10.11.2013 08:01, Anssi Hannula kirjoitti:
10.11.2013 07:42, Olivier Langlois kirjoitti:
On Sat, 2013-11-09 at 00:03 +0200, Anssi Hannula wrote:
08.11.2013 23:28, Olivier Langlois kirjoitti:
lano1106@whippet2 ~ $ speaker-test -D hdmi:CARD=HDMI,DEV=0 -c8 -r192000 -F S32_LE
speaker-test 1.0.27.2
Playback device is hdmi:CARD=HDMI,DEV=0 Stream parameters are 192000Hz, S32_LE, 8 channels Using 16 octaves of pink noise Rate set to 192000Hz (requested 192000Hz) Buffer size range from 8 to 131072 Period size range from 4 to 65536 Using max buffer size 131072 Periods = 4 was set period_size = 32768 was set buffer_size = 131072 0 - Front Left 4 - Center 1 - Front Right 7 - Side Right 7 - Side Right 6 - Side Left 6 - Side Left 5 - LFE
[...]
OK, speaker-test bug (affecting all generic HDMI codecs).
This is triggered by the fact that the standard CEA/HDMI 8 channel mapping contains RL/RR and RLC/RRC instead of SL/SR and RL/RR. I.e. what one usually considers side speakers is RL/RR and then the rear speakers are RLC/RRC (rear-left-center, rear-right-center).
Speaker-test tries to play back channels in the following order: 0, /* Front Left */ 4, /* Center */ 1, /* Front Right */ 7, /* Side Right */ 3, /* Rear Right */ 2, /* Rear Left */ 6, /* Side Left */ 5, /* LFE */
When it is time to play back Side Left/Right, speaker-test tries to look for SL/SR in the chmap, but doesn't find it, so it just plays back channels 6/7 which are actually "SL/SR" but with a different name (RL/RR).
When it becomes time to playback Rear Left/Right, speaker-test again tries to find RL/RR in the chmap, and this time it does. However, of course the chmap RL/RR corresponds to the HDMI/CEA RL/RR which is traditionally SL/SR. So SL/SR is outputted again.
AFAICS the most straight-forward way to fix this is to check that the chmap actually comprises of the assumed regular channels before trying to apply the predefined playback order, and then fallback to playing channels in-order 0-7 (maybe also provide a predefined ordering for the CEA/HDMI 8ch chmap, though). I can do that later, right now I'm too tired.
BTW, currently speaker-test always uses its own hardcoded channel number=>name mappings for showing the channel name, unless a channel map was manually specified. This is clearly wrong since there may be a device-specific non-standard channelmap.
I can fix that as well, but of course that would mess up the output for HDMI when this is not applied: https://git.kernel.org/cgit/linux/kernel/git/tiwai/sound.git/commit/sound/pc...
So what to do, can we show wrong channel names on broken kernels or should we wait a bit? I guess that one really should be pushed to -stable in any case...
(There is also the question if we want to change the HDMI channel mapping to appear as traditional side+rear instead of the CEA rear+rearcenter. I don't have a strong preference on this so I'm fine with status quo.)