drm/xe/devcoredump: Defer devcoredump initialization during probe
authorBalasubramani Vivekanandan <balasubramani.vivekanandan@intel.com>
Fri, 1 Aug 2025 05:23:56 +0000 (10:53 +0530)
committerRodrigo Vivi <rodrigo.vivi@intel.com>
Mon, 4 Aug 2025 15:58:56 +0000 (11:58 -0400)
commit465f1dba74c010995190bff267ae5a75afcdcfea
tree65503af931a8bec770e42f4aa3c66793cddb674d
parentdf9bdd4381be100a778c63a5dd810dc82b8c2705
drm/xe/devcoredump: Defer devcoredump initialization during probe

Doing devcoredump initializing before GT though look harmless, it leads
to problem during driver unbind. Because of this order, GT/Engine
release functions will be called before xe devcoredump release function
(xe_driver_devcoredump_fini) leading to the following kernel crash[1]
because the devcoredump functions might still use GT/Engine
datastructures after those are freed.

The following crash is observed while running the IGT
xe_wedged@wedged-at-any-timeout. The test forces a wedged state by
submitting a workload which hangs. Then does a unbind/rebind of the
driver to recover from the wedged state.
The hanged workload leads to a devcoredump. The following crash is
noticed when the devcoredump capture races with the driver unbind.
During driver unbind, the release function hw_engine_fini() will be
called which assigns NULL to hwe->gt. But the same data structure is
accessed during the coredump capture in the function
xe_engine_snapshot_print by reading snapshot->hwe->gt.

With this patch, we make sure the devcoredump is stopped before
deinitializing the core driver functions.

[1]:
BUG: kernel NULL pointer dereference, address: 0000000000000000
Workqueue: events_unbound xe_devcoredump_deferred_snap_work [xe]
RIP: 0010:xe_engine_snapshot_print+0x47/0x420 [xe]
Call Trace:
 <TASK>
 ? drm_printf+0x64/0x90
 __xe_devcoredump_read+0x23f/0x2d0 [xe]
 ? __pfx___drm_printfn_coredump+0x10/0x10
 ? __pfx___drm_puts_coredump+0x10/0x10
 xe_devcoredump_deferred_snap_work+0x17a/0x190 [xe]
 process_one_work+0x22e/0x6f0
 worker_thread+0x1e8/0x3d0
 ? __pfx_worker_thread+0x10/0x10
 kthread+0x11f/0x250
 ? __pfx_kthread+0x10/0x10
 ret_from_fork+0x47/0x70
 ? __pfx_kthread+0x10/0x10
 ret_from_fork_asm+0x1a/0x30

v2: Detailed commit description (Rodrigo)
v3: FIXME added (Rodrigo, Stuart)

Fixes: 4209d635a823 ("drm/xe: Remove devcoredump during driver release")
Reviewed-by: Stuart Summers <stuart.summers@intel.com>
Link: https://lore.kernel.org/r/20250731061300.14320-1-balasubramani.vivekanandan@intel.com
Signed-off-by: Balasubramani Vivekanandan <balasubramani.vivekanandan@intel.com>
Link: https://lore.kernel.org/r/20250801052356.21885-1-balasubramani.vivekanandan@intel.com
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
(cherry picked from commit 1fdc4c381ff765479d76ccf3134717c430c871b8)
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
drivers/gpu/drm/xe/xe_device.c
drivers/gpu/drm/xe/xe_guc_capture.c