On Tue, Feb 27, 2024 at 10:14 PM David Hildenbrand <david@xxxxxxxxxx> wrote:
>
> On 27.02.24 10:07, Ryan Roberts wrote:
> > On 27/02/2024 02:40, Barry Song wrote:
> >> From: Barry Song <v-songbaohua@xxxxxxxx>
> >>
> >> madvise and some others might need folio_pte_batch to check if a range
> >> of PTEs are completely mapped to a large folio with contiguous physcial
> >> addresses. Let's export it for others to use.
> >>
> >> Cc: Lance Yang <ioworker0@xxxxxxxxx>
> >> Cc: Ryan Roberts <ryan.roberts@xxxxxxx>
> >> Cc: David Hildenbrand <david@xxxxxxxxxx>
> >> Cc: Yin Fengwei <fengwei.yin@xxxxxxxxx>
> >> Signed-off-by: Barry Song <v-songbaohua@xxxxxxxx>
> >> ---
> >> -v1:
> >> at least two jobs madv_free and madv_pageout depend on it. To avoid
> >> conflicts and dependencies, after discussing with Lance, we prefer
> >> this one can land earlier.
> >
> > I think this will also ultimately be useful for mprotect too, though I haven't
> > looked at it properly yet.
> >
>
> Yes, I think we briefly discussed that.
>
> >>
> >> mm/internal.h | 13 +++++++++++++
> >> mm/memory.c | 11 +----------
> >> 2 files changed, 14 insertions(+), 10 deletions(-)
> >>
> >> diff --git a/mm/internal.h b/mm/internal.h
> >> index 13b59d384845..8e2bc304f671 100644
> >> --- a/mm/internal.h
> >> +++ b/mm/internal.h
> >> @@ -83,6 +83,19 @@ static inline void *folio_raw_mapping(struct folio *folio)
> >> return (void *)(mapping & ~PAGE_MAPPING_FLAGS);
> >> }
> >>
> >> +/* Flags for folio_pte_batch(). */
> >> +typedef int __bitwise fpb_t;
> >> +
> >> +/* Compare PTEs after pte_mkclean(), ignoring the dirty bit. */
> >> +#define FPB_IGNORE_DIRTY ((__force fpb_t)BIT(0))
> >> +
> >> +/* Compare PTEs after pte_clear_soft_dirty(), ignoring the soft-dirty bit. */
> >> +#define FPB_IGNORE_SOFT_DIRTY ((__force fpb_t)BIT(1))
> >> +
> >> +extern int folio_pte_batch(struct folio *folio, unsigned long addr,
> >> + pte_t *start_ptep, pte_t pte, int max_nr, fpb_t flags,
> >> + bool *any_writable);
> >> +
> >> void __acct_reclaim_writeback(pg_data_t *pgdat, struct folio *folio,
> >> int nr_throttled);
> >> static inline void acct_reclaim_writeback(struct folio *folio)
> >> diff --git a/mm/memory.c b/mm/memory.c
> >> index 1c45b6a42a1b..319b3be05e75 100644
> >> --- a/mm/memory.c
> >> +++ b/mm/memory.c
> >> @@ -953,15 +953,6 @@ static __always_inline void __copy_present_ptes(struct vm_area_struct *dst_vma,
> >> set_ptes(dst_vma->vm_mm, addr, dst_pte, pte, nr);
> >> }
> >>
> >> -/* Flags for folio_pte_batch(). */
> >> -typedef int __bitwise fpb_t;
> >> -
> >> -/* Compare PTEs after pte_mkclean(), ignoring the dirty bit. */
> >> -#define FPB_IGNORE_DIRTY ((__force fpb_t)BIT(0))
> >> -
> >> -/* Compare PTEs after pte_clear_soft_dirty(), ignoring the soft-dirty bit. */
> >> -#define FPB_IGNORE_SOFT_DIRTY ((__force fpb_t)BIT(1))
> >> -
> >> static inline pte_t __pte_batch_clear_ignored(pte_t pte, fpb_t flags)
> >> {
> >> if (flags & FPB_IGNORE_DIRTY)
> >> @@ -982,7 +973,7 @@ static inline pte_t __pte_batch_clear_ignored(pte_t pte, fpb_t flags)
> >> * If "any_writable" is set, it will indicate if any other PTE besides the
> >> * first (given) PTE is writable.
> >> */
> >
> > David was talking in Lance's patch thread, about improving the docs for this
> > function now that its exported. Might be worth syncing on that.
>
> Here is my take:
>
> Signed-off-by: David Hildenbrand <david@xxxxxxxxxx>
> ---
> mm/memory.c | 22 ++++++++++++++++++----
> 1 file changed, 18 insertions(+), 4 deletions(-)
>
> diff --git a/mm/memory.c b/mm/memory.c
> index d0b855a1837a8..098356b8805ae 100644
> --- a/mm/memory.c
> +++ b/mm/memory.c
> @@ -971,16 +971,28 @@ static inline pte_t __pte_batch_clear_ignored(pte_t pte, fpb_t flags)
> return pte_wrprotect(pte_mkold(pte));
> }
>
> -/*
> +/**
> + * folio_pte_batch - detect a PTE batch for a large folio
> + * @folio: The large folio to detect a PTE batch for.
> + * @addr: The user virtual address the first page is mapped at.
> + * @start_ptep: Page table pointer for the first entry.
> + * @pte: Page table entry for the first page.
> + * @max_nr: The maximum number of table entries to consider.
> + * @flags: Flags to modify the PTE batch semantics.
> + * @any_writable: Optional pointer to indicate whether any entry except the
> + * first one is writable.
> + *
> * Detect a PTE batch: consecutive (present) PTEs that map consecutive
> - * pages of the same folio.
> + * pages of the same large folio.
> *
> * All PTEs inside a PTE batch have the same PTE bits set, excluding the PFN,
> * the accessed bit, writable bit, dirty bit (with FPB_IGNORE_DIRTY) and
> * soft-dirty bit (with FPB_IGNORE_SOFT_DIRTY).
> *
> - * If "any_writable" is set, it will indicate if any other PTE besides the
> - * first (given) PTE is writable.
> + * start_ptep must map any page of the folio. max_nr must be at least one and
> + * must be limited by the caller so scanning cannot exceed a single page table.
> + *
> + * Return: the number of table entries in the batch.
> */
> static inline int folio_pte_batch(struct folio *folio, unsigned long addr,
> pte_t *start_ptep, pte_t pte, int max_nr, fpb_t flags,
> @@ -996,6 +1008,8 @@ static inline int folio_pte_batch(struct folio *folio, unsigned long addr,
> *any_writable = false;
>
> VM_WARN_ON_FOLIO(!pte_present(pte), folio);
> + VM_WARN_ON_FOLIO(!folio_test_large(folio) || max_nr < 1, folio);
> + VM_WARN_ON_FOLIO(page_folio(pfn_to_page(pte_pfn(pte))) != folio, folio);
>
> nr = pte_batch_hint(start_ptep, pte);
> expected_pte = __pte_batch_clear_ignored(pte_advance_pfn(pte, nr), flags);
> --
> 2.43.2
>
>
> >
> >> -static inline int folio_pte_batch(struct folio *folio, unsigned long addr,
> >> +int folio_pte_batch(struct folio *folio, unsigned long addr,
> >
> > fork() is very performance sensitive. Is there a risk we are regressing
> > performance by making this out-of-line? Although its in the same compilation
> > unit so the compiler may well inline it anyway?
>
> Easy to verify by looking at the generated asm I guess?
my aarch64-linux-gnu-gcc didn't inline it
$ aarch64-linux-gnu-gcc --version
aarch64-linux-gnu-gcc (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0
Copyright (C) 2021 Free Software Foundation, Inc.
$ nm -S -s vmlinux.a | grep folio_pte_batch
0000000000003818 0000000000000204 T folio_pte_batch
>
> >
> > Either way, perhaps we are better off making it inline in the header? That would
> > avoid needing to rerun David's micro-benchmarks for fork() and munmap().
actually tried this before trying extern, the problem is that we have to add
others into internal.h, for example __pte_batch_clear_ignored, which
seems not API. are we comfortable to move that one to internal.h too?
>
> That way, the compiler can most certainly better optimize it also outside of mm/memory.c
>
> --
> Cheers,
>
> David / dhildenb
Thanks
Barry
