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, 128 insertions, 19 deletions

diff --git a/drivers/gpu/drm/v3d/v3d_sched.c b/drivers/gpu/drm/v3d/v3d_sched.c
index c66d0ce21435..f7508e907536 100644
--- a/drivers/gpu/drm/v3d/v3d_sched.c
+++ b/drivers/gpu/drm/v3d/v3d_sched.c
@@ -30,6 +30,12 @@ to_v3d_job(struct drm_sched_job *sched_job)
 	return container_of(sched_job, struct v3d_job, base);
 }
 
+static struct v3d_tfu_job *
+to_tfu_job(struct drm_sched_job *sched_job)
+{
+	return container_of(sched_job, struct v3d_tfu_job, base);
+}
+
 static void
 v3d_job_free(struct drm_sched_job *sched_job)
 {
@@ -40,6 +46,16 @@ v3d_job_free(struct drm_sched_job *sched_job)
 	v3d_exec_put(job->exec);
 }
 
+static void
+v3d_tfu_job_free(struct drm_sched_job *sched_job)
+{
+	struct v3d_tfu_job *job = to_tfu_job(sched_job);
+
+	drm_sched_job_cleanup(sched_job);
+
+	v3d_tfu_job_put(job);
+}
+
 /**
  * Returns the fences that the bin or render job depends on, one by one.
  * v3d_job_run() won't be called until all of them have been signaled.
@@ -78,6 +94,27 @@ v3d_job_dependency(struct drm_sched_job *sched_job,
 	return fence;
 }
 
+/**
+ * Returns the fences that the TFU job depends on, one by one.
+ * v3d_tfu_job_run() won't be called until all of them have been
+ * signaled.
+ */
+static struct dma_fence *
+v3d_tfu_job_dependency(struct drm_sched_job *sched_job,
+		       struct drm_sched_entity *s_entity)
+{
+	struct v3d_tfu_job *job = to_tfu_job(sched_job);
+	struct dma_fence *fence;
+
+	fence = job->in_fence;
+	if (fence) {
+		job->in_fence = NULL;
+		return fence;
+	}
+
+	return NULL;
+}
+
 static struct dma_fence *v3d_job_run(struct drm_sched_job *sched_job)
 {
 	struct v3d_job *job = to_v3d_job(sched_job);
@@ -149,28 +186,47 @@ static struct dma_fence *v3d_job_run(struct drm_sched_job *sched_job)
 	return fence;
 }
 
-static void
-v3d_job_timedout(struct drm_sched_job *sched_job)
+static struct dma_fence *
+v3d_tfu_job_run(struct drm_sched_job *sched_job)
 {
-	struct v3d_job *job = to_v3d_job(sched_job);
-	struct v3d_exec_info *exec = job->exec;
-	struct v3d_dev *v3d = exec->v3d;
-	enum v3d_queue job_q = job == &exec->bin ? V3D_BIN : V3D_RENDER;
-	enum v3d_queue q;
-	u32 ctca = V3D_CORE_READ(0, V3D_CLE_CTNCA(job_q));
-	u32 ctra = V3D_CORE_READ(0, V3D_CLE_CTNRA(job_q));
+	struct v3d_tfu_job *job = to_tfu_job(sched_job);
+	struct v3d_dev *v3d = job->v3d;
+	struct drm_device *dev = &v3d->drm;
+	struct dma_fence *fence;
 
-	/* If the current address or return address have changed, then
-	 * the GPU has probably made progress and we should delay the
-	 * reset.  This could fail if the GPU got in an infinite loop
-	 * in the CL, but that is pretty unlikely outside of an i-g-t
-	 * testcase.
-	 */
-	if (job->timedout_ctca != ctca || job->timedout_ctra != ctra) {
-		job->timedout_ctca = ctca;
-		job->timedout_ctra = ctra;
-		return;
+	fence = v3d_fence_create(v3d, V3D_TFU);
+	if (IS_ERR(fence))
+		return NULL;
+
+	v3d->tfu_job = job;
+	if (job->done_fence)
+		dma_fence_put(job->done_fence);
+	job->done_fence = dma_fence_get(fence);
+
+	trace_v3d_submit_tfu(dev, to_v3d_fence(fence)->seqno);
+
+	V3D_WRITE(V3D_TFU_IIA, job->args.iia);
+	V3D_WRITE(V3D_TFU_IIS, job->args.iis);
+	V3D_WRITE(V3D_TFU_ICA, job->args.ica);
+	V3D_WRITE(V3D_TFU_IUA, job->args.iua);
+	V3D_WRITE(V3D_TFU_IOA, job->args.ioa);
+	V3D_WRITE(V3D_TFU_IOS, job->args.ios);
+	V3D_WRITE(V3D_TFU_COEF0, job->args.coef[0]);
+	if (job->args.coef[0] & V3D_TFU_COEF0_USECOEF) {
+		V3D_WRITE(V3D_TFU_COEF1, job->args.coef[1]);
+		V3D_WRITE(V3D_TFU_COEF2, job->args.coef[2]);
+		V3D_WRITE(V3D_TFU_COEF3, job->args.coef[3]);
 	}
+	/* ICFG kicks off the job. */
+	V3D_WRITE(V3D_TFU_ICFG, job->args.icfg | V3D_TFU_ICFG_IOC);
+
+	return fence;
+}
+
+static void
+v3d_gpu_reset_for_timeout(struct v3d_dev *v3d, struct drm_sched_job *sched_job)
+{
+	enum v3d_queue q;
 
 	mutex_lock(&v3d->reset_lock);
 
@@ -195,6 +251,39 @@ v3d_job_timedout(struct drm_sched_job *sched_job)
 	mutex_unlock(&v3d->reset_lock);
 }
 
+static void
+v3d_job_timedout(struct drm_sched_job *sched_job)
+{
+	struct v3d_job *job = to_v3d_job(sched_job);
+	struct v3d_exec_info *exec = job->exec;
+	struct v3d_dev *v3d = exec->v3d;
+	enum v3d_queue job_q = job == &exec->bin ? V3D_BIN : V3D_RENDER;
+	u32 ctca = V3D_CORE_READ(0, V3D_CLE_CTNCA(job_q));
+	u32 ctra = V3D_CORE_READ(0, V3D_CLE_CTNRA(job_q));
+
+	/* If the current address or return address have changed, then
+	 * the GPU has probably made progress and we should delay the
+	 * reset.  This could fail if the GPU got in an infinite loop
+	 * in the CL, but that is pretty unlikely outside of an i-g-t
+	 * testcase.
+	 */
+	if (job->timedout_ctca != ctca || job->timedout_ctra != ctra) {
+		job->timedout_ctca = ctca;
+		job->timedout_ctra = ctra;
+		return;
+	}
+
+	v3d_gpu_reset_for_timeout(v3d, sched_job);
+}
+
+static void
+v3d_tfu_job_timedout(struct drm_sched_job *sched_job)
+{
+	struct v3d_tfu_job *job = to_tfu_job(sched_job);
+
+	v3d_gpu_reset_for_timeout(job->v3d, sched_job);
+}
+
 static const struct drm_sched_backend_ops v3d_sched_ops = {
 	.dependency = v3d_job_dependency,
 	.run_job = v3d_job_run,
@@ -202,6 +291,13 @@ static const struct drm_sched_backend_ops v3d_sched_ops = {
 	.free_job = v3d_job_free
 };
 
+static const struct drm_sched_backend_ops v3d_tfu_sched_ops = {
+	.dependency = v3d_tfu_job_dependency,
+	.run_job = v3d_tfu_job_run,
+	.timedout_job = v3d_tfu_job_timedout,
+	.free_job = v3d_tfu_job_free
+};
+
 int
 v3d_sched_init(struct v3d_dev *v3d)
 {
@@ -232,6 +328,19 @@ v3d_sched_init(struct v3d_dev *v3d)
 		return ret;
 	}
 
+	ret = drm_sched_init(&v3d->queue[V3D_TFU].sched,
+			     &v3d_tfu_sched_ops,
+			     hw_jobs_limit, job_hang_limit,
+			     msecs_to_jiffies(hang_limit_ms),
+			     "v3d_tfu");
+	if (ret) {
+		dev_err(v3d->dev, "Failed to create TFU scheduler: %d.",
+			ret);
+		drm_sched_fini(&v3d->queue[V3D_RENDER].sched);
+		drm_sched_fini(&v3d->queue[V3D_BIN].sched);
+		return ret;
+	}
+
 	return 0;
 }