Re: [RFC PATCH v3 20/28] target/arm: Set cortex-a57 as default cpu for KVM-only build

[Fabiano Rosas]

Peter Maydell <peter.maydell@linaro.org> writes:

> On Mon, 16 Jan 2023 at 13:45, Fabiano Rosas <farosas@suse.de> wrote:
>>
>> Richard Henderson <richard.henderson@linaro.org> writes:
>>
>> > On 1/13/23 06:04, Fabiano Rosas wrote:
>> >> The cortex-a15 is not present anymore when CONFIG_TCG=n, so use the
>> >> cortex-a57 as default cpu for KVM.
>> >>
>> >> Signed-off-by: Fabiano Rosas <farosas@suse.de>
>> >
>> > Ideally there would not be a default at all, requiring the command-line 
>> > option to be used.
>>
>> We could probably do that now without impact to users, since KVM would
>> always require a -cpu option due to the current default being
>> cortex-a15.
>>
>> >
>> > Second choice would be "host", since that's the only value that's actually 
>> > usable (except
>> > for the off-chance that you're actually running on an a57, which is less 
>> > and less likely
>> > as time moves on).
>> >
>>
>> I'll have to go around fixing qtest first, either to add -cpu or to add
>> -accel kvm, otherwise we get:
>>
>> The 'host' CPU type can only be used with KVM or HVF
>
> For a CPU type that will work with either KVM or TCG, that would
> be "max".

Yes, although the issue here is more that there are tests running with
!kvm_enabled (no -accel kvm given) and !tcg_enabled (--disable-tcg).

The "max" cpu does in fact work with qtest because even when
CONFIG_TCG=n, it ends up configuring a "cortex-a57 + extra things" in
aarch64_max_initfn. But that seems a bit too implicit to me, it would be
better for the tests to explicitly set the accel and cpu options.