Yes, I think some problems when migrating from KVM->TCG are expected. I
keep getting page faults in TCG when it tries to handle local APIC timer
interrupts, which it probably doesn't understand. But that will be a
different question to be asked here.
Thanks for all the help, Peter.
Arnab
On Fri, Oct 2, 2020 at 11:39 PM Peter Maydell <peter.mayd...@linaro.org>
wrote:
> On Fri, 2 Oct 2020 at 17:50, Arnabjyoti Kalita
> <akal...@cs.stonybrook.edu> wrote:
> > What I can understand is that when we restore the VM State, the same
> interrupt state will be restored which was before we did a savevm. And then
> these interrupts will be executed appropriately.
> >
> > If there is an interrupt at the start of restore, then the interrupt
> will be executed first before the process of executing and finding the TB
> starts, based on the below logic. Am I correct ?
> >
> > while (!cpu_has_exception()) {
> > while (!cpu_has_interrupt()) {
> > tb = tb_find()
> > ....
> > }
> > }
>
> Yes, the interrupt will probably be the first thing to happen.
> (If the VM save/restore had not happened, it would have been
> the next thing to happen in the original execution.)
>
> > Is it possible that the same interrupt could have different numbers
> > in the KVM mode and in TCG mode? Like say, an interrupt is 0x15 in
> > KVM, but is 0x75 in TCG.
>
> I don't think that should happen. However, I'm not sure whether
> the x86 guest code supports migration from KVM to TCG (I know that
> it doesn't work for Arm guest code). KVM->KVM and TCG->TCG should
> definitely both work.
>
> thanks
> -- PMM
>
