pmap_emulate_reference() panics still ongoing
Andrew Gallatin
gallatin at cs.duke.edu
Wed Apr 28 07:16:06 PDT 2004
Ruslan Ermilov writes:
> On Wed, Apr 28, 2004 at 08:39:11AM -0400, Andrew Gallatin wrote:
> >
> > Ruslan Ermilov writes:
> > > Andrew,
> > >
> > > This is with latest src/sys/alpha/alpha/pmap.c,v 1.146.
> > >
> > > db> where
> > > Debugger() at Debugger+0x38
> > > __panic() at __panic+0x228
> > > pmap_emulate_reference() at pmap_emulate_reference+0x15c
> > > trap() at trap+0x3dc
> > > XentMM() at XentMM+0x2c
> > > --- memory management fault (from ipl 0) ---
> > > --- user mode ---
> >
> > Damn. Did the change make it any worse for you?
> >
> No, I can't say this made things any worse. ;)
Good.
> This is not a very fast runner (166MHz EV4, 64MB RAM, swap
> is used). Since I've upgraded it to 5-CURRENT, running the
> GENERIC kernel, my 5 attempts to build world ended up with
> this panic (often the panic message was unreadable, this one,
> after I've upgraded kernel to the latest pmap.c, was probably
> a good exception). Now I've built the debug version of the
> kernel. Will it be of some help?
Wow, and I thought my 600MHz ev67 was slow. At least you can
reproduce this -- I can't.
Given that I always run a bwx-enabled kernel, and I never see it, I
wonder if there might be some corruption caused by 8 or 16 bit data
fields in the vm or pmap systems which share a 32-bit word, but which
are separately locked. I'm thinking that the read-modify-write which a
non bwx kernel needs to do could lead to data-corruption in this case.
I don't see anything immediately, but then I don't know the vm system
very well.
> (Hardly relevant, but it was happy running 4.10-RC.)
Very relevant -- it means the bug is new to 5.x
Drew
More information about the freebsd-alpha
mailing list