Re: [PATCH 26/28] kdb,panic,debug_core: Allow the debug core to receive a panic before smp_send_stop()

[Eric W. Biederman]

Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> writes:

> On Fri, 12 Feb 2010 17:43:39 -0600 Jason Wessel <jason.wessel@xxxxxxxxxxxxx> wrote:
>
>> >>  	printk(KERN_EMERG "Kernel panic - not syncing: %s\n",buf);
>> >>  #ifdef CONFIG_DEBUG_BUGVERBOSE
>> >>  	dump_stack();
>> >> @@ -91,8 +94,6 @@ NORET_TYPE void panic(const char * fmt, ...)
>> >>  	 */
>> >>  	smp_send_stop();
>> >>  
>> >> -	atomic_notifier_call_chain(&panic_notifier_list, 0, buf);
>> >> -
>> >>  	bust_spinlocks(0);
>> >>  
>> >>  	if (!panic_blink)
>> >>     
>> >
>> > So the notifier call now happens before all the printks and the kexec
>> > and kmsg_dump handling.  What effect does this have upon the code which
>> > implements kexec and kmsg_dump?
>> >
>> >   
>> 
>> I certainly don't want to break kexec or alter any behavior, does that
>> mean kgdb / kdb should hook the kexec for notification?
>> 
>> I think ideally it is a end user's preference as to if they want in via
>> kexec or the kernel debugger.  Calling the smp_send_stop() prior to the
>> notifier was a death sentence for the kernel debugger. 
>> 
>> Perhaps I can move the notifier before smp_send_stop()?
>
> Well.  My question can be simplified to "does this break existing code"?

Yes.  Removing the bust_spinlocks(1) and moving the panic notification up
will measurably decrease the reliability of crash dump capture.  I expect
it will even result in dead_locks where we fail to print the panic message
for people who are trying to debug via the console.

Eric
--
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/