Christoffer Dall <christoffer.d...@arm.com> writes:
> On Fri, Apr 20, 2018 at 03:54:09PM +0100, Punit Agrawal wrote:
>> KVM only supports PMD hugepages at stage 2. Extend the stage 2 fault
>> handling to add support for PUD hugepages.
>>
>> Addition of pud hugepage support enables additional hugepage
>> sizes (e.g., 1G with 4K granule) which can be useful on cores that
>> support mapping larger block sizes in the TLB entries.
>>
>> Signed-off-by: Punit Agrawal <punit.agra...@arm.com>
>> Cc: Christoffer Dall <christoffer.d...@arm.com>
>> Cc: Marc Zyngier <marc.zyng...@arm.com>
>> Cc: Russell King <li...@armlinux.org.uk>
>> Cc: Catalin Marinas <catalin.mari...@arm.com>
>> Cc: Will Deacon <will.dea...@arm.com>
>> ---
>> arch/arm/include/asm/kvm_mmu.h | 19 +++++++++
>> arch/arm64/include/asm/kvm_mmu.h | 15 +++++++
>> arch/arm64/include/asm/pgtable-hwdef.h | 4 ++
>> arch/arm64/include/asm/pgtable.h | 2 +
>> virt/kvm/arm/mmu.c | 54 ++++++++++++++++++++------
>> 5 files changed, 83 insertions(+), 11 deletions(-)
>>
[...]
>> diff --git a/virt/kvm/arm/mmu.c b/virt/kvm/arm/mmu.c
[...]
>> @@ -1452,9 +1472,12 @@ static int user_mem_abort(struct kvm_vcpu *vcpu,
>> phys_addr_t fault_ipa,
>> }
>>
>> vma_pagesize = vma_kernel_pagesize(vma);
>> - if (vma_pagesize == PMD_SIZE && !logging_active) {
>> + if ((vma_pagesize == PMD_SIZE || vma_pagesize == PUD_SIZE) &&
>> + !logging_active) {
>> + struct hstate *h = hstate_vma(vma);
>> +
>> hugetlb = true;
>> - gfn = (fault_ipa & PMD_MASK) >> PAGE_SHIFT;
>> + gfn = (fault_ipa & huge_page_mask(h)) >> PAGE_SHIFT;
>> } else {
>> /*
>> * Pages belonging to memslots that don't have the same
>> @@ -1521,15 +1544,8 @@ 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) {
>> - /*
>> - * Only PMD_SIZE transparent hugepages(THP) are
>> - * currently supported. This code will need to be
>> - * updated if other THP sizes are supported.
>> - */
>> + if (!hugetlb && !force_pte)
>> hugetlb = transparent_hugepage_adjust(&pfn, &fault_ipa);
>> - vma_pagesize = PMD_SIZE;
>
> Why this change? Won't you end up trying to map THPs as individual
> pages now?
Argh - that's a rebase gone awry. Thanks for spotting.
There's another issue with this hunk - hugetlb can be false after the
call to transparent_hugepage_adjust(). I've fixed that up for the next
update.
>
>> - }
>>
>> if (writable)
>> kvm_set_pfn_dirty(pfn);
>> @@ -1540,7 +1556,23 @@ static int user_mem_abort(struct kvm_vcpu *vcpu,
>> phys_addr_t fault_ipa,
>> if (exec_fault)
>> invalidate_icache_guest_page(pfn, vma_pagesize);
>>
>> - if (hugetlb) {
>> + if (vma_pagesize == PUD_SIZE) {
>> + pud_t new_pud = kvm_pfn_pud(pfn, mem_type);
>> +
>> + new_pud = kvm_pud_mkhuge(new_pud);
>> + if (writable)
>> + new_pud = kvm_s2pud_mkwrite(new_pud);
>> +
>> + if (exec_fault) {
>> + new_pud = kvm_s2pud_mkexec(new_pud);
>> + } else if (fault_status == FSC_PERM) {
>> + /* Preserve execute if XN was already cleared */
>> + if (stage2_is_exec(kvm, fault_ipa))
>> + new_pud = kvm_s2pud_mkexec(new_pud);
>> + }
>
> aha, another reason for my suggestion in the other patch.
Ack! Already fixed locally.
>
>> +
>> + ret = stage2_set_pud_huge(kvm, memcache, fault_ipa, &new_pud);
>> + } else if (vma_pagesize == PMD_SIZE) {
>> pmd_t new_pmd = kvm_pfn_pmd(pfn, mem_type);
>>
>> new_pmd = kvm_pmd_mkhuge(new_pmd);
>> --
>> 2.17.0
>>
>
> Otherwise, this patch looks fine.
Thanks a lot for reviewing the patches. I'll send out an update
incorporating your suggestions.
Punit
