Re: 10.0-CURRENT: kernel: Fatal trap 12: page fault while in kernel mode

On 07/06/12 00:21, O. Hartmann wrote:
> On 07/05/12 23:27, John Baldwin wrote:
>> On Tuesday, July 03, 2012 1:43:30 am O. Hartmann wrote:
>>> The most recent build of FreeBSD 10.0-CURRENT crashes on one of our
>>> boxes with recent Intel hardware, see dmesg extract below. FreeBSD does
>>> obviously only crash on hardware with modern "Sandy-Bridge" hardware,
>>> the very same kernel config and a very similar setup does work very well
>>> on an older Intel Core2Dua architecture.
>>>
>>> Machine in question runs an older kernel (set via nextboot -k
>>> kernel.old) very well:
>>> FreeBSD 10.0-CURRENT #0 r237697: Thu Jun 28 11:45:08 CEST 2012
>>>
>>> Most recent buildworls/kernels crash after going into multiuser mode
>>> (/etc/rc-scripts getting executed). Below the kernel trap message.
>>>
>>> I do this report on the fly, so if there is need for deeper
>>> investigations let me know, I will provide necessary detail requested.
>>
>> We would need a stack trace to debug this further.  Do you have a crashdump?
>>
> 
> Well, I built a GENERIC kernel and bootet that, it also crashes, but I
> didn't find any coredump (as found in the crashes with the custom
> kernel). I was able to do a backtrace with the GENERIC kernel, but
> incapable of saving the trace - I did this in a rush, sorry.
> 
> I went back, just on a hunch, to FreeBSD 10.0-CURRENT #1 r237000, which
> works, but this isn't the last working revision.
> 
> I will look tomorrow for how to generate a dump or backtrace and save
> that somehow (if available in the handbook). Otherwise, if not not much
> efford, I would appreciate any hint (for the impatient) how to generate
> suiatbel debug informations.
> 
> Regards,
> Oliver
> 
Yesterday, I made a buildworld with the most recent sources and the
system is now running well and shiny with

FreeBSD 10.0-CURRENT #0 r238164: Fri Jul  6 15:07:59 CEST 201 (amd64)

I saw a lot of pmap.h stuff being changed for amd64 last days.


If it would be of interest, I can go back in the source's revision and
build again a faulty kernel.

Regards,
Oliver