Re: [PATCH V3] arm/kvm: add support for MTE

[Cornelia Huck <cohuck@redhat.com>]

On Thu, Sep 26 2024, Gustavo Romero <gustavo.romero@linaro.org> wrote:

> Hi Cornelia and Ganapatrao,
>
> On 9/25/24 14:54, Cornelia Huck wrote:
>> On Fri, Sep 20 2024, Ganapatrao Kulkarni <gankulkarni@os.amperecomputing.com> wrote:
>>> +
>>> +        /*
>>> +         * Clear MTE bits, if not enabled in KVM mode.
>> Maybe add "This matches the MTE bits being masked by KVM in that case."?
>
> Clearing the MTE bits is also necessary when MTE is supported by the
> host (and so KVM can enable the MTE capability - so won't mask the MTE
> bits, but the user didn't want MTE enabled in the guest (mte=on no given
> or explicitly set to =off), so this comment is not always true?
>
> How about something like:
>
> "If MTE is supported by the host but could not be enabled on KVM mode or
> MTE should not be enabled on the guest (e.i. mte=off), clear guest's MTE 
> bits."

s/e.i./i.e./ :)

Otherwise fine with me.

>
> I do assume MTE is supported by the host (i.e. MTE bits >= 2 in the host)
> because otherwise condition "if (cpu_isar_feature(aa64_mte, cpu)) { ... 
> }" is not
> taken; and at this point cpu->isar->id_aa64pfr1 is set from the host's 
> bits via
> kvm_arm_set_cpu_features_from_host() and kvm_arm_get_host_cpu_features ().
>
>
>>> +         */
>>> +        if (kvm_enabled() && !cpu->kvm_mte) {
>>> +                FIELD_DP64(cpu->isar.id_aa64pfr1, ID_AA64PFR1, MTE, 0);
>>> +        }
>>>   #endif
>>>       }