On 02/20/2012 07:17 AM, Alexander Graf wrote: > > On 18.02.2012, at 00:00, Scott Wood wrote: > >> It would be simpler (both here and in the idle hcall) if we could just >> drop support for CONFIG_PREEMPT=n. :-P > > When running with CONFIG_PREEMPT=n we don't have to worry about interrupts being enabled though, That's exactly when we *do* need to worry. Interrupts can cause the "need resched" condition to become true. > since we only preempt on known good checkpoints, right? Yes, and "entering the guest" is supposed to be one of those known good checkpoints, similar to returning to userspace. If we miss a reschedule here, we're may not have another chance until the next timer interrupt. The code for returning to userspace is similarly structured. -Scott -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
