git.ipfire.org Git - thirdparty/kernel/stable-queue.git/blob

   1 From 0e0e9bd5f7b9d40fd03b70092367247d52da1db0 Mon Sep 17 00:00:00 2001
   2 From: Yin Fengwei <fengwei.yin@intel.com>
   3 Date: Tue, 8 Aug 2023 10:09:17 +0800
   4 Subject: madvise:madvise_free_pte_range(): don't use mapcount() against large folio for sharing check
   5
   6 From: Yin Fengwei <fengwei.yin@intel.com>
   7
   8 commit 0e0e9bd5f7b9d40fd03b70092367247d52da1db0 upstream.
   9
  10 Commit 98b211d6415f ("madvise: convert madvise_free_pte_range() to use a
  11 folio") replaced the page_mapcount() with folio_mapcount() to check
  12 whether the folio is shared by other mapping.
  13
  14 It's not correct for large folios. folio_mapcount() returns the total
  15 mapcount of large folio which is not suitable to detect whether the folio
  16 is shared.
  17
  18 Use folio_estimated_sharers() which returns a estimated number of shares.
  19 That means it's not 100% correct. It should be OK for madvise case here.
  20
  21 User-visible effects is that the THP is skipped when user call madvise.
  22 But the correct behavior is THP should be split and processed then.
  23
  24 NOTE: this change is a temporary fix to reduce the user-visible effects
  25 before the long term fix from David is ready.
  26
  27 Link: https://lkml.kernel.org/r/20230808020917.2230692-4-fengwei.yin@intel.com
  28 Fixes: 98b211d6415f ("madvise: convert madvise_free_pte_range() to use a folio")
  29 Signed-off-by: Yin Fengwei <fengwei.yin@intel.com>
  30 Reviewed-by: Yu Zhao <yuzhao@google.com>
  31 Reviewed-by: Ryan Roberts <ryan.roberts@arm.com>
  32 Cc: David Hildenbrand <david@redhat.com>
  33 Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
  34 Cc: Matthew Wilcox <willy@infradead.org>
  35 Cc: Minchan Kim <minchan@kernel.org>
  36 Cc: Vishal Moola (Oracle) <vishal.moola@gmail.com>
  37 Cc: Yang Shi <shy828301@gmail.com>
  38 Cc: <stable@vger.kernel.org>
  39 Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
  40 Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
  41 ---
  42  include/linux/mm.h |   19 +++++++++++++++++++
  43  mm/madvise.c       |    4 ++--
  44  2 files changed, 21 insertions(+), 2 deletions(-)
  45
  46 --- a/include/linux/mm.h
  47 +++ b/include/linux/mm.h
  48 @@ -1727,6 +1727,25 @@ static inline size_t folio_size(struct f
  49         return PAGE_SIZE << folio_order(folio);
  50  }
  51
  52 +/**
  53 + * folio_estimated_sharers - Estimate the number of sharers of a folio.
  54 + * @folio: The folio.
  55 + *
  56 + * folio_estimated_sharers() aims to serve as a function to efficiently
  57 + * estimate the number of processes sharing a folio. This is done by
  58 + * looking at the precise mapcount of the first subpage in the folio, and
  59 + * assuming the other subpages are the same. This may not be true for large
  60 + * folios. If you want exact mapcounts for exact calculations, look at
  61 + * page_mapcount() or folio_total_mapcount().
  62 + *
  63 + * Return: The estimated number of processes sharing a folio.
  64 + */
  65 +static inline int folio_estimated_sharers(struct folio *folio)
  66 +{
  67 +       return page_mapcount(folio_page(folio, 0));
  68 +}
  69 +
  70 +
  71  #ifndef HAVE_ARCH_MAKE_PAGE_ACCESSIBLE
  72  static inline int arch_make_page_accessible(struct page *page)
  73  {
  74 --- a/mm/madvise.c
  75 +++ b/mm/madvise.c
  76 @@ -654,8 +654,8 @@ static int madvise_free_pte_range(pmd_t
  77                  * deactivate all pages.
  78                  */
  79                 if (folio_test_large(folio)) {
  80 -                       if (folio_mapcount(folio) != 1)
  81 -                               goto out;
  82 +                       if (folio_estimated_sharers(folio) != 1)
  83 +                               break;
  84                         folio_get(folio);
  85                         if (!folio_trylock(folio)) {
  86                                 folio_put(folio);