> On 4. Aug 2025, at 22:40, Pierrick Bouvier <pierrick.bouv...@linaro.org>
> wrote:
>
> On 8/4/25 12:56 PM, Mohamed Mediouni wrote:
>>> On 4. Aug 2025, at 21:50, Pierrick Bouvier <pierrick.bouv...@linaro.org>
>>> wrote:
>>>
>>> Is it equivalent to simply using its=off, or is there a difference?
>>> The info_report seems to imply it's not the same.
>> Not equivalent.
>> Regular system: GICv3 + ITS
>> This configuration (for the newest machine version): GICv3 with no MSIs
>> And its=off explicitly: GICv3 + GICv2m
>> It became not equivalent since the intro of GICv3 + GICv2m in patch 2 of
>> this series.
>>
>> Thank you,
>
> I see. It could be worth adding this information to commit message.
> With that,
> Reviewed-by: Pierrick Bouvier <pierrick.bouv...@linaro.org>
>
> Slightly off topic for this commit, is there any downside to always have
> GICv3 + GICv2m setup enabled? Do some systems don't support GICv2m?
>
GICv2m is modelled as a device external from the GIC, and so it can be emulated
everywhere.
It can be done but then it supposes that the same Qemu command line will expose
a different device model (instead of just MSI support missing) when ran on a
different system. Not sure that’s the right thing to do…
That’s because the current setup already has ITS on by default when a GICv3 is
chosen.
Note that GICv3 + GICv2m* hardware exists in the wild (Graviton1 notably) but
VMs have a GICv3 w/ ITS configuration there.
* not quite, but gets shoehorned in with Linux quirks
Thank you,
-m
