[F][PATCH 0/2] Fixes for memory corruption and a deadlock in i915

[Sultan Alsawaf]

In 5.4, i915 suffers from a mutex deadlock that occurs when kswapd runs i915's
shrinker to free memory. This was fixed in 5.5, but the 5.5 fix is too invasive
to backport (see the commit message for details).

Another bug, memory corruption caused by a race in i915, affects 5.4+ and has
not been fixed by upstream. On 5.4, this crash causes a kernel panic, but on
5.5+ the memory corruption from the race doesn't cause an immediate panic.

The attached fixes for these bugs were submitted to LKML [1][2] and will
hopefully be addressed soon. In the meantime, let's pick them into Focal to fix
the crashes by release time.

Note that the attached "drm/i915: Synchronize active and retire callbacks"
patch varies from the upstream submission for it because it is the backport of
the fix to 5.4.

[1] https://lore.kernel.org/stable/20200407203222.2493-1-sultan@kerneltoast.com
[2] https://lore.kernel.org/stable/20200407064007.7599-1-sultan@kerneltoast.com

Sultan

Sultan Alsawaf (2):
  drm/i915: Fix ref->mutex deadlock in i915_active_wait()
  drm/i915: Synchronize active and retire callbacks

 drivers/gpu/drm/i915/i915_active.c       | 77 ++++++++++++++++++------
 drivers/gpu/drm/i915/i915_active.h       | 14 ++---
 drivers/gpu/drm/i915/i915_active_types.h |  3 +
 3 files changed, 69 insertions(+), 25 deletions(-)

-- 
2.20.1