[CentOS] CentOS-4 Xen kernel with low RAM and Badness in local_bh_enable at kernel/softirq.c:141

Wed Dec 3 14:36:40 UTC 2008
Ross Walker <rswwalker at gmail.com>

On Dec 3, 2008, at 4:09 AM, Kingsly John <member+centos at kingsly.net>  
wrote:

> Sorry about the multiple mails that came through to the list.
>
> I'd been trying to send them out for over 20 hours and they wouldn't  
> relay
> through a openvpn tunnel because of the "Badness".
>
> I had moved the queue manually and forgot to remove the mail that went
> through.
>
>> Dec  2 16:05:21 noc kernel: Badness in local_bh_enable at kernel/ 
>> softirq.c:141
>> Dec  2 16:05:21 noc kernel:  [<c0121178>] local_bh_enable+0x47/0x6f
>> Dec  2 16:05:21 noc kernel:  [<c02177d5>] skb_checksum+0x133/0x25e
>> Dec  2 16:05:21 noc kernel:  [<c0250f06>] udp_poll+0x66/0x113
>> Dec  2 16:05:21 noc kernel:  [<c02135fd>] sock_poll+0x19/0x1d
>> Dec  2 16:05:21 noc kernel:  [<c016d1a6>] do_select+0x190/0x2c7
>> Dec  2 16:05:21 noc kernel:  [<c016ceb5>] __pollwait+0x0/0x9b
>> Dec  2 16:05:21 noc kernel:  [<c0144ae4>] __kmalloc+0x56/0xd3
>> Dec  2 16:05:21 noc kernel:  [<c016d5dc>] sys_select+0x2e7/0x45c
>> Dec  2 16:05:21 noc kernel:  [<c010740f>] syscall_call+0x7/0xb
>
> This seems to happen on certain mails which are relayed through an  
> openvpn
> connection on the router VM.(possibly everytime sendmail tries to  
> clear the
> queue.) And the mail never gets out and the the connection times out.
>
> All test mails (one liners) went out without any problems via the VPN.
>
> After sending one mail by-passing the VPN, i went ahead and switched  
> the
> kernel.. and now all my mails are relaying.

As you found out you need more memory for the RH 2.6.18 kernel then  
then stock one which Xen.org uses because RH backports later kernel  
features and enables advanced features.

The errors you are seeing are network related so it might be checksum  
offloading in dom0 and domU is getting in the way.

Use ethtool to disable checksum offloading in dom0 and the domU.

This is a newer Xen "feature" then what was in your older kernel you  
could just try disabling checksum offloading in the domU only.

-Ross