Re: [PATCH v2 4/5] target/arm: Always add pmu property

[Akihiko Odaki]

On 2024/07/16 20:32, Peter Maydell wrote:
> On Tue, 16 Jul 2024 at 09:28, Akihiko Odaki <akihiko.odaki@daynix.com> wrote:
> > kvm-steal-time and sve properties are added for KVM even if the
> > corresponding features are not available. Always add pmu property too.
> >
> > Signed-off-by: Akihiko Odaki <akihiko.odaki@daynix.com>
> > ---
> >   target/arm/cpu.c | 3 ++-
> >   1 file changed, 2 insertions(+), 1 deletion(-)
> >
> > diff --git a/target/arm/cpu.c b/target/arm/cpu.c
> > index 9e1d15701468..32508644aee7 100644
> > --- a/target/arm/cpu.c
> > +++ b/target/arm/cpu.c
> > @@ -1781,9 +1781,10 @@ void arm_cpu_post_init(Object *obj)
> >
> >       if (arm_feature(&cpu->env, ARM_FEATURE_PMU)) {
> >           cpu->has_pmu = true;
> > -        object_property_add_bool(obj, "pmu", arm_get_pmu, arm_set_pmu);
> >       }
> >
> > +    object_property_add_bool(obj, "pmu", arm_get_pmu, arm_set_pmu);
> > +
> >       /*
> >        * Allow user to turn off VFP and Neon support, but only for TCG --
> >        * KVM does not currently allow us to lie to the guest about its
>
> Before we do this we need to do something to forbid setting
> the pmu property to true on CPUs which don't have it. That is:
>
>   * for CPUs which do have a PMU, we should default to present, and
>     allow the user to turn it on and off with pmu=on/off
>   * for CPUs which do not have a PMU, we should not let the user
>     turn it on and off (either by not providing the property, or
>     else by making the property-set method raise an error, or by
>     having realize detect the discrepancy and raise an error)

I don't think there is any reason to prohibit adding a PMU to a CPU that 
doesn't have when you allow to remove one. For example, neoverse-v1 
should always have PMU in the real world.

Perhaps it may make sense to prohibit adding a PMU when the CPU is not 
Armv8 as the PMU we emulate is apparently PMUv3, which is part of Armv8.

Regards,
Akihiko Odaki