On Friday 18 January 2013 08:05 PM, Arnd Bergmann wrote:
> On Friday 18 January 2013, Vineet Gupta wrote:
>> +void cpu_idle(void)
>> +{
>> + /* Since we SLEEP in idle loop, TIF_POLLING_NRFLAG can't be set */
>> +
>> + /* endless idle loop with no priority at all */
>> + while (1) {
>> + tick_nohz_idle_enter();
>> + rcu_idle_enter();
>> +
>> + while (!need_resched())
>> + arch_idle();
>> +
>> + rcu_idle_exit();
>> + tick_nohz_idle_exit();
>> +
>> + schedule_preempt_disabled();
>> + }
>> +}
> Unless I'm mistaken, you have introduced the classic sleep race
> here, where an interrupt can happen between the check for
> need_resched() and the sleep instruction in arch_idle().
Hmm... there is indeed a race - so we could end up missing a reschedule
opportunity, for until next interrupt (so next tick, if NOHZ is not enabled, and
indefinitely if system has no other interrupting sources).
> To avoid that, you need to disable interrupts around
> the inner loop. The sleep instruction should return with
> interrupts implicitly enabled if ARC behaves like most
> other architectures doing this.
>
> Arnd
Indeed sleep has option to turn on interrupts before it commits - so this is
certainly doable.
static inline void arch_idle(void)
{
- __asm__("sleep");
+ /* sleep, but enable all interrupts before committing */
+ __asm__("sleep 0x3");
}
- while (!need_resched())
+doze:
+ local_irq_disable();
+ if (!need_resched()) {
arch_idle();
+ goto doze;
+ } else {
+ local_irq_enable();
+ }
Thx,
-Vineet
--
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/
