Re: [Xen-devel] [PATCH net v2 2/3] xen-netback: don't stop dealloc kthread too early

On 11/08/14 14:58, David Vrabel wrote:

On 11/08/14 14:48, Wei Liu wrote:

quoted

On Mon, Aug 11, 2014 at 01:10:12PM +0100, David Vrabel wrote:

quoted

On 08/08/14 17:37, Wei Liu wrote:

[...]

quoted

quoted

  	if (tx_evtchn == rx_evtchn) {
  		/* feature-split-event-channels == 0 */

@@ -687,6 +700,9 @@ void xenvif_disconnect(struct xenvif *vif)
  			queue->task = NULL;
  		}
  
+		wait_event(queue->inflight_wq,
+			   atomic_read(&queue->inflight_packets) == 0);

Just make the dealloc task not stop unless it has deallocated all
outstanding requests.  There's no need for another wait queue here.

Are you suggesting something like

   while (atomic_read(&queue->inflight_packets) !=0)
   	schedule_timeout(SOME_TIMEOUT);

No. That would be awful!

Add the test to the kthread itself:

int xenvif_dealloc_kthread(void *data)
{
	struct xenvif_queue *queue = data;

	while (atomic_read(&queue->inflight_packets) > 0
                || !kthread_should_stop()) {
             [...]

etc.

Although, the main loop is a bit confused, so I suggest adding:

static bool xenvif_dealloc_thread_should_stop(struct xenvif_queue *q)
{
     /* Dealloc thread must remain running if there are any inflight
      * packets so it can properly dealloc them when they complete.
      */
     return atomic_read(&queue->inflight_packets) == 0
         && kthread_should_stop();
}

And cleaning it up a bit (the while() could be a for(;;)).

I would recommend this:
---

@@ -2066,7 +2066,7 @@ int xenvif_dealloc_kthread(void *data)
 		wait_event_interruptible(queue->dealloc_wq,
 					 tx_dealloc_work_todo(queue) ||
 					 kthread_should_stop());
-		if (kthread_should_stop())
+		if (kthread_should_stop() && !atomic_read(&queue->inflight_packets))
 			break;
 
 		xenvif_tx_dealloc_action(queue);

---
If kthread_stop called, this will keep the main loop running until all callbacks are called.
Then it proceeds to the exit branch, otherwise doesn't disrupt normal operation.