drm/v3d: Add support for submitting jobs to the TFU.

The TFU can copy from raster, UIF, and SAND input images to UIF output
images, with optional mipmap generation.  This will certainly be
useful for media EGL image input, but is also useful immediately for
mipmap generation without bogging the V3D core down.

For now we only run the queue 1 job deep, and don't have any hang
recovery (though I don't think we should need it, with TFU).  Queuing
multiple jobs in the HW will require synchronizing the YUV coefficient
regs updates since they don't get FIFOed with the job.

v2: Change the ioctl to IOW instead of IOWR, always set COEF0, explain
    why TFU is AUTH, clarify the syncing docs, drop the unused TFU
    interrupt regs (you're expected to use the hub's), don't take
    &bo->base for NULL bos.
v3: Fix a little whitespace alignment (noticed by checkpatch), rebase
    on drm_sched_job_cleanup() changes.

Signed-off-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Dave Emett <david.emett@broadcom.com> (v2)
Link: https://patchwork.freedesktop.org/patch/264607/

1 files changed, 158 insertions, 20 deletions

diff --git a/drivers/gpu/drm/v3d/v3d_gem.c b/drivers/gpu/drm/v3d/v3d_gem.c
index 1e8947c7d954..6abe2fa43306 100644
--- a/drivers/gpu/drm/v3d/v3d_gem.c
+++ b/drivers/gpu/drm/v3d/v3d_gem.c
@@ -207,26 +207,27 @@ v3d_flush_caches(struct v3d_dev *v3d)
 }
 
 static void
-v3d_attach_object_fences(struct v3d_exec_info *exec)
+v3d_attach_object_fences(struct v3d_bo **bos, int bo_count,
+			 struct dma_fence *fence)
 {
-	struct dma_fence *out_fence = exec->render_done_fence;
 	int i;
 
-	for (i = 0; i < exec->bo_count; i++) {
+	for (i = 0; i < bo_count; i++) {
 		/* XXX: Use shared fences for read-only objects. */
-		reservation_object_add_excl_fence(exec->bo[i]->resv, out_fence);
+		reservation_object_add_excl_fence(bos[i]->resv, fence);
 	}
 }
 
 static void
 v3d_unlock_bo_reservations(struct drm_device *dev,
-			   struct v3d_exec_info *exec,
+			   struct v3d_bo **bos,
+			   int bo_count,
 			   struct ww_acquire_ctx *acquire_ctx)
 {
 	int i;
 
-	for (i = 0; i < exec->bo_count; i++)
-		ww_mutex_unlock(&exec->bo[i]->resv->lock);
+	for (i = 0; i < bo_count; i++)
+		ww_mutex_unlock(&bos[i]->resv->lock);
 
 	ww_acquire_fini(acquire_ctx);
 }
@@ -240,7 +241,8 @@ v3d_unlock_bo_reservations(struct drm_device *dev,
  */
 static int
 v3d_lock_bo_reservations(struct drm_device *dev,
-			 struct v3d_exec_info *exec,
+			 struct v3d_bo **bos,
+			 int bo_count,
 			 struct ww_acquire_ctx *acquire_ctx)
 {
 	int contended_lock = -1;
@@ -250,7 +252,7 @@ v3d_lock_bo_reservations(struct drm_device *dev,
 
 retry:
 	if (contended_lock != -1) {
-		struct v3d_bo *bo = exec->bo[contended_lock];
+		struct v3d_bo *bo = bos[contended_lock];
 
 		ret = ww_mutex_lock_slow_interruptible(&bo->resv->lock,
 						       acquire_ctx);
@@ -260,20 +262,20 @@ retry:
 		}
 	}
 
-	for (i = 0; i < exec->bo_count; i++) {
+	for (i = 0; i < bo_count; i++) {
 		if (i == contended_lock)
 			continue;
 
-		ret = ww_mutex_lock_interruptible(&exec->bo[i]->resv->lock,
+		ret = ww_mutex_lock_interruptible(&bos[i]->resv->lock,
 						  acquire_ctx);
 		if (ret) {
 			int j;
 
 			for (j = 0; j < i; j++)
-				ww_mutex_unlock(&exec->bo[j]->resv->lock);
+				ww_mutex_unlock(&bos[j]->resv->lock);
 
 			if (contended_lock != -1 && contended_lock >= i) {
-				struct v3d_bo *bo = exec->bo[contended_lock];
+				struct v3d_bo *bo = bos[contended_lock];
 
 				ww_mutex_unlock(&bo->resv->lock);
 			}
@@ -293,10 +295,11 @@ retry:
 	/* Reserve space for our shared (read-only) fence references,
 	 * before we commit the CL to the hardware.
 	 */
-	for (i = 0; i < exec->bo_count; i++) {
-		ret = reservation_object_reserve_shared(exec->bo[i]->resv, 1);
+	for (i = 0; i < bo_count; i++) {
+		ret = reservation_object_reserve_shared(bos[i]->resv, 1);
 		if (ret) {
-			v3d_unlock_bo_reservations(dev, exec, acquire_ctx);
+			v3d_unlock_bo_reservations(dev, bos, bo_count,
+						   acquire_ctx);
 			return ret;
 		}
 	}
@@ -419,6 +422,33 @@ void v3d_exec_put(struct v3d_exec_info *exec)
 	kref_put(&exec->refcount, v3d_exec_cleanup);
 }
 
+static void
+v3d_tfu_job_cleanup(struct kref *ref)
+{
+	struct v3d_tfu_job *job = container_of(ref, struct v3d_tfu_job,
+					       refcount);
+	struct v3d_dev *v3d = job->v3d;
+	unsigned int i;
+
+	dma_fence_put(job->in_fence);
+	dma_fence_put(job->done_fence);
+
+	for (i = 0; i < ARRAY_SIZE(job->bo); i++) {
+		if (job->bo[i])
+			drm_gem_object_put_unlocked(&job->bo[i]->base);
+	}
+
+	pm_runtime_mark_last_busy(v3d->dev);
+	pm_runtime_put_autosuspend(v3d->dev);
+
+	kfree(job);
+}
+
+void v3d_tfu_job_put(struct v3d_tfu_job *job)
+{
+	kref_put(&job->refcount, v3d_tfu_job_cleanup);
+}
+
 int
 v3d_wait_bo_ioctl(struct drm_device *dev, void *data,
 		  struct drm_file *file_priv)
@@ -536,7 +566,8 @@ v3d_submit_cl_ioctl(struct drm_device *dev, void *data,
 	if (ret)
 		goto fail;
 
-	ret = v3d_lock_bo_reservations(dev, exec, &acquire_ctx);
+	ret = v3d_lock_bo_reservations(dev, exec->bo, exec->bo_count,
+				       &acquire_ctx);
 	if (ret)
 		goto fail;
 
@@ -570,9 +601,10 @@ v3d_submit_cl_ioctl(struct drm_device *dev, void *data,
 				  &v3d_priv->sched_entity[V3D_RENDER]);
 	mutex_unlock(&v3d->sched_lock);
 
-	v3d_attach_object_fences(exec);
+	v3d_attach_object_fences(exec->bo, exec->bo_count,
+				 exec->render_done_fence);
 
-	v3d_unlock_bo_reservations(dev, exec, &acquire_ctx);
+	v3d_unlock_bo_reservations(dev, exec->bo, exec->bo_count, &acquire_ctx);
 
 	/* Update the return sync object for the */
 	sync_out = drm_syncobj_find(file_priv, args->out_sync);
@@ -588,13 +620,119 @@ v3d_submit_cl_ioctl(struct drm_device *dev, void *data,
 
 fail_unreserve:
 	mutex_unlock(&v3d->sched_lock);
-	v3d_unlock_bo_reservations(dev, exec, &acquire_ctx);
+	v3d_unlock_bo_reservations(dev, exec->bo, exec->bo_count, &acquire_ctx);
 fail:
 	v3d_exec_put(exec);
 
 	return ret;
 }
 
+/**
+ * v3d_submit_tfu_ioctl() - Submits a TFU (texture formatting) job to the V3D.
+ * @dev: DRM device
+ * @data: ioctl argument
+ * @file_priv: DRM file for this fd
+ *
+ * Userspace provides the register setup for the TFU, which we don't
+ * need to validate since the TFU is behind the MMU.
+ */
+int
+v3d_submit_tfu_ioctl(struct drm_device *dev, void *data,
+		     struct drm_file *file_priv)
+{
+	struct v3d_dev *v3d = to_v3d_dev(dev);
+	struct v3d_file_priv *v3d_priv = file_priv->driver_priv;
+	struct drm_v3d_submit_tfu *args = data;
+	struct v3d_tfu_job *job;
+	struct ww_acquire_ctx acquire_ctx;
+	struct drm_syncobj *sync_out;
+	struct dma_fence *sched_done_fence;
+	int ret = 0;
+	int bo_count;
+
+	job = kcalloc(1, sizeof(*job), GFP_KERNEL);
+	if (!job)
+		return -ENOMEM;
+
+	ret = pm_runtime_get_sync(v3d->dev);
+	if (ret < 0) {
+		kfree(job);
+		return ret;
+	}
+
+	kref_init(&job->refcount);
+
+	ret = drm_syncobj_find_fence(file_priv, args->in_sync,
+				     0, 0, &job->in_fence);
+	if (ret == -EINVAL)
+		goto fail;
+
+	job->args = *args;
+	job->v3d = v3d;
+
+	spin_lock(&file_priv->table_lock);
+	for (bo_count = 0; bo_count < ARRAY_SIZE(job->bo); bo_count++) {
+		struct drm_gem_object *bo;
+
+		if (!args->bo_handles[bo_count])
+			break;
+
+		bo = idr_find(&file_priv->object_idr,
+			      args->bo_handles[bo_count]);
+		if (!bo) {
+			DRM_DEBUG("Failed to look up GEM BO %d: %d\n",
+				  bo_count, args->bo_handles[bo_count]);
+			ret = -ENOENT;
+			spin_unlock(&file_priv->table_lock);
+			goto fail;
+		}
+		drm_gem_object_get(bo);
+		job->bo[bo_count] = to_v3d_bo(bo);
+	}
+	spin_unlock(&file_priv->table_lock);
+
+	ret = v3d_lock_bo_reservations(dev, job->bo, bo_count, &acquire_ctx);
+	if (ret)
+		goto fail;
+
+	mutex_lock(&v3d->sched_lock);
+	ret = drm_sched_job_init(&job->base,
+				 &v3d_priv->sched_entity[V3D_TFU],
+				 v3d_priv);
+	if (ret)
+		goto fail_unreserve;
+
+	sched_done_fence = dma_fence_get(&job->base.s_fence->finished);
+
+	kref_get(&job->refcount); /* put by scheduler job completion */
+	drm_sched_entity_push_job(&job->base, &v3d_priv->sched_entity[V3D_TFU]);
+	mutex_unlock(&v3d->sched_lock);
+
+	v3d_attach_object_fences(job->bo, bo_count, sched_done_fence);
+
+	v3d_unlock_bo_reservations(dev, job->bo, bo_count, &acquire_ctx);
+
+	/* Update the return sync object */
+	sync_out = drm_syncobj_find(file_priv, args->out_sync);
+	if (sync_out) {
+		drm_syncobj_replace_fence(sync_out, 0, sched_done_fence);
+		drm_syncobj_put(sync_out);
+	}
+	dma_fence_put(sched_done_fence);
+
+	v3d_tfu_job_put(job);
+
+	return 0;
+
+fail_unreserve:
+	mutex_unlock(&v3d->sched_lock);
+	v3d_unlock_bo_reservations(dev, job->bo, bo_count, &acquire_ctx);
+fail:
+	v3d_tfu_job_put(job);
+
+	return ret;
+}
+
 int
 v3d_gem_init(struct drm_device *dev)
 {