Re: kern/43893: Hard lock under moderate filtering web-proxy use

[Stephen Borrill <netbsd%precedence.co.uk@localhost>]

On Sat, 16 Oct 2010, Mindaugas Rasiukevicius wrote:
> netbsd%precedence.co.uk@localhost wrote:
> > db{0}> bt
> > tcp_output(c5830a04,c9e69c2e,0,d80c4c9c,2c,0,d80c4bcc,ff,0,0) at
> > netbsd:tcp_output+0x81 tcp_input
> > (c58c4500,14,6,1,d80c0010,d80c0010,d5975ce0,d80c0154,14,c58c4500) at
> > netbsd:tcp_input+0x1fb4 ip_input
> > (c58c4500,0,43b83278,0,0,d80c01ec,d80c4d80,c04ed8df,0,d96c0010) at
> > netbsd:ip_input+0x6a1 ipintr
> > (0,d96c0010,d80c0030,10,d80c0010,0,6cc360,d9723bac,0,d80c4da0) at
> > netbsd:ipintr+0x4d softint_dispatch
> > (d96cc860,4,0,0,0,0,d80c4d90,d80c4bd8,d80c4c30,0) at
> > netbsd:softint_dispatch+0x9f DDB lost frame for netbsd:Xsoftintr+0x3d,
> > trying 0xd80c4d88 Xsoftintr() at netbsd:Xsoftintr+0x3d
>
> Can you objdump -D what is on tcp_output+0x81?

I've done objdump -D netbsd and searched for tcp_output. I guess you are 
looking around c0161bc1:

c0161b40 <tcp_output>:
[snip] 
c0161bb3:       8b ba d0 00 00 00       mov    0xd0(%edx),%edi
c0161bb9:       85 ff                   test   %edi,%edi
c0161bbb:       0f 84 33 10 00 00       je     c0162bf4 <tcp_output+0x10b4>
c0161bc1:       c7 45 a8 00 00 00 00    movl   $0x0,0xffffffa8(%ebp)
c0161bc8:       31 c0                   xor    %eax,%eax
c0161bca:       c7 45 a0 28 00 00 00    movl   $0x28,0xffffffa0(%ebp)
c0161bd1:       85 ff                   test   %edi,%edi

> Also, is it always 0x81 on this trace when it lock-ups?

I'll have to get back to you on that. The end user is currently running 
-current - they have reported that it seems to be much more stable than 
netbsd-5 (which is an interesting datapoint in itself). We're going to 
switch back to netbsd-5 today and wait for another lock up.

--
Stephen