[PATCH 2/2] KVM: arm64: nv: fixup! Support multiple nested Stage-2 mmu structures

[Ganapatrao Kulkarni]

On 27-11-2021 12:50 am, Marc Zyngier wrote:
> [resending after having sorted my email config]
> 
> On Fri, 26 Nov 2021 05:59:00 +0000,
> Ganapatrao Kulkarni <gankulkarni at os.amperecomputing.com> wrote:
>>
>> Hi Marc,
>>
>> On 25-11-2021 07:53 pm, Marc Zyngier wrote:
>>> On Mon, 22 Nov 2021 09:58:03 +0000,
>>> Ganapatrao Kulkarni <gankulkarni at os.amperecomputing.com> wrote:
>>>>
>>>> Commit 1776c91346b6 ("KVM: arm64: nv: Support multiple nested Stage-2 mmu
>>>> structures")[1] added a function kvm_vcpu_init_nested which expands the
>>>> stage-2 mmu structures array when ever a new vCPU is created. The array
>>>> is expanded using krealloc() and results in a stale mmu address pointer
>>>> in pgt->mmu. Adding a fix to update the pointer with the new address after
>>>> successful krealloc.
>>>>
>>>> [1] https://git.kernel.org/pub/scm/linux/kernel/git/maz/arm-platforms.git/
>>>> branch kvm-arm64/nv-5.13
>>>>
>>>> Signed-off-by: Ganapatrao Kulkarni <gankulkarni at os.amperecomputing.com>
>>>> ---
>>>>    arch/arm64/kvm/nested.c | 9 +++++++++
>>>>    1 file changed, 9 insertions(+)
>>>>
>>>> diff --git a/arch/arm64/kvm/nested.c b/arch/arm64/kvm/nested.c
>>>> index 4ffbc14d0245..57ad8d8f4ee5 100644
>>>> --- a/arch/arm64/kvm/nested.c
>>>> +++ b/arch/arm64/kvm/nested.c
>>>> @@ -68,6 +68,8 @@ int kvm_vcpu_init_nested(struct kvm_vcpu *vcpu)
>>>>    		       num_mmus * sizeof(*kvm->arch.nested_mmus),
>>>>    		       GFP_KERNEL | __GFP_ZERO);
>>>>    	if (tmp) {
>>>> +		int i;
>>>> +
>>>>    		if (kvm_init_stage2_mmu(kvm, &tmp[num_mmus - 1]) ||
>>>>    		    kvm_init_stage2_mmu(kvm, &tmp[num_mmus - 2])) {
>>>>    			kvm_free_stage2_pgd(&tmp[num_mmus - 1]);
>>>> @@ -80,6 +82,13 @@ int kvm_vcpu_init_nested(struct kvm_vcpu *vcpu)
>>>>    		}
>>>>      		kvm->arch.nested_mmus = tmp;
>>>> +
>>>> +		/* Fixup pgt->mmu after krealloc */
>>>> +		for (i = 0; i < kvm->arch.nested_mmus_size; i++) {
>>>> +			struct kvm_s2_mmu *mmu = &kvm->arch.nested_mmus[i];
>>>> +
>>>> +			mmu->pgt->mmu = mmu;
>>>> +		}
>>>>    	}
>>>>      	mutex_unlock(&kvm->lock);
>>>
>>> Another good catch. I've tweaked a bit to avoid some unnecessary
>>> repainting, see below.
>>>
>>> Thanks again,
>>>
>>> 	M.
>>>
>>> diff --git a/arch/arm64/kvm/nested.c b/arch/arm64/kvm/nested.c
>>> index a4dfffa1dae0..92b225db59ac 100644
>>> --- a/arch/arm64/kvm/nested.c
>>> +++ b/arch/arm64/kvm/nested.c
>>> @@ -66,8 +66,19 @@ int kvm_vcpu_init_nested(struct kvm_vcpu *vcpu)
>>>    	num_mmus = atomic_read(&kvm->online_vcpus) * 2;
>>>    	tmp = krealloc(kvm->arch.nested_mmus,
>>>    		       num_mmus * sizeof(*kvm->arch.nested_mmus),
>>> -		       GFP_KERNEL | __GFP_ZERO);
>>> +		       GFP_KERNEL_ACCOUNT | __GFP_ZERO);
>>>    	if (tmp) {
>>> +		/*
>>> +		 * If we went through a realocation, adjust the MMU
>>
>> Is it more precise to say?
>>> +		 * back-pointers in the pg_table structures.
>> * back-pointers in the pg_table structures of previous inits.
> 
> Yes. I have added something along those lines.
> 
>>> +		 */
>>> +		if (kvm->arch.nested_mmus != tmp) {
>>> +			int i;
>>> +
>>> +			for (i = 0; i < num_mms - 2; i++)
>>> +				tmp[i].pgt->mmu = &tmp[i];
>>> +		}
>>
>> Thanks for this optimization, it saves 2 redundant iterations.
>>> +
>>>    		if (kvm_init_stage2_mmu(kvm, &tmp[num_mmus - 1]) ||
>>>    		    kvm_init_stage2_mmu(kvm, &tmp[num_mmus - 2])) {
>>>    			kvm_free_stage2_pgd(&tmp[num_mmus - 1]);
>>>
>>
>> Feel free to add,
>> Reviewed-by: Ganapatrao Kulkarni <gankulkarni at os.amperecomputing.com>
> 
> Given that this was a fixup, I haven't taken this tag. I will Cc you

no problem, makes sense to fold this in to original patch.

> on the whole series, and you can give you tag on the whole patch if
> you are happy with it.

Sure.
> 
> BTW, I have now fixed the bug that was preventing L2 userspace from
> running (bad interaction with the pgtable code which was unhappy about
> my use of the SW bits when relaxing the permissions). You should now
> be able to test the whole series.

Yes, I have rebased to latest branch kvm-arm64/nv-5.16 and I am able to 
boot L1 and L2.

> 
> Thanks,
> 
> 	M.
> 

Thanks,
Ganapat