From mboxrd@z Thu Jan  1 00:00:00 1970
From: Rusty Russell <rusty@rustcorp.com.au>
Subject: Re: [PATCH 2/4] virtio_net: return NETDEV_TX_BUSY instead of queueing an extra skb.
Date: Tue, 2 Jun 2009 23:34:28 +0930
Message-ID: <200906022334.29428.rusty@rustcorp.com.au>
References: <200905292346.04815.rusty@rustcorp.com.au> <1243930050.9146.53.camel@blaa>
Mime-Version: 1.0
Content-Type: Text/Plain;
  charset="utf-8"
Content-Transfer-Encoding: 7bit
Cc: netdev@vger.kernel.org, virtualization@lists.linux-foundation.org,
	Herbert Xu <herbert@gondor.apana.org.au>
To: Mark McLoughlin <markmc@redhat.com>
Return-path: <netdev-owner@vger.kernel.org>
Received: from ozlabs.org ([203.10.76.45]:38614 "EHLO ozlabs.org"
	rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP
	id S1752795AbZFBOGq (ORCPT <rfc822;netdev@vger.kernel.org>);
	Tue, 2 Jun 2009 10:06:46 -0400
In-Reply-To: <1243930050.9146.53.camel@blaa>
Content-Disposition: inline
Sender: netdev-owner@vger.kernel.org
List-ID: <netdev.vger.kernel.org>

On Tue, 2 Jun 2009 05:37:30 pm Mark McLoughlin wrote:
> On Fri, 2009-05-29 at 23:46 +0930, Rusty Russell wrote:
> > This effectively reverts 99ffc696d10b28580fe93441d627cf290ac4484c
> > "virtio: wean net driver off NETDEV_TX_BUSY".
> >
> > The complexity of queuing an skb (setting a tasklet to re-xmit) is
> > questionable,
>
> It certainly adds some subtle complexities to start_xmit()
>
> >  especially once we get rid of the other reason for the
> > tasklet in the next patch.
> >
> > If the skb won't fit in the tx queue, just return NETDEV_TX_BUSY.  It
> > might be frowned upon, but it's common and not going away any time
> > soon.
> >
> > Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
> > Cc: Herbert Xu <herbert@gondor.apana.org.au>
> > ---
> >  drivers/net/virtio_net.c |   49
> > ++++++++++------------------------------------- 1 file changed, 11
> > insertions(+), 38 deletions(-)
> >
> > diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
> > --- a/drivers/net/virtio_net.c
> > +++ b/drivers/net/virtio_net.c
> >
> > @@ -526,27 +517,14 @@ again:
> >  	/* Free up any pending old buffers before queueing new ones. */
> >  	free_old_xmit_skbs(vi);
> >
> > -	/* If we has a buffer left over from last time, send it now. */
> > -	if (unlikely(vi->last_xmit_skb) &&
> > -	    xmit_skb(vi, vi->last_xmit_skb) != 0)
> > -		goto stop_queue;
> > +	/* Put new one in send queue and do transmit */
> > +	__skb_queue_head(&vi->send, skb);
> > +	if (likely(xmit_skb(vi, skb) == 0)) {
> > +		vi->svq->vq_ops->kick(vi->svq);
> > +		return NETDEV_TX_OK;
> > +	}
>
> Hmm, is it okay to leave the skb on the send queue if we return
> NETDEV_TX_BUSY?

Certainly not.  That's a bug.  Incremental fix is:

diff -u b/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
--- b/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c
@@ -524,8 +524,9 @@
 		return NETDEV_TX_OK;
 	}
 
-	/* Ring too full for this packet. */
+	/* Ring too full for this packet, remove it from queue again. */
 	pr_debug("%s: virtio not prepared to send\n", dev->name);
+	__skb_unlink(skb, &vi->send);
 	netif_stop_queue(dev);
 
 	/* Activate callback for using skbs: if this returns false it

Thanks!
Rusty.