]> git.proxmox.com Git - mirror_ubuntu-disco-kernel.git/commitdiff
mm: memcontrol: eliminate raw access to stat and event counters
authorJohannes Weiner <hannes@cmpxchg.org>
Thu, 1 Feb 2018 00:16:37 +0000 (16:16 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Thu, 1 Feb 2018 01:18:36 +0000 (17:18 -0800)
Replace all raw 'this_cpu_' modifications of the stat and event per-cpu
counters with API functions such as mod_memcg_state().

This makes the code easier to read, but is also in preparation for the
next patch, which changes the per-cpu implementation of those counters.

Link: http://lkml.kernel.org/r/20171103153336.24044-1-hannes@cmpxchg.org
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Michal Hocko <mhocko@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
include/linux/memcontrol.h
mm/memcontrol.c

index 69966c461d1c1164c155f3c5ff2ee9d4c7220911..2c80b69dd26614ad0e4f7d9510a9691b0b1b3f53 100644 (file)
@@ -272,13 +272,6 @@ static inline bool mem_cgroup_disabled(void)
        return !cgroup_subsys_enabled(memory_cgrp_subsys);
 }
 
-static inline void mem_cgroup_event(struct mem_cgroup *memcg,
-                                   enum memcg_event_item event)
-{
-       this_cpu_inc(memcg->stat->events[event]);
-       cgroup_file_notify(&memcg->events_file);
-}
-
 bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg);
 
 int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm,
@@ -627,15 +620,23 @@ unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order,
                                                gfp_t gfp_mask,
                                                unsigned long *total_scanned);
 
+/* idx can be of type enum memcg_event_item or vm_event_item */
+static inline void __count_memcg_events(struct mem_cgroup *memcg,
+                                       int idx, unsigned long count)
+{
+       if (!mem_cgroup_disabled())
+               __this_cpu_add(memcg->stat->events[idx], count);
+}
+
+/* idx can be of type enum memcg_event_item or vm_event_item */
 static inline void count_memcg_events(struct mem_cgroup *memcg,
-                                     enum vm_event_item idx,
-                                     unsigned long count)
+                                     int idx, unsigned long count)
 {
        if (!mem_cgroup_disabled())
                this_cpu_add(memcg->stat->events[idx], count);
 }
 
-/* idx can be of type enum memcg_stat_item or node_stat_item */
+/* idx can be of type enum memcg_event_item or vm_event_item */
 static inline void count_memcg_page_event(struct page *page,
                                          int idx)
 {
@@ -654,12 +655,20 @@ static inline void count_memcg_event_mm(struct mm_struct *mm,
        rcu_read_lock();
        memcg = mem_cgroup_from_task(rcu_dereference(mm->owner));
        if (likely(memcg)) {
-               this_cpu_inc(memcg->stat->events[idx]);
+               count_memcg_events(memcg, idx, 1);
                if (idx == OOM_KILL)
                        cgroup_file_notify(&memcg->events_file);
        }
        rcu_read_unlock();
 }
+
+static inline void mem_cgroup_event(struct mem_cgroup *memcg,
+                                   enum memcg_event_item event)
+{
+       count_memcg_events(memcg, event, 1);
+       cgroup_file_notify(&memcg->events_file);
+}
+
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 void mem_cgroup_split_huge_fixup(struct page *head);
 #endif
index 9011997d8a5c75c79768b2739d738b9b5fec9dfa..23841af1d756f53273a96ddb890db86b2f349134 100644 (file)
@@ -586,23 +586,23 @@ static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg,
         * counted as CACHE even if it's on ANON LRU.
         */
        if (PageAnon(page))
-               __this_cpu_add(memcg->stat->count[MEMCG_RSS], nr_pages);
+               __mod_memcg_state(memcg, MEMCG_RSS, nr_pages);
        else {
-               __this_cpu_add(memcg->stat->count[MEMCG_CACHE], nr_pages);
+               __mod_memcg_state(memcg, MEMCG_CACHE, nr_pages);
                if (PageSwapBacked(page))
-                       __this_cpu_add(memcg->stat->count[NR_SHMEM], nr_pages);
+                       __mod_memcg_state(memcg, NR_SHMEM, nr_pages);
        }
 
        if (compound) {
                VM_BUG_ON_PAGE(!PageTransHuge(page), page);
-               __this_cpu_add(memcg->stat->count[MEMCG_RSS_HUGE], nr_pages);
+               __mod_memcg_state(memcg, MEMCG_RSS_HUGE, nr_pages);
        }
 
        /* pagein of a big page is an event. So, ignore page size */
        if (nr_pages > 0)
-               __this_cpu_inc(memcg->stat->events[PGPGIN]);
+               __count_memcg_events(memcg, PGPGIN, 1);
        else {
-               __this_cpu_inc(memcg->stat->events[PGPGOUT]);
+               __count_memcg_events(memcg, PGPGOUT, 1);
                nr_pages = -nr_pages; /* for event */
        }
 
@@ -2415,18 +2415,11 @@ void mem_cgroup_split_huge_fixup(struct page *head)
        for (i = 1; i < HPAGE_PMD_NR; i++)
                head[i].mem_cgroup = head->mem_cgroup;
 
-       __this_cpu_sub(head->mem_cgroup->stat->count[MEMCG_RSS_HUGE],
-                      HPAGE_PMD_NR);
+       __mod_memcg_state(head->mem_cgroup, MEMCG_RSS_HUGE, -HPAGE_PMD_NR);
 }
 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 
 #ifdef CONFIG_MEMCG_SWAP
-static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg,
-                                      int nr_entries)
-{
-       this_cpu_add(memcg->stat->count[MEMCG_SWAP], nr_entries);
-}
-
 /**
  * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
  * @entry: swap entry to be moved
@@ -2450,8 +2443,8 @@ static int mem_cgroup_move_swap_account(swp_entry_t entry,
        new_id = mem_cgroup_id(to);
 
        if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) {
-               mem_cgroup_swap_statistics(from, -1);
-               mem_cgroup_swap_statistics(to, 1);
+               mod_memcg_state(from, MEMCG_SWAP, -1);
+               mod_memcg_state(to, MEMCG_SWAP, 1);
                return 0;
        }
        return -EINVAL;
@@ -4584,8 +4577,8 @@ static int mem_cgroup_move_account(struct page *page,
        spin_lock_irqsave(&from->move_lock, flags);
 
        if (!anon && page_mapped(page)) {
-               __this_cpu_sub(from->stat->count[NR_FILE_MAPPED], nr_pages);
-               __this_cpu_add(to->stat->count[NR_FILE_MAPPED], nr_pages);
+               __mod_memcg_state(from, NR_FILE_MAPPED, -nr_pages);
+               __mod_memcg_state(to, NR_FILE_MAPPED, nr_pages);
        }
 
        /*
@@ -4597,16 +4590,14 @@ static int mem_cgroup_move_account(struct page *page,
                struct address_space *mapping = page_mapping(page);
 
                if (mapping_cap_account_dirty(mapping)) {
-                       __this_cpu_sub(from->stat->count[NR_FILE_DIRTY],
-                                      nr_pages);
-                       __this_cpu_add(to->stat->count[NR_FILE_DIRTY],
-                                      nr_pages);
+                       __mod_memcg_state(from, NR_FILE_DIRTY, -nr_pages);
+                       __mod_memcg_state(to, NR_FILE_DIRTY, nr_pages);
                }
        }
 
        if (PageWriteback(page)) {
-               __this_cpu_sub(from->stat->count[NR_WRITEBACK], nr_pages);
-               __this_cpu_add(to->stat->count[NR_WRITEBACK], nr_pages);
+               __mod_memcg_state(from, NR_WRITEBACK, -nr_pages);
+               __mod_memcg_state(to, NR_WRITEBACK, nr_pages);
        }
 
        /*
@@ -5642,11 +5633,11 @@ static void uncharge_batch(const struct uncharge_gather *ug)
        }
 
        local_irq_save(flags);
-       __this_cpu_sub(ug->memcg->stat->count[MEMCG_RSS], ug->nr_anon);
-       __this_cpu_sub(ug->memcg->stat->count[MEMCG_CACHE], ug->nr_file);
-       __this_cpu_sub(ug->memcg->stat->count[MEMCG_RSS_HUGE], ug->nr_huge);
-       __this_cpu_sub(ug->memcg->stat->count[NR_SHMEM], ug->nr_shmem);
-       __this_cpu_add(ug->memcg->stat->events[PGPGOUT], ug->pgpgout);
+       __mod_memcg_state(ug->memcg, MEMCG_RSS, -ug->nr_anon);
+       __mod_memcg_state(ug->memcg, MEMCG_CACHE, -ug->nr_file);
+       __mod_memcg_state(ug->memcg, MEMCG_RSS_HUGE, -ug->nr_huge);
+       __mod_memcg_state(ug->memcg, NR_SHMEM, -ug->nr_shmem);
+       __count_memcg_events(ug->memcg, PGPGOUT, ug->pgpgout);
        __this_cpu_add(ug->memcg->stat->nr_page_events, nr_pages);
        memcg_check_events(ug->memcg, ug->dummy_page);
        local_irq_restore(flags);
@@ -5874,7 +5865,7 @@ bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages)
        if (in_softirq())
                gfp_mask = GFP_NOWAIT;
 
-       this_cpu_add(memcg->stat->count[MEMCG_SOCK], nr_pages);
+       mod_memcg_state(memcg, MEMCG_SOCK, nr_pages);
 
        if (try_charge(memcg, gfp_mask, nr_pages) == 0)
                return true;
@@ -5895,7 +5886,7 @@ void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages)
                return;
        }
 
-       this_cpu_sub(memcg->stat->count[MEMCG_SOCK], nr_pages);
+       mod_memcg_state(memcg, MEMCG_SOCK, -nr_pages);
 
        refill_stock(memcg, nr_pages);
 }
@@ -6019,7 +6010,7 @@ void mem_cgroup_swapout(struct page *page, swp_entry_t entry)
        oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg),
                                   nr_entries);
        VM_BUG_ON_PAGE(oldid, page);
-       mem_cgroup_swap_statistics(swap_memcg, nr_entries);
+       mod_memcg_state(swap_memcg, MEMCG_SWAP, nr_entries);
 
        page->mem_cgroup = NULL;
 
@@ -6085,7 +6076,7 @@ int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry)
                mem_cgroup_id_get_many(memcg, nr_pages - 1);
        oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg), nr_pages);
        VM_BUG_ON_PAGE(oldid, page);
-       mem_cgroup_swap_statistics(memcg, nr_pages);
+       mod_memcg_state(memcg, MEMCG_SWAP, nr_pages);
 
        return 0;
 }
@@ -6113,7 +6104,7 @@ void mem_cgroup_uncharge_swap(swp_entry_t entry, unsigned int nr_pages)
                        else
                                page_counter_uncharge(&memcg->memsw, nr_pages);
                }
-               mem_cgroup_swap_statistics(memcg, -nr_pages);
+               mod_memcg_state(memcg, MEMCG_SWAP, -nr_pages);
                mem_cgroup_id_put_many(memcg, nr_pages);
        }
        rcu_read_unlock();