DragonFlyBSD

crash trace attached.

Nuno.

:New submission from Nuno Antunes <nuno.antunes@gmail.com>:
:
:Hi,
:
:I got this crash while make buildworld was running inside a vkernel and xor=
:g was
:compiling in the real host.
:
:I'm attaching screenshots of the crash and trace.
:
:Regards,
:Nuno

That is one I haven't seen before.  I will attempt to reproduce it.

A kernel crash dump would be nice, it shouldn't be possible for it
    to get a null-pointer derefence there.  Hmmm.

-Matt

::Hi,
::
::I got this crash while make buildworld was running inside a vkernel and xor=
::g was
::compiling in the real host.
::
::I'm attaching screenshots of the crash and trace.
::
::Regards,
::Nuno
:
: That is one I haven't seen before. I will attempt to reproduce it.
:
: A kernel crash dump would be nice, it shouldn't be possible for it
: to get a null-pointer derefence there. Hmmm.
:
: -Matt

I think I know what is going on here.  There are a couple of places
    in the pmap code where the PV list is being scanned and in the
    particular place shown in the snapshot you took, it is possible that
    a page with active I/O on it (i.e. when the system is under a heavy
    paging load) can cause the calling process to block.  This can cause
    the list scan to go stale and blow up in our faces.

I will commit a workaround.

-Matt
                    Matthew Dillon 
                    <dillon@backplane.com>

Fixed by the following commit:

http://leaf.dragonflybsd.org/mailarchive/commits/2007-01/msg00229.html

Nuno.