2012/10/29 Matthijs Kooijman matthijs@stdin.nl:

I'm using some evergreen systems with HDMI output, which got broken by a recent commit. In commit 1c3439f (drm/radeon/hdmi: update modesetting), the HDMI_ACR_SOURCE bit is set in the HDMI_ACR_PACKET_CONTROL register.

However, setting this bit breaks my HDMI audio when a non-default resolution is used. This problem only occurs on some TV's (it's broken on a Philips, Sharp and Sony TV and seems to be working on a few LG TV's).

On the Sony, there are short pulses of distorted audio (couple of hundred ms audio, then half a second or so silence, then some audio again, etc.). On the Philips and Shart TV's, the audio is completely muted.

It seems setting this bit causes the audio infoframe timings to be controlled by software instead. I suspect that the timings are not properly (re)calculated when a non-default resolution is used, causing the audio to break.

Alex Deucher mentioned on IRC that the proper timing code needs to be ported over. I might have a look at that, if I can find enough time and understanding, but that might take a while. Would it make sense to disable this particular bit for the time being, to make audio working again? Is there any particular reason to not keep the bit disabled and let the timings be determined by the hardware in any case?

I've (finally) taken a look at fglrx and it also doesn't set that bit: RREG32(0x0001203c); -> 0x00010000 WREG32(0x0001203c, 0x00011000);

0x0001203c = HDMI_ACR_PACKET_CONTROL 0x00011000 = HDMI_ACR_AUTO_SEND | HDMI_ACR_N_MULTIPLE(1)

I think maybe we should ignore Alex's code [0] based on (probably) some internal specs and just copy fglrx behavior. I guess the fglrx team discovered some issues with the specs and fixed just the code.

[0] http://people.freedesktop.org/~agd5f/0001-WIP-port-of-hdmi-dp-audio-code-to-...