[PATCH v2] xen/netback: Wake dealloc thread after completing zerocopy work

[PATCH v2] xen/netback: Wake dealloc thread after completing zerocopy work

[Ross Lagerwall]

Waking the dealloc thread before decrementing inflight_packets is racy
because it means the thread may go to sleep before inflight_packets is
decremented. If kthread_stop() has already been called, the dealloc
thread may wait forever with nothing to wake it. Instead, wake the
thread only after decrementing inflight_packets.

Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
---
Changed in V2: Move wakeup into zerocopy_complete function.

 drivers/net/xen-netback/interface.c | 6 ++++++
 drivers/net/xen-netback/netback.c   | 1 -
 2 files changed, 6 insertions(+), 1 deletion(-)

diff --git a/drivers/net/xen-netback/interface.c b/drivers/net/xen-netback/interface.c
index 1a83e19..28577a3 100644
--- a/drivers/net/xen-netback/interface.c
+++ b/drivers/net/xen-netback/interface.c
@@ -61,6 +61,12 @@ void xenvif_skb_zerocopy_prepare(struct xenvif_queue *queue,
 void xenvif_skb_zerocopy_complete(struct xenvif_queue *queue)
 {
 	atomic_dec(&queue->inflight_packets);
+
+	/* Wake the dealloc thread _after_ decrementing inflight_packets so
+	 * that if kthread_stop() has already been called, the dealloc thread
+	 * does not wait forever with nothing to wake it.
+	 */
+	wake_up(&queue->dealloc_wq);
 }
 
 int xenvif_schedulable(struct xenvif *vif)
diff --git a/drivers/net/xen-netback/netback.c b/drivers/net/xen-netback/netback.c
index 7d50711..09ffda4 100644
--- a/drivers/net/xen-netback/netback.c
+++ b/drivers/net/xen-netback/netback.c
@@ -1536,7 +1536,6 @@ void xenvif_zerocopy_callback(struct ubuf_info *ubuf, bool zerocopy_success)
 		smp_wmb();
 		queue->dealloc_prod++;
 	} while (ubuf);
-	wake_up(&queue->dealloc_wq);
 	spin_unlock_irqrestore(&queue->callback_lock, flags);
 
 	if (likely(zerocopy_success))
-- 
2.1.0

Re: [PATCH v2] xen/netback: Wake dealloc thread after completing zerocopy work

[Wei Liu]

On Tue, Aug 04, 2015 at 03:40:59PM +0100, Ross Lagerwall wrote:
> Waking the dealloc thread before decrementing inflight_packets is racy
> because it means the thread may go to sleep before inflight_packets is
> decremented. If kthread_stop() has already been called, the dealloc
> thread may wait forever with nothing to wake it. Instead, wake the
> thread only after decrementing inflight_packets.
> 
> Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>

Acked-by: Wei Liu <wei.liu2@citrix.com>

> ---
> Changed in V2: Move wakeup into zerocopy_complete function.
> 
>  drivers/net/xen-netback/interface.c | 6 ++++++
>  drivers/net/xen-netback/netback.c   | 1 -
>  2 files changed, 6 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/net/xen-netback/interface.c b/drivers/net/xen-netback/interface.c
> index 1a83e19..28577a3 100644
> --- a/drivers/net/xen-netback/interface.c
> +++ b/drivers/net/xen-netback/interface.c
> @@ -61,6 +61,12 @@ void xenvif_skb_zerocopy_prepare(struct xenvif_queue *queue,
>  void xenvif_skb_zerocopy_complete(struct xenvif_queue *queue)
>  {
>  	atomic_dec(&queue->inflight_packets);
> +
> +	/* Wake the dealloc thread _after_ decrementing inflight_packets so
> +	 * that if kthread_stop() has already been called, the dealloc thread
> +	 * does not wait forever with nothing to wake it.
> +	 */
> +	wake_up(&queue->dealloc_wq);
>  }
>  
>  int xenvif_schedulable(struct xenvif *vif)
> diff --git a/drivers/net/xen-netback/netback.c b/drivers/net/xen-netback/netback.c
> index 7d50711..09ffda4 100644
> --- a/drivers/net/xen-netback/netback.c
> +++ b/drivers/net/xen-netback/netback.c
> @@ -1536,7 +1536,6 @@ void xenvif_zerocopy_callback(struct ubuf_info *ubuf, bool zerocopy_success)
>  		smp_wmb();
>  		queue->dealloc_prod++;
>  	} while (ubuf);
> -	wake_up(&queue->dealloc_wq);
>  	spin_unlock_irqrestore(&queue->callback_lock, flags);
>  
>  	if (likely(zerocopy_success))
> -- 
> 2.1.0

Re: [PATCH v2] xen/netback: Wake dealloc thread after completing zerocopy work

[David Miller]

From: Ross Lagerwall <ross.lagerwall@citrix.com>
Date: Tue, 4 Aug 2015 15:40:59 +0100

> Waking the dealloc thread before decrementing inflight_packets is racy
> because it means the thread may go to sleep before inflight_packets is
> decremented. If kthread_stop() has already been called, the dealloc
> thread may wait forever with nothing to wake it. Instead, wake the
> thread only after decrementing inflight_packets.
> 
> Signed-off-by: Ross Lagerwall <ross.lagerwall@citrix.com>
> ---
> Changed in V2: Move wakeup into zerocopy_complete function.

Applied, thanks.