[Intel-gfx] [PATCH 00/23] huge gtt pages

[Chris Wilson]

Quoting Chris Wilson (2017-08-22 15:21:10)
> Quoting Matthew Auld (2017-08-21 19:34:40)
> > Some more improvements as per Chris' comments.
> > 
> > Matthew Auld (23):
> >   mm/shmem: introduce shmem_file_setup_with_mnt
> >   drm/i915: introduce simple gemfs
> >   drm/i915/gemfs: enable THP
> >   drm/i915: introduce page_size_mask to dev_info
> >   drm/i915: push set_pages down to the callers
> >   drm/i915: introduce page_size members
> >   drm/i915: introduce vm set_pages/clear_pages
> >   drm/i915: align the vma start to the largest gtt page size
> >   drm/i915: align 64K objects to 2M
> >   drm/i915: enable IPS bit for 64K pages
> >   drm/i915: disable GTT cache for 2M/1G pages
> >   drm/i915: support 1G pages for the 48b PPGTT
> >   drm/i915: support 2M pages for the 48b PPGTT
> >   drm/i915: add support for 64K scratch page
> >   drm/i915: support 64K pages for the 48b PPGTT
> >   drm/i915: accurate page size tracking for the ppgtt
> >   drm/i915/debugfs: include some gtt page size metrics
> >   drm/i915/selftests: huge page tests
> >   drm/i915/selftests: mix huge pages
> >   drm/i915: disable platform support for vGPU huge gtt pages
> >   drm/i915: enable platform support for 64K pages
> >   drm/i915: enable platform support for 2M pages
> >   drm/i915: enable platform support for 1G pages
> 
> Ok, looking good. I keep pushing off the selftest review, sorry. I did
> start going through it and felt that it could do with a lot more
> permutations of pages sizes within an object to try and exercise the
> different boundaries and levels. Everything else looks excellent, I just
> haven't yet put together my list of questions for the st...

And the other pencilled in suggestion I have is that you grab the
THP/memfd work and put together a userptr igt that can allocate and
utilize huge pages. For that we will need a I915_PARAM to report the
supported pages sizes (and then have to mask that against the THP
supported pages sizes).
-Chris