Re: LOCKDEP complaints in l2tp_xmit_skb()

[Eric Dumazet <eric.dumazet@gmail.com>]

On Thu, 2012-06-28 at 10:57 +0200, Eric Dumazet wrote:
> On Thu, 2012-06-28 at 08:56 +0200, Eric Dumazet wrote:
> 
> > [PATCH] net: Qdisc busylock gets its own lockdep class
> > 
> > Tom Parkin reported following LOCKDEP splat :
> ..
> > 
> > Instruct lockdep that each Qdisc busylock is independant, or else
> > bonding or various tunnels can trigger a splat.
> > 
> > Reported-by: Tom Parkin <tparkin@katalix.com>
> > Signed-off-by: Eric Dumazet <edumazet@google.com>
> > ---
> 
> I reproduced the bug using a bond0 device, adding a qdisc on it,
> (one Qdisc on bond0, and a Qdisc on the slave too)
> 
> Problem with this patch is I have following message :
> 
> BUG: key ffff88..... not in .data!
> 
> No more LOCKDEP splat, but patch not good as is.

I tested the alternative following patch with my bonding setup,
could you test it with l2tp ?

 net/core/dev.c |   12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

diff --git a/net/core/dev.c b/net/core/dev.c
index 6df2140..d265c67 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -2412,13 +2412,13 @@ static inline int __dev_xmit_skb(struct sk_buff *skb, struct Qdisc *q,
 
 		qdisc_bstats_update(q, skb);
 
-		if (sch_direct_xmit(skb, q, dev, txq, root_lock)) {
-			if (unlikely(contended)) {
-				spin_unlock(&q->busylock);
-				contended = false;
-			}
+		if (unlikely(contended)) {
+			spin_unlock(&q->busylock);
+			contended = false;
+		}
+		if (sch_direct_xmit(skb, q, dev, txq, root_lock))
 			__qdisc_run(q);
-		} else
+		else
 			qdisc_run_end(q);
 
 		rc = NET_XMIT_SUCCESS;