[RESEND,v2,1/4] mm: hugetlb: free the 2nd vmemmap page associated with each HugeTLB page

Commit Message

Muchun Song Sept. 17, 2021, 3:48 a.m. UTC

Currently, we only free 6 vmemmap pages associated with a 2MB HugeTLB
page. However, we can remap all tail vmemmap pages to the page frame
mapped to with the head vmemmap page. Finally, we can free 7 vmemmap
pages for a 2MB HugeTLB page. It is a fine gain (e.g. we can save
extra 2GB memory when there is 1TB HugeTLB pages in the system
compared with the current implementation).

But the head vmemmap page is not freed to the buddy allocator and all
tail vmemmap pages are mapped to the head vmemmap page frame. So we
can see more than one struct page struct with PG_head (e.g. 8 per 2 MB
HugeTLB page) associated with each HugeTLB page. We should adjust
compound_head() to make it returns the real head struct page when the
parameter is the tail struct page but with PG_head flag.

Signed-off-by: Muchun Song <[e-mail]>
---
 Documentation/admin-guide/kernel-parameters.txt |  2 +-
 include/linux/page-flags.h                      | 75 +++++++++++++++++++++++--
 mm/hugetlb_vmemmap.c                            | 60 +++++++++++---------
 mm/sparse-vmemmap.c                             | 21 +++++++
 4 files changed, 126 insertions(+), 32 deletions(-)

Comments

Barry Song Sept. 18, 2021, 4:38 a.m. UTC | #1

On Sat, Sep 18, 2021 at 12:08 AM Muchun Song <[e-mail]> wrote:
>
> Currently, we only free 6 vmemmap pages associated with a 2MB HugeTLB
> page. However, we can remap all tail vmemmap pages to the page frame
> mapped to with the head vmemmap page. Finally, we can free 7 vmemmap
> pages for a 2MB HugeTLB page. It is a fine gain (e.g. we can save
> extra 2GB memory when there is 1TB HugeTLB pages in the system
> compared with the current implementation).
>
> But the head vmemmap page is not freed to the buddy allocator and all
> tail vmemmap pages are mapped to the head vmemmap page frame. So we
> can see more than one struct page struct with PG_head (e.g. 8 per 2 MB
> HugeTLB page) associated with each HugeTLB page. We should adjust
> compound_head() to make it returns the real head struct page when the
> parameter is the tail struct page but with PG_head flag.
>
> Signed-off-by: Muchun Song <[e-mail]>
> ---
>  Documentation/admin-guide/kernel-parameters.txt |  2 +-
>  include/linux/page-flags.h                      | 75 +++++++++++++++++++++++--
>  mm/hugetlb_vmemmap.c                            | 60 +++++++++++---------
>  mm/sparse-vmemmap.c                             | 21 +++++++
>  4 files changed, 126 insertions(+), 32 deletions(-)
>
> diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
> index bdb22006f713..a154a7b3b9a5 100644
> --- a/Documentation/admin-guide/kernel-parameters.txt
> +++ b/Documentation/admin-guide/kernel-parameters.txt
> @@ -1606,7 +1606,7 @@
>                         [KNL] Reguires CONFIG_HUGETLB_PAGE_FREE_VMEMMAP
>                         enabled.
>                         Allows heavy hugetlb users to free up some more
> -                       memory (6 * PAGE_SIZE for each 2MB hugetlb page).
> +                       memory (7 * PAGE_SIZE for each 2MB hugetlb page).
>                         Format: { on | off (default) }
>
>                         on:  enable the feature
> diff --git a/include/linux/page-flags.h b/include/linux/page-flags.h
> index 8e1d97d8f3bd..7b1a918ebd43 100644
> --- a/include/linux/page-flags.h
> +++ b/include/linux/page-flags.h
> @@ -184,13 +184,64 @@ enum pageflags {
>
>  #ifndef __GENERATING_BOUNDS_H
>
> +#ifdef CONFIG_HUGETLB_PAGE_FREE_VMEMMAP
> +extern bool hugetlb_free_vmemmap_enabled;
> +
> +/*
> + * If the feature of freeing some vmemmap pages associated with each HugeTLB
> + * page is enabled, the head vmemmap page frame is reused and all of the tail
> + * vmemmap addresses map to the head vmemmap page frame (furture details can
> + * refer to the figure at the head of the mm/hugetlb_vmemmap.c).  In other
> + * word, there are more than one page struct with PG_head associated with each
> + * HugeTLB page.  We __know__ that there is only one head page struct, the tail
> + * page structs with PG_head are fake head page structs.  We need an approach
> + * to distinguish between those two different types of page structs so that
> + * compound_head() can return the real head page struct when the parameter is
> + * the tail page struct but with PG_head.
> + *
> + * The page_head_if_fake() returns the real head page struct iff the @page may
> + * be fake, otherwise, returns the @page if it cannot be a fake page struct.
> + */
> +static __always_inline const struct page *page_head_if_fake(const struct page *page)
> +{
> +       if (!hugetlb_free_vmemmap_enabled)
> +               return page;
> +
> +       /*
> +        * Only addresses aligned with PAGE_SIZE of struct page may be fake head
> +        * struct page. The alignment check aims to avoid access the fields (
> +        * e.g. compound_head) of the @page[1]. It can avoid touch a (possibly)
> +        * cold cacheline in some cases.
> +        */
> +       if (IS_ALIGNED((unsigned long)page, PAGE_SIZE) &&
> +           test_bit(PG_head, &page->flags)) {
> +               /*
> +                * We can safely access the field of the @page[1] with PG_head
> +                * because the @page is a compound page composed with at least
> +                * two contiguous pages.
> +                */
> +               unsigned long head = READ_ONCE(page[1].compound_head);
> +
> +               if (likely(head & 1))
> +                       return (const struct page *)(head - 1);
> +       }
> +
> +       return page;
> +}
> +#else
> +static __always_inline const struct page *page_head_if_fake(const struct page *page)
> +{
> +       return page;
> +}
> +#endif
> +
>  static inline unsigned long _compound_head(const struct page *page)
>  {
>         unsigned long head = READ_ONCE(page->compound_head);
>
>         if (unlikely(head & 1))
>                 return head - 1;
> -       return (unsigned long)page;
> +       return (unsigned long)page_head_if_fake(page);

hard to read. page_head_if_fake,  what is the other side of
page_head_if_not_fake?
I would expect something like
page_to_page_head()
or
get_page_head()

Anyway, I am not quite sure what is the best name. but page_head_if_fake(page)
sounds odd to me. just like the things have two sides, but if_fake  presents
one side only.

>  }
>
>  #define compound_head(page)    ((typeof(page))_compound_head(page))
> @@ -225,12 +276,14 @@ static inline unsigned long _compound_head(const struct page *page)
>
>  static __always_inline int PageTail(struct page *page)
>  {
> -       return READ_ONCE(page->compound_head) & 1;
> +       return READ_ONCE(page->compound_head) & 1 ||
> +              page_head_if_fake(page) != page;

i would expect a wrapper like:
page_is_fake_head()

and the above page_to_page_head() can leverage the wrapper.
here too.

>  }
>
>  static __always_inline int PageCompound(struct page *page)
>  {
> -       return test_bit(PG_head, &page->flags) || PageTail(page);
> +       return test_bit(PG_head, &page->flags) ||
> +              READ_ONCE(page->compound_head) & 1;

hard to read. could it be something like the below?
return PageHead(page) || PageTail(page);

or do we really need to change this function? even a fake head still has
the true test_bit(PG_head, &page->flags), though it is not a real head, it
is still a pagecompound, right?


>  }
>
>  #define        PAGE_POISON_PATTERN     -1l
> @@ -675,7 +728,21 @@ static inline bool test_set_page_writeback(struct page *page)
>         return set_page_writeback(page);
>  }
>
> -__PAGEFLAG(Head, head, PF_ANY) CLEARPAGEFLAG(Head, head, PF_ANY)
> +static __always_inline bool folio_test_head(struct folio *folio)
> +{
> +       return test_bit(PG_head, folio_flags(folio, FOLIO_PF_ANY));
> +}
> +
> +static __always_inline int PageHead(struct page *page)
> +{
> +       PF_POISONED_CHECK(page);
> +       return test_bit(PG_head, &page->flags) &&
> +              page_head_if_fake(page) == page;
> +}
> +
> +__SETPAGEFLAG(Head, head, PF_ANY)
> +__CLEARPAGEFLAG(Head, head, PF_ANY)
> +CLEARPAGEFLAG(Head, head, PF_ANY)
>
>  /* Whether there are one or multiple pages in a folio */
>  static inline bool folio_single(struct folio *folio)
> diff --git a/mm/hugetlb_vmemmap.c b/mm/hugetlb_vmemmap.c
> index c540c21e26f5..527bcaa44a48 100644
> --- a/mm/hugetlb_vmemmap.c
> +++ b/mm/hugetlb_vmemmap.c
> @@ -124,9 +124,9 @@
>   * page of page structs (page 0) associated with the HugeTLB page contains the 4
>   * page structs necessary to describe the HugeTLB. The only use of the remaining
>   * pages of page structs (page 1 to page 7) is to point to page->compound_head.
> - * Therefore, we can remap pages 2 to 7 to page 1. Only 2 pages of page structs
> + * Therefore, we can remap pages 1 to 7 to page 0. Only 1 pages of page structs
>   * will be used for each HugeTLB page. This will allow us to free the remaining
> - * 6 pages to the buddy allocator.
> + * 7 pages to the buddy allocator.
>   *
>   * Here is how things look after remapping.
>   *
> @@ -134,30 +134,30 @@
>   * +-----------+ ---virt_to_page---> +-----------+   mapping to   +-----------+
>   * |           |                     |     0     | -------------> |     0     |
>   * |           |                     +-----------+                +-----------+
> - * |           |                     |     1     | -------------> |     1     |
> - * |           |                     +-----------+                +-----------+
> - * |           |                     |     2     | ----------------^ ^ ^ ^ ^ ^
> - * |           |                     +-----------+                   | | | | |
> - * |           |                     |     3     | ------------------+ | | | |
> - * |           |                     +-----------+                     | | | |
> - * |           |                     |     4     | --------------------+ | | |
> - * |    PMD    |                     +-----------+                       | | |
> - * |   level   |                     |     5     | ----------------------+ | |
> - * |  mapping  |                     +-----------+                         | |
> - * |           |                     |     6     | ------------------------+ |
> - * |           |                     +-----------+                           |
> - * |           |                     |     7     | --------------------------+
> + * |           |                     |     1     | ---------------^ ^ ^ ^ ^ ^ ^
> + * |           |                     +-----------+                  | | | | | |
> + * |           |                     |     2     | -----------------+ | | | | |
> + * |           |                     +-----------+                    | | | | |
> + * |           |                     |     3     | -------------------+ | | | |
> + * |           |                     +-----------+                      | | | |
> + * |           |                     |     4     | ---------------------+ | | |
> + * |    PMD    |                     +-----------+                        | | |
> + * |   level   |                     |     5     | -----------------------+ | |
> + * |  mapping  |                     +-----------+                          | |
> + * |           |                     |     6     | -------------------------+ |
> + * |           |                     +-----------+                            |
> + * |           |                     |     7     | ---------------------------+
>   * |           |                     +-----------+
>   * |           |
>   * |           |
>   * |           |
>   * +-----------+
>   *
> - * When a HugeTLB is freed to the buddy system, we should allocate 6 pages for
> + * When a HugeTLB is freed to the buddy system, we should allocate 7 pages for
>   * vmemmap pages and restore the previous mapping relationship.
>   *
>   * For the HugeTLB page of the pud level mapping. It is similar to the former.
> - * We also can use this approach to free (PAGE_SIZE - 2) vmemmap pages.
> + * We also can use this approach to free (PAGE_SIZE - 1) vmemmap pages.
>   *
>   * Apart from the HugeTLB page of the pmd/pud level mapping, some architectures
>   * (e.g. aarch64) provides a contiguous bit in the translation table entries
> @@ -166,7 +166,13 @@
>   *
>   * The contiguous bit is used to increase the mapping size at the pmd and pte
>   * (last) level. So this type of HugeTLB page can be optimized only when its
> - * size of the struct page structs is greater than 2 pages.
> + * size of the struct page structs is greater than 1 pages.
> + *
> + * Notice: The head vmemmap page is not freed to the buddy allocator and all
> + * tail vmemmap pages are mapped to the head vmemmap page frame. So we can see
> + * more than one struct page struct with PG_head (e.g. 8 per 2 MB HugeTLB page)
> + * associated with each HugeTLB page. The compound_head() can handle this
> + * correctly (more details refer to the comment above compound_head()).
>   */
>  #define pr_fmt(fmt)    "HugeTLB: " fmt
>
> @@ -175,14 +181,16 @@
>  /*
>   * There are a lot of struct page structures associated with each HugeTLB page.
>   * For tail pages, the value of compound_head is the same. So we can reuse first
> - * page of tail page structures. We map the virtual addresses of the remaining
> - * pages of tail page structures to the first tail page struct, and then free
> - * these page frames. Therefore, we need to reserve two pages as vmemmap areas.
> + * page of head page structures. We map the virtual addresses of all the pages
> + * of tail page structures to the head page struct, and then free these page
> + * frames. Therefore, we need to reserve one pages as vmemmap areas.
>   */
> -#define RESERVE_VMEMMAP_NR             2U
> +#define RESERVE_VMEMMAP_NR             1U
>  #define RESERVE_VMEMMAP_SIZE           (RESERVE_VMEMMAP_NR << PAGE_SHIFT)
>
> -bool hugetlb_free_vmemmap_enabled = IS_ENABLED(CONFIG_HUGETLB_PAGE_FREE_VMEMMAP_DEFAULT_ON);
> +bool hugetlb_free_vmemmap_enabled __read_mostly =
> +       IS_ENABLED(CONFIG_HUGETLB_PAGE_FREE_VMEMMAP_DEFAULT_ON);
> +EXPORT_SYMBOL(hugetlb_free_vmemmap_enabled);
>
>  static int __init early_hugetlb_free_vmemmap_param(char *buf)
>  {
> @@ -236,7 +244,6 @@ int alloc_huge_page_vmemmap(struct hstate *h, struct page *head)
>          */
>         ret = vmemmap_remap_alloc(vmemmap_addr, vmemmap_end, vmemmap_reuse,
>                                   GFP_KERNEL | __GFP_NORETRY | __GFP_THISNODE);
> -
>         if (!ret)
>                 ClearHPageVmemmapOptimized(head);
>
> @@ -282,9 +289,8 @@ void __init hugetlb_vmemmap_init(struct hstate *h)
>
>         vmemmap_pages = (nr_pages * sizeof(struct page)) >> PAGE_SHIFT;
>         /*
> -        * The head page and the first tail page are not to be freed to buddy
> -        * allocator, the other pages will map to the first tail page, so they
> -        * can be freed.
> +        * The head page is not to be freed to buddy allocator, the other tail
> +        * pages will map to the head page, so they can be freed.
>          *
>          * Could RESERVE_VMEMMAP_NR be greater than @vmemmap_pages? It is true
>          * on some architectures (e.g. aarch64). See Documentation/arm64/
> diff --git a/mm/sparse-vmemmap.c b/mm/sparse-vmemmap.c
> index bdce883f9286..62e3d20648ce 100644
> --- a/mm/sparse-vmemmap.c
> +++ b/mm/sparse-vmemmap.c
> @@ -53,6 +53,17 @@ struct vmemmap_remap_walk {
>         struct list_head *vmemmap_pages;
>  };
>
> +/*
> + * How many struct page structs need to be reset. When we reuse the head
> + * struct page, the special metadata (e.g. page->flags or page->mapping)
> + * cannot copy to the tail struct page structs. The invalid value will be
> + * checked in the free_tail_pages_check(). In order to avoid the message
> + * of "corrupted mapping in tail page". We need to reset at least 3 (one
> + * head struct page struct and two tail struct page structs) struct page
> + * structs.
> + */
> +#define NR_RESET_STRUCT_PAGE           3
> +
>  static int split_vmemmap_huge_pmd(pmd_t *pmd, unsigned long start,
>                                   struct vmemmap_remap_walk *walk)
>  {
> @@ -245,6 +256,15 @@ static void vmemmap_remap_pte(pte_t *pte, unsigned long addr,
>         set_pte_at(&init_mm, addr, pte, entry);
>  }
>
> +static inline void reset_struct_pages(struct page *start)
> +{
> +       int i;
> +       struct page *from = start + NR_RESET_STRUCT_PAGE;
> +
> +       for (i = 0; i < NR_RESET_STRUCT_PAGE; i++)
> +               memcpy(start + i, from, sizeof(*from));
> +}
> +
>  static void vmemmap_restore_pte(pte_t *pte, unsigned long addr,
>                                 struct vmemmap_remap_walk *walk)
>  {
> @@ -258,6 +278,7 @@ static void vmemmap_restore_pte(pte_t *pte, unsigned long addr,
>         list_del(&page->lru);
>         to = page_to_virt(page);
>         copy_page(to, (void *)walk->reuse_addr);
> +       reset_struct_pages(to);
>
>         set_pte_at(&init_mm, addr, pte, mk_pte(page, pgprot));
>  }
> --
> 2.11.0
>

Thanks
barry

Muchun Song Sept. 18, 2021, 10:06 a.m. UTC | #2

On Sat, Sep 18, 2021 at 12:39 PM Barry Song <[e-mail]> wrote:
>
> On Sat, Sep 18, 2021 at 12:08 AM Muchun Song <[e-mail]> wrote:
> >
> > Currently, we only free 6 vmemmap pages associated with a 2MB HugeTLB
> > page. However, we can remap all tail vmemmap pages to the page frame
> > mapped to with the head vmemmap page. Finally, we can free 7 vmemmap
> > pages for a 2MB HugeTLB page. It is a fine gain (e.g. we can save
> > extra 2GB memory when there is 1TB HugeTLB pages in the system
> > compared with the current implementation).
> >
> > But the head vmemmap page is not freed to the buddy allocator and all
> > tail vmemmap pages are mapped to the head vmemmap page frame. So we
> > can see more than one struct page struct with PG_head (e.g. 8 per 2 MB
> > HugeTLB page) associated with each HugeTLB page. We should adjust
> > compound_head() to make it returns the real head struct page when the
> > parameter is the tail struct page but with PG_head flag.
> >
> > Signed-off-by: Muchun Song <[e-mail]>
> > ---
> >  Documentation/admin-guide/kernel-parameters.txt |  2 +-
> >  include/linux/page-flags.h                      | 75 +++++++++++++++++++++++--
> >  mm/hugetlb_vmemmap.c                            | 60 +++++++++++---------
> >  mm/sparse-vmemmap.c                             | 21 +++++++
> >  4 files changed, 126 insertions(+), 32 deletions(-)
> >
> > diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
> > index bdb22006f713..a154a7b3b9a5 100644
> > --- a/Documentation/admin-guide/kernel-parameters.txt
> > +++ b/Documentation/admin-guide/kernel-parameters.txt
> > @@ -1606,7 +1606,7 @@
> >                         [KNL] Reguires CONFIG_HUGETLB_PAGE_FREE_VMEMMAP
> >                         enabled.
> >                         Allows heavy hugetlb users to free up some more
> > -                       memory (6 * PAGE_SIZE for each 2MB hugetlb page).
> > +                       memory (7 * PAGE_SIZE for each 2MB hugetlb page).
> >                         Format: { on | off (default) }
> >
> >                         on:  enable the feature
> > diff --git a/include/linux/page-flags.h b/include/linux/page-flags.h
> > index 8e1d97d8f3bd..7b1a918ebd43 100644
> > --- a/include/linux/page-flags.h
> > +++ b/include/linux/page-flags.h
> > @@ -184,13 +184,64 @@ enum pageflags {
> >
> >  #ifndef __GENERATING_BOUNDS_H
> >
> > +#ifdef CONFIG_HUGETLB_PAGE_FREE_VMEMMAP
> > +extern bool hugetlb_free_vmemmap_enabled;
> > +
> > +/*
> > + * If the feature of freeing some vmemmap pages associated with each HugeTLB
> > + * page is enabled, the head vmemmap page frame is reused and all of the tail
> > + * vmemmap addresses map to the head vmemmap page frame (furture details can
> > + * refer to the figure at the head of the mm/hugetlb_vmemmap.c).  In other
> > + * word, there are more than one page struct with PG_head associated with each
> > + * HugeTLB page.  We __know__ that there is only one head page struct, the tail
> > + * page structs with PG_head are fake head page structs.  We need an approach
> > + * to distinguish between those two different types of page structs so that
> > + * compound_head() can return the real head page struct when the parameter is
> > + * the tail page struct but with PG_head.
> > + *
> > + * The page_head_if_fake() returns the real head page struct iff the @page may
> > + * be fake, otherwise, returns the @page if it cannot be a fake page struct.
> > + */
> > +static __always_inline const struct page *page_head_if_fake(const struct page *page)
> > +{
> > +       if (!hugetlb_free_vmemmap_enabled)
> > +               return page;
> > +
> > +       /*
> > +        * Only addresses aligned with PAGE_SIZE of struct page may be fake head
> > +        * struct page. The alignment check aims to avoid access the fields (
> > +        * e.g. compound_head) of the @page[1]. It can avoid touch a (possibly)
> > +        * cold cacheline in some cases.
> > +        */
> > +       if (IS_ALIGNED((unsigned long)page, PAGE_SIZE) &&
> > +           test_bit(PG_head, &page->flags)) {
> > +               /*
> > +                * We can safely access the field of the @page[1] with PG_head
> > +                * because the @page is a compound page composed with at least
> > +                * two contiguous pages.
> > +                */
> > +               unsigned long head = READ_ONCE(page[1].compound_head);
> > +
> > +               if (likely(head & 1))
> > +                       return (const struct page *)(head - 1);
> > +       }
> > +
> > +       return page;
> > +}
> > +#else
> > +static __always_inline const struct page *page_head_if_fake(const struct page *page)
> > +{
> > +       return page;
> > +}
> > +#endif
> > +
> >  static inline unsigned long _compound_head(const struct page *page)
> >  {
> >         unsigned long head = READ_ONCE(page->compound_head);
> >
> >         if (unlikely(head & 1))
> >                 return head - 1;
> > -       return (unsigned long)page;
> > +       return (unsigned long)page_head_if_fake(page);
>
> hard to read. page_head_if_fake,  what is the other side of
> page_head_if_not_fake?

1) return itself if the @page is not a fake head page.
2) return head page if @page is a fake head page.

So I want to express that page_head_if_fake returns a
head page only and only if the parameter of @page is a
fake head page. Otherwise, it returns itself.

> I would expect something like
> page_to_page_head()
> or
> get_page_head()
>

Those names seem to be not appropriate as well, because
its functionality does not make sure it can return a head
page. If the parameter is a head page, it definitely
returns a head page, otherwise, it may return itself which
may be a tail page.

From this point of view, I still prefer page_head_if_fake.

> Anyway, I am not quite sure what is the best name. but page_head_if_fake(page)
> sounds odd to me. just like the things have two sides, but if_fake  presents
> one side only.

If others have any ideas, comments are welcome.

>
> >  }
> >
> >  #define compound_head(page)    ((typeof(page))_compound_head(page))
> > @@ -225,12 +276,14 @@ static inline unsigned long _compound_head(const struct page *page)
> >
> >  static __always_inline int PageTail(struct page *page)
> >  {
> > -       return READ_ONCE(page->compound_head) & 1;
> > +       return READ_ONCE(page->compound_head) & 1 ||
> > +              page_head_if_fake(page) != page;
>
> i would expect a wrapper like:
> page_is_fake_head()

Good point. Will do.

>
> and the above page_to_page_head() can leverage the wrapper.
> here too.
>
> >  }
> >
> >  static __always_inline int PageCompound(struct page *page)
> >  {
> > -       return test_bit(PG_head, &page->flags) || PageTail(page);
> > +       return test_bit(PG_head, &page->flags) ||
> > +              READ_ONCE(page->compound_head) & 1;
>
> hard to read. could it be something like the below?
> return PageHead(page) || PageTail(page);
>
> or do we really need to change this function? even a fake head still has
> the true test_bit(PG_head, &page->flags), though it is not a real head, it
> is still a pagecompound, right?

Right. PageCompound() can not be changed.  It is odd but
efficient because calling page_head_if_fake is eliminated.
So I select performance not readability. I'm not sure if it's
worth it.

Thanks.

Barry Song Sept. 21, 2021, 12:11 a.m. UTC | #3

On Tue, Sep 21, 2021 at 10:23 PM Muchun Song <[e-mail]> wrote:
>
> On Sat, Sep 18, 2021 at 6:06 PM Muchun Song <[e-mail]> wrote:
> >
> > On Sat, Sep 18, 2021 at 12:39 PM Barry Song <[e-mail]> wrote:
> > >
> > > On Sat, Sep 18, 2021 at 12:08 AM Muchun Song <[e-mail]> wrote:
> > > >
> > > > Currently, we only free 6 vmemmap pages associated with a 2MB HugeTLB
> > > > page. However, we can remap all tail vmemmap pages to the page frame
> > > > mapped to with the head vmemmap page. Finally, we can free 7 vmemmap
> > > > pages for a 2MB HugeTLB page. It is a fine gain (e.g. we can save
> > > > extra 2GB memory when there is 1TB HugeTLB pages in the system
> > > > compared with the current implementation).
> > > >
> > > > But the head vmemmap page is not freed to the buddy allocator and all
> > > > tail vmemmap pages are mapped to the head vmemmap page frame. So we
> > > > can see more than one struct page struct with PG_head (e.g. 8 per 2 MB
> > > > HugeTLB page) associated with each HugeTLB page. We should adjust
> > > > compound_head() to make it returns the real head struct page when the
> > > > parameter is the tail struct page but with PG_head flag.
> > > >
> > > > Signed-off-by: Muchun Song <[e-mail]>
> > > > ---
> > > >  Documentation/admin-guide/kernel-parameters.txt |  2 +-
> > > >  include/linux/page-flags.h                      | 75 +++++++++++++++++++++++--
> > > >  mm/hugetlb_vmemmap.c                            | 60 +++++++++++---------
> > > >  mm/sparse-vmemmap.c                             | 21 +++++++
> > > >  4 files changed, 126 insertions(+), 32 deletions(-)
> > > >
> > > > diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
> > > > index bdb22006f713..a154a7b3b9a5 100644
> > > > --- a/Documentation/admin-guide/kernel-parameters.txt
> > > > +++ b/Documentation/admin-guide/kernel-parameters.txt
> > > > @@ -1606,7 +1606,7 @@
> > > >                         [KNL] Reguires CONFIG_HUGETLB_PAGE_FREE_VMEMMAP
> > > >                         enabled.
> > > >                         Allows heavy hugetlb users to free up some more
> > > > -                       memory (6 * PAGE_SIZE for each 2MB hugetlb page).
> > > > +                       memory (7 * PAGE_SIZE for each 2MB hugetlb page).
> > > >                         Format: { on | off (default) }
> > > >
> > > >                         on:  enable the feature
> > > > diff --git a/include/linux/page-flags.h b/include/linux/page-flags.h
> > > > index 8e1d97d8f3bd..7b1a918ebd43 100644
> > > > --- a/include/linux/page-flags.h
> > > > +++ b/include/linux/page-flags.h
> > > > @@ -184,13 +184,64 @@ enum pageflags {
> > > >
> > > >  #ifndef __GENERATING_BOUNDS_H
> > > >
> > > > +#ifdef CONFIG_HUGETLB_PAGE_FREE_VMEMMAP
> > > > +extern bool hugetlb_free_vmemmap_enabled;
> > > > +
> > > > +/*
> > > > + * If the feature of freeing some vmemmap pages associated with each HugeTLB
> > > > + * page is enabled, the head vmemmap page frame is reused and all of the tail
> > > > + * vmemmap addresses map to the head vmemmap page frame (furture details can
> > > > + * refer to the figure at the head of the mm/hugetlb_vmemmap.c).  In other
> > > > + * word, there are more than one page struct with PG_head associated with each
> > > > + * HugeTLB page.  We __know__ that there is only one head page struct, the tail
> > > > + * page structs with PG_head are fake head page structs.  We need an approach
> > > > + * to distinguish between those two different types of page structs so that
> > > > + * compound_head() can return the real head page struct when the parameter is
> > > > + * the tail page struct but with PG_head.
> > > > + *
> > > > + * The page_head_if_fake() returns the real head page struct iff the @page may
> > > > + * be fake, otherwise, returns the @page if it cannot be a fake page struct.
> > > > + */
> > > > +static __always_inline const struct page *page_head_if_fake(const struct page *page)
> > > > +{
> > > > +       if (!hugetlb_free_vmemmap_enabled)
> > > > +               return page;
> > > > +
> > > > +       /*
> > > > +        * Only addresses aligned with PAGE_SIZE of struct page may be fake head
> > > > +        * struct page. The alignment check aims to avoid access the fields (
> > > > +        * e.g. compound_head) of the @page[1]. It can avoid touch a (possibly)
> > > > +        * cold cacheline in some cases.
> > > > +        */
> > > > +       if (IS_ALIGNED((unsigned long)page, PAGE_SIZE) &&
> > > > +           test_bit(PG_head, &page->flags)) {
> > > > +               /*
> > > > +                * We can safely access the field of the @page[1] with PG_head
> > > > +                * because the @page is a compound page composed with at least
> > > > +                * two contiguous pages.
> > > > +                */
> > > > +               unsigned long head = READ_ONCE(page[1].compound_head);
> > > > +
> > > > +               if (likely(head & 1))
> > > > +                       return (const struct page *)(head - 1);
> > > > +       }
> > > > +
> > > > +       return page;
> > > > +}
> > > > +#else
> > > > +static __always_inline const struct page *page_head_if_fake(const struct page *page)
> > > > +{
> > > > +       return page;
> > > > +}
> > > > +#endif
> > > > +
> > > >  static inline unsigned long _compound_head(const struct page *page)
> > > >  {
> > > >         unsigned long head = READ_ONCE(page->compound_head);
> > > >
> > > >         if (unlikely(head & 1))
> > > >                 return head - 1;
> > > > -       return (unsigned long)page;
> > > > +       return (unsigned long)page_head_if_fake(page);
> > >
> > > hard to read. page_head_if_fake,  what is the other side of
> > > page_head_if_not_fake?
> >
> > 1) return itself if the @page is not a fake head page.
> > 2) return head page if @page is a fake head page.
> >
> > So I want to express that page_head_if_fake returns a
> > head page only and only if the parameter of @page is a
> > fake head page. Otherwise, it returns itself.
> >
> > > I would expect something like
> > > page_to_page_head()
> > > or
> > > get_page_head()
> > >
> >
> > Those names seem to be not appropriate as well, because
> > its functionality does not make sure it can return a head
> > page. If the parameter is a head page, it definitely
> > returns a head page, otherwise, it may return itself which
> > may be a tail page.
> >
> > From this point of view, I still prefer page_head_if_fake.
> >
> > > Anyway, I am not quite sure what is the best name. but page_head_if_fake(page)
> > > sounds odd to me. just like the things have two sides, but if_fake  presents
> > > one side only.
> >
> > If others have any ideas, comments are welcome.
> >
> > >
> > > >  }
> > > >
> > > >  #define compound_head(page)    ((typeof(page))_compound_head(page))
> > > > @@ -225,12 +276,14 @@ static inline unsigned long _compound_head(const struct page *page)
> > > >
> > > >  static __always_inline int PageTail(struct page *page)
> > > >  {
> > > > -       return READ_ONCE(page->compound_head) & 1;
> > > > +       return READ_ONCE(page->compound_head) & 1 ||
> > > > +              page_head_if_fake(page) != page;
> > >
> > > i would expect a wrapper like:
> > > page_is_fake_head()
> >
> > Good point. Will do.
> >
> > >
> > > and the above page_to_page_head() can leverage the wrapper.
> > > here too.
> > >
> > > >  }
> > > >
> > > >  static __always_inline int PageCompound(struct page *page)
> > > >  {
> > > > -       return test_bit(PG_head, &page->flags) || PageTail(page);
> > > > +       return test_bit(PG_head, &page->flags) ||
> > > > +              READ_ONCE(page->compound_head) & 1;
> > >
> > > hard to read. could it be something like the below?
> > > return PageHead(page) || PageTail(page);
> > >
> > > or do we really need to change this function? even a fake head still has
> > > the true test_bit(PG_head, &page->flags), though it is not a real head, it
> > > is still a pagecompound, right?
> >
> > Right. PageCompound() can not be changed.  It is odd but
> > efficient because calling page_head_if_fake is eliminated.
> > So I select performance not readability. I'm not sure if it's
> > worth it.
>
> In order to improve readability, I'll introduce 3 helpers as follows.
>
> 1) page_head_or_fake(), which returns true for the head page
>    or fake head page.
> 2) page_head_is_fake(), which returns true for fake head page.
> 3) page_tail_not_fake_head(), which returns true for the tail page
>    except the fake head page.
>
> In the end, PageHead(), PageTail() and PageCompound() become
> the following.
>
> static __always_inline int PageHead(struct page *page)
> {
>     return page_head_or_fake(page) && !page_head_is_fake(page);
> }
>
> static __always_inline int PageTail(struct page *page)
> {
>     return page_tail_not_fake_head(page) || page_head_is_fake(page);
> }
>
> static __always_inline int PageCompound(struct page *page)
> {
>     return page_head_or_fake(page) || page_tail_not_fake_head(page);
> }
>
> Do those look more readable?
>

still not good enough. After a second thought, page_head_if_fake seems
to have the best performance though this function returns an odd value.
i just made a little bit refine on your code in doc:

diff --git a/include/linux/page-flags.h b/include/linux/page-flags.h
index 2c0d11e71e26..240c2fca13c7 100644
--- a/include/linux/page-flags.h
+++ b/include/linux/page-flags.h
@@ -197,8 +197,9 @@ extern bool hugetlb_free_vmemmap_enabled;
  * compound_head() can return the real head page struct when the parameter is
  * the tail page struct but with PG_head.
  *
- * The page_head_if_fake() returns the real head page struct iff the @page may
- * be fake, otherwise, returns the @page if it cannot be a fake page struct.
+ * The page_head_if_fake() returns the real head page struct if the @page is
+ * fake page_head, otherwise, returns @page which can either be a true page_
+ * head or tail.
  */
 static __always_inline const struct page *page_head_if_fake(const
struct page *page)
 {
@@ -226,6 +227,12 @@ static __always_inline const struct page
*page_head_if_fake(const struct page *p

        return page;
 }
+
+static __always_inline const struct page *page_is_fake_head(const
struct page *page)
+{
+       return page_head_if_fake(page) != page;
+}
+
 #else
 static __always_inline const struct page *page_head_if_fake(const
struct page *page)
 {
@@ -247,7 +254,7 @@ static inline unsigned long _compound_head(const
struct page *page)
 static __always_inline int PageTail(struct page *page)
 {
        return READ_ONCE(page->compound_head) & 1 ||
-              page_head_if_fake(page) != page;
+              page_is_fake_head(page);
 }

 static __always_inline int PageCompound(struct page *page)

> Thanks.

Thanks
barry

Muchun Song Sept. 21, 2021, 6:43 a.m. UTC | #4

On Sat, Sep 18, 2021 at 6:06 PM Muchun Song <[e-mail]> wrote:
>
> On Sat, Sep 18, 2021 at 12:39 PM Barry Song <[e-mail]> wrote:
> >
> > On Sat, Sep 18, 2021 at 12:08 AM Muchun Song <[e-mail]> wrote:
> > >
> > > Currently, we only free 6 vmemmap pages associated with a 2MB HugeTLB
> > > page. However, we can remap all tail vmemmap pages to the page frame
> > > mapped to with the head vmemmap page. Finally, we can free 7 vmemmap
> > > pages for a 2MB HugeTLB page. It is a fine gain (e.g. we can save
> > > extra 2GB memory when there is 1TB HugeTLB pages in the system
> > > compared with the current implementation).
> > >
> > > But the head vmemmap page is not freed to the buddy allocator and all
> > > tail vmemmap pages are mapped to the head vmemmap page frame. So we
> > > can see more than one struct page struct with PG_head (e.g. 8 per 2 MB
> > > HugeTLB page) associated with each HugeTLB page. We should adjust
> > > compound_head() to make it returns the real head struct page when the
> > > parameter is the tail struct page but with PG_head flag.
> > >
> > > Signed-off-by: Muchun Song <[e-mail]>
> > > ---
> > >  Documentation/admin-guide/kernel-parameters.txt |  2 +-
> > >  include/linux/page-flags.h                      | 75 +++++++++++++++++++++++--
> > >  mm/hugetlb_vmemmap.c                            | 60 +++++++++++---------
> > >  mm/sparse-vmemmap.c                             | 21 +++++++
> > >  4 files changed, 126 insertions(+), 32 deletions(-)
> > >
> > > diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
> > > index bdb22006f713..a154a7b3b9a5 100644
> > > --- a/Documentation/admin-guide/kernel-parameters.txt
> > > +++ b/Documentation/admin-guide/kernel-parameters.txt
> > > @@ -1606,7 +1606,7 @@
> > >                         [KNL] Reguires CONFIG_HUGETLB_PAGE_FREE_VMEMMAP
> > >                         enabled.
> > >                         Allows heavy hugetlb users to free up some more
> > > -                       memory (6 * PAGE_SIZE for each 2MB hugetlb page).
> > > +                       memory (7 * PAGE_SIZE for each 2MB hugetlb page).
> > >                         Format: { on | off (default) }
> > >
> > >                         on:  enable the feature
> > > diff --git a/include/linux/page-flags.h b/include/linux/page-flags.h
> > > index 8e1d97d8f3bd..7b1a918ebd43 100644
> > > --- a/include/linux/page-flags.h
> > > +++ b/include/linux/page-flags.h
> > > @@ -184,13 +184,64 @@ enum pageflags {
> > >
> > >  #ifndef __GENERATING_BOUNDS_H
> > >
> > > +#ifdef CONFIG_HUGETLB_PAGE_FREE_VMEMMAP
> > > +extern bool hugetlb_free_vmemmap_enabled;
> > > +
> > > +/*
> > > + * If the feature of freeing some vmemmap pages associated with each HugeTLB
> > > + * page is enabled, the head vmemmap page frame is reused and all of the tail
> > > + * vmemmap addresses map to the head vmemmap page frame (furture details can
> > > + * refer to the figure at the head of the mm/hugetlb_vmemmap.c).  In other
> > > + * word, there are more than one page struct with PG_head associated with each
> > > + * HugeTLB page.  We __know__ that there is only one head page struct, the tail
> > > + * page structs with PG_head are fake head page structs.  We need an approach
> > > + * to distinguish between those two different types of page structs so that
> > > + * compound_head() can return the real head page struct when the parameter is
> > > + * the tail page struct but with PG_head.
> > > + *
> > > + * The page_head_if_fake() returns the real head page struct iff the @page may
> > > + * be fake, otherwise, returns the @page if it cannot be a fake page struct.
> > > + */
> > > +static __always_inline const struct page *page_head_if_fake(const struct page *page)
> > > +{
> > > +       if (!hugetlb_free_vmemmap_enabled)
> > > +               return page;
> > > +
> > > +       /*
> > > +        * Only addresses aligned with PAGE_SIZE of struct page may be fake head
> > > +        * struct page. The alignment check aims to avoid access the fields (
> > > +        * e.g. compound_head) of the @page[1]. It can avoid touch a (possibly)
> > > +        * cold cacheline in some cases.
> > > +        */
> > > +       if (IS_ALIGNED((unsigned long)page, PAGE_SIZE) &&
> > > +           test_bit(PG_head, &page->flags)) {
> > > +               /*
> > > +                * We can safely access the field of the @page[1] with PG_head
> > > +                * because the @page is a compound page composed with at least
> > > +                * two contiguous pages.
> > > +                */
> > > +               unsigned long head = READ_ONCE(page[1].compound_head);
> > > +
> > > +               if (likely(head & 1))
> > > +                       return (const struct page *)(head - 1);
> > > +       }
> > > +
> > > +       return page;
> > > +}
> > > +#else
> > > +static __always_inline const struct page *page_head_if_fake(const struct page *page)
> > > +{
> > > +       return page;
> > > +}
> > > +#endif
> > > +
> > >  static inline unsigned long _compound_head(const struct page *page)
> > >  {
> > >         unsigned long head = READ_ONCE(page->compound_head);
> > >
> > >         if (unlikely(head & 1))
> > >                 return head - 1;
> > > -       return (unsigned long)page;
> > > +       return (unsigned long)page_head_if_fake(page);
> >
> > hard to read. page_head_if_fake,  what is the other side of
> > page_head_if_not_fake?
>
> 1) return itself if the @page is not a fake head page.
> 2) return head page if @page is a fake head page.
>
> So I want to express that page_head_if_fake returns a
> head page only and only if the parameter of @page is a
> fake head page. Otherwise, it returns itself.
>
> > I would expect something like
> > page_to_page_head()
> > or
> > get_page_head()
> >
>
> Those names seem to be not appropriate as well, because
> its functionality does not make sure it can return a head
> page. If the parameter is a head page, it definitely
> returns a head page, otherwise, it may return itself which
> may be a tail page.
>
> From this point of view, I still prefer page_head_if_fake.

After some thinking, I figured out 2 names.

page_head_if_fake() always returns a head page if the parameter
of @page is not a compound page or its ->flags has PG_head set
(you can think the head page is itself if the page is not a compound
page). All the callers of it already guarantee this. It means it has to
return a head page unless the @page is a tail page (except fake
head page). So I propose two names as follows.

1) page_head_unless_tail
2) page_head_filter_fake

The former means it always returns a head page unless the
caller passes a tail page as a parameter. The latter means
it always returns a head page but filtering out the fake head
page. The former is inspired by get_page_unless_zero.

What do you think?

Thanks.

>
> > Anyway, I am not quite sure what is the best name. but page_head_if_fake(page)
> > sounds odd to me. just like the things have two sides, but if_fake  presents
> > one side only.
>
> If others have any ideas, comments are welcome.
>
> >
> > >  }
> > >
> > >  #define compound_head(page)    ((typeof(page))_compound_head(page))
> > > @@ -225,12 +276,14 @@ static inline unsigned long _compound_head(const struct page *page)
> > >
> > >  static __always_inline int PageTail(struct page *page)
> > >  {
> > > -       return READ_ONCE(page->compound_head) & 1;
> > > +       return READ_ONCE(page->compound_head) & 1 ||
> > > +              page_head_if_fake(page) != page;
> >
> > i would expect a wrapper like:
> > page_is_fake_head()
>
> Good point. Will do.
>
> >
> > and the above page_to_page_head() can leverage the wrapper.
> > here too.
> >
> > >  }
> > >
> > >  static __always_inline int PageCompound(struct page *page)
> > >  {
> > > -       return test_bit(PG_head, &page->flags) || PageTail(page);
> > > +       return test_bit(PG_head, &page->flags) ||
> > > +              READ_ONCE(page->compound_head) & 1;
> >
> > hard to read. could it be something like the below?
> > return PageHead(page) || PageTail(page);
> >
> > or do we really need to change this function? even a fake head still has
> > the true test_bit(PG_head, &page->flags), though it is not a real head, it
> > is still a pagecompound, right?
>
> Right. PageCompound() can not be changed.  It is odd but
> efficient because calling page_head_if_fake is eliminated.
> So I select performance not readability. I'm not sure if it's
> worth it.
>
> Thanks.

Muchun Song Sept. 21, 2021, 10:22 a.m. UTC | #5

On Sat, Sep 18, 2021 at 6:06 PM Muchun Song <[e-mail]> wrote:
>
> On Sat, Sep 18, 2021 at 12:39 PM Barry Song <[e-mail]> wrote:
> >
> > On Sat, Sep 18, 2021 at 12:08 AM Muchun Song <[e-mail]> wrote:
> > >
> > > Currently, we only free 6 vmemmap pages associated with a 2MB HugeTLB
> > > page. However, we can remap all tail vmemmap pages to the page frame
> > > mapped to with the head vmemmap page. Finally, we can free 7 vmemmap
> > > pages for a 2MB HugeTLB page. It is a fine gain (e.g. we can save
> > > extra 2GB memory when there is 1TB HugeTLB pages in the system
> > > compared with the current implementation).
> > >
> > > But the head vmemmap page is not freed to the buddy allocator and all
> > > tail vmemmap pages are mapped to the head vmemmap page frame. So we
> > > can see more than one struct page struct with PG_head (e.g. 8 per 2 MB
> > > HugeTLB page) associated with each HugeTLB page. We should adjust
> > > compound_head() to make it returns the real head struct page when the
> > > parameter is the tail struct page but with PG_head flag.
> > >
> > > Signed-off-by: Muchun Song <[e-mail]>
> > > ---
> > >  Documentation/admin-guide/kernel-parameters.txt |  2 +-
> > >  include/linux/page-flags.h                      | 75 +++++++++++++++++++++++--
> > >  mm/hugetlb_vmemmap.c                            | 60 +++++++++++---------
> > >  mm/sparse-vmemmap.c                             | 21 +++++++
> > >  4 files changed, 126 insertions(+), 32 deletions(-)
> > >
> > > diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
> > > index bdb22006f713..a154a7b3b9a5 100644
> > > --- a/Documentation/admin-guide/kernel-parameters.txt
> > > +++ b/Documentation/admin-guide/kernel-parameters.txt
> > > @@ -1606,7 +1606,7 @@
> > >                         [KNL] Reguires CONFIG_HUGETLB_PAGE_FREE_VMEMMAP
> > >                         enabled.
> > >                         Allows heavy hugetlb users to free up some more
> > > -                       memory (6 * PAGE_SIZE for each 2MB hugetlb page).
> > > +                       memory (7 * PAGE_SIZE for each 2MB hugetlb page).
> > >                         Format: { on | off (default) }
> > >
> > >                         on:  enable the feature
> > > diff --git a/include/linux/page-flags.h b/include/linux/page-flags.h
> > > index 8e1d97d8f3bd..7b1a918ebd43 100644
> > > --- a/include/linux/page-flags.h
> > > +++ b/include/linux/page-flags.h
> > > @@ -184,13 +184,64 @@ enum pageflags {
> > >
> > >  #ifndef __GENERATING_BOUNDS_H
> > >
> > > +#ifdef CONFIG_HUGETLB_PAGE_FREE_VMEMMAP
> > > +extern bool hugetlb_free_vmemmap_enabled;
> > > +
> > > +/*
> > > + * If the feature of freeing some vmemmap pages associated with each HugeTLB
> > > + * page is enabled, the head vmemmap page frame is reused and all of the tail
> > > + * vmemmap addresses map to the head vmemmap page frame (furture details can
> > > + * refer to the figure at the head of the mm/hugetlb_vmemmap.c).  In other
> > > + * word, there are more than one page struct with PG_head associated with each
> > > + * HugeTLB page.  We __know__ that there is only one head page struct, the tail
> > > + * page structs with PG_head are fake head page structs.  We need an approach
> > > + * to distinguish between those two different types of page structs so that
> > > + * compound_head() can return the real head page struct when the parameter is
> > > + * the tail page struct but with PG_head.
> > > + *
> > > + * The page_head_if_fake() returns the real head page struct iff the @page may
> > > + * be fake, otherwise, returns the @page if it cannot be a fake page struct.
> > > + */
> > > +static __always_inline const struct page *page_head_if_fake(const struct page *page)
> > > +{
> > > +       if (!hugetlb_free_vmemmap_enabled)
> > > +               return page;
> > > +
> > > +       /*
> > > +        * Only addresses aligned with PAGE_SIZE of struct page may be fake head
> > > +        * struct page. The alignment check aims to avoid access the fields (
> > > +        * e.g. compound_head) of the @page[1]. It can avoid touch a (possibly)
> > > +        * cold cacheline in some cases.
> > > +        */
> > > +       if (IS_ALIGNED((unsigned long)page, PAGE_SIZE) &&
> > > +           test_bit(PG_head, &page->flags)) {
> > > +               /*
> > > +                * We can safely access the field of the @page[1] with PG_head
> > > +                * because the @page is a compound page composed with at least
> > > +                * two contiguous pages.
> > > +                */
> > > +               unsigned long head = READ_ONCE(page[1].compound_head);
> > > +
> > > +               if (likely(head & 1))
> > > +                       return (const struct page *)(head - 1);
> > > +       }
> > > +
> > > +       return page;
> > > +}
> > > +#else
> > > +static __always_inline const struct page *page_head_if_fake(const struct page *page)
> > > +{
> > > +       return page;
> > > +}
> > > +#endif
> > > +
> > >  static inline unsigned long _compound_head(const struct page *page)
> > >  {
> > >         unsigned long head = READ_ONCE(page->compound_head);
> > >
> > >         if (unlikely(head & 1))
> > >                 return head - 1;
> > > -       return (unsigned long)page;
> > > +       return (unsigned long)page_head_if_fake(page);
> >
> > hard to read. page_head_if_fake,  what is the other side of
> > page_head_if_not_fake?
>
> 1) return itself if the @page is not a fake head page.
> 2) return head page if @page is a fake head page.
>
> So I want to express that page_head_if_fake returns a
> head page only and only if the parameter of @page is a
> fake head page. Otherwise, it returns itself.
>
> > I would expect something like
> > page_to_page_head()
> > or
> > get_page_head()
> >
>
> Those names seem to be not appropriate as well, because
> its functionality does not make sure it can return a head
> page. If the parameter is a head page, it definitely
> returns a head page, otherwise, it may return itself which
> may be a tail page.
>
> From this point of view, I still prefer page_head_if_fake.
>
> > Anyway, I am not quite sure what is the best name. but page_head_if_fake(page)
> > sounds odd to me. just like the things have two sides, but if_fake  presents
> > one side only.
>
> If others have any ideas, comments are welcome.
>
> >
> > >  }
> > >
> > >  #define compound_head(page)    ((typeof(page))_compound_head(page))
> > > @@ -225,12 +276,14 @@ static inline unsigned long _compound_head(const struct page *page)
> > >
> > >  static __always_inline int PageTail(struct page *page)
> > >  {
> > > -       return READ_ONCE(page->compound_head) & 1;
> > > +       return READ_ONCE(page->compound_head) & 1 ||
> > > +              page_head_if_fake(page) != page;
> >
> > i would expect a wrapper like:
> > page_is_fake_head()
>
> Good point. Will do.
>
> >
> > and the above page_to_page_head() can leverage the wrapper.
> > here too.
> >
> > >  }
> > >
> > >  static __always_inline int PageCompound(struct page *page)
> > >  {
> > > -       return test_bit(PG_head, &page->flags) || PageTail(page);
> > > +       return test_bit(PG_head, &page->flags) ||
> > > +              READ_ONCE(page->compound_head) & 1;
> >
> > hard to read. could it be something like the below?
> > return PageHead(page) || PageTail(page);
> >
> > or do we really need to change this function? even a fake head still has
> > the true test_bit(PG_head, &page->flags), though it is not a real head, it
> > is still a pagecompound, right?
>
> Right. PageCompound() can not be changed.  It is odd but
> efficient because calling page_head_if_fake is eliminated.
> So I select performance not readability. I'm not sure if it's
> worth it.

In order to improve readability, I'll introduce 3 helpers as follows.

1) page_head_or_fake(), which returns true for the head page
   or fake head page.
2) page_head_is_fake(), which returns true for fake head page.
3) page_tail_not_fake_head(), which returns true for the tail page
   except the fake head page.

In the end, PageHead(), PageTail() and PageCompound() become
the following.

static __always_inline int PageHead(struct page *page)
{
    return page_head_or_fake(page) && !page_head_is_fake(page);
}

static __always_inline int PageTail(struct page *page)
{
    return page_tail_not_fake_head(page) || page_head_is_fake(page);
}

static __always_inline int PageCompound(struct page *page)
{
    return page_head_or_fake(page) || page_tail_not_fake_head(page);
}

Do those look more readable?

Thanks.

Muchun Song Sept. 21, 2021, 1:46 p.m. UTC | #6

On Tue, Sep 21, 2021 at 8:11 PM Barry Song <[e-mail]> wrote:
>
> On Tue, Sep 21, 2021 at 10:23 PM Muchun Song <[e-mail]> wrote:
> >
> > On Sat, Sep 18, 2021 at 6:06 PM Muchun Song <[e-mail]> wrote:
> > >
> > > On Sat, Sep 18, 2021 at 12:39 PM Barry Song <[e-mail]> wrote:
> > > >
> > > > On Sat, Sep 18, 2021 at 12:08 AM Muchun Song <[e-mail]> wrote:
> > > > >
> > > > > Currently, we only free 6 vmemmap pages associated with a 2MB HugeTLB
> > > > > page. However, we can remap all tail vmemmap pages to the page frame
> > > > > mapped to with the head vmemmap page. Finally, we can free 7 vmemmap
> > > > > pages for a 2MB HugeTLB page. It is a fine gain (e.g. we can save
> > > > > extra 2GB memory when there is 1TB HugeTLB pages in the system
> > > > > compared with the current implementation).
> > > > >
> > > > > But the head vmemmap page is not freed to the buddy allocator and all
> > > > > tail vmemmap pages are mapped to the head vmemmap page frame. So we
> > > > > can see more than one struct page struct with PG_head (e.g. 8 per 2 MB
> > > > > HugeTLB page) associated with each HugeTLB page. We should adjust
> > > > > compound_head() to make it returns the real head struct page when the
> > > > > parameter is the tail struct page but with PG_head flag.
> > > > >
> > > > > Signed-off-by: Muchun Song <[e-mail]>
> > > > > ---
> > > > >  Documentation/admin-guide/kernel-parameters.txt |  2 +-
> > > > >  include/linux/page-flags.h                      | 75 +++++++++++++++++++++++--
> > > > >  mm/hugetlb_vmemmap.c                            | 60 +++++++++++---------
> > > > >  mm/sparse-vmemmap.c                             | 21 +++++++
> > > > >  4 files changed, 126 insertions(+), 32 deletions(-)
> > > > >
> > > > > diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
> > > > > index bdb22006f713..a154a7b3b9a5 100644
> > > > > --- a/Documentation/admin-guide/kernel-parameters.txt
> > > > > +++ b/Documentation/admin-guide/kernel-parameters.txt
> > > > > @@ -1606,7 +1606,7 @@
> > > > >                         [KNL] Reguires CONFIG_HUGETLB_PAGE_FREE_VMEMMAP
> > > > >                         enabled.
> > > > >                         Allows heavy hugetlb users to free up some more
> > > > > -                       memory (6 * PAGE_SIZE for each 2MB hugetlb page).
> > > > > +                       memory (7 * PAGE_SIZE for each 2MB hugetlb page).
> > > > >                         Format: { on | off (default) }
> > > > >
> > > > >                         on:  enable the feature
> > > > > diff --git a/include/linux/page-flags.h b/include/linux/page-flags.h
> > > > > index 8e1d97d8f3bd..7b1a918ebd43 100644
> > > > > --- a/include/linux/page-flags.h
> > > > > +++ b/include/linux/page-flags.h
> > > > > @@ -184,13 +184,64 @@ enum pageflags {
> > > > >
> > > > >  #ifndef __GENERATING_BOUNDS_H
> > > > >
> > > > > +#ifdef CONFIG_HUGETLB_PAGE_FREE_VMEMMAP
> > > > > +extern bool hugetlb_free_vmemmap_enabled;
> > > > > +
> > > > > +/*
> > > > > + * If the feature of freeing some vmemmap pages associated with each HugeTLB
> > > > > + * page is enabled, the head vmemmap page frame is reused and all of the tail
> > > > > + * vmemmap addresses map to the head vmemmap page frame (furture details can
> > > > > + * refer to the figure at the head of the mm/hugetlb_vmemmap.c).  In other
> > > > > + * word, there are more than one page struct with PG_head associated with each
> > > > > + * HugeTLB page.  We __know__ that there is only one head page struct, the tail
> > > > > + * page structs with PG_head are fake head page structs.  We need an approach
> > > > > + * to distinguish between those two different types of page structs so that
> > > > > + * compound_head() can return the real head page struct when the parameter is
> > > > > + * the tail page struct but with PG_head.
> > > > > + *
> > > > > + * The page_head_if_fake() returns the real head page struct iff the @page may
> > > > > + * be fake, otherwise, returns the @page if it cannot be a fake page struct.
> > > > > + */
> > > > > +static __always_inline const struct page *page_head_if_fake(const struct page *page)
> > > > > +{
> > > > > +       if (!hugetlb_free_vmemmap_enabled)
> > > > > +               return page;
> > > > > +
> > > > > +       /*
> > > > > +        * Only addresses aligned with PAGE_SIZE of struct page may be fake head
> > > > > +        * struct page. The alignment check aims to avoid access the fields (
> > > > > +        * e.g. compound_head) of the @page[1]. It can avoid touch a (possibly)
> > > > > +        * cold cacheline in some cases.
> > > > > +        */
> > > > > +       if (IS_ALIGNED((unsigned long)page, PAGE_SIZE) &&
> > > > > +           test_bit(PG_head, &page->flags)) {
> > > > > +               /*
> > > > > +                * We can safely access the field of the @page[1] with PG_head
> > > > > +                * because the @page is a compound page composed with at least
> > > > > +                * two contiguous pages.
> > > > > +                */
> > > > > +               unsigned long head = READ_ONCE(page[1].compound_head);
> > > > > +
> > > > > +               if (likely(head & 1))
> > > > > +                       return (const struct page *)(head - 1);
> > > > > +       }
> > > > > +
> > > > > +       return page;
> > > > > +}
> > > > > +#else
> > > > > +static __always_inline const struct page *page_head_if_fake(const struct page *page)
> > > > > +{
> > > > > +       return page;
> > > > > +}
> > > > > +#endif
> > > > > +
> > > > >  static inline unsigned long _compound_head(const struct page *page)
> > > > >  {
> > > > >         unsigned long head = READ_ONCE(page->compound_head);
> > > > >
> > > > >         if (unlikely(head & 1))
> > > > >                 return head - 1;
> > > > > -       return (unsigned long)page;
> > > > > +       return (unsigned long)page_head_if_fake(page);
> > > >
> > > > hard to read. page_head_if_fake,  what is the other side of
> > > > page_head_if_not_fake?
> > >
> > > 1) return itself if the @page is not a fake head page.
> > > 2) return head page if @page is a fake head page.
> > >
> > > So I want to express that page_head_if_fake returns a
> > > head page only and only if the parameter of @page is a
> > > fake head page. Otherwise, it returns itself.
> > >
> > > > I would expect something like
> > > > page_to_page_head()
> > > > or
> > > > get_page_head()
> > > >
> > >
> > > Those names seem to be not appropriate as well, because
> > > its functionality does not make sure it can return a head
> > > page. If the parameter is a head page, it definitely
> > > returns a head page, otherwise, it may return itself which
> > > may be a tail page.
> > >
> > > From this point of view, I still prefer page_head_if_fake.
> > >
> > > > Anyway, I am not quite sure what is the best name. but page_head_if_fake(page)
> > > > sounds odd to me. just like the things have two sides, but if_fake  presents
> > > > one side only.
> > >
> > > If others have any ideas, comments are welcome.
> > >
> > > >
> > > > >  }
> > > > >
> > > > >  #define compound_head(page)    ((typeof(page))_compound_head(page))
> > > > > @@ -225,12 +276,14 @@ static inline unsigned long _compound_head(const struct page *page)
> > > > >
> > > > >  static __always_inline int PageTail(struct page *page)
> > > > >  {
> > > > > -       return READ_ONCE(page->compound_head) & 1;
> > > > > +       return READ_ONCE(page->compound_head) & 1 ||
> > > > > +              page_head_if_fake(page) != page;
> > > >
> > > > i would expect a wrapper like:
> > > > page_is_fake_head()
> > >
> > > Good point. Will do.
> > >
> > > >
> > > > and the above page_to_page_head() can leverage the wrapper.
> > > > here too.
> > > >
> > > > >  }
> > > > >
> > > > >  static __always_inline int PageCompound(struct page *page)
> > > > >  {
> > > > > -       return test_bit(PG_head, &page->flags) || PageTail(page);
> > > > > +       return test_bit(PG_head, &page->flags) ||
> > > > > +              READ_ONCE(page->compound_head) & 1;
> > > >
> > > > hard to read. could it be something like the below?
> > > > return PageHead(page) || PageTail(page);
> > > >
> > > > or do we really need to change this function? even a fake head still has
> > > > the true test_bit(PG_head, &page->flags), though it is not a real head, it
> > > > is still a pagecompound, right?
> > >
> > > Right. PageCompound() can not be changed.  It is odd but
> > > efficient because calling page_head_if_fake is eliminated.
> > > So I select performance not readability. I'm not sure if it's
> > > worth it.
> >
> > In order to improve readability, I'll introduce 3 helpers as follows.
> >
> > 1) page_head_or_fake(), which returns true for the head page
> >    or fake head page.
> > 2) page_head_is_fake(), which returns true for fake head page.
> > 3) page_tail_not_fake_head(), which returns true for the tail page
> >    except the fake head page.
> >
> > In the end, PageHead(), PageTail() and PageCompound() become
> > the following.
> >
> > static __always_inline int PageHead(struct page *page)
> > {
> >     return page_head_or_fake(page) && !page_head_is_fake(page);
> > }
> >
> > static __always_inline int PageTail(struct page *page)
> > {
> >     return page_tail_not_fake_head(page) || page_head_is_fake(page);
> > }
> >
> > static __always_inline int PageCompound(struct page *page)
> > {
> >     return page_head_or_fake(page) || page_tail_not_fake_head(page);
> > }
> >
> > Do those look more readable?
> >
>
> still not good enough. After a second thought, page_head_if_fake seems
> to have the best performance though this function returns an odd value.
> i just made a little bit refine on your code in doc:

Right. page_head_if_fake is the choice for performance.

>
> diff --git a/include/linux/page-flags.h b/include/linux/page-flags.h
> index 2c0d11e71e26..240c2fca13c7 100644
> --- a/include/linux/page-flags.h
> +++ b/include/linux/page-flags.h
> @@ -197,8 +197,9 @@ extern bool hugetlb_free_vmemmap_enabled;
>   * compound_head() can return the real head page struct when the parameter is
>   * the tail page struct but with PG_head.
>   *
> - * The page_head_if_fake() returns the real head page struct iff the @page may
> - * be fake, otherwise, returns the @page if it cannot be a fake page struct.
> + * The page_head_if_fake() returns the real head page struct if the @page is
> + * fake page_head, otherwise, returns @page which can either be a true page_
> + * head or tail.
>   */

Good annotation.

>  static __always_inline const struct page *page_head_if_fake(const
> struct page *page)
>  {
> @@ -226,6 +227,12 @@ static __always_inline const struct page
> *page_head_if_fake(const struct page *p
>
>         return page;
>  }
> +
> +static __always_inline const struct page *page_is_fake_head(const
> struct page *page)
> +{
> +       return page_head_if_fake(page) != page;
> +}
> +
>  #else
>  static __always_inline const struct page *page_head_if_fake(const
> struct page *page)
>  {
> @@ -247,7 +254,7 @@ static inline unsigned long _compound_head(const
> struct page *page)
>  static __always_inline int PageTail(struct page *page)
>  {
>         return READ_ONCE(page->compound_head) & 1 ||
> -              page_head_if_fake(page) != page;
> +              page_is_fake_head(page);
>  }

Yeah, this makes PageTail more readable. In your previous thread,
you proposed that why not use PageTail in PageCompound directly
to improve code readability. So I want to introduce 2 more helpers
besides page_is_fake_head().

static __always_inline int page_tail_not_fake_head(struct page *page)
{
    return READ_ONCE(page->compound_head) & 1;
}

static __always_inline int page_head_or_fake(struct page *page)
{
    return test_bit(PG_head, &page->flags);
}

Then PageTail() and PageCompound() change to the following.

static __always_inline int PageTail(struct page *page)
{
    return page_tail_not_fake_head(page) || page_is_fake_head(page);
}

static __always_inline int PageCompound(struct page *page)
{
    return page_head_or_fake(page) || page_tail_not_fake_head(page);
}

From the point of names of helpers, they act as self-annotation.
So I think PageTail and PageCompound become readable
as well. But you said "still not good enough". Is it because of
the names of helpers or introducing more complexity?

Thanks.

Barry Song Sept. 21, 2021, 8:43 p.m. UTC | #7

On Wed, Sep 22, 2021 at 1:46 AM Muchun Song <[e-mail]> wrote:
>
> On Tue, Sep 21, 2021 at 8:11 PM Barry Song <[e-mail]> wrote:
> >
> > On Tue, Sep 21, 2021 at 10:23 PM Muchun Song <[e-mail]> wrote:
> > >
> > > On Sat, Sep 18, 2021 at 6:06 PM Muchun Song <[e-mail]> wrote:
> > > >
> > > > On Sat, Sep 18, 2021 at 12:39 PM Barry Song <[e-mail]> wrote:
> > > > >
> > > > > On Sat, Sep 18, 2021 at 12:08 AM Muchun Song <[e-mail]> wrote:
> > > > > >
> > > > > > Currently, we only free 6 vmemmap pages associated with a 2MB HugeTLB
> > > > > > page. However, we can remap all tail vmemmap pages to the page frame
> > > > > > mapped to with the head vmemmap page. Finally, we can free 7 vmemmap
> > > > > > pages for a 2MB HugeTLB page. It is a fine gain (e.g. we can save
> > > > > > extra 2GB memory when there is 1TB HugeTLB pages in the system
> > > > > > compared with the current implementation).
> > > > > >
> > > > > > But the head vmemmap page is not freed to the buddy allocator and all
> > > > > > tail vmemmap pages are mapped to the head vmemmap page frame. So we
> > > > > > can see more than one struct page struct with PG_head (e.g. 8 per 2 MB
> > > > > > HugeTLB page) associated with each HugeTLB page. We should adjust
> > > > > > compound_head() to make it returns the real head struct page when the
> > > > > > parameter is the tail struct page but with PG_head flag.
> > > > > >
> > > > > > Signed-off-by: Muchun Song <[e-mail]>
> > > > > > ---
> > > > > >  Documentation/admin-guide/kernel-parameters.txt |  2 +-
> > > > > >  include/linux/page-flags.h                      | 75 +++++++++++++++++++++++--
> > > > > >  mm/hugetlb_vmemmap.c                            | 60 +++++++++++---------
> > > > > >  mm/sparse-vmemmap.c                             | 21 +++++++
> > > > > >  4 files changed, 126 insertions(+), 32 deletions(-)
> > > > > >
> > > > > > diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
> > > > > > index bdb22006f713..a154a7b3b9a5 100644
> > > > > > --- a/Documentation/admin-guide/kernel-parameters.txt
> > > > > > +++ b/Documentation/admin-guide/kernel-parameters.txt
> > > > > > @@ -1606,7 +1606,7 @@
> > > > > >                         [KNL] Reguires CONFIG_HUGETLB_PAGE_FREE_VMEMMAP
> > > > > >                         enabled.
> > > > > >                         Allows heavy hugetlb users to free up some more
> > > > > > -                       memory (6 * PAGE_SIZE for each 2MB hugetlb page).
> > > > > > +                       memory (7 * PAGE_SIZE for each 2MB hugetlb page).
> > > > > >                         Format: { on | off (default) }
> > > > > >
> > > > > >                         on:  enable the feature
> > > > > > diff --git a/include/linux/page-flags.h b/include/linux/page-flags.h
> > > > > > index 8e1d97d8f3bd..7b1a918ebd43 100644
> > > > > > --- a/include/linux/page-flags.h
> > > > > > +++ b/include/linux/page-flags.h
> > > > > > @@ -184,13 +184,64 @@ enum pageflags {
> > > > > >
> > > > > >  #ifndef __GENERATING_BOUNDS_H
> > > > > >
> > > > > > +#ifdef CONFIG_HUGETLB_PAGE_FREE_VMEMMAP
> > > > > > +extern bool hugetlb_free_vmemmap_enabled;
> > > > > > +
> > > > > > +/*
> > > > > > + * If the feature of freeing some vmemmap pages associated with each HugeTLB
> > > > > > + * page is enabled, the head vmemmap page frame is reused and all of the tail
> > > > > > + * vmemmap addresses map to the head vmemmap page frame (furture details can
> > > > > > + * refer to the figure at the head of the mm/hugetlb_vmemmap.c).  In other
> > > > > > + * word, there are more than one page struct with PG_head associated with each
> > > > > > + * HugeTLB page.  We __know__ that there is only one head page struct, the tail
> > > > > > + * page structs with PG_head are fake head page structs.  We need an approach
> > > > > > + * to distinguish between those two different types of page structs so that
> > > > > > + * compound_head() can return the real head page struct when the parameter is
> > > > > > + * the tail page struct but with PG_head.
> > > > > > + *
> > > > > > + * The page_head_if_fake() returns the real head page struct iff the @page may
> > > > > > + * be fake, otherwise, returns the @page if it cannot be a fake page struct.
> > > > > > + */
> > > > > > +static __always_inline const struct page *page_head_if_fake(const struct page *page)
> > > > > > +{
> > > > > > +       if (!hugetlb_free_vmemmap_enabled)
> > > > > > +               return page;
> > > > > > +
> > > > > > +       /*
> > > > > > +        * Only addresses aligned with PAGE_SIZE of struct page may be fake head
> > > > > > +        * struct page. The alignment check aims to avoid access the fields (
> > > > > > +        * e.g. compound_head) of the @page[1]. It can avoid touch a (possibly)
> > > > > > +        * cold cacheline in some cases.
> > > > > > +        */
> > > > > > +       if (IS_ALIGNED((unsigned long)page, PAGE_SIZE) &&
> > > > > > +           test_bit(PG_head, &page->flags)) {
> > > > > > +               /*
> > > > > > +                * We can safely access the field of the @page[1] with PG_head
> > > > > > +                * because the @page is a compound page composed with at least
> > > > > > +                * two contiguous pages.
> > > > > > +                */
> > > > > > +               unsigned long head = READ_ONCE(page[1].compound_head);
> > > > > > +
> > > > > > +               if (likely(head & 1))
> > > > > > +                       return (const struct page *)(head - 1);
> > > > > > +       }
> > > > > > +
> > > > > > +       return page;
> > > > > > +}
> > > > > > +#else
> > > > > > +static __always_inline const struct page *page_head_if_fake(const struct page *page)
> > > > > > +{
> > > > > > +       return page;
> > > > > > +}
> > > > > > +#endif
> > > > > > +
> > > > > >  static inline unsigned long _compound_head(const struct page *page)
> > > > > >  {
> > > > > >         unsigned long head = READ_ONCE(page->compound_head);
> > > > > >
> > > > > >         if (unlikely(head & 1))
> > > > > >                 return head - 1;
> > > > > > -       return (unsigned long)page;
> > > > > > +       return (unsigned long)page_head_if_fake(page);
> > > > >
> > > > > hard to read. page_head_if_fake,  what is the other side of
> > > > > page_head_if_not_fake?
> > > >
> > > > 1) return itself if the @page is not a fake head page.
> > > > 2) return head page if @page is a fake head page.
> > > >
> > > > So I want to express that page_head_if_fake returns a
> > > > head page only and only if the parameter of @page is a
> > > > fake head page. Otherwise, it returns itself.
> > > >
> > > > > I would expect something like
> > > > > page_to_page_head()
> > > > > or
> > > > > get_page_head()
> > > > >
> > > >
> > > > Those names seem to be not appropriate as well, because
> > > > its functionality does not make sure it can return a head
> > > > page. If the parameter is a head page, it definitely
> > > > returns a head page, otherwise, it may return itself which
> > > > may be a tail page.
> > > >
> > > > From this point of view, I still prefer page_head_if_fake.
> > > >
> > > > > Anyway, I am not quite sure what is the best name. but page_head_if_fake(page)
> > > > > sounds odd to me. just like the things have two sides, but if_fake  presents
> > > > > one side only.
> > > >
> > > > If others have any ideas, comments are welcome.
> > > >
> > > > >
> > > > > >  }
> > > > > >
> > > > > >  #define compound_head(page)    ((typeof(page))_compound_head(page))
> > > > > > @@ -225,12 +276,14 @@ static inline unsigned long _compound_head(const struct page *page)
> > > > > >
> > > > > >  static __always_inline int PageTail(struct page *page)
> > > > > >  {
> > > > > > -       return READ_ONCE(page->compound_head) & 1;
> > > > > > +       return READ_ONCE(page->compound_head) & 1 ||
> > > > > > +              page_head_if_fake(page) != page;
> > > > >
> > > > > i would expect a wrapper like:
> > > > > page_is_fake_head()
> > > >
> > > > Good point. Will do.
> > > >
> > > > >
> > > > > and the above page_to_page_head() can leverage the wrapper.
> > > > > here too.
> > > > >
> > > > > >  }
> > > > > >
> > > > > >  static __always_inline int PageCompound(struct page *page)
> > > > > >  {
> > > > > > -       return test_bit(PG_head, &page->flags) || PageTail(page);
> > > > > > +       return test_bit(PG_head, &page->flags) ||
> > > > > > +              READ_ONCE(page->compound_head) & 1;
> > > > >
> > > > > hard to read. could it be something like the below?
> > > > > return PageHead(page) || PageTail(page);
> > > > >
> > > > > or do we really need to change this function? even a fake head still has
> > > > > the true test_bit(PG_head, &page->flags), though it is not a real head, it
> > > > > is still a pagecompound, right?
> > > >
> > > > Right. PageCompound() can not be changed.  It is odd but
> > > > efficient because calling page_head_if_fake is eliminated.
> > > > So I select performance not readability. I'm not sure if it's
> > > > worth it.
> > >
> > > In order to improve readability, I'll introduce 3 helpers as follows.
> > >
> > > 1) page_head_or_fake(), which returns true for the head page
> > >    or fake head page.
> > > 2) page_head_is_fake(), which returns true for fake head page.
> > > 3) page_tail_not_fake_head(), which returns true for the tail page
> > >    except the fake head page.
> > >
> > > In the end, PageHead(), PageTail() and PageCompound() become
> > > the following.
> > >
> > > static __always_inline int PageHead(struct page *page)
> > > {
> > >     return page_head_or_fake(page) && !page_head_is_fake(page);
> > > }
> > >
> > > static __always_inline int PageTail(struct page *page)
> > > {
> > >     return page_tail_not_fake_head(page) || page_head_is_fake(page);
> > > }
> > >
> > > static __always_inline int PageCompound(struct page *page)
> > > {
> > >     return page_head_or_fake(page) || page_tail_not_fake_head(page);
> > > }
> > >
> > > Do those look more readable?
> > >
> >
> > still not good enough. After a second thought, page_head_if_fake seems
> > to have the best performance though this function returns an odd value.
> > i just made a little bit refine on your code in doc:
>
> Right. page_head_if_fake is the choice for performance.
>
> >
> > diff --git a/include/linux/page-flags.h b/include/linux/page-flags.h
> > index 2c0d11e71e26..240c2fca13c7 100644
> > --- a/include/linux/page-flags.h
> > +++ b/include/linux/page-flags.h
> > @@ -197,8 +197,9 @@ extern bool hugetlb_free_vmemmap_enabled;
> >   * compound_head() can return the real head page struct when the parameter is
> >   * the tail page struct but with PG_head.
> >   *
> > - * The page_head_if_fake() returns the real head page struct iff the @page may
> > - * be fake, otherwise, returns the @page if it cannot be a fake page struct.
> > + * The page_head_if_fake() returns the real head page struct if the @page is
> > + * fake page_head, otherwise, returns @page which can either be a true page_
> > + * head or tail.
> >   */
>
> Good annotation.
>
> >  static __always_inline const struct page *page_head_if_fake(const
> > struct page *page)
> >  {
> > @@ -226,6 +227,12 @@ static __always_inline const struct page
> > *page_head_if_fake(const struct page *p
> >
> >         return page;
> >  }
> > +
> > +static __always_inline const struct page *page_is_fake_head(const
> > struct page *page)
> > +{
> > +       return page_head_if_fake(page) != page;
> > +}
> > +
> >  #else
> >  static __always_inline const struct page *page_head_if_fake(const
> > struct page *page)
> >  {
> > @@ -247,7 +254,7 @@ static inline unsigned long _compound_head(const
> > struct page *page)
> >  static __always_inline int PageTail(struct page *page)
> >  {
> >         return READ_ONCE(page->compound_head) & 1 ||
> > -              page_head_if_fake(page) != page;
> > +              page_is_fake_head(page);
> >  }
>
> Yeah, this makes PageTail more readable. In your previous thread,
> you proposed that why not use PageTail in PageCompound directly
> to improve code readability. So I want to introduce 2 more helpers
> besides page_is_fake_head().
>
> static __always_inline int page_tail_not_fake_head(struct page *page)
> {
>     return READ_ONCE(page->compound_head) & 1;
> }
>
> static __always_inline int page_head_or_fake(struct page *page)
> {
>     return test_bit(PG_head, &page->flags);
> }
>
> Then PageTail() and PageCompound() change to the following.
>
> static __always_inline int PageTail(struct page *page)
> {
>     return page_tail_not_fake_head(page) || page_is_fake_head(page);
> }
>
> static __always_inline int PageCompound(struct page *page)
> {
>     return page_head_or_fake(page) || page_tail_not_fake_head(page);
> }
>
> From the point of names of helpers, they act as self-annotation.
> So I think PageTail and PageCompound become readable
> as well. But you said "still not good enough". Is it because of
> the names of helpers or introducing more complexity?

I really don't think it is worth this complexity. If there is anything to make
the code more readable, I would rename page_head_if_fake() to
page_fixed_dup_head().

this function fixes up the page:
1. if the page is a fake head, we need to return its true head (things
get fixed.)
2. if the page is not a fake head, in other words, it is either true
head or tail, no need to fix anything.

>
> Thanks.

Thanks
barry

Muchun Song Sept. 22, 2021, 2:38 a.m. UTC | #8

On Wed, Sep 22, 2021 at 4:43 AM Barry Song <[e-mail]> wrote:
>
> On Wed, Sep 22, 2021 at 1:46 AM Muchun Song <[e-mail]> wrote:
> >
> > On Tue, Sep 21, 2021 at 8:11 PM Barry Song <[e-mail]> wrote:
> > >
> > > On Tue, Sep 21, 2021 at 10:23 PM Muchun Song <[e-mail]> wrote:
> > > >
> > > > On Sat, Sep 18, 2021 at 6:06 PM Muchun Song <[e-mail]> wrote:
> > > > >
> > > > > On Sat, Sep 18, 2021 at 12:39 PM Barry Song <[e-mail]> wrote:
> > > > > >
> > > > > > On Sat, Sep 18, 2021 at 12:08 AM Muchun Song <[e-mail]> wrote:
> > > > > > >
> > > > > > > Currently, we only free 6 vmemmap pages associated with a 2MB HugeTLB
> > > > > > > page. However, we can remap all tail vmemmap pages to the page frame
> > > > > > > mapped to with the head vmemmap page. Finally, we can free 7 vmemmap
> > > > > > > pages for a 2MB HugeTLB page. It is a fine gain (e.g. we can save
> > > > > > > extra 2GB memory when there is 1TB HugeTLB pages in the system
> > > > > > > compared with the current implementation).
> > > > > > >
> > > > > > > But the head vmemmap page is not freed to the buddy allocator and all
> > > > > > > tail vmemmap pages are mapped to the head vmemmap page frame. So we
> > > > > > > can see more than one struct page struct with PG_head (e.g. 8 per 2 MB
> > > > > > > HugeTLB page) associated with each HugeTLB page. We should adjust
> > > > > > > compound_head() to make it returns the real head struct page when the
> > > > > > > parameter is the tail struct page but with PG_head flag.
> > > > > > >
> > > > > > > Signed-off-by: Muchun Song <[e-mail]>
> > > > > > > ---
> > > > > > >  Documentation/admin-guide/kernel-parameters.txt |  2 +-
> > > > > > >  include/linux/page-flags.h                      | 75 +++++++++++++++++++++++--
> > > > > > >  mm/hugetlb_vmemmap.c                            | 60 +++++++++++---------
> > > > > > >  mm/sparse-vmemmap.c                             | 21 +++++++
> > > > > > >  4 files changed, 126 insertions(+), 32 deletions(-)
> > > > > > >
> > > > > > > diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
> > > > > > > index bdb22006f713..a154a7b3b9a5 100644
> > > > > > > --- a/Documentation/admin-guide/kernel-parameters.txt
> > > > > > > +++ b/Documentation/admin-guide/kernel-parameters.txt
> > > > > > > @@ -1606,7 +1606,7 @@
> > > > > > >                         [KNL] Reguires CONFIG_HUGETLB_PAGE_FREE_VMEMMAP
> > > > > > >                         enabled.
> > > > > > >                         Allows heavy hugetlb users to free up some more
> > > > > > > -                       memory (6 * PAGE_SIZE for each 2MB hugetlb page).
> > > > > > > +                       memory (7 * PAGE_SIZE for each 2MB hugetlb page).
> > > > > > >                         Format: { on | off (default) }
> > > > > > >
> > > > > > >                         on:  enable the feature
> > > > > > > diff --git a/include/linux/page-flags.h b/include/linux/page-flags.h
> > > > > > > index 8e1d97d8f3bd..7b1a918ebd43 100644
> > > > > > > --- a/include/linux/page-flags.h
> > > > > > > +++ b/include/linux/page-flags.h
> > > > > > > @@ -184,13 +184,64 @@ enum pageflags {
> > > > > > >
> > > > > > >  #ifndef __GENERATING_BOUNDS_H
> > > > > > >
> > > > > > > +#ifdef CONFIG_HUGETLB_PAGE_FREE_VMEMMAP
> > > > > > > +extern bool hugetlb_free_vmemmap_enabled;
> > > > > > > +
> > > > > > > +/*
> > > > > > > + * If the feature of freeing some vmemmap pages associated with each HugeTLB
> > > > > > > + * page is enabled, the head vmemmap page frame is reused and all of the tail
> > > > > > > + * vmemmap addresses map to the head vmemmap page frame (furture details can
> > > > > > > + * refer to the figure at the head of the mm/hugetlb_vmemmap.c).  In other
> > > > > > > + * word, there are more than one page struct with PG_head associated with each
> > > > > > > + * HugeTLB page.  We __know__ that there is only one head page struct, the tail
> > > > > > > + * page structs with PG_head are fake head page structs.  We need an approach
> > > > > > > + * to distinguish between those two different types of page structs so that
> > > > > > > + * compound_head() can return the real head page struct when the parameter is
> > > > > > > + * the tail page struct but with PG_head.
> > > > > > > + *
> > > > > > > + * The page_head_if_fake() returns the real head page struct iff the @page may
> > > > > > > + * be fake, otherwise, returns the @page if it cannot be a fake page struct.
> > > > > > > + */
> > > > > > > +static __always_inline const struct page *page_head_if_fake(const struct page *page)
> > > > > > > +{
> > > > > > > +       if (!hugetlb_free_vmemmap_enabled)
> > > > > > > +               return page;
> > > > > > > +
> > > > > > > +       /*
> > > > > > > +        * Only addresses aligned with PAGE_SIZE of struct page may be fake head
> > > > > > > +        * struct page. The alignment check aims to avoid access the fields (
> > > > > > > +        * e.g. compound_head) of the @page[1]. It can avoid touch a (possibly)
> > > > > > > +        * cold cacheline in some cases.
> > > > > > > +        */
> > > > > > > +       if (IS_ALIGNED((unsigned long)page, PAGE_SIZE) &&
> > > > > > > +           test_bit(PG_head, &page->flags)) {
> > > > > > > +               /*
> > > > > > > +                * We can safely access the field of the @page[1] with PG_head
> > > > > > > +                * because the @page is a compound page composed with at least
> > > > > > > +                * two contiguous pages.
> > > > > > > +                */
> > > > > > > +               unsigned long head = READ_ONCE(page[1].compound_head);
> > > > > > > +
> > > > > > > +               if (likely(head & 1))
> > > > > > > +                       return (const struct page *)(head - 1);
> > > > > > > +       }
> > > > > > > +
> > > > > > > +       return page;
> > > > > > > +}
> > > > > > > +#else
> > > > > > > +static __always_inline const struct page *page_head_if_fake(const struct page *page)
> > > > > > > +{
> > > > > > > +       return page;
> > > > > > > +}
> > > > > > > +#endif
> > > > > > > +
> > > > > > >  static inline unsigned long _compound_head(const struct page *page)
> > > > > > >  {
> > > > > > >         unsigned long head = READ_ONCE(page->compound_head);
> > > > > > >
> > > > > > >         if (unlikely(head & 1))
> > > > > > >                 return head - 1;
> > > > > > > -       return (unsigned long)page;
> > > > > > > +       return (unsigned long)page_head_if_fake(page);
> > > > > >
> > > > > > hard to read. page_head_if_fake,  what is the other side of
> > > > > > page_head_if_not_fake?
> > > > >
> > > > > 1) return itself if the @page is not a fake head page.
> > > > > 2) return head page if @page is a fake head page.
> > > > >
> > > > > So I want to express that page_head_if_fake returns a
> > > > > head page only and only if the parameter of @page is a
> > > > > fake head page. Otherwise, it returns itself.
> > > > >
> > > > > > I would expect something like
> > > > > > page_to_page_head()
> > > > > > or
> > > > > > get_page_head()
> > > > > >
> > > > >
> > > > > Those names seem to be not appropriate as well, because
> > > > > its functionality does not make sure it can return a head
> > > > > page. If the parameter is a head page, it definitely
> > > > > returns a head page, otherwise, it may return itself which
> > > > > may be a tail page.
> > > > >
> > > > > From this point of view, I still prefer page_head_if_fake.
> > > > >
> > > > > > Anyway, I am not quite sure what is the best name. but page_head_if_fake(page)
> > > > > > sounds odd to me. just like the things have two sides, but if_fake  presents
> > > > > > one side only.
> > > > >
> > > > > If others have any ideas, comments are welcome.
> > > > >
> > > > > >
> > > > > > >  }
> > > > > > >
> > > > > > >  #define compound_head(page)    ((typeof(page))_compound_head(page))
> > > > > > > @@ -225,12 +276,14 @@ static inline unsigned long _compound_head(const struct page *page)
> > > > > > >
> > > > > > >  static __always_inline int PageTail(struct page *page)
> > > > > > >  {
> > > > > > > -       return READ_ONCE(page->compound_head) & 1;
> > > > > > > +       return READ_ONCE(page->compound_head) & 1 ||
> > > > > > > +              page_head_if_fake(page) != page;
> > > > > >
> > > > > > i would expect a wrapper like:
> > > > > > page_is_fake_head()
> > > > >
> > > > > Good point. Will do.
> > > > >
> > > > > >
> > > > > > and the above page_to_page_head() can leverage the wrapper.
> > > > > > here too.
> > > > > >
> > > > > > >  }
> > > > > > >
> > > > > > >  static __always_inline int PageCompound(struct page *page)
> > > > > > >  {
> > > > > > > -       return test_bit(PG_head, &page->flags) || PageTail(page);
> > > > > > > +       return test_bit(PG_head, &page->flags) ||
> > > > > > > +              READ_ONCE(page->compound_head) & 1;
> > > > > >
> > > > > > hard to read. could it be something like the below?
> > > > > > return PageHead(page) || PageTail(page);
> > > > > >
> > > > > > or do we really need to change this function? even a fake head still has
> > > > > > the true test_bit(PG_head, &page->flags), though it is not a real head, it
> > > > > > is still a pagecompound, right?
> > > > >
> > > > > Right. PageCompound() can not be changed.  It is odd but
> > > > > efficient because calling page_head_if_fake is eliminated.
> > > > > So I select performance not readability. I'm not sure if it's
> > > > > worth it.
> > > >
> > > > In order to improve readability, I'll introduce 3 helpers as follows.
> > > >
> > > > 1) page_head_or_fake(), which returns true for the head page
> > > >    or fake head page.
> > > > 2) page_head_is_fake(), which returns true for fake head page.
> > > > 3) page_tail_not_fake_head(), which returns true for the tail page
> > > >    except the fake head page.
> > > >
> > > > In the end, PageHead(), PageTail() and PageCompound() become
> > > > the following.
> > > >
> > > > static __always_inline int PageHead(struct page *page)
> > > > {
> > > >     return page_head_or_fake(page) && !page_head_is_fake(page);
> > > > }
> > > >
> > > > static __always_inline int PageTail(struct page *page)
> > > > {
> > > >     return page_tail_not_fake_head(page) || page_head_is_fake(page);
> > > > }
> > > >
> > > > static __always_inline int PageCompound(struct page *page)
> > > > {
> > > >     return page_head_or_fake(page) || page_tail_not_fake_head(page);
> > > > }
> > > >
> > > > Do those look more readable?
> > > >
> > >
> > > still not good enough. After a second thought, page_head_if_fake seems
> > > to have the best performance though this function returns an odd value.
> > > i just made a little bit refine on your code in doc:
> >
> > Right. page_head_if_fake is the choice for performance.
> >
> > >
> > > diff --git a/include/linux/page-flags.h b/include/linux/page-flags.h
> > > index 2c0d11e71e26..240c2fca13c7 100644
> > > --- a/include/linux/page-flags.h
> > > +++ b/include/linux/page-flags.h
> > > @@ -197,8 +197,9 @@ extern bool hugetlb_free_vmemmap_enabled;
> > >   * compound_head() can return the real head page struct when the parameter is
> > >   * the tail page struct but with PG_head.
> > >   *
> > > - * The page_head_if_fake() returns the real head page struct iff the @page may
> > > - * be fake, otherwise, returns the @page if it cannot be a fake page struct.
> > > + * The page_head_if_fake() returns the real head page struct if the @page is
> > > + * fake page_head, otherwise, returns @page which can either be a true page_
> > > + * head or tail.
> > >   */
> >
> > Good annotation.
> >
> > >  static __always_inline const struct page *page_head_if_fake(const
> > > struct page *page)
> > >  {
> > > @@ -226,6 +227,12 @@ static __always_inline const struct page
> > > *page_head_if_fake(const struct page *p
> > >
> > >         return page;
> > >  }
> > > +
> > > +static __always_inline const struct page *page_is_fake_head(const
> > > struct page *page)
> > > +{
> > > +       return page_head_if_fake(page) != page;
> > > +}
> > > +
> > >  #else
> > >  static __always_inline const struct page *page_head_if_fake(const
> > > struct page *page)
> > >  {
> > > @@ -247,7 +254,7 @@ static inline unsigned long _compound_head(const
> > > struct page *page)
> > >  static __always_inline int PageTail(struct page *page)
> > >  {
> > >         return READ_ONCE(page->compound_head) & 1 ||
> > > -              page_head_if_fake(page) != page;
> > > +              page_is_fake_head(page);
> > >  }
> >
> > Yeah, this makes PageTail more readable. In your previous thread,
> > you proposed that why not use PageTail in PageCompound directly
> > to improve code readability. So I want to introduce 2 more helpers
> > besides page_is_fake_head().
> >
> > static __always_inline int page_tail_not_fake_head(struct page *page)
> > {
> >     return READ_ONCE(page->compound_head) & 1;
> > }
> >
> > static __always_inline int page_head_or_fake(struct page *page)
> > {
> >     return test_bit(PG_head, &page->flags);
> > }
> >
> > Then PageTail() and PageCompound() change to the following.
> >
> > static __always_inline int PageTail(struct page *page)
> > {
> >     return page_tail_not_fake_head(page) || page_is_fake_head(page);
> > }
> >
> > static __always_inline int PageCompound(struct page *page)
> > {
> >     return page_head_or_fake(page) || page_tail_not_fake_head(page);
> > }
> >
> > From the point of names of helpers, they act as self-annotation.
> > So I think PageTail and PageCompound become readable
> > as well. But you said "still not good enough". Is it because of
> > the names of helpers or introducing more complexity?
>
> I really don't think it is worth this complexity. If there is anything to make

Got it.

> the code more readable, I would rename page_head_if_fake() to
> page_fixed_dup_head().

Here means page_fixed_up_head, right? Is it a typo?

Thanks.

>
> this function fixes up the page:
> 1. if the page is a fake head, we need to return its true head (things
> get fixed.)
> 2. if the page is not a fake head, in other words, it is either true
> head or tail, no need to fix anything.
>
> >
> > Thanks.
>
> Thanks
> barry

Barry Song Sept. 22, 2021, 7:36 a.m. UTC | #9

On Wed, Sep 22, 2021 at 2:39 PM Muchun Song <[e-mail]> wrote:
>
> On Wed, Sep 22, 2021 at 4:43 AM Barry Song <[e-mail]> wrote:
> >
> > On Wed, Sep 22, 2021 at 1:46 AM Muchun Song <[e-mail]> wrote:
> > >
> > > On Tue, Sep 21, 2021 at 8:11 PM Barry Song <[e-mail]> wrote:
> > > >
> > > > On Tue, Sep 21, 2021 at 10:23 PM Muchun Song <[e-mail]> wrote:
> > > > >
> > > > > On Sat, Sep 18, 2021 at 6:06 PM Muchun Song <[e-mail]> wrote:
> > > > > >
> > > > > > On Sat, Sep 18, 2021 at 12:39 PM Barry Song <[e-mail]> wrote:
> > > > > > >
> > > > > > > On Sat, Sep 18, 2021 at 12:08 AM Muchun Song <[e-mail]> wrote:
> > > > > > > >
> > > > > > > > Currently, we only free 6 vmemmap pages associated with a 2MB HugeTLB
> > > > > > > > page. However, we can remap all tail vmemmap pages to the page frame
> > > > > > > > mapped to with the head vmemmap page. Finally, we can free 7 vmemmap
> > > > > > > > pages for a 2MB HugeTLB page. It is a fine gain (e.g. we can save
> > > > > > > > extra 2GB memory when there is 1TB HugeTLB pages in the system
> > > > > > > > compared with the current implementation).
> > > > > > > >
> > > > > > > > But the head vmemmap page is not freed to the buddy allocator and all
> > > > > > > > tail vmemmap pages are mapped to the head vmemmap page frame. So we
> > > > > > > > can see more than one struct page struct with PG_head (e.g. 8 per 2 MB
> > > > > > > > HugeTLB page) associated with each HugeTLB page. We should adjust
> > > > > > > > compound_head() to make it returns the real head struct page when the
> > > > > > > > parameter is the tail struct page but with PG_head flag.
> > > > > > > >
> > > > > > > > Signed-off-by: Muchun Song <[e-mail]>
> > > > > > > > ---
> > > > > > > >  Documentation/admin-guide/kernel-parameters.txt |  2 +-
> > > > > > > >  include/linux/page-flags.h                      | 75 +++++++++++++++++++++++--
> > > > > > > >  mm/hugetlb_vmemmap.c                            | 60 +++++++++++---------
> > > > > > > >  mm/sparse-vmemmap.c                             | 21 +++++++
> > > > > > > >  4 files changed, 126 insertions(+), 32 deletions(-)
> > > > > > > >
> > > > > > > > diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
> > > > > > > > index bdb22006f713..a154a7b3b9a5 100644
> > > > > > > > --- a/Documentation/admin-guide/kernel-parameters.txt
> > > > > > > > +++ b/Documentation/admin-guide/kernel-parameters.txt
> > > > > > > > @@ -1606,7 +1606,7 @@
> > > > > > > >                         [KNL] Reguires CONFIG_HUGETLB_PAGE_FREE_VMEMMAP
> > > > > > > >                         enabled.
> > > > > > > >                         Allows heavy hugetlb users to free up some more
> > > > > > > > -                       memory (6 * PAGE_SIZE for each 2MB hugetlb page).
> > > > > > > > +                       memory (7 * PAGE_SIZE for each 2MB hugetlb page).
> > > > > > > >                         Format: { on | off (default) }
> > > > > > > >
> > > > > > > >                         on:  enable the feature
> > > > > > > > diff --git a/include/linux/page-flags.h b/include/linux/page-flags.h
> > > > > > > > index 8e1d97d8f3bd..7b1a918ebd43 100644
> > > > > > > > --- a/include/linux/page-flags.h
> > > > > > > > +++ b/include/linux/page-flags.h
> > > > > > > > @@ -184,13 +184,64 @@ enum pageflags {
> > > > > > > >
> > > > > > > >  #ifndef __GENERATING_BOUNDS_H
> > > > > > > >
> > > > > > > > +#ifdef CONFIG_HUGETLB_PAGE_FREE_VMEMMAP
> > > > > > > > +extern bool hugetlb_free_vmemmap_enabled;
> > > > > > > > +
> > > > > > > > +/*
> > > > > > > > + * If the feature of freeing some vmemmap pages associated with each HugeTLB
> > > > > > > > + * page is enabled, the head vmemmap page frame is reused and all of the tail
> > > > > > > > + * vmemmap addresses map to the head vmemmap page frame (furture details can
> > > > > > > > + * refer to the figure at the head of the mm/hugetlb_vmemmap.c).  In other
> > > > > > > > + * word, there are more than one page struct with PG_head associated with each
> > > > > > > > + * HugeTLB page.  We __know__ that there is only one head page struct, the tail
> > > > > > > > + * page structs with PG_head are fake head page structs.  We need an approach
> > > > > > > > + * to distinguish between those two different types of page structs so that
> > > > > > > > + * compound_head() can return the real head page struct when the parameter is
> > > > > > > > + * the tail page struct but with PG_head.
> > > > > > > > + *
> > > > > > > > + * The page_head_if_fake() returns the real head page struct iff the @page may
> > > > > > > > + * be fake, otherwise, returns the @page if it cannot be a fake page struct.
> > > > > > > > + */
> > > > > > > > +static __always_inline const struct page *page_head_if_fake(const struct page *page)
> > > > > > > > +{
> > > > > > > > +       if (!hugetlb_free_vmemmap_enabled)
> > > > > > > > +               return page;
> > > > > > > > +
> > > > > > > > +       /*
> > > > > > > > +        * Only addresses aligned with PAGE_SIZE of struct page may be fake head
> > > > > > > > +        * struct page. The alignment check aims to avoid access the fields (
> > > > > > > > +        * e.g. compound_head) of the @page[1]. It can avoid touch a (possibly)
> > > > > > > > +        * cold cacheline in some cases.
> > > > > > > > +        */
> > > > > > > > +       if (IS_ALIGNED((unsigned long)page, PAGE_SIZE) &&
> > > > > > > > +           test_bit(PG_head, &page->flags)) {
> > > > > > > > +               /*
> > > > > > > > +                * We can safely access the field of the @page[1] with PG_head
> > > > > > > > +                * because the @page is a compound page composed with at least
> > > > > > > > +                * two contiguous pages.
> > > > > > > > +                */
> > > > > > > > +               unsigned long head = READ_ONCE(page[1].compound_head);
> > > > > > > > +
> > > > > > > > +               if (likely(head & 1))
> > > > > > > > +                       return (const struct page *)(head - 1);
> > > > > > > > +       }
> > > > > > > > +
> > > > > > > > +       return page;
> > > > > > > > +}
> > > > > > > > +#else
> > > > > > > > +static __always_inline const struct page *page_head_if_fake(const struct page *page)
> > > > > > > > +{
> > > > > > > > +       return page;
> > > > > > > > +}
> > > > > > > > +#endif
> > > > > > > > +
> > > > > > > >  static inline unsigned long _compound_head(const struct page *page)
> > > > > > > >  {
> > > > > > > >         unsigned long head = READ_ONCE(page->compound_head);
> > > > > > > >
> > > > > > > >         if (unlikely(head & 1))
> > > > > > > >                 return head - 1;
> > > > > > > > -       return (unsigned long)page;
> > > > > > > > +       return (unsigned long)page_head_if_fake(page);
> > > > > > >
> > > > > > > hard to read. page_head_if_fake,  what is the other side of
> > > > > > > page_head_if_not_fake?
> > > > > >
> > > > > > 1) return itself if the @page is not a fake head page.
> > > > > > 2) return head page if @page is a fake head page.
> > > > > >
> > > > > > So I want to express that page_head_if_fake returns a
> > > > > > head page only and only if the parameter of @page is a
> > > > > > fake head page. Otherwise, it returns itself.
> > > > > >
> > > > > > > I would expect something like
> > > > > > > page_to_page_head()
> > > > > > > or
> > > > > > > get_page_head()
> > > > > > >
> > > > > >
> > > > > > Those names seem to be not appropriate as well, because
> > > > > > its functionality does not make sure it can return a head
> > > > > > page. If the parameter is a head page, it definitely
> > > > > > returns a head page, otherwise, it may return itself which
> > > > > > may be a tail page.
> > > > > >
> > > > > > From this point of view, I still prefer page_head_if_fake.
> > > > > >
> > > > > > > Anyway, I am not quite sure what is the best name. but page_head_if_fake(page)
> > > > > > > sounds odd to me. just like the things have two sides, but if_fake  presents
> > > > > > > one side only.
> > > > > >
> > > > > > If others have any ideas, comments are welcome.
> > > > > >
> > > > > > >
> > > > > > > >  }
> > > > > > > >
> > > > > > > >  #define compound_head(page)    ((typeof(page))_compound_head(page))
> > > > > > > > @@ -225,12 +276,14 @@ static inline unsigned long _compound_head(const struct page *page)
> > > > > > > >
> > > > > > > >  static __always_inline int PageTail(struct page *page)
> > > > > > > >  {
> > > > > > > > -       return READ_ONCE(page->compound_head) & 1;
> > > > > > > > +       return READ_ONCE(page->compound_head) & 1 ||
> > > > > > > > +              page_head_if_fake(page) != page;
> > > > > > >
> > > > > > > i would expect a wrapper like:
> > > > > > > page_is_fake_head()
> > > > > >
> > > > > > Good point. Will do.
> > > > > >
> > > > > > >
> > > > > > > and the above page_to_page_head() can leverage the wrapper.
> > > > > > > here too.
> > > > > > >
> > > > > > > >  }
> > > > > > > >
> > > > > > > >  static __always_inline int PageCompound(struct page *page)
> > > > > > > >  {
> > > > > > > > -       return test_bit(PG_head, &page->flags) || PageTail(page);
> > > > > > > > +       return test_bit(PG_head, &page->flags) ||
> > > > > > > > +              READ_ONCE(page->compound_head) & 1;
> > > > > > >
> > > > > > > hard to read. could it be something like the below?
> > > > > > > return PageHead(page) || PageTail(page);
> > > > > > >
> > > > > > > or do we really need to change this function? even a fake head still has
> > > > > > > the true test_bit(PG_head, &page->flags), though it is not a real head, it
> > > > > > > is still a pagecompound, right?
> > > > > >
> > > > > > Right. PageCompound() can not be changed.  It is odd but
> > > > > > efficient because calling page_head_if_fake is eliminated.
> > > > > > So I select performance not readability. I'm not sure if it's
> > > > > > worth it.
> > > > >
> > > > > In order to improve readability, I'll introduce 3 helpers as follows.
> > > > >
> > > > > 1) page_head_or_fake(), which returns true for the head page
> > > > >    or fake head page.
> > > > > 2) page_head_is_fake(), which returns true for fake head page.
> > > > > 3) page_tail_not_fake_head(), which returns true for the tail page
> > > > >    except the fake head page.
> > > > >
> > > > > In the end, PageHead(), PageTail() and PageCompound() become
> > > > > the following.
> > > > >
> > > > > static __always_inline int PageHead(struct page *page)
> > > > > {
> > > > >     return page_head_or_fake(page) && !page_head_is_fake(page);
> > > > > }
> > > > >
> > > > > static __always_inline int PageTail(struct page *page)
> > > > > {
> > > > >     return page_tail_not_fake_head(page) || page_head_is_fake(page);
> > > > > }
> > > > >
> > > > > static __always_inline int PageCompound(struct page *page)
> > > > > {
> > > > >     return page_head_or_fake(page) || page_tail_not_fake_head(page);
> > > > > }
> > > > >
> > > > > Do those look more readable?
> > > > >
> > > >
> > > > still not good enough. After a second thought, page_head_if_fake seems
> > > > to have the best performance though this function returns an odd value.
> > > > i just made a little bit refine on your code in doc:
> > >
> > > Right. page_head_if_fake is the choice for performance.
> > >
> > > >
> > > > diff --git a/include/linux/page-flags.h b/include/linux/page-flags.h
> > > > index 2c0d11e71e26..240c2fca13c7 100644
> > > > --- a/include/linux/page-flags.h
> > > > +++ b/include/linux/page-flags.h
> > > > @@ -197,8 +197,9 @@ extern bool hugetlb_free_vmemmap_enabled;
> > > >   * compound_head() can return the real head page struct when the parameter is
> > > >   * the tail page struct but with PG_head.
> > > >   *
> > > > - * The page_head_if_fake() returns the real head page struct iff the @page may
> > > > - * be fake, otherwise, returns the @page if it cannot be a fake page struct.
> > > > + * The page_head_if_fake() returns the real head page struct if the @page is
> > > > + * fake page_head, otherwise, returns @page which can either be a true page_
> > > > + * head or tail.
> > > >   */
> > >
> > > Good annotation.
> > >
> > > >  static __always_inline const struct page *page_head_if_fake(const
> > > > struct page *page)
> > > >  {
> > > > @@ -226,6 +227,12 @@ static __always_inline const struct page
> > > > *page_head_if_fake(const struct page *p
> > > >
> > > >         return page;
> > > >  }
> > > > +
> > > > +static __always_inline const struct page *page_is_fake_head(const
> > > > struct page *page)
> > > > +{
> > > > +       return page_head_if_fake(page) != page;
> > > > +}
> > > > +
> > > >  #else
> > > >  static __always_inline const struct page *page_head_if_fake(const
> > > > struct page *page)
> > > >  {
> > > > @@ -247,7 +254,7 @@ static inline unsigned long _compound_head(const
> > > > struct page *page)
> > > >  static __always_inline int PageTail(struct page *page)
> > > >  {
> > > >         return READ_ONCE(page->compound_head) & 1 ||
> > > > -              page_head_if_fake(page) != page;
> > > > +              page_is_fake_head(page);
> > > >  }
> > >
> > > Yeah, this makes PageTail more readable. In your previous thread,
> > > you proposed that why not use PageTail in PageCompound directly
> > > to improve code readability. So I want to introduce 2 more helpers
> > > besides page_is_fake_head().
> > >
> > > static __always_inline int page_tail_not_fake_head(struct page *page)
> > > {
> > >     return READ_ONCE(page->compound_head) & 1;
> > > }
> > >
> > > static __always_inline int page_head_or_fake(struct page *page)
> > > {
> > >     return test_bit(PG_head, &page->flags);
> > > }
> > >
> > > Then PageTail() and PageCompound() change to the following.
> > >
> > > static __always_inline int PageTail(struct page *page)
> > > {
> > >     return page_tail_not_fake_head(page) || page_is_fake_head(page);
> > > }
> > >
> > > static __always_inline int PageCompound(struct page *page)
> > > {
> > >     return page_head_or_fake(page) || page_tail_not_fake_head(page);
> > > }
> > >
> > > From the point of names of helpers, they act as self-annotation.
> > > So I think PageTail and PageCompound become readable
> > > as well. But you said "still not good enough". Is it because of
> > > the names of helpers or introducing more complexity?
> >
> > I really don't think it is worth this complexity. If there is anything to make
>
> Got it.
>
> > the code more readable, I would rename page_head_if_fake() to
> > page_fixed_dup_head().
>
> Here means page_fixed_up_head, right? Is it a typo?

I actually meant "duplicated", but in your case, it is "fake".
It doesn't matter too much. Both are ok.

>
> Thanks.
>
> >
> > this function fixes up the page:
> > 1. if the page is a fake head, we need to return its true head (things
> > get fixed.)
> > 2. if the page is not a fake head, in other words, it is either true
> > head or tail, no need to fix anything.
> >
> > >
> > > Thanks.

Thanks
barry

Patch

diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
index bdb22006f713..a154a7b3b9a5 100644
--- a/Documentation/admin-guide/kernel-parameters.txt
+++ b/Documentation/admin-guide/kernel-parameters.txt
@@ -1606,7 +1606,7 @@ 
 			[KNL] Reguires CONFIG_HUGETLB_PAGE_FREE_VMEMMAP
 			enabled.
 			Allows heavy hugetlb users to free up some more
-			memory (6 * PAGE_SIZE for each 2MB hugetlb page).
+			memory (7 * PAGE_SIZE for each 2MB hugetlb page).
 			Format: { on | off (default) }
 
 			on:  enable the feature
diff --git a/include/linux/page-flags.h b/include/linux/page-flags.h
index 8e1d97d8f3bd..7b1a918ebd43 100644
--- a/include/linux/page-flags.h
+++ b/include/linux/page-flags.h
@@ -184,13 +184,64 @@  enum pageflags {
 
 #ifndef __GENERATING_BOUNDS_H
 
+#ifdef CONFIG_HUGETLB_PAGE_FREE_VMEMMAP
+extern bool hugetlb_free_vmemmap_enabled;
+
+/*
+ * If the feature of freeing some vmemmap pages associated with each HugeTLB
+ * page is enabled, the head vmemmap page frame is reused and all of the tail
+ * vmemmap addresses map to the head vmemmap page frame (furture details can
+ * refer to the figure at the head of the mm/hugetlb_vmemmap.c).  In other
+ * word, there are more than one page struct with PG_head associated with each
+ * HugeTLB page.  We __know__ that there is only one head page struct, the tail
+ * page structs with PG_head are fake head page structs.  We need an approach
+ * to distinguish between those two different types of page structs so that
+ * compound_head() can return the real head page struct when the parameter is
+ * the tail page struct but with PG_head.
+ *
+ * The page_head_if_fake() returns the real head page struct iff the @page may
+ * be fake, otherwise, returns the @page if it cannot be a fake page struct.
+ */
+static __always_inline const struct page *page_head_if_fake(const struct page *page)
+{
+	if (!hugetlb_free_vmemmap_enabled)
+		return page;
+
+	/*
+	 * Only addresses aligned with PAGE_SIZE of struct page may be fake head
+	 * struct page. The alignment check aims to avoid access the fields (
+	 * e.g. compound_head) of the @page[1]. It can avoid touch a (possibly)
+	 * cold cacheline in some cases.
+	 */
+	if (IS_ALIGNED((unsigned long)page, PAGE_SIZE) &&
+	    test_bit(PG_head, &page->flags)) {
+		/*
+		 * We can safely access the field of the @page[1] with PG_head
+		 * because the @page is a compound page composed with at least
+		 * two contiguous pages.
+		 */
+		unsigned long head = READ_ONCE(page[1].compound_head);
+
+		if (likely(head & 1))
+			return (const struct page *)(head - 1);
+	}
+
+	return page;
+}
+#else
+static __always_inline const struct page *page_head_if_fake(const struct page *page)
+{
+	return page;
+}
+#endif
+
 static inline unsigned long _compound_head(const struct page *page)
 {
 	unsigned long head = READ_ONCE(page->compound_head);
 
 	if (unlikely(head & 1))
 		return head - 1;
-	return (unsigned long)page;
+	return (unsigned long)page_head_if_fake(page);
 }
 
 #define compound_head(page)	((typeof(page))_compound_head(page))
@@ -225,12 +276,14 @@  static inline unsigned long _compound_head(const struct page *page)
 
 static __always_inline int PageTail(struct page *page)
 {
-	return READ_ONCE(page->compound_head) & 1;
+	return READ_ONCE(page->compound_head) & 1 ||
+	       page_head_if_fake(page) != page;
 }
 
 static __always_inline int PageCompound(struct page *page)
 {
-	return test_bit(PG_head, &page->flags) || PageTail(page);
+	return test_bit(PG_head, &page->flags) ||
+	       READ_ONCE(page->compound_head) & 1;
 }
 
 #define	PAGE_POISON_PATTERN	-1l
@@ -675,7 +728,21 @@  static inline bool test_set_page_writeback(struct page *page)
 	return set_page_writeback(page);
 }
 
-__PAGEFLAG(Head, head, PF_ANY) CLEARPAGEFLAG(Head, head, PF_ANY)
+static __always_inline bool folio_test_head(struct folio *folio)
+{
+	return test_bit(PG_head, folio_flags(folio, FOLIO_PF_ANY));
+}
+
+static __always_inline int PageHead(struct page *page)
+{
+	PF_POISONED_CHECK(page);
+	return test_bit(PG_head, &page->flags) &&
+	       page_head_if_fake(page) == page;
+}
+
+__SETPAGEFLAG(Head, head, PF_ANY)
+__CLEARPAGEFLAG(Head, head, PF_ANY)
+CLEARPAGEFLAG(Head, head, PF_ANY)
 
 /* Whether there are one or multiple pages in a folio */
 static inline bool folio_single(struct folio *folio)
diff --git a/mm/hugetlb_vmemmap.c b/mm/hugetlb_vmemmap.c
index c540c21e26f5..527bcaa44a48 100644
--- a/mm/hugetlb_vmemmap.c
+++ b/mm/hugetlb_vmemmap.c
@@ -124,9 +124,9 @@ 
  * page of page structs (page 0) associated with the HugeTLB page contains the 4
  * page structs necessary to describe the HugeTLB. The only use of the remaining
  * pages of page structs (page 1 to page 7) is to point to page->compound_head.
- * Therefore, we can remap pages 2 to 7 to page 1. Only 2 pages of page structs
+ * Therefore, we can remap pages 1 to 7 to page 0. Only 1 pages of page structs
  * will be used for each HugeTLB page. This will allow us to free the remaining
- * 6 pages to the buddy allocator.
+ * 7 pages to the buddy allocator.
  *
  * Here is how things look after remapping.
  *
@@ -134,30 +134,30 @@ 
  * +-----------+ ---virt_to_page---> +-----------+   mapping to   +-----------+
  * |           |                     |     0     | -------------> |     0     |
  * |           |                     +-----------+                +-----------+
- * |           |                     |     1     | -------------> |     1     |
- * |           |                     +-----------+                +-----------+
- * |           |                     |     2     | ----------------^ ^ ^ ^ ^ ^
- * |           |                     +-----------+                   | | | | |
- * |           |                     |     3     | ------------------+ | | | |
- * |           |                     +-----------+                     | | | |
- * |           |                     |     4     | --------------------+ | | |
- * |    PMD    |                     +-----------+                       | | |
- * |   level   |                     |     5     | ----------------------+ | |
- * |  mapping  |                     +-----------+                         | |
- * |           |                     |     6     | ------------------------+ |
- * |           |                     +-----------+                           |
- * |           |                     |     7     | --------------------------+
+ * |           |                     |     1     | ---------------^ ^ ^ ^ ^ ^ ^
+ * |           |                     +-----------+                  | | | | | |
+ * |           |                     |     2     | -----------------+ | | | | |
+ * |           |                     +-----------+                    | | | | |
+ * |           |                     |     3     | -------------------+ | | | |
+ * |           |                     +-----------+                      | | | |
+ * |           |                     |     4     | ---------------------+ | | |
+ * |    PMD    |                     +-----------+                        | | |
+ * |   level   |                     |     5     | -----------------------+ | |
+ * |  mapping  |                     +-----------+                          | |
+ * |           |                     |     6     | -------------------------+ |
+ * |           |                     +-----------+                            |
+ * |           |                     |     7     | ---------------------------+
  * |           |                     +-----------+
  * |           |
  * |           |
  * |           |
  * +-----------+
  *
- * When a HugeTLB is freed to the buddy system, we should allocate 6 pages for
+ * When a HugeTLB is freed to the buddy system, we should allocate 7 pages for
  * vmemmap pages and restore the previous mapping relationship.
  *
  * For the HugeTLB page of the pud level mapping. It is similar to the former.
- * We also can use this approach to free (PAGE_SIZE - 2) vmemmap pages.
+ * We also can use this approach to free (PAGE_SIZE - 1) vmemmap pages.
  *
  * Apart from the HugeTLB page of the pmd/pud level mapping, some architectures
  * (e.g. aarch64) provides a contiguous bit in the translation table entries
@@ -166,7 +166,13 @@ 
  *
  * The contiguous bit is used to increase the mapping size at the pmd and pte
  * (last) level. So this type of HugeTLB page can be optimized only when its
- * size of the struct page structs is greater than 2 pages.
+ * size of the struct page structs is greater than 1 pages.
+ *
+ * Notice: The head vmemmap page is not freed to the buddy allocator and all
+ * tail vmemmap pages are mapped to the head vmemmap page frame. So we can see
+ * more than one struct page struct with PG_head (e.g. 8 per 2 MB HugeTLB page)
+ * associated with each HugeTLB page. The compound_head() can handle this
+ * correctly (more details refer to the comment above compound_head()).
  */
 #define pr_fmt(fmt)	"HugeTLB: " fmt
 
@@ -175,14 +181,16 @@ 
 /*
  * There are a lot of struct page structures associated with each HugeTLB page.
  * For tail pages, the value of compound_head is the same. So we can reuse first
- * page of tail page structures. We map the virtual addresses of the remaining
- * pages of tail page structures to the first tail page struct, and then free
- * these page frames. Therefore, we need to reserve two pages as vmemmap areas.
+ * page of head page structures. We map the virtual addresses of all the pages
+ * of tail page structures to the head page struct, and then free these page
+ * frames. Therefore, we need to reserve one pages as vmemmap areas.
  */
-#define RESERVE_VMEMMAP_NR		2U
+#define RESERVE_VMEMMAP_NR		1U
 #define RESERVE_VMEMMAP_SIZE		(RESERVE_VMEMMAP_NR << PAGE_SHIFT)
 
-bool hugetlb_free_vmemmap_enabled = IS_ENABLED(CONFIG_HUGETLB_PAGE_FREE_VMEMMAP_DEFAULT_ON);
+bool hugetlb_free_vmemmap_enabled __read_mostly =
+	IS_ENABLED(CONFIG_HUGETLB_PAGE_FREE_VMEMMAP_DEFAULT_ON);
+EXPORT_SYMBOL(hugetlb_free_vmemmap_enabled);
 
 static int __init early_hugetlb_free_vmemmap_param(char *buf)
 {
@@ -236,7 +244,6 @@  int alloc_huge_page_vmemmap(struct hstate *h, struct page *head)
 	 */
 	ret = vmemmap_remap_alloc(vmemmap_addr, vmemmap_end, vmemmap_reuse,
 				  GFP_KERNEL | __GFP_NORETRY | __GFP_THISNODE);
-
 	if (!ret)
 		ClearHPageVmemmapOptimized(head);
 
@@ -282,9 +289,8 @@  void __init hugetlb_vmemmap_init(struct hstate *h)
 
 	vmemmap_pages = (nr_pages * sizeof(struct page)) >> PAGE_SHIFT;
 	/*
-	 * The head page and the first tail page are not to be freed to buddy
-	 * allocator, the other pages will map to the first tail page, so they
-	 * can be freed.
+	 * The head page is not to be freed to buddy allocator, the other tail
+	 * pages will map to the head page, so they can be freed.
 	 *
 	 * Could RESERVE_VMEMMAP_NR be greater than @vmemmap_pages? It is true
 	 * on some architectures (e.g. aarch64). See Documentation/arm64/
diff --git a/mm/sparse-vmemmap.c b/mm/sparse-vmemmap.c
index bdce883f9286..62e3d20648ce 100644
--- a/mm/sparse-vmemmap.c
+++ b/mm/sparse-vmemmap.c
@@ -53,6 +53,17 @@  struct vmemmap_remap_walk {
 	struct list_head *vmemmap_pages;
 };
 
+/*
+ * How many struct page structs need to be reset. When we reuse the head
+ * struct page, the special metadata (e.g. page->flags or page->mapping)
+ * cannot copy to the tail struct page structs. The invalid value will be
+ * checked in the free_tail_pages_check(). In order to avoid the message
+ * of "corrupted mapping in tail page". We need to reset at least 3 (one
+ * head struct page struct and two tail struct page structs) struct page
+ * structs.
+ */
+#define NR_RESET_STRUCT_PAGE		3
+
 static int split_vmemmap_huge_pmd(pmd_t *pmd, unsigned long start,
 				  struct vmemmap_remap_walk *walk)
 {
@@ -245,6 +256,15 @@  static void vmemmap_remap_pte(pte_t *pte, unsigned long addr,
 	set_pte_at(&init_mm, addr, pte, entry);
 }
 
+static inline void reset_struct_pages(struct page *start)
+{
+	int i;
+	struct page *from = start + NR_RESET_STRUCT_PAGE;
+
+	for (i = 0; i < NR_RESET_STRUCT_PAGE; i++)
+		memcpy(start + i, from, sizeof(*from));
+}
+
 static void vmemmap_restore_pte(pte_t *pte, unsigned long addr,
 				struct vmemmap_remap_walk *walk)
 {
@@ -258,6 +278,7 @@  static void vmemmap_restore_pte(pte_t *pte, unsigned long addr,
 	list_del(&page->lru);
 	to = page_to_virt(page);
 	copy_page(to, (void *)walk->reuse_addr);
+	reset_struct_pages(to);
 
 	set_pte_at(&init_mm, addr, pte, mk_pte(page, pgprot));
 }