Re: [RFC/RFT] HDMI/DP CEA-861-E+ channel allocations 0x20+

[Anssi Hannula <anssi.hannula@iki.fi>]

05.11.2013 16:18, Anssi Hannula kirjoitti:
> 05.11.2013 15:43, Takashi Iwai kirjoitti:
>> At Fri, 01 Nov 2013 23:43:12 +0200,
>> Anssi Hannula wrote:
>>>
>>> Hi all!
>>>
>>> Currently we allow the use of CA values up to 0x31 as defined in
>>> CEA-861-E/F. However, only CA values up to 0x19 are defined in
>>> CEA-861-B/C/D.
>>>
>>> HDMI specification 1.4b specifies that the CA field is to be filled
>>> according to CEA-861-D, and DisplayPort 1.1a according to CEA-861-C.
>>>
>>> The ELD (EDID-Like Data) format as specified by Intel HDA specification
>>> 1.0a has a speaker allocation bitmask that only accommodates speakers
>>> present in CEA-861-D; all of the 0x20+ CAs contain speakers that do not
>>> have a corresponding bit in ELD.
>>>
>>> Using a CA value unsupported by sink will cause either a completely
>>> silent output or stereo output, so I think we should try to prevent
>>> selecting such channel maps, if feasible.
>>>
>>> However, before doing anything, I wonder if they are actually supported
>>> by some newer receivers (mine is 4 years old). It'd be good if someone
>>> with a newish receiver could try the below :)
>>> To test this, one can run (replacing XX and YY with appropriate values
>>> from "aplay -L") on *sound git master*:
>>> speaker-test -c6 -Dhdmi:CARD=XX,DEV=YY -m FL,FR,RL,RR,FLH,FRH
>>> If you get some output for the RL/RR speakers, that should mean that
>>> 0x20+ CAs are supported. If there is no output except on FL/FR and there
>>> is proper output without the "-m FL,FR,RL,RR,FLH,FRH", this means that
>>> 0x20+ CAs are not supported.
>>>
>>> I think there are about these options for us to take:
>>> a) drop 0x20+ CAs from channel_allocations altogether
>>> b) put the 0x20+ CAs under a module parameter
>>> c) only allow 0x20+ CAs if any CEA-861-E+ only speakers are specified
>>>    in EDID. However, as ELD doesn't have bits for these, we'd have to
>>>    employ some non-standard bits in ELD or communicate directly with
>>>    video driver.
>>> d) do nothing, allow the 0x20+ CAs.
>>>
>>> IMHO we should do something, since players using ALSA channel mapping
>>> could just automatically select a manual channel map that uses an
>>> unsupported 0x20+ CA if the source audio stream contains such channels...
>>>
>>> WDYT?
>>
>> Practical options as of now are either (b) or (d).
>>
>> Extending EDID in a non-standard way is no-go.  Better to put finger
>> away.  OTOH, if we're going to that direction, we should rather build
>> a better / more direct communication way with the graphics driver,
>> instead.  This would make things in Intel graphics easier, too, for
>> example.
> 
> Yeah, another benefit of that would be getting the sink
> manufacturer/product/name etc on the older (i.e. all but the very
> latest) ATI/AMD cards. Of course that is a rather limited benefit, and
> I'm not sure this 0x20+ CAs stuff is worth it either...

BTW, regarding ELD from HDA spec:
"Vendor Defined Block:
The vendor defined block of the ELD memory structure byte offset starts
from 4 + Baseline_ELD_Len * 4 to ELD buffer size – 1. This structure is
vendor specific. OS class driver will not interpret this block. Only the
associated vendor defined graphic/audio driver will be able to
understand and enumerate these features based on the specific vendor ELD
version number."

So it is valid to have vendor-defined data in the end of ELD, so we
could put an extended speaker mask for CEA-861-E+ positions there, which
we could use to determine if E+ chmaps should be allowed.
Though there are some problems with this approach (wouldn't work
directly with AMD/ATI nor with 3rdparty video drivers)...

I'm not saying this is a better option than the others, just wanted to
add this for the record :)

>> And (a) isn't the best option, obviously.
>>
>> Now the question is whether (b) or (d).  Can 0x20+ CA be chosen by
>> default from any applications without extra setup?  If it can be done
>> only via user's manual configuration or option, it's essentially
>> user's responsibility.  If so, adding an option to the module is
>> nothing but one more annoyance.  Then I'd take (d).
>>
>> If 0x20+ can be selected automatically in some situations, it'd make
>> sense to block it as default, so (b) would be the choice.  But I guess
>> it won't happen normally.
> 
> Well, I guess not many applications have ALSA chmap support yet, so the
> answer is "not chosen by default" for now.
> 
> I'm going to probably be writing ALSA chmap support for XBMC, and I will
> probably try to set a channel map according to the source audio stream
> channel map (if possible) to avoid software remapping, though. I don't
> have the exact behavior planned out yet, though, but unless 0x20+ CAs
> are blocked by ALSA, I think I'll just avoid using the E+ speaker
> positions by default for HDMI in XBMC ALSA code.
> 
> For the record, the E+ speaker positions are FLH, FRH, FCH, FLW, FRW,
> TC, so they aren't very common. And since HDMI 1.x supports just 8 PCM
> channels, using those in discrete PCM mode means dropping some of the
> standard 7.1 speakers (passthrough is of course a different story).
> 
> Also, even though Peter's receiver accepted the CAs, the receiver
> channel mapping was somewhat weird (e.g. front wide channels were mapped
> to rear speakers).
> 
> I guess I'm leaning towards (d)...
> 


-- 
Anssi Hannula