[PATCH 2/6] drm/radeon: remove HDMI interrupts on Evergreen

[Rafał Miłecki]

2013/4/15 Alex Deucher <alexdeucher at gmail.com>:
> On Mon, Apr 15, 2013 at 4:08 AM, Christian König
> <deathsimple at vodafone.de> wrote:
>> Am 14.04.2013 20:02, schrieb Rafał Miłecki:
>>
>>> 2013/4/14 Alex Deucher <alexdeucher at gmail.com>:
>>>>
>>>> On Sun, Apr 14, 2013 at 11:55 AM, Rafał Miłecki <zajec5 at gmail.com> wrote:
>>>>>
>>>>> 2013/4/14 Alex Deucher <alexdeucher at gmail.com>:
>>>>>>
>>>>>> On Sat, Apr 13, 2013 at 7:26 PM, Rafał Miłecki <zajec5 at gmail.com>
>>>>>> wrote:
>>>>>>>
>>>>>>> We need interrupts on format change for R6xx only, where hardware
>>>>>>> seems
>>>>>>> to be somehow bugged and requires setting audio info manually.
>>>>>>
>>>>>> Can you confirm that this is actually needed on older chips?  AFAIK,
>>>>>> it shouldn't be required for any chips.  It's mainly for debugging.
>>>>>
>>>>> I can't really right now :( My notebook with RV620 died (hard disk
>>>>> ended it's life and power cable got broken). I hope to resurrect him
>>>>> in about a week.
>>>>>
>>>>> If that isn't needed on R6xx, I'm not sure why we implemented it in
>>>>> first place at all. Christian? Do you have idea why this was required?
>>>>> I remember than in first place we were using timer, then we switched
>>>>> to the interrupts. But why we needed it at all?
>>>>
>>>> I suspect it was just assumed to be necessary due to the original RE.
>>>
>>> I'm OK with removing that from R6xx too, if it's not needed. I just
>>> want to check that first, to don't break audio accidentally. In case
>>> of Evergreen I was able to test it, so I dares to submit this patch ;)
>>
>>
>> Well, originally I was just imitating fglrx behavior with this, but since I
>> now have access to the AMD documentation I can't find a reason why fglrx was
>> actually doing it like this. In theory format changes should work on their
>> own, but it is still possible they did this because of some bug or something
>> like this.
>>
>> I can't really test it anymore either, so no idea if it is really required
>> or not.
>
> For both evergreen and older asics, maybe rather than removing the
> code, we could just disable the interrupt src.  I.e., remove the call
> to radeon_irq_kms_enable_afmt().  That way we can always re-enable it
> if we need it for testing or debugging.

Even removing the code still keeps it in git history :)

Just give me some more time, so I can resurrect my old RV620 and make
some tests.

-- 
Rafał