Re: [PATCH v8 2/9] KVM: arm/arm64: Share common code in user_mem_abort()

[Punit Agrawal]

Marc Zyngier <marc.zyngier@xxxxxxx> writes:

> Hi Punit,
>
> On 01/10/18 16:54, Punit Agrawal wrote:
>> The code for operations such as marking the pfn as dirty, and
>> dcache/icache maintenance during stage 2 fault handling is duplicated
>> between normal pages and PMD hugepages.
>>
>> Instead of creating another copy of the operations when we introduce
>> PUD hugepages, let's share them across the different pagesizes.
>>
>> Signed-off-by: Punit Agrawal <punit.agrawal@xxxxxxx>
>> Cc: Suzuki K Poulose <suzuki.poulose@xxxxxxx>
>> Cc: Christoffer Dall <christoffer.dall@xxxxxxx>
>> Cc: Marc Zyngier <marc.zyngier@xxxxxxx>
>> ---
>>   virt/kvm/arm/mmu.c | 45 +++++++++++++++++++++++++++++----------------
>>   1 file changed, 29 insertions(+), 16 deletions(-)
>>
>> diff --git a/virt/kvm/arm/mmu.c b/virt/kvm/arm/mmu.c
>> index c23a1b323aad..5b76ee204000 100644
>> --- a/virt/kvm/arm/mmu.c
>> +++ b/virt/kvm/arm/mmu.c
>> @@ -1490,7 +1490,7 @@ static int user_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa,
>>   	kvm_pfn_t pfn;
>>   	pgprot_t mem_type = PAGE_S2;
>>   	bool logging_active = memslot_is_logging(memslot);
>> -	unsigned long flags = 0;
>> +	unsigned long vma_pagesize, flags = 0;
>>     	write_fault = kvm_is_write_fault(vcpu);
>>   	exec_fault = kvm_vcpu_trap_is_iabt(vcpu);
>> @@ -1510,10 +1510,17 @@ static int user_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa,
>>   		return -EFAULT;
>>   	}
>>   -	if (vma_kernel_pagesize(vma) == PMD_SIZE && !logging_active) {
>> +	vma_pagesize = vma_kernel_pagesize(vma);
>> +	if (vma_pagesize == PMD_SIZE && !logging_active) {
>>   		hugetlb = true;
>>   		gfn = (fault_ipa & PMD_MASK) >> PAGE_SHIFT;
>>   	} else {
>> +		/*
>> +		 * Fallback to PTE if it's not one of the Stage 2
>> +		 * supported hugepage sizes
>> +		 */
>> +		vma_pagesize = PAGE_SIZE;
>> +
>>   		/*
>>   		 * Pages belonging to memslots that don't have the same
>>   		 * alignment for userspace and IPA cannot be mapped using
>> @@ -1579,23 +1586,34 @@ static int user_mem_abort(struct kvm_vcpu *vcpu, phys_addr_t fault_ipa,
>>   	if (mmu_notifier_retry(kvm, mmu_seq))
>>   		goto out_unlock;
>>   -	if (!hugetlb && !force_pte)
>> +	if (!hugetlb && !force_pte) {
>> +		/*
>> +		 * Only PMD_SIZE transparent hugepages(THP) are
>> +		 * currently supported. This code will need to be
>> +		 * updated to support other THP sizes.
>> +		 */
>>   		hugetlb = transparent_hugepage_adjust(&pfn, &fault_ipa);
>> +		if (hugetlb)
>> +			vma_pagesize = PMD_SIZE;
>> +	}
>> +
>> +	if (writable)
>> +		kvm_set_pfn_dirty(pfn);
>>   -	if (hugetlb) {
>> +	if (fault_status != FSC_PERM)
>> +		clean_dcache_guest_page(pfn, vma_pagesize);
>> +
>> +	if (exec_fault)
>> +		invalidate_icache_guest_page(pfn, vma_pagesize);
>> +
>> +	if (hugetlb && vma_pagesize == PMD_SIZE) {
>
> Can you end-up in a situation where hugetlb==false and vma_pagesize ==
> PMD_SIZE? If that's the case, then the above CMOs are not done on the
> same granularity as they were done before this patch. If that cannot
> happen, then the above condition can be simplified.
>
> Which one is it?

hugetlb is a hangover from when we didn't have vma_pagesize. I think we
can drop it and rely on the pagesize to control the size of mapping we
put down.

Let me give that a try.

Thanks for taking a look.

>
>
> Thanks,
>
> 	M.