Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * kernel/cpuset.c |
| 3 | * |
| 4 | * Processor and Memory placement constraints for sets of tasks. |
| 5 | * |
| 6 | * Copyright (C) 2003 BULL SA. |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 7 | * Copyright (C) 2004-2007 Silicon Graphics, Inc. |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 8 | * Copyright (C) 2006 Google, Inc |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 9 | * |
| 10 | * Portions derived from Patrick Mochel's sysfs code. |
| 11 | * sysfs is Copyright (c) 2001-3 Patrick Mochel |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 12 | * |
Paul Jackson | 825a46a | 2006-03-24 03:16:03 -0800 | [diff] [blame] | 13 | * 2003-10-10 Written by Simon Derr. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 14 | * 2003-10-22 Updates by Stephen Hemminger. |
Paul Jackson | 825a46a | 2006-03-24 03:16:03 -0800 | [diff] [blame] | 15 | * 2004 May-July Rework by Paul Jackson. |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 16 | * 2006 Rework by Paul Menage to use generic cgroups |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 17 | * |
| 18 | * This file is subject to the terms and conditions of the GNU General Public |
| 19 | * License. See the file COPYING in the main directory of the Linux |
| 20 | * distribution for more details. |
| 21 | */ |
| 22 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 23 | #include <linux/cpu.h> |
| 24 | #include <linux/cpumask.h> |
| 25 | #include <linux/cpuset.h> |
| 26 | #include <linux/err.h> |
| 27 | #include <linux/errno.h> |
| 28 | #include <linux/file.h> |
| 29 | #include <linux/fs.h> |
| 30 | #include <linux/init.h> |
| 31 | #include <linux/interrupt.h> |
| 32 | #include <linux/kernel.h> |
| 33 | #include <linux/kmod.h> |
| 34 | #include <linux/list.h> |
Paul Jackson | 68860ec | 2005-10-30 15:02:36 -0800 | [diff] [blame] | 35 | #include <linux/mempolicy.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 36 | #include <linux/mm.h> |
| 37 | #include <linux/module.h> |
| 38 | #include <linux/mount.h> |
| 39 | #include <linux/namei.h> |
| 40 | #include <linux/pagemap.h> |
| 41 | #include <linux/proc_fs.h> |
Paul Jackson | 6b9c260 | 2006-01-08 01:02:02 -0800 | [diff] [blame] | 42 | #include <linux/rcupdate.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 43 | #include <linux/sched.h> |
| 44 | #include <linux/seq_file.h> |
David Quigley | 22fb52d | 2006-06-23 02:04:00 -0700 | [diff] [blame] | 45 | #include <linux/security.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 46 | #include <linux/slab.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 47 | #include <linux/spinlock.h> |
| 48 | #include <linux/stat.h> |
| 49 | #include <linux/string.h> |
| 50 | #include <linux/time.h> |
| 51 | #include <linux/backing-dev.h> |
| 52 | #include <linux/sort.h> |
| 53 | |
| 54 | #include <asm/uaccess.h> |
| 55 | #include <asm/atomic.h> |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 56 | #include <linux/mutex.h> |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 57 | #include <linux/kfifo.h> |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 58 | #include <linux/workqueue.h> |
| 59 | #include <linux/cgroup.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 60 | |
Paul Jackson | 202f72d | 2006-01-08 01:01:57 -0800 | [diff] [blame] | 61 | /* |
| 62 | * Tracks how many cpusets are currently defined in system. |
| 63 | * When there is only one cpuset (the root cpuset) we can |
| 64 | * short circuit some hooks. |
| 65 | */ |
Paul Jackson | 7edc596 | 2006-01-08 01:02:03 -0800 | [diff] [blame] | 66 | int number_of_cpusets __read_mostly; |
Paul Jackson | 202f72d | 2006-01-08 01:01:57 -0800 | [diff] [blame] | 67 | |
Paul Menage | 2df167a | 2008-02-07 00:14:45 -0800 | [diff] [blame] | 68 | /* Forward declare cgroup structures */ |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 69 | struct cgroup_subsys cpuset_subsys; |
| 70 | struct cpuset; |
| 71 | |
Paul Jackson | 3e0d98b | 2006-01-08 01:01:49 -0800 | [diff] [blame] | 72 | /* See "Frequency meter" comments, below. */ |
| 73 | |
| 74 | struct fmeter { |
| 75 | int cnt; /* unprocessed events count */ |
| 76 | int val; /* most recent output value */ |
| 77 | time_t time; /* clock (secs) when val computed */ |
| 78 | spinlock_t lock; /* guards read or write of above */ |
| 79 | }; |
| 80 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 81 | struct cpuset { |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 82 | struct cgroup_subsys_state css; |
| 83 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 84 | unsigned long flags; /* "unsigned long" so bitops work */ |
| 85 | cpumask_t cpus_allowed; /* CPUs allowed to tasks in cpuset */ |
| 86 | nodemask_t mems_allowed; /* Memory Nodes allowed to tasks */ |
| 87 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 88 | struct cpuset *parent; /* my parent */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 89 | |
| 90 | /* |
| 91 | * Copy of global cpuset_mems_generation as of the most |
| 92 | * recent time this cpuset changed its mems_allowed. |
| 93 | */ |
Paul Jackson | 3e0d98b | 2006-01-08 01:01:49 -0800 | [diff] [blame] | 94 | int mems_generation; |
| 95 | |
| 96 | struct fmeter fmeter; /* memory_pressure filter */ |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 97 | |
| 98 | /* partition number for rebuild_sched_domains() */ |
| 99 | int pn; |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 100 | |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 101 | /* for custom sched domain */ |
| 102 | int relax_domain_level; |
| 103 | |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 104 | /* used for walking a cpuset heirarchy */ |
| 105 | struct list_head stack_list; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 106 | }; |
| 107 | |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 108 | /* Retrieve the cpuset for a cgroup */ |
| 109 | static inline struct cpuset *cgroup_cs(struct cgroup *cont) |
| 110 | { |
| 111 | return container_of(cgroup_subsys_state(cont, cpuset_subsys_id), |
| 112 | struct cpuset, css); |
| 113 | } |
| 114 | |
| 115 | /* Retrieve the cpuset for a task */ |
| 116 | static inline struct cpuset *task_cs(struct task_struct *task) |
| 117 | { |
| 118 | return container_of(task_subsys_state(task, cpuset_subsys_id), |
| 119 | struct cpuset, css); |
| 120 | } |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 121 | struct cpuset_hotplug_scanner { |
| 122 | struct cgroup_scanner scan; |
| 123 | struct cgroup *to; |
| 124 | }; |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 125 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 126 | /* bits in struct cpuset flags field */ |
| 127 | typedef enum { |
| 128 | CS_CPU_EXCLUSIVE, |
| 129 | CS_MEM_EXCLUSIVE, |
Paul Menage | 7860836 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 130 | CS_MEM_HARDWALL, |
Paul Jackson | 45b07ef | 2006-01-08 01:00:56 -0800 | [diff] [blame] | 131 | CS_MEMORY_MIGRATE, |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 132 | CS_SCHED_LOAD_BALANCE, |
Paul Jackson | 825a46a | 2006-03-24 03:16:03 -0800 | [diff] [blame] | 133 | CS_SPREAD_PAGE, |
| 134 | CS_SPREAD_SLAB, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 135 | } cpuset_flagbits_t; |
| 136 | |
| 137 | /* convenient tests for these bits */ |
| 138 | static inline int is_cpu_exclusive(const struct cpuset *cs) |
| 139 | { |
Paul Jackson | 7b5b9ef | 2006-03-24 03:16:00 -0800 | [diff] [blame] | 140 | return test_bit(CS_CPU_EXCLUSIVE, &cs->flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 141 | } |
| 142 | |
| 143 | static inline int is_mem_exclusive(const struct cpuset *cs) |
| 144 | { |
Paul Jackson | 7b5b9ef | 2006-03-24 03:16:00 -0800 | [diff] [blame] | 145 | return test_bit(CS_MEM_EXCLUSIVE, &cs->flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 146 | } |
| 147 | |
Paul Menage | 7860836 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 148 | static inline int is_mem_hardwall(const struct cpuset *cs) |
| 149 | { |
| 150 | return test_bit(CS_MEM_HARDWALL, &cs->flags); |
| 151 | } |
| 152 | |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 153 | static inline int is_sched_load_balance(const struct cpuset *cs) |
| 154 | { |
| 155 | return test_bit(CS_SCHED_LOAD_BALANCE, &cs->flags); |
| 156 | } |
| 157 | |
Paul Jackson | 45b07ef | 2006-01-08 01:00:56 -0800 | [diff] [blame] | 158 | static inline int is_memory_migrate(const struct cpuset *cs) |
| 159 | { |
Paul Jackson | 7b5b9ef | 2006-03-24 03:16:00 -0800 | [diff] [blame] | 160 | return test_bit(CS_MEMORY_MIGRATE, &cs->flags); |
Paul Jackson | 45b07ef | 2006-01-08 01:00:56 -0800 | [diff] [blame] | 161 | } |
| 162 | |
Paul Jackson | 825a46a | 2006-03-24 03:16:03 -0800 | [diff] [blame] | 163 | static inline int is_spread_page(const struct cpuset *cs) |
| 164 | { |
| 165 | return test_bit(CS_SPREAD_PAGE, &cs->flags); |
| 166 | } |
| 167 | |
| 168 | static inline int is_spread_slab(const struct cpuset *cs) |
| 169 | { |
| 170 | return test_bit(CS_SPREAD_SLAB, &cs->flags); |
| 171 | } |
| 172 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 173 | /* |
Paul Jackson | 151a442 | 2006-03-24 03:16:11 -0800 | [diff] [blame] | 174 | * Increment this integer everytime any cpuset changes its |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 175 | * mems_allowed value. Users of cpusets can track this generation |
| 176 | * number, and avoid having to lock and reload mems_allowed unless |
| 177 | * the cpuset they're using changes generation. |
| 178 | * |
Paul Menage | 2df167a | 2008-02-07 00:14:45 -0800 | [diff] [blame] | 179 | * A single, global generation is needed because cpuset_attach_task() could |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 180 | * reattach a task to a different cpuset, which must not have its |
| 181 | * generation numbers aliased with those of that tasks previous cpuset. |
| 182 | * |
| 183 | * Generations are needed for mems_allowed because one task cannot |
Paul Menage | 2df167a | 2008-02-07 00:14:45 -0800 | [diff] [blame] | 184 | * modify another's memory placement. So we must enable every task, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 185 | * on every visit to __alloc_pages(), to efficiently check whether |
| 186 | * its current->cpuset->mems_allowed has changed, requiring an update |
| 187 | * of its current->mems_allowed. |
Paul Jackson | 151a442 | 2006-03-24 03:16:11 -0800 | [diff] [blame] | 188 | * |
Paul Menage | 2df167a | 2008-02-07 00:14:45 -0800 | [diff] [blame] | 189 | * Since writes to cpuset_mems_generation are guarded by the cgroup lock |
Paul Jackson | 151a442 | 2006-03-24 03:16:11 -0800 | [diff] [blame] | 190 | * there is no need to mark it atomic. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 191 | */ |
Paul Jackson | 151a442 | 2006-03-24 03:16:11 -0800 | [diff] [blame] | 192 | static int cpuset_mems_generation; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 193 | |
| 194 | static struct cpuset top_cpuset = { |
| 195 | .flags = ((1 << CS_CPU_EXCLUSIVE) | (1 << CS_MEM_EXCLUSIVE)), |
| 196 | .cpus_allowed = CPU_MASK_ALL, |
| 197 | .mems_allowed = NODE_MASK_ALL, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 198 | }; |
| 199 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 200 | /* |
Paul Menage | 2df167a | 2008-02-07 00:14:45 -0800 | [diff] [blame] | 201 | * There are two global mutexes guarding cpuset structures. The first |
| 202 | * is the main control groups cgroup_mutex, accessed via |
| 203 | * cgroup_lock()/cgroup_unlock(). The second is the cpuset-specific |
| 204 | * callback_mutex, below. They can nest. It is ok to first take |
| 205 | * cgroup_mutex, then nest callback_mutex. We also require taking |
| 206 | * task_lock() when dereferencing a task's cpuset pointer. See "The |
| 207 | * task_lock() exception", at the end of this comment. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 208 | * |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 209 | * A task must hold both mutexes to modify cpusets. If a task |
Paul Menage | 2df167a | 2008-02-07 00:14:45 -0800 | [diff] [blame] | 210 | * holds cgroup_mutex, then it blocks others wanting that mutex, |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 211 | * ensuring that it is the only task able to also acquire callback_mutex |
Paul Jackson | 053199e | 2005-10-30 15:02:30 -0800 | [diff] [blame] | 212 | * and be able to modify cpusets. It can perform various checks on |
| 213 | * the cpuset structure first, knowing nothing will change. It can |
Paul Menage | 2df167a | 2008-02-07 00:14:45 -0800 | [diff] [blame] | 214 | * also allocate memory while just holding cgroup_mutex. While it is |
Paul Jackson | 053199e | 2005-10-30 15:02:30 -0800 | [diff] [blame] | 215 | * performing these checks, various callback routines can briefly |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 216 | * acquire callback_mutex to query cpusets. Once it is ready to make |
| 217 | * the changes, it takes callback_mutex, blocking everyone else. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 218 | * |
Paul Jackson | 053199e | 2005-10-30 15:02:30 -0800 | [diff] [blame] | 219 | * Calls to the kernel memory allocator can not be made while holding |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 220 | * callback_mutex, as that would risk double tripping on callback_mutex |
Paul Jackson | 053199e | 2005-10-30 15:02:30 -0800 | [diff] [blame] | 221 | * from one of the callbacks into the cpuset code from within |
| 222 | * __alloc_pages(). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 223 | * |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 224 | * If a task is only holding callback_mutex, then it has read-only |
Paul Jackson | 053199e | 2005-10-30 15:02:30 -0800 | [diff] [blame] | 225 | * access to cpusets. |
| 226 | * |
| 227 | * The task_struct fields mems_allowed and mems_generation may only |
| 228 | * be accessed in the context of that task, so require no locks. |
| 229 | * |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 230 | * The cpuset_common_file_read() handlers only hold callback_mutex across |
Paul Jackson | 053199e | 2005-10-30 15:02:30 -0800 | [diff] [blame] | 231 | * small pieces of code, such as when reading out possibly multi-word |
| 232 | * cpumasks and nodemasks. |
| 233 | * |
Paul Menage | 2df167a | 2008-02-07 00:14:45 -0800 | [diff] [blame] | 234 | * Accessing a task's cpuset should be done in accordance with the |
| 235 | * guidelines for accessing subsystem state in kernel/cgroup.c |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 236 | */ |
| 237 | |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 238 | static DEFINE_MUTEX(callback_mutex); |
Paul Jackson | 4247bdc | 2005-09-10 00:26:06 -0700 | [diff] [blame] | 239 | |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 240 | /* This is ugly, but preserves the userspace API for existing cpuset |
| 241 | * users. If someone tries to mount the "cpuset" filesystem, we |
| 242 | * silently switch it to mount "cgroup" instead */ |
David Howells | 454e239 | 2006-06-23 02:02:57 -0700 | [diff] [blame] | 243 | static int cpuset_get_sb(struct file_system_type *fs_type, |
| 244 | int flags, const char *unused_dev_name, |
| 245 | void *data, struct vfsmount *mnt) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 246 | { |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 247 | struct file_system_type *cgroup_fs = get_fs_type("cgroup"); |
| 248 | int ret = -ENODEV; |
| 249 | if (cgroup_fs) { |
| 250 | char mountopts[] = |
| 251 | "cpuset,noprefix," |
| 252 | "release_agent=/sbin/cpuset_release_agent"; |
| 253 | ret = cgroup_fs->get_sb(cgroup_fs, flags, |
| 254 | unused_dev_name, mountopts, mnt); |
| 255 | put_filesystem(cgroup_fs); |
| 256 | } |
| 257 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 258 | } |
| 259 | |
| 260 | static struct file_system_type cpuset_fs_type = { |
| 261 | .name = "cpuset", |
| 262 | .get_sb = cpuset_get_sb, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 263 | }; |
| 264 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 265 | /* |
| 266 | * Return in *pmask the portion of a cpusets's cpus_allowed that |
| 267 | * are online. If none are online, walk up the cpuset hierarchy |
| 268 | * until we find one that does have some online cpus. If we get |
| 269 | * all the way to the top and still haven't found any online cpus, |
| 270 | * return cpu_online_map. Or if passed a NULL cs from an exit'ing |
| 271 | * task, return cpu_online_map. |
| 272 | * |
| 273 | * One way or another, we guarantee to return some non-empty subset |
| 274 | * of cpu_online_map. |
| 275 | * |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 276 | * Call with callback_mutex held. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 277 | */ |
| 278 | |
| 279 | static void guarantee_online_cpus(const struct cpuset *cs, cpumask_t *pmask) |
| 280 | { |
| 281 | while (cs && !cpus_intersects(cs->cpus_allowed, cpu_online_map)) |
| 282 | cs = cs->parent; |
| 283 | if (cs) |
| 284 | cpus_and(*pmask, cs->cpus_allowed, cpu_online_map); |
| 285 | else |
| 286 | *pmask = cpu_online_map; |
| 287 | BUG_ON(!cpus_intersects(*pmask, cpu_online_map)); |
| 288 | } |
| 289 | |
| 290 | /* |
| 291 | * Return in *pmask the portion of a cpusets's mems_allowed that |
Christoph Lameter | 0e1e7c7 | 2007-10-16 01:25:38 -0700 | [diff] [blame] | 292 | * are online, with memory. If none are online with memory, walk |
| 293 | * up the cpuset hierarchy until we find one that does have some |
| 294 | * online mems. If we get all the way to the top and still haven't |
| 295 | * found any online mems, return node_states[N_HIGH_MEMORY]. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 296 | * |
| 297 | * One way or another, we guarantee to return some non-empty subset |
Christoph Lameter | 0e1e7c7 | 2007-10-16 01:25:38 -0700 | [diff] [blame] | 298 | * of node_states[N_HIGH_MEMORY]. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 299 | * |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 300 | * Call with callback_mutex held. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 301 | */ |
| 302 | |
| 303 | static void guarantee_online_mems(const struct cpuset *cs, nodemask_t *pmask) |
| 304 | { |
Christoph Lameter | 0e1e7c7 | 2007-10-16 01:25:38 -0700 | [diff] [blame] | 305 | while (cs && !nodes_intersects(cs->mems_allowed, |
| 306 | node_states[N_HIGH_MEMORY])) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 307 | cs = cs->parent; |
| 308 | if (cs) |
Christoph Lameter | 0e1e7c7 | 2007-10-16 01:25:38 -0700 | [diff] [blame] | 309 | nodes_and(*pmask, cs->mems_allowed, |
| 310 | node_states[N_HIGH_MEMORY]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 311 | else |
Christoph Lameter | 0e1e7c7 | 2007-10-16 01:25:38 -0700 | [diff] [blame] | 312 | *pmask = node_states[N_HIGH_MEMORY]; |
| 313 | BUG_ON(!nodes_intersects(*pmask, node_states[N_HIGH_MEMORY])); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 314 | } |
| 315 | |
Paul Jackson | cf2a473c | 2006-01-08 01:01:54 -0800 | [diff] [blame] | 316 | /** |
| 317 | * cpuset_update_task_memory_state - update task memory placement |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 318 | * |
Paul Jackson | cf2a473c | 2006-01-08 01:01:54 -0800 | [diff] [blame] | 319 | * If the current tasks cpusets mems_allowed changed behind our |
| 320 | * backs, update current->mems_allowed, mems_generation and task NUMA |
| 321 | * mempolicy to the new value. |
| 322 | * |
| 323 | * Task mempolicy is updated by rebinding it relative to the |
| 324 | * current->cpuset if a task has its memory placement changed. |
| 325 | * Do not call this routine if in_interrupt(). |
| 326 | * |
Paul Jackson | 4a01c8d | 2006-03-31 02:30:50 -0800 | [diff] [blame] | 327 | * Call without callback_mutex or task_lock() held. May be |
Paul Menage | 2df167a | 2008-02-07 00:14:45 -0800 | [diff] [blame] | 328 | * called with or without cgroup_mutex held. Thanks in part to |
| 329 | * 'the_top_cpuset_hack', the task's cpuset pointer will never |
David Rientjes | 41f7f60 | 2008-03-04 23:32:38 -0800 | [diff] [blame] | 330 | * be NULL. This routine also might acquire callback_mutex during |
| 331 | * call. |
Paul Jackson | 5aa15b5 | 2005-10-30 15:02:28 -0800 | [diff] [blame] | 332 | * |
Paul Jackson | 6b9c260 | 2006-01-08 01:02:02 -0800 | [diff] [blame] | 333 | * Reading current->cpuset->mems_generation doesn't need task_lock |
| 334 | * to guard the current->cpuset derefence, because it is guarded |
Paul Menage | 2df167a | 2008-02-07 00:14:45 -0800 | [diff] [blame] | 335 | * from concurrent freeing of current->cpuset using RCU. |
Paul Jackson | 6b9c260 | 2006-01-08 01:02:02 -0800 | [diff] [blame] | 336 | * |
| 337 | * The rcu_dereference() is technically probably not needed, |
| 338 | * as I don't actually mind if I see a new cpuset pointer but |
| 339 | * an old value of mems_generation. However this really only |
| 340 | * matters on alpha systems using cpusets heavily. If I dropped |
| 341 | * that rcu_dereference(), it would save them a memory barrier. |
| 342 | * For all other arch's, rcu_dereference is a no-op anyway, and for |
| 343 | * alpha systems not using cpusets, another planned optimization, |
| 344 | * avoiding the rcu critical section for tasks in the root cpuset |
| 345 | * which is statically allocated, so can't vanish, will make this |
| 346 | * irrelevant. Better to use RCU as intended, than to engage in |
| 347 | * some cute trick to save a memory barrier that is impossible to |
| 348 | * test, for alpha systems using cpusets heavily, which might not |
| 349 | * even exist. |
Paul Jackson | 053199e | 2005-10-30 15:02:30 -0800 | [diff] [blame] | 350 | * |
| 351 | * This routine is needed to update the per-task mems_allowed data, |
| 352 | * within the tasks context, when it is trying to allocate memory |
| 353 | * (in various mm/mempolicy.c routines) and notices that some other |
| 354 | * task has been modifying its cpuset. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 355 | */ |
| 356 | |
Randy Dunlap | fe85a99 | 2006-02-03 03:04:23 -0800 | [diff] [blame] | 357 | void cpuset_update_task_memory_state(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 358 | { |
Paul Jackson | 053199e | 2005-10-30 15:02:30 -0800 | [diff] [blame] | 359 | int my_cpusets_mem_gen; |
Paul Jackson | cf2a473c | 2006-01-08 01:01:54 -0800 | [diff] [blame] | 360 | struct task_struct *tsk = current; |
Paul Jackson | 6b9c260 | 2006-01-08 01:02:02 -0800 | [diff] [blame] | 361 | struct cpuset *cs; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 362 | |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 363 | if (task_cs(tsk) == &top_cpuset) { |
Paul Jackson | 03a285f | 2006-01-08 01:02:04 -0800 | [diff] [blame] | 364 | /* Don't need rcu for top_cpuset. It's never freed. */ |
| 365 | my_cpusets_mem_gen = top_cpuset.mems_generation; |
| 366 | } else { |
| 367 | rcu_read_lock(); |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 368 | my_cpusets_mem_gen = task_cs(current)->mems_generation; |
Paul Jackson | 03a285f | 2006-01-08 01:02:04 -0800 | [diff] [blame] | 369 | rcu_read_unlock(); |
| 370 | } |
Paul Jackson | cf2a473c | 2006-01-08 01:01:54 -0800 | [diff] [blame] | 371 | |
| 372 | if (my_cpusets_mem_gen != tsk->cpuset_mems_generation) { |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 373 | mutex_lock(&callback_mutex); |
Paul Jackson | cf2a473c | 2006-01-08 01:01:54 -0800 | [diff] [blame] | 374 | task_lock(tsk); |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 375 | cs = task_cs(tsk); /* Maybe changed when task not locked */ |
Paul Jackson | cf2a473c | 2006-01-08 01:01:54 -0800 | [diff] [blame] | 376 | guarantee_online_mems(cs, &tsk->mems_allowed); |
| 377 | tsk->cpuset_mems_generation = cs->mems_generation; |
Paul Jackson | 825a46a | 2006-03-24 03:16:03 -0800 | [diff] [blame] | 378 | if (is_spread_page(cs)) |
| 379 | tsk->flags |= PF_SPREAD_PAGE; |
| 380 | else |
| 381 | tsk->flags &= ~PF_SPREAD_PAGE; |
| 382 | if (is_spread_slab(cs)) |
| 383 | tsk->flags |= PF_SPREAD_SLAB; |
| 384 | else |
| 385 | tsk->flags &= ~PF_SPREAD_SLAB; |
Paul Jackson | cf2a473c | 2006-01-08 01:01:54 -0800 | [diff] [blame] | 386 | task_unlock(tsk); |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 387 | mutex_unlock(&callback_mutex); |
Paul Jackson | 74cb215 | 2006-01-08 01:01:56 -0800 | [diff] [blame] | 388 | mpol_rebind_task(tsk, &tsk->mems_allowed); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 389 | } |
| 390 | } |
| 391 | |
| 392 | /* |
| 393 | * is_cpuset_subset(p, q) - Is cpuset p a subset of cpuset q? |
| 394 | * |
| 395 | * One cpuset is a subset of another if all its allowed CPUs and |
| 396 | * Memory Nodes are a subset of the other, and its exclusive flags |
Paul Menage | 2df167a | 2008-02-07 00:14:45 -0800 | [diff] [blame] | 397 | * are only set if the other's are set. Call holding cgroup_mutex. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 398 | */ |
| 399 | |
| 400 | static int is_cpuset_subset(const struct cpuset *p, const struct cpuset *q) |
| 401 | { |
| 402 | return cpus_subset(p->cpus_allowed, q->cpus_allowed) && |
| 403 | nodes_subset(p->mems_allowed, q->mems_allowed) && |
| 404 | is_cpu_exclusive(p) <= is_cpu_exclusive(q) && |
| 405 | is_mem_exclusive(p) <= is_mem_exclusive(q); |
| 406 | } |
| 407 | |
| 408 | /* |
| 409 | * validate_change() - Used to validate that any proposed cpuset change |
| 410 | * follows the structural rules for cpusets. |
| 411 | * |
| 412 | * If we replaced the flag and mask values of the current cpuset |
| 413 | * (cur) with those values in the trial cpuset (trial), would |
| 414 | * our various subset and exclusive rules still be valid? Presumes |
Paul Menage | 2df167a | 2008-02-07 00:14:45 -0800 | [diff] [blame] | 415 | * cgroup_mutex held. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 416 | * |
| 417 | * 'cur' is the address of an actual, in-use cpuset. Operations |
| 418 | * such as list traversal that depend on the actual address of the |
| 419 | * cpuset in the list must use cur below, not trial. |
| 420 | * |
| 421 | * 'trial' is the address of bulk structure copy of cur, with |
| 422 | * perhaps one or more of the fields cpus_allowed, mems_allowed, |
| 423 | * or flags changed to new, trial values. |
| 424 | * |
| 425 | * Return 0 if valid, -errno if not. |
| 426 | */ |
| 427 | |
| 428 | static int validate_change(const struct cpuset *cur, const struct cpuset *trial) |
| 429 | { |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 430 | struct cgroup *cont; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 431 | struct cpuset *c, *par; |
| 432 | |
| 433 | /* Each of our child cpusets must be a subset of us */ |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 434 | list_for_each_entry(cont, &cur->css.cgroup->children, sibling) { |
| 435 | if (!is_cpuset_subset(cgroup_cs(cont), trial)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 436 | return -EBUSY; |
| 437 | } |
| 438 | |
| 439 | /* Remaining checks don't apply to root cpuset */ |
Paul Jackson | 6960406 | 2006-12-06 20:36:15 -0800 | [diff] [blame] | 440 | if (cur == &top_cpuset) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 441 | return 0; |
| 442 | |
Paul Jackson | 6960406 | 2006-12-06 20:36:15 -0800 | [diff] [blame] | 443 | par = cur->parent; |
| 444 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 445 | /* We must be a subset of our parent cpuset */ |
| 446 | if (!is_cpuset_subset(trial, par)) |
| 447 | return -EACCES; |
| 448 | |
Paul Menage | 2df167a | 2008-02-07 00:14:45 -0800 | [diff] [blame] | 449 | /* |
| 450 | * If either I or some sibling (!= me) is exclusive, we can't |
| 451 | * overlap |
| 452 | */ |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 453 | list_for_each_entry(cont, &par->css.cgroup->children, sibling) { |
| 454 | c = cgroup_cs(cont); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 455 | if ((is_cpu_exclusive(trial) || is_cpu_exclusive(c)) && |
| 456 | c != cur && |
| 457 | cpus_intersects(trial->cpus_allowed, c->cpus_allowed)) |
| 458 | return -EINVAL; |
| 459 | if ((is_mem_exclusive(trial) || is_mem_exclusive(c)) && |
| 460 | c != cur && |
| 461 | nodes_intersects(trial->mems_allowed, c->mems_allowed)) |
| 462 | return -EINVAL; |
| 463 | } |
| 464 | |
Paul Jackson | 020958b | 2007-10-18 23:40:21 -0700 | [diff] [blame] | 465 | /* Cpusets with tasks can't have empty cpus_allowed or mems_allowed */ |
| 466 | if (cgroup_task_count(cur->css.cgroup)) { |
| 467 | if (cpus_empty(trial->cpus_allowed) || |
| 468 | nodes_empty(trial->mems_allowed)) { |
| 469 | return -ENOSPC; |
| 470 | } |
| 471 | } |
| 472 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 473 | return 0; |
| 474 | } |
| 475 | |
Dinakar Guniguntala | 85d7b94 | 2005-06-25 14:57:34 -0700 | [diff] [blame] | 476 | /* |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 477 | * Helper routine for rebuild_sched_domains(). |
| 478 | * Do cpusets a, b have overlapping cpus_allowed masks? |
| 479 | */ |
| 480 | |
| 481 | static int cpusets_overlap(struct cpuset *a, struct cpuset *b) |
| 482 | { |
| 483 | return cpus_intersects(a->cpus_allowed, b->cpus_allowed); |
| 484 | } |
| 485 | |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 486 | static void |
| 487 | update_domain_attr(struct sched_domain_attr *dattr, struct cpuset *c) |
| 488 | { |
| 489 | if (!dattr) |
| 490 | return; |
| 491 | if (dattr->relax_domain_level < c->relax_domain_level) |
| 492 | dattr->relax_domain_level = c->relax_domain_level; |
| 493 | return; |
| 494 | } |
| 495 | |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 496 | /* |
| 497 | * rebuild_sched_domains() |
| 498 | * |
Li Zefan | c372e81 | 2008-07-25 01:47:23 -0700 | [diff] [blame^] | 499 | * This routine will be called to rebuild the scheduler's dynamic |
| 500 | * sched domains: |
| 501 | * - if the flag 'sched_load_balance' of any cpuset with non-empty |
| 502 | * 'cpus' changes, |
| 503 | * - or if the 'cpus' allowed changes in any cpuset which has that |
| 504 | * flag enabled, |
| 505 | * - or if the 'sched_relax_domain_level' of any cpuset which has |
| 506 | * that flag enabled and with non-empty 'cpus' changes, |
| 507 | * - or if any cpuset with non-empty 'cpus' is removed, |
| 508 | * - or if a cpu gets offlined. |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 509 | * |
| 510 | * This routine builds a partial partition of the systems CPUs |
| 511 | * (the set of non-overlappping cpumask_t's in the array 'part' |
| 512 | * below), and passes that partial partition to the kernel/sched.c |
| 513 | * partition_sched_domains() routine, which will rebuild the |
| 514 | * schedulers load balancing domains (sched domains) as specified |
| 515 | * by that partial partition. A 'partial partition' is a set of |
| 516 | * non-overlapping subsets whose union is a subset of that set. |
| 517 | * |
| 518 | * See "What is sched_load_balance" in Documentation/cpusets.txt |
| 519 | * for a background explanation of this. |
| 520 | * |
| 521 | * Does not return errors, on the theory that the callers of this |
| 522 | * routine would rather not worry about failures to rebuild sched |
| 523 | * domains when operating in the severe memory shortage situations |
| 524 | * that could cause allocation failures below. |
| 525 | * |
| 526 | * Call with cgroup_mutex held. May take callback_mutex during |
| 527 | * call due to the kfifo_alloc() and kmalloc() calls. May nest |
Gautham R Shenoy | 86ef5c9 | 2008-01-25 21:08:02 +0100 | [diff] [blame] | 528 | * a call to the get_online_cpus()/put_online_cpus() pair. |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 529 | * Must not be called holding callback_mutex, because we must not |
Gautham R Shenoy | 86ef5c9 | 2008-01-25 21:08:02 +0100 | [diff] [blame] | 530 | * call get_online_cpus() while holding callback_mutex. Elsewhere |
| 531 | * the kernel nests callback_mutex inside get_online_cpus() calls. |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 532 | * So the reverse nesting would risk an ABBA deadlock. |
| 533 | * |
| 534 | * The three key local variables below are: |
| 535 | * q - a kfifo queue of cpuset pointers, used to implement a |
| 536 | * top-down scan of all cpusets. This scan loads a pointer |
| 537 | * to each cpuset marked is_sched_load_balance into the |
| 538 | * array 'csa'. For our purposes, rebuilding the schedulers |
| 539 | * sched domains, we can ignore !is_sched_load_balance cpusets. |
| 540 | * csa - (for CpuSet Array) Array of pointers to all the cpusets |
| 541 | * that need to be load balanced, for convenient iterative |
| 542 | * access by the subsequent code that finds the best partition, |
| 543 | * i.e the set of domains (subsets) of CPUs such that the |
| 544 | * cpus_allowed of every cpuset marked is_sched_load_balance |
| 545 | * is a subset of one of these domains, while there are as |
| 546 | * many such domains as possible, each as small as possible. |
| 547 | * doms - Conversion of 'csa' to an array of cpumasks, for passing to |
| 548 | * the kernel/sched.c routine partition_sched_domains() in a |
| 549 | * convenient format, that can be easily compared to the prior |
| 550 | * value to determine what partition elements (sched domains) |
| 551 | * were changed (added or removed.) |
| 552 | * |
| 553 | * Finding the best partition (set of domains): |
| 554 | * The triple nested loops below over i, j, k scan over the |
| 555 | * load balanced cpusets (using the array of cpuset pointers in |
| 556 | * csa[]) looking for pairs of cpusets that have overlapping |
| 557 | * cpus_allowed, but which don't have the same 'pn' partition |
| 558 | * number and gives them in the same partition number. It keeps |
| 559 | * looping on the 'restart' label until it can no longer find |
| 560 | * any such pairs. |
| 561 | * |
| 562 | * The union of the cpus_allowed masks from the set of |
| 563 | * all cpusets having the same 'pn' value then form the one |
| 564 | * element of the partition (one sched domain) to be passed to |
| 565 | * partition_sched_domains(). |
| 566 | */ |
| 567 | |
Max Krasnyansky | e761b77 | 2008-07-15 04:43:49 -0700 | [diff] [blame] | 568 | void rebuild_sched_domains(void) |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 569 | { |
| 570 | struct kfifo *q; /* queue of cpusets to be scanned */ |
| 571 | struct cpuset *cp; /* scans q */ |
| 572 | struct cpuset **csa; /* array of all cpuset ptrs */ |
| 573 | int csn; /* how many cpuset ptrs in csa so far */ |
| 574 | int i, j, k; /* indices for partition finding loops */ |
| 575 | cpumask_t *doms; /* resulting partition; i.e. sched domains */ |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 576 | struct sched_domain_attr *dattr; /* attributes for custom domains */ |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 577 | int ndoms; /* number of sched domains in result */ |
| 578 | int nslot; /* next empty doms[] cpumask_t slot */ |
| 579 | |
| 580 | q = NULL; |
| 581 | csa = NULL; |
| 582 | doms = NULL; |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 583 | dattr = NULL; |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 584 | |
| 585 | /* Special case for the 99% of systems with one, full, sched domain */ |
| 586 | if (is_sched_load_balance(&top_cpuset)) { |
| 587 | ndoms = 1; |
| 588 | doms = kmalloc(sizeof(cpumask_t), GFP_KERNEL); |
| 589 | if (!doms) |
| 590 | goto rebuild; |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 591 | dattr = kmalloc(sizeof(struct sched_domain_attr), GFP_KERNEL); |
| 592 | if (dattr) { |
| 593 | *dattr = SD_ATTR_INIT; |
| 594 | update_domain_attr(dattr, &top_cpuset); |
| 595 | } |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 596 | *doms = top_cpuset.cpus_allowed; |
| 597 | goto rebuild; |
| 598 | } |
| 599 | |
| 600 | q = kfifo_alloc(number_of_cpusets * sizeof(cp), GFP_KERNEL, NULL); |
| 601 | if (IS_ERR(q)) |
| 602 | goto done; |
| 603 | csa = kmalloc(number_of_cpusets * sizeof(cp), GFP_KERNEL); |
| 604 | if (!csa) |
| 605 | goto done; |
| 606 | csn = 0; |
| 607 | |
| 608 | cp = &top_cpuset; |
| 609 | __kfifo_put(q, (void *)&cp, sizeof(cp)); |
| 610 | while (__kfifo_get(q, (void *)&cp, sizeof(cp))) { |
| 611 | struct cgroup *cont; |
| 612 | struct cpuset *child; /* scans child cpusets of cp */ |
| 613 | if (is_sched_load_balance(cp)) |
| 614 | csa[csn++] = cp; |
| 615 | list_for_each_entry(cont, &cp->css.cgroup->children, sibling) { |
| 616 | child = cgroup_cs(cont); |
| 617 | __kfifo_put(q, (void *)&child, sizeof(cp)); |
| 618 | } |
| 619 | } |
| 620 | |
| 621 | for (i = 0; i < csn; i++) |
| 622 | csa[i]->pn = i; |
| 623 | ndoms = csn; |
| 624 | |
| 625 | restart: |
| 626 | /* Find the best partition (set of sched domains) */ |
| 627 | for (i = 0; i < csn; i++) { |
| 628 | struct cpuset *a = csa[i]; |
| 629 | int apn = a->pn; |
| 630 | |
| 631 | for (j = 0; j < csn; j++) { |
| 632 | struct cpuset *b = csa[j]; |
| 633 | int bpn = b->pn; |
| 634 | |
| 635 | if (apn != bpn && cpusets_overlap(a, b)) { |
| 636 | for (k = 0; k < csn; k++) { |
| 637 | struct cpuset *c = csa[k]; |
| 638 | |
| 639 | if (c->pn == bpn) |
| 640 | c->pn = apn; |
| 641 | } |
| 642 | ndoms--; /* one less element */ |
| 643 | goto restart; |
| 644 | } |
| 645 | } |
| 646 | } |
| 647 | |
| 648 | /* Convert <csn, csa> to <ndoms, doms> */ |
| 649 | doms = kmalloc(ndoms * sizeof(cpumask_t), GFP_KERNEL); |
| 650 | if (!doms) |
| 651 | goto rebuild; |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 652 | dattr = kmalloc(ndoms * sizeof(struct sched_domain_attr), GFP_KERNEL); |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 653 | |
| 654 | for (nslot = 0, i = 0; i < csn; i++) { |
| 655 | struct cpuset *a = csa[i]; |
| 656 | int apn = a->pn; |
| 657 | |
| 658 | if (apn >= 0) { |
| 659 | cpumask_t *dp = doms + nslot; |
| 660 | |
| 661 | if (nslot == ndoms) { |
| 662 | static int warnings = 10; |
| 663 | if (warnings) { |
| 664 | printk(KERN_WARNING |
| 665 | "rebuild_sched_domains confused:" |
| 666 | " nslot %d, ndoms %d, csn %d, i %d," |
| 667 | " apn %d\n", |
| 668 | nslot, ndoms, csn, i, apn); |
| 669 | warnings--; |
| 670 | } |
| 671 | continue; |
| 672 | } |
| 673 | |
| 674 | cpus_clear(*dp); |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 675 | if (dattr) |
| 676 | *(dattr + nslot) = SD_ATTR_INIT; |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 677 | for (j = i; j < csn; j++) { |
| 678 | struct cpuset *b = csa[j]; |
| 679 | |
| 680 | if (apn == b->pn) { |
| 681 | cpus_or(*dp, *dp, b->cpus_allowed); |
| 682 | b->pn = -1; |
Miao Xie | 91cd4d6 | 2008-07-21 14:21:35 -0700 | [diff] [blame] | 683 | if (dattr) |
| 684 | update_domain_attr(dattr |
| 685 | + nslot, b); |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 686 | } |
| 687 | } |
| 688 | nslot++; |
| 689 | } |
| 690 | } |
| 691 | BUG_ON(nslot != ndoms); |
| 692 | |
| 693 | rebuild: |
| 694 | /* Have scheduler rebuild sched domains */ |
Gautham R Shenoy | 86ef5c9 | 2008-01-25 21:08:02 +0100 | [diff] [blame] | 695 | get_online_cpus(); |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 696 | partition_sched_domains(ndoms, doms, dattr); |
Gautham R Shenoy | 86ef5c9 | 2008-01-25 21:08:02 +0100 | [diff] [blame] | 697 | put_online_cpus(); |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 698 | |
| 699 | done: |
| 700 | if (q && !IS_ERR(q)) |
| 701 | kfifo_free(q); |
| 702 | kfree(csa); |
| 703 | /* Don't kfree(doms) -- partition_sched_domains() does that. */ |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 704 | /* Don't kfree(dattr) -- partition_sched_domains() does that. */ |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 705 | } |
| 706 | |
Paul Menage | 8707d8b | 2007-10-18 23:40:22 -0700 | [diff] [blame] | 707 | static inline int started_after_time(struct task_struct *t1, |
| 708 | struct timespec *time, |
| 709 | struct task_struct *t2) |
| 710 | { |
| 711 | int start_diff = timespec_compare(&t1->start_time, time); |
| 712 | if (start_diff > 0) { |
| 713 | return 1; |
| 714 | } else if (start_diff < 0) { |
| 715 | return 0; |
| 716 | } else { |
| 717 | /* |
| 718 | * Arbitrarily, if two processes started at the same |
| 719 | * time, we'll say that the lower pointer value |
| 720 | * started first. Note that t2 may have exited by now |
| 721 | * so this may not be a valid pointer any longer, but |
| 722 | * that's fine - it still serves to distinguish |
| 723 | * between two tasks started (effectively) |
| 724 | * simultaneously. |
| 725 | */ |
| 726 | return t1 > t2; |
| 727 | } |
| 728 | } |
| 729 | |
| 730 | static inline int started_after(void *p1, void *p2) |
| 731 | { |
| 732 | struct task_struct *t1 = p1; |
| 733 | struct task_struct *t2 = p2; |
| 734 | return started_after_time(t1, &t2->start_time, t2); |
| 735 | } |
| 736 | |
Cliff Wickman | 58f4790 | 2008-02-07 00:14:44 -0800 | [diff] [blame] | 737 | /** |
| 738 | * cpuset_test_cpumask - test a task's cpus_allowed versus its cpuset's |
| 739 | * @tsk: task to test |
| 740 | * @scan: struct cgroup_scanner contained in its struct cpuset_hotplug_scanner |
| 741 | * |
Paul Menage | 2df167a | 2008-02-07 00:14:45 -0800 | [diff] [blame] | 742 | * Call with cgroup_mutex held. May take callback_mutex during call. |
Cliff Wickman | 58f4790 | 2008-02-07 00:14:44 -0800 | [diff] [blame] | 743 | * Called for each task in a cgroup by cgroup_scan_tasks(). |
| 744 | * Return nonzero if this tasks's cpus_allowed mask should be changed (in other |
| 745 | * words, if its mask is not equal to its cpuset's mask). |
Paul Jackson | 053199e | 2005-10-30 15:02:30 -0800 | [diff] [blame] | 746 | */ |
Adrian Bunk | 9e0c914 | 2008-04-29 01:00:25 -0700 | [diff] [blame] | 747 | static int cpuset_test_cpumask(struct task_struct *tsk, |
| 748 | struct cgroup_scanner *scan) |
Cliff Wickman | 58f4790 | 2008-02-07 00:14:44 -0800 | [diff] [blame] | 749 | { |
| 750 | return !cpus_equal(tsk->cpus_allowed, |
| 751 | (cgroup_cs(scan->cg))->cpus_allowed); |
| 752 | } |
Paul Jackson | 053199e | 2005-10-30 15:02:30 -0800 | [diff] [blame] | 753 | |
Cliff Wickman | 58f4790 | 2008-02-07 00:14:44 -0800 | [diff] [blame] | 754 | /** |
| 755 | * cpuset_change_cpumask - make a task's cpus_allowed the same as its cpuset's |
| 756 | * @tsk: task to test |
| 757 | * @scan: struct cgroup_scanner containing the cgroup of the task |
| 758 | * |
| 759 | * Called by cgroup_scan_tasks() for each task in a cgroup whose |
| 760 | * cpus_allowed mask needs to be changed. |
| 761 | * |
| 762 | * We don't need to re-check for the cgroup/cpuset membership, since we're |
| 763 | * holding cgroup_lock() at this point. |
| 764 | */ |
Adrian Bunk | 9e0c914 | 2008-04-29 01:00:25 -0700 | [diff] [blame] | 765 | static void cpuset_change_cpumask(struct task_struct *tsk, |
| 766 | struct cgroup_scanner *scan) |
Cliff Wickman | 58f4790 | 2008-02-07 00:14:44 -0800 | [diff] [blame] | 767 | { |
Mike Travis | f9a86fc | 2008-04-04 18:11:07 -0700 | [diff] [blame] | 768 | set_cpus_allowed_ptr(tsk, &((cgroup_cs(scan->cg))->cpus_allowed)); |
Cliff Wickman | 58f4790 | 2008-02-07 00:14:44 -0800 | [diff] [blame] | 769 | } |
| 770 | |
| 771 | /** |
Miao Xie | 0b2f630 | 2008-07-25 01:47:21 -0700 | [diff] [blame] | 772 | * update_tasks_cpumask - Update the cpumasks of tasks in the cpuset. |
| 773 | * @cs: the cpuset in which each task's cpus_allowed mask needs to be changed |
| 774 | * |
| 775 | * Called with cgroup_mutex held |
| 776 | * |
| 777 | * The cgroup_scan_tasks() function will scan all the tasks in a cgroup, |
| 778 | * calling callback functions for each. |
| 779 | * |
| 780 | * Return 0 if successful, -errno if not. |
| 781 | */ |
| 782 | static int update_tasks_cpumask(struct cpuset *cs) |
| 783 | { |
| 784 | struct cgroup_scanner scan; |
| 785 | struct ptr_heap heap; |
| 786 | int retval; |
| 787 | |
| 788 | retval = heap_init(&heap, PAGE_SIZE, GFP_KERNEL, &started_after); |
| 789 | if (retval) |
| 790 | return retval; |
| 791 | |
| 792 | scan.cg = cs->css.cgroup; |
| 793 | scan.test_task = cpuset_test_cpumask; |
| 794 | scan.process_task = cpuset_change_cpumask; |
| 795 | scan.heap = &heap; |
| 796 | retval = cgroup_scan_tasks(&scan); |
| 797 | |
| 798 | heap_free(&heap); |
| 799 | return retval; |
| 800 | } |
| 801 | |
| 802 | /** |
Cliff Wickman | 58f4790 | 2008-02-07 00:14:44 -0800 | [diff] [blame] | 803 | * update_cpumask - update the cpus_allowed mask of a cpuset and all tasks in it |
| 804 | * @cs: the cpuset to consider |
| 805 | * @buf: buffer of cpu numbers written to this cpuset |
| 806 | */ |
Paul Menage | e371239 | 2008-07-25 01:47:02 -0700 | [diff] [blame] | 807 | static int update_cpumask(struct cpuset *cs, const char *buf) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 808 | { |
| 809 | struct cpuset trialcs; |
Cliff Wickman | 58f4790 | 2008-02-07 00:14:44 -0800 | [diff] [blame] | 810 | int retval; |
| 811 | int is_load_balanced; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 812 | |
Paul Jackson | 4c4d50f | 2006-08-27 01:23:51 -0700 | [diff] [blame] | 813 | /* top_cpuset.cpus_allowed tracks cpu_online_map; it's read-only */ |
| 814 | if (cs == &top_cpuset) |
| 815 | return -EACCES; |
| 816 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 817 | trialcs = *cs; |
David Rientjes | 6f7f02e | 2007-05-08 00:31:43 -0700 | [diff] [blame] | 818 | |
| 819 | /* |
Paul Jackson | c8d9c90 | 2008-02-07 00:14:46 -0800 | [diff] [blame] | 820 | * An empty cpus_allowed is ok only if the cpuset has no tasks. |
Paul Jackson | 020958b | 2007-10-18 23:40:21 -0700 | [diff] [blame] | 821 | * Since cpulist_parse() fails on an empty mask, we special case |
| 822 | * that parsing. The validate_change() call ensures that cpusets |
| 823 | * with tasks have cpus. |
David Rientjes | 6f7f02e | 2007-05-08 00:31:43 -0700 | [diff] [blame] | 824 | */ |
Paul Jackson | 020958b | 2007-10-18 23:40:21 -0700 | [diff] [blame] | 825 | if (!*buf) { |
David Rientjes | 6f7f02e | 2007-05-08 00:31:43 -0700 | [diff] [blame] | 826 | cpus_clear(trialcs.cpus_allowed); |
| 827 | } else { |
| 828 | retval = cpulist_parse(buf, trialcs.cpus_allowed); |
| 829 | if (retval < 0) |
| 830 | return retval; |
Lai Jiangshan | 3734074 | 2008-06-05 22:46:32 -0700 | [diff] [blame] | 831 | |
| 832 | if (!cpus_subset(trialcs.cpus_allowed, cpu_online_map)) |
| 833 | return -EINVAL; |
David Rientjes | 6f7f02e | 2007-05-08 00:31:43 -0700 | [diff] [blame] | 834 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 835 | retval = validate_change(cs, &trialcs); |
Dinakar Guniguntala | 85d7b94 | 2005-06-25 14:57:34 -0700 | [diff] [blame] | 836 | if (retval < 0) |
| 837 | return retval; |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 838 | |
Paul Menage | 8707d8b | 2007-10-18 23:40:22 -0700 | [diff] [blame] | 839 | /* Nothing to do if the cpus didn't change */ |
| 840 | if (cpus_equal(cs->cpus_allowed, trialcs.cpus_allowed)) |
| 841 | return 0; |
Cliff Wickman | 58f4790 | 2008-02-07 00:14:44 -0800 | [diff] [blame] | 842 | |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 843 | is_load_balanced = is_sched_load_balance(&trialcs); |
| 844 | |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 845 | mutex_lock(&callback_mutex); |
Dinakar Guniguntala | 85d7b94 | 2005-06-25 14:57:34 -0700 | [diff] [blame] | 846 | cs->cpus_allowed = trialcs.cpus_allowed; |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 847 | mutex_unlock(&callback_mutex); |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 848 | |
Paul Menage | 8707d8b | 2007-10-18 23:40:22 -0700 | [diff] [blame] | 849 | /* |
| 850 | * Scan tasks in the cpuset, and update the cpumasks of any |
Cliff Wickman | 58f4790 | 2008-02-07 00:14:44 -0800 | [diff] [blame] | 851 | * that need an update. |
Paul Menage | 8707d8b | 2007-10-18 23:40:22 -0700 | [diff] [blame] | 852 | */ |
Miao Xie | 0b2f630 | 2008-07-25 01:47:21 -0700 | [diff] [blame] | 853 | retval = update_tasks_cpumask(cs); |
| 854 | if (retval < 0) |
| 855 | return retval; |
Cliff Wickman | 58f4790 | 2008-02-07 00:14:44 -0800 | [diff] [blame] | 856 | |
Paul Menage | 8707d8b | 2007-10-18 23:40:22 -0700 | [diff] [blame] | 857 | if (is_load_balanced) |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 858 | rebuild_sched_domains(); |
Dinakar Guniguntala | 85d7b94 | 2005-06-25 14:57:34 -0700 | [diff] [blame] | 859 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 860 | } |
| 861 | |
Paul Jackson | 053199e | 2005-10-30 15:02:30 -0800 | [diff] [blame] | 862 | /* |
Paul Jackson | e4e364e | 2006-03-31 02:30:52 -0800 | [diff] [blame] | 863 | * cpuset_migrate_mm |
| 864 | * |
| 865 | * Migrate memory region from one set of nodes to another. |
| 866 | * |
| 867 | * Temporarilly set tasks mems_allowed to target nodes of migration, |
| 868 | * so that the migration code can allocate pages on these nodes. |
| 869 | * |
Paul Menage | 2df167a | 2008-02-07 00:14:45 -0800 | [diff] [blame] | 870 | * Call holding cgroup_mutex, so current's cpuset won't change |
Paul Jackson | c8d9c90 | 2008-02-07 00:14:46 -0800 | [diff] [blame] | 871 | * during this call, as manage_mutex holds off any cpuset_attach() |
Paul Jackson | e4e364e | 2006-03-31 02:30:52 -0800 | [diff] [blame] | 872 | * calls. Therefore we don't need to take task_lock around the |
| 873 | * call to guarantee_online_mems(), as we know no one is changing |
Paul Menage | 2df167a | 2008-02-07 00:14:45 -0800 | [diff] [blame] | 874 | * our task's cpuset. |
Paul Jackson | e4e364e | 2006-03-31 02:30:52 -0800 | [diff] [blame] | 875 | * |
| 876 | * Hold callback_mutex around the two modifications of our tasks |
| 877 | * mems_allowed to synchronize with cpuset_mems_allowed(). |
| 878 | * |
| 879 | * While the mm_struct we are migrating is typically from some |
| 880 | * other task, the task_struct mems_allowed that we are hacking |
| 881 | * is for our current task, which must allocate new pages for that |
| 882 | * migrating memory region. |
| 883 | * |
| 884 | * We call cpuset_update_task_memory_state() before hacking |
| 885 | * our tasks mems_allowed, so that we are assured of being in |
| 886 | * sync with our tasks cpuset, and in particular, callbacks to |
| 887 | * cpuset_update_task_memory_state() from nested page allocations |
| 888 | * won't see any mismatch of our cpuset and task mems_generation |
| 889 | * values, so won't overwrite our hacked tasks mems_allowed |
| 890 | * nodemask. |
| 891 | */ |
| 892 | |
| 893 | static void cpuset_migrate_mm(struct mm_struct *mm, const nodemask_t *from, |
| 894 | const nodemask_t *to) |
| 895 | { |
| 896 | struct task_struct *tsk = current; |
| 897 | |
| 898 | cpuset_update_task_memory_state(); |
| 899 | |
| 900 | mutex_lock(&callback_mutex); |
| 901 | tsk->mems_allowed = *to; |
| 902 | mutex_unlock(&callback_mutex); |
| 903 | |
| 904 | do_migrate_pages(mm, from, to, MPOL_MF_MOVE_ALL); |
| 905 | |
| 906 | mutex_lock(&callback_mutex); |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 907 | guarantee_online_mems(task_cs(tsk),&tsk->mems_allowed); |
Paul Jackson | e4e364e | 2006-03-31 02:30:52 -0800 | [diff] [blame] | 908 | mutex_unlock(&callback_mutex); |
| 909 | } |
| 910 | |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 911 | static void *cpuset_being_rebound; |
| 912 | |
Miao Xie | 0b2f630 | 2008-07-25 01:47:21 -0700 | [diff] [blame] | 913 | /** |
| 914 | * update_tasks_nodemask - Update the nodemasks of tasks in the cpuset. |
| 915 | * @cs: the cpuset in which each task's mems_allowed mask needs to be changed |
| 916 | * @oldmem: old mems_allowed of cpuset cs |
| 917 | * |
| 918 | * Called with cgroup_mutex held |
| 919 | * Return 0 if successful, -errno if not. |
| 920 | */ |
| 921 | static int update_tasks_nodemask(struct cpuset *cs, const nodemask_t *oldmem) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 922 | { |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 923 | struct task_struct *p; |
Paul Jackson | 4225399 | 2006-01-08 01:01:59 -0800 | [diff] [blame] | 924 | struct mm_struct **mmarray; |
| 925 | int i, n, ntasks; |
Paul Jackson | 04c19fa | 2006-01-08 01:02:00 -0800 | [diff] [blame] | 926 | int migrate; |
Paul Jackson | 4225399 | 2006-01-08 01:01:59 -0800 | [diff] [blame] | 927 | int fudge; |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 928 | struct cgroup_iter it; |
Miao Xie | 0b2f630 | 2008-07-25 01:47:21 -0700 | [diff] [blame] | 929 | int retval; |
Paul Jackson | 59dac16 | 2006-01-08 01:01:52 -0800 | [diff] [blame] | 930 | |
Lee Schermerhorn | 846a16b | 2008-04-28 02:13:09 -0700 | [diff] [blame] | 931 | cpuset_being_rebound = cs; /* causes mpol_dup() rebind */ |
Paul Jackson | 4225399 | 2006-01-08 01:01:59 -0800 | [diff] [blame] | 932 | |
| 933 | fudge = 10; /* spare mmarray[] slots */ |
| 934 | fudge += cpus_weight(cs->cpus_allowed); /* imagine one fork-bomb/cpu */ |
| 935 | retval = -ENOMEM; |
| 936 | |
| 937 | /* |
| 938 | * Allocate mmarray[] to hold mm reference for each task |
| 939 | * in cpuset cs. Can't kmalloc GFP_KERNEL while holding |
| 940 | * tasklist_lock. We could use GFP_ATOMIC, but with a |
| 941 | * few more lines of code, we can retry until we get a big |
| 942 | * enough mmarray[] w/o using GFP_ATOMIC. |
| 943 | */ |
| 944 | while (1) { |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 945 | ntasks = cgroup_task_count(cs->css.cgroup); /* guess */ |
Paul Jackson | 4225399 | 2006-01-08 01:01:59 -0800 | [diff] [blame] | 946 | ntasks += fudge; |
| 947 | mmarray = kmalloc(ntasks * sizeof(*mmarray), GFP_KERNEL); |
| 948 | if (!mmarray) |
| 949 | goto done; |
Paul Menage | c2aef33 | 2007-07-15 23:40:11 -0700 | [diff] [blame] | 950 | read_lock(&tasklist_lock); /* block fork */ |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 951 | if (cgroup_task_count(cs->css.cgroup) <= ntasks) |
Paul Jackson | 4225399 | 2006-01-08 01:01:59 -0800 | [diff] [blame] | 952 | break; /* got enough */ |
Paul Menage | c2aef33 | 2007-07-15 23:40:11 -0700 | [diff] [blame] | 953 | read_unlock(&tasklist_lock); /* try again */ |
Paul Jackson | 4225399 | 2006-01-08 01:01:59 -0800 | [diff] [blame] | 954 | kfree(mmarray); |
| 955 | } |
| 956 | |
| 957 | n = 0; |
| 958 | |
| 959 | /* Load up mmarray[] with mm reference for each task in cpuset. */ |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 960 | cgroup_iter_start(cs->css.cgroup, &it); |
| 961 | while ((p = cgroup_iter_next(cs->css.cgroup, &it))) { |
Paul Jackson | 4225399 | 2006-01-08 01:01:59 -0800 | [diff] [blame] | 962 | struct mm_struct *mm; |
| 963 | |
| 964 | if (n >= ntasks) { |
| 965 | printk(KERN_WARNING |
| 966 | "Cpuset mempolicy rebind incomplete.\n"); |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 967 | break; |
Paul Jackson | 4225399 | 2006-01-08 01:01:59 -0800 | [diff] [blame] | 968 | } |
Paul Jackson | 4225399 | 2006-01-08 01:01:59 -0800 | [diff] [blame] | 969 | mm = get_task_mm(p); |
| 970 | if (!mm) |
| 971 | continue; |
| 972 | mmarray[n++] = mm; |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 973 | } |
| 974 | cgroup_iter_end(cs->css.cgroup, &it); |
Paul Menage | c2aef33 | 2007-07-15 23:40:11 -0700 | [diff] [blame] | 975 | read_unlock(&tasklist_lock); |
Paul Jackson | 4225399 | 2006-01-08 01:01:59 -0800 | [diff] [blame] | 976 | |
| 977 | /* |
| 978 | * Now that we've dropped the tasklist spinlock, we can |
| 979 | * rebind the vma mempolicies of each mm in mmarray[] to their |
| 980 | * new cpuset, and release that mm. The mpol_rebind_mm() |
| 981 | * call takes mmap_sem, which we couldn't take while holding |
Lee Schermerhorn | 846a16b | 2008-04-28 02:13:09 -0700 | [diff] [blame] | 982 | * tasklist_lock. Forks can happen again now - the mpol_dup() |
Paul Jackson | 4225399 | 2006-01-08 01:01:59 -0800 | [diff] [blame] | 983 | * cpuset_being_rebound check will catch such forks, and rebind |
| 984 | * their vma mempolicies too. Because we still hold the global |
Paul Menage | 2df167a | 2008-02-07 00:14:45 -0800 | [diff] [blame] | 985 | * cgroup_mutex, we know that no other rebind effort will |
Paul Jackson | 4225399 | 2006-01-08 01:01:59 -0800 | [diff] [blame] | 986 | * be contending for the global variable cpuset_being_rebound. |
| 987 | * It's ok if we rebind the same mm twice; mpol_rebind_mm() |
Paul Jackson | 04c19fa | 2006-01-08 01:02:00 -0800 | [diff] [blame] | 988 | * is idempotent. Also migrate pages in each mm to new nodes. |
Paul Jackson | 4225399 | 2006-01-08 01:01:59 -0800 | [diff] [blame] | 989 | */ |
Paul Jackson | 04c19fa | 2006-01-08 01:02:00 -0800 | [diff] [blame] | 990 | migrate = is_memory_migrate(cs); |
Paul Jackson | 4225399 | 2006-01-08 01:01:59 -0800 | [diff] [blame] | 991 | for (i = 0; i < n; i++) { |
| 992 | struct mm_struct *mm = mmarray[i]; |
| 993 | |
| 994 | mpol_rebind_mm(mm, &cs->mems_allowed); |
Paul Jackson | e4e364e | 2006-03-31 02:30:52 -0800 | [diff] [blame] | 995 | if (migrate) |
Miao Xie | 0b2f630 | 2008-07-25 01:47:21 -0700 | [diff] [blame] | 996 | cpuset_migrate_mm(mm, oldmem, &cs->mems_allowed); |
Paul Jackson | 4225399 | 2006-01-08 01:01:59 -0800 | [diff] [blame] | 997 | mmput(mm); |
| 998 | } |
| 999 | |
Paul Menage | 2df167a | 2008-02-07 00:14:45 -0800 | [diff] [blame] | 1000 | /* We're done rebinding vmas to this cpuset's new mems_allowed. */ |
Paul Jackson | 4225399 | 2006-01-08 01:01:59 -0800 | [diff] [blame] | 1001 | kfree(mmarray); |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1002 | cpuset_being_rebound = NULL; |
Paul Jackson | 4225399 | 2006-01-08 01:01:59 -0800 | [diff] [blame] | 1003 | retval = 0; |
Paul Jackson | 59dac16 | 2006-01-08 01:01:52 -0800 | [diff] [blame] | 1004 | done: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1005 | return retval; |
| 1006 | } |
| 1007 | |
Miao Xie | 0b2f630 | 2008-07-25 01:47:21 -0700 | [diff] [blame] | 1008 | /* |
| 1009 | * Handle user request to change the 'mems' memory placement |
| 1010 | * of a cpuset. Needs to validate the request, update the |
| 1011 | * cpusets mems_allowed and mems_generation, and for each |
| 1012 | * task in the cpuset, rebind any vma mempolicies and if |
| 1013 | * the cpuset is marked 'memory_migrate', migrate the tasks |
| 1014 | * pages to the new memory. |
| 1015 | * |
| 1016 | * Call with cgroup_mutex held. May take callback_mutex during call. |
| 1017 | * Will take tasklist_lock, scan tasklist for tasks in cpuset cs, |
| 1018 | * lock each such tasks mm->mmap_sem, scan its vma's and rebind |
| 1019 | * their mempolicies to the cpusets new mems_allowed. |
| 1020 | */ |
| 1021 | static int update_nodemask(struct cpuset *cs, const char *buf) |
| 1022 | { |
| 1023 | struct cpuset trialcs; |
| 1024 | nodemask_t oldmem; |
| 1025 | int retval; |
| 1026 | |
| 1027 | /* |
| 1028 | * top_cpuset.mems_allowed tracks node_stats[N_HIGH_MEMORY]; |
| 1029 | * it's read-only |
| 1030 | */ |
| 1031 | if (cs == &top_cpuset) |
| 1032 | return -EACCES; |
| 1033 | |
| 1034 | trialcs = *cs; |
| 1035 | |
| 1036 | /* |
| 1037 | * An empty mems_allowed is ok iff there are no tasks in the cpuset. |
| 1038 | * Since nodelist_parse() fails on an empty mask, we special case |
| 1039 | * that parsing. The validate_change() call ensures that cpusets |
| 1040 | * with tasks have memory. |
| 1041 | */ |
| 1042 | if (!*buf) { |
| 1043 | nodes_clear(trialcs.mems_allowed); |
| 1044 | } else { |
| 1045 | retval = nodelist_parse(buf, trialcs.mems_allowed); |
| 1046 | if (retval < 0) |
| 1047 | goto done; |
| 1048 | |
| 1049 | if (!nodes_subset(trialcs.mems_allowed, |
| 1050 | node_states[N_HIGH_MEMORY])) |
| 1051 | return -EINVAL; |
| 1052 | } |
| 1053 | oldmem = cs->mems_allowed; |
| 1054 | if (nodes_equal(oldmem, trialcs.mems_allowed)) { |
| 1055 | retval = 0; /* Too easy - nothing to do */ |
| 1056 | goto done; |
| 1057 | } |
| 1058 | retval = validate_change(cs, &trialcs); |
| 1059 | if (retval < 0) |
| 1060 | goto done; |
| 1061 | |
| 1062 | mutex_lock(&callback_mutex); |
| 1063 | cs->mems_allowed = trialcs.mems_allowed; |
| 1064 | cs->mems_generation = cpuset_mems_generation++; |
| 1065 | mutex_unlock(&callback_mutex); |
| 1066 | |
| 1067 | retval = update_tasks_nodemask(cs, &oldmem); |
| 1068 | done: |
| 1069 | return retval; |
| 1070 | } |
| 1071 | |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1072 | int current_cpuset_is_being_rebound(void) |
| 1073 | { |
| 1074 | return task_cs(current) == cpuset_being_rebound; |
| 1075 | } |
| 1076 | |
Paul Menage | 5be7a47 | 2008-05-06 20:42:41 -0700 | [diff] [blame] | 1077 | static int update_relax_domain_level(struct cpuset *cs, s64 val) |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 1078 | { |
Li Zefan | 30e0e17 | 2008-05-13 10:27:17 +0800 | [diff] [blame] | 1079 | if (val < -1 || val >= SD_LV_MAX) |
| 1080 | return -EINVAL; |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 1081 | |
| 1082 | if (val != cs->relax_domain_level) { |
| 1083 | cs->relax_domain_level = val; |
Li Zefan | c372e81 | 2008-07-25 01:47:23 -0700 | [diff] [blame^] | 1084 | if (!cpus_empty(cs->cpus_allowed) && is_sched_load_balance(cs)) |
| 1085 | rebuild_sched_domains(); |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 1086 | } |
| 1087 | |
| 1088 | return 0; |
| 1089 | } |
| 1090 | |
Paul Jackson | 3e0d98b | 2006-01-08 01:01:49 -0800 | [diff] [blame] | 1091 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1092 | * update_flag - read a 0 or a 1 in a file and update associated flag |
Paul Menage | 7860836 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 1093 | * bit: the bit to update (see cpuset_flagbits_t) |
| 1094 | * cs: the cpuset to update |
| 1095 | * turning_on: whether the flag is being set or cleared |
Paul Jackson | 053199e | 2005-10-30 15:02:30 -0800 | [diff] [blame] | 1096 | * |
Paul Menage | 2df167a | 2008-02-07 00:14:45 -0800 | [diff] [blame] | 1097 | * Call with cgroup_mutex held. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1098 | */ |
| 1099 | |
Paul Menage | 700fe1a | 2008-04-29 01:00:00 -0700 | [diff] [blame] | 1100 | static int update_flag(cpuset_flagbits_t bit, struct cpuset *cs, |
| 1101 | int turning_on) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1102 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1103 | struct cpuset trialcs; |
Paul Jackson | 607717a | 2007-10-16 01:27:43 -0700 | [diff] [blame] | 1104 | int err; |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 1105 | int cpus_nonempty, balance_flag_changed; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1106 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1107 | trialcs = *cs; |
| 1108 | if (turning_on) |
| 1109 | set_bit(bit, &trialcs.flags); |
| 1110 | else |
| 1111 | clear_bit(bit, &trialcs.flags); |
| 1112 | |
| 1113 | err = validate_change(cs, &trialcs); |
Dinakar Guniguntala | 85d7b94 | 2005-06-25 14:57:34 -0700 | [diff] [blame] | 1114 | if (err < 0) |
| 1115 | return err; |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 1116 | |
| 1117 | cpus_nonempty = !cpus_empty(trialcs.cpus_allowed); |
| 1118 | balance_flag_changed = (is_sched_load_balance(cs) != |
| 1119 | is_sched_load_balance(&trialcs)); |
| 1120 | |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 1121 | mutex_lock(&callback_mutex); |
Paul Jackson | 6960406 | 2006-12-06 20:36:15 -0800 | [diff] [blame] | 1122 | cs->flags = trialcs.flags; |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 1123 | mutex_unlock(&callback_mutex); |
Dinakar Guniguntala | 85d7b94 | 2005-06-25 14:57:34 -0700 | [diff] [blame] | 1124 | |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 1125 | if (cpus_nonempty && balance_flag_changed) |
| 1126 | rebuild_sched_domains(); |
| 1127 | |
Dinakar Guniguntala | 85d7b94 | 2005-06-25 14:57:34 -0700 | [diff] [blame] | 1128 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1129 | } |
| 1130 | |
Paul Jackson | 053199e | 2005-10-30 15:02:30 -0800 | [diff] [blame] | 1131 | /* |
Adrian Bunk | 80f7228 | 2006-06-30 18:27:16 +0200 | [diff] [blame] | 1132 | * Frequency meter - How fast is some event occurring? |
Paul Jackson | 3e0d98b | 2006-01-08 01:01:49 -0800 | [diff] [blame] | 1133 | * |
| 1134 | * These routines manage a digitally filtered, constant time based, |
| 1135 | * event frequency meter. There are four routines: |
| 1136 | * fmeter_init() - initialize a frequency meter. |
| 1137 | * fmeter_markevent() - called each time the event happens. |
| 1138 | * fmeter_getrate() - returns the recent rate of such events. |
| 1139 | * fmeter_update() - internal routine used to update fmeter. |
| 1140 | * |
| 1141 | * A common data structure is passed to each of these routines, |
| 1142 | * which is used to keep track of the state required to manage the |
| 1143 | * frequency meter and its digital filter. |
| 1144 | * |
| 1145 | * The filter works on the number of events marked per unit time. |
| 1146 | * The filter is single-pole low-pass recursive (IIR). The time unit |
| 1147 | * is 1 second. Arithmetic is done using 32-bit integers scaled to |
| 1148 | * simulate 3 decimal digits of precision (multiplied by 1000). |
| 1149 | * |
| 1150 | * With an FM_COEF of 933, and a time base of 1 second, the filter |
| 1151 | * has a half-life of 10 seconds, meaning that if the events quit |
| 1152 | * happening, then the rate returned from the fmeter_getrate() |
| 1153 | * will be cut in half each 10 seconds, until it converges to zero. |
| 1154 | * |
| 1155 | * It is not worth doing a real infinitely recursive filter. If more |
| 1156 | * than FM_MAXTICKS ticks have elapsed since the last filter event, |
| 1157 | * just compute FM_MAXTICKS ticks worth, by which point the level |
| 1158 | * will be stable. |
| 1159 | * |
| 1160 | * Limit the count of unprocessed events to FM_MAXCNT, so as to avoid |
| 1161 | * arithmetic overflow in the fmeter_update() routine. |
| 1162 | * |
| 1163 | * Given the simple 32 bit integer arithmetic used, this meter works |
| 1164 | * best for reporting rates between one per millisecond (msec) and |
| 1165 | * one per 32 (approx) seconds. At constant rates faster than one |
| 1166 | * per msec it maxes out at values just under 1,000,000. At constant |
| 1167 | * rates between one per msec, and one per second it will stabilize |
| 1168 | * to a value N*1000, where N is the rate of events per second. |
| 1169 | * At constant rates between one per second and one per 32 seconds, |
| 1170 | * it will be choppy, moving up on the seconds that have an event, |
| 1171 | * and then decaying until the next event. At rates slower than |
| 1172 | * about one in 32 seconds, it decays all the way back to zero between |
| 1173 | * each event. |
| 1174 | */ |
| 1175 | |
| 1176 | #define FM_COEF 933 /* coefficient for half-life of 10 secs */ |
| 1177 | #define FM_MAXTICKS ((time_t)99) /* useless computing more ticks than this */ |
| 1178 | #define FM_MAXCNT 1000000 /* limit cnt to avoid overflow */ |
| 1179 | #define FM_SCALE 1000 /* faux fixed point scale */ |
| 1180 | |
| 1181 | /* Initialize a frequency meter */ |
| 1182 | static void fmeter_init(struct fmeter *fmp) |
| 1183 | { |
| 1184 | fmp->cnt = 0; |
| 1185 | fmp->val = 0; |
| 1186 | fmp->time = 0; |
| 1187 | spin_lock_init(&fmp->lock); |
| 1188 | } |
| 1189 | |
| 1190 | /* Internal meter update - process cnt events and update value */ |
| 1191 | static void fmeter_update(struct fmeter *fmp) |
| 1192 | { |
| 1193 | time_t now = get_seconds(); |
| 1194 | time_t ticks = now - fmp->time; |
| 1195 | |
| 1196 | if (ticks == 0) |
| 1197 | return; |
| 1198 | |
| 1199 | ticks = min(FM_MAXTICKS, ticks); |
| 1200 | while (ticks-- > 0) |
| 1201 | fmp->val = (FM_COEF * fmp->val) / FM_SCALE; |
| 1202 | fmp->time = now; |
| 1203 | |
| 1204 | fmp->val += ((FM_SCALE - FM_COEF) * fmp->cnt) / FM_SCALE; |
| 1205 | fmp->cnt = 0; |
| 1206 | } |
| 1207 | |
| 1208 | /* Process any previous ticks, then bump cnt by one (times scale). */ |
| 1209 | static void fmeter_markevent(struct fmeter *fmp) |
| 1210 | { |
| 1211 | spin_lock(&fmp->lock); |
| 1212 | fmeter_update(fmp); |
| 1213 | fmp->cnt = min(FM_MAXCNT, fmp->cnt + FM_SCALE); |
| 1214 | spin_unlock(&fmp->lock); |
| 1215 | } |
| 1216 | |
| 1217 | /* Process any previous ticks, then return current value. */ |
| 1218 | static int fmeter_getrate(struct fmeter *fmp) |
| 1219 | { |
| 1220 | int val; |
| 1221 | |
| 1222 | spin_lock(&fmp->lock); |
| 1223 | fmeter_update(fmp); |
| 1224 | val = fmp->val; |
| 1225 | spin_unlock(&fmp->lock); |
| 1226 | return val; |
| 1227 | } |
| 1228 | |
Paul Menage | 2df167a | 2008-02-07 00:14:45 -0800 | [diff] [blame] | 1229 | /* Called by cgroups to determine if a cpuset is usable; cgroup_mutex held */ |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1230 | static int cpuset_can_attach(struct cgroup_subsys *ss, |
| 1231 | struct cgroup *cont, struct task_struct *tsk) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1232 | { |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1233 | struct cpuset *cs = cgroup_cs(cont); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1234 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1235 | if (cpus_empty(cs->cpus_allowed) || nodes_empty(cs->mems_allowed)) |
| 1236 | return -ENOSPC; |
David Rientjes | 9985b0b | 2008-06-05 12:57:11 -0700 | [diff] [blame] | 1237 | if (tsk->flags & PF_THREAD_BOUND) { |
| 1238 | cpumask_t mask; |
| 1239 | |
| 1240 | mutex_lock(&callback_mutex); |
| 1241 | mask = cs->cpus_allowed; |
| 1242 | mutex_unlock(&callback_mutex); |
| 1243 | if (!cpus_equal(tsk->cpus_allowed, mask)) |
| 1244 | return -EINVAL; |
| 1245 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1246 | |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1247 | return security_task_setscheduler(tsk, 0, NULL); |
| 1248 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1249 | |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1250 | static void cpuset_attach(struct cgroup_subsys *ss, |
| 1251 | struct cgroup *cont, struct cgroup *oldcont, |
| 1252 | struct task_struct *tsk) |
| 1253 | { |
| 1254 | cpumask_t cpus; |
| 1255 | nodemask_t from, to; |
| 1256 | struct mm_struct *mm; |
| 1257 | struct cpuset *cs = cgroup_cs(cont); |
| 1258 | struct cpuset *oldcs = cgroup_cs(oldcont); |
David Rientjes | 9985b0b | 2008-06-05 12:57:11 -0700 | [diff] [blame] | 1259 | int err; |
David Quigley | 22fb52d | 2006-06-23 02:04:00 -0700 | [diff] [blame] | 1260 | |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 1261 | mutex_lock(&callback_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1262 | guarantee_online_cpus(cs, &cpus); |
David Rientjes | 9985b0b | 2008-06-05 12:57:11 -0700 | [diff] [blame] | 1263 | err = set_cpus_allowed_ptr(tsk, &cpus); |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1264 | mutex_unlock(&callback_mutex); |
David Rientjes | 9985b0b | 2008-06-05 12:57:11 -0700 | [diff] [blame] | 1265 | if (err) |
| 1266 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1267 | |
Paul Jackson | 45b07ef | 2006-01-08 01:00:56 -0800 | [diff] [blame] | 1268 | from = oldcs->mems_allowed; |
| 1269 | to = cs->mems_allowed; |
Paul Jackson | 4225399 | 2006-01-08 01:01:59 -0800 | [diff] [blame] | 1270 | mm = get_task_mm(tsk); |
| 1271 | if (mm) { |
| 1272 | mpol_rebind_mm(mm, &to); |
Paul Jackson | 2741a55 | 2006-03-31 02:30:51 -0800 | [diff] [blame] | 1273 | if (is_memory_migrate(cs)) |
Paul Jackson | e4e364e | 2006-03-31 02:30:52 -0800 | [diff] [blame] | 1274 | cpuset_migrate_mm(mm, &from, &to); |
Paul Jackson | 4225399 | 2006-01-08 01:01:59 -0800 | [diff] [blame] | 1275 | mmput(mm); |
| 1276 | } |
| 1277 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1278 | } |
| 1279 | |
| 1280 | /* The various types of files and directories in a cpuset file system */ |
| 1281 | |
| 1282 | typedef enum { |
Paul Jackson | 45b07ef | 2006-01-08 01:00:56 -0800 | [diff] [blame] | 1283 | FILE_MEMORY_MIGRATE, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1284 | FILE_CPULIST, |
| 1285 | FILE_MEMLIST, |
| 1286 | FILE_CPU_EXCLUSIVE, |
| 1287 | FILE_MEM_EXCLUSIVE, |
Paul Menage | 7860836 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 1288 | FILE_MEM_HARDWALL, |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 1289 | FILE_SCHED_LOAD_BALANCE, |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 1290 | FILE_SCHED_RELAX_DOMAIN_LEVEL, |
Paul Jackson | 3e0d98b | 2006-01-08 01:01:49 -0800 | [diff] [blame] | 1291 | FILE_MEMORY_PRESSURE_ENABLED, |
| 1292 | FILE_MEMORY_PRESSURE, |
Paul Jackson | 825a46a | 2006-03-24 03:16:03 -0800 | [diff] [blame] | 1293 | FILE_SPREAD_PAGE, |
| 1294 | FILE_SPREAD_SLAB, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1295 | } cpuset_filetype_t; |
| 1296 | |
Paul Menage | 700fe1a | 2008-04-29 01:00:00 -0700 | [diff] [blame] | 1297 | static int cpuset_write_u64(struct cgroup *cgrp, struct cftype *cft, u64 val) |
| 1298 | { |
| 1299 | int retval = 0; |
| 1300 | struct cpuset *cs = cgroup_cs(cgrp); |
| 1301 | cpuset_filetype_t type = cft->private; |
| 1302 | |
Paul Menage | e371239 | 2008-07-25 01:47:02 -0700 | [diff] [blame] | 1303 | if (!cgroup_lock_live_group(cgrp)) |
Paul Menage | 700fe1a | 2008-04-29 01:00:00 -0700 | [diff] [blame] | 1304 | return -ENODEV; |
Paul Menage | 700fe1a | 2008-04-29 01:00:00 -0700 | [diff] [blame] | 1305 | |
| 1306 | switch (type) { |
| 1307 | case FILE_CPU_EXCLUSIVE: |
| 1308 | retval = update_flag(CS_CPU_EXCLUSIVE, cs, val); |
| 1309 | break; |
| 1310 | case FILE_MEM_EXCLUSIVE: |
| 1311 | retval = update_flag(CS_MEM_EXCLUSIVE, cs, val); |
| 1312 | break; |
Paul Menage | 7860836 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 1313 | case FILE_MEM_HARDWALL: |
| 1314 | retval = update_flag(CS_MEM_HARDWALL, cs, val); |
| 1315 | break; |
Paul Menage | 700fe1a | 2008-04-29 01:00:00 -0700 | [diff] [blame] | 1316 | case FILE_SCHED_LOAD_BALANCE: |
| 1317 | retval = update_flag(CS_SCHED_LOAD_BALANCE, cs, val); |
| 1318 | break; |
| 1319 | case FILE_MEMORY_MIGRATE: |
| 1320 | retval = update_flag(CS_MEMORY_MIGRATE, cs, val); |
| 1321 | break; |
| 1322 | case FILE_MEMORY_PRESSURE_ENABLED: |
| 1323 | cpuset_memory_pressure_enabled = !!val; |
| 1324 | break; |
| 1325 | case FILE_MEMORY_PRESSURE: |
| 1326 | retval = -EACCES; |
| 1327 | break; |
| 1328 | case FILE_SPREAD_PAGE: |
| 1329 | retval = update_flag(CS_SPREAD_PAGE, cs, val); |
| 1330 | cs->mems_generation = cpuset_mems_generation++; |
| 1331 | break; |
| 1332 | case FILE_SPREAD_SLAB: |
| 1333 | retval = update_flag(CS_SPREAD_SLAB, cs, val); |
| 1334 | cs->mems_generation = cpuset_mems_generation++; |
| 1335 | break; |
| 1336 | default: |
| 1337 | retval = -EINVAL; |
| 1338 | break; |
| 1339 | } |
| 1340 | cgroup_unlock(); |
| 1341 | return retval; |
| 1342 | } |
| 1343 | |
Paul Menage | 5be7a47 | 2008-05-06 20:42:41 -0700 | [diff] [blame] | 1344 | static int cpuset_write_s64(struct cgroup *cgrp, struct cftype *cft, s64 val) |
| 1345 | { |
| 1346 | int retval = 0; |
| 1347 | struct cpuset *cs = cgroup_cs(cgrp); |
| 1348 | cpuset_filetype_t type = cft->private; |
| 1349 | |
Paul Menage | e371239 | 2008-07-25 01:47:02 -0700 | [diff] [blame] | 1350 | if (!cgroup_lock_live_group(cgrp)) |
Paul Menage | 5be7a47 | 2008-05-06 20:42:41 -0700 | [diff] [blame] | 1351 | return -ENODEV; |
Paul Menage | e371239 | 2008-07-25 01:47:02 -0700 | [diff] [blame] | 1352 | |
Paul Menage | 5be7a47 | 2008-05-06 20:42:41 -0700 | [diff] [blame] | 1353 | switch (type) { |
| 1354 | case FILE_SCHED_RELAX_DOMAIN_LEVEL: |
| 1355 | retval = update_relax_domain_level(cs, val); |
| 1356 | break; |
| 1357 | default: |
| 1358 | retval = -EINVAL; |
| 1359 | break; |
| 1360 | } |
| 1361 | cgroup_unlock(); |
| 1362 | return retval; |
| 1363 | } |
| 1364 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1365 | /* |
Paul Menage | e371239 | 2008-07-25 01:47:02 -0700 | [diff] [blame] | 1366 | * Common handling for a write to a "cpus" or "mems" file. |
| 1367 | */ |
| 1368 | static int cpuset_write_resmask(struct cgroup *cgrp, struct cftype *cft, |
| 1369 | const char *buf) |
| 1370 | { |
| 1371 | int retval = 0; |
| 1372 | |
| 1373 | if (!cgroup_lock_live_group(cgrp)) |
| 1374 | return -ENODEV; |
| 1375 | |
| 1376 | switch (cft->private) { |
| 1377 | case FILE_CPULIST: |
| 1378 | retval = update_cpumask(cgroup_cs(cgrp), buf); |
| 1379 | break; |
| 1380 | case FILE_MEMLIST: |
| 1381 | retval = update_nodemask(cgroup_cs(cgrp), buf); |
| 1382 | break; |
| 1383 | default: |
| 1384 | retval = -EINVAL; |
| 1385 | break; |
| 1386 | } |
| 1387 | cgroup_unlock(); |
| 1388 | return retval; |
| 1389 | } |
| 1390 | |
| 1391 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1392 | * These ascii lists should be read in a single call, by using a user |
| 1393 | * buffer large enough to hold the entire map. If read in smaller |
| 1394 | * chunks, there is no guarantee of atomicity. Since the display format |
| 1395 | * used, list of ranges of sequential numbers, is variable length, |
| 1396 | * and since these maps can change value dynamically, one could read |
| 1397 | * gibberish by doing partial reads while a list was changing. |
| 1398 | * A single large read to a buffer that crosses a page boundary is |
| 1399 | * ok, because the result being copied to user land is not recomputed |
| 1400 | * across a page fault. |
| 1401 | */ |
| 1402 | |
| 1403 | static int cpuset_sprintf_cpulist(char *page, struct cpuset *cs) |
| 1404 | { |
| 1405 | cpumask_t mask; |
| 1406 | |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 1407 | mutex_lock(&callback_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1408 | mask = cs->cpus_allowed; |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 1409 | mutex_unlock(&callback_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1410 | |
| 1411 | return cpulist_scnprintf(page, PAGE_SIZE, mask); |
| 1412 | } |
| 1413 | |
| 1414 | static int cpuset_sprintf_memlist(char *page, struct cpuset *cs) |
| 1415 | { |
| 1416 | nodemask_t mask; |
| 1417 | |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 1418 | mutex_lock(&callback_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1419 | mask = cs->mems_allowed; |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 1420 | mutex_unlock(&callback_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1421 | |
| 1422 | return nodelist_scnprintf(page, PAGE_SIZE, mask); |
| 1423 | } |
| 1424 | |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1425 | static ssize_t cpuset_common_file_read(struct cgroup *cont, |
| 1426 | struct cftype *cft, |
| 1427 | struct file *file, |
| 1428 | char __user *buf, |
| 1429 | size_t nbytes, loff_t *ppos) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1430 | { |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1431 | struct cpuset *cs = cgroup_cs(cont); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1432 | cpuset_filetype_t type = cft->private; |
| 1433 | char *page; |
| 1434 | ssize_t retval = 0; |
| 1435 | char *s; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1436 | |
Mel Gorman | e12ba74 | 2007-10-16 01:25:52 -0700 | [diff] [blame] | 1437 | if (!(page = (char *)__get_free_page(GFP_TEMPORARY))) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1438 | return -ENOMEM; |
| 1439 | |
| 1440 | s = page; |
| 1441 | |
| 1442 | switch (type) { |
| 1443 | case FILE_CPULIST: |
| 1444 | s += cpuset_sprintf_cpulist(s, cs); |
| 1445 | break; |
| 1446 | case FILE_MEMLIST: |
| 1447 | s += cpuset_sprintf_memlist(s, cs); |
| 1448 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1449 | default: |
| 1450 | retval = -EINVAL; |
| 1451 | goto out; |
| 1452 | } |
| 1453 | *s++ = '\n'; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1454 | |
Al Viro | eacaa1f | 2005-09-30 03:26:43 +0100 | [diff] [blame] | 1455 | retval = simple_read_from_buffer(buf, nbytes, ppos, page, s - page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1456 | out: |
| 1457 | free_page((unsigned long)page); |
| 1458 | return retval; |
| 1459 | } |
| 1460 | |
Paul Menage | 700fe1a | 2008-04-29 01:00:00 -0700 | [diff] [blame] | 1461 | static u64 cpuset_read_u64(struct cgroup *cont, struct cftype *cft) |
| 1462 | { |
| 1463 | struct cpuset *cs = cgroup_cs(cont); |
| 1464 | cpuset_filetype_t type = cft->private; |
| 1465 | switch (type) { |
| 1466 | case FILE_CPU_EXCLUSIVE: |
| 1467 | return is_cpu_exclusive(cs); |
| 1468 | case FILE_MEM_EXCLUSIVE: |
| 1469 | return is_mem_exclusive(cs); |
Paul Menage | 7860836 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 1470 | case FILE_MEM_HARDWALL: |
| 1471 | return is_mem_hardwall(cs); |
Paul Menage | 700fe1a | 2008-04-29 01:00:00 -0700 | [diff] [blame] | 1472 | case FILE_SCHED_LOAD_BALANCE: |
| 1473 | return is_sched_load_balance(cs); |
| 1474 | case FILE_MEMORY_MIGRATE: |
| 1475 | return is_memory_migrate(cs); |
| 1476 | case FILE_MEMORY_PRESSURE_ENABLED: |
| 1477 | return cpuset_memory_pressure_enabled; |
| 1478 | case FILE_MEMORY_PRESSURE: |
| 1479 | return fmeter_getrate(&cs->fmeter); |
| 1480 | case FILE_SPREAD_PAGE: |
| 1481 | return is_spread_page(cs); |
| 1482 | case FILE_SPREAD_SLAB: |
| 1483 | return is_spread_slab(cs); |
| 1484 | default: |
| 1485 | BUG(); |
| 1486 | } |
| 1487 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1488 | |
Paul Menage | 5be7a47 | 2008-05-06 20:42:41 -0700 | [diff] [blame] | 1489 | static s64 cpuset_read_s64(struct cgroup *cont, struct cftype *cft) |
| 1490 | { |
| 1491 | struct cpuset *cs = cgroup_cs(cont); |
| 1492 | cpuset_filetype_t type = cft->private; |
| 1493 | switch (type) { |
| 1494 | case FILE_SCHED_RELAX_DOMAIN_LEVEL: |
| 1495 | return cs->relax_domain_level; |
| 1496 | default: |
| 1497 | BUG(); |
| 1498 | } |
| 1499 | } |
| 1500 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1501 | |
| 1502 | /* |
| 1503 | * for the common functions, 'private' gives the type of file |
| 1504 | */ |
| 1505 | |
Paul Menage | addf2c7 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 1506 | static struct cftype files[] = { |
| 1507 | { |
| 1508 | .name = "cpus", |
| 1509 | .read = cpuset_common_file_read, |
Paul Menage | e371239 | 2008-07-25 01:47:02 -0700 | [diff] [blame] | 1510 | .write_string = cpuset_write_resmask, |
| 1511 | .max_write_len = (100U + 6 * NR_CPUS), |
Paul Menage | addf2c7 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 1512 | .private = FILE_CPULIST, |
| 1513 | }, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1514 | |
Paul Menage | addf2c7 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 1515 | { |
| 1516 | .name = "mems", |
| 1517 | .read = cpuset_common_file_read, |
Paul Menage | e371239 | 2008-07-25 01:47:02 -0700 | [diff] [blame] | 1518 | .write_string = cpuset_write_resmask, |
| 1519 | .max_write_len = (100U + 6 * MAX_NUMNODES), |
Paul Menage | addf2c7 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 1520 | .private = FILE_MEMLIST, |
| 1521 | }, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1522 | |
Paul Menage | addf2c7 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 1523 | { |
| 1524 | .name = "cpu_exclusive", |
| 1525 | .read_u64 = cpuset_read_u64, |
| 1526 | .write_u64 = cpuset_write_u64, |
| 1527 | .private = FILE_CPU_EXCLUSIVE, |
| 1528 | }, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1529 | |
Paul Menage | addf2c7 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 1530 | { |
| 1531 | .name = "mem_exclusive", |
| 1532 | .read_u64 = cpuset_read_u64, |
| 1533 | .write_u64 = cpuset_write_u64, |
| 1534 | .private = FILE_MEM_EXCLUSIVE, |
| 1535 | }, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1536 | |
Paul Menage | addf2c7 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 1537 | { |
Paul Menage | 7860836 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 1538 | .name = "mem_hardwall", |
| 1539 | .read_u64 = cpuset_read_u64, |
| 1540 | .write_u64 = cpuset_write_u64, |
| 1541 | .private = FILE_MEM_HARDWALL, |
| 1542 | }, |
| 1543 | |
| 1544 | { |
Paul Menage | addf2c7 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 1545 | .name = "sched_load_balance", |
| 1546 | .read_u64 = cpuset_read_u64, |
| 1547 | .write_u64 = cpuset_write_u64, |
| 1548 | .private = FILE_SCHED_LOAD_BALANCE, |
| 1549 | }, |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 1550 | |
Paul Menage | addf2c7 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 1551 | { |
| 1552 | .name = "sched_relax_domain_level", |
Paul Menage | 5be7a47 | 2008-05-06 20:42:41 -0700 | [diff] [blame] | 1553 | .read_s64 = cpuset_read_s64, |
| 1554 | .write_s64 = cpuset_write_s64, |
Paul Menage | addf2c7 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 1555 | .private = FILE_SCHED_RELAX_DOMAIN_LEVEL, |
| 1556 | }, |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 1557 | |
Paul Menage | addf2c7 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 1558 | { |
| 1559 | .name = "memory_migrate", |
| 1560 | .read_u64 = cpuset_read_u64, |
| 1561 | .write_u64 = cpuset_write_u64, |
| 1562 | .private = FILE_MEMORY_MIGRATE, |
| 1563 | }, |
| 1564 | |
| 1565 | { |
| 1566 | .name = "memory_pressure", |
| 1567 | .read_u64 = cpuset_read_u64, |
| 1568 | .write_u64 = cpuset_write_u64, |
| 1569 | .private = FILE_MEMORY_PRESSURE, |
| 1570 | }, |
| 1571 | |
| 1572 | { |
| 1573 | .name = "memory_spread_page", |
| 1574 | .read_u64 = cpuset_read_u64, |
| 1575 | .write_u64 = cpuset_write_u64, |
| 1576 | .private = FILE_SPREAD_PAGE, |
| 1577 | }, |
| 1578 | |
| 1579 | { |
| 1580 | .name = "memory_spread_slab", |
| 1581 | .read_u64 = cpuset_read_u64, |
| 1582 | .write_u64 = cpuset_write_u64, |
| 1583 | .private = FILE_SPREAD_SLAB, |
| 1584 | }, |
Paul Jackson | 45b07ef | 2006-01-08 01:00:56 -0800 | [diff] [blame] | 1585 | }; |
| 1586 | |
Paul Jackson | 3e0d98b | 2006-01-08 01:01:49 -0800 | [diff] [blame] | 1587 | static struct cftype cft_memory_pressure_enabled = { |
| 1588 | .name = "memory_pressure_enabled", |
Paul Menage | 700fe1a | 2008-04-29 01:00:00 -0700 | [diff] [blame] | 1589 | .read_u64 = cpuset_read_u64, |
| 1590 | .write_u64 = cpuset_write_u64, |
Paul Jackson | 3e0d98b | 2006-01-08 01:01:49 -0800 | [diff] [blame] | 1591 | .private = FILE_MEMORY_PRESSURE_ENABLED, |
| 1592 | }; |
| 1593 | |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1594 | static int cpuset_populate(struct cgroup_subsys *ss, struct cgroup *cont) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1595 | { |
| 1596 | int err; |
| 1597 | |
Paul Menage | addf2c7 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 1598 | err = cgroup_add_files(cont, ss, files, ARRAY_SIZE(files)); |
| 1599 | if (err) |
Paul Jackson | 825a46a | 2006-03-24 03:16:03 -0800 | [diff] [blame] | 1600 | return err; |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1601 | /* memory_pressure_enabled is in root cpuset only */ |
Paul Menage | addf2c7 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 1602 | if (!cont->parent) |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1603 | err = cgroup_add_file(cont, ss, |
Paul Menage | addf2c7 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 1604 | &cft_memory_pressure_enabled); |
| 1605 | return err; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1606 | } |
| 1607 | |
| 1608 | /* |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1609 | * post_clone() is called at the end of cgroup_clone(). |
| 1610 | * 'cgroup' was just created automatically as a result of |
| 1611 | * a cgroup_clone(), and the current task is about to |
| 1612 | * be moved into 'cgroup'. |
| 1613 | * |
| 1614 | * Currently we refuse to set up the cgroup - thereby |
| 1615 | * refusing the task to be entered, and as a result refusing |
| 1616 | * the sys_unshare() or clone() which initiated it - if any |
| 1617 | * sibling cpusets have exclusive cpus or mem. |
| 1618 | * |
| 1619 | * If this becomes a problem for some users who wish to |
| 1620 | * allow that scenario, then cpuset_post_clone() could be |
| 1621 | * changed to grant parent->cpus_allowed-sibling_cpus_exclusive |
Paul Menage | 2df167a | 2008-02-07 00:14:45 -0800 | [diff] [blame] | 1622 | * (and likewise for mems) to the new cgroup. Called with cgroup_mutex |
| 1623 | * held. |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1624 | */ |
| 1625 | static void cpuset_post_clone(struct cgroup_subsys *ss, |
| 1626 | struct cgroup *cgroup) |
| 1627 | { |
| 1628 | struct cgroup *parent, *child; |
| 1629 | struct cpuset *cs, *parent_cs; |
| 1630 | |
| 1631 | parent = cgroup->parent; |
| 1632 | list_for_each_entry(child, &parent->children, sibling) { |
| 1633 | cs = cgroup_cs(child); |
| 1634 | if (is_mem_exclusive(cs) || is_cpu_exclusive(cs)) |
| 1635 | return; |
| 1636 | } |
| 1637 | cs = cgroup_cs(cgroup); |
| 1638 | parent_cs = cgroup_cs(parent); |
| 1639 | |
| 1640 | cs->mems_allowed = parent_cs->mems_allowed; |
| 1641 | cs->cpus_allowed = parent_cs->cpus_allowed; |
| 1642 | return; |
| 1643 | } |
| 1644 | |
| 1645 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1646 | * cpuset_create - create a cpuset |
Paul Menage | 2df167a | 2008-02-07 00:14:45 -0800 | [diff] [blame] | 1647 | * ss: cpuset cgroup subsystem |
| 1648 | * cont: control group that the new cpuset will be part of |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1649 | */ |
| 1650 | |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1651 | static struct cgroup_subsys_state *cpuset_create( |
| 1652 | struct cgroup_subsys *ss, |
| 1653 | struct cgroup *cont) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1654 | { |
| 1655 | struct cpuset *cs; |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1656 | struct cpuset *parent; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1657 | |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1658 | if (!cont->parent) { |
| 1659 | /* This is early initialization for the top cgroup */ |
| 1660 | top_cpuset.mems_generation = cpuset_mems_generation++; |
| 1661 | return &top_cpuset.css; |
| 1662 | } |
| 1663 | parent = cgroup_cs(cont->parent); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1664 | cs = kmalloc(sizeof(*cs), GFP_KERNEL); |
| 1665 | if (!cs) |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1666 | return ERR_PTR(-ENOMEM); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1667 | |
Paul Jackson | cf2a473c | 2006-01-08 01:01:54 -0800 | [diff] [blame] | 1668 | cpuset_update_task_memory_state(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1669 | cs->flags = 0; |
Paul Jackson | 825a46a | 2006-03-24 03:16:03 -0800 | [diff] [blame] | 1670 | if (is_spread_page(parent)) |
| 1671 | set_bit(CS_SPREAD_PAGE, &cs->flags); |
| 1672 | if (is_spread_slab(parent)) |
| 1673 | set_bit(CS_SPREAD_SLAB, &cs->flags); |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 1674 | set_bit(CS_SCHED_LOAD_BALANCE, &cs->flags); |
Mike Travis | f9a86fc | 2008-04-04 18:11:07 -0700 | [diff] [blame] | 1675 | cpus_clear(cs->cpus_allowed); |
| 1676 | nodes_clear(cs->mems_allowed); |
Paul Jackson | 151a442 | 2006-03-24 03:16:11 -0800 | [diff] [blame] | 1677 | cs->mems_generation = cpuset_mems_generation++; |
Paul Jackson | 3e0d98b | 2006-01-08 01:01:49 -0800 | [diff] [blame] | 1678 | fmeter_init(&cs->fmeter); |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 1679 | cs->relax_domain_level = -1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1680 | |
| 1681 | cs->parent = parent; |
Paul Jackson | 202f72d | 2006-01-08 01:01:57 -0800 | [diff] [blame] | 1682 | number_of_cpusets++; |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1683 | return &cs->css ; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1684 | } |
| 1685 | |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 1686 | /* |
| 1687 | * Locking note on the strange update_flag() call below: |
| 1688 | * |
| 1689 | * If the cpuset being removed has its flag 'sched_load_balance' |
| 1690 | * enabled, then simulate turning sched_load_balance off, which |
Gautham R Shenoy | 86ef5c9 | 2008-01-25 21:08:02 +0100 | [diff] [blame] | 1691 | * will call rebuild_sched_domains(). The get_online_cpus() |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 1692 | * call in rebuild_sched_domains() must not be made while holding |
| 1693 | * callback_mutex. Elsewhere the kernel nests callback_mutex inside |
Gautham R Shenoy | 86ef5c9 | 2008-01-25 21:08:02 +0100 | [diff] [blame] | 1694 | * get_online_cpus() calls. So the reverse nesting would risk an |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 1695 | * ABBA deadlock. |
| 1696 | */ |
| 1697 | |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1698 | static void cpuset_destroy(struct cgroup_subsys *ss, struct cgroup *cont) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1699 | { |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1700 | struct cpuset *cs = cgroup_cs(cont); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1701 | |
Paul Jackson | cf2a473c | 2006-01-08 01:01:54 -0800 | [diff] [blame] | 1702 | cpuset_update_task_memory_state(); |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 1703 | |
| 1704 | if (is_sched_load_balance(cs)) |
Paul Menage | 700fe1a | 2008-04-29 01:00:00 -0700 | [diff] [blame] | 1705 | update_flag(CS_SCHED_LOAD_BALANCE, cs, 0); |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 1706 | |
Paul Jackson | 202f72d | 2006-01-08 01:01:57 -0800 | [diff] [blame] | 1707 | number_of_cpusets--; |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1708 | kfree(cs); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1709 | } |
| 1710 | |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1711 | struct cgroup_subsys cpuset_subsys = { |
| 1712 | .name = "cpuset", |
| 1713 | .create = cpuset_create, |
| 1714 | .destroy = cpuset_destroy, |
| 1715 | .can_attach = cpuset_can_attach, |
| 1716 | .attach = cpuset_attach, |
| 1717 | .populate = cpuset_populate, |
| 1718 | .post_clone = cpuset_post_clone, |
| 1719 | .subsys_id = cpuset_subsys_id, |
| 1720 | .early_init = 1, |
| 1721 | }; |
| 1722 | |
Paul Jackson | c417f02 | 2006-01-08 01:02:01 -0800 | [diff] [blame] | 1723 | /* |
| 1724 | * cpuset_init_early - just enough so that the calls to |
| 1725 | * cpuset_update_task_memory_state() in early init code |
| 1726 | * are harmless. |
| 1727 | */ |
| 1728 | |
| 1729 | int __init cpuset_init_early(void) |
| 1730 | { |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1731 | top_cpuset.mems_generation = cpuset_mems_generation++; |
Paul Jackson | c417f02 | 2006-01-08 01:02:01 -0800 | [diff] [blame] | 1732 | return 0; |
| 1733 | } |
| 1734 | |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1735 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1736 | /** |
| 1737 | * cpuset_init - initialize cpusets at system boot |
| 1738 | * |
| 1739 | * Description: Initialize top_cpuset and the cpuset internal file system, |
| 1740 | **/ |
| 1741 | |
| 1742 | int __init cpuset_init(void) |
| 1743 | { |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1744 | int err = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1745 | |
Mike Travis | f9a86fc | 2008-04-04 18:11:07 -0700 | [diff] [blame] | 1746 | cpus_setall(top_cpuset.cpus_allowed); |
| 1747 | nodes_setall(top_cpuset.mems_allowed); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1748 | |
Paul Jackson | 3e0d98b | 2006-01-08 01:01:49 -0800 | [diff] [blame] | 1749 | fmeter_init(&top_cpuset.fmeter); |
Paul Jackson | 151a442 | 2006-03-24 03:16:11 -0800 | [diff] [blame] | 1750 | top_cpuset.mems_generation = cpuset_mems_generation++; |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 1751 | set_bit(CS_SCHED_LOAD_BALANCE, &top_cpuset.flags); |
Hidetoshi Seto | 1d3504f | 2008-04-15 14:04:23 +0900 | [diff] [blame] | 1752 | top_cpuset.relax_domain_level = -1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1753 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1754 | err = register_filesystem(&cpuset_fs_type); |
| 1755 | if (err < 0) |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1756 | return err; |
| 1757 | |
Paul Jackson | 202f72d | 2006-01-08 01:01:57 -0800 | [diff] [blame] | 1758 | number_of_cpusets = 1; |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1759 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1760 | } |
| 1761 | |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1762 | /** |
| 1763 | * cpuset_do_move_task - move a given task to another cpuset |
| 1764 | * @tsk: pointer to task_struct the task to move |
| 1765 | * @scan: struct cgroup_scanner contained in its struct cpuset_hotplug_scanner |
| 1766 | * |
| 1767 | * Called by cgroup_scan_tasks() for each task in a cgroup. |
| 1768 | * Return nonzero to stop the walk through the tasks. |
| 1769 | */ |
Adrian Bunk | 9e0c914 | 2008-04-29 01:00:25 -0700 | [diff] [blame] | 1770 | static void cpuset_do_move_task(struct task_struct *tsk, |
| 1771 | struct cgroup_scanner *scan) |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1772 | { |
| 1773 | struct cpuset_hotplug_scanner *chsp; |
| 1774 | |
| 1775 | chsp = container_of(scan, struct cpuset_hotplug_scanner, scan); |
| 1776 | cgroup_attach_task(chsp->to, tsk); |
| 1777 | } |
| 1778 | |
| 1779 | /** |
| 1780 | * move_member_tasks_to_cpuset - move tasks from one cpuset to another |
| 1781 | * @from: cpuset in which the tasks currently reside |
| 1782 | * @to: cpuset to which the tasks will be moved |
| 1783 | * |
Paul Jackson | c8d9c90 | 2008-02-07 00:14:46 -0800 | [diff] [blame] | 1784 | * Called with cgroup_mutex held |
| 1785 | * callback_mutex must not be held, as cpuset_attach() will take it. |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1786 | * |
| 1787 | * The cgroup_scan_tasks() function will scan all the tasks in a cgroup, |
| 1788 | * calling callback functions for each. |
| 1789 | */ |
| 1790 | static void move_member_tasks_to_cpuset(struct cpuset *from, struct cpuset *to) |
| 1791 | { |
| 1792 | struct cpuset_hotplug_scanner scan; |
| 1793 | |
| 1794 | scan.scan.cg = from->css.cgroup; |
| 1795 | scan.scan.test_task = NULL; /* select all tasks in cgroup */ |
| 1796 | scan.scan.process_task = cpuset_do_move_task; |
| 1797 | scan.scan.heap = NULL; |
| 1798 | scan.to = to->css.cgroup; |
| 1799 | |
| 1800 | if (cgroup_scan_tasks((struct cgroup_scanner *)&scan)) |
| 1801 | printk(KERN_ERR "move_member_tasks_to_cpuset: " |
| 1802 | "cgroup_scan_tasks failed\n"); |
| 1803 | } |
| 1804 | |
Paul Jackson | b1aac8b | 2006-09-29 02:01:17 -0700 | [diff] [blame] | 1805 | /* |
| 1806 | * If common_cpu_mem_hotplug_unplug(), below, unplugs any CPUs |
| 1807 | * or memory nodes, we need to walk over the cpuset hierarchy, |
| 1808 | * removing that CPU or node from all cpusets. If this removes the |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1809 | * last CPU or node from a cpuset, then move the tasks in the empty |
| 1810 | * cpuset to its next-highest non-empty parent. |
Paul Jackson | b1aac8b | 2006-09-29 02:01:17 -0700 | [diff] [blame] | 1811 | * |
Paul Jackson | c8d9c90 | 2008-02-07 00:14:46 -0800 | [diff] [blame] | 1812 | * Called with cgroup_mutex held |
| 1813 | * callback_mutex must not be held, as cpuset_attach() will take it. |
Paul Jackson | b1aac8b | 2006-09-29 02:01:17 -0700 | [diff] [blame] | 1814 | */ |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1815 | static void remove_tasks_in_empty_cpuset(struct cpuset *cs) |
Paul Jackson | b1aac8b | 2006-09-29 02:01:17 -0700 | [diff] [blame] | 1816 | { |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1817 | struct cpuset *parent; |
Paul Jackson | b1aac8b | 2006-09-29 02:01:17 -0700 | [diff] [blame] | 1818 | |
Paul Jackson | c8d9c90 | 2008-02-07 00:14:46 -0800 | [diff] [blame] | 1819 | /* |
| 1820 | * The cgroup's css_sets list is in use if there are tasks |
| 1821 | * in the cpuset; the list is empty if there are none; |
| 1822 | * the cs->css.refcnt seems always 0. |
| 1823 | */ |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1824 | if (list_empty(&cs->css.cgroup->css_sets)) |
| 1825 | return; |
| 1826 | |
| 1827 | /* |
| 1828 | * Find its next-highest non-empty parent, (top cpuset |
| 1829 | * has online cpus, so can't be empty). |
| 1830 | */ |
| 1831 | parent = cs->parent; |
Paul Jackson | b450129 | 2008-02-07 00:14:47 -0800 | [diff] [blame] | 1832 | while (cpus_empty(parent->cpus_allowed) || |
| 1833 | nodes_empty(parent->mems_allowed)) |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1834 | parent = parent->parent; |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1835 | |
| 1836 | move_member_tasks_to_cpuset(cs, parent); |
| 1837 | } |
| 1838 | |
| 1839 | /* |
| 1840 | * Walk the specified cpuset subtree and look for empty cpusets. |
| 1841 | * The tasks of such cpuset must be moved to a parent cpuset. |
| 1842 | * |
Paul Menage | 2df167a | 2008-02-07 00:14:45 -0800 | [diff] [blame] | 1843 | * Called with cgroup_mutex held. We take callback_mutex to modify |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1844 | * cpus_allowed and mems_allowed. |
| 1845 | * |
| 1846 | * This walk processes the tree from top to bottom, completing one layer |
| 1847 | * before dropping down to the next. It always processes a node before |
| 1848 | * any of its children. |
| 1849 | * |
| 1850 | * For now, since we lack memory hot unplug, we'll never see a cpuset |
| 1851 | * that has tasks along with an empty 'mems'. But if we did see such |
| 1852 | * a cpuset, we'd handle it just like we do if its 'cpus' was empty. |
| 1853 | */ |
| 1854 | static void scan_for_empty_cpusets(const struct cpuset *root) |
| 1855 | { |
| 1856 | struct cpuset *cp; /* scans cpusets being updated */ |
| 1857 | struct cpuset *child; /* scans child cpusets of cp */ |
| 1858 | struct list_head queue; |
| 1859 | struct cgroup *cont; |
Miao Xie | f9b4fb8 | 2008-07-25 01:47:22 -0700 | [diff] [blame] | 1860 | nodemask_t oldmems; |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1861 | |
| 1862 | INIT_LIST_HEAD(&queue); |
| 1863 | |
| 1864 | list_add_tail((struct list_head *)&root->stack_list, &queue); |
| 1865 | |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1866 | while (!list_empty(&queue)) { |
| 1867 | cp = container_of(queue.next, struct cpuset, stack_list); |
| 1868 | list_del(queue.next); |
| 1869 | list_for_each_entry(cont, &cp->css.cgroup->children, sibling) { |
| 1870 | child = cgroup_cs(cont); |
| 1871 | list_add_tail(&child->stack_list, &queue); |
| 1872 | } |
| 1873 | cont = cp->css.cgroup; |
Paul Jackson | b450129 | 2008-02-07 00:14:47 -0800 | [diff] [blame] | 1874 | |
| 1875 | /* Continue past cpusets with all cpus, mems online */ |
| 1876 | if (cpus_subset(cp->cpus_allowed, cpu_online_map) && |
| 1877 | nodes_subset(cp->mems_allowed, node_states[N_HIGH_MEMORY])) |
| 1878 | continue; |
| 1879 | |
Miao Xie | f9b4fb8 | 2008-07-25 01:47:22 -0700 | [diff] [blame] | 1880 | oldmems = cp->mems_allowed; |
| 1881 | |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1882 | /* Remove offline cpus and mems from this cpuset. */ |
Paul Jackson | b450129 | 2008-02-07 00:14:47 -0800 | [diff] [blame] | 1883 | mutex_lock(&callback_mutex); |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1884 | cpus_and(cp->cpus_allowed, cp->cpus_allowed, cpu_online_map); |
| 1885 | nodes_and(cp->mems_allowed, cp->mems_allowed, |
| 1886 | node_states[N_HIGH_MEMORY]); |
Paul Jackson | b450129 | 2008-02-07 00:14:47 -0800 | [diff] [blame] | 1887 | mutex_unlock(&callback_mutex); |
| 1888 | |
| 1889 | /* Move tasks from the empty cpuset to a parent */ |
Paul Jackson | c8d9c90 | 2008-02-07 00:14:46 -0800 | [diff] [blame] | 1890 | if (cpus_empty(cp->cpus_allowed) || |
Paul Jackson | b450129 | 2008-02-07 00:14:47 -0800 | [diff] [blame] | 1891 | nodes_empty(cp->mems_allowed)) |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1892 | remove_tasks_in_empty_cpuset(cp); |
Miao Xie | f9b4fb8 | 2008-07-25 01:47:22 -0700 | [diff] [blame] | 1893 | else { |
| 1894 | update_tasks_cpumask(cp); |
| 1895 | update_tasks_nodemask(cp, &oldmems); |
| 1896 | } |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1897 | } |
Paul Jackson | b1aac8b | 2006-09-29 02:01:17 -0700 | [diff] [blame] | 1898 | } |
| 1899 | |
| 1900 | /* |
| 1901 | * The cpus_allowed and mems_allowed nodemasks in the top_cpuset track |
Christoph Lameter | 0e1e7c7 | 2007-10-16 01:25:38 -0700 | [diff] [blame] | 1902 | * cpu_online_map and node_states[N_HIGH_MEMORY]. Force the top cpuset to |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1903 | * track what's online after any CPU or memory node hotplug or unplug event. |
Paul Jackson | b1aac8b | 2006-09-29 02:01:17 -0700 | [diff] [blame] | 1904 | * |
| 1905 | * Since there are two callers of this routine, one for CPU hotplug |
| 1906 | * events and one for memory node hotplug events, we could have coded |
| 1907 | * two separate routines here. We code it as a single common routine |
| 1908 | * in order to minimize text size. |
| 1909 | */ |
| 1910 | |
Dmitry Adamushko | 3e84050 | 2008-07-13 02:10:29 +0200 | [diff] [blame] | 1911 | static void common_cpu_mem_hotplug_unplug(int rebuild_sd) |
Paul Jackson | b1aac8b | 2006-09-29 02:01:17 -0700 | [diff] [blame] | 1912 | { |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1913 | cgroup_lock(); |
Paul Jackson | b1aac8b | 2006-09-29 02:01:17 -0700 | [diff] [blame] | 1914 | |
Paul Jackson | b1aac8b | 2006-09-29 02:01:17 -0700 | [diff] [blame] | 1915 | top_cpuset.cpus_allowed = cpu_online_map; |
Christoph Lameter | 0e1e7c7 | 2007-10-16 01:25:38 -0700 | [diff] [blame] | 1916 | top_cpuset.mems_allowed = node_states[N_HIGH_MEMORY]; |
Cliff Wickman | 956db3c | 2008-02-07 00:14:43 -0800 | [diff] [blame] | 1917 | scan_for_empty_cpusets(&top_cpuset); |
Paul Jackson | b1aac8b | 2006-09-29 02:01:17 -0700 | [diff] [blame] | 1918 | |
Max Krasnyansky | 5c8e1ed | 2008-05-29 11:17:01 -0700 | [diff] [blame] | 1919 | /* |
| 1920 | * Scheduler destroys domains on hotplug events. |
| 1921 | * Rebuild them based on the current settings. |
| 1922 | */ |
Dmitry Adamushko | 3e84050 | 2008-07-13 02:10:29 +0200 | [diff] [blame] | 1923 | if (rebuild_sd) |
| 1924 | rebuild_sched_domains(); |
Max Krasnyansky | 5c8e1ed | 2008-05-29 11:17:01 -0700 | [diff] [blame] | 1925 | |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 1926 | cgroup_unlock(); |
Paul Jackson | b1aac8b | 2006-09-29 02:01:17 -0700 | [diff] [blame] | 1927 | } |
Paul Jackson | b1aac8b | 2006-09-29 02:01:17 -0700 | [diff] [blame] | 1928 | |
Paul Jackson | 4c4d50f | 2006-08-27 01:23:51 -0700 | [diff] [blame] | 1929 | /* |
| 1930 | * The top_cpuset tracks what CPUs and Memory Nodes are online, |
| 1931 | * period. This is necessary in order to make cpusets transparent |
| 1932 | * (of no affect) on systems that are actively using CPU hotplug |
| 1933 | * but making no active use of cpusets. |
| 1934 | * |
Paul Jackson | 38837fc | 2006-09-29 02:01:16 -0700 | [diff] [blame] | 1935 | * This routine ensures that top_cpuset.cpus_allowed tracks |
| 1936 | * cpu_online_map on each CPU hotplug (cpuhp) event. |
Paul Jackson | 4c4d50f | 2006-08-27 01:23:51 -0700 | [diff] [blame] | 1937 | */ |
| 1938 | |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 1939 | static int cpuset_handle_cpuhp(struct notifier_block *unused_nb, |
| 1940 | unsigned long phase, void *unused_cpu) |
Paul Jackson | 4c4d50f | 2006-08-27 01:23:51 -0700 | [diff] [blame] | 1941 | { |
Dmitry Adamushko | 3e84050 | 2008-07-13 02:10:29 +0200 | [diff] [blame] | 1942 | switch (phase) { |
| 1943 | case CPU_UP_CANCELED: |
| 1944 | case CPU_UP_CANCELED_FROZEN: |
| 1945 | case CPU_DOWN_FAILED: |
| 1946 | case CPU_DOWN_FAILED_FROZEN: |
| 1947 | case CPU_ONLINE: |
| 1948 | case CPU_ONLINE_FROZEN: |
| 1949 | case CPU_DEAD: |
| 1950 | case CPU_DEAD_FROZEN: |
| 1951 | common_cpu_mem_hotplug_unplug(1); |
| 1952 | break; |
| 1953 | default: |
Avi Kivity | ac07675 | 2007-05-24 12:33:15 +0300 | [diff] [blame] | 1954 | return NOTIFY_DONE; |
Dmitry Adamushko | 3e84050 | 2008-07-13 02:10:29 +0200 | [diff] [blame] | 1955 | } |
Avi Kivity | ac07675 | 2007-05-24 12:33:15 +0300 | [diff] [blame] | 1956 | |
Dmitry Adamushko | 3e84050 | 2008-07-13 02:10:29 +0200 | [diff] [blame] | 1957 | return NOTIFY_OK; |
Paul Jackson | 4c4d50f | 2006-08-27 01:23:51 -0700 | [diff] [blame] | 1958 | } |
Paul Jackson | 4c4d50f | 2006-08-27 01:23:51 -0700 | [diff] [blame] | 1959 | |
Paul Jackson | b1aac8b | 2006-09-29 02:01:17 -0700 | [diff] [blame] | 1960 | #ifdef CONFIG_MEMORY_HOTPLUG |
Paul Jackson | 38837fc | 2006-09-29 02:01:16 -0700 | [diff] [blame] | 1961 | /* |
Christoph Lameter | 0e1e7c7 | 2007-10-16 01:25:38 -0700 | [diff] [blame] | 1962 | * Keep top_cpuset.mems_allowed tracking node_states[N_HIGH_MEMORY]. |
| 1963 | * Call this routine anytime after you change |
| 1964 | * node_states[N_HIGH_MEMORY]. |
Paul Jackson | 38837fc | 2006-09-29 02:01:16 -0700 | [diff] [blame] | 1965 | * See also the previous routine cpuset_handle_cpuhp(). |
| 1966 | */ |
| 1967 | |
Al Viro | 1af9892 | 2006-10-10 22:48:57 +0100 | [diff] [blame] | 1968 | void cpuset_track_online_nodes(void) |
Paul Jackson | 38837fc | 2006-09-29 02:01:16 -0700 | [diff] [blame] | 1969 | { |
Dmitry Adamushko | 3e84050 | 2008-07-13 02:10:29 +0200 | [diff] [blame] | 1970 | common_cpu_mem_hotplug_unplug(0); |
Paul Jackson | 38837fc | 2006-09-29 02:01:16 -0700 | [diff] [blame] | 1971 | } |
| 1972 | #endif |
| 1973 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1974 | /** |
| 1975 | * cpuset_init_smp - initialize cpus_allowed |
| 1976 | * |
| 1977 | * Description: Finish top cpuset after cpu, node maps are initialized |
| 1978 | **/ |
| 1979 | |
| 1980 | void __init cpuset_init_smp(void) |
| 1981 | { |
| 1982 | top_cpuset.cpus_allowed = cpu_online_map; |
Christoph Lameter | 0e1e7c7 | 2007-10-16 01:25:38 -0700 | [diff] [blame] | 1983 | top_cpuset.mems_allowed = node_states[N_HIGH_MEMORY]; |
Paul Jackson | 4c4d50f | 2006-08-27 01:23:51 -0700 | [diff] [blame] | 1984 | |
| 1985 | hotcpu_notifier(cpuset_handle_cpuhp, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1986 | } |
| 1987 | |
| 1988 | /** |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1989 | * cpuset_cpus_allowed - return cpus_allowed mask from a tasks cpuset. |
| 1990 | * @tsk: pointer to task_struct from which to obtain cpuset->cpus_allowed. |
Mike Travis | f9a86fc | 2008-04-04 18:11:07 -0700 | [diff] [blame] | 1991 | * @pmask: pointer to cpumask_t variable to receive cpus_allowed set. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1992 | * |
| 1993 | * Description: Returns the cpumask_t cpus_allowed of the cpuset |
| 1994 | * attached to the specified @tsk. Guaranteed to return some non-empty |
| 1995 | * subset of cpu_online_map, even if this means going outside the |
| 1996 | * tasks cpuset. |
| 1997 | **/ |
| 1998 | |
Mike Travis | f9a86fc | 2008-04-04 18:11:07 -0700 | [diff] [blame] | 1999 | void cpuset_cpus_allowed(struct task_struct *tsk, cpumask_t *pmask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2000 | { |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 2001 | mutex_lock(&callback_mutex); |
Mike Travis | f9a86fc | 2008-04-04 18:11:07 -0700 | [diff] [blame] | 2002 | cpuset_cpus_allowed_locked(tsk, pmask); |
Cliff Wickman | 470fd64 | 2007-10-18 23:40:46 -0700 | [diff] [blame] | 2003 | mutex_unlock(&callback_mutex); |
Cliff Wickman | 470fd64 | 2007-10-18 23:40:46 -0700 | [diff] [blame] | 2004 | } |
| 2005 | |
| 2006 | /** |
| 2007 | * cpuset_cpus_allowed_locked - return cpus_allowed mask from a tasks cpuset. |
Paul Menage | 2df167a | 2008-02-07 00:14:45 -0800 | [diff] [blame] | 2008 | * Must be called with callback_mutex held. |
Cliff Wickman | 470fd64 | 2007-10-18 23:40:46 -0700 | [diff] [blame] | 2009 | **/ |
Mike Travis | f9a86fc | 2008-04-04 18:11:07 -0700 | [diff] [blame] | 2010 | void cpuset_cpus_allowed_locked(struct task_struct *tsk, cpumask_t *pmask) |
Cliff Wickman | 470fd64 | 2007-10-18 23:40:46 -0700 | [diff] [blame] | 2011 | { |
Paul Jackson | 909d75a | 2006-01-08 01:01:55 -0800 | [diff] [blame] | 2012 | task_lock(tsk); |
Mike Travis | f9a86fc | 2008-04-04 18:11:07 -0700 | [diff] [blame] | 2013 | guarantee_online_cpus(task_cs(tsk), pmask); |
Paul Jackson | 909d75a | 2006-01-08 01:01:55 -0800 | [diff] [blame] | 2014 | task_unlock(tsk); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2015 | } |
| 2016 | |
| 2017 | void cpuset_init_current_mems_allowed(void) |
| 2018 | { |
Mike Travis | f9a86fc | 2008-04-04 18:11:07 -0700 | [diff] [blame] | 2019 | nodes_setall(current->mems_allowed); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2020 | } |
| 2021 | |
Randy Dunlap | d9fd8a6 | 2005-07-27 11:45:11 -0700 | [diff] [blame] | 2022 | /** |
Paul Jackson | 909d75a | 2006-01-08 01:01:55 -0800 | [diff] [blame] | 2023 | * cpuset_mems_allowed - return mems_allowed mask from a tasks cpuset. |
| 2024 | * @tsk: pointer to task_struct from which to obtain cpuset->mems_allowed. |
| 2025 | * |
| 2026 | * Description: Returns the nodemask_t mems_allowed of the cpuset |
| 2027 | * attached to the specified @tsk. Guaranteed to return some non-empty |
Christoph Lameter | 0e1e7c7 | 2007-10-16 01:25:38 -0700 | [diff] [blame] | 2028 | * subset of node_states[N_HIGH_MEMORY], even if this means going outside the |
Paul Jackson | 909d75a | 2006-01-08 01:01:55 -0800 | [diff] [blame] | 2029 | * tasks cpuset. |
| 2030 | **/ |
| 2031 | |
| 2032 | nodemask_t cpuset_mems_allowed(struct task_struct *tsk) |
| 2033 | { |
| 2034 | nodemask_t mask; |
| 2035 | |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 2036 | mutex_lock(&callback_mutex); |
Paul Jackson | 909d75a | 2006-01-08 01:01:55 -0800 | [diff] [blame] | 2037 | task_lock(tsk); |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 2038 | guarantee_online_mems(task_cs(tsk), &mask); |
Paul Jackson | 909d75a | 2006-01-08 01:01:55 -0800 | [diff] [blame] | 2039 | task_unlock(tsk); |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 2040 | mutex_unlock(&callback_mutex); |
Paul Jackson | 909d75a | 2006-01-08 01:01:55 -0800 | [diff] [blame] | 2041 | |
| 2042 | return mask; |
| 2043 | } |
| 2044 | |
| 2045 | /** |
Mel Gorman | 19770b3 | 2008-04-28 02:12:18 -0700 | [diff] [blame] | 2046 | * cpuset_nodemask_valid_mems_allowed - check nodemask vs. curremt mems_allowed |
| 2047 | * @nodemask: the nodemask to be checked |
Randy Dunlap | d9fd8a6 | 2005-07-27 11:45:11 -0700 | [diff] [blame] | 2048 | * |
Mel Gorman | 19770b3 | 2008-04-28 02:12:18 -0700 | [diff] [blame] | 2049 | * Are any of the nodes in the nodemask allowed in current->mems_allowed? |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2050 | */ |
Mel Gorman | 19770b3 | 2008-04-28 02:12:18 -0700 | [diff] [blame] | 2051 | int cpuset_nodemask_valid_mems_allowed(nodemask_t *nodemask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2052 | { |
Mel Gorman | 19770b3 | 2008-04-28 02:12:18 -0700 | [diff] [blame] | 2053 | return nodes_intersects(*nodemask, current->mems_allowed); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2054 | } |
| 2055 | |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 2056 | /* |
Paul Menage | 7860836 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 2057 | * nearest_hardwall_ancestor() - Returns the nearest mem_exclusive or |
| 2058 | * mem_hardwall ancestor to the specified cpuset. Call holding |
| 2059 | * callback_mutex. If no ancestor is mem_exclusive or mem_hardwall |
| 2060 | * (an unusual configuration), then returns the root cpuset. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2061 | */ |
Paul Menage | 7860836 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 2062 | static const struct cpuset *nearest_hardwall_ancestor(const struct cpuset *cs) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2063 | { |
Paul Menage | 7860836 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 2064 | while (!(is_mem_exclusive(cs) || is_mem_hardwall(cs)) && cs->parent) |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 2065 | cs = cs->parent; |
| 2066 | return cs; |
| 2067 | } |
| 2068 | |
| 2069 | /** |
Paul Jackson | 02a0e53 | 2006-12-13 00:34:25 -0800 | [diff] [blame] | 2070 | * cpuset_zone_allowed_softwall - Can we allocate on zone z's memory node? |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 2071 | * @z: is this zone on an allowed node? |
Paul Jackson | 02a0e53 | 2006-12-13 00:34:25 -0800 | [diff] [blame] | 2072 | * @gfp_mask: memory allocation flags |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 2073 | * |
Paul Jackson | 02a0e53 | 2006-12-13 00:34:25 -0800 | [diff] [blame] | 2074 | * If we're in interrupt, yes, we can always allocate. If |
| 2075 | * __GFP_THISNODE is set, yes, we can always allocate. If zone |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 2076 | * z's node is in our tasks mems_allowed, yes. If it's not a |
| 2077 | * __GFP_HARDWALL request and this zone's nodes is in the nearest |
Paul Menage | 7860836 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 2078 | * hardwalled cpuset ancestor to this tasks cpuset, yes. |
David Rientjes | c596d9f | 2007-05-06 14:49:32 -0700 | [diff] [blame] | 2079 | * If the task has been OOM killed and has access to memory reserves |
| 2080 | * as specified by the TIF_MEMDIE flag, yes. |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 2081 | * Otherwise, no. |
| 2082 | * |
Paul Jackson | 02a0e53 | 2006-12-13 00:34:25 -0800 | [diff] [blame] | 2083 | * If __GFP_HARDWALL is set, cpuset_zone_allowed_softwall() |
| 2084 | * reduces to cpuset_zone_allowed_hardwall(). Otherwise, |
| 2085 | * cpuset_zone_allowed_softwall() might sleep, and might allow a zone |
| 2086 | * from an enclosing cpuset. |
| 2087 | * |
| 2088 | * cpuset_zone_allowed_hardwall() only handles the simpler case of |
| 2089 | * hardwall cpusets, and never sleeps. |
| 2090 | * |
| 2091 | * The __GFP_THISNODE placement logic is really handled elsewhere, |
| 2092 | * by forcibly using a zonelist starting at a specified node, and by |
| 2093 | * (in get_page_from_freelist()) refusing to consider the zones for |
| 2094 | * any node on the zonelist except the first. By the time any such |
| 2095 | * calls get to this routine, we should just shut up and say 'yes'. |
| 2096 | * |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 2097 | * GFP_USER allocations are marked with the __GFP_HARDWALL bit, |
David Rientjes | c596d9f | 2007-05-06 14:49:32 -0700 | [diff] [blame] | 2098 | * and do not allow allocations outside the current tasks cpuset |
| 2099 | * unless the task has been OOM killed as is marked TIF_MEMDIE. |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 2100 | * GFP_KERNEL allocations are not so marked, so can escape to the |
Paul Menage | 7860836 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 2101 | * nearest enclosing hardwalled ancestor cpuset. |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 2102 | * |
Paul Jackson | 02a0e53 | 2006-12-13 00:34:25 -0800 | [diff] [blame] | 2103 | * Scanning up parent cpusets requires callback_mutex. The |
| 2104 | * __alloc_pages() routine only calls here with __GFP_HARDWALL bit |
| 2105 | * _not_ set if it's a GFP_KERNEL allocation, and all nodes in the |
| 2106 | * current tasks mems_allowed came up empty on the first pass over |
| 2107 | * the zonelist. So only GFP_KERNEL allocations, if all nodes in the |
| 2108 | * cpuset are short of memory, might require taking the callback_mutex |
| 2109 | * mutex. |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 2110 | * |
Paul Jackson | 36be57f | 2006-05-20 15:00:10 -0700 | [diff] [blame] | 2111 | * The first call here from mm/page_alloc:get_page_from_freelist() |
Paul Jackson | 02a0e53 | 2006-12-13 00:34:25 -0800 | [diff] [blame] | 2112 | * has __GFP_HARDWALL set in gfp_mask, enforcing hardwall cpusets, |
| 2113 | * so no allocation on a node outside the cpuset is allowed (unless |
| 2114 | * in interrupt, of course). |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 2115 | * |
Paul Jackson | 36be57f | 2006-05-20 15:00:10 -0700 | [diff] [blame] | 2116 | * The second pass through get_page_from_freelist() doesn't even call |
| 2117 | * here for GFP_ATOMIC calls. For those calls, the __alloc_pages() |
| 2118 | * variable 'wait' is not set, and the bit ALLOC_CPUSET is not set |
| 2119 | * in alloc_flags. That logic and the checks below have the combined |
| 2120 | * affect that: |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 2121 | * in_interrupt - any node ok (current task context irrelevant) |
| 2122 | * GFP_ATOMIC - any node ok |
David Rientjes | c596d9f | 2007-05-06 14:49:32 -0700 | [diff] [blame] | 2123 | * TIF_MEMDIE - any node ok |
Paul Menage | 7860836 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 2124 | * GFP_KERNEL - any node in enclosing hardwalled cpuset ok |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 2125 | * GFP_USER - only nodes in current tasks mems allowed ok. |
Paul Jackson | 36be57f | 2006-05-20 15:00:10 -0700 | [diff] [blame] | 2126 | * |
| 2127 | * Rule: |
Paul Jackson | 02a0e53 | 2006-12-13 00:34:25 -0800 | [diff] [blame] | 2128 | * Don't call cpuset_zone_allowed_softwall if you can't sleep, unless you |
Paul Jackson | 36be57f | 2006-05-20 15:00:10 -0700 | [diff] [blame] | 2129 | * pass in the __GFP_HARDWALL flag set in gfp_flag, which disables |
| 2130 | * the code that might scan up ancestor cpusets and sleep. |
Paul Jackson | 02a0e53 | 2006-12-13 00:34:25 -0800 | [diff] [blame] | 2131 | */ |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 2132 | |
Paul Jackson | 02a0e53 | 2006-12-13 00:34:25 -0800 | [diff] [blame] | 2133 | int __cpuset_zone_allowed_softwall(struct zone *z, gfp_t gfp_mask) |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 2134 | { |
| 2135 | int node; /* node that zone z is on */ |
| 2136 | const struct cpuset *cs; /* current cpuset ancestors */ |
Paul Jackson | 29afd49 | 2006-03-24 03:16:12 -0800 | [diff] [blame] | 2137 | int allowed; /* is allocation in zone z allowed? */ |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 2138 | |
Christoph Lameter | 9b819d2 | 2006-09-25 23:31:40 -0700 | [diff] [blame] | 2139 | if (in_interrupt() || (gfp_mask & __GFP_THISNODE)) |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 2140 | return 1; |
Christoph Lameter | 89fa302 | 2006-09-25 23:31:55 -0700 | [diff] [blame] | 2141 | node = zone_to_nid(z); |
Paul Jackson | 92d1dbd | 2006-05-20 15:00:11 -0700 | [diff] [blame] | 2142 | might_sleep_if(!(gfp_mask & __GFP_HARDWALL)); |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 2143 | if (node_isset(node, current->mems_allowed)) |
| 2144 | return 1; |
David Rientjes | c596d9f | 2007-05-06 14:49:32 -0700 | [diff] [blame] | 2145 | /* |
| 2146 | * Allow tasks that have access to memory reserves because they have |
| 2147 | * been OOM killed to get memory anywhere. |
| 2148 | */ |
| 2149 | if (unlikely(test_thread_flag(TIF_MEMDIE))) |
| 2150 | return 1; |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 2151 | if (gfp_mask & __GFP_HARDWALL) /* If hardwall request, stop here */ |
| 2152 | return 0; |
| 2153 | |
Bob Picco | 5563e77 | 2005-11-13 16:06:35 -0800 | [diff] [blame] | 2154 | if (current->flags & PF_EXITING) /* Let dying task have memory */ |
| 2155 | return 1; |
| 2156 | |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 2157 | /* Not hardwall and node outside mems_allowed: scan up cpusets */ |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 2158 | mutex_lock(&callback_mutex); |
Paul Jackson | 053199e | 2005-10-30 15:02:30 -0800 | [diff] [blame] | 2159 | |
Paul Jackson | 053199e | 2005-10-30 15:02:30 -0800 | [diff] [blame] | 2160 | task_lock(current); |
Paul Menage | 7860836 | 2008-04-29 01:00:26 -0700 | [diff] [blame] | 2161 | cs = nearest_hardwall_ancestor(task_cs(current)); |
Paul Jackson | 053199e | 2005-10-30 15:02:30 -0800 | [diff] [blame] | 2162 | task_unlock(current); |
| 2163 | |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 2164 | allowed = node_isset(node, cs->mems_allowed); |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 2165 | mutex_unlock(&callback_mutex); |
Paul Jackson | 9bf2229 | 2005-09-06 15:18:12 -0700 | [diff] [blame] | 2166 | return allowed; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2167 | } |
| 2168 | |
Paul Jackson | 02a0e53 | 2006-12-13 00:34:25 -0800 | [diff] [blame] | 2169 | /* |
| 2170 | * cpuset_zone_allowed_hardwall - Can we allocate on zone z's memory node? |
| 2171 | * @z: is this zone on an allowed node? |
| 2172 | * @gfp_mask: memory allocation flags |
| 2173 | * |
| 2174 | * If we're in interrupt, yes, we can always allocate. |
| 2175 | * If __GFP_THISNODE is set, yes, we can always allocate. If zone |
David Rientjes | c596d9f | 2007-05-06 14:49:32 -0700 | [diff] [blame] | 2176 | * z's node is in our tasks mems_allowed, yes. If the task has been |
| 2177 | * OOM killed and has access to memory reserves as specified by the |
| 2178 | * TIF_MEMDIE flag, yes. Otherwise, no. |
Paul Jackson | 02a0e53 | 2006-12-13 00:34:25 -0800 | [diff] [blame] | 2179 | * |
| 2180 | * The __GFP_THISNODE placement logic is really handled elsewhere, |
| 2181 | * by forcibly using a zonelist starting at a specified node, and by |
| 2182 | * (in get_page_from_freelist()) refusing to consider the zones for |
| 2183 | * any node on the zonelist except the first. By the time any such |
| 2184 | * calls get to this routine, we should just shut up and say 'yes'. |
| 2185 | * |
| 2186 | * Unlike the cpuset_zone_allowed_softwall() variant, above, |
| 2187 | * this variant requires that the zone be in the current tasks |
| 2188 | * mems_allowed or that we're in interrupt. It does not scan up the |
| 2189 | * cpuset hierarchy for the nearest enclosing mem_exclusive cpuset. |
| 2190 | * It never sleeps. |
| 2191 | */ |
| 2192 | |
| 2193 | int __cpuset_zone_allowed_hardwall(struct zone *z, gfp_t gfp_mask) |
| 2194 | { |
| 2195 | int node; /* node that zone z is on */ |
| 2196 | |
| 2197 | if (in_interrupt() || (gfp_mask & __GFP_THISNODE)) |
| 2198 | return 1; |
| 2199 | node = zone_to_nid(z); |
| 2200 | if (node_isset(node, current->mems_allowed)) |
| 2201 | return 1; |
Daniel Walker | dedf8b7 | 2007-10-18 03:06:04 -0700 | [diff] [blame] | 2202 | /* |
| 2203 | * Allow tasks that have access to memory reserves because they have |
| 2204 | * been OOM killed to get memory anywhere. |
| 2205 | */ |
| 2206 | if (unlikely(test_thread_flag(TIF_MEMDIE))) |
| 2207 | return 1; |
Paul Jackson | 02a0e53 | 2006-12-13 00:34:25 -0800 | [diff] [blame] | 2208 | return 0; |
| 2209 | } |
| 2210 | |
Paul Jackson | ef08e3b | 2005-09-06 15:18:13 -0700 | [diff] [blame] | 2211 | /** |
Paul Jackson | 505970b | 2006-01-14 13:21:06 -0800 | [diff] [blame] | 2212 | * cpuset_lock - lock out any changes to cpuset structures |
| 2213 | * |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 2214 | * The out of memory (oom) code needs to mutex_lock cpusets |
Paul Jackson | 505970b | 2006-01-14 13:21:06 -0800 | [diff] [blame] | 2215 | * from being changed while it scans the tasklist looking for a |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 2216 | * task in an overlapping cpuset. Expose callback_mutex via this |
Paul Jackson | 505970b | 2006-01-14 13:21:06 -0800 | [diff] [blame] | 2217 | * cpuset_lock() routine, so the oom code can lock it, before |
| 2218 | * locking the task list. The tasklist_lock is a spinlock, so |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 2219 | * must be taken inside callback_mutex. |
Paul Jackson | 505970b | 2006-01-14 13:21:06 -0800 | [diff] [blame] | 2220 | */ |
| 2221 | |
| 2222 | void cpuset_lock(void) |
| 2223 | { |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 2224 | mutex_lock(&callback_mutex); |
Paul Jackson | 505970b | 2006-01-14 13:21:06 -0800 | [diff] [blame] | 2225 | } |
| 2226 | |
| 2227 | /** |
| 2228 | * cpuset_unlock - release lock on cpuset changes |
| 2229 | * |
| 2230 | * Undo the lock taken in a previous cpuset_lock() call. |
| 2231 | */ |
| 2232 | |
| 2233 | void cpuset_unlock(void) |
| 2234 | { |
Ingo Molnar | 3d3f26a | 2006-03-23 03:00:18 -0800 | [diff] [blame] | 2235 | mutex_unlock(&callback_mutex); |
Paul Jackson | 505970b | 2006-01-14 13:21:06 -0800 | [diff] [blame] | 2236 | } |
| 2237 | |
| 2238 | /** |
Paul Jackson | 825a46a | 2006-03-24 03:16:03 -0800 | [diff] [blame] | 2239 | * cpuset_mem_spread_node() - On which node to begin search for a page |
| 2240 | * |
| 2241 | * If a task is marked PF_SPREAD_PAGE or PF_SPREAD_SLAB (as for |
| 2242 | * tasks in a cpuset with is_spread_page or is_spread_slab set), |
| 2243 | * and if the memory allocation used cpuset_mem_spread_node() |
| 2244 | * to determine on which node to start looking, as it will for |
| 2245 | * certain page cache or slab cache pages such as used for file |
| 2246 | * system buffers and inode caches, then instead of starting on the |
| 2247 | * local node to look for a free page, rather spread the starting |
| 2248 | * node around the tasks mems_allowed nodes. |
| 2249 | * |
| 2250 | * We don't have to worry about the returned node being offline |
| 2251 | * because "it can't happen", and even if it did, it would be ok. |
| 2252 | * |
| 2253 | * The routines calling guarantee_online_mems() are careful to |
| 2254 | * only set nodes in task->mems_allowed that are online. So it |
| 2255 | * should not be possible for the following code to return an |
| 2256 | * offline node. But if it did, that would be ok, as this routine |
| 2257 | * is not returning the node where the allocation must be, only |
| 2258 | * the node where the search should start. The zonelist passed to |
| 2259 | * __alloc_pages() will include all nodes. If the slab allocator |
| 2260 | * is passed an offline node, it will fall back to the local node. |
| 2261 | * See kmem_cache_alloc_node(). |
| 2262 | */ |
| 2263 | |
| 2264 | int cpuset_mem_spread_node(void) |
| 2265 | { |
| 2266 | int node; |
| 2267 | |
| 2268 | node = next_node(current->cpuset_mem_spread_rotor, current->mems_allowed); |
| 2269 | if (node == MAX_NUMNODES) |
| 2270 | node = first_node(current->mems_allowed); |
| 2271 | current->cpuset_mem_spread_rotor = node; |
| 2272 | return node; |
| 2273 | } |
| 2274 | EXPORT_SYMBOL_GPL(cpuset_mem_spread_node); |
| 2275 | |
| 2276 | /** |
David Rientjes | bbe373f | 2007-10-16 23:25:58 -0700 | [diff] [blame] | 2277 | * cpuset_mems_allowed_intersects - Does @tsk1's mems_allowed intersect @tsk2's? |
| 2278 | * @tsk1: pointer to task_struct of some task. |
| 2279 | * @tsk2: pointer to task_struct of some other task. |
Paul Jackson | ef08e3b | 2005-09-06 15:18:13 -0700 | [diff] [blame] | 2280 | * |
David Rientjes | bbe373f | 2007-10-16 23:25:58 -0700 | [diff] [blame] | 2281 | * Description: Return true if @tsk1's mems_allowed intersects the |
| 2282 | * mems_allowed of @tsk2. Used by the OOM killer to determine if |
| 2283 | * one of the task's memory usage might impact the memory available |
| 2284 | * to the other. |
Paul Jackson | ef08e3b | 2005-09-06 15:18:13 -0700 | [diff] [blame] | 2285 | **/ |
| 2286 | |
David Rientjes | bbe373f | 2007-10-16 23:25:58 -0700 | [diff] [blame] | 2287 | int cpuset_mems_allowed_intersects(const struct task_struct *tsk1, |
| 2288 | const struct task_struct *tsk2) |
Paul Jackson | ef08e3b | 2005-09-06 15:18:13 -0700 | [diff] [blame] | 2289 | { |
David Rientjes | bbe373f | 2007-10-16 23:25:58 -0700 | [diff] [blame] | 2290 | return nodes_intersects(tsk1->mems_allowed, tsk2->mems_allowed); |
Paul Jackson | ef08e3b | 2005-09-06 15:18:13 -0700 | [diff] [blame] | 2291 | } |
| 2292 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2293 | /* |
Paul Jackson | 3e0d98b | 2006-01-08 01:01:49 -0800 | [diff] [blame] | 2294 | * Collection of memory_pressure is suppressed unless |
| 2295 | * this flag is enabled by writing "1" to the special |
| 2296 | * cpuset file 'memory_pressure_enabled' in the root cpuset. |
| 2297 | */ |
| 2298 | |
Paul Jackson | c5b2aff | 2006-01-08 01:01:51 -0800 | [diff] [blame] | 2299 | int cpuset_memory_pressure_enabled __read_mostly; |
Paul Jackson | 3e0d98b | 2006-01-08 01:01:49 -0800 | [diff] [blame] | 2300 | |
| 2301 | /** |
| 2302 | * cpuset_memory_pressure_bump - keep stats of per-cpuset reclaims. |
| 2303 | * |
| 2304 | * Keep a running average of the rate of synchronous (direct) |
| 2305 | * page reclaim efforts initiated by tasks in each cpuset. |
| 2306 | * |
| 2307 | * This represents the rate at which some task in the cpuset |
| 2308 | * ran low on memory on all nodes it was allowed to use, and |
| 2309 | * had to enter the kernels page reclaim code in an effort to |
| 2310 | * create more free memory by tossing clean pages or swapping |
| 2311 | * or writing dirty pages. |
| 2312 | * |
| 2313 | * Display to user space in the per-cpuset read-only file |
| 2314 | * "memory_pressure". Value displayed is an integer |
| 2315 | * representing the recent rate of entry into the synchronous |
| 2316 | * (direct) page reclaim by any task attached to the cpuset. |
| 2317 | **/ |
| 2318 | |
| 2319 | void __cpuset_memory_pressure_bump(void) |
| 2320 | { |
Paul Jackson | 3e0d98b | 2006-01-08 01:01:49 -0800 | [diff] [blame] | 2321 | task_lock(current); |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 2322 | fmeter_markevent(&task_cs(current)->fmeter); |
Paul Jackson | 3e0d98b | 2006-01-08 01:01:49 -0800 | [diff] [blame] | 2323 | task_unlock(current); |
| 2324 | } |
| 2325 | |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 2326 | #ifdef CONFIG_PROC_PID_CPUSET |
Paul Jackson | 3e0d98b | 2006-01-08 01:01:49 -0800 | [diff] [blame] | 2327 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2328 | * proc_cpuset_show() |
| 2329 | * - Print tasks cpuset path into seq_file. |
| 2330 | * - Used for /proc/<pid>/cpuset. |
Paul Jackson | 053199e | 2005-10-30 15:02:30 -0800 | [diff] [blame] | 2331 | * - No need to task_lock(tsk) on this tsk->cpuset reference, as it |
| 2332 | * doesn't really matter if tsk->cpuset changes after we read it, |
Paul Jackson | c8d9c90 | 2008-02-07 00:14:46 -0800 | [diff] [blame] | 2333 | * and we take cgroup_mutex, keeping cpuset_attach() from changing it |
Paul Menage | 2df167a | 2008-02-07 00:14:45 -0800 | [diff] [blame] | 2334 | * anyway. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2335 | */ |
Paul Jackson | 029190c | 2007-10-18 23:40:20 -0700 | [diff] [blame] | 2336 | static int proc_cpuset_show(struct seq_file *m, void *unused_v) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2337 | { |
Eric W. Biederman | 13b41b0 | 2006-06-26 00:25:56 -0700 | [diff] [blame] | 2338 | struct pid *pid; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2339 | struct task_struct *tsk; |
| 2340 | char *buf; |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 2341 | struct cgroup_subsys_state *css; |
Eric W. Biederman | 99f8955 | 2006-06-26 00:25:55 -0700 | [diff] [blame] | 2342 | int retval; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2343 | |
Eric W. Biederman | 99f8955 | 2006-06-26 00:25:55 -0700 | [diff] [blame] | 2344 | retval = -ENOMEM; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2345 | buf = kmalloc(PAGE_SIZE, GFP_KERNEL); |
| 2346 | if (!buf) |
Eric W. Biederman | 99f8955 | 2006-06-26 00:25:55 -0700 | [diff] [blame] | 2347 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2348 | |
Eric W. Biederman | 99f8955 | 2006-06-26 00:25:55 -0700 | [diff] [blame] | 2349 | retval = -ESRCH; |
Eric W. Biederman | 13b41b0 | 2006-06-26 00:25:56 -0700 | [diff] [blame] | 2350 | pid = m->private; |
| 2351 | tsk = get_pid_task(pid, PIDTYPE_PID); |
Eric W. Biederman | 99f8955 | 2006-06-26 00:25:55 -0700 | [diff] [blame] | 2352 | if (!tsk) |
| 2353 | goto out_free; |
| 2354 | |
| 2355 | retval = -EINVAL; |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 2356 | cgroup_lock(); |
| 2357 | css = task_subsys_state(tsk, cpuset_subsys_id); |
| 2358 | retval = cgroup_path(css->cgroup, buf, PAGE_SIZE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2359 | if (retval < 0) |
Eric W. Biederman | 99f8955 | 2006-06-26 00:25:55 -0700 | [diff] [blame] | 2360 | goto out_unlock; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2361 | seq_puts(m, buf); |
| 2362 | seq_putc(m, '\n'); |
Eric W. Biederman | 99f8955 | 2006-06-26 00:25:55 -0700 | [diff] [blame] | 2363 | out_unlock: |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 2364 | cgroup_unlock(); |
Eric W. Biederman | 99f8955 | 2006-06-26 00:25:55 -0700 | [diff] [blame] | 2365 | put_task_struct(tsk); |
| 2366 | out_free: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2367 | kfree(buf); |
Eric W. Biederman | 99f8955 | 2006-06-26 00:25:55 -0700 | [diff] [blame] | 2368 | out: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2369 | return retval; |
| 2370 | } |
| 2371 | |
| 2372 | static int cpuset_open(struct inode *inode, struct file *file) |
| 2373 | { |
Eric W. Biederman | 13b41b0 | 2006-06-26 00:25:56 -0700 | [diff] [blame] | 2374 | struct pid *pid = PROC_I(inode)->pid; |
| 2375 | return single_open(file, proc_cpuset_show, pid); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2376 | } |
| 2377 | |
Arjan van de Ven | 9a32144 | 2007-02-12 00:55:35 -0800 | [diff] [blame] | 2378 | const struct file_operations proc_cpuset_operations = { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2379 | .open = cpuset_open, |
| 2380 | .read = seq_read, |
| 2381 | .llseek = seq_lseek, |
| 2382 | .release = single_release, |
| 2383 | }; |
Paul Menage | 8793d85 | 2007-10-18 23:39:39 -0700 | [diff] [blame] | 2384 | #endif /* CONFIG_PROC_PID_CPUSET */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2385 | |
| 2386 | /* Display task cpus_allowed, mems_allowed in /proc/<pid>/status file. */ |
Eric W. Biederman | df5f831 | 2008-02-08 04:18:33 -0800 | [diff] [blame] | 2387 | void cpuset_task_status_allowed(struct seq_file *m, struct task_struct *task) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2388 | { |
Eric W. Biederman | df5f831 | 2008-02-08 04:18:33 -0800 | [diff] [blame] | 2389 | seq_printf(m, "Cpus_allowed:\t"); |
| 2390 | m->count += cpumask_scnprintf(m->buf + m->count, m->size - m->count, |
| 2391 | task->cpus_allowed); |
| 2392 | seq_printf(m, "\n"); |
Mike Travis | 39106dc | 2008-04-08 11:43:03 -0700 | [diff] [blame] | 2393 | seq_printf(m, "Cpus_allowed_list:\t"); |
| 2394 | m->count += cpulist_scnprintf(m->buf + m->count, m->size - m->count, |
| 2395 | task->cpus_allowed); |
| 2396 | seq_printf(m, "\n"); |
Eric W. Biederman | df5f831 | 2008-02-08 04:18:33 -0800 | [diff] [blame] | 2397 | seq_printf(m, "Mems_allowed:\t"); |
| 2398 | m->count += nodemask_scnprintf(m->buf + m->count, m->size - m->count, |
| 2399 | task->mems_allowed); |
| 2400 | seq_printf(m, "\n"); |
Mike Travis | 39106dc | 2008-04-08 11:43:03 -0700 | [diff] [blame] | 2401 | seq_printf(m, "Mems_allowed_list:\t"); |
| 2402 | m->count += nodelist_scnprintf(m->buf + m->count, m->size - m->count, |
| 2403 | task->mems_allowed); |
| 2404 | seq_printf(m, "\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2405 | } |