Re: ftrace: preemptoff selftest not working

[Heiko Carstens <heiko.carstens@de.ibm.com>]

On Wed, Nov 19, 2008 at 10:02:23AM +0100, Ingo Molnar wrote:
> 
> >  #ifdef CONFIG_FTRACE_STARTUP_TEST
> > +	/*
> > +	 * When this gets called we hold the BKL which means that preemption
> > +	 * is disabled. Various trace selftests however need to disable
> > +	 * and enable preemption for successful tests. So we drop the BKL here
> > +	 * and grab it after the tests again.
> > +	 */
> > +	unlock_kernel();
> >  	if (type->selftest) {
> >  		struct tracer *saved_tracer = current_trace;
> >  		struct trace_array *tr = &global_trace;
> > @@ -562,6 +569,7 @@ int register_tracer(struct tracer *type)
> >  		}
> >  		printk(KERN_CONT "PASSED\n");
> >  	}
> > +	lock_kernel();
> 
> dropping the BKL was a good idea, but the code flow was not 
> investigated thoroughly enough, which caused this crash to trigger in 
> -tip testing:

Yes, I came to the same conlcusion this morning after reading the patch
again and wanted to send a follow-up patch. But you were faster ;)

Anyway, what bothers me more is the question if the idea to drop the BKL
in register_tracer is good. It's probably just a question of time until
the first tracers come in modules. And then the unlock_kernel()/lock_kernel()
sequence would be broken.
So it probably might make more sense to drop and grab the BKL in the
init functions that register a tracer?

But.. maybe the BKL is gone until this is an issue ;)