[alsa-devel] [PATCH] ALSA: hda - fix broken HDMI jack detection after S3

[David Henningsson]

On 08/22/2012 03:52 PM, Takashi Iwai wrote:
> At Wed, 22 Aug 2012 15:46:59 +0200,
> David Henningsson wrote:
>>
>> On 08/22/2012 02:58 PM, Takashi Iwai wrote:
>>> At Wed, 22 Aug 2012 14:39:17 +0200,
>>> David Henningsson wrote:
>>>>
>>>> On 08/22/2012 02:22 PM, Takashi Iwai wrote:
>>>>> At Wed, 22 Aug 2012 14:01:41 +0200,
>>>>> David Henningsson wrote:
>>>>>>
>>>>>> The HDMI codec (an NVIDIA one in this case) forgot that its pins
>>>>>> were unsol enabled, while it was suspended. Therefore jack detection
>>>>>> was broken after S3.
>>>>>> With this patch, we reenable the unsol events on resume,
>>>>>> and also do an extra check afterwards, to see if the HDMI monitor was
>>>>>> plugged/unplugged while in S3.
>>>>>>
>>>>>> Cc: stable at kernel.org (3.3+)
>>>>>> BugLink: https://bugs.launchpad.net/bugs/1040030
>>>>>> Signed-off-by: David Henningsson <david.henningsson at canonical.com>
>>>>>> ---
>>>>>>     sound/pci/hda/patch_hdmi.c |   13 +++++++++++++
>>>>>>     1 file changed, 13 insertions(+)
>>>>>>
>>>>>> diff --git a/sound/pci/hda/patch_hdmi.c b/sound/pci/hda/patch_hdmi.c
>>>>>> index 8f23374..6a3ac05 100644
>>>>>> --- a/sound/pci/hda/patch_hdmi.c
>>>>>> +++ b/sound/pci/hda/patch_hdmi.c
>>>>>> @@ -1315,6 +1315,16 @@ static int generic_hdmi_init(struct hda_codec *codec)
>>>>>>     	return 0;
>>>>>>     }
>>>>>>
>>>>>> +#ifdef CONFIG_PM
>>>>>> +static int generic_hdmi_resume(struct hda_codec *codec)
>>>>>> +{
>>>>>> +	snd_hda_codec_resume_cache(codec);
>>>>>> +	snd_hda_jack_set_dirty_all(codec);
>>>>>> +	snd_hda_jack_report_sync(codec);
>>>>>> +	return 0;
>>>>>
>>>>> Hm, is this really needed?
>>>>>
>>>>> snd_hda_jack_set_dirty_all() is already called in
>>>>> hda_call_codec_resume(), and snd_hda_jack_report_sync() is called in
>>>>> the init callback.
>>>>
>>>> The tester (who has the hardware) has gone for the day, so I can't
>>>> really verify different scenarios right now, but after having looked at
>>>> hda_call_codec_resume I see what you mean...
>>>>
>>>> I do notice one difference though - the order.
>>>> snd_hda_codec_resume_cache, which is what writes the unsol_enable verbs,
>>>> should probably be before the set_dirty_all / report_sync. If not for
>>>> anything else, so for the race condition of somebody plugging/unplugging
>>>> the monitor after checking the jack but before the unsol is enabled.
>>>
>>> Calling snd_hda_jack_report_sync() in init means that all jacks are
>>> checked at first, then the caches are resumed.  So this won't change
>>> ENABLE_UNSOL verb setups.
>>
>> I'm not sure I'm following. Here's the order in hda_call_codec_resume:
>>
>> snd_hda_jack_set_dirty_all()
>> generic_hdmi_init -> snd_hda_jack_report_sync() - get_pin_sense
>> snd_hda_codec_resume_cache() - set_unsol_enable
>>
>> With the (theoretical?) race condition being a change of pin sense
>> between snd_hda_jack_report_sync() and snd_hda_codec_resume_cache(),
>> which will not be picked up.
>
> In that race, yes.  But this should be irrelevant with this bug :)

Yes, the bug needs further verification.

>> The patch changed the order to:
>>
>> snd_hda_codec_resume_cache() - set_unsol_enable
>> snd_hda_jack_set_dirty_all()
>> snd_hda_jack_report_sync() - get_pin_sense
>>
>> Which eliminates that race condition.
>
> Well, what I'm thinking now is rather to call
> snd_hda_jack_report_sync() generically in hda_call_codec_resume().
> The call doesn't have to be in all places.
> Ditto for the initialization case.  It'll clean up many lines.

Fine with me - the more code being generic, the better.

-- 
David Henningsson, Canonical Ltd.
https://launchpad.net/~diwic