drm/xe/bo: sync kernel fences for KMD buffers
authorMatthew Auld <matthew.auld@intel.com>
Wed, 25 Oct 2023 17:39:41 +0000 (18:39 +0100)
committerRodrigo Vivi <rodrigo.vivi@intel.com>
Thu, 21 Dec 2023 16:43:32 +0000 (11:43 -0500)
commit503a6f4e4f961acbbcac8d36f51226f3d3cfa7b7
tree58c2e911ae6c512ec0271c4eb01d01fb0047fc92
parenta667cf56dbd64e35f8f34ec47549888fa28878fb
drm/xe/bo: sync kernel fences for KMD buffers

With things like pipelined evictions, VRAM pages can be marked as free
and yet still have some active kernel fences, with the idea that the
next caller to allocate the memory will respect them. However it looks
like we are missing synchronisation for KMD internal buffers, like
page-tables, lrc etc. For userspace objects we should already have the
required synchronisation for CPU access via the fault handler, and
likewise for GPU access when vm_binding them.

To fix this synchronise against any kernel fences for all KMD objects at
creation. This should resolve some severe corruption seen during
evictions.

v2 (Matt B):
  - Revamp the comment explaining this. Also mention why USAGE_KERNEL is
    correct here.
v3 (Thomas):
  - Make sure to use ctx.interruptible for the wait.

Testcase: igt@xe-evict-ccs
Closes: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/853
Closes: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/855
Reported-by: Zbigniew Kempczyński <zbigniew.kempczynski@intel.com>
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Tested-by: Zbigniew Kempczyński <zbigniew.kempczynski@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
drivers/gpu/drm/xe/xe_bo.c