Re: upgrade to 3.8.1 : BUG Scheduling while atomic in bonding driver:

On 03/07/2013 04:50 PM, Linda Walsh wrote:

I am *not* seeing the bug in 3.8.2 with the 2nd patch applied (in
addition to the first)...

So that means bond lock is the reason, nice, but this is really not a
good fix if we just unlock it...

The better way is to move the cycle wait logical out of the
bond_update_speed_duplex() IMO, I think we need the folk who work on
this driver to make the decision ;-)

Regards,
Michael Wang

Michael Wang wrote:

quoted

And both bond_enslave() and bond_mii_monitor() are using bond_update_speed_duplex()
with preempt disabled.

Along with the changes in bond_enslave(), I think you also need this (untested):

diff --git a/drivers/net/bonding/bond_main.c b/drivers/net/bonding/bond_main.c
index 11d01d6..9af143a 100644
--- a/drivers/net/bonding/bond_main.c
+++ b/drivers/net/bonding/bond_main.c

@@ -2373,7 +2373,9 @@ static void bond_miimon_commit(struct bonding *bond)
                                bond_set_backup_slave(slave);
                        }

+                       read_unlock(&bond->lock);
                        bond_update_speed_duplex(slave);
+                       read_lock(&bond->lock);

                        pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
                                bond->dev->name, slave->dev->name,

Regards,
Michael Wang