On Sat, 2007-02-24 at 21:13 -0800, David Miller wrote:
> As I suspected, the one-shot code wasn't very well tested and I'd be
> the one to debug this thing on sparc64 :-)
>
> When a timer exceeds the timer period, the one-shot handling code does
> the following loop:
>
> for (;;) {
> ktime_t next = ktime_add(dev->next_event, tick_period);
>
> if (!clockevents_program_event(dev, next, ktime_get()))
> return;
> tick_periodic(cpu);
> }
>
> So it just keeps running tick_periodic() until we "catch up".
>
> Problem is, if clockevents_program_event() gets a "next" time in the
> past, the very case where we'll loop, it DOES NOT update
> dev->next_event. It returns the error before doing so.
Yep, that's caused by a late change for the *!&#ed broadcast stuff for
x86.
> As a result of this, we'll loop forever here, the softlockup watchdog
> will trigger, and the system will wedge completely.
>
> I was getting a softlockup and immediate system hang, so to debug this
> I kept a history of the last 8 TSC values when tick_periodic() was
> invoked. At softlockup trigger, I'd dump the log. And what I saw
> were TSC deltas that we so tiny as to be just enough to indicate
> tick_periodic() was running in a tight loop :-)
>
> I propose the following fix, which I'm about to test.
Yep
> Signed-off-by: David S. Miller <[EMAIL PROTECTED]>
Acked-by: Thomas Gleixner <[EMAIL PROTECTED]>
> diff --git a/kernel/time/tick-common.c b/kernel/time/tick-common.c
> index 4500e34..0986a2b 100644
> --- a/kernel/time/tick-common.c
> +++ b/kernel/time/tick-common.c
> @@ -77,6 +77,7 @@ static void tick_periodic(int cpu)
> void tick_handle_periodic(struct clock_event_device *dev)
> {
> int cpu = smp_processor_id();
> + ktime_t next;
>
> tick_periodic(cpu);
>
> @@ -86,12 +87,12 @@ void tick_handle_periodic(struct clock_event_device *dev)
> * Setup the next period for devices, which do not have
> * periodic mode:
> */
> + next = ktime_add(dev->next_event, tick_period);
> for (;;) {
> - ktime_t next = ktime_add(dev->next_event, tick_period);
> -
> if (!clockevents_program_event(dev, next, ktime_get()))
> return;
> tick_periodic(cpu);
> + next = ktime_add(next, tick_period);
> }
> }
>
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [EMAIL PROTECTED]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
