At Sat, 16 Feb 2013 22:59:40 +0200, Anssi Hannula wrote:
16.02.2013 18:40, Anssi Hannula kirjoitti:
Some HDMI codecs (at least NVIDIA 0x10de000b:0x10de0101:0x100100) start transmitting an empty audio stream as soon as PIN_OUT and AC_DIG1_ENABLE are enabled.
Since commit 6169b673618bf0b2518ce413b54925782a603f06 ("ALSA: hda - Always turn on pins for HDMI/DP") this happens at first open() time, and will continue even after close().
Additionally, some codecs (at least Intel PantherPoint HDMI) currently continue transmitting HDMI audio even after close() in case some actual audio was output after open() (this happens regardless of PIN_OUT).
Empty HDMI audio transmission when not intended has the effect that a possible HDMI audio sink/receiver may prefer the empty HDMI audio stream over an actual audio stream on its S/PDIF inputs.
To avoid the issue before first prepare(), set stream format to 0 on codec initialization. 0 is not a valid format value for HDMI and will prevent the audio stream from being output.
Additionally, at close() time, make sure that the stream is cleaned up. This will ensure that the format is reset to 0 at that time, preventing audio from being output in that case.
Thanks to OpenELEC developers and users for their help in investigating this issue on the affected NVIDIA "ION2" hardware, and for the initial bug report of the issue. Testing of the final version on NVIDIA ION2 was done by OpenELEC user "MrXIII". Testing on Intel PantherPoint was done by myself.
Signed-off-by: Anssi Hannula anssi.hannula@iki.fi Cc: stable@vger.kernel.org
This also supersedes the patch I attached yesterday as it fixes both cases.
I guess the alternative to this approach would be to fiddle with AC_DIG1_ENABLE when preparing and closing the device, but with a quick look that seemed to be possibly more complex since AC_DIG1_ENABLE is already meddled with at quite a few places in hda_codec.c.
Hmm... actually, that would not be so much more complex, just making sure AC_DIG1_ENABLE is off in hdmi_add_cvt() (by e.g. zeroing AC_VERB_SET_DIGI_CONVERT_1) and making snd_hda_spdif_ctls_unassign() call "set_spdif_ctls(codec, spdif->nid, 0, -1);".
However, it would still mean that just a simple snd_pcm_open() could cause an empty stream to be output (if format is valid), since iec958 controls are assigned at open() time, and iec958 mute controls AC_DIG1_ENABLE. Or we could add additional code in hda_codec.c to prevent AC_DIG1_ENABLE being set if prepare() has not been called.
Is the problem fixed if you set codec->no_sticky_stream = 1?
Actually, the current behavior is intentional. There have been bug reports that just reopening a device causes the receiver not reacting immediately. In the worst case, it takes a few seconds to sync. So we introduced a mechanism to keep the PCM stream assignment.
If your problem is solved by clearing codec->no_stick_stream, we may provide an additional mixer control to turn it on/off, for example, because we can't blindly disable/enable it globally if user's receiver has a problem with the stream sync.
thanks,
Takashi