On Tue, 16 Jan 2007, Manuel Bouyer wrote:

> 
> On Tue, Jan 16, 2007 at 01:52:19PM +0000, Matthias Scheler wrote:
>> On Tue, Jan 16, 2007 at 03:26:48AM +0000, reed@reedmedia.net wrote:
>>> Anyone seen anything like this?
>>>
>>> My kernel is netbsd-3.
>>
>> This pullup might fix your problem:
>>
>> http://mail-index.netbsd.org/source-changes/2007/01/15/0016.html
>>
>
> I don't think so, this pullup was to fix a possible panic at boot with more
> than one xbd.
> This commit could be the fix:
> http://mail-index.netbsd.org/source-changes/2007/01/01/0033.html
>
> but the author has not submitted a pullup request to netbsd-3 yet ...

Hm.  I am the xen provider in question  (we are a small shop, and I have a 
dayjob, so I haven't gotten around to really looking at it until now)  but 
using the same kernel, I do get occasional heavy disk load panics. 
Actually I just triggered one right now (incidentally while I was 
attempting to manually apply the patch you suggested;  all my NetBSD boxes 
at the moment are XenUs. )

here's the crash:

uvm_fault(0xc044ec80, 0xcaa8a000, 0, 2) -> 0xe
kernel: page fault trap, code=0
Stopped in pid 8.1 (aiodoned) at        netbsd:softdep_pageiodone+0x248: 
m
ovl     %edx,0(%eax)
softdep_pageiodone(c09e30fc,3415c,c0745380,c9bab108,c9bab318) at 
netbsd:softdep_
pageiodone+0x248
uvm_aio_aiodone(c09e30fc,204,c03f755f,0,c047a7e0) at 
netbsd:uvm_aio_aiodone+0x4d
b
uvm_aiodone_daemon(c9bab318,4fc000,c0506000,0,c01001cc) at 
netbsd:uvm_aiodone_da
emon+0xc9
ds          0x11
es          0x11
fs          0x31
gs          0x11
edi         0x800
esi         0x3fff
ebp         0xca504efc
ebx         0
edx         0
ecx         0xcca0bf08
eax         0xcaa8a130
eip         0xc025e0b8  softdep_pageiodone+0x248
cs          0x9
eflags      0x10246
esp         0xca504e94
ss          0x11
netbsd:softdep_pageiodone+0x248:        movl    %edx,0(%eax)
Stopped in pid 8.1 (aiodoned) at        netbsd:softdep_pageiodone+0x248: 
m
ovl     %edx,0(%eax)


and the bt
db> bt
softdep_pageiodone(c09e30fc,3415c,c0745380,c9bab108,c9bab318) at 
netbsd:softdep_
pageiodone+0x248
uvm_aio_aiodone(c09e30fc,204,c03f755f,0,c047a7e0) at 
netbsd:uvm_aio_aiodone+0x4d
b
uvm_aiodone_daemon(c9bab318,4fc000,c0506000,0,c01001cc) at 
netbsd:uvm_aiodone_da
emon+0xc9
db>


(the disk, incidentally, was full, and it does have two xbd devices, and 
heavy activity on one.)


I've extracted and applied the patch in question; (It applies cleanly, 
btw, save for the comments)   I'm re-compiling 3.1 with the patch applied 
presently.