Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 1 | /* memcontrol.c - Memory Controller |
| 2 | * |
| 3 | * Copyright IBM Corporation, 2007 |
| 4 | * Author Balbir Singh <balbir@linux.vnet.ibm.com> |
| 5 | * |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 6 | * Copyright 2007 OpenVZ SWsoft Inc |
| 7 | * Author: Pavel Emelianov <xemul@openvz.org> |
| 8 | * |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 9 | * Memory thresholds |
| 10 | * Copyright (C) 2009 Nokia Corporation |
| 11 | * Author: Kirill A. Shutemov |
| 12 | * |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 13 | * Kernel Memory Controller |
| 14 | * Copyright (C) 2012 Parallels Inc. and Google Inc. |
| 15 | * Authors: Glauber Costa and Suleiman Souhlal |
| 16 | * |
Johannes Weiner | 1575e68 | 2015-04-14 15:44:51 -0700 | [diff] [blame] | 17 | * Native page reclaim |
| 18 | * Charge lifetime sanitation |
| 19 | * Lockless page tracking & accounting |
| 20 | * Unified hierarchy configuration model |
| 21 | * Copyright (C) 2015 Red Hat, Inc., Johannes Weiner |
| 22 | * |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 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. |
| 27 | * |
| 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. |
| 32 | */ |
| 33 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 34 | #include <linux/page_counter.h> |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 35 | #include <linux/memcontrol.h> |
| 36 | #include <linux/cgroup.h> |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 37 | #include <linux/mm.h> |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 38 | #include <linux/hugetlb.h> |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 39 | #include <linux/pagemap.h> |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 40 | #include <linux/smp.h> |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 41 | #include <linux/page-flags.h> |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 42 | #include <linux/backing-dev.h> |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 43 | #include <linux/bit_spinlock.h> |
| 44 | #include <linux/rcupdate.h> |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 45 | #include <linux/limits.h> |
Paul Gortmaker | b9e15ba | 2011-05-26 16:00:52 -0400 | [diff] [blame] | 46 | #include <linux/export.h> |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 47 | #include <linux/mutex.h> |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 48 | #include <linux/rbtree.h> |
Balbir Singh | b6ac57d | 2008-04-29 01:00:19 -0700 | [diff] [blame] | 49 | #include <linux/slab.h> |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 50 | #include <linux/swap.h> |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 51 | #include <linux/swapops.h> |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 52 | #include <linux/spinlock.h> |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 53 | #include <linux/eventfd.h> |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 54 | #include <linux/poll.h> |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 55 | #include <linux/sort.h> |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 56 | #include <linux/fs.h> |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 57 | #include <linux/seq_file.h> |
Anton Vorontsov | 70ddf63 | 2013-04-29 15:08:31 -0700 | [diff] [blame] | 58 | #include <linux/vmpressure.h> |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 59 | #include <linux/mm_inline.h> |
Johannes Weiner | 5d1ea48 | 2014-12-10 15:44:55 -0800 | [diff] [blame] | 60 | #include <linux/swap_cgroup.h> |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 61 | #include <linux/cpu.h> |
KAMEZAWA Hiroyuki | 158e0a2 | 2010-08-10 18:03:00 -0700 | [diff] [blame] | 62 | #include <linux/oom.h> |
Johannes Weiner | 0056f4e | 2013-10-31 16:34:14 -0700 | [diff] [blame] | 63 | #include <linux/lockdep.h> |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 64 | #include <linux/file.h> |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 65 | #include "internal.h" |
Glauber Costa | d1a4c0b | 2011-12-11 21:47:04 +0000 | [diff] [blame] | 66 | #include <net/sock.h> |
Michal Hocko | 4bd2c1e | 2012-10-08 16:33:10 -0700 | [diff] [blame] | 67 | #include <net/ip.h> |
Glauber Costa | d1a4c0b | 2011-12-11 21:47:04 +0000 | [diff] [blame] | 68 | #include <net/tcp_memcontrol.h> |
Qiang Huang | f35c3a8 | 2013-11-12 15:08:22 -0800 | [diff] [blame] | 69 | #include "slab.h" |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 70 | |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 71 | #include <asm/uaccess.h> |
| 72 | |
KOSAKI Motohiro | cc8e970 | 2010-08-09 17:19:57 -0700 | [diff] [blame] | 73 | #include <trace/events/vmscan.h> |
| 74 | |
Tejun Heo | 073219e | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 75 | struct cgroup_subsys memory_cgrp_subsys __read_mostly; |
| 76 | EXPORT_SYMBOL(memory_cgrp_subsys); |
David Rientjes | 68ae564 | 2012-12-12 13:51:57 -0800 | [diff] [blame] | 77 | |
KAMEZAWA Hiroyuki | a181b0e | 2008-07-25 01:47:08 -0700 | [diff] [blame] | 78 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
Kirill A. Shutemov | 6bbda35 | 2012-05-29 15:06:55 -0700 | [diff] [blame] | 79 | static struct mem_cgroup *root_mem_cgroup __read_mostly; |
Tejun Heo | 5616163 | 2015-05-22 17:13:20 -0400 | [diff] [blame] | 80 | struct cgroup_subsys_state *mem_cgroup_root_css __read_mostly; |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 81 | |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 82 | /* Whether the swap controller is active */ |
Andrew Morton | c255a45 | 2012-07-31 16:43:02 -0700 | [diff] [blame] | 83 | #ifdef CONFIG_MEMCG_SWAP |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 84 | int do_swap_account __read_mostly; |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 85 | #else |
Kirill A. Shutemov | a0db00f | 2012-05-29 15:06:56 -0700 | [diff] [blame] | 86 | #define do_swap_account 0 |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 87 | #endif |
| 88 | |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 89 | static const char * const mem_cgroup_stat_names[] = { |
| 90 | "cache", |
| 91 | "rss", |
David Rientjes | b070e65 | 2013-05-07 16:18:09 -0700 | [diff] [blame] | 92 | "rss_huge", |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 93 | "mapped_file", |
Greg Thelen | c4843a7 | 2015-05-22 17:13:16 -0400 | [diff] [blame] | 94 | "dirty", |
Sha Zhengju | 3ea67d0 | 2013-09-12 15:13:53 -0700 | [diff] [blame] | 95 | "writeback", |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 96 | "swap", |
| 97 | }; |
| 98 | |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 99 | static const char * const mem_cgroup_events_names[] = { |
| 100 | "pgpgin", |
| 101 | "pgpgout", |
| 102 | "pgfault", |
| 103 | "pgmajfault", |
| 104 | }; |
| 105 | |
Sha Zhengju | 58cf188 | 2013-02-22 16:32:05 -0800 | [diff] [blame] | 106 | static const char * const mem_cgroup_lru_names[] = { |
| 107 | "inactive_anon", |
| 108 | "active_anon", |
| 109 | "inactive_file", |
| 110 | "active_file", |
| 111 | "unevictable", |
| 112 | }; |
| 113 | |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 114 | /* |
| 115 | * Per memcg event counter is incremented at every pagein/pageout. With THP, |
| 116 | * it will be incremated by the number of pages. This counter is used for |
| 117 | * for trigger some periodic events. This is straightforward and better |
| 118 | * than using jiffies etc. to handle periodic memcg event. |
| 119 | */ |
| 120 | enum mem_cgroup_events_target { |
| 121 | MEM_CGROUP_TARGET_THRESH, |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 122 | MEM_CGROUP_TARGET_SOFTLIMIT, |
KAMEZAWA Hiroyuki | 453a9bf | 2011-07-08 15:39:43 -0700 | [diff] [blame] | 123 | MEM_CGROUP_TARGET_NUMAINFO, |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 124 | MEM_CGROUP_NTARGETS, |
| 125 | }; |
Kirill A. Shutemov | a0db00f | 2012-05-29 15:06:56 -0700 | [diff] [blame] | 126 | #define THRESHOLDS_EVENTS_TARGET 128 |
| 127 | #define SOFTLIMIT_EVENTS_TARGET 1024 |
| 128 | #define NUMAINFO_EVENTS_TARGET 1024 |
Johannes Weiner | e9f8974 | 2011-03-23 16:42:37 -0700 | [diff] [blame] | 129 | |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 130 | struct mem_cgroup_stat_cpu { |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 131 | long count[MEM_CGROUP_STAT_NSTATS]; |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 132 | unsigned long events[MEMCG_NR_EVENTS]; |
Johannes Weiner | 1311471 | 2012-05-29 15:07:07 -0700 | [diff] [blame] | 133 | unsigned long nr_page_events; |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 134 | unsigned long targets[MEM_CGROUP_NTARGETS]; |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 135 | }; |
| 136 | |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 137 | struct reclaim_iter { |
| 138 | struct mem_cgroup *position; |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 139 | /* scan generation, increased every round-trip */ |
| 140 | unsigned int generation; |
| 141 | }; |
| 142 | |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 143 | /* |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 144 | * per-zone information in memory controller. |
| 145 | */ |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 146 | struct mem_cgroup_per_zone { |
Johannes Weiner | 6290df5 | 2012-01-12 17:18:10 -0800 | [diff] [blame] | 147 | struct lruvec lruvec; |
Hugh Dickins | 1eb4927 | 2012-03-21 16:34:19 -0700 | [diff] [blame] | 148 | unsigned long lru_size[NR_LRU_LISTS]; |
KOSAKI Motohiro | 3e2f41f | 2009-01-07 18:08:20 -0800 | [diff] [blame] | 149 | |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 150 | struct reclaim_iter iter[DEF_PRIORITY + 1]; |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 151 | |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 152 | struct rb_node tree_node; /* RB tree node */ |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 153 | unsigned long usage_in_excess;/* Set to the value by which */ |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 154 | /* the soft limit is exceeded*/ |
| 155 | bool on_tree; |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 156 | struct mem_cgroup *memcg; /* Back pointer, we cannot */ |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 157 | /* use container_of */ |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 158 | }; |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 159 | |
| 160 | struct mem_cgroup_per_node { |
| 161 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; |
| 162 | }; |
| 163 | |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 164 | /* |
| 165 | * Cgroups above their limits are maintained in a RB-Tree, independent of |
| 166 | * their hierarchy representation |
| 167 | */ |
| 168 | |
| 169 | struct mem_cgroup_tree_per_zone { |
| 170 | struct rb_root rb_root; |
| 171 | spinlock_t lock; |
| 172 | }; |
| 173 | |
| 174 | struct mem_cgroup_tree_per_node { |
| 175 | struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES]; |
| 176 | }; |
| 177 | |
| 178 | struct mem_cgroup_tree { |
| 179 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; |
| 180 | }; |
| 181 | |
| 182 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; |
| 183 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 184 | struct mem_cgroup_threshold { |
| 185 | struct eventfd_ctx *eventfd; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 186 | unsigned long threshold; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 187 | }; |
| 188 | |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 189 | /* For threshold */ |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 190 | struct mem_cgroup_threshold_ary { |
Sha Zhengju | 748dad3 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 191 | /* An array index points to threshold just below or equal to usage. */ |
Phil Carmody | 5407a56 | 2010-05-26 14:42:42 -0700 | [diff] [blame] | 192 | int current_threshold; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 193 | /* Size of entries[] */ |
| 194 | unsigned int size; |
| 195 | /* Array of thresholds */ |
| 196 | struct mem_cgroup_threshold entries[0]; |
| 197 | }; |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 198 | |
| 199 | struct mem_cgroup_thresholds { |
| 200 | /* Primary thresholds array */ |
| 201 | struct mem_cgroup_threshold_ary *primary; |
| 202 | /* |
| 203 | * Spare threshold array. |
| 204 | * This is needed to make mem_cgroup_unregister_event() "never fail". |
| 205 | * It must be able to store at least primary->size - 1 entries. |
| 206 | */ |
| 207 | struct mem_cgroup_threshold_ary *spare; |
| 208 | }; |
| 209 | |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 210 | /* for OOM */ |
| 211 | struct mem_cgroup_eventfd_list { |
| 212 | struct list_head list; |
| 213 | struct eventfd_ctx *eventfd; |
| 214 | }; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 215 | |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 216 | /* |
| 217 | * cgroup_event represents events which userspace want to receive. |
| 218 | */ |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 219 | struct mem_cgroup_event { |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 220 | /* |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 221 | * memcg which the event belongs to. |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 222 | */ |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 223 | struct mem_cgroup *memcg; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 224 | /* |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 225 | * eventfd to signal userspace about the event. |
| 226 | */ |
| 227 | struct eventfd_ctx *eventfd; |
| 228 | /* |
| 229 | * Each of these stored in a list by the cgroup. |
| 230 | */ |
| 231 | struct list_head list; |
| 232 | /* |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 233 | * register_event() callback will be used to add new userspace |
| 234 | * waiter for changes related to this event. Use eventfd_signal() |
| 235 | * on eventfd to send notification to userspace. |
| 236 | */ |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 237 | int (*register_event)(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 238 | struct eventfd_ctx *eventfd, const char *args); |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 239 | /* |
| 240 | * unregister_event() callback will be called when userspace closes |
| 241 | * the eventfd or on cgroup removing. This callback must be set, |
| 242 | * if you want provide notification functionality. |
| 243 | */ |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 244 | void (*unregister_event)(struct mem_cgroup *memcg, |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 245 | struct eventfd_ctx *eventfd); |
| 246 | /* |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 247 | * All fields below needed to unregister event when |
| 248 | * userspace closes eventfd. |
| 249 | */ |
| 250 | poll_table pt; |
| 251 | wait_queue_head_t *wqh; |
| 252 | wait_queue_t wait; |
| 253 | struct work_struct remove; |
| 254 | }; |
| 255 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 256 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
| 257 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 258 | |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 259 | /* |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 260 | * The memory controller data structure. The memory controller controls both |
| 261 | * page cache and RSS per cgroup. We would eventually like to provide |
| 262 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, |
| 263 | * to help the administrator determine what knobs to tune. |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 264 | */ |
| 265 | struct mem_cgroup { |
| 266 | struct cgroup_subsys_state css; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 267 | |
| 268 | /* Accounted resources */ |
| 269 | struct page_counter memory; |
| 270 | struct page_counter memsw; |
| 271 | struct page_counter kmem; |
| 272 | |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 273 | /* Normal memory consumption range */ |
| 274 | unsigned long low; |
| 275 | unsigned long high; |
| 276 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 277 | unsigned long soft_limit; |
Hugh Dickins | 59927fb | 2012-03-15 15:17:07 -0700 | [diff] [blame] | 278 | |
Anton Vorontsov | 70ddf63 | 2013-04-29 15:08:31 -0700 | [diff] [blame] | 279 | /* vmpressure notifications */ |
| 280 | struct vmpressure vmpressure; |
| 281 | |
Johannes Weiner | 2f7dd7a | 2014-10-02 16:16:57 -0700 | [diff] [blame] | 282 | /* css_online() has been completed */ |
| 283 | int initialized; |
| 284 | |
Li Zefan | 465939a | 2013-07-08 16:00:38 -0700 | [diff] [blame] | 285 | /* |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 286 | * Should the accounting and control be hierarchical, per subtree? |
| 287 | */ |
| 288 | bool use_hierarchy; |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 289 | |
Tejun Heo | c2b42d3 | 2015-06-24 16:58:23 -0700 | [diff] [blame] | 290 | /* protected by memcg_oom_lock */ |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 291 | bool oom_lock; |
Tejun Heo | c2b42d3 | 2015-06-24 16:58:23 -0700 | [diff] [blame] | 292 | int under_oom; |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 293 | |
KAMEZAWA Hiroyuki | 1f4c025 | 2011-07-26 16:08:21 -0700 | [diff] [blame] | 294 | int swappiness; |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 295 | /* OOM-Killer disable */ |
| 296 | int oom_kill_disable; |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 297 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 298 | /* protect arrays of thresholds */ |
| 299 | struct mutex thresholds_lock; |
| 300 | |
| 301 | /* thresholds for memory usage. RCU-protected */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 302 | struct mem_cgroup_thresholds thresholds; |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 303 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 304 | /* thresholds for mem+swap usage. RCU-protected */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 305 | struct mem_cgroup_thresholds memsw_thresholds; |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 306 | |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 307 | /* For oom notifier event fd */ |
| 308 | struct list_head oom_notify; |
Johannes Weiner | 185efc0 | 2011-09-14 16:21:58 -0700 | [diff] [blame] | 309 | |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 310 | /* |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 311 | * Should we move charges of a task when a task is moved into this |
| 312 | * mem_cgroup ? And what type of charges should we move ? |
| 313 | */ |
Andrew Morton | f894ffa | 2013-09-12 15:13:35 -0700 | [diff] [blame] | 314 | unsigned long move_charge_at_immigrate; |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 315 | /* |
KAMEZAWA Hiroyuki | 619d094 | 2012-03-21 16:34:23 -0700 | [diff] [blame] | 316 | * set > 0 if pages under this cgroup are moving to other cgroup. |
| 317 | */ |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 318 | atomic_t moving_account; |
KAMEZAWA Hiroyuki | 312734c0 | 2012-03-21 16:34:24 -0700 | [diff] [blame] | 319 | /* taken only while moving_account > 0 */ |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 320 | spinlock_t move_lock; |
| 321 | struct task_struct *move_lock_task; |
| 322 | unsigned long move_lock_flags; |
KAMEZAWA Hiroyuki | 619d094 | 2012-03-21 16:34:23 -0700 | [diff] [blame] | 323 | /* |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 324 | * percpu counter. |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 325 | */ |
Kirill A. Shutemov | 3a7951b | 2012-05-29 15:06:56 -0700 | [diff] [blame] | 326 | struct mem_cgroup_stat_cpu __percpu *stat; |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 327 | spinlock_t pcp_counter_lock; |
Glauber Costa | d1a4c0b | 2011-12-11 21:47:04 +0000 | [diff] [blame] | 328 | |
Michal Hocko | 4bd2c1e | 2012-10-08 16:33:10 -0700 | [diff] [blame] | 329 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_INET) |
Eric W. Biederman | 2e685cad | 2013-10-19 16:26:19 -0700 | [diff] [blame] | 330 | struct cg_proto tcp_mem; |
Glauber Costa | d1a4c0b | 2011-12-11 21:47:04 +0000 | [diff] [blame] | 331 | #endif |
Glauber Costa | 2633d7a | 2012-12-18 14:22:34 -0800 | [diff] [blame] | 332 | #if defined(CONFIG_MEMCG_KMEM) |
Vladimir Davydov | f7ce319 | 2015-02-12 14:59:20 -0800 | [diff] [blame] | 333 | /* Index in the kmem_cache->memcg_params.memcg_caches array */ |
Glauber Costa | 2633d7a | 2012-12-18 14:22:34 -0800 | [diff] [blame] | 334 | int kmemcg_id; |
Vladimir Davydov | 2788cf0 | 2015-02-12 14:59:38 -0800 | [diff] [blame] | 335 | bool kmem_acct_activated; |
Vladimir Davydov | 2a4db7e | 2015-02-12 14:59:32 -0800 | [diff] [blame] | 336 | bool kmem_acct_active; |
Glauber Costa | 2633d7a | 2012-12-18 14:22:34 -0800 | [diff] [blame] | 337 | #endif |
Glauber Costa | 45cf7eb | 2013-02-22 16:34:49 -0800 | [diff] [blame] | 338 | |
| 339 | int last_scanned_node; |
| 340 | #if MAX_NUMNODES > 1 |
| 341 | nodemask_t scan_nodes; |
| 342 | atomic_t numainfo_events; |
| 343 | atomic_t numainfo_updating; |
| 344 | #endif |
Anton Vorontsov | 70ddf63 | 2013-04-29 15:08:31 -0700 | [diff] [blame] | 345 | |
Tejun Heo | 52ebea7 | 2015-05-22 17:13:37 -0400 | [diff] [blame] | 346 | #ifdef CONFIG_CGROUP_WRITEBACK |
| 347 | struct list_head cgwb_list; |
Tejun Heo | 841710a | 2015-05-22 18:23:33 -0400 | [diff] [blame] | 348 | struct wb_domain cgwb_domain; |
Tejun Heo | 52ebea7 | 2015-05-22 17:13:37 -0400 | [diff] [blame] | 349 | #endif |
| 350 | |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 351 | /* List of events which userspace want to receive */ |
| 352 | struct list_head event_list; |
| 353 | spinlock_t event_list_lock; |
| 354 | |
Johannes Weiner | 54f72fe | 2013-07-08 15:59:49 -0700 | [diff] [blame] | 355 | struct mem_cgroup_per_node *nodeinfo[0]; |
| 356 | /* WARNING: nodeinfo must be the last member here */ |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 357 | }; |
| 358 | |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 359 | #ifdef CONFIG_MEMCG_KMEM |
Vladimir Davydov | cb731d6 | 2015-02-12 14:58:54 -0800 | [diff] [blame] | 360 | bool memcg_kmem_is_active(struct mem_cgroup *memcg) |
Glauber Costa | 7de3768 | 2012-12-18 14:22:07 -0800 | [diff] [blame] | 361 | { |
Vladimir Davydov | 2a4db7e | 2015-02-12 14:59:32 -0800 | [diff] [blame] | 362 | return memcg->kmem_acct_active; |
Glauber Costa | 7de3768 | 2012-12-18 14:22:07 -0800 | [diff] [blame] | 363 | } |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 364 | #endif |
| 365 | |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 366 | /* Stuffs for move charges at task migration. */ |
| 367 | /* |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 368 | * Types of charges to be moved. |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 369 | */ |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 370 | #define MOVE_ANON 0x1U |
| 371 | #define MOVE_FILE 0x2U |
| 372 | #define MOVE_MASK (MOVE_ANON | MOVE_FILE) |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 373 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 374 | /* "mc" and its members are protected by cgroup_mutex */ |
| 375 | static struct move_charge_struct { |
Daisuke Nishimura | b1dd693 | 2010-11-24 12:57:06 -0800 | [diff] [blame] | 376 | spinlock_t lock; /* for from, to */ |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 377 | struct mem_cgroup *from; |
| 378 | struct mem_cgroup *to; |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 379 | unsigned long flags; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 380 | unsigned long precharge; |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 381 | unsigned long moved_charge; |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 382 | unsigned long moved_swap; |
Daisuke Nishimura | 8033b97 | 2010-03-10 15:22:16 -0800 | [diff] [blame] | 383 | struct task_struct *moving_task; /* a task moving charges */ |
| 384 | wait_queue_head_t waitq; /* a waitq for other context */ |
| 385 | } mc = { |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 386 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
Daisuke Nishimura | 8033b97 | 2010-03-10 15:22:16 -0800 | [diff] [blame] | 387 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
| 388 | }; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 389 | |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 390 | /* |
| 391 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft |
| 392 | * limit reclaim to prevent infinite loops, if they ever occur. |
| 393 | */ |
Kirill A. Shutemov | a0db00f | 2012-05-29 15:06:56 -0700 | [diff] [blame] | 394 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 395 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 396 | |
KAMEZAWA Hiroyuki | 217bc31 | 2008-02-07 00:14:17 -0800 | [diff] [blame] | 397 | enum charge_type { |
| 398 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, |
Kamezawa Hiroyuki | 41326c1 | 2012-07-31 16:41:40 -0700 | [diff] [blame] | 399 | MEM_CGROUP_CHARGE_TYPE_ANON, |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 400 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
KAMEZAWA Hiroyuki | 8a9478c | 2009-06-17 16:27:17 -0700 | [diff] [blame] | 401 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
KAMEZAWA Hiroyuki | c05555b | 2008-10-18 20:28:11 -0700 | [diff] [blame] | 402 | NR_CHARGE_TYPE, |
| 403 | }; |
| 404 | |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 405 | /* for encoding cft->private value on file */ |
Glauber Costa | 86ae53e | 2012-12-18 14:21:45 -0800 | [diff] [blame] | 406 | enum res_type { |
| 407 | _MEM, |
| 408 | _MEMSWAP, |
| 409 | _OOM_TYPE, |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 410 | _KMEM, |
Glauber Costa | 86ae53e | 2012-12-18 14:21:45 -0800 | [diff] [blame] | 411 | }; |
| 412 | |
Kirill A. Shutemov | a0db00f | 2012-05-29 15:06:56 -0700 | [diff] [blame] | 413 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
| 414 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 415 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 416 | /* Used for OOM nofiier */ |
| 417 | #define OOM_CONTROL (0) |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 418 | |
Balbir Singh | 75822b4 | 2009-09-23 15:56:38 -0700 | [diff] [blame] | 419 | /* |
Glauber Costa | 0999821 | 2013-02-22 16:34:55 -0800 | [diff] [blame] | 420 | * The memcg_create_mutex will be held whenever a new cgroup is created. |
| 421 | * As a consequence, any change that needs to protect against new child cgroups |
| 422 | * appearing has to hold it as well. |
| 423 | */ |
| 424 | static DEFINE_MUTEX(memcg_create_mutex); |
| 425 | |
Wanpeng Li | b214514 | 2012-07-31 16:46:01 -0700 | [diff] [blame] | 426 | struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *s) |
| 427 | { |
Tejun Heo | a7c6d55 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 428 | return s ? container_of(s, struct mem_cgroup, css) : NULL; |
Wanpeng Li | b214514 | 2012-07-31 16:46:01 -0700 | [diff] [blame] | 429 | } |
| 430 | |
Anton Vorontsov | 70ddf63 | 2013-04-29 15:08:31 -0700 | [diff] [blame] | 431 | /* Some nice accessors for the vmpressure. */ |
| 432 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) |
| 433 | { |
| 434 | if (!memcg) |
| 435 | memcg = root_mem_cgroup; |
| 436 | return &memcg->vmpressure; |
| 437 | } |
| 438 | |
| 439 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) |
| 440 | { |
| 441 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; |
| 442 | } |
| 443 | |
Michal Hocko | 7ffc0ed | 2012-10-08 16:33:13 -0700 | [diff] [blame] | 444 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
| 445 | { |
| 446 | return (memcg == root_mem_cgroup); |
| 447 | } |
| 448 | |
Li Zefan | 4219b2d | 2013-09-23 16:56:29 +0800 | [diff] [blame] | 449 | /* |
| 450 | * We restrict the id in the range of [1, 65535], so it can fit into |
| 451 | * an unsigned short. |
| 452 | */ |
| 453 | #define MEM_CGROUP_ID_MAX USHRT_MAX |
| 454 | |
Li Zefan | 34c00c3 | 2013-09-23 16:56:01 +0800 | [diff] [blame] | 455 | static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg) |
| 456 | { |
Tejun Heo | 15a4c83 | 2014-05-04 15:09:14 -0400 | [diff] [blame] | 457 | return memcg->css.id; |
Li Zefan | 34c00c3 | 2013-09-23 16:56:01 +0800 | [diff] [blame] | 458 | } |
| 459 | |
Vladimir Davydov | adbe427 | 2015-04-15 16:13:00 -0700 | [diff] [blame] | 460 | /* |
| 461 | * A helper function to get mem_cgroup from ID. must be called under |
| 462 | * rcu_read_lock(). The caller is responsible for calling |
| 463 | * css_tryget_online() if the mem_cgroup is used for charging. (dropping |
| 464 | * refcnt from swap can be called against removed memcg.) |
| 465 | */ |
Li Zefan | 34c00c3 | 2013-09-23 16:56:01 +0800 | [diff] [blame] | 466 | static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id) |
| 467 | { |
| 468 | struct cgroup_subsys_state *css; |
| 469 | |
Tejun Heo | 7d699dd | 2014-05-04 15:09:13 -0400 | [diff] [blame] | 470 | css = css_from_id(id, &memory_cgrp_subsys); |
Li Zefan | 34c00c3 | 2013-09-23 16:56:01 +0800 | [diff] [blame] | 471 | return mem_cgroup_from_css(css); |
| 472 | } |
| 473 | |
Glauber Costa | e1aab16 | 2011-12-11 21:47:03 +0000 | [diff] [blame] | 474 | /* Writing them here to avoid exposing memcg's inner layout */ |
Michal Hocko | 4bd2c1e | 2012-10-08 16:33:10 -0700 | [diff] [blame] | 475 | #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM) |
Glauber Costa | e1aab16 | 2011-12-11 21:47:03 +0000 | [diff] [blame] | 476 | |
Glauber Costa | e1aab16 | 2011-12-11 21:47:03 +0000 | [diff] [blame] | 477 | void sock_update_memcg(struct sock *sk) |
| 478 | { |
Glauber Costa | 376be5f | 2012-01-20 04:57:14 +0000 | [diff] [blame] | 479 | if (mem_cgroup_sockets_enabled) { |
Glauber Costa | e1aab16 | 2011-12-11 21:47:03 +0000 | [diff] [blame] | 480 | struct mem_cgroup *memcg; |
Glauber Costa | 3f13461 | 2012-05-29 15:07:11 -0700 | [diff] [blame] | 481 | struct cg_proto *cg_proto; |
Glauber Costa | e1aab16 | 2011-12-11 21:47:03 +0000 | [diff] [blame] | 482 | |
| 483 | BUG_ON(!sk->sk_prot->proto_cgroup); |
| 484 | |
Glauber Costa | f3f511e | 2012-01-05 20:16:39 +0000 | [diff] [blame] | 485 | /* Socket cloning can throw us here with sk_cgrp already |
| 486 | * filled. It won't however, necessarily happen from |
| 487 | * process context. So the test for root memcg given |
| 488 | * the current task's memcg won't help us in this case. |
| 489 | * |
| 490 | * Respecting the original socket's memcg is a better |
| 491 | * decision in this case. |
| 492 | */ |
| 493 | if (sk->sk_cgrp) { |
| 494 | BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg)); |
Li Zefan | 5347e5a | 2013-07-08 16:00:30 -0700 | [diff] [blame] | 495 | css_get(&sk->sk_cgrp->memcg->css); |
Glauber Costa | f3f511e | 2012-01-05 20:16:39 +0000 | [diff] [blame] | 496 | return; |
| 497 | } |
| 498 | |
Glauber Costa | e1aab16 | 2011-12-11 21:47:03 +0000 | [diff] [blame] | 499 | rcu_read_lock(); |
| 500 | memcg = mem_cgroup_from_task(current); |
Glauber Costa | 3f13461 | 2012-05-29 15:07:11 -0700 | [diff] [blame] | 501 | cg_proto = sk->sk_prot->proto_cgroup(memcg); |
Li Zefan | 5347e5a | 2013-07-08 16:00:30 -0700 | [diff] [blame] | 502 | if (!mem_cgroup_is_root(memcg) && |
Tejun Heo | ec903c0 | 2014-05-13 12:11:01 -0400 | [diff] [blame] | 503 | memcg_proto_active(cg_proto) && |
| 504 | css_tryget_online(&memcg->css)) { |
Glauber Costa | 3f13461 | 2012-05-29 15:07:11 -0700 | [diff] [blame] | 505 | sk->sk_cgrp = cg_proto; |
Glauber Costa | e1aab16 | 2011-12-11 21:47:03 +0000 | [diff] [blame] | 506 | } |
| 507 | rcu_read_unlock(); |
| 508 | } |
| 509 | } |
| 510 | EXPORT_SYMBOL(sock_update_memcg); |
| 511 | |
| 512 | void sock_release_memcg(struct sock *sk) |
| 513 | { |
Glauber Costa | 376be5f | 2012-01-20 04:57:14 +0000 | [diff] [blame] | 514 | if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { |
Glauber Costa | e1aab16 | 2011-12-11 21:47:03 +0000 | [diff] [blame] | 515 | struct mem_cgroup *memcg; |
| 516 | WARN_ON(!sk->sk_cgrp->memcg); |
| 517 | memcg = sk->sk_cgrp->memcg; |
Li Zefan | 5347e5a | 2013-07-08 16:00:30 -0700 | [diff] [blame] | 518 | css_put(&sk->sk_cgrp->memcg->css); |
Glauber Costa | e1aab16 | 2011-12-11 21:47:03 +0000 | [diff] [blame] | 519 | } |
| 520 | } |
Glauber Costa | d1a4c0b | 2011-12-11 21:47:04 +0000 | [diff] [blame] | 521 | |
| 522 | struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg) |
| 523 | { |
| 524 | if (!memcg || mem_cgroup_is_root(memcg)) |
| 525 | return NULL; |
| 526 | |
Eric W. Biederman | 2e685cad | 2013-10-19 16:26:19 -0700 | [diff] [blame] | 527 | return &memcg->tcp_mem; |
Glauber Costa | d1a4c0b | 2011-12-11 21:47:04 +0000 | [diff] [blame] | 528 | } |
| 529 | EXPORT_SYMBOL(tcp_proto_cgroup); |
Glauber Costa | e1aab16 | 2011-12-11 21:47:03 +0000 | [diff] [blame] | 530 | |
Glauber Costa | 3f13461 | 2012-05-29 15:07:11 -0700 | [diff] [blame] | 531 | #endif |
| 532 | |
Glauber Costa | a8964b9 | 2012-12-18 14:22:09 -0800 | [diff] [blame] | 533 | #ifdef CONFIG_MEMCG_KMEM |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 534 | /* |
Vladimir Davydov | f7ce319 | 2015-02-12 14:59:20 -0800 | [diff] [blame] | 535 | * This will be the memcg's index in each cache's ->memcg_params.memcg_caches. |
Li Zefan | b862783 | 2013-09-23 16:56:47 +0800 | [diff] [blame] | 536 | * The main reason for not using cgroup id for this: |
| 537 | * this works better in sparse environments, where we have a lot of memcgs, |
| 538 | * but only a few kmem-limited. Or also, if we have, for instance, 200 |
| 539 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a |
| 540 | * 200 entry array for that. |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 541 | * |
Vladimir Davydov | dbcf73e | 2015-02-12 14:58:57 -0800 | [diff] [blame] | 542 | * The current size of the caches array is stored in memcg_nr_cache_ids. It |
| 543 | * will double each time we have to increase it. |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 544 | */ |
Vladimir Davydov | dbcf73e | 2015-02-12 14:58:57 -0800 | [diff] [blame] | 545 | static DEFINE_IDA(memcg_cache_ida); |
| 546 | int memcg_nr_cache_ids; |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 547 | |
Vladimir Davydov | 05257a1 | 2015-02-12 14:59:01 -0800 | [diff] [blame] | 548 | /* Protects memcg_nr_cache_ids */ |
| 549 | static DECLARE_RWSEM(memcg_cache_ids_sem); |
| 550 | |
| 551 | void memcg_get_cache_ids(void) |
| 552 | { |
| 553 | down_read(&memcg_cache_ids_sem); |
| 554 | } |
| 555 | |
| 556 | void memcg_put_cache_ids(void) |
| 557 | { |
| 558 | up_read(&memcg_cache_ids_sem); |
| 559 | } |
| 560 | |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 561 | /* |
| 562 | * MIN_SIZE is different than 1, because we would like to avoid going through |
| 563 | * the alloc/free process all the time. In a small machine, 4 kmem-limited |
| 564 | * cgroups is a reasonable guess. In the future, it could be a parameter or |
| 565 | * tunable, but that is strictly not necessary. |
| 566 | * |
Li Zefan | b862783 | 2013-09-23 16:56:47 +0800 | [diff] [blame] | 567 | * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 568 | * this constant directly from cgroup, but it is understandable that this is |
| 569 | * better kept as an internal representation in cgroup.c. In any case, the |
Li Zefan | b862783 | 2013-09-23 16:56:47 +0800 | [diff] [blame] | 570 | * cgrp_id space is not getting any smaller, and we don't have to necessarily |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 571 | * increase ours as well if it increases. |
| 572 | */ |
| 573 | #define MEMCG_CACHES_MIN_SIZE 4 |
Li Zefan | b862783 | 2013-09-23 16:56:47 +0800 | [diff] [blame] | 574 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 575 | |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 576 | /* |
| 577 | * A lot of the calls to the cache allocation functions are expected to be |
| 578 | * inlined by the compiler. Since the calls to memcg_kmem_get_cache are |
| 579 | * conditional to this static branch, we'll have to allow modules that does |
| 580 | * kmem_cache_alloc and the such to see this symbol as well |
| 581 | */ |
Glauber Costa | a8964b9 | 2012-12-18 14:22:09 -0800 | [diff] [blame] | 582 | struct static_key memcg_kmem_enabled_key; |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 583 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
Glauber Costa | a8964b9 | 2012-12-18 14:22:09 -0800 | [diff] [blame] | 584 | |
Glauber Costa | a8964b9 | 2012-12-18 14:22:09 -0800 | [diff] [blame] | 585 | #endif /* CONFIG_MEMCG_KMEM */ |
| 586 | |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 587 | static struct mem_cgroup_per_zone * |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 588 | mem_cgroup_zone_zoneinfo(struct mem_cgroup *memcg, struct zone *zone) |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 589 | { |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 590 | int nid = zone_to_nid(zone); |
| 591 | int zid = zone_idx(zone); |
| 592 | |
Johannes Weiner | 54f72fe | 2013-07-08 15:59:49 -0700 | [diff] [blame] | 593 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 594 | } |
| 595 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 596 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg) |
Wu Fengguang | d324236 | 2009-12-16 12:19:59 +0100 | [diff] [blame] | 597 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 598 | return &memcg->css; |
Wu Fengguang | d324236 | 2009-12-16 12:19:59 +0100 | [diff] [blame] | 599 | } |
| 600 | |
Tejun Heo | ad7fa85 | 2015-05-27 20:00:02 -0400 | [diff] [blame] | 601 | /** |
| 602 | * mem_cgroup_css_from_page - css of the memcg associated with a page |
| 603 | * @page: page of interest |
| 604 | * |
| 605 | * If memcg is bound to the default hierarchy, css of the memcg associated |
| 606 | * with @page is returned. The returned css remains associated with @page |
| 607 | * until it is released. |
| 608 | * |
| 609 | * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup |
| 610 | * is returned. |
| 611 | * |
| 612 | * XXX: The above description of behavior on the default hierarchy isn't |
| 613 | * strictly true yet as replace_page_cache_page() can modify the |
| 614 | * association before @page is released even on the default hierarchy; |
| 615 | * however, the current and planned usages don't mix the the two functions |
| 616 | * and replace_page_cache_page() will soon be updated to make the invariant |
| 617 | * actually true. |
| 618 | */ |
| 619 | struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page) |
| 620 | { |
| 621 | struct mem_cgroup *memcg; |
| 622 | |
| 623 | rcu_read_lock(); |
| 624 | |
| 625 | memcg = page->mem_cgroup; |
| 626 | |
| 627 | if (!memcg || !cgroup_on_dfl(memcg->css.cgroup)) |
| 628 | memcg = root_mem_cgroup; |
| 629 | |
| 630 | rcu_read_unlock(); |
| 631 | return &memcg->css; |
| 632 | } |
| 633 | |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 634 | static struct mem_cgroup_per_zone * |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 635 | mem_cgroup_page_zoneinfo(struct mem_cgroup *memcg, struct page *page) |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 636 | { |
Johannes Weiner | 97a6c37 | 2011-03-23 16:42:27 -0700 | [diff] [blame] | 637 | int nid = page_to_nid(page); |
| 638 | int zid = page_zonenum(page); |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 639 | |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 640 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 641 | } |
| 642 | |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 643 | static struct mem_cgroup_tree_per_zone * |
| 644 | soft_limit_tree_node_zone(int nid, int zid) |
| 645 | { |
| 646 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; |
| 647 | } |
| 648 | |
| 649 | static struct mem_cgroup_tree_per_zone * |
| 650 | soft_limit_tree_from_page(struct page *page) |
| 651 | { |
| 652 | int nid = page_to_nid(page); |
| 653 | int zid = page_zonenum(page); |
| 654 | |
| 655 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; |
| 656 | } |
| 657 | |
Johannes Weiner | cf2c812 | 2014-06-06 14:38:21 -0700 | [diff] [blame] | 658 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_zone *mz, |
| 659 | struct mem_cgroup_tree_per_zone *mctz, |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 660 | unsigned long new_usage_in_excess) |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 661 | { |
| 662 | struct rb_node **p = &mctz->rb_root.rb_node; |
| 663 | struct rb_node *parent = NULL; |
| 664 | struct mem_cgroup_per_zone *mz_node; |
| 665 | |
| 666 | if (mz->on_tree) |
| 667 | return; |
| 668 | |
| 669 | mz->usage_in_excess = new_usage_in_excess; |
| 670 | if (!mz->usage_in_excess) |
| 671 | return; |
| 672 | while (*p) { |
| 673 | parent = *p; |
| 674 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, |
| 675 | tree_node); |
| 676 | if (mz->usage_in_excess < mz_node->usage_in_excess) |
| 677 | p = &(*p)->rb_left; |
| 678 | /* |
| 679 | * We can't avoid mem cgroups that are over their soft |
| 680 | * limit by the same amount |
| 681 | */ |
| 682 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) |
| 683 | p = &(*p)->rb_right; |
| 684 | } |
| 685 | rb_link_node(&mz->tree_node, parent, p); |
| 686 | rb_insert_color(&mz->tree_node, &mctz->rb_root); |
| 687 | mz->on_tree = true; |
| 688 | } |
| 689 | |
Johannes Weiner | cf2c812 | 2014-06-06 14:38:21 -0700 | [diff] [blame] | 690 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz, |
| 691 | struct mem_cgroup_tree_per_zone *mctz) |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 692 | { |
| 693 | if (!mz->on_tree) |
| 694 | return; |
| 695 | rb_erase(&mz->tree_node, &mctz->rb_root); |
| 696 | mz->on_tree = false; |
| 697 | } |
| 698 | |
Johannes Weiner | cf2c812 | 2014-06-06 14:38:21 -0700 | [diff] [blame] | 699 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone *mz, |
| 700 | struct mem_cgroup_tree_per_zone *mctz) |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 701 | { |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 702 | unsigned long flags; |
| 703 | |
| 704 | spin_lock_irqsave(&mctz->lock, flags); |
Johannes Weiner | cf2c812 | 2014-06-06 14:38:21 -0700 | [diff] [blame] | 705 | __mem_cgroup_remove_exceeded(mz, mctz); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 706 | spin_unlock_irqrestore(&mctz->lock, flags); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 707 | } |
| 708 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 709 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
| 710 | { |
| 711 | unsigned long nr_pages = page_counter_read(&memcg->memory); |
Jason Low | 4db0c3c | 2015-04-15 16:14:08 -0700 | [diff] [blame] | 712 | unsigned long soft_limit = READ_ONCE(memcg->soft_limit); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 713 | unsigned long excess = 0; |
| 714 | |
| 715 | if (nr_pages > soft_limit) |
| 716 | excess = nr_pages - soft_limit; |
| 717 | |
| 718 | return excess; |
| 719 | } |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 720 | |
| 721 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) |
| 722 | { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 723 | unsigned long excess; |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 724 | struct mem_cgroup_per_zone *mz; |
| 725 | struct mem_cgroup_tree_per_zone *mctz; |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 726 | |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 727 | mctz = soft_limit_tree_from_page(page); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 728 | /* |
| 729 | * Necessary to update all ancestors when hierarchy is used. |
| 730 | * because their event counter is not touched. |
| 731 | */ |
| 732 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 733 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 734 | excess = soft_limit_excess(memcg); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 735 | /* |
| 736 | * We have to update the tree if mz is on RB-tree or |
| 737 | * mem is over its softlimit. |
| 738 | */ |
| 739 | if (excess || mz->on_tree) { |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 740 | unsigned long flags; |
| 741 | |
| 742 | spin_lock_irqsave(&mctz->lock, flags); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 743 | /* if on-tree, remove it */ |
| 744 | if (mz->on_tree) |
Johannes Weiner | cf2c812 | 2014-06-06 14:38:21 -0700 | [diff] [blame] | 745 | __mem_cgroup_remove_exceeded(mz, mctz); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 746 | /* |
| 747 | * Insert again. mz->usage_in_excess will be updated. |
| 748 | * If excess is 0, no tree ops. |
| 749 | */ |
Johannes Weiner | cf2c812 | 2014-06-06 14:38:21 -0700 | [diff] [blame] | 750 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 751 | spin_unlock_irqrestore(&mctz->lock, flags); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 752 | } |
| 753 | } |
| 754 | } |
| 755 | |
| 756 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) |
| 757 | { |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 758 | struct mem_cgroup_tree_per_zone *mctz; |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 759 | struct mem_cgroup_per_zone *mz; |
| 760 | int nid, zid; |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 761 | |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 762 | for_each_node(nid) { |
| 763 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
| 764 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; |
| 765 | mctz = soft_limit_tree_node_zone(nid, zid); |
Johannes Weiner | cf2c812 | 2014-06-06 14:38:21 -0700 | [diff] [blame] | 766 | mem_cgroup_remove_exceeded(mz, mctz); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 767 | } |
| 768 | } |
| 769 | } |
| 770 | |
| 771 | static struct mem_cgroup_per_zone * |
| 772 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) |
| 773 | { |
| 774 | struct rb_node *rightmost = NULL; |
| 775 | struct mem_cgroup_per_zone *mz; |
| 776 | |
| 777 | retry: |
| 778 | mz = NULL; |
| 779 | rightmost = rb_last(&mctz->rb_root); |
| 780 | if (!rightmost) |
| 781 | goto done; /* Nothing to reclaim from */ |
| 782 | |
| 783 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); |
| 784 | /* |
| 785 | * Remove the node now but someone else can add it back, |
| 786 | * we will to add it back at the end of reclaim to its correct |
| 787 | * position in the tree. |
| 788 | */ |
Johannes Weiner | cf2c812 | 2014-06-06 14:38:21 -0700 | [diff] [blame] | 789 | __mem_cgroup_remove_exceeded(mz, mctz); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 790 | if (!soft_limit_excess(mz->memcg) || |
Tejun Heo | ec903c0 | 2014-05-13 12:11:01 -0400 | [diff] [blame] | 791 | !css_tryget_online(&mz->memcg->css)) |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 792 | goto retry; |
| 793 | done: |
| 794 | return mz; |
| 795 | } |
| 796 | |
| 797 | static struct mem_cgroup_per_zone * |
| 798 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) |
| 799 | { |
| 800 | struct mem_cgroup_per_zone *mz; |
| 801 | |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 802 | spin_lock_irq(&mctz->lock); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 803 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 804 | spin_unlock_irq(&mctz->lock); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 805 | return mz; |
| 806 | } |
| 807 | |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 808 | /* |
| 809 | * Implementation Note: reading percpu statistics for memcg. |
| 810 | * |
| 811 | * Both of vmstat[] and percpu_counter has threshold and do periodic |
| 812 | * synchronization to implement "quick" read. There are trade-off between |
| 813 | * reading cost and precision of value. Then, we may have a chance to implement |
| 814 | * a periodic synchronizion of counter in memcg's counter. |
| 815 | * |
| 816 | * But this _read() function is used for user interface now. The user accounts |
| 817 | * memory usage by memory cgroup and he _always_ requires exact value because |
| 818 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always |
| 819 | * have to visit all online cpus and make sum. So, for now, unnecessary |
| 820 | * synchronization is not implemented. (just implemented for cpu hotplug) |
| 821 | * |
| 822 | * If there are kernel internal actions which can make use of some not-exact |
| 823 | * value, and reading all cpu value can be performance bottleneck in some |
| 824 | * common workload, threashold and synchonization as vmstat[] should be |
| 825 | * implemented. |
| 826 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 827 | static long mem_cgroup_read_stat(struct mem_cgroup *memcg, |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 828 | enum mem_cgroup_stat_index idx) |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 829 | { |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 830 | long val = 0; |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 831 | int cpu; |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 832 | |
Tejun Heo | 733a572 | 2015-05-22 18:23:18 -0400 | [diff] [blame] | 833 | for_each_possible_cpu(cpu) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 834 | val += per_cpu(memcg->stat->count[idx], cpu); |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 835 | return val; |
| 836 | } |
| 837 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 838 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
Johannes Weiner | e9f8974 | 2011-03-23 16:42:37 -0700 | [diff] [blame] | 839 | enum mem_cgroup_events_index idx) |
| 840 | { |
| 841 | unsigned long val = 0; |
| 842 | int cpu; |
| 843 | |
Tejun Heo | 733a572 | 2015-05-22 18:23:18 -0400 | [diff] [blame] | 844 | for_each_possible_cpu(cpu) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 845 | val += per_cpu(memcg->stat->events[idx], cpu); |
Johannes Weiner | e9f8974 | 2011-03-23 16:42:37 -0700 | [diff] [blame] | 846 | return val; |
| 847 | } |
| 848 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 849 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
David Rientjes | b070e65 | 2013-05-07 16:18:09 -0700 | [diff] [blame] | 850 | struct page *page, |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 851 | int nr_pages) |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 852 | { |
KAMEZAWA Hiroyuki | b2402857 | 2012-03-21 16:34:22 -0700 | [diff] [blame] | 853 | /* |
| 854 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is |
| 855 | * counted as CACHE even if it's on ANON LRU. |
| 856 | */ |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 857 | if (PageAnon(page)) |
KAMEZAWA Hiroyuki | b2402857 | 2012-03-21 16:34:22 -0700 | [diff] [blame] | 858 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 859 | nr_pages); |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 860 | else |
KAMEZAWA Hiroyuki | b2402857 | 2012-03-21 16:34:22 -0700 | [diff] [blame] | 861 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 862 | nr_pages); |
Balaji Rao | 55e462b | 2008-05-01 04:35:12 -0700 | [diff] [blame] | 863 | |
David Rientjes | b070e65 | 2013-05-07 16:18:09 -0700 | [diff] [blame] | 864 | if (PageTransHuge(page)) |
| 865 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
| 866 | nr_pages); |
| 867 | |
KAMEZAWA Hiroyuki | e401f176 | 2011-01-20 14:44:23 -0800 | [diff] [blame] | 868 | /* pagein of a big page is an event. So, ignore page size */ |
| 869 | if (nr_pages > 0) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 870 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
KAMEZAWA Hiroyuki | 3751d60 | 2011-02-01 15:52:45 -0800 | [diff] [blame] | 871 | else { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 872 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
KAMEZAWA Hiroyuki | 3751d60 | 2011-02-01 15:52:45 -0800 | [diff] [blame] | 873 | nr_pages = -nr_pages; /* for event */ |
| 874 | } |
KAMEZAWA Hiroyuki | e401f176 | 2011-01-20 14:44:23 -0800 | [diff] [blame] | 875 | |
Johannes Weiner | 1311471 | 2012-05-29 15:07:07 -0700 | [diff] [blame] | 876 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 877 | } |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 878 | |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 879 | unsigned long mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru) |
Konstantin Khlebnikov | 074291f | 2012-05-29 15:07:00 -0700 | [diff] [blame] | 880 | { |
| 881 | struct mem_cgroup_per_zone *mz; |
| 882 | |
| 883 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
| 884 | return mz->lru_size[lru]; |
| 885 | } |
| 886 | |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 887 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
| 888 | int nid, |
| 889 | unsigned int lru_mask) |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 890 | { |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 891 | unsigned long nr = 0; |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 892 | int zid; |
| 893 | |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 894 | VM_BUG_ON((unsigned)nid >= nr_node_ids); |
KAMEZAWA Hiroyuki | bb2a0de | 2011-07-26 16:08:22 -0700 | [diff] [blame] | 895 | |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 896 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
| 897 | struct mem_cgroup_per_zone *mz; |
| 898 | enum lru_list lru; |
| 899 | |
| 900 | for_each_lru(lru) { |
| 901 | if (!(BIT(lru) & lru_mask)) |
| 902 | continue; |
| 903 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; |
| 904 | nr += mz->lru_size[lru]; |
| 905 | } |
| 906 | } |
| 907 | return nr; |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 908 | } |
KAMEZAWA Hiroyuki | bb2a0de | 2011-07-26 16:08:22 -0700 | [diff] [blame] | 909 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 910 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
KAMEZAWA Hiroyuki | bb2a0de | 2011-07-26 16:08:22 -0700 | [diff] [blame] | 911 | unsigned int lru_mask) |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 912 | { |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 913 | unsigned long nr = 0; |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 914 | int nid; |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 915 | |
Lai Jiangshan | 31aaea4 | 2012-12-12 13:51:27 -0800 | [diff] [blame] | 916 | for_each_node_state(nid, N_MEMORY) |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 917 | nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
| 918 | return nr; |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 919 | } |
| 920 | |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 921 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
| 922 | enum mem_cgroup_events_target target) |
KAMEZAWA Hiroyuki | d2265e6 | 2010-03-10 15:22:31 -0800 | [diff] [blame] | 923 | { |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 924 | unsigned long val, next; |
| 925 | |
Johannes Weiner | 1311471 | 2012-05-29 15:07:07 -0700 | [diff] [blame] | 926 | val = __this_cpu_read(memcg->stat->nr_page_events); |
Steven Rostedt | 4799401 | 2011-11-02 13:38:33 -0700 | [diff] [blame] | 927 | next = __this_cpu_read(memcg->stat->targets[target]); |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 928 | /* from time_after() in jiffies.h */ |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 929 | if ((long)next - (long)val < 0) { |
| 930 | switch (target) { |
| 931 | case MEM_CGROUP_TARGET_THRESH: |
| 932 | next = val + THRESHOLDS_EVENTS_TARGET; |
| 933 | break; |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 934 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
| 935 | next = val + SOFTLIMIT_EVENTS_TARGET; |
| 936 | break; |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 937 | case MEM_CGROUP_TARGET_NUMAINFO: |
| 938 | next = val + NUMAINFO_EVENTS_TARGET; |
| 939 | break; |
| 940 | default: |
| 941 | break; |
| 942 | } |
| 943 | __this_cpu_write(memcg->stat->targets[target], next); |
| 944 | return true; |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 945 | } |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 946 | return false; |
KAMEZAWA Hiroyuki | d2265e6 | 2010-03-10 15:22:31 -0800 | [diff] [blame] | 947 | } |
| 948 | |
| 949 | /* |
| 950 | * Check events in order. |
| 951 | * |
| 952 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 953 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
KAMEZAWA Hiroyuki | d2265e6 | 2010-03-10 15:22:31 -0800 | [diff] [blame] | 954 | { |
| 955 | /* threshold event is triggered in finer grain than soft limit */ |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 956 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
| 957 | MEM_CGROUP_TARGET_THRESH))) { |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 958 | bool do_softlimit; |
Andrew Morton | 82b3f2a | 2012-02-03 15:37:14 -0800 | [diff] [blame] | 959 | bool do_numainfo __maybe_unused; |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 960 | |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 961 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
| 962 | MEM_CGROUP_TARGET_SOFTLIMIT); |
KAMEZAWA Hiroyuki | 453a9bf | 2011-07-08 15:39:43 -0700 | [diff] [blame] | 963 | #if MAX_NUMNODES > 1 |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 964 | do_numainfo = mem_cgroup_event_ratelimit(memcg, |
| 965 | MEM_CGROUP_TARGET_NUMAINFO); |
KAMEZAWA Hiroyuki | 453a9bf | 2011-07-08 15:39:43 -0700 | [diff] [blame] | 966 | #endif |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 967 | mem_cgroup_threshold(memcg); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 968 | if (unlikely(do_softlimit)) |
| 969 | mem_cgroup_update_tree(memcg, page); |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 970 | #if MAX_NUMNODES > 1 |
| 971 | if (unlikely(do_numainfo)) |
| 972 | atomic_inc(&memcg->numainfo_events); |
| 973 | #endif |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 974 | } |
KAMEZAWA Hiroyuki | d2265e6 | 2010-03-10 15:22:31 -0800 | [diff] [blame] | 975 | } |
| 976 | |
Balbir Singh | cf475ad | 2008-04-29 01:00:16 -0700 | [diff] [blame] | 977 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 978 | { |
Balbir Singh | 31a78f2 | 2008-09-28 23:09:31 +0100 | [diff] [blame] | 979 | /* |
| 980 | * mm_update_next_owner() may clear mm->owner to NULL |
| 981 | * if it races with swapoff, page migration, etc. |
| 982 | * So this can be called with p == NULL. |
| 983 | */ |
| 984 | if (unlikely(!p)) |
| 985 | return NULL; |
| 986 | |
Tejun Heo | 073219e | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 987 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 988 | } |
| 989 | |
Johannes Weiner | df38197 | 2014-04-07 15:37:43 -0700 | [diff] [blame] | 990 | static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 991 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 992 | struct mem_cgroup *memcg = NULL; |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 993 | |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 994 | rcu_read_lock(); |
| 995 | do { |
Michal Hocko | 6f6acb0 | 2014-05-22 11:54:19 -0700 | [diff] [blame] | 996 | /* |
| 997 | * Page cache insertions can happen withou an |
| 998 | * actual mm context, e.g. during disk probing |
| 999 | * on boot, loopback IO, acct() writes etc. |
| 1000 | */ |
| 1001 | if (unlikely(!mm)) |
Johannes Weiner | df38197 | 2014-04-07 15:37:43 -0700 | [diff] [blame] | 1002 | memcg = root_mem_cgroup; |
Michal Hocko | 6f6acb0 | 2014-05-22 11:54:19 -0700 | [diff] [blame] | 1003 | else { |
| 1004 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
| 1005 | if (unlikely(!memcg)) |
| 1006 | memcg = root_mem_cgroup; |
| 1007 | } |
Tejun Heo | ec903c0 | 2014-05-13 12:11:01 -0400 | [diff] [blame] | 1008 | } while (!css_tryget_online(&memcg->css)); |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 1009 | rcu_read_unlock(); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1010 | return memcg; |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 1011 | } |
| 1012 | |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 1013 | /** |
| 1014 | * mem_cgroup_iter - iterate over memory cgroup hierarchy |
| 1015 | * @root: hierarchy root |
| 1016 | * @prev: previously returned memcg, NULL on first invocation |
| 1017 | * @reclaim: cookie for shared reclaim walks, NULL for full walks |
| 1018 | * |
| 1019 | * Returns references to children of the hierarchy below @root, or |
| 1020 | * @root itself, or %NULL after a full round-trip. |
| 1021 | * |
| 1022 | * Caller must pass the return value in @prev on subsequent |
| 1023 | * invocations for reference counting, or use mem_cgroup_iter_break() |
| 1024 | * to cancel a hierarchy walk before the round-trip is complete. |
| 1025 | * |
| 1026 | * Reclaimers can specify a zone and a priority level in @reclaim to |
| 1027 | * divide up the memcgs in the hierarchy among all concurrent |
| 1028 | * reclaimers operating on the same zone and priority. |
| 1029 | */ |
Andrew Morton | 694fbc0 | 2013-09-24 15:27:37 -0700 | [diff] [blame] | 1030 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 1031 | struct mem_cgroup *prev, |
Andrew Morton | 694fbc0 | 2013-09-24 15:27:37 -0700 | [diff] [blame] | 1032 | struct mem_cgroup_reclaim_cookie *reclaim) |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 1033 | { |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 1034 | struct reclaim_iter *uninitialized_var(iter); |
| 1035 | struct cgroup_subsys_state *css = NULL; |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1036 | struct mem_cgroup *memcg = NULL; |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 1037 | struct mem_cgroup *pos = NULL; |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1038 | |
Andrew Morton | 694fbc0 | 2013-09-24 15:27:37 -0700 | [diff] [blame] | 1039 | if (mem_cgroup_disabled()) |
| 1040 | return NULL; |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 1041 | |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 1042 | if (!root) |
| 1043 | root = root_mem_cgroup; |
| 1044 | |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1045 | if (prev && !reclaim) |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 1046 | pos = prev; |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1047 | |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1048 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
| 1049 | if (prev) |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 1050 | goto out; |
Andrew Morton | 694fbc0 | 2013-09-24 15:27:37 -0700 | [diff] [blame] | 1051 | return root; |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1052 | } |
| 1053 | |
Michal Hocko | 542f85f | 2013-04-29 15:07:15 -0700 | [diff] [blame] | 1054 | rcu_read_lock(); |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1055 | |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 1056 | if (reclaim) { |
| 1057 | struct mem_cgroup_per_zone *mz; |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 1058 | |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 1059 | mz = mem_cgroup_zone_zoneinfo(root, reclaim->zone); |
| 1060 | iter = &mz->iter[reclaim->priority]; |
Michal Hocko | 5f57816 | 2013-04-29 15:07:17 -0700 | [diff] [blame] | 1061 | |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 1062 | if (prev && reclaim->generation != iter->generation) |
Michal Hocko | 542f85f | 2013-04-29 15:07:15 -0700 | [diff] [blame] | 1063 | goto out_unlock; |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 1064 | |
| 1065 | do { |
Jason Low | 4db0c3c | 2015-04-15 16:14:08 -0700 | [diff] [blame] | 1066 | pos = READ_ONCE(iter->position); |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 1067 | /* |
| 1068 | * A racing update may change the position and |
| 1069 | * put the last reference, hence css_tryget(), |
| 1070 | * or retry to see the updated position. |
| 1071 | */ |
| 1072 | } while (pos && !css_tryget(&pos->css)); |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1073 | } |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 1074 | |
| 1075 | if (pos) |
| 1076 | css = &pos->css; |
| 1077 | |
| 1078 | for (;;) { |
| 1079 | css = css_next_descendant_pre(css, &root->css); |
| 1080 | if (!css) { |
| 1081 | /* |
| 1082 | * Reclaimers share the hierarchy walk, and a |
| 1083 | * new one might jump in right at the end of |
| 1084 | * the hierarchy - make sure they see at least |
| 1085 | * one group and restart from the beginning. |
| 1086 | */ |
| 1087 | if (!prev) |
| 1088 | continue; |
| 1089 | break; |
| 1090 | } |
| 1091 | |
| 1092 | /* |
| 1093 | * Verify the css and acquire a reference. The root |
| 1094 | * is provided by the caller, so we know it's alive |
| 1095 | * and kicking, and don't take an extra reference. |
| 1096 | */ |
| 1097 | memcg = mem_cgroup_from_css(css); |
| 1098 | |
| 1099 | if (css == &root->css) |
| 1100 | break; |
| 1101 | |
Johannes Weiner | b205256 | 2014-12-10 15:42:48 -0800 | [diff] [blame] | 1102 | if (css_tryget(css)) { |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 1103 | /* |
| 1104 | * Make sure the memcg is initialized: |
| 1105 | * mem_cgroup_css_online() orders the the |
| 1106 | * initialization against setting the flag. |
| 1107 | */ |
| 1108 | if (smp_load_acquire(&memcg->initialized)) |
| 1109 | break; |
| 1110 | |
| 1111 | css_put(css); |
| 1112 | } |
| 1113 | |
| 1114 | memcg = NULL; |
| 1115 | } |
| 1116 | |
| 1117 | if (reclaim) { |
| 1118 | if (cmpxchg(&iter->position, pos, memcg) == pos) { |
| 1119 | if (memcg) |
| 1120 | css_get(&memcg->css); |
| 1121 | if (pos) |
| 1122 | css_put(&pos->css); |
| 1123 | } |
| 1124 | |
| 1125 | /* |
| 1126 | * pairs with css_tryget when dereferencing iter->position |
| 1127 | * above. |
| 1128 | */ |
| 1129 | if (pos) |
| 1130 | css_put(&pos->css); |
| 1131 | |
| 1132 | if (!memcg) |
| 1133 | iter->generation++; |
| 1134 | else if (!prev) |
| 1135 | reclaim->generation = iter->generation; |
| 1136 | } |
| 1137 | |
Michal Hocko | 542f85f | 2013-04-29 15:07:15 -0700 | [diff] [blame] | 1138 | out_unlock: |
| 1139 | rcu_read_unlock(); |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 1140 | out: |
Michal Hocko | c40046f | 2013-04-29 15:07:14 -0700 | [diff] [blame] | 1141 | if (prev && prev != root) |
| 1142 | css_put(&prev->css); |
| 1143 | |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1144 | return memcg; |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1145 | } |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1146 | |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 1147 | /** |
| 1148 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely |
| 1149 | * @root: hierarchy root |
| 1150 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() |
| 1151 | */ |
| 1152 | void mem_cgroup_iter_break(struct mem_cgroup *root, |
| 1153 | struct mem_cgroup *prev) |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1154 | { |
| 1155 | if (!root) |
| 1156 | root = root_mem_cgroup; |
| 1157 | if (prev && prev != root) |
| 1158 | css_put(&prev->css); |
| 1159 | } |
| 1160 | |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1161 | /* |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1162 | * Iteration constructs for visiting all cgroups (under a tree). If |
| 1163 | * loops are exited prematurely (break), mem_cgroup_iter_break() must |
| 1164 | * be used for reference counting. |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1165 | */ |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1166 | #define for_each_mem_cgroup_tree(iter, root) \ |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 1167 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1168 | iter != NULL; \ |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 1169 | iter = mem_cgroup_iter(root, iter, NULL)) |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1170 | |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1171 | #define for_each_mem_cgroup(iter) \ |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 1172 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1173 | iter != NULL; \ |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 1174 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1175 | |
David Rientjes | 68ae564 | 2012-12-12 13:51:57 -0800 | [diff] [blame] | 1176 | void __mem_cgroup_count_vm_event(struct mm_struct *mm, enum vm_event_item idx) |
Ying Han | 456f998 | 2011-05-26 16:25:38 -0700 | [diff] [blame] | 1177 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1178 | struct mem_cgroup *memcg; |
Ying Han | 456f998 | 2011-05-26 16:25:38 -0700 | [diff] [blame] | 1179 | |
Ying Han | 456f998 | 2011-05-26 16:25:38 -0700 | [diff] [blame] | 1180 | rcu_read_lock(); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1181 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
| 1182 | if (unlikely(!memcg)) |
Ying Han | 456f998 | 2011-05-26 16:25:38 -0700 | [diff] [blame] | 1183 | goto out; |
| 1184 | |
| 1185 | switch (idx) { |
Ying Han | 456f998 | 2011-05-26 16:25:38 -0700 | [diff] [blame] | 1186 | case PGFAULT: |
Johannes Weiner | 0e574a9 | 2012-01-12 17:18:35 -0800 | [diff] [blame] | 1187 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]); |
| 1188 | break; |
| 1189 | case PGMAJFAULT: |
| 1190 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]); |
Ying Han | 456f998 | 2011-05-26 16:25:38 -0700 | [diff] [blame] | 1191 | break; |
| 1192 | default: |
| 1193 | BUG(); |
| 1194 | } |
| 1195 | out: |
| 1196 | rcu_read_unlock(); |
| 1197 | } |
David Rientjes | 68ae564 | 2012-12-12 13:51:57 -0800 | [diff] [blame] | 1198 | EXPORT_SYMBOL(__mem_cgroup_count_vm_event); |
Ying Han | 456f998 | 2011-05-26 16:25:38 -0700 | [diff] [blame] | 1199 | |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1200 | /** |
| 1201 | * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg |
| 1202 | * @zone: zone of the wanted lruvec |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1203 | * @memcg: memcg of the wanted lruvec |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1204 | * |
| 1205 | * Returns the lru list vector holding pages for the given @zone and |
| 1206 | * @mem. This can be the global zone lruvec, if the memory controller |
| 1207 | * is disabled. |
| 1208 | */ |
| 1209 | struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone, |
| 1210 | struct mem_cgroup *memcg) |
| 1211 | { |
| 1212 | struct mem_cgroup_per_zone *mz; |
Hugh Dickins | bea8c15 | 2012-11-16 14:14:54 -0800 | [diff] [blame] | 1213 | struct lruvec *lruvec; |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1214 | |
Hugh Dickins | bea8c15 | 2012-11-16 14:14:54 -0800 | [diff] [blame] | 1215 | if (mem_cgroup_disabled()) { |
| 1216 | lruvec = &zone->lruvec; |
| 1217 | goto out; |
| 1218 | } |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1219 | |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 1220 | mz = mem_cgroup_zone_zoneinfo(memcg, zone); |
Hugh Dickins | bea8c15 | 2012-11-16 14:14:54 -0800 | [diff] [blame] | 1221 | lruvec = &mz->lruvec; |
| 1222 | out: |
| 1223 | /* |
| 1224 | * Since a node can be onlined after the mem_cgroup was created, |
| 1225 | * we have to be prepared to initialize lruvec->zone here; |
| 1226 | * and if offlined then reonlined, we need to reinitialize it. |
| 1227 | */ |
| 1228 | if (unlikely(lruvec->zone != zone)) |
| 1229 | lruvec->zone = zone; |
| 1230 | return lruvec; |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1231 | } |
| 1232 | |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1233 | /** |
Johannes Weiner | dfe0e77 | 2014-12-10 15:43:43 -0800 | [diff] [blame] | 1234 | * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1235 | * @page: the page |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1236 | * @zone: zone of the page |
Johannes Weiner | dfe0e77 | 2014-12-10 15:43:43 -0800 | [diff] [blame] | 1237 | * |
| 1238 | * This function is only safe when following the LRU page isolation |
| 1239 | * and putback protocol: the LRU lock must be held, and the page must |
| 1240 | * either be PageLRU() or the caller must have isolated/allocated it. |
Minchan Kim | 3f58a82 | 2011-03-22 16:32:53 -0700 | [diff] [blame] | 1241 | */ |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1242 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone) |
Minchan Kim | 3f58a82 | 2011-03-22 16:32:53 -0700 | [diff] [blame] | 1243 | { |
| 1244 | struct mem_cgroup_per_zone *mz; |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1245 | struct mem_cgroup *memcg; |
Hugh Dickins | bea8c15 | 2012-11-16 14:14:54 -0800 | [diff] [blame] | 1246 | struct lruvec *lruvec; |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 1247 | |
Hugh Dickins | bea8c15 | 2012-11-16 14:14:54 -0800 | [diff] [blame] | 1248 | if (mem_cgroup_disabled()) { |
| 1249 | lruvec = &zone->lruvec; |
| 1250 | goto out; |
| 1251 | } |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 1252 | |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 1253 | memcg = page->mem_cgroup; |
Hugh Dickins | 7512102 | 2012-03-05 14:59:18 -0800 | [diff] [blame] | 1254 | /* |
Johannes Weiner | dfe0e77 | 2014-12-10 15:43:43 -0800 | [diff] [blame] | 1255 | * Swapcache readahead pages are added to the LRU - and |
Johannes Weiner | 2983331 | 2014-12-10 15:44:02 -0800 | [diff] [blame] | 1256 | * possibly migrated - before they are charged. |
Hugh Dickins | 7512102 | 2012-03-05 14:59:18 -0800 | [diff] [blame] | 1257 | */ |
Johannes Weiner | 2983331 | 2014-12-10 15:44:02 -0800 | [diff] [blame] | 1258 | if (!memcg) |
| 1259 | memcg = root_mem_cgroup; |
Hugh Dickins | 7512102 | 2012-03-05 14:59:18 -0800 | [diff] [blame] | 1260 | |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 1261 | mz = mem_cgroup_page_zoneinfo(memcg, page); |
Hugh Dickins | bea8c15 | 2012-11-16 14:14:54 -0800 | [diff] [blame] | 1262 | lruvec = &mz->lruvec; |
| 1263 | out: |
| 1264 | /* |
| 1265 | * Since a node can be onlined after the mem_cgroup was created, |
| 1266 | * we have to be prepared to initialize lruvec->zone here; |
| 1267 | * and if offlined then reonlined, we need to reinitialize it. |
| 1268 | */ |
| 1269 | if (unlikely(lruvec->zone != zone)) |
| 1270 | lruvec->zone = zone; |
| 1271 | return lruvec; |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1272 | } |
| 1273 | |
| 1274 | /** |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1275 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
| 1276 | * @lruvec: mem_cgroup per zone lru vector |
| 1277 | * @lru: index of lru list the page is sitting on |
| 1278 | * @nr_pages: positive when adding or negative when removing |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1279 | * |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1280 | * This function must be called when a page is added to or removed from an |
| 1281 | * lru list. |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1282 | */ |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1283 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
| 1284 | int nr_pages) |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1285 | { |
| 1286 | struct mem_cgroup_per_zone *mz; |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1287 | unsigned long *lru_size; |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1288 | |
| 1289 | if (mem_cgroup_disabled()) |
| 1290 | return; |
| 1291 | |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1292 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
| 1293 | lru_size = mz->lru_size + lru; |
| 1294 | *lru_size += nr_pages; |
| 1295 | VM_BUG_ON((long)(*lru_size) < 0); |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 1296 | } |
KAMEZAWA Hiroyuki | 544122e | 2009-01-07 18:08:34 -0800 | [diff] [blame] | 1297 | |
Johannes Weiner | 2314b42 | 2014-12-10 15:44:33 -0800 | [diff] [blame] | 1298 | bool mem_cgroup_is_descendant(struct mem_cgroup *memcg, struct mem_cgroup *root) |
Michal Hocko | 3e92041 | 2011-07-26 16:08:29 -0700 | [diff] [blame] | 1299 | { |
Johannes Weiner | 2314b42 | 2014-12-10 15:44:33 -0800 | [diff] [blame] | 1300 | if (root == memcg) |
Johannes Weiner | 91c63734 | 2012-05-29 15:06:24 -0700 | [diff] [blame] | 1301 | return true; |
Johannes Weiner | 2314b42 | 2014-12-10 15:44:33 -0800 | [diff] [blame] | 1302 | if (!root->use_hierarchy) |
Johannes Weiner | 91c63734 | 2012-05-29 15:06:24 -0700 | [diff] [blame] | 1303 | return false; |
Johannes Weiner | 2314b42 | 2014-12-10 15:44:33 -0800 | [diff] [blame] | 1304 | return cgroup_is_descendant(memcg->css.cgroup, root->css.cgroup); |
Johannes Weiner | c3ac9a8 | 2012-05-29 15:06:25 -0700 | [diff] [blame] | 1305 | } |
| 1306 | |
Johannes Weiner | 2314b42 | 2014-12-10 15:44:33 -0800 | [diff] [blame] | 1307 | bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg) |
Johannes Weiner | c3ac9a8 | 2012-05-29 15:06:25 -0700 | [diff] [blame] | 1308 | { |
Johannes Weiner | 2314b42 | 2014-12-10 15:44:33 -0800 | [diff] [blame] | 1309 | struct mem_cgroup *task_memcg; |
KAMEZAWA Hiroyuki | 158e0a2 | 2010-08-10 18:03:00 -0700 | [diff] [blame] | 1310 | struct task_struct *p; |
David Rientjes | ffbdccf | 2013-07-03 15:01:23 -0700 | [diff] [blame] | 1311 | bool ret; |
David Rientjes | 4c4a221 | 2008-02-07 00:14:06 -0800 | [diff] [blame] | 1312 | |
KAMEZAWA Hiroyuki | 158e0a2 | 2010-08-10 18:03:00 -0700 | [diff] [blame] | 1313 | p = find_lock_task_mm(task); |
David Rientjes | de077d2 | 2012-01-12 17:18:52 -0800 | [diff] [blame] | 1314 | if (p) { |
Johannes Weiner | 2314b42 | 2014-12-10 15:44:33 -0800 | [diff] [blame] | 1315 | task_memcg = get_mem_cgroup_from_mm(p->mm); |
David Rientjes | de077d2 | 2012-01-12 17:18:52 -0800 | [diff] [blame] | 1316 | task_unlock(p); |
| 1317 | } else { |
| 1318 | /* |
| 1319 | * All threads may have already detached their mm's, but the oom |
| 1320 | * killer still needs to detect if they have already been oom |
| 1321 | * killed to prevent needlessly killing additional tasks. |
| 1322 | */ |
David Rientjes | ffbdccf | 2013-07-03 15:01:23 -0700 | [diff] [blame] | 1323 | rcu_read_lock(); |
Johannes Weiner | 2314b42 | 2014-12-10 15:44:33 -0800 | [diff] [blame] | 1324 | task_memcg = mem_cgroup_from_task(task); |
| 1325 | css_get(&task_memcg->css); |
David Rientjes | ffbdccf | 2013-07-03 15:01:23 -0700 | [diff] [blame] | 1326 | rcu_read_unlock(); |
David Rientjes | de077d2 | 2012-01-12 17:18:52 -0800 | [diff] [blame] | 1327 | } |
Johannes Weiner | 2314b42 | 2014-12-10 15:44:33 -0800 | [diff] [blame] | 1328 | ret = mem_cgroup_is_descendant(task_memcg, memcg); |
| 1329 | css_put(&task_memcg->css); |
David Rientjes | 4c4a221 | 2008-02-07 00:14:06 -0800 | [diff] [blame] | 1330 | return ret; |
| 1331 | } |
| 1332 | |
Konstantin Khlebnikov | c56d5c7 | 2012-05-29 15:07:00 -0700 | [diff] [blame] | 1333 | int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec) |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 1334 | { |
KOSAKI Motohiro | c772be9 | 2009-01-07 18:08:25 -0800 | [diff] [blame] | 1335 | unsigned long inactive_ratio; |
Johannes Weiner | 9b27297 | 2011-11-02 13:38:23 -0700 | [diff] [blame] | 1336 | unsigned long inactive; |
| 1337 | unsigned long active; |
| 1338 | unsigned long gb; |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 1339 | |
Hugh Dickins | 4d7dcca | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 1340 | inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON); |
| 1341 | active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON); |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 1342 | |
KOSAKI Motohiro | c772be9 | 2009-01-07 18:08:25 -0800 | [diff] [blame] | 1343 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
| 1344 | if (gb) |
| 1345 | inactive_ratio = int_sqrt(10 * gb); |
| 1346 | else |
| 1347 | inactive_ratio = 1; |
| 1348 | |
Johannes Weiner | 9b27297 | 2011-11-02 13:38:23 -0700 | [diff] [blame] | 1349 | return inactive * inactive_ratio < active; |
KOSAKI Motohiro | c772be9 | 2009-01-07 18:08:25 -0800 | [diff] [blame] | 1350 | } |
| 1351 | |
Vladimir Davydov | 90cbc25 | 2015-02-11 15:25:55 -0800 | [diff] [blame] | 1352 | bool mem_cgroup_lruvec_online(struct lruvec *lruvec) |
| 1353 | { |
| 1354 | struct mem_cgroup_per_zone *mz; |
| 1355 | struct mem_cgroup *memcg; |
| 1356 | |
| 1357 | if (mem_cgroup_disabled()) |
| 1358 | return true; |
| 1359 | |
| 1360 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
| 1361 | memcg = mz->memcg; |
| 1362 | |
| 1363 | return !!(memcg->css.flags & CSS_ONLINE); |
| 1364 | } |
| 1365 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1366 | #define mem_cgroup_from_counter(counter, member) \ |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1367 | container_of(counter, struct mem_cgroup, member) |
| 1368 | |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 1369 | /** |
Johannes Weiner | 9d11ea9 | 2011-03-23 16:42:21 -0700 | [diff] [blame] | 1370 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
Wanpeng Li | dad7557 | 2012-06-20 12:53:01 -0700 | [diff] [blame] | 1371 | * @memcg: the memory cgroup |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 1372 | * |
Johannes Weiner | 9d11ea9 | 2011-03-23 16:42:21 -0700 | [diff] [blame] | 1373 | * Returns the maximum amount of memory @mem can be charged with, in |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 1374 | * pages. |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 1375 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1376 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 1377 | { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1378 | unsigned long margin = 0; |
| 1379 | unsigned long count; |
| 1380 | unsigned long limit; |
Johannes Weiner | 9d11ea9 | 2011-03-23 16:42:21 -0700 | [diff] [blame] | 1381 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1382 | count = page_counter_read(&memcg->memory); |
Jason Low | 4db0c3c | 2015-04-15 16:14:08 -0700 | [diff] [blame] | 1383 | limit = READ_ONCE(memcg->memory.limit); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1384 | if (count < limit) |
| 1385 | margin = limit - count; |
| 1386 | |
| 1387 | if (do_swap_account) { |
| 1388 | count = page_counter_read(&memcg->memsw); |
Jason Low | 4db0c3c | 2015-04-15 16:14:08 -0700 | [diff] [blame] | 1389 | limit = READ_ONCE(memcg->memsw.limit); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1390 | if (count <= limit) |
| 1391 | margin = min(margin, limit - count); |
| 1392 | } |
| 1393 | |
| 1394 | return margin; |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 1395 | } |
| 1396 | |
KAMEZAWA Hiroyuki | 1f4c025 | 2011-07-26 16:08:21 -0700 | [diff] [blame] | 1397 | int mem_cgroup_swappiness(struct mem_cgroup *memcg) |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 1398 | { |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 1399 | /* root ? */ |
Linus Torvalds | 14208b0 | 2014-06-09 15:03:33 -0700 | [diff] [blame] | 1400 | if (mem_cgroup_disabled() || !memcg->css.parent) |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 1401 | return vm_swappiness; |
| 1402 | |
Johannes Weiner | bf1ff26 | 2011-03-23 16:42:32 -0700 | [diff] [blame] | 1403 | return memcg->swappiness; |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 1404 | } |
| 1405 | |
KAMEZAWA Hiroyuki | 619d094 | 2012-03-21 16:34:23 -0700 | [diff] [blame] | 1406 | /* |
Qiang Huang | bdcbb65 | 2014-06-04 16:08:21 -0700 | [diff] [blame] | 1407 | * A routine for checking "mem" is under move_account() or not. |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1408 | * |
Qiang Huang | bdcbb65 | 2014-06-04 16:08:21 -0700 | [diff] [blame] | 1409 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
| 1410 | * moving cgroups. This is for waiting at high-memory pressure |
| 1411 | * caused by "move". |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1412 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1413 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1414 | { |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 1415 | struct mem_cgroup *from; |
| 1416 | struct mem_cgroup *to; |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1417 | bool ret = false; |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 1418 | /* |
| 1419 | * Unlike task_move routines, we access mc.to, mc.from not under |
| 1420 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. |
| 1421 | */ |
| 1422 | spin_lock(&mc.lock); |
| 1423 | from = mc.from; |
| 1424 | to = mc.to; |
| 1425 | if (!from) |
| 1426 | goto unlock; |
Michal Hocko | 3e92041 | 2011-07-26 16:08:29 -0700 | [diff] [blame] | 1427 | |
Johannes Weiner | 2314b42 | 2014-12-10 15:44:33 -0800 | [diff] [blame] | 1428 | ret = mem_cgroup_is_descendant(from, memcg) || |
| 1429 | mem_cgroup_is_descendant(to, memcg); |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 1430 | unlock: |
| 1431 | spin_unlock(&mc.lock); |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1432 | return ret; |
| 1433 | } |
| 1434 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1435 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1436 | { |
| 1437 | if (mc.moving_task && current != mc.moving_task) { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1438 | if (mem_cgroup_under_move(memcg)) { |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1439 | DEFINE_WAIT(wait); |
| 1440 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); |
| 1441 | /* moving charge context might have finished. */ |
| 1442 | if (mc.moving_task) |
| 1443 | schedule(); |
| 1444 | finish_wait(&mc.waitq, &wait); |
| 1445 | return true; |
| 1446 | } |
| 1447 | } |
| 1448 | return false; |
| 1449 | } |
| 1450 | |
Sha Zhengju | 58cf188 | 2013-02-22 16:32:05 -0800 | [diff] [blame] | 1451 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1452 | /** |
Sha Zhengju | 58cf188 | 2013-02-22 16:32:05 -0800 | [diff] [blame] | 1453 | * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1454 | * @memcg: The memory cgroup that went over limit |
| 1455 | * @p: Task that is going to be killed |
| 1456 | * |
| 1457 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is |
| 1458 | * enabled |
| 1459 | */ |
| 1460 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) |
| 1461 | { |
Tejun Heo | e61734c | 2014-02-12 09:29:50 -0500 | [diff] [blame] | 1462 | /* oom_info_lock ensures that parallel ooms do not interleave */ |
Michal Hocko | 08088cb | 2014-02-25 15:01:44 -0800 | [diff] [blame] | 1463 | static DEFINE_MUTEX(oom_info_lock); |
Sha Zhengju | 58cf188 | 2013-02-22 16:32:05 -0800 | [diff] [blame] | 1464 | struct mem_cgroup *iter; |
| 1465 | unsigned int i; |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1466 | |
Michal Hocko | 08088cb | 2014-02-25 15:01:44 -0800 | [diff] [blame] | 1467 | mutex_lock(&oom_info_lock); |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1468 | rcu_read_lock(); |
| 1469 | |
Balasubramani Vivekanandan | 2415b9f | 2015-04-14 15:48:18 -0700 | [diff] [blame] | 1470 | if (p) { |
| 1471 | pr_info("Task in "); |
| 1472 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); |
| 1473 | pr_cont(" killed as a result of limit of "); |
| 1474 | } else { |
| 1475 | pr_info("Memory limit reached of cgroup "); |
| 1476 | } |
| 1477 | |
Tejun Heo | e61734c | 2014-02-12 09:29:50 -0500 | [diff] [blame] | 1478 | pr_cont_cgroup_path(memcg->css.cgroup); |
Greg Thelen | 0346dad | 2015-01-26 12:58:38 -0800 | [diff] [blame] | 1479 | pr_cont("\n"); |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1480 | |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1481 | rcu_read_unlock(); |
| 1482 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1483 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
| 1484 | K((u64)page_counter_read(&memcg->memory)), |
| 1485 | K((u64)memcg->memory.limit), memcg->memory.failcnt); |
| 1486 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", |
| 1487 | K((u64)page_counter_read(&memcg->memsw)), |
| 1488 | K((u64)memcg->memsw.limit), memcg->memsw.failcnt); |
| 1489 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", |
| 1490 | K((u64)page_counter_read(&memcg->kmem)), |
| 1491 | K((u64)memcg->kmem.limit), memcg->kmem.failcnt); |
Sha Zhengju | 58cf188 | 2013-02-22 16:32:05 -0800 | [diff] [blame] | 1492 | |
| 1493 | for_each_mem_cgroup_tree(iter, memcg) { |
Tejun Heo | e61734c | 2014-02-12 09:29:50 -0500 | [diff] [blame] | 1494 | pr_info("Memory cgroup stats for "); |
| 1495 | pr_cont_cgroup_path(iter->css.cgroup); |
Sha Zhengju | 58cf188 | 2013-02-22 16:32:05 -0800 | [diff] [blame] | 1496 | pr_cont(":"); |
| 1497 | |
| 1498 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
| 1499 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
| 1500 | continue; |
| 1501 | pr_cont(" %s:%ldKB", mem_cgroup_stat_names[i], |
| 1502 | K(mem_cgroup_read_stat(iter, i))); |
| 1503 | } |
| 1504 | |
| 1505 | for (i = 0; i < NR_LRU_LISTS; i++) |
| 1506 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], |
| 1507 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); |
| 1508 | |
| 1509 | pr_cont("\n"); |
| 1510 | } |
Michal Hocko | 08088cb | 2014-02-25 15:01:44 -0800 | [diff] [blame] | 1511 | mutex_unlock(&oom_info_lock); |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1512 | } |
| 1513 | |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 1514 | /* |
| 1515 | * This function returns the number of memcg under hierarchy tree. Returns |
| 1516 | * 1(self count) if no children. |
| 1517 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1518 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 1519 | { |
| 1520 | int num = 0; |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1521 | struct mem_cgroup *iter; |
| 1522 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1523 | for_each_mem_cgroup_tree(iter, memcg) |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1524 | num++; |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 1525 | return num; |
| 1526 | } |
| 1527 | |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1528 | /* |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 1529 | * Return the memory (and swap, if configured) limit for a memcg. |
| 1530 | */ |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1531 | static unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg) |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 1532 | { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1533 | unsigned long limit; |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 1534 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1535 | limit = memcg->memory.limit; |
Michal Hocko | 9a5a8f1 | 2012-11-16 14:14:49 -0800 | [diff] [blame] | 1536 | if (mem_cgroup_swappiness(memcg)) { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1537 | unsigned long memsw_limit; |
Michal Hocko | 9a5a8f1 | 2012-11-16 14:14:49 -0800 | [diff] [blame] | 1538 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1539 | memsw_limit = memcg->memsw.limit; |
| 1540 | limit = min(limit + total_swap_pages, memsw_limit); |
Michal Hocko | 9a5a8f1 | 2012-11-16 14:14:49 -0800 | [diff] [blame] | 1541 | } |
Michal Hocko | 9a5a8f1 | 2012-11-16 14:14:49 -0800 | [diff] [blame] | 1542 | return limit; |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 1543 | } |
| 1544 | |
David Rientjes | 1996546 | 2012-12-11 16:00:26 -0800 | [diff] [blame] | 1545 | static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
| 1546 | int order) |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 1547 | { |
| 1548 | struct mem_cgroup *iter; |
| 1549 | unsigned long chosen_points = 0; |
| 1550 | unsigned long totalpages; |
| 1551 | unsigned int points = 0; |
| 1552 | struct task_struct *chosen = NULL; |
| 1553 | |
Johannes Weiner | dc56401 | 2015-06-24 16:57:19 -0700 | [diff] [blame] | 1554 | mutex_lock(&oom_lock); |
| 1555 | |
David Rientjes | 876aafb | 2012-07-31 16:43:48 -0700 | [diff] [blame] | 1556 | /* |
David Rientjes | 465adcf | 2013-04-29 15:08:45 -0700 | [diff] [blame] | 1557 | * If current has a pending SIGKILL or is exiting, then automatically |
| 1558 | * select it. The goal is to allow it to allocate so that it may |
| 1559 | * quickly exit and free its memory. |
David Rientjes | 876aafb | 2012-07-31 16:43:48 -0700 | [diff] [blame] | 1560 | */ |
Oleg Nesterov | d003f37 | 2014-12-12 16:56:24 -0800 | [diff] [blame] | 1561 | if (fatal_signal_pending(current) || task_will_free_mem(current)) { |
Johannes Weiner | 16e9519 | 2015-06-24 16:57:07 -0700 | [diff] [blame] | 1562 | mark_oom_victim(current); |
Johannes Weiner | dc56401 | 2015-06-24 16:57:19 -0700 | [diff] [blame] | 1563 | goto unlock; |
David Rientjes | 876aafb | 2012-07-31 16:43:48 -0700 | [diff] [blame] | 1564 | } |
| 1565 | |
Balasubramani Vivekanandan | 2415b9f | 2015-04-14 15:48:18 -0700 | [diff] [blame] | 1566 | check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL, memcg); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1567 | totalpages = mem_cgroup_get_limit(memcg) ? : 1; |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 1568 | for_each_mem_cgroup_tree(iter, memcg) { |
Tejun Heo | 72ec702 | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 1569 | struct css_task_iter it; |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 1570 | struct task_struct *task; |
| 1571 | |
Tejun Heo | 72ec702 | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 1572 | css_task_iter_start(&iter->css, &it); |
| 1573 | while ((task = css_task_iter_next(&it))) { |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 1574 | switch (oom_scan_process_thread(task, totalpages, NULL, |
| 1575 | false)) { |
| 1576 | case OOM_SCAN_SELECT: |
| 1577 | if (chosen) |
| 1578 | put_task_struct(chosen); |
| 1579 | chosen = task; |
| 1580 | chosen_points = ULONG_MAX; |
| 1581 | get_task_struct(chosen); |
| 1582 | /* fall through */ |
| 1583 | case OOM_SCAN_CONTINUE: |
| 1584 | continue; |
| 1585 | case OOM_SCAN_ABORT: |
Tejun Heo | 72ec702 | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 1586 | css_task_iter_end(&it); |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 1587 | mem_cgroup_iter_break(memcg, iter); |
| 1588 | if (chosen) |
| 1589 | put_task_struct(chosen); |
Johannes Weiner | dc56401 | 2015-06-24 16:57:19 -0700 | [diff] [blame] | 1590 | goto unlock; |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 1591 | case OOM_SCAN_OK: |
| 1592 | break; |
| 1593 | }; |
| 1594 | points = oom_badness(task, memcg, NULL, totalpages); |
David Rientjes | d49ad93 | 2014-01-23 15:53:34 -0800 | [diff] [blame] | 1595 | if (!points || points < chosen_points) |
| 1596 | continue; |
| 1597 | /* Prefer thread group leaders for display purposes */ |
| 1598 | if (points == chosen_points && |
| 1599 | thread_group_leader(chosen)) |
| 1600 | continue; |
| 1601 | |
| 1602 | if (chosen) |
| 1603 | put_task_struct(chosen); |
| 1604 | chosen = task; |
| 1605 | chosen_points = points; |
| 1606 | get_task_struct(chosen); |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 1607 | } |
Tejun Heo | 72ec702 | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 1608 | css_task_iter_end(&it); |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 1609 | } |
| 1610 | |
Johannes Weiner | dc56401 | 2015-06-24 16:57:19 -0700 | [diff] [blame] | 1611 | if (chosen) { |
| 1612 | points = chosen_points * 1000 / totalpages; |
| 1613 | oom_kill_process(chosen, gfp_mask, order, points, totalpages, |
| 1614 | memcg, NULL, "Memory cgroup out of memory"); |
| 1615 | } |
| 1616 | unlock: |
| 1617 | mutex_unlock(&oom_lock); |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 1618 | } |
| 1619 | |
Michele Curti | ae6e71d | 2014-12-12 16:56:35 -0800 | [diff] [blame] | 1620 | #if MAX_NUMNODES > 1 |
| 1621 | |
KAMEZAWA Hiroyuki | 4d0c066 | 2011-07-08 15:39:42 -0700 | [diff] [blame] | 1622 | /** |
| 1623 | * test_mem_cgroup_node_reclaimable |
Wanpeng Li | dad7557 | 2012-06-20 12:53:01 -0700 | [diff] [blame] | 1624 | * @memcg: the target memcg |
KAMEZAWA Hiroyuki | 4d0c066 | 2011-07-08 15:39:42 -0700 | [diff] [blame] | 1625 | * @nid: the node ID to be checked. |
| 1626 | * @noswap : specify true here if the user wants flle only information. |
| 1627 | * |
| 1628 | * This function returns whether the specified memcg contains any |
| 1629 | * reclaimable pages on a node. Returns true if there are any reclaimable |
| 1630 | * pages in the node. |
| 1631 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1632 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
KAMEZAWA Hiroyuki | 4d0c066 | 2011-07-08 15:39:42 -0700 | [diff] [blame] | 1633 | int nid, bool noswap) |
| 1634 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1635 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
KAMEZAWA Hiroyuki | 4d0c066 | 2011-07-08 15:39:42 -0700 | [diff] [blame] | 1636 | return true; |
| 1637 | if (noswap || !total_swap_pages) |
| 1638 | return false; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1639 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
KAMEZAWA Hiroyuki | 4d0c066 | 2011-07-08 15:39:42 -0700 | [diff] [blame] | 1640 | return true; |
| 1641 | return false; |
| 1642 | |
| 1643 | } |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1644 | |
| 1645 | /* |
| 1646 | * Always updating the nodemask is not very good - even if we have an empty |
| 1647 | * list or the wrong list here, we can start from some node and traverse all |
| 1648 | * nodes based on the zonelist. So update the list loosely once per 10 secs. |
| 1649 | * |
| 1650 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1651 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1652 | { |
| 1653 | int nid; |
KAMEZAWA Hiroyuki | 453a9bf | 2011-07-08 15:39:43 -0700 | [diff] [blame] | 1654 | /* |
| 1655 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET |
| 1656 | * pagein/pageout changes since the last update. |
| 1657 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1658 | if (!atomic_read(&memcg->numainfo_events)) |
KAMEZAWA Hiroyuki | 453a9bf | 2011-07-08 15:39:43 -0700 | [diff] [blame] | 1659 | return; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1660 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1661 | return; |
| 1662 | |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1663 | /* make a nodemask where this memcg uses memory from */ |
Lai Jiangshan | 31aaea4 | 2012-12-12 13:51:27 -0800 | [diff] [blame] | 1664 | memcg->scan_nodes = node_states[N_MEMORY]; |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1665 | |
Lai Jiangshan | 31aaea4 | 2012-12-12 13:51:27 -0800 | [diff] [blame] | 1666 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1667 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1668 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
| 1669 | node_clear(nid, memcg->scan_nodes); |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1670 | } |
KAMEZAWA Hiroyuki | 453a9bf | 2011-07-08 15:39:43 -0700 | [diff] [blame] | 1671 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1672 | atomic_set(&memcg->numainfo_events, 0); |
| 1673 | atomic_set(&memcg->numainfo_updating, 0); |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1674 | } |
| 1675 | |
| 1676 | /* |
| 1677 | * Selecting a node where we start reclaim from. Because what we need is just |
| 1678 | * reducing usage counter, start from anywhere is O,K. Considering |
| 1679 | * memory reclaim from current node, there are pros. and cons. |
| 1680 | * |
| 1681 | * Freeing memory from current node means freeing memory from a node which |
| 1682 | * we'll use or we've used. So, it may make LRU bad. And if several threads |
| 1683 | * hit limits, it will see a contention on a node. But freeing from remote |
| 1684 | * node means more costs for memory reclaim because of memory latency. |
| 1685 | * |
| 1686 | * Now, we use round-robin. Better algorithm is welcomed. |
| 1687 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1688 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1689 | { |
| 1690 | int node; |
| 1691 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1692 | mem_cgroup_may_update_nodemask(memcg); |
| 1693 | node = memcg->last_scanned_node; |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1694 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1695 | node = next_node(node, memcg->scan_nodes); |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1696 | if (node == MAX_NUMNODES) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1697 | node = first_node(memcg->scan_nodes); |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1698 | /* |
| 1699 | * We call this when we hit limit, not when pages are added to LRU. |
| 1700 | * No LRU may hold pages because all pages are UNEVICTABLE or |
| 1701 | * memcg is too small and all pages are not on LRU. In that case, |
| 1702 | * we use curret node. |
| 1703 | */ |
| 1704 | if (unlikely(node == MAX_NUMNODES)) |
| 1705 | node = numa_node_id(); |
| 1706 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1707 | memcg->last_scanned_node = node; |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1708 | return node; |
| 1709 | } |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1710 | #else |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1711 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1712 | { |
| 1713 | return 0; |
| 1714 | } |
| 1715 | #endif |
| 1716 | |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 1717 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
| 1718 | struct zone *zone, |
| 1719 | gfp_t gfp_mask, |
| 1720 | unsigned long *total_scanned) |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1721 | { |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 1722 | struct mem_cgroup *victim = NULL; |
| 1723 | int total = 0; |
| 1724 | int loop = 0; |
| 1725 | unsigned long excess; |
| 1726 | unsigned long nr_scanned; |
| 1727 | struct mem_cgroup_reclaim_cookie reclaim = { |
| 1728 | .zone = zone, |
| 1729 | .priority = 0, |
| 1730 | }; |
Johannes Weiner | 9d11ea9 | 2011-03-23 16:42:21 -0700 | [diff] [blame] | 1731 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1732 | excess = soft_limit_excess(root_memcg); |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1733 | |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 1734 | while (1) { |
| 1735 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); |
| 1736 | if (!victim) { |
| 1737 | loop++; |
| 1738 | if (loop >= 2) { |
| 1739 | /* |
| 1740 | * If we have not been able to reclaim |
| 1741 | * anything, it might because there are |
| 1742 | * no reclaimable pages under this hierarchy |
| 1743 | */ |
| 1744 | if (!total) |
| 1745 | break; |
| 1746 | /* |
| 1747 | * We want to do more targeted reclaim. |
| 1748 | * excess >> 2 is not to excessive so as to |
| 1749 | * reclaim too much, nor too less that we keep |
| 1750 | * coming back to reclaim from this cgroup |
| 1751 | */ |
| 1752 | if (total >= (excess >> 2) || |
| 1753 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) |
| 1754 | break; |
| 1755 | } |
| 1756 | continue; |
| 1757 | } |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 1758 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, |
| 1759 | zone, &nr_scanned); |
| 1760 | *total_scanned += nr_scanned; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1761 | if (!soft_limit_excess(root_memcg)) |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 1762 | break; |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1763 | } |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 1764 | mem_cgroup_iter_break(root_memcg, victim); |
| 1765 | return total; |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1766 | } |
| 1767 | |
Johannes Weiner | 0056f4e | 2013-10-31 16:34:14 -0700 | [diff] [blame] | 1768 | #ifdef CONFIG_LOCKDEP |
| 1769 | static struct lockdep_map memcg_oom_lock_dep_map = { |
| 1770 | .name = "memcg_oom_lock", |
| 1771 | }; |
| 1772 | #endif |
| 1773 | |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1774 | static DEFINE_SPINLOCK(memcg_oom_lock); |
| 1775 | |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1776 | /* |
| 1777 | * Check OOM-Killer is already running under our hierarchy. |
| 1778 | * If someone is running, return false. |
| 1779 | */ |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1780 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1781 | { |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1782 | struct mem_cgroup *iter, *failed = NULL; |
KAMEZAWA Hiroyuki | a636b32 | 2009-01-07 18:08:08 -0800 | [diff] [blame] | 1783 | |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1784 | spin_lock(&memcg_oom_lock); |
| 1785 | |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1786 | for_each_mem_cgroup_tree(iter, memcg) { |
Johannes Weiner | 23751be | 2011-08-25 15:59:16 -0700 | [diff] [blame] | 1787 | if (iter->oom_lock) { |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1788 | /* |
| 1789 | * this subtree of our hierarchy is already locked |
| 1790 | * so we cannot give a lock. |
| 1791 | */ |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1792 | failed = iter; |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1793 | mem_cgroup_iter_break(memcg, iter); |
| 1794 | break; |
Johannes Weiner | 23751be | 2011-08-25 15:59:16 -0700 | [diff] [blame] | 1795 | } else |
| 1796 | iter->oom_lock = true; |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1797 | } |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1798 | |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1799 | if (failed) { |
| 1800 | /* |
| 1801 | * OK, we failed to lock the whole subtree so we have |
| 1802 | * to clean up what we set up to the failing subtree |
| 1803 | */ |
| 1804 | for_each_mem_cgroup_tree(iter, memcg) { |
| 1805 | if (iter == failed) { |
| 1806 | mem_cgroup_iter_break(memcg, iter); |
| 1807 | break; |
| 1808 | } |
| 1809 | iter->oom_lock = false; |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1810 | } |
Johannes Weiner | 0056f4e | 2013-10-31 16:34:14 -0700 | [diff] [blame] | 1811 | } else |
| 1812 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1813 | |
| 1814 | spin_unlock(&memcg_oom_lock); |
| 1815 | |
| 1816 | return !failed; |
KAMEZAWA Hiroyuki | a636b32 | 2009-01-07 18:08:08 -0800 | [diff] [blame] | 1817 | } |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 1818 | |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1819 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 1820 | { |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1821 | struct mem_cgroup *iter; |
| 1822 | |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1823 | spin_lock(&memcg_oom_lock); |
Johannes Weiner | 0056f4e | 2013-10-31 16:34:14 -0700 | [diff] [blame] | 1824 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1825 | for_each_mem_cgroup_tree(iter, memcg) |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1826 | iter->oom_lock = false; |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1827 | spin_unlock(&memcg_oom_lock); |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1828 | } |
| 1829 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1830 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1831 | { |
| 1832 | struct mem_cgroup *iter; |
| 1833 | |
Tejun Heo | c2b42d3 | 2015-06-24 16:58:23 -0700 | [diff] [blame] | 1834 | spin_lock(&memcg_oom_lock); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1835 | for_each_mem_cgroup_tree(iter, memcg) |
Tejun Heo | c2b42d3 | 2015-06-24 16:58:23 -0700 | [diff] [blame] | 1836 | iter->under_oom++; |
| 1837 | spin_unlock(&memcg_oom_lock); |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1838 | } |
| 1839 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1840 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1841 | { |
| 1842 | struct mem_cgroup *iter; |
| 1843 | |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1844 | /* |
| 1845 | * When a new child is created while the hierarchy is under oom, |
Tejun Heo | c2b42d3 | 2015-06-24 16:58:23 -0700 | [diff] [blame] | 1846 | * mem_cgroup_oom_lock() may not be called. Watch for underflow. |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1847 | */ |
Tejun Heo | c2b42d3 | 2015-06-24 16:58:23 -0700 | [diff] [blame] | 1848 | spin_lock(&memcg_oom_lock); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1849 | for_each_mem_cgroup_tree(iter, memcg) |
Tejun Heo | c2b42d3 | 2015-06-24 16:58:23 -0700 | [diff] [blame] | 1850 | if (iter->under_oom > 0) |
| 1851 | iter->under_oom--; |
| 1852 | spin_unlock(&memcg_oom_lock); |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 1853 | } |
| 1854 | |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1855 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
| 1856 | |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1857 | struct oom_wait_info { |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 1858 | struct mem_cgroup *memcg; |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1859 | wait_queue_t wait; |
| 1860 | }; |
| 1861 | |
| 1862 | static int memcg_oom_wake_function(wait_queue_t *wait, |
| 1863 | unsigned mode, int sync, void *arg) |
| 1864 | { |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 1865 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
| 1866 | struct mem_cgroup *oom_wait_memcg; |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1867 | struct oom_wait_info *oom_wait_info; |
| 1868 | |
| 1869 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 1870 | oom_wait_memcg = oom_wait_info->memcg; |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1871 | |
Johannes Weiner | 2314b42 | 2014-12-10 15:44:33 -0800 | [diff] [blame] | 1872 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
| 1873 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1874 | return 0; |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1875 | return autoremove_wake_function(wait, mode, sync, arg); |
| 1876 | } |
| 1877 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1878 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 1879 | { |
Tejun Heo | c2b42d3 | 2015-06-24 16:58:23 -0700 | [diff] [blame] | 1880 | /* |
| 1881 | * For the following lockless ->under_oom test, the only required |
| 1882 | * guarantee is that it must see the state asserted by an OOM when |
| 1883 | * this function is called as a result of userland actions |
| 1884 | * triggered by the notification of the OOM. This is trivially |
| 1885 | * achieved by invoking mem_cgroup_mark_under_oom() before |
| 1886 | * triggering notification. |
| 1887 | */ |
| 1888 | if (memcg && memcg->under_oom) |
Tejun Heo | f4b90b70 | 2015-06-24 16:58:21 -0700 | [diff] [blame] | 1889 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 1890 | } |
| 1891 | |
Johannes Weiner | 3812c8c | 2013-09-12 15:13:44 -0700 | [diff] [blame] | 1892 | static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1893 | { |
Johannes Weiner | 3812c8c | 2013-09-12 15:13:44 -0700 | [diff] [blame] | 1894 | if (!current->memcg_oom.may_oom) |
| 1895 | return; |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1896 | /* |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 1897 | * We are in the middle of the charge context here, so we |
| 1898 | * don't want to block when potentially sitting on a callstack |
| 1899 | * that holds all kinds of filesystem and mm locks. |
| 1900 | * |
| 1901 | * Also, the caller may handle a failed allocation gracefully |
| 1902 | * (like optional page cache readahead) and so an OOM killer |
| 1903 | * invocation might not even be necessary. |
| 1904 | * |
| 1905 | * That's why we don't do anything here except remember the |
| 1906 | * OOM context and then deal with it at the end of the page |
| 1907 | * fault when the stack is unwound, the locks are released, |
| 1908 | * and when we know whether the fault was overall successful. |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1909 | */ |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 1910 | css_get(&memcg->css); |
| 1911 | current->memcg_oom.memcg = memcg; |
| 1912 | current->memcg_oom.gfp_mask = mask; |
| 1913 | current->memcg_oom.order = order; |
| 1914 | } |
| 1915 | |
| 1916 | /** |
| 1917 | * mem_cgroup_oom_synchronize - complete memcg OOM handling |
| 1918 | * @handle: actually kill/wait or just clean up the OOM state |
| 1919 | * |
| 1920 | * This has to be called at the end of a page fault if the memcg OOM |
| 1921 | * handler was enabled. |
| 1922 | * |
| 1923 | * Memcg supports userspace OOM handling where failed allocations must |
| 1924 | * sleep on a waitqueue until the userspace task resolves the |
| 1925 | * situation. Sleeping directly in the charge context with all kinds |
| 1926 | * of locks held is not a good idea, instead we remember an OOM state |
| 1927 | * in the task and mem_cgroup_oom_synchronize() has to be called at |
| 1928 | * the end of the page fault to complete the OOM handling. |
| 1929 | * |
| 1930 | * Returns %true if an ongoing memcg OOM situation was detected and |
| 1931 | * completed, %false otherwise. |
| 1932 | */ |
| 1933 | bool mem_cgroup_oom_synchronize(bool handle) |
| 1934 | { |
| 1935 | struct mem_cgroup *memcg = current->memcg_oom.memcg; |
| 1936 | struct oom_wait_info owait; |
| 1937 | bool locked; |
| 1938 | |
| 1939 | /* OOM is global, do not handle */ |
| 1940 | if (!memcg) |
| 1941 | return false; |
| 1942 | |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 1943 | if (!handle || oom_killer_disabled) |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 1944 | goto cleanup; |
| 1945 | |
| 1946 | owait.memcg = memcg; |
| 1947 | owait.wait.flags = 0; |
| 1948 | owait.wait.func = memcg_oom_wake_function; |
| 1949 | owait.wait.private = current; |
| 1950 | INIT_LIST_HEAD(&owait.wait.task_list); |
| 1951 | |
| 1952 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1953 | mem_cgroup_mark_under_oom(memcg); |
| 1954 | |
| 1955 | locked = mem_cgroup_oom_trylock(memcg); |
| 1956 | |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 1957 | if (locked) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1958 | mem_cgroup_oom_notify(memcg); |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1959 | |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1960 | if (locked && !memcg->oom_kill_disable) { |
| 1961 | mem_cgroup_unmark_under_oom(memcg); |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 1962 | finish_wait(&memcg_oom_waitq, &owait.wait); |
| 1963 | mem_cgroup_out_of_memory(memcg, current->memcg_oom.gfp_mask, |
| 1964 | current->memcg_oom.order); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 1965 | } else { |
Johannes Weiner | 3812c8c | 2013-09-12 15:13:44 -0700 | [diff] [blame] | 1966 | schedule(); |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 1967 | mem_cgroup_unmark_under_oom(memcg); |
| 1968 | finish_wait(&memcg_oom_waitq, &owait.wait); |
| 1969 | } |
| 1970 | |
| 1971 | if (locked) { |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1972 | mem_cgroup_oom_unlock(memcg); |
| 1973 | /* |
| 1974 | * There is no guarantee that an OOM-lock contender |
| 1975 | * sees the wakeups triggered by the OOM kill |
| 1976 | * uncharges. Wake any sleepers explicitely. |
| 1977 | */ |
| 1978 | memcg_oom_recover(memcg); |
| 1979 | } |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 1980 | cleanup: |
| 1981 | current->memcg_oom.memcg = NULL; |
Johannes Weiner | 3812c8c | 2013-09-12 15:13:44 -0700 | [diff] [blame] | 1982 | css_put(&memcg->css); |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1983 | return true; |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 1984 | } |
| 1985 | |
Johannes Weiner | d7365e7 | 2014-10-29 14:50:48 -0700 | [diff] [blame] | 1986 | /** |
| 1987 | * mem_cgroup_begin_page_stat - begin a page state statistics transaction |
| 1988 | * @page: page that is going to change accounted state |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1989 | * |
Johannes Weiner | d7365e7 | 2014-10-29 14:50:48 -0700 | [diff] [blame] | 1990 | * This function must mark the beginning of an accounted page state |
| 1991 | * change to prevent double accounting when the page is concurrently |
| 1992 | * being moved to another memcg: |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1993 | * |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 1994 | * memcg = mem_cgroup_begin_page_stat(page); |
Johannes Weiner | d7365e7 | 2014-10-29 14:50:48 -0700 | [diff] [blame] | 1995 | * if (TestClearPageState(page)) |
| 1996 | * mem_cgroup_update_page_stat(memcg, state, -1); |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 1997 | * mem_cgroup_end_page_stat(memcg); |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 1998 | */ |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 1999 | struct mem_cgroup *mem_cgroup_begin_page_stat(struct page *page) |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 2000 | { |
| 2001 | struct mem_cgroup *memcg; |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 2002 | unsigned long flags; |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 2003 | |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 2004 | /* |
| 2005 | * The RCU lock is held throughout the transaction. The fast |
| 2006 | * path can get away without acquiring the memcg->move_lock |
| 2007 | * because page moving starts with an RCU grace period. |
| 2008 | * |
| 2009 | * The RCU lock also protects the memcg from being freed when |
| 2010 | * the page state that is going to change is the only thing |
| 2011 | * preventing the page from being uncharged. |
| 2012 | * E.g. end-writeback clearing PageWriteback(), which allows |
| 2013 | * migration to go ahead and uncharge the page before the |
| 2014 | * account transaction might be complete. |
| 2015 | */ |
Johannes Weiner | d7365e7 | 2014-10-29 14:50:48 -0700 | [diff] [blame] | 2016 | rcu_read_lock(); |
| 2017 | |
| 2018 | if (mem_cgroup_disabled()) |
| 2019 | return NULL; |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 2020 | again: |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 2021 | memcg = page->mem_cgroup; |
Johannes Weiner | 2983331 | 2014-12-10 15:44:02 -0800 | [diff] [blame] | 2022 | if (unlikely(!memcg)) |
Johannes Weiner | d7365e7 | 2014-10-29 14:50:48 -0700 | [diff] [blame] | 2023 | return NULL; |
| 2024 | |
Qiang Huang | bdcbb65 | 2014-06-04 16:08:21 -0700 | [diff] [blame] | 2025 | if (atomic_read(&memcg->moving_account) <= 0) |
Johannes Weiner | d7365e7 | 2014-10-29 14:50:48 -0700 | [diff] [blame] | 2026 | return memcg; |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 2027 | |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 2028 | spin_lock_irqsave(&memcg->move_lock, flags); |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 2029 | if (memcg != page->mem_cgroup) { |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 2030 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 2031 | goto again; |
| 2032 | } |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 2033 | |
| 2034 | /* |
| 2035 | * When charge migration first begins, we can have locked and |
| 2036 | * unlocked page stat updates happening concurrently. Track |
| 2037 | * the task who has the lock for mem_cgroup_end_page_stat(). |
| 2038 | */ |
| 2039 | memcg->move_lock_task = current; |
| 2040 | memcg->move_lock_flags = flags; |
Johannes Weiner | d7365e7 | 2014-10-29 14:50:48 -0700 | [diff] [blame] | 2041 | |
| 2042 | return memcg; |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 2043 | } |
Greg Thelen | c4843a7 | 2015-05-22 17:13:16 -0400 | [diff] [blame] | 2044 | EXPORT_SYMBOL(mem_cgroup_begin_page_stat); |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 2045 | |
Johannes Weiner | d7365e7 | 2014-10-29 14:50:48 -0700 | [diff] [blame] | 2046 | /** |
| 2047 | * mem_cgroup_end_page_stat - finish a page state statistics transaction |
| 2048 | * @memcg: the memcg that was accounted against |
Johannes Weiner | d7365e7 | 2014-10-29 14:50:48 -0700 | [diff] [blame] | 2049 | */ |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 2050 | void mem_cgroup_end_page_stat(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 2051 | { |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 2052 | if (memcg && memcg->move_lock_task == current) { |
| 2053 | unsigned long flags = memcg->move_lock_flags; |
| 2054 | |
| 2055 | memcg->move_lock_task = NULL; |
| 2056 | memcg->move_lock_flags = 0; |
| 2057 | |
| 2058 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
| 2059 | } |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 2060 | |
Johannes Weiner | d7365e7 | 2014-10-29 14:50:48 -0700 | [diff] [blame] | 2061 | rcu_read_unlock(); |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 2062 | } |
Greg Thelen | c4843a7 | 2015-05-22 17:13:16 -0400 | [diff] [blame] | 2063 | EXPORT_SYMBOL(mem_cgroup_end_page_stat); |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 2064 | |
Johannes Weiner | d7365e7 | 2014-10-29 14:50:48 -0700 | [diff] [blame] | 2065 | /** |
| 2066 | * mem_cgroup_update_page_stat - update page state statistics |
| 2067 | * @memcg: memcg to account against |
| 2068 | * @idx: page state item to account |
| 2069 | * @val: number of pages (positive or negative) |
| 2070 | * |
| 2071 | * See mem_cgroup_begin_page_stat() for locking requirements. |
| 2072 | */ |
| 2073 | void mem_cgroup_update_page_stat(struct mem_cgroup *memcg, |
Sha Zhengju | 68b4876 | 2013-09-12 15:13:50 -0700 | [diff] [blame] | 2074 | enum mem_cgroup_stat_index idx, int val) |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 2075 | { |
Sha Zhengju | 658b72c | 2013-09-12 15:13:52 -0700 | [diff] [blame] | 2076 | VM_BUG_ON(!rcu_read_lock_held()); |
KAMEZAWA Hiroyuki | 26174ef | 2010-10-27 15:33:43 -0700 | [diff] [blame] | 2077 | |
Johannes Weiner | d7365e7 | 2014-10-29 14:50:48 -0700 | [diff] [blame] | 2078 | if (memcg) |
| 2079 | this_cpu_add(memcg->stat->count[idx], val); |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 2080 | } |
KAMEZAWA Hiroyuki | 26174ef | 2010-10-27 15:33:43 -0700 | [diff] [blame] | 2081 | |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2082 | /* |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2083 | * size of first charge trial. "32" comes from vmscan.c's magic value. |
| 2084 | * TODO: maybe necessary to use big numbers in big irons. |
| 2085 | */ |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 2086 | #define CHARGE_BATCH 32U |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2087 | struct memcg_stock_pcp { |
| 2088 | struct mem_cgroup *cached; /* this never be root cgroup */ |
Johannes Weiner | 11c9ea4 | 2011-03-23 16:42:34 -0700 | [diff] [blame] | 2089 | unsigned int nr_pages; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2090 | struct work_struct work; |
KAMEZAWA Hiroyuki | 26fe616 | 2011-06-15 15:08:45 -0700 | [diff] [blame] | 2091 | unsigned long flags; |
Kirill A. Shutemov | a0db00f | 2012-05-29 15:06:56 -0700 | [diff] [blame] | 2092 | #define FLUSHING_CACHED_CHARGE 0 |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2093 | }; |
| 2094 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); |
Michal Hocko | 9f50fad | 2011-08-09 11:56:26 +0200 | [diff] [blame] | 2095 | static DEFINE_MUTEX(percpu_charge_mutex); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2096 | |
Suleiman Souhlal | a0956d5 | 2012-12-18 14:21:36 -0800 | [diff] [blame] | 2097 | /** |
| 2098 | * consume_stock: Try to consume stocked charge on this cpu. |
| 2099 | * @memcg: memcg to consume from. |
| 2100 | * @nr_pages: how many pages to charge. |
| 2101 | * |
| 2102 | * The charges will only happen if @memcg matches the current cpu's memcg |
| 2103 | * stock, and at least @nr_pages are available in that stock. Failure to |
| 2104 | * service an allocation will refill the stock. |
| 2105 | * |
| 2106 | * returns true if successful, false otherwise. |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2107 | */ |
Suleiman Souhlal | a0956d5 | 2012-12-18 14:21:36 -0800 | [diff] [blame] | 2108 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2109 | { |
| 2110 | struct memcg_stock_pcp *stock; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2111 | bool ret = false; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2112 | |
Suleiman Souhlal | a0956d5 | 2012-12-18 14:21:36 -0800 | [diff] [blame] | 2113 | if (nr_pages > CHARGE_BATCH) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2114 | return ret; |
Suleiman Souhlal | a0956d5 | 2012-12-18 14:21:36 -0800 | [diff] [blame] | 2115 | |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2116 | stock = &get_cpu_var(memcg_stock); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2117 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
Suleiman Souhlal | a0956d5 | 2012-12-18 14:21:36 -0800 | [diff] [blame] | 2118 | stock->nr_pages -= nr_pages; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2119 | ret = true; |
| 2120 | } |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2121 | put_cpu_var(memcg_stock); |
| 2122 | return ret; |
| 2123 | } |
| 2124 | |
| 2125 | /* |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2126 | * Returns stocks cached in percpu and reset cached information. |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2127 | */ |
| 2128 | static void drain_stock(struct memcg_stock_pcp *stock) |
| 2129 | { |
| 2130 | struct mem_cgroup *old = stock->cached; |
| 2131 | |
Johannes Weiner | 11c9ea4 | 2011-03-23 16:42:34 -0700 | [diff] [blame] | 2132 | if (stock->nr_pages) { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2133 | page_counter_uncharge(&old->memory, stock->nr_pages); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2134 | if (do_swap_account) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2135 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
Johannes Weiner | e8ea14c | 2014-12-10 15:42:42 -0800 | [diff] [blame] | 2136 | css_put_many(&old->css, stock->nr_pages); |
Johannes Weiner | 11c9ea4 | 2011-03-23 16:42:34 -0700 | [diff] [blame] | 2137 | stock->nr_pages = 0; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2138 | } |
| 2139 | stock->cached = NULL; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2140 | } |
| 2141 | |
| 2142 | /* |
| 2143 | * This must be called under preempt disabled or must be called by |
| 2144 | * a thread which is pinned to local cpu. |
| 2145 | */ |
| 2146 | static void drain_local_stock(struct work_struct *dummy) |
| 2147 | { |
Christoph Lameter | 7c8e018 | 2014-06-04 16:07:56 -0700 | [diff] [blame] | 2148 | struct memcg_stock_pcp *stock = this_cpu_ptr(&memcg_stock); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2149 | drain_stock(stock); |
KAMEZAWA Hiroyuki | 26fe616 | 2011-06-15 15:08:45 -0700 | [diff] [blame] | 2150 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2151 | } |
| 2152 | |
| 2153 | /* |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2154 | * Cache charges(val) to local per_cpu area. |
Greg Thelen | 320cc51 | 2010-03-15 15:27:28 +0100 | [diff] [blame] | 2155 | * This will be consumed by consume_stock() function, later. |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2156 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2157 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2158 | { |
| 2159 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); |
| 2160 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2161 | if (stock->cached != memcg) { /* reset if necessary */ |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2162 | drain_stock(stock); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2163 | stock->cached = memcg; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2164 | } |
Johannes Weiner | 11c9ea4 | 2011-03-23 16:42:34 -0700 | [diff] [blame] | 2165 | stock->nr_pages += nr_pages; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2166 | put_cpu_var(memcg_stock); |
| 2167 | } |
| 2168 | |
| 2169 | /* |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2170 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
Johannes Weiner | 6d3d6aa | 2014-12-10 15:42:50 -0800 | [diff] [blame] | 2171 | * of the hierarchy under it. |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2172 | */ |
Johannes Weiner | 6d3d6aa | 2014-12-10 15:42:50 -0800 | [diff] [blame] | 2173 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2174 | { |
KAMEZAWA Hiroyuki | 26fe616 | 2011-06-15 15:08:45 -0700 | [diff] [blame] | 2175 | int cpu, curcpu; |
Michal Hocko | d38144b | 2011-07-26 16:08:28 -0700 | [diff] [blame] | 2176 | |
Johannes Weiner | 6d3d6aa | 2014-12-10 15:42:50 -0800 | [diff] [blame] | 2177 | /* If someone's already draining, avoid adding running more workers. */ |
| 2178 | if (!mutex_trylock(&percpu_charge_mutex)) |
| 2179 | return; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2180 | /* Notify other cpus that system-wide "drain" is running */ |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2181 | get_online_cpus(); |
Johannes Weiner | 5af12d0 | 2011-08-25 15:59:07 -0700 | [diff] [blame] | 2182 | curcpu = get_cpu(); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2183 | for_each_online_cpu(cpu) { |
| 2184 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2185 | struct mem_cgroup *memcg; |
KAMEZAWA Hiroyuki | 26fe616 | 2011-06-15 15:08:45 -0700 | [diff] [blame] | 2186 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2187 | memcg = stock->cached; |
| 2188 | if (!memcg || !stock->nr_pages) |
KAMEZAWA Hiroyuki | 26fe616 | 2011-06-15 15:08:45 -0700 | [diff] [blame] | 2189 | continue; |
Johannes Weiner | 2314b42 | 2014-12-10 15:44:33 -0800 | [diff] [blame] | 2190 | if (!mem_cgroup_is_descendant(memcg, root_memcg)) |
Michal Hocko | 3e92041 | 2011-07-26 16:08:29 -0700 | [diff] [blame] | 2191 | continue; |
Michal Hocko | d1a05b6 | 2011-07-26 16:08:27 -0700 | [diff] [blame] | 2192 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
| 2193 | if (cpu == curcpu) |
| 2194 | drain_local_stock(&stock->work); |
| 2195 | else |
| 2196 | schedule_work_on(cpu, &stock->work); |
| 2197 | } |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2198 | } |
Johannes Weiner | 5af12d0 | 2011-08-25 15:59:07 -0700 | [diff] [blame] | 2199 | put_cpu(); |
Andrew Morton | f894ffa | 2013-09-12 15:13:35 -0700 | [diff] [blame] | 2200 | put_online_cpus(); |
Michal Hocko | 9f50fad | 2011-08-09 11:56:26 +0200 | [diff] [blame] | 2201 | mutex_unlock(&percpu_charge_mutex); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2202 | } |
| 2203 | |
Paul Gortmaker | 0db0628 | 2013-06-19 14:53:51 -0400 | [diff] [blame] | 2204 | static int memcg_cpu_hotplug_callback(struct notifier_block *nb, |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2205 | unsigned long action, |
| 2206 | void *hcpu) |
| 2207 | { |
| 2208 | int cpu = (unsigned long)hcpu; |
| 2209 | struct memcg_stock_pcp *stock; |
| 2210 | |
KAMEZAWA Hiroyuki | 619d094 | 2012-03-21 16:34:23 -0700 | [diff] [blame] | 2211 | if (action == CPU_ONLINE) |
KAMEZAWA Hiroyuki | 1489eba | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 2212 | return NOTIFY_OK; |
KAMEZAWA Hiroyuki | 1489eba | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 2213 | |
Kirill A. Shutemov | d833049 | 2012-04-12 12:49:11 -0700 | [diff] [blame] | 2214 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2215 | return NOTIFY_OK; |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 2216 | |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2217 | stock = &per_cpu(memcg_stock, cpu); |
| 2218 | drain_stock(stock); |
| 2219 | return NOTIFY_OK; |
| 2220 | } |
| 2221 | |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 2222 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
| 2223 | unsigned int nr_pages) |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2224 | { |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 2225 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
Johannes Weiner | 9b13061 | 2014-08-06 16:05:51 -0700 | [diff] [blame] | 2226 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2227 | struct mem_cgroup *mem_over_limit; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2228 | struct page_counter *counter; |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2229 | unsigned long nr_reclaimed; |
Johannes Weiner | b70a2a2 | 2014-10-09 15:28:56 -0700 | [diff] [blame] | 2230 | bool may_swap = true; |
| 2231 | bool drained = false; |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 2232 | int ret = 0; |
KAMEZAWA Hiroyuki | a636b32 | 2009-01-07 18:08:08 -0800 | [diff] [blame] | 2233 | |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 2234 | if (mem_cgroup_is_root(memcg)) |
| 2235 | goto done; |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2236 | retry: |
Michal Hocko | b6b6cc7 | 2014-04-07 15:37:44 -0700 | [diff] [blame] | 2237 | if (consume_stock(memcg, nr_pages)) |
| 2238 | goto done; |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2239 | |
Johannes Weiner | 3fbe724 | 2014-10-09 15:28:54 -0700 | [diff] [blame] | 2240 | if (!do_swap_account || |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2241 | !page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
| 2242 | if (!page_counter_try_charge(&memcg->memory, batch, &counter)) |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2243 | goto done_restock; |
Johannes Weiner | 3fbe724 | 2014-10-09 15:28:54 -0700 | [diff] [blame] | 2244 | if (do_swap_account) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2245 | page_counter_uncharge(&memcg->memsw, batch); |
| 2246 | mem_over_limit = mem_cgroup_from_counter(counter, memory); |
Johannes Weiner | 3fbe724 | 2014-10-09 15:28:54 -0700 | [diff] [blame] | 2247 | } else { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2248 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
Johannes Weiner | b70a2a2 | 2014-10-09 15:28:56 -0700 | [diff] [blame] | 2249 | may_swap = false; |
Johannes Weiner | 3fbe724 | 2014-10-09 15:28:54 -0700 | [diff] [blame] | 2250 | } |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2251 | |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2252 | if (batch > nr_pages) { |
| 2253 | batch = nr_pages; |
| 2254 | goto retry; |
| 2255 | } |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2256 | |
Johannes Weiner | 06b078f | 2014-08-06 16:05:44 -0700 | [diff] [blame] | 2257 | /* |
| 2258 | * Unlike in global OOM situations, memcg is not in a physical |
| 2259 | * memory shortage. Allow dying and OOM-killed tasks to |
| 2260 | * bypass the last charges so that they can exit quickly and |
| 2261 | * free their memory. |
| 2262 | */ |
| 2263 | if (unlikely(test_thread_flag(TIF_MEMDIE) || |
| 2264 | fatal_signal_pending(current) || |
| 2265 | current->flags & PF_EXITING)) |
| 2266 | goto bypass; |
| 2267 | |
| 2268 | if (unlikely(task_in_memcg_oom(current))) |
| 2269 | goto nomem; |
| 2270 | |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2271 | if (!(gfp_mask & __GFP_WAIT)) |
| 2272 | goto nomem; |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2273 | |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 2274 | mem_cgroup_events(mem_over_limit, MEMCG_MAX, 1); |
| 2275 | |
Johannes Weiner | b70a2a2 | 2014-10-09 15:28:56 -0700 | [diff] [blame] | 2276 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
| 2277 | gfp_mask, may_swap); |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2278 | |
Johannes Weiner | 61e02c7 | 2014-08-06 16:08:16 -0700 | [diff] [blame] | 2279 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2280 | goto retry; |
Johannes Weiner | 28c34c2 | 2014-08-06 16:05:47 -0700 | [diff] [blame] | 2281 | |
Johannes Weiner | b70a2a2 | 2014-10-09 15:28:56 -0700 | [diff] [blame] | 2282 | if (!drained) { |
Johannes Weiner | 6d3d6aa | 2014-12-10 15:42:50 -0800 | [diff] [blame] | 2283 | drain_all_stock(mem_over_limit); |
Johannes Weiner | b70a2a2 | 2014-10-09 15:28:56 -0700 | [diff] [blame] | 2284 | drained = true; |
| 2285 | goto retry; |
| 2286 | } |
| 2287 | |
Johannes Weiner | 28c34c2 | 2014-08-06 16:05:47 -0700 | [diff] [blame] | 2288 | if (gfp_mask & __GFP_NORETRY) |
| 2289 | goto nomem; |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2290 | /* |
| 2291 | * Even though the limit is exceeded at this point, reclaim |
| 2292 | * may have been able to free some pages. Retry the charge |
| 2293 | * before killing the task. |
| 2294 | * |
| 2295 | * Only for regular pages, though: huge pages are rather |
| 2296 | * unlikely to succeed so close to the limit, and we fall back |
| 2297 | * to regular pages anyway in case of failure. |
| 2298 | */ |
Johannes Weiner | 61e02c7 | 2014-08-06 16:08:16 -0700 | [diff] [blame] | 2299 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2300 | goto retry; |
| 2301 | /* |
| 2302 | * At task move, charge accounts can be doubly counted. So, it's |
| 2303 | * better to wait until the end of task_move if something is going on. |
| 2304 | */ |
| 2305 | if (mem_cgroup_wait_acct_move(mem_over_limit)) |
| 2306 | goto retry; |
| 2307 | |
Johannes Weiner | 9b13061 | 2014-08-06 16:05:51 -0700 | [diff] [blame] | 2308 | if (nr_retries--) |
| 2309 | goto retry; |
| 2310 | |
Johannes Weiner | 06b078f | 2014-08-06 16:05:44 -0700 | [diff] [blame] | 2311 | if (gfp_mask & __GFP_NOFAIL) |
| 2312 | goto bypass; |
| 2313 | |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2314 | if (fatal_signal_pending(current)) |
| 2315 | goto bypass; |
| 2316 | |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 2317 | mem_cgroup_events(mem_over_limit, MEMCG_OOM, 1); |
| 2318 | |
Johannes Weiner | 61e02c7 | 2014-08-06 16:08:16 -0700 | [diff] [blame] | 2319 | mem_cgroup_oom(mem_over_limit, gfp_mask, get_order(nr_pages)); |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2320 | nomem: |
Johannes Weiner | 6d1fdc4 | 2014-04-07 15:37:45 -0700 | [diff] [blame] | 2321 | if (!(gfp_mask & __GFP_NOFAIL)) |
Johannes Weiner | 3168ecb | 2013-10-31 16:34:13 -0700 | [diff] [blame] | 2322 | return -ENOMEM; |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 2323 | bypass: |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 2324 | return -EINTR; |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2325 | |
| 2326 | done_restock: |
Johannes Weiner | e8ea14c | 2014-12-10 15:42:42 -0800 | [diff] [blame] | 2327 | css_get_many(&memcg->css, batch); |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2328 | if (batch > nr_pages) |
| 2329 | refill_stock(memcg, batch - nr_pages); |
Vladimir Davydov | 7d63809 | 2015-06-10 11:14:46 -0700 | [diff] [blame] | 2330 | if (!(gfp_mask & __GFP_WAIT)) |
| 2331 | goto done; |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 2332 | /* |
| 2333 | * If the hierarchy is above the normal consumption range, |
| 2334 | * make the charging task trim their excess contribution. |
| 2335 | */ |
| 2336 | do { |
| 2337 | if (page_counter_read(&memcg->memory) <= memcg->high) |
| 2338 | continue; |
| 2339 | mem_cgroup_events(memcg, MEMCG_HIGH, 1); |
| 2340 | try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true); |
| 2341 | } while ((memcg = parent_mem_cgroup(memcg))); |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2342 | done: |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 2343 | return ret; |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2344 | } |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2345 | |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 2346 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
Daisuke Nishimura | a3032a2 | 2009-12-15 16:47:10 -0800 | [diff] [blame] | 2347 | { |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 2348 | if (mem_cgroup_is_root(memcg)) |
| 2349 | return; |
| 2350 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2351 | page_counter_uncharge(&memcg->memory, nr_pages); |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 2352 | if (do_swap_account) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2353 | page_counter_uncharge(&memcg->memsw, nr_pages); |
Daisuke Nishimura | a3032a2 | 2009-12-15 16:47:10 -0800 | [diff] [blame] | 2354 | |
Johannes Weiner | e8ea14c | 2014-12-10 15:42:42 -0800 | [diff] [blame] | 2355 | css_put_many(&memcg->css, nr_pages); |
KAMEZAWA Hiroyuki | d01dd17 | 2012-05-29 15:07:03 -0700 | [diff] [blame] | 2356 | } |
| 2357 | |
| 2358 | /* |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2359 | * try_get_mem_cgroup_from_page - look up page's memcg association |
| 2360 | * @page: the page |
| 2361 | * |
| 2362 | * Look up, get a css reference, and return the memcg that owns @page. |
| 2363 | * |
| 2364 | * The page must be locked to prevent racing with swap-in and page |
| 2365 | * cache charges. If coming from an unlocked page table, the caller |
| 2366 | * must ensure the page is on the LRU or this can race with charging. |
| 2367 | */ |
Wu Fengguang | e42d9d5 | 2009-12-16 12:19:59 +0100 | [diff] [blame] | 2368 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
KAMEZAWA Hiroyuki | b5a8431 | 2009-01-07 18:08:35 -0800 | [diff] [blame] | 2369 | { |
Johannes Weiner | 2983331 | 2014-12-10 15:44:02 -0800 | [diff] [blame] | 2370 | struct mem_cgroup *memcg; |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2371 | unsigned short id; |
KAMEZAWA Hiroyuki | b5a8431 | 2009-01-07 18:08:35 -0800 | [diff] [blame] | 2372 | swp_entry_t ent; |
| 2373 | |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 2374 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
Daisuke Nishimura | 3c776e6 | 2009-04-02 16:57:43 -0700 | [diff] [blame] | 2375 | |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 2376 | memcg = page->mem_cgroup; |
Johannes Weiner | 2983331 | 2014-12-10 15:44:02 -0800 | [diff] [blame] | 2377 | if (memcg) { |
| 2378 | if (!css_tryget_online(&memcg->css)) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2379 | memcg = NULL; |
Wu Fengguang | e42d9d5 | 2009-12-16 12:19:59 +0100 | [diff] [blame] | 2380 | } else if (PageSwapCache(page)) { |
Daisuke Nishimura | 3c776e6 | 2009-04-02 16:57:43 -0700 | [diff] [blame] | 2381 | ent.val = page_private(page); |
Bob Liu | 9fb4b7c | 2012-01-12 17:18:48 -0800 | [diff] [blame] | 2382 | id = lookup_swap_cgroup_id(ent); |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2383 | rcu_read_lock(); |
Vladimir Davydov | adbe427 | 2015-04-15 16:13:00 -0700 | [diff] [blame] | 2384 | memcg = mem_cgroup_from_id(id); |
Tejun Heo | ec903c0 | 2014-05-13 12:11:01 -0400 | [diff] [blame] | 2385 | if (memcg && !css_tryget_online(&memcg->css)) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2386 | memcg = NULL; |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2387 | rcu_read_unlock(); |
Daisuke Nishimura | 3c776e6 | 2009-04-02 16:57:43 -0700 | [diff] [blame] | 2388 | } |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2389 | return memcg; |
KAMEZAWA Hiroyuki | b5a8431 | 2009-01-07 18:08:35 -0800 | [diff] [blame] | 2390 | } |
| 2391 | |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2392 | static void lock_page_lru(struct page *page, int *isolated) |
| 2393 | { |
| 2394 | struct zone *zone = page_zone(page); |
| 2395 | |
| 2396 | spin_lock_irq(&zone->lru_lock); |
| 2397 | if (PageLRU(page)) { |
| 2398 | struct lruvec *lruvec; |
| 2399 | |
| 2400 | lruvec = mem_cgroup_page_lruvec(page, zone); |
| 2401 | ClearPageLRU(page); |
| 2402 | del_page_from_lru_list(page, lruvec, page_lru(page)); |
| 2403 | *isolated = 1; |
| 2404 | } else |
| 2405 | *isolated = 0; |
| 2406 | } |
| 2407 | |
| 2408 | static void unlock_page_lru(struct page *page, int isolated) |
| 2409 | { |
| 2410 | struct zone *zone = page_zone(page); |
| 2411 | |
| 2412 | if (isolated) { |
| 2413 | struct lruvec *lruvec; |
| 2414 | |
| 2415 | lruvec = mem_cgroup_page_lruvec(page, zone); |
| 2416 | VM_BUG_ON_PAGE(PageLRU(page), page); |
| 2417 | SetPageLRU(page); |
| 2418 | add_page_to_lru_list(page, lruvec, page_lru(page)); |
| 2419 | } |
| 2420 | spin_unlock_irq(&zone->lru_lock); |
| 2421 | } |
| 2422 | |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 2423 | static void commit_charge(struct page *page, struct mem_cgroup *memcg, |
Johannes Weiner | 6abb5a8 | 2014-08-08 14:19:33 -0700 | [diff] [blame] | 2424 | bool lrucare) |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2425 | { |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2426 | int isolated; |
Hugh Dickins | 9ce70c0 | 2012-03-05 14:59:16 -0800 | [diff] [blame] | 2427 | |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 2428 | VM_BUG_ON_PAGE(page->mem_cgroup, page); |
Hugh Dickins | 9ce70c0 | 2012-03-05 14:59:16 -0800 | [diff] [blame] | 2429 | |
| 2430 | /* |
| 2431 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page |
| 2432 | * may already be on some other mem_cgroup's LRU. Take care of it. |
| 2433 | */ |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2434 | if (lrucare) |
| 2435 | lock_page_lru(page, &isolated); |
Hugh Dickins | 9ce70c0 | 2012-03-05 14:59:16 -0800 | [diff] [blame] | 2436 | |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2437 | /* |
| 2438 | * Nobody should be changing or seriously looking at |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 2439 | * page->mem_cgroup at this point: |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2440 | * |
| 2441 | * - the page is uncharged |
| 2442 | * |
| 2443 | * - the page is off-LRU |
| 2444 | * |
| 2445 | * - an anonymous fault has exclusive page access, except for |
| 2446 | * a locked page table |
| 2447 | * |
| 2448 | * - a page cache insertion, a swapin fault, or a migration |
| 2449 | * have the page locked |
| 2450 | */ |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 2451 | page->mem_cgroup = memcg; |
Hugh Dickins | 3be9127 | 2008-02-07 00:14:19 -0800 | [diff] [blame] | 2452 | |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2453 | if (lrucare) |
| 2454 | unlock_page_lru(page, isolated); |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2455 | } |
| 2456 | |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2457 | #ifdef CONFIG_MEMCG_KMEM |
Vladimir Davydov | dbf22eb | 2015-02-10 14:11:41 -0800 | [diff] [blame] | 2458 | int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp, |
| 2459 | unsigned long nr_pages) |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 2460 | { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2461 | struct page_counter *counter; |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2462 | int ret = 0; |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2463 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2464 | ret = page_counter_try_charge(&memcg->kmem, nr_pages, &counter); |
| 2465 | if (ret < 0) |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2466 | return ret; |
| 2467 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2468 | ret = try_charge(memcg, gfp, nr_pages); |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2469 | if (ret == -EINTR) { |
| 2470 | /* |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 2471 | * try_charge() chose to bypass to root due to OOM kill or |
| 2472 | * fatal signal. Since our only options are to either fail |
| 2473 | * the allocation or charge it to this cgroup, do it as a |
| 2474 | * temporary condition. But we can't fail. From a kmem/slab |
| 2475 | * perspective, the cache has already been selected, by |
| 2476 | * mem_cgroup_kmem_get_cache(), so it is too late to change |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2477 | * our minds. |
| 2478 | * |
| 2479 | * This condition will only trigger if the task entered |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 2480 | * memcg_charge_kmem in a sane state, but was OOM-killed |
| 2481 | * during try_charge() above. Tasks that were already dying |
| 2482 | * when the allocation triggers should have been already |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2483 | * directed to the root cgroup in memcontrol.h |
| 2484 | */ |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2485 | page_counter_charge(&memcg->memory, nr_pages); |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2486 | if (do_swap_account) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2487 | page_counter_charge(&memcg->memsw, nr_pages); |
Johannes Weiner | e8ea14c | 2014-12-10 15:42:42 -0800 | [diff] [blame] | 2488 | css_get_many(&memcg->css, nr_pages); |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2489 | ret = 0; |
| 2490 | } else if (ret) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2491 | page_counter_uncharge(&memcg->kmem, nr_pages); |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2492 | |
| 2493 | return ret; |
| 2494 | } |
| 2495 | |
Vladimir Davydov | dbf22eb | 2015-02-10 14:11:41 -0800 | [diff] [blame] | 2496 | void memcg_uncharge_kmem(struct mem_cgroup *memcg, unsigned long nr_pages) |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2497 | { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2498 | page_counter_uncharge(&memcg->memory, nr_pages); |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2499 | if (do_swap_account) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2500 | page_counter_uncharge(&memcg->memsw, nr_pages); |
Glauber Costa | 7de3768 | 2012-12-18 14:22:07 -0800 | [diff] [blame] | 2501 | |
Johannes Weiner | 64f2199 | 2014-12-10 15:42:45 -0800 | [diff] [blame] | 2502 | page_counter_uncharge(&memcg->kmem, nr_pages); |
Glauber Costa | 7de3768 | 2012-12-18 14:22:07 -0800 | [diff] [blame] | 2503 | |
Johannes Weiner | e8ea14c | 2014-12-10 15:42:42 -0800 | [diff] [blame] | 2504 | css_put_many(&memcg->css, nr_pages); |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2505 | } |
| 2506 | |
Glauber Costa | 2633d7a | 2012-12-18 14:22:34 -0800 | [diff] [blame] | 2507 | /* |
| 2508 | * helper for acessing a memcg's index. It will be used as an index in the |
| 2509 | * child cache array in kmem_cache, and also to derive its name. This function |
| 2510 | * will return -1 when this is not a kmem-limited memcg. |
| 2511 | */ |
| 2512 | int memcg_cache_id(struct mem_cgroup *memcg) |
| 2513 | { |
| 2514 | return memcg ? memcg->kmemcg_id : -1; |
| 2515 | } |
| 2516 | |
Vladimir Davydov | f3bb304 | 2014-10-09 15:28:45 -0700 | [diff] [blame] | 2517 | static int memcg_alloc_cache_id(void) |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 2518 | { |
Vladimir Davydov | f3bb304 | 2014-10-09 15:28:45 -0700 | [diff] [blame] | 2519 | int id, size; |
| 2520 | int err; |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 2521 | |
Vladimir Davydov | dbcf73e | 2015-02-12 14:58:57 -0800 | [diff] [blame] | 2522 | id = ida_simple_get(&memcg_cache_ida, |
Vladimir Davydov | f3bb304 | 2014-10-09 15:28:45 -0700 | [diff] [blame] | 2523 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); |
| 2524 | if (id < 0) |
| 2525 | return id; |
| 2526 | |
Vladimir Davydov | dbcf73e | 2015-02-12 14:58:57 -0800 | [diff] [blame] | 2527 | if (id < memcg_nr_cache_ids) |
Vladimir Davydov | f3bb304 | 2014-10-09 15:28:45 -0700 | [diff] [blame] | 2528 | return id; |
| 2529 | |
| 2530 | /* |
| 2531 | * There's no space for the new id in memcg_caches arrays, |
| 2532 | * so we have to grow them. |
| 2533 | */ |
Vladimir Davydov | 05257a1 | 2015-02-12 14:59:01 -0800 | [diff] [blame] | 2534 | down_write(&memcg_cache_ids_sem); |
Vladimir Davydov | f3bb304 | 2014-10-09 15:28:45 -0700 | [diff] [blame] | 2535 | |
| 2536 | size = 2 * (id + 1); |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 2537 | if (size < MEMCG_CACHES_MIN_SIZE) |
| 2538 | size = MEMCG_CACHES_MIN_SIZE; |
| 2539 | else if (size > MEMCG_CACHES_MAX_SIZE) |
| 2540 | size = MEMCG_CACHES_MAX_SIZE; |
| 2541 | |
Vladimir Davydov | f3bb304 | 2014-10-09 15:28:45 -0700 | [diff] [blame] | 2542 | err = memcg_update_all_caches(size); |
Vladimir Davydov | 05257a1 | 2015-02-12 14:59:01 -0800 | [diff] [blame] | 2543 | if (!err) |
Vladimir Davydov | 60d3fd3 | 2015-02-12 14:59:10 -0800 | [diff] [blame] | 2544 | err = memcg_update_all_list_lrus(size); |
| 2545 | if (!err) |
Vladimir Davydov | 05257a1 | 2015-02-12 14:59:01 -0800 | [diff] [blame] | 2546 | memcg_nr_cache_ids = size; |
| 2547 | |
| 2548 | up_write(&memcg_cache_ids_sem); |
| 2549 | |
Vladimir Davydov | f3bb304 | 2014-10-09 15:28:45 -0700 | [diff] [blame] | 2550 | if (err) { |
Vladimir Davydov | dbcf73e | 2015-02-12 14:58:57 -0800 | [diff] [blame] | 2551 | ida_simple_remove(&memcg_cache_ida, id); |
Vladimir Davydov | f3bb304 | 2014-10-09 15:28:45 -0700 | [diff] [blame] | 2552 | return err; |
| 2553 | } |
| 2554 | return id; |
| 2555 | } |
| 2556 | |
| 2557 | static void memcg_free_cache_id(int id) |
| 2558 | { |
Vladimir Davydov | dbcf73e | 2015-02-12 14:58:57 -0800 | [diff] [blame] | 2559 | ida_simple_remove(&memcg_cache_ida, id); |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 2560 | } |
| 2561 | |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 2562 | struct memcg_kmem_cache_create_work { |
Vladimir Davydov | 5722d09 | 2014-04-07 15:39:24 -0700 | [diff] [blame] | 2563 | struct mem_cgroup *memcg; |
| 2564 | struct kmem_cache *cachep; |
| 2565 | struct work_struct work; |
| 2566 | }; |
| 2567 | |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 2568 | static void memcg_kmem_cache_create_func(struct work_struct *w) |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2569 | { |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 2570 | struct memcg_kmem_cache_create_work *cw = |
| 2571 | container_of(w, struct memcg_kmem_cache_create_work, work); |
Vladimir Davydov | 5722d09 | 2014-04-07 15:39:24 -0700 | [diff] [blame] | 2572 | struct mem_cgroup *memcg = cw->memcg; |
| 2573 | struct kmem_cache *cachep = cw->cachep; |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2574 | |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 2575 | memcg_create_kmem_cache(memcg, cachep); |
Vladimir Davydov | bd67314 | 2014-06-04 16:07:40 -0700 | [diff] [blame] | 2576 | |
Vladimir Davydov | 5722d09 | 2014-04-07 15:39:24 -0700 | [diff] [blame] | 2577 | css_put(&memcg->css); |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2578 | kfree(cw); |
| 2579 | } |
| 2580 | |
| 2581 | /* |
| 2582 | * Enqueue the creation of a per-memcg kmem_cache. |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2583 | */ |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 2584 | static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
| 2585 | struct kmem_cache *cachep) |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2586 | { |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 2587 | struct memcg_kmem_cache_create_work *cw; |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2588 | |
Vladimir Davydov | 776ed0f | 2014-06-04 16:10:02 -0700 | [diff] [blame] | 2589 | cw = kmalloc(sizeof(*cw), GFP_NOWAIT); |
Vladimir Davydov | 8135be5 | 2014-12-12 16:56:38 -0800 | [diff] [blame] | 2590 | if (!cw) |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2591 | return; |
Vladimir Davydov | 8135be5 | 2014-12-12 16:56:38 -0800 | [diff] [blame] | 2592 | |
| 2593 | css_get(&memcg->css); |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2594 | |
| 2595 | cw->memcg = memcg; |
| 2596 | cw->cachep = cachep; |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 2597 | INIT_WORK(&cw->work, memcg_kmem_cache_create_func); |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2598 | |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2599 | schedule_work(&cw->work); |
| 2600 | } |
| 2601 | |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 2602 | static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
| 2603 | struct kmem_cache *cachep) |
Glauber Costa | 0e9d92f | 2012-12-18 14:22:42 -0800 | [diff] [blame] | 2604 | { |
| 2605 | /* |
| 2606 | * We need to stop accounting when we kmalloc, because if the |
| 2607 | * corresponding kmalloc cache is not yet created, the first allocation |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 2608 | * in __memcg_schedule_kmem_cache_create will recurse. |
Glauber Costa | 0e9d92f | 2012-12-18 14:22:42 -0800 | [diff] [blame] | 2609 | * |
| 2610 | * However, it is better to enclose the whole function. Depending on |
| 2611 | * the debugging options enabled, INIT_WORK(), for instance, can |
| 2612 | * trigger an allocation. This too, will make us recurse. Because at |
| 2613 | * this point we can't allow ourselves back into memcg_kmem_get_cache, |
| 2614 | * the safest choice is to do it like this, wrapping the whole function. |
| 2615 | */ |
Vladimir Davydov | 6f185c2 | 2014-12-12 16:55:15 -0800 | [diff] [blame] | 2616 | current->memcg_kmem_skip_account = 1; |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 2617 | __memcg_schedule_kmem_cache_create(memcg, cachep); |
Vladimir Davydov | 6f185c2 | 2014-12-12 16:55:15 -0800 | [diff] [blame] | 2618 | current->memcg_kmem_skip_account = 0; |
Glauber Costa | 0e9d92f | 2012-12-18 14:22:42 -0800 | [diff] [blame] | 2619 | } |
Vladimir Davydov | c67a8a6 | 2014-06-04 16:07:39 -0700 | [diff] [blame] | 2620 | |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2621 | /* |
| 2622 | * Return the kmem_cache we're supposed to use for a slab allocation. |
| 2623 | * We try to use the current memcg's version of the cache. |
| 2624 | * |
| 2625 | * If the cache does not exist yet, if we are the first user of it, |
| 2626 | * we either create it immediately, if possible, or create it asynchronously |
| 2627 | * in a workqueue. |
| 2628 | * In the latter case, we will let the current allocation go through with |
| 2629 | * the original cache. |
| 2630 | * |
| 2631 | * Can't be called in interrupt context or from kernel threads. |
| 2632 | * This function needs to be called with rcu_read_lock() held. |
| 2633 | */ |
Zhang Zhen | 056b7cc | 2014-12-12 16:55:38 -0800 | [diff] [blame] | 2634 | struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep) |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2635 | { |
| 2636 | struct mem_cgroup *memcg; |
Vladimir Davydov | 959c896 | 2014-01-23 15:52:59 -0800 | [diff] [blame] | 2637 | struct kmem_cache *memcg_cachep; |
Vladimir Davydov | 2a4db7e | 2015-02-12 14:59:32 -0800 | [diff] [blame] | 2638 | int kmemcg_id; |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2639 | |
Vladimir Davydov | f7ce319 | 2015-02-12 14:59:20 -0800 | [diff] [blame] | 2640 | VM_BUG_ON(!is_root_cache(cachep)); |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2641 | |
Vladimir Davydov | 9d100c5 | 2014-12-12 16:54:53 -0800 | [diff] [blame] | 2642 | if (current->memcg_kmem_skip_account) |
Glauber Costa | 0e9d92f | 2012-12-18 14:22:42 -0800 | [diff] [blame] | 2643 | return cachep; |
| 2644 | |
Vladimir Davydov | 8135be5 | 2014-12-12 16:56:38 -0800 | [diff] [blame] | 2645 | memcg = get_mem_cgroup_from_mm(current->mm); |
Jason Low | 4db0c3c | 2015-04-15 16:14:08 -0700 | [diff] [blame] | 2646 | kmemcg_id = READ_ONCE(memcg->kmemcg_id); |
Vladimir Davydov | 2a4db7e | 2015-02-12 14:59:32 -0800 | [diff] [blame] | 2647 | if (kmemcg_id < 0) |
Li Zefan | ca0dde9 | 2013-04-29 15:08:57 -0700 | [diff] [blame] | 2648 | goto out; |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2649 | |
Vladimir Davydov | 2a4db7e | 2015-02-12 14:59:32 -0800 | [diff] [blame] | 2650 | memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id); |
Vladimir Davydov | 8135be5 | 2014-12-12 16:56:38 -0800 | [diff] [blame] | 2651 | if (likely(memcg_cachep)) |
| 2652 | return memcg_cachep; |
Li Zefan | ca0dde9 | 2013-04-29 15:08:57 -0700 | [diff] [blame] | 2653 | |
| 2654 | /* |
| 2655 | * If we are in a safe context (can wait, and not in interrupt |
| 2656 | * context), we could be be predictable and return right away. |
| 2657 | * This would guarantee that the allocation being performed |
| 2658 | * already belongs in the new cache. |
| 2659 | * |
| 2660 | * However, there are some clashes that can arrive from locking. |
| 2661 | * For instance, because we acquire the slab_mutex while doing |
Vladimir Davydov | 776ed0f | 2014-06-04 16:10:02 -0700 | [diff] [blame] | 2662 | * memcg_create_kmem_cache, this means no further allocation |
| 2663 | * could happen with the slab_mutex held. So it's better to |
| 2664 | * defer everything. |
Li Zefan | ca0dde9 | 2013-04-29 15:08:57 -0700 | [diff] [blame] | 2665 | */ |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 2666 | memcg_schedule_kmem_cache_create(memcg, cachep); |
Li Zefan | ca0dde9 | 2013-04-29 15:08:57 -0700 | [diff] [blame] | 2667 | out: |
Vladimir Davydov | 8135be5 | 2014-12-12 16:56:38 -0800 | [diff] [blame] | 2668 | css_put(&memcg->css); |
Li Zefan | ca0dde9 | 2013-04-29 15:08:57 -0700 | [diff] [blame] | 2669 | return cachep; |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2670 | } |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2671 | |
Vladimir Davydov | 8135be5 | 2014-12-12 16:56:38 -0800 | [diff] [blame] | 2672 | void __memcg_kmem_put_cache(struct kmem_cache *cachep) |
| 2673 | { |
| 2674 | if (!is_root_cache(cachep)) |
Vladimir Davydov | f7ce319 | 2015-02-12 14:59:20 -0800 | [diff] [blame] | 2675 | css_put(&cachep->memcg_params.memcg->css); |
Vladimir Davydov | 8135be5 | 2014-12-12 16:56:38 -0800 | [diff] [blame] | 2676 | } |
| 2677 | |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2678 | /* |
| 2679 | * We need to verify if the allocation against current->mm->owner's memcg is |
| 2680 | * possible for the given order. But the page is not allocated yet, so we'll |
| 2681 | * need a further commit step to do the final arrangements. |
| 2682 | * |
| 2683 | * It is possible for the task to switch cgroups in this mean time, so at |
| 2684 | * commit time, we can't rely on task conversion any longer. We'll then use |
| 2685 | * the handle argument to return to the caller which cgroup we should commit |
| 2686 | * against. We could also return the memcg directly and avoid the pointer |
| 2687 | * passing, but a boolean return value gives better semantics considering |
| 2688 | * the compiled-out case as well. |
| 2689 | * |
| 2690 | * Returning true means the allocation is possible. |
| 2691 | */ |
| 2692 | bool |
| 2693 | __memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **_memcg, int order) |
| 2694 | { |
| 2695 | struct mem_cgroup *memcg; |
| 2696 | int ret; |
| 2697 | |
| 2698 | *_memcg = NULL; |
Glauber Costa | 6d42c23 | 2013-07-08 16:00:00 -0700 | [diff] [blame] | 2699 | |
Johannes Weiner | df38197 | 2014-04-07 15:37:43 -0700 | [diff] [blame] | 2700 | memcg = get_mem_cgroup_from_mm(current->mm); |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2701 | |
Vladimir Davydov | cf2b8fb | 2014-10-09 15:28:59 -0700 | [diff] [blame] | 2702 | if (!memcg_kmem_is_active(memcg)) { |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2703 | css_put(&memcg->css); |
| 2704 | return true; |
| 2705 | } |
| 2706 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2707 | ret = memcg_charge_kmem(memcg, gfp, 1 << order); |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2708 | if (!ret) |
| 2709 | *_memcg = memcg; |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2710 | |
| 2711 | css_put(&memcg->css); |
| 2712 | return (ret == 0); |
| 2713 | } |
| 2714 | |
| 2715 | void __memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, |
| 2716 | int order) |
| 2717 | { |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2718 | VM_BUG_ON(mem_cgroup_is_root(memcg)); |
| 2719 | |
| 2720 | /* The page allocation failed. Revert */ |
| 2721 | if (!page) { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2722 | memcg_uncharge_kmem(memcg, 1 << order); |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2723 | return; |
| 2724 | } |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 2725 | page->mem_cgroup = memcg; |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2726 | } |
| 2727 | |
| 2728 | void __memcg_kmem_uncharge_pages(struct page *page, int order) |
| 2729 | { |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 2730 | struct mem_cgroup *memcg = page->mem_cgroup; |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2731 | |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2732 | if (!memcg) |
| 2733 | return; |
| 2734 | |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 2735 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
Johannes Weiner | 2983331 | 2014-12-10 15:44:02 -0800 | [diff] [blame] | 2736 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2737 | memcg_uncharge_kmem(memcg, 1 << order); |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 2738 | page->mem_cgroup = NULL; |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2739 | } |
Vladimir Davydov | 60d3fd3 | 2015-02-12 14:59:10 -0800 | [diff] [blame] | 2740 | |
| 2741 | struct mem_cgroup *__mem_cgroup_from_kmem(void *ptr) |
| 2742 | { |
| 2743 | struct mem_cgroup *memcg = NULL; |
| 2744 | struct kmem_cache *cachep; |
| 2745 | struct page *page; |
| 2746 | |
| 2747 | page = virt_to_head_page(ptr); |
| 2748 | if (PageSlab(page)) { |
| 2749 | cachep = page->slab_cache; |
| 2750 | if (!is_root_cache(cachep)) |
Vladimir Davydov | f7ce319 | 2015-02-12 14:59:20 -0800 | [diff] [blame] | 2751 | memcg = cachep->memcg_params.memcg; |
Vladimir Davydov | 60d3fd3 | 2015-02-12 14:59:10 -0800 | [diff] [blame] | 2752 | } else |
| 2753 | /* page allocated by alloc_kmem_pages */ |
| 2754 | memcg = page->mem_cgroup; |
| 2755 | |
| 2756 | return memcg; |
| 2757 | } |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2758 | #endif /* CONFIG_MEMCG_KMEM */ |
| 2759 | |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2760 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
| 2761 | |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2762 | /* |
| 2763 | * Because tail pages are not marked as "used", set it. We're under |
KAMEZAWA Hiroyuki | e94c8a9 | 2012-01-12 17:18:20 -0800 | [diff] [blame] | 2764 | * zone->lru_lock, 'splitting on pmd' and compound_lock. |
| 2765 | * charge/uncharge will be never happen and move_account() is done under |
| 2766 | * compound_lock(), so we don't have to take care of races. |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2767 | */ |
KAMEZAWA Hiroyuki | e94c8a9 | 2012-01-12 17:18:20 -0800 | [diff] [blame] | 2768 | void mem_cgroup_split_huge_fixup(struct page *head) |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2769 | { |
KAMEZAWA Hiroyuki | e94c8a9 | 2012-01-12 17:18:20 -0800 | [diff] [blame] | 2770 | int i; |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2771 | |
KAMEZAWA Hiroyuki | 3d37c4a | 2011-01-25 15:07:28 -0800 | [diff] [blame] | 2772 | if (mem_cgroup_disabled()) |
| 2773 | return; |
David Rientjes | b070e65 | 2013-05-07 16:18:09 -0700 | [diff] [blame] | 2774 | |
Johannes Weiner | 2983331 | 2014-12-10 15:44:02 -0800 | [diff] [blame] | 2775 | for (i = 1; i < HPAGE_PMD_NR; i++) |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 2776 | head[i].mem_cgroup = head->mem_cgroup; |
Michal Hocko | b9982f8 | 2014-12-10 15:43:51 -0800 | [diff] [blame] | 2777 | |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 2778 | __this_cpu_sub(head->mem_cgroup->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
David Rientjes | b070e65 | 2013-05-07 16:18:09 -0700 | [diff] [blame] | 2779 | HPAGE_PMD_NR); |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2780 | } |
Hugh Dickins | 12d2710 | 2012-01-12 17:19:52 -0800 | [diff] [blame] | 2781 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2782 | |
Andrew Morton | c255a45 | 2012-07-31 16:43:02 -0700 | [diff] [blame] | 2783 | #ifdef CONFIG_MEMCG_SWAP |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2784 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
| 2785 | bool charge) |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2786 | { |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2787 | int val = (charge) ? 1 : -1; |
| 2788 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2789 | } |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 2790 | |
| 2791 | /** |
| 2792 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. |
| 2793 | * @entry: swap entry to be moved |
| 2794 | * @from: mem_cgroup which the entry is moved from |
| 2795 | * @to: mem_cgroup which the entry is moved to |
| 2796 | * |
| 2797 | * It succeeds only when the swap_cgroup's record for this entry is the same |
| 2798 | * as the mem_cgroup's id of @from. |
| 2799 | * |
| 2800 | * Returns 0 on success, -EINVAL on failure. |
| 2801 | * |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2802 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 2803 | * both res and memsw, and called css_get(). |
| 2804 | */ |
| 2805 | static int mem_cgroup_move_swap_account(swp_entry_t entry, |
Hugh Dickins | e91cbb4 | 2012-05-29 15:06:51 -0700 | [diff] [blame] | 2806 | struct mem_cgroup *from, struct mem_cgroup *to) |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 2807 | { |
| 2808 | unsigned short old_id, new_id; |
| 2809 | |
Li Zefan | 34c00c3 | 2013-09-23 16:56:01 +0800 | [diff] [blame] | 2810 | old_id = mem_cgroup_id(from); |
| 2811 | new_id = mem_cgroup_id(to); |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 2812 | |
| 2813 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 2814 | mem_cgroup_swap_statistics(from, false); |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 2815 | mem_cgroup_swap_statistics(to, true); |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 2816 | return 0; |
| 2817 | } |
| 2818 | return -EINVAL; |
| 2819 | } |
| 2820 | #else |
| 2821 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, |
Hugh Dickins | e91cbb4 | 2012-05-29 15:06:51 -0700 | [diff] [blame] | 2822 | struct mem_cgroup *from, struct mem_cgroup *to) |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 2823 | { |
| 2824 | return -EINVAL; |
| 2825 | } |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2826 | #endif |
| 2827 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2828 | static DEFINE_MUTEX(memcg_limit_mutex); |
Daisuke Nishimura | f212ad7 | 2011-03-23 16:42:25 -0700 | [diff] [blame] | 2829 | |
KOSAKI Motohiro | d38d2a7 | 2009-01-06 14:39:44 -0800 | [diff] [blame] | 2830 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2831 | unsigned long limit) |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 2832 | { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2833 | unsigned long curusage; |
| 2834 | unsigned long oldusage; |
| 2835 | bool enlarge = false; |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 2836 | int retry_count; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2837 | int ret; |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 2838 | |
| 2839 | /* |
| 2840 | * For keeping hierarchical_reclaim simple, how long we should retry |
| 2841 | * is depends on callers. We set our retry-count to be function |
| 2842 | * of # of children which we should visit in this loop. |
| 2843 | */ |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2844 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
| 2845 | mem_cgroup_count_children(memcg); |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 2846 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2847 | oldusage = page_counter_read(&memcg->memory); |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 2848 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2849 | do { |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 2850 | if (signal_pending(current)) { |
| 2851 | ret = -EINTR; |
| 2852 | break; |
| 2853 | } |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2854 | |
| 2855 | mutex_lock(&memcg_limit_mutex); |
| 2856 | if (limit > memcg->memsw.limit) { |
| 2857 | mutex_unlock(&memcg_limit_mutex); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2858 | ret = -EINVAL; |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 2859 | break; |
| 2860 | } |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2861 | if (limit > memcg->memory.limit) |
| 2862 | enlarge = true; |
| 2863 | ret = page_counter_limit(&memcg->memory, limit); |
| 2864 | mutex_unlock(&memcg_limit_mutex); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2865 | |
| 2866 | if (!ret) |
| 2867 | break; |
| 2868 | |
Johannes Weiner | b70a2a2 | 2014-10-09 15:28:56 -0700 | [diff] [blame] | 2869 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true); |
| 2870 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2871 | curusage = page_counter_read(&memcg->memory); |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 2872 | /* Usage is reduced ? */ |
Andrew Morton | f894ffa | 2013-09-12 15:13:35 -0700 | [diff] [blame] | 2873 | if (curusage >= oldusage) |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 2874 | retry_count--; |
| 2875 | else |
| 2876 | oldusage = curusage; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2877 | } while (retry_count); |
| 2878 | |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 2879 | if (!ret && enlarge) |
| 2880 | memcg_oom_recover(memcg); |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 2881 | |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2882 | return ret; |
| 2883 | } |
| 2884 | |
Li Zefan | 338c843 | 2009-06-17 16:27:15 -0700 | [diff] [blame] | 2885 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2886 | unsigned long limit) |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2887 | { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2888 | unsigned long curusage; |
| 2889 | unsigned long oldusage; |
| 2890 | bool enlarge = false; |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 2891 | int retry_count; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2892 | int ret; |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2893 | |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 2894 | /* see mem_cgroup_resize_res_limit */ |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2895 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
| 2896 | mem_cgroup_count_children(memcg); |
| 2897 | |
| 2898 | oldusage = page_counter_read(&memcg->memsw); |
| 2899 | |
| 2900 | do { |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2901 | if (signal_pending(current)) { |
| 2902 | ret = -EINTR; |
| 2903 | break; |
| 2904 | } |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2905 | |
| 2906 | mutex_lock(&memcg_limit_mutex); |
| 2907 | if (limit < memcg->memory.limit) { |
| 2908 | mutex_unlock(&memcg_limit_mutex); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2909 | ret = -EINVAL; |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2910 | break; |
| 2911 | } |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2912 | if (limit > memcg->memsw.limit) |
| 2913 | enlarge = true; |
| 2914 | ret = page_counter_limit(&memcg->memsw, limit); |
| 2915 | mutex_unlock(&memcg_limit_mutex); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2916 | |
| 2917 | if (!ret) |
| 2918 | break; |
| 2919 | |
Johannes Weiner | b70a2a2 | 2014-10-09 15:28:56 -0700 | [diff] [blame] | 2920 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false); |
| 2921 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2922 | curusage = page_counter_read(&memcg->memsw); |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 2923 | /* Usage is reduced ? */ |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2924 | if (curusage >= oldusage) |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 2925 | retry_count--; |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 2926 | else |
| 2927 | oldusage = curusage; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2928 | } while (retry_count); |
| 2929 | |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 2930 | if (!ret && enlarge) |
| 2931 | memcg_oom_recover(memcg); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2932 | |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 2933 | return ret; |
| 2934 | } |
| 2935 | |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 2936 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
| 2937 | gfp_t gfp_mask, |
| 2938 | unsigned long *total_scanned) |
| 2939 | { |
| 2940 | unsigned long nr_reclaimed = 0; |
| 2941 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; |
| 2942 | unsigned long reclaimed; |
| 2943 | int loop = 0; |
| 2944 | struct mem_cgroup_tree_per_zone *mctz; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2945 | unsigned long excess; |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 2946 | unsigned long nr_scanned; |
| 2947 | |
| 2948 | if (order > 0) |
| 2949 | return 0; |
| 2950 | |
| 2951 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); |
| 2952 | /* |
| 2953 | * This loop can run a while, specially if mem_cgroup's continuously |
| 2954 | * keep exceeding their soft limit and putting the system under |
| 2955 | * pressure |
| 2956 | */ |
| 2957 | do { |
| 2958 | if (next_mz) |
| 2959 | mz = next_mz; |
| 2960 | else |
| 2961 | mz = mem_cgroup_largest_soft_limit_node(mctz); |
| 2962 | if (!mz) |
| 2963 | break; |
| 2964 | |
| 2965 | nr_scanned = 0; |
| 2966 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone, |
| 2967 | gfp_mask, &nr_scanned); |
| 2968 | nr_reclaimed += reclaimed; |
| 2969 | *total_scanned += nr_scanned; |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2970 | spin_lock_irq(&mctz->lock); |
Vladimir Davydov | bc2f2e7 | 2014-12-10 15:43:40 -0800 | [diff] [blame] | 2971 | __mem_cgroup_remove_exceeded(mz, mctz); |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 2972 | |
| 2973 | /* |
| 2974 | * If we failed to reclaim anything from this memory cgroup |
| 2975 | * it is time to move on to the next cgroup |
| 2976 | */ |
| 2977 | next_mz = NULL; |
Vladimir Davydov | bc2f2e7 | 2014-12-10 15:43:40 -0800 | [diff] [blame] | 2978 | if (!reclaimed) |
| 2979 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); |
| 2980 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2981 | excess = soft_limit_excess(mz->memcg); |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 2982 | /* |
| 2983 | * One school of thought says that we should not add |
| 2984 | * back the node to the tree if reclaim returns 0. |
| 2985 | * But our reclaim could return 0, simply because due |
| 2986 | * to priority we are exposing a smaller subset of |
| 2987 | * memory to reclaim from. Consider this as a longer |
| 2988 | * term TODO. |
| 2989 | */ |
| 2990 | /* If excess == 0, no tree ops */ |
Johannes Weiner | cf2c812 | 2014-06-06 14:38:21 -0700 | [diff] [blame] | 2991 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2992 | spin_unlock_irq(&mctz->lock); |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 2993 | css_put(&mz->memcg->css); |
| 2994 | loop++; |
| 2995 | /* |
| 2996 | * Could not reclaim anything and there are no more |
| 2997 | * mem cgroups to try or we seem to be looping without |
| 2998 | * reclaiming anything. |
| 2999 | */ |
| 3000 | if (!nr_reclaimed && |
| 3001 | (next_mz == NULL || |
| 3002 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) |
| 3003 | break; |
| 3004 | } while (!nr_reclaimed); |
| 3005 | if (next_mz) |
| 3006 | css_put(&next_mz->memcg->css); |
| 3007 | return nr_reclaimed; |
| 3008 | } |
| 3009 | |
Tejun Heo | ea280e7 | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 3010 | /* |
| 3011 | * Test whether @memcg has children, dead or alive. Note that this |
| 3012 | * function doesn't care whether @memcg has use_hierarchy enabled and |
| 3013 | * returns %true if there are child csses according to the cgroup |
| 3014 | * hierarchy. Testing use_hierarchy is the caller's responsiblity. |
| 3015 | */ |
Glauber Costa | b5f99b5 | 2013-02-22 16:34:53 -0800 | [diff] [blame] | 3016 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
| 3017 | { |
Tejun Heo | ea280e7 | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 3018 | bool ret; |
| 3019 | |
Johannes Weiner | 696ac17 | 2013-10-31 16:34:15 -0700 | [diff] [blame] | 3020 | /* |
Tejun Heo | ea280e7 | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 3021 | * The lock does not prevent addition or deletion of children, but |
| 3022 | * it prevents a new child from being initialized based on this |
| 3023 | * parent in css_online(), so it's enough to decide whether |
| 3024 | * hierarchically inherited attributes can still be changed or not. |
Johannes Weiner | 696ac17 | 2013-10-31 16:34:15 -0700 | [diff] [blame] | 3025 | */ |
Tejun Heo | ea280e7 | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 3026 | lockdep_assert_held(&memcg_create_mutex); |
| 3027 | |
| 3028 | rcu_read_lock(); |
| 3029 | ret = css_next_child(NULL, &memcg->css); |
| 3030 | rcu_read_unlock(); |
| 3031 | return ret; |
Glauber Costa | b5f99b5 | 2013-02-22 16:34:53 -0800 | [diff] [blame] | 3032 | } |
| 3033 | |
| 3034 | /* |
Michal Hocko | c26251f | 2012-10-26 13:37:28 +0200 | [diff] [blame] | 3035 | * Reclaims as many pages from the given memcg as possible and moves |
| 3036 | * the rest to the parent. |
| 3037 | * |
| 3038 | * Caller is responsible for holding css reference for memcg. |
| 3039 | */ |
| 3040 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) |
| 3041 | { |
| 3042 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
Michal Hocko | c26251f | 2012-10-26 13:37:28 +0200 | [diff] [blame] | 3043 | |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3044 | /* we call try-to-free pages for make this cgroup empty */ |
| 3045 | lru_add_drain_all(); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3046 | /* try to free all pages in this cgroup */ |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3047 | while (nr_retries && page_counter_read(&memcg->memory)) { |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3048 | int progress; |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3049 | |
Michal Hocko | c26251f | 2012-10-26 13:37:28 +0200 | [diff] [blame] | 3050 | if (signal_pending(current)) |
| 3051 | return -EINTR; |
| 3052 | |
Johannes Weiner | b70a2a2 | 2014-10-09 15:28:56 -0700 | [diff] [blame] | 3053 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
| 3054 | GFP_KERNEL, true); |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3055 | if (!progress) { |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3056 | nr_retries--; |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3057 | /* maybe some writeback is necessary */ |
Jens Axboe | 8aa7e84 | 2009-07-09 14:52:32 +0200 | [diff] [blame] | 3058 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3059 | } |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3060 | |
| 3061 | } |
Michal Hocko | ab5196c | 2012-10-26 13:37:32 +0200 | [diff] [blame] | 3062 | |
| 3063 | return 0; |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 3064 | } |
| 3065 | |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3066 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
| 3067 | char *buf, size_t nbytes, |
| 3068 | loff_t off) |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3069 | { |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3070 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
Michal Hocko | c26251f | 2012-10-26 13:37:28 +0200 | [diff] [blame] | 3071 | |
Michal Hocko | d842301 | 2012-10-26 13:37:29 +0200 | [diff] [blame] | 3072 | if (mem_cgroup_is_root(memcg)) |
| 3073 | return -EINVAL; |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3074 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3075 | } |
| 3076 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3077 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
| 3078 | struct cftype *cft) |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 3079 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3080 | return mem_cgroup_from_css(css)->use_hierarchy; |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 3081 | } |
| 3082 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3083 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
| 3084 | struct cftype *cft, u64 val) |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 3085 | { |
| 3086 | int retval = 0; |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3087 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Tejun Heo | 5c9d535 | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 3088 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 3089 | |
Glauber Costa | 0999821 | 2013-02-22 16:34:55 -0800 | [diff] [blame] | 3090 | mutex_lock(&memcg_create_mutex); |
Glauber Costa | 567fb43 | 2012-07-31 16:43:07 -0700 | [diff] [blame] | 3091 | |
| 3092 | if (memcg->use_hierarchy == val) |
| 3093 | goto out; |
| 3094 | |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 3095 | /* |
André Goddard Rosa | af901ca | 2009-11-14 13:09:05 -0200 | [diff] [blame] | 3096 | * If parent's use_hierarchy is set, we can't make any modifications |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 3097 | * in the child subtrees. If it is unset, then the change can |
| 3098 | * occur, provided the current cgroup has no children. |
| 3099 | * |
| 3100 | * For the root cgroup, parent_mem is NULL, we allow value to be |
| 3101 | * set if there are no children. |
| 3102 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3103 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 3104 | (val == 1 || val == 0)) { |
Tejun Heo | ea280e7 | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 3105 | if (!memcg_has_children(memcg)) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3106 | memcg->use_hierarchy = val; |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 3107 | else |
| 3108 | retval = -EBUSY; |
| 3109 | } else |
| 3110 | retval = -EINVAL; |
Glauber Costa | 567fb43 | 2012-07-31 16:43:07 -0700 | [diff] [blame] | 3111 | |
| 3112 | out: |
Glauber Costa | 0999821 | 2013-02-22 16:34:55 -0800 | [diff] [blame] | 3113 | mutex_unlock(&memcg_create_mutex); |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 3114 | |
| 3115 | return retval; |
| 3116 | } |
| 3117 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3118 | static unsigned long tree_stat(struct mem_cgroup *memcg, |
| 3119 | enum mem_cgroup_stat_index idx) |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 3120 | { |
| 3121 | struct mem_cgroup *iter; |
| 3122 | long val = 0; |
| 3123 | |
| 3124 | /* Per-cpu values can be negative, use a signed accumulator */ |
| 3125 | for_each_mem_cgroup_tree(iter, memcg) |
| 3126 | val += mem_cgroup_read_stat(iter, idx); |
| 3127 | |
| 3128 | if (val < 0) /* race ? */ |
| 3129 | val = 0; |
| 3130 | return val; |
| 3131 | } |
| 3132 | |
| 3133 | static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
| 3134 | { |
| 3135 | u64 val; |
| 3136 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3137 | if (mem_cgroup_is_root(memcg)) { |
| 3138 | val = tree_stat(memcg, MEM_CGROUP_STAT_CACHE); |
| 3139 | val += tree_stat(memcg, MEM_CGROUP_STAT_RSS); |
| 3140 | if (swap) |
| 3141 | val += tree_stat(memcg, MEM_CGROUP_STAT_SWAP); |
| 3142 | } else { |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 3143 | if (!swap) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3144 | val = page_counter_read(&memcg->memory); |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 3145 | else |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3146 | val = page_counter_read(&memcg->memsw); |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 3147 | } |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 3148 | return val << PAGE_SHIFT; |
| 3149 | } |
| 3150 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3151 | enum { |
| 3152 | RES_USAGE, |
| 3153 | RES_LIMIT, |
| 3154 | RES_MAX_USAGE, |
| 3155 | RES_FAILCNT, |
| 3156 | RES_SOFT_LIMIT, |
| 3157 | }; |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 3158 | |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 3159 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 3160 | struct cftype *cft) |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 3161 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3162 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3163 | struct page_counter *counter; |
Tejun Heo | af36f90 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 3164 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3165 | switch (MEMFILE_TYPE(cft->private)) { |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3166 | case _MEM: |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3167 | counter = &memcg->memory; |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 3168 | break; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3169 | case _MEMSWAP: |
| 3170 | counter = &memcg->memsw; |
| 3171 | break; |
| 3172 | case _KMEM: |
| 3173 | counter = &memcg->kmem; |
| 3174 | break; |
| 3175 | default: |
| 3176 | BUG(); |
| 3177 | } |
| 3178 | |
| 3179 | switch (MEMFILE_ATTR(cft->private)) { |
| 3180 | case RES_USAGE: |
| 3181 | if (counter == &memcg->memory) |
| 3182 | return mem_cgroup_usage(memcg, false); |
| 3183 | if (counter == &memcg->memsw) |
| 3184 | return mem_cgroup_usage(memcg, true); |
| 3185 | return (u64)page_counter_read(counter) * PAGE_SIZE; |
| 3186 | case RES_LIMIT: |
| 3187 | return (u64)counter->limit * PAGE_SIZE; |
| 3188 | case RES_MAX_USAGE: |
| 3189 | return (u64)counter->watermark * PAGE_SIZE; |
| 3190 | case RES_FAILCNT: |
| 3191 | return counter->failcnt; |
| 3192 | case RES_SOFT_LIMIT: |
| 3193 | return (u64)memcg->soft_limit * PAGE_SIZE; |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3194 | default: |
| 3195 | BUG(); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3196 | } |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 3197 | } |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 3198 | |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 3199 | #ifdef CONFIG_MEMCG_KMEM |
Vladimir Davydov | 8c0145b | 2014-12-10 15:43:48 -0800 | [diff] [blame] | 3200 | static int memcg_activate_kmem(struct mem_cgroup *memcg, |
| 3201 | unsigned long nr_pages) |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3202 | { |
| 3203 | int err = 0; |
| 3204 | int memcg_id; |
| 3205 | |
Vladimir Davydov | 2a4db7e | 2015-02-12 14:59:32 -0800 | [diff] [blame] | 3206 | BUG_ON(memcg->kmemcg_id >= 0); |
Vladimir Davydov | 2788cf0 | 2015-02-12 14:59:38 -0800 | [diff] [blame] | 3207 | BUG_ON(memcg->kmem_acct_activated); |
Vladimir Davydov | 2a4db7e | 2015-02-12 14:59:32 -0800 | [diff] [blame] | 3208 | BUG_ON(memcg->kmem_acct_active); |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3209 | |
| 3210 | /* |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 3211 | * For simplicity, we won't allow this to be disabled. It also can't |
| 3212 | * be changed if the cgroup has children already, or if tasks had |
| 3213 | * already joined. |
| 3214 | * |
| 3215 | * If tasks join before we set the limit, a person looking at |
| 3216 | * kmem.usage_in_bytes will have no way to determine when it took |
| 3217 | * place, which makes the value quite meaningless. |
| 3218 | * |
| 3219 | * After it first became limited, changes in the value of the limit are |
| 3220 | * of course permitted. |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 3221 | */ |
Glauber Costa | 0999821 | 2013-02-22 16:34:55 -0800 | [diff] [blame] | 3222 | mutex_lock(&memcg_create_mutex); |
Tejun Heo | ea280e7 | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 3223 | if (cgroup_has_tasks(memcg->css.cgroup) || |
| 3224 | (memcg->use_hierarchy && memcg_has_children(memcg))) |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3225 | err = -EBUSY; |
Glauber Costa | 0999821 | 2013-02-22 16:34:55 -0800 | [diff] [blame] | 3226 | mutex_unlock(&memcg_create_mutex); |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3227 | if (err) |
| 3228 | goto out; |
| 3229 | |
Vladimir Davydov | f3bb304 | 2014-10-09 15:28:45 -0700 | [diff] [blame] | 3230 | memcg_id = memcg_alloc_cache_id(); |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3231 | if (memcg_id < 0) { |
| 3232 | err = memcg_id; |
| 3233 | goto out; |
| 3234 | } |
| 3235 | |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3236 | /* |
Vladimir Davydov | 900a38f | 2014-12-12 16:55:10 -0800 | [diff] [blame] | 3237 | * We couldn't have accounted to this cgroup, because it hasn't got |
| 3238 | * activated yet, so this should succeed. |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3239 | */ |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3240 | err = page_counter_limit(&memcg->kmem, nr_pages); |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3241 | VM_BUG_ON(err); |
| 3242 | |
| 3243 | static_key_slow_inc(&memcg_kmem_enabled_key); |
| 3244 | /* |
Vladimir Davydov | 900a38f | 2014-12-12 16:55:10 -0800 | [diff] [blame] | 3245 | * A memory cgroup is considered kmem-active as soon as it gets |
| 3246 | * kmemcg_id. Setting the id after enabling static branching will |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3247 | * guarantee no one starts accounting before all call sites are |
| 3248 | * patched. |
| 3249 | */ |
Vladimir Davydov | 900a38f | 2014-12-12 16:55:10 -0800 | [diff] [blame] | 3250 | memcg->kmemcg_id = memcg_id; |
Vladimir Davydov | 2788cf0 | 2015-02-12 14:59:38 -0800 | [diff] [blame] | 3251 | memcg->kmem_acct_activated = true; |
Vladimir Davydov | 2a4db7e | 2015-02-12 14:59:32 -0800 | [diff] [blame] | 3252 | memcg->kmem_acct_active = true; |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3253 | out: |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3254 | return err; |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3255 | } |
| 3256 | |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3257 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3258 | unsigned long limit) |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3259 | { |
| 3260 | int ret; |
| 3261 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3262 | mutex_lock(&memcg_limit_mutex); |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3263 | if (!memcg_kmem_is_active(memcg)) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3264 | ret = memcg_activate_kmem(memcg, limit); |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3265 | else |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3266 | ret = page_counter_limit(&memcg->kmem, limit); |
| 3267 | mutex_unlock(&memcg_limit_mutex); |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3268 | return ret; |
| 3269 | } |
| 3270 | |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 3271 | static int memcg_propagate_kmem(struct mem_cgroup *memcg) |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 3272 | { |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 3273 | int ret = 0; |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 3274 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3275 | |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 3276 | if (!parent) |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3277 | return 0; |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 3278 | |
Vladimir Davydov | 8c0145b | 2014-12-10 15:43:48 -0800 | [diff] [blame] | 3279 | mutex_lock(&memcg_limit_mutex); |
Glauber Costa | a8964b9 | 2012-12-18 14:22:09 -0800 | [diff] [blame] | 3280 | /* |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3281 | * If the parent cgroup is not kmem-active now, it cannot be activated |
| 3282 | * after this point, because it has at least one child already. |
Glauber Costa | a8964b9 | 2012-12-18 14:22:09 -0800 | [diff] [blame] | 3283 | */ |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3284 | if (memcg_kmem_is_active(parent)) |
Vladimir Davydov | 8c0145b | 2014-12-10 15:43:48 -0800 | [diff] [blame] | 3285 | ret = memcg_activate_kmem(memcg, PAGE_COUNTER_MAX); |
| 3286 | mutex_unlock(&memcg_limit_mutex); |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 3287 | return ret; |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 3288 | } |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3289 | #else |
| 3290 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3291 | unsigned long limit) |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3292 | { |
| 3293 | return -EINVAL; |
| 3294 | } |
Hugh Dickins | 6d043990 | 2013-02-22 16:35:50 -0800 | [diff] [blame] | 3295 | #endif /* CONFIG_MEMCG_KMEM */ |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 3296 | |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3297 | /* |
| 3298 | * The user of this function is... |
| 3299 | * RES_LIMIT. |
| 3300 | */ |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3301 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
| 3302 | char *buf, size_t nbytes, loff_t off) |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 3303 | { |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3304 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3305 | unsigned long nr_pages; |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3306 | int ret; |
| 3307 | |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3308 | buf = strstrip(buf); |
Johannes Weiner | 650c5e5 | 2015-02-11 15:26:03 -0800 | [diff] [blame] | 3309 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3310 | if (ret) |
| 3311 | return ret; |
Tejun Heo | af36f90 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 3312 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3313 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3314 | case RES_LIMIT: |
Balbir Singh | 4b3bde4 | 2009-09-23 15:56:32 -0700 | [diff] [blame] | 3315 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
| 3316 | ret = -EINVAL; |
| 3317 | break; |
| 3318 | } |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3319 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
| 3320 | case _MEM: |
| 3321 | ret = mem_cgroup_resize_limit(memcg, nr_pages); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3322 | break; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3323 | case _MEMSWAP: |
| 3324 | ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages); |
| 3325 | break; |
| 3326 | case _KMEM: |
| 3327 | ret = memcg_update_kmem_limit(memcg, nr_pages); |
| 3328 | break; |
| 3329 | } |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3330 | break; |
Balbir Singh | 296c81d | 2009-09-23 15:56:36 -0700 | [diff] [blame] | 3331 | case RES_SOFT_LIMIT: |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3332 | memcg->soft_limit = nr_pages; |
| 3333 | ret = 0; |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3334 | break; |
| 3335 | } |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3336 | return ret ?: nbytes; |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 3337 | } |
| 3338 | |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3339 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
| 3340 | size_t nbytes, loff_t off) |
Pavel Emelyanov | c84872e | 2008-04-29 01:00:17 -0700 | [diff] [blame] | 3341 | { |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3342 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3343 | struct page_counter *counter; |
Pavel Emelyanov | c84872e | 2008-04-29 01:00:17 -0700 | [diff] [blame] | 3344 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3345 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
| 3346 | case _MEM: |
| 3347 | counter = &memcg->memory; |
| 3348 | break; |
| 3349 | case _MEMSWAP: |
| 3350 | counter = &memcg->memsw; |
| 3351 | break; |
| 3352 | case _KMEM: |
| 3353 | counter = &memcg->kmem; |
| 3354 | break; |
| 3355 | default: |
| 3356 | BUG(); |
| 3357 | } |
Tejun Heo | af36f90 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 3358 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3359 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
Pavel Emelyanov | 29f2a4d | 2008-04-29 01:00:21 -0700 | [diff] [blame] | 3360 | case RES_MAX_USAGE: |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3361 | page_counter_reset_watermark(counter); |
Pavel Emelyanov | 29f2a4d | 2008-04-29 01:00:21 -0700 | [diff] [blame] | 3362 | break; |
| 3363 | case RES_FAILCNT: |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3364 | counter->failcnt = 0; |
Pavel Emelyanov | 29f2a4d | 2008-04-29 01:00:21 -0700 | [diff] [blame] | 3365 | break; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3366 | default: |
| 3367 | BUG(); |
Pavel Emelyanov | 29f2a4d | 2008-04-29 01:00:21 -0700 | [diff] [blame] | 3368 | } |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 3369 | |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3370 | return nbytes; |
Pavel Emelyanov | c84872e | 2008-04-29 01:00:17 -0700 | [diff] [blame] | 3371 | } |
| 3372 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3373 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3374 | struct cftype *cft) |
| 3375 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3376 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3377 | } |
| 3378 | |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 3379 | #ifdef CONFIG_MMU |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3380 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3381 | struct cftype *cft, u64 val) |
| 3382 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3383 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3384 | |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 3385 | if (val & ~MOVE_MASK) |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3386 | return -EINVAL; |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3387 | |
Glauber Costa | ee5e847 | 2013-02-22 16:34:50 -0800 | [diff] [blame] | 3388 | /* |
| 3389 | * No kind of locking is needed in here, because ->can_attach() will |
| 3390 | * check this value once in the beginning of the process, and then carry |
| 3391 | * on with stale data. This means that changes to this value will only |
| 3392 | * affect task migrations starting after the change. |
| 3393 | */ |
| 3394 | memcg->move_charge_at_immigrate = val; |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3395 | return 0; |
| 3396 | } |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 3397 | #else |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3398 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 3399 | struct cftype *cft, u64 val) |
| 3400 | { |
| 3401 | return -ENOSYS; |
| 3402 | } |
| 3403 | #endif |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3404 | |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3405 | #ifdef CONFIG_NUMA |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 3406 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3407 | { |
Greg Thelen | 25485de | 2013-11-12 15:07:40 -0800 | [diff] [blame] | 3408 | struct numa_stat { |
| 3409 | const char *name; |
| 3410 | unsigned int lru_mask; |
| 3411 | }; |
| 3412 | |
| 3413 | static const struct numa_stat stats[] = { |
| 3414 | { "total", LRU_ALL }, |
| 3415 | { "file", LRU_ALL_FILE }, |
| 3416 | { "anon", LRU_ALL_ANON }, |
| 3417 | { "unevictable", BIT(LRU_UNEVICTABLE) }, |
| 3418 | }; |
| 3419 | const struct numa_stat *stat; |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3420 | int nid; |
Greg Thelen | 25485de | 2013-11-12 15:07:40 -0800 | [diff] [blame] | 3421 | unsigned long nr; |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 3422 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3423 | |
Greg Thelen | 25485de | 2013-11-12 15:07:40 -0800 | [diff] [blame] | 3424 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
| 3425 | nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask); |
| 3426 | seq_printf(m, "%s=%lu", stat->name, nr); |
| 3427 | for_each_node_state(nid, N_MEMORY) { |
| 3428 | nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
| 3429 | stat->lru_mask); |
| 3430 | seq_printf(m, " N%d=%lu", nid, nr); |
| 3431 | } |
| 3432 | seq_putc(m, '\n'); |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3433 | } |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3434 | |
Ying Han | 071aee1 | 2013-11-12 15:07:41 -0800 | [diff] [blame] | 3435 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
| 3436 | struct mem_cgroup *iter; |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3437 | |
Ying Han | 071aee1 | 2013-11-12 15:07:41 -0800 | [diff] [blame] | 3438 | nr = 0; |
| 3439 | for_each_mem_cgroup_tree(iter, memcg) |
| 3440 | nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask); |
| 3441 | seq_printf(m, "hierarchical_%s=%lu", stat->name, nr); |
| 3442 | for_each_node_state(nid, N_MEMORY) { |
| 3443 | nr = 0; |
| 3444 | for_each_mem_cgroup_tree(iter, memcg) |
| 3445 | nr += mem_cgroup_node_nr_lru_pages( |
| 3446 | iter, nid, stat->lru_mask); |
| 3447 | seq_printf(m, " N%d=%lu", nid, nr); |
| 3448 | } |
| 3449 | seq_putc(m, '\n'); |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3450 | } |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3451 | |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3452 | return 0; |
| 3453 | } |
| 3454 | #endif /* CONFIG_NUMA */ |
| 3455 | |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 3456 | static int memcg_stat_show(struct seq_file *m, void *v) |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 3457 | { |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 3458 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3459 | unsigned long memory, memsw; |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 3460 | struct mem_cgroup *mi; |
| 3461 | unsigned int i; |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 3462 | |
Greg Thelen | 0ca44b1 | 2015-02-11 15:25:58 -0800 | [diff] [blame] | 3463 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_stat_names) != |
| 3464 | MEM_CGROUP_STAT_NSTATS); |
| 3465 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_events_names) != |
| 3466 | MEM_CGROUP_EVENTS_NSTATS); |
Rickard Strandqvist | 70bc068 | 2014-12-12 16:56:41 -0800 | [diff] [blame] | 3467 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); |
| 3468 | |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 3469 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
Kamezawa Hiroyuki | bff6bb8 | 2012-07-31 16:41:38 -0700 | [diff] [blame] | 3470 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
Daisuke Nishimura | 1dd3a27 | 2009-09-23 15:56:43 -0700 | [diff] [blame] | 3471 | continue; |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 3472 | seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i], |
| 3473 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); |
Daisuke Nishimura | 1dd3a27 | 2009-09-23 15:56:43 -0700 | [diff] [blame] | 3474 | } |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 3475 | |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 3476 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
| 3477 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], |
| 3478 | mem_cgroup_read_events(memcg, i)); |
| 3479 | |
| 3480 | for (i = 0; i < NR_LRU_LISTS; i++) |
| 3481 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], |
| 3482 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); |
| 3483 | |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3484 | /* Hierarchical information */ |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3485 | memory = memsw = PAGE_COUNTER_MAX; |
| 3486 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { |
| 3487 | memory = min(memory, mi->memory.limit); |
| 3488 | memsw = min(memsw, mi->memsw.limit); |
KAMEZAWA Hiroyuki | fee7b54 | 2009-01-07 18:08:26 -0800 | [diff] [blame] | 3489 | } |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3490 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
| 3491 | (u64)memory * PAGE_SIZE); |
| 3492 | if (do_swap_account) |
| 3493 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
| 3494 | (u64)memsw * PAGE_SIZE); |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 3495 | |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 3496 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
| 3497 | long long val = 0; |
| 3498 | |
Kamezawa Hiroyuki | bff6bb8 | 2012-07-31 16:41:38 -0700 | [diff] [blame] | 3499 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
Daisuke Nishimura | 1dd3a27 | 2009-09-23 15:56:43 -0700 | [diff] [blame] | 3500 | continue; |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 3501 | for_each_mem_cgroup_tree(mi, memcg) |
| 3502 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; |
| 3503 | seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val); |
| 3504 | } |
| 3505 | |
| 3506 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { |
| 3507 | unsigned long long val = 0; |
| 3508 | |
| 3509 | for_each_mem_cgroup_tree(mi, memcg) |
| 3510 | val += mem_cgroup_read_events(mi, i); |
| 3511 | seq_printf(m, "total_%s %llu\n", |
| 3512 | mem_cgroup_events_names[i], val); |
| 3513 | } |
| 3514 | |
| 3515 | for (i = 0; i < NR_LRU_LISTS; i++) { |
| 3516 | unsigned long long val = 0; |
| 3517 | |
| 3518 | for_each_mem_cgroup_tree(mi, memcg) |
| 3519 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; |
| 3520 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); |
Daisuke Nishimura | 1dd3a27 | 2009-09-23 15:56:43 -0700 | [diff] [blame] | 3521 | } |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3522 | |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 3523 | #ifdef CONFIG_DEBUG_VM |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 3524 | { |
| 3525 | int nid, zid; |
| 3526 | struct mem_cgroup_per_zone *mz; |
Hugh Dickins | 89abfab | 2012-05-29 15:06:53 -0700 | [diff] [blame] | 3527 | struct zone_reclaim_stat *rstat; |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 3528 | unsigned long recent_rotated[2] = {0, 0}; |
| 3529 | unsigned long recent_scanned[2] = {0, 0}; |
| 3530 | |
| 3531 | for_each_online_node(nid) |
| 3532 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 3533 | mz = &memcg->nodeinfo[nid]->zoneinfo[zid]; |
Hugh Dickins | 89abfab | 2012-05-29 15:06:53 -0700 | [diff] [blame] | 3534 | rstat = &mz->lruvec.reclaim_stat; |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 3535 | |
Hugh Dickins | 89abfab | 2012-05-29 15:06:53 -0700 | [diff] [blame] | 3536 | recent_rotated[0] += rstat->recent_rotated[0]; |
| 3537 | recent_rotated[1] += rstat->recent_rotated[1]; |
| 3538 | recent_scanned[0] += rstat->recent_scanned[0]; |
| 3539 | recent_scanned[1] += rstat->recent_scanned[1]; |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 3540 | } |
Johannes Weiner | 78ccf5b | 2012-05-29 15:07:06 -0700 | [diff] [blame] | 3541 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
| 3542 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); |
| 3543 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); |
| 3544 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 3545 | } |
| 3546 | #endif |
| 3547 | |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 3548 | return 0; |
| 3549 | } |
| 3550 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3551 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
| 3552 | struct cftype *cft) |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 3553 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3554 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 3555 | |
KAMEZAWA Hiroyuki | 1f4c025 | 2011-07-26 16:08:21 -0700 | [diff] [blame] | 3556 | return mem_cgroup_swappiness(memcg); |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 3557 | } |
| 3558 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3559 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
| 3560 | struct cftype *cft, u64 val) |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 3561 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3562 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Li Zefan | 068b38c | 2009-01-15 13:51:26 -0800 | [diff] [blame] | 3563 | |
Johannes Weiner | 3dae7fe | 2014-06-04 16:07:01 -0700 | [diff] [blame] | 3564 | if (val > 100) |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 3565 | return -EINVAL; |
| 3566 | |
Linus Torvalds | 14208b0 | 2014-06-09 15:03:33 -0700 | [diff] [blame] | 3567 | if (css->parent) |
Johannes Weiner | 3dae7fe | 2014-06-04 16:07:01 -0700 | [diff] [blame] | 3568 | memcg->swappiness = val; |
| 3569 | else |
| 3570 | vm_swappiness = val; |
Li Zefan | 068b38c | 2009-01-15 13:51:26 -0800 | [diff] [blame] | 3571 | |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 3572 | return 0; |
| 3573 | } |
| 3574 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3575 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
| 3576 | { |
| 3577 | struct mem_cgroup_threshold_ary *t; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3578 | unsigned long usage; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3579 | int i; |
| 3580 | |
| 3581 | rcu_read_lock(); |
| 3582 | if (!swap) |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3583 | t = rcu_dereference(memcg->thresholds.primary); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3584 | else |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3585 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3586 | |
| 3587 | if (!t) |
| 3588 | goto unlock; |
| 3589 | |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 3590 | usage = mem_cgroup_usage(memcg, swap); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3591 | |
| 3592 | /* |
Sha Zhengju | 748dad3 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 3593 | * current_threshold points to threshold just below or equal to usage. |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3594 | * If it's not true, a threshold was crossed after last |
| 3595 | * call of __mem_cgroup_threshold(). |
| 3596 | */ |
Phil Carmody | 5407a56 | 2010-05-26 14:42:42 -0700 | [diff] [blame] | 3597 | i = t->current_threshold; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3598 | |
| 3599 | /* |
| 3600 | * Iterate backward over array of thresholds starting from |
| 3601 | * current_threshold and check if a threshold is crossed. |
| 3602 | * If none of thresholds below usage is crossed, we read |
| 3603 | * only one element of the array here. |
| 3604 | */ |
| 3605 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) |
| 3606 | eventfd_signal(t->entries[i].eventfd, 1); |
| 3607 | |
| 3608 | /* i = current_threshold + 1 */ |
| 3609 | i++; |
| 3610 | |
| 3611 | /* |
| 3612 | * Iterate forward over array of thresholds starting from |
| 3613 | * current_threshold+1 and check if a threshold is crossed. |
| 3614 | * If none of thresholds above usage is crossed, we read |
| 3615 | * only one element of the array here. |
| 3616 | */ |
| 3617 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) |
| 3618 | eventfd_signal(t->entries[i].eventfd, 1); |
| 3619 | |
| 3620 | /* Update current_threshold */ |
Phil Carmody | 5407a56 | 2010-05-26 14:42:42 -0700 | [diff] [blame] | 3621 | t->current_threshold = i - 1; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3622 | unlock: |
| 3623 | rcu_read_unlock(); |
| 3624 | } |
| 3625 | |
| 3626 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) |
| 3627 | { |
Kirill A. Shutemov | ad4ca5f | 2010-10-07 12:59:27 -0700 | [diff] [blame] | 3628 | while (memcg) { |
| 3629 | __mem_cgroup_threshold(memcg, false); |
| 3630 | if (do_swap_account) |
| 3631 | __mem_cgroup_threshold(memcg, true); |
| 3632 | |
| 3633 | memcg = parent_mem_cgroup(memcg); |
| 3634 | } |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3635 | } |
| 3636 | |
| 3637 | static int compare_thresholds(const void *a, const void *b) |
| 3638 | { |
| 3639 | const struct mem_cgroup_threshold *_a = a; |
| 3640 | const struct mem_cgroup_threshold *_b = b; |
| 3641 | |
Greg Thelen | 2bff24a | 2013-09-11 14:23:08 -0700 | [diff] [blame] | 3642 | if (_a->threshold > _b->threshold) |
| 3643 | return 1; |
| 3644 | |
| 3645 | if (_a->threshold < _b->threshold) |
| 3646 | return -1; |
| 3647 | |
| 3648 | return 0; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3649 | } |
| 3650 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3651 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3652 | { |
| 3653 | struct mem_cgroup_eventfd_list *ev; |
| 3654 | |
Michal Hocko | 2bcf2e9 | 2014-07-30 16:08:33 -0700 | [diff] [blame] | 3655 | spin_lock(&memcg_oom_lock); |
| 3656 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3657 | list_for_each_entry(ev, &memcg->oom_notify, list) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3658 | eventfd_signal(ev->eventfd, 1); |
Michal Hocko | 2bcf2e9 | 2014-07-30 16:08:33 -0700 | [diff] [blame] | 3659 | |
| 3660 | spin_unlock(&memcg_oom_lock); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3661 | return 0; |
| 3662 | } |
| 3663 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3664 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3665 | { |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 3666 | struct mem_cgroup *iter; |
| 3667 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3668 | for_each_mem_cgroup_tree(iter, memcg) |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 3669 | mem_cgroup_oom_notify_cb(iter); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3670 | } |
| 3671 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3672 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3673 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3674 | { |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3675 | struct mem_cgroup_thresholds *thresholds; |
| 3676 | struct mem_cgroup_threshold_ary *new; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3677 | unsigned long threshold; |
| 3678 | unsigned long usage; |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3679 | int i, size, ret; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3680 | |
Johannes Weiner | 650c5e5 | 2015-02-11 15:26:03 -0800 | [diff] [blame] | 3681 | ret = page_counter_memparse(args, "-1", &threshold); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3682 | if (ret) |
| 3683 | return ret; |
| 3684 | |
| 3685 | mutex_lock(&memcg->thresholds_lock); |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3686 | |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 3687 | if (type == _MEM) { |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3688 | thresholds = &memcg->thresholds; |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 3689 | usage = mem_cgroup_usage(memcg, false); |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 3690 | } else if (type == _MEMSWAP) { |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3691 | thresholds = &memcg->memsw_thresholds; |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 3692 | usage = mem_cgroup_usage(memcg, true); |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 3693 | } else |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3694 | BUG(); |
| 3695 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3696 | /* Check if a threshold crossed before adding a new one */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3697 | if (thresholds->primary) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3698 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
| 3699 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3700 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3701 | |
| 3702 | /* Allocate memory for new array of thresholds */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3703 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3704 | GFP_KERNEL); |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3705 | if (!new) { |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3706 | ret = -ENOMEM; |
| 3707 | goto unlock; |
| 3708 | } |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3709 | new->size = size; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3710 | |
| 3711 | /* Copy thresholds (if any) to new array */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3712 | if (thresholds->primary) { |
| 3713 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3714 | sizeof(struct mem_cgroup_threshold)); |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3715 | } |
| 3716 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3717 | /* Add new threshold */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3718 | new->entries[size - 1].eventfd = eventfd; |
| 3719 | new->entries[size - 1].threshold = threshold; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3720 | |
| 3721 | /* Sort thresholds. Registering of new threshold isn't time-critical */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3722 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3723 | compare_thresholds, NULL); |
| 3724 | |
| 3725 | /* Find current threshold */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3726 | new->current_threshold = -1; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3727 | for (i = 0; i < size; i++) { |
Sha Zhengju | 748dad3 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 3728 | if (new->entries[i].threshold <= usage) { |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3729 | /* |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3730 | * new->current_threshold will not be used until |
| 3731 | * rcu_assign_pointer(), so it's safe to increment |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3732 | * it here. |
| 3733 | */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3734 | ++new->current_threshold; |
Sha Zhengju | 748dad3 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 3735 | } else |
| 3736 | break; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3737 | } |
| 3738 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3739 | /* Free old spare buffer and save old primary buffer as spare */ |
| 3740 | kfree(thresholds->spare); |
| 3741 | thresholds->spare = thresholds->primary; |
| 3742 | |
| 3743 | rcu_assign_pointer(thresholds->primary, new); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3744 | |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3745 | /* To be sure that nobody uses thresholds */ |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3746 | synchronize_rcu(); |
| 3747 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3748 | unlock: |
| 3749 | mutex_unlock(&memcg->thresholds_lock); |
| 3750 | |
| 3751 | return ret; |
| 3752 | } |
| 3753 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3754 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3755 | struct eventfd_ctx *eventfd, const char *args) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3756 | { |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3757 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3758 | } |
| 3759 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3760 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3761 | struct eventfd_ctx *eventfd, const char *args) |
| 3762 | { |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3763 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3764 | } |
| 3765 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3766 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3767 | struct eventfd_ctx *eventfd, enum res_type type) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3768 | { |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3769 | struct mem_cgroup_thresholds *thresholds; |
| 3770 | struct mem_cgroup_threshold_ary *new; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3771 | unsigned long usage; |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3772 | int i, j, size; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3773 | |
| 3774 | mutex_lock(&memcg->thresholds_lock); |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 3775 | |
| 3776 | if (type == _MEM) { |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3777 | thresholds = &memcg->thresholds; |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 3778 | usage = mem_cgroup_usage(memcg, false); |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 3779 | } else if (type == _MEMSWAP) { |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3780 | thresholds = &memcg->memsw_thresholds; |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 3781 | usage = mem_cgroup_usage(memcg, true); |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 3782 | } else |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3783 | BUG(); |
| 3784 | |
Anton Vorontsov | 371528c | 2012-02-24 05:14:46 +0400 | [diff] [blame] | 3785 | if (!thresholds->primary) |
| 3786 | goto unlock; |
| 3787 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3788 | /* Check if a threshold crossed before removing */ |
| 3789 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
| 3790 | |
| 3791 | /* Calculate new number of threshold */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3792 | size = 0; |
| 3793 | for (i = 0; i < thresholds->primary->size; i++) { |
| 3794 | if (thresholds->primary->entries[i].eventfd != eventfd) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3795 | size++; |
| 3796 | } |
| 3797 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3798 | new = thresholds->spare; |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3799 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3800 | /* Set thresholds array to NULL if we don't have thresholds */ |
| 3801 | if (!size) { |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3802 | kfree(new); |
| 3803 | new = NULL; |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3804 | goto swap_buffers; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3805 | } |
| 3806 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3807 | new->size = size; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3808 | |
| 3809 | /* Copy thresholds and find current threshold */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3810 | new->current_threshold = -1; |
| 3811 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { |
| 3812 | if (thresholds->primary->entries[i].eventfd == eventfd) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3813 | continue; |
| 3814 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3815 | new->entries[j] = thresholds->primary->entries[i]; |
Sha Zhengju | 748dad3 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 3816 | if (new->entries[j].threshold <= usage) { |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3817 | /* |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3818 | * new->current_threshold will not be used |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3819 | * until rcu_assign_pointer(), so it's safe to increment |
| 3820 | * it here. |
| 3821 | */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3822 | ++new->current_threshold; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3823 | } |
| 3824 | j++; |
| 3825 | } |
| 3826 | |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3827 | swap_buffers: |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3828 | /* Swap primary and spare array */ |
| 3829 | thresholds->spare = thresholds->primary; |
Sha Zhengju | 8c75776 | 2012-05-10 13:01:45 -0700 | [diff] [blame] | 3830 | /* If all events are unregistered, free the spare array */ |
| 3831 | if (!new) { |
| 3832 | kfree(thresholds->spare); |
| 3833 | thresholds->spare = NULL; |
| 3834 | } |
| 3835 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3836 | rcu_assign_pointer(thresholds->primary, new); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3837 | |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3838 | /* To be sure that nobody uses thresholds */ |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3839 | synchronize_rcu(); |
Anton Vorontsov | 371528c | 2012-02-24 05:14:46 +0400 | [diff] [blame] | 3840 | unlock: |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3841 | mutex_unlock(&memcg->thresholds_lock); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3842 | } |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3843 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3844 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3845 | struct eventfd_ctx *eventfd) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3846 | { |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3847 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3848 | } |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3849 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3850 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3851 | struct eventfd_ctx *eventfd) |
| 3852 | { |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3853 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3854 | } |
| 3855 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3856 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3857 | struct eventfd_ctx *eventfd, const char *args) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3858 | { |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3859 | struct mem_cgroup_eventfd_list *event; |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3860 | |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3861 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
| 3862 | if (!event) |
| 3863 | return -ENOMEM; |
| 3864 | |
Michal Hocko | 1af8efe | 2011-07-26 16:08:24 -0700 | [diff] [blame] | 3865 | spin_lock(&memcg_oom_lock); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3866 | |
| 3867 | event->eventfd = eventfd; |
| 3868 | list_add(&event->list, &memcg->oom_notify); |
| 3869 | |
| 3870 | /* already in OOM ? */ |
Tejun Heo | c2b42d3 | 2015-06-24 16:58:23 -0700 | [diff] [blame] | 3871 | if (memcg->under_oom) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3872 | eventfd_signal(eventfd, 1); |
Michal Hocko | 1af8efe | 2011-07-26 16:08:24 -0700 | [diff] [blame] | 3873 | spin_unlock(&memcg_oom_lock); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3874 | |
| 3875 | return 0; |
| 3876 | } |
| 3877 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3878 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3879 | struct eventfd_ctx *eventfd) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3880 | { |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3881 | struct mem_cgroup_eventfd_list *ev, *tmp; |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3882 | |
Michal Hocko | 1af8efe | 2011-07-26 16:08:24 -0700 | [diff] [blame] | 3883 | spin_lock(&memcg_oom_lock); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3884 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3885 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3886 | if (ev->eventfd == eventfd) { |
| 3887 | list_del(&ev->list); |
| 3888 | kfree(ev); |
| 3889 | } |
| 3890 | } |
| 3891 | |
Michal Hocko | 1af8efe | 2011-07-26 16:08:24 -0700 | [diff] [blame] | 3892 | spin_unlock(&memcg_oom_lock); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3893 | } |
| 3894 | |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 3895 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3896 | { |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 3897 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf)); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3898 | |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 3899 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
Tejun Heo | c2b42d3 | 2015-06-24 16:58:23 -0700 | [diff] [blame] | 3900 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3901 | return 0; |
| 3902 | } |
| 3903 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3904 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3905 | struct cftype *cft, u64 val) |
| 3906 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3907 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3908 | |
| 3909 | /* cannot set to root cgroup and only 0 and 1 are allowed */ |
Linus Torvalds | 14208b0 | 2014-06-09 15:03:33 -0700 | [diff] [blame] | 3910 | if (!css->parent || !((val == 0) || (val == 1))) |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3911 | return -EINVAL; |
| 3912 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3913 | memcg->oom_kill_disable = val; |
KAMEZAWA Hiroyuki | 4d845eb | 2010-06-29 15:05:18 -0700 | [diff] [blame] | 3914 | if (!val) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3915 | memcg_oom_recover(memcg); |
Johannes Weiner | 3dae7fe | 2014-06-04 16:07:01 -0700 | [diff] [blame] | 3916 | |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3917 | return 0; |
| 3918 | } |
| 3919 | |
Andrew Morton | c255a45 | 2012-07-31 16:43:02 -0700 | [diff] [blame] | 3920 | #ifdef CONFIG_MEMCG_KMEM |
Glauber Costa | cbe128e3 | 2012-04-09 19:36:34 -0300 | [diff] [blame] | 3921 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
Glauber Costa | e5671df | 2011-12-11 21:47:01 +0000 | [diff] [blame] | 3922 | { |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 3923 | int ret; |
| 3924 | |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 3925 | ret = memcg_propagate_kmem(memcg); |
| 3926 | if (ret) |
| 3927 | return ret; |
Glauber Costa | 2633d7a | 2012-12-18 14:22:34 -0800 | [diff] [blame] | 3928 | |
Glauber Costa | 1d62e43 | 2012-04-09 19:36:33 -0300 | [diff] [blame] | 3929 | return mem_cgroup_sockets_init(memcg, ss); |
Michel Lespinasse | 573b400 | 2013-04-29 15:08:13 -0700 | [diff] [blame] | 3930 | } |
Glauber Costa | e5671df | 2011-12-11 21:47:01 +0000 | [diff] [blame] | 3931 | |
Vladimir Davydov | 2a4db7e | 2015-02-12 14:59:32 -0800 | [diff] [blame] | 3932 | static void memcg_deactivate_kmem(struct mem_cgroup *memcg) |
| 3933 | { |
Vladimir Davydov | 2788cf0 | 2015-02-12 14:59:38 -0800 | [diff] [blame] | 3934 | struct cgroup_subsys_state *css; |
| 3935 | struct mem_cgroup *parent, *child; |
| 3936 | int kmemcg_id; |
| 3937 | |
Vladimir Davydov | 2a4db7e | 2015-02-12 14:59:32 -0800 | [diff] [blame] | 3938 | if (!memcg->kmem_acct_active) |
| 3939 | return; |
| 3940 | |
| 3941 | /* |
| 3942 | * Clear the 'active' flag before clearing memcg_caches arrays entries. |
| 3943 | * Since we take the slab_mutex in memcg_deactivate_kmem_caches(), it |
| 3944 | * guarantees no cache will be created for this cgroup after we are |
| 3945 | * done (see memcg_create_kmem_cache()). |
| 3946 | */ |
| 3947 | memcg->kmem_acct_active = false; |
| 3948 | |
| 3949 | memcg_deactivate_kmem_caches(memcg); |
Vladimir Davydov | 2788cf0 | 2015-02-12 14:59:38 -0800 | [diff] [blame] | 3950 | |
| 3951 | kmemcg_id = memcg->kmemcg_id; |
| 3952 | BUG_ON(kmemcg_id < 0); |
| 3953 | |
| 3954 | parent = parent_mem_cgroup(memcg); |
| 3955 | if (!parent) |
| 3956 | parent = root_mem_cgroup; |
| 3957 | |
| 3958 | /* |
| 3959 | * Change kmemcg_id of this cgroup and all its descendants to the |
| 3960 | * parent's id, and then move all entries from this cgroup's list_lrus |
| 3961 | * to ones of the parent. After we have finished, all list_lrus |
| 3962 | * corresponding to this cgroup are guaranteed to remain empty. The |
| 3963 | * ordering is imposed by list_lru_node->lock taken by |
| 3964 | * memcg_drain_all_list_lrus(). |
| 3965 | */ |
| 3966 | css_for_each_descendant_pre(css, &memcg->css) { |
| 3967 | child = mem_cgroup_from_css(css); |
| 3968 | BUG_ON(child->kmemcg_id != kmemcg_id); |
| 3969 | child->kmemcg_id = parent->kmemcg_id; |
| 3970 | if (!memcg->use_hierarchy) |
| 3971 | break; |
| 3972 | } |
| 3973 | memcg_drain_all_list_lrus(kmemcg_id, parent->kmemcg_id); |
| 3974 | |
| 3975 | memcg_free_cache_id(kmemcg_id); |
Vladimir Davydov | 2a4db7e | 2015-02-12 14:59:32 -0800 | [diff] [blame] | 3976 | } |
| 3977 | |
Li Zefan | 10d5ebf | 2013-07-08 16:00:33 -0700 | [diff] [blame] | 3978 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
Glauber Costa | d1a4c0b | 2011-12-11 21:47:04 +0000 | [diff] [blame] | 3979 | { |
Vladimir Davydov | f48b80a | 2015-02-12 14:59:56 -0800 | [diff] [blame] | 3980 | if (memcg->kmem_acct_activated) { |
| 3981 | memcg_destroy_kmem_caches(memcg); |
| 3982 | static_key_slow_dec(&memcg_kmem_enabled_key); |
| 3983 | WARN_ON(page_counter_read(&memcg->kmem)); |
| 3984 | } |
Glauber Costa | 1d62e43 | 2012-04-09 19:36:33 -0300 | [diff] [blame] | 3985 | mem_cgroup_sockets_destroy(memcg); |
Li Zefan | 10d5ebf | 2013-07-08 16:00:33 -0700 | [diff] [blame] | 3986 | } |
Glauber Costa | e5671df | 2011-12-11 21:47:01 +0000 | [diff] [blame] | 3987 | #else |
Glauber Costa | cbe128e3 | 2012-04-09 19:36:34 -0300 | [diff] [blame] | 3988 | static int memcg_init_kmem(struct mem_cgroup *memcg, struct cgroup_subsys *ss) |
Glauber Costa | e5671df | 2011-12-11 21:47:01 +0000 | [diff] [blame] | 3989 | { |
| 3990 | return 0; |
| 3991 | } |
Glauber Costa | d1a4c0b | 2011-12-11 21:47:04 +0000 | [diff] [blame] | 3992 | |
Vladimir Davydov | 2a4db7e | 2015-02-12 14:59:32 -0800 | [diff] [blame] | 3993 | static void memcg_deactivate_kmem(struct mem_cgroup *memcg) |
| 3994 | { |
| 3995 | } |
| 3996 | |
Li Zefan | 10d5ebf | 2013-07-08 16:00:33 -0700 | [diff] [blame] | 3997 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
| 3998 | { |
| 3999 | } |
Glauber Costa | e5671df | 2011-12-11 21:47:01 +0000 | [diff] [blame] | 4000 | #endif |
| 4001 | |
Tejun Heo | 52ebea7 | 2015-05-22 17:13:37 -0400 | [diff] [blame] | 4002 | #ifdef CONFIG_CGROUP_WRITEBACK |
| 4003 | |
| 4004 | struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg) |
| 4005 | { |
| 4006 | return &memcg->cgwb_list; |
| 4007 | } |
| 4008 | |
Tejun Heo | 841710a | 2015-05-22 18:23:33 -0400 | [diff] [blame] | 4009 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
| 4010 | { |
| 4011 | return wb_domain_init(&memcg->cgwb_domain, gfp); |
| 4012 | } |
| 4013 | |
| 4014 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) |
| 4015 | { |
| 4016 | wb_domain_exit(&memcg->cgwb_domain); |
| 4017 | } |
| 4018 | |
Tejun Heo | 2529bb3 | 2015-05-22 18:23:34 -0400 | [diff] [blame] | 4019 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
| 4020 | { |
| 4021 | wb_domain_size_changed(&memcg->cgwb_domain); |
| 4022 | } |
| 4023 | |
Tejun Heo | 841710a | 2015-05-22 18:23:33 -0400 | [diff] [blame] | 4024 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
| 4025 | { |
| 4026 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); |
| 4027 | |
| 4028 | if (!memcg->css.parent) |
| 4029 | return NULL; |
| 4030 | |
| 4031 | return &memcg->cgwb_domain; |
| 4032 | } |
| 4033 | |
Tejun Heo | c2aa723 | 2015-05-22 18:23:35 -0400 | [diff] [blame] | 4034 | /** |
| 4035 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg |
| 4036 | * @wb: bdi_writeback in question |
| 4037 | * @pavail: out parameter for number of available pages |
| 4038 | * @pdirty: out parameter for number of dirty pages |
| 4039 | * @pwriteback: out parameter for number of pages under writeback |
| 4040 | * |
| 4041 | * Determine the numbers of available, dirty, and writeback pages in @wb's |
| 4042 | * memcg. Dirty and writeback are self-explanatory. Available is a bit |
| 4043 | * more involved. |
| 4044 | * |
| 4045 | * A memcg's headroom is "min(max, high) - used". The available memory is |
| 4046 | * calculated as the lowest headroom of itself and the ancestors plus the |
| 4047 | * number of pages already being used for file pages. Note that this |
| 4048 | * doesn't consider the actual amount of available memory in the system. |
| 4049 | * The caller should further cap *@pavail accordingly. |
| 4050 | */ |
| 4051 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pavail, |
| 4052 | unsigned long *pdirty, unsigned long *pwriteback) |
| 4053 | { |
| 4054 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); |
| 4055 | struct mem_cgroup *parent; |
| 4056 | unsigned long head_room = PAGE_COUNTER_MAX; |
| 4057 | unsigned long file_pages; |
| 4058 | |
| 4059 | *pdirty = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_DIRTY); |
| 4060 | |
| 4061 | /* this should eventually include NR_UNSTABLE_NFS */ |
| 4062 | *pwriteback = mem_cgroup_read_stat(memcg, MEM_CGROUP_STAT_WRITEBACK); |
| 4063 | |
| 4064 | file_pages = mem_cgroup_nr_lru_pages(memcg, (1 << LRU_INACTIVE_FILE) | |
| 4065 | (1 << LRU_ACTIVE_FILE)); |
| 4066 | while ((parent = parent_mem_cgroup(memcg))) { |
| 4067 | unsigned long ceiling = min(memcg->memory.limit, memcg->high); |
| 4068 | unsigned long used = page_counter_read(&memcg->memory); |
| 4069 | |
| 4070 | head_room = min(head_room, ceiling - min(ceiling, used)); |
| 4071 | memcg = parent; |
| 4072 | } |
| 4073 | |
| 4074 | *pavail = file_pages + head_room; |
| 4075 | } |
| 4076 | |
Tejun Heo | 841710a | 2015-05-22 18:23:33 -0400 | [diff] [blame] | 4077 | #else /* CONFIG_CGROUP_WRITEBACK */ |
| 4078 | |
| 4079 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
| 4080 | { |
| 4081 | return 0; |
| 4082 | } |
| 4083 | |
| 4084 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) |
| 4085 | { |
| 4086 | } |
| 4087 | |
Tejun Heo | 2529bb3 | 2015-05-22 18:23:34 -0400 | [diff] [blame] | 4088 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
| 4089 | { |
| 4090 | } |
| 4091 | |
Tejun Heo | 52ebea7 | 2015-05-22 17:13:37 -0400 | [diff] [blame] | 4092 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
| 4093 | |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4094 | /* |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 4095 | * DO NOT USE IN NEW FILES. |
| 4096 | * |
| 4097 | * "cgroup.event_control" implementation. |
| 4098 | * |
| 4099 | * This is way over-engineered. It tries to support fully configurable |
| 4100 | * events for each user. Such level of flexibility is completely |
| 4101 | * unnecessary especially in the light of the planned unified hierarchy. |
| 4102 | * |
| 4103 | * Please deprecate this and replace with something simpler if at all |
| 4104 | * possible. |
| 4105 | */ |
| 4106 | |
| 4107 | /* |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4108 | * Unregister event and free resources. |
| 4109 | * |
| 4110 | * Gets called from workqueue. |
| 4111 | */ |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 4112 | static void memcg_event_remove(struct work_struct *work) |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4113 | { |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 4114 | struct mem_cgroup_event *event = |
| 4115 | container_of(work, struct mem_cgroup_event, remove); |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4116 | struct mem_cgroup *memcg = event->memcg; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4117 | |
| 4118 | remove_wait_queue(event->wqh, &event->wait); |
| 4119 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4120 | event->unregister_event(memcg, event->eventfd); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4121 | |
| 4122 | /* Notify userspace the event is going away. */ |
| 4123 | eventfd_signal(event->eventfd, 1); |
| 4124 | |
| 4125 | eventfd_ctx_put(event->eventfd); |
| 4126 | kfree(event); |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4127 | css_put(&memcg->css); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4128 | } |
| 4129 | |
| 4130 | /* |
| 4131 | * Gets called on POLLHUP on eventfd when user closes it. |
| 4132 | * |
| 4133 | * Called with wqh->lock held and interrupts disabled. |
| 4134 | */ |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 4135 | static int memcg_event_wake(wait_queue_t *wait, unsigned mode, |
| 4136 | int sync, void *key) |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4137 | { |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 4138 | struct mem_cgroup_event *event = |
| 4139 | container_of(wait, struct mem_cgroup_event, wait); |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4140 | struct mem_cgroup *memcg = event->memcg; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4141 | unsigned long flags = (unsigned long)key; |
| 4142 | |
| 4143 | if (flags & POLLHUP) { |
| 4144 | /* |
| 4145 | * If the event has been detached at cgroup removal, we |
| 4146 | * can simply return knowing the other side will cleanup |
| 4147 | * for us. |
| 4148 | * |
| 4149 | * We can't race against event freeing since the other |
| 4150 | * side will require wqh->lock via remove_wait_queue(), |
| 4151 | * which we hold. |
| 4152 | */ |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4153 | spin_lock(&memcg->event_list_lock); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4154 | if (!list_empty(&event->list)) { |
| 4155 | list_del_init(&event->list); |
| 4156 | /* |
| 4157 | * We are in atomic context, but cgroup_event_remove() |
| 4158 | * may sleep, so we have to call it in workqueue. |
| 4159 | */ |
| 4160 | schedule_work(&event->remove); |
| 4161 | } |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4162 | spin_unlock(&memcg->event_list_lock); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4163 | } |
| 4164 | |
| 4165 | return 0; |
| 4166 | } |
| 4167 | |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 4168 | static void memcg_event_ptable_queue_proc(struct file *file, |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4169 | wait_queue_head_t *wqh, poll_table *pt) |
| 4170 | { |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 4171 | struct mem_cgroup_event *event = |
| 4172 | container_of(pt, struct mem_cgroup_event, pt); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4173 | |
| 4174 | event->wqh = wqh; |
| 4175 | add_wait_queue(wqh, &event->wait); |
| 4176 | } |
| 4177 | |
| 4178 | /* |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 4179 | * DO NOT USE IN NEW FILES. |
| 4180 | * |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4181 | * Parse input and register new cgroup event handler. |
| 4182 | * |
| 4183 | * Input must be in format '<event_fd> <control_fd> <args>'. |
| 4184 | * Interpretation of args is defined by control file implementation. |
| 4185 | */ |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4186 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
| 4187 | char *buf, size_t nbytes, loff_t off) |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4188 | { |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4189 | struct cgroup_subsys_state *css = of_css(of); |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4190 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 4191 | struct mem_cgroup_event *event; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4192 | struct cgroup_subsys_state *cfile_css; |
| 4193 | unsigned int efd, cfd; |
| 4194 | struct fd efile; |
| 4195 | struct fd cfile; |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4196 | const char *name; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4197 | char *endp; |
| 4198 | int ret; |
| 4199 | |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4200 | buf = strstrip(buf); |
| 4201 | |
| 4202 | efd = simple_strtoul(buf, &endp, 10); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4203 | if (*endp != ' ') |
| 4204 | return -EINVAL; |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4205 | buf = endp + 1; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4206 | |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4207 | cfd = simple_strtoul(buf, &endp, 10); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4208 | if ((*endp != ' ') && (*endp != '\0')) |
| 4209 | return -EINVAL; |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4210 | buf = endp + 1; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4211 | |
| 4212 | event = kzalloc(sizeof(*event), GFP_KERNEL); |
| 4213 | if (!event) |
| 4214 | return -ENOMEM; |
| 4215 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4216 | event->memcg = memcg; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4217 | INIT_LIST_HEAD(&event->list); |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 4218 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
| 4219 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); |
| 4220 | INIT_WORK(&event->remove, memcg_event_remove); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4221 | |
| 4222 | efile = fdget(efd); |
| 4223 | if (!efile.file) { |
| 4224 | ret = -EBADF; |
| 4225 | goto out_kfree; |
| 4226 | } |
| 4227 | |
| 4228 | event->eventfd = eventfd_ctx_fileget(efile.file); |
| 4229 | if (IS_ERR(event->eventfd)) { |
| 4230 | ret = PTR_ERR(event->eventfd); |
| 4231 | goto out_put_efile; |
| 4232 | } |
| 4233 | |
| 4234 | cfile = fdget(cfd); |
| 4235 | if (!cfile.file) { |
| 4236 | ret = -EBADF; |
| 4237 | goto out_put_eventfd; |
| 4238 | } |
| 4239 | |
| 4240 | /* the process need read permission on control file */ |
| 4241 | /* AV: shouldn't we check that it's been opened for read instead? */ |
| 4242 | ret = inode_permission(file_inode(cfile.file), MAY_READ); |
| 4243 | if (ret < 0) |
| 4244 | goto out_put_cfile; |
| 4245 | |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4246 | /* |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4247 | * Determine the event callbacks and set them in @event. This used |
| 4248 | * to be done via struct cftype but cgroup core no longer knows |
| 4249 | * about these events. The following is crude but the whole thing |
| 4250 | * is for compatibility anyway. |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 4251 | * |
| 4252 | * DO NOT ADD NEW FILES. |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4253 | */ |
Al Viro | b583043 | 2014-10-31 01:22:04 -0400 | [diff] [blame] | 4254 | name = cfile.file->f_path.dentry->d_name.name; |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4255 | |
| 4256 | if (!strcmp(name, "memory.usage_in_bytes")) { |
| 4257 | event->register_event = mem_cgroup_usage_register_event; |
| 4258 | event->unregister_event = mem_cgroup_usage_unregister_event; |
| 4259 | } else if (!strcmp(name, "memory.oom_control")) { |
| 4260 | event->register_event = mem_cgroup_oom_register_event; |
| 4261 | event->unregister_event = mem_cgroup_oom_unregister_event; |
| 4262 | } else if (!strcmp(name, "memory.pressure_level")) { |
| 4263 | event->register_event = vmpressure_register_event; |
| 4264 | event->unregister_event = vmpressure_unregister_event; |
| 4265 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4266 | event->register_event = memsw_cgroup_usage_register_event; |
| 4267 | event->unregister_event = memsw_cgroup_usage_unregister_event; |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4268 | } else { |
| 4269 | ret = -EINVAL; |
| 4270 | goto out_put_cfile; |
| 4271 | } |
| 4272 | |
| 4273 | /* |
Tejun Heo | b5557c4 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4274 | * Verify @cfile should belong to @css. Also, remaining events are |
| 4275 | * automatically removed on cgroup destruction but the removal is |
| 4276 | * asynchronous, so take an extra ref on @css. |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4277 | */ |
Al Viro | b583043 | 2014-10-31 01:22:04 -0400 | [diff] [blame] | 4278 | cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, |
Tejun Heo | ec903c0 | 2014-05-13 12:11:01 -0400 | [diff] [blame] | 4279 | &memory_cgrp_subsys); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4280 | ret = -EINVAL; |
Tejun Heo | 5a17f54 | 2014-02-11 11:52:47 -0500 | [diff] [blame] | 4281 | if (IS_ERR(cfile_css)) |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4282 | goto out_put_cfile; |
Tejun Heo | 5a17f54 | 2014-02-11 11:52:47 -0500 | [diff] [blame] | 4283 | if (cfile_css != css) { |
| 4284 | css_put(cfile_css); |
| 4285 | goto out_put_cfile; |
| 4286 | } |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4287 | |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4288 | ret = event->register_event(memcg, event->eventfd, buf); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4289 | if (ret) |
| 4290 | goto out_put_css; |
| 4291 | |
| 4292 | efile.file->f_op->poll(efile.file, &event->pt); |
| 4293 | |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4294 | spin_lock(&memcg->event_list_lock); |
| 4295 | list_add(&event->list, &memcg->event_list); |
| 4296 | spin_unlock(&memcg->event_list_lock); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4297 | |
| 4298 | fdput(cfile); |
| 4299 | fdput(efile); |
| 4300 | |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4301 | return nbytes; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4302 | |
| 4303 | out_put_css: |
Tejun Heo | b5557c4 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4304 | css_put(css); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4305 | out_put_cfile: |
| 4306 | fdput(cfile); |
| 4307 | out_put_eventfd: |
| 4308 | eventfd_ctx_put(event->eventfd); |
| 4309 | out_put_efile: |
| 4310 | fdput(efile); |
| 4311 | out_kfree: |
| 4312 | kfree(event); |
| 4313 | |
| 4314 | return ret; |
| 4315 | } |
| 4316 | |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 4317 | static struct cftype mem_cgroup_legacy_files[] = { |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4318 | { |
Balbir Singh | 0eea103 | 2008-02-07 00:13:57 -0800 | [diff] [blame] | 4319 | .name = "usage_in_bytes", |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4320 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 4321 | .read_u64 = mem_cgroup_read_u64, |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4322 | }, |
| 4323 | { |
Pavel Emelyanov | c84872e | 2008-04-29 01:00:17 -0700 | [diff] [blame] | 4324 | .name = "max_usage_in_bytes", |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4325 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4326 | .write = mem_cgroup_reset, |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 4327 | .read_u64 = mem_cgroup_read_u64, |
Pavel Emelyanov | c84872e | 2008-04-29 01:00:17 -0700 | [diff] [blame] | 4328 | }, |
| 4329 | { |
Balbir Singh | 0eea103 | 2008-02-07 00:13:57 -0800 | [diff] [blame] | 4330 | .name = "limit_in_bytes", |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4331 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4332 | .write = mem_cgroup_write, |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 4333 | .read_u64 = mem_cgroup_read_u64, |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4334 | }, |
| 4335 | { |
Balbir Singh | 296c81d | 2009-09-23 15:56:36 -0700 | [diff] [blame] | 4336 | .name = "soft_limit_in_bytes", |
| 4337 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4338 | .write = mem_cgroup_write, |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 4339 | .read_u64 = mem_cgroup_read_u64, |
Balbir Singh | 296c81d | 2009-09-23 15:56:36 -0700 | [diff] [blame] | 4340 | }, |
| 4341 | { |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4342 | .name = "failcnt", |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4343 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4344 | .write = mem_cgroup_reset, |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 4345 | .read_u64 = mem_cgroup_read_u64, |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4346 | }, |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 4347 | { |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 4348 | .name = "stat", |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 4349 | .seq_show = memcg_stat_show, |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 4350 | }, |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 4351 | { |
| 4352 | .name = "force_empty", |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4353 | .write = mem_cgroup_force_empty_write, |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 4354 | }, |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 4355 | { |
| 4356 | .name = "use_hierarchy", |
| 4357 | .write_u64 = mem_cgroup_hierarchy_write, |
| 4358 | .read_u64 = mem_cgroup_hierarchy_read, |
| 4359 | }, |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 4360 | { |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 4361 | .name = "cgroup.event_control", /* XXX: for compat */ |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4362 | .write = memcg_write_event_control, |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4363 | .flags = CFTYPE_NO_PREFIX, |
| 4364 | .mode = S_IWUGO, |
| 4365 | }, |
| 4366 | { |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 4367 | .name = "swappiness", |
| 4368 | .read_u64 = mem_cgroup_swappiness_read, |
| 4369 | .write_u64 = mem_cgroup_swappiness_write, |
| 4370 | }, |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4371 | { |
| 4372 | .name = "move_charge_at_immigrate", |
| 4373 | .read_u64 = mem_cgroup_move_charge_read, |
| 4374 | .write_u64 = mem_cgroup_move_charge_write, |
| 4375 | }, |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4376 | { |
| 4377 | .name = "oom_control", |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 4378 | .seq_show = mem_cgroup_oom_control_read, |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 4379 | .write_u64 = mem_cgroup_oom_control_write, |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4380 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
| 4381 | }, |
Anton Vorontsov | 70ddf63 | 2013-04-29 15:08:31 -0700 | [diff] [blame] | 4382 | { |
| 4383 | .name = "pressure_level", |
Anton Vorontsov | 70ddf63 | 2013-04-29 15:08:31 -0700 | [diff] [blame] | 4384 | }, |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 4385 | #ifdef CONFIG_NUMA |
| 4386 | { |
| 4387 | .name = "numa_stat", |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 4388 | .seq_show = memcg_numa_stat_show, |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 4389 | }, |
| 4390 | #endif |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 4391 | #ifdef CONFIG_MEMCG_KMEM |
| 4392 | { |
| 4393 | .name = "kmem.limit_in_bytes", |
| 4394 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4395 | .write = mem_cgroup_write, |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 4396 | .read_u64 = mem_cgroup_read_u64, |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 4397 | }, |
| 4398 | { |
| 4399 | .name = "kmem.usage_in_bytes", |
| 4400 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 4401 | .read_u64 = mem_cgroup_read_u64, |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 4402 | }, |
| 4403 | { |
| 4404 | .name = "kmem.failcnt", |
| 4405 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4406 | .write = mem_cgroup_reset, |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 4407 | .read_u64 = mem_cgroup_read_u64, |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 4408 | }, |
| 4409 | { |
| 4410 | .name = "kmem.max_usage_in_bytes", |
| 4411 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4412 | .write = mem_cgroup_reset, |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 4413 | .read_u64 = mem_cgroup_read_u64, |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 4414 | }, |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 4415 | #ifdef CONFIG_SLABINFO |
| 4416 | { |
| 4417 | .name = "kmem.slabinfo", |
Vladimir Davydov | b047501 | 2014-12-10 15:44:19 -0800 | [diff] [blame] | 4418 | .seq_start = slab_start, |
| 4419 | .seq_next = slab_next, |
| 4420 | .seq_stop = slab_stop, |
| 4421 | .seq_show = memcg_slab_show, |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 4422 | }, |
| 4423 | #endif |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 4424 | #endif |
Tejun Heo | 6bc1034 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 4425 | { }, /* terminate */ |
Tejun Heo | af36f90 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 4426 | }; |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4427 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4428 | static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 4429 | { |
| 4430 | struct mem_cgroup_per_node *pn; |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 4431 | struct mem_cgroup_per_zone *mz; |
KAMEZAWA Hiroyuki | 41e3355 | 2008-04-08 17:41:54 -0700 | [diff] [blame] | 4432 | int zone, tmp = node; |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 4433 | /* |
| 4434 | * This routine is called against possible nodes. |
| 4435 | * But it's BUG to call kmalloc() against offline node. |
| 4436 | * |
| 4437 | * TODO: this routine can waste much memory for nodes which will |
| 4438 | * never be onlined. It's better to use memory hotplug callback |
| 4439 | * function. |
| 4440 | */ |
KAMEZAWA Hiroyuki | 41e3355 | 2008-04-08 17:41:54 -0700 | [diff] [blame] | 4441 | if (!node_state(node, N_NORMAL_MEMORY)) |
| 4442 | tmp = -1; |
Jesper Juhl | 17295c8 | 2011-01-13 15:47:42 -0800 | [diff] [blame] | 4443 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 4444 | if (!pn) |
| 4445 | return 1; |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 4446 | |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 4447 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
| 4448 | mz = &pn->zoneinfo[zone]; |
Hugh Dickins | bea8c15 | 2012-11-16 14:14:54 -0800 | [diff] [blame] | 4449 | lruvec_init(&mz->lruvec); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 4450 | mz->usage_in_excess = 0; |
| 4451 | mz->on_tree = false; |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 4452 | mz->memcg = memcg; |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 4453 | } |
Johannes Weiner | 54f72fe | 2013-07-08 15:59:49 -0700 | [diff] [blame] | 4454 | memcg->nodeinfo[node] = pn; |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 4455 | return 0; |
| 4456 | } |
| 4457 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4458 | static void free_mem_cgroup_per_zone_info(struct mem_cgroup *memcg, int node) |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 4459 | { |
Johannes Weiner | 54f72fe | 2013-07-08 15:59:49 -0700 | [diff] [blame] | 4460 | kfree(memcg->nodeinfo[node]); |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 4461 | } |
| 4462 | |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 4463 | static struct mem_cgroup *mem_cgroup_alloc(void) |
| 4464 | { |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 4465 | struct mem_cgroup *memcg; |
Vladimir Davydov | 8ff69e2 | 2014-01-23 15:52:52 -0800 | [diff] [blame] | 4466 | size_t size; |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 4467 | |
Vladimir Davydov | 8ff69e2 | 2014-01-23 15:52:52 -0800 | [diff] [blame] | 4468 | size = sizeof(struct mem_cgroup); |
| 4469 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 4470 | |
Vladimir Davydov | 8ff69e2 | 2014-01-23 15:52:52 -0800 | [diff] [blame] | 4471 | memcg = kzalloc(size, GFP_KERNEL); |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 4472 | if (!memcg) |
Dan Carpenter | e7bbcdf | 2010-03-23 13:35:12 -0700 | [diff] [blame] | 4473 | return NULL; |
| 4474 | |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 4475 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
| 4476 | if (!memcg->stat) |
Dan Carpenter | d2e61b8 | 2010-11-11 14:05:12 -0800 | [diff] [blame] | 4477 | goto out_free; |
Tejun Heo | 841710a | 2015-05-22 18:23:33 -0400 | [diff] [blame] | 4478 | |
| 4479 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) |
| 4480 | goto out_free_stat; |
| 4481 | |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 4482 | spin_lock_init(&memcg->pcp_counter_lock); |
| 4483 | return memcg; |
Dan Carpenter | d2e61b8 | 2010-11-11 14:05:12 -0800 | [diff] [blame] | 4484 | |
Tejun Heo | 841710a | 2015-05-22 18:23:33 -0400 | [diff] [blame] | 4485 | out_free_stat: |
| 4486 | free_percpu(memcg->stat); |
Dan Carpenter | d2e61b8 | 2010-11-11 14:05:12 -0800 | [diff] [blame] | 4487 | out_free: |
Vladimir Davydov | 8ff69e2 | 2014-01-23 15:52:52 -0800 | [diff] [blame] | 4488 | kfree(memcg); |
Dan Carpenter | d2e61b8 | 2010-11-11 14:05:12 -0800 | [diff] [blame] | 4489 | return NULL; |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 4490 | } |
| 4491 | |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4492 | /* |
Glauber Costa | c8b2a36 | 2012-12-18 14:22:13 -0800 | [diff] [blame] | 4493 | * At destroying mem_cgroup, references from swap_cgroup can remain. |
| 4494 | * (scanning all at force_empty is too costly...) |
| 4495 | * |
| 4496 | * Instead of clearing all references at force_empty, we remember |
| 4497 | * the number of reference from swap_cgroup and free mem_cgroup when |
| 4498 | * it goes down to 0. |
| 4499 | * |
| 4500 | * Removal of cgroup itself succeeds regardless of refs from swap. |
Hugh Dickins | 59927fb | 2012-03-15 15:17:07 -0700 | [diff] [blame] | 4501 | */ |
Glauber Costa | c8b2a36 | 2012-12-18 14:22:13 -0800 | [diff] [blame] | 4502 | |
| 4503 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
Hugh Dickins | 59927fb | 2012-03-15 15:17:07 -0700 | [diff] [blame] | 4504 | { |
Glauber Costa | c8b2a36 | 2012-12-18 14:22:13 -0800 | [diff] [blame] | 4505 | int node; |
Hugh Dickins | 59927fb | 2012-03-15 15:17:07 -0700 | [diff] [blame] | 4506 | |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 4507 | mem_cgroup_remove_from_trees(memcg); |
Glauber Costa | c8b2a36 | 2012-12-18 14:22:13 -0800 | [diff] [blame] | 4508 | |
| 4509 | for_each_node(node) |
| 4510 | free_mem_cgroup_per_zone_info(memcg, node); |
| 4511 | |
| 4512 | free_percpu(memcg->stat); |
Tejun Heo | 841710a | 2015-05-22 18:23:33 -0400 | [diff] [blame] | 4513 | memcg_wb_domain_exit(memcg); |
Vladimir Davydov | 8ff69e2 | 2014-01-23 15:52:52 -0800 | [diff] [blame] | 4514 | kfree(memcg); |
Hugh Dickins | 59927fb | 2012-03-15 15:17:07 -0700 | [diff] [blame] | 4515 | } |
Glauber Costa | 3afe36b | 2012-05-29 15:07:10 -0700 | [diff] [blame] | 4516 | |
Daisuke Nishimura | 7bcc1bb | 2009-01-29 14:25:11 -0800 | [diff] [blame] | 4517 | /* |
| 4518 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. |
| 4519 | */ |
Glauber Costa | e1aab16 | 2011-12-11 21:47:03 +0000 | [diff] [blame] | 4520 | struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) |
Daisuke Nishimura | 7bcc1bb | 2009-01-29 14:25:11 -0800 | [diff] [blame] | 4521 | { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 4522 | if (!memcg->memory.parent) |
Daisuke Nishimura | 7bcc1bb | 2009-01-29 14:25:11 -0800 | [diff] [blame] | 4523 | return NULL; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 4524 | return mem_cgroup_from_counter(memcg->memory.parent, memory); |
Daisuke Nishimura | 7bcc1bb | 2009-01-29 14:25:11 -0800 | [diff] [blame] | 4525 | } |
Glauber Costa | e1aab16 | 2011-12-11 21:47:03 +0000 | [diff] [blame] | 4526 | EXPORT_SYMBOL(parent_mem_cgroup); |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 4527 | |
Li Zefan | 0eb253e | 2009-01-15 13:51:25 -0800 | [diff] [blame] | 4528 | static struct cgroup_subsys_state * __ref |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 4529 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4530 | { |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 4531 | struct mem_cgroup *memcg; |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 4532 | long error = -ENOMEM; |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 4533 | int node; |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4534 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4535 | memcg = mem_cgroup_alloc(); |
| 4536 | if (!memcg) |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 4537 | return ERR_PTR(error); |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 4538 | |
Bob Liu | 3ed28fa | 2012-01-12 17:19:04 -0800 | [diff] [blame] | 4539 | for_each_node(node) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4540 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 4541 | goto free_out; |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 4542 | |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 4543 | /* root ? */ |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 4544 | if (parent_css == NULL) { |
Hillf Danton | a41c58a | 2011-12-19 17:11:57 -0800 | [diff] [blame] | 4545 | root_mem_cgroup = memcg; |
Tejun Heo | 5616163 | 2015-05-22 17:13:20 -0400 | [diff] [blame] | 4546 | mem_cgroup_root_css = &memcg->css; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 4547 | page_counter_init(&memcg->memory, NULL); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 4548 | memcg->high = PAGE_COUNTER_MAX; |
Johannes Weiner | 24d404d | 2015-01-08 14:32:35 -0800 | [diff] [blame] | 4549 | memcg->soft_limit = PAGE_COUNTER_MAX; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 4550 | page_counter_init(&memcg->memsw, NULL); |
| 4551 | page_counter_init(&memcg->kmem, NULL); |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 4552 | } |
Balbir Singh | 28dbc4b | 2009-01-07 18:08:05 -0800 | [diff] [blame] | 4553 | |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 4554 | memcg->last_scanned_node = MAX_NUMNODES; |
| 4555 | INIT_LIST_HEAD(&memcg->oom_notify); |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 4556 | memcg->move_charge_at_immigrate = 0; |
| 4557 | mutex_init(&memcg->thresholds_lock); |
| 4558 | spin_lock_init(&memcg->move_lock); |
Anton Vorontsov | 70ddf63 | 2013-04-29 15:08:31 -0700 | [diff] [blame] | 4559 | vmpressure_init(&memcg->vmpressure); |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4560 | INIT_LIST_HEAD(&memcg->event_list); |
| 4561 | spin_lock_init(&memcg->event_list_lock); |
Vladimir Davydov | 900a38f | 2014-12-12 16:55:10 -0800 | [diff] [blame] | 4562 | #ifdef CONFIG_MEMCG_KMEM |
| 4563 | memcg->kmemcg_id = -1; |
Vladimir Davydov | 900a38f | 2014-12-12 16:55:10 -0800 | [diff] [blame] | 4564 | #endif |
Tejun Heo | 52ebea7 | 2015-05-22 17:13:37 -0400 | [diff] [blame] | 4565 | #ifdef CONFIG_CGROUP_WRITEBACK |
| 4566 | INIT_LIST_HEAD(&memcg->cgwb_list); |
| 4567 | #endif |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 4568 | return &memcg->css; |
| 4569 | |
| 4570 | free_out: |
| 4571 | __mem_cgroup_free(memcg); |
| 4572 | return ERR_PTR(error); |
| 4573 | } |
| 4574 | |
| 4575 | static int |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 4576 | mem_cgroup_css_online(struct cgroup_subsys_state *css) |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 4577 | { |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 4578 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Tejun Heo | 5c9d535 | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 4579 | struct mem_cgroup *parent = mem_cgroup_from_css(css->parent); |
Johannes Weiner | 2f7dd7a | 2014-10-02 16:16:57 -0700 | [diff] [blame] | 4580 | int ret; |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 4581 | |
Tejun Heo | 15a4c83 | 2014-05-04 15:09:14 -0400 | [diff] [blame] | 4582 | if (css->id > MEM_CGROUP_ID_MAX) |
Li Zefan | 4219b2d | 2013-09-23 16:56:29 +0800 | [diff] [blame] | 4583 | return -ENOSPC; |
| 4584 | |
Tejun Heo | 6387698 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 4585 | if (!parent) |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 4586 | return 0; |
| 4587 | |
Glauber Costa | 0999821 | 2013-02-22 16:34:55 -0800 | [diff] [blame] | 4588 | mutex_lock(&memcg_create_mutex); |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 4589 | |
| 4590 | memcg->use_hierarchy = parent->use_hierarchy; |
| 4591 | memcg->oom_kill_disable = parent->oom_kill_disable; |
| 4592 | memcg->swappiness = mem_cgroup_swappiness(parent); |
| 4593 | |
| 4594 | if (parent->use_hierarchy) { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 4595 | page_counter_init(&memcg->memory, &parent->memory); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 4596 | memcg->high = PAGE_COUNTER_MAX; |
Johannes Weiner | 24d404d | 2015-01-08 14:32:35 -0800 | [diff] [blame] | 4597 | memcg->soft_limit = PAGE_COUNTER_MAX; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 4598 | page_counter_init(&memcg->memsw, &parent->memsw); |
| 4599 | page_counter_init(&memcg->kmem, &parent->kmem); |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 4600 | |
Daisuke Nishimura | 7bcc1bb | 2009-01-29 14:25:11 -0800 | [diff] [blame] | 4601 | /* |
Li Zefan | 8d76a97 | 2013-07-08 16:00:36 -0700 | [diff] [blame] | 4602 | * No need to take a reference to the parent because cgroup |
| 4603 | * core guarantees its existence. |
Daisuke Nishimura | 7bcc1bb | 2009-01-29 14:25:11 -0800 | [diff] [blame] | 4604 | */ |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 4605 | } else { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 4606 | page_counter_init(&memcg->memory, NULL); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 4607 | memcg->high = PAGE_COUNTER_MAX; |
Johannes Weiner | 24d404d | 2015-01-08 14:32:35 -0800 | [diff] [blame] | 4608 | memcg->soft_limit = PAGE_COUNTER_MAX; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 4609 | page_counter_init(&memcg->memsw, NULL); |
| 4610 | page_counter_init(&memcg->kmem, NULL); |
Tejun Heo | 8c7f6ed | 2012-09-13 12:20:58 -0700 | [diff] [blame] | 4611 | /* |
| 4612 | * Deeper hierachy with use_hierarchy == false doesn't make |
| 4613 | * much sense so let cgroup subsystem know about this |
| 4614 | * unfortunate state in our controller. |
| 4615 | */ |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 4616 | if (parent != root_mem_cgroup) |
Tejun Heo | 073219e | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 4617 | memory_cgrp_subsys.broken_hierarchy = true; |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 4618 | } |
Glauber Costa | 0999821 | 2013-02-22 16:34:55 -0800 | [diff] [blame] | 4619 | mutex_unlock(&memcg_create_mutex); |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 4620 | |
Johannes Weiner | 2f7dd7a | 2014-10-02 16:16:57 -0700 | [diff] [blame] | 4621 | ret = memcg_init_kmem(memcg, &memory_cgrp_subsys); |
| 4622 | if (ret) |
| 4623 | return ret; |
| 4624 | |
| 4625 | /* |
| 4626 | * Make sure the memcg is initialized: mem_cgroup_iter() |
| 4627 | * orders reading memcg->initialized against its callers |
| 4628 | * reading the memcg members. |
| 4629 | */ |
| 4630 | smp_store_release(&memcg->initialized, 1); |
| 4631 | |
| 4632 | return 0; |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4633 | } |
| 4634 | |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 4635 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
KAMEZAWA Hiroyuki | df878fb | 2008-02-07 00:14:28 -0800 | [diff] [blame] | 4636 | { |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 4637 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 4638 | struct mem_cgroup_event *event, *tmp; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4639 | |
| 4640 | /* |
| 4641 | * Unregister events and notify userspace. |
| 4642 | * Notify userspace about cgroup removing only after rmdir of cgroup |
| 4643 | * directory to avoid race between userspace and kernelspace. |
| 4644 | */ |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4645 | spin_lock(&memcg->event_list_lock); |
| 4646 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4647 | list_del_init(&event->list); |
| 4648 | schedule_work(&event->remove); |
| 4649 | } |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4650 | spin_unlock(&memcg->event_list_lock); |
KAMEZAWA Hiroyuki | ec64f51 | 2009-04-02 16:57:26 -0700 | [diff] [blame] | 4651 | |
Michal Hocko | 33cb876 | 2013-07-31 13:53:51 -0700 | [diff] [blame] | 4652 | vmpressure_cleanup(&memcg->vmpressure); |
Vladimir Davydov | 2a4db7e | 2015-02-12 14:59:32 -0800 | [diff] [blame] | 4653 | |
| 4654 | memcg_deactivate_kmem(memcg); |
Tejun Heo | 52ebea7 | 2015-05-22 17:13:37 -0400 | [diff] [blame] | 4655 | |
| 4656 | wb_memcg_offline(memcg); |
KAMEZAWA Hiroyuki | df878fb | 2008-02-07 00:14:28 -0800 | [diff] [blame] | 4657 | } |
| 4658 | |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 4659 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4660 | { |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 4661 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Daisuke Nishimura | c268e99 | 2009-01-15 13:51:13 -0800 | [diff] [blame] | 4662 | |
Li Zefan | 10d5ebf | 2013-07-08 16:00:33 -0700 | [diff] [blame] | 4663 | memcg_destroy_kmem(memcg); |
Li Zefan | 465939a | 2013-07-08 16:00:38 -0700 | [diff] [blame] | 4664 | __mem_cgroup_free(memcg); |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4665 | } |
| 4666 | |
Tejun Heo | 1ced953 | 2014-07-08 18:02:57 -0400 | [diff] [blame] | 4667 | /** |
| 4668 | * mem_cgroup_css_reset - reset the states of a mem_cgroup |
| 4669 | * @css: the target css |
| 4670 | * |
| 4671 | * Reset the states of the mem_cgroup associated with @css. This is |
| 4672 | * invoked when the userland requests disabling on the default hierarchy |
| 4673 | * but the memcg is pinned through dependency. The memcg should stop |
| 4674 | * applying policies and should revert to the vanilla state as it may be |
| 4675 | * made visible again. |
| 4676 | * |
| 4677 | * The current implementation only resets the essential configurations. |
| 4678 | * This needs to be expanded to cover all the visible parts. |
| 4679 | */ |
| 4680 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) |
| 4681 | { |
| 4682 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
| 4683 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 4684 | mem_cgroup_resize_limit(memcg, PAGE_COUNTER_MAX); |
| 4685 | mem_cgroup_resize_memsw_limit(memcg, PAGE_COUNTER_MAX); |
| 4686 | memcg_update_kmem_limit(memcg, PAGE_COUNTER_MAX); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 4687 | memcg->low = 0; |
| 4688 | memcg->high = PAGE_COUNTER_MAX; |
Johannes Weiner | 24d404d | 2015-01-08 14:32:35 -0800 | [diff] [blame] | 4689 | memcg->soft_limit = PAGE_COUNTER_MAX; |
Tejun Heo | 2529bb3 | 2015-05-22 18:23:34 -0400 | [diff] [blame] | 4690 | memcg_wb_domain_size_changed(memcg); |
Tejun Heo | 1ced953 | 2014-07-08 18:02:57 -0400 | [diff] [blame] | 4691 | } |
| 4692 | |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4693 | #ifdef CONFIG_MMU |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4694 | /* Handlers for move charge at task migration. */ |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4695 | static int mem_cgroup_do_precharge(unsigned long count) |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4696 | { |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 4697 | int ret; |
Johannes Weiner | 9476db9 | 2014-08-06 16:05:55 -0700 | [diff] [blame] | 4698 | |
| 4699 | /* Try a single bulk charge without reclaim first */ |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 4700 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_WAIT, count); |
Johannes Weiner | 9476db9 | 2014-08-06 16:05:55 -0700 | [diff] [blame] | 4701 | if (!ret) { |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4702 | mc.precharge += count; |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4703 | return ret; |
| 4704 | } |
Johannes Weiner | 692e7c4 | 2014-08-06 16:05:57 -0700 | [diff] [blame] | 4705 | if (ret == -EINTR) { |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 4706 | cancel_charge(root_mem_cgroup, count); |
Johannes Weiner | 692e7c4 | 2014-08-06 16:05:57 -0700 | [diff] [blame] | 4707 | return ret; |
| 4708 | } |
Johannes Weiner | 9476db9 | 2014-08-06 16:05:55 -0700 | [diff] [blame] | 4709 | |
| 4710 | /* Try charges one by one with reclaim */ |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4711 | while (count--) { |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 4712 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_NORETRY, 1); |
Johannes Weiner | 9476db9 | 2014-08-06 16:05:55 -0700 | [diff] [blame] | 4713 | /* |
| 4714 | * In case of failure, any residual charges against |
| 4715 | * mc.to will be dropped by mem_cgroup_clear_mc() |
Johannes Weiner | 692e7c4 | 2014-08-06 16:05:57 -0700 | [diff] [blame] | 4716 | * later on. However, cancel any charges that are |
| 4717 | * bypassed to root right away or they'll be lost. |
Johannes Weiner | 9476db9 | 2014-08-06 16:05:55 -0700 | [diff] [blame] | 4718 | */ |
Johannes Weiner | 692e7c4 | 2014-08-06 16:05:57 -0700 | [diff] [blame] | 4719 | if (ret == -EINTR) |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 4720 | cancel_charge(root_mem_cgroup, 1); |
KAMEZAWA Hiroyuki | 38c5d72 | 2012-01-12 17:19:01 -0800 | [diff] [blame] | 4721 | if (ret) |
KAMEZAWA Hiroyuki | 38c5d72 | 2012-01-12 17:19:01 -0800 | [diff] [blame] | 4722 | return ret; |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4723 | mc.precharge++; |
Johannes Weiner | 9476db9 | 2014-08-06 16:05:55 -0700 | [diff] [blame] | 4724 | cond_resched(); |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4725 | } |
Johannes Weiner | 9476db9 | 2014-08-06 16:05:55 -0700 | [diff] [blame] | 4726 | return 0; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4727 | } |
| 4728 | |
| 4729 | /** |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 4730 | * get_mctgt_type - get target type of moving charge |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4731 | * @vma: the vma the pte to be checked belongs |
| 4732 | * @addr: the address corresponding to the pte to be checked |
| 4733 | * @ptent: the pte to be checked |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4734 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4735 | * |
| 4736 | * Returns |
| 4737 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. |
| 4738 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for |
| 4739 | * move charge. if @target is not NULL, the page is stored in target->page |
| 4740 | * with extra refcnt got(Callers should handle it). |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4741 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
| 4742 | * target for charge migration. if @target is not NULL, the entry is stored |
| 4743 | * in target->ent. |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4744 | * |
| 4745 | * Called with pte lock held. |
| 4746 | */ |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4747 | union mc_target { |
| 4748 | struct page *page; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4749 | swp_entry_t ent; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4750 | }; |
| 4751 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4752 | enum mc_target_type { |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 4753 | MC_TARGET_NONE = 0, |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4754 | MC_TARGET_PAGE, |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4755 | MC_TARGET_SWAP, |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4756 | }; |
| 4757 | |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4758 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
| 4759 | unsigned long addr, pte_t ptent) |
| 4760 | { |
| 4761 | struct page *page = vm_normal_page(vma, addr, ptent); |
| 4762 | |
| 4763 | if (!page || !page_mapped(page)) |
| 4764 | return NULL; |
| 4765 | if (PageAnon(page)) { |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 4766 | if (!(mc.flags & MOVE_ANON)) |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4767 | return NULL; |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 4768 | } else { |
| 4769 | if (!(mc.flags & MOVE_FILE)) |
| 4770 | return NULL; |
| 4771 | } |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4772 | if (!get_page_unless_zero(page)) |
| 4773 | return NULL; |
| 4774 | |
| 4775 | return page; |
| 4776 | } |
| 4777 | |
KAMEZAWA Hiroyuki | 4b91355 | 2012-05-29 15:06:51 -0700 | [diff] [blame] | 4778 | #ifdef CONFIG_SWAP |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4779 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
| 4780 | unsigned long addr, pte_t ptent, swp_entry_t *entry) |
| 4781 | { |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4782 | struct page *page = NULL; |
| 4783 | swp_entry_t ent = pte_to_swp_entry(ptent); |
| 4784 | |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 4785 | if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent)) |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4786 | return NULL; |
KAMEZAWA Hiroyuki | 4b91355 | 2012-05-29 15:06:51 -0700 | [diff] [blame] | 4787 | /* |
| 4788 | * Because lookup_swap_cache() updates some statistics counter, |
| 4789 | * we call find_get_page() with swapper_space directly. |
| 4790 | */ |
Shaohua Li | 33806f0 | 2013-02-22 16:34:37 -0800 | [diff] [blame] | 4791 | page = find_get_page(swap_address_space(ent), ent.val); |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4792 | if (do_swap_account) |
| 4793 | entry->val = ent.val; |
| 4794 | |
| 4795 | return page; |
| 4796 | } |
KAMEZAWA Hiroyuki | 4b91355 | 2012-05-29 15:06:51 -0700 | [diff] [blame] | 4797 | #else |
| 4798 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
| 4799 | unsigned long addr, pte_t ptent, swp_entry_t *entry) |
| 4800 | { |
| 4801 | return NULL; |
| 4802 | } |
| 4803 | #endif |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4804 | |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 4805 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
| 4806 | unsigned long addr, pte_t ptent, swp_entry_t *entry) |
| 4807 | { |
| 4808 | struct page *page = NULL; |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 4809 | struct address_space *mapping; |
| 4810 | pgoff_t pgoff; |
| 4811 | |
| 4812 | if (!vma->vm_file) /* anonymous vma */ |
| 4813 | return NULL; |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 4814 | if (!(mc.flags & MOVE_FILE)) |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 4815 | return NULL; |
| 4816 | |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 4817 | mapping = vma->vm_file->f_mapping; |
Kirill A. Shutemov | 0661a33 | 2015-02-10 14:10:04 -0800 | [diff] [blame] | 4818 | pgoff = linear_page_index(vma, addr); |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 4819 | |
| 4820 | /* page is moved even if it's not RSS of this task(page-faulted). */ |
Hugh Dickins | aa3b189 | 2011-08-03 16:21:24 -0700 | [diff] [blame] | 4821 | #ifdef CONFIG_SWAP |
| 4822 | /* shmem/tmpfs may report page out on swap: account for that too. */ |
Johannes Weiner | 139b6a6 | 2014-05-06 12:50:05 -0700 | [diff] [blame] | 4823 | if (shmem_mapping(mapping)) { |
| 4824 | page = find_get_entry(mapping, pgoff); |
| 4825 | if (radix_tree_exceptional_entry(page)) { |
| 4826 | swp_entry_t swp = radix_to_swp_entry(page); |
| 4827 | if (do_swap_account) |
| 4828 | *entry = swp; |
| 4829 | page = find_get_page(swap_address_space(swp), swp.val); |
| 4830 | } |
| 4831 | } else |
| 4832 | page = find_get_page(mapping, pgoff); |
| 4833 | #else |
| 4834 | page = find_get_page(mapping, pgoff); |
Hugh Dickins | aa3b189 | 2011-08-03 16:21:24 -0700 | [diff] [blame] | 4835 | #endif |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 4836 | return page; |
| 4837 | } |
| 4838 | |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 4839 | /** |
| 4840 | * mem_cgroup_move_account - move account of the page |
| 4841 | * @page: the page |
| 4842 | * @nr_pages: number of regular pages (>1 for huge pages) |
| 4843 | * @from: mem_cgroup which the page is moved from. |
| 4844 | * @to: mem_cgroup which the page is moved to. @from != @to. |
| 4845 | * |
| 4846 | * The caller must confirm following. |
| 4847 | * - page is not on LRU (isolate_page() is useful.) |
| 4848 | * - compound_lock is held when nr_pages > 1 |
| 4849 | * |
| 4850 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" |
| 4851 | * from old cgroup. |
| 4852 | */ |
| 4853 | static int mem_cgroup_move_account(struct page *page, |
| 4854 | unsigned int nr_pages, |
| 4855 | struct mem_cgroup *from, |
| 4856 | struct mem_cgroup *to) |
| 4857 | { |
| 4858 | unsigned long flags; |
| 4859 | int ret; |
Greg Thelen | c4843a7 | 2015-05-22 17:13:16 -0400 | [diff] [blame] | 4860 | bool anon; |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 4861 | |
| 4862 | VM_BUG_ON(from == to); |
| 4863 | VM_BUG_ON_PAGE(PageLRU(page), page); |
| 4864 | /* |
| 4865 | * The page is isolated from LRU. So, collapse function |
| 4866 | * will not handle this page. But page splitting can happen. |
| 4867 | * Do this check under compound_page_lock(). The caller should |
| 4868 | * hold it. |
| 4869 | */ |
| 4870 | ret = -EBUSY; |
| 4871 | if (nr_pages > 1 && !PageTransHuge(page)) |
| 4872 | goto out; |
| 4873 | |
| 4874 | /* |
| 4875 | * Prevent mem_cgroup_migrate() from looking at page->mem_cgroup |
| 4876 | * of its source page while we change it: page migration takes |
| 4877 | * both pages off the LRU, but page cache replacement doesn't. |
| 4878 | */ |
| 4879 | if (!trylock_page(page)) |
| 4880 | goto out; |
| 4881 | |
| 4882 | ret = -EINVAL; |
| 4883 | if (page->mem_cgroup != from) |
| 4884 | goto out_unlock; |
| 4885 | |
Greg Thelen | c4843a7 | 2015-05-22 17:13:16 -0400 | [diff] [blame] | 4886 | anon = PageAnon(page); |
| 4887 | |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 4888 | spin_lock_irqsave(&from->move_lock, flags); |
| 4889 | |
Greg Thelen | c4843a7 | 2015-05-22 17:13:16 -0400 | [diff] [blame] | 4890 | if (!anon && page_mapped(page)) { |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 4891 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], |
| 4892 | nr_pages); |
| 4893 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], |
| 4894 | nr_pages); |
| 4895 | } |
| 4896 | |
Greg Thelen | c4843a7 | 2015-05-22 17:13:16 -0400 | [diff] [blame] | 4897 | /* |
| 4898 | * move_lock grabbed above and caller set from->moving_account, so |
| 4899 | * mem_cgroup_update_page_stat() will serialize updates to PageDirty. |
| 4900 | * So mapping should be stable for dirty pages. |
| 4901 | */ |
| 4902 | if (!anon && PageDirty(page)) { |
| 4903 | struct address_space *mapping = page_mapping(page); |
| 4904 | |
| 4905 | if (mapping_cap_account_dirty(mapping)) { |
| 4906 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_DIRTY], |
| 4907 | nr_pages); |
| 4908 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_DIRTY], |
| 4909 | nr_pages); |
| 4910 | } |
| 4911 | } |
| 4912 | |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 4913 | if (PageWriteback(page)) { |
| 4914 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK], |
| 4915 | nr_pages); |
| 4916 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK], |
| 4917 | nr_pages); |
| 4918 | } |
| 4919 | |
| 4920 | /* |
| 4921 | * It is safe to change page->mem_cgroup here because the page |
| 4922 | * is referenced, charged, and isolated - we can't race with |
| 4923 | * uncharging, charging, migration, or LRU putback. |
| 4924 | */ |
| 4925 | |
| 4926 | /* caller should have done css_get */ |
| 4927 | page->mem_cgroup = to; |
| 4928 | spin_unlock_irqrestore(&from->move_lock, flags); |
| 4929 | |
| 4930 | ret = 0; |
| 4931 | |
| 4932 | local_irq_disable(); |
| 4933 | mem_cgroup_charge_statistics(to, page, nr_pages); |
| 4934 | memcg_check_events(to, page); |
| 4935 | mem_cgroup_charge_statistics(from, page, -nr_pages); |
| 4936 | memcg_check_events(from, page); |
| 4937 | local_irq_enable(); |
| 4938 | out_unlock: |
| 4939 | unlock_page(page); |
| 4940 | out: |
| 4941 | return ret; |
| 4942 | } |
| 4943 | |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 4944 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4945 | unsigned long addr, pte_t ptent, union mc_target *target) |
| 4946 | { |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4947 | struct page *page = NULL; |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 4948 | enum mc_target_type ret = MC_TARGET_NONE; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4949 | swp_entry_t ent = { .val = 0 }; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4950 | |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4951 | if (pte_present(ptent)) |
| 4952 | page = mc_handle_present_pte(vma, addr, ptent); |
| 4953 | else if (is_swap_pte(ptent)) |
| 4954 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); |
Kirill A. Shutemov | 0661a33 | 2015-02-10 14:10:04 -0800 | [diff] [blame] | 4955 | else if (pte_none(ptent)) |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 4956 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4957 | |
| 4958 | if (!page && !ent.val) |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 4959 | return ret; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4960 | if (page) { |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4961 | /* |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 4962 | * Do only loose check w/o serialization. |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 4963 | * mem_cgroup_move_account() checks the page is valid or |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 4964 | * not under LRU exclusion. |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4965 | */ |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 4966 | if (page->mem_cgroup == mc.from) { |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4967 | ret = MC_TARGET_PAGE; |
| 4968 | if (target) |
| 4969 | target->page = page; |
| 4970 | } |
| 4971 | if (!ret || !target) |
| 4972 | put_page(page); |
| 4973 | } |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4974 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
| 4975 | if (ent.val && !ret && |
Li Zefan | 34c00c3 | 2013-09-23 16:56:01 +0800 | [diff] [blame] | 4976 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
KAMEZAWA Hiroyuki | 7f0f154 | 2010-05-11 14:06:58 -0700 | [diff] [blame] | 4977 | ret = MC_TARGET_SWAP; |
| 4978 | if (target) |
| 4979 | target->ent = ent; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4980 | } |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4981 | return ret; |
| 4982 | } |
| 4983 | |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 4984 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
| 4985 | /* |
| 4986 | * We don't consider swapping or file mapped pages because THP does not |
| 4987 | * support them for now. |
| 4988 | * Caller should make sure that pmd_trans_huge(pmd) is true. |
| 4989 | */ |
| 4990 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, |
| 4991 | unsigned long addr, pmd_t pmd, union mc_target *target) |
| 4992 | { |
| 4993 | struct page *page = NULL; |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 4994 | enum mc_target_type ret = MC_TARGET_NONE; |
| 4995 | |
| 4996 | page = pmd_page(pmd); |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 4997 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 4998 | if (!(mc.flags & MOVE_ANON)) |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 4999 | return ret; |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 5000 | if (page->mem_cgroup == mc.from) { |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 5001 | ret = MC_TARGET_PAGE; |
| 5002 | if (target) { |
| 5003 | get_page(page); |
| 5004 | target->page = page; |
| 5005 | } |
| 5006 | } |
| 5007 | return ret; |
| 5008 | } |
| 5009 | #else |
| 5010 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, |
| 5011 | unsigned long addr, pmd_t pmd, union mc_target *target) |
| 5012 | { |
| 5013 | return MC_TARGET_NONE; |
| 5014 | } |
| 5015 | #endif |
| 5016 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5017 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
| 5018 | unsigned long addr, unsigned long end, |
| 5019 | struct mm_walk *walk) |
| 5020 | { |
Naoya Horiguchi | 26bcd64 | 2015-02-11 15:27:57 -0800 | [diff] [blame] | 5021 | struct vm_area_struct *vma = walk->vma; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5022 | pte_t *pte; |
| 5023 | spinlock_t *ptl; |
| 5024 | |
Kirill A. Shutemov | bf92915 | 2013-11-14 14:30:54 -0800 | [diff] [blame] | 5025 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 5026 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
| 5027 | mc.precharge += HPAGE_PMD_NR; |
Kirill A. Shutemov | bf92915 | 2013-11-14 14:30:54 -0800 | [diff] [blame] | 5028 | spin_unlock(ptl); |
Andrea Arcangeli | 1a5a990 | 2012-03-21 16:33:42 -0700 | [diff] [blame] | 5029 | return 0; |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 5030 | } |
Dave Hansen | 0331932 | 2011-03-22 16:32:56 -0700 | [diff] [blame] | 5031 | |
Andrea Arcangeli | 45f83ce | 2012-03-28 14:42:40 -0700 | [diff] [blame] | 5032 | if (pmd_trans_unstable(pmd)) |
| 5033 | return 0; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5034 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
| 5035 | for (; addr != end; pte++, addr += PAGE_SIZE) |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 5036 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5037 | mc.precharge++; /* increment precharge temporarily */ |
| 5038 | pte_unmap_unlock(pte - 1, ptl); |
| 5039 | cond_resched(); |
| 5040 | |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5041 | return 0; |
| 5042 | } |
| 5043 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5044 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
| 5045 | { |
| 5046 | unsigned long precharge; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5047 | |
Naoya Horiguchi | 26bcd64 | 2015-02-11 15:27:57 -0800 | [diff] [blame] | 5048 | struct mm_walk mem_cgroup_count_precharge_walk = { |
| 5049 | .pmd_entry = mem_cgroup_count_precharge_pte_range, |
| 5050 | .mm = mm, |
| 5051 | }; |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 5052 | down_read(&mm->mmap_sem); |
Naoya Horiguchi | 26bcd64 | 2015-02-11 15:27:57 -0800 | [diff] [blame] | 5053 | walk_page_range(0, ~0UL, &mem_cgroup_count_precharge_walk); |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 5054 | up_read(&mm->mmap_sem); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5055 | |
| 5056 | precharge = mc.precharge; |
| 5057 | mc.precharge = 0; |
| 5058 | |
| 5059 | return precharge; |
| 5060 | } |
| 5061 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5062 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
| 5063 | { |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 5064 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
| 5065 | |
| 5066 | VM_BUG_ON(mc.moving_task); |
| 5067 | mc.moving_task = current; |
| 5068 | return mem_cgroup_do_precharge(precharge); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5069 | } |
| 5070 | |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 5071 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
| 5072 | static void __mem_cgroup_clear_mc(void) |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5073 | { |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 5074 | struct mem_cgroup *from = mc.from; |
| 5075 | struct mem_cgroup *to = mc.to; |
| 5076 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5077 | /* we must uncharge all the leftover precharges from mc.to */ |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 5078 | if (mc.precharge) { |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5079 | cancel_charge(mc.to, mc.precharge); |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 5080 | mc.precharge = 0; |
| 5081 | } |
| 5082 | /* |
| 5083 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so |
| 5084 | * we must uncharge here. |
| 5085 | */ |
| 5086 | if (mc.moved_charge) { |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5087 | cancel_charge(mc.from, mc.moved_charge); |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 5088 | mc.moved_charge = 0; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5089 | } |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 5090 | /* we must fixup refcnts and charges */ |
| 5091 | if (mc.moved_swap) { |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 5092 | /* uncharge swap account from the old cgroup */ |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 5093 | if (!mem_cgroup_is_root(mc.from)) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 5094 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 5095 | |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 5096 | /* |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 5097 | * we charged both to->memory and to->memsw, so we |
| 5098 | * should uncharge to->memory. |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 5099 | */ |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 5100 | if (!mem_cgroup_is_root(mc.to)) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 5101 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 5102 | |
Johannes Weiner | e8ea14c | 2014-12-10 15:42:42 -0800 | [diff] [blame] | 5103 | css_put_many(&mc.from->css, mc.moved_swap); |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 5104 | |
Li Zefan | 4050377 | 2013-07-08 16:00:34 -0700 | [diff] [blame] | 5105 | /* we've already done css_get(mc.to) */ |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 5106 | mc.moved_swap = 0; |
| 5107 | } |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 5108 | memcg_oom_recover(from); |
| 5109 | memcg_oom_recover(to); |
| 5110 | wake_up_all(&mc.waitq); |
| 5111 | } |
| 5112 | |
| 5113 | static void mem_cgroup_clear_mc(void) |
| 5114 | { |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 5115 | /* |
| 5116 | * we must clear moving_task before waking up waiters at the end of |
| 5117 | * task migration. |
| 5118 | */ |
| 5119 | mc.moving_task = NULL; |
| 5120 | __mem_cgroup_clear_mc(); |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 5121 | spin_lock(&mc.lock); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5122 | mc.from = NULL; |
| 5123 | mc.to = NULL; |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 5124 | spin_unlock(&mc.lock); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5125 | } |
| 5126 | |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 5127 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
Li Zefan | 761b3ef | 2012-01-31 13:47:36 +0800 | [diff] [blame] | 5128 | struct cgroup_taskset *tset) |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5129 | { |
Tejun Heo | 2f7ee56 | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 5130 | struct task_struct *p = cgroup_taskset_first(tset); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5131 | int ret = 0; |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 5132 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 5133 | unsigned long move_flags; |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5134 | |
Glauber Costa | ee5e847 | 2013-02-22 16:34:50 -0800 | [diff] [blame] | 5135 | /* |
| 5136 | * We are now commited to this value whatever it is. Changes in this |
| 5137 | * tunable will only affect upcoming migrations, not the current one. |
| 5138 | * So we need to save it, and keep it going. |
| 5139 | */ |
Jason Low | 4db0c3c | 2015-04-15 16:14:08 -0700 | [diff] [blame] | 5140 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 5141 | if (move_flags) { |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5142 | struct mm_struct *mm; |
| 5143 | struct mem_cgroup *from = mem_cgroup_from_task(p); |
| 5144 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 5145 | VM_BUG_ON(from == memcg); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5146 | |
| 5147 | mm = get_task_mm(p); |
| 5148 | if (!mm) |
| 5149 | return 0; |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5150 | /* We move charges only when we move a owner of the mm */ |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5151 | if (mm->owner == p) { |
| 5152 | VM_BUG_ON(mc.from); |
| 5153 | VM_BUG_ON(mc.to); |
| 5154 | VM_BUG_ON(mc.precharge); |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 5155 | VM_BUG_ON(mc.moved_charge); |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 5156 | VM_BUG_ON(mc.moved_swap); |
Johannes Weiner | 247b144 | 2014-12-10 15:44:11 -0800 | [diff] [blame] | 5157 | |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 5158 | spin_lock(&mc.lock); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5159 | mc.from = from; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 5160 | mc.to = memcg; |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 5161 | mc.flags = move_flags; |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 5162 | spin_unlock(&mc.lock); |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 5163 | /* We set mc.moving_task later */ |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5164 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5165 | ret = mem_cgroup_precharge_mc(mm); |
| 5166 | if (ret) |
| 5167 | mem_cgroup_clear_mc(); |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 5168 | } |
| 5169 | mmput(mm); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5170 | } |
| 5171 | return ret; |
| 5172 | } |
| 5173 | |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 5174 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
Li Zefan | 761b3ef | 2012-01-31 13:47:36 +0800 | [diff] [blame] | 5175 | struct cgroup_taskset *tset) |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5176 | { |
Johannes Weiner | 4e2f245 | 2014-12-10 15:44:08 -0800 | [diff] [blame] | 5177 | if (mc.to) |
| 5178 | mem_cgroup_clear_mc(); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5179 | } |
| 5180 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5181 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
| 5182 | unsigned long addr, unsigned long end, |
| 5183 | struct mm_walk *walk) |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5184 | { |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5185 | int ret = 0; |
Naoya Horiguchi | 26bcd64 | 2015-02-11 15:27:57 -0800 | [diff] [blame] | 5186 | struct vm_area_struct *vma = walk->vma; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5187 | pte_t *pte; |
| 5188 | spinlock_t *ptl; |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 5189 | enum mc_target_type target_type; |
| 5190 | union mc_target target; |
| 5191 | struct page *page; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5192 | |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 5193 | /* |
| 5194 | * We don't take compound_lock() here but no race with splitting thp |
| 5195 | * happens because: |
| 5196 | * - if pmd_trans_huge_lock() returns 1, the relevant thp is not |
| 5197 | * under splitting, which means there's no concurrent thp split, |
| 5198 | * - if another thread runs into split_huge_page() just after we |
| 5199 | * entered this if-block, the thread must wait for page table lock |
| 5200 | * to be unlocked in __split_huge_page_splitting(), where the main |
| 5201 | * part of thp split is not executed yet. |
| 5202 | */ |
Kirill A. Shutemov | bf92915 | 2013-11-14 14:30:54 -0800 | [diff] [blame] | 5203 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
Hugh Dickins | 62ade86 | 2012-05-18 11:28:34 -0700 | [diff] [blame] | 5204 | if (mc.precharge < HPAGE_PMD_NR) { |
Kirill A. Shutemov | bf92915 | 2013-11-14 14:30:54 -0800 | [diff] [blame] | 5205 | spin_unlock(ptl); |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 5206 | return 0; |
| 5207 | } |
| 5208 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); |
| 5209 | if (target_type == MC_TARGET_PAGE) { |
| 5210 | page = target.page; |
| 5211 | if (!isolate_lru_page(page)) { |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 5212 | if (!mem_cgroup_move_account(page, HPAGE_PMD_NR, |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 5213 | mc.from, mc.to)) { |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 5214 | mc.precharge -= HPAGE_PMD_NR; |
| 5215 | mc.moved_charge += HPAGE_PMD_NR; |
| 5216 | } |
| 5217 | putback_lru_page(page); |
| 5218 | } |
| 5219 | put_page(page); |
| 5220 | } |
Kirill A. Shutemov | bf92915 | 2013-11-14 14:30:54 -0800 | [diff] [blame] | 5221 | spin_unlock(ptl); |
Andrea Arcangeli | 1a5a990 | 2012-03-21 16:33:42 -0700 | [diff] [blame] | 5222 | return 0; |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 5223 | } |
| 5224 | |
Andrea Arcangeli | 45f83ce | 2012-03-28 14:42:40 -0700 | [diff] [blame] | 5225 | if (pmd_trans_unstable(pmd)) |
| 5226 | return 0; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5227 | retry: |
| 5228 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
| 5229 | for (; addr != end; addr += PAGE_SIZE) { |
| 5230 | pte_t ptent = *(pte++); |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 5231 | swp_entry_t ent; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5232 | |
| 5233 | if (!mc.precharge) |
| 5234 | break; |
| 5235 | |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 5236 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5237 | case MC_TARGET_PAGE: |
| 5238 | page = target.page; |
| 5239 | if (isolate_lru_page(page)) |
| 5240 | goto put; |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 5241 | if (!mem_cgroup_move_account(page, 1, mc.from, mc.to)) { |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5242 | mc.precharge--; |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 5243 | /* we uncharge from mc.from later. */ |
| 5244 | mc.moved_charge++; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5245 | } |
| 5246 | putback_lru_page(page); |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 5247 | put: /* get_mctgt_type() gets the page */ |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5248 | put_page(page); |
| 5249 | break; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 5250 | case MC_TARGET_SWAP: |
| 5251 | ent = target.ent; |
Hugh Dickins | e91cbb4 | 2012-05-29 15:06:51 -0700 | [diff] [blame] | 5252 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 5253 | mc.precharge--; |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 5254 | /* we fixup refcnts and charges later. */ |
| 5255 | mc.moved_swap++; |
| 5256 | } |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 5257 | break; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5258 | default: |
| 5259 | break; |
| 5260 | } |
| 5261 | } |
| 5262 | pte_unmap_unlock(pte - 1, ptl); |
| 5263 | cond_resched(); |
| 5264 | |
| 5265 | if (addr != end) { |
| 5266 | /* |
| 5267 | * We have consumed all precharges we got in can_attach(). |
| 5268 | * We try charge one by one, but don't do any additional |
| 5269 | * charges to mc.to if we have failed in charge once in attach() |
| 5270 | * phase. |
| 5271 | */ |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 5272 | ret = mem_cgroup_do_precharge(1); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5273 | if (!ret) |
| 5274 | goto retry; |
| 5275 | } |
| 5276 | |
| 5277 | return ret; |
| 5278 | } |
| 5279 | |
| 5280 | static void mem_cgroup_move_charge(struct mm_struct *mm) |
| 5281 | { |
Naoya Horiguchi | 26bcd64 | 2015-02-11 15:27:57 -0800 | [diff] [blame] | 5282 | struct mm_walk mem_cgroup_move_charge_walk = { |
| 5283 | .pmd_entry = mem_cgroup_move_charge_pte_range, |
| 5284 | .mm = mm, |
| 5285 | }; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5286 | |
| 5287 | lru_add_drain_all(); |
Johannes Weiner | 312722c | 2014-12-10 15:44:25 -0800 | [diff] [blame] | 5288 | /* |
| 5289 | * Signal mem_cgroup_begin_page_stat() to take the memcg's |
| 5290 | * move_lock while we're moving its pages to another memcg. |
| 5291 | * Then wait for already started RCU-only updates to finish. |
| 5292 | */ |
| 5293 | atomic_inc(&mc.from->moving_account); |
| 5294 | synchronize_rcu(); |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 5295 | retry: |
| 5296 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { |
| 5297 | /* |
| 5298 | * Someone who are holding the mmap_sem might be waiting in |
| 5299 | * waitq. So we cancel all extra charges, wake up all waiters, |
| 5300 | * and retry. Because we cancel precharges, we might not be able |
| 5301 | * to move enough charges, but moving charge is a best-effort |
| 5302 | * feature anyway, so it wouldn't be a big problem. |
| 5303 | */ |
| 5304 | __mem_cgroup_clear_mc(); |
| 5305 | cond_resched(); |
| 5306 | goto retry; |
| 5307 | } |
Naoya Horiguchi | 26bcd64 | 2015-02-11 15:27:57 -0800 | [diff] [blame] | 5308 | /* |
| 5309 | * When we have consumed all precharges and failed in doing |
| 5310 | * additional charge, the page walk just aborts. |
| 5311 | */ |
| 5312 | walk_page_range(0, ~0UL, &mem_cgroup_move_charge_walk); |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 5313 | up_read(&mm->mmap_sem); |
Johannes Weiner | 312722c | 2014-12-10 15:44:25 -0800 | [diff] [blame] | 5314 | atomic_dec(&mc.from->moving_account); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5315 | } |
| 5316 | |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 5317 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
Li Zefan | 761b3ef | 2012-01-31 13:47:36 +0800 | [diff] [blame] | 5318 | struct cgroup_taskset *tset) |
Balbir Singh | 67e465a | 2008-02-07 00:13:54 -0800 | [diff] [blame] | 5319 | { |
Tejun Heo | 2f7ee56 | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 5320 | struct task_struct *p = cgroup_taskset_first(tset); |
KOSAKI Motohiro | a433658 | 2011-06-15 15:08:13 -0700 | [diff] [blame] | 5321 | struct mm_struct *mm = get_task_mm(p); |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 5322 | |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 5323 | if (mm) { |
KOSAKI Motohiro | a433658 | 2011-06-15 15:08:13 -0700 | [diff] [blame] | 5324 | if (mc.to) |
| 5325 | mem_cgroup_move_charge(mm); |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 5326 | mmput(mm); |
| 5327 | } |
KOSAKI Motohiro | a433658 | 2011-06-15 15:08:13 -0700 | [diff] [blame] | 5328 | if (mc.to) |
| 5329 | mem_cgroup_clear_mc(); |
Balbir Singh | 67e465a | 2008-02-07 00:13:54 -0800 | [diff] [blame] | 5330 | } |
Daisuke Nishimura | 5cfb80a | 2010-03-23 13:35:11 -0700 | [diff] [blame] | 5331 | #else /* !CONFIG_MMU */ |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 5332 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
Li Zefan | 761b3ef | 2012-01-31 13:47:36 +0800 | [diff] [blame] | 5333 | struct cgroup_taskset *tset) |
Daisuke Nishimura | 5cfb80a | 2010-03-23 13:35:11 -0700 | [diff] [blame] | 5334 | { |
| 5335 | return 0; |
| 5336 | } |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 5337 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
Li Zefan | 761b3ef | 2012-01-31 13:47:36 +0800 | [diff] [blame] | 5338 | struct cgroup_taskset *tset) |
Daisuke Nishimura | 5cfb80a | 2010-03-23 13:35:11 -0700 | [diff] [blame] | 5339 | { |
| 5340 | } |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 5341 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
Li Zefan | 761b3ef | 2012-01-31 13:47:36 +0800 | [diff] [blame] | 5342 | struct cgroup_taskset *tset) |
Daisuke Nishimura | 5cfb80a | 2010-03-23 13:35:11 -0700 | [diff] [blame] | 5343 | { |
| 5344 | } |
| 5345 | #endif |
Balbir Singh | 67e465a | 2008-02-07 00:13:54 -0800 | [diff] [blame] | 5346 | |
Tejun Heo | f00baae | 2013-04-15 13:41:15 -0700 | [diff] [blame] | 5347 | /* |
| 5348 | * Cgroup retains root cgroups across [un]mount cycles making it necessary |
Tejun Heo | aa6ec29 | 2014-07-09 10:08:08 -0400 | [diff] [blame] | 5349 | * to verify whether we're attached to the default hierarchy on each mount |
| 5350 | * attempt. |
Tejun Heo | f00baae | 2013-04-15 13:41:15 -0700 | [diff] [blame] | 5351 | */ |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 5352 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
Tejun Heo | f00baae | 2013-04-15 13:41:15 -0700 | [diff] [blame] | 5353 | { |
| 5354 | /* |
Tejun Heo | aa6ec29 | 2014-07-09 10:08:08 -0400 | [diff] [blame] | 5355 | * use_hierarchy is forced on the default hierarchy. cgroup core |
Tejun Heo | f00baae | 2013-04-15 13:41:15 -0700 | [diff] [blame] | 5356 | * guarantees that @root doesn't have any children, so turning it |
| 5357 | * on for the root memcg is enough. |
| 5358 | */ |
Tejun Heo | aa6ec29 | 2014-07-09 10:08:08 -0400 | [diff] [blame] | 5359 | if (cgroup_on_dfl(root_css->cgroup)) |
Vladimir Davydov | 7feee59 | 2015-03-12 16:26:19 -0700 | [diff] [blame] | 5360 | root_mem_cgroup->use_hierarchy = true; |
| 5361 | else |
| 5362 | root_mem_cgroup->use_hierarchy = false; |
Tejun Heo | f00baae | 2013-04-15 13:41:15 -0700 | [diff] [blame] | 5363 | } |
| 5364 | |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5365 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
| 5366 | struct cftype *cft) |
| 5367 | { |
| 5368 | return mem_cgroup_usage(mem_cgroup_from_css(css), false); |
| 5369 | } |
| 5370 | |
| 5371 | static int memory_low_show(struct seq_file *m, void *v) |
| 5372 | { |
| 5373 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
Jason Low | 4db0c3c | 2015-04-15 16:14:08 -0700 | [diff] [blame] | 5374 | unsigned long low = READ_ONCE(memcg->low); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5375 | |
| 5376 | if (low == PAGE_COUNTER_MAX) |
Johannes Weiner | d297369 | 2015-02-27 15:52:04 -0800 | [diff] [blame] | 5377 | seq_puts(m, "max\n"); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5378 | else |
| 5379 | seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE); |
| 5380 | |
| 5381 | return 0; |
| 5382 | } |
| 5383 | |
| 5384 | static ssize_t memory_low_write(struct kernfs_open_file *of, |
| 5385 | char *buf, size_t nbytes, loff_t off) |
| 5386 | { |
| 5387 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
| 5388 | unsigned long low; |
| 5389 | int err; |
| 5390 | |
| 5391 | buf = strstrip(buf); |
Johannes Weiner | d297369 | 2015-02-27 15:52:04 -0800 | [diff] [blame] | 5392 | err = page_counter_memparse(buf, "max", &low); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5393 | if (err) |
| 5394 | return err; |
| 5395 | |
| 5396 | memcg->low = low; |
| 5397 | |
| 5398 | return nbytes; |
| 5399 | } |
| 5400 | |
| 5401 | static int memory_high_show(struct seq_file *m, void *v) |
| 5402 | { |
| 5403 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
Jason Low | 4db0c3c | 2015-04-15 16:14:08 -0700 | [diff] [blame] | 5404 | unsigned long high = READ_ONCE(memcg->high); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5405 | |
| 5406 | if (high == PAGE_COUNTER_MAX) |
Johannes Weiner | d297369 | 2015-02-27 15:52:04 -0800 | [diff] [blame] | 5407 | seq_puts(m, "max\n"); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5408 | else |
| 5409 | seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE); |
| 5410 | |
| 5411 | return 0; |
| 5412 | } |
| 5413 | |
| 5414 | static ssize_t memory_high_write(struct kernfs_open_file *of, |
| 5415 | char *buf, size_t nbytes, loff_t off) |
| 5416 | { |
| 5417 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
| 5418 | unsigned long high; |
| 5419 | int err; |
| 5420 | |
| 5421 | buf = strstrip(buf); |
Johannes Weiner | d297369 | 2015-02-27 15:52:04 -0800 | [diff] [blame] | 5422 | err = page_counter_memparse(buf, "max", &high); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5423 | if (err) |
| 5424 | return err; |
| 5425 | |
| 5426 | memcg->high = high; |
| 5427 | |
Tejun Heo | 2529bb3 | 2015-05-22 18:23:34 -0400 | [diff] [blame] | 5428 | memcg_wb_domain_size_changed(memcg); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5429 | return nbytes; |
| 5430 | } |
| 5431 | |
| 5432 | static int memory_max_show(struct seq_file *m, void *v) |
| 5433 | { |
| 5434 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
Jason Low | 4db0c3c | 2015-04-15 16:14:08 -0700 | [diff] [blame] | 5435 | unsigned long max = READ_ONCE(memcg->memory.limit); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5436 | |
| 5437 | if (max == PAGE_COUNTER_MAX) |
Johannes Weiner | d297369 | 2015-02-27 15:52:04 -0800 | [diff] [blame] | 5438 | seq_puts(m, "max\n"); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5439 | else |
| 5440 | seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE); |
| 5441 | |
| 5442 | return 0; |
| 5443 | } |
| 5444 | |
| 5445 | static ssize_t memory_max_write(struct kernfs_open_file *of, |
| 5446 | char *buf, size_t nbytes, loff_t off) |
| 5447 | { |
| 5448 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
| 5449 | unsigned long max; |
| 5450 | int err; |
| 5451 | |
| 5452 | buf = strstrip(buf); |
Johannes Weiner | d297369 | 2015-02-27 15:52:04 -0800 | [diff] [blame] | 5453 | err = page_counter_memparse(buf, "max", &max); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5454 | if (err) |
| 5455 | return err; |
| 5456 | |
| 5457 | err = mem_cgroup_resize_limit(memcg, max); |
| 5458 | if (err) |
| 5459 | return err; |
| 5460 | |
Tejun Heo | 2529bb3 | 2015-05-22 18:23:34 -0400 | [diff] [blame] | 5461 | memcg_wb_domain_size_changed(memcg); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5462 | return nbytes; |
| 5463 | } |
| 5464 | |
| 5465 | static int memory_events_show(struct seq_file *m, void *v) |
| 5466 | { |
| 5467 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
| 5468 | |
| 5469 | seq_printf(m, "low %lu\n", mem_cgroup_read_events(memcg, MEMCG_LOW)); |
| 5470 | seq_printf(m, "high %lu\n", mem_cgroup_read_events(memcg, MEMCG_HIGH)); |
| 5471 | seq_printf(m, "max %lu\n", mem_cgroup_read_events(memcg, MEMCG_MAX)); |
| 5472 | seq_printf(m, "oom %lu\n", mem_cgroup_read_events(memcg, MEMCG_OOM)); |
| 5473 | |
| 5474 | return 0; |
| 5475 | } |
| 5476 | |
| 5477 | static struct cftype memory_files[] = { |
| 5478 | { |
| 5479 | .name = "current", |
| 5480 | .read_u64 = memory_current_read, |
| 5481 | }, |
| 5482 | { |
| 5483 | .name = "low", |
| 5484 | .flags = CFTYPE_NOT_ON_ROOT, |
| 5485 | .seq_show = memory_low_show, |
| 5486 | .write = memory_low_write, |
| 5487 | }, |
| 5488 | { |
| 5489 | .name = "high", |
| 5490 | .flags = CFTYPE_NOT_ON_ROOT, |
| 5491 | .seq_show = memory_high_show, |
| 5492 | .write = memory_high_write, |
| 5493 | }, |
| 5494 | { |
| 5495 | .name = "max", |
| 5496 | .flags = CFTYPE_NOT_ON_ROOT, |
| 5497 | .seq_show = memory_max_show, |
| 5498 | .write = memory_max_write, |
| 5499 | }, |
| 5500 | { |
| 5501 | .name = "events", |
| 5502 | .flags = CFTYPE_NOT_ON_ROOT, |
| 5503 | .seq_show = memory_events_show, |
| 5504 | }, |
| 5505 | { } /* terminate */ |
| 5506 | }; |
| 5507 | |
Tejun Heo | 073219e | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 5508 | struct cgroup_subsys memory_cgrp_subsys = { |
Tejun Heo | 92fb974 | 2012-11-19 08:13:38 -0800 | [diff] [blame] | 5509 | .css_alloc = mem_cgroup_css_alloc, |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 5510 | .css_online = mem_cgroup_css_online, |
Tejun Heo | 92fb974 | 2012-11-19 08:13:38 -0800 | [diff] [blame] | 5511 | .css_offline = mem_cgroup_css_offline, |
| 5512 | .css_free = mem_cgroup_css_free, |
Tejun Heo | 1ced953 | 2014-07-08 18:02:57 -0400 | [diff] [blame] | 5513 | .css_reset = mem_cgroup_css_reset, |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5514 | .can_attach = mem_cgroup_can_attach, |
| 5515 | .cancel_attach = mem_cgroup_cancel_attach, |
Balbir Singh | 67e465a | 2008-02-07 00:13:54 -0800 | [diff] [blame] | 5516 | .attach = mem_cgroup_move_task, |
Tejun Heo | f00baae | 2013-04-15 13:41:15 -0700 | [diff] [blame] | 5517 | .bind = mem_cgroup_bind, |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5518 | .dfl_cftypes = memory_files, |
| 5519 | .legacy_cftypes = mem_cgroup_legacy_files, |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 5520 | .early_init = 0, |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 5521 | }; |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 5522 | |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5523 | /** |
| 5524 | * mem_cgroup_events - count memory events against a cgroup |
| 5525 | * @memcg: the memory cgroup |
| 5526 | * @idx: the event index |
| 5527 | * @nr: the number of events to account for |
| 5528 | */ |
| 5529 | void mem_cgroup_events(struct mem_cgroup *memcg, |
| 5530 | enum mem_cgroup_events_index idx, |
| 5531 | unsigned int nr) |
| 5532 | { |
| 5533 | this_cpu_add(memcg->stat->events[idx], nr); |
| 5534 | } |
| 5535 | |
| 5536 | /** |
| 5537 | * mem_cgroup_low - check if memory consumption is below the normal range |
| 5538 | * @root: the highest ancestor to consider |
| 5539 | * @memcg: the memory cgroup to check |
| 5540 | * |
| 5541 | * Returns %true if memory consumption of @memcg, and that of all |
| 5542 | * configurable ancestors up to @root, is below the normal range. |
| 5543 | */ |
| 5544 | bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg) |
| 5545 | { |
| 5546 | if (mem_cgroup_disabled()) |
| 5547 | return false; |
| 5548 | |
| 5549 | /* |
| 5550 | * The toplevel group doesn't have a configurable range, so |
| 5551 | * it's never low when looked at directly, and it is not |
| 5552 | * considered an ancestor when assessing the hierarchy. |
| 5553 | */ |
| 5554 | |
| 5555 | if (memcg == root_mem_cgroup) |
| 5556 | return false; |
| 5557 | |
Michal Hocko | 4e54ded | 2015-02-27 15:51:46 -0800 | [diff] [blame] | 5558 | if (page_counter_read(&memcg->memory) >= memcg->low) |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5559 | return false; |
| 5560 | |
| 5561 | while (memcg != root) { |
| 5562 | memcg = parent_mem_cgroup(memcg); |
| 5563 | |
| 5564 | if (memcg == root_mem_cgroup) |
| 5565 | break; |
| 5566 | |
Michal Hocko | 4e54ded | 2015-02-27 15:51:46 -0800 | [diff] [blame] | 5567 | if (page_counter_read(&memcg->memory) >= memcg->low) |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5568 | return false; |
| 5569 | } |
| 5570 | return true; |
| 5571 | } |
| 5572 | |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5573 | /** |
| 5574 | * mem_cgroup_try_charge - try charging a page |
| 5575 | * @page: page to charge |
| 5576 | * @mm: mm context of the victim |
| 5577 | * @gfp_mask: reclaim mode |
| 5578 | * @memcgp: charged memcg return |
| 5579 | * |
| 5580 | * Try to charge @page to the memcg that @mm belongs to, reclaiming |
| 5581 | * pages according to @gfp_mask if necessary. |
| 5582 | * |
| 5583 | * Returns 0 on success, with *@memcgp pointing to the charged memcg. |
| 5584 | * Otherwise, an error code is returned. |
| 5585 | * |
| 5586 | * After page->mapping has been set up, the caller must finalize the |
| 5587 | * charge with mem_cgroup_commit_charge(). Or abort the transaction |
| 5588 | * with mem_cgroup_cancel_charge() in case page instantiation fails. |
| 5589 | */ |
| 5590 | int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm, |
| 5591 | gfp_t gfp_mask, struct mem_cgroup **memcgp) |
| 5592 | { |
| 5593 | struct mem_cgroup *memcg = NULL; |
| 5594 | unsigned int nr_pages = 1; |
| 5595 | int ret = 0; |
| 5596 | |
| 5597 | if (mem_cgroup_disabled()) |
| 5598 | goto out; |
| 5599 | |
| 5600 | if (PageSwapCache(page)) { |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5601 | /* |
| 5602 | * Every swap fault against a single page tries to charge the |
| 5603 | * page, bail as early as possible. shmem_unuse() encounters |
| 5604 | * already charged pages, too. The USED bit is protected by |
| 5605 | * the page lock, which serializes swap cache removal, which |
| 5606 | * in turn serializes uncharging. |
| 5607 | */ |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 5608 | if (page->mem_cgroup) |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5609 | goto out; |
| 5610 | } |
| 5611 | |
| 5612 | if (PageTransHuge(page)) { |
| 5613 | nr_pages <<= compound_order(page); |
| 5614 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); |
| 5615 | } |
| 5616 | |
| 5617 | if (do_swap_account && PageSwapCache(page)) |
| 5618 | memcg = try_get_mem_cgroup_from_page(page); |
| 5619 | if (!memcg) |
| 5620 | memcg = get_mem_cgroup_from_mm(mm); |
| 5621 | |
| 5622 | ret = try_charge(memcg, gfp_mask, nr_pages); |
| 5623 | |
| 5624 | css_put(&memcg->css); |
| 5625 | |
| 5626 | if (ret == -EINTR) { |
| 5627 | memcg = root_mem_cgroup; |
| 5628 | ret = 0; |
| 5629 | } |
| 5630 | out: |
| 5631 | *memcgp = memcg; |
| 5632 | return ret; |
| 5633 | } |
| 5634 | |
| 5635 | /** |
| 5636 | * mem_cgroup_commit_charge - commit a page charge |
| 5637 | * @page: page to charge |
| 5638 | * @memcg: memcg to charge the page to |
| 5639 | * @lrucare: page might be on LRU already |
| 5640 | * |
| 5641 | * Finalize a charge transaction started by mem_cgroup_try_charge(), |
| 5642 | * after page->mapping has been set up. This must happen atomically |
| 5643 | * as part of the page instantiation, i.e. under the page table lock |
| 5644 | * for anonymous pages, under the page lock for page and swap cache. |
| 5645 | * |
| 5646 | * In addition, the page must not be on the LRU during the commit, to |
| 5647 | * prevent racing with task migration. If it might be, use @lrucare. |
| 5648 | * |
| 5649 | * Use mem_cgroup_cancel_charge() to cancel the transaction instead. |
| 5650 | */ |
| 5651 | void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg, |
| 5652 | bool lrucare) |
| 5653 | { |
| 5654 | unsigned int nr_pages = 1; |
| 5655 | |
| 5656 | VM_BUG_ON_PAGE(!page->mapping, page); |
| 5657 | VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page); |
| 5658 | |
| 5659 | if (mem_cgroup_disabled()) |
| 5660 | return; |
| 5661 | /* |
| 5662 | * Swap faults will attempt to charge the same page multiple |
| 5663 | * times. But reuse_swap_page() might have removed the page |
| 5664 | * from swapcache already, so we can't check PageSwapCache(). |
| 5665 | */ |
| 5666 | if (!memcg) |
| 5667 | return; |
| 5668 | |
Johannes Weiner | 6abb5a8 | 2014-08-08 14:19:33 -0700 | [diff] [blame] | 5669 | commit_charge(page, memcg, lrucare); |
| 5670 | |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5671 | if (PageTransHuge(page)) { |
| 5672 | nr_pages <<= compound_order(page); |
| 5673 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); |
| 5674 | } |
| 5675 | |
Johannes Weiner | 6abb5a8 | 2014-08-08 14:19:33 -0700 | [diff] [blame] | 5676 | local_irq_disable(); |
| 5677 | mem_cgroup_charge_statistics(memcg, page, nr_pages); |
| 5678 | memcg_check_events(memcg, page); |
| 5679 | local_irq_enable(); |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5680 | |
| 5681 | if (do_swap_account && PageSwapCache(page)) { |
| 5682 | swp_entry_t entry = { .val = page_private(page) }; |
| 5683 | /* |
| 5684 | * The swap entry might not get freed for a long time, |
| 5685 | * let's not wait for it. The page already received a |
| 5686 | * memory+swap charge, drop the swap entry duplicate. |
| 5687 | */ |
| 5688 | mem_cgroup_uncharge_swap(entry); |
| 5689 | } |
| 5690 | } |
| 5691 | |
| 5692 | /** |
| 5693 | * mem_cgroup_cancel_charge - cancel a page charge |
| 5694 | * @page: page to charge |
| 5695 | * @memcg: memcg to charge the page to |
| 5696 | * |
| 5697 | * Cancel a charge transaction started by mem_cgroup_try_charge(). |
| 5698 | */ |
| 5699 | void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg) |
| 5700 | { |
| 5701 | unsigned int nr_pages = 1; |
| 5702 | |
| 5703 | if (mem_cgroup_disabled()) |
| 5704 | return; |
| 5705 | /* |
| 5706 | * Swap faults will attempt to charge the same page multiple |
| 5707 | * times. But reuse_swap_page() might have removed the page |
| 5708 | * from swapcache already, so we can't check PageSwapCache(). |
| 5709 | */ |
| 5710 | if (!memcg) |
| 5711 | return; |
| 5712 | |
| 5713 | if (PageTransHuge(page)) { |
| 5714 | nr_pages <<= compound_order(page); |
| 5715 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); |
| 5716 | } |
| 5717 | |
| 5718 | cancel_charge(memcg, nr_pages); |
| 5719 | } |
| 5720 | |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5721 | static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout, |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5722 | unsigned long nr_anon, unsigned long nr_file, |
| 5723 | unsigned long nr_huge, struct page *dummy_page) |
| 5724 | { |
Johannes Weiner | 18eca2e | 2014-12-10 15:43:57 -0800 | [diff] [blame] | 5725 | unsigned long nr_pages = nr_anon + nr_file; |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5726 | unsigned long flags; |
| 5727 | |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 5728 | if (!mem_cgroup_is_root(memcg)) { |
Johannes Weiner | 18eca2e | 2014-12-10 15:43:57 -0800 | [diff] [blame] | 5729 | page_counter_uncharge(&memcg->memory, nr_pages); |
| 5730 | if (do_swap_account) |
| 5731 | page_counter_uncharge(&memcg->memsw, nr_pages); |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 5732 | memcg_oom_recover(memcg); |
| 5733 | } |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5734 | |
| 5735 | local_irq_save(flags); |
| 5736 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS], nr_anon); |
| 5737 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_CACHE], nr_file); |
| 5738 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], nr_huge); |
| 5739 | __this_cpu_add(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT], pgpgout); |
Johannes Weiner | 18eca2e | 2014-12-10 15:43:57 -0800 | [diff] [blame] | 5740 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5741 | memcg_check_events(memcg, dummy_page); |
| 5742 | local_irq_restore(flags); |
Johannes Weiner | e8ea14c | 2014-12-10 15:42:42 -0800 | [diff] [blame] | 5743 | |
| 5744 | if (!mem_cgroup_is_root(memcg)) |
Johannes Weiner | 18eca2e | 2014-12-10 15:43:57 -0800 | [diff] [blame] | 5745 | css_put_many(&memcg->css, nr_pages); |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5746 | } |
| 5747 | |
| 5748 | static void uncharge_list(struct list_head *page_list) |
| 5749 | { |
| 5750 | struct mem_cgroup *memcg = NULL; |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5751 | unsigned long nr_anon = 0; |
| 5752 | unsigned long nr_file = 0; |
| 5753 | unsigned long nr_huge = 0; |
| 5754 | unsigned long pgpgout = 0; |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5755 | struct list_head *next; |
| 5756 | struct page *page; |
| 5757 | |
| 5758 | next = page_list->next; |
| 5759 | do { |
| 5760 | unsigned int nr_pages = 1; |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5761 | |
| 5762 | page = list_entry(next, struct page, lru); |
| 5763 | next = page->lru.next; |
| 5764 | |
| 5765 | VM_BUG_ON_PAGE(PageLRU(page), page); |
| 5766 | VM_BUG_ON_PAGE(page_count(page), page); |
| 5767 | |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 5768 | if (!page->mem_cgroup) |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5769 | continue; |
| 5770 | |
| 5771 | /* |
| 5772 | * Nobody should be changing or seriously looking at |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 5773 | * page->mem_cgroup at this point, we have fully |
Johannes Weiner | 2983331 | 2014-12-10 15:44:02 -0800 | [diff] [blame] | 5774 | * exclusive access to the page. |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5775 | */ |
| 5776 | |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 5777 | if (memcg != page->mem_cgroup) { |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5778 | if (memcg) { |
Johannes Weiner | 18eca2e | 2014-12-10 15:43:57 -0800 | [diff] [blame] | 5779 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
| 5780 | nr_huge, page); |
| 5781 | pgpgout = nr_anon = nr_file = nr_huge = 0; |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5782 | } |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 5783 | memcg = page->mem_cgroup; |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5784 | } |
| 5785 | |
| 5786 | if (PageTransHuge(page)) { |
| 5787 | nr_pages <<= compound_order(page); |
| 5788 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); |
| 5789 | nr_huge += nr_pages; |
| 5790 | } |
| 5791 | |
| 5792 | if (PageAnon(page)) |
| 5793 | nr_anon += nr_pages; |
| 5794 | else |
| 5795 | nr_file += nr_pages; |
| 5796 | |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 5797 | page->mem_cgroup = NULL; |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5798 | |
| 5799 | pgpgout++; |
| 5800 | } while (next != page_list); |
| 5801 | |
| 5802 | if (memcg) |
Johannes Weiner | 18eca2e | 2014-12-10 15:43:57 -0800 | [diff] [blame] | 5803 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
| 5804 | nr_huge, page); |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5805 | } |
| 5806 | |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5807 | /** |
| 5808 | * mem_cgroup_uncharge - uncharge a page |
| 5809 | * @page: page to uncharge |
| 5810 | * |
| 5811 | * Uncharge a page previously charged with mem_cgroup_try_charge() and |
| 5812 | * mem_cgroup_commit_charge(). |
| 5813 | */ |
| 5814 | void mem_cgroup_uncharge(struct page *page) |
| 5815 | { |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5816 | if (mem_cgroup_disabled()) |
| 5817 | return; |
| 5818 | |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5819 | /* Don't touch page->lru of any random page, pre-check: */ |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 5820 | if (!page->mem_cgroup) |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5821 | return; |
| 5822 | |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5823 | INIT_LIST_HEAD(&page->lru); |
| 5824 | uncharge_list(&page->lru); |
| 5825 | } |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5826 | |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5827 | /** |
| 5828 | * mem_cgroup_uncharge_list - uncharge a list of page |
| 5829 | * @page_list: list of pages to uncharge |
| 5830 | * |
| 5831 | * Uncharge a list of pages previously charged with |
| 5832 | * mem_cgroup_try_charge() and mem_cgroup_commit_charge(). |
| 5833 | */ |
| 5834 | void mem_cgroup_uncharge_list(struct list_head *page_list) |
| 5835 | { |
| 5836 | if (mem_cgroup_disabled()) |
| 5837 | return; |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5838 | |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5839 | if (!list_empty(page_list)) |
| 5840 | uncharge_list(page_list); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5841 | } |
| 5842 | |
| 5843 | /** |
| 5844 | * mem_cgroup_migrate - migrate a charge to another page |
| 5845 | * @oldpage: currently charged page |
| 5846 | * @newpage: page to transfer the charge to |
Michal Hocko | f5e03a4 | 2015-02-05 12:25:14 -0800 | [diff] [blame] | 5847 | * @lrucare: either or both pages might be on the LRU already |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5848 | * |
| 5849 | * Migrate the charge from @oldpage to @newpage. |
| 5850 | * |
| 5851 | * Both pages must be locked, @newpage->mapping must be set up. |
| 5852 | */ |
| 5853 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage, |
| 5854 | bool lrucare) |
| 5855 | { |
Johannes Weiner | 2983331 | 2014-12-10 15:44:02 -0800 | [diff] [blame] | 5856 | struct mem_cgroup *memcg; |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5857 | int isolated; |
| 5858 | |
| 5859 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); |
| 5860 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); |
| 5861 | VM_BUG_ON_PAGE(!lrucare && PageLRU(oldpage), oldpage); |
| 5862 | VM_BUG_ON_PAGE(!lrucare && PageLRU(newpage), newpage); |
| 5863 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); |
Johannes Weiner | 6abb5a8 | 2014-08-08 14:19:33 -0700 | [diff] [blame] | 5864 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
| 5865 | newpage); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5866 | |
| 5867 | if (mem_cgroup_disabled()) |
| 5868 | return; |
| 5869 | |
| 5870 | /* Page cache replacement: new page already charged? */ |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 5871 | if (newpage->mem_cgroup) |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5872 | return; |
| 5873 | |
Johannes Weiner | 7d5e324 | 2014-12-10 15:43:46 -0800 | [diff] [blame] | 5874 | /* |
| 5875 | * Swapcache readahead pages can get migrated before being |
| 5876 | * charged, and migration from compaction can happen to an |
| 5877 | * uncharged page when the PFN walker finds a page that |
| 5878 | * reclaim just put back on the LRU but has not released yet. |
| 5879 | */ |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 5880 | memcg = oldpage->mem_cgroup; |
Johannes Weiner | 2983331 | 2014-12-10 15:44:02 -0800 | [diff] [blame] | 5881 | if (!memcg) |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5882 | return; |
| 5883 | |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5884 | if (lrucare) |
| 5885 | lock_page_lru(oldpage, &isolated); |
| 5886 | |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 5887 | oldpage->mem_cgroup = NULL; |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5888 | |
| 5889 | if (lrucare) |
| 5890 | unlock_page_lru(oldpage, isolated); |
| 5891 | |
Johannes Weiner | 2983331 | 2014-12-10 15:44:02 -0800 | [diff] [blame] | 5892 | commit_charge(newpage, memcg, lrucare); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5893 | } |
| 5894 | |
Michal Hocko | 2d11085 | 2013-02-22 16:34:43 -0800 | [diff] [blame] | 5895 | /* |
Michal Hocko | 1081312 | 2013-02-22 16:35:41 -0800 | [diff] [blame] | 5896 | * subsys_initcall() for memory controller. |
| 5897 | * |
| 5898 | * Some parts like hotcpu_notifier() have to be initialized from this context |
| 5899 | * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically |
| 5900 | * everything that doesn't depend on a specific mem_cgroup structure should |
| 5901 | * be initialized from here. |
Michal Hocko | 2d11085 | 2013-02-22 16:34:43 -0800 | [diff] [blame] | 5902 | */ |
| 5903 | static int __init mem_cgroup_init(void) |
| 5904 | { |
Johannes Weiner | 95a045f | 2015-02-11 15:26:33 -0800 | [diff] [blame] | 5905 | int cpu, node; |
| 5906 | |
Michal Hocko | 2d11085 | 2013-02-22 16:34:43 -0800 | [diff] [blame] | 5907 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); |
Johannes Weiner | 95a045f | 2015-02-11 15:26:33 -0800 | [diff] [blame] | 5908 | |
| 5909 | for_each_possible_cpu(cpu) |
| 5910 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, |
| 5911 | drain_local_stock); |
| 5912 | |
| 5913 | for_each_node(node) { |
| 5914 | struct mem_cgroup_tree_per_node *rtpn; |
| 5915 | int zone; |
| 5916 | |
| 5917 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, |
| 5918 | node_online(node) ? node : NUMA_NO_NODE); |
| 5919 | |
| 5920 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
| 5921 | struct mem_cgroup_tree_per_zone *rtpz; |
| 5922 | |
| 5923 | rtpz = &rtpn->rb_tree_per_zone[zone]; |
| 5924 | rtpz->rb_root = RB_ROOT; |
| 5925 | spin_lock_init(&rtpz->lock); |
| 5926 | } |
| 5927 | soft_limit_tree.rb_tree_per_node[node] = rtpn; |
| 5928 | } |
| 5929 | |
Michal Hocko | 2d11085 | 2013-02-22 16:34:43 -0800 | [diff] [blame] | 5930 | return 0; |
| 5931 | } |
| 5932 | subsys_initcall(mem_cgroup_init); |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 5933 | |
| 5934 | #ifdef CONFIG_MEMCG_SWAP |
| 5935 | /** |
| 5936 | * mem_cgroup_swapout - transfer a memsw charge to swap |
| 5937 | * @page: page whose memsw charge to transfer |
| 5938 | * @entry: swap entry to move the charge to |
| 5939 | * |
| 5940 | * Transfer the memsw charge of @page to @entry. |
| 5941 | */ |
| 5942 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) |
| 5943 | { |
| 5944 | struct mem_cgroup *memcg; |
| 5945 | unsigned short oldid; |
| 5946 | |
| 5947 | VM_BUG_ON_PAGE(PageLRU(page), page); |
| 5948 | VM_BUG_ON_PAGE(page_count(page), page); |
| 5949 | |
| 5950 | if (!do_swap_account) |
| 5951 | return; |
| 5952 | |
| 5953 | memcg = page->mem_cgroup; |
| 5954 | |
| 5955 | /* Readahead page, never charged */ |
| 5956 | if (!memcg) |
| 5957 | return; |
| 5958 | |
| 5959 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg)); |
| 5960 | VM_BUG_ON_PAGE(oldid, page); |
| 5961 | mem_cgroup_swap_statistics(memcg, true); |
| 5962 | |
| 5963 | page->mem_cgroup = NULL; |
| 5964 | |
| 5965 | if (!mem_cgroup_is_root(memcg)) |
| 5966 | page_counter_uncharge(&memcg->memory, 1); |
| 5967 | |
Sebastian Andrzej Siewior | ce9ce66 | 2015-09-04 15:47:50 -0700 | [diff] [blame] | 5968 | /* |
| 5969 | * Interrupts should be disabled here because the caller holds the |
| 5970 | * mapping->tree_lock lock which is taken with interrupts-off. It is |
| 5971 | * important here to have the interrupts disabled because it is the |
| 5972 | * only synchronisation we have for udpating the per-CPU variables. |
| 5973 | */ |
| 5974 | VM_BUG_ON(!irqs_disabled()); |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 5975 | mem_cgroup_charge_statistics(memcg, page, -1); |
| 5976 | memcg_check_events(memcg, page); |
| 5977 | } |
| 5978 | |
| 5979 | /** |
| 5980 | * mem_cgroup_uncharge_swap - uncharge a swap entry |
| 5981 | * @entry: swap entry to uncharge |
| 5982 | * |
| 5983 | * Drop the memsw charge associated with @entry. |
| 5984 | */ |
| 5985 | void mem_cgroup_uncharge_swap(swp_entry_t entry) |
| 5986 | { |
| 5987 | struct mem_cgroup *memcg; |
| 5988 | unsigned short id; |
| 5989 | |
| 5990 | if (!do_swap_account) |
| 5991 | return; |
| 5992 | |
| 5993 | id = swap_cgroup_record(entry, 0); |
| 5994 | rcu_read_lock(); |
Vladimir Davydov | adbe427 | 2015-04-15 16:13:00 -0700 | [diff] [blame] | 5995 | memcg = mem_cgroup_from_id(id); |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 5996 | if (memcg) { |
| 5997 | if (!mem_cgroup_is_root(memcg)) |
| 5998 | page_counter_uncharge(&memcg->memsw, 1); |
| 5999 | mem_cgroup_swap_statistics(memcg, false); |
| 6000 | css_put(&memcg->css); |
| 6001 | } |
| 6002 | rcu_read_unlock(); |
| 6003 | } |
| 6004 | |
| 6005 | /* for remember boot option*/ |
| 6006 | #ifdef CONFIG_MEMCG_SWAP_ENABLED |
| 6007 | static int really_do_swap_account __initdata = 1; |
| 6008 | #else |
| 6009 | static int really_do_swap_account __initdata; |
| 6010 | #endif |
| 6011 | |
| 6012 | static int __init enable_swap_account(char *s) |
| 6013 | { |
| 6014 | if (!strcmp(s, "1")) |
| 6015 | really_do_swap_account = 1; |
| 6016 | else if (!strcmp(s, "0")) |
| 6017 | really_do_swap_account = 0; |
| 6018 | return 1; |
| 6019 | } |
| 6020 | __setup("swapaccount=", enable_swap_account); |
| 6021 | |
| 6022 | static struct cftype memsw_cgroup_files[] = { |
| 6023 | { |
| 6024 | .name = "memsw.usage_in_bytes", |
| 6025 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), |
| 6026 | .read_u64 = mem_cgroup_read_u64, |
| 6027 | }, |
| 6028 | { |
| 6029 | .name = "memsw.max_usage_in_bytes", |
| 6030 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), |
| 6031 | .write = mem_cgroup_reset, |
| 6032 | .read_u64 = mem_cgroup_read_u64, |
| 6033 | }, |
| 6034 | { |
| 6035 | .name = "memsw.limit_in_bytes", |
| 6036 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), |
| 6037 | .write = mem_cgroup_write, |
| 6038 | .read_u64 = mem_cgroup_read_u64, |
| 6039 | }, |
| 6040 | { |
| 6041 | .name = "memsw.failcnt", |
| 6042 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), |
| 6043 | .write = mem_cgroup_reset, |
| 6044 | .read_u64 = mem_cgroup_read_u64, |
| 6045 | }, |
| 6046 | { }, /* terminate */ |
| 6047 | }; |
| 6048 | |
| 6049 | static int __init mem_cgroup_swap_init(void) |
| 6050 | { |
| 6051 | if (!mem_cgroup_disabled() && really_do_swap_account) { |
| 6052 | do_swap_account = 1; |
| 6053 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, |
| 6054 | memsw_cgroup_files)); |
| 6055 | } |
| 6056 | return 0; |
| 6057 | } |
| 6058 | subsys_initcall(mem_cgroup_swap_init); |
| 6059 | |
| 6060 | #endif /* CONFIG_MEMCG_SWAP */ |