The entries were only initialized once in amdgpu_sa_bo_new. If a fence
wasn't signalled yet in the first amdgpu_sa_bo_next_hole call, but then
got signalled before a later amdgpu_sa_bo_next_hole call, it could
destroy the fence but leave its pointer in the array, resulting in
use-after-free in amdgpu_sa_bo_new.
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Michel Dänzer <michel.daenzer@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
        for (i = 0; i < AMDGPU_SA_NUM_FENCE_LISTS; ++i) {
                struct amdgpu_sa_bo *sa_bo;
 
+               fences[i] = NULL;
+
                if (list_empty(&sa_manager->flist[i]))
                        continue;
 
 
        spin_lock(&sa_manager->wq.lock);
        do {
-               for (i = 0; i < AMDGPU_SA_NUM_FENCE_LISTS; ++i) {
-                       fences[i] = NULL;
+               for (i = 0; i < AMDGPU_SA_NUM_FENCE_LISTS; ++i)
                        tries[i] = 0;
-               }
 
                do {
                        amdgpu_sa_bo_try_free(sa_manager);