need to learn this.
When set_tiling fails, keep tiling type at the type we already had, do
NOT use the parameters that we passed to the ioctl.
Noticed when i botched a kernel diff (locally) so that that ioctl failed
more often.
ok matthieu@
This diff seems larger than it is since upstream reindented some
sources. This updates libdrm_intel to -current (which only affect the
GEM code) and libdrm to 2.4.15. bumping both minor versions.
Discussed with matthieu@
We always map through the gtt, never using the cpu mapping (this decreases
complexity a lot). Keep use of the original ioctl for this. Also don't bother
with userland domain setting, the kernel handles this correctly.
Finally, don't persist mapping across calls to unmap and remapping, this aids
with debugging, but may change later.
ok matthieu@.
are to some assertions and a small change to modesetting code).
bump libdrm_intel minor due to added symbols, libdrm doesn't get bumped,
no change to symbol list.
ok matthieu@.
This is needed for us to be able to update Mesa and xf86-video-intel.
Includes a few fixes, as well as the intel bufmgr interface and the
modesetting interface (which currently errors when you check if it's
enabled).
ok matthieu@
