When a GPU dma-fence signals, drivers often need to access userspace
memory from a kthread context — either to write a fence completion
value to a userspace VA (XE) or to signal a per-queue eventfd (AMDGPU).
Both require borrowing the process MM via kthread_use_mm(), and both
use the same kref-managed dma-fence-callback-to-workqueue pattern.

Extracting this pattern into a shared DRM helper allows both XE and
AMDGPU to use it instead of maintaining independent open-coded
implementations.

This series does that:

Patch 1 introduces drm_user_fence — an embeddable base structure with
driver-supplied worker/destroy callbacks. The common code handles:
 - process MM grab at init (mmgrab)
 - dma-fence callback registration
 - workqueue dispatch on fence signal
 - mmget_not_zero/kthread_use_mm/mmput in the worker
 - kref lifetime management
 - cancellation via drm_user_fence_cancel_sync() for safe driver teardown

Patch 2 converts XE to use the new helper. struct xe_user_fence embeds
struct drm_user_fence as its base. XE-specific fields (xe_device pointer
for ufence_wq wake-up, userspace VA, expected value, signalled flag)
remain in the wrapper. No behavioral change is intended.
A follow-on patch (not in this series) will wire AMDGPU's render-node
EOP eventfd signaling path to the same helper.

v3:
 - Store fence reference inside drm_user_fence during add_callback so
   drm_user_fence_cancel() can be called safely without the caller
   holding a separate fence reference, fixing a potential UAF when a
   foreign dma-fence signals after driver teardown. (Sashiko review)
 - Fix contradictory kernel-doc for drm_user_fence_cancel() which
   incorrectly instructed callers to put references that the function
   already puts internally. (Sashiko review)
 - Add drm_user_fence_cancel_sync() to guarantee the worker has fully
   completed before returning. Wire into xe_sync_entry_cleanup() and
   xe_vma_destroy_late() for safe XE teardown. (Sashiko review)
 - Add xe_sync_ufence_cancel_sync() wrapper in xe_sync.c so xe_vm.c
   does not need to include drm_user_fence.h or access struct
   xe_user_fence internals directly.
 - Add smp_wmb() in xe_ufence_worker() between WRITE_ONCE(signalled)
   and copy_to_user() to prevent store reordering on weakly-ordered
   architectures such as ARM64. (Sashiko review)
 - Replace XE_WARN_ON() on copy_to_user() failure with drm_dbg() to
   prevent unprivileged userspace from triggering a kernel warning by
   unmapping the user fence memory before it signals. (Sashiko review)
v2:
 - Move INIT_WORK() to drm_user_fence_init() fixing lockdep class
   divergence caused by two different INIT_WORK() call sites.
   (Sashiko review)
 - drm_user_fence_add_callback() now consumes the fence reference in
   all paths: transferred to the callback on success, put immediately
   on the already-signaled and error paths. (Sashiko review)
 - Add drm_user_fence_cancel() to allow drivers to safely detach a
   pending callback before context teardown. (Sashiko review)
 - Fix premature dma_fence_put() after drm_user_fence_add_callback()
   in xe_sync_entry_signal(). (Sashiko review)
 - Fix missing newline at end of new files.

Suggested-by: Christian König <[email protected]>
Cc: Alex Deucher <[email protected]>
Cc: Matthew Brost <[email protected]>
Cc: Mika Kuoppala <[email protected]>
Cc: Thomas Hellström <[email protected]>
Cc: Rodrigo Vivi <[email protected]>
Cc: Maarten Lankhorst <[email protected]>
Cc: Maxime Ripard <[email protected]>
Cc: Thomas Zimmermann <[email protected]>
Cc: David Airlie <[email protected]>
Cc: Simona Vetter <[email protected]>
Cc: Sumit Semwal <[email protected]>
Cc: [email protected]
Cc: [email protected]
Cc: [email protected]
Cc: [email protected]
Cc: [email protected]
Cc: [email protected]

Srinivasan Shanmugam (2):
  drm: Add common drm_user_fence helper
  drm/xe: Convert xe_user_fence to drm_user_fence

 drivers/gpu/drm/Makefile           |   1 +
 drivers/gpu/drm/drm_user_fence.c   | 210 +++++++++++++++++++++++++++++
 drivers/gpu/drm/xe/xe_sync.c       | 140 +++++++++----------
 drivers/gpu/drm/xe/xe_sync.h       |   1 +
 drivers/gpu/drm/xe/xe_sync_types.h |   3 +-
 drivers/gpu/drm/xe/xe_vm.c         |   1 +
 include/drm/drm_user_fence.h       |  76 +++++++++++
 7 files changed, 360 insertions(+), 72 deletions(-)
 create mode 100644 drivers/gpu/drm/drm_user_fence.c
 create mode 100644 include/drm/drm_user_fence.h

-- 
2.34.1

Reply via email to