2.6.14-rt4: __get_nsec_offset() false positives

[john stultz]

Hey Ingo,
	I just booted 2.6.14-rt4 and I'm getting lots of the
__get_nsec_offset() warnings where there isn't really a problem.

The main issue is that the clocksource may not be a TSC (acpi_pm in my
case), so the check to see if the cycle value ever goes backwards will
falsely trigger when the 24bit wide ACPI PM counter wraps.

To properly check for algorithmic inconsistencies, the checks should
probably be similar to what you had earlier inside
__get_monotonic_clock(), since at __get_nsec_offset() you really don't
have enough information to sort out if an inconsistency has occurred.

If we want to watch for hardware inconsistencies (like unsynced TSCs),
those checks really need to go inside the clocksource drivers
themselves. 

I'll write up a paranoid debug patch that provides similar checks for
both cases and include it in my patch set so you don't have to keep
forward porting your own versions. 

thanks
-john






-
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/