[Intel-gfx] [PATCH 2/3] drm/i915: Fix i915_vma_pin_iomap()

[Juha-Pekka Heikkila]

On 21.6.2022 13.53, Matthew Auld wrote:
> On Mon, 20 Jun 2022 at 10:38, Juha-Pekka Heikkila
> <juhapekka.heikkila at gmail.com> wrote:
>>
>> On 10.6.2022 20.43, Matthew Auld wrote:
>>> On Fri, 10 Jun 2022 at 15:53, Matthew Auld
>>> <matthew.william.auld at gmail.com> wrote:
>>>>
>>>> On Fri, 10 Jun 2022 at 13:12, Juha-Pekka Heikkila
>>>> <juhapekka.heikkila at gmail.com> wrote:
>>>>>
>>>>> From: CQ Tang <cq.tang at intel.com>
>>>>>
>>>>> Display might allocate a smem object and call
>>>>> i915_vma_pin_iomap(), the existing code will fail.
>>>>>
>>>>> This fix was suggested by Chris P Wilson, that we pin
>>>>> the smem with i915_gem_object_pin_map_unlocked().
>>>>>
>>>>> v2 (jheikkil): Change i915_gem_object_pin_map_unlocked to
>>>>>                  i915_gem_object_pin_map
>>>>>
>>>>> Signed-off-by: CQ Tang <cq.tang at intel.com>
>>>>> Signed-off-by: Juha-Pekka Heikkila <juhapekka.heikkila at gmail.com>
>>>>> Cc: Chris Wilson <chris.p.wilson at intel.com>
>>>>> Cc: Jari Tahvanainen <jari.tahvanainen at intel.com>
>>>> Reviewed-by: Matthew Auld <matthew.auld at intel.com>
>>>
>>> Although maybe consider putting this as patch 1, and then reword the
>>> commit title/message to be more like "drm/i915: extend
>>> i915_vma_iomap()" or so, which then becomes a prep patch for
>>> supporting the dpt fallback to smem. Otherwise it looks like this
>>> patch is basically just fixing the first patch to not trigger the
>>> WARN_ON(), which seems iffy IMO. Each patch by itself should ideally
>>> be functional.
>>
>> Probably easiest is to put patch 1 in as last, it's the final customer
>> for these two other patches. This way if someone will end up doing
>> bisecting there would be nothing interesting to see with these.
>>
>> I did finally get ci to look all green after last weeks outages. I'd
>> guess these patches are ready to be pushed but I don't have commit
>> rights to drm-tip. Are you able to help with these or I'll go bother
>> Imre or Jani to get them into tip?
> 
> Ok, if no objections I will go ahead and push this to
> drm-intel-gt-next, with the tweaked patch ordering.

No objections. I had this set yet on test run on Imre's wish on try-bot 
with forcing adlp (on bat) to use smem and results were all clean.

/Juha-Pekka