On Tue, Aug 29, 2023 at 11:18:01AM -0700, Steve Sistare wrote:
> During RUN_STATE_SUSPENDED, the cpu clock remains enabled, so the
> timers_state saved to the migration stream is stale, causing time errors
> in the guest when it wakes from suspend.
Instead of having this, I'm wondering whether we should just let:
ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
stop the vm for suspended too - I think we reached a consensus that
SUSPENDED should be treated the same as running here (except the vcpu
beingg running or not).
So the more risky change is we should make runstate_is_running() cover
SUSPENDED, but of course that again can affect many other call sites.. and
I'm not sure whether it's 100% working everywhere.
I think I mentioned the other "easier" way, which is to modify
vm_stop_force_state() to take suspended:
int vm_stop_force_state(RunState state)
{
- if (runstate_is_running()) {
+ if (runstate_is_running() || runstate_is_suspended()) {
return vm_stop(state);
That resides in cpus.c but it really only affects migration, so much less
risky. Do you think this should be the better (and correct) way to go?
--
Peter Xu
