On Mon, Dec 16, 2019 at 03:06:57PM +0000, Peter Maydell wrote: > Incidentally, if I understand things correctly, for TCG the > behaviour is (and has always been) that VM-stopped time is > not counted, because we run the emulated versions of these counters > off QEMU_CLOCK_VIRTUAL. So having the KVM default be the same as > the TCG default is nicely consistent.
It's correct that the vtimer has always been this way for TCG. However TCG and KVM still won't be the same with the adjvtime patches. The reason is that TCG also bases the ptimer off QEMU_CLOCK_VIRTUAL. This means on TCG ptimer == vtimer and both will lag a clocksource outside the VM. I'm not sure if TCG people want to change that to match KVM. If so, then I guess the ptimer should be based off QEMU_CLOCK_REALTIME. Thanks, drew
