Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * kernel/cpuset.c | |
3 | * | |
4 | * Processor and Memory placement constraints for sets of tasks. | |
5 | * | |
6 | * Copyright (C) 2003 BULL SA. | |
029190c5 | 7 | * Copyright (C) 2004-2007 Silicon Graphics, Inc. |
8793d854 | 8 | * Copyright (C) 2006 Google, Inc |
1da177e4 LT |
9 | * |
10 | * Portions derived from Patrick Mochel's sysfs code. | |
11 | * sysfs is Copyright (c) 2001-3 Patrick Mochel | |
1da177e4 | 12 | * |
825a46af | 13 | * 2003-10-10 Written by Simon Derr. |
1da177e4 | 14 | * 2003-10-22 Updates by Stephen Hemminger. |
825a46af | 15 | * 2004 May-July Rework by Paul Jackson. |
8793d854 | 16 | * 2006 Rework by Paul Menage to use generic cgroups |
cf417141 MK |
17 | * 2008 Rework of the scheduler domains and CPU hotplug handling |
18 | * by Max Krasnyansky | |
1da177e4 LT |
19 | * |
20 | * This file is subject to the terms and conditions of the GNU General Public | |
21 | * License. See the file COPYING in the main directory of the Linux | |
22 | * distribution for more details. | |
23 | */ | |
24 | ||
1da177e4 LT |
25 | #include <linux/cpu.h> |
26 | #include <linux/cpumask.h> | |
27 | #include <linux/cpuset.h> | |
28 | #include <linux/err.h> | |
29 | #include <linux/errno.h> | |
30 | #include <linux/file.h> | |
31 | #include <linux/fs.h> | |
32 | #include <linux/init.h> | |
33 | #include <linux/interrupt.h> | |
34 | #include <linux/kernel.h> | |
35 | #include <linux/kmod.h> | |
36 | #include <linux/list.h> | |
68860ec1 | 37 | #include <linux/mempolicy.h> |
1da177e4 | 38 | #include <linux/mm.h> |
f481891f | 39 | #include <linux/memory.h> |
1da177e4 LT |
40 | #include <linux/module.h> |
41 | #include <linux/mount.h> | |
42 | #include <linux/namei.h> | |
43 | #include <linux/pagemap.h> | |
44 | #include <linux/proc_fs.h> | |
6b9c2603 | 45 | #include <linux/rcupdate.h> |
1da177e4 LT |
46 | #include <linux/sched.h> |
47 | #include <linux/seq_file.h> | |
22fb52dd | 48 | #include <linux/security.h> |
1da177e4 | 49 | #include <linux/slab.h> |
1da177e4 LT |
50 | #include <linux/spinlock.h> |
51 | #include <linux/stat.h> | |
52 | #include <linux/string.h> | |
53 | #include <linux/time.h> | |
54 | #include <linux/backing-dev.h> | |
55 | #include <linux/sort.h> | |
56 | ||
57 | #include <asm/uaccess.h> | |
58 | #include <asm/atomic.h> | |
3d3f26a7 | 59 | #include <linux/mutex.h> |
956db3ca CW |
60 | #include <linux/workqueue.h> |
61 | #include <linux/cgroup.h> | |
1da177e4 | 62 | |
f90d4118 MX |
63 | /* |
64 | * Workqueue for cpuset related tasks. | |
65 | * | |
66 | * Using kevent workqueue may cause deadlock when memory_migrate | |
67 | * is set. So we create a separate workqueue thread for cpuset. | |
68 | */ | |
69 | static struct workqueue_struct *cpuset_wq; | |
70 | ||
202f72d5 PJ |
71 | /* |
72 | * Tracks how many cpusets are currently defined in system. | |
73 | * When there is only one cpuset (the root cpuset) we can | |
74 | * short circuit some hooks. | |
75 | */ | |
7edc5962 | 76 | int number_of_cpusets __read_mostly; |
202f72d5 | 77 | |
2df167a3 | 78 | /* Forward declare cgroup structures */ |
8793d854 PM |
79 | struct cgroup_subsys cpuset_subsys; |
80 | struct cpuset; | |
81 | ||
3e0d98b9 PJ |
82 | /* See "Frequency meter" comments, below. */ |
83 | ||
84 | struct fmeter { | |
85 | int cnt; /* unprocessed events count */ | |
86 | int val; /* most recent output value */ | |
87 | time_t time; /* clock (secs) when val computed */ | |
88 | spinlock_t lock; /* guards read or write of above */ | |
89 | }; | |
90 | ||
1da177e4 | 91 | struct cpuset { |
8793d854 PM |
92 | struct cgroup_subsys_state css; |
93 | ||
1da177e4 | 94 | unsigned long flags; /* "unsigned long" so bitops work */ |
300ed6cb | 95 | cpumask_var_t cpus_allowed; /* CPUs allowed to tasks in cpuset */ |
1da177e4 LT |
96 | nodemask_t mems_allowed; /* Memory Nodes allowed to tasks */ |
97 | ||
1da177e4 | 98 | struct cpuset *parent; /* my parent */ |
1da177e4 | 99 | |
3e0d98b9 | 100 | struct fmeter fmeter; /* memory_pressure filter */ |
029190c5 PJ |
101 | |
102 | /* partition number for rebuild_sched_domains() */ | |
103 | int pn; | |
956db3ca | 104 | |
1d3504fc HS |
105 | /* for custom sched domain */ |
106 | int relax_domain_level; | |
107 | ||
956db3ca CW |
108 | /* used for walking a cpuset heirarchy */ |
109 | struct list_head stack_list; | |
1da177e4 LT |
110 | }; |
111 | ||
8793d854 PM |
112 | /* Retrieve the cpuset for a cgroup */ |
113 | static inline struct cpuset *cgroup_cs(struct cgroup *cont) | |
114 | { | |
115 | return container_of(cgroup_subsys_state(cont, cpuset_subsys_id), | |
116 | struct cpuset, css); | |
117 | } | |
118 | ||
119 | /* Retrieve the cpuset for a task */ | |
120 | static inline struct cpuset *task_cs(struct task_struct *task) | |
121 | { | |
122 | return container_of(task_subsys_state(task, cpuset_subsys_id), | |
123 | struct cpuset, css); | |
124 | } | |
8793d854 | 125 | |
1da177e4 LT |
126 | /* bits in struct cpuset flags field */ |
127 | typedef enum { | |
128 | CS_CPU_EXCLUSIVE, | |
129 | CS_MEM_EXCLUSIVE, | |
78608366 | 130 | CS_MEM_HARDWALL, |
45b07ef3 | 131 | CS_MEMORY_MIGRATE, |
029190c5 | 132 | CS_SCHED_LOAD_BALANCE, |
825a46af PJ |
133 | CS_SPREAD_PAGE, |
134 | CS_SPREAD_SLAB, | |
1da177e4 LT |
135 | } cpuset_flagbits_t; |
136 | ||
137 | /* convenient tests for these bits */ | |
138 | static inline int is_cpu_exclusive(const struct cpuset *cs) | |
139 | { | |
7b5b9ef0 | 140 | return test_bit(CS_CPU_EXCLUSIVE, &cs->flags); |
1da177e4 LT |
141 | } |
142 | ||
143 | static inline int is_mem_exclusive(const struct cpuset *cs) | |
144 | { | |
7b5b9ef0 | 145 | return test_bit(CS_MEM_EXCLUSIVE, &cs->flags); |
1da177e4 LT |
146 | } |
147 | ||
78608366 PM |
148 | static inline int is_mem_hardwall(const struct cpuset *cs) |
149 | { | |
150 | return test_bit(CS_MEM_HARDWALL, &cs->flags); | |
151 | } | |
152 | ||
029190c5 PJ |
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 | ||
45b07ef3 PJ |
158 | static inline int is_memory_migrate(const struct cpuset *cs) |
159 | { | |
7b5b9ef0 | 160 | return test_bit(CS_MEMORY_MIGRATE, &cs->flags); |
45b07ef3 PJ |
161 | } |
162 | ||
825a46af PJ |
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 | ||
1da177e4 LT |
173 | static struct cpuset top_cpuset = { |
174 | .flags = ((1 << CS_CPU_EXCLUSIVE) | (1 << CS_MEM_EXCLUSIVE)), | |
1da177e4 LT |
175 | }; |
176 | ||
1da177e4 | 177 | /* |
2df167a3 PM |
178 | * There are two global mutexes guarding cpuset structures. The first |
179 | * is the main control groups cgroup_mutex, accessed via | |
180 | * cgroup_lock()/cgroup_unlock(). The second is the cpuset-specific | |
181 | * callback_mutex, below. They can nest. It is ok to first take | |
182 | * cgroup_mutex, then nest callback_mutex. We also require taking | |
183 | * task_lock() when dereferencing a task's cpuset pointer. See "The | |
184 | * task_lock() exception", at the end of this comment. | |
053199ed | 185 | * |
3d3f26a7 | 186 | * A task must hold both mutexes to modify cpusets. If a task |
2df167a3 | 187 | * holds cgroup_mutex, then it blocks others wanting that mutex, |
3d3f26a7 | 188 | * ensuring that it is the only task able to also acquire callback_mutex |
053199ed PJ |
189 | * and be able to modify cpusets. It can perform various checks on |
190 | * the cpuset structure first, knowing nothing will change. It can | |
2df167a3 | 191 | * also allocate memory while just holding cgroup_mutex. While it is |
053199ed | 192 | * performing these checks, various callback routines can briefly |
3d3f26a7 IM |
193 | * acquire callback_mutex to query cpusets. Once it is ready to make |
194 | * the changes, it takes callback_mutex, blocking everyone else. | |
053199ed PJ |
195 | * |
196 | * Calls to the kernel memory allocator can not be made while holding | |
3d3f26a7 | 197 | * callback_mutex, as that would risk double tripping on callback_mutex |
053199ed PJ |
198 | * from one of the callbacks into the cpuset code from within |
199 | * __alloc_pages(). | |
200 | * | |
3d3f26a7 | 201 | * If a task is only holding callback_mutex, then it has read-only |
053199ed PJ |
202 | * access to cpusets. |
203 | * | |
58568d2a MX |
204 | * Now, the task_struct fields mems_allowed and mempolicy may be changed |
205 | * by other task, we use alloc_lock in the task_struct fields to protect | |
206 | * them. | |
053199ed | 207 | * |
3d3f26a7 | 208 | * The cpuset_common_file_read() handlers only hold callback_mutex across |
053199ed PJ |
209 | * small pieces of code, such as when reading out possibly multi-word |
210 | * cpumasks and nodemasks. | |
211 | * | |
2df167a3 PM |
212 | * Accessing a task's cpuset should be done in accordance with the |
213 | * guidelines for accessing subsystem state in kernel/cgroup.c | |
1da177e4 LT |
214 | */ |
215 | ||
3d3f26a7 | 216 | static DEFINE_MUTEX(callback_mutex); |
4247bdc6 | 217 | |
75aa1994 DR |
218 | /* |
219 | * cpuset_buffer_lock protects both the cpuset_name and cpuset_nodelist | |
220 | * buffers. They are statically allocated to prevent using excess stack | |
221 | * when calling cpuset_print_task_mems_allowed(). | |
222 | */ | |
223 | #define CPUSET_NAME_LEN (128) | |
224 | #define CPUSET_NODELIST_LEN (256) | |
225 | static char cpuset_name[CPUSET_NAME_LEN]; | |
226 | static char cpuset_nodelist[CPUSET_NODELIST_LEN]; | |
227 | static DEFINE_SPINLOCK(cpuset_buffer_lock); | |
228 | ||
cf417141 MK |
229 | /* |
230 | * This is ugly, but preserves the userspace API for existing cpuset | |
8793d854 | 231 | * users. If someone tries to mount the "cpuset" filesystem, we |
cf417141 MK |
232 | * silently switch it to mount "cgroup" instead |
233 | */ | |
454e2398 DH |
234 | static int cpuset_get_sb(struct file_system_type *fs_type, |
235 | int flags, const char *unused_dev_name, | |
236 | void *data, struct vfsmount *mnt) | |
1da177e4 | 237 | { |
8793d854 PM |
238 | struct file_system_type *cgroup_fs = get_fs_type("cgroup"); |
239 | int ret = -ENODEV; | |
240 | if (cgroup_fs) { | |
241 | char mountopts[] = | |
242 | "cpuset,noprefix," | |
243 | "release_agent=/sbin/cpuset_release_agent"; | |
244 | ret = cgroup_fs->get_sb(cgroup_fs, flags, | |
245 | unused_dev_name, mountopts, mnt); | |
246 | put_filesystem(cgroup_fs); | |
247 | } | |
248 | return ret; | |
1da177e4 LT |
249 | } |
250 | ||
251 | static struct file_system_type cpuset_fs_type = { | |
252 | .name = "cpuset", | |
253 | .get_sb = cpuset_get_sb, | |
1da177e4 LT |
254 | }; |
255 | ||
1da177e4 | 256 | /* |
300ed6cb | 257 | * Return in pmask the portion of a cpusets's cpus_allowed that |
1da177e4 LT |
258 | * are online. If none are online, walk up the cpuset hierarchy |
259 | * until we find one that does have some online cpus. If we get | |
260 | * all the way to the top and still haven't found any online cpus, | |
261 | * return cpu_online_map. Or if passed a NULL cs from an exit'ing | |
262 | * task, return cpu_online_map. | |
263 | * | |
264 | * One way or another, we guarantee to return some non-empty subset | |
265 | * of cpu_online_map. | |
266 | * | |
3d3f26a7 | 267 | * Call with callback_mutex held. |
1da177e4 LT |
268 | */ |
269 | ||
6af866af LZ |
270 | static void guarantee_online_cpus(const struct cpuset *cs, |
271 | struct cpumask *pmask) | |
1da177e4 | 272 | { |
300ed6cb | 273 | while (cs && !cpumask_intersects(cs->cpus_allowed, cpu_online_mask)) |
1da177e4 LT |
274 | cs = cs->parent; |
275 | if (cs) | |
300ed6cb | 276 | cpumask_and(pmask, cs->cpus_allowed, cpu_online_mask); |
1da177e4 | 277 | else |
300ed6cb LZ |
278 | cpumask_copy(pmask, cpu_online_mask); |
279 | BUG_ON(!cpumask_intersects(pmask, cpu_online_mask)); | |
1da177e4 LT |
280 | } |
281 | ||
282 | /* | |
283 | * Return in *pmask the portion of a cpusets's mems_allowed that | |
0e1e7c7a CL |
284 | * are online, with memory. If none are online with memory, walk |
285 | * up the cpuset hierarchy until we find one that does have some | |
286 | * online mems. If we get all the way to the top and still haven't | |
287 | * found any online mems, return node_states[N_HIGH_MEMORY]. | |
1da177e4 LT |
288 | * |
289 | * One way or another, we guarantee to return some non-empty subset | |
0e1e7c7a | 290 | * of node_states[N_HIGH_MEMORY]. |
1da177e4 | 291 | * |
3d3f26a7 | 292 | * Call with callback_mutex held. |
1da177e4 LT |
293 | */ |
294 | ||
295 | static void guarantee_online_mems(const struct cpuset *cs, nodemask_t *pmask) | |
296 | { | |
0e1e7c7a CL |
297 | while (cs && !nodes_intersects(cs->mems_allowed, |
298 | node_states[N_HIGH_MEMORY])) | |
1da177e4 LT |
299 | cs = cs->parent; |
300 | if (cs) | |
0e1e7c7a CL |
301 | nodes_and(*pmask, cs->mems_allowed, |
302 | node_states[N_HIGH_MEMORY]); | |
1da177e4 | 303 | else |
0e1e7c7a CL |
304 | *pmask = node_states[N_HIGH_MEMORY]; |
305 | BUG_ON(!nodes_intersects(*pmask, node_states[N_HIGH_MEMORY])); | |
1da177e4 LT |
306 | } |
307 | ||
f3b39d47 MX |
308 | /* |
309 | * update task's spread flag if cpuset's page/slab spread flag is set | |
310 | * | |
311 | * Called with callback_mutex/cgroup_mutex held | |
312 | */ | |
313 | static void cpuset_update_task_spread_flag(struct cpuset *cs, | |
314 | struct task_struct *tsk) | |
315 | { | |
316 | if (is_spread_page(cs)) | |
317 | tsk->flags |= PF_SPREAD_PAGE; | |
318 | else | |
319 | tsk->flags &= ~PF_SPREAD_PAGE; | |
320 | if (is_spread_slab(cs)) | |
321 | tsk->flags |= PF_SPREAD_SLAB; | |
322 | else | |
323 | tsk->flags &= ~PF_SPREAD_SLAB; | |
324 | } | |
325 | ||
1da177e4 LT |
326 | /* |
327 | * is_cpuset_subset(p, q) - Is cpuset p a subset of cpuset q? | |
328 | * | |
329 | * One cpuset is a subset of another if all its allowed CPUs and | |
330 | * Memory Nodes are a subset of the other, and its exclusive flags | |
2df167a3 | 331 | * are only set if the other's are set. Call holding cgroup_mutex. |
1da177e4 LT |
332 | */ |
333 | ||
334 | static int is_cpuset_subset(const struct cpuset *p, const struct cpuset *q) | |
335 | { | |
300ed6cb | 336 | return cpumask_subset(p->cpus_allowed, q->cpus_allowed) && |
1da177e4 LT |
337 | nodes_subset(p->mems_allowed, q->mems_allowed) && |
338 | is_cpu_exclusive(p) <= is_cpu_exclusive(q) && | |
339 | is_mem_exclusive(p) <= is_mem_exclusive(q); | |
340 | } | |
341 | ||
645fcc9d LZ |
342 | /** |
343 | * alloc_trial_cpuset - allocate a trial cpuset | |
344 | * @cs: the cpuset that the trial cpuset duplicates | |
345 | */ | |
346 | static struct cpuset *alloc_trial_cpuset(const struct cpuset *cs) | |
347 | { | |
300ed6cb LZ |
348 | struct cpuset *trial; |
349 | ||
350 | trial = kmemdup(cs, sizeof(*cs), GFP_KERNEL); | |
351 | if (!trial) | |
352 | return NULL; | |
353 | ||
354 | if (!alloc_cpumask_var(&trial->cpus_allowed, GFP_KERNEL)) { | |
355 | kfree(trial); | |
356 | return NULL; | |
357 | } | |
358 | cpumask_copy(trial->cpus_allowed, cs->cpus_allowed); | |
359 | ||
360 | return trial; | |
645fcc9d LZ |
361 | } |
362 | ||
363 | /** | |
364 | * free_trial_cpuset - free the trial cpuset | |
365 | * @trial: the trial cpuset to be freed | |
366 | */ | |
367 | static void free_trial_cpuset(struct cpuset *trial) | |
368 | { | |
300ed6cb | 369 | free_cpumask_var(trial->cpus_allowed); |
645fcc9d LZ |
370 | kfree(trial); |
371 | } | |
372 | ||
1da177e4 LT |
373 | /* |
374 | * validate_change() - Used to validate that any proposed cpuset change | |
375 | * follows the structural rules for cpusets. | |
376 | * | |
377 | * If we replaced the flag and mask values of the current cpuset | |
378 | * (cur) with those values in the trial cpuset (trial), would | |
379 | * our various subset and exclusive rules still be valid? Presumes | |
2df167a3 | 380 | * cgroup_mutex held. |
1da177e4 LT |
381 | * |
382 | * 'cur' is the address of an actual, in-use cpuset. Operations | |
383 | * such as list traversal that depend on the actual address of the | |
384 | * cpuset in the list must use cur below, not trial. | |
385 | * | |
386 | * 'trial' is the address of bulk structure copy of cur, with | |
387 | * perhaps one or more of the fields cpus_allowed, mems_allowed, | |
388 | * or flags changed to new, trial values. | |
389 | * | |
390 | * Return 0 if valid, -errno if not. | |
391 | */ | |
392 | ||
393 | static int validate_change(const struct cpuset *cur, const struct cpuset *trial) | |
394 | { | |
8793d854 | 395 | struct cgroup *cont; |
1da177e4 LT |
396 | struct cpuset *c, *par; |
397 | ||
398 | /* Each of our child cpusets must be a subset of us */ | |
8793d854 PM |
399 | list_for_each_entry(cont, &cur->css.cgroup->children, sibling) { |
400 | if (!is_cpuset_subset(cgroup_cs(cont), trial)) | |
1da177e4 LT |
401 | return -EBUSY; |
402 | } | |
403 | ||
404 | /* Remaining checks don't apply to root cpuset */ | |
69604067 | 405 | if (cur == &top_cpuset) |
1da177e4 LT |
406 | return 0; |
407 | ||
69604067 PJ |
408 | par = cur->parent; |
409 | ||
1da177e4 LT |
410 | /* We must be a subset of our parent cpuset */ |
411 | if (!is_cpuset_subset(trial, par)) | |
412 | return -EACCES; | |
413 | ||
2df167a3 PM |
414 | /* |
415 | * If either I or some sibling (!= me) is exclusive, we can't | |
416 | * overlap | |
417 | */ | |
8793d854 PM |
418 | list_for_each_entry(cont, &par->css.cgroup->children, sibling) { |
419 | c = cgroup_cs(cont); | |
1da177e4 LT |
420 | if ((is_cpu_exclusive(trial) || is_cpu_exclusive(c)) && |
421 | c != cur && | |
300ed6cb | 422 | cpumask_intersects(trial->cpus_allowed, c->cpus_allowed)) |
1da177e4 LT |
423 | return -EINVAL; |
424 | if ((is_mem_exclusive(trial) || is_mem_exclusive(c)) && | |
425 | c != cur && | |
426 | nodes_intersects(trial->mems_allowed, c->mems_allowed)) | |
427 | return -EINVAL; | |
428 | } | |
429 | ||
020958b6 PJ |
430 | /* Cpusets with tasks can't have empty cpus_allowed or mems_allowed */ |
431 | if (cgroup_task_count(cur->css.cgroup)) { | |
300ed6cb | 432 | if (cpumask_empty(trial->cpus_allowed) || |
020958b6 PJ |
433 | nodes_empty(trial->mems_allowed)) { |
434 | return -ENOSPC; | |
435 | } | |
436 | } | |
437 | ||
1da177e4 LT |
438 | return 0; |
439 | } | |
440 | ||
db7f47cf | 441 | #ifdef CONFIG_SMP |
029190c5 | 442 | /* |
cf417141 | 443 | * Helper routine for generate_sched_domains(). |
029190c5 PJ |
444 | * Do cpusets a, b have overlapping cpus_allowed masks? |
445 | */ | |
029190c5 PJ |
446 | static int cpusets_overlap(struct cpuset *a, struct cpuset *b) |
447 | { | |
300ed6cb | 448 | return cpumask_intersects(a->cpus_allowed, b->cpus_allowed); |
029190c5 PJ |
449 | } |
450 | ||
1d3504fc HS |
451 | static void |
452 | update_domain_attr(struct sched_domain_attr *dattr, struct cpuset *c) | |
453 | { | |
1d3504fc HS |
454 | if (dattr->relax_domain_level < c->relax_domain_level) |
455 | dattr->relax_domain_level = c->relax_domain_level; | |
456 | return; | |
457 | } | |
458 | ||
f5393693 LJ |
459 | static void |
460 | update_domain_attr_tree(struct sched_domain_attr *dattr, struct cpuset *c) | |
461 | { | |
462 | LIST_HEAD(q); | |
463 | ||
464 | list_add(&c->stack_list, &q); | |
465 | while (!list_empty(&q)) { | |
466 | struct cpuset *cp; | |
467 | struct cgroup *cont; | |
468 | struct cpuset *child; | |
469 | ||
470 | cp = list_first_entry(&q, struct cpuset, stack_list); | |
471 | list_del(q.next); | |
472 | ||
300ed6cb | 473 | if (cpumask_empty(cp->cpus_allowed)) |
f5393693 LJ |
474 | continue; |
475 | ||
476 | if (is_sched_load_balance(cp)) | |
477 | update_domain_attr(dattr, cp); | |
478 | ||
479 | list_for_each_entry(cont, &cp->css.cgroup->children, sibling) { | |
480 | child = cgroup_cs(cont); | |
481 | list_add_tail(&child->stack_list, &q); | |
482 | } | |
483 | } | |
484 | } | |
485 | ||
029190c5 | 486 | /* |
cf417141 MK |
487 | * generate_sched_domains() |
488 | * | |
489 | * This function builds a partial partition of the systems CPUs | |
490 | * A 'partial partition' is a set of non-overlapping subsets whose | |
491 | * union is a subset of that set. | |
492 | * The output of this function needs to be passed to kernel/sched.c | |
493 | * partition_sched_domains() routine, which will rebuild the scheduler's | |
494 | * load balancing domains (sched domains) as specified by that partial | |
495 | * partition. | |
029190c5 | 496 | * |
45ce80fb | 497 | * See "What is sched_load_balance" in Documentation/cgroups/cpusets.txt |
029190c5 PJ |
498 | * for a background explanation of this. |
499 | * | |
500 | * Does not return errors, on the theory that the callers of this | |
501 | * routine would rather not worry about failures to rebuild sched | |
502 | * domains when operating in the severe memory shortage situations | |
503 | * that could cause allocation failures below. | |
504 | * | |
cf417141 | 505 | * Must be called with cgroup_lock held. |
029190c5 PJ |
506 | * |
507 | * The three key local variables below are: | |
aeed6824 | 508 | * q - a linked-list queue of cpuset pointers, used to implement a |
029190c5 PJ |
509 | * top-down scan of all cpusets. This scan loads a pointer |
510 | * to each cpuset marked is_sched_load_balance into the | |
511 | * array 'csa'. For our purposes, rebuilding the schedulers | |
512 | * sched domains, we can ignore !is_sched_load_balance cpusets. | |
513 | * csa - (for CpuSet Array) Array of pointers to all the cpusets | |
514 | * that need to be load balanced, for convenient iterative | |
515 | * access by the subsequent code that finds the best partition, | |
516 | * i.e the set of domains (subsets) of CPUs such that the | |
517 | * cpus_allowed of every cpuset marked is_sched_load_balance | |
518 | * is a subset of one of these domains, while there are as | |
519 | * many such domains as possible, each as small as possible. | |
520 | * doms - Conversion of 'csa' to an array of cpumasks, for passing to | |
521 | * the kernel/sched.c routine partition_sched_domains() in a | |
522 | * convenient format, that can be easily compared to the prior | |
523 | * value to determine what partition elements (sched domains) | |
524 | * were changed (added or removed.) | |
525 | * | |
526 | * Finding the best partition (set of domains): | |
527 | * The triple nested loops below over i, j, k scan over the | |
528 | * load balanced cpusets (using the array of cpuset pointers in | |
529 | * csa[]) looking for pairs of cpusets that have overlapping | |
530 | * cpus_allowed, but which don't have the same 'pn' partition | |
531 | * number and gives them in the same partition number. It keeps | |
532 | * looping on the 'restart' label until it can no longer find | |
533 | * any such pairs. | |
534 | * | |
535 | * The union of the cpus_allowed masks from the set of | |
536 | * all cpusets having the same 'pn' value then form the one | |
537 | * element of the partition (one sched domain) to be passed to | |
538 | * partition_sched_domains(). | |
539 | */ | |
acc3f5d7 | 540 | static int generate_sched_domains(cpumask_var_t **domains, |
cf417141 | 541 | struct sched_domain_attr **attributes) |
029190c5 | 542 | { |
cf417141 | 543 | LIST_HEAD(q); /* queue of cpusets to be scanned */ |
029190c5 PJ |
544 | struct cpuset *cp; /* scans q */ |
545 | struct cpuset **csa; /* array of all cpuset ptrs */ | |
546 | int csn; /* how many cpuset ptrs in csa so far */ | |
547 | int i, j, k; /* indices for partition finding loops */ | |
acc3f5d7 | 548 | cpumask_var_t *doms; /* resulting partition; i.e. sched domains */ |
1d3504fc | 549 | struct sched_domain_attr *dattr; /* attributes for custom domains */ |
1583715d | 550 | int ndoms = 0; /* number of sched domains in result */ |
6af866af | 551 | int nslot; /* next empty doms[] struct cpumask slot */ |
029190c5 | 552 | |
029190c5 | 553 | doms = NULL; |
1d3504fc | 554 | dattr = NULL; |
cf417141 | 555 | csa = NULL; |
029190c5 PJ |
556 | |
557 | /* Special case for the 99% of systems with one, full, sched domain */ | |
558 | if (is_sched_load_balance(&top_cpuset)) { | |
acc3f5d7 RR |
559 | ndoms = 1; |
560 | doms = alloc_sched_domains(ndoms); | |
029190c5 | 561 | if (!doms) |
cf417141 MK |
562 | goto done; |
563 | ||
1d3504fc HS |
564 | dattr = kmalloc(sizeof(struct sched_domain_attr), GFP_KERNEL); |
565 | if (dattr) { | |
566 | *dattr = SD_ATTR_INIT; | |
93a65575 | 567 | update_domain_attr_tree(dattr, &top_cpuset); |
1d3504fc | 568 | } |
acc3f5d7 | 569 | cpumask_copy(doms[0], top_cpuset.cpus_allowed); |
cf417141 | 570 | |
cf417141 | 571 | goto done; |
029190c5 PJ |
572 | } |
573 | ||
029190c5 PJ |
574 | csa = kmalloc(number_of_cpusets * sizeof(cp), GFP_KERNEL); |
575 | if (!csa) | |
576 | goto done; | |
577 | csn = 0; | |
578 | ||
aeed6824 LZ |
579 | list_add(&top_cpuset.stack_list, &q); |
580 | while (!list_empty(&q)) { | |
029190c5 PJ |
581 | struct cgroup *cont; |
582 | struct cpuset *child; /* scans child cpusets of cp */ | |
489a5393 | 583 | |
aeed6824 LZ |
584 | cp = list_first_entry(&q, struct cpuset, stack_list); |
585 | list_del(q.next); | |
586 | ||
300ed6cb | 587 | if (cpumask_empty(cp->cpus_allowed)) |
489a5393 LJ |
588 | continue; |
589 | ||
f5393693 LJ |
590 | /* |
591 | * All child cpusets contain a subset of the parent's cpus, so | |
592 | * just skip them, and then we call update_domain_attr_tree() | |
593 | * to calc relax_domain_level of the corresponding sched | |
594 | * domain. | |
595 | */ | |
596 | if (is_sched_load_balance(cp)) { | |
029190c5 | 597 | csa[csn++] = cp; |
f5393693 LJ |
598 | continue; |
599 | } | |
489a5393 | 600 | |
029190c5 PJ |
601 | list_for_each_entry(cont, &cp->css.cgroup->children, sibling) { |
602 | child = cgroup_cs(cont); | |
aeed6824 | 603 | list_add_tail(&child->stack_list, &q); |
029190c5 PJ |
604 | } |
605 | } | |
606 | ||
607 | for (i = 0; i < csn; i++) | |
608 | csa[i]->pn = i; | |
609 | ndoms = csn; | |
610 | ||
611 | restart: | |
612 | /* Find the best partition (set of sched domains) */ | |
613 | for (i = 0; i < csn; i++) { | |
614 | struct cpuset *a = csa[i]; | |
615 | int apn = a->pn; | |
616 | ||
617 | for (j = 0; j < csn; j++) { | |
618 | struct cpuset *b = csa[j]; | |
619 | int bpn = b->pn; | |
620 | ||
621 | if (apn != bpn && cpusets_overlap(a, b)) { | |
622 | for (k = 0; k < csn; k++) { | |
623 | struct cpuset *c = csa[k]; | |
624 | ||
625 | if (c->pn == bpn) | |
626 | c->pn = apn; | |
627 | } | |
628 | ndoms--; /* one less element */ | |
629 | goto restart; | |
630 | } | |
631 | } | |
632 | } | |
633 | ||
cf417141 MK |
634 | /* |
635 | * Now we know how many domains to create. | |
636 | * Convert <csn, csa> to <ndoms, doms> and populate cpu masks. | |
637 | */ | |
acc3f5d7 | 638 | doms = alloc_sched_domains(ndoms); |
700018e0 | 639 | if (!doms) |
cf417141 | 640 | goto done; |
cf417141 MK |
641 | |
642 | /* | |
643 | * The rest of the code, including the scheduler, can deal with | |
644 | * dattr==NULL case. No need to abort if alloc fails. | |
645 | */ | |
1d3504fc | 646 | dattr = kmalloc(ndoms * sizeof(struct sched_domain_attr), GFP_KERNEL); |
029190c5 PJ |
647 | |
648 | for (nslot = 0, i = 0; i < csn; i++) { | |
649 | struct cpuset *a = csa[i]; | |
6af866af | 650 | struct cpumask *dp; |
029190c5 PJ |
651 | int apn = a->pn; |
652 | ||
cf417141 MK |
653 | if (apn < 0) { |
654 | /* Skip completed partitions */ | |
655 | continue; | |
656 | } | |
657 | ||
acc3f5d7 | 658 | dp = doms[nslot]; |
cf417141 MK |
659 | |
660 | if (nslot == ndoms) { | |
661 | static int warnings = 10; | |
662 | if (warnings) { | |
663 | printk(KERN_WARNING | |
664 | "rebuild_sched_domains confused:" | |
665 | " nslot %d, ndoms %d, csn %d, i %d," | |
666 | " apn %d\n", | |
667 | nslot, ndoms, csn, i, apn); | |
668 | warnings--; | |
029190c5 | 669 | } |
cf417141 MK |
670 | continue; |
671 | } | |
029190c5 | 672 | |
6af866af | 673 | cpumask_clear(dp); |
cf417141 MK |
674 | if (dattr) |
675 | *(dattr + nslot) = SD_ATTR_INIT; | |
676 | for (j = i; j < csn; j++) { | |
677 | struct cpuset *b = csa[j]; | |
678 | ||
679 | if (apn == b->pn) { | |
300ed6cb | 680 | cpumask_or(dp, dp, b->cpus_allowed); |
cf417141 MK |
681 | if (dattr) |
682 | update_domain_attr_tree(dattr + nslot, b); | |
683 | ||
684 | /* Done with this partition */ | |
685 | b->pn = -1; | |
029190c5 | 686 | } |
029190c5 | 687 | } |
cf417141 | 688 | nslot++; |
029190c5 PJ |
689 | } |
690 | BUG_ON(nslot != ndoms); | |
691 | ||
cf417141 MK |
692 | done: |
693 | kfree(csa); | |
694 | ||
700018e0 LZ |
695 | /* |
696 | * Fallback to the default domain if kmalloc() failed. | |
697 | * See comments in partition_sched_domains(). | |
698 | */ | |
699 | if (doms == NULL) | |
700 | ndoms = 1; | |
701 | ||
cf417141 MK |
702 | *domains = doms; |
703 | *attributes = dattr; | |
704 | return ndoms; | |
705 | } | |
706 | ||
707 | /* | |
708 | * Rebuild scheduler domains. | |
709 | * | |
710 | * Call with neither cgroup_mutex held nor within get_online_cpus(). | |
711 | * Takes both cgroup_mutex and get_online_cpus(). | |
712 | * | |
713 | * Cannot be directly called from cpuset code handling changes | |
714 | * to the cpuset pseudo-filesystem, because it cannot be called | |
715 | * from code that already holds cgroup_mutex. | |
716 | */ | |
717 | static void do_rebuild_sched_domains(struct work_struct *unused) | |
718 | { | |
719 | struct sched_domain_attr *attr; | |
acc3f5d7 | 720 | cpumask_var_t *doms; |
cf417141 MK |
721 | int ndoms; |
722 | ||
86ef5c9a | 723 | get_online_cpus(); |
cf417141 MK |
724 | |
725 | /* Generate domain masks and attrs */ | |
726 | cgroup_lock(); | |
727 | ndoms = generate_sched_domains(&doms, &attr); | |
728 | cgroup_unlock(); | |
729 | ||
730 | /* Have scheduler rebuild the domains */ | |
731 | partition_sched_domains(ndoms, doms, attr); | |
732 | ||
86ef5c9a | 733 | put_online_cpus(); |
cf417141 | 734 | } |
db7f47cf PM |
735 | #else /* !CONFIG_SMP */ |
736 | static void do_rebuild_sched_domains(struct work_struct *unused) | |
737 | { | |
738 | } | |
739 | ||
e1b8090b | 740 | static int generate_sched_domains(cpumask_var_t **domains, |
db7f47cf PM |
741 | struct sched_domain_attr **attributes) |
742 | { | |
743 | *domains = NULL; | |
744 | return 1; | |
745 | } | |
746 | #endif /* CONFIG_SMP */ | |
029190c5 | 747 | |
cf417141 MK |
748 | static DECLARE_WORK(rebuild_sched_domains_work, do_rebuild_sched_domains); |
749 | ||
750 | /* | |
751 | * Rebuild scheduler domains, asynchronously via workqueue. | |
752 | * | |
753 | * If the flag 'sched_load_balance' of any cpuset with non-empty | |
754 | * 'cpus' changes, or if the 'cpus' allowed changes in any cpuset | |
755 | * which has that flag enabled, or if any cpuset with a non-empty | |
756 | * 'cpus' is removed, then call this routine to rebuild the | |
757 | * scheduler's dynamic sched domains. | |
758 | * | |
759 | * The rebuild_sched_domains() and partition_sched_domains() | |
760 | * routines must nest cgroup_lock() inside get_online_cpus(), | |
761 | * but such cpuset changes as these must nest that locking the | |
762 | * other way, holding cgroup_lock() for much of the code. | |
763 | * | |
764 | * So in order to avoid an ABBA deadlock, the cpuset code handling | |
765 | * these user changes delegates the actual sched domain rebuilding | |
766 | * to a separate workqueue thread, which ends up processing the | |
767 | * above do_rebuild_sched_domains() function. | |
768 | */ | |
769 | static void async_rebuild_sched_domains(void) | |
770 | { | |
f90d4118 | 771 | queue_work(cpuset_wq, &rebuild_sched_domains_work); |
cf417141 MK |
772 | } |
773 | ||
774 | /* | |
775 | * Accomplishes the same scheduler domain rebuild as the above | |
776 | * async_rebuild_sched_domains(), however it directly calls the | |
777 | * rebuild routine synchronously rather than calling it via an | |
778 | * asynchronous work thread. | |
779 | * | |
780 | * This can only be called from code that is not holding | |
781 | * cgroup_mutex (not nested in a cgroup_lock() call.) | |
782 | */ | |
783 | void rebuild_sched_domains(void) | |
784 | { | |
785 | do_rebuild_sched_domains(NULL); | |
029190c5 PJ |
786 | } |
787 | ||
58f4790b CW |
788 | /** |
789 | * cpuset_test_cpumask - test a task's cpus_allowed versus its cpuset's | |
790 | * @tsk: task to test | |
791 | * @scan: struct cgroup_scanner contained in its struct cpuset_hotplug_scanner | |
792 | * | |
2df167a3 | 793 | * Call with cgroup_mutex held. May take callback_mutex during call. |
58f4790b CW |
794 | * Called for each task in a cgroup by cgroup_scan_tasks(). |
795 | * Return nonzero if this tasks's cpus_allowed mask should be changed (in other | |
796 | * words, if its mask is not equal to its cpuset's mask). | |
053199ed | 797 | */ |
9e0c914c AB |
798 | static int cpuset_test_cpumask(struct task_struct *tsk, |
799 | struct cgroup_scanner *scan) | |
58f4790b | 800 | { |
300ed6cb | 801 | return !cpumask_equal(&tsk->cpus_allowed, |
58f4790b CW |
802 | (cgroup_cs(scan->cg))->cpus_allowed); |
803 | } | |
053199ed | 804 | |
58f4790b CW |
805 | /** |
806 | * cpuset_change_cpumask - make a task's cpus_allowed the same as its cpuset's | |
807 | * @tsk: task to test | |
808 | * @scan: struct cgroup_scanner containing the cgroup of the task | |
809 | * | |
810 | * Called by cgroup_scan_tasks() for each task in a cgroup whose | |
811 | * cpus_allowed mask needs to be changed. | |
812 | * | |
813 | * We don't need to re-check for the cgroup/cpuset membership, since we're | |
814 | * holding cgroup_lock() at this point. | |
815 | */ | |
9e0c914c AB |
816 | static void cpuset_change_cpumask(struct task_struct *tsk, |
817 | struct cgroup_scanner *scan) | |
58f4790b | 818 | { |
300ed6cb | 819 | set_cpus_allowed_ptr(tsk, ((cgroup_cs(scan->cg))->cpus_allowed)); |
58f4790b CW |
820 | } |
821 | ||
0b2f630a MX |
822 | /** |
823 | * update_tasks_cpumask - Update the cpumasks of tasks in the cpuset. | |
824 | * @cs: the cpuset in which each task's cpus_allowed mask needs to be changed | |
4e74339a | 825 | * @heap: if NULL, defer allocating heap memory to cgroup_scan_tasks() |
0b2f630a MX |
826 | * |
827 | * Called with cgroup_mutex held | |
828 | * | |
829 | * The cgroup_scan_tasks() function will scan all the tasks in a cgroup, | |
830 | * calling callback functions for each. | |
831 | * | |
4e74339a LZ |
832 | * No return value. It's guaranteed that cgroup_scan_tasks() always returns 0 |
833 | * if @heap != NULL. | |
0b2f630a | 834 | */ |
4e74339a | 835 | static void update_tasks_cpumask(struct cpuset *cs, struct ptr_heap *heap) |
0b2f630a MX |
836 | { |
837 | struct cgroup_scanner scan; | |
0b2f630a MX |
838 | |
839 | scan.cg = cs->css.cgroup; | |
840 | scan.test_task = cpuset_test_cpumask; | |
841 | scan.process_task = cpuset_change_cpumask; | |
4e74339a LZ |
842 | scan.heap = heap; |
843 | cgroup_scan_tasks(&scan); | |
0b2f630a MX |
844 | } |
845 | ||
58f4790b CW |
846 | /** |
847 | * update_cpumask - update the cpus_allowed mask of a cpuset and all tasks in it | |
848 | * @cs: the cpuset to consider | |
849 | * @buf: buffer of cpu numbers written to this cpuset | |
850 | */ | |
645fcc9d LZ |
851 | static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs, |
852 | const char *buf) | |
1da177e4 | 853 | { |
4e74339a | 854 | struct ptr_heap heap; |
58f4790b CW |
855 | int retval; |
856 | int is_load_balanced; | |
1da177e4 | 857 | |
4c4d50f7 PJ |
858 | /* top_cpuset.cpus_allowed tracks cpu_online_map; it's read-only */ |
859 | if (cs == &top_cpuset) | |
860 | return -EACCES; | |
861 | ||
6f7f02e7 | 862 | /* |
c8d9c90c | 863 | * An empty cpus_allowed is ok only if the cpuset has no tasks. |
020958b6 PJ |
864 | * Since cpulist_parse() fails on an empty mask, we special case |
865 | * that parsing. The validate_change() call ensures that cpusets | |
866 | * with tasks have cpus. | |
6f7f02e7 | 867 | */ |
020958b6 | 868 | if (!*buf) { |
300ed6cb | 869 | cpumask_clear(trialcs->cpus_allowed); |
6f7f02e7 | 870 | } else { |
300ed6cb | 871 | retval = cpulist_parse(buf, trialcs->cpus_allowed); |
6f7f02e7 DR |
872 | if (retval < 0) |
873 | return retval; | |
37340746 | 874 | |
6ad4c188 | 875 | if (!cpumask_subset(trialcs->cpus_allowed, cpu_active_mask)) |
37340746 | 876 | return -EINVAL; |
6f7f02e7 | 877 | } |
645fcc9d | 878 | retval = validate_change(cs, trialcs); |
85d7b949 DG |
879 | if (retval < 0) |
880 | return retval; | |
029190c5 | 881 | |
8707d8b8 | 882 | /* Nothing to do if the cpus didn't change */ |
300ed6cb | 883 | if (cpumask_equal(cs->cpus_allowed, trialcs->cpus_allowed)) |
8707d8b8 | 884 | return 0; |
58f4790b | 885 | |
4e74339a LZ |
886 | retval = heap_init(&heap, PAGE_SIZE, GFP_KERNEL, NULL); |
887 | if (retval) | |
888 | return retval; | |
889 | ||
645fcc9d | 890 | is_load_balanced = is_sched_load_balance(trialcs); |
029190c5 | 891 | |
3d3f26a7 | 892 | mutex_lock(&callback_mutex); |
300ed6cb | 893 | cpumask_copy(cs->cpus_allowed, trialcs->cpus_allowed); |
3d3f26a7 | 894 | mutex_unlock(&callback_mutex); |
029190c5 | 895 | |
8707d8b8 PM |
896 | /* |
897 | * Scan tasks in the cpuset, and update the cpumasks of any | |
58f4790b | 898 | * that need an update. |
8707d8b8 | 899 | */ |
4e74339a LZ |
900 | update_tasks_cpumask(cs, &heap); |
901 | ||
902 | heap_free(&heap); | |
58f4790b | 903 | |
8707d8b8 | 904 | if (is_load_balanced) |
cf417141 | 905 | async_rebuild_sched_domains(); |
85d7b949 | 906 | return 0; |
1da177e4 LT |
907 | } |
908 | ||
e4e364e8 PJ |
909 | /* |
910 | * cpuset_migrate_mm | |
911 | * | |
912 | * Migrate memory region from one set of nodes to another. | |
913 | * | |
914 | * Temporarilly set tasks mems_allowed to target nodes of migration, | |
915 | * so that the migration code can allocate pages on these nodes. | |
916 | * | |
2df167a3 | 917 | * Call holding cgroup_mutex, so current's cpuset won't change |
c8d9c90c | 918 | * during this call, as manage_mutex holds off any cpuset_attach() |
e4e364e8 PJ |
919 | * calls. Therefore we don't need to take task_lock around the |
920 | * call to guarantee_online_mems(), as we know no one is changing | |
2df167a3 | 921 | * our task's cpuset. |
e4e364e8 PJ |
922 | * |
923 | * Hold callback_mutex around the two modifications of our tasks | |
924 | * mems_allowed to synchronize with cpuset_mems_allowed(). | |
925 | * | |
926 | * While the mm_struct we are migrating is typically from some | |
927 | * other task, the task_struct mems_allowed that we are hacking | |
928 | * is for our current task, which must allocate new pages for that | |
929 | * migrating memory region. | |
e4e364e8 PJ |
930 | */ |
931 | ||
932 | static void cpuset_migrate_mm(struct mm_struct *mm, const nodemask_t *from, | |
933 | const nodemask_t *to) | |
934 | { | |
935 | struct task_struct *tsk = current; | |
936 | ||
e4e364e8 | 937 | tsk->mems_allowed = *to; |
e4e364e8 PJ |
938 | |
939 | do_migrate_pages(mm, from, to, MPOL_MF_MOVE_ALL); | |
940 | ||
8793d854 | 941 | guarantee_online_mems(task_cs(tsk),&tsk->mems_allowed); |
e4e364e8 PJ |
942 | } |
943 | ||
3b6766fe | 944 | /* |
58568d2a MX |
945 | * cpuset_change_task_nodemask - change task's mems_allowed and mempolicy |
946 | * @tsk: the task to change | |
947 | * @newmems: new nodes that the task will be set | |
948 | * | |
949 | * In order to avoid seeing no nodes if the old and new nodes are disjoint, | |
950 | * we structure updates as setting all new allowed nodes, then clearing newly | |
951 | * disallowed ones. | |
952 | * | |
953 | * Called with task's alloc_lock held | |
954 | */ | |
955 | static void cpuset_change_task_nodemask(struct task_struct *tsk, | |
956 | nodemask_t *newmems) | |
957 | { | |
958 | nodes_or(tsk->mems_allowed, tsk->mems_allowed, *newmems); | |
959 | mpol_rebind_task(tsk, &tsk->mems_allowed); | |
960 | mpol_rebind_task(tsk, newmems); | |
961 | tsk->mems_allowed = *newmems; | |
962 | } | |
963 | ||
964 | /* | |
965 | * Update task's mems_allowed and rebind its mempolicy and vmas' mempolicy | |
966 | * of it to cpuset's new mems_allowed, and migrate pages to new nodes if | |
967 | * memory_migrate flag is set. Called with cgroup_mutex held. | |
3b6766fe LZ |
968 | */ |
969 | static void cpuset_change_nodemask(struct task_struct *p, | |
970 | struct cgroup_scanner *scan) | |
971 | { | |
972 | struct mm_struct *mm; | |
973 | struct cpuset *cs; | |
974 | int migrate; | |
975 | const nodemask_t *oldmem = scan->data; | |
58568d2a MX |
976 | nodemask_t newmems; |
977 | ||
978 | cs = cgroup_cs(scan->cg); | |
979 | guarantee_online_mems(cs, &newmems); | |
980 | ||
981 | task_lock(p); | |
982 | cpuset_change_task_nodemask(p, &newmems); | |
983 | task_unlock(p); | |
3b6766fe LZ |
984 | |
985 | mm = get_task_mm(p); | |
986 | if (!mm) | |
987 | return; | |
988 | ||
3b6766fe LZ |
989 | migrate = is_memory_migrate(cs); |
990 | ||
991 | mpol_rebind_mm(mm, &cs->mems_allowed); | |
992 | if (migrate) | |
993 | cpuset_migrate_mm(mm, oldmem, &cs->mems_allowed); | |
994 | mmput(mm); | |
995 | } | |
996 | ||
8793d854 PM |
997 | static void *cpuset_being_rebound; |
998 | ||
0b2f630a MX |
999 | /** |
1000 | * update_tasks_nodemask - Update the nodemasks of tasks in the cpuset. | |
1001 | * @cs: the cpuset in which each task's mems_allowed mask needs to be changed | |
1002 | * @oldmem: old mems_allowed of cpuset cs | |
010cfac4 | 1003 | * @heap: if NULL, defer allocating heap memory to cgroup_scan_tasks() |
0b2f630a MX |
1004 | * |
1005 | * Called with cgroup_mutex held | |
010cfac4 LZ |
1006 | * No return value. It's guaranteed that cgroup_scan_tasks() always returns 0 |
1007 | * if @heap != NULL. | |
0b2f630a | 1008 | */ |
010cfac4 LZ |
1009 | static void update_tasks_nodemask(struct cpuset *cs, const nodemask_t *oldmem, |
1010 | struct ptr_heap *heap) | |
1da177e4 | 1011 | { |
3b6766fe | 1012 | struct cgroup_scanner scan; |
59dac16f | 1013 | |
846a16bf | 1014 | cpuset_being_rebound = cs; /* causes mpol_dup() rebind */ |
4225399a | 1015 | |
3b6766fe LZ |
1016 | scan.cg = cs->css.cgroup; |
1017 | scan.test_task = NULL; | |
1018 | scan.process_task = cpuset_change_nodemask; | |
010cfac4 | 1019 | scan.heap = heap; |
3b6766fe | 1020 | scan.data = (nodemask_t *)oldmem; |
4225399a PJ |
1021 | |
1022 | /* | |
3b6766fe LZ |
1023 | * The mpol_rebind_mm() call takes mmap_sem, which we couldn't |
1024 | * take while holding tasklist_lock. Forks can happen - the | |
1025 | * mpol_dup() cpuset_being_rebound check will catch such forks, | |
1026 | * and rebind their vma mempolicies too. Because we still hold | |
1027 | * the global cgroup_mutex, we know that no other rebind effort | |
1028 | * will be contending for the global variable cpuset_being_rebound. | |
4225399a | 1029 | * It's ok if we rebind the same mm twice; mpol_rebind_mm() |
04c19fa6 | 1030 | * is idempotent. Also migrate pages in each mm to new nodes. |
4225399a | 1031 | */ |
010cfac4 | 1032 | cgroup_scan_tasks(&scan); |
4225399a | 1033 | |
2df167a3 | 1034 | /* We're done rebinding vmas to this cpuset's new mems_allowed. */ |
8793d854 | 1035 | cpuset_being_rebound = NULL; |
1da177e4 LT |
1036 | } |
1037 | ||
0b2f630a MX |
1038 | /* |
1039 | * Handle user request to change the 'mems' memory placement | |
1040 | * of a cpuset. Needs to validate the request, update the | |
58568d2a MX |
1041 | * cpusets mems_allowed, and for each task in the cpuset, |
1042 | * update mems_allowed and rebind task's mempolicy and any vma | |
1043 | * mempolicies and if the cpuset is marked 'memory_migrate', | |
1044 | * migrate the tasks pages to the new memory. | |
0b2f630a MX |
1045 | * |
1046 | * Call with cgroup_mutex held. May take callback_mutex during call. | |
1047 | * Will take tasklist_lock, scan tasklist for tasks in cpuset cs, | |
1048 | * lock each such tasks mm->mmap_sem, scan its vma's and rebind | |
1049 | * their mempolicies to the cpusets new mems_allowed. | |
1050 | */ | |
645fcc9d LZ |
1051 | static int update_nodemask(struct cpuset *cs, struct cpuset *trialcs, |
1052 | const char *buf) | |
0b2f630a | 1053 | { |
0b2f630a MX |
1054 | nodemask_t oldmem; |
1055 | int retval; | |
010cfac4 | 1056 | struct ptr_heap heap; |
0b2f630a MX |
1057 | |
1058 | /* | |
1059 | * top_cpuset.mems_allowed tracks node_stats[N_HIGH_MEMORY]; | |
1060 | * it's read-only | |
1061 | */ | |
1062 | if (cs == &top_cpuset) | |
1063 | return -EACCES; | |
1064 | ||
0b2f630a MX |
1065 | /* |
1066 | * An empty mems_allowed is ok iff there are no tasks in the cpuset. | |
1067 | * Since nodelist_parse() fails on an empty mask, we special case | |
1068 | * that parsing. The validate_change() call ensures that cpusets | |
1069 | * with tasks have memory. | |
1070 | */ | |
1071 | if (!*buf) { | |
645fcc9d | 1072 | nodes_clear(trialcs->mems_allowed); |
0b2f630a | 1073 | } else { |
645fcc9d | 1074 | retval = nodelist_parse(buf, trialcs->mems_allowed); |
0b2f630a MX |
1075 | if (retval < 0) |
1076 | goto done; | |
1077 | ||
645fcc9d | 1078 | if (!nodes_subset(trialcs->mems_allowed, |
0b2f630a MX |
1079 | node_states[N_HIGH_MEMORY])) |
1080 | return -EINVAL; | |
1081 | } | |
1082 | oldmem = cs->mems_allowed; | |
645fcc9d | 1083 | if (nodes_equal(oldmem, trialcs->mems_allowed)) { |
0b2f630a MX |
1084 | retval = 0; /* Too easy - nothing to do */ |
1085 | goto done; | |
1086 | } | |
645fcc9d | 1087 | retval = validate_change(cs, trialcs); |
0b2f630a MX |
1088 | if (retval < 0) |
1089 | goto done; | |
1090 | ||
010cfac4 LZ |
1091 | retval = heap_init(&heap, PAGE_SIZE, GFP_KERNEL, NULL); |
1092 | if (retval < 0) | |
1093 | goto done; | |
1094 | ||
0b2f630a | 1095 | mutex_lock(&callback_mutex); |
645fcc9d | 1096 | cs->mems_allowed = trialcs->mems_allowed; |
0b2f630a MX |
1097 | mutex_unlock(&callback_mutex); |
1098 | ||
010cfac4 LZ |
1099 | update_tasks_nodemask(cs, &oldmem, &heap); |
1100 | ||
1101 | heap_free(&heap); | |
0b2f630a MX |
1102 | done: |
1103 | return retval; | |
1104 | } | |
1105 | ||
8793d854 PM |
1106 | int current_cpuset_is_being_rebound(void) |
1107 | { | |
1108 | return task_cs(current) == cpuset_being_rebound; | |
1109 | } | |
1110 | ||
5be7a479 | 1111 | static int update_relax_domain_level(struct cpuset *cs, s64 val) |
1d3504fc | 1112 | { |
db7f47cf | 1113 | #ifdef CONFIG_SMP |
30e0e178 LZ |
1114 | if (val < -1 || val >= SD_LV_MAX) |
1115 | return -EINVAL; | |
db7f47cf | 1116 | #endif |
1d3504fc HS |
1117 | |
1118 | if (val != cs->relax_domain_level) { | |
1119 | cs->relax_domain_level = val; | |
300ed6cb LZ |
1120 | if (!cpumask_empty(cs->cpus_allowed) && |
1121 | is_sched_load_balance(cs)) | |
cf417141 | 1122 | async_rebuild_sched_domains(); |
1d3504fc HS |
1123 | } |
1124 | ||
1125 | return 0; | |
1126 | } | |
1127 | ||
950592f7 MX |
1128 | /* |
1129 | * cpuset_change_flag - make a task's spread flags the same as its cpuset's | |
1130 | * @tsk: task to be updated | |
1131 | * @scan: struct cgroup_scanner containing the cgroup of the task | |
1132 | * | |
1133 | * Called by cgroup_scan_tasks() for each task in a cgroup. | |
1134 | * | |
1135 | * We don't need to re-check for the cgroup/cpuset membership, since we're | |
1136 | * holding cgroup_lock() at this point. | |
1137 | */ | |
1138 | static void cpuset_change_flag(struct task_struct *tsk, | |
1139 | struct cgroup_scanner *scan) | |
1140 | { | |
1141 | cpuset_update_task_spread_flag(cgroup_cs(scan->cg), tsk); | |
1142 | } | |
1143 | ||
1144 | /* | |
1145 | * update_tasks_flags - update the spread flags of tasks in the cpuset. | |
1146 | * @cs: the cpuset in which each task's spread flags needs to be changed | |
1147 | * @heap: if NULL, defer allocating heap memory to cgroup_scan_tasks() | |
1148 | * | |
1149 | * Called with cgroup_mutex held | |
1150 | * | |
1151 | * The cgroup_scan_tasks() function will scan all the tasks in a cgroup, | |
1152 | * calling callback functions for each. | |
1153 | * | |
1154 | * No return value. It's guaranteed that cgroup_scan_tasks() always returns 0 | |
1155 | * if @heap != NULL. | |
1156 | */ | |
1157 | static void update_tasks_flags(struct cpuset *cs, struct ptr_heap *heap) | |
1158 | { | |
1159 | struct cgroup_scanner scan; | |
1160 | ||
1161 | scan.cg = cs->css.cgroup; | |
1162 | scan.test_task = NULL; | |
1163 | scan.process_task = cpuset_change_flag; | |
1164 | scan.heap = heap; | |
1165 | cgroup_scan_tasks(&scan); | |
1166 | } | |
1167 | ||
1da177e4 LT |
1168 | /* |
1169 | * update_flag - read a 0 or a 1 in a file and update associated flag | |
78608366 PM |
1170 | * bit: the bit to update (see cpuset_flagbits_t) |
1171 | * cs: the cpuset to update | |
1172 | * turning_on: whether the flag is being set or cleared | |
053199ed | 1173 | * |
2df167a3 | 1174 | * Call with cgroup_mutex held. |
1da177e4 LT |
1175 | */ |
1176 | ||
700fe1ab PM |
1177 | static int update_flag(cpuset_flagbits_t bit, struct cpuset *cs, |
1178 | int turning_on) | |
1da177e4 | 1179 | { |
645fcc9d | 1180 | struct cpuset *trialcs; |
40b6a762 | 1181 | int balance_flag_changed; |
950592f7 MX |
1182 | int spread_flag_changed; |
1183 | struct ptr_heap heap; | |
1184 | int err; | |
1da177e4 | 1185 | |
645fcc9d LZ |
1186 | trialcs = alloc_trial_cpuset(cs); |
1187 | if (!trialcs) | |
1188 | return -ENOMEM; | |
1189 | ||
1da177e4 | 1190 | if (turning_on) |
645fcc9d | 1191 | set_bit(bit, &trialcs->flags); |
1da177e4 | 1192 | else |
645fcc9d | 1193 | clear_bit(bit, &trialcs->flags); |
1da177e4 | 1194 | |
645fcc9d | 1195 | err = validate_change(cs, trialcs); |
85d7b949 | 1196 | if (err < 0) |
645fcc9d | 1197 | goto out; |
029190c5 | 1198 | |
950592f7 MX |
1199 | err = heap_init(&heap, PAGE_SIZE, GFP_KERNEL, NULL); |
1200 | if (err < 0) | |
1201 | goto out; | |
1202 | ||
029190c5 | 1203 | balance_flag_changed = (is_sched_load_balance(cs) != |
645fcc9d | 1204 | is_sched_load_balance(trialcs)); |
029190c5 | 1205 | |
950592f7 MX |
1206 | spread_flag_changed = ((is_spread_slab(cs) != is_spread_slab(trialcs)) |
1207 | || (is_spread_page(cs) != is_spread_page(trialcs))); | |
1208 | ||
3d3f26a7 | 1209 | mutex_lock(&callback_mutex); |
645fcc9d | 1210 | cs->flags = trialcs->flags; |
3d3f26a7 | 1211 | mutex_unlock(&callback_mutex); |
85d7b949 | 1212 | |
300ed6cb | 1213 | if (!cpumask_empty(trialcs->cpus_allowed) && balance_flag_changed) |
cf417141 | 1214 | async_rebuild_sched_domains(); |
029190c5 | 1215 | |
950592f7 MX |
1216 | if (spread_flag_changed) |
1217 | update_tasks_flags(cs, &heap); | |
1218 | heap_free(&heap); | |
645fcc9d LZ |
1219 | out: |
1220 | free_trial_cpuset(trialcs); | |
1221 | return err; | |
1da177e4 LT |
1222 | } |
1223 | ||
3e0d98b9 | 1224 | /* |
80f7228b | 1225 | * Frequency meter - How fast is some event occurring? |
3e0d98b9 PJ |
1226 | * |
1227 | * These routines manage a digitally filtered, constant time based, | |
1228 | * event frequency meter. There are four routines: | |
1229 | * fmeter_init() - initialize a frequency meter. | |
1230 | * fmeter_markevent() - called each time the event happens. | |
1231 | * fmeter_getrate() - returns the recent rate of such events. | |
1232 | * fmeter_update() - internal routine used to update fmeter. | |
1233 | * | |
1234 | * A common data structure is passed to each of these routines, | |
1235 | * which is used to keep track of the state required to manage the | |
1236 | * frequency meter and its digital filter. | |
1237 | * | |
1238 | * The filter works on the number of events marked per unit time. | |
1239 | * The filter is single-pole low-pass recursive (IIR). The time unit | |
1240 | * is 1 second. Arithmetic is done using 32-bit integers scaled to | |
1241 | * simulate 3 decimal digits of precision (multiplied by 1000). | |
1242 | * | |
1243 | * With an FM_COEF of 933, and a time base of 1 second, the filter | |
1244 | * has a half-life of 10 seconds, meaning that if the events quit | |
1245 | * happening, then the rate returned from the fmeter_getrate() | |
1246 | * will be cut in half each 10 seconds, until it converges to zero. | |
1247 | * | |
1248 | * It is not worth doing a real infinitely recursive filter. If more | |
1249 | * than FM_MAXTICKS ticks have elapsed since the last filter event, | |
1250 | * just compute FM_MAXTICKS ticks worth, by which point the level | |
1251 | * will be stable. | |
1252 | * | |
1253 | * Limit the count of unprocessed events to FM_MAXCNT, so as to avoid | |
1254 | * arithmetic overflow in the fmeter_update() routine. | |
1255 | * | |
1256 | * Given the simple 32 bit integer arithmetic used, this meter works | |
1257 | * best for reporting rates between one per millisecond (msec) and | |
1258 | * one per 32 (approx) seconds. At constant rates faster than one | |
1259 | * per msec it maxes out at values just under 1,000,000. At constant | |
1260 | * rates between one per msec, and one per second it will stabilize | |
1261 | * to a value N*1000, where N is the rate of events per second. | |
1262 | * At constant rates between one per second and one per 32 seconds, | |
1263 | * it will be choppy, moving up on the seconds that have an event, | |
1264 | * and then decaying until the next event. At rates slower than | |
1265 | * about one in 32 seconds, it decays all the way back to zero between | |
1266 | * each event. | |
1267 | */ | |
1268 | ||
1269 | #define FM_COEF 933 /* coefficient for half-life of 10 secs */ | |
1270 | #define FM_MAXTICKS ((time_t)99) /* useless computing more ticks than this */ | |
1271 | #define FM_MAXCNT 1000000 /* limit cnt to avoid overflow */ | |
1272 | #define FM_SCALE 1000 /* faux fixed point scale */ | |
1273 | ||
1274 | /* Initialize a frequency meter */ | |
1275 | static void fmeter_init(struct fmeter *fmp) | |
1276 | { | |
1277 | fmp->cnt = 0; | |
1278 | fmp->val = 0; | |
1279 | fmp->time = 0; | |
1280 | spin_lock_init(&fmp->lock); | |
1281 | } | |
1282 | ||
1283 | /* Internal meter update - process cnt events and update value */ | |
1284 | static void fmeter_update(struct fmeter *fmp) | |
1285 | { | |
1286 | time_t now = get_seconds(); | |
1287 | time_t ticks = now - fmp->time; | |
1288 | ||
1289 | if (ticks == 0) | |
1290 | return; | |
1291 | ||
1292 | ticks = min(FM_MAXTICKS, ticks); | |
1293 | while (ticks-- > 0) | |
1294 | fmp->val = (FM_COEF * fmp->val) / FM_SCALE; | |
1295 | fmp->time = now; | |
1296 | ||
1297 | fmp->val += ((FM_SCALE - FM_COEF) * fmp->cnt) / FM_SCALE; | |
1298 | fmp->cnt = 0; | |
1299 | } | |
1300 | ||
1301 | /* Process any previous ticks, then bump cnt by one (times scale). */ | |
1302 | static void fmeter_markevent(struct fmeter *fmp) | |
1303 | { | |
1304 | spin_lock(&fmp->lock); | |
1305 | fmeter_update(fmp); | |
1306 | fmp->cnt = min(FM_MAXCNT, fmp->cnt + FM_SCALE); | |
1307 | spin_unlock(&fmp->lock); | |
1308 | } | |
1309 | ||
1310 | /* Process any previous ticks, then return current value. */ | |
1311 | static int fmeter_getrate(struct fmeter *fmp) | |
1312 | { | |
1313 | int val; | |
1314 | ||
1315 | spin_lock(&fmp->lock); | |
1316 | fmeter_update(fmp); | |
1317 | val = fmp->val; | |
1318 | spin_unlock(&fmp->lock); | |
1319 | return val; | |
1320 | } | |
1321 | ||
2341d1b6 LZ |
1322 | /* Protected by cgroup_lock */ |
1323 | static cpumask_var_t cpus_attach; | |
1324 | ||
2df167a3 | 1325 | /* Called by cgroups to determine if a cpuset is usable; cgroup_mutex held */ |
be367d09 BB |
1326 | static int cpuset_can_attach(struct cgroup_subsys *ss, struct cgroup *cont, |
1327 | struct task_struct *tsk, bool threadgroup) | |
1da177e4 | 1328 | { |
be367d09 | 1329 | int ret; |
8793d854 | 1330 | struct cpuset *cs = cgroup_cs(cont); |
1da177e4 | 1331 | |
300ed6cb | 1332 | if (cpumask_empty(cs->cpus_allowed) || nodes_empty(cs->mems_allowed)) |
1da177e4 | 1333 | return -ENOSPC; |
9985b0ba | 1334 | |
6d7b2f5f DR |
1335 | /* |
1336 | * Kthreads bound to specific cpus cannot be moved to a new cpuset; we | |
1337 | * cannot change their cpu affinity and isolating such threads by their | |
1338 | * set of allowed nodes is unnecessary. Thus, cpusets are not | |
1339 | * applicable for such threads. This prevents checking for success of | |
1340 | * set_cpus_allowed_ptr() on all attached tasks before cpus_allowed may | |
1341 | * be changed. | |
1342 | */ | |
1343 | if (tsk->flags & PF_THREAD_BOUND) | |
1344 | return -EINVAL; | |
1da177e4 | 1345 | |
be367d09 BB |
1346 | ret = security_task_setscheduler(tsk, 0, NULL); |
1347 | if (ret) | |
1348 | return ret; | |
1349 | if (threadgroup) { | |
1350 | struct task_struct *c; | |
1351 | ||
1352 | rcu_read_lock(); | |
1353 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
1354 | ret = security_task_setscheduler(c, 0, NULL); | |
1355 | if (ret) { | |
1356 | rcu_read_unlock(); | |
1357 | return ret; | |
1358 | } | |
1359 | } | |
1360 | rcu_read_unlock(); | |
1361 | } | |
1362 | return 0; | |
1363 | } | |
1364 | ||
1365 | static void cpuset_attach_task(struct task_struct *tsk, nodemask_t *to, | |
1366 | struct cpuset *cs) | |
1367 | { | |
1368 | int err; | |
1369 | /* | |
1370 | * can_attach beforehand should guarantee that this doesn't fail. | |
1371 | * TODO: have a better way to handle failure here | |
1372 | */ | |
1373 | err = set_cpus_allowed_ptr(tsk, cpus_attach); | |
1374 | WARN_ON_ONCE(err); | |
1375 | ||
1376 | task_lock(tsk); | |
1377 | cpuset_change_task_nodemask(tsk, to); | |
1378 | task_unlock(tsk); | |
1379 | cpuset_update_task_spread_flag(cs, tsk); | |
1380 | ||
8793d854 | 1381 | } |
1da177e4 | 1382 | |
be367d09 BB |
1383 | static void cpuset_attach(struct cgroup_subsys *ss, struct cgroup *cont, |
1384 | struct cgroup *oldcont, struct task_struct *tsk, | |
1385 | bool threadgroup) | |
8793d854 | 1386 | { |
8793d854 PM |
1387 | nodemask_t from, to; |
1388 | struct mm_struct *mm; | |
1389 | struct cpuset *cs = cgroup_cs(cont); | |
1390 | struct cpuset *oldcs = cgroup_cs(oldcont); | |
22fb52dd | 1391 | |
f5813d94 | 1392 | if (cs == &top_cpuset) { |
2341d1b6 | 1393 | cpumask_copy(cpus_attach, cpu_possible_mask); |
58568d2a | 1394 | to = node_possible_map; |
f5813d94 | 1395 | } else { |
2341d1b6 | 1396 | guarantee_online_cpus(cs, cpus_attach); |
58568d2a | 1397 | guarantee_online_mems(cs, &to); |
f5813d94 | 1398 | } |
1da177e4 | 1399 | |
be367d09 BB |
1400 | /* do per-task migration stuff possibly for each in the threadgroup */ |
1401 | cpuset_attach_task(tsk, &to, cs); | |
1402 | if (threadgroup) { | |
1403 | struct task_struct *c; | |
1404 | rcu_read_lock(); | |
1405 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
1406 | cpuset_attach_task(c, &to, cs); | |
1407 | } | |
1408 | rcu_read_unlock(); | |
1409 | } | |
950592f7 | 1410 | |
be367d09 | 1411 | /* change mm; only needs to be done once even if threadgroup */ |
45b07ef3 PJ |
1412 | from = oldcs->mems_allowed; |
1413 | to = cs->mems_allowed; | |
4225399a PJ |
1414 | mm = get_task_mm(tsk); |
1415 | if (mm) { | |
1416 | mpol_rebind_mm(mm, &to); | |
2741a559 | 1417 | if (is_memory_migrate(cs)) |
e4e364e8 | 1418 | cpuset_migrate_mm(mm, &from, &to); |
4225399a PJ |
1419 | mmput(mm); |
1420 | } | |
1da177e4 LT |
1421 | } |
1422 | ||
1423 | /* The various types of files and directories in a cpuset file system */ | |
1424 | ||
1425 | typedef enum { | |
45b07ef3 | 1426 | FILE_MEMORY_MIGRATE, |
1da177e4 LT |
1427 | FILE_CPULIST, |
1428 | FILE_MEMLIST, | |
1429 | FILE_CPU_EXCLUSIVE, | |
1430 | FILE_MEM_EXCLUSIVE, | |
78608366 | 1431 | FILE_MEM_HARDWALL, |
029190c5 | 1432 | FILE_SCHED_LOAD_BALANCE, |
1d3504fc | 1433 | FILE_SCHED_RELAX_DOMAIN_LEVEL, |
3e0d98b9 PJ |
1434 | FILE_MEMORY_PRESSURE_ENABLED, |
1435 | FILE_MEMORY_PRESSURE, | |
825a46af PJ |
1436 | FILE_SPREAD_PAGE, |
1437 | FILE_SPREAD_SLAB, | |
1da177e4 LT |
1438 | } cpuset_filetype_t; |
1439 | ||
700fe1ab PM |
1440 | static int cpuset_write_u64(struct cgroup *cgrp, struct cftype *cft, u64 val) |
1441 | { | |
1442 | int retval = 0; | |
1443 | struct cpuset *cs = cgroup_cs(cgrp); | |
1444 | cpuset_filetype_t type = cft->private; | |
1445 | ||
e3712395 | 1446 | if (!cgroup_lock_live_group(cgrp)) |
700fe1ab | 1447 | return -ENODEV; |
700fe1ab PM |
1448 | |
1449 | switch (type) { | |
1da177e4 | 1450 | case FILE_CPU_EXCLUSIVE: |
700fe1ab | 1451 | retval = update_flag(CS_CPU_EXCLUSIVE, cs, val); |
1da177e4 LT |
1452 | break; |
1453 | case FILE_MEM_EXCLUSIVE: | |
700fe1ab | 1454 | retval = update_flag(CS_MEM_EXCLUSIVE, cs, val); |
1da177e4 | 1455 | break; |
78608366 PM |
1456 | case FILE_MEM_HARDWALL: |
1457 | retval = update_flag(CS_MEM_HARDWALL, cs, val); | |
1458 | break; | |
029190c5 | 1459 | case FILE_SCHED_LOAD_BALANCE: |
700fe1ab | 1460 | retval = update_flag(CS_SCHED_LOAD_BALANCE, cs, val); |
1d3504fc | 1461 | break; |
45b07ef3 | 1462 | case FILE_MEMORY_MIGRATE: |
700fe1ab | 1463 | retval = update_flag(CS_MEMORY_MIGRATE, cs, val); |
45b07ef3 | 1464 | break; |
3e0d98b9 | 1465 | case FILE_MEMORY_PRESSURE_ENABLED: |
700fe1ab | 1466 | cpuset_memory_pressure_enabled = !!val; |
3e0d98b9 PJ |
1467 | break; |
1468 | case FILE_MEMORY_PRESSURE: | |
1469 | retval = -EACCES; | |
1470 | break; | |
825a46af | 1471 | case FILE_SPREAD_PAGE: |
700fe1ab | 1472 | retval = update_flag(CS_SPREAD_PAGE, cs, val); |
825a46af PJ |
1473 | break; |
1474 | case FILE_SPREAD_SLAB: | |
700fe1ab | 1475 | retval = update_flag(CS_SPREAD_SLAB, cs, val); |
825a46af | 1476 | break; |
1da177e4 LT |
1477 | default: |
1478 | retval = -EINVAL; | |
700fe1ab | 1479 | break; |
1da177e4 | 1480 | } |
8793d854 | 1481 | cgroup_unlock(); |
1da177e4 LT |
1482 | return retval; |
1483 | } | |
1484 | ||
5be7a479 PM |
1485 | static int cpuset_write_s64(struct cgroup *cgrp, struct cftype *cft, s64 val) |
1486 | { | |
1487 | int retval = 0; | |
1488 | struct cpuset *cs = cgroup_cs(cgrp); | |
1489 | cpuset_filetype_t type = cft->private; | |
1490 | ||
e3712395 | 1491 | if (!cgroup_lock_live_group(cgrp)) |
5be7a479 | 1492 | return -ENODEV; |
e3712395 | 1493 | |
5be7a479 PM |
1494 | switch (type) { |
1495 | case FILE_SCHED_RELAX_DOMAIN_LEVEL: | |
1496 | retval = update_relax_domain_level(cs, val); | |
1497 | break; | |
1498 | default: | |
1499 | retval = -EINVAL; | |
1500 | break; | |
1501 | } | |
1502 | cgroup_unlock(); | |
1503 | return retval; | |
1504 | } | |
1505 | ||
e3712395 PM |
1506 | /* |
1507 | * Common handling for a write to a "cpus" or "mems" file. | |
1508 | */ | |
1509 | static int cpuset_write_resmask(struct cgroup *cgrp, struct cftype *cft, | |
1510 | const char *buf) | |
1511 | { | |
1512 | int retval = 0; | |
645fcc9d LZ |
1513 | struct cpuset *cs = cgroup_cs(cgrp); |
1514 | struct cpuset *trialcs; | |
e3712395 PM |
1515 | |
1516 | if (!cgroup_lock_live_group(cgrp)) | |
1517 | return -ENODEV; | |
1518 | ||
645fcc9d LZ |
1519 | trialcs = alloc_trial_cpuset(cs); |
1520 | if (!trialcs) | |
1521 | return -ENOMEM; | |
1522 | ||
e3712395 PM |
1523 | switch (cft->private) { |
1524 | case FILE_CPULIST: | |
645fcc9d | 1525 | retval = update_cpumask(cs, trialcs, buf); |
e3712395 PM |
1526 | break; |
1527 | case FILE_MEMLIST: | |
645fcc9d | 1528 | retval = update_nodemask(cs, trialcs, buf); |
e3712395 PM |
1529 | break; |
1530 | default: | |
1531 | retval = -EINVAL; | |
1532 | break; | |
1533 | } | |
645fcc9d LZ |
1534 | |
1535 | free_trial_cpuset(trialcs); | |
e3712395 PM |
1536 | cgroup_unlock(); |
1537 | return retval; | |
1538 | } | |
1539 | ||
1da177e4 LT |
1540 | /* |
1541 | * These ascii lists should be read in a single call, by using a user | |
1542 | * buffer large enough to hold the entire map. If read in smaller | |
1543 | * chunks, there is no guarantee of atomicity. Since the display format | |
1544 | * used, list of ranges of sequential numbers, is variable length, | |
1545 | * and since these maps can change value dynamically, one could read | |
1546 | * gibberish by doing partial reads while a list was changing. | |
1547 | * A single large read to a buffer that crosses a page boundary is | |
1548 | * ok, because the result being copied to user land is not recomputed | |
1549 | * across a page fault. | |
1550 | */ | |
1551 | ||
1552 | static int cpuset_sprintf_cpulist(char *page, struct cpuset *cs) | |
1553 | { | |
5a7625df | 1554 | int ret; |
1da177e4 | 1555 | |
3d3f26a7 | 1556 | mutex_lock(&callback_mutex); |
300ed6cb | 1557 | ret = cpulist_scnprintf(page, PAGE_SIZE, cs->cpus_allowed); |
3d3f26a7 | 1558 | mutex_unlock(&callback_mutex); |
1da177e4 | 1559 | |
5a7625df | 1560 | return ret; |
1da177e4 LT |
1561 | } |
1562 | ||
1563 | static int cpuset_sprintf_memlist(char *page, struct cpuset *cs) | |
1564 | { | |
1565 | nodemask_t mask; | |
1566 | ||
3d3f26a7 | 1567 | mutex_lock(&callback_mutex); |
1da177e4 | 1568 | mask = cs->mems_allowed; |
3d3f26a7 | 1569 | mutex_unlock(&callback_mutex); |
1da177e4 LT |
1570 | |
1571 | return nodelist_scnprintf(page, PAGE_SIZE, mask); | |
1572 | } | |
1573 | ||
8793d854 PM |
1574 | static ssize_t cpuset_common_file_read(struct cgroup *cont, |
1575 | struct cftype *cft, | |
1576 | struct file *file, | |
1577 | char __user *buf, | |
1578 | size_t nbytes, loff_t *ppos) | |
1da177e4 | 1579 | { |
8793d854 | 1580 | struct cpuset *cs = cgroup_cs(cont); |
1da177e4 LT |
1581 | cpuset_filetype_t type = cft->private; |
1582 | char *page; | |
1583 | ssize_t retval = 0; | |
1584 | char *s; | |
1da177e4 | 1585 | |
e12ba74d | 1586 | if (!(page = (char *)__get_free_page(GFP_TEMPORARY))) |
1da177e4 LT |
1587 | return -ENOMEM; |
1588 | ||
1589 | s = page; | |
1590 | ||
1591 | switch (type) { | |
1592 | case FILE_CPULIST: | |
1593 | s += cpuset_sprintf_cpulist(s, cs); | |
1594 | break; | |
1595 | case FILE_MEMLIST: | |
1596 | s += cpuset_sprintf_memlist(s, cs); | |
1597 | break; | |
1da177e4 LT |
1598 | default: |
1599 | retval = -EINVAL; | |
1600 | goto out; | |
1601 | } | |
1602 | *s++ = '\n'; | |
1da177e4 | 1603 | |
eacaa1f5 | 1604 | retval = simple_read_from_buffer(buf, nbytes, ppos, page, s - page); |
1da177e4 LT |
1605 | out: |
1606 | free_page((unsigned long)page); | |
1607 | return retval; | |
1608 | } | |
1609 | ||
700fe1ab PM |
1610 | static u64 cpuset_read_u64(struct cgroup *cont, struct cftype *cft) |
1611 | { | |
1612 | struct cpuset *cs = cgroup_cs(cont); | |
1613 | cpuset_filetype_t type = cft->private; | |
1614 | switch (type) { | |
1615 | case FILE_CPU_EXCLUSIVE: | |
1616 | return is_cpu_exclusive(cs); | |
1617 | case FILE_MEM_EXCLUSIVE: | |
1618 | return is_mem_exclusive(cs); | |
78608366 PM |
1619 | case FILE_MEM_HARDWALL: |
1620 | return is_mem_hardwall(cs); | |
700fe1ab PM |
1621 | case FILE_SCHED_LOAD_BALANCE: |
1622 | return is_sched_load_balance(cs); | |
1623 | case FILE_MEMORY_MIGRATE: | |
1624 | return is_memory_migrate(cs); | |
1625 | case FILE_MEMORY_PRESSURE_ENABLED: | |
1626 | return cpuset_memory_pressure_enabled; | |
1627 | case FILE_MEMORY_PRESSURE: | |
1628 | return fmeter_getrate(&cs->fmeter); | |
1629 | case FILE_SPREAD_PAGE: | |
1630 | return is_spread_page(cs); | |
1631 | case FILE_SPREAD_SLAB: | |
1632 | return is_spread_slab(cs); | |
1633 | default: | |
1634 | BUG(); | |
1635 | } | |
cf417141 MK |
1636 | |
1637 | /* Unreachable but makes gcc happy */ | |
1638 | return 0; | |
700fe1ab | 1639 | } |
1da177e4 | 1640 | |
5be7a479 PM |
1641 | static s64 cpuset_read_s64(struct cgroup *cont, struct cftype *cft) |
1642 | { | |
1643 | struct cpuset *cs = cgroup_cs(cont); | |
1644 | cpuset_filetype_t type = cft->private; | |
1645 | switch (type) { | |
1646 | case FILE_SCHED_RELAX_DOMAIN_LEVEL: | |
1647 | return cs->relax_domain_level; | |
1648 | default: | |
1649 | BUG(); | |
1650 | } | |
cf417141 MK |
1651 | |
1652 | /* Unrechable but makes gcc happy */ | |
1653 | return 0; | |
5be7a479 PM |
1654 | } |
1655 | ||
1da177e4 LT |
1656 | |
1657 | /* | |
1658 | * for the common functions, 'private' gives the type of file | |
1659 | */ | |
1660 | ||
addf2c73 PM |
1661 | static struct cftype files[] = { |
1662 | { | |
1663 | .name = "cpus", | |
1664 | .read = cpuset_common_file_read, | |
e3712395 PM |
1665 | .write_string = cpuset_write_resmask, |
1666 | .max_write_len = (100U + 6 * NR_CPUS), | |
addf2c73 PM |
1667 | .private = FILE_CPULIST, |
1668 | }, | |
1669 | ||
1670 | { | |
1671 | .name = "mems", | |
1672 | .read = cpuset_common_file_read, | |
e3712395 PM |
1673 | .write_string = cpuset_write_resmask, |
1674 | .max_write_len = (100U + 6 * MAX_NUMNODES), | |
addf2c73 PM |
1675 | .private = FILE_MEMLIST, |
1676 | }, | |
1677 | ||
1678 | { | |
1679 | .name = "cpu_exclusive", | |
1680 | .read_u64 = cpuset_read_u64, | |
1681 | .write_u64 = cpuset_write_u64, | |
1682 | .private = FILE_CPU_EXCLUSIVE, | |
1683 | }, | |
1684 | ||
1685 | { | |
1686 | .name = "mem_exclusive", | |
1687 | .read_u64 = cpuset_read_u64, | |
1688 | .write_u64 = cpuset_write_u64, | |
1689 | .private = FILE_MEM_EXCLUSIVE, | |
1690 | }, | |
1691 | ||
78608366 PM |
1692 | { |
1693 | .name = "mem_hardwall", | |
1694 | .read_u64 = cpuset_read_u64, | |
1695 | .write_u64 = cpuset_write_u64, | |
1696 | .private = FILE_MEM_HARDWALL, | |
1697 | }, | |
1698 | ||
addf2c73 PM |
1699 | { |
1700 | .name = "sched_load_balance", | |
1701 | .read_u64 = cpuset_read_u64, | |
1702 | .write_u64 = cpuset_write_u64, | |
1703 | .private = FILE_SCHED_LOAD_BALANCE, | |
1704 | }, | |
1705 | ||
1706 | { | |
1707 | .name = "sched_relax_domain_level", | |
5be7a479 PM |
1708 | .read_s64 = cpuset_read_s64, |
1709 | .write_s64 = cpuset_write_s64, | |
addf2c73 PM |
1710 | .private = FILE_SCHED_RELAX_DOMAIN_LEVEL, |
1711 | }, | |
1712 | ||
1713 | { | |
1714 | .name = "memory_migrate", | |
1715 | .read_u64 = cpuset_read_u64, | |
1716 | .write_u64 = cpuset_write_u64, | |
1717 | .private = FILE_MEMORY_MIGRATE, | |
1718 | }, | |
1719 | ||
1720 | { | |
1721 | .name = "memory_pressure", | |
1722 | .read_u64 = cpuset_read_u64, | |
1723 | .write_u64 = cpuset_write_u64, | |
1724 | .private = FILE_MEMORY_PRESSURE, | |
099fca32 | 1725 | .mode = S_IRUGO, |
addf2c73 PM |
1726 | }, |
1727 | ||
1728 | { | |
1729 | .name = "memory_spread_page", | |
1730 | .read_u64 = cpuset_read_u64, | |
1731 | .write_u64 = cpuset_write_u64, | |
1732 | .private = FILE_SPREAD_PAGE, | |
1733 | }, | |
1734 | ||
1735 | { | |
1736 | .name = "memory_spread_slab", | |
1737 | .read_u64 = cpuset_read_u64, | |
1738 | .write_u64 = cpuset_write_u64, | |
1739 | .private = FILE_SPREAD_SLAB, | |
1740 | }, | |
45b07ef3 PJ |
1741 | }; |
1742 | ||
3e0d98b9 PJ |
1743 | static struct cftype cft_memory_pressure_enabled = { |
1744 | .name = "memory_pressure_enabled", | |
700fe1ab PM |
1745 | .read_u64 = cpuset_read_u64, |
1746 | .write_u64 = cpuset_write_u64, | |
3e0d98b9 PJ |
1747 | .private = FILE_MEMORY_PRESSURE_ENABLED, |
1748 | }; | |
1749 | ||
8793d854 | 1750 | static int cpuset_populate(struct cgroup_subsys *ss, struct cgroup *cont) |
1da177e4 LT |
1751 | { |
1752 | int err; | |
1753 | ||
addf2c73 PM |
1754 | err = cgroup_add_files(cont, ss, files, ARRAY_SIZE(files)); |
1755 | if (err) | |
1da177e4 | 1756 | return err; |
8793d854 | 1757 | /* memory_pressure_enabled is in root cpuset only */ |
addf2c73 | 1758 | if (!cont->parent) |
8793d854 | 1759 | err = cgroup_add_file(cont, ss, |
addf2c73 PM |
1760 | &cft_memory_pressure_enabled); |
1761 | return err; | |
1da177e4 LT |
1762 | } |
1763 | ||
8793d854 PM |
1764 | /* |
1765 | * post_clone() is called at the end of cgroup_clone(). | |
1766 | * 'cgroup' was just created automatically as a result of | |
1767 | * a cgroup_clone(), and the current task is about to | |
1768 | * be moved into 'cgroup'. | |
1769 | * | |
1770 | * Currently we refuse to set up the cgroup - thereby | |
1771 | * refusing the task to be entered, and as a result refusing | |
1772 | * the sys_unshare() or clone() which initiated it - if any | |
1773 | * sibling cpusets have exclusive cpus or mem. | |
1774 | * | |
1775 | * If this becomes a problem for some users who wish to | |
1776 | * allow that scenario, then cpuset_post_clone() could be | |
1777 | * changed to grant parent->cpus_allowed-sibling_cpus_exclusive | |
2df167a3 PM |
1778 | * (and likewise for mems) to the new cgroup. Called with cgroup_mutex |
1779 | * held. | |
8793d854 PM |
1780 | */ |
1781 | static void cpuset_post_clone(struct cgroup_subsys *ss, | |
1782 | struct cgroup *cgroup) | |
1783 | { | |
1784 | struct cgroup *parent, *child; | |
1785 | struct cpuset *cs, *parent_cs; | |
1786 | ||
1787 | parent = cgroup->parent; | |
1788 | list_for_each_entry(child, &parent->children, sibling) { | |
1789 | cs = cgroup_cs(child); | |
1790 | if (is_mem_exclusive(cs) || is_cpu_exclusive(cs)) | |
1791 | return; | |
1792 | } | |
1793 | cs = cgroup_cs(cgroup); | |
1794 | parent_cs = cgroup_cs(parent); | |
1795 | ||
1796 | cs->mems_allowed = parent_cs->mems_allowed; | |
300ed6cb | 1797 | cpumask_copy(cs->cpus_allowed, parent_cs->cpus_allowed); |
8793d854 PM |
1798 | return; |
1799 | } | |
1800 | ||
1da177e4 LT |
1801 | /* |
1802 | * cpuset_create - create a cpuset | |
2df167a3 PM |
1803 | * ss: cpuset cgroup subsystem |
1804 | * cont: control group that the new cpuset will be part of | |
1da177e4 LT |
1805 | */ |
1806 | ||
8793d854 PM |
1807 | static struct cgroup_subsys_state *cpuset_create( |
1808 | struct cgroup_subsys *ss, | |
1809 | struct cgroup *cont) | |
1da177e4 LT |
1810 | { |
1811 | struct cpuset *cs; | |
8793d854 | 1812 | struct cpuset *parent; |
1da177e4 | 1813 | |
8793d854 | 1814 | if (!cont->parent) { |
8793d854 PM |
1815 | return &top_cpuset.css; |
1816 | } | |
1817 | parent = cgroup_cs(cont->parent); | |
1da177e4 LT |
1818 | cs = kmalloc(sizeof(*cs), GFP_KERNEL); |
1819 | if (!cs) | |
8793d854 | 1820 | return ERR_PTR(-ENOMEM); |
300ed6cb LZ |
1821 | if (!alloc_cpumask_var(&cs->cpus_allowed, GFP_KERNEL)) { |
1822 | kfree(cs); | |
1823 | return ERR_PTR(-ENOMEM); | |
1824 | } | |
1da177e4 | 1825 | |
1da177e4 | 1826 | cs->flags = 0; |
825a46af PJ |
1827 | if (is_spread_page(parent)) |
1828 | set_bit(CS_SPREAD_PAGE, &cs->flags); | |
1829 | if (is_spread_slab(parent)) | |
1830 | set_bit(CS_SPREAD_SLAB, &cs->flags); | |
029190c5 | 1831 | set_bit(CS_SCHED_LOAD_BALANCE, &cs->flags); |
300ed6cb | 1832 | cpumask_clear(cs->cpus_allowed); |
f9a86fcb | 1833 | nodes_clear(cs->mems_allowed); |
3e0d98b9 | 1834 | fmeter_init(&cs->fmeter); |
1d3504fc | 1835 | cs->relax_domain_level = -1; |
1da177e4 LT |
1836 | |
1837 | cs->parent = parent; | |
202f72d5 | 1838 | number_of_cpusets++; |
8793d854 | 1839 | return &cs->css ; |
1da177e4 LT |
1840 | } |
1841 | ||
029190c5 | 1842 | /* |
029190c5 PJ |
1843 | * If the cpuset being removed has its flag 'sched_load_balance' |
1844 | * enabled, then simulate turning sched_load_balance off, which | |
cf417141 | 1845 | * will call async_rebuild_sched_domains(). |
029190c5 PJ |
1846 | */ |
1847 | ||
8793d854 | 1848 | static void cpuset_destroy(struct cgroup_subsys *ss, struct cgroup *cont) |
1da177e4 | 1849 | { |
8793d854 | 1850 | struct cpuset *cs = cgroup_cs(cont); |
1da177e4 | 1851 | |
029190c5 | 1852 | if (is_sched_load_balance(cs)) |
700fe1ab | 1853 | update_flag(CS_SCHED_LOAD_BALANCE, cs, 0); |
029190c5 | 1854 | |
202f72d5 | 1855 | number_of_cpusets--; |
300ed6cb | 1856 | free_cpumask_var(cs->cpus_allowed); |
8793d854 | 1857 | kfree(cs); |
1da177e4 LT |
1858 | } |
1859 | ||
8793d854 PM |
1860 | struct cgroup_subsys cpuset_subsys = { |
1861 | .name = "cpuset", | |
1862 | .create = cpuset_create, | |
cf417141 | 1863 | .destroy = cpuset_destroy, |
8793d854 PM |
1864 | .can_attach = cpuset_can_attach, |
1865 | .attach = cpuset_attach, | |
1866 | .populate = cpuset_populate, | |
1867 | .post_clone = cpuset_post_clone, | |
1868 | .subsys_id = cpuset_subsys_id, | |
1869 | .early_init = 1, | |
1870 | }; | |
1871 | ||
1da177e4 LT |
1872 | /** |
1873 | * cpuset_init - initialize cpusets at system boot | |
1874 | * | |
1875 | * Description: Initialize top_cpuset and the cpuset internal file system, | |
1876 | **/ | |
1877 | ||
1878 | int __init cpuset_init(void) | |
1879 | { | |
8793d854 | 1880 | int err = 0; |
1da177e4 | 1881 | |
58568d2a MX |
1882 | if (!alloc_cpumask_var(&top_cpuset.cpus_allowed, GFP_KERNEL)) |
1883 | BUG(); | |
1884 | ||
300ed6cb | 1885 | cpumask_setall(top_cpuset.cpus_allowed); |
f9a86fcb | 1886 | nodes_setall(top_cpuset.mems_allowed); |
1da177e4 | 1887 | |
3e0d98b9 | 1888 | fmeter_init(&top_cpuset.fmeter); |
029190c5 | 1889 | set_bit(CS_SCHED_LOAD_BALANCE, &top_cpuset.flags); |
1d3504fc | 1890 | top_cpuset.relax_domain_level = -1; |
1da177e4 | 1891 | |
1da177e4 LT |
1892 | err = register_filesystem(&cpuset_fs_type); |
1893 | if (err < 0) | |
8793d854 PM |
1894 | return err; |
1895 | ||
2341d1b6 LZ |
1896 | if (!alloc_cpumask_var(&cpus_attach, GFP_KERNEL)) |
1897 | BUG(); | |
1898 | ||
202f72d5 | 1899 | number_of_cpusets = 1; |
8793d854 | 1900 | return 0; |
1da177e4 LT |
1901 | } |
1902 | ||
956db3ca CW |
1903 | /** |
1904 | * cpuset_do_move_task - move a given task to another cpuset | |
1905 | * @tsk: pointer to task_struct the task to move | |
1906 | * @scan: struct cgroup_scanner contained in its struct cpuset_hotplug_scanner | |
1907 | * | |
1908 | * Called by cgroup_scan_tasks() for each task in a cgroup. | |
1909 | * Return nonzero to stop the walk through the tasks. | |
1910 | */ | |
9e0c914c AB |
1911 | static void cpuset_do_move_task(struct task_struct *tsk, |
1912 | struct cgroup_scanner *scan) | |
956db3ca | 1913 | { |
7f81b1ae | 1914 | struct cgroup *new_cgroup = scan->data; |
956db3ca | 1915 | |
7f81b1ae | 1916 | cgroup_attach_task(new_cgroup, tsk); |
956db3ca CW |
1917 | } |
1918 | ||
1919 | /** | |
1920 | * move_member_tasks_to_cpuset - move tasks from one cpuset to another | |
1921 | * @from: cpuset in which the tasks currently reside | |
1922 | * @to: cpuset to which the tasks will be moved | |
1923 | * | |
c8d9c90c PJ |
1924 | * Called with cgroup_mutex held |
1925 | * callback_mutex must not be held, as cpuset_attach() will take it. | |
956db3ca CW |
1926 | * |
1927 | * The cgroup_scan_tasks() function will scan all the tasks in a cgroup, | |
1928 | * calling callback functions for each. | |
1929 | */ | |
1930 | static void move_member_tasks_to_cpuset(struct cpuset *from, struct cpuset *to) | |
1931 | { | |
7f81b1ae | 1932 | struct cgroup_scanner scan; |
956db3ca | 1933 | |
7f81b1ae LZ |
1934 | scan.cg = from->css.cgroup; |
1935 | scan.test_task = NULL; /* select all tasks in cgroup */ | |
1936 | scan.process_task = cpuset_do_move_task; | |
1937 | scan.heap = NULL; | |
1938 | scan.data = to->css.cgroup; | |
956db3ca | 1939 | |
7f81b1ae | 1940 | if (cgroup_scan_tasks(&scan)) |
956db3ca CW |
1941 | printk(KERN_ERR "move_member_tasks_to_cpuset: " |
1942 | "cgroup_scan_tasks failed\n"); | |
1943 | } | |
1944 | ||
b1aac8bb | 1945 | /* |
cf417141 | 1946 | * If CPU and/or memory hotplug handlers, below, unplug any CPUs |
b1aac8bb PJ |
1947 | * or memory nodes, we need to walk over the cpuset hierarchy, |
1948 | * removing that CPU or node from all cpusets. If this removes the | |
956db3ca CW |
1949 | * last CPU or node from a cpuset, then move the tasks in the empty |
1950 | * cpuset to its next-highest non-empty parent. | |
b1aac8bb | 1951 | * |
c8d9c90c PJ |
1952 | * Called with cgroup_mutex held |
1953 | * callback_mutex must not be held, as cpuset_attach() will take it. | |
b1aac8bb | 1954 | */ |
956db3ca CW |
1955 | static void remove_tasks_in_empty_cpuset(struct cpuset *cs) |
1956 | { | |
1957 | struct cpuset *parent; | |
1958 | ||
c8d9c90c PJ |
1959 | /* |
1960 | * The cgroup's css_sets list is in use if there are tasks | |
1961 | * in the cpuset; the list is empty if there are none; | |
1962 | * the cs->css.refcnt seems always 0. | |
1963 | */ | |
956db3ca CW |
1964 | if (list_empty(&cs->css.cgroup->css_sets)) |
1965 | return; | |
b1aac8bb | 1966 | |
956db3ca CW |
1967 | /* |
1968 | * Find its next-highest non-empty parent, (top cpuset | |
1969 | * has online cpus, so can't be empty). | |
1970 | */ | |
1971 | parent = cs->parent; | |
300ed6cb | 1972 | while (cpumask_empty(parent->cpus_allowed) || |
b4501295 | 1973 | nodes_empty(parent->mems_allowed)) |
956db3ca | 1974 | parent = parent->parent; |
956db3ca CW |
1975 | |
1976 | move_member_tasks_to_cpuset(cs, parent); | |
1977 | } | |
1978 | ||
1979 | /* | |
1980 | * Walk the specified cpuset subtree and look for empty cpusets. | |
1981 | * The tasks of such cpuset must be moved to a parent cpuset. | |
1982 | * | |
2df167a3 | 1983 | * Called with cgroup_mutex held. We take callback_mutex to modify |
956db3ca CW |
1984 | * cpus_allowed and mems_allowed. |
1985 | * | |
1986 | * This walk processes the tree from top to bottom, completing one layer | |
1987 | * before dropping down to the next. It always processes a node before | |
1988 | * any of its children. | |
1989 | * | |
1990 | * For now, since we lack memory hot unplug, we'll never see a cpuset | |
1991 | * that has tasks along with an empty 'mems'. But if we did see such | |
1992 | * a cpuset, we'd handle it just like we do if its 'cpus' was empty. | |
1993 | */ | |
d294eb83 | 1994 | static void scan_for_empty_cpusets(struct cpuset *root) |
b1aac8bb | 1995 | { |
8d1e6266 | 1996 | LIST_HEAD(queue); |
956db3ca CW |
1997 | struct cpuset *cp; /* scans cpusets being updated */ |
1998 | struct cpuset *child; /* scans child cpusets of cp */ | |
8793d854 | 1999 | struct cgroup *cont; |
f9b4fb8d | 2000 | nodemask_t oldmems; |
b1aac8bb | 2001 | |
956db3ca CW |
2002 | list_add_tail((struct list_head *)&root->stack_list, &queue); |
2003 | ||
956db3ca | 2004 | while (!list_empty(&queue)) { |
8d1e6266 | 2005 | cp = list_first_entry(&queue, struct cpuset, stack_list); |
956db3ca CW |
2006 | list_del(queue.next); |
2007 | list_for_each_entry(cont, &cp->css.cgroup->children, sibling) { | |
2008 | child = cgroup_cs(cont); | |
2009 | list_add_tail(&child->stack_list, &queue); | |
2010 | } | |
b4501295 PJ |
2011 | |
2012 | /* Continue past cpusets with all cpus, mems online */ | |
6ad4c188 | 2013 | if (cpumask_subset(cp->cpus_allowed, cpu_active_mask) && |
b4501295 PJ |
2014 | nodes_subset(cp->mems_allowed, node_states[N_HIGH_MEMORY])) |
2015 | continue; | |
2016 | ||
f9b4fb8d MX |
2017 | oldmems = cp->mems_allowed; |
2018 | ||
956db3ca | 2019 | /* Remove offline cpus and mems from this cpuset. */ |
b4501295 | 2020 | mutex_lock(&callback_mutex); |
300ed6cb | 2021 | cpumask_and(cp->cpus_allowed, cp->cpus_allowed, |
6ad4c188 | 2022 | cpu_active_mask); |
956db3ca CW |
2023 | nodes_and(cp->mems_allowed, cp->mems_allowed, |
2024 | node_states[N_HIGH_MEMORY]); | |
b4501295 PJ |
2025 | mutex_unlock(&callback_mutex); |
2026 | ||
2027 | /* Move tasks from the empty cpuset to a parent */ | |
300ed6cb | 2028 | if (cpumask_empty(cp->cpus_allowed) || |
b4501295 | 2029 | nodes_empty(cp->mems_allowed)) |
956db3ca | 2030 | remove_tasks_in_empty_cpuset(cp); |
f9b4fb8d | 2031 | else { |
4e74339a | 2032 | update_tasks_cpumask(cp, NULL); |
010cfac4 | 2033 | update_tasks_nodemask(cp, &oldmems, NULL); |
f9b4fb8d | 2034 | } |
b1aac8bb PJ |
2035 | } |
2036 | } | |
2037 | ||
4c4d50f7 PJ |
2038 | /* |
2039 | * The top_cpuset tracks what CPUs and Memory Nodes are online, | |
2040 | * period. This is necessary in order to make cpusets transparent | |
2041 | * (of no affect) on systems that are actively using CPU hotplug | |
2042 | * but making no active use of cpusets. | |
2043 | * | |
38837fc7 PJ |
2044 | * This routine ensures that top_cpuset.cpus_allowed tracks |
2045 | * cpu_online_map on each CPU hotplug (cpuhp) event. | |
cf417141 MK |
2046 | * |
2047 | * Called within get_online_cpus(). Needs to call cgroup_lock() | |
2048 | * before calling generate_sched_domains(). | |
4c4d50f7 | 2049 | */ |
cf417141 | 2050 | static int cpuset_track_online_cpus(struct notifier_block *unused_nb, |
029190c5 | 2051 | unsigned long phase, void *unused_cpu) |
4c4d50f7 | 2052 | { |
cf417141 | 2053 | struct sched_domain_attr *attr; |
acc3f5d7 | 2054 | cpumask_var_t *doms; |
cf417141 MK |
2055 | int ndoms; |
2056 | ||
3e84050c | 2057 | switch (phase) { |
3e84050c DA |
2058 | case CPU_ONLINE: |
2059 | case CPU_ONLINE_FROZEN: | |
6ad4c188 PZ |
2060 | case CPU_DOWN_PREPARE: |
2061 | case CPU_DOWN_PREPARE_FROZEN: | |
2062 | case CPU_DOWN_FAILED: | |
2063 | case CPU_DOWN_FAILED_FROZEN: | |
3e84050c | 2064 | break; |
cf417141 | 2065 | |
3e84050c | 2066 | default: |
ac076758 | 2067 | return NOTIFY_DONE; |
3e84050c | 2068 | } |
ac076758 | 2069 | |
cf417141 | 2070 | cgroup_lock(); |
0b4217b3 | 2071 | mutex_lock(&callback_mutex); |
6ad4c188 | 2072 | cpumask_copy(top_cpuset.cpus_allowed, cpu_active_mask); |
0b4217b3 | 2073 | mutex_unlock(&callback_mutex); |
cf417141 MK |
2074 | scan_for_empty_cpusets(&top_cpuset); |
2075 | ndoms = generate_sched_domains(&doms, &attr); | |
2076 | cgroup_unlock(); | |
2077 | ||
2078 | /* Have scheduler rebuild the domains */ | |
2079 | partition_sched_domains(ndoms, doms, attr); | |
2080 | ||
3e84050c | 2081 | return NOTIFY_OK; |
4c4d50f7 | 2082 | } |
4c4d50f7 | 2083 | |
b1aac8bb | 2084 | #ifdef CONFIG_MEMORY_HOTPLUG |
38837fc7 | 2085 | /* |
0e1e7c7a | 2086 | * Keep top_cpuset.mems_allowed tracking node_states[N_HIGH_MEMORY]. |
cf417141 MK |
2087 | * Call this routine anytime after node_states[N_HIGH_MEMORY] changes. |
2088 | * See also the previous routine cpuset_track_online_cpus(). | |
38837fc7 | 2089 | */ |
f481891f MX |
2090 | static int cpuset_track_online_nodes(struct notifier_block *self, |
2091 | unsigned long action, void *arg) | |
38837fc7 | 2092 | { |
cf417141 | 2093 | cgroup_lock(); |
f481891f MX |
2094 | switch (action) { |
2095 | case MEM_ONLINE: | |
f481891f | 2096 | case MEM_OFFLINE: |
0b4217b3 | 2097 | mutex_lock(&callback_mutex); |
f481891f | 2098 | top_cpuset.mems_allowed = node_states[N_HIGH_MEMORY]; |
0b4217b3 LZ |
2099 | mutex_unlock(&callback_mutex); |
2100 | if (action == MEM_OFFLINE) | |
2101 | scan_for_empty_cpusets(&top_cpuset); | |
f481891f MX |
2102 | break; |
2103 | default: | |
2104 | break; | |
2105 | } | |
cf417141 | 2106 | cgroup_unlock(); |
f481891f | 2107 | return NOTIFY_OK; |
38837fc7 PJ |
2108 | } |
2109 | #endif | |
2110 | ||
1da177e4 LT |
2111 | /** |
2112 | * cpuset_init_smp - initialize cpus_allowed | |
2113 | * | |
2114 | * Description: Finish top cpuset after cpu, node maps are initialized | |
2115 | **/ | |
2116 | ||
2117 | void __init cpuset_init_smp(void) | |
2118 | { | |
6ad4c188 | 2119 | cpumask_copy(top_cpuset.cpus_allowed, cpu_active_mask); |
0e1e7c7a | 2120 | top_cpuset.mems_allowed = node_states[N_HIGH_MEMORY]; |
4c4d50f7 | 2121 | |
cf417141 | 2122 | hotcpu_notifier(cpuset_track_online_cpus, 0); |
f481891f | 2123 | hotplug_memory_notifier(cpuset_track_online_nodes, 10); |
f90d4118 MX |
2124 | |
2125 | cpuset_wq = create_singlethread_workqueue("cpuset"); | |
2126 | BUG_ON(!cpuset_wq); | |
1da177e4 LT |
2127 | } |
2128 | ||
2129 | /** | |
1da177e4 LT |
2130 | * cpuset_cpus_allowed - return cpus_allowed mask from a tasks cpuset. |
2131 | * @tsk: pointer to task_struct from which to obtain cpuset->cpus_allowed. | |
6af866af | 2132 | * @pmask: pointer to struct cpumask variable to receive cpus_allowed set. |
1da177e4 | 2133 | * |
300ed6cb | 2134 | * Description: Returns the cpumask_var_t cpus_allowed of the cpuset |
1da177e4 LT |
2135 | * attached to the specified @tsk. Guaranteed to return some non-empty |
2136 | * subset of cpu_online_map, even if this means going outside the | |
2137 | * tasks cpuset. | |
2138 | **/ | |
2139 | ||
6af866af | 2140 | void cpuset_cpus_allowed(struct task_struct *tsk, struct cpumask *pmask) |
1da177e4 | 2141 | { |
3d3f26a7 | 2142 | mutex_lock(&callback_mutex); |
f9a86fcb | 2143 | cpuset_cpus_allowed_locked(tsk, pmask); |
470fd646 | 2144 | mutex_unlock(&callback_mutex); |
470fd646 CW |
2145 | } |
2146 | ||
2147 | /** | |
2148 | * cpuset_cpus_allowed_locked - return cpus_allowed mask from a tasks cpuset. | |
2df167a3 | 2149 | * Must be called with callback_mutex held. |
470fd646 | 2150 | **/ |
6af866af | 2151 | void cpuset_cpus_allowed_locked(struct task_struct *tsk, struct cpumask *pmask) |
470fd646 | 2152 | { |
909d75a3 | 2153 | task_lock(tsk); |
f9a86fcb | 2154 | guarantee_online_cpus(task_cs(tsk), pmask); |
909d75a3 | 2155 | task_unlock(tsk); |
1da177e4 LT |
2156 | } |
2157 | ||
2158 | void cpuset_init_current_mems_allowed(void) | |
2159 | { | |
f9a86fcb | 2160 | nodes_setall(current->mems_allowed); |
1da177e4 LT |
2161 | } |
2162 | ||
909d75a3 PJ |
2163 | /** |
2164 | * cpuset_mems_allowed - return mems_allowed mask from a tasks cpuset. | |
2165 | * @tsk: pointer to task_struct from which to obtain cpuset->mems_allowed. | |
2166 | * | |
2167 | * Description: Returns the nodemask_t mems_allowed of the cpuset | |
2168 | * attached to the specified @tsk. Guaranteed to return some non-empty | |
0e1e7c7a | 2169 | * subset of node_states[N_HIGH_MEMORY], even if this means going outside the |
909d75a3 PJ |
2170 | * tasks cpuset. |
2171 | **/ | |
2172 | ||
2173 | nodemask_t cpuset_mems_allowed(struct task_struct *tsk) | |
2174 | { | |
2175 | nodemask_t mask; | |
2176 | ||
3d3f26a7 | 2177 | mutex_lock(&callback_mutex); |
909d75a3 | 2178 | task_lock(tsk); |
8793d854 | 2179 | guarantee_online_mems(task_cs(tsk), &mask); |
909d75a3 | 2180 | task_unlock(tsk); |
3d3f26a7 | 2181 | mutex_unlock(&callback_mutex); |
909d75a3 PJ |
2182 | |
2183 | return mask; | |
2184 | } | |
2185 | ||
d9fd8a6d | 2186 | /** |
19770b32 MG |
2187 | * cpuset_nodemask_valid_mems_allowed - check nodemask vs. curremt mems_allowed |
2188 | * @nodemask: the nodemask to be checked | |
d9fd8a6d | 2189 | * |
19770b32 | 2190 | * Are any of the nodes in the nodemask allowed in current->mems_allowed? |
1da177e4 | 2191 | */ |
19770b32 | 2192 | int cpuset_nodemask_valid_mems_allowed(nodemask_t *nodemask) |
1da177e4 | 2193 | { |
19770b32 | 2194 | return nodes_intersects(*nodemask, current->mems_allowed); |
1da177e4 LT |
2195 | } |
2196 | ||
9bf2229f | 2197 | /* |
78608366 PM |
2198 | * nearest_hardwall_ancestor() - Returns the nearest mem_exclusive or |
2199 | * mem_hardwall ancestor to the specified cpuset. Call holding | |
2200 | * callback_mutex. If no ancestor is mem_exclusive or mem_hardwall | |
2201 | * (an unusual configuration), then returns the root cpuset. | |
9bf2229f | 2202 | */ |
78608366 | 2203 | static const struct cpuset *nearest_hardwall_ancestor(const struct cpuset *cs) |
9bf2229f | 2204 | { |
78608366 | 2205 | while (!(is_mem_exclusive(cs) || is_mem_hardwall(cs)) && cs->parent) |
9bf2229f PJ |
2206 | cs = cs->parent; |
2207 | return cs; | |
2208 | } | |
2209 | ||
d9fd8a6d | 2210 | /** |
a1bc5a4e DR |
2211 | * cpuset_node_allowed_softwall - Can we allocate on a memory node? |
2212 | * @node: is this an allowed node? | |
02a0e53d | 2213 | * @gfp_mask: memory allocation flags |
d9fd8a6d | 2214 | * |
a1bc5a4e DR |
2215 | * If we're in interrupt, yes, we can always allocate. If __GFP_THISNODE is |
2216 | * set, yes, we can always allocate. If node is in our task's mems_allowed, | |
2217 | * yes. If it's not a __GFP_HARDWALL request and this node is in the nearest | |
2218 | * hardwalled cpuset ancestor to this task's cpuset, yes. If the task has been | |
2219 | * OOM killed and has access to memory reserves as specified by the TIF_MEMDIE | |
2220 | * flag, yes. | |
9bf2229f PJ |
2221 | * Otherwise, no. |
2222 | * | |
a1bc5a4e DR |
2223 | * If __GFP_HARDWALL is set, cpuset_node_allowed_softwall() reduces to |
2224 | * cpuset_node_allowed_hardwall(). Otherwise, cpuset_node_allowed_softwall() | |
2225 | * might sleep, and might allow a node from an enclosing cpuset. | |
02a0e53d | 2226 | * |
a1bc5a4e DR |
2227 | * cpuset_node_allowed_hardwall() only handles the simpler case of hardwall |
2228 | * cpusets, and never sleeps. | |
02a0e53d PJ |
2229 | * |
2230 | * The __GFP_THISNODE placement logic is really handled elsewhere, | |
2231 | * by forcibly using a zonelist starting at a specified node, and by | |
2232 | * (in get_page_from_freelist()) refusing to consider the zones for | |
2233 | * any node on the zonelist except the first. By the time any such | |
2234 | * calls get to this routine, we should just shut up and say 'yes'. | |
2235 | * | |
9bf2229f | 2236 | * GFP_USER allocations are marked with the __GFP_HARDWALL bit, |
c596d9f3 DR |
2237 | * and do not allow allocations outside the current tasks cpuset |
2238 | * unless the task has been OOM killed as is marked TIF_MEMDIE. | |
9bf2229f | 2239 | * GFP_KERNEL allocations are not so marked, so can escape to the |
78608366 | 2240 | * nearest enclosing hardwalled ancestor cpuset. |
9bf2229f | 2241 | * |
02a0e53d PJ |
2242 | * Scanning up parent cpusets requires callback_mutex. The |
2243 | * __alloc_pages() routine only calls here with __GFP_HARDWALL bit | |
2244 | * _not_ set if it's a GFP_KERNEL allocation, and all nodes in the | |
2245 | * current tasks mems_allowed came up empty on the first pass over | |
2246 | * the zonelist. So only GFP_KERNEL allocations, if all nodes in the | |
2247 | * cpuset are short of memory, might require taking the callback_mutex | |
2248 | * mutex. | |
9bf2229f | 2249 | * |
36be57ff | 2250 | * The first call here from mm/page_alloc:get_page_from_freelist() |
02a0e53d PJ |
2251 | * has __GFP_HARDWALL set in gfp_mask, enforcing hardwall cpusets, |
2252 | * so no allocation on a node outside the cpuset is allowed (unless | |
2253 | * in interrupt, of course). | |
36be57ff PJ |
2254 | * |
2255 | * The second pass through get_page_from_freelist() doesn't even call | |
2256 | * here for GFP_ATOMIC calls. For those calls, the __alloc_pages() | |
2257 | * variable 'wait' is not set, and the bit ALLOC_CPUSET is not set | |
2258 | * in alloc_flags. That logic and the checks below have the combined | |
2259 | * affect that: | |
9bf2229f PJ |
2260 | * in_interrupt - any node ok (current task context irrelevant) |
2261 | * GFP_ATOMIC - any node ok | |
c596d9f3 | 2262 | * TIF_MEMDIE - any node ok |
78608366 | 2263 | * GFP_KERNEL - any node in enclosing hardwalled cpuset ok |
9bf2229f | 2264 | * GFP_USER - only nodes in current tasks mems allowed ok. |
36be57ff PJ |
2265 | * |
2266 | * Rule: | |
a1bc5a4e | 2267 | * Don't call cpuset_node_allowed_softwall if you can't sleep, unless you |
36be57ff PJ |
2268 | * pass in the __GFP_HARDWALL flag set in gfp_flag, which disables |
2269 | * the code that might scan up ancestor cpusets and sleep. | |
02a0e53d | 2270 | */ |
a1bc5a4e | 2271 | int __cpuset_node_allowed_softwall(int node, gfp_t gfp_mask) |
1da177e4 | 2272 | { |
9bf2229f | 2273 | const struct cpuset *cs; /* current cpuset ancestors */ |
29afd49b | 2274 | int allowed; /* is allocation in zone z allowed? */ |
9bf2229f | 2275 | |
9b819d20 | 2276 | if (in_interrupt() || (gfp_mask & __GFP_THISNODE)) |
9bf2229f | 2277 | return 1; |
92d1dbd2 | 2278 | might_sleep_if(!(gfp_mask & __GFP_HARDWALL)); |
9bf2229f PJ |
2279 | if (node_isset(node, current->mems_allowed)) |
2280 | return 1; | |
c596d9f3 DR |
2281 | /* |
2282 | * Allow tasks that have access to memory reserves because they have | |
2283 | * been OOM killed to get memory anywhere. | |
2284 | */ | |
2285 | if (unlikely(test_thread_flag(TIF_MEMDIE))) | |
2286 | return 1; | |
9bf2229f PJ |
2287 | if (gfp_mask & __GFP_HARDWALL) /* If hardwall request, stop here */ |
2288 | return 0; | |
2289 | ||
5563e770 BP |
2290 | if (current->flags & PF_EXITING) /* Let dying task have memory */ |
2291 | return 1; | |
2292 | ||
9bf2229f | 2293 | /* Not hardwall and node outside mems_allowed: scan up cpusets */ |
3d3f26a7 | 2294 | mutex_lock(&callback_mutex); |
053199ed | 2295 | |
053199ed | 2296 | task_lock(current); |
78608366 | 2297 | cs = nearest_hardwall_ancestor(task_cs(current)); |
053199ed PJ |
2298 | task_unlock(current); |
2299 | ||
9bf2229f | 2300 | allowed = node_isset(node, cs->mems_allowed); |
3d3f26a7 | 2301 | mutex_unlock(&callback_mutex); |
9bf2229f | 2302 | return allowed; |
1da177e4 LT |
2303 | } |
2304 | ||
02a0e53d | 2305 | /* |
a1bc5a4e DR |
2306 | * cpuset_node_allowed_hardwall - Can we allocate on a memory node? |
2307 | * @node: is this an allowed node? | |
02a0e53d PJ |
2308 | * @gfp_mask: memory allocation flags |
2309 | * | |
a1bc5a4e DR |
2310 | * If we're in interrupt, yes, we can always allocate. If __GFP_THISNODE is |
2311 | * set, yes, we can always allocate. If node is in our task's mems_allowed, | |
2312 | * yes. If the task has been OOM killed and has access to memory reserves as | |
2313 | * specified by the TIF_MEMDIE flag, yes. | |
2314 | * Otherwise, no. | |
02a0e53d PJ |
2315 | * |
2316 | * The __GFP_THISNODE placement logic is really handled elsewhere, | |
2317 | * by forcibly using a zonelist starting at a specified node, and by | |
2318 | * (in get_page_from_freelist()) refusing to consider the zones for | |
2319 | * any node on the zonelist except the first. By the time any such | |
2320 | * calls get to this routine, we should just shut up and say 'yes'. | |
2321 | * | |
a1bc5a4e DR |
2322 | * Unlike the cpuset_node_allowed_softwall() variant, above, |
2323 | * this variant requires that the node be in the current task's | |
02a0e53d PJ |
2324 | * mems_allowed or that we're in interrupt. It does not scan up the |
2325 | * cpuset hierarchy for the nearest enclosing mem_exclusive cpuset. | |
2326 | * It never sleeps. | |
2327 | */ | |
a1bc5a4e | 2328 | int __cpuset_node_allowed_hardwall(int node, gfp_t gfp_mask) |
02a0e53d | 2329 | { |
02a0e53d PJ |
2330 | if (in_interrupt() || (gfp_mask & __GFP_THISNODE)) |
2331 | return 1; | |
02a0e53d PJ |
2332 | if (node_isset(node, current->mems_allowed)) |
2333 | return 1; | |
dedf8b79 DW |
2334 | /* |
2335 | * Allow tasks that have access to memory reserves because they have | |
2336 | * been OOM killed to get memory anywhere. | |
2337 | */ | |
2338 | if (unlikely(test_thread_flag(TIF_MEMDIE))) | |
2339 | return 1; | |
02a0e53d PJ |
2340 | return 0; |
2341 | } | |
2342 | ||
505970b9 PJ |
2343 | /** |
2344 | * cpuset_lock - lock out any changes to cpuset structures | |
2345 | * | |
3d3f26a7 | 2346 | * The out of memory (oom) code needs to mutex_lock cpusets |
505970b9 | 2347 | * from being changed while it scans the tasklist looking for a |
3d3f26a7 | 2348 | * task in an overlapping cpuset. Expose callback_mutex via this |
505970b9 PJ |
2349 | * cpuset_lock() routine, so the oom code can lock it, before |
2350 | * locking the task list. The tasklist_lock is a spinlock, so | |
3d3f26a7 | 2351 | * must be taken inside callback_mutex. |
505970b9 PJ |
2352 | */ |
2353 | ||
2354 | void cpuset_lock(void) | |
2355 | { | |
3d3f26a7 | 2356 | mutex_lock(&callback_mutex); |
505970b9 PJ |
2357 | } |
2358 | ||
2359 | /** | |
2360 | * cpuset_unlock - release lock on cpuset changes | |
2361 | * | |
2362 | * Undo the lock taken in a previous cpuset_lock() call. | |
2363 | */ | |
2364 | ||
2365 | void cpuset_unlock(void) | |
2366 | { | |
3d3f26a7 | 2367 | mutex_unlock(&callback_mutex); |
505970b9 PJ |
2368 | } |
2369 | ||
825a46af PJ |
2370 | /** |
2371 | * cpuset_mem_spread_node() - On which node to begin search for a page | |
2372 | * | |
2373 | * If a task is marked PF_SPREAD_PAGE or PF_SPREAD_SLAB (as for | |
2374 | * tasks in a cpuset with is_spread_page or is_spread_slab set), | |
2375 | * and if the memory allocation used cpuset_mem_spread_node() | |
2376 | * to determine on which node to start looking, as it will for | |
2377 | * certain page cache or slab cache pages such as used for file | |
2378 | * system buffers and inode caches, then instead of starting on the | |
2379 | * local node to look for a free page, rather spread the starting | |
2380 | * node around the tasks mems_allowed nodes. | |
2381 | * | |
2382 | * We don't have to worry about the returned node being offline | |
2383 | * because "it can't happen", and even if it did, it would be ok. | |
2384 | * | |
2385 | * The routines calling guarantee_online_mems() are careful to | |
2386 | * only set nodes in task->mems_allowed that are online. So it | |
2387 | * should not be possible for the following code to return an | |
2388 | * offline node. But if it did, that would be ok, as this routine | |
2389 | * is not returning the node where the allocation must be, only | |
2390 | * the node where the search should start. The zonelist passed to | |
2391 | * __alloc_pages() will include all nodes. If the slab allocator | |
2392 | * is passed an offline node, it will fall back to the local node. | |
2393 | * See kmem_cache_alloc_node(). | |
2394 | */ | |
2395 | ||
2396 | int cpuset_mem_spread_node(void) | |
2397 | { | |
2398 | int node; | |
2399 | ||
2400 | node = next_node(current->cpuset_mem_spread_rotor, current->mems_allowed); | |
2401 | if (node == MAX_NUMNODES) | |
2402 | node = first_node(current->mems_allowed); | |
2403 | current->cpuset_mem_spread_rotor = node; | |
2404 | return node; | |
2405 | } | |
2406 | EXPORT_SYMBOL_GPL(cpuset_mem_spread_node); | |
2407 | ||
ef08e3b4 | 2408 | /** |
bbe373f2 DR |
2409 | * cpuset_mems_allowed_intersects - Does @tsk1's mems_allowed intersect @tsk2's? |
2410 | * @tsk1: pointer to task_struct of some task. | |
2411 | * @tsk2: pointer to task_struct of some other task. | |
2412 | * | |
2413 | * Description: Return true if @tsk1's mems_allowed intersects the | |
2414 | * mems_allowed of @tsk2. Used by the OOM killer to determine if | |
2415 | * one of the task's memory usage might impact the memory available | |
2416 | * to the other. | |
ef08e3b4 PJ |
2417 | **/ |
2418 | ||
bbe373f2 DR |
2419 | int cpuset_mems_allowed_intersects(const struct task_struct *tsk1, |
2420 | const struct task_struct *tsk2) | |
ef08e3b4 | 2421 | { |
bbe373f2 | 2422 | return nodes_intersects(tsk1->mems_allowed, tsk2->mems_allowed); |
ef08e3b4 PJ |
2423 | } |
2424 | ||
75aa1994 DR |
2425 | /** |
2426 | * cpuset_print_task_mems_allowed - prints task's cpuset and mems_allowed | |
2427 | * @task: pointer to task_struct of some task. | |
2428 | * | |
2429 | * Description: Prints @task's name, cpuset name, and cached copy of its | |
2430 | * mems_allowed to the kernel log. Must hold task_lock(task) to allow | |
2431 | * dereferencing task_cs(task). | |
2432 | */ | |
2433 | void cpuset_print_task_mems_allowed(struct task_struct *tsk) | |
2434 | { | |
2435 | struct dentry *dentry; | |
2436 | ||
2437 | dentry = task_cs(tsk)->css.cgroup->dentry; | |
2438 | spin_lock(&cpuset_buffer_lock); | |
2439 | snprintf(cpuset_name, CPUSET_NAME_LEN, | |
2440 | dentry ? (const char *)dentry->d_name.name : "/"); | |
2441 | nodelist_scnprintf(cpuset_nodelist, CPUSET_NODELIST_LEN, | |
2442 | tsk->mems_allowed); | |
2443 | printk(KERN_INFO "%s cpuset=%s mems_allowed=%s\n", | |
2444 | tsk->comm, cpuset_name, cpuset_nodelist); | |
2445 | spin_unlock(&cpuset_buffer_lock); | |
2446 | } | |
2447 | ||
3e0d98b9 PJ |
2448 | /* |
2449 | * Collection of memory_pressure is suppressed unless | |
2450 | * this flag is enabled by writing "1" to the special | |
2451 | * cpuset file 'memory_pressure_enabled' in the root cpuset. | |
2452 | */ | |
2453 | ||
c5b2aff8 | 2454 | int cpuset_memory_pressure_enabled __read_mostly; |
3e0d98b9 PJ |
2455 | |
2456 | /** | |
2457 | * cpuset_memory_pressure_bump - keep stats of per-cpuset reclaims. | |
2458 | * | |
2459 | * Keep a running average of the rate of synchronous (direct) | |
2460 | * page reclaim efforts initiated by tasks in each cpuset. | |
2461 | * | |
2462 | * This represents the rate at which some task in the cpuset | |
2463 | * ran low on memory on all nodes it was allowed to use, and | |
2464 | * had to enter the kernels page reclaim code in an effort to | |
2465 | * create more free memory by tossing clean pages or swapping | |
2466 | * or writing dirty pages. | |
2467 | * | |
2468 | * Display to user space in the per-cpuset read-only file | |
2469 | * "memory_pressure". Value displayed is an integer | |
2470 | * representing the recent rate of entry into the synchronous | |
2471 | * (direct) page reclaim by any task attached to the cpuset. | |
2472 | **/ | |
2473 | ||
2474 | void __cpuset_memory_pressure_bump(void) | |
2475 | { | |
3e0d98b9 | 2476 | task_lock(current); |
8793d854 | 2477 | fmeter_markevent(&task_cs(current)->fmeter); |
3e0d98b9 PJ |
2478 | task_unlock(current); |
2479 | } | |
2480 | ||
8793d854 | 2481 | #ifdef CONFIG_PROC_PID_CPUSET |
1da177e4 LT |
2482 | /* |
2483 | * proc_cpuset_show() | |
2484 | * - Print tasks cpuset path into seq_file. | |
2485 | * - Used for /proc/<pid>/cpuset. | |
053199ed PJ |
2486 | * - No need to task_lock(tsk) on this tsk->cpuset reference, as it |
2487 | * doesn't really matter if tsk->cpuset changes after we read it, | |
c8d9c90c | 2488 | * and we take cgroup_mutex, keeping cpuset_attach() from changing it |
2df167a3 | 2489 | * anyway. |
1da177e4 | 2490 | */ |
029190c5 | 2491 | static int proc_cpuset_show(struct seq_file *m, void *unused_v) |
1da177e4 | 2492 | { |
13b41b09 | 2493 | struct pid *pid; |
1da177e4 LT |
2494 | struct task_struct *tsk; |
2495 | char *buf; | |
8793d854 | 2496 | struct cgroup_subsys_state *css; |
99f89551 | 2497 | int retval; |
1da177e4 | 2498 | |
99f89551 | 2499 | retval = -ENOMEM; |
1da177e4 LT |
2500 | buf = kmalloc(PAGE_SIZE, GFP_KERNEL); |
2501 | if (!buf) | |
99f89551 EB |
2502 | goto out; |
2503 | ||
2504 | retval = -ESRCH; | |
13b41b09 EB |
2505 | pid = m->private; |
2506 | tsk = get_pid_task(pid, PIDTYPE_PID); | |
99f89551 EB |
2507 | if (!tsk) |
2508 | goto out_free; | |
1da177e4 | 2509 | |
99f89551 | 2510 | retval = -EINVAL; |
8793d854 PM |
2511 | cgroup_lock(); |
2512 | css = task_subsys_state(tsk, cpuset_subsys_id); | |
2513 | retval = cgroup_path(css->cgroup, buf, PAGE_SIZE); | |
1da177e4 | 2514 | if (retval < 0) |
99f89551 | 2515 | goto out_unlock; |
1da177e4 LT |
2516 | seq_puts(m, buf); |
2517 | seq_putc(m, '\n'); | |
99f89551 | 2518 | out_unlock: |
8793d854 | 2519 | cgroup_unlock(); |
99f89551 EB |
2520 | put_task_struct(tsk); |
2521 | out_free: | |
1da177e4 | 2522 | kfree(buf); |
99f89551 | 2523 | out: |
1da177e4 LT |
2524 | return retval; |
2525 | } | |
2526 | ||
2527 | static int cpuset_open(struct inode *inode, struct file *file) | |
2528 | { | |
13b41b09 EB |
2529 | struct pid *pid = PROC_I(inode)->pid; |
2530 | return single_open(file, proc_cpuset_show, pid); | |
1da177e4 LT |
2531 | } |
2532 | ||
9a32144e | 2533 | const struct file_operations proc_cpuset_operations = { |
1da177e4 LT |
2534 | .open = cpuset_open, |
2535 | .read = seq_read, | |
2536 | .llseek = seq_lseek, | |
2537 | .release = single_release, | |
2538 | }; | |
8793d854 | 2539 | #endif /* CONFIG_PROC_PID_CPUSET */ |
1da177e4 | 2540 | |
d01d4827 | 2541 | /* Display task mems_allowed in /proc/<pid>/status file. */ |
df5f8314 EB |
2542 | void cpuset_task_status_allowed(struct seq_file *m, struct task_struct *task) |
2543 | { | |
df5f8314 | 2544 | seq_printf(m, "Mems_allowed:\t"); |
30e8e136 | 2545 | seq_nodemask(m, &task->mems_allowed); |
df5f8314 | 2546 | seq_printf(m, "\n"); |
39106dcf | 2547 | seq_printf(m, "Mems_allowed_list:\t"); |
30e8e136 | 2548 | seq_nodemask_list(m, &task->mems_allowed); |
39106dcf | 2549 | seq_printf(m, "\n"); |
1da177e4 | 2550 | } |