| Message ID | 52EA8BD4.6020803@linaro.org |
|---|---|
| State | New |
| Headers | show |
On Thu, 30 Jan 2014, Daniel Lezcano wrote: > On 01/30/2014 05:07 PM, Nicolas Pitre wrote: > > On Thu, 30 Jan 2014, Daniel Lezcano wrote: > > > But what I don't get with your comment is the local_irq_enable is done > > > from > > > the cpuidle common framework in 'cpuidle_enter_state' it is not done from > > > the > > > arch specific backend cpuidle driver. > > > > Oh well... This certainly means we'll have to clean this mess as some > > drivers do it on their own while some others don't. Some drivers also > > loop on !need_resched() while some others simply return on the first > > interrupt. > > Ok, I think the mess is coming from 'default_idle' which does not re-enable > the local_irq but used from different places like amd_e400_idle and > apm_cpu_idle. Yet if you look at the code path before my patches you'll see that IRQs were enabled only after cpuidle_idle_call() had returned success. > void default_idle(void) > { > trace_cpu_idle_rcuidle(1, smp_processor_id()); > safe_halt(); > trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id()); > } > > Considering the system configured without cpuidle because this one *always* > enable the local irq, Yet this is discutable. Given that some hardware do have IRQs turned on upon exiting idle mode, I think we should generalize it so that the explicit enabling of IRQs, when needed, should be done as close as possible to the operation that caused idle mode to be entered. > we have the different cases: > > x86_idle = default_idle(); > ==> local_irq_enable is missing According to Peter it is not. > x86_idle = amd_e400_idle(); > ==> it calls local_irq_disable(); but in the idle loop context where the > local irqs are already disabled. Since it returned from default_idle() then IRQs are enabled. > ==> if amd_e400_c1e_detected is true, the local_irq are enabled > ==> otherwise no > ==> default_idle is called from there and does not enable local_irqs Again, it does. > > > So the code above could be: > > > > > > if (cpuidle_idle_call()) > > > x86_idle(); > > > > > > without the else section, this local_irq_enable is pointless. Or may be I > > > missed something ? > > > > A later patch removes it anyway. But if it is really necessary to > > enable interrupts then the core will do it but with a warning now. > > This WARN should disappear. It was there because it was up to the backend > cpuidle driver to enable the irq. But in the meantime, that was consolidated > into a single place in the cpuidle framework so no need to try to catch > errors. And that consolidation was a mistake IMHO. We should assume that the exiting of idle mode has IRQs enabled already, and do so manually in the backend driver if it is not the case on particular hardware. That's the only way to ensure uniformity at a higher level. Yet, if a code path is buggy in that regard, whether this is through cpuidle when enabled, or the default idle function otherwise, then the warning is there in cpu_idle_loop() to catch them all. > What about (based on this patchset). > > diff --git a/arch/x86/kernel/process.c b/arch/x86/kernel/process.c > index 4505e2a..2d60cbb 100644 > --- a/arch/x86/kernel/process.c > +++ b/arch/x86/kernel/process.c > @@ -299,6 +299,7 @@ void arch_cpu_idle_dead(void) > void arch_cpu_idle(void) > { > x86_idle(); > + local_irq_enable(); > } Again this is redundant. Nicolas -- To unsubscribe from this list: send the line "unsubscribe linux-pm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/arch/x86/kernel/process.c b/arch/x86/kernel/process.c index 4505e2a..2d60cbb 100644 --- a/arch/x86/kernel/process.c +++ b/arch/x86/kernel/process.c @@ -299,6 +299,7 @@ void arch_cpu_idle_dead(void) void arch_cpu_idle(void) { x86_idle(); + local_irq_enable(); } /*