2.6.12-rc1-RT-V0.7.41-15: it_real_fn oops on boot in run_timer_softirq

2.6.12-rc1-RT-V0.7.41-15: it_real_fn oops on boot in run_timer_softirq

[Lee Revell]

Since 2.6.12-rc1-RT something I get this Oops on boot about 50% of the
time.  It's clearly some kind of race because if I just reboot again it
works.  Seems to happen shortly after ksoftirqd startup (maybe the first
time we hit the timer softirq?).

This is (lazily) hand copied and incomplete, but hopefully is enough
info...

EIP is at it_real_fn+0x2f/0x70
eax 0 ebx df9020f0  ecx 1 edx c0019b49
esi 0 edi "       " ebp dffd5f44 esp  dffd5f34
ds 7b es 7b ss 68 preempt 1
process ksoftirqd/0, pid 2, tid dffd4000, task dffd1110

call trace
 show_stack
 show_registers
 die
 do_page_fault
 error_code
 run_timer_softirq
 ___do_softirq
 _do_softirq
 ksoftirqd
 kthread
 kthread_helper

Lee

Re: 2.6.12-rc1-RT-V0.7.41-15: it_real_fn oops on boot in run_timer_softirq

[Ingo Molnar]

* Lee Revell <rlrevell@joe-job.com> wrote:

> Since 2.6.12-rc1-RT something I get this Oops on boot about 50% of the 
> time.  It's clearly some kind of race because if I just reboot again 
> it works.  Seems to happen shortly after ksoftirqd startup (maybe the 
> first time we hit the timer softirq?).
> 
> This is (lazily) hand copied and incomplete, but hopefully is enough 
> info...

hm, does the patch below help?

If not then please try to capture the full oops via serial logging, or 
if it's not possible, here are some guidelines wrt. how to minimize work 
when writing down oopses by hand:

1)

write down the EIP of the crash, and do:

	objdump -d vmlinux > vmlinux.asm
	grep -C 100 c0xxxxxx vmlinux.asm > toingo.asm

c0xxxxxx is the EIP of the oops. This way i'll know precisely which 
instruction crashed, and can match it up in my kernel image, even if i 
have a different .config and different build environment.

2)

Or write down the 'EIP is at ...' line plus ~10 bytes of the 'Code: ' 
line of the oops too, starting at the byte that is enclosed by <>. This 
info is enough in most cases.

	Ingo

--- linux/kernel/itimer.c.orig
+++ linux/kernel/itimer.c
@@ -138,7 +138,8 @@ void it_real_fn(unsigned long __data)
 	 * here because do_setitimer makes sure we have finished running
 	 * before it touches anything.
 	 */
-	it_real_arm(p, p->signal->it_real_incr);
+	if (p->signal)
+		it_real_arm(p, p->signal->it_real_incr);
 }
 
 int do_setitimer(int which, struct itimerval *value, struct itimerval *ovalue)