On Thu, Dec 03, 2020 at 11:55:17AM +0100, Alexander Graf wrote:
>
> On 03.12.20 02:52, Roman Bolshakov wrote:
> > On Wed, Dec 02, 2020 at 08:04:03PM +0100, Alexander Graf wrote:
> > > When clearing internal state of a CPU, we should also make sure that HVF
> > > knows about it and can push the new values down to vcpu state.
> > >
> > I'm sorry if I'm asking something dumb. But isn't
> > cpu_synchronize_all_post_reset() is supposed to push QEMU state into HVF
> > (or any other accel) after reset?
> >
> > For x86 it used to work:
> >
> > static void do_hvf_cpu_synchronize_post_reset(CPUState *cpu,
> > run_on_cpu_data arg)
> > {
> > hvf_put_registers(cpu);
> > cpu->vcpu_dirty = false;
> > }
>
>
> Yes, it works because after the reset is done, there is no other register
> modification happening. With the PSCI emulation code in QEMU, we still do
> modify CPU state after reset though.
>
Maybe I miss something but that doesn't seem correct. Why PSCI reset is
split from machine reset?
> Different question though: Why do we need the post_reset() call at all here
> to push state?
My understanding that post_reset is akin to a commit of the CPU state
after all reset actions have been done to QEMU CPU Arch env state. i.e.
arch/machine reset modifies env state and then the env is pushed to
accel. cpu->vcpu_dirty is cleared because env is in-sync with vcpu.
> We would just push it on the next run anyways, right?
That's correct (at least for x86 HVF).
> So if we don't set vcpu_dirty to false then, we wouldn't need this
> patch here I think.
>
That's right but the same post-reset approach is used for all accels,
including KVM. But I haven't found this for TCG.
Regards,
Roman
