Message-ID: <CAJuCfpE0tJfq8juxD+jeStnhQ2PTUH6DqjL7AP_E+Pw++8L35w@mail.gmail.com>
Date: Thu, 28 Nov 2024 11:53:28 -0800
From: Suren Baghdasaryan <surenb@...gle.com>
To: David Wang <00107082@....com>
Cc: kent.overstreet@...ux.dev, yuzhao@...gle.com, akpm@...ux-foundation.org, 
	linux-mm@...ck.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH] mm/codetag: swap tags when migrate pages

On Thu, Nov 28, 2024 at 2:26 AM David Wang <00107082@....com> wrote:
>
> The initial solution for codetag adjustment during page migration
> uses three kinds of low level plumbings, those steps can be replaced
> by swapping tags, which only needs one kind of low level plumbing,
> and code is more clear.

This description does not explain the real issue. I would suggest
something like:

Current solution to adjust codetag references during page migration is
done in 3 steps:
1. sets the codetag reference of the old page as empty (not pointing
to any codetag);
2. subtracts counters of the new page to compensate for its own allocation;
3. sets codetag reference of the new page to point to the codetag of
the old page.
This does not work if CONFIG_MEM_ALLOC_PROFILING_DEBUG=n because
set_codetag_empty() becomes NOOP. Instead, let's simply swap codetag
references so that the new page is referencing the old codetag and the
old page is referencing the new codetag. This way accounting stays
valid and the logic makes more sense.

Fixes: e0a955bf7f61 ("mm/codetag: add pgalloc_tag_copy()")

>
> Signed-off-by: David Wang <00107082@....com>
> Link: https://lore.kernel.org/lkml/20241124074318.399027-1-00107082@163.com/
 This above Link: seems unusual. Maybe uses Closes instead like this:

Closed: https://lore.kernel.org/lkml/20241124074318.399027-1-00107082@163.com/

> ---
>  include/linux/pgalloc_tag.h |  4 ++--
>  lib/alloc_tag.c             | 35 +++++++++++++++++++----------------
>  mm/migrate.c                |  2 +-
>  3 files changed, 22 insertions(+), 19 deletions(-)
>
> diff --git a/include/linux/pgalloc_tag.h b/include/linux/pgalloc_tag.h
> index 0e43ab653ab6..3469c4b20105 100644
> --- a/include/linux/pgalloc_tag.h
> +++ b/include/linux/pgalloc_tag.h
> @@ -231,7 +231,7 @@ static inline void pgalloc_tag_sub_pages(struct alloc_tag *tag, unsigned int nr)
>  }
>
>  void pgalloc_tag_split(struct folio *folio, int old_order, int new_order);
> -void pgalloc_tag_copy(struct folio *new, struct folio *old);
> +void pgalloc_tag_swap(struct folio *new, struct folio *old);
>
>  void __init alloc_tag_sec_init(void);
>
> @@ -245,7 +245,7 @@ static inline struct alloc_tag *pgalloc_tag_get(struct page *page) { return NULL
>  static inline void pgalloc_tag_sub_pages(struct alloc_tag *tag, unsigned int nr) {}
>  static inline void alloc_tag_sec_init(void) {}
>  static inline void pgalloc_tag_split(struct folio *folio, int old_order, int new_order) {}
> -static inline void pgalloc_tag_copy(struct folio *new, struct folio *old) {}
> +static inline void pgalloc_tag_swap(struct folio *new, struct folio *old) {}
>
>  #endif /* CONFIG_MEM_ALLOC_PROFILING */
>
> diff --git a/lib/alloc_tag.c b/lib/alloc_tag.c
> index 2414a7ee7ec7..b45efde50c40 100644
> --- a/lib/alloc_tag.c
> +++ b/lib/alloc_tag.c
> @@ -189,26 +189,29 @@ void pgalloc_tag_split(struct folio *folio, int old_order, int new_order)
>         }
>  }
>
> -void pgalloc_tag_copy(struct folio *new, struct folio *old)
> +void pgalloc_tag_swap(struct folio *new, struct folio *old)
>  {
> -       union pgtag_ref_handle handle;
> -       union codetag_ref ref;
> -       struct alloc_tag *tag;
> +       union pgtag_ref_handle handles[2];
> +       union codetag_ref refs[2];
> +       struct alloc_tag *tags[2];
> +       struct folio *folios[2] = {new, old};
> +       int i;
>
> -       tag = pgalloc_tag_get(&old->page);
> -       if (!tag)
> -               return;
> +       for (i = 0; i < 2; i++) {
> +               tags[i] = pgalloc_tag_get(&folios[i]->page);
> +               if (!tags[i])
> +                       return;
> +               if (!get_page_tag_ref(&folios[i]->page, &refs[i], &handles[i]))
> +                       return;

If any of the above getters fail on the second cycle, you will miss
calling put_page_tag_ref(handles[0]) and releasing the page_tag_ref
you obtained on the first cycle. It might be cleaner to drop the use
of arrays and use new/old.

> +       }
>
> -       if (!get_page_tag_ref(&new->page, &ref, &handle))
> -               return;
> +       swap(tags[0], tags[1]);
>
> -       /* Clear the old ref to the original allocation tag. */
> -       clear_page_tag_ref(&old->page);
> -       /* Decrement the counters of the tag on get_new_folio. */
> -       alloc_tag_sub(&ref, folio_size(new));
> -       __alloc_tag_ref_set(&ref, tag);
> -       update_page_tag_ref(handle, &ref);
> -       put_page_tag_ref(handle);
> +       for (i = 0; i < 2; i++) {
> +               __alloc_tag_ref_set(&refs[i], tags[i]);
> +               update_page_tag_ref(handles[i], &refs[i]);
> +               put_page_tag_ref(handles[i]);
> +       }
>  }
>
>  static void shutdown_mem_profiling(bool remove_file)
> diff --git a/mm/migrate.c b/mm/migrate.c
> index 2ce6b4b814df..cc68583c86f9 100644
> --- a/mm/migrate.c
> +++ b/mm/migrate.c
> @@ -745,7 +745,7 @@ void folio_migrate_flags(struct folio *newfolio, struct folio *folio)
>                 folio_set_readahead(newfolio);
>
>         folio_copy_owner(newfolio, folio);
> -       pgalloc_tag_copy(newfolio, folio);
> +       pgalloc_tag_swap(newfolio, folio);
>
>         mem_cgroup_migrate(folio, newfolio);
>  }
> --
> 2.39.2
>

Powered by blists - more mailing lists