Skip to contentSkip to conversation

hrtimers: calculate expires_next after all timers are executed

PATCH [email protected] 3 participants 3 messages Jan 21, 2015 – Jan 22, 2015 High-resolution timers, timer wheel, clockevents

lore
Share & tools
lore
#0 lore raw

Hi Thomas,

In this patch, we are only setting the flag cpu_base->in_hrtirq = 1; in hrtimer_interrupt() and not resetting it.

Review void hrtimer_interrupt(struct clock_even · line 1254

Quoted 5 lines
(quoted diff) +1 -0
@@ -1251,9 +1251,9 @@ void hrtimer_interrupt(struct clock_even
 dev->next_event.tv64 = KTIME_MAX;
 
 raw_spin_lock(&cpu_base->lock);
+ cpu_base->in_hrtirq = 1;

Since this is the case, after the hrtimer_interrupt routine completes and new timer is started and say it happens to be leftmost,
we don't allow hrtimer_reprogram() since we bail out given that cpu_base->in_hrtirq=1 :

timer irq comes
--->hrtimer_interrupt()
---->cpu_base->in_hrtirq flag is set
------>Interrupt is serviced

After sometime

-->__hrtimer_start_range_ns()
--->happens to be leftmost hence call hrtimer_enqueue_reprogram() ----->hrtimer_reprogram()
------->bail out since cpu_base->in_hrtirq flag is stil set

At the end of hrtimer interrupt routine, i think we should also clear this flag.

Something like the following can be added ?

kernel/time/hrtimer.c +2 -1
index 8c94a55..b663653 100644
@@ -1251,9 +1251,9 @@ void hrtimer_interrupt(struct clock_event_device *dev)
        dev->next_event.tv64 = KTIME_MAX;
 
        raw_spin_lock(&cpu_base->lock);
-       cpu_base->in_hrtirq = 1;
        entry_time = now = hrtimer_update_base(cpu_base);
 retry:
+       cpu_base->in_hrtirq = 1;
        /*
         * We set expires_next to KTIME_MAX here with cpu_base->lock
         * held to prevent that a timer is enqueued in our queue via
@@ -1304,6 +1304,7 @@ retry:
         * against it.
         */
        cpu_base->expires_next = expires_next;
+       cpu_base->in_hrtirq = 0;
        raw_spin_unlock(&cpu_base->lock);
 
        /* Reprogramming necessary ? */

Thanks and regards,
vignesh Radhakrishnan

Signature · 2 lines
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project
#1 lore raw
Quoted 24 lines · On Wed, 21 Jan 2015, [email protected] wrote:

Hi Thomas,

In this patch, we are only setting the flag cpu_base->in_hrtirq = 1; in hrtimer_interrupt() and not resetting it.

kernel/time/hrtimer.c void hrtimer_interrupt(struct clock_even · line 1254

Quoted 5 lines
kernel/time/hrtimer.c +1 -0
@@ -1251,9 +1251,9 @@ void hrtimer_interrupt(struct clock_even
 dev->next_event.tv64 = KTIME_MAX;
 
 raw_spin_lock(&cpu_base->lock);
+ cpu_base->in_hrtirq = 1;

Since this is the case, after the hrtimer_interrupt routine completes and new timer is started and say it happens to be leftmost,
we don't allow hrtimer_reprogram() since we bail out given that cpu_base->in_hrtirq=1 :

timer irq comes
--->hrtimer_interrupt()
---->cpu_base->in_hrtirq flag is set
------>Interrupt is serviced

After sometime

Quoted 7 lines

-->__hrtimer_start_range_ns()
--->happens to be leftmost hence call hrtimer_enqueue_reprogram() ----->hrtimer_reprogram()
------->bail out since cpu_base->in_hrtirq flag is stil set

At the end of hrtimer interrupt routine, i think we should also clear this flag.

Bah, yes. I moved it around several times and then dropped it.

Thanks for spotting it!

tglx

#2 lore raw

Hi Thomas,

Slightly tested your version (with Vignesh's fixes), no issues so far, thanks!