Re: [PATCH] let WARN_ON() output the condition

[Ingo Molnar <mingo@elte.hu>]

* Jiri Kosina <jikos@jikos.cz> wrote:

> On Wed, 6 Dec 2006, Ingo Molnar wrote:
> 
> > a WARN_ON() also triggers a stack dump, which should pinpoint the exact 
> > location. (especially if combined with kallsyms) For example:
> 
> Actually, I was referring to something a little bit different. For example 
> kernel/mutex.c:__mutex_lock_common() calls spin_lock_mutex() on line 132. 
> spin_lock_mutex() contains
> 
>                 DEBUG_LOCKS_WARN_ON(in_interrupt());    \
>                 local_irq_save(flags);                  \
>                 __raw_spin_lock(&(lock)->raw_lock);     \
>                 DEBUG_LOCKS_WARN_ON(l->magic != l);     \
> 
> When one of these two WARN_ONs trigger, we get only
> 
> 	WARNING at kernel/mutex.c:132 __mutex_lock_common()

no, that's not all we get - we should also get a stackdump. Are you not 
getting a stackdump perhaps?

but i agree with you in theory that your proposed output is better, but 
the side-effect issue is a killer i think. Could you try to rework it to 
not evaluate the condition twice and to make it dependent on 
CONFIG_DEBUG_BUGVERBOSE? You can avoid the evaluation side-effect issue 
by doing something like:

	int __c = (c);							\
                                                                        \
        if (unlikely(__c)) {                                            \
                if (debug_locks_off())                                  \
                        WARN_ON(__c);                                   \
                __ret = 1;                                              \

	Ingo