Re: BUG: tracer_alloc_buffers returned with preemption imbalance

["Paul E. McKenney" <paulmck@linux.vnet.ibm.com>]

On Sun, Jun 24, 2012 at 04:37:28PM +0800, Cong Wang wrote:
> On Sun, Jun 24, 2012 at 1:48 AM, Fengguang Wu <fengguang.wu@intel.com> wrote:
> >> >  static inline int rcu_blocking_is_gp(void)
> >> >  {
> >> >         might_sleep();  /* Check for RCU read-side critical section. */
> >> > +       preempt_disable();
> >> >         return num_online_cpus() <= 1;
> >> > +       preempt_enable();
> >> >  }
> >>
> >> Thank you!  I have no idea how a preempt_disable() causes that badness
> >> to happen, but this commit is not yet critically important, so I will
> >> drop it.
> >
> > preempt_enable() becomes dead code because of the return statement?
> > I wonder why gcc didn't issue a warning (or I failed to catch it)...
> >
> 
> gcc has an option -Wunreachable-code, but we don't enable it
> when building kernel, nor it will be enabled with -Wall. If we enable it,
> we will have many false-positives as we have lots of debugging code
> which is not reachable unless we enable some debugging option.
> 
> However, when I test it manually with the following code:
> 
> ~% cat /tmp/unreachable.c
> int main(void)
> {
> 	int a = 0;
> 	a++;
> 	return ++a;
> 	a++;
> }
> ~% gcc -Wunreachable-code -O0 -c /tmp/unreachable.c
> 
> gcc still doesn't give me any warning for the last line of the code,
> gcc optimizes it out silently, I am wondering if this is a gcc bug.

But in my case, the trailing preempt_enable() should not have been
optimized away, right?  Wouldn't it be more like the following?

int a = 0;
int main(void)
{
	a++;
	return ++a;
	a++;
}

Hmmm...  But this -still- doesn't emit any warnings.

							Thanx, Paul