On Wed, 2011-10-05 at 00:05 +0200, Thomas Gleixner wrote:

quoted

peter_zijlstra-frob-rcu.patch

      Looks OK.  Hmmm...  Should this one go to mainline?
      Oh, looks equivalent, actually.  So why the change?

Peter ? 

-       if (in_irq() || in_serving_softirq()) {
+       if (preempt_count() & (HARDIRQ_MASK | SOFTIRQ_OFFSET)) {


For !rt its equivalent yes, for rt otoh its not:

int in_serving_softirq(void)
{
        int res;

        preempt_disable();
        res = __get_cpu_var(local_softirq_runner) == current;
        preempt_enable();
        return res;
}

However invoke_softirq() will still add SOFTIRQ_OFFSET so we need to
look at that to avoid recursion issues.

The changelog describes this. So this change is a direct consequence of
-rt frobbing the softirq stuff and thus isn't needed upstream.