Re: [PATCH v2 2/2] mm: Allocate THP on hugezeropage wp-fault

[Ryan Roberts <ryan.roberts@arm.com>]

On 06/09/2024 10:00, Dev Jain wrote:
> 
> On 9/6/24 14:13, Ryan Roberts wrote:
>> On 06/09/2024 08:05, Dev Jain wrote:
>>> On 9/5/24 18:44, Ryan Roberts wrote:
>>>> On 04/09/2024 11:09, Dev Jain wrote:
>>>>> Introduce do_huge_zero_wp_pmd() to handle wp-fault on a hugezeropage and
>>>>> replace it with a PMD-mapped THP. Change the helpers introduced in the
>>>>> previous patch to flush TLB entry corresponding to the hugezeropage,
>>>>> and preserve PMD uffd-wp marker. In case of failure, fallback to
>>>>> splitting the PMD.
>>>>>
>>>>> Signed-off-by: Dev Jain <dev.jain@arm.com>
>>>>> ---
>>>>>    include/linux/huge_mm.h |  6 ++++
>>>>>    mm/huge_memory.c        | 79 +++++++++++++++++++++++++++++++++++------
>>>>>    mm/memory.c             |  5 +--
>>>>>    3 files changed, 78 insertions(+), 12 deletions(-)
>>>>>
>>>>> diff --git a/include/linux/huge_mm.h b/include/linux/huge_mm.h
>>>>> index e25d9ebfdf89..fdd2cf473a3c 100644
>>>>> --- a/include/linux/huge_mm.h
>>>>> +++ b/include/linux/huge_mm.h
>>>>> @@ -9,6 +9,12 @@
>>>>>    #include <linux/kobject.h>
>>>>>      vm_fault_t do_huge_pmd_anonymous_page(struct vm_fault *vmf);
>>>>> +vm_fault_t thp_fault_alloc(gfp_t gfp, int order, struct vm_area_struct *vma,
>>>>> +               unsigned long haddr, struct folio **foliop,
>>>>> +               unsigned long addr);
>>>>> +void map_pmd_thp(struct folio *folio, struct vm_fault *vmf,
>>>>> +         struct vm_area_struct *vma, unsigned long haddr,
>>>>> +         pgtable_t pgtable);
>>>> I don't think you are using either of these outside of huge_memory.c, so not
>>>> sure you need to declare them here or make them non-static?
>>> As pointed out by Kirill, you are right, I forgot to drop these from my previous
>>> approach.
>>>
>>>>>    int copy_huge_pmd(struct mm_struct *dst_mm, struct mm_struct *src_mm,
>>>>>              pmd_t *dst_pmd, pmd_t *src_pmd, unsigned long addr,
>>>>>              struct vm_area_struct *dst_vma, struct vm_area_struct *src_vma);
>>>>> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
>>>>> index 58125fbcc532..150163ad77d3 100644
>>>>> --- a/mm/huge_memory.c
>>>>> +++ b/mm/huge_memory.c
>>>>> @@ -943,9 +943,9 @@ unsigned long thp_get_unmapped_area(struct file *filp,
>>>>> unsigned long addr,
>>>>>    }
>>>>>    EXPORT_SYMBOL_GPL(thp_get_unmapped_area);
>>>>>    -static vm_fault_t thp_fault_alloc(gfp_t gfp, int order, struct
>>>>> vm_area_struct *vma,
>>>>> -                  unsigned long haddr, struct folio **foliop,
>>>>> -                  unsigned long addr)
>>>>> +vm_fault_t thp_fault_alloc(gfp_t gfp, int order, struct vm_area_struct *vma,
>>>>> +               unsigned long haddr, struct folio **foliop,
>>>>> +               unsigned long addr)
>>>>>    {
>>>>>        struct folio *folio = vma_alloc_folio(gfp, order, vma, haddr, true);
>>>>>    @@ -984,21 +984,29 @@ static void __thp_fault_success_stats(struct
>>>>> vm_area_struct *vma, int order)
>>>>>        count_memcg_event_mm(vma->vm_mm, THP_FAULT_ALLOC);
>>>>>    }
>>>>>    -static void map_pmd_thp(struct folio *folio, struct vm_fault *vmf,
>>>>> -            struct vm_area_struct *vma, unsigned long haddr,
>>>>> -            pgtable_t pgtable)
>>>>> +void map_pmd_thp(struct folio *folio, struct vm_fault *vmf,
>>>>> +         struct vm_area_struct *vma, unsigned long haddr,
>>>>> +         pgtable_t pgtable)
>>>>>    {
>>>>> -    pmd_t entry;
>>>>> +    pmd_t entry, old_pmd;
>>>>> +    bool is_pmd_none = pmd_none(*vmf->pmd);
>>>>>          entry = mk_huge_pmd(&folio->page, vma->vm_page_prot);
>>>>>        entry = maybe_pmd_mkwrite(pmd_mkdirty(entry), vma);
>>>>>        folio_add_new_anon_rmap(folio, vma, haddr, RMAP_EXCLUSIVE);
>>>>>        folio_add_lru_vma(folio, vma);
>>>>> -    pgtable_trans_huge_deposit(vma->vm_mm, vmf->pmd, pgtable);
>>>>> +    if (!is_pmd_none) {
>>>>> +        old_pmd = pmdp_huge_clear_flush(vma, haddr, vmf->pmd);
>>>>> +        if (pmd_uffd_wp(old_pmd))
>>>>> +            entry = pmd_mkuffd_wp(entry);
>>>> I don't really get this; entry is writable, so I wouldn't expect to also be
>>>> setting uffd-wp here? That combination is not allowed and is checked for in
>>>> page_table_check_pte_flags().
>>>>
>>>> It looks like you expect to get here in the uffd-wp-async case, which used to
>>>> cause the pmd to be split to ptes. I'm guessing (but don't know for sure) that
>>>> would cause the uffd-wp bit to be set in each of the new ptes, then during
>>>> fallback to handling the wp fault on the pte, uffd would handle it?
>>> I guess you are correct; I missed the WARN_ON() in page_table_check_pmd_flags(),
>>> but I did see, if I read the uffd code correctly, that mfill_atomic() will just
>>> return in case of pmd_trans_huge(*dst_pmd) while doing a uffd_copy to the
>>> destination
>>> location. So preserving pmd_uffd_wp is useless in case a THP is mapped, but I
>>> did not
>>> know that in fact it is supposed to be an invalid combination. So, I will
>>> drop it,
>>> unless someone has some other objection.
>> So what's the correct way to handle uffd-wp-async in wp_huge_pmd()? Just split
>> it? If so, you can revert your changes in memory.c.
> 
> I think so.
> 
>>
>>>>> +    }
>>>>> +    if (pgtable)
>>>>> +        pgtable_trans_huge_deposit(vma->vm_mm, vmf->pmd, pgtable);
>>>> Should this call be moved outside of here? It doesn't really feel like it
>>>> belongs. Could it be called before calling map_pmd_thp() for the site that
>>>> has a
>>>> pgtable?
>>> Every other place I checked, they are doing this: make the entry -> deposit
>>> pgtable ->
>>> set_pmd_at(). I guess the general flow is to do the deposit based on the old
>>> pmd, before
>>> setting the new one. Which implies: I should at least move this check before
>>> I call
>>> pmdp_huge_clear_flush(). And, since vmf->pmd and creating the new entry has no
>>> relation,
>>> I am inclined to do what you are saying.
>> pgtable_trans_huge_deposit() is just adding the pgtable to a list so that if the
>> THP needs to be split in future, then we have preallocated the pte pgtable so
>> the operation can't fail.
> 
> Yes.
> 
>> And enqueing it is just under the protection of the
>> PTL as far as I can tell. So I think the only ordering requirement is that you
>> both set the pmd and deposit the pgtable under the lock (without dropping it in
>> between). So you can deposit either before or after map_pmd_thp().
> 
> Yes I'll do that before.
> 
>> And
>> pmdp_huge_clear_flush() is irrelavent, I think?
> 
> You mean, in this context? Everywhere, pgtable deposit uses the old pmd
> value to be replaced as its input, that is, it is called before set_pmd_at().
> So calling pgtable deposit after clear_flush() will violate this ordering.
> I do not think this ordering is really required but I'd rather be safe :)

The pmd pointer is just used to get the pmd table (the pointer points to an
entry inside the table so its just a case of backwards aligning the pointer).
The pointer is never dereferenced, so the value of the entry is irrelevant.

> 
>>
>>>>>        set_pmd_at(vma->vm_mm, haddr, vmf->pmd, entry);
>>>>>        update_mmu_cache_pmd(vma, vmf->address, vmf->pmd);
>>>>>        add_mm_counter(vma->vm_mm, MM_ANONPAGES, HPAGE_PMD_NR);
>>>>> -    mm_inc_nr_ptes(vma->vm_mm);
>>>>> +    if (is_pmd_none)
>>>>> +        mm_inc_nr_ptes(vma->vm_mm);
>>>>>    }
>>>>>      static vm_fault_t __do_huge_pmd_anonymous_page(struct vm_fault *vmf)
>>>>> @@ -1576,6 +1584,50 @@ void huge_pmd_set_accessed(struct vm_fault *vmf)
>>>>>        spin_unlock(vmf->ptl);
>>>>>    }
>>>>>    +static vm_fault_t do_huge_zero_wp_pmd_locked(struct vm_fault *vmf,
>>>>> +                         unsigned long haddr,
>>>>> +                         struct folio *folio)
>>>>> +{
>>>>> +    struct vm_area_struct *vma = vmf->vma;
>>>>> +    vm_fault_t ret = 0;
>>>>> +
>>>>> +    ret = check_stable_address_space(vma->vm_mm);
>>>>> +    if (ret)
>>>>> +        goto out;
>>>>> +    map_pmd_thp(folio, vmf, vma, haddr, NULL);
>>>>> +out:
>>>>> +    return ret;
>>>>> +}
>>>>> +
>>>>> +static vm_fault_t do_huge_zero_wp_pmd(struct vm_fault *vmf, unsigned long
>>>>> haddr)
>>>>> +{
>>>>> +    struct vm_area_struct *vma = vmf->vma;
>>>>> +    gfp_t gfp = vma_thp_gfp_mask(vma);
>>>>> +    struct mmu_notifier_range range;
>>>>> +    struct folio *folio = NULL;
>>>>> +    vm_fault_t ret = 0;
>>>>> +
>>>>> +    ret = thp_fault_alloc(gfp, HPAGE_PMD_ORDER, vma, haddr, &folio,
>>>>> +                  vmf->address);
>>>> Just checking: the PTE table was already allocated during the read fault,
>>>> right?
>>>> So we don't have to allocate it here.
>>> Correct, that happens in set_huge_zero_folio(). Thanks for checking.
>>>
>>>>> +    if (ret)
>>>>> +        goto out;
>>>>> +
>>>>> +    mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, vma->vm_mm, haddr,
>>>>> +                haddr + HPAGE_PMD_SIZE);
>>>>> +    mmu_notifier_invalidate_range_start(&range);
>>>>> +    vmf->ptl = pmd_lock(vma->vm_mm, vmf->pmd);
>>>>> +    if (unlikely(!pmd_same(pmdp_get(vmf->pmd), vmf->orig_pmd)))
>>>>> +        goto unlock;
>>>>> +    ret = do_huge_zero_wp_pmd_locked(vmf, haddr, folio);
>>>>> +    if (!ret)
>>>>> +        __thp_fault_success_stats(vma, HPAGE_PMD_ORDER);
>>>>> +unlock:
>>>>> +    spin_unlock(vmf->ptl);
>>>>> +    mmu_notifier_invalidate_range_end(&range);
>>>> I'll confess I don't understand all the mmu notifier rules.
>>> I confess the same :)
>>>
>>>> But the doc at
>>>> Documentation/mm/mmu_notifier.rst implies that the notification must be done
>>>> while holding the PTL. Although that's not how wp_page_copy(). Are you
>>>> confident
>>>> what you have done is correct?
>>> Everywhere else, invalidate_range_end() is getting called after dropping the
>>> lock,
>>> one reason is that it has a might_sleep(), and therefore we cannot call it while
>>> holding a spinlock. I still don't know what exactly these calls mean...but I
>>> think
>>> what I have done is correct.
>> High level; they are notifying secondary MMUs (e.g. IOMMU) of a change so the
>> tables of those secondary MMUs can be kept in sync. I don't understand all the
>> ordering requirement details though.
>>
>> I think what you have is probably correct, would be good for someone that knows
>> what they are talking about to confirm though :)
> 
> Exactly.
> 
>>
>>>> Thanks,
>>>> Ryan
>>>>
>>>>> +out:
>>>>> +    return ret;
>>>>> +}
>>>>> +
>>>>>    vm_fault_t do_huge_pmd_wp_page(struct vm_fault *vmf)
>>>>>    {
>>>>>        const bool unshare = vmf->flags & FAULT_FLAG_UNSHARE;
>>>>> @@ -1588,8 +1640,15 @@ vm_fault_t do_huge_pmd_wp_page(struct vm_fault *vmf)
>>>>>        vmf->ptl = pmd_lockptr(vma->vm_mm, vmf->pmd);
>>>>>        VM_BUG_ON_VMA(!vma->anon_vma, vma);
>>>>>    -    if (is_huge_zero_pmd(orig_pmd))
>>>>> +    if (is_huge_zero_pmd(orig_pmd)) {
>>>>> +        vm_fault_t ret = do_huge_zero_wp_pmd(vmf, haddr);
>>>>> +
>>>>> +        if (!(ret & VM_FAULT_FALLBACK))
>>>>> +            return ret;
>>>>> +
>>>>> +        /* Fallback to splitting PMD if THP cannot be allocated */
>>>>>            goto fallback;
>>>>> +    }
>>>>>          spin_lock(vmf->ptl);
>>>>>    diff --git a/mm/memory.c b/mm/memory.c
>>>>> index 3c01d68065be..c081a25f5173 100644
>>>>> --- a/mm/memory.c
>>>>> +++ b/mm/memory.c
>>>>> @@ -5409,9 +5409,10 @@ static inline vm_fault_t wp_huge_pmd(struct vm_fault
>>>>> *vmf)
>>>>>        if (vma_is_anonymous(vma)) {
>>>>>            if (likely(!unshare) &&
>>>>>                userfaultfd_huge_pmd_wp(vma, vmf->orig_pmd)) {
>>>>> -            if (userfaultfd_wp_async(vmf->vma))
>>>>> +            if (!userfaultfd_wp_async(vmf->vma))
>>>>> +                return handle_userfault(vmf, VM_UFFD_WP);
>>>>> +            if (!is_huge_zero_pmd(vmf->orig_pmd))
>>>>>                    goto split;
>>>>> -            return handle_userfault(vmf, VM_UFFD_WP);
>>>>>            }
>>>>>            return do_huge_pmd_wp_page(vmf);
>>>>>        }