Re: [PATCH net-next-2.6] bonding: fix user-controlled queuing issues

[Phil Oester <kernel@linuxace.com>]

On Mon, Feb 21, 2011 at 05:53:03PM -0500, Andy Gospodarek wrote:
> Users noticed the following messages were filling their logs when using
> queue 16 for user-mode bonding:
> 
> kernel: bond0 selects TX queue 16, but real number of TX queues is 16
> 
> ---
> 
> My tests all seem to work well, but more testing/feedback is obviously
> appreciated.

Sorry, this patch does not fix the issue:

Feb 22 10:46:50 foo kernel: bond0 selects TX queue 16, but real number of TX queues is 16

Adding the below debugging patch confirms:

--- a/net/core/dev.c.orig       2011-02-17 11:42:09.110280300 -0500
+++ b/net/core/dev.c    2011-02-17 11:40:42.110280300 -0500
@@ -2060,6 +2060,11 @@
        int queue_index;
        const struct net_device_ops *ops = dev->netdev_ops;
 
+                if (net_ratelimit()) {
+                        pr_warning("dev_pick_tx: %s queue_mapping=%d, real_num=%d\n",
+                                dev->name, skb->queue_mapping, dev->real_num_tx_queues);
+                }
+
        if (ops->ndo_select_queue) {
                queue_index = ops->ndo_select_queue(dev, skb);
                queue_index = dev_cap_txqueue(dev, queue_index);


Produces a number of these in the logs:

Feb 22 10:46:50 foo kernel: dev_pick_tx: bond0 queue_mapping=16, real_num=16

Phil