Red Hat kernel says "Clocksource tsc unstable" but I'm using kvm-clock

I have a couple of virtualized RHEL 6 running on a RHEV host. Both host and VMs have >1year uptime.

Only in the last two days they started seldom throwing these kernel messages at random times

kernel: Clocksource tsc unstable (delta = -17179878652 ns).  Enable clocksource failover by adding clocksource_failover kernel parameter.

Fine: I'll follow the advice. The strange thing is, though, I'm not using tsc as a clocksource

[~]# cat /sys/devices/system/clocksource/clocksource0/current_clocksource 
kvm-clock

Studying the documentation of the clocksource framework, I came up with more questions than answers.

1. First of all, why is the kernel complaining about a clocksource that the system is not supposed to use? Is the kernel cycling anyway through all the available clocksources, perhaps? If yes, what for?

2. As far as I know, the message means that the current clocksource is drifted away from the watchdog clocksource more than WATCHDOG_THRESHOLD but what is the current watchdog clocksource? It cannot be kvm-clock, since it usually has CLOCK_SOURCE_MUST_VERIFY set. Is there a way to display it at runtime?

3. I understand that tsc isn't considered to be accurate with modern CPUs because of throttling/power saving features but /proc/cpuinfo tells that the processors seen by the VM have the constant_tsc flag. So I would assume tsc can also provide a quite reliable timing. Is that flag a "bypass" to the underlying real CPU or is it just an emulated one, running its own counter? In the first case, I could speculate that the host's scheduler switched the VM process to a different core having a drifted counter. Could this make sense?