Dear Russell, On Sat, 4 Jul 2015 17:35:33 +0800 Jisheng Zhang <jszh...@marvell.com> wrote:
> Dear Russell, > > On Sat, 4 Jul 2015 10:25:46 +0100 > Russell King - ARM Linux <li...@arm.linux.org.uk> wrote: > > > On Sat, Jul 04, 2015 at 04:50:07PM +0800, Jisheng Zhang wrote: > > > Dear Russell, > > > > > > On Sat, 4 Jul 2015 09:26:23 +0100 > > > Russell King - ARM Linux <li...@arm.linux.org.uk> wrote: > > > > > > > On Sat, Jul 04, 2015 at 01:19:30PM +0800, Jisheng Zhang wrote: > > > > > On Marvell Berlin SoCs, the cpu's local timer is shutdown when the cpu > > > > > goes to a deep idle state, then the timer framework will be notified > > > > > to > > > > > use a broadcast timer instead. The broadcast timer uses dw-apb-ictl as > > > > > interrupt chip, this patch adds irq_set_affinity support so that the > > > > > going to deep idle state cpu can set the interrupt affinity of the > > > > > broadcast interrupt to avoid unnecessary wakeups and IPIs. > > > > > > > > NAK to this patch. > > > > > > > > The real question is - if CPU0 is the CPU going offline, why is it > > > > still receiving _any_ interrupts - all interrupts should be migrated > > > > off it, including the chained interrupts. > > > > > > I think it's due to broadcast timer interrupt. Let me describe the > > > situation: > > > > > > 1. cpu1 is going offline > > > 2. cpuidle notify timer framework to use a broadcast timer instead due to > > > localtimer > > > is CLOCK_EVT_FEAT_C3STOP > > > 3. when timer is expired, CPU0 will be waken up by the timer interrupt if > > > it has > > > gone offline > > > 4. CPU0 sends broadcast timer IPI to CPU1 > > > > If CPU1 is going offline, then CPU1 should have no interrupts delivered to > > "sleep $i" on CPU1 will make the timer framework to wake up it at > current_time+$i > in the future then go offline, currently this is achieved by programming one > broadcast timer. In our case, the broadcast timer interrupt will be routed to > CPU0 by default, so CPU0 has to send broadcast timer IPI to CPU1. > > > it. However, this is not the situation you're testing - in your results > > below, and your "simple test" you never take CPU1 offline. > > > > when cpu1 executes "sleep $i", cpu1 will go to deepest cpuidle level, in our > case it will go offline. > I may misread your emails. I guess the "offline" means "cpu hot unplug"? Sorry for misunderstanding. This patch doesn't try to improve anything related with "hog unplug", it tries to improve the following situation instead: 1. cpu1 is entering deepest cpuidle level, shutdown in our case. 2. cpuidle notify timer framework to use a broadcast timer instead due to localtimer is CLOCK_EVT_FEAT_C3STOP 3. when timer is expired, CPU0 will be waken up by the timer interrupt if it has been shutdown 4. CPU0 sends broadcast timer IPI to CPU1 Thanks, Jisheng -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
