[Intel-xe] [PATCH 2/2] drm/xe: Drop coalescing of GGTT invalidations

[Maarten Lankhorst]

On 2023-04-06 06:23, Niranjana Vishwanathapura wrote:
> On Thu, Apr 06, 2023 at 03:44:33AM +0000, Matthew Brost wrote:
>> On Wed, Apr 05, 2023 at 08:42:36PM -0700, Niranjana Vishwanathapura 
>> wrote:
>>> On Wed, Apr 05, 2023 at 04:34:03PM -0700, Matthew Brost wrote:
>>> > Causing a bunch of hangs driver load and in user space. Benefit is
>>> > rather small too, so just drop this for to stablize the stack.
>>> >
>>>
>>> Looks like this should instead be revert of following patches?
>>>
>>> drm/xe: Coalesce GGTT invalidations
>>> drm/xe: Pad GGTT mapping with an extra page pointing to scratch
>>>
>>
>> git revert conflicts...
>>
>
> Looks like a minor conflict.
> I am seeing reverting above patches brings back some XE_BUG_ON()
> checks and a xe_ggtt_invalidate() in __xe_ggtt_insert_bo_at().
> Wondering if we need them as well.

What is the exact bug? Might be not rejecting non-64k aligned pages.

~Maarten