Hello,
On Sat, 23 May 2015, Eric Dumazet wrote:
> > Not sure if that can help but both RCU_INIT_POINTER
> > calls in veth_dellink can be before the unregister_netdevice_queue
> > ones. May be it will help for the case when veth_dellink is
> > called with head=NULL. As result, veth_xmit will free all
> > flying skbs to peer that is unregistered first.
>
> Good point, but note that we never call dellink() method with head==NULL
> in current kernels.
Can it be a general problem with process_backlog
delivering packets even after the ndo_stop call?
It can happen even for devices that use napi_disable.
synchronize_net simply misses packets processed on
another CPU because process_backlog is not always in
RCU read section.
What about moving flush_backlog before
NETDEV_UNREGISTER, it was after NETDEV_UNREGISTER_FINAL.
It will work like a synchronize_net call.
rollback_registered_many:
1. __dev_close_many: Commit netif_running, ndo_stop
2. on_each_cpu(flush_backlog, dev, 1): may be its place
can be additionally tuned...
3. synchronize_net: not needed anymore?
4. call_netdevice_notifiers(NETDEV_UNREGISTER, dev)
Compile-tested only:
diff --git a/net/core/dev.c b/net/core/dev.c
index c7ba038..e402661 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -6034,6 +6036,7 @@ static void rollback_registered_many(struct list_head
*head)
unlist_netdevice(dev);
dev->reg_state = NETREG_UNREGISTERING;
+ on_each_cpu(flush_backlog, dev, 1);
}
synchronize_net();
@@ -6657,8 +6660,6 @@ void netdev_run_todo(void)
dev->reg_state = NETREG_UNREGISTERED;
- on_each_cpu(flush_backlog, dev, 1);
-
netdev_wait_allrefs(dev);
/* paranoia */
Regards
--
Julian Anastasov <j...@ssi.bg>
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
