Re: mtrash_ctor panic caused by ACPI and ATAng was Re: Interrupts, ATAng, and wayward pointers, oh my!

On Fri, Feb 06, 2004 at 10:46:03PM -0800, Doug White wrote:
> On Fri, 6 Feb 2004, Steve Kargl wrote:
> 
> > Via trial and error, I have determined that the above panic is
> > caused by ACPI and ATAng.  This is a Dell 4150 laptop.  I can
> > build a working kernel with sources checked out via cvsup with
> > date=2004.01.30.19.00.00.  If I use a date of 2004.01.30.20.00.00,
> > I retrieve only revision 1.203 of ata-all.c and revision 1.19 of
> > ata-queue.h.  No other files are changed in sys/ and the resulting
> > kernel produces the above panic.  Finally, if I set
> > hint.acpi.0.disabled="1" in /boot/loader.conf.  The kernel that
> > previously panicked will boot fine.
> 
> Intriguing.

Yes, intriguing.  I used the better part of 1 day to eliminate 
possibilities for the problem.  I forgot to mention the problem
does not depend on APIC or no APIC support in the kernel.

> > PS: No, I can't get a core dump because the disk subsystem isn't
> >     ready for a dump when the panics occurred and I can't hook
> >     up a serial console.  I can panic the machine as needed.
> 
> You can say 'set dumpdev=/dev/foo0s1b' in loader to set the dump target.
> 

This doesn't work unless I can get a near instantaneous dump. That is,
db> call dumpsys
returns in under a second.  Revision 1.19 of ata-queue.c (note the
typo of ata-queue.h above) reworked the locking of ATA.  I suspect
that my laptop is tickling some race that Soren hasn't seen.

-- 
Steve