Comment # 2 on bug 1220119 from Jiri Wiesner

(In reply to Petr Vorel from comment #0)
> This is on x86_64 on my private openQA worker by ltp_net_tcp_cmds, triggered
> for 2 tests: arp and host (both tests did not fail):

The current clocksource, the one used for providing time - kvm-clock, remained
active after the error appeared. The error should have no effect on how
kvm-clock works and I do not see a reason for why the error should have an
effect on the workload. The clocksource watchdog gets executed every 500
milliseconds in softirq context (a timer handler). The kvm-clock clocksource
actually reported a value close to 500 milliseconds so kvm-clock is probably
fine.

> [  410.818828][    C0] clocksource: timekeeping watchdog on CPU0: Marking clocksource 'tsc' as unstable because the skew is too large:
> [  410.820365][    C0] clocksource:                       'kvm-clock' wd_nsec: 506724586 wd_now: 610448443a wd_last: 60e6144350 mask: ffffffffffffffff
> [  410.821824][    C0] clocksource:                       'tsc' cs_nsec: 49302930542 cs_now: 17c1364ba0a cs_last: 153fdc32526 mask: ffffffffffffffff
> [  410.818845][    C0] clocksource: timekeeping watchdog on CPU0: Marking clocksource 'tsc' as unstable because the skew is too large:
> [  410.820687][    C0] clocksource:                       'kvm-clock' wd_nsec: 506741379 wd_now: 61044885d3 wd_last: 60e6144350 mask: ffffffffffffffff
> [  410.821794][    C0] clocksource:                       'tsc' cs_nsec: 804834172 cs_now: 154a546a082 cs_last: 153fdc32526 mask: ffffffffffffffff

There is something rather conspicuous about the errors - there are two them.
The error is supposed to be printed only once per clocksource being marked
unstable. It is as if the clocksource watchdog ran in two threads (there is
supposed to be only one).

> Also, originally I noted this on ltp_net_tcp_cmds running on ppc64le on o3
> (kernel tests in https://openqa.opensuse.org/group_overview/32)

I'd be interested to see the kernel log from that test. While logged in, I can
see a failed ltp_net_tcp_cmds/host test but I don't know how to display the
kernel log.

> Others found that systemd-networkd test on Tumbleweed on x86_64
> (https://openqa.opensuse.org/group_overview/1):
> [  311.682306][    C0] clocksource: timekeeping watchdog on CPU0: Marking clocksource 'tsc' as unstable because the skew is too large:
> [  311.684482][    C0] clocksource:                       'kvm-clock' wd_nsec: 621509306 wd_now: 49d9caac63 wd_last: 49b4bf31a9 mask: ffffffffffffffff
> [  311.686513][    C0] clocksource:                       'tsc' cs_nsec: 952336665 cs_now: a3a422af12 cs_last: a327415f8e mask: ffffffffffffffff

The interval, as reported by kvm-clock, is much longer than 500 milliseconds
indicating a delayed expiration of the timer, which could be a consequence of
the system being under high load.

> Is it a problem of openQA setup? There are other bugs.

There is almost certainly something wrong with the tsc clocksource (not
kvm-clock). Perhaps, the TSC counter is somehow virtualized (or
para-virtualized by Qemu, I don't know) and the virtualized TSC misbehaves
under certain conditions.