git.openwrt.org Git - openwrt/staging/ldir.git/blob

   1 From 397624e12244ec038f51cb1f178ccb7a2ec562e5 Mon Sep 17 00:00:00 2001
   2 From: "T.J. Alumbaugh" <talumbau@google.com>
   3 Date: Wed, 18 Jan 2023 00:18:23 +0000
   4 Subject: [PATCH 14/19] UPSTREAM: mm: multi-gen LRU: section for Bloom filters
   5
   6 Move Bloom filters code into a dedicated section.  Improve the design doc
   7 to explain Bloom filter usage and connection between aging and eviction in
   8 their use.
   9
  10 Link: https://lkml.kernel.org/r/20230118001827.1040870-4-talumbau@google.com
  11 Change-Id: I73e866f687c1ed9f5c8538086aa39408b79897db
  12 Signed-off-by: T.J. Alumbaugh <talumbau@google.com>
  13 Cc: Yu Zhao <yuzhao@google.com>
  14 Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
  15 (cherry picked from commit ccbbbb85945d8f0255aa9dbc1b617017e2294f2c)
  16 Bug: 274865848
  17 Signed-off-by: T.J. Mercier <tjmercier@google.com>
  18 ---
  19  Documentation/mm/multigen_lru.rst |  16 +++
  20  mm/vmscan.c                       | 180 +++++++++++++++---------------
  21  2 files changed, 108 insertions(+), 88 deletions(-)
  22
  23 --- a/Documentation/mm/multigen_lru.rst
  24 +++ b/Documentation/mm/multigen_lru.rst
  25 @@ -170,6 +170,22 @@ promotes hot pages. If the scan was done
  26  adds the PMD entry pointing to the PTE table to the Bloom filter. This
  27  forms a feedback loop between the eviction and the aging.
  28
  29 +Bloom Filters
  30 +-------------
  31 +Bloom filters are a space and memory efficient data structure for set
  32 +membership test, i.e., test if an element is not in the set or may be
  33 +in the set.
  34 +
  35 +In the eviction path, specifically, in ``lru_gen_look_around()``, if a
  36 +PMD has a sufficient number of hot pages, its address is placed in the
  37 +filter. In the aging path, set membership means that the PTE range
  38 +will be scanned for young pages.
  39 +
  40 +Note that Bloom filters are probabilistic on set membership. If a test
  41 +is false positive, the cost is an additional scan of a range of PTEs,
  42 +which may yield hot pages anyway. Parameters of the filter itself can
  43 +control the false positive rate in the limit.
  44 +
  45  Summary
  46  -------
  47  The multi-gen LRU can be disassembled into the following parts:
  48 --- a/mm/vmscan.c
  49 +++ b/mm/vmscan.c
  50 @@ -3209,6 +3209,98 @@ static bool __maybe_unused seq_is_valid(
  51  }
  52
  53  /******************************************************************************
  54 + *                          Bloom filters
  55 + ******************************************************************************/
  56 +
  57 +/*
  58 + * Bloom filters with m=1<<15, k=2 and the false positive rates of ~1/5 when
  59 + * n=10,000 and ~1/2 when n=20,000, where, conventionally, m is the number of
  60 + * bits in a bitmap, k is the number of hash functions and n is the number of
  61 + * inserted items.
  62 + *
  63 + * Page table walkers use one of the two filters to reduce their search space.
  64 + * To get rid of non-leaf entries that no longer have enough leaf entries, the
  65 + * aging uses the double-buffering technique to flip to the other filter each
  66 + * time it produces a new generation. For non-leaf entries that have enough
  67 + * leaf entries, the aging carries them over to the next generation in
  68 + * walk_pmd_range(); the eviction also report them when walking the rmap
  69 + * in lru_gen_look_around().
  70 + *
  71 + * For future optimizations:
  72 + * 1. It's not necessary to keep both filters all the time. The spare one can be
  73 + *    freed after the RCU grace period and reallocated if needed again.
  74 + * 2. And when reallocating, it's worth scaling its size according to the number
  75 + *    of inserted entries in the other filter, to reduce the memory overhead on
  76 + *    small systems and false positives on large systems.
  77 + * 3. Jenkins' hash function is an alternative to Knuth's.
  78 + */
  79 +#define BLOOM_FILTER_SHIFT     15
  80 +
  81 +static inline int filter_gen_from_seq(unsigned long seq)
  82 +{
  83 +       return seq % NR_BLOOM_FILTERS;
  84 +}
  85 +
  86 +static void get_item_key(void *item, int *key)
  87 +{
  88 +       u32 hash = hash_ptr(item, BLOOM_FILTER_SHIFT * 2);
  89 +
  90 +       BUILD_BUG_ON(BLOOM_FILTER_SHIFT * 2 > BITS_PER_TYPE(u32));
  91 +
  92 +       key[0] = hash & (BIT(BLOOM_FILTER_SHIFT) - 1);
  93 +       key[1] = hash >> BLOOM_FILTER_SHIFT;
  94 +}
  95 +
  96 +static bool test_bloom_filter(struct lruvec *lruvec, unsigned long seq, void *item)
  97 +{
  98 +       int key[2];
  99 +       unsigned long *filter;
 100 +       int gen = filter_gen_from_seq(seq);
 101 +
 102 +       filter = READ_ONCE(lruvec->mm_state.filters[gen]);
 103 +       if (!filter)
 104 +               return true;
 105 +
 106 +       get_item_key(item, key);
 107 +
 108 +       return test_bit(key[0], filter) && test_bit(key[1], filter);
 109 +}
 110 +
 111 +static void update_bloom_filter(struct lruvec *lruvec, unsigned long seq, void *item)
 112 +{
 113 +       int key[2];
 114 +       unsigned long *filter;
 115 +       int gen = filter_gen_from_seq(seq);
 116 +
 117 +       filter = READ_ONCE(lruvec->mm_state.filters[gen]);
 118 +       if (!filter)
 119 +               return;
 120 +
 121 +       get_item_key(item, key);
 122 +
 123 +       if (!test_bit(key[0], filter))
 124 +               set_bit(key[0], filter);
 125 +       if (!test_bit(key[1], filter))
 126 +               set_bit(key[1], filter);
 127 +}
 128 +
 129 +static void reset_bloom_filter(struct lruvec *lruvec, unsigned long seq)
 130 +{
 131 +       unsigned long *filter;
 132 +       int gen = filter_gen_from_seq(seq);
 133 +
 134 +       filter = lruvec->mm_state.filters[gen];
 135 +       if (filter) {
 136 +               bitmap_clear(filter, 0, BIT(BLOOM_FILTER_SHIFT));
 137 +               return;
 138 +       }
 139 +
 140 +       filter = bitmap_zalloc(BIT(BLOOM_FILTER_SHIFT),
 141 +                              __GFP_HIGH | __GFP_NOMEMALLOC | __GFP_NOWARN);
 142 +       WRITE_ONCE(lruvec->mm_state.filters[gen], filter);
 143 +}
 144 +
 145 +/******************************************************************************
 146   *                          mm_struct list
 147   ******************************************************************************/
 148
 149 @@ -3333,94 +3425,6 @@ void lru_gen_migrate_mm(struct mm_struct
 150  }
 151  #endif
 152
 153 -/*
 154 - * Bloom filters with m=1<<15, k=2 and the false positive rates of ~1/5 when
 155 - * n=10,000 and ~1/2 when n=20,000, where, conventionally, m is the number of
 156 - * bits in a bitmap, k is the number of hash functions and n is the number of
 157 - * inserted items.
 158 - *
 159 - * Page table walkers use one of the two filters to reduce their search space.
 160 - * To get rid of non-leaf entries that no longer have enough leaf entries, the
 161 - * aging uses the double-buffering technique to flip to the other filter each
 162 - * time it produces a new generation. For non-leaf entries that have enough
 163 - * leaf entries, the aging carries them over to the next generation in
 164 - * walk_pmd_range(); the eviction also report them when walking the rmap
 165 - * in lru_gen_look_around().
 166 - *
 167 - * For future optimizations:
 168 - * 1. It's not necessary to keep both filters all the time. The spare one can be
 169 - *    freed after the RCU grace period and reallocated if needed again.
 170 - * 2. And when reallocating, it's worth scaling its size according to the number
 171 - *    of inserted entries in the other filter, to reduce the memory overhead on
 172 - *    small systems and false positives on large systems.
 173 - * 3. Jenkins' hash function is an alternative to Knuth's.
 174 - */
 175 -#define BLOOM_FILTER_SHIFT     15
 176 -
 177 -static inline int filter_gen_from_seq(unsigned long seq)
 178 -{
 179 -       return seq % NR_BLOOM_FILTERS;
 180 -}
 181 -
 182 -static void get_item_key(void *item, int *key)
 183 -{
 184 -       u32 hash = hash_ptr(item, BLOOM_FILTER_SHIFT * 2);
 185 -
 186 -       BUILD_BUG_ON(BLOOM_FILTER_SHIFT * 2 > BITS_PER_TYPE(u32));
 187 -
 188 -       key[0] = hash & (BIT(BLOOM_FILTER_SHIFT) - 1);
 189 -       key[1] = hash >> BLOOM_FILTER_SHIFT;
 190 -}
 191 -
 192 -static void reset_bloom_filter(struct lruvec *lruvec, unsigned long seq)
 193 -{
 194 -       unsigned long *filter;
 195 -       int gen = filter_gen_from_seq(seq);
 196 -
 197 -       filter = lruvec->mm_state.filters[gen];
 198 -       if (filter) {
 199 -               bitmap_clear(filter, 0, BIT(BLOOM_FILTER_SHIFT));
 200 -               return;
 201 -       }
 202 -
 203 -       filter = bitmap_zalloc(BIT(BLOOM_FILTER_SHIFT),
 204 -                              __GFP_HIGH | __GFP_NOMEMALLOC | __GFP_NOWARN);
 205 -       WRITE_ONCE(lruvec->mm_state.filters[gen], filter);
 206 -}
 207 -
 208 -static void update_bloom_filter(struct lruvec *lruvec, unsigned long seq, void *item)
 209 -{
 210 -       int key[2];
 211 -       unsigned long *filter;
 212 -       int gen = filter_gen_from_seq(seq);
 213 -
 214 -       filter = READ_ONCE(lruvec->mm_state.filters[gen]);
 215 -       if (!filter)
 216 -               return;
 217 -
 218 -       get_item_key(item, key);
 219 -
 220 -       if (!test_bit(key[0], filter))
 221 -               set_bit(key[0], filter);
 222 -       if (!test_bit(key[1], filter))
 223 -               set_bit(key[1], filter);
 224 -}
 225 -
 226 -static bool test_bloom_filter(struct lruvec *lruvec, unsigned long seq, void *item)
 227 -{
 228 -       int key[2];
 229 -       unsigned long *filter;
 230 -       int gen = filter_gen_from_seq(seq);
 231 -
 232 -       filter = READ_ONCE(lruvec->mm_state.filters[gen]);
 233 -       if (!filter)
 234 -               return true;
 235 -
 236 -       get_item_key(item, key);
 237 -
 238 -       return test_bit(key[0], filter) && test_bit(key[1], filter);
 239 -}
 240 -
 241  static void reset_mm_stats(struct lruvec *lruvec, struct lru_gen_mm_walk *walk, bool last)
 242  {
 243         int i;