re: drm/ttm: switch to per device LRU lock

From: Colin Ian King
Date: Thu Mar 25 2021 - 11:01:19 EST


Hi,

Static analysis with Coverity in linux-next has detected an issue in
drivers/gpu/drm/ttm/ttm_bo.c with the follow commit:

commit a1f091f8ef2b680a5184db065527612247cb4cae
Author: Christian König <christian.koenig@xxxxxxx>
Date: Tue Oct 6 17:26:42 2020 +0200

drm/ttm: switch to per device LRU lock

Instead of having a global lock for potentially less contention.


The analysis is as follows:

617 int ttm_mem_evict_first(struct ttm_device *bdev,
618 struct ttm_resource_manager *man,
619 const struct ttm_place *place,
620 struct ttm_operation_ctx *ctx,
621 struct ww_acquire_ctx *ticket)
622 {
1. assign_zero: Assigning: bo = NULL.

623 struct ttm_buffer_object *bo = NULL, *busy_bo = NULL;
624 bool locked = false;
625 unsigned i;
626 int ret;
627

Explicit null dereferenced (FORWARD_NULL)2. var_deref_op:
Dereferencing null pointer bo.

628 spin_lock(&bo->bdev->lru_lock);
629 for (i = 0; i < TTM_MAX_BO_PRIORITY; ++i) {

The spin_lock on bo is dereferencing a null bo pointer.

Colin