1 /* memcontrol.c - Memory Controller
3 * Copyright IBM Corporation, 2007
4 * Author Balbir Singh <balbir@linux.vnet.ibm.com>
6 * Copyright 2007 OpenVZ SWsoft Inc
7 * Author: Pavel Emelianov <xemul@openvz.org>
10 * Copyright (C) 2009 Nokia Corporation
11 * Author: Kirill A. Shutemov
13 * This program is free software; you can redistribute it and/or modify
14 * it under the terms of the GNU General Public License as published by
15 * the Free Software Foundation; either version 2 of the License, or
16 * (at your option) any later version.
18 * This program is distributed in the hope that it will be useful,
19 * but WITHOUT ANY WARRANTY; without even the implied warranty of
20 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
21 * GNU General Public License for more details.
24 #include <linux/res_counter.h>
25 #include <linux/memcontrol.h>
26 #include <linux/cgroup.h>
28 #include <linux/hugetlb.h>
29 #include <linux/pagemap.h>
30 #include <linux/smp.h>
31 #include <linux/page-flags.h>
32 #include <linux/backing-dev.h>
33 #include <linux/bit_spinlock.h>
34 #include <linux/rcupdate.h>
35 #include <linux/limits.h>
36 #include <linux/mutex.h>
37 #include <linux/rbtree.h>
38 #include <linux/slab.h>
39 #include <linux/swap.h>
40 #include <linux/swapops.h>
41 #include <linux/spinlock.h>
42 #include <linux/eventfd.h>
43 #include <linux/sort.h>
45 #include <linux/seq_file.h>
46 #include <linux/vmalloc.h>
47 #include <linux/mm_inline.h>
48 #include <linux/page_cgroup.h>
49 #include <linux/cpu.h>
52 #include <asm/uaccess.h>
54 struct cgroup_subsys mem_cgroup_subsys __read_mostly
;
55 #define MEM_CGROUP_RECLAIM_RETRIES 5
56 struct mem_cgroup
*root_mem_cgroup __read_mostly
;
58 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
59 /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
60 int do_swap_account __read_mostly
;
61 static int really_do_swap_account __initdata
= 1; /* for remember boot option*/
63 #define do_swap_account (0)
67 * Per memcg event counter is incremented at every pagein/pageout. This counter
68 * is used for trigger some periodic events. This is straightforward and better
69 * than using jiffies etc. to handle periodic memcg event.
71 * These values will be used as !((event) & ((1 <<(thresh)) - 1))
73 #define THRESHOLDS_EVENTS_THRESH (7) /* once in 128 */
74 #define SOFTLIMIT_EVENTS_THRESH (10) /* once in 1024 */
77 * Statistics for memory cgroup.
79 enum mem_cgroup_stat_index
{
81 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
83 MEM_CGROUP_STAT_CACHE
, /* # of pages charged as cache */
84 MEM_CGROUP_STAT_RSS
, /* # of pages charged as anon rss */
85 MEM_CGROUP_STAT_FILE_MAPPED
, /* # of pages charged as file rss */
86 MEM_CGROUP_STAT_PGPGIN_COUNT
, /* # of pages paged in */
87 MEM_CGROUP_STAT_PGPGOUT_COUNT
, /* # of pages paged out */
88 MEM_CGROUP_STAT_SWAPOUT
, /* # of pages, swapped out */
89 MEM_CGROUP_EVENTS
, /* incremented at every pagein/pageout */
91 MEM_CGROUP_STAT_NSTATS
,
94 struct mem_cgroup_stat_cpu
{
95 s64 count
[MEM_CGROUP_STAT_NSTATS
];
99 * per-zone information in memory controller.
101 struct mem_cgroup_per_zone
{
103 * spin_lock to protect the per cgroup LRU
105 struct list_head lists
[NR_LRU_LISTS
];
106 unsigned long count
[NR_LRU_LISTS
];
108 struct zone_reclaim_stat reclaim_stat
;
109 struct rb_node tree_node
; /* RB tree node */
110 unsigned long long usage_in_excess
;/* Set to the value by which */
111 /* the soft limit is exceeded*/
113 struct mem_cgroup
*mem
; /* Back pointer, we cannot */
114 /* use container_of */
116 /* Macro for accessing counter */
117 #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
119 struct mem_cgroup_per_node
{
120 struct mem_cgroup_per_zone zoneinfo
[MAX_NR_ZONES
];
123 struct mem_cgroup_lru_info
{
124 struct mem_cgroup_per_node
*nodeinfo
[MAX_NUMNODES
];
128 * Cgroups above their limits are maintained in a RB-Tree, independent of
129 * their hierarchy representation
132 struct mem_cgroup_tree_per_zone
{
133 struct rb_root rb_root
;
137 struct mem_cgroup_tree_per_node
{
138 struct mem_cgroup_tree_per_zone rb_tree_per_zone
[MAX_NR_ZONES
];
141 struct mem_cgroup_tree
{
142 struct mem_cgroup_tree_per_node
*rb_tree_per_node
[MAX_NUMNODES
];
145 static struct mem_cgroup_tree soft_limit_tree __read_mostly
;
147 struct mem_cgroup_threshold
{
148 struct eventfd_ctx
*eventfd
;
153 struct mem_cgroup_threshold_ary
{
154 /* An array index points to threshold just below usage. */
155 int current_threshold
;
156 /* Size of entries[] */
158 /* Array of thresholds */
159 struct mem_cgroup_threshold entries
[0];
162 struct mem_cgroup_thresholds
{
163 /* Primary thresholds array */
164 struct mem_cgroup_threshold_ary
*primary
;
166 * Spare threshold array.
167 * This is needed to make mem_cgroup_unregister_event() "never fail".
168 * It must be able to store at least primary->size - 1 entries.
170 struct mem_cgroup_threshold_ary
*spare
;
174 struct mem_cgroup_eventfd_list
{
175 struct list_head list
;
176 struct eventfd_ctx
*eventfd
;
179 static void mem_cgroup_threshold(struct mem_cgroup
*mem
);
180 static void mem_cgroup_oom_notify(struct mem_cgroup
*mem
);
183 * The memory controller data structure. The memory controller controls both
184 * page cache and RSS per cgroup. We would eventually like to provide
185 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
186 * to help the administrator determine what knobs to tune.
188 * TODO: Add a water mark for the memory controller. Reclaim will begin when
189 * we hit the water mark. May be even add a low water mark, such that
190 * no reclaim occurs from a cgroup at it's low water mark, this is
191 * a feature that will be implemented much later in the future.
194 struct cgroup_subsys_state css
;
196 * the counter to account for memory usage
198 struct res_counter res
;
200 * the counter to account for mem+swap usage.
202 struct res_counter memsw
;
204 * Per cgroup active and inactive list, similar to the
205 * per zone LRU lists.
207 struct mem_cgroup_lru_info info
;
210 protect against reclaim related member.
212 spinlock_t reclaim_param_lock
;
215 * While reclaiming in a hierarchy, we cache the last child we
218 int last_scanned_child
;
220 * Should the accounting and control be hierarchical, per subtree?
226 unsigned int swappiness
;
227 /* OOM-Killer disable */
228 int oom_kill_disable
;
230 /* set when res.limit == memsw.limit */
231 bool memsw_is_minimum
;
233 /* protect arrays of thresholds */
234 struct mutex thresholds_lock
;
236 /* thresholds for memory usage. RCU-protected */
237 struct mem_cgroup_thresholds thresholds
;
239 /* thresholds for mem+swap usage. RCU-protected */
240 struct mem_cgroup_thresholds memsw_thresholds
;
242 /* For oom notifier event fd */
243 struct list_head oom_notify
;
246 * Should we move charges of a task when a task is moved into this
247 * mem_cgroup ? And what type of charges should we move ?
249 unsigned long move_charge_at_immigrate
;
253 struct mem_cgroup_stat_cpu
*stat
;
256 /* Stuffs for move charges at task migration. */
258 * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a
259 * left-shifted bitmap of these types.
262 MOVE_CHARGE_TYPE_ANON
, /* private anonymous page and swap of it */
263 MOVE_CHARGE_TYPE_FILE
, /* file page(including tmpfs) and swap of it */
267 /* "mc" and its members are protected by cgroup_mutex */
268 static struct move_charge_struct
{
269 struct mem_cgroup
*from
;
270 struct mem_cgroup
*to
;
271 unsigned long precharge
;
272 unsigned long moved_charge
;
273 unsigned long moved_swap
;
274 struct task_struct
*moving_task
; /* a task moving charges */
275 wait_queue_head_t waitq
; /* a waitq for other context */
277 .waitq
= __WAIT_QUEUE_HEAD_INITIALIZER(mc
.waitq
),
280 static bool move_anon(void)
282 return test_bit(MOVE_CHARGE_TYPE_ANON
,
283 &mc
.to
->move_charge_at_immigrate
);
286 static bool move_file(void)
288 return test_bit(MOVE_CHARGE_TYPE_FILE
,
289 &mc
.to
->move_charge_at_immigrate
);
293 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
294 * limit reclaim to prevent infinite loops, if they ever occur.
296 #define MEM_CGROUP_MAX_RECLAIM_LOOPS (100)
297 #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2)
300 MEM_CGROUP_CHARGE_TYPE_CACHE
= 0,
301 MEM_CGROUP_CHARGE_TYPE_MAPPED
,
302 MEM_CGROUP_CHARGE_TYPE_SHMEM
, /* used by page migration of shmem */
303 MEM_CGROUP_CHARGE_TYPE_FORCE
, /* used by force_empty */
304 MEM_CGROUP_CHARGE_TYPE_SWAPOUT
, /* for accounting swapcache */
305 MEM_CGROUP_CHARGE_TYPE_DROP
, /* a page was unused swap cache */
309 /* only for here (for easy reading.) */
310 #define PCGF_CACHE (1UL << PCG_CACHE)
311 #define PCGF_USED (1UL << PCG_USED)
312 #define PCGF_LOCK (1UL << PCG_LOCK)
313 /* Not used, but added here for completeness */
314 #define PCGF_ACCT (1UL << PCG_ACCT)
316 /* for encoding cft->private value on file */
319 #define _OOM_TYPE (2)
320 #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val))
321 #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff)
322 #define MEMFILE_ATTR(val) ((val) & 0xffff)
323 /* Used for OOM nofiier */
324 #define OOM_CONTROL (0)
327 * Reclaim flags for mem_cgroup_hierarchical_reclaim
329 #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
330 #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
331 #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
332 #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
333 #define MEM_CGROUP_RECLAIM_SOFT_BIT 0x2
334 #define MEM_CGROUP_RECLAIM_SOFT (1 << MEM_CGROUP_RECLAIM_SOFT_BIT)
336 static void mem_cgroup_get(struct mem_cgroup
*mem
);
337 static void mem_cgroup_put(struct mem_cgroup
*mem
);
338 static struct mem_cgroup
*parent_mem_cgroup(struct mem_cgroup
*mem
);
339 static void drain_all_stock_async(void);
341 static struct mem_cgroup_per_zone
*
342 mem_cgroup_zoneinfo(struct mem_cgroup
*mem
, int nid
, int zid
)
344 return &mem
->info
.nodeinfo
[nid
]->zoneinfo
[zid
];
347 struct cgroup_subsys_state
*mem_cgroup_css(struct mem_cgroup
*mem
)
352 static struct mem_cgroup_per_zone
*
353 page_cgroup_zoneinfo(struct page_cgroup
*pc
)
355 struct mem_cgroup
*mem
= pc
->mem_cgroup
;
356 int nid
= page_cgroup_nid(pc
);
357 int zid
= page_cgroup_zid(pc
);
362 return mem_cgroup_zoneinfo(mem
, nid
, zid
);
365 static struct mem_cgroup_tree_per_zone
*
366 soft_limit_tree_node_zone(int nid
, int zid
)
368 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
371 static struct mem_cgroup_tree_per_zone
*
372 soft_limit_tree_from_page(struct page
*page
)
374 int nid
= page_to_nid(page
);
375 int zid
= page_zonenum(page
);
377 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
381 __mem_cgroup_insert_exceeded(struct mem_cgroup
*mem
,
382 struct mem_cgroup_per_zone
*mz
,
383 struct mem_cgroup_tree_per_zone
*mctz
,
384 unsigned long long new_usage_in_excess
)
386 struct rb_node
**p
= &mctz
->rb_root
.rb_node
;
387 struct rb_node
*parent
= NULL
;
388 struct mem_cgroup_per_zone
*mz_node
;
393 mz
->usage_in_excess
= new_usage_in_excess
;
394 if (!mz
->usage_in_excess
)
398 mz_node
= rb_entry(parent
, struct mem_cgroup_per_zone
,
400 if (mz
->usage_in_excess
< mz_node
->usage_in_excess
)
403 * We can't avoid mem cgroups that are over their soft
404 * limit by the same amount
406 else if (mz
->usage_in_excess
>= mz_node
->usage_in_excess
)
409 rb_link_node(&mz
->tree_node
, parent
, p
);
410 rb_insert_color(&mz
->tree_node
, &mctz
->rb_root
);
415 __mem_cgroup_remove_exceeded(struct mem_cgroup
*mem
,
416 struct mem_cgroup_per_zone
*mz
,
417 struct mem_cgroup_tree_per_zone
*mctz
)
421 rb_erase(&mz
->tree_node
, &mctz
->rb_root
);
426 mem_cgroup_remove_exceeded(struct mem_cgroup
*mem
,
427 struct mem_cgroup_per_zone
*mz
,
428 struct mem_cgroup_tree_per_zone
*mctz
)
430 spin_lock(&mctz
->lock
);
431 __mem_cgroup_remove_exceeded(mem
, mz
, mctz
);
432 spin_unlock(&mctz
->lock
);
436 static void mem_cgroup_update_tree(struct mem_cgroup
*mem
, struct page
*page
)
438 unsigned long long excess
;
439 struct mem_cgroup_per_zone
*mz
;
440 struct mem_cgroup_tree_per_zone
*mctz
;
441 int nid
= page_to_nid(page
);
442 int zid
= page_zonenum(page
);
443 mctz
= soft_limit_tree_from_page(page
);
446 * Necessary to update all ancestors when hierarchy is used.
447 * because their event counter is not touched.
449 for (; mem
; mem
= parent_mem_cgroup(mem
)) {
450 mz
= mem_cgroup_zoneinfo(mem
, nid
, zid
);
451 excess
= res_counter_soft_limit_excess(&mem
->res
);
453 * We have to update the tree if mz is on RB-tree or
454 * mem is over its softlimit.
456 if (excess
|| mz
->on_tree
) {
457 spin_lock(&mctz
->lock
);
458 /* if on-tree, remove it */
460 __mem_cgroup_remove_exceeded(mem
, mz
, mctz
);
462 * Insert again. mz->usage_in_excess will be updated.
463 * If excess is 0, no tree ops.
465 __mem_cgroup_insert_exceeded(mem
, mz
, mctz
, excess
);
466 spin_unlock(&mctz
->lock
);
471 static void mem_cgroup_remove_from_trees(struct mem_cgroup
*mem
)
474 struct mem_cgroup_per_zone
*mz
;
475 struct mem_cgroup_tree_per_zone
*mctz
;
477 for_each_node_state(node
, N_POSSIBLE
) {
478 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
479 mz
= mem_cgroup_zoneinfo(mem
, node
, zone
);
480 mctz
= soft_limit_tree_node_zone(node
, zone
);
481 mem_cgroup_remove_exceeded(mem
, mz
, mctz
);
486 static inline unsigned long mem_cgroup_get_excess(struct mem_cgroup
*mem
)
488 return res_counter_soft_limit_excess(&mem
->res
) >> PAGE_SHIFT
;
491 static struct mem_cgroup_per_zone
*
492 __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
494 struct rb_node
*rightmost
= NULL
;
495 struct mem_cgroup_per_zone
*mz
;
499 rightmost
= rb_last(&mctz
->rb_root
);
501 goto done
; /* Nothing to reclaim from */
503 mz
= rb_entry(rightmost
, struct mem_cgroup_per_zone
, tree_node
);
505 * Remove the node now but someone else can add it back,
506 * we will to add it back at the end of reclaim to its correct
507 * position in the tree.
509 __mem_cgroup_remove_exceeded(mz
->mem
, mz
, mctz
);
510 if (!res_counter_soft_limit_excess(&mz
->mem
->res
) ||
511 !css_tryget(&mz
->mem
->css
))
517 static struct mem_cgroup_per_zone
*
518 mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
520 struct mem_cgroup_per_zone
*mz
;
522 spin_lock(&mctz
->lock
);
523 mz
= __mem_cgroup_largest_soft_limit_node(mctz
);
524 spin_unlock(&mctz
->lock
);
528 static s64
mem_cgroup_read_stat(struct mem_cgroup
*mem
,
529 enum mem_cgroup_stat_index idx
)
534 for_each_possible_cpu(cpu
)
535 val
+= per_cpu(mem
->stat
->count
[idx
], cpu
);
539 static s64
mem_cgroup_local_usage(struct mem_cgroup
*mem
)
543 ret
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_RSS
);
544 ret
+= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_CACHE
);
548 static void mem_cgroup_swap_statistics(struct mem_cgroup
*mem
,
551 int val
= (charge
) ? 1 : -1;
552 this_cpu_add(mem
->stat
->count
[MEM_CGROUP_STAT_SWAPOUT
], val
);
555 static void mem_cgroup_charge_statistics(struct mem_cgroup
*mem
,
556 struct page_cgroup
*pc
,
559 int val
= (charge
) ? 1 : -1;
563 if (PageCgroupCache(pc
))
564 __this_cpu_add(mem
->stat
->count
[MEM_CGROUP_STAT_CACHE
], val
);
566 __this_cpu_add(mem
->stat
->count
[MEM_CGROUP_STAT_RSS
], val
);
569 __this_cpu_inc(mem
->stat
->count
[MEM_CGROUP_STAT_PGPGIN_COUNT
]);
571 __this_cpu_inc(mem
->stat
->count
[MEM_CGROUP_STAT_PGPGOUT_COUNT
]);
572 __this_cpu_inc(mem
->stat
->count
[MEM_CGROUP_EVENTS
]);
577 static unsigned long mem_cgroup_get_local_zonestat(struct mem_cgroup
*mem
,
581 struct mem_cgroup_per_zone
*mz
;
584 for_each_online_node(nid
)
585 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
586 mz
= mem_cgroup_zoneinfo(mem
, nid
, zid
);
587 total
+= MEM_CGROUP_ZSTAT(mz
, idx
);
592 static bool __memcg_event_check(struct mem_cgroup
*mem
, int event_mask_shift
)
596 val
= this_cpu_read(mem
->stat
->count
[MEM_CGROUP_EVENTS
]);
598 return !(val
& ((1 << event_mask_shift
) - 1));
602 * Check events in order.
605 static void memcg_check_events(struct mem_cgroup
*mem
, struct page
*page
)
607 /* threshold event is triggered in finer grain than soft limit */
608 if (unlikely(__memcg_event_check(mem
, THRESHOLDS_EVENTS_THRESH
))) {
609 mem_cgroup_threshold(mem
);
610 if (unlikely(__memcg_event_check(mem
, SOFTLIMIT_EVENTS_THRESH
)))
611 mem_cgroup_update_tree(mem
, page
);
615 static struct mem_cgroup
*mem_cgroup_from_cont(struct cgroup
*cont
)
617 return container_of(cgroup_subsys_state(cont
,
618 mem_cgroup_subsys_id
), struct mem_cgroup
,
622 struct mem_cgroup
*mem_cgroup_from_task(struct task_struct
*p
)
625 * mm_update_next_owner() may clear mm->owner to NULL
626 * if it races with swapoff, page migration, etc.
627 * So this can be called with p == NULL.
632 return container_of(task_subsys_state(p
, mem_cgroup_subsys_id
),
633 struct mem_cgroup
, css
);
636 static struct mem_cgroup
*try_get_mem_cgroup_from_mm(struct mm_struct
*mm
)
638 struct mem_cgroup
*mem
= NULL
;
643 * Because we have no locks, mm->owner's may be being moved to other
644 * cgroup. We use css_tryget() here even if this looks
645 * pessimistic (rather than adding locks here).
649 mem
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
652 } while (!css_tryget(&mem
->css
));
658 * Call callback function against all cgroup under hierarchy tree.
660 static int mem_cgroup_walk_tree(struct mem_cgroup
*root
, void *data
,
661 int (*func
)(struct mem_cgroup
*, void *))
663 int found
, ret
, nextid
;
664 struct cgroup_subsys_state
*css
;
665 struct mem_cgroup
*mem
;
667 if (!root
->use_hierarchy
)
668 return (*func
)(root
, data
);
676 css
= css_get_next(&mem_cgroup_subsys
, nextid
, &root
->css
,
678 if (css
&& css_tryget(css
))
679 mem
= container_of(css
, struct mem_cgroup
, css
);
683 ret
= (*func
)(mem
, data
);
687 } while (!ret
&& css
);
692 static inline bool mem_cgroup_is_root(struct mem_cgroup
*mem
)
694 return (mem
== root_mem_cgroup
);
698 * Following LRU functions are allowed to be used without PCG_LOCK.
699 * Operations are called by routine of global LRU independently from memcg.
700 * What we have to take care of here is validness of pc->mem_cgroup.
702 * Changes to pc->mem_cgroup happens when
705 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
706 * It is added to LRU before charge.
707 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
708 * When moving account, the page is not on LRU. It's isolated.
711 void mem_cgroup_del_lru_list(struct page
*page
, enum lru_list lru
)
713 struct page_cgroup
*pc
;
714 struct mem_cgroup_per_zone
*mz
;
716 if (mem_cgroup_disabled())
718 pc
= lookup_page_cgroup(page
);
719 /* can happen while we handle swapcache. */
720 if (!TestClearPageCgroupAcctLRU(pc
))
722 VM_BUG_ON(!pc
->mem_cgroup
);
724 * We don't check PCG_USED bit. It's cleared when the "page" is finally
725 * removed from global LRU.
727 mz
= page_cgroup_zoneinfo(pc
);
728 MEM_CGROUP_ZSTAT(mz
, lru
) -= 1;
729 if (mem_cgroup_is_root(pc
->mem_cgroup
))
731 VM_BUG_ON(list_empty(&pc
->lru
));
732 list_del_init(&pc
->lru
);
736 void mem_cgroup_del_lru(struct page
*page
)
738 mem_cgroup_del_lru_list(page
, page_lru(page
));
741 void mem_cgroup_rotate_lru_list(struct page
*page
, enum lru_list lru
)
743 struct mem_cgroup_per_zone
*mz
;
744 struct page_cgroup
*pc
;
746 if (mem_cgroup_disabled())
749 pc
= lookup_page_cgroup(page
);
751 * Used bit is set without atomic ops but after smp_wmb().
752 * For making pc->mem_cgroup visible, insert smp_rmb() here.
755 /* unused or root page is not rotated. */
756 if (!PageCgroupUsed(pc
) || mem_cgroup_is_root(pc
->mem_cgroup
))
758 mz
= page_cgroup_zoneinfo(pc
);
759 list_move(&pc
->lru
, &mz
->lists
[lru
]);
762 void mem_cgroup_add_lru_list(struct page
*page
, enum lru_list lru
)
764 struct page_cgroup
*pc
;
765 struct mem_cgroup_per_zone
*mz
;
767 if (mem_cgroup_disabled())
769 pc
= lookup_page_cgroup(page
);
770 VM_BUG_ON(PageCgroupAcctLRU(pc
));
772 * Used bit is set without atomic ops but after smp_wmb().
773 * For making pc->mem_cgroup visible, insert smp_rmb() here.
776 if (!PageCgroupUsed(pc
))
779 mz
= page_cgroup_zoneinfo(pc
);
780 MEM_CGROUP_ZSTAT(mz
, lru
) += 1;
781 SetPageCgroupAcctLRU(pc
);
782 if (mem_cgroup_is_root(pc
->mem_cgroup
))
784 list_add(&pc
->lru
, &mz
->lists
[lru
]);
788 * At handling SwapCache, pc->mem_cgroup may be changed while it's linked to
789 * lru because the page may.be reused after it's fully uncharged (because of
790 * SwapCache behavior).To handle that, unlink page_cgroup from LRU when charge
791 * it again. This function is only used to charge SwapCache. It's done under
792 * lock_page and expected that zone->lru_lock is never held.
794 static void mem_cgroup_lru_del_before_commit_swapcache(struct page
*page
)
797 struct zone
*zone
= page_zone(page
);
798 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
800 spin_lock_irqsave(&zone
->lru_lock
, flags
);
802 * Forget old LRU when this page_cgroup is *not* used. This Used bit
803 * is guarded by lock_page() because the page is SwapCache.
805 if (!PageCgroupUsed(pc
))
806 mem_cgroup_del_lru_list(page
, page_lru(page
));
807 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
810 static void mem_cgroup_lru_add_after_commit_swapcache(struct page
*page
)
813 struct zone
*zone
= page_zone(page
);
814 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
816 spin_lock_irqsave(&zone
->lru_lock
, flags
);
817 /* link when the page is linked to LRU but page_cgroup isn't */
818 if (PageLRU(page
) && !PageCgroupAcctLRU(pc
))
819 mem_cgroup_add_lru_list(page
, page_lru(page
));
820 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
824 void mem_cgroup_move_lists(struct page
*page
,
825 enum lru_list from
, enum lru_list to
)
827 if (mem_cgroup_disabled())
829 mem_cgroup_del_lru_list(page
, from
);
830 mem_cgroup_add_lru_list(page
, to
);
833 int task_in_mem_cgroup(struct task_struct
*task
, const struct mem_cgroup
*mem
)
836 struct mem_cgroup
*curr
= NULL
;
840 curr
= try_get_mem_cgroup_from_mm(task
->mm
);
846 * We should check use_hierarchy of "mem" not "curr". Because checking
847 * use_hierarchy of "curr" here make this function true if hierarchy is
848 * enabled in "curr" and "curr" is a child of "mem" in *cgroup*
849 * hierarchy(even if use_hierarchy is disabled in "mem").
851 if (mem
->use_hierarchy
)
852 ret
= css_is_ancestor(&curr
->css
, &mem
->css
);
859 static int calc_inactive_ratio(struct mem_cgroup
*memcg
, unsigned long *present_pages
)
861 unsigned long active
;
862 unsigned long inactive
;
864 unsigned long inactive_ratio
;
866 inactive
= mem_cgroup_get_local_zonestat(memcg
, LRU_INACTIVE_ANON
);
867 active
= mem_cgroup_get_local_zonestat(memcg
, LRU_ACTIVE_ANON
);
869 gb
= (inactive
+ active
) >> (30 - PAGE_SHIFT
);
871 inactive_ratio
= int_sqrt(10 * gb
);
876 present_pages
[0] = inactive
;
877 present_pages
[1] = active
;
880 return inactive_ratio
;
883 int mem_cgroup_inactive_anon_is_low(struct mem_cgroup
*memcg
)
885 unsigned long active
;
886 unsigned long inactive
;
887 unsigned long present_pages
[2];
888 unsigned long inactive_ratio
;
890 inactive_ratio
= calc_inactive_ratio(memcg
, present_pages
);
892 inactive
= present_pages
[0];
893 active
= present_pages
[1];
895 if (inactive
* inactive_ratio
< active
)
901 int mem_cgroup_inactive_file_is_low(struct mem_cgroup
*memcg
)
903 unsigned long active
;
904 unsigned long inactive
;
906 inactive
= mem_cgroup_get_local_zonestat(memcg
, LRU_INACTIVE_FILE
);
907 active
= mem_cgroup_get_local_zonestat(memcg
, LRU_ACTIVE_FILE
);
909 return (active
> inactive
);
912 unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup
*memcg
,
916 int nid
= zone
->zone_pgdat
->node_id
;
917 int zid
= zone_idx(zone
);
918 struct mem_cgroup_per_zone
*mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
920 return MEM_CGROUP_ZSTAT(mz
, lru
);
923 struct zone_reclaim_stat
*mem_cgroup_get_reclaim_stat(struct mem_cgroup
*memcg
,
926 int nid
= zone
->zone_pgdat
->node_id
;
927 int zid
= zone_idx(zone
);
928 struct mem_cgroup_per_zone
*mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
930 return &mz
->reclaim_stat
;
933 struct zone_reclaim_stat
*
934 mem_cgroup_get_reclaim_stat_from_page(struct page
*page
)
936 struct page_cgroup
*pc
;
937 struct mem_cgroup_per_zone
*mz
;
939 if (mem_cgroup_disabled())
942 pc
= lookup_page_cgroup(page
);
944 * Used bit is set without atomic ops but after smp_wmb().
945 * For making pc->mem_cgroup visible, insert smp_rmb() here.
948 if (!PageCgroupUsed(pc
))
951 mz
= page_cgroup_zoneinfo(pc
);
955 return &mz
->reclaim_stat
;
958 unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan
,
959 struct list_head
*dst
,
960 unsigned long *scanned
, int order
,
961 int mode
, struct zone
*z
,
962 struct mem_cgroup
*mem_cont
,
963 int active
, int file
)
965 unsigned long nr_taken
= 0;
969 struct list_head
*src
;
970 struct page_cgroup
*pc
, *tmp
;
971 int nid
= z
->zone_pgdat
->node_id
;
972 int zid
= zone_idx(z
);
973 struct mem_cgroup_per_zone
*mz
;
974 int lru
= LRU_FILE
* file
+ active
;
978 mz
= mem_cgroup_zoneinfo(mem_cont
, nid
, zid
);
979 src
= &mz
->lists
[lru
];
982 list_for_each_entry_safe_reverse(pc
, tmp
, src
, lru
) {
983 if (scan
>= nr_to_scan
)
987 if (unlikely(!PageCgroupUsed(pc
)))
989 if (unlikely(!PageLRU(page
)))
993 ret
= __isolate_lru_page(page
, mode
, file
);
996 list_move(&page
->lru
, dst
);
997 mem_cgroup_del_lru(page
);
1001 /* we don't affect global LRU but rotate in our LRU */
1002 mem_cgroup_rotate_lru_list(page
, page_lru(page
));
1013 #define mem_cgroup_from_res_counter(counter, member) \
1014 container_of(counter, struct mem_cgroup, member)
1016 static bool mem_cgroup_check_under_limit(struct mem_cgroup
*mem
)
1018 if (do_swap_account
) {
1019 if (res_counter_check_under_limit(&mem
->res
) &&
1020 res_counter_check_under_limit(&mem
->memsw
))
1023 if (res_counter_check_under_limit(&mem
->res
))
1028 static unsigned int get_swappiness(struct mem_cgroup
*memcg
)
1030 struct cgroup
*cgrp
= memcg
->css
.cgroup
;
1031 unsigned int swappiness
;
1034 if (cgrp
->parent
== NULL
)
1035 return vm_swappiness
;
1037 spin_lock(&memcg
->reclaim_param_lock
);
1038 swappiness
= memcg
->swappiness
;
1039 spin_unlock(&memcg
->reclaim_param_lock
);
1044 static int mem_cgroup_count_children_cb(struct mem_cgroup
*mem
, void *data
)
1052 * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode.
1053 * @memcg: The memory cgroup that went over limit
1054 * @p: Task that is going to be killed
1056 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1059 void mem_cgroup_print_oom_info(struct mem_cgroup
*memcg
, struct task_struct
*p
)
1061 struct cgroup
*task_cgrp
;
1062 struct cgroup
*mem_cgrp
;
1064 * Need a buffer in BSS, can't rely on allocations. The code relies
1065 * on the assumption that OOM is serialized for memory controller.
1066 * If this assumption is broken, revisit this code.
1068 static char memcg_name
[PATH_MAX
];
1077 mem_cgrp
= memcg
->css
.cgroup
;
1078 task_cgrp
= task_cgroup(p
, mem_cgroup_subsys_id
);
1080 ret
= cgroup_path(task_cgrp
, memcg_name
, PATH_MAX
);
1083 * Unfortunately, we are unable to convert to a useful name
1084 * But we'll still print out the usage information
1091 printk(KERN_INFO
"Task in %s killed", memcg_name
);
1094 ret
= cgroup_path(mem_cgrp
, memcg_name
, PATH_MAX
);
1102 * Continues from above, so we don't need an KERN_ level
1104 printk(KERN_CONT
" as a result of limit of %s\n", memcg_name
);
1107 printk(KERN_INFO
"memory: usage %llukB, limit %llukB, failcnt %llu\n",
1108 res_counter_read_u64(&memcg
->res
, RES_USAGE
) >> 10,
1109 res_counter_read_u64(&memcg
->res
, RES_LIMIT
) >> 10,
1110 res_counter_read_u64(&memcg
->res
, RES_FAILCNT
));
1111 printk(KERN_INFO
"memory+swap: usage %llukB, limit %llukB, "
1113 res_counter_read_u64(&memcg
->memsw
, RES_USAGE
) >> 10,
1114 res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
) >> 10,
1115 res_counter_read_u64(&memcg
->memsw
, RES_FAILCNT
));
1119 * This function returns the number of memcg under hierarchy tree. Returns
1120 * 1(self count) if no children.
1122 static int mem_cgroup_count_children(struct mem_cgroup
*mem
)
1125 mem_cgroup_walk_tree(mem
, &num
, mem_cgroup_count_children_cb
);
1130 * Visit the first child (need not be the first child as per the ordering
1131 * of the cgroup list, since we track last_scanned_child) of @mem and use
1132 * that to reclaim free pages from.
1134 static struct mem_cgroup
*
1135 mem_cgroup_select_victim(struct mem_cgroup
*root_mem
)
1137 struct mem_cgroup
*ret
= NULL
;
1138 struct cgroup_subsys_state
*css
;
1141 if (!root_mem
->use_hierarchy
) {
1142 css_get(&root_mem
->css
);
1148 nextid
= root_mem
->last_scanned_child
+ 1;
1149 css
= css_get_next(&mem_cgroup_subsys
, nextid
, &root_mem
->css
,
1151 if (css
&& css_tryget(css
))
1152 ret
= container_of(css
, struct mem_cgroup
, css
);
1155 /* Updates scanning parameter */
1156 spin_lock(&root_mem
->reclaim_param_lock
);
1158 /* this means start scan from ID:1 */
1159 root_mem
->last_scanned_child
= 0;
1161 root_mem
->last_scanned_child
= found
;
1162 spin_unlock(&root_mem
->reclaim_param_lock
);
1169 * Scan the hierarchy if needed to reclaim memory. We remember the last child
1170 * we reclaimed from, so that we don't end up penalizing one child extensively
1171 * based on its position in the children list.
1173 * root_mem is the original ancestor that we've been reclaim from.
1175 * We give up and return to the caller when we visit root_mem twice.
1176 * (other groups can be removed while we're walking....)
1178 * If shrink==true, for avoiding to free too much, this returns immedieately.
1180 static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup
*root_mem
,
1183 unsigned long reclaim_options
)
1185 struct mem_cgroup
*victim
;
1188 bool noswap
= reclaim_options
& MEM_CGROUP_RECLAIM_NOSWAP
;
1189 bool shrink
= reclaim_options
& MEM_CGROUP_RECLAIM_SHRINK
;
1190 bool check_soft
= reclaim_options
& MEM_CGROUP_RECLAIM_SOFT
;
1191 unsigned long excess
= mem_cgroup_get_excess(root_mem
);
1193 /* If memsw_is_minimum==1, swap-out is of-no-use. */
1194 if (root_mem
->memsw_is_minimum
)
1198 victim
= mem_cgroup_select_victim(root_mem
);
1199 if (victim
== root_mem
) {
1202 drain_all_stock_async();
1205 * If we have not been able to reclaim
1206 * anything, it might because there are
1207 * no reclaimable pages under this hierarchy
1209 if (!check_soft
|| !total
) {
1210 css_put(&victim
->css
);
1214 * We want to do more targetted reclaim.
1215 * excess >> 2 is not to excessive so as to
1216 * reclaim too much, nor too less that we keep
1217 * coming back to reclaim from this cgroup
1219 if (total
>= (excess
>> 2) ||
1220 (loop
> MEM_CGROUP_MAX_RECLAIM_LOOPS
)) {
1221 css_put(&victim
->css
);
1226 if (!mem_cgroup_local_usage(victim
)) {
1227 /* this cgroup's local usage == 0 */
1228 css_put(&victim
->css
);
1231 /* we use swappiness of local cgroup */
1233 ret
= mem_cgroup_shrink_node_zone(victim
, gfp_mask
,
1234 noswap
, get_swappiness(victim
), zone
,
1235 zone
->zone_pgdat
->node_id
);
1237 ret
= try_to_free_mem_cgroup_pages(victim
, gfp_mask
,
1238 noswap
, get_swappiness(victim
));
1239 css_put(&victim
->css
);
1241 * At shrinking usage, we can't check we should stop here or
1242 * reclaim more. It's depends on callers. last_scanned_child
1243 * will work enough for keeping fairness under tree.
1249 if (res_counter_check_under_soft_limit(&root_mem
->res
))
1251 } else if (mem_cgroup_check_under_limit(root_mem
))
1257 static int mem_cgroup_oom_lock_cb(struct mem_cgroup
*mem
, void *data
)
1259 int *val
= (int *)data
;
1262 * Logically, we can stop scanning immediately when we find
1263 * a memcg is already locked. But condidering unlock ops and
1264 * creation/removal of memcg, scan-all is simple operation.
1266 x
= atomic_inc_return(&mem
->oom_lock
);
1267 *val
= max(x
, *val
);
1271 * Check OOM-Killer is already running under our hierarchy.
1272 * If someone is running, return false.
1274 static bool mem_cgroup_oom_lock(struct mem_cgroup
*mem
)
1278 mem_cgroup_walk_tree(mem
, &lock_count
, mem_cgroup_oom_lock_cb
);
1280 if (lock_count
== 1)
1285 static int mem_cgroup_oom_unlock_cb(struct mem_cgroup
*mem
, void *data
)
1288 * When a new child is created while the hierarchy is under oom,
1289 * mem_cgroup_oom_lock() may not be called. We have to use
1290 * atomic_add_unless() here.
1292 atomic_add_unless(&mem
->oom_lock
, -1, 0);
1296 static void mem_cgroup_oom_unlock(struct mem_cgroup
*mem
)
1298 mem_cgroup_walk_tree(mem
, NULL
, mem_cgroup_oom_unlock_cb
);
1301 static DEFINE_MUTEX(memcg_oom_mutex
);
1302 static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq
);
1304 struct oom_wait_info
{
1305 struct mem_cgroup
*mem
;
1309 static int memcg_oom_wake_function(wait_queue_t
*wait
,
1310 unsigned mode
, int sync
, void *arg
)
1312 struct mem_cgroup
*wake_mem
= (struct mem_cgroup
*)arg
;
1313 struct oom_wait_info
*oom_wait_info
;
1315 oom_wait_info
= container_of(wait
, struct oom_wait_info
, wait
);
1317 if (oom_wait_info
->mem
== wake_mem
)
1319 /* if no hierarchy, no match */
1320 if (!oom_wait_info
->mem
->use_hierarchy
|| !wake_mem
->use_hierarchy
)
1323 * Both of oom_wait_info->mem and wake_mem are stable under us.
1324 * Then we can use css_is_ancestor without taking care of RCU.
1326 if (!css_is_ancestor(&oom_wait_info
->mem
->css
, &wake_mem
->css
) &&
1327 !css_is_ancestor(&wake_mem
->css
, &oom_wait_info
->mem
->css
))
1331 return autoremove_wake_function(wait
, mode
, sync
, arg
);
1334 static void memcg_wakeup_oom(struct mem_cgroup
*mem
)
1336 /* for filtering, pass "mem" as argument. */
1337 __wake_up(&memcg_oom_waitq
, TASK_NORMAL
, 0, mem
);
1340 static void memcg_oom_recover(struct mem_cgroup
*mem
)
1342 if (atomic_read(&mem
->oom_lock
))
1343 memcg_wakeup_oom(mem
);
1347 * try to call OOM killer. returns false if we should exit memory-reclaim loop.
1349 bool mem_cgroup_handle_oom(struct mem_cgroup
*mem
, gfp_t mask
)
1351 struct oom_wait_info owait
;
1352 bool locked
, need_to_kill
;
1355 owait
.wait
.flags
= 0;
1356 owait
.wait
.func
= memcg_oom_wake_function
;
1357 owait
.wait
.private = current
;
1358 INIT_LIST_HEAD(&owait
.wait
.task_list
);
1359 need_to_kill
= true;
1360 /* At first, try to OOM lock hierarchy under mem.*/
1361 mutex_lock(&memcg_oom_mutex
);
1362 locked
= mem_cgroup_oom_lock(mem
);
1364 * Even if signal_pending(), we can't quit charge() loop without
1365 * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
1366 * under OOM is always welcomed, use TASK_KILLABLE here.
1368 prepare_to_wait(&memcg_oom_waitq
, &owait
.wait
, TASK_KILLABLE
);
1369 if (!locked
|| mem
->oom_kill_disable
)
1370 need_to_kill
= false;
1372 mem_cgroup_oom_notify(mem
);
1373 mutex_unlock(&memcg_oom_mutex
);
1376 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1377 mem_cgroup_out_of_memory(mem
, mask
);
1380 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1382 mutex_lock(&memcg_oom_mutex
);
1383 mem_cgroup_oom_unlock(mem
);
1384 memcg_wakeup_oom(mem
);
1385 mutex_unlock(&memcg_oom_mutex
);
1387 if (test_thread_flag(TIF_MEMDIE
) || fatal_signal_pending(current
))
1389 /* Give chance to dying process */
1390 schedule_timeout(1);
1395 * Currently used to update mapped file statistics, but the routine can be
1396 * generalized to update other statistics as well.
1398 void mem_cgroup_update_file_mapped(struct page
*page
, int val
)
1400 struct mem_cgroup
*mem
;
1401 struct page_cgroup
*pc
;
1403 pc
= lookup_page_cgroup(page
);
1407 lock_page_cgroup(pc
);
1408 mem
= pc
->mem_cgroup
;
1409 if (!mem
|| !PageCgroupUsed(pc
))
1413 * Preemption is already disabled. We can use __this_cpu_xxx
1416 __this_cpu_inc(mem
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
1417 SetPageCgroupFileMapped(pc
);
1419 __this_cpu_dec(mem
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
1420 ClearPageCgroupFileMapped(pc
);
1424 unlock_page_cgroup(pc
);
1428 * size of first charge trial. "32" comes from vmscan.c's magic value.
1429 * TODO: maybe necessary to use big numbers in big irons.
1431 #define CHARGE_SIZE (32 * PAGE_SIZE)
1432 struct memcg_stock_pcp
{
1433 struct mem_cgroup
*cached
; /* this never be root cgroup */
1435 struct work_struct work
;
1437 static DEFINE_PER_CPU(struct memcg_stock_pcp
, memcg_stock
);
1438 static atomic_t memcg_drain_count
;
1441 * Try to consume stocked charge on this cpu. If success, PAGE_SIZE is consumed
1442 * from local stock and true is returned. If the stock is 0 or charges from a
1443 * cgroup which is not current target, returns false. This stock will be
1446 static bool consume_stock(struct mem_cgroup
*mem
)
1448 struct memcg_stock_pcp
*stock
;
1451 stock
= &get_cpu_var(memcg_stock
);
1452 if (mem
== stock
->cached
&& stock
->charge
)
1453 stock
->charge
-= PAGE_SIZE
;
1454 else /* need to call res_counter_charge */
1456 put_cpu_var(memcg_stock
);
1461 * Returns stocks cached in percpu to res_counter and reset cached information.
1463 static void drain_stock(struct memcg_stock_pcp
*stock
)
1465 struct mem_cgroup
*old
= stock
->cached
;
1467 if (stock
->charge
) {
1468 res_counter_uncharge(&old
->res
, stock
->charge
);
1469 if (do_swap_account
)
1470 res_counter_uncharge(&old
->memsw
, stock
->charge
);
1472 stock
->cached
= NULL
;
1477 * This must be called under preempt disabled or must be called by
1478 * a thread which is pinned to local cpu.
1480 static void drain_local_stock(struct work_struct
*dummy
)
1482 struct memcg_stock_pcp
*stock
= &__get_cpu_var(memcg_stock
);
1487 * Cache charges(val) which is from res_counter, to local per_cpu area.
1488 * This will be consumed by consume_stock() function, later.
1490 static void refill_stock(struct mem_cgroup
*mem
, int val
)
1492 struct memcg_stock_pcp
*stock
= &get_cpu_var(memcg_stock
);
1494 if (stock
->cached
!= mem
) { /* reset if necessary */
1496 stock
->cached
= mem
;
1498 stock
->charge
+= val
;
1499 put_cpu_var(memcg_stock
);
1503 * Tries to drain stocked charges in other cpus. This function is asynchronous
1504 * and just put a work per cpu for draining localy on each cpu. Caller can
1505 * expects some charges will be back to res_counter later but cannot wait for
1508 static void drain_all_stock_async(void)
1511 /* This function is for scheduling "drain" in asynchronous way.
1512 * The result of "drain" is not directly handled by callers. Then,
1513 * if someone is calling drain, we don't have to call drain more.
1514 * Anyway, WORK_STRUCT_PENDING check in queue_work_on() will catch if
1515 * there is a race. We just do loose check here.
1517 if (atomic_read(&memcg_drain_count
))
1519 /* Notify other cpus that system-wide "drain" is running */
1520 atomic_inc(&memcg_drain_count
);
1522 for_each_online_cpu(cpu
) {
1523 struct memcg_stock_pcp
*stock
= &per_cpu(memcg_stock
, cpu
);
1524 schedule_work_on(cpu
, &stock
->work
);
1527 atomic_dec(&memcg_drain_count
);
1528 /* We don't wait for flush_work */
1531 /* This is a synchronous drain interface. */
1532 static void drain_all_stock_sync(void)
1534 /* called when force_empty is called */
1535 atomic_inc(&memcg_drain_count
);
1536 schedule_on_each_cpu(drain_local_stock
);
1537 atomic_dec(&memcg_drain_count
);
1540 static int __cpuinit
memcg_stock_cpu_callback(struct notifier_block
*nb
,
1541 unsigned long action
,
1544 int cpu
= (unsigned long)hcpu
;
1545 struct memcg_stock_pcp
*stock
;
1547 if (action
!= CPU_DEAD
)
1549 stock
= &per_cpu(memcg_stock
, cpu
);
1555 * Unlike exported interface, "oom" parameter is added. if oom==true,
1556 * oom-killer can be invoked.
1558 static int __mem_cgroup_try_charge(struct mm_struct
*mm
,
1559 gfp_t gfp_mask
, struct mem_cgroup
**memcg
, bool oom
)
1561 struct mem_cgroup
*mem
, *mem_over_limit
;
1562 int nr_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
1563 struct res_counter
*fail_res
;
1564 int csize
= CHARGE_SIZE
;
1567 * Unlike gloval-vm's OOM-kill, we're not in memory shortage
1568 * in system level. So, allow to go ahead dying process in addition to
1571 if (unlikely(test_thread_flag(TIF_MEMDIE
)
1572 || fatal_signal_pending(current
)))
1576 * We always charge the cgroup the mm_struct belongs to.
1577 * The mm_struct's mem_cgroup changes on task migration if the
1578 * thread group leader migrates. It's possible that mm is not
1579 * set, if so charge the init_mm (happens for pagecache usage).
1583 mem
= try_get_mem_cgroup_from_mm(mm
);
1591 VM_BUG_ON(css_is_removed(&mem
->css
));
1592 if (mem_cgroup_is_root(mem
))
1597 unsigned long flags
= 0;
1599 if (consume_stock(mem
))
1602 ret
= res_counter_charge(&mem
->res
, csize
, &fail_res
);
1604 if (!do_swap_account
)
1606 ret
= res_counter_charge(&mem
->memsw
, csize
, &fail_res
);
1609 /* mem+swap counter fails */
1610 res_counter_uncharge(&mem
->res
, csize
);
1611 flags
|= MEM_CGROUP_RECLAIM_NOSWAP
;
1612 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
,
1615 /* mem counter fails */
1616 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
,
1619 /* reduce request size and retry */
1620 if (csize
> PAGE_SIZE
) {
1624 if (!(gfp_mask
& __GFP_WAIT
))
1627 ret
= mem_cgroup_hierarchical_reclaim(mem_over_limit
, NULL
,
1633 * try_to_free_mem_cgroup_pages() might not give us a full
1634 * picture of reclaim. Some pages are reclaimed and might be
1635 * moved to swap cache or just unmapped from the cgroup.
1636 * Check the limit again to see if the reclaim reduced the
1637 * current usage of the cgroup before giving up
1640 if (mem_cgroup_check_under_limit(mem_over_limit
))
1643 /* try to avoid oom while someone is moving charge */
1644 if (mc
.moving_task
&& current
!= mc
.moving_task
) {
1645 struct mem_cgroup
*from
, *to
;
1646 bool do_continue
= false;
1648 * There is a small race that "from" or "to" can be
1649 * freed by rmdir, so we use css_tryget().
1653 if (from
&& css_tryget(&from
->css
)) {
1654 if (mem_over_limit
->use_hierarchy
)
1655 do_continue
= css_is_ancestor(
1657 &mem_over_limit
->css
);
1659 do_continue
= (from
== mem_over_limit
);
1660 css_put(&from
->css
);
1662 if (!do_continue
&& to
&& css_tryget(&to
->css
)) {
1663 if (mem_over_limit
->use_hierarchy
)
1664 do_continue
= css_is_ancestor(
1666 &mem_over_limit
->css
);
1668 do_continue
= (to
== mem_over_limit
);
1673 prepare_to_wait(&mc
.waitq
, &wait
,
1674 TASK_INTERRUPTIBLE
);
1675 /* moving charge context might have finished. */
1678 finish_wait(&mc
.waitq
, &wait
);
1683 if (!nr_retries
--) {
1686 if (mem_cgroup_handle_oom(mem_over_limit
, gfp_mask
)) {
1687 nr_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
1690 /* When we reach here, current task is dying .*/
1695 if (csize
> PAGE_SIZE
)
1696 refill_stock(mem
, csize
- PAGE_SIZE
);
1708 * Somemtimes we have to undo a charge we got by try_charge().
1709 * This function is for that and do uncharge, put css's refcnt.
1710 * gotten by try_charge().
1712 static void __mem_cgroup_cancel_charge(struct mem_cgroup
*mem
,
1713 unsigned long count
)
1715 if (!mem_cgroup_is_root(mem
)) {
1716 res_counter_uncharge(&mem
->res
, PAGE_SIZE
* count
);
1717 if (do_swap_account
)
1718 res_counter_uncharge(&mem
->memsw
, PAGE_SIZE
* count
);
1719 VM_BUG_ON(test_bit(CSS_ROOT
, &mem
->css
.flags
));
1720 WARN_ON_ONCE(count
> INT_MAX
);
1721 __css_put(&mem
->css
, (int)count
);
1723 /* we don't need css_put for root */
1726 static void mem_cgroup_cancel_charge(struct mem_cgroup
*mem
)
1728 __mem_cgroup_cancel_charge(mem
, 1);
1732 * A helper function to get mem_cgroup from ID. must be called under
1733 * rcu_read_lock(). The caller must check css_is_removed() or some if
1734 * it's concern. (dropping refcnt from swap can be called against removed
1737 static struct mem_cgroup
*mem_cgroup_lookup(unsigned short id
)
1739 struct cgroup_subsys_state
*css
;
1741 /* ID 0 is unused ID */
1744 css
= css_lookup(&mem_cgroup_subsys
, id
);
1747 return container_of(css
, struct mem_cgroup
, css
);
1750 struct mem_cgroup
*try_get_mem_cgroup_from_page(struct page
*page
)
1752 struct mem_cgroup
*mem
= NULL
;
1753 struct page_cgroup
*pc
;
1757 VM_BUG_ON(!PageLocked(page
));
1759 pc
= lookup_page_cgroup(page
);
1760 lock_page_cgroup(pc
);
1761 if (PageCgroupUsed(pc
)) {
1762 mem
= pc
->mem_cgroup
;
1763 if (mem
&& !css_tryget(&mem
->css
))
1765 } else if (PageSwapCache(page
)) {
1766 ent
.val
= page_private(page
);
1767 id
= lookup_swap_cgroup(ent
);
1769 mem
= mem_cgroup_lookup(id
);
1770 if (mem
&& !css_tryget(&mem
->css
))
1774 unlock_page_cgroup(pc
);
1779 * commit a charge got by __mem_cgroup_try_charge() and makes page_cgroup to be
1780 * USED state. If already USED, uncharge and return.
1783 static void __mem_cgroup_commit_charge(struct mem_cgroup
*mem
,
1784 struct page_cgroup
*pc
,
1785 enum charge_type ctype
)
1787 /* try_charge() can return NULL to *memcg, taking care of it. */
1791 lock_page_cgroup(pc
);
1792 if (unlikely(PageCgroupUsed(pc
))) {
1793 unlock_page_cgroup(pc
);
1794 mem_cgroup_cancel_charge(mem
);
1798 pc
->mem_cgroup
= mem
;
1800 * We access a page_cgroup asynchronously without lock_page_cgroup().
1801 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
1802 * is accessed after testing USED bit. To make pc->mem_cgroup visible
1803 * before USED bit, we need memory barrier here.
1804 * See mem_cgroup_add_lru_list(), etc.
1808 case MEM_CGROUP_CHARGE_TYPE_CACHE
:
1809 case MEM_CGROUP_CHARGE_TYPE_SHMEM
:
1810 SetPageCgroupCache(pc
);
1811 SetPageCgroupUsed(pc
);
1813 case MEM_CGROUP_CHARGE_TYPE_MAPPED
:
1814 ClearPageCgroupCache(pc
);
1815 SetPageCgroupUsed(pc
);
1821 mem_cgroup_charge_statistics(mem
, pc
, true);
1823 unlock_page_cgroup(pc
);
1825 * "charge_statistics" updated event counter. Then, check it.
1826 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
1827 * if they exceeds softlimit.
1829 memcg_check_events(mem
, pc
->page
);
1833 * __mem_cgroup_move_account - move account of the page
1834 * @pc: page_cgroup of the page.
1835 * @from: mem_cgroup which the page is moved from.
1836 * @to: mem_cgroup which the page is moved to. @from != @to.
1837 * @uncharge: whether we should call uncharge and css_put against @from.
1839 * The caller must confirm following.
1840 * - page is not on LRU (isolate_page() is useful.)
1841 * - the pc is locked, used, and ->mem_cgroup points to @from.
1843 * This function doesn't do "charge" nor css_get to new cgroup. It should be
1844 * done by a caller(__mem_cgroup_try_charge would be usefull). If @uncharge is
1845 * true, this function does "uncharge" from old cgroup, but it doesn't if
1846 * @uncharge is false, so a caller should do "uncharge".
1849 static void __mem_cgroup_move_account(struct page_cgroup
*pc
,
1850 struct mem_cgroup
*from
, struct mem_cgroup
*to
, bool uncharge
)
1852 VM_BUG_ON(from
== to
);
1853 VM_BUG_ON(PageLRU(pc
->page
));
1854 VM_BUG_ON(!PageCgroupLocked(pc
));
1855 VM_BUG_ON(!PageCgroupUsed(pc
));
1856 VM_BUG_ON(pc
->mem_cgroup
!= from
);
1858 if (PageCgroupFileMapped(pc
)) {
1859 /* Update mapped_file data for mem_cgroup */
1861 __this_cpu_dec(from
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
1862 __this_cpu_inc(to
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
1865 mem_cgroup_charge_statistics(from
, pc
, false);
1867 /* This is not "cancel", but cancel_charge does all we need. */
1868 mem_cgroup_cancel_charge(from
);
1870 /* caller should have done css_get */
1871 pc
->mem_cgroup
= to
;
1872 mem_cgroup_charge_statistics(to
, pc
, true);
1874 * We charges against "to" which may not have any tasks. Then, "to"
1875 * can be under rmdir(). But in current implementation, caller of
1876 * this function is just force_empty() and move charge, so it's
1877 * garanteed that "to" is never removed. So, we don't check rmdir
1883 * check whether the @pc is valid for moving account and call
1884 * __mem_cgroup_move_account()
1886 static int mem_cgroup_move_account(struct page_cgroup
*pc
,
1887 struct mem_cgroup
*from
, struct mem_cgroup
*to
, bool uncharge
)
1890 lock_page_cgroup(pc
);
1891 if (PageCgroupUsed(pc
) && pc
->mem_cgroup
== from
) {
1892 __mem_cgroup_move_account(pc
, from
, to
, uncharge
);
1895 unlock_page_cgroup(pc
);
1899 memcg_check_events(to
, pc
->page
);
1900 memcg_check_events(from
, pc
->page
);
1905 * move charges to its parent.
1908 static int mem_cgroup_move_parent(struct page_cgroup
*pc
,
1909 struct mem_cgroup
*child
,
1912 struct page
*page
= pc
->page
;
1913 struct cgroup
*cg
= child
->css
.cgroup
;
1914 struct cgroup
*pcg
= cg
->parent
;
1915 struct mem_cgroup
*parent
;
1923 if (!get_page_unless_zero(page
))
1925 if (isolate_lru_page(page
))
1928 parent
= mem_cgroup_from_cont(pcg
);
1929 ret
= __mem_cgroup_try_charge(NULL
, gfp_mask
, &parent
, false);
1933 ret
= mem_cgroup_move_account(pc
, child
, parent
, true);
1935 mem_cgroup_cancel_charge(parent
);
1937 putback_lru_page(page
);
1945 * Charge the memory controller for page usage.
1947 * 0 if the charge was successful
1948 * < 0 if the cgroup is over its limit
1950 static int mem_cgroup_charge_common(struct page
*page
, struct mm_struct
*mm
,
1951 gfp_t gfp_mask
, enum charge_type ctype
,
1952 struct mem_cgroup
*memcg
)
1954 struct mem_cgroup
*mem
;
1955 struct page_cgroup
*pc
;
1958 pc
= lookup_page_cgroup(page
);
1959 /* can happen at boot */
1965 ret
= __mem_cgroup_try_charge(mm
, gfp_mask
, &mem
, true);
1969 __mem_cgroup_commit_charge(mem
, pc
, ctype
);
1973 int mem_cgroup_newpage_charge(struct page
*page
,
1974 struct mm_struct
*mm
, gfp_t gfp_mask
)
1976 if (mem_cgroup_disabled())
1978 if (PageCompound(page
))
1981 * If already mapped, we don't have to account.
1982 * If page cache, page->mapping has address_space.
1983 * But page->mapping may have out-of-use anon_vma pointer,
1984 * detecit it by PageAnon() check. newly-mapped-anon's page->mapping
1987 if (page_mapped(page
) || (page
->mapping
&& !PageAnon(page
)))
1991 return mem_cgroup_charge_common(page
, mm
, gfp_mask
,
1992 MEM_CGROUP_CHARGE_TYPE_MAPPED
, NULL
);
1996 __mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*ptr
,
1997 enum charge_type ctype
);
1999 int mem_cgroup_cache_charge(struct page
*page
, struct mm_struct
*mm
,
2002 struct mem_cgroup
*mem
= NULL
;
2005 if (mem_cgroup_disabled())
2007 if (PageCompound(page
))
2010 * Corner case handling. This is called from add_to_page_cache()
2011 * in usual. But some FS (shmem) precharges this page before calling it
2012 * and call add_to_page_cache() with GFP_NOWAIT.
2014 * For GFP_NOWAIT case, the page may be pre-charged before calling
2015 * add_to_page_cache(). (See shmem.c) check it here and avoid to call
2016 * charge twice. (It works but has to pay a bit larger cost.)
2017 * And when the page is SwapCache, it should take swap information
2018 * into account. This is under lock_page() now.
2020 if (!(gfp_mask
& __GFP_WAIT
)) {
2021 struct page_cgroup
*pc
;
2024 pc
= lookup_page_cgroup(page
);
2027 lock_page_cgroup(pc
);
2028 if (PageCgroupUsed(pc
)) {
2029 unlock_page_cgroup(pc
);
2032 unlock_page_cgroup(pc
);
2035 if (unlikely(!mm
&& !mem
))
2038 if (page_is_file_cache(page
))
2039 return mem_cgroup_charge_common(page
, mm
, gfp_mask
,
2040 MEM_CGROUP_CHARGE_TYPE_CACHE
, NULL
);
2043 if (PageSwapCache(page
)) {
2044 ret
= mem_cgroup_try_charge_swapin(mm
, page
, gfp_mask
, &mem
);
2046 __mem_cgroup_commit_charge_swapin(page
, mem
,
2047 MEM_CGROUP_CHARGE_TYPE_SHMEM
);
2049 ret
= mem_cgroup_charge_common(page
, mm
, gfp_mask
,
2050 MEM_CGROUP_CHARGE_TYPE_SHMEM
, mem
);
2056 * While swap-in, try_charge -> commit or cancel, the page is locked.
2057 * And when try_charge() successfully returns, one refcnt to memcg without
2058 * struct page_cgroup is acquired. This refcnt will be consumed by
2059 * "commit()" or removed by "cancel()"
2061 int mem_cgroup_try_charge_swapin(struct mm_struct
*mm
,
2063 gfp_t mask
, struct mem_cgroup
**ptr
)
2065 struct mem_cgroup
*mem
;
2068 if (mem_cgroup_disabled())
2071 if (!do_swap_account
)
2074 * A racing thread's fault, or swapoff, may have already updated
2075 * the pte, and even removed page from swap cache: in those cases
2076 * do_swap_page()'s pte_same() test will fail; but there's also a
2077 * KSM case which does need to charge the page.
2079 if (!PageSwapCache(page
))
2081 mem
= try_get_mem_cgroup_from_page(page
);
2085 ret
= __mem_cgroup_try_charge(NULL
, mask
, ptr
, true);
2086 /* drop extra refcnt from tryget */
2092 return __mem_cgroup_try_charge(mm
, mask
, ptr
, true);
2096 __mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*ptr
,
2097 enum charge_type ctype
)
2099 struct page_cgroup
*pc
;
2101 if (mem_cgroup_disabled())
2105 cgroup_exclude_rmdir(&ptr
->css
);
2106 pc
= lookup_page_cgroup(page
);
2107 mem_cgroup_lru_del_before_commit_swapcache(page
);
2108 __mem_cgroup_commit_charge(ptr
, pc
, ctype
);
2109 mem_cgroup_lru_add_after_commit_swapcache(page
);
2111 * Now swap is on-memory. This means this page may be
2112 * counted both as mem and swap....double count.
2113 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
2114 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
2115 * may call delete_from_swap_cache() before reach here.
2117 if (do_swap_account
&& PageSwapCache(page
)) {
2118 swp_entry_t ent
= {.val
= page_private(page
)};
2120 struct mem_cgroup
*memcg
;
2122 id
= swap_cgroup_record(ent
, 0);
2124 memcg
= mem_cgroup_lookup(id
);
2127 * This recorded memcg can be obsolete one. So, avoid
2128 * calling css_tryget
2130 if (!mem_cgroup_is_root(memcg
))
2131 res_counter_uncharge(&memcg
->memsw
, PAGE_SIZE
);
2132 mem_cgroup_swap_statistics(memcg
, false);
2133 mem_cgroup_put(memcg
);
2138 * At swapin, we may charge account against cgroup which has no tasks.
2139 * So, rmdir()->pre_destroy() can be called while we do this charge.
2140 * In that case, we need to call pre_destroy() again. check it here.
2142 cgroup_release_and_wakeup_rmdir(&ptr
->css
);
2145 void mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*ptr
)
2147 __mem_cgroup_commit_charge_swapin(page
, ptr
,
2148 MEM_CGROUP_CHARGE_TYPE_MAPPED
);
2151 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup
*mem
)
2153 if (mem_cgroup_disabled())
2157 mem_cgroup_cancel_charge(mem
);
2161 __do_uncharge(struct mem_cgroup
*mem
, const enum charge_type ctype
)
2163 struct memcg_batch_info
*batch
= NULL
;
2164 bool uncharge_memsw
= true;
2165 /* If swapout, usage of swap doesn't decrease */
2166 if (!do_swap_account
|| ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
)
2167 uncharge_memsw
= false;
2169 batch
= ¤t
->memcg_batch
;
2171 * In usual, we do css_get() when we remember memcg pointer.
2172 * But in this case, we keep res->usage until end of a series of
2173 * uncharges. Then, it's ok to ignore memcg's refcnt.
2178 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
2179 * In those cases, all pages freed continously can be expected to be in
2180 * the same cgroup and we have chance to coalesce uncharges.
2181 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
2182 * because we want to do uncharge as soon as possible.
2185 if (!batch
->do_batch
|| test_thread_flag(TIF_MEMDIE
))
2186 goto direct_uncharge
;
2189 * In typical case, batch->memcg == mem. This means we can
2190 * merge a series of uncharges to an uncharge of res_counter.
2191 * If not, we uncharge res_counter ony by one.
2193 if (batch
->memcg
!= mem
)
2194 goto direct_uncharge
;
2195 /* remember freed charge and uncharge it later */
2196 batch
->bytes
+= PAGE_SIZE
;
2198 batch
->memsw_bytes
+= PAGE_SIZE
;
2201 res_counter_uncharge(&mem
->res
, PAGE_SIZE
);
2203 res_counter_uncharge(&mem
->memsw
, PAGE_SIZE
);
2204 if (unlikely(batch
->memcg
!= mem
))
2205 memcg_oom_recover(mem
);
2210 * uncharge if !page_mapped(page)
2212 static struct mem_cgroup
*
2213 __mem_cgroup_uncharge_common(struct page
*page
, enum charge_type ctype
)
2215 struct page_cgroup
*pc
;
2216 struct mem_cgroup
*mem
= NULL
;
2217 struct mem_cgroup_per_zone
*mz
;
2219 if (mem_cgroup_disabled())
2222 if (PageSwapCache(page
))
2226 * Check if our page_cgroup is valid
2228 pc
= lookup_page_cgroup(page
);
2229 if (unlikely(!pc
|| !PageCgroupUsed(pc
)))
2232 lock_page_cgroup(pc
);
2234 mem
= pc
->mem_cgroup
;
2236 if (!PageCgroupUsed(pc
))
2240 case MEM_CGROUP_CHARGE_TYPE_MAPPED
:
2241 case MEM_CGROUP_CHARGE_TYPE_DROP
:
2242 /* See mem_cgroup_prepare_migration() */
2243 if (page_mapped(page
) || PageCgroupMigration(pc
))
2246 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT
:
2247 if (!PageAnon(page
)) { /* Shared memory */
2248 if (page
->mapping
&& !page_is_file_cache(page
))
2250 } else if (page_mapped(page
)) /* Anon */
2257 if (!mem_cgroup_is_root(mem
))
2258 __do_uncharge(mem
, ctype
);
2259 if (ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
)
2260 mem_cgroup_swap_statistics(mem
, true);
2261 mem_cgroup_charge_statistics(mem
, pc
, false);
2263 ClearPageCgroupUsed(pc
);
2265 * pc->mem_cgroup is not cleared here. It will be accessed when it's
2266 * freed from LRU. This is safe because uncharged page is expected not
2267 * to be reused (freed soon). Exception is SwapCache, it's handled by
2268 * special functions.
2271 mz
= page_cgroup_zoneinfo(pc
);
2272 unlock_page_cgroup(pc
);
2274 memcg_check_events(mem
, page
);
2275 /* at swapout, this memcg will be accessed to record to swap */
2276 if (ctype
!= MEM_CGROUP_CHARGE_TYPE_SWAPOUT
)
2282 unlock_page_cgroup(pc
);
2286 void mem_cgroup_uncharge_page(struct page
*page
)
2289 if (page_mapped(page
))
2291 if (page
->mapping
&& !PageAnon(page
))
2293 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_MAPPED
);
2296 void mem_cgroup_uncharge_cache_page(struct page
*page
)
2298 VM_BUG_ON(page_mapped(page
));
2299 VM_BUG_ON(page
->mapping
);
2300 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_CACHE
);
2304 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
2305 * In that cases, pages are freed continuously and we can expect pages
2306 * are in the same memcg. All these calls itself limits the number of
2307 * pages freed at once, then uncharge_start/end() is called properly.
2308 * This may be called prural(2) times in a context,
2311 void mem_cgroup_uncharge_start(void)
2313 current
->memcg_batch
.do_batch
++;
2314 /* We can do nest. */
2315 if (current
->memcg_batch
.do_batch
== 1) {
2316 current
->memcg_batch
.memcg
= NULL
;
2317 current
->memcg_batch
.bytes
= 0;
2318 current
->memcg_batch
.memsw_bytes
= 0;
2322 void mem_cgroup_uncharge_end(void)
2324 struct memcg_batch_info
*batch
= ¤t
->memcg_batch
;
2326 if (!batch
->do_batch
)
2330 if (batch
->do_batch
) /* If stacked, do nothing. */
2336 * This "batch->memcg" is valid without any css_get/put etc...
2337 * bacause we hide charges behind us.
2340 res_counter_uncharge(&batch
->memcg
->res
, batch
->bytes
);
2341 if (batch
->memsw_bytes
)
2342 res_counter_uncharge(&batch
->memcg
->memsw
, batch
->memsw_bytes
);
2343 memcg_oom_recover(batch
->memcg
);
2344 /* forget this pointer (for sanity check) */
2345 batch
->memcg
= NULL
;
2350 * called after __delete_from_swap_cache() and drop "page" account.
2351 * memcg information is recorded to swap_cgroup of "ent"
2354 mem_cgroup_uncharge_swapcache(struct page
*page
, swp_entry_t ent
, bool swapout
)
2356 struct mem_cgroup
*memcg
;
2357 int ctype
= MEM_CGROUP_CHARGE_TYPE_SWAPOUT
;
2359 if (!swapout
) /* this was a swap cache but the swap is unused ! */
2360 ctype
= MEM_CGROUP_CHARGE_TYPE_DROP
;
2362 memcg
= __mem_cgroup_uncharge_common(page
, ctype
);
2364 /* record memcg information */
2365 if (do_swap_account
&& swapout
&& memcg
) {
2366 swap_cgroup_record(ent
, css_id(&memcg
->css
));
2367 mem_cgroup_get(memcg
);
2369 if (swapout
&& memcg
)
2370 css_put(&memcg
->css
);
2374 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2376 * called from swap_entry_free(). remove record in swap_cgroup and
2377 * uncharge "memsw" account.
2379 void mem_cgroup_uncharge_swap(swp_entry_t ent
)
2381 struct mem_cgroup
*memcg
;
2384 if (!do_swap_account
)
2387 id
= swap_cgroup_record(ent
, 0);
2389 memcg
= mem_cgroup_lookup(id
);
2392 * We uncharge this because swap is freed.
2393 * This memcg can be obsolete one. We avoid calling css_tryget
2395 if (!mem_cgroup_is_root(memcg
))
2396 res_counter_uncharge(&memcg
->memsw
, PAGE_SIZE
);
2397 mem_cgroup_swap_statistics(memcg
, false);
2398 mem_cgroup_put(memcg
);
2404 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
2405 * @entry: swap entry to be moved
2406 * @from: mem_cgroup which the entry is moved from
2407 * @to: mem_cgroup which the entry is moved to
2408 * @need_fixup: whether we should fixup res_counters and refcounts.
2410 * It succeeds only when the swap_cgroup's record for this entry is the same
2411 * as the mem_cgroup's id of @from.
2413 * Returns 0 on success, -EINVAL on failure.
2415 * The caller must have charged to @to, IOW, called res_counter_charge() about
2416 * both res and memsw, and called css_get().
2418 static int mem_cgroup_move_swap_account(swp_entry_t entry
,
2419 struct mem_cgroup
*from
, struct mem_cgroup
*to
, bool need_fixup
)
2421 unsigned short old_id
, new_id
;
2423 old_id
= css_id(&from
->css
);
2424 new_id
= css_id(&to
->css
);
2426 if (swap_cgroup_cmpxchg(entry
, old_id
, new_id
) == old_id
) {
2427 mem_cgroup_swap_statistics(from
, false);
2428 mem_cgroup_swap_statistics(to
, true);
2430 * This function is only called from task migration context now.
2431 * It postpones res_counter and refcount handling till the end
2432 * of task migration(mem_cgroup_clear_mc()) for performance
2433 * improvement. But we cannot postpone mem_cgroup_get(to)
2434 * because if the process that has been moved to @to does
2435 * swap-in, the refcount of @to might be decreased to 0.
2439 if (!mem_cgroup_is_root(from
))
2440 res_counter_uncharge(&from
->memsw
, PAGE_SIZE
);
2441 mem_cgroup_put(from
);
2443 * we charged both to->res and to->memsw, so we should
2446 if (!mem_cgroup_is_root(to
))
2447 res_counter_uncharge(&to
->res
, PAGE_SIZE
);
2455 static inline int mem_cgroup_move_swap_account(swp_entry_t entry
,
2456 struct mem_cgroup
*from
, struct mem_cgroup
*to
, bool need_fixup
)
2463 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
2466 int mem_cgroup_prepare_migration(struct page
*page
,
2467 struct page
*newpage
, struct mem_cgroup
**ptr
)
2469 struct page_cgroup
*pc
;
2470 struct mem_cgroup
*mem
= NULL
;
2471 enum charge_type ctype
;
2474 if (mem_cgroup_disabled())
2477 pc
= lookup_page_cgroup(page
);
2478 lock_page_cgroup(pc
);
2479 if (PageCgroupUsed(pc
)) {
2480 mem
= pc
->mem_cgroup
;
2483 * At migrating an anonymous page, its mapcount goes down
2484 * to 0 and uncharge() will be called. But, even if it's fully
2485 * unmapped, migration may fail and this page has to be
2486 * charged again. We set MIGRATION flag here and delay uncharge
2487 * until end_migration() is called
2489 * Corner Case Thinking
2491 * When the old page was mapped as Anon and it's unmap-and-freed
2492 * while migration was ongoing.
2493 * If unmap finds the old page, uncharge() of it will be delayed
2494 * until end_migration(). If unmap finds a new page, it's
2495 * uncharged when it make mapcount to be 1->0. If unmap code
2496 * finds swap_migration_entry, the new page will not be mapped
2497 * and end_migration() will find it(mapcount==0).
2500 * When the old page was mapped but migraion fails, the kernel
2501 * remaps it. A charge for it is kept by MIGRATION flag even
2502 * if mapcount goes down to 0. We can do remap successfully
2503 * without charging it again.
2506 * The "old" page is under lock_page() until the end of
2507 * migration, so, the old page itself will not be swapped-out.
2508 * If the new page is swapped out before end_migraton, our
2509 * hook to usual swap-out path will catch the event.
2512 SetPageCgroupMigration(pc
);
2514 unlock_page_cgroup(pc
);
2516 * If the page is not charged at this point,
2523 ret
= __mem_cgroup_try_charge(NULL
, GFP_KERNEL
, ptr
, false);
2524 css_put(&mem
->css
);/* drop extra refcnt */
2525 if (ret
|| *ptr
== NULL
) {
2526 if (PageAnon(page
)) {
2527 lock_page_cgroup(pc
);
2528 ClearPageCgroupMigration(pc
);
2529 unlock_page_cgroup(pc
);
2531 * The old page may be fully unmapped while we kept it.
2533 mem_cgroup_uncharge_page(page
);
2538 * We charge new page before it's used/mapped. So, even if unlock_page()
2539 * is called before end_migration, we can catch all events on this new
2540 * page. In the case new page is migrated but not remapped, new page's
2541 * mapcount will be finally 0 and we call uncharge in end_migration().
2543 pc
= lookup_page_cgroup(newpage
);
2545 ctype
= MEM_CGROUP_CHARGE_TYPE_MAPPED
;
2546 else if (page_is_file_cache(page
))
2547 ctype
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
2549 ctype
= MEM_CGROUP_CHARGE_TYPE_SHMEM
;
2550 __mem_cgroup_commit_charge(mem
, pc
, ctype
);
2554 /* remove redundant charge if migration failed*/
2555 void mem_cgroup_end_migration(struct mem_cgroup
*mem
,
2556 struct page
*oldpage
, struct page
*newpage
)
2558 struct page
*used
, *unused
;
2559 struct page_cgroup
*pc
;
2563 /* blocks rmdir() */
2564 cgroup_exclude_rmdir(&mem
->css
);
2565 /* at migration success, oldpage->mapping is NULL. */
2566 if (oldpage
->mapping
) {
2574 * We disallowed uncharge of pages under migration because mapcount
2575 * of the page goes down to zero, temporarly.
2576 * Clear the flag and check the page should be charged.
2578 pc
= lookup_page_cgroup(oldpage
);
2579 lock_page_cgroup(pc
);
2580 ClearPageCgroupMigration(pc
);
2581 unlock_page_cgroup(pc
);
2583 if (unused
!= oldpage
)
2584 pc
= lookup_page_cgroup(unused
);
2585 __mem_cgroup_uncharge_common(unused
, MEM_CGROUP_CHARGE_TYPE_FORCE
);
2587 pc
= lookup_page_cgroup(used
);
2589 * If a page is a file cache, radix-tree replacement is very atomic
2590 * and we can skip this check. When it was an Anon page, its mapcount
2591 * goes down to 0. But because we added MIGRATION flage, it's not
2592 * uncharged yet. There are several case but page->mapcount check
2593 * and USED bit check in mem_cgroup_uncharge_page() will do enough
2594 * check. (see prepare_charge() also)
2597 mem_cgroup_uncharge_page(used
);
2599 * At migration, we may charge account against cgroup which has no
2601 * So, rmdir()->pre_destroy() can be called while we do this charge.
2602 * In that case, we need to call pre_destroy() again. check it here.
2604 cgroup_release_and_wakeup_rmdir(&mem
->css
);
2608 * A call to try to shrink memory usage on charge failure at shmem's swapin.
2609 * Calling hierarchical_reclaim is not enough because we should update
2610 * last_oom_jiffies to prevent pagefault_out_of_memory from invoking global OOM.
2611 * Moreover considering hierarchy, we should reclaim from the mem_over_limit,
2612 * not from the memcg which this page would be charged to.
2613 * try_charge_swapin does all of these works properly.
2615 int mem_cgroup_shmem_charge_fallback(struct page
*page
,
2616 struct mm_struct
*mm
,
2619 struct mem_cgroup
*mem
= NULL
;
2622 if (mem_cgroup_disabled())
2625 ret
= mem_cgroup_try_charge_swapin(mm
, page
, gfp_mask
, &mem
);
2627 mem_cgroup_cancel_charge_swapin(mem
); /* it does !mem check */
2632 static DEFINE_MUTEX(set_limit_mutex
);
2634 static int mem_cgroup_resize_limit(struct mem_cgroup
*memcg
,
2635 unsigned long long val
)
2638 u64 memswlimit
, memlimit
;
2640 int children
= mem_cgroup_count_children(memcg
);
2641 u64 curusage
, oldusage
;
2645 * For keeping hierarchical_reclaim simple, how long we should retry
2646 * is depends on callers. We set our retry-count to be function
2647 * of # of children which we should visit in this loop.
2649 retry_count
= MEM_CGROUP_RECLAIM_RETRIES
* children
;
2651 oldusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
2654 while (retry_count
) {
2655 if (signal_pending(current
)) {
2660 * Rather than hide all in some function, I do this in
2661 * open coded manner. You see what this really does.
2662 * We have to guarantee mem->res.limit < mem->memsw.limit.
2664 mutex_lock(&set_limit_mutex
);
2665 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
2666 if (memswlimit
< val
) {
2668 mutex_unlock(&set_limit_mutex
);
2672 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
2676 ret
= res_counter_set_limit(&memcg
->res
, val
);
2678 if (memswlimit
== val
)
2679 memcg
->memsw_is_minimum
= true;
2681 memcg
->memsw_is_minimum
= false;
2683 mutex_unlock(&set_limit_mutex
);
2688 mem_cgroup_hierarchical_reclaim(memcg
, NULL
, GFP_KERNEL
,
2689 MEM_CGROUP_RECLAIM_SHRINK
);
2690 curusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
2691 /* Usage is reduced ? */
2692 if (curusage
>= oldusage
)
2695 oldusage
= curusage
;
2697 if (!ret
&& enlarge
)
2698 memcg_oom_recover(memcg
);
2703 static int mem_cgroup_resize_memsw_limit(struct mem_cgroup
*memcg
,
2704 unsigned long long val
)
2707 u64 memlimit
, memswlimit
, oldusage
, curusage
;
2708 int children
= mem_cgroup_count_children(memcg
);
2712 /* see mem_cgroup_resize_res_limit */
2713 retry_count
= children
* MEM_CGROUP_RECLAIM_RETRIES
;
2714 oldusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
2715 while (retry_count
) {
2716 if (signal_pending(current
)) {
2721 * Rather than hide all in some function, I do this in
2722 * open coded manner. You see what this really does.
2723 * We have to guarantee mem->res.limit < mem->memsw.limit.
2725 mutex_lock(&set_limit_mutex
);
2726 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
2727 if (memlimit
> val
) {
2729 mutex_unlock(&set_limit_mutex
);
2732 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
2733 if (memswlimit
< val
)
2735 ret
= res_counter_set_limit(&memcg
->memsw
, val
);
2737 if (memlimit
== val
)
2738 memcg
->memsw_is_minimum
= true;
2740 memcg
->memsw_is_minimum
= false;
2742 mutex_unlock(&set_limit_mutex
);
2747 mem_cgroup_hierarchical_reclaim(memcg
, NULL
, GFP_KERNEL
,
2748 MEM_CGROUP_RECLAIM_NOSWAP
|
2749 MEM_CGROUP_RECLAIM_SHRINK
);
2750 curusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
2751 /* Usage is reduced ? */
2752 if (curusage
>= oldusage
)
2755 oldusage
= curusage
;
2757 if (!ret
&& enlarge
)
2758 memcg_oom_recover(memcg
);
2762 unsigned long mem_cgroup_soft_limit_reclaim(struct zone
*zone
, int order
,
2763 gfp_t gfp_mask
, int nid
,
2766 unsigned long nr_reclaimed
= 0;
2767 struct mem_cgroup_per_zone
*mz
, *next_mz
= NULL
;
2768 unsigned long reclaimed
;
2770 struct mem_cgroup_tree_per_zone
*mctz
;
2771 unsigned long long excess
;
2776 mctz
= soft_limit_tree_node_zone(nid
, zid
);
2778 * This loop can run a while, specially if mem_cgroup's continuously
2779 * keep exceeding their soft limit and putting the system under
2786 mz
= mem_cgroup_largest_soft_limit_node(mctz
);
2790 reclaimed
= mem_cgroup_hierarchical_reclaim(mz
->mem
, zone
,
2792 MEM_CGROUP_RECLAIM_SOFT
);
2793 nr_reclaimed
+= reclaimed
;
2794 spin_lock(&mctz
->lock
);
2797 * If we failed to reclaim anything from this memory cgroup
2798 * it is time to move on to the next cgroup
2804 * Loop until we find yet another one.
2806 * By the time we get the soft_limit lock
2807 * again, someone might have aded the
2808 * group back on the RB tree. Iterate to
2809 * make sure we get a different mem.
2810 * mem_cgroup_largest_soft_limit_node returns
2811 * NULL if no other cgroup is present on
2815 __mem_cgroup_largest_soft_limit_node(mctz
);
2816 if (next_mz
== mz
) {
2817 css_put(&next_mz
->mem
->css
);
2819 } else /* next_mz == NULL or other memcg */
2823 __mem_cgroup_remove_exceeded(mz
->mem
, mz
, mctz
);
2824 excess
= res_counter_soft_limit_excess(&mz
->mem
->res
);
2826 * One school of thought says that we should not add
2827 * back the node to the tree if reclaim returns 0.
2828 * But our reclaim could return 0, simply because due
2829 * to priority we are exposing a smaller subset of
2830 * memory to reclaim from. Consider this as a longer
2833 /* If excess == 0, no tree ops */
2834 __mem_cgroup_insert_exceeded(mz
->mem
, mz
, mctz
, excess
);
2835 spin_unlock(&mctz
->lock
);
2836 css_put(&mz
->mem
->css
);
2839 * Could not reclaim anything and there are no more
2840 * mem cgroups to try or we seem to be looping without
2841 * reclaiming anything.
2843 if (!nr_reclaimed
&&
2845 loop
> MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS
))
2847 } while (!nr_reclaimed
);
2849 css_put(&next_mz
->mem
->css
);
2850 return nr_reclaimed
;
2854 * This routine traverse page_cgroup in given list and drop them all.
2855 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
2857 static int mem_cgroup_force_empty_list(struct mem_cgroup
*mem
,
2858 int node
, int zid
, enum lru_list lru
)
2861 struct mem_cgroup_per_zone
*mz
;
2862 struct page_cgroup
*pc
, *busy
;
2863 unsigned long flags
, loop
;
2864 struct list_head
*list
;
2867 zone
= &NODE_DATA(node
)->node_zones
[zid
];
2868 mz
= mem_cgroup_zoneinfo(mem
, node
, zid
);
2869 list
= &mz
->lists
[lru
];
2871 loop
= MEM_CGROUP_ZSTAT(mz
, lru
);
2872 /* give some margin against EBUSY etc...*/
2877 spin_lock_irqsave(&zone
->lru_lock
, flags
);
2878 if (list_empty(list
)) {
2879 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
2882 pc
= list_entry(list
->prev
, struct page_cgroup
, lru
);
2884 list_move(&pc
->lru
, list
);
2886 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
2889 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
2891 ret
= mem_cgroup_move_parent(pc
, mem
, GFP_KERNEL
);
2895 if (ret
== -EBUSY
|| ret
== -EINVAL
) {
2896 /* found lock contention or "pc" is obsolete. */
2903 if (!ret
&& !list_empty(list
))
2909 * make mem_cgroup's charge to be 0 if there is no task.
2910 * This enables deleting this mem_cgroup.
2912 static int mem_cgroup_force_empty(struct mem_cgroup
*mem
, bool free_all
)
2915 int node
, zid
, shrink
;
2916 int nr_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
2917 struct cgroup
*cgrp
= mem
->css
.cgroup
;
2922 /* should free all ? */
2928 if (cgroup_task_count(cgrp
) || !list_empty(&cgrp
->children
))
2931 if (signal_pending(current
))
2933 /* This is for making all *used* pages to be on LRU. */
2934 lru_add_drain_all();
2935 drain_all_stock_sync();
2937 for_each_node_state(node
, N_HIGH_MEMORY
) {
2938 for (zid
= 0; !ret
&& zid
< MAX_NR_ZONES
; zid
++) {
2941 ret
= mem_cgroup_force_empty_list(mem
,
2950 memcg_oom_recover(mem
);
2951 /* it seems parent cgroup doesn't have enough mem */
2955 /* "ret" should also be checked to ensure all lists are empty. */
2956 } while (mem
->res
.usage
> 0 || ret
);
2962 /* returns EBUSY if there is a task or if we come here twice. */
2963 if (cgroup_task_count(cgrp
) || !list_empty(&cgrp
->children
) || shrink
) {
2967 /* we call try-to-free pages for make this cgroup empty */
2968 lru_add_drain_all();
2969 /* try to free all pages in this cgroup */
2971 while (nr_retries
&& mem
->res
.usage
> 0) {
2974 if (signal_pending(current
)) {
2978 progress
= try_to_free_mem_cgroup_pages(mem
, GFP_KERNEL
,
2979 false, get_swappiness(mem
));
2982 /* maybe some writeback is necessary */
2983 congestion_wait(BLK_RW_ASYNC
, HZ
/10);
2988 /* try move_account...there may be some *locked* pages. */
2992 int mem_cgroup_force_empty_write(struct cgroup
*cont
, unsigned int event
)
2994 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont
), true);
2998 static u64
mem_cgroup_hierarchy_read(struct cgroup
*cont
, struct cftype
*cft
)
3000 return mem_cgroup_from_cont(cont
)->use_hierarchy
;
3003 static int mem_cgroup_hierarchy_write(struct cgroup
*cont
, struct cftype
*cft
,
3007 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
3008 struct cgroup
*parent
= cont
->parent
;
3009 struct mem_cgroup
*parent_mem
= NULL
;
3012 parent_mem
= mem_cgroup_from_cont(parent
);
3016 * If parent's use_hierarchy is set, we can't make any modifications
3017 * in the child subtrees. If it is unset, then the change can
3018 * occur, provided the current cgroup has no children.
3020 * For the root cgroup, parent_mem is NULL, we allow value to be
3021 * set if there are no children.
3023 if ((!parent_mem
|| !parent_mem
->use_hierarchy
) &&
3024 (val
== 1 || val
== 0)) {
3025 if (list_empty(&cont
->children
))
3026 mem
->use_hierarchy
= val
;
3036 struct mem_cgroup_idx_data
{
3038 enum mem_cgroup_stat_index idx
;
3042 mem_cgroup_get_idx_stat(struct mem_cgroup
*mem
, void *data
)
3044 struct mem_cgroup_idx_data
*d
= data
;
3045 d
->val
+= mem_cgroup_read_stat(mem
, d
->idx
);
3050 mem_cgroup_get_recursive_idx_stat(struct mem_cgroup
*mem
,
3051 enum mem_cgroup_stat_index idx
, s64
*val
)
3053 struct mem_cgroup_idx_data d
;
3056 mem_cgroup_walk_tree(mem
, &d
, mem_cgroup_get_idx_stat
);
3060 static inline u64
mem_cgroup_usage(struct mem_cgroup
*mem
, bool swap
)
3064 if (!mem_cgroup_is_root(mem
)) {
3066 return res_counter_read_u64(&mem
->res
, RES_USAGE
);
3068 return res_counter_read_u64(&mem
->memsw
, RES_USAGE
);
3071 mem_cgroup_get_recursive_idx_stat(mem
, MEM_CGROUP_STAT_CACHE
, &idx_val
);
3073 mem_cgroup_get_recursive_idx_stat(mem
, MEM_CGROUP_STAT_RSS
, &idx_val
);
3077 mem_cgroup_get_recursive_idx_stat(mem
,
3078 MEM_CGROUP_STAT_SWAPOUT
, &idx_val
);
3082 return val
<< PAGE_SHIFT
;
3085 static u64
mem_cgroup_read(struct cgroup
*cont
, struct cftype
*cft
)
3087 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
3091 type
= MEMFILE_TYPE(cft
->private);
3092 name
= MEMFILE_ATTR(cft
->private);
3095 if (name
== RES_USAGE
)
3096 val
= mem_cgroup_usage(mem
, false);
3098 val
= res_counter_read_u64(&mem
->res
, name
);
3101 if (name
== RES_USAGE
)
3102 val
= mem_cgroup_usage(mem
, true);
3104 val
= res_counter_read_u64(&mem
->memsw
, name
);
3113 * The user of this function is...
3116 static int mem_cgroup_write(struct cgroup
*cont
, struct cftype
*cft
,
3119 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3121 unsigned long long val
;
3124 type
= MEMFILE_TYPE(cft
->private);
3125 name
= MEMFILE_ATTR(cft
->private);
3128 if (mem_cgroup_is_root(memcg
)) { /* Can't set limit on root */
3132 /* This function does all necessary parse...reuse it */
3133 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
3137 ret
= mem_cgroup_resize_limit(memcg
, val
);
3139 ret
= mem_cgroup_resize_memsw_limit(memcg
, val
);
3141 case RES_SOFT_LIMIT
:
3142 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
3146 * For memsw, soft limits are hard to implement in terms
3147 * of semantics, for now, we support soft limits for
3148 * control without swap
3151 ret
= res_counter_set_soft_limit(&memcg
->res
, val
);
3156 ret
= -EINVAL
; /* should be BUG() ? */
3162 static void memcg_get_hierarchical_limit(struct mem_cgroup
*memcg
,
3163 unsigned long long *mem_limit
, unsigned long long *memsw_limit
)
3165 struct cgroup
*cgroup
;
3166 unsigned long long min_limit
, min_memsw_limit
, tmp
;
3168 min_limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3169 min_memsw_limit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3170 cgroup
= memcg
->css
.cgroup
;
3171 if (!memcg
->use_hierarchy
)
3174 while (cgroup
->parent
) {
3175 cgroup
= cgroup
->parent
;
3176 memcg
= mem_cgroup_from_cont(cgroup
);
3177 if (!memcg
->use_hierarchy
)
3179 tmp
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3180 min_limit
= min(min_limit
, tmp
);
3181 tmp
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3182 min_memsw_limit
= min(min_memsw_limit
, tmp
);
3185 *mem_limit
= min_limit
;
3186 *memsw_limit
= min_memsw_limit
;
3190 static int mem_cgroup_reset(struct cgroup
*cont
, unsigned int event
)
3192 struct mem_cgroup
*mem
;
3195 mem
= mem_cgroup_from_cont(cont
);
3196 type
= MEMFILE_TYPE(event
);
3197 name
= MEMFILE_ATTR(event
);
3201 res_counter_reset_max(&mem
->res
);
3203 res_counter_reset_max(&mem
->memsw
);
3207 res_counter_reset_failcnt(&mem
->res
);
3209 res_counter_reset_failcnt(&mem
->memsw
);
3216 static u64
mem_cgroup_move_charge_read(struct cgroup
*cgrp
,
3219 return mem_cgroup_from_cont(cgrp
)->move_charge_at_immigrate
;
3223 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
3224 struct cftype
*cft
, u64 val
)
3226 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgrp
);
3228 if (val
>= (1 << NR_MOVE_TYPE
))
3231 * We check this value several times in both in can_attach() and
3232 * attach(), so we need cgroup lock to prevent this value from being
3236 mem
->move_charge_at_immigrate
= val
;
3242 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
3243 struct cftype
*cft
, u64 val
)
3250 /* For read statistics */
3266 struct mcs_total_stat
{
3267 s64 stat
[NR_MCS_STAT
];
3273 } memcg_stat_strings
[NR_MCS_STAT
] = {
3274 {"cache", "total_cache"},
3275 {"rss", "total_rss"},
3276 {"mapped_file", "total_mapped_file"},
3277 {"pgpgin", "total_pgpgin"},
3278 {"pgpgout", "total_pgpgout"},
3279 {"swap", "total_swap"},
3280 {"inactive_anon", "total_inactive_anon"},
3281 {"active_anon", "total_active_anon"},
3282 {"inactive_file", "total_inactive_file"},
3283 {"active_file", "total_active_file"},
3284 {"unevictable", "total_unevictable"}
3288 static int mem_cgroup_get_local_stat(struct mem_cgroup
*mem
, void *data
)
3290 struct mcs_total_stat
*s
= data
;
3294 val
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_CACHE
);
3295 s
->stat
[MCS_CACHE
] += val
* PAGE_SIZE
;
3296 val
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_RSS
);
3297 s
->stat
[MCS_RSS
] += val
* PAGE_SIZE
;
3298 val
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_FILE_MAPPED
);
3299 s
->stat
[MCS_FILE_MAPPED
] += val
* PAGE_SIZE
;
3300 val
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_PGPGIN_COUNT
);
3301 s
->stat
[MCS_PGPGIN
] += val
;
3302 val
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_PGPGOUT_COUNT
);
3303 s
->stat
[MCS_PGPGOUT
] += val
;
3304 if (do_swap_account
) {
3305 val
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_SWAPOUT
);
3306 s
->stat
[MCS_SWAP
] += val
* PAGE_SIZE
;
3310 val
= mem_cgroup_get_local_zonestat(mem
, LRU_INACTIVE_ANON
);
3311 s
->stat
[MCS_INACTIVE_ANON
] += val
* PAGE_SIZE
;
3312 val
= mem_cgroup_get_local_zonestat(mem
, LRU_ACTIVE_ANON
);
3313 s
->stat
[MCS_ACTIVE_ANON
] += val
* PAGE_SIZE
;
3314 val
= mem_cgroup_get_local_zonestat(mem
, LRU_INACTIVE_FILE
);
3315 s
->stat
[MCS_INACTIVE_FILE
] += val
* PAGE_SIZE
;
3316 val
= mem_cgroup_get_local_zonestat(mem
, LRU_ACTIVE_FILE
);
3317 s
->stat
[MCS_ACTIVE_FILE
] += val
* PAGE_SIZE
;
3318 val
= mem_cgroup_get_local_zonestat(mem
, LRU_UNEVICTABLE
);
3319 s
->stat
[MCS_UNEVICTABLE
] += val
* PAGE_SIZE
;
3324 mem_cgroup_get_total_stat(struct mem_cgroup
*mem
, struct mcs_total_stat
*s
)
3326 mem_cgroup_walk_tree(mem
, s
, mem_cgroup_get_local_stat
);
3329 static int mem_control_stat_show(struct cgroup
*cont
, struct cftype
*cft
,
3330 struct cgroup_map_cb
*cb
)
3332 struct mem_cgroup
*mem_cont
= mem_cgroup_from_cont(cont
);
3333 struct mcs_total_stat mystat
;
3336 memset(&mystat
, 0, sizeof(mystat
));
3337 mem_cgroup_get_local_stat(mem_cont
, &mystat
);
3339 for (i
= 0; i
< NR_MCS_STAT
; i
++) {
3340 if (i
== MCS_SWAP
&& !do_swap_account
)
3342 cb
->fill(cb
, memcg_stat_strings
[i
].local_name
, mystat
.stat
[i
]);
3345 /* Hierarchical information */
3347 unsigned long long limit
, memsw_limit
;
3348 memcg_get_hierarchical_limit(mem_cont
, &limit
, &memsw_limit
);
3349 cb
->fill(cb
, "hierarchical_memory_limit", limit
);
3350 if (do_swap_account
)
3351 cb
->fill(cb
, "hierarchical_memsw_limit", memsw_limit
);
3354 memset(&mystat
, 0, sizeof(mystat
));
3355 mem_cgroup_get_total_stat(mem_cont
, &mystat
);
3356 for (i
= 0; i
< NR_MCS_STAT
; i
++) {
3357 if (i
== MCS_SWAP
&& !do_swap_account
)
3359 cb
->fill(cb
, memcg_stat_strings
[i
].total_name
, mystat
.stat
[i
]);
3362 #ifdef CONFIG_DEBUG_VM
3363 cb
->fill(cb
, "inactive_ratio", calc_inactive_ratio(mem_cont
, NULL
));
3367 struct mem_cgroup_per_zone
*mz
;
3368 unsigned long recent_rotated
[2] = {0, 0};
3369 unsigned long recent_scanned
[2] = {0, 0};
3371 for_each_online_node(nid
)
3372 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
3373 mz
= mem_cgroup_zoneinfo(mem_cont
, nid
, zid
);
3375 recent_rotated
[0] +=
3376 mz
->reclaim_stat
.recent_rotated
[0];
3377 recent_rotated
[1] +=
3378 mz
->reclaim_stat
.recent_rotated
[1];
3379 recent_scanned
[0] +=
3380 mz
->reclaim_stat
.recent_scanned
[0];
3381 recent_scanned
[1] +=
3382 mz
->reclaim_stat
.recent_scanned
[1];
3384 cb
->fill(cb
, "recent_rotated_anon", recent_rotated
[0]);
3385 cb
->fill(cb
, "recent_rotated_file", recent_rotated
[1]);
3386 cb
->fill(cb
, "recent_scanned_anon", recent_scanned
[0]);
3387 cb
->fill(cb
, "recent_scanned_file", recent_scanned
[1]);
3394 static u64
mem_cgroup_swappiness_read(struct cgroup
*cgrp
, struct cftype
*cft
)
3396 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
3398 return get_swappiness(memcg
);
3401 static int mem_cgroup_swappiness_write(struct cgroup
*cgrp
, struct cftype
*cft
,
3404 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
3405 struct mem_cgroup
*parent
;
3410 if (cgrp
->parent
== NULL
)
3413 parent
= mem_cgroup_from_cont(cgrp
->parent
);
3417 /* If under hierarchy, only empty-root can set this value */
3418 if ((parent
->use_hierarchy
) ||
3419 (memcg
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
3424 spin_lock(&memcg
->reclaim_param_lock
);
3425 memcg
->swappiness
= val
;
3426 spin_unlock(&memcg
->reclaim_param_lock
);
3433 static void __mem_cgroup_threshold(struct mem_cgroup
*memcg
, bool swap
)
3435 struct mem_cgroup_threshold_ary
*t
;
3441 t
= rcu_dereference(memcg
->thresholds
.primary
);
3443 t
= rcu_dereference(memcg
->memsw_thresholds
.primary
);
3448 usage
= mem_cgroup_usage(memcg
, swap
);
3451 * current_threshold points to threshold just below usage.
3452 * If it's not true, a threshold was crossed after last
3453 * call of __mem_cgroup_threshold().
3455 i
= t
->current_threshold
;
3458 * Iterate backward over array of thresholds starting from
3459 * current_threshold and check if a threshold is crossed.
3460 * If none of thresholds below usage is crossed, we read
3461 * only one element of the array here.
3463 for (; i
>= 0 && unlikely(t
->entries
[i
].threshold
> usage
); i
--)
3464 eventfd_signal(t
->entries
[i
].eventfd
, 1);
3466 /* i = current_threshold + 1 */
3470 * Iterate forward over array of thresholds starting from
3471 * current_threshold+1 and check if a threshold is crossed.
3472 * If none of thresholds above usage is crossed, we read
3473 * only one element of the array here.
3475 for (; i
< t
->size
&& unlikely(t
->entries
[i
].threshold
<= usage
); i
++)
3476 eventfd_signal(t
->entries
[i
].eventfd
, 1);
3478 /* Update current_threshold */
3479 t
->current_threshold
= i
- 1;
3484 static void mem_cgroup_threshold(struct mem_cgroup
*memcg
)
3486 __mem_cgroup_threshold(memcg
, false);
3487 if (do_swap_account
)
3488 __mem_cgroup_threshold(memcg
, true);
3491 static int compare_thresholds(const void *a
, const void *b
)
3493 const struct mem_cgroup_threshold
*_a
= a
;
3494 const struct mem_cgroup_threshold
*_b
= b
;
3496 return _a
->threshold
- _b
->threshold
;
3499 static int mem_cgroup_oom_notify_cb(struct mem_cgroup
*mem
, void *data
)
3501 struct mem_cgroup_eventfd_list
*ev
;
3503 list_for_each_entry(ev
, &mem
->oom_notify
, list
)
3504 eventfd_signal(ev
->eventfd
, 1);
3508 static void mem_cgroup_oom_notify(struct mem_cgroup
*mem
)
3510 mem_cgroup_walk_tree(mem
, NULL
, mem_cgroup_oom_notify_cb
);
3513 static int mem_cgroup_usage_register_event(struct cgroup
*cgrp
,
3514 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
3516 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
3517 struct mem_cgroup_thresholds
*thresholds
;
3518 struct mem_cgroup_threshold_ary
*new;
3519 int type
= MEMFILE_TYPE(cft
->private);
3520 u64 threshold
, usage
;
3523 ret
= res_counter_memparse_write_strategy(args
, &threshold
);
3527 mutex_lock(&memcg
->thresholds_lock
);
3530 thresholds
= &memcg
->thresholds
;
3531 else if (type
== _MEMSWAP
)
3532 thresholds
= &memcg
->memsw_thresholds
;
3536 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
3538 /* Check if a threshold crossed before adding a new one */
3539 if (thresholds
->primary
)
3540 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
3542 size
= thresholds
->primary
? thresholds
->primary
->size
+ 1 : 1;
3544 /* Allocate memory for new array of thresholds */
3545 new = kmalloc(sizeof(*new) + size
* sizeof(struct mem_cgroup_threshold
),
3553 /* Copy thresholds (if any) to new array */
3554 if (thresholds
->primary
) {
3555 memcpy(new->entries
, thresholds
->primary
->entries
, (size
- 1) *
3556 sizeof(struct mem_cgroup_threshold
));
3559 /* Add new threshold */
3560 new->entries
[size
- 1].eventfd
= eventfd
;
3561 new->entries
[size
- 1].threshold
= threshold
;
3563 /* Sort thresholds. Registering of new threshold isn't time-critical */
3564 sort(new->entries
, size
, sizeof(struct mem_cgroup_threshold
),
3565 compare_thresholds
, NULL
);
3567 /* Find current threshold */
3568 new->current_threshold
= -1;
3569 for (i
= 0; i
< size
; i
++) {
3570 if (new->entries
[i
].threshold
< usage
) {
3572 * new->current_threshold will not be used until
3573 * rcu_assign_pointer(), so it's safe to increment
3576 ++new->current_threshold
;
3580 /* Free old spare buffer and save old primary buffer as spare */
3581 kfree(thresholds
->spare
);
3582 thresholds
->spare
= thresholds
->primary
;
3584 rcu_assign_pointer(thresholds
->primary
, new);
3586 /* To be sure that nobody uses thresholds */
3590 mutex_unlock(&memcg
->thresholds_lock
);
3595 static void mem_cgroup_usage_unregister_event(struct cgroup
*cgrp
,
3596 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
3598 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
3599 struct mem_cgroup_thresholds
*thresholds
;
3600 struct mem_cgroup_threshold_ary
*new;
3601 int type
= MEMFILE_TYPE(cft
->private);
3605 mutex_lock(&memcg
->thresholds_lock
);
3607 thresholds
= &memcg
->thresholds
;
3608 else if (type
== _MEMSWAP
)
3609 thresholds
= &memcg
->memsw_thresholds
;
3614 * Something went wrong if we trying to unregister a threshold
3615 * if we don't have thresholds
3617 BUG_ON(!thresholds
);
3619 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
3621 /* Check if a threshold crossed before removing */
3622 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
3624 /* Calculate new number of threshold */
3626 for (i
= 0; i
< thresholds
->primary
->size
; i
++) {
3627 if (thresholds
->primary
->entries
[i
].eventfd
!= eventfd
)
3631 new = thresholds
->spare
;
3633 /* Set thresholds array to NULL if we don't have thresholds */
3642 /* Copy thresholds and find current threshold */
3643 new->current_threshold
= -1;
3644 for (i
= 0, j
= 0; i
< thresholds
->primary
->size
; i
++) {
3645 if (thresholds
->primary
->entries
[i
].eventfd
== eventfd
)
3648 new->entries
[j
] = thresholds
->primary
->entries
[i
];
3649 if (new->entries
[j
].threshold
< usage
) {
3651 * new->current_threshold will not be used
3652 * until rcu_assign_pointer(), so it's safe to increment
3655 ++new->current_threshold
;
3661 /* Swap primary and spare array */
3662 thresholds
->spare
= thresholds
->primary
;
3663 rcu_assign_pointer(thresholds
->primary
, new);
3665 /* To be sure that nobody uses thresholds */
3668 mutex_unlock(&memcg
->thresholds_lock
);
3671 static int mem_cgroup_oom_register_event(struct cgroup
*cgrp
,
3672 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
3674 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
3675 struct mem_cgroup_eventfd_list
*event
;
3676 int type
= MEMFILE_TYPE(cft
->private);
3678 BUG_ON(type
!= _OOM_TYPE
);
3679 event
= kmalloc(sizeof(*event
), GFP_KERNEL
);
3683 mutex_lock(&memcg_oom_mutex
);
3685 event
->eventfd
= eventfd
;
3686 list_add(&event
->list
, &memcg
->oom_notify
);
3688 /* already in OOM ? */
3689 if (atomic_read(&memcg
->oom_lock
))
3690 eventfd_signal(eventfd
, 1);
3691 mutex_unlock(&memcg_oom_mutex
);
3696 static void mem_cgroup_oom_unregister_event(struct cgroup
*cgrp
,
3697 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
3699 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgrp
);
3700 struct mem_cgroup_eventfd_list
*ev
, *tmp
;
3701 int type
= MEMFILE_TYPE(cft
->private);
3703 BUG_ON(type
!= _OOM_TYPE
);
3705 mutex_lock(&memcg_oom_mutex
);
3707 list_for_each_entry_safe(ev
, tmp
, &mem
->oom_notify
, list
) {
3708 if (ev
->eventfd
== eventfd
) {
3709 list_del(&ev
->list
);
3714 mutex_unlock(&memcg_oom_mutex
);
3717 static int mem_cgroup_oom_control_read(struct cgroup
*cgrp
,
3718 struct cftype
*cft
, struct cgroup_map_cb
*cb
)
3720 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgrp
);
3722 cb
->fill(cb
, "oom_kill_disable", mem
->oom_kill_disable
);
3724 if (atomic_read(&mem
->oom_lock
))
3725 cb
->fill(cb
, "under_oom", 1);
3727 cb
->fill(cb
, "under_oom", 0);
3733 static int mem_cgroup_oom_control_write(struct cgroup
*cgrp
,
3734 struct cftype
*cft
, u64 val
)
3736 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgrp
);
3737 struct mem_cgroup
*parent
;
3739 /* cannot set to root cgroup and only 0 and 1 are allowed */
3740 if (!cgrp
->parent
|| !((val
== 0) || (val
== 1)))
3743 parent
= mem_cgroup_from_cont(cgrp
->parent
);
3746 /* oom-kill-disable is a flag for subhierarchy. */
3747 if ((parent
->use_hierarchy
) ||
3748 (mem
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
3752 mem
->oom_kill_disable
= val
;
3754 memcg_oom_recover(mem
);
3759 static struct cftype mem_cgroup_files
[] = {
3761 .name
= "usage_in_bytes",
3762 .private = MEMFILE_PRIVATE(_MEM
, RES_USAGE
),
3763 .read_u64
= mem_cgroup_read
,
3764 .register_event
= mem_cgroup_usage_register_event
,
3765 .unregister_event
= mem_cgroup_usage_unregister_event
,
3768 .name
= "max_usage_in_bytes",
3769 .private = MEMFILE_PRIVATE(_MEM
, RES_MAX_USAGE
),
3770 .trigger
= mem_cgroup_reset
,
3771 .read_u64
= mem_cgroup_read
,
3774 .name
= "limit_in_bytes",
3775 .private = MEMFILE_PRIVATE(_MEM
, RES_LIMIT
),
3776 .write_string
= mem_cgroup_write
,
3777 .read_u64
= mem_cgroup_read
,
3780 .name
= "soft_limit_in_bytes",
3781 .private = MEMFILE_PRIVATE(_MEM
, RES_SOFT_LIMIT
),
3782 .write_string
= mem_cgroup_write
,
3783 .read_u64
= mem_cgroup_read
,
3787 .private = MEMFILE_PRIVATE(_MEM
, RES_FAILCNT
),
3788 .trigger
= mem_cgroup_reset
,
3789 .read_u64
= mem_cgroup_read
,
3793 .read_map
= mem_control_stat_show
,
3796 .name
= "force_empty",
3797 .trigger
= mem_cgroup_force_empty_write
,
3800 .name
= "use_hierarchy",
3801 .write_u64
= mem_cgroup_hierarchy_write
,
3802 .read_u64
= mem_cgroup_hierarchy_read
,
3805 .name
= "swappiness",
3806 .read_u64
= mem_cgroup_swappiness_read
,
3807 .write_u64
= mem_cgroup_swappiness_write
,
3810 .name
= "move_charge_at_immigrate",
3811 .read_u64
= mem_cgroup_move_charge_read
,
3812 .write_u64
= mem_cgroup_move_charge_write
,
3815 .name
= "oom_control",
3816 .read_map
= mem_cgroup_oom_control_read
,
3817 .write_u64
= mem_cgroup_oom_control_write
,
3818 .register_event
= mem_cgroup_oom_register_event
,
3819 .unregister_event
= mem_cgroup_oom_unregister_event
,
3820 .private = MEMFILE_PRIVATE(_OOM_TYPE
, OOM_CONTROL
),
3824 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
3825 static struct cftype memsw_cgroup_files
[] = {
3827 .name
= "memsw.usage_in_bytes",
3828 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_USAGE
),
3829 .read_u64
= mem_cgroup_read
,
3830 .register_event
= mem_cgroup_usage_register_event
,
3831 .unregister_event
= mem_cgroup_usage_unregister_event
,
3834 .name
= "memsw.max_usage_in_bytes",
3835 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_MAX_USAGE
),
3836 .trigger
= mem_cgroup_reset
,
3837 .read_u64
= mem_cgroup_read
,
3840 .name
= "memsw.limit_in_bytes",
3841 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_LIMIT
),
3842 .write_string
= mem_cgroup_write
,
3843 .read_u64
= mem_cgroup_read
,
3846 .name
= "memsw.failcnt",
3847 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_FAILCNT
),
3848 .trigger
= mem_cgroup_reset
,
3849 .read_u64
= mem_cgroup_read
,
3853 static int register_memsw_files(struct cgroup
*cont
, struct cgroup_subsys
*ss
)
3855 if (!do_swap_account
)
3857 return cgroup_add_files(cont
, ss
, memsw_cgroup_files
,
3858 ARRAY_SIZE(memsw_cgroup_files
));
3861 static int register_memsw_files(struct cgroup
*cont
, struct cgroup_subsys
*ss
)
3867 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup
*mem
, int node
)
3869 struct mem_cgroup_per_node
*pn
;
3870 struct mem_cgroup_per_zone
*mz
;
3872 int zone
, tmp
= node
;
3874 * This routine is called against possible nodes.
3875 * But it's BUG to call kmalloc() against offline node.
3877 * TODO: this routine can waste much memory for nodes which will
3878 * never be onlined. It's better to use memory hotplug callback
3881 if (!node_state(node
, N_NORMAL_MEMORY
))
3883 pn
= kmalloc_node(sizeof(*pn
), GFP_KERNEL
, tmp
);
3887 mem
->info
.nodeinfo
[node
] = pn
;
3888 memset(pn
, 0, sizeof(*pn
));
3890 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
3891 mz
= &pn
->zoneinfo
[zone
];
3893 INIT_LIST_HEAD(&mz
->lists
[l
]);
3894 mz
->usage_in_excess
= 0;
3895 mz
->on_tree
= false;
3901 static void free_mem_cgroup_per_zone_info(struct mem_cgroup
*mem
, int node
)
3903 kfree(mem
->info
.nodeinfo
[node
]);
3906 static struct mem_cgroup
*mem_cgroup_alloc(void)
3908 struct mem_cgroup
*mem
;
3909 int size
= sizeof(struct mem_cgroup
);
3911 /* Can be very big if MAX_NUMNODES is very big */
3912 if (size
< PAGE_SIZE
)
3913 mem
= kmalloc(size
, GFP_KERNEL
);
3915 mem
= vmalloc(size
);
3920 memset(mem
, 0, size
);
3921 mem
->stat
= alloc_percpu(struct mem_cgroup_stat_cpu
);
3923 if (size
< PAGE_SIZE
)
3933 * At destroying mem_cgroup, references from swap_cgroup can remain.
3934 * (scanning all at force_empty is too costly...)
3936 * Instead of clearing all references at force_empty, we remember
3937 * the number of reference from swap_cgroup and free mem_cgroup when
3938 * it goes down to 0.
3940 * Removal of cgroup itself succeeds regardless of refs from swap.
3943 static void __mem_cgroup_free(struct mem_cgroup
*mem
)
3947 mem_cgroup_remove_from_trees(mem
);
3948 free_css_id(&mem_cgroup_subsys
, &mem
->css
);
3950 for_each_node_state(node
, N_POSSIBLE
)
3951 free_mem_cgroup_per_zone_info(mem
, node
);
3953 free_percpu(mem
->stat
);
3954 if (sizeof(struct mem_cgroup
) < PAGE_SIZE
)
3960 static void mem_cgroup_get(struct mem_cgroup
*mem
)
3962 atomic_inc(&mem
->refcnt
);
3965 static void __mem_cgroup_put(struct mem_cgroup
*mem
, int count
)
3967 if (atomic_sub_and_test(count
, &mem
->refcnt
)) {
3968 struct mem_cgroup
*parent
= parent_mem_cgroup(mem
);
3969 __mem_cgroup_free(mem
);
3971 mem_cgroup_put(parent
);
3975 static void mem_cgroup_put(struct mem_cgroup
*mem
)
3977 __mem_cgroup_put(mem
, 1);
3981 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
3983 static struct mem_cgroup
*parent_mem_cgroup(struct mem_cgroup
*mem
)
3985 if (!mem
->res
.parent
)
3987 return mem_cgroup_from_res_counter(mem
->res
.parent
, res
);
3990 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
3991 static void __init
enable_swap_cgroup(void)
3993 if (!mem_cgroup_disabled() && really_do_swap_account
)
3994 do_swap_account
= 1;
3997 static void __init
enable_swap_cgroup(void)
4002 static int mem_cgroup_soft_limit_tree_init(void)
4004 struct mem_cgroup_tree_per_node
*rtpn
;
4005 struct mem_cgroup_tree_per_zone
*rtpz
;
4006 int tmp
, node
, zone
;
4008 for_each_node_state(node
, N_POSSIBLE
) {
4010 if (!node_state(node
, N_NORMAL_MEMORY
))
4012 rtpn
= kzalloc_node(sizeof(*rtpn
), GFP_KERNEL
, tmp
);
4016 soft_limit_tree
.rb_tree_per_node
[node
] = rtpn
;
4018 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
4019 rtpz
= &rtpn
->rb_tree_per_zone
[zone
];
4020 rtpz
->rb_root
= RB_ROOT
;
4021 spin_lock_init(&rtpz
->lock
);
4027 static struct cgroup_subsys_state
* __ref
4028 mem_cgroup_create(struct cgroup_subsys
*ss
, struct cgroup
*cont
)
4030 struct mem_cgroup
*mem
, *parent
;
4031 long error
= -ENOMEM
;
4034 mem
= mem_cgroup_alloc();
4036 return ERR_PTR(error
);
4038 for_each_node_state(node
, N_POSSIBLE
)
4039 if (alloc_mem_cgroup_per_zone_info(mem
, node
))
4043 if (cont
->parent
== NULL
) {
4045 enable_swap_cgroup();
4047 root_mem_cgroup
= mem
;
4048 if (mem_cgroup_soft_limit_tree_init())
4050 for_each_possible_cpu(cpu
) {
4051 struct memcg_stock_pcp
*stock
=
4052 &per_cpu(memcg_stock
, cpu
);
4053 INIT_WORK(&stock
->work
, drain_local_stock
);
4055 hotcpu_notifier(memcg_stock_cpu_callback
, 0);
4057 parent
= mem_cgroup_from_cont(cont
->parent
);
4058 mem
->use_hierarchy
= parent
->use_hierarchy
;
4059 mem
->oom_kill_disable
= parent
->oom_kill_disable
;
4062 if (parent
&& parent
->use_hierarchy
) {
4063 res_counter_init(&mem
->res
, &parent
->res
);
4064 res_counter_init(&mem
->memsw
, &parent
->memsw
);
4066 * We increment refcnt of the parent to ensure that we can
4067 * safely access it on res_counter_charge/uncharge.
4068 * This refcnt will be decremented when freeing this
4069 * mem_cgroup(see mem_cgroup_put).
4071 mem_cgroup_get(parent
);
4073 res_counter_init(&mem
->res
, NULL
);
4074 res_counter_init(&mem
->memsw
, NULL
);
4076 mem
->last_scanned_child
= 0;
4077 spin_lock_init(&mem
->reclaim_param_lock
);
4078 INIT_LIST_HEAD(&mem
->oom_notify
);
4081 mem
->swappiness
= get_swappiness(parent
);
4082 atomic_set(&mem
->refcnt
, 1);
4083 mem
->move_charge_at_immigrate
= 0;
4084 mutex_init(&mem
->thresholds_lock
);
4087 __mem_cgroup_free(mem
);
4088 root_mem_cgroup
= NULL
;
4089 return ERR_PTR(error
);
4092 static int mem_cgroup_pre_destroy(struct cgroup_subsys
*ss
,
4093 struct cgroup
*cont
)
4095 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
4097 return mem_cgroup_force_empty(mem
, false);
4100 static void mem_cgroup_destroy(struct cgroup_subsys
*ss
,
4101 struct cgroup
*cont
)
4103 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
4105 mem_cgroup_put(mem
);
4108 static int mem_cgroup_populate(struct cgroup_subsys
*ss
,
4109 struct cgroup
*cont
)
4113 ret
= cgroup_add_files(cont
, ss
, mem_cgroup_files
,
4114 ARRAY_SIZE(mem_cgroup_files
));
4117 ret
= register_memsw_files(cont
, ss
);
4122 /* Handlers for move charge at task migration. */
4123 #define PRECHARGE_COUNT_AT_ONCE 256
4124 static int mem_cgroup_do_precharge(unsigned long count
)
4127 int batch_count
= PRECHARGE_COUNT_AT_ONCE
;
4128 struct mem_cgroup
*mem
= mc
.to
;
4130 if (mem_cgroup_is_root(mem
)) {
4131 mc
.precharge
+= count
;
4132 /* we don't need css_get for root */
4135 /* try to charge at once */
4137 struct res_counter
*dummy
;
4139 * "mem" cannot be under rmdir() because we've already checked
4140 * by cgroup_lock_live_cgroup() that it is not removed and we
4141 * are still under the same cgroup_mutex. So we can postpone
4144 if (res_counter_charge(&mem
->res
, PAGE_SIZE
* count
, &dummy
))
4146 if (do_swap_account
&& res_counter_charge(&mem
->memsw
,
4147 PAGE_SIZE
* count
, &dummy
)) {
4148 res_counter_uncharge(&mem
->res
, PAGE_SIZE
* count
);
4151 mc
.precharge
+= count
;
4152 VM_BUG_ON(test_bit(CSS_ROOT
, &mem
->css
.flags
));
4153 WARN_ON_ONCE(count
> INT_MAX
);
4154 __css_get(&mem
->css
, (int)count
);
4158 /* fall back to one by one charge */
4160 if (signal_pending(current
)) {
4164 if (!batch_count
--) {
4165 batch_count
= PRECHARGE_COUNT_AT_ONCE
;
4168 ret
= __mem_cgroup_try_charge(NULL
, GFP_KERNEL
, &mem
, false);
4170 /* mem_cgroup_clear_mc() will do uncharge later */
4178 * is_target_pte_for_mc - check a pte whether it is valid for move charge
4179 * @vma: the vma the pte to be checked belongs
4180 * @addr: the address corresponding to the pte to be checked
4181 * @ptent: the pte to be checked
4182 * @target: the pointer the target page or swap ent will be stored(can be NULL)
4185 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
4186 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
4187 * move charge. if @target is not NULL, the page is stored in target->page
4188 * with extra refcnt got(Callers should handle it).
4189 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
4190 * target for charge migration. if @target is not NULL, the entry is stored
4193 * Called with pte lock held.
4200 enum mc_target_type
{
4201 MC_TARGET_NONE
, /* not used */
4206 static struct page
*mc_handle_present_pte(struct vm_area_struct
*vma
,
4207 unsigned long addr
, pte_t ptent
)
4209 struct page
*page
= vm_normal_page(vma
, addr
, ptent
);
4211 if (!page
|| !page_mapped(page
))
4213 if (PageAnon(page
)) {
4214 /* we don't move shared anon */
4215 if (!move_anon() || page_mapcount(page
) > 2)
4217 } else if (!move_file())
4218 /* we ignore mapcount for file pages */
4220 if (!get_page_unless_zero(page
))
4226 static struct page
*mc_handle_swap_pte(struct vm_area_struct
*vma
,
4227 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
4230 struct page
*page
= NULL
;
4231 swp_entry_t ent
= pte_to_swp_entry(ptent
);
4233 if (!move_anon() || non_swap_entry(ent
))
4235 usage_count
= mem_cgroup_count_swap_user(ent
, &page
);
4236 if (usage_count
> 1) { /* we don't move shared anon */
4241 if (do_swap_account
)
4242 entry
->val
= ent
.val
;
4247 static struct page
*mc_handle_file_pte(struct vm_area_struct
*vma
,
4248 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
4250 struct page
*page
= NULL
;
4251 struct inode
*inode
;
4252 struct address_space
*mapping
;
4255 if (!vma
->vm_file
) /* anonymous vma */
4260 inode
= vma
->vm_file
->f_path
.dentry
->d_inode
;
4261 mapping
= vma
->vm_file
->f_mapping
;
4262 if (pte_none(ptent
))
4263 pgoff
= linear_page_index(vma
, addr
);
4264 else /* pte_file(ptent) is true */
4265 pgoff
= pte_to_pgoff(ptent
);
4267 /* page is moved even if it's not RSS of this task(page-faulted). */
4268 if (!mapping_cap_swap_backed(mapping
)) { /* normal file */
4269 page
= find_get_page(mapping
, pgoff
);
4270 } else { /* shmem/tmpfs file. we should take account of swap too. */
4272 mem_cgroup_get_shmem_target(inode
, pgoff
, &page
, &ent
);
4273 if (do_swap_account
)
4274 entry
->val
= ent
.val
;
4280 static int is_target_pte_for_mc(struct vm_area_struct
*vma
,
4281 unsigned long addr
, pte_t ptent
, union mc_target
*target
)
4283 struct page
*page
= NULL
;
4284 struct page_cgroup
*pc
;
4286 swp_entry_t ent
= { .val
= 0 };
4288 if (pte_present(ptent
))
4289 page
= mc_handle_present_pte(vma
, addr
, ptent
);
4290 else if (is_swap_pte(ptent
))
4291 page
= mc_handle_swap_pte(vma
, addr
, ptent
, &ent
);
4292 else if (pte_none(ptent
) || pte_file(ptent
))
4293 page
= mc_handle_file_pte(vma
, addr
, ptent
, &ent
);
4295 if (!page
&& !ent
.val
)
4298 pc
= lookup_page_cgroup(page
);
4300 * Do only loose check w/o page_cgroup lock.
4301 * mem_cgroup_move_account() checks the pc is valid or not under
4304 if (PageCgroupUsed(pc
) && pc
->mem_cgroup
== mc
.from
) {
4305 ret
= MC_TARGET_PAGE
;
4307 target
->page
= page
;
4309 if (!ret
|| !target
)
4312 /* There is a swap entry and a page doesn't exist or isn't charged */
4313 if (ent
.val
&& !ret
&&
4314 css_id(&mc
.from
->css
) == lookup_swap_cgroup(ent
)) {
4315 ret
= MC_TARGET_SWAP
;
4322 static int mem_cgroup_count_precharge_pte_range(pmd_t
*pmd
,
4323 unsigned long addr
, unsigned long end
,
4324 struct mm_walk
*walk
)
4326 struct vm_area_struct
*vma
= walk
->private;
4330 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
4331 for (; addr
!= end
; pte
++, addr
+= PAGE_SIZE
)
4332 if (is_target_pte_for_mc(vma
, addr
, *pte
, NULL
))
4333 mc
.precharge
++; /* increment precharge temporarily */
4334 pte_unmap_unlock(pte
- 1, ptl
);
4340 static unsigned long mem_cgroup_count_precharge(struct mm_struct
*mm
)
4342 unsigned long precharge
;
4343 struct vm_area_struct
*vma
;
4345 down_read(&mm
->mmap_sem
);
4346 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
4347 struct mm_walk mem_cgroup_count_precharge_walk
= {
4348 .pmd_entry
= mem_cgroup_count_precharge_pte_range
,
4352 if (is_vm_hugetlb_page(vma
))
4354 walk_page_range(vma
->vm_start
, vma
->vm_end
,
4355 &mem_cgroup_count_precharge_walk
);
4357 up_read(&mm
->mmap_sem
);
4359 precharge
= mc
.precharge
;
4365 static int mem_cgroup_precharge_mc(struct mm_struct
*mm
)
4367 return mem_cgroup_do_precharge(mem_cgroup_count_precharge(mm
));
4370 static void mem_cgroup_clear_mc(void)
4372 /* we must uncharge all the leftover precharges from mc.to */
4374 __mem_cgroup_cancel_charge(mc
.to
, mc
.precharge
);
4376 memcg_oom_recover(mc
.to
);
4379 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
4380 * we must uncharge here.
4382 if (mc
.moved_charge
) {
4383 __mem_cgroup_cancel_charge(mc
.from
, mc
.moved_charge
);
4384 mc
.moved_charge
= 0;
4385 memcg_oom_recover(mc
.from
);
4387 /* we must fixup refcnts and charges */
4388 if (mc
.moved_swap
) {
4389 WARN_ON_ONCE(mc
.moved_swap
> INT_MAX
);
4390 /* uncharge swap account from the old cgroup */
4391 if (!mem_cgroup_is_root(mc
.from
))
4392 res_counter_uncharge(&mc
.from
->memsw
,
4393 PAGE_SIZE
* mc
.moved_swap
);
4394 __mem_cgroup_put(mc
.from
, mc
.moved_swap
);
4396 if (!mem_cgroup_is_root(mc
.to
)) {
4398 * we charged both to->res and to->memsw, so we should
4401 res_counter_uncharge(&mc
.to
->res
,
4402 PAGE_SIZE
* mc
.moved_swap
);
4403 VM_BUG_ON(test_bit(CSS_ROOT
, &mc
.to
->css
.flags
));
4404 __css_put(&mc
.to
->css
, mc
.moved_swap
);
4406 /* we've already done mem_cgroup_get(mc.to) */
4412 mc
.moving_task
= NULL
;
4413 wake_up_all(&mc
.waitq
);
4416 static int mem_cgroup_can_attach(struct cgroup_subsys
*ss
,
4417 struct cgroup
*cgroup
,
4418 struct task_struct
*p
,
4422 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgroup
);
4424 if (mem
->move_charge_at_immigrate
) {
4425 struct mm_struct
*mm
;
4426 struct mem_cgroup
*from
= mem_cgroup_from_task(p
);
4428 VM_BUG_ON(from
== mem
);
4430 mm
= get_task_mm(p
);
4433 /* We move charges only when we move a owner of the mm */
4434 if (mm
->owner
== p
) {
4437 VM_BUG_ON(mc
.precharge
);
4438 VM_BUG_ON(mc
.moved_charge
);
4439 VM_BUG_ON(mc
.moved_swap
);
4440 VM_BUG_ON(mc
.moving_task
);
4444 mc
.moved_charge
= 0;
4446 mc
.moving_task
= current
;
4448 ret
= mem_cgroup_precharge_mc(mm
);
4450 mem_cgroup_clear_mc();
4457 static void mem_cgroup_cancel_attach(struct cgroup_subsys
*ss
,
4458 struct cgroup
*cgroup
,
4459 struct task_struct
*p
,
4462 mem_cgroup_clear_mc();
4465 static int mem_cgroup_move_charge_pte_range(pmd_t
*pmd
,
4466 unsigned long addr
, unsigned long end
,
4467 struct mm_walk
*walk
)
4470 struct vm_area_struct
*vma
= walk
->private;
4475 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
4476 for (; addr
!= end
; addr
+= PAGE_SIZE
) {
4477 pte_t ptent
= *(pte
++);
4478 union mc_target target
;
4481 struct page_cgroup
*pc
;
4487 type
= is_target_pte_for_mc(vma
, addr
, ptent
, &target
);
4489 case MC_TARGET_PAGE
:
4491 if (isolate_lru_page(page
))
4493 pc
= lookup_page_cgroup(page
);
4494 if (!mem_cgroup_move_account(pc
,
4495 mc
.from
, mc
.to
, false)) {
4497 /* we uncharge from mc.from later. */
4500 putback_lru_page(page
);
4501 put
: /* is_target_pte_for_mc() gets the page */
4504 case MC_TARGET_SWAP
:
4506 if (!mem_cgroup_move_swap_account(ent
,
4507 mc
.from
, mc
.to
, false)) {
4509 /* we fixup refcnts and charges later. */
4517 pte_unmap_unlock(pte
- 1, ptl
);
4522 * We have consumed all precharges we got in can_attach().
4523 * We try charge one by one, but don't do any additional
4524 * charges to mc.to if we have failed in charge once in attach()
4527 ret
= mem_cgroup_do_precharge(1);
4535 static void mem_cgroup_move_charge(struct mm_struct
*mm
)
4537 struct vm_area_struct
*vma
;
4539 lru_add_drain_all();
4540 down_read(&mm
->mmap_sem
);
4541 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
4543 struct mm_walk mem_cgroup_move_charge_walk
= {
4544 .pmd_entry
= mem_cgroup_move_charge_pte_range
,
4548 if (is_vm_hugetlb_page(vma
))
4550 ret
= walk_page_range(vma
->vm_start
, vma
->vm_end
,
4551 &mem_cgroup_move_charge_walk
);
4554 * means we have consumed all precharges and failed in
4555 * doing additional charge. Just abandon here.
4559 up_read(&mm
->mmap_sem
);
4562 static void mem_cgroup_move_task(struct cgroup_subsys
*ss
,
4563 struct cgroup
*cont
,
4564 struct cgroup
*old_cont
,
4565 struct task_struct
*p
,
4568 struct mm_struct
*mm
;
4571 /* no need to move charge */
4574 mm
= get_task_mm(p
);
4576 mem_cgroup_move_charge(mm
);
4579 mem_cgroup_clear_mc();
4581 #else /* !CONFIG_MMU */
4582 static int mem_cgroup_can_attach(struct cgroup_subsys
*ss
,
4583 struct cgroup
*cgroup
,
4584 struct task_struct
*p
,
4589 static void mem_cgroup_cancel_attach(struct cgroup_subsys
*ss
,
4590 struct cgroup
*cgroup
,
4591 struct task_struct
*p
,
4595 static void mem_cgroup_move_task(struct cgroup_subsys
*ss
,
4596 struct cgroup
*cont
,
4597 struct cgroup
*old_cont
,
4598 struct task_struct
*p
,
4604 struct cgroup_subsys mem_cgroup_subsys
= {
4606 .subsys_id
= mem_cgroup_subsys_id
,
4607 .create
= mem_cgroup_create
,
4608 .pre_destroy
= mem_cgroup_pre_destroy
,
4609 .destroy
= mem_cgroup_destroy
,
4610 .populate
= mem_cgroup_populate
,
4611 .can_attach
= mem_cgroup_can_attach
,
4612 .cancel_attach
= mem_cgroup_cancel_attach
,
4613 .attach
= mem_cgroup_move_task
,
4618 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4620 static int __init
disable_swap_account(char *s
)
4622 really_do_swap_account
= 0;
4625 __setup("noswapaccount", disable_swap_account
);