4 files changed, 47 insertions, 17 deletions

diff --git a/drivers/gpu/drm/amd/scheduler/gpu_sched_trace.h b/drivers/gpu/drm/amd/scheduler/gpu_sched_trace.h
index 144f50acc971..c89dc777768f 100644
--- a/drivers/gpu/drm/amd/scheduler/gpu_sched_trace.h
+++ b/drivers/gpu/drm/amd/scheduler/gpu_sched_trace.h
@@ -16,6 +16,8 @@ TRACE_EVENT(amd_sched_job,
             TP_ARGS(sched_job),
             TP_STRUCT__entry(
                              __field(struct amd_sched_entity *, entity)
+                             __field(struct amd_sched_job *, sched_job)
+                             __field(struct fence *, fence)
                              __field(const char *, name)
                              __field(u32, job_count)
                              __field(int, hw_job_count)
@@ -23,16 +25,32 @@ TRACE_EVENT(amd_sched_job,
 
             TP_fast_assign(
                            __entry->entity = sched_job->s_entity;
+                           __entry->sched_job = sched_job;
+                           __entry->fence = &sched_job->s_fence->base;
                            __entry->name = sched_job->sched->name;
                            __entry->job_count = kfifo_len(
                                    &sched_job->s_entity->job_queue) / sizeof(sched_job);
                            __entry->hw_job_count = atomic_read(
                                    &sched_job->sched->hw_rq_count);
                            ),
-            TP_printk("entity=%p, ring=%s, job count:%u, hw job count:%d",
-                      __entry->entity, __entry->name, __entry->job_count,
-                      __entry->hw_job_count)
+            TP_printk("entity=%p, sched job=%p, fence=%p, ring=%s, job count:%u, hw job count:%d",
+                      __entry->entity, __entry->sched_job, __entry->fence, __entry->name,
+                      __entry->job_count, __entry->hw_job_count)
 );
+
+TRACE_EVENT(amd_sched_process_job,
+            TP_PROTO(struct amd_sched_fence *fence),
+            TP_ARGS(fence),
+            TP_STRUCT__entry(
+                    __field(struct fence *, fence)
+                    ),
+
+            TP_fast_assign(
+                    __entry->fence = &fence->base;
+                    ),
+            TP_printk("fence=%p signaled", __entry->fence)
+);
+
 #endif
 
 /* This part must be outside protection */
diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c
index 89619a5a4289..ea30d6ad4c13 100644
--- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c
+++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c
@@ -34,6 +34,9 @@ static struct amd_sched_job *
 amd_sched_entity_pop_job(struct amd_sched_entity *entity);
 static void amd_sched_wakeup(struct amd_gpu_scheduler *sched);
 
+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)
 {
@@ -273,22 +276,13 @@ static bool amd_sched_entity_in(struct amd_sched_job *sched_job)
  *
  * Returns 0 for success, negative error code otherwise.
  */
-int amd_sched_entity_push_job(struct amd_sched_job *sched_job)
+void amd_sched_entity_push_job(struct amd_sched_job *sched_job)
 {
         struct amd_sched_entity *entity = sched_job->s_entity;
-        struct amd_sched_fence *fence = amd_sched_fence_create(
-                entity, sched_job->owner);
-
-        if (!fence)
-                return -ENOMEM;
-
-        fence_get(&fence->base);
-        sched_job->s_fence = fence;
 
         wait_event(entity->sched->job_scheduled,
                    amd_sched_entity_in(sched_job));
         trace_amd_sched_job(sched_job);
-        return 0;
 }
 
 /**
@@ -343,6 +337,7 @@ static void amd_sched_process_job(struct fence *f, struct fence_cb *cb)
                 list_del_init(&s_fence->list);
                 spin_unlock_irqrestore(&sched->fence_list_lock, flags);
         }
+        trace_amd_sched_process_job(s_fence);
         fence_put(&s_fence->base);
         wake_up_interruptible(&sched->wake_up_worker);
 }
@@ -450,6 +445,13 @@ int amd_sched_init(struct amd_gpu_scheduler *sched,
         init_waitqueue_head(&sched->wake_up_worker);
         init_waitqueue_head(&sched->job_scheduled);
         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);
@@ -470,4 +472,6 @@ void amd_sched_fini(struct amd_gpu_scheduler *sched)
 {
         if (sched->thread)
                 kthread_stop(sched->thread);
+        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 929e9aced041..939692b14f4b 100644
--- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h
+++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h
@@ -30,6 +30,9 @@
 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 
@@ -76,7 +79,6 @@ struct amd_sched_job {
         struct amd_gpu_scheduler        *sched;
         struct amd_sched_entity         *s_entity;
         struct amd_sched_fence          *s_fence;
-        void                            *owner;
 };
 
 extern const struct fence_ops amd_sched_fence_ops;
@@ -128,7 +130,7 @@ int amd_sched_entity_init(struct amd_gpu_scheduler *sched,
                           uint32_t jobs);
 void amd_sched_entity_fini(struct amd_gpu_scheduler *sched,
                            struct amd_sched_entity *entity);
-int amd_sched_entity_push_job(struct amd_sched_job *sched_job);
+void amd_sched_entity_push_job(struct amd_sched_job *sched_job);
 
 struct amd_sched_fence *amd_sched_fence_create(
         struct amd_sched_entity *s_entity, void *owner);
diff --git a/drivers/gpu/drm/amd/scheduler/sched_fence.c b/drivers/gpu/drm/amd/scheduler/sched_fence.c
index d802638094f4..8d2130b9ff05 100644
--- a/drivers/gpu/drm/amd/scheduler/sched_fence.c
+++ b/drivers/gpu/drm/amd/scheduler/sched_fence.c
@@ -32,7 +32,7 @@ struct amd_sched_fence *amd_sched_fence_create(struct amd_sched_entity *s_entity
         struct amd_sched_fence *fence = NULL;
         unsigned seq;
 
-        fence = kzalloc(sizeof(struct amd_sched_fence), GFP_KERNEL);
+        fence = kmem_cache_zalloc(sched_fence_slab, GFP_KERNEL);
         if (fence == NULL)
                 return NULL;
         fence->owner = owner;
@@ -71,11 +71,17 @@ static bool amd_sched_fence_enable_signaling(struct fence *f)
         return true;
 }
 
+static void amd_sched_fence_release(struct fence *f)
+{
+        struct amd_sched_fence *fence = to_amd_sched_fence(f);
+        kmem_cache_free(sched_fence_slab, fence);
+}
+
 const struct fence_ops amd_sched_fence_ops = {
         .get_driver_name = amd_sched_fence_get_driver_name,
         .get_timeline_name = amd_sched_fence_get_timeline_name,
         .enable_signaling = amd_sched_fence_enable_signaling,
         .signaled = NULL,
         .wait = fence_default_wait,
-        .release = NULL,
+        .release = amd_sched_fence_release,
 };