[PATCH] drm/amdgpu: Simplify amdgpu_lockup_timeout usage.

[Christian König]

Am 21.12.2017 um 18:09 schrieb Michel Dänzer:
> On 2017-12-19 06:47 PM, Christian König wrote:
>> Yeah, that is a known issue which came to front again because Andrey's
>> patch is slightly buggy.
>>
>> Please test and review the attached (only compile tested) fix for
>> Andrey's patch.
>>
>> Still working on finding the root cause, but so far didn't had time for
>> that.
> Unfortunately, I haven't been able to narrow it down to any particular
> piglit tests. What I can say though is that it's triggered much more
> quickly when running piglit with --process-isolation false.

I think I'm able to reproduce it now as well.

I thought a bit about what could be the difference between my test 
machines and the system you guys where able to reproduce it on, because 
of the lack of VM faults on my side no matter what I do.

My conclusion was that it is somehow related to the installed amount of 
system memory and reducing TTMs swap limit indeed resulted in a VM fault 
rather quickly.

I was only able to reproduce this once now, but that was so quick that I 
might have found the root cause.

Christian.