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 * Kernel Memory Controller
14 * Copyright (C) 2012 Parallels Inc. and Google Inc.
15 * Authors: Glauber Costa and Suleiman Souhlal
18 * Charge lifetime sanitation
19 * Lockless page tracking & accounting
20 * Unified hierarchy configuration model
21 * Copyright (C) 2015 Red Hat, Inc., Johannes Weiner
23 * This program is free software; you can redistribute it and/or modify
24 * it under the terms of the GNU General Public License as published by
25 * the Free Software Foundation; either version 2 of the License, or
26 * (at your option) any later version.
28 * This program is distributed in the hope that it will be useful,
29 * but WITHOUT ANY WARRANTY; without even the implied warranty of
30 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
31 * GNU General Public License for more details.
34 #include <linux/page_counter.h>
35 #include <linux/memcontrol.h>
36 #include <linux/cgroup.h>
38 #include <linux/sched/mm.h>
39 #include <linux/shmem_fs.h>
40 #include <linux/hugetlb.h>
41 #include <linux/pagemap.h>
42 #include <linux/smp.h>
43 #include <linux/page-flags.h>
44 #include <linux/backing-dev.h>
45 #include <linux/bit_spinlock.h>
46 #include <linux/rcupdate.h>
47 #include <linux/limits.h>
48 #include <linux/export.h>
49 #include <linux/mutex.h>
50 #include <linux/rbtree.h>
51 #include <linux/slab.h>
52 #include <linux/swap.h>
53 #include <linux/swapops.h>
54 #include <linux/spinlock.h>
55 #include <linux/eventfd.h>
56 #include <linux/poll.h>
57 #include <linux/sort.h>
59 #include <linux/seq_file.h>
60 #include <linux/vmpressure.h>
61 #include <linux/mm_inline.h>
62 #include <linux/swap_cgroup.h>
63 #include <linux/cpu.h>
64 #include <linux/oom.h>
65 #include <linux/lockdep.h>
66 #include <linux/file.h>
67 #include <linux/tracehook.h>
73 #include <linux/uaccess.h>
75 #include <trace/events/vmscan.h>
77 struct cgroup_subsys memory_cgrp_subsys __read_mostly
;
78 EXPORT_SYMBOL(memory_cgrp_subsys
);
80 struct mem_cgroup
*root_mem_cgroup __read_mostly
;
82 #define MEM_CGROUP_RECLAIM_RETRIES 5
84 /* Socket memory accounting disabled? */
85 static bool cgroup_memory_nosocket
;
87 /* Kernel memory accounting disabled? */
88 static bool cgroup_memory_nokmem
;
90 /* Whether the swap controller is active */
91 #ifdef CONFIG_MEMCG_SWAP
92 int do_swap_account __read_mostly
;
94 #define do_swap_account 0
97 /* Whether legacy memory+swap accounting is active */
98 static bool do_memsw_account(void)
100 return !cgroup_subsys_on_dfl(memory_cgrp_subsys
) && do_swap_account
;
103 static const char *const mem_cgroup_lru_names
[] = {
111 #define THRESHOLDS_EVENTS_TARGET 128
112 #define SOFTLIMIT_EVENTS_TARGET 1024
113 #define NUMAINFO_EVENTS_TARGET 1024
116 * Cgroups above their limits are maintained in a RB-Tree, independent of
117 * their hierarchy representation
120 struct mem_cgroup_tree_per_node
{
121 struct rb_root rb_root
;
122 struct rb_node
*rb_rightmost
;
126 struct mem_cgroup_tree
{
127 struct mem_cgroup_tree_per_node
*rb_tree_per_node
[MAX_NUMNODES
];
130 static struct mem_cgroup_tree soft_limit_tree __read_mostly
;
133 struct mem_cgroup_eventfd_list
{
134 struct list_head list
;
135 struct eventfd_ctx
*eventfd
;
139 * cgroup_event represents events which userspace want to receive.
141 struct mem_cgroup_event
{
143 * memcg which the event belongs to.
145 struct mem_cgroup
*memcg
;
147 * eventfd to signal userspace about the event.
149 struct eventfd_ctx
*eventfd
;
151 * Each of these stored in a list by the cgroup.
153 struct list_head list
;
155 * register_event() callback will be used to add new userspace
156 * waiter for changes related to this event. Use eventfd_signal()
157 * on eventfd to send notification to userspace.
159 int (*register_event
)(struct mem_cgroup
*memcg
,
160 struct eventfd_ctx
*eventfd
, const char *args
);
162 * unregister_event() callback will be called when userspace closes
163 * the eventfd or on cgroup removing. This callback must be set,
164 * if you want provide notification functionality.
166 void (*unregister_event
)(struct mem_cgroup
*memcg
,
167 struct eventfd_ctx
*eventfd
);
169 * All fields below needed to unregister event when
170 * userspace closes eventfd.
173 wait_queue_head_t
*wqh
;
174 wait_queue_entry_t wait
;
175 struct work_struct remove
;
178 static void mem_cgroup_threshold(struct mem_cgroup
*memcg
);
179 static void mem_cgroup_oom_notify(struct mem_cgroup
*memcg
);
181 /* Stuffs for move charges at task migration. */
183 * Types of charges to be moved.
185 #define MOVE_ANON 0x1U
186 #define MOVE_FILE 0x2U
187 #define MOVE_MASK (MOVE_ANON | MOVE_FILE)
189 /* "mc" and its members are protected by cgroup_mutex */
190 static struct move_charge_struct
{
191 spinlock_t lock
; /* for from, to */
192 struct mm_struct
*mm
;
193 struct mem_cgroup
*from
;
194 struct mem_cgroup
*to
;
196 unsigned long precharge
;
197 unsigned long moved_charge
;
198 unsigned long moved_swap
;
199 struct task_struct
*moving_task
; /* a task moving charges */
200 wait_queue_head_t waitq
; /* a waitq for other context */
202 .lock
= __SPIN_LOCK_UNLOCKED(mc
.lock
),
203 .waitq
= __WAIT_QUEUE_HEAD_INITIALIZER(mc
.waitq
),
207 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
208 * limit reclaim to prevent infinite loops, if they ever occur.
210 #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100
211 #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2
214 MEM_CGROUP_CHARGE_TYPE_CACHE
= 0,
215 MEM_CGROUP_CHARGE_TYPE_ANON
,
216 MEM_CGROUP_CHARGE_TYPE_SWAPOUT
, /* for accounting swapcache */
217 MEM_CGROUP_CHARGE_TYPE_DROP
, /* a page was unused swap cache */
221 /* for encoding cft->private value on file */
230 #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val))
231 #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff)
232 #define MEMFILE_ATTR(val) ((val) & 0xffff)
233 /* Used for OOM nofiier */
234 #define OOM_CONTROL (0)
237 * Iteration constructs for visiting all cgroups (under a tree). If
238 * loops are exited prematurely (break), mem_cgroup_iter_break() must
239 * be used for reference counting.
241 #define for_each_mem_cgroup_tree(iter, root) \
242 for (iter = mem_cgroup_iter(root, NULL, NULL); \
244 iter = mem_cgroup_iter(root, iter, NULL))
246 #define for_each_mem_cgroup(iter) \
247 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
249 iter = mem_cgroup_iter(NULL, iter, NULL))
251 /* Some nice accessors for the vmpressure. */
252 struct vmpressure
*memcg_to_vmpressure(struct mem_cgroup
*memcg
)
255 memcg
= root_mem_cgroup
;
256 return &memcg
->vmpressure
;
259 struct cgroup_subsys_state
*vmpressure_to_css(struct vmpressure
*vmpr
)
261 return &container_of(vmpr
, struct mem_cgroup
, vmpressure
)->css
;
264 #ifdef CONFIG_MEMCG_KMEM
266 * This will be the memcg's index in each cache's ->memcg_params.memcg_caches.
267 * The main reason for not using cgroup id for this:
268 * this works better in sparse environments, where we have a lot of memcgs,
269 * but only a few kmem-limited. Or also, if we have, for instance, 200
270 * memcgs, and none but the 200th is kmem-limited, we'd have to have a
271 * 200 entry array for that.
273 * The current size of the caches array is stored in memcg_nr_cache_ids. It
274 * will double each time we have to increase it.
276 static DEFINE_IDA(memcg_cache_ida
);
277 int memcg_nr_cache_ids
;
279 /* Protects memcg_nr_cache_ids */
280 static DECLARE_RWSEM(memcg_cache_ids_sem
);
282 void memcg_get_cache_ids(void)
284 down_read(&memcg_cache_ids_sem
);
287 void memcg_put_cache_ids(void)
289 up_read(&memcg_cache_ids_sem
);
293 * MIN_SIZE is different than 1, because we would like to avoid going through
294 * the alloc/free process all the time. In a small machine, 4 kmem-limited
295 * cgroups is a reasonable guess. In the future, it could be a parameter or
296 * tunable, but that is strictly not necessary.
298 * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get
299 * this constant directly from cgroup, but it is understandable that this is
300 * better kept as an internal representation in cgroup.c. In any case, the
301 * cgrp_id space is not getting any smaller, and we don't have to necessarily
302 * increase ours as well if it increases.
304 #define MEMCG_CACHES_MIN_SIZE 4
305 #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX
308 * A lot of the calls to the cache allocation functions are expected to be
309 * inlined by the compiler. Since the calls to memcg_kmem_get_cache are
310 * conditional to this static branch, we'll have to allow modules that does
311 * kmem_cache_alloc and the such to see this symbol as well
313 DEFINE_STATIC_KEY_FALSE(memcg_kmem_enabled_key
);
314 EXPORT_SYMBOL(memcg_kmem_enabled_key
);
316 struct workqueue_struct
*memcg_kmem_cache_wq
;
318 static int memcg_shrinker_map_size
;
319 static DEFINE_MUTEX(memcg_shrinker_map_mutex
);
321 static void memcg_free_shrinker_map_rcu(struct rcu_head
*head
)
323 kvfree(container_of(head
, struct memcg_shrinker_map
, rcu
));
326 static int memcg_expand_one_shrinker_map(struct mem_cgroup
*memcg
,
327 int size
, int old_size
)
329 struct memcg_shrinker_map
*new, *old
;
332 lockdep_assert_held(&memcg_shrinker_map_mutex
);
335 old
= rcu_dereference_protected(
336 mem_cgroup_nodeinfo(memcg
, nid
)->shrinker_map
, true);
337 /* Not yet online memcg */
341 new = kvmalloc(sizeof(*new) + size
, GFP_KERNEL
);
345 /* Set all old bits, clear all new bits */
346 memset(new->map
, (int)0xff, old_size
);
347 memset((void *)new->map
+ old_size
, 0, size
- old_size
);
349 rcu_assign_pointer(memcg
->nodeinfo
[nid
]->shrinker_map
, new);
350 call_rcu(&old
->rcu
, memcg_free_shrinker_map_rcu
);
356 static void memcg_free_shrinker_maps(struct mem_cgroup
*memcg
)
358 struct mem_cgroup_per_node
*pn
;
359 struct memcg_shrinker_map
*map
;
362 if (mem_cgroup_is_root(memcg
))
366 pn
= mem_cgroup_nodeinfo(memcg
, nid
);
367 map
= rcu_dereference_protected(pn
->shrinker_map
, true);
370 rcu_assign_pointer(pn
->shrinker_map
, NULL
);
374 static int memcg_alloc_shrinker_maps(struct mem_cgroup
*memcg
)
376 struct memcg_shrinker_map
*map
;
377 int nid
, size
, ret
= 0;
379 if (mem_cgroup_is_root(memcg
))
382 mutex_lock(&memcg_shrinker_map_mutex
);
383 size
= memcg_shrinker_map_size
;
385 map
= kvzalloc(sizeof(*map
) + size
, GFP_KERNEL
);
387 memcg_free_shrinker_maps(memcg
);
391 rcu_assign_pointer(memcg
->nodeinfo
[nid
]->shrinker_map
, map
);
393 mutex_unlock(&memcg_shrinker_map_mutex
);
398 int memcg_expand_shrinker_maps(int new_id
)
400 int size
, old_size
, ret
= 0;
401 struct mem_cgroup
*memcg
;
403 size
= DIV_ROUND_UP(new_id
+ 1, BITS_PER_LONG
) * sizeof(unsigned long);
404 old_size
= memcg_shrinker_map_size
;
405 if (size
<= old_size
)
408 mutex_lock(&memcg_shrinker_map_mutex
);
409 if (!root_mem_cgroup
)
412 for_each_mem_cgroup(memcg
) {
413 if (mem_cgroup_is_root(memcg
))
415 ret
= memcg_expand_one_shrinker_map(memcg
, size
, old_size
);
421 memcg_shrinker_map_size
= size
;
422 mutex_unlock(&memcg_shrinker_map_mutex
);
426 void memcg_set_shrinker_bit(struct mem_cgroup
*memcg
, int nid
, int shrinker_id
)
428 if (shrinker_id
>= 0 && memcg
&& !mem_cgroup_is_root(memcg
)) {
429 struct memcg_shrinker_map
*map
;
432 map
= rcu_dereference(memcg
->nodeinfo
[nid
]->shrinker_map
);
433 /* Pairs with smp mb in shrink_slab() */
434 smp_mb__before_atomic();
435 set_bit(shrinker_id
, map
->map
);
440 #else /* CONFIG_MEMCG_KMEM */
441 static int memcg_alloc_shrinker_maps(struct mem_cgroup
*memcg
)
445 static void memcg_free_shrinker_maps(struct mem_cgroup
*memcg
) { }
446 #endif /* CONFIG_MEMCG_KMEM */
449 * mem_cgroup_css_from_page - css of the memcg associated with a page
450 * @page: page of interest
452 * If memcg is bound to the default hierarchy, css of the memcg associated
453 * with @page is returned. The returned css remains associated with @page
454 * until it is released.
456 * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup
459 struct cgroup_subsys_state
*mem_cgroup_css_from_page(struct page
*page
)
461 struct mem_cgroup
*memcg
;
463 memcg
= page
->mem_cgroup
;
465 if (!memcg
|| !cgroup_subsys_on_dfl(memory_cgrp_subsys
))
466 memcg
= root_mem_cgroup
;
472 * page_cgroup_ino - return inode number of the memcg a page is charged to
475 * Look up the closest online ancestor of the memory cgroup @page is charged to
476 * and return its inode number or 0 if @page is not charged to any cgroup. It
477 * is safe to call this function without holding a reference to @page.
479 * Note, this function is inherently racy, because there is nothing to prevent
480 * the cgroup inode from getting torn down and potentially reallocated a moment
481 * after page_cgroup_ino() returns, so it only should be used by callers that
482 * do not care (such as procfs interfaces).
484 ino_t
page_cgroup_ino(struct page
*page
)
486 struct mem_cgroup
*memcg
;
487 unsigned long ino
= 0;
490 memcg
= READ_ONCE(page
->mem_cgroup
);
491 while (memcg
&& !(memcg
->css
.flags
& CSS_ONLINE
))
492 memcg
= parent_mem_cgroup(memcg
);
494 ino
= cgroup_ino(memcg
->css
.cgroup
);
499 static struct mem_cgroup_per_node
*
500 mem_cgroup_page_nodeinfo(struct mem_cgroup
*memcg
, struct page
*page
)
502 int nid
= page_to_nid(page
);
504 return memcg
->nodeinfo
[nid
];
507 static struct mem_cgroup_tree_per_node
*
508 soft_limit_tree_node(int nid
)
510 return soft_limit_tree
.rb_tree_per_node
[nid
];
513 static struct mem_cgroup_tree_per_node
*
514 soft_limit_tree_from_page(struct page
*page
)
516 int nid
= page_to_nid(page
);
518 return soft_limit_tree
.rb_tree_per_node
[nid
];
521 static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_node
*mz
,
522 struct mem_cgroup_tree_per_node
*mctz
,
523 unsigned long new_usage_in_excess
)
525 struct rb_node
**p
= &mctz
->rb_root
.rb_node
;
526 struct rb_node
*parent
= NULL
;
527 struct mem_cgroup_per_node
*mz_node
;
528 bool rightmost
= true;
533 mz
->usage_in_excess
= new_usage_in_excess
;
534 if (!mz
->usage_in_excess
)
538 mz_node
= rb_entry(parent
, struct mem_cgroup_per_node
,
540 if (mz
->usage_in_excess
< mz_node
->usage_in_excess
) {
546 * We can't avoid mem cgroups that are over their soft
547 * limit by the same amount
549 else if (mz
->usage_in_excess
>= mz_node
->usage_in_excess
)
554 mctz
->rb_rightmost
= &mz
->tree_node
;
556 rb_link_node(&mz
->tree_node
, parent
, p
);
557 rb_insert_color(&mz
->tree_node
, &mctz
->rb_root
);
561 static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_node
*mz
,
562 struct mem_cgroup_tree_per_node
*mctz
)
567 if (&mz
->tree_node
== mctz
->rb_rightmost
)
568 mctz
->rb_rightmost
= rb_prev(&mz
->tree_node
);
570 rb_erase(&mz
->tree_node
, &mctz
->rb_root
);
574 static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_node
*mz
,
575 struct mem_cgroup_tree_per_node
*mctz
)
579 spin_lock_irqsave(&mctz
->lock
, flags
);
580 __mem_cgroup_remove_exceeded(mz
, mctz
);
581 spin_unlock_irqrestore(&mctz
->lock
, flags
);
584 static unsigned long soft_limit_excess(struct mem_cgroup
*memcg
)
586 unsigned long nr_pages
= page_counter_read(&memcg
->memory
);
587 unsigned long soft_limit
= READ_ONCE(memcg
->soft_limit
);
588 unsigned long excess
= 0;
590 if (nr_pages
> soft_limit
)
591 excess
= nr_pages
- soft_limit
;
596 static void mem_cgroup_update_tree(struct mem_cgroup
*memcg
, struct page
*page
)
598 unsigned long excess
;
599 struct mem_cgroup_per_node
*mz
;
600 struct mem_cgroup_tree_per_node
*mctz
;
602 mctz
= soft_limit_tree_from_page(page
);
606 * Necessary to update all ancestors when hierarchy is used.
607 * because their event counter is not touched.
609 for (; memcg
; memcg
= parent_mem_cgroup(memcg
)) {
610 mz
= mem_cgroup_page_nodeinfo(memcg
, page
);
611 excess
= soft_limit_excess(memcg
);
613 * We have to update the tree if mz is on RB-tree or
614 * mem is over its softlimit.
616 if (excess
|| mz
->on_tree
) {
619 spin_lock_irqsave(&mctz
->lock
, flags
);
620 /* if on-tree, remove it */
622 __mem_cgroup_remove_exceeded(mz
, mctz
);
624 * Insert again. mz->usage_in_excess will be updated.
625 * If excess is 0, no tree ops.
627 __mem_cgroup_insert_exceeded(mz
, mctz
, excess
);
628 spin_unlock_irqrestore(&mctz
->lock
, flags
);
633 static void mem_cgroup_remove_from_trees(struct mem_cgroup
*memcg
)
635 struct mem_cgroup_tree_per_node
*mctz
;
636 struct mem_cgroup_per_node
*mz
;
640 mz
= mem_cgroup_nodeinfo(memcg
, nid
);
641 mctz
= soft_limit_tree_node(nid
);
643 mem_cgroup_remove_exceeded(mz
, mctz
);
647 static struct mem_cgroup_per_node
*
648 __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node
*mctz
)
650 struct mem_cgroup_per_node
*mz
;
654 if (!mctz
->rb_rightmost
)
655 goto done
; /* Nothing to reclaim from */
657 mz
= rb_entry(mctz
->rb_rightmost
,
658 struct mem_cgroup_per_node
, tree_node
);
660 * Remove the node now but someone else can add it back,
661 * we will to add it back at the end of reclaim to its correct
662 * position in the tree.
664 __mem_cgroup_remove_exceeded(mz
, mctz
);
665 if (!soft_limit_excess(mz
->memcg
) ||
666 !css_tryget_online(&mz
->memcg
->css
))
672 static struct mem_cgroup_per_node
*
673 mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node
*mctz
)
675 struct mem_cgroup_per_node
*mz
;
677 spin_lock_irq(&mctz
->lock
);
678 mz
= __mem_cgroup_largest_soft_limit_node(mctz
);
679 spin_unlock_irq(&mctz
->lock
);
683 static unsigned long memcg_sum_events(struct mem_cgroup
*memcg
,
686 return atomic_long_read(&memcg
->events
[event
]);
689 static void mem_cgroup_charge_statistics(struct mem_cgroup
*memcg
,
691 bool compound
, int nr_pages
)
694 * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
695 * counted as CACHE even if it's on ANON LRU.
698 __mod_memcg_state(memcg
, MEMCG_RSS
, nr_pages
);
700 __mod_memcg_state(memcg
, MEMCG_CACHE
, nr_pages
);
701 if (PageSwapBacked(page
))
702 __mod_memcg_state(memcg
, NR_SHMEM
, nr_pages
);
706 VM_BUG_ON_PAGE(!PageTransHuge(page
), page
);
707 __mod_memcg_state(memcg
, MEMCG_RSS_HUGE
, nr_pages
);
710 /* pagein of a big page is an event. So, ignore page size */
712 __count_memcg_events(memcg
, PGPGIN
, 1);
714 __count_memcg_events(memcg
, PGPGOUT
, 1);
715 nr_pages
= -nr_pages
; /* for event */
718 __this_cpu_add(memcg
->stat_cpu
->nr_page_events
, nr_pages
);
721 unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup
*memcg
,
722 int nid
, unsigned int lru_mask
)
724 struct lruvec
*lruvec
= mem_cgroup_lruvec(NODE_DATA(nid
), memcg
);
725 unsigned long nr
= 0;
728 VM_BUG_ON((unsigned)nid
>= nr_node_ids
);
731 if (!(BIT(lru
) & lru_mask
))
733 nr
+= mem_cgroup_get_lru_size(lruvec
, lru
);
738 static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup
*memcg
,
739 unsigned int lru_mask
)
741 unsigned long nr
= 0;
744 for_each_node_state(nid
, N_MEMORY
)
745 nr
+= mem_cgroup_node_nr_lru_pages(memcg
, nid
, lru_mask
);
749 static bool mem_cgroup_event_ratelimit(struct mem_cgroup
*memcg
,
750 enum mem_cgroup_events_target target
)
752 unsigned long val
, next
;
754 val
= __this_cpu_read(memcg
->stat_cpu
->nr_page_events
);
755 next
= __this_cpu_read(memcg
->stat_cpu
->targets
[target
]);
756 /* from time_after() in jiffies.h */
757 if ((long)(next
- val
) < 0) {
759 case MEM_CGROUP_TARGET_THRESH
:
760 next
= val
+ THRESHOLDS_EVENTS_TARGET
;
762 case MEM_CGROUP_TARGET_SOFTLIMIT
:
763 next
= val
+ SOFTLIMIT_EVENTS_TARGET
;
765 case MEM_CGROUP_TARGET_NUMAINFO
:
766 next
= val
+ NUMAINFO_EVENTS_TARGET
;
771 __this_cpu_write(memcg
->stat_cpu
->targets
[target
], next
);
778 * Check events in order.
781 static void memcg_check_events(struct mem_cgroup
*memcg
, struct page
*page
)
783 /* threshold event is triggered in finer grain than soft limit */
784 if (unlikely(mem_cgroup_event_ratelimit(memcg
,
785 MEM_CGROUP_TARGET_THRESH
))) {
787 bool do_numainfo __maybe_unused
;
789 do_softlimit
= mem_cgroup_event_ratelimit(memcg
,
790 MEM_CGROUP_TARGET_SOFTLIMIT
);
792 do_numainfo
= mem_cgroup_event_ratelimit(memcg
,
793 MEM_CGROUP_TARGET_NUMAINFO
);
795 mem_cgroup_threshold(memcg
);
796 if (unlikely(do_softlimit
))
797 mem_cgroup_update_tree(memcg
, page
);
799 if (unlikely(do_numainfo
))
800 atomic_inc(&memcg
->numainfo_events
);
805 struct mem_cgroup
*mem_cgroup_from_task(struct task_struct
*p
)
808 * mm_update_next_owner() may clear mm->owner to NULL
809 * if it races with swapoff, page migration, etc.
810 * So this can be called with p == NULL.
815 return mem_cgroup_from_css(task_css(p
, memory_cgrp_id
));
817 EXPORT_SYMBOL(mem_cgroup_from_task
);
820 * get_mem_cgroup_from_mm: Obtain a reference on given mm_struct's memcg.
821 * @mm: mm from which memcg should be extracted. It can be NULL.
823 * Obtain a reference on mm->memcg and returns it if successful. Otherwise
824 * root_mem_cgroup is returned. However if mem_cgroup is disabled, NULL is
827 struct mem_cgroup
*get_mem_cgroup_from_mm(struct mm_struct
*mm
)
829 struct mem_cgroup
*memcg
;
831 if (mem_cgroup_disabled())
837 * Page cache insertions can happen withou an
838 * actual mm context, e.g. during disk probing
839 * on boot, loopback IO, acct() writes etc.
842 memcg
= root_mem_cgroup
;
844 memcg
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
845 if (unlikely(!memcg
))
846 memcg
= root_mem_cgroup
;
848 } while (!css_tryget_online(&memcg
->css
));
852 EXPORT_SYMBOL(get_mem_cgroup_from_mm
);
855 * get_mem_cgroup_from_page: Obtain a reference on given page's memcg.
856 * @page: page from which memcg should be extracted.
858 * Obtain a reference on page->memcg and returns it if successful. Otherwise
859 * root_mem_cgroup is returned.
861 struct mem_cgroup
*get_mem_cgroup_from_page(struct page
*page
)
863 struct mem_cgroup
*memcg
= page
->mem_cgroup
;
865 if (mem_cgroup_disabled())
869 if (!memcg
|| !css_tryget_online(&memcg
->css
))
870 memcg
= root_mem_cgroup
;
874 EXPORT_SYMBOL(get_mem_cgroup_from_page
);
877 * If current->active_memcg is non-NULL, do not fallback to current->mm->memcg.
879 static __always_inline
struct mem_cgroup
*get_mem_cgroup_from_current(void)
881 if (unlikely(current
->active_memcg
)) {
882 struct mem_cgroup
*memcg
= root_mem_cgroup
;
885 if (css_tryget_online(¤t
->active_memcg
->css
))
886 memcg
= current
->active_memcg
;
890 return get_mem_cgroup_from_mm(current
->mm
);
894 * mem_cgroup_iter - iterate over memory cgroup hierarchy
895 * @root: hierarchy root
896 * @prev: previously returned memcg, NULL on first invocation
897 * @reclaim: cookie for shared reclaim walks, NULL for full walks
899 * Returns references to children of the hierarchy below @root, or
900 * @root itself, or %NULL after a full round-trip.
902 * Caller must pass the return value in @prev on subsequent
903 * invocations for reference counting, or use mem_cgroup_iter_break()
904 * to cancel a hierarchy walk before the round-trip is complete.
906 * Reclaimers can specify a node and a priority level in @reclaim to
907 * divide up the memcgs in the hierarchy among all concurrent
908 * reclaimers operating on the same node and priority.
910 struct mem_cgroup
*mem_cgroup_iter(struct mem_cgroup
*root
,
911 struct mem_cgroup
*prev
,
912 struct mem_cgroup_reclaim_cookie
*reclaim
)
914 struct mem_cgroup_reclaim_iter
*uninitialized_var(iter
);
915 struct cgroup_subsys_state
*css
= NULL
;
916 struct mem_cgroup
*memcg
= NULL
;
917 struct mem_cgroup
*pos
= NULL
;
919 if (mem_cgroup_disabled())
923 root
= root_mem_cgroup
;
925 if (prev
&& !reclaim
)
928 if (!root
->use_hierarchy
&& root
!= root_mem_cgroup
) {
937 struct mem_cgroup_per_node
*mz
;
939 mz
= mem_cgroup_nodeinfo(root
, reclaim
->pgdat
->node_id
);
940 iter
= &mz
->iter
[reclaim
->priority
];
942 if (prev
&& reclaim
->generation
!= iter
->generation
)
946 pos
= READ_ONCE(iter
->position
);
947 if (!pos
|| css_tryget(&pos
->css
))
950 * css reference reached zero, so iter->position will
951 * be cleared by ->css_released. However, we should not
952 * rely on this happening soon, because ->css_released
953 * is called from a work queue, and by busy-waiting we
954 * might block it. So we clear iter->position right
957 (void)cmpxchg(&iter
->position
, pos
, NULL
);
965 css
= css_next_descendant_pre(css
, &root
->css
);
968 * Reclaimers share the hierarchy walk, and a
969 * new one might jump in right at the end of
970 * the hierarchy - make sure they see at least
971 * one group and restart from the beginning.
979 * Verify the css and acquire a reference. The root
980 * is provided by the caller, so we know it's alive
981 * and kicking, and don't take an extra reference.
983 memcg
= mem_cgroup_from_css(css
);
985 if (css
== &root
->css
)
996 * The position could have already been updated by a competing
997 * thread, so check that the value hasn't changed since we read
998 * it to avoid reclaiming from the same cgroup twice.
1000 (void)cmpxchg(&iter
->position
, pos
, memcg
);
1008 reclaim
->generation
= iter
->generation
;
1014 if (prev
&& prev
!= root
)
1015 css_put(&prev
->css
);
1021 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
1022 * @root: hierarchy root
1023 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
1025 void mem_cgroup_iter_break(struct mem_cgroup
*root
,
1026 struct mem_cgroup
*prev
)
1029 root
= root_mem_cgroup
;
1030 if (prev
&& prev
!= root
)
1031 css_put(&prev
->css
);
1034 static void invalidate_reclaim_iterators(struct mem_cgroup
*dead_memcg
)
1036 struct mem_cgroup
*memcg
= dead_memcg
;
1037 struct mem_cgroup_reclaim_iter
*iter
;
1038 struct mem_cgroup_per_node
*mz
;
1042 for (; memcg
; memcg
= parent_mem_cgroup(memcg
)) {
1043 for_each_node(nid
) {
1044 mz
= mem_cgroup_nodeinfo(memcg
, nid
);
1045 for (i
= 0; i
<= DEF_PRIORITY
; i
++) {
1046 iter
= &mz
->iter
[i
];
1047 cmpxchg(&iter
->position
,
1055 * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy
1056 * @memcg: hierarchy root
1057 * @fn: function to call for each task
1058 * @arg: argument passed to @fn
1060 * This function iterates over tasks attached to @memcg or to any of its
1061 * descendants and calls @fn for each task. If @fn returns a non-zero
1062 * value, the function breaks the iteration loop and returns the value.
1063 * Otherwise, it will iterate over all tasks and return 0.
1065 * This function must not be called for the root memory cgroup.
1067 int mem_cgroup_scan_tasks(struct mem_cgroup
*memcg
,
1068 int (*fn
)(struct task_struct
*, void *), void *arg
)
1070 struct mem_cgroup
*iter
;
1073 BUG_ON(memcg
== root_mem_cgroup
);
1075 for_each_mem_cgroup_tree(iter
, memcg
) {
1076 struct css_task_iter it
;
1077 struct task_struct
*task
;
1079 css_task_iter_start(&iter
->css
, 0, &it
);
1080 while (!ret
&& (task
= css_task_iter_next(&it
)))
1081 ret
= fn(task
, arg
);
1082 css_task_iter_end(&it
);
1084 mem_cgroup_iter_break(memcg
, iter
);
1092 * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page
1094 * @pgdat: pgdat of the page
1096 * This function is only safe when following the LRU page isolation
1097 * and putback protocol: the LRU lock must be held, and the page must
1098 * either be PageLRU() or the caller must have isolated/allocated it.
1100 struct lruvec
*mem_cgroup_page_lruvec(struct page
*page
, struct pglist_data
*pgdat
)
1102 struct mem_cgroup_per_node
*mz
;
1103 struct mem_cgroup
*memcg
;
1104 struct lruvec
*lruvec
;
1106 if (mem_cgroup_disabled()) {
1107 lruvec
= &pgdat
->lruvec
;
1111 memcg
= page
->mem_cgroup
;
1113 * Swapcache readahead pages are added to the LRU - and
1114 * possibly migrated - before they are charged.
1117 memcg
= root_mem_cgroup
;
1119 mz
= mem_cgroup_page_nodeinfo(memcg
, page
);
1120 lruvec
= &mz
->lruvec
;
1123 * Since a node can be onlined after the mem_cgroup was created,
1124 * we have to be prepared to initialize lruvec->zone here;
1125 * and if offlined then reonlined, we need to reinitialize it.
1127 if (unlikely(lruvec
->pgdat
!= pgdat
))
1128 lruvec
->pgdat
= pgdat
;
1133 * mem_cgroup_update_lru_size - account for adding or removing an lru page
1134 * @lruvec: mem_cgroup per zone lru vector
1135 * @lru: index of lru list the page is sitting on
1136 * @zid: zone id of the accounted pages
1137 * @nr_pages: positive when adding or negative when removing
1139 * This function must be called under lru_lock, just before a page is added
1140 * to or just after a page is removed from an lru list (that ordering being
1141 * so as to allow it to check that lru_size 0 is consistent with list_empty).
1143 void mem_cgroup_update_lru_size(struct lruvec
*lruvec
, enum lru_list lru
,
1144 int zid
, int nr_pages
)
1146 struct mem_cgroup_per_node
*mz
;
1147 unsigned long *lru_size
;
1150 if (mem_cgroup_disabled())
1153 mz
= container_of(lruvec
, struct mem_cgroup_per_node
, lruvec
);
1154 lru_size
= &mz
->lru_zone_size
[zid
][lru
];
1157 *lru_size
+= nr_pages
;
1160 if (WARN_ONCE(size
< 0,
1161 "%s(%p, %d, %d): lru_size %ld\n",
1162 __func__
, lruvec
, lru
, nr_pages
, size
)) {
1168 *lru_size
+= nr_pages
;
1171 bool task_in_mem_cgroup(struct task_struct
*task
, struct mem_cgroup
*memcg
)
1173 struct mem_cgroup
*task_memcg
;
1174 struct task_struct
*p
;
1177 p
= find_lock_task_mm(task
);
1179 task_memcg
= get_mem_cgroup_from_mm(p
->mm
);
1183 * All threads may have already detached their mm's, but the oom
1184 * killer still needs to detect if they have already been oom
1185 * killed to prevent needlessly killing additional tasks.
1188 task_memcg
= mem_cgroup_from_task(task
);
1189 css_get(&task_memcg
->css
);
1192 ret
= mem_cgroup_is_descendant(task_memcg
, memcg
);
1193 css_put(&task_memcg
->css
);
1198 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
1199 * @memcg: the memory cgroup
1201 * Returns the maximum amount of memory @mem can be charged with, in
1204 static unsigned long mem_cgroup_margin(struct mem_cgroup
*memcg
)
1206 unsigned long margin
= 0;
1207 unsigned long count
;
1208 unsigned long limit
;
1210 count
= page_counter_read(&memcg
->memory
);
1211 limit
= READ_ONCE(memcg
->memory
.max
);
1213 margin
= limit
- count
;
1215 if (do_memsw_account()) {
1216 count
= page_counter_read(&memcg
->memsw
);
1217 limit
= READ_ONCE(memcg
->memsw
.max
);
1219 margin
= min(margin
, limit
- count
);
1228 * A routine for checking "mem" is under move_account() or not.
1230 * Checking a cgroup is mc.from or mc.to or under hierarchy of
1231 * moving cgroups. This is for waiting at high-memory pressure
1234 static bool mem_cgroup_under_move(struct mem_cgroup
*memcg
)
1236 struct mem_cgroup
*from
;
1237 struct mem_cgroup
*to
;
1240 * Unlike task_move routines, we access mc.to, mc.from not under
1241 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1243 spin_lock(&mc
.lock
);
1249 ret
= mem_cgroup_is_descendant(from
, memcg
) ||
1250 mem_cgroup_is_descendant(to
, memcg
);
1252 spin_unlock(&mc
.lock
);
1256 static bool mem_cgroup_wait_acct_move(struct mem_cgroup
*memcg
)
1258 if (mc
.moving_task
&& current
!= mc
.moving_task
) {
1259 if (mem_cgroup_under_move(memcg
)) {
1261 prepare_to_wait(&mc
.waitq
, &wait
, TASK_INTERRUPTIBLE
);
1262 /* moving charge context might have finished. */
1265 finish_wait(&mc
.waitq
, &wait
);
1272 static const unsigned int memcg1_stats
[] = {
1283 static const char *const memcg1_stat_names
[] = {
1294 #define K(x) ((x) << (PAGE_SHIFT-10))
1296 * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller.
1297 * @memcg: The memory cgroup that went over limit
1298 * @p: Task that is going to be killed
1300 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1303 void mem_cgroup_print_oom_info(struct mem_cgroup
*memcg
, struct task_struct
*p
)
1305 struct mem_cgroup
*iter
;
1311 pr_info("Task in ");
1312 pr_cont_cgroup_path(task_cgroup(p
, memory_cgrp_id
));
1313 pr_cont(" killed as a result of limit of ");
1315 pr_info("Memory limit reached of cgroup ");
1318 pr_cont_cgroup_path(memcg
->css
.cgroup
);
1323 pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n",
1324 K((u64
)page_counter_read(&memcg
->memory
)),
1325 K((u64
)memcg
->memory
.max
), memcg
->memory
.failcnt
);
1326 pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n",
1327 K((u64
)page_counter_read(&memcg
->memsw
)),
1328 K((u64
)memcg
->memsw
.max
), memcg
->memsw
.failcnt
);
1329 pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n",
1330 K((u64
)page_counter_read(&memcg
->kmem
)),
1331 K((u64
)memcg
->kmem
.max
), memcg
->kmem
.failcnt
);
1333 for_each_mem_cgroup_tree(iter
, memcg
) {
1334 pr_info("Memory cgroup stats for ");
1335 pr_cont_cgroup_path(iter
->css
.cgroup
);
1338 for (i
= 0; i
< ARRAY_SIZE(memcg1_stats
); i
++) {
1339 if (memcg1_stats
[i
] == MEMCG_SWAP
&& !do_swap_account
)
1341 pr_cont(" %s:%luKB", memcg1_stat_names
[i
],
1342 K(memcg_page_state(iter
, memcg1_stats
[i
])));
1345 for (i
= 0; i
< NR_LRU_LISTS
; i
++)
1346 pr_cont(" %s:%luKB", mem_cgroup_lru_names
[i
],
1347 K(mem_cgroup_nr_lru_pages(iter
, BIT(i
))));
1354 * Return the memory (and swap, if configured) limit for a memcg.
1356 unsigned long mem_cgroup_get_max(struct mem_cgroup
*memcg
)
1360 max
= memcg
->memory
.max
;
1361 if (mem_cgroup_swappiness(memcg
)) {
1362 unsigned long memsw_max
;
1363 unsigned long swap_max
;
1365 memsw_max
= memcg
->memsw
.max
;
1366 swap_max
= memcg
->swap
.max
;
1367 swap_max
= min(swap_max
, (unsigned long)total_swap_pages
);
1368 max
= min(max
+ swap_max
, memsw_max
);
1373 static bool mem_cgroup_out_of_memory(struct mem_cgroup
*memcg
, gfp_t gfp_mask
,
1376 struct oom_control oc
= {
1380 .gfp_mask
= gfp_mask
,
1385 mutex_lock(&oom_lock
);
1386 ret
= out_of_memory(&oc
);
1387 mutex_unlock(&oom_lock
);
1391 #if MAX_NUMNODES > 1
1394 * test_mem_cgroup_node_reclaimable
1395 * @memcg: the target memcg
1396 * @nid: the node ID to be checked.
1397 * @noswap : specify true here if the user wants flle only information.
1399 * This function returns whether the specified memcg contains any
1400 * reclaimable pages on a node. Returns true if there are any reclaimable
1401 * pages in the node.
1403 static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup
*memcg
,
1404 int nid
, bool noswap
)
1406 if (mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL_FILE
))
1408 if (noswap
|| !total_swap_pages
)
1410 if (mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL_ANON
))
1417 * Always updating the nodemask is not very good - even if we have an empty
1418 * list or the wrong list here, we can start from some node and traverse all
1419 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1422 static void mem_cgroup_may_update_nodemask(struct mem_cgroup
*memcg
)
1426 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1427 * pagein/pageout changes since the last update.
1429 if (!atomic_read(&memcg
->numainfo_events
))
1431 if (atomic_inc_return(&memcg
->numainfo_updating
) > 1)
1434 /* make a nodemask where this memcg uses memory from */
1435 memcg
->scan_nodes
= node_states
[N_MEMORY
];
1437 for_each_node_mask(nid
, node_states
[N_MEMORY
]) {
1439 if (!test_mem_cgroup_node_reclaimable(memcg
, nid
, false))
1440 node_clear(nid
, memcg
->scan_nodes
);
1443 atomic_set(&memcg
->numainfo_events
, 0);
1444 atomic_set(&memcg
->numainfo_updating
, 0);
1448 * Selecting a node where we start reclaim from. Because what we need is just
1449 * reducing usage counter, start from anywhere is O,K. Considering
1450 * memory reclaim from current node, there are pros. and cons.
1452 * Freeing memory from current node means freeing memory from a node which
1453 * we'll use or we've used. So, it may make LRU bad. And if several threads
1454 * hit limits, it will see a contention on a node. But freeing from remote
1455 * node means more costs for memory reclaim because of memory latency.
1457 * Now, we use round-robin. Better algorithm is welcomed.
1459 int mem_cgroup_select_victim_node(struct mem_cgroup
*memcg
)
1463 mem_cgroup_may_update_nodemask(memcg
);
1464 node
= memcg
->last_scanned_node
;
1466 node
= next_node_in(node
, memcg
->scan_nodes
);
1468 * mem_cgroup_may_update_nodemask might have seen no reclaimmable pages
1469 * last time it really checked all the LRUs due to rate limiting.
1470 * Fallback to the current node in that case for simplicity.
1472 if (unlikely(node
== MAX_NUMNODES
))
1473 node
= numa_node_id();
1475 memcg
->last_scanned_node
= node
;
1479 int mem_cgroup_select_victim_node(struct mem_cgroup
*memcg
)
1485 static int mem_cgroup_soft_reclaim(struct mem_cgroup
*root_memcg
,
1488 unsigned long *total_scanned
)
1490 struct mem_cgroup
*victim
= NULL
;
1493 unsigned long excess
;
1494 unsigned long nr_scanned
;
1495 struct mem_cgroup_reclaim_cookie reclaim
= {
1500 excess
= soft_limit_excess(root_memcg
);
1503 victim
= mem_cgroup_iter(root_memcg
, victim
, &reclaim
);
1508 * If we have not been able to reclaim
1509 * anything, it might because there are
1510 * no reclaimable pages under this hierarchy
1515 * We want to do more targeted reclaim.
1516 * excess >> 2 is not to excessive so as to
1517 * reclaim too much, nor too less that we keep
1518 * coming back to reclaim from this cgroup
1520 if (total
>= (excess
>> 2) ||
1521 (loop
> MEM_CGROUP_MAX_RECLAIM_LOOPS
))
1526 total
+= mem_cgroup_shrink_node(victim
, gfp_mask
, false,
1527 pgdat
, &nr_scanned
);
1528 *total_scanned
+= nr_scanned
;
1529 if (!soft_limit_excess(root_memcg
))
1532 mem_cgroup_iter_break(root_memcg
, victim
);
1536 #ifdef CONFIG_LOCKDEP
1537 static struct lockdep_map memcg_oom_lock_dep_map
= {
1538 .name
= "memcg_oom_lock",
1542 static DEFINE_SPINLOCK(memcg_oom_lock
);
1545 * Check OOM-Killer is already running under our hierarchy.
1546 * If someone is running, return false.
1548 static bool mem_cgroup_oom_trylock(struct mem_cgroup
*memcg
)
1550 struct mem_cgroup
*iter
, *failed
= NULL
;
1552 spin_lock(&memcg_oom_lock
);
1554 for_each_mem_cgroup_tree(iter
, memcg
) {
1555 if (iter
->oom_lock
) {
1557 * this subtree of our hierarchy is already locked
1558 * so we cannot give a lock.
1561 mem_cgroup_iter_break(memcg
, iter
);
1564 iter
->oom_lock
= true;
1569 * OK, we failed to lock the whole subtree so we have
1570 * to clean up what we set up to the failing subtree
1572 for_each_mem_cgroup_tree(iter
, memcg
) {
1573 if (iter
== failed
) {
1574 mem_cgroup_iter_break(memcg
, iter
);
1577 iter
->oom_lock
= false;
1580 mutex_acquire(&memcg_oom_lock_dep_map
, 0, 1, _RET_IP_
);
1582 spin_unlock(&memcg_oom_lock
);
1587 static void mem_cgroup_oom_unlock(struct mem_cgroup
*memcg
)
1589 struct mem_cgroup
*iter
;
1591 spin_lock(&memcg_oom_lock
);
1592 mutex_release(&memcg_oom_lock_dep_map
, 1, _RET_IP_
);
1593 for_each_mem_cgroup_tree(iter
, memcg
)
1594 iter
->oom_lock
= false;
1595 spin_unlock(&memcg_oom_lock
);
1598 static void mem_cgroup_mark_under_oom(struct mem_cgroup
*memcg
)
1600 struct mem_cgroup
*iter
;
1602 spin_lock(&memcg_oom_lock
);
1603 for_each_mem_cgroup_tree(iter
, memcg
)
1605 spin_unlock(&memcg_oom_lock
);
1608 static void mem_cgroup_unmark_under_oom(struct mem_cgroup
*memcg
)
1610 struct mem_cgroup
*iter
;
1613 * When a new child is created while the hierarchy is under oom,
1614 * mem_cgroup_oom_lock() may not be called. Watch for underflow.
1616 spin_lock(&memcg_oom_lock
);
1617 for_each_mem_cgroup_tree(iter
, memcg
)
1618 if (iter
->under_oom
> 0)
1620 spin_unlock(&memcg_oom_lock
);
1623 static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq
);
1625 struct oom_wait_info
{
1626 struct mem_cgroup
*memcg
;
1627 wait_queue_entry_t wait
;
1630 static int memcg_oom_wake_function(wait_queue_entry_t
*wait
,
1631 unsigned mode
, int sync
, void *arg
)
1633 struct mem_cgroup
*wake_memcg
= (struct mem_cgroup
*)arg
;
1634 struct mem_cgroup
*oom_wait_memcg
;
1635 struct oom_wait_info
*oom_wait_info
;
1637 oom_wait_info
= container_of(wait
, struct oom_wait_info
, wait
);
1638 oom_wait_memcg
= oom_wait_info
->memcg
;
1640 if (!mem_cgroup_is_descendant(wake_memcg
, oom_wait_memcg
) &&
1641 !mem_cgroup_is_descendant(oom_wait_memcg
, wake_memcg
))
1643 return autoremove_wake_function(wait
, mode
, sync
, arg
);
1646 static void memcg_oom_recover(struct mem_cgroup
*memcg
)
1649 * For the following lockless ->under_oom test, the only required
1650 * guarantee is that it must see the state asserted by an OOM when
1651 * this function is called as a result of userland actions
1652 * triggered by the notification of the OOM. This is trivially
1653 * achieved by invoking mem_cgroup_mark_under_oom() before
1654 * triggering notification.
1656 if (memcg
&& memcg
->under_oom
)
1657 __wake_up(&memcg_oom_waitq
, TASK_NORMAL
, 0, memcg
);
1667 static enum oom_status
mem_cgroup_oom(struct mem_cgroup
*memcg
, gfp_t mask
, int order
)
1669 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
1672 memcg_memory_event(memcg
, MEMCG_OOM
);
1675 * We are in the middle of the charge context here, so we
1676 * don't want to block when potentially sitting on a callstack
1677 * that holds all kinds of filesystem and mm locks.
1679 * cgroup1 allows disabling the OOM killer and waiting for outside
1680 * handling until the charge can succeed; remember the context and put
1681 * the task to sleep at the end of the page fault when all locks are
1684 * On the other hand, in-kernel OOM killer allows for an async victim
1685 * memory reclaim (oom_reaper) and that means that we are not solely
1686 * relying on the oom victim to make a forward progress and we can
1687 * invoke the oom killer here.
1689 * Please note that mem_cgroup_out_of_memory might fail to find a
1690 * victim and then we have to bail out from the charge path.
1692 if (memcg
->oom_kill_disable
) {
1693 if (!current
->in_user_fault
)
1695 css_get(&memcg
->css
);
1696 current
->memcg_in_oom
= memcg
;
1697 current
->memcg_oom_gfp_mask
= mask
;
1698 current
->memcg_oom_order
= order
;
1703 if (mem_cgroup_out_of_memory(memcg
, mask
, order
))
1710 * mem_cgroup_oom_synchronize - complete memcg OOM handling
1711 * @handle: actually kill/wait or just clean up the OOM state
1713 * This has to be called at the end of a page fault if the memcg OOM
1714 * handler was enabled.
1716 * Memcg supports userspace OOM handling where failed allocations must
1717 * sleep on a waitqueue until the userspace task resolves the
1718 * situation. Sleeping directly in the charge context with all kinds
1719 * of locks held is not a good idea, instead we remember an OOM state
1720 * in the task and mem_cgroup_oom_synchronize() has to be called at
1721 * the end of the page fault to complete the OOM handling.
1723 * Returns %true if an ongoing memcg OOM situation was detected and
1724 * completed, %false otherwise.
1726 bool mem_cgroup_oom_synchronize(bool handle
)
1728 struct mem_cgroup
*memcg
= current
->memcg_in_oom
;
1729 struct oom_wait_info owait
;
1732 /* OOM is global, do not handle */
1739 owait
.memcg
= memcg
;
1740 owait
.wait
.flags
= 0;
1741 owait
.wait
.func
= memcg_oom_wake_function
;
1742 owait
.wait
.private = current
;
1743 INIT_LIST_HEAD(&owait
.wait
.entry
);
1745 prepare_to_wait(&memcg_oom_waitq
, &owait
.wait
, TASK_KILLABLE
);
1746 mem_cgroup_mark_under_oom(memcg
);
1748 locked
= mem_cgroup_oom_trylock(memcg
);
1751 mem_cgroup_oom_notify(memcg
);
1753 if (locked
&& !memcg
->oom_kill_disable
) {
1754 mem_cgroup_unmark_under_oom(memcg
);
1755 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1756 mem_cgroup_out_of_memory(memcg
, current
->memcg_oom_gfp_mask
,
1757 current
->memcg_oom_order
);
1760 mem_cgroup_unmark_under_oom(memcg
);
1761 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1765 mem_cgroup_oom_unlock(memcg
);
1767 * There is no guarantee that an OOM-lock contender
1768 * sees the wakeups triggered by the OOM kill
1769 * uncharges. Wake any sleepers explicitely.
1771 memcg_oom_recover(memcg
);
1774 current
->memcg_in_oom
= NULL
;
1775 css_put(&memcg
->css
);
1780 * mem_cgroup_get_oom_group - get a memory cgroup to clean up after OOM
1781 * @victim: task to be killed by the OOM killer
1782 * @oom_domain: memcg in case of memcg OOM, NULL in case of system-wide OOM
1784 * Returns a pointer to a memory cgroup, which has to be cleaned up
1785 * by killing all belonging OOM-killable tasks.
1787 * Caller has to call mem_cgroup_put() on the returned non-NULL memcg.
1789 struct mem_cgroup
*mem_cgroup_get_oom_group(struct task_struct
*victim
,
1790 struct mem_cgroup
*oom_domain
)
1792 struct mem_cgroup
*oom_group
= NULL
;
1793 struct mem_cgroup
*memcg
;
1795 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys
))
1799 oom_domain
= root_mem_cgroup
;
1803 memcg
= mem_cgroup_from_task(victim
);
1804 if (memcg
== root_mem_cgroup
)
1808 * Traverse the memory cgroup hierarchy from the victim task's
1809 * cgroup up to the OOMing cgroup (or root) to find the
1810 * highest-level memory cgroup with oom.group set.
1812 for (; memcg
; memcg
= parent_mem_cgroup(memcg
)) {
1813 if (memcg
->oom_group
)
1816 if (memcg
== oom_domain
)
1821 css_get(&oom_group
->css
);
1828 void mem_cgroup_print_oom_group(struct mem_cgroup
*memcg
)
1830 pr_info("Tasks in ");
1831 pr_cont_cgroup_path(memcg
->css
.cgroup
);
1832 pr_cont(" are going to be killed due to memory.oom.group set\n");
1836 * lock_page_memcg - lock a page->mem_cgroup binding
1839 * This function protects unlocked LRU pages from being moved to
1842 * It ensures lifetime of the returned memcg. Caller is responsible
1843 * for the lifetime of the page; __unlock_page_memcg() is available
1844 * when @page might get freed inside the locked section.
1846 struct mem_cgroup
*lock_page_memcg(struct page
*page
)
1848 struct mem_cgroup
*memcg
;
1849 unsigned long flags
;
1852 * The RCU lock is held throughout the transaction. The fast
1853 * path can get away without acquiring the memcg->move_lock
1854 * because page moving starts with an RCU grace period.
1856 * The RCU lock also protects the memcg from being freed when
1857 * the page state that is going to change is the only thing
1858 * preventing the page itself from being freed. E.g. writeback
1859 * doesn't hold a page reference and relies on PG_writeback to
1860 * keep off truncation, migration and so forth.
1864 if (mem_cgroup_disabled())
1867 memcg
= page
->mem_cgroup
;
1868 if (unlikely(!memcg
))
1871 if (atomic_read(&memcg
->moving_account
) <= 0)
1874 spin_lock_irqsave(&memcg
->move_lock
, flags
);
1875 if (memcg
!= page
->mem_cgroup
) {
1876 spin_unlock_irqrestore(&memcg
->move_lock
, flags
);
1881 * When charge migration first begins, we can have locked and
1882 * unlocked page stat updates happening concurrently. Track
1883 * the task who has the lock for unlock_page_memcg().
1885 memcg
->move_lock_task
= current
;
1886 memcg
->move_lock_flags
= flags
;
1890 EXPORT_SYMBOL(lock_page_memcg
);
1893 * __unlock_page_memcg - unlock and unpin a memcg
1896 * Unlock and unpin a memcg returned by lock_page_memcg().
1898 void __unlock_page_memcg(struct mem_cgroup
*memcg
)
1900 if (memcg
&& memcg
->move_lock_task
== current
) {
1901 unsigned long flags
= memcg
->move_lock_flags
;
1903 memcg
->move_lock_task
= NULL
;
1904 memcg
->move_lock_flags
= 0;
1906 spin_unlock_irqrestore(&memcg
->move_lock
, flags
);
1913 * unlock_page_memcg - unlock a page->mem_cgroup binding
1916 void unlock_page_memcg(struct page
*page
)
1918 __unlock_page_memcg(page
->mem_cgroup
);
1920 EXPORT_SYMBOL(unlock_page_memcg
);
1922 struct memcg_stock_pcp
{
1923 struct mem_cgroup
*cached
; /* this never be root cgroup */
1924 unsigned int nr_pages
;
1925 struct work_struct work
;
1926 unsigned long flags
;
1927 #define FLUSHING_CACHED_CHARGE 0
1929 static DEFINE_PER_CPU(struct memcg_stock_pcp
, memcg_stock
);
1930 static DEFINE_MUTEX(percpu_charge_mutex
);
1933 * consume_stock: Try to consume stocked charge on this cpu.
1934 * @memcg: memcg to consume from.
1935 * @nr_pages: how many pages to charge.
1937 * The charges will only happen if @memcg matches the current cpu's memcg
1938 * stock, and at least @nr_pages are available in that stock. Failure to
1939 * service an allocation will refill the stock.
1941 * returns true if successful, false otherwise.
1943 static bool consume_stock(struct mem_cgroup
*memcg
, unsigned int nr_pages
)
1945 struct memcg_stock_pcp
*stock
;
1946 unsigned long flags
;
1949 if (nr_pages
> MEMCG_CHARGE_BATCH
)
1952 local_irq_save(flags
);
1954 stock
= this_cpu_ptr(&memcg_stock
);
1955 if (memcg
== stock
->cached
&& stock
->nr_pages
>= nr_pages
) {
1956 stock
->nr_pages
-= nr_pages
;
1960 local_irq_restore(flags
);
1966 * Returns stocks cached in percpu and reset cached information.
1968 static void drain_stock(struct memcg_stock_pcp
*stock
)
1970 struct mem_cgroup
*old
= stock
->cached
;
1972 if (stock
->nr_pages
) {
1973 page_counter_uncharge(&old
->memory
, stock
->nr_pages
);
1974 if (do_memsw_account())
1975 page_counter_uncharge(&old
->memsw
, stock
->nr_pages
);
1976 css_put_many(&old
->css
, stock
->nr_pages
);
1977 stock
->nr_pages
= 0;
1979 stock
->cached
= NULL
;
1982 static void drain_local_stock(struct work_struct
*dummy
)
1984 struct memcg_stock_pcp
*stock
;
1985 unsigned long flags
;
1988 * The only protection from memory hotplug vs. drain_stock races is
1989 * that we always operate on local CPU stock here with IRQ disabled
1991 local_irq_save(flags
);
1993 stock
= this_cpu_ptr(&memcg_stock
);
1995 clear_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
);
1997 local_irq_restore(flags
);
2001 * Cache charges(val) to local per_cpu area.
2002 * This will be consumed by consume_stock() function, later.
2004 static void refill_stock(struct mem_cgroup
*memcg
, unsigned int nr_pages
)
2006 struct memcg_stock_pcp
*stock
;
2007 unsigned long flags
;
2009 local_irq_save(flags
);
2011 stock
= this_cpu_ptr(&memcg_stock
);
2012 if (stock
->cached
!= memcg
) { /* reset if necessary */
2014 stock
->cached
= memcg
;
2016 stock
->nr_pages
+= nr_pages
;
2018 if (stock
->nr_pages
> MEMCG_CHARGE_BATCH
)
2021 local_irq_restore(flags
);
2025 * Drains all per-CPU charge caches for given root_memcg resp. subtree
2026 * of the hierarchy under it.
2028 static void drain_all_stock(struct mem_cgroup
*root_memcg
)
2032 /* If someone's already draining, avoid adding running more workers. */
2033 if (!mutex_trylock(&percpu_charge_mutex
))
2036 * Notify other cpus that system-wide "drain" is running
2037 * We do not care about races with the cpu hotplug because cpu down
2038 * as well as workers from this path always operate on the local
2039 * per-cpu data. CPU up doesn't touch memcg_stock at all.
2042 for_each_online_cpu(cpu
) {
2043 struct memcg_stock_pcp
*stock
= &per_cpu(memcg_stock
, cpu
);
2044 struct mem_cgroup
*memcg
;
2046 memcg
= stock
->cached
;
2047 if (!memcg
|| !stock
->nr_pages
|| !css_tryget(&memcg
->css
))
2049 if (!mem_cgroup_is_descendant(memcg
, root_memcg
)) {
2050 css_put(&memcg
->css
);
2053 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
)) {
2055 drain_local_stock(&stock
->work
);
2057 schedule_work_on(cpu
, &stock
->work
);
2059 css_put(&memcg
->css
);
2062 mutex_unlock(&percpu_charge_mutex
);
2065 static int memcg_hotplug_cpu_dead(unsigned int cpu
)
2067 struct memcg_stock_pcp
*stock
;
2068 struct mem_cgroup
*memcg
;
2070 stock
= &per_cpu(memcg_stock
, cpu
);
2073 for_each_mem_cgroup(memcg
) {
2076 for (i
= 0; i
< MEMCG_NR_STAT
; i
++) {
2080 x
= this_cpu_xchg(memcg
->stat_cpu
->count
[i
], 0);
2082 atomic_long_add(x
, &memcg
->stat
[i
]);
2084 if (i
>= NR_VM_NODE_STAT_ITEMS
)
2087 for_each_node(nid
) {
2088 struct mem_cgroup_per_node
*pn
;
2090 pn
= mem_cgroup_nodeinfo(memcg
, nid
);
2091 x
= this_cpu_xchg(pn
->lruvec_stat_cpu
->count
[i
], 0);
2093 atomic_long_add(x
, &pn
->lruvec_stat
[i
]);
2097 for (i
= 0; i
< NR_VM_EVENT_ITEMS
; i
++) {
2100 x
= this_cpu_xchg(memcg
->stat_cpu
->events
[i
], 0);
2102 atomic_long_add(x
, &memcg
->events
[i
]);
2109 static void reclaim_high(struct mem_cgroup
*memcg
,
2110 unsigned int nr_pages
,
2114 if (page_counter_read(&memcg
->memory
) <= memcg
->high
)
2116 memcg_memory_event(memcg
, MEMCG_HIGH
);
2117 try_to_free_mem_cgroup_pages(memcg
, nr_pages
, gfp_mask
, true);
2118 } while ((memcg
= parent_mem_cgroup(memcg
)));
2121 static void high_work_func(struct work_struct
*work
)
2123 struct mem_cgroup
*memcg
;
2125 memcg
= container_of(work
, struct mem_cgroup
, high_work
);
2126 reclaim_high(memcg
, MEMCG_CHARGE_BATCH
, GFP_KERNEL
);
2130 * Scheduled by try_charge() to be executed from the userland return path
2131 * and reclaims memory over the high limit.
2133 void mem_cgroup_handle_over_high(void)
2135 unsigned int nr_pages
= current
->memcg_nr_pages_over_high
;
2136 struct mem_cgroup
*memcg
;
2138 if (likely(!nr_pages
))
2141 memcg
= get_mem_cgroup_from_mm(current
->mm
);
2142 reclaim_high(memcg
, nr_pages
, GFP_KERNEL
);
2143 css_put(&memcg
->css
);
2144 current
->memcg_nr_pages_over_high
= 0;
2147 static int try_charge(struct mem_cgroup
*memcg
, gfp_t gfp_mask
,
2148 unsigned int nr_pages
)
2150 unsigned int batch
= max(MEMCG_CHARGE_BATCH
, nr_pages
);
2151 int nr_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
2152 struct mem_cgroup
*mem_over_limit
;
2153 struct page_counter
*counter
;
2154 unsigned long nr_reclaimed
;
2155 bool may_swap
= true;
2156 bool drained
= false;
2158 enum oom_status oom_status
;
2160 if (mem_cgroup_is_root(memcg
))
2163 if (consume_stock(memcg
, nr_pages
))
2166 if (!do_memsw_account() ||
2167 page_counter_try_charge(&memcg
->memsw
, batch
, &counter
)) {
2168 if (page_counter_try_charge(&memcg
->memory
, batch
, &counter
))
2170 if (do_memsw_account())
2171 page_counter_uncharge(&memcg
->memsw
, batch
);
2172 mem_over_limit
= mem_cgroup_from_counter(counter
, memory
);
2174 mem_over_limit
= mem_cgroup_from_counter(counter
, memsw
);
2178 if (batch
> nr_pages
) {
2184 * Unlike in global OOM situations, memcg is not in a physical
2185 * memory shortage. Allow dying and OOM-killed tasks to
2186 * bypass the last charges so that they can exit quickly and
2187 * free their memory.
2189 if (unlikely(tsk_is_oom_victim(current
) ||
2190 fatal_signal_pending(current
) ||
2191 current
->flags
& PF_EXITING
))
2195 * Prevent unbounded recursion when reclaim operations need to
2196 * allocate memory. This might exceed the limits temporarily,
2197 * but we prefer facilitating memory reclaim and getting back
2198 * under the limit over triggering OOM kills in these cases.
2200 if (unlikely(current
->flags
& PF_MEMALLOC
))
2203 if (unlikely(task_in_memcg_oom(current
)))
2206 if (!gfpflags_allow_blocking(gfp_mask
))
2209 memcg_memory_event(mem_over_limit
, MEMCG_MAX
);
2211 nr_reclaimed
= try_to_free_mem_cgroup_pages(mem_over_limit
, nr_pages
,
2212 gfp_mask
, may_swap
);
2214 if (mem_cgroup_margin(mem_over_limit
) >= nr_pages
)
2218 drain_all_stock(mem_over_limit
);
2223 if (gfp_mask
& __GFP_NORETRY
)
2226 * Even though the limit is exceeded at this point, reclaim
2227 * may have been able to free some pages. Retry the charge
2228 * before killing the task.
2230 * Only for regular pages, though: huge pages are rather
2231 * unlikely to succeed so close to the limit, and we fall back
2232 * to regular pages anyway in case of failure.
2234 if (nr_reclaimed
&& nr_pages
<= (1 << PAGE_ALLOC_COSTLY_ORDER
))
2237 * At task move, charge accounts can be doubly counted. So, it's
2238 * better to wait until the end of task_move if something is going on.
2240 if (mem_cgroup_wait_acct_move(mem_over_limit
))
2246 if (gfp_mask
& __GFP_RETRY_MAYFAIL
&& oomed
)
2249 if (gfp_mask
& __GFP_NOFAIL
)
2252 if (fatal_signal_pending(current
))
2256 * keep retrying as long as the memcg oom killer is able to make
2257 * a forward progress or bypass the charge if the oom killer
2258 * couldn't make any progress.
2260 oom_status
= mem_cgroup_oom(mem_over_limit
, gfp_mask
,
2261 get_order(nr_pages
* PAGE_SIZE
));
2262 switch (oom_status
) {
2264 nr_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
2273 if (!(gfp_mask
& __GFP_NOFAIL
))
2277 * The allocation either can't fail or will lead to more memory
2278 * being freed very soon. Allow memory usage go over the limit
2279 * temporarily by force charging it.
2281 page_counter_charge(&memcg
->memory
, nr_pages
);
2282 if (do_memsw_account())
2283 page_counter_charge(&memcg
->memsw
, nr_pages
);
2284 css_get_many(&memcg
->css
, nr_pages
);
2289 css_get_many(&memcg
->css
, batch
);
2290 if (batch
> nr_pages
)
2291 refill_stock(memcg
, batch
- nr_pages
);
2294 * If the hierarchy is above the normal consumption range, schedule
2295 * reclaim on returning to userland. We can perform reclaim here
2296 * if __GFP_RECLAIM but let's always punt for simplicity and so that
2297 * GFP_KERNEL can consistently be used during reclaim. @memcg is
2298 * not recorded as it most likely matches current's and won't
2299 * change in the meantime. As high limit is checked again before
2300 * reclaim, the cost of mismatch is negligible.
2303 if (page_counter_read(&memcg
->memory
) > memcg
->high
) {
2304 /* Don't bother a random interrupted task */
2305 if (in_interrupt()) {
2306 schedule_work(&memcg
->high_work
);
2309 current
->memcg_nr_pages_over_high
+= batch
;
2310 set_notify_resume(current
);
2313 } while ((memcg
= parent_mem_cgroup(memcg
)));
2318 static void cancel_charge(struct mem_cgroup
*memcg
, unsigned int nr_pages
)
2320 if (mem_cgroup_is_root(memcg
))
2323 page_counter_uncharge(&memcg
->memory
, nr_pages
);
2324 if (do_memsw_account())
2325 page_counter_uncharge(&memcg
->memsw
, nr_pages
);
2327 css_put_many(&memcg
->css
, nr_pages
);
2330 static void lock_page_lru(struct page
*page
, int *isolated
)
2332 struct zone
*zone
= page_zone(page
);
2334 spin_lock_irq(zone_lru_lock(zone
));
2335 if (PageLRU(page
)) {
2336 struct lruvec
*lruvec
;
2338 lruvec
= mem_cgroup_page_lruvec(page
, zone
->zone_pgdat
);
2340 del_page_from_lru_list(page
, lruvec
, page_lru(page
));
2346 static void unlock_page_lru(struct page
*page
, int isolated
)
2348 struct zone
*zone
= page_zone(page
);
2351 struct lruvec
*lruvec
;
2353 lruvec
= mem_cgroup_page_lruvec(page
, zone
->zone_pgdat
);
2354 VM_BUG_ON_PAGE(PageLRU(page
), page
);
2356 add_page_to_lru_list(page
, lruvec
, page_lru(page
));
2358 spin_unlock_irq(zone_lru_lock(zone
));
2361 static void commit_charge(struct page
*page
, struct mem_cgroup
*memcg
,
2366 VM_BUG_ON_PAGE(page
->mem_cgroup
, page
);
2369 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2370 * may already be on some other mem_cgroup's LRU. Take care of it.
2373 lock_page_lru(page
, &isolated
);
2376 * Nobody should be changing or seriously looking at
2377 * page->mem_cgroup at this point:
2379 * - the page is uncharged
2381 * - the page is off-LRU
2383 * - an anonymous fault has exclusive page access, except for
2384 * a locked page table
2386 * - a page cache insertion, a swapin fault, or a migration
2387 * have the page locked
2389 page
->mem_cgroup
= memcg
;
2392 unlock_page_lru(page
, isolated
);
2395 #ifdef CONFIG_MEMCG_KMEM
2396 static int memcg_alloc_cache_id(void)
2401 id
= ida_simple_get(&memcg_cache_ida
,
2402 0, MEMCG_CACHES_MAX_SIZE
, GFP_KERNEL
);
2406 if (id
< memcg_nr_cache_ids
)
2410 * There's no space for the new id in memcg_caches arrays,
2411 * so we have to grow them.
2413 down_write(&memcg_cache_ids_sem
);
2415 size
= 2 * (id
+ 1);
2416 if (size
< MEMCG_CACHES_MIN_SIZE
)
2417 size
= MEMCG_CACHES_MIN_SIZE
;
2418 else if (size
> MEMCG_CACHES_MAX_SIZE
)
2419 size
= MEMCG_CACHES_MAX_SIZE
;
2421 err
= memcg_update_all_caches(size
);
2423 err
= memcg_update_all_list_lrus(size
);
2425 memcg_nr_cache_ids
= size
;
2427 up_write(&memcg_cache_ids_sem
);
2430 ida_simple_remove(&memcg_cache_ida
, id
);
2436 static void memcg_free_cache_id(int id
)
2438 ida_simple_remove(&memcg_cache_ida
, id
);
2441 struct memcg_kmem_cache_create_work
{
2442 struct mem_cgroup
*memcg
;
2443 struct kmem_cache
*cachep
;
2444 struct work_struct work
;
2447 static void memcg_kmem_cache_create_func(struct work_struct
*w
)
2449 struct memcg_kmem_cache_create_work
*cw
=
2450 container_of(w
, struct memcg_kmem_cache_create_work
, work
);
2451 struct mem_cgroup
*memcg
= cw
->memcg
;
2452 struct kmem_cache
*cachep
= cw
->cachep
;
2454 memcg_create_kmem_cache(memcg
, cachep
);
2456 css_put(&memcg
->css
);
2461 * Enqueue the creation of a per-memcg kmem_cache.
2463 static void memcg_schedule_kmem_cache_create(struct mem_cgroup
*memcg
,
2464 struct kmem_cache
*cachep
)
2466 struct memcg_kmem_cache_create_work
*cw
;
2468 cw
= kmalloc(sizeof(*cw
), GFP_NOWAIT
| __GFP_NOWARN
);
2472 css_get(&memcg
->css
);
2475 cw
->cachep
= cachep
;
2476 INIT_WORK(&cw
->work
, memcg_kmem_cache_create_func
);
2478 queue_work(memcg_kmem_cache_wq
, &cw
->work
);
2481 static inline bool memcg_kmem_bypass(void)
2483 if (in_interrupt() || !current
->mm
|| (current
->flags
& PF_KTHREAD
))
2489 * memcg_kmem_get_cache: select the correct per-memcg cache for allocation
2490 * @cachep: the original global kmem cache
2492 * Return the kmem_cache we're supposed to use for a slab allocation.
2493 * We try to use the current memcg's version of the cache.
2495 * If the cache does not exist yet, if we are the first user of it, we
2496 * create it asynchronously in a workqueue and let the current allocation
2497 * go through with the original cache.
2499 * This function takes a reference to the cache it returns to assure it
2500 * won't get destroyed while we are working with it. Once the caller is
2501 * done with it, memcg_kmem_put_cache() must be called to release the
2504 struct kmem_cache
*memcg_kmem_get_cache(struct kmem_cache
*cachep
)
2506 struct mem_cgroup
*memcg
;
2507 struct kmem_cache
*memcg_cachep
;
2510 VM_BUG_ON(!is_root_cache(cachep
));
2512 if (memcg_kmem_bypass())
2515 memcg
= get_mem_cgroup_from_current();
2516 kmemcg_id
= READ_ONCE(memcg
->kmemcg_id
);
2520 memcg_cachep
= cache_from_memcg_idx(cachep
, kmemcg_id
);
2521 if (likely(memcg_cachep
))
2522 return memcg_cachep
;
2525 * If we are in a safe context (can wait, and not in interrupt
2526 * context), we could be be predictable and return right away.
2527 * This would guarantee that the allocation being performed
2528 * already belongs in the new cache.
2530 * However, there are some clashes that can arrive from locking.
2531 * For instance, because we acquire the slab_mutex while doing
2532 * memcg_create_kmem_cache, this means no further allocation
2533 * could happen with the slab_mutex held. So it's better to
2536 memcg_schedule_kmem_cache_create(memcg
, cachep
);
2538 css_put(&memcg
->css
);
2543 * memcg_kmem_put_cache: drop reference taken by memcg_kmem_get_cache
2544 * @cachep: the cache returned by memcg_kmem_get_cache
2546 void memcg_kmem_put_cache(struct kmem_cache
*cachep
)
2548 if (!is_root_cache(cachep
))
2549 css_put(&cachep
->memcg_params
.memcg
->css
);
2553 * memcg_kmem_charge_memcg: charge a kmem page
2554 * @page: page to charge
2555 * @gfp: reclaim mode
2556 * @order: allocation order
2557 * @memcg: memory cgroup to charge
2559 * Returns 0 on success, an error code on failure.
2561 int memcg_kmem_charge_memcg(struct page
*page
, gfp_t gfp
, int order
,
2562 struct mem_cgroup
*memcg
)
2564 unsigned int nr_pages
= 1 << order
;
2565 struct page_counter
*counter
;
2568 ret
= try_charge(memcg
, gfp
, nr_pages
);
2572 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys
) &&
2573 !page_counter_try_charge(&memcg
->kmem
, nr_pages
, &counter
)) {
2574 cancel_charge(memcg
, nr_pages
);
2578 page
->mem_cgroup
= memcg
;
2584 * memcg_kmem_charge: charge a kmem page to the current memory cgroup
2585 * @page: page to charge
2586 * @gfp: reclaim mode
2587 * @order: allocation order
2589 * Returns 0 on success, an error code on failure.
2591 int memcg_kmem_charge(struct page
*page
, gfp_t gfp
, int order
)
2593 struct mem_cgroup
*memcg
;
2596 if (memcg_kmem_bypass())
2599 memcg
= get_mem_cgroup_from_current();
2600 if (!mem_cgroup_is_root(memcg
)) {
2601 ret
= memcg_kmem_charge_memcg(page
, gfp
, order
, memcg
);
2603 __SetPageKmemcg(page
);
2605 css_put(&memcg
->css
);
2609 * memcg_kmem_uncharge: uncharge a kmem page
2610 * @page: page to uncharge
2611 * @order: allocation order
2613 void memcg_kmem_uncharge(struct page
*page
, int order
)
2615 struct mem_cgroup
*memcg
= page
->mem_cgroup
;
2616 unsigned int nr_pages
= 1 << order
;
2621 VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg
), page
);
2623 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys
))
2624 page_counter_uncharge(&memcg
->kmem
, nr_pages
);
2626 page_counter_uncharge(&memcg
->memory
, nr_pages
);
2627 if (do_memsw_account())
2628 page_counter_uncharge(&memcg
->memsw
, nr_pages
);
2630 page
->mem_cgroup
= NULL
;
2632 /* slab pages do not have PageKmemcg flag set */
2633 if (PageKmemcg(page
))
2634 __ClearPageKmemcg(page
);
2636 css_put_many(&memcg
->css
, nr_pages
);
2638 #endif /* CONFIG_MEMCG_KMEM */
2640 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2643 * Because tail pages are not marked as "used", set it. We're under
2644 * zone_lru_lock and migration entries setup in all page mappings.
2646 void mem_cgroup_split_huge_fixup(struct page
*head
)
2650 if (mem_cgroup_disabled())
2653 for (i
= 1; i
< HPAGE_PMD_NR
; i
++)
2654 head
[i
].mem_cgroup
= head
->mem_cgroup
;
2656 __mod_memcg_state(head
->mem_cgroup
, MEMCG_RSS_HUGE
, -HPAGE_PMD_NR
);
2658 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
2660 #ifdef CONFIG_MEMCG_SWAP
2662 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
2663 * @entry: swap entry to be moved
2664 * @from: mem_cgroup which the entry is moved from
2665 * @to: mem_cgroup which the entry is moved to
2667 * It succeeds only when the swap_cgroup's record for this entry is the same
2668 * as the mem_cgroup's id of @from.
2670 * Returns 0 on success, -EINVAL on failure.
2672 * The caller must have charged to @to, IOW, called page_counter_charge() about
2673 * both res and memsw, and called css_get().
2675 static int mem_cgroup_move_swap_account(swp_entry_t entry
,
2676 struct mem_cgroup
*from
, struct mem_cgroup
*to
)
2678 unsigned short old_id
, new_id
;
2680 old_id
= mem_cgroup_id(from
);
2681 new_id
= mem_cgroup_id(to
);
2683 if (swap_cgroup_cmpxchg(entry
, old_id
, new_id
) == old_id
) {
2684 mod_memcg_state(from
, MEMCG_SWAP
, -1);
2685 mod_memcg_state(to
, MEMCG_SWAP
, 1);
2691 static inline int mem_cgroup_move_swap_account(swp_entry_t entry
,
2692 struct mem_cgroup
*from
, struct mem_cgroup
*to
)
2698 static DEFINE_MUTEX(memcg_max_mutex
);
2700 static int mem_cgroup_resize_max(struct mem_cgroup
*memcg
,
2701 unsigned long max
, bool memsw
)
2703 bool enlarge
= false;
2704 bool drained
= false;
2706 bool limits_invariant
;
2707 struct page_counter
*counter
= memsw
? &memcg
->memsw
: &memcg
->memory
;
2710 if (signal_pending(current
)) {
2715 mutex_lock(&memcg_max_mutex
);
2717 * Make sure that the new limit (memsw or memory limit) doesn't
2718 * break our basic invariant rule memory.max <= memsw.max.
2720 limits_invariant
= memsw
? max
>= memcg
->memory
.max
:
2721 max
<= memcg
->memsw
.max
;
2722 if (!limits_invariant
) {
2723 mutex_unlock(&memcg_max_mutex
);
2727 if (max
> counter
->max
)
2729 ret
= page_counter_set_max(counter
, max
);
2730 mutex_unlock(&memcg_max_mutex
);
2736 drain_all_stock(memcg
);
2741 if (!try_to_free_mem_cgroup_pages(memcg
, 1,
2742 GFP_KERNEL
, !memsw
)) {
2748 if (!ret
&& enlarge
)
2749 memcg_oom_recover(memcg
);
2754 unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t
*pgdat
, int order
,
2756 unsigned long *total_scanned
)
2758 unsigned long nr_reclaimed
= 0;
2759 struct mem_cgroup_per_node
*mz
, *next_mz
= NULL
;
2760 unsigned long reclaimed
;
2762 struct mem_cgroup_tree_per_node
*mctz
;
2763 unsigned long excess
;
2764 unsigned long nr_scanned
;
2769 mctz
= soft_limit_tree_node(pgdat
->node_id
);
2772 * Do not even bother to check the largest node if the root
2773 * is empty. Do it lockless to prevent lock bouncing. Races
2774 * are acceptable as soft limit is best effort anyway.
2776 if (!mctz
|| RB_EMPTY_ROOT(&mctz
->rb_root
))
2780 * This loop can run a while, specially if mem_cgroup's continuously
2781 * keep exceeding their soft limit and putting the system under
2788 mz
= mem_cgroup_largest_soft_limit_node(mctz
);
2793 reclaimed
= mem_cgroup_soft_reclaim(mz
->memcg
, pgdat
,
2794 gfp_mask
, &nr_scanned
);
2795 nr_reclaimed
+= reclaimed
;
2796 *total_scanned
+= nr_scanned
;
2797 spin_lock_irq(&mctz
->lock
);
2798 __mem_cgroup_remove_exceeded(mz
, mctz
);
2801 * If we failed to reclaim anything from this memory cgroup
2802 * it is time to move on to the next cgroup
2806 next_mz
= __mem_cgroup_largest_soft_limit_node(mctz
);
2808 excess
= soft_limit_excess(mz
->memcg
);
2810 * One school of thought says that we should not add
2811 * back the node to the tree if reclaim returns 0.
2812 * But our reclaim could return 0, simply because due
2813 * to priority we are exposing a smaller subset of
2814 * memory to reclaim from. Consider this as a longer
2817 /* If excess == 0, no tree ops */
2818 __mem_cgroup_insert_exceeded(mz
, mctz
, excess
);
2819 spin_unlock_irq(&mctz
->lock
);
2820 css_put(&mz
->memcg
->css
);
2823 * Could not reclaim anything and there are no more
2824 * mem cgroups to try or we seem to be looping without
2825 * reclaiming anything.
2827 if (!nr_reclaimed
&&
2829 loop
> MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS
))
2831 } while (!nr_reclaimed
);
2833 css_put(&next_mz
->memcg
->css
);
2834 return nr_reclaimed
;
2838 * Test whether @memcg has children, dead or alive. Note that this
2839 * function doesn't care whether @memcg has use_hierarchy enabled and
2840 * returns %true if there are child csses according to the cgroup
2841 * hierarchy. Testing use_hierarchy is the caller's responsiblity.
2843 static inline bool memcg_has_children(struct mem_cgroup
*memcg
)
2848 ret
= css_next_child(NULL
, &memcg
->css
);
2854 * Reclaims as many pages from the given memcg as possible.
2856 * Caller is responsible for holding css reference for memcg.
2858 static int mem_cgroup_force_empty(struct mem_cgroup
*memcg
)
2860 int nr_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
2862 /* we call try-to-free pages for make this cgroup empty */
2863 lru_add_drain_all();
2865 drain_all_stock(memcg
);
2867 /* try to free all pages in this cgroup */
2868 while (nr_retries
&& page_counter_read(&memcg
->memory
)) {
2871 if (signal_pending(current
))
2874 progress
= try_to_free_mem_cgroup_pages(memcg
, 1,
2878 /* maybe some writeback is necessary */
2879 congestion_wait(BLK_RW_ASYNC
, HZ
/10);
2887 static ssize_t
mem_cgroup_force_empty_write(struct kernfs_open_file
*of
,
2888 char *buf
, size_t nbytes
,
2891 struct mem_cgroup
*memcg
= mem_cgroup_from_css(of_css(of
));
2893 if (mem_cgroup_is_root(memcg
))
2895 return mem_cgroup_force_empty(memcg
) ?: nbytes
;
2898 static u64
mem_cgroup_hierarchy_read(struct cgroup_subsys_state
*css
,
2901 return mem_cgroup_from_css(css
)->use_hierarchy
;
2904 static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state
*css
,
2905 struct cftype
*cft
, u64 val
)
2908 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
2909 struct mem_cgroup
*parent_memcg
= mem_cgroup_from_css(memcg
->css
.parent
);
2911 if (memcg
->use_hierarchy
== val
)
2915 * If parent's use_hierarchy is set, we can't make any modifications
2916 * in the child subtrees. If it is unset, then the change can
2917 * occur, provided the current cgroup has no children.
2919 * For the root cgroup, parent_mem is NULL, we allow value to be
2920 * set if there are no children.
2922 if ((!parent_memcg
|| !parent_memcg
->use_hierarchy
) &&
2923 (val
== 1 || val
== 0)) {
2924 if (!memcg_has_children(memcg
))
2925 memcg
->use_hierarchy
= val
;
2934 struct accumulated_stats
{
2935 unsigned long stat
[MEMCG_NR_STAT
];
2936 unsigned long events
[NR_VM_EVENT_ITEMS
];
2937 unsigned long lru_pages
[NR_LRU_LISTS
];
2938 const unsigned int *stats_array
;
2939 const unsigned int *events_array
;
2944 static void accumulate_memcg_tree(struct mem_cgroup
*memcg
,
2945 struct accumulated_stats
*acc
)
2947 struct mem_cgroup
*mi
;
2950 for_each_mem_cgroup_tree(mi
, memcg
) {
2951 for (i
= 0; i
< acc
->stats_size
; i
++)
2952 acc
->stat
[i
] += memcg_page_state(mi
,
2953 acc
->stats_array
? acc
->stats_array
[i
] : i
);
2955 for (i
= 0; i
< acc
->events_size
; i
++)
2956 acc
->events
[i
] += memcg_sum_events(mi
,
2957 acc
->events_array
? acc
->events_array
[i
] : i
);
2959 for (i
= 0; i
< NR_LRU_LISTS
; i
++)
2960 acc
->lru_pages
[i
] +=
2961 mem_cgroup_nr_lru_pages(mi
, BIT(i
));
2965 static unsigned long mem_cgroup_usage(struct mem_cgroup
*memcg
, bool swap
)
2967 unsigned long val
= 0;
2969 if (mem_cgroup_is_root(memcg
)) {
2970 struct mem_cgroup
*iter
;
2972 for_each_mem_cgroup_tree(iter
, memcg
) {
2973 val
+= memcg_page_state(iter
, MEMCG_CACHE
);
2974 val
+= memcg_page_state(iter
, MEMCG_RSS
);
2976 val
+= memcg_page_state(iter
, MEMCG_SWAP
);
2980 val
= page_counter_read(&memcg
->memory
);
2982 val
= page_counter_read(&memcg
->memsw
);
2995 static u64
mem_cgroup_read_u64(struct cgroup_subsys_state
*css
,
2998 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
2999 struct page_counter
*counter
;
3001 switch (MEMFILE_TYPE(cft
->private)) {
3003 counter
= &memcg
->memory
;
3006 counter
= &memcg
->memsw
;
3009 counter
= &memcg
->kmem
;
3012 counter
= &memcg
->tcpmem
;
3018 switch (MEMFILE_ATTR(cft
->private)) {
3020 if (counter
== &memcg
->memory
)
3021 return (u64
)mem_cgroup_usage(memcg
, false) * PAGE_SIZE
;
3022 if (counter
== &memcg
->memsw
)
3023 return (u64
)mem_cgroup_usage(memcg
, true) * PAGE_SIZE
;
3024 return (u64
)page_counter_read(counter
) * PAGE_SIZE
;
3026 return (u64
)counter
->max
* PAGE_SIZE
;
3028 return (u64
)counter
->watermark
* PAGE_SIZE
;
3030 return counter
->failcnt
;
3031 case RES_SOFT_LIMIT
:
3032 return (u64
)memcg
->soft_limit
* PAGE_SIZE
;
3038 #ifdef CONFIG_MEMCG_KMEM
3039 static int memcg_online_kmem(struct mem_cgroup
*memcg
)
3043 if (cgroup_memory_nokmem
)
3046 BUG_ON(memcg
->kmemcg_id
>= 0);
3047 BUG_ON(memcg
->kmem_state
);
3049 memcg_id
= memcg_alloc_cache_id();
3053 static_branch_inc(&memcg_kmem_enabled_key
);
3055 * A memory cgroup is considered kmem-online as soon as it gets
3056 * kmemcg_id. Setting the id after enabling static branching will
3057 * guarantee no one starts accounting before all call sites are
3060 memcg
->kmemcg_id
= memcg_id
;
3061 memcg
->kmem_state
= KMEM_ONLINE
;
3062 INIT_LIST_HEAD(&memcg
->kmem_caches
);
3067 static void memcg_offline_kmem(struct mem_cgroup
*memcg
)
3069 struct cgroup_subsys_state
*css
;
3070 struct mem_cgroup
*parent
, *child
;
3073 if (memcg
->kmem_state
!= KMEM_ONLINE
)
3076 * Clear the online state before clearing memcg_caches array
3077 * entries. The slab_mutex in memcg_deactivate_kmem_caches()
3078 * guarantees that no cache will be created for this cgroup
3079 * after we are done (see memcg_create_kmem_cache()).
3081 memcg
->kmem_state
= KMEM_ALLOCATED
;
3083 memcg_deactivate_kmem_caches(memcg
);
3085 kmemcg_id
= memcg
->kmemcg_id
;
3086 BUG_ON(kmemcg_id
< 0);
3088 parent
= parent_mem_cgroup(memcg
);
3090 parent
= root_mem_cgroup
;
3093 * Change kmemcg_id of this cgroup and all its descendants to the
3094 * parent's id, and then move all entries from this cgroup's list_lrus
3095 * to ones of the parent. After we have finished, all list_lrus
3096 * corresponding to this cgroup are guaranteed to remain empty. The
3097 * ordering is imposed by list_lru_node->lock taken by
3098 * memcg_drain_all_list_lrus().
3100 rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */
3101 css_for_each_descendant_pre(css
, &memcg
->css
) {
3102 child
= mem_cgroup_from_css(css
);
3103 BUG_ON(child
->kmemcg_id
!= kmemcg_id
);
3104 child
->kmemcg_id
= parent
->kmemcg_id
;
3105 if (!memcg
->use_hierarchy
)
3110 memcg_drain_all_list_lrus(kmemcg_id
, parent
);
3112 memcg_free_cache_id(kmemcg_id
);
3115 static void memcg_free_kmem(struct mem_cgroup
*memcg
)
3117 /* css_alloc() failed, offlining didn't happen */
3118 if (unlikely(memcg
->kmem_state
== KMEM_ONLINE
))
3119 memcg_offline_kmem(memcg
);
3121 if (memcg
->kmem_state
== KMEM_ALLOCATED
) {
3122 memcg_destroy_kmem_caches(memcg
);
3123 static_branch_dec(&memcg_kmem_enabled_key
);
3124 WARN_ON(page_counter_read(&memcg
->kmem
));
3128 static int memcg_online_kmem(struct mem_cgroup
*memcg
)
3132 static void memcg_offline_kmem(struct mem_cgroup
*memcg
)
3135 static void memcg_free_kmem(struct mem_cgroup
*memcg
)
3138 #endif /* CONFIG_MEMCG_KMEM */
3140 static int memcg_update_kmem_max(struct mem_cgroup
*memcg
,
3145 mutex_lock(&memcg_max_mutex
);
3146 ret
= page_counter_set_max(&memcg
->kmem
, max
);
3147 mutex_unlock(&memcg_max_mutex
);
3151 static int memcg_update_tcp_max(struct mem_cgroup
*memcg
, unsigned long max
)
3155 mutex_lock(&memcg_max_mutex
);
3157 ret
= page_counter_set_max(&memcg
->tcpmem
, max
);
3161 if (!memcg
->tcpmem_active
) {
3163 * The active flag needs to be written after the static_key
3164 * update. This is what guarantees that the socket activation
3165 * function is the last one to run. See mem_cgroup_sk_alloc()
3166 * for details, and note that we don't mark any socket as
3167 * belonging to this memcg until that flag is up.
3169 * We need to do this, because static_keys will span multiple
3170 * sites, but we can't control their order. If we mark a socket
3171 * as accounted, but the accounting functions are not patched in
3172 * yet, we'll lose accounting.
3174 * We never race with the readers in mem_cgroup_sk_alloc(),
3175 * because when this value change, the code to process it is not
3178 static_branch_inc(&memcg_sockets_enabled_key
);
3179 memcg
->tcpmem_active
= true;
3182 mutex_unlock(&memcg_max_mutex
);
3187 * The user of this function is...
3190 static ssize_t
mem_cgroup_write(struct kernfs_open_file
*of
,
3191 char *buf
, size_t nbytes
, loff_t off
)
3193 struct mem_cgroup
*memcg
= mem_cgroup_from_css(of_css(of
));
3194 unsigned long nr_pages
;
3197 buf
= strstrip(buf
);
3198 ret
= page_counter_memparse(buf
, "-1", &nr_pages
);
3202 switch (MEMFILE_ATTR(of_cft(of
)->private)) {
3204 if (mem_cgroup_is_root(memcg
)) { /* Can't set limit on root */
3208 switch (MEMFILE_TYPE(of_cft(of
)->private)) {
3210 ret
= mem_cgroup_resize_max(memcg
, nr_pages
, false);
3213 ret
= mem_cgroup_resize_max(memcg
, nr_pages
, true);
3216 ret
= memcg_update_kmem_max(memcg
, nr_pages
);
3219 ret
= memcg_update_tcp_max(memcg
, nr_pages
);
3223 case RES_SOFT_LIMIT
:
3224 memcg
->soft_limit
= nr_pages
;
3228 return ret
?: nbytes
;
3231 static ssize_t
mem_cgroup_reset(struct kernfs_open_file
*of
, char *buf
,
3232 size_t nbytes
, loff_t off
)
3234 struct mem_cgroup
*memcg
= mem_cgroup_from_css(of_css(of
));
3235 struct page_counter
*counter
;
3237 switch (MEMFILE_TYPE(of_cft(of
)->private)) {
3239 counter
= &memcg
->memory
;
3242 counter
= &memcg
->memsw
;
3245 counter
= &memcg
->kmem
;
3248 counter
= &memcg
->tcpmem
;
3254 switch (MEMFILE_ATTR(of_cft(of
)->private)) {
3256 page_counter_reset_watermark(counter
);
3259 counter
->failcnt
= 0;
3268 static u64
mem_cgroup_move_charge_read(struct cgroup_subsys_state
*css
,
3271 return mem_cgroup_from_css(css
)->move_charge_at_immigrate
;
3275 static int mem_cgroup_move_charge_write(struct cgroup_subsys_state
*css
,
3276 struct cftype
*cft
, u64 val
)
3278 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
3280 if (val
& ~MOVE_MASK
)
3284 * No kind of locking is needed in here, because ->can_attach() will
3285 * check this value once in the beginning of the process, and then carry
3286 * on with stale data. This means that changes to this value will only
3287 * affect task migrations starting after the change.
3289 memcg
->move_charge_at_immigrate
= val
;
3293 static int mem_cgroup_move_charge_write(struct cgroup_subsys_state
*css
,
3294 struct cftype
*cft
, u64 val
)
3301 static int memcg_numa_stat_show(struct seq_file
*m
, void *v
)
3305 unsigned int lru_mask
;
3308 static const struct numa_stat stats
[] = {
3309 { "total", LRU_ALL
},
3310 { "file", LRU_ALL_FILE
},
3311 { "anon", LRU_ALL_ANON
},
3312 { "unevictable", BIT(LRU_UNEVICTABLE
) },
3314 const struct numa_stat
*stat
;
3317 struct mem_cgroup
*memcg
= mem_cgroup_from_css(seq_css(m
));
3319 for (stat
= stats
; stat
< stats
+ ARRAY_SIZE(stats
); stat
++) {
3320 nr
= mem_cgroup_nr_lru_pages(memcg
, stat
->lru_mask
);
3321 seq_printf(m
, "%s=%lu", stat
->name
, nr
);
3322 for_each_node_state(nid
, N_MEMORY
) {
3323 nr
= mem_cgroup_node_nr_lru_pages(memcg
, nid
,
3325 seq_printf(m
, " N%d=%lu", nid
, nr
);
3330 for (stat
= stats
; stat
< stats
+ ARRAY_SIZE(stats
); stat
++) {
3331 struct mem_cgroup
*iter
;
3334 for_each_mem_cgroup_tree(iter
, memcg
)
3335 nr
+= mem_cgroup_nr_lru_pages(iter
, stat
->lru_mask
);
3336 seq_printf(m
, "hierarchical_%s=%lu", stat
->name
, nr
);
3337 for_each_node_state(nid
, N_MEMORY
) {
3339 for_each_mem_cgroup_tree(iter
, memcg
)
3340 nr
+= mem_cgroup_node_nr_lru_pages(
3341 iter
, nid
, stat
->lru_mask
);
3342 seq_printf(m
, " N%d=%lu", nid
, nr
);
3349 #endif /* CONFIG_NUMA */
3351 /* Universal VM events cgroup1 shows, original sort order */
3352 static const unsigned int memcg1_events
[] = {
3359 static const char *const memcg1_event_names
[] = {
3366 static int memcg_stat_show(struct seq_file
*m
, void *v
)
3368 struct mem_cgroup
*memcg
= mem_cgroup_from_css(seq_css(m
));
3369 unsigned long memory
, memsw
;
3370 struct mem_cgroup
*mi
;
3372 struct accumulated_stats acc
;
3374 BUILD_BUG_ON(ARRAY_SIZE(memcg1_stat_names
) != ARRAY_SIZE(memcg1_stats
));
3375 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names
) != NR_LRU_LISTS
);
3377 for (i
= 0; i
< ARRAY_SIZE(memcg1_stats
); i
++) {
3378 if (memcg1_stats
[i
] == MEMCG_SWAP
&& !do_memsw_account())
3380 seq_printf(m
, "%s %lu\n", memcg1_stat_names
[i
],
3381 memcg_page_state(memcg
, memcg1_stats
[i
]) *
3385 for (i
= 0; i
< ARRAY_SIZE(memcg1_events
); i
++)
3386 seq_printf(m
, "%s %lu\n", memcg1_event_names
[i
],
3387 memcg_sum_events(memcg
, memcg1_events
[i
]));
3389 for (i
= 0; i
< NR_LRU_LISTS
; i
++)
3390 seq_printf(m
, "%s %lu\n", mem_cgroup_lru_names
[i
],
3391 mem_cgroup_nr_lru_pages(memcg
, BIT(i
)) * PAGE_SIZE
);
3393 /* Hierarchical information */
3394 memory
= memsw
= PAGE_COUNTER_MAX
;
3395 for (mi
= memcg
; mi
; mi
= parent_mem_cgroup(mi
)) {
3396 memory
= min(memory
, mi
->memory
.max
);
3397 memsw
= min(memsw
, mi
->memsw
.max
);
3399 seq_printf(m
, "hierarchical_memory_limit %llu\n",
3400 (u64
)memory
* PAGE_SIZE
);
3401 if (do_memsw_account())
3402 seq_printf(m
, "hierarchical_memsw_limit %llu\n",
3403 (u64
)memsw
* PAGE_SIZE
);
3405 memset(&acc
, 0, sizeof(acc
));
3406 acc
.stats_size
= ARRAY_SIZE(memcg1_stats
);
3407 acc
.stats_array
= memcg1_stats
;
3408 acc
.events_size
= ARRAY_SIZE(memcg1_events
);
3409 acc
.events_array
= memcg1_events
;
3410 accumulate_memcg_tree(memcg
, &acc
);
3412 for (i
= 0; i
< ARRAY_SIZE(memcg1_stats
); i
++) {
3413 if (memcg1_stats
[i
] == MEMCG_SWAP
&& !do_memsw_account())
3415 seq_printf(m
, "total_%s %llu\n", memcg1_stat_names
[i
],
3416 (u64
)acc
.stat
[i
] * PAGE_SIZE
);
3419 for (i
= 0; i
< ARRAY_SIZE(memcg1_events
); i
++)
3420 seq_printf(m
, "total_%s %llu\n", memcg1_event_names
[i
],
3421 (u64
)acc
.events
[i
]);
3423 for (i
= 0; i
< NR_LRU_LISTS
; i
++)
3424 seq_printf(m
, "total_%s %llu\n", mem_cgroup_lru_names
[i
],
3425 (u64
)acc
.lru_pages
[i
] * PAGE_SIZE
);
3427 #ifdef CONFIG_DEBUG_VM
3430 struct mem_cgroup_per_node
*mz
;
3431 struct zone_reclaim_stat
*rstat
;
3432 unsigned long recent_rotated
[2] = {0, 0};
3433 unsigned long recent_scanned
[2] = {0, 0};
3435 for_each_online_pgdat(pgdat
) {
3436 mz
= mem_cgroup_nodeinfo(memcg
, pgdat
->node_id
);
3437 rstat
= &mz
->lruvec
.reclaim_stat
;
3439 recent_rotated
[0] += rstat
->recent_rotated
[0];
3440 recent_rotated
[1] += rstat
->recent_rotated
[1];
3441 recent_scanned
[0] += rstat
->recent_scanned
[0];
3442 recent_scanned
[1] += rstat
->recent_scanned
[1];
3444 seq_printf(m
, "recent_rotated_anon %lu\n", recent_rotated
[0]);
3445 seq_printf(m
, "recent_rotated_file %lu\n", recent_rotated
[1]);
3446 seq_printf(m
, "recent_scanned_anon %lu\n", recent_scanned
[0]);
3447 seq_printf(m
, "recent_scanned_file %lu\n", recent_scanned
[1]);
3454 static u64
mem_cgroup_swappiness_read(struct cgroup_subsys_state
*css
,
3457 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
3459 return mem_cgroup_swappiness(memcg
);
3462 static int mem_cgroup_swappiness_write(struct cgroup_subsys_state
*css
,
3463 struct cftype
*cft
, u64 val
)
3465 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
3471 memcg
->swappiness
= val
;
3473 vm_swappiness
= val
;
3478 static void __mem_cgroup_threshold(struct mem_cgroup
*memcg
, bool swap
)
3480 struct mem_cgroup_threshold_ary
*t
;
3481 unsigned long usage
;
3486 t
= rcu_dereference(memcg
->thresholds
.primary
);
3488 t
= rcu_dereference(memcg
->memsw_thresholds
.primary
);
3493 usage
= mem_cgroup_usage(memcg
, swap
);
3496 * current_threshold points to threshold just below or equal to usage.
3497 * If it's not true, a threshold was crossed after last
3498 * call of __mem_cgroup_threshold().
3500 i
= t
->current_threshold
;
3503 * Iterate backward over array of thresholds starting from
3504 * current_threshold and check if a threshold is crossed.
3505 * If none of thresholds below usage is crossed, we read
3506 * only one element of the array here.
3508 for (; i
>= 0 && unlikely(t
->entries
[i
].threshold
> usage
); i
--)
3509 eventfd_signal(t
->entries
[i
].eventfd
, 1);
3511 /* i = current_threshold + 1 */
3515 * Iterate forward over array of thresholds starting from
3516 * current_threshold+1 and check if a threshold is crossed.
3517 * If none of thresholds above usage is crossed, we read
3518 * only one element of the array here.
3520 for (; i
< t
->size
&& unlikely(t
->entries
[i
].threshold
<= usage
); i
++)
3521 eventfd_signal(t
->entries
[i
].eventfd
, 1);
3523 /* Update current_threshold */
3524 t
->current_threshold
= i
- 1;
3529 static void mem_cgroup_threshold(struct mem_cgroup
*memcg
)
3532 __mem_cgroup_threshold(memcg
, false);
3533 if (do_memsw_account())
3534 __mem_cgroup_threshold(memcg
, true);
3536 memcg
= parent_mem_cgroup(memcg
);
3540 static int compare_thresholds(const void *a
, const void *b
)
3542 const struct mem_cgroup_threshold
*_a
= a
;
3543 const struct mem_cgroup_threshold
*_b
= b
;
3545 if (_a
->threshold
> _b
->threshold
)
3548 if (_a
->threshold
< _b
->threshold
)
3554 static int mem_cgroup_oom_notify_cb(struct mem_cgroup
*memcg
)
3556 struct mem_cgroup_eventfd_list
*ev
;
3558 spin_lock(&memcg_oom_lock
);
3560 list_for_each_entry(ev
, &memcg
->oom_notify
, list
)
3561 eventfd_signal(ev
->eventfd
, 1);
3563 spin_unlock(&memcg_oom_lock
);
3567 static void mem_cgroup_oom_notify(struct mem_cgroup
*memcg
)
3569 struct mem_cgroup
*iter
;
3571 for_each_mem_cgroup_tree(iter
, memcg
)
3572 mem_cgroup_oom_notify_cb(iter
);
3575 static int __mem_cgroup_usage_register_event(struct mem_cgroup
*memcg
,
3576 struct eventfd_ctx
*eventfd
, const char *args
, enum res_type type
)
3578 struct mem_cgroup_thresholds
*thresholds
;
3579 struct mem_cgroup_threshold_ary
*new;
3580 unsigned long threshold
;
3581 unsigned long usage
;
3584 ret
= page_counter_memparse(args
, "-1", &threshold
);
3588 mutex_lock(&memcg
->thresholds_lock
);
3591 thresholds
= &memcg
->thresholds
;
3592 usage
= mem_cgroup_usage(memcg
, false);
3593 } else if (type
== _MEMSWAP
) {
3594 thresholds
= &memcg
->memsw_thresholds
;
3595 usage
= mem_cgroup_usage(memcg
, true);
3599 /* Check if a threshold crossed before adding a new one */
3600 if (thresholds
->primary
)
3601 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
3603 size
= thresholds
->primary
? thresholds
->primary
->size
+ 1 : 1;
3605 /* Allocate memory for new array of thresholds */
3606 new = kmalloc(sizeof(*new) + size
* sizeof(struct mem_cgroup_threshold
),
3614 /* Copy thresholds (if any) to new array */
3615 if (thresholds
->primary
) {
3616 memcpy(new->entries
, thresholds
->primary
->entries
, (size
- 1) *
3617 sizeof(struct mem_cgroup_threshold
));
3620 /* Add new threshold */
3621 new->entries
[size
- 1].eventfd
= eventfd
;
3622 new->entries
[size
- 1].threshold
= threshold
;
3624 /* Sort thresholds. Registering of new threshold isn't time-critical */
3625 sort(new->entries
, size
, sizeof(struct mem_cgroup_threshold
),
3626 compare_thresholds
, NULL
);
3628 /* Find current threshold */
3629 new->current_threshold
= -1;
3630 for (i
= 0; i
< size
; i
++) {
3631 if (new->entries
[i
].threshold
<= usage
) {
3633 * new->current_threshold will not be used until
3634 * rcu_assign_pointer(), so it's safe to increment
3637 ++new->current_threshold
;
3642 /* Free old spare buffer and save old primary buffer as spare */
3643 kfree(thresholds
->spare
);
3644 thresholds
->spare
= thresholds
->primary
;
3646 rcu_assign_pointer(thresholds
->primary
, new);
3648 /* To be sure that nobody uses thresholds */
3652 mutex_unlock(&memcg
->thresholds_lock
);
3657 static int mem_cgroup_usage_register_event(struct mem_cgroup
*memcg
,
3658 struct eventfd_ctx
*eventfd
, const char *args
)
3660 return __mem_cgroup_usage_register_event(memcg
, eventfd
, args
, _MEM
);
3663 static int memsw_cgroup_usage_register_event(struct mem_cgroup
*memcg
,
3664 struct eventfd_ctx
*eventfd
, const char *args
)
3666 return __mem_cgroup_usage_register_event(memcg
, eventfd
, args
, _MEMSWAP
);
3669 static void __mem_cgroup_usage_unregister_event(struct mem_cgroup
*memcg
,
3670 struct eventfd_ctx
*eventfd
, enum res_type type
)
3672 struct mem_cgroup_thresholds
*thresholds
;
3673 struct mem_cgroup_threshold_ary
*new;
3674 unsigned long usage
;
3677 mutex_lock(&memcg
->thresholds_lock
);
3680 thresholds
= &memcg
->thresholds
;
3681 usage
= mem_cgroup_usage(memcg
, false);
3682 } else if (type
== _MEMSWAP
) {
3683 thresholds
= &memcg
->memsw_thresholds
;
3684 usage
= mem_cgroup_usage(memcg
, true);
3688 if (!thresholds
->primary
)
3691 /* Check if a threshold crossed before removing */
3692 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
3694 /* Calculate new number of threshold */
3696 for (i
= 0; i
< thresholds
->primary
->size
; i
++) {
3697 if (thresholds
->primary
->entries
[i
].eventfd
!= eventfd
)
3701 new = thresholds
->spare
;
3703 /* Set thresholds array to NULL if we don't have thresholds */
3712 /* Copy thresholds and find current threshold */
3713 new->current_threshold
= -1;
3714 for (i
= 0, j
= 0; i
< thresholds
->primary
->size
; i
++) {
3715 if (thresholds
->primary
->entries
[i
].eventfd
== eventfd
)
3718 new->entries
[j
] = thresholds
->primary
->entries
[i
];
3719 if (new->entries
[j
].threshold
<= usage
) {
3721 * new->current_threshold will not be used
3722 * until rcu_assign_pointer(), so it's safe to increment
3725 ++new->current_threshold
;
3731 /* Swap primary and spare array */
3732 thresholds
->spare
= thresholds
->primary
;
3734 rcu_assign_pointer(thresholds
->primary
, new);
3736 /* To be sure that nobody uses thresholds */
3739 /* If all events are unregistered, free the spare array */
3741 kfree(thresholds
->spare
);
3742 thresholds
->spare
= NULL
;
3745 mutex_unlock(&memcg
->thresholds_lock
);
3748 static void mem_cgroup_usage_unregister_event(struct mem_cgroup
*memcg
,
3749 struct eventfd_ctx
*eventfd
)
3751 return __mem_cgroup_usage_unregister_event(memcg
, eventfd
, _MEM
);
3754 static void memsw_cgroup_usage_unregister_event(struct mem_cgroup
*memcg
,
3755 struct eventfd_ctx
*eventfd
)
3757 return __mem_cgroup_usage_unregister_event(memcg
, eventfd
, _MEMSWAP
);
3760 static int mem_cgroup_oom_register_event(struct mem_cgroup
*memcg
,
3761 struct eventfd_ctx
*eventfd
, const char *args
)
3763 struct mem_cgroup_eventfd_list
*event
;
3765 event
= kmalloc(sizeof(*event
), GFP_KERNEL
);
3769 spin_lock(&memcg_oom_lock
);
3771 event
->eventfd
= eventfd
;
3772 list_add(&event
->list
, &memcg
->oom_notify
);
3774 /* already in OOM ? */
3775 if (memcg
->under_oom
)
3776 eventfd_signal(eventfd
, 1);
3777 spin_unlock(&memcg_oom_lock
);
3782 static void mem_cgroup_oom_unregister_event(struct mem_cgroup
*memcg
,
3783 struct eventfd_ctx
*eventfd
)
3785 struct mem_cgroup_eventfd_list
*ev
, *tmp
;
3787 spin_lock(&memcg_oom_lock
);
3789 list_for_each_entry_safe(ev
, tmp
, &memcg
->oom_notify
, list
) {
3790 if (ev
->eventfd
== eventfd
) {
3791 list_del(&ev
->list
);
3796 spin_unlock(&memcg_oom_lock
);
3799 static int mem_cgroup_oom_control_read(struct seq_file
*sf
, void *v
)
3801 struct mem_cgroup
*memcg
= mem_cgroup_from_css(seq_css(sf
));
3803 seq_printf(sf
, "oom_kill_disable %d\n", memcg
->oom_kill_disable
);
3804 seq_printf(sf
, "under_oom %d\n", (bool)memcg
->under_oom
);
3805 seq_printf(sf
, "oom_kill %lu\n",
3806 atomic_long_read(&memcg
->memory_events
[MEMCG_OOM_KILL
]));
3810 static int mem_cgroup_oom_control_write(struct cgroup_subsys_state
*css
,
3811 struct cftype
*cft
, u64 val
)
3813 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
3815 /* cannot set to root cgroup and only 0 and 1 are allowed */
3816 if (!css
->parent
|| !((val
== 0) || (val
== 1)))
3819 memcg
->oom_kill_disable
= val
;
3821 memcg_oom_recover(memcg
);
3826 #ifdef CONFIG_CGROUP_WRITEBACK
3828 static int memcg_wb_domain_init(struct mem_cgroup
*memcg
, gfp_t gfp
)
3830 return wb_domain_init(&memcg
->cgwb_domain
, gfp
);
3833 static void memcg_wb_domain_exit(struct mem_cgroup
*memcg
)
3835 wb_domain_exit(&memcg
->cgwb_domain
);
3838 static void memcg_wb_domain_size_changed(struct mem_cgroup
*memcg
)
3840 wb_domain_size_changed(&memcg
->cgwb_domain
);
3843 struct wb_domain
*mem_cgroup_wb_domain(struct bdi_writeback
*wb
)
3845 struct mem_cgroup
*memcg
= mem_cgroup_from_css(wb
->memcg_css
);
3847 if (!memcg
->css
.parent
)
3850 return &memcg
->cgwb_domain
;
3854 * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg
3855 * @wb: bdi_writeback in question
3856 * @pfilepages: out parameter for number of file pages
3857 * @pheadroom: out parameter for number of allocatable pages according to memcg
3858 * @pdirty: out parameter for number of dirty pages
3859 * @pwriteback: out parameter for number of pages under writeback
3861 * Determine the numbers of file, headroom, dirty, and writeback pages in
3862 * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom
3863 * is a bit more involved.
3865 * A memcg's headroom is "min(max, high) - used". In the hierarchy, the
3866 * headroom is calculated as the lowest headroom of itself and the
3867 * ancestors. Note that this doesn't consider the actual amount of
3868 * available memory in the system. The caller should further cap
3869 * *@pheadroom accordingly.
3871 void mem_cgroup_wb_stats(struct bdi_writeback
*wb
, unsigned long *pfilepages
,
3872 unsigned long *pheadroom
, unsigned long *pdirty
,
3873 unsigned long *pwriteback
)
3875 struct mem_cgroup
*memcg
= mem_cgroup_from_css(wb
->memcg_css
);
3876 struct mem_cgroup
*parent
;
3878 *pdirty
= memcg_page_state(memcg
, NR_FILE_DIRTY
);
3880 /* this should eventually include NR_UNSTABLE_NFS */
3881 *pwriteback
= memcg_page_state(memcg
, NR_WRITEBACK
);
3882 *pfilepages
= mem_cgroup_nr_lru_pages(memcg
, (1 << LRU_INACTIVE_FILE
) |
3883 (1 << LRU_ACTIVE_FILE
));
3884 *pheadroom
= PAGE_COUNTER_MAX
;
3886 while ((parent
= parent_mem_cgroup(memcg
))) {
3887 unsigned long ceiling
= min(memcg
->memory
.max
, memcg
->high
);
3888 unsigned long used
= page_counter_read(&memcg
->memory
);
3890 *pheadroom
= min(*pheadroom
, ceiling
- min(ceiling
, used
));
3895 #else /* CONFIG_CGROUP_WRITEBACK */
3897 static int memcg_wb_domain_init(struct mem_cgroup
*memcg
, gfp_t gfp
)
3902 static void memcg_wb_domain_exit(struct mem_cgroup
*memcg
)
3906 static void memcg_wb_domain_size_changed(struct mem_cgroup
*memcg
)
3910 #endif /* CONFIG_CGROUP_WRITEBACK */
3913 * DO NOT USE IN NEW FILES.
3915 * "cgroup.event_control" implementation.
3917 * This is way over-engineered. It tries to support fully configurable
3918 * events for each user. Such level of flexibility is completely
3919 * unnecessary especially in the light of the planned unified hierarchy.
3921 * Please deprecate this and replace with something simpler if at all
3926 * Unregister event and free resources.
3928 * Gets called from workqueue.
3930 static void memcg_event_remove(struct work_struct
*work
)
3932 struct mem_cgroup_event
*event
=
3933 container_of(work
, struct mem_cgroup_event
, remove
);
3934 struct mem_cgroup
*memcg
= event
->memcg
;
3936 remove_wait_queue(event
->wqh
, &event
->wait
);
3938 event
->unregister_event(memcg
, event
->eventfd
);
3940 /* Notify userspace the event is going away. */
3941 eventfd_signal(event
->eventfd
, 1);
3943 eventfd_ctx_put(event
->eventfd
);
3945 css_put(&memcg
->css
);
3949 * Gets called on EPOLLHUP on eventfd when user closes it.
3951 * Called with wqh->lock held and interrupts disabled.
3953 static int memcg_event_wake(wait_queue_entry_t
*wait
, unsigned mode
,
3954 int sync
, void *key
)
3956 struct mem_cgroup_event
*event
=
3957 container_of(wait
, struct mem_cgroup_event
, wait
);
3958 struct mem_cgroup
*memcg
= event
->memcg
;
3959 __poll_t flags
= key_to_poll(key
);
3961 if (flags
& EPOLLHUP
) {
3963 * If the event has been detached at cgroup removal, we
3964 * can simply return knowing the other side will cleanup
3967 * We can't race against event freeing since the other
3968 * side will require wqh->lock via remove_wait_queue(),
3971 spin_lock(&memcg
->event_list_lock
);
3972 if (!list_empty(&event
->list
)) {
3973 list_del_init(&event
->list
);
3975 * We are in atomic context, but cgroup_event_remove()
3976 * may sleep, so we have to call it in workqueue.
3978 schedule_work(&event
->remove
);
3980 spin_unlock(&memcg
->event_list_lock
);
3986 static void memcg_event_ptable_queue_proc(struct file
*file
,
3987 wait_queue_head_t
*wqh
, poll_table
*pt
)
3989 struct mem_cgroup_event
*event
=
3990 container_of(pt
, struct mem_cgroup_event
, pt
);
3993 add_wait_queue(wqh
, &event
->wait
);
3997 * DO NOT USE IN NEW FILES.
3999 * Parse input and register new cgroup event handler.
4001 * Input must be in format '<event_fd> <control_fd> <args>'.
4002 * Interpretation of args is defined by control file implementation.
4004 static ssize_t
memcg_write_event_control(struct kernfs_open_file
*of
,
4005 char *buf
, size_t nbytes
, loff_t off
)
4007 struct cgroup_subsys_state
*css
= of_css(of
);
4008 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
4009 struct mem_cgroup_event
*event
;
4010 struct cgroup_subsys_state
*cfile_css
;
4011 unsigned int efd
, cfd
;
4018 buf
= strstrip(buf
);
4020 efd
= simple_strtoul(buf
, &endp
, 10);
4025 cfd
= simple_strtoul(buf
, &endp
, 10);
4026 if ((*endp
!= ' ') && (*endp
!= '\0'))
4030 event
= kzalloc(sizeof(*event
), GFP_KERNEL
);
4034 event
->memcg
= memcg
;
4035 INIT_LIST_HEAD(&event
->list
);
4036 init_poll_funcptr(&event
->pt
, memcg_event_ptable_queue_proc
);
4037 init_waitqueue_func_entry(&event
->wait
, memcg_event_wake
);
4038 INIT_WORK(&event
->remove
, memcg_event_remove
);
4046 event
->eventfd
= eventfd_ctx_fileget(efile
.file
);
4047 if (IS_ERR(event
->eventfd
)) {
4048 ret
= PTR_ERR(event
->eventfd
);
4055 goto out_put_eventfd
;
4058 /* the process need read permission on control file */
4059 /* AV: shouldn't we check that it's been opened for read instead? */
4060 ret
= inode_permission(file_inode(cfile
.file
), MAY_READ
);
4065 * Determine the event callbacks and set them in @event. This used
4066 * to be done via struct cftype but cgroup core no longer knows
4067 * about these events. The following is crude but the whole thing
4068 * is for compatibility anyway.
4070 * DO NOT ADD NEW FILES.
4072 name
= cfile
.file
->f_path
.dentry
->d_name
.name
;
4074 if (!strcmp(name
, "memory.usage_in_bytes")) {
4075 event
->register_event
= mem_cgroup_usage_register_event
;
4076 event
->unregister_event
= mem_cgroup_usage_unregister_event
;
4077 } else if (!strcmp(name
, "memory.oom_control")) {
4078 event
->register_event
= mem_cgroup_oom_register_event
;
4079 event
->unregister_event
= mem_cgroup_oom_unregister_event
;
4080 } else if (!strcmp(name
, "memory.pressure_level")) {
4081 event
->register_event
= vmpressure_register_event
;
4082 event
->unregister_event
= vmpressure_unregister_event
;
4083 } else if (!strcmp(name
, "memory.memsw.usage_in_bytes")) {
4084 event
->register_event
= memsw_cgroup_usage_register_event
;
4085 event
->unregister_event
= memsw_cgroup_usage_unregister_event
;
4092 * Verify @cfile should belong to @css. Also, remaining events are
4093 * automatically removed on cgroup destruction but the removal is
4094 * asynchronous, so take an extra ref on @css.
4096 cfile_css
= css_tryget_online_from_dir(cfile
.file
->f_path
.dentry
->d_parent
,
4097 &memory_cgrp_subsys
);
4099 if (IS_ERR(cfile_css
))
4101 if (cfile_css
!= css
) {
4106 ret
= event
->register_event(memcg
, event
->eventfd
, buf
);
4110 vfs_poll(efile
.file
, &event
->pt
);
4112 spin_lock(&memcg
->event_list_lock
);
4113 list_add(&event
->list
, &memcg
->event_list
);
4114 spin_unlock(&memcg
->event_list_lock
);
4126 eventfd_ctx_put(event
->eventfd
);
4135 static struct cftype mem_cgroup_legacy_files
[] = {
4137 .name
= "usage_in_bytes",
4138 .private = MEMFILE_PRIVATE(_MEM
, RES_USAGE
),
4139 .read_u64
= mem_cgroup_read_u64
,
4142 .name
= "max_usage_in_bytes",
4143 .private = MEMFILE_PRIVATE(_MEM
, RES_MAX_USAGE
),
4144 .write
= mem_cgroup_reset
,
4145 .read_u64
= mem_cgroup_read_u64
,
4148 .name
= "limit_in_bytes",
4149 .private = MEMFILE_PRIVATE(_MEM
, RES_LIMIT
),
4150 .write
= mem_cgroup_write
,
4151 .read_u64
= mem_cgroup_read_u64
,
4154 .name
= "soft_limit_in_bytes",
4155 .private = MEMFILE_PRIVATE(_MEM
, RES_SOFT_LIMIT
),
4156 .write
= mem_cgroup_write
,
4157 .read_u64
= mem_cgroup_read_u64
,
4161 .private = MEMFILE_PRIVATE(_MEM
, RES_FAILCNT
),
4162 .write
= mem_cgroup_reset
,
4163 .read_u64
= mem_cgroup_read_u64
,
4167 .seq_show
= memcg_stat_show
,
4170 .name
= "force_empty",
4171 .write
= mem_cgroup_force_empty_write
,
4174 .name
= "use_hierarchy",
4175 .write_u64
= mem_cgroup_hierarchy_write
,
4176 .read_u64
= mem_cgroup_hierarchy_read
,
4179 .name
= "cgroup.event_control", /* XXX: for compat */
4180 .write
= memcg_write_event_control
,
4181 .flags
= CFTYPE_NO_PREFIX
| CFTYPE_WORLD_WRITABLE
,
4184 .name
= "swappiness",
4185 .read_u64
= mem_cgroup_swappiness_read
,
4186 .write_u64
= mem_cgroup_swappiness_write
,
4189 .name
= "move_charge_at_immigrate",
4190 .read_u64
= mem_cgroup_move_charge_read
,
4191 .write_u64
= mem_cgroup_move_charge_write
,
4194 .name
= "oom_control",
4195 .seq_show
= mem_cgroup_oom_control_read
,
4196 .write_u64
= mem_cgroup_oom_control_write
,
4197 .private = MEMFILE_PRIVATE(_OOM_TYPE
, OOM_CONTROL
),
4200 .name
= "pressure_level",
4204 .name
= "numa_stat",
4205 .seq_show
= memcg_numa_stat_show
,
4209 .name
= "kmem.limit_in_bytes",
4210 .private = MEMFILE_PRIVATE(_KMEM
, RES_LIMIT
),
4211 .write
= mem_cgroup_write
,
4212 .read_u64
= mem_cgroup_read_u64
,
4215 .name
= "kmem.usage_in_bytes",
4216 .private = MEMFILE_PRIVATE(_KMEM
, RES_USAGE
),
4217 .read_u64
= mem_cgroup_read_u64
,
4220 .name
= "kmem.failcnt",
4221 .private = MEMFILE_PRIVATE(_KMEM
, RES_FAILCNT
),
4222 .write
= mem_cgroup_reset
,
4223 .read_u64
= mem_cgroup_read_u64
,
4226 .name
= "kmem.max_usage_in_bytes",
4227 .private = MEMFILE_PRIVATE(_KMEM
, RES_MAX_USAGE
),
4228 .write
= mem_cgroup_reset
,
4229 .read_u64
= mem_cgroup_read_u64
,
4231 #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
4233 .name
= "kmem.slabinfo",
4234 .seq_start
= memcg_slab_start
,
4235 .seq_next
= memcg_slab_next
,
4236 .seq_stop
= memcg_slab_stop
,
4237 .seq_show
= memcg_slab_show
,
4241 .name
= "kmem.tcp.limit_in_bytes",
4242 .private = MEMFILE_PRIVATE(_TCP
, RES_LIMIT
),
4243 .write
= mem_cgroup_write
,
4244 .read_u64
= mem_cgroup_read_u64
,
4247 .name
= "kmem.tcp.usage_in_bytes",
4248 .private = MEMFILE_PRIVATE(_TCP
, RES_USAGE
),
4249 .read_u64
= mem_cgroup_read_u64
,
4252 .name
= "kmem.tcp.failcnt",
4253 .private = MEMFILE_PRIVATE(_TCP
, RES_FAILCNT
),
4254 .write
= mem_cgroup_reset
,
4255 .read_u64
= mem_cgroup_read_u64
,
4258 .name
= "kmem.tcp.max_usage_in_bytes",
4259 .private = MEMFILE_PRIVATE(_TCP
, RES_MAX_USAGE
),
4260 .write
= mem_cgroup_reset
,
4261 .read_u64
= mem_cgroup_read_u64
,
4263 { }, /* terminate */
4267 * Private memory cgroup IDR
4269 * Swap-out records and page cache shadow entries need to store memcg
4270 * references in constrained space, so we maintain an ID space that is
4271 * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of
4272 * memory-controlled cgroups to 64k.
4274 * However, there usually are many references to the oflline CSS after
4275 * the cgroup has been destroyed, such as page cache or reclaimable
4276 * slab objects, that don't need to hang on to the ID. We want to keep
4277 * those dead CSS from occupying IDs, or we might quickly exhaust the
4278 * relatively small ID space and prevent the creation of new cgroups
4279 * even when there are much fewer than 64k cgroups - possibly none.
4281 * Maintain a private 16-bit ID space for memcg, and allow the ID to
4282 * be freed and recycled when it's no longer needed, which is usually
4283 * when the CSS is offlined.
4285 * The only exception to that are records of swapped out tmpfs/shmem
4286 * pages that need to be attributed to live ancestors on swapin. But
4287 * those references are manageable from userspace.
4290 static DEFINE_IDR(mem_cgroup_idr
);
4292 static void mem_cgroup_id_remove(struct mem_cgroup
*memcg
)
4294 if (memcg
->id
.id
> 0) {
4295 idr_remove(&mem_cgroup_idr
, memcg
->id
.id
);
4300 static void mem_cgroup_id_get_many(struct mem_cgroup
*memcg
, unsigned int n
)
4302 refcount_add(n
, &memcg
->id
.ref
);
4305 static void mem_cgroup_id_put_many(struct mem_cgroup
*memcg
, unsigned int n
)
4307 if (refcount_sub_and_test(n
, &memcg
->id
.ref
)) {
4308 mem_cgroup_id_remove(memcg
);
4310 /* Memcg ID pins CSS */
4311 css_put(&memcg
->css
);
4315 static inline void mem_cgroup_id_get(struct mem_cgroup
*memcg
)
4317 mem_cgroup_id_get_many(memcg
, 1);
4320 static inline void mem_cgroup_id_put(struct mem_cgroup
*memcg
)
4322 mem_cgroup_id_put_many(memcg
, 1);
4326 * mem_cgroup_from_id - look up a memcg from a memcg id
4327 * @id: the memcg id to look up
4329 * Caller must hold rcu_read_lock().
4331 struct mem_cgroup
*mem_cgroup_from_id(unsigned short id
)
4333 WARN_ON_ONCE(!rcu_read_lock_held());
4334 return idr_find(&mem_cgroup_idr
, id
);
4337 static int alloc_mem_cgroup_per_node_info(struct mem_cgroup
*memcg
, int node
)
4339 struct mem_cgroup_per_node
*pn
;
4342 * This routine is called against possible nodes.
4343 * But it's BUG to call kmalloc() against offline node.
4345 * TODO: this routine can waste much memory for nodes which will
4346 * never be onlined. It's better to use memory hotplug callback
4349 if (!node_state(node
, N_NORMAL_MEMORY
))
4351 pn
= kzalloc_node(sizeof(*pn
), GFP_KERNEL
, tmp
);
4355 pn
->lruvec_stat_cpu
= alloc_percpu(struct lruvec_stat
);
4356 if (!pn
->lruvec_stat_cpu
) {
4361 lruvec_init(&pn
->lruvec
);
4362 pn
->usage_in_excess
= 0;
4363 pn
->on_tree
= false;
4366 memcg
->nodeinfo
[node
] = pn
;
4370 static void free_mem_cgroup_per_node_info(struct mem_cgroup
*memcg
, int node
)
4372 struct mem_cgroup_per_node
*pn
= memcg
->nodeinfo
[node
];
4377 free_percpu(pn
->lruvec_stat_cpu
);
4381 static void __mem_cgroup_free(struct mem_cgroup
*memcg
)
4386 free_mem_cgroup_per_node_info(memcg
, node
);
4387 free_percpu(memcg
->stat_cpu
);
4391 static void mem_cgroup_free(struct mem_cgroup
*memcg
)
4393 memcg_wb_domain_exit(memcg
);
4394 __mem_cgroup_free(memcg
);
4397 static struct mem_cgroup
*mem_cgroup_alloc(void)
4399 struct mem_cgroup
*memcg
;
4403 size
= sizeof(struct mem_cgroup
);
4404 size
+= nr_node_ids
* sizeof(struct mem_cgroup_per_node
*);
4406 memcg
= kzalloc(size
, GFP_KERNEL
);
4410 memcg
->id
.id
= idr_alloc(&mem_cgroup_idr
, NULL
,
4411 1, MEM_CGROUP_ID_MAX
,
4413 if (memcg
->id
.id
< 0)
4416 memcg
->stat_cpu
= alloc_percpu(struct mem_cgroup_stat_cpu
);
4417 if (!memcg
->stat_cpu
)
4421 if (alloc_mem_cgroup_per_node_info(memcg
, node
))
4424 if (memcg_wb_domain_init(memcg
, GFP_KERNEL
))
4427 INIT_WORK(&memcg
->high_work
, high_work_func
);
4428 memcg
->last_scanned_node
= MAX_NUMNODES
;
4429 INIT_LIST_HEAD(&memcg
->oom_notify
);
4430 mutex_init(&memcg
->thresholds_lock
);
4431 spin_lock_init(&memcg
->move_lock
);
4432 vmpressure_init(&memcg
->vmpressure
);
4433 INIT_LIST_HEAD(&memcg
->event_list
);
4434 spin_lock_init(&memcg
->event_list_lock
);
4435 memcg
->socket_pressure
= jiffies
;
4436 #ifdef CONFIG_MEMCG_KMEM
4437 memcg
->kmemcg_id
= -1;
4439 #ifdef CONFIG_CGROUP_WRITEBACK
4440 INIT_LIST_HEAD(&memcg
->cgwb_list
);
4442 idr_replace(&mem_cgroup_idr
, memcg
, memcg
->id
.id
);
4445 mem_cgroup_id_remove(memcg
);
4446 __mem_cgroup_free(memcg
);
4450 static struct cgroup_subsys_state
* __ref
4451 mem_cgroup_css_alloc(struct cgroup_subsys_state
*parent_css
)
4453 struct mem_cgroup
*parent
= mem_cgroup_from_css(parent_css
);
4454 struct mem_cgroup
*memcg
;
4455 long error
= -ENOMEM
;
4457 memcg
= mem_cgroup_alloc();
4459 return ERR_PTR(error
);
4461 memcg
->high
= PAGE_COUNTER_MAX
;
4462 memcg
->soft_limit
= PAGE_COUNTER_MAX
;
4464 memcg
->swappiness
= mem_cgroup_swappiness(parent
);
4465 memcg
->oom_kill_disable
= parent
->oom_kill_disable
;
4467 if (parent
&& parent
->use_hierarchy
) {
4468 memcg
->use_hierarchy
= true;
4469 page_counter_init(&memcg
->memory
, &parent
->memory
);
4470 page_counter_init(&memcg
->swap
, &parent
->swap
);
4471 page_counter_init(&memcg
->memsw
, &parent
->memsw
);
4472 page_counter_init(&memcg
->kmem
, &parent
->kmem
);
4473 page_counter_init(&memcg
->tcpmem
, &parent
->tcpmem
);
4475 page_counter_init(&memcg
->memory
, NULL
);
4476 page_counter_init(&memcg
->swap
, NULL
);
4477 page_counter_init(&memcg
->memsw
, NULL
);
4478 page_counter_init(&memcg
->kmem
, NULL
);
4479 page_counter_init(&memcg
->tcpmem
, NULL
);
4481 * Deeper hierachy with use_hierarchy == false doesn't make
4482 * much sense so let cgroup subsystem know about this
4483 * unfortunate state in our controller.
4485 if (parent
!= root_mem_cgroup
)
4486 memory_cgrp_subsys
.broken_hierarchy
= true;
4489 /* The following stuff does not apply to the root */
4491 root_mem_cgroup
= memcg
;
4495 error
= memcg_online_kmem(memcg
);
4499 if (cgroup_subsys_on_dfl(memory_cgrp_subsys
) && !cgroup_memory_nosocket
)
4500 static_branch_inc(&memcg_sockets_enabled_key
);
4504 mem_cgroup_id_remove(memcg
);
4505 mem_cgroup_free(memcg
);
4506 return ERR_PTR(-ENOMEM
);
4509 static int mem_cgroup_css_online(struct cgroup_subsys_state
*css
)
4511 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
4514 * A memcg must be visible for memcg_expand_shrinker_maps()
4515 * by the time the maps are allocated. So, we allocate maps
4516 * here, when for_each_mem_cgroup() can't skip it.
4518 if (memcg_alloc_shrinker_maps(memcg
)) {
4519 mem_cgroup_id_remove(memcg
);
4523 /* Online state pins memcg ID, memcg ID pins CSS */
4524 refcount_set(&memcg
->id
.ref
, 1);
4529 static void mem_cgroup_css_offline(struct cgroup_subsys_state
*css
)
4531 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
4532 struct mem_cgroup_event
*event
, *tmp
;
4535 * Unregister events and notify userspace.
4536 * Notify userspace about cgroup removing only after rmdir of cgroup
4537 * directory to avoid race between userspace and kernelspace.
4539 spin_lock(&memcg
->event_list_lock
);
4540 list_for_each_entry_safe(event
, tmp
, &memcg
->event_list
, list
) {
4541 list_del_init(&event
->list
);
4542 schedule_work(&event
->remove
);
4544 spin_unlock(&memcg
->event_list_lock
);
4546 page_counter_set_min(&memcg
->memory
, 0);
4547 page_counter_set_low(&memcg
->memory
, 0);
4549 memcg_offline_kmem(memcg
);
4550 wb_memcg_offline(memcg
);
4552 drain_all_stock(memcg
);
4554 mem_cgroup_id_put(memcg
);
4557 static void mem_cgroup_css_released(struct cgroup_subsys_state
*css
)
4559 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
4561 invalidate_reclaim_iterators(memcg
);
4564 static void mem_cgroup_css_free(struct cgroup_subsys_state
*css
)
4566 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
4568 if (cgroup_subsys_on_dfl(memory_cgrp_subsys
) && !cgroup_memory_nosocket
)
4569 static_branch_dec(&memcg_sockets_enabled_key
);
4571 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys
) && memcg
->tcpmem_active
)
4572 static_branch_dec(&memcg_sockets_enabled_key
);
4574 vmpressure_cleanup(&memcg
->vmpressure
);
4575 cancel_work_sync(&memcg
->high_work
);
4576 mem_cgroup_remove_from_trees(memcg
);
4577 memcg_free_shrinker_maps(memcg
);
4578 memcg_free_kmem(memcg
);
4579 mem_cgroup_free(memcg
);
4583 * mem_cgroup_css_reset - reset the states of a mem_cgroup
4584 * @css: the target css
4586 * Reset the states of the mem_cgroup associated with @css. This is
4587 * invoked when the userland requests disabling on the default hierarchy
4588 * but the memcg is pinned through dependency. The memcg should stop
4589 * applying policies and should revert to the vanilla state as it may be
4590 * made visible again.
4592 * The current implementation only resets the essential configurations.
4593 * This needs to be expanded to cover all the visible parts.
4595 static void mem_cgroup_css_reset(struct cgroup_subsys_state
*css
)
4597 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
4599 page_counter_set_max(&memcg
->memory
, PAGE_COUNTER_MAX
);
4600 page_counter_set_max(&memcg
->swap
, PAGE_COUNTER_MAX
);
4601 page_counter_set_max(&memcg
->memsw
, PAGE_COUNTER_MAX
);
4602 page_counter_set_max(&memcg
->kmem
, PAGE_COUNTER_MAX
);
4603 page_counter_set_max(&memcg
->tcpmem
, PAGE_COUNTER_MAX
);
4604 page_counter_set_min(&memcg
->memory
, 0);
4605 page_counter_set_low(&memcg
->memory
, 0);
4606 memcg
->high
= PAGE_COUNTER_MAX
;
4607 memcg
->soft_limit
= PAGE_COUNTER_MAX
;
4608 memcg_wb_domain_size_changed(memcg
);
4612 /* Handlers for move charge at task migration. */
4613 static int mem_cgroup_do_precharge(unsigned long count
)
4617 /* Try a single bulk charge without reclaim first, kswapd may wake */
4618 ret
= try_charge(mc
.to
, GFP_KERNEL
& ~__GFP_DIRECT_RECLAIM
, count
);
4620 mc
.precharge
+= count
;
4624 /* Try charges one by one with reclaim, but do not retry */
4626 ret
= try_charge(mc
.to
, GFP_KERNEL
| __GFP_NORETRY
, 1);
4640 enum mc_target_type
{
4647 static struct page
*mc_handle_present_pte(struct vm_area_struct
*vma
,
4648 unsigned long addr
, pte_t ptent
)
4650 struct page
*page
= _vm_normal_page(vma
, addr
, ptent
, true);
4652 if (!page
|| !page_mapped(page
))
4654 if (PageAnon(page
)) {
4655 if (!(mc
.flags
& MOVE_ANON
))
4658 if (!(mc
.flags
& MOVE_FILE
))
4661 if (!get_page_unless_zero(page
))
4667 #if defined(CONFIG_SWAP) || defined(CONFIG_DEVICE_PRIVATE)
4668 static struct page
*mc_handle_swap_pte(struct vm_area_struct
*vma
,
4669 pte_t ptent
, swp_entry_t
*entry
)
4671 struct page
*page
= NULL
;
4672 swp_entry_t ent
= pte_to_swp_entry(ptent
);
4674 if (!(mc
.flags
& MOVE_ANON
) || non_swap_entry(ent
))
4678 * Handle MEMORY_DEVICE_PRIVATE which are ZONE_DEVICE page belonging to
4679 * a device and because they are not accessible by CPU they are store
4680 * as special swap entry in the CPU page table.
4682 if (is_device_private_entry(ent
)) {
4683 page
= device_private_entry_to_page(ent
);
4685 * MEMORY_DEVICE_PRIVATE means ZONE_DEVICE page and which have
4686 * a refcount of 1 when free (unlike normal page)
4688 if (!page_ref_add_unless(page
, 1, 1))
4694 * Because lookup_swap_cache() updates some statistics counter,
4695 * we call find_get_page() with swapper_space directly.
4697 page
= find_get_page(swap_address_space(ent
), swp_offset(ent
));
4698 if (do_memsw_account())
4699 entry
->val
= ent
.val
;
4704 static struct page
*mc_handle_swap_pte(struct vm_area_struct
*vma
,
4705 pte_t ptent
, swp_entry_t
*entry
)
4711 static struct page
*mc_handle_file_pte(struct vm_area_struct
*vma
,
4712 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
4714 struct page
*page
= NULL
;
4715 struct address_space
*mapping
;
4718 if (!vma
->vm_file
) /* anonymous vma */
4720 if (!(mc
.flags
& MOVE_FILE
))
4723 mapping
= vma
->vm_file
->f_mapping
;
4724 pgoff
= linear_page_index(vma
, addr
);
4726 /* page is moved even if it's not RSS of this task(page-faulted). */
4728 /* shmem/tmpfs may report page out on swap: account for that too. */
4729 if (shmem_mapping(mapping
)) {
4730 page
= find_get_entry(mapping
, pgoff
);
4731 if (radix_tree_exceptional_entry(page
)) {
4732 swp_entry_t swp
= radix_to_swp_entry(page
);
4733 if (do_memsw_account())
4735 page
= find_get_page(swap_address_space(swp
),
4739 page
= find_get_page(mapping
, pgoff
);
4741 page
= find_get_page(mapping
, pgoff
);
4747 * mem_cgroup_move_account - move account of the page
4749 * @compound: charge the page as compound or small page
4750 * @from: mem_cgroup which the page is moved from.
4751 * @to: mem_cgroup which the page is moved to. @from != @to.
4753 * The caller must make sure the page is not on LRU (isolate_page() is useful.)
4755 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
4758 static int mem_cgroup_move_account(struct page
*page
,
4760 struct mem_cgroup
*from
,
4761 struct mem_cgroup
*to
)
4763 unsigned long flags
;
4764 unsigned int nr_pages
= compound
? hpage_nr_pages(page
) : 1;
4768 VM_BUG_ON(from
== to
);
4769 VM_BUG_ON_PAGE(PageLRU(page
), page
);
4770 VM_BUG_ON(compound
&& !PageTransHuge(page
));
4773 * Prevent mem_cgroup_migrate() from looking at
4774 * page->mem_cgroup of its source page while we change it.
4777 if (!trylock_page(page
))
4781 if (page
->mem_cgroup
!= from
)
4784 anon
= PageAnon(page
);
4786 spin_lock_irqsave(&from
->move_lock
, flags
);
4788 if (!anon
&& page_mapped(page
)) {
4789 __mod_memcg_state(from
, NR_FILE_MAPPED
, -nr_pages
);
4790 __mod_memcg_state(to
, NR_FILE_MAPPED
, nr_pages
);
4794 * move_lock grabbed above and caller set from->moving_account, so
4795 * mod_memcg_page_state will serialize updates to PageDirty.
4796 * So mapping should be stable for dirty pages.
4798 if (!anon
&& PageDirty(page
)) {
4799 struct address_space
*mapping
= page_mapping(page
);
4801 if (mapping_cap_account_dirty(mapping
)) {
4802 __mod_memcg_state(from
, NR_FILE_DIRTY
, -nr_pages
);
4803 __mod_memcg_state(to
, NR_FILE_DIRTY
, nr_pages
);
4807 if (PageWriteback(page
)) {
4808 __mod_memcg_state(from
, NR_WRITEBACK
, -nr_pages
);
4809 __mod_memcg_state(to
, NR_WRITEBACK
, nr_pages
);
4813 * It is safe to change page->mem_cgroup here because the page
4814 * is referenced, charged, and isolated - we can't race with
4815 * uncharging, charging, migration, or LRU putback.
4818 /* caller should have done css_get */
4819 page
->mem_cgroup
= to
;
4820 spin_unlock_irqrestore(&from
->move_lock
, flags
);
4824 local_irq_disable();
4825 mem_cgroup_charge_statistics(to
, page
, compound
, nr_pages
);
4826 memcg_check_events(to
, page
);
4827 mem_cgroup_charge_statistics(from
, page
, compound
, -nr_pages
);
4828 memcg_check_events(from
, page
);
4837 * get_mctgt_type - get target type of moving charge
4838 * @vma: the vma the pte to be checked belongs
4839 * @addr: the address corresponding to the pte to be checked
4840 * @ptent: the pte to be checked
4841 * @target: the pointer the target page or swap ent will be stored(can be NULL)
4844 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
4845 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
4846 * move charge. if @target is not NULL, the page is stored in target->page
4847 * with extra refcnt got(Callers should handle it).
4848 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
4849 * target for charge migration. if @target is not NULL, the entry is stored
4851 * 3(MC_TARGET_DEVICE): like MC_TARGET_PAGE but page is MEMORY_DEVICE_PUBLIC
4852 * or MEMORY_DEVICE_PRIVATE (so ZONE_DEVICE page and thus not on the lru).
4853 * For now we such page is charge like a regular page would be as for all
4854 * intent and purposes it is just special memory taking the place of a
4857 * See Documentations/vm/hmm.txt and include/linux/hmm.h
4859 * Called with pte lock held.
4862 static enum mc_target_type
get_mctgt_type(struct vm_area_struct
*vma
,
4863 unsigned long addr
, pte_t ptent
, union mc_target
*target
)
4865 struct page
*page
= NULL
;
4866 enum mc_target_type ret
= MC_TARGET_NONE
;
4867 swp_entry_t ent
= { .val
= 0 };
4869 if (pte_present(ptent
))
4870 page
= mc_handle_present_pte(vma
, addr
, ptent
);
4871 else if (is_swap_pte(ptent
))
4872 page
= mc_handle_swap_pte(vma
, ptent
, &ent
);
4873 else if (pte_none(ptent
))
4874 page
= mc_handle_file_pte(vma
, addr
, ptent
, &ent
);
4876 if (!page
&& !ent
.val
)
4880 * Do only loose check w/o serialization.
4881 * mem_cgroup_move_account() checks the page is valid or
4882 * not under LRU exclusion.
4884 if (page
->mem_cgroup
== mc
.from
) {
4885 ret
= MC_TARGET_PAGE
;
4886 if (is_device_private_page(page
) ||
4887 is_device_public_page(page
))
4888 ret
= MC_TARGET_DEVICE
;
4890 target
->page
= page
;
4892 if (!ret
|| !target
)
4896 * There is a swap entry and a page doesn't exist or isn't charged.
4897 * But we cannot move a tail-page in a THP.
4899 if (ent
.val
&& !ret
&& (!page
|| !PageTransCompound(page
)) &&
4900 mem_cgroup_id(mc
.from
) == lookup_swap_cgroup_id(ent
)) {
4901 ret
= MC_TARGET_SWAP
;
4908 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4910 * We don't consider PMD mapped swapping or file mapped pages because THP does
4911 * not support them for now.
4912 * Caller should make sure that pmd_trans_huge(pmd) is true.
4914 static enum mc_target_type
get_mctgt_type_thp(struct vm_area_struct
*vma
,
4915 unsigned long addr
, pmd_t pmd
, union mc_target
*target
)
4917 struct page
*page
= NULL
;
4918 enum mc_target_type ret
= MC_TARGET_NONE
;
4920 if (unlikely(is_swap_pmd(pmd
))) {
4921 VM_BUG_ON(thp_migration_supported() &&
4922 !is_pmd_migration_entry(pmd
));
4925 page
= pmd_page(pmd
);
4926 VM_BUG_ON_PAGE(!page
|| !PageHead(page
), page
);
4927 if (!(mc
.flags
& MOVE_ANON
))
4929 if (page
->mem_cgroup
== mc
.from
) {
4930 ret
= MC_TARGET_PAGE
;
4933 target
->page
= page
;
4939 static inline enum mc_target_type
get_mctgt_type_thp(struct vm_area_struct
*vma
,
4940 unsigned long addr
, pmd_t pmd
, union mc_target
*target
)
4942 return MC_TARGET_NONE
;
4946 static int mem_cgroup_count_precharge_pte_range(pmd_t
*pmd
,
4947 unsigned long addr
, unsigned long end
,
4948 struct mm_walk
*walk
)
4950 struct vm_area_struct
*vma
= walk
->vma
;
4954 ptl
= pmd_trans_huge_lock(pmd
, vma
);
4957 * Note their can not be MC_TARGET_DEVICE for now as we do not
4958 * support transparent huge page with MEMORY_DEVICE_PUBLIC or
4959 * MEMORY_DEVICE_PRIVATE but this might change.
4961 if (get_mctgt_type_thp(vma
, addr
, *pmd
, NULL
) == MC_TARGET_PAGE
)
4962 mc
.precharge
+= HPAGE_PMD_NR
;
4967 if (pmd_trans_unstable(pmd
))
4969 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
4970 for (; addr
!= end
; pte
++, addr
+= PAGE_SIZE
)
4971 if (get_mctgt_type(vma
, addr
, *pte
, NULL
))
4972 mc
.precharge
++; /* increment precharge temporarily */
4973 pte_unmap_unlock(pte
- 1, ptl
);
4979 static unsigned long mem_cgroup_count_precharge(struct mm_struct
*mm
)
4981 unsigned long precharge
;
4983 struct mm_walk mem_cgroup_count_precharge_walk
= {
4984 .pmd_entry
= mem_cgroup_count_precharge_pte_range
,
4987 down_read(&mm
->mmap_sem
);
4988 walk_page_range(0, mm
->highest_vm_end
,
4989 &mem_cgroup_count_precharge_walk
);
4990 up_read(&mm
->mmap_sem
);
4992 precharge
= mc
.precharge
;
4998 static int mem_cgroup_precharge_mc(struct mm_struct
*mm
)
5000 unsigned long precharge
= mem_cgroup_count_precharge(mm
);
5002 VM_BUG_ON(mc
.moving_task
);
5003 mc
.moving_task
= current
;
5004 return mem_cgroup_do_precharge(precharge
);
5007 /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
5008 static void __mem_cgroup_clear_mc(void)
5010 struct mem_cgroup
*from
= mc
.from
;
5011 struct mem_cgroup
*to
= mc
.to
;
5013 /* we must uncharge all the leftover precharges from mc.to */
5015 cancel_charge(mc
.to
, mc
.precharge
);
5019 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
5020 * we must uncharge here.
5022 if (mc
.moved_charge
) {
5023 cancel_charge(mc
.from
, mc
.moved_charge
);
5024 mc
.moved_charge
= 0;
5026 /* we must fixup refcnts and charges */
5027 if (mc
.moved_swap
) {
5028 /* uncharge swap account from the old cgroup */
5029 if (!mem_cgroup_is_root(mc
.from
))
5030 page_counter_uncharge(&mc
.from
->memsw
, mc
.moved_swap
);
5032 mem_cgroup_id_put_many(mc
.from
, mc
.moved_swap
);
5035 * we charged both to->memory and to->memsw, so we
5036 * should uncharge to->memory.
5038 if (!mem_cgroup_is_root(mc
.to
))
5039 page_counter_uncharge(&mc
.to
->memory
, mc
.moved_swap
);
5041 mem_cgroup_id_get_many(mc
.to
, mc
.moved_swap
);
5042 css_put_many(&mc
.to
->css
, mc
.moved_swap
);
5046 memcg_oom_recover(from
);
5047 memcg_oom_recover(to
);
5048 wake_up_all(&mc
.waitq
);
5051 static void mem_cgroup_clear_mc(void)
5053 struct mm_struct
*mm
= mc
.mm
;
5056 * we must clear moving_task before waking up waiters at the end of
5059 mc
.moving_task
= NULL
;
5060 __mem_cgroup_clear_mc();
5061 spin_lock(&mc
.lock
);
5065 spin_unlock(&mc
.lock
);
5070 static int mem_cgroup_can_attach(struct cgroup_taskset
*tset
)
5072 struct cgroup_subsys_state
*css
;
5073 struct mem_cgroup
*memcg
= NULL
; /* unneeded init to make gcc happy */
5074 struct mem_cgroup
*from
;
5075 struct task_struct
*leader
, *p
;
5076 struct mm_struct
*mm
;
5077 unsigned long move_flags
;
5080 /* charge immigration isn't supported on the default hierarchy */
5081 if (cgroup_subsys_on_dfl(memory_cgrp_subsys
))
5085 * Multi-process migrations only happen on the default hierarchy
5086 * where charge immigration is not used. Perform charge
5087 * immigration if @tset contains a leader and whine if there are
5091 cgroup_taskset_for_each_leader(leader
, css
, tset
) {
5094 memcg
= mem_cgroup_from_css(css
);
5100 * We are now commited to this value whatever it is. Changes in this
5101 * tunable will only affect upcoming migrations, not the current one.
5102 * So we need to save it, and keep it going.
5104 move_flags
= READ_ONCE(memcg
->move_charge_at_immigrate
);
5108 from
= mem_cgroup_from_task(p
);
5110 VM_BUG_ON(from
== memcg
);
5112 mm
= get_task_mm(p
);
5115 /* We move charges only when we move a owner of the mm */
5116 if (mm
->owner
== p
) {
5119 VM_BUG_ON(mc
.precharge
);
5120 VM_BUG_ON(mc
.moved_charge
);
5121 VM_BUG_ON(mc
.moved_swap
);
5123 spin_lock(&mc
.lock
);
5127 mc
.flags
= move_flags
;
5128 spin_unlock(&mc
.lock
);
5129 /* We set mc.moving_task later */
5131 ret
= mem_cgroup_precharge_mc(mm
);
5133 mem_cgroup_clear_mc();
5140 static void mem_cgroup_cancel_attach(struct cgroup_taskset
*tset
)
5143 mem_cgroup_clear_mc();
5146 static int mem_cgroup_move_charge_pte_range(pmd_t
*pmd
,
5147 unsigned long addr
, unsigned long end
,
5148 struct mm_walk
*walk
)
5151 struct vm_area_struct
*vma
= walk
->vma
;
5154 enum mc_target_type target_type
;
5155 union mc_target target
;
5158 ptl
= pmd_trans_huge_lock(pmd
, vma
);
5160 if (mc
.precharge
< HPAGE_PMD_NR
) {
5164 target_type
= get_mctgt_type_thp(vma
, addr
, *pmd
, &target
);
5165 if (target_type
== MC_TARGET_PAGE
) {
5167 if (!isolate_lru_page(page
)) {
5168 if (!mem_cgroup_move_account(page
, true,
5170 mc
.precharge
-= HPAGE_PMD_NR
;
5171 mc
.moved_charge
+= HPAGE_PMD_NR
;
5173 putback_lru_page(page
);
5176 } else if (target_type
== MC_TARGET_DEVICE
) {
5178 if (!mem_cgroup_move_account(page
, true,
5180 mc
.precharge
-= HPAGE_PMD_NR
;
5181 mc
.moved_charge
+= HPAGE_PMD_NR
;
5189 if (pmd_trans_unstable(pmd
))
5192 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
5193 for (; addr
!= end
; addr
+= PAGE_SIZE
) {
5194 pte_t ptent
= *(pte
++);
5195 bool device
= false;
5201 switch (get_mctgt_type(vma
, addr
, ptent
, &target
)) {
5202 case MC_TARGET_DEVICE
:
5205 case MC_TARGET_PAGE
:
5208 * We can have a part of the split pmd here. Moving it
5209 * can be done but it would be too convoluted so simply
5210 * ignore such a partial THP and keep it in original
5211 * memcg. There should be somebody mapping the head.
5213 if (PageTransCompound(page
))
5215 if (!device
&& isolate_lru_page(page
))
5217 if (!mem_cgroup_move_account(page
, false,
5220 /* we uncharge from mc.from later. */
5224 putback_lru_page(page
);
5225 put
: /* get_mctgt_type() gets the page */
5228 case MC_TARGET_SWAP
:
5230 if (!mem_cgroup_move_swap_account(ent
, mc
.from
, mc
.to
)) {
5232 /* we fixup refcnts and charges later. */
5240 pte_unmap_unlock(pte
- 1, ptl
);
5245 * We have consumed all precharges we got in can_attach().
5246 * We try charge one by one, but don't do any additional
5247 * charges to mc.to if we have failed in charge once in attach()
5250 ret
= mem_cgroup_do_precharge(1);
5258 static void mem_cgroup_move_charge(void)
5260 struct mm_walk mem_cgroup_move_charge_walk
= {
5261 .pmd_entry
= mem_cgroup_move_charge_pte_range
,
5265 lru_add_drain_all();
5267 * Signal lock_page_memcg() to take the memcg's move_lock
5268 * while we're moving its pages to another memcg. Then wait
5269 * for already started RCU-only updates to finish.
5271 atomic_inc(&mc
.from
->moving_account
);
5274 if (unlikely(!down_read_trylock(&mc
.mm
->mmap_sem
))) {
5276 * Someone who are holding the mmap_sem might be waiting in
5277 * waitq. So we cancel all extra charges, wake up all waiters,
5278 * and retry. Because we cancel precharges, we might not be able
5279 * to move enough charges, but moving charge is a best-effort
5280 * feature anyway, so it wouldn't be a big problem.
5282 __mem_cgroup_clear_mc();
5287 * When we have consumed all precharges and failed in doing
5288 * additional charge, the page walk just aborts.
5290 walk_page_range(0, mc
.mm
->highest_vm_end
, &mem_cgroup_move_charge_walk
);
5292 up_read(&mc
.mm
->mmap_sem
);
5293 atomic_dec(&mc
.from
->moving_account
);
5296 static void mem_cgroup_move_task(void)
5299 mem_cgroup_move_charge();
5300 mem_cgroup_clear_mc();
5303 #else /* !CONFIG_MMU */
5304 static int mem_cgroup_can_attach(struct cgroup_taskset
*tset
)
5308 static void mem_cgroup_cancel_attach(struct cgroup_taskset
*tset
)
5311 static void mem_cgroup_move_task(void)
5317 * Cgroup retains root cgroups across [un]mount cycles making it necessary
5318 * to verify whether we're attached to the default hierarchy on each mount
5321 static void mem_cgroup_bind(struct cgroup_subsys_state
*root_css
)
5324 * use_hierarchy is forced on the default hierarchy. cgroup core
5325 * guarantees that @root doesn't have any children, so turning it
5326 * on for the root memcg is enough.
5328 if (cgroup_subsys_on_dfl(memory_cgrp_subsys
))
5329 root_mem_cgroup
->use_hierarchy
= true;
5331 root_mem_cgroup
->use_hierarchy
= false;
5334 static u64
memory_current_read(struct cgroup_subsys_state
*css
,
5337 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
5339 return (u64
)page_counter_read(&memcg
->memory
) * PAGE_SIZE
;
5342 static int memory_min_show(struct seq_file
*m
, void *v
)
5344 struct mem_cgroup
*memcg
= mem_cgroup_from_css(seq_css(m
));
5345 unsigned long min
= READ_ONCE(memcg
->memory
.min
);
5347 if (min
== PAGE_COUNTER_MAX
)
5348 seq_puts(m
, "max\n");
5350 seq_printf(m
, "%llu\n", (u64
)min
* PAGE_SIZE
);
5355 static ssize_t
memory_min_write(struct kernfs_open_file
*of
,
5356 char *buf
, size_t nbytes
, loff_t off
)
5358 struct mem_cgroup
*memcg
= mem_cgroup_from_css(of_css(of
));
5362 buf
= strstrip(buf
);
5363 err
= page_counter_memparse(buf
, "max", &min
);
5367 page_counter_set_min(&memcg
->memory
, min
);
5372 static int memory_low_show(struct seq_file
*m
, void *v
)
5374 struct mem_cgroup
*memcg
= mem_cgroup_from_css(seq_css(m
));
5375 unsigned long low
= READ_ONCE(memcg
->memory
.low
);
5377 if (low
== PAGE_COUNTER_MAX
)
5378 seq_puts(m
, "max\n");
5380 seq_printf(m
, "%llu\n", (u64
)low
* PAGE_SIZE
);
5385 static ssize_t
memory_low_write(struct kernfs_open_file
*of
,
5386 char *buf
, size_t nbytes
, loff_t off
)
5388 struct mem_cgroup
*memcg
= mem_cgroup_from_css(of_css(of
));
5392 buf
= strstrip(buf
);
5393 err
= page_counter_memparse(buf
, "max", &low
);
5397 page_counter_set_low(&memcg
->memory
, low
);
5402 static int memory_high_show(struct seq_file
*m
, void *v
)
5404 struct mem_cgroup
*memcg
= mem_cgroup_from_css(seq_css(m
));
5405 unsigned long high
= READ_ONCE(memcg
->high
);
5407 if (high
== PAGE_COUNTER_MAX
)
5408 seq_puts(m
, "max\n");
5410 seq_printf(m
, "%llu\n", (u64
)high
* PAGE_SIZE
);
5415 static ssize_t
memory_high_write(struct kernfs_open_file
*of
,
5416 char *buf
, size_t nbytes
, loff_t off
)
5418 struct mem_cgroup
*memcg
= mem_cgroup_from_css(of_css(of
));
5419 unsigned long nr_pages
;
5423 buf
= strstrip(buf
);
5424 err
= page_counter_memparse(buf
, "max", &high
);
5430 nr_pages
= page_counter_read(&memcg
->memory
);
5431 if (nr_pages
> high
)
5432 try_to_free_mem_cgroup_pages(memcg
, nr_pages
- high
,
5435 memcg_wb_domain_size_changed(memcg
);
5439 static int memory_max_show(struct seq_file
*m
, void *v
)
5441 struct mem_cgroup
*memcg
= mem_cgroup_from_css(seq_css(m
));
5442 unsigned long max
= READ_ONCE(memcg
->memory
.max
);
5444 if (max
== PAGE_COUNTER_MAX
)
5445 seq_puts(m
, "max\n");
5447 seq_printf(m
, "%llu\n", (u64
)max
* PAGE_SIZE
);
5452 static ssize_t
memory_max_write(struct kernfs_open_file
*of
,
5453 char *buf
, size_t nbytes
, loff_t off
)
5455 struct mem_cgroup
*memcg
= mem_cgroup_from_css(of_css(of
));
5456 unsigned int nr_reclaims
= MEM_CGROUP_RECLAIM_RETRIES
;
5457 bool drained
= false;
5461 buf
= strstrip(buf
);
5462 err
= page_counter_memparse(buf
, "max", &max
);
5466 xchg(&memcg
->memory
.max
, max
);
5469 unsigned long nr_pages
= page_counter_read(&memcg
->memory
);
5471 if (nr_pages
<= max
)
5474 if (signal_pending(current
)) {
5480 drain_all_stock(memcg
);
5486 if (!try_to_free_mem_cgroup_pages(memcg
, nr_pages
- max
,
5492 memcg_memory_event(memcg
, MEMCG_OOM
);
5493 if (!mem_cgroup_out_of_memory(memcg
, GFP_KERNEL
, 0))
5497 memcg_wb_domain_size_changed(memcg
);
5501 static int memory_events_show(struct seq_file
*m
, void *v
)
5503 struct mem_cgroup
*memcg
= mem_cgroup_from_css(seq_css(m
));
5505 seq_printf(m
, "low %lu\n",
5506 atomic_long_read(&memcg
->memory_events
[MEMCG_LOW
]));
5507 seq_printf(m
, "high %lu\n",
5508 atomic_long_read(&memcg
->memory_events
[MEMCG_HIGH
]));
5509 seq_printf(m
, "max %lu\n",
5510 atomic_long_read(&memcg
->memory_events
[MEMCG_MAX
]));
5511 seq_printf(m
, "oom %lu\n",
5512 atomic_long_read(&memcg
->memory_events
[MEMCG_OOM
]));
5513 seq_printf(m
, "oom_kill %lu\n",
5514 atomic_long_read(&memcg
->memory_events
[MEMCG_OOM_KILL
]));
5519 static int memory_stat_show(struct seq_file
*m
, void *v
)
5521 struct mem_cgroup
*memcg
= mem_cgroup_from_css(seq_css(m
));
5522 struct accumulated_stats acc
;
5526 * Provide statistics on the state of the memory subsystem as
5527 * well as cumulative event counters that show past behavior.
5529 * This list is ordered following a combination of these gradients:
5530 * 1) generic big picture -> specifics and details
5531 * 2) reflecting userspace activity -> reflecting kernel heuristics
5533 * Current memory state:
5536 memset(&acc
, 0, sizeof(acc
));
5537 acc
.stats_size
= MEMCG_NR_STAT
;
5538 acc
.events_size
= NR_VM_EVENT_ITEMS
;
5539 accumulate_memcg_tree(memcg
, &acc
);
5541 seq_printf(m
, "anon %llu\n",
5542 (u64
)acc
.stat
[MEMCG_RSS
] * PAGE_SIZE
);
5543 seq_printf(m
, "file %llu\n",
5544 (u64
)acc
.stat
[MEMCG_CACHE
] * PAGE_SIZE
);
5545 seq_printf(m
, "kernel_stack %llu\n",
5546 (u64
)acc
.stat
[MEMCG_KERNEL_STACK_KB
] * 1024);
5547 seq_printf(m
, "slab %llu\n",
5548 (u64
)(acc
.stat
[NR_SLAB_RECLAIMABLE
] +
5549 acc
.stat
[NR_SLAB_UNRECLAIMABLE
]) * PAGE_SIZE
);
5550 seq_printf(m
, "sock %llu\n",
5551 (u64
)acc
.stat
[MEMCG_SOCK
] * PAGE_SIZE
);
5553 seq_printf(m
, "shmem %llu\n",
5554 (u64
)acc
.stat
[NR_SHMEM
] * PAGE_SIZE
);
5555 seq_printf(m
, "file_mapped %llu\n",
5556 (u64
)acc
.stat
[NR_FILE_MAPPED
] * PAGE_SIZE
);
5557 seq_printf(m
, "file_dirty %llu\n",
5558 (u64
)acc
.stat
[NR_FILE_DIRTY
] * PAGE_SIZE
);
5559 seq_printf(m
, "file_writeback %llu\n",
5560 (u64
)acc
.stat
[NR_WRITEBACK
] * PAGE_SIZE
);
5562 for (i
= 0; i
< NR_LRU_LISTS
; i
++)
5563 seq_printf(m
, "%s %llu\n", mem_cgroup_lru_names
[i
],
5564 (u64
)acc
.lru_pages
[i
] * PAGE_SIZE
);
5566 seq_printf(m
, "slab_reclaimable %llu\n",
5567 (u64
)acc
.stat
[NR_SLAB_RECLAIMABLE
] * PAGE_SIZE
);
5568 seq_printf(m
, "slab_unreclaimable %llu\n",
5569 (u64
)acc
.stat
[NR_SLAB_UNRECLAIMABLE
] * PAGE_SIZE
);
5571 /* Accumulated memory events */
5573 seq_printf(m
, "pgfault %lu\n", acc
.events
[PGFAULT
]);
5574 seq_printf(m
, "pgmajfault %lu\n", acc
.events
[PGMAJFAULT
]);
5576 seq_printf(m
, "workingset_refault %lu\n",
5577 acc
.stat
[WORKINGSET_REFAULT
]);
5578 seq_printf(m
, "workingset_activate %lu\n",
5579 acc
.stat
[WORKINGSET_ACTIVATE
]);
5580 seq_printf(m
, "workingset_nodereclaim %lu\n",
5581 acc
.stat
[WORKINGSET_NODERECLAIM
]);
5583 seq_printf(m
, "pgrefill %lu\n", acc
.events
[PGREFILL
]);
5584 seq_printf(m
, "pgscan %lu\n", acc
.events
[PGSCAN_KSWAPD
] +
5585 acc
.events
[PGSCAN_DIRECT
]);
5586 seq_printf(m
, "pgsteal %lu\n", acc
.events
[PGSTEAL_KSWAPD
] +
5587 acc
.events
[PGSTEAL_DIRECT
]);
5588 seq_printf(m
, "pgactivate %lu\n", acc
.events
[PGACTIVATE
]);
5589 seq_printf(m
, "pgdeactivate %lu\n", acc
.events
[PGDEACTIVATE
]);
5590 seq_printf(m
, "pglazyfree %lu\n", acc
.events
[PGLAZYFREE
]);
5591 seq_printf(m
, "pglazyfreed %lu\n", acc
.events
[PGLAZYFREED
]);
5596 static int memory_oom_group_show(struct seq_file
*m
, void *v
)
5598 struct mem_cgroup
*memcg
= mem_cgroup_from_css(seq_css(m
));
5600 seq_printf(m
, "%d\n", memcg
->oom_group
);
5605 static ssize_t
memory_oom_group_write(struct kernfs_open_file
*of
,
5606 char *buf
, size_t nbytes
, loff_t off
)
5608 struct mem_cgroup
*memcg
= mem_cgroup_from_css(of_css(of
));
5611 buf
= strstrip(buf
);
5615 ret
= kstrtoint(buf
, 0, &oom_group
);
5619 if (oom_group
!= 0 && oom_group
!= 1)
5622 memcg
->oom_group
= oom_group
;
5627 static struct cftype memory_files
[] = {
5630 .flags
= CFTYPE_NOT_ON_ROOT
,
5631 .read_u64
= memory_current_read
,
5635 .flags
= CFTYPE_NOT_ON_ROOT
,
5636 .seq_show
= memory_min_show
,
5637 .write
= memory_min_write
,
5641 .flags
= CFTYPE_NOT_ON_ROOT
,
5642 .seq_show
= memory_low_show
,
5643 .write
= memory_low_write
,
5647 .flags
= CFTYPE_NOT_ON_ROOT
,
5648 .seq_show
= memory_high_show
,
5649 .write
= memory_high_write
,
5653 .flags
= CFTYPE_NOT_ON_ROOT
,
5654 .seq_show
= memory_max_show
,
5655 .write
= memory_max_write
,
5659 .flags
= CFTYPE_NOT_ON_ROOT
,
5660 .file_offset
= offsetof(struct mem_cgroup
, events_file
),
5661 .seq_show
= memory_events_show
,
5665 .flags
= CFTYPE_NOT_ON_ROOT
,
5666 .seq_show
= memory_stat_show
,
5669 .name
= "oom.group",
5670 .flags
= CFTYPE_NOT_ON_ROOT
| CFTYPE_NS_DELEGATABLE
,
5671 .seq_show
= memory_oom_group_show
,
5672 .write
= memory_oom_group_write
,
5677 struct cgroup_subsys memory_cgrp_subsys
= {
5678 .css_alloc
= mem_cgroup_css_alloc
,
5679 .css_online
= mem_cgroup_css_online
,
5680 .css_offline
= mem_cgroup_css_offline
,
5681 .css_released
= mem_cgroup_css_released
,
5682 .css_free
= mem_cgroup_css_free
,
5683 .css_reset
= mem_cgroup_css_reset
,
5684 .can_attach
= mem_cgroup_can_attach
,
5685 .cancel_attach
= mem_cgroup_cancel_attach
,
5686 .post_attach
= mem_cgroup_move_task
,
5687 .bind
= mem_cgroup_bind
,
5688 .dfl_cftypes
= memory_files
,
5689 .legacy_cftypes
= mem_cgroup_legacy_files
,
5694 * mem_cgroup_protected - check if memory consumption is in the normal range
5695 * @root: the top ancestor of the sub-tree being checked
5696 * @memcg: the memory cgroup to check
5698 * WARNING: This function is not stateless! It can only be used as part
5699 * of a top-down tree iteration, not for isolated queries.
5701 * Returns one of the following:
5702 * MEMCG_PROT_NONE: cgroup memory is not protected
5703 * MEMCG_PROT_LOW: cgroup memory is protected as long there is
5704 * an unprotected supply of reclaimable memory from other cgroups.
5705 * MEMCG_PROT_MIN: cgroup memory is protected
5707 * @root is exclusive; it is never protected when looked at directly
5709 * To provide a proper hierarchical behavior, effective memory.min/low values
5710 * are used. Below is the description of how effective memory.low is calculated.
5711 * Effective memory.min values is calculated in the same way.
5713 * Effective memory.low is always equal or less than the original memory.low.
5714 * If there is no memory.low overcommittment (which is always true for
5715 * top-level memory cgroups), these two values are equal.
5716 * Otherwise, it's a part of parent's effective memory.low,
5717 * calculated as a cgroup's memory.low usage divided by sum of sibling's
5718 * memory.low usages, where memory.low usage is the size of actually
5722 * elow = min( memory.low, parent->elow * ------------------ ),
5723 * siblings_low_usage
5725 * | memory.current, if memory.current < memory.low
5730 * Such definition of the effective memory.low provides the expected
5731 * hierarchical behavior: parent's memory.low value is limiting
5732 * children, unprotected memory is reclaimed first and cgroups,
5733 * which are not using their guarantee do not affect actual memory
5736 * For example, if there are memcgs A, A/B, A/C, A/D and A/E:
5738 * A A/memory.low = 2G, A/memory.current = 6G
5740 * BC DE B/memory.low = 3G B/memory.current = 2G
5741 * C/memory.low = 1G C/memory.current = 2G
5742 * D/memory.low = 0 D/memory.current = 2G
5743 * E/memory.low = 10G E/memory.current = 0
5745 * and the memory pressure is applied, the following memory distribution
5746 * is expected (approximately):
5748 * A/memory.current = 2G
5750 * B/memory.current = 1.3G
5751 * C/memory.current = 0.6G
5752 * D/memory.current = 0
5753 * E/memory.current = 0
5755 * These calculations require constant tracking of the actual low usages
5756 * (see propagate_protected_usage()), as well as recursive calculation of
5757 * effective memory.low values. But as we do call mem_cgroup_protected()
5758 * path for each memory cgroup top-down from the reclaim,
5759 * it's possible to optimize this part, and save calculated elow
5760 * for next usage. This part is intentionally racy, but it's ok,
5761 * as memory.low is a best-effort mechanism.
5763 enum mem_cgroup_protection
mem_cgroup_protected(struct mem_cgroup
*root
,
5764 struct mem_cgroup
*memcg
)
5766 struct mem_cgroup
*parent
;
5767 unsigned long emin
, parent_emin
;
5768 unsigned long elow
, parent_elow
;
5769 unsigned long usage
;
5771 if (mem_cgroup_disabled())
5772 return MEMCG_PROT_NONE
;
5775 root
= root_mem_cgroup
;
5777 return MEMCG_PROT_NONE
;
5779 usage
= page_counter_read(&memcg
->memory
);
5781 return MEMCG_PROT_NONE
;
5783 emin
= memcg
->memory
.min
;
5784 elow
= memcg
->memory
.low
;
5786 parent
= parent_mem_cgroup(memcg
);
5787 /* No parent means a non-hierarchical mode on v1 memcg */
5789 return MEMCG_PROT_NONE
;
5794 parent_emin
= READ_ONCE(parent
->memory
.emin
);
5795 emin
= min(emin
, parent_emin
);
5796 if (emin
&& parent_emin
) {
5797 unsigned long min_usage
, siblings_min_usage
;
5799 min_usage
= min(usage
, memcg
->memory
.min
);
5800 siblings_min_usage
= atomic_long_read(
5801 &parent
->memory
.children_min_usage
);
5803 if (min_usage
&& siblings_min_usage
)
5804 emin
= min(emin
, parent_emin
* min_usage
/
5805 siblings_min_usage
);
5808 parent_elow
= READ_ONCE(parent
->memory
.elow
);
5809 elow
= min(elow
, parent_elow
);
5810 if (elow
&& parent_elow
) {
5811 unsigned long low_usage
, siblings_low_usage
;
5813 low_usage
= min(usage
, memcg
->memory
.low
);
5814 siblings_low_usage
= atomic_long_read(
5815 &parent
->memory
.children_low_usage
);
5817 if (low_usage
&& siblings_low_usage
)
5818 elow
= min(elow
, parent_elow
* low_usage
/
5819 siblings_low_usage
);
5823 memcg
->memory
.emin
= emin
;
5824 memcg
->memory
.elow
= elow
;
5827 return MEMCG_PROT_MIN
;
5828 else if (usage
<= elow
)
5829 return MEMCG_PROT_LOW
;
5831 return MEMCG_PROT_NONE
;
5835 * mem_cgroup_try_charge - try charging a page
5836 * @page: page to charge
5837 * @mm: mm context of the victim
5838 * @gfp_mask: reclaim mode
5839 * @memcgp: charged memcg return
5840 * @compound: charge the page as compound or small page
5842 * Try to charge @page to the memcg that @mm belongs to, reclaiming
5843 * pages according to @gfp_mask if necessary.
5845 * Returns 0 on success, with *@memcgp pointing to the charged memcg.
5846 * Otherwise, an error code is returned.
5848 * After page->mapping has been set up, the caller must finalize the
5849 * charge with mem_cgroup_commit_charge(). Or abort the transaction
5850 * with mem_cgroup_cancel_charge() in case page instantiation fails.
5852 int mem_cgroup_try_charge(struct page
*page
, struct mm_struct
*mm
,
5853 gfp_t gfp_mask
, struct mem_cgroup
**memcgp
,
5856 struct mem_cgroup
*memcg
= NULL
;
5857 unsigned int nr_pages
= compound
? hpage_nr_pages(page
) : 1;
5860 if (mem_cgroup_disabled())
5863 if (PageSwapCache(page
)) {
5865 * Every swap fault against a single page tries to charge the
5866 * page, bail as early as possible. shmem_unuse() encounters
5867 * already charged pages, too. The USED bit is protected by
5868 * the page lock, which serializes swap cache removal, which
5869 * in turn serializes uncharging.
5871 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
5872 if (compound_head(page
)->mem_cgroup
)
5875 if (do_swap_account
) {
5876 swp_entry_t ent
= { .val
= page_private(page
), };
5877 unsigned short id
= lookup_swap_cgroup_id(ent
);
5880 memcg
= mem_cgroup_from_id(id
);
5881 if (memcg
&& !css_tryget_online(&memcg
->css
))
5888 memcg
= get_mem_cgroup_from_mm(mm
);
5890 ret
= try_charge(memcg
, gfp_mask
, nr_pages
);
5892 css_put(&memcg
->css
);
5898 int mem_cgroup_try_charge_delay(struct page
*page
, struct mm_struct
*mm
,
5899 gfp_t gfp_mask
, struct mem_cgroup
**memcgp
,
5902 struct mem_cgroup
*memcg
;
5905 ret
= mem_cgroup_try_charge(page
, mm
, gfp_mask
, memcgp
, compound
);
5907 mem_cgroup_throttle_swaprate(memcg
, page_to_nid(page
), gfp_mask
);
5912 * mem_cgroup_commit_charge - commit a page charge
5913 * @page: page to charge
5914 * @memcg: memcg to charge the page to
5915 * @lrucare: page might be on LRU already
5916 * @compound: charge the page as compound or small page
5918 * Finalize a charge transaction started by mem_cgroup_try_charge(),
5919 * after page->mapping has been set up. This must happen atomically
5920 * as part of the page instantiation, i.e. under the page table lock
5921 * for anonymous pages, under the page lock for page and swap cache.
5923 * In addition, the page must not be on the LRU during the commit, to
5924 * prevent racing with task migration. If it might be, use @lrucare.
5926 * Use mem_cgroup_cancel_charge() to cancel the transaction instead.
5928 void mem_cgroup_commit_charge(struct page
*page
, struct mem_cgroup
*memcg
,
5929 bool lrucare
, bool compound
)
5931 unsigned int nr_pages
= compound
? hpage_nr_pages(page
) : 1;
5933 VM_BUG_ON_PAGE(!page
->mapping
, page
);
5934 VM_BUG_ON_PAGE(PageLRU(page
) && !lrucare
, page
);
5936 if (mem_cgroup_disabled())
5939 * Swap faults will attempt to charge the same page multiple
5940 * times. But reuse_swap_page() might have removed the page
5941 * from swapcache already, so we can't check PageSwapCache().
5946 commit_charge(page
, memcg
, lrucare
);
5948 local_irq_disable();
5949 mem_cgroup_charge_statistics(memcg
, page
, compound
, nr_pages
);
5950 memcg_check_events(memcg
, page
);
5953 if (do_memsw_account() && PageSwapCache(page
)) {
5954 swp_entry_t entry
= { .val
= page_private(page
) };
5956 * The swap entry might not get freed for a long time,
5957 * let's not wait for it. The page already received a
5958 * memory+swap charge, drop the swap entry duplicate.
5960 mem_cgroup_uncharge_swap(entry
, nr_pages
);
5965 * mem_cgroup_cancel_charge - cancel a page charge
5966 * @page: page to charge
5967 * @memcg: memcg to charge the page to
5968 * @compound: charge the page as compound or small page
5970 * Cancel a charge transaction started by mem_cgroup_try_charge().
5972 void mem_cgroup_cancel_charge(struct page
*page
, struct mem_cgroup
*memcg
,
5975 unsigned int nr_pages
= compound
? hpage_nr_pages(page
) : 1;
5977 if (mem_cgroup_disabled())
5980 * Swap faults will attempt to charge the same page multiple
5981 * times. But reuse_swap_page() might have removed the page
5982 * from swapcache already, so we can't check PageSwapCache().
5987 cancel_charge(memcg
, nr_pages
);
5990 struct uncharge_gather
{
5991 struct mem_cgroup
*memcg
;
5992 unsigned long pgpgout
;
5993 unsigned long nr_anon
;
5994 unsigned long nr_file
;
5995 unsigned long nr_kmem
;
5996 unsigned long nr_huge
;
5997 unsigned long nr_shmem
;
5998 struct page
*dummy_page
;
6001 static inline void uncharge_gather_clear(struct uncharge_gather
*ug
)
6003 memset(ug
, 0, sizeof(*ug
));
6006 static void uncharge_batch(const struct uncharge_gather
*ug
)
6008 unsigned long nr_pages
= ug
->nr_anon
+ ug
->nr_file
+ ug
->nr_kmem
;
6009 unsigned long flags
;
6011 if (!mem_cgroup_is_root(ug
->memcg
)) {
6012 page_counter_uncharge(&ug
->memcg
->memory
, nr_pages
);
6013 if (do_memsw_account())
6014 page_counter_uncharge(&ug
->memcg
->memsw
, nr_pages
);
6015 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys
) && ug
->nr_kmem
)
6016 page_counter_uncharge(&ug
->memcg
->kmem
, ug
->nr_kmem
);
6017 memcg_oom_recover(ug
->memcg
);
6020 local_irq_save(flags
);
6021 __mod_memcg_state(ug
->memcg
, MEMCG_RSS
, -ug
->nr_anon
);
6022 __mod_memcg_state(ug
->memcg
, MEMCG_CACHE
, -ug
->nr_file
);
6023 __mod_memcg_state(ug
->memcg
, MEMCG_RSS_HUGE
, -ug
->nr_huge
);
6024 __mod_memcg_state(ug
->memcg
, NR_SHMEM
, -ug
->nr_shmem
);
6025 __count_memcg_events(ug
->memcg
, PGPGOUT
, ug
->pgpgout
);
6026 __this_cpu_add(ug
->memcg
->stat_cpu
->nr_page_events
, nr_pages
);
6027 memcg_check_events(ug
->memcg
, ug
->dummy_page
);
6028 local_irq_restore(flags
);
6030 if (!mem_cgroup_is_root(ug
->memcg
))
6031 css_put_many(&ug
->memcg
->css
, nr_pages
);
6034 static void uncharge_page(struct page
*page
, struct uncharge_gather
*ug
)
6036 VM_BUG_ON_PAGE(PageLRU(page
), page
);
6037 VM_BUG_ON_PAGE(page_count(page
) && !is_zone_device_page(page
) &&
6038 !PageHWPoison(page
) , page
);
6040 if (!page
->mem_cgroup
)
6044 * Nobody should be changing or seriously looking at
6045 * page->mem_cgroup at this point, we have fully
6046 * exclusive access to the page.
6049 if (ug
->memcg
!= page
->mem_cgroup
) {
6052 uncharge_gather_clear(ug
);
6054 ug
->memcg
= page
->mem_cgroup
;
6057 if (!PageKmemcg(page
)) {
6058 unsigned int nr_pages
= 1;
6060 if (PageTransHuge(page
)) {
6061 nr_pages
<<= compound_order(page
);
6062 ug
->nr_huge
+= nr_pages
;
6065 ug
->nr_anon
+= nr_pages
;
6067 ug
->nr_file
+= nr_pages
;
6068 if (PageSwapBacked(page
))
6069 ug
->nr_shmem
+= nr_pages
;
6073 ug
->nr_kmem
+= 1 << compound_order(page
);
6074 __ClearPageKmemcg(page
);
6077 ug
->dummy_page
= page
;
6078 page
->mem_cgroup
= NULL
;
6081 static void uncharge_list(struct list_head
*page_list
)
6083 struct uncharge_gather ug
;
6084 struct list_head
*next
;
6086 uncharge_gather_clear(&ug
);
6089 * Note that the list can be a single page->lru; hence the
6090 * do-while loop instead of a simple list_for_each_entry().
6092 next
= page_list
->next
;
6096 page
= list_entry(next
, struct page
, lru
);
6097 next
= page
->lru
.next
;
6099 uncharge_page(page
, &ug
);
6100 } while (next
!= page_list
);
6103 uncharge_batch(&ug
);
6107 * mem_cgroup_uncharge - uncharge a page
6108 * @page: page to uncharge
6110 * Uncharge a page previously charged with mem_cgroup_try_charge() and
6111 * mem_cgroup_commit_charge().
6113 void mem_cgroup_uncharge(struct page
*page
)
6115 struct uncharge_gather ug
;
6117 if (mem_cgroup_disabled())
6120 /* Don't touch page->lru of any random page, pre-check: */
6121 if (!page
->mem_cgroup
)
6124 uncharge_gather_clear(&ug
);
6125 uncharge_page(page
, &ug
);
6126 uncharge_batch(&ug
);
6130 * mem_cgroup_uncharge_list - uncharge a list of page
6131 * @page_list: list of pages to uncharge
6133 * Uncharge a list of pages previously charged with
6134 * mem_cgroup_try_charge() and mem_cgroup_commit_charge().
6136 void mem_cgroup_uncharge_list(struct list_head
*page_list
)
6138 if (mem_cgroup_disabled())
6141 if (!list_empty(page_list
))
6142 uncharge_list(page_list
);
6146 * mem_cgroup_migrate - charge a page's replacement
6147 * @oldpage: currently circulating page
6148 * @newpage: replacement page
6150 * Charge @newpage as a replacement page for @oldpage. @oldpage will
6151 * be uncharged upon free.
6153 * Both pages must be locked, @newpage->mapping must be set up.
6155 void mem_cgroup_migrate(struct page
*oldpage
, struct page
*newpage
)
6157 struct mem_cgroup
*memcg
;
6158 unsigned int nr_pages
;
6160 unsigned long flags
;
6162 VM_BUG_ON_PAGE(!PageLocked(oldpage
), oldpage
);
6163 VM_BUG_ON_PAGE(!PageLocked(newpage
), newpage
);
6164 VM_BUG_ON_PAGE(PageAnon(oldpage
) != PageAnon(newpage
), newpage
);
6165 VM_BUG_ON_PAGE(PageTransHuge(oldpage
) != PageTransHuge(newpage
),
6168 if (mem_cgroup_disabled())
6171 /* Page cache replacement: new page already charged? */
6172 if (newpage
->mem_cgroup
)
6175 /* Swapcache readahead pages can get replaced before being charged */
6176 memcg
= oldpage
->mem_cgroup
;
6180 /* Force-charge the new page. The old one will be freed soon */
6181 compound
= PageTransHuge(newpage
);
6182 nr_pages
= compound
? hpage_nr_pages(newpage
) : 1;
6184 page_counter_charge(&memcg
->memory
, nr_pages
);
6185 if (do_memsw_account())
6186 page_counter_charge(&memcg
->memsw
, nr_pages
);
6187 css_get_many(&memcg
->css
, nr_pages
);
6189 commit_charge(newpage
, memcg
, false);
6191 local_irq_save(flags
);
6192 mem_cgroup_charge_statistics(memcg
, newpage
, compound
, nr_pages
);
6193 memcg_check_events(memcg
, newpage
);
6194 local_irq_restore(flags
);
6197 DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key
);
6198 EXPORT_SYMBOL(memcg_sockets_enabled_key
);
6200 void mem_cgroup_sk_alloc(struct sock
*sk
)
6202 struct mem_cgroup
*memcg
;
6204 if (!mem_cgroup_sockets_enabled
)
6208 * Socket cloning can throw us here with sk_memcg already
6209 * filled. It won't however, necessarily happen from
6210 * process context. So the test for root memcg given
6211 * the current task's memcg won't help us in this case.
6213 * Respecting the original socket's memcg is a better
6214 * decision in this case.
6217 css_get(&sk
->sk_memcg
->css
);
6222 memcg
= mem_cgroup_from_task(current
);
6223 if (memcg
== root_mem_cgroup
)
6225 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys
) && !memcg
->tcpmem_active
)
6227 if (css_tryget_online(&memcg
->css
))
6228 sk
->sk_memcg
= memcg
;
6233 void mem_cgroup_sk_free(struct sock
*sk
)
6236 css_put(&sk
->sk_memcg
->css
);
6240 * mem_cgroup_charge_skmem - charge socket memory
6241 * @memcg: memcg to charge
6242 * @nr_pages: number of pages to charge
6244 * Charges @nr_pages to @memcg. Returns %true if the charge fit within
6245 * @memcg's configured limit, %false if the charge had to be forced.
6247 bool mem_cgroup_charge_skmem(struct mem_cgroup
*memcg
, unsigned int nr_pages
)
6249 gfp_t gfp_mask
= GFP_KERNEL
;
6251 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys
)) {
6252 struct page_counter
*fail
;
6254 if (page_counter_try_charge(&memcg
->tcpmem
, nr_pages
, &fail
)) {
6255 memcg
->tcpmem_pressure
= 0;
6258 page_counter_charge(&memcg
->tcpmem
, nr_pages
);
6259 memcg
->tcpmem_pressure
= 1;
6263 /* Don't block in the packet receive path */
6265 gfp_mask
= GFP_NOWAIT
;
6267 mod_memcg_state(memcg
, MEMCG_SOCK
, nr_pages
);
6269 if (try_charge(memcg
, gfp_mask
, nr_pages
) == 0)
6272 try_charge(memcg
, gfp_mask
|__GFP_NOFAIL
, nr_pages
);
6277 * mem_cgroup_uncharge_skmem - uncharge socket memory
6278 * @memcg: memcg to uncharge
6279 * @nr_pages: number of pages to uncharge
6281 void mem_cgroup_uncharge_skmem(struct mem_cgroup
*memcg
, unsigned int nr_pages
)
6283 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys
)) {
6284 page_counter_uncharge(&memcg
->tcpmem
, nr_pages
);
6288 mod_memcg_state(memcg
, MEMCG_SOCK
, -nr_pages
);
6290 refill_stock(memcg
, nr_pages
);
6293 static int __init
cgroup_memory(char *s
)
6297 while ((token
= strsep(&s
, ",")) != NULL
) {
6300 if (!strcmp(token
, "nosocket"))
6301 cgroup_memory_nosocket
= true;
6302 if (!strcmp(token
, "nokmem"))
6303 cgroup_memory_nokmem
= true;
6307 __setup("cgroup.memory=", cgroup_memory
);
6310 * subsys_initcall() for memory controller.
6312 * Some parts like memcg_hotplug_cpu_dead() have to be initialized from this
6313 * context because of lock dependencies (cgroup_lock -> cpu hotplug) but
6314 * basically everything that doesn't depend on a specific mem_cgroup structure
6315 * should be initialized from here.
6317 static int __init
mem_cgroup_init(void)
6321 #ifdef CONFIG_MEMCG_KMEM
6323 * Kmem cache creation is mostly done with the slab_mutex held,
6324 * so use a workqueue with limited concurrency to avoid stalling
6325 * all worker threads in case lots of cgroups are created and
6326 * destroyed simultaneously.
6328 memcg_kmem_cache_wq
= alloc_workqueue("memcg_kmem_cache", 0, 1);
6329 BUG_ON(!memcg_kmem_cache_wq
);
6332 cpuhp_setup_state_nocalls(CPUHP_MM_MEMCQ_DEAD
, "mm/memctrl:dead", NULL
,
6333 memcg_hotplug_cpu_dead
);
6335 for_each_possible_cpu(cpu
)
6336 INIT_WORK(&per_cpu_ptr(&memcg_stock
, cpu
)->work
,
6339 for_each_node(node
) {
6340 struct mem_cgroup_tree_per_node
*rtpn
;
6342 rtpn
= kzalloc_node(sizeof(*rtpn
), GFP_KERNEL
,
6343 node_online(node
) ? node
: NUMA_NO_NODE
);
6345 rtpn
->rb_root
= RB_ROOT
;
6346 rtpn
->rb_rightmost
= NULL
;
6347 spin_lock_init(&rtpn
->lock
);
6348 soft_limit_tree
.rb_tree_per_node
[node
] = rtpn
;
6353 subsys_initcall(mem_cgroup_init
);
6355 #ifdef CONFIG_MEMCG_SWAP
6356 static struct mem_cgroup
*mem_cgroup_id_get_online(struct mem_cgroup
*memcg
)
6358 while (!refcount_inc_not_zero(&memcg
->id
.ref
)) {
6360 * The root cgroup cannot be destroyed, so it's refcount must
6363 if (WARN_ON_ONCE(memcg
== root_mem_cgroup
)) {
6367 memcg
= parent_mem_cgroup(memcg
);
6369 memcg
= root_mem_cgroup
;
6375 * mem_cgroup_swapout - transfer a memsw charge to swap
6376 * @page: page whose memsw charge to transfer
6377 * @entry: swap entry to move the charge to
6379 * Transfer the memsw charge of @page to @entry.
6381 void mem_cgroup_swapout(struct page
*page
, swp_entry_t entry
)
6383 struct mem_cgroup
*memcg
, *swap_memcg
;
6384 unsigned int nr_entries
;
6385 unsigned short oldid
;
6387 VM_BUG_ON_PAGE(PageLRU(page
), page
);
6388 VM_BUG_ON_PAGE(page_count(page
), page
);
6390 if (!do_memsw_account())
6393 memcg
= page
->mem_cgroup
;
6395 /* Readahead page, never charged */
6400 * In case the memcg owning these pages has been offlined and doesn't
6401 * have an ID allocated to it anymore, charge the closest online
6402 * ancestor for the swap instead and transfer the memory+swap charge.
6404 swap_memcg
= mem_cgroup_id_get_online(memcg
);
6405 nr_entries
= hpage_nr_pages(page
);
6406 /* Get references for the tail pages, too */
6408 mem_cgroup_id_get_many(swap_memcg
, nr_entries
- 1);
6409 oldid
= swap_cgroup_record(entry
, mem_cgroup_id(swap_memcg
),
6411 VM_BUG_ON_PAGE(oldid
, page
);
6412 mod_memcg_state(swap_memcg
, MEMCG_SWAP
, nr_entries
);
6414 page
->mem_cgroup
= NULL
;
6416 if (!mem_cgroup_is_root(memcg
))
6417 page_counter_uncharge(&memcg
->memory
, nr_entries
);
6419 if (memcg
!= swap_memcg
) {
6420 if (!mem_cgroup_is_root(swap_memcg
))
6421 page_counter_charge(&swap_memcg
->memsw
, nr_entries
);
6422 page_counter_uncharge(&memcg
->memsw
, nr_entries
);
6426 * Interrupts should be disabled here because the caller holds the
6427 * i_pages lock which is taken with interrupts-off. It is
6428 * important here to have the interrupts disabled because it is the
6429 * only synchronisation we have for updating the per-CPU variables.
6431 VM_BUG_ON(!irqs_disabled());
6432 mem_cgroup_charge_statistics(memcg
, page
, PageTransHuge(page
),
6434 memcg_check_events(memcg
, page
);
6436 if (!mem_cgroup_is_root(memcg
))
6437 css_put_many(&memcg
->css
, nr_entries
);
6441 * mem_cgroup_try_charge_swap - try charging swap space for a page
6442 * @page: page being added to swap
6443 * @entry: swap entry to charge
6445 * Try to charge @page's memcg for the swap space at @entry.
6447 * Returns 0 on success, -ENOMEM on failure.
6449 int mem_cgroup_try_charge_swap(struct page
*page
, swp_entry_t entry
)
6451 unsigned int nr_pages
= hpage_nr_pages(page
);
6452 struct page_counter
*counter
;
6453 struct mem_cgroup
*memcg
;
6454 unsigned short oldid
;
6456 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys
) || !do_swap_account
)
6459 memcg
= page
->mem_cgroup
;
6461 /* Readahead page, never charged */
6466 memcg_memory_event(memcg
, MEMCG_SWAP_FAIL
);
6470 memcg
= mem_cgroup_id_get_online(memcg
);
6472 if (!mem_cgroup_is_root(memcg
) &&
6473 !page_counter_try_charge(&memcg
->swap
, nr_pages
, &counter
)) {
6474 memcg_memory_event(memcg
, MEMCG_SWAP_MAX
);
6475 memcg_memory_event(memcg
, MEMCG_SWAP_FAIL
);
6476 mem_cgroup_id_put(memcg
);
6480 /* Get references for the tail pages, too */
6482 mem_cgroup_id_get_many(memcg
, nr_pages
- 1);
6483 oldid
= swap_cgroup_record(entry
, mem_cgroup_id(memcg
), nr_pages
);
6484 VM_BUG_ON_PAGE(oldid
, page
);
6485 mod_memcg_state(memcg
, MEMCG_SWAP
, nr_pages
);
6491 * mem_cgroup_uncharge_swap - uncharge swap space
6492 * @entry: swap entry to uncharge
6493 * @nr_pages: the amount of swap space to uncharge
6495 void mem_cgroup_uncharge_swap(swp_entry_t entry
, unsigned int nr_pages
)
6497 struct mem_cgroup
*memcg
;
6500 if (!do_swap_account
)
6503 id
= swap_cgroup_record(entry
, 0, nr_pages
);
6505 memcg
= mem_cgroup_from_id(id
);
6507 if (!mem_cgroup_is_root(memcg
)) {
6508 if (cgroup_subsys_on_dfl(memory_cgrp_subsys
))
6509 page_counter_uncharge(&memcg
->swap
, nr_pages
);
6511 page_counter_uncharge(&memcg
->memsw
, nr_pages
);
6513 mod_memcg_state(memcg
, MEMCG_SWAP
, -nr_pages
);
6514 mem_cgroup_id_put_many(memcg
, nr_pages
);
6519 long mem_cgroup_get_nr_swap_pages(struct mem_cgroup
*memcg
)
6521 long nr_swap_pages
= get_nr_swap_pages();
6523 if (!do_swap_account
|| !cgroup_subsys_on_dfl(memory_cgrp_subsys
))
6524 return nr_swap_pages
;
6525 for (; memcg
!= root_mem_cgroup
; memcg
= parent_mem_cgroup(memcg
))
6526 nr_swap_pages
= min_t(long, nr_swap_pages
,
6527 READ_ONCE(memcg
->swap
.max
) -
6528 page_counter_read(&memcg
->swap
));
6529 return nr_swap_pages
;
6532 bool mem_cgroup_swap_full(struct page
*page
)
6534 struct mem_cgroup
*memcg
;
6536 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
6540 if (!do_swap_account
|| !cgroup_subsys_on_dfl(memory_cgrp_subsys
))
6543 memcg
= page
->mem_cgroup
;
6547 for (; memcg
!= root_mem_cgroup
; memcg
= parent_mem_cgroup(memcg
))
6548 if (page_counter_read(&memcg
->swap
) * 2 >= memcg
->swap
.max
)
6554 /* for remember boot option*/
6555 #ifdef CONFIG_MEMCG_SWAP_ENABLED
6556 static int really_do_swap_account __initdata
= 1;
6558 static int really_do_swap_account __initdata
;
6561 static int __init
enable_swap_account(char *s
)
6563 if (!strcmp(s
, "1"))
6564 really_do_swap_account
= 1;
6565 else if (!strcmp(s
, "0"))
6566 really_do_swap_account
= 0;
6569 __setup("swapaccount=", enable_swap_account
);
6571 static u64
swap_current_read(struct cgroup_subsys_state
*css
,
6574 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
6576 return (u64
)page_counter_read(&memcg
->swap
) * PAGE_SIZE
;
6579 static int swap_max_show(struct seq_file
*m
, void *v
)
6581 struct mem_cgroup
*memcg
= mem_cgroup_from_css(seq_css(m
));
6582 unsigned long max
= READ_ONCE(memcg
->swap
.max
);
6584 if (max
== PAGE_COUNTER_MAX
)
6585 seq_puts(m
, "max\n");
6587 seq_printf(m
, "%llu\n", (u64
)max
* PAGE_SIZE
);
6592 static ssize_t
swap_max_write(struct kernfs_open_file
*of
,
6593 char *buf
, size_t nbytes
, loff_t off
)
6595 struct mem_cgroup
*memcg
= mem_cgroup_from_css(of_css(of
));
6599 buf
= strstrip(buf
);
6600 err
= page_counter_memparse(buf
, "max", &max
);
6604 xchg(&memcg
->swap
.max
, max
);
6609 static int swap_events_show(struct seq_file
*m
, void *v
)
6611 struct mem_cgroup
*memcg
= mem_cgroup_from_css(seq_css(m
));
6613 seq_printf(m
, "max %lu\n",
6614 atomic_long_read(&memcg
->memory_events
[MEMCG_SWAP_MAX
]));
6615 seq_printf(m
, "fail %lu\n",
6616 atomic_long_read(&memcg
->memory_events
[MEMCG_SWAP_FAIL
]));
6621 static struct cftype swap_files
[] = {
6623 .name
= "swap.current",
6624 .flags
= CFTYPE_NOT_ON_ROOT
,
6625 .read_u64
= swap_current_read
,
6629 .flags
= CFTYPE_NOT_ON_ROOT
,
6630 .seq_show
= swap_max_show
,
6631 .write
= swap_max_write
,
6634 .name
= "swap.events",
6635 .flags
= CFTYPE_NOT_ON_ROOT
,
6636 .file_offset
= offsetof(struct mem_cgroup
, swap_events_file
),
6637 .seq_show
= swap_events_show
,
6642 static struct cftype memsw_cgroup_files
[] = {
6644 .name
= "memsw.usage_in_bytes",
6645 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_USAGE
),
6646 .read_u64
= mem_cgroup_read_u64
,
6649 .name
= "memsw.max_usage_in_bytes",
6650 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_MAX_USAGE
),
6651 .write
= mem_cgroup_reset
,
6652 .read_u64
= mem_cgroup_read_u64
,
6655 .name
= "memsw.limit_in_bytes",
6656 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_LIMIT
),
6657 .write
= mem_cgroup_write
,
6658 .read_u64
= mem_cgroup_read_u64
,
6661 .name
= "memsw.failcnt",
6662 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_FAILCNT
),
6663 .write
= mem_cgroup_reset
,
6664 .read_u64
= mem_cgroup_read_u64
,
6666 { }, /* terminate */
6669 static int __init
mem_cgroup_swap_init(void)
6671 if (!mem_cgroup_disabled() && really_do_swap_account
) {
6672 do_swap_account
= 1;
6673 WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys
,
6675 WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys
,
6676 memsw_cgroup_files
));
6680 subsys_initcall(mem_cgroup_swap_init
);
6682 #endif /* CONFIG_MEMCG_SWAP */