Re: -rt IA64 update

[Simon Derr]

On Tue, 30 May 2006, john stultz wrote:

> On Tue, 2006-05-30 at 08:15 +0200, Ingo Molnar wrote:
> > * Simon Derr <Simon.Derr@xxxxxxxx> wrote:
> > > * This kernel, when booting, prints:
> > > 
> > > 	BUG in check_monotonic_clock at kernel/time/timeofday.c:164
> > > 
> > > But I think this happens because two get_monotonic_clock() are racing 
> > > on two cpus. There is a lock to prevent the race, but it is a seqlock. 
> > > That means that it is okay if the race happens since another try will 
> > > be attempted, but the message that has been printed on the console 
> > > can't be removed, and the user is unnecessarily scared.
> 
> Simon, I suspect here you actually have unsynced ITCs, as the
> check_monotonic_clock values are all locked w/ spinlocks.
> 
> You should probably add a cmpxchg in clocksource_itc_read() where you
> currently do the if (last_itc < now), or you'll race on setting
> last_itc. Let me know if you still see the issue w/ that fix.

I have already tried to add a raw_spinlock_t to avoid this race, and the 
message still appears.

What I believe is happening is that is that somewhere between 
__get_nsec_offset() and ktime_add_ns(), while in

static ktime_t __get_monotonic_clock(void)
{
        s64 offset = __get_nsec_offset();
#ifdef CONFIG_PARANOID_GENERIC_TIME
        ktime_t check = get_check_value();
#endif
        ktime_t ret;

        ret = ktime_add_ns(system_time, offset);
        check_monotonic_clock(check,ret);
        return ret;
}

the cycle_last and system_time are being updated by a concurrent 
timeofday_periodic_hook().

This can happen since system_time_lock is "only" a seqlock.

>From a correctness point of view this is okay as get_monotonic_clock() 
will detect the race and call __get_monotonic_clock() again. But the 
warning message has been already been printed.

	Simon.

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/