OSDN Git Service

mm: memcontrol: make the slab calculation consistent
authorMuchun Song <songmuchun@bytedance.com>
Wed, 24 Feb 2021 20:03:43 +0000 (12:03 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Wed, 24 Feb 2021 21:38:29 +0000 (13:38 -0800)
Although the ratio of the slab is one, we also should read the ratio from
the related memory_stats instead of hard-coding.  And the local variable
of size is already the value of slab_unreclaimable.  So we do not need to
read again.

To do this we need some code like below:

if (unlikely(memory_stats[i].idx == NR_SLAB_UNRECLAIMABLE_B)) {
- size = memcg_page_state(memcg, NR_SLAB_RECLAIMABLE_B) +
-        memcg_page_state(memcg, NR_SLAB_UNRECLAIMABLE_B);
+       VM_BUG_ON(i < 1);
+       VM_BUG_ON(memory_stats[i - 1].idx != NR_SLAB_RECLAIMABLE_B);
+ size += memcg_page_state(memcg, memory_stats[i - 1].idx) *
+ memory_stats[i - 1].ratio;

It requires a series of VM_BUG_ONs or comments to ensure these two items
are actually adjacent and in the right order.  So it would probably be
easier to implement this using a wrapper that has a big switch() for unit
conversion.

More details about this discussion can refer to:

    https://lore.kernel.org/patchwork/patch/1348611/

This would fix the ratio inconsistency and get rid of the order
guarantee.

Link: https://lkml.kernel.org/r/20201228164110.2838-8-songmuchun@bytedance.com
Signed-off-by: Muchun Song <songmuchun@bytedance.com>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Cc: Feng Tang <feng.tang@intel.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Hugh Dickins <hughd@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: NeilBrown <neilb@suse.de>
Cc: Pankaj Gupta <pankaj.gupta@cloud.ionos.com>
Cc: Rafael. J. Wysocki <rafael@kernel.org>
Cc: Randy Dunlap <rdunlap@infradead.org>
Cc: Roman Gushchin <guro@fb.com>
Cc: Sami Tolvanen <samitolvanen@google.com>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/memcontrol.c

index d3a0c59..f27e862 100644 (file)
@@ -1512,49 +1512,71 @@ static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg)
 
 struct memory_stat {
        const char *name;
-       unsigned int ratio;
        unsigned int idx;
 };
 
 static const struct memory_stat memory_stats[] = {
-       { "anon", PAGE_SIZE, NR_ANON_MAPPED },
-       { "file", PAGE_SIZE, NR_FILE_PAGES },
-       { "kernel_stack", 1024, NR_KERNEL_STACK_KB },
-       { "pagetables", PAGE_SIZE, NR_PAGETABLE },
-       { "percpu", 1, MEMCG_PERCPU_B },
-       { "sock", PAGE_SIZE, MEMCG_SOCK },
-       { "shmem", PAGE_SIZE, NR_SHMEM },
-       { "file_mapped", PAGE_SIZE, NR_FILE_MAPPED },
-       { "file_dirty", PAGE_SIZE, NR_FILE_DIRTY },
-       { "file_writeback", PAGE_SIZE, NR_WRITEBACK },
+       { "anon",                       NR_ANON_MAPPED                  },
+       { "file",                       NR_FILE_PAGES                   },
+       { "kernel_stack",               NR_KERNEL_STACK_KB              },
+       { "pagetables",                 NR_PAGETABLE                    },
+       { "percpu",                     MEMCG_PERCPU_B                  },
+       { "sock",                       MEMCG_SOCK                      },
+       { "shmem",                      NR_SHMEM                        },
+       { "file_mapped",                NR_FILE_MAPPED                  },
+       { "file_dirty",                 NR_FILE_DIRTY                   },
+       { "file_writeback",             NR_WRITEBACK                    },
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
-       { "anon_thp", PAGE_SIZE, NR_ANON_THPS },
-       { "file_thp", PAGE_SIZE, NR_FILE_THPS },
-       { "shmem_thp", PAGE_SIZE, NR_SHMEM_THPS },
+       { "anon_thp",                   NR_ANON_THPS                    },
+       { "file_thp",                   NR_FILE_THPS                    },
+       { "shmem_thp",                  NR_SHMEM_THPS                   },
 #endif
-       { "inactive_anon", PAGE_SIZE, NR_INACTIVE_ANON },
-       { "active_anon", PAGE_SIZE, NR_ACTIVE_ANON },
-       { "inactive_file", PAGE_SIZE, NR_INACTIVE_FILE },
-       { "active_file", PAGE_SIZE, NR_ACTIVE_FILE },
-       { "unevictable", PAGE_SIZE, NR_UNEVICTABLE },
-
-       /*
-        * Note: The slab_reclaimable and slab_unreclaimable must be
-        * together and slab_reclaimable must be in front.
-        */
-       { "slab_reclaimable", 1, NR_SLAB_RECLAIMABLE_B },
-       { "slab_unreclaimable", 1, NR_SLAB_UNRECLAIMABLE_B },
+       { "inactive_anon",              NR_INACTIVE_ANON                },
+       { "active_anon",                NR_ACTIVE_ANON                  },
+       { "inactive_file",              NR_INACTIVE_FILE                },
+       { "active_file",                NR_ACTIVE_FILE                  },
+       { "unevictable",                NR_UNEVICTABLE                  },
+       { "slab_reclaimable",           NR_SLAB_RECLAIMABLE_B           },
+       { "slab_unreclaimable",         NR_SLAB_UNRECLAIMABLE_B         },
 
        /* The memory events */
-       { "workingset_refault_anon", 1, WORKINGSET_REFAULT_ANON },
-       { "workingset_refault_file", 1, WORKINGSET_REFAULT_FILE },
-       { "workingset_activate_anon", 1, WORKINGSET_ACTIVATE_ANON },
-       { "workingset_activate_file", 1, WORKINGSET_ACTIVATE_FILE },
-       { "workingset_restore_anon", 1, WORKINGSET_RESTORE_ANON },
-       { "workingset_restore_file", 1, WORKINGSET_RESTORE_FILE },
-       { "workingset_nodereclaim", 1, WORKINGSET_NODERECLAIM },
+       { "workingset_refault_anon",    WORKINGSET_REFAULT_ANON         },
+       { "workingset_refault_file",    WORKINGSET_REFAULT_FILE         },
+       { "workingset_activate_anon",   WORKINGSET_ACTIVATE_ANON        },
+       { "workingset_activate_file",   WORKINGSET_ACTIVATE_FILE        },
+       { "workingset_restore_anon",    WORKINGSET_RESTORE_ANON         },
+       { "workingset_restore_file",    WORKINGSET_RESTORE_FILE         },
+       { "workingset_nodereclaim",     WORKINGSET_NODERECLAIM          },
 };
 
+/* Translate stat items to the correct unit for memory.stat output */
+static int memcg_page_state_unit(int item)
+{
+       switch (item) {
+       case MEMCG_PERCPU_B:
+       case NR_SLAB_RECLAIMABLE_B:
+       case NR_SLAB_UNRECLAIMABLE_B:
+       case WORKINGSET_REFAULT_ANON:
+       case WORKINGSET_REFAULT_FILE:
+       case WORKINGSET_ACTIVATE_ANON:
+       case WORKINGSET_ACTIVATE_FILE:
+       case WORKINGSET_RESTORE_ANON:
+       case WORKINGSET_RESTORE_FILE:
+       case WORKINGSET_NODERECLAIM:
+               return 1;
+       case NR_KERNEL_STACK_KB:
+               return SZ_1K;
+       default:
+               return PAGE_SIZE;
+       }
+}
+
+static inline unsigned long memcg_page_state_output(struct mem_cgroup *memcg,
+                                                   int item)
+{
+       return memcg_page_state(memcg, item) * memcg_page_state_unit(item);
+}
+
 static char *memory_stat_format(struct mem_cgroup *memcg)
 {
        struct seq_buf s;
@@ -1578,13 +1600,12 @@ static char *memory_stat_format(struct mem_cgroup *memcg)
        for (i = 0; i < ARRAY_SIZE(memory_stats); i++) {
                u64 size;
 
-               size = memcg_page_state(memcg, memory_stats[i].idx);
-               size *= memory_stats[i].ratio;
+               size = memcg_page_state_output(memcg, memory_stats[i].idx);
                seq_buf_printf(&s, "%s %llu\n", memory_stats[i].name, size);
 
                if (unlikely(memory_stats[i].idx == NR_SLAB_UNRECLAIMABLE_B)) {
-                       size = memcg_page_state(memcg, NR_SLAB_RECLAIMABLE_B) +
-                              memcg_page_state(memcg, NR_SLAB_UNRECLAIMABLE_B);
+                       size += memcg_page_state_output(memcg,
+                                                       NR_SLAB_RECLAIMABLE_B);
                        seq_buf_printf(&s, "slab %llu\n", size);
                }
        }
@@ -6375,6 +6396,12 @@ static int memory_stat_show(struct seq_file *m, void *v)
 }
 
 #ifdef CONFIG_NUMA
+static inline unsigned long lruvec_page_state_output(struct lruvec *lruvec,
+                                                    int item)
+{
+       return lruvec_page_state(lruvec, item) * memcg_page_state_unit(item);
+}
+
 static int memory_numa_stat_show(struct seq_file *m, void *v)
 {
        int i;
@@ -6392,8 +6419,8 @@ static int memory_numa_stat_show(struct seq_file *m, void *v)
                        struct lruvec *lruvec;
 
                        lruvec = mem_cgroup_lruvec(memcg, NODE_DATA(nid));
-                       size = lruvec_page_state(lruvec, memory_stats[i].idx);
-                       size *= memory_stats[i].ratio;
+                       size = lruvec_page_state_output(lruvec,
+                                                       memory_stats[i].idx);
                        seq_printf(m, " N%d=%llu", nid, size);
                }
                seq_putc(m, '\n');