Re: [RFC PATCH 1/2] KVM: arm64: nv: selftests: Add guest hypervisor test

[Ganapatrao Kulkarni <gankulkarni@os.amperecomputing.com>]

On 07-02-2025 07:29 pm, Marc Zyngier wrote:
> On Fri, 07 Feb 2025 13:26:41 +0000,
> Ganapatrao Kulkarni <gankulkarni@os.amperecomputing.com> wrote:
>>
>>>> +	if (is_vcpu_nested(vcpu)) {
>>>> +		vcpu_set_reg(vcpu, KVM_ARM64_SYS_REG(SYS_CPTR_EL2), fpen);
>>>> +		vcpu_set_reg(vcpu, KVM_ARM64_SYS_REG(SYS_SCTLR_EL2), sctlr_el1);
>>>> +		vcpu_set_reg(vcpu, KVM_ARM64_SYS_REG(SYS_TCR_EL2), tcr_el1);
>>>> +		vcpu_set_reg(vcpu, KVM_ARM64_SYS_REG(SYS_MAIR_EL2), DEFAULT_MAIR_EL1);
>>>> +		vcpu_set_reg(vcpu, KVM_ARM64_SYS_REG(SYS_TTBR0_EL2), ttbr0_el1);
>>>> +		vcpu_set_reg(vcpu, KVM_ARM64_SYS_REG(SYS_TPIDR_EL2), vcpu->id);
>>>
>>> How about some of the basics such as HCR_EL2, MDCR_EL2? A bunch of
>>> things there do have an impact on how the guest behaves, and relying
>>> on defaults feels like a bad idea.
>>
>> Sure, I will try to have these registers also set to required value
>> explicitly.
>>
>>>
>>> This also assumes VHE, without trying to enforce it.
>>
>> Ok, I will try to set specific bits of HCR_EL2 to force it run in VHE.
>>
>>>
>>> Finally, how to you plan to make all the existing tests run as EL2
>>> guests if TPIDR_EL1 isn't populated with the expected value? Surely
>>> you need to change the read side...
>>
>> OK, I have not yet tried running existing tests modifying to run as
>> EL2 guests. I will try to run them modifying to run in vEL2.
> 
> You shouldn't try to modify them. Just make them take a parameter so
> that the initialisation is done by configuring everything at EL2.
> 

OK, make sense, thanks.
I will work on one of the test case and post it in the v2.
>>
>> Should we modify them to run as EL2 guests by default, if the host
>> supports/detected NV? or command line argument based run? either in
>> El1(default) or in EL2?
> 
> EL1 by default.

OK.
> 
>>
>> BTW, I have also ran all existing tests on L1, most of the tests are
>> passing(atleast I did not see any failure prints).
>>
>> arm64/debug-exceptions is failing on L1 and needs to be
>> debugged/fixed/skipped.
>> arm64/arch_timer_edge_cases fails on both L0 and L1.
> 
> Then I guess you have some work to do to debug these problems, and it
> once more means that NV is not ready for merging.

I will debug and share the findings/fix at the earliest.

-- 
Thanks,
Ganapat/GK