Re: [PATCH v2 1/2] target/arm: kvm: Inject events at the last stage of sync

[Andrew Jones]

On Thu, Feb 06, 2020 at 09:41:10PM +0000, Beata Michalska wrote:
> On Tue, 4 Feb 2020 at 10:34, Andrew Jones <address@hidden> wrote:
> >
> > On Wed, Jan 29, 2020 at 08:24:40PM +0000, Beata Michalska wrote:
> > > KVM_SET_VCPU_EVENTS might actually lead to vcpu registers being modified.
> > > As such this should be the last step of sync to avoid potential 
> > > overwriting
> > > of whatever changes KVM might have done.
> > >
> > > Signed-off-by: Beata Michalska <address@hidden>
> > > ---
> > >  target/arm/kvm32.c | 20 ++++++++++----------
> > >  target/arm/kvm64.c | 20 ++++++++++----------
> > >  2 files changed, 20 insertions(+), 20 deletions(-)
> > >
> > > diff --git a/target/arm/kvm32.c b/target/arm/kvm32.c
> > > index 32bf8d6..cf2b47f 100644
> > > --- a/target/arm/kvm32.c
> > > +++ b/target/arm/kvm32.c
> > > @@ -386,17 +386,17 @@ int kvm_arch_put_registers(CPUState *cs, int level)
> > >          return ret;
> > >      }
> > >
> > > -    ret = kvm_put_vcpu_events(cpu);
> > > -    if (ret) {
> > > -        return ret;
> > > -    }
> > > -
> > >      write_cpustate_to_list(cpu, true);
> > >
> > >      if (!write_list_to_kvmstate(cpu, level)) {
> > >          return EINVAL;
> > >      }
> > >
> > > +    ret = kvm_put_vcpu_events(cpu);
> > > +    if (ret) {
> > > +        return ret;
> > > +    }
> > > +
> >
> > I think we should put a comment above this that says basically the same
> > thing as the commit message in order to explain why kvm_put_vcpu_events()
> > *must* be after write_list_to_kvmstate().
> >
> Will do that.
> 
> > >      kvm_arm_sync_mpstate_to_kvm(cpu);
> > >
> > >      return ret;
> > > @@ -462,11 +462,6 @@ int kvm_arch_get_registers(CPUState *cs)
> > >      }
> > >      vfp_set_fpscr(env, fpscr);
> > >
> > > -    ret = kvm_get_vcpu_events(cpu);
> > > -    if (ret) {
> > > -        return ret;
> > > -    }
> > > -
> > >      if (!write_kvmstate_to_list(cpu)) {
> > >          return EINVAL;
> > >      }
> > > @@ -475,6 +470,11 @@ int kvm_arch_get_registers(CPUState *cs)
> > >       */
> > >      write_list_to_cpustate(cpu);
> > >
> > > +    ret = kvm_get_vcpu_events(cpu);
> > > +    if (ret) {
> > > +        return ret;
> > > +    }
> > > +
> >
> > Why are we moving kvm_get_vcpu_events()?
> 
> This is only to make things consistent with put_registeres.
> There is no functional change per se.

Without a functional change I wouldn't move it. It's much
more appealing to have the final state writes at the bottom
of this function.

Thanks,
drew