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 | * |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 17 | * This program is free software; you can redistribute it and/or modify |
| 18 | * it under the terms of the GNU General Public License as published by |
| 19 | * the Free Software Foundation; either version 2 of the License, or |
| 20 | * (at your option) any later version. |
| 21 | * |
| 22 | * This program is distributed in the hope that it will be useful, |
| 23 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 24 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 25 | * GNU General Public License for more details. |
| 26 | */ |
| 27 | |
| 28 | #include <linux/res_counter.h> |
| 29 | #include <linux/memcontrol.h> |
| 30 | #include <linux/cgroup.h> |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 31 | #include <linux/mm.h> |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 32 | #include <linux/hugetlb.h> |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 33 | #include <linux/pagemap.h> |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 34 | #include <linux/smp.h> |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 35 | #include <linux/page-flags.h> |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 36 | #include <linux/backing-dev.h> |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 37 | #include <linux/bit_spinlock.h> |
| 38 | #include <linux/rcupdate.h> |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 39 | #include <linux/limits.h> |
Paul Gortmaker | b9e15ba | 2011-05-26 16:00:52 -0400 | [diff] [blame] | 40 | #include <linux/export.h> |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 41 | #include <linux/mutex.h> |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 42 | #include <linux/rbtree.h> |
Balbir Singh | b6ac57d | 2008-04-29 01:00:19 -0700 | [diff] [blame] | 43 | #include <linux/slab.h> |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 44 | #include <linux/swap.h> |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 45 | #include <linux/swapops.h> |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 46 | #include <linux/spinlock.h> |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 47 | #include <linux/eventfd.h> |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 48 | #include <linux/poll.h> |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 49 | #include <linux/sort.h> |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 50 | #include <linux/fs.h> |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 51 | #include <linux/seq_file.h> |
Anton Vorontsov | 70ddf63 | 2013-04-29 15:08:31 -0700 | [diff] [blame] | 52 | #include <linux/vmpressure.h> |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 53 | #include <linux/mm_inline.h> |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 54 | #include <linux/page_cgroup.h> |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 55 | #include <linux/cpu.h> |
KAMEZAWA Hiroyuki | 158e0a2 | 2010-08-10 18:03:00 -0700 | [diff] [blame] | 56 | #include <linux/oom.h> |
Johannes Weiner | 0056f4e | 2013-10-31 16:34:14 -0700 | [diff] [blame] | 57 | #include <linux/lockdep.h> |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 58 | #include <linux/file.h> |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 59 | #include "internal.h" |
Glauber Costa | d1a4c0b | 2011-12-11 21:47:04 +0000 | [diff] [blame] | 60 | #include <net/sock.h> |
Michal Hocko | 4bd2c1e | 2012-10-08 16:33:10 -0700 | [diff] [blame] | 61 | #include <net/ip.h> |
Glauber Costa | d1a4c0b | 2011-12-11 21:47:04 +0000 | [diff] [blame] | 62 | #include <net/tcp_memcontrol.h> |
Qiang Huang | f35c3a8 | 2013-11-12 15:08:22 -0800 | [diff] [blame] | 63 | #include "slab.h" |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 64 | |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 65 | #include <asm/uaccess.h> |
| 66 | |
KOSAKI Motohiro | cc8e970 | 2010-08-09 17:19:57 -0700 | [diff] [blame] | 67 | #include <trace/events/vmscan.h> |
| 68 | |
Tejun Heo | 073219e | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 69 | struct cgroup_subsys memory_cgrp_subsys __read_mostly; |
| 70 | EXPORT_SYMBOL(memory_cgrp_subsys); |
David Rientjes | 68ae564 | 2012-12-12 13:51:57 -0800 | [diff] [blame] | 71 | |
KAMEZAWA Hiroyuki | a181b0e | 2008-07-25 01:47:08 -0700 | [diff] [blame] | 72 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
Kirill A. Shutemov | 6bbda35 | 2012-05-29 15:06:55 -0700 | [diff] [blame] | 73 | static struct mem_cgroup *root_mem_cgroup __read_mostly; |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 74 | |
Andrew Morton | c255a45 | 2012-07-31 16:43:02 -0700 | [diff] [blame] | 75 | #ifdef CONFIG_MEMCG_SWAP |
Li Zefan | 338c843 | 2009-06-17 16:27:15 -0700 | [diff] [blame] | 76 | /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */ |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 77 | int do_swap_account __read_mostly; |
Michal Hocko | a42c390 | 2010-11-24 12:57:08 -0800 | [diff] [blame] | 78 | |
| 79 | /* for remember boot option*/ |
Andrew Morton | c255a45 | 2012-07-31 16:43:02 -0700 | [diff] [blame] | 80 | #ifdef CONFIG_MEMCG_SWAP_ENABLED |
Michal Hocko | a42c390 | 2010-11-24 12:57:08 -0800 | [diff] [blame] | 81 | static int really_do_swap_account __initdata = 1; |
| 82 | #else |
Fabian Frederick | ada4ba5 | 2014-06-04 16:08:08 -0700 | [diff] [blame] | 83 | static int really_do_swap_account __initdata; |
Michal Hocko | a42c390 | 2010-11-24 12:57:08 -0800 | [diff] [blame] | 84 | #endif |
| 85 | |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 86 | #else |
Kirill A. Shutemov | a0db00f | 2012-05-29 15:06:56 -0700 | [diff] [blame] | 87 | #define do_swap_account 0 |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 88 | #endif |
| 89 | |
| 90 | |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 91 | static const char * const mem_cgroup_stat_names[] = { |
| 92 | "cache", |
| 93 | "rss", |
David Rientjes | b070e65 | 2013-05-07 16:18:09 -0700 | [diff] [blame] | 94 | "rss_huge", |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 95 | "mapped_file", |
Sha Zhengju | 3ea67d0 | 2013-09-12 15:13:53 -0700 | [diff] [blame] | 96 | "writeback", |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 97 | "swap", |
| 98 | }; |
| 99 | |
Johannes Weiner | e9f8974 | 2011-03-23 16:42:37 -0700 | [diff] [blame] | 100 | enum mem_cgroup_events_index { |
| 101 | MEM_CGROUP_EVENTS_PGPGIN, /* # of pages paged in */ |
| 102 | MEM_CGROUP_EVENTS_PGPGOUT, /* # of pages paged out */ |
Ying Han | 456f998 | 2011-05-26 16:25:38 -0700 | [diff] [blame] | 103 | MEM_CGROUP_EVENTS_PGFAULT, /* # of page-faults */ |
| 104 | MEM_CGROUP_EVENTS_PGMAJFAULT, /* # of major page-faults */ |
Johannes Weiner | e9f8974 | 2011-03-23 16:42:37 -0700 | [diff] [blame] | 105 | MEM_CGROUP_EVENTS_NSTATS, |
| 106 | }; |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 107 | |
| 108 | static const char * const mem_cgroup_events_names[] = { |
| 109 | "pgpgin", |
| 110 | "pgpgout", |
| 111 | "pgfault", |
| 112 | "pgmajfault", |
| 113 | }; |
| 114 | |
Sha Zhengju | 58cf188 | 2013-02-22 16:32:05 -0800 | [diff] [blame] | 115 | static const char * const mem_cgroup_lru_names[] = { |
| 116 | "inactive_anon", |
| 117 | "active_anon", |
| 118 | "inactive_file", |
| 119 | "active_file", |
| 120 | "unevictable", |
| 121 | }; |
| 122 | |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 123 | /* |
| 124 | * Per memcg event counter is incremented at every pagein/pageout. With THP, |
| 125 | * it will be incremated by the number of pages. This counter is used for |
| 126 | * for trigger some periodic events. This is straightforward and better |
| 127 | * than using jiffies etc. to handle periodic memcg event. |
| 128 | */ |
| 129 | enum mem_cgroup_events_target { |
| 130 | MEM_CGROUP_TARGET_THRESH, |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 131 | MEM_CGROUP_TARGET_SOFTLIMIT, |
KAMEZAWA Hiroyuki | 453a9bf | 2011-07-08 15:39:43 -0700 | [diff] [blame] | 132 | MEM_CGROUP_TARGET_NUMAINFO, |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 133 | MEM_CGROUP_NTARGETS, |
| 134 | }; |
Kirill A. Shutemov | a0db00f | 2012-05-29 15:06:56 -0700 | [diff] [blame] | 135 | #define THRESHOLDS_EVENTS_TARGET 128 |
| 136 | #define SOFTLIMIT_EVENTS_TARGET 1024 |
| 137 | #define NUMAINFO_EVENTS_TARGET 1024 |
Johannes Weiner | e9f8974 | 2011-03-23 16:42:37 -0700 | [diff] [blame] | 138 | |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 139 | struct mem_cgroup_stat_cpu { |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 140 | long count[MEM_CGROUP_STAT_NSTATS]; |
Johannes Weiner | e9f8974 | 2011-03-23 16:42:37 -0700 | [diff] [blame] | 141 | unsigned long events[MEM_CGROUP_EVENTS_NSTATS]; |
Johannes Weiner | 1311471 | 2012-05-29 15:07:07 -0700 | [diff] [blame] | 142 | unsigned long nr_page_events; |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 143 | unsigned long targets[MEM_CGROUP_NTARGETS]; |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 144 | }; |
| 145 | |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 146 | struct mem_cgroup_reclaim_iter { |
Michal Hocko | 5f57816 | 2013-04-29 15:07:17 -0700 | [diff] [blame] | 147 | /* |
| 148 | * last scanned hierarchy member. Valid only if last_dead_count |
| 149 | * matches memcg->dead_count of the hierarchy root group. |
| 150 | */ |
Michal Hocko | 542f85f | 2013-04-29 15:07:15 -0700 | [diff] [blame] | 151 | struct mem_cgroup *last_visited; |
Hugh Dickins | d2ab70a | 2014-01-23 15:53:30 -0800 | [diff] [blame] | 152 | int last_dead_count; |
Michal Hocko | 5f57816 | 2013-04-29 15:07:17 -0700 | [diff] [blame] | 153 | |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 154 | /* scan generation, increased every round-trip */ |
| 155 | unsigned int generation; |
| 156 | }; |
| 157 | |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 158 | /* |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 159 | * per-zone information in memory controller. |
| 160 | */ |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 161 | struct mem_cgroup_per_zone { |
Johannes Weiner | 6290df5 | 2012-01-12 17:18:10 -0800 | [diff] [blame] | 162 | struct lruvec lruvec; |
Hugh Dickins | 1eb4927 | 2012-03-21 16:34:19 -0700 | [diff] [blame] | 163 | unsigned long lru_size[NR_LRU_LISTS]; |
KOSAKI Motohiro | 3e2f41f | 2009-01-07 18:08:20 -0800 | [diff] [blame] | 164 | |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 165 | struct mem_cgroup_reclaim_iter reclaim_iter[DEF_PRIORITY + 1]; |
| 166 | |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 167 | struct rb_node tree_node; /* RB tree node */ |
| 168 | unsigned long long usage_in_excess;/* Set to the value by which */ |
| 169 | /* the soft limit is exceeded*/ |
| 170 | bool on_tree; |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 171 | struct mem_cgroup *memcg; /* Back pointer, we cannot */ |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 172 | /* use container_of */ |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 173 | }; |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 174 | |
| 175 | struct mem_cgroup_per_node { |
| 176 | struct mem_cgroup_per_zone zoneinfo[MAX_NR_ZONES]; |
| 177 | }; |
| 178 | |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 179 | /* |
| 180 | * Cgroups above their limits are maintained in a RB-Tree, independent of |
| 181 | * their hierarchy representation |
| 182 | */ |
| 183 | |
| 184 | struct mem_cgroup_tree_per_zone { |
| 185 | struct rb_root rb_root; |
| 186 | spinlock_t lock; |
| 187 | }; |
| 188 | |
| 189 | struct mem_cgroup_tree_per_node { |
| 190 | struct mem_cgroup_tree_per_zone rb_tree_per_zone[MAX_NR_ZONES]; |
| 191 | }; |
| 192 | |
| 193 | struct mem_cgroup_tree { |
| 194 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; |
| 195 | }; |
| 196 | |
| 197 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; |
| 198 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 199 | struct mem_cgroup_threshold { |
| 200 | struct eventfd_ctx *eventfd; |
| 201 | u64 threshold; |
| 202 | }; |
| 203 | |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 204 | /* For threshold */ |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 205 | struct mem_cgroup_threshold_ary { |
Sha Zhengju | 748dad3 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 206 | /* An array index points to threshold just below or equal to usage. */ |
Phil Carmody | 5407a56 | 2010-05-26 14:42:42 -0700 | [diff] [blame] | 207 | int current_threshold; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 208 | /* Size of entries[] */ |
| 209 | unsigned int size; |
| 210 | /* Array of thresholds */ |
| 211 | struct mem_cgroup_threshold entries[0]; |
| 212 | }; |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 213 | |
| 214 | struct mem_cgroup_thresholds { |
| 215 | /* Primary thresholds array */ |
| 216 | struct mem_cgroup_threshold_ary *primary; |
| 217 | /* |
| 218 | * Spare threshold array. |
| 219 | * This is needed to make mem_cgroup_unregister_event() "never fail". |
| 220 | * It must be able to store at least primary->size - 1 entries. |
| 221 | */ |
| 222 | struct mem_cgroup_threshold_ary *spare; |
| 223 | }; |
| 224 | |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 225 | /* for OOM */ |
| 226 | struct mem_cgroup_eventfd_list { |
| 227 | struct list_head list; |
| 228 | struct eventfd_ctx *eventfd; |
| 229 | }; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 230 | |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 231 | /* |
| 232 | * cgroup_event represents events which userspace want to receive. |
| 233 | */ |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 234 | struct mem_cgroup_event { |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 235 | /* |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 236 | * memcg which the event belongs to. |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 237 | */ |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 238 | struct mem_cgroup *memcg; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 239 | /* |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 240 | * eventfd to signal userspace about the event. |
| 241 | */ |
| 242 | struct eventfd_ctx *eventfd; |
| 243 | /* |
| 244 | * Each of these stored in a list by the cgroup. |
| 245 | */ |
| 246 | struct list_head list; |
| 247 | /* |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 248 | * register_event() callback will be used to add new userspace |
| 249 | * waiter for changes related to this event. Use eventfd_signal() |
| 250 | * on eventfd to send notification to userspace. |
| 251 | */ |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 252 | int (*register_event)(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 253 | struct eventfd_ctx *eventfd, const char *args); |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 254 | /* |
| 255 | * unregister_event() callback will be called when userspace closes |
| 256 | * the eventfd or on cgroup removing. This callback must be set, |
| 257 | * if you want provide notification functionality. |
| 258 | */ |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 259 | void (*unregister_event)(struct mem_cgroup *memcg, |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 260 | struct eventfd_ctx *eventfd); |
| 261 | /* |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 262 | * All fields below needed to unregister event when |
| 263 | * userspace closes eventfd. |
| 264 | */ |
| 265 | poll_table pt; |
| 266 | wait_queue_head_t *wqh; |
| 267 | wait_queue_t wait; |
| 268 | struct work_struct remove; |
| 269 | }; |
| 270 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 271 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
| 272 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 273 | |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 274 | /* |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 275 | * The memory controller data structure. The memory controller controls both |
| 276 | * page cache and RSS per cgroup. We would eventually like to provide |
| 277 | * statistics based on the statistics developed by Rik Van Riel for clock-pro, |
| 278 | * to help the administrator determine what knobs to tune. |
| 279 | * |
| 280 | * TODO: Add a water mark for the memory controller. Reclaim will begin when |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 281 | * we hit the water mark. May be even add a low water mark, such that |
| 282 | * no reclaim occurs from a cgroup at it's low water mark, this is |
| 283 | * a feature that will be implemented much later in the future. |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 284 | */ |
| 285 | struct mem_cgroup { |
| 286 | struct cgroup_subsys_state css; |
| 287 | /* |
| 288 | * the counter to account for memory usage |
| 289 | */ |
| 290 | struct res_counter res; |
Hugh Dickins | 59927fb | 2012-03-15 15:17:07 -0700 | [diff] [blame] | 291 | |
Anton Vorontsov | 70ddf63 | 2013-04-29 15:08:31 -0700 | [diff] [blame] | 292 | /* vmpressure notifications */ |
| 293 | struct vmpressure vmpressure; |
| 294 | |
Li Zefan | 465939a | 2013-07-08 16:00:38 -0700 | [diff] [blame] | 295 | /* |
| 296 | * the counter to account for mem+swap usage. |
| 297 | */ |
| 298 | struct res_counter memsw; |
Hugh Dickins | 59927fb | 2012-03-15 15:17:07 -0700 | [diff] [blame] | 299 | |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 300 | /* |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 301 | * the counter to account for kernel memory usage. |
| 302 | */ |
| 303 | struct res_counter kmem; |
| 304 | /* |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 305 | * Should the accounting and control be hierarchical, per subtree? |
| 306 | */ |
| 307 | bool use_hierarchy; |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 308 | unsigned long kmem_account_flags; /* See KMEM_ACCOUNTED_*, below */ |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 309 | |
| 310 | bool oom_lock; |
| 311 | atomic_t under_oom; |
Johannes Weiner | 3812c8c | 2013-09-12 15:13:44 -0700 | [diff] [blame] | 312 | atomic_t oom_wakeups; |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 313 | |
KAMEZAWA Hiroyuki | 1f4c025 | 2011-07-26 16:08:21 -0700 | [diff] [blame] | 314 | int swappiness; |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 315 | /* OOM-Killer disable */ |
| 316 | int oom_kill_disable; |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 317 | |
KAMEZAWA Hiroyuki | 22a668d | 2009-06-17 16:27:19 -0700 | [diff] [blame] | 318 | /* set when res.limit == memsw.limit */ |
| 319 | bool memsw_is_minimum; |
| 320 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 321 | /* protect arrays of thresholds */ |
| 322 | struct mutex thresholds_lock; |
| 323 | |
| 324 | /* thresholds for memory usage. RCU-protected */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 325 | struct mem_cgroup_thresholds thresholds; |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 326 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 327 | /* thresholds for mem+swap usage. RCU-protected */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 328 | struct mem_cgroup_thresholds memsw_thresholds; |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 329 | |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 330 | /* For oom notifier event fd */ |
| 331 | struct list_head oom_notify; |
Johannes Weiner | 185efc0 | 2011-09-14 16:21:58 -0700 | [diff] [blame] | 332 | |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 333 | /* |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 334 | * Should we move charges of a task when a task is moved into this |
| 335 | * mem_cgroup ? And what type of charges should we move ? |
| 336 | */ |
Andrew Morton | f894ffa | 2013-09-12 15:13:35 -0700 | [diff] [blame] | 337 | unsigned long move_charge_at_immigrate; |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 338 | /* |
KAMEZAWA Hiroyuki | 619d094 | 2012-03-21 16:34:23 -0700 | [diff] [blame] | 339 | * set > 0 if pages under this cgroup are moving to other cgroup. |
| 340 | */ |
| 341 | atomic_t moving_account; |
KAMEZAWA Hiroyuki | 312734c0 | 2012-03-21 16:34:24 -0700 | [diff] [blame] | 342 | /* taken only while moving_account > 0 */ |
| 343 | spinlock_t move_lock; |
KAMEZAWA Hiroyuki | 619d094 | 2012-03-21 16:34:23 -0700 | [diff] [blame] | 344 | /* |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 345 | * percpu counter. |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 346 | */ |
Kirill A. Shutemov | 3a7951b | 2012-05-29 15:06:56 -0700 | [diff] [blame] | 347 | struct mem_cgroup_stat_cpu __percpu *stat; |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 348 | /* |
| 349 | * used when a cpu is offlined or other synchronizations |
| 350 | * See mem_cgroup_read_stat(). |
| 351 | */ |
| 352 | struct mem_cgroup_stat_cpu nocpu_base; |
| 353 | spinlock_t pcp_counter_lock; |
Glauber Costa | d1a4c0b | 2011-12-11 21:47:04 +0000 | [diff] [blame] | 354 | |
Michal Hocko | 5f57816 | 2013-04-29 15:07:17 -0700 | [diff] [blame] | 355 | atomic_t dead_count; |
Michal Hocko | 4bd2c1e | 2012-10-08 16:33:10 -0700 | [diff] [blame] | 356 | #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_INET) |
Eric W. Biederman | 2e685ca | 2013-10-19 16:26:19 -0700 | [diff] [blame] | 357 | struct cg_proto tcp_mem; |
Glauber Costa | d1a4c0b | 2011-12-11 21:47:04 +0000 | [diff] [blame] | 358 | #endif |
Glauber Costa | 2633d7a | 2012-12-18 14:22:34 -0800 | [diff] [blame] | 359 | #if defined(CONFIG_MEMCG_KMEM) |
Vladimir Davydov | bd67314 | 2014-06-04 16:07:40 -0700 | [diff] [blame] | 360 | /* analogous to slab_common's slab_caches list, but per-memcg; |
| 361 | * protected by memcg_slab_mutex */ |
Glauber Costa | 2633d7a | 2012-12-18 14:22:34 -0800 | [diff] [blame] | 362 | struct list_head memcg_slab_caches; |
Glauber Costa | 2633d7a | 2012-12-18 14:22:34 -0800 | [diff] [blame] | 363 | /* Index in the kmem_cache->memcg_params->memcg_caches array */ |
| 364 | int kmemcg_id; |
| 365 | #endif |
Glauber Costa | 45cf7eb | 2013-02-22 16:34:49 -0800 | [diff] [blame] | 366 | |
| 367 | int last_scanned_node; |
| 368 | #if MAX_NUMNODES > 1 |
| 369 | nodemask_t scan_nodes; |
| 370 | atomic_t numainfo_events; |
| 371 | atomic_t numainfo_updating; |
| 372 | #endif |
Anton Vorontsov | 70ddf63 | 2013-04-29 15:08:31 -0700 | [diff] [blame] | 373 | |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 374 | /* List of events which userspace want to receive */ |
| 375 | struct list_head event_list; |
| 376 | spinlock_t event_list_lock; |
| 377 | |
Johannes Weiner | 54f72fe | 2013-07-08 15:59:49 -0700 | [diff] [blame] | 378 | struct mem_cgroup_per_node *nodeinfo[0]; |
| 379 | /* WARNING: nodeinfo must be the last member here */ |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 380 | }; |
| 381 | |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 382 | /* internal only representation about the status of kmem accounting. */ |
| 383 | enum { |
Vladimir Davydov | 6de64be | 2014-01-23 15:53:08 -0800 | [diff] [blame] | 384 | KMEM_ACCOUNTED_ACTIVE, /* accounted by this cgroup itself */ |
Glauber Costa | 7de3768 | 2012-12-18 14:22:07 -0800 | [diff] [blame] | 385 | KMEM_ACCOUNTED_DEAD, /* dead memcg with pending kmem charges */ |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 386 | }; |
| 387 | |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 388 | #ifdef CONFIG_MEMCG_KMEM |
| 389 | static inline void memcg_kmem_set_active(struct mem_cgroup *memcg) |
| 390 | { |
| 391 | set_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags); |
| 392 | } |
Glauber Costa | 7de3768 | 2012-12-18 14:22:07 -0800 | [diff] [blame] | 393 | |
| 394 | static bool memcg_kmem_is_active(struct mem_cgroup *memcg) |
| 395 | { |
| 396 | return test_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags); |
| 397 | } |
| 398 | |
| 399 | static void memcg_kmem_mark_dead(struct mem_cgroup *memcg) |
| 400 | { |
Li Zefan | 10d5ebf | 2013-07-08 16:00:33 -0700 | [diff] [blame] | 401 | /* |
| 402 | * Our caller must use css_get() first, because memcg_uncharge_kmem() |
| 403 | * will call css_put() if it sees the memcg is dead. |
| 404 | */ |
| 405 | smp_wmb(); |
Glauber Costa | 7de3768 | 2012-12-18 14:22:07 -0800 | [diff] [blame] | 406 | if (test_bit(KMEM_ACCOUNTED_ACTIVE, &memcg->kmem_account_flags)) |
| 407 | set_bit(KMEM_ACCOUNTED_DEAD, &memcg->kmem_account_flags); |
| 408 | } |
| 409 | |
| 410 | static bool memcg_kmem_test_and_clear_dead(struct mem_cgroup *memcg) |
| 411 | { |
| 412 | return test_and_clear_bit(KMEM_ACCOUNTED_DEAD, |
| 413 | &memcg->kmem_account_flags); |
| 414 | } |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 415 | #endif |
| 416 | |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 417 | /* Stuffs for move charges at task migration. */ |
| 418 | /* |
Glauber Costa | ee5e847 | 2013-02-22 16:34:50 -0800 | [diff] [blame] | 419 | * Types of charges to be moved. "move_charge_at_immitgrate" and |
| 420 | * "immigrate_flags" are treated as a left-shifted bitmap of these types. |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 421 | */ |
| 422 | enum move_type { |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 423 | MOVE_CHARGE_TYPE_ANON, /* private anonymous page and swap of it */ |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 424 | MOVE_CHARGE_TYPE_FILE, /* file page(including tmpfs) and swap of it */ |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 425 | NR_MOVE_TYPE, |
| 426 | }; |
| 427 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 428 | /* "mc" and its members are protected by cgroup_mutex */ |
| 429 | static struct move_charge_struct { |
Daisuke Nishimura | b1dd693 | 2010-11-24 12:57:06 -0800 | [diff] [blame] | 430 | spinlock_t lock; /* for from, to */ |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 431 | struct mem_cgroup *from; |
| 432 | struct mem_cgroup *to; |
Glauber Costa | ee5e847 | 2013-02-22 16:34:50 -0800 | [diff] [blame] | 433 | unsigned long immigrate_flags; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 434 | unsigned long precharge; |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 435 | unsigned long moved_charge; |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 436 | unsigned long moved_swap; |
Daisuke Nishimura | 8033b97 | 2010-03-10 15:22:16 -0800 | [diff] [blame] | 437 | struct task_struct *moving_task; /* a task moving charges */ |
| 438 | wait_queue_head_t waitq; /* a waitq for other context */ |
| 439 | } mc = { |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 440 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
Daisuke Nishimura | 8033b97 | 2010-03-10 15:22:16 -0800 | [diff] [blame] | 441 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
| 442 | }; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 443 | |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 444 | static bool move_anon(void) |
| 445 | { |
Glauber Costa | ee5e847 | 2013-02-22 16:34:50 -0800 | [diff] [blame] | 446 | return test_bit(MOVE_CHARGE_TYPE_ANON, &mc.immigrate_flags); |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 447 | } |
| 448 | |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 449 | static bool move_file(void) |
| 450 | { |
Glauber Costa | ee5e847 | 2013-02-22 16:34:50 -0800 | [diff] [blame] | 451 | return test_bit(MOVE_CHARGE_TYPE_FILE, &mc.immigrate_flags); |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 452 | } |
| 453 | |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 454 | /* |
| 455 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft |
| 456 | * limit reclaim to prevent infinite loops, if they ever occur. |
| 457 | */ |
Kirill A. Shutemov | a0db00f | 2012-05-29 15:06:56 -0700 | [diff] [blame] | 458 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 459 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 460 | |
KAMEZAWA Hiroyuki | 217bc31 | 2008-02-07 00:14:17 -0800 | [diff] [blame] | 461 | enum charge_type { |
| 462 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, |
Kamezawa Hiroyuki | 41326c1 | 2012-07-31 16:41:40 -0700 | [diff] [blame] | 463 | MEM_CGROUP_CHARGE_TYPE_ANON, |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 464 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
KAMEZAWA Hiroyuki | 8a9478c | 2009-06-17 16:27:17 -0700 | [diff] [blame] | 465 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
KAMEZAWA Hiroyuki | c05555b | 2008-10-18 20:28:11 -0700 | [diff] [blame] | 466 | NR_CHARGE_TYPE, |
| 467 | }; |
| 468 | |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 469 | /* for encoding cft->private value on file */ |
Glauber Costa | 86ae53e | 2012-12-18 14:21:45 -0800 | [diff] [blame] | 470 | enum res_type { |
| 471 | _MEM, |
| 472 | _MEMSWAP, |
| 473 | _OOM_TYPE, |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 474 | _KMEM, |
Glauber Costa | 86ae53e | 2012-12-18 14:21:45 -0800 | [diff] [blame] | 475 | }; |
| 476 | |
Kirill A. Shutemov | a0db00f | 2012-05-29 15:06:56 -0700 | [diff] [blame] | 477 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
| 478 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 479 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 480 | /* Used for OOM nofiier */ |
| 481 | #define OOM_CONTROL (0) |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 482 | |
Balbir Singh | 75822b4 | 2009-09-23 15:56:38 -0700 | [diff] [blame] | 483 | /* |
| 484 | * Reclaim flags for mem_cgroup_hierarchical_reclaim |
| 485 | */ |
| 486 | #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0 |
| 487 | #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT) |
| 488 | #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1 |
| 489 | #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT) |
| 490 | |
Glauber Costa | 0999821 | 2013-02-22 16:34:55 -0800 | [diff] [blame] | 491 | /* |
| 492 | * The memcg_create_mutex will be held whenever a new cgroup is created. |
| 493 | * As a consequence, any change that needs to protect against new child cgroups |
| 494 | * appearing has to hold it as well. |
| 495 | */ |
| 496 | static DEFINE_MUTEX(memcg_create_mutex); |
| 497 | |
Wanpeng Li | b214514 | 2012-07-31 16:46:01 -0700 | [diff] [blame] | 498 | struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *s) |
| 499 | { |
Tejun Heo | a7c6d55 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 500 | return s ? container_of(s, struct mem_cgroup, css) : NULL; |
Wanpeng Li | b214514 | 2012-07-31 16:46:01 -0700 | [diff] [blame] | 501 | } |
| 502 | |
Anton Vorontsov | 70ddf63 | 2013-04-29 15:08:31 -0700 | [diff] [blame] | 503 | /* Some nice accessors for the vmpressure. */ |
| 504 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) |
| 505 | { |
| 506 | if (!memcg) |
| 507 | memcg = root_mem_cgroup; |
| 508 | return &memcg->vmpressure; |
| 509 | } |
| 510 | |
| 511 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) |
| 512 | { |
| 513 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; |
| 514 | } |
| 515 | |
Michal Hocko | 7ffc0ed | 2012-10-08 16:33:13 -0700 | [diff] [blame] | 516 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
| 517 | { |
| 518 | return (memcg == root_mem_cgroup); |
| 519 | } |
| 520 | |
Li Zefan | 4219b2d | 2013-09-23 16:56:29 +0800 | [diff] [blame] | 521 | /* |
| 522 | * We restrict the id in the range of [1, 65535], so it can fit into |
| 523 | * an unsigned short. |
| 524 | */ |
| 525 | #define MEM_CGROUP_ID_MAX USHRT_MAX |
| 526 | |
Li Zefan | 34c00c3 | 2013-09-23 16:56:01 +0800 | [diff] [blame] | 527 | static inline unsigned short mem_cgroup_id(struct mem_cgroup *memcg) |
| 528 | { |
| 529 | /* |
| 530 | * The ID of the root cgroup is 0, but memcg treat 0 as an |
| 531 | * invalid ID, so we return (cgroup_id + 1). |
| 532 | */ |
| 533 | return memcg->css.cgroup->id + 1; |
| 534 | } |
| 535 | |
| 536 | static inline struct mem_cgroup *mem_cgroup_from_id(unsigned short id) |
| 537 | { |
| 538 | struct cgroup_subsys_state *css; |
| 539 | |
Tejun Heo | 073219e | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 540 | css = css_from_id(id - 1, &memory_cgrp_subsys); |
Li Zefan | 34c00c3 | 2013-09-23 16:56:01 +0800 | [diff] [blame] | 541 | return mem_cgroup_from_css(css); |
| 542 | } |
| 543 | |
Glauber Costa | e1aab16 | 2011-12-11 21:47:03 +0000 | [diff] [blame] | 544 | /* Writing them here to avoid exposing memcg's inner layout */ |
Michal Hocko | 4bd2c1e | 2012-10-08 16:33:10 -0700 | [diff] [blame] | 545 | #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM) |
Glauber Costa | e1aab16 | 2011-12-11 21:47:03 +0000 | [diff] [blame] | 546 | |
Glauber Costa | e1aab16 | 2011-12-11 21:47:03 +0000 | [diff] [blame] | 547 | void sock_update_memcg(struct sock *sk) |
| 548 | { |
Glauber Costa | 376be5f | 2012-01-20 04:57:14 +0000 | [diff] [blame] | 549 | if (mem_cgroup_sockets_enabled) { |
Glauber Costa | e1aab16 | 2011-12-11 21:47:03 +0000 | [diff] [blame] | 550 | struct mem_cgroup *memcg; |
Glauber Costa | 3f13461 | 2012-05-29 15:07:11 -0700 | [diff] [blame] | 551 | struct cg_proto *cg_proto; |
Glauber Costa | e1aab16 | 2011-12-11 21:47:03 +0000 | [diff] [blame] | 552 | |
| 553 | BUG_ON(!sk->sk_prot->proto_cgroup); |
| 554 | |
Glauber Costa | f3f511e | 2012-01-05 20:16:39 +0000 | [diff] [blame] | 555 | /* Socket cloning can throw us here with sk_cgrp already |
| 556 | * filled. It won't however, necessarily happen from |
| 557 | * process context. So the test for root memcg given |
| 558 | * the current task's memcg won't help us in this case. |
| 559 | * |
| 560 | * Respecting the original socket's memcg is a better |
| 561 | * decision in this case. |
| 562 | */ |
| 563 | if (sk->sk_cgrp) { |
| 564 | BUG_ON(mem_cgroup_is_root(sk->sk_cgrp->memcg)); |
Li Zefan | 5347e5a | 2013-07-08 16:00:30 -0700 | [diff] [blame] | 565 | css_get(&sk->sk_cgrp->memcg->css); |
Glauber Costa | f3f511e | 2012-01-05 20:16:39 +0000 | [diff] [blame] | 566 | return; |
| 567 | } |
| 568 | |
Glauber Costa | e1aab16 | 2011-12-11 21:47:03 +0000 | [diff] [blame] | 569 | rcu_read_lock(); |
| 570 | memcg = mem_cgroup_from_task(current); |
Glauber Costa | 3f13461 | 2012-05-29 15:07:11 -0700 | [diff] [blame] | 571 | cg_proto = sk->sk_prot->proto_cgroup(memcg); |
Li Zefan | 5347e5a | 2013-07-08 16:00:30 -0700 | [diff] [blame] | 572 | if (!mem_cgroup_is_root(memcg) && |
| 573 | memcg_proto_active(cg_proto) && css_tryget(&memcg->css)) { |
Glauber Costa | 3f13461 | 2012-05-29 15:07:11 -0700 | [diff] [blame] | 574 | sk->sk_cgrp = cg_proto; |
Glauber Costa | e1aab16 | 2011-12-11 21:47:03 +0000 | [diff] [blame] | 575 | } |
| 576 | rcu_read_unlock(); |
| 577 | } |
| 578 | } |
| 579 | EXPORT_SYMBOL(sock_update_memcg); |
| 580 | |
| 581 | void sock_release_memcg(struct sock *sk) |
| 582 | { |
Glauber Costa | 376be5f | 2012-01-20 04:57:14 +0000 | [diff] [blame] | 583 | if (mem_cgroup_sockets_enabled && sk->sk_cgrp) { |
Glauber Costa | e1aab16 | 2011-12-11 21:47:03 +0000 | [diff] [blame] | 584 | struct mem_cgroup *memcg; |
| 585 | WARN_ON(!sk->sk_cgrp->memcg); |
| 586 | memcg = sk->sk_cgrp->memcg; |
Li Zefan | 5347e5a | 2013-07-08 16:00:30 -0700 | [diff] [blame] | 587 | css_put(&sk->sk_cgrp->memcg->css); |
Glauber Costa | e1aab16 | 2011-12-11 21:47:03 +0000 | [diff] [blame] | 588 | } |
| 589 | } |
Glauber Costa | d1a4c0b | 2011-12-11 21:47:04 +0000 | [diff] [blame] | 590 | |
| 591 | struct cg_proto *tcp_proto_cgroup(struct mem_cgroup *memcg) |
| 592 | { |
| 593 | if (!memcg || mem_cgroup_is_root(memcg)) |
| 594 | return NULL; |
| 595 | |
Eric W. Biederman | 2e685ca | 2013-10-19 16:26:19 -0700 | [diff] [blame] | 596 | return &memcg->tcp_mem; |
Glauber Costa | d1a4c0b | 2011-12-11 21:47:04 +0000 | [diff] [blame] | 597 | } |
| 598 | EXPORT_SYMBOL(tcp_proto_cgroup); |
Glauber Costa | e1aab16 | 2011-12-11 21:47:03 +0000 | [diff] [blame] | 599 | |
Glauber Costa | 3f13461 | 2012-05-29 15:07:11 -0700 | [diff] [blame] | 600 | static void disarm_sock_keys(struct mem_cgroup *memcg) |
| 601 | { |
Eric W. Biederman | 2e685ca | 2013-10-19 16:26:19 -0700 | [diff] [blame] | 602 | if (!memcg_proto_activated(&memcg->tcp_mem)) |
Glauber Costa | 3f13461 | 2012-05-29 15:07:11 -0700 | [diff] [blame] | 603 | return; |
| 604 | static_key_slow_dec(&memcg_socket_limit_enabled); |
| 605 | } |
| 606 | #else |
| 607 | static void disarm_sock_keys(struct mem_cgroup *memcg) |
| 608 | { |
| 609 | } |
| 610 | #endif |
| 611 | |
Glauber Costa | a8964b9 | 2012-12-18 14:22:09 -0800 | [diff] [blame] | 612 | #ifdef CONFIG_MEMCG_KMEM |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 613 | /* |
| 614 | * 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] | 615 | * The main reason for not using cgroup id for this: |
| 616 | * this works better in sparse environments, where we have a lot of memcgs, |
| 617 | * but only a few kmem-limited. Or also, if we have, for instance, 200 |
| 618 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a |
| 619 | * 200 entry array for that. |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 620 | * |
| 621 | * The current size of the caches array is stored in |
| 622 | * memcg_limited_groups_array_size. It will double each time we have to |
| 623 | * increase it. |
| 624 | */ |
| 625 | static DEFINE_IDA(kmem_limited_groups); |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 626 | int memcg_limited_groups_array_size; |
| 627 | |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 628 | /* |
| 629 | * MIN_SIZE is different than 1, because we would like to avoid going through |
| 630 | * the alloc/free process all the time. In a small machine, 4 kmem-limited |
| 631 | * cgroups is a reasonable guess. In the future, it could be a parameter or |
| 632 | * tunable, but that is strictly not necessary. |
| 633 | * |
Li Zefan | b862783 | 2013-09-23 16:56:47 +0800 | [diff] [blame] | 634 | * 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] | 635 | * this constant directly from cgroup, but it is understandable that this is |
| 636 | * 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] | 637 | * 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] | 638 | * increase ours as well if it increases. |
| 639 | */ |
| 640 | #define MEMCG_CACHES_MIN_SIZE 4 |
Li Zefan | b862783 | 2013-09-23 16:56:47 +0800 | [diff] [blame] | 641 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 642 | |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 643 | /* |
| 644 | * A lot of the calls to the cache allocation functions are expected to be |
| 645 | * inlined by the compiler. Since the calls to memcg_kmem_get_cache are |
| 646 | * conditional to this static branch, we'll have to allow modules that does |
| 647 | * kmem_cache_alloc and the such to see this symbol as well |
| 648 | */ |
Glauber Costa | a8964b9 | 2012-12-18 14:22:09 -0800 | [diff] [blame] | 649 | struct static_key memcg_kmem_enabled_key; |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 650 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
Glauber Costa | a8964b9 | 2012-12-18 14:22:09 -0800 | [diff] [blame] | 651 | |
| 652 | static void disarm_kmem_keys(struct mem_cgroup *memcg) |
| 653 | { |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 654 | if (memcg_kmem_is_active(memcg)) { |
Glauber Costa | a8964b9 | 2012-12-18 14:22:09 -0800 | [diff] [blame] | 655 | static_key_slow_dec(&memcg_kmem_enabled_key); |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 656 | ida_simple_remove(&kmem_limited_groups, memcg->kmemcg_id); |
| 657 | } |
Glauber Costa | bea207c | 2012-12-18 14:22:11 -0800 | [diff] [blame] | 658 | /* |
| 659 | * This check can't live in kmem destruction function, |
| 660 | * since the charges will outlive the cgroup |
| 661 | */ |
| 662 | WARN_ON(res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0); |
Glauber Costa | a8964b9 | 2012-12-18 14:22:09 -0800 | [diff] [blame] | 663 | } |
| 664 | #else |
| 665 | static void disarm_kmem_keys(struct mem_cgroup *memcg) |
| 666 | { |
| 667 | } |
| 668 | #endif /* CONFIG_MEMCG_KMEM */ |
| 669 | |
| 670 | static void disarm_static_keys(struct mem_cgroup *memcg) |
| 671 | { |
| 672 | disarm_sock_keys(memcg); |
| 673 | disarm_kmem_keys(memcg); |
| 674 | } |
| 675 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 676 | static void drain_all_stock_async(struct mem_cgroup *memcg); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 677 | |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 678 | static struct mem_cgroup_per_zone * |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 679 | mem_cgroup_zoneinfo(struct mem_cgroup *memcg, int nid, int zid) |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 680 | { |
Glauber Costa | 45cf7eb | 2013-02-22 16:34:49 -0800 | [diff] [blame] | 681 | VM_BUG_ON((unsigned)nid >= nr_node_ids); |
Johannes Weiner | 54f72fe | 2013-07-08 15:59:49 -0700 | [diff] [blame] | 682 | return &memcg->nodeinfo[nid]->zoneinfo[zid]; |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 683 | } |
| 684 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 685 | struct cgroup_subsys_state *mem_cgroup_css(struct mem_cgroup *memcg) |
Wu Fengguang | d324236 | 2009-12-16 12:19:59 +0100 | [diff] [blame] | 686 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 687 | return &memcg->css; |
Wu Fengguang | d324236 | 2009-12-16 12:19:59 +0100 | [diff] [blame] | 688 | } |
| 689 | |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 690 | static struct mem_cgroup_per_zone * |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 691 | page_cgroup_zoneinfo(struct mem_cgroup *memcg, struct page *page) |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 692 | { |
Johannes Weiner | 97a6c37 | 2011-03-23 16:42:27 -0700 | [diff] [blame] | 693 | int nid = page_to_nid(page); |
| 694 | int zid = page_zonenum(page); |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 695 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 696 | return mem_cgroup_zoneinfo(memcg, nid, zid); |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 697 | } |
| 698 | |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 699 | static struct mem_cgroup_tree_per_zone * |
| 700 | soft_limit_tree_node_zone(int nid, int zid) |
| 701 | { |
| 702 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; |
| 703 | } |
| 704 | |
| 705 | static struct mem_cgroup_tree_per_zone * |
| 706 | soft_limit_tree_from_page(struct page *page) |
| 707 | { |
| 708 | int nid = page_to_nid(page); |
| 709 | int zid = page_zonenum(page); |
| 710 | |
| 711 | return &soft_limit_tree.rb_tree_per_node[nid]->rb_tree_per_zone[zid]; |
| 712 | } |
| 713 | |
| 714 | static void |
| 715 | __mem_cgroup_insert_exceeded(struct mem_cgroup *memcg, |
| 716 | struct mem_cgroup_per_zone *mz, |
| 717 | struct mem_cgroup_tree_per_zone *mctz, |
| 718 | unsigned long long new_usage_in_excess) |
| 719 | { |
| 720 | struct rb_node **p = &mctz->rb_root.rb_node; |
| 721 | struct rb_node *parent = NULL; |
| 722 | struct mem_cgroup_per_zone *mz_node; |
| 723 | |
| 724 | if (mz->on_tree) |
| 725 | return; |
| 726 | |
| 727 | mz->usage_in_excess = new_usage_in_excess; |
| 728 | if (!mz->usage_in_excess) |
| 729 | return; |
| 730 | while (*p) { |
| 731 | parent = *p; |
| 732 | mz_node = rb_entry(parent, struct mem_cgroup_per_zone, |
| 733 | tree_node); |
| 734 | if (mz->usage_in_excess < mz_node->usage_in_excess) |
| 735 | p = &(*p)->rb_left; |
| 736 | /* |
| 737 | * We can't avoid mem cgroups that are over their soft |
| 738 | * limit by the same amount |
| 739 | */ |
| 740 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) |
| 741 | p = &(*p)->rb_right; |
| 742 | } |
| 743 | rb_link_node(&mz->tree_node, parent, p); |
| 744 | rb_insert_color(&mz->tree_node, &mctz->rb_root); |
| 745 | mz->on_tree = true; |
| 746 | } |
| 747 | |
| 748 | static void |
| 749 | __mem_cgroup_remove_exceeded(struct mem_cgroup *memcg, |
| 750 | struct mem_cgroup_per_zone *mz, |
| 751 | struct mem_cgroup_tree_per_zone *mctz) |
| 752 | { |
| 753 | if (!mz->on_tree) |
| 754 | return; |
| 755 | rb_erase(&mz->tree_node, &mctz->rb_root); |
| 756 | mz->on_tree = false; |
| 757 | } |
| 758 | |
| 759 | static void |
| 760 | mem_cgroup_remove_exceeded(struct mem_cgroup *memcg, |
| 761 | struct mem_cgroup_per_zone *mz, |
| 762 | struct mem_cgroup_tree_per_zone *mctz) |
| 763 | { |
| 764 | spin_lock(&mctz->lock); |
| 765 | __mem_cgroup_remove_exceeded(memcg, mz, mctz); |
| 766 | spin_unlock(&mctz->lock); |
| 767 | } |
| 768 | |
| 769 | |
| 770 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) |
| 771 | { |
| 772 | unsigned long long excess; |
| 773 | struct mem_cgroup_per_zone *mz; |
| 774 | struct mem_cgroup_tree_per_zone *mctz; |
| 775 | int nid = page_to_nid(page); |
| 776 | int zid = page_zonenum(page); |
| 777 | mctz = soft_limit_tree_from_page(page); |
| 778 | |
| 779 | /* |
| 780 | * Necessary to update all ancestors when hierarchy is used. |
| 781 | * because their event counter is not touched. |
| 782 | */ |
| 783 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { |
| 784 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
| 785 | excess = res_counter_soft_limit_excess(&memcg->res); |
| 786 | /* |
| 787 | * We have to update the tree if mz is on RB-tree or |
| 788 | * mem is over its softlimit. |
| 789 | */ |
| 790 | if (excess || mz->on_tree) { |
| 791 | spin_lock(&mctz->lock); |
| 792 | /* if on-tree, remove it */ |
| 793 | if (mz->on_tree) |
| 794 | __mem_cgroup_remove_exceeded(memcg, mz, mctz); |
| 795 | /* |
| 796 | * Insert again. mz->usage_in_excess will be updated. |
| 797 | * If excess is 0, no tree ops. |
| 798 | */ |
| 799 | __mem_cgroup_insert_exceeded(memcg, mz, mctz, excess); |
| 800 | spin_unlock(&mctz->lock); |
| 801 | } |
| 802 | } |
| 803 | } |
| 804 | |
| 805 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) |
| 806 | { |
| 807 | int node, zone; |
| 808 | struct mem_cgroup_per_zone *mz; |
| 809 | struct mem_cgroup_tree_per_zone *mctz; |
| 810 | |
| 811 | for_each_node(node) { |
| 812 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
| 813 | mz = mem_cgroup_zoneinfo(memcg, node, zone); |
| 814 | mctz = soft_limit_tree_node_zone(node, zone); |
| 815 | mem_cgroup_remove_exceeded(memcg, mz, mctz); |
| 816 | } |
| 817 | } |
| 818 | } |
| 819 | |
| 820 | static struct mem_cgroup_per_zone * |
| 821 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) |
| 822 | { |
| 823 | struct rb_node *rightmost = NULL; |
| 824 | struct mem_cgroup_per_zone *mz; |
| 825 | |
| 826 | retry: |
| 827 | mz = NULL; |
| 828 | rightmost = rb_last(&mctz->rb_root); |
| 829 | if (!rightmost) |
| 830 | goto done; /* Nothing to reclaim from */ |
| 831 | |
| 832 | mz = rb_entry(rightmost, struct mem_cgroup_per_zone, tree_node); |
| 833 | /* |
| 834 | * Remove the node now but someone else can add it back, |
| 835 | * we will to add it back at the end of reclaim to its correct |
| 836 | * position in the tree. |
| 837 | */ |
| 838 | __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz); |
| 839 | if (!res_counter_soft_limit_excess(&mz->memcg->res) || |
| 840 | !css_tryget(&mz->memcg->css)) |
| 841 | goto retry; |
| 842 | done: |
| 843 | return mz; |
| 844 | } |
| 845 | |
| 846 | static struct mem_cgroup_per_zone * |
| 847 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz) |
| 848 | { |
| 849 | struct mem_cgroup_per_zone *mz; |
| 850 | |
| 851 | spin_lock(&mctz->lock); |
| 852 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
| 853 | spin_unlock(&mctz->lock); |
| 854 | return mz; |
| 855 | } |
| 856 | |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 857 | /* |
| 858 | * Implementation Note: reading percpu statistics for memcg. |
| 859 | * |
| 860 | * Both of vmstat[] and percpu_counter has threshold and do periodic |
| 861 | * synchronization to implement "quick" read. There are trade-off between |
| 862 | * reading cost and precision of value. Then, we may have a chance to implement |
| 863 | * a periodic synchronizion of counter in memcg's counter. |
| 864 | * |
| 865 | * But this _read() function is used for user interface now. The user accounts |
| 866 | * memory usage by memory cgroup and he _always_ requires exact value because |
| 867 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always |
| 868 | * have to visit all online cpus and make sum. So, for now, unnecessary |
| 869 | * synchronization is not implemented. (just implemented for cpu hotplug) |
| 870 | * |
| 871 | * If there are kernel internal actions which can make use of some not-exact |
| 872 | * value, and reading all cpu value can be performance bottleneck in some |
| 873 | * common workload, threashold and synchonization as vmstat[] should be |
| 874 | * implemented. |
| 875 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 876 | static long mem_cgroup_read_stat(struct mem_cgroup *memcg, |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 877 | enum mem_cgroup_stat_index idx) |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 878 | { |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 879 | long val = 0; |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 880 | int cpu; |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 881 | |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 882 | get_online_cpus(); |
| 883 | for_each_online_cpu(cpu) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 884 | val += per_cpu(memcg->stat->count[idx], cpu); |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 885 | #ifdef CONFIG_HOTPLUG_CPU |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 886 | spin_lock(&memcg->pcp_counter_lock); |
| 887 | val += memcg->nocpu_base.count[idx]; |
| 888 | spin_unlock(&memcg->pcp_counter_lock); |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 889 | #endif |
| 890 | put_online_cpus(); |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 891 | return val; |
| 892 | } |
| 893 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 894 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 895 | bool charge) |
| 896 | { |
| 897 | int val = (charge) ? 1 : -1; |
Kamezawa Hiroyuki | bff6bb8 | 2012-07-31 16:41:38 -0700 | [diff] [blame] | 898 | this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_SWAP], val); |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 899 | } |
| 900 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 901 | static unsigned long mem_cgroup_read_events(struct mem_cgroup *memcg, |
Johannes Weiner | e9f8974 | 2011-03-23 16:42:37 -0700 | [diff] [blame] | 902 | enum mem_cgroup_events_index idx) |
| 903 | { |
| 904 | unsigned long val = 0; |
| 905 | int cpu; |
| 906 | |
David Rientjes | 9c56751 | 2013-10-16 13:46:43 -0700 | [diff] [blame] | 907 | get_online_cpus(); |
Johannes Weiner | e9f8974 | 2011-03-23 16:42:37 -0700 | [diff] [blame] | 908 | for_each_online_cpu(cpu) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 909 | val += per_cpu(memcg->stat->events[idx], cpu); |
Johannes Weiner | e9f8974 | 2011-03-23 16:42:37 -0700 | [diff] [blame] | 910 | #ifdef CONFIG_HOTPLUG_CPU |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 911 | spin_lock(&memcg->pcp_counter_lock); |
| 912 | val += memcg->nocpu_base.events[idx]; |
| 913 | spin_unlock(&memcg->pcp_counter_lock); |
Johannes Weiner | e9f8974 | 2011-03-23 16:42:37 -0700 | [diff] [blame] | 914 | #endif |
David Rientjes | 9c56751 | 2013-10-16 13:46:43 -0700 | [diff] [blame] | 915 | put_online_cpus(); |
Johannes Weiner | e9f8974 | 2011-03-23 16:42:37 -0700 | [diff] [blame] | 916 | return val; |
| 917 | } |
| 918 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 919 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
David Rientjes | b070e65 | 2013-05-07 16:18:09 -0700 | [diff] [blame] | 920 | struct page *page, |
KAMEZAWA Hiroyuki | b2402857 | 2012-03-21 16:34:22 -0700 | [diff] [blame] | 921 | bool anon, int nr_pages) |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 922 | { |
KAMEZAWA Hiroyuki | b2402857 | 2012-03-21 16:34:22 -0700 | [diff] [blame] | 923 | /* |
| 924 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is |
| 925 | * counted as CACHE even if it's on ANON LRU. |
| 926 | */ |
| 927 | if (anon) |
| 928 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS], |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 929 | nr_pages); |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 930 | else |
KAMEZAWA Hiroyuki | b2402857 | 2012-03-21 16:34:22 -0700 | [diff] [blame] | 931 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_CACHE], |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 932 | nr_pages); |
Balaji Rao | 55e462b | 2008-05-01 04:35:12 -0700 | [diff] [blame] | 933 | |
David Rientjes | b070e65 | 2013-05-07 16:18:09 -0700 | [diff] [blame] | 934 | if (PageTransHuge(page)) |
| 935 | __this_cpu_add(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
| 936 | nr_pages); |
| 937 | |
KAMEZAWA Hiroyuki | e401f17 | 2011-01-20 14:44:23 -0800 | [diff] [blame] | 938 | /* pagein of a big page is an event. So, ignore page size */ |
| 939 | if (nr_pages > 0) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 940 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); |
KAMEZAWA Hiroyuki | 3751d60 | 2011-02-01 15:52:45 -0800 | [diff] [blame] | 941 | else { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 942 | __this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); |
KAMEZAWA Hiroyuki | 3751d60 | 2011-02-01 15:52:45 -0800 | [diff] [blame] | 943 | nr_pages = -nr_pages; /* for event */ |
| 944 | } |
KAMEZAWA Hiroyuki | e401f17 | 2011-01-20 14:44:23 -0800 | [diff] [blame] | 945 | |
Johannes Weiner | 1311471 | 2012-05-29 15:07:07 -0700 | [diff] [blame] | 946 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 947 | } |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 948 | |
KAMEZAWA Hiroyuki | bb2a0de | 2011-07-26 16:08:22 -0700 | [diff] [blame] | 949 | unsigned long |
Hugh Dickins | 4d7dcca | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 950 | mem_cgroup_get_lru_size(struct lruvec *lruvec, enum lru_list lru) |
Konstantin Khlebnikov | 074291f | 2012-05-29 15:07:00 -0700 | [diff] [blame] | 951 | { |
| 952 | struct mem_cgroup_per_zone *mz; |
| 953 | |
| 954 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
| 955 | return mz->lru_size[lru]; |
| 956 | } |
| 957 | |
| 958 | static unsigned long |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 959 | mem_cgroup_zone_nr_lru_pages(struct mem_cgroup *memcg, int nid, int zid, |
KAMEZAWA Hiroyuki | bb2a0de | 2011-07-26 16:08:22 -0700 | [diff] [blame] | 960 | unsigned int lru_mask) |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 961 | { |
| 962 | struct mem_cgroup_per_zone *mz; |
Hugh Dickins | f156ab9 | 2012-03-21 16:34:19 -0700 | [diff] [blame] | 963 | enum lru_list lru; |
KAMEZAWA Hiroyuki | bb2a0de | 2011-07-26 16:08:22 -0700 | [diff] [blame] | 964 | unsigned long ret = 0; |
| 965 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 966 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
KAMEZAWA Hiroyuki | bb2a0de | 2011-07-26 16:08:22 -0700 | [diff] [blame] | 967 | |
Hugh Dickins | f156ab9 | 2012-03-21 16:34:19 -0700 | [diff] [blame] | 968 | for_each_lru(lru) { |
| 969 | if (BIT(lru) & lru_mask) |
| 970 | ret += mz->lru_size[lru]; |
KAMEZAWA Hiroyuki | bb2a0de | 2011-07-26 16:08:22 -0700 | [diff] [blame] | 971 | } |
| 972 | return ret; |
| 973 | } |
| 974 | |
| 975 | static unsigned long |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 976 | mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
KAMEZAWA Hiroyuki | bb2a0de | 2011-07-26 16:08:22 -0700 | [diff] [blame] | 977 | int nid, unsigned int lru_mask) |
| 978 | { |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 979 | u64 total = 0; |
| 980 | int zid; |
| 981 | |
KAMEZAWA Hiroyuki | bb2a0de | 2011-07-26 16:08:22 -0700 | [diff] [blame] | 982 | for (zid = 0; zid < MAX_NR_ZONES; zid++) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 983 | total += mem_cgroup_zone_nr_lru_pages(memcg, |
| 984 | nid, zid, lru_mask); |
KAMEZAWA Hiroyuki | bb2a0de | 2011-07-26 16:08:22 -0700 | [diff] [blame] | 985 | |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 986 | return total; |
| 987 | } |
KAMEZAWA Hiroyuki | bb2a0de | 2011-07-26 16:08:22 -0700 | [diff] [blame] | 988 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 989 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
KAMEZAWA Hiroyuki | bb2a0de | 2011-07-26 16:08:22 -0700 | [diff] [blame] | 990 | unsigned int lru_mask) |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 991 | { |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 992 | int nid; |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 993 | u64 total = 0; |
| 994 | |
Lai Jiangshan | 31aaea4 | 2012-12-12 13:51:27 -0800 | [diff] [blame] | 995 | for_each_node_state(nid, N_MEMORY) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 996 | total += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 997 | return total; |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 998 | } |
| 999 | |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 1000 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
| 1001 | enum mem_cgroup_events_target target) |
KAMEZAWA Hiroyuki | d2265e6 | 2010-03-10 15:22:31 -0800 | [diff] [blame] | 1002 | { |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 1003 | unsigned long val, next; |
| 1004 | |
Johannes Weiner | 1311471 | 2012-05-29 15:07:07 -0700 | [diff] [blame] | 1005 | val = __this_cpu_read(memcg->stat->nr_page_events); |
Steven Rostedt | 4799401 | 2011-11-02 13:38:33 -0700 | [diff] [blame] | 1006 | next = __this_cpu_read(memcg->stat->targets[target]); |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 1007 | /* from time_after() in jiffies.h */ |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 1008 | if ((long)next - (long)val < 0) { |
| 1009 | switch (target) { |
| 1010 | case MEM_CGROUP_TARGET_THRESH: |
| 1011 | next = val + THRESHOLDS_EVENTS_TARGET; |
| 1012 | break; |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 1013 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
| 1014 | next = val + SOFTLIMIT_EVENTS_TARGET; |
| 1015 | break; |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 1016 | case MEM_CGROUP_TARGET_NUMAINFO: |
| 1017 | next = val + NUMAINFO_EVENTS_TARGET; |
| 1018 | break; |
| 1019 | default: |
| 1020 | break; |
| 1021 | } |
| 1022 | __this_cpu_write(memcg->stat->targets[target], next); |
| 1023 | return true; |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 1024 | } |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 1025 | return false; |
KAMEZAWA Hiroyuki | d2265e6 | 2010-03-10 15:22:31 -0800 | [diff] [blame] | 1026 | } |
| 1027 | |
| 1028 | /* |
| 1029 | * Check events in order. |
| 1030 | * |
| 1031 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1032 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
KAMEZAWA Hiroyuki | d2265e6 | 2010-03-10 15:22:31 -0800 | [diff] [blame] | 1033 | { |
Steven Rostedt | 4799401 | 2011-11-02 13:38:33 -0700 | [diff] [blame] | 1034 | preempt_disable(); |
KAMEZAWA Hiroyuki | d2265e6 | 2010-03-10 15:22:31 -0800 | [diff] [blame] | 1035 | /* threshold event is triggered in finer grain than soft limit */ |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 1036 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
| 1037 | MEM_CGROUP_TARGET_THRESH))) { |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 1038 | bool do_softlimit; |
Andrew Morton | 82b3f2a | 2012-02-03 15:37:14 -0800 | [diff] [blame] | 1039 | bool do_numainfo __maybe_unused; |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 1040 | |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 1041 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
| 1042 | MEM_CGROUP_TARGET_SOFTLIMIT); |
KAMEZAWA Hiroyuki | 453a9bf | 2011-07-08 15:39:43 -0700 | [diff] [blame] | 1043 | #if MAX_NUMNODES > 1 |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 1044 | do_numainfo = mem_cgroup_event_ratelimit(memcg, |
| 1045 | MEM_CGROUP_TARGET_NUMAINFO); |
KAMEZAWA Hiroyuki | 453a9bf | 2011-07-08 15:39:43 -0700 | [diff] [blame] | 1046 | #endif |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 1047 | preempt_enable(); |
| 1048 | |
| 1049 | mem_cgroup_threshold(memcg); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 1050 | if (unlikely(do_softlimit)) |
| 1051 | mem_cgroup_update_tree(memcg, page); |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 1052 | #if MAX_NUMNODES > 1 |
| 1053 | if (unlikely(do_numainfo)) |
| 1054 | atomic_inc(&memcg->numainfo_events); |
| 1055 | #endif |
| 1056 | } else |
| 1057 | preempt_enable(); |
KAMEZAWA Hiroyuki | d2265e6 | 2010-03-10 15:22:31 -0800 | [diff] [blame] | 1058 | } |
| 1059 | |
Balbir Singh | cf475ad | 2008-04-29 01:00:16 -0700 | [diff] [blame] | 1060 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 1061 | { |
Balbir Singh | 31a78f2 | 2008-09-28 23:09:31 +0100 | [diff] [blame] | 1062 | /* |
| 1063 | * mm_update_next_owner() may clear mm->owner to NULL |
| 1064 | * if it races with swapoff, page migration, etc. |
| 1065 | * So this can be called with p == NULL. |
| 1066 | */ |
| 1067 | if (unlikely(!p)) |
| 1068 | return NULL; |
| 1069 | |
Tejun Heo | 073219e | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 1070 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 1071 | } |
| 1072 | |
Johannes Weiner | df38197 | 2014-04-07 15:37:43 -0700 | [diff] [blame] | 1073 | 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] | 1074 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1075 | struct mem_cgroup *memcg = NULL; |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 1076 | |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 1077 | rcu_read_lock(); |
| 1078 | do { |
Michal Hocko | 6f6acb0 | 2014-05-22 11:54:19 -0700 | [diff] [blame] | 1079 | /* |
| 1080 | * Page cache insertions can happen withou an |
| 1081 | * actual mm context, e.g. during disk probing |
| 1082 | * on boot, loopback IO, acct() writes etc. |
| 1083 | */ |
| 1084 | if (unlikely(!mm)) |
Johannes Weiner | df38197 | 2014-04-07 15:37:43 -0700 | [diff] [blame] | 1085 | memcg = root_mem_cgroup; |
Michal Hocko | 6f6acb0 | 2014-05-22 11:54:19 -0700 | [diff] [blame] | 1086 | else { |
| 1087 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
| 1088 | if (unlikely(!memcg)) |
| 1089 | memcg = root_mem_cgroup; |
| 1090 | } |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1091 | } while (!css_tryget(&memcg->css)); |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 1092 | rcu_read_unlock(); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1093 | return memcg; |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 1094 | } |
| 1095 | |
Michal Hocko | 16248d8 | 2013-04-29 15:07:19 -0700 | [diff] [blame] | 1096 | /* |
| 1097 | * Returns a next (in a pre-order walk) alive memcg (with elevated css |
| 1098 | * ref. count) or NULL if the whole root's subtree has been visited. |
| 1099 | * |
| 1100 | * helper function to be used by mem_cgroup_iter |
| 1101 | */ |
| 1102 | static struct mem_cgroup *__mem_cgroup_iter_next(struct mem_cgroup *root, |
Andrew Morton | 694fbc0 | 2013-09-24 15:27:37 -0700 | [diff] [blame] | 1103 | struct mem_cgroup *last_visited) |
Michal Hocko | 16248d8 | 2013-04-29 15:07:19 -0700 | [diff] [blame] | 1104 | { |
Tejun Heo | 492eb21 | 2013-08-08 20:11:25 -0400 | [diff] [blame] | 1105 | struct cgroup_subsys_state *prev_css, *next_css; |
Michal Hocko | 16248d8 | 2013-04-29 15:07:19 -0700 | [diff] [blame] | 1106 | |
Tejun Heo | bd8815a | 2013-08-08 20:11:27 -0400 | [diff] [blame] | 1107 | prev_css = last_visited ? &last_visited->css : NULL; |
Michal Hocko | 16248d8 | 2013-04-29 15:07:19 -0700 | [diff] [blame] | 1108 | skip_node: |
Tejun Heo | 492eb21 | 2013-08-08 20:11:25 -0400 | [diff] [blame] | 1109 | next_css = css_next_descendant_pre(prev_css, &root->css); |
Michal Hocko | 16248d8 | 2013-04-29 15:07:19 -0700 | [diff] [blame] | 1110 | |
| 1111 | /* |
| 1112 | * Even if we found a group we have to make sure it is |
| 1113 | * alive. css && !memcg means that the groups should be |
| 1114 | * skipped and we should continue the tree walk. |
| 1115 | * last_visited css is safe to use because it is |
| 1116 | * protected by css_get and the tree walk is rcu safe. |
Michal Hocko | 0eef615 | 2014-01-23 15:53:37 -0800 | [diff] [blame] | 1117 | * |
| 1118 | * We do not take a reference on the root of the tree walk |
| 1119 | * because we might race with the root removal when it would |
| 1120 | * be the only node in the iterated hierarchy and mem_cgroup_iter |
| 1121 | * would end up in an endless loop because it expects that at |
| 1122 | * least one valid node will be returned. Root cannot disappear |
| 1123 | * because caller of the iterator should hold it already so |
| 1124 | * skipping css reference should be safe. |
Michal Hocko | 16248d8 | 2013-04-29 15:07:19 -0700 | [diff] [blame] | 1125 | */ |
Tejun Heo | 492eb21 | 2013-08-08 20:11:25 -0400 | [diff] [blame] | 1126 | if (next_css) { |
Hugh Dickins | ce48225 | 2014-03-03 15:38:24 -0800 | [diff] [blame] | 1127 | if ((next_css == &root->css) || |
| 1128 | ((next_css->flags & CSS_ONLINE) && css_tryget(next_css))) |
Hugh Dickins | d8ad305 | 2014-01-23 15:53:32 -0800 | [diff] [blame] | 1129 | return mem_cgroup_from_css(next_css); |
Michal Hocko | 0eef615 | 2014-01-23 15:53:37 -0800 | [diff] [blame] | 1130 | |
| 1131 | prev_css = next_css; |
| 1132 | goto skip_node; |
Michal Hocko | 16248d8 | 2013-04-29 15:07:19 -0700 | [diff] [blame] | 1133 | } |
| 1134 | |
| 1135 | return NULL; |
| 1136 | } |
| 1137 | |
Johannes Weiner | 519ebea | 2013-07-03 15:04:51 -0700 | [diff] [blame] | 1138 | static void mem_cgroup_iter_invalidate(struct mem_cgroup *root) |
| 1139 | { |
| 1140 | /* |
| 1141 | * When a group in the hierarchy below root is destroyed, the |
| 1142 | * hierarchy iterator can no longer be trusted since it might |
| 1143 | * have pointed to the destroyed group. Invalidate it. |
| 1144 | */ |
| 1145 | atomic_inc(&root->dead_count); |
| 1146 | } |
| 1147 | |
| 1148 | static struct mem_cgroup * |
| 1149 | mem_cgroup_iter_load(struct mem_cgroup_reclaim_iter *iter, |
| 1150 | struct mem_cgroup *root, |
| 1151 | int *sequence) |
| 1152 | { |
| 1153 | struct mem_cgroup *position = NULL; |
| 1154 | /* |
| 1155 | * A cgroup destruction happens in two stages: offlining and |
| 1156 | * release. They are separated by a RCU grace period. |
| 1157 | * |
| 1158 | * If the iterator is valid, we may still race with an |
| 1159 | * offlining. The RCU lock ensures the object won't be |
| 1160 | * released, tryget will fail if we lost the race. |
| 1161 | */ |
| 1162 | *sequence = atomic_read(&root->dead_count); |
| 1163 | if (iter->last_dead_count == *sequence) { |
| 1164 | smp_rmb(); |
| 1165 | position = iter->last_visited; |
Michal Hocko | ecc736f | 2014-01-23 15:53:35 -0800 | [diff] [blame] | 1166 | |
| 1167 | /* |
| 1168 | * We cannot take a reference to root because we might race |
| 1169 | * with root removal and returning NULL would end up in |
| 1170 | * an endless loop on the iterator user level when root |
| 1171 | * would be returned all the time. |
| 1172 | */ |
| 1173 | if (position && position != root && |
| 1174 | !css_tryget(&position->css)) |
Johannes Weiner | 519ebea | 2013-07-03 15:04:51 -0700 | [diff] [blame] | 1175 | position = NULL; |
| 1176 | } |
| 1177 | return position; |
| 1178 | } |
| 1179 | |
| 1180 | static void mem_cgroup_iter_update(struct mem_cgroup_reclaim_iter *iter, |
| 1181 | struct mem_cgroup *last_visited, |
| 1182 | struct mem_cgroup *new_position, |
Michal Hocko | ecc736f | 2014-01-23 15:53:35 -0800 | [diff] [blame] | 1183 | struct mem_cgroup *root, |
Johannes Weiner | 519ebea | 2013-07-03 15:04:51 -0700 | [diff] [blame] | 1184 | int sequence) |
| 1185 | { |
Michal Hocko | ecc736f | 2014-01-23 15:53:35 -0800 | [diff] [blame] | 1186 | /* root reference counting symmetric to mem_cgroup_iter_load */ |
| 1187 | if (last_visited && last_visited != root) |
Johannes Weiner | 519ebea | 2013-07-03 15:04:51 -0700 | [diff] [blame] | 1188 | css_put(&last_visited->css); |
| 1189 | /* |
| 1190 | * We store the sequence count from the time @last_visited was |
| 1191 | * loaded successfully instead of rereading it here so that we |
| 1192 | * don't lose destruction events in between. We could have |
| 1193 | * raced with the destruction of @new_position after all. |
| 1194 | */ |
| 1195 | iter->last_visited = new_position; |
| 1196 | smp_wmb(); |
| 1197 | iter->last_dead_count = sequence; |
| 1198 | } |
| 1199 | |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 1200 | /** |
| 1201 | * mem_cgroup_iter - iterate over memory cgroup hierarchy |
| 1202 | * @root: hierarchy root |
| 1203 | * @prev: previously returned memcg, NULL on first invocation |
| 1204 | * @reclaim: cookie for shared reclaim walks, NULL for full walks |
| 1205 | * |
| 1206 | * Returns references to children of the hierarchy below @root, or |
| 1207 | * @root itself, or %NULL after a full round-trip. |
| 1208 | * |
| 1209 | * Caller must pass the return value in @prev on subsequent |
| 1210 | * invocations for reference counting, or use mem_cgroup_iter_break() |
| 1211 | * to cancel a hierarchy walk before the round-trip is complete. |
| 1212 | * |
| 1213 | * Reclaimers can specify a zone and a priority level in @reclaim to |
| 1214 | * divide up the memcgs in the hierarchy among all concurrent |
| 1215 | * reclaimers operating on the same zone and priority. |
| 1216 | */ |
Andrew Morton | 694fbc0 | 2013-09-24 15:27:37 -0700 | [diff] [blame] | 1217 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 1218 | struct mem_cgroup *prev, |
Andrew Morton | 694fbc0 | 2013-09-24 15:27:37 -0700 | [diff] [blame] | 1219 | struct mem_cgroup_reclaim_cookie *reclaim) |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 1220 | { |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1221 | struct mem_cgroup *memcg = NULL; |
Michal Hocko | 542f85f | 2013-04-29 15:07:15 -0700 | [diff] [blame] | 1222 | struct mem_cgroup *last_visited = NULL; |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1223 | |
Andrew Morton | 694fbc0 | 2013-09-24 15:27:37 -0700 | [diff] [blame] | 1224 | if (mem_cgroup_disabled()) |
| 1225 | return NULL; |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 1226 | |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 1227 | if (!root) |
| 1228 | root = root_mem_cgroup; |
| 1229 | |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1230 | if (prev && !reclaim) |
Michal Hocko | 542f85f | 2013-04-29 15:07:15 -0700 | [diff] [blame] | 1231 | last_visited = prev; |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1232 | |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1233 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
| 1234 | if (prev) |
Michal Hocko | c40046f | 2013-04-29 15:07:14 -0700 | [diff] [blame] | 1235 | goto out_css_put; |
Andrew Morton | 694fbc0 | 2013-09-24 15:27:37 -0700 | [diff] [blame] | 1236 | return root; |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1237 | } |
| 1238 | |
Michal Hocko | 542f85f | 2013-04-29 15:07:15 -0700 | [diff] [blame] | 1239 | rcu_read_lock(); |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1240 | while (!memcg) { |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 1241 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
Johannes Weiner | 519ebea | 2013-07-03 15:04:51 -0700 | [diff] [blame] | 1242 | int uninitialized_var(seq); |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1243 | |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 1244 | if (reclaim) { |
| 1245 | int nid = zone_to_nid(reclaim->zone); |
| 1246 | int zid = zone_idx(reclaim->zone); |
| 1247 | struct mem_cgroup_per_zone *mz; |
| 1248 | |
| 1249 | mz = mem_cgroup_zoneinfo(root, nid, zid); |
| 1250 | iter = &mz->reclaim_iter[reclaim->priority]; |
Michal Hocko | 542f85f | 2013-04-29 15:07:15 -0700 | [diff] [blame] | 1251 | if (prev && reclaim->generation != iter->generation) { |
Michal Hocko | 5f57816 | 2013-04-29 15:07:17 -0700 | [diff] [blame] | 1252 | iter->last_visited = NULL; |
Michal Hocko | 542f85f | 2013-04-29 15:07:15 -0700 | [diff] [blame] | 1253 | goto out_unlock; |
| 1254 | } |
Michal Hocko | 5f57816 | 2013-04-29 15:07:17 -0700 | [diff] [blame] | 1255 | |
Johannes Weiner | 519ebea | 2013-07-03 15:04:51 -0700 | [diff] [blame] | 1256 | last_visited = mem_cgroup_iter_load(iter, root, &seq); |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 1257 | } |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1258 | |
Andrew Morton | 694fbc0 | 2013-09-24 15:27:37 -0700 | [diff] [blame] | 1259 | memcg = __mem_cgroup_iter_next(root, last_visited); |
Michal Hocko | 542f85f | 2013-04-29 15:07:15 -0700 | [diff] [blame] | 1260 | |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 1261 | if (reclaim) { |
Michal Hocko | ecc736f | 2014-01-23 15:53:35 -0800 | [diff] [blame] | 1262 | mem_cgroup_iter_update(iter, last_visited, memcg, root, |
| 1263 | seq); |
Michal Hocko | 542f85f | 2013-04-29 15:07:15 -0700 | [diff] [blame] | 1264 | |
Michal Hocko | 19f3940 | 2013-04-29 15:07:18 -0700 | [diff] [blame] | 1265 | if (!memcg) |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 1266 | iter->generation++; |
| 1267 | else if (!prev && memcg) |
| 1268 | reclaim->generation = iter->generation; |
| 1269 | } |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1270 | |
Andrew Morton | 694fbc0 | 2013-09-24 15:27:37 -0700 | [diff] [blame] | 1271 | if (prev && !memcg) |
Michal Hocko | 542f85f | 2013-04-29 15:07:15 -0700 | [diff] [blame] | 1272 | goto out_unlock; |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1273 | } |
Michal Hocko | 542f85f | 2013-04-29 15:07:15 -0700 | [diff] [blame] | 1274 | out_unlock: |
| 1275 | rcu_read_unlock(); |
Michal Hocko | c40046f | 2013-04-29 15:07:14 -0700 | [diff] [blame] | 1276 | out_css_put: |
| 1277 | if (prev && prev != root) |
| 1278 | css_put(&prev->css); |
| 1279 | |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1280 | return memcg; |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1281 | } |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1282 | |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 1283 | /** |
| 1284 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely |
| 1285 | * @root: hierarchy root |
| 1286 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() |
| 1287 | */ |
| 1288 | void mem_cgroup_iter_break(struct mem_cgroup *root, |
| 1289 | struct mem_cgroup *prev) |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1290 | { |
| 1291 | if (!root) |
| 1292 | root = root_mem_cgroup; |
| 1293 | if (prev && prev != root) |
| 1294 | css_put(&prev->css); |
| 1295 | } |
| 1296 | |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1297 | /* |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1298 | * Iteration constructs for visiting all cgroups (under a tree). If |
| 1299 | * loops are exited prematurely (break), mem_cgroup_iter_break() must |
| 1300 | * be used for reference counting. |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1301 | */ |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1302 | #define for_each_mem_cgroup_tree(iter, root) \ |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 1303 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1304 | iter != NULL; \ |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 1305 | iter = mem_cgroup_iter(root, iter, NULL)) |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1306 | |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1307 | #define for_each_mem_cgroup(iter) \ |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 1308 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1309 | iter != NULL; \ |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 1310 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1311 | |
David Rientjes | 68ae564 | 2012-12-12 13:51:57 -0800 | [diff] [blame] | 1312 | 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] | 1313 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1314 | struct mem_cgroup *memcg; |
Ying Han | 456f998 | 2011-05-26 16:25:38 -0700 | [diff] [blame] | 1315 | |
Ying Han | 456f998 | 2011-05-26 16:25:38 -0700 | [diff] [blame] | 1316 | rcu_read_lock(); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1317 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
| 1318 | if (unlikely(!memcg)) |
Ying Han | 456f998 | 2011-05-26 16:25:38 -0700 | [diff] [blame] | 1319 | goto out; |
| 1320 | |
| 1321 | switch (idx) { |
Ying Han | 456f998 | 2011-05-26 16:25:38 -0700 | [diff] [blame] | 1322 | case PGFAULT: |
Johannes Weiner | 0e574a9 | 2012-01-12 17:18:35 -0800 | [diff] [blame] | 1323 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGFAULT]); |
| 1324 | break; |
| 1325 | case PGMAJFAULT: |
| 1326 | this_cpu_inc(memcg->stat->events[MEM_CGROUP_EVENTS_PGMAJFAULT]); |
Ying Han | 456f998 | 2011-05-26 16:25:38 -0700 | [diff] [blame] | 1327 | break; |
| 1328 | default: |
| 1329 | BUG(); |
| 1330 | } |
| 1331 | out: |
| 1332 | rcu_read_unlock(); |
| 1333 | } |
David Rientjes | 68ae564 | 2012-12-12 13:51:57 -0800 | [diff] [blame] | 1334 | EXPORT_SYMBOL(__mem_cgroup_count_vm_event); |
Ying Han | 456f998 | 2011-05-26 16:25:38 -0700 | [diff] [blame] | 1335 | |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1336 | /** |
| 1337 | * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg |
| 1338 | * @zone: zone of the wanted lruvec |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1339 | * @memcg: memcg of the wanted lruvec |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1340 | * |
| 1341 | * Returns the lru list vector holding pages for the given @zone and |
| 1342 | * @mem. This can be the global zone lruvec, if the memory controller |
| 1343 | * is disabled. |
| 1344 | */ |
| 1345 | struct lruvec *mem_cgroup_zone_lruvec(struct zone *zone, |
| 1346 | struct mem_cgroup *memcg) |
| 1347 | { |
| 1348 | struct mem_cgroup_per_zone *mz; |
Hugh Dickins | bea8c15 | 2012-11-16 14:14:54 -0800 | [diff] [blame] | 1349 | struct lruvec *lruvec; |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1350 | |
Hugh Dickins | bea8c15 | 2012-11-16 14:14:54 -0800 | [diff] [blame] | 1351 | if (mem_cgroup_disabled()) { |
| 1352 | lruvec = &zone->lruvec; |
| 1353 | goto out; |
| 1354 | } |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1355 | |
| 1356 | mz = mem_cgroup_zoneinfo(memcg, zone_to_nid(zone), zone_idx(zone)); |
Hugh Dickins | bea8c15 | 2012-11-16 14:14:54 -0800 | [diff] [blame] | 1357 | lruvec = &mz->lruvec; |
| 1358 | out: |
| 1359 | /* |
| 1360 | * Since a node can be onlined after the mem_cgroup was created, |
| 1361 | * we have to be prepared to initialize lruvec->zone here; |
| 1362 | * and if offlined then reonlined, we need to reinitialize it. |
| 1363 | */ |
| 1364 | if (unlikely(lruvec->zone != zone)) |
| 1365 | lruvec->zone = zone; |
| 1366 | return lruvec; |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1367 | } |
| 1368 | |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 1369 | /* |
| 1370 | * Following LRU functions are allowed to be used without PCG_LOCK. |
| 1371 | * Operations are called by routine of global LRU independently from memcg. |
| 1372 | * What we have to take care of here is validness of pc->mem_cgroup. |
| 1373 | * |
| 1374 | * Changes to pc->mem_cgroup happens when |
| 1375 | * 1. charge |
| 1376 | * 2. moving account |
| 1377 | * In typical case, "charge" is done before add-to-lru. Exception is SwapCache. |
| 1378 | * It is added to LRU before charge. |
| 1379 | * If PCG_USED bit is not set, page_cgroup is not added to this private LRU. |
| 1380 | * When moving account, the page is not on LRU. It's isolated. |
| 1381 | */ |
| 1382 | |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1383 | /** |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1384 | * mem_cgroup_page_lruvec - return lruvec for adding an lru page |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1385 | * @page: the page |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1386 | * @zone: zone of the page |
Minchan Kim | 3f58a82 | 2011-03-22 16:32:53 -0700 | [diff] [blame] | 1387 | */ |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1388 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct zone *zone) |
Minchan Kim | 3f58a82 | 2011-03-22 16:32:53 -0700 | [diff] [blame] | 1389 | { |
| 1390 | struct mem_cgroup_per_zone *mz; |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1391 | struct mem_cgroup *memcg; |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 1392 | struct page_cgroup *pc; |
Hugh Dickins | bea8c15 | 2012-11-16 14:14:54 -0800 | [diff] [blame] | 1393 | struct lruvec *lruvec; |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 1394 | |
Hugh Dickins | bea8c15 | 2012-11-16 14:14:54 -0800 | [diff] [blame] | 1395 | if (mem_cgroup_disabled()) { |
| 1396 | lruvec = &zone->lruvec; |
| 1397 | goto out; |
| 1398 | } |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 1399 | |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 1400 | pc = lookup_page_cgroup(page); |
KAMEZAWA Hiroyuki | 38c5d72 | 2012-01-12 17:19:01 -0800 | [diff] [blame] | 1401 | memcg = pc->mem_cgroup; |
Hugh Dickins | 7512102 | 2012-03-05 14:59:18 -0800 | [diff] [blame] | 1402 | |
| 1403 | /* |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1404 | * Surreptitiously switch any uncharged offlist page to root: |
Hugh Dickins | 7512102 | 2012-03-05 14:59:18 -0800 | [diff] [blame] | 1405 | * an uncharged page off lru does nothing to secure |
| 1406 | * its former mem_cgroup from sudden removal. |
| 1407 | * |
| 1408 | * Our caller holds lru_lock, and PageCgroupUsed is updated |
| 1409 | * under page_cgroup lock: between them, they make all uses |
| 1410 | * of pc->mem_cgroup safe. |
| 1411 | */ |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1412 | if (!PageLRU(page) && !PageCgroupUsed(pc) && memcg != root_mem_cgroup) |
Hugh Dickins | 7512102 | 2012-03-05 14:59:18 -0800 | [diff] [blame] | 1413 | pc->mem_cgroup = memcg = root_mem_cgroup; |
| 1414 | |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1415 | mz = page_cgroup_zoneinfo(memcg, page); |
Hugh Dickins | bea8c15 | 2012-11-16 14:14:54 -0800 | [diff] [blame] | 1416 | lruvec = &mz->lruvec; |
| 1417 | out: |
| 1418 | /* |
| 1419 | * Since a node can be onlined after the mem_cgroup was created, |
| 1420 | * we have to be prepared to initialize lruvec->zone here; |
| 1421 | * and if offlined then reonlined, we need to reinitialize it. |
| 1422 | */ |
| 1423 | if (unlikely(lruvec->zone != zone)) |
| 1424 | lruvec->zone = zone; |
| 1425 | return lruvec; |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1426 | } |
| 1427 | |
| 1428 | /** |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1429 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
| 1430 | * @lruvec: mem_cgroup per zone lru vector |
| 1431 | * @lru: index of lru list the page is sitting on |
| 1432 | * @nr_pages: positive when adding or negative when removing |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1433 | * |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1434 | * This function must be called when a page is added to or removed from an |
| 1435 | * lru list. |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1436 | */ |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1437 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
| 1438 | int nr_pages) |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1439 | { |
| 1440 | struct mem_cgroup_per_zone *mz; |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1441 | unsigned long *lru_size; |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1442 | |
| 1443 | if (mem_cgroup_disabled()) |
| 1444 | return; |
| 1445 | |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1446 | mz = container_of(lruvec, struct mem_cgroup_per_zone, lruvec); |
| 1447 | lru_size = mz->lru_size + lru; |
| 1448 | *lru_size += nr_pages; |
| 1449 | VM_BUG_ON((long)(*lru_size) < 0); |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 1450 | } |
KAMEZAWA Hiroyuki | 544122e | 2009-01-07 18:08:34 -0800 | [diff] [blame] | 1451 | |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 1452 | /* |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1453 | * Checks whether given mem is same or in the root_mem_cgroup's |
Michal Hocko | 3e92041 | 2011-07-26 16:08:29 -0700 | [diff] [blame] | 1454 | * hierarchy subtree |
| 1455 | */ |
Johannes Weiner | c3ac9a8 | 2012-05-29 15:06:25 -0700 | [diff] [blame] | 1456 | bool __mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, |
| 1457 | struct mem_cgroup *memcg) |
Michal Hocko | 3e92041 | 2011-07-26 16:08:29 -0700 | [diff] [blame] | 1458 | { |
Johannes Weiner | 91c63734 | 2012-05-29 15:06:24 -0700 | [diff] [blame] | 1459 | if (root_memcg == memcg) |
| 1460 | return true; |
Hugh Dickins | 3a981f4 | 2012-06-20 12:52:58 -0700 | [diff] [blame] | 1461 | if (!root_memcg->use_hierarchy || !memcg) |
Johannes Weiner | 91c63734 | 2012-05-29 15:06:24 -0700 | [diff] [blame] | 1462 | return false; |
Li Zefan | b47f77b | 2013-09-23 16:55:43 +0800 | [diff] [blame] | 1463 | return cgroup_is_descendant(memcg->css.cgroup, root_memcg->css.cgroup); |
Johannes Weiner | c3ac9a8 | 2012-05-29 15:06:25 -0700 | [diff] [blame] | 1464 | } |
| 1465 | |
| 1466 | static bool mem_cgroup_same_or_subtree(const struct mem_cgroup *root_memcg, |
| 1467 | struct mem_cgroup *memcg) |
| 1468 | { |
| 1469 | bool ret; |
| 1470 | |
Johannes Weiner | 91c63734 | 2012-05-29 15:06:24 -0700 | [diff] [blame] | 1471 | rcu_read_lock(); |
Johannes Weiner | c3ac9a8 | 2012-05-29 15:06:25 -0700 | [diff] [blame] | 1472 | ret = __mem_cgroup_same_or_subtree(root_memcg, memcg); |
Johannes Weiner | 91c63734 | 2012-05-29 15:06:24 -0700 | [diff] [blame] | 1473 | rcu_read_unlock(); |
| 1474 | return ret; |
Michal Hocko | 3e92041 | 2011-07-26 16:08:29 -0700 | [diff] [blame] | 1475 | } |
| 1476 | |
David Rientjes | ffbdccf | 2013-07-03 15:01:23 -0700 | [diff] [blame] | 1477 | bool task_in_mem_cgroup(struct task_struct *task, |
| 1478 | const struct mem_cgroup *memcg) |
David Rientjes | 4c4a221 | 2008-02-07 00:14:06 -0800 | [diff] [blame] | 1479 | { |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 1480 | struct mem_cgroup *curr = NULL; |
KAMEZAWA Hiroyuki | 158e0a2 | 2010-08-10 18:03:00 -0700 | [diff] [blame] | 1481 | struct task_struct *p; |
David Rientjes | ffbdccf | 2013-07-03 15:01:23 -0700 | [diff] [blame] | 1482 | bool ret; |
David Rientjes | 4c4a221 | 2008-02-07 00:14:06 -0800 | [diff] [blame] | 1483 | |
KAMEZAWA Hiroyuki | 158e0a2 | 2010-08-10 18:03:00 -0700 | [diff] [blame] | 1484 | p = find_lock_task_mm(task); |
David Rientjes | de077d2 | 2012-01-12 17:18:52 -0800 | [diff] [blame] | 1485 | if (p) { |
Johannes Weiner | df38197 | 2014-04-07 15:37:43 -0700 | [diff] [blame] | 1486 | curr = get_mem_cgroup_from_mm(p->mm); |
David Rientjes | de077d2 | 2012-01-12 17:18:52 -0800 | [diff] [blame] | 1487 | task_unlock(p); |
| 1488 | } else { |
| 1489 | /* |
| 1490 | * All threads may have already detached their mm's, but the oom |
| 1491 | * killer still needs to detect if they have already been oom |
| 1492 | * killed to prevent needlessly killing additional tasks. |
| 1493 | */ |
David Rientjes | ffbdccf | 2013-07-03 15:01:23 -0700 | [diff] [blame] | 1494 | rcu_read_lock(); |
David Rientjes | de077d2 | 2012-01-12 17:18:52 -0800 | [diff] [blame] | 1495 | curr = mem_cgroup_from_task(task); |
| 1496 | if (curr) |
| 1497 | css_get(&curr->css); |
David Rientjes | ffbdccf | 2013-07-03 15:01:23 -0700 | [diff] [blame] | 1498 | rcu_read_unlock(); |
David Rientjes | de077d2 | 2012-01-12 17:18:52 -0800 | [diff] [blame] | 1499 | } |
Daisuke Nishimura | d31f56d | 2009-12-15 16:47:12 -0800 | [diff] [blame] | 1500 | /* |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1501 | * We should check use_hierarchy of "memcg" not "curr". Because checking |
Daisuke Nishimura | d31f56d | 2009-12-15 16:47:12 -0800 | [diff] [blame] | 1502 | * use_hierarchy of "curr" here make this function true if hierarchy is |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1503 | * enabled in "curr" and "curr" is a child of "memcg" in *cgroup* |
| 1504 | * hierarchy(even if use_hierarchy is disabled in "memcg"). |
Daisuke Nishimura | d31f56d | 2009-12-15 16:47:12 -0800 | [diff] [blame] | 1505 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1506 | ret = mem_cgroup_same_or_subtree(memcg, curr); |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 1507 | css_put(&curr->css); |
David Rientjes | 4c4a221 | 2008-02-07 00:14:06 -0800 | [diff] [blame] | 1508 | return ret; |
| 1509 | } |
| 1510 | |
Konstantin Khlebnikov | c56d5c7 | 2012-05-29 15:07:00 -0700 | [diff] [blame] | 1511 | int mem_cgroup_inactive_anon_is_low(struct lruvec *lruvec) |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 1512 | { |
KOSAKI Motohiro | c772be9 | 2009-01-07 18:08:25 -0800 | [diff] [blame] | 1513 | unsigned long inactive_ratio; |
Johannes Weiner | 9b27297 | 2011-11-02 13:38:23 -0700 | [diff] [blame] | 1514 | unsigned long inactive; |
| 1515 | unsigned long active; |
| 1516 | unsigned long gb; |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 1517 | |
Hugh Dickins | 4d7dcca | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 1518 | inactive = mem_cgroup_get_lru_size(lruvec, LRU_INACTIVE_ANON); |
| 1519 | active = mem_cgroup_get_lru_size(lruvec, LRU_ACTIVE_ANON); |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 1520 | |
KOSAKI Motohiro | c772be9 | 2009-01-07 18:08:25 -0800 | [diff] [blame] | 1521 | gb = (inactive + active) >> (30 - PAGE_SHIFT); |
| 1522 | if (gb) |
| 1523 | inactive_ratio = int_sqrt(10 * gb); |
| 1524 | else |
| 1525 | inactive_ratio = 1; |
| 1526 | |
Johannes Weiner | 9b27297 | 2011-11-02 13:38:23 -0700 | [diff] [blame] | 1527 | return inactive * inactive_ratio < active; |
KOSAKI Motohiro | c772be9 | 2009-01-07 18:08:25 -0800 | [diff] [blame] | 1528 | } |
| 1529 | |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1530 | #define mem_cgroup_from_res_counter(counter, member) \ |
| 1531 | container_of(counter, struct mem_cgroup, member) |
| 1532 | |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 1533 | /** |
Johannes Weiner | 9d11ea9 | 2011-03-23 16:42:21 -0700 | [diff] [blame] | 1534 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
Wanpeng Li | dad7557 | 2012-06-20 12:53:01 -0700 | [diff] [blame] | 1535 | * @memcg: the memory cgroup |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 1536 | * |
Johannes Weiner | 9d11ea9 | 2011-03-23 16:42:21 -0700 | [diff] [blame] | 1537 | * Returns the maximum amount of memory @mem can be charged with, in |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 1538 | * pages. |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 1539 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1540 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 1541 | { |
Johannes Weiner | 9d11ea9 | 2011-03-23 16:42:21 -0700 | [diff] [blame] | 1542 | unsigned long long margin; |
| 1543 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1544 | margin = res_counter_margin(&memcg->res); |
Johannes Weiner | 9d11ea9 | 2011-03-23 16:42:21 -0700 | [diff] [blame] | 1545 | if (do_swap_account) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1546 | margin = min(margin, res_counter_margin(&memcg->memsw)); |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 1547 | return margin >> PAGE_SHIFT; |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 1548 | } |
| 1549 | |
KAMEZAWA Hiroyuki | 1f4c025 | 2011-07-26 16:08:21 -0700 | [diff] [blame] | 1550 | int mem_cgroup_swappiness(struct mem_cgroup *memcg) |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 1551 | { |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 1552 | /* root ? */ |
Tejun Heo | 6387698 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 1553 | if (!css_parent(&memcg->css)) |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 1554 | return vm_swappiness; |
| 1555 | |
Johannes Weiner | bf1ff26 | 2011-03-23 16:42:32 -0700 | [diff] [blame] | 1556 | return memcg->swappiness; |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 1557 | } |
| 1558 | |
KAMEZAWA Hiroyuki | 619d094 | 2012-03-21 16:34:23 -0700 | [diff] [blame] | 1559 | /* |
| 1560 | * memcg->moving_account is used for checking possibility that some thread is |
| 1561 | * calling move_account(). When a thread on CPU-A starts moving pages under |
| 1562 | * a memcg, other threads should check memcg->moving_account under |
| 1563 | * rcu_read_lock(), like this: |
| 1564 | * |
| 1565 | * CPU-A CPU-B |
| 1566 | * rcu_read_lock() |
| 1567 | * memcg->moving_account+1 if (memcg->mocing_account) |
| 1568 | * take heavy locks. |
| 1569 | * synchronize_rcu() update something. |
| 1570 | * rcu_read_unlock() |
| 1571 | * start move here. |
| 1572 | */ |
KAMEZAWA Hiroyuki | 4331f7d | 2012-03-21 16:34:26 -0700 | [diff] [blame] | 1573 | |
| 1574 | /* for quick checking without looking up memcg */ |
| 1575 | atomic_t memcg_moving __read_mostly; |
| 1576 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1577 | static void mem_cgroup_start_move(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1578 | { |
KAMEZAWA Hiroyuki | 4331f7d | 2012-03-21 16:34:26 -0700 | [diff] [blame] | 1579 | atomic_inc(&memcg_moving); |
KAMEZAWA Hiroyuki | 619d094 | 2012-03-21 16:34:23 -0700 | [diff] [blame] | 1580 | atomic_inc(&memcg->moving_account); |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1581 | synchronize_rcu(); |
| 1582 | } |
| 1583 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1584 | static void mem_cgroup_end_move(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1585 | { |
KAMEZAWA Hiroyuki | 619d094 | 2012-03-21 16:34:23 -0700 | [diff] [blame] | 1586 | /* |
| 1587 | * Now, mem_cgroup_clear_mc() may call this function with NULL. |
| 1588 | * We check NULL in callee rather than caller. |
| 1589 | */ |
KAMEZAWA Hiroyuki | 4331f7d | 2012-03-21 16:34:26 -0700 | [diff] [blame] | 1590 | if (memcg) { |
| 1591 | atomic_dec(&memcg_moving); |
KAMEZAWA Hiroyuki | 619d094 | 2012-03-21 16:34:23 -0700 | [diff] [blame] | 1592 | atomic_dec(&memcg->moving_account); |
KAMEZAWA Hiroyuki | 4331f7d | 2012-03-21 16:34:26 -0700 | [diff] [blame] | 1593 | } |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1594 | } |
KAMEZAWA Hiroyuki | 619d094 | 2012-03-21 16:34:23 -0700 | [diff] [blame] | 1595 | |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1596 | /* |
Qiang Huang | bdcbb65 | 2014-06-04 16:08:21 -0700 | [diff] [blame] | 1597 | * A routine for checking "mem" is under move_account() or not. |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1598 | * |
Qiang Huang | bdcbb65 | 2014-06-04 16:08:21 -0700 | [diff] [blame] | 1599 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
| 1600 | * moving cgroups. This is for waiting at high-memory pressure |
| 1601 | * caused by "move". |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1602 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1603 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1604 | { |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 1605 | struct mem_cgroup *from; |
| 1606 | struct mem_cgroup *to; |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1607 | bool ret = false; |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 1608 | /* |
| 1609 | * Unlike task_move routines, we access mc.to, mc.from not under |
| 1610 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. |
| 1611 | */ |
| 1612 | spin_lock(&mc.lock); |
| 1613 | from = mc.from; |
| 1614 | to = mc.to; |
| 1615 | if (!from) |
| 1616 | goto unlock; |
Michal Hocko | 3e92041 | 2011-07-26 16:08:29 -0700 | [diff] [blame] | 1617 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1618 | ret = mem_cgroup_same_or_subtree(memcg, from) |
| 1619 | || mem_cgroup_same_or_subtree(memcg, to); |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 1620 | unlock: |
| 1621 | spin_unlock(&mc.lock); |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1622 | return ret; |
| 1623 | } |
| 1624 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1625 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1626 | { |
| 1627 | if (mc.moving_task && current != mc.moving_task) { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1628 | if (mem_cgroup_under_move(memcg)) { |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1629 | DEFINE_WAIT(wait); |
| 1630 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); |
| 1631 | /* moving charge context might have finished. */ |
| 1632 | if (mc.moving_task) |
| 1633 | schedule(); |
| 1634 | finish_wait(&mc.waitq, &wait); |
| 1635 | return true; |
| 1636 | } |
| 1637 | } |
| 1638 | return false; |
| 1639 | } |
| 1640 | |
KAMEZAWA Hiroyuki | 312734c0 | 2012-03-21 16:34:24 -0700 | [diff] [blame] | 1641 | /* |
| 1642 | * Take this lock when |
| 1643 | * - a code tries to modify page's memcg while it's USED. |
| 1644 | * - a code tries to modify page state accounting in a memcg. |
KAMEZAWA Hiroyuki | 312734c0 | 2012-03-21 16:34:24 -0700 | [diff] [blame] | 1645 | */ |
| 1646 | static void move_lock_mem_cgroup(struct mem_cgroup *memcg, |
| 1647 | unsigned long *flags) |
| 1648 | { |
| 1649 | spin_lock_irqsave(&memcg->move_lock, *flags); |
| 1650 | } |
| 1651 | |
| 1652 | static void move_unlock_mem_cgroup(struct mem_cgroup *memcg, |
| 1653 | unsigned long *flags) |
| 1654 | { |
| 1655 | spin_unlock_irqrestore(&memcg->move_lock, *flags); |
| 1656 | } |
| 1657 | |
Sha Zhengju | 58cf188 | 2013-02-22 16:32:05 -0800 | [diff] [blame] | 1658 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1659 | /** |
Sha Zhengju | 58cf188 | 2013-02-22 16:32:05 -0800 | [diff] [blame] | 1660 | * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1661 | * @memcg: The memory cgroup that went over limit |
| 1662 | * @p: Task that is going to be killed |
| 1663 | * |
| 1664 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is |
| 1665 | * enabled |
| 1666 | */ |
| 1667 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) |
| 1668 | { |
Tejun Heo | e61734c | 2014-02-12 09:29:50 -0500 | [diff] [blame] | 1669 | /* oom_info_lock ensures that parallel ooms do not interleave */ |
Michal Hocko | 08088cb | 2014-02-25 15:01:44 -0800 | [diff] [blame] | 1670 | static DEFINE_MUTEX(oom_info_lock); |
Sha Zhengju | 58cf188 | 2013-02-22 16:32:05 -0800 | [diff] [blame] | 1671 | struct mem_cgroup *iter; |
| 1672 | unsigned int i; |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1673 | |
Sha Zhengju | 58cf188 | 2013-02-22 16:32:05 -0800 | [diff] [blame] | 1674 | if (!p) |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1675 | return; |
| 1676 | |
Michal Hocko | 08088cb | 2014-02-25 15:01:44 -0800 | [diff] [blame] | 1677 | mutex_lock(&oom_info_lock); |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1678 | rcu_read_lock(); |
| 1679 | |
Tejun Heo | e61734c | 2014-02-12 09:29:50 -0500 | [diff] [blame] | 1680 | pr_info("Task in "); |
| 1681 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); |
| 1682 | pr_info(" killed as a result of limit of "); |
| 1683 | pr_cont_cgroup_path(memcg->css.cgroup); |
| 1684 | pr_info("\n"); |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1685 | |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1686 | rcu_read_unlock(); |
| 1687 | |
Andrew Morton | d045197 | 2013-02-22 16:32:06 -0800 | [diff] [blame] | 1688 | pr_info("memory: usage %llukB, limit %llukB, failcnt %llu\n", |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1689 | res_counter_read_u64(&memcg->res, RES_USAGE) >> 10, |
| 1690 | res_counter_read_u64(&memcg->res, RES_LIMIT) >> 10, |
| 1691 | res_counter_read_u64(&memcg->res, RES_FAILCNT)); |
Andrew Morton | d045197 | 2013-02-22 16:32:06 -0800 | [diff] [blame] | 1692 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %llu\n", |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1693 | res_counter_read_u64(&memcg->memsw, RES_USAGE) >> 10, |
| 1694 | res_counter_read_u64(&memcg->memsw, RES_LIMIT) >> 10, |
| 1695 | res_counter_read_u64(&memcg->memsw, RES_FAILCNT)); |
Andrew Morton | d045197 | 2013-02-22 16:32:06 -0800 | [diff] [blame] | 1696 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %llu\n", |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 1697 | res_counter_read_u64(&memcg->kmem, RES_USAGE) >> 10, |
| 1698 | res_counter_read_u64(&memcg->kmem, RES_LIMIT) >> 10, |
| 1699 | res_counter_read_u64(&memcg->kmem, RES_FAILCNT)); |
Sha Zhengju | 58cf188 | 2013-02-22 16:32:05 -0800 | [diff] [blame] | 1700 | |
| 1701 | for_each_mem_cgroup_tree(iter, memcg) { |
Tejun Heo | e61734c | 2014-02-12 09:29:50 -0500 | [diff] [blame] | 1702 | pr_info("Memory cgroup stats for "); |
| 1703 | pr_cont_cgroup_path(iter->css.cgroup); |
Sha Zhengju | 58cf188 | 2013-02-22 16:32:05 -0800 | [diff] [blame] | 1704 | pr_cont(":"); |
| 1705 | |
| 1706 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
| 1707 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
| 1708 | continue; |
| 1709 | pr_cont(" %s:%ldKB", mem_cgroup_stat_names[i], |
| 1710 | K(mem_cgroup_read_stat(iter, i))); |
| 1711 | } |
| 1712 | |
| 1713 | for (i = 0; i < NR_LRU_LISTS; i++) |
| 1714 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], |
| 1715 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); |
| 1716 | |
| 1717 | pr_cont("\n"); |
| 1718 | } |
Michal Hocko | 08088cb | 2014-02-25 15:01:44 -0800 | [diff] [blame] | 1719 | mutex_unlock(&oom_info_lock); |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1720 | } |
| 1721 | |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 1722 | /* |
| 1723 | * This function returns the number of memcg under hierarchy tree. Returns |
| 1724 | * 1(self count) if no children. |
| 1725 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1726 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 1727 | { |
| 1728 | int num = 0; |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1729 | struct mem_cgroup *iter; |
| 1730 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1731 | for_each_mem_cgroup_tree(iter, memcg) |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1732 | num++; |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 1733 | return num; |
| 1734 | } |
| 1735 | |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1736 | /* |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 1737 | * Return the memory (and swap, if configured) limit for a memcg. |
| 1738 | */ |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 1739 | static u64 mem_cgroup_get_limit(struct mem_cgroup *memcg) |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 1740 | { |
| 1741 | u64 limit; |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 1742 | |
Johannes Weiner | f3e8eb7 | 2011-01-13 15:47:39 -0800 | [diff] [blame] | 1743 | limit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
Johannes Weiner | f3e8eb7 | 2011-01-13 15:47:39 -0800 | [diff] [blame] | 1744 | |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 1745 | /* |
Michal Hocko | 9a5a8f1 | 2012-11-16 14:14:49 -0800 | [diff] [blame] | 1746 | * Do not consider swap space if we cannot swap due to swappiness |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 1747 | */ |
Michal Hocko | 9a5a8f1 | 2012-11-16 14:14:49 -0800 | [diff] [blame] | 1748 | if (mem_cgroup_swappiness(memcg)) { |
| 1749 | u64 memsw; |
| 1750 | |
| 1751 | limit += total_swap_pages << PAGE_SHIFT; |
| 1752 | memsw = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
| 1753 | |
| 1754 | /* |
| 1755 | * If memsw is finite and limits the amount of swap space |
| 1756 | * available to this memcg, return that limit. |
| 1757 | */ |
| 1758 | limit = min(limit, memsw); |
| 1759 | } |
| 1760 | |
| 1761 | return limit; |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 1762 | } |
| 1763 | |
David Rientjes | 1996546 | 2012-12-11 16:00:26 -0800 | [diff] [blame] | 1764 | static void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
| 1765 | int order) |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 1766 | { |
| 1767 | struct mem_cgroup *iter; |
| 1768 | unsigned long chosen_points = 0; |
| 1769 | unsigned long totalpages; |
| 1770 | unsigned int points = 0; |
| 1771 | struct task_struct *chosen = NULL; |
| 1772 | |
David Rientjes | 876aafb | 2012-07-31 16:43:48 -0700 | [diff] [blame] | 1773 | /* |
David Rientjes | 465adcf | 2013-04-29 15:08:45 -0700 | [diff] [blame] | 1774 | * If current has a pending SIGKILL or is exiting, then automatically |
| 1775 | * select it. The goal is to allow it to allocate so that it may |
| 1776 | * quickly exit and free its memory. |
David Rientjes | 876aafb | 2012-07-31 16:43:48 -0700 | [diff] [blame] | 1777 | */ |
David Rientjes | 465adcf | 2013-04-29 15:08:45 -0700 | [diff] [blame] | 1778 | if (fatal_signal_pending(current) || current->flags & PF_EXITING) { |
David Rientjes | 876aafb | 2012-07-31 16:43:48 -0700 | [diff] [blame] | 1779 | set_thread_flag(TIF_MEMDIE); |
| 1780 | return; |
| 1781 | } |
| 1782 | |
| 1783 | check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL); |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 1784 | totalpages = mem_cgroup_get_limit(memcg) >> PAGE_SHIFT ? : 1; |
| 1785 | for_each_mem_cgroup_tree(iter, memcg) { |
Tejun Heo | 72ec702 | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 1786 | struct css_task_iter it; |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 1787 | struct task_struct *task; |
| 1788 | |
Tejun Heo | 72ec702 | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 1789 | css_task_iter_start(&iter->css, &it); |
| 1790 | while ((task = css_task_iter_next(&it))) { |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 1791 | switch (oom_scan_process_thread(task, totalpages, NULL, |
| 1792 | false)) { |
| 1793 | case OOM_SCAN_SELECT: |
| 1794 | if (chosen) |
| 1795 | put_task_struct(chosen); |
| 1796 | chosen = task; |
| 1797 | chosen_points = ULONG_MAX; |
| 1798 | get_task_struct(chosen); |
| 1799 | /* fall through */ |
| 1800 | case OOM_SCAN_CONTINUE: |
| 1801 | continue; |
| 1802 | case OOM_SCAN_ABORT: |
Tejun Heo | 72ec702 | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 1803 | css_task_iter_end(&it); |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 1804 | mem_cgroup_iter_break(memcg, iter); |
| 1805 | if (chosen) |
| 1806 | put_task_struct(chosen); |
| 1807 | return; |
| 1808 | case OOM_SCAN_OK: |
| 1809 | break; |
| 1810 | }; |
| 1811 | points = oom_badness(task, memcg, NULL, totalpages); |
David Rientjes | d49ad93 | 2014-01-23 15:53:34 -0800 | [diff] [blame] | 1812 | if (!points || points < chosen_points) |
| 1813 | continue; |
| 1814 | /* Prefer thread group leaders for display purposes */ |
| 1815 | if (points == chosen_points && |
| 1816 | thread_group_leader(chosen)) |
| 1817 | continue; |
| 1818 | |
| 1819 | if (chosen) |
| 1820 | put_task_struct(chosen); |
| 1821 | chosen = task; |
| 1822 | chosen_points = points; |
| 1823 | get_task_struct(chosen); |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 1824 | } |
Tejun Heo | 72ec702 | 2013-08-08 20:11:26 -0400 | [diff] [blame] | 1825 | css_task_iter_end(&it); |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 1826 | } |
| 1827 | |
| 1828 | if (!chosen) |
| 1829 | return; |
| 1830 | points = chosen_points * 1000 / totalpages; |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 1831 | oom_kill_process(chosen, gfp_mask, order, points, totalpages, memcg, |
| 1832 | NULL, "Memory cgroup out of memory"); |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 1833 | } |
| 1834 | |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 1835 | static unsigned long mem_cgroup_reclaim(struct mem_cgroup *memcg, |
| 1836 | gfp_t gfp_mask, |
| 1837 | unsigned long flags) |
| 1838 | { |
| 1839 | unsigned long total = 0; |
| 1840 | bool noswap = false; |
| 1841 | int loop; |
| 1842 | |
| 1843 | if (flags & MEM_CGROUP_RECLAIM_NOSWAP) |
| 1844 | noswap = true; |
| 1845 | if (!(flags & MEM_CGROUP_RECLAIM_SHRINK) && memcg->memsw_is_minimum) |
| 1846 | noswap = true; |
| 1847 | |
| 1848 | for (loop = 0; loop < MEM_CGROUP_MAX_RECLAIM_LOOPS; loop++) { |
| 1849 | if (loop) |
| 1850 | drain_all_stock_async(memcg); |
| 1851 | total += try_to_free_mem_cgroup_pages(memcg, gfp_mask, noswap); |
| 1852 | /* |
| 1853 | * Allow limit shrinkers, which are triggered directly |
| 1854 | * by userspace, to catch signals and stop reclaim |
| 1855 | * after minimal progress, regardless of the margin. |
| 1856 | */ |
| 1857 | if (total && (flags & MEM_CGROUP_RECLAIM_SHRINK)) |
| 1858 | break; |
| 1859 | if (mem_cgroup_margin(memcg)) |
| 1860 | break; |
| 1861 | /* |
| 1862 | * If nothing was reclaimed after two attempts, there |
| 1863 | * may be no reclaimable pages in this hierarchy. |
| 1864 | */ |
| 1865 | if (loop && !total) |
| 1866 | break; |
| 1867 | } |
| 1868 | return total; |
| 1869 | } |
| 1870 | |
KAMEZAWA Hiroyuki | 4d0c066 | 2011-07-08 15:39:42 -0700 | [diff] [blame] | 1871 | /** |
| 1872 | * test_mem_cgroup_node_reclaimable |
Wanpeng Li | dad7557 | 2012-06-20 12:53:01 -0700 | [diff] [blame] | 1873 | * @memcg: the target memcg |
KAMEZAWA Hiroyuki | 4d0c066 | 2011-07-08 15:39:42 -0700 | [diff] [blame] | 1874 | * @nid: the node ID to be checked. |
| 1875 | * @noswap : specify true here if the user wants flle only information. |
| 1876 | * |
| 1877 | * This function returns whether the specified memcg contains any |
| 1878 | * reclaimable pages on a node. Returns true if there are any reclaimable |
| 1879 | * pages in the node. |
| 1880 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1881 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
KAMEZAWA Hiroyuki | 4d0c066 | 2011-07-08 15:39:42 -0700 | [diff] [blame] | 1882 | int nid, bool noswap) |
| 1883 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1884 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
KAMEZAWA Hiroyuki | 4d0c066 | 2011-07-08 15:39:42 -0700 | [diff] [blame] | 1885 | return true; |
| 1886 | if (noswap || !total_swap_pages) |
| 1887 | return false; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1888 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
KAMEZAWA Hiroyuki | 4d0c066 | 2011-07-08 15:39:42 -0700 | [diff] [blame] | 1889 | return true; |
| 1890 | return false; |
| 1891 | |
| 1892 | } |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 1893 | #if MAX_NUMNODES > 1 |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1894 | |
| 1895 | /* |
| 1896 | * Always updating the nodemask is not very good - even if we have an empty |
| 1897 | * list or the wrong list here, we can start from some node and traverse all |
| 1898 | * nodes based on the zonelist. So update the list loosely once per 10 secs. |
| 1899 | * |
| 1900 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1901 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1902 | { |
| 1903 | int nid; |
KAMEZAWA Hiroyuki | 453a9bf | 2011-07-08 15:39:43 -0700 | [diff] [blame] | 1904 | /* |
| 1905 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET |
| 1906 | * pagein/pageout changes since the last update. |
| 1907 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1908 | if (!atomic_read(&memcg->numainfo_events)) |
KAMEZAWA Hiroyuki | 453a9bf | 2011-07-08 15:39:43 -0700 | [diff] [blame] | 1909 | return; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1910 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1911 | return; |
| 1912 | |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1913 | /* make a nodemask where this memcg uses memory from */ |
Lai Jiangshan | 31aaea4 | 2012-12-12 13:51:27 -0800 | [diff] [blame] | 1914 | memcg->scan_nodes = node_states[N_MEMORY]; |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1915 | |
Lai Jiangshan | 31aaea4 | 2012-12-12 13:51:27 -0800 | [diff] [blame] | 1916 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1917 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1918 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
| 1919 | node_clear(nid, memcg->scan_nodes); |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1920 | } |
KAMEZAWA Hiroyuki | 453a9bf | 2011-07-08 15:39:43 -0700 | [diff] [blame] | 1921 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1922 | atomic_set(&memcg->numainfo_events, 0); |
| 1923 | atomic_set(&memcg->numainfo_updating, 0); |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1924 | } |
| 1925 | |
| 1926 | /* |
| 1927 | * Selecting a node where we start reclaim from. Because what we need is just |
| 1928 | * reducing usage counter, start from anywhere is O,K. Considering |
| 1929 | * memory reclaim from current node, there are pros. and cons. |
| 1930 | * |
| 1931 | * Freeing memory from current node means freeing memory from a node which |
| 1932 | * we'll use or we've used. So, it may make LRU bad. And if several threads |
| 1933 | * hit limits, it will see a contention on a node. But freeing from remote |
| 1934 | * node means more costs for memory reclaim because of memory latency. |
| 1935 | * |
| 1936 | * Now, we use round-robin. Better algorithm is welcomed. |
| 1937 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1938 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1939 | { |
| 1940 | int node; |
| 1941 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1942 | mem_cgroup_may_update_nodemask(memcg); |
| 1943 | node = memcg->last_scanned_node; |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1944 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1945 | node = next_node(node, memcg->scan_nodes); |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1946 | if (node == MAX_NUMNODES) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1947 | node = first_node(memcg->scan_nodes); |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1948 | /* |
| 1949 | * We call this when we hit limit, not when pages are added to LRU. |
| 1950 | * No LRU may hold pages because all pages are UNEVICTABLE or |
| 1951 | * memcg is too small and all pages are not on LRU. In that case, |
| 1952 | * we use curret node. |
| 1953 | */ |
| 1954 | if (unlikely(node == MAX_NUMNODES)) |
| 1955 | node = numa_node_id(); |
| 1956 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1957 | memcg->last_scanned_node = node; |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1958 | return node; |
| 1959 | } |
| 1960 | |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 1961 | /* |
| 1962 | * Check all nodes whether it contains reclaimable pages or not. |
| 1963 | * For quick scan, we make use of scan_nodes. This will allow us to skip |
| 1964 | * unused nodes. But scan_nodes is lazily updated and may not cotain |
| 1965 | * enough new information. We need to do double check. |
| 1966 | */ |
| 1967 | static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) |
| 1968 | { |
| 1969 | int nid; |
| 1970 | |
| 1971 | /* |
| 1972 | * quick check...making use of scan_node. |
| 1973 | * We can skip unused nodes. |
| 1974 | */ |
| 1975 | if (!nodes_empty(memcg->scan_nodes)) { |
| 1976 | for (nid = first_node(memcg->scan_nodes); |
| 1977 | nid < MAX_NUMNODES; |
| 1978 | nid = next_node(nid, memcg->scan_nodes)) { |
| 1979 | |
| 1980 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) |
| 1981 | return true; |
| 1982 | } |
| 1983 | } |
| 1984 | /* |
| 1985 | * Check rest of nodes. |
| 1986 | */ |
| 1987 | for_each_node_state(nid, N_MEMORY) { |
| 1988 | if (node_isset(nid, memcg->scan_nodes)) |
| 1989 | continue; |
| 1990 | if (test_mem_cgroup_node_reclaimable(memcg, nid, noswap)) |
| 1991 | return true; |
| 1992 | } |
| 1993 | return false; |
| 1994 | } |
| 1995 | |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1996 | #else |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1997 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1998 | { |
| 1999 | return 0; |
| 2000 | } |
KAMEZAWA Hiroyuki | 4d0c066 | 2011-07-08 15:39:42 -0700 | [diff] [blame] | 2001 | |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 2002 | static bool mem_cgroup_reclaimable(struct mem_cgroup *memcg, bool noswap) |
| 2003 | { |
| 2004 | return test_mem_cgroup_node_reclaimable(memcg, 0, noswap); |
| 2005 | } |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 2006 | #endif |
| 2007 | |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 2008 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
| 2009 | struct zone *zone, |
| 2010 | gfp_t gfp_mask, |
| 2011 | unsigned long *total_scanned) |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 2012 | { |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 2013 | struct mem_cgroup *victim = NULL; |
| 2014 | int total = 0; |
| 2015 | int loop = 0; |
| 2016 | unsigned long excess; |
| 2017 | unsigned long nr_scanned; |
| 2018 | struct mem_cgroup_reclaim_cookie reclaim = { |
| 2019 | .zone = zone, |
| 2020 | .priority = 0, |
| 2021 | }; |
Johannes Weiner | 9d11ea9 | 2011-03-23 16:42:21 -0700 | [diff] [blame] | 2022 | |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 2023 | excess = res_counter_soft_limit_excess(&root_memcg->res) >> PAGE_SHIFT; |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 2024 | |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 2025 | while (1) { |
| 2026 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); |
| 2027 | if (!victim) { |
| 2028 | loop++; |
| 2029 | if (loop >= 2) { |
| 2030 | /* |
| 2031 | * If we have not been able to reclaim |
| 2032 | * anything, it might because there are |
| 2033 | * no reclaimable pages under this hierarchy |
| 2034 | */ |
| 2035 | if (!total) |
| 2036 | break; |
| 2037 | /* |
| 2038 | * We want to do more targeted reclaim. |
| 2039 | * excess >> 2 is not to excessive so as to |
| 2040 | * reclaim too much, nor too less that we keep |
| 2041 | * coming back to reclaim from this cgroup |
| 2042 | */ |
| 2043 | if (total >= (excess >> 2) || |
| 2044 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) |
| 2045 | break; |
| 2046 | } |
| 2047 | continue; |
| 2048 | } |
| 2049 | if (!mem_cgroup_reclaimable(victim, false)) |
| 2050 | continue; |
| 2051 | total += mem_cgroup_shrink_node_zone(victim, gfp_mask, false, |
| 2052 | zone, &nr_scanned); |
| 2053 | *total_scanned += nr_scanned; |
| 2054 | if (!res_counter_soft_limit_excess(&root_memcg->res)) |
| 2055 | break; |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 2056 | } |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 2057 | mem_cgroup_iter_break(root_memcg, victim); |
| 2058 | return total; |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 2059 | } |
| 2060 | |
Johannes Weiner | 0056f4e | 2013-10-31 16:34:14 -0700 | [diff] [blame] | 2061 | #ifdef CONFIG_LOCKDEP |
| 2062 | static struct lockdep_map memcg_oom_lock_dep_map = { |
| 2063 | .name = "memcg_oom_lock", |
| 2064 | }; |
| 2065 | #endif |
| 2066 | |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 2067 | static DEFINE_SPINLOCK(memcg_oom_lock); |
| 2068 | |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 2069 | /* |
| 2070 | * Check OOM-Killer is already running under our hierarchy. |
| 2071 | * If someone is running, return false. |
| 2072 | */ |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 2073 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 2074 | { |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 2075 | struct mem_cgroup *iter, *failed = NULL; |
KAMEZAWA Hiroyuki | a636b32 | 2009-01-07 18:08:08 -0800 | [diff] [blame] | 2076 | |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 2077 | spin_lock(&memcg_oom_lock); |
| 2078 | |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 2079 | for_each_mem_cgroup_tree(iter, memcg) { |
Johannes Weiner | 23751be | 2011-08-25 15:59:16 -0700 | [diff] [blame] | 2080 | if (iter->oom_lock) { |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 2081 | /* |
| 2082 | * this subtree of our hierarchy is already locked |
| 2083 | * so we cannot give a lock. |
| 2084 | */ |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 2085 | failed = iter; |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 2086 | mem_cgroup_iter_break(memcg, iter); |
| 2087 | break; |
Johannes Weiner | 23751be | 2011-08-25 15:59:16 -0700 | [diff] [blame] | 2088 | } else |
| 2089 | iter->oom_lock = true; |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 2090 | } |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 2091 | |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 2092 | if (failed) { |
| 2093 | /* |
| 2094 | * OK, we failed to lock the whole subtree so we have |
| 2095 | * to clean up what we set up to the failing subtree |
| 2096 | */ |
| 2097 | for_each_mem_cgroup_tree(iter, memcg) { |
| 2098 | if (iter == failed) { |
| 2099 | mem_cgroup_iter_break(memcg, iter); |
| 2100 | break; |
| 2101 | } |
| 2102 | iter->oom_lock = false; |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 2103 | } |
Johannes Weiner | 0056f4e | 2013-10-31 16:34:14 -0700 | [diff] [blame] | 2104 | } else |
| 2105 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 2106 | |
| 2107 | spin_unlock(&memcg_oom_lock); |
| 2108 | |
| 2109 | return !failed; |
KAMEZAWA Hiroyuki | a636b32 | 2009-01-07 18:08:08 -0800 | [diff] [blame] | 2110 | } |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 2111 | |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 2112 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 2113 | { |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 2114 | struct mem_cgroup *iter; |
| 2115 | |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 2116 | spin_lock(&memcg_oom_lock); |
Johannes Weiner | 0056f4e | 2013-10-31 16:34:14 -0700 | [diff] [blame] | 2117 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2118 | for_each_mem_cgroup_tree(iter, memcg) |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 2119 | iter->oom_lock = false; |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 2120 | spin_unlock(&memcg_oom_lock); |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 2121 | } |
| 2122 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2123 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 2124 | { |
| 2125 | struct mem_cgroup *iter; |
| 2126 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2127 | for_each_mem_cgroup_tree(iter, memcg) |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 2128 | atomic_inc(&iter->under_oom); |
| 2129 | } |
| 2130 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2131 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 2132 | { |
| 2133 | struct mem_cgroup *iter; |
| 2134 | |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 2135 | /* |
| 2136 | * When a new child is created while the hierarchy is under oom, |
| 2137 | * mem_cgroup_oom_lock() may not be called. We have to use |
| 2138 | * atomic_add_unless() here. |
| 2139 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2140 | for_each_mem_cgroup_tree(iter, memcg) |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 2141 | atomic_add_unless(&iter->under_oom, -1, 0); |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 2142 | } |
| 2143 | |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 2144 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
| 2145 | |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 2146 | struct oom_wait_info { |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 2147 | struct mem_cgroup *memcg; |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 2148 | wait_queue_t wait; |
| 2149 | }; |
| 2150 | |
| 2151 | static int memcg_oom_wake_function(wait_queue_t *wait, |
| 2152 | unsigned mode, int sync, void *arg) |
| 2153 | { |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 2154 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
| 2155 | struct mem_cgroup *oom_wait_memcg; |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 2156 | struct oom_wait_info *oom_wait_info; |
| 2157 | |
| 2158 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 2159 | oom_wait_memcg = oom_wait_info->memcg; |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 2160 | |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 2161 | /* |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 2162 | * Both of oom_wait_info->memcg and wake_memcg are stable under us. |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 2163 | * Then we can use css_is_ancestor without taking care of RCU. |
| 2164 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2165 | if (!mem_cgroup_same_or_subtree(oom_wait_memcg, wake_memcg) |
| 2166 | && !mem_cgroup_same_or_subtree(wake_memcg, oom_wait_memcg)) |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 2167 | return 0; |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 2168 | return autoremove_wake_function(wait, mode, sync, arg); |
| 2169 | } |
| 2170 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2171 | static void memcg_wakeup_oom(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 2172 | { |
Johannes Weiner | 3812c8c | 2013-09-12 15:13:44 -0700 | [diff] [blame] | 2173 | atomic_inc(&memcg->oom_wakeups); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2174 | /* for filtering, pass "memcg" as argument. */ |
| 2175 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 2176 | } |
| 2177 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2178 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 2179 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2180 | if (memcg && atomic_read(&memcg->under_oom)) |
| 2181 | memcg_wakeup_oom(memcg); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 2182 | } |
| 2183 | |
Johannes Weiner | 3812c8c | 2013-09-12 15:13:44 -0700 | [diff] [blame] | 2184 | 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] | 2185 | { |
Johannes Weiner | 3812c8c | 2013-09-12 15:13:44 -0700 | [diff] [blame] | 2186 | if (!current->memcg_oom.may_oom) |
| 2187 | return; |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 2188 | /* |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 2189 | * We are in the middle of the charge context here, so we |
| 2190 | * don't want to block when potentially sitting on a callstack |
| 2191 | * that holds all kinds of filesystem and mm locks. |
| 2192 | * |
| 2193 | * Also, the caller may handle a failed allocation gracefully |
| 2194 | * (like optional page cache readahead) and so an OOM killer |
| 2195 | * invocation might not even be necessary. |
| 2196 | * |
| 2197 | * That's why we don't do anything here except remember the |
| 2198 | * OOM context and then deal with it at the end of the page |
| 2199 | * fault when the stack is unwound, the locks are released, |
| 2200 | * and when we know whether the fault was overall successful. |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 2201 | */ |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 2202 | css_get(&memcg->css); |
| 2203 | current->memcg_oom.memcg = memcg; |
| 2204 | current->memcg_oom.gfp_mask = mask; |
| 2205 | current->memcg_oom.order = order; |
| 2206 | } |
| 2207 | |
| 2208 | /** |
| 2209 | * mem_cgroup_oom_synchronize - complete memcg OOM handling |
| 2210 | * @handle: actually kill/wait or just clean up the OOM state |
| 2211 | * |
| 2212 | * This has to be called at the end of a page fault if the memcg OOM |
| 2213 | * handler was enabled. |
| 2214 | * |
| 2215 | * Memcg supports userspace OOM handling where failed allocations must |
| 2216 | * sleep on a waitqueue until the userspace task resolves the |
| 2217 | * situation. Sleeping directly in the charge context with all kinds |
| 2218 | * of locks held is not a good idea, instead we remember an OOM state |
| 2219 | * in the task and mem_cgroup_oom_synchronize() has to be called at |
| 2220 | * the end of the page fault to complete the OOM handling. |
| 2221 | * |
| 2222 | * Returns %true if an ongoing memcg OOM situation was detected and |
| 2223 | * completed, %false otherwise. |
| 2224 | */ |
| 2225 | bool mem_cgroup_oom_synchronize(bool handle) |
| 2226 | { |
| 2227 | struct mem_cgroup *memcg = current->memcg_oom.memcg; |
| 2228 | struct oom_wait_info owait; |
| 2229 | bool locked; |
| 2230 | |
| 2231 | /* OOM is global, do not handle */ |
| 2232 | if (!memcg) |
| 2233 | return false; |
| 2234 | |
| 2235 | if (!handle) |
| 2236 | goto cleanup; |
| 2237 | |
| 2238 | owait.memcg = memcg; |
| 2239 | owait.wait.flags = 0; |
| 2240 | owait.wait.func = memcg_oom_wake_function; |
| 2241 | owait.wait.private = current; |
| 2242 | INIT_LIST_HEAD(&owait.wait.task_list); |
| 2243 | |
| 2244 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 2245 | mem_cgroup_mark_under_oom(memcg); |
| 2246 | |
| 2247 | locked = mem_cgroup_oom_trylock(memcg); |
| 2248 | |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 2249 | if (locked) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2250 | mem_cgroup_oom_notify(memcg); |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 2251 | |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 2252 | if (locked && !memcg->oom_kill_disable) { |
| 2253 | mem_cgroup_unmark_under_oom(memcg); |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 2254 | finish_wait(&memcg_oom_waitq, &owait.wait); |
| 2255 | mem_cgroup_out_of_memory(memcg, current->memcg_oom.gfp_mask, |
| 2256 | current->memcg_oom.order); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 2257 | } else { |
Johannes Weiner | 3812c8c | 2013-09-12 15:13:44 -0700 | [diff] [blame] | 2258 | schedule(); |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 2259 | mem_cgroup_unmark_under_oom(memcg); |
| 2260 | finish_wait(&memcg_oom_waitq, &owait.wait); |
| 2261 | } |
| 2262 | |
| 2263 | if (locked) { |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 2264 | mem_cgroup_oom_unlock(memcg); |
| 2265 | /* |
| 2266 | * There is no guarantee that an OOM-lock contender |
| 2267 | * sees the wakeups triggered by the OOM kill |
| 2268 | * uncharges. Wake any sleepers explicitely. |
| 2269 | */ |
| 2270 | memcg_oom_recover(memcg); |
| 2271 | } |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 2272 | cleanup: |
| 2273 | current->memcg_oom.memcg = NULL; |
Johannes Weiner | 3812c8c | 2013-09-12 15:13:44 -0700 | [diff] [blame] | 2274 | css_put(&memcg->css); |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 2275 | return true; |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 2276 | } |
| 2277 | |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 2278 | /* |
Qiang Huang | b5ffc85 | 2014-06-04 16:08:22 -0700 | [diff] [blame^] | 2279 | * Used to update mapped file or writeback or other statistics. |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 2280 | * |
| 2281 | * Notes: Race condition |
| 2282 | * |
Qiang Huang | b5ffc85 | 2014-06-04 16:08:22 -0700 | [diff] [blame^] | 2283 | * We usually use lock_page_cgroup() for accessing page_cgroup member but |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 2284 | * it tends to be costly. But considering some conditions, we doesn't need |
| 2285 | * to do so _always_. |
| 2286 | * |
| 2287 | * Considering "charge", lock_page_cgroup() is not required because all |
| 2288 | * file-stat operations happen after a page is attached to radix-tree. There |
| 2289 | * are no race with "charge". |
| 2290 | * |
| 2291 | * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup |
| 2292 | * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even |
| 2293 | * if there are race with "uncharge". Statistics itself is properly handled |
| 2294 | * by flags. |
| 2295 | * |
| 2296 | * Considering "move", this is an only case we see a race. To make the race |
Qiang Huang | b5ffc85 | 2014-06-04 16:08:22 -0700 | [diff] [blame^] | 2297 | * small, we check memcg->moving_account and detect there are possibility |
| 2298 | * of race or not. If there is, we take a lock. |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 2299 | */ |
KAMEZAWA Hiroyuki | 26174ef | 2010-10-27 15:33:43 -0700 | [diff] [blame] | 2300 | |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 2301 | void __mem_cgroup_begin_update_page_stat(struct page *page, |
| 2302 | bool *locked, unsigned long *flags) |
| 2303 | { |
| 2304 | struct mem_cgroup *memcg; |
| 2305 | struct page_cgroup *pc; |
| 2306 | |
| 2307 | pc = lookup_page_cgroup(page); |
| 2308 | again: |
| 2309 | memcg = pc->mem_cgroup; |
| 2310 | if (unlikely(!memcg || !PageCgroupUsed(pc))) |
| 2311 | return; |
| 2312 | /* |
| 2313 | * If this memory cgroup is not under account moving, we don't |
Wanpeng Li | da92c47 | 2012-07-31 16:43:26 -0700 | [diff] [blame] | 2314 | * need to take move_lock_mem_cgroup(). Because we already hold |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 2315 | * rcu_read_lock(), any calls to move_account will be delayed until |
Qiang Huang | bdcbb65 | 2014-06-04 16:08:21 -0700 | [diff] [blame] | 2316 | * rcu_read_unlock(). |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 2317 | */ |
Qiang Huang | bdcbb65 | 2014-06-04 16:08:21 -0700 | [diff] [blame] | 2318 | VM_BUG_ON(!rcu_read_lock_held()); |
| 2319 | if (atomic_read(&memcg->moving_account) <= 0) |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 2320 | return; |
| 2321 | |
| 2322 | move_lock_mem_cgroup(memcg, flags); |
| 2323 | if (memcg != pc->mem_cgroup || !PageCgroupUsed(pc)) { |
| 2324 | move_unlock_mem_cgroup(memcg, flags); |
| 2325 | goto again; |
| 2326 | } |
| 2327 | *locked = true; |
| 2328 | } |
| 2329 | |
| 2330 | void __mem_cgroup_end_update_page_stat(struct page *page, unsigned long *flags) |
| 2331 | { |
| 2332 | struct page_cgroup *pc = lookup_page_cgroup(page); |
| 2333 | |
| 2334 | /* |
| 2335 | * It's guaranteed that pc->mem_cgroup never changes while |
| 2336 | * lock is held because a routine modifies pc->mem_cgroup |
Wanpeng Li | da92c47 | 2012-07-31 16:43:26 -0700 | [diff] [blame] | 2337 | * should take move_lock_mem_cgroup(). |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 2338 | */ |
| 2339 | move_unlock_mem_cgroup(pc->mem_cgroup, flags); |
| 2340 | } |
| 2341 | |
Greg Thelen | 2a7106f | 2011-01-13 15:47:37 -0800 | [diff] [blame] | 2342 | void mem_cgroup_update_page_stat(struct page *page, |
Sha Zhengju | 68b4876 | 2013-09-12 15:13:50 -0700 | [diff] [blame] | 2343 | enum mem_cgroup_stat_index idx, int val) |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 2344 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2345 | struct mem_cgroup *memcg; |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 2346 | struct page_cgroup *pc = lookup_page_cgroup(page); |
KAMEZAWA Hiroyuki | dbd4ea7 | 2011-01-13 15:47:38 -0800 | [diff] [blame] | 2347 | unsigned long uninitialized_var(flags); |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 2348 | |
Johannes Weiner | cfa4494 | 2012-01-12 17:18:38 -0800 | [diff] [blame] | 2349 | if (mem_cgroup_disabled()) |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 2350 | return; |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 2351 | |
Sha Zhengju | 658b72c | 2013-09-12 15:13:52 -0700 | [diff] [blame] | 2352 | VM_BUG_ON(!rcu_read_lock_held()); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2353 | memcg = pc->mem_cgroup; |
| 2354 | if (unlikely(!memcg || !PageCgroupUsed(pc))) |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 2355 | return; |
KAMEZAWA Hiroyuki | 26174ef | 2010-10-27 15:33:43 -0700 | [diff] [blame] | 2356 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2357 | this_cpu_add(memcg->stat->count[idx], val); |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 2358 | } |
KAMEZAWA Hiroyuki | 26174ef | 2010-10-27 15:33:43 -0700 | [diff] [blame] | 2359 | |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2360 | /* |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2361 | * size of first charge trial. "32" comes from vmscan.c's magic value. |
| 2362 | * TODO: maybe necessary to use big numbers in big irons. |
| 2363 | */ |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 2364 | #define CHARGE_BATCH 32U |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2365 | struct memcg_stock_pcp { |
| 2366 | struct mem_cgroup *cached; /* this never be root cgroup */ |
Johannes Weiner | 11c9ea4 | 2011-03-23 16:42:34 -0700 | [diff] [blame] | 2367 | unsigned int nr_pages; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2368 | struct work_struct work; |
KAMEZAWA Hiroyuki | 26fe616 | 2011-06-15 15:08:45 -0700 | [diff] [blame] | 2369 | unsigned long flags; |
Kirill A. Shutemov | a0db00f | 2012-05-29 15:06:56 -0700 | [diff] [blame] | 2370 | #define FLUSHING_CACHED_CHARGE 0 |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2371 | }; |
| 2372 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); |
Michal Hocko | 9f50fad | 2011-08-09 11:56:26 +0200 | [diff] [blame] | 2373 | static DEFINE_MUTEX(percpu_charge_mutex); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2374 | |
Suleiman Souhlal | a0956d5 | 2012-12-18 14:21:36 -0800 | [diff] [blame] | 2375 | /** |
| 2376 | * consume_stock: Try to consume stocked charge on this cpu. |
| 2377 | * @memcg: memcg to consume from. |
| 2378 | * @nr_pages: how many pages to charge. |
| 2379 | * |
| 2380 | * The charges will only happen if @memcg matches the current cpu's memcg |
| 2381 | * stock, and at least @nr_pages are available in that stock. Failure to |
| 2382 | * service an allocation will refill the stock. |
| 2383 | * |
| 2384 | * returns true if successful, false otherwise. |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2385 | */ |
Suleiman Souhlal | a0956d5 | 2012-12-18 14:21:36 -0800 | [diff] [blame] | 2386 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2387 | { |
| 2388 | struct memcg_stock_pcp *stock; |
| 2389 | bool ret = true; |
| 2390 | |
Suleiman Souhlal | a0956d5 | 2012-12-18 14:21:36 -0800 | [diff] [blame] | 2391 | if (nr_pages > CHARGE_BATCH) |
| 2392 | return false; |
| 2393 | |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2394 | stock = &get_cpu_var(memcg_stock); |
Suleiman Souhlal | a0956d5 | 2012-12-18 14:21:36 -0800 | [diff] [blame] | 2395 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) |
| 2396 | stock->nr_pages -= nr_pages; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2397 | else /* need to call res_counter_charge */ |
| 2398 | ret = false; |
| 2399 | put_cpu_var(memcg_stock); |
| 2400 | return ret; |
| 2401 | } |
| 2402 | |
| 2403 | /* |
| 2404 | * Returns stocks cached in percpu to res_counter and reset cached information. |
| 2405 | */ |
| 2406 | static void drain_stock(struct memcg_stock_pcp *stock) |
| 2407 | { |
| 2408 | struct mem_cgroup *old = stock->cached; |
| 2409 | |
Johannes Weiner | 11c9ea4 | 2011-03-23 16:42:34 -0700 | [diff] [blame] | 2410 | if (stock->nr_pages) { |
| 2411 | unsigned long bytes = stock->nr_pages * PAGE_SIZE; |
| 2412 | |
| 2413 | res_counter_uncharge(&old->res, bytes); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2414 | if (do_swap_account) |
Johannes Weiner | 11c9ea4 | 2011-03-23 16:42:34 -0700 | [diff] [blame] | 2415 | res_counter_uncharge(&old->memsw, bytes); |
| 2416 | stock->nr_pages = 0; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2417 | } |
| 2418 | stock->cached = NULL; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2419 | } |
| 2420 | |
| 2421 | /* |
| 2422 | * This must be called under preempt disabled or must be called by |
| 2423 | * a thread which is pinned to local cpu. |
| 2424 | */ |
| 2425 | static void drain_local_stock(struct work_struct *dummy) |
| 2426 | { |
Christoph Lameter | 7c8e018 | 2014-06-04 16:07:56 -0700 | [diff] [blame] | 2427 | struct memcg_stock_pcp *stock = this_cpu_ptr(&memcg_stock); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2428 | drain_stock(stock); |
KAMEZAWA Hiroyuki | 26fe616 | 2011-06-15 15:08:45 -0700 | [diff] [blame] | 2429 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2430 | } |
| 2431 | |
Michal Hocko | e477749 | 2013-02-22 16:35:40 -0800 | [diff] [blame] | 2432 | static void __init memcg_stock_init(void) |
| 2433 | { |
| 2434 | int cpu; |
| 2435 | |
| 2436 | for_each_possible_cpu(cpu) { |
| 2437 | struct memcg_stock_pcp *stock = |
| 2438 | &per_cpu(memcg_stock, cpu); |
| 2439 | INIT_WORK(&stock->work, drain_local_stock); |
| 2440 | } |
| 2441 | } |
| 2442 | |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2443 | /* |
| 2444 | * Cache charges(val) which is from res_counter, to local per_cpu area. |
Greg Thelen | 320cc51 | 2010-03-15 15:27:28 +0100 | [diff] [blame] | 2445 | * This will be consumed by consume_stock() function, later. |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2446 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2447 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2448 | { |
| 2449 | struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock); |
| 2450 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2451 | if (stock->cached != memcg) { /* reset if necessary */ |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2452 | drain_stock(stock); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2453 | stock->cached = memcg; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2454 | } |
Johannes Weiner | 11c9ea4 | 2011-03-23 16:42:34 -0700 | [diff] [blame] | 2455 | stock->nr_pages += nr_pages; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2456 | put_cpu_var(memcg_stock); |
| 2457 | } |
| 2458 | |
| 2459 | /* |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2460 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
Michal Hocko | d38144b | 2011-07-26 16:08:28 -0700 | [diff] [blame] | 2461 | * of the hierarchy under it. sync flag says whether we should block |
| 2462 | * until the work is done. |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2463 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2464 | static void drain_all_stock(struct mem_cgroup *root_memcg, bool sync) |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2465 | { |
KAMEZAWA Hiroyuki | 26fe616 | 2011-06-15 15:08:45 -0700 | [diff] [blame] | 2466 | int cpu, curcpu; |
Michal Hocko | d38144b | 2011-07-26 16:08:28 -0700 | [diff] [blame] | 2467 | |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2468 | /* Notify other cpus that system-wide "drain" is running */ |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2469 | get_online_cpus(); |
Johannes Weiner | 5af12d0 | 2011-08-25 15:59:07 -0700 | [diff] [blame] | 2470 | curcpu = get_cpu(); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2471 | for_each_online_cpu(cpu) { |
| 2472 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2473 | struct mem_cgroup *memcg; |
KAMEZAWA Hiroyuki | 26fe616 | 2011-06-15 15:08:45 -0700 | [diff] [blame] | 2474 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2475 | memcg = stock->cached; |
| 2476 | if (!memcg || !stock->nr_pages) |
KAMEZAWA Hiroyuki | 26fe616 | 2011-06-15 15:08:45 -0700 | [diff] [blame] | 2477 | continue; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2478 | if (!mem_cgroup_same_or_subtree(root_memcg, memcg)) |
Michal Hocko | 3e92041 | 2011-07-26 16:08:29 -0700 | [diff] [blame] | 2479 | continue; |
Michal Hocko | d1a05b6 | 2011-07-26 16:08:27 -0700 | [diff] [blame] | 2480 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
| 2481 | if (cpu == curcpu) |
| 2482 | drain_local_stock(&stock->work); |
| 2483 | else |
| 2484 | schedule_work_on(cpu, &stock->work); |
| 2485 | } |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2486 | } |
Johannes Weiner | 5af12d0 | 2011-08-25 15:59:07 -0700 | [diff] [blame] | 2487 | put_cpu(); |
Michal Hocko | d38144b | 2011-07-26 16:08:28 -0700 | [diff] [blame] | 2488 | |
| 2489 | if (!sync) |
| 2490 | goto out; |
| 2491 | |
| 2492 | for_each_online_cpu(cpu) { |
| 2493 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); |
Michal Hocko | 9f50fad | 2011-08-09 11:56:26 +0200 | [diff] [blame] | 2494 | if (test_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) |
Michal Hocko | d38144b | 2011-07-26 16:08:28 -0700 | [diff] [blame] | 2495 | flush_work(&stock->work); |
| 2496 | } |
| 2497 | out: |
Andrew Morton | f894ffa | 2013-09-12 15:13:35 -0700 | [diff] [blame] | 2498 | put_online_cpus(); |
Michal Hocko | d38144b | 2011-07-26 16:08:28 -0700 | [diff] [blame] | 2499 | } |
| 2500 | |
| 2501 | /* |
| 2502 | * Tries to drain stocked charges in other cpus. This function is asynchronous |
| 2503 | * and just put a work per cpu for draining localy on each cpu. Caller can |
| 2504 | * expects some charges will be back to res_counter later but cannot wait for |
| 2505 | * it. |
| 2506 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2507 | static void drain_all_stock_async(struct mem_cgroup *root_memcg) |
Michal Hocko | d38144b | 2011-07-26 16:08:28 -0700 | [diff] [blame] | 2508 | { |
Michal Hocko | 9f50fad | 2011-08-09 11:56:26 +0200 | [diff] [blame] | 2509 | /* |
| 2510 | * If someone calls draining, avoid adding more kworker runs. |
| 2511 | */ |
| 2512 | if (!mutex_trylock(&percpu_charge_mutex)) |
| 2513 | return; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2514 | drain_all_stock(root_memcg, false); |
Michal Hocko | 9f50fad | 2011-08-09 11:56:26 +0200 | [diff] [blame] | 2515 | mutex_unlock(&percpu_charge_mutex); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2516 | } |
| 2517 | |
| 2518 | /* This is a synchronous drain interface. */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2519 | static void drain_all_stock_sync(struct mem_cgroup *root_memcg) |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2520 | { |
| 2521 | /* called when force_empty is called */ |
Michal Hocko | 9f50fad | 2011-08-09 11:56:26 +0200 | [diff] [blame] | 2522 | mutex_lock(&percpu_charge_mutex); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2523 | drain_all_stock(root_memcg, true); |
Michal Hocko | 9f50fad | 2011-08-09 11:56:26 +0200 | [diff] [blame] | 2524 | mutex_unlock(&percpu_charge_mutex); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2525 | } |
| 2526 | |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 2527 | /* |
| 2528 | * This function drains percpu counter value from DEAD cpu and |
| 2529 | * move it to local cpu. Note that this function can be preempted. |
| 2530 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2531 | static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *memcg, int cpu) |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 2532 | { |
| 2533 | int i; |
| 2534 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2535 | spin_lock(&memcg->pcp_counter_lock); |
Johannes Weiner | 6104621 | 2012-05-29 15:07:05 -0700 | [diff] [blame] | 2536 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2537 | long x = per_cpu(memcg->stat->count[i], cpu); |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 2538 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2539 | per_cpu(memcg->stat->count[i], cpu) = 0; |
| 2540 | memcg->nocpu_base.count[i] += x; |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 2541 | } |
Johannes Weiner | e9f8974 | 2011-03-23 16:42:37 -0700 | [diff] [blame] | 2542 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2543 | unsigned long x = per_cpu(memcg->stat->events[i], cpu); |
Johannes Weiner | e9f8974 | 2011-03-23 16:42:37 -0700 | [diff] [blame] | 2544 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2545 | per_cpu(memcg->stat->events[i], cpu) = 0; |
| 2546 | memcg->nocpu_base.events[i] += x; |
Johannes Weiner | e9f8974 | 2011-03-23 16:42:37 -0700 | [diff] [blame] | 2547 | } |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2548 | spin_unlock(&memcg->pcp_counter_lock); |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 2549 | } |
| 2550 | |
Paul Gortmaker | 0db0628 | 2013-06-19 14:53:51 -0400 | [diff] [blame] | 2551 | static int memcg_cpu_hotplug_callback(struct notifier_block *nb, |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2552 | unsigned long action, |
| 2553 | void *hcpu) |
| 2554 | { |
| 2555 | int cpu = (unsigned long)hcpu; |
| 2556 | struct memcg_stock_pcp *stock; |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 2557 | struct mem_cgroup *iter; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2558 | |
KAMEZAWA Hiroyuki | 619d094 | 2012-03-21 16:34:23 -0700 | [diff] [blame] | 2559 | if (action == CPU_ONLINE) |
KAMEZAWA Hiroyuki | 1489eba | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 2560 | return NOTIFY_OK; |
KAMEZAWA Hiroyuki | 1489eba | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 2561 | |
Kirill A. Shutemov | d833049 | 2012-04-12 12:49:11 -0700 | [diff] [blame] | 2562 | if (action != CPU_DEAD && action != CPU_DEAD_FROZEN) |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2563 | return NOTIFY_OK; |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 2564 | |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 2565 | for_each_mem_cgroup(iter) |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 2566 | mem_cgroup_drain_pcp_counter(iter, cpu); |
| 2567 | |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2568 | stock = &per_cpu(memcg_stock, cpu); |
| 2569 | drain_stock(stock); |
| 2570 | return NOTIFY_OK; |
| 2571 | } |
| 2572 | |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2573 | |
Johannes Weiner | 6d1fdc4 | 2014-04-07 15:37:45 -0700 | [diff] [blame] | 2574 | /* See mem_cgroup_try_charge() for details */ |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2575 | enum { |
| 2576 | CHARGE_OK, /* success */ |
| 2577 | CHARGE_RETRY, /* need to retry but retry is not bad */ |
| 2578 | CHARGE_NOMEM, /* we can't do more. return -ENOMEM */ |
| 2579 | CHARGE_WOULDBLOCK, /* GFP_WAIT wasn't set and no enough res. */ |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2580 | }; |
| 2581 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2582 | static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
Suleiman Souhlal | 4c9c535 | 2012-12-18 14:21:41 -0800 | [diff] [blame] | 2583 | unsigned int nr_pages, unsigned int min_pages, |
Johannes Weiner | 3812c8c | 2013-09-12 15:13:44 -0700 | [diff] [blame] | 2584 | bool invoke_oom) |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2585 | { |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 2586 | unsigned long csize = nr_pages * PAGE_SIZE; |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2587 | struct mem_cgroup *mem_over_limit; |
| 2588 | struct res_counter *fail_res; |
| 2589 | unsigned long flags = 0; |
| 2590 | int ret; |
| 2591 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2592 | ret = res_counter_charge(&memcg->res, csize, &fail_res); |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2593 | |
| 2594 | if (likely(!ret)) { |
| 2595 | if (!do_swap_account) |
| 2596 | return CHARGE_OK; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2597 | ret = res_counter_charge(&memcg->memsw, csize, &fail_res); |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2598 | if (likely(!ret)) |
| 2599 | return CHARGE_OK; |
| 2600 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2601 | res_counter_uncharge(&memcg->res, csize); |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2602 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, memsw); |
| 2603 | flags |= MEM_CGROUP_RECLAIM_NOSWAP; |
| 2604 | } else |
| 2605 | mem_over_limit = mem_cgroup_from_res_counter(fail_res, res); |
Johannes Weiner | 9221edb | 2011-02-01 15:52:42 -0800 | [diff] [blame] | 2606 | /* |
Johannes Weiner | 9221edb | 2011-02-01 15:52:42 -0800 | [diff] [blame] | 2607 | * Never reclaim on behalf of optional batching, retry with a |
| 2608 | * single page instead. |
| 2609 | */ |
Suleiman Souhlal | 4c9c535 | 2012-12-18 14:21:41 -0800 | [diff] [blame] | 2610 | if (nr_pages > min_pages) |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2611 | return CHARGE_RETRY; |
| 2612 | |
| 2613 | if (!(gfp_mask & __GFP_WAIT)) |
| 2614 | return CHARGE_WOULDBLOCK; |
| 2615 | |
Suleiman Souhlal | 4c9c535 | 2012-12-18 14:21:41 -0800 | [diff] [blame] | 2616 | if (gfp_mask & __GFP_NORETRY) |
| 2617 | return CHARGE_NOMEM; |
| 2618 | |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 2619 | ret = mem_cgroup_reclaim(mem_over_limit, gfp_mask, flags); |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 2620 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 2621 | return CHARGE_RETRY; |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2622 | /* |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 2623 | * Even though the limit is exceeded at this point, reclaim |
| 2624 | * may have been able to free some pages. Retry the charge |
| 2625 | * before killing the task. |
| 2626 | * |
| 2627 | * Only for regular pages, though: huge pages are rather |
| 2628 | * unlikely to succeed so close to the limit, and we fall back |
| 2629 | * to regular pages anyway in case of failure. |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2630 | */ |
Suleiman Souhlal | 4c9c535 | 2012-12-18 14:21:41 -0800 | [diff] [blame] | 2631 | if (nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER) && ret) |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2632 | return CHARGE_RETRY; |
| 2633 | |
| 2634 | /* |
| 2635 | * At task move, charge accounts can be doubly counted. So, it's |
| 2636 | * better to wait until the end of task_move if something is going on. |
| 2637 | */ |
| 2638 | if (mem_cgroup_wait_acct_move(mem_over_limit)) |
| 2639 | return CHARGE_RETRY; |
| 2640 | |
Johannes Weiner | 3812c8c | 2013-09-12 15:13:44 -0700 | [diff] [blame] | 2641 | if (invoke_oom) |
| 2642 | mem_cgroup_oom(mem_over_limit, gfp_mask, get_order(csize)); |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2643 | |
Johannes Weiner | 3812c8c | 2013-09-12 15:13:44 -0700 | [diff] [blame] | 2644 | return CHARGE_NOMEM; |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2645 | } |
| 2646 | |
Johannes Weiner | 6d1fdc4 | 2014-04-07 15:37:45 -0700 | [diff] [blame] | 2647 | /** |
| 2648 | * mem_cgroup_try_charge - try charging a memcg |
| 2649 | * @memcg: memcg to charge |
| 2650 | * @nr_pages: number of pages to charge |
| 2651 | * @oom: trigger OOM if reclaim fails |
KAMEZAWA Hiroyuki | 38c5d72 | 2012-01-12 17:19:01 -0800 | [diff] [blame] | 2652 | * |
Johannes Weiner | 6d1fdc4 | 2014-04-07 15:37:45 -0700 | [diff] [blame] | 2653 | * Returns 0 if @memcg was charged successfully, -EINTR if the charge |
| 2654 | * was bypassed to root_mem_cgroup, and -ENOMEM if the charge failed. |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2655 | */ |
Johannes Weiner | 6d1fdc4 | 2014-04-07 15:37:45 -0700 | [diff] [blame] | 2656 | static int mem_cgroup_try_charge(struct mem_cgroup *memcg, |
| 2657 | gfp_t gfp_mask, |
| 2658 | unsigned int nr_pages, |
| 2659 | bool oom) |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2660 | { |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 2661 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2662 | int nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES; |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2663 | int ret; |
KAMEZAWA Hiroyuki | a636b32 | 2009-01-07 18:08:08 -0800 | [diff] [blame] | 2664 | |
Johannes Weiner | 6d1fdc4 | 2014-04-07 15:37:45 -0700 | [diff] [blame] | 2665 | if (mem_cgroup_is_root(memcg)) |
| 2666 | goto done; |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 2667 | /* |
Johannes Weiner | 6d1fdc4 | 2014-04-07 15:37:45 -0700 | [diff] [blame] | 2668 | * Unlike in global OOM situations, memcg is not in a physical |
| 2669 | * memory shortage. Allow dying and OOM-killed tasks to |
| 2670 | * bypass the last charges so that they can exit quickly and |
| 2671 | * free their memory. |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 2672 | */ |
Johannes Weiner | 6d1fdc4 | 2014-04-07 15:37:45 -0700 | [diff] [blame] | 2673 | if (unlikely(test_thread_flag(TIF_MEMDIE) || |
Michal Hocko | d8dc595 | 2014-06-04 16:07:36 -0700 | [diff] [blame] | 2674 | fatal_signal_pending(current) || |
| 2675 | current->flags & PF_EXITING)) |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 2676 | goto bypass; |
KAMEZAWA Hiroyuki | a636b32 | 2009-01-07 18:08:08 -0800 | [diff] [blame] | 2677 | |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 2678 | if (unlikely(task_in_memcg_oom(current))) |
Johannes Weiner | 1f14c1a | 2013-12-12 17:12:35 -0800 | [diff] [blame] | 2679 | goto nomem; |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 2680 | |
Johannes Weiner | a0d8b00 | 2013-12-12 17:12:20 -0800 | [diff] [blame] | 2681 | if (gfp_mask & __GFP_NOFAIL) |
| 2682 | oom = false; |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 2683 | again: |
Michal Hocko | b6b6cc7 | 2014-04-07 15:37:44 -0700 | [diff] [blame] | 2684 | if (consume_stock(memcg, nr_pages)) |
| 2685 | goto done; |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2686 | |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2687 | do { |
Johannes Weiner | 3812c8c | 2013-09-12 15:13:44 -0700 | [diff] [blame] | 2688 | bool invoke_oom = oom && !nr_oom_retries; |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2689 | |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2690 | /* If killed, bypass charge */ |
Johannes Weiner | 6d1fdc4 | 2014-04-07 15:37:45 -0700 | [diff] [blame] | 2691 | if (fatal_signal_pending(current)) |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2692 | goto bypass; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2693 | |
Johannes Weiner | 3812c8c | 2013-09-12 15:13:44 -0700 | [diff] [blame] | 2694 | ret = mem_cgroup_do_charge(memcg, gfp_mask, batch, |
| 2695 | nr_pages, invoke_oom); |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2696 | switch (ret) { |
| 2697 | case CHARGE_OK: |
| 2698 | break; |
| 2699 | case CHARGE_RETRY: /* not in OOM situation but retry */ |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 2700 | batch = nr_pages; |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 2701 | goto again; |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2702 | case CHARGE_WOULDBLOCK: /* !__GFP_WAIT */ |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2703 | goto nomem; |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2704 | case CHARGE_NOMEM: /* OOM routine works */ |
Johannes Weiner | 6d1fdc4 | 2014-04-07 15:37:45 -0700 | [diff] [blame] | 2705 | if (!oom || invoke_oom) |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 2706 | goto nomem; |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2707 | nr_oom_retries--; |
| 2708 | break; |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 2709 | } |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2710 | } while (ret != CHARGE_OK); |
| 2711 | |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 2712 | if (batch > nr_pages) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2713 | refill_stock(memcg, batch - nr_pages); |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 2714 | done: |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2715 | return 0; |
| 2716 | nomem: |
Johannes Weiner | 6d1fdc4 | 2014-04-07 15:37:45 -0700 | [diff] [blame] | 2717 | if (!(gfp_mask & __GFP_NOFAIL)) |
Johannes Weiner | 3168ecb | 2013-10-31 16:34:13 -0700 | [diff] [blame] | 2718 | return -ENOMEM; |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 2719 | bypass: |
KAMEZAWA Hiroyuki | 38c5d72 | 2012-01-12 17:19:01 -0800 | [diff] [blame] | 2720 | return -EINTR; |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2721 | } |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2722 | |
Johannes Weiner | 6d1fdc4 | 2014-04-07 15:37:45 -0700 | [diff] [blame] | 2723 | /** |
| 2724 | * mem_cgroup_try_charge_mm - try charging a mm |
| 2725 | * @mm: mm_struct to charge |
| 2726 | * @nr_pages: number of pages to charge |
| 2727 | * @oom: trigger OOM if reclaim fails |
| 2728 | * |
| 2729 | * Returns the charged mem_cgroup associated with the given mm_struct or |
| 2730 | * NULL the charge failed. |
| 2731 | */ |
| 2732 | static struct mem_cgroup *mem_cgroup_try_charge_mm(struct mm_struct *mm, |
| 2733 | gfp_t gfp_mask, |
| 2734 | unsigned int nr_pages, |
| 2735 | bool oom) |
| 2736 | |
| 2737 | { |
| 2738 | struct mem_cgroup *memcg; |
| 2739 | int ret; |
| 2740 | |
| 2741 | memcg = get_mem_cgroup_from_mm(mm); |
| 2742 | ret = mem_cgroup_try_charge(memcg, gfp_mask, nr_pages, oom); |
| 2743 | css_put(&memcg->css); |
| 2744 | if (ret == -EINTR) |
| 2745 | memcg = root_mem_cgroup; |
| 2746 | else if (ret) |
| 2747 | memcg = NULL; |
| 2748 | |
| 2749 | return memcg; |
| 2750 | } |
| 2751 | |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2752 | /* |
Daisuke Nishimura | a3032a2 | 2009-12-15 16:47:10 -0800 | [diff] [blame] | 2753 | * Somemtimes we have to undo a charge we got by try_charge(). |
| 2754 | * This function is for that and do uncharge, put css's refcnt. |
| 2755 | * gotten by try_charge(). |
| 2756 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2757 | static void __mem_cgroup_cancel_charge(struct mem_cgroup *memcg, |
Johannes Weiner | e7018b8d | 2011-03-23 16:42:33 -0700 | [diff] [blame] | 2758 | unsigned int nr_pages) |
Daisuke Nishimura | a3032a2 | 2009-12-15 16:47:10 -0800 | [diff] [blame] | 2759 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2760 | if (!mem_cgroup_is_root(memcg)) { |
Johannes Weiner | e7018b8d | 2011-03-23 16:42:33 -0700 | [diff] [blame] | 2761 | unsigned long bytes = nr_pages * PAGE_SIZE; |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 2762 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2763 | res_counter_uncharge(&memcg->res, bytes); |
Johannes Weiner | e7018b8d | 2011-03-23 16:42:33 -0700 | [diff] [blame] | 2764 | if (do_swap_account) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2765 | res_counter_uncharge(&memcg->memsw, bytes); |
Johannes Weiner | e7018b8d | 2011-03-23 16:42:33 -0700 | [diff] [blame] | 2766 | } |
Daisuke Nishimura | a3032a2 | 2009-12-15 16:47:10 -0800 | [diff] [blame] | 2767 | } |
| 2768 | |
| 2769 | /* |
KAMEZAWA Hiroyuki | d01dd17 | 2012-05-29 15:07:03 -0700 | [diff] [blame] | 2770 | * Cancel chrages in this cgroup....doesn't propagate to parent cgroup. |
| 2771 | * This is useful when moving usage to parent cgroup. |
| 2772 | */ |
| 2773 | static void __mem_cgroup_cancel_local_charge(struct mem_cgroup *memcg, |
| 2774 | unsigned int nr_pages) |
| 2775 | { |
| 2776 | unsigned long bytes = nr_pages * PAGE_SIZE; |
| 2777 | |
| 2778 | if (mem_cgroup_is_root(memcg)) |
| 2779 | return; |
| 2780 | |
| 2781 | res_counter_uncharge_until(&memcg->res, memcg->res.parent, bytes); |
| 2782 | if (do_swap_account) |
| 2783 | res_counter_uncharge_until(&memcg->memsw, |
| 2784 | memcg->memsw.parent, bytes); |
| 2785 | } |
| 2786 | |
| 2787 | /* |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2788 | * A helper function to get mem_cgroup from ID. must be called under |
Tejun Heo | e931608 | 2012-11-05 09:16:58 -0800 | [diff] [blame] | 2789 | * rcu_read_lock(). The caller is responsible for calling css_tryget if |
| 2790 | * the mem_cgroup is used for charging. (dropping refcnt from swap can be |
| 2791 | * called against removed memcg.) |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2792 | */ |
| 2793 | static struct mem_cgroup *mem_cgroup_lookup(unsigned short id) |
| 2794 | { |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2795 | /* ID 0 is unused ID */ |
| 2796 | if (!id) |
| 2797 | return NULL; |
Li Zefan | 34c00c3 | 2013-09-23 16:56:01 +0800 | [diff] [blame] | 2798 | return mem_cgroup_from_id(id); |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2799 | } |
| 2800 | |
Wu Fengguang | e42d9d5 | 2009-12-16 12:19:59 +0100 | [diff] [blame] | 2801 | struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page) |
KAMEZAWA Hiroyuki | b5a8431 | 2009-01-07 18:08:35 -0800 | [diff] [blame] | 2802 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2803 | struct mem_cgroup *memcg = NULL; |
Daisuke Nishimura | 3c776e6 | 2009-04-02 16:57:43 -0700 | [diff] [blame] | 2804 | struct page_cgroup *pc; |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2805 | unsigned short id; |
KAMEZAWA Hiroyuki | b5a8431 | 2009-01-07 18:08:35 -0800 | [diff] [blame] | 2806 | swp_entry_t ent; |
| 2807 | |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 2808 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
Daisuke Nishimura | 3c776e6 | 2009-04-02 16:57:43 -0700 | [diff] [blame] | 2809 | |
Daisuke Nishimura | 3c776e6 | 2009-04-02 16:57:43 -0700 | [diff] [blame] | 2810 | pc = lookup_page_cgroup(page); |
Daisuke Nishimura | c0bd3f6 | 2009-04-30 15:08:11 -0700 | [diff] [blame] | 2811 | lock_page_cgroup(pc); |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2812 | if (PageCgroupUsed(pc)) { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2813 | memcg = pc->mem_cgroup; |
| 2814 | if (memcg && !css_tryget(&memcg->css)) |
| 2815 | memcg = NULL; |
Wu Fengguang | e42d9d5 | 2009-12-16 12:19:59 +0100 | [diff] [blame] | 2816 | } else if (PageSwapCache(page)) { |
Daisuke Nishimura | 3c776e6 | 2009-04-02 16:57:43 -0700 | [diff] [blame] | 2817 | ent.val = page_private(page); |
Bob Liu | 9fb4b7c | 2012-01-12 17:18:48 -0800 | [diff] [blame] | 2818 | id = lookup_swap_cgroup_id(ent); |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2819 | rcu_read_lock(); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2820 | memcg = mem_cgroup_lookup(id); |
| 2821 | if (memcg && !css_tryget(&memcg->css)) |
| 2822 | memcg = NULL; |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 2823 | rcu_read_unlock(); |
Daisuke Nishimura | 3c776e6 | 2009-04-02 16:57:43 -0700 | [diff] [blame] | 2824 | } |
Daisuke Nishimura | c0bd3f6 | 2009-04-30 15:08:11 -0700 | [diff] [blame] | 2825 | unlock_page_cgroup(pc); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2826 | return memcg; |
KAMEZAWA Hiroyuki | b5a8431 | 2009-01-07 18:08:35 -0800 | [diff] [blame] | 2827 | } |
| 2828 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2829 | static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg, |
Johannes Weiner | 5564e88 | 2011-03-23 16:42:29 -0700 | [diff] [blame] | 2830 | struct page *page, |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 2831 | unsigned int nr_pages, |
Hugh Dickins | 9ce70c0 | 2012-03-05 14:59:16 -0800 | [diff] [blame] | 2832 | enum charge_type ctype, |
| 2833 | bool lrucare) |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2834 | { |
Johannes Weiner | ce587e6 | 2012-04-24 20:22:33 +0200 | [diff] [blame] | 2835 | struct page_cgroup *pc = lookup_page_cgroup(page); |
Hugh Dickins | 9ce70c0 | 2012-03-05 14:59:16 -0800 | [diff] [blame] | 2836 | struct zone *uninitialized_var(zone); |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 2837 | struct lruvec *lruvec; |
Hugh Dickins | 9ce70c0 | 2012-03-05 14:59:16 -0800 | [diff] [blame] | 2838 | bool was_on_lru = false; |
KAMEZAWA Hiroyuki | b2402857 | 2012-03-21 16:34:22 -0700 | [diff] [blame] | 2839 | bool anon; |
Hugh Dickins | 9ce70c0 | 2012-03-05 14:59:16 -0800 | [diff] [blame] | 2840 | |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2841 | lock_page_cgroup(pc); |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 2842 | VM_BUG_ON_PAGE(PageCgroupUsed(pc), page); |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2843 | /* |
| 2844 | * we don't need page_cgroup_lock about tail pages, becase they are not |
| 2845 | * accessed by any other context at this point. |
| 2846 | */ |
Hugh Dickins | 9ce70c0 | 2012-03-05 14:59:16 -0800 | [diff] [blame] | 2847 | |
| 2848 | /* |
| 2849 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page |
| 2850 | * may already be on some other mem_cgroup's LRU. Take care of it. |
| 2851 | */ |
| 2852 | if (lrucare) { |
| 2853 | zone = page_zone(page); |
| 2854 | spin_lock_irq(&zone->lru_lock); |
| 2855 | if (PageLRU(page)) { |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 2856 | lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup); |
Hugh Dickins | 9ce70c0 | 2012-03-05 14:59:16 -0800 | [diff] [blame] | 2857 | ClearPageLRU(page); |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 2858 | del_page_from_lru_list(page, lruvec, page_lru(page)); |
Hugh Dickins | 9ce70c0 | 2012-03-05 14:59:16 -0800 | [diff] [blame] | 2859 | was_on_lru = true; |
| 2860 | } |
| 2861 | } |
| 2862 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2863 | pc->mem_cgroup = memcg; |
KAMEZAWA Hiroyuki | 261fb61 | 2009-09-23 15:56:33 -0700 | [diff] [blame] | 2864 | /* |
| 2865 | * We access a page_cgroup asynchronously without lock_page_cgroup(). |
| 2866 | * Especially when a page_cgroup is taken from a page, pc->mem_cgroup |
| 2867 | * is accessed after testing USED bit. To make pc->mem_cgroup visible |
| 2868 | * before USED bit, we need memory barrier here. |
| 2869 | * See mem_cgroup_add_lru_list(), etc. |
Andrew Morton | f894ffa | 2013-09-12 15:13:35 -0700 | [diff] [blame] | 2870 | */ |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 2871 | smp_wmb(); |
KAMEZAWA Hiroyuki | b2402857 | 2012-03-21 16:34:22 -0700 | [diff] [blame] | 2872 | SetPageCgroupUsed(pc); |
Hugh Dickins | 3be9127 | 2008-02-07 00:14:19 -0800 | [diff] [blame] | 2873 | |
Hugh Dickins | 9ce70c0 | 2012-03-05 14:59:16 -0800 | [diff] [blame] | 2874 | if (lrucare) { |
| 2875 | if (was_on_lru) { |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 2876 | lruvec = mem_cgroup_zone_lruvec(zone, pc->mem_cgroup); |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 2877 | VM_BUG_ON_PAGE(PageLRU(page), page); |
Hugh Dickins | 9ce70c0 | 2012-03-05 14:59:16 -0800 | [diff] [blame] | 2878 | SetPageLRU(page); |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 2879 | add_page_to_lru_list(page, lruvec, page_lru(page)); |
Hugh Dickins | 9ce70c0 | 2012-03-05 14:59:16 -0800 | [diff] [blame] | 2880 | } |
| 2881 | spin_unlock_irq(&zone->lru_lock); |
| 2882 | } |
| 2883 | |
Kamezawa Hiroyuki | 41326c1 | 2012-07-31 16:41:40 -0700 | [diff] [blame] | 2884 | if (ctype == MEM_CGROUP_CHARGE_TYPE_ANON) |
KAMEZAWA Hiroyuki | b2402857 | 2012-03-21 16:34:22 -0700 | [diff] [blame] | 2885 | anon = true; |
| 2886 | else |
| 2887 | anon = false; |
| 2888 | |
David Rientjes | b070e65 | 2013-05-07 16:18:09 -0700 | [diff] [blame] | 2889 | mem_cgroup_charge_statistics(memcg, page, anon, nr_pages); |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 2890 | unlock_page_cgroup(pc); |
Hugh Dickins | 9ce70c0 | 2012-03-05 14:59:16 -0800 | [diff] [blame] | 2891 | |
KAMEZAWA Hiroyuki | 430e4863 | 2010-03-10 15:22:30 -0800 | [diff] [blame] | 2892 | /* |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 2893 | * "charge_statistics" updated event counter. Then, check it. |
| 2894 | * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree. |
| 2895 | * if they exceeds softlimit. |
KAMEZAWA Hiroyuki | 430e4863 | 2010-03-10 15:22:30 -0800 | [diff] [blame] | 2896 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2897 | memcg_check_events(memcg, page); |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2898 | } |
| 2899 | |
Glauber Costa | 7cf2798 | 2012-12-18 14:22:55 -0800 | [diff] [blame] | 2900 | static DEFINE_MUTEX(set_limit_mutex); |
| 2901 | |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2902 | #ifdef CONFIG_MEMCG_KMEM |
Vladimir Davydov | bd67314 | 2014-06-04 16:07:40 -0700 | [diff] [blame] | 2903 | /* |
| 2904 | * The memcg_slab_mutex is held whenever a per memcg kmem cache is created or |
| 2905 | * destroyed. It protects memcg_caches arrays and memcg_slab_caches lists. |
| 2906 | */ |
| 2907 | static DEFINE_MUTEX(memcg_slab_mutex); |
| 2908 | |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 2909 | static DEFINE_MUTEX(activate_kmem_mutex); |
| 2910 | |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2911 | static inline bool memcg_can_account_kmem(struct mem_cgroup *memcg) |
| 2912 | { |
| 2913 | return !mem_cgroup_disabled() && !mem_cgroup_is_root(memcg) && |
Vladimir Davydov | 6de64be | 2014-01-23 15:53:08 -0800 | [diff] [blame] | 2914 | memcg_kmem_is_active(memcg); |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2915 | } |
| 2916 | |
Glauber Costa | 1f458cb | 2012-12-18 14:22:50 -0800 | [diff] [blame] | 2917 | /* |
| 2918 | * This is a bit cumbersome, but it is rarely used and avoids a backpointer |
| 2919 | * in the memcg_cache_params struct. |
| 2920 | */ |
| 2921 | static struct kmem_cache *memcg_params_to_cache(struct memcg_cache_params *p) |
| 2922 | { |
| 2923 | struct kmem_cache *cachep; |
| 2924 | |
| 2925 | VM_BUG_ON(p->is_root_cache); |
| 2926 | cachep = p->root_cache; |
Qiang Huang | 7a67d7a | 2013-11-12 15:08:24 -0800 | [diff] [blame] | 2927 | return cache_from_memcg_idx(cachep, memcg_cache_id(p->memcg)); |
Glauber Costa | 1f458cb | 2012-12-18 14:22:50 -0800 | [diff] [blame] | 2928 | } |
| 2929 | |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 2930 | #ifdef CONFIG_SLABINFO |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 2931 | static int mem_cgroup_slabinfo_read(struct seq_file *m, void *v) |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 2932 | { |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 2933 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 2934 | struct memcg_cache_params *params; |
| 2935 | |
| 2936 | if (!memcg_can_account_kmem(memcg)) |
| 2937 | return -EIO; |
| 2938 | |
| 2939 | print_slabinfo_header(m); |
| 2940 | |
Vladimir Davydov | bd67314 | 2014-06-04 16:07:40 -0700 | [diff] [blame] | 2941 | mutex_lock(&memcg_slab_mutex); |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 2942 | list_for_each_entry(params, &memcg->memcg_slab_caches, list) |
| 2943 | cache_show(memcg_params_to_cache(params), m); |
Vladimir Davydov | bd67314 | 2014-06-04 16:07:40 -0700 | [diff] [blame] | 2944 | mutex_unlock(&memcg_slab_mutex); |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 2945 | |
| 2946 | return 0; |
| 2947 | } |
| 2948 | #endif |
| 2949 | |
Vladimir Davydov | c67a8a6 | 2014-06-04 16:07:39 -0700 | [diff] [blame] | 2950 | static int memcg_charge_kmem(struct mem_cgroup *memcg, gfp_t gfp, u64 size) |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2951 | { |
| 2952 | struct res_counter *fail_res; |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2953 | int ret = 0; |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2954 | |
| 2955 | ret = res_counter_charge(&memcg->kmem, size, &fail_res); |
| 2956 | if (ret) |
| 2957 | return ret; |
| 2958 | |
Johannes Weiner | 6d1fdc4 | 2014-04-07 15:37:45 -0700 | [diff] [blame] | 2959 | ret = mem_cgroup_try_charge(memcg, gfp, size >> PAGE_SHIFT, |
| 2960 | oom_gfp_allowed(gfp)); |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2961 | if (ret == -EINTR) { |
| 2962 | /* |
Johannes Weiner | 6d1fdc4 | 2014-04-07 15:37:45 -0700 | [diff] [blame] | 2963 | * mem_cgroup_try_charge() chosed to bypass to root due to |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2964 | * OOM kill or fatal signal. Since our only options are to |
| 2965 | * either fail the allocation or charge it to this cgroup, do |
| 2966 | * it as a temporary condition. But we can't fail. From a |
| 2967 | * kmem/slab perspective, the cache has already been selected, |
| 2968 | * by mem_cgroup_kmem_get_cache(), so it is too late to change |
| 2969 | * our minds. |
| 2970 | * |
| 2971 | * This condition will only trigger if the task entered |
| 2972 | * memcg_charge_kmem in a sane state, but was OOM-killed during |
Johannes Weiner | 6d1fdc4 | 2014-04-07 15:37:45 -0700 | [diff] [blame] | 2973 | * mem_cgroup_try_charge() above. Tasks that were already |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2974 | * dying when the allocation triggers should have been already |
| 2975 | * directed to the root cgroup in memcontrol.h |
| 2976 | */ |
| 2977 | res_counter_charge_nofail(&memcg->res, size, &fail_res); |
| 2978 | if (do_swap_account) |
| 2979 | res_counter_charge_nofail(&memcg->memsw, size, |
| 2980 | &fail_res); |
| 2981 | ret = 0; |
| 2982 | } else if (ret) |
| 2983 | res_counter_uncharge(&memcg->kmem, size); |
| 2984 | |
| 2985 | return ret; |
| 2986 | } |
| 2987 | |
Vladimir Davydov | c67a8a6 | 2014-06-04 16:07:39 -0700 | [diff] [blame] | 2988 | static void memcg_uncharge_kmem(struct mem_cgroup *memcg, u64 size) |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2989 | { |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2990 | res_counter_uncharge(&memcg->res, size); |
| 2991 | if (do_swap_account) |
| 2992 | res_counter_uncharge(&memcg->memsw, size); |
Glauber Costa | 7de3768 | 2012-12-18 14:22:07 -0800 | [diff] [blame] | 2993 | |
| 2994 | /* Not down to 0 */ |
| 2995 | if (res_counter_uncharge(&memcg->kmem, size)) |
| 2996 | return; |
| 2997 | |
Li Zefan | 10d5ebf | 2013-07-08 16:00:33 -0700 | [diff] [blame] | 2998 | /* |
| 2999 | * Releases a reference taken in kmem_cgroup_css_offline in case |
| 3000 | * this last uncharge is racing with the offlining code or it is |
| 3001 | * outliving the memcg existence. |
| 3002 | * |
| 3003 | * The memory barrier imposed by test&clear is paired with the |
| 3004 | * explicit one in memcg_kmem_mark_dead(). |
| 3005 | */ |
Glauber Costa | 7de3768 | 2012-12-18 14:22:07 -0800 | [diff] [blame] | 3006 | if (memcg_kmem_test_and_clear_dead(memcg)) |
Li Zefan | 10d5ebf | 2013-07-08 16:00:33 -0700 | [diff] [blame] | 3007 | css_put(&memcg->css); |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 3008 | } |
| 3009 | |
Glauber Costa | 2633d7a | 2012-12-18 14:22:34 -0800 | [diff] [blame] | 3010 | /* |
| 3011 | * helper for acessing a memcg's index. It will be used as an index in the |
| 3012 | * child cache array in kmem_cache, and also to derive its name. This function |
| 3013 | * will return -1 when this is not a kmem-limited memcg. |
| 3014 | */ |
| 3015 | int memcg_cache_id(struct mem_cgroup *memcg) |
| 3016 | { |
| 3017 | return memcg ? memcg->kmemcg_id : -1; |
| 3018 | } |
| 3019 | |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 3020 | static size_t memcg_caches_array_size(int num_groups) |
| 3021 | { |
| 3022 | ssize_t size; |
| 3023 | if (num_groups <= 0) |
| 3024 | return 0; |
| 3025 | |
| 3026 | size = 2 * num_groups; |
| 3027 | if (size < MEMCG_CACHES_MIN_SIZE) |
| 3028 | size = MEMCG_CACHES_MIN_SIZE; |
| 3029 | else if (size > MEMCG_CACHES_MAX_SIZE) |
| 3030 | size = MEMCG_CACHES_MAX_SIZE; |
| 3031 | |
| 3032 | return size; |
| 3033 | } |
| 3034 | |
| 3035 | /* |
| 3036 | * We should update the current array size iff all caches updates succeed. This |
| 3037 | * can only be done from the slab side. The slab mutex needs to be held when |
| 3038 | * calling this. |
| 3039 | */ |
| 3040 | void memcg_update_array_size(int num) |
| 3041 | { |
| 3042 | if (num > memcg_limited_groups_array_size) |
| 3043 | memcg_limited_groups_array_size = memcg_caches_array_size(num); |
| 3044 | } |
| 3045 | |
| 3046 | int memcg_update_cache_size(struct kmem_cache *s, int num_groups) |
| 3047 | { |
| 3048 | struct memcg_cache_params *cur_params = s->memcg_params; |
| 3049 | |
Qiang Huang | f35c3a8 | 2013-11-12 15:08:22 -0800 | [diff] [blame] | 3050 | VM_BUG_ON(!is_root_cache(s)); |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 3051 | |
| 3052 | if (num_groups > memcg_limited_groups_array_size) { |
| 3053 | int i; |
Vladimir Davydov | f857026 | 2014-01-23 15:53:06 -0800 | [diff] [blame] | 3054 | struct memcg_cache_params *new_params; |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 3055 | ssize_t size = memcg_caches_array_size(num_groups); |
| 3056 | |
| 3057 | size *= sizeof(void *); |
Andrey Vagin | 90c7a79 | 2013-09-11 14:22:18 -0700 | [diff] [blame] | 3058 | size += offsetof(struct memcg_cache_params, memcg_caches); |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 3059 | |
Vladimir Davydov | f857026 | 2014-01-23 15:53:06 -0800 | [diff] [blame] | 3060 | new_params = kzalloc(size, GFP_KERNEL); |
| 3061 | if (!new_params) |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 3062 | return -ENOMEM; |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 3063 | |
Vladimir Davydov | f857026 | 2014-01-23 15:53:06 -0800 | [diff] [blame] | 3064 | new_params->is_root_cache = true; |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 3065 | |
| 3066 | /* |
| 3067 | * There is the chance it will be bigger than |
| 3068 | * memcg_limited_groups_array_size, if we failed an allocation |
| 3069 | * in a cache, in which case all caches updated before it, will |
| 3070 | * have a bigger array. |
| 3071 | * |
| 3072 | * But if that is the case, the data after |
| 3073 | * memcg_limited_groups_array_size is certainly unused |
| 3074 | */ |
| 3075 | for (i = 0; i < memcg_limited_groups_array_size; i++) { |
| 3076 | if (!cur_params->memcg_caches[i]) |
| 3077 | continue; |
Vladimir Davydov | f857026 | 2014-01-23 15:53:06 -0800 | [diff] [blame] | 3078 | new_params->memcg_caches[i] = |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 3079 | cur_params->memcg_caches[i]; |
| 3080 | } |
| 3081 | |
| 3082 | /* |
| 3083 | * Ideally, we would wait until all caches succeed, and only |
| 3084 | * then free the old one. But this is not worth the extra |
| 3085 | * pointer per-cache we'd have to have for this. |
| 3086 | * |
| 3087 | * It is not a big deal if some caches are left with a size |
| 3088 | * bigger than the others. And all updates will reset this |
| 3089 | * anyway. |
| 3090 | */ |
Vladimir Davydov | f857026 | 2014-01-23 15:53:06 -0800 | [diff] [blame] | 3091 | rcu_assign_pointer(s->memcg_params, new_params); |
| 3092 | if (cur_params) |
| 3093 | kfree_rcu(cur_params, rcu_head); |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 3094 | } |
| 3095 | return 0; |
| 3096 | } |
| 3097 | |
Vladimir Davydov | 5722d09 | 2014-04-07 15:39:24 -0700 | [diff] [blame] | 3098 | char *memcg_create_cache_name(struct mem_cgroup *memcg, |
| 3099 | struct kmem_cache *root_cache) |
| 3100 | { |
Fabian Frederick | ada4ba5 | 2014-06-04 16:08:08 -0700 | [diff] [blame] | 3101 | static char *buf; |
Vladimir Davydov | 5722d09 | 2014-04-07 15:39:24 -0700 | [diff] [blame] | 3102 | |
| 3103 | /* |
| 3104 | * We need a mutex here to protect the shared buffer. Since this is |
| 3105 | * expected to be called only on cache creation, we can employ the |
| 3106 | * slab_mutex for that purpose. |
| 3107 | */ |
| 3108 | lockdep_assert_held(&slab_mutex); |
| 3109 | |
| 3110 | if (!buf) { |
| 3111 | buf = kmalloc(NAME_MAX + 1, GFP_KERNEL); |
| 3112 | if (!buf) |
| 3113 | return NULL; |
| 3114 | } |
| 3115 | |
| 3116 | cgroup_name(memcg->css.cgroup, buf, NAME_MAX + 1); |
| 3117 | return kasprintf(GFP_KERNEL, "%s(%d:%s)", root_cache->name, |
| 3118 | memcg_cache_id(memcg), buf); |
| 3119 | } |
| 3120 | |
Vladimir Davydov | 363a044 | 2014-01-23 15:52:56 -0800 | [diff] [blame] | 3121 | int memcg_alloc_cache_params(struct mem_cgroup *memcg, struct kmem_cache *s, |
| 3122 | struct kmem_cache *root_cache) |
Glauber Costa | 2633d7a | 2012-12-18 14:22:34 -0800 | [diff] [blame] | 3123 | { |
Andrey Vagin | 90c7a79 | 2013-09-11 14:22:18 -0700 | [diff] [blame] | 3124 | size_t size; |
Glauber Costa | 2633d7a | 2012-12-18 14:22:34 -0800 | [diff] [blame] | 3125 | |
| 3126 | if (!memcg_kmem_enabled()) |
| 3127 | return 0; |
| 3128 | |
Andrey Vagin | 90c7a79 | 2013-09-11 14:22:18 -0700 | [diff] [blame] | 3129 | if (!memcg) { |
| 3130 | size = offsetof(struct memcg_cache_params, memcg_caches); |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 3131 | size += memcg_limited_groups_array_size * sizeof(void *); |
Andrey Vagin | 90c7a79 | 2013-09-11 14:22:18 -0700 | [diff] [blame] | 3132 | } else |
| 3133 | size = sizeof(struct memcg_cache_params); |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 3134 | |
Glauber Costa | 2633d7a | 2012-12-18 14:22:34 -0800 | [diff] [blame] | 3135 | s->memcg_params = kzalloc(size, GFP_KERNEL); |
| 3136 | if (!s->memcg_params) |
| 3137 | return -ENOMEM; |
| 3138 | |
Glauber Costa | 943a451 | 2012-12-18 14:23:03 -0800 | [diff] [blame] | 3139 | if (memcg) { |
Glauber Costa | 2633d7a | 2012-12-18 14:22:34 -0800 | [diff] [blame] | 3140 | s->memcg_params->memcg = memcg; |
Glauber Costa | 943a451 | 2012-12-18 14:23:03 -0800 | [diff] [blame] | 3141 | s->memcg_params->root_cache = root_cache; |
Vladimir Davydov | 051dd46 | 2014-04-07 15:39:27 -0700 | [diff] [blame] | 3142 | css_get(&memcg->css); |
Glauber Costa | 4ba902b | 2013-02-12 13:46:22 -0800 | [diff] [blame] | 3143 | } else |
| 3144 | s->memcg_params->is_root_cache = true; |
| 3145 | |
Glauber Costa | 2633d7a | 2012-12-18 14:22:34 -0800 | [diff] [blame] | 3146 | return 0; |
| 3147 | } |
| 3148 | |
Vladimir Davydov | 363a044 | 2014-01-23 15:52:56 -0800 | [diff] [blame] | 3149 | void memcg_free_cache_params(struct kmem_cache *s) |
| 3150 | { |
Vladimir Davydov | 051dd46 | 2014-04-07 15:39:27 -0700 | [diff] [blame] | 3151 | if (!s->memcg_params) |
| 3152 | return; |
| 3153 | if (!s->memcg_params->is_root_cache) |
| 3154 | css_put(&s->memcg_params->memcg->css); |
Vladimir Davydov | 363a044 | 2014-01-23 15:52:56 -0800 | [diff] [blame] | 3155 | kfree(s->memcg_params); |
| 3156 | } |
| 3157 | |
Vladimir Davydov | bd67314 | 2014-06-04 16:07:40 -0700 | [diff] [blame] | 3158 | static void memcg_kmem_create_cache(struct mem_cgroup *memcg, |
| 3159 | struct kmem_cache *root_cache) |
Glauber Costa | 2633d7a | 2012-12-18 14:22:34 -0800 | [diff] [blame] | 3160 | { |
Vladimir Davydov | bd67314 | 2014-06-04 16:07:40 -0700 | [diff] [blame] | 3161 | struct kmem_cache *cachep; |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 3162 | int id; |
| 3163 | |
Vladimir Davydov | bd67314 | 2014-06-04 16:07:40 -0700 | [diff] [blame] | 3164 | lockdep_assert_held(&memcg_slab_mutex); |
| 3165 | |
| 3166 | id = memcg_cache_id(memcg); |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 3167 | |
Vladimir Davydov | 2edefe1 | 2014-01-23 15:53:02 -0800 | [diff] [blame] | 3168 | /* |
Vladimir Davydov | bd67314 | 2014-06-04 16:07:40 -0700 | [diff] [blame] | 3169 | * Since per-memcg caches are created asynchronously on first |
| 3170 | * allocation (see memcg_kmem_get_cache()), several threads can try to |
| 3171 | * create the same cache, but only one of them may succeed. |
Vladimir Davydov | 2edefe1 | 2014-01-23 15:53:02 -0800 | [diff] [blame] | 3172 | */ |
Vladimir Davydov | bd67314 | 2014-06-04 16:07:40 -0700 | [diff] [blame] | 3173 | if (cache_from_memcg_idx(root_cache, id)) |
| 3174 | return; |
Vladimir Davydov | 2edefe1 | 2014-01-23 15:53:02 -0800 | [diff] [blame] | 3175 | |
Vladimir Davydov | bd67314 | 2014-06-04 16:07:40 -0700 | [diff] [blame] | 3176 | cachep = kmem_cache_create_memcg(memcg, root_cache); |
| 3177 | /* |
| 3178 | * If we could not create a memcg cache, do not complain, because |
| 3179 | * that's not critical at all as we can always proceed with the root |
| 3180 | * cache. |
| 3181 | */ |
| 3182 | if (!cachep) |
| 3183 | return; |
| 3184 | |
| 3185 | list_add(&cachep->memcg_params->list, &memcg->memcg_slab_caches); |
Vladimir Davydov | 1aa1325 | 2014-01-23 15:52:58 -0800 | [diff] [blame] | 3186 | |
Vladimir Davydov | 1aa1325 | 2014-01-23 15:52:58 -0800 | [diff] [blame] | 3187 | /* |
Vladimir Davydov | 959c896 | 2014-01-23 15:52:59 -0800 | [diff] [blame] | 3188 | * Since readers won't lock (see cache_from_memcg_idx()), we need a |
| 3189 | * barrier here to ensure nobody will see the kmem_cache partially |
| 3190 | * initialized. |
Vladimir Davydov | 1aa1325 | 2014-01-23 15:52:58 -0800 | [diff] [blame] | 3191 | */ |
Vladimir Davydov | 959c896 | 2014-01-23 15:52:59 -0800 | [diff] [blame] | 3192 | smp_wmb(); |
| 3193 | |
Vladimir Davydov | bd67314 | 2014-06-04 16:07:40 -0700 | [diff] [blame] | 3194 | BUG_ON(root_cache->memcg_params->memcg_caches[id]); |
| 3195 | root_cache->memcg_params->memcg_caches[id] = cachep; |
Vladimir Davydov | 1aa1325 | 2014-01-23 15:52:58 -0800 | [diff] [blame] | 3196 | } |
| 3197 | |
Vladimir Davydov | bd67314 | 2014-06-04 16:07:40 -0700 | [diff] [blame] | 3198 | static void memcg_kmem_destroy_cache(struct kmem_cache *cachep) |
Vladimir Davydov | 1aa1325 | 2014-01-23 15:52:58 -0800 | [diff] [blame] | 3199 | { |
Vladimir Davydov | bd67314 | 2014-06-04 16:07:40 -0700 | [diff] [blame] | 3200 | struct kmem_cache *root_cache; |
Vladimir Davydov | 1aa1325 | 2014-01-23 15:52:58 -0800 | [diff] [blame] | 3201 | struct mem_cgroup *memcg; |
| 3202 | int id; |
| 3203 | |
Vladimir Davydov | bd67314 | 2014-06-04 16:07:40 -0700 | [diff] [blame] | 3204 | lockdep_assert_held(&memcg_slab_mutex); |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 3205 | |
Vladimir Davydov | bd67314 | 2014-06-04 16:07:40 -0700 | [diff] [blame] | 3206 | BUG_ON(is_root_cache(cachep)); |
Vladimir Davydov | 2edefe1 | 2014-01-23 15:53:02 -0800 | [diff] [blame] | 3207 | |
Vladimir Davydov | bd67314 | 2014-06-04 16:07:40 -0700 | [diff] [blame] | 3208 | root_cache = cachep->memcg_params->root_cache; |
| 3209 | memcg = cachep->memcg_params->memcg; |
Vladimir Davydov | 96403da | 2014-01-23 15:53:01 -0800 | [diff] [blame] | 3210 | id = memcg_cache_id(memcg); |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 3211 | |
Vladimir Davydov | bd67314 | 2014-06-04 16:07:40 -0700 | [diff] [blame] | 3212 | BUG_ON(root_cache->memcg_params->memcg_caches[id] != cachep); |
| 3213 | root_cache->memcg_params->memcg_caches[id] = NULL; |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 3214 | |
Vladimir Davydov | bd67314 | 2014-06-04 16:07:40 -0700 | [diff] [blame] | 3215 | list_del(&cachep->memcg_params->list); |
| 3216 | |
| 3217 | kmem_cache_destroy(cachep); |
Glauber Costa | 2633d7a | 2012-12-18 14:22:34 -0800 | [diff] [blame] | 3218 | } |
| 3219 | |
Glauber Costa | 0e9d92f | 2012-12-18 14:22:42 -0800 | [diff] [blame] | 3220 | /* |
| 3221 | * During the creation a new cache, we need to disable our accounting mechanism |
| 3222 | * altogether. This is true even if we are not creating, but rather just |
| 3223 | * enqueing new caches to be created. |
| 3224 | * |
| 3225 | * This is because that process will trigger allocations; some visible, like |
| 3226 | * explicit kmallocs to auxiliary data structures, name strings and internal |
| 3227 | * cache structures; some well concealed, like INIT_WORK() that can allocate |
| 3228 | * objects during debug. |
| 3229 | * |
| 3230 | * If any allocation happens during memcg_kmem_get_cache, we will recurse back |
| 3231 | * to it. This may not be a bounded recursion: since the first cache creation |
| 3232 | * failed to complete (waiting on the allocation), we'll just try to create the |
| 3233 | * cache again, failing at the same point. |
| 3234 | * |
| 3235 | * memcg_kmem_get_cache is prepared to abort after seeing a positive count of |
| 3236 | * memcg_kmem_skip_account. So we enclose anything that might allocate memory |
| 3237 | * inside the following two functions. |
| 3238 | */ |
| 3239 | static inline void memcg_stop_kmem_account(void) |
| 3240 | { |
| 3241 | VM_BUG_ON(!current->mm); |
| 3242 | current->memcg_kmem_skip_account++; |
| 3243 | } |
| 3244 | |
| 3245 | static inline void memcg_resume_kmem_account(void) |
| 3246 | { |
| 3247 | VM_BUG_ON(!current->mm); |
| 3248 | current->memcg_kmem_skip_account--; |
| 3249 | } |
| 3250 | |
Vladimir Davydov | b852990 | 2014-04-07 15:39:28 -0700 | [diff] [blame] | 3251 | int __kmem_cache_destroy_memcg_children(struct kmem_cache *s) |
Glauber Costa | 7cf2798 | 2012-12-18 14:22:55 -0800 | [diff] [blame] | 3252 | { |
| 3253 | struct kmem_cache *c; |
Vladimir Davydov | b852990 | 2014-04-07 15:39:28 -0700 | [diff] [blame] | 3254 | int i, failed = 0; |
Glauber Costa | 7cf2798 | 2012-12-18 14:22:55 -0800 | [diff] [blame] | 3255 | |
Vladimir Davydov | bd67314 | 2014-06-04 16:07:40 -0700 | [diff] [blame] | 3256 | mutex_lock(&memcg_slab_mutex); |
Qiang Huang | 7a67d7a | 2013-11-12 15:08:24 -0800 | [diff] [blame] | 3257 | for_each_memcg_cache_index(i) { |
| 3258 | c = cache_from_memcg_idx(s, i); |
Glauber Costa | 7cf2798 | 2012-12-18 14:22:55 -0800 | [diff] [blame] | 3259 | if (!c) |
| 3260 | continue; |
| 3261 | |
Vladimir Davydov | bd67314 | 2014-06-04 16:07:40 -0700 | [diff] [blame] | 3262 | memcg_kmem_destroy_cache(c); |
Vladimir Davydov | b852990 | 2014-04-07 15:39:28 -0700 | [diff] [blame] | 3263 | |
| 3264 | if (cache_from_memcg_idx(s, i)) |
| 3265 | failed++; |
Glauber Costa | 7cf2798 | 2012-12-18 14:22:55 -0800 | [diff] [blame] | 3266 | } |
Vladimir Davydov | bd67314 | 2014-06-04 16:07:40 -0700 | [diff] [blame] | 3267 | mutex_unlock(&memcg_slab_mutex); |
Vladimir Davydov | b852990 | 2014-04-07 15:39:28 -0700 | [diff] [blame] | 3268 | return failed; |
Glauber Costa | 7cf2798 | 2012-12-18 14:22:55 -0800 | [diff] [blame] | 3269 | } |
| 3270 | |
Glauber Costa | 1f458cb | 2012-12-18 14:22:50 -0800 | [diff] [blame] | 3271 | static void mem_cgroup_destroy_all_caches(struct mem_cgroup *memcg) |
| 3272 | { |
| 3273 | struct kmem_cache *cachep; |
Vladimir Davydov | bd67314 | 2014-06-04 16:07:40 -0700 | [diff] [blame] | 3274 | struct memcg_cache_params *params, *tmp; |
Glauber Costa | 1f458cb | 2012-12-18 14:22:50 -0800 | [diff] [blame] | 3275 | |
| 3276 | if (!memcg_kmem_is_active(memcg)) |
| 3277 | return; |
| 3278 | |
Vladimir Davydov | bd67314 | 2014-06-04 16:07:40 -0700 | [diff] [blame] | 3279 | mutex_lock(&memcg_slab_mutex); |
| 3280 | list_for_each_entry_safe(params, tmp, &memcg->memcg_slab_caches, list) { |
Glauber Costa | 1f458cb | 2012-12-18 14:22:50 -0800 | [diff] [blame] | 3281 | cachep = memcg_params_to_cache(params); |
Vladimir Davydov | bd67314 | 2014-06-04 16:07:40 -0700 | [diff] [blame] | 3282 | kmem_cache_shrink(cachep); |
| 3283 | if (atomic_read(&cachep->memcg_params->nr_pages) == 0) |
| 3284 | memcg_kmem_destroy_cache(cachep); |
Glauber Costa | 1f458cb | 2012-12-18 14:22:50 -0800 | [diff] [blame] | 3285 | } |
Vladimir Davydov | bd67314 | 2014-06-04 16:07:40 -0700 | [diff] [blame] | 3286 | mutex_unlock(&memcg_slab_mutex); |
Glauber Costa | 1f458cb | 2012-12-18 14:22:50 -0800 | [diff] [blame] | 3287 | } |
| 3288 | |
Vladimir Davydov | 5722d09 | 2014-04-07 15:39:24 -0700 | [diff] [blame] | 3289 | struct create_work { |
| 3290 | struct mem_cgroup *memcg; |
| 3291 | struct kmem_cache *cachep; |
| 3292 | struct work_struct work; |
| 3293 | }; |
| 3294 | |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 3295 | static void memcg_create_cache_work_func(struct work_struct *w) |
| 3296 | { |
Vladimir Davydov | 5722d09 | 2014-04-07 15:39:24 -0700 | [diff] [blame] | 3297 | struct create_work *cw = container_of(w, struct create_work, work); |
| 3298 | struct mem_cgroup *memcg = cw->memcg; |
| 3299 | struct kmem_cache *cachep = cw->cachep; |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 3300 | |
Vladimir Davydov | bd67314 | 2014-06-04 16:07:40 -0700 | [diff] [blame] | 3301 | mutex_lock(&memcg_slab_mutex); |
| 3302 | memcg_kmem_create_cache(memcg, cachep); |
| 3303 | mutex_unlock(&memcg_slab_mutex); |
| 3304 | |
Vladimir Davydov | 5722d09 | 2014-04-07 15:39:24 -0700 | [diff] [blame] | 3305 | css_put(&memcg->css); |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 3306 | kfree(cw); |
| 3307 | } |
| 3308 | |
| 3309 | /* |
| 3310 | * Enqueue the creation of a per-memcg kmem_cache. |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 3311 | */ |
Glauber Costa | 0e9d92f | 2012-12-18 14:22:42 -0800 | [diff] [blame] | 3312 | static void __memcg_create_cache_enqueue(struct mem_cgroup *memcg, |
| 3313 | struct kmem_cache *cachep) |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 3314 | { |
| 3315 | struct create_work *cw; |
| 3316 | |
| 3317 | cw = kmalloc(sizeof(struct create_work), GFP_NOWAIT); |
Li Zefan | ca0dde9 | 2013-04-29 15:08:57 -0700 | [diff] [blame] | 3318 | if (cw == NULL) { |
| 3319 | css_put(&memcg->css); |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 3320 | return; |
| 3321 | } |
| 3322 | |
| 3323 | cw->memcg = memcg; |
| 3324 | cw->cachep = cachep; |
| 3325 | |
| 3326 | INIT_WORK(&cw->work, memcg_create_cache_work_func); |
| 3327 | schedule_work(&cw->work); |
| 3328 | } |
| 3329 | |
Glauber Costa | 0e9d92f | 2012-12-18 14:22:42 -0800 | [diff] [blame] | 3330 | static void memcg_create_cache_enqueue(struct mem_cgroup *memcg, |
| 3331 | struct kmem_cache *cachep) |
| 3332 | { |
| 3333 | /* |
| 3334 | * We need to stop accounting when we kmalloc, because if the |
| 3335 | * corresponding kmalloc cache is not yet created, the first allocation |
| 3336 | * in __memcg_create_cache_enqueue will recurse. |
| 3337 | * |
| 3338 | * However, it is better to enclose the whole function. Depending on |
| 3339 | * the debugging options enabled, INIT_WORK(), for instance, can |
| 3340 | * trigger an allocation. This too, will make us recurse. Because at |
| 3341 | * this point we can't allow ourselves back into memcg_kmem_get_cache, |
| 3342 | * the safest choice is to do it like this, wrapping the whole function. |
| 3343 | */ |
| 3344 | memcg_stop_kmem_account(); |
| 3345 | __memcg_create_cache_enqueue(memcg, cachep); |
| 3346 | memcg_resume_kmem_account(); |
| 3347 | } |
Vladimir Davydov | c67a8a6 | 2014-06-04 16:07:39 -0700 | [diff] [blame] | 3348 | |
| 3349 | int __memcg_charge_slab(struct kmem_cache *cachep, gfp_t gfp, int order) |
| 3350 | { |
| 3351 | int res; |
| 3352 | |
| 3353 | res = memcg_charge_kmem(cachep->memcg_params->memcg, gfp, |
| 3354 | PAGE_SIZE << order); |
| 3355 | if (!res) |
| 3356 | atomic_add(1 << order, &cachep->memcg_params->nr_pages); |
| 3357 | return res; |
| 3358 | } |
| 3359 | |
| 3360 | void __memcg_uncharge_slab(struct kmem_cache *cachep, int order) |
| 3361 | { |
| 3362 | memcg_uncharge_kmem(cachep->memcg_params->memcg, PAGE_SIZE << order); |
| 3363 | atomic_sub(1 << order, &cachep->memcg_params->nr_pages); |
| 3364 | } |
| 3365 | |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 3366 | /* |
| 3367 | * Return the kmem_cache we're supposed to use for a slab allocation. |
| 3368 | * We try to use the current memcg's version of the cache. |
| 3369 | * |
| 3370 | * If the cache does not exist yet, if we are the first user of it, |
| 3371 | * we either create it immediately, if possible, or create it asynchronously |
| 3372 | * in a workqueue. |
| 3373 | * In the latter case, we will let the current allocation go through with |
| 3374 | * the original cache. |
| 3375 | * |
| 3376 | * Can't be called in interrupt context or from kernel threads. |
| 3377 | * This function needs to be called with rcu_read_lock() held. |
| 3378 | */ |
| 3379 | struct kmem_cache *__memcg_kmem_get_cache(struct kmem_cache *cachep, |
| 3380 | gfp_t gfp) |
| 3381 | { |
| 3382 | struct mem_cgroup *memcg; |
Vladimir Davydov | 959c896 | 2014-01-23 15:52:59 -0800 | [diff] [blame] | 3383 | struct kmem_cache *memcg_cachep; |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 3384 | |
| 3385 | VM_BUG_ON(!cachep->memcg_params); |
| 3386 | VM_BUG_ON(!cachep->memcg_params->is_root_cache); |
| 3387 | |
Glauber Costa | 0e9d92f | 2012-12-18 14:22:42 -0800 | [diff] [blame] | 3388 | if (!current->mm || current->memcg_kmem_skip_account) |
| 3389 | return cachep; |
| 3390 | |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 3391 | rcu_read_lock(); |
| 3392 | memcg = mem_cgroup_from_task(rcu_dereference(current->mm->owner)); |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 3393 | |
| 3394 | if (!memcg_can_account_kmem(memcg)) |
Li Zefan | ca0dde9 | 2013-04-29 15:08:57 -0700 | [diff] [blame] | 3395 | goto out; |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 3396 | |
Vladimir Davydov | 959c896 | 2014-01-23 15:52:59 -0800 | [diff] [blame] | 3397 | memcg_cachep = cache_from_memcg_idx(cachep, memcg_cache_id(memcg)); |
| 3398 | if (likely(memcg_cachep)) { |
| 3399 | cachep = memcg_cachep; |
Li Zefan | ca0dde9 | 2013-04-29 15:08:57 -0700 | [diff] [blame] | 3400 | goto out; |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 3401 | } |
| 3402 | |
Li Zefan | ca0dde9 | 2013-04-29 15:08:57 -0700 | [diff] [blame] | 3403 | /* The corresponding put will be done in the workqueue. */ |
| 3404 | if (!css_tryget(&memcg->css)) |
| 3405 | goto out; |
| 3406 | rcu_read_unlock(); |
| 3407 | |
| 3408 | /* |
| 3409 | * If we are in a safe context (can wait, and not in interrupt |
| 3410 | * context), we could be be predictable and return right away. |
| 3411 | * This would guarantee that the allocation being performed |
| 3412 | * already belongs in the new cache. |
| 3413 | * |
| 3414 | * However, there are some clashes that can arrive from locking. |
| 3415 | * For instance, because we acquire the slab_mutex while doing |
| 3416 | * kmem_cache_dup, this means no further allocation could happen |
| 3417 | * with the slab_mutex held. |
| 3418 | * |
| 3419 | * Also, because cache creation issue get_online_cpus(), this |
| 3420 | * creates a lock chain: memcg_slab_mutex -> cpu_hotplug_mutex, |
| 3421 | * that ends up reversed during cpu hotplug. (cpuset allocates |
| 3422 | * a bunch of GFP_KERNEL memory during cpuup). Due to all that, |
| 3423 | * better to defer everything. |
| 3424 | */ |
| 3425 | memcg_create_cache_enqueue(memcg, cachep); |
| 3426 | return cachep; |
| 3427 | out: |
| 3428 | rcu_read_unlock(); |
| 3429 | return cachep; |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 3430 | } |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 3431 | |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 3432 | /* |
| 3433 | * We need to verify if the allocation against current->mm->owner's memcg is |
| 3434 | * possible for the given order. But the page is not allocated yet, so we'll |
| 3435 | * need a further commit step to do the final arrangements. |
| 3436 | * |
| 3437 | * It is possible for the task to switch cgroups in this mean time, so at |
| 3438 | * commit time, we can't rely on task conversion any longer. We'll then use |
| 3439 | * the handle argument to return to the caller which cgroup we should commit |
| 3440 | * against. We could also return the memcg directly and avoid the pointer |
| 3441 | * passing, but a boolean return value gives better semantics considering |
| 3442 | * the compiled-out case as well. |
| 3443 | * |
| 3444 | * Returning true means the allocation is possible. |
| 3445 | */ |
| 3446 | bool |
| 3447 | __memcg_kmem_newpage_charge(gfp_t gfp, struct mem_cgroup **_memcg, int order) |
| 3448 | { |
| 3449 | struct mem_cgroup *memcg; |
| 3450 | int ret; |
| 3451 | |
| 3452 | *_memcg = NULL; |
Glauber Costa | 6d42c23 | 2013-07-08 16:00:00 -0700 | [diff] [blame] | 3453 | |
| 3454 | /* |
| 3455 | * Disabling accounting is only relevant for some specific memcg |
| 3456 | * internal allocations. Therefore we would initially not have such |
Vladimir Davydov | 5238343 | 2014-06-04 16:06:39 -0700 | [diff] [blame] | 3457 | * check here, since direct calls to the page allocator that are |
| 3458 | * accounted to kmemcg (alloc_kmem_pages and friends) only happen |
| 3459 | * outside memcg core. We are mostly concerned with cache allocations, |
| 3460 | * and by having this test at memcg_kmem_get_cache, we are already able |
| 3461 | * to relay the allocation to the root cache and bypass the memcg cache |
| 3462 | * altogether. |
Glauber Costa | 6d42c23 | 2013-07-08 16:00:00 -0700 | [diff] [blame] | 3463 | * |
| 3464 | * There is one exception, though: the SLUB allocator does not create |
| 3465 | * large order caches, but rather service large kmallocs directly from |
| 3466 | * the page allocator. Therefore, the following sequence when backed by |
| 3467 | * the SLUB allocator: |
| 3468 | * |
Andrew Morton | f894ffa | 2013-09-12 15:13:35 -0700 | [diff] [blame] | 3469 | * memcg_stop_kmem_account(); |
| 3470 | * kmalloc(<large_number>) |
| 3471 | * memcg_resume_kmem_account(); |
Glauber Costa | 6d42c23 | 2013-07-08 16:00:00 -0700 | [diff] [blame] | 3472 | * |
| 3473 | * would effectively ignore the fact that we should skip accounting, |
| 3474 | * since it will drive us directly to this function without passing |
| 3475 | * through the cache selector memcg_kmem_get_cache. Such large |
| 3476 | * allocations are extremely rare but can happen, for instance, for the |
| 3477 | * cache arrays. We bring this test here. |
| 3478 | */ |
| 3479 | if (!current->mm || current->memcg_kmem_skip_account) |
| 3480 | return true; |
| 3481 | |
Johannes Weiner | df38197 | 2014-04-07 15:37:43 -0700 | [diff] [blame] | 3482 | memcg = get_mem_cgroup_from_mm(current->mm); |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 3483 | |
| 3484 | if (!memcg_can_account_kmem(memcg)) { |
| 3485 | css_put(&memcg->css); |
| 3486 | return true; |
| 3487 | } |
| 3488 | |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 3489 | ret = memcg_charge_kmem(memcg, gfp, PAGE_SIZE << order); |
| 3490 | if (!ret) |
| 3491 | *_memcg = memcg; |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 3492 | |
| 3493 | css_put(&memcg->css); |
| 3494 | return (ret == 0); |
| 3495 | } |
| 3496 | |
| 3497 | void __memcg_kmem_commit_charge(struct page *page, struct mem_cgroup *memcg, |
| 3498 | int order) |
| 3499 | { |
| 3500 | struct page_cgroup *pc; |
| 3501 | |
| 3502 | VM_BUG_ON(mem_cgroup_is_root(memcg)); |
| 3503 | |
| 3504 | /* The page allocation failed. Revert */ |
| 3505 | if (!page) { |
| 3506 | memcg_uncharge_kmem(memcg, PAGE_SIZE << order); |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 3507 | return; |
| 3508 | } |
| 3509 | |
| 3510 | pc = lookup_page_cgroup(page); |
| 3511 | lock_page_cgroup(pc); |
| 3512 | pc->mem_cgroup = memcg; |
| 3513 | SetPageCgroupUsed(pc); |
| 3514 | unlock_page_cgroup(pc); |
| 3515 | } |
| 3516 | |
| 3517 | void __memcg_kmem_uncharge_pages(struct page *page, int order) |
| 3518 | { |
| 3519 | struct mem_cgroup *memcg = NULL; |
| 3520 | struct page_cgroup *pc; |
| 3521 | |
| 3522 | |
| 3523 | pc = lookup_page_cgroup(page); |
| 3524 | /* |
| 3525 | * Fast unlocked return. Theoretically might have changed, have to |
| 3526 | * check again after locking. |
| 3527 | */ |
| 3528 | if (!PageCgroupUsed(pc)) |
| 3529 | return; |
| 3530 | |
| 3531 | lock_page_cgroup(pc); |
| 3532 | if (PageCgroupUsed(pc)) { |
| 3533 | memcg = pc->mem_cgroup; |
| 3534 | ClearPageCgroupUsed(pc); |
| 3535 | } |
| 3536 | unlock_page_cgroup(pc); |
| 3537 | |
| 3538 | /* |
| 3539 | * We trust that only if there is a memcg associated with the page, it |
| 3540 | * is a valid allocation |
| 3541 | */ |
| 3542 | if (!memcg) |
| 3543 | return; |
| 3544 | |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 3545 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 3546 | memcg_uncharge_kmem(memcg, PAGE_SIZE << order); |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 3547 | } |
Glauber Costa | 1f458cb | 2012-12-18 14:22:50 -0800 | [diff] [blame] | 3548 | #else |
| 3549 | static inline void mem_cgroup_destroy_all_caches(struct mem_cgroup *memcg) |
| 3550 | { |
| 3551 | } |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 3552 | #endif /* CONFIG_MEMCG_KMEM */ |
| 3553 | |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 3554 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
| 3555 | |
Kirill A. Shutemov | a0db00f | 2012-05-29 15:06:56 -0700 | [diff] [blame] | 3556 | #define PCGF_NOCOPY_AT_SPLIT (1 << PCG_LOCK | 1 << PCG_MIGRATION) |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 3557 | /* |
| 3558 | * 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] | 3559 | * zone->lru_lock, 'splitting on pmd' and compound_lock. |
| 3560 | * charge/uncharge will be never happen and move_account() is done under |
| 3561 | * compound_lock(), so we don't have to take care of races. |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 3562 | */ |
KAMEZAWA Hiroyuki | e94c8a9 | 2012-01-12 17:18:20 -0800 | [diff] [blame] | 3563 | void mem_cgroup_split_huge_fixup(struct page *head) |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 3564 | { |
| 3565 | struct page_cgroup *head_pc = lookup_page_cgroup(head); |
KAMEZAWA Hiroyuki | e94c8a9 | 2012-01-12 17:18:20 -0800 | [diff] [blame] | 3566 | struct page_cgroup *pc; |
David Rientjes | b070e65 | 2013-05-07 16:18:09 -0700 | [diff] [blame] | 3567 | struct mem_cgroup *memcg; |
KAMEZAWA Hiroyuki | e94c8a9 | 2012-01-12 17:18:20 -0800 | [diff] [blame] | 3568 | int i; |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 3569 | |
KAMEZAWA Hiroyuki | 3d37c4a | 2011-01-25 15:07:28 -0800 | [diff] [blame] | 3570 | if (mem_cgroup_disabled()) |
| 3571 | return; |
David Rientjes | b070e65 | 2013-05-07 16:18:09 -0700 | [diff] [blame] | 3572 | |
| 3573 | memcg = head_pc->mem_cgroup; |
KAMEZAWA Hiroyuki | e94c8a9 | 2012-01-12 17:18:20 -0800 | [diff] [blame] | 3574 | for (i = 1; i < HPAGE_PMD_NR; i++) { |
| 3575 | pc = head_pc + i; |
David Rientjes | b070e65 | 2013-05-07 16:18:09 -0700 | [diff] [blame] | 3576 | pc->mem_cgroup = memcg; |
KAMEZAWA Hiroyuki | e94c8a9 | 2012-01-12 17:18:20 -0800 | [diff] [blame] | 3577 | smp_wmb();/* see __commit_charge() */ |
KAMEZAWA Hiroyuki | e94c8a9 | 2012-01-12 17:18:20 -0800 | [diff] [blame] | 3578 | pc->flags = head_pc->flags & ~PCGF_NOCOPY_AT_SPLIT; |
| 3579 | } |
David Rientjes | b070e65 | 2013-05-07 16:18:09 -0700 | [diff] [blame] | 3580 | __this_cpu_sub(memcg->stat->count[MEM_CGROUP_STAT_RSS_HUGE], |
| 3581 | HPAGE_PMD_NR); |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 3582 | } |
Hugh Dickins | 12d2710 | 2012-01-12 17:19:52 -0800 | [diff] [blame] | 3583 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 3584 | |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3585 | /** |
Johannes Weiner | de3638d | 2011-03-23 16:42:28 -0700 | [diff] [blame] | 3586 | * mem_cgroup_move_account - move account of the page |
Johannes Weiner | 5564e88 | 2011-03-23 16:42:29 -0700 | [diff] [blame] | 3587 | * @page: the page |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 3588 | * @nr_pages: number of regular pages (>1 for huge pages) |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3589 | * @pc: page_cgroup of the page. |
| 3590 | * @from: mem_cgroup which the page is moved from. |
| 3591 | * @to: mem_cgroup which the page is moved to. @from != @to. |
| 3592 | * |
| 3593 | * The caller must confirm following. |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 3594 | * - page is not on LRU (isolate_page() is useful.) |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 3595 | * - compound_lock is held when nr_pages > 1 |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3596 | * |
KAMEZAWA Hiroyuki | 2f3479b | 2012-05-29 15:07:04 -0700 | [diff] [blame] | 3597 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" |
| 3598 | * from old cgroup. |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3599 | */ |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 3600 | static int mem_cgroup_move_account(struct page *page, |
| 3601 | unsigned int nr_pages, |
| 3602 | struct page_cgroup *pc, |
| 3603 | struct mem_cgroup *from, |
KAMEZAWA Hiroyuki | 2f3479b | 2012-05-29 15:07:04 -0700 | [diff] [blame] | 3604 | struct mem_cgroup *to) |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3605 | { |
Johannes Weiner | de3638d | 2011-03-23 16:42:28 -0700 | [diff] [blame] | 3606 | unsigned long flags; |
| 3607 | int ret; |
KAMEZAWA Hiroyuki | b2402857 | 2012-03-21 16:34:22 -0700 | [diff] [blame] | 3608 | bool anon = PageAnon(page); |
KAMEZAWA Hiroyuki | 987eba6 | 2011-01-20 14:44:25 -0800 | [diff] [blame] | 3609 | |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3610 | VM_BUG_ON(from == to); |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 3611 | VM_BUG_ON_PAGE(PageLRU(page), page); |
Johannes Weiner | de3638d | 2011-03-23 16:42:28 -0700 | [diff] [blame] | 3612 | /* |
| 3613 | * The page is isolated from LRU. So, collapse function |
| 3614 | * will not handle this page. But page splitting can happen. |
| 3615 | * Do this check under compound_page_lock(). The caller should |
| 3616 | * hold it. |
| 3617 | */ |
| 3618 | ret = -EBUSY; |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 3619 | if (nr_pages > 1 && !PageTransHuge(page)) |
Johannes Weiner | de3638d | 2011-03-23 16:42:28 -0700 | [diff] [blame] | 3620 | goto out; |
| 3621 | |
| 3622 | lock_page_cgroup(pc); |
| 3623 | |
| 3624 | ret = -EINVAL; |
| 3625 | if (!PageCgroupUsed(pc) || pc->mem_cgroup != from) |
| 3626 | goto unlock; |
| 3627 | |
KAMEZAWA Hiroyuki | 312734c0 | 2012-03-21 16:34:24 -0700 | [diff] [blame] | 3628 | move_lock_mem_cgroup(from, &flags); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3629 | |
Johannes Weiner | 59d1d25 | 2014-04-07 15:37:40 -0700 | [diff] [blame] | 3630 | if (!anon && page_mapped(page)) { |
| 3631 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], |
| 3632 | nr_pages); |
| 3633 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED], |
| 3634 | nr_pages); |
| 3635 | } |
Sha Zhengju | 3ea67d0 | 2013-09-12 15:13:53 -0700 | [diff] [blame] | 3636 | |
Johannes Weiner | 59d1d25 | 2014-04-07 15:37:40 -0700 | [diff] [blame] | 3637 | if (PageWriteback(page)) { |
| 3638 | __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK], |
| 3639 | nr_pages); |
| 3640 | __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK], |
| 3641 | nr_pages); |
| 3642 | } |
Sha Zhengju | 3ea67d0 | 2013-09-12 15:13:53 -0700 | [diff] [blame] | 3643 | |
David Rientjes | b070e65 | 2013-05-07 16:18:09 -0700 | [diff] [blame] | 3644 | mem_cgroup_charge_statistics(from, page, anon, -nr_pages); |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 3645 | |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 3646 | /* caller should have done css_get */ |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 3647 | pc->mem_cgroup = to; |
David Rientjes | b070e65 | 2013-05-07 16:18:09 -0700 | [diff] [blame] | 3648 | mem_cgroup_charge_statistics(to, page, anon, nr_pages); |
KAMEZAWA Hiroyuki | 312734c0 | 2012-03-21 16:34:24 -0700 | [diff] [blame] | 3649 | move_unlock_mem_cgroup(from, &flags); |
Johannes Weiner | de3638d | 2011-03-23 16:42:28 -0700 | [diff] [blame] | 3650 | ret = 0; |
| 3651 | unlock: |
Daisuke Nishimura | 57f9fd7d | 2009-12-15 16:47:11 -0800 | [diff] [blame] | 3652 | unlock_page_cgroup(pc); |
KAMEZAWA Hiroyuki | d2265e6 | 2010-03-10 15:22:31 -0800 | [diff] [blame] | 3653 | /* |
| 3654 | * check events |
| 3655 | */ |
Johannes Weiner | 5564e88 | 2011-03-23 16:42:29 -0700 | [diff] [blame] | 3656 | memcg_check_events(to, page); |
| 3657 | memcg_check_events(from, page); |
Johannes Weiner | de3638d | 2011-03-23 16:42:28 -0700 | [diff] [blame] | 3658 | out: |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3659 | return ret; |
| 3660 | } |
| 3661 | |
Michal Hocko | 2ef37d3 | 2012-10-26 13:37:30 +0200 | [diff] [blame] | 3662 | /** |
| 3663 | * mem_cgroup_move_parent - moves page to the parent group |
| 3664 | * @page: the page to move |
| 3665 | * @pc: page_cgroup of the page |
| 3666 | * @child: page's cgroup |
| 3667 | * |
| 3668 | * move charges to its parent or the root cgroup if the group has no |
| 3669 | * parent (aka use_hierarchy==0). |
| 3670 | * Although this might fail (get_page_unless_zero, isolate_lru_page or |
| 3671 | * mem_cgroup_move_account fails) the failure is always temporary and |
| 3672 | * it signals a race with a page removal/uncharge or migration. In the |
| 3673 | * first case the page is on the way out and it will vanish from the LRU |
| 3674 | * on the next attempt and the call should be retried later. |
| 3675 | * Isolation from the LRU fails only if page has been isolated from |
| 3676 | * the LRU since we looked at it and that usually means either global |
| 3677 | * reclaim or migration going on. The page will either get back to the |
| 3678 | * LRU or vanish. |
| 3679 | * Finaly mem_cgroup_move_account fails only if the page got uncharged |
| 3680 | * (!PageCgroupUsed) or moved to a different group. The page will |
| 3681 | * disappear in the next attempt. |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3682 | */ |
Johannes Weiner | 5564e88 | 2011-03-23 16:42:29 -0700 | [diff] [blame] | 3683 | static int mem_cgroup_move_parent(struct page *page, |
| 3684 | struct page_cgroup *pc, |
KAMEZAWA Hiroyuki | 6068bf0 | 2012-07-31 16:42:45 -0700 | [diff] [blame] | 3685 | struct mem_cgroup *child) |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3686 | { |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3687 | struct mem_cgroup *parent; |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 3688 | unsigned int nr_pages; |
Andrew Morton | 4be4489 | 2011-03-23 16:42:39 -0700 | [diff] [blame] | 3689 | unsigned long uninitialized_var(flags); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3690 | int ret; |
| 3691 | |
Michal Hocko | d842301 | 2012-10-26 13:37:29 +0200 | [diff] [blame] | 3692 | VM_BUG_ON(mem_cgroup_is_root(child)); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3693 | |
Daisuke Nishimura | 57f9fd7d | 2009-12-15 16:47:11 -0800 | [diff] [blame] | 3694 | ret = -EBUSY; |
| 3695 | if (!get_page_unless_zero(page)) |
| 3696 | goto out; |
| 3697 | if (isolate_lru_page(page)) |
| 3698 | goto put; |
KAMEZAWA Hiroyuki | 52dbb90 | 2011-01-25 15:07:29 -0800 | [diff] [blame] | 3699 | |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 3700 | nr_pages = hpage_nr_pages(page); |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 3701 | |
KAMEZAWA Hiroyuki | cc926f7 | 2012-05-29 15:07:04 -0700 | [diff] [blame] | 3702 | parent = parent_mem_cgroup(child); |
| 3703 | /* |
| 3704 | * If no parent, move charges to root cgroup. |
| 3705 | */ |
| 3706 | if (!parent) |
| 3707 | parent = root_mem_cgroup; |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 3708 | |
Michal Hocko | 2ef37d3 | 2012-10-26 13:37:30 +0200 | [diff] [blame] | 3709 | if (nr_pages > 1) { |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 3710 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); |
KAMEZAWA Hiroyuki | 987eba6 | 2011-01-20 14:44:25 -0800 | [diff] [blame] | 3711 | flags = compound_lock_irqsave(page); |
Michal Hocko | 2ef37d3 | 2012-10-26 13:37:30 +0200 | [diff] [blame] | 3712 | } |
KAMEZAWA Hiroyuki | 987eba6 | 2011-01-20 14:44:25 -0800 | [diff] [blame] | 3713 | |
KAMEZAWA Hiroyuki | cc926f7 | 2012-05-29 15:07:04 -0700 | [diff] [blame] | 3714 | ret = mem_cgroup_move_account(page, nr_pages, |
KAMEZAWA Hiroyuki | 2f3479b | 2012-05-29 15:07:04 -0700 | [diff] [blame] | 3715 | pc, child, parent); |
KAMEZAWA Hiroyuki | cc926f7 | 2012-05-29 15:07:04 -0700 | [diff] [blame] | 3716 | if (!ret) |
| 3717 | __mem_cgroup_cancel_local_charge(child, nr_pages); |
Jesper Juhl | 8dba474 | 2011-01-25 15:07:24 -0800 | [diff] [blame] | 3718 | |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 3719 | if (nr_pages > 1) |
KAMEZAWA Hiroyuki | 987eba6 | 2011-01-20 14:44:25 -0800 | [diff] [blame] | 3720 | compound_unlock_irqrestore(page, flags); |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 3721 | putback_lru_page(page); |
Daisuke Nishimura | 57f9fd7d | 2009-12-15 16:47:11 -0800 | [diff] [blame] | 3722 | put: |
Daisuke Nishimura | 40d5813 | 2009-01-15 13:51:12 -0800 | [diff] [blame] | 3723 | put_page(page); |
Daisuke Nishimura | 57f9fd7d | 2009-12-15 16:47:11 -0800 | [diff] [blame] | 3724 | out: |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3725 | return ret; |
| 3726 | } |
| 3727 | |
Michal Hocko | d715ae0 | 2014-04-07 15:37:46 -0700 | [diff] [blame] | 3728 | int mem_cgroup_charge_anon(struct page *page, |
Johannes Weiner | 1bec6b3 | 2014-04-07 15:37:41 -0700 | [diff] [blame] | 3729 | struct mm_struct *mm, gfp_t gfp_mask) |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 3730 | { |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 3731 | unsigned int nr_pages = 1; |
Johannes Weiner | 6d1fdc4 | 2014-04-07 15:37:45 -0700 | [diff] [blame] | 3732 | struct mem_cgroup *memcg; |
Johannes Weiner | 8493ae4 | 2011-02-01 15:52:44 -0800 | [diff] [blame] | 3733 | bool oom = true; |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 3734 | |
Johannes Weiner | 1bec6b3 | 2014-04-07 15:37:41 -0700 | [diff] [blame] | 3735 | if (mem_cgroup_disabled()) |
| 3736 | return 0; |
| 3737 | |
| 3738 | VM_BUG_ON_PAGE(page_mapped(page), page); |
| 3739 | VM_BUG_ON_PAGE(page->mapping && !PageAnon(page), page); |
| 3740 | VM_BUG_ON(!mm); |
| 3741 | |
Andrea Arcangeli | 37c2ac7 | 2011-01-13 15:47:16 -0800 | [diff] [blame] | 3742 | if (PageTransHuge(page)) { |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 3743 | nr_pages <<= compound_order(page); |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 3744 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); |
Johannes Weiner | 8493ae4 | 2011-02-01 15:52:44 -0800 | [diff] [blame] | 3745 | /* |
| 3746 | * Never OOM-kill a process for a huge page. The |
| 3747 | * fault handler will fall back to regular pages. |
| 3748 | */ |
| 3749 | oom = false; |
Andrea Arcangeli | 37c2ac7 | 2011-01-13 15:47:16 -0800 | [diff] [blame] | 3750 | } |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 3751 | |
Johannes Weiner | 6d1fdc4 | 2014-04-07 15:37:45 -0700 | [diff] [blame] | 3752 | memcg = mem_cgroup_try_charge_mm(mm, gfp_mask, nr_pages, oom); |
| 3753 | if (!memcg) |
| 3754 | return -ENOMEM; |
Johannes Weiner | 1bec6b3 | 2014-04-07 15:37:41 -0700 | [diff] [blame] | 3755 | __mem_cgroup_commit_charge(memcg, page, nr_pages, |
| 3756 | MEM_CGROUP_CHARGE_TYPE_ANON, false); |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 3757 | return 0; |
| 3758 | } |
| 3759 | |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 3760 | /* |
| 3761 | * While swap-in, try_charge -> commit or cancel, the page is locked. |
| 3762 | * And when try_charge() successfully returns, one refcnt to memcg without |
Uwe Kleine-König | 21ae295 | 2009-10-07 15:21:09 +0200 | [diff] [blame] | 3763 | * struct page_cgroup is acquired. This refcnt will be consumed by |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 3764 | * "commit()" or removed by "cancel()" |
| 3765 | */ |
Johannes Weiner | 0435a2f | 2012-07-31 16:45:43 -0700 | [diff] [blame] | 3766 | static int __mem_cgroup_try_charge_swapin(struct mm_struct *mm, |
| 3767 | struct page *page, |
| 3768 | gfp_t mask, |
| 3769 | struct mem_cgroup **memcgp) |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3770 | { |
Johannes Weiner | 6d1fdc4 | 2014-04-07 15:37:45 -0700 | [diff] [blame] | 3771 | struct mem_cgroup *memcg = NULL; |
Johannes Weiner | 90deb78 | 2012-07-31 16:45:47 -0700 | [diff] [blame] | 3772 | struct page_cgroup *pc; |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 3773 | int ret; |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3774 | |
Johannes Weiner | 90deb78 | 2012-07-31 16:45:47 -0700 | [diff] [blame] | 3775 | pc = lookup_page_cgroup(page); |
| 3776 | /* |
| 3777 | * Every swap fault against a single page tries to charge the |
| 3778 | * page, bail as early as possible. shmem_unuse() encounters |
| 3779 | * already charged pages, too. The USED bit is protected by |
| 3780 | * the page lock, which serializes swap cache removal, which |
| 3781 | * in turn serializes uncharging. |
| 3782 | */ |
| 3783 | if (PageCgroupUsed(pc)) |
Johannes Weiner | 6d1fdc4 | 2014-04-07 15:37:45 -0700 | [diff] [blame] | 3784 | goto out; |
| 3785 | if (do_swap_account) |
| 3786 | memcg = try_get_mem_cgroup_from_page(page); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3787 | if (!memcg) |
Johannes Weiner | 6d1fdc4 | 2014-04-07 15:37:45 -0700 | [diff] [blame] | 3788 | memcg = get_mem_cgroup_from_mm(mm); |
| 3789 | ret = mem_cgroup_try_charge(memcg, mask, 1, true); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3790 | css_put(&memcg->css); |
KAMEZAWA Hiroyuki | 38c5d72 | 2012-01-12 17:19:01 -0800 | [diff] [blame] | 3791 | if (ret == -EINTR) |
Johannes Weiner | 6d1fdc4 | 2014-04-07 15:37:45 -0700 | [diff] [blame] | 3792 | memcg = root_mem_cgroup; |
| 3793 | else if (ret) |
| 3794 | return ret; |
| 3795 | out: |
| 3796 | *memcgp = memcg; |
| 3797 | return 0; |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3798 | } |
| 3799 | |
Johannes Weiner | 0435a2f | 2012-07-31 16:45:43 -0700 | [diff] [blame] | 3800 | int mem_cgroup_try_charge_swapin(struct mm_struct *mm, struct page *page, |
| 3801 | gfp_t gfp_mask, struct mem_cgroup **memcgp) |
| 3802 | { |
Johannes Weiner | 6d1fdc4 | 2014-04-07 15:37:45 -0700 | [diff] [blame] | 3803 | if (mem_cgroup_disabled()) { |
| 3804 | *memcgp = NULL; |
Johannes Weiner | 0435a2f | 2012-07-31 16:45:43 -0700 | [diff] [blame] | 3805 | return 0; |
Johannes Weiner | 6d1fdc4 | 2014-04-07 15:37:45 -0700 | [diff] [blame] | 3806 | } |
Johannes Weiner | bdf4f4d | 2012-07-31 16:45:50 -0700 | [diff] [blame] | 3807 | /* |
| 3808 | * A racing thread's fault, or swapoff, may have already |
| 3809 | * updated the pte, and even removed page from swap cache: in |
| 3810 | * those cases unuse_pte()'s pte_same() test will fail; but |
| 3811 | * there's also a KSM case which does need to charge the page. |
| 3812 | */ |
| 3813 | if (!PageSwapCache(page)) { |
Johannes Weiner | 6d1fdc4 | 2014-04-07 15:37:45 -0700 | [diff] [blame] | 3814 | struct mem_cgroup *memcg; |
Johannes Weiner | bdf4f4d | 2012-07-31 16:45:50 -0700 | [diff] [blame] | 3815 | |
Johannes Weiner | 6d1fdc4 | 2014-04-07 15:37:45 -0700 | [diff] [blame] | 3816 | memcg = mem_cgroup_try_charge_mm(mm, gfp_mask, 1, true); |
| 3817 | if (!memcg) |
| 3818 | return -ENOMEM; |
| 3819 | *memcgp = memcg; |
| 3820 | return 0; |
Johannes Weiner | bdf4f4d | 2012-07-31 16:45:50 -0700 | [diff] [blame] | 3821 | } |
Johannes Weiner | 0435a2f | 2012-07-31 16:45:43 -0700 | [diff] [blame] | 3822 | return __mem_cgroup_try_charge_swapin(mm, page, gfp_mask, memcgp); |
| 3823 | } |
| 3824 | |
Johannes Weiner | 827a03d | 2012-07-31 16:45:36 -0700 | [diff] [blame] | 3825 | void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg) |
| 3826 | { |
| 3827 | if (mem_cgroup_disabled()) |
| 3828 | return; |
| 3829 | if (!memcg) |
| 3830 | return; |
| 3831 | __mem_cgroup_cancel_charge(memcg, 1); |
| 3832 | } |
| 3833 | |
Daisuke Nishimura | 83aae4c | 2009-04-02 16:57:48 -0700 | [diff] [blame] | 3834 | static void |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 3835 | __mem_cgroup_commit_charge_swapin(struct page *page, struct mem_cgroup *memcg, |
Daisuke Nishimura | 83aae4c | 2009-04-02 16:57:48 -0700 | [diff] [blame] | 3836 | enum charge_type ctype) |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 3837 | { |
Hirokazu Takahashi | f8d66542 | 2009-01-07 18:08:02 -0800 | [diff] [blame] | 3838 | if (mem_cgroup_disabled()) |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 3839 | return; |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 3840 | if (!memcg) |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 3841 | return; |
KAMEZAWA Hiroyuki | 5a6475a | 2011-03-23 16:42:42 -0700 | [diff] [blame] | 3842 | |
Johannes Weiner | ce587e6 | 2012-04-24 20:22:33 +0200 | [diff] [blame] | 3843 | __mem_cgroup_commit_charge(memcg, page, 1, ctype, true); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3844 | /* |
| 3845 | * Now swap is on-memory. This means this page may be |
| 3846 | * counted both as mem and swap....double count. |
KAMEZAWA Hiroyuki | 03f3c43 | 2009-01-07 18:08:31 -0800 | [diff] [blame] | 3847 | * Fix it by uncharging from memsw. Basically, this SwapCache is stable |
| 3848 | * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page() |
| 3849 | * may call delete_from_swap_cache() before reach here. |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3850 | */ |
KAMEZAWA Hiroyuki | 03f3c43 | 2009-01-07 18:08:31 -0800 | [diff] [blame] | 3851 | if (do_swap_account && PageSwapCache(page)) { |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3852 | swp_entry_t ent = {.val = page_private(page)}; |
Hugh Dickins | 8649300 | 2012-05-29 15:06:52 -0700 | [diff] [blame] | 3853 | mem_cgroup_uncharge_swap(ent); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3854 | } |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 3855 | } |
| 3856 | |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 3857 | void mem_cgroup_commit_charge_swapin(struct page *page, |
| 3858 | struct mem_cgroup *memcg) |
Daisuke Nishimura | 83aae4c | 2009-04-02 16:57:48 -0700 | [diff] [blame] | 3859 | { |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 3860 | __mem_cgroup_commit_charge_swapin(page, memcg, |
Kamezawa Hiroyuki | 41326c1 | 2012-07-31 16:41:40 -0700 | [diff] [blame] | 3861 | MEM_CGROUP_CHARGE_TYPE_ANON); |
Daisuke Nishimura | 83aae4c | 2009-04-02 16:57:48 -0700 | [diff] [blame] | 3862 | } |
| 3863 | |
Michal Hocko | d715ae0 | 2014-04-07 15:37:46 -0700 | [diff] [blame] | 3864 | int mem_cgroup_charge_file(struct page *page, struct mm_struct *mm, |
Johannes Weiner | 827a03d | 2012-07-31 16:45:36 -0700 | [diff] [blame] | 3865 | gfp_t gfp_mask) |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 3866 | { |
Johannes Weiner | 827a03d | 2012-07-31 16:45:36 -0700 | [diff] [blame] | 3867 | enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE; |
Johannes Weiner | 6d1fdc4 | 2014-04-07 15:37:45 -0700 | [diff] [blame] | 3868 | struct mem_cgroup *memcg; |
Johannes Weiner | 827a03d | 2012-07-31 16:45:36 -0700 | [diff] [blame] | 3869 | int ret; |
| 3870 | |
Hirokazu Takahashi | f8d66542 | 2009-01-07 18:08:02 -0800 | [diff] [blame] | 3871 | if (mem_cgroup_disabled()) |
Johannes Weiner | 827a03d | 2012-07-31 16:45:36 -0700 | [diff] [blame] | 3872 | return 0; |
| 3873 | if (PageCompound(page)) |
| 3874 | return 0; |
| 3875 | |
Johannes Weiner | 6d1fdc4 | 2014-04-07 15:37:45 -0700 | [diff] [blame] | 3876 | if (PageSwapCache(page)) { /* shmem */ |
Johannes Weiner | 0435a2f | 2012-07-31 16:45:43 -0700 | [diff] [blame] | 3877 | ret = __mem_cgroup_try_charge_swapin(mm, page, |
| 3878 | gfp_mask, &memcg); |
Johannes Weiner | 6d1fdc4 | 2014-04-07 15:37:45 -0700 | [diff] [blame] | 3879 | if (ret) |
| 3880 | return ret; |
| 3881 | __mem_cgroup_commit_charge_swapin(page, memcg, type); |
| 3882 | return 0; |
Johannes Weiner | 827a03d | 2012-07-31 16:45:36 -0700 | [diff] [blame] | 3883 | } |
Johannes Weiner | 6d1fdc4 | 2014-04-07 15:37:45 -0700 | [diff] [blame] | 3884 | |
Michal Hocko | 6f6acb0 | 2014-05-22 11:54:19 -0700 | [diff] [blame] | 3885 | memcg = mem_cgroup_try_charge_mm(mm, gfp_mask, 1, true); |
| 3886 | if (!memcg) |
| 3887 | return -ENOMEM; |
Johannes Weiner | 6d1fdc4 | 2014-04-07 15:37:45 -0700 | [diff] [blame] | 3888 | __mem_cgroup_commit_charge(memcg, page, 1, type, false); |
| 3889 | return 0; |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 3890 | } |
| 3891 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3892 | static void mem_cgroup_do_uncharge(struct mem_cgroup *memcg, |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 3893 | unsigned int nr_pages, |
| 3894 | const enum charge_type ctype) |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 3895 | { |
| 3896 | struct memcg_batch_info *batch = NULL; |
| 3897 | bool uncharge_memsw = true; |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 3898 | |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 3899 | /* If swapout, usage of swap doesn't decrease */ |
| 3900 | if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) |
| 3901 | uncharge_memsw = false; |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 3902 | |
| 3903 | batch = ¤t->memcg_batch; |
| 3904 | /* |
| 3905 | * In usual, we do css_get() when we remember memcg pointer. |
| 3906 | * But in this case, we keep res->usage until end of a series of |
| 3907 | * uncharges. Then, it's ok to ignore memcg's refcnt. |
| 3908 | */ |
| 3909 | if (!batch->memcg) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3910 | batch->memcg = memcg; |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 3911 | /* |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3912 | * do_batch > 0 when unmapping pages or inode invalidate/truncate. |
Lucas De Marchi | 25985ed | 2011-03-30 22:57:33 -0300 | [diff] [blame] | 3913 | * In those cases, all pages freed continuously can be expected to be in |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3914 | * the same cgroup and we have chance to coalesce uncharges. |
| 3915 | * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE) |
| 3916 | * because we want to do uncharge as soon as possible. |
| 3917 | */ |
| 3918 | |
| 3919 | if (!batch->do_batch || test_thread_flag(TIF_MEMDIE)) |
| 3920 | goto direct_uncharge; |
| 3921 | |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 3922 | if (nr_pages > 1) |
Andrea Arcangeli | ec16851 | 2011-01-13 15:46:56 -0800 | [diff] [blame] | 3923 | goto direct_uncharge; |
| 3924 | |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3925 | /* |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 3926 | * In typical case, batch->memcg == mem. This means we can |
| 3927 | * merge a series of uncharges to an uncharge of res_counter. |
| 3928 | * If not, we uncharge res_counter ony by one. |
| 3929 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3930 | if (batch->memcg != memcg) |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 3931 | goto direct_uncharge; |
| 3932 | /* remember freed charge and uncharge it later */ |
Johannes Weiner | 7ffd4ca | 2011-03-23 16:42:35 -0700 | [diff] [blame] | 3933 | batch->nr_pages++; |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 3934 | if (uncharge_memsw) |
Johannes Weiner | 7ffd4ca | 2011-03-23 16:42:35 -0700 | [diff] [blame] | 3935 | batch->memsw_nr_pages++; |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 3936 | return; |
| 3937 | direct_uncharge: |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3938 | res_counter_uncharge(&memcg->res, nr_pages * PAGE_SIZE); |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 3939 | if (uncharge_memsw) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3940 | res_counter_uncharge(&memcg->memsw, nr_pages * PAGE_SIZE); |
| 3941 | if (unlikely(batch->memcg != memcg)) |
| 3942 | memcg_oom_recover(memcg); |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 3943 | } |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 3944 | |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 3945 | /* |
KAMEZAWA Hiroyuki | 69029cd | 2008-07-25 01:47:14 -0700 | [diff] [blame] | 3946 | * uncharge if !page_mapped(page) |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 3947 | */ |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3948 | static struct mem_cgroup * |
Johannes Weiner | 0030f53 | 2012-07-31 16:45:25 -0700 | [diff] [blame] | 3949 | __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype, |
| 3950 | bool end_migration) |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 3951 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3952 | struct mem_cgroup *memcg = NULL; |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 3953 | unsigned int nr_pages = 1; |
| 3954 | struct page_cgroup *pc; |
KAMEZAWA Hiroyuki | b2402857 | 2012-03-21 16:34:22 -0700 | [diff] [blame] | 3955 | bool anon; |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 3956 | |
Hirokazu Takahashi | f8d66542 | 2009-01-07 18:08:02 -0800 | [diff] [blame] | 3957 | if (mem_cgroup_disabled()) |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3958 | return NULL; |
Balbir Singh | 4077960 | 2008-04-04 14:29:59 -0700 | [diff] [blame] | 3959 | |
Andrea Arcangeli | 37c2ac7 | 2011-01-13 15:47:16 -0800 | [diff] [blame] | 3960 | if (PageTransHuge(page)) { |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 3961 | nr_pages <<= compound_order(page); |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 3962 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); |
Andrea Arcangeli | 37c2ac7 | 2011-01-13 15:47:16 -0800 | [diff] [blame] | 3963 | } |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 3964 | /* |
Balbir Singh | 3c541e1 | 2008-02-07 00:14:41 -0800 | [diff] [blame] | 3965 | * Check if our page_cgroup is valid |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 3966 | */ |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 3967 | pc = lookup_page_cgroup(page); |
Johannes Weiner | cfa4494 | 2012-01-12 17:18:38 -0800 | [diff] [blame] | 3968 | if (unlikely(!PageCgroupUsed(pc))) |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3969 | return NULL; |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 3970 | |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 3971 | lock_page_cgroup(pc); |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 3972 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3973 | memcg = pc->mem_cgroup; |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3974 | |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 3975 | if (!PageCgroupUsed(pc)) |
| 3976 | goto unlock_out; |
| 3977 | |
KAMEZAWA Hiroyuki | b2402857 | 2012-03-21 16:34:22 -0700 | [diff] [blame] | 3978 | anon = PageAnon(page); |
| 3979 | |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 3980 | switch (ctype) { |
Kamezawa Hiroyuki | 41326c1 | 2012-07-31 16:41:40 -0700 | [diff] [blame] | 3981 | case MEM_CGROUP_CHARGE_TYPE_ANON: |
KAMEZAWA Hiroyuki | 2ff76f1 | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 3982 | /* |
| 3983 | * Generally PageAnon tells if it's the anon statistics to be |
| 3984 | * updated; but sometimes e.g. mem_cgroup_uncharge_page() is |
| 3985 | * used before page reached the stage of being marked PageAnon. |
| 3986 | */ |
KAMEZAWA Hiroyuki | b2402857 | 2012-03-21 16:34:22 -0700 | [diff] [blame] | 3987 | anon = true; |
| 3988 | /* fallthrough */ |
KAMEZAWA Hiroyuki | 8a9478c | 2009-06-17 16:27:17 -0700 | [diff] [blame] | 3989 | case MEM_CGROUP_CHARGE_TYPE_DROP: |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3990 | /* See mem_cgroup_prepare_migration() */ |
Johannes Weiner | 0030f53 | 2012-07-31 16:45:25 -0700 | [diff] [blame] | 3991 | if (page_mapped(page)) |
| 3992 | goto unlock_out; |
| 3993 | /* |
| 3994 | * Pages under migration may not be uncharged. But |
| 3995 | * end_migration() /must/ be the one uncharging the |
| 3996 | * unused post-migration page and so it has to call |
| 3997 | * here with the migration bit still set. See the |
| 3998 | * res_counter handling below. |
| 3999 | */ |
| 4000 | if (!end_migration && PageCgroupMigration(pc)) |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 4001 | goto unlock_out; |
| 4002 | break; |
| 4003 | case MEM_CGROUP_CHARGE_TYPE_SWAPOUT: |
| 4004 | if (!PageAnon(page)) { /* Shared memory */ |
| 4005 | if (page->mapping && !page_is_file_cache(page)) |
| 4006 | goto unlock_out; |
| 4007 | } else if (page_mapped(page)) /* Anon */ |
| 4008 | goto unlock_out; |
| 4009 | break; |
| 4010 | default: |
| 4011 | break; |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 4012 | } |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 4013 | |
David Rientjes | b070e65 | 2013-05-07 16:18:09 -0700 | [diff] [blame] | 4014 | mem_cgroup_charge_statistics(memcg, page, anon, -nr_pages); |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 4015 | |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 4016 | ClearPageCgroupUsed(pc); |
KAMEZAWA Hiroyuki | 544122e | 2009-01-07 18:08:34 -0800 | [diff] [blame] | 4017 | /* |
| 4018 | * pc->mem_cgroup is not cleared here. It will be accessed when it's |
| 4019 | * freed from LRU. This is safe because uncharged page is expected not |
| 4020 | * to be reused (freed soon). Exception is SwapCache, it's handled by |
| 4021 | * special functions. |
| 4022 | */ |
Hugh Dickins | b9c565d | 2008-03-04 14:29:11 -0800 | [diff] [blame] | 4023 | |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 4024 | unlock_page_cgroup(pc); |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 4025 | /* |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4026 | * even after unlock, we have memcg->res.usage here and this memcg |
Li Zefan | 4050377 | 2013-07-08 16:00:34 -0700 | [diff] [blame] | 4027 | * will never be freed, so it's safe to call css_get(). |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 4028 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4029 | memcg_check_events(memcg, page); |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 4030 | if (do_swap_account && ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT) { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4031 | mem_cgroup_swap_statistics(memcg, true); |
Li Zefan | 4050377 | 2013-07-08 16:00:34 -0700 | [diff] [blame] | 4032 | css_get(&memcg->css); |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 4033 | } |
Johannes Weiner | 0030f53 | 2012-07-31 16:45:25 -0700 | [diff] [blame] | 4034 | /* |
| 4035 | * Migration does not charge the res_counter for the |
| 4036 | * replacement page, so leave it alone when phasing out the |
| 4037 | * page that is unused after the migration. |
| 4038 | */ |
| 4039 | if (!end_migration && !mem_cgroup_is_root(memcg)) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4040 | mem_cgroup_do_uncharge(memcg, nr_pages, ctype); |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 4041 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4042 | return memcg; |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 4043 | |
| 4044 | unlock_out: |
| 4045 | unlock_page_cgroup(pc); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4046 | return NULL; |
Balbir Singh | 3c541e1 | 2008-02-07 00:14:41 -0800 | [diff] [blame] | 4047 | } |
| 4048 | |
KAMEZAWA Hiroyuki | 69029cd | 2008-07-25 01:47:14 -0700 | [diff] [blame] | 4049 | void mem_cgroup_uncharge_page(struct page *page) |
| 4050 | { |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 4051 | /* early check. */ |
| 4052 | if (page_mapped(page)) |
| 4053 | return; |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 4054 | VM_BUG_ON_PAGE(page->mapping && !PageAnon(page), page); |
Johannes Weiner | 28ccddf | 2013-05-24 15:55:15 -0700 | [diff] [blame] | 4055 | /* |
| 4056 | * If the page is in swap cache, uncharge should be deferred |
| 4057 | * to the swap path, which also properly accounts swap usage |
| 4058 | * and handles memcg lifetime. |
| 4059 | * |
| 4060 | * Note that this check is not stable and reclaim may add the |
| 4061 | * page to swap cache at any time after this. However, if the |
| 4062 | * page is not in swap cache by the time page->mapcount hits |
| 4063 | * 0, there won't be any page table references to the swap |
| 4064 | * slot, and reclaim will free it and not actually write the |
| 4065 | * page to disk. |
| 4066 | */ |
Johannes Weiner | 0c59b89 | 2012-07-31 16:45:31 -0700 | [diff] [blame] | 4067 | if (PageSwapCache(page)) |
| 4068 | return; |
Johannes Weiner | 0030f53 | 2012-07-31 16:45:25 -0700 | [diff] [blame] | 4069 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_ANON, false); |
KAMEZAWA Hiroyuki | 69029cd | 2008-07-25 01:47:14 -0700 | [diff] [blame] | 4070 | } |
| 4071 | |
| 4072 | void mem_cgroup_uncharge_cache_page(struct page *page) |
| 4073 | { |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 4074 | VM_BUG_ON_PAGE(page_mapped(page), page); |
| 4075 | VM_BUG_ON_PAGE(page->mapping, page); |
Johannes Weiner | 0030f53 | 2012-07-31 16:45:25 -0700 | [diff] [blame] | 4076 | __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE, false); |
KAMEZAWA Hiroyuki | 69029cd | 2008-07-25 01:47:14 -0700 | [diff] [blame] | 4077 | } |
| 4078 | |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 4079 | /* |
| 4080 | * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate. |
| 4081 | * In that cases, pages are freed continuously and we can expect pages |
| 4082 | * are in the same memcg. All these calls itself limits the number of |
| 4083 | * pages freed at once, then uncharge_start/end() is called properly. |
| 4084 | * This may be called prural(2) times in a context, |
| 4085 | */ |
| 4086 | |
| 4087 | void mem_cgroup_uncharge_start(void) |
| 4088 | { |
| 4089 | current->memcg_batch.do_batch++; |
| 4090 | /* We can do nest. */ |
| 4091 | if (current->memcg_batch.do_batch == 1) { |
| 4092 | current->memcg_batch.memcg = NULL; |
Johannes Weiner | 7ffd4ca | 2011-03-23 16:42:35 -0700 | [diff] [blame] | 4093 | current->memcg_batch.nr_pages = 0; |
| 4094 | current->memcg_batch.memsw_nr_pages = 0; |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 4095 | } |
| 4096 | } |
| 4097 | |
| 4098 | void mem_cgroup_uncharge_end(void) |
| 4099 | { |
| 4100 | struct memcg_batch_info *batch = ¤t->memcg_batch; |
| 4101 | |
| 4102 | if (!batch->do_batch) |
| 4103 | return; |
| 4104 | |
| 4105 | batch->do_batch--; |
| 4106 | if (batch->do_batch) /* If stacked, do nothing. */ |
| 4107 | return; |
| 4108 | |
| 4109 | if (!batch->memcg) |
| 4110 | return; |
| 4111 | /* |
| 4112 | * This "batch->memcg" is valid without any css_get/put etc... |
| 4113 | * bacause we hide charges behind us. |
| 4114 | */ |
Johannes Weiner | 7ffd4ca | 2011-03-23 16:42:35 -0700 | [diff] [blame] | 4115 | if (batch->nr_pages) |
| 4116 | res_counter_uncharge(&batch->memcg->res, |
| 4117 | batch->nr_pages * PAGE_SIZE); |
| 4118 | if (batch->memsw_nr_pages) |
| 4119 | res_counter_uncharge(&batch->memcg->memsw, |
| 4120 | batch->memsw_nr_pages * PAGE_SIZE); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 4121 | memcg_oom_recover(batch->memcg); |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 4122 | /* forget this pointer (for sanity check) */ |
| 4123 | batch->memcg = NULL; |
| 4124 | } |
| 4125 | |
Daisuke Nishimura | e767e05 | 2009-05-28 14:34:28 -0700 | [diff] [blame] | 4126 | #ifdef CONFIG_SWAP |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4127 | /* |
Daisuke Nishimura | e767e05 | 2009-05-28 14:34:28 -0700 | [diff] [blame] | 4128 | * called after __delete_from_swap_cache() and drop "page" account. |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4129 | * memcg information is recorded to swap_cgroup of "ent" |
| 4130 | */ |
KAMEZAWA Hiroyuki | 8a9478c | 2009-06-17 16:27:17 -0700 | [diff] [blame] | 4131 | void |
| 4132 | mem_cgroup_uncharge_swapcache(struct page *page, swp_entry_t ent, bool swapout) |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 4133 | { |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4134 | struct mem_cgroup *memcg; |
KAMEZAWA Hiroyuki | 8a9478c | 2009-06-17 16:27:17 -0700 | [diff] [blame] | 4135 | int ctype = MEM_CGROUP_CHARGE_TYPE_SWAPOUT; |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4136 | |
KAMEZAWA Hiroyuki | 8a9478c | 2009-06-17 16:27:17 -0700 | [diff] [blame] | 4137 | if (!swapout) /* this was a swap cache but the swap is unused ! */ |
| 4138 | ctype = MEM_CGROUP_CHARGE_TYPE_DROP; |
| 4139 | |
Johannes Weiner | 0030f53 | 2012-07-31 16:45:25 -0700 | [diff] [blame] | 4140 | memcg = __mem_cgroup_uncharge_common(page, ctype, false); |
KAMEZAWA Hiroyuki | 8a9478c | 2009-06-17 16:27:17 -0700 | [diff] [blame] | 4141 | |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 4142 | /* |
| 4143 | * record memcg information, if swapout && memcg != NULL, |
Li Zefan | 4050377 | 2013-07-08 16:00:34 -0700 | [diff] [blame] | 4144 | * css_get() was called in uncharge(). |
KAMEZAWA Hiroyuki | f75ca96 | 2010-08-10 18:03:02 -0700 | [diff] [blame] | 4145 | */ |
| 4146 | if (do_swap_account && swapout && memcg) |
Li Zefan | 34c00c3 | 2013-09-23 16:56:01 +0800 | [diff] [blame] | 4147 | swap_cgroup_record(ent, mem_cgroup_id(memcg)); |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 4148 | } |
Daisuke Nishimura | e767e05 | 2009-05-28 14:34:28 -0700 | [diff] [blame] | 4149 | #endif |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 4150 | |
Andrew Morton | c255a45 | 2012-07-31 16:43:02 -0700 | [diff] [blame] | 4151 | #ifdef CONFIG_MEMCG_SWAP |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4152 | /* |
| 4153 | * called from swap_entry_free(). remove record in swap_cgroup and |
| 4154 | * uncharge "memsw" account. |
| 4155 | */ |
| 4156 | void mem_cgroup_uncharge_swap(swp_entry_t ent) |
| 4157 | { |
| 4158 | struct mem_cgroup *memcg; |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 4159 | unsigned short id; |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4160 | |
| 4161 | if (!do_swap_account) |
| 4162 | return; |
| 4163 | |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 4164 | id = swap_cgroup_record(ent, 0); |
| 4165 | rcu_read_lock(); |
| 4166 | memcg = mem_cgroup_lookup(id); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4167 | if (memcg) { |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 4168 | /* |
| 4169 | * We uncharge this because swap is freed. |
| 4170 | * This memcg can be obsolete one. We avoid calling css_tryget |
| 4171 | */ |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 4172 | if (!mem_cgroup_is_root(memcg)) |
KAMEZAWA Hiroyuki | 4e64915 | 2009-10-01 15:44:11 -0700 | [diff] [blame] | 4173 | res_counter_uncharge(&memcg->memsw, PAGE_SIZE); |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 4174 | mem_cgroup_swap_statistics(memcg, false); |
Li Zefan | 4050377 | 2013-07-08 16:00:34 -0700 | [diff] [blame] | 4175 | css_put(&memcg->css); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4176 | } |
KAMEZAWA Hiroyuki | a3b2d69 | 2009-04-02 16:57:45 -0700 | [diff] [blame] | 4177 | rcu_read_unlock(); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4178 | } |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4179 | |
| 4180 | /** |
| 4181 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. |
| 4182 | * @entry: swap entry to be moved |
| 4183 | * @from: mem_cgroup which the entry is moved from |
| 4184 | * @to: mem_cgroup which the entry is moved to |
| 4185 | * |
| 4186 | * It succeeds only when the swap_cgroup's record for this entry is the same |
| 4187 | * as the mem_cgroup's id of @from. |
| 4188 | * |
| 4189 | * Returns 0 on success, -EINVAL on failure. |
| 4190 | * |
| 4191 | * The caller must have charged to @to, IOW, called res_counter_charge() about |
| 4192 | * both res and memsw, and called css_get(). |
| 4193 | */ |
| 4194 | static int mem_cgroup_move_swap_account(swp_entry_t entry, |
Hugh Dickins | e91cbb4 | 2012-05-29 15:06:51 -0700 | [diff] [blame] | 4195 | struct mem_cgroup *from, struct mem_cgroup *to) |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4196 | { |
| 4197 | unsigned short old_id, new_id; |
| 4198 | |
Li Zefan | 34c00c3 | 2013-09-23 16:56:01 +0800 | [diff] [blame] | 4199 | old_id = mem_cgroup_id(from); |
| 4200 | new_id = mem_cgroup_id(to); |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4201 | |
| 4202 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4203 | mem_cgroup_swap_statistics(from, false); |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4204 | mem_cgroup_swap_statistics(to, true); |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 4205 | /* |
| 4206 | * This function is only called from task migration context now. |
| 4207 | * It postpones res_counter and refcount handling till the end |
| 4208 | * of task migration(mem_cgroup_clear_mc()) for performance |
Li Zefan | 4050377 | 2013-07-08 16:00:34 -0700 | [diff] [blame] | 4209 | * improvement. But we cannot postpone css_get(to) because if |
| 4210 | * the process that has been moved to @to does swap-in, the |
| 4211 | * refcount of @to might be decreased to 0. |
| 4212 | * |
| 4213 | * We are in attach() phase, so the cgroup is guaranteed to be |
| 4214 | * alive, so we can just call css_get(). |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 4215 | */ |
Li Zefan | 4050377 | 2013-07-08 16:00:34 -0700 | [diff] [blame] | 4216 | css_get(&to->css); |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4217 | return 0; |
| 4218 | } |
| 4219 | return -EINVAL; |
| 4220 | } |
| 4221 | #else |
| 4222 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, |
Hugh Dickins | e91cbb4 | 2012-05-29 15:06:51 -0700 | [diff] [blame] | 4223 | struct mem_cgroup *from, struct mem_cgroup *to) |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4224 | { |
| 4225 | return -EINVAL; |
| 4226 | } |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4227 | #endif |
| 4228 | |
KAMEZAWA Hiroyuki | ae41be3 | 2008-02-07 00:14:10 -0800 | [diff] [blame] | 4229 | /* |
KAMEZAWA Hiroyuki | 01b1ae6 | 2009-01-07 18:07:50 -0800 | [diff] [blame] | 4230 | * Before starting migration, account PAGE_SIZE to mem_cgroup that the old |
| 4231 | * page belongs to. |
KAMEZAWA Hiroyuki | ae41be3 | 2008-02-07 00:14:10 -0800 | [diff] [blame] | 4232 | */ |
Johannes Weiner | 0030f53 | 2012-07-31 16:45:25 -0700 | [diff] [blame] | 4233 | void mem_cgroup_prepare_migration(struct page *page, struct page *newpage, |
| 4234 | struct mem_cgroup **memcgp) |
KAMEZAWA Hiroyuki | ae41be3 | 2008-02-07 00:14:10 -0800 | [diff] [blame] | 4235 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4236 | struct mem_cgroup *memcg = NULL; |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 4237 | unsigned int nr_pages = 1; |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 4238 | struct page_cgroup *pc; |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 4239 | enum charge_type ctype; |
Hugh Dickins | 8869b8f | 2008-03-04 14:29:09 -0800 | [diff] [blame] | 4240 | |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 4241 | *memcgp = NULL; |
KAMEZAWA Hiroyuki | 56039ef | 2011-03-23 16:42:19 -0700 | [diff] [blame] | 4242 | |
Hirokazu Takahashi | f8d66542 | 2009-01-07 18:08:02 -0800 | [diff] [blame] | 4243 | if (mem_cgroup_disabled()) |
Johannes Weiner | 0030f53 | 2012-07-31 16:45:25 -0700 | [diff] [blame] | 4244 | return; |
Balbir Singh | 4077960 | 2008-04-04 14:29:59 -0700 | [diff] [blame] | 4245 | |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 4246 | if (PageTransHuge(page)) |
| 4247 | nr_pages <<= compound_order(page); |
| 4248 | |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 4249 | pc = lookup_page_cgroup(page); |
| 4250 | lock_page_cgroup(pc); |
| 4251 | if (PageCgroupUsed(pc)) { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4252 | memcg = pc->mem_cgroup; |
| 4253 | css_get(&memcg->css); |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 4254 | /* |
| 4255 | * At migrating an anonymous page, its mapcount goes down |
| 4256 | * to 0 and uncharge() will be called. But, even if it's fully |
| 4257 | * unmapped, migration may fail and this page has to be |
| 4258 | * charged again. We set MIGRATION flag here and delay uncharge |
| 4259 | * until end_migration() is called |
| 4260 | * |
| 4261 | * Corner Case Thinking |
| 4262 | * A) |
| 4263 | * When the old page was mapped as Anon and it's unmap-and-freed |
| 4264 | * while migration was ongoing. |
| 4265 | * If unmap finds the old page, uncharge() of it will be delayed |
| 4266 | * until end_migration(). If unmap finds a new page, it's |
| 4267 | * uncharged when it make mapcount to be 1->0. If unmap code |
| 4268 | * finds swap_migration_entry, the new page will not be mapped |
| 4269 | * and end_migration() will find it(mapcount==0). |
| 4270 | * |
| 4271 | * B) |
| 4272 | * When the old page was mapped but migraion fails, the kernel |
| 4273 | * remaps it. A charge for it is kept by MIGRATION flag even |
| 4274 | * if mapcount goes down to 0. We can do remap successfully |
| 4275 | * without charging it again. |
| 4276 | * |
| 4277 | * C) |
| 4278 | * The "old" page is under lock_page() until the end of |
| 4279 | * migration, so, the old page itself will not be swapped-out. |
| 4280 | * If the new page is swapped out before end_migraton, our |
| 4281 | * hook to usual swap-out path will catch the event. |
| 4282 | */ |
| 4283 | if (PageAnon(page)) |
| 4284 | SetPageCgroupMigration(pc); |
Hugh Dickins | b9c565d | 2008-03-04 14:29:11 -0800 | [diff] [blame] | 4285 | } |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 4286 | unlock_page_cgroup(pc); |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 4287 | /* |
| 4288 | * If the page is not charged at this point, |
| 4289 | * we return here. |
| 4290 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4291 | if (!memcg) |
Johannes Weiner | 0030f53 | 2012-07-31 16:45:25 -0700 | [diff] [blame] | 4292 | return; |
KAMEZAWA Hiroyuki | 01b1ae6 | 2009-01-07 18:07:50 -0800 | [diff] [blame] | 4293 | |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 4294 | *memcgp = memcg; |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 4295 | /* |
| 4296 | * We charge new page before it's used/mapped. So, even if unlock_page() |
| 4297 | * is called before end_migration, we can catch all events on this new |
| 4298 | * page. In the case new page is migrated but not remapped, new page's |
| 4299 | * mapcount will be finally 0 and we call uncharge in end_migration(). |
| 4300 | */ |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 4301 | if (PageAnon(page)) |
Kamezawa Hiroyuki | 41326c1 | 2012-07-31 16:41:40 -0700 | [diff] [blame] | 4302 | ctype = MEM_CGROUP_CHARGE_TYPE_ANON; |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 4303 | else |
Johannes Weiner | 62ba744 | 2012-07-31 16:45:39 -0700 | [diff] [blame] | 4304 | ctype = MEM_CGROUP_CHARGE_TYPE_CACHE; |
Johannes Weiner | 0030f53 | 2012-07-31 16:45:25 -0700 | [diff] [blame] | 4305 | /* |
| 4306 | * The page is committed to the memcg, but it's not actually |
| 4307 | * charged to the res_counter since we plan on replacing the |
| 4308 | * old one and only one page is going to be left afterwards. |
| 4309 | */ |
Mel Gorman | b32967f | 2012-11-19 12:35:47 +0000 | [diff] [blame] | 4310 | __mem_cgroup_commit_charge(memcg, newpage, nr_pages, ctype, false); |
KAMEZAWA Hiroyuki | e8589cc | 2008-07-25 01:47:10 -0700 | [diff] [blame] | 4311 | } |
Hugh Dickins | fb59e9f | 2008-03-04 14:29:16 -0800 | [diff] [blame] | 4312 | |
KAMEZAWA Hiroyuki | 69029cd | 2008-07-25 01:47:14 -0700 | [diff] [blame] | 4313 | /* remove redundant charge if migration failed*/ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4314 | void mem_cgroup_end_migration(struct mem_cgroup *memcg, |
Daisuke Nishimura | 50de1dd | 2011-01-13 15:47:43 -0800 | [diff] [blame] | 4315 | struct page *oldpage, struct page *newpage, bool migration_ok) |
KAMEZAWA Hiroyuki | e8589cc | 2008-07-25 01:47:10 -0700 | [diff] [blame] | 4316 | { |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 4317 | struct page *used, *unused; |
KAMEZAWA Hiroyuki | 01b1ae6 | 2009-01-07 18:07:50 -0800 | [diff] [blame] | 4318 | struct page_cgroup *pc; |
KAMEZAWA Hiroyuki | b2402857 | 2012-03-21 16:34:22 -0700 | [diff] [blame] | 4319 | bool anon; |
KAMEZAWA Hiroyuki | 01b1ae6 | 2009-01-07 18:07:50 -0800 | [diff] [blame] | 4320 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4321 | if (!memcg) |
KAMEZAWA Hiroyuki | 01b1ae6 | 2009-01-07 18:07:50 -0800 | [diff] [blame] | 4322 | return; |
Tejun Heo | b25ed60 | 2012-11-05 09:16:59 -0800 | [diff] [blame] | 4323 | |
Daisuke Nishimura | 50de1dd | 2011-01-13 15:47:43 -0800 | [diff] [blame] | 4324 | if (!migration_ok) { |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 4325 | used = oldpage; |
| 4326 | unused = newpage; |
KAMEZAWA Hiroyuki | 01b1ae6 | 2009-01-07 18:07:50 -0800 | [diff] [blame] | 4327 | } else { |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 4328 | used = newpage; |
KAMEZAWA Hiroyuki | 01b1ae6 | 2009-01-07 18:07:50 -0800 | [diff] [blame] | 4329 | unused = oldpage; |
| 4330 | } |
Johannes Weiner | 0030f53 | 2012-07-31 16:45:25 -0700 | [diff] [blame] | 4331 | anon = PageAnon(used); |
Johannes Weiner | 7d18895 | 2012-07-31 16:45:34 -0700 | [diff] [blame] | 4332 | __mem_cgroup_uncharge_common(unused, |
| 4333 | anon ? MEM_CGROUP_CHARGE_TYPE_ANON |
| 4334 | : MEM_CGROUP_CHARGE_TYPE_CACHE, |
| 4335 | true); |
Johannes Weiner | 0030f53 | 2012-07-31 16:45:25 -0700 | [diff] [blame] | 4336 | css_put(&memcg->css); |
KAMEZAWA Hiroyuki | 69029cd | 2008-07-25 01:47:14 -0700 | [diff] [blame] | 4337 | /* |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 4338 | * We disallowed uncharge of pages under migration because mapcount |
| 4339 | * of the page goes down to zero, temporarly. |
| 4340 | * Clear the flag and check the page should be charged. |
KAMEZAWA Hiroyuki | 69029cd | 2008-07-25 01:47:14 -0700 | [diff] [blame] | 4341 | */ |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 4342 | pc = lookup_page_cgroup(oldpage); |
| 4343 | lock_page_cgroup(pc); |
| 4344 | ClearPageCgroupMigration(pc); |
| 4345 | unlock_page_cgroup(pc); |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 4346 | |
KAMEZAWA Hiroyuki | 01b1ae6 | 2009-01-07 18:07:50 -0800 | [diff] [blame] | 4347 | /* |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 4348 | * If a page is a file cache, radix-tree replacement is very atomic |
| 4349 | * and we can skip this check. When it was an Anon page, its mapcount |
| 4350 | * goes down to 0. But because we added MIGRATION flage, it's not |
| 4351 | * uncharged yet. There are several case but page->mapcount check |
| 4352 | * and USED bit check in mem_cgroup_uncharge_page() will do enough |
| 4353 | * check. (see prepare_charge() also) |
KAMEZAWA Hiroyuki | 01b1ae6 | 2009-01-07 18:07:50 -0800 | [diff] [blame] | 4354 | */ |
KAMEZAWA Hiroyuki | b2402857 | 2012-03-21 16:34:22 -0700 | [diff] [blame] | 4355 | if (anon) |
akpm@linux-foundation.org | ac39cf8 | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 4356 | mem_cgroup_uncharge_page(used); |
KAMEZAWA Hiroyuki | ae41be3 | 2008-02-07 00:14:10 -0800 | [diff] [blame] | 4357 | } |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 4358 | |
KAMEZAWA Hiroyuki | ab936cb | 2012-01-12 17:17:44 -0800 | [diff] [blame] | 4359 | /* |
| 4360 | * At replace page cache, newpage is not under any memcg but it's on |
| 4361 | * LRU. So, this function doesn't touch res_counter but handles LRU |
| 4362 | * in correct way. Both pages are locked so we cannot race with uncharge. |
| 4363 | */ |
| 4364 | void mem_cgroup_replace_page_cache(struct page *oldpage, |
| 4365 | struct page *newpage) |
| 4366 | { |
Hugh Dickins | bde05d1 | 2012-05-29 15:06:38 -0700 | [diff] [blame] | 4367 | struct mem_cgroup *memcg = NULL; |
KAMEZAWA Hiroyuki | ab936cb | 2012-01-12 17:17:44 -0800 | [diff] [blame] | 4368 | struct page_cgroup *pc; |
KAMEZAWA Hiroyuki | ab936cb | 2012-01-12 17:17:44 -0800 | [diff] [blame] | 4369 | enum charge_type type = MEM_CGROUP_CHARGE_TYPE_CACHE; |
KAMEZAWA Hiroyuki | ab936cb | 2012-01-12 17:17:44 -0800 | [diff] [blame] | 4370 | |
| 4371 | if (mem_cgroup_disabled()) |
| 4372 | return; |
| 4373 | |
| 4374 | pc = lookup_page_cgroup(oldpage); |
| 4375 | /* fix accounting on old pages */ |
| 4376 | lock_page_cgroup(pc); |
Hugh Dickins | bde05d1 | 2012-05-29 15:06:38 -0700 | [diff] [blame] | 4377 | if (PageCgroupUsed(pc)) { |
| 4378 | memcg = pc->mem_cgroup; |
David Rientjes | b070e65 | 2013-05-07 16:18:09 -0700 | [diff] [blame] | 4379 | mem_cgroup_charge_statistics(memcg, oldpage, false, -1); |
Hugh Dickins | bde05d1 | 2012-05-29 15:06:38 -0700 | [diff] [blame] | 4380 | ClearPageCgroupUsed(pc); |
| 4381 | } |
KAMEZAWA Hiroyuki | ab936cb | 2012-01-12 17:17:44 -0800 | [diff] [blame] | 4382 | unlock_page_cgroup(pc); |
| 4383 | |
Hugh Dickins | bde05d1 | 2012-05-29 15:06:38 -0700 | [diff] [blame] | 4384 | /* |
| 4385 | * When called from shmem_replace_page(), in some cases the |
| 4386 | * oldpage has already been charged, and in some cases not. |
| 4387 | */ |
| 4388 | if (!memcg) |
| 4389 | return; |
KAMEZAWA Hiroyuki | ab936cb | 2012-01-12 17:17:44 -0800 | [diff] [blame] | 4390 | /* |
| 4391 | * Even if newpage->mapping was NULL before starting replacement, |
| 4392 | * the newpage may be on LRU(or pagevec for LRU) already. We lock |
| 4393 | * LRU while we overwrite pc->mem_cgroup. |
| 4394 | */ |
Johannes Weiner | ce587e6 | 2012-04-24 20:22:33 +0200 | [diff] [blame] | 4395 | __mem_cgroup_commit_charge(memcg, newpage, 1, type, true); |
KAMEZAWA Hiroyuki | ab936cb | 2012-01-12 17:17:44 -0800 | [diff] [blame] | 4396 | } |
| 4397 | |
Daisuke Nishimura | f212ad7 | 2011-03-23 16:42:25 -0700 | [diff] [blame] | 4398 | #ifdef CONFIG_DEBUG_VM |
| 4399 | static struct page_cgroup *lookup_page_cgroup_used(struct page *page) |
| 4400 | { |
| 4401 | struct page_cgroup *pc; |
| 4402 | |
| 4403 | pc = lookup_page_cgroup(page); |
Johannes Weiner | cfa4494 | 2012-01-12 17:18:38 -0800 | [diff] [blame] | 4404 | /* |
| 4405 | * Can be NULL while feeding pages into the page allocator for |
| 4406 | * the first time, i.e. during boot or memory hotplug; |
| 4407 | * or when mem_cgroup_disabled(). |
| 4408 | */ |
Daisuke Nishimura | f212ad7 | 2011-03-23 16:42:25 -0700 | [diff] [blame] | 4409 | if (likely(pc) && PageCgroupUsed(pc)) |
| 4410 | return pc; |
| 4411 | return NULL; |
| 4412 | } |
| 4413 | |
| 4414 | bool mem_cgroup_bad_page_check(struct page *page) |
| 4415 | { |
| 4416 | if (mem_cgroup_disabled()) |
| 4417 | return false; |
| 4418 | |
| 4419 | return lookup_page_cgroup_used(page) != NULL; |
| 4420 | } |
| 4421 | |
| 4422 | void mem_cgroup_print_bad_page(struct page *page) |
| 4423 | { |
| 4424 | struct page_cgroup *pc; |
| 4425 | |
| 4426 | pc = lookup_page_cgroup_used(page); |
| 4427 | if (pc) { |
Andrew Morton | d045197 | 2013-02-22 16:32:06 -0800 | [diff] [blame] | 4428 | pr_alert("pc:%p pc->flags:%lx pc->mem_cgroup:%p\n", |
| 4429 | pc, pc->flags, pc->mem_cgroup); |
Daisuke Nishimura | f212ad7 | 2011-03-23 16:42:25 -0700 | [diff] [blame] | 4430 | } |
| 4431 | } |
| 4432 | #endif |
| 4433 | |
KOSAKI Motohiro | d38d2a7 | 2009-01-06 14:39:44 -0800 | [diff] [blame] | 4434 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4435 | unsigned long long val) |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 4436 | { |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 4437 | int retry_count; |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 4438 | u64 memswlimit, memlimit; |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 4439 | int ret = 0; |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 4440 | int children = mem_cgroup_count_children(memcg); |
| 4441 | u64 curusage, oldusage; |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 4442 | int enlarge; |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 4443 | |
| 4444 | /* |
| 4445 | * For keeping hierarchical_reclaim simple, how long we should retry |
| 4446 | * is depends on callers. We set our retry-count to be function |
| 4447 | * of # of children which we should visit in this loop. |
| 4448 | */ |
| 4449 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * children; |
| 4450 | |
| 4451 | oldusage = res_counter_read_u64(&memcg->res, RES_USAGE); |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 4452 | |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 4453 | enlarge = 0; |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4454 | while (retry_count) { |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 4455 | if (signal_pending(current)) { |
| 4456 | ret = -EINTR; |
| 4457 | break; |
| 4458 | } |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4459 | /* |
| 4460 | * Rather than hide all in some function, I do this in |
| 4461 | * open coded manner. You see what this really does. |
Wanpeng Li | aaad153 | 2012-07-31 16:43:23 -0700 | [diff] [blame] | 4462 | * We have to guarantee memcg->res.limit <= memcg->memsw.limit. |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4463 | */ |
| 4464 | mutex_lock(&set_limit_mutex); |
| 4465 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
| 4466 | if (memswlimit < val) { |
| 4467 | ret = -EINVAL; |
| 4468 | mutex_unlock(&set_limit_mutex); |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 4469 | break; |
| 4470 | } |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 4471 | |
| 4472 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
| 4473 | if (memlimit < val) |
| 4474 | enlarge = 1; |
| 4475 | |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4476 | ret = res_counter_set_limit(&memcg->res, val); |
KAMEZAWA Hiroyuki | 22a668d | 2009-06-17 16:27:19 -0700 | [diff] [blame] | 4477 | if (!ret) { |
| 4478 | if (memswlimit == val) |
| 4479 | memcg->memsw_is_minimum = true; |
| 4480 | else |
| 4481 | memcg->memsw_is_minimum = false; |
| 4482 | } |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4483 | mutex_unlock(&set_limit_mutex); |
| 4484 | |
| 4485 | if (!ret) |
| 4486 | break; |
| 4487 | |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 4488 | mem_cgroup_reclaim(memcg, GFP_KERNEL, |
| 4489 | MEM_CGROUP_RECLAIM_SHRINK); |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 4490 | curusage = res_counter_read_u64(&memcg->res, RES_USAGE); |
| 4491 | /* Usage is reduced ? */ |
Andrew Morton | f894ffa | 2013-09-12 15:13:35 -0700 | [diff] [blame] | 4492 | if (curusage >= oldusage) |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 4493 | retry_count--; |
| 4494 | else |
| 4495 | oldusage = curusage; |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4496 | } |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 4497 | if (!ret && enlarge) |
| 4498 | memcg_oom_recover(memcg); |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 4499 | |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4500 | return ret; |
| 4501 | } |
| 4502 | |
Li Zefan | 338c843 | 2009-06-17 16:27:15 -0700 | [diff] [blame] | 4503 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
| 4504 | unsigned long long val) |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4505 | { |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 4506 | int retry_count; |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 4507 | u64 memlimit, memswlimit, oldusage, curusage; |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 4508 | int children = mem_cgroup_count_children(memcg); |
| 4509 | int ret = -EBUSY; |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 4510 | int enlarge = 0; |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4511 | |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 4512 | /* see mem_cgroup_resize_res_limit */ |
Andrew Morton | f894ffa | 2013-09-12 15:13:35 -0700 | [diff] [blame] | 4513 | retry_count = children * MEM_CGROUP_RECLAIM_RETRIES; |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 4514 | oldusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4515 | while (retry_count) { |
| 4516 | if (signal_pending(current)) { |
| 4517 | ret = -EINTR; |
| 4518 | break; |
| 4519 | } |
| 4520 | /* |
| 4521 | * Rather than hide all in some function, I do this in |
| 4522 | * open coded manner. You see what this really does. |
Wanpeng Li | aaad153 | 2012-07-31 16:43:23 -0700 | [diff] [blame] | 4523 | * We have to guarantee memcg->res.limit <= memcg->memsw.limit. |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4524 | */ |
| 4525 | mutex_lock(&set_limit_mutex); |
| 4526 | memlimit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
| 4527 | if (memlimit > val) { |
| 4528 | ret = -EINVAL; |
| 4529 | mutex_unlock(&set_limit_mutex); |
| 4530 | break; |
| 4531 | } |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 4532 | memswlimit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
| 4533 | if (memswlimit < val) |
| 4534 | enlarge = 1; |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4535 | ret = res_counter_set_limit(&memcg->memsw, val); |
KAMEZAWA Hiroyuki | 22a668d | 2009-06-17 16:27:19 -0700 | [diff] [blame] | 4536 | if (!ret) { |
| 4537 | if (memlimit == val) |
| 4538 | memcg->memsw_is_minimum = true; |
| 4539 | else |
| 4540 | memcg->memsw_is_minimum = false; |
| 4541 | } |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4542 | mutex_unlock(&set_limit_mutex); |
| 4543 | |
| 4544 | if (!ret) |
| 4545 | break; |
| 4546 | |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 4547 | mem_cgroup_reclaim(memcg, GFP_KERNEL, |
| 4548 | MEM_CGROUP_RECLAIM_NOSWAP | |
| 4549 | MEM_CGROUP_RECLAIM_SHRINK); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4550 | curusage = res_counter_read_u64(&memcg->memsw, RES_USAGE); |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 4551 | /* Usage is reduced ? */ |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4552 | if (curusage >= oldusage) |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 4553 | retry_count--; |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 4554 | else |
| 4555 | oldusage = curusage; |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 4556 | } |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 4557 | if (!ret && enlarge) |
| 4558 | memcg_oom_recover(memcg); |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 4559 | return ret; |
| 4560 | } |
| 4561 | |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 4562 | unsigned long mem_cgroup_soft_limit_reclaim(struct zone *zone, int order, |
| 4563 | gfp_t gfp_mask, |
| 4564 | unsigned long *total_scanned) |
| 4565 | { |
| 4566 | unsigned long nr_reclaimed = 0; |
| 4567 | struct mem_cgroup_per_zone *mz, *next_mz = NULL; |
| 4568 | unsigned long reclaimed; |
| 4569 | int loop = 0; |
| 4570 | struct mem_cgroup_tree_per_zone *mctz; |
| 4571 | unsigned long long excess; |
| 4572 | unsigned long nr_scanned; |
| 4573 | |
| 4574 | if (order > 0) |
| 4575 | return 0; |
| 4576 | |
| 4577 | mctz = soft_limit_tree_node_zone(zone_to_nid(zone), zone_idx(zone)); |
| 4578 | /* |
| 4579 | * This loop can run a while, specially if mem_cgroup's continuously |
| 4580 | * keep exceeding their soft limit and putting the system under |
| 4581 | * pressure |
| 4582 | */ |
| 4583 | do { |
| 4584 | if (next_mz) |
| 4585 | mz = next_mz; |
| 4586 | else |
| 4587 | mz = mem_cgroup_largest_soft_limit_node(mctz); |
| 4588 | if (!mz) |
| 4589 | break; |
| 4590 | |
| 4591 | nr_scanned = 0; |
| 4592 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, zone, |
| 4593 | gfp_mask, &nr_scanned); |
| 4594 | nr_reclaimed += reclaimed; |
| 4595 | *total_scanned += nr_scanned; |
| 4596 | spin_lock(&mctz->lock); |
| 4597 | |
| 4598 | /* |
| 4599 | * If we failed to reclaim anything from this memory cgroup |
| 4600 | * it is time to move on to the next cgroup |
| 4601 | */ |
| 4602 | next_mz = NULL; |
| 4603 | if (!reclaimed) { |
| 4604 | do { |
| 4605 | /* |
| 4606 | * Loop until we find yet another one. |
| 4607 | * |
| 4608 | * By the time we get the soft_limit lock |
| 4609 | * again, someone might have aded the |
| 4610 | * group back on the RB tree. Iterate to |
| 4611 | * make sure we get a different mem. |
| 4612 | * mem_cgroup_largest_soft_limit_node returns |
| 4613 | * NULL if no other cgroup is present on |
| 4614 | * the tree |
| 4615 | */ |
| 4616 | next_mz = |
| 4617 | __mem_cgroup_largest_soft_limit_node(mctz); |
| 4618 | if (next_mz == mz) |
| 4619 | css_put(&next_mz->memcg->css); |
| 4620 | else /* next_mz == NULL or other memcg */ |
| 4621 | break; |
| 4622 | } while (1); |
| 4623 | } |
| 4624 | __mem_cgroup_remove_exceeded(mz->memcg, mz, mctz); |
| 4625 | excess = res_counter_soft_limit_excess(&mz->memcg->res); |
| 4626 | /* |
| 4627 | * One school of thought says that we should not add |
| 4628 | * back the node to the tree if reclaim returns 0. |
| 4629 | * But our reclaim could return 0, simply because due |
| 4630 | * to priority we are exposing a smaller subset of |
| 4631 | * memory to reclaim from. Consider this as a longer |
| 4632 | * term TODO. |
| 4633 | */ |
| 4634 | /* If excess == 0, no tree ops */ |
| 4635 | __mem_cgroup_insert_exceeded(mz->memcg, mz, mctz, excess); |
| 4636 | spin_unlock(&mctz->lock); |
| 4637 | css_put(&mz->memcg->css); |
| 4638 | loop++; |
| 4639 | /* |
| 4640 | * Could not reclaim anything and there are no more |
| 4641 | * mem cgroups to try or we seem to be looping without |
| 4642 | * reclaiming anything. |
| 4643 | */ |
| 4644 | if (!nr_reclaimed && |
| 4645 | (next_mz == NULL || |
| 4646 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) |
| 4647 | break; |
| 4648 | } while (!nr_reclaimed); |
| 4649 | if (next_mz) |
| 4650 | css_put(&next_mz->memcg->css); |
| 4651 | return nr_reclaimed; |
| 4652 | } |
| 4653 | |
Michal Hocko | 2ef37d3 | 2012-10-26 13:37:30 +0200 | [diff] [blame] | 4654 | /** |
| 4655 | * mem_cgroup_force_empty_list - clears LRU of a group |
| 4656 | * @memcg: group to clear |
| 4657 | * @node: NUMA node |
| 4658 | * @zid: zone id |
| 4659 | * @lru: lru to to clear |
| 4660 | * |
KAMEZAWA Hiroyuki | 3c935d1 | 2012-07-31 16:42:46 -0700 | [diff] [blame] | 4661 | * Traverse a specified page_cgroup list and try to drop them all. This doesn't |
Michal Hocko | 2ef37d3 | 2012-10-26 13:37:30 +0200 | [diff] [blame] | 4662 | * reclaim the pages page themselves - pages are moved to the parent (or root) |
| 4663 | * group. |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 4664 | */ |
Michal Hocko | 2ef37d3 | 2012-10-26 13:37:30 +0200 | [diff] [blame] | 4665 | static void mem_cgroup_force_empty_list(struct mem_cgroup *memcg, |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 4666 | int node, int zid, enum lru_list lru) |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 4667 | { |
Hugh Dickins | bea8c15 | 2012-11-16 14:14:54 -0800 | [diff] [blame] | 4668 | struct lruvec *lruvec; |
Michal Hocko | 2ef37d3 | 2012-10-26 13:37:30 +0200 | [diff] [blame] | 4669 | unsigned long flags; |
KAMEZAWA Hiroyuki | 072c56c1 | 2008-02-07 00:14:39 -0800 | [diff] [blame] | 4670 | struct list_head *list; |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 4671 | struct page *busy; |
| 4672 | struct zone *zone; |
KAMEZAWA Hiroyuki | 072c56c1 | 2008-02-07 00:14:39 -0800 | [diff] [blame] | 4673 | |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 4674 | zone = &NODE_DATA(node)->node_zones[zid]; |
Hugh Dickins | bea8c15 | 2012-11-16 14:14:54 -0800 | [diff] [blame] | 4675 | lruvec = mem_cgroup_zone_lruvec(zone, memcg); |
| 4676 | list = &lruvec->lists[lru]; |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 4677 | |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 4678 | busy = NULL; |
Michal Hocko | 2ef37d3 | 2012-10-26 13:37:30 +0200 | [diff] [blame] | 4679 | do { |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 4680 | struct page_cgroup *pc; |
Johannes Weiner | 5564e88 | 2011-03-23 16:42:29 -0700 | [diff] [blame] | 4681 | struct page *page; |
| 4682 | |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 4683 | spin_lock_irqsave(&zone->lru_lock, flags); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 4684 | if (list_empty(list)) { |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 4685 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 4686 | break; |
| 4687 | } |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 4688 | page = list_entry(list->prev, struct page, lru); |
| 4689 | if (busy == page) { |
| 4690 | list_move(&page->lru, list); |
Thiago Farina | 648bcc7 | 2010-03-05 13:42:04 -0800 | [diff] [blame] | 4691 | busy = NULL; |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 4692 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 4693 | continue; |
| 4694 | } |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 4695 | spin_unlock_irqrestore(&zone->lru_lock, flags); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 4696 | |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 4697 | pc = lookup_page_cgroup(page); |
Johannes Weiner | 5564e88 | 2011-03-23 16:42:29 -0700 | [diff] [blame] | 4698 | |
KAMEZAWA Hiroyuki | 3c935d1 | 2012-07-31 16:42:46 -0700 | [diff] [blame] | 4699 | if (mem_cgroup_move_parent(page, pc, memcg)) { |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 4700 | /* found lock contention or "pc" is obsolete. */ |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 4701 | busy = page; |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 4702 | cond_resched(); |
| 4703 | } else |
| 4704 | busy = NULL; |
Michal Hocko | 2ef37d3 | 2012-10-26 13:37:30 +0200 | [diff] [blame] | 4705 | } while (!list_empty(list)); |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 4706 | } |
| 4707 | |
| 4708 | /* |
Michal Hocko | c26251f | 2012-10-26 13:37:28 +0200 | [diff] [blame] | 4709 | * make mem_cgroup's charge to be 0 if there is no task by moving |
| 4710 | * all the charges and pages to the parent. |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 4711 | * This enables deleting this mem_cgroup. |
Michal Hocko | c26251f | 2012-10-26 13:37:28 +0200 | [diff] [blame] | 4712 | * |
| 4713 | * Caller is responsible for holding css reference on the memcg. |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 4714 | */ |
Michal Hocko | ab5196c | 2012-10-26 13:37:32 +0200 | [diff] [blame] | 4715 | static void mem_cgroup_reparent_charges(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 4716 | { |
Michal Hocko | c26251f | 2012-10-26 13:37:28 +0200 | [diff] [blame] | 4717 | int node, zid; |
Glauber Costa | bea207c | 2012-12-18 14:22:11 -0800 | [diff] [blame] | 4718 | u64 usage; |
Hugh Dickins | 8869b8f | 2008-03-04 14:29:09 -0800 | [diff] [blame] | 4719 | |
Daisuke Nishimura | fce6647 | 2010-01-15 17:01:30 -0800 | [diff] [blame] | 4720 | do { |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 4721 | /* This is for making all *used* pages to be on LRU. */ |
| 4722 | lru_add_drain_all(); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4723 | drain_all_stock_sync(memcg); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4724 | mem_cgroup_start_move(memcg); |
Lai Jiangshan | 31aaea4 | 2012-12-12 13:51:27 -0800 | [diff] [blame] | 4725 | for_each_node_state(node, N_MEMORY) { |
Michal Hocko | 2ef37d3 | 2012-10-26 13:37:30 +0200 | [diff] [blame] | 4726 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
Hugh Dickins | f156ab9 | 2012-03-21 16:34:19 -0700 | [diff] [blame] | 4727 | enum lru_list lru; |
| 4728 | for_each_lru(lru) { |
Michal Hocko | 2ef37d3 | 2012-10-26 13:37:30 +0200 | [diff] [blame] | 4729 | mem_cgroup_force_empty_list(memcg, |
Hugh Dickins | f156ab9 | 2012-03-21 16:34:19 -0700 | [diff] [blame] | 4730 | node, zid, lru); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 4731 | } |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 4732 | } |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 4733 | } |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4734 | mem_cgroup_end_move(memcg); |
| 4735 | memcg_oom_recover(memcg); |
KAMEZAWA Hiroyuki | 52d4b9a | 2008-10-18 20:28:16 -0700 | [diff] [blame] | 4736 | cond_resched(); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 4737 | |
Michal Hocko | 2ef37d3 | 2012-10-26 13:37:30 +0200 | [diff] [blame] | 4738 | /* |
Glauber Costa | bea207c | 2012-12-18 14:22:11 -0800 | [diff] [blame] | 4739 | * Kernel memory may not necessarily be trackable to a specific |
| 4740 | * process. So they are not migrated, and therefore we can't |
| 4741 | * expect their value to drop to 0 here. |
| 4742 | * Having res filled up with kmem only is enough. |
| 4743 | * |
Michal Hocko | 2ef37d3 | 2012-10-26 13:37:30 +0200 | [diff] [blame] | 4744 | * This is a safety check because mem_cgroup_force_empty_list |
| 4745 | * could have raced with mem_cgroup_replace_page_cache callers |
| 4746 | * so the lru seemed empty but the page could have been added |
| 4747 | * right after the check. RES_USAGE should be safe as we always |
| 4748 | * charge before adding to the LRU. |
| 4749 | */ |
Glauber Costa | bea207c | 2012-12-18 14:22:11 -0800 | [diff] [blame] | 4750 | usage = res_counter_read_u64(&memcg->res, RES_USAGE) - |
| 4751 | res_counter_read_u64(&memcg->kmem, RES_USAGE); |
| 4752 | } while (usage > 0); |
Michal Hocko | c26251f | 2012-10-26 13:37:28 +0200 | [diff] [blame] | 4753 | } |
| 4754 | |
Glauber Costa | b5f99b5 | 2013-02-22 16:34:53 -0800 | [diff] [blame] | 4755 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
| 4756 | { |
Johannes Weiner | 696ac17 | 2013-10-31 16:34:15 -0700 | [diff] [blame] | 4757 | lockdep_assert_held(&memcg_create_mutex); |
| 4758 | /* |
| 4759 | * The lock does not prevent addition or deletion to the list |
| 4760 | * of children, but it prevents a new child from being |
| 4761 | * initialized based on this parent in css_online(), so it's |
| 4762 | * enough to decide whether hierarchically inherited |
| 4763 | * attributes can still be changed or not. |
| 4764 | */ |
| 4765 | return memcg->use_hierarchy && |
| 4766 | !list_empty(&memcg->css.cgroup->children); |
Glauber Costa | b5f99b5 | 2013-02-22 16:34:53 -0800 | [diff] [blame] | 4767 | } |
| 4768 | |
| 4769 | /* |
Michal Hocko | c26251f | 2012-10-26 13:37:28 +0200 | [diff] [blame] | 4770 | * Reclaims as many pages from the given memcg as possible and moves |
| 4771 | * the rest to the parent. |
| 4772 | * |
| 4773 | * Caller is responsible for holding css reference for memcg. |
| 4774 | */ |
| 4775 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) |
| 4776 | { |
| 4777 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
| 4778 | struct cgroup *cgrp = memcg->css.cgroup; |
| 4779 | |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 4780 | /* returns EBUSY if there is a task or if we come here twice. */ |
Tejun Heo | 07bc356 | 2014-02-13 06:58:39 -0500 | [diff] [blame] | 4781 | if (cgroup_has_tasks(cgrp) || !list_empty(&cgrp->children)) |
Michal Hocko | c26251f | 2012-10-26 13:37:28 +0200 | [diff] [blame] | 4782 | return -EBUSY; |
| 4783 | |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 4784 | /* we call try-to-free pages for make this cgroup empty */ |
| 4785 | lru_add_drain_all(); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 4786 | /* try to free all pages in this cgroup */ |
Glauber Costa | 569530f | 2012-04-12 12:49:13 -0700 | [diff] [blame] | 4787 | while (nr_retries && res_counter_read_u64(&memcg->res, RES_USAGE) > 0) { |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 4788 | int progress; |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 4789 | |
Michal Hocko | c26251f | 2012-10-26 13:37:28 +0200 | [diff] [blame] | 4790 | if (signal_pending(current)) |
| 4791 | return -EINTR; |
| 4792 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4793 | progress = try_to_free_mem_cgroup_pages(memcg, GFP_KERNEL, |
Johannes Weiner | 185efc0 | 2011-09-14 16:21:58 -0700 | [diff] [blame] | 4794 | false); |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 4795 | if (!progress) { |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 4796 | nr_retries--; |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 4797 | /* maybe some writeback is necessary */ |
Jens Axboe | 8aa7e84 | 2009-07-09 14:52:32 +0200 | [diff] [blame] | 4798 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 4799 | } |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 4800 | |
| 4801 | } |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 4802 | lru_add_drain(); |
Michal Hocko | ab5196c | 2012-10-26 13:37:32 +0200 | [diff] [blame] | 4803 | mem_cgroup_reparent_charges(memcg); |
| 4804 | |
| 4805 | return 0; |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 4806 | } |
| 4807 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 4808 | static int mem_cgroup_force_empty_write(struct cgroup_subsys_state *css, |
| 4809 | unsigned int event) |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 4810 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 4811 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Michal Hocko | c26251f | 2012-10-26 13:37:28 +0200 | [diff] [blame] | 4812 | |
Michal Hocko | d842301 | 2012-10-26 13:37:29 +0200 | [diff] [blame] | 4813 | if (mem_cgroup_is_root(memcg)) |
| 4814 | return -EINVAL; |
Li Zefan | c33bd83 | 2013-09-12 15:13:19 -0700 | [diff] [blame] | 4815 | return mem_cgroup_force_empty(memcg); |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 4816 | } |
| 4817 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 4818 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
| 4819 | struct cftype *cft) |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 4820 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 4821 | return mem_cgroup_from_css(css)->use_hierarchy; |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 4822 | } |
| 4823 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 4824 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
| 4825 | struct cftype *cft, u64 val) |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 4826 | { |
| 4827 | int retval = 0; |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 4828 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Tejun Heo | 6387698 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 4829 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(css_parent(&memcg->css)); |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 4830 | |
Glauber Costa | 0999821 | 2013-02-22 16:34:55 -0800 | [diff] [blame] | 4831 | mutex_lock(&memcg_create_mutex); |
Glauber Costa | 567fb43 | 2012-07-31 16:43:07 -0700 | [diff] [blame] | 4832 | |
| 4833 | if (memcg->use_hierarchy == val) |
| 4834 | goto out; |
| 4835 | |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 4836 | /* |
André Goddard Rosa | af901ca | 2009-11-14 13:09:05 -0200 | [diff] [blame] | 4837 | * 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] | 4838 | * in the child subtrees. If it is unset, then the change can |
| 4839 | * occur, provided the current cgroup has no children. |
| 4840 | * |
| 4841 | * For the root cgroup, parent_mem is NULL, we allow value to be |
| 4842 | * set if there are no children. |
| 4843 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4844 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 4845 | (val == 1 || val == 0)) { |
Johannes Weiner | 696ac17 | 2013-10-31 16:34:15 -0700 | [diff] [blame] | 4846 | if (list_empty(&memcg->css.cgroup->children)) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4847 | memcg->use_hierarchy = val; |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 4848 | else |
| 4849 | retval = -EBUSY; |
| 4850 | } else |
| 4851 | retval = -EINVAL; |
Glauber Costa | 567fb43 | 2012-07-31 16:43:07 -0700 | [diff] [blame] | 4852 | |
| 4853 | out: |
Glauber Costa | 0999821 | 2013-02-22 16:34:55 -0800 | [diff] [blame] | 4854 | mutex_unlock(&memcg_create_mutex); |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 4855 | |
| 4856 | return retval; |
| 4857 | } |
| 4858 | |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 4859 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4860 | static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *memcg, |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 4861 | enum mem_cgroup_stat_index idx) |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 4862 | { |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 4863 | struct mem_cgroup *iter; |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 4864 | long val = 0; |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 4865 | |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 4866 | /* Per-cpu values can be negative, use a signed accumulator */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4867 | for_each_mem_cgroup_tree(iter, memcg) |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 4868 | val += mem_cgroup_read_stat(iter, idx); |
| 4869 | |
| 4870 | if (val < 0) /* race ? */ |
| 4871 | val = 0; |
| 4872 | return val; |
Balbir Singh | 0c3e73e | 2009-09-23 15:56:42 -0700 | [diff] [blame] | 4873 | } |
| 4874 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4875 | static inline u64 mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
Kirill A. Shutemov | 104f392 | 2010-03-10 15:22:21 -0800 | [diff] [blame] | 4876 | { |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 4877 | u64 val; |
Kirill A. Shutemov | 104f392 | 2010-03-10 15:22:21 -0800 | [diff] [blame] | 4878 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4879 | if (!mem_cgroup_is_root(memcg)) { |
Kirill A. Shutemov | 104f392 | 2010-03-10 15:22:21 -0800 | [diff] [blame] | 4880 | if (!swap) |
Glauber Costa | 65c64ce | 2011-12-22 01:02:27 +0000 | [diff] [blame] | 4881 | return res_counter_read_u64(&memcg->res, RES_USAGE); |
Kirill A. Shutemov | 104f392 | 2010-03-10 15:22:21 -0800 | [diff] [blame] | 4882 | else |
Glauber Costa | 65c64ce | 2011-12-22 01:02:27 +0000 | [diff] [blame] | 4883 | return res_counter_read_u64(&memcg->memsw, RES_USAGE); |
Kirill A. Shutemov | 104f392 | 2010-03-10 15:22:21 -0800 | [diff] [blame] | 4884 | } |
| 4885 | |
David Rientjes | b070e65 | 2013-05-07 16:18:09 -0700 | [diff] [blame] | 4886 | /* |
| 4887 | * Transparent hugepages are still accounted for in MEM_CGROUP_STAT_RSS |
| 4888 | * as well as in MEM_CGROUP_STAT_RSS_HUGE. |
| 4889 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4890 | val = mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_CACHE); |
| 4891 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_RSS); |
Kirill A. Shutemov | 104f392 | 2010-03-10 15:22:21 -0800 | [diff] [blame] | 4892 | |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 4893 | if (swap) |
Kamezawa Hiroyuki | bff6bb8 | 2012-07-31 16:41:38 -0700 | [diff] [blame] | 4894 | val += mem_cgroup_recursive_stat(memcg, MEM_CGROUP_STAT_SWAP); |
Kirill A. Shutemov | 104f392 | 2010-03-10 15:22:21 -0800 | [diff] [blame] | 4895 | |
| 4896 | return val << PAGE_SHIFT; |
| 4897 | } |
| 4898 | |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 4899 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
| 4900 | struct cftype *cft) |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4901 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 4902 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Kirill A. Shutemov | 104f392 | 2010-03-10 15:22:21 -0800 | [diff] [blame] | 4903 | u64 val; |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 4904 | int name; |
Glauber Costa | 86ae53e | 2012-12-18 14:21:45 -0800 | [diff] [blame] | 4905 | enum res_type type; |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4906 | |
| 4907 | type = MEMFILE_TYPE(cft->private); |
| 4908 | name = MEMFILE_ATTR(cft->private); |
Tejun Heo | af36f90 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 4909 | |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4910 | switch (type) { |
| 4911 | case _MEM: |
Kirill A. Shutemov | 104f392 | 2010-03-10 15:22:21 -0800 | [diff] [blame] | 4912 | if (name == RES_USAGE) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4913 | val = mem_cgroup_usage(memcg, false); |
Kirill A. Shutemov | 104f392 | 2010-03-10 15:22:21 -0800 | [diff] [blame] | 4914 | else |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4915 | val = res_counter_read_u64(&memcg->res, name); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4916 | break; |
| 4917 | case _MEMSWAP: |
Kirill A. Shutemov | 104f392 | 2010-03-10 15:22:21 -0800 | [diff] [blame] | 4918 | if (name == RES_USAGE) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4919 | val = mem_cgroup_usage(memcg, true); |
Kirill A. Shutemov | 104f392 | 2010-03-10 15:22:21 -0800 | [diff] [blame] | 4920 | else |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4921 | val = res_counter_read_u64(&memcg->memsw, name); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4922 | break; |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 4923 | case _KMEM: |
| 4924 | val = res_counter_read_u64(&memcg->kmem, name); |
| 4925 | break; |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4926 | default: |
| 4927 | BUG(); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4928 | } |
Tejun Heo | af36f90 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 4929 | |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 4930 | return val; |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4931 | } |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 4932 | |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 4933 | #ifdef CONFIG_MEMCG_KMEM |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 4934 | /* should be called with activate_kmem_mutex held */ |
| 4935 | static int __memcg_activate_kmem(struct mem_cgroup *memcg, |
| 4936 | unsigned long long limit) |
| 4937 | { |
| 4938 | int err = 0; |
| 4939 | int memcg_id; |
| 4940 | |
| 4941 | if (memcg_kmem_is_active(memcg)) |
| 4942 | return 0; |
| 4943 | |
| 4944 | /* |
| 4945 | * We are going to allocate memory for data shared by all memory |
| 4946 | * cgroups so let's stop accounting here. |
| 4947 | */ |
| 4948 | memcg_stop_kmem_account(); |
| 4949 | |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 4950 | /* |
| 4951 | * For simplicity, we won't allow this to be disabled. It also can't |
| 4952 | * be changed if the cgroup has children already, or if tasks had |
| 4953 | * already joined. |
| 4954 | * |
| 4955 | * If tasks join before we set the limit, a person looking at |
| 4956 | * kmem.usage_in_bytes will have no way to determine when it took |
| 4957 | * place, which makes the value quite meaningless. |
| 4958 | * |
| 4959 | * After it first became limited, changes in the value of the limit are |
| 4960 | * of course permitted. |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 4961 | */ |
Glauber Costa | 0999821 | 2013-02-22 16:34:55 -0800 | [diff] [blame] | 4962 | mutex_lock(&memcg_create_mutex); |
Tejun Heo | 07bc356 | 2014-02-13 06:58:39 -0500 | [diff] [blame] | 4963 | if (cgroup_has_tasks(memcg->css.cgroup) || memcg_has_children(memcg)) |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 4964 | err = -EBUSY; |
Glauber Costa | 0999821 | 2013-02-22 16:34:55 -0800 | [diff] [blame] | 4965 | mutex_unlock(&memcg_create_mutex); |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 4966 | if (err) |
| 4967 | goto out; |
| 4968 | |
| 4969 | memcg_id = ida_simple_get(&kmem_limited_groups, |
| 4970 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); |
| 4971 | if (memcg_id < 0) { |
| 4972 | err = memcg_id; |
| 4973 | goto out; |
| 4974 | } |
| 4975 | |
| 4976 | /* |
| 4977 | * Make sure we have enough space for this cgroup in each root cache's |
| 4978 | * memcg_params. |
| 4979 | */ |
Vladimir Davydov | bd67314 | 2014-06-04 16:07:40 -0700 | [diff] [blame] | 4980 | mutex_lock(&memcg_slab_mutex); |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 4981 | err = memcg_update_all_caches(memcg_id + 1); |
Vladimir Davydov | bd67314 | 2014-06-04 16:07:40 -0700 | [diff] [blame] | 4982 | mutex_unlock(&memcg_slab_mutex); |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 4983 | if (err) |
| 4984 | goto out_rmid; |
| 4985 | |
| 4986 | memcg->kmemcg_id = memcg_id; |
| 4987 | INIT_LIST_HEAD(&memcg->memcg_slab_caches); |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 4988 | |
| 4989 | /* |
| 4990 | * We couldn't have accounted to this cgroup, because it hasn't got the |
| 4991 | * active bit set yet, so this should succeed. |
| 4992 | */ |
| 4993 | err = res_counter_set_limit(&memcg->kmem, limit); |
| 4994 | VM_BUG_ON(err); |
| 4995 | |
| 4996 | static_key_slow_inc(&memcg_kmem_enabled_key); |
| 4997 | /* |
| 4998 | * Setting the active bit after enabling static branching will |
| 4999 | * guarantee no one starts accounting before all call sites are |
| 5000 | * patched. |
| 5001 | */ |
| 5002 | memcg_kmem_set_active(memcg); |
| 5003 | out: |
| 5004 | memcg_resume_kmem_account(); |
| 5005 | return err; |
| 5006 | |
| 5007 | out_rmid: |
| 5008 | ida_simple_remove(&kmem_limited_groups, memcg_id); |
| 5009 | goto out; |
| 5010 | } |
| 5011 | |
| 5012 | static int memcg_activate_kmem(struct mem_cgroup *memcg, |
| 5013 | unsigned long long limit) |
| 5014 | { |
| 5015 | int ret; |
| 5016 | |
| 5017 | mutex_lock(&activate_kmem_mutex); |
| 5018 | ret = __memcg_activate_kmem(memcg, limit); |
| 5019 | mutex_unlock(&activate_kmem_mutex); |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 5020 | return ret; |
| 5021 | } |
| 5022 | |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 5023 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, |
| 5024 | unsigned long long val) |
| 5025 | { |
| 5026 | int ret; |
| 5027 | |
| 5028 | if (!memcg_kmem_is_active(memcg)) |
| 5029 | ret = memcg_activate_kmem(memcg, val); |
| 5030 | else |
| 5031 | ret = res_counter_set_limit(&memcg->kmem, val); |
| 5032 | return ret; |
| 5033 | } |
| 5034 | |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 5035 | static int memcg_propagate_kmem(struct mem_cgroup *memcg) |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 5036 | { |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 5037 | int ret = 0; |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 5038 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 5039 | |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 5040 | if (!parent) |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 5041 | return 0; |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 5042 | |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 5043 | mutex_lock(&activate_kmem_mutex); |
Glauber Costa | a8964b9 | 2012-12-18 14:22:09 -0800 | [diff] [blame] | 5044 | /* |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 5045 | * If the parent cgroup is not kmem-active now, it cannot be activated |
| 5046 | * after this point, because it has at least one child already. |
Glauber Costa | a8964b9 | 2012-12-18 14:22:09 -0800 | [diff] [blame] | 5047 | */ |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 5048 | if (memcg_kmem_is_active(parent)) |
| 5049 | ret = __memcg_activate_kmem(memcg, RES_COUNTER_MAX); |
| 5050 | mutex_unlock(&activate_kmem_mutex); |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 5051 | return ret; |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 5052 | } |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 5053 | #else |
| 5054 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, |
| 5055 | unsigned long long val) |
| 5056 | { |
| 5057 | return -EINVAL; |
| 5058 | } |
Hugh Dickins | 6d043990 | 2013-02-22 16:35:50 -0800 | [diff] [blame] | 5059 | #endif /* CONFIG_MEMCG_KMEM */ |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 5060 | |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 5061 | /* |
| 5062 | * The user of this function is... |
| 5063 | * RES_LIMIT. |
| 5064 | */ |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 5065 | static int mem_cgroup_write(struct cgroup_subsys_state *css, struct cftype *cft, |
Tejun Heo | 4d3bb51 | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 5066 | char *buffer) |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 5067 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 5068 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Glauber Costa | 86ae53e | 2012-12-18 14:21:45 -0800 | [diff] [blame] | 5069 | enum res_type type; |
| 5070 | int name; |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 5071 | unsigned long long val; |
| 5072 | int ret; |
| 5073 | |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 5074 | type = MEMFILE_TYPE(cft->private); |
| 5075 | name = MEMFILE_ATTR(cft->private); |
Tejun Heo | af36f90 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 5076 | |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 5077 | switch (name) { |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 5078 | case RES_LIMIT: |
Balbir Singh | 4b3bde4 | 2009-09-23 15:56:32 -0700 | [diff] [blame] | 5079 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
| 5080 | ret = -EINVAL; |
| 5081 | break; |
| 5082 | } |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 5083 | /* This function does all necessary parse...reuse it */ |
| 5084 | ret = res_counter_memparse_write_strategy(buffer, &val); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 5085 | if (ret) |
| 5086 | break; |
| 5087 | if (type == _MEM) |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 5088 | ret = mem_cgroup_resize_limit(memcg, val); |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 5089 | else if (type == _MEMSWAP) |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 5090 | ret = mem_cgroup_resize_memsw_limit(memcg, val); |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 5091 | else if (type == _KMEM) |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 5092 | ret = memcg_update_kmem_limit(memcg, val); |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 5093 | else |
| 5094 | return -EINVAL; |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 5095 | break; |
Balbir Singh | 296c81d | 2009-09-23 15:56:36 -0700 | [diff] [blame] | 5096 | case RES_SOFT_LIMIT: |
| 5097 | ret = res_counter_memparse_write_strategy(buffer, &val); |
| 5098 | if (ret) |
| 5099 | break; |
| 5100 | /* |
| 5101 | * For memsw, soft limits are hard to implement in terms |
| 5102 | * of semantics, for now, we support soft limits for |
| 5103 | * control without swap |
| 5104 | */ |
| 5105 | if (type == _MEM) |
| 5106 | ret = res_counter_set_soft_limit(&memcg->res, val); |
| 5107 | else |
| 5108 | ret = -EINVAL; |
| 5109 | break; |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 5110 | default: |
| 5111 | ret = -EINVAL; /* should be BUG() ? */ |
| 5112 | break; |
| 5113 | } |
| 5114 | return ret; |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 5115 | } |
| 5116 | |
KAMEZAWA Hiroyuki | fee7b54 | 2009-01-07 18:08:26 -0800 | [diff] [blame] | 5117 | static void memcg_get_hierarchical_limit(struct mem_cgroup *memcg, |
| 5118 | unsigned long long *mem_limit, unsigned long long *memsw_limit) |
| 5119 | { |
KAMEZAWA Hiroyuki | fee7b54 | 2009-01-07 18:08:26 -0800 | [diff] [blame] | 5120 | unsigned long long min_limit, min_memsw_limit, tmp; |
| 5121 | |
| 5122 | min_limit = res_counter_read_u64(&memcg->res, RES_LIMIT); |
| 5123 | min_memsw_limit = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
KAMEZAWA Hiroyuki | fee7b54 | 2009-01-07 18:08:26 -0800 | [diff] [blame] | 5124 | if (!memcg->use_hierarchy) |
| 5125 | goto out; |
| 5126 | |
Tejun Heo | 6387698 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 5127 | while (css_parent(&memcg->css)) { |
| 5128 | memcg = mem_cgroup_from_css(css_parent(&memcg->css)); |
KAMEZAWA Hiroyuki | fee7b54 | 2009-01-07 18:08:26 -0800 | [diff] [blame] | 5129 | if (!memcg->use_hierarchy) |
| 5130 | break; |
| 5131 | tmp = res_counter_read_u64(&memcg->res, RES_LIMIT); |
| 5132 | min_limit = min(min_limit, tmp); |
| 5133 | tmp = res_counter_read_u64(&memcg->memsw, RES_LIMIT); |
| 5134 | min_memsw_limit = min(min_memsw_limit, tmp); |
| 5135 | } |
| 5136 | out: |
| 5137 | *mem_limit = min_limit; |
| 5138 | *memsw_limit = min_memsw_limit; |
KAMEZAWA Hiroyuki | fee7b54 | 2009-01-07 18:08:26 -0800 | [diff] [blame] | 5139 | } |
| 5140 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 5141 | static int mem_cgroup_reset(struct cgroup_subsys_state *css, unsigned int event) |
Pavel Emelyanov | c84872e | 2008-04-29 01:00:17 -0700 | [diff] [blame] | 5142 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 5143 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Glauber Costa | 86ae53e | 2012-12-18 14:21:45 -0800 | [diff] [blame] | 5144 | int name; |
| 5145 | enum res_type type; |
Pavel Emelyanov | c84872e | 2008-04-29 01:00:17 -0700 | [diff] [blame] | 5146 | |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 5147 | type = MEMFILE_TYPE(event); |
| 5148 | name = MEMFILE_ATTR(event); |
Tejun Heo | af36f90 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 5149 | |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 5150 | switch (name) { |
Pavel Emelyanov | 29f2a4d | 2008-04-29 01:00:21 -0700 | [diff] [blame] | 5151 | case RES_MAX_USAGE: |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 5152 | if (type == _MEM) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 5153 | res_counter_reset_max(&memcg->res); |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 5154 | else if (type == _MEMSWAP) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 5155 | res_counter_reset_max(&memcg->memsw); |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 5156 | else if (type == _KMEM) |
| 5157 | res_counter_reset_max(&memcg->kmem); |
| 5158 | else |
| 5159 | return -EINVAL; |
Pavel Emelyanov | 29f2a4d | 2008-04-29 01:00:21 -0700 | [diff] [blame] | 5160 | break; |
| 5161 | case RES_FAILCNT: |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 5162 | if (type == _MEM) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 5163 | res_counter_reset_failcnt(&memcg->res); |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 5164 | else if (type == _MEMSWAP) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 5165 | res_counter_reset_failcnt(&memcg->memsw); |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 5166 | else if (type == _KMEM) |
| 5167 | res_counter_reset_failcnt(&memcg->kmem); |
| 5168 | else |
| 5169 | return -EINVAL; |
Pavel Emelyanov | 29f2a4d | 2008-04-29 01:00:21 -0700 | [diff] [blame] | 5170 | break; |
| 5171 | } |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 5172 | |
Pavel Emelyanov | 85cc59d | 2008-04-29 01:00:20 -0700 | [diff] [blame] | 5173 | return 0; |
Pavel Emelyanov | c84872e | 2008-04-29 01:00:17 -0700 | [diff] [blame] | 5174 | } |
| 5175 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 5176 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5177 | struct cftype *cft) |
| 5178 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 5179 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5180 | } |
| 5181 | |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 5182 | #ifdef CONFIG_MMU |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 5183 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5184 | struct cftype *cft, u64 val) |
| 5185 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 5186 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5187 | |
| 5188 | if (val >= (1 << NR_MOVE_TYPE)) |
| 5189 | return -EINVAL; |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5190 | |
Glauber Costa | ee5e847 | 2013-02-22 16:34:50 -0800 | [diff] [blame] | 5191 | /* |
| 5192 | * No kind of locking is needed in here, because ->can_attach() will |
| 5193 | * check this value once in the beginning of the process, and then carry |
| 5194 | * on with stale data. This means that changes to this value will only |
| 5195 | * affect task migrations starting after the change. |
| 5196 | */ |
| 5197 | memcg->move_charge_at_immigrate = val; |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5198 | return 0; |
| 5199 | } |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 5200 | #else |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 5201 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 5202 | struct cftype *cft, u64 val) |
| 5203 | { |
| 5204 | return -ENOSYS; |
| 5205 | } |
| 5206 | #endif |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5207 | |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 5208 | #ifdef CONFIG_NUMA |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 5209 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 5210 | { |
Greg Thelen | 25485de | 2013-11-12 15:07:40 -0800 | [diff] [blame] | 5211 | struct numa_stat { |
| 5212 | const char *name; |
| 5213 | unsigned int lru_mask; |
| 5214 | }; |
| 5215 | |
| 5216 | static const struct numa_stat stats[] = { |
| 5217 | { "total", LRU_ALL }, |
| 5218 | { "file", LRU_ALL_FILE }, |
| 5219 | { "anon", LRU_ALL_ANON }, |
| 5220 | { "unevictable", BIT(LRU_UNEVICTABLE) }, |
| 5221 | }; |
| 5222 | const struct numa_stat *stat; |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 5223 | int nid; |
Greg Thelen | 25485de | 2013-11-12 15:07:40 -0800 | [diff] [blame] | 5224 | unsigned long nr; |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 5225 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 5226 | |
Greg Thelen | 25485de | 2013-11-12 15:07:40 -0800 | [diff] [blame] | 5227 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
| 5228 | nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask); |
| 5229 | seq_printf(m, "%s=%lu", stat->name, nr); |
| 5230 | for_each_node_state(nid, N_MEMORY) { |
| 5231 | nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
| 5232 | stat->lru_mask); |
| 5233 | seq_printf(m, " N%d=%lu", nid, nr); |
| 5234 | } |
| 5235 | seq_putc(m, '\n'); |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 5236 | } |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 5237 | |
Ying Han | 071aee1 | 2013-11-12 15:07:41 -0800 | [diff] [blame] | 5238 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
| 5239 | struct mem_cgroup *iter; |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 5240 | |
Ying Han | 071aee1 | 2013-11-12 15:07:41 -0800 | [diff] [blame] | 5241 | nr = 0; |
| 5242 | for_each_mem_cgroup_tree(iter, memcg) |
| 5243 | nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask); |
| 5244 | seq_printf(m, "hierarchical_%s=%lu", stat->name, nr); |
| 5245 | for_each_node_state(nid, N_MEMORY) { |
| 5246 | nr = 0; |
| 5247 | for_each_mem_cgroup_tree(iter, memcg) |
| 5248 | nr += mem_cgroup_node_nr_lru_pages( |
| 5249 | iter, nid, stat->lru_mask); |
| 5250 | seq_printf(m, " N%d=%lu", nid, nr); |
| 5251 | } |
| 5252 | seq_putc(m, '\n'); |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 5253 | } |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 5254 | |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 5255 | return 0; |
| 5256 | } |
| 5257 | #endif /* CONFIG_NUMA */ |
| 5258 | |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 5259 | static inline void mem_cgroup_lru_names_not_uptodate(void) |
| 5260 | { |
| 5261 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); |
| 5262 | } |
| 5263 | |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 5264 | static int memcg_stat_show(struct seq_file *m, void *v) |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 5265 | { |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 5266 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 5267 | struct mem_cgroup *mi; |
| 5268 | unsigned int i; |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 5269 | |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 5270 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
Kamezawa Hiroyuki | bff6bb8 | 2012-07-31 16:41:38 -0700 | [diff] [blame] | 5271 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
Daisuke Nishimura | 1dd3a27 | 2009-09-23 15:56:43 -0700 | [diff] [blame] | 5272 | continue; |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 5273 | seq_printf(m, "%s %ld\n", mem_cgroup_stat_names[i], |
| 5274 | mem_cgroup_read_stat(memcg, i) * PAGE_SIZE); |
Daisuke Nishimura | 1dd3a27 | 2009-09-23 15:56:43 -0700 | [diff] [blame] | 5275 | } |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 5276 | |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 5277 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) |
| 5278 | seq_printf(m, "%s %lu\n", mem_cgroup_events_names[i], |
| 5279 | mem_cgroup_read_events(memcg, i)); |
| 5280 | |
| 5281 | for (i = 0; i < NR_LRU_LISTS; i++) |
| 5282 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], |
| 5283 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); |
| 5284 | |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 5285 | /* Hierarchical information */ |
KAMEZAWA Hiroyuki | fee7b54 | 2009-01-07 18:08:26 -0800 | [diff] [blame] | 5286 | { |
| 5287 | unsigned long long limit, memsw_limit; |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 5288 | memcg_get_hierarchical_limit(memcg, &limit, &memsw_limit); |
Johannes Weiner | 78ccf5b | 2012-05-29 15:07:06 -0700 | [diff] [blame] | 5289 | seq_printf(m, "hierarchical_memory_limit %llu\n", limit); |
KAMEZAWA Hiroyuki | fee7b54 | 2009-01-07 18:08:26 -0800 | [diff] [blame] | 5290 | if (do_swap_account) |
Johannes Weiner | 78ccf5b | 2012-05-29 15:07:06 -0700 | [diff] [blame] | 5291 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
| 5292 | memsw_limit); |
KAMEZAWA Hiroyuki | fee7b54 | 2009-01-07 18:08:26 -0800 | [diff] [blame] | 5293 | } |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 5294 | |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 5295 | for (i = 0; i < MEM_CGROUP_STAT_NSTATS; i++) { |
| 5296 | long long val = 0; |
| 5297 | |
Kamezawa Hiroyuki | bff6bb8 | 2012-07-31 16:41:38 -0700 | [diff] [blame] | 5298 | if (i == MEM_CGROUP_STAT_SWAP && !do_swap_account) |
Daisuke Nishimura | 1dd3a27 | 2009-09-23 15:56:43 -0700 | [diff] [blame] | 5299 | continue; |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 5300 | for_each_mem_cgroup_tree(mi, memcg) |
| 5301 | val += mem_cgroup_read_stat(mi, i) * PAGE_SIZE; |
| 5302 | seq_printf(m, "total_%s %lld\n", mem_cgroup_stat_names[i], val); |
| 5303 | } |
| 5304 | |
| 5305 | for (i = 0; i < MEM_CGROUP_EVENTS_NSTATS; i++) { |
| 5306 | unsigned long long val = 0; |
| 5307 | |
| 5308 | for_each_mem_cgroup_tree(mi, memcg) |
| 5309 | val += mem_cgroup_read_events(mi, i); |
| 5310 | seq_printf(m, "total_%s %llu\n", |
| 5311 | mem_cgroup_events_names[i], val); |
| 5312 | } |
| 5313 | |
| 5314 | for (i = 0; i < NR_LRU_LISTS; i++) { |
| 5315 | unsigned long long val = 0; |
| 5316 | |
| 5317 | for_each_mem_cgroup_tree(mi, memcg) |
| 5318 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; |
| 5319 | 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] | 5320 | } |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 5321 | |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 5322 | #ifdef CONFIG_DEBUG_VM |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 5323 | { |
| 5324 | int nid, zid; |
| 5325 | struct mem_cgroup_per_zone *mz; |
Hugh Dickins | 89abfab | 2012-05-29 15:06:53 -0700 | [diff] [blame] | 5326 | struct zone_reclaim_stat *rstat; |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 5327 | unsigned long recent_rotated[2] = {0, 0}; |
| 5328 | unsigned long recent_scanned[2] = {0, 0}; |
| 5329 | |
| 5330 | for_each_online_node(nid) |
| 5331 | for (zid = 0; zid < MAX_NR_ZONES; zid++) { |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 5332 | mz = mem_cgroup_zoneinfo(memcg, nid, zid); |
Hugh Dickins | 89abfab | 2012-05-29 15:06:53 -0700 | [diff] [blame] | 5333 | rstat = &mz->lruvec.reclaim_stat; |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 5334 | |
Hugh Dickins | 89abfab | 2012-05-29 15:06:53 -0700 | [diff] [blame] | 5335 | recent_rotated[0] += rstat->recent_rotated[0]; |
| 5336 | recent_rotated[1] += rstat->recent_rotated[1]; |
| 5337 | recent_scanned[0] += rstat->recent_scanned[0]; |
| 5338 | recent_scanned[1] += rstat->recent_scanned[1]; |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 5339 | } |
Johannes Weiner | 78ccf5b | 2012-05-29 15:07:06 -0700 | [diff] [blame] | 5340 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
| 5341 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); |
| 5342 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); |
| 5343 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 5344 | } |
| 5345 | #endif |
| 5346 | |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 5347 | return 0; |
| 5348 | } |
| 5349 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 5350 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
| 5351 | struct cftype *cft) |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 5352 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 5353 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 5354 | |
KAMEZAWA Hiroyuki | 1f4c025 | 2011-07-26 16:08:21 -0700 | [diff] [blame] | 5355 | return mem_cgroup_swappiness(memcg); |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 5356 | } |
| 5357 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 5358 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
| 5359 | struct cftype *cft, u64 val) |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 5360 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 5361 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Li Zefan | 068b38c | 2009-01-15 13:51:26 -0800 | [diff] [blame] | 5362 | |
Johannes Weiner | 3dae7fe | 2014-06-04 16:07:01 -0700 | [diff] [blame] | 5363 | if (val > 100) |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 5364 | return -EINVAL; |
| 5365 | |
Johannes Weiner | 3dae7fe | 2014-06-04 16:07:01 -0700 | [diff] [blame] | 5366 | if (css_parent(css)) |
| 5367 | memcg->swappiness = val; |
| 5368 | else |
| 5369 | vm_swappiness = val; |
Li Zefan | 068b38c | 2009-01-15 13:51:26 -0800 | [diff] [blame] | 5370 | |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 5371 | return 0; |
| 5372 | } |
| 5373 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5374 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
| 5375 | { |
| 5376 | struct mem_cgroup_threshold_ary *t; |
| 5377 | u64 usage; |
| 5378 | int i; |
| 5379 | |
| 5380 | rcu_read_lock(); |
| 5381 | if (!swap) |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 5382 | t = rcu_dereference(memcg->thresholds.primary); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5383 | else |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 5384 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5385 | |
| 5386 | if (!t) |
| 5387 | goto unlock; |
| 5388 | |
| 5389 | usage = mem_cgroup_usage(memcg, swap); |
| 5390 | |
| 5391 | /* |
Sha Zhengju | 748dad3 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 5392 | * current_threshold points to threshold just below or equal to usage. |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5393 | * If it's not true, a threshold was crossed after last |
| 5394 | * call of __mem_cgroup_threshold(). |
| 5395 | */ |
Phil Carmody | 5407a56 | 2010-05-26 14:42:42 -0700 | [diff] [blame] | 5396 | i = t->current_threshold; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5397 | |
| 5398 | /* |
| 5399 | * Iterate backward over array of thresholds starting from |
| 5400 | * current_threshold and check if a threshold is crossed. |
| 5401 | * If none of thresholds below usage is crossed, we read |
| 5402 | * only one element of the array here. |
| 5403 | */ |
| 5404 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) |
| 5405 | eventfd_signal(t->entries[i].eventfd, 1); |
| 5406 | |
| 5407 | /* i = current_threshold + 1 */ |
| 5408 | i++; |
| 5409 | |
| 5410 | /* |
| 5411 | * Iterate forward over array of thresholds starting from |
| 5412 | * current_threshold+1 and check if a threshold is crossed. |
| 5413 | * If none of thresholds above usage is crossed, we read |
| 5414 | * only one element of the array here. |
| 5415 | */ |
| 5416 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) |
| 5417 | eventfd_signal(t->entries[i].eventfd, 1); |
| 5418 | |
| 5419 | /* Update current_threshold */ |
Phil Carmody | 5407a56 | 2010-05-26 14:42:42 -0700 | [diff] [blame] | 5420 | t->current_threshold = i - 1; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5421 | unlock: |
| 5422 | rcu_read_unlock(); |
| 5423 | } |
| 5424 | |
| 5425 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) |
| 5426 | { |
Kirill A. Shutemov | ad4ca5f | 2010-10-07 12:59:27 -0700 | [diff] [blame] | 5427 | while (memcg) { |
| 5428 | __mem_cgroup_threshold(memcg, false); |
| 5429 | if (do_swap_account) |
| 5430 | __mem_cgroup_threshold(memcg, true); |
| 5431 | |
| 5432 | memcg = parent_mem_cgroup(memcg); |
| 5433 | } |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5434 | } |
| 5435 | |
| 5436 | static int compare_thresholds(const void *a, const void *b) |
| 5437 | { |
| 5438 | const struct mem_cgroup_threshold *_a = a; |
| 5439 | const struct mem_cgroup_threshold *_b = b; |
| 5440 | |
Greg Thelen | 2bff24a | 2013-09-11 14:23:08 -0700 | [diff] [blame] | 5441 | if (_a->threshold > _b->threshold) |
| 5442 | return 1; |
| 5443 | |
| 5444 | if (_a->threshold < _b->threshold) |
| 5445 | return -1; |
| 5446 | |
| 5447 | return 0; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5448 | } |
| 5449 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 5450 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 5451 | { |
| 5452 | struct mem_cgroup_eventfd_list *ev; |
| 5453 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 5454 | list_for_each_entry(ev, &memcg->oom_notify, list) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 5455 | eventfd_signal(ev->eventfd, 1); |
| 5456 | return 0; |
| 5457 | } |
| 5458 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 5459 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 5460 | { |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 5461 | struct mem_cgroup *iter; |
| 5462 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 5463 | for_each_mem_cgroup_tree(iter, memcg) |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 5464 | mem_cgroup_oom_notify_cb(iter); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 5465 | } |
| 5466 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 5467 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 5468 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5469 | { |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 5470 | struct mem_cgroup_thresholds *thresholds; |
| 5471 | struct mem_cgroup_threshold_ary *new; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5472 | u64 threshold, usage; |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 5473 | int i, size, ret; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5474 | |
| 5475 | ret = res_counter_memparse_write_strategy(args, &threshold); |
| 5476 | if (ret) |
| 5477 | return ret; |
| 5478 | |
| 5479 | mutex_lock(&memcg->thresholds_lock); |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 5480 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5481 | if (type == _MEM) |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 5482 | thresholds = &memcg->thresholds; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5483 | else if (type == _MEMSWAP) |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 5484 | thresholds = &memcg->memsw_thresholds; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5485 | else |
| 5486 | BUG(); |
| 5487 | |
| 5488 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); |
| 5489 | |
| 5490 | /* Check if a threshold crossed before adding a new one */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 5491 | if (thresholds->primary) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5492 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
| 5493 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 5494 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5495 | |
| 5496 | /* Allocate memory for new array of thresholds */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 5497 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5498 | GFP_KERNEL); |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 5499 | if (!new) { |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5500 | ret = -ENOMEM; |
| 5501 | goto unlock; |
| 5502 | } |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 5503 | new->size = size; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5504 | |
| 5505 | /* Copy thresholds (if any) to new array */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 5506 | if (thresholds->primary) { |
| 5507 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5508 | sizeof(struct mem_cgroup_threshold)); |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 5509 | } |
| 5510 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5511 | /* Add new threshold */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 5512 | new->entries[size - 1].eventfd = eventfd; |
| 5513 | new->entries[size - 1].threshold = threshold; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5514 | |
| 5515 | /* Sort thresholds. Registering of new threshold isn't time-critical */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 5516 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5517 | compare_thresholds, NULL); |
| 5518 | |
| 5519 | /* Find current threshold */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 5520 | new->current_threshold = -1; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5521 | for (i = 0; i < size; i++) { |
Sha Zhengju | 748dad3 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 5522 | if (new->entries[i].threshold <= usage) { |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5523 | /* |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 5524 | * new->current_threshold will not be used until |
| 5525 | * rcu_assign_pointer(), so it's safe to increment |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5526 | * it here. |
| 5527 | */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 5528 | ++new->current_threshold; |
Sha Zhengju | 748dad3 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 5529 | } else |
| 5530 | break; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5531 | } |
| 5532 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 5533 | /* Free old spare buffer and save old primary buffer as spare */ |
| 5534 | kfree(thresholds->spare); |
| 5535 | thresholds->spare = thresholds->primary; |
| 5536 | |
| 5537 | rcu_assign_pointer(thresholds->primary, new); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5538 | |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 5539 | /* To be sure that nobody uses thresholds */ |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5540 | synchronize_rcu(); |
| 5541 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5542 | unlock: |
| 5543 | mutex_unlock(&memcg->thresholds_lock); |
| 5544 | |
| 5545 | return ret; |
| 5546 | } |
| 5547 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 5548 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 5549 | struct eventfd_ctx *eventfd, const char *args) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5550 | { |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 5551 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 5552 | } |
| 5553 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 5554 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 5555 | struct eventfd_ctx *eventfd, const char *args) |
| 5556 | { |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 5557 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 5558 | } |
| 5559 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 5560 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 5561 | struct eventfd_ctx *eventfd, enum res_type type) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5562 | { |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 5563 | struct mem_cgroup_thresholds *thresholds; |
| 5564 | struct mem_cgroup_threshold_ary *new; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5565 | u64 usage; |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 5566 | int i, j, size; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5567 | |
| 5568 | mutex_lock(&memcg->thresholds_lock); |
| 5569 | if (type == _MEM) |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 5570 | thresholds = &memcg->thresholds; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5571 | else if (type == _MEMSWAP) |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 5572 | thresholds = &memcg->memsw_thresholds; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5573 | else |
| 5574 | BUG(); |
| 5575 | |
Anton Vorontsov | 371528c | 2012-02-24 05:14:46 +0400 | [diff] [blame] | 5576 | if (!thresholds->primary) |
| 5577 | goto unlock; |
| 5578 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5579 | usage = mem_cgroup_usage(memcg, type == _MEMSWAP); |
| 5580 | |
| 5581 | /* Check if a threshold crossed before removing */ |
| 5582 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
| 5583 | |
| 5584 | /* Calculate new number of threshold */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 5585 | size = 0; |
| 5586 | for (i = 0; i < thresholds->primary->size; i++) { |
| 5587 | if (thresholds->primary->entries[i].eventfd != eventfd) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5588 | size++; |
| 5589 | } |
| 5590 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 5591 | new = thresholds->spare; |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 5592 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5593 | /* Set thresholds array to NULL if we don't have thresholds */ |
| 5594 | if (!size) { |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 5595 | kfree(new); |
| 5596 | new = NULL; |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 5597 | goto swap_buffers; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5598 | } |
| 5599 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 5600 | new->size = size; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5601 | |
| 5602 | /* Copy thresholds and find current threshold */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 5603 | new->current_threshold = -1; |
| 5604 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { |
| 5605 | if (thresholds->primary->entries[i].eventfd == eventfd) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5606 | continue; |
| 5607 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 5608 | new->entries[j] = thresholds->primary->entries[i]; |
Sha Zhengju | 748dad3 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 5609 | if (new->entries[j].threshold <= usage) { |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5610 | /* |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 5611 | * new->current_threshold will not be used |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5612 | * until rcu_assign_pointer(), so it's safe to increment |
| 5613 | * it here. |
| 5614 | */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 5615 | ++new->current_threshold; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5616 | } |
| 5617 | j++; |
| 5618 | } |
| 5619 | |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 5620 | swap_buffers: |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 5621 | /* Swap primary and spare array */ |
| 5622 | thresholds->spare = thresholds->primary; |
Sha Zhengju | 8c75776 | 2012-05-10 13:01:45 -0700 | [diff] [blame] | 5623 | /* If all events are unregistered, free the spare array */ |
| 5624 | if (!new) { |
| 5625 | kfree(thresholds->spare); |
| 5626 | thresholds->spare = NULL; |
| 5627 | } |
| 5628 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 5629 | rcu_assign_pointer(thresholds->primary, new); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5630 | |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 5631 | /* To be sure that nobody uses thresholds */ |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5632 | synchronize_rcu(); |
Anton Vorontsov | 371528c | 2012-02-24 05:14:46 +0400 | [diff] [blame] | 5633 | unlock: |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5634 | mutex_unlock(&memcg->thresholds_lock); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 5635 | } |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 5636 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 5637 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 5638 | struct eventfd_ctx *eventfd) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 5639 | { |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 5640 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 5641 | } |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 5642 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 5643 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 5644 | struct eventfd_ctx *eventfd) |
| 5645 | { |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 5646 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 5647 | } |
| 5648 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 5649 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 5650 | struct eventfd_ctx *eventfd, const char *args) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 5651 | { |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 5652 | struct mem_cgroup_eventfd_list *event; |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 5653 | |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 5654 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
| 5655 | if (!event) |
| 5656 | return -ENOMEM; |
| 5657 | |
Michal Hocko | 1af8efe | 2011-07-26 16:08:24 -0700 | [diff] [blame] | 5658 | spin_lock(&memcg_oom_lock); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 5659 | |
| 5660 | event->eventfd = eventfd; |
| 5661 | list_add(&event->list, &memcg->oom_notify); |
| 5662 | |
| 5663 | /* already in OOM ? */ |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 5664 | if (atomic_read(&memcg->under_oom)) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 5665 | eventfd_signal(eventfd, 1); |
Michal Hocko | 1af8efe | 2011-07-26 16:08:24 -0700 | [diff] [blame] | 5666 | spin_unlock(&memcg_oom_lock); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 5667 | |
| 5668 | return 0; |
| 5669 | } |
| 5670 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 5671 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 5672 | struct eventfd_ctx *eventfd) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 5673 | { |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 5674 | struct mem_cgroup_eventfd_list *ev, *tmp; |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 5675 | |
Michal Hocko | 1af8efe | 2011-07-26 16:08:24 -0700 | [diff] [blame] | 5676 | spin_lock(&memcg_oom_lock); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 5677 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 5678 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 5679 | if (ev->eventfd == eventfd) { |
| 5680 | list_del(&ev->list); |
| 5681 | kfree(ev); |
| 5682 | } |
| 5683 | } |
| 5684 | |
Michal Hocko | 1af8efe | 2011-07-26 16:08:24 -0700 | [diff] [blame] | 5685 | spin_unlock(&memcg_oom_lock); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 5686 | } |
| 5687 | |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 5688 | 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] | 5689 | { |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 5690 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf)); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 5691 | |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 5692 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
| 5693 | seq_printf(sf, "under_oom %d\n", (bool)atomic_read(&memcg->under_oom)); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 5694 | return 0; |
| 5695 | } |
| 5696 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 5697 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 5698 | struct cftype *cft, u64 val) |
| 5699 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 5700 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 5701 | |
| 5702 | /* cannot set to root cgroup and only 0 and 1 are allowed */ |
Johannes Weiner | 3dae7fe | 2014-06-04 16:07:01 -0700 | [diff] [blame] | 5703 | if (!css_parent(css) || !((val == 0) || (val == 1))) |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 5704 | return -EINVAL; |
| 5705 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 5706 | memcg->oom_kill_disable = val; |
KAMEZAWA Hiroyuki | 4d845eb | 2010-06-29 15:05:18 -0700 | [diff] [blame] | 5707 | if (!val) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 5708 | memcg_oom_recover(memcg); |
Johannes Weiner | 3dae7fe | 2014-06-04 16:07:01 -0700 | [diff] [blame] | 5709 | |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 5710 | return 0; |
| 5711 | } |
| 5712 | |
Andrew Morton | c255a45 | 2012-07-31 16:43:02 -0700 | [diff] [blame] | 5713 | #ifdef CONFIG_MEMCG_KMEM |
Glauber Costa | cbe128e3 | 2012-04-09 19:36:34 -0300 | [diff] [blame] | 5714 | 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] | 5715 | { |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 5716 | int ret; |
| 5717 | |
Glauber Costa | 2633d7a | 2012-12-18 14:22:34 -0800 | [diff] [blame] | 5718 | memcg->kmemcg_id = -1; |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 5719 | ret = memcg_propagate_kmem(memcg); |
| 5720 | if (ret) |
| 5721 | return ret; |
Glauber Costa | 2633d7a | 2012-12-18 14:22:34 -0800 | [diff] [blame] | 5722 | |
Glauber Costa | 1d62e43 | 2012-04-09 19:36:33 -0300 | [diff] [blame] | 5723 | return mem_cgroup_sockets_init(memcg, ss); |
Michel Lespinasse | 573b400 | 2013-04-29 15:08:13 -0700 | [diff] [blame] | 5724 | } |
Glauber Costa | e5671df | 2011-12-11 21:47:01 +0000 | [diff] [blame] | 5725 | |
Li Zefan | 10d5ebf | 2013-07-08 16:00:33 -0700 | [diff] [blame] | 5726 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
Glauber Costa | d1a4c0b | 2011-12-11 21:47:04 +0000 | [diff] [blame] | 5727 | { |
Glauber Costa | 1d62e43 | 2012-04-09 19:36:33 -0300 | [diff] [blame] | 5728 | mem_cgroup_sockets_destroy(memcg); |
Li Zefan | 10d5ebf | 2013-07-08 16:00:33 -0700 | [diff] [blame] | 5729 | } |
| 5730 | |
| 5731 | static void kmem_cgroup_css_offline(struct mem_cgroup *memcg) |
| 5732 | { |
| 5733 | if (!memcg_kmem_is_active(memcg)) |
| 5734 | return; |
| 5735 | |
| 5736 | /* |
| 5737 | * kmem charges can outlive the cgroup. In the case of slab |
| 5738 | * pages, for instance, a page contain objects from various |
| 5739 | * processes. As we prevent from taking a reference for every |
| 5740 | * such allocation we have to be careful when doing uncharge |
| 5741 | * (see memcg_uncharge_kmem) and here during offlining. |
| 5742 | * |
| 5743 | * The idea is that that only the _last_ uncharge which sees |
| 5744 | * the dead memcg will drop the last reference. An additional |
| 5745 | * reference is taken here before the group is marked dead |
| 5746 | * which is then paired with css_put during uncharge resp. here. |
| 5747 | * |
| 5748 | * Although this might sound strange as this path is called from |
| 5749 | * css_offline() when the referencemight have dropped down to 0 |
| 5750 | * and shouldn't be incremented anymore (css_tryget would fail) |
| 5751 | * we do not have other options because of the kmem allocations |
| 5752 | * lifetime. |
| 5753 | */ |
| 5754 | css_get(&memcg->css); |
Glauber Costa | 7de3768 | 2012-12-18 14:22:07 -0800 | [diff] [blame] | 5755 | |
| 5756 | memcg_kmem_mark_dead(memcg); |
| 5757 | |
| 5758 | if (res_counter_read_u64(&memcg->kmem, RES_USAGE) != 0) |
| 5759 | return; |
| 5760 | |
Glauber Costa | 7de3768 | 2012-12-18 14:22:07 -0800 | [diff] [blame] | 5761 | if (memcg_kmem_test_and_clear_dead(memcg)) |
Li Zefan | 10d5ebf | 2013-07-08 16:00:33 -0700 | [diff] [blame] | 5762 | css_put(&memcg->css); |
Glauber Costa | d1a4c0b | 2011-12-11 21:47:04 +0000 | [diff] [blame] | 5763 | } |
Glauber Costa | e5671df | 2011-12-11 21:47:01 +0000 | [diff] [blame] | 5764 | #else |
Glauber Costa | cbe128e3 | 2012-04-09 19:36:34 -0300 | [diff] [blame] | 5765 | 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] | 5766 | { |
| 5767 | return 0; |
| 5768 | } |
Glauber Costa | d1a4c0b | 2011-12-11 21:47:04 +0000 | [diff] [blame] | 5769 | |
Li Zefan | 10d5ebf | 2013-07-08 16:00:33 -0700 | [diff] [blame] | 5770 | static void memcg_destroy_kmem(struct mem_cgroup *memcg) |
| 5771 | { |
| 5772 | } |
| 5773 | |
| 5774 | static void kmem_cgroup_css_offline(struct mem_cgroup *memcg) |
Glauber Costa | d1a4c0b | 2011-12-11 21:47:04 +0000 | [diff] [blame] | 5775 | { |
| 5776 | } |
Glauber Costa | e5671df | 2011-12-11 21:47:01 +0000 | [diff] [blame] | 5777 | #endif |
| 5778 | |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 5779 | /* |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 5780 | * DO NOT USE IN NEW FILES. |
| 5781 | * |
| 5782 | * "cgroup.event_control" implementation. |
| 5783 | * |
| 5784 | * This is way over-engineered. It tries to support fully configurable |
| 5785 | * events for each user. Such level of flexibility is completely |
| 5786 | * unnecessary especially in the light of the planned unified hierarchy. |
| 5787 | * |
| 5788 | * Please deprecate this and replace with something simpler if at all |
| 5789 | * possible. |
| 5790 | */ |
| 5791 | |
| 5792 | /* |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 5793 | * Unregister event and free resources. |
| 5794 | * |
| 5795 | * Gets called from workqueue. |
| 5796 | */ |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 5797 | static void memcg_event_remove(struct work_struct *work) |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 5798 | { |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 5799 | struct mem_cgroup_event *event = |
| 5800 | container_of(work, struct mem_cgroup_event, remove); |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 5801 | struct mem_cgroup *memcg = event->memcg; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 5802 | |
| 5803 | remove_wait_queue(event->wqh, &event->wait); |
| 5804 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 5805 | event->unregister_event(memcg, event->eventfd); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 5806 | |
| 5807 | /* Notify userspace the event is going away. */ |
| 5808 | eventfd_signal(event->eventfd, 1); |
| 5809 | |
| 5810 | eventfd_ctx_put(event->eventfd); |
| 5811 | kfree(event); |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 5812 | css_put(&memcg->css); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 5813 | } |
| 5814 | |
| 5815 | /* |
| 5816 | * Gets called on POLLHUP on eventfd when user closes it. |
| 5817 | * |
| 5818 | * Called with wqh->lock held and interrupts disabled. |
| 5819 | */ |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 5820 | static int memcg_event_wake(wait_queue_t *wait, unsigned mode, |
| 5821 | int sync, void *key) |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 5822 | { |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 5823 | struct mem_cgroup_event *event = |
| 5824 | container_of(wait, struct mem_cgroup_event, wait); |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 5825 | struct mem_cgroup *memcg = event->memcg; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 5826 | unsigned long flags = (unsigned long)key; |
| 5827 | |
| 5828 | if (flags & POLLHUP) { |
| 5829 | /* |
| 5830 | * If the event has been detached at cgroup removal, we |
| 5831 | * can simply return knowing the other side will cleanup |
| 5832 | * for us. |
| 5833 | * |
| 5834 | * We can't race against event freeing since the other |
| 5835 | * side will require wqh->lock via remove_wait_queue(), |
| 5836 | * which we hold. |
| 5837 | */ |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 5838 | spin_lock(&memcg->event_list_lock); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 5839 | if (!list_empty(&event->list)) { |
| 5840 | list_del_init(&event->list); |
| 5841 | /* |
| 5842 | * We are in atomic context, but cgroup_event_remove() |
| 5843 | * may sleep, so we have to call it in workqueue. |
| 5844 | */ |
| 5845 | schedule_work(&event->remove); |
| 5846 | } |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 5847 | spin_unlock(&memcg->event_list_lock); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 5848 | } |
| 5849 | |
| 5850 | return 0; |
| 5851 | } |
| 5852 | |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 5853 | static void memcg_event_ptable_queue_proc(struct file *file, |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 5854 | wait_queue_head_t *wqh, poll_table *pt) |
| 5855 | { |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 5856 | struct mem_cgroup_event *event = |
| 5857 | container_of(pt, struct mem_cgroup_event, pt); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 5858 | |
| 5859 | event->wqh = wqh; |
| 5860 | add_wait_queue(wqh, &event->wait); |
| 5861 | } |
| 5862 | |
| 5863 | /* |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 5864 | * DO NOT USE IN NEW FILES. |
| 5865 | * |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 5866 | * Parse input and register new cgroup event handler. |
| 5867 | * |
| 5868 | * Input must be in format '<event_fd> <control_fd> <args>'. |
| 5869 | * Interpretation of args is defined by control file implementation. |
| 5870 | */ |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 5871 | static int memcg_write_event_control(struct cgroup_subsys_state *css, |
Tejun Heo | 4d3bb51 | 2014-03-19 10:23:54 -0400 | [diff] [blame] | 5872 | struct cftype *cft, char *buffer) |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 5873 | { |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 5874 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 5875 | struct mem_cgroup_event *event; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 5876 | struct cgroup_subsys_state *cfile_css; |
| 5877 | unsigned int efd, cfd; |
| 5878 | struct fd efile; |
| 5879 | struct fd cfile; |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 5880 | const char *name; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 5881 | char *endp; |
| 5882 | int ret; |
| 5883 | |
| 5884 | efd = simple_strtoul(buffer, &endp, 10); |
| 5885 | if (*endp != ' ') |
| 5886 | return -EINVAL; |
| 5887 | buffer = endp + 1; |
| 5888 | |
| 5889 | cfd = simple_strtoul(buffer, &endp, 10); |
| 5890 | if ((*endp != ' ') && (*endp != '\0')) |
| 5891 | return -EINVAL; |
| 5892 | buffer = endp + 1; |
| 5893 | |
| 5894 | event = kzalloc(sizeof(*event), GFP_KERNEL); |
| 5895 | if (!event) |
| 5896 | return -ENOMEM; |
| 5897 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 5898 | event->memcg = memcg; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 5899 | INIT_LIST_HEAD(&event->list); |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 5900 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
| 5901 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); |
| 5902 | INIT_WORK(&event->remove, memcg_event_remove); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 5903 | |
| 5904 | efile = fdget(efd); |
| 5905 | if (!efile.file) { |
| 5906 | ret = -EBADF; |
| 5907 | goto out_kfree; |
| 5908 | } |
| 5909 | |
| 5910 | event->eventfd = eventfd_ctx_fileget(efile.file); |
| 5911 | if (IS_ERR(event->eventfd)) { |
| 5912 | ret = PTR_ERR(event->eventfd); |
| 5913 | goto out_put_efile; |
| 5914 | } |
| 5915 | |
| 5916 | cfile = fdget(cfd); |
| 5917 | if (!cfile.file) { |
| 5918 | ret = -EBADF; |
| 5919 | goto out_put_eventfd; |
| 5920 | } |
| 5921 | |
| 5922 | /* the process need read permission on control file */ |
| 5923 | /* AV: shouldn't we check that it's been opened for read instead? */ |
| 5924 | ret = inode_permission(file_inode(cfile.file), MAY_READ); |
| 5925 | if (ret < 0) |
| 5926 | goto out_put_cfile; |
| 5927 | |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 5928 | /* |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 5929 | * Determine the event callbacks and set them in @event. This used |
| 5930 | * to be done via struct cftype but cgroup core no longer knows |
| 5931 | * about these events. The following is crude but the whole thing |
| 5932 | * is for compatibility anyway. |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 5933 | * |
| 5934 | * DO NOT ADD NEW FILES. |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 5935 | */ |
| 5936 | name = cfile.file->f_dentry->d_name.name; |
| 5937 | |
| 5938 | if (!strcmp(name, "memory.usage_in_bytes")) { |
| 5939 | event->register_event = mem_cgroup_usage_register_event; |
| 5940 | event->unregister_event = mem_cgroup_usage_unregister_event; |
| 5941 | } else if (!strcmp(name, "memory.oom_control")) { |
| 5942 | event->register_event = mem_cgroup_oom_register_event; |
| 5943 | event->unregister_event = mem_cgroup_oom_unregister_event; |
| 5944 | } else if (!strcmp(name, "memory.pressure_level")) { |
| 5945 | event->register_event = vmpressure_register_event; |
| 5946 | event->unregister_event = vmpressure_unregister_event; |
| 5947 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 5948 | event->register_event = memsw_cgroup_usage_register_event; |
| 5949 | event->unregister_event = memsw_cgroup_usage_unregister_event; |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 5950 | } else { |
| 5951 | ret = -EINVAL; |
| 5952 | goto out_put_cfile; |
| 5953 | } |
| 5954 | |
| 5955 | /* |
Tejun Heo | b5557c4 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 5956 | * Verify @cfile should belong to @css. Also, remaining events are |
| 5957 | * automatically removed on cgroup destruction but the removal is |
| 5958 | * asynchronous, so take an extra ref on @css. |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 5959 | */ |
Tejun Heo | 5a17f54 | 2014-02-11 11:52:47 -0500 | [diff] [blame] | 5960 | cfile_css = css_tryget_from_dir(cfile.file->f_dentry->d_parent, |
| 5961 | &memory_cgrp_subsys); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 5962 | ret = -EINVAL; |
Tejun Heo | 5a17f54 | 2014-02-11 11:52:47 -0500 | [diff] [blame] | 5963 | if (IS_ERR(cfile_css)) |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 5964 | goto out_put_cfile; |
Tejun Heo | 5a17f54 | 2014-02-11 11:52:47 -0500 | [diff] [blame] | 5965 | if (cfile_css != css) { |
| 5966 | css_put(cfile_css); |
| 5967 | goto out_put_cfile; |
| 5968 | } |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 5969 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 5970 | ret = event->register_event(memcg, event->eventfd, buffer); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 5971 | if (ret) |
| 5972 | goto out_put_css; |
| 5973 | |
| 5974 | efile.file->f_op->poll(efile.file, &event->pt); |
| 5975 | |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 5976 | spin_lock(&memcg->event_list_lock); |
| 5977 | list_add(&event->list, &memcg->event_list); |
| 5978 | spin_unlock(&memcg->event_list_lock); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 5979 | |
| 5980 | fdput(cfile); |
| 5981 | fdput(efile); |
| 5982 | |
| 5983 | return 0; |
| 5984 | |
| 5985 | out_put_css: |
Tejun Heo | b5557c4 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 5986 | css_put(css); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 5987 | out_put_cfile: |
| 5988 | fdput(cfile); |
| 5989 | out_put_eventfd: |
| 5990 | eventfd_ctx_put(event->eventfd); |
| 5991 | out_put_efile: |
| 5992 | fdput(efile); |
| 5993 | out_kfree: |
| 5994 | kfree(event); |
| 5995 | |
| 5996 | return ret; |
| 5997 | } |
| 5998 | |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 5999 | static struct cftype mem_cgroup_files[] = { |
| 6000 | { |
Balbir Singh | 0eea103 | 2008-02-07 00:13:57 -0800 | [diff] [blame] | 6001 | .name = "usage_in_bytes", |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 6002 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 6003 | .read_u64 = mem_cgroup_read_u64, |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 6004 | }, |
| 6005 | { |
Pavel Emelyanov | c84872e | 2008-04-29 01:00:17 -0700 | [diff] [blame] | 6006 | .name = "max_usage_in_bytes", |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 6007 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
Pavel Emelyanov | 29f2a4d | 2008-04-29 01:00:21 -0700 | [diff] [blame] | 6008 | .trigger = mem_cgroup_reset, |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 6009 | .read_u64 = mem_cgroup_read_u64, |
Pavel Emelyanov | c84872e | 2008-04-29 01:00:17 -0700 | [diff] [blame] | 6010 | }, |
| 6011 | { |
Balbir Singh | 0eea103 | 2008-02-07 00:13:57 -0800 | [diff] [blame] | 6012 | .name = "limit_in_bytes", |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 6013 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
Paul Menage | 856c13a | 2008-07-25 01:47:04 -0700 | [diff] [blame] | 6014 | .write_string = mem_cgroup_write, |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 6015 | .read_u64 = mem_cgroup_read_u64, |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 6016 | }, |
| 6017 | { |
Balbir Singh | 296c81d | 2009-09-23 15:56:36 -0700 | [diff] [blame] | 6018 | .name = "soft_limit_in_bytes", |
| 6019 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), |
| 6020 | .write_string = mem_cgroup_write, |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 6021 | .read_u64 = mem_cgroup_read_u64, |
Balbir Singh | 296c81d | 2009-09-23 15:56:36 -0700 | [diff] [blame] | 6022 | }, |
| 6023 | { |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 6024 | .name = "failcnt", |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 6025 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
Pavel Emelyanov | 29f2a4d | 2008-04-29 01:00:21 -0700 | [diff] [blame] | 6026 | .trigger = mem_cgroup_reset, |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 6027 | .read_u64 = mem_cgroup_read_u64, |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 6028 | }, |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 6029 | { |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 6030 | .name = "stat", |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 6031 | .seq_show = memcg_stat_show, |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 6032 | }, |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 6033 | { |
| 6034 | .name = "force_empty", |
| 6035 | .trigger = mem_cgroup_force_empty_write, |
| 6036 | }, |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 6037 | { |
| 6038 | .name = "use_hierarchy", |
Tejun Heo | f00baae | 2013-04-15 13:41:15 -0700 | [diff] [blame] | 6039 | .flags = CFTYPE_INSANE, |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 6040 | .write_u64 = mem_cgroup_hierarchy_write, |
| 6041 | .read_u64 = mem_cgroup_hierarchy_read, |
| 6042 | }, |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 6043 | { |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 6044 | .name = "cgroup.event_control", /* XXX: for compat */ |
| 6045 | .write_string = memcg_write_event_control, |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 6046 | .flags = CFTYPE_NO_PREFIX, |
| 6047 | .mode = S_IWUGO, |
| 6048 | }, |
| 6049 | { |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 6050 | .name = "swappiness", |
| 6051 | .read_u64 = mem_cgroup_swappiness_read, |
| 6052 | .write_u64 = mem_cgroup_swappiness_write, |
| 6053 | }, |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 6054 | { |
| 6055 | .name = "move_charge_at_immigrate", |
| 6056 | .read_u64 = mem_cgroup_move_charge_read, |
| 6057 | .write_u64 = mem_cgroup_move_charge_write, |
| 6058 | }, |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 6059 | { |
| 6060 | .name = "oom_control", |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 6061 | .seq_show = mem_cgroup_oom_control_read, |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 6062 | .write_u64 = mem_cgroup_oom_control_write, |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 6063 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
| 6064 | }, |
Anton Vorontsov | 70ddf63 | 2013-04-29 15:08:31 -0700 | [diff] [blame] | 6065 | { |
| 6066 | .name = "pressure_level", |
Anton Vorontsov | 70ddf63 | 2013-04-29 15:08:31 -0700 | [diff] [blame] | 6067 | }, |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 6068 | #ifdef CONFIG_NUMA |
| 6069 | { |
| 6070 | .name = "numa_stat", |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 6071 | .seq_show = memcg_numa_stat_show, |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 6072 | }, |
| 6073 | #endif |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 6074 | #ifdef CONFIG_MEMCG_KMEM |
| 6075 | { |
| 6076 | .name = "kmem.limit_in_bytes", |
| 6077 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), |
| 6078 | .write_string = mem_cgroup_write, |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 6079 | .read_u64 = mem_cgroup_read_u64, |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 6080 | }, |
| 6081 | { |
| 6082 | .name = "kmem.usage_in_bytes", |
| 6083 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 6084 | .read_u64 = mem_cgroup_read_u64, |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 6085 | }, |
| 6086 | { |
| 6087 | .name = "kmem.failcnt", |
| 6088 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), |
| 6089 | .trigger = mem_cgroup_reset, |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 6090 | .read_u64 = mem_cgroup_read_u64, |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 6091 | }, |
| 6092 | { |
| 6093 | .name = "kmem.max_usage_in_bytes", |
| 6094 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), |
| 6095 | .trigger = mem_cgroup_reset, |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 6096 | .read_u64 = mem_cgroup_read_u64, |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 6097 | }, |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 6098 | #ifdef CONFIG_SLABINFO |
| 6099 | { |
| 6100 | .name = "kmem.slabinfo", |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 6101 | .seq_show = mem_cgroup_slabinfo_read, |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 6102 | }, |
| 6103 | #endif |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 6104 | #endif |
Tejun Heo | 6bc1034 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 6105 | { }, /* terminate */ |
Tejun Heo | af36f90 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 6106 | }; |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 6107 | |
Michal Hocko | 2d11085 | 2013-02-22 16:34:43 -0800 | [diff] [blame] | 6108 | #ifdef CONFIG_MEMCG_SWAP |
| 6109 | static struct cftype memsw_cgroup_files[] = { |
| 6110 | { |
| 6111 | .name = "memsw.usage_in_bytes", |
| 6112 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 6113 | .read_u64 = mem_cgroup_read_u64, |
Michal Hocko | 2d11085 | 2013-02-22 16:34:43 -0800 | [diff] [blame] | 6114 | }, |
| 6115 | { |
| 6116 | .name = "memsw.max_usage_in_bytes", |
| 6117 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), |
| 6118 | .trigger = mem_cgroup_reset, |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 6119 | .read_u64 = mem_cgroup_read_u64, |
Michal Hocko | 2d11085 | 2013-02-22 16:34:43 -0800 | [diff] [blame] | 6120 | }, |
| 6121 | { |
| 6122 | .name = "memsw.limit_in_bytes", |
| 6123 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), |
| 6124 | .write_string = mem_cgroup_write, |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 6125 | .read_u64 = mem_cgroup_read_u64, |
Michal Hocko | 2d11085 | 2013-02-22 16:34:43 -0800 | [diff] [blame] | 6126 | }, |
| 6127 | { |
| 6128 | .name = "memsw.failcnt", |
| 6129 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), |
| 6130 | .trigger = mem_cgroup_reset, |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 6131 | .read_u64 = mem_cgroup_read_u64, |
Michal Hocko | 2d11085 | 2013-02-22 16:34:43 -0800 | [diff] [blame] | 6132 | }, |
| 6133 | { }, /* terminate */ |
| 6134 | }; |
| 6135 | #endif |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 6136 | 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] | 6137 | { |
| 6138 | struct mem_cgroup_per_node *pn; |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 6139 | struct mem_cgroup_per_zone *mz; |
KAMEZAWA Hiroyuki | 41e3355 | 2008-04-08 17:41:54 -0700 | [diff] [blame] | 6140 | int zone, tmp = node; |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 6141 | /* |
| 6142 | * This routine is called against possible nodes. |
| 6143 | * But it's BUG to call kmalloc() against offline node. |
| 6144 | * |
| 6145 | * TODO: this routine can waste much memory for nodes which will |
| 6146 | * never be onlined. It's better to use memory hotplug callback |
| 6147 | * function. |
| 6148 | */ |
KAMEZAWA Hiroyuki | 41e3355 | 2008-04-08 17:41:54 -0700 | [diff] [blame] | 6149 | if (!node_state(node, N_NORMAL_MEMORY)) |
| 6150 | tmp = -1; |
Jesper Juhl | 17295c8 | 2011-01-13 15:47:42 -0800 | [diff] [blame] | 6151 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 6152 | if (!pn) |
| 6153 | return 1; |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 6154 | |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 6155 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
| 6156 | mz = &pn->zoneinfo[zone]; |
Hugh Dickins | bea8c15 | 2012-11-16 14:14:54 -0800 | [diff] [blame] | 6157 | lruvec_init(&mz->lruvec); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 6158 | mz->usage_in_excess = 0; |
| 6159 | mz->on_tree = false; |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 6160 | mz->memcg = memcg; |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 6161 | } |
Johannes Weiner | 54f72fe | 2013-07-08 15:59:49 -0700 | [diff] [blame] | 6162 | memcg->nodeinfo[node] = pn; |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 6163 | return 0; |
| 6164 | } |
| 6165 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 6166 | 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] | 6167 | { |
Johannes Weiner | 54f72fe | 2013-07-08 15:59:49 -0700 | [diff] [blame] | 6168 | kfree(memcg->nodeinfo[node]); |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 6169 | } |
| 6170 | |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 6171 | static struct mem_cgroup *mem_cgroup_alloc(void) |
| 6172 | { |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 6173 | struct mem_cgroup *memcg; |
Vladimir Davydov | 8ff69e2 | 2014-01-23 15:52:52 -0800 | [diff] [blame] | 6174 | size_t size; |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 6175 | |
Vladimir Davydov | 8ff69e2 | 2014-01-23 15:52:52 -0800 | [diff] [blame] | 6176 | size = sizeof(struct mem_cgroup); |
| 6177 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 6178 | |
Vladimir Davydov | 8ff69e2 | 2014-01-23 15:52:52 -0800 | [diff] [blame] | 6179 | memcg = kzalloc(size, GFP_KERNEL); |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 6180 | if (!memcg) |
Dan Carpenter | e7bbcdf | 2010-03-23 13:35:12 -0700 | [diff] [blame] | 6181 | return NULL; |
| 6182 | |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 6183 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
| 6184 | if (!memcg->stat) |
Dan Carpenter | d2e61b8 | 2010-11-11 14:05:12 -0800 | [diff] [blame] | 6185 | goto out_free; |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 6186 | spin_lock_init(&memcg->pcp_counter_lock); |
| 6187 | return memcg; |
Dan Carpenter | d2e61b8 | 2010-11-11 14:05:12 -0800 | [diff] [blame] | 6188 | |
| 6189 | out_free: |
Vladimir Davydov | 8ff69e2 | 2014-01-23 15:52:52 -0800 | [diff] [blame] | 6190 | kfree(memcg); |
Dan Carpenter | d2e61b8 | 2010-11-11 14:05:12 -0800 | [diff] [blame] | 6191 | return NULL; |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 6192 | } |
| 6193 | |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 6194 | /* |
Glauber Costa | c8b2a36 | 2012-12-18 14:22:13 -0800 | [diff] [blame] | 6195 | * At destroying mem_cgroup, references from swap_cgroup can remain. |
| 6196 | * (scanning all at force_empty is too costly...) |
| 6197 | * |
| 6198 | * Instead of clearing all references at force_empty, we remember |
| 6199 | * the number of reference from swap_cgroup and free mem_cgroup when |
| 6200 | * it goes down to 0. |
| 6201 | * |
| 6202 | * Removal of cgroup itself succeeds regardless of refs from swap. |
Hugh Dickins | 59927fb | 2012-03-15 15:17:07 -0700 | [diff] [blame] | 6203 | */ |
Glauber Costa | c8b2a36 | 2012-12-18 14:22:13 -0800 | [diff] [blame] | 6204 | |
| 6205 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
Hugh Dickins | 59927fb | 2012-03-15 15:17:07 -0700 | [diff] [blame] | 6206 | { |
Glauber Costa | c8b2a36 | 2012-12-18 14:22:13 -0800 | [diff] [blame] | 6207 | int node; |
Hugh Dickins | 59927fb | 2012-03-15 15:17:07 -0700 | [diff] [blame] | 6208 | |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 6209 | mem_cgroup_remove_from_trees(memcg); |
Glauber Costa | c8b2a36 | 2012-12-18 14:22:13 -0800 | [diff] [blame] | 6210 | |
| 6211 | for_each_node(node) |
| 6212 | free_mem_cgroup_per_zone_info(memcg, node); |
| 6213 | |
| 6214 | free_percpu(memcg->stat); |
| 6215 | |
Glauber Costa | 3f13461 | 2012-05-29 15:07:11 -0700 | [diff] [blame] | 6216 | /* |
| 6217 | * We need to make sure that (at least for now), the jump label |
| 6218 | * destruction code runs outside of the cgroup lock. This is because |
| 6219 | * get_online_cpus(), which is called from the static_branch update, |
| 6220 | * can't be called inside the cgroup_lock. cpusets are the ones |
| 6221 | * enforcing this dependency, so if they ever change, we might as well. |
| 6222 | * |
| 6223 | * schedule_work() will guarantee this happens. Be careful if you need |
| 6224 | * to move this code around, and make sure it is outside |
| 6225 | * the cgroup_lock. |
| 6226 | */ |
Glauber Costa | a8964b9 | 2012-12-18 14:22:09 -0800 | [diff] [blame] | 6227 | disarm_static_keys(memcg); |
Vladimir Davydov | 8ff69e2 | 2014-01-23 15:52:52 -0800 | [diff] [blame] | 6228 | kfree(memcg); |
Hugh Dickins | 59927fb | 2012-03-15 15:17:07 -0700 | [diff] [blame] | 6229 | } |
Glauber Costa | 3afe36b | 2012-05-29 15:07:10 -0700 | [diff] [blame] | 6230 | |
Daisuke Nishimura | 7bcc1bb | 2009-01-29 14:25:11 -0800 | [diff] [blame] | 6231 | /* |
| 6232 | * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled. |
| 6233 | */ |
Glauber Costa | e1aab16 | 2011-12-11 21:47:03 +0000 | [diff] [blame] | 6234 | struct mem_cgroup *parent_mem_cgroup(struct mem_cgroup *memcg) |
Daisuke Nishimura | 7bcc1bb | 2009-01-29 14:25:11 -0800 | [diff] [blame] | 6235 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 6236 | if (!memcg->res.parent) |
Daisuke Nishimura | 7bcc1bb | 2009-01-29 14:25:11 -0800 | [diff] [blame] | 6237 | return NULL; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 6238 | return mem_cgroup_from_res_counter(memcg->res.parent, res); |
Daisuke Nishimura | 7bcc1bb | 2009-01-29 14:25:11 -0800 | [diff] [blame] | 6239 | } |
Glauber Costa | e1aab16 | 2011-12-11 21:47:03 +0000 | [diff] [blame] | 6240 | EXPORT_SYMBOL(parent_mem_cgroup); |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 6241 | |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 6242 | static void __init mem_cgroup_soft_limit_tree_init(void) |
| 6243 | { |
| 6244 | struct mem_cgroup_tree_per_node *rtpn; |
| 6245 | struct mem_cgroup_tree_per_zone *rtpz; |
| 6246 | int tmp, node, zone; |
| 6247 | |
| 6248 | for_each_node(node) { |
| 6249 | tmp = node; |
| 6250 | if (!node_state(node, N_NORMAL_MEMORY)) |
| 6251 | tmp = -1; |
| 6252 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, tmp); |
| 6253 | BUG_ON(!rtpn); |
| 6254 | |
| 6255 | soft_limit_tree.rb_tree_per_node[node] = rtpn; |
| 6256 | |
| 6257 | for (zone = 0; zone < MAX_NR_ZONES; zone++) { |
| 6258 | rtpz = &rtpn->rb_tree_per_zone[zone]; |
| 6259 | rtpz->rb_root = RB_ROOT; |
| 6260 | spin_lock_init(&rtpz->lock); |
| 6261 | } |
| 6262 | } |
| 6263 | } |
| 6264 | |
Li Zefan | 0eb253e | 2009-01-15 13:51:25 -0800 | [diff] [blame] | 6265 | static struct cgroup_subsys_state * __ref |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 6266 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 6267 | { |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 6268 | struct mem_cgroup *memcg; |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 6269 | long error = -ENOMEM; |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 6270 | int node; |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 6271 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 6272 | memcg = mem_cgroup_alloc(); |
| 6273 | if (!memcg) |
KAMEZAWA Hiroyuki | 04046e1 | 2009-04-02 16:57:33 -0700 | [diff] [blame] | 6274 | return ERR_PTR(error); |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 6275 | |
Bob Liu | 3ed28fa | 2012-01-12 17:19:04 -0800 | [diff] [blame] | 6276 | for_each_node(node) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 6277 | if (alloc_mem_cgroup_per_zone_info(memcg, node)) |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 6278 | goto free_out; |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 6279 | |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 6280 | /* root ? */ |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 6281 | if (parent_css == NULL) { |
Hillf Danton | a41c58a | 2011-12-19 17:11:57 -0800 | [diff] [blame] | 6282 | root_mem_cgroup = memcg; |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 6283 | res_counter_init(&memcg->res, NULL); |
| 6284 | res_counter_init(&memcg->memsw, NULL); |
| 6285 | res_counter_init(&memcg->kmem, NULL); |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 6286 | } |
Balbir Singh | 28dbc4b | 2009-01-07 18:08:05 -0800 | [diff] [blame] | 6287 | |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 6288 | memcg->last_scanned_node = MAX_NUMNODES; |
| 6289 | INIT_LIST_HEAD(&memcg->oom_notify); |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 6290 | memcg->move_charge_at_immigrate = 0; |
| 6291 | mutex_init(&memcg->thresholds_lock); |
| 6292 | spin_lock_init(&memcg->move_lock); |
Anton Vorontsov | 70ddf63 | 2013-04-29 15:08:31 -0700 | [diff] [blame] | 6293 | vmpressure_init(&memcg->vmpressure); |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 6294 | INIT_LIST_HEAD(&memcg->event_list); |
| 6295 | spin_lock_init(&memcg->event_list_lock); |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 6296 | |
| 6297 | return &memcg->css; |
| 6298 | |
| 6299 | free_out: |
| 6300 | __mem_cgroup_free(memcg); |
| 6301 | return ERR_PTR(error); |
| 6302 | } |
| 6303 | |
| 6304 | static int |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 6305 | mem_cgroup_css_online(struct cgroup_subsys_state *css) |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 6306 | { |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 6307 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
| 6308 | struct mem_cgroup *parent = mem_cgroup_from_css(css_parent(css)); |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 6309 | |
Li Zefan | 4219b2d | 2013-09-23 16:56:29 +0800 | [diff] [blame] | 6310 | if (css->cgroup->id > MEM_CGROUP_ID_MAX) |
| 6311 | return -ENOSPC; |
| 6312 | |
Tejun Heo | 6387698 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 6313 | if (!parent) |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 6314 | return 0; |
| 6315 | |
Glauber Costa | 0999821 | 2013-02-22 16:34:55 -0800 | [diff] [blame] | 6316 | mutex_lock(&memcg_create_mutex); |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 6317 | |
| 6318 | memcg->use_hierarchy = parent->use_hierarchy; |
| 6319 | memcg->oom_kill_disable = parent->oom_kill_disable; |
| 6320 | memcg->swappiness = mem_cgroup_swappiness(parent); |
| 6321 | |
| 6322 | if (parent->use_hierarchy) { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 6323 | res_counter_init(&memcg->res, &parent->res); |
| 6324 | res_counter_init(&memcg->memsw, &parent->memsw); |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 6325 | res_counter_init(&memcg->kmem, &parent->kmem); |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 6326 | |
Daisuke Nishimura | 7bcc1bb | 2009-01-29 14:25:11 -0800 | [diff] [blame] | 6327 | /* |
Li Zefan | 8d76a97 | 2013-07-08 16:00:36 -0700 | [diff] [blame] | 6328 | * No need to take a reference to the parent because cgroup |
| 6329 | * core guarantees its existence. |
Daisuke Nishimura | 7bcc1bb | 2009-01-29 14:25:11 -0800 | [diff] [blame] | 6330 | */ |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 6331 | } else { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 6332 | res_counter_init(&memcg->res, NULL); |
| 6333 | res_counter_init(&memcg->memsw, NULL); |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 6334 | res_counter_init(&memcg->kmem, NULL); |
Tejun Heo | 8c7f6ed | 2012-09-13 12:20:58 -0700 | [diff] [blame] | 6335 | /* |
| 6336 | * Deeper hierachy with use_hierarchy == false doesn't make |
| 6337 | * much sense so let cgroup subsystem know about this |
| 6338 | * unfortunate state in our controller. |
| 6339 | */ |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 6340 | if (parent != root_mem_cgroup) |
Tejun Heo | 073219e | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 6341 | memory_cgrp_subsys.broken_hierarchy = true; |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 6342 | } |
Glauber Costa | 0999821 | 2013-02-22 16:34:55 -0800 | [diff] [blame] | 6343 | mutex_unlock(&memcg_create_mutex); |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 6344 | |
Tejun Heo | 073219e | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 6345 | return memcg_init_kmem(memcg, &memory_cgrp_subsys); |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 6346 | } |
| 6347 | |
Michal Hocko | 5f57816 | 2013-04-29 15:07:17 -0700 | [diff] [blame] | 6348 | /* |
| 6349 | * Announce all parents that a group from their hierarchy is gone. |
| 6350 | */ |
| 6351 | static void mem_cgroup_invalidate_reclaim_iterators(struct mem_cgroup *memcg) |
| 6352 | { |
| 6353 | struct mem_cgroup *parent = memcg; |
| 6354 | |
| 6355 | while ((parent = parent_mem_cgroup(parent))) |
Johannes Weiner | 519ebea | 2013-07-03 15:04:51 -0700 | [diff] [blame] | 6356 | mem_cgroup_iter_invalidate(parent); |
Michal Hocko | 5f57816 | 2013-04-29 15:07:17 -0700 | [diff] [blame] | 6357 | |
| 6358 | /* |
| 6359 | * if the root memcg is not hierarchical we have to check it |
| 6360 | * explicitely. |
| 6361 | */ |
| 6362 | if (!root_mem_cgroup->use_hierarchy) |
Johannes Weiner | 519ebea | 2013-07-03 15:04:51 -0700 | [diff] [blame] | 6363 | mem_cgroup_iter_invalidate(root_mem_cgroup); |
Michal Hocko | 5f57816 | 2013-04-29 15:07:17 -0700 | [diff] [blame] | 6364 | } |
| 6365 | |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 6366 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
KAMEZAWA Hiroyuki | df878fb | 2008-02-07 00:14:28 -0800 | [diff] [blame] | 6367 | { |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 6368 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 6369 | struct mem_cgroup_event *event, *tmp; |
Filipe Brandenburger | 4fb1a86 | 2014-03-03 15:38:25 -0800 | [diff] [blame] | 6370 | struct cgroup_subsys_state *iter; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 6371 | |
| 6372 | /* |
| 6373 | * Unregister events and notify userspace. |
| 6374 | * Notify userspace about cgroup removing only after rmdir of cgroup |
| 6375 | * directory to avoid race between userspace and kernelspace. |
| 6376 | */ |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 6377 | spin_lock(&memcg->event_list_lock); |
| 6378 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 6379 | list_del_init(&event->list); |
| 6380 | schedule_work(&event->remove); |
| 6381 | } |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 6382 | spin_unlock(&memcg->event_list_lock); |
KAMEZAWA Hiroyuki | ec64f51 | 2009-04-02 16:57:26 -0700 | [diff] [blame] | 6383 | |
Li Zefan | 10d5ebf | 2013-07-08 16:00:33 -0700 | [diff] [blame] | 6384 | kmem_cgroup_css_offline(memcg); |
| 6385 | |
Michal Hocko | 5f57816 | 2013-04-29 15:07:17 -0700 | [diff] [blame] | 6386 | mem_cgroup_invalidate_reclaim_iterators(memcg); |
Filipe Brandenburger | 4fb1a86 | 2014-03-03 15:38:25 -0800 | [diff] [blame] | 6387 | |
| 6388 | /* |
| 6389 | * This requires that offlining is serialized. Right now that is |
| 6390 | * guaranteed because css_killed_work_fn() holds the cgroup_mutex. |
| 6391 | */ |
| 6392 | css_for_each_descendant_post(iter, css) |
| 6393 | mem_cgroup_reparent_charges(mem_cgroup_from_css(iter)); |
| 6394 | |
Glauber Costa | 1f458cb | 2012-12-18 14:22:50 -0800 | [diff] [blame] | 6395 | mem_cgroup_destroy_all_caches(memcg); |
Michal Hocko | 33cb876 | 2013-07-31 13:53:51 -0700 | [diff] [blame] | 6396 | vmpressure_cleanup(&memcg->vmpressure); |
KAMEZAWA Hiroyuki | df878fb | 2008-02-07 00:14:28 -0800 | [diff] [blame] | 6397 | } |
| 6398 | |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 6399 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 6400 | { |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 6401 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Johannes Weiner | 96f1c58 | 2013-12-12 17:12:34 -0800 | [diff] [blame] | 6402 | /* |
| 6403 | * XXX: css_offline() would be where we should reparent all |
| 6404 | * memory to prepare the cgroup for destruction. However, |
| 6405 | * memcg does not do css_tryget() and res_counter charging |
| 6406 | * under the same RCU lock region, which means that charging |
| 6407 | * could race with offlining. Offlining only happens to |
| 6408 | * cgroups with no tasks in them but charges can show up |
| 6409 | * without any tasks from the swapin path when the target |
| 6410 | * memcg is looked up from the swapout record and not from the |
| 6411 | * current task as it usually is. A race like this can leak |
| 6412 | * charges and put pages with stale cgroup pointers into |
| 6413 | * circulation: |
| 6414 | * |
| 6415 | * #0 #1 |
| 6416 | * lookup_swap_cgroup_id() |
| 6417 | * rcu_read_lock() |
| 6418 | * mem_cgroup_lookup() |
| 6419 | * css_tryget() |
| 6420 | * rcu_read_unlock() |
| 6421 | * disable css_tryget() |
| 6422 | * call_rcu() |
| 6423 | * offline_css() |
| 6424 | * reparent_charges() |
| 6425 | * res_counter_charge() |
| 6426 | * css_put() |
| 6427 | * css_free() |
| 6428 | * pc->mem_cgroup = dead memcg |
| 6429 | * add page to lru |
| 6430 | * |
| 6431 | * The bulk of the charges are still moved in offline_css() to |
| 6432 | * avoid pinning a lot of pages in case a long-term reference |
| 6433 | * like a swapout record is deferring the css_free() to long |
| 6434 | * after offlining. But this makes sure we catch any charges |
| 6435 | * made after offlining: |
| 6436 | */ |
| 6437 | mem_cgroup_reparent_charges(memcg); |
Daisuke Nishimura | c268e99 | 2009-01-15 13:51:13 -0800 | [diff] [blame] | 6438 | |
Li Zefan | 10d5ebf | 2013-07-08 16:00:33 -0700 | [diff] [blame] | 6439 | memcg_destroy_kmem(memcg); |
Li Zefan | 465939a | 2013-07-08 16:00:38 -0700 | [diff] [blame] | 6440 | __mem_cgroup_free(memcg); |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 6441 | } |
| 6442 | |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 6443 | #ifdef CONFIG_MMU |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 6444 | /* Handlers for move charge at task migration. */ |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 6445 | #define PRECHARGE_COUNT_AT_ONCE 256 |
| 6446 | static int mem_cgroup_do_precharge(unsigned long count) |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 6447 | { |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 6448 | int ret = 0; |
| 6449 | int batch_count = PRECHARGE_COUNT_AT_ONCE; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 6450 | struct mem_cgroup *memcg = mc.to; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6451 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 6452 | if (mem_cgroup_is_root(memcg)) { |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 6453 | mc.precharge += count; |
| 6454 | /* we don't need css_get for root */ |
| 6455 | return ret; |
| 6456 | } |
| 6457 | /* try to charge at once */ |
| 6458 | if (count > 1) { |
| 6459 | struct res_counter *dummy; |
| 6460 | /* |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 6461 | * "memcg" cannot be under rmdir() because we've already checked |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 6462 | * by cgroup_lock_live_cgroup() that it is not removed and we |
| 6463 | * are still under the same cgroup_mutex. So we can postpone |
| 6464 | * css_get(). |
| 6465 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 6466 | if (res_counter_charge(&memcg->res, PAGE_SIZE * count, &dummy)) |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 6467 | goto one_by_one; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 6468 | if (do_swap_account && res_counter_charge(&memcg->memsw, |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 6469 | PAGE_SIZE * count, &dummy)) { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 6470 | res_counter_uncharge(&memcg->res, PAGE_SIZE * count); |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 6471 | goto one_by_one; |
| 6472 | } |
| 6473 | mc.precharge += count; |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 6474 | return ret; |
| 6475 | } |
| 6476 | one_by_one: |
| 6477 | /* fall back to one by one charge */ |
| 6478 | while (count--) { |
| 6479 | if (signal_pending(current)) { |
| 6480 | ret = -EINTR; |
| 6481 | break; |
| 6482 | } |
| 6483 | if (!batch_count--) { |
| 6484 | batch_count = PRECHARGE_COUNT_AT_ONCE; |
| 6485 | cond_resched(); |
| 6486 | } |
Johannes Weiner | 6d1fdc4 | 2014-04-07 15:37:45 -0700 | [diff] [blame] | 6487 | ret = mem_cgroup_try_charge(memcg, GFP_KERNEL, 1, false); |
KAMEZAWA Hiroyuki | 38c5d72 | 2012-01-12 17:19:01 -0800 | [diff] [blame] | 6488 | if (ret) |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 6489 | /* mem_cgroup_clear_mc() will do uncharge later */ |
KAMEZAWA Hiroyuki | 38c5d72 | 2012-01-12 17:19:01 -0800 | [diff] [blame] | 6490 | return ret; |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 6491 | mc.precharge++; |
| 6492 | } |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6493 | return ret; |
| 6494 | } |
| 6495 | |
| 6496 | /** |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 6497 | * get_mctgt_type - get target type of moving charge |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6498 | * @vma: the vma the pte to be checked belongs |
| 6499 | * @addr: the address corresponding to the pte to be checked |
| 6500 | * @ptent: the pte to be checked |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 6501 | * @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] | 6502 | * |
| 6503 | * Returns |
| 6504 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. |
| 6505 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for |
| 6506 | * move charge. if @target is not NULL, the page is stored in target->page |
| 6507 | * with extra refcnt got(Callers should handle it). |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 6508 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
| 6509 | * target for charge migration. if @target is not NULL, the entry is stored |
| 6510 | * in target->ent. |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6511 | * |
| 6512 | * Called with pte lock held. |
| 6513 | */ |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6514 | union mc_target { |
| 6515 | struct page *page; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 6516 | swp_entry_t ent; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6517 | }; |
| 6518 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6519 | enum mc_target_type { |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 6520 | MC_TARGET_NONE = 0, |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6521 | MC_TARGET_PAGE, |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 6522 | MC_TARGET_SWAP, |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6523 | }; |
| 6524 | |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 6525 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
| 6526 | unsigned long addr, pte_t ptent) |
| 6527 | { |
| 6528 | struct page *page = vm_normal_page(vma, addr, ptent); |
| 6529 | |
| 6530 | if (!page || !page_mapped(page)) |
| 6531 | return NULL; |
| 6532 | if (PageAnon(page)) { |
| 6533 | /* we don't move shared anon */ |
KAMEZAWA Hiroyuki | 4b91355 | 2012-05-29 15:06:51 -0700 | [diff] [blame] | 6534 | if (!move_anon()) |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 6535 | return NULL; |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 6536 | } else if (!move_file()) |
| 6537 | /* we ignore mapcount for file pages */ |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 6538 | return NULL; |
| 6539 | if (!get_page_unless_zero(page)) |
| 6540 | return NULL; |
| 6541 | |
| 6542 | return page; |
| 6543 | } |
| 6544 | |
KAMEZAWA Hiroyuki | 4b91355 | 2012-05-29 15:06:51 -0700 | [diff] [blame] | 6545 | #ifdef CONFIG_SWAP |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 6546 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
| 6547 | unsigned long addr, pte_t ptent, swp_entry_t *entry) |
| 6548 | { |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 6549 | struct page *page = NULL; |
| 6550 | swp_entry_t ent = pte_to_swp_entry(ptent); |
| 6551 | |
| 6552 | if (!move_anon() || non_swap_entry(ent)) |
| 6553 | return NULL; |
KAMEZAWA Hiroyuki | 4b91355 | 2012-05-29 15:06:51 -0700 | [diff] [blame] | 6554 | /* |
| 6555 | * Because lookup_swap_cache() updates some statistics counter, |
| 6556 | * we call find_get_page() with swapper_space directly. |
| 6557 | */ |
Shaohua Li | 33806f0 | 2013-02-22 16:34:37 -0800 | [diff] [blame] | 6558 | page = find_get_page(swap_address_space(ent), ent.val); |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 6559 | if (do_swap_account) |
| 6560 | entry->val = ent.val; |
| 6561 | |
| 6562 | return page; |
| 6563 | } |
KAMEZAWA Hiroyuki | 4b91355 | 2012-05-29 15:06:51 -0700 | [diff] [blame] | 6564 | #else |
| 6565 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
| 6566 | unsigned long addr, pte_t ptent, swp_entry_t *entry) |
| 6567 | { |
| 6568 | return NULL; |
| 6569 | } |
| 6570 | #endif |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 6571 | |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 6572 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
| 6573 | unsigned long addr, pte_t ptent, swp_entry_t *entry) |
| 6574 | { |
| 6575 | struct page *page = NULL; |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 6576 | struct address_space *mapping; |
| 6577 | pgoff_t pgoff; |
| 6578 | |
| 6579 | if (!vma->vm_file) /* anonymous vma */ |
| 6580 | return NULL; |
| 6581 | if (!move_file()) |
| 6582 | return NULL; |
| 6583 | |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 6584 | mapping = vma->vm_file->f_mapping; |
| 6585 | if (pte_none(ptent)) |
| 6586 | pgoff = linear_page_index(vma, addr); |
| 6587 | else /* pte_file(ptent) is true */ |
| 6588 | pgoff = pte_to_pgoff(ptent); |
| 6589 | |
| 6590 | /* 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] | 6591 | #ifdef CONFIG_SWAP |
| 6592 | /* shmem/tmpfs may report page out on swap: account for that too. */ |
Johannes Weiner | 139b6a6 | 2014-05-06 12:50:05 -0700 | [diff] [blame] | 6593 | if (shmem_mapping(mapping)) { |
| 6594 | page = find_get_entry(mapping, pgoff); |
| 6595 | if (radix_tree_exceptional_entry(page)) { |
| 6596 | swp_entry_t swp = radix_to_swp_entry(page); |
| 6597 | if (do_swap_account) |
| 6598 | *entry = swp; |
| 6599 | page = find_get_page(swap_address_space(swp), swp.val); |
| 6600 | } |
| 6601 | } else |
| 6602 | page = find_get_page(mapping, pgoff); |
| 6603 | #else |
| 6604 | page = find_get_page(mapping, pgoff); |
Hugh Dickins | aa3b189 | 2011-08-03 16:21:24 -0700 | [diff] [blame] | 6605 | #endif |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 6606 | return page; |
| 6607 | } |
| 6608 | |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 6609 | 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] | 6610 | unsigned long addr, pte_t ptent, union mc_target *target) |
| 6611 | { |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 6612 | struct page *page = NULL; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6613 | struct page_cgroup *pc; |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 6614 | enum mc_target_type ret = MC_TARGET_NONE; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 6615 | swp_entry_t ent = { .val = 0 }; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6616 | |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 6617 | if (pte_present(ptent)) |
| 6618 | page = mc_handle_present_pte(vma, addr, ptent); |
| 6619 | else if (is_swap_pte(ptent)) |
| 6620 | page = mc_handle_swap_pte(vma, addr, ptent, &ent); |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 6621 | else if (pte_none(ptent) || pte_file(ptent)) |
| 6622 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 6623 | |
| 6624 | if (!page && !ent.val) |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 6625 | return ret; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 6626 | if (page) { |
| 6627 | pc = lookup_page_cgroup(page); |
| 6628 | /* |
| 6629 | * Do only loose check w/o page_cgroup lock. |
| 6630 | * mem_cgroup_move_account() checks the pc is valid or not under |
| 6631 | * the lock. |
| 6632 | */ |
| 6633 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { |
| 6634 | ret = MC_TARGET_PAGE; |
| 6635 | if (target) |
| 6636 | target->page = page; |
| 6637 | } |
| 6638 | if (!ret || !target) |
| 6639 | put_page(page); |
| 6640 | } |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 6641 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
| 6642 | if (ent.val && !ret && |
Li Zefan | 34c00c3 | 2013-09-23 16:56:01 +0800 | [diff] [blame] | 6643 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
KAMEZAWA Hiroyuki | 7f0f154 | 2010-05-11 14:06:58 -0700 | [diff] [blame] | 6644 | ret = MC_TARGET_SWAP; |
| 6645 | if (target) |
| 6646 | target->ent = ent; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 6647 | } |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6648 | return ret; |
| 6649 | } |
| 6650 | |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 6651 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
| 6652 | /* |
| 6653 | * We don't consider swapping or file mapped pages because THP does not |
| 6654 | * support them for now. |
| 6655 | * Caller should make sure that pmd_trans_huge(pmd) is true. |
| 6656 | */ |
| 6657 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, |
| 6658 | unsigned long addr, pmd_t pmd, union mc_target *target) |
| 6659 | { |
| 6660 | struct page *page = NULL; |
| 6661 | struct page_cgroup *pc; |
| 6662 | enum mc_target_type ret = MC_TARGET_NONE; |
| 6663 | |
| 6664 | page = pmd_page(pmd); |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 6665 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 6666 | if (!move_anon()) |
| 6667 | return ret; |
| 6668 | pc = lookup_page_cgroup(page); |
| 6669 | if (PageCgroupUsed(pc) && pc->mem_cgroup == mc.from) { |
| 6670 | ret = MC_TARGET_PAGE; |
| 6671 | if (target) { |
| 6672 | get_page(page); |
| 6673 | target->page = page; |
| 6674 | } |
| 6675 | } |
| 6676 | return ret; |
| 6677 | } |
| 6678 | #else |
| 6679 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, |
| 6680 | unsigned long addr, pmd_t pmd, union mc_target *target) |
| 6681 | { |
| 6682 | return MC_TARGET_NONE; |
| 6683 | } |
| 6684 | #endif |
| 6685 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6686 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
| 6687 | unsigned long addr, unsigned long end, |
| 6688 | struct mm_walk *walk) |
| 6689 | { |
| 6690 | struct vm_area_struct *vma = walk->private; |
| 6691 | pte_t *pte; |
| 6692 | spinlock_t *ptl; |
| 6693 | |
Kirill A. Shutemov | bf92915 | 2013-11-14 14:30:54 -0800 | [diff] [blame] | 6694 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 6695 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
| 6696 | mc.precharge += HPAGE_PMD_NR; |
Kirill A. Shutemov | bf92915 | 2013-11-14 14:30:54 -0800 | [diff] [blame] | 6697 | spin_unlock(ptl); |
Andrea Arcangeli | 1a5a990 | 2012-03-21 16:33:42 -0700 | [diff] [blame] | 6698 | return 0; |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 6699 | } |
Dave Hansen | 0331932 | 2011-03-22 16:32:56 -0700 | [diff] [blame] | 6700 | |
Andrea Arcangeli | 45f83ce | 2012-03-28 14:42:40 -0700 | [diff] [blame] | 6701 | if (pmd_trans_unstable(pmd)) |
| 6702 | return 0; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6703 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
| 6704 | for (; addr != end; pte++, addr += PAGE_SIZE) |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 6705 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6706 | mc.precharge++; /* increment precharge temporarily */ |
| 6707 | pte_unmap_unlock(pte - 1, ptl); |
| 6708 | cond_resched(); |
| 6709 | |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 6710 | return 0; |
| 6711 | } |
| 6712 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6713 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
| 6714 | { |
| 6715 | unsigned long precharge; |
| 6716 | struct vm_area_struct *vma; |
| 6717 | |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 6718 | down_read(&mm->mmap_sem); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6719 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
| 6720 | struct mm_walk mem_cgroup_count_precharge_walk = { |
| 6721 | .pmd_entry = mem_cgroup_count_precharge_pte_range, |
| 6722 | .mm = mm, |
| 6723 | .private = vma, |
| 6724 | }; |
| 6725 | if (is_vm_hugetlb_page(vma)) |
| 6726 | continue; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6727 | walk_page_range(vma->vm_start, vma->vm_end, |
| 6728 | &mem_cgroup_count_precharge_walk); |
| 6729 | } |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 6730 | up_read(&mm->mmap_sem); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6731 | |
| 6732 | precharge = mc.precharge; |
| 6733 | mc.precharge = 0; |
| 6734 | |
| 6735 | return precharge; |
| 6736 | } |
| 6737 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6738 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
| 6739 | { |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 6740 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
| 6741 | |
| 6742 | VM_BUG_ON(mc.moving_task); |
| 6743 | mc.moving_task = current; |
| 6744 | return mem_cgroup_do_precharge(precharge); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6745 | } |
| 6746 | |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 6747 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
| 6748 | static void __mem_cgroup_clear_mc(void) |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6749 | { |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 6750 | struct mem_cgroup *from = mc.from; |
| 6751 | struct mem_cgroup *to = mc.to; |
Li Zefan | 4050377 | 2013-07-08 16:00:34 -0700 | [diff] [blame] | 6752 | int i; |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 6753 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6754 | /* we must uncharge all the leftover precharges from mc.to */ |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 6755 | if (mc.precharge) { |
| 6756 | __mem_cgroup_cancel_charge(mc.to, mc.precharge); |
| 6757 | mc.precharge = 0; |
| 6758 | } |
| 6759 | /* |
| 6760 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so |
| 6761 | * we must uncharge here. |
| 6762 | */ |
| 6763 | if (mc.moved_charge) { |
| 6764 | __mem_cgroup_cancel_charge(mc.from, mc.moved_charge); |
| 6765 | mc.moved_charge = 0; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6766 | } |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 6767 | /* we must fixup refcnts and charges */ |
| 6768 | if (mc.moved_swap) { |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 6769 | /* uncharge swap account from the old cgroup */ |
| 6770 | if (!mem_cgroup_is_root(mc.from)) |
| 6771 | res_counter_uncharge(&mc.from->memsw, |
| 6772 | PAGE_SIZE * mc.moved_swap); |
Li Zefan | 4050377 | 2013-07-08 16:00:34 -0700 | [diff] [blame] | 6773 | |
| 6774 | for (i = 0; i < mc.moved_swap; i++) |
| 6775 | css_put(&mc.from->css); |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 6776 | |
| 6777 | if (!mem_cgroup_is_root(mc.to)) { |
| 6778 | /* |
| 6779 | * we charged both to->res and to->memsw, so we should |
| 6780 | * uncharge to->res. |
| 6781 | */ |
| 6782 | res_counter_uncharge(&mc.to->res, |
| 6783 | PAGE_SIZE * mc.moved_swap); |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 6784 | } |
Li Zefan | 4050377 | 2013-07-08 16:00:34 -0700 | [diff] [blame] | 6785 | /* we've already done css_get(mc.to) */ |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 6786 | mc.moved_swap = 0; |
| 6787 | } |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 6788 | memcg_oom_recover(from); |
| 6789 | memcg_oom_recover(to); |
| 6790 | wake_up_all(&mc.waitq); |
| 6791 | } |
| 6792 | |
| 6793 | static void mem_cgroup_clear_mc(void) |
| 6794 | { |
| 6795 | struct mem_cgroup *from = mc.from; |
| 6796 | |
| 6797 | /* |
| 6798 | * we must clear moving_task before waking up waiters at the end of |
| 6799 | * task migration. |
| 6800 | */ |
| 6801 | mc.moving_task = NULL; |
| 6802 | __mem_cgroup_clear_mc(); |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 6803 | spin_lock(&mc.lock); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6804 | mc.from = NULL; |
| 6805 | mc.to = NULL; |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 6806 | spin_unlock(&mc.lock); |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 6807 | mem_cgroup_end_move(from); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6808 | } |
| 6809 | |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 6810 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
Li Zefan | 761b3ef | 2012-01-31 13:47:36 +0800 | [diff] [blame] | 6811 | struct cgroup_taskset *tset) |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 6812 | { |
Tejun Heo | 2f7ee56 | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 6813 | struct task_struct *p = cgroup_taskset_first(tset); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 6814 | int ret = 0; |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 6815 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Glauber Costa | ee5e847 | 2013-02-22 16:34:50 -0800 | [diff] [blame] | 6816 | unsigned long move_charge_at_immigrate; |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 6817 | |
Glauber Costa | ee5e847 | 2013-02-22 16:34:50 -0800 | [diff] [blame] | 6818 | /* |
| 6819 | * We are now commited to this value whatever it is. Changes in this |
| 6820 | * tunable will only affect upcoming migrations, not the current one. |
| 6821 | * So we need to save it, and keep it going. |
| 6822 | */ |
| 6823 | move_charge_at_immigrate = memcg->move_charge_at_immigrate; |
| 6824 | if (move_charge_at_immigrate) { |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 6825 | struct mm_struct *mm; |
| 6826 | struct mem_cgroup *from = mem_cgroup_from_task(p); |
| 6827 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 6828 | VM_BUG_ON(from == memcg); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 6829 | |
| 6830 | mm = get_task_mm(p); |
| 6831 | if (!mm) |
| 6832 | return 0; |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 6833 | /* We move charges only when we move a owner of the mm */ |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6834 | if (mm->owner == p) { |
| 6835 | VM_BUG_ON(mc.from); |
| 6836 | VM_BUG_ON(mc.to); |
| 6837 | VM_BUG_ON(mc.precharge); |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 6838 | VM_BUG_ON(mc.moved_charge); |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 6839 | VM_BUG_ON(mc.moved_swap); |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 6840 | mem_cgroup_start_move(from); |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 6841 | spin_lock(&mc.lock); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6842 | mc.from = from; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 6843 | mc.to = memcg; |
Glauber Costa | ee5e847 | 2013-02-22 16:34:50 -0800 | [diff] [blame] | 6844 | mc.immigrate_flags = move_charge_at_immigrate; |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 6845 | spin_unlock(&mc.lock); |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 6846 | /* We set mc.moving_task later */ |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 6847 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6848 | ret = mem_cgroup_precharge_mc(mm); |
| 6849 | if (ret) |
| 6850 | mem_cgroup_clear_mc(); |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 6851 | } |
| 6852 | mmput(mm); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 6853 | } |
| 6854 | return ret; |
| 6855 | } |
| 6856 | |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 6857 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
Li Zefan | 761b3ef | 2012-01-31 13:47:36 +0800 | [diff] [blame] | 6858 | struct cgroup_taskset *tset) |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 6859 | { |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6860 | mem_cgroup_clear_mc(); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 6861 | } |
| 6862 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6863 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
| 6864 | unsigned long addr, unsigned long end, |
| 6865 | struct mm_walk *walk) |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 6866 | { |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6867 | int ret = 0; |
| 6868 | struct vm_area_struct *vma = walk->private; |
| 6869 | pte_t *pte; |
| 6870 | spinlock_t *ptl; |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 6871 | enum mc_target_type target_type; |
| 6872 | union mc_target target; |
| 6873 | struct page *page; |
| 6874 | struct page_cgroup *pc; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6875 | |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 6876 | /* |
| 6877 | * We don't take compound_lock() here but no race with splitting thp |
| 6878 | * happens because: |
| 6879 | * - if pmd_trans_huge_lock() returns 1, the relevant thp is not |
| 6880 | * under splitting, which means there's no concurrent thp split, |
| 6881 | * - if another thread runs into split_huge_page() just after we |
| 6882 | * entered this if-block, the thread must wait for page table lock |
| 6883 | * to be unlocked in __split_huge_page_splitting(), where the main |
| 6884 | * part of thp split is not executed yet. |
| 6885 | */ |
Kirill A. Shutemov | bf92915 | 2013-11-14 14:30:54 -0800 | [diff] [blame] | 6886 | if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { |
Hugh Dickins | 62ade86 | 2012-05-18 11:28:34 -0700 | [diff] [blame] | 6887 | if (mc.precharge < HPAGE_PMD_NR) { |
Kirill A. Shutemov | bf92915 | 2013-11-14 14:30:54 -0800 | [diff] [blame] | 6888 | spin_unlock(ptl); |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 6889 | return 0; |
| 6890 | } |
| 6891 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); |
| 6892 | if (target_type == MC_TARGET_PAGE) { |
| 6893 | page = target.page; |
| 6894 | if (!isolate_lru_page(page)) { |
| 6895 | pc = lookup_page_cgroup(page); |
| 6896 | if (!mem_cgroup_move_account(page, HPAGE_PMD_NR, |
KAMEZAWA Hiroyuki | 2f3479b | 2012-05-29 15:07:04 -0700 | [diff] [blame] | 6897 | pc, mc.from, mc.to)) { |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 6898 | mc.precharge -= HPAGE_PMD_NR; |
| 6899 | mc.moved_charge += HPAGE_PMD_NR; |
| 6900 | } |
| 6901 | putback_lru_page(page); |
| 6902 | } |
| 6903 | put_page(page); |
| 6904 | } |
Kirill A. Shutemov | bf92915 | 2013-11-14 14:30:54 -0800 | [diff] [blame] | 6905 | spin_unlock(ptl); |
Andrea Arcangeli | 1a5a990 | 2012-03-21 16:33:42 -0700 | [diff] [blame] | 6906 | return 0; |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 6907 | } |
| 6908 | |
Andrea Arcangeli | 45f83ce | 2012-03-28 14:42:40 -0700 | [diff] [blame] | 6909 | if (pmd_trans_unstable(pmd)) |
| 6910 | return 0; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6911 | retry: |
| 6912 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
| 6913 | for (; addr != end; addr += PAGE_SIZE) { |
| 6914 | pte_t ptent = *(pte++); |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 6915 | swp_entry_t ent; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6916 | |
| 6917 | if (!mc.precharge) |
| 6918 | break; |
| 6919 | |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 6920 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6921 | case MC_TARGET_PAGE: |
| 6922 | page = target.page; |
| 6923 | if (isolate_lru_page(page)) |
| 6924 | goto put; |
| 6925 | pc = lookup_page_cgroup(page); |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 6926 | if (!mem_cgroup_move_account(page, 1, pc, |
KAMEZAWA Hiroyuki | 2f3479b | 2012-05-29 15:07:04 -0700 | [diff] [blame] | 6927 | mc.from, mc.to)) { |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6928 | mc.precharge--; |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 6929 | /* we uncharge from mc.from later. */ |
| 6930 | mc.moved_charge++; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6931 | } |
| 6932 | putback_lru_page(page); |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 6933 | put: /* get_mctgt_type() gets the page */ |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6934 | put_page(page); |
| 6935 | break; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 6936 | case MC_TARGET_SWAP: |
| 6937 | ent = target.ent; |
Hugh Dickins | e91cbb4 | 2012-05-29 15:06:51 -0700 | [diff] [blame] | 6938 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 6939 | mc.precharge--; |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 6940 | /* we fixup refcnts and charges later. */ |
| 6941 | mc.moved_swap++; |
| 6942 | } |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 6943 | break; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6944 | default: |
| 6945 | break; |
| 6946 | } |
| 6947 | } |
| 6948 | pte_unmap_unlock(pte - 1, ptl); |
| 6949 | cond_resched(); |
| 6950 | |
| 6951 | if (addr != end) { |
| 6952 | /* |
| 6953 | * We have consumed all precharges we got in can_attach(). |
| 6954 | * We try charge one by one, but don't do any additional |
| 6955 | * charges to mc.to if we have failed in charge once in attach() |
| 6956 | * phase. |
| 6957 | */ |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 6958 | ret = mem_cgroup_do_precharge(1); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6959 | if (!ret) |
| 6960 | goto retry; |
| 6961 | } |
| 6962 | |
| 6963 | return ret; |
| 6964 | } |
| 6965 | |
| 6966 | static void mem_cgroup_move_charge(struct mm_struct *mm) |
| 6967 | { |
| 6968 | struct vm_area_struct *vma; |
| 6969 | |
| 6970 | lru_add_drain_all(); |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 6971 | retry: |
| 6972 | if (unlikely(!down_read_trylock(&mm->mmap_sem))) { |
| 6973 | /* |
| 6974 | * Someone who are holding the mmap_sem might be waiting in |
| 6975 | * waitq. So we cancel all extra charges, wake up all waiters, |
| 6976 | * and retry. Because we cancel precharges, we might not be able |
| 6977 | * to move enough charges, but moving charge is a best-effort |
| 6978 | * feature anyway, so it wouldn't be a big problem. |
| 6979 | */ |
| 6980 | __mem_cgroup_clear_mc(); |
| 6981 | cond_resched(); |
| 6982 | goto retry; |
| 6983 | } |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6984 | for (vma = mm->mmap; vma; vma = vma->vm_next) { |
| 6985 | int ret; |
| 6986 | struct mm_walk mem_cgroup_move_charge_walk = { |
| 6987 | .pmd_entry = mem_cgroup_move_charge_pte_range, |
| 6988 | .mm = mm, |
| 6989 | .private = vma, |
| 6990 | }; |
| 6991 | if (is_vm_hugetlb_page(vma)) |
| 6992 | continue; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6993 | ret = walk_page_range(vma->vm_start, vma->vm_end, |
| 6994 | &mem_cgroup_move_charge_walk); |
| 6995 | if (ret) |
| 6996 | /* |
| 6997 | * means we have consumed all precharges and failed in |
| 6998 | * doing additional charge. Just abandon here. |
| 6999 | */ |
| 7000 | break; |
| 7001 | } |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 7002 | up_read(&mm->mmap_sem); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 7003 | } |
| 7004 | |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 7005 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
Li Zefan | 761b3ef | 2012-01-31 13:47:36 +0800 | [diff] [blame] | 7006 | struct cgroup_taskset *tset) |
Balbir Singh | 67e465a | 2008-02-07 00:13:54 -0800 | [diff] [blame] | 7007 | { |
Tejun Heo | 2f7ee56 | 2011-12-12 18:12:21 -0800 | [diff] [blame] | 7008 | struct task_struct *p = cgroup_taskset_first(tset); |
KOSAKI Motohiro | a433658 | 2011-06-15 15:08:13 -0700 | [diff] [blame] | 7009 | struct mm_struct *mm = get_task_mm(p); |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 7010 | |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 7011 | if (mm) { |
KOSAKI Motohiro | a433658 | 2011-06-15 15:08:13 -0700 | [diff] [blame] | 7012 | if (mc.to) |
| 7013 | mem_cgroup_move_charge(mm); |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 7014 | mmput(mm); |
| 7015 | } |
KOSAKI Motohiro | a433658 | 2011-06-15 15:08:13 -0700 | [diff] [blame] | 7016 | if (mc.to) |
| 7017 | mem_cgroup_clear_mc(); |
Balbir Singh | 67e465a | 2008-02-07 00:13:54 -0800 | [diff] [blame] | 7018 | } |
Daisuke Nishimura | 5cfb80a | 2010-03-23 13:35:11 -0700 | [diff] [blame] | 7019 | #else /* !CONFIG_MMU */ |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 7020 | static int mem_cgroup_can_attach(struct cgroup_subsys_state *css, |
Li Zefan | 761b3ef | 2012-01-31 13:47:36 +0800 | [diff] [blame] | 7021 | struct cgroup_taskset *tset) |
Daisuke Nishimura | 5cfb80a | 2010-03-23 13:35:11 -0700 | [diff] [blame] | 7022 | { |
| 7023 | return 0; |
| 7024 | } |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 7025 | static void mem_cgroup_cancel_attach(struct cgroup_subsys_state *css, |
Li Zefan | 761b3ef | 2012-01-31 13:47:36 +0800 | [diff] [blame] | 7026 | struct cgroup_taskset *tset) |
Daisuke Nishimura | 5cfb80a | 2010-03-23 13:35:11 -0700 | [diff] [blame] | 7027 | { |
| 7028 | } |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 7029 | static void mem_cgroup_move_task(struct cgroup_subsys_state *css, |
Li Zefan | 761b3ef | 2012-01-31 13:47:36 +0800 | [diff] [blame] | 7030 | struct cgroup_taskset *tset) |
Daisuke Nishimura | 5cfb80a | 2010-03-23 13:35:11 -0700 | [diff] [blame] | 7031 | { |
| 7032 | } |
| 7033 | #endif |
Balbir Singh | 67e465a | 2008-02-07 00:13:54 -0800 | [diff] [blame] | 7034 | |
Tejun Heo | f00baae | 2013-04-15 13:41:15 -0700 | [diff] [blame] | 7035 | /* |
| 7036 | * Cgroup retains root cgroups across [un]mount cycles making it necessary |
| 7037 | * to verify sane_behavior flag on each mount attempt. |
| 7038 | */ |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 7039 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
Tejun Heo | f00baae | 2013-04-15 13:41:15 -0700 | [diff] [blame] | 7040 | { |
| 7041 | /* |
| 7042 | * use_hierarchy is forced with sane_behavior. cgroup core |
| 7043 | * guarantees that @root doesn't have any children, so turning it |
| 7044 | * on for the root memcg is enough. |
| 7045 | */ |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 7046 | if (cgroup_sane_behavior(root_css->cgroup)) |
| 7047 | mem_cgroup_from_css(root_css)->use_hierarchy = true; |
Tejun Heo | f00baae | 2013-04-15 13:41:15 -0700 | [diff] [blame] | 7048 | } |
| 7049 | |
Tejun Heo | 073219e | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 7050 | struct cgroup_subsys memory_cgrp_subsys = { |
Tejun Heo | 92fb974 | 2012-11-19 08:13:38 -0800 | [diff] [blame] | 7051 | .css_alloc = mem_cgroup_css_alloc, |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 7052 | .css_online = mem_cgroup_css_online, |
Tejun Heo | 92fb974 | 2012-11-19 08:13:38 -0800 | [diff] [blame] | 7053 | .css_offline = mem_cgroup_css_offline, |
| 7054 | .css_free = mem_cgroup_css_free, |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 7055 | .can_attach = mem_cgroup_can_attach, |
| 7056 | .cancel_attach = mem_cgroup_cancel_attach, |
Balbir Singh | 67e465a | 2008-02-07 00:13:54 -0800 | [diff] [blame] | 7057 | .attach = mem_cgroup_move_task, |
Tejun Heo | f00baae | 2013-04-15 13:41:15 -0700 | [diff] [blame] | 7058 | .bind = mem_cgroup_bind, |
Tejun Heo | 6bc1034 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 7059 | .base_cftypes = mem_cgroup_files, |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 7060 | .early_init = 0, |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 7061 | }; |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 7062 | |
Andrew Morton | c255a45 | 2012-07-31 16:43:02 -0700 | [diff] [blame] | 7063 | #ifdef CONFIG_MEMCG_SWAP |
Michal Hocko | a42c390 | 2010-11-24 12:57:08 -0800 | [diff] [blame] | 7064 | static int __init enable_swap_account(char *s) |
| 7065 | { |
Michal Hocko | a2c8990 | 2011-05-24 17:12:50 -0700 | [diff] [blame] | 7066 | if (!strcmp(s, "1")) |
Michal Hocko | a42c390 | 2010-11-24 12:57:08 -0800 | [diff] [blame] | 7067 | really_do_swap_account = 1; |
Michal Hocko | a2c8990 | 2011-05-24 17:12:50 -0700 | [diff] [blame] | 7068 | else if (!strcmp(s, "0")) |
Michal Hocko | a42c390 | 2010-11-24 12:57:08 -0800 | [diff] [blame] | 7069 | really_do_swap_account = 0; |
| 7070 | return 1; |
| 7071 | } |
Michal Hocko | a2c8990 | 2011-05-24 17:12:50 -0700 | [diff] [blame] | 7072 | __setup("swapaccount=", enable_swap_account); |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 7073 | |
Michal Hocko | 2d11085 | 2013-02-22 16:34:43 -0800 | [diff] [blame] | 7074 | static void __init memsw_file_init(void) |
| 7075 | { |
Tejun Heo | 073219e | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 7076 | WARN_ON(cgroup_add_cftypes(&memory_cgrp_subsys, memsw_cgroup_files)); |
Michal Hocko | 2d11085 | 2013-02-22 16:34:43 -0800 | [diff] [blame] | 7077 | } |
Michal Hocko | 6acc8b0 | 2013-02-22 16:34:45 -0800 | [diff] [blame] | 7078 | |
| 7079 | static void __init enable_swap_cgroup(void) |
| 7080 | { |
| 7081 | if (!mem_cgroup_disabled() && really_do_swap_account) { |
| 7082 | do_swap_account = 1; |
| 7083 | memsw_file_init(); |
| 7084 | } |
| 7085 | } |
| 7086 | |
Michal Hocko | 2d11085 | 2013-02-22 16:34:43 -0800 | [diff] [blame] | 7087 | #else |
Michal Hocko | 6acc8b0 | 2013-02-22 16:34:45 -0800 | [diff] [blame] | 7088 | static void __init enable_swap_cgroup(void) |
Michal Hocko | 2d11085 | 2013-02-22 16:34:43 -0800 | [diff] [blame] | 7089 | { |
| 7090 | } |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 7091 | #endif |
Michal Hocko | 2d11085 | 2013-02-22 16:34:43 -0800 | [diff] [blame] | 7092 | |
| 7093 | /* |
Michal Hocko | 1081312 | 2013-02-22 16:35:41 -0800 | [diff] [blame] | 7094 | * subsys_initcall() for memory controller. |
| 7095 | * |
| 7096 | * Some parts like hotcpu_notifier() have to be initialized from this context |
| 7097 | * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically |
| 7098 | * everything that doesn't depend on a specific mem_cgroup structure should |
| 7099 | * be initialized from here. |
Michal Hocko | 2d11085 | 2013-02-22 16:34:43 -0800 | [diff] [blame] | 7100 | */ |
| 7101 | static int __init mem_cgroup_init(void) |
| 7102 | { |
| 7103 | hotcpu_notifier(memcg_cpu_hotplug_callback, 0); |
Michal Hocko | 6acc8b0 | 2013-02-22 16:34:45 -0800 | [diff] [blame] | 7104 | enable_swap_cgroup(); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 7105 | mem_cgroup_soft_limit_tree_init(); |
Michal Hocko | e477749 | 2013-02-22 16:35:40 -0800 | [diff] [blame] | 7106 | memcg_stock_init(); |
Michal Hocko | 2d11085 | 2013-02-22 16:34:43 -0800 | [diff] [blame] | 7107 | return 0; |
| 7108 | } |
| 7109 | subsys_initcall(mem_cgroup_init); |