Re: [PATCH -v2 1/2] mm: add spurious fault fixing support for huge pmd

["Huang, Ying" <ying.huang@linux.alibaba.com>]

Lorenzo Stoakes <lorenzo.stoakes@oracle.com> writes:

> On Wed, Oct 15, 2025 at 04:43:14PM +0800, Huang, Ying wrote:
>> Hi, Lorenzo,
>>
>> Thanks for comments!
>>
>> Lorenzo Stoakes <lorenzo.stoakes@oracle.com> writes:
>>
>> > On Mon, Oct 13, 2025 at 05:20:37PM +0800, Huang Ying wrote:
>> >> In the current kernel, there is spurious fault fixing support for pte,
>> >> but not for huge pmd because no architectures need it. But in the
>> >> next patch in the series, we will change the write protection fault
>> >> handling logic on arm64, so that some stale huge pmd entries may
>> >> remain in the TLB. These entries need to be flushed via the huge pmd
>> >> spurious fault fixing mechanism.
>> >>
>> >> Signed-off-by: Huang Ying <ying.huang@linux.alibaba.com>
>> >
>> > Right now the PTE level spurious fault handling is dealt with in
>> > handle_pte_fault() when ptep_set_access_flags() returns false.
>> >
>> > Now you're updating touch_pmd() which is invoked by follow_huge_pmd() and
>> > huge_pmd_set_accessed().
>> >
>> > 1 - Why are you not adding handling to GUP?
>> >
>> > 2 - Is this the correct level of abstraction? It's really not obvious but
>> >     huge_pmd_set_accessed() is invoked by __handle_mm_fault() on a non-WP,
>> >     non-NUMA hint huge page fault where a page table entry already exists
>> >     but we are faulting anyway (e.g. non-present or read-only writable).
>> >
>> > You don't mention any of this in the commit message, which you need to do
>> > and really need to explain how spurious faults can arise, why you can only
>> > do this at the point of abstraction you do (if you are unable to put it in
>> > actual fault handing-code), and you need to add a bunch more comments to
>> > explain this.
>>
>> This patch adds the spurious PMD page fault fixing based on the spurious
>> PTE page fault fixing.  So, I assumed that the spurious page fault
>> fixing has been documented already.  But you are right, nothing prevents
>> us from improving it further.  Let's try to do that.
>
> I wouldn't make any kind of assumption like this in the kernel :) sadly our
> documentation is often incomplete.
>
>>
>> The page faults may be spurious because of the racy access to the page
>> table.  For example, a non-populated virtual page is accessed on 2 CPUs
>> simultaneously, thus the page faults are triggered on both CPUs.
>> However, it's possible that one CPU (say CPU A) cannot find the reason
>> for the page fault if the other CPU (say CPU B) has changed the page
>> table before the PTE is checked on CPU A.  Most of the time, the
>> spurious page faults can be ignored safely.  However, if the page fault
>> is for the write access, it's possible that a stale read-only TLB entry
>> exists in the local CPU and needs to be flushed on some architectures.
>> This is called the spurious page fault fixing.
>>
>> The spurious page fault fixing only makes sense during page fault
>> handling, so we don't need to do it for GUP.  In fact, I plan to avoid
>> it in all GUP paths in another followup patch.
>
> OK this is great, let's put it all in the kdoc for the new shared spurious
> faulting function! :) and additionally add it to the commit message.

Sure.  Will do it in the next version.

[snip]

>> >>
>> >> diff --git a/include/linux/pgtable.h b/include/linux/pgtable.h
>> >> index 32e8457ad535..341622ec80e4 100644
>> >> --- a/include/linux/pgtable.h
>> >> +++ b/include/linux/pgtable.h
>> >> @@ -1232,6 +1232,10 @@ static inline void arch_swap_restore(swp_entry_t entry, struct folio *folio)
>> >>  #define flush_tlb_fix_spurious_fault(vma, address, ptep) flush_tlb_page(vma, address)
>> >>  #endif
>> >>
>> >> +#ifndef flush_tlb_fix_spurious_fault_pmd
>> >> +#define flush_tlb_fix_spurious_fault_pmd(vma, address, ptep) do { } while (0)
>> >> +#endif
>> >
>> > flush_tlb_fix_spurious_fault(), when the arch doesn't declare it, defaults to
>> > flush_tlb_page() - why do we just do nothing in this case here?
>>
>> Because all architectures do nothing for the spurious PMD page fault
>> fixing until the [2/2] of this series.  Where, we make it necessary to
>> flush the local TLB for spurious PMD page fault fixing on arm64
>> architecture.
>>
>> If we follow the design of flush_tlb_fix_spurious_fault(), we need to
>> change all architecture implementation to do nothing in this patch to
>> keep the current behavior.  I don't think that it's a good idea.  Do
>> you agree?
>
> Yeah probably we should keep the same behaviour as before, which is
> obviously, prior to this series, we did nothing.
>
> I guess in the PTE case we _always_ want to flush the TLB, whereas in the
> PMD case we otherwise don't have any need to at the point at which the
> spurious flush is performed.
>
> But from your explanation above re: the stale TLB entry this _only_ needs
> to be done for architectures which might encounter this problem rather than
> needing a TLB flush in general.
>
> Given we're generalising the code and one case always flushes the TLB and
> the other doesn't maybe it's worth putting a comment in the generalised
> function mentioning this?

I'm not sure whether it's a good idea to document architecture behaviors
in the general code.  The behavior may be changed architecture by
architecture in the future.

[snip]

---
Best Regards,
Huang, Ying