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 LT |
46 | #include <linux/sched.h> |
47 | #include <linux/seq_file.h> | |
22fb52dd | 48 | #include <linux/security.h> |
1da177e4 | 49 | #include <linux/slab.h> |
1da177e4 LT |
50 | #include <linux/spinlock.h> |
51 | #include <linux/stat.h> | |
52 | #include <linux/string.h> | |
53 | #include <linux/time.h> | |
d2b43658 | 54 | #include <linux/time64.h> |
1da177e4 LT |
55 | #include <linux/backing-dev.h> |
56 | #include <linux/sort.h> | |
57 | ||
58 | #include <asm/uaccess.h> | |
60063497 | 59 | #include <linux/atomic.h> |
3d3f26a7 | 60 | #include <linux/mutex.h> |
956db3ca | 61 | #include <linux/cgroup.h> |
e44193d3 | 62 | #include <linux/wait.h> |
1da177e4 | 63 | |
002f2906 | 64 | DEFINE_STATIC_KEY_FALSE(cpusets_enabled_key); |
202f72d5 | 65 | |
3e0d98b9 PJ |
66 | /* See "Frequency meter" comments, below. */ |
67 | ||
68 | struct fmeter { | |
69 | int cnt; /* unprocessed events count */ | |
70 | int val; /* most recent output value */ | |
d2b43658 | 71 | time64_t time; /* clock (secs) when val computed */ |
3e0d98b9 PJ |
72 | spinlock_t lock; /* guards read or write of above */ |
73 | }; | |
74 | ||
1da177e4 | 75 | struct cpuset { |
8793d854 PM |
76 | struct cgroup_subsys_state css; |
77 | ||
1da177e4 | 78 | unsigned long flags; /* "unsigned long" so bitops work */ |
e2b9a3d7 | 79 | |
7e88291b LZ |
80 | /* |
81 | * On default hierarchy: | |
82 | * | |
83 | * The user-configured masks can only be changed by writing to | |
84 | * cpuset.cpus and cpuset.mems, and won't be limited by the | |
85 | * parent masks. | |
86 | * | |
87 | * The effective masks is the real masks that apply to the tasks | |
88 | * in the cpuset. They may be changed if the configured masks are | |
89 | * changed or hotplug happens. | |
90 | * | |
91 | * effective_mask == configured_mask & parent's effective_mask, | |
92 | * and if it ends up empty, it will inherit the parent's mask. | |
93 | * | |
94 | * | |
95 | * On legacy hierachy: | |
96 | * | |
97 | * The user-configured masks are always the same with effective masks. | |
98 | */ | |
99 | ||
e2b9a3d7 LZ |
100 | /* user-configured CPUs and Memory Nodes allow to tasks */ |
101 | cpumask_var_t cpus_allowed; | |
102 | nodemask_t mems_allowed; | |
103 | ||
104 | /* effective CPUs and Memory Nodes allow to tasks */ | |
105 | cpumask_var_t effective_cpus; | |
106 | nodemask_t effective_mems; | |
1da177e4 | 107 | |
33ad801d LZ |
108 | /* |
109 | * This is old Memory Nodes tasks took on. | |
110 | * | |
111 | * - top_cpuset.old_mems_allowed is initialized to mems_allowed. | |
112 | * - A new cpuset's old_mems_allowed is initialized when some | |
113 | * task is moved into it. | |
114 | * - old_mems_allowed is used in cpuset_migrate_mm() when we change | |
115 | * cpuset.mems_allowed and have tasks' nodemask updated, and | |
116 | * then old_mems_allowed is updated to mems_allowed. | |
117 | */ | |
118 | nodemask_t old_mems_allowed; | |
119 | ||
3e0d98b9 | 120 | struct fmeter fmeter; /* memory_pressure filter */ |
029190c5 | 121 | |
452477fa TH |
122 | /* |
123 | * Tasks are being attached to this cpuset. Used to prevent | |
124 | * zeroing cpus/mems_allowed between ->can_attach() and ->attach(). | |
125 | */ | |
126 | int attach_in_progress; | |
127 | ||
029190c5 PJ |
128 | /* partition number for rebuild_sched_domains() */ |
129 | int pn; | |
956db3ca | 130 | |
1d3504fc HS |
131 | /* for custom sched domain */ |
132 | int relax_domain_level; | |
1da177e4 LT |
133 | }; |
134 | ||
a7c6d554 | 135 | static inline struct cpuset *css_cs(struct cgroup_subsys_state *css) |
8793d854 | 136 | { |
a7c6d554 | 137 | return css ? container_of(css, struct cpuset, css) : NULL; |
8793d854 PM |
138 | } |
139 | ||
140 | /* Retrieve the cpuset for a task */ | |
141 | static inline struct cpuset *task_cs(struct task_struct *task) | |
142 | { | |
073219e9 | 143 | return css_cs(task_css(task, cpuset_cgrp_id)); |
8793d854 | 144 | } |
8793d854 | 145 | |
c9710d80 | 146 | static inline struct cpuset *parent_cs(struct cpuset *cs) |
c431069f | 147 | { |
5c9d535b | 148 | return css_cs(cs->css.parent); |
c431069f TH |
149 | } |
150 | ||
b246272e DR |
151 | #ifdef CONFIG_NUMA |
152 | static inline bool task_has_mempolicy(struct task_struct *task) | |
153 | { | |
154 | return task->mempolicy; | |
155 | } | |
156 | #else | |
157 | static inline bool task_has_mempolicy(struct task_struct *task) | |
158 | { | |
159 | return false; | |
160 | } | |
161 | #endif | |
162 | ||
163 | ||
1da177e4 LT |
164 | /* bits in struct cpuset flags field */ |
165 | typedef enum { | |
efeb77b2 | 166 | CS_ONLINE, |
1da177e4 LT |
167 | CS_CPU_EXCLUSIVE, |
168 | CS_MEM_EXCLUSIVE, | |
78608366 | 169 | CS_MEM_HARDWALL, |
45b07ef3 | 170 | CS_MEMORY_MIGRATE, |
029190c5 | 171 | CS_SCHED_LOAD_BALANCE, |
825a46af PJ |
172 | CS_SPREAD_PAGE, |
173 | CS_SPREAD_SLAB, | |
1da177e4 LT |
174 | } cpuset_flagbits_t; |
175 | ||
176 | /* convenient tests for these bits */ | |
efeb77b2 TH |
177 | static inline bool is_cpuset_online(const struct cpuset *cs) |
178 | { | |
179 | return test_bit(CS_ONLINE, &cs->flags); | |
180 | } | |
181 | ||
1da177e4 LT |
182 | static inline int is_cpu_exclusive(const struct cpuset *cs) |
183 | { | |
7b5b9ef0 | 184 | return test_bit(CS_CPU_EXCLUSIVE, &cs->flags); |
1da177e4 LT |
185 | } |
186 | ||
187 | static inline int is_mem_exclusive(const struct cpuset *cs) | |
188 | { | |
7b5b9ef0 | 189 | return test_bit(CS_MEM_EXCLUSIVE, &cs->flags); |
1da177e4 LT |
190 | } |
191 | ||
78608366 PM |
192 | static inline int is_mem_hardwall(const struct cpuset *cs) |
193 | { | |
194 | return test_bit(CS_MEM_HARDWALL, &cs->flags); | |
195 | } | |
196 | ||
029190c5 PJ |
197 | static inline int is_sched_load_balance(const struct cpuset *cs) |
198 | { | |
199 | return test_bit(CS_SCHED_LOAD_BALANCE, &cs->flags); | |
200 | } | |
201 | ||
45b07ef3 PJ |
202 | static inline int is_memory_migrate(const struct cpuset *cs) |
203 | { | |
7b5b9ef0 | 204 | return test_bit(CS_MEMORY_MIGRATE, &cs->flags); |
45b07ef3 PJ |
205 | } |
206 | ||
825a46af PJ |
207 | static inline int is_spread_page(const struct cpuset *cs) |
208 | { | |
209 | return test_bit(CS_SPREAD_PAGE, &cs->flags); | |
210 | } | |
211 | ||
212 | static inline int is_spread_slab(const struct cpuset *cs) | |
213 | { | |
214 | return test_bit(CS_SPREAD_SLAB, &cs->flags); | |
215 | } | |
216 | ||
1da177e4 | 217 | static struct cpuset top_cpuset = { |
efeb77b2 TH |
218 | .flags = ((1 << CS_ONLINE) | (1 << CS_CPU_EXCLUSIVE) | |
219 | (1 << CS_MEM_EXCLUSIVE)), | |
1da177e4 LT |
220 | }; |
221 | ||
ae8086ce TH |
222 | /** |
223 | * cpuset_for_each_child - traverse online children of a cpuset | |
224 | * @child_cs: loop cursor pointing to the current child | |
492eb21b | 225 | * @pos_css: used for iteration |
ae8086ce TH |
226 | * @parent_cs: target cpuset to walk children of |
227 | * | |
228 | * Walk @child_cs through the online children of @parent_cs. Must be used | |
229 | * with RCU read locked. | |
230 | */ | |
492eb21b TH |
231 | #define cpuset_for_each_child(child_cs, pos_css, parent_cs) \ |
232 | css_for_each_child((pos_css), &(parent_cs)->css) \ | |
233 | if (is_cpuset_online(((child_cs) = css_cs((pos_css))))) | |
ae8086ce | 234 | |
fc560a26 TH |
235 | /** |
236 | * cpuset_for_each_descendant_pre - pre-order walk of a cpuset's descendants | |
237 | * @des_cs: loop cursor pointing to the current descendant | |
492eb21b | 238 | * @pos_css: used for iteration |
fc560a26 TH |
239 | * @root_cs: target cpuset to walk ancestor of |
240 | * | |
241 | * Walk @des_cs through the online descendants of @root_cs. Must be used | |
492eb21b | 242 | * with RCU read locked. The caller may modify @pos_css by calling |
bd8815a6 TH |
243 | * css_rightmost_descendant() to skip subtree. @root_cs is included in the |
244 | * iteration and the first node to be visited. | |
fc560a26 | 245 | */ |
492eb21b TH |
246 | #define cpuset_for_each_descendant_pre(des_cs, pos_css, root_cs) \ |
247 | css_for_each_descendant_pre((pos_css), &(root_cs)->css) \ | |
248 | if (is_cpuset_online(((des_cs) = css_cs((pos_css))))) | |
fc560a26 | 249 | |
1da177e4 | 250 | /* |
8447a0fe VD |
251 | * There are two global locks guarding cpuset structures - cpuset_mutex and |
252 | * callback_lock. We also require taking task_lock() when dereferencing a | |
253 | * task's cpuset pointer. See "The task_lock() exception", at the end of this | |
254 | * comment. | |
5d21cc2d | 255 | * |
8447a0fe | 256 | * A task must hold both locks to modify cpusets. If a task holds |
5d21cc2d | 257 | * cpuset_mutex, then it blocks others wanting that mutex, ensuring that it |
8447a0fe | 258 | * is the only task able to also acquire callback_lock and be able to |
5d21cc2d TH |
259 | * modify cpusets. It can perform various checks on the cpuset structure |
260 | * first, knowing nothing will change. It can also allocate memory while | |
261 | * just holding cpuset_mutex. While it is performing these checks, various | |
8447a0fe VD |
262 | * callback routines can briefly acquire callback_lock to query cpusets. |
263 | * Once it is ready to make the changes, it takes callback_lock, blocking | |
5d21cc2d | 264 | * everyone else. |
053199ed PJ |
265 | * |
266 | * Calls to the kernel memory allocator can not be made while holding | |
8447a0fe | 267 | * callback_lock, as that would risk double tripping on callback_lock |
053199ed PJ |
268 | * from one of the callbacks into the cpuset code from within |
269 | * __alloc_pages(). | |
270 | * | |
8447a0fe | 271 | * If a task is only holding callback_lock, then it has read-only |
053199ed PJ |
272 | * access to cpusets. |
273 | * | |
58568d2a MX |
274 | * Now, the task_struct fields mems_allowed and mempolicy may be changed |
275 | * by other task, we use alloc_lock in the task_struct fields to protect | |
276 | * them. | |
053199ed | 277 | * |
8447a0fe | 278 | * The cpuset_common_file_read() handlers only hold callback_lock across |
053199ed PJ |
279 | * small pieces of code, such as when reading out possibly multi-word |
280 | * cpumasks and nodemasks. | |
281 | * | |
2df167a3 PM |
282 | * Accessing a task's cpuset should be done in accordance with the |
283 | * guidelines for accessing subsystem state in kernel/cgroup.c | |
1da177e4 LT |
284 | */ |
285 | ||
5d21cc2d | 286 | static DEFINE_MUTEX(cpuset_mutex); |
8447a0fe | 287 | static DEFINE_SPINLOCK(callback_lock); |
4247bdc6 | 288 | |
e93ad19d TH |
289 | static struct workqueue_struct *cpuset_migrate_mm_wq; |
290 | ||
3a5a6d0c TH |
291 | /* |
292 | * CPU / memory hotplug is handled asynchronously. | |
293 | */ | |
294 | static void cpuset_hotplug_workfn(struct work_struct *work); | |
3a5a6d0c TH |
295 | static DECLARE_WORK(cpuset_hotplug_work, cpuset_hotplug_workfn); |
296 | ||
e44193d3 LZ |
297 | static DECLARE_WAIT_QUEUE_HEAD(cpuset_attach_wq); |
298 | ||
cf417141 MK |
299 | /* |
300 | * This is ugly, but preserves the userspace API for existing cpuset | |
8793d854 | 301 | * users. If someone tries to mount the "cpuset" filesystem, we |
cf417141 MK |
302 | * silently switch it to mount "cgroup" instead |
303 | */ | |
f7e83571 AV |
304 | static struct dentry *cpuset_mount(struct file_system_type *fs_type, |
305 | int flags, const char *unused_dev_name, void *data) | |
1da177e4 | 306 | { |
8793d854 | 307 | struct file_system_type *cgroup_fs = get_fs_type("cgroup"); |
f7e83571 | 308 | struct dentry *ret = ERR_PTR(-ENODEV); |
8793d854 PM |
309 | if (cgroup_fs) { |
310 | char mountopts[] = | |
311 | "cpuset,noprefix," | |
312 | "release_agent=/sbin/cpuset_release_agent"; | |
f7e83571 AV |
313 | ret = cgroup_fs->mount(cgroup_fs, flags, |
314 | unused_dev_name, mountopts); | |
8793d854 PM |
315 | put_filesystem(cgroup_fs); |
316 | } | |
317 | return ret; | |
1da177e4 LT |
318 | } |
319 | ||
320 | static struct file_system_type cpuset_fs_type = { | |
321 | .name = "cpuset", | |
f7e83571 | 322 | .mount = cpuset_mount, |
1da177e4 LT |
323 | }; |
324 | ||
1da177e4 | 325 | /* |
300ed6cb | 326 | * Return in pmask the portion of a cpusets's cpus_allowed that |
1da177e4 | 327 | * are online. If none are online, walk up the cpuset hierarchy |
40df2deb LZ |
328 | * until we find one that does have some online cpus. The top |
329 | * cpuset always has some cpus online. | |
1da177e4 LT |
330 | * |
331 | * One way or another, we guarantee to return some non-empty subset | |
5f054e31 | 332 | * of cpu_online_mask. |
1da177e4 | 333 | * |
8447a0fe | 334 | * Call with callback_lock or cpuset_mutex held. |
1da177e4 | 335 | */ |
c9710d80 | 336 | static void guarantee_online_cpus(struct cpuset *cs, struct cpumask *pmask) |
1da177e4 | 337 | { |
ae1c8023 | 338 | while (!cpumask_intersects(cs->effective_cpus, cpu_online_mask)) |
c431069f | 339 | cs = parent_cs(cs); |
ae1c8023 | 340 | cpumask_and(pmask, cs->effective_cpus, cpu_online_mask); |
1da177e4 LT |
341 | } |
342 | ||
343 | /* | |
344 | * Return in *pmask the portion of a cpusets's mems_allowed that | |
0e1e7c7a CL |
345 | * are online, with memory. If none are online with memory, walk |
346 | * up the cpuset hierarchy until we find one that does have some | |
40df2deb | 347 | * online mems. The top cpuset always has some mems online. |
1da177e4 LT |
348 | * |
349 | * One way or another, we guarantee to return some non-empty subset | |
38d7bee9 | 350 | * of node_states[N_MEMORY]. |
1da177e4 | 351 | * |
8447a0fe | 352 | * Call with callback_lock or cpuset_mutex held. |
1da177e4 | 353 | */ |
c9710d80 | 354 | static void guarantee_online_mems(struct cpuset *cs, nodemask_t *pmask) |
1da177e4 | 355 | { |
ae1c8023 | 356 | while (!nodes_intersects(cs->effective_mems, node_states[N_MEMORY])) |
c431069f | 357 | cs = parent_cs(cs); |
ae1c8023 | 358 | nodes_and(*pmask, cs->effective_mems, node_states[N_MEMORY]); |
1da177e4 LT |
359 | } |
360 | ||
f3b39d47 MX |
361 | /* |
362 | * update task's spread flag if cpuset's page/slab spread flag is set | |
363 | * | |
8447a0fe | 364 | * Call with callback_lock or cpuset_mutex held. |
f3b39d47 MX |
365 | */ |
366 | static void cpuset_update_task_spread_flag(struct cpuset *cs, | |
367 | struct task_struct *tsk) | |
368 | { | |
369 | if (is_spread_page(cs)) | |
2ad654bc | 370 | task_set_spread_page(tsk); |
f3b39d47 | 371 | else |
2ad654bc ZL |
372 | task_clear_spread_page(tsk); |
373 | ||
f3b39d47 | 374 | if (is_spread_slab(cs)) |
2ad654bc | 375 | task_set_spread_slab(tsk); |
f3b39d47 | 376 | else |
2ad654bc | 377 | task_clear_spread_slab(tsk); |
f3b39d47 MX |
378 | } |
379 | ||
1da177e4 LT |
380 | /* |
381 | * is_cpuset_subset(p, q) - Is cpuset p a subset of cpuset q? | |
382 | * | |
383 | * One cpuset is a subset of another if all its allowed CPUs and | |
384 | * Memory Nodes are a subset of the other, and its exclusive flags | |
5d21cc2d | 385 | * are only set if the other's are set. Call holding cpuset_mutex. |
1da177e4 LT |
386 | */ |
387 | ||
388 | static int is_cpuset_subset(const struct cpuset *p, const struct cpuset *q) | |
389 | { | |
300ed6cb | 390 | return cpumask_subset(p->cpus_allowed, q->cpus_allowed) && |
1da177e4 LT |
391 | nodes_subset(p->mems_allowed, q->mems_allowed) && |
392 | is_cpu_exclusive(p) <= is_cpu_exclusive(q) && | |
393 | is_mem_exclusive(p) <= is_mem_exclusive(q); | |
394 | } | |
395 | ||
645fcc9d LZ |
396 | /** |
397 | * alloc_trial_cpuset - allocate a trial cpuset | |
398 | * @cs: the cpuset that the trial cpuset duplicates | |
399 | */ | |
c9710d80 | 400 | static struct cpuset *alloc_trial_cpuset(struct cpuset *cs) |
645fcc9d | 401 | { |
300ed6cb LZ |
402 | struct cpuset *trial; |
403 | ||
404 | trial = kmemdup(cs, sizeof(*cs), GFP_KERNEL); | |
405 | if (!trial) | |
406 | return NULL; | |
407 | ||
e2b9a3d7 LZ |
408 | if (!alloc_cpumask_var(&trial->cpus_allowed, GFP_KERNEL)) |
409 | goto free_cs; | |
410 | if (!alloc_cpumask_var(&trial->effective_cpus, GFP_KERNEL)) | |
411 | goto free_cpus; | |
300ed6cb | 412 | |
e2b9a3d7 LZ |
413 | cpumask_copy(trial->cpus_allowed, cs->cpus_allowed); |
414 | cpumask_copy(trial->effective_cpus, cs->effective_cpus); | |
300ed6cb | 415 | return trial; |
e2b9a3d7 LZ |
416 | |
417 | free_cpus: | |
418 | free_cpumask_var(trial->cpus_allowed); | |
419 | free_cs: | |
420 | kfree(trial); | |
421 | return NULL; | |
645fcc9d LZ |
422 | } |
423 | ||
424 | /** | |
425 | * free_trial_cpuset - free the trial cpuset | |
426 | * @trial: the trial cpuset to be freed | |
427 | */ | |
428 | static void free_trial_cpuset(struct cpuset *trial) | |
429 | { | |
e2b9a3d7 | 430 | free_cpumask_var(trial->effective_cpus); |
300ed6cb | 431 | free_cpumask_var(trial->cpus_allowed); |
645fcc9d LZ |
432 | kfree(trial); |
433 | } | |
434 | ||
1da177e4 LT |
435 | /* |
436 | * validate_change() - Used to validate that any proposed cpuset change | |
437 | * follows the structural rules for cpusets. | |
438 | * | |
439 | * If we replaced the flag and mask values of the current cpuset | |
440 | * (cur) with those values in the trial cpuset (trial), would | |
441 | * our various subset and exclusive rules still be valid? Presumes | |
5d21cc2d | 442 | * cpuset_mutex held. |
1da177e4 LT |
443 | * |
444 | * 'cur' is the address of an actual, in-use cpuset. Operations | |
445 | * such as list traversal that depend on the actual address of the | |
446 | * cpuset in the list must use cur below, not trial. | |
447 | * | |
448 | * 'trial' is the address of bulk structure copy of cur, with | |
449 | * perhaps one or more of the fields cpus_allowed, mems_allowed, | |
450 | * or flags changed to new, trial values. | |
451 | * | |
452 | * Return 0 if valid, -errno if not. | |
453 | */ | |
454 | ||
c9710d80 | 455 | static int validate_change(struct cpuset *cur, struct cpuset *trial) |
1da177e4 | 456 | { |
492eb21b | 457 | struct cgroup_subsys_state *css; |
1da177e4 | 458 | struct cpuset *c, *par; |
ae8086ce TH |
459 | int ret; |
460 | ||
461 | rcu_read_lock(); | |
1da177e4 LT |
462 | |
463 | /* Each of our child cpusets must be a subset of us */ | |
ae8086ce | 464 | ret = -EBUSY; |
492eb21b | 465 | cpuset_for_each_child(c, css, cur) |
ae8086ce TH |
466 | if (!is_cpuset_subset(c, trial)) |
467 | goto out; | |
1da177e4 LT |
468 | |
469 | /* Remaining checks don't apply to root cpuset */ | |
ae8086ce | 470 | ret = 0; |
69604067 | 471 | if (cur == &top_cpuset) |
ae8086ce | 472 | goto out; |
1da177e4 | 473 | |
c431069f | 474 | par = parent_cs(cur); |
69604067 | 475 | |
7e88291b | 476 | /* On legacy hiearchy, we must be a subset of our parent cpuset. */ |
ae8086ce | 477 | ret = -EACCES; |
9e10a130 TH |
478 | if (!cgroup_subsys_on_dfl(cpuset_cgrp_subsys) && |
479 | !is_cpuset_subset(trial, par)) | |
ae8086ce | 480 | goto out; |
1da177e4 | 481 | |
2df167a3 PM |
482 | /* |
483 | * If either I or some sibling (!= me) is exclusive, we can't | |
484 | * overlap | |
485 | */ | |
ae8086ce | 486 | ret = -EINVAL; |
492eb21b | 487 | cpuset_for_each_child(c, css, par) { |
1da177e4 LT |
488 | if ((is_cpu_exclusive(trial) || is_cpu_exclusive(c)) && |
489 | c != cur && | |
300ed6cb | 490 | cpumask_intersects(trial->cpus_allowed, c->cpus_allowed)) |
ae8086ce | 491 | goto out; |
1da177e4 LT |
492 | if ((is_mem_exclusive(trial) || is_mem_exclusive(c)) && |
493 | c != cur && | |
494 | nodes_intersects(trial->mems_allowed, c->mems_allowed)) | |
ae8086ce | 495 | goto out; |
1da177e4 LT |
496 | } |
497 | ||
452477fa TH |
498 | /* |
499 | * Cpusets with tasks - existing or newly being attached - can't | |
1c09b195 | 500 | * be changed to have empty cpus_allowed or mems_allowed. |
452477fa | 501 | */ |
ae8086ce | 502 | ret = -ENOSPC; |
27bd4dbb | 503 | if ((cgroup_is_populated(cur->css.cgroup) || cur->attach_in_progress)) { |
1c09b195 LZ |
504 | if (!cpumask_empty(cur->cpus_allowed) && |
505 | cpumask_empty(trial->cpus_allowed)) | |
506 | goto out; | |
507 | if (!nodes_empty(cur->mems_allowed) && | |
508 | nodes_empty(trial->mems_allowed)) | |
509 | goto out; | |
510 | } | |
020958b6 | 511 | |
f82f8042 JL |
512 | /* |
513 | * We can't shrink if we won't have enough room for SCHED_DEADLINE | |
514 | * tasks. | |
515 | */ | |
516 | ret = -EBUSY; | |
517 | if (is_cpu_exclusive(cur) && | |
518 | !cpuset_cpumask_can_shrink(cur->cpus_allowed, | |
519 | trial->cpus_allowed)) | |
520 | goto out; | |
521 | ||
ae8086ce TH |
522 | ret = 0; |
523 | out: | |
524 | rcu_read_unlock(); | |
525 | return ret; | |
1da177e4 LT |
526 | } |
527 | ||
db7f47cf | 528 | #ifdef CONFIG_SMP |
029190c5 | 529 | /* |
cf417141 | 530 | * Helper routine for generate_sched_domains(). |
8b5f1c52 | 531 | * Do cpusets a, b have overlapping effective cpus_allowed masks? |
029190c5 | 532 | */ |
029190c5 PJ |
533 | static int cpusets_overlap(struct cpuset *a, struct cpuset *b) |
534 | { | |
8b5f1c52 | 535 | return cpumask_intersects(a->effective_cpus, b->effective_cpus); |
029190c5 PJ |
536 | } |
537 | ||
1d3504fc HS |
538 | static void |
539 | update_domain_attr(struct sched_domain_attr *dattr, struct cpuset *c) | |
540 | { | |
1d3504fc HS |
541 | if (dattr->relax_domain_level < c->relax_domain_level) |
542 | dattr->relax_domain_level = c->relax_domain_level; | |
543 | return; | |
544 | } | |
545 | ||
fc560a26 TH |
546 | static void update_domain_attr_tree(struct sched_domain_attr *dattr, |
547 | struct cpuset *root_cs) | |
f5393693 | 548 | { |
fc560a26 | 549 | struct cpuset *cp; |
492eb21b | 550 | struct cgroup_subsys_state *pos_css; |
f5393693 | 551 | |
fc560a26 | 552 | rcu_read_lock(); |
492eb21b | 553 | cpuset_for_each_descendant_pre(cp, pos_css, root_cs) { |
fc560a26 TH |
554 | /* skip the whole subtree if @cp doesn't have any CPU */ |
555 | if (cpumask_empty(cp->cpus_allowed)) { | |
492eb21b | 556 | pos_css = css_rightmost_descendant(pos_css); |
f5393693 | 557 | continue; |
fc560a26 | 558 | } |
f5393693 LJ |
559 | |
560 | if (is_sched_load_balance(cp)) | |
561 | update_domain_attr(dattr, cp); | |
f5393693 | 562 | } |
fc560a26 | 563 | rcu_read_unlock(); |
f5393693 LJ |
564 | } |
565 | ||
029190c5 | 566 | /* |
cf417141 MK |
567 | * generate_sched_domains() |
568 | * | |
569 | * This function builds a partial partition of the systems CPUs | |
570 | * A 'partial partition' is a set of non-overlapping subsets whose | |
571 | * union is a subset of that set. | |
0a0fca9d | 572 | * The output of this function needs to be passed to kernel/sched/core.c |
cf417141 MK |
573 | * partition_sched_domains() routine, which will rebuild the scheduler's |
574 | * load balancing domains (sched domains) as specified by that partial | |
575 | * partition. | |
029190c5 | 576 | * |
45ce80fb | 577 | * See "What is sched_load_balance" in Documentation/cgroups/cpusets.txt |
029190c5 PJ |
578 | * for a background explanation of this. |
579 | * | |
580 | * Does not return errors, on the theory that the callers of this | |
581 | * routine would rather not worry about failures to rebuild sched | |
582 | * domains when operating in the severe memory shortage situations | |
583 | * that could cause allocation failures below. | |
584 | * | |
5d21cc2d | 585 | * Must be called with cpuset_mutex held. |
029190c5 PJ |
586 | * |
587 | * The three key local variables below are: | |
aeed6824 | 588 | * q - a linked-list queue of cpuset pointers, used to implement a |
029190c5 PJ |
589 | * top-down scan of all cpusets. This scan loads a pointer |
590 | * to each cpuset marked is_sched_load_balance into the | |
591 | * array 'csa'. For our purposes, rebuilding the schedulers | |
592 | * sched domains, we can ignore !is_sched_load_balance cpusets. | |
593 | * csa - (for CpuSet Array) Array of pointers to all the cpusets | |
594 | * that need to be load balanced, for convenient iterative | |
595 | * access by the subsequent code that finds the best partition, | |
596 | * i.e the set of domains (subsets) of CPUs such that the | |
597 | * cpus_allowed of every cpuset marked is_sched_load_balance | |
598 | * is a subset of one of these domains, while there are as | |
599 | * many such domains as possible, each as small as possible. | |
600 | * doms - Conversion of 'csa' to an array of cpumasks, for passing to | |
0a0fca9d | 601 | * the kernel/sched/core.c routine partition_sched_domains() in a |
029190c5 PJ |
602 | * convenient format, that can be easily compared to the prior |
603 | * value to determine what partition elements (sched domains) | |
604 | * were changed (added or removed.) | |
605 | * | |
606 | * Finding the best partition (set of domains): | |
607 | * The triple nested loops below over i, j, k scan over the | |
608 | * load balanced cpusets (using the array of cpuset pointers in | |
609 | * csa[]) looking for pairs of cpusets that have overlapping | |
610 | * cpus_allowed, but which don't have the same 'pn' partition | |
611 | * number and gives them in the same partition number. It keeps | |
612 | * looping on the 'restart' label until it can no longer find | |
613 | * any such pairs. | |
614 | * | |
615 | * The union of the cpus_allowed masks from the set of | |
616 | * all cpusets having the same 'pn' value then form the one | |
617 | * element of the partition (one sched domain) to be passed to | |
618 | * partition_sched_domains(). | |
619 | */ | |
acc3f5d7 | 620 | static int generate_sched_domains(cpumask_var_t **domains, |
cf417141 | 621 | struct sched_domain_attr **attributes) |
029190c5 | 622 | { |
029190c5 PJ |
623 | struct cpuset *cp; /* scans q */ |
624 | struct cpuset **csa; /* array of all cpuset ptrs */ | |
625 | int csn; /* how many cpuset ptrs in csa so far */ | |
626 | int i, j, k; /* indices for partition finding loops */ | |
acc3f5d7 | 627 | cpumask_var_t *doms; /* resulting partition; i.e. sched domains */ |
47b8ea71 | 628 | cpumask_var_t non_isolated_cpus; /* load balanced CPUs */ |
1d3504fc | 629 | struct sched_domain_attr *dattr; /* attributes for custom domains */ |
1583715d | 630 | int ndoms = 0; /* number of sched domains in result */ |
6af866af | 631 | int nslot; /* next empty doms[] struct cpumask slot */ |
492eb21b | 632 | struct cgroup_subsys_state *pos_css; |
029190c5 | 633 | |
029190c5 | 634 | doms = NULL; |
1d3504fc | 635 | dattr = NULL; |
cf417141 | 636 | csa = NULL; |
029190c5 | 637 | |
47b8ea71 RR |
638 | if (!alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL)) |
639 | goto done; | |
640 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); | |
641 | ||
029190c5 PJ |
642 | /* Special case for the 99% of systems with one, full, sched domain */ |
643 | if (is_sched_load_balance(&top_cpuset)) { | |
acc3f5d7 RR |
644 | ndoms = 1; |
645 | doms = alloc_sched_domains(ndoms); | |
029190c5 | 646 | if (!doms) |
cf417141 MK |
647 | goto done; |
648 | ||
1d3504fc HS |
649 | dattr = kmalloc(sizeof(struct sched_domain_attr), GFP_KERNEL); |
650 | if (dattr) { | |
651 | *dattr = SD_ATTR_INIT; | |
93a65575 | 652 | update_domain_attr_tree(dattr, &top_cpuset); |
1d3504fc | 653 | } |
47b8ea71 RR |
654 | cpumask_and(doms[0], top_cpuset.effective_cpus, |
655 | non_isolated_cpus); | |
cf417141 | 656 | |
cf417141 | 657 | goto done; |
029190c5 PJ |
658 | } |
659 | ||
664eedde | 660 | csa = kmalloc(nr_cpusets() * sizeof(cp), GFP_KERNEL); |
029190c5 PJ |
661 | if (!csa) |
662 | goto done; | |
663 | csn = 0; | |
664 | ||
fc560a26 | 665 | rcu_read_lock(); |
492eb21b | 666 | cpuset_for_each_descendant_pre(cp, pos_css, &top_cpuset) { |
bd8815a6 TH |
667 | if (cp == &top_cpuset) |
668 | continue; | |
f5393693 | 669 | /* |
fc560a26 TH |
670 | * Continue traversing beyond @cp iff @cp has some CPUs and |
671 | * isn't load balancing. The former is obvious. The | |
672 | * latter: All child cpusets contain a subset of the | |
673 | * parent's cpus, so just skip them, and then we call | |
674 | * update_domain_attr_tree() to calc relax_domain_level of | |
675 | * the corresponding sched domain. | |
f5393693 | 676 | */ |
fc560a26 | 677 | if (!cpumask_empty(cp->cpus_allowed) && |
47b8ea71 RR |
678 | !(is_sched_load_balance(cp) && |
679 | cpumask_intersects(cp->cpus_allowed, non_isolated_cpus))) | |
f5393693 | 680 | continue; |
489a5393 | 681 | |
fc560a26 TH |
682 | if (is_sched_load_balance(cp)) |
683 | csa[csn++] = cp; | |
684 | ||
685 | /* skip @cp's subtree */ | |
492eb21b | 686 | pos_css = css_rightmost_descendant(pos_css); |
fc560a26 TH |
687 | } |
688 | rcu_read_unlock(); | |
029190c5 PJ |
689 | |
690 | for (i = 0; i < csn; i++) | |
691 | csa[i]->pn = i; | |
692 | ndoms = csn; | |
693 | ||
694 | restart: | |
695 | /* Find the best partition (set of sched domains) */ | |
696 | for (i = 0; i < csn; i++) { | |
697 | struct cpuset *a = csa[i]; | |
698 | int apn = a->pn; | |
699 | ||
700 | for (j = 0; j < csn; j++) { | |
701 | struct cpuset *b = csa[j]; | |
702 | int bpn = b->pn; | |
703 | ||
704 | if (apn != bpn && cpusets_overlap(a, b)) { | |
705 | for (k = 0; k < csn; k++) { | |
706 | struct cpuset *c = csa[k]; | |
707 | ||
708 | if (c->pn == bpn) | |
709 | c->pn = apn; | |
710 | } | |
711 | ndoms--; /* one less element */ | |
712 | goto restart; | |
713 | } | |
714 | } | |
715 | } | |
716 | ||
cf417141 MK |
717 | /* |
718 | * Now we know how many domains to create. | |
719 | * Convert <csn, csa> to <ndoms, doms> and populate cpu masks. | |
720 | */ | |
acc3f5d7 | 721 | doms = alloc_sched_domains(ndoms); |
700018e0 | 722 | if (!doms) |
cf417141 | 723 | goto done; |
cf417141 MK |
724 | |
725 | /* | |
726 | * The rest of the code, including the scheduler, can deal with | |
727 | * dattr==NULL case. No need to abort if alloc fails. | |
728 | */ | |
1d3504fc | 729 | dattr = kmalloc(ndoms * sizeof(struct sched_domain_attr), GFP_KERNEL); |
029190c5 PJ |
730 | |
731 | for (nslot = 0, i = 0; i < csn; i++) { | |
732 | struct cpuset *a = csa[i]; | |
6af866af | 733 | struct cpumask *dp; |
029190c5 PJ |
734 | int apn = a->pn; |
735 | ||
cf417141 MK |
736 | if (apn < 0) { |
737 | /* Skip completed partitions */ | |
738 | continue; | |
739 | } | |
740 | ||
acc3f5d7 | 741 | dp = doms[nslot]; |
cf417141 MK |
742 | |
743 | if (nslot == ndoms) { | |
744 | static int warnings = 10; | |
745 | if (warnings) { | |
12d3089c FF |
746 | pr_warn("rebuild_sched_domains confused: nslot %d, ndoms %d, csn %d, i %d, apn %d\n", |
747 | nslot, ndoms, csn, i, apn); | |
cf417141 | 748 | warnings--; |
029190c5 | 749 | } |
cf417141 MK |
750 | continue; |
751 | } | |
029190c5 | 752 | |
6af866af | 753 | cpumask_clear(dp); |
cf417141 MK |
754 | if (dattr) |
755 | *(dattr + nslot) = SD_ATTR_INIT; | |
756 | for (j = i; j < csn; j++) { | |
757 | struct cpuset *b = csa[j]; | |
758 | ||
759 | if (apn == b->pn) { | |
8b5f1c52 | 760 | cpumask_or(dp, dp, b->effective_cpus); |
47b8ea71 | 761 | cpumask_and(dp, dp, non_isolated_cpus); |
cf417141 MK |
762 | if (dattr) |
763 | update_domain_attr_tree(dattr + nslot, b); | |
764 | ||
765 | /* Done with this partition */ | |
766 | b->pn = -1; | |
029190c5 | 767 | } |
029190c5 | 768 | } |
cf417141 | 769 | nslot++; |
029190c5 PJ |
770 | } |
771 | BUG_ON(nslot != ndoms); | |
772 | ||
cf417141 | 773 | done: |
47b8ea71 | 774 | free_cpumask_var(non_isolated_cpus); |
cf417141 MK |
775 | kfree(csa); |
776 | ||
700018e0 LZ |
777 | /* |
778 | * Fallback to the default domain if kmalloc() failed. | |
779 | * See comments in partition_sched_domains(). | |
780 | */ | |
781 | if (doms == NULL) | |
782 | ndoms = 1; | |
783 | ||
cf417141 MK |
784 | *domains = doms; |
785 | *attributes = dattr; | |
786 | return ndoms; | |
787 | } | |
788 | ||
789 | /* | |
790 | * Rebuild scheduler domains. | |
791 | * | |
699140ba TH |
792 | * If the flag 'sched_load_balance' of any cpuset with non-empty |
793 | * 'cpus' changes, or if the 'cpus' allowed changes in any cpuset | |
794 | * which has that flag enabled, or if any cpuset with a non-empty | |
795 | * 'cpus' is removed, then call this routine to rebuild the | |
796 | * scheduler's dynamic sched domains. | |
cf417141 | 797 | * |
5d21cc2d | 798 | * Call with cpuset_mutex held. Takes get_online_cpus(). |
cf417141 | 799 | */ |
699140ba | 800 | static void rebuild_sched_domains_locked(void) |
cf417141 MK |
801 | { |
802 | struct sched_domain_attr *attr; | |
acc3f5d7 | 803 | cpumask_var_t *doms; |
cf417141 MK |
804 | int ndoms; |
805 | ||
5d21cc2d | 806 | lockdep_assert_held(&cpuset_mutex); |
86ef5c9a | 807 | get_online_cpus(); |
cf417141 | 808 | |
5b16c2a4 LZ |
809 | /* |
810 | * We have raced with CPU hotplug. Don't do anything to avoid | |
811 | * passing doms with offlined cpu to partition_sched_domains(). | |
812 | * Anyways, hotplug work item will rebuild sched domains. | |
813 | */ | |
8b5f1c52 | 814 | if (!cpumask_equal(top_cpuset.effective_cpus, cpu_active_mask)) |
5b16c2a4 LZ |
815 | goto out; |
816 | ||
cf417141 | 817 | /* Generate domain masks and attrs */ |
cf417141 | 818 | ndoms = generate_sched_domains(&doms, &attr); |
cf417141 MK |
819 | |
820 | /* Have scheduler rebuild the domains */ | |
821 | partition_sched_domains(ndoms, doms, attr); | |
5b16c2a4 | 822 | out: |
86ef5c9a | 823 | put_online_cpus(); |
cf417141 | 824 | } |
db7f47cf | 825 | #else /* !CONFIG_SMP */ |
699140ba | 826 | static void rebuild_sched_domains_locked(void) |
db7f47cf PM |
827 | { |
828 | } | |
db7f47cf | 829 | #endif /* CONFIG_SMP */ |
029190c5 | 830 | |
cf417141 MK |
831 | void rebuild_sched_domains(void) |
832 | { | |
5d21cc2d | 833 | mutex_lock(&cpuset_mutex); |
699140ba | 834 | rebuild_sched_domains_locked(); |
5d21cc2d | 835 | mutex_unlock(&cpuset_mutex); |
029190c5 PJ |
836 | } |
837 | ||
0b2f630a MX |
838 | /** |
839 | * update_tasks_cpumask - Update the cpumasks of tasks in the cpuset. | |
840 | * @cs: the cpuset in which each task's cpus_allowed mask needs to be changed | |
0b2f630a | 841 | * |
d66393e5 TH |
842 | * Iterate through each task of @cs updating its cpus_allowed to the |
843 | * effective cpuset's. As this function is called with cpuset_mutex held, | |
844 | * cpuset membership stays stable. | |
0b2f630a | 845 | */ |
d66393e5 | 846 | static void update_tasks_cpumask(struct cpuset *cs) |
0b2f630a | 847 | { |
d66393e5 TH |
848 | struct css_task_iter it; |
849 | struct task_struct *task; | |
850 | ||
851 | css_task_iter_start(&cs->css, &it); | |
852 | while ((task = css_task_iter_next(&it))) | |
ae1c8023 | 853 | set_cpus_allowed_ptr(task, cs->effective_cpus); |
d66393e5 | 854 | css_task_iter_end(&it); |
0b2f630a MX |
855 | } |
856 | ||
5c5cc623 | 857 | /* |
734d4513 LZ |
858 | * update_cpumasks_hier - Update effective cpumasks and tasks in the subtree |
859 | * @cs: the cpuset to consider | |
860 | * @new_cpus: temp variable for calculating new effective_cpus | |
861 | * | |
862 | * When congifured cpumask is changed, the effective cpumasks of this cpuset | |
863 | * and all its descendants need to be updated. | |
5c5cc623 | 864 | * |
734d4513 | 865 | * On legacy hierachy, effective_cpus will be the same with cpu_allowed. |
5c5cc623 LZ |
866 | * |
867 | * Called with cpuset_mutex held | |
868 | */ | |
734d4513 | 869 | static void update_cpumasks_hier(struct cpuset *cs, struct cpumask *new_cpus) |
5c5cc623 LZ |
870 | { |
871 | struct cpuset *cp; | |
492eb21b | 872 | struct cgroup_subsys_state *pos_css; |
8b5f1c52 | 873 | bool need_rebuild_sched_domains = false; |
5c5cc623 LZ |
874 | |
875 | rcu_read_lock(); | |
734d4513 LZ |
876 | cpuset_for_each_descendant_pre(cp, pos_css, cs) { |
877 | struct cpuset *parent = parent_cs(cp); | |
878 | ||
879 | cpumask_and(new_cpus, cp->cpus_allowed, parent->effective_cpus); | |
880 | ||
554b0d1c LZ |
881 | /* |
882 | * If it becomes empty, inherit the effective mask of the | |
883 | * parent, which is guaranteed to have some CPUs. | |
884 | */ | |
9e10a130 TH |
885 | if (cgroup_subsys_on_dfl(cpuset_cgrp_subsys) && |
886 | cpumask_empty(new_cpus)) | |
554b0d1c LZ |
887 | cpumask_copy(new_cpus, parent->effective_cpus); |
888 | ||
734d4513 LZ |
889 | /* Skip the whole subtree if the cpumask remains the same. */ |
890 | if (cpumask_equal(new_cpus, cp->effective_cpus)) { | |
891 | pos_css = css_rightmost_descendant(pos_css); | |
892 | continue; | |
5c5cc623 | 893 | } |
734d4513 | 894 | |
ec903c0c | 895 | if (!css_tryget_online(&cp->css)) |
5c5cc623 LZ |
896 | continue; |
897 | rcu_read_unlock(); | |
898 | ||
8447a0fe | 899 | spin_lock_irq(&callback_lock); |
734d4513 | 900 | cpumask_copy(cp->effective_cpus, new_cpus); |
8447a0fe | 901 | spin_unlock_irq(&callback_lock); |
734d4513 | 902 | |
9e10a130 | 903 | WARN_ON(!cgroup_subsys_on_dfl(cpuset_cgrp_subsys) && |
734d4513 LZ |
904 | !cpumask_equal(cp->cpus_allowed, cp->effective_cpus)); |
905 | ||
d66393e5 | 906 | update_tasks_cpumask(cp); |
5c5cc623 | 907 | |
8b5f1c52 LZ |
908 | /* |
909 | * If the effective cpumask of any non-empty cpuset is changed, | |
910 | * we need to rebuild sched domains. | |
911 | */ | |
912 | if (!cpumask_empty(cp->cpus_allowed) && | |
913 | is_sched_load_balance(cp)) | |
914 | need_rebuild_sched_domains = true; | |
915 | ||
5c5cc623 LZ |
916 | rcu_read_lock(); |
917 | css_put(&cp->css); | |
918 | } | |
919 | rcu_read_unlock(); | |
8b5f1c52 LZ |
920 | |
921 | if (need_rebuild_sched_domains) | |
922 | rebuild_sched_domains_locked(); | |
5c5cc623 LZ |
923 | } |
924 | ||
58f4790b CW |
925 | /** |
926 | * update_cpumask - update the cpus_allowed mask of a cpuset and all tasks in it | |
927 | * @cs: the cpuset to consider | |
fc34ac1d | 928 | * @trialcs: trial cpuset |
58f4790b CW |
929 | * @buf: buffer of cpu numbers written to this cpuset |
930 | */ | |
645fcc9d LZ |
931 | static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs, |
932 | const char *buf) | |
1da177e4 | 933 | { |
58f4790b | 934 | int retval; |
1da177e4 | 935 | |
5f054e31 | 936 | /* top_cpuset.cpus_allowed tracks cpu_online_mask; it's read-only */ |
4c4d50f7 PJ |
937 | if (cs == &top_cpuset) |
938 | return -EACCES; | |
939 | ||
6f7f02e7 | 940 | /* |
c8d9c90c | 941 | * An empty cpus_allowed is ok only if the cpuset has no tasks. |
020958b6 PJ |
942 | * Since cpulist_parse() fails on an empty mask, we special case |
943 | * that parsing. The validate_change() call ensures that cpusets | |
944 | * with tasks have cpus. | |
6f7f02e7 | 945 | */ |
020958b6 | 946 | if (!*buf) { |
300ed6cb | 947 | cpumask_clear(trialcs->cpus_allowed); |
6f7f02e7 | 948 | } else { |
300ed6cb | 949 | retval = cpulist_parse(buf, trialcs->cpus_allowed); |
6f7f02e7 DR |
950 | if (retval < 0) |
951 | return retval; | |
37340746 | 952 | |
5d8ba82c LZ |
953 | if (!cpumask_subset(trialcs->cpus_allowed, |
954 | top_cpuset.cpus_allowed)) | |
37340746 | 955 | return -EINVAL; |
6f7f02e7 | 956 | } |
029190c5 | 957 | |
8707d8b8 | 958 | /* Nothing to do if the cpus didn't change */ |
300ed6cb | 959 | if (cpumask_equal(cs->cpus_allowed, trialcs->cpus_allowed)) |
8707d8b8 | 960 | return 0; |
58f4790b | 961 | |
a73456f3 LZ |
962 | retval = validate_change(cs, trialcs); |
963 | if (retval < 0) | |
964 | return retval; | |
965 | ||
8447a0fe | 966 | spin_lock_irq(&callback_lock); |
300ed6cb | 967 | cpumask_copy(cs->cpus_allowed, trialcs->cpus_allowed); |
8447a0fe | 968 | spin_unlock_irq(&callback_lock); |
029190c5 | 969 | |
734d4513 LZ |
970 | /* use trialcs->cpus_allowed as a temp variable */ |
971 | update_cpumasks_hier(cs, trialcs->cpus_allowed); | |
85d7b949 | 972 | return 0; |
1da177e4 LT |
973 | } |
974 | ||
e4e364e8 | 975 | /* |
e93ad19d TH |
976 | * Migrate memory region from one set of nodes to another. This is |
977 | * performed asynchronously as it can be called from process migration path | |
978 | * holding locks involved in process management. All mm migrations are | |
979 | * performed in the queued order and can be waited for by flushing | |
980 | * cpuset_migrate_mm_wq. | |
e4e364e8 PJ |
981 | */ |
982 | ||
e93ad19d TH |
983 | struct cpuset_migrate_mm_work { |
984 | struct work_struct work; | |
985 | struct mm_struct *mm; | |
986 | nodemask_t from; | |
987 | nodemask_t to; | |
988 | }; | |
989 | ||
990 | static void cpuset_migrate_mm_workfn(struct work_struct *work) | |
991 | { | |
992 | struct cpuset_migrate_mm_work *mwork = | |
993 | container_of(work, struct cpuset_migrate_mm_work, work); | |
994 | ||
995 | /* on a wq worker, no need to worry about %current's mems_allowed */ | |
996 | do_migrate_pages(mwork->mm, &mwork->from, &mwork->to, MPOL_MF_MOVE_ALL); | |
997 | mmput(mwork->mm); | |
998 | kfree(mwork); | |
999 | } | |
1000 | ||
e4e364e8 PJ |
1001 | static void cpuset_migrate_mm(struct mm_struct *mm, const nodemask_t *from, |
1002 | const nodemask_t *to) | |
1003 | { | |
e93ad19d | 1004 | struct cpuset_migrate_mm_work *mwork; |
e4e364e8 | 1005 | |
e93ad19d TH |
1006 | mwork = kzalloc(sizeof(*mwork), GFP_KERNEL); |
1007 | if (mwork) { | |
1008 | mwork->mm = mm; | |
1009 | mwork->from = *from; | |
1010 | mwork->to = *to; | |
1011 | INIT_WORK(&mwork->work, cpuset_migrate_mm_workfn); | |
1012 | queue_work(cpuset_migrate_mm_wq, &mwork->work); | |
1013 | } else { | |
1014 | mmput(mm); | |
1015 | } | |
1016 | } | |
e4e364e8 | 1017 | |
5cf1cacb | 1018 | static void cpuset_post_attach(void) |
e93ad19d TH |
1019 | { |
1020 | flush_workqueue(cpuset_migrate_mm_wq); | |
e4e364e8 PJ |
1021 | } |
1022 | ||
3b6766fe | 1023 | /* |
58568d2a MX |
1024 | * cpuset_change_task_nodemask - change task's mems_allowed and mempolicy |
1025 | * @tsk: the task to change | |
1026 | * @newmems: new nodes that the task will be set | |
1027 | * | |
1028 | * In order to avoid seeing no nodes if the old and new nodes are disjoint, | |
1029 | * we structure updates as setting all new allowed nodes, then clearing newly | |
1030 | * disallowed ones. | |
58568d2a MX |
1031 | */ |
1032 | static void cpuset_change_task_nodemask(struct task_struct *tsk, | |
1033 | nodemask_t *newmems) | |
1034 | { | |
b246272e | 1035 | bool need_loop; |
89e8a244 | 1036 | |
c0ff7453 | 1037 | task_lock(tsk); |
b246272e DR |
1038 | /* |
1039 | * Determine if a loop is necessary if another thread is doing | |
d26914d1 | 1040 | * read_mems_allowed_begin(). If at least one node remains unchanged and |
b246272e DR |
1041 | * tsk does not have a mempolicy, then an empty nodemask will not be |
1042 | * possible when mems_allowed is larger than a word. | |
1043 | */ | |
1044 | need_loop = task_has_mempolicy(tsk) || | |
1045 | !nodes_intersects(*newmems, tsk->mems_allowed); | |
c0ff7453 | 1046 | |
0fc0287c PZ |
1047 | if (need_loop) { |
1048 | local_irq_disable(); | |
cc9a6c87 | 1049 | write_seqcount_begin(&tsk->mems_allowed_seq); |
0fc0287c | 1050 | } |
c0ff7453 | 1051 | |
cc9a6c87 MG |
1052 | nodes_or(tsk->mems_allowed, tsk->mems_allowed, *newmems); |
1053 | mpol_rebind_task(tsk, newmems, MPOL_REBIND_STEP1); | |
c0ff7453 MX |
1054 | |
1055 | mpol_rebind_task(tsk, newmems, MPOL_REBIND_STEP2); | |
58568d2a | 1056 | tsk->mems_allowed = *newmems; |
cc9a6c87 | 1057 | |
0fc0287c | 1058 | if (need_loop) { |
cc9a6c87 | 1059 | write_seqcount_end(&tsk->mems_allowed_seq); |
0fc0287c PZ |
1060 | local_irq_enable(); |
1061 | } | |
cc9a6c87 | 1062 | |
c0ff7453 | 1063 | task_unlock(tsk); |
58568d2a MX |
1064 | } |
1065 | ||
8793d854 PM |
1066 | static void *cpuset_being_rebound; |
1067 | ||
0b2f630a MX |
1068 | /** |
1069 | * update_tasks_nodemask - Update the nodemasks of tasks in the cpuset. | |
1070 | * @cs: the cpuset in which each task's mems_allowed mask needs to be changed | |
0b2f630a | 1071 | * |
d66393e5 TH |
1072 | * Iterate through each task of @cs updating its mems_allowed to the |
1073 | * effective cpuset's. As this function is called with cpuset_mutex held, | |
1074 | * cpuset membership stays stable. | |
0b2f630a | 1075 | */ |
d66393e5 | 1076 | static void update_tasks_nodemask(struct cpuset *cs) |
1da177e4 | 1077 | { |
33ad801d | 1078 | static nodemask_t newmems; /* protected by cpuset_mutex */ |
d66393e5 TH |
1079 | struct css_task_iter it; |
1080 | struct task_struct *task; | |
59dac16f | 1081 | |
846a16bf | 1082 | cpuset_being_rebound = cs; /* causes mpol_dup() rebind */ |
4225399a | 1083 | |
ae1c8023 | 1084 | guarantee_online_mems(cs, &newmems); |
33ad801d | 1085 | |
4225399a | 1086 | /* |
3b6766fe LZ |
1087 | * The mpol_rebind_mm() call takes mmap_sem, which we couldn't |
1088 | * take while holding tasklist_lock. Forks can happen - the | |
1089 | * mpol_dup() cpuset_being_rebound check will catch such forks, | |
1090 | * and rebind their vma mempolicies too. Because we still hold | |
5d21cc2d | 1091 | * the global cpuset_mutex, we know that no other rebind effort |
3b6766fe | 1092 | * will be contending for the global variable cpuset_being_rebound. |
4225399a | 1093 | * It's ok if we rebind the same mm twice; mpol_rebind_mm() |
04c19fa6 | 1094 | * is idempotent. Also migrate pages in each mm to new nodes. |
4225399a | 1095 | */ |
d66393e5 TH |
1096 | css_task_iter_start(&cs->css, &it); |
1097 | while ((task = css_task_iter_next(&it))) { | |
1098 | struct mm_struct *mm; | |
1099 | bool migrate; | |
1100 | ||
1101 | cpuset_change_task_nodemask(task, &newmems); | |
1102 | ||
1103 | mm = get_task_mm(task); | |
1104 | if (!mm) | |
1105 | continue; | |
1106 | ||
1107 | migrate = is_memory_migrate(cs); | |
1108 | ||
1109 | mpol_rebind_mm(mm, &cs->mems_allowed); | |
1110 | if (migrate) | |
1111 | cpuset_migrate_mm(mm, &cs->old_mems_allowed, &newmems); | |
e93ad19d TH |
1112 | else |
1113 | mmput(mm); | |
d66393e5 TH |
1114 | } |
1115 | css_task_iter_end(&it); | |
4225399a | 1116 | |
33ad801d LZ |
1117 | /* |
1118 | * All the tasks' nodemasks have been updated, update | |
1119 | * cs->old_mems_allowed. | |
1120 | */ | |
1121 | cs->old_mems_allowed = newmems; | |
1122 | ||
2df167a3 | 1123 | /* We're done rebinding vmas to this cpuset's new mems_allowed. */ |
8793d854 | 1124 | cpuset_being_rebound = NULL; |
1da177e4 LT |
1125 | } |
1126 | ||
5c5cc623 | 1127 | /* |
734d4513 LZ |
1128 | * update_nodemasks_hier - Update effective nodemasks and tasks in the subtree |
1129 | * @cs: the cpuset to consider | |
1130 | * @new_mems: a temp variable for calculating new effective_mems | |
5c5cc623 | 1131 | * |
734d4513 LZ |
1132 | * When configured nodemask is changed, the effective nodemasks of this cpuset |
1133 | * and all its descendants need to be updated. | |
5c5cc623 | 1134 | * |
734d4513 | 1135 | * On legacy hiearchy, effective_mems will be the same with mems_allowed. |
5c5cc623 LZ |
1136 | * |
1137 | * Called with cpuset_mutex held | |
1138 | */ | |
734d4513 | 1139 | static void update_nodemasks_hier(struct cpuset *cs, nodemask_t *new_mems) |
5c5cc623 LZ |
1140 | { |
1141 | struct cpuset *cp; | |
492eb21b | 1142 | struct cgroup_subsys_state *pos_css; |
5c5cc623 LZ |
1143 | |
1144 | rcu_read_lock(); | |
734d4513 LZ |
1145 | cpuset_for_each_descendant_pre(cp, pos_css, cs) { |
1146 | struct cpuset *parent = parent_cs(cp); | |
1147 | ||
1148 | nodes_and(*new_mems, cp->mems_allowed, parent->effective_mems); | |
1149 | ||
554b0d1c LZ |
1150 | /* |
1151 | * If it becomes empty, inherit the effective mask of the | |
1152 | * parent, which is guaranteed to have some MEMs. | |
1153 | */ | |
9e10a130 TH |
1154 | if (cgroup_subsys_on_dfl(cpuset_cgrp_subsys) && |
1155 | nodes_empty(*new_mems)) | |
554b0d1c LZ |
1156 | *new_mems = parent->effective_mems; |
1157 | ||
734d4513 LZ |
1158 | /* Skip the whole subtree if the nodemask remains the same. */ |
1159 | if (nodes_equal(*new_mems, cp->effective_mems)) { | |
1160 | pos_css = css_rightmost_descendant(pos_css); | |
1161 | continue; | |
5c5cc623 | 1162 | } |
734d4513 | 1163 | |
ec903c0c | 1164 | if (!css_tryget_online(&cp->css)) |
5c5cc623 LZ |
1165 | continue; |
1166 | rcu_read_unlock(); | |
1167 | ||
8447a0fe | 1168 | spin_lock_irq(&callback_lock); |
734d4513 | 1169 | cp->effective_mems = *new_mems; |
8447a0fe | 1170 | spin_unlock_irq(&callback_lock); |
734d4513 | 1171 | |
9e10a130 | 1172 | WARN_ON(!cgroup_subsys_on_dfl(cpuset_cgrp_subsys) && |
a1381268 | 1173 | !nodes_equal(cp->mems_allowed, cp->effective_mems)); |
734d4513 | 1174 | |
d66393e5 | 1175 | update_tasks_nodemask(cp); |
5c5cc623 LZ |
1176 | |
1177 | rcu_read_lock(); | |
1178 | css_put(&cp->css); | |
1179 | } | |
1180 | rcu_read_unlock(); | |
1181 | } | |
1182 | ||
0b2f630a MX |
1183 | /* |
1184 | * Handle user request to change the 'mems' memory placement | |
1185 | * of a cpuset. Needs to validate the request, update the | |
58568d2a MX |
1186 | * cpusets mems_allowed, and for each task in the cpuset, |
1187 | * update mems_allowed and rebind task's mempolicy and any vma | |
1188 | * mempolicies and if the cpuset is marked 'memory_migrate', | |
1189 | * migrate the tasks pages to the new memory. | |
0b2f630a | 1190 | * |
8447a0fe | 1191 | * Call with cpuset_mutex held. May take callback_lock during call. |
0b2f630a MX |
1192 | * Will take tasklist_lock, scan tasklist for tasks in cpuset cs, |
1193 | * lock each such tasks mm->mmap_sem, scan its vma's and rebind | |
1194 | * their mempolicies to the cpusets new mems_allowed. | |
1195 | */ | |
645fcc9d LZ |
1196 | static int update_nodemask(struct cpuset *cs, struct cpuset *trialcs, |
1197 | const char *buf) | |
0b2f630a | 1198 | { |
0b2f630a MX |
1199 | int retval; |
1200 | ||
1201 | /* | |
38d7bee9 | 1202 | * top_cpuset.mems_allowed tracks node_stats[N_MEMORY]; |
0b2f630a MX |
1203 | * it's read-only |
1204 | */ | |
53feb297 MX |
1205 | if (cs == &top_cpuset) { |
1206 | retval = -EACCES; | |
1207 | goto done; | |
1208 | } | |
0b2f630a | 1209 | |
0b2f630a MX |
1210 | /* |
1211 | * An empty mems_allowed is ok iff there are no tasks in the cpuset. | |
1212 | * Since nodelist_parse() fails on an empty mask, we special case | |
1213 | * that parsing. The validate_change() call ensures that cpusets | |
1214 | * with tasks have memory. | |
1215 | */ | |
1216 | if (!*buf) { | |
645fcc9d | 1217 | nodes_clear(trialcs->mems_allowed); |
0b2f630a | 1218 | } else { |
645fcc9d | 1219 | retval = nodelist_parse(buf, trialcs->mems_allowed); |
0b2f630a MX |
1220 | if (retval < 0) |
1221 | goto done; | |
1222 | ||
645fcc9d | 1223 | if (!nodes_subset(trialcs->mems_allowed, |
5d8ba82c LZ |
1224 | top_cpuset.mems_allowed)) { |
1225 | retval = -EINVAL; | |
53feb297 MX |
1226 | goto done; |
1227 | } | |
0b2f630a | 1228 | } |
33ad801d LZ |
1229 | |
1230 | if (nodes_equal(cs->mems_allowed, trialcs->mems_allowed)) { | |
0b2f630a MX |
1231 | retval = 0; /* Too easy - nothing to do */ |
1232 | goto done; | |
1233 | } | |
645fcc9d | 1234 | retval = validate_change(cs, trialcs); |
0b2f630a MX |
1235 | if (retval < 0) |
1236 | goto done; | |
1237 | ||
8447a0fe | 1238 | spin_lock_irq(&callback_lock); |
645fcc9d | 1239 | cs->mems_allowed = trialcs->mems_allowed; |
8447a0fe | 1240 | spin_unlock_irq(&callback_lock); |
0b2f630a | 1241 | |
734d4513 | 1242 | /* use trialcs->mems_allowed as a temp variable */ |
24ee3cf8 | 1243 | update_nodemasks_hier(cs, &trialcs->mems_allowed); |
0b2f630a MX |
1244 | done: |
1245 | return retval; | |
1246 | } | |
1247 | ||
8793d854 PM |
1248 | int current_cpuset_is_being_rebound(void) |
1249 | { | |
391acf97 GZ |
1250 | int ret; |
1251 | ||
1252 | rcu_read_lock(); | |
1253 | ret = task_cs(current) == cpuset_being_rebound; | |
1254 | rcu_read_unlock(); | |
1255 | ||
1256 | return ret; | |
8793d854 PM |
1257 | } |
1258 | ||
5be7a479 | 1259 | static int update_relax_domain_level(struct cpuset *cs, s64 val) |
1d3504fc | 1260 | { |
db7f47cf | 1261 | #ifdef CONFIG_SMP |
60495e77 | 1262 | if (val < -1 || val >= sched_domain_level_max) |
30e0e178 | 1263 | return -EINVAL; |
db7f47cf | 1264 | #endif |
1d3504fc HS |
1265 | |
1266 | if (val != cs->relax_domain_level) { | |
1267 | cs->relax_domain_level = val; | |
300ed6cb LZ |
1268 | if (!cpumask_empty(cs->cpus_allowed) && |
1269 | is_sched_load_balance(cs)) | |
699140ba | 1270 | rebuild_sched_domains_locked(); |
1d3504fc HS |
1271 | } |
1272 | ||
1273 | return 0; | |
1274 | } | |
1275 | ||
72ec7029 | 1276 | /** |
950592f7 MX |
1277 | * update_tasks_flags - update the spread flags of tasks in the cpuset. |
1278 | * @cs: the cpuset in which each task's spread flags needs to be changed | |
950592f7 | 1279 | * |
d66393e5 TH |
1280 | * Iterate through each task of @cs updating its spread flags. As this |
1281 | * function is called with cpuset_mutex held, cpuset membership stays | |
1282 | * stable. | |
950592f7 | 1283 | */ |
d66393e5 | 1284 | static void update_tasks_flags(struct cpuset *cs) |
950592f7 | 1285 | { |
d66393e5 TH |
1286 | struct css_task_iter it; |
1287 | struct task_struct *task; | |
1288 | ||
1289 | css_task_iter_start(&cs->css, &it); | |
1290 | while ((task = css_task_iter_next(&it))) | |
1291 | cpuset_update_task_spread_flag(cs, task); | |
1292 | css_task_iter_end(&it); | |
950592f7 MX |
1293 | } |
1294 | ||
1da177e4 LT |
1295 | /* |
1296 | * update_flag - read a 0 or a 1 in a file and update associated flag | |
78608366 PM |
1297 | * bit: the bit to update (see cpuset_flagbits_t) |
1298 | * cs: the cpuset to update | |
1299 | * turning_on: whether the flag is being set or cleared | |
053199ed | 1300 | * |
5d21cc2d | 1301 | * Call with cpuset_mutex held. |
1da177e4 LT |
1302 | */ |
1303 | ||
700fe1ab PM |
1304 | static int update_flag(cpuset_flagbits_t bit, struct cpuset *cs, |
1305 | int turning_on) | |
1da177e4 | 1306 | { |
645fcc9d | 1307 | struct cpuset *trialcs; |
40b6a762 | 1308 | int balance_flag_changed; |
950592f7 | 1309 | int spread_flag_changed; |
950592f7 | 1310 | int err; |
1da177e4 | 1311 | |
645fcc9d LZ |
1312 | trialcs = alloc_trial_cpuset(cs); |
1313 | if (!trialcs) | |
1314 | return -ENOMEM; | |
1315 | ||
1da177e4 | 1316 | if (turning_on) |
645fcc9d | 1317 | set_bit(bit, &trialcs->flags); |
1da177e4 | 1318 | else |
645fcc9d | 1319 | clear_bit(bit, &trialcs->flags); |
1da177e4 | 1320 | |
645fcc9d | 1321 | err = validate_change(cs, trialcs); |
85d7b949 | 1322 | if (err < 0) |
645fcc9d | 1323 | goto out; |
029190c5 | 1324 | |
029190c5 | 1325 | balance_flag_changed = (is_sched_load_balance(cs) != |
645fcc9d | 1326 | is_sched_load_balance(trialcs)); |
029190c5 | 1327 | |
950592f7 MX |
1328 | spread_flag_changed = ((is_spread_slab(cs) != is_spread_slab(trialcs)) |
1329 | || (is_spread_page(cs) != is_spread_page(trialcs))); | |
1330 | ||
8447a0fe | 1331 | spin_lock_irq(&callback_lock); |
645fcc9d | 1332 | cs->flags = trialcs->flags; |
8447a0fe | 1333 | spin_unlock_irq(&callback_lock); |
85d7b949 | 1334 | |
300ed6cb | 1335 | if (!cpumask_empty(trialcs->cpus_allowed) && balance_flag_changed) |
699140ba | 1336 | rebuild_sched_domains_locked(); |
029190c5 | 1337 | |
950592f7 | 1338 | if (spread_flag_changed) |
d66393e5 | 1339 | update_tasks_flags(cs); |
645fcc9d LZ |
1340 | out: |
1341 | free_trial_cpuset(trialcs); | |
1342 | return err; | |
1da177e4 LT |
1343 | } |
1344 | ||
3e0d98b9 | 1345 | /* |
80f7228b | 1346 | * Frequency meter - How fast is some event occurring? |
3e0d98b9 PJ |
1347 | * |
1348 | * These routines manage a digitally filtered, constant time based, | |
1349 | * event frequency meter. There are four routines: | |
1350 | * fmeter_init() - initialize a frequency meter. | |
1351 | * fmeter_markevent() - called each time the event happens. | |
1352 | * fmeter_getrate() - returns the recent rate of such events. | |
1353 | * fmeter_update() - internal routine used to update fmeter. | |
1354 | * | |
1355 | * A common data structure is passed to each of these routines, | |
1356 | * which is used to keep track of the state required to manage the | |
1357 | * frequency meter and its digital filter. | |
1358 | * | |
1359 | * The filter works on the number of events marked per unit time. | |
1360 | * The filter is single-pole low-pass recursive (IIR). The time unit | |
1361 | * is 1 second. Arithmetic is done using 32-bit integers scaled to | |
1362 | * simulate 3 decimal digits of precision (multiplied by 1000). | |
1363 | * | |
1364 | * With an FM_COEF of 933, and a time base of 1 second, the filter | |
1365 | * has a half-life of 10 seconds, meaning that if the events quit | |
1366 | * happening, then the rate returned from the fmeter_getrate() | |
1367 | * will be cut in half each 10 seconds, until it converges to zero. | |
1368 | * | |
1369 | * It is not worth doing a real infinitely recursive filter. If more | |
1370 | * than FM_MAXTICKS ticks have elapsed since the last filter event, | |
1371 | * just compute FM_MAXTICKS ticks worth, by which point the level | |
1372 | * will be stable. | |
1373 | * | |
1374 | * Limit the count of unprocessed events to FM_MAXCNT, so as to avoid | |
1375 | * arithmetic overflow in the fmeter_update() routine. | |
1376 | * | |
1377 | * Given the simple 32 bit integer arithmetic used, this meter works | |
1378 | * best for reporting rates between one per millisecond (msec) and | |
1379 | * one per 32 (approx) seconds. At constant rates faster than one | |
1380 | * per msec it maxes out at values just under 1,000,000. At constant | |
1381 | * rates between one per msec, and one per second it will stabilize | |
1382 | * to a value N*1000, where N is the rate of events per second. | |
1383 | * At constant rates between one per second and one per 32 seconds, | |
1384 | * it will be choppy, moving up on the seconds that have an event, | |
1385 | * and then decaying until the next event. At rates slower than | |
1386 | * about one in 32 seconds, it decays all the way back to zero between | |
1387 | * each event. | |
1388 | */ | |
1389 | ||
1390 | #define FM_COEF 933 /* coefficient for half-life of 10 secs */ | |
d2b43658 | 1391 | #define FM_MAXTICKS ((u32)99) /* useless computing more ticks than this */ |
3e0d98b9 PJ |
1392 | #define FM_MAXCNT 1000000 /* limit cnt to avoid overflow */ |
1393 | #define FM_SCALE 1000 /* faux fixed point scale */ | |
1394 | ||
1395 | /* Initialize a frequency meter */ | |
1396 | static void fmeter_init(struct fmeter *fmp) | |
1397 | { | |
1398 | fmp->cnt = 0; | |
1399 | fmp->val = 0; | |
1400 | fmp->time = 0; | |
1401 | spin_lock_init(&fmp->lock); | |
1402 | } | |
1403 | ||
1404 | /* Internal meter update - process cnt events and update value */ | |
1405 | static void fmeter_update(struct fmeter *fmp) | |
1406 | { | |
d2b43658 AB |
1407 | time64_t now; |
1408 | u32 ticks; | |
1409 | ||
1410 | now = ktime_get_seconds(); | |
1411 | ticks = now - fmp->time; | |
3e0d98b9 PJ |
1412 | |
1413 | if (ticks == 0) | |
1414 | return; | |
1415 | ||
1416 | ticks = min(FM_MAXTICKS, ticks); | |
1417 | while (ticks-- > 0) | |
1418 | fmp->val = (FM_COEF * fmp->val) / FM_SCALE; | |
1419 | fmp->time = now; | |
1420 | ||
1421 | fmp->val += ((FM_SCALE - FM_COEF) * fmp->cnt) / FM_SCALE; | |
1422 | fmp->cnt = 0; | |
1423 | } | |
1424 | ||
1425 | /* Process any previous ticks, then bump cnt by one (times scale). */ | |
1426 | static void fmeter_markevent(struct fmeter *fmp) | |
1427 | { | |
1428 | spin_lock(&fmp->lock); | |
1429 | fmeter_update(fmp); | |
1430 | fmp->cnt = min(FM_MAXCNT, fmp->cnt + FM_SCALE); | |
1431 | spin_unlock(&fmp->lock); | |
1432 | } | |
1433 | ||
1434 | /* Process any previous ticks, then return current value. */ | |
1435 | static int fmeter_getrate(struct fmeter *fmp) | |
1436 | { | |
1437 | int val; | |
1438 | ||
1439 | spin_lock(&fmp->lock); | |
1440 | fmeter_update(fmp); | |
1441 | val = fmp->val; | |
1442 | spin_unlock(&fmp->lock); | |
1443 | return val; | |
1444 | } | |
1445 | ||
57fce0a6 TH |
1446 | static struct cpuset *cpuset_attach_old_cs; |
1447 | ||
5d21cc2d | 1448 | /* Called by cgroups to determine if a cpuset is usable; cpuset_mutex held */ |
1f7dd3e5 | 1449 | static int cpuset_can_attach(struct cgroup_taskset *tset) |
f780bdb7 | 1450 | { |
1f7dd3e5 TH |
1451 | struct cgroup_subsys_state *css; |
1452 | struct cpuset *cs; | |
bb9d97b6 TH |
1453 | struct task_struct *task; |
1454 | int ret; | |
1da177e4 | 1455 | |
57fce0a6 | 1456 | /* used later by cpuset_attach() */ |
1f7dd3e5 TH |
1457 | cpuset_attach_old_cs = task_cs(cgroup_taskset_first(tset, &css)); |
1458 | cs = css_cs(css); | |
57fce0a6 | 1459 | |
5d21cc2d TH |
1460 | mutex_lock(&cpuset_mutex); |
1461 | ||
aa6ec29b | 1462 | /* allow moving tasks into an empty cpuset if on default hierarchy */ |
5d21cc2d | 1463 | ret = -ENOSPC; |
9e10a130 | 1464 | if (!cgroup_subsys_on_dfl(cpuset_cgrp_subsys) && |
88fa523b | 1465 | (cpumask_empty(cs->cpus_allowed) || nodes_empty(cs->mems_allowed))) |
5d21cc2d | 1466 | goto out_unlock; |
9985b0ba | 1467 | |
1f7dd3e5 | 1468 | cgroup_taskset_for_each(task, css, tset) { |
7f51412a JL |
1469 | ret = task_can_attach(task, cs->cpus_allowed); |
1470 | if (ret) | |
5d21cc2d TH |
1471 | goto out_unlock; |
1472 | ret = security_task_setscheduler(task); | |
1473 | if (ret) | |
1474 | goto out_unlock; | |
bb9d97b6 | 1475 | } |
f780bdb7 | 1476 | |
452477fa TH |
1477 | /* |
1478 | * Mark attach is in progress. This makes validate_change() fail | |
1479 | * changes which zero cpus/mems_allowed. | |
1480 | */ | |
1481 | cs->attach_in_progress++; | |
5d21cc2d TH |
1482 | ret = 0; |
1483 | out_unlock: | |
1484 | mutex_unlock(&cpuset_mutex); | |
1485 | return ret; | |
8793d854 | 1486 | } |
f780bdb7 | 1487 | |
1f7dd3e5 | 1488 | static void cpuset_cancel_attach(struct cgroup_taskset *tset) |
452477fa | 1489 | { |
1f7dd3e5 TH |
1490 | struct cgroup_subsys_state *css; |
1491 | struct cpuset *cs; | |
1492 | ||
1493 | cgroup_taskset_first(tset, &css); | |
1494 | cs = css_cs(css); | |
1495 | ||
5d21cc2d | 1496 | mutex_lock(&cpuset_mutex); |
eb95419b | 1497 | css_cs(css)->attach_in_progress--; |
5d21cc2d | 1498 | mutex_unlock(&cpuset_mutex); |
8793d854 | 1499 | } |
1da177e4 | 1500 | |
4e4c9a14 | 1501 | /* |
5d21cc2d | 1502 | * Protected by cpuset_mutex. cpus_attach is used only by cpuset_attach() |
4e4c9a14 TH |
1503 | * but we can't allocate it dynamically there. Define it global and |
1504 | * allocate from cpuset_init(). | |
1505 | */ | |
1506 | static cpumask_var_t cpus_attach; | |
1507 | ||
1f7dd3e5 | 1508 | static void cpuset_attach(struct cgroup_taskset *tset) |
8793d854 | 1509 | { |
67bd2c59 | 1510 | /* static buf protected by cpuset_mutex */ |
4e4c9a14 | 1511 | static nodemask_t cpuset_attach_nodemask_to; |
bb9d97b6 | 1512 | struct task_struct *task; |
4530eddb | 1513 | struct task_struct *leader; |
1f7dd3e5 TH |
1514 | struct cgroup_subsys_state *css; |
1515 | struct cpuset *cs; | |
57fce0a6 | 1516 | struct cpuset *oldcs = cpuset_attach_old_cs; |
22fb52dd | 1517 | |
1f7dd3e5 TH |
1518 | cgroup_taskset_first(tset, &css); |
1519 | cs = css_cs(css); | |
1520 | ||
5d21cc2d TH |
1521 | mutex_lock(&cpuset_mutex); |
1522 | ||
4e4c9a14 TH |
1523 | /* prepare for attach */ |
1524 | if (cs == &top_cpuset) | |
1525 | cpumask_copy(cpus_attach, cpu_possible_mask); | |
1526 | else | |
ae1c8023 | 1527 | guarantee_online_cpus(cs, cpus_attach); |
4e4c9a14 | 1528 | |
ae1c8023 | 1529 | guarantee_online_mems(cs, &cpuset_attach_nodemask_to); |
4e4c9a14 | 1530 | |
1f7dd3e5 | 1531 | cgroup_taskset_for_each(task, css, tset) { |
bb9d97b6 TH |
1532 | /* |
1533 | * can_attach beforehand should guarantee that this doesn't | |
1534 | * fail. TODO: have a better way to handle failure here | |
1535 | */ | |
1536 | WARN_ON_ONCE(set_cpus_allowed_ptr(task, cpus_attach)); | |
1537 | ||
1538 | cpuset_change_task_nodemask(task, &cpuset_attach_nodemask_to); | |
1539 | cpuset_update_task_spread_flag(cs, task); | |
1540 | } | |
22fb52dd | 1541 | |
f780bdb7 | 1542 | /* |
4530eddb TH |
1543 | * Change mm for all threadgroup leaders. This is expensive and may |
1544 | * sleep and should be moved outside migration path proper. | |
f780bdb7 | 1545 | */ |
ae1c8023 | 1546 | cpuset_attach_nodemask_to = cs->effective_mems; |
1f7dd3e5 | 1547 | cgroup_taskset_for_each_leader(leader, css, tset) { |
3df9ca0a TH |
1548 | struct mm_struct *mm = get_task_mm(leader); |
1549 | ||
1550 | if (mm) { | |
1551 | mpol_rebind_mm(mm, &cpuset_attach_nodemask_to); | |
1552 | ||
1553 | /* | |
1554 | * old_mems_allowed is the same with mems_allowed | |
1555 | * here, except if this task is being moved | |
1556 | * automatically due to hotplug. In that case | |
1557 | * @mems_allowed has been updated and is empty, so | |
1558 | * @old_mems_allowed is the right nodesets that we | |
1559 | * migrate mm from. | |
1560 | */ | |
e93ad19d | 1561 | if (is_memory_migrate(cs)) |
3df9ca0a TH |
1562 | cpuset_migrate_mm(mm, &oldcs->old_mems_allowed, |
1563 | &cpuset_attach_nodemask_to); | |
e93ad19d TH |
1564 | else |
1565 | mmput(mm); | |
f047cecf | 1566 | } |
4225399a | 1567 | } |
452477fa | 1568 | |
33ad801d | 1569 | cs->old_mems_allowed = cpuset_attach_nodemask_to; |
02bb5863 | 1570 | |
452477fa | 1571 | cs->attach_in_progress--; |
e44193d3 LZ |
1572 | if (!cs->attach_in_progress) |
1573 | wake_up(&cpuset_attach_wq); | |
5d21cc2d TH |
1574 | |
1575 | mutex_unlock(&cpuset_mutex); | |
1da177e4 LT |
1576 | } |
1577 | ||
1578 | /* The various types of files and directories in a cpuset file system */ | |
1579 | ||
1580 | typedef enum { | |
45b07ef3 | 1581 | FILE_MEMORY_MIGRATE, |
1da177e4 LT |
1582 | FILE_CPULIST, |
1583 | FILE_MEMLIST, | |
afd1a8b3 LZ |
1584 | FILE_EFFECTIVE_CPULIST, |
1585 | FILE_EFFECTIVE_MEMLIST, | |
1da177e4 LT |
1586 | FILE_CPU_EXCLUSIVE, |
1587 | FILE_MEM_EXCLUSIVE, | |
78608366 | 1588 | FILE_MEM_HARDWALL, |
029190c5 | 1589 | FILE_SCHED_LOAD_BALANCE, |
1d3504fc | 1590 | FILE_SCHED_RELAX_DOMAIN_LEVEL, |
3e0d98b9 PJ |
1591 | FILE_MEMORY_PRESSURE_ENABLED, |
1592 | FILE_MEMORY_PRESSURE, | |
825a46af PJ |
1593 | FILE_SPREAD_PAGE, |
1594 | FILE_SPREAD_SLAB, | |
1da177e4 LT |
1595 | } cpuset_filetype_t; |
1596 | ||
182446d0 TH |
1597 | static int cpuset_write_u64(struct cgroup_subsys_state *css, struct cftype *cft, |
1598 | u64 val) | |
700fe1ab | 1599 | { |
182446d0 | 1600 | struct cpuset *cs = css_cs(css); |
700fe1ab | 1601 | cpuset_filetype_t type = cft->private; |
a903f086 | 1602 | int retval = 0; |
700fe1ab | 1603 | |
5d21cc2d | 1604 | mutex_lock(&cpuset_mutex); |
a903f086 LZ |
1605 | if (!is_cpuset_online(cs)) { |
1606 | retval = -ENODEV; | |
5d21cc2d | 1607 | goto out_unlock; |
a903f086 | 1608 | } |
700fe1ab PM |
1609 | |
1610 | switch (type) { | |
1da177e4 | 1611 | case FILE_CPU_EXCLUSIVE: |
700fe1ab | 1612 | retval = update_flag(CS_CPU_EXCLUSIVE, cs, val); |
1da177e4 LT |
1613 | break; |
1614 | case FILE_MEM_EXCLUSIVE: | |
700fe1ab | 1615 | retval = update_flag(CS_MEM_EXCLUSIVE, cs, val); |
1da177e4 | 1616 | break; |
78608366 PM |
1617 | case FILE_MEM_HARDWALL: |
1618 | retval = update_flag(CS_MEM_HARDWALL, cs, val); | |
1619 | break; | |
029190c5 | 1620 | case FILE_SCHED_LOAD_BALANCE: |
700fe1ab | 1621 | retval = update_flag(CS_SCHED_LOAD_BALANCE, cs, val); |
1d3504fc | 1622 | break; |
45b07ef3 | 1623 | case FILE_MEMORY_MIGRATE: |
700fe1ab | 1624 | retval = update_flag(CS_MEMORY_MIGRATE, cs, val); |
45b07ef3 | 1625 | break; |
3e0d98b9 | 1626 | case FILE_MEMORY_PRESSURE_ENABLED: |
700fe1ab | 1627 | cpuset_memory_pressure_enabled = !!val; |
3e0d98b9 | 1628 | break; |
825a46af | 1629 | case FILE_SPREAD_PAGE: |
700fe1ab | 1630 | retval = update_flag(CS_SPREAD_PAGE, cs, val); |
825a46af PJ |
1631 | break; |
1632 | case FILE_SPREAD_SLAB: | |
700fe1ab | 1633 | retval = update_flag(CS_SPREAD_SLAB, cs, val); |
825a46af | 1634 | break; |
1da177e4 LT |
1635 | default: |
1636 | retval = -EINVAL; | |
700fe1ab | 1637 | break; |
1da177e4 | 1638 | } |
5d21cc2d TH |
1639 | out_unlock: |
1640 | mutex_unlock(&cpuset_mutex); | |
1da177e4 LT |
1641 | return retval; |
1642 | } | |
1643 | ||
182446d0 TH |
1644 | static int cpuset_write_s64(struct cgroup_subsys_state *css, struct cftype *cft, |
1645 | s64 val) | |
5be7a479 | 1646 | { |
182446d0 | 1647 | struct cpuset *cs = css_cs(css); |
5be7a479 | 1648 | cpuset_filetype_t type = cft->private; |
5d21cc2d | 1649 | int retval = -ENODEV; |
5be7a479 | 1650 | |
5d21cc2d TH |
1651 | mutex_lock(&cpuset_mutex); |
1652 | if (!is_cpuset_online(cs)) | |
1653 | goto out_unlock; | |
e3712395 | 1654 | |
5be7a479 PM |
1655 | switch (type) { |
1656 | case FILE_SCHED_RELAX_DOMAIN_LEVEL: | |
1657 | retval = update_relax_domain_level(cs, val); | |
1658 | break; | |
1659 | default: | |
1660 | retval = -EINVAL; | |
1661 | break; | |
1662 | } | |
5d21cc2d TH |
1663 | out_unlock: |
1664 | mutex_unlock(&cpuset_mutex); | |
5be7a479 PM |
1665 | return retval; |
1666 | } | |
1667 | ||
e3712395 PM |
1668 | /* |
1669 | * Common handling for a write to a "cpus" or "mems" file. | |
1670 | */ | |
451af504 TH |
1671 | static ssize_t cpuset_write_resmask(struct kernfs_open_file *of, |
1672 | char *buf, size_t nbytes, loff_t off) | |
e3712395 | 1673 | { |
451af504 | 1674 | struct cpuset *cs = css_cs(of_css(of)); |
645fcc9d | 1675 | struct cpuset *trialcs; |
5d21cc2d | 1676 | int retval = -ENODEV; |
e3712395 | 1677 | |
451af504 TH |
1678 | buf = strstrip(buf); |
1679 | ||
3a5a6d0c TH |
1680 | /* |
1681 | * CPU or memory hotunplug may leave @cs w/o any execution | |
1682 | * resources, in which case the hotplug code asynchronously updates | |
1683 | * configuration and transfers all tasks to the nearest ancestor | |
1684 | * which can execute. | |
1685 | * | |
1686 | * As writes to "cpus" or "mems" may restore @cs's execution | |
1687 | * resources, wait for the previously scheduled operations before | |
1688 | * proceeding, so that we don't end up keep removing tasks added | |
1689 | * after execution capability is restored. | |
76bb5ab8 TH |
1690 | * |
1691 | * cpuset_hotplug_work calls back into cgroup core via | |
1692 | * cgroup_transfer_tasks() and waiting for it from a cgroupfs | |
1693 | * operation like this one can lead to a deadlock through kernfs | |
1694 | * active_ref protection. Let's break the protection. Losing the | |
1695 | * protection is okay as we check whether @cs is online after | |
1696 | * grabbing cpuset_mutex anyway. This only happens on the legacy | |
1697 | * hierarchies. | |
3a5a6d0c | 1698 | */ |
76bb5ab8 TH |
1699 | css_get(&cs->css); |
1700 | kernfs_break_active_protection(of->kn); | |
3a5a6d0c TH |
1701 | flush_work(&cpuset_hotplug_work); |
1702 | ||
5d21cc2d TH |
1703 | mutex_lock(&cpuset_mutex); |
1704 | if (!is_cpuset_online(cs)) | |
1705 | goto out_unlock; | |
e3712395 | 1706 | |
645fcc9d | 1707 | trialcs = alloc_trial_cpuset(cs); |
b75f38d6 LZ |
1708 | if (!trialcs) { |
1709 | retval = -ENOMEM; | |
5d21cc2d | 1710 | goto out_unlock; |
b75f38d6 | 1711 | } |
645fcc9d | 1712 | |
451af504 | 1713 | switch (of_cft(of)->private) { |
e3712395 | 1714 | case FILE_CPULIST: |
645fcc9d | 1715 | retval = update_cpumask(cs, trialcs, buf); |
e3712395 PM |
1716 | break; |
1717 | case FILE_MEMLIST: | |
645fcc9d | 1718 | retval = update_nodemask(cs, trialcs, buf); |
e3712395 PM |
1719 | break; |
1720 | default: | |
1721 | retval = -EINVAL; | |
1722 | break; | |
1723 | } | |
645fcc9d LZ |
1724 | |
1725 | free_trial_cpuset(trialcs); | |
5d21cc2d TH |
1726 | out_unlock: |
1727 | mutex_unlock(&cpuset_mutex); | |
76bb5ab8 TH |
1728 | kernfs_unbreak_active_protection(of->kn); |
1729 | css_put(&cs->css); | |
e93ad19d | 1730 | flush_workqueue(cpuset_migrate_mm_wq); |
451af504 | 1731 | return retval ?: nbytes; |
e3712395 PM |
1732 | } |
1733 | ||
1da177e4 LT |
1734 | /* |
1735 | * These ascii lists should be read in a single call, by using a user | |
1736 | * buffer large enough to hold the entire map. If read in smaller | |
1737 | * chunks, there is no guarantee of atomicity. Since the display format | |
1738 | * used, list of ranges of sequential numbers, is variable length, | |
1739 | * and since these maps can change value dynamically, one could read | |
1740 | * gibberish by doing partial reads while a list was changing. | |
1da177e4 | 1741 | */ |
2da8ca82 | 1742 | static int cpuset_common_seq_show(struct seq_file *sf, void *v) |
1da177e4 | 1743 | { |
2da8ca82 TH |
1744 | struct cpuset *cs = css_cs(seq_css(sf)); |
1745 | cpuset_filetype_t type = seq_cft(sf)->private; | |
51ffe411 | 1746 | int ret = 0; |
1da177e4 | 1747 | |
8447a0fe | 1748 | spin_lock_irq(&callback_lock); |
1da177e4 LT |
1749 | |
1750 | switch (type) { | |
1751 | case FILE_CPULIST: | |
e8e6d97c | 1752 | seq_printf(sf, "%*pbl\n", cpumask_pr_args(cs->cpus_allowed)); |
1da177e4 LT |
1753 | break; |
1754 | case FILE_MEMLIST: | |
e8e6d97c | 1755 | seq_printf(sf, "%*pbl\n", nodemask_pr_args(&cs->mems_allowed)); |
1da177e4 | 1756 | break; |
afd1a8b3 | 1757 | case FILE_EFFECTIVE_CPULIST: |
e8e6d97c | 1758 | seq_printf(sf, "%*pbl\n", cpumask_pr_args(cs->effective_cpus)); |
afd1a8b3 LZ |
1759 | break; |
1760 | case FILE_EFFECTIVE_MEMLIST: | |
e8e6d97c | 1761 | seq_printf(sf, "%*pbl\n", nodemask_pr_args(&cs->effective_mems)); |
afd1a8b3 | 1762 | break; |
1da177e4 | 1763 | default: |
51ffe411 | 1764 | ret = -EINVAL; |
1da177e4 | 1765 | } |
1da177e4 | 1766 | |
8447a0fe | 1767 | spin_unlock_irq(&callback_lock); |
51ffe411 | 1768 | return ret; |
1da177e4 LT |
1769 | } |
1770 | ||
182446d0 | 1771 | static u64 cpuset_read_u64(struct cgroup_subsys_state *css, struct cftype *cft) |
700fe1ab | 1772 | { |
182446d0 | 1773 | struct cpuset *cs = css_cs(css); |
700fe1ab PM |
1774 | cpuset_filetype_t type = cft->private; |
1775 | switch (type) { | |
1776 | case FILE_CPU_EXCLUSIVE: | |
1777 | return is_cpu_exclusive(cs); | |
1778 | case FILE_MEM_EXCLUSIVE: | |
1779 | return is_mem_exclusive(cs); | |
78608366 PM |
1780 | case FILE_MEM_HARDWALL: |
1781 | return is_mem_hardwall(cs); | |
700fe1ab PM |
1782 | case FILE_SCHED_LOAD_BALANCE: |
1783 | return is_sched_load_balance(cs); | |
1784 | case FILE_MEMORY_MIGRATE: | |
1785 | return is_memory_migrate(cs); | |
1786 | case FILE_MEMORY_PRESSURE_ENABLED: | |
1787 | return cpuset_memory_pressure_enabled; | |
1788 | case FILE_MEMORY_PRESSURE: | |
1789 | return fmeter_getrate(&cs->fmeter); | |
1790 | case FILE_SPREAD_PAGE: | |
1791 | return is_spread_page(cs); | |
1792 | case FILE_SPREAD_SLAB: | |
1793 | return is_spread_slab(cs); | |
1794 | default: | |
1795 | BUG(); | |
1796 | } | |
cf417141 MK |
1797 | |
1798 | /* Unreachable but makes gcc happy */ | |
1799 | return 0; | |
700fe1ab | 1800 | } |
1da177e4 | 1801 | |
182446d0 | 1802 | static s64 cpuset_read_s64(struct cgroup_subsys_state *css, struct cftype *cft) |
5be7a479 | 1803 | { |
182446d0 | 1804 | struct cpuset *cs = css_cs(css); |
5be7a479 PM |
1805 | cpuset_filetype_t type = cft->private; |
1806 | switch (type) { | |
1807 | case FILE_SCHED_RELAX_DOMAIN_LEVEL: | |
1808 | return cs->relax_domain_level; | |
1809 | default: | |
1810 | BUG(); | |
1811 | } | |
cf417141 MK |
1812 | |
1813 | /* Unrechable but makes gcc happy */ | |
1814 | return 0; | |
5be7a479 PM |
1815 | } |
1816 | ||
1da177e4 LT |
1817 | |
1818 | /* | |
1819 | * for the common functions, 'private' gives the type of file | |
1820 | */ | |
1821 | ||
addf2c73 PM |
1822 | static struct cftype files[] = { |
1823 | { | |
1824 | .name = "cpus", | |
2da8ca82 | 1825 | .seq_show = cpuset_common_seq_show, |
451af504 | 1826 | .write = cpuset_write_resmask, |
e3712395 | 1827 | .max_write_len = (100U + 6 * NR_CPUS), |
addf2c73 PM |
1828 | .private = FILE_CPULIST, |
1829 | }, | |
1830 | ||
1831 | { | |
1832 | .name = "mems", | |
2da8ca82 | 1833 | .seq_show = cpuset_common_seq_show, |
451af504 | 1834 | .write = cpuset_write_resmask, |
e3712395 | 1835 | .max_write_len = (100U + 6 * MAX_NUMNODES), |
addf2c73 PM |
1836 | .private = FILE_MEMLIST, |
1837 | }, | |
1838 | ||
afd1a8b3 LZ |
1839 | { |
1840 | .name = "effective_cpus", | |
1841 | .seq_show = cpuset_common_seq_show, | |
1842 | .private = FILE_EFFECTIVE_CPULIST, | |
1843 | }, | |
1844 | ||
1845 | { | |
1846 | .name = "effective_mems", | |
1847 | .seq_show = cpuset_common_seq_show, | |
1848 | .private = FILE_EFFECTIVE_MEMLIST, | |
1849 | }, | |
1850 | ||
addf2c73 PM |
1851 | { |
1852 | .name = "cpu_exclusive", | |
1853 | .read_u64 = cpuset_read_u64, | |
1854 | .write_u64 = cpuset_write_u64, | |
1855 | .private = FILE_CPU_EXCLUSIVE, | |
1856 | }, | |
1857 | ||
1858 | { | |
1859 | .name = "mem_exclusive", | |
1860 | .read_u64 = cpuset_read_u64, | |
1861 | .write_u64 = cpuset_write_u64, | |
1862 | .private = FILE_MEM_EXCLUSIVE, | |
1863 | }, | |
1864 | ||
78608366 PM |
1865 | { |
1866 | .name = "mem_hardwall", | |
1867 | .read_u64 = cpuset_read_u64, | |
1868 | .write_u64 = cpuset_write_u64, | |
1869 | .private = FILE_MEM_HARDWALL, | |
1870 | }, | |
1871 | ||
addf2c73 PM |
1872 | { |
1873 | .name = "sched_load_balance", | |
1874 | .read_u64 = cpuset_read_u64, | |
1875 | .write_u64 = cpuset_write_u64, | |
1876 | .private = FILE_SCHED_LOAD_BALANCE, | |
1877 | }, | |
1878 | ||
1879 | { | |
1880 | .name = "sched_relax_domain_level", | |
5be7a479 PM |
1881 | .read_s64 = cpuset_read_s64, |
1882 | .write_s64 = cpuset_write_s64, | |
addf2c73 PM |
1883 | .private = FILE_SCHED_RELAX_DOMAIN_LEVEL, |
1884 | }, | |
1885 | ||
1886 | { | |
1887 | .name = "memory_migrate", | |
1888 | .read_u64 = cpuset_read_u64, | |
1889 | .write_u64 = cpuset_write_u64, | |
1890 | .private = FILE_MEMORY_MIGRATE, | |
1891 | }, | |
1892 | ||
1893 | { | |
1894 | .name = "memory_pressure", | |
1895 | .read_u64 = cpuset_read_u64, | |
addf2c73 PM |
1896 | }, |
1897 | ||
1898 | { | |
1899 | .name = "memory_spread_page", | |
1900 | .read_u64 = cpuset_read_u64, | |
1901 | .write_u64 = cpuset_write_u64, | |
1902 | .private = FILE_SPREAD_PAGE, | |
1903 | }, | |
1904 | ||
1905 | { | |
1906 | .name = "memory_spread_slab", | |
1907 | .read_u64 = cpuset_read_u64, | |
1908 | .write_u64 = cpuset_write_u64, | |
1909 | .private = FILE_SPREAD_SLAB, | |
1910 | }, | |
3e0d98b9 | 1911 | |
4baf6e33 TH |
1912 | { |
1913 | .name = "memory_pressure_enabled", | |
1914 | .flags = CFTYPE_ONLY_ON_ROOT, | |
1915 | .read_u64 = cpuset_read_u64, | |
1916 | .write_u64 = cpuset_write_u64, | |
1917 | .private = FILE_MEMORY_PRESSURE_ENABLED, | |
1918 | }, | |
1da177e4 | 1919 | |
4baf6e33 TH |
1920 | { } /* terminate */ |
1921 | }; | |
1da177e4 LT |
1922 | |
1923 | /* | |
92fb9748 | 1924 | * cpuset_css_alloc - allocate a cpuset css |
c9e5fe66 | 1925 | * cgrp: control group that the new cpuset will be part of |
1da177e4 LT |
1926 | */ |
1927 | ||
eb95419b TH |
1928 | static struct cgroup_subsys_state * |
1929 | cpuset_css_alloc(struct cgroup_subsys_state *parent_css) | |
1da177e4 | 1930 | { |
c8f699bb | 1931 | struct cpuset *cs; |
1da177e4 | 1932 | |
eb95419b | 1933 | if (!parent_css) |
8793d854 | 1934 | return &top_cpuset.css; |
033fa1c5 | 1935 | |
c8f699bb | 1936 | cs = kzalloc(sizeof(*cs), GFP_KERNEL); |
1da177e4 | 1937 | if (!cs) |
8793d854 | 1938 | return ERR_PTR(-ENOMEM); |
e2b9a3d7 LZ |
1939 | if (!alloc_cpumask_var(&cs->cpus_allowed, GFP_KERNEL)) |
1940 | goto free_cs; | |
1941 | if (!alloc_cpumask_var(&cs->effective_cpus, GFP_KERNEL)) | |
1942 | goto free_cpus; | |
1da177e4 | 1943 | |
029190c5 | 1944 | set_bit(CS_SCHED_LOAD_BALANCE, &cs->flags); |
300ed6cb | 1945 | cpumask_clear(cs->cpus_allowed); |
f9a86fcb | 1946 | nodes_clear(cs->mems_allowed); |
e2b9a3d7 LZ |
1947 | cpumask_clear(cs->effective_cpus); |
1948 | nodes_clear(cs->effective_mems); | |
3e0d98b9 | 1949 | fmeter_init(&cs->fmeter); |
1d3504fc | 1950 | cs->relax_domain_level = -1; |
1da177e4 | 1951 | |
c8f699bb | 1952 | return &cs->css; |
e2b9a3d7 LZ |
1953 | |
1954 | free_cpus: | |
1955 | free_cpumask_var(cs->cpus_allowed); | |
1956 | free_cs: | |
1957 | kfree(cs); | |
1958 | return ERR_PTR(-ENOMEM); | |
c8f699bb TH |
1959 | } |
1960 | ||
eb95419b | 1961 | static int cpuset_css_online(struct cgroup_subsys_state *css) |
c8f699bb | 1962 | { |
eb95419b | 1963 | struct cpuset *cs = css_cs(css); |
c431069f | 1964 | struct cpuset *parent = parent_cs(cs); |
ae8086ce | 1965 | struct cpuset *tmp_cs; |
492eb21b | 1966 | struct cgroup_subsys_state *pos_css; |
c8f699bb TH |
1967 | |
1968 | if (!parent) | |
1969 | return 0; | |
1970 | ||
5d21cc2d TH |
1971 | mutex_lock(&cpuset_mutex); |
1972 | ||
efeb77b2 | 1973 | set_bit(CS_ONLINE, &cs->flags); |
c8f699bb TH |
1974 | if (is_spread_page(parent)) |
1975 | set_bit(CS_SPREAD_PAGE, &cs->flags); | |
1976 | if (is_spread_slab(parent)) | |
1977 | set_bit(CS_SPREAD_SLAB, &cs->flags); | |
1da177e4 | 1978 | |
664eedde | 1979 | cpuset_inc(); |
033fa1c5 | 1980 | |
8447a0fe | 1981 | spin_lock_irq(&callback_lock); |
9e10a130 | 1982 | if (cgroup_subsys_on_dfl(cpuset_cgrp_subsys)) { |
e2b9a3d7 LZ |
1983 | cpumask_copy(cs->effective_cpus, parent->effective_cpus); |
1984 | cs->effective_mems = parent->effective_mems; | |
1985 | } | |
8447a0fe | 1986 | spin_unlock_irq(&callback_lock); |
e2b9a3d7 | 1987 | |
eb95419b | 1988 | if (!test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags)) |
5d21cc2d | 1989 | goto out_unlock; |
033fa1c5 TH |
1990 | |
1991 | /* | |
1992 | * Clone @parent's configuration if CGRP_CPUSET_CLONE_CHILDREN is | |
1993 | * set. This flag handling is implemented in cgroup core for | |
1994 | * histrical reasons - the flag may be specified during mount. | |
1995 | * | |
1996 | * Currently, if any sibling cpusets have exclusive cpus or mem, we | |
1997 | * refuse to clone the configuration - thereby refusing the task to | |
1998 | * be entered, and as a result refusing the sys_unshare() or | |
1999 | * clone() which initiated it. If this becomes a problem for some | |
2000 | * users who wish to allow that scenario, then this could be | |
2001 | * changed to grant parent->cpus_allowed-sibling_cpus_exclusive | |
2002 | * (and likewise for mems) to the new cgroup. | |
2003 | */ | |
ae8086ce | 2004 | rcu_read_lock(); |
492eb21b | 2005 | cpuset_for_each_child(tmp_cs, pos_css, parent) { |
ae8086ce TH |
2006 | if (is_mem_exclusive(tmp_cs) || is_cpu_exclusive(tmp_cs)) { |
2007 | rcu_read_unlock(); | |
5d21cc2d | 2008 | goto out_unlock; |
ae8086ce | 2009 | } |
033fa1c5 | 2010 | } |
ae8086ce | 2011 | rcu_read_unlock(); |
033fa1c5 | 2012 | |
8447a0fe | 2013 | spin_lock_irq(&callback_lock); |
033fa1c5 | 2014 | cs->mems_allowed = parent->mems_allowed; |
790317e1 | 2015 | cs->effective_mems = parent->mems_allowed; |
033fa1c5 | 2016 | cpumask_copy(cs->cpus_allowed, parent->cpus_allowed); |
790317e1 | 2017 | cpumask_copy(cs->effective_cpus, parent->cpus_allowed); |
cea74465 | 2018 | spin_unlock_irq(&callback_lock); |
5d21cc2d TH |
2019 | out_unlock: |
2020 | mutex_unlock(&cpuset_mutex); | |
c8f699bb TH |
2021 | return 0; |
2022 | } | |
2023 | ||
0b9e6965 ZH |
2024 | /* |
2025 | * If the cpuset being removed has its flag 'sched_load_balance' | |
2026 | * enabled, then simulate turning sched_load_balance off, which | |
2027 | * will call rebuild_sched_domains_locked(). | |
2028 | */ | |
2029 | ||
eb95419b | 2030 | static void cpuset_css_offline(struct cgroup_subsys_state *css) |
c8f699bb | 2031 | { |
eb95419b | 2032 | struct cpuset *cs = css_cs(css); |
c8f699bb | 2033 | |
5d21cc2d | 2034 | mutex_lock(&cpuset_mutex); |
c8f699bb TH |
2035 | |
2036 | if (is_sched_load_balance(cs)) | |
2037 | update_flag(CS_SCHED_LOAD_BALANCE, cs, 0); | |
2038 | ||
664eedde | 2039 | cpuset_dec(); |
efeb77b2 | 2040 | clear_bit(CS_ONLINE, &cs->flags); |
c8f699bb | 2041 | |
5d21cc2d | 2042 | mutex_unlock(&cpuset_mutex); |
1da177e4 LT |
2043 | } |
2044 | ||
eb95419b | 2045 | static void cpuset_css_free(struct cgroup_subsys_state *css) |
1da177e4 | 2046 | { |
eb95419b | 2047 | struct cpuset *cs = css_cs(css); |
1da177e4 | 2048 | |
e2b9a3d7 | 2049 | free_cpumask_var(cs->effective_cpus); |
300ed6cb | 2050 | free_cpumask_var(cs->cpus_allowed); |
8793d854 | 2051 | kfree(cs); |
1da177e4 LT |
2052 | } |
2053 | ||
39bd0d15 LZ |
2054 | static void cpuset_bind(struct cgroup_subsys_state *root_css) |
2055 | { | |
2056 | mutex_lock(&cpuset_mutex); | |
8447a0fe | 2057 | spin_lock_irq(&callback_lock); |
39bd0d15 | 2058 | |
9e10a130 | 2059 | if (cgroup_subsys_on_dfl(cpuset_cgrp_subsys)) { |
39bd0d15 LZ |
2060 | cpumask_copy(top_cpuset.cpus_allowed, cpu_possible_mask); |
2061 | top_cpuset.mems_allowed = node_possible_map; | |
2062 | } else { | |
2063 | cpumask_copy(top_cpuset.cpus_allowed, | |
2064 | top_cpuset.effective_cpus); | |
2065 | top_cpuset.mems_allowed = top_cpuset.effective_mems; | |
2066 | } | |
2067 | ||
8447a0fe | 2068 | spin_unlock_irq(&callback_lock); |
39bd0d15 LZ |
2069 | mutex_unlock(&cpuset_mutex); |
2070 | } | |
2071 | ||
06f4e948 ZL |
2072 | /* |
2073 | * Make sure the new task conform to the current state of its parent, | |
2074 | * which could have been changed by cpuset just after it inherits the | |
2075 | * state from the parent and before it sits on the cgroup's task list. | |
2076 | */ | |
2077 | void cpuset_fork(struct task_struct *task) | |
2078 | { | |
2079 | if (task_css_is_root(task, cpuset_cgrp_id)) | |
2080 | return; | |
2081 | ||
2082 | set_cpus_allowed_ptr(task, ¤t->cpus_allowed); | |
2083 | task->mems_allowed = current->mems_allowed; | |
2084 | } | |
2085 | ||
073219e9 | 2086 | struct cgroup_subsys cpuset_cgrp_subsys = { |
39bd0d15 LZ |
2087 | .css_alloc = cpuset_css_alloc, |
2088 | .css_online = cpuset_css_online, | |
2089 | .css_offline = cpuset_css_offline, | |
2090 | .css_free = cpuset_css_free, | |
2091 | .can_attach = cpuset_can_attach, | |
2092 | .cancel_attach = cpuset_cancel_attach, | |
2093 | .attach = cpuset_attach, | |
5cf1cacb | 2094 | .post_attach = cpuset_post_attach, |
39bd0d15 | 2095 | .bind = cpuset_bind, |
06f4e948 | 2096 | .fork = cpuset_fork, |
5577964e | 2097 | .legacy_cftypes = files, |
b38e42e9 | 2098 | .early_init = true, |
8793d854 PM |
2099 | }; |
2100 | ||
1da177e4 LT |
2101 | /** |
2102 | * cpuset_init - initialize cpusets at system boot | |
2103 | * | |
2104 | * Description: Initialize top_cpuset and the cpuset internal file system, | |
2105 | **/ | |
2106 | ||
2107 | int __init cpuset_init(void) | |
2108 | { | |
8793d854 | 2109 | int err = 0; |
1da177e4 | 2110 | |
58568d2a MX |
2111 | if (!alloc_cpumask_var(&top_cpuset.cpus_allowed, GFP_KERNEL)) |
2112 | BUG(); | |
e2b9a3d7 LZ |
2113 | if (!alloc_cpumask_var(&top_cpuset.effective_cpus, GFP_KERNEL)) |
2114 | BUG(); | |
58568d2a | 2115 | |
300ed6cb | 2116 | cpumask_setall(top_cpuset.cpus_allowed); |
f9a86fcb | 2117 | nodes_setall(top_cpuset.mems_allowed); |
e2b9a3d7 LZ |
2118 | cpumask_setall(top_cpuset.effective_cpus); |
2119 | nodes_setall(top_cpuset.effective_mems); | |
1da177e4 | 2120 | |
3e0d98b9 | 2121 | fmeter_init(&top_cpuset.fmeter); |
029190c5 | 2122 | set_bit(CS_SCHED_LOAD_BALANCE, &top_cpuset.flags); |
1d3504fc | 2123 | top_cpuset.relax_domain_level = -1; |
1da177e4 | 2124 | |
1da177e4 LT |
2125 | err = register_filesystem(&cpuset_fs_type); |
2126 | if (err < 0) | |
8793d854 PM |
2127 | return err; |
2128 | ||
2341d1b6 LZ |
2129 | if (!alloc_cpumask_var(&cpus_attach, GFP_KERNEL)) |
2130 | BUG(); | |
2131 | ||
8793d854 | 2132 | return 0; |
1da177e4 LT |
2133 | } |
2134 | ||
b1aac8bb | 2135 | /* |
cf417141 | 2136 | * If CPU and/or memory hotplug handlers, below, unplug any CPUs |
b1aac8bb PJ |
2137 | * or memory nodes, we need to walk over the cpuset hierarchy, |
2138 | * removing that CPU or node from all cpusets. If this removes the | |
956db3ca CW |
2139 | * last CPU or node from a cpuset, then move the tasks in the empty |
2140 | * cpuset to its next-highest non-empty parent. | |
b1aac8bb | 2141 | */ |
956db3ca CW |
2142 | static void remove_tasks_in_empty_cpuset(struct cpuset *cs) |
2143 | { | |
2144 | struct cpuset *parent; | |
2145 | ||
956db3ca CW |
2146 | /* |
2147 | * Find its next-highest non-empty parent, (top cpuset | |
2148 | * has online cpus, so can't be empty). | |
2149 | */ | |
c431069f | 2150 | parent = parent_cs(cs); |
300ed6cb | 2151 | while (cpumask_empty(parent->cpus_allowed) || |
b4501295 | 2152 | nodes_empty(parent->mems_allowed)) |
c431069f | 2153 | parent = parent_cs(parent); |
956db3ca | 2154 | |
8cc99345 | 2155 | if (cgroup_transfer_tasks(parent->css.cgroup, cs->css.cgroup)) { |
12d3089c | 2156 | pr_err("cpuset: failed to transfer tasks out of empty cpuset "); |
e61734c5 TH |
2157 | pr_cont_cgroup_name(cs->css.cgroup); |
2158 | pr_cont("\n"); | |
8cc99345 | 2159 | } |
956db3ca CW |
2160 | } |
2161 | ||
be4c9dd7 LZ |
2162 | static void |
2163 | hotplug_update_tasks_legacy(struct cpuset *cs, | |
2164 | struct cpumask *new_cpus, nodemask_t *new_mems, | |
2165 | bool cpus_updated, bool mems_updated) | |
390a36aa LZ |
2166 | { |
2167 | bool is_empty; | |
2168 | ||
8447a0fe | 2169 | spin_lock_irq(&callback_lock); |
be4c9dd7 LZ |
2170 | cpumask_copy(cs->cpus_allowed, new_cpus); |
2171 | cpumask_copy(cs->effective_cpus, new_cpus); | |
2172 | cs->mems_allowed = *new_mems; | |
2173 | cs->effective_mems = *new_mems; | |
8447a0fe | 2174 | spin_unlock_irq(&callback_lock); |
390a36aa LZ |
2175 | |
2176 | /* | |
2177 | * Don't call update_tasks_cpumask() if the cpuset becomes empty, | |
2178 | * as the tasks will be migratecd to an ancestor. | |
2179 | */ | |
be4c9dd7 | 2180 | if (cpus_updated && !cpumask_empty(cs->cpus_allowed)) |
390a36aa | 2181 | update_tasks_cpumask(cs); |
be4c9dd7 | 2182 | if (mems_updated && !nodes_empty(cs->mems_allowed)) |
390a36aa LZ |
2183 | update_tasks_nodemask(cs); |
2184 | ||
2185 | is_empty = cpumask_empty(cs->cpus_allowed) || | |
2186 | nodes_empty(cs->mems_allowed); | |
2187 | ||
2188 | mutex_unlock(&cpuset_mutex); | |
2189 | ||
2190 | /* | |
2191 | * Move tasks to the nearest ancestor with execution resources, | |
2192 | * This is full cgroup operation which will also call back into | |
2193 | * cpuset. Should be done outside any lock. | |
2194 | */ | |
2195 | if (is_empty) | |
2196 | remove_tasks_in_empty_cpuset(cs); | |
2197 | ||
2198 | mutex_lock(&cpuset_mutex); | |
2199 | } | |
2200 | ||
be4c9dd7 LZ |
2201 | static void |
2202 | hotplug_update_tasks(struct cpuset *cs, | |
2203 | struct cpumask *new_cpus, nodemask_t *new_mems, | |
2204 | bool cpus_updated, bool mems_updated) | |
390a36aa | 2205 | { |
be4c9dd7 LZ |
2206 | if (cpumask_empty(new_cpus)) |
2207 | cpumask_copy(new_cpus, parent_cs(cs)->effective_cpus); | |
2208 | if (nodes_empty(*new_mems)) | |
2209 | *new_mems = parent_cs(cs)->effective_mems; | |
2210 | ||
8447a0fe | 2211 | spin_lock_irq(&callback_lock); |
be4c9dd7 LZ |
2212 | cpumask_copy(cs->effective_cpus, new_cpus); |
2213 | cs->effective_mems = *new_mems; | |
8447a0fe | 2214 | spin_unlock_irq(&callback_lock); |
390a36aa | 2215 | |
be4c9dd7 | 2216 | if (cpus_updated) |
390a36aa | 2217 | update_tasks_cpumask(cs); |
be4c9dd7 | 2218 | if (mems_updated) |
390a36aa LZ |
2219 | update_tasks_nodemask(cs); |
2220 | } | |
2221 | ||
deb7aa30 | 2222 | /** |
388afd85 | 2223 | * cpuset_hotplug_update_tasks - update tasks in a cpuset for hotunplug |
deb7aa30 | 2224 | * @cs: cpuset in interest |
956db3ca | 2225 | * |
deb7aa30 TH |
2226 | * Compare @cs's cpu and mem masks against top_cpuset and if some have gone |
2227 | * offline, update @cs accordingly. If @cs ends up with no CPU or memory, | |
2228 | * all its tasks are moved to the nearest ancestor with both resources. | |
80d1fa64 | 2229 | */ |
388afd85 | 2230 | static void cpuset_hotplug_update_tasks(struct cpuset *cs) |
80d1fa64 | 2231 | { |
be4c9dd7 LZ |
2232 | static cpumask_t new_cpus; |
2233 | static nodemask_t new_mems; | |
2234 | bool cpus_updated; | |
2235 | bool mems_updated; | |
e44193d3 LZ |
2236 | retry: |
2237 | wait_event(cpuset_attach_wq, cs->attach_in_progress == 0); | |
80d1fa64 | 2238 | |
5d21cc2d | 2239 | mutex_lock(&cpuset_mutex); |
7ddf96b0 | 2240 | |
e44193d3 LZ |
2241 | /* |
2242 | * We have raced with task attaching. We wait until attaching | |
2243 | * is finished, so we won't attach a task to an empty cpuset. | |
2244 | */ | |
2245 | if (cs->attach_in_progress) { | |
2246 | mutex_unlock(&cpuset_mutex); | |
2247 | goto retry; | |
2248 | } | |
2249 | ||
be4c9dd7 LZ |
2250 | cpumask_and(&new_cpus, cs->cpus_allowed, parent_cs(cs)->effective_cpus); |
2251 | nodes_and(new_mems, cs->mems_allowed, parent_cs(cs)->effective_mems); | |
80d1fa64 | 2252 | |
be4c9dd7 LZ |
2253 | cpus_updated = !cpumask_equal(&new_cpus, cs->effective_cpus); |
2254 | mems_updated = !nodes_equal(new_mems, cs->effective_mems); | |
deb7aa30 | 2255 | |
9e10a130 | 2256 | if (cgroup_subsys_on_dfl(cpuset_cgrp_subsys)) |
be4c9dd7 LZ |
2257 | hotplug_update_tasks(cs, &new_cpus, &new_mems, |
2258 | cpus_updated, mems_updated); | |
390a36aa | 2259 | else |
be4c9dd7 LZ |
2260 | hotplug_update_tasks_legacy(cs, &new_cpus, &new_mems, |
2261 | cpus_updated, mems_updated); | |
8d033948 | 2262 | |
5d21cc2d | 2263 | mutex_unlock(&cpuset_mutex); |
b1aac8bb PJ |
2264 | } |
2265 | ||
deb7aa30 | 2266 | /** |
3a5a6d0c | 2267 | * cpuset_hotplug_workfn - handle CPU/memory hotunplug for a cpuset |
956db3ca | 2268 | * |
deb7aa30 TH |
2269 | * This function is called after either CPU or memory configuration has |
2270 | * changed and updates cpuset accordingly. The top_cpuset is always | |
2271 | * synchronized to cpu_active_mask and N_MEMORY, which is necessary in | |
2272 | * order to make cpusets transparent (of no affect) on systems that are | |
2273 | * actively using CPU hotplug but making no active use of cpusets. | |
956db3ca | 2274 | * |
deb7aa30 | 2275 | * Non-root cpusets are only affected by offlining. If any CPUs or memory |
388afd85 LZ |
2276 | * nodes have been taken down, cpuset_hotplug_update_tasks() is invoked on |
2277 | * all descendants. | |
956db3ca | 2278 | * |
deb7aa30 TH |
2279 | * Note that CPU offlining during suspend is ignored. We don't modify |
2280 | * cpusets across suspend/resume cycles at all. | |
956db3ca | 2281 | */ |
3a5a6d0c | 2282 | static void cpuset_hotplug_workfn(struct work_struct *work) |
b1aac8bb | 2283 | { |
5c5cc623 LZ |
2284 | static cpumask_t new_cpus; |
2285 | static nodemask_t new_mems; | |
deb7aa30 | 2286 | bool cpus_updated, mems_updated; |
9e10a130 | 2287 | bool on_dfl = cgroup_subsys_on_dfl(cpuset_cgrp_subsys); |
b1aac8bb | 2288 | |
5d21cc2d | 2289 | mutex_lock(&cpuset_mutex); |
956db3ca | 2290 | |
deb7aa30 TH |
2291 | /* fetch the available cpus/mems and find out which changed how */ |
2292 | cpumask_copy(&new_cpus, cpu_active_mask); | |
2293 | new_mems = node_states[N_MEMORY]; | |
7ddf96b0 | 2294 | |
7e88291b LZ |
2295 | cpus_updated = !cpumask_equal(top_cpuset.effective_cpus, &new_cpus); |
2296 | mems_updated = !nodes_equal(top_cpuset.effective_mems, new_mems); | |
7ddf96b0 | 2297 | |
deb7aa30 TH |
2298 | /* synchronize cpus_allowed to cpu_active_mask */ |
2299 | if (cpus_updated) { | |
8447a0fe | 2300 | spin_lock_irq(&callback_lock); |
7e88291b LZ |
2301 | if (!on_dfl) |
2302 | cpumask_copy(top_cpuset.cpus_allowed, &new_cpus); | |
1344ab9c | 2303 | cpumask_copy(top_cpuset.effective_cpus, &new_cpus); |
8447a0fe | 2304 | spin_unlock_irq(&callback_lock); |
deb7aa30 TH |
2305 | /* we don't mess with cpumasks of tasks in top_cpuset */ |
2306 | } | |
b4501295 | 2307 | |
deb7aa30 TH |
2308 | /* synchronize mems_allowed to N_MEMORY */ |
2309 | if (mems_updated) { | |
8447a0fe | 2310 | spin_lock_irq(&callback_lock); |
7e88291b LZ |
2311 | if (!on_dfl) |
2312 | top_cpuset.mems_allowed = new_mems; | |
1344ab9c | 2313 | top_cpuset.effective_mems = new_mems; |
8447a0fe | 2314 | spin_unlock_irq(&callback_lock); |
d66393e5 | 2315 | update_tasks_nodemask(&top_cpuset); |
deb7aa30 | 2316 | } |
b4501295 | 2317 | |
388afd85 LZ |
2318 | mutex_unlock(&cpuset_mutex); |
2319 | ||
5c5cc623 LZ |
2320 | /* if cpus or mems changed, we need to propagate to descendants */ |
2321 | if (cpus_updated || mems_updated) { | |
deb7aa30 | 2322 | struct cpuset *cs; |
492eb21b | 2323 | struct cgroup_subsys_state *pos_css; |
f9b4fb8d | 2324 | |
fc560a26 | 2325 | rcu_read_lock(); |
492eb21b | 2326 | cpuset_for_each_descendant_pre(cs, pos_css, &top_cpuset) { |
ec903c0c | 2327 | if (cs == &top_cpuset || !css_tryget_online(&cs->css)) |
388afd85 LZ |
2328 | continue; |
2329 | rcu_read_unlock(); | |
7ddf96b0 | 2330 | |
388afd85 | 2331 | cpuset_hotplug_update_tasks(cs); |
b4501295 | 2332 | |
388afd85 LZ |
2333 | rcu_read_lock(); |
2334 | css_put(&cs->css); | |
2335 | } | |
2336 | rcu_read_unlock(); | |
2337 | } | |
8d033948 | 2338 | |
deb7aa30 | 2339 | /* rebuild sched domains if cpus_allowed has changed */ |
e0e80a02 LZ |
2340 | if (cpus_updated) |
2341 | rebuild_sched_domains(); | |
b1aac8bb PJ |
2342 | } |
2343 | ||
7ddf96b0 | 2344 | void cpuset_update_active_cpus(bool cpu_online) |
4c4d50f7 | 2345 | { |
3a5a6d0c TH |
2346 | /* |
2347 | * We're inside cpu hotplug critical region which usually nests | |
2348 | * inside cgroup synchronization. Bounce actual hotplug processing | |
2349 | * to a work item to avoid reverse locking order. | |
2350 | * | |
2351 | * We still need to do partition_sched_domains() synchronously; | |
2352 | * otherwise, the scheduler will get confused and put tasks to the | |
2353 | * dead CPU. Fall back to the default single domain. | |
2354 | * cpuset_hotplug_workfn() will rebuild it as necessary. | |
2355 | */ | |
2356 | partition_sched_domains(1, NULL, NULL); | |
2357 | schedule_work(&cpuset_hotplug_work); | |
4c4d50f7 | 2358 | } |
4c4d50f7 | 2359 | |
38837fc7 | 2360 | /* |
38d7bee9 LJ |
2361 | * Keep top_cpuset.mems_allowed tracking node_states[N_MEMORY]. |
2362 | * Call this routine anytime after node_states[N_MEMORY] changes. | |
a1cd2b13 | 2363 | * See cpuset_update_active_cpus() for CPU hotplug handling. |
38837fc7 | 2364 | */ |
f481891f MX |
2365 | static int cpuset_track_online_nodes(struct notifier_block *self, |
2366 | unsigned long action, void *arg) | |
38837fc7 | 2367 | { |
3a5a6d0c | 2368 | schedule_work(&cpuset_hotplug_work); |
f481891f | 2369 | return NOTIFY_OK; |
38837fc7 | 2370 | } |
d8f10cb3 AM |
2371 | |
2372 | static struct notifier_block cpuset_track_online_nodes_nb = { | |
2373 | .notifier_call = cpuset_track_online_nodes, | |
2374 | .priority = 10, /* ??! */ | |
2375 | }; | |
38837fc7 | 2376 | |
1da177e4 LT |
2377 | /** |
2378 | * cpuset_init_smp - initialize cpus_allowed | |
2379 | * | |
2380 | * Description: Finish top cpuset after cpu, node maps are initialized | |
d8f10cb3 | 2381 | */ |
1da177e4 LT |
2382 | void __init cpuset_init_smp(void) |
2383 | { | |
6ad4c188 | 2384 | cpumask_copy(top_cpuset.cpus_allowed, cpu_active_mask); |
38d7bee9 | 2385 | top_cpuset.mems_allowed = node_states[N_MEMORY]; |
33ad801d | 2386 | top_cpuset.old_mems_allowed = top_cpuset.mems_allowed; |
4c4d50f7 | 2387 | |
e2b9a3d7 LZ |
2388 | cpumask_copy(top_cpuset.effective_cpus, cpu_active_mask); |
2389 | top_cpuset.effective_mems = node_states[N_MEMORY]; | |
2390 | ||
d8f10cb3 | 2391 | register_hotmemory_notifier(&cpuset_track_online_nodes_nb); |
e93ad19d TH |
2392 | |
2393 | cpuset_migrate_mm_wq = alloc_ordered_workqueue("cpuset_migrate_mm", 0); | |
2394 | BUG_ON(!cpuset_migrate_mm_wq); | |
1da177e4 LT |
2395 | } |
2396 | ||
2397 | /** | |
1da177e4 LT |
2398 | * cpuset_cpus_allowed - return cpus_allowed mask from a tasks cpuset. |
2399 | * @tsk: pointer to task_struct from which to obtain cpuset->cpus_allowed. | |
6af866af | 2400 | * @pmask: pointer to struct cpumask variable to receive cpus_allowed set. |
1da177e4 | 2401 | * |
300ed6cb | 2402 | * Description: Returns the cpumask_var_t cpus_allowed of the cpuset |
1da177e4 | 2403 | * attached to the specified @tsk. Guaranteed to return some non-empty |
5f054e31 | 2404 | * subset of cpu_online_mask, even if this means going outside the |
1da177e4 LT |
2405 | * tasks cpuset. |
2406 | **/ | |
2407 | ||
6af866af | 2408 | void cpuset_cpus_allowed(struct task_struct *tsk, struct cpumask *pmask) |
1da177e4 | 2409 | { |
8447a0fe VD |
2410 | unsigned long flags; |
2411 | ||
2412 | spin_lock_irqsave(&callback_lock, flags); | |
b8dadcb5 | 2413 | rcu_read_lock(); |
ae1c8023 | 2414 | guarantee_online_cpus(task_cs(tsk), pmask); |
b8dadcb5 | 2415 | rcu_read_unlock(); |
8447a0fe | 2416 | spin_unlock_irqrestore(&callback_lock, flags); |
1da177e4 LT |
2417 | } |
2418 | ||
2baab4e9 | 2419 | void cpuset_cpus_allowed_fallback(struct task_struct *tsk) |
9084bb82 | 2420 | { |
9084bb82 | 2421 | rcu_read_lock(); |
ae1c8023 | 2422 | do_set_cpus_allowed(tsk, task_cs(tsk)->effective_cpus); |
9084bb82 ON |
2423 | rcu_read_unlock(); |
2424 | ||
2425 | /* | |
2426 | * We own tsk->cpus_allowed, nobody can change it under us. | |
2427 | * | |
2428 | * But we used cs && cs->cpus_allowed lockless and thus can | |
2429 | * race with cgroup_attach_task() or update_cpumask() and get | |
2430 | * the wrong tsk->cpus_allowed. However, both cases imply the | |
2431 | * subsequent cpuset_change_cpumask()->set_cpus_allowed_ptr() | |
2432 | * which takes task_rq_lock(). | |
2433 | * | |
2434 | * If we are called after it dropped the lock we must see all | |
2435 | * changes in tsk_cs()->cpus_allowed. Otherwise we can temporary | |
2436 | * set any mask even if it is not right from task_cs() pov, | |
2437 | * the pending set_cpus_allowed_ptr() will fix things. | |
2baab4e9 PZ |
2438 | * |
2439 | * select_fallback_rq() will fix things ups and set cpu_possible_mask | |
2440 | * if required. | |
9084bb82 | 2441 | */ |
9084bb82 ON |
2442 | } |
2443 | ||
8f4ab07f | 2444 | void __init cpuset_init_current_mems_allowed(void) |
1da177e4 | 2445 | { |
f9a86fcb | 2446 | nodes_setall(current->mems_allowed); |
1da177e4 LT |
2447 | } |
2448 | ||
909d75a3 PJ |
2449 | /** |
2450 | * cpuset_mems_allowed - return mems_allowed mask from a tasks cpuset. | |
2451 | * @tsk: pointer to task_struct from which to obtain cpuset->mems_allowed. | |
2452 | * | |
2453 | * Description: Returns the nodemask_t mems_allowed of the cpuset | |
2454 | * attached to the specified @tsk. Guaranteed to return some non-empty | |
38d7bee9 | 2455 | * subset of node_states[N_MEMORY], even if this means going outside the |
909d75a3 PJ |
2456 | * tasks cpuset. |
2457 | **/ | |
2458 | ||
2459 | nodemask_t cpuset_mems_allowed(struct task_struct *tsk) | |
2460 | { | |
2461 | nodemask_t mask; | |
8447a0fe | 2462 | unsigned long flags; |
909d75a3 | 2463 | |
8447a0fe | 2464 | spin_lock_irqsave(&callback_lock, flags); |
b8dadcb5 | 2465 | rcu_read_lock(); |
ae1c8023 | 2466 | guarantee_online_mems(task_cs(tsk), &mask); |
b8dadcb5 | 2467 | rcu_read_unlock(); |
8447a0fe | 2468 | spin_unlock_irqrestore(&callback_lock, flags); |
909d75a3 PJ |
2469 | |
2470 | return mask; | |
2471 | } | |
2472 | ||
d9fd8a6d | 2473 | /** |
19770b32 MG |
2474 | * cpuset_nodemask_valid_mems_allowed - check nodemask vs. curremt mems_allowed |
2475 | * @nodemask: the nodemask to be checked | |
d9fd8a6d | 2476 | * |
19770b32 | 2477 | * Are any of the nodes in the nodemask allowed in current->mems_allowed? |
1da177e4 | 2478 | */ |
19770b32 | 2479 | int cpuset_nodemask_valid_mems_allowed(nodemask_t *nodemask) |
1da177e4 | 2480 | { |
19770b32 | 2481 | return nodes_intersects(*nodemask, current->mems_allowed); |
1da177e4 LT |
2482 | } |
2483 | ||
9bf2229f | 2484 | /* |
78608366 PM |
2485 | * nearest_hardwall_ancestor() - Returns the nearest mem_exclusive or |
2486 | * mem_hardwall ancestor to the specified cpuset. Call holding | |
8447a0fe | 2487 | * callback_lock. If no ancestor is mem_exclusive or mem_hardwall |
78608366 | 2488 | * (an unusual configuration), then returns the root cpuset. |
9bf2229f | 2489 | */ |
c9710d80 | 2490 | static struct cpuset *nearest_hardwall_ancestor(struct cpuset *cs) |
9bf2229f | 2491 | { |
c431069f TH |
2492 | while (!(is_mem_exclusive(cs) || is_mem_hardwall(cs)) && parent_cs(cs)) |
2493 | cs = parent_cs(cs); | |
9bf2229f PJ |
2494 | return cs; |
2495 | } | |
2496 | ||
d9fd8a6d | 2497 | /** |
344736f2 | 2498 | * cpuset_node_allowed - Can we allocate on a memory node? |
a1bc5a4e | 2499 | * @node: is this an allowed node? |
02a0e53d | 2500 | * @gfp_mask: memory allocation flags |
d9fd8a6d | 2501 | * |
6e276d2a DR |
2502 | * If we're in interrupt, yes, we can always allocate. If @node is set in |
2503 | * current's mems_allowed, yes. If it's not a __GFP_HARDWALL request and this | |
2504 | * node is set in the nearest hardwalled cpuset ancestor to current's cpuset, | |
2505 | * yes. If current has access to memory reserves due to TIF_MEMDIE, yes. | |
9bf2229f PJ |
2506 | * Otherwise, no. |
2507 | * | |
2508 | * GFP_USER allocations are marked with the __GFP_HARDWALL bit, | |
c596d9f3 DR |
2509 | * and do not allow allocations outside the current tasks cpuset |
2510 | * unless the task has been OOM killed as is marked TIF_MEMDIE. | |
9bf2229f | 2511 | * GFP_KERNEL allocations are not so marked, so can escape to the |
78608366 | 2512 | * nearest enclosing hardwalled ancestor cpuset. |
9bf2229f | 2513 | * |
8447a0fe | 2514 | * Scanning up parent cpusets requires callback_lock. The |
02a0e53d PJ |
2515 | * __alloc_pages() routine only calls here with __GFP_HARDWALL bit |
2516 | * _not_ set if it's a GFP_KERNEL allocation, and all nodes in the | |
2517 | * current tasks mems_allowed came up empty on the first pass over | |
2518 | * the zonelist. So only GFP_KERNEL allocations, if all nodes in the | |
8447a0fe | 2519 | * cpuset are short of memory, might require taking the callback_lock. |
9bf2229f | 2520 | * |
36be57ff | 2521 | * The first call here from mm/page_alloc:get_page_from_freelist() |
02a0e53d PJ |
2522 | * has __GFP_HARDWALL set in gfp_mask, enforcing hardwall cpusets, |
2523 | * so no allocation on a node outside the cpuset is allowed (unless | |
2524 | * in interrupt, of course). | |
36be57ff PJ |
2525 | * |
2526 | * The second pass through get_page_from_freelist() doesn't even call | |
2527 | * here for GFP_ATOMIC calls. For those calls, the __alloc_pages() | |
2528 | * variable 'wait' is not set, and the bit ALLOC_CPUSET is not set | |
2529 | * in alloc_flags. That logic and the checks below have the combined | |
2530 | * affect that: | |
9bf2229f PJ |
2531 | * in_interrupt - any node ok (current task context irrelevant) |
2532 | * GFP_ATOMIC - any node ok | |
c596d9f3 | 2533 | * TIF_MEMDIE - any node ok |
78608366 | 2534 | * GFP_KERNEL - any node in enclosing hardwalled cpuset ok |
9bf2229f | 2535 | * GFP_USER - only nodes in current tasks mems allowed ok. |
02a0e53d | 2536 | */ |
002f2906 | 2537 | bool __cpuset_node_allowed(int node, gfp_t gfp_mask) |
1da177e4 | 2538 | { |
c9710d80 | 2539 | struct cpuset *cs; /* current cpuset ancestors */ |
29afd49b | 2540 | int allowed; /* is allocation in zone z allowed? */ |
8447a0fe | 2541 | unsigned long flags; |
9bf2229f | 2542 | |
6e276d2a | 2543 | if (in_interrupt()) |
002f2906 | 2544 | return true; |
9bf2229f | 2545 | if (node_isset(node, current->mems_allowed)) |
002f2906 | 2546 | return true; |
c596d9f3 DR |
2547 | /* |
2548 | * Allow tasks that have access to memory reserves because they have | |
2549 | * been OOM killed to get memory anywhere. | |
2550 | */ | |
2551 | if (unlikely(test_thread_flag(TIF_MEMDIE))) | |
002f2906 | 2552 | return true; |
9bf2229f | 2553 | if (gfp_mask & __GFP_HARDWALL) /* If hardwall request, stop here */ |
002f2906 | 2554 | return false; |
9bf2229f | 2555 | |
5563e770 | 2556 | if (current->flags & PF_EXITING) /* Let dying task have memory */ |
002f2906 | 2557 | return true; |
5563e770 | 2558 | |
9bf2229f | 2559 | /* Not hardwall and node outside mems_allowed: scan up cpusets */ |
8447a0fe | 2560 | spin_lock_irqsave(&callback_lock, flags); |
053199ed | 2561 | |
b8dadcb5 | 2562 | rcu_read_lock(); |
78608366 | 2563 | cs = nearest_hardwall_ancestor(task_cs(current)); |
99afb0fd | 2564 | allowed = node_isset(node, cs->mems_allowed); |
b8dadcb5 | 2565 | rcu_read_unlock(); |
053199ed | 2566 | |
8447a0fe | 2567 | spin_unlock_irqrestore(&callback_lock, flags); |
9bf2229f | 2568 | return allowed; |
1da177e4 LT |
2569 | } |
2570 | ||
825a46af | 2571 | /** |
6adef3eb JS |
2572 | * cpuset_mem_spread_node() - On which node to begin search for a file page |
2573 | * cpuset_slab_spread_node() - On which node to begin search for a slab page | |
825a46af PJ |
2574 | * |
2575 | * If a task is marked PF_SPREAD_PAGE or PF_SPREAD_SLAB (as for | |
2576 | * tasks in a cpuset with is_spread_page or is_spread_slab set), | |
2577 | * and if the memory allocation used cpuset_mem_spread_node() | |
2578 | * to determine on which node to start looking, as it will for | |
2579 | * certain page cache or slab cache pages such as used for file | |
2580 | * system buffers and inode caches, then instead of starting on the | |
2581 | * local node to look for a free page, rather spread the starting | |
2582 | * node around the tasks mems_allowed nodes. | |
2583 | * | |
2584 | * We don't have to worry about the returned node being offline | |
2585 | * because "it can't happen", and even if it did, it would be ok. | |
2586 | * | |
2587 | * The routines calling guarantee_online_mems() are careful to | |
2588 | * only set nodes in task->mems_allowed that are online. So it | |
2589 | * should not be possible for the following code to return an | |
2590 | * offline node. But if it did, that would be ok, as this routine | |
2591 | * is not returning the node where the allocation must be, only | |
2592 | * the node where the search should start. The zonelist passed to | |
2593 | * __alloc_pages() will include all nodes. If the slab allocator | |
2594 | * is passed an offline node, it will fall back to the local node. | |
2595 | * See kmem_cache_alloc_node(). | |
2596 | */ | |
2597 | ||
6adef3eb | 2598 | static int cpuset_spread_node(int *rotor) |
825a46af | 2599 | { |
0edaf86c | 2600 | return *rotor = next_node_in(*rotor, current->mems_allowed); |
825a46af | 2601 | } |
6adef3eb JS |
2602 | |
2603 | int cpuset_mem_spread_node(void) | |
2604 | { | |
778d3b0f MH |
2605 | if (current->cpuset_mem_spread_rotor == NUMA_NO_NODE) |
2606 | current->cpuset_mem_spread_rotor = | |
2607 | node_random(¤t->mems_allowed); | |
2608 | ||
6adef3eb JS |
2609 | return cpuset_spread_node(¤t->cpuset_mem_spread_rotor); |
2610 | } | |
2611 | ||
2612 | int cpuset_slab_spread_node(void) | |
2613 | { | |
778d3b0f MH |
2614 | if (current->cpuset_slab_spread_rotor == NUMA_NO_NODE) |
2615 | current->cpuset_slab_spread_rotor = | |
2616 | node_random(¤t->mems_allowed); | |
2617 | ||
6adef3eb JS |
2618 | return cpuset_spread_node(¤t->cpuset_slab_spread_rotor); |
2619 | } | |
2620 | ||
825a46af PJ |
2621 | EXPORT_SYMBOL_GPL(cpuset_mem_spread_node); |
2622 | ||
ef08e3b4 | 2623 | /** |
bbe373f2 DR |
2624 | * cpuset_mems_allowed_intersects - Does @tsk1's mems_allowed intersect @tsk2's? |
2625 | * @tsk1: pointer to task_struct of some task. | |
2626 | * @tsk2: pointer to task_struct of some other task. | |
2627 | * | |
2628 | * Description: Return true if @tsk1's mems_allowed intersects the | |
2629 | * mems_allowed of @tsk2. Used by the OOM killer to determine if | |
2630 | * one of the task's memory usage might impact the memory available | |
2631 | * to the other. | |
ef08e3b4 PJ |
2632 | **/ |
2633 | ||
bbe373f2 DR |
2634 | int cpuset_mems_allowed_intersects(const struct task_struct *tsk1, |
2635 | const struct task_struct *tsk2) | |
ef08e3b4 | 2636 | { |
bbe373f2 | 2637 | return nodes_intersects(tsk1->mems_allowed, tsk2->mems_allowed); |
ef08e3b4 PJ |
2638 | } |
2639 | ||
75aa1994 | 2640 | /** |
da39da3a | 2641 | * cpuset_print_current_mems_allowed - prints current's cpuset and mems_allowed |
75aa1994 | 2642 | * |
da39da3a | 2643 | * Description: Prints current's name, cpuset name, and cached copy of its |
b8dadcb5 | 2644 | * mems_allowed to the kernel log. |
75aa1994 | 2645 | */ |
da39da3a | 2646 | void cpuset_print_current_mems_allowed(void) |
75aa1994 | 2647 | { |
b8dadcb5 | 2648 | struct cgroup *cgrp; |
75aa1994 | 2649 | |
b8dadcb5 | 2650 | rcu_read_lock(); |
63f43f55 | 2651 | |
da39da3a DR |
2652 | cgrp = task_cs(current)->css.cgroup; |
2653 | pr_info("%s cpuset=", current->comm); | |
e61734c5 | 2654 | pr_cont_cgroup_name(cgrp); |
da39da3a DR |
2655 | pr_cont(" mems_allowed=%*pbl\n", |
2656 | nodemask_pr_args(¤t->mems_allowed)); | |
f440d98f | 2657 | |
cfb5966b | 2658 | rcu_read_unlock(); |
75aa1994 DR |
2659 | } |
2660 | ||
3e0d98b9 PJ |
2661 | /* |
2662 | * Collection of memory_pressure is suppressed unless | |
2663 | * this flag is enabled by writing "1" to the special | |
2664 | * cpuset file 'memory_pressure_enabled' in the root cpuset. | |
2665 | */ | |
2666 | ||
c5b2aff8 | 2667 | int cpuset_memory_pressure_enabled __read_mostly; |
3e0d98b9 PJ |
2668 | |
2669 | /** | |
2670 | * cpuset_memory_pressure_bump - keep stats of per-cpuset reclaims. | |
2671 | * | |
2672 | * Keep a running average of the rate of synchronous (direct) | |
2673 | * page reclaim efforts initiated by tasks in each cpuset. | |
2674 | * | |
2675 | * This represents the rate at which some task in the cpuset | |
2676 | * ran low on memory on all nodes it was allowed to use, and | |
2677 | * had to enter the kernels page reclaim code in an effort to | |
2678 | * create more free memory by tossing clean pages or swapping | |
2679 | * or writing dirty pages. | |
2680 | * | |
2681 | * Display to user space in the per-cpuset read-only file | |
2682 | * "memory_pressure". Value displayed is an integer | |
2683 | * representing the recent rate of entry into the synchronous | |
2684 | * (direct) page reclaim by any task attached to the cpuset. | |
2685 | **/ | |
2686 | ||
2687 | void __cpuset_memory_pressure_bump(void) | |
2688 | { | |
b8dadcb5 | 2689 | rcu_read_lock(); |
8793d854 | 2690 | fmeter_markevent(&task_cs(current)->fmeter); |
b8dadcb5 | 2691 | rcu_read_unlock(); |
3e0d98b9 PJ |
2692 | } |
2693 | ||
8793d854 | 2694 | #ifdef CONFIG_PROC_PID_CPUSET |
1da177e4 LT |
2695 | /* |
2696 | * proc_cpuset_show() | |
2697 | * - Print tasks cpuset path into seq_file. | |
2698 | * - Used for /proc/<pid>/cpuset. | |
053199ed PJ |
2699 | * - No need to task_lock(tsk) on this tsk->cpuset reference, as it |
2700 | * doesn't really matter if tsk->cpuset changes after we read it, | |
5d21cc2d | 2701 | * and we take cpuset_mutex, keeping cpuset_attach() from changing it |
2df167a3 | 2702 | * anyway. |
1da177e4 | 2703 | */ |
52de4779 ZL |
2704 | int proc_cpuset_show(struct seq_file *m, struct pid_namespace *ns, |
2705 | struct pid *pid, struct task_struct *tsk) | |
1da177e4 | 2706 | { |
e61734c5 | 2707 | char *buf, *p; |
8793d854 | 2708 | struct cgroup_subsys_state *css; |
99f89551 | 2709 | int retval; |
1da177e4 | 2710 | |
99f89551 | 2711 | retval = -ENOMEM; |
e61734c5 | 2712 | buf = kmalloc(PATH_MAX, GFP_KERNEL); |
1da177e4 | 2713 | if (!buf) |
99f89551 EB |
2714 | goto out; |
2715 | ||
e61734c5 | 2716 | retval = -ENAMETOOLONG; |
a79a908f AK |
2717 | css = task_get_css(tsk, cpuset_cgrp_id); |
2718 | p = cgroup_path_ns(css->cgroup, buf, PATH_MAX, | |
2719 | current->nsproxy->cgroup_ns); | |
2720 | css_put(css); | |
e61734c5 | 2721 | if (!p) |
52de4779 | 2722 | goto out_free; |
e61734c5 | 2723 | seq_puts(m, p); |
1da177e4 | 2724 | seq_putc(m, '\n'); |
e61734c5 | 2725 | retval = 0; |
99f89551 | 2726 | out_free: |
1da177e4 | 2727 | kfree(buf); |
99f89551 | 2728 | out: |
1da177e4 LT |
2729 | return retval; |
2730 | } | |
8793d854 | 2731 | #endif /* CONFIG_PROC_PID_CPUSET */ |
1da177e4 | 2732 | |
d01d4827 | 2733 | /* Display task mems_allowed in /proc/<pid>/status file. */ |
df5f8314 EB |
2734 | void cpuset_task_status_allowed(struct seq_file *m, struct task_struct *task) |
2735 | { | |
e8e6d97c TH |
2736 | seq_printf(m, "Mems_allowed:\t%*pb\n", |
2737 | nodemask_pr_args(&task->mems_allowed)); | |
2738 | seq_printf(m, "Mems_allowed_list:\t%*pbl\n", | |
2739 | nodemask_pr_args(&task->mems_allowed)); | |
1da177e4 | 2740 | } |