[Mesa-dev] [PATCH v2] intel/ppgtt: memory address alignment

[Lionel Landwerlin]

On 24/07/18 18:34, Kenneth Graunke wrote:
> On Tuesday, July 24, 2018 5:34:57 AM PDT Lionel Landwerlin wrote:
>> That looks correct to me (and we do the same in Anv).
>> Also a bit baffled that we haven't run into issues earlier :(
>>
>> But would be good to have Ken's Rb too.
>>
>> Thanks a lot!
>>
>> Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin at intel.com>
> Yeah, this is probably for the best...we used to just ask the kernel
> for memory and it would do this for us.  Now that we're doing it
> ourselves, we ought to be defensive here.
>
> Reviewed-by: Kenneth Graunke <kenneth at whitecape.org>
>
> But I agree with Chris, I'm surprised that this would actually fix
> anything...all of our allocations ought to multiples of PAGE_SIZE,
> so unless we're starting at a funny address, they ought to remain
> that way...
>
> I wonder if this isn't papering over another bug.

Sergii,

If you can reproduce this bug locally, would you mind adding

  assert(size % 4096 == 0);

at the top of vma_alloc() and see if it hits the asserts.
Having a backtrace would be great :)

Thanks a lot,

-
Lionel