9.2 + ZFS + i386 = panic: pmap_enter: attempted pmap_enter on 4MB page

Konstantin Belousov kostikbel at gmail.com
Tue Oct 22 17:36:09 UTC 2013


On Tue, Oct 22, 2013 at 08:07:53PM +0300, Andriy Gapon wrote:
> on 22/10/2013 15:46 J David said the following:
> > On Sun, Oct 20, 2013 at 9:02 AM, J David <j.david.lists at gmail.com> wrote:
> >> The "attempted pmap_enter on 4MB page" panic occurred again.
> > 
> > Since a machine that crashes 3-4x a day is not useful, we have
> > captured a few viable core images for this panic and have now set it
> > to vm.pmap.pg_ps_enabled=0 in the hopes of stopping the crashes.
> > 
> > If there is anything else we can do to aid the debugging, please let us know.
> 
> Both panics you posted actually happened because of bad pointer
> dereference in ZFS.
>
> It appears that for some reason trap_pfault does not just call panic
> right there as it should, but instead attempts to "fault in" a
> corresponding page.
>
> All in all, these look like a memory corruption of some sort (or use
> after free).

I just do not have time left for this right now.

One thing that catched my eye initially are very strange pointers in the
backtrace, like 'zio=0x883ca8b8'. The standard i386 address space map
splits user/kernel mode at 3GB, so 0x883ca8b8 is definitely usermode,
unless non-standard kernel config is used.

This is why I asked for the gdb backtrace.
why 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 834 bytes
Desc: not available
URL: <http://lists.freebsd.org/pipermail/freebsd-fs/attachments/20131022/2db72b2d/attachment.sig>


More information about the freebsd-fs mailing list