-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Steven Rostedt wrote:
> On Thu, 11 May 2006, Clark Williams wrote:
>
>> Robert Crocombe wrote:
>>> testing NMI watchdog ... OK.
>>>
>>> where we croak.  On the non-realtime version, it is instead like so:
>> Yeah, this is where my Athlon64x2 goes into the weeds.  I followed it
>> down into the routines that calculate processor migration costs and
>> got lost in the context switches.  I suspect that the forced
>> migrations aggravate a problem down in the 64-bit arch specific code
>> that hasn't been looked at in a while (I believe most people running
>> AMD are doing so in 32-bit mode).
>>
>> Maybe it's time for round two...
>>
>
> Hmm, I run my AMDx2 in 64bit kernel mode and 32bit user space.  But since
> it is my main machine I don't usually test the new -rt kernels on it that
> often.  Looks like when I'm back in the States, I'll have to test it
> again.
>
> -- Steve
>
I spent some time this week trying to dig a bit deeper into why a
64-bit -rtXX won't boot on my Athlon64x2.

I've attached the output from one of my kernel boots (2.6.16-rt23)
that has lots of extra printks in kernel/sched.c and some in
arch/x86_64/kernel/smpboot.c.  After spending time looking at the code
around the migration measurement, it occurred to me that things didn't
really look right for the migration thread on processor 1. I stuck a
print in the routine migration_thread() and noticed that I never
actually saw the migration thread for processor 1 get started (I did
see the print when migration_init() started the migration thread on
processor 0). Since the migration measurement code depends on all
migration threads being present, I suspect that somehow the kernel is
not completing initialization for the second processor.

My only thought at this point is to start comparing
arch/x86_64/kernel/smpboot.c to arch/i386/kernel/smpboot.c and see if
something isn't being done on the x86_64 side.

Clark



-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.3 (GNU/Linux)
Comment: Using GnuPG with Fedora - http://enigmail.mozdev.org

iD8DBQFEbfdnHyuj/+TTEp0RAnIQAKDEYjJgVho6RXLllgDolZKGRGpPKACfTL/j
s4N1Cg9R3Sh/v87xfx45Vcc=
=WV8C
-----END PGP SIGNATURE-----