Re: [PATCH v4 2/2] kvm: Use huge pages for DAX-backed files

[Liran Alon <liran.alon@oracle.com>]

> On 12 Dec 2019, at 18:54, Dan Williams <dan.j.williams@intel.com> wrote:
> 
> On Thu, Dec 12, 2019 at 4:34 AM Liran Alon <liran.alon@oracle.com> wrote:
>> 
>> 
>> 
>>> On 11 Dec 2019, at 23:32, Barret Rhoden <brho@google.com> wrote:
>>> 
>>> This change allows KVM to map DAX-backed files made of huge pages with
>>> huge mappings in the EPT/TDP.
>>> 
>>> DAX pages are not PageTransCompound.  The existing check is trying to
>>> determine if the mapping for the pfn is a huge mapping or not.  For
>>> non-DAX maps, e.g. hugetlbfs, that means checking PageTransCompound.
>>> For DAX, we can check the page table itself.
>> 
>> For hugetlbfs pages, tdp_page_fault() -> mapping_level() -> host_mapping_level() -> kvm_host_page_size() -> vma_kernel_pagesize()
>> will return the page-size of the hugetlbfs without the need to parse the page-tables.
>> See vma->vm_ops->pagesize() callback implementation at hugetlb_vm_ops->pagesize()==hugetlb_vm_op_pagesize().
>> 
>> Only for pages that were originally mapped as small-pages and later merged to larger pages by THP, there is a need to check for PageTransCompound(). Again, instead of parsing page-tables.
>> 
>> Therefore, it seems more logical to me that:
>> (a) If DAX-backed files are mapped as large-pages to userspace, it should be reflected in vma->vm_ops->page_size() of that mapping. Causing kvm_host_page_size() to return the right size without the need to parse the page-tables.
> 
> A given dax-mapped vma may have mixed page sizes so ->page_size()
> can't be used reliably to enumerating the mapping size.

Naive question: Why don’t split the VMA in this case to multiple VMAs with different results for ->page_size()?
What you are describing sounds like DAX is breaking this callback semantics in an unpredictable manner.

> 
>> (b) If DAX-backed files small-pages can be later merged to large-pages by THP, then the “struct page” of these pages should be modified as usual to make PageTransCompound() return true for them. I’m not highly familiar with this mechanism, but I would expect THP to be able to merge DAX-backed files small-pages to large-pages in case DAX provides “struct page” for the DAX pages.
> 
> DAX pages do not participate in THP and do not have the
> PageTransCompound accounting. The only mechanism that records the
> mapping size for dax is the page tables themselves.

What is the rational behind this? Given that DAX pages can be described with “struct page” (i.e. ZONE_DEVICE), what prevents THP from manipulating page-tables to merge multiple DAX PFNs to a larger page?

-Liran

> 
> 
>> 
>>> 
>>> Note that KVM already faulted in the page (or huge page) in the host's
>>> page table, and we hold the KVM mmu spinlock.  We grabbed that lock in
>>> kvm_mmu_notifier_invalidate_range_end, before checking the mmu seq.
>>> 
>>> Signed-off-by: Barret Rhoden <brho@google.com>
>>> ---
>>> arch/x86/kvm/mmu/mmu.c | 36 ++++++++++++++++++++++++++++++++----
>>> 1 file changed, 32 insertions(+), 4 deletions(-)
>>> 
>>> diff --git a/arch/x86/kvm/mmu/mmu.c b/arch/x86/kvm/mmu/mmu.c
>>> index 6f92b40d798c..cd07bc4e595f 100644
>>> --- a/arch/x86/kvm/mmu/mmu.c
>>> +++ b/arch/x86/kvm/mmu/mmu.c
>>> @@ -3384,6 +3384,35 @@ static int kvm_handle_bad_page(struct kvm_vcpu *vcpu, gfn_t gfn, kvm_pfn_t pfn)
>>>      return -EFAULT;
>>> }
>>> 
>>> +static bool pfn_is_huge_mapped(struct kvm *kvm, gfn_t gfn, kvm_pfn_t pfn)
>>> +{
>>> +     struct page *page = pfn_to_page(pfn);
>>> +     unsigned long hva;
>>> +
>>> +     if (!is_zone_device_page(page))
>>> +             return PageTransCompoundMap(page);
>>> +
>>> +     /*
>>> +      * DAX pages do not use compound pages.  The page should have already
>>> +      * been mapped into the host-side page table during try_async_pf(), so
>>> +      * we can check the page tables directly.
>>> +      */
>>> +     hva = gfn_to_hva(kvm, gfn);
>>> +     if (kvm_is_error_hva(hva))
>>> +             return false;
>>> +
>>> +     /*
>>> +      * Our caller grabbed the KVM mmu_lock with a successful
>>> +      * mmu_notifier_retry, so we're safe to walk the page table.
>>> +      */
>>> +     switch (dev_pagemap_mapping_shift(hva, current->mm)) {
>> 
>> Doesn’t dev_pagemap_mapping_shift() get “struct page” as first parameter?
>> Was this changed by a commit I missed?
>> 
>> -Liran
>> 
>>> +     case PMD_SHIFT:
>>> +     case PUD_SIZE:
>>> +             return true;
>>> +     }
>>> +     return false;
>>> +}
>>> +
>>> static void transparent_hugepage_adjust(struct kvm_vcpu *vcpu,
>>>                                      gfn_t gfn, kvm_pfn_t *pfnp,
>>>                                      int *levelp)
>>> @@ -3398,8 +3427,8 @@ static void transparent_hugepage_adjust(struct kvm_vcpu *vcpu,
>>>       * here.
>>>       */
>>>      if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn) &&
>>> -         !kvm_is_zone_device_pfn(pfn) && level == PT_PAGE_TABLE_LEVEL &&
>>> -         PageTransCompoundMap(pfn_to_page(pfn)) &&
>>> +         level == PT_PAGE_TABLE_LEVEL &&
>>> +         pfn_is_huge_mapped(vcpu->kvm, gfn, pfn) &&
>>>          !mmu_gfn_lpage_is_disallowed(vcpu, gfn, PT_DIRECTORY_LEVEL)) {
>>>              unsigned long mask;
>>>              /*
>>> @@ -6015,8 +6044,7 @@ static bool kvm_mmu_zap_collapsible_spte(struct kvm *kvm,
>>>               * mapping if the indirect sp has level = 1.
>>>               */
>>>              if (sp->role.direct && !kvm_is_reserved_pfn(pfn) &&
>>> -                 !kvm_is_zone_device_pfn(pfn) &&
>>> -                 PageTransCompoundMap(pfn_to_page(pfn))) {
>>> +                 pfn_is_huge_mapped(kvm, sp->gfn, pfn)) {
>>>                      pte_list_remove(rmap_head, sptep);
>>> 
>>>                      if (kvm_available_flush_tlb_with_range())
>>> --
>>> 2.24.0.525.g8f36a354ae-goog
>>> 
>>