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>
50 #include <linux/oom.h>
53 #include <asm/uaccess.h>
55 #include <trace/events/vmscan.h>
57 struct cgroup_subsys mem_cgroup_subsys __read_mostly
;
58 #define MEM_CGROUP_RECLAIM_RETRIES 5
59 struct mem_cgroup
*root_mem_cgroup __read_mostly
;
61 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
62 /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
63 int do_swap_account __read_mostly
;
65 /* for remember boot option*/
66 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP_ENABLED
67 static int really_do_swap_account __initdata
= 1;
69 static int really_do_swap_account __initdata
= 0;
73 #define do_swap_account (0)
77 * Per memcg event counter is incremented at every pagein/pageout. This counter
78 * is used for trigger some periodic events. This is straightforward and better
79 * than using jiffies etc. to handle periodic memcg event.
81 * These values will be used as !((event) & ((1 <<(thresh)) - 1))
83 #define THRESHOLDS_EVENTS_THRESH (7) /* once in 128 */
84 #define SOFTLIMIT_EVENTS_THRESH (10) /* once in 1024 */
87 * Statistics for memory cgroup.
89 enum mem_cgroup_stat_index
{
91 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
93 MEM_CGROUP_STAT_CACHE
, /* # of pages charged as cache */
94 MEM_CGROUP_STAT_RSS
, /* # of pages charged as anon rss */
95 MEM_CGROUP_STAT_FILE_MAPPED
, /* # of pages charged as file rss */
96 MEM_CGROUP_STAT_PGPGIN_COUNT
, /* # of pages paged in */
97 MEM_CGROUP_STAT_PGPGOUT_COUNT
, /* # of pages paged out */
98 MEM_CGROUP_STAT_SWAPOUT
, /* # of pages, swapped out */
99 MEM_CGROUP_STAT_DATA
, /* end of data requires synchronization */
100 /* incremented at every pagein/pageout */
101 MEM_CGROUP_EVENTS
= MEM_CGROUP_STAT_DATA
,
102 MEM_CGROUP_ON_MOVE
, /* someone is moving account between groups */
104 MEM_CGROUP_STAT_NSTATS
,
107 struct mem_cgroup_stat_cpu
{
108 s64 count
[MEM_CGROUP_STAT_NSTATS
];
112 * per-zone information in memory controller.
114 struct mem_cgroup_per_zone
{
116 * spin_lock to protect the per cgroup LRU
118 struct list_head lists
[NR_LRU_LISTS
];
119 unsigned long count
[NR_LRU_LISTS
];
121 struct zone_reclaim_stat reclaim_stat
;
122 struct rb_node tree_node
; /* RB tree node */
123 unsigned long long usage_in_excess
;/* Set to the value by which */
124 /* the soft limit is exceeded*/
126 struct mem_cgroup
*mem
; /* Back pointer, we cannot */
127 /* use container_of */
129 /* Macro for accessing counter */
130 #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
132 struct mem_cgroup_per_node
{
133 struct mem_cgroup_per_zone zoneinfo
[MAX_NR_ZONES
];
136 struct mem_cgroup_lru_info
{
137 struct mem_cgroup_per_node
*nodeinfo
[MAX_NUMNODES
];
141 * Cgroups above their limits are maintained in a RB-Tree, independent of
142 * their hierarchy representation
145 struct mem_cgroup_tree_per_zone
{
146 struct rb_root rb_root
;
150 struct mem_cgroup_tree_per_node
{
151 struct mem_cgroup_tree_per_zone rb_tree_per_zone
[MAX_NR_ZONES
];
154 struct mem_cgroup_tree
{
155 struct mem_cgroup_tree_per_node
*rb_tree_per_node
[MAX_NUMNODES
];
158 static struct mem_cgroup_tree soft_limit_tree __read_mostly
;
160 struct mem_cgroup_threshold
{
161 struct eventfd_ctx
*eventfd
;
166 struct mem_cgroup_threshold_ary
{
167 /* An array index points to threshold just below usage. */
168 int current_threshold
;
169 /* Size of entries[] */
171 /* Array of thresholds */
172 struct mem_cgroup_threshold entries
[0];
175 struct mem_cgroup_thresholds
{
176 /* Primary thresholds array */
177 struct mem_cgroup_threshold_ary
*primary
;
179 * Spare threshold array.
180 * This is needed to make mem_cgroup_unregister_event() "never fail".
181 * It must be able to store at least primary->size - 1 entries.
183 struct mem_cgroup_threshold_ary
*spare
;
187 struct mem_cgroup_eventfd_list
{
188 struct list_head list
;
189 struct eventfd_ctx
*eventfd
;
192 static void mem_cgroup_threshold(struct mem_cgroup
*mem
);
193 static void mem_cgroup_oom_notify(struct mem_cgroup
*mem
);
196 * The memory controller data structure. The memory controller controls both
197 * page cache and RSS per cgroup. We would eventually like to provide
198 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
199 * to help the administrator determine what knobs to tune.
201 * TODO: Add a water mark for the memory controller. Reclaim will begin when
202 * we hit the water mark. May be even add a low water mark, such that
203 * no reclaim occurs from a cgroup at it's low water mark, this is
204 * a feature that will be implemented much later in the future.
207 struct cgroup_subsys_state css
;
209 * the counter to account for memory usage
211 struct res_counter res
;
213 * the counter to account for mem+swap usage.
215 struct res_counter memsw
;
217 * Per cgroup active and inactive list, similar to the
218 * per zone LRU lists.
220 struct mem_cgroup_lru_info info
;
223 protect against reclaim related member.
225 spinlock_t reclaim_param_lock
;
228 * While reclaiming in a hierarchy, we cache the last child we
231 int last_scanned_child
;
233 * Should the accounting and control be hierarchical, per subtree?
239 unsigned int swappiness
;
240 /* OOM-Killer disable */
241 int oom_kill_disable
;
243 /* set when res.limit == memsw.limit */
244 bool memsw_is_minimum
;
246 /* protect arrays of thresholds */
247 struct mutex thresholds_lock
;
249 /* thresholds for memory usage. RCU-protected */
250 struct mem_cgroup_thresholds thresholds
;
252 /* thresholds for mem+swap usage. RCU-protected */
253 struct mem_cgroup_thresholds memsw_thresholds
;
255 /* For oom notifier event fd */
256 struct list_head oom_notify
;
259 * Should we move charges of a task when a task is moved into this
260 * mem_cgroup ? And what type of charges should we move ?
262 unsigned long move_charge_at_immigrate
;
266 struct mem_cgroup_stat_cpu
*stat
;
268 * used when a cpu is offlined or other synchronizations
269 * See mem_cgroup_read_stat().
271 struct mem_cgroup_stat_cpu nocpu_base
;
272 spinlock_t pcp_counter_lock
;
275 /* Stuffs for move charges at task migration. */
277 * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a
278 * left-shifted bitmap of these types.
281 MOVE_CHARGE_TYPE_ANON
, /* private anonymous page and swap of it */
282 MOVE_CHARGE_TYPE_FILE
, /* file page(including tmpfs) and swap of it */
286 /* "mc" and its members are protected by cgroup_mutex */
287 static struct move_charge_struct
{
288 spinlock_t lock
; /* for from, to */
289 struct mem_cgroup
*from
;
290 struct mem_cgroup
*to
;
291 unsigned long precharge
;
292 unsigned long moved_charge
;
293 unsigned long moved_swap
;
294 struct task_struct
*moving_task
; /* a task moving charges */
295 wait_queue_head_t waitq
; /* a waitq for other context */
297 .lock
= __SPIN_LOCK_UNLOCKED(mc
.lock
),
298 .waitq
= __WAIT_QUEUE_HEAD_INITIALIZER(mc
.waitq
),
301 static bool move_anon(void)
303 return test_bit(MOVE_CHARGE_TYPE_ANON
,
304 &mc
.to
->move_charge_at_immigrate
);
307 static bool move_file(void)
309 return test_bit(MOVE_CHARGE_TYPE_FILE
,
310 &mc
.to
->move_charge_at_immigrate
);
314 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
315 * limit reclaim to prevent infinite loops, if they ever occur.
317 #define MEM_CGROUP_MAX_RECLAIM_LOOPS (100)
318 #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2)
321 MEM_CGROUP_CHARGE_TYPE_CACHE
= 0,
322 MEM_CGROUP_CHARGE_TYPE_MAPPED
,
323 MEM_CGROUP_CHARGE_TYPE_SHMEM
, /* used by page migration of shmem */
324 MEM_CGROUP_CHARGE_TYPE_FORCE
, /* used by force_empty */
325 MEM_CGROUP_CHARGE_TYPE_SWAPOUT
, /* for accounting swapcache */
326 MEM_CGROUP_CHARGE_TYPE_DROP
, /* a page was unused swap cache */
330 /* only for here (for easy reading.) */
331 #define PCGF_CACHE (1UL << PCG_CACHE)
332 #define PCGF_USED (1UL << PCG_USED)
333 #define PCGF_LOCK (1UL << PCG_LOCK)
334 /* Not used, but added here for completeness */
335 #define PCGF_ACCT (1UL << PCG_ACCT)
337 /* for encoding cft->private value on file */
340 #define _OOM_TYPE (2)
341 #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val))
342 #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff)
343 #define MEMFILE_ATTR(val) ((val) & 0xffff)
344 /* Used for OOM nofiier */
345 #define OOM_CONTROL (0)
348 * Reclaim flags for mem_cgroup_hierarchical_reclaim
350 #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
351 #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
352 #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
353 #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
354 #define MEM_CGROUP_RECLAIM_SOFT_BIT 0x2
355 #define MEM_CGROUP_RECLAIM_SOFT (1 << MEM_CGROUP_RECLAIM_SOFT_BIT)
357 static void mem_cgroup_get(struct mem_cgroup
*mem
);
358 static void mem_cgroup_put(struct mem_cgroup
*mem
);
359 static struct mem_cgroup
*parent_mem_cgroup(struct mem_cgroup
*mem
);
360 static void drain_all_stock_async(void);
362 static struct mem_cgroup_per_zone
*
363 mem_cgroup_zoneinfo(struct mem_cgroup
*mem
, int nid
, int zid
)
365 return &mem
->info
.nodeinfo
[nid
]->zoneinfo
[zid
];
368 struct cgroup_subsys_state
*mem_cgroup_css(struct mem_cgroup
*mem
)
373 static struct mem_cgroup_per_zone
*
374 page_cgroup_zoneinfo(struct page_cgroup
*pc
)
376 struct mem_cgroup
*mem
= pc
->mem_cgroup
;
377 int nid
= page_cgroup_nid(pc
);
378 int zid
= page_cgroup_zid(pc
);
383 return mem_cgroup_zoneinfo(mem
, nid
, zid
);
386 static struct mem_cgroup_tree_per_zone
*
387 soft_limit_tree_node_zone(int nid
, int zid
)
389 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
392 static struct mem_cgroup_tree_per_zone
*
393 soft_limit_tree_from_page(struct page
*page
)
395 int nid
= page_to_nid(page
);
396 int zid
= page_zonenum(page
);
398 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
402 __mem_cgroup_insert_exceeded(struct mem_cgroup
*mem
,
403 struct mem_cgroup_per_zone
*mz
,
404 struct mem_cgroup_tree_per_zone
*mctz
,
405 unsigned long long new_usage_in_excess
)
407 struct rb_node
**p
= &mctz
->rb_root
.rb_node
;
408 struct rb_node
*parent
= NULL
;
409 struct mem_cgroup_per_zone
*mz_node
;
414 mz
->usage_in_excess
= new_usage_in_excess
;
415 if (!mz
->usage_in_excess
)
419 mz_node
= rb_entry(parent
, struct mem_cgroup_per_zone
,
421 if (mz
->usage_in_excess
< mz_node
->usage_in_excess
)
424 * We can't avoid mem cgroups that are over their soft
425 * limit by the same amount
427 else if (mz
->usage_in_excess
>= mz_node
->usage_in_excess
)
430 rb_link_node(&mz
->tree_node
, parent
, p
);
431 rb_insert_color(&mz
->tree_node
, &mctz
->rb_root
);
436 __mem_cgroup_remove_exceeded(struct mem_cgroup
*mem
,
437 struct mem_cgroup_per_zone
*mz
,
438 struct mem_cgroup_tree_per_zone
*mctz
)
442 rb_erase(&mz
->tree_node
, &mctz
->rb_root
);
447 mem_cgroup_remove_exceeded(struct mem_cgroup
*mem
,
448 struct mem_cgroup_per_zone
*mz
,
449 struct mem_cgroup_tree_per_zone
*mctz
)
451 spin_lock(&mctz
->lock
);
452 __mem_cgroup_remove_exceeded(mem
, mz
, mctz
);
453 spin_unlock(&mctz
->lock
);
457 static void mem_cgroup_update_tree(struct mem_cgroup
*mem
, struct page
*page
)
459 unsigned long long excess
;
460 struct mem_cgroup_per_zone
*mz
;
461 struct mem_cgroup_tree_per_zone
*mctz
;
462 int nid
= page_to_nid(page
);
463 int zid
= page_zonenum(page
);
464 mctz
= soft_limit_tree_from_page(page
);
467 * Necessary to update all ancestors when hierarchy is used.
468 * because their event counter is not touched.
470 for (; mem
; mem
= parent_mem_cgroup(mem
)) {
471 mz
= mem_cgroup_zoneinfo(mem
, nid
, zid
);
472 excess
= res_counter_soft_limit_excess(&mem
->res
);
474 * We have to update the tree if mz is on RB-tree or
475 * mem is over its softlimit.
477 if (excess
|| mz
->on_tree
) {
478 spin_lock(&mctz
->lock
);
479 /* if on-tree, remove it */
481 __mem_cgroup_remove_exceeded(mem
, mz
, mctz
);
483 * Insert again. mz->usage_in_excess will be updated.
484 * If excess is 0, no tree ops.
486 __mem_cgroup_insert_exceeded(mem
, mz
, mctz
, excess
);
487 spin_unlock(&mctz
->lock
);
492 static void mem_cgroup_remove_from_trees(struct mem_cgroup
*mem
)
495 struct mem_cgroup_per_zone
*mz
;
496 struct mem_cgroup_tree_per_zone
*mctz
;
498 for_each_node_state(node
, N_POSSIBLE
) {
499 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
500 mz
= mem_cgroup_zoneinfo(mem
, node
, zone
);
501 mctz
= soft_limit_tree_node_zone(node
, zone
);
502 mem_cgroup_remove_exceeded(mem
, mz
, mctz
);
507 static struct mem_cgroup_per_zone
*
508 __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
510 struct rb_node
*rightmost
= NULL
;
511 struct mem_cgroup_per_zone
*mz
;
515 rightmost
= rb_last(&mctz
->rb_root
);
517 goto done
; /* Nothing to reclaim from */
519 mz
= rb_entry(rightmost
, struct mem_cgroup_per_zone
, tree_node
);
521 * Remove the node now but someone else can add it back,
522 * we will to add it back at the end of reclaim to its correct
523 * position in the tree.
525 __mem_cgroup_remove_exceeded(mz
->mem
, mz
, mctz
);
526 if (!res_counter_soft_limit_excess(&mz
->mem
->res
) ||
527 !css_tryget(&mz
->mem
->css
))
533 static struct mem_cgroup_per_zone
*
534 mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
536 struct mem_cgroup_per_zone
*mz
;
538 spin_lock(&mctz
->lock
);
539 mz
= __mem_cgroup_largest_soft_limit_node(mctz
);
540 spin_unlock(&mctz
->lock
);
545 * Implementation Note: reading percpu statistics for memcg.
547 * Both of vmstat[] and percpu_counter has threshold and do periodic
548 * synchronization to implement "quick" read. There are trade-off between
549 * reading cost and precision of value. Then, we may have a chance to implement
550 * a periodic synchronizion of counter in memcg's counter.
552 * But this _read() function is used for user interface now. The user accounts
553 * memory usage by memory cgroup and he _always_ requires exact value because
554 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
555 * have to visit all online cpus and make sum. So, for now, unnecessary
556 * synchronization is not implemented. (just implemented for cpu hotplug)
558 * If there are kernel internal actions which can make use of some not-exact
559 * value, and reading all cpu value can be performance bottleneck in some
560 * common workload, threashold and synchonization as vmstat[] should be
563 static s64
mem_cgroup_read_stat(struct mem_cgroup
*mem
,
564 enum mem_cgroup_stat_index idx
)
570 for_each_online_cpu(cpu
)
571 val
+= per_cpu(mem
->stat
->count
[idx
], cpu
);
572 #ifdef CONFIG_HOTPLUG_CPU
573 spin_lock(&mem
->pcp_counter_lock
);
574 val
+= mem
->nocpu_base
.count
[idx
];
575 spin_unlock(&mem
->pcp_counter_lock
);
581 static s64
mem_cgroup_local_usage(struct mem_cgroup
*mem
)
585 ret
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_RSS
);
586 ret
+= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_CACHE
);
590 static void mem_cgroup_swap_statistics(struct mem_cgroup
*mem
,
593 int val
= (charge
) ? 1 : -1;
594 this_cpu_add(mem
->stat
->count
[MEM_CGROUP_STAT_SWAPOUT
], val
);
597 static void mem_cgroup_charge_statistics(struct mem_cgroup
*mem
,
598 bool file
, int nr_pages
)
603 __this_cpu_add(mem
->stat
->count
[MEM_CGROUP_STAT_CACHE
], nr_pages
);
605 __this_cpu_add(mem
->stat
->count
[MEM_CGROUP_STAT_RSS
], nr_pages
);
607 /* pagein of a big page is an event. So, ignore page size */
609 __this_cpu_inc(mem
->stat
->count
[MEM_CGROUP_STAT_PGPGIN_COUNT
]);
611 __this_cpu_inc(mem
->stat
->count
[MEM_CGROUP_STAT_PGPGOUT_COUNT
]);
612 nr_pages
= -nr_pages
; /* for event */
615 __this_cpu_add(mem
->stat
->count
[MEM_CGROUP_EVENTS
], nr_pages
);
620 static unsigned long mem_cgroup_get_local_zonestat(struct mem_cgroup
*mem
,
624 struct mem_cgroup_per_zone
*mz
;
627 for_each_online_node(nid
)
628 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
629 mz
= mem_cgroup_zoneinfo(mem
, nid
, zid
);
630 total
+= MEM_CGROUP_ZSTAT(mz
, idx
);
635 static bool __memcg_event_check(struct mem_cgroup
*mem
, int event_mask_shift
)
639 val
= this_cpu_read(mem
->stat
->count
[MEM_CGROUP_EVENTS
]);
641 return !(val
& ((1 << event_mask_shift
) - 1));
645 * Check events in order.
648 static void memcg_check_events(struct mem_cgroup
*mem
, struct page
*page
)
650 /* threshold event is triggered in finer grain than soft limit */
651 if (unlikely(__memcg_event_check(mem
, THRESHOLDS_EVENTS_THRESH
))) {
652 mem_cgroup_threshold(mem
);
653 if (unlikely(__memcg_event_check(mem
, SOFTLIMIT_EVENTS_THRESH
)))
654 mem_cgroup_update_tree(mem
, page
);
658 static struct mem_cgroup
*mem_cgroup_from_cont(struct cgroup
*cont
)
660 return container_of(cgroup_subsys_state(cont
,
661 mem_cgroup_subsys_id
), struct mem_cgroup
,
665 struct mem_cgroup
*mem_cgroup_from_task(struct task_struct
*p
)
668 * mm_update_next_owner() may clear mm->owner to NULL
669 * if it races with swapoff, page migration, etc.
670 * So this can be called with p == NULL.
675 return container_of(task_subsys_state(p
, mem_cgroup_subsys_id
),
676 struct mem_cgroup
, css
);
679 static struct mem_cgroup
*try_get_mem_cgroup_from_mm(struct mm_struct
*mm
)
681 struct mem_cgroup
*mem
= NULL
;
686 * Because we have no locks, mm->owner's may be being moved to other
687 * cgroup. We use css_tryget() here even if this looks
688 * pessimistic (rather than adding locks here).
692 mem
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
695 } while (!css_tryget(&mem
->css
));
700 /* The caller has to guarantee "mem" exists before calling this */
701 static struct mem_cgroup
*mem_cgroup_start_loop(struct mem_cgroup
*mem
)
703 struct cgroup_subsys_state
*css
;
706 if (!mem
) /* ROOT cgroup has the smallest ID */
707 return root_mem_cgroup
; /*css_put/get against root is ignored*/
708 if (!mem
->use_hierarchy
) {
709 if (css_tryget(&mem
->css
))
715 * searching a memory cgroup which has the smallest ID under given
716 * ROOT cgroup. (ID >= 1)
718 css
= css_get_next(&mem_cgroup_subsys
, 1, &mem
->css
, &found
);
719 if (css
&& css_tryget(css
))
720 mem
= container_of(css
, struct mem_cgroup
, css
);
727 static struct mem_cgroup
*mem_cgroup_get_next(struct mem_cgroup
*iter
,
728 struct mem_cgroup
*root
,
731 int nextid
= css_id(&iter
->css
) + 1;
734 struct cgroup_subsys_state
*css
;
736 hierarchy_used
= iter
->use_hierarchy
;
739 /* If no ROOT, walk all, ignore hierarchy */
740 if (!cond
|| (root
&& !hierarchy_used
))
744 root
= root_mem_cgroup
;
750 css
= css_get_next(&mem_cgroup_subsys
, nextid
,
752 if (css
&& css_tryget(css
))
753 iter
= container_of(css
, struct mem_cgroup
, css
);
755 /* If css is NULL, no more cgroups will be found */
757 } while (css
&& !iter
);
762 * for_eacn_mem_cgroup_tree() for visiting all cgroup under tree. Please
763 * be careful that "break" loop is not allowed. We have reference count.
764 * Instead of that modify "cond" to be false and "continue" to exit the loop.
766 #define for_each_mem_cgroup_tree_cond(iter, root, cond) \
767 for (iter = mem_cgroup_start_loop(root);\
769 iter = mem_cgroup_get_next(iter, root, cond))
771 #define for_each_mem_cgroup_tree(iter, root) \
772 for_each_mem_cgroup_tree_cond(iter, root, true)
774 #define for_each_mem_cgroup_all(iter) \
775 for_each_mem_cgroup_tree_cond(iter, NULL, true)
778 static inline bool mem_cgroup_is_root(struct mem_cgroup
*mem
)
780 return (mem
== root_mem_cgroup
);
784 * Following LRU functions are allowed to be used without PCG_LOCK.
785 * Operations are called by routine of global LRU independently from memcg.
786 * What we have to take care of here is validness of pc->mem_cgroup.
788 * Changes to pc->mem_cgroup happens when
791 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
792 * It is added to LRU before charge.
793 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
794 * When moving account, the page is not on LRU. It's isolated.
797 void mem_cgroup_del_lru_list(struct page
*page
, enum lru_list lru
)
799 struct page_cgroup
*pc
;
800 struct mem_cgroup_per_zone
*mz
;
802 if (mem_cgroup_disabled())
804 pc
= lookup_page_cgroup(page
);
805 /* can happen while we handle swapcache. */
806 if (!TestClearPageCgroupAcctLRU(pc
))
808 VM_BUG_ON(!pc
->mem_cgroup
);
810 * We don't check PCG_USED bit. It's cleared when the "page" is finally
811 * removed from global LRU.
813 mz
= page_cgroup_zoneinfo(pc
);
814 /* huge page split is done under lru_lock. so, we have no races. */
815 MEM_CGROUP_ZSTAT(mz
, lru
) -= 1 << compound_order(page
);
816 if (mem_cgroup_is_root(pc
->mem_cgroup
))
818 VM_BUG_ON(list_empty(&pc
->lru
));
819 list_del_init(&pc
->lru
);
822 void mem_cgroup_del_lru(struct page
*page
)
824 mem_cgroup_del_lru_list(page
, page_lru(page
));
828 * Writeback is about to end against a page which has been marked for immediate
829 * reclaim. If it still appears to be reclaimable, move it to the tail of the
832 void mem_cgroup_rotate_reclaimable_page(struct page
*page
)
834 struct mem_cgroup_per_zone
*mz
;
835 struct page_cgroup
*pc
;
836 enum lru_list lru
= page_lru(page
);
838 if (mem_cgroup_disabled())
841 pc
= lookup_page_cgroup(page
);
842 /* unused or root page is not rotated. */
843 if (!PageCgroupUsed(pc
))
845 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
847 if (mem_cgroup_is_root(pc
->mem_cgroup
))
849 mz
= page_cgroup_zoneinfo(pc
);
850 list_move_tail(&pc
->lru
, &mz
->lists
[lru
]);
853 void mem_cgroup_rotate_lru_list(struct page
*page
, enum lru_list lru
)
855 struct mem_cgroup_per_zone
*mz
;
856 struct page_cgroup
*pc
;
858 if (mem_cgroup_disabled())
861 pc
= lookup_page_cgroup(page
);
862 /* unused or root page is not rotated. */
863 if (!PageCgroupUsed(pc
))
865 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
867 if (mem_cgroup_is_root(pc
->mem_cgroup
))
869 mz
= page_cgroup_zoneinfo(pc
);
870 list_move(&pc
->lru
, &mz
->lists
[lru
]);
873 void mem_cgroup_add_lru_list(struct page
*page
, enum lru_list lru
)
875 struct page_cgroup
*pc
;
876 struct mem_cgroup_per_zone
*mz
;
878 if (mem_cgroup_disabled())
880 pc
= lookup_page_cgroup(page
);
881 VM_BUG_ON(PageCgroupAcctLRU(pc
));
882 if (!PageCgroupUsed(pc
))
884 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
886 mz
= page_cgroup_zoneinfo(pc
);
887 /* huge page split is done under lru_lock. so, we have no races. */
888 MEM_CGROUP_ZSTAT(mz
, lru
) += 1 << compound_order(page
);
889 SetPageCgroupAcctLRU(pc
);
890 if (mem_cgroup_is_root(pc
->mem_cgroup
))
892 list_add(&pc
->lru
, &mz
->lists
[lru
]);
896 * At handling SwapCache, pc->mem_cgroup may be changed while it's linked to
897 * lru because the page may.be reused after it's fully uncharged (because of
898 * SwapCache behavior).To handle that, unlink page_cgroup from LRU when charge
899 * it again. This function is only used to charge SwapCache. It's done under
900 * lock_page and expected that zone->lru_lock is never held.
902 static void mem_cgroup_lru_del_before_commit_swapcache(struct page
*page
)
905 struct zone
*zone
= page_zone(page
);
906 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
908 spin_lock_irqsave(&zone
->lru_lock
, flags
);
910 * Forget old LRU when this page_cgroup is *not* used. This Used bit
911 * is guarded by lock_page() because the page is SwapCache.
913 if (!PageCgroupUsed(pc
))
914 mem_cgroup_del_lru_list(page
, page_lru(page
));
915 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
918 static void mem_cgroup_lru_add_after_commit_swapcache(struct page
*page
)
921 struct zone
*zone
= page_zone(page
);
922 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
924 spin_lock_irqsave(&zone
->lru_lock
, flags
);
925 /* link when the page is linked to LRU but page_cgroup isn't */
926 if (PageLRU(page
) && !PageCgroupAcctLRU(pc
))
927 mem_cgroup_add_lru_list(page
, page_lru(page
));
928 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
932 void mem_cgroup_move_lists(struct page
*page
,
933 enum lru_list from
, enum lru_list to
)
935 if (mem_cgroup_disabled())
937 mem_cgroup_del_lru_list(page
, from
);
938 mem_cgroup_add_lru_list(page
, to
);
941 int task_in_mem_cgroup(struct task_struct
*task
, const struct mem_cgroup
*mem
)
944 struct mem_cgroup
*curr
= NULL
;
945 struct task_struct
*p
;
947 p
= find_lock_task_mm(task
);
950 curr
= try_get_mem_cgroup_from_mm(p
->mm
);
955 * We should check use_hierarchy of "mem" not "curr". Because checking
956 * use_hierarchy of "curr" here make this function true if hierarchy is
957 * enabled in "curr" and "curr" is a child of "mem" in *cgroup*
958 * hierarchy(even if use_hierarchy is disabled in "mem").
960 if (mem
->use_hierarchy
)
961 ret
= css_is_ancestor(&curr
->css
, &mem
->css
);
968 static int calc_inactive_ratio(struct mem_cgroup
*memcg
, unsigned long *present_pages
)
970 unsigned long active
;
971 unsigned long inactive
;
973 unsigned long inactive_ratio
;
975 inactive
= mem_cgroup_get_local_zonestat(memcg
, LRU_INACTIVE_ANON
);
976 active
= mem_cgroup_get_local_zonestat(memcg
, LRU_ACTIVE_ANON
);
978 gb
= (inactive
+ active
) >> (30 - PAGE_SHIFT
);
980 inactive_ratio
= int_sqrt(10 * gb
);
985 present_pages
[0] = inactive
;
986 present_pages
[1] = active
;
989 return inactive_ratio
;
992 int mem_cgroup_inactive_anon_is_low(struct mem_cgroup
*memcg
)
994 unsigned long active
;
995 unsigned long inactive
;
996 unsigned long present_pages
[2];
997 unsigned long inactive_ratio
;
999 inactive_ratio
= calc_inactive_ratio(memcg
, present_pages
);
1001 inactive
= present_pages
[0];
1002 active
= present_pages
[1];
1004 if (inactive
* inactive_ratio
< active
)
1010 int mem_cgroup_inactive_file_is_low(struct mem_cgroup
*memcg
)
1012 unsigned long active
;
1013 unsigned long inactive
;
1015 inactive
= mem_cgroup_get_local_zonestat(memcg
, LRU_INACTIVE_FILE
);
1016 active
= mem_cgroup_get_local_zonestat(memcg
, LRU_ACTIVE_FILE
);
1018 return (active
> inactive
);
1021 unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup
*memcg
,
1025 int nid
= zone_to_nid(zone
);
1026 int zid
= zone_idx(zone
);
1027 struct mem_cgroup_per_zone
*mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
1029 return MEM_CGROUP_ZSTAT(mz
, lru
);
1032 struct zone_reclaim_stat
*mem_cgroup_get_reclaim_stat(struct mem_cgroup
*memcg
,
1035 int nid
= zone_to_nid(zone
);
1036 int zid
= zone_idx(zone
);
1037 struct mem_cgroup_per_zone
*mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
1039 return &mz
->reclaim_stat
;
1042 struct zone_reclaim_stat
*
1043 mem_cgroup_get_reclaim_stat_from_page(struct page
*page
)
1045 struct page_cgroup
*pc
;
1046 struct mem_cgroup_per_zone
*mz
;
1048 if (mem_cgroup_disabled())
1051 pc
= lookup_page_cgroup(page
);
1052 if (!PageCgroupUsed(pc
))
1054 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
1056 mz
= page_cgroup_zoneinfo(pc
);
1060 return &mz
->reclaim_stat
;
1063 unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan
,
1064 struct list_head
*dst
,
1065 unsigned long *scanned
, int order
,
1066 int mode
, struct zone
*z
,
1067 struct mem_cgroup
*mem_cont
,
1068 int active
, int file
)
1070 unsigned long nr_taken
= 0;
1074 struct list_head
*src
;
1075 struct page_cgroup
*pc
, *tmp
;
1076 int nid
= zone_to_nid(z
);
1077 int zid
= zone_idx(z
);
1078 struct mem_cgroup_per_zone
*mz
;
1079 int lru
= LRU_FILE
* file
+ active
;
1083 mz
= mem_cgroup_zoneinfo(mem_cont
, nid
, zid
);
1084 src
= &mz
->lists
[lru
];
1087 list_for_each_entry_safe_reverse(pc
, tmp
, src
, lru
) {
1088 if (scan
>= nr_to_scan
)
1092 if (unlikely(!PageCgroupUsed(pc
)))
1094 if (unlikely(!PageLRU(page
)))
1098 ret
= __isolate_lru_page(page
, mode
, file
);
1101 list_move(&page
->lru
, dst
);
1102 mem_cgroup_del_lru(page
);
1103 nr_taken
+= hpage_nr_pages(page
);
1106 /* we don't affect global LRU but rotate in our LRU */
1107 mem_cgroup_rotate_lru_list(page
, page_lru(page
));
1116 trace_mm_vmscan_memcg_isolate(0, nr_to_scan
, scan
, nr_taken
,
1122 #define mem_cgroup_from_res_counter(counter, member) \
1123 container_of(counter, struct mem_cgroup, member)
1126 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
1127 * @mem: the memory cgroup
1129 * Returns the maximum amount of memory @mem can be charged with, in
1132 static unsigned long long mem_cgroup_margin(struct mem_cgroup
*mem
)
1134 unsigned long long margin
;
1136 margin
= res_counter_margin(&mem
->res
);
1137 if (do_swap_account
)
1138 margin
= min(margin
, res_counter_margin(&mem
->memsw
));
1142 static unsigned int get_swappiness(struct mem_cgroup
*memcg
)
1144 struct cgroup
*cgrp
= memcg
->css
.cgroup
;
1145 unsigned int swappiness
;
1148 if (cgrp
->parent
== NULL
)
1149 return vm_swappiness
;
1151 spin_lock(&memcg
->reclaim_param_lock
);
1152 swappiness
= memcg
->swappiness
;
1153 spin_unlock(&memcg
->reclaim_param_lock
);
1158 static void mem_cgroup_start_move(struct mem_cgroup
*mem
)
1163 spin_lock(&mem
->pcp_counter_lock
);
1164 for_each_online_cpu(cpu
)
1165 per_cpu(mem
->stat
->count
[MEM_CGROUP_ON_MOVE
], cpu
) += 1;
1166 mem
->nocpu_base
.count
[MEM_CGROUP_ON_MOVE
] += 1;
1167 spin_unlock(&mem
->pcp_counter_lock
);
1173 static void mem_cgroup_end_move(struct mem_cgroup
*mem
)
1180 spin_lock(&mem
->pcp_counter_lock
);
1181 for_each_online_cpu(cpu
)
1182 per_cpu(mem
->stat
->count
[MEM_CGROUP_ON_MOVE
], cpu
) -= 1;
1183 mem
->nocpu_base
.count
[MEM_CGROUP_ON_MOVE
] -= 1;
1184 spin_unlock(&mem
->pcp_counter_lock
);
1188 * 2 routines for checking "mem" is under move_account() or not.
1190 * mem_cgroup_stealed() - checking a cgroup is mc.from or not. This is used
1191 * for avoiding race in accounting. If true,
1192 * pc->mem_cgroup may be overwritten.
1194 * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or
1195 * under hierarchy of moving cgroups. This is for
1196 * waiting at hith-memory prressure caused by "move".
1199 static bool mem_cgroup_stealed(struct mem_cgroup
*mem
)
1201 VM_BUG_ON(!rcu_read_lock_held());
1202 return this_cpu_read(mem
->stat
->count
[MEM_CGROUP_ON_MOVE
]) > 0;
1205 static bool mem_cgroup_under_move(struct mem_cgroup
*mem
)
1207 struct mem_cgroup
*from
;
1208 struct mem_cgroup
*to
;
1211 * Unlike task_move routines, we access mc.to, mc.from not under
1212 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1214 spin_lock(&mc
.lock
);
1219 if (from
== mem
|| to
== mem
1220 || (mem
->use_hierarchy
&& css_is_ancestor(&from
->css
, &mem
->css
))
1221 || (mem
->use_hierarchy
&& css_is_ancestor(&to
->css
, &mem
->css
)))
1224 spin_unlock(&mc
.lock
);
1228 static bool mem_cgroup_wait_acct_move(struct mem_cgroup
*mem
)
1230 if (mc
.moving_task
&& current
!= mc
.moving_task
) {
1231 if (mem_cgroup_under_move(mem
)) {
1233 prepare_to_wait(&mc
.waitq
, &wait
, TASK_INTERRUPTIBLE
);
1234 /* moving charge context might have finished. */
1237 finish_wait(&mc
.waitq
, &wait
);
1245 * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode.
1246 * @memcg: The memory cgroup that went over limit
1247 * @p: Task that is going to be killed
1249 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1252 void mem_cgroup_print_oom_info(struct mem_cgroup
*memcg
, struct task_struct
*p
)
1254 struct cgroup
*task_cgrp
;
1255 struct cgroup
*mem_cgrp
;
1257 * Need a buffer in BSS, can't rely on allocations. The code relies
1258 * on the assumption that OOM is serialized for memory controller.
1259 * If this assumption is broken, revisit this code.
1261 static char memcg_name
[PATH_MAX
];
1270 mem_cgrp
= memcg
->css
.cgroup
;
1271 task_cgrp
= task_cgroup(p
, mem_cgroup_subsys_id
);
1273 ret
= cgroup_path(task_cgrp
, memcg_name
, PATH_MAX
);
1276 * Unfortunately, we are unable to convert to a useful name
1277 * But we'll still print out the usage information
1284 printk(KERN_INFO
"Task in %s killed", memcg_name
);
1287 ret
= cgroup_path(mem_cgrp
, memcg_name
, PATH_MAX
);
1295 * Continues from above, so we don't need an KERN_ level
1297 printk(KERN_CONT
" as a result of limit of %s\n", memcg_name
);
1300 printk(KERN_INFO
"memory: usage %llukB, limit %llukB, failcnt %llu\n",
1301 res_counter_read_u64(&memcg
->res
, RES_USAGE
) >> 10,
1302 res_counter_read_u64(&memcg
->res
, RES_LIMIT
) >> 10,
1303 res_counter_read_u64(&memcg
->res
, RES_FAILCNT
));
1304 printk(KERN_INFO
"memory+swap: usage %llukB, limit %llukB, "
1306 res_counter_read_u64(&memcg
->memsw
, RES_USAGE
) >> 10,
1307 res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
) >> 10,
1308 res_counter_read_u64(&memcg
->memsw
, RES_FAILCNT
));
1312 * This function returns the number of memcg under hierarchy tree. Returns
1313 * 1(self count) if no children.
1315 static int mem_cgroup_count_children(struct mem_cgroup
*mem
)
1318 struct mem_cgroup
*iter
;
1320 for_each_mem_cgroup_tree(iter
, mem
)
1326 * Return the memory (and swap, if configured) limit for a memcg.
1328 u64
mem_cgroup_get_limit(struct mem_cgroup
*memcg
)
1333 limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
1334 limit
+= total_swap_pages
<< PAGE_SHIFT
;
1336 memsw
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
1338 * If memsw is finite and limits the amount of swap space available
1339 * to this memcg, return that limit.
1341 return min(limit
, memsw
);
1345 * Visit the first child (need not be the first child as per the ordering
1346 * of the cgroup list, since we track last_scanned_child) of @mem and use
1347 * that to reclaim free pages from.
1349 static struct mem_cgroup
*
1350 mem_cgroup_select_victim(struct mem_cgroup
*root_mem
)
1352 struct mem_cgroup
*ret
= NULL
;
1353 struct cgroup_subsys_state
*css
;
1356 if (!root_mem
->use_hierarchy
) {
1357 css_get(&root_mem
->css
);
1363 nextid
= root_mem
->last_scanned_child
+ 1;
1364 css
= css_get_next(&mem_cgroup_subsys
, nextid
, &root_mem
->css
,
1366 if (css
&& css_tryget(css
))
1367 ret
= container_of(css
, struct mem_cgroup
, css
);
1370 /* Updates scanning parameter */
1371 spin_lock(&root_mem
->reclaim_param_lock
);
1373 /* this means start scan from ID:1 */
1374 root_mem
->last_scanned_child
= 0;
1376 root_mem
->last_scanned_child
= found
;
1377 spin_unlock(&root_mem
->reclaim_param_lock
);
1384 * Scan the hierarchy if needed to reclaim memory. We remember the last child
1385 * we reclaimed from, so that we don't end up penalizing one child extensively
1386 * based on its position in the children list.
1388 * root_mem is the original ancestor that we've been reclaim from.
1390 * We give up and return to the caller when we visit root_mem twice.
1391 * (other groups can be removed while we're walking....)
1393 * If shrink==true, for avoiding to free too much, this returns immedieately.
1395 static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup
*root_mem
,
1398 unsigned long reclaim_options
)
1400 struct mem_cgroup
*victim
;
1403 bool noswap
= reclaim_options
& MEM_CGROUP_RECLAIM_NOSWAP
;
1404 bool shrink
= reclaim_options
& MEM_CGROUP_RECLAIM_SHRINK
;
1405 bool check_soft
= reclaim_options
& MEM_CGROUP_RECLAIM_SOFT
;
1406 unsigned long excess
;
1408 excess
= res_counter_soft_limit_excess(&root_mem
->res
) >> PAGE_SHIFT
;
1410 /* If memsw_is_minimum==1, swap-out is of-no-use. */
1411 if (root_mem
->memsw_is_minimum
)
1415 victim
= mem_cgroup_select_victim(root_mem
);
1416 if (victim
== root_mem
) {
1419 drain_all_stock_async();
1422 * If we have not been able to reclaim
1423 * anything, it might because there are
1424 * no reclaimable pages under this hierarchy
1426 if (!check_soft
|| !total
) {
1427 css_put(&victim
->css
);
1431 * We want to do more targetted reclaim.
1432 * excess >> 2 is not to excessive so as to
1433 * reclaim too much, nor too less that we keep
1434 * coming back to reclaim from this cgroup
1436 if (total
>= (excess
>> 2) ||
1437 (loop
> MEM_CGROUP_MAX_RECLAIM_LOOPS
)) {
1438 css_put(&victim
->css
);
1443 if (!mem_cgroup_local_usage(victim
)) {
1444 /* this cgroup's local usage == 0 */
1445 css_put(&victim
->css
);
1448 /* we use swappiness of local cgroup */
1450 ret
= mem_cgroup_shrink_node_zone(victim
, gfp_mask
,
1451 noswap
, get_swappiness(victim
), zone
);
1453 ret
= try_to_free_mem_cgroup_pages(victim
, gfp_mask
,
1454 noswap
, get_swappiness(victim
));
1455 css_put(&victim
->css
);
1457 * At shrinking usage, we can't check we should stop here or
1458 * reclaim more. It's depends on callers. last_scanned_child
1459 * will work enough for keeping fairness under tree.
1465 if (!res_counter_soft_limit_excess(&root_mem
->res
))
1467 } else if (mem_cgroup_margin(root_mem
))
1474 * Check OOM-Killer is already running under our hierarchy.
1475 * If someone is running, return false.
1477 static bool mem_cgroup_oom_lock(struct mem_cgroup
*mem
)
1479 int x
, lock_count
= 0;
1480 struct mem_cgroup
*iter
;
1482 for_each_mem_cgroup_tree(iter
, mem
) {
1483 x
= atomic_inc_return(&iter
->oom_lock
);
1484 lock_count
= max(x
, lock_count
);
1487 if (lock_count
== 1)
1492 static int mem_cgroup_oom_unlock(struct mem_cgroup
*mem
)
1494 struct mem_cgroup
*iter
;
1497 * When a new child is created while the hierarchy is under oom,
1498 * mem_cgroup_oom_lock() may not be called. We have to use
1499 * atomic_add_unless() here.
1501 for_each_mem_cgroup_tree(iter
, mem
)
1502 atomic_add_unless(&iter
->oom_lock
, -1, 0);
1507 static DEFINE_MUTEX(memcg_oom_mutex
);
1508 static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq
);
1510 struct oom_wait_info
{
1511 struct mem_cgroup
*mem
;
1515 static int memcg_oom_wake_function(wait_queue_t
*wait
,
1516 unsigned mode
, int sync
, void *arg
)
1518 struct mem_cgroup
*wake_mem
= (struct mem_cgroup
*)arg
;
1519 struct oom_wait_info
*oom_wait_info
;
1521 oom_wait_info
= container_of(wait
, struct oom_wait_info
, wait
);
1523 if (oom_wait_info
->mem
== wake_mem
)
1525 /* if no hierarchy, no match */
1526 if (!oom_wait_info
->mem
->use_hierarchy
|| !wake_mem
->use_hierarchy
)
1529 * Both of oom_wait_info->mem and wake_mem are stable under us.
1530 * Then we can use css_is_ancestor without taking care of RCU.
1532 if (!css_is_ancestor(&oom_wait_info
->mem
->css
, &wake_mem
->css
) &&
1533 !css_is_ancestor(&wake_mem
->css
, &oom_wait_info
->mem
->css
))
1537 return autoremove_wake_function(wait
, mode
, sync
, arg
);
1540 static void memcg_wakeup_oom(struct mem_cgroup
*mem
)
1542 /* for filtering, pass "mem" as argument. */
1543 __wake_up(&memcg_oom_waitq
, TASK_NORMAL
, 0, mem
);
1546 static void memcg_oom_recover(struct mem_cgroup
*mem
)
1548 if (mem
&& atomic_read(&mem
->oom_lock
))
1549 memcg_wakeup_oom(mem
);
1553 * try to call OOM killer. returns false if we should exit memory-reclaim loop.
1555 bool mem_cgroup_handle_oom(struct mem_cgroup
*mem
, gfp_t mask
)
1557 struct oom_wait_info owait
;
1558 bool locked
, need_to_kill
;
1561 owait
.wait
.flags
= 0;
1562 owait
.wait
.func
= memcg_oom_wake_function
;
1563 owait
.wait
.private = current
;
1564 INIT_LIST_HEAD(&owait
.wait
.task_list
);
1565 need_to_kill
= true;
1566 /* At first, try to OOM lock hierarchy under mem.*/
1567 mutex_lock(&memcg_oom_mutex
);
1568 locked
= mem_cgroup_oom_lock(mem
);
1570 * Even if signal_pending(), we can't quit charge() loop without
1571 * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
1572 * under OOM is always welcomed, use TASK_KILLABLE here.
1574 prepare_to_wait(&memcg_oom_waitq
, &owait
.wait
, TASK_KILLABLE
);
1575 if (!locked
|| mem
->oom_kill_disable
)
1576 need_to_kill
= false;
1578 mem_cgroup_oom_notify(mem
);
1579 mutex_unlock(&memcg_oom_mutex
);
1582 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1583 mem_cgroup_out_of_memory(mem
, mask
);
1586 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1588 mutex_lock(&memcg_oom_mutex
);
1589 mem_cgroup_oom_unlock(mem
);
1590 memcg_wakeup_oom(mem
);
1591 mutex_unlock(&memcg_oom_mutex
);
1593 if (test_thread_flag(TIF_MEMDIE
) || fatal_signal_pending(current
))
1595 /* Give chance to dying process */
1596 schedule_timeout(1);
1601 * Currently used to update mapped file statistics, but the routine can be
1602 * generalized to update other statistics as well.
1604 * Notes: Race condition
1606 * We usually use page_cgroup_lock() for accessing page_cgroup member but
1607 * it tends to be costly. But considering some conditions, we doesn't need
1608 * to do so _always_.
1610 * Considering "charge", lock_page_cgroup() is not required because all
1611 * file-stat operations happen after a page is attached to radix-tree. There
1612 * are no race with "charge".
1614 * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup
1615 * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even
1616 * if there are race with "uncharge". Statistics itself is properly handled
1619 * Considering "move", this is an only case we see a race. To make the race
1620 * small, we check MEM_CGROUP_ON_MOVE percpu value and detect there are
1621 * possibility of race condition. If there is, we take a lock.
1624 void mem_cgroup_update_page_stat(struct page
*page
,
1625 enum mem_cgroup_page_stat_item idx
, int val
)
1627 struct mem_cgroup
*mem
;
1628 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
1629 bool need_unlock
= false;
1630 unsigned long uninitialized_var(flags
);
1636 mem
= pc
->mem_cgroup
;
1637 if (unlikely(!mem
|| !PageCgroupUsed(pc
)))
1639 /* pc->mem_cgroup is unstable ? */
1640 if (unlikely(mem_cgroup_stealed(mem
)) || PageTransHuge(page
)) {
1641 /* take a lock against to access pc->mem_cgroup */
1642 move_lock_page_cgroup(pc
, &flags
);
1644 mem
= pc
->mem_cgroup
;
1645 if (!mem
|| !PageCgroupUsed(pc
))
1650 case MEMCG_NR_FILE_MAPPED
:
1652 SetPageCgroupFileMapped(pc
);
1653 else if (!page_mapped(page
))
1654 ClearPageCgroupFileMapped(pc
);
1655 idx
= MEM_CGROUP_STAT_FILE_MAPPED
;
1661 this_cpu_add(mem
->stat
->count
[idx
], val
);
1664 if (unlikely(need_unlock
))
1665 move_unlock_page_cgroup(pc
, &flags
);
1669 EXPORT_SYMBOL(mem_cgroup_update_page_stat
);
1672 * size of first charge trial. "32" comes from vmscan.c's magic value.
1673 * TODO: maybe necessary to use big numbers in big irons.
1675 #define CHARGE_SIZE (32 * PAGE_SIZE)
1676 struct memcg_stock_pcp
{
1677 struct mem_cgroup
*cached
; /* this never be root cgroup */
1679 struct work_struct work
;
1681 static DEFINE_PER_CPU(struct memcg_stock_pcp
, memcg_stock
);
1682 static atomic_t memcg_drain_count
;
1685 * Try to consume stocked charge on this cpu. If success, PAGE_SIZE is consumed
1686 * from local stock and true is returned. If the stock is 0 or charges from a
1687 * cgroup which is not current target, returns false. This stock will be
1690 static bool consume_stock(struct mem_cgroup
*mem
)
1692 struct memcg_stock_pcp
*stock
;
1695 stock
= &get_cpu_var(memcg_stock
);
1696 if (mem
== stock
->cached
&& stock
->charge
)
1697 stock
->charge
-= PAGE_SIZE
;
1698 else /* need to call res_counter_charge */
1700 put_cpu_var(memcg_stock
);
1705 * Returns stocks cached in percpu to res_counter and reset cached information.
1707 static void drain_stock(struct memcg_stock_pcp
*stock
)
1709 struct mem_cgroup
*old
= stock
->cached
;
1711 if (stock
->charge
) {
1712 res_counter_uncharge(&old
->res
, stock
->charge
);
1713 if (do_swap_account
)
1714 res_counter_uncharge(&old
->memsw
, stock
->charge
);
1716 stock
->cached
= NULL
;
1721 * This must be called under preempt disabled or must be called by
1722 * a thread which is pinned to local cpu.
1724 static void drain_local_stock(struct work_struct
*dummy
)
1726 struct memcg_stock_pcp
*stock
= &__get_cpu_var(memcg_stock
);
1731 * Cache charges(val) which is from res_counter, to local per_cpu area.
1732 * This will be consumed by consume_stock() function, later.
1734 static void refill_stock(struct mem_cgroup
*mem
, int val
)
1736 struct memcg_stock_pcp
*stock
= &get_cpu_var(memcg_stock
);
1738 if (stock
->cached
!= mem
) { /* reset if necessary */
1740 stock
->cached
= mem
;
1742 stock
->charge
+= val
;
1743 put_cpu_var(memcg_stock
);
1747 * Tries to drain stocked charges in other cpus. This function is asynchronous
1748 * and just put a work per cpu for draining localy on each cpu. Caller can
1749 * expects some charges will be back to res_counter later but cannot wait for
1752 static void drain_all_stock_async(void)
1755 /* This function is for scheduling "drain" in asynchronous way.
1756 * The result of "drain" is not directly handled by callers. Then,
1757 * if someone is calling drain, we don't have to call drain more.
1758 * Anyway, WORK_STRUCT_PENDING check in queue_work_on() will catch if
1759 * there is a race. We just do loose check here.
1761 if (atomic_read(&memcg_drain_count
))
1763 /* Notify other cpus that system-wide "drain" is running */
1764 atomic_inc(&memcg_drain_count
);
1766 for_each_online_cpu(cpu
) {
1767 struct memcg_stock_pcp
*stock
= &per_cpu(memcg_stock
, cpu
);
1768 schedule_work_on(cpu
, &stock
->work
);
1771 atomic_dec(&memcg_drain_count
);
1772 /* We don't wait for flush_work */
1775 /* This is a synchronous drain interface. */
1776 static void drain_all_stock_sync(void)
1778 /* called when force_empty is called */
1779 atomic_inc(&memcg_drain_count
);
1780 schedule_on_each_cpu(drain_local_stock
);
1781 atomic_dec(&memcg_drain_count
);
1785 * This function drains percpu counter value from DEAD cpu and
1786 * move it to local cpu. Note that this function can be preempted.
1788 static void mem_cgroup_drain_pcp_counter(struct mem_cgroup
*mem
, int cpu
)
1792 spin_lock(&mem
->pcp_counter_lock
);
1793 for (i
= 0; i
< MEM_CGROUP_STAT_DATA
; i
++) {
1794 s64 x
= per_cpu(mem
->stat
->count
[i
], cpu
);
1796 per_cpu(mem
->stat
->count
[i
], cpu
) = 0;
1797 mem
->nocpu_base
.count
[i
] += x
;
1799 /* need to clear ON_MOVE value, works as a kind of lock. */
1800 per_cpu(mem
->stat
->count
[MEM_CGROUP_ON_MOVE
], cpu
) = 0;
1801 spin_unlock(&mem
->pcp_counter_lock
);
1804 static void synchronize_mem_cgroup_on_move(struct mem_cgroup
*mem
, int cpu
)
1806 int idx
= MEM_CGROUP_ON_MOVE
;
1808 spin_lock(&mem
->pcp_counter_lock
);
1809 per_cpu(mem
->stat
->count
[idx
], cpu
) = mem
->nocpu_base
.count
[idx
];
1810 spin_unlock(&mem
->pcp_counter_lock
);
1813 static int __cpuinit
memcg_cpu_hotplug_callback(struct notifier_block
*nb
,
1814 unsigned long action
,
1817 int cpu
= (unsigned long)hcpu
;
1818 struct memcg_stock_pcp
*stock
;
1819 struct mem_cgroup
*iter
;
1821 if ((action
== CPU_ONLINE
)) {
1822 for_each_mem_cgroup_all(iter
)
1823 synchronize_mem_cgroup_on_move(iter
, cpu
);
1827 if ((action
!= CPU_DEAD
) || action
!= CPU_DEAD_FROZEN
)
1830 for_each_mem_cgroup_all(iter
)
1831 mem_cgroup_drain_pcp_counter(iter
, cpu
);
1833 stock
= &per_cpu(memcg_stock
, cpu
);
1839 /* See __mem_cgroup_try_charge() for details */
1841 CHARGE_OK
, /* success */
1842 CHARGE_RETRY
, /* need to retry but retry is not bad */
1843 CHARGE_NOMEM
, /* we can't do more. return -ENOMEM */
1844 CHARGE_WOULDBLOCK
, /* GFP_WAIT wasn't set and no enough res. */
1845 CHARGE_OOM_DIE
, /* the current is killed because of OOM */
1848 static int __mem_cgroup_do_charge(struct mem_cgroup
*mem
, gfp_t gfp_mask
,
1849 int csize
, bool oom_check
)
1851 struct mem_cgroup
*mem_over_limit
;
1852 struct res_counter
*fail_res
;
1853 unsigned long flags
= 0;
1856 ret
= res_counter_charge(&mem
->res
, csize
, &fail_res
);
1859 if (!do_swap_account
)
1861 ret
= res_counter_charge(&mem
->memsw
, csize
, &fail_res
);
1865 res_counter_uncharge(&mem
->res
, csize
);
1866 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
, memsw
);
1867 flags
|= MEM_CGROUP_RECLAIM_NOSWAP
;
1869 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
, res
);
1871 * csize can be either a huge page (HPAGE_SIZE), a batch of
1872 * regular pages (CHARGE_SIZE), or a single regular page
1875 * Never reclaim on behalf of optional batching, retry with a
1876 * single page instead.
1878 if (csize
== CHARGE_SIZE
)
1879 return CHARGE_RETRY
;
1881 if (!(gfp_mask
& __GFP_WAIT
))
1882 return CHARGE_WOULDBLOCK
;
1884 ret
= mem_cgroup_hierarchical_reclaim(mem_over_limit
, NULL
,
1886 if (mem_cgroup_margin(mem_over_limit
) >= csize
)
1887 return CHARGE_RETRY
;
1889 * Even though the limit is exceeded at this point, reclaim
1890 * may have been able to free some pages. Retry the charge
1891 * before killing the task.
1893 * Only for regular pages, though: huge pages are rather
1894 * unlikely to succeed so close to the limit, and we fall back
1895 * to regular pages anyway in case of failure.
1897 if (csize
== PAGE_SIZE
&& ret
)
1898 return CHARGE_RETRY
;
1901 * At task move, charge accounts can be doubly counted. So, it's
1902 * better to wait until the end of task_move if something is going on.
1904 if (mem_cgroup_wait_acct_move(mem_over_limit
))
1905 return CHARGE_RETRY
;
1907 /* If we don't need to call oom-killer at el, return immediately */
1909 return CHARGE_NOMEM
;
1911 if (!mem_cgroup_handle_oom(mem_over_limit
, gfp_mask
))
1912 return CHARGE_OOM_DIE
;
1914 return CHARGE_RETRY
;
1918 * Unlike exported interface, "oom" parameter is added. if oom==true,
1919 * oom-killer can be invoked.
1921 static int __mem_cgroup_try_charge(struct mm_struct
*mm
,
1923 struct mem_cgroup
**memcg
, bool oom
,
1926 int nr_oom_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
1927 struct mem_cgroup
*mem
= NULL
;
1929 int csize
= max(CHARGE_SIZE
, (unsigned long) page_size
);
1932 * Unlike gloval-vm's OOM-kill, we're not in memory shortage
1933 * in system level. So, allow to go ahead dying process in addition to
1936 if (unlikely(test_thread_flag(TIF_MEMDIE
)
1937 || fatal_signal_pending(current
)))
1941 * We always charge the cgroup the mm_struct belongs to.
1942 * The mm_struct's mem_cgroup changes on task migration if the
1943 * thread group leader migrates. It's possible that mm is not
1944 * set, if so charge the init_mm (happens for pagecache usage).
1949 if (*memcg
) { /* css should be a valid one */
1951 VM_BUG_ON(css_is_removed(&mem
->css
));
1952 if (mem_cgroup_is_root(mem
))
1954 if (page_size
== PAGE_SIZE
&& consume_stock(mem
))
1958 struct task_struct
*p
;
1961 p
= rcu_dereference(mm
->owner
);
1963 * Because we don't have task_lock(), "p" can exit.
1964 * In that case, "mem" can point to root or p can be NULL with
1965 * race with swapoff. Then, we have small risk of mis-accouning.
1966 * But such kind of mis-account by race always happens because
1967 * we don't have cgroup_mutex(). It's overkill and we allo that
1969 * (*) swapoff at el will charge against mm-struct not against
1970 * task-struct. So, mm->owner can be NULL.
1972 mem
= mem_cgroup_from_task(p
);
1973 if (!mem
|| mem_cgroup_is_root(mem
)) {
1977 if (page_size
== PAGE_SIZE
&& consume_stock(mem
)) {
1979 * It seems dagerous to access memcg without css_get().
1980 * But considering how consume_stok works, it's not
1981 * necessary. If consume_stock success, some charges
1982 * from this memcg are cached on this cpu. So, we
1983 * don't need to call css_get()/css_tryget() before
1984 * calling consume_stock().
1989 /* after here, we may be blocked. we need to get refcnt */
1990 if (!css_tryget(&mem
->css
)) {
2000 /* If killed, bypass charge */
2001 if (fatal_signal_pending(current
)) {
2007 if (oom
&& !nr_oom_retries
) {
2009 nr_oom_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
2012 ret
= __mem_cgroup_do_charge(mem
, gfp_mask
, csize
, oom_check
);
2017 case CHARGE_RETRY
: /* not in OOM situation but retry */
2022 case CHARGE_WOULDBLOCK
: /* !__GFP_WAIT */
2025 case CHARGE_NOMEM
: /* OOM routine works */
2030 /* If oom, we never return -ENOMEM */
2033 case CHARGE_OOM_DIE
: /* Killed by OOM Killer */
2037 } while (ret
!= CHARGE_OK
);
2039 if (csize
> page_size
)
2040 refill_stock(mem
, csize
- page_size
);
2054 * Somemtimes we have to undo a charge we got by try_charge().
2055 * This function is for that and do uncharge, put css's refcnt.
2056 * gotten by try_charge().
2058 static void __mem_cgroup_cancel_charge(struct mem_cgroup
*mem
,
2059 unsigned long count
)
2061 if (!mem_cgroup_is_root(mem
)) {
2062 res_counter_uncharge(&mem
->res
, PAGE_SIZE
* count
);
2063 if (do_swap_account
)
2064 res_counter_uncharge(&mem
->memsw
, PAGE_SIZE
* count
);
2068 static void mem_cgroup_cancel_charge(struct mem_cgroup
*mem
,
2071 __mem_cgroup_cancel_charge(mem
, page_size
>> PAGE_SHIFT
);
2075 * A helper function to get mem_cgroup from ID. must be called under
2076 * rcu_read_lock(). The caller must check css_is_removed() or some if
2077 * it's concern. (dropping refcnt from swap can be called against removed
2080 static struct mem_cgroup
*mem_cgroup_lookup(unsigned short id
)
2082 struct cgroup_subsys_state
*css
;
2084 /* ID 0 is unused ID */
2087 css
= css_lookup(&mem_cgroup_subsys
, id
);
2090 return container_of(css
, struct mem_cgroup
, css
);
2093 struct mem_cgroup
*try_get_mem_cgroup_from_page(struct page
*page
)
2095 struct mem_cgroup
*mem
= NULL
;
2096 struct page_cgroup
*pc
;
2100 VM_BUG_ON(!PageLocked(page
));
2102 pc
= lookup_page_cgroup(page
);
2103 lock_page_cgroup(pc
);
2104 if (PageCgroupUsed(pc
)) {
2105 mem
= pc
->mem_cgroup
;
2106 if (mem
&& !css_tryget(&mem
->css
))
2108 } else if (PageSwapCache(page
)) {
2109 ent
.val
= page_private(page
);
2110 id
= lookup_swap_cgroup(ent
);
2112 mem
= mem_cgroup_lookup(id
);
2113 if (mem
&& !css_tryget(&mem
->css
))
2117 unlock_page_cgroup(pc
);
2121 static void __mem_cgroup_commit_charge(struct mem_cgroup
*mem
,
2122 struct page_cgroup
*pc
,
2123 enum charge_type ctype
,
2126 int nr_pages
= page_size
>> PAGE_SHIFT
;
2128 /* try_charge() can return NULL to *memcg, taking care of it. */
2132 lock_page_cgroup(pc
);
2133 if (unlikely(PageCgroupUsed(pc
))) {
2134 unlock_page_cgroup(pc
);
2135 mem_cgroup_cancel_charge(mem
, page_size
);
2139 * we don't need page_cgroup_lock about tail pages, becase they are not
2140 * accessed by any other context at this point.
2142 pc
->mem_cgroup
= mem
;
2144 * We access a page_cgroup asynchronously without lock_page_cgroup().
2145 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
2146 * is accessed after testing USED bit. To make pc->mem_cgroup visible
2147 * before USED bit, we need memory barrier here.
2148 * See mem_cgroup_add_lru_list(), etc.
2152 case MEM_CGROUP_CHARGE_TYPE_CACHE
:
2153 case MEM_CGROUP_CHARGE_TYPE_SHMEM
:
2154 SetPageCgroupCache(pc
);
2155 SetPageCgroupUsed(pc
);
2157 case MEM_CGROUP_CHARGE_TYPE_MAPPED
:
2158 ClearPageCgroupCache(pc
);
2159 SetPageCgroupUsed(pc
);
2165 mem_cgroup_charge_statistics(mem
, PageCgroupCache(pc
), nr_pages
);
2166 unlock_page_cgroup(pc
);
2168 * "charge_statistics" updated event counter. Then, check it.
2169 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
2170 * if they exceeds softlimit.
2172 memcg_check_events(mem
, pc
->page
);
2175 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2177 #define PCGF_NOCOPY_AT_SPLIT ((1 << PCG_LOCK) | (1 << PCG_MOVE_LOCK) |\
2178 (1 << PCG_ACCT_LRU) | (1 << PCG_MIGRATION))
2180 * Because tail pages are not marked as "used", set it. We're under
2181 * zone->lru_lock, 'splitting on pmd' and compund_lock.
2183 void mem_cgroup_split_huge_fixup(struct page
*head
, struct page
*tail
)
2185 struct page_cgroup
*head_pc
= lookup_page_cgroup(head
);
2186 struct page_cgroup
*tail_pc
= lookup_page_cgroup(tail
);
2187 unsigned long flags
;
2189 if (mem_cgroup_disabled())
2192 * We have no races with charge/uncharge but will have races with
2193 * page state accounting.
2195 move_lock_page_cgroup(head_pc
, &flags
);
2197 tail_pc
->mem_cgroup
= head_pc
->mem_cgroup
;
2198 smp_wmb(); /* see __commit_charge() */
2199 if (PageCgroupAcctLRU(head_pc
)) {
2201 struct mem_cgroup_per_zone
*mz
;
2204 * LRU flags cannot be copied because we need to add tail
2205 *.page to LRU by generic call and our hook will be called.
2206 * We hold lru_lock, then, reduce counter directly.
2208 lru
= page_lru(head
);
2209 mz
= page_cgroup_zoneinfo(head_pc
);
2210 MEM_CGROUP_ZSTAT(mz
, lru
) -= 1;
2212 tail_pc
->flags
= head_pc
->flags
& ~PCGF_NOCOPY_AT_SPLIT
;
2213 move_unlock_page_cgroup(head_pc
, &flags
);
2218 * __mem_cgroup_move_account - move account of the page
2219 * @pc: page_cgroup of the page.
2220 * @from: mem_cgroup which the page is moved from.
2221 * @to: mem_cgroup which the page is moved to. @from != @to.
2222 * @uncharge: whether we should call uncharge and css_put against @from.
2224 * The caller must confirm following.
2225 * - page is not on LRU (isolate_page() is useful.)
2226 * - the pc is locked, used, and ->mem_cgroup points to @from.
2228 * This function doesn't do "charge" nor css_get to new cgroup. It should be
2229 * done by a caller(__mem_cgroup_try_charge would be usefull). If @uncharge is
2230 * true, this function does "uncharge" from old cgroup, but it doesn't if
2231 * @uncharge is false, so a caller should do "uncharge".
2234 static void __mem_cgroup_move_account(struct page_cgroup
*pc
,
2235 struct mem_cgroup
*from
, struct mem_cgroup
*to
, bool uncharge
,
2238 int nr_pages
= charge_size
>> PAGE_SHIFT
;
2240 VM_BUG_ON(from
== to
);
2241 VM_BUG_ON(PageLRU(pc
->page
));
2242 VM_BUG_ON(!page_is_cgroup_locked(pc
));
2243 VM_BUG_ON(!PageCgroupUsed(pc
));
2244 VM_BUG_ON(pc
->mem_cgroup
!= from
);
2246 if (PageCgroupFileMapped(pc
)) {
2247 /* Update mapped_file data for mem_cgroup */
2249 __this_cpu_dec(from
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
2250 __this_cpu_inc(to
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
2253 mem_cgroup_charge_statistics(from
, PageCgroupCache(pc
), -nr_pages
);
2255 /* This is not "cancel", but cancel_charge does all we need. */
2256 mem_cgroup_cancel_charge(from
, charge_size
);
2258 /* caller should have done css_get */
2259 pc
->mem_cgroup
= to
;
2260 mem_cgroup_charge_statistics(to
, PageCgroupCache(pc
), nr_pages
);
2262 * We charges against "to" which may not have any tasks. Then, "to"
2263 * can be under rmdir(). But in current implementation, caller of
2264 * this function is just force_empty() and move charge, so it's
2265 * garanteed that "to" is never removed. So, we don't check rmdir
2271 * check whether the @pc is valid for moving account and call
2272 * __mem_cgroup_move_account()
2274 static int mem_cgroup_move_account(struct page_cgroup
*pc
,
2275 struct mem_cgroup
*from
, struct mem_cgroup
*to
,
2276 bool uncharge
, int charge_size
)
2279 unsigned long flags
;
2281 * The page is isolated from LRU. So, collapse function
2282 * will not handle this page. But page splitting can happen.
2283 * Do this check under compound_page_lock(). The caller should
2286 if ((charge_size
> PAGE_SIZE
) && !PageTransHuge(pc
->page
))
2289 lock_page_cgroup(pc
);
2290 if (PageCgroupUsed(pc
) && pc
->mem_cgroup
== from
) {
2291 move_lock_page_cgroup(pc
, &flags
);
2292 __mem_cgroup_move_account(pc
, from
, to
, uncharge
, charge_size
);
2293 move_unlock_page_cgroup(pc
, &flags
);
2296 unlock_page_cgroup(pc
);
2300 memcg_check_events(to
, pc
->page
);
2301 memcg_check_events(from
, pc
->page
);
2306 * move charges to its parent.
2309 static int mem_cgroup_move_parent(struct page_cgroup
*pc
,
2310 struct mem_cgroup
*child
,
2313 struct page
*page
= pc
->page
;
2314 struct cgroup
*cg
= child
->css
.cgroup
;
2315 struct cgroup
*pcg
= cg
->parent
;
2316 struct mem_cgroup
*parent
;
2317 int page_size
= PAGE_SIZE
;
2318 unsigned long flags
;
2326 if (!get_page_unless_zero(page
))
2328 if (isolate_lru_page(page
))
2331 if (PageTransHuge(page
))
2332 page_size
= HPAGE_SIZE
;
2334 parent
= mem_cgroup_from_cont(pcg
);
2335 ret
= __mem_cgroup_try_charge(NULL
, gfp_mask
,
2336 &parent
, false, page_size
);
2340 if (page_size
> PAGE_SIZE
)
2341 flags
= compound_lock_irqsave(page
);
2343 ret
= mem_cgroup_move_account(pc
, child
, parent
, true, page_size
);
2345 mem_cgroup_cancel_charge(parent
, page_size
);
2347 if (page_size
> PAGE_SIZE
)
2348 compound_unlock_irqrestore(page
, flags
);
2350 putback_lru_page(page
);
2358 * Charge the memory controller for page usage.
2360 * 0 if the charge was successful
2361 * < 0 if the cgroup is over its limit
2363 static int mem_cgroup_charge_common(struct page
*page
, struct mm_struct
*mm
,
2364 gfp_t gfp_mask
, enum charge_type ctype
)
2366 struct mem_cgroup
*mem
= NULL
;
2367 int page_size
= PAGE_SIZE
;
2368 struct page_cgroup
*pc
;
2372 if (PageTransHuge(page
)) {
2373 page_size
<<= compound_order(page
);
2374 VM_BUG_ON(!PageTransHuge(page
));
2376 * Never OOM-kill a process for a huge page. The
2377 * fault handler will fall back to regular pages.
2382 pc
= lookup_page_cgroup(page
);
2383 /* can happen at boot */
2388 ret
= __mem_cgroup_try_charge(mm
, gfp_mask
, &mem
, oom
, page_size
);
2392 __mem_cgroup_commit_charge(mem
, pc
, ctype
, page_size
);
2396 int mem_cgroup_newpage_charge(struct page
*page
,
2397 struct mm_struct
*mm
, gfp_t gfp_mask
)
2399 if (mem_cgroup_disabled())
2402 * If already mapped, we don't have to account.
2403 * If page cache, page->mapping has address_space.
2404 * But page->mapping may have out-of-use anon_vma pointer,
2405 * detecit it by PageAnon() check. newly-mapped-anon's page->mapping
2408 if (page_mapped(page
) || (page
->mapping
&& !PageAnon(page
)))
2412 return mem_cgroup_charge_common(page
, mm
, gfp_mask
,
2413 MEM_CGROUP_CHARGE_TYPE_MAPPED
);
2417 __mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*ptr
,
2418 enum charge_type ctype
);
2420 int mem_cgroup_cache_charge(struct page
*page
, struct mm_struct
*mm
,
2425 if (mem_cgroup_disabled())
2427 if (PageCompound(page
))
2430 * Corner case handling. This is called from add_to_page_cache()
2431 * in usual. But some FS (shmem) precharges this page before calling it
2432 * and call add_to_page_cache() with GFP_NOWAIT.
2434 * For GFP_NOWAIT case, the page may be pre-charged before calling
2435 * add_to_page_cache(). (See shmem.c) check it here and avoid to call
2436 * charge twice. (It works but has to pay a bit larger cost.)
2437 * And when the page is SwapCache, it should take swap information
2438 * into account. This is under lock_page() now.
2440 if (!(gfp_mask
& __GFP_WAIT
)) {
2441 struct page_cgroup
*pc
;
2443 pc
= lookup_page_cgroup(page
);
2446 lock_page_cgroup(pc
);
2447 if (PageCgroupUsed(pc
)) {
2448 unlock_page_cgroup(pc
);
2451 unlock_page_cgroup(pc
);
2457 if (page_is_file_cache(page
))
2458 return mem_cgroup_charge_common(page
, mm
, gfp_mask
,
2459 MEM_CGROUP_CHARGE_TYPE_CACHE
);
2462 if (PageSwapCache(page
)) {
2463 struct mem_cgroup
*mem
;
2465 ret
= mem_cgroup_try_charge_swapin(mm
, page
, gfp_mask
, &mem
);
2467 __mem_cgroup_commit_charge_swapin(page
, mem
,
2468 MEM_CGROUP_CHARGE_TYPE_SHMEM
);
2470 ret
= mem_cgroup_charge_common(page
, mm
, gfp_mask
,
2471 MEM_CGROUP_CHARGE_TYPE_SHMEM
);
2477 * While swap-in, try_charge -> commit or cancel, the page is locked.
2478 * And when try_charge() successfully returns, one refcnt to memcg without
2479 * struct page_cgroup is acquired. This refcnt will be consumed by
2480 * "commit()" or removed by "cancel()"
2482 int mem_cgroup_try_charge_swapin(struct mm_struct
*mm
,
2484 gfp_t mask
, struct mem_cgroup
**ptr
)
2486 struct mem_cgroup
*mem
;
2491 if (mem_cgroup_disabled())
2494 if (!do_swap_account
)
2497 * A racing thread's fault, or swapoff, may have already updated
2498 * the pte, and even removed page from swap cache: in those cases
2499 * do_swap_page()'s pte_same() test will fail; but there's also a
2500 * KSM case which does need to charge the page.
2502 if (!PageSwapCache(page
))
2504 mem
= try_get_mem_cgroup_from_page(page
);
2508 ret
= __mem_cgroup_try_charge(NULL
, mask
, ptr
, true, PAGE_SIZE
);
2514 return __mem_cgroup_try_charge(mm
, mask
, ptr
, true, PAGE_SIZE
);
2518 __mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*ptr
,
2519 enum charge_type ctype
)
2521 struct page_cgroup
*pc
;
2523 if (mem_cgroup_disabled())
2527 cgroup_exclude_rmdir(&ptr
->css
);
2528 pc
= lookup_page_cgroup(page
);
2529 mem_cgroup_lru_del_before_commit_swapcache(page
);
2530 __mem_cgroup_commit_charge(ptr
, pc
, ctype
, PAGE_SIZE
);
2531 mem_cgroup_lru_add_after_commit_swapcache(page
);
2533 * Now swap is on-memory. This means this page may be
2534 * counted both as mem and swap....double count.
2535 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
2536 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
2537 * may call delete_from_swap_cache() before reach here.
2539 if (do_swap_account
&& PageSwapCache(page
)) {
2540 swp_entry_t ent
= {.val
= page_private(page
)};
2542 struct mem_cgroup
*memcg
;
2544 id
= swap_cgroup_record(ent
, 0);
2546 memcg
= mem_cgroup_lookup(id
);
2549 * This recorded memcg can be obsolete one. So, avoid
2550 * calling css_tryget
2552 if (!mem_cgroup_is_root(memcg
))
2553 res_counter_uncharge(&memcg
->memsw
, PAGE_SIZE
);
2554 mem_cgroup_swap_statistics(memcg
, false);
2555 mem_cgroup_put(memcg
);
2560 * At swapin, we may charge account against cgroup which has no tasks.
2561 * So, rmdir()->pre_destroy() can be called while we do this charge.
2562 * In that case, we need to call pre_destroy() again. check it here.
2564 cgroup_release_and_wakeup_rmdir(&ptr
->css
);
2567 void mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*ptr
)
2569 __mem_cgroup_commit_charge_swapin(page
, ptr
,
2570 MEM_CGROUP_CHARGE_TYPE_MAPPED
);
2573 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup
*mem
)
2575 if (mem_cgroup_disabled())
2579 mem_cgroup_cancel_charge(mem
, PAGE_SIZE
);
2583 __do_uncharge(struct mem_cgroup
*mem
, const enum charge_type ctype
,
2586 struct memcg_batch_info
*batch
= NULL
;
2587 bool uncharge_memsw
= true;
2588 /* If swapout, usage of swap doesn't decrease */
2589 if (!do_swap_account
|| ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
)
2590 uncharge_memsw
= false;
2592 batch
= ¤t
->memcg_batch
;
2594 * In usual, we do css_get() when we remember memcg pointer.
2595 * But in this case, we keep res->usage until end of a series of
2596 * uncharges. Then, it's ok to ignore memcg's refcnt.
2601 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
2602 * In those cases, all pages freed continously can be expected to be in
2603 * the same cgroup and we have chance to coalesce uncharges.
2604 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
2605 * because we want to do uncharge as soon as possible.
2608 if (!batch
->do_batch
|| test_thread_flag(TIF_MEMDIE
))
2609 goto direct_uncharge
;
2611 if (page_size
!= PAGE_SIZE
)
2612 goto direct_uncharge
;
2615 * In typical case, batch->memcg == mem. This means we can
2616 * merge a series of uncharges to an uncharge of res_counter.
2617 * If not, we uncharge res_counter ony by one.
2619 if (batch
->memcg
!= mem
)
2620 goto direct_uncharge
;
2621 /* remember freed charge and uncharge it later */
2622 batch
->bytes
+= PAGE_SIZE
;
2624 batch
->memsw_bytes
+= PAGE_SIZE
;
2627 res_counter_uncharge(&mem
->res
, page_size
);
2629 res_counter_uncharge(&mem
->memsw
, page_size
);
2630 if (unlikely(batch
->memcg
!= mem
))
2631 memcg_oom_recover(mem
);
2636 * uncharge if !page_mapped(page)
2638 static struct mem_cgroup
*
2639 __mem_cgroup_uncharge_common(struct page
*page
, enum charge_type ctype
)
2642 struct page_cgroup
*pc
;
2643 struct mem_cgroup
*mem
= NULL
;
2644 int page_size
= PAGE_SIZE
;
2646 if (mem_cgroup_disabled())
2649 if (PageSwapCache(page
))
2652 if (PageTransHuge(page
)) {
2653 page_size
<<= compound_order(page
);
2654 VM_BUG_ON(!PageTransHuge(page
));
2657 count
= page_size
>> PAGE_SHIFT
;
2659 * Check if our page_cgroup is valid
2661 pc
= lookup_page_cgroup(page
);
2662 if (unlikely(!pc
|| !PageCgroupUsed(pc
)))
2665 lock_page_cgroup(pc
);
2667 mem
= pc
->mem_cgroup
;
2669 if (!PageCgroupUsed(pc
))
2673 case MEM_CGROUP_CHARGE_TYPE_MAPPED
:
2674 case MEM_CGROUP_CHARGE_TYPE_DROP
:
2675 /* See mem_cgroup_prepare_migration() */
2676 if (page_mapped(page
) || PageCgroupMigration(pc
))
2679 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT
:
2680 if (!PageAnon(page
)) { /* Shared memory */
2681 if (page
->mapping
&& !page_is_file_cache(page
))
2683 } else if (page_mapped(page
)) /* Anon */
2690 mem_cgroup_charge_statistics(mem
, PageCgroupCache(pc
), -count
);
2692 ClearPageCgroupUsed(pc
);
2694 * pc->mem_cgroup is not cleared here. It will be accessed when it's
2695 * freed from LRU. This is safe because uncharged page is expected not
2696 * to be reused (freed soon). Exception is SwapCache, it's handled by
2697 * special functions.
2700 unlock_page_cgroup(pc
);
2702 * even after unlock, we have mem->res.usage here and this memcg
2703 * will never be freed.
2705 memcg_check_events(mem
, page
);
2706 if (do_swap_account
&& ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
) {
2707 mem_cgroup_swap_statistics(mem
, true);
2708 mem_cgroup_get(mem
);
2710 if (!mem_cgroup_is_root(mem
))
2711 __do_uncharge(mem
, ctype
, page_size
);
2716 unlock_page_cgroup(pc
);
2720 void mem_cgroup_uncharge_page(struct page
*page
)
2723 if (page_mapped(page
))
2725 if (page
->mapping
&& !PageAnon(page
))
2727 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_MAPPED
);
2730 void mem_cgroup_uncharge_cache_page(struct page
*page
)
2732 VM_BUG_ON(page_mapped(page
));
2733 VM_BUG_ON(page
->mapping
);
2734 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_CACHE
);
2738 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
2739 * In that cases, pages are freed continuously and we can expect pages
2740 * are in the same memcg. All these calls itself limits the number of
2741 * pages freed at once, then uncharge_start/end() is called properly.
2742 * This may be called prural(2) times in a context,
2745 void mem_cgroup_uncharge_start(void)
2747 current
->memcg_batch
.do_batch
++;
2748 /* We can do nest. */
2749 if (current
->memcg_batch
.do_batch
== 1) {
2750 current
->memcg_batch
.memcg
= NULL
;
2751 current
->memcg_batch
.bytes
= 0;
2752 current
->memcg_batch
.memsw_bytes
= 0;
2756 void mem_cgroup_uncharge_end(void)
2758 struct memcg_batch_info
*batch
= ¤t
->memcg_batch
;
2760 if (!batch
->do_batch
)
2764 if (batch
->do_batch
) /* If stacked, do nothing. */
2770 * This "batch->memcg" is valid without any css_get/put etc...
2771 * bacause we hide charges behind us.
2774 res_counter_uncharge(&batch
->memcg
->res
, batch
->bytes
);
2775 if (batch
->memsw_bytes
)
2776 res_counter_uncharge(&batch
->memcg
->memsw
, batch
->memsw_bytes
);
2777 memcg_oom_recover(batch
->memcg
);
2778 /* forget this pointer (for sanity check) */
2779 batch
->memcg
= NULL
;
2784 * called after __delete_from_swap_cache() and drop "page" account.
2785 * memcg information is recorded to swap_cgroup of "ent"
2788 mem_cgroup_uncharge_swapcache(struct page
*page
, swp_entry_t ent
, bool swapout
)
2790 struct mem_cgroup
*memcg
;
2791 int ctype
= MEM_CGROUP_CHARGE_TYPE_SWAPOUT
;
2793 if (!swapout
) /* this was a swap cache but the swap is unused ! */
2794 ctype
= MEM_CGROUP_CHARGE_TYPE_DROP
;
2796 memcg
= __mem_cgroup_uncharge_common(page
, ctype
);
2799 * record memcg information, if swapout && memcg != NULL,
2800 * mem_cgroup_get() was called in uncharge().
2802 if (do_swap_account
&& swapout
&& memcg
)
2803 swap_cgroup_record(ent
, css_id(&memcg
->css
));
2807 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2809 * called from swap_entry_free(). remove record in swap_cgroup and
2810 * uncharge "memsw" account.
2812 void mem_cgroup_uncharge_swap(swp_entry_t ent
)
2814 struct mem_cgroup
*memcg
;
2817 if (!do_swap_account
)
2820 id
= swap_cgroup_record(ent
, 0);
2822 memcg
= mem_cgroup_lookup(id
);
2825 * We uncharge this because swap is freed.
2826 * This memcg can be obsolete one. We avoid calling css_tryget
2828 if (!mem_cgroup_is_root(memcg
))
2829 res_counter_uncharge(&memcg
->memsw
, PAGE_SIZE
);
2830 mem_cgroup_swap_statistics(memcg
, false);
2831 mem_cgroup_put(memcg
);
2837 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
2838 * @entry: swap entry to be moved
2839 * @from: mem_cgroup which the entry is moved from
2840 * @to: mem_cgroup which the entry is moved to
2841 * @need_fixup: whether we should fixup res_counters and refcounts.
2843 * It succeeds only when the swap_cgroup's record for this entry is the same
2844 * as the mem_cgroup's id of @from.
2846 * Returns 0 on success, -EINVAL on failure.
2848 * The caller must have charged to @to, IOW, called res_counter_charge() about
2849 * both res and memsw, and called css_get().
2851 static int mem_cgroup_move_swap_account(swp_entry_t entry
,
2852 struct mem_cgroup
*from
, struct mem_cgroup
*to
, bool need_fixup
)
2854 unsigned short old_id
, new_id
;
2856 old_id
= css_id(&from
->css
);
2857 new_id
= css_id(&to
->css
);
2859 if (swap_cgroup_cmpxchg(entry
, old_id
, new_id
) == old_id
) {
2860 mem_cgroup_swap_statistics(from
, false);
2861 mem_cgroup_swap_statistics(to
, true);
2863 * This function is only called from task migration context now.
2864 * It postpones res_counter and refcount handling till the end
2865 * of task migration(mem_cgroup_clear_mc()) for performance
2866 * improvement. But we cannot postpone mem_cgroup_get(to)
2867 * because if the process that has been moved to @to does
2868 * swap-in, the refcount of @to might be decreased to 0.
2872 if (!mem_cgroup_is_root(from
))
2873 res_counter_uncharge(&from
->memsw
, PAGE_SIZE
);
2874 mem_cgroup_put(from
);
2876 * we charged both to->res and to->memsw, so we should
2879 if (!mem_cgroup_is_root(to
))
2880 res_counter_uncharge(&to
->res
, PAGE_SIZE
);
2887 static inline int mem_cgroup_move_swap_account(swp_entry_t entry
,
2888 struct mem_cgroup
*from
, struct mem_cgroup
*to
, bool need_fixup
)
2895 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
2898 int mem_cgroup_prepare_migration(struct page
*page
,
2899 struct page
*newpage
, struct mem_cgroup
**ptr
, gfp_t gfp_mask
)
2901 struct page_cgroup
*pc
;
2902 struct mem_cgroup
*mem
= NULL
;
2903 enum charge_type ctype
;
2908 VM_BUG_ON(PageTransHuge(page
));
2909 if (mem_cgroup_disabled())
2912 pc
= lookup_page_cgroup(page
);
2913 lock_page_cgroup(pc
);
2914 if (PageCgroupUsed(pc
)) {
2915 mem
= pc
->mem_cgroup
;
2918 * At migrating an anonymous page, its mapcount goes down
2919 * to 0 and uncharge() will be called. But, even if it's fully
2920 * unmapped, migration may fail and this page has to be
2921 * charged again. We set MIGRATION flag here and delay uncharge
2922 * until end_migration() is called
2924 * Corner Case Thinking
2926 * When the old page was mapped as Anon and it's unmap-and-freed
2927 * while migration was ongoing.
2928 * If unmap finds the old page, uncharge() of it will be delayed
2929 * until end_migration(). If unmap finds a new page, it's
2930 * uncharged when it make mapcount to be 1->0. If unmap code
2931 * finds swap_migration_entry, the new page will not be mapped
2932 * and end_migration() will find it(mapcount==0).
2935 * When the old page was mapped but migraion fails, the kernel
2936 * remaps it. A charge for it is kept by MIGRATION flag even
2937 * if mapcount goes down to 0. We can do remap successfully
2938 * without charging it again.
2941 * The "old" page is under lock_page() until the end of
2942 * migration, so, the old page itself will not be swapped-out.
2943 * If the new page is swapped out before end_migraton, our
2944 * hook to usual swap-out path will catch the event.
2947 SetPageCgroupMigration(pc
);
2949 unlock_page_cgroup(pc
);
2951 * If the page is not charged at this point,
2958 ret
= __mem_cgroup_try_charge(NULL
, gfp_mask
, ptr
, false, PAGE_SIZE
);
2959 css_put(&mem
->css
);/* drop extra refcnt */
2960 if (ret
|| *ptr
== NULL
) {
2961 if (PageAnon(page
)) {
2962 lock_page_cgroup(pc
);
2963 ClearPageCgroupMigration(pc
);
2964 unlock_page_cgroup(pc
);
2966 * The old page may be fully unmapped while we kept it.
2968 mem_cgroup_uncharge_page(page
);
2973 * We charge new page before it's used/mapped. So, even if unlock_page()
2974 * is called before end_migration, we can catch all events on this new
2975 * page. In the case new page is migrated but not remapped, new page's
2976 * mapcount will be finally 0 and we call uncharge in end_migration().
2978 pc
= lookup_page_cgroup(newpage
);
2980 ctype
= MEM_CGROUP_CHARGE_TYPE_MAPPED
;
2981 else if (page_is_file_cache(page
))
2982 ctype
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
2984 ctype
= MEM_CGROUP_CHARGE_TYPE_SHMEM
;
2985 __mem_cgroup_commit_charge(mem
, pc
, ctype
, PAGE_SIZE
);
2989 /* remove redundant charge if migration failed*/
2990 void mem_cgroup_end_migration(struct mem_cgroup
*mem
,
2991 struct page
*oldpage
, struct page
*newpage
, bool migration_ok
)
2993 struct page
*used
, *unused
;
2994 struct page_cgroup
*pc
;
2998 /* blocks rmdir() */
2999 cgroup_exclude_rmdir(&mem
->css
);
3000 if (!migration_ok
) {
3008 * We disallowed uncharge of pages under migration because mapcount
3009 * of the page goes down to zero, temporarly.
3010 * Clear the flag and check the page should be charged.
3012 pc
= lookup_page_cgroup(oldpage
);
3013 lock_page_cgroup(pc
);
3014 ClearPageCgroupMigration(pc
);
3015 unlock_page_cgroup(pc
);
3017 __mem_cgroup_uncharge_common(unused
, MEM_CGROUP_CHARGE_TYPE_FORCE
);
3020 * If a page is a file cache, radix-tree replacement is very atomic
3021 * and we can skip this check. When it was an Anon page, its mapcount
3022 * goes down to 0. But because we added MIGRATION flage, it's not
3023 * uncharged yet. There are several case but page->mapcount check
3024 * and USED bit check in mem_cgroup_uncharge_page() will do enough
3025 * check. (see prepare_charge() also)
3028 mem_cgroup_uncharge_page(used
);
3030 * At migration, we may charge account against cgroup which has no
3032 * So, rmdir()->pre_destroy() can be called while we do this charge.
3033 * In that case, we need to call pre_destroy() again. check it here.
3035 cgroup_release_and_wakeup_rmdir(&mem
->css
);
3039 * A call to try to shrink memory usage on charge failure at shmem's swapin.
3040 * Calling hierarchical_reclaim is not enough because we should update
3041 * last_oom_jiffies to prevent pagefault_out_of_memory from invoking global OOM.
3042 * Moreover considering hierarchy, we should reclaim from the mem_over_limit,
3043 * not from the memcg which this page would be charged to.
3044 * try_charge_swapin does all of these works properly.
3046 int mem_cgroup_shmem_charge_fallback(struct page
*page
,
3047 struct mm_struct
*mm
,
3050 struct mem_cgroup
*mem
;
3053 if (mem_cgroup_disabled())
3056 ret
= mem_cgroup_try_charge_swapin(mm
, page
, gfp_mask
, &mem
);
3058 mem_cgroup_cancel_charge_swapin(mem
); /* it does !mem check */
3063 static DEFINE_MUTEX(set_limit_mutex
);
3065 static int mem_cgroup_resize_limit(struct mem_cgroup
*memcg
,
3066 unsigned long long val
)
3069 u64 memswlimit
, memlimit
;
3071 int children
= mem_cgroup_count_children(memcg
);
3072 u64 curusage
, oldusage
;
3076 * For keeping hierarchical_reclaim simple, how long we should retry
3077 * is depends on callers. We set our retry-count to be function
3078 * of # of children which we should visit in this loop.
3080 retry_count
= MEM_CGROUP_RECLAIM_RETRIES
* children
;
3082 oldusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3085 while (retry_count
) {
3086 if (signal_pending(current
)) {
3091 * Rather than hide all in some function, I do this in
3092 * open coded manner. You see what this really does.
3093 * We have to guarantee mem->res.limit < mem->memsw.limit.
3095 mutex_lock(&set_limit_mutex
);
3096 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3097 if (memswlimit
< val
) {
3099 mutex_unlock(&set_limit_mutex
);
3103 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3107 ret
= res_counter_set_limit(&memcg
->res
, val
);
3109 if (memswlimit
== val
)
3110 memcg
->memsw_is_minimum
= true;
3112 memcg
->memsw_is_minimum
= false;
3114 mutex_unlock(&set_limit_mutex
);
3119 mem_cgroup_hierarchical_reclaim(memcg
, NULL
, GFP_KERNEL
,
3120 MEM_CGROUP_RECLAIM_SHRINK
);
3121 curusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3122 /* Usage is reduced ? */
3123 if (curusage
>= oldusage
)
3126 oldusage
= curusage
;
3128 if (!ret
&& enlarge
)
3129 memcg_oom_recover(memcg
);
3134 static int mem_cgroup_resize_memsw_limit(struct mem_cgroup
*memcg
,
3135 unsigned long long val
)
3138 u64 memlimit
, memswlimit
, oldusage
, curusage
;
3139 int children
= mem_cgroup_count_children(memcg
);
3143 /* see mem_cgroup_resize_res_limit */
3144 retry_count
= children
* MEM_CGROUP_RECLAIM_RETRIES
;
3145 oldusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3146 while (retry_count
) {
3147 if (signal_pending(current
)) {
3152 * Rather than hide all in some function, I do this in
3153 * open coded manner. You see what this really does.
3154 * We have to guarantee mem->res.limit < mem->memsw.limit.
3156 mutex_lock(&set_limit_mutex
);
3157 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3158 if (memlimit
> val
) {
3160 mutex_unlock(&set_limit_mutex
);
3163 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3164 if (memswlimit
< val
)
3166 ret
= res_counter_set_limit(&memcg
->memsw
, val
);
3168 if (memlimit
== val
)
3169 memcg
->memsw_is_minimum
= true;
3171 memcg
->memsw_is_minimum
= false;
3173 mutex_unlock(&set_limit_mutex
);
3178 mem_cgroup_hierarchical_reclaim(memcg
, NULL
, GFP_KERNEL
,
3179 MEM_CGROUP_RECLAIM_NOSWAP
|
3180 MEM_CGROUP_RECLAIM_SHRINK
);
3181 curusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3182 /* Usage is reduced ? */
3183 if (curusage
>= oldusage
)
3186 oldusage
= curusage
;
3188 if (!ret
&& enlarge
)
3189 memcg_oom_recover(memcg
);
3193 unsigned long mem_cgroup_soft_limit_reclaim(struct zone
*zone
, int order
,
3196 unsigned long nr_reclaimed
= 0;
3197 struct mem_cgroup_per_zone
*mz
, *next_mz
= NULL
;
3198 unsigned long reclaimed
;
3200 struct mem_cgroup_tree_per_zone
*mctz
;
3201 unsigned long long excess
;
3206 mctz
= soft_limit_tree_node_zone(zone_to_nid(zone
), zone_idx(zone
));
3208 * This loop can run a while, specially if mem_cgroup's continuously
3209 * keep exceeding their soft limit and putting the system under
3216 mz
= mem_cgroup_largest_soft_limit_node(mctz
);
3220 reclaimed
= mem_cgroup_hierarchical_reclaim(mz
->mem
, zone
,
3222 MEM_CGROUP_RECLAIM_SOFT
);
3223 nr_reclaimed
+= reclaimed
;
3224 spin_lock(&mctz
->lock
);
3227 * If we failed to reclaim anything from this memory cgroup
3228 * it is time to move on to the next cgroup
3234 * Loop until we find yet another one.
3236 * By the time we get the soft_limit lock
3237 * again, someone might have aded the
3238 * group back on the RB tree. Iterate to
3239 * make sure we get a different mem.
3240 * mem_cgroup_largest_soft_limit_node returns
3241 * NULL if no other cgroup is present on
3245 __mem_cgroup_largest_soft_limit_node(mctz
);
3246 if (next_mz
== mz
) {
3247 css_put(&next_mz
->mem
->css
);
3249 } else /* next_mz == NULL or other memcg */
3253 __mem_cgroup_remove_exceeded(mz
->mem
, mz
, mctz
);
3254 excess
= res_counter_soft_limit_excess(&mz
->mem
->res
);
3256 * One school of thought says that we should not add
3257 * back the node to the tree if reclaim returns 0.
3258 * But our reclaim could return 0, simply because due
3259 * to priority we are exposing a smaller subset of
3260 * memory to reclaim from. Consider this as a longer
3263 /* If excess == 0, no tree ops */
3264 __mem_cgroup_insert_exceeded(mz
->mem
, mz
, mctz
, excess
);
3265 spin_unlock(&mctz
->lock
);
3266 css_put(&mz
->mem
->css
);
3269 * Could not reclaim anything and there are no more
3270 * mem cgroups to try or we seem to be looping without
3271 * reclaiming anything.
3273 if (!nr_reclaimed
&&
3275 loop
> MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS
))
3277 } while (!nr_reclaimed
);
3279 css_put(&next_mz
->mem
->css
);
3280 return nr_reclaimed
;
3284 * This routine traverse page_cgroup in given list and drop them all.
3285 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
3287 static int mem_cgroup_force_empty_list(struct mem_cgroup
*mem
,
3288 int node
, int zid
, enum lru_list lru
)
3291 struct mem_cgroup_per_zone
*mz
;
3292 struct page_cgroup
*pc
, *busy
;
3293 unsigned long flags
, loop
;
3294 struct list_head
*list
;
3297 zone
= &NODE_DATA(node
)->node_zones
[zid
];
3298 mz
= mem_cgroup_zoneinfo(mem
, node
, zid
);
3299 list
= &mz
->lists
[lru
];
3301 loop
= MEM_CGROUP_ZSTAT(mz
, lru
);
3302 /* give some margin against EBUSY etc...*/
3307 spin_lock_irqsave(&zone
->lru_lock
, flags
);
3308 if (list_empty(list
)) {
3309 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3312 pc
= list_entry(list
->prev
, struct page_cgroup
, lru
);
3314 list_move(&pc
->lru
, list
);
3316 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3319 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3321 ret
= mem_cgroup_move_parent(pc
, mem
, GFP_KERNEL
);
3325 if (ret
== -EBUSY
|| ret
== -EINVAL
) {
3326 /* found lock contention or "pc" is obsolete. */
3333 if (!ret
&& !list_empty(list
))
3339 * make mem_cgroup's charge to be 0 if there is no task.
3340 * This enables deleting this mem_cgroup.
3342 static int mem_cgroup_force_empty(struct mem_cgroup
*mem
, bool free_all
)
3345 int node
, zid
, shrink
;
3346 int nr_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
3347 struct cgroup
*cgrp
= mem
->css
.cgroup
;
3352 /* should free all ? */
3358 if (cgroup_task_count(cgrp
) || !list_empty(&cgrp
->children
))
3361 if (signal_pending(current
))
3363 /* This is for making all *used* pages to be on LRU. */
3364 lru_add_drain_all();
3365 drain_all_stock_sync();
3367 mem_cgroup_start_move(mem
);
3368 for_each_node_state(node
, N_HIGH_MEMORY
) {
3369 for (zid
= 0; !ret
&& zid
< MAX_NR_ZONES
; zid
++) {
3372 ret
= mem_cgroup_force_empty_list(mem
,
3381 mem_cgroup_end_move(mem
);
3382 memcg_oom_recover(mem
);
3383 /* it seems parent cgroup doesn't have enough mem */
3387 /* "ret" should also be checked to ensure all lists are empty. */
3388 } while (mem
->res
.usage
> 0 || ret
);
3394 /* returns EBUSY if there is a task or if we come here twice. */
3395 if (cgroup_task_count(cgrp
) || !list_empty(&cgrp
->children
) || shrink
) {
3399 /* we call try-to-free pages for make this cgroup empty */
3400 lru_add_drain_all();
3401 /* try to free all pages in this cgroup */
3403 while (nr_retries
&& mem
->res
.usage
> 0) {
3406 if (signal_pending(current
)) {
3410 progress
= try_to_free_mem_cgroup_pages(mem
, GFP_KERNEL
,
3411 false, get_swappiness(mem
));
3414 /* maybe some writeback is necessary */
3415 congestion_wait(BLK_RW_ASYNC
, HZ
/10);
3420 /* try move_account...there may be some *locked* pages. */
3424 int mem_cgroup_force_empty_write(struct cgroup
*cont
, unsigned int event
)
3426 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont
), true);
3430 static u64
mem_cgroup_hierarchy_read(struct cgroup
*cont
, struct cftype
*cft
)
3432 return mem_cgroup_from_cont(cont
)->use_hierarchy
;
3435 static int mem_cgroup_hierarchy_write(struct cgroup
*cont
, struct cftype
*cft
,
3439 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
3440 struct cgroup
*parent
= cont
->parent
;
3441 struct mem_cgroup
*parent_mem
= NULL
;
3444 parent_mem
= mem_cgroup_from_cont(parent
);
3448 * If parent's use_hierarchy is set, we can't make any modifications
3449 * in the child subtrees. If it is unset, then the change can
3450 * occur, provided the current cgroup has no children.
3452 * For the root cgroup, parent_mem is NULL, we allow value to be
3453 * set if there are no children.
3455 if ((!parent_mem
|| !parent_mem
->use_hierarchy
) &&
3456 (val
== 1 || val
== 0)) {
3457 if (list_empty(&cont
->children
))
3458 mem
->use_hierarchy
= val
;
3469 static u64
mem_cgroup_get_recursive_idx_stat(struct mem_cgroup
*mem
,
3470 enum mem_cgroup_stat_index idx
)
3472 struct mem_cgroup
*iter
;
3475 /* each per cpu's value can be minus.Then, use s64 */
3476 for_each_mem_cgroup_tree(iter
, mem
)
3477 val
+= mem_cgroup_read_stat(iter
, idx
);
3479 if (val
< 0) /* race ? */
3484 static inline u64
mem_cgroup_usage(struct mem_cgroup
*mem
, bool swap
)
3488 if (!mem_cgroup_is_root(mem
)) {
3490 return res_counter_read_u64(&mem
->res
, RES_USAGE
);
3492 return res_counter_read_u64(&mem
->memsw
, RES_USAGE
);
3495 val
= mem_cgroup_get_recursive_idx_stat(mem
, MEM_CGROUP_STAT_CACHE
);
3496 val
+= mem_cgroup_get_recursive_idx_stat(mem
, MEM_CGROUP_STAT_RSS
);
3499 val
+= mem_cgroup_get_recursive_idx_stat(mem
,
3500 MEM_CGROUP_STAT_SWAPOUT
);
3502 return val
<< PAGE_SHIFT
;
3505 static u64
mem_cgroup_read(struct cgroup
*cont
, struct cftype
*cft
)
3507 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
3511 type
= MEMFILE_TYPE(cft
->private);
3512 name
= MEMFILE_ATTR(cft
->private);
3515 if (name
== RES_USAGE
)
3516 val
= mem_cgroup_usage(mem
, false);
3518 val
= res_counter_read_u64(&mem
->res
, name
);
3521 if (name
== RES_USAGE
)
3522 val
= mem_cgroup_usage(mem
, true);
3524 val
= res_counter_read_u64(&mem
->memsw
, name
);
3533 * The user of this function is...
3536 static int mem_cgroup_write(struct cgroup
*cont
, struct cftype
*cft
,
3539 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3541 unsigned long long val
;
3544 type
= MEMFILE_TYPE(cft
->private);
3545 name
= MEMFILE_ATTR(cft
->private);
3548 if (mem_cgroup_is_root(memcg
)) { /* Can't set limit on root */
3552 /* This function does all necessary parse...reuse it */
3553 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
3557 ret
= mem_cgroup_resize_limit(memcg
, val
);
3559 ret
= mem_cgroup_resize_memsw_limit(memcg
, val
);
3561 case RES_SOFT_LIMIT
:
3562 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
3566 * For memsw, soft limits are hard to implement in terms
3567 * of semantics, for now, we support soft limits for
3568 * control without swap
3571 ret
= res_counter_set_soft_limit(&memcg
->res
, val
);
3576 ret
= -EINVAL
; /* should be BUG() ? */
3582 static void memcg_get_hierarchical_limit(struct mem_cgroup
*memcg
,
3583 unsigned long long *mem_limit
, unsigned long long *memsw_limit
)
3585 struct cgroup
*cgroup
;
3586 unsigned long long min_limit
, min_memsw_limit
, tmp
;
3588 min_limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3589 min_memsw_limit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3590 cgroup
= memcg
->css
.cgroup
;
3591 if (!memcg
->use_hierarchy
)
3594 while (cgroup
->parent
) {
3595 cgroup
= cgroup
->parent
;
3596 memcg
= mem_cgroup_from_cont(cgroup
);
3597 if (!memcg
->use_hierarchy
)
3599 tmp
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3600 min_limit
= min(min_limit
, tmp
);
3601 tmp
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3602 min_memsw_limit
= min(min_memsw_limit
, tmp
);
3605 *mem_limit
= min_limit
;
3606 *memsw_limit
= min_memsw_limit
;
3610 static int mem_cgroup_reset(struct cgroup
*cont
, unsigned int event
)
3612 struct mem_cgroup
*mem
;
3615 mem
= mem_cgroup_from_cont(cont
);
3616 type
= MEMFILE_TYPE(event
);
3617 name
= MEMFILE_ATTR(event
);
3621 res_counter_reset_max(&mem
->res
);
3623 res_counter_reset_max(&mem
->memsw
);
3627 res_counter_reset_failcnt(&mem
->res
);
3629 res_counter_reset_failcnt(&mem
->memsw
);
3636 static u64
mem_cgroup_move_charge_read(struct cgroup
*cgrp
,
3639 return mem_cgroup_from_cont(cgrp
)->move_charge_at_immigrate
;
3643 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
3644 struct cftype
*cft
, u64 val
)
3646 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgrp
);
3648 if (val
>= (1 << NR_MOVE_TYPE
))
3651 * We check this value several times in both in can_attach() and
3652 * attach(), so we need cgroup lock to prevent this value from being
3656 mem
->move_charge_at_immigrate
= val
;
3662 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
3663 struct cftype
*cft
, u64 val
)
3670 /* For read statistics */
3686 struct mcs_total_stat
{
3687 s64 stat
[NR_MCS_STAT
];
3693 } memcg_stat_strings
[NR_MCS_STAT
] = {
3694 {"cache", "total_cache"},
3695 {"rss", "total_rss"},
3696 {"mapped_file", "total_mapped_file"},
3697 {"pgpgin", "total_pgpgin"},
3698 {"pgpgout", "total_pgpgout"},
3699 {"swap", "total_swap"},
3700 {"inactive_anon", "total_inactive_anon"},
3701 {"active_anon", "total_active_anon"},
3702 {"inactive_file", "total_inactive_file"},
3703 {"active_file", "total_active_file"},
3704 {"unevictable", "total_unevictable"}
3709 mem_cgroup_get_local_stat(struct mem_cgroup
*mem
, struct mcs_total_stat
*s
)
3714 val
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_CACHE
);
3715 s
->stat
[MCS_CACHE
] += val
* PAGE_SIZE
;
3716 val
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_RSS
);
3717 s
->stat
[MCS_RSS
] += val
* PAGE_SIZE
;
3718 val
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_FILE_MAPPED
);
3719 s
->stat
[MCS_FILE_MAPPED
] += val
* PAGE_SIZE
;
3720 val
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_PGPGIN_COUNT
);
3721 s
->stat
[MCS_PGPGIN
] += val
;
3722 val
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_PGPGOUT_COUNT
);
3723 s
->stat
[MCS_PGPGOUT
] += val
;
3724 if (do_swap_account
) {
3725 val
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_SWAPOUT
);
3726 s
->stat
[MCS_SWAP
] += val
* PAGE_SIZE
;
3730 val
= mem_cgroup_get_local_zonestat(mem
, LRU_INACTIVE_ANON
);
3731 s
->stat
[MCS_INACTIVE_ANON
] += val
* PAGE_SIZE
;
3732 val
= mem_cgroup_get_local_zonestat(mem
, LRU_ACTIVE_ANON
);
3733 s
->stat
[MCS_ACTIVE_ANON
] += val
* PAGE_SIZE
;
3734 val
= mem_cgroup_get_local_zonestat(mem
, LRU_INACTIVE_FILE
);
3735 s
->stat
[MCS_INACTIVE_FILE
] += val
* PAGE_SIZE
;
3736 val
= mem_cgroup_get_local_zonestat(mem
, LRU_ACTIVE_FILE
);
3737 s
->stat
[MCS_ACTIVE_FILE
] += val
* PAGE_SIZE
;
3738 val
= mem_cgroup_get_local_zonestat(mem
, LRU_UNEVICTABLE
);
3739 s
->stat
[MCS_UNEVICTABLE
] += val
* PAGE_SIZE
;
3743 mem_cgroup_get_total_stat(struct mem_cgroup
*mem
, struct mcs_total_stat
*s
)
3745 struct mem_cgroup
*iter
;
3747 for_each_mem_cgroup_tree(iter
, mem
)
3748 mem_cgroup_get_local_stat(iter
, s
);
3751 static int mem_control_stat_show(struct cgroup
*cont
, struct cftype
*cft
,
3752 struct cgroup_map_cb
*cb
)
3754 struct mem_cgroup
*mem_cont
= mem_cgroup_from_cont(cont
);
3755 struct mcs_total_stat mystat
;
3758 memset(&mystat
, 0, sizeof(mystat
));
3759 mem_cgroup_get_local_stat(mem_cont
, &mystat
);
3761 for (i
= 0; i
< NR_MCS_STAT
; i
++) {
3762 if (i
== MCS_SWAP
&& !do_swap_account
)
3764 cb
->fill(cb
, memcg_stat_strings
[i
].local_name
, mystat
.stat
[i
]);
3767 /* Hierarchical information */
3769 unsigned long long limit
, memsw_limit
;
3770 memcg_get_hierarchical_limit(mem_cont
, &limit
, &memsw_limit
);
3771 cb
->fill(cb
, "hierarchical_memory_limit", limit
);
3772 if (do_swap_account
)
3773 cb
->fill(cb
, "hierarchical_memsw_limit", memsw_limit
);
3776 memset(&mystat
, 0, sizeof(mystat
));
3777 mem_cgroup_get_total_stat(mem_cont
, &mystat
);
3778 for (i
= 0; i
< NR_MCS_STAT
; i
++) {
3779 if (i
== MCS_SWAP
&& !do_swap_account
)
3781 cb
->fill(cb
, memcg_stat_strings
[i
].total_name
, mystat
.stat
[i
]);
3784 #ifdef CONFIG_DEBUG_VM
3785 cb
->fill(cb
, "inactive_ratio", calc_inactive_ratio(mem_cont
, NULL
));
3789 struct mem_cgroup_per_zone
*mz
;
3790 unsigned long recent_rotated
[2] = {0, 0};
3791 unsigned long recent_scanned
[2] = {0, 0};
3793 for_each_online_node(nid
)
3794 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
3795 mz
= mem_cgroup_zoneinfo(mem_cont
, nid
, zid
);
3797 recent_rotated
[0] +=
3798 mz
->reclaim_stat
.recent_rotated
[0];
3799 recent_rotated
[1] +=
3800 mz
->reclaim_stat
.recent_rotated
[1];
3801 recent_scanned
[0] +=
3802 mz
->reclaim_stat
.recent_scanned
[0];
3803 recent_scanned
[1] +=
3804 mz
->reclaim_stat
.recent_scanned
[1];
3806 cb
->fill(cb
, "recent_rotated_anon", recent_rotated
[0]);
3807 cb
->fill(cb
, "recent_rotated_file", recent_rotated
[1]);
3808 cb
->fill(cb
, "recent_scanned_anon", recent_scanned
[0]);
3809 cb
->fill(cb
, "recent_scanned_file", recent_scanned
[1]);
3816 static u64
mem_cgroup_swappiness_read(struct cgroup
*cgrp
, struct cftype
*cft
)
3818 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
3820 return get_swappiness(memcg
);
3823 static int mem_cgroup_swappiness_write(struct cgroup
*cgrp
, struct cftype
*cft
,
3826 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
3827 struct mem_cgroup
*parent
;
3832 if (cgrp
->parent
== NULL
)
3835 parent
= mem_cgroup_from_cont(cgrp
->parent
);
3839 /* If under hierarchy, only empty-root can set this value */
3840 if ((parent
->use_hierarchy
) ||
3841 (memcg
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
3846 spin_lock(&memcg
->reclaim_param_lock
);
3847 memcg
->swappiness
= val
;
3848 spin_unlock(&memcg
->reclaim_param_lock
);
3855 static void __mem_cgroup_threshold(struct mem_cgroup
*memcg
, bool swap
)
3857 struct mem_cgroup_threshold_ary
*t
;
3863 t
= rcu_dereference(memcg
->thresholds
.primary
);
3865 t
= rcu_dereference(memcg
->memsw_thresholds
.primary
);
3870 usage
= mem_cgroup_usage(memcg
, swap
);
3873 * current_threshold points to threshold just below usage.
3874 * If it's not true, a threshold was crossed after last
3875 * call of __mem_cgroup_threshold().
3877 i
= t
->current_threshold
;
3880 * Iterate backward over array of thresholds starting from
3881 * current_threshold and check if a threshold is crossed.
3882 * If none of thresholds below usage is crossed, we read
3883 * only one element of the array here.
3885 for (; i
>= 0 && unlikely(t
->entries
[i
].threshold
> usage
); i
--)
3886 eventfd_signal(t
->entries
[i
].eventfd
, 1);
3888 /* i = current_threshold + 1 */
3892 * Iterate forward over array of thresholds starting from
3893 * current_threshold+1 and check if a threshold is crossed.
3894 * If none of thresholds above usage is crossed, we read
3895 * only one element of the array here.
3897 for (; i
< t
->size
&& unlikely(t
->entries
[i
].threshold
<= usage
); i
++)
3898 eventfd_signal(t
->entries
[i
].eventfd
, 1);
3900 /* Update current_threshold */
3901 t
->current_threshold
= i
- 1;
3906 static void mem_cgroup_threshold(struct mem_cgroup
*memcg
)
3909 __mem_cgroup_threshold(memcg
, false);
3910 if (do_swap_account
)
3911 __mem_cgroup_threshold(memcg
, true);
3913 memcg
= parent_mem_cgroup(memcg
);
3917 static int compare_thresholds(const void *a
, const void *b
)
3919 const struct mem_cgroup_threshold
*_a
= a
;
3920 const struct mem_cgroup_threshold
*_b
= b
;
3922 return _a
->threshold
- _b
->threshold
;
3925 static int mem_cgroup_oom_notify_cb(struct mem_cgroup
*mem
)
3927 struct mem_cgroup_eventfd_list
*ev
;
3929 list_for_each_entry(ev
, &mem
->oom_notify
, list
)
3930 eventfd_signal(ev
->eventfd
, 1);
3934 static void mem_cgroup_oom_notify(struct mem_cgroup
*mem
)
3936 struct mem_cgroup
*iter
;
3938 for_each_mem_cgroup_tree(iter
, mem
)
3939 mem_cgroup_oom_notify_cb(iter
);
3942 static int mem_cgroup_usage_register_event(struct cgroup
*cgrp
,
3943 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
3945 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
3946 struct mem_cgroup_thresholds
*thresholds
;
3947 struct mem_cgroup_threshold_ary
*new;
3948 int type
= MEMFILE_TYPE(cft
->private);
3949 u64 threshold
, usage
;
3952 ret
= res_counter_memparse_write_strategy(args
, &threshold
);
3956 mutex_lock(&memcg
->thresholds_lock
);
3959 thresholds
= &memcg
->thresholds
;
3960 else if (type
== _MEMSWAP
)
3961 thresholds
= &memcg
->memsw_thresholds
;
3965 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
3967 /* Check if a threshold crossed before adding a new one */
3968 if (thresholds
->primary
)
3969 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
3971 size
= thresholds
->primary
? thresholds
->primary
->size
+ 1 : 1;
3973 /* Allocate memory for new array of thresholds */
3974 new = kmalloc(sizeof(*new) + size
* sizeof(struct mem_cgroup_threshold
),
3982 /* Copy thresholds (if any) to new array */
3983 if (thresholds
->primary
) {
3984 memcpy(new->entries
, thresholds
->primary
->entries
, (size
- 1) *
3985 sizeof(struct mem_cgroup_threshold
));
3988 /* Add new threshold */
3989 new->entries
[size
- 1].eventfd
= eventfd
;
3990 new->entries
[size
- 1].threshold
= threshold
;
3992 /* Sort thresholds. Registering of new threshold isn't time-critical */
3993 sort(new->entries
, size
, sizeof(struct mem_cgroup_threshold
),
3994 compare_thresholds
, NULL
);
3996 /* Find current threshold */
3997 new->current_threshold
= -1;
3998 for (i
= 0; i
< size
; i
++) {
3999 if (new->entries
[i
].threshold
< usage
) {
4001 * new->current_threshold will not be used until
4002 * rcu_assign_pointer(), so it's safe to increment
4005 ++new->current_threshold
;
4009 /* Free old spare buffer and save old primary buffer as spare */
4010 kfree(thresholds
->spare
);
4011 thresholds
->spare
= thresholds
->primary
;
4013 rcu_assign_pointer(thresholds
->primary
, new);
4015 /* To be sure that nobody uses thresholds */
4019 mutex_unlock(&memcg
->thresholds_lock
);
4024 static void mem_cgroup_usage_unregister_event(struct cgroup
*cgrp
,
4025 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
4027 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4028 struct mem_cgroup_thresholds
*thresholds
;
4029 struct mem_cgroup_threshold_ary
*new;
4030 int type
= MEMFILE_TYPE(cft
->private);
4034 mutex_lock(&memcg
->thresholds_lock
);
4036 thresholds
= &memcg
->thresholds
;
4037 else if (type
== _MEMSWAP
)
4038 thresholds
= &memcg
->memsw_thresholds
;
4043 * Something went wrong if we trying to unregister a threshold
4044 * if we don't have thresholds
4046 BUG_ON(!thresholds
);
4048 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
4050 /* Check if a threshold crossed before removing */
4051 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
4053 /* Calculate new number of threshold */
4055 for (i
= 0; i
< thresholds
->primary
->size
; i
++) {
4056 if (thresholds
->primary
->entries
[i
].eventfd
!= eventfd
)
4060 new = thresholds
->spare
;
4062 /* Set thresholds array to NULL if we don't have thresholds */
4071 /* Copy thresholds and find current threshold */
4072 new->current_threshold
= -1;
4073 for (i
= 0, j
= 0; i
< thresholds
->primary
->size
; i
++) {
4074 if (thresholds
->primary
->entries
[i
].eventfd
== eventfd
)
4077 new->entries
[j
] = thresholds
->primary
->entries
[i
];
4078 if (new->entries
[j
].threshold
< usage
) {
4080 * new->current_threshold will not be used
4081 * until rcu_assign_pointer(), so it's safe to increment
4084 ++new->current_threshold
;
4090 /* Swap primary and spare array */
4091 thresholds
->spare
= thresholds
->primary
;
4092 rcu_assign_pointer(thresholds
->primary
, new);
4094 /* To be sure that nobody uses thresholds */
4097 mutex_unlock(&memcg
->thresholds_lock
);
4100 static int mem_cgroup_oom_register_event(struct cgroup
*cgrp
,
4101 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
4103 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4104 struct mem_cgroup_eventfd_list
*event
;
4105 int type
= MEMFILE_TYPE(cft
->private);
4107 BUG_ON(type
!= _OOM_TYPE
);
4108 event
= kmalloc(sizeof(*event
), GFP_KERNEL
);
4112 mutex_lock(&memcg_oom_mutex
);
4114 event
->eventfd
= eventfd
;
4115 list_add(&event
->list
, &memcg
->oom_notify
);
4117 /* already in OOM ? */
4118 if (atomic_read(&memcg
->oom_lock
))
4119 eventfd_signal(eventfd
, 1);
4120 mutex_unlock(&memcg_oom_mutex
);
4125 static void mem_cgroup_oom_unregister_event(struct cgroup
*cgrp
,
4126 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
4128 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgrp
);
4129 struct mem_cgroup_eventfd_list
*ev
, *tmp
;
4130 int type
= MEMFILE_TYPE(cft
->private);
4132 BUG_ON(type
!= _OOM_TYPE
);
4134 mutex_lock(&memcg_oom_mutex
);
4136 list_for_each_entry_safe(ev
, tmp
, &mem
->oom_notify
, list
) {
4137 if (ev
->eventfd
== eventfd
) {
4138 list_del(&ev
->list
);
4143 mutex_unlock(&memcg_oom_mutex
);
4146 static int mem_cgroup_oom_control_read(struct cgroup
*cgrp
,
4147 struct cftype
*cft
, struct cgroup_map_cb
*cb
)
4149 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgrp
);
4151 cb
->fill(cb
, "oom_kill_disable", mem
->oom_kill_disable
);
4153 if (atomic_read(&mem
->oom_lock
))
4154 cb
->fill(cb
, "under_oom", 1);
4156 cb
->fill(cb
, "under_oom", 0);
4160 static int mem_cgroup_oom_control_write(struct cgroup
*cgrp
,
4161 struct cftype
*cft
, u64 val
)
4163 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgrp
);
4164 struct mem_cgroup
*parent
;
4166 /* cannot set to root cgroup and only 0 and 1 are allowed */
4167 if (!cgrp
->parent
|| !((val
== 0) || (val
== 1)))
4170 parent
= mem_cgroup_from_cont(cgrp
->parent
);
4173 /* oom-kill-disable is a flag for subhierarchy. */
4174 if ((parent
->use_hierarchy
) ||
4175 (mem
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
4179 mem
->oom_kill_disable
= val
;
4181 memcg_oom_recover(mem
);
4186 static struct cftype mem_cgroup_files
[] = {
4188 .name
= "usage_in_bytes",
4189 .private = MEMFILE_PRIVATE(_MEM
, RES_USAGE
),
4190 .read_u64
= mem_cgroup_read
,
4191 .register_event
= mem_cgroup_usage_register_event
,
4192 .unregister_event
= mem_cgroup_usage_unregister_event
,
4195 .name
= "max_usage_in_bytes",
4196 .private = MEMFILE_PRIVATE(_MEM
, RES_MAX_USAGE
),
4197 .trigger
= mem_cgroup_reset
,
4198 .read_u64
= mem_cgroup_read
,
4201 .name
= "limit_in_bytes",
4202 .private = MEMFILE_PRIVATE(_MEM
, RES_LIMIT
),
4203 .write_string
= mem_cgroup_write
,
4204 .read_u64
= mem_cgroup_read
,
4207 .name
= "soft_limit_in_bytes",
4208 .private = MEMFILE_PRIVATE(_MEM
, RES_SOFT_LIMIT
),
4209 .write_string
= mem_cgroup_write
,
4210 .read_u64
= mem_cgroup_read
,
4214 .private = MEMFILE_PRIVATE(_MEM
, RES_FAILCNT
),
4215 .trigger
= mem_cgroup_reset
,
4216 .read_u64
= mem_cgroup_read
,
4220 .read_map
= mem_control_stat_show
,
4223 .name
= "force_empty",
4224 .trigger
= mem_cgroup_force_empty_write
,
4227 .name
= "use_hierarchy",
4228 .write_u64
= mem_cgroup_hierarchy_write
,
4229 .read_u64
= mem_cgroup_hierarchy_read
,
4232 .name
= "swappiness",
4233 .read_u64
= mem_cgroup_swappiness_read
,
4234 .write_u64
= mem_cgroup_swappiness_write
,
4237 .name
= "move_charge_at_immigrate",
4238 .read_u64
= mem_cgroup_move_charge_read
,
4239 .write_u64
= mem_cgroup_move_charge_write
,
4242 .name
= "oom_control",
4243 .read_map
= mem_cgroup_oom_control_read
,
4244 .write_u64
= mem_cgroup_oom_control_write
,
4245 .register_event
= mem_cgroup_oom_register_event
,
4246 .unregister_event
= mem_cgroup_oom_unregister_event
,
4247 .private = MEMFILE_PRIVATE(_OOM_TYPE
, OOM_CONTROL
),
4251 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4252 static struct cftype memsw_cgroup_files
[] = {
4254 .name
= "memsw.usage_in_bytes",
4255 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_USAGE
),
4256 .read_u64
= mem_cgroup_read
,
4257 .register_event
= mem_cgroup_usage_register_event
,
4258 .unregister_event
= mem_cgroup_usage_unregister_event
,
4261 .name
= "memsw.max_usage_in_bytes",
4262 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_MAX_USAGE
),
4263 .trigger
= mem_cgroup_reset
,
4264 .read_u64
= mem_cgroup_read
,
4267 .name
= "memsw.limit_in_bytes",
4268 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_LIMIT
),
4269 .write_string
= mem_cgroup_write
,
4270 .read_u64
= mem_cgroup_read
,
4273 .name
= "memsw.failcnt",
4274 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_FAILCNT
),
4275 .trigger
= mem_cgroup_reset
,
4276 .read_u64
= mem_cgroup_read
,
4280 static int register_memsw_files(struct cgroup
*cont
, struct cgroup_subsys
*ss
)
4282 if (!do_swap_account
)
4284 return cgroup_add_files(cont
, ss
, memsw_cgroup_files
,
4285 ARRAY_SIZE(memsw_cgroup_files
));
4288 static int register_memsw_files(struct cgroup
*cont
, struct cgroup_subsys
*ss
)
4294 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup
*mem
, int node
)
4296 struct mem_cgroup_per_node
*pn
;
4297 struct mem_cgroup_per_zone
*mz
;
4299 int zone
, tmp
= node
;
4301 * This routine is called against possible nodes.
4302 * But it's BUG to call kmalloc() against offline node.
4304 * TODO: this routine can waste much memory for nodes which will
4305 * never be onlined. It's better to use memory hotplug callback
4308 if (!node_state(node
, N_NORMAL_MEMORY
))
4310 pn
= kzalloc_node(sizeof(*pn
), GFP_KERNEL
, tmp
);
4314 mem
->info
.nodeinfo
[node
] = pn
;
4315 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
4316 mz
= &pn
->zoneinfo
[zone
];
4318 INIT_LIST_HEAD(&mz
->lists
[l
]);
4319 mz
->usage_in_excess
= 0;
4320 mz
->on_tree
= false;
4326 static void free_mem_cgroup_per_zone_info(struct mem_cgroup
*mem
, int node
)
4328 kfree(mem
->info
.nodeinfo
[node
]);
4331 static struct mem_cgroup
*mem_cgroup_alloc(void)
4333 struct mem_cgroup
*mem
;
4334 int size
= sizeof(struct mem_cgroup
);
4336 /* Can be very big if MAX_NUMNODES is very big */
4337 if (size
< PAGE_SIZE
)
4338 mem
= kzalloc(size
, GFP_KERNEL
);
4340 mem
= vzalloc(size
);
4345 mem
->stat
= alloc_percpu(struct mem_cgroup_stat_cpu
);
4348 spin_lock_init(&mem
->pcp_counter_lock
);
4352 if (size
< PAGE_SIZE
)
4360 * At destroying mem_cgroup, references from swap_cgroup can remain.
4361 * (scanning all at force_empty is too costly...)
4363 * Instead of clearing all references at force_empty, we remember
4364 * the number of reference from swap_cgroup and free mem_cgroup when
4365 * it goes down to 0.
4367 * Removal of cgroup itself succeeds regardless of refs from swap.
4370 static void __mem_cgroup_free(struct mem_cgroup
*mem
)
4374 mem_cgroup_remove_from_trees(mem
);
4375 free_css_id(&mem_cgroup_subsys
, &mem
->css
);
4377 for_each_node_state(node
, N_POSSIBLE
)
4378 free_mem_cgroup_per_zone_info(mem
, node
);
4380 free_percpu(mem
->stat
);
4381 if (sizeof(struct mem_cgroup
) < PAGE_SIZE
)
4387 static void mem_cgroup_get(struct mem_cgroup
*mem
)
4389 atomic_inc(&mem
->refcnt
);
4392 static void __mem_cgroup_put(struct mem_cgroup
*mem
, int count
)
4394 if (atomic_sub_and_test(count
, &mem
->refcnt
)) {
4395 struct mem_cgroup
*parent
= parent_mem_cgroup(mem
);
4396 __mem_cgroup_free(mem
);
4398 mem_cgroup_put(parent
);
4402 static void mem_cgroup_put(struct mem_cgroup
*mem
)
4404 __mem_cgroup_put(mem
, 1);
4408 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
4410 static struct mem_cgroup
*parent_mem_cgroup(struct mem_cgroup
*mem
)
4412 if (!mem
->res
.parent
)
4414 return mem_cgroup_from_res_counter(mem
->res
.parent
, res
);
4417 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4418 static void __init
enable_swap_cgroup(void)
4420 if (!mem_cgroup_disabled() && really_do_swap_account
)
4421 do_swap_account
= 1;
4424 static void __init
enable_swap_cgroup(void)
4429 static int mem_cgroup_soft_limit_tree_init(void)
4431 struct mem_cgroup_tree_per_node
*rtpn
;
4432 struct mem_cgroup_tree_per_zone
*rtpz
;
4433 int tmp
, node
, zone
;
4435 for_each_node_state(node
, N_POSSIBLE
) {
4437 if (!node_state(node
, N_NORMAL_MEMORY
))
4439 rtpn
= kzalloc_node(sizeof(*rtpn
), GFP_KERNEL
, tmp
);
4443 soft_limit_tree
.rb_tree_per_node
[node
] = rtpn
;
4445 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
4446 rtpz
= &rtpn
->rb_tree_per_zone
[zone
];
4447 rtpz
->rb_root
= RB_ROOT
;
4448 spin_lock_init(&rtpz
->lock
);
4454 static struct cgroup_subsys_state
* __ref
4455 mem_cgroup_create(struct cgroup_subsys
*ss
, struct cgroup
*cont
)
4457 struct mem_cgroup
*mem
, *parent
;
4458 long error
= -ENOMEM
;
4461 mem
= mem_cgroup_alloc();
4463 return ERR_PTR(error
);
4465 for_each_node_state(node
, N_POSSIBLE
)
4466 if (alloc_mem_cgroup_per_zone_info(mem
, node
))
4470 if (cont
->parent
== NULL
) {
4472 enable_swap_cgroup();
4474 root_mem_cgroup
= mem
;
4475 if (mem_cgroup_soft_limit_tree_init())
4477 for_each_possible_cpu(cpu
) {
4478 struct memcg_stock_pcp
*stock
=
4479 &per_cpu(memcg_stock
, cpu
);
4480 INIT_WORK(&stock
->work
, drain_local_stock
);
4482 hotcpu_notifier(memcg_cpu_hotplug_callback
, 0);
4484 parent
= mem_cgroup_from_cont(cont
->parent
);
4485 mem
->use_hierarchy
= parent
->use_hierarchy
;
4486 mem
->oom_kill_disable
= parent
->oom_kill_disable
;
4489 if (parent
&& parent
->use_hierarchy
) {
4490 res_counter_init(&mem
->res
, &parent
->res
);
4491 res_counter_init(&mem
->memsw
, &parent
->memsw
);
4493 * We increment refcnt of the parent to ensure that we can
4494 * safely access it on res_counter_charge/uncharge.
4495 * This refcnt will be decremented when freeing this
4496 * mem_cgroup(see mem_cgroup_put).
4498 mem_cgroup_get(parent
);
4500 res_counter_init(&mem
->res
, NULL
);
4501 res_counter_init(&mem
->memsw
, NULL
);
4503 mem
->last_scanned_child
= 0;
4504 spin_lock_init(&mem
->reclaim_param_lock
);
4505 INIT_LIST_HEAD(&mem
->oom_notify
);
4508 mem
->swappiness
= get_swappiness(parent
);
4509 atomic_set(&mem
->refcnt
, 1);
4510 mem
->move_charge_at_immigrate
= 0;
4511 mutex_init(&mem
->thresholds_lock
);
4514 __mem_cgroup_free(mem
);
4515 root_mem_cgroup
= NULL
;
4516 return ERR_PTR(error
);
4519 static int mem_cgroup_pre_destroy(struct cgroup_subsys
*ss
,
4520 struct cgroup
*cont
)
4522 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
4524 return mem_cgroup_force_empty(mem
, false);
4527 static void mem_cgroup_destroy(struct cgroup_subsys
*ss
,
4528 struct cgroup
*cont
)
4530 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
4532 mem_cgroup_put(mem
);
4535 static int mem_cgroup_populate(struct cgroup_subsys
*ss
,
4536 struct cgroup
*cont
)
4540 ret
= cgroup_add_files(cont
, ss
, mem_cgroup_files
,
4541 ARRAY_SIZE(mem_cgroup_files
));
4544 ret
= register_memsw_files(cont
, ss
);
4549 /* Handlers for move charge at task migration. */
4550 #define PRECHARGE_COUNT_AT_ONCE 256
4551 static int mem_cgroup_do_precharge(unsigned long count
)
4554 int batch_count
= PRECHARGE_COUNT_AT_ONCE
;
4555 struct mem_cgroup
*mem
= mc
.to
;
4557 if (mem_cgroup_is_root(mem
)) {
4558 mc
.precharge
+= count
;
4559 /* we don't need css_get for root */
4562 /* try to charge at once */
4564 struct res_counter
*dummy
;
4566 * "mem" cannot be under rmdir() because we've already checked
4567 * by cgroup_lock_live_cgroup() that it is not removed and we
4568 * are still under the same cgroup_mutex. So we can postpone
4571 if (res_counter_charge(&mem
->res
, PAGE_SIZE
* count
, &dummy
))
4573 if (do_swap_account
&& res_counter_charge(&mem
->memsw
,
4574 PAGE_SIZE
* count
, &dummy
)) {
4575 res_counter_uncharge(&mem
->res
, PAGE_SIZE
* count
);
4578 mc
.precharge
+= count
;
4582 /* fall back to one by one charge */
4584 if (signal_pending(current
)) {
4588 if (!batch_count
--) {
4589 batch_count
= PRECHARGE_COUNT_AT_ONCE
;
4592 ret
= __mem_cgroup_try_charge(NULL
, GFP_KERNEL
, &mem
, false,
4595 /* mem_cgroup_clear_mc() will do uncharge later */
4603 * is_target_pte_for_mc - check a pte whether it is valid for move charge
4604 * @vma: the vma the pte to be checked belongs
4605 * @addr: the address corresponding to the pte to be checked
4606 * @ptent: the pte to be checked
4607 * @target: the pointer the target page or swap ent will be stored(can be NULL)
4610 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
4611 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
4612 * move charge. if @target is not NULL, the page is stored in target->page
4613 * with extra refcnt got(Callers should handle it).
4614 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
4615 * target for charge migration. if @target is not NULL, the entry is stored
4618 * Called with pte lock held.
4625 enum mc_target_type
{
4626 MC_TARGET_NONE
, /* not used */
4631 static struct page
*mc_handle_present_pte(struct vm_area_struct
*vma
,
4632 unsigned long addr
, pte_t ptent
)
4634 struct page
*page
= vm_normal_page(vma
, addr
, ptent
);
4636 if (!page
|| !page_mapped(page
))
4638 if (PageAnon(page
)) {
4639 /* we don't move shared anon */
4640 if (!move_anon() || page_mapcount(page
) > 2)
4642 } else if (!move_file())
4643 /* we ignore mapcount for file pages */
4645 if (!get_page_unless_zero(page
))
4651 static struct page
*mc_handle_swap_pte(struct vm_area_struct
*vma
,
4652 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
4655 struct page
*page
= NULL
;
4656 swp_entry_t ent
= pte_to_swp_entry(ptent
);
4658 if (!move_anon() || non_swap_entry(ent
))
4660 usage_count
= mem_cgroup_count_swap_user(ent
, &page
);
4661 if (usage_count
> 1) { /* we don't move shared anon */
4666 if (do_swap_account
)
4667 entry
->val
= ent
.val
;
4672 static struct page
*mc_handle_file_pte(struct vm_area_struct
*vma
,
4673 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
4675 struct page
*page
= NULL
;
4676 struct inode
*inode
;
4677 struct address_space
*mapping
;
4680 if (!vma
->vm_file
) /* anonymous vma */
4685 inode
= vma
->vm_file
->f_path
.dentry
->d_inode
;
4686 mapping
= vma
->vm_file
->f_mapping
;
4687 if (pte_none(ptent
))
4688 pgoff
= linear_page_index(vma
, addr
);
4689 else /* pte_file(ptent) is true */
4690 pgoff
= pte_to_pgoff(ptent
);
4692 /* page is moved even if it's not RSS of this task(page-faulted). */
4693 if (!mapping_cap_swap_backed(mapping
)) { /* normal file */
4694 page
= find_get_page(mapping
, pgoff
);
4695 } else { /* shmem/tmpfs file. we should take account of swap too. */
4697 mem_cgroup_get_shmem_target(inode
, pgoff
, &page
, &ent
);
4698 if (do_swap_account
)
4699 entry
->val
= ent
.val
;
4705 static int is_target_pte_for_mc(struct vm_area_struct
*vma
,
4706 unsigned long addr
, pte_t ptent
, union mc_target
*target
)
4708 struct page
*page
= NULL
;
4709 struct page_cgroup
*pc
;
4711 swp_entry_t ent
= { .val
= 0 };
4713 if (pte_present(ptent
))
4714 page
= mc_handle_present_pte(vma
, addr
, ptent
);
4715 else if (is_swap_pte(ptent
))
4716 page
= mc_handle_swap_pte(vma
, addr
, ptent
, &ent
);
4717 else if (pte_none(ptent
) || pte_file(ptent
))
4718 page
= mc_handle_file_pte(vma
, addr
, ptent
, &ent
);
4720 if (!page
&& !ent
.val
)
4723 pc
= lookup_page_cgroup(page
);
4725 * Do only loose check w/o page_cgroup lock.
4726 * mem_cgroup_move_account() checks the pc is valid or not under
4729 if (PageCgroupUsed(pc
) && pc
->mem_cgroup
== mc
.from
) {
4730 ret
= MC_TARGET_PAGE
;
4732 target
->page
= page
;
4734 if (!ret
|| !target
)
4737 /* There is a swap entry and a page doesn't exist or isn't charged */
4738 if (ent
.val
&& !ret
&&
4739 css_id(&mc
.from
->css
) == lookup_swap_cgroup(ent
)) {
4740 ret
= MC_TARGET_SWAP
;
4747 static int mem_cgroup_count_precharge_pte_range(pmd_t
*pmd
,
4748 unsigned long addr
, unsigned long end
,
4749 struct mm_walk
*walk
)
4751 struct vm_area_struct
*vma
= walk
->private;
4755 split_huge_page_pmd(walk
->mm
, pmd
);
4757 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
4758 for (; addr
!= end
; pte
++, addr
+= PAGE_SIZE
)
4759 if (is_target_pte_for_mc(vma
, addr
, *pte
, NULL
))
4760 mc
.precharge
++; /* increment precharge temporarily */
4761 pte_unmap_unlock(pte
- 1, ptl
);
4767 static unsigned long mem_cgroup_count_precharge(struct mm_struct
*mm
)
4769 unsigned long precharge
;
4770 struct vm_area_struct
*vma
;
4772 down_read(&mm
->mmap_sem
);
4773 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
4774 struct mm_walk mem_cgroup_count_precharge_walk
= {
4775 .pmd_entry
= mem_cgroup_count_precharge_pte_range
,
4779 if (is_vm_hugetlb_page(vma
))
4781 walk_page_range(vma
->vm_start
, vma
->vm_end
,
4782 &mem_cgroup_count_precharge_walk
);
4784 up_read(&mm
->mmap_sem
);
4786 precharge
= mc
.precharge
;
4792 static int mem_cgroup_precharge_mc(struct mm_struct
*mm
)
4794 unsigned long precharge
= mem_cgroup_count_precharge(mm
);
4796 VM_BUG_ON(mc
.moving_task
);
4797 mc
.moving_task
= current
;
4798 return mem_cgroup_do_precharge(precharge
);
4801 /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
4802 static void __mem_cgroup_clear_mc(void)
4804 struct mem_cgroup
*from
= mc
.from
;
4805 struct mem_cgroup
*to
= mc
.to
;
4807 /* we must uncharge all the leftover precharges from mc.to */
4809 __mem_cgroup_cancel_charge(mc
.to
, mc
.precharge
);
4813 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
4814 * we must uncharge here.
4816 if (mc
.moved_charge
) {
4817 __mem_cgroup_cancel_charge(mc
.from
, mc
.moved_charge
);
4818 mc
.moved_charge
= 0;
4820 /* we must fixup refcnts and charges */
4821 if (mc
.moved_swap
) {
4822 /* uncharge swap account from the old cgroup */
4823 if (!mem_cgroup_is_root(mc
.from
))
4824 res_counter_uncharge(&mc
.from
->memsw
,
4825 PAGE_SIZE
* mc
.moved_swap
);
4826 __mem_cgroup_put(mc
.from
, mc
.moved_swap
);
4828 if (!mem_cgroup_is_root(mc
.to
)) {
4830 * we charged both to->res and to->memsw, so we should
4833 res_counter_uncharge(&mc
.to
->res
,
4834 PAGE_SIZE
* mc
.moved_swap
);
4836 /* we've already done mem_cgroup_get(mc.to) */
4839 memcg_oom_recover(from
);
4840 memcg_oom_recover(to
);
4841 wake_up_all(&mc
.waitq
);
4844 static void mem_cgroup_clear_mc(void)
4846 struct mem_cgroup
*from
= mc
.from
;
4849 * we must clear moving_task before waking up waiters at the end of
4852 mc
.moving_task
= NULL
;
4853 __mem_cgroup_clear_mc();
4854 spin_lock(&mc
.lock
);
4857 spin_unlock(&mc
.lock
);
4858 mem_cgroup_end_move(from
);
4861 static int mem_cgroup_can_attach(struct cgroup_subsys
*ss
,
4862 struct cgroup
*cgroup
,
4863 struct task_struct
*p
,
4867 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgroup
);
4869 if (mem
->move_charge_at_immigrate
) {
4870 struct mm_struct
*mm
;
4871 struct mem_cgroup
*from
= mem_cgroup_from_task(p
);
4873 VM_BUG_ON(from
== mem
);
4875 mm
= get_task_mm(p
);
4878 /* We move charges only when we move a owner of the mm */
4879 if (mm
->owner
== p
) {
4882 VM_BUG_ON(mc
.precharge
);
4883 VM_BUG_ON(mc
.moved_charge
);
4884 VM_BUG_ON(mc
.moved_swap
);
4885 mem_cgroup_start_move(from
);
4886 spin_lock(&mc
.lock
);
4889 spin_unlock(&mc
.lock
);
4890 /* We set mc.moving_task later */
4892 ret
= mem_cgroup_precharge_mc(mm
);
4894 mem_cgroup_clear_mc();
4901 static void mem_cgroup_cancel_attach(struct cgroup_subsys
*ss
,
4902 struct cgroup
*cgroup
,
4903 struct task_struct
*p
,
4906 mem_cgroup_clear_mc();
4909 static int mem_cgroup_move_charge_pte_range(pmd_t
*pmd
,
4910 unsigned long addr
, unsigned long end
,
4911 struct mm_walk
*walk
)
4914 struct vm_area_struct
*vma
= walk
->private;
4918 split_huge_page_pmd(walk
->mm
, pmd
);
4920 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
4921 for (; addr
!= end
; addr
+= PAGE_SIZE
) {
4922 pte_t ptent
= *(pte
++);
4923 union mc_target target
;
4926 struct page_cgroup
*pc
;
4932 type
= is_target_pte_for_mc(vma
, addr
, ptent
, &target
);
4934 case MC_TARGET_PAGE
:
4936 if (isolate_lru_page(page
))
4938 pc
= lookup_page_cgroup(page
);
4939 if (!mem_cgroup_move_account(pc
,
4940 mc
.from
, mc
.to
, false, PAGE_SIZE
)) {
4942 /* we uncharge from mc.from later. */
4945 putback_lru_page(page
);
4946 put
: /* is_target_pte_for_mc() gets the page */
4949 case MC_TARGET_SWAP
:
4951 if (!mem_cgroup_move_swap_account(ent
,
4952 mc
.from
, mc
.to
, false)) {
4954 /* we fixup refcnts and charges later. */
4962 pte_unmap_unlock(pte
- 1, ptl
);
4967 * We have consumed all precharges we got in can_attach().
4968 * We try charge one by one, but don't do any additional
4969 * charges to mc.to if we have failed in charge once in attach()
4972 ret
= mem_cgroup_do_precharge(1);
4980 static void mem_cgroup_move_charge(struct mm_struct
*mm
)
4982 struct vm_area_struct
*vma
;
4984 lru_add_drain_all();
4986 if (unlikely(!down_read_trylock(&mm
->mmap_sem
))) {
4988 * Someone who are holding the mmap_sem might be waiting in
4989 * waitq. So we cancel all extra charges, wake up all waiters,
4990 * and retry. Because we cancel precharges, we might not be able
4991 * to move enough charges, but moving charge is a best-effort
4992 * feature anyway, so it wouldn't be a big problem.
4994 __mem_cgroup_clear_mc();
4998 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
5000 struct mm_walk mem_cgroup_move_charge_walk
= {
5001 .pmd_entry
= mem_cgroup_move_charge_pte_range
,
5005 if (is_vm_hugetlb_page(vma
))
5007 ret
= walk_page_range(vma
->vm_start
, vma
->vm_end
,
5008 &mem_cgroup_move_charge_walk
);
5011 * means we have consumed all precharges and failed in
5012 * doing additional charge. Just abandon here.
5016 up_read(&mm
->mmap_sem
);
5019 static void mem_cgroup_move_task(struct cgroup_subsys
*ss
,
5020 struct cgroup
*cont
,
5021 struct cgroup
*old_cont
,
5022 struct task_struct
*p
,
5025 struct mm_struct
*mm
;
5028 /* no need to move charge */
5031 mm
= get_task_mm(p
);
5033 mem_cgroup_move_charge(mm
);
5036 mem_cgroup_clear_mc();
5038 #else /* !CONFIG_MMU */
5039 static int mem_cgroup_can_attach(struct cgroup_subsys
*ss
,
5040 struct cgroup
*cgroup
,
5041 struct task_struct
*p
,
5046 static void mem_cgroup_cancel_attach(struct cgroup_subsys
*ss
,
5047 struct cgroup
*cgroup
,
5048 struct task_struct
*p
,
5052 static void mem_cgroup_move_task(struct cgroup_subsys
*ss
,
5053 struct cgroup
*cont
,
5054 struct cgroup
*old_cont
,
5055 struct task_struct
*p
,
5061 struct cgroup_subsys mem_cgroup_subsys
= {
5063 .subsys_id
= mem_cgroup_subsys_id
,
5064 .create
= mem_cgroup_create
,
5065 .pre_destroy
= mem_cgroup_pre_destroy
,
5066 .destroy
= mem_cgroup_destroy
,
5067 .populate
= mem_cgroup_populate
,
5068 .can_attach
= mem_cgroup_can_attach
,
5069 .cancel_attach
= mem_cgroup_cancel_attach
,
5070 .attach
= mem_cgroup_move_task
,
5075 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
5076 static int __init
enable_swap_account(char *s
)
5078 /* consider enabled if no parameter or 1 is given */
5079 if (!(*s
) || !strcmp(s
, "=1"))
5080 really_do_swap_account
= 1;
5081 else if (!strcmp(s
, "=0"))
5082 really_do_swap_account
= 0;
5085 __setup("swapaccount", enable_swap_account
);
5087 static int __init
disable_swap_account(char *s
)
5089 printk_once("noswapaccount is deprecated and will be removed in 2.6.40. Use swapaccount=0 instead\n");
5090 enable_swap_account("=0");
5093 __setup("noswapaccount", disable_swap_account
);