With the current cleanup flow, we could trigger a NULL pointer dereference if there is a delayed destruction of a BO with a system resource that gets executed on drain_workqueue() call, as we attempt to free a resource using an already released resource manager. Remove the device from the device list and drain its workqueue before releasing the system domain manager in ttm_device_fini(). Signed-off-by: Karolina Stolarek <karolina.stolarek@xxxxxxxxx> --- This is actually a reiteration of a patch sent in [1], but the solution and commit message changed significantly, so I decided not to send it as v2. [1] - 20231013143423.1503088-1-karolina.stolarek@xxxxxxxxx drivers/gpu/drm/ttm/ttm_device.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/drivers/gpu/drm/ttm/ttm_device.c b/drivers/gpu/drm/ttm/ttm_device.c index 7726a72befc5..d48b39132b32 100644 --- a/drivers/gpu/drm/ttm/ttm_device.c +++ b/drivers/gpu/drm/ttm/ttm_device.c @@ -232,10 +232,6 @@ void ttm_device_fini(struct ttm_device *bdev) struct ttm_resource_manager *man; unsigned i; - man = ttm_manager_type(bdev, TTM_PL_SYSTEM); - ttm_resource_manager_set_used(man, false); - ttm_set_driver_manager(bdev, TTM_PL_SYSTEM, NULL); - mutex_lock(&ttm_global_mutex); list_del(&bdev->device_list); mutex_unlock(&ttm_global_mutex); @@ -243,6 +239,10 @@ void ttm_device_fini(struct ttm_device *bdev) drain_workqueue(bdev->wq); destroy_workqueue(bdev->wq); + man = ttm_manager_type(bdev, TTM_PL_SYSTEM); + ttm_resource_manager_set_used(man, false); + ttm_set_driver_manager(bdev, TTM_PL_SYSTEM, NULL); + spin_lock(&bdev->lru_lock); for (i = 0; i < TTM_MAX_BO_PRIORITY; ++i) if (list_empty(&man->lru[0])) -- 2.25.1
