Re: [patch] x86: some lock annotations for user copy paths

[Nick Piggin]

On Wed, Sep 10, 2008 at 05:01:04PM +0200, Peter Zijlstra wrote:
> On Wed, 2008-09-10 at 16:48 +0200, Nick Piggin wrote:
> 
> > @@ -3016,3 +3016,18 @@ void print_vma_addr(char *prefix, unsign
> >  	}
> >  	up_read(&current->mm->mmap_sem);
> >  }
> > +
> > +void might_fault(void)
> > +{
> > +	/*
> > +	 * it would be nicer only to annotatea paths which are not under
> > +	 * pagefault_disable, however that requires a larger audit and
> > +	 * providing helpers like get_user_atomic.
> > +	 */
> > +	if (!in_atomic()) {
> > +		might_sleep();
> > +		if (current->mm)
> > +			might_lock_read(&current->mm->mmap_sem);
> > +	}
> > +}
> > +EXPORT_SYMBOL(might_fault);
> 
> >From the nitpick squad :-), I prefer the form:
> 
> void might_fault(void)
> {
> 	if (in_atomic())
> 		return;
> 
> 	might_sleep();
> 
> 	if (!current->mm)
> 		might_lock_read(&current->mm->mmap_sem);
> }
> 
> 
> Due to it being one nesting level less.

Well... yeah. I find it doesn't matter unless the function is complex or
nesting level high in absolute terms ;)

 
> > Index: linux-2.6/include/linux/kernel.h
> > ===================================================================
> > --- linux-2.6.orig/include/linux/kernel.h
> > +++ linux-2.6/include/linux/kernel.h
> > @@ -140,6 +140,15 @@ extern int _cond_resched(void);
> >  		(__x < 0) ? -__x : __x;		\
> >  	})
> >  
> > +#ifdef CONFIG_LOCKDEP
> > +void might_fault(void);
> > +#else
> > +static inline void might_fault(void)
> > +{
> > +	might_sleep();
> > +}
> > +#endif
> > +
> >  extern struct atomic_notifier_head panic_notifier_list;
> >  extern long (*panic_blink)(long time);
> >  NORET_TYPE void panic(const char * fmt, ...)
> 
> This forgets that in_atomic() again - possibly triggering might_sleep()
> where not appropriate.
 
OK, drat... maybe we'll just noop that path and lose a couple of
might_sleep() points -- testing and developing should be done mostly
under lockdep anyway.


> I'm not sure its worth it to out-of-line the thing though (its only big
> on debug builds), and CONFIG_LOCKDEP is the wrong CONFIG_* variable, I
> think CONFIG_PROVE_LOCKING would be the appropriate one.

It is so that we don't need to pull sched.h and lockdep.h into those low
level user access functions.
--
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/