ULE status, invalid load, buildkernel times.

[Peter Wemm]

On Friday 27 July 2007, Milos Vyletel wrote:
> On Fri, Jul 27, 2007 at 09:26:40AM -0400, Kris Kennaway wrote:
> > On Fri, Jul 27, 2007 at 09:48:32AM +0200, Milos Vyletel wrote:
> > > On Thu, Jul 26, 2007 at 06:06:20PM -0700, Peter Wemm wrote:
> > > > The other option is to find the kernel.debug for this crash,
> > > > and do this:
> > > > kgdb kernel.debug
> > > > gdb>  l *0xffffffff8033953c
> > > > This will tell us the file and line number that the crash
> > > > happened in. There is no need to reboot for this unless you no
> > > > longer have a crashing kernel.
> > >
> > > I've played with this a little while, and after turning
> > > INVARIANTS on, it paniced in lapic_ipi_raw() on the
> > > KASSERT(lapic != NULL, ("%s called too early", __func__));
> > >
> > > so I assume, that this function was called before lapic_init(),
> > > where lapic is initialized, which is wrong.
> > >
> > > It was clean current kernel with no other patches, now I don't
> > > have local access to that machine so I can test it in few days.
> > >
> > > btw. how can one get trace in text form, I mean syslog stop after
> > > panic and all I got logged is that it paniced. Anything I type in
> > > db> is lost. I know that this can be done by remote gdb, but
> > > unfortunatelly this isn't possible.
> >
> > If you trigger a dump (call doadump) then some amount of the DDB
> > session will usually be saved with the dump and displayed by kgdb.
>
> Yes, I forgot about that. I have zfs swap partition and I can't
> configure my dumpdev. Have anyone succesfully acomplish this?

Please abandon the original patch and try this one:

http://people.freebsd.org/~peter/topology.diff

This should fix it once and for all.  I have tested it on amd64.  I have 
compile tested on i386 (an old laptop), but not booted it on an i386 
smp box - I dont have one.

As such, I'd very much appreciate any confirmation from i386 users that 
it works.  Both from HTT and non-HTT users.  Please check that
sysctl kern.sched.topology is right.  On an Athlon64 X2 or dual-core 
opteron, it should be "0".  A HTT p4 or xeon should be non-zero.  It 
should be the same in both i386 and amd64 mode.
-- 
Peter Wemm - peter at wemm.org; peter at FreeBSD.org; peter at yahoo-inc.com
"All of this is for nothing if we don't go to the stars" - JMS/B5