Re: [PATCH v17 10/20] KVM: arm64: coco: Disable Steal time accounting for coco guests

[Suzuki K Poulose <suzuki.poulose@arm.com>]

Hi Fuad

On 09/09/2026 13:23, Fuad Tabba wrote:
> Hi Suzuki,
> 
> On Wed, 9 Sept 2026 at 12:52, Suzuki K Poulose <suzuki.poulose@arm.com> wrote:
> ...
> 
>>> I think it would be better to leave KVM_CAP_STEAL_TIME reported and
>>> instead reject the configuration for these VMs, returning -EPERM from
>>> the KVM_ARM_VCPU_PVTIME_CTRL attr. A capability query should report
>>> what the kernel supports, not whether a given VM may use it; gating it
>>> per-VM here is also inconsistent, since the !kvm clause leaves the
>>> system-fd cap at 1 while the VM-fd cap reads 0. And -EPERM says that
>>
>>
>> But isn't that we want from a VMM perspective ? i.e., enable PVTIME if
>> it is supported for the given VM (type). Rather than PVTIME is supported
>> by the KVM, so I can go enable it by default. But, we endup getting a
>> failure while we enable it. I encountered this with kvmtool, where
>> CAP_STEAL_TIME is reported true and even the HAS_DEVICE_ATTR goes
>> through fine, but SET_DEVICE_ATTR seemed to fail.
>>
>> Isn't it better to allow the VM to decide if the capability is supported
>> for the given VM type, like we are doing for other capabilities ?
> 
> You're right, per-VM capability reporting. My wires got crossed from
> the pKVM series and the kvmtool fixes. However, I'd still return
> -EPERM rather than -ENXIO from the attribute. The capability exists,

Just to make sure I understand the picture:

For coco VMs:

  - check extension on coco KVM, KVM_CAP_STEAL_TIME => 0

And a follow up DEVICE_HAS_ATTR => -EPERM ?

But we tell the VMM, look you don't have the capabilit on this VM.
But if you try to set the attribute isn't it more appropriate to
say:
  -ENXIO - No you don't have that feature on this VM (-ENXIO)
rather than
  -EPERM - No, I won't let you do that.

Remember we are doing only in the "VM" context. It is a minor nit
but I wanted to make sure we understand what we are thinking.

Cheers
Suzuki

> it's just not permitted for a confidential VM, and -EPERM says exactly
> that, whereas -ENXIO reads as "no such feature".
> 
> Cheers,
> /fuad
> 
>>
>>> steal-time exists but isn't permitted for a confidential VM, whereas
>>> -ENXIO reads as "no such feature".
>>
>> Cheers
>> Suzuki
>>
>>
>>>
>>> Cheers,
>>> /fuad
>>>
>>>>
>>>>    int kvm_arm_pvtime_set_attr(struct kvm_vcpu *vcpu,
>>>> @@ -81,7 +81,7 @@ int kvm_arm_pvtime_set_attr(struct kvm_vcpu *vcpu,
>>>>           int ret = 0;
>>>>           int idx;
>>>>
>>>> -       if (!kvm_arm_pvtime_supported() ||
>>>> +       if (!kvm_arm_pvtime_supported(kvm) ||
>>>>               attr->attr != KVM_ARM_VCPU_PVTIME_IPA)
>>>>                   return -ENXIO;
>>>>
>>>> @@ -110,7 +110,7 @@ int kvm_arm_pvtime_get_attr(struct kvm_vcpu *vcpu,
>>>>           u64 __user *user = (u64 __user *)attr->addr;
>>>>           u64 ipa;
>>>>
>>>> -       if (!kvm_arm_pvtime_supported() ||
>>>> +       if (!kvm_arm_pvtime_supported(vcpu->kvm) ||
>>>>               attr->attr != KVM_ARM_VCPU_PVTIME_IPA)
>>>>                   return -ENXIO;
>>>>
>>>> @@ -126,7 +126,7 @@ int kvm_arm_pvtime_has_attr(struct kvm_vcpu *vcpu,
>>>>    {
>>>>           switch (attr->attr) {
>>>>           case KVM_ARM_VCPU_PVTIME_IPA:
>>>> -               if (kvm_arm_pvtime_supported())
>>>> +               if (kvm_arm_pvtime_supported(vcpu->kvm))
>>>>                           return 0;
>>>>           }
>>>>           return -ENXIO;
>>>> --
>>>> 2.43.0
>>>>
>>