[Intel-gfx] [PATCH cont'd] intel: non-blocking mmaps on the cheap

[Ben Widawsky]

The drm patch goes on top of Daniel's patch. I'll squash before pushing,
it's just separate for review.

The kernel stuff is straight forward, just an IOCTL to query the cache
level.

Both patches will get better commit messages when resubmitted.

I ran a modified version of the test I posted earlier.
On Sandybridge I see no improvements on the test I posted earlier.
On Ironlake, improvement over regular GTT mappings:
21.7486% +/- 1.17445%

These results are similar to my earlier patch series.

Overall, I like this series much better. Do we want to sneak in a set
cache level with the same IOCTL while we're at it?

Ben