Re: [PATCH v6 1/2] arm/kvm: add support for MTE

[Cornelia Huck <cohuck@redhat.com>]

On Wed, Mar 01 2023, Andrea Bolognani <abologna@redhat.com> wrote:

> On Wed, Mar 01, 2023 at 03:15:17PM +0100, Cornelia Huck wrote:
>> On Wed, Mar 01 2023, Andrea Bolognani <abologna@redhat.com> wrote:
>> > I'm actually a bit confused. The documentation for the mte property
>> > of the virt machine type says
>> >
>> >   mte
>> >     Set on/off to enable/disable emulating a guest CPU which implements
>> >     the Arm Memory Tagging Extensions. The default is off.
>> >
>> > So why is there a need to have a CPU property in addition to the
>> > existing machine type property?
>>
>> I think the state prior to my patches is actually a bit confusing: the
>> user needs to set a machine type property (causing tag memory to be
>> allocated), which in turn enables a cpu feature. Supporting the machine
>> type property for KVM does not make much sense IMHO: we don't allocate
>> tag memory for KVM (in fact, that would not work). We have to keep the
>> previous behaviour, and explicitly instructing QEMU to create cpus with
>> a certain feature via a cpu property makes the most sense to me.
>
> I agree that a CPU feature makes more sense.
>
>> We might want to tweak the documentation for the machine property to
>> indicate that it creates tag memory and only implicitly enables mte but
>> is a pre-req for it -- thoughts?
>
> I wonder if it would be possible to flip things around, so that the
> machine property is retained with its existing behavior for backwards
> compatibility, but both for KVM and for TCG the CPU property can be
> used on its own?
>
> Basically, keeping the default of machine.mte to off when cpu.mte is
> not specified, but switching it to on when it is. This way, you'd be
> able to simply use
>
>   -machine virt -cpu xxx,mte=on
>
> to enable MTE, regardless of whether you're using KVM or TCG, instead
> of requiring the above for KVM and
>
>   -machine virt,mte=on -cpu xxx
>
> for TCG.

The machine prop is a bool; that means that we cannot distinguish
between "user did not set mte at all" and "user explicitly set
mte=off" -- I think we want

  -machine virt,mte=off -cpu xxx,mte=on

to generate an error, but that would still imply that we'd need to error
out for

  -machine virt -cpu xxx,mte=on

as well.

We could make the machine prop OnOffAuto, but that looks like overkill
to me.

>
> Note that, from libvirt's point of view, there's no advantage to
> doing things that way instead of what you already have. Handling the
> additional machine property is a complete non-issue. But it would
> make things nicer for people running QEMU directly, I think.

I'm tempted to simply consider this to be another wart of the QEMU
command line :)