]> git.proxmox.com Git - mirror_ubuntu-eoan-kernel.git/commitdiff
memcg: unify charge/uncharge quantities to units of pages
authorJohannes Weiner <hannes@cmpxchg.org>
Wed, 23 Mar 2011 23:42:36 +0000 (16:42 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Thu, 24 Mar 2011 02:46:30 +0000 (19:46 -0700)
There is no clear pattern when we pass a page count and when we pass a
byte count that is a multiple of PAGE_SIZE.

We never charge or uncharge subpage quantities, so convert it all to page
counts.

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/memcontrol.c

index 9dfbed2aacc9b6b5fa43e4a8ef9d007b0fde3418..bc02218eab012428dfe193315c716255cf70aa79 100644 (file)
@@ -1109,16 +1109,16 @@ unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan,
  * @mem: the memory cgroup
  *
  * Returns the maximum amount of memory @mem can be charged with, in
- * bytes.
+ * pages.
  */
-static unsigned long long mem_cgroup_margin(struct mem_cgroup *mem)
+static unsigned long mem_cgroup_margin(struct mem_cgroup *mem)
 {
        unsigned long long margin;
 
        margin = res_counter_margin(&mem->res);
        if (do_swap_account)
                margin = min(margin, res_counter_margin(&mem->memsw));
-       return margin;
+       return margin >> PAGE_SHIFT;
 }
 
 static unsigned int get_swappiness(struct mem_cgroup *memcg)
@@ -1647,7 +1647,7 @@ EXPORT_SYMBOL(mem_cgroup_update_page_stat);
  * size of first charge trial. "32" comes from vmscan.c's magic value.
  * TODO: maybe necessary to use big numbers in big irons.
  */
-#define CHARGE_SIZE    (32 * PAGE_SIZE)
+#define CHARGE_BATCH   32U
 struct memcg_stock_pcp {
        struct mem_cgroup *cached; /* this never be root cgroup */
        unsigned int nr_pages;
@@ -1822,9 +1822,10 @@ enum {
        CHARGE_OOM_DIE,         /* the current is killed because of OOM */
 };
 
-static int __mem_cgroup_do_charge(struct mem_cgroup *mem, gfp_t gfp_mask,
-                               int csize, bool oom_check)
+static int mem_cgroup_do_charge(struct mem_cgroup *mem, gfp_t gfp_mask,
+                               unsigned int nr_pages, bool oom_check)
 {
+       unsigned long csize = nr_pages * PAGE_SIZE;
        struct mem_cgroup *mem_over_limit;
        struct res_counter *fail_res;
        unsigned long flags = 0;
@@ -1845,14 +1846,13 @@ static int __mem_cgroup_do_charge(struct mem_cgroup *mem, gfp_t gfp_mask,
        } else
                mem_over_limit = mem_cgroup_from_res_counter(fail_res, res);
        /*
-        * csize can be either a huge page (HPAGE_SIZE), a batch of
-        * regular pages (CHARGE_SIZE), or a single regular page
-        * (PAGE_SIZE).
+        * nr_pages can be either a huge page (HPAGE_PMD_NR), a batch
+        * of regular pages (CHARGE_BATCH), or a single regular page (1).
         *
         * Never reclaim on behalf of optional batching, retry with a
         * single page instead.
         */
-       if (csize == CHARGE_SIZE)
+       if (nr_pages == CHARGE_BATCH)
                return CHARGE_RETRY;
 
        if (!(gfp_mask & __GFP_WAIT))
@@ -1860,7 +1860,7 @@ static int __mem_cgroup_do_charge(struct mem_cgroup *mem, gfp_t gfp_mask,
 
        ret = mem_cgroup_hierarchical_reclaim(mem_over_limit, NULL,
                                              gfp_mask, flags);
-       if (mem_cgroup_margin(mem_over_limit) >= csize)
+       if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
                return CHARGE_RETRY;
        /*
         * Even though the limit is exceeded at this point, reclaim
@@ -1871,7 +1871,7 @@ static int __mem_cgroup_do_charge(struct mem_cgroup *mem, gfp_t gfp_mask,
         * unlikely to succeed so close to the limit, and we fall back
         * to regular pages anyway in case of failure.
         */
-       if (csize == PAGE_SIZE && ret)
+       if (nr_pages == 1 && ret)
                return CHARGE_RETRY;
 
        /*
@@ -1897,13 +1897,14 @@ static int __mem_cgroup_do_charge(struct mem_cgroup *mem, gfp_t gfp_mask,
  */
 static int __mem_cgroup_try_charge(struct mm_struct *mm,
                                   gfp_t gfp_mask,
-                                  struct mem_cgroup **memcg, bool oom,
-                                  int page_size)
+                                  unsigned int nr_pages,
+                                  struct mem_cgroup **memcg,
+                                  bool oom)
 {
+       unsigned int batch = max(CHARGE_BATCH, nr_pages);
        int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
        struct mem_cgroup *mem = NULL;
        int ret;
-       int csize = max(CHARGE_SIZE, (unsigned long) page_size);
 
        /*
         * Unlike gloval-vm's OOM-kill, we're not in memory shortage
@@ -1928,7 +1929,7 @@ again:
                VM_BUG_ON(css_is_removed(&mem->css));
                if (mem_cgroup_is_root(mem))
                        goto done;
-               if (page_size == PAGE_SIZE && consume_stock(mem))
+               if (nr_pages == 1 && consume_stock(mem))
                        goto done;
                css_get(&mem->css);
        } else {
@@ -1951,7 +1952,7 @@ again:
                        rcu_read_unlock();
                        goto done;
                }
-               if (page_size == PAGE_SIZE && consume_stock(mem)) {
+               if (nr_pages == 1 && consume_stock(mem)) {
                        /*
                         * It seems dagerous to access memcg without css_get().
                         * But considering how consume_stok works, it's not
@@ -1986,13 +1987,12 @@ again:
                        nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
                }
 
-               ret = __mem_cgroup_do_charge(mem, gfp_mask, csize, oom_check);
-
+               ret = mem_cgroup_do_charge(mem, gfp_mask, batch, oom_check);
                switch (ret) {
                case CHARGE_OK:
                        break;
                case CHARGE_RETRY: /* not in OOM situation but retry */
-                       csize = page_size;
+                       batch = nr_pages;
                        css_put(&mem->css);
                        mem = NULL;
                        goto again;
@@ -2013,8 +2013,8 @@ again:
                }
        } while (ret != CHARGE_OK);
 
-       if (csize > page_size)
-               refill_stock(mem, (csize - page_size) >> PAGE_SHIFT);
+       if (batch > nr_pages)
+               refill_stock(mem, batch - nr_pages);
        css_put(&mem->css);
 done:
        *memcg = mem;
@@ -2093,12 +2093,10 @@ struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page)
 
 static void __mem_cgroup_commit_charge(struct mem_cgroup *mem,
                                       struct page *page,
+                                      unsigned int nr_pages,
                                       struct page_cgroup *pc,
-                                      enum charge_type ctype,
-                                      int page_size)
+                                      enum charge_type ctype)
 {
-       int nr_pages = page_size >> PAGE_SHIFT;
-
        lock_page_cgroup(pc);
        if (unlikely(PageCgroupUsed(pc))) {
                unlock_page_cgroup(pc);
@@ -2187,26 +2185,28 @@ void mem_cgroup_split_huge_fixup(struct page *head, struct page *tail)
 /**
  * mem_cgroup_move_account - move account of the page
  * @page: the page
+ * @nr_pages: number of regular pages (>1 for huge pages)
  * @pc:        page_cgroup of the page.
  * @from: mem_cgroup which the page is moved from.
  * @to:        mem_cgroup which the page is moved to. @from != @to.
  * @uncharge: whether we should call uncharge and css_put against @from.
- * @charge_size: number of bytes to charge (regular or huge page)
  *
  * The caller must confirm following.
  * - page is not on LRU (isolate_page() is useful.)
- * - compound_lock is held when charge_size > PAGE_SIZE
+ * - compound_lock is held when nr_pages > 1
  *
  * This function doesn't do "charge" nor css_get to new cgroup. It should be
  * done by a caller(__mem_cgroup_try_charge would be usefull). If @uncharge is
  * true, this function does "uncharge" from old cgroup, but it doesn't if
  * @uncharge is false, so a caller should do "uncharge".
  */
-static int mem_cgroup_move_account(struct page *page, struct page_cgroup *pc,
-                                  struct mem_cgroup *from, struct mem_cgroup *to,
-                                  bool uncharge, int charge_size)
+static int mem_cgroup_move_account(struct page *page,
+                                  unsigned int nr_pages,
+                                  struct page_cgroup *pc,
+                                  struct mem_cgroup *from,
+                                  struct mem_cgroup *to,
+                                  bool uncharge)
 {
-       int nr_pages = charge_size >> PAGE_SHIFT;
        unsigned long flags;
        int ret;
 
@@ -2219,7 +2219,7 @@ static int mem_cgroup_move_account(struct page *page, struct page_cgroup *pc,
         * hold it.
         */
        ret = -EBUSY;
-       if (charge_size > PAGE_SIZE && !PageTransHuge(page))
+       if (nr_pages > 1 && !PageTransHuge(page))
                goto out;
 
        lock_page_cgroup(pc);
@@ -2277,7 +2277,7 @@ static int mem_cgroup_move_parent(struct page *page,
        struct cgroup *cg = child->css.cgroup;
        struct cgroup *pcg = cg->parent;
        struct mem_cgroup *parent;
-       int page_size = PAGE_SIZE;
+       unsigned int nr_pages;
        unsigned long flags;
        int ret;
 
@@ -2291,23 +2291,21 @@ static int mem_cgroup_move_parent(struct page *page,
        if (isolate_lru_page(page))
                goto put;
 
-       if (PageTransHuge(page))
-               page_size = HPAGE_SIZE;
+       nr_pages = hpage_nr_pages(page);
 
        parent = mem_cgroup_from_cont(pcg);
-       ret = __mem_cgroup_try_charge(NULL, gfp_mask,
-                               &parent, false, page_size);
+       ret = __mem_cgroup_try_charge(NULL, gfp_mask, nr_pages, &parent, false);
        if (ret || !parent)
                goto put_back;
 
-       if (page_size > PAGE_SIZE)
+       if (nr_pages > 1)
                flags = compound_lock_irqsave(page);
 
-       ret = mem_cgroup_move_account(page, pc, child, parent, true, page_size);
+       ret = mem_cgroup_move_account(page, nr_pages, pc, child, parent, true);
        if (ret)
-               __mem_cgroup_cancel_charge(parent, page_size >> PAGE_SHIFT);
+               __mem_cgroup_cancel_charge(parent, nr_pages);
 
-       if (page_size > PAGE_SIZE)
+       if (nr_pages > 1)
                compound_unlock_irqrestore(page, flags);
 put_back:
        putback_lru_page(page);
@@ -2327,13 +2325,13 @@ static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
                                gfp_t gfp_mask, enum charge_type ctype)
 {
        struct mem_cgroup *mem = NULL;
-       int page_size = PAGE_SIZE;
+       unsigned int nr_pages = 1;
        struct page_cgroup *pc;
        bool oom = true;
        int ret;
 
        if (PageTransHuge(page)) {
-               page_size <<= compound_order(page);
+               nr_pages <<= compound_order(page);
                VM_BUG_ON(!PageTransHuge(page));
                /*
                 * Never OOM-kill a process for a huge page.  The
@@ -2345,11 +2343,11 @@ static int mem_cgroup_charge_common(struct page *page, struct mm_struct *mm,
        pc = lookup_page_cgroup(page);
        BUG_ON(!pc); /* XXX: remove this and move pc lookup into commit */
 
-       ret = __mem_cgroup_try_charge(mm, gfp_mask, &mem, oom, page_size);
+       ret = __mem_cgroup_try_charge(mm, gfp_mask, nr_pages, &mem, oom);
        if (ret || !mem)
                return ret;
 
-       __mem_cgroup_commit_charge(mem, page, pc, ctype, page_size);
+       __mem_cgroup_commit_charge(mem, page, nr_pages, pc, ctype);
        return 0;
 }
 
@@ -2465,13 +2463,13 @@ int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
        if (!mem)
                goto charge_cur_mm;
        *ptr = mem;
-       ret = __mem_cgroup_try_charge(NULL, mask, ptr, true, PAGE_SIZE);
+       ret = __mem_cgroup_try_charge(NULL, mask, 1, ptr, true);
        css_put(&mem->css);
        return ret;
 charge_cur_mm:
        if (unlikely(!mm))
                mm = &init_mm;
-       return __mem_cgroup_try_charge(mm, mask, ptr, true, PAGE_SIZE);
+       return __mem_cgroup_try_charge(mm, mask, 1, ptr, true);
 }
 
 static void
@@ -2487,7 +2485,7 @@ __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *ptr,
        cgroup_exclude_rmdir(&ptr->css);
        pc = lookup_page_cgroup(page);
        mem_cgroup_lru_del_before_commit_swapcache(page);
-       __mem_cgroup_commit_charge(ptr, page, pc, ctype, PAGE_SIZE);
+       __mem_cgroup_commit_charge(ptr, page, 1, pc, ctype);
        mem_cgroup_lru_add_after_commit_swapcache(page);
        /*
         * Now swap is on-memory. This means this page may be
@@ -2539,12 +2537,13 @@ void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem)
        __mem_cgroup_cancel_charge(mem, 1);
 }
 
-static void
-__do_uncharge(struct mem_cgroup *mem, const enum charge_type ctype,
-             int page_size)
+static void mem_cgroup_do_uncharge(struct mem_cgroup *mem,
+                                  unsigned int nr_pages,
+                                  const enum charge_type ctype)
 {
        struct memcg_batch_info *batch = NULL;
        bool uncharge_memsw = true;
+
        /* If swapout, usage of swap doesn't decrease */
        if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
                uncharge_memsw = false;
@@ -2568,7 +2567,7 @@ __do_uncharge(struct mem_cgroup *mem, const enum charge_type ctype,
        if (!batch->do_batch || test_thread_flag(TIF_MEMDIE))
                goto direct_uncharge;
 
-       if (page_size != PAGE_SIZE)
+       if (nr_pages > 1)
                goto direct_uncharge;
 
        /*
@@ -2584,9 +2583,9 @@ __do_uncharge(struct mem_cgroup *mem, const enum charge_type ctype,
                batch->memsw_nr_pages++;
        return;
 direct_uncharge:
-       res_counter_uncharge(&mem->res, page_size);
+       res_counter_uncharge(&mem->res, nr_pages * PAGE_SIZE);
        if (uncharge_memsw)
-               res_counter_uncharge(&mem->memsw, page_size);
+               res_counter_uncharge(&mem->memsw, nr_pages * PAGE_SIZE);
        if (unlikely(batch->memcg != mem))
                memcg_oom_recover(mem);
        return;
@@ -2598,10 +2597,9 @@ direct_uncharge:
 static struct mem_cgroup *
 __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
 {
-       int count;
-       struct page_cgroup *pc;
        struct mem_cgroup *mem = NULL;
-       int page_size = PAGE_SIZE;
+       unsigned int nr_pages = 1;
+       struct page_cgroup *pc;
 
        if (mem_cgroup_disabled())
                return NULL;
@@ -2610,11 +2608,9 @@ __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
                return NULL;
 
        if (PageTransHuge(page)) {
-               page_size <<= compound_order(page);
+               nr_pages <<= compound_order(page);
                VM_BUG_ON(!PageTransHuge(page));
        }
-
-       count = page_size >> PAGE_SHIFT;
        /*
         * Check if our page_cgroup is valid
         */
@@ -2647,7 +2643,7 @@ __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
                break;
        }
 
-       mem_cgroup_charge_statistics(mem, PageCgroupCache(pc), -count);
+       mem_cgroup_charge_statistics(mem, PageCgroupCache(pc), -nr_pages);
 
        ClearPageCgroupUsed(pc);
        /*
@@ -2668,7 +2664,7 @@ __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
                mem_cgroup_get(mem);
        }
        if (!mem_cgroup_is_root(mem))
-               __do_uncharge(mem, ctype, page_size);
+               mem_cgroup_do_uncharge(mem, nr_pages, ctype);
 
        return mem;
 
@@ -2860,8 +2856,8 @@ static inline int mem_cgroup_move_swap_account(swp_entry_t entry,
 int mem_cgroup_prepare_migration(struct page *page,
        struct page *newpage, struct mem_cgroup **ptr, gfp_t gfp_mask)
 {
-       struct page_cgroup *pc;
        struct mem_cgroup *mem = NULL;
+       struct page_cgroup *pc;
        enum charge_type ctype;
        int ret = 0;
 
@@ -2917,7 +2913,7 @@ int mem_cgroup_prepare_migration(struct page *page,
                return 0;
 
        *ptr = mem;
-       ret = __mem_cgroup_try_charge(NULL, gfp_mask, ptr, false, PAGE_SIZE);
+       ret = __mem_cgroup_try_charge(NULL, gfp_mask, 1, ptr, false);
        css_put(&mem->css);/* drop extra refcnt */
        if (ret || *ptr == NULL) {
                if (PageAnon(page)) {
@@ -2944,7 +2940,7 @@ int mem_cgroup_prepare_migration(struct page *page,
                ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
        else
                ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM;
-       __mem_cgroup_commit_charge(mem, page, pc, ctype, PAGE_SIZE);
+       __mem_cgroup_commit_charge(mem, page, 1, pc, ctype);
        return ret;
 }
 
@@ -4598,8 +4594,7 @@ one_by_one:
                        batch_count = PRECHARGE_COUNT_AT_ONCE;
                        cond_resched();
                }
-               ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, &mem, false,
-                                             PAGE_SIZE);
+               ret = __mem_cgroup_try_charge(NULL, GFP_KERNEL, 1, &mem, false);
                if (ret || !mem)
                        /* mem_cgroup_clear_mc() will do uncharge later */
                        return -ENOMEM;
@@ -4945,8 +4940,8 @@ retry:
                        if (isolate_lru_page(page))
                                goto put;
                        pc = lookup_page_cgroup(page);
-                       if (!mem_cgroup_move_account(page, pc,
-                                       mc.from, mc.to, false, PAGE_SIZE)) {
+                       if (!mem_cgroup_move_account(page, 1, pc,
+                                                    mc.from, mc.to, false)) {
                                mc.precharge--;
                                /* we uncharge from mc.from later. */
                                mc.moved_charge++;