Re: [Xen-devel] Badness in local_bh_enable

[Nick Craig-Wood <nick@xxxxxxxxxxxxxx>]

(re-opening an old thread)

On Tue, Mar 08, 2005 at 11:05:01AM -0000, Ian Pratt wrote:
>  > nic@stateless:~$ strings 
> > /lib/modules/2.6.10-xen0/kernel/net/rxrpc/rxrpc.ko | grep vermagic=
> > vermagic=2.6.10-xen0 preempt PENTIUM4 gcc-3.3 
> > 
> > These seems to be compiled with ARCH=xen.
> 
> Not necessarily. You may have used that -xen0 tree, but not specified
> ARCH=xen. There's no way to tell :-(
> 
> > nic@stateless:~$ strings 
> > /lib/modules/2.6.10-xenU/kernel/net/ipv4/netfilter/ipt_REJECT.
> > ko | grep vermagic=
> > vermagic=2.6.10-xenU preempt PENTIUM4 gcc-3.3
> > nic@stateless:~$ objdump -d 
> > /lib/modules/2.6.10-xenU/kernel/net/ipv4/netfilter/ipt_REJECT.
> > ko | egrep  -e sti
> >  b3f:   fb                      sti
> 
> Just finding a cli/sti in the disassembled output does not necessarily
> indicate a problem -- objdump frequently gets confused and disassembles
> things that are data, particularly after an undefined instruction e.g.
> the uda2 used for BUG messages.
> 
> You'll need to look at the instructions around the cli/sti to determine
> whether they're real or not.

Was this problem ever got to the bottom of?

We are seeing it too.  Lots of 

Badness in local_bh_enable at kernel/softirq.c:140
 [<c011fb12>] local_bh_enable+0x82/0x90
 [<c031fcfd>] skb_checksum+0x13d/0x2d0
 [<c016ac5c>] __pollwait+0x8c/0xd0
 [<c0360d3a>] udp_poll+0x9a/0x160
 [<c031af49>] sock_poll+0x29/0x40
 [<c016b635>] do_pollfd+0x95/0xa0
 [<c016b6aa>] do_poll+0x6a/0xd0
 [<c016b871>] sys_poll+0x161/0x240
 [<c011f14c>] sys_gettimeofday+0x3c/0x90
 [<c016abd0>] __pollwait+0x0/0xd0
 [<c0109758>] syscall_call+0x7/0xb

In the logs.

I've checked the loaded modules...

Module                  Size  Used by
sch_sfq                 4992  88 
cls_u32                 8068  101 
sch_cbq                15616  13 
ipt_state               1664  2 
ipt_REJECT              5888  3 
ipt_LOG                 6528  4 
ipt_limit               2176  4 
iptable_mangle          2432  0 
iptable_filter          3200  1 
ip_nat_ftp              4464  0 
ip_conntrack_ftp       71600  1 ip_nat_ftp
iptable_nat            23496  1 ip_nat_ftp
ip_conntrack           41364  4 
ipt_state,ip_nat_ftp,ip_conntrack_ftp,iptable_nat
ip_tables              17024  7 
ipt_state,ipt_REJECT,ipt_LOG,ipt_limit,iptable_mangle,iptable_filter,iptable_nat
e1000                  84788  0 

And these all have the right vermagic

# strings `cat /tmp/modules` | grep vermagic
vermagic=2.6.10-x1d-p4 preempt PENTIUM4 gcc-3.3
vermagic=2.6.10-x1d-p4 preempt PENTIUM4 gcc-3.3
vermagic=2.6.10-x1d-p4 preempt PENTIUM4 gcc-3.3
vermagic=2.6.10-x1d-p4 preempt PENTIUM4 gcc-3.3
vermagic=2.6.10-x1d-p4 preempt PENTIUM4 gcc-3.3
vermagic=2.6.10-x1d-p4 preempt PENTIUM4 gcc-3.3
vermagic=2.6.10-x1d-p4 preempt PENTIUM4 gcc-3.3
vermagic=2.6.10-x1d-p4 preempt PENTIUM4 gcc-3.3
vermagic=2.6.10-x1d-p4 preempt PENTIUM4 gcc-3.3
vermagic=2.6.10-x1d-p4 preempt PENTIUM4 gcc-3.3
vermagic=2.6.10-x1d-p4 preempt PENTIUM4 gcc-3.3
vermagic=2.6.10-x1d-p4 preempt PENTIUM4 gcc-3.3
vermagic=2.6.10-x1d-p4 preempt PENTIUM4 gcc-3.3
vermagic=2.6.10-x1d-p4 preempt PENTIUM4 gcc-3.3
vermagic=2.6.10-x1d-p4 preempt PENTIUM4 gcc-3.3

# objdump -d `cat /tmp/modules` | grep -c cli
0

# objdump -d `cat /tmp/modules` | grep -c sti
2

Those two are clearly caused by bodged dissasembly, eg

 b65:   ff 8b 5c 24 38 85       decl   0x8538245c(%ebx)
 b6b:   db 0f                   fisttpl (%edi)
 b6d:   84 d2                   test   %dl,%dl
 b6f:   fb                      sti    
 b70:   ff                      (bad)  
 b71:   ff 8b 43 04 85 c0       decl   0xc0850443(%ebx)

I built the kernel using debian's make-kpkg which I wouldn't have
thought would have made a mistake.

Any other ideas?

This is kernel 2.6.10 running with xen 2.04.

-- 
Nick Craig-Wood <nick@xxxxxxxxxxxxxx> -- http://www.craig-wood.com/nick

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel