target/linux/generic/patches-3.3/120-ppp_txqueue_restart.patch

   1 commit 9a5d2bd99e0dfe9a31b3c160073ac445ba3d773f
   2 Author: David Woodhouse <dwmw2@infradead.org>
   3 Date:   Sun Apr 8 10:01:44 2012 +0000
   4
   5     ppp: Fix race condition with queue start/stop
   6
   7     Commit e675f0cc9a872fd152edc0c77acfed19bf28b81e ("ppp: Don't stop and
   8     restart queue on every TX packet") introduced a race condition which
   9     could leave the net queue stopped even when the channel is no longer
  10     busy. By calling netif_stop_queue() from ppp_start_xmit(), based on the
  11     return value from ppp_xmit_process() but *after* all the locks have been
  12     dropped, we could potentially do so *after* the channel has actually
  13     finished transmitting and attempted to re-wake the queue.
  14
  15     Fix this by moving the netif_stop_queue() into ppp_xmit_process() under
  16     the xmit lock. I hadn't done this previously, because it gets called
  17     from other places than ppp_start_xmit(). But I now think it's the better
  18     option. The net queue *should* be stopped if the channel becomes
  19     congested due to writes from pppd, anyway.
  20
  21     Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
  22     Signed-off-by: David S. Miller <davem@davemloft.net>
  23
  24 commit e675f0cc9a872fd152edc0c77acfed19bf28b81e
  25 Author: David Woodhouse <dwmw2@infradead.org>
  26 Date:   Mon Mar 26 00:03:42 2012 +0000
  27
  28     ppp: Don't stop and restart queue on every TX packet
  29
  30     For every transmitted packet, ppp_start_xmit() will stop the netdev
  31     queue and then, if appropriate, restart it. This causes the TX softirq
  32     to run, entirely gratuitously.
  33
  34     This is "only" a waste of CPU time in the normal case, but it's actively
  35     harmful when the PPP device is a TEQL slave — the wakeup will cause the
  36     offending device to receive the next TX packet from the TEQL queue, when
  37     it *should* have gone to the next slave in the list. We end up seeing
  38     large bursts of packets on just *one* slave device, rather than using
  39     the full available bandwidth over all slaves.
  40
  41     This patch fixes the problem by *not* unconditionally stopping the queue
  42     in ppp_start_xmit(). It adds a return value from ppp_xmit_process()
  43     which indicates whether the queue should be stopped or not.
  44
  45     It *doesn't* remove the call to netif_wake_queue() from
  46     ppp_xmit_process(), because other code paths (especially from
  47     ppp_output_wakeup()) need it there and it's messy to push it out to the
  48     other callers to do it based on the return value. So we leave it in
  49     place — it's a no-op in the case where the queue wasn't stopped, so it's
  50     harmless in the TX path.
  51
  52     Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
  53     Signed-off-by: David S. Miller <davem@davemloft.net>
  54
  55
  56
  57 diff --git a/drivers/net/ppp/ppp_generic.c b/drivers/net/ppp/ppp_generic.c
  58 index 159da29..21d7151 100644
  59 --- a/drivers/net/ppp/ppp_generic.c
  60 +++ b/drivers/net/ppp/ppp_generic.c
  61 @@ -968,7 +968,6 @@ ppp_start_xmit(struct sk_buff *skb, struct net_device *dev)
  62         proto = npindex_to_proto[npi];
  63         put_unaligned_be16(proto, pp);
  64
  65 -       netif_stop_queue(dev);
  66         skb_queue_tail(&ppp->file.xq, skb);
  67         ppp_xmit_process(ppp);
  68         return NETDEV_TX_OK;
  69 @@ -1063,6 +1062,8 @@ ppp_xmit_process(struct ppp *ppp)
  70                    code that we can accept some more. */
  71                 if (!ppp->xmit_pending && !skb_peek(&ppp->file.xq))
  72                         netif_wake_queue(ppp->dev);
  73 +               else
  74 +                       netif_stop_queue(ppp->dev);
  75         }
  76         ppp_xmit_unlock(ppp);
  77  }