drm/amd: fix scheduler fence teardown order v2

Some fences might be alive even after we have stopped the scheduler leading
to warnings about leaked objects from the SLUB allocator.

Fix this by allocating/freeing the SLUB allocator from the module
init/fini functions just like we do it for hw fences.

v2: make variable static, add link to bug

Fixes: https://bugs.freedesktop.org/show_bug.cgi?id=97500

Reported-by: Grazvydas Ignotas <notasas@gmail.com>
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com> (v1)
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org

4 files changed, 24 insertions, 16 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
index 71ed27eb3dde..73f2415630f8 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
@@ -737,6 +737,7 @@ static int __init amdgpu_init(void)
 {
         amdgpu_sync_init();
         amdgpu_fence_slab_init();
+        amd_sched_fence_slab_init();
         if (vgacon_text_force()) {
                 DRM_ERROR("VGACON disables amdgpu kernel modesetting.\n");
                 return -EINVAL;
@@ -756,6 +757,7 @@ static void __exit amdgpu_exit(void)
         drm_pci_exit(driver, pdriver);
         amdgpu_unregister_atpx_handler();
         amdgpu_sync_fini();
+        amd_sched_fence_slab_fini();
         amdgpu_fence_slab_fini();
 }
 
diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c
index 910b8d5b21c5..ffe1f85ce300 100644
--- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c
+++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c
@@ -34,9 +34,6 @@ static bool amd_sched_entity_is_ready(struct amd_sched_entity *entity);
 static void amd_sched_wakeup(struct amd_gpu_scheduler *sched);
 static void amd_sched_process_job(struct fence *f, struct fence_cb *cb);
 
-struct kmem_cache *sched_fence_slab;
-atomic_t sched_fence_slab_ref = ATOMIC_INIT(0);
-
 /* Initialize a given run queue struct */
 static void amd_sched_rq_init(struct amd_sched_rq *rq)
 {
@@ -618,13 +615,6 @@ int amd_sched_init(struct amd_gpu_scheduler *sched,
         INIT_LIST_HEAD(&sched->ring_mirror_list);
         spin_lock_init(&sched->job_list_lock);
         atomic_set(&sched->hw_rq_count, 0);
-        if (atomic_inc_return(&sched_fence_slab_ref) == 1) {
-                sched_fence_slab = kmem_cache_create(
-                        "amd_sched_fence", sizeof(struct amd_sched_fence), 0,
-                        SLAB_HWCACHE_ALIGN, NULL);
-                if (!sched_fence_slab)
-                        return -ENOMEM;
-        }
 
         /* Each scheduler will run on a seperate kernel thread */
         sched->thread = kthread_run(amd_sched_main, sched, sched->name);
@@ -645,7 +635,4 @@ void amd_sched_fini(struct amd_gpu_scheduler *sched)
 {
         if (sched->thread)
                 kthread_stop(sched->thread);
-        rcu_barrier();
-        if (atomic_dec_and_test(&sched_fence_slab_ref))
-                kmem_cache_destroy(sched_fence_slab);
 }
diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h
index 7cbbbfb502ef..51068e6c3d9a 100644
--- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h
+++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h
@@ -30,9 +30,6 @@
 struct amd_gpu_scheduler;
 struct amd_sched_rq;
 
-extern struct kmem_cache *sched_fence_slab;
-extern atomic_t sched_fence_slab_ref;
-
 /**
  * A scheduler entity is a wrapper around a job queue or a group
  * of other entities. Entities take turns emitting jobs from their
@@ -145,6 +142,9 @@ void amd_sched_entity_fini(struct amd_gpu_scheduler *sched,
                            struct amd_sched_entity *entity);
 void amd_sched_entity_push_job(struct amd_sched_job *sched_job);
 
+int amd_sched_fence_slab_init(void);
+void amd_sched_fence_slab_fini(void);
+
 struct amd_sched_fence *amd_sched_fence_create(
         struct amd_sched_entity *s_entity, void *owner);
 void amd_sched_fence_scheduled(struct amd_sched_fence *fence);
diff --git a/drivers/gpu/drm/amd/scheduler/sched_fence.c b/drivers/gpu/drm/amd/scheduler/sched_fence.c
index 3653b5a40494..88fc2d662579 100644
--- a/drivers/gpu/drm/amd/scheduler/sched_fence.c
+++ b/drivers/gpu/drm/amd/scheduler/sched_fence.c
@@ -27,6 +27,25 @@
 #include <drm/drmP.h>
 #include "gpu_scheduler.h"
 
+static struct kmem_cache *sched_fence_slab;
+
+int amd_sched_fence_slab_init(void)
+{
+        sched_fence_slab = kmem_cache_create(
+                "amd_sched_fence", sizeof(struct amd_sched_fence), 0,
+                SLAB_HWCACHE_ALIGN, NULL);
+        if (!sched_fence_slab)
+                return -ENOMEM;
+
+        return 0;
+}
+
+void amd_sched_fence_slab_fini(void)
+{
+        rcu_barrier();
+        kmem_cache_destroy(sched_fence_slab);
+}
+
 struct amd_sched_fence *amd_sched_fence_create(struct amd_sched_entity *entity,
                                                void *owner)
 {