Re: [PATCH v6 05/13] hw/arm, target/arm: nested virtualisation on HVF

["Philippe Mathieu-Daudé" <philmd@linaro.org>]

On 11/8/25 14:56, Philippe Mathieu-Daudé wrote:
> On 8/8/25 09:01, Mohamed Mediouni wrote:
>> Signed-off-by: Mohamed Mediouni <mohamed@unpredictable.fr>
>> ---
>>   hw/arm/virt.c         |  9 ++++++---
>>   target/arm/hvf-stub.c | 15 +++++++++++++++
>>   target/arm/hvf/hvf.c  | 41 +++++++++++++++++++++++++++++++++++++++--
>>   target/arm/hvf_arm.h  |  3 +++
>>   4 files changed, 63 insertions(+), 5 deletions(-)
>>
>> diff --git a/hw/arm/virt.c b/hw/arm/virt.c
>> index 59e6c2b9df..91d8cd9363 100644
>> --- a/hw/arm/virt.c
>> +++ b/hw/arm/virt.c
>> @@ -817,8 +817,7 @@ static void create_gic(VirtMachineState *vms, 
>> MemoryRegion *mem)
>>           g_assert_not_reached();
>>       }
>> -    if (kvm_enabled() && vms->virt &&
>> -        (revision != 3 || !kvm_irqchip_in_kernel())) {
>> +    if (kvm_enabled() && vms->virt && (revision != 3 || ! 
>> kvm_irqchip_in_kernel())) {
>>           error_report("KVM EL2 is only supported with in-kernel GICv3");
>>           exit(1);
>>       }
>> @@ -2281,7 +2280,8 @@ static void machvirt_init(MachineState *machine)
>>           exit(1);
>>       }
>> -    if (vms->virt && !kvm_enabled() && !tcg_enabled() && ! 
>> qtest_enabled()) {
>> +    if (vms->virt && !kvm_enabled() && !tcg_enabled()
>> +       && !hvf_enabled() && !qtest_enabled()) {
>>           error_report("mach-virt: %s does not support providing "
>>                        "Virtualization extensions to the guest CPU",
>>                        current_accel_name());
>> @@ -2553,6 +2553,9 @@ static void virt_set_virt(Object *obj, bool 
>> value, Error **errp)
>>       VirtMachineState *vms = VIRT_MACHINE(obj);
>>       vms->virt = value;
>> +#if defined(CONFIG_HVF) && defined(__aarch64__)
>> +    hvf_arm_el2_enable(value);
>> +#endif
>>   }

> We don't have hvf_arm_el1_enable(). I think we just want to always
> provide the most performant configuration to users.
> IOW, we don't need hvf_arm_el2_enable(). I'd just always enable EL2
> (if supported) in hvf_arm_get_host_cpu_features().

We create vCPUs in machvirt_init(). Upon vCPU INITialization, we check
host features. Then during REALIZation, we check the QOM properties.
If 'virt' flag is disabled, machvirt_init() sets "has_el2" property
to false, effectively setting ARMCPU::has_el2 field to %false.
In arm_cpu_realizefn(), if ARMCPU::has_el2 we unset ARM_FEATURE_EL2.