Re: Panic on boot with new ACPI-CA

On Thursday 03 November 2005 11:14 am, Peter Holm wrote:
> On Thu, Nov 03, 2005 at 10:43:10AM -0500, John Baldwin wrote:
> > On Thursday 03 November 2005 09:40 am, Peter Holm wrote:
> > > On Thu, Nov 03, 2005 at 04:24:46PM +0200, Giorgos Keramidas wrote:
> > > > On 2005-11-03 03:47, Giorgos Keramidas <keramida_at_ceid.upatras.gr> 
wrote:
> > > > >On 2005-11-02 17:03, Nate Lawson <nate_at_root.org> wrote:
> > > > >> As I mentioned to Jung-uk, the problem is likely an error in
> > > > >> acpi-ca modifying memory after it has freed it.  The way to
> > > > >> track this down is to enable memguard(9).  See the man page for
> > > > >> info.  You need to add options DEBUG_MEMGUARD to your kernel,
> > > > >> set the malloc type to watch to M_ACPICA, and rebuild your
> > > > >> kernel and modules.  Memguard sets page permissions so we can
> > > > >> catch the culprit who is modifying the memory.
> > > > >
> > > > > This is exactly the messgae printed on my console at panic time
> > > > > -- of memory modified after free.  I'm building a kernel with
> > > > > MEMGUARD now, but it's probably going to be a bit hard to get a
> > > > > kernel dump, because the panic happens before disks are
> > > > > available and I don't have a serial console here.
> > > >
> > > > This is definitely something that is ACPI-related.  I updated my
> > > > sources to the last commit before the start of the ACPI import:
> > > >
> > > >     build_at_flame:/home/build/src$ cvs -qR up -APd -D '2005/11/01
> > > > 22:00:00 UTC'
> > > >
> > > > Rebuilt everything and I see no panics now.
> > > >
> > > > I'll use the watchpoint trick Nate posted when I have a new build
> > > > to test.
> > >
> > > I've had the same problem with two of my boxes. Here's the result of a
> > > watchpoint:
> > >
> > > http://people.freebsd.org/~pho/stress/log/acpi.html
> > >
> > > I too came to the conclusion that the damage happened between
> > >
> > > 2005-11-01 22:00:00 UTC OK
> > > 2005-11-01 22:45:00 UTC panic
> >
> > Does this diff make a difference perhaps?
>
> It did not make any difference as far as I can tell.

Ok, it was more of a guess anyway that perhaps ACPICA was reusing the garbage 
in buf.Pointer.

-- 
John Baldwin <jhb_at_FreeBSD.org>  <><  http://www.FreeBSD.org/~jhb/
"Power Users Use the Power to Serve"  =  http://www.FreeBSD.org