Here is an older kernel which seems to die the same way.
I actually can not see mtx_enter-loop, but I trigger crash the same way. Remote
console via IPMI just blinks like a X-mass tree.
OpenBSD 5.2-current (GENERIC.MP) #0: Tue Nov 6 17:03:43 CET 2012
[email protected]:/usr/src/sys/arch/amd64/compile/GENERIC.MP
On 3 jan 2013, at 20:15, mxb <[email protected]> wrote:
>
> Now, after several tests I can state that problem is there.
> I made sure that /usr/src is clean and up to date (I had Hennings diffs on
> test).
>
> The stock -current kernel crashes with this behavior, eg. panic and loop with
> mtx_enter on console.
> This bug is triggered then I transfer /bsd over IPSec. While scp from the
> same machine, but on public side goes fine.
>
> client (network1)<-----> fw1 ---> [---- ospf over gre over ipsec] <----- fw2
> <-------->network2
>
> client does 'scp fw2.network2_ip:/bsd .' - results in panic.
> client does 'scp fw2.public_ip:/bsd .' - all fine.
>
> Of cause I have CARP both on external and internal sides of fw2.
> CARP for internal network on fw1.
>
> Side note is that I noticed drastic speed drop just before system goes in
> panic.
> Normally I have decent speed between two networks, e.g. transfers from
> clients on network2 to client on network1.
>
> Any ideas?
>
> //mxb
>
> On 3 jan 2013, at 13:39, Stuart Henderson <[email protected]> wrote:
>
>> On 2013-01-03, mxb <[email protected]> wrote:
>>> Sorry for the noise. I think I'v found the problem.
>>
>> Any more information on what it was??
>> Even if it was something you've done, that should't happen..
>>
>>
>>
>>>> On 1 jan 2013, at 19:11, mxb <[email protected]> wrote:
>>>>>
>>>>> I'v got yet another panic.
>>>>> This time, after applying Martin Pelikans' diff, catched a pointer.
>>>>> However, machine never drops to ddb, even sysctl.conf says it should.
>>>>>
>>>>> panic: mxt_enter: locking against myself, 0xffffffff80a2d540
>>>>> kernel: privileged instruction fault trap, code=0
>>>>> kernel: double fault trap, code=0
>>>>>
>>>>> I was able to reproduce this several times and whenever I wanted it.
>>>>>
>>>>> Here is scenario(or setup) then I was able to trigger:
>>>>> Two networks with OSPF-routing on top of GRE on top of IPSec.
>>>>> Client on network1 starts scp-download from a firewall(fw1) on network2.
>>>>> fw1 acts as a VPN/OSPF/GRE end-point for network2.
>>>>> After some time(1min or so) fw1 goes down with panic above.
>>>>> Basically I tried to scp down kernel from this machine.
>>>>>
>>>>> However, fw1 never goes down then I "scp up" from an Internal network
>>>>> behind fw1.
>>>>> I my case this was a patched kernel with Martins' diff from a VM-machine
>>>>> sitting behind fw1.
