On 29 July 2016 at 07:54, Andrew Jones <drjo...@redhat.com> wrote:
> OK, so this property will be exposed to all ARM cpu types, and if a user
> turns it on, then it will stay on for all types, except when using KVM
> with an aarch64 cpu type, and KVM doesn't support it. This could mislead
> users to believe they'll get a pmu, by simply adding pmu=on, even when
> they can't. I think we'd ideally keep has_pmu, and the current code that
> sets it, and then add code like
>
> if (enable_pmu && !has_pmu) {
> error_report("Warning: ...")
> }
>
> somewhere.
I think we should probably follow the existing model used by
has_el3, where the property only exists if it's valid to set it.
thanks
-- PMM
