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 /* for encoding cft->private value on file */
333 #define _OOM_TYPE (2)
334 #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val))
335 #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff)
336 #define MEMFILE_ATTR(val) ((val) & 0xffff)
337 /* Used for OOM nofiier */
338 #define OOM_CONTROL (0)
341 * Reclaim flags for mem_cgroup_hierarchical_reclaim
343 #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
344 #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
345 #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
346 #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
347 #define MEM_CGROUP_RECLAIM_SOFT_BIT 0x2
348 #define MEM_CGROUP_RECLAIM_SOFT (1 << MEM_CGROUP_RECLAIM_SOFT_BIT)
350 static void mem_cgroup_get(struct mem_cgroup
*mem
);
351 static void mem_cgroup_put(struct mem_cgroup
*mem
);
352 static struct mem_cgroup
*parent_mem_cgroup(struct mem_cgroup
*mem
);
353 static void drain_all_stock_async(void);
355 static struct mem_cgroup_per_zone
*
356 mem_cgroup_zoneinfo(struct mem_cgroup
*mem
, int nid
, int zid
)
358 return &mem
->info
.nodeinfo
[nid
]->zoneinfo
[zid
];
361 struct cgroup_subsys_state
*mem_cgroup_css(struct mem_cgroup
*mem
)
366 static struct mem_cgroup_per_zone
*
367 page_cgroup_zoneinfo(struct mem_cgroup
*mem
, struct page
*page
)
369 int nid
= page_to_nid(page
);
370 int zid
= page_zonenum(page
);
372 return mem_cgroup_zoneinfo(mem
, nid
, zid
);
375 static struct mem_cgroup_tree_per_zone
*
376 soft_limit_tree_node_zone(int nid
, int zid
)
378 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
381 static struct mem_cgroup_tree_per_zone
*
382 soft_limit_tree_from_page(struct page
*page
)
384 int nid
= page_to_nid(page
);
385 int zid
= page_zonenum(page
);
387 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
391 __mem_cgroup_insert_exceeded(struct mem_cgroup
*mem
,
392 struct mem_cgroup_per_zone
*mz
,
393 struct mem_cgroup_tree_per_zone
*mctz
,
394 unsigned long long new_usage_in_excess
)
396 struct rb_node
**p
= &mctz
->rb_root
.rb_node
;
397 struct rb_node
*parent
= NULL
;
398 struct mem_cgroup_per_zone
*mz_node
;
403 mz
->usage_in_excess
= new_usage_in_excess
;
404 if (!mz
->usage_in_excess
)
408 mz_node
= rb_entry(parent
, struct mem_cgroup_per_zone
,
410 if (mz
->usage_in_excess
< mz_node
->usage_in_excess
)
413 * We can't avoid mem cgroups that are over their soft
414 * limit by the same amount
416 else if (mz
->usage_in_excess
>= mz_node
->usage_in_excess
)
419 rb_link_node(&mz
->tree_node
, parent
, p
);
420 rb_insert_color(&mz
->tree_node
, &mctz
->rb_root
);
425 __mem_cgroup_remove_exceeded(struct mem_cgroup
*mem
,
426 struct mem_cgroup_per_zone
*mz
,
427 struct mem_cgroup_tree_per_zone
*mctz
)
431 rb_erase(&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
)
440 spin_lock(&mctz
->lock
);
441 __mem_cgroup_remove_exceeded(mem
, mz
, mctz
);
442 spin_unlock(&mctz
->lock
);
446 static void mem_cgroup_update_tree(struct mem_cgroup
*mem
, struct page
*page
)
448 unsigned long long excess
;
449 struct mem_cgroup_per_zone
*mz
;
450 struct mem_cgroup_tree_per_zone
*mctz
;
451 int nid
= page_to_nid(page
);
452 int zid
= page_zonenum(page
);
453 mctz
= soft_limit_tree_from_page(page
);
456 * Necessary to update all ancestors when hierarchy is used.
457 * because their event counter is not touched.
459 for (; mem
; mem
= parent_mem_cgroup(mem
)) {
460 mz
= mem_cgroup_zoneinfo(mem
, nid
, zid
);
461 excess
= res_counter_soft_limit_excess(&mem
->res
);
463 * We have to update the tree if mz is on RB-tree or
464 * mem is over its softlimit.
466 if (excess
|| mz
->on_tree
) {
467 spin_lock(&mctz
->lock
);
468 /* if on-tree, remove it */
470 __mem_cgroup_remove_exceeded(mem
, mz
, mctz
);
472 * Insert again. mz->usage_in_excess will be updated.
473 * If excess is 0, no tree ops.
475 __mem_cgroup_insert_exceeded(mem
, mz
, mctz
, excess
);
476 spin_unlock(&mctz
->lock
);
481 static void mem_cgroup_remove_from_trees(struct mem_cgroup
*mem
)
484 struct mem_cgroup_per_zone
*mz
;
485 struct mem_cgroup_tree_per_zone
*mctz
;
487 for_each_node_state(node
, N_POSSIBLE
) {
488 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
489 mz
= mem_cgroup_zoneinfo(mem
, node
, zone
);
490 mctz
= soft_limit_tree_node_zone(node
, zone
);
491 mem_cgroup_remove_exceeded(mem
, mz
, mctz
);
496 static struct mem_cgroup_per_zone
*
497 __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
499 struct rb_node
*rightmost
= NULL
;
500 struct mem_cgroup_per_zone
*mz
;
504 rightmost
= rb_last(&mctz
->rb_root
);
506 goto done
; /* Nothing to reclaim from */
508 mz
= rb_entry(rightmost
, struct mem_cgroup_per_zone
, tree_node
);
510 * Remove the node now but someone else can add it back,
511 * we will to add it back at the end of reclaim to its correct
512 * position in the tree.
514 __mem_cgroup_remove_exceeded(mz
->mem
, mz
, mctz
);
515 if (!res_counter_soft_limit_excess(&mz
->mem
->res
) ||
516 !css_tryget(&mz
->mem
->css
))
522 static struct mem_cgroup_per_zone
*
523 mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
525 struct mem_cgroup_per_zone
*mz
;
527 spin_lock(&mctz
->lock
);
528 mz
= __mem_cgroup_largest_soft_limit_node(mctz
);
529 spin_unlock(&mctz
->lock
);
534 * Implementation Note: reading percpu statistics for memcg.
536 * Both of vmstat[] and percpu_counter has threshold and do periodic
537 * synchronization to implement "quick" read. There are trade-off between
538 * reading cost and precision of value. Then, we may have a chance to implement
539 * a periodic synchronizion of counter in memcg's counter.
541 * But this _read() function is used for user interface now. The user accounts
542 * memory usage by memory cgroup and he _always_ requires exact value because
543 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
544 * have to visit all online cpus and make sum. So, for now, unnecessary
545 * synchronization is not implemented. (just implemented for cpu hotplug)
547 * If there are kernel internal actions which can make use of some not-exact
548 * value, and reading all cpu value can be performance bottleneck in some
549 * common workload, threashold and synchonization as vmstat[] should be
552 static s64
mem_cgroup_read_stat(struct mem_cgroup
*mem
,
553 enum mem_cgroup_stat_index idx
)
559 for_each_online_cpu(cpu
)
560 val
+= per_cpu(mem
->stat
->count
[idx
], cpu
);
561 #ifdef CONFIG_HOTPLUG_CPU
562 spin_lock(&mem
->pcp_counter_lock
);
563 val
+= mem
->nocpu_base
.count
[idx
];
564 spin_unlock(&mem
->pcp_counter_lock
);
570 static s64
mem_cgroup_local_usage(struct mem_cgroup
*mem
)
574 ret
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_RSS
);
575 ret
+= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_CACHE
);
579 static void mem_cgroup_swap_statistics(struct mem_cgroup
*mem
,
582 int val
= (charge
) ? 1 : -1;
583 this_cpu_add(mem
->stat
->count
[MEM_CGROUP_STAT_SWAPOUT
], val
);
586 static void mem_cgroup_charge_statistics(struct mem_cgroup
*mem
,
587 bool file
, int nr_pages
)
592 __this_cpu_add(mem
->stat
->count
[MEM_CGROUP_STAT_CACHE
], nr_pages
);
594 __this_cpu_add(mem
->stat
->count
[MEM_CGROUP_STAT_RSS
], nr_pages
);
596 /* pagein of a big page is an event. So, ignore page size */
598 __this_cpu_inc(mem
->stat
->count
[MEM_CGROUP_STAT_PGPGIN_COUNT
]);
600 __this_cpu_inc(mem
->stat
->count
[MEM_CGROUP_STAT_PGPGOUT_COUNT
]);
601 nr_pages
= -nr_pages
; /* for event */
604 __this_cpu_add(mem
->stat
->count
[MEM_CGROUP_EVENTS
], nr_pages
);
609 static unsigned long mem_cgroup_get_local_zonestat(struct mem_cgroup
*mem
,
613 struct mem_cgroup_per_zone
*mz
;
616 for_each_online_node(nid
)
617 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
618 mz
= mem_cgroup_zoneinfo(mem
, nid
, zid
);
619 total
+= MEM_CGROUP_ZSTAT(mz
, idx
);
624 static bool __memcg_event_check(struct mem_cgroup
*mem
, int event_mask_shift
)
628 val
= this_cpu_read(mem
->stat
->count
[MEM_CGROUP_EVENTS
]);
630 return !(val
& ((1 << event_mask_shift
) - 1));
634 * Check events in order.
637 static void memcg_check_events(struct mem_cgroup
*mem
, struct page
*page
)
639 /* threshold event is triggered in finer grain than soft limit */
640 if (unlikely(__memcg_event_check(mem
, THRESHOLDS_EVENTS_THRESH
))) {
641 mem_cgroup_threshold(mem
);
642 if (unlikely(__memcg_event_check(mem
, SOFTLIMIT_EVENTS_THRESH
)))
643 mem_cgroup_update_tree(mem
, page
);
647 static struct mem_cgroup
*mem_cgroup_from_cont(struct cgroup
*cont
)
649 return container_of(cgroup_subsys_state(cont
,
650 mem_cgroup_subsys_id
), struct mem_cgroup
,
654 struct mem_cgroup
*mem_cgroup_from_task(struct task_struct
*p
)
657 * mm_update_next_owner() may clear mm->owner to NULL
658 * if it races with swapoff, page migration, etc.
659 * So this can be called with p == NULL.
664 return container_of(task_subsys_state(p
, mem_cgroup_subsys_id
),
665 struct mem_cgroup
, css
);
668 static struct mem_cgroup
*try_get_mem_cgroup_from_mm(struct mm_struct
*mm
)
670 struct mem_cgroup
*mem
= NULL
;
675 * Because we have no locks, mm->owner's may be being moved to other
676 * cgroup. We use css_tryget() here even if this looks
677 * pessimistic (rather than adding locks here).
681 mem
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
684 } while (!css_tryget(&mem
->css
));
689 /* The caller has to guarantee "mem" exists before calling this */
690 static struct mem_cgroup
*mem_cgroup_start_loop(struct mem_cgroup
*mem
)
692 struct cgroup_subsys_state
*css
;
695 if (!mem
) /* ROOT cgroup has the smallest ID */
696 return root_mem_cgroup
; /*css_put/get against root is ignored*/
697 if (!mem
->use_hierarchy
) {
698 if (css_tryget(&mem
->css
))
704 * searching a memory cgroup which has the smallest ID under given
705 * ROOT cgroup. (ID >= 1)
707 css
= css_get_next(&mem_cgroup_subsys
, 1, &mem
->css
, &found
);
708 if (css
&& css_tryget(css
))
709 mem
= container_of(css
, struct mem_cgroup
, css
);
716 static struct mem_cgroup
*mem_cgroup_get_next(struct mem_cgroup
*iter
,
717 struct mem_cgroup
*root
,
720 int nextid
= css_id(&iter
->css
) + 1;
723 struct cgroup_subsys_state
*css
;
725 hierarchy_used
= iter
->use_hierarchy
;
728 /* If no ROOT, walk all, ignore hierarchy */
729 if (!cond
|| (root
&& !hierarchy_used
))
733 root
= root_mem_cgroup
;
739 css
= css_get_next(&mem_cgroup_subsys
, nextid
,
741 if (css
&& css_tryget(css
))
742 iter
= container_of(css
, struct mem_cgroup
, css
);
744 /* If css is NULL, no more cgroups will be found */
746 } while (css
&& !iter
);
751 * for_eacn_mem_cgroup_tree() for visiting all cgroup under tree. Please
752 * be careful that "break" loop is not allowed. We have reference count.
753 * Instead of that modify "cond" to be false and "continue" to exit the loop.
755 #define for_each_mem_cgroup_tree_cond(iter, root, cond) \
756 for (iter = mem_cgroup_start_loop(root);\
758 iter = mem_cgroup_get_next(iter, root, cond))
760 #define for_each_mem_cgroup_tree(iter, root) \
761 for_each_mem_cgroup_tree_cond(iter, root, true)
763 #define for_each_mem_cgroup_all(iter) \
764 for_each_mem_cgroup_tree_cond(iter, NULL, true)
767 static inline bool mem_cgroup_is_root(struct mem_cgroup
*mem
)
769 return (mem
== root_mem_cgroup
);
773 * Following LRU functions are allowed to be used without PCG_LOCK.
774 * Operations are called by routine of global LRU independently from memcg.
775 * What we have to take care of here is validness of pc->mem_cgroup.
777 * Changes to pc->mem_cgroup happens when
780 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
781 * It is added to LRU before charge.
782 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
783 * When moving account, the page is not on LRU. It's isolated.
786 void mem_cgroup_del_lru_list(struct page
*page
, enum lru_list lru
)
788 struct page_cgroup
*pc
;
789 struct mem_cgroup_per_zone
*mz
;
791 if (mem_cgroup_disabled())
793 pc
= lookup_page_cgroup(page
);
794 /* can happen while we handle swapcache. */
795 if (!TestClearPageCgroupAcctLRU(pc
))
797 VM_BUG_ON(!pc
->mem_cgroup
);
799 * We don't check PCG_USED bit. It's cleared when the "page" is finally
800 * removed from global LRU.
802 mz
= page_cgroup_zoneinfo(pc
->mem_cgroup
, page
);
803 /* huge page split is done under lru_lock. so, we have no races. */
804 MEM_CGROUP_ZSTAT(mz
, lru
) -= 1 << compound_order(page
);
805 if (mem_cgroup_is_root(pc
->mem_cgroup
))
807 VM_BUG_ON(list_empty(&pc
->lru
));
808 list_del_init(&pc
->lru
);
811 void mem_cgroup_del_lru(struct page
*page
)
813 mem_cgroup_del_lru_list(page
, page_lru(page
));
817 * Writeback is about to end against a page which has been marked for immediate
818 * reclaim. If it still appears to be reclaimable, move it to the tail of the
821 void mem_cgroup_rotate_reclaimable_page(struct page
*page
)
823 struct mem_cgroup_per_zone
*mz
;
824 struct page_cgroup
*pc
;
825 enum lru_list lru
= page_lru(page
);
827 if (mem_cgroup_disabled())
830 pc
= lookup_page_cgroup(page
);
831 /* unused or root page is not rotated. */
832 if (!PageCgroupUsed(pc
))
834 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
836 if (mem_cgroup_is_root(pc
->mem_cgroup
))
838 mz
= page_cgroup_zoneinfo(pc
->mem_cgroup
, page
);
839 list_move_tail(&pc
->lru
, &mz
->lists
[lru
]);
842 void mem_cgroup_rotate_lru_list(struct page
*page
, enum lru_list lru
)
844 struct mem_cgroup_per_zone
*mz
;
845 struct page_cgroup
*pc
;
847 if (mem_cgroup_disabled())
850 pc
= lookup_page_cgroup(page
);
851 /* unused or root page is not rotated. */
852 if (!PageCgroupUsed(pc
))
854 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
856 if (mem_cgroup_is_root(pc
->mem_cgroup
))
858 mz
= page_cgroup_zoneinfo(pc
->mem_cgroup
, page
);
859 list_move(&pc
->lru
, &mz
->lists
[lru
]);
862 void mem_cgroup_add_lru_list(struct page
*page
, enum lru_list lru
)
864 struct page_cgroup
*pc
;
865 struct mem_cgroup_per_zone
*mz
;
867 if (mem_cgroup_disabled())
869 pc
= lookup_page_cgroup(page
);
870 VM_BUG_ON(PageCgroupAcctLRU(pc
));
871 if (!PageCgroupUsed(pc
))
873 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
875 mz
= page_cgroup_zoneinfo(pc
->mem_cgroup
, page
);
876 /* huge page split is done under lru_lock. so, we have no races. */
877 MEM_CGROUP_ZSTAT(mz
, lru
) += 1 << compound_order(page
);
878 SetPageCgroupAcctLRU(pc
);
879 if (mem_cgroup_is_root(pc
->mem_cgroup
))
881 list_add(&pc
->lru
, &mz
->lists
[lru
]);
885 * At handling SwapCache, pc->mem_cgroup may be changed while it's linked to
886 * lru because the page may.be reused after it's fully uncharged (because of
887 * SwapCache behavior).To handle that, unlink page_cgroup from LRU when charge
888 * it again. This function is only used to charge SwapCache. It's done under
889 * lock_page and expected that zone->lru_lock is never held.
891 static void mem_cgroup_lru_del_before_commit_swapcache(struct page
*page
)
894 struct zone
*zone
= page_zone(page
);
895 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
897 spin_lock_irqsave(&zone
->lru_lock
, flags
);
899 * Forget old LRU when this page_cgroup is *not* used. This Used bit
900 * is guarded by lock_page() because the page is SwapCache.
902 if (!PageCgroupUsed(pc
))
903 mem_cgroup_del_lru_list(page
, page_lru(page
));
904 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
907 static void mem_cgroup_lru_add_after_commit_swapcache(struct page
*page
)
910 struct zone
*zone
= page_zone(page
);
911 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
913 spin_lock_irqsave(&zone
->lru_lock
, flags
);
914 /* link when the page is linked to LRU but page_cgroup isn't */
915 if (PageLRU(page
) && !PageCgroupAcctLRU(pc
))
916 mem_cgroup_add_lru_list(page
, page_lru(page
));
917 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
921 void mem_cgroup_move_lists(struct page
*page
,
922 enum lru_list from
, enum lru_list to
)
924 if (mem_cgroup_disabled())
926 mem_cgroup_del_lru_list(page
, from
);
927 mem_cgroup_add_lru_list(page
, to
);
930 int task_in_mem_cgroup(struct task_struct
*task
, const struct mem_cgroup
*mem
)
933 struct mem_cgroup
*curr
= NULL
;
934 struct task_struct
*p
;
936 p
= find_lock_task_mm(task
);
939 curr
= try_get_mem_cgroup_from_mm(p
->mm
);
944 * We should check use_hierarchy of "mem" not "curr". Because checking
945 * use_hierarchy of "curr" here make this function true if hierarchy is
946 * enabled in "curr" and "curr" is a child of "mem" in *cgroup*
947 * hierarchy(even if use_hierarchy is disabled in "mem").
949 if (mem
->use_hierarchy
)
950 ret
= css_is_ancestor(&curr
->css
, &mem
->css
);
957 static int calc_inactive_ratio(struct mem_cgroup
*memcg
, unsigned long *present_pages
)
959 unsigned long active
;
960 unsigned long inactive
;
962 unsigned long inactive_ratio
;
964 inactive
= mem_cgroup_get_local_zonestat(memcg
, LRU_INACTIVE_ANON
);
965 active
= mem_cgroup_get_local_zonestat(memcg
, LRU_ACTIVE_ANON
);
967 gb
= (inactive
+ active
) >> (30 - PAGE_SHIFT
);
969 inactive_ratio
= int_sqrt(10 * gb
);
974 present_pages
[0] = inactive
;
975 present_pages
[1] = active
;
978 return inactive_ratio
;
981 int mem_cgroup_inactive_anon_is_low(struct mem_cgroup
*memcg
)
983 unsigned long active
;
984 unsigned long inactive
;
985 unsigned long present_pages
[2];
986 unsigned long inactive_ratio
;
988 inactive_ratio
= calc_inactive_ratio(memcg
, present_pages
);
990 inactive
= present_pages
[0];
991 active
= present_pages
[1];
993 if (inactive
* inactive_ratio
< active
)
999 int mem_cgroup_inactive_file_is_low(struct mem_cgroup
*memcg
)
1001 unsigned long active
;
1002 unsigned long inactive
;
1004 inactive
= mem_cgroup_get_local_zonestat(memcg
, LRU_INACTIVE_FILE
);
1005 active
= mem_cgroup_get_local_zonestat(memcg
, LRU_ACTIVE_FILE
);
1007 return (active
> inactive
);
1010 unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup
*memcg
,
1014 int nid
= zone_to_nid(zone
);
1015 int zid
= zone_idx(zone
);
1016 struct mem_cgroup_per_zone
*mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
1018 return MEM_CGROUP_ZSTAT(mz
, lru
);
1021 struct zone_reclaim_stat
*mem_cgroup_get_reclaim_stat(struct mem_cgroup
*memcg
,
1024 int nid
= zone_to_nid(zone
);
1025 int zid
= zone_idx(zone
);
1026 struct mem_cgroup_per_zone
*mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
1028 return &mz
->reclaim_stat
;
1031 struct zone_reclaim_stat
*
1032 mem_cgroup_get_reclaim_stat_from_page(struct page
*page
)
1034 struct page_cgroup
*pc
;
1035 struct mem_cgroup_per_zone
*mz
;
1037 if (mem_cgroup_disabled())
1040 pc
= lookup_page_cgroup(page
);
1041 if (!PageCgroupUsed(pc
))
1043 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
1045 mz
= page_cgroup_zoneinfo(pc
->mem_cgroup
, page
);
1049 return &mz
->reclaim_stat
;
1052 unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan
,
1053 struct list_head
*dst
,
1054 unsigned long *scanned
, int order
,
1055 int mode
, struct zone
*z
,
1056 struct mem_cgroup
*mem_cont
,
1057 int active
, int file
)
1059 unsigned long nr_taken
= 0;
1063 struct list_head
*src
;
1064 struct page_cgroup
*pc
, *tmp
;
1065 int nid
= zone_to_nid(z
);
1066 int zid
= zone_idx(z
);
1067 struct mem_cgroup_per_zone
*mz
;
1068 int lru
= LRU_FILE
* file
+ active
;
1072 mz
= mem_cgroup_zoneinfo(mem_cont
, nid
, zid
);
1073 src
= &mz
->lists
[lru
];
1076 list_for_each_entry_safe_reverse(pc
, tmp
, src
, lru
) {
1077 if (scan
>= nr_to_scan
)
1081 if (unlikely(!PageCgroupUsed(pc
)))
1083 if (unlikely(!PageLRU(page
)))
1087 ret
= __isolate_lru_page(page
, mode
, file
);
1090 list_move(&page
->lru
, dst
);
1091 mem_cgroup_del_lru(page
);
1092 nr_taken
+= hpage_nr_pages(page
);
1095 /* we don't affect global LRU but rotate in our LRU */
1096 mem_cgroup_rotate_lru_list(page
, page_lru(page
));
1105 trace_mm_vmscan_memcg_isolate(0, nr_to_scan
, scan
, nr_taken
,
1111 #define mem_cgroup_from_res_counter(counter, member) \
1112 container_of(counter, struct mem_cgroup, member)
1115 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
1116 * @mem: the memory cgroup
1118 * Returns the maximum amount of memory @mem can be charged with, in
1121 static unsigned long long mem_cgroup_margin(struct mem_cgroup
*mem
)
1123 unsigned long long margin
;
1125 margin
= res_counter_margin(&mem
->res
);
1126 if (do_swap_account
)
1127 margin
= min(margin
, res_counter_margin(&mem
->memsw
));
1131 static unsigned int get_swappiness(struct mem_cgroup
*memcg
)
1133 struct cgroup
*cgrp
= memcg
->css
.cgroup
;
1134 unsigned int swappiness
;
1137 if (cgrp
->parent
== NULL
)
1138 return vm_swappiness
;
1140 spin_lock(&memcg
->reclaim_param_lock
);
1141 swappiness
= memcg
->swappiness
;
1142 spin_unlock(&memcg
->reclaim_param_lock
);
1147 static void mem_cgroup_start_move(struct mem_cgroup
*mem
)
1152 spin_lock(&mem
->pcp_counter_lock
);
1153 for_each_online_cpu(cpu
)
1154 per_cpu(mem
->stat
->count
[MEM_CGROUP_ON_MOVE
], cpu
) += 1;
1155 mem
->nocpu_base
.count
[MEM_CGROUP_ON_MOVE
] += 1;
1156 spin_unlock(&mem
->pcp_counter_lock
);
1162 static void mem_cgroup_end_move(struct mem_cgroup
*mem
)
1169 spin_lock(&mem
->pcp_counter_lock
);
1170 for_each_online_cpu(cpu
)
1171 per_cpu(mem
->stat
->count
[MEM_CGROUP_ON_MOVE
], cpu
) -= 1;
1172 mem
->nocpu_base
.count
[MEM_CGROUP_ON_MOVE
] -= 1;
1173 spin_unlock(&mem
->pcp_counter_lock
);
1177 * 2 routines for checking "mem" is under move_account() or not.
1179 * mem_cgroup_stealed() - checking a cgroup is mc.from or not. This is used
1180 * for avoiding race in accounting. If true,
1181 * pc->mem_cgroup may be overwritten.
1183 * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or
1184 * under hierarchy of moving cgroups. This is for
1185 * waiting at hith-memory prressure caused by "move".
1188 static bool mem_cgroup_stealed(struct mem_cgroup
*mem
)
1190 VM_BUG_ON(!rcu_read_lock_held());
1191 return this_cpu_read(mem
->stat
->count
[MEM_CGROUP_ON_MOVE
]) > 0;
1194 static bool mem_cgroup_under_move(struct mem_cgroup
*mem
)
1196 struct mem_cgroup
*from
;
1197 struct mem_cgroup
*to
;
1200 * Unlike task_move routines, we access mc.to, mc.from not under
1201 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1203 spin_lock(&mc
.lock
);
1208 if (from
== mem
|| to
== mem
1209 || (mem
->use_hierarchy
&& css_is_ancestor(&from
->css
, &mem
->css
))
1210 || (mem
->use_hierarchy
&& css_is_ancestor(&to
->css
, &mem
->css
)))
1213 spin_unlock(&mc
.lock
);
1217 static bool mem_cgroup_wait_acct_move(struct mem_cgroup
*mem
)
1219 if (mc
.moving_task
&& current
!= mc
.moving_task
) {
1220 if (mem_cgroup_under_move(mem
)) {
1222 prepare_to_wait(&mc
.waitq
, &wait
, TASK_INTERRUPTIBLE
);
1223 /* moving charge context might have finished. */
1226 finish_wait(&mc
.waitq
, &wait
);
1234 * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode.
1235 * @memcg: The memory cgroup that went over limit
1236 * @p: Task that is going to be killed
1238 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1241 void mem_cgroup_print_oom_info(struct mem_cgroup
*memcg
, struct task_struct
*p
)
1243 struct cgroup
*task_cgrp
;
1244 struct cgroup
*mem_cgrp
;
1246 * Need a buffer in BSS, can't rely on allocations. The code relies
1247 * on the assumption that OOM is serialized for memory controller.
1248 * If this assumption is broken, revisit this code.
1250 static char memcg_name
[PATH_MAX
];
1259 mem_cgrp
= memcg
->css
.cgroup
;
1260 task_cgrp
= task_cgroup(p
, mem_cgroup_subsys_id
);
1262 ret
= cgroup_path(task_cgrp
, memcg_name
, PATH_MAX
);
1265 * Unfortunately, we are unable to convert to a useful name
1266 * But we'll still print out the usage information
1273 printk(KERN_INFO
"Task in %s killed", memcg_name
);
1276 ret
= cgroup_path(mem_cgrp
, memcg_name
, PATH_MAX
);
1284 * Continues from above, so we don't need an KERN_ level
1286 printk(KERN_CONT
" as a result of limit of %s\n", memcg_name
);
1289 printk(KERN_INFO
"memory: usage %llukB, limit %llukB, failcnt %llu\n",
1290 res_counter_read_u64(&memcg
->res
, RES_USAGE
) >> 10,
1291 res_counter_read_u64(&memcg
->res
, RES_LIMIT
) >> 10,
1292 res_counter_read_u64(&memcg
->res
, RES_FAILCNT
));
1293 printk(KERN_INFO
"memory+swap: usage %llukB, limit %llukB, "
1295 res_counter_read_u64(&memcg
->memsw
, RES_USAGE
) >> 10,
1296 res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
) >> 10,
1297 res_counter_read_u64(&memcg
->memsw
, RES_FAILCNT
));
1301 * This function returns the number of memcg under hierarchy tree. Returns
1302 * 1(self count) if no children.
1304 static int mem_cgroup_count_children(struct mem_cgroup
*mem
)
1307 struct mem_cgroup
*iter
;
1309 for_each_mem_cgroup_tree(iter
, mem
)
1315 * Return the memory (and swap, if configured) limit for a memcg.
1317 u64
mem_cgroup_get_limit(struct mem_cgroup
*memcg
)
1322 limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
1323 limit
+= total_swap_pages
<< PAGE_SHIFT
;
1325 memsw
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
1327 * If memsw is finite and limits the amount of swap space available
1328 * to this memcg, return that limit.
1330 return min(limit
, memsw
);
1334 * Visit the first child (need not be the first child as per the ordering
1335 * of the cgroup list, since we track last_scanned_child) of @mem and use
1336 * that to reclaim free pages from.
1338 static struct mem_cgroup
*
1339 mem_cgroup_select_victim(struct mem_cgroup
*root_mem
)
1341 struct mem_cgroup
*ret
= NULL
;
1342 struct cgroup_subsys_state
*css
;
1345 if (!root_mem
->use_hierarchy
) {
1346 css_get(&root_mem
->css
);
1352 nextid
= root_mem
->last_scanned_child
+ 1;
1353 css
= css_get_next(&mem_cgroup_subsys
, nextid
, &root_mem
->css
,
1355 if (css
&& css_tryget(css
))
1356 ret
= container_of(css
, struct mem_cgroup
, css
);
1359 /* Updates scanning parameter */
1360 spin_lock(&root_mem
->reclaim_param_lock
);
1362 /* this means start scan from ID:1 */
1363 root_mem
->last_scanned_child
= 0;
1365 root_mem
->last_scanned_child
= found
;
1366 spin_unlock(&root_mem
->reclaim_param_lock
);
1373 * Scan the hierarchy if needed to reclaim memory. We remember the last child
1374 * we reclaimed from, so that we don't end up penalizing one child extensively
1375 * based on its position in the children list.
1377 * root_mem is the original ancestor that we've been reclaim from.
1379 * We give up and return to the caller when we visit root_mem twice.
1380 * (other groups can be removed while we're walking....)
1382 * If shrink==true, for avoiding to free too much, this returns immedieately.
1384 static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup
*root_mem
,
1387 unsigned long reclaim_options
)
1389 struct mem_cgroup
*victim
;
1392 bool noswap
= reclaim_options
& MEM_CGROUP_RECLAIM_NOSWAP
;
1393 bool shrink
= reclaim_options
& MEM_CGROUP_RECLAIM_SHRINK
;
1394 bool check_soft
= reclaim_options
& MEM_CGROUP_RECLAIM_SOFT
;
1395 unsigned long excess
;
1397 excess
= res_counter_soft_limit_excess(&root_mem
->res
) >> PAGE_SHIFT
;
1399 /* If memsw_is_minimum==1, swap-out is of-no-use. */
1400 if (root_mem
->memsw_is_minimum
)
1404 victim
= mem_cgroup_select_victim(root_mem
);
1405 if (victim
== root_mem
) {
1408 drain_all_stock_async();
1411 * If we have not been able to reclaim
1412 * anything, it might because there are
1413 * no reclaimable pages under this hierarchy
1415 if (!check_soft
|| !total
) {
1416 css_put(&victim
->css
);
1420 * We want to do more targetted reclaim.
1421 * excess >> 2 is not to excessive so as to
1422 * reclaim too much, nor too less that we keep
1423 * coming back to reclaim from this cgroup
1425 if (total
>= (excess
>> 2) ||
1426 (loop
> MEM_CGROUP_MAX_RECLAIM_LOOPS
)) {
1427 css_put(&victim
->css
);
1432 if (!mem_cgroup_local_usage(victim
)) {
1433 /* this cgroup's local usage == 0 */
1434 css_put(&victim
->css
);
1437 /* we use swappiness of local cgroup */
1439 ret
= mem_cgroup_shrink_node_zone(victim
, gfp_mask
,
1440 noswap
, get_swappiness(victim
), zone
);
1442 ret
= try_to_free_mem_cgroup_pages(victim
, gfp_mask
,
1443 noswap
, get_swappiness(victim
));
1444 css_put(&victim
->css
);
1446 * At shrinking usage, we can't check we should stop here or
1447 * reclaim more. It's depends on callers. last_scanned_child
1448 * will work enough for keeping fairness under tree.
1454 if (!res_counter_soft_limit_excess(&root_mem
->res
))
1456 } else if (mem_cgroup_margin(root_mem
))
1463 * Check OOM-Killer is already running under our hierarchy.
1464 * If someone is running, return false.
1466 static bool mem_cgroup_oom_lock(struct mem_cgroup
*mem
)
1468 int x
, lock_count
= 0;
1469 struct mem_cgroup
*iter
;
1471 for_each_mem_cgroup_tree(iter
, mem
) {
1472 x
= atomic_inc_return(&iter
->oom_lock
);
1473 lock_count
= max(x
, lock_count
);
1476 if (lock_count
== 1)
1481 static int mem_cgroup_oom_unlock(struct mem_cgroup
*mem
)
1483 struct mem_cgroup
*iter
;
1486 * When a new child is created while the hierarchy is under oom,
1487 * mem_cgroup_oom_lock() may not be called. We have to use
1488 * atomic_add_unless() here.
1490 for_each_mem_cgroup_tree(iter
, mem
)
1491 atomic_add_unless(&iter
->oom_lock
, -1, 0);
1496 static DEFINE_MUTEX(memcg_oom_mutex
);
1497 static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq
);
1499 struct oom_wait_info
{
1500 struct mem_cgroup
*mem
;
1504 static int memcg_oom_wake_function(wait_queue_t
*wait
,
1505 unsigned mode
, int sync
, void *arg
)
1507 struct mem_cgroup
*wake_mem
= (struct mem_cgroup
*)arg
;
1508 struct oom_wait_info
*oom_wait_info
;
1510 oom_wait_info
= container_of(wait
, struct oom_wait_info
, wait
);
1512 if (oom_wait_info
->mem
== wake_mem
)
1514 /* if no hierarchy, no match */
1515 if (!oom_wait_info
->mem
->use_hierarchy
|| !wake_mem
->use_hierarchy
)
1518 * Both of oom_wait_info->mem and wake_mem are stable under us.
1519 * Then we can use css_is_ancestor without taking care of RCU.
1521 if (!css_is_ancestor(&oom_wait_info
->mem
->css
, &wake_mem
->css
) &&
1522 !css_is_ancestor(&wake_mem
->css
, &oom_wait_info
->mem
->css
))
1526 return autoremove_wake_function(wait
, mode
, sync
, arg
);
1529 static void memcg_wakeup_oom(struct mem_cgroup
*mem
)
1531 /* for filtering, pass "mem" as argument. */
1532 __wake_up(&memcg_oom_waitq
, TASK_NORMAL
, 0, mem
);
1535 static void memcg_oom_recover(struct mem_cgroup
*mem
)
1537 if (mem
&& atomic_read(&mem
->oom_lock
))
1538 memcg_wakeup_oom(mem
);
1542 * try to call OOM killer. returns false if we should exit memory-reclaim loop.
1544 bool mem_cgroup_handle_oom(struct mem_cgroup
*mem
, gfp_t mask
)
1546 struct oom_wait_info owait
;
1547 bool locked
, need_to_kill
;
1550 owait
.wait
.flags
= 0;
1551 owait
.wait
.func
= memcg_oom_wake_function
;
1552 owait
.wait
.private = current
;
1553 INIT_LIST_HEAD(&owait
.wait
.task_list
);
1554 need_to_kill
= true;
1555 /* At first, try to OOM lock hierarchy under mem.*/
1556 mutex_lock(&memcg_oom_mutex
);
1557 locked
= mem_cgroup_oom_lock(mem
);
1559 * Even if signal_pending(), we can't quit charge() loop without
1560 * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
1561 * under OOM is always welcomed, use TASK_KILLABLE here.
1563 prepare_to_wait(&memcg_oom_waitq
, &owait
.wait
, TASK_KILLABLE
);
1564 if (!locked
|| mem
->oom_kill_disable
)
1565 need_to_kill
= false;
1567 mem_cgroup_oom_notify(mem
);
1568 mutex_unlock(&memcg_oom_mutex
);
1571 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1572 mem_cgroup_out_of_memory(mem
, mask
);
1575 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1577 mutex_lock(&memcg_oom_mutex
);
1578 mem_cgroup_oom_unlock(mem
);
1579 memcg_wakeup_oom(mem
);
1580 mutex_unlock(&memcg_oom_mutex
);
1582 if (test_thread_flag(TIF_MEMDIE
) || fatal_signal_pending(current
))
1584 /* Give chance to dying process */
1585 schedule_timeout(1);
1590 * Currently used to update mapped file statistics, but the routine can be
1591 * generalized to update other statistics as well.
1593 * Notes: Race condition
1595 * We usually use page_cgroup_lock() for accessing page_cgroup member but
1596 * it tends to be costly. But considering some conditions, we doesn't need
1597 * to do so _always_.
1599 * Considering "charge", lock_page_cgroup() is not required because all
1600 * file-stat operations happen after a page is attached to radix-tree. There
1601 * are no race with "charge".
1603 * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup
1604 * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even
1605 * if there are race with "uncharge". Statistics itself is properly handled
1608 * Considering "move", this is an only case we see a race. To make the race
1609 * small, we check MEM_CGROUP_ON_MOVE percpu value and detect there are
1610 * possibility of race condition. If there is, we take a lock.
1613 void mem_cgroup_update_page_stat(struct page
*page
,
1614 enum mem_cgroup_page_stat_item idx
, int val
)
1616 struct mem_cgroup
*mem
;
1617 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
1618 bool need_unlock
= false;
1619 unsigned long uninitialized_var(flags
);
1625 mem
= pc
->mem_cgroup
;
1626 if (unlikely(!mem
|| !PageCgroupUsed(pc
)))
1628 /* pc->mem_cgroup is unstable ? */
1629 if (unlikely(mem_cgroup_stealed(mem
)) || PageTransHuge(page
)) {
1630 /* take a lock against to access pc->mem_cgroup */
1631 move_lock_page_cgroup(pc
, &flags
);
1633 mem
= pc
->mem_cgroup
;
1634 if (!mem
|| !PageCgroupUsed(pc
))
1639 case MEMCG_NR_FILE_MAPPED
:
1641 SetPageCgroupFileMapped(pc
);
1642 else if (!page_mapped(page
))
1643 ClearPageCgroupFileMapped(pc
);
1644 idx
= MEM_CGROUP_STAT_FILE_MAPPED
;
1650 this_cpu_add(mem
->stat
->count
[idx
], val
);
1653 if (unlikely(need_unlock
))
1654 move_unlock_page_cgroup(pc
, &flags
);
1658 EXPORT_SYMBOL(mem_cgroup_update_page_stat
);
1661 * size of first charge trial. "32" comes from vmscan.c's magic value.
1662 * TODO: maybe necessary to use big numbers in big irons.
1664 #define CHARGE_SIZE (32 * PAGE_SIZE)
1665 struct memcg_stock_pcp
{
1666 struct mem_cgroup
*cached
; /* this never be root cgroup */
1668 struct work_struct work
;
1670 static DEFINE_PER_CPU(struct memcg_stock_pcp
, memcg_stock
);
1671 static atomic_t memcg_drain_count
;
1674 * Try to consume stocked charge on this cpu. If success, PAGE_SIZE is consumed
1675 * from local stock and true is returned. If the stock is 0 or charges from a
1676 * cgroup which is not current target, returns false. This stock will be
1679 static bool consume_stock(struct mem_cgroup
*mem
)
1681 struct memcg_stock_pcp
*stock
;
1684 stock
= &get_cpu_var(memcg_stock
);
1685 if (mem
== stock
->cached
&& stock
->charge
)
1686 stock
->charge
-= PAGE_SIZE
;
1687 else /* need to call res_counter_charge */
1689 put_cpu_var(memcg_stock
);
1694 * Returns stocks cached in percpu to res_counter and reset cached information.
1696 static void drain_stock(struct memcg_stock_pcp
*stock
)
1698 struct mem_cgroup
*old
= stock
->cached
;
1700 if (stock
->charge
) {
1701 res_counter_uncharge(&old
->res
, stock
->charge
);
1702 if (do_swap_account
)
1703 res_counter_uncharge(&old
->memsw
, stock
->charge
);
1705 stock
->cached
= NULL
;
1710 * This must be called under preempt disabled or must be called by
1711 * a thread which is pinned to local cpu.
1713 static void drain_local_stock(struct work_struct
*dummy
)
1715 struct memcg_stock_pcp
*stock
= &__get_cpu_var(memcg_stock
);
1720 * Cache charges(val) which is from res_counter, to local per_cpu area.
1721 * This will be consumed by consume_stock() function, later.
1723 static void refill_stock(struct mem_cgroup
*mem
, int val
)
1725 struct memcg_stock_pcp
*stock
= &get_cpu_var(memcg_stock
);
1727 if (stock
->cached
!= mem
) { /* reset if necessary */
1729 stock
->cached
= mem
;
1731 stock
->charge
+= val
;
1732 put_cpu_var(memcg_stock
);
1736 * Tries to drain stocked charges in other cpus. This function is asynchronous
1737 * and just put a work per cpu for draining localy on each cpu. Caller can
1738 * expects some charges will be back to res_counter later but cannot wait for
1741 static void drain_all_stock_async(void)
1744 /* This function is for scheduling "drain" in asynchronous way.
1745 * The result of "drain" is not directly handled by callers. Then,
1746 * if someone is calling drain, we don't have to call drain more.
1747 * Anyway, WORK_STRUCT_PENDING check in queue_work_on() will catch if
1748 * there is a race. We just do loose check here.
1750 if (atomic_read(&memcg_drain_count
))
1752 /* Notify other cpus that system-wide "drain" is running */
1753 atomic_inc(&memcg_drain_count
);
1755 for_each_online_cpu(cpu
) {
1756 struct memcg_stock_pcp
*stock
= &per_cpu(memcg_stock
, cpu
);
1757 schedule_work_on(cpu
, &stock
->work
);
1760 atomic_dec(&memcg_drain_count
);
1761 /* We don't wait for flush_work */
1764 /* This is a synchronous drain interface. */
1765 static void drain_all_stock_sync(void)
1767 /* called when force_empty is called */
1768 atomic_inc(&memcg_drain_count
);
1769 schedule_on_each_cpu(drain_local_stock
);
1770 atomic_dec(&memcg_drain_count
);
1774 * This function drains percpu counter value from DEAD cpu and
1775 * move it to local cpu. Note that this function can be preempted.
1777 static void mem_cgroup_drain_pcp_counter(struct mem_cgroup
*mem
, int cpu
)
1781 spin_lock(&mem
->pcp_counter_lock
);
1782 for (i
= 0; i
< MEM_CGROUP_STAT_DATA
; i
++) {
1783 s64 x
= per_cpu(mem
->stat
->count
[i
], cpu
);
1785 per_cpu(mem
->stat
->count
[i
], cpu
) = 0;
1786 mem
->nocpu_base
.count
[i
] += x
;
1788 /* need to clear ON_MOVE value, works as a kind of lock. */
1789 per_cpu(mem
->stat
->count
[MEM_CGROUP_ON_MOVE
], cpu
) = 0;
1790 spin_unlock(&mem
->pcp_counter_lock
);
1793 static void synchronize_mem_cgroup_on_move(struct mem_cgroup
*mem
, int cpu
)
1795 int idx
= MEM_CGROUP_ON_MOVE
;
1797 spin_lock(&mem
->pcp_counter_lock
);
1798 per_cpu(mem
->stat
->count
[idx
], cpu
) = mem
->nocpu_base
.count
[idx
];
1799 spin_unlock(&mem
->pcp_counter_lock
);
1802 static int __cpuinit
memcg_cpu_hotplug_callback(struct notifier_block
*nb
,
1803 unsigned long action
,
1806 int cpu
= (unsigned long)hcpu
;
1807 struct memcg_stock_pcp
*stock
;
1808 struct mem_cgroup
*iter
;
1810 if ((action
== CPU_ONLINE
)) {
1811 for_each_mem_cgroup_all(iter
)
1812 synchronize_mem_cgroup_on_move(iter
, cpu
);
1816 if ((action
!= CPU_DEAD
) || action
!= CPU_DEAD_FROZEN
)
1819 for_each_mem_cgroup_all(iter
)
1820 mem_cgroup_drain_pcp_counter(iter
, cpu
);
1822 stock
= &per_cpu(memcg_stock
, cpu
);
1828 /* See __mem_cgroup_try_charge() for details */
1830 CHARGE_OK
, /* success */
1831 CHARGE_RETRY
, /* need to retry but retry is not bad */
1832 CHARGE_NOMEM
, /* we can't do more. return -ENOMEM */
1833 CHARGE_WOULDBLOCK
, /* GFP_WAIT wasn't set and no enough res. */
1834 CHARGE_OOM_DIE
, /* the current is killed because of OOM */
1837 static int __mem_cgroup_do_charge(struct mem_cgroup
*mem
, gfp_t gfp_mask
,
1838 int csize
, bool oom_check
)
1840 struct mem_cgroup
*mem_over_limit
;
1841 struct res_counter
*fail_res
;
1842 unsigned long flags
= 0;
1845 ret
= res_counter_charge(&mem
->res
, csize
, &fail_res
);
1848 if (!do_swap_account
)
1850 ret
= res_counter_charge(&mem
->memsw
, csize
, &fail_res
);
1854 res_counter_uncharge(&mem
->res
, csize
);
1855 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
, memsw
);
1856 flags
|= MEM_CGROUP_RECLAIM_NOSWAP
;
1858 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
, res
);
1860 * csize can be either a huge page (HPAGE_SIZE), a batch of
1861 * regular pages (CHARGE_SIZE), or a single regular page
1864 * Never reclaim on behalf of optional batching, retry with a
1865 * single page instead.
1867 if (csize
== CHARGE_SIZE
)
1868 return CHARGE_RETRY
;
1870 if (!(gfp_mask
& __GFP_WAIT
))
1871 return CHARGE_WOULDBLOCK
;
1873 ret
= mem_cgroup_hierarchical_reclaim(mem_over_limit
, NULL
,
1875 if (mem_cgroup_margin(mem_over_limit
) >= csize
)
1876 return CHARGE_RETRY
;
1878 * Even though the limit is exceeded at this point, reclaim
1879 * may have been able to free some pages. Retry the charge
1880 * before killing the task.
1882 * Only for regular pages, though: huge pages are rather
1883 * unlikely to succeed so close to the limit, and we fall back
1884 * to regular pages anyway in case of failure.
1886 if (csize
== PAGE_SIZE
&& ret
)
1887 return CHARGE_RETRY
;
1890 * At task move, charge accounts can be doubly counted. So, it's
1891 * better to wait until the end of task_move if something is going on.
1893 if (mem_cgroup_wait_acct_move(mem_over_limit
))
1894 return CHARGE_RETRY
;
1896 /* If we don't need to call oom-killer at el, return immediately */
1898 return CHARGE_NOMEM
;
1900 if (!mem_cgroup_handle_oom(mem_over_limit
, gfp_mask
))
1901 return CHARGE_OOM_DIE
;
1903 return CHARGE_RETRY
;
1907 * Unlike exported interface, "oom" parameter is added. if oom==true,
1908 * oom-killer can be invoked.
1910 static int __mem_cgroup_try_charge(struct mm_struct
*mm
,
1912 struct mem_cgroup
**memcg
, bool oom
,
1915 int nr_oom_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
1916 struct mem_cgroup
*mem
= NULL
;
1918 int csize
= max(CHARGE_SIZE
, (unsigned long) page_size
);
1921 * Unlike gloval-vm's OOM-kill, we're not in memory shortage
1922 * in system level. So, allow to go ahead dying process in addition to
1925 if (unlikely(test_thread_flag(TIF_MEMDIE
)
1926 || fatal_signal_pending(current
)))
1930 * We always charge the cgroup the mm_struct belongs to.
1931 * The mm_struct's mem_cgroup changes on task migration if the
1932 * thread group leader migrates. It's possible that mm is not
1933 * set, if so charge the init_mm (happens for pagecache usage).
1938 if (*memcg
) { /* css should be a valid one */
1940 VM_BUG_ON(css_is_removed(&mem
->css
));
1941 if (mem_cgroup_is_root(mem
))
1943 if (page_size
== PAGE_SIZE
&& consume_stock(mem
))
1947 struct task_struct
*p
;
1950 p
= rcu_dereference(mm
->owner
);
1952 * Because we don't have task_lock(), "p" can exit.
1953 * In that case, "mem" can point to root or p can be NULL with
1954 * race with swapoff. Then, we have small risk of mis-accouning.
1955 * But such kind of mis-account by race always happens because
1956 * we don't have cgroup_mutex(). It's overkill and we allo that
1958 * (*) swapoff at el will charge against mm-struct not against
1959 * task-struct. So, mm->owner can be NULL.
1961 mem
= mem_cgroup_from_task(p
);
1962 if (!mem
|| mem_cgroup_is_root(mem
)) {
1966 if (page_size
== PAGE_SIZE
&& consume_stock(mem
)) {
1968 * It seems dagerous to access memcg without css_get().
1969 * But considering how consume_stok works, it's not
1970 * necessary. If consume_stock success, some charges
1971 * from this memcg are cached on this cpu. So, we
1972 * don't need to call css_get()/css_tryget() before
1973 * calling consume_stock().
1978 /* after here, we may be blocked. we need to get refcnt */
1979 if (!css_tryget(&mem
->css
)) {
1989 /* If killed, bypass charge */
1990 if (fatal_signal_pending(current
)) {
1996 if (oom
&& !nr_oom_retries
) {
1998 nr_oom_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
2001 ret
= __mem_cgroup_do_charge(mem
, gfp_mask
, csize
, oom_check
);
2006 case CHARGE_RETRY
: /* not in OOM situation but retry */
2011 case CHARGE_WOULDBLOCK
: /* !__GFP_WAIT */
2014 case CHARGE_NOMEM
: /* OOM routine works */
2019 /* If oom, we never return -ENOMEM */
2022 case CHARGE_OOM_DIE
: /* Killed by OOM Killer */
2026 } while (ret
!= CHARGE_OK
);
2028 if (csize
> page_size
)
2029 refill_stock(mem
, csize
- page_size
);
2043 * Somemtimes we have to undo a charge we got by try_charge().
2044 * This function is for that and do uncharge, put css's refcnt.
2045 * gotten by try_charge().
2047 static void __mem_cgroup_cancel_charge(struct mem_cgroup
*mem
,
2048 unsigned long count
)
2050 if (!mem_cgroup_is_root(mem
)) {
2051 res_counter_uncharge(&mem
->res
, PAGE_SIZE
* count
);
2052 if (do_swap_account
)
2053 res_counter_uncharge(&mem
->memsw
, PAGE_SIZE
* count
);
2057 static void mem_cgroup_cancel_charge(struct mem_cgroup
*mem
,
2060 __mem_cgroup_cancel_charge(mem
, page_size
>> PAGE_SHIFT
);
2064 * A helper function to get mem_cgroup from ID. must be called under
2065 * rcu_read_lock(). The caller must check css_is_removed() or some if
2066 * it's concern. (dropping refcnt from swap can be called against removed
2069 static struct mem_cgroup
*mem_cgroup_lookup(unsigned short id
)
2071 struct cgroup_subsys_state
*css
;
2073 /* ID 0 is unused ID */
2076 css
= css_lookup(&mem_cgroup_subsys
, id
);
2079 return container_of(css
, struct mem_cgroup
, css
);
2082 struct mem_cgroup
*try_get_mem_cgroup_from_page(struct page
*page
)
2084 struct mem_cgroup
*mem
= NULL
;
2085 struct page_cgroup
*pc
;
2089 VM_BUG_ON(!PageLocked(page
));
2091 pc
= lookup_page_cgroup(page
);
2092 lock_page_cgroup(pc
);
2093 if (PageCgroupUsed(pc
)) {
2094 mem
= pc
->mem_cgroup
;
2095 if (mem
&& !css_tryget(&mem
->css
))
2097 } else if (PageSwapCache(page
)) {
2098 ent
.val
= page_private(page
);
2099 id
= lookup_swap_cgroup(ent
);
2101 mem
= mem_cgroup_lookup(id
);
2102 if (mem
&& !css_tryget(&mem
->css
))
2106 unlock_page_cgroup(pc
);
2110 static void __mem_cgroup_commit_charge(struct mem_cgroup
*mem
,
2111 struct page_cgroup
*pc
,
2112 enum charge_type ctype
,
2115 int nr_pages
= page_size
>> PAGE_SHIFT
;
2117 lock_page_cgroup(pc
);
2118 if (unlikely(PageCgroupUsed(pc
))) {
2119 unlock_page_cgroup(pc
);
2120 mem_cgroup_cancel_charge(mem
, page_size
);
2124 * we don't need page_cgroup_lock about tail pages, becase they are not
2125 * accessed by any other context at this point.
2127 pc
->mem_cgroup
= mem
;
2129 * We access a page_cgroup asynchronously without lock_page_cgroup().
2130 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
2131 * is accessed after testing USED bit. To make pc->mem_cgroup visible
2132 * before USED bit, we need memory barrier here.
2133 * See mem_cgroup_add_lru_list(), etc.
2137 case MEM_CGROUP_CHARGE_TYPE_CACHE
:
2138 case MEM_CGROUP_CHARGE_TYPE_SHMEM
:
2139 SetPageCgroupCache(pc
);
2140 SetPageCgroupUsed(pc
);
2142 case MEM_CGROUP_CHARGE_TYPE_MAPPED
:
2143 ClearPageCgroupCache(pc
);
2144 SetPageCgroupUsed(pc
);
2150 mem_cgroup_charge_statistics(mem
, PageCgroupCache(pc
), nr_pages
);
2151 unlock_page_cgroup(pc
);
2153 * "charge_statistics" updated event counter. Then, check it.
2154 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
2155 * if they exceeds softlimit.
2157 memcg_check_events(mem
, pc
->page
);
2160 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2162 #define PCGF_NOCOPY_AT_SPLIT ((1 << PCG_LOCK) | (1 << PCG_MOVE_LOCK) |\
2163 (1 << PCG_ACCT_LRU) | (1 << PCG_MIGRATION))
2165 * Because tail pages are not marked as "used", set it. We're under
2166 * zone->lru_lock, 'splitting on pmd' and compund_lock.
2168 void mem_cgroup_split_huge_fixup(struct page
*head
, struct page
*tail
)
2170 struct page_cgroup
*head_pc
= lookup_page_cgroup(head
);
2171 struct page_cgroup
*tail_pc
= lookup_page_cgroup(tail
);
2172 unsigned long flags
;
2174 if (mem_cgroup_disabled())
2177 * We have no races with charge/uncharge but will have races with
2178 * page state accounting.
2180 move_lock_page_cgroup(head_pc
, &flags
);
2182 tail_pc
->mem_cgroup
= head_pc
->mem_cgroup
;
2183 smp_wmb(); /* see __commit_charge() */
2184 if (PageCgroupAcctLRU(head_pc
)) {
2186 struct mem_cgroup_per_zone
*mz
;
2189 * LRU flags cannot be copied because we need to add tail
2190 *.page to LRU by generic call and our hook will be called.
2191 * We hold lru_lock, then, reduce counter directly.
2193 lru
= page_lru(head
);
2194 mz
= page_cgroup_zoneinfo(head_pc
->mem_cgroup
, head
);
2195 MEM_CGROUP_ZSTAT(mz
, lru
) -= 1;
2197 tail_pc
->flags
= head_pc
->flags
& ~PCGF_NOCOPY_AT_SPLIT
;
2198 move_unlock_page_cgroup(head_pc
, &flags
);
2203 * __mem_cgroup_move_account - move account of the page
2204 * @pc: page_cgroup of the page.
2205 * @from: mem_cgroup which the page is moved from.
2206 * @to: mem_cgroup which the page is moved to. @from != @to.
2207 * @uncharge: whether we should call uncharge and css_put against @from.
2209 * The caller must confirm following.
2210 * - page is not on LRU (isolate_page() is useful.)
2211 * - the pc is locked, used, and ->mem_cgroup points to @from.
2213 * This function doesn't do "charge" nor css_get to new cgroup. It should be
2214 * done by a caller(__mem_cgroup_try_charge would be usefull). If @uncharge is
2215 * true, this function does "uncharge" from old cgroup, but it doesn't if
2216 * @uncharge is false, so a caller should do "uncharge".
2219 static void __mem_cgroup_move_account(struct page_cgroup
*pc
,
2220 struct mem_cgroup
*from
, struct mem_cgroup
*to
, bool uncharge
,
2223 int nr_pages
= charge_size
>> PAGE_SHIFT
;
2225 VM_BUG_ON(from
== to
);
2226 VM_BUG_ON(PageLRU(pc
->page
));
2227 VM_BUG_ON(!page_is_cgroup_locked(pc
));
2228 VM_BUG_ON(!PageCgroupUsed(pc
));
2229 VM_BUG_ON(pc
->mem_cgroup
!= from
);
2231 if (PageCgroupFileMapped(pc
)) {
2232 /* Update mapped_file data for mem_cgroup */
2234 __this_cpu_dec(from
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
2235 __this_cpu_inc(to
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
2238 mem_cgroup_charge_statistics(from
, PageCgroupCache(pc
), -nr_pages
);
2240 /* This is not "cancel", but cancel_charge does all we need. */
2241 mem_cgroup_cancel_charge(from
, charge_size
);
2243 /* caller should have done css_get */
2244 pc
->mem_cgroup
= to
;
2245 mem_cgroup_charge_statistics(to
, PageCgroupCache(pc
), nr_pages
);
2247 * We charges against "to" which may not have any tasks. Then, "to"
2248 * can be under rmdir(). But in current implementation, caller of
2249 * this function is just force_empty() and move charge, so it's
2250 * garanteed that "to" is never removed. So, we don't check rmdir
2256 * check whether the @pc is valid for moving account and call
2257 * __mem_cgroup_move_account()
2259 static int mem_cgroup_move_account(struct page_cgroup
*pc
,
2260 struct mem_cgroup
*from
, struct mem_cgroup
*to
,
2261 bool uncharge
, int charge_size
)
2264 unsigned long flags
;
2266 * The page is isolated from LRU. So, collapse function
2267 * will not handle this page. But page splitting can happen.
2268 * Do this check under compound_page_lock(). The caller should
2271 if ((charge_size
> PAGE_SIZE
) && !PageTransHuge(pc
->page
))
2274 lock_page_cgroup(pc
);
2275 if (PageCgroupUsed(pc
) && pc
->mem_cgroup
== from
) {
2276 move_lock_page_cgroup(pc
, &flags
);
2277 __mem_cgroup_move_account(pc
, from
, to
, uncharge
, charge_size
);
2278 move_unlock_page_cgroup(pc
, &flags
);
2281 unlock_page_cgroup(pc
);
2285 memcg_check_events(to
, pc
->page
);
2286 memcg_check_events(from
, pc
->page
);
2291 * move charges to its parent.
2294 static int mem_cgroup_move_parent(struct page_cgroup
*pc
,
2295 struct mem_cgroup
*child
,
2298 struct page
*page
= pc
->page
;
2299 struct cgroup
*cg
= child
->css
.cgroup
;
2300 struct cgroup
*pcg
= cg
->parent
;
2301 struct mem_cgroup
*parent
;
2302 int page_size
= PAGE_SIZE
;
2303 unsigned long flags
;
2311 if (!get_page_unless_zero(page
))
2313 if (isolate_lru_page(page
))
2316 if (PageTransHuge(page
))
2317 page_size
= HPAGE_SIZE
;
2319 parent
= mem_cgroup_from_cont(pcg
);
2320 ret
= __mem_cgroup_try_charge(NULL
, gfp_mask
,
2321 &parent
, false, page_size
);
2325 if (page_size
> PAGE_SIZE
)
2326 flags
= compound_lock_irqsave(page
);
2328 ret
= mem_cgroup_move_account(pc
, child
, parent
, true, page_size
);
2330 mem_cgroup_cancel_charge(parent
, page_size
);
2332 if (page_size
> PAGE_SIZE
)
2333 compound_unlock_irqrestore(page
, flags
);
2335 putback_lru_page(page
);
2343 * Charge the memory controller for page usage.
2345 * 0 if the charge was successful
2346 * < 0 if the cgroup is over its limit
2348 static int mem_cgroup_charge_common(struct page
*page
, struct mm_struct
*mm
,
2349 gfp_t gfp_mask
, enum charge_type ctype
)
2351 struct mem_cgroup
*mem
= NULL
;
2352 int page_size
= PAGE_SIZE
;
2353 struct page_cgroup
*pc
;
2357 if (PageTransHuge(page
)) {
2358 page_size
<<= compound_order(page
);
2359 VM_BUG_ON(!PageTransHuge(page
));
2361 * Never OOM-kill a process for a huge page. The
2362 * fault handler will fall back to regular pages.
2367 pc
= lookup_page_cgroup(page
);
2368 BUG_ON(!pc
); /* XXX: remove this and move pc lookup into commit */
2370 ret
= __mem_cgroup_try_charge(mm
, gfp_mask
, &mem
, oom
, page_size
);
2374 __mem_cgroup_commit_charge(mem
, pc
, ctype
, page_size
);
2378 int mem_cgroup_newpage_charge(struct page
*page
,
2379 struct mm_struct
*mm
, gfp_t gfp_mask
)
2381 if (mem_cgroup_disabled())
2384 * If already mapped, we don't have to account.
2385 * If page cache, page->mapping has address_space.
2386 * But page->mapping may have out-of-use anon_vma pointer,
2387 * detecit it by PageAnon() check. newly-mapped-anon's page->mapping
2390 if (page_mapped(page
) || (page
->mapping
&& !PageAnon(page
)))
2394 return mem_cgroup_charge_common(page
, mm
, gfp_mask
,
2395 MEM_CGROUP_CHARGE_TYPE_MAPPED
);
2399 __mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*ptr
,
2400 enum charge_type ctype
);
2402 int mem_cgroup_cache_charge(struct page
*page
, struct mm_struct
*mm
,
2407 if (mem_cgroup_disabled())
2409 if (PageCompound(page
))
2412 * Corner case handling. This is called from add_to_page_cache()
2413 * in usual. But some FS (shmem) precharges this page before calling it
2414 * and call add_to_page_cache() with GFP_NOWAIT.
2416 * For GFP_NOWAIT case, the page may be pre-charged before calling
2417 * add_to_page_cache(). (See shmem.c) check it here and avoid to call
2418 * charge twice. (It works but has to pay a bit larger cost.)
2419 * And when the page is SwapCache, it should take swap information
2420 * into account. This is under lock_page() now.
2422 if (!(gfp_mask
& __GFP_WAIT
)) {
2423 struct page_cgroup
*pc
;
2425 pc
= lookup_page_cgroup(page
);
2428 lock_page_cgroup(pc
);
2429 if (PageCgroupUsed(pc
)) {
2430 unlock_page_cgroup(pc
);
2433 unlock_page_cgroup(pc
);
2439 if (page_is_file_cache(page
))
2440 return mem_cgroup_charge_common(page
, mm
, gfp_mask
,
2441 MEM_CGROUP_CHARGE_TYPE_CACHE
);
2444 if (PageSwapCache(page
)) {
2445 struct mem_cgroup
*mem
;
2447 ret
= mem_cgroup_try_charge_swapin(mm
, page
, gfp_mask
, &mem
);
2449 __mem_cgroup_commit_charge_swapin(page
, mem
,
2450 MEM_CGROUP_CHARGE_TYPE_SHMEM
);
2452 ret
= mem_cgroup_charge_common(page
, mm
, gfp_mask
,
2453 MEM_CGROUP_CHARGE_TYPE_SHMEM
);
2459 * While swap-in, try_charge -> commit or cancel, the page is locked.
2460 * And when try_charge() successfully returns, one refcnt to memcg without
2461 * struct page_cgroup is acquired. This refcnt will be consumed by
2462 * "commit()" or removed by "cancel()"
2464 int mem_cgroup_try_charge_swapin(struct mm_struct
*mm
,
2466 gfp_t mask
, struct mem_cgroup
**ptr
)
2468 struct mem_cgroup
*mem
;
2473 if (mem_cgroup_disabled())
2476 if (!do_swap_account
)
2479 * A racing thread's fault, or swapoff, may have already updated
2480 * the pte, and even removed page from swap cache: in those cases
2481 * do_swap_page()'s pte_same() test will fail; but there's also a
2482 * KSM case which does need to charge the page.
2484 if (!PageSwapCache(page
))
2486 mem
= try_get_mem_cgroup_from_page(page
);
2490 ret
= __mem_cgroup_try_charge(NULL
, mask
, ptr
, true, PAGE_SIZE
);
2496 return __mem_cgroup_try_charge(mm
, mask
, ptr
, true, PAGE_SIZE
);
2500 __mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*ptr
,
2501 enum charge_type ctype
)
2503 struct page_cgroup
*pc
;
2505 if (mem_cgroup_disabled())
2509 cgroup_exclude_rmdir(&ptr
->css
);
2510 pc
= lookup_page_cgroup(page
);
2511 mem_cgroup_lru_del_before_commit_swapcache(page
);
2512 __mem_cgroup_commit_charge(ptr
, pc
, ctype
, PAGE_SIZE
);
2513 mem_cgroup_lru_add_after_commit_swapcache(page
);
2515 * Now swap is on-memory. This means this page may be
2516 * counted both as mem and swap....double count.
2517 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
2518 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
2519 * may call delete_from_swap_cache() before reach here.
2521 if (do_swap_account
&& PageSwapCache(page
)) {
2522 swp_entry_t ent
= {.val
= page_private(page
)};
2524 struct mem_cgroup
*memcg
;
2526 id
= swap_cgroup_record(ent
, 0);
2528 memcg
= mem_cgroup_lookup(id
);
2531 * This recorded memcg can be obsolete one. So, avoid
2532 * calling css_tryget
2534 if (!mem_cgroup_is_root(memcg
))
2535 res_counter_uncharge(&memcg
->memsw
, PAGE_SIZE
);
2536 mem_cgroup_swap_statistics(memcg
, false);
2537 mem_cgroup_put(memcg
);
2542 * At swapin, we may charge account against cgroup which has no tasks.
2543 * So, rmdir()->pre_destroy() can be called while we do this charge.
2544 * In that case, we need to call pre_destroy() again. check it here.
2546 cgroup_release_and_wakeup_rmdir(&ptr
->css
);
2549 void mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*ptr
)
2551 __mem_cgroup_commit_charge_swapin(page
, ptr
,
2552 MEM_CGROUP_CHARGE_TYPE_MAPPED
);
2555 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup
*mem
)
2557 if (mem_cgroup_disabled())
2561 mem_cgroup_cancel_charge(mem
, PAGE_SIZE
);
2565 __do_uncharge(struct mem_cgroup
*mem
, const enum charge_type ctype
,
2568 struct memcg_batch_info
*batch
= NULL
;
2569 bool uncharge_memsw
= true;
2570 /* If swapout, usage of swap doesn't decrease */
2571 if (!do_swap_account
|| ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
)
2572 uncharge_memsw
= false;
2574 batch
= ¤t
->memcg_batch
;
2576 * In usual, we do css_get() when we remember memcg pointer.
2577 * But in this case, we keep res->usage until end of a series of
2578 * uncharges. Then, it's ok to ignore memcg's refcnt.
2583 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
2584 * In those cases, all pages freed continously can be expected to be in
2585 * the same cgroup and we have chance to coalesce uncharges.
2586 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
2587 * because we want to do uncharge as soon as possible.
2590 if (!batch
->do_batch
|| test_thread_flag(TIF_MEMDIE
))
2591 goto direct_uncharge
;
2593 if (page_size
!= PAGE_SIZE
)
2594 goto direct_uncharge
;
2597 * In typical case, batch->memcg == mem. This means we can
2598 * merge a series of uncharges to an uncharge of res_counter.
2599 * If not, we uncharge res_counter ony by one.
2601 if (batch
->memcg
!= mem
)
2602 goto direct_uncharge
;
2603 /* remember freed charge and uncharge it later */
2604 batch
->bytes
+= PAGE_SIZE
;
2606 batch
->memsw_bytes
+= PAGE_SIZE
;
2609 res_counter_uncharge(&mem
->res
, page_size
);
2611 res_counter_uncharge(&mem
->memsw
, page_size
);
2612 if (unlikely(batch
->memcg
!= mem
))
2613 memcg_oom_recover(mem
);
2618 * uncharge if !page_mapped(page)
2620 static struct mem_cgroup
*
2621 __mem_cgroup_uncharge_common(struct page
*page
, enum charge_type ctype
)
2624 struct page_cgroup
*pc
;
2625 struct mem_cgroup
*mem
= NULL
;
2626 int page_size
= PAGE_SIZE
;
2628 if (mem_cgroup_disabled())
2631 if (PageSwapCache(page
))
2634 if (PageTransHuge(page
)) {
2635 page_size
<<= compound_order(page
);
2636 VM_BUG_ON(!PageTransHuge(page
));
2639 count
= page_size
>> PAGE_SHIFT
;
2641 * Check if our page_cgroup is valid
2643 pc
= lookup_page_cgroup(page
);
2644 if (unlikely(!pc
|| !PageCgroupUsed(pc
)))
2647 lock_page_cgroup(pc
);
2649 mem
= pc
->mem_cgroup
;
2651 if (!PageCgroupUsed(pc
))
2655 case MEM_CGROUP_CHARGE_TYPE_MAPPED
:
2656 case MEM_CGROUP_CHARGE_TYPE_DROP
:
2657 /* See mem_cgroup_prepare_migration() */
2658 if (page_mapped(page
) || PageCgroupMigration(pc
))
2661 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT
:
2662 if (!PageAnon(page
)) { /* Shared memory */
2663 if (page
->mapping
&& !page_is_file_cache(page
))
2665 } else if (page_mapped(page
)) /* Anon */
2672 mem_cgroup_charge_statistics(mem
, PageCgroupCache(pc
), -count
);
2674 ClearPageCgroupUsed(pc
);
2676 * pc->mem_cgroup is not cleared here. It will be accessed when it's
2677 * freed from LRU. This is safe because uncharged page is expected not
2678 * to be reused (freed soon). Exception is SwapCache, it's handled by
2679 * special functions.
2682 unlock_page_cgroup(pc
);
2684 * even after unlock, we have mem->res.usage here and this memcg
2685 * will never be freed.
2687 memcg_check_events(mem
, page
);
2688 if (do_swap_account
&& ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
) {
2689 mem_cgroup_swap_statistics(mem
, true);
2690 mem_cgroup_get(mem
);
2692 if (!mem_cgroup_is_root(mem
))
2693 __do_uncharge(mem
, ctype
, page_size
);
2698 unlock_page_cgroup(pc
);
2702 void mem_cgroup_uncharge_page(struct page
*page
)
2705 if (page_mapped(page
))
2707 if (page
->mapping
&& !PageAnon(page
))
2709 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_MAPPED
);
2712 void mem_cgroup_uncharge_cache_page(struct page
*page
)
2714 VM_BUG_ON(page_mapped(page
));
2715 VM_BUG_ON(page
->mapping
);
2716 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_CACHE
);
2720 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
2721 * In that cases, pages are freed continuously and we can expect pages
2722 * are in the same memcg. All these calls itself limits the number of
2723 * pages freed at once, then uncharge_start/end() is called properly.
2724 * This may be called prural(2) times in a context,
2727 void mem_cgroup_uncharge_start(void)
2729 current
->memcg_batch
.do_batch
++;
2730 /* We can do nest. */
2731 if (current
->memcg_batch
.do_batch
== 1) {
2732 current
->memcg_batch
.memcg
= NULL
;
2733 current
->memcg_batch
.bytes
= 0;
2734 current
->memcg_batch
.memsw_bytes
= 0;
2738 void mem_cgroup_uncharge_end(void)
2740 struct memcg_batch_info
*batch
= ¤t
->memcg_batch
;
2742 if (!batch
->do_batch
)
2746 if (batch
->do_batch
) /* If stacked, do nothing. */
2752 * This "batch->memcg" is valid without any css_get/put etc...
2753 * bacause we hide charges behind us.
2756 res_counter_uncharge(&batch
->memcg
->res
, batch
->bytes
);
2757 if (batch
->memsw_bytes
)
2758 res_counter_uncharge(&batch
->memcg
->memsw
, batch
->memsw_bytes
);
2759 memcg_oom_recover(batch
->memcg
);
2760 /* forget this pointer (for sanity check) */
2761 batch
->memcg
= NULL
;
2766 * called after __delete_from_swap_cache() and drop "page" account.
2767 * memcg information is recorded to swap_cgroup of "ent"
2770 mem_cgroup_uncharge_swapcache(struct page
*page
, swp_entry_t ent
, bool swapout
)
2772 struct mem_cgroup
*memcg
;
2773 int ctype
= MEM_CGROUP_CHARGE_TYPE_SWAPOUT
;
2775 if (!swapout
) /* this was a swap cache but the swap is unused ! */
2776 ctype
= MEM_CGROUP_CHARGE_TYPE_DROP
;
2778 memcg
= __mem_cgroup_uncharge_common(page
, ctype
);
2781 * record memcg information, if swapout && memcg != NULL,
2782 * mem_cgroup_get() was called in uncharge().
2784 if (do_swap_account
&& swapout
&& memcg
)
2785 swap_cgroup_record(ent
, css_id(&memcg
->css
));
2789 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2791 * called from swap_entry_free(). remove record in swap_cgroup and
2792 * uncharge "memsw" account.
2794 void mem_cgroup_uncharge_swap(swp_entry_t ent
)
2796 struct mem_cgroup
*memcg
;
2799 if (!do_swap_account
)
2802 id
= swap_cgroup_record(ent
, 0);
2804 memcg
= mem_cgroup_lookup(id
);
2807 * We uncharge this because swap is freed.
2808 * This memcg can be obsolete one. We avoid calling css_tryget
2810 if (!mem_cgroup_is_root(memcg
))
2811 res_counter_uncharge(&memcg
->memsw
, PAGE_SIZE
);
2812 mem_cgroup_swap_statistics(memcg
, false);
2813 mem_cgroup_put(memcg
);
2819 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
2820 * @entry: swap entry to be moved
2821 * @from: mem_cgroup which the entry is moved from
2822 * @to: mem_cgroup which the entry is moved to
2823 * @need_fixup: whether we should fixup res_counters and refcounts.
2825 * It succeeds only when the swap_cgroup's record for this entry is the same
2826 * as the mem_cgroup's id of @from.
2828 * Returns 0 on success, -EINVAL on failure.
2830 * The caller must have charged to @to, IOW, called res_counter_charge() about
2831 * both res and memsw, and called css_get().
2833 static int mem_cgroup_move_swap_account(swp_entry_t entry
,
2834 struct mem_cgroup
*from
, struct mem_cgroup
*to
, bool need_fixup
)
2836 unsigned short old_id
, new_id
;
2838 old_id
= css_id(&from
->css
);
2839 new_id
= css_id(&to
->css
);
2841 if (swap_cgroup_cmpxchg(entry
, old_id
, new_id
) == old_id
) {
2842 mem_cgroup_swap_statistics(from
, false);
2843 mem_cgroup_swap_statistics(to
, true);
2845 * This function is only called from task migration context now.
2846 * It postpones res_counter and refcount handling till the end
2847 * of task migration(mem_cgroup_clear_mc()) for performance
2848 * improvement. But we cannot postpone mem_cgroup_get(to)
2849 * because if the process that has been moved to @to does
2850 * swap-in, the refcount of @to might be decreased to 0.
2854 if (!mem_cgroup_is_root(from
))
2855 res_counter_uncharge(&from
->memsw
, PAGE_SIZE
);
2856 mem_cgroup_put(from
);
2858 * we charged both to->res and to->memsw, so we should
2861 if (!mem_cgroup_is_root(to
))
2862 res_counter_uncharge(&to
->res
, PAGE_SIZE
);
2869 static inline int mem_cgroup_move_swap_account(swp_entry_t entry
,
2870 struct mem_cgroup
*from
, struct mem_cgroup
*to
, bool need_fixup
)
2877 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
2880 int mem_cgroup_prepare_migration(struct page
*page
,
2881 struct page
*newpage
, struct mem_cgroup
**ptr
, gfp_t gfp_mask
)
2883 struct page_cgroup
*pc
;
2884 struct mem_cgroup
*mem
= NULL
;
2885 enum charge_type ctype
;
2890 VM_BUG_ON(PageTransHuge(page
));
2891 if (mem_cgroup_disabled())
2894 pc
= lookup_page_cgroup(page
);
2895 lock_page_cgroup(pc
);
2896 if (PageCgroupUsed(pc
)) {
2897 mem
= pc
->mem_cgroup
;
2900 * At migrating an anonymous page, its mapcount goes down
2901 * to 0 and uncharge() will be called. But, even if it's fully
2902 * unmapped, migration may fail and this page has to be
2903 * charged again. We set MIGRATION flag here and delay uncharge
2904 * until end_migration() is called
2906 * Corner Case Thinking
2908 * When the old page was mapped as Anon and it's unmap-and-freed
2909 * while migration was ongoing.
2910 * If unmap finds the old page, uncharge() of it will be delayed
2911 * until end_migration(). If unmap finds a new page, it's
2912 * uncharged when it make mapcount to be 1->0. If unmap code
2913 * finds swap_migration_entry, the new page will not be mapped
2914 * and end_migration() will find it(mapcount==0).
2917 * When the old page was mapped but migraion fails, the kernel
2918 * remaps it. A charge for it is kept by MIGRATION flag even
2919 * if mapcount goes down to 0. We can do remap successfully
2920 * without charging it again.
2923 * The "old" page is under lock_page() until the end of
2924 * migration, so, the old page itself will not be swapped-out.
2925 * If the new page is swapped out before end_migraton, our
2926 * hook to usual swap-out path will catch the event.
2929 SetPageCgroupMigration(pc
);
2931 unlock_page_cgroup(pc
);
2933 * If the page is not charged at this point,
2940 ret
= __mem_cgroup_try_charge(NULL
, gfp_mask
, ptr
, false, PAGE_SIZE
);
2941 css_put(&mem
->css
);/* drop extra refcnt */
2942 if (ret
|| *ptr
== NULL
) {
2943 if (PageAnon(page
)) {
2944 lock_page_cgroup(pc
);
2945 ClearPageCgroupMigration(pc
);
2946 unlock_page_cgroup(pc
);
2948 * The old page may be fully unmapped while we kept it.
2950 mem_cgroup_uncharge_page(page
);
2955 * We charge new page before it's used/mapped. So, even if unlock_page()
2956 * is called before end_migration, we can catch all events on this new
2957 * page. In the case new page is migrated but not remapped, new page's
2958 * mapcount will be finally 0 and we call uncharge in end_migration().
2960 pc
= lookup_page_cgroup(newpage
);
2962 ctype
= MEM_CGROUP_CHARGE_TYPE_MAPPED
;
2963 else if (page_is_file_cache(page
))
2964 ctype
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
2966 ctype
= MEM_CGROUP_CHARGE_TYPE_SHMEM
;
2967 __mem_cgroup_commit_charge(mem
, pc
, ctype
, PAGE_SIZE
);
2971 /* remove redundant charge if migration failed*/
2972 void mem_cgroup_end_migration(struct mem_cgroup
*mem
,
2973 struct page
*oldpage
, struct page
*newpage
, bool migration_ok
)
2975 struct page
*used
, *unused
;
2976 struct page_cgroup
*pc
;
2980 /* blocks rmdir() */
2981 cgroup_exclude_rmdir(&mem
->css
);
2982 if (!migration_ok
) {
2990 * We disallowed uncharge of pages under migration because mapcount
2991 * of the page goes down to zero, temporarly.
2992 * Clear the flag and check the page should be charged.
2994 pc
= lookup_page_cgroup(oldpage
);
2995 lock_page_cgroup(pc
);
2996 ClearPageCgroupMigration(pc
);
2997 unlock_page_cgroup(pc
);
2999 __mem_cgroup_uncharge_common(unused
, MEM_CGROUP_CHARGE_TYPE_FORCE
);
3002 * If a page is a file cache, radix-tree replacement is very atomic
3003 * and we can skip this check. When it was an Anon page, its mapcount
3004 * goes down to 0. But because we added MIGRATION flage, it's not
3005 * uncharged yet. There are several case but page->mapcount check
3006 * and USED bit check in mem_cgroup_uncharge_page() will do enough
3007 * check. (see prepare_charge() also)
3010 mem_cgroup_uncharge_page(used
);
3012 * At migration, we may charge account against cgroup which has no
3014 * So, rmdir()->pre_destroy() can be called while we do this charge.
3015 * In that case, we need to call pre_destroy() again. check it here.
3017 cgroup_release_and_wakeup_rmdir(&mem
->css
);
3021 * A call to try to shrink memory usage on charge failure at shmem's swapin.
3022 * Calling hierarchical_reclaim is not enough because we should update
3023 * last_oom_jiffies to prevent pagefault_out_of_memory from invoking global OOM.
3024 * Moreover considering hierarchy, we should reclaim from the mem_over_limit,
3025 * not from the memcg which this page would be charged to.
3026 * try_charge_swapin does all of these works properly.
3028 int mem_cgroup_shmem_charge_fallback(struct page
*page
,
3029 struct mm_struct
*mm
,
3032 struct mem_cgroup
*mem
;
3035 if (mem_cgroup_disabled())
3038 ret
= mem_cgroup_try_charge_swapin(mm
, page
, gfp_mask
, &mem
);
3040 mem_cgroup_cancel_charge_swapin(mem
); /* it does !mem check */
3045 #ifdef CONFIG_DEBUG_VM
3046 static struct page_cgroup
*lookup_page_cgroup_used(struct page
*page
)
3048 struct page_cgroup
*pc
;
3050 pc
= lookup_page_cgroup(page
);
3051 if (likely(pc
) && PageCgroupUsed(pc
))
3056 bool mem_cgroup_bad_page_check(struct page
*page
)
3058 if (mem_cgroup_disabled())
3061 return lookup_page_cgroup_used(page
) != NULL
;
3064 void mem_cgroup_print_bad_page(struct page
*page
)
3066 struct page_cgroup
*pc
;
3068 pc
= lookup_page_cgroup_used(page
);
3073 printk(KERN_ALERT
"pc:%p pc->flags:%lx pc->mem_cgroup:%p",
3074 pc
, pc
->flags
, pc
->mem_cgroup
);
3076 path
= kmalloc(PATH_MAX
, GFP_KERNEL
);
3079 ret
= cgroup_path(pc
->mem_cgroup
->css
.cgroup
,
3084 printk(KERN_CONT
"(%s)\n",
3085 (ret
< 0) ? "cannot get the path" : path
);
3091 static DEFINE_MUTEX(set_limit_mutex
);
3093 static int mem_cgroup_resize_limit(struct mem_cgroup
*memcg
,
3094 unsigned long long val
)
3097 u64 memswlimit
, memlimit
;
3099 int children
= mem_cgroup_count_children(memcg
);
3100 u64 curusage
, oldusage
;
3104 * For keeping hierarchical_reclaim simple, how long we should retry
3105 * is depends on callers. We set our retry-count to be function
3106 * of # of children which we should visit in this loop.
3108 retry_count
= MEM_CGROUP_RECLAIM_RETRIES
* children
;
3110 oldusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3113 while (retry_count
) {
3114 if (signal_pending(current
)) {
3119 * Rather than hide all in some function, I do this in
3120 * open coded manner. You see what this really does.
3121 * We have to guarantee mem->res.limit < mem->memsw.limit.
3123 mutex_lock(&set_limit_mutex
);
3124 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3125 if (memswlimit
< val
) {
3127 mutex_unlock(&set_limit_mutex
);
3131 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3135 ret
= res_counter_set_limit(&memcg
->res
, val
);
3137 if (memswlimit
== val
)
3138 memcg
->memsw_is_minimum
= true;
3140 memcg
->memsw_is_minimum
= false;
3142 mutex_unlock(&set_limit_mutex
);
3147 mem_cgroup_hierarchical_reclaim(memcg
, NULL
, GFP_KERNEL
,
3148 MEM_CGROUP_RECLAIM_SHRINK
);
3149 curusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3150 /* Usage is reduced ? */
3151 if (curusage
>= oldusage
)
3154 oldusage
= curusage
;
3156 if (!ret
&& enlarge
)
3157 memcg_oom_recover(memcg
);
3162 static int mem_cgroup_resize_memsw_limit(struct mem_cgroup
*memcg
,
3163 unsigned long long val
)
3166 u64 memlimit
, memswlimit
, oldusage
, curusage
;
3167 int children
= mem_cgroup_count_children(memcg
);
3171 /* see mem_cgroup_resize_res_limit */
3172 retry_count
= children
* MEM_CGROUP_RECLAIM_RETRIES
;
3173 oldusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3174 while (retry_count
) {
3175 if (signal_pending(current
)) {
3180 * Rather than hide all in some function, I do this in
3181 * open coded manner. You see what this really does.
3182 * We have to guarantee mem->res.limit < mem->memsw.limit.
3184 mutex_lock(&set_limit_mutex
);
3185 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3186 if (memlimit
> val
) {
3188 mutex_unlock(&set_limit_mutex
);
3191 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3192 if (memswlimit
< val
)
3194 ret
= res_counter_set_limit(&memcg
->memsw
, val
);
3196 if (memlimit
== val
)
3197 memcg
->memsw_is_minimum
= true;
3199 memcg
->memsw_is_minimum
= false;
3201 mutex_unlock(&set_limit_mutex
);
3206 mem_cgroup_hierarchical_reclaim(memcg
, NULL
, GFP_KERNEL
,
3207 MEM_CGROUP_RECLAIM_NOSWAP
|
3208 MEM_CGROUP_RECLAIM_SHRINK
);
3209 curusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3210 /* Usage is reduced ? */
3211 if (curusage
>= oldusage
)
3214 oldusage
= curusage
;
3216 if (!ret
&& enlarge
)
3217 memcg_oom_recover(memcg
);
3221 unsigned long mem_cgroup_soft_limit_reclaim(struct zone
*zone
, int order
,
3224 unsigned long nr_reclaimed
= 0;
3225 struct mem_cgroup_per_zone
*mz
, *next_mz
= NULL
;
3226 unsigned long reclaimed
;
3228 struct mem_cgroup_tree_per_zone
*mctz
;
3229 unsigned long long excess
;
3234 mctz
= soft_limit_tree_node_zone(zone_to_nid(zone
), zone_idx(zone
));
3236 * This loop can run a while, specially if mem_cgroup's continuously
3237 * keep exceeding their soft limit and putting the system under
3244 mz
= mem_cgroup_largest_soft_limit_node(mctz
);
3248 reclaimed
= mem_cgroup_hierarchical_reclaim(mz
->mem
, zone
,
3250 MEM_CGROUP_RECLAIM_SOFT
);
3251 nr_reclaimed
+= reclaimed
;
3252 spin_lock(&mctz
->lock
);
3255 * If we failed to reclaim anything from this memory cgroup
3256 * it is time to move on to the next cgroup
3262 * Loop until we find yet another one.
3264 * By the time we get the soft_limit lock
3265 * again, someone might have aded the
3266 * group back on the RB tree. Iterate to
3267 * make sure we get a different mem.
3268 * mem_cgroup_largest_soft_limit_node returns
3269 * NULL if no other cgroup is present on
3273 __mem_cgroup_largest_soft_limit_node(mctz
);
3274 if (next_mz
== mz
) {
3275 css_put(&next_mz
->mem
->css
);
3277 } else /* next_mz == NULL or other memcg */
3281 __mem_cgroup_remove_exceeded(mz
->mem
, mz
, mctz
);
3282 excess
= res_counter_soft_limit_excess(&mz
->mem
->res
);
3284 * One school of thought says that we should not add
3285 * back the node to the tree if reclaim returns 0.
3286 * But our reclaim could return 0, simply because due
3287 * to priority we are exposing a smaller subset of
3288 * memory to reclaim from. Consider this as a longer
3291 /* If excess == 0, no tree ops */
3292 __mem_cgroup_insert_exceeded(mz
->mem
, mz
, mctz
, excess
);
3293 spin_unlock(&mctz
->lock
);
3294 css_put(&mz
->mem
->css
);
3297 * Could not reclaim anything and there are no more
3298 * mem cgroups to try or we seem to be looping without
3299 * reclaiming anything.
3301 if (!nr_reclaimed
&&
3303 loop
> MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS
))
3305 } while (!nr_reclaimed
);
3307 css_put(&next_mz
->mem
->css
);
3308 return nr_reclaimed
;
3312 * This routine traverse page_cgroup in given list and drop them all.
3313 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
3315 static int mem_cgroup_force_empty_list(struct mem_cgroup
*mem
,
3316 int node
, int zid
, enum lru_list lru
)
3319 struct mem_cgroup_per_zone
*mz
;
3320 struct page_cgroup
*pc
, *busy
;
3321 unsigned long flags
, loop
;
3322 struct list_head
*list
;
3325 zone
= &NODE_DATA(node
)->node_zones
[zid
];
3326 mz
= mem_cgroup_zoneinfo(mem
, node
, zid
);
3327 list
= &mz
->lists
[lru
];
3329 loop
= MEM_CGROUP_ZSTAT(mz
, lru
);
3330 /* give some margin against EBUSY etc...*/
3335 spin_lock_irqsave(&zone
->lru_lock
, flags
);
3336 if (list_empty(list
)) {
3337 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3340 pc
= list_entry(list
->prev
, struct page_cgroup
, lru
);
3342 list_move(&pc
->lru
, list
);
3344 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3347 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3349 ret
= mem_cgroup_move_parent(pc
, mem
, GFP_KERNEL
);
3353 if (ret
== -EBUSY
|| ret
== -EINVAL
) {
3354 /* found lock contention or "pc" is obsolete. */
3361 if (!ret
&& !list_empty(list
))
3367 * make mem_cgroup's charge to be 0 if there is no task.
3368 * This enables deleting this mem_cgroup.
3370 static int mem_cgroup_force_empty(struct mem_cgroup
*mem
, bool free_all
)
3373 int node
, zid
, shrink
;
3374 int nr_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
3375 struct cgroup
*cgrp
= mem
->css
.cgroup
;
3380 /* should free all ? */
3386 if (cgroup_task_count(cgrp
) || !list_empty(&cgrp
->children
))
3389 if (signal_pending(current
))
3391 /* This is for making all *used* pages to be on LRU. */
3392 lru_add_drain_all();
3393 drain_all_stock_sync();
3395 mem_cgroup_start_move(mem
);
3396 for_each_node_state(node
, N_HIGH_MEMORY
) {
3397 for (zid
= 0; !ret
&& zid
< MAX_NR_ZONES
; zid
++) {
3400 ret
= mem_cgroup_force_empty_list(mem
,
3409 mem_cgroup_end_move(mem
);
3410 memcg_oom_recover(mem
);
3411 /* it seems parent cgroup doesn't have enough mem */
3415 /* "ret" should also be checked to ensure all lists are empty. */
3416 } while (mem
->res
.usage
> 0 || ret
);
3422 /* returns EBUSY if there is a task or if we come here twice. */
3423 if (cgroup_task_count(cgrp
) || !list_empty(&cgrp
->children
) || shrink
) {
3427 /* we call try-to-free pages for make this cgroup empty */
3428 lru_add_drain_all();
3429 /* try to free all pages in this cgroup */
3431 while (nr_retries
&& mem
->res
.usage
> 0) {
3434 if (signal_pending(current
)) {
3438 progress
= try_to_free_mem_cgroup_pages(mem
, GFP_KERNEL
,
3439 false, get_swappiness(mem
));
3442 /* maybe some writeback is necessary */
3443 congestion_wait(BLK_RW_ASYNC
, HZ
/10);
3448 /* try move_account...there may be some *locked* pages. */
3452 int mem_cgroup_force_empty_write(struct cgroup
*cont
, unsigned int event
)
3454 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont
), true);
3458 static u64
mem_cgroup_hierarchy_read(struct cgroup
*cont
, struct cftype
*cft
)
3460 return mem_cgroup_from_cont(cont
)->use_hierarchy
;
3463 static int mem_cgroup_hierarchy_write(struct cgroup
*cont
, struct cftype
*cft
,
3467 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
3468 struct cgroup
*parent
= cont
->parent
;
3469 struct mem_cgroup
*parent_mem
= NULL
;
3472 parent_mem
= mem_cgroup_from_cont(parent
);
3476 * If parent's use_hierarchy is set, we can't make any modifications
3477 * in the child subtrees. If it is unset, then the change can
3478 * occur, provided the current cgroup has no children.
3480 * For the root cgroup, parent_mem is NULL, we allow value to be
3481 * set if there are no children.
3483 if ((!parent_mem
|| !parent_mem
->use_hierarchy
) &&
3484 (val
== 1 || val
== 0)) {
3485 if (list_empty(&cont
->children
))
3486 mem
->use_hierarchy
= val
;
3497 static u64
mem_cgroup_get_recursive_idx_stat(struct mem_cgroup
*mem
,
3498 enum mem_cgroup_stat_index idx
)
3500 struct mem_cgroup
*iter
;
3503 /* each per cpu's value can be minus.Then, use s64 */
3504 for_each_mem_cgroup_tree(iter
, mem
)
3505 val
+= mem_cgroup_read_stat(iter
, idx
);
3507 if (val
< 0) /* race ? */
3512 static inline u64
mem_cgroup_usage(struct mem_cgroup
*mem
, bool swap
)
3516 if (!mem_cgroup_is_root(mem
)) {
3518 return res_counter_read_u64(&mem
->res
, RES_USAGE
);
3520 return res_counter_read_u64(&mem
->memsw
, RES_USAGE
);
3523 val
= mem_cgroup_get_recursive_idx_stat(mem
, MEM_CGROUP_STAT_CACHE
);
3524 val
+= mem_cgroup_get_recursive_idx_stat(mem
, MEM_CGROUP_STAT_RSS
);
3527 val
+= mem_cgroup_get_recursive_idx_stat(mem
,
3528 MEM_CGROUP_STAT_SWAPOUT
);
3530 return val
<< PAGE_SHIFT
;
3533 static u64
mem_cgroup_read(struct cgroup
*cont
, struct cftype
*cft
)
3535 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
3539 type
= MEMFILE_TYPE(cft
->private);
3540 name
= MEMFILE_ATTR(cft
->private);
3543 if (name
== RES_USAGE
)
3544 val
= mem_cgroup_usage(mem
, false);
3546 val
= res_counter_read_u64(&mem
->res
, name
);
3549 if (name
== RES_USAGE
)
3550 val
= mem_cgroup_usage(mem
, true);
3552 val
= res_counter_read_u64(&mem
->memsw
, name
);
3561 * The user of this function is...
3564 static int mem_cgroup_write(struct cgroup
*cont
, struct cftype
*cft
,
3567 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3569 unsigned long long val
;
3572 type
= MEMFILE_TYPE(cft
->private);
3573 name
= MEMFILE_ATTR(cft
->private);
3576 if (mem_cgroup_is_root(memcg
)) { /* Can't set limit on root */
3580 /* This function does all necessary parse...reuse it */
3581 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
3585 ret
= mem_cgroup_resize_limit(memcg
, val
);
3587 ret
= mem_cgroup_resize_memsw_limit(memcg
, val
);
3589 case RES_SOFT_LIMIT
:
3590 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
3594 * For memsw, soft limits are hard to implement in terms
3595 * of semantics, for now, we support soft limits for
3596 * control without swap
3599 ret
= res_counter_set_soft_limit(&memcg
->res
, val
);
3604 ret
= -EINVAL
; /* should be BUG() ? */
3610 static void memcg_get_hierarchical_limit(struct mem_cgroup
*memcg
,
3611 unsigned long long *mem_limit
, unsigned long long *memsw_limit
)
3613 struct cgroup
*cgroup
;
3614 unsigned long long min_limit
, min_memsw_limit
, tmp
;
3616 min_limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3617 min_memsw_limit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3618 cgroup
= memcg
->css
.cgroup
;
3619 if (!memcg
->use_hierarchy
)
3622 while (cgroup
->parent
) {
3623 cgroup
= cgroup
->parent
;
3624 memcg
= mem_cgroup_from_cont(cgroup
);
3625 if (!memcg
->use_hierarchy
)
3627 tmp
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3628 min_limit
= min(min_limit
, tmp
);
3629 tmp
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3630 min_memsw_limit
= min(min_memsw_limit
, tmp
);
3633 *mem_limit
= min_limit
;
3634 *memsw_limit
= min_memsw_limit
;
3638 static int mem_cgroup_reset(struct cgroup
*cont
, unsigned int event
)
3640 struct mem_cgroup
*mem
;
3643 mem
= mem_cgroup_from_cont(cont
);
3644 type
= MEMFILE_TYPE(event
);
3645 name
= MEMFILE_ATTR(event
);
3649 res_counter_reset_max(&mem
->res
);
3651 res_counter_reset_max(&mem
->memsw
);
3655 res_counter_reset_failcnt(&mem
->res
);
3657 res_counter_reset_failcnt(&mem
->memsw
);
3664 static u64
mem_cgroup_move_charge_read(struct cgroup
*cgrp
,
3667 return mem_cgroup_from_cont(cgrp
)->move_charge_at_immigrate
;
3671 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
3672 struct cftype
*cft
, u64 val
)
3674 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgrp
);
3676 if (val
>= (1 << NR_MOVE_TYPE
))
3679 * We check this value several times in both in can_attach() and
3680 * attach(), so we need cgroup lock to prevent this value from being
3684 mem
->move_charge_at_immigrate
= val
;
3690 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
3691 struct cftype
*cft
, u64 val
)
3698 /* For read statistics */
3714 struct mcs_total_stat
{
3715 s64 stat
[NR_MCS_STAT
];
3721 } memcg_stat_strings
[NR_MCS_STAT
] = {
3722 {"cache", "total_cache"},
3723 {"rss", "total_rss"},
3724 {"mapped_file", "total_mapped_file"},
3725 {"pgpgin", "total_pgpgin"},
3726 {"pgpgout", "total_pgpgout"},
3727 {"swap", "total_swap"},
3728 {"inactive_anon", "total_inactive_anon"},
3729 {"active_anon", "total_active_anon"},
3730 {"inactive_file", "total_inactive_file"},
3731 {"active_file", "total_active_file"},
3732 {"unevictable", "total_unevictable"}
3737 mem_cgroup_get_local_stat(struct mem_cgroup
*mem
, struct mcs_total_stat
*s
)
3742 val
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_CACHE
);
3743 s
->stat
[MCS_CACHE
] += val
* PAGE_SIZE
;
3744 val
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_RSS
);
3745 s
->stat
[MCS_RSS
] += val
* PAGE_SIZE
;
3746 val
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_FILE_MAPPED
);
3747 s
->stat
[MCS_FILE_MAPPED
] += val
* PAGE_SIZE
;
3748 val
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_PGPGIN_COUNT
);
3749 s
->stat
[MCS_PGPGIN
] += val
;
3750 val
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_PGPGOUT_COUNT
);
3751 s
->stat
[MCS_PGPGOUT
] += val
;
3752 if (do_swap_account
) {
3753 val
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_SWAPOUT
);
3754 s
->stat
[MCS_SWAP
] += val
* PAGE_SIZE
;
3758 val
= mem_cgroup_get_local_zonestat(mem
, LRU_INACTIVE_ANON
);
3759 s
->stat
[MCS_INACTIVE_ANON
] += val
* PAGE_SIZE
;
3760 val
= mem_cgroup_get_local_zonestat(mem
, LRU_ACTIVE_ANON
);
3761 s
->stat
[MCS_ACTIVE_ANON
] += val
* PAGE_SIZE
;
3762 val
= mem_cgroup_get_local_zonestat(mem
, LRU_INACTIVE_FILE
);
3763 s
->stat
[MCS_INACTIVE_FILE
] += val
* PAGE_SIZE
;
3764 val
= mem_cgroup_get_local_zonestat(mem
, LRU_ACTIVE_FILE
);
3765 s
->stat
[MCS_ACTIVE_FILE
] += val
* PAGE_SIZE
;
3766 val
= mem_cgroup_get_local_zonestat(mem
, LRU_UNEVICTABLE
);
3767 s
->stat
[MCS_UNEVICTABLE
] += val
* PAGE_SIZE
;
3771 mem_cgroup_get_total_stat(struct mem_cgroup
*mem
, struct mcs_total_stat
*s
)
3773 struct mem_cgroup
*iter
;
3775 for_each_mem_cgroup_tree(iter
, mem
)
3776 mem_cgroup_get_local_stat(iter
, s
);
3779 static int mem_control_stat_show(struct cgroup
*cont
, struct cftype
*cft
,
3780 struct cgroup_map_cb
*cb
)
3782 struct mem_cgroup
*mem_cont
= mem_cgroup_from_cont(cont
);
3783 struct mcs_total_stat mystat
;
3786 memset(&mystat
, 0, sizeof(mystat
));
3787 mem_cgroup_get_local_stat(mem_cont
, &mystat
);
3789 for (i
= 0; i
< NR_MCS_STAT
; i
++) {
3790 if (i
== MCS_SWAP
&& !do_swap_account
)
3792 cb
->fill(cb
, memcg_stat_strings
[i
].local_name
, mystat
.stat
[i
]);
3795 /* Hierarchical information */
3797 unsigned long long limit
, memsw_limit
;
3798 memcg_get_hierarchical_limit(mem_cont
, &limit
, &memsw_limit
);
3799 cb
->fill(cb
, "hierarchical_memory_limit", limit
);
3800 if (do_swap_account
)
3801 cb
->fill(cb
, "hierarchical_memsw_limit", memsw_limit
);
3804 memset(&mystat
, 0, sizeof(mystat
));
3805 mem_cgroup_get_total_stat(mem_cont
, &mystat
);
3806 for (i
= 0; i
< NR_MCS_STAT
; i
++) {
3807 if (i
== MCS_SWAP
&& !do_swap_account
)
3809 cb
->fill(cb
, memcg_stat_strings
[i
].total_name
, mystat
.stat
[i
]);
3812 #ifdef CONFIG_DEBUG_VM
3813 cb
->fill(cb
, "inactive_ratio", calc_inactive_ratio(mem_cont
, NULL
));
3817 struct mem_cgroup_per_zone
*mz
;
3818 unsigned long recent_rotated
[2] = {0, 0};
3819 unsigned long recent_scanned
[2] = {0, 0};
3821 for_each_online_node(nid
)
3822 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
3823 mz
= mem_cgroup_zoneinfo(mem_cont
, nid
, zid
);
3825 recent_rotated
[0] +=
3826 mz
->reclaim_stat
.recent_rotated
[0];
3827 recent_rotated
[1] +=
3828 mz
->reclaim_stat
.recent_rotated
[1];
3829 recent_scanned
[0] +=
3830 mz
->reclaim_stat
.recent_scanned
[0];
3831 recent_scanned
[1] +=
3832 mz
->reclaim_stat
.recent_scanned
[1];
3834 cb
->fill(cb
, "recent_rotated_anon", recent_rotated
[0]);
3835 cb
->fill(cb
, "recent_rotated_file", recent_rotated
[1]);
3836 cb
->fill(cb
, "recent_scanned_anon", recent_scanned
[0]);
3837 cb
->fill(cb
, "recent_scanned_file", recent_scanned
[1]);
3844 static u64
mem_cgroup_swappiness_read(struct cgroup
*cgrp
, struct cftype
*cft
)
3846 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
3848 return get_swappiness(memcg
);
3851 static int mem_cgroup_swappiness_write(struct cgroup
*cgrp
, struct cftype
*cft
,
3854 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
3855 struct mem_cgroup
*parent
;
3860 if (cgrp
->parent
== NULL
)
3863 parent
= mem_cgroup_from_cont(cgrp
->parent
);
3867 /* If under hierarchy, only empty-root can set this value */
3868 if ((parent
->use_hierarchy
) ||
3869 (memcg
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
3874 spin_lock(&memcg
->reclaim_param_lock
);
3875 memcg
->swappiness
= val
;
3876 spin_unlock(&memcg
->reclaim_param_lock
);
3883 static void __mem_cgroup_threshold(struct mem_cgroup
*memcg
, bool swap
)
3885 struct mem_cgroup_threshold_ary
*t
;
3891 t
= rcu_dereference(memcg
->thresholds
.primary
);
3893 t
= rcu_dereference(memcg
->memsw_thresholds
.primary
);
3898 usage
= mem_cgroup_usage(memcg
, swap
);
3901 * current_threshold points to threshold just below usage.
3902 * If it's not true, a threshold was crossed after last
3903 * call of __mem_cgroup_threshold().
3905 i
= t
->current_threshold
;
3908 * Iterate backward over array of thresholds starting from
3909 * current_threshold and check if a threshold is crossed.
3910 * If none of thresholds below usage is crossed, we read
3911 * only one element of the array here.
3913 for (; i
>= 0 && unlikely(t
->entries
[i
].threshold
> usage
); i
--)
3914 eventfd_signal(t
->entries
[i
].eventfd
, 1);
3916 /* i = current_threshold + 1 */
3920 * Iterate forward over array of thresholds starting from
3921 * current_threshold+1 and check if a threshold is crossed.
3922 * If none of thresholds above usage is crossed, we read
3923 * only one element of the array here.
3925 for (; i
< t
->size
&& unlikely(t
->entries
[i
].threshold
<= usage
); i
++)
3926 eventfd_signal(t
->entries
[i
].eventfd
, 1);
3928 /* Update current_threshold */
3929 t
->current_threshold
= i
- 1;
3934 static void mem_cgroup_threshold(struct mem_cgroup
*memcg
)
3937 __mem_cgroup_threshold(memcg
, false);
3938 if (do_swap_account
)
3939 __mem_cgroup_threshold(memcg
, true);
3941 memcg
= parent_mem_cgroup(memcg
);
3945 static int compare_thresholds(const void *a
, const void *b
)
3947 const struct mem_cgroup_threshold
*_a
= a
;
3948 const struct mem_cgroup_threshold
*_b
= b
;
3950 return _a
->threshold
- _b
->threshold
;
3953 static int mem_cgroup_oom_notify_cb(struct mem_cgroup
*mem
)
3955 struct mem_cgroup_eventfd_list
*ev
;
3957 list_for_each_entry(ev
, &mem
->oom_notify
, list
)
3958 eventfd_signal(ev
->eventfd
, 1);
3962 static void mem_cgroup_oom_notify(struct mem_cgroup
*mem
)
3964 struct mem_cgroup
*iter
;
3966 for_each_mem_cgroup_tree(iter
, mem
)
3967 mem_cgroup_oom_notify_cb(iter
);
3970 static int mem_cgroup_usage_register_event(struct cgroup
*cgrp
,
3971 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
3973 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
3974 struct mem_cgroup_thresholds
*thresholds
;
3975 struct mem_cgroup_threshold_ary
*new;
3976 int type
= MEMFILE_TYPE(cft
->private);
3977 u64 threshold
, usage
;
3980 ret
= res_counter_memparse_write_strategy(args
, &threshold
);
3984 mutex_lock(&memcg
->thresholds_lock
);
3987 thresholds
= &memcg
->thresholds
;
3988 else if (type
== _MEMSWAP
)
3989 thresholds
= &memcg
->memsw_thresholds
;
3993 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
3995 /* Check if a threshold crossed before adding a new one */
3996 if (thresholds
->primary
)
3997 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
3999 size
= thresholds
->primary
? thresholds
->primary
->size
+ 1 : 1;
4001 /* Allocate memory for new array of thresholds */
4002 new = kmalloc(sizeof(*new) + size
* sizeof(struct mem_cgroup_threshold
),
4010 /* Copy thresholds (if any) to new array */
4011 if (thresholds
->primary
) {
4012 memcpy(new->entries
, thresholds
->primary
->entries
, (size
- 1) *
4013 sizeof(struct mem_cgroup_threshold
));
4016 /* Add new threshold */
4017 new->entries
[size
- 1].eventfd
= eventfd
;
4018 new->entries
[size
- 1].threshold
= threshold
;
4020 /* Sort thresholds. Registering of new threshold isn't time-critical */
4021 sort(new->entries
, size
, sizeof(struct mem_cgroup_threshold
),
4022 compare_thresholds
, NULL
);
4024 /* Find current threshold */
4025 new->current_threshold
= -1;
4026 for (i
= 0; i
< size
; i
++) {
4027 if (new->entries
[i
].threshold
< usage
) {
4029 * new->current_threshold will not be used until
4030 * rcu_assign_pointer(), so it's safe to increment
4033 ++new->current_threshold
;
4037 /* Free old spare buffer and save old primary buffer as spare */
4038 kfree(thresholds
->spare
);
4039 thresholds
->spare
= thresholds
->primary
;
4041 rcu_assign_pointer(thresholds
->primary
, new);
4043 /* To be sure that nobody uses thresholds */
4047 mutex_unlock(&memcg
->thresholds_lock
);
4052 static void mem_cgroup_usage_unregister_event(struct cgroup
*cgrp
,
4053 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
4055 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4056 struct mem_cgroup_thresholds
*thresholds
;
4057 struct mem_cgroup_threshold_ary
*new;
4058 int type
= MEMFILE_TYPE(cft
->private);
4062 mutex_lock(&memcg
->thresholds_lock
);
4064 thresholds
= &memcg
->thresholds
;
4065 else if (type
== _MEMSWAP
)
4066 thresholds
= &memcg
->memsw_thresholds
;
4071 * Something went wrong if we trying to unregister a threshold
4072 * if we don't have thresholds
4074 BUG_ON(!thresholds
);
4076 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
4078 /* Check if a threshold crossed before removing */
4079 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
4081 /* Calculate new number of threshold */
4083 for (i
= 0; i
< thresholds
->primary
->size
; i
++) {
4084 if (thresholds
->primary
->entries
[i
].eventfd
!= eventfd
)
4088 new = thresholds
->spare
;
4090 /* Set thresholds array to NULL if we don't have thresholds */
4099 /* Copy thresholds and find current threshold */
4100 new->current_threshold
= -1;
4101 for (i
= 0, j
= 0; i
< thresholds
->primary
->size
; i
++) {
4102 if (thresholds
->primary
->entries
[i
].eventfd
== eventfd
)
4105 new->entries
[j
] = thresholds
->primary
->entries
[i
];
4106 if (new->entries
[j
].threshold
< usage
) {
4108 * new->current_threshold will not be used
4109 * until rcu_assign_pointer(), so it's safe to increment
4112 ++new->current_threshold
;
4118 /* Swap primary and spare array */
4119 thresholds
->spare
= thresholds
->primary
;
4120 rcu_assign_pointer(thresholds
->primary
, new);
4122 /* To be sure that nobody uses thresholds */
4125 mutex_unlock(&memcg
->thresholds_lock
);
4128 static int mem_cgroup_oom_register_event(struct cgroup
*cgrp
,
4129 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
4131 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4132 struct mem_cgroup_eventfd_list
*event
;
4133 int type
= MEMFILE_TYPE(cft
->private);
4135 BUG_ON(type
!= _OOM_TYPE
);
4136 event
= kmalloc(sizeof(*event
), GFP_KERNEL
);
4140 mutex_lock(&memcg_oom_mutex
);
4142 event
->eventfd
= eventfd
;
4143 list_add(&event
->list
, &memcg
->oom_notify
);
4145 /* already in OOM ? */
4146 if (atomic_read(&memcg
->oom_lock
))
4147 eventfd_signal(eventfd
, 1);
4148 mutex_unlock(&memcg_oom_mutex
);
4153 static void mem_cgroup_oom_unregister_event(struct cgroup
*cgrp
,
4154 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
4156 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgrp
);
4157 struct mem_cgroup_eventfd_list
*ev
, *tmp
;
4158 int type
= MEMFILE_TYPE(cft
->private);
4160 BUG_ON(type
!= _OOM_TYPE
);
4162 mutex_lock(&memcg_oom_mutex
);
4164 list_for_each_entry_safe(ev
, tmp
, &mem
->oom_notify
, list
) {
4165 if (ev
->eventfd
== eventfd
) {
4166 list_del(&ev
->list
);
4171 mutex_unlock(&memcg_oom_mutex
);
4174 static int mem_cgroup_oom_control_read(struct cgroup
*cgrp
,
4175 struct cftype
*cft
, struct cgroup_map_cb
*cb
)
4177 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgrp
);
4179 cb
->fill(cb
, "oom_kill_disable", mem
->oom_kill_disable
);
4181 if (atomic_read(&mem
->oom_lock
))
4182 cb
->fill(cb
, "under_oom", 1);
4184 cb
->fill(cb
, "under_oom", 0);
4188 static int mem_cgroup_oom_control_write(struct cgroup
*cgrp
,
4189 struct cftype
*cft
, u64 val
)
4191 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgrp
);
4192 struct mem_cgroup
*parent
;
4194 /* cannot set to root cgroup and only 0 and 1 are allowed */
4195 if (!cgrp
->parent
|| !((val
== 0) || (val
== 1)))
4198 parent
= mem_cgroup_from_cont(cgrp
->parent
);
4201 /* oom-kill-disable is a flag for subhierarchy. */
4202 if ((parent
->use_hierarchy
) ||
4203 (mem
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
4207 mem
->oom_kill_disable
= val
;
4209 memcg_oom_recover(mem
);
4214 static struct cftype mem_cgroup_files
[] = {
4216 .name
= "usage_in_bytes",
4217 .private = MEMFILE_PRIVATE(_MEM
, RES_USAGE
),
4218 .read_u64
= mem_cgroup_read
,
4219 .register_event
= mem_cgroup_usage_register_event
,
4220 .unregister_event
= mem_cgroup_usage_unregister_event
,
4223 .name
= "max_usage_in_bytes",
4224 .private = MEMFILE_PRIVATE(_MEM
, RES_MAX_USAGE
),
4225 .trigger
= mem_cgroup_reset
,
4226 .read_u64
= mem_cgroup_read
,
4229 .name
= "limit_in_bytes",
4230 .private = MEMFILE_PRIVATE(_MEM
, RES_LIMIT
),
4231 .write_string
= mem_cgroup_write
,
4232 .read_u64
= mem_cgroup_read
,
4235 .name
= "soft_limit_in_bytes",
4236 .private = MEMFILE_PRIVATE(_MEM
, RES_SOFT_LIMIT
),
4237 .write_string
= mem_cgroup_write
,
4238 .read_u64
= mem_cgroup_read
,
4242 .private = MEMFILE_PRIVATE(_MEM
, RES_FAILCNT
),
4243 .trigger
= mem_cgroup_reset
,
4244 .read_u64
= mem_cgroup_read
,
4248 .read_map
= mem_control_stat_show
,
4251 .name
= "force_empty",
4252 .trigger
= mem_cgroup_force_empty_write
,
4255 .name
= "use_hierarchy",
4256 .write_u64
= mem_cgroup_hierarchy_write
,
4257 .read_u64
= mem_cgroup_hierarchy_read
,
4260 .name
= "swappiness",
4261 .read_u64
= mem_cgroup_swappiness_read
,
4262 .write_u64
= mem_cgroup_swappiness_write
,
4265 .name
= "move_charge_at_immigrate",
4266 .read_u64
= mem_cgroup_move_charge_read
,
4267 .write_u64
= mem_cgroup_move_charge_write
,
4270 .name
= "oom_control",
4271 .read_map
= mem_cgroup_oom_control_read
,
4272 .write_u64
= mem_cgroup_oom_control_write
,
4273 .register_event
= mem_cgroup_oom_register_event
,
4274 .unregister_event
= mem_cgroup_oom_unregister_event
,
4275 .private = MEMFILE_PRIVATE(_OOM_TYPE
, OOM_CONTROL
),
4279 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4280 static struct cftype memsw_cgroup_files
[] = {
4282 .name
= "memsw.usage_in_bytes",
4283 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_USAGE
),
4284 .read_u64
= mem_cgroup_read
,
4285 .register_event
= mem_cgroup_usage_register_event
,
4286 .unregister_event
= mem_cgroup_usage_unregister_event
,
4289 .name
= "memsw.max_usage_in_bytes",
4290 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_MAX_USAGE
),
4291 .trigger
= mem_cgroup_reset
,
4292 .read_u64
= mem_cgroup_read
,
4295 .name
= "memsw.limit_in_bytes",
4296 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_LIMIT
),
4297 .write_string
= mem_cgroup_write
,
4298 .read_u64
= mem_cgroup_read
,
4301 .name
= "memsw.failcnt",
4302 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_FAILCNT
),
4303 .trigger
= mem_cgroup_reset
,
4304 .read_u64
= mem_cgroup_read
,
4308 static int register_memsw_files(struct cgroup
*cont
, struct cgroup_subsys
*ss
)
4310 if (!do_swap_account
)
4312 return cgroup_add_files(cont
, ss
, memsw_cgroup_files
,
4313 ARRAY_SIZE(memsw_cgroup_files
));
4316 static int register_memsw_files(struct cgroup
*cont
, struct cgroup_subsys
*ss
)
4322 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup
*mem
, int node
)
4324 struct mem_cgroup_per_node
*pn
;
4325 struct mem_cgroup_per_zone
*mz
;
4327 int zone
, tmp
= node
;
4329 * This routine is called against possible nodes.
4330 * But it's BUG to call kmalloc() against offline node.
4332 * TODO: this routine can waste much memory for nodes which will
4333 * never be onlined. It's better to use memory hotplug callback
4336 if (!node_state(node
, N_NORMAL_MEMORY
))
4338 pn
= kzalloc_node(sizeof(*pn
), GFP_KERNEL
, tmp
);
4342 mem
->info
.nodeinfo
[node
] = pn
;
4343 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
4344 mz
= &pn
->zoneinfo
[zone
];
4346 INIT_LIST_HEAD(&mz
->lists
[l
]);
4347 mz
->usage_in_excess
= 0;
4348 mz
->on_tree
= false;
4354 static void free_mem_cgroup_per_zone_info(struct mem_cgroup
*mem
, int node
)
4356 kfree(mem
->info
.nodeinfo
[node
]);
4359 static struct mem_cgroup
*mem_cgroup_alloc(void)
4361 struct mem_cgroup
*mem
;
4362 int size
= sizeof(struct mem_cgroup
);
4364 /* Can be very big if MAX_NUMNODES is very big */
4365 if (size
< PAGE_SIZE
)
4366 mem
= kzalloc(size
, GFP_KERNEL
);
4368 mem
= vzalloc(size
);
4373 mem
->stat
= alloc_percpu(struct mem_cgroup_stat_cpu
);
4376 spin_lock_init(&mem
->pcp_counter_lock
);
4380 if (size
< PAGE_SIZE
)
4388 * At destroying mem_cgroup, references from swap_cgroup can remain.
4389 * (scanning all at force_empty is too costly...)
4391 * Instead of clearing all references at force_empty, we remember
4392 * the number of reference from swap_cgroup and free mem_cgroup when
4393 * it goes down to 0.
4395 * Removal of cgroup itself succeeds regardless of refs from swap.
4398 static void __mem_cgroup_free(struct mem_cgroup
*mem
)
4402 mem_cgroup_remove_from_trees(mem
);
4403 free_css_id(&mem_cgroup_subsys
, &mem
->css
);
4405 for_each_node_state(node
, N_POSSIBLE
)
4406 free_mem_cgroup_per_zone_info(mem
, node
);
4408 free_percpu(mem
->stat
);
4409 if (sizeof(struct mem_cgroup
) < PAGE_SIZE
)
4415 static void mem_cgroup_get(struct mem_cgroup
*mem
)
4417 atomic_inc(&mem
->refcnt
);
4420 static void __mem_cgroup_put(struct mem_cgroup
*mem
, int count
)
4422 if (atomic_sub_and_test(count
, &mem
->refcnt
)) {
4423 struct mem_cgroup
*parent
= parent_mem_cgroup(mem
);
4424 __mem_cgroup_free(mem
);
4426 mem_cgroup_put(parent
);
4430 static void mem_cgroup_put(struct mem_cgroup
*mem
)
4432 __mem_cgroup_put(mem
, 1);
4436 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
4438 static struct mem_cgroup
*parent_mem_cgroup(struct mem_cgroup
*mem
)
4440 if (!mem
->res
.parent
)
4442 return mem_cgroup_from_res_counter(mem
->res
.parent
, res
);
4445 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4446 static void __init
enable_swap_cgroup(void)
4448 if (!mem_cgroup_disabled() && really_do_swap_account
)
4449 do_swap_account
= 1;
4452 static void __init
enable_swap_cgroup(void)
4457 static int mem_cgroup_soft_limit_tree_init(void)
4459 struct mem_cgroup_tree_per_node
*rtpn
;
4460 struct mem_cgroup_tree_per_zone
*rtpz
;
4461 int tmp
, node
, zone
;
4463 for_each_node_state(node
, N_POSSIBLE
) {
4465 if (!node_state(node
, N_NORMAL_MEMORY
))
4467 rtpn
= kzalloc_node(sizeof(*rtpn
), GFP_KERNEL
, tmp
);
4471 soft_limit_tree
.rb_tree_per_node
[node
] = rtpn
;
4473 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
4474 rtpz
= &rtpn
->rb_tree_per_zone
[zone
];
4475 rtpz
->rb_root
= RB_ROOT
;
4476 spin_lock_init(&rtpz
->lock
);
4482 static struct cgroup_subsys_state
* __ref
4483 mem_cgroup_create(struct cgroup_subsys
*ss
, struct cgroup
*cont
)
4485 struct mem_cgroup
*mem
, *parent
;
4486 long error
= -ENOMEM
;
4489 mem
= mem_cgroup_alloc();
4491 return ERR_PTR(error
);
4493 for_each_node_state(node
, N_POSSIBLE
)
4494 if (alloc_mem_cgroup_per_zone_info(mem
, node
))
4498 if (cont
->parent
== NULL
) {
4500 enable_swap_cgroup();
4502 root_mem_cgroup
= mem
;
4503 if (mem_cgroup_soft_limit_tree_init())
4505 for_each_possible_cpu(cpu
) {
4506 struct memcg_stock_pcp
*stock
=
4507 &per_cpu(memcg_stock
, cpu
);
4508 INIT_WORK(&stock
->work
, drain_local_stock
);
4510 hotcpu_notifier(memcg_cpu_hotplug_callback
, 0);
4512 parent
= mem_cgroup_from_cont(cont
->parent
);
4513 mem
->use_hierarchy
= parent
->use_hierarchy
;
4514 mem
->oom_kill_disable
= parent
->oom_kill_disable
;
4517 if (parent
&& parent
->use_hierarchy
) {
4518 res_counter_init(&mem
->res
, &parent
->res
);
4519 res_counter_init(&mem
->memsw
, &parent
->memsw
);
4521 * We increment refcnt of the parent to ensure that we can
4522 * safely access it on res_counter_charge/uncharge.
4523 * This refcnt will be decremented when freeing this
4524 * mem_cgroup(see mem_cgroup_put).
4526 mem_cgroup_get(parent
);
4528 res_counter_init(&mem
->res
, NULL
);
4529 res_counter_init(&mem
->memsw
, NULL
);
4531 mem
->last_scanned_child
= 0;
4532 spin_lock_init(&mem
->reclaim_param_lock
);
4533 INIT_LIST_HEAD(&mem
->oom_notify
);
4536 mem
->swappiness
= get_swappiness(parent
);
4537 atomic_set(&mem
->refcnt
, 1);
4538 mem
->move_charge_at_immigrate
= 0;
4539 mutex_init(&mem
->thresholds_lock
);
4542 __mem_cgroup_free(mem
);
4543 root_mem_cgroup
= NULL
;
4544 return ERR_PTR(error
);
4547 static int mem_cgroup_pre_destroy(struct cgroup_subsys
*ss
,
4548 struct cgroup
*cont
)
4550 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
4552 return mem_cgroup_force_empty(mem
, false);
4555 static void mem_cgroup_destroy(struct cgroup_subsys
*ss
,
4556 struct cgroup
*cont
)
4558 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
4560 mem_cgroup_put(mem
);
4563 static int mem_cgroup_populate(struct cgroup_subsys
*ss
,
4564 struct cgroup
*cont
)
4568 ret
= cgroup_add_files(cont
, ss
, mem_cgroup_files
,
4569 ARRAY_SIZE(mem_cgroup_files
));
4572 ret
= register_memsw_files(cont
, ss
);
4577 /* Handlers for move charge at task migration. */
4578 #define PRECHARGE_COUNT_AT_ONCE 256
4579 static int mem_cgroup_do_precharge(unsigned long count
)
4582 int batch_count
= PRECHARGE_COUNT_AT_ONCE
;
4583 struct mem_cgroup
*mem
= mc
.to
;
4585 if (mem_cgroup_is_root(mem
)) {
4586 mc
.precharge
+= count
;
4587 /* we don't need css_get for root */
4590 /* try to charge at once */
4592 struct res_counter
*dummy
;
4594 * "mem" cannot be under rmdir() because we've already checked
4595 * by cgroup_lock_live_cgroup() that it is not removed and we
4596 * are still under the same cgroup_mutex. So we can postpone
4599 if (res_counter_charge(&mem
->res
, PAGE_SIZE
* count
, &dummy
))
4601 if (do_swap_account
&& res_counter_charge(&mem
->memsw
,
4602 PAGE_SIZE
* count
, &dummy
)) {
4603 res_counter_uncharge(&mem
->res
, PAGE_SIZE
* count
);
4606 mc
.precharge
+= count
;
4610 /* fall back to one by one charge */
4612 if (signal_pending(current
)) {
4616 if (!batch_count
--) {
4617 batch_count
= PRECHARGE_COUNT_AT_ONCE
;
4620 ret
= __mem_cgroup_try_charge(NULL
, GFP_KERNEL
, &mem
, false,
4623 /* mem_cgroup_clear_mc() will do uncharge later */
4631 * is_target_pte_for_mc - check a pte whether it is valid for move charge
4632 * @vma: the vma the pte to be checked belongs
4633 * @addr: the address corresponding to the pte to be checked
4634 * @ptent: the pte to be checked
4635 * @target: the pointer the target page or swap ent will be stored(can be NULL)
4638 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
4639 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
4640 * move charge. if @target is not NULL, the page is stored in target->page
4641 * with extra refcnt got(Callers should handle it).
4642 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
4643 * target for charge migration. if @target is not NULL, the entry is stored
4646 * Called with pte lock held.
4653 enum mc_target_type
{
4654 MC_TARGET_NONE
, /* not used */
4659 static struct page
*mc_handle_present_pte(struct vm_area_struct
*vma
,
4660 unsigned long addr
, pte_t ptent
)
4662 struct page
*page
= vm_normal_page(vma
, addr
, ptent
);
4664 if (!page
|| !page_mapped(page
))
4666 if (PageAnon(page
)) {
4667 /* we don't move shared anon */
4668 if (!move_anon() || page_mapcount(page
) > 2)
4670 } else if (!move_file())
4671 /* we ignore mapcount for file pages */
4673 if (!get_page_unless_zero(page
))
4679 static struct page
*mc_handle_swap_pte(struct vm_area_struct
*vma
,
4680 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
4683 struct page
*page
= NULL
;
4684 swp_entry_t ent
= pte_to_swp_entry(ptent
);
4686 if (!move_anon() || non_swap_entry(ent
))
4688 usage_count
= mem_cgroup_count_swap_user(ent
, &page
);
4689 if (usage_count
> 1) { /* we don't move shared anon */
4694 if (do_swap_account
)
4695 entry
->val
= ent
.val
;
4700 static struct page
*mc_handle_file_pte(struct vm_area_struct
*vma
,
4701 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
4703 struct page
*page
= NULL
;
4704 struct inode
*inode
;
4705 struct address_space
*mapping
;
4708 if (!vma
->vm_file
) /* anonymous vma */
4713 inode
= vma
->vm_file
->f_path
.dentry
->d_inode
;
4714 mapping
= vma
->vm_file
->f_mapping
;
4715 if (pte_none(ptent
))
4716 pgoff
= linear_page_index(vma
, addr
);
4717 else /* pte_file(ptent) is true */
4718 pgoff
= pte_to_pgoff(ptent
);
4720 /* page is moved even if it's not RSS of this task(page-faulted). */
4721 if (!mapping_cap_swap_backed(mapping
)) { /* normal file */
4722 page
= find_get_page(mapping
, pgoff
);
4723 } else { /* shmem/tmpfs file. we should take account of swap too. */
4725 mem_cgroup_get_shmem_target(inode
, pgoff
, &page
, &ent
);
4726 if (do_swap_account
)
4727 entry
->val
= ent
.val
;
4733 static int is_target_pte_for_mc(struct vm_area_struct
*vma
,
4734 unsigned long addr
, pte_t ptent
, union mc_target
*target
)
4736 struct page
*page
= NULL
;
4737 struct page_cgroup
*pc
;
4739 swp_entry_t ent
= { .val
= 0 };
4741 if (pte_present(ptent
))
4742 page
= mc_handle_present_pte(vma
, addr
, ptent
);
4743 else if (is_swap_pte(ptent
))
4744 page
= mc_handle_swap_pte(vma
, addr
, ptent
, &ent
);
4745 else if (pte_none(ptent
) || pte_file(ptent
))
4746 page
= mc_handle_file_pte(vma
, addr
, ptent
, &ent
);
4748 if (!page
&& !ent
.val
)
4751 pc
= lookup_page_cgroup(page
);
4753 * Do only loose check w/o page_cgroup lock.
4754 * mem_cgroup_move_account() checks the pc is valid or not under
4757 if (PageCgroupUsed(pc
) && pc
->mem_cgroup
== mc
.from
) {
4758 ret
= MC_TARGET_PAGE
;
4760 target
->page
= page
;
4762 if (!ret
|| !target
)
4765 /* There is a swap entry and a page doesn't exist or isn't charged */
4766 if (ent
.val
&& !ret
&&
4767 css_id(&mc
.from
->css
) == lookup_swap_cgroup(ent
)) {
4768 ret
= MC_TARGET_SWAP
;
4775 static int mem_cgroup_count_precharge_pte_range(pmd_t
*pmd
,
4776 unsigned long addr
, unsigned long end
,
4777 struct mm_walk
*walk
)
4779 struct vm_area_struct
*vma
= walk
->private;
4783 split_huge_page_pmd(walk
->mm
, pmd
);
4785 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
4786 for (; addr
!= end
; pte
++, addr
+= PAGE_SIZE
)
4787 if (is_target_pte_for_mc(vma
, addr
, *pte
, NULL
))
4788 mc
.precharge
++; /* increment precharge temporarily */
4789 pte_unmap_unlock(pte
- 1, ptl
);
4795 static unsigned long mem_cgroup_count_precharge(struct mm_struct
*mm
)
4797 unsigned long precharge
;
4798 struct vm_area_struct
*vma
;
4800 down_read(&mm
->mmap_sem
);
4801 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
4802 struct mm_walk mem_cgroup_count_precharge_walk
= {
4803 .pmd_entry
= mem_cgroup_count_precharge_pte_range
,
4807 if (is_vm_hugetlb_page(vma
))
4809 walk_page_range(vma
->vm_start
, vma
->vm_end
,
4810 &mem_cgroup_count_precharge_walk
);
4812 up_read(&mm
->mmap_sem
);
4814 precharge
= mc
.precharge
;
4820 static int mem_cgroup_precharge_mc(struct mm_struct
*mm
)
4822 unsigned long precharge
= mem_cgroup_count_precharge(mm
);
4824 VM_BUG_ON(mc
.moving_task
);
4825 mc
.moving_task
= current
;
4826 return mem_cgroup_do_precharge(precharge
);
4829 /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
4830 static void __mem_cgroup_clear_mc(void)
4832 struct mem_cgroup
*from
= mc
.from
;
4833 struct mem_cgroup
*to
= mc
.to
;
4835 /* we must uncharge all the leftover precharges from mc.to */
4837 __mem_cgroup_cancel_charge(mc
.to
, mc
.precharge
);
4841 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
4842 * we must uncharge here.
4844 if (mc
.moved_charge
) {
4845 __mem_cgroup_cancel_charge(mc
.from
, mc
.moved_charge
);
4846 mc
.moved_charge
= 0;
4848 /* we must fixup refcnts and charges */
4849 if (mc
.moved_swap
) {
4850 /* uncharge swap account from the old cgroup */
4851 if (!mem_cgroup_is_root(mc
.from
))
4852 res_counter_uncharge(&mc
.from
->memsw
,
4853 PAGE_SIZE
* mc
.moved_swap
);
4854 __mem_cgroup_put(mc
.from
, mc
.moved_swap
);
4856 if (!mem_cgroup_is_root(mc
.to
)) {
4858 * we charged both to->res and to->memsw, so we should
4861 res_counter_uncharge(&mc
.to
->res
,
4862 PAGE_SIZE
* mc
.moved_swap
);
4864 /* we've already done mem_cgroup_get(mc.to) */
4867 memcg_oom_recover(from
);
4868 memcg_oom_recover(to
);
4869 wake_up_all(&mc
.waitq
);
4872 static void mem_cgroup_clear_mc(void)
4874 struct mem_cgroup
*from
= mc
.from
;
4877 * we must clear moving_task before waking up waiters at the end of
4880 mc
.moving_task
= NULL
;
4881 __mem_cgroup_clear_mc();
4882 spin_lock(&mc
.lock
);
4885 spin_unlock(&mc
.lock
);
4886 mem_cgroup_end_move(from
);
4889 static int mem_cgroup_can_attach(struct cgroup_subsys
*ss
,
4890 struct cgroup
*cgroup
,
4891 struct task_struct
*p
,
4895 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgroup
);
4897 if (mem
->move_charge_at_immigrate
) {
4898 struct mm_struct
*mm
;
4899 struct mem_cgroup
*from
= mem_cgroup_from_task(p
);
4901 VM_BUG_ON(from
== mem
);
4903 mm
= get_task_mm(p
);
4906 /* We move charges only when we move a owner of the mm */
4907 if (mm
->owner
== p
) {
4910 VM_BUG_ON(mc
.precharge
);
4911 VM_BUG_ON(mc
.moved_charge
);
4912 VM_BUG_ON(mc
.moved_swap
);
4913 mem_cgroup_start_move(from
);
4914 spin_lock(&mc
.lock
);
4917 spin_unlock(&mc
.lock
);
4918 /* We set mc.moving_task later */
4920 ret
= mem_cgroup_precharge_mc(mm
);
4922 mem_cgroup_clear_mc();
4929 static void mem_cgroup_cancel_attach(struct cgroup_subsys
*ss
,
4930 struct cgroup
*cgroup
,
4931 struct task_struct
*p
,
4934 mem_cgroup_clear_mc();
4937 static int mem_cgroup_move_charge_pte_range(pmd_t
*pmd
,
4938 unsigned long addr
, unsigned long end
,
4939 struct mm_walk
*walk
)
4942 struct vm_area_struct
*vma
= walk
->private;
4946 split_huge_page_pmd(walk
->mm
, pmd
);
4948 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
4949 for (; addr
!= end
; addr
+= PAGE_SIZE
) {
4950 pte_t ptent
= *(pte
++);
4951 union mc_target target
;
4954 struct page_cgroup
*pc
;
4960 type
= is_target_pte_for_mc(vma
, addr
, ptent
, &target
);
4962 case MC_TARGET_PAGE
:
4964 if (isolate_lru_page(page
))
4966 pc
= lookup_page_cgroup(page
);
4967 if (!mem_cgroup_move_account(pc
,
4968 mc
.from
, mc
.to
, false, PAGE_SIZE
)) {
4970 /* we uncharge from mc.from later. */
4973 putback_lru_page(page
);
4974 put
: /* is_target_pte_for_mc() gets the page */
4977 case MC_TARGET_SWAP
:
4979 if (!mem_cgroup_move_swap_account(ent
,
4980 mc
.from
, mc
.to
, false)) {
4982 /* we fixup refcnts and charges later. */
4990 pte_unmap_unlock(pte
- 1, ptl
);
4995 * We have consumed all precharges we got in can_attach().
4996 * We try charge one by one, but don't do any additional
4997 * charges to mc.to if we have failed in charge once in attach()
5000 ret
= mem_cgroup_do_precharge(1);
5008 static void mem_cgroup_move_charge(struct mm_struct
*mm
)
5010 struct vm_area_struct
*vma
;
5012 lru_add_drain_all();
5014 if (unlikely(!down_read_trylock(&mm
->mmap_sem
))) {
5016 * Someone who are holding the mmap_sem might be waiting in
5017 * waitq. So we cancel all extra charges, wake up all waiters,
5018 * and retry. Because we cancel precharges, we might not be able
5019 * to move enough charges, but moving charge is a best-effort
5020 * feature anyway, so it wouldn't be a big problem.
5022 __mem_cgroup_clear_mc();
5026 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
5028 struct mm_walk mem_cgroup_move_charge_walk
= {
5029 .pmd_entry
= mem_cgroup_move_charge_pte_range
,
5033 if (is_vm_hugetlb_page(vma
))
5035 ret
= walk_page_range(vma
->vm_start
, vma
->vm_end
,
5036 &mem_cgroup_move_charge_walk
);
5039 * means we have consumed all precharges and failed in
5040 * doing additional charge. Just abandon here.
5044 up_read(&mm
->mmap_sem
);
5047 static void mem_cgroup_move_task(struct cgroup_subsys
*ss
,
5048 struct cgroup
*cont
,
5049 struct cgroup
*old_cont
,
5050 struct task_struct
*p
,
5053 struct mm_struct
*mm
;
5056 /* no need to move charge */
5059 mm
= get_task_mm(p
);
5061 mem_cgroup_move_charge(mm
);
5064 mem_cgroup_clear_mc();
5066 #else /* !CONFIG_MMU */
5067 static int mem_cgroup_can_attach(struct cgroup_subsys
*ss
,
5068 struct cgroup
*cgroup
,
5069 struct task_struct
*p
,
5074 static void mem_cgroup_cancel_attach(struct cgroup_subsys
*ss
,
5075 struct cgroup
*cgroup
,
5076 struct task_struct
*p
,
5080 static void mem_cgroup_move_task(struct cgroup_subsys
*ss
,
5081 struct cgroup
*cont
,
5082 struct cgroup
*old_cont
,
5083 struct task_struct
*p
,
5089 struct cgroup_subsys mem_cgroup_subsys
= {
5091 .subsys_id
= mem_cgroup_subsys_id
,
5092 .create
= mem_cgroup_create
,
5093 .pre_destroy
= mem_cgroup_pre_destroy
,
5094 .destroy
= mem_cgroup_destroy
,
5095 .populate
= mem_cgroup_populate
,
5096 .can_attach
= mem_cgroup_can_attach
,
5097 .cancel_attach
= mem_cgroup_cancel_attach
,
5098 .attach
= mem_cgroup_move_task
,
5103 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
5104 static int __init
enable_swap_account(char *s
)
5106 /* consider enabled if no parameter or 1 is given */
5107 if (!(*s
) || !strcmp(s
, "=1"))
5108 really_do_swap_account
= 1;
5109 else if (!strcmp(s
, "=0"))
5110 really_do_swap_account
= 0;
5113 __setup("swapaccount", enable_swap_account
);
5115 static int __init
disable_swap_account(char *s
)
5117 printk_once("noswapaccount is deprecated and will be removed in 2.6.40. Use swapaccount=0 instead\n");
5118 enable_swap_account("=0");
5121 __setup("noswapaccount", disable_swap_account
);