Re: Recording and replaying HPET timer values

[Pavel Dovgalyuk]

On 21.05.2021 05:15, Arnabjyoti Kalita wrote:
> Thank you for your reply, Pavel.
>
> Yes, I understand. Interrupts might be injected at different times and
> this would be a source of indeterminism. But, let's say that I know
> when to inject the interrupts.
>
> Is this enough to ensure that the replay of the HPET clocks would be
> deterministic? Do you see any other potential causes of concern?

Right, I don't see any other issues.

> Best Regards,
> Arnabjyoti Kalita
>
> On Mon, May 17, 2021 at 1:05 PM Pavel Dovgalyuk
> <pavel.dovgalyuk@ispras.ru> wrote:
> > On 14.05.2021 06:30, Arnabjyoti Kalita wrote:
> > > Hello all,
> > >
> > > I am recording and replaying HPET timer values for my guest.
> > >
> > > I am running QEMU version 5.0.1. The guest kernel is a
> > > 4.4.0-21-generic Ubuntu flavor. The host and target architecture is
> > > both x86-64.
> > >
> > > To record, I start QEMU in KVM mode as follows -
> > >
> > > sudo ./qemu-system-x86_64 -m 1024 --machine pc-i440fx-2.5 -cpu
> > > qemu64,-kvmclock -enable-kvm -clock_replay
> > > mode=record,file=clock_record9.txt -netdev
> > > tap,id=tap1,ifname=tap0,script=no,downscript=no -device
> > > virtio-net-pci,netdev=tap1,mac=00:00:00:00:00:00 -drive
> > > file=~/os_images_for_qemu/ubuntu-16.04-desktop-amd64.qcow2,format=qcow2,if=none,id=img-direct
> > > -device virtio-blk-pci,drive=img-direct
> > >
> > > After this, I force the guest to switch to the "hpet" clocksource.
> > >
> > > The clock_record9.txt file records the values of the HPET counter. To
> > > perform the recording, I have changed the hpet_ram_read function in
> > > the "hw/timer/hpet.c" file.
> > >
> > > static uint64_t hpet_ram_read() {
> > > .....
> > >       case HPET_COUNTER:
> > >           if (hpet_enabled(s)) {
> > >               cur_tick = hpet_get_ticks(s);
> > >           } else {
> > >               cur_tick = s->hpet_counter;
> > >           }
> > >           record_hpet_counter((int64_t)cur_tick);  <-----  record here
> > >           return cur_tick;
> > > }
> > >
> > > In replay mode, I start the guest in TCG mode as below-
> > >
> > > sudo ./qemu-system-x86_64 -m 1024 -clock_replay
> > > mode=replay,file=clock_record9.txt -machine pc-i440fx-2.5 -cpu qemu64
> > > -netdev tap,id=tap1,ifname=tap0,script=no,downscript=no -device
> > > virtio-net-pci,netdev=tap1,mac=00:00:00:00:00:00 -drive
> > > file=~/os_images_for_qemu/ubuntu-16.04-desktop-amd64.qcow2,format=qcow2,if=none,id=img-direct
> > > -device virtio-blk-pci,drive=img-direct -loadvm snapshot1
> > >
> > > This is supposed to replay and mimic the HPET counter values. The
> > > snapshot is there to ensure that both record and replay start from the
> > > same guest state.
> > >
> > > Is the guest replay going to be deterministic, with respect to the
> > > HPET clock? Are there other factors about the HPET clock that need to
> > > be taken into account while replaying it ?
> >
> > I don't sure that this is enough.
> > There is qemu_timer in hpet.c, which is used for injecting the interrupts.
> > And interrupt handler code may read HPET clock. I mean that there may be
> > different frequency or timer reads in different runs.
> >
> > > I am not worried about other I/O events right now and I just want to
> > > ensure that HPET timer values are being replayed correctly.
> > >
> > > Best Regards,
> > > Arnabjyoti Kalita