Re: 3.18-rc regression: drm/nouveau: use shared fences for readable objects

Hi,

Sorry to but in like this but I'm suffering from the same kind of deadlocks with nouveau...

The really odd thing is that i could boot some -rc6+ kernel without problems but it hung while playing video and then it refused to start properly again.

Anyway, to quote Maarten: Ok that most likely means the interrupt based wait is borked somehow, so lets find \ out why..

I fear that this happens because of a race in the interface, so my first attempt will \ rule out abuse of the nvif api by nouveau_fence.c

Can you test below patch with the default wait function? ---

I tried the patch, straight on Linus' master and it didn't change anything for me :(