Re: [BUG 2.6.30-rc1] panic when loading oprofile

From: Andi Kleen
Date: Fri May 15 2009 - 08:41:42 EST


"Brandeburg, Jesse" <jesse.brandeburg@xxxxxxxxx> writes:

Hi Jesse,

> when starting a profile run on the latest net-next kernel, I'm currently
> trying to reproduce on 2.6.30-rc5 stock.

Were you able to reproduce it?

>
> config available upon request, arch=x86_64, recent (F10 or newer) oprofile
> userspace.

it looks like two bugs: oprofile didn't catch a NMI that belongs to
it (most likely) and the NMI watchdog referenced a NULL pointer
while processing an NMI.

Did you have the nmi watchdog enabled on the command line?

>
> BUG: unable to handle kernel NULL pointer dereference at (null)
> IP: [<ffffffff8066080a>] nmi_watchdog_tick+0xa1/0x1d6

I don't get the same code as you. But the oopsing instruction in your
oops is

2b:* 44 0f a3 28 bt %r13d,(%rax) <-- trapping instruction

with rax == 0 and I suspect it's one of the new cpu mask checks
I would try reverting

fcc5c4a2feea3886dc058498b28508b2731720d5
2f537a9f8e82f55c241b002c8cfbf34303b45ada
fcef8576d8a64fc603e719c97d423f9f6d4e0e8b

and see which one causes it. That would only fix the NMI watchdog bug
of course.

The oprofile not catching a event problem would be still open then.
I think the checks for overflowed counters are not 100% perfect
so that could happen. I have some patches in the works to use the new
global status register on arch perfmon 2, with that the overflow
check is somewhat more reliable. But that's more work.

-Andi

--
ak@xxxxxxxxxxxxxxx -- Speaking for myself only.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/