On Tue, Jul 19, 2011 at 10:57:37AM +1000, Benjamin Herrenschmidt wrote: > On Mon, 2011-07-18 at 17:39 -0700, Paul E. McKenney wrote: > > > Hrm, no I don't see that happening no. The preempt count when > > exiting an > > > irq or softirq stack should be the exact same as when entering it, > > which > > > is why we don't bother copying it over. Do you see any case where > > that > > > wouldn't hold ? > > > > Nope, other than seeing preempt_count() transition from zero to three > > across a spin_unlock_irqrestore() for no good reason that I could see. > > Do you have a nice repro-case ? :-) > > That sounds really nasty ... smells really like something bad's > happening from an interrupt, but we don't copy back the preempt-count > from the interrupt stacks at all, so that's really really odd. Good question... The system I reproduced on four times over the weekend is out of commission. Trying the same test on another system with a minimal patch -- will let you know how it goes. Thanx, Paul -- To unsubscribe from this list: send the line "unsubscribe linux-arch" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
