target/linux/generic/backport-6.1/020-v6.3-07-BACKPORT-mm-multi-gen-LRU-clarify-scan_control-flags.patch

   1 From 11b14ee8cbbbebd8204609076a9327a1171cd253 Mon Sep 17 00:00:00 2001
   2 From: Yu Zhao <yuzhao@google.com>
   3 Date: Wed, 21 Dec 2022 21:19:05 -0700
   4 Subject: [PATCH 07/19] BACKPORT: mm: multi-gen LRU: clarify scan_control flags
   5
   6 Among the flags in scan_control:
   7 1. sc->may_swap, which indicates swap constraint due to memsw.max, is
   8    supported as usual.
   9 2. sc->proactive, which indicates reclaim by memory.reclaim, may not
  10    opportunistically skip the aging path, since it is considered less
  11    latency sensitive.
  12 3. !(sc->gfp_mask & __GFP_IO), which indicates IO constraint, lowers
  13    swappiness to prioritize file LRU, since clean file folios are more
  14    likely to exist.
  15 4. sc->may_writepage and sc->may_unmap, which indicates opportunistic
  16    reclaim, are rejected, since unmapped clean folios are already
  17    prioritized. Scanning for more of them is likely futile and can
  18    cause high reclaim latency when there is a large number of memcgs.
  19
  20 The rest are handled by the existing code.
  21
  22 Link: https://lkml.kernel.org/r/20221222041905.2431096-8-yuzhao@google.com
  23 Signed-off-by: Yu Zhao <yuzhao@google.com>
  24 Cc: Johannes Weiner <hannes@cmpxchg.org>
  25 Cc: Jonathan Corbet <corbet@lwn.net>
  26 Cc: Michael Larabel <Michael@MichaelLarabel.com>
  27 Cc: Michal Hocko <mhocko@kernel.org>
  28 Cc: Mike Rapoport <rppt@kernel.org>
  29 Cc: Roman Gushchin <roman.gushchin@linux.dev>
  30 Cc: Suren Baghdasaryan <surenb@google.com>
  31 Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
  32 Bug: 274865848
  33 (cherry picked from commit e9d4e1ee788097484606c32122f146d802a9c5fb)
  34 [TJ: Resolved conflict with older function signature for min_cgroup_below_min, and over
  35 cdded861182142ac4488a4d64c571107aeb77f53 ("ANDROID: MGLRU: Don't skip anon reclaim if swap low")]
  36 Change-Id: Ic2e779eaf4e91a3921831b4e2fa10c740dc59d50
  37 Signed-off-by: T.J. Mercier <tjmercier@google.com>
  38 ---
  39  mm/vmscan.c | 55 +++++++++++++++++++++++++++--------------------------
  40  1 file changed, 28 insertions(+), 27 deletions(-)
  41
  42 --- a/mm/vmscan.c
  43 +++ b/mm/vmscan.c
  44 @@ -3185,6 +3185,9 @@ static int get_swappiness(struct lruvec
  45         struct mem_cgroup *memcg = lruvec_memcg(lruvec);
  46         struct pglist_data *pgdat = lruvec_pgdat(lruvec);
  47
  48 +       if (!sc->may_swap)
  49 +               return 0;
  50 +
  51         if (!can_demote(pgdat->node_id, sc) &&
  52             mem_cgroup_get_nr_swap_pages(memcg) < MIN_LRU_BATCH)
  53                 return 0;
  54 @@ -4223,7 +4226,7 @@ static void walk_mm(struct lruvec *lruve
  55         } while (err == -EAGAIN);
  56  }
  57
  58 -static struct lru_gen_mm_walk *set_mm_walk(struct pglist_data *pgdat)
  59 +static struct lru_gen_mm_walk *set_mm_walk(struct pglist_data *pgdat, bool force_alloc)
  60  {
  61         struct lru_gen_mm_walk *walk = current->reclaim_state->mm_walk;
  62
  63 @@ -4231,7 +4234,7 @@ static struct lru_gen_mm_walk *set_mm_wa
  64                 VM_WARN_ON_ONCE(walk);
  65
  66                 walk = &pgdat->mm_walk;
  67 -       } else if (!pgdat && !walk) {
  68 +       } else if (!walk && force_alloc) {
  69                 VM_WARN_ON_ONCE(current_is_kswapd());
  70
  71                 walk = kzalloc(sizeof(*walk), __GFP_HIGH | __GFP_NOMEMALLOC | __GFP_NOWARN);
  72 @@ -4419,7 +4422,7 @@ static bool try_to_inc_max_seq(struct lr
  73                 goto done;
  74         }
  75
  76 -       walk = set_mm_walk(NULL);
  77 +       walk = set_mm_walk(NULL, true);
  78         if (!walk) {
  79                 success = iterate_mm_list_nowalk(lruvec, max_seq);
  80                 goto done;
  81 @@ -4488,8 +4491,6 @@ static bool lruvec_is_reclaimable(struct
  82         struct mem_cgroup *memcg = lruvec_memcg(lruvec);
  83         DEFINE_MIN_SEQ(lruvec);
  84
  85 -       VM_WARN_ON_ONCE(sc->memcg_low_reclaim);
  86 -
  87         /* see the comment on lru_gen_folio */
  88         gen = lru_gen_from_seq(min_seq[LRU_GEN_FILE]);
  89         birth = READ_ONCE(lruvec->lrugen.timestamps[gen]);
  90 @@ -4753,12 +4754,8 @@ static bool isolate_folio(struct lruvec
  91  {
  92         bool success;
  93
  94 -       /* unmapping inhibited */
  95 -       if (!sc->may_unmap && folio_mapped(folio))
  96 -               return false;
  97 -
  98         /* swapping inhibited */
  99 -       if (!(sc->may_writepage && (sc->gfp_mask & __GFP_IO)) &&
 100 +       if (!(sc->gfp_mask & __GFP_IO) &&
 101             (folio_test_dirty(folio) ||
 102              (folio_test_anon(folio) && !folio_test_swapcache(folio))))
 103                 return false;
 104 @@ -4857,9 +4854,8 @@ static int scan_folios(struct lruvec *lr
 105         __count_vm_events(PGSCAN_ANON + type, isolated);
 106
 107         /*
 108 -        * There might not be eligible pages due to reclaim_idx, may_unmap and
 109 -        * may_writepage. Check the remaining to prevent livelock if it's not
 110 -        * making progress.
 111 +        * There might not be eligible folios due to reclaim_idx. Check the
 112 +        * remaining to prevent livelock if it's not making progress.
 113          */
 114         return isolated || !remaining ? scanned : 0;
 115  }
 116 @@ -5119,8 +5115,7 @@ static long get_nr_to_scan(struct lruvec
 117         struct mem_cgroup *memcg = lruvec_memcg(lruvec);
 118         DEFINE_MAX_SEQ(lruvec);
 119
 120 -       if (mem_cgroup_below_min(memcg) ||
 121 -           (mem_cgroup_below_low(memcg) && !sc->memcg_low_reclaim))
 122 +       if (mem_cgroup_below_min(memcg))
 123                 return 0;
 124
 125         if (!should_run_aging(lruvec, max_seq, sc, can_swap, &nr_to_scan))
 126 @@ -5148,17 +5143,14 @@ static bool try_to_shrink_lruvec(struct
 127         long nr_to_scan;
 128         unsigned long scanned = 0;
 129         unsigned long nr_to_reclaim = get_nr_to_reclaim(sc);
 130 +       int swappiness = get_swappiness(lruvec, sc);
 131 +
 132 +       /* clean file folios are more likely to exist */
 133 +       if (swappiness && !(sc->gfp_mask & __GFP_IO))
 134 +               swappiness = 1;
 135
 136         while (true) {
 137                 int delta;
 138 -               int swappiness;
 139 -
 140 -               if (sc->may_swap)
 141 -                       swappiness = get_swappiness(lruvec, sc);
 142 -               else if (!cgroup_reclaim(sc) && get_swappiness(lruvec, sc))
 143 -                       swappiness = 1;
 144 -               else
 145 -                       swappiness = 0;
 146
 147                 nr_to_scan = get_nr_to_scan(lruvec, sc, swappiness);
 148                 if (nr_to_scan <= 0)
 149 @@ -5289,12 +5281,13 @@ static void lru_gen_shrink_lruvec(struct
 150         struct blk_plug plug;
 151
 152         VM_WARN_ON_ONCE(global_reclaim(sc));
 153 +       VM_WARN_ON_ONCE(!sc->may_writepage || !sc->may_unmap);
 154
 155         lru_add_drain();
 156
 157         blk_start_plug(&plug);
 158
 159 -       set_mm_walk(lruvec_pgdat(lruvec));
 160 +       set_mm_walk(NULL, sc->proactive);
 161
 162         if (try_to_shrink_lruvec(lruvec, sc))
 163                 lru_gen_rotate_memcg(lruvec, MEMCG_LRU_YOUNG);
 164 @@ -5350,11 +5343,19 @@ static void lru_gen_shrink_node(struct p
 165
 166         VM_WARN_ON_ONCE(!global_reclaim(sc));
 167
 168 +       /*
 169 +        * Unmapped clean folios are already prioritized. Scanning for more of
 170 +        * them is likely futile and can cause high reclaim latency when there
 171 +        * is a large number of memcgs.
 172 +        */
 173 +       if (!sc->may_writepage || !sc->may_unmap)
 174 +               goto done;
 175 +
 176         lru_add_drain();
 177
 178         blk_start_plug(&plug);
 179
 180 -       set_mm_walk(pgdat);
 181 +       set_mm_walk(pgdat, sc->proactive);
 182
 183         set_initial_priority(pgdat, sc);
 184
 185 @@ -5372,7 +5373,7 @@ static void lru_gen_shrink_node(struct p
 186         clear_mm_walk();
 187
 188         blk_finish_plug(&plug);
 189 -
 190 +done:
 191         /* kswapd should never fail */
 192         pgdat->kswapd_failures = 0;
 193  }
 194 @@ -5944,7 +5945,7 @@ static ssize_t lru_gen_seq_write(struct
 195         set_task_reclaim_state(current, &sc.reclaim_state);
 196         flags = memalloc_noreclaim_save();
 197         blk_start_plug(&plug);
 198 -       if (!set_mm_walk(NULL)) {
 199 +       if (!set_mm_walk(NULL, true)) {
 200                 err = -ENOMEM;
 201                 goto done;
 202         }