Hi Radim,
On 2017/6/13 3:52, Radim Krčmář wrote:
2017-06-12 21:23+0800, Jay Zhou:
Guest using kvmclock will be hanged when migrating from unstable
tsc host to stable tsc host occasionally.
Sometimes, the tsc timestamp saved at the source side will be
backward when the guest stopped, and this value is transferred
to the destination side. The guest at the destination side thought
kvmclock is stable, so the protection mechanism against time
going backwards is not used.
When the first time vcpu0 enters the guest at the destination
side to update the wall clock, the result of
pvclock_clocksource_read will be backward occasionally,
which results in the wall clock drift.
Signed-off-by: Jay Zhou <jianjay.zhou@xxxxxxxxxx>
---
Hm, are you using KVM that has e3fd9a93a12a (4.9+)?
I'm using KVM (4.4.11) and QEMU (2.6.0). I thought the newest KVM
and QEMU have the same problem, maybe I'm wrong.
If you get a timestamp from KVM_GET_CLOCK() and pass that to
KVM_SET_CLOCK(), then kvmclock should not jump backwards anymore
(it could before 4.9, but only if the host had stable tsc).
If migrating from old KVMs to newer KVMs (v4.9+), the clock member
of struct KVMClockState is migrated to the destination side, but the
clock_is_reliable isn't migrated. The value of clock_is_reliable is false
at the source side, and it is true at the destination side. Is it possible
the clock member of struct KVMClockState migrated is smaller than the
last_value saved in the guest theoretically? Could you explain more
why it should not jump backwards in the scenario of migrating from unstable tsc
host to stable tsc host?
A possible source of this bug in when QEMU recomputes the timestamp to
pass to KVM_SET_CLOCK() using TSC and kvmclock page.
Yes, this bug happened in this scenario in my tests.
Can you provide the values used for KVM_GET_CLOCK and KVM_SET_CLOCK when
the bug occurs?
I didn't print these values out when the bug occured, and I do not have the
physical machine with unstable tsc clocksource now. But I will try to reproduce
with the newest KVM and QEMU once I have.
Regards,
Jay Zhou
