drm/v3d: Address race-condition between per-fd GPU stats and fd release

drivers/gpu/drm/v3d/v3d_drv.c

@@ -150,12 +150,24 @@ v3d_open(struct drm_device *dev, struct drm_file *file)
 static void
 v3d_postclose(struct drm_device *dev, struct drm_file *file)
 {
+	struct v3d_dev *v3d = to_v3d_dev(dev);
 	struct v3d_file_priv *v3d_priv = file->driver_priv;
+	unsigned long irqflags;
 	enum v3d_queue q;
 
-	for (q = 0; q < V3D_MAX_QUEUES; q++)
+	for (q = 0; q < V3D_MAX_QUEUES; q++) {
+		struct v3d_queue_state *queue = &v3d->queue[q];
+		struct v3d_job *job = queue->active_job;
+
 		drm_sched_entity_destroy(&v3d_priv->sched_entity[q]);
 
+		if (job && job->base.entity == &v3d_priv->sched_entity[q]) {
+			spin_lock_irqsave(&queue->queue_lock, irqflags);
+			job->file_priv = NULL;
+			spin_unlock_irqrestore(&queue->queue_lock, irqflags);
+		}
+	}
+
 	v3d_perfmon_close_file(v3d_priv);
 	kfree(v3d_priv);
 }

drivers/gpu/drm/v3d/v3d_drv.h

@@ -58,6 +58,10 @@ struct v3d_queue_state {
 
 	/* Stores the GPU stats for this queue in the global context. */
 	struct v3d_stats stats;
+
+	/* Currently active job for this queue */
+	struct v3d_job *active_job;
+	spinlock_t queue_lock;
 };
 
 /* Performance monitor object. The perform lifetime is controlled by userspace
@@ -165,18 +169,8 @@ struct v3d_dev {
 
 	struct work_struct overflow_mem_work;
 
-	struct v3d_bin_job *bin_job;
-	struct v3d_render_job *render_job;
-	struct v3d_tfu_job *tfu_job;
-	struct v3d_csd_job *csd_job;
-
 	struct v3d_queue_state queue[V3D_MAX_QUEUES];
 
-	/* Spinlock used to synchronize the overflow memory
-	 * management against bin job submission.
-	 */
-	spinlock_t job_lock;
-
 	/* Used to track the active perfmon if any. */
 	struct v3d_perfmon *active_perfmon;
 
@@ -322,9 +316,9 @@ struct v3d_job {
 	struct v3d_perfmon *perfmon;
 
 	/* File descriptor of the process that submitted the job that could be used
-	 * for collecting stats by process of GPU usage.
+	 * to collect per-process information about the GPU.
 	 */
-	struct drm_file *file;
+	struct v3d_file_priv *file_priv;
 
 	/* Callback for the freeing of the job on refcount going to 0. */
 	void (*free)(struct kref *ref);
@@ -565,7 +559,7 @@ void v3d_get_stats(const struct v3d_stats *stats, u64 timestamp,
 
 /* v3d_fence.c */
 extern const struct dma_fence_ops v3d_fence_ops;
-struct dma_fence *v3d_fence_create(struct v3d_dev *v3d, enum v3d_queue queue);
+struct dma_fence *v3d_fence_create(struct v3d_dev *v3d, enum v3d_queue q);
 
 /* v3d_gem.c */
 int v3d_gem_init(struct drm_device *dev);
@@ -609,7 +603,7 @@ void v3d_timestamp_query_info_free(struct v3d_timestamp_query_info *query_info,
 				   unsigned int count);
 void v3d_performance_query_info_free(struct v3d_performance_query_info *query_info,
 				     unsigned int count);
-void v3d_job_update_stats(struct v3d_job *job, enum v3d_queue queue);
+void v3d_job_update_stats(struct v3d_job *job, enum v3d_queue q);
 int v3d_sched_init(struct v3d_dev *v3d);
 void v3d_sched_fini(struct v3d_dev *v3d);
 

drivers/gpu/drm/v3d/v3d_fence.c

@@ -3,19 +3,20 @@
 
 #include "v3d_drv.h"
 
-struct dma_fence *v3d_fence_create(struct v3d_dev *v3d, enum v3d_queue queue)
+struct dma_fence *v3d_fence_create(struct v3d_dev *v3d, enum v3d_queue q)
 {
+	struct v3d_queue_state *queue = &v3d->queue[q];
 	struct v3d_fence *fence;
 
 	fence = kzalloc(sizeof(*fence), GFP_KERNEL);
 	if (!fence)
 		return ERR_PTR(-ENOMEM);
 
 	fence->dev = &v3d->drm;
-	fence->queue = queue;
-	fence->seqno = ++v3d->queue[queue].emit_seqno;
-	dma_fence_init(&fence->base, &v3d_fence_ops, &v3d->job_lock,
-		       v3d->queue[queue].fence_context, fence->seqno);
+	fence->queue = q;
+	fence->seqno = ++queue->emit_seqno;
+	dma_fence_init(&fence->base, &v3d_fence_ops, &queue->queue_lock,
+		       queue->fence_context, fence->seqno);
 
 	return &fence->base;
 }

drivers/gpu/drm/v3d/v3d_gem.c

@@ -272,10 +272,11 @@ v3d_gem_init(struct drm_device *dev)
 		queue->fence_context = dma_fence_context_alloc(1);
 		memset(&queue->stats, 0, sizeof(queue->stats));
 		seqcount_init(&queue->stats.lock);
+
+		spin_lock_init(&queue->queue_lock);
 	}
 
 	spin_lock_init(&v3d->mm_lock);
-	spin_lock_init(&v3d->job_lock);
 	ret = drmm_mutex_init(dev, &v3d->bo_lock);
 	if (ret)
 		return ret;
@@ -327,17 +328,16 @@ void
 v3d_gem_destroy(struct drm_device *dev)
 {
 	struct v3d_dev *v3d = to_v3d_dev(dev);
+	enum v3d_queue q;
 
 	v3d_sched_fini(v3d);
 	v3d_gemfs_fini(v3d);
 
 	/* Waiting for jobs to finish would need to be done before
 	 * unregistering V3D.
 	 */
-	WARN_ON(v3d->bin_job);
-	WARN_ON(v3d->render_job);
-	WARN_ON(v3d->tfu_job);
-	WARN_ON(v3d->csd_job);
+	for (q = 0; q < V3D_MAX_QUEUES; q++)
+		WARN_ON(v3d->queue[q].active_job);
 
 	drm_mm_takedown(&v3d->mm);
 

drivers/gpu/drm/v3d/v3d_irq.c

@@ -42,6 +42,8 @@ v3d_overflow_mem_work(struct work_struct *work)
 		container_of(work, struct v3d_dev, overflow_mem_work);
 	struct drm_device *dev = &v3d->drm;
 	struct v3d_bo *bo = v3d_bo_create(dev, NULL /* XXX: GMP */, 256 * 1024);
+	struct v3d_queue_state *queue = &v3d->queue[V3D_BIN];
+	struct v3d_bin_job *bin_job;
 	struct drm_gem_object *obj;
 	unsigned long irqflags;
 
@@ -60,15 +62,17 @@ v3d_overflow_mem_work(struct work_struct *work)
 	 * bin job got scheduled, that's fine.  We'll just give them
 	 * some binner pool anyway.
 	 */
-	spin_lock_irqsave(&v3d->job_lock, irqflags);
-	if (!v3d->bin_job) {
-		spin_unlock_irqrestore(&v3d->job_lock, irqflags);
+	spin_lock_irqsave(&queue->queue_lock, irqflags);
+	bin_job = (struct v3d_bin_job *)queue->active_job;
+
+	if (!bin_job) {
+		spin_unlock_irqrestore(&queue->queue_lock, irqflags);
 		goto out;
 	}
 
 	drm_gem_object_get(obj);
-	list_add_tail(&bo->unref_head, &v3d->bin_job->render->unref_list);
-	spin_unlock_irqrestore(&v3d->job_lock, irqflags);
+	list_add_tail(&bo->unref_head, &bin_job->render->unref_list);
+	spin_unlock_irqrestore(&queue->queue_lock, irqflags);
 
 	v3d_mmu_flush_all(v3d);
 
@@ -79,6 +83,20 @@ v3d_overflow_mem_work(struct work_struct *work)
 	drm_gem_object_put(obj);
 }
 
+static void
+v3d_irq_signal_fence(struct v3d_dev *v3d, enum v3d_queue q,
+		     void (*trace_irq)(struct drm_device *, uint64_t))
+{
+	struct v3d_queue_state *queue = &v3d->queue[q];
+	struct v3d_fence *fence = to_v3d_fence(queue->active_job->irq_fence);
+
+	v3d_job_update_stats(queue->active_job, q);
+	trace_irq(&v3d->drm, fence->seqno);
+
+	queue->active_job = NULL;
+	dma_fence_signal(&fence->base);
+}
+
 static irqreturn_t
 v3d_irq(int irq, void *arg)
 {
@@ -102,41 +120,17 @@ v3d_irq(int irq, void *arg)
 	}
 
 	if (intsts & V3D_INT_FLDONE) {
-		struct v3d_fence *fence =
-			to_v3d_fence(v3d->bin_job->base.irq_fence);
-
-		v3d_job_update_stats(&v3d->bin_job->base, V3D_BIN);
-		trace_v3d_bcl_irq(&v3d->drm, fence->seqno);
-
-		v3d->bin_job = NULL;
-		dma_fence_signal(&fence->base);
-
+		v3d_irq_signal_fence(v3d, V3D_BIN, trace_v3d_bcl_irq);
 		status = IRQ_HANDLED;
 	}
 
 	if (intsts & V3D_INT_FRDONE) {
-		struct v3d_fence *fence =
-			to_v3d_fence(v3d->render_job->base.irq_fence);
-
-		v3d_job_update_stats(&v3d->render_job->base, V3D_RENDER);
-		trace_v3d_rcl_irq(&v3d->drm, fence->seqno);
-
-		v3d->render_job = NULL;
-		dma_fence_signal(&fence->base);
-
+		v3d_irq_signal_fence(v3d, V3D_RENDER, trace_v3d_rcl_irq);
 		status = IRQ_HANDLED;
 	}
 
 	if (intsts & V3D_INT_CSDDONE(v3d->ver)) {
-		struct v3d_fence *fence =
-			to_v3d_fence(v3d->csd_job->base.irq_fence);
-
-		v3d_job_update_stats(&v3d->csd_job->base, V3D_CSD);
-		trace_v3d_csd_irq(&v3d->drm, fence->seqno);
-
-		v3d->csd_job = NULL;
-		dma_fence_signal(&fence->base);
-
+		v3d_irq_signal_fence(v3d, V3D_CSD, trace_v3d_csd_irq);
 		status = IRQ_HANDLED;
 	}
 
@@ -168,15 +162,7 @@ v3d_hub_irq(int irq, void *arg)
 	V3D_WRITE(V3D_HUB_INT_CLR, intsts);
 
 	if (intsts & V3D_HUB_INT_TFUC) {
-		struct v3d_fence *fence =
-			to_v3d_fence(v3d->tfu_job->base.irq_fence);
-
-		v3d_job_update_stats(&v3d->tfu_job->base, V3D_TFU);
-		trace_v3d_tfu_irq(&v3d->drm, fence->seqno);
-
-		v3d->tfu_job = NULL;
-		dma_fence_signal(&fence->base);
-
+		v3d_irq_signal_fence(v3d, V3D_TFU, trace_v3d_tfu_irq);
 		status = IRQ_HANDLED;
 	}
 

drivers/gpu/drm/v3d/v3d_sched.c

@@ -139,7 +139,7 @@ static void
 v3d_job_start_stats(struct v3d_job *job, enum v3d_queue queue)
 {
 	struct v3d_dev *v3d = job->v3d;
-	struct v3d_file_priv *file = job->file->driver_priv;
+	struct v3d_file_priv *file = job->file_priv;
 	struct v3d_stats *global_stats = &v3d->queue[queue].stats;
 	struct v3d_stats *local_stats = &file->stats[queue];
 	u64 now = local_clock();
@@ -194,11 +194,11 @@ v3d_stats_update(struct v3d_stats *stats, u64 now)
 }
 
 void
-v3d_job_update_stats(struct v3d_job *job, enum v3d_queue queue)
+v3d_job_update_stats(struct v3d_job *job, enum v3d_queue q)
 {
 	struct v3d_dev *v3d = job->v3d;
-	struct v3d_file_priv *file = job->file->driver_priv;
-	struct v3d_stats *global_stats = &v3d->queue[queue].stats;
+	struct v3d_queue_state *queue = &v3d->queue[q];
+	struct v3d_stats *global_stats = &queue->stats;
 	u64 now = local_clock();
 	unsigned long flags;
 
@@ -209,10 +209,10 @@ v3d_job_update_stats(struct v3d_job *job, enum v3d_queue queue)
 		preempt_disable();
 
 	/* Don't update the local stats if the file context has already closed */
-	if (file)
-		v3d_stats_update(&file->stats[queue], now);
-	else
-		drm_dbg(&v3d->drm, "The file descriptor was closed before job completion\n");
+	spin_lock(&queue->queue_lock);
+	if (job->file_priv)
+		v3d_stats_update(&job->file_priv->stats[q], now);
+	spin_unlock(&queue->queue_lock);
 
 	v3d_stats_update(global_stats, now);
 
@@ -226,27 +226,28 @@ static struct dma_fence *v3d_bin_job_run(struct drm_sched_job *sched_job)
 {
 	struct v3d_bin_job *job = to_bin_job(sched_job);
 	struct v3d_dev *v3d = job->base.v3d;
+	struct v3d_queue_state *queue = &v3d->queue[V3D_BIN];
 	struct drm_device *dev = &v3d->drm;
 	struct dma_fence *fence;
 	unsigned long irqflags;
 
 	if (unlikely(job->base.base.s_fence->finished.error)) {
-		spin_lock_irqsave(&v3d->job_lock, irqflags);
-		v3d->bin_job = NULL;
-		spin_unlock_irqrestore(&v3d->job_lock, irqflags);
+		spin_lock_irqsave(&queue->queue_lock, irqflags);
+		queue->active_job = NULL;
+		spin_unlock_irqrestore(&queue->queue_lock, irqflags);
 		return NULL;
 	}
 
 	/* Lock required around bin_job update vs
 	 * v3d_overflow_mem_work().
 	 */
-	spin_lock_irqsave(&v3d->job_lock, irqflags);
-	v3d->bin_job = job;
+	spin_lock_irqsave(&queue->queue_lock, irqflags);
+	queue->active_job = &job->base;
 	/* Clear out the overflow allocation, so we don't
 	 * reuse the overflow attached to a previous job.
 	 */
 	V3D_CORE_WRITE(0, V3D_PTB_BPOS, 0);
-	spin_unlock_irqrestore(&v3d->job_lock, irqflags);
+	spin_unlock_irqrestore(&queue->queue_lock, irqflags);
 
 	v3d_invalidate_caches(v3d);
 
@@ -290,11 +291,11 @@ static struct dma_fence *v3d_render_job_run(struct drm_sched_job *sched_job)
 	struct dma_fence *fence;
 
 	if (unlikely(job->base.base.s_fence->finished.error)) {
-		v3d->render_job = NULL;
+		v3d->queue[V3D_RENDER].active_job = NULL;
 		return NULL;
 	}
 
-	v3d->render_job = job;
+	v3d->queue[V3D_RENDER].active_job = &job->base;
 
 	/* Can we avoid this flush?  We need to be careful of
 	 * scheduling, though -- imagine job0 rendering to texture and
@@ -338,11 +339,11 @@ v3d_tfu_job_run(struct drm_sched_job *sched_job)
 	struct dma_fence *fence;
 
 	if (unlikely(job->base.base.s_fence->finished.error)) {
-		v3d->tfu_job = NULL;
+		v3d->queue[V3D_TFU].active_job = NULL;
 		return NULL;
 	}
 
-	v3d->tfu_job = job;
+	v3d->queue[V3D_TFU].active_job = &job->base;
 
 	fence = v3d_fence_create(v3d, V3D_TFU);
 	if (IS_ERR(fence))
@@ -386,11 +387,11 @@ v3d_csd_job_run(struct drm_sched_job *sched_job)
 	int i, csd_cfg0_reg;
 
 	if (unlikely(job->base.base.s_fence->finished.error)) {
-		v3d->csd_job = NULL;
+		v3d->queue[V3D_CSD].active_job = NULL;
 		return NULL;
 	}
 
-	v3d->csd_job = job;
+	v3d->queue[V3D_CSD].active_job = &job->base;
 
 	v3d_invalidate_caches(v3d);
 
@@ -574,7 +575,7 @@ static void
 v3d_reset_performance_queries(struct v3d_cpu_job *job)
 {
 	struct v3d_performance_query_info *performance_query = &job->performance_query;
-	struct v3d_file_priv *v3d_priv = job->base.file->driver_priv;
+	struct v3d_file_priv *v3d_priv = job->base.file_priv;
 	struct v3d_dev *v3d = job->base.v3d;
 	struct v3d_perfmon *perfmon;
 
@@ -604,7 +605,7 @@ v3d_write_performance_query_result(struct v3d_cpu_job *job, void *data,
 {
 	struct v3d_performance_query_info *performance_query =
 						&job->performance_query;
-	struct v3d_file_priv *v3d_priv = job->base.file->driver_priv;
+	struct v3d_file_priv *v3d_priv = job->base.file_priv;
 	struct v3d_performance_query *perf_query =
 			&performance_query->queries[query];
 	struct v3d_dev *v3d = job->base.v3d;