]>
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> |
9984de1a | 40 | #include <linux/export.h> |
1da177e4 LT |
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 | 46 | #include <linux/sched.h> |
6e84f315 | 47 | #include <linux/sched/mm.h> |
f719ff9b | 48 | #include <linux/sched/task.h> |
1da177e4 | 49 | #include <linux/seq_file.h> |
22fb52dd | 50 | #include <linux/security.h> |
1da177e4 | 51 | #include <linux/slab.h> |
1da177e4 LT |
52 | #include <linux/spinlock.h> |
53 | #include <linux/stat.h> | |
54 | #include <linux/string.h> | |
55 | #include <linux/time.h> | |
d2b43658 | 56 | #include <linux/time64.h> |
1da177e4 LT |
57 | #include <linux/backing-dev.h> |
58 | #include <linux/sort.h> | |
59 | ||
7c0f6ba6 | 60 | #include <linux/uaccess.h> |
60063497 | 61 | #include <linux/atomic.h> |
3d3f26a7 | 62 | #include <linux/mutex.h> |
956db3ca | 63 | #include <linux/cgroup.h> |
e44193d3 | 64 | #include <linux/wait.h> |
1da177e4 | 65 | |
89affbf5 | 66 | DEFINE_STATIC_KEY_FALSE(cpusets_pre_enable_key); |
002f2906 | 67 | DEFINE_STATIC_KEY_FALSE(cpusets_enabled_key); |
202f72d5 | 68 | |
3e0d98b9 PJ |
69 | /* See "Frequency meter" comments, below. */ |
70 | ||
71 | struct fmeter { | |
72 | int cnt; /* unprocessed events count */ | |
73 | int val; /* most recent output value */ | |
d2b43658 | 74 | time64_t time; /* clock (secs) when val computed */ |
3e0d98b9 PJ |
75 | spinlock_t lock; /* guards read or write of above */ |
76 | }; | |
77 | ||
1da177e4 | 78 | struct cpuset { |
8793d854 PM |
79 | struct cgroup_subsys_state css; |
80 | ||
1da177e4 | 81 | unsigned long flags; /* "unsigned long" so bitops work */ |
e2b9a3d7 | 82 | |
7e88291b LZ |
83 | /* |
84 | * On default hierarchy: | |
85 | * | |
86 | * The user-configured masks can only be changed by writing to | |
87 | * cpuset.cpus and cpuset.mems, and won't be limited by the | |
88 | * parent masks. | |
89 | * | |
90 | * The effective masks is the real masks that apply to the tasks | |
91 | * in the cpuset. They may be changed if the configured masks are | |
92 | * changed or hotplug happens. | |
93 | * | |
94 | * effective_mask == configured_mask & parent's effective_mask, | |
95 | * and if it ends up empty, it will inherit the parent's mask. | |
96 | * | |
97 | * | |
98 | * On legacy hierachy: | |
99 | * | |
100 | * The user-configured masks are always the same with effective masks. | |
101 | */ | |
102 | ||
e2b9a3d7 LZ |
103 | /* user-configured CPUs and Memory Nodes allow to tasks */ |
104 | cpumask_var_t cpus_allowed; | |
105 | nodemask_t mems_allowed; | |
106 | ||
107 | /* effective CPUs and Memory Nodes allow to tasks */ | |
108 | cpumask_var_t effective_cpus; | |
109 | nodemask_t effective_mems; | |
1da177e4 | 110 | |
33ad801d LZ |
111 | /* |
112 | * This is old Memory Nodes tasks took on. | |
113 | * | |
114 | * - top_cpuset.old_mems_allowed is initialized to mems_allowed. | |
115 | * - A new cpuset's old_mems_allowed is initialized when some | |
116 | * task is moved into it. | |
117 | * - old_mems_allowed is used in cpuset_migrate_mm() when we change | |
118 | * cpuset.mems_allowed and have tasks' nodemask updated, and | |
119 | * then old_mems_allowed is updated to mems_allowed. | |
120 | */ | |
121 | nodemask_t old_mems_allowed; | |
122 | ||
3e0d98b9 | 123 | struct fmeter fmeter; /* memory_pressure filter */ |
029190c5 | 124 | |
452477fa TH |
125 | /* |
126 | * Tasks are being attached to this cpuset. Used to prevent | |
127 | * zeroing cpus/mems_allowed between ->can_attach() and ->attach(). | |
128 | */ | |
129 | int attach_in_progress; | |
130 | ||
029190c5 PJ |
131 | /* partition number for rebuild_sched_domains() */ |
132 | int pn; | |
956db3ca | 133 | |
1d3504fc HS |
134 | /* for custom sched domain */ |
135 | int relax_domain_level; | |
1da177e4 LT |
136 | }; |
137 | ||
a7c6d554 | 138 | static inline struct cpuset *css_cs(struct cgroup_subsys_state *css) |
8793d854 | 139 | { |
a7c6d554 | 140 | return css ? container_of(css, struct cpuset, css) : NULL; |
8793d854 PM |
141 | } |
142 | ||
143 | /* Retrieve the cpuset for a task */ | |
144 | static inline struct cpuset *task_cs(struct task_struct *task) | |
145 | { | |
073219e9 | 146 | return css_cs(task_css(task, cpuset_cgrp_id)); |
8793d854 | 147 | } |
8793d854 | 148 | |
c9710d80 | 149 | static inline struct cpuset *parent_cs(struct cpuset *cs) |
c431069f | 150 | { |
5c9d535b | 151 | return css_cs(cs->css.parent); |
c431069f TH |
152 | } |
153 | ||
b246272e DR |
154 | #ifdef CONFIG_NUMA |
155 | static inline bool task_has_mempolicy(struct task_struct *task) | |
156 | { | |
157 | return task->mempolicy; | |
158 | } | |
159 | #else | |
160 | static inline bool task_has_mempolicy(struct task_struct *task) | |
161 | { | |
162 | return false; | |
163 | } | |
164 | #endif | |
165 | ||
166 | ||
1da177e4 LT |
167 | /* bits in struct cpuset flags field */ |
168 | typedef enum { | |
efeb77b2 | 169 | CS_ONLINE, |
1da177e4 LT |
170 | CS_CPU_EXCLUSIVE, |
171 | CS_MEM_EXCLUSIVE, | |
78608366 | 172 | CS_MEM_HARDWALL, |
45b07ef3 | 173 | CS_MEMORY_MIGRATE, |
029190c5 | 174 | CS_SCHED_LOAD_BALANCE, |
825a46af PJ |
175 | CS_SPREAD_PAGE, |
176 | CS_SPREAD_SLAB, | |
1da177e4 LT |
177 | } cpuset_flagbits_t; |
178 | ||
179 | /* convenient tests for these bits */ | |
41c25707 | 180 | static inline bool is_cpuset_online(struct cpuset *cs) |
efeb77b2 | 181 | { |
41c25707 | 182 | return test_bit(CS_ONLINE, &cs->flags) && !css_is_dying(&cs->css); |
efeb77b2 TH |
183 | } |
184 | ||
1da177e4 LT |
185 | static inline int is_cpu_exclusive(const struct cpuset *cs) |
186 | { | |
7b5b9ef0 | 187 | return test_bit(CS_CPU_EXCLUSIVE, &cs->flags); |
1da177e4 LT |
188 | } |
189 | ||
190 | static inline int is_mem_exclusive(const struct cpuset *cs) | |
191 | { | |
7b5b9ef0 | 192 | return test_bit(CS_MEM_EXCLUSIVE, &cs->flags); |
1da177e4 LT |
193 | } |
194 | ||
78608366 PM |
195 | static inline int is_mem_hardwall(const struct cpuset *cs) |
196 | { | |
197 | return test_bit(CS_MEM_HARDWALL, &cs->flags); | |
198 | } | |
199 | ||
029190c5 PJ |
200 | static inline int is_sched_load_balance(const struct cpuset *cs) |
201 | { | |
202 | return test_bit(CS_SCHED_LOAD_BALANCE, &cs->flags); | |
203 | } | |
204 | ||
45b07ef3 PJ |
205 | static inline int is_memory_migrate(const struct cpuset *cs) |
206 | { | |
7b5b9ef0 | 207 | return test_bit(CS_MEMORY_MIGRATE, &cs->flags); |
45b07ef3 PJ |
208 | } |
209 | ||
825a46af PJ |
210 | static inline int is_spread_page(const struct cpuset *cs) |
211 | { | |
212 | return test_bit(CS_SPREAD_PAGE, &cs->flags); | |
213 | } | |
214 | ||
215 | static inline int is_spread_slab(const struct cpuset *cs) | |
216 | { | |
217 | return test_bit(CS_SPREAD_SLAB, &cs->flags); | |
218 | } | |
219 | ||
1da177e4 | 220 | static struct cpuset top_cpuset = { |
efeb77b2 TH |
221 | .flags = ((1 << CS_ONLINE) | (1 << CS_CPU_EXCLUSIVE) | |
222 | (1 << CS_MEM_EXCLUSIVE)), | |
1da177e4 LT |
223 | }; |
224 | ||
ae8086ce TH |
225 | /** |
226 | * cpuset_for_each_child - traverse online children of a cpuset | |
227 | * @child_cs: loop cursor pointing to the current child | |
492eb21b | 228 | * @pos_css: used for iteration |
ae8086ce TH |
229 | * @parent_cs: target cpuset to walk children of |
230 | * | |
231 | * Walk @child_cs through the online children of @parent_cs. Must be used | |
232 | * with RCU read locked. | |
233 | */ | |
492eb21b TH |
234 | #define cpuset_for_each_child(child_cs, pos_css, parent_cs) \ |
235 | css_for_each_child((pos_css), &(parent_cs)->css) \ | |
236 | if (is_cpuset_online(((child_cs) = css_cs((pos_css))))) | |
ae8086ce | 237 | |
fc560a26 TH |
238 | /** |
239 | * cpuset_for_each_descendant_pre - pre-order walk of a cpuset's descendants | |
240 | * @des_cs: loop cursor pointing to the current descendant | |
492eb21b | 241 | * @pos_css: used for iteration |
fc560a26 TH |
242 | * @root_cs: target cpuset to walk ancestor of |
243 | * | |
244 | * Walk @des_cs through the online descendants of @root_cs. Must be used | |
492eb21b | 245 | * with RCU read locked. The caller may modify @pos_css by calling |
bd8815a6 TH |
246 | * css_rightmost_descendant() to skip subtree. @root_cs is included in the |
247 | * iteration and the first node to be visited. | |
fc560a26 | 248 | */ |
492eb21b TH |
249 | #define cpuset_for_each_descendant_pre(des_cs, pos_css, root_cs) \ |
250 | css_for_each_descendant_pre((pos_css), &(root_cs)->css) \ | |
251 | if (is_cpuset_online(((des_cs) = css_cs((pos_css))))) | |
fc560a26 | 252 | |
1da177e4 | 253 | /* |
8447a0fe VD |
254 | * There are two global locks guarding cpuset structures - cpuset_mutex and |
255 | * callback_lock. We also require taking task_lock() when dereferencing a | |
256 | * task's cpuset pointer. See "The task_lock() exception", at the end of this | |
257 | * comment. | |
5d21cc2d | 258 | * |
8447a0fe | 259 | * A task must hold both locks to modify cpusets. If a task holds |
5d21cc2d | 260 | * cpuset_mutex, then it blocks others wanting that mutex, ensuring that it |
8447a0fe | 261 | * is the only task able to also acquire callback_lock and be able to |
5d21cc2d TH |
262 | * modify cpusets. It can perform various checks on the cpuset structure |
263 | * first, knowing nothing will change. It can also allocate memory while | |
264 | * just holding cpuset_mutex. While it is performing these checks, various | |
8447a0fe VD |
265 | * callback routines can briefly acquire callback_lock to query cpusets. |
266 | * Once it is ready to make the changes, it takes callback_lock, blocking | |
5d21cc2d | 267 | * everyone else. |
053199ed PJ |
268 | * |
269 | * Calls to the kernel memory allocator can not be made while holding | |
8447a0fe | 270 | * callback_lock, as that would risk double tripping on callback_lock |
053199ed PJ |
271 | * from one of the callbacks into the cpuset code from within |
272 | * __alloc_pages(). | |
273 | * | |
8447a0fe | 274 | * If a task is only holding callback_lock, then it has read-only |
053199ed PJ |
275 | * access to cpusets. |
276 | * | |
58568d2a MX |
277 | * Now, the task_struct fields mems_allowed and mempolicy may be changed |
278 | * by other task, we use alloc_lock in the task_struct fields to protect | |
279 | * them. | |
053199ed | 280 | * |
8447a0fe | 281 | * The cpuset_common_file_read() handlers only hold callback_lock across |
053199ed PJ |
282 | * small pieces of code, such as when reading out possibly multi-word |
283 | * cpumasks and nodemasks. | |
284 | * | |
2df167a3 PM |
285 | * Accessing a task's cpuset should be done in accordance with the |
286 | * guidelines for accessing subsystem state in kernel/cgroup.c | |
1da177e4 LT |
287 | */ |
288 | ||
5d21cc2d | 289 | static DEFINE_MUTEX(cpuset_mutex); |
8447a0fe | 290 | static DEFINE_SPINLOCK(callback_lock); |
4247bdc6 | 291 | |
e93ad19d TH |
292 | static struct workqueue_struct *cpuset_migrate_mm_wq; |
293 | ||
3a5a6d0c TH |
294 | /* |
295 | * CPU / memory hotplug is handled asynchronously. | |
296 | */ | |
297 | static void cpuset_hotplug_workfn(struct work_struct *work); | |
3a5a6d0c TH |
298 | static DECLARE_WORK(cpuset_hotplug_work, cpuset_hotplug_workfn); |
299 | ||
e44193d3 LZ |
300 | static DECLARE_WAIT_QUEUE_HEAD(cpuset_attach_wq); |
301 | ||
cf417141 MK |
302 | /* |
303 | * This is ugly, but preserves the userspace API for existing cpuset | |
8793d854 | 304 | * users. If someone tries to mount the "cpuset" filesystem, we |
cf417141 MK |
305 | * silently switch it to mount "cgroup" instead |
306 | */ | |
f7e83571 AV |
307 | static struct dentry *cpuset_mount(struct file_system_type *fs_type, |
308 | int flags, const char *unused_dev_name, void *data) | |
1da177e4 | 309 | { |
8793d854 | 310 | struct file_system_type *cgroup_fs = get_fs_type("cgroup"); |
f7e83571 | 311 | struct dentry *ret = ERR_PTR(-ENODEV); |
8793d854 PM |
312 | if (cgroup_fs) { |
313 | char mountopts[] = | |
314 | "cpuset,noprefix," | |
315 | "release_agent=/sbin/cpuset_release_agent"; | |
f7e83571 AV |
316 | ret = cgroup_fs->mount(cgroup_fs, flags, |
317 | unused_dev_name, mountopts); | |
8793d854 PM |
318 | put_filesystem(cgroup_fs); |
319 | } | |
320 | return ret; | |
1da177e4 LT |
321 | } |
322 | ||
323 | static struct file_system_type cpuset_fs_type = { | |
324 | .name = "cpuset", | |
f7e83571 | 325 | .mount = cpuset_mount, |
1da177e4 LT |
326 | }; |
327 | ||
1da177e4 | 328 | /* |
300ed6cb | 329 | * Return in pmask the portion of a cpusets's cpus_allowed that |
1da177e4 | 330 | * are online. If none are online, walk up the cpuset hierarchy |
28b89b9e | 331 | * until we find one that does have some online cpus. |
1da177e4 LT |
332 | * |
333 | * One way or another, we guarantee to return some non-empty subset | |
5f054e31 | 334 | * of cpu_online_mask. |
1da177e4 | 335 | * |
8447a0fe | 336 | * Call with callback_lock or cpuset_mutex held. |
1da177e4 | 337 | */ |
c9710d80 | 338 | static void guarantee_online_cpus(struct cpuset *cs, struct cpumask *pmask) |
1da177e4 | 339 | { |
28b89b9e | 340 | while (!cpumask_intersects(cs->effective_cpus, cpu_online_mask)) { |
c431069f | 341 | cs = parent_cs(cs); |
28b89b9e JP |
342 | if (unlikely(!cs)) { |
343 | /* | |
344 | * The top cpuset doesn't have any online cpu as a | |
345 | * consequence of a race between cpuset_hotplug_work | |
346 | * and cpu hotplug notifier. But we know the top | |
347 | * cpuset's effective_cpus is on its way to to be | |
348 | * identical to cpu_online_mask. | |
349 | */ | |
350 | cpumask_copy(pmask, cpu_online_mask); | |
351 | return; | |
352 | } | |
353 | } | |
ae1c8023 | 354 | cpumask_and(pmask, cs->effective_cpus, cpu_online_mask); |
1da177e4 LT |
355 | } |
356 | ||
357 | /* | |
358 | * Return in *pmask the portion of a cpusets's mems_allowed that | |
0e1e7c7a CL |
359 | * are online, with memory. If none are online with memory, walk |
360 | * up the cpuset hierarchy until we find one that does have some | |
40df2deb | 361 | * online mems. The top cpuset always has some mems online. |
1da177e4 LT |
362 | * |
363 | * One way or another, we guarantee to return some non-empty subset | |
38d7bee9 | 364 | * of node_states[N_MEMORY]. |
1da177e4 | 365 | * |
8447a0fe | 366 | * Call with callback_lock or cpuset_mutex held. |
1da177e4 | 367 | */ |
c9710d80 | 368 | static void guarantee_online_mems(struct cpuset *cs, nodemask_t *pmask) |
1da177e4 | 369 | { |
ae1c8023 | 370 | while (!nodes_intersects(cs->effective_mems, node_states[N_MEMORY])) |
c431069f | 371 | cs = parent_cs(cs); |
ae1c8023 | 372 | nodes_and(*pmask, cs->effective_mems, node_states[N_MEMORY]); |
1da177e4 LT |
373 | } |
374 | ||
f3b39d47 MX |
375 | /* |
376 | * update task's spread flag if cpuset's page/slab spread flag is set | |
377 | * | |
8447a0fe | 378 | * Call with callback_lock or cpuset_mutex held. |
f3b39d47 MX |
379 | */ |
380 | static void cpuset_update_task_spread_flag(struct cpuset *cs, | |
381 | struct task_struct *tsk) | |
382 | { | |
383 | if (is_spread_page(cs)) | |
2ad654bc | 384 | task_set_spread_page(tsk); |
f3b39d47 | 385 | else |
2ad654bc ZL |
386 | task_clear_spread_page(tsk); |
387 | ||
f3b39d47 | 388 | if (is_spread_slab(cs)) |
2ad654bc | 389 | task_set_spread_slab(tsk); |
f3b39d47 | 390 | else |
2ad654bc | 391 | task_clear_spread_slab(tsk); |
f3b39d47 MX |
392 | } |
393 | ||
1da177e4 LT |
394 | /* |
395 | * is_cpuset_subset(p, q) - Is cpuset p a subset of cpuset q? | |
396 | * | |
397 | * One cpuset is a subset of another if all its allowed CPUs and | |
398 | * Memory Nodes are a subset of the other, and its exclusive flags | |
5d21cc2d | 399 | * are only set if the other's are set. Call holding cpuset_mutex. |
1da177e4 LT |
400 | */ |
401 | ||
402 | static int is_cpuset_subset(const struct cpuset *p, const struct cpuset *q) | |
403 | { | |
300ed6cb | 404 | return cpumask_subset(p->cpus_allowed, q->cpus_allowed) && |
1da177e4 LT |
405 | nodes_subset(p->mems_allowed, q->mems_allowed) && |
406 | is_cpu_exclusive(p) <= is_cpu_exclusive(q) && | |
407 | is_mem_exclusive(p) <= is_mem_exclusive(q); | |
408 | } | |
409 | ||
645fcc9d LZ |
410 | /** |
411 | * alloc_trial_cpuset - allocate a trial cpuset | |
412 | * @cs: the cpuset that the trial cpuset duplicates | |
413 | */ | |
c9710d80 | 414 | static struct cpuset *alloc_trial_cpuset(struct cpuset *cs) |
645fcc9d | 415 | { |
300ed6cb LZ |
416 | struct cpuset *trial; |
417 | ||
418 | trial = kmemdup(cs, sizeof(*cs), GFP_KERNEL); | |
419 | if (!trial) | |
420 | return NULL; | |
421 | ||
e2b9a3d7 LZ |
422 | if (!alloc_cpumask_var(&trial->cpus_allowed, GFP_KERNEL)) |
423 | goto free_cs; | |
424 | if (!alloc_cpumask_var(&trial->effective_cpus, GFP_KERNEL)) | |
425 | goto free_cpus; | |
300ed6cb | 426 | |
e2b9a3d7 LZ |
427 | cpumask_copy(trial->cpus_allowed, cs->cpus_allowed); |
428 | cpumask_copy(trial->effective_cpus, cs->effective_cpus); | |
300ed6cb | 429 | return trial; |
e2b9a3d7 LZ |
430 | |
431 | free_cpus: | |
432 | free_cpumask_var(trial->cpus_allowed); | |
433 | free_cs: | |
434 | kfree(trial); | |
435 | return NULL; | |
645fcc9d LZ |
436 | } |
437 | ||
438 | /** | |
439 | * free_trial_cpuset - free the trial cpuset | |
440 | * @trial: the trial cpuset to be freed | |
441 | */ | |
442 | static void free_trial_cpuset(struct cpuset *trial) | |
443 | { | |
e2b9a3d7 | 444 | free_cpumask_var(trial->effective_cpus); |
300ed6cb | 445 | free_cpumask_var(trial->cpus_allowed); |
645fcc9d LZ |
446 | kfree(trial); |
447 | } | |
448 | ||
1da177e4 LT |
449 | /* |
450 | * validate_change() - Used to validate that any proposed cpuset change | |
451 | * follows the structural rules for cpusets. | |
452 | * | |
453 | * If we replaced the flag and mask values of the current cpuset | |
454 | * (cur) with those values in the trial cpuset (trial), would | |
455 | * our various subset and exclusive rules still be valid? Presumes | |
5d21cc2d | 456 | * cpuset_mutex held. |
1da177e4 LT |
457 | * |
458 | * 'cur' is the address of an actual, in-use cpuset. Operations | |
459 | * such as list traversal that depend on the actual address of the | |
460 | * cpuset in the list must use cur below, not trial. | |
461 | * | |
462 | * 'trial' is the address of bulk structure copy of cur, with | |
463 | * perhaps one or more of the fields cpus_allowed, mems_allowed, | |
464 | * or flags changed to new, trial values. | |
465 | * | |
466 | * Return 0 if valid, -errno if not. | |
467 | */ | |
468 | ||
c9710d80 | 469 | static int validate_change(struct cpuset *cur, struct cpuset *trial) |
1da177e4 | 470 | { |
492eb21b | 471 | struct cgroup_subsys_state *css; |
1da177e4 | 472 | struct cpuset *c, *par; |
ae8086ce TH |
473 | int ret; |
474 | ||
475 | rcu_read_lock(); | |
1da177e4 LT |
476 | |
477 | /* Each of our child cpusets must be a subset of us */ | |
ae8086ce | 478 | ret = -EBUSY; |
492eb21b | 479 | cpuset_for_each_child(c, css, cur) |
ae8086ce TH |
480 | if (!is_cpuset_subset(c, trial)) |
481 | goto out; | |
1da177e4 LT |
482 | |
483 | /* Remaining checks don't apply to root cpuset */ | |
ae8086ce | 484 | ret = 0; |
69604067 | 485 | if (cur == &top_cpuset) |
ae8086ce | 486 | goto out; |
1da177e4 | 487 | |
c431069f | 488 | par = parent_cs(cur); |
69604067 | 489 | |
7e88291b | 490 | /* On legacy hiearchy, we must be a subset of our parent cpuset. */ |
ae8086ce | 491 | ret = -EACCES; |
9e10a130 TH |
492 | if (!cgroup_subsys_on_dfl(cpuset_cgrp_subsys) && |
493 | !is_cpuset_subset(trial, par)) | |
ae8086ce | 494 | goto out; |
1da177e4 | 495 | |
2df167a3 PM |
496 | /* |
497 | * If either I or some sibling (!= me) is exclusive, we can't | |
498 | * overlap | |
499 | */ | |
ae8086ce | 500 | ret = -EINVAL; |
492eb21b | 501 | cpuset_for_each_child(c, css, par) { |
1da177e4 LT |
502 | if ((is_cpu_exclusive(trial) || is_cpu_exclusive(c)) && |
503 | c != cur && | |
300ed6cb | 504 | cpumask_intersects(trial->cpus_allowed, c->cpus_allowed)) |
ae8086ce | 505 | goto out; |
1da177e4 LT |
506 | if ((is_mem_exclusive(trial) || is_mem_exclusive(c)) && |
507 | c != cur && | |
508 | nodes_intersects(trial->mems_allowed, c->mems_allowed)) | |
ae8086ce | 509 | goto out; |
1da177e4 LT |
510 | } |
511 | ||
452477fa TH |
512 | /* |
513 | * Cpusets with tasks - existing or newly being attached - can't | |
1c09b195 | 514 | * be changed to have empty cpus_allowed or mems_allowed. |
452477fa | 515 | */ |
ae8086ce | 516 | ret = -ENOSPC; |
27bd4dbb | 517 | if ((cgroup_is_populated(cur->css.cgroup) || cur->attach_in_progress)) { |
1c09b195 LZ |
518 | if (!cpumask_empty(cur->cpus_allowed) && |
519 | cpumask_empty(trial->cpus_allowed)) | |
520 | goto out; | |
521 | if (!nodes_empty(cur->mems_allowed) && | |
522 | nodes_empty(trial->mems_allowed)) | |
523 | goto out; | |
524 | } | |
020958b6 | 525 | |
f82f8042 JL |
526 | /* |
527 | * We can't shrink if we won't have enough room for SCHED_DEADLINE | |
528 | * tasks. | |
529 | */ | |
530 | ret = -EBUSY; | |
531 | if (is_cpu_exclusive(cur) && | |
532 | !cpuset_cpumask_can_shrink(cur->cpus_allowed, | |
533 | trial->cpus_allowed)) | |
534 | goto out; | |
535 | ||
ae8086ce TH |
536 | ret = 0; |
537 | out: | |
538 | rcu_read_unlock(); | |
539 | return ret; | |
1da177e4 LT |
540 | } |
541 | ||
db7f47cf | 542 | #ifdef CONFIG_SMP |
029190c5 | 543 | /* |
cf417141 | 544 | * Helper routine for generate_sched_domains(). |
8b5f1c52 | 545 | * Do cpusets a, b have overlapping effective cpus_allowed masks? |
029190c5 | 546 | */ |
029190c5 PJ |
547 | static int cpusets_overlap(struct cpuset *a, struct cpuset *b) |
548 | { | |
8b5f1c52 | 549 | return cpumask_intersects(a->effective_cpus, b->effective_cpus); |
029190c5 PJ |
550 | } |
551 | ||
1d3504fc HS |
552 | static void |
553 | update_domain_attr(struct sched_domain_attr *dattr, struct cpuset *c) | |
554 | { | |
1d3504fc HS |
555 | if (dattr->relax_domain_level < c->relax_domain_level) |
556 | dattr->relax_domain_level = c->relax_domain_level; | |
557 | return; | |
558 | } | |
559 | ||
fc560a26 TH |
560 | static void update_domain_attr_tree(struct sched_domain_attr *dattr, |
561 | struct cpuset *root_cs) | |
f5393693 | 562 | { |
fc560a26 | 563 | struct cpuset *cp; |
492eb21b | 564 | struct cgroup_subsys_state *pos_css; |
f5393693 | 565 | |
fc560a26 | 566 | rcu_read_lock(); |
492eb21b | 567 | cpuset_for_each_descendant_pre(cp, pos_css, root_cs) { |
fc560a26 TH |
568 | /* skip the whole subtree if @cp doesn't have any CPU */ |
569 | if (cpumask_empty(cp->cpus_allowed)) { | |
492eb21b | 570 | pos_css = css_rightmost_descendant(pos_css); |
f5393693 | 571 | continue; |
fc560a26 | 572 | } |
f5393693 LJ |
573 | |
574 | if (is_sched_load_balance(cp)) | |
575 | update_domain_attr(dattr, cp); | |
f5393693 | 576 | } |
fc560a26 | 577 | rcu_read_unlock(); |
f5393693 LJ |
578 | } |
579 | ||
be040bea PB |
580 | /* Must be called with cpuset_mutex held. */ |
581 | static inline int nr_cpusets(void) | |
582 | { | |
583 | /* jump label reference count + the top-level cpuset */ | |
584 | return static_key_count(&cpusets_enabled_key.key) + 1; | |
585 | } | |
586 | ||
029190c5 | 587 | /* |
cf417141 MK |
588 | * generate_sched_domains() |
589 | * | |
590 | * This function builds a partial partition of the systems CPUs | |
591 | * A 'partial partition' is a set of non-overlapping subsets whose | |
592 | * union is a subset of that set. | |
0a0fca9d | 593 | * The output of this function needs to be passed to kernel/sched/core.c |
cf417141 MK |
594 | * partition_sched_domains() routine, which will rebuild the scheduler's |
595 | * load balancing domains (sched domains) as specified by that partial | |
596 | * partition. | |
029190c5 | 597 | * |
45ce80fb | 598 | * See "What is sched_load_balance" in Documentation/cgroups/cpusets.txt |
029190c5 PJ |
599 | * for a background explanation of this. |
600 | * | |
601 | * Does not return errors, on the theory that the callers of this | |
602 | * routine would rather not worry about failures to rebuild sched | |
603 | * domains when operating in the severe memory shortage situations | |
604 | * that could cause allocation failures below. | |
605 | * | |
5d21cc2d | 606 | * Must be called with cpuset_mutex held. |
029190c5 PJ |
607 | * |
608 | * The three key local variables below are: | |
aeed6824 | 609 | * q - a linked-list queue of cpuset pointers, used to implement a |
029190c5 PJ |
610 | * top-down scan of all cpusets. This scan loads a pointer |
611 | * to each cpuset marked is_sched_load_balance into the | |
612 | * array 'csa'. For our purposes, rebuilding the schedulers | |
613 | * sched domains, we can ignore !is_sched_load_balance cpusets. | |
614 | * csa - (for CpuSet Array) Array of pointers to all the cpusets | |
615 | * that need to be load balanced, for convenient iterative | |
616 | * access by the subsequent code that finds the best partition, | |
617 | * i.e the set of domains (subsets) of CPUs such that the | |
618 | * cpus_allowed of every cpuset marked is_sched_load_balance | |
619 | * is a subset of one of these domains, while there are as | |
620 | * many such domains as possible, each as small as possible. | |
621 | * doms - Conversion of 'csa' to an array of cpumasks, for passing to | |
0a0fca9d | 622 | * the kernel/sched/core.c routine partition_sched_domains() in a |
029190c5 PJ |
623 | * convenient format, that can be easily compared to the prior |
624 | * value to determine what partition elements (sched domains) | |
625 | * were changed (added or removed.) | |
626 | * | |
627 | * Finding the best partition (set of domains): | |
628 | * The triple nested loops below over i, j, k scan over the | |
629 | * load balanced cpusets (using the array of cpuset pointers in | |
630 | * csa[]) looking for pairs of cpusets that have overlapping | |
631 | * cpus_allowed, but which don't have the same 'pn' partition | |
632 | * number and gives them in the same partition number. It keeps | |
633 | * looping on the 'restart' label until it can no longer find | |
634 | * any such pairs. | |
635 | * | |
636 | * The union of the cpus_allowed masks from the set of | |
637 | * all cpusets having the same 'pn' value then form the one | |
638 | * element of the partition (one sched domain) to be passed to | |
639 | * partition_sched_domains(). | |
640 | */ | |
acc3f5d7 | 641 | static int generate_sched_domains(cpumask_var_t **domains, |
cf417141 | 642 | struct sched_domain_attr **attributes) |
029190c5 | 643 | { |
029190c5 PJ |
644 | struct cpuset *cp; /* scans q */ |
645 | struct cpuset **csa; /* array of all cpuset ptrs */ | |
646 | int csn; /* how many cpuset ptrs in csa so far */ | |
647 | int i, j, k; /* indices for partition finding loops */ | |
acc3f5d7 | 648 | cpumask_var_t *doms; /* resulting partition; i.e. sched domains */ |
47b8ea71 | 649 | cpumask_var_t non_isolated_cpus; /* load balanced CPUs */ |
1d3504fc | 650 | struct sched_domain_attr *dattr; /* attributes for custom domains */ |
1583715d | 651 | int ndoms = 0; /* number of sched domains in result */ |
6af866af | 652 | int nslot; /* next empty doms[] struct cpumask slot */ |
492eb21b | 653 | struct cgroup_subsys_state *pos_css; |
029190c5 | 654 | |
029190c5 | 655 | doms = NULL; |
1d3504fc | 656 | dattr = NULL; |
cf417141 | 657 | csa = NULL; |
029190c5 | 658 | |
47b8ea71 RR |
659 | if (!alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL)) |
660 | goto done; | |
661 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); | |
662 | ||
029190c5 PJ |
663 | /* Special case for the 99% of systems with one, full, sched domain */ |
664 | if (is_sched_load_balance(&top_cpuset)) { | |
acc3f5d7 RR |
665 | ndoms = 1; |
666 | doms = alloc_sched_domains(ndoms); | |
029190c5 | 667 | if (!doms) |
cf417141 MK |
668 | goto done; |
669 | ||
1d3504fc HS |
670 | dattr = kmalloc(sizeof(struct sched_domain_attr), GFP_KERNEL); |
671 | if (dattr) { | |
672 | *dattr = SD_ATTR_INIT; | |
93a65575 | 673 | update_domain_attr_tree(dattr, &top_cpuset); |
1d3504fc | 674 | } |
47b8ea71 RR |
675 | cpumask_and(doms[0], top_cpuset.effective_cpus, |
676 | non_isolated_cpus); | |
cf417141 | 677 | |
cf417141 | 678 | goto done; |
029190c5 PJ |
679 | } |
680 | ||
664eedde | 681 | csa = kmalloc(nr_cpusets() * sizeof(cp), GFP_KERNEL); |
029190c5 PJ |
682 | if (!csa) |
683 | goto done; | |
684 | csn = 0; | |
685 | ||
fc560a26 | 686 | rcu_read_lock(); |
492eb21b | 687 | cpuset_for_each_descendant_pre(cp, pos_css, &top_cpuset) { |
bd8815a6 TH |
688 | if (cp == &top_cpuset) |
689 | continue; | |
f5393693 | 690 | /* |
fc560a26 TH |
691 | * Continue traversing beyond @cp iff @cp has some CPUs and |
692 | * isn't load balancing. The former is obvious. The | |
693 | * latter: All child cpusets contain a subset of the | |
694 | * parent's cpus, so just skip them, and then we call | |
695 | * update_domain_attr_tree() to calc relax_domain_level of | |
696 | * the corresponding sched domain. | |
f5393693 | 697 | */ |
fc560a26 | 698 | if (!cpumask_empty(cp->cpus_allowed) && |
47b8ea71 RR |
699 | !(is_sched_load_balance(cp) && |
700 | cpumask_intersects(cp->cpus_allowed, non_isolated_cpus))) | |
f5393693 | 701 | continue; |
489a5393 | 702 | |
fc560a26 TH |
703 | if (is_sched_load_balance(cp)) |
704 | csa[csn++] = cp; | |
705 | ||
706 | /* skip @cp's subtree */ | |
492eb21b | 707 | pos_css = css_rightmost_descendant(pos_css); |
fc560a26 TH |
708 | } |
709 | rcu_read_unlock(); | |
029190c5 PJ |
710 | |
711 | for (i = 0; i < csn; i++) | |
712 | csa[i]->pn = i; | |
713 | ndoms = csn; | |
714 | ||
715 | restart: | |
716 | /* Find the best partition (set of sched domains) */ | |
717 | for (i = 0; i < csn; i++) { | |
718 | struct cpuset *a = csa[i]; | |
719 | int apn = a->pn; | |
720 | ||
721 | for (j = 0; j < csn; j++) { | |
722 | struct cpuset *b = csa[j]; | |
723 | int bpn = b->pn; | |
724 | ||
725 | if (apn != bpn && cpusets_overlap(a, b)) { | |
726 | for (k = 0; k < csn; k++) { | |
727 | struct cpuset *c = csa[k]; | |
728 | ||
729 | if (c->pn == bpn) | |
730 | c->pn = apn; | |
731 | } | |
732 | ndoms--; /* one less element */ | |
733 | goto restart; | |
734 | } | |
735 | } | |
736 | } | |
737 | ||
cf417141 MK |
738 | /* |
739 | * Now we know how many domains to create. | |
740 | * Convert <csn, csa> to <ndoms, doms> and populate cpu masks. | |
741 | */ | |
acc3f5d7 | 742 | doms = alloc_sched_domains(ndoms); |
700018e0 | 743 | if (!doms) |
cf417141 | 744 | goto done; |
cf417141 MK |
745 | |
746 | /* | |
747 | * The rest of the code, including the scheduler, can deal with | |
748 | * dattr==NULL case. No need to abort if alloc fails. | |
749 | */ | |
1d3504fc | 750 | dattr = kmalloc(ndoms * sizeof(struct sched_domain_attr), GFP_KERNEL); |
029190c5 PJ |
751 | |
752 | for (nslot = 0, i = 0; i < csn; i++) { | |
753 | struct cpuset *a = csa[i]; | |
6af866af | 754 | struct cpumask *dp; |
029190c5 PJ |
755 | int apn = a->pn; |
756 | ||
cf417141 MK |
757 | if (apn < 0) { |
758 | /* Skip completed partitions */ | |
759 | continue; | |
760 | } | |
761 | ||
acc3f5d7 | 762 | dp = doms[nslot]; |
cf417141 MK |
763 | |
764 | if (nslot == ndoms) { | |
765 | static int warnings = 10; | |
766 | if (warnings) { | |
12d3089c FF |
767 | pr_warn("rebuild_sched_domains confused: nslot %d, ndoms %d, csn %d, i %d, apn %d\n", |
768 | nslot, ndoms, csn, i, apn); | |
cf417141 | 769 | warnings--; |
029190c5 | 770 | } |
cf417141 MK |
771 | continue; |
772 | } | |
029190c5 | 773 | |
6af866af | 774 | cpumask_clear(dp); |
cf417141 MK |
775 | if (dattr) |
776 | *(dattr + nslot) = SD_ATTR_INIT; | |
777 | for (j = i; j < csn; j++) { | |
778 | struct cpuset *b = csa[j]; | |
779 | ||
780 | if (apn == b->pn) { | |
8b5f1c52 | 781 | cpumask_or(dp, dp, b->effective_cpus); |
47b8ea71 | 782 | cpumask_and(dp, dp, non_isolated_cpus); |
cf417141 MK |
783 | if (dattr) |
784 | update_domain_attr_tree(dattr + nslot, b); | |
785 | ||
786 | /* Done with this partition */ | |
787 | b->pn = -1; | |
029190c5 | 788 | } |
029190c5 | 789 | } |
cf417141 | 790 | nslot++; |
029190c5 PJ |
791 | } |
792 | BUG_ON(nslot != ndoms); | |
793 | ||
cf417141 | 794 | done: |
47b8ea71 | 795 | free_cpumask_var(non_isolated_cpus); |
cf417141 MK |
796 | kfree(csa); |
797 | ||
700018e0 LZ |
798 | /* |
799 | * Fallback to the default domain if kmalloc() failed. | |
800 | * See comments in partition_sched_domains(). | |
801 | */ | |
802 | if (doms == NULL) | |
803 | ndoms = 1; | |
804 | ||
cf417141 MK |
805 | *domains = doms; |
806 | *attributes = dattr; | |
807 | return ndoms; | |
808 | } | |
809 | ||
810 | /* | |
811 | * Rebuild scheduler domains. | |
812 | * | |
699140ba TH |
813 | * If the flag 'sched_load_balance' of any cpuset with non-empty |
814 | * 'cpus' changes, or if the 'cpus' allowed changes in any cpuset | |
815 | * which has that flag enabled, or if any cpuset with a non-empty | |
816 | * 'cpus' is removed, then call this routine to rebuild the | |
817 | * scheduler's dynamic sched domains. | |
cf417141 | 818 | * |
5d21cc2d | 819 | * Call with cpuset_mutex held. Takes get_online_cpus(). |
cf417141 | 820 | */ |
699140ba | 821 | static void rebuild_sched_domains_locked(void) |
cf417141 MK |
822 | { |
823 | struct sched_domain_attr *attr; | |
acc3f5d7 | 824 | cpumask_var_t *doms; |
cf417141 MK |
825 | int ndoms; |
826 | ||
5d21cc2d | 827 | lockdep_assert_held(&cpuset_mutex); |
86ef5c9a | 828 | get_online_cpus(); |
cf417141 | 829 | |
5b16c2a4 LZ |
830 | /* |
831 | * We have raced with CPU hotplug. Don't do anything to avoid | |
832 | * passing doms with offlined cpu to partition_sched_domains(). | |
833 | * Anyways, hotplug work item will rebuild sched domains. | |
834 | */ | |
8b5f1c52 | 835 | if (!cpumask_equal(top_cpuset.effective_cpus, cpu_active_mask)) |
5b16c2a4 LZ |
836 | goto out; |
837 | ||
cf417141 | 838 | /* Generate domain masks and attrs */ |
cf417141 | 839 | ndoms = generate_sched_domains(&doms, &attr); |
cf417141 MK |
840 | |
841 | /* Have scheduler rebuild the domains */ | |
842 | partition_sched_domains(ndoms, doms, attr); | |
5b16c2a4 | 843 | out: |
86ef5c9a | 844 | put_online_cpus(); |
cf417141 | 845 | } |
db7f47cf | 846 | #else /* !CONFIG_SMP */ |
699140ba | 847 | static void rebuild_sched_domains_locked(void) |
db7f47cf PM |
848 | { |
849 | } | |
db7f47cf | 850 | #endif /* CONFIG_SMP */ |
029190c5 | 851 | |
cf417141 MK |
852 | void rebuild_sched_domains(void) |
853 | { | |
5d21cc2d | 854 | mutex_lock(&cpuset_mutex); |
699140ba | 855 | rebuild_sched_domains_locked(); |
5d21cc2d | 856 | mutex_unlock(&cpuset_mutex); |
029190c5 PJ |
857 | } |
858 | ||
0b2f630a MX |
859 | /** |
860 | * update_tasks_cpumask - Update the cpumasks of tasks in the cpuset. | |
861 | * @cs: the cpuset in which each task's cpus_allowed mask needs to be changed | |
0b2f630a | 862 | * |
d66393e5 TH |
863 | * Iterate through each task of @cs updating its cpus_allowed to the |
864 | * effective cpuset's. As this function is called with cpuset_mutex held, | |
865 | * cpuset membership stays stable. | |
0b2f630a | 866 | */ |
d66393e5 | 867 | static void update_tasks_cpumask(struct cpuset *cs) |
0b2f630a | 868 | { |
d66393e5 TH |
869 | struct css_task_iter it; |
870 | struct task_struct *task; | |
871 | ||
872 | css_task_iter_start(&cs->css, &it); | |
873 | while ((task = css_task_iter_next(&it))) | |
ae1c8023 | 874 | set_cpus_allowed_ptr(task, cs->effective_cpus); |
d66393e5 | 875 | css_task_iter_end(&it); |
0b2f630a MX |
876 | } |
877 | ||
5c5cc623 | 878 | /* |
734d4513 LZ |
879 | * update_cpumasks_hier - Update effective cpumasks and tasks in the subtree |
880 | * @cs: the cpuset to consider | |
881 | * @new_cpus: temp variable for calculating new effective_cpus | |
882 | * | |
883 | * When congifured cpumask is changed, the effective cpumasks of this cpuset | |
884 | * and all its descendants need to be updated. | |
5c5cc623 | 885 | * |
734d4513 | 886 | * On legacy hierachy, effective_cpus will be the same with cpu_allowed. |
5c5cc623 LZ |
887 | * |
888 | * Called with cpuset_mutex held | |
889 | */ | |
734d4513 | 890 | static void update_cpumasks_hier(struct cpuset *cs, struct cpumask *new_cpus) |
5c5cc623 LZ |
891 | { |
892 | struct cpuset *cp; | |
492eb21b | 893 | struct cgroup_subsys_state *pos_css; |
8b5f1c52 | 894 | bool need_rebuild_sched_domains = false; |
5c5cc623 LZ |
895 | |
896 | rcu_read_lock(); | |
734d4513 LZ |
897 | cpuset_for_each_descendant_pre(cp, pos_css, cs) { |
898 | struct cpuset *parent = parent_cs(cp); | |
899 | ||
900 | cpumask_and(new_cpus, cp->cpus_allowed, parent->effective_cpus); | |
901 | ||
554b0d1c LZ |
902 | /* |
903 | * If it becomes empty, inherit the effective mask of the | |
904 | * parent, which is guaranteed to have some CPUs. | |
905 | */ | |
9e10a130 TH |
906 | if (cgroup_subsys_on_dfl(cpuset_cgrp_subsys) && |
907 | cpumask_empty(new_cpus)) | |
554b0d1c LZ |
908 | cpumask_copy(new_cpus, parent->effective_cpus); |
909 | ||
734d4513 LZ |
910 | /* Skip the whole subtree if the cpumask remains the same. */ |
911 | if (cpumask_equal(new_cpus, cp->effective_cpus)) { | |
912 | pos_css = css_rightmost_descendant(pos_css); | |
913 | continue; | |
5c5cc623 | 914 | } |
734d4513 | 915 | |
ec903c0c | 916 | if (!css_tryget_online(&cp->css)) |
5c5cc623 LZ |
917 | continue; |
918 | rcu_read_unlock(); | |
919 | ||
8447a0fe | 920 | spin_lock_irq(&callback_lock); |
734d4513 | 921 | cpumask_copy(cp->effective_cpus, new_cpus); |
8447a0fe | 922 | spin_unlock_irq(&callback_lock); |
734d4513 | 923 | |
9e10a130 | 924 | WARN_ON(!cgroup_subsys_on_dfl(cpuset_cgrp_subsys) && |
734d4513 LZ |
925 | !cpumask_equal(cp->cpus_allowed, cp->effective_cpus)); |
926 | ||
d66393e5 | 927 | update_tasks_cpumask(cp); |
5c5cc623 | 928 | |
8b5f1c52 LZ |
929 | /* |
930 | * If the effective cpumask of any non-empty cpuset is changed, | |
931 | * we need to rebuild sched domains. | |
932 | */ | |
933 | if (!cpumask_empty(cp->cpus_allowed) && | |
934 | is_sched_load_balance(cp)) | |
935 | need_rebuild_sched_domains = true; | |
936 | ||
5c5cc623 LZ |
937 | rcu_read_lock(); |
938 | css_put(&cp->css); | |
939 | } | |
940 | rcu_read_unlock(); | |
8b5f1c52 LZ |
941 | |
942 | if (need_rebuild_sched_domains) | |
943 | rebuild_sched_domains_locked(); | |
5c5cc623 LZ |
944 | } |
945 | ||
58f4790b CW |
946 | /** |
947 | * update_cpumask - update the cpus_allowed mask of a cpuset and all tasks in it | |
948 | * @cs: the cpuset to consider | |
fc34ac1d | 949 | * @trialcs: trial cpuset |
58f4790b CW |
950 | * @buf: buffer of cpu numbers written to this cpuset |
951 | */ | |
645fcc9d LZ |
952 | static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs, |
953 | const char *buf) | |
1da177e4 | 954 | { |
58f4790b | 955 | int retval; |
1da177e4 | 956 | |
5f054e31 | 957 | /* top_cpuset.cpus_allowed tracks cpu_online_mask; it's read-only */ |
4c4d50f7 PJ |
958 | if (cs == &top_cpuset) |
959 | return -EACCES; | |
960 | ||
6f7f02e7 | 961 | /* |
c8d9c90c | 962 | * An empty cpus_allowed is ok only if the cpuset has no tasks. |
020958b6 PJ |
963 | * Since cpulist_parse() fails on an empty mask, we special case |
964 | * that parsing. The validate_change() call ensures that cpusets | |
965 | * with tasks have cpus. | |
6f7f02e7 | 966 | */ |
020958b6 | 967 | if (!*buf) { |
300ed6cb | 968 | cpumask_clear(trialcs->cpus_allowed); |
6f7f02e7 | 969 | } else { |
300ed6cb | 970 | retval = cpulist_parse(buf, trialcs->cpus_allowed); |
6f7f02e7 DR |
971 | if (retval < 0) |
972 | return retval; | |
37340746 | 973 | |
5d8ba82c LZ |
974 | if (!cpumask_subset(trialcs->cpus_allowed, |
975 | top_cpuset.cpus_allowed)) | |
37340746 | 976 | return -EINVAL; |
6f7f02e7 | 977 | } |
029190c5 | 978 | |
8707d8b8 | 979 | /* Nothing to do if the cpus didn't change */ |
300ed6cb | 980 | if (cpumask_equal(cs->cpus_allowed, trialcs->cpus_allowed)) |
8707d8b8 | 981 | return 0; |
58f4790b | 982 | |
a73456f3 LZ |
983 | retval = validate_change(cs, trialcs); |
984 | if (retval < 0) | |
985 | return retval; | |
986 | ||
8447a0fe | 987 | spin_lock_irq(&callback_lock); |
300ed6cb | 988 | cpumask_copy(cs->cpus_allowed, trialcs->cpus_allowed); |
8447a0fe | 989 | spin_unlock_irq(&callback_lock); |
029190c5 | 990 | |
734d4513 LZ |
991 | /* use trialcs->cpus_allowed as a temp variable */ |
992 | update_cpumasks_hier(cs, trialcs->cpus_allowed); | |
85d7b949 | 993 | return 0; |
1da177e4 LT |
994 | } |
995 | ||
e4e364e8 | 996 | /* |
e93ad19d TH |
997 | * Migrate memory region from one set of nodes to another. This is |
998 | * performed asynchronously as it can be called from process migration path | |
999 | * holding locks involved in process management. All mm migrations are | |
1000 | * performed in the queued order and can be waited for by flushing | |
1001 | * cpuset_migrate_mm_wq. | |
e4e364e8 PJ |
1002 | */ |
1003 | ||
e93ad19d TH |
1004 | struct cpuset_migrate_mm_work { |
1005 | struct work_struct work; | |
1006 | struct mm_struct *mm; | |
1007 | nodemask_t from; | |
1008 | nodemask_t to; | |
1009 | }; | |
1010 | ||
1011 | static void cpuset_migrate_mm_workfn(struct work_struct *work) | |
1012 | { | |
1013 | struct cpuset_migrate_mm_work *mwork = | |
1014 | container_of(work, struct cpuset_migrate_mm_work, work); | |
1015 | ||
1016 | /* on a wq worker, no need to worry about %current's mems_allowed */ | |
1017 | do_migrate_pages(mwork->mm, &mwork->from, &mwork->to, MPOL_MF_MOVE_ALL); | |
1018 | mmput(mwork->mm); | |
1019 | kfree(mwork); | |
1020 | } | |
1021 | ||
e4e364e8 PJ |
1022 | static void cpuset_migrate_mm(struct mm_struct *mm, const nodemask_t *from, |
1023 | const nodemask_t *to) | |
1024 | { | |
e93ad19d | 1025 | struct cpuset_migrate_mm_work *mwork; |
e4e364e8 | 1026 | |
e93ad19d TH |
1027 | mwork = kzalloc(sizeof(*mwork), GFP_KERNEL); |
1028 | if (mwork) { | |
1029 | mwork->mm = mm; | |
1030 | mwork->from = *from; | |
1031 | mwork->to = *to; | |
1032 | INIT_WORK(&mwork->work, cpuset_migrate_mm_workfn); | |
1033 | queue_work(cpuset_migrate_mm_wq, &mwork->work); | |
1034 | } else { | |
1035 | mmput(mm); | |
1036 | } | |
1037 | } | |
e4e364e8 | 1038 | |
5cf1cacb | 1039 | static void cpuset_post_attach(void) |
e93ad19d TH |
1040 | { |
1041 | flush_workqueue(cpuset_migrate_mm_wq); | |
e4e364e8 PJ |
1042 | } |
1043 | ||
3b6766fe | 1044 | /* |
58568d2a MX |
1045 | * cpuset_change_task_nodemask - change task's mems_allowed and mempolicy |
1046 | * @tsk: the task to change | |
1047 | * @newmems: new nodes that the task will be set | |
1048 | * | |
5f155f27 VB |
1049 | * We use the mems_allowed_seq seqlock to safely update both tsk->mems_allowed |
1050 | * and rebind an eventual tasks' mempolicy. If the task is allocating in | |
1051 | * parallel, it might temporarily see an empty intersection, which results in | |
1052 | * a seqlock check and retry before OOM or allocation failure. | |
58568d2a MX |
1053 | */ |
1054 | static void cpuset_change_task_nodemask(struct task_struct *tsk, | |
1055 | nodemask_t *newmems) | |
1056 | { | |
c0ff7453 | 1057 | task_lock(tsk); |
c0ff7453 | 1058 | |
5f155f27 VB |
1059 | local_irq_disable(); |
1060 | write_seqcount_begin(&tsk->mems_allowed_seq); | |
c0ff7453 | 1061 | |
cc9a6c87 | 1062 | nodes_or(tsk->mems_allowed, tsk->mems_allowed, *newmems); |
213980c0 | 1063 | mpol_rebind_task(tsk, newmems); |
58568d2a | 1064 | tsk->mems_allowed = *newmems; |
cc9a6c87 | 1065 | |
5f155f27 VB |
1066 | write_seqcount_end(&tsk->mems_allowed_seq); |
1067 | local_irq_enable(); | |
cc9a6c87 | 1068 | |
c0ff7453 | 1069 | task_unlock(tsk); |
58568d2a MX |
1070 | } |
1071 | ||
8793d854 PM |
1072 | static void *cpuset_being_rebound; |
1073 | ||
0b2f630a MX |
1074 | /** |
1075 | * update_tasks_nodemask - Update the nodemasks of tasks in the cpuset. | |
1076 | * @cs: the cpuset in which each task's mems_allowed mask needs to be changed | |
0b2f630a | 1077 | * |
d66393e5 TH |
1078 | * Iterate through each task of @cs updating its mems_allowed to the |
1079 | * effective cpuset's. As this function is called with cpuset_mutex held, | |
1080 | * cpuset membership stays stable. | |
0b2f630a | 1081 | */ |
d66393e5 | 1082 | static void update_tasks_nodemask(struct cpuset *cs) |
1da177e4 | 1083 | { |
33ad801d | 1084 | static nodemask_t newmems; /* protected by cpuset_mutex */ |
d66393e5 TH |
1085 | struct css_task_iter it; |
1086 | struct task_struct *task; | |
59dac16f | 1087 | |
846a16bf | 1088 | cpuset_being_rebound = cs; /* causes mpol_dup() rebind */ |
4225399a | 1089 | |
ae1c8023 | 1090 | guarantee_online_mems(cs, &newmems); |
33ad801d | 1091 | |
4225399a | 1092 | /* |
3b6766fe LZ |
1093 | * The mpol_rebind_mm() call takes mmap_sem, which we couldn't |
1094 | * take while holding tasklist_lock. Forks can happen - the | |
1095 | * mpol_dup() cpuset_being_rebound check will catch such forks, | |
1096 | * and rebind their vma mempolicies too. Because we still hold | |
5d21cc2d | 1097 | * the global cpuset_mutex, we know that no other rebind effort |
3b6766fe | 1098 | * will be contending for the global variable cpuset_being_rebound. |
4225399a | 1099 | * It's ok if we rebind the same mm twice; mpol_rebind_mm() |
04c19fa6 | 1100 | * is idempotent. Also migrate pages in each mm to new nodes. |
4225399a | 1101 | */ |
d66393e5 TH |
1102 | css_task_iter_start(&cs->css, &it); |
1103 | while ((task = css_task_iter_next(&it))) { | |
1104 | struct mm_struct *mm; | |
1105 | bool migrate; | |
1106 | ||
1107 | cpuset_change_task_nodemask(task, &newmems); | |
1108 | ||
1109 | mm = get_task_mm(task); | |
1110 | if (!mm) | |
1111 | continue; | |
1112 | ||
1113 | migrate = is_memory_migrate(cs); | |
1114 | ||
1115 | mpol_rebind_mm(mm, &cs->mems_allowed); | |
1116 | if (migrate) | |
1117 | cpuset_migrate_mm(mm, &cs->old_mems_allowed, &newmems); | |
e93ad19d TH |
1118 | else |
1119 | mmput(mm); | |
d66393e5 TH |
1120 | } |
1121 | css_task_iter_end(&it); | |
4225399a | 1122 | |
33ad801d LZ |
1123 | /* |
1124 | * All the tasks' nodemasks have been updated, update | |
1125 | * cs->old_mems_allowed. | |
1126 | */ | |
1127 | cs->old_mems_allowed = newmems; | |
1128 | ||
2df167a3 | 1129 | /* We're done rebinding vmas to this cpuset's new mems_allowed. */ |
8793d854 | 1130 | cpuset_being_rebound = NULL; |
1da177e4 LT |
1131 | } |
1132 | ||
5c5cc623 | 1133 | /* |
734d4513 LZ |
1134 | * update_nodemasks_hier - Update effective nodemasks and tasks in the subtree |
1135 | * @cs: the cpuset to consider | |
1136 | * @new_mems: a temp variable for calculating new effective_mems | |
5c5cc623 | 1137 | * |
734d4513 LZ |
1138 | * When configured nodemask is changed, the effective nodemasks of this cpuset |
1139 | * and all its descendants need to be updated. | |
5c5cc623 | 1140 | * |
734d4513 | 1141 | * On legacy hiearchy, effective_mems will be the same with mems_allowed. |
5c5cc623 LZ |
1142 | * |
1143 | * Called with cpuset_mutex held | |
1144 | */ | |
734d4513 | 1145 | static void update_nodemasks_hier(struct cpuset *cs, nodemask_t *new_mems) |
5c5cc623 LZ |
1146 | { |
1147 | struct cpuset *cp; | |
492eb21b | 1148 | struct cgroup_subsys_state *pos_css; |
5c5cc623 LZ |
1149 | |
1150 | rcu_read_lock(); | |
734d4513 LZ |
1151 | cpuset_for_each_descendant_pre(cp, pos_css, cs) { |
1152 | struct cpuset *parent = parent_cs(cp); | |
1153 | ||
1154 | nodes_and(*new_mems, cp->mems_allowed, parent->effective_mems); | |
1155 | ||
554b0d1c LZ |
1156 | /* |
1157 | * If it becomes empty, inherit the effective mask of the | |
1158 | * parent, which is guaranteed to have some MEMs. | |
1159 | */ | |
9e10a130 TH |
1160 | if (cgroup_subsys_on_dfl(cpuset_cgrp_subsys) && |
1161 | nodes_empty(*new_mems)) | |
554b0d1c LZ |
1162 | *new_mems = parent->effective_mems; |
1163 | ||
734d4513 LZ |
1164 | /* Skip the whole subtree if the nodemask remains the same. */ |
1165 | if (nodes_equal(*new_mems, cp->effective_mems)) { | |
1166 | pos_css = css_rightmost_descendant(pos_css); | |
1167 | continue; | |
5c5cc623 | 1168 | } |
734d4513 | 1169 | |
ec903c0c | 1170 | if (!css_tryget_online(&cp->css)) |
5c5cc623 LZ |
1171 | continue; |
1172 | rcu_read_unlock(); | |
1173 | ||
8447a0fe | 1174 | spin_lock_irq(&callback_lock); |
734d4513 | 1175 | cp->effective_mems = *new_mems; |
8447a0fe | 1176 | spin_unlock_irq(&callback_lock); |
734d4513 | 1177 | |
9e10a130 | 1178 | WARN_ON(!cgroup_subsys_on_dfl(cpuset_cgrp_subsys) && |
a1381268 | 1179 | !nodes_equal(cp->mems_allowed, cp->effective_mems)); |
734d4513 | 1180 | |
d66393e5 | 1181 | update_tasks_nodemask(cp); |
5c5cc623 LZ |
1182 | |
1183 | rcu_read_lock(); | |
1184 | css_put(&cp->css); | |
1185 | } | |
1186 | rcu_read_unlock(); | |
1187 | } | |
1188 | ||
0b2f630a MX |
1189 | /* |
1190 | * Handle user request to change the 'mems' memory placement | |
1191 | * of a cpuset. Needs to validate the request, update the | |
58568d2a MX |
1192 | * cpusets mems_allowed, and for each task in the cpuset, |
1193 | * update mems_allowed and rebind task's mempolicy and any vma | |
1194 | * mempolicies and if the cpuset is marked 'memory_migrate', | |
1195 | * migrate the tasks pages to the new memory. | |
0b2f630a | 1196 | * |
8447a0fe | 1197 | * Call with cpuset_mutex held. May take callback_lock during call. |
0b2f630a MX |
1198 | * Will take tasklist_lock, scan tasklist for tasks in cpuset cs, |
1199 | * lock each such tasks mm->mmap_sem, scan its vma's and rebind | |
1200 | * their mempolicies to the cpusets new mems_allowed. | |
1201 | */ | |
645fcc9d LZ |
1202 | static int update_nodemask(struct cpuset *cs, struct cpuset *trialcs, |
1203 | const char *buf) | |
0b2f630a | 1204 | { |
0b2f630a MX |
1205 | int retval; |
1206 | ||
1207 | /* | |
38d7bee9 | 1208 | * top_cpuset.mems_allowed tracks node_stats[N_MEMORY]; |
0b2f630a MX |
1209 | * it's read-only |
1210 | */ | |
53feb297 MX |
1211 | if (cs == &top_cpuset) { |
1212 | retval = -EACCES; | |
1213 | goto done; | |
1214 | } | |
0b2f630a | 1215 | |
0b2f630a MX |
1216 | /* |
1217 | * An empty mems_allowed is ok iff there are no tasks in the cpuset. | |
1218 | * Since nodelist_parse() fails on an empty mask, we special case | |
1219 | * that parsing. The validate_change() call ensures that cpusets | |
1220 | * with tasks have memory. | |
1221 | */ | |
1222 | if (!*buf) { | |
645fcc9d | 1223 | nodes_clear(trialcs->mems_allowed); |
0b2f630a | 1224 | } else { |
645fcc9d | 1225 | retval = nodelist_parse(buf, trialcs->mems_allowed); |
0b2f630a MX |
1226 | if (retval < 0) |
1227 | goto done; | |
1228 | ||
645fcc9d | 1229 | if (!nodes_subset(trialcs->mems_allowed, |
5d8ba82c LZ |
1230 | top_cpuset.mems_allowed)) { |
1231 | retval = -EINVAL; | |
53feb297 MX |
1232 | goto done; |
1233 | } | |
0b2f630a | 1234 | } |
33ad801d LZ |
1235 | |
1236 | if (nodes_equal(cs->mems_allowed, trialcs->mems_allowed)) { | |
0b2f630a MX |
1237 | retval = 0; /* Too easy - nothing to do */ |
1238 | goto done; | |
1239 | } | |
645fcc9d | 1240 | retval = validate_change(cs, trialcs); |
0b2f630a MX |
1241 | if (retval < 0) |
1242 | goto done; | |
1243 | ||
8447a0fe | 1244 | spin_lock_irq(&callback_lock); |
645fcc9d | 1245 | cs->mems_allowed = trialcs->mems_allowed; |
8447a0fe | 1246 | spin_unlock_irq(&callback_lock); |
0b2f630a | 1247 | |
734d4513 | 1248 | /* use trialcs->mems_allowed as a temp variable */ |
24ee3cf8 | 1249 | update_nodemasks_hier(cs, &trialcs->mems_allowed); |
0b2f630a MX |
1250 | done: |
1251 | return retval; | |
1252 | } | |
1253 | ||
8793d854 PM |
1254 | int current_cpuset_is_being_rebound(void) |
1255 | { | |
391acf97 GZ |
1256 | int ret; |
1257 | ||
1258 | rcu_read_lock(); | |
1259 | ret = task_cs(current) == cpuset_being_rebound; | |
1260 | rcu_read_unlock(); | |
1261 | ||
1262 | return ret; | |
8793d854 PM |
1263 | } |
1264 | ||
5be7a479 | 1265 | static int update_relax_domain_level(struct cpuset *cs, s64 val) |
1d3504fc | 1266 | { |
db7f47cf | 1267 | #ifdef CONFIG_SMP |
60495e77 | 1268 | if (val < -1 || val >= sched_domain_level_max) |
30e0e178 | 1269 | return -EINVAL; |
db7f47cf | 1270 | #endif |
1d3504fc HS |
1271 | |
1272 | if (val != cs->relax_domain_level) { | |
1273 | cs->relax_domain_level = val; | |
300ed6cb LZ |
1274 | if (!cpumask_empty(cs->cpus_allowed) && |
1275 | is_sched_load_balance(cs)) | |
699140ba | 1276 | rebuild_sched_domains_locked(); |
1d3504fc HS |
1277 | } |
1278 | ||
1279 | return 0; | |
1280 | } | |
1281 | ||
72ec7029 | 1282 | /** |
950592f7 MX |
1283 | * update_tasks_flags - update the spread flags of tasks in the cpuset. |
1284 | * @cs: the cpuset in which each task's spread flags needs to be changed | |
950592f7 | 1285 | * |
d66393e5 TH |
1286 | * Iterate through each task of @cs updating its spread flags. As this |
1287 | * function is called with cpuset_mutex held, cpuset membership stays | |
1288 | * stable. | |
950592f7 | 1289 | */ |
d66393e5 | 1290 | static void update_tasks_flags(struct cpuset *cs) |
950592f7 | 1291 | { |
d66393e5 TH |
1292 | struct css_task_iter it; |
1293 | struct task_struct *task; | |
1294 | ||
1295 | css_task_iter_start(&cs->css, &it); | |
1296 | while ((task = css_task_iter_next(&it))) | |
1297 | cpuset_update_task_spread_flag(cs, task); | |
1298 | css_task_iter_end(&it); | |
950592f7 MX |
1299 | } |
1300 | ||
1da177e4 LT |
1301 | /* |
1302 | * update_flag - read a 0 or a 1 in a file and update associated flag | |
78608366 PM |
1303 | * bit: the bit to update (see cpuset_flagbits_t) |
1304 | * cs: the cpuset to update | |
1305 | * turning_on: whether the flag is being set or cleared | |
053199ed | 1306 | * |
5d21cc2d | 1307 | * Call with cpuset_mutex held. |
1da177e4 LT |
1308 | */ |
1309 | ||
700fe1ab PM |
1310 | static int update_flag(cpuset_flagbits_t bit, struct cpuset *cs, |
1311 | int turning_on) | |
1da177e4 | 1312 | { |
645fcc9d | 1313 | struct cpuset *trialcs; |
40b6a762 | 1314 | int balance_flag_changed; |
950592f7 | 1315 | int spread_flag_changed; |
950592f7 | 1316 | int err; |
1da177e4 | 1317 | |
645fcc9d LZ |
1318 | trialcs = alloc_trial_cpuset(cs); |
1319 | if (!trialcs) | |
1320 | return -ENOMEM; | |
1321 | ||
1da177e4 | 1322 | if (turning_on) |
645fcc9d | 1323 | set_bit(bit, &trialcs->flags); |
1da177e4 | 1324 | else |
645fcc9d | 1325 | clear_bit(bit, &trialcs->flags); |
1da177e4 | 1326 | |
645fcc9d | 1327 | err = validate_change(cs, trialcs); |
85d7b949 | 1328 | if (err < 0) |
645fcc9d | 1329 | goto out; |
029190c5 | 1330 | |
029190c5 | 1331 | balance_flag_changed = (is_sched_load_balance(cs) != |
645fcc9d | 1332 | is_sched_load_balance(trialcs)); |
029190c5 | 1333 | |
950592f7 MX |
1334 | spread_flag_changed = ((is_spread_slab(cs) != is_spread_slab(trialcs)) |
1335 | || (is_spread_page(cs) != is_spread_page(trialcs))); | |
1336 | ||
8447a0fe | 1337 | spin_lock_irq(&callback_lock); |
645fcc9d | 1338 | cs->flags = trialcs->flags; |
8447a0fe | 1339 | spin_unlock_irq(&callback_lock); |
85d7b949 | 1340 | |
300ed6cb | 1341 | if (!cpumask_empty(trialcs->cpus_allowed) && balance_flag_changed) |
699140ba | 1342 | rebuild_sched_domains_locked(); |
029190c5 | 1343 | |
950592f7 | 1344 | if (spread_flag_changed) |
d66393e5 | 1345 | update_tasks_flags(cs); |
645fcc9d LZ |
1346 | out: |
1347 | free_trial_cpuset(trialcs); | |
1348 | return err; | |
1da177e4 LT |
1349 | } |
1350 | ||
3e0d98b9 | 1351 | /* |
80f7228b | 1352 | * Frequency meter - How fast is some event occurring? |
3e0d98b9 PJ |
1353 | * |
1354 | * These routines manage a digitally filtered, constant time based, | |
1355 | * event frequency meter. There are four routines: | |
1356 | * fmeter_init() - initialize a frequency meter. | |
1357 | * fmeter_markevent() - called each time the event happens. | |
1358 | * fmeter_getrate() - returns the recent rate of such events. | |
1359 | * fmeter_update() - internal routine used to update fmeter. | |
1360 | * | |
1361 | * A common data structure is passed to each of these routines, | |
1362 | * which is used to keep track of the state required to manage the | |
1363 | * frequency meter and its digital filter. | |
1364 | * | |
1365 | * The filter works on the number of events marked per unit time. | |
1366 | * The filter is single-pole low-pass recursive (IIR). The time unit | |
1367 | * is 1 second. Arithmetic is done using 32-bit integers scaled to | |
1368 | * simulate 3 decimal digits of precision (multiplied by 1000). | |
1369 | * | |
1370 | * With an FM_COEF of 933, and a time base of 1 second, the filter | |
1371 | * has a half-life of 10 seconds, meaning that if the events quit | |
1372 | * happening, then the rate returned from the fmeter_getrate() | |
1373 | * will be cut in half each 10 seconds, until it converges to zero. | |
1374 | * | |
1375 | * It is not worth doing a real infinitely recursive filter. If more | |
1376 | * than FM_MAXTICKS ticks have elapsed since the last filter event, | |
1377 | * just compute FM_MAXTICKS ticks worth, by which point the level | |
1378 | * will be stable. | |
1379 | * | |
1380 | * Limit the count of unprocessed events to FM_MAXCNT, so as to avoid | |
1381 | * arithmetic overflow in the fmeter_update() routine. | |
1382 | * | |
1383 | * Given the simple 32 bit integer arithmetic used, this meter works | |
1384 | * best for reporting rates between one per millisecond (msec) and | |
1385 | * one per 32 (approx) seconds. At constant rates faster than one | |
1386 | * per msec it maxes out at values just under 1,000,000. At constant | |
1387 | * rates between one per msec, and one per second it will stabilize | |
1388 | * to a value N*1000, where N is the rate of events per second. | |
1389 | * At constant rates between one per second and one per 32 seconds, | |
1390 | * it will be choppy, moving up on the seconds that have an event, | |
1391 | * and then decaying until the next event. At rates slower than | |
1392 | * about one in 32 seconds, it decays all the way back to zero between | |
1393 | * each event. | |
1394 | */ | |
1395 | ||
1396 | #define FM_COEF 933 /* coefficient for half-life of 10 secs */ | |
d2b43658 | 1397 | #define FM_MAXTICKS ((u32)99) /* useless computing more ticks than this */ |
3e0d98b9 PJ |
1398 | #define FM_MAXCNT 1000000 /* limit cnt to avoid overflow */ |
1399 | #define FM_SCALE 1000 /* faux fixed point scale */ | |
1400 | ||
1401 | /* Initialize a frequency meter */ | |
1402 | static void fmeter_init(struct fmeter *fmp) | |
1403 | { | |
1404 | fmp->cnt = 0; | |
1405 | fmp->val = 0; | |
1406 | fmp->time = 0; | |
1407 | spin_lock_init(&fmp->lock); | |
1408 | } | |
1409 | ||
1410 | /* Internal meter update - process cnt events and update value */ | |
1411 | static void fmeter_update(struct fmeter *fmp) | |
1412 | { | |
d2b43658 AB |
1413 | time64_t now; |
1414 | u32 ticks; | |
1415 | ||
1416 | now = ktime_get_seconds(); | |
1417 | ticks = now - fmp->time; | |
3e0d98b9 PJ |
1418 | |
1419 | if (ticks == 0) | |
1420 | return; | |
1421 | ||
1422 | ticks = min(FM_MAXTICKS, ticks); | |
1423 | while (ticks-- > 0) | |
1424 | fmp->val = (FM_COEF * fmp->val) / FM_SCALE; | |
1425 | fmp->time = now; | |
1426 | ||
1427 | fmp->val += ((FM_SCALE - FM_COEF) * fmp->cnt) / FM_SCALE; | |
1428 | fmp->cnt = 0; | |
1429 | } | |
1430 | ||
1431 | /* Process any previous ticks, then bump cnt by one (times scale). */ | |
1432 | static void fmeter_markevent(struct fmeter *fmp) | |
1433 | { | |
1434 | spin_lock(&fmp->lock); | |
1435 | fmeter_update(fmp); | |
1436 | fmp->cnt = min(FM_MAXCNT, fmp->cnt + FM_SCALE); | |
1437 | spin_unlock(&fmp->lock); | |
1438 | } | |
1439 | ||
1440 | /* Process any previous ticks, then return current value. */ | |
1441 | static int fmeter_getrate(struct fmeter *fmp) | |
1442 | { | |
1443 | int val; | |
1444 | ||
1445 | spin_lock(&fmp->lock); | |
1446 | fmeter_update(fmp); | |
1447 | val = fmp->val; | |
1448 | spin_unlock(&fmp->lock); | |
1449 | return val; | |
1450 | } | |
1451 | ||
57fce0a6 TH |
1452 | static struct cpuset *cpuset_attach_old_cs; |
1453 | ||
5d21cc2d | 1454 | /* Called by cgroups to determine if a cpuset is usable; cpuset_mutex held */ |
1f7dd3e5 | 1455 | static int cpuset_can_attach(struct cgroup_taskset *tset) |
f780bdb7 | 1456 | { |
1f7dd3e5 TH |
1457 | struct cgroup_subsys_state *css; |
1458 | struct cpuset *cs; | |
bb9d97b6 TH |
1459 | struct task_struct *task; |
1460 | int ret; | |
1da177e4 | 1461 | |
57fce0a6 | 1462 | /* used later by cpuset_attach() */ |
1f7dd3e5 TH |
1463 | cpuset_attach_old_cs = task_cs(cgroup_taskset_first(tset, &css)); |
1464 | cs = css_cs(css); | |
57fce0a6 | 1465 | |
5d21cc2d TH |
1466 | mutex_lock(&cpuset_mutex); |
1467 | ||
aa6ec29b | 1468 | /* allow moving tasks into an empty cpuset if on default hierarchy */ |
5d21cc2d | 1469 | ret = -ENOSPC; |
9e10a130 | 1470 | if (!cgroup_subsys_on_dfl(cpuset_cgrp_subsys) && |
88fa523b | 1471 | (cpumask_empty(cs->cpus_allowed) || nodes_empty(cs->mems_allowed))) |
5d21cc2d | 1472 | goto out_unlock; |
9985b0ba | 1473 | |
1f7dd3e5 | 1474 | cgroup_taskset_for_each(task, css, tset) { |
7f51412a JL |
1475 | ret = task_can_attach(task, cs->cpus_allowed); |
1476 | if (ret) | |
5d21cc2d TH |
1477 | goto out_unlock; |
1478 | ret = security_task_setscheduler(task); | |
1479 | if (ret) | |
1480 | goto out_unlock; | |
bb9d97b6 | 1481 | } |
f780bdb7 | 1482 | |
452477fa TH |
1483 | /* |
1484 | * Mark attach is in progress. This makes validate_change() fail | |
1485 | * changes which zero cpus/mems_allowed. | |
1486 | */ | |
1487 | cs->attach_in_progress++; | |
5d21cc2d TH |
1488 | ret = 0; |
1489 | out_unlock: | |
1490 | mutex_unlock(&cpuset_mutex); | |
1491 | return ret; | |
8793d854 | 1492 | } |
f780bdb7 | 1493 | |
1f7dd3e5 | 1494 | static void cpuset_cancel_attach(struct cgroup_taskset *tset) |
452477fa | 1495 | { |
1f7dd3e5 TH |
1496 | struct cgroup_subsys_state *css; |
1497 | struct cpuset *cs; | |
1498 | ||
1499 | cgroup_taskset_first(tset, &css); | |
1500 | cs = css_cs(css); | |
1501 | ||
5d21cc2d | 1502 | mutex_lock(&cpuset_mutex); |
eb95419b | 1503 | css_cs(css)->attach_in_progress--; |
5d21cc2d | 1504 | mutex_unlock(&cpuset_mutex); |
8793d854 | 1505 | } |
1da177e4 | 1506 | |
4e4c9a14 | 1507 | /* |
5d21cc2d | 1508 | * Protected by cpuset_mutex. cpus_attach is used only by cpuset_attach() |
4e4c9a14 TH |
1509 | * but we can't allocate it dynamically there. Define it global and |
1510 | * allocate from cpuset_init(). | |
1511 | */ | |
1512 | static cpumask_var_t cpus_attach; | |
1513 | ||
1f7dd3e5 | 1514 | static void cpuset_attach(struct cgroup_taskset *tset) |
8793d854 | 1515 | { |
67bd2c59 | 1516 | /* static buf protected by cpuset_mutex */ |
4e4c9a14 | 1517 | static nodemask_t cpuset_attach_nodemask_to; |
bb9d97b6 | 1518 | struct task_struct *task; |
4530eddb | 1519 | struct task_struct *leader; |
1f7dd3e5 TH |
1520 | struct cgroup_subsys_state *css; |
1521 | struct cpuset *cs; | |
57fce0a6 | 1522 | struct cpuset *oldcs = cpuset_attach_old_cs; |
22fb52dd | 1523 | |
1f7dd3e5 TH |
1524 | cgroup_taskset_first(tset, &css); |
1525 | cs = css_cs(css); | |
1526 | ||
5d21cc2d TH |
1527 | mutex_lock(&cpuset_mutex); |
1528 | ||
4e4c9a14 TH |
1529 | /* prepare for attach */ |
1530 | if (cs == &top_cpuset) | |
1531 | cpumask_copy(cpus_attach, cpu_possible_mask); | |
1532 | else | |
ae1c8023 | 1533 | guarantee_online_cpus(cs, cpus_attach); |
4e4c9a14 | 1534 | |
ae1c8023 | 1535 | guarantee_online_mems(cs, &cpuset_attach_nodemask_to); |
4e4c9a14 | 1536 | |
1f7dd3e5 | 1537 | cgroup_taskset_for_each(task, css, tset) { |
bb9d97b6 TH |
1538 | /* |
1539 | * can_attach beforehand should guarantee that this doesn't | |
1540 | * fail. TODO: have a better way to handle failure here | |
1541 | */ | |
1542 | WARN_ON_ONCE(set_cpus_allowed_ptr(task, cpus_attach)); | |
1543 | ||
1544 | cpuset_change_task_nodemask(task, &cpuset_attach_nodemask_to); | |
1545 | cpuset_update_task_spread_flag(cs, task); | |
1546 | } | |
22fb52dd | 1547 | |
f780bdb7 | 1548 | /* |
4530eddb TH |
1549 | * Change mm for all threadgroup leaders. This is expensive and may |
1550 | * sleep and should be moved outside migration path proper. | |
f780bdb7 | 1551 | */ |
ae1c8023 | 1552 | cpuset_attach_nodemask_to = cs->effective_mems; |
1f7dd3e5 | 1553 | cgroup_taskset_for_each_leader(leader, css, tset) { |
3df9ca0a TH |
1554 | struct mm_struct *mm = get_task_mm(leader); |
1555 | ||
1556 | if (mm) { | |
1557 | mpol_rebind_mm(mm, &cpuset_attach_nodemask_to); | |
1558 | ||
1559 | /* | |
1560 | * old_mems_allowed is the same with mems_allowed | |
1561 | * here, except if this task is being moved | |
1562 | * automatically due to hotplug. In that case | |
1563 | * @mems_allowed has been updated and is empty, so | |
1564 | * @old_mems_allowed is the right nodesets that we | |
1565 | * migrate mm from. | |
1566 | */ | |
e93ad19d | 1567 | if (is_memory_migrate(cs)) |
3df9ca0a TH |
1568 | cpuset_migrate_mm(mm, &oldcs->old_mems_allowed, |
1569 | &cpuset_attach_nodemask_to); | |
e93ad19d TH |
1570 | else |
1571 | mmput(mm); | |
f047cecf | 1572 | } |
4225399a | 1573 | } |
452477fa | 1574 | |
33ad801d | 1575 | cs->old_mems_allowed = cpuset_attach_nodemask_to; |
02bb5863 | 1576 | |
452477fa | 1577 | cs->attach_in_progress--; |
e44193d3 LZ |
1578 | if (!cs->attach_in_progress) |
1579 | wake_up(&cpuset_attach_wq); | |
5d21cc2d TH |
1580 | |
1581 | mutex_unlock(&cpuset_mutex); | |
1da177e4 LT |
1582 | } |
1583 | ||
1584 | /* The various types of files and directories in a cpuset file system */ | |
1585 | ||
1586 | typedef enum { | |
45b07ef3 | 1587 | FILE_MEMORY_MIGRATE, |
1da177e4 LT |
1588 | FILE_CPULIST, |
1589 | FILE_MEMLIST, | |
afd1a8b3 LZ |
1590 | FILE_EFFECTIVE_CPULIST, |
1591 | FILE_EFFECTIVE_MEMLIST, | |
1da177e4 LT |
1592 | FILE_CPU_EXCLUSIVE, |
1593 | FILE_MEM_EXCLUSIVE, | |
78608366 | 1594 | FILE_MEM_HARDWALL, |
029190c5 | 1595 | FILE_SCHED_LOAD_BALANCE, |
1d3504fc | 1596 | FILE_SCHED_RELAX_DOMAIN_LEVEL, |
3e0d98b9 PJ |
1597 | FILE_MEMORY_PRESSURE_ENABLED, |
1598 | FILE_MEMORY_PRESSURE, | |
825a46af PJ |
1599 | FILE_SPREAD_PAGE, |
1600 | FILE_SPREAD_SLAB, | |
1da177e4 LT |
1601 | } cpuset_filetype_t; |
1602 | ||
182446d0 TH |
1603 | static int cpuset_write_u64(struct cgroup_subsys_state *css, struct cftype *cft, |
1604 | u64 val) | |
700fe1ab | 1605 | { |
182446d0 | 1606 | struct cpuset *cs = css_cs(css); |
700fe1ab | 1607 | cpuset_filetype_t type = cft->private; |
a903f086 | 1608 | int retval = 0; |
700fe1ab | 1609 | |
5d21cc2d | 1610 | mutex_lock(&cpuset_mutex); |
a903f086 LZ |
1611 | if (!is_cpuset_online(cs)) { |
1612 | retval = -ENODEV; | |
5d21cc2d | 1613 | goto out_unlock; |
a903f086 | 1614 | } |
700fe1ab PM |
1615 | |
1616 | switch (type) { | |
1da177e4 | 1617 | case FILE_CPU_EXCLUSIVE: |
700fe1ab | 1618 | retval = update_flag(CS_CPU_EXCLUSIVE, cs, val); |
1da177e4 LT |
1619 | break; |
1620 | case FILE_MEM_EXCLUSIVE: | |
700fe1ab | 1621 | retval = update_flag(CS_MEM_EXCLUSIVE, cs, val); |
1da177e4 | 1622 | break; |
78608366 PM |
1623 | case FILE_MEM_HARDWALL: |
1624 | retval = update_flag(CS_MEM_HARDWALL, cs, val); | |
1625 | break; | |
029190c5 | 1626 | case FILE_SCHED_LOAD_BALANCE: |
700fe1ab | 1627 | retval = update_flag(CS_SCHED_LOAD_BALANCE, cs, val); |
1d3504fc | 1628 | break; |
45b07ef3 | 1629 | case FILE_MEMORY_MIGRATE: |
700fe1ab | 1630 | retval = update_flag(CS_MEMORY_MIGRATE, cs, val); |
45b07ef3 | 1631 | break; |
3e0d98b9 | 1632 | case FILE_MEMORY_PRESSURE_ENABLED: |
700fe1ab | 1633 | cpuset_memory_pressure_enabled = !!val; |
3e0d98b9 | 1634 | break; |
825a46af | 1635 | case FILE_SPREAD_PAGE: |
700fe1ab | 1636 | retval = update_flag(CS_SPREAD_PAGE, cs, val); |
825a46af PJ |
1637 | break; |
1638 | case FILE_SPREAD_SLAB: | |
700fe1ab | 1639 | retval = update_flag(CS_SPREAD_SLAB, cs, val); |
825a46af | 1640 | break; |
1da177e4 LT |
1641 | default: |
1642 | retval = -EINVAL; | |
700fe1ab | 1643 | break; |
1da177e4 | 1644 | } |
5d21cc2d TH |
1645 | out_unlock: |
1646 | mutex_unlock(&cpuset_mutex); | |
1da177e4 LT |
1647 | return retval; |
1648 | } | |
1649 | ||
182446d0 TH |
1650 | static int cpuset_write_s64(struct cgroup_subsys_state *css, struct cftype *cft, |
1651 | s64 val) | |
5be7a479 | 1652 | { |
182446d0 | 1653 | struct cpuset *cs = css_cs(css); |
5be7a479 | 1654 | cpuset_filetype_t type = cft->private; |
5d21cc2d | 1655 | int retval = -ENODEV; |
5be7a479 | 1656 | |
5d21cc2d TH |
1657 | mutex_lock(&cpuset_mutex); |
1658 | if (!is_cpuset_online(cs)) | |
1659 | goto out_unlock; | |
e3712395 | 1660 | |
5be7a479 PM |
1661 | switch (type) { |
1662 | case FILE_SCHED_RELAX_DOMAIN_LEVEL: | |
1663 | retval = update_relax_domain_level(cs, val); | |
1664 | break; | |
1665 | default: | |
1666 | retval = -EINVAL; | |
1667 | break; | |
1668 | } | |
5d21cc2d TH |
1669 | out_unlock: |
1670 | mutex_unlock(&cpuset_mutex); | |
5be7a479 PM |
1671 | return retval; |
1672 | } | |
1673 | ||
e3712395 PM |
1674 | /* |
1675 | * Common handling for a write to a "cpus" or "mems" file. | |
1676 | */ | |
451af504 TH |
1677 | static ssize_t cpuset_write_resmask(struct kernfs_open_file *of, |
1678 | char *buf, size_t nbytes, loff_t off) | |
e3712395 | 1679 | { |
451af504 | 1680 | struct cpuset *cs = css_cs(of_css(of)); |
645fcc9d | 1681 | struct cpuset *trialcs; |
5d21cc2d | 1682 | int retval = -ENODEV; |
e3712395 | 1683 | |
451af504 TH |
1684 | buf = strstrip(buf); |
1685 | ||
3a5a6d0c TH |
1686 | /* |
1687 | * CPU or memory hotunplug may leave @cs w/o any execution | |
1688 | * resources, in which case the hotplug code asynchronously updates | |
1689 | * configuration and transfers all tasks to the nearest ancestor | |
1690 | * which can execute. | |
1691 | * | |
1692 | * As writes to "cpus" or "mems" may restore @cs's execution | |
1693 | * resources, wait for the previously scheduled operations before | |
1694 | * proceeding, so that we don't end up keep removing tasks added | |
1695 | * after execution capability is restored. | |
76bb5ab8 TH |
1696 | * |
1697 | * cpuset_hotplug_work calls back into cgroup core via | |
1698 | * cgroup_transfer_tasks() and waiting for it from a cgroupfs | |
1699 | * operation like this one can lead to a deadlock through kernfs | |
1700 | * active_ref protection. Let's break the protection. Losing the | |
1701 | * protection is okay as we check whether @cs is online after | |
1702 | * grabbing cpuset_mutex anyway. This only happens on the legacy | |
1703 | * hierarchies. | |
3a5a6d0c | 1704 | */ |
76bb5ab8 TH |
1705 | css_get(&cs->css); |
1706 | kernfs_break_active_protection(of->kn); | |
3a5a6d0c TH |
1707 | flush_work(&cpuset_hotplug_work); |
1708 | ||
5d21cc2d TH |
1709 | mutex_lock(&cpuset_mutex); |
1710 | if (!is_cpuset_online(cs)) | |
1711 | goto out_unlock; | |
e3712395 | 1712 | |
645fcc9d | 1713 | trialcs = alloc_trial_cpuset(cs); |
b75f38d6 LZ |
1714 | if (!trialcs) { |
1715 | retval = -ENOMEM; | |
5d21cc2d | 1716 | goto out_unlock; |
b75f38d6 | 1717 | } |
645fcc9d | 1718 | |
451af504 | 1719 | switch (of_cft(of)->private) { |
e3712395 | 1720 | case FILE_CPULIST: |
645fcc9d | 1721 | retval = update_cpumask(cs, trialcs, buf); |
e3712395 PM |
1722 | break; |
1723 | case FILE_MEMLIST: | |
645fcc9d | 1724 | retval = update_nodemask(cs, trialcs, buf); |
e3712395 PM |
1725 | break; |
1726 | default: | |
1727 | retval = -EINVAL; | |
1728 | break; | |
1729 | } | |
645fcc9d LZ |
1730 | |
1731 | free_trial_cpuset(trialcs); | |
5d21cc2d TH |
1732 | out_unlock: |
1733 | mutex_unlock(&cpuset_mutex); | |
76bb5ab8 TH |
1734 | kernfs_unbreak_active_protection(of->kn); |
1735 | css_put(&cs->css); | |
e93ad19d | 1736 | flush_workqueue(cpuset_migrate_mm_wq); |
451af504 | 1737 | return retval ?: nbytes; |
e3712395 PM |
1738 | } |
1739 | ||
1da177e4 LT |
1740 | /* |
1741 | * These ascii lists should be read in a single call, by using a user | |
1742 | * buffer large enough to hold the entire map. If read in smaller | |
1743 | * chunks, there is no guarantee of atomicity. Since the display format | |
1744 | * used, list of ranges of sequential numbers, is variable length, | |
1745 | * and since these maps can change value dynamically, one could read | |
1746 | * gibberish by doing partial reads while a list was changing. | |
1da177e4 | 1747 | */ |
2da8ca82 | 1748 | static int cpuset_common_seq_show(struct seq_file *sf, void *v) |
1da177e4 | 1749 | { |
2da8ca82 TH |
1750 | struct cpuset *cs = css_cs(seq_css(sf)); |
1751 | cpuset_filetype_t type = seq_cft(sf)->private; | |
51ffe411 | 1752 | int ret = 0; |
1da177e4 | 1753 | |
8447a0fe | 1754 | spin_lock_irq(&callback_lock); |
1da177e4 LT |
1755 | |
1756 | switch (type) { | |
1757 | case FILE_CPULIST: | |
e8e6d97c | 1758 | seq_printf(sf, "%*pbl\n", cpumask_pr_args(cs->cpus_allowed)); |
1da177e4 LT |
1759 | break; |
1760 | case FILE_MEMLIST: | |
e8e6d97c | 1761 | seq_printf(sf, "%*pbl\n", nodemask_pr_args(&cs->mems_allowed)); |
1da177e4 | 1762 | break; |
afd1a8b3 | 1763 | case FILE_EFFECTIVE_CPULIST: |
e8e6d97c | 1764 | seq_printf(sf, "%*pbl\n", cpumask_pr_args(cs->effective_cpus)); |
afd1a8b3 LZ |
1765 | break; |
1766 | case FILE_EFFECTIVE_MEMLIST: | |
e8e6d97c | 1767 | seq_printf(sf, "%*pbl\n", nodemask_pr_args(&cs->effective_mems)); |
afd1a8b3 | 1768 | break; |
1da177e4 | 1769 | default: |
51ffe411 | 1770 | ret = -EINVAL; |
1da177e4 | 1771 | } |
1da177e4 | 1772 | |
8447a0fe | 1773 | spin_unlock_irq(&callback_lock); |
51ffe411 | 1774 | return ret; |
1da177e4 LT |
1775 | } |
1776 | ||
182446d0 | 1777 | static u64 cpuset_read_u64(struct cgroup_subsys_state *css, struct cftype *cft) |
700fe1ab | 1778 | { |
182446d0 | 1779 | struct cpuset *cs = css_cs(css); |
700fe1ab PM |
1780 | cpuset_filetype_t type = cft->private; |
1781 | switch (type) { | |
1782 | case FILE_CPU_EXCLUSIVE: | |
1783 | return is_cpu_exclusive(cs); | |
1784 | case FILE_MEM_EXCLUSIVE: | |
1785 | return is_mem_exclusive(cs); | |
78608366 PM |
1786 | case FILE_MEM_HARDWALL: |
1787 | return is_mem_hardwall(cs); | |
700fe1ab PM |
1788 | case FILE_SCHED_LOAD_BALANCE: |
1789 | return is_sched_load_balance(cs); | |
1790 | case FILE_MEMORY_MIGRATE: | |
1791 | return is_memory_migrate(cs); | |
1792 | case FILE_MEMORY_PRESSURE_ENABLED: | |
1793 | return cpuset_memory_pressure_enabled; | |
1794 | case FILE_MEMORY_PRESSURE: | |
1795 | return fmeter_getrate(&cs->fmeter); | |
1796 | case FILE_SPREAD_PAGE: | |
1797 | return is_spread_page(cs); | |
1798 | case FILE_SPREAD_SLAB: | |
1799 | return is_spread_slab(cs); | |
1800 | default: | |
1801 | BUG(); | |
1802 | } | |
cf417141 MK |
1803 | |
1804 | /* Unreachable but makes gcc happy */ | |
1805 | return 0; | |
700fe1ab | 1806 | } |
1da177e4 | 1807 | |
182446d0 | 1808 | static s64 cpuset_read_s64(struct cgroup_subsys_state *css, struct cftype *cft) |
5be7a479 | 1809 | { |
182446d0 | 1810 | struct cpuset *cs = css_cs(css); |
5be7a479 PM |
1811 | cpuset_filetype_t type = cft->private; |
1812 | switch (type) { | |
1813 | case FILE_SCHED_RELAX_DOMAIN_LEVEL: | |
1814 | return cs->relax_domain_level; | |
1815 | default: | |
1816 | BUG(); | |
1817 | } | |
cf417141 MK |
1818 | |
1819 | /* Unrechable but makes gcc happy */ | |
1820 | return 0; | |
5be7a479 PM |
1821 | } |
1822 | ||
1da177e4 LT |
1823 | |
1824 | /* | |
1825 | * for the common functions, 'private' gives the type of file | |
1826 | */ | |
1827 | ||
addf2c73 PM |
1828 | static struct cftype files[] = { |
1829 | { | |
1830 | .name = "cpus", | |
2da8ca82 | 1831 | .seq_show = cpuset_common_seq_show, |
451af504 | 1832 | .write = cpuset_write_resmask, |
e3712395 | 1833 | .max_write_len = (100U + 6 * NR_CPUS), |
addf2c73 PM |
1834 | .private = FILE_CPULIST, |
1835 | }, | |
1836 | ||
1837 | { | |
1838 | .name = "mems", | |
2da8ca82 | 1839 | .seq_show = cpuset_common_seq_show, |
451af504 | 1840 | .write = cpuset_write_resmask, |
e3712395 | 1841 | .max_write_len = (100U + 6 * MAX_NUMNODES), |
addf2c73 PM |
1842 | .private = FILE_MEMLIST, |
1843 | }, | |
1844 | ||
afd1a8b3 LZ |
1845 | { |
1846 | .name = "effective_cpus", | |
1847 | .seq_show = cpuset_common_seq_show, | |
1848 | .private = FILE_EFFECTIVE_CPULIST, | |
1849 | }, | |
1850 | ||
1851 | { | |
1852 | .name = "effective_mems", | |
1853 | .seq_show = cpuset_common_seq_show, | |
1854 | .private = FILE_EFFECTIVE_MEMLIST, | |
1855 | }, | |
1856 | ||
addf2c73 PM |
1857 | { |
1858 | .name = "cpu_exclusive", | |
1859 | .read_u64 = cpuset_read_u64, | |
1860 | .write_u64 = cpuset_write_u64, | |
1861 | .private = FILE_CPU_EXCLUSIVE, | |
1862 | }, | |
1863 | ||
1864 | { | |
1865 | .name = "mem_exclusive", | |
1866 | .read_u64 = cpuset_read_u64, | |
1867 | .write_u64 = cpuset_write_u64, | |
1868 | .private = FILE_MEM_EXCLUSIVE, | |
1869 | }, | |
1870 | ||
78608366 PM |
1871 | { |
1872 | .name = "mem_hardwall", | |
1873 | .read_u64 = cpuset_read_u64, | |
1874 | .write_u64 = cpuset_write_u64, | |
1875 | .private = FILE_MEM_HARDWALL, | |
1876 | }, | |
1877 | ||
addf2c73 PM |
1878 | { |
1879 | .name = "sched_load_balance", | |
1880 | .read_u64 = cpuset_read_u64, | |
1881 | .write_u64 = cpuset_write_u64, | |
1882 | .private = FILE_SCHED_LOAD_BALANCE, | |
1883 | }, | |
1884 | ||
1885 | { | |
1886 | .name = "sched_relax_domain_level", | |
5be7a479 PM |
1887 | .read_s64 = cpuset_read_s64, |
1888 | .write_s64 = cpuset_write_s64, | |
addf2c73 PM |
1889 | .private = FILE_SCHED_RELAX_DOMAIN_LEVEL, |
1890 | }, | |
1891 | ||
1892 | { | |
1893 | .name = "memory_migrate", | |
1894 | .read_u64 = cpuset_read_u64, | |
1895 | .write_u64 = cpuset_write_u64, | |
1896 | .private = FILE_MEMORY_MIGRATE, | |
1897 | }, | |
1898 | ||
1899 | { | |
1900 | .name = "memory_pressure", | |
1901 | .read_u64 = cpuset_read_u64, | |
1c08c22c | 1902 | .private = FILE_MEMORY_PRESSURE, |
addf2c73 PM |
1903 | }, |
1904 | ||
1905 | { | |
1906 | .name = "memory_spread_page", | |
1907 | .read_u64 = cpuset_read_u64, | |
1908 | .write_u64 = cpuset_write_u64, | |
1909 | .private = FILE_SPREAD_PAGE, | |
1910 | }, | |
1911 | ||
1912 | { | |
1913 | .name = "memory_spread_slab", | |
1914 | .read_u64 = cpuset_read_u64, | |
1915 | .write_u64 = cpuset_write_u64, | |
1916 | .private = FILE_SPREAD_SLAB, | |
1917 | }, | |
3e0d98b9 | 1918 | |
4baf6e33 TH |
1919 | { |
1920 | .name = "memory_pressure_enabled", | |
1921 | .flags = CFTYPE_ONLY_ON_ROOT, | |
1922 | .read_u64 = cpuset_read_u64, | |
1923 | .write_u64 = cpuset_write_u64, | |
1924 | .private = FILE_MEMORY_PRESSURE_ENABLED, | |
1925 | }, | |
1da177e4 | 1926 | |
4baf6e33 TH |
1927 | { } /* terminate */ |
1928 | }; | |
1da177e4 LT |
1929 | |
1930 | /* | |
92fb9748 | 1931 | * cpuset_css_alloc - allocate a cpuset css |
c9e5fe66 | 1932 | * cgrp: control group that the new cpuset will be part of |
1da177e4 LT |
1933 | */ |
1934 | ||
eb95419b TH |
1935 | static struct cgroup_subsys_state * |
1936 | cpuset_css_alloc(struct cgroup_subsys_state *parent_css) | |
1da177e4 | 1937 | { |
c8f699bb | 1938 | struct cpuset *cs; |
1da177e4 | 1939 | |
eb95419b | 1940 | if (!parent_css) |
8793d854 | 1941 | return &top_cpuset.css; |
033fa1c5 | 1942 | |
c8f699bb | 1943 | cs = kzalloc(sizeof(*cs), GFP_KERNEL); |
1da177e4 | 1944 | if (!cs) |
8793d854 | 1945 | return ERR_PTR(-ENOMEM); |
e2b9a3d7 LZ |
1946 | if (!alloc_cpumask_var(&cs->cpus_allowed, GFP_KERNEL)) |
1947 | goto free_cs; | |
1948 | if (!alloc_cpumask_var(&cs->effective_cpus, GFP_KERNEL)) | |
1949 | goto free_cpus; | |
1da177e4 | 1950 | |
029190c5 | 1951 | set_bit(CS_SCHED_LOAD_BALANCE, &cs->flags); |
300ed6cb | 1952 | cpumask_clear(cs->cpus_allowed); |
f9a86fcb | 1953 | nodes_clear(cs->mems_allowed); |
e2b9a3d7 LZ |
1954 | cpumask_clear(cs->effective_cpus); |
1955 | nodes_clear(cs->effective_mems); | |
3e0d98b9 | 1956 | fmeter_init(&cs->fmeter); |
1d3504fc | 1957 | cs->relax_domain_level = -1; |
1da177e4 | 1958 | |
c8f699bb | 1959 | return &cs->css; |
e2b9a3d7 LZ |
1960 | |
1961 | free_cpus: | |
1962 | free_cpumask_var(cs->cpus_allowed); | |
1963 | free_cs: | |
1964 | kfree(cs); | |
1965 | return ERR_PTR(-ENOMEM); | |
c8f699bb TH |
1966 | } |
1967 | ||
eb95419b | 1968 | static int cpuset_css_online(struct cgroup_subsys_state *css) |
c8f699bb | 1969 | { |
eb95419b | 1970 | struct cpuset *cs = css_cs(css); |
c431069f | 1971 | struct cpuset *parent = parent_cs(cs); |
ae8086ce | 1972 | struct cpuset *tmp_cs; |
492eb21b | 1973 | struct cgroup_subsys_state *pos_css; |
c8f699bb TH |
1974 | |
1975 | if (!parent) | |
1976 | return 0; | |
1977 | ||
5d21cc2d TH |
1978 | mutex_lock(&cpuset_mutex); |
1979 | ||
efeb77b2 | 1980 | set_bit(CS_ONLINE, &cs->flags); |
c8f699bb TH |
1981 | if (is_spread_page(parent)) |
1982 | set_bit(CS_SPREAD_PAGE, &cs->flags); | |
1983 | if (is_spread_slab(parent)) | |
1984 | set_bit(CS_SPREAD_SLAB, &cs->flags); | |
1da177e4 | 1985 | |
664eedde | 1986 | cpuset_inc(); |
033fa1c5 | 1987 | |
8447a0fe | 1988 | spin_lock_irq(&callback_lock); |
9e10a130 | 1989 | if (cgroup_subsys_on_dfl(cpuset_cgrp_subsys)) { |
e2b9a3d7 LZ |
1990 | cpumask_copy(cs->effective_cpus, parent->effective_cpus); |
1991 | cs->effective_mems = parent->effective_mems; | |
1992 | } | |
8447a0fe | 1993 | spin_unlock_irq(&callback_lock); |
e2b9a3d7 | 1994 | |
eb95419b | 1995 | if (!test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags)) |
5d21cc2d | 1996 | goto out_unlock; |
033fa1c5 TH |
1997 | |
1998 | /* | |
1999 | * Clone @parent's configuration if CGRP_CPUSET_CLONE_CHILDREN is | |
2000 | * set. This flag handling is implemented in cgroup core for | |
2001 | * histrical reasons - the flag may be specified during mount. | |
2002 | * | |
2003 | * Currently, if any sibling cpusets have exclusive cpus or mem, we | |
2004 | * refuse to clone the configuration - thereby refusing the task to | |
2005 | * be entered, and as a result refusing the sys_unshare() or | |
2006 | * clone() which initiated it. If this becomes a problem for some | |
2007 | * users who wish to allow that scenario, then this could be | |
2008 | * changed to grant parent->cpus_allowed-sibling_cpus_exclusive | |
2009 | * (and likewise for mems) to the new cgroup. | |
2010 | */ | |
ae8086ce | 2011 | rcu_read_lock(); |
492eb21b | 2012 | cpuset_for_each_child(tmp_cs, pos_css, parent) { |
ae8086ce TH |
2013 | if (is_mem_exclusive(tmp_cs) || is_cpu_exclusive(tmp_cs)) { |
2014 | rcu_read_unlock(); | |
5d21cc2d | 2015 | goto out_unlock; |
ae8086ce | 2016 | } |
033fa1c5 | 2017 | } |
ae8086ce | 2018 | rcu_read_unlock(); |
033fa1c5 | 2019 | |
8447a0fe | 2020 | spin_lock_irq(&callback_lock); |
033fa1c5 | 2021 | cs->mems_allowed = parent->mems_allowed; |
790317e1 | 2022 | cs->effective_mems = parent->mems_allowed; |
033fa1c5 | 2023 | cpumask_copy(cs->cpus_allowed, parent->cpus_allowed); |
790317e1 | 2024 | cpumask_copy(cs->effective_cpus, parent->cpus_allowed); |
cea74465 | 2025 | spin_unlock_irq(&callback_lock); |
5d21cc2d TH |
2026 | out_unlock: |
2027 | mutex_unlock(&cpuset_mutex); | |
c8f699bb TH |
2028 | return 0; |
2029 | } | |
2030 | ||
0b9e6965 ZH |
2031 | /* |
2032 | * If the cpuset being removed has its flag 'sched_load_balance' | |
2033 | * enabled, then simulate turning sched_load_balance off, which | |
2034 | * will call rebuild_sched_domains_locked(). | |
2035 | */ | |
2036 | ||
eb95419b | 2037 | static void cpuset_css_offline(struct cgroup_subsys_state *css) |
c8f699bb | 2038 | { |
eb95419b | 2039 | struct cpuset *cs = css_cs(css); |
c8f699bb | 2040 | |
5d21cc2d | 2041 | mutex_lock(&cpuset_mutex); |
c8f699bb TH |
2042 | |
2043 | if (is_sched_load_balance(cs)) | |
2044 | update_flag(CS_SCHED_LOAD_BALANCE, cs, 0); | |
2045 | ||
664eedde | 2046 | cpuset_dec(); |
efeb77b2 | 2047 | clear_bit(CS_ONLINE, &cs->flags); |
c8f699bb | 2048 | |
5d21cc2d | 2049 | mutex_unlock(&cpuset_mutex); |
1da177e4 LT |
2050 | } |
2051 | ||
eb95419b | 2052 | static void cpuset_css_free(struct cgroup_subsys_state *css) |
1da177e4 | 2053 | { |
eb95419b | 2054 | struct cpuset *cs = css_cs(css); |
1da177e4 | 2055 | |
e2b9a3d7 | 2056 | free_cpumask_var(cs->effective_cpus); |
300ed6cb | 2057 | free_cpumask_var(cs->cpus_allowed); |
8793d854 | 2058 | kfree(cs); |
1da177e4 LT |
2059 | } |
2060 | ||
39bd0d15 LZ |
2061 | static void cpuset_bind(struct cgroup_subsys_state *root_css) |
2062 | { | |
2063 | mutex_lock(&cpuset_mutex); | |
8447a0fe | 2064 | spin_lock_irq(&callback_lock); |
39bd0d15 | 2065 | |
9e10a130 | 2066 | if (cgroup_subsys_on_dfl(cpuset_cgrp_subsys)) { |
39bd0d15 LZ |
2067 | cpumask_copy(top_cpuset.cpus_allowed, cpu_possible_mask); |
2068 | top_cpuset.mems_allowed = node_possible_map; | |
2069 | } else { | |
2070 | cpumask_copy(top_cpuset.cpus_allowed, | |
2071 | top_cpuset.effective_cpus); | |
2072 | top_cpuset.mems_allowed = top_cpuset.effective_mems; | |
2073 | } | |
2074 | ||
8447a0fe | 2075 | spin_unlock_irq(&callback_lock); |
39bd0d15 LZ |
2076 | mutex_unlock(&cpuset_mutex); |
2077 | } | |
2078 | ||
06f4e948 ZL |
2079 | /* |
2080 | * Make sure the new task conform to the current state of its parent, | |
2081 | * which could have been changed by cpuset just after it inherits the | |
2082 | * state from the parent and before it sits on the cgroup's task list. | |
2083 | */ | |
8a15b817 | 2084 | static void cpuset_fork(struct task_struct *task) |
06f4e948 ZL |
2085 | { |
2086 | if (task_css_is_root(task, cpuset_cgrp_id)) | |
2087 | return; | |
2088 | ||
2089 | set_cpus_allowed_ptr(task, ¤t->cpus_allowed); | |
2090 | task->mems_allowed = current->mems_allowed; | |
2091 | } | |
2092 | ||
073219e9 | 2093 | struct cgroup_subsys cpuset_cgrp_subsys = { |
39bd0d15 LZ |
2094 | .css_alloc = cpuset_css_alloc, |
2095 | .css_online = cpuset_css_online, | |
2096 | .css_offline = cpuset_css_offline, | |
2097 | .css_free = cpuset_css_free, | |
2098 | .can_attach = cpuset_can_attach, | |
2099 | .cancel_attach = cpuset_cancel_attach, | |
2100 | .attach = cpuset_attach, | |
5cf1cacb | 2101 | .post_attach = cpuset_post_attach, |
39bd0d15 | 2102 | .bind = cpuset_bind, |
06f4e948 | 2103 | .fork = cpuset_fork, |
5577964e | 2104 | .legacy_cftypes = files, |
b38e42e9 | 2105 | .early_init = true, |
8793d854 PM |
2106 | }; |
2107 | ||
1da177e4 LT |
2108 | /** |
2109 | * cpuset_init - initialize cpusets at system boot | |
2110 | * | |
2111 | * Description: Initialize top_cpuset and the cpuset internal file system, | |
2112 | **/ | |
2113 | ||
2114 | int __init cpuset_init(void) | |
2115 | { | |
8793d854 | 2116 | int err = 0; |
1da177e4 | 2117 | |
75fa8e5d NMG |
2118 | BUG_ON(!alloc_cpumask_var(&top_cpuset.cpus_allowed, GFP_KERNEL)); |
2119 | BUG_ON(!alloc_cpumask_var(&top_cpuset.effective_cpus, GFP_KERNEL)); | |
58568d2a | 2120 | |
300ed6cb | 2121 | cpumask_setall(top_cpuset.cpus_allowed); |
f9a86fcb | 2122 | nodes_setall(top_cpuset.mems_allowed); |
e2b9a3d7 LZ |
2123 | cpumask_setall(top_cpuset.effective_cpus); |
2124 | nodes_setall(top_cpuset.effective_mems); | |
1da177e4 | 2125 | |
3e0d98b9 | 2126 | fmeter_init(&top_cpuset.fmeter); |
029190c5 | 2127 | set_bit(CS_SCHED_LOAD_BALANCE, &top_cpuset.flags); |
1d3504fc | 2128 | top_cpuset.relax_domain_level = -1; |
1da177e4 | 2129 | |
1da177e4 LT |
2130 | err = register_filesystem(&cpuset_fs_type); |
2131 | if (err < 0) | |
8793d854 PM |
2132 | return err; |
2133 | ||
75fa8e5d | 2134 | BUG_ON(!alloc_cpumask_var(&cpus_attach, GFP_KERNEL)); |
2341d1b6 | 2135 | |
8793d854 | 2136 | return 0; |
1da177e4 LT |
2137 | } |
2138 | ||
b1aac8bb | 2139 | /* |
cf417141 | 2140 | * If CPU and/or memory hotplug handlers, below, unplug any CPUs |
b1aac8bb PJ |
2141 | * or memory nodes, we need to walk over the cpuset hierarchy, |
2142 | * removing that CPU or node from all cpusets. If this removes the | |
956db3ca CW |
2143 | * last CPU or node from a cpuset, then move the tasks in the empty |
2144 | * cpuset to its next-highest non-empty parent. | |
b1aac8bb | 2145 | */ |
956db3ca CW |
2146 | static void remove_tasks_in_empty_cpuset(struct cpuset *cs) |
2147 | { | |
2148 | struct cpuset *parent; | |
2149 | ||
956db3ca CW |
2150 | /* |
2151 | * Find its next-highest non-empty parent, (top cpuset | |
2152 | * has online cpus, so can't be empty). | |
2153 | */ | |
c431069f | 2154 | parent = parent_cs(cs); |
300ed6cb | 2155 | while (cpumask_empty(parent->cpus_allowed) || |
b4501295 | 2156 | nodes_empty(parent->mems_allowed)) |
c431069f | 2157 | parent = parent_cs(parent); |
956db3ca | 2158 | |
8cc99345 | 2159 | if (cgroup_transfer_tasks(parent->css.cgroup, cs->css.cgroup)) { |
12d3089c | 2160 | pr_err("cpuset: failed to transfer tasks out of empty cpuset "); |
e61734c5 TH |
2161 | pr_cont_cgroup_name(cs->css.cgroup); |
2162 | pr_cont("\n"); | |
8cc99345 | 2163 | } |
956db3ca CW |
2164 | } |
2165 | ||
be4c9dd7 LZ |
2166 | static void |
2167 | hotplug_update_tasks_legacy(struct cpuset *cs, | |
2168 | struct cpumask *new_cpus, nodemask_t *new_mems, | |
2169 | bool cpus_updated, bool mems_updated) | |
390a36aa LZ |
2170 | { |
2171 | bool is_empty; | |
2172 | ||
8447a0fe | 2173 | spin_lock_irq(&callback_lock); |
be4c9dd7 LZ |
2174 | cpumask_copy(cs->cpus_allowed, new_cpus); |
2175 | cpumask_copy(cs->effective_cpus, new_cpus); | |
2176 | cs->mems_allowed = *new_mems; | |
2177 | cs->effective_mems = *new_mems; | |
8447a0fe | 2178 | spin_unlock_irq(&callback_lock); |
390a36aa LZ |
2179 | |
2180 | /* | |
2181 | * Don't call update_tasks_cpumask() if the cpuset becomes empty, | |
2182 | * as the tasks will be migratecd to an ancestor. | |
2183 | */ | |
be4c9dd7 | 2184 | if (cpus_updated && !cpumask_empty(cs->cpus_allowed)) |
390a36aa | 2185 | update_tasks_cpumask(cs); |
be4c9dd7 | 2186 | if (mems_updated && !nodes_empty(cs->mems_allowed)) |
390a36aa LZ |
2187 | update_tasks_nodemask(cs); |
2188 | ||
2189 | is_empty = cpumask_empty(cs->cpus_allowed) || | |
2190 | nodes_empty(cs->mems_allowed); | |
2191 | ||
2192 | mutex_unlock(&cpuset_mutex); | |
2193 | ||
2194 | /* | |
2195 | * Move tasks to the nearest ancestor with execution resources, | |
2196 | * This is full cgroup operation which will also call back into | |
2197 | * cpuset. Should be done outside any lock. | |
2198 | */ | |
2199 | if (is_empty) | |
2200 | remove_tasks_in_empty_cpuset(cs); | |
2201 | ||
2202 | mutex_lock(&cpuset_mutex); | |
2203 | } | |
2204 | ||
be4c9dd7 LZ |
2205 | static void |
2206 | hotplug_update_tasks(struct cpuset *cs, | |
2207 | struct cpumask *new_cpus, nodemask_t *new_mems, | |
2208 | bool cpus_updated, bool mems_updated) | |
390a36aa | 2209 | { |
be4c9dd7 LZ |
2210 | if (cpumask_empty(new_cpus)) |
2211 | cpumask_copy(new_cpus, parent_cs(cs)->effective_cpus); | |
2212 | if (nodes_empty(*new_mems)) | |
2213 | *new_mems = parent_cs(cs)->effective_mems; | |
2214 | ||
8447a0fe | 2215 | spin_lock_irq(&callback_lock); |
be4c9dd7 LZ |
2216 | cpumask_copy(cs->effective_cpus, new_cpus); |
2217 | cs->effective_mems = *new_mems; | |
8447a0fe | 2218 | spin_unlock_irq(&callback_lock); |
390a36aa | 2219 | |
be4c9dd7 | 2220 | if (cpus_updated) |
390a36aa | 2221 | update_tasks_cpumask(cs); |
be4c9dd7 | 2222 | if (mems_updated) |
390a36aa LZ |
2223 | update_tasks_nodemask(cs); |
2224 | } | |
2225 | ||
deb7aa30 | 2226 | /** |
388afd85 | 2227 | * cpuset_hotplug_update_tasks - update tasks in a cpuset for hotunplug |
deb7aa30 | 2228 | * @cs: cpuset in interest |
956db3ca | 2229 | * |
deb7aa30 TH |
2230 | * Compare @cs's cpu and mem masks against top_cpuset and if some have gone |
2231 | * offline, update @cs accordingly. If @cs ends up with no CPU or memory, | |
2232 | * all its tasks are moved to the nearest ancestor with both resources. | |
80d1fa64 | 2233 | */ |
388afd85 | 2234 | static void cpuset_hotplug_update_tasks(struct cpuset *cs) |
80d1fa64 | 2235 | { |
be4c9dd7 LZ |
2236 | static cpumask_t new_cpus; |
2237 | static nodemask_t new_mems; | |
2238 | bool cpus_updated; | |
2239 | bool mems_updated; | |
e44193d3 LZ |
2240 | retry: |
2241 | wait_event(cpuset_attach_wq, cs->attach_in_progress == 0); | |
80d1fa64 | 2242 | |
5d21cc2d | 2243 | mutex_lock(&cpuset_mutex); |
7ddf96b0 | 2244 | |
e44193d3 LZ |
2245 | /* |
2246 | * We have raced with task attaching. We wait until attaching | |
2247 | * is finished, so we won't attach a task to an empty cpuset. | |
2248 | */ | |
2249 | if (cs->attach_in_progress) { | |
2250 | mutex_unlock(&cpuset_mutex); | |
2251 | goto retry; | |
2252 | } | |
2253 | ||
be4c9dd7 LZ |
2254 | cpumask_and(&new_cpus, cs->cpus_allowed, parent_cs(cs)->effective_cpus); |
2255 | nodes_and(new_mems, cs->mems_allowed, parent_cs(cs)->effective_mems); | |
80d1fa64 | 2256 | |
be4c9dd7 LZ |
2257 | cpus_updated = !cpumask_equal(&new_cpus, cs->effective_cpus); |
2258 | mems_updated = !nodes_equal(new_mems, cs->effective_mems); | |
deb7aa30 | 2259 | |
9e10a130 | 2260 | if (cgroup_subsys_on_dfl(cpuset_cgrp_subsys)) |
be4c9dd7 LZ |
2261 | hotplug_update_tasks(cs, &new_cpus, &new_mems, |
2262 | cpus_updated, mems_updated); | |
390a36aa | 2263 | else |
be4c9dd7 LZ |
2264 | hotplug_update_tasks_legacy(cs, &new_cpus, &new_mems, |
2265 | cpus_updated, mems_updated); | |
8d033948 | 2266 | |
5d21cc2d | 2267 | mutex_unlock(&cpuset_mutex); |
b1aac8bb PJ |
2268 | } |
2269 | ||
deb7aa30 | 2270 | /** |
3a5a6d0c | 2271 | * cpuset_hotplug_workfn - handle CPU/memory hotunplug for a cpuset |
956db3ca | 2272 | * |
deb7aa30 TH |
2273 | * This function is called after either CPU or memory configuration has |
2274 | * changed and updates cpuset accordingly. The top_cpuset is always | |
2275 | * synchronized to cpu_active_mask and N_MEMORY, which is necessary in | |
2276 | * order to make cpusets transparent (of no affect) on systems that are | |
2277 | * actively using CPU hotplug but making no active use of cpusets. | |
956db3ca | 2278 | * |
deb7aa30 | 2279 | * Non-root cpusets are only affected by offlining. If any CPUs or memory |
388afd85 LZ |
2280 | * nodes have been taken down, cpuset_hotplug_update_tasks() is invoked on |
2281 | * all descendants. | |
956db3ca | 2282 | * |
deb7aa30 TH |
2283 | * Note that CPU offlining during suspend is ignored. We don't modify |
2284 | * cpusets across suspend/resume cycles at all. | |
956db3ca | 2285 | */ |
3a5a6d0c | 2286 | static void cpuset_hotplug_workfn(struct work_struct *work) |
b1aac8bb | 2287 | { |
5c5cc623 LZ |
2288 | static cpumask_t new_cpus; |
2289 | static nodemask_t new_mems; | |
deb7aa30 | 2290 | bool cpus_updated, mems_updated; |
9e10a130 | 2291 | bool on_dfl = cgroup_subsys_on_dfl(cpuset_cgrp_subsys); |
b1aac8bb | 2292 | |
5d21cc2d | 2293 | mutex_lock(&cpuset_mutex); |
956db3ca | 2294 | |
deb7aa30 TH |
2295 | /* fetch the available cpus/mems and find out which changed how */ |
2296 | cpumask_copy(&new_cpus, cpu_active_mask); | |
2297 | new_mems = node_states[N_MEMORY]; | |
7ddf96b0 | 2298 | |
7e88291b LZ |
2299 | cpus_updated = !cpumask_equal(top_cpuset.effective_cpus, &new_cpus); |
2300 | mems_updated = !nodes_equal(top_cpuset.effective_mems, new_mems); | |
7ddf96b0 | 2301 | |
deb7aa30 TH |
2302 | /* synchronize cpus_allowed to cpu_active_mask */ |
2303 | if (cpus_updated) { | |
8447a0fe | 2304 | spin_lock_irq(&callback_lock); |
7e88291b LZ |
2305 | if (!on_dfl) |
2306 | cpumask_copy(top_cpuset.cpus_allowed, &new_cpus); | |
1344ab9c | 2307 | cpumask_copy(top_cpuset.effective_cpus, &new_cpus); |
8447a0fe | 2308 | spin_unlock_irq(&callback_lock); |
deb7aa30 TH |
2309 | /* we don't mess with cpumasks of tasks in top_cpuset */ |
2310 | } | |
b4501295 | 2311 | |
deb7aa30 TH |
2312 | /* synchronize mems_allowed to N_MEMORY */ |
2313 | if (mems_updated) { | |
8447a0fe | 2314 | spin_lock_irq(&callback_lock); |
7e88291b LZ |
2315 | if (!on_dfl) |
2316 | top_cpuset.mems_allowed = new_mems; | |
1344ab9c | 2317 | top_cpuset.effective_mems = new_mems; |
8447a0fe | 2318 | spin_unlock_irq(&callback_lock); |
d66393e5 | 2319 | update_tasks_nodemask(&top_cpuset); |
deb7aa30 | 2320 | } |
b4501295 | 2321 | |
388afd85 LZ |
2322 | mutex_unlock(&cpuset_mutex); |
2323 | ||
5c5cc623 LZ |
2324 | /* if cpus or mems changed, we need to propagate to descendants */ |
2325 | if (cpus_updated || mems_updated) { | |
deb7aa30 | 2326 | struct cpuset *cs; |
492eb21b | 2327 | struct cgroup_subsys_state *pos_css; |
f9b4fb8d | 2328 | |
fc560a26 | 2329 | rcu_read_lock(); |
492eb21b | 2330 | cpuset_for_each_descendant_pre(cs, pos_css, &top_cpuset) { |
ec903c0c | 2331 | if (cs == &top_cpuset || !css_tryget_online(&cs->css)) |
388afd85 LZ |
2332 | continue; |
2333 | rcu_read_unlock(); | |
7ddf96b0 | 2334 | |
388afd85 | 2335 | cpuset_hotplug_update_tasks(cs); |
b4501295 | 2336 | |
388afd85 LZ |
2337 | rcu_read_lock(); |
2338 | css_put(&cs->css); | |
2339 | } | |
2340 | rcu_read_unlock(); | |
2341 | } | |
8d033948 | 2342 | |
deb7aa30 | 2343 | /* rebuild sched domains if cpus_allowed has changed */ |
e0e80a02 LZ |
2344 | if (cpus_updated) |
2345 | rebuild_sched_domains(); | |
b1aac8bb PJ |
2346 | } |
2347 | ||
30e03acd | 2348 | void cpuset_update_active_cpus(void) |
4c4d50f7 | 2349 | { |
3a5a6d0c TH |
2350 | /* |
2351 | * We're inside cpu hotplug critical region which usually nests | |
2352 | * inside cgroup synchronization. Bounce actual hotplug processing | |
2353 | * to a work item to avoid reverse locking order. | |
3a5a6d0c | 2354 | */ |
3a5a6d0c | 2355 | schedule_work(&cpuset_hotplug_work); |
4c4d50f7 | 2356 | } |
4c4d50f7 | 2357 | |
38837fc7 | 2358 | /* |
38d7bee9 LJ |
2359 | * Keep top_cpuset.mems_allowed tracking node_states[N_MEMORY]. |
2360 | * Call this routine anytime after node_states[N_MEMORY] changes. | |
a1cd2b13 | 2361 | * See cpuset_update_active_cpus() for CPU hotplug handling. |
38837fc7 | 2362 | */ |
f481891f MX |
2363 | static int cpuset_track_online_nodes(struct notifier_block *self, |
2364 | unsigned long action, void *arg) | |
38837fc7 | 2365 | { |
3a5a6d0c | 2366 | schedule_work(&cpuset_hotplug_work); |
f481891f | 2367 | return NOTIFY_OK; |
38837fc7 | 2368 | } |
d8f10cb3 AM |
2369 | |
2370 | static struct notifier_block cpuset_track_online_nodes_nb = { | |
2371 | .notifier_call = cpuset_track_online_nodes, | |
2372 | .priority = 10, /* ??! */ | |
2373 | }; | |
38837fc7 | 2374 | |
1da177e4 LT |
2375 | /** |
2376 | * cpuset_init_smp - initialize cpus_allowed | |
2377 | * | |
2378 | * Description: Finish top cpuset after cpu, node maps are initialized | |
d8f10cb3 | 2379 | */ |
1da177e4 LT |
2380 | void __init cpuset_init_smp(void) |
2381 | { | |
6ad4c188 | 2382 | cpumask_copy(top_cpuset.cpus_allowed, cpu_active_mask); |
38d7bee9 | 2383 | top_cpuset.mems_allowed = node_states[N_MEMORY]; |
33ad801d | 2384 | top_cpuset.old_mems_allowed = top_cpuset.mems_allowed; |
4c4d50f7 | 2385 | |
e2b9a3d7 LZ |
2386 | cpumask_copy(top_cpuset.effective_cpus, cpu_active_mask); |
2387 | top_cpuset.effective_mems = node_states[N_MEMORY]; | |
2388 | ||
d8f10cb3 | 2389 | register_hotmemory_notifier(&cpuset_track_online_nodes_nb); |
e93ad19d TH |
2390 | |
2391 | cpuset_migrate_mm_wq = alloc_ordered_workqueue("cpuset_migrate_mm", 0); | |
2392 | BUG_ON(!cpuset_migrate_mm_wq); | |
1da177e4 LT |
2393 | } |
2394 | ||
2395 | /** | |
1da177e4 LT |
2396 | * cpuset_cpus_allowed - return cpus_allowed mask from a tasks cpuset. |
2397 | * @tsk: pointer to task_struct from which to obtain cpuset->cpus_allowed. | |
6af866af | 2398 | * @pmask: pointer to struct cpumask variable to receive cpus_allowed set. |
1da177e4 | 2399 | * |
300ed6cb | 2400 | * Description: Returns the cpumask_var_t cpus_allowed of the cpuset |
1da177e4 | 2401 | * attached to the specified @tsk. Guaranteed to return some non-empty |
5f054e31 | 2402 | * subset of cpu_online_mask, even if this means going outside the |
1da177e4 LT |
2403 | * tasks cpuset. |
2404 | **/ | |
2405 | ||
6af866af | 2406 | void cpuset_cpus_allowed(struct task_struct *tsk, struct cpumask *pmask) |
1da177e4 | 2407 | { |
8447a0fe VD |
2408 | unsigned long flags; |
2409 | ||
2410 | spin_lock_irqsave(&callback_lock, flags); | |
b8dadcb5 | 2411 | rcu_read_lock(); |
ae1c8023 | 2412 | guarantee_online_cpus(task_cs(tsk), pmask); |
b8dadcb5 | 2413 | rcu_read_unlock(); |
8447a0fe | 2414 | spin_unlock_irqrestore(&callback_lock, flags); |
1da177e4 LT |
2415 | } |
2416 | ||
2baab4e9 | 2417 | void cpuset_cpus_allowed_fallback(struct task_struct *tsk) |
9084bb82 | 2418 | { |
9084bb82 | 2419 | rcu_read_lock(); |
ae1c8023 | 2420 | do_set_cpus_allowed(tsk, task_cs(tsk)->effective_cpus); |
9084bb82 ON |
2421 | rcu_read_unlock(); |
2422 | ||
2423 | /* | |
2424 | * We own tsk->cpus_allowed, nobody can change it under us. | |
2425 | * | |
2426 | * But we used cs && cs->cpus_allowed lockless and thus can | |
2427 | * race with cgroup_attach_task() or update_cpumask() and get | |
2428 | * the wrong tsk->cpus_allowed. However, both cases imply the | |
2429 | * subsequent cpuset_change_cpumask()->set_cpus_allowed_ptr() | |
2430 | * which takes task_rq_lock(). | |
2431 | * | |
2432 | * If we are called after it dropped the lock we must see all | |
2433 | * changes in tsk_cs()->cpus_allowed. Otherwise we can temporary | |
2434 | * set any mask even if it is not right from task_cs() pov, | |
2435 | * the pending set_cpus_allowed_ptr() will fix things. | |
2baab4e9 PZ |
2436 | * |
2437 | * select_fallback_rq() will fix things ups and set cpu_possible_mask | |
2438 | * if required. | |
9084bb82 | 2439 | */ |
9084bb82 ON |
2440 | } |
2441 | ||
8f4ab07f | 2442 | void __init cpuset_init_current_mems_allowed(void) |
1da177e4 | 2443 | { |
f9a86fcb | 2444 | nodes_setall(current->mems_allowed); |
1da177e4 LT |
2445 | } |
2446 | ||
909d75a3 PJ |
2447 | /** |
2448 | * cpuset_mems_allowed - return mems_allowed mask from a tasks cpuset. | |
2449 | * @tsk: pointer to task_struct from which to obtain cpuset->mems_allowed. | |
2450 | * | |
2451 | * Description: Returns the nodemask_t mems_allowed of the cpuset | |
2452 | * attached to the specified @tsk. Guaranteed to return some non-empty | |
38d7bee9 | 2453 | * subset of node_states[N_MEMORY], even if this means going outside the |
909d75a3 PJ |
2454 | * tasks cpuset. |
2455 | **/ | |
2456 | ||
2457 | nodemask_t cpuset_mems_allowed(struct task_struct *tsk) | |
2458 | { | |
2459 | nodemask_t mask; | |
8447a0fe | 2460 | unsigned long flags; |
909d75a3 | 2461 | |
8447a0fe | 2462 | spin_lock_irqsave(&callback_lock, flags); |
b8dadcb5 | 2463 | rcu_read_lock(); |
ae1c8023 | 2464 | guarantee_online_mems(task_cs(tsk), &mask); |
b8dadcb5 | 2465 | rcu_read_unlock(); |
8447a0fe | 2466 | spin_unlock_irqrestore(&callback_lock, flags); |
909d75a3 PJ |
2467 | |
2468 | return mask; | |
2469 | } | |
2470 | ||
d9fd8a6d | 2471 | /** |
19770b32 MG |
2472 | * cpuset_nodemask_valid_mems_allowed - check nodemask vs. curremt mems_allowed |
2473 | * @nodemask: the nodemask to be checked | |
d9fd8a6d | 2474 | * |
19770b32 | 2475 | * Are any of the nodes in the nodemask allowed in current->mems_allowed? |
1da177e4 | 2476 | */ |
19770b32 | 2477 | int cpuset_nodemask_valid_mems_allowed(nodemask_t *nodemask) |
1da177e4 | 2478 | { |
19770b32 | 2479 | return nodes_intersects(*nodemask, current->mems_allowed); |
1da177e4 LT |
2480 | } |
2481 | ||
9bf2229f | 2482 | /* |
78608366 PM |
2483 | * nearest_hardwall_ancestor() - Returns the nearest mem_exclusive or |
2484 | * mem_hardwall ancestor to the specified cpuset. Call holding | |
8447a0fe | 2485 | * callback_lock. If no ancestor is mem_exclusive or mem_hardwall |
78608366 | 2486 | * (an unusual configuration), then returns the root cpuset. |
9bf2229f | 2487 | */ |
c9710d80 | 2488 | static struct cpuset *nearest_hardwall_ancestor(struct cpuset *cs) |
9bf2229f | 2489 | { |
c431069f TH |
2490 | while (!(is_mem_exclusive(cs) || is_mem_hardwall(cs)) && parent_cs(cs)) |
2491 | cs = parent_cs(cs); | |
9bf2229f PJ |
2492 | return cs; |
2493 | } | |
2494 | ||
d9fd8a6d | 2495 | /** |
344736f2 | 2496 | * cpuset_node_allowed - Can we allocate on a memory node? |
a1bc5a4e | 2497 | * @node: is this an allowed node? |
02a0e53d | 2498 | * @gfp_mask: memory allocation flags |
d9fd8a6d | 2499 | * |
6e276d2a DR |
2500 | * If we're in interrupt, yes, we can always allocate. If @node is set in |
2501 | * current's mems_allowed, yes. If it's not a __GFP_HARDWALL request and this | |
2502 | * node is set in the nearest hardwalled cpuset ancestor to current's cpuset, | |
2503 | * yes. If current has access to memory reserves due to TIF_MEMDIE, yes. | |
9bf2229f PJ |
2504 | * Otherwise, no. |
2505 | * | |
2506 | * GFP_USER allocations are marked with the __GFP_HARDWALL bit, | |
c596d9f3 DR |
2507 | * and do not allow allocations outside the current tasks cpuset |
2508 | * unless the task has been OOM killed as is marked TIF_MEMDIE. | |
9bf2229f | 2509 | * GFP_KERNEL allocations are not so marked, so can escape to the |
78608366 | 2510 | * nearest enclosing hardwalled ancestor cpuset. |
9bf2229f | 2511 | * |
8447a0fe | 2512 | * Scanning up parent cpusets requires callback_lock. The |
02a0e53d PJ |
2513 | * __alloc_pages() routine only calls here with __GFP_HARDWALL bit |
2514 | * _not_ set if it's a GFP_KERNEL allocation, and all nodes in the | |
2515 | * current tasks mems_allowed came up empty on the first pass over | |
2516 | * the zonelist. So only GFP_KERNEL allocations, if all nodes in the | |
8447a0fe | 2517 | * cpuset are short of memory, might require taking the callback_lock. |
9bf2229f | 2518 | * |
36be57ff | 2519 | * The first call here from mm/page_alloc:get_page_from_freelist() |
02a0e53d PJ |
2520 | * has __GFP_HARDWALL set in gfp_mask, enforcing hardwall cpusets, |
2521 | * so no allocation on a node outside the cpuset is allowed (unless | |
2522 | * in interrupt, of course). | |
36be57ff PJ |
2523 | * |
2524 | * The second pass through get_page_from_freelist() doesn't even call | |
2525 | * here for GFP_ATOMIC calls. For those calls, the __alloc_pages() | |
2526 | * variable 'wait' is not set, and the bit ALLOC_CPUSET is not set | |
2527 | * in alloc_flags. That logic and the checks below have the combined | |
2528 | * affect that: | |
9bf2229f PJ |
2529 | * in_interrupt - any node ok (current task context irrelevant) |
2530 | * GFP_ATOMIC - any node ok | |
c596d9f3 | 2531 | * TIF_MEMDIE - any node ok |
78608366 | 2532 | * GFP_KERNEL - any node in enclosing hardwalled cpuset ok |
9bf2229f | 2533 | * GFP_USER - only nodes in current tasks mems allowed ok. |
02a0e53d | 2534 | */ |
002f2906 | 2535 | bool __cpuset_node_allowed(int node, gfp_t gfp_mask) |
1da177e4 | 2536 | { |
c9710d80 | 2537 | struct cpuset *cs; /* current cpuset ancestors */ |
29afd49b | 2538 | int allowed; /* is allocation in zone z allowed? */ |
8447a0fe | 2539 | unsigned long flags; |
9bf2229f | 2540 | |
6e276d2a | 2541 | if (in_interrupt()) |
002f2906 | 2542 | return true; |
9bf2229f | 2543 | if (node_isset(node, current->mems_allowed)) |
002f2906 | 2544 | return true; |
c596d9f3 DR |
2545 | /* |
2546 | * Allow tasks that have access to memory reserves because they have | |
2547 | * been OOM killed to get memory anywhere. | |
2548 | */ | |
2549 | if (unlikely(test_thread_flag(TIF_MEMDIE))) | |
002f2906 | 2550 | return true; |
9bf2229f | 2551 | if (gfp_mask & __GFP_HARDWALL) /* If hardwall request, stop here */ |
002f2906 | 2552 | return false; |
9bf2229f | 2553 | |
5563e770 | 2554 | if (current->flags & PF_EXITING) /* Let dying task have memory */ |
002f2906 | 2555 | return true; |
5563e770 | 2556 | |
9bf2229f | 2557 | /* Not hardwall and node outside mems_allowed: scan up cpusets */ |
8447a0fe | 2558 | spin_lock_irqsave(&callback_lock, flags); |
053199ed | 2559 | |
b8dadcb5 | 2560 | rcu_read_lock(); |
78608366 | 2561 | cs = nearest_hardwall_ancestor(task_cs(current)); |
99afb0fd | 2562 | allowed = node_isset(node, cs->mems_allowed); |
b8dadcb5 | 2563 | rcu_read_unlock(); |
053199ed | 2564 | |
8447a0fe | 2565 | spin_unlock_irqrestore(&callback_lock, flags); |
9bf2229f | 2566 | return allowed; |
1da177e4 LT |
2567 | } |
2568 | ||
825a46af | 2569 | /** |
6adef3eb JS |
2570 | * cpuset_mem_spread_node() - On which node to begin search for a file page |
2571 | * cpuset_slab_spread_node() - On which node to begin search for a slab page | |
825a46af PJ |
2572 | * |
2573 | * If a task is marked PF_SPREAD_PAGE or PF_SPREAD_SLAB (as for | |
2574 | * tasks in a cpuset with is_spread_page or is_spread_slab set), | |
2575 | * and if the memory allocation used cpuset_mem_spread_node() | |
2576 | * to determine on which node to start looking, as it will for | |
2577 | * certain page cache or slab cache pages such as used for file | |
2578 | * system buffers and inode caches, then instead of starting on the | |
2579 | * local node to look for a free page, rather spread the starting | |
2580 | * node around the tasks mems_allowed nodes. | |
2581 | * | |
2582 | * We don't have to worry about the returned node being offline | |
2583 | * because "it can't happen", and even if it did, it would be ok. | |
2584 | * | |
2585 | * The routines calling guarantee_online_mems() are careful to | |
2586 | * only set nodes in task->mems_allowed that are online. So it | |
2587 | * should not be possible for the following code to return an | |
2588 | * offline node. But if it did, that would be ok, as this routine | |
2589 | * is not returning the node where the allocation must be, only | |
2590 | * the node where the search should start. The zonelist passed to | |
2591 | * __alloc_pages() will include all nodes. If the slab allocator | |
2592 | * is passed an offline node, it will fall back to the local node. | |
2593 | * See kmem_cache_alloc_node(). | |
2594 | */ | |
2595 | ||
6adef3eb | 2596 | static int cpuset_spread_node(int *rotor) |
825a46af | 2597 | { |
0edaf86c | 2598 | return *rotor = next_node_in(*rotor, current->mems_allowed); |
825a46af | 2599 | } |
6adef3eb JS |
2600 | |
2601 | int cpuset_mem_spread_node(void) | |
2602 | { | |
778d3b0f MH |
2603 | if (current->cpuset_mem_spread_rotor == NUMA_NO_NODE) |
2604 | current->cpuset_mem_spread_rotor = | |
2605 | node_random(¤t->mems_allowed); | |
2606 | ||
6adef3eb JS |
2607 | return cpuset_spread_node(¤t->cpuset_mem_spread_rotor); |
2608 | } | |
2609 | ||
2610 | int cpuset_slab_spread_node(void) | |
2611 | { | |
778d3b0f MH |
2612 | if (current->cpuset_slab_spread_rotor == NUMA_NO_NODE) |
2613 | current->cpuset_slab_spread_rotor = | |
2614 | node_random(¤t->mems_allowed); | |
2615 | ||
6adef3eb JS |
2616 | return cpuset_spread_node(¤t->cpuset_slab_spread_rotor); |
2617 | } | |
2618 | ||
825a46af PJ |
2619 | EXPORT_SYMBOL_GPL(cpuset_mem_spread_node); |
2620 | ||
ef08e3b4 | 2621 | /** |
bbe373f2 DR |
2622 | * cpuset_mems_allowed_intersects - Does @tsk1's mems_allowed intersect @tsk2's? |
2623 | * @tsk1: pointer to task_struct of some task. | |
2624 | * @tsk2: pointer to task_struct of some other task. | |
2625 | * | |
2626 | * Description: Return true if @tsk1's mems_allowed intersects the | |
2627 | * mems_allowed of @tsk2. Used by the OOM killer to determine if | |
2628 | * one of the task's memory usage might impact the memory available | |
2629 | * to the other. | |
ef08e3b4 PJ |
2630 | **/ |
2631 | ||
bbe373f2 DR |
2632 | int cpuset_mems_allowed_intersects(const struct task_struct *tsk1, |
2633 | const struct task_struct *tsk2) | |
ef08e3b4 | 2634 | { |
bbe373f2 | 2635 | return nodes_intersects(tsk1->mems_allowed, tsk2->mems_allowed); |
ef08e3b4 PJ |
2636 | } |
2637 | ||
75aa1994 | 2638 | /** |
da39da3a | 2639 | * cpuset_print_current_mems_allowed - prints current's cpuset and mems_allowed |
75aa1994 | 2640 | * |
da39da3a | 2641 | * Description: Prints current's name, cpuset name, and cached copy of its |
b8dadcb5 | 2642 | * mems_allowed to the kernel log. |
75aa1994 | 2643 | */ |
da39da3a | 2644 | void cpuset_print_current_mems_allowed(void) |
75aa1994 | 2645 | { |
b8dadcb5 | 2646 | struct cgroup *cgrp; |
75aa1994 | 2647 | |
b8dadcb5 | 2648 | rcu_read_lock(); |
63f43f55 | 2649 | |
da39da3a DR |
2650 | cgrp = task_cs(current)->css.cgroup; |
2651 | pr_info("%s cpuset=", current->comm); | |
e61734c5 | 2652 | pr_cont_cgroup_name(cgrp); |
da39da3a DR |
2653 | pr_cont(" mems_allowed=%*pbl\n", |
2654 | nodemask_pr_args(¤t->mems_allowed)); | |
f440d98f | 2655 | |
cfb5966b | 2656 | rcu_read_unlock(); |
75aa1994 DR |
2657 | } |
2658 | ||
3e0d98b9 PJ |
2659 | /* |
2660 | * Collection of memory_pressure is suppressed unless | |
2661 | * this flag is enabled by writing "1" to the special | |
2662 | * cpuset file 'memory_pressure_enabled' in the root cpuset. | |
2663 | */ | |
2664 | ||
c5b2aff8 | 2665 | int cpuset_memory_pressure_enabled __read_mostly; |
3e0d98b9 PJ |
2666 | |
2667 | /** | |
2668 | * cpuset_memory_pressure_bump - keep stats of per-cpuset reclaims. | |
2669 | * | |
2670 | * Keep a running average of the rate of synchronous (direct) | |
2671 | * page reclaim efforts initiated by tasks in each cpuset. | |
2672 | * | |
2673 | * This represents the rate at which some task in the cpuset | |
2674 | * ran low on memory on all nodes it was allowed to use, and | |
2675 | * had to enter the kernels page reclaim code in an effort to | |
2676 | * create more free memory by tossing clean pages or swapping | |
2677 | * or writing dirty pages. | |
2678 | * | |
2679 | * Display to user space in the per-cpuset read-only file | |
2680 | * "memory_pressure". Value displayed is an integer | |
2681 | * representing the recent rate of entry into the synchronous | |
2682 | * (direct) page reclaim by any task attached to the cpuset. | |
2683 | **/ | |
2684 | ||
2685 | void __cpuset_memory_pressure_bump(void) | |
2686 | { | |
b8dadcb5 | 2687 | rcu_read_lock(); |
8793d854 | 2688 | fmeter_markevent(&task_cs(current)->fmeter); |
b8dadcb5 | 2689 | rcu_read_unlock(); |
3e0d98b9 PJ |
2690 | } |
2691 | ||
8793d854 | 2692 | #ifdef CONFIG_PROC_PID_CPUSET |
1da177e4 LT |
2693 | /* |
2694 | * proc_cpuset_show() | |
2695 | * - Print tasks cpuset path into seq_file. | |
2696 | * - Used for /proc/<pid>/cpuset. | |
053199ed PJ |
2697 | * - No need to task_lock(tsk) on this tsk->cpuset reference, as it |
2698 | * doesn't really matter if tsk->cpuset changes after we read it, | |
5d21cc2d | 2699 | * and we take cpuset_mutex, keeping cpuset_attach() from changing it |
2df167a3 | 2700 | * anyway. |
1da177e4 | 2701 | */ |
52de4779 ZL |
2702 | int proc_cpuset_show(struct seq_file *m, struct pid_namespace *ns, |
2703 | struct pid *pid, struct task_struct *tsk) | |
1da177e4 | 2704 | { |
4c737b41 | 2705 | char *buf; |
8793d854 | 2706 | struct cgroup_subsys_state *css; |
99f89551 | 2707 | int retval; |
1da177e4 | 2708 | |
99f89551 | 2709 | retval = -ENOMEM; |
e61734c5 | 2710 | buf = kmalloc(PATH_MAX, GFP_KERNEL); |
1da177e4 | 2711 | if (!buf) |
99f89551 EB |
2712 | goto out; |
2713 | ||
a79a908f | 2714 | css = task_get_css(tsk, cpuset_cgrp_id); |
4c737b41 TH |
2715 | retval = cgroup_path_ns(css->cgroup, buf, PATH_MAX, |
2716 | current->nsproxy->cgroup_ns); | |
a79a908f | 2717 | css_put(css); |
4c737b41 | 2718 | if (retval >= PATH_MAX) |
679a5e3f TH |
2719 | retval = -ENAMETOOLONG; |
2720 | if (retval < 0) | |
52de4779 | 2721 | goto out_free; |
4c737b41 | 2722 | seq_puts(m, buf); |
1da177e4 | 2723 | seq_putc(m, '\n'); |
e61734c5 | 2724 | retval = 0; |
99f89551 | 2725 | out_free: |
1da177e4 | 2726 | kfree(buf); |
99f89551 | 2727 | out: |
1da177e4 LT |
2728 | return retval; |
2729 | } | |
8793d854 | 2730 | #endif /* CONFIG_PROC_PID_CPUSET */ |
1da177e4 | 2731 | |
d01d4827 | 2732 | /* Display task mems_allowed in /proc/<pid>/status file. */ |
df5f8314 EB |
2733 | void cpuset_task_status_allowed(struct seq_file *m, struct task_struct *task) |
2734 | { | |
e8e6d97c TH |
2735 | seq_printf(m, "Mems_allowed:\t%*pb\n", |
2736 | nodemask_pr_args(&task->mems_allowed)); | |
2737 | seq_printf(m, "Mems_allowed_list:\t%*pbl\n", | |
2738 | nodemask_pr_args(&task->mems_allowed)); | |
1da177e4 | 2739 | } |