devices, Inc. [AMD/ATI] RV635
> > HDMI Audio [Radeon HD 3600 Series]
> > >>
> > >> These are the HDMI outputs of *two* graphics cards. The first one's
> > >> model is so new that the driver doesn't know it.
> > >
> > > the 3600 doesn't have hdmi socket so I think it is irrelevant for my issue.
> > 
> > The RV635 chip does support HDMI outputs (even if none are actually
> > connected), so the corresponding PCI device shows up here.
> > 
> > >>> If you update the kernel, you will get two devices named "HDA ATI
> > HDMI".
> > >
> > > upgrade to where?
> > 
> > Oops.  Even the latest driver lacks the PCI IDs of many HDMI HDA devices
> > (aa50/aa58/aa60/aa68/aa80/aa88/aa90/aa98:
> > Cypress/Juniper/Redwood/Cedar/
> > Cayman+Antilles/Barts/Turks/Caicos).
> > 
> > Alex, are these devices that require disabling snooping?
> 
> Sorry what does snooping mean in this context?

Most of HD-audio controller have a snoop control, but the recent AMD
HDMI audio controller don't have the feature, thus the driver must use
the non-cached pages.  At least, PCI ID 1002:aaa0, aaa8 and aab0 are
known to require such.

Now the question is whether all new chips above (aa50,...) require the
same workaround, too.  Could you check this or forward the inquiry to
someone in AMD?

FWIW, such a chip needs to set AZX_DRIVER_ATIHDMI_NS bits in
driver_data of pci entries in sound/pci/hda/hda_intel.c.


thanks,

Takashi