[PATCH V2 03/29] staging: bcm2835-camera: Do not bulk receive from service thread

Stefan Wahren wahrenst at gmx.net
Sat Jun 29 12:13:18 UTC 2019


From: Dave Stevenson <dave.stevenson at raspberrypi.org>

vchi_bulk_queue_receive will queue up to a default of 4
bulk receives on a connection before blocking.
If called from the VCHI service_callback thread, then
that thread is unable to service the VCHI_CALLBACK_BULK_RECEIVED
events that would enable the queue call to succeed.

Add a workqueue to schedule the call vchi_bulk_queue_receive
in an alternate context to avoid the lock up.

Signed-off-by: Dave Stevenson <dave.stevenson at raspberrypi.org>
Signed-off-by: Stefan Wahren <wahrenst at gmx.net>
Acked-by: Hans Verkuil <hverkuil-cisco at xs4all.nl>
Acked-by: Mauro Carvalho Chehab <mchehab+samsung at kernel.org>
---
 .../vc04_services/bcm2835-camera/mmal-vchiq.c      | 101 ++++++++++++---------
 1 file changed, 59 insertions(+), 42 deletions(-)

diff --git a/drivers/staging/vc04_services/bcm2835-camera/mmal-vchiq.c b/drivers/staging/vc04_services/bcm2835-camera/mmal-vchiq.c
index f1bb900..1a343d8 100644
--- a/drivers/staging/vc04_services/bcm2835-camera/mmal-vchiq.c
+++ b/drivers/staging/vc04_services/bcm2835-camera/mmal-vchiq.c
@@ -117,8 +117,10 @@ struct mmal_msg_context {

 	union {
 		struct {
-			/* work struct for defered callback - must come first */
+			/* work struct for buffer_cb callback */
 			struct work_struct work;
+			/* work struct for deferred callback */
+			struct work_struct buffer_to_host_work;
 			/* mmal instance */
 			struct vchiq_mmal_instance *instance;
 			/* mmal port */
@@ -167,6 +169,9 @@ struct vchiq_mmal_instance {
 	/* component to use next */
 	int component_idx;
 	struct vchiq_mmal_component component[VCHIQ_MMAL_MAX_COMPONENTS];
+
+	/* ordered workqueue to process all bulk operations */
+	struct workqueue_struct *bulk_wq;
 };

 static struct mmal_msg_context *
@@ -248,7 +253,44 @@ static void buffer_work_cb(struct work_struct *work)
 					    msg_context->u.bulk.mmal_flags,
 					    msg_context->u.bulk.dts,
 					    msg_context->u.bulk.pts);
+}

+/* workqueue scheduled callback to handle receiving buffers
+ *
+ * VCHI will allow up to 4 bulk receives to be scheduled before blocking.
+ * If we block in the service_callback context then we can't process the
+ * VCHI_CALLBACK_BULK_RECEIVED message that would otherwise allow the blocked
+ * vchi_bulk_queue_receive() call to complete.
+ */
+static void buffer_to_host_work_cb(struct work_struct *work)
+{
+	struct mmal_msg_context *msg_context =
+		container_of(work, struct mmal_msg_context,
+			     u.bulk.buffer_to_host_work);
+	struct vchiq_mmal_instance *instance = msg_context->instance;
+	unsigned long len = msg_context->u.bulk.buffer_used;
+	int ret;
+
+	if (!len)
+		/* Dummy receive to ensure the buffers remain in order */
+		len = 8;
+	/* queue the bulk submission */
+	vchi_service_use(instance->handle);
+	ret = vchi_bulk_queue_receive(instance->handle,
+				      msg_context->u.bulk.buffer->buffer,
+				      /* Actual receive needs to be a multiple
+				       * of 4 bytes
+				       */
+				      (len + 3) & ~3,
+				      VCHI_FLAGS_CALLBACK_WHEN_OP_COMPLETE |
+				      VCHI_FLAGS_BLOCK_UNTIL_QUEUED,
+				      msg_context);
+
+	vchi_service_release(instance->handle);
+
+	if (ret != 0)
+		pr_err("%s: ctx: %p, vchi_bulk_queue_receive failed %d\n",
+		       __func__, msg_context, ret);
 }

 /* enqueue a bulk receive for a given message context */
@@ -257,7 +299,6 @@ static int bulk_receive(struct vchiq_mmal_instance *instance,
 			struct mmal_msg_context *msg_context)
 {
 	unsigned long rd_len;
-	int ret;

 	rd_len = msg->u.buffer_from_host.buffer_header.length;

@@ -293,45 +334,10 @@ static int bulk_receive(struct vchiq_mmal_instance *instance,
 	msg_context->u.bulk.dts = msg->u.buffer_from_host.buffer_header.dts;
 	msg_context->u.bulk.pts = msg->u.buffer_from_host.buffer_header.pts;

-	/* queue the bulk submission */
-	vchi_service_use(instance->handle);
-	ret = vchi_bulk_queue_receive(instance->handle,
-				      msg_context->u.bulk.buffer->buffer,
-				      /* Actual receive needs to be a multiple
-				       * of 4 bytes
-				       */
-				      (rd_len + 3) & ~3,
-				      VCHI_FLAGS_CALLBACK_WHEN_OP_COMPLETE |
-				      VCHI_FLAGS_BLOCK_UNTIL_QUEUED,
-				      msg_context);
-
-	vchi_service_release(instance->handle);
+	queue_work(msg_context->instance->bulk_wq,
+		   &msg_context->u.bulk.buffer_to_host_work);

-	return ret;
-}
-
-/* enque a dummy bulk receive for a given message context */
-static int dummy_bulk_receive(struct vchiq_mmal_instance *instance,
-			      struct mmal_msg_context *msg_context)
-{
-	int ret;
-
-	/* zero length indicates this was a dummy transfer */
-	msg_context->u.bulk.buffer_used = 0;
-
-	/* queue the bulk submission */
-	vchi_service_use(instance->handle);
-
-	ret = vchi_bulk_queue_receive(instance->handle,
-				      instance->bulk_scratch,
-				      8,
-				      VCHI_FLAGS_CALLBACK_WHEN_OP_COMPLETE |
-				      VCHI_FLAGS_BLOCK_UNTIL_QUEUED,
-				      msg_context);
-
-	vchi_service_release(instance->handle);
-
-	return ret;
+	return 0;
 }

 /* data in message, memcpy from packet into output buffer */
@@ -379,6 +385,8 @@ buffer_from_host(struct vchiq_mmal_instance *instance,

 	/* initialise work structure ready to schedule callback */
 	INIT_WORK(&msg_context->u.bulk.work, buffer_work_cb);
+	INIT_WORK(&msg_context->u.bulk.buffer_to_host_work,
+		  buffer_to_host_work_cb);

 	/* prep the buffer from host message */
 	memset(&m, 0xbc, sizeof(m));	/* just to make debug clearer */
@@ -459,7 +467,7 @@ static void buffer_to_host_cb(struct vchiq_mmal_instance *instance,
 		if (msg->u.buffer_from_host.buffer_header.flags &
 		    MMAL_BUFFER_HEADER_FLAG_EOS) {
 			msg_context->u.bulk.status =
-			    dummy_bulk_receive(instance, msg_context);
+			    bulk_receive(instance, msg, msg_context);
 			if (msg_context->u.bulk.status == 0)
 				return;	/* successful bulk submission, bulk
 					 * completion will trigger callback
@@ -1793,6 +1801,9 @@ int vchiq_mmal_finalise(struct vchiq_mmal_instance *instance)

 	mutex_unlock(&instance->vchiq_mutex);

+	flush_workqueue(instance->bulk_wq);
+	destroy_workqueue(instance->bulk_wq);
+
 	vfree(instance->bulk_scratch);

 	idr_destroy(&instance->context_map);
@@ -1855,6 +1866,11 @@ int vchiq_mmal_init(struct vchiq_mmal_instance **out_instance)

 	params.callback_param = instance;

+	instance->bulk_wq = alloc_ordered_workqueue("mmal-vchiq",
+						    WQ_MEM_RECLAIM);
+	if (!instance->bulk_wq)
+		goto err_free;
+
 	status = vchi_service_open(vchi_instance, &params, &instance->handle);
 	if (status) {
 		pr_err("Failed to open VCHI service connection (status=%d)\n",
@@ -1869,8 +1885,9 @@ int vchiq_mmal_init(struct vchiq_mmal_instance **out_instance)
 	return 0;

 err_close_services:
-
 	vchi_service_close(instance->handle);
+	destroy_workqueue(instance->bulk_wq);
+err_free:
 	vfree(instance->bulk_scratch);
 	kfree(instance);
 	return -ENODEV;
--
2.7.4



More information about the devel mailing list