Re: clocksource watchdog false detection

[Thomas Gleixner]

On Tue, 17 Jan 2017, Yinghai Lu wrote:
> one socket system with running cpu/memory stress test, got:
> 
> --
> [45477.565642] timekeeping watchdog: Marking clocksource 'tsc' as
> unstable, because the skew is too large:
> [45477.565643]     'acpi_pm' wd_now: 32e642 wd_last: 71c8c5 mask: ffffff
> [45477.565644]     'tsc' cs_now: 4b80842681e0 cs_last: 4a48ac7d5a0a
> mask: ffffffffffffffff
> [45477.623779] Switched to clocksource acpi_pm
> 
> so acpi_pm reading between two timer are wrapped.
> 
> When that happen, should just disable clocksource_watchdog instead?

When that happens then something prevented the expiry of the watchdog timer
for more than 4.5 seconds.

So instead of blaming the messenger, you should fix that problem.

Thanks,

	tglx