[Intel-gfx] [PATCH 3/6] drm/i915: Don't emit semaphore wait if wrap happened

[Daniel Vetter]

On Thu, Dec 6, 2012 at 12:41 PM, Paulo Zanoni <przanoni at gmail.com> wrote:
> 2012/12/6 Daniel Vetter <daniel at ffwll.ch>:
>> On Wed, Dec 5, 2012 at 9:44 PM, Paulo Zanoni <przanoni at gmail.com> wrote:
>>> 2012/12/4 Mika Kuoppala <mika.kuoppala at linux.intel.com>:
>>>> If wrap just happened we need to prevent emitting waits for
>>>> pre wrap values. Detect this and emit no-ops instead.
>>>>
>>>> v2: Use olr > seqno to detect wrap instead of *seqno == 0
>>>> as suggested by Chris Wilson.
>>>
>>> This commit introduces a bug on Haswell. Now when I'm typing my
>>> password on GDM the screen keeps doing wrong rendering. It "blinks
>>> blue". After logging in I don't see more prodrm/i915: Set initial seqno value close to wrap boundaryblems.
>>
>> Just now I've taken out "drm/i915: Set initial seqno value close to
>> wrap boundary" since QA complained that it regresses things. Does that
>> help for you, too?
>
> It helps: besides the "wrong rendering at GDM screen" I was also
> getting  GPU hangs (when starting X, when running dmesg, when alt+tab,
> etc), and it seems with today's dinq I don't get the gpu hangs
> anymore. I still get the "wrong rendering" problem and it goes away if
> we revert the "Don't emit semaphore wait  if wrap happened".

Ok, looks like we have still some fish left to fry here. I've backed
out the 2nd patch, too. And I guess we need some more tests in i-g-t
to check semaphore correctness, we seem to have some serious gaps ...
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch