[PATCH] conditionalize call from panic() to smp_send_stop()

[PATCH] conditionalize call from panic() to smp_send_stop()

[Jan Beulich]

... avoiding nested warnings and/or oopses as much as possible (namely
durin early boot).

Signed-off-by: Jan Beulich <jbeulich@novell.com>

---
 kernel/panic.c |    3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

--- linux-2.6.29-rc1/kernel/panic.c	2009-01-14 11:36:29.000000000 +0100
+++ 2.6.29-rc1-panic-conditional-stop/kernel/panic.c	2009-01-06 17:02:49.000000000 +0100
@@ -89,7 +89,8 @@ NORET_TYPE void panic(const char * fmt, 
 	 * unfortunately means it may not be hardened to work in a panic
 	 * situation.
 	 */
-	smp_send_stop();
+	if (num_online_cpus() > 1)
+		smp_send_stop();
 #endif
 
 	atomic_notifier_call_chain(&panic_notifier_list, 0, buf);

Re: [PATCH] conditionalize call from panic() to smp_send_stop()

[Andi Kleen]

"Jan Beulich" <jbeulich@novell.com> writes:

> ... avoiding nested warnings and/or oopses as much as possible (namely
> durin early boot).

Better would be to fix smp_send_stop than to work around it for this
very limited case. I've been running into this with machine checks
too. The new fancy smp_call_function() that is there now just doesn't
work in any unusual situation like early boot or fatal exceptions,

My thinking was to use a separate vector, or possible just the NMI.

-Andi

-- 
ak@linux.intel.com -- Speaking for myself only.

Re: [PATCH] conditionalize call from panic() to smp_send_stop()

[Jan Beulich]

>>> Andi Kleen <andi@firstfloor.org> 15.01.09 04:34 >>>
>"Jan Beulich" <jbeulich@novell.com> writes:
>
>> ... avoiding nested warnings and/or oopses as much as possible (namely
>> durin early boot).
>
>Better would be to fix smp_send_stop than to work around it for this
>very limited case. I've been running into this with machine checks
>too. The new fancy smp_call_function() that is there now just doesn't
>work in any unusual situation like early boot or fatal exceptions,

Indeed - I wasn't used to smp_call_function...() to be generic functions
that can be fixed for all architectures at once. I simply wanted to avoid
needing individual changes in all architectures...

Jan