[PATCH 0/2] Deny unbinds if user fence pending

[Mika Kuoppala <mika.kuoppala@linux.intel.com>]

If we ever want to intercept OP_MAP through
user fence, we need to make sure that sequences
of OP_MAP and OP_UNMAP can be tracked without them
coalescing and thus vanishing.

Enforce vm_bind_ioctl by enforcing wait on associated
user fences on bind, before unbinding: If there is
non signalled user fence pending when the unmap is
attempted, deny it with -EBUSY.

[1] states that -EINVAL should be returned if
the ufence was part of the previous OP_MAP inside
this ioctl. This can be added if there is need to
distinguish these apart. This series returns -EBUSY 
regardless if the OP_MAP was part of this or previous
ioctl.

[1] https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/1159

Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: Jani Nikula <jani.nikula@linux.intel.com>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>

Mika Kuoppala (2):
  drm/xe: Expose user fence from xe_sync_entry
  drm/xe: Deny unbinds if uapi ufence pending

 drivers/gpu/drm/xe/xe_sync.c       | 56 +++++++++++++++++++++++++-----
 drivers/gpu/drm/xe/xe_sync.h       |  4 +++
 drivers/gpu/drm/xe/xe_sync_types.h |  2 +-
 drivers/gpu/drm/xe/xe_vm.c         | 35 +++++++++++++++++++
 drivers/gpu/drm/xe/xe_vm_types.h   |  7 ++++
 5 files changed, 94 insertions(+), 10 deletions(-)

-- 
2.34.1