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> | |
54 | #include <linux/backing-dev.h> | |
55 | #include <linux/sort.h> | |
56 | ||
57 | #include <asm/uaccess.h> | |
60063497 | 58 | #include <linux/atomic.h> |
3d3f26a7 | 59 | #include <linux/mutex.h> |
956db3ca CW |
60 | #include <linux/workqueue.h> |
61 | #include <linux/cgroup.h> | |
e44193d3 | 62 | #include <linux/wait.h> |
1da177e4 | 63 | |
664eedde | 64 | struct static_key cpusets_enabled_key __read_mostly = STATIC_KEY_INIT_FALSE; |
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 */ | |
71 | time_t time; /* clock (secs) when val computed */ | |
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 | /* |
5d21cc2d TH |
251 | * There are two global mutexes guarding cpuset structures - cpuset_mutex |
252 | * and callback_mutex. The latter may nest inside the former. We also | |
253 | * require taking task_lock() when dereferencing a task's cpuset pointer. | |
254 | * See "The task_lock() exception", at the end of this comment. | |
255 | * | |
256 | * A task must hold both mutexes to modify cpusets. If a task holds | |
257 | * cpuset_mutex, then it blocks others wanting that mutex, ensuring that it | |
258 | * is the only task able to also acquire callback_mutex and be able to | |
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 | |
262 | * callback routines can briefly acquire callback_mutex to query cpusets. | |
263 | * Once it is ready to make the changes, it takes callback_mutex, blocking | |
264 | * everyone else. | |
053199ed PJ |
265 | * |
266 | * Calls to the kernel memory allocator can not be made while holding | |
3d3f26a7 | 267 | * callback_mutex, as that would risk double tripping on callback_mutex |
053199ed PJ |
268 | * from one of the callbacks into the cpuset code from within |
269 | * __alloc_pages(). | |
270 | * | |
3d3f26a7 | 271 | * If a task is only holding callback_mutex, 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 | * |
3d3f26a7 | 278 | * The cpuset_common_file_read() handlers only hold callback_mutex 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); |
3d3f26a7 | 287 | static DEFINE_MUTEX(callback_mutex); |
4247bdc6 | 288 | |
3a5a6d0c TH |
289 | /* |
290 | * CPU / memory hotplug is handled asynchronously. | |
291 | */ | |
292 | static void cpuset_hotplug_workfn(struct work_struct *work); | |
3a5a6d0c TH |
293 | static DECLARE_WORK(cpuset_hotplug_work, cpuset_hotplug_workfn); |
294 | ||
e44193d3 LZ |
295 | static DECLARE_WAIT_QUEUE_HEAD(cpuset_attach_wq); |
296 | ||
cf417141 MK |
297 | /* |
298 | * This is ugly, but preserves the userspace API for existing cpuset | |
8793d854 | 299 | * users. If someone tries to mount the "cpuset" filesystem, we |
cf417141 MK |
300 | * silently switch it to mount "cgroup" instead |
301 | */ | |
f7e83571 AV |
302 | static struct dentry *cpuset_mount(struct file_system_type *fs_type, |
303 | int flags, const char *unused_dev_name, void *data) | |
1da177e4 | 304 | { |
8793d854 | 305 | struct file_system_type *cgroup_fs = get_fs_type("cgroup"); |
f7e83571 | 306 | struct dentry *ret = ERR_PTR(-ENODEV); |
8793d854 PM |
307 | if (cgroup_fs) { |
308 | char mountopts[] = | |
309 | "cpuset,noprefix," | |
310 | "release_agent=/sbin/cpuset_release_agent"; | |
f7e83571 AV |
311 | ret = cgroup_fs->mount(cgroup_fs, flags, |
312 | unused_dev_name, mountopts); | |
8793d854 PM |
313 | put_filesystem(cgroup_fs); |
314 | } | |
315 | return ret; | |
1da177e4 LT |
316 | } |
317 | ||
318 | static struct file_system_type cpuset_fs_type = { | |
319 | .name = "cpuset", | |
f7e83571 | 320 | .mount = cpuset_mount, |
1da177e4 LT |
321 | }; |
322 | ||
1da177e4 | 323 | /* |
300ed6cb | 324 | * Return in pmask the portion of a cpusets's cpus_allowed that |
1da177e4 | 325 | * are online. If none are online, walk up the cpuset hierarchy |
40df2deb LZ |
326 | * until we find one that does have some online cpus. The top |
327 | * cpuset always has some cpus online. | |
1da177e4 LT |
328 | * |
329 | * One way or another, we guarantee to return some non-empty subset | |
5f054e31 | 330 | * of cpu_online_mask. |
1da177e4 | 331 | * |
3d3f26a7 | 332 | * Call with callback_mutex held. |
1da177e4 | 333 | */ |
c9710d80 | 334 | static void guarantee_online_cpus(struct cpuset *cs, struct cpumask *pmask) |
1da177e4 | 335 | { |
ae1c8023 | 336 | while (!cpumask_intersects(cs->effective_cpus, cpu_online_mask)) |
c431069f | 337 | cs = parent_cs(cs); |
ae1c8023 | 338 | cpumask_and(pmask, cs->effective_cpus, cpu_online_mask); |
1da177e4 LT |
339 | } |
340 | ||
341 | /* | |
342 | * Return in *pmask the portion of a cpusets's mems_allowed that | |
0e1e7c7a CL |
343 | * are online, with memory. If none are online with memory, walk |
344 | * up the cpuset hierarchy until we find one that does have some | |
40df2deb | 345 | * online mems. The top cpuset always has some mems online. |
1da177e4 LT |
346 | * |
347 | * One way or another, we guarantee to return some non-empty subset | |
38d7bee9 | 348 | * of node_states[N_MEMORY]. |
1da177e4 | 349 | * |
3d3f26a7 | 350 | * Call with callback_mutex held. |
1da177e4 | 351 | */ |
c9710d80 | 352 | static void guarantee_online_mems(struct cpuset *cs, nodemask_t *pmask) |
1da177e4 | 353 | { |
ae1c8023 | 354 | while (!nodes_intersects(cs->effective_mems, node_states[N_MEMORY])) |
c431069f | 355 | cs = parent_cs(cs); |
ae1c8023 | 356 | nodes_and(*pmask, cs->effective_mems, node_states[N_MEMORY]); |
1da177e4 LT |
357 | } |
358 | ||
f3b39d47 MX |
359 | /* |
360 | * update task's spread flag if cpuset's page/slab spread flag is set | |
361 | * | |
5d21cc2d | 362 | * Called with callback_mutex/cpuset_mutex held |
f3b39d47 MX |
363 | */ |
364 | static void cpuset_update_task_spread_flag(struct cpuset *cs, | |
365 | struct task_struct *tsk) | |
366 | { | |
367 | if (is_spread_page(cs)) | |
2ad654bc | 368 | task_set_spread_page(tsk); |
f3b39d47 | 369 | else |
2ad654bc ZL |
370 | task_clear_spread_page(tsk); |
371 | ||
f3b39d47 | 372 | if (is_spread_slab(cs)) |
2ad654bc | 373 | task_set_spread_slab(tsk); |
f3b39d47 | 374 | else |
2ad654bc | 375 | task_clear_spread_slab(tsk); |
f3b39d47 MX |
376 | } |
377 | ||
1da177e4 LT |
378 | /* |
379 | * is_cpuset_subset(p, q) - Is cpuset p a subset of cpuset q? | |
380 | * | |
381 | * One cpuset is a subset of another if all its allowed CPUs and | |
382 | * Memory Nodes are a subset of the other, and its exclusive flags | |
5d21cc2d | 383 | * are only set if the other's are set. Call holding cpuset_mutex. |
1da177e4 LT |
384 | */ |
385 | ||
386 | static int is_cpuset_subset(const struct cpuset *p, const struct cpuset *q) | |
387 | { | |
300ed6cb | 388 | return cpumask_subset(p->cpus_allowed, q->cpus_allowed) && |
1da177e4 LT |
389 | nodes_subset(p->mems_allowed, q->mems_allowed) && |
390 | is_cpu_exclusive(p) <= is_cpu_exclusive(q) && | |
391 | is_mem_exclusive(p) <= is_mem_exclusive(q); | |
392 | } | |
393 | ||
645fcc9d LZ |
394 | /** |
395 | * alloc_trial_cpuset - allocate a trial cpuset | |
396 | * @cs: the cpuset that the trial cpuset duplicates | |
397 | */ | |
c9710d80 | 398 | static struct cpuset *alloc_trial_cpuset(struct cpuset *cs) |
645fcc9d | 399 | { |
300ed6cb LZ |
400 | struct cpuset *trial; |
401 | ||
402 | trial = kmemdup(cs, sizeof(*cs), GFP_KERNEL); | |
403 | if (!trial) | |
404 | return NULL; | |
405 | ||
e2b9a3d7 LZ |
406 | if (!alloc_cpumask_var(&trial->cpus_allowed, GFP_KERNEL)) |
407 | goto free_cs; | |
408 | if (!alloc_cpumask_var(&trial->effective_cpus, GFP_KERNEL)) | |
409 | goto free_cpus; | |
300ed6cb | 410 | |
e2b9a3d7 LZ |
411 | cpumask_copy(trial->cpus_allowed, cs->cpus_allowed); |
412 | cpumask_copy(trial->effective_cpus, cs->effective_cpus); | |
300ed6cb | 413 | return trial; |
e2b9a3d7 LZ |
414 | |
415 | free_cpus: | |
416 | free_cpumask_var(trial->cpus_allowed); | |
417 | free_cs: | |
418 | kfree(trial); | |
419 | return NULL; | |
645fcc9d LZ |
420 | } |
421 | ||
422 | /** | |
423 | * free_trial_cpuset - free the trial cpuset | |
424 | * @trial: the trial cpuset to be freed | |
425 | */ | |
426 | static void free_trial_cpuset(struct cpuset *trial) | |
427 | { | |
e2b9a3d7 | 428 | free_cpumask_var(trial->effective_cpus); |
300ed6cb | 429 | free_cpumask_var(trial->cpus_allowed); |
645fcc9d LZ |
430 | kfree(trial); |
431 | } | |
432 | ||
1da177e4 LT |
433 | /* |
434 | * validate_change() - Used to validate that any proposed cpuset change | |
435 | * follows the structural rules for cpusets. | |
436 | * | |
437 | * If we replaced the flag and mask values of the current cpuset | |
438 | * (cur) with those values in the trial cpuset (trial), would | |
439 | * our various subset and exclusive rules still be valid? Presumes | |
5d21cc2d | 440 | * cpuset_mutex held. |
1da177e4 LT |
441 | * |
442 | * 'cur' is the address of an actual, in-use cpuset. Operations | |
443 | * such as list traversal that depend on the actual address of the | |
444 | * cpuset in the list must use cur below, not trial. | |
445 | * | |
446 | * 'trial' is the address of bulk structure copy of cur, with | |
447 | * perhaps one or more of the fields cpus_allowed, mems_allowed, | |
448 | * or flags changed to new, trial values. | |
449 | * | |
450 | * Return 0 if valid, -errno if not. | |
451 | */ | |
452 | ||
c9710d80 | 453 | static int validate_change(struct cpuset *cur, struct cpuset *trial) |
1da177e4 | 454 | { |
492eb21b | 455 | struct cgroup_subsys_state *css; |
1da177e4 | 456 | struct cpuset *c, *par; |
ae8086ce TH |
457 | int ret; |
458 | ||
459 | rcu_read_lock(); | |
1da177e4 LT |
460 | |
461 | /* Each of our child cpusets must be a subset of us */ | |
ae8086ce | 462 | ret = -EBUSY; |
492eb21b | 463 | cpuset_for_each_child(c, css, cur) |
ae8086ce TH |
464 | if (!is_cpuset_subset(c, trial)) |
465 | goto out; | |
1da177e4 LT |
466 | |
467 | /* Remaining checks don't apply to root cpuset */ | |
ae8086ce | 468 | ret = 0; |
69604067 | 469 | if (cur == &top_cpuset) |
ae8086ce | 470 | goto out; |
1da177e4 | 471 | |
c431069f | 472 | par = parent_cs(cur); |
69604067 | 473 | |
7e88291b | 474 | /* On legacy hiearchy, we must be a subset of our parent cpuset. */ |
ae8086ce | 475 | ret = -EACCES; |
7e88291b | 476 | if (!cgroup_on_dfl(cur->css.cgroup) && !is_cpuset_subset(trial, par)) |
ae8086ce | 477 | goto out; |
1da177e4 | 478 | |
2df167a3 PM |
479 | /* |
480 | * If either I or some sibling (!= me) is exclusive, we can't | |
481 | * overlap | |
482 | */ | |
ae8086ce | 483 | ret = -EINVAL; |
492eb21b | 484 | cpuset_for_each_child(c, css, par) { |
1da177e4 LT |
485 | if ((is_cpu_exclusive(trial) || is_cpu_exclusive(c)) && |
486 | c != cur && | |
300ed6cb | 487 | cpumask_intersects(trial->cpus_allowed, c->cpus_allowed)) |
ae8086ce | 488 | goto out; |
1da177e4 LT |
489 | if ((is_mem_exclusive(trial) || is_mem_exclusive(c)) && |
490 | c != cur && | |
491 | nodes_intersects(trial->mems_allowed, c->mems_allowed)) | |
ae8086ce | 492 | goto out; |
1da177e4 LT |
493 | } |
494 | ||
452477fa TH |
495 | /* |
496 | * Cpusets with tasks - existing or newly being attached - can't | |
1c09b195 | 497 | * be changed to have empty cpus_allowed or mems_allowed. |
452477fa | 498 | */ |
ae8086ce | 499 | ret = -ENOSPC; |
07bc356e | 500 | if ((cgroup_has_tasks(cur->css.cgroup) || cur->attach_in_progress)) { |
1c09b195 LZ |
501 | if (!cpumask_empty(cur->cpus_allowed) && |
502 | cpumask_empty(trial->cpus_allowed)) | |
503 | goto out; | |
504 | if (!nodes_empty(cur->mems_allowed) && | |
505 | nodes_empty(trial->mems_allowed)) | |
506 | goto out; | |
507 | } | |
020958b6 | 508 | |
f82f8042 JL |
509 | /* |
510 | * We can't shrink if we won't have enough room for SCHED_DEADLINE | |
511 | * tasks. | |
512 | */ | |
513 | ret = -EBUSY; | |
514 | if (is_cpu_exclusive(cur) && | |
515 | !cpuset_cpumask_can_shrink(cur->cpus_allowed, | |
516 | trial->cpus_allowed)) | |
517 | goto out; | |
518 | ||
ae8086ce TH |
519 | ret = 0; |
520 | out: | |
521 | rcu_read_unlock(); | |
522 | return ret; | |
1da177e4 LT |
523 | } |
524 | ||
db7f47cf | 525 | #ifdef CONFIG_SMP |
029190c5 | 526 | /* |
cf417141 | 527 | * Helper routine for generate_sched_domains(). |
8b5f1c52 | 528 | * Do cpusets a, b have overlapping effective cpus_allowed masks? |
029190c5 | 529 | */ |
029190c5 PJ |
530 | static int cpusets_overlap(struct cpuset *a, struct cpuset *b) |
531 | { | |
8b5f1c52 | 532 | return cpumask_intersects(a->effective_cpus, b->effective_cpus); |
029190c5 PJ |
533 | } |
534 | ||
1d3504fc HS |
535 | static void |
536 | update_domain_attr(struct sched_domain_attr *dattr, struct cpuset *c) | |
537 | { | |
1d3504fc HS |
538 | if (dattr->relax_domain_level < c->relax_domain_level) |
539 | dattr->relax_domain_level = c->relax_domain_level; | |
540 | return; | |
541 | } | |
542 | ||
fc560a26 TH |
543 | static void update_domain_attr_tree(struct sched_domain_attr *dattr, |
544 | struct cpuset *root_cs) | |
f5393693 | 545 | { |
fc560a26 | 546 | struct cpuset *cp; |
492eb21b | 547 | struct cgroup_subsys_state *pos_css; |
f5393693 | 548 | |
fc560a26 | 549 | rcu_read_lock(); |
492eb21b | 550 | cpuset_for_each_descendant_pre(cp, pos_css, root_cs) { |
bd8815a6 TH |
551 | if (cp == root_cs) |
552 | continue; | |
553 | ||
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 */ |
1d3504fc | 628 | struct sched_domain_attr *dattr; /* attributes for custom domains */ |
1583715d | 629 | int ndoms = 0; /* number of sched domains in result */ |
6af866af | 630 | int nslot; /* next empty doms[] struct cpumask slot */ |
492eb21b | 631 | struct cgroup_subsys_state *pos_css; |
029190c5 | 632 | |
029190c5 | 633 | doms = NULL; |
1d3504fc | 634 | dattr = NULL; |
cf417141 | 635 | csa = NULL; |
029190c5 PJ |
636 | |
637 | /* Special case for the 99% of systems with one, full, sched domain */ | |
638 | if (is_sched_load_balance(&top_cpuset)) { | |
acc3f5d7 RR |
639 | ndoms = 1; |
640 | doms = alloc_sched_domains(ndoms); | |
029190c5 | 641 | if (!doms) |
cf417141 MK |
642 | goto done; |
643 | ||
1d3504fc HS |
644 | dattr = kmalloc(sizeof(struct sched_domain_attr), GFP_KERNEL); |
645 | if (dattr) { | |
646 | *dattr = SD_ATTR_INIT; | |
93a65575 | 647 | update_domain_attr_tree(dattr, &top_cpuset); |
1d3504fc | 648 | } |
8b5f1c52 | 649 | cpumask_copy(doms[0], top_cpuset.effective_cpus); |
cf417141 | 650 | |
cf417141 | 651 | goto done; |
029190c5 PJ |
652 | } |
653 | ||
664eedde | 654 | csa = kmalloc(nr_cpusets() * sizeof(cp), GFP_KERNEL); |
029190c5 PJ |
655 | if (!csa) |
656 | goto done; | |
657 | csn = 0; | |
658 | ||
fc560a26 | 659 | rcu_read_lock(); |
492eb21b | 660 | cpuset_for_each_descendant_pre(cp, pos_css, &top_cpuset) { |
bd8815a6 TH |
661 | if (cp == &top_cpuset) |
662 | continue; | |
f5393693 | 663 | /* |
fc560a26 TH |
664 | * Continue traversing beyond @cp iff @cp has some CPUs and |
665 | * isn't load balancing. The former is obvious. The | |
666 | * latter: All child cpusets contain a subset of the | |
667 | * parent's cpus, so just skip them, and then we call | |
668 | * update_domain_attr_tree() to calc relax_domain_level of | |
669 | * the corresponding sched domain. | |
f5393693 | 670 | */ |
fc560a26 TH |
671 | if (!cpumask_empty(cp->cpus_allowed) && |
672 | !is_sched_load_balance(cp)) | |
f5393693 | 673 | continue; |
489a5393 | 674 | |
fc560a26 TH |
675 | if (is_sched_load_balance(cp)) |
676 | csa[csn++] = cp; | |
677 | ||
678 | /* skip @cp's subtree */ | |
492eb21b | 679 | pos_css = css_rightmost_descendant(pos_css); |
fc560a26 TH |
680 | } |
681 | rcu_read_unlock(); | |
029190c5 PJ |
682 | |
683 | for (i = 0; i < csn; i++) | |
684 | csa[i]->pn = i; | |
685 | ndoms = csn; | |
686 | ||
687 | restart: | |
688 | /* Find the best partition (set of sched domains) */ | |
689 | for (i = 0; i < csn; i++) { | |
690 | struct cpuset *a = csa[i]; | |
691 | int apn = a->pn; | |
692 | ||
693 | for (j = 0; j < csn; j++) { | |
694 | struct cpuset *b = csa[j]; | |
695 | int bpn = b->pn; | |
696 | ||
697 | if (apn != bpn && cpusets_overlap(a, b)) { | |
698 | for (k = 0; k < csn; k++) { | |
699 | struct cpuset *c = csa[k]; | |
700 | ||
701 | if (c->pn == bpn) | |
702 | c->pn = apn; | |
703 | } | |
704 | ndoms--; /* one less element */ | |
705 | goto restart; | |
706 | } | |
707 | } | |
708 | } | |
709 | ||
cf417141 MK |
710 | /* |
711 | * Now we know how many domains to create. | |
712 | * Convert <csn, csa> to <ndoms, doms> and populate cpu masks. | |
713 | */ | |
acc3f5d7 | 714 | doms = alloc_sched_domains(ndoms); |
700018e0 | 715 | if (!doms) |
cf417141 | 716 | goto done; |
cf417141 MK |
717 | |
718 | /* | |
719 | * The rest of the code, including the scheduler, can deal with | |
720 | * dattr==NULL case. No need to abort if alloc fails. | |
721 | */ | |
1d3504fc | 722 | dattr = kmalloc(ndoms * sizeof(struct sched_domain_attr), GFP_KERNEL); |
029190c5 PJ |
723 | |
724 | for (nslot = 0, i = 0; i < csn; i++) { | |
725 | struct cpuset *a = csa[i]; | |
6af866af | 726 | struct cpumask *dp; |
029190c5 PJ |
727 | int apn = a->pn; |
728 | ||
cf417141 MK |
729 | if (apn < 0) { |
730 | /* Skip completed partitions */ | |
731 | continue; | |
732 | } | |
733 | ||
acc3f5d7 | 734 | dp = doms[nslot]; |
cf417141 MK |
735 | |
736 | if (nslot == ndoms) { | |
737 | static int warnings = 10; | |
738 | if (warnings) { | |
12d3089c FF |
739 | pr_warn("rebuild_sched_domains confused: nslot %d, ndoms %d, csn %d, i %d, apn %d\n", |
740 | nslot, ndoms, csn, i, apn); | |
cf417141 | 741 | warnings--; |
029190c5 | 742 | } |
cf417141 MK |
743 | continue; |
744 | } | |
029190c5 | 745 | |
6af866af | 746 | cpumask_clear(dp); |
cf417141 MK |
747 | if (dattr) |
748 | *(dattr + nslot) = SD_ATTR_INIT; | |
749 | for (j = i; j < csn; j++) { | |
750 | struct cpuset *b = csa[j]; | |
751 | ||
752 | if (apn == b->pn) { | |
8b5f1c52 | 753 | cpumask_or(dp, dp, b->effective_cpus); |
cf417141 MK |
754 | if (dattr) |
755 | update_domain_attr_tree(dattr + nslot, b); | |
756 | ||
757 | /* Done with this partition */ | |
758 | b->pn = -1; | |
029190c5 | 759 | } |
029190c5 | 760 | } |
cf417141 | 761 | nslot++; |
029190c5 PJ |
762 | } |
763 | BUG_ON(nslot != ndoms); | |
764 | ||
cf417141 MK |
765 | done: |
766 | kfree(csa); | |
767 | ||
700018e0 LZ |
768 | /* |
769 | * Fallback to the default domain if kmalloc() failed. | |
770 | * See comments in partition_sched_domains(). | |
771 | */ | |
772 | if (doms == NULL) | |
773 | ndoms = 1; | |
774 | ||
cf417141 MK |
775 | *domains = doms; |
776 | *attributes = dattr; | |
777 | return ndoms; | |
778 | } | |
779 | ||
780 | /* | |
781 | * Rebuild scheduler domains. | |
782 | * | |
699140ba TH |
783 | * If the flag 'sched_load_balance' of any cpuset with non-empty |
784 | * 'cpus' changes, or if the 'cpus' allowed changes in any cpuset | |
785 | * which has that flag enabled, or if any cpuset with a non-empty | |
786 | * 'cpus' is removed, then call this routine to rebuild the | |
787 | * scheduler's dynamic sched domains. | |
cf417141 | 788 | * |
5d21cc2d | 789 | * Call with cpuset_mutex held. Takes get_online_cpus(). |
cf417141 | 790 | */ |
699140ba | 791 | static void rebuild_sched_domains_locked(void) |
cf417141 MK |
792 | { |
793 | struct sched_domain_attr *attr; | |
acc3f5d7 | 794 | cpumask_var_t *doms; |
cf417141 MK |
795 | int ndoms; |
796 | ||
5d21cc2d | 797 | lockdep_assert_held(&cpuset_mutex); |
86ef5c9a | 798 | get_online_cpus(); |
cf417141 | 799 | |
5b16c2a4 LZ |
800 | /* |
801 | * We have raced with CPU hotplug. Don't do anything to avoid | |
802 | * passing doms with offlined cpu to partition_sched_domains(). | |
803 | * Anyways, hotplug work item will rebuild sched domains. | |
804 | */ | |
8b5f1c52 | 805 | if (!cpumask_equal(top_cpuset.effective_cpus, cpu_active_mask)) |
5b16c2a4 LZ |
806 | goto out; |
807 | ||
cf417141 | 808 | /* Generate domain masks and attrs */ |
cf417141 | 809 | ndoms = generate_sched_domains(&doms, &attr); |
cf417141 MK |
810 | |
811 | /* Have scheduler rebuild the domains */ | |
812 | partition_sched_domains(ndoms, doms, attr); | |
5b16c2a4 | 813 | out: |
86ef5c9a | 814 | put_online_cpus(); |
cf417141 | 815 | } |
db7f47cf | 816 | #else /* !CONFIG_SMP */ |
699140ba | 817 | static void rebuild_sched_domains_locked(void) |
db7f47cf PM |
818 | { |
819 | } | |
db7f47cf | 820 | #endif /* CONFIG_SMP */ |
029190c5 | 821 | |
cf417141 MK |
822 | void rebuild_sched_domains(void) |
823 | { | |
5d21cc2d | 824 | mutex_lock(&cpuset_mutex); |
699140ba | 825 | rebuild_sched_domains_locked(); |
5d21cc2d | 826 | mutex_unlock(&cpuset_mutex); |
029190c5 PJ |
827 | } |
828 | ||
0b2f630a MX |
829 | /** |
830 | * update_tasks_cpumask - Update the cpumasks of tasks in the cpuset. | |
831 | * @cs: the cpuset in which each task's cpus_allowed mask needs to be changed | |
0b2f630a | 832 | * |
d66393e5 TH |
833 | * Iterate through each task of @cs updating its cpus_allowed to the |
834 | * effective cpuset's. As this function is called with cpuset_mutex held, | |
835 | * cpuset membership stays stable. | |
0b2f630a | 836 | */ |
d66393e5 | 837 | static void update_tasks_cpumask(struct cpuset *cs) |
0b2f630a | 838 | { |
d66393e5 TH |
839 | struct css_task_iter it; |
840 | struct task_struct *task; | |
841 | ||
842 | css_task_iter_start(&cs->css, &it); | |
843 | while ((task = css_task_iter_next(&it))) | |
ae1c8023 | 844 | set_cpus_allowed_ptr(task, cs->effective_cpus); |
d66393e5 | 845 | css_task_iter_end(&it); |
0b2f630a MX |
846 | } |
847 | ||
5c5cc623 | 848 | /* |
734d4513 LZ |
849 | * update_cpumasks_hier - Update effective cpumasks and tasks in the subtree |
850 | * @cs: the cpuset to consider | |
851 | * @new_cpus: temp variable for calculating new effective_cpus | |
852 | * | |
853 | * When congifured cpumask is changed, the effective cpumasks of this cpuset | |
854 | * and all its descendants need to be updated. | |
5c5cc623 | 855 | * |
734d4513 | 856 | * On legacy hierachy, effective_cpus will be the same with cpu_allowed. |
5c5cc623 LZ |
857 | * |
858 | * Called with cpuset_mutex held | |
859 | */ | |
734d4513 | 860 | static void update_cpumasks_hier(struct cpuset *cs, struct cpumask *new_cpus) |
5c5cc623 LZ |
861 | { |
862 | struct cpuset *cp; | |
492eb21b | 863 | struct cgroup_subsys_state *pos_css; |
8b5f1c52 | 864 | bool need_rebuild_sched_domains = false; |
5c5cc623 LZ |
865 | |
866 | rcu_read_lock(); | |
734d4513 LZ |
867 | cpuset_for_each_descendant_pre(cp, pos_css, cs) { |
868 | struct cpuset *parent = parent_cs(cp); | |
869 | ||
870 | cpumask_and(new_cpus, cp->cpus_allowed, parent->effective_cpus); | |
871 | ||
554b0d1c LZ |
872 | /* |
873 | * If it becomes empty, inherit the effective mask of the | |
874 | * parent, which is guaranteed to have some CPUs. | |
875 | */ | |
876 | if (cpumask_empty(new_cpus)) | |
877 | cpumask_copy(new_cpus, parent->effective_cpus); | |
878 | ||
734d4513 LZ |
879 | /* Skip the whole subtree if the cpumask remains the same. */ |
880 | if (cpumask_equal(new_cpus, cp->effective_cpus)) { | |
881 | pos_css = css_rightmost_descendant(pos_css); | |
882 | continue; | |
5c5cc623 | 883 | } |
734d4513 | 884 | |
ec903c0c | 885 | if (!css_tryget_online(&cp->css)) |
5c5cc623 LZ |
886 | continue; |
887 | rcu_read_unlock(); | |
888 | ||
734d4513 LZ |
889 | mutex_lock(&callback_mutex); |
890 | cpumask_copy(cp->effective_cpus, new_cpus); | |
891 | mutex_unlock(&callback_mutex); | |
892 | ||
893 | WARN_ON(!cgroup_on_dfl(cp->css.cgroup) && | |
894 | !cpumask_equal(cp->cpus_allowed, cp->effective_cpus)); | |
895 | ||
d66393e5 | 896 | update_tasks_cpumask(cp); |
5c5cc623 | 897 | |
8b5f1c52 LZ |
898 | /* |
899 | * If the effective cpumask of any non-empty cpuset is changed, | |
900 | * we need to rebuild sched domains. | |
901 | */ | |
902 | if (!cpumask_empty(cp->cpus_allowed) && | |
903 | is_sched_load_balance(cp)) | |
904 | need_rebuild_sched_domains = true; | |
905 | ||
5c5cc623 LZ |
906 | rcu_read_lock(); |
907 | css_put(&cp->css); | |
908 | } | |
909 | rcu_read_unlock(); | |
8b5f1c52 LZ |
910 | |
911 | if (need_rebuild_sched_domains) | |
912 | rebuild_sched_domains_locked(); | |
5c5cc623 LZ |
913 | } |
914 | ||
58f4790b CW |
915 | /** |
916 | * update_cpumask - update the cpus_allowed mask of a cpuset and all tasks in it | |
917 | * @cs: the cpuset to consider | |
fc34ac1d | 918 | * @trialcs: trial cpuset |
58f4790b CW |
919 | * @buf: buffer of cpu numbers written to this cpuset |
920 | */ | |
645fcc9d LZ |
921 | static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs, |
922 | const char *buf) | |
1da177e4 | 923 | { |
58f4790b | 924 | int retval; |
1da177e4 | 925 | |
5f054e31 | 926 | /* top_cpuset.cpus_allowed tracks cpu_online_mask; it's read-only */ |
4c4d50f7 PJ |
927 | if (cs == &top_cpuset) |
928 | return -EACCES; | |
929 | ||
6f7f02e7 | 930 | /* |
c8d9c90c | 931 | * An empty cpus_allowed is ok only if the cpuset has no tasks. |
020958b6 PJ |
932 | * Since cpulist_parse() fails on an empty mask, we special case |
933 | * that parsing. The validate_change() call ensures that cpusets | |
934 | * with tasks have cpus. | |
6f7f02e7 | 935 | */ |
020958b6 | 936 | if (!*buf) { |
300ed6cb | 937 | cpumask_clear(trialcs->cpus_allowed); |
6f7f02e7 | 938 | } else { |
300ed6cb | 939 | retval = cpulist_parse(buf, trialcs->cpus_allowed); |
6f7f02e7 DR |
940 | if (retval < 0) |
941 | return retval; | |
37340746 | 942 | |
5d8ba82c LZ |
943 | if (!cpumask_subset(trialcs->cpus_allowed, |
944 | top_cpuset.cpus_allowed)) | |
37340746 | 945 | return -EINVAL; |
6f7f02e7 | 946 | } |
029190c5 | 947 | |
8707d8b8 | 948 | /* Nothing to do if the cpus didn't change */ |
300ed6cb | 949 | if (cpumask_equal(cs->cpus_allowed, trialcs->cpus_allowed)) |
8707d8b8 | 950 | return 0; |
58f4790b | 951 | |
a73456f3 LZ |
952 | retval = validate_change(cs, trialcs); |
953 | if (retval < 0) | |
954 | return retval; | |
955 | ||
3d3f26a7 | 956 | mutex_lock(&callback_mutex); |
300ed6cb | 957 | cpumask_copy(cs->cpus_allowed, trialcs->cpus_allowed); |
3d3f26a7 | 958 | mutex_unlock(&callback_mutex); |
029190c5 | 959 | |
734d4513 LZ |
960 | /* use trialcs->cpus_allowed as a temp variable */ |
961 | update_cpumasks_hier(cs, trialcs->cpus_allowed); | |
85d7b949 | 962 | return 0; |
1da177e4 LT |
963 | } |
964 | ||
e4e364e8 PJ |
965 | /* |
966 | * cpuset_migrate_mm | |
967 | * | |
968 | * Migrate memory region from one set of nodes to another. | |
969 | * | |
970 | * Temporarilly set tasks mems_allowed to target nodes of migration, | |
971 | * so that the migration code can allocate pages on these nodes. | |
972 | * | |
e4e364e8 PJ |
973 | * While the mm_struct we are migrating is typically from some |
974 | * other task, the task_struct mems_allowed that we are hacking | |
975 | * is for our current task, which must allocate new pages for that | |
976 | * migrating memory region. | |
e4e364e8 PJ |
977 | */ |
978 | ||
979 | static void cpuset_migrate_mm(struct mm_struct *mm, const nodemask_t *from, | |
980 | const nodemask_t *to) | |
981 | { | |
982 | struct task_struct *tsk = current; | |
983 | ||
e4e364e8 | 984 | tsk->mems_allowed = *to; |
e4e364e8 PJ |
985 | |
986 | do_migrate_pages(mm, from, to, MPOL_MF_MOVE_ALL); | |
987 | ||
47295830 | 988 | rcu_read_lock(); |
ae1c8023 | 989 | guarantee_online_mems(task_cs(tsk), &tsk->mems_allowed); |
47295830 | 990 | rcu_read_unlock(); |
e4e364e8 PJ |
991 | } |
992 | ||
3b6766fe | 993 | /* |
58568d2a MX |
994 | * cpuset_change_task_nodemask - change task's mems_allowed and mempolicy |
995 | * @tsk: the task to change | |
996 | * @newmems: new nodes that the task will be set | |
997 | * | |
998 | * In order to avoid seeing no nodes if the old and new nodes are disjoint, | |
999 | * we structure updates as setting all new allowed nodes, then clearing newly | |
1000 | * disallowed ones. | |
58568d2a MX |
1001 | */ |
1002 | static void cpuset_change_task_nodemask(struct task_struct *tsk, | |
1003 | nodemask_t *newmems) | |
1004 | { | |
b246272e | 1005 | bool need_loop; |
89e8a244 | 1006 | |
c0ff7453 MX |
1007 | /* |
1008 | * Allow tasks that have access to memory reserves because they have | |
1009 | * been OOM killed to get memory anywhere. | |
1010 | */ | |
1011 | if (unlikely(test_thread_flag(TIF_MEMDIE))) | |
1012 | return; | |
1013 | if (current->flags & PF_EXITING) /* Let dying task have memory */ | |
1014 | return; | |
1015 | ||
1016 | task_lock(tsk); | |
b246272e DR |
1017 | /* |
1018 | * Determine if a loop is necessary if another thread is doing | |
d26914d1 | 1019 | * read_mems_allowed_begin(). If at least one node remains unchanged and |
b246272e DR |
1020 | * tsk does not have a mempolicy, then an empty nodemask will not be |
1021 | * possible when mems_allowed is larger than a word. | |
1022 | */ | |
1023 | need_loop = task_has_mempolicy(tsk) || | |
1024 | !nodes_intersects(*newmems, tsk->mems_allowed); | |
c0ff7453 | 1025 | |
0fc0287c PZ |
1026 | if (need_loop) { |
1027 | local_irq_disable(); | |
cc9a6c87 | 1028 | write_seqcount_begin(&tsk->mems_allowed_seq); |
0fc0287c | 1029 | } |
c0ff7453 | 1030 | |
cc9a6c87 MG |
1031 | nodes_or(tsk->mems_allowed, tsk->mems_allowed, *newmems); |
1032 | mpol_rebind_task(tsk, newmems, MPOL_REBIND_STEP1); | |
c0ff7453 MX |
1033 | |
1034 | mpol_rebind_task(tsk, newmems, MPOL_REBIND_STEP2); | |
58568d2a | 1035 | tsk->mems_allowed = *newmems; |
cc9a6c87 | 1036 | |
0fc0287c | 1037 | if (need_loop) { |
cc9a6c87 | 1038 | write_seqcount_end(&tsk->mems_allowed_seq); |
0fc0287c PZ |
1039 | local_irq_enable(); |
1040 | } | |
cc9a6c87 | 1041 | |
c0ff7453 | 1042 | task_unlock(tsk); |
58568d2a MX |
1043 | } |
1044 | ||
8793d854 PM |
1045 | static void *cpuset_being_rebound; |
1046 | ||
0b2f630a MX |
1047 | /** |
1048 | * update_tasks_nodemask - Update the nodemasks of tasks in the cpuset. | |
1049 | * @cs: the cpuset in which each task's mems_allowed mask needs to be changed | |
0b2f630a | 1050 | * |
d66393e5 TH |
1051 | * Iterate through each task of @cs updating its mems_allowed to the |
1052 | * effective cpuset's. As this function is called with cpuset_mutex held, | |
1053 | * cpuset membership stays stable. | |
0b2f630a | 1054 | */ |
d66393e5 | 1055 | static void update_tasks_nodemask(struct cpuset *cs) |
1da177e4 | 1056 | { |
33ad801d | 1057 | static nodemask_t newmems; /* protected by cpuset_mutex */ |
d66393e5 TH |
1058 | struct css_task_iter it; |
1059 | struct task_struct *task; | |
59dac16f | 1060 | |
846a16bf | 1061 | cpuset_being_rebound = cs; /* causes mpol_dup() rebind */ |
4225399a | 1062 | |
ae1c8023 | 1063 | guarantee_online_mems(cs, &newmems); |
33ad801d | 1064 | |
4225399a | 1065 | /* |
3b6766fe LZ |
1066 | * The mpol_rebind_mm() call takes mmap_sem, which we couldn't |
1067 | * take while holding tasklist_lock. Forks can happen - the | |
1068 | * mpol_dup() cpuset_being_rebound check will catch such forks, | |
1069 | * and rebind their vma mempolicies too. Because we still hold | |
5d21cc2d | 1070 | * the global cpuset_mutex, we know that no other rebind effort |
3b6766fe | 1071 | * will be contending for the global variable cpuset_being_rebound. |
4225399a | 1072 | * It's ok if we rebind the same mm twice; mpol_rebind_mm() |
04c19fa6 | 1073 | * is idempotent. Also migrate pages in each mm to new nodes. |
4225399a | 1074 | */ |
d66393e5 TH |
1075 | css_task_iter_start(&cs->css, &it); |
1076 | while ((task = css_task_iter_next(&it))) { | |
1077 | struct mm_struct *mm; | |
1078 | bool migrate; | |
1079 | ||
1080 | cpuset_change_task_nodemask(task, &newmems); | |
1081 | ||
1082 | mm = get_task_mm(task); | |
1083 | if (!mm) | |
1084 | continue; | |
1085 | ||
1086 | migrate = is_memory_migrate(cs); | |
1087 | ||
1088 | mpol_rebind_mm(mm, &cs->mems_allowed); | |
1089 | if (migrate) | |
1090 | cpuset_migrate_mm(mm, &cs->old_mems_allowed, &newmems); | |
1091 | mmput(mm); | |
1092 | } | |
1093 | css_task_iter_end(&it); | |
4225399a | 1094 | |
33ad801d LZ |
1095 | /* |
1096 | * All the tasks' nodemasks have been updated, update | |
1097 | * cs->old_mems_allowed. | |
1098 | */ | |
1099 | cs->old_mems_allowed = newmems; | |
1100 | ||
2df167a3 | 1101 | /* We're done rebinding vmas to this cpuset's new mems_allowed. */ |
8793d854 | 1102 | cpuset_being_rebound = NULL; |
1da177e4 LT |
1103 | } |
1104 | ||
5c5cc623 | 1105 | /* |
734d4513 LZ |
1106 | * update_nodemasks_hier - Update effective nodemasks and tasks in the subtree |
1107 | * @cs: the cpuset to consider | |
1108 | * @new_mems: a temp variable for calculating new effective_mems | |
5c5cc623 | 1109 | * |
734d4513 LZ |
1110 | * When configured nodemask is changed, the effective nodemasks of this cpuset |
1111 | * and all its descendants need to be updated. | |
5c5cc623 | 1112 | * |
734d4513 | 1113 | * On legacy hiearchy, effective_mems will be the same with mems_allowed. |
5c5cc623 LZ |
1114 | * |
1115 | * Called with cpuset_mutex held | |
1116 | */ | |
734d4513 | 1117 | static void update_nodemasks_hier(struct cpuset *cs, nodemask_t *new_mems) |
5c5cc623 LZ |
1118 | { |
1119 | struct cpuset *cp; | |
492eb21b | 1120 | struct cgroup_subsys_state *pos_css; |
5c5cc623 LZ |
1121 | |
1122 | rcu_read_lock(); | |
734d4513 LZ |
1123 | cpuset_for_each_descendant_pre(cp, pos_css, cs) { |
1124 | struct cpuset *parent = parent_cs(cp); | |
1125 | ||
1126 | nodes_and(*new_mems, cp->mems_allowed, parent->effective_mems); | |
1127 | ||
554b0d1c LZ |
1128 | /* |
1129 | * If it becomes empty, inherit the effective mask of the | |
1130 | * parent, which is guaranteed to have some MEMs. | |
1131 | */ | |
1132 | if (nodes_empty(*new_mems)) | |
1133 | *new_mems = parent->effective_mems; | |
1134 | ||
734d4513 LZ |
1135 | /* Skip the whole subtree if the nodemask remains the same. */ |
1136 | if (nodes_equal(*new_mems, cp->effective_mems)) { | |
1137 | pos_css = css_rightmost_descendant(pos_css); | |
1138 | continue; | |
5c5cc623 | 1139 | } |
734d4513 | 1140 | |
ec903c0c | 1141 | if (!css_tryget_online(&cp->css)) |
5c5cc623 LZ |
1142 | continue; |
1143 | rcu_read_unlock(); | |
1144 | ||
734d4513 LZ |
1145 | mutex_lock(&callback_mutex); |
1146 | cp->effective_mems = *new_mems; | |
1147 | mutex_unlock(&callback_mutex); | |
1148 | ||
1149 | WARN_ON(!cgroup_on_dfl(cp->css.cgroup) && | |
a1381268 | 1150 | !nodes_equal(cp->mems_allowed, cp->effective_mems)); |
734d4513 | 1151 | |
d66393e5 | 1152 | update_tasks_nodemask(cp); |
5c5cc623 LZ |
1153 | |
1154 | rcu_read_lock(); | |
1155 | css_put(&cp->css); | |
1156 | } | |
1157 | rcu_read_unlock(); | |
1158 | } | |
1159 | ||
0b2f630a MX |
1160 | /* |
1161 | * Handle user request to change the 'mems' memory placement | |
1162 | * of a cpuset. Needs to validate the request, update the | |
58568d2a MX |
1163 | * cpusets mems_allowed, and for each task in the cpuset, |
1164 | * update mems_allowed and rebind task's mempolicy and any vma | |
1165 | * mempolicies and if the cpuset is marked 'memory_migrate', | |
1166 | * migrate the tasks pages to the new memory. | |
0b2f630a | 1167 | * |
5d21cc2d | 1168 | * Call with cpuset_mutex held. May take callback_mutex during call. |
0b2f630a MX |
1169 | * Will take tasklist_lock, scan tasklist for tasks in cpuset cs, |
1170 | * lock each such tasks mm->mmap_sem, scan its vma's and rebind | |
1171 | * their mempolicies to the cpusets new mems_allowed. | |
1172 | */ | |
645fcc9d LZ |
1173 | static int update_nodemask(struct cpuset *cs, struct cpuset *trialcs, |
1174 | const char *buf) | |
0b2f630a | 1175 | { |
0b2f630a MX |
1176 | int retval; |
1177 | ||
1178 | /* | |
38d7bee9 | 1179 | * top_cpuset.mems_allowed tracks node_stats[N_MEMORY]; |
0b2f630a MX |
1180 | * it's read-only |
1181 | */ | |
53feb297 MX |
1182 | if (cs == &top_cpuset) { |
1183 | retval = -EACCES; | |
1184 | goto done; | |
1185 | } | |
0b2f630a | 1186 | |
0b2f630a MX |
1187 | /* |
1188 | * An empty mems_allowed is ok iff there are no tasks in the cpuset. | |
1189 | * Since nodelist_parse() fails on an empty mask, we special case | |
1190 | * that parsing. The validate_change() call ensures that cpusets | |
1191 | * with tasks have memory. | |
1192 | */ | |
1193 | if (!*buf) { | |
645fcc9d | 1194 | nodes_clear(trialcs->mems_allowed); |
0b2f630a | 1195 | } else { |
645fcc9d | 1196 | retval = nodelist_parse(buf, trialcs->mems_allowed); |
0b2f630a MX |
1197 | if (retval < 0) |
1198 | goto done; | |
1199 | ||
645fcc9d | 1200 | if (!nodes_subset(trialcs->mems_allowed, |
5d8ba82c LZ |
1201 | top_cpuset.mems_allowed)) { |
1202 | retval = -EINVAL; | |
53feb297 MX |
1203 | goto done; |
1204 | } | |
0b2f630a | 1205 | } |
33ad801d LZ |
1206 | |
1207 | if (nodes_equal(cs->mems_allowed, trialcs->mems_allowed)) { | |
0b2f630a MX |
1208 | retval = 0; /* Too easy - nothing to do */ |
1209 | goto done; | |
1210 | } | |
645fcc9d | 1211 | retval = validate_change(cs, trialcs); |
0b2f630a MX |
1212 | if (retval < 0) |
1213 | goto done; | |
1214 | ||
1215 | mutex_lock(&callback_mutex); | |
645fcc9d | 1216 | cs->mems_allowed = trialcs->mems_allowed; |
0b2f630a MX |
1217 | mutex_unlock(&callback_mutex); |
1218 | ||
734d4513 LZ |
1219 | /* use trialcs->mems_allowed as a temp variable */ |
1220 | update_nodemasks_hier(cs, &cs->mems_allowed); | |
0b2f630a MX |
1221 | done: |
1222 | return retval; | |
1223 | } | |
1224 | ||
8793d854 PM |
1225 | int current_cpuset_is_being_rebound(void) |
1226 | { | |
391acf97 GZ |
1227 | int ret; |
1228 | ||
1229 | rcu_read_lock(); | |
1230 | ret = task_cs(current) == cpuset_being_rebound; | |
1231 | rcu_read_unlock(); | |
1232 | ||
1233 | return ret; | |
8793d854 PM |
1234 | } |
1235 | ||
5be7a479 | 1236 | static int update_relax_domain_level(struct cpuset *cs, s64 val) |
1d3504fc | 1237 | { |
db7f47cf | 1238 | #ifdef CONFIG_SMP |
60495e77 | 1239 | if (val < -1 || val >= sched_domain_level_max) |
30e0e178 | 1240 | return -EINVAL; |
db7f47cf | 1241 | #endif |
1d3504fc HS |
1242 | |
1243 | if (val != cs->relax_domain_level) { | |
1244 | cs->relax_domain_level = val; | |
300ed6cb LZ |
1245 | if (!cpumask_empty(cs->cpus_allowed) && |
1246 | is_sched_load_balance(cs)) | |
699140ba | 1247 | rebuild_sched_domains_locked(); |
1d3504fc HS |
1248 | } |
1249 | ||
1250 | return 0; | |
1251 | } | |
1252 | ||
72ec7029 | 1253 | /** |
950592f7 MX |
1254 | * update_tasks_flags - update the spread flags of tasks in the cpuset. |
1255 | * @cs: the cpuset in which each task's spread flags needs to be changed | |
950592f7 | 1256 | * |
d66393e5 TH |
1257 | * Iterate through each task of @cs updating its spread flags. As this |
1258 | * function is called with cpuset_mutex held, cpuset membership stays | |
1259 | * stable. | |
950592f7 | 1260 | */ |
d66393e5 | 1261 | static void update_tasks_flags(struct cpuset *cs) |
950592f7 | 1262 | { |
d66393e5 TH |
1263 | struct css_task_iter it; |
1264 | struct task_struct *task; | |
1265 | ||
1266 | css_task_iter_start(&cs->css, &it); | |
1267 | while ((task = css_task_iter_next(&it))) | |
1268 | cpuset_update_task_spread_flag(cs, task); | |
1269 | css_task_iter_end(&it); | |
950592f7 MX |
1270 | } |
1271 | ||
1da177e4 LT |
1272 | /* |
1273 | * update_flag - read a 0 or a 1 in a file and update associated flag | |
78608366 PM |
1274 | * bit: the bit to update (see cpuset_flagbits_t) |
1275 | * cs: the cpuset to update | |
1276 | * turning_on: whether the flag is being set or cleared | |
053199ed | 1277 | * |
5d21cc2d | 1278 | * Call with cpuset_mutex held. |
1da177e4 LT |
1279 | */ |
1280 | ||
700fe1ab PM |
1281 | static int update_flag(cpuset_flagbits_t bit, struct cpuset *cs, |
1282 | int turning_on) | |
1da177e4 | 1283 | { |
645fcc9d | 1284 | struct cpuset *trialcs; |
40b6a762 | 1285 | int balance_flag_changed; |
950592f7 | 1286 | int spread_flag_changed; |
950592f7 | 1287 | int err; |
1da177e4 | 1288 | |
645fcc9d LZ |
1289 | trialcs = alloc_trial_cpuset(cs); |
1290 | if (!trialcs) | |
1291 | return -ENOMEM; | |
1292 | ||
1da177e4 | 1293 | if (turning_on) |
645fcc9d | 1294 | set_bit(bit, &trialcs->flags); |
1da177e4 | 1295 | else |
645fcc9d | 1296 | clear_bit(bit, &trialcs->flags); |
1da177e4 | 1297 | |
645fcc9d | 1298 | err = validate_change(cs, trialcs); |
85d7b949 | 1299 | if (err < 0) |
645fcc9d | 1300 | goto out; |
029190c5 | 1301 | |
029190c5 | 1302 | balance_flag_changed = (is_sched_load_balance(cs) != |
645fcc9d | 1303 | is_sched_load_balance(trialcs)); |
029190c5 | 1304 | |
950592f7 MX |
1305 | spread_flag_changed = ((is_spread_slab(cs) != is_spread_slab(trialcs)) |
1306 | || (is_spread_page(cs) != is_spread_page(trialcs))); | |
1307 | ||
3d3f26a7 | 1308 | mutex_lock(&callback_mutex); |
645fcc9d | 1309 | cs->flags = trialcs->flags; |
3d3f26a7 | 1310 | mutex_unlock(&callback_mutex); |
85d7b949 | 1311 | |
300ed6cb | 1312 | if (!cpumask_empty(trialcs->cpus_allowed) && balance_flag_changed) |
699140ba | 1313 | rebuild_sched_domains_locked(); |
029190c5 | 1314 | |
950592f7 | 1315 | if (spread_flag_changed) |
d66393e5 | 1316 | update_tasks_flags(cs); |
645fcc9d LZ |
1317 | out: |
1318 | free_trial_cpuset(trialcs); | |
1319 | return err; | |
1da177e4 LT |
1320 | } |
1321 | ||
3e0d98b9 | 1322 | /* |
80f7228b | 1323 | * Frequency meter - How fast is some event occurring? |
3e0d98b9 PJ |
1324 | * |
1325 | * These routines manage a digitally filtered, constant time based, | |
1326 | * event frequency meter. There are four routines: | |
1327 | * fmeter_init() - initialize a frequency meter. | |
1328 | * fmeter_markevent() - called each time the event happens. | |
1329 | * fmeter_getrate() - returns the recent rate of such events. | |
1330 | * fmeter_update() - internal routine used to update fmeter. | |
1331 | * | |
1332 | * A common data structure is passed to each of these routines, | |
1333 | * which is used to keep track of the state required to manage the | |
1334 | * frequency meter and its digital filter. | |
1335 | * | |
1336 | * The filter works on the number of events marked per unit time. | |
1337 | * The filter is single-pole low-pass recursive (IIR). The time unit | |
1338 | * is 1 second. Arithmetic is done using 32-bit integers scaled to | |
1339 | * simulate 3 decimal digits of precision (multiplied by 1000). | |
1340 | * | |
1341 | * With an FM_COEF of 933, and a time base of 1 second, the filter | |
1342 | * has a half-life of 10 seconds, meaning that if the events quit | |
1343 | * happening, then the rate returned from the fmeter_getrate() | |
1344 | * will be cut in half each 10 seconds, until it converges to zero. | |
1345 | * | |
1346 | * It is not worth doing a real infinitely recursive filter. If more | |
1347 | * than FM_MAXTICKS ticks have elapsed since the last filter event, | |
1348 | * just compute FM_MAXTICKS ticks worth, by which point the level | |
1349 | * will be stable. | |
1350 | * | |
1351 | * Limit the count of unprocessed events to FM_MAXCNT, so as to avoid | |
1352 | * arithmetic overflow in the fmeter_update() routine. | |
1353 | * | |
1354 | * Given the simple 32 bit integer arithmetic used, this meter works | |
1355 | * best for reporting rates between one per millisecond (msec) and | |
1356 | * one per 32 (approx) seconds. At constant rates faster than one | |
1357 | * per msec it maxes out at values just under 1,000,000. At constant | |
1358 | * rates between one per msec, and one per second it will stabilize | |
1359 | * to a value N*1000, where N is the rate of events per second. | |
1360 | * At constant rates between one per second and one per 32 seconds, | |
1361 | * it will be choppy, moving up on the seconds that have an event, | |
1362 | * and then decaying until the next event. At rates slower than | |
1363 | * about one in 32 seconds, it decays all the way back to zero between | |
1364 | * each event. | |
1365 | */ | |
1366 | ||
1367 | #define FM_COEF 933 /* coefficient for half-life of 10 secs */ | |
1368 | #define FM_MAXTICKS ((time_t)99) /* useless computing more ticks than this */ | |
1369 | #define FM_MAXCNT 1000000 /* limit cnt to avoid overflow */ | |
1370 | #define FM_SCALE 1000 /* faux fixed point scale */ | |
1371 | ||
1372 | /* Initialize a frequency meter */ | |
1373 | static void fmeter_init(struct fmeter *fmp) | |
1374 | { | |
1375 | fmp->cnt = 0; | |
1376 | fmp->val = 0; | |
1377 | fmp->time = 0; | |
1378 | spin_lock_init(&fmp->lock); | |
1379 | } | |
1380 | ||
1381 | /* Internal meter update - process cnt events and update value */ | |
1382 | static void fmeter_update(struct fmeter *fmp) | |
1383 | { | |
1384 | time_t now = get_seconds(); | |
1385 | time_t ticks = now - fmp->time; | |
1386 | ||
1387 | if (ticks == 0) | |
1388 | return; | |
1389 | ||
1390 | ticks = min(FM_MAXTICKS, ticks); | |
1391 | while (ticks-- > 0) | |
1392 | fmp->val = (FM_COEF * fmp->val) / FM_SCALE; | |
1393 | fmp->time = now; | |
1394 | ||
1395 | fmp->val += ((FM_SCALE - FM_COEF) * fmp->cnt) / FM_SCALE; | |
1396 | fmp->cnt = 0; | |
1397 | } | |
1398 | ||
1399 | /* Process any previous ticks, then bump cnt by one (times scale). */ | |
1400 | static void fmeter_markevent(struct fmeter *fmp) | |
1401 | { | |
1402 | spin_lock(&fmp->lock); | |
1403 | fmeter_update(fmp); | |
1404 | fmp->cnt = min(FM_MAXCNT, fmp->cnt + FM_SCALE); | |
1405 | spin_unlock(&fmp->lock); | |
1406 | } | |
1407 | ||
1408 | /* Process any previous ticks, then return current value. */ | |
1409 | static int fmeter_getrate(struct fmeter *fmp) | |
1410 | { | |
1411 | int val; | |
1412 | ||
1413 | spin_lock(&fmp->lock); | |
1414 | fmeter_update(fmp); | |
1415 | val = fmp->val; | |
1416 | spin_unlock(&fmp->lock); | |
1417 | return val; | |
1418 | } | |
1419 | ||
57fce0a6 TH |
1420 | static struct cpuset *cpuset_attach_old_cs; |
1421 | ||
5d21cc2d | 1422 | /* Called by cgroups to determine if a cpuset is usable; cpuset_mutex held */ |
eb95419b TH |
1423 | static int cpuset_can_attach(struct cgroup_subsys_state *css, |
1424 | struct cgroup_taskset *tset) | |
f780bdb7 | 1425 | { |
eb95419b | 1426 | struct cpuset *cs = css_cs(css); |
bb9d97b6 TH |
1427 | struct task_struct *task; |
1428 | int ret; | |
1da177e4 | 1429 | |
57fce0a6 TH |
1430 | /* used later by cpuset_attach() */ |
1431 | cpuset_attach_old_cs = task_cs(cgroup_taskset_first(tset)); | |
1432 | ||
5d21cc2d TH |
1433 | mutex_lock(&cpuset_mutex); |
1434 | ||
aa6ec29b | 1435 | /* allow moving tasks into an empty cpuset if on default hierarchy */ |
5d21cc2d | 1436 | ret = -ENOSPC; |
aa6ec29b | 1437 | if (!cgroup_on_dfl(css->cgroup) && |
88fa523b | 1438 | (cpumask_empty(cs->cpus_allowed) || nodes_empty(cs->mems_allowed))) |
5d21cc2d | 1439 | goto out_unlock; |
9985b0ba | 1440 | |
924f0d9a | 1441 | cgroup_taskset_for_each(task, tset) { |
7f51412a JL |
1442 | ret = task_can_attach(task, cs->cpus_allowed); |
1443 | if (ret) | |
5d21cc2d TH |
1444 | goto out_unlock; |
1445 | ret = security_task_setscheduler(task); | |
1446 | if (ret) | |
1447 | goto out_unlock; | |
bb9d97b6 | 1448 | } |
f780bdb7 | 1449 | |
452477fa TH |
1450 | /* |
1451 | * Mark attach is in progress. This makes validate_change() fail | |
1452 | * changes which zero cpus/mems_allowed. | |
1453 | */ | |
1454 | cs->attach_in_progress++; | |
5d21cc2d TH |
1455 | ret = 0; |
1456 | out_unlock: | |
1457 | mutex_unlock(&cpuset_mutex); | |
1458 | return ret; | |
8793d854 | 1459 | } |
f780bdb7 | 1460 | |
eb95419b | 1461 | static void cpuset_cancel_attach(struct cgroup_subsys_state *css, |
452477fa TH |
1462 | struct cgroup_taskset *tset) |
1463 | { | |
5d21cc2d | 1464 | mutex_lock(&cpuset_mutex); |
eb95419b | 1465 | css_cs(css)->attach_in_progress--; |
5d21cc2d | 1466 | mutex_unlock(&cpuset_mutex); |
8793d854 | 1467 | } |
1da177e4 | 1468 | |
4e4c9a14 | 1469 | /* |
5d21cc2d | 1470 | * Protected by cpuset_mutex. cpus_attach is used only by cpuset_attach() |
4e4c9a14 TH |
1471 | * but we can't allocate it dynamically there. Define it global and |
1472 | * allocate from cpuset_init(). | |
1473 | */ | |
1474 | static cpumask_var_t cpus_attach; | |
1475 | ||
eb95419b TH |
1476 | static void cpuset_attach(struct cgroup_subsys_state *css, |
1477 | struct cgroup_taskset *tset) | |
8793d854 | 1478 | { |
67bd2c59 | 1479 | /* static buf protected by cpuset_mutex */ |
4e4c9a14 | 1480 | static nodemask_t cpuset_attach_nodemask_to; |
8793d854 | 1481 | struct mm_struct *mm; |
bb9d97b6 TH |
1482 | struct task_struct *task; |
1483 | struct task_struct *leader = cgroup_taskset_first(tset); | |
eb95419b | 1484 | struct cpuset *cs = css_cs(css); |
57fce0a6 | 1485 | struct cpuset *oldcs = cpuset_attach_old_cs; |
22fb52dd | 1486 | |
5d21cc2d TH |
1487 | mutex_lock(&cpuset_mutex); |
1488 | ||
4e4c9a14 TH |
1489 | /* prepare for attach */ |
1490 | if (cs == &top_cpuset) | |
1491 | cpumask_copy(cpus_attach, cpu_possible_mask); | |
1492 | else | |
ae1c8023 | 1493 | guarantee_online_cpus(cs, cpus_attach); |
4e4c9a14 | 1494 | |
ae1c8023 | 1495 | guarantee_online_mems(cs, &cpuset_attach_nodemask_to); |
4e4c9a14 | 1496 | |
924f0d9a | 1497 | cgroup_taskset_for_each(task, tset) { |
bb9d97b6 TH |
1498 | /* |
1499 | * can_attach beforehand should guarantee that this doesn't | |
1500 | * fail. TODO: have a better way to handle failure here | |
1501 | */ | |
1502 | WARN_ON_ONCE(set_cpus_allowed_ptr(task, cpus_attach)); | |
1503 | ||
1504 | cpuset_change_task_nodemask(task, &cpuset_attach_nodemask_to); | |
1505 | cpuset_update_task_spread_flag(cs, task); | |
1506 | } | |
22fb52dd | 1507 | |
f780bdb7 BB |
1508 | /* |
1509 | * Change mm, possibly for multiple threads in a threadgroup. This is | |
1510 | * expensive and may sleep. | |
1511 | */ | |
ae1c8023 | 1512 | cpuset_attach_nodemask_to = cs->effective_mems; |
bb9d97b6 | 1513 | mm = get_task_mm(leader); |
4225399a | 1514 | if (mm) { |
f780bdb7 | 1515 | mpol_rebind_mm(mm, &cpuset_attach_nodemask_to); |
f047cecf LZ |
1516 | |
1517 | /* | |
1518 | * old_mems_allowed is the same with mems_allowed here, except | |
1519 | * if this task is being moved automatically due to hotplug. | |
1520 | * In that case @mems_allowed has been updated and is empty, | |
1521 | * so @old_mems_allowed is the right nodesets that we migrate | |
1522 | * mm from. | |
1523 | */ | |
1524 | if (is_memory_migrate(cs)) { | |
ae1c8023 | 1525 | cpuset_migrate_mm(mm, &oldcs->old_mems_allowed, |
f780bdb7 | 1526 | &cpuset_attach_nodemask_to); |
f047cecf | 1527 | } |
4225399a PJ |
1528 | mmput(mm); |
1529 | } | |
452477fa | 1530 | |
33ad801d | 1531 | cs->old_mems_allowed = cpuset_attach_nodemask_to; |
02bb5863 | 1532 | |
452477fa | 1533 | cs->attach_in_progress--; |
e44193d3 LZ |
1534 | if (!cs->attach_in_progress) |
1535 | wake_up(&cpuset_attach_wq); | |
5d21cc2d TH |
1536 | |
1537 | mutex_unlock(&cpuset_mutex); | |
1da177e4 LT |
1538 | } |
1539 | ||
1540 | /* The various types of files and directories in a cpuset file system */ | |
1541 | ||
1542 | typedef enum { | |
45b07ef3 | 1543 | FILE_MEMORY_MIGRATE, |
1da177e4 LT |
1544 | FILE_CPULIST, |
1545 | FILE_MEMLIST, | |
afd1a8b3 LZ |
1546 | FILE_EFFECTIVE_CPULIST, |
1547 | FILE_EFFECTIVE_MEMLIST, | |
1da177e4 LT |
1548 | FILE_CPU_EXCLUSIVE, |
1549 | FILE_MEM_EXCLUSIVE, | |
78608366 | 1550 | FILE_MEM_HARDWALL, |
029190c5 | 1551 | FILE_SCHED_LOAD_BALANCE, |
1d3504fc | 1552 | FILE_SCHED_RELAX_DOMAIN_LEVEL, |
3e0d98b9 PJ |
1553 | FILE_MEMORY_PRESSURE_ENABLED, |
1554 | FILE_MEMORY_PRESSURE, | |
825a46af PJ |
1555 | FILE_SPREAD_PAGE, |
1556 | FILE_SPREAD_SLAB, | |
1da177e4 LT |
1557 | } cpuset_filetype_t; |
1558 | ||
182446d0 TH |
1559 | static int cpuset_write_u64(struct cgroup_subsys_state *css, struct cftype *cft, |
1560 | u64 val) | |
700fe1ab | 1561 | { |
182446d0 | 1562 | struct cpuset *cs = css_cs(css); |
700fe1ab | 1563 | cpuset_filetype_t type = cft->private; |
a903f086 | 1564 | int retval = 0; |
700fe1ab | 1565 | |
5d21cc2d | 1566 | mutex_lock(&cpuset_mutex); |
a903f086 LZ |
1567 | if (!is_cpuset_online(cs)) { |
1568 | retval = -ENODEV; | |
5d21cc2d | 1569 | goto out_unlock; |
a903f086 | 1570 | } |
700fe1ab PM |
1571 | |
1572 | switch (type) { | |
1da177e4 | 1573 | case FILE_CPU_EXCLUSIVE: |
700fe1ab | 1574 | retval = update_flag(CS_CPU_EXCLUSIVE, cs, val); |
1da177e4 LT |
1575 | break; |
1576 | case FILE_MEM_EXCLUSIVE: | |
700fe1ab | 1577 | retval = update_flag(CS_MEM_EXCLUSIVE, cs, val); |
1da177e4 | 1578 | break; |
78608366 PM |
1579 | case FILE_MEM_HARDWALL: |
1580 | retval = update_flag(CS_MEM_HARDWALL, cs, val); | |
1581 | break; | |
029190c5 | 1582 | case FILE_SCHED_LOAD_BALANCE: |
700fe1ab | 1583 | retval = update_flag(CS_SCHED_LOAD_BALANCE, cs, val); |
1d3504fc | 1584 | break; |
45b07ef3 | 1585 | case FILE_MEMORY_MIGRATE: |
700fe1ab | 1586 | retval = update_flag(CS_MEMORY_MIGRATE, cs, val); |
45b07ef3 | 1587 | break; |
3e0d98b9 | 1588 | case FILE_MEMORY_PRESSURE_ENABLED: |
700fe1ab | 1589 | cpuset_memory_pressure_enabled = !!val; |
3e0d98b9 PJ |
1590 | break; |
1591 | case FILE_MEMORY_PRESSURE: | |
1592 | retval = -EACCES; | |
1593 | break; | |
825a46af | 1594 | case FILE_SPREAD_PAGE: |
700fe1ab | 1595 | retval = update_flag(CS_SPREAD_PAGE, cs, val); |
825a46af PJ |
1596 | break; |
1597 | case FILE_SPREAD_SLAB: | |
700fe1ab | 1598 | retval = update_flag(CS_SPREAD_SLAB, cs, val); |
825a46af | 1599 | break; |
1da177e4 LT |
1600 | default: |
1601 | retval = -EINVAL; | |
700fe1ab | 1602 | break; |
1da177e4 | 1603 | } |
5d21cc2d TH |
1604 | out_unlock: |
1605 | mutex_unlock(&cpuset_mutex); | |
1da177e4 LT |
1606 | return retval; |
1607 | } | |
1608 | ||
182446d0 TH |
1609 | static int cpuset_write_s64(struct cgroup_subsys_state *css, struct cftype *cft, |
1610 | s64 val) | |
5be7a479 | 1611 | { |
182446d0 | 1612 | struct cpuset *cs = css_cs(css); |
5be7a479 | 1613 | cpuset_filetype_t type = cft->private; |
5d21cc2d | 1614 | int retval = -ENODEV; |
5be7a479 | 1615 | |
5d21cc2d TH |
1616 | mutex_lock(&cpuset_mutex); |
1617 | if (!is_cpuset_online(cs)) | |
1618 | goto out_unlock; | |
e3712395 | 1619 | |
5be7a479 PM |
1620 | switch (type) { |
1621 | case FILE_SCHED_RELAX_DOMAIN_LEVEL: | |
1622 | retval = update_relax_domain_level(cs, val); | |
1623 | break; | |
1624 | default: | |
1625 | retval = -EINVAL; | |
1626 | break; | |
1627 | } | |
5d21cc2d TH |
1628 | out_unlock: |
1629 | mutex_unlock(&cpuset_mutex); | |
5be7a479 PM |
1630 | return retval; |
1631 | } | |
1632 | ||
e3712395 PM |
1633 | /* |
1634 | * Common handling for a write to a "cpus" or "mems" file. | |
1635 | */ | |
451af504 TH |
1636 | static ssize_t cpuset_write_resmask(struct kernfs_open_file *of, |
1637 | char *buf, size_t nbytes, loff_t off) | |
e3712395 | 1638 | { |
451af504 | 1639 | struct cpuset *cs = css_cs(of_css(of)); |
645fcc9d | 1640 | struct cpuset *trialcs; |
5d21cc2d | 1641 | int retval = -ENODEV; |
e3712395 | 1642 | |
451af504 TH |
1643 | buf = strstrip(buf); |
1644 | ||
3a5a6d0c TH |
1645 | /* |
1646 | * CPU or memory hotunplug may leave @cs w/o any execution | |
1647 | * resources, in which case the hotplug code asynchronously updates | |
1648 | * configuration and transfers all tasks to the nearest ancestor | |
1649 | * which can execute. | |
1650 | * | |
1651 | * As writes to "cpus" or "mems" may restore @cs's execution | |
1652 | * resources, wait for the previously scheduled operations before | |
1653 | * proceeding, so that we don't end up keep removing tasks added | |
1654 | * after execution capability is restored. | |
76bb5ab8 TH |
1655 | * |
1656 | * cpuset_hotplug_work calls back into cgroup core via | |
1657 | * cgroup_transfer_tasks() and waiting for it from a cgroupfs | |
1658 | * operation like this one can lead to a deadlock through kernfs | |
1659 | * active_ref protection. Let's break the protection. Losing the | |
1660 | * protection is okay as we check whether @cs is online after | |
1661 | * grabbing cpuset_mutex anyway. This only happens on the legacy | |
1662 | * hierarchies. | |
3a5a6d0c | 1663 | */ |
76bb5ab8 TH |
1664 | css_get(&cs->css); |
1665 | kernfs_break_active_protection(of->kn); | |
3a5a6d0c TH |
1666 | flush_work(&cpuset_hotplug_work); |
1667 | ||
5d21cc2d TH |
1668 | mutex_lock(&cpuset_mutex); |
1669 | if (!is_cpuset_online(cs)) | |
1670 | goto out_unlock; | |
e3712395 | 1671 | |
645fcc9d | 1672 | trialcs = alloc_trial_cpuset(cs); |
b75f38d6 LZ |
1673 | if (!trialcs) { |
1674 | retval = -ENOMEM; | |
5d21cc2d | 1675 | goto out_unlock; |
b75f38d6 | 1676 | } |
645fcc9d | 1677 | |
451af504 | 1678 | switch (of_cft(of)->private) { |
e3712395 | 1679 | case FILE_CPULIST: |
645fcc9d | 1680 | retval = update_cpumask(cs, trialcs, buf); |
e3712395 PM |
1681 | break; |
1682 | case FILE_MEMLIST: | |
645fcc9d | 1683 | retval = update_nodemask(cs, trialcs, buf); |
e3712395 PM |
1684 | break; |
1685 | default: | |
1686 | retval = -EINVAL; | |
1687 | break; | |
1688 | } | |
645fcc9d LZ |
1689 | |
1690 | free_trial_cpuset(trialcs); | |
5d21cc2d TH |
1691 | out_unlock: |
1692 | mutex_unlock(&cpuset_mutex); | |
76bb5ab8 TH |
1693 | kernfs_unbreak_active_protection(of->kn); |
1694 | css_put(&cs->css); | |
451af504 | 1695 | return retval ?: nbytes; |
e3712395 PM |
1696 | } |
1697 | ||
1da177e4 LT |
1698 | /* |
1699 | * These ascii lists should be read in a single call, by using a user | |
1700 | * buffer large enough to hold the entire map. If read in smaller | |
1701 | * chunks, there is no guarantee of atomicity. Since the display format | |
1702 | * used, list of ranges of sequential numbers, is variable length, | |
1703 | * and since these maps can change value dynamically, one could read | |
1704 | * gibberish by doing partial reads while a list was changing. | |
1da177e4 | 1705 | */ |
2da8ca82 | 1706 | static int cpuset_common_seq_show(struct seq_file *sf, void *v) |
1da177e4 | 1707 | { |
2da8ca82 TH |
1708 | struct cpuset *cs = css_cs(seq_css(sf)); |
1709 | cpuset_filetype_t type = seq_cft(sf)->private; | |
51ffe411 TH |
1710 | ssize_t count; |
1711 | char *buf, *s; | |
1712 | int ret = 0; | |
1da177e4 | 1713 | |
51ffe411 TH |
1714 | count = seq_get_buf(sf, &buf); |
1715 | s = buf; | |
1da177e4 | 1716 | |
51ffe411 | 1717 | mutex_lock(&callback_mutex); |
1da177e4 LT |
1718 | |
1719 | switch (type) { | |
1720 | case FILE_CPULIST: | |
51ffe411 | 1721 | s += cpulist_scnprintf(s, count, cs->cpus_allowed); |
1da177e4 LT |
1722 | break; |
1723 | case FILE_MEMLIST: | |
51ffe411 | 1724 | s += nodelist_scnprintf(s, count, cs->mems_allowed); |
1da177e4 | 1725 | break; |
afd1a8b3 LZ |
1726 | case FILE_EFFECTIVE_CPULIST: |
1727 | s += cpulist_scnprintf(s, count, cs->effective_cpus); | |
1728 | break; | |
1729 | case FILE_EFFECTIVE_MEMLIST: | |
1730 | s += nodelist_scnprintf(s, count, cs->effective_mems); | |
1731 | break; | |
1da177e4 | 1732 | default: |
51ffe411 TH |
1733 | ret = -EINVAL; |
1734 | goto out_unlock; | |
1da177e4 | 1735 | } |
1da177e4 | 1736 | |
51ffe411 TH |
1737 | if (s < buf + count - 1) { |
1738 | *s++ = '\n'; | |
1739 | seq_commit(sf, s - buf); | |
1740 | } else { | |
1741 | seq_commit(sf, -1); | |
1742 | } | |
1743 | out_unlock: | |
1744 | mutex_unlock(&callback_mutex); | |
1745 | return ret; | |
1da177e4 LT |
1746 | } |
1747 | ||
182446d0 | 1748 | static u64 cpuset_read_u64(struct cgroup_subsys_state *css, struct cftype *cft) |
700fe1ab | 1749 | { |
182446d0 | 1750 | struct cpuset *cs = css_cs(css); |
700fe1ab PM |
1751 | cpuset_filetype_t type = cft->private; |
1752 | switch (type) { | |
1753 | case FILE_CPU_EXCLUSIVE: | |
1754 | return is_cpu_exclusive(cs); | |
1755 | case FILE_MEM_EXCLUSIVE: | |
1756 | return is_mem_exclusive(cs); | |
78608366 PM |
1757 | case FILE_MEM_HARDWALL: |
1758 | return is_mem_hardwall(cs); | |
700fe1ab PM |
1759 | case FILE_SCHED_LOAD_BALANCE: |
1760 | return is_sched_load_balance(cs); | |
1761 | case FILE_MEMORY_MIGRATE: | |
1762 | return is_memory_migrate(cs); | |
1763 | case FILE_MEMORY_PRESSURE_ENABLED: | |
1764 | return cpuset_memory_pressure_enabled; | |
1765 | case FILE_MEMORY_PRESSURE: | |
1766 | return fmeter_getrate(&cs->fmeter); | |
1767 | case FILE_SPREAD_PAGE: | |
1768 | return is_spread_page(cs); | |
1769 | case FILE_SPREAD_SLAB: | |
1770 | return is_spread_slab(cs); | |
1771 | default: | |
1772 | BUG(); | |
1773 | } | |
cf417141 MK |
1774 | |
1775 | /* Unreachable but makes gcc happy */ | |
1776 | return 0; | |
700fe1ab | 1777 | } |
1da177e4 | 1778 | |
182446d0 | 1779 | static s64 cpuset_read_s64(struct cgroup_subsys_state *css, struct cftype *cft) |
5be7a479 | 1780 | { |
182446d0 | 1781 | struct cpuset *cs = css_cs(css); |
5be7a479 PM |
1782 | cpuset_filetype_t type = cft->private; |
1783 | switch (type) { | |
1784 | case FILE_SCHED_RELAX_DOMAIN_LEVEL: | |
1785 | return cs->relax_domain_level; | |
1786 | default: | |
1787 | BUG(); | |
1788 | } | |
cf417141 MK |
1789 | |
1790 | /* Unrechable but makes gcc happy */ | |
1791 | return 0; | |
5be7a479 PM |
1792 | } |
1793 | ||
1da177e4 LT |
1794 | |
1795 | /* | |
1796 | * for the common functions, 'private' gives the type of file | |
1797 | */ | |
1798 | ||
addf2c73 PM |
1799 | static struct cftype files[] = { |
1800 | { | |
1801 | .name = "cpus", | |
2da8ca82 | 1802 | .seq_show = cpuset_common_seq_show, |
451af504 | 1803 | .write = cpuset_write_resmask, |
e3712395 | 1804 | .max_write_len = (100U + 6 * NR_CPUS), |
addf2c73 PM |
1805 | .private = FILE_CPULIST, |
1806 | }, | |
1807 | ||
1808 | { | |
1809 | .name = "mems", | |
2da8ca82 | 1810 | .seq_show = cpuset_common_seq_show, |
451af504 | 1811 | .write = cpuset_write_resmask, |
e3712395 | 1812 | .max_write_len = (100U + 6 * MAX_NUMNODES), |
addf2c73 PM |
1813 | .private = FILE_MEMLIST, |
1814 | }, | |
1815 | ||
afd1a8b3 LZ |
1816 | { |
1817 | .name = "effective_cpus", | |
1818 | .seq_show = cpuset_common_seq_show, | |
1819 | .private = FILE_EFFECTIVE_CPULIST, | |
1820 | }, | |
1821 | ||
1822 | { | |
1823 | .name = "effective_mems", | |
1824 | .seq_show = cpuset_common_seq_show, | |
1825 | .private = FILE_EFFECTIVE_MEMLIST, | |
1826 | }, | |
1827 | ||
addf2c73 PM |
1828 | { |
1829 | .name = "cpu_exclusive", | |
1830 | .read_u64 = cpuset_read_u64, | |
1831 | .write_u64 = cpuset_write_u64, | |
1832 | .private = FILE_CPU_EXCLUSIVE, | |
1833 | }, | |
1834 | ||
1835 | { | |
1836 | .name = "mem_exclusive", | |
1837 | .read_u64 = cpuset_read_u64, | |
1838 | .write_u64 = cpuset_write_u64, | |
1839 | .private = FILE_MEM_EXCLUSIVE, | |
1840 | }, | |
1841 | ||
78608366 PM |
1842 | { |
1843 | .name = "mem_hardwall", | |
1844 | .read_u64 = cpuset_read_u64, | |
1845 | .write_u64 = cpuset_write_u64, | |
1846 | .private = FILE_MEM_HARDWALL, | |
1847 | }, | |
1848 | ||
addf2c73 PM |
1849 | { |
1850 | .name = "sched_load_balance", | |
1851 | .read_u64 = cpuset_read_u64, | |
1852 | .write_u64 = cpuset_write_u64, | |
1853 | .private = FILE_SCHED_LOAD_BALANCE, | |
1854 | }, | |
1855 | ||
1856 | { | |
1857 | .name = "sched_relax_domain_level", | |
5be7a479 PM |
1858 | .read_s64 = cpuset_read_s64, |
1859 | .write_s64 = cpuset_write_s64, | |
addf2c73 PM |
1860 | .private = FILE_SCHED_RELAX_DOMAIN_LEVEL, |
1861 | }, | |
1862 | ||
1863 | { | |
1864 | .name = "memory_migrate", | |
1865 | .read_u64 = cpuset_read_u64, | |
1866 | .write_u64 = cpuset_write_u64, | |
1867 | .private = FILE_MEMORY_MIGRATE, | |
1868 | }, | |
1869 | ||
1870 | { | |
1871 | .name = "memory_pressure", | |
1872 | .read_u64 = cpuset_read_u64, | |
1873 | .write_u64 = cpuset_write_u64, | |
1874 | .private = FILE_MEMORY_PRESSURE, | |
099fca32 | 1875 | .mode = S_IRUGO, |
addf2c73 PM |
1876 | }, |
1877 | ||
1878 | { | |
1879 | .name = "memory_spread_page", | |
1880 | .read_u64 = cpuset_read_u64, | |
1881 | .write_u64 = cpuset_write_u64, | |
1882 | .private = FILE_SPREAD_PAGE, | |
1883 | }, | |
1884 | ||
1885 | { | |
1886 | .name = "memory_spread_slab", | |
1887 | .read_u64 = cpuset_read_u64, | |
1888 | .write_u64 = cpuset_write_u64, | |
1889 | .private = FILE_SPREAD_SLAB, | |
1890 | }, | |
3e0d98b9 | 1891 | |
4baf6e33 TH |
1892 | { |
1893 | .name = "memory_pressure_enabled", | |
1894 | .flags = CFTYPE_ONLY_ON_ROOT, | |
1895 | .read_u64 = cpuset_read_u64, | |
1896 | .write_u64 = cpuset_write_u64, | |
1897 | .private = FILE_MEMORY_PRESSURE_ENABLED, | |
1898 | }, | |
1da177e4 | 1899 | |
4baf6e33 TH |
1900 | { } /* terminate */ |
1901 | }; | |
1da177e4 LT |
1902 | |
1903 | /* | |
92fb9748 | 1904 | * cpuset_css_alloc - allocate a cpuset css |
c9e5fe66 | 1905 | * cgrp: control group that the new cpuset will be part of |
1da177e4 LT |
1906 | */ |
1907 | ||
eb95419b TH |
1908 | static struct cgroup_subsys_state * |
1909 | cpuset_css_alloc(struct cgroup_subsys_state *parent_css) | |
1da177e4 | 1910 | { |
c8f699bb | 1911 | struct cpuset *cs; |
1da177e4 | 1912 | |
eb95419b | 1913 | if (!parent_css) |
8793d854 | 1914 | return &top_cpuset.css; |
033fa1c5 | 1915 | |
c8f699bb | 1916 | cs = kzalloc(sizeof(*cs), GFP_KERNEL); |
1da177e4 | 1917 | if (!cs) |
8793d854 | 1918 | return ERR_PTR(-ENOMEM); |
e2b9a3d7 LZ |
1919 | if (!alloc_cpumask_var(&cs->cpus_allowed, GFP_KERNEL)) |
1920 | goto free_cs; | |
1921 | if (!alloc_cpumask_var(&cs->effective_cpus, GFP_KERNEL)) | |
1922 | goto free_cpus; | |
1da177e4 | 1923 | |
029190c5 | 1924 | set_bit(CS_SCHED_LOAD_BALANCE, &cs->flags); |
300ed6cb | 1925 | cpumask_clear(cs->cpus_allowed); |
f9a86fcb | 1926 | nodes_clear(cs->mems_allowed); |
e2b9a3d7 LZ |
1927 | cpumask_clear(cs->effective_cpus); |
1928 | nodes_clear(cs->effective_mems); | |
3e0d98b9 | 1929 | fmeter_init(&cs->fmeter); |
1d3504fc | 1930 | cs->relax_domain_level = -1; |
1da177e4 | 1931 | |
c8f699bb | 1932 | return &cs->css; |
e2b9a3d7 LZ |
1933 | |
1934 | free_cpus: | |
1935 | free_cpumask_var(cs->cpus_allowed); | |
1936 | free_cs: | |
1937 | kfree(cs); | |
1938 | return ERR_PTR(-ENOMEM); | |
c8f699bb TH |
1939 | } |
1940 | ||
eb95419b | 1941 | static int cpuset_css_online(struct cgroup_subsys_state *css) |
c8f699bb | 1942 | { |
eb95419b | 1943 | struct cpuset *cs = css_cs(css); |
c431069f | 1944 | struct cpuset *parent = parent_cs(cs); |
ae8086ce | 1945 | struct cpuset *tmp_cs; |
492eb21b | 1946 | struct cgroup_subsys_state *pos_css; |
c8f699bb TH |
1947 | |
1948 | if (!parent) | |
1949 | return 0; | |
1950 | ||
5d21cc2d TH |
1951 | mutex_lock(&cpuset_mutex); |
1952 | ||
efeb77b2 | 1953 | set_bit(CS_ONLINE, &cs->flags); |
c8f699bb TH |
1954 | if (is_spread_page(parent)) |
1955 | set_bit(CS_SPREAD_PAGE, &cs->flags); | |
1956 | if (is_spread_slab(parent)) | |
1957 | set_bit(CS_SPREAD_SLAB, &cs->flags); | |
1da177e4 | 1958 | |
664eedde | 1959 | cpuset_inc(); |
033fa1c5 | 1960 | |
e2b9a3d7 LZ |
1961 | mutex_lock(&callback_mutex); |
1962 | if (cgroup_on_dfl(cs->css.cgroup)) { | |
1963 | cpumask_copy(cs->effective_cpus, parent->effective_cpus); | |
1964 | cs->effective_mems = parent->effective_mems; | |
1965 | } | |
1966 | mutex_unlock(&callback_mutex); | |
1967 | ||
eb95419b | 1968 | if (!test_bit(CGRP_CPUSET_CLONE_CHILDREN, &css->cgroup->flags)) |
5d21cc2d | 1969 | goto out_unlock; |
033fa1c5 TH |
1970 | |
1971 | /* | |
1972 | * Clone @parent's configuration if CGRP_CPUSET_CLONE_CHILDREN is | |
1973 | * set. This flag handling is implemented in cgroup core for | |
1974 | * histrical reasons - the flag may be specified during mount. | |
1975 | * | |
1976 | * Currently, if any sibling cpusets have exclusive cpus or mem, we | |
1977 | * refuse to clone the configuration - thereby refusing the task to | |
1978 | * be entered, and as a result refusing the sys_unshare() or | |
1979 | * clone() which initiated it. If this becomes a problem for some | |
1980 | * users who wish to allow that scenario, then this could be | |
1981 | * changed to grant parent->cpus_allowed-sibling_cpus_exclusive | |
1982 | * (and likewise for mems) to the new cgroup. | |
1983 | */ | |
ae8086ce | 1984 | rcu_read_lock(); |
492eb21b | 1985 | cpuset_for_each_child(tmp_cs, pos_css, parent) { |
ae8086ce TH |
1986 | if (is_mem_exclusive(tmp_cs) || is_cpu_exclusive(tmp_cs)) { |
1987 | rcu_read_unlock(); | |
5d21cc2d | 1988 | goto out_unlock; |
ae8086ce | 1989 | } |
033fa1c5 | 1990 | } |
ae8086ce | 1991 | rcu_read_unlock(); |
033fa1c5 TH |
1992 | |
1993 | mutex_lock(&callback_mutex); | |
1994 | cs->mems_allowed = parent->mems_allowed; | |
1995 | cpumask_copy(cs->cpus_allowed, parent->cpus_allowed); | |
1996 | mutex_unlock(&callback_mutex); | |
5d21cc2d TH |
1997 | out_unlock: |
1998 | mutex_unlock(&cpuset_mutex); | |
c8f699bb TH |
1999 | return 0; |
2000 | } | |
2001 | ||
0b9e6965 ZH |
2002 | /* |
2003 | * If the cpuset being removed has its flag 'sched_load_balance' | |
2004 | * enabled, then simulate turning sched_load_balance off, which | |
2005 | * will call rebuild_sched_domains_locked(). | |
2006 | */ | |
2007 | ||
eb95419b | 2008 | static void cpuset_css_offline(struct cgroup_subsys_state *css) |
c8f699bb | 2009 | { |
eb95419b | 2010 | struct cpuset *cs = css_cs(css); |
c8f699bb | 2011 | |
5d21cc2d | 2012 | mutex_lock(&cpuset_mutex); |
c8f699bb TH |
2013 | |
2014 | if (is_sched_load_balance(cs)) | |
2015 | update_flag(CS_SCHED_LOAD_BALANCE, cs, 0); | |
2016 | ||
664eedde | 2017 | cpuset_dec(); |
efeb77b2 | 2018 | clear_bit(CS_ONLINE, &cs->flags); |
c8f699bb | 2019 | |
5d21cc2d | 2020 | mutex_unlock(&cpuset_mutex); |
1da177e4 LT |
2021 | } |
2022 | ||
eb95419b | 2023 | static void cpuset_css_free(struct cgroup_subsys_state *css) |
1da177e4 | 2024 | { |
eb95419b | 2025 | struct cpuset *cs = css_cs(css); |
1da177e4 | 2026 | |
e2b9a3d7 | 2027 | free_cpumask_var(cs->effective_cpus); |
300ed6cb | 2028 | free_cpumask_var(cs->cpus_allowed); |
8793d854 | 2029 | kfree(cs); |
1da177e4 LT |
2030 | } |
2031 | ||
39bd0d15 LZ |
2032 | static void cpuset_bind(struct cgroup_subsys_state *root_css) |
2033 | { | |
2034 | mutex_lock(&cpuset_mutex); | |
2035 | mutex_lock(&callback_mutex); | |
2036 | ||
2037 | if (cgroup_on_dfl(root_css->cgroup)) { | |
2038 | cpumask_copy(top_cpuset.cpus_allowed, cpu_possible_mask); | |
2039 | top_cpuset.mems_allowed = node_possible_map; | |
2040 | } else { | |
2041 | cpumask_copy(top_cpuset.cpus_allowed, | |
2042 | top_cpuset.effective_cpus); | |
2043 | top_cpuset.mems_allowed = top_cpuset.effective_mems; | |
2044 | } | |
2045 | ||
2046 | mutex_unlock(&callback_mutex); | |
2047 | mutex_unlock(&cpuset_mutex); | |
2048 | } | |
2049 | ||
073219e9 | 2050 | struct cgroup_subsys cpuset_cgrp_subsys = { |
39bd0d15 LZ |
2051 | .css_alloc = cpuset_css_alloc, |
2052 | .css_online = cpuset_css_online, | |
2053 | .css_offline = cpuset_css_offline, | |
2054 | .css_free = cpuset_css_free, | |
2055 | .can_attach = cpuset_can_attach, | |
2056 | .cancel_attach = cpuset_cancel_attach, | |
2057 | .attach = cpuset_attach, | |
2058 | .bind = cpuset_bind, | |
5577964e | 2059 | .legacy_cftypes = files, |
39bd0d15 | 2060 | .early_init = 1, |
8793d854 PM |
2061 | }; |
2062 | ||
1da177e4 LT |
2063 | /** |
2064 | * cpuset_init - initialize cpusets at system boot | |
2065 | * | |
2066 | * Description: Initialize top_cpuset and the cpuset internal file system, | |
2067 | **/ | |
2068 | ||
2069 | int __init cpuset_init(void) | |
2070 | { | |
8793d854 | 2071 | int err = 0; |
1da177e4 | 2072 | |
58568d2a MX |
2073 | if (!alloc_cpumask_var(&top_cpuset.cpus_allowed, GFP_KERNEL)) |
2074 | BUG(); | |
e2b9a3d7 LZ |
2075 | if (!alloc_cpumask_var(&top_cpuset.effective_cpus, GFP_KERNEL)) |
2076 | BUG(); | |
58568d2a | 2077 | |
300ed6cb | 2078 | cpumask_setall(top_cpuset.cpus_allowed); |
f9a86fcb | 2079 | nodes_setall(top_cpuset.mems_allowed); |
e2b9a3d7 LZ |
2080 | cpumask_setall(top_cpuset.effective_cpus); |
2081 | nodes_setall(top_cpuset.effective_mems); | |
1da177e4 | 2082 | |
3e0d98b9 | 2083 | fmeter_init(&top_cpuset.fmeter); |
029190c5 | 2084 | set_bit(CS_SCHED_LOAD_BALANCE, &top_cpuset.flags); |
1d3504fc | 2085 | top_cpuset.relax_domain_level = -1; |
1da177e4 | 2086 | |
1da177e4 LT |
2087 | err = register_filesystem(&cpuset_fs_type); |
2088 | if (err < 0) | |
8793d854 PM |
2089 | return err; |
2090 | ||
2341d1b6 LZ |
2091 | if (!alloc_cpumask_var(&cpus_attach, GFP_KERNEL)) |
2092 | BUG(); | |
2093 | ||
8793d854 | 2094 | return 0; |
1da177e4 LT |
2095 | } |
2096 | ||
b1aac8bb | 2097 | /* |
cf417141 | 2098 | * If CPU and/or memory hotplug handlers, below, unplug any CPUs |
b1aac8bb PJ |
2099 | * or memory nodes, we need to walk over the cpuset hierarchy, |
2100 | * removing that CPU or node from all cpusets. If this removes the | |
956db3ca CW |
2101 | * last CPU or node from a cpuset, then move the tasks in the empty |
2102 | * cpuset to its next-highest non-empty parent. | |
b1aac8bb | 2103 | */ |
956db3ca CW |
2104 | static void remove_tasks_in_empty_cpuset(struct cpuset *cs) |
2105 | { | |
2106 | struct cpuset *parent; | |
2107 | ||
956db3ca CW |
2108 | /* |
2109 | * Find its next-highest non-empty parent, (top cpuset | |
2110 | * has online cpus, so can't be empty). | |
2111 | */ | |
c431069f | 2112 | parent = parent_cs(cs); |
300ed6cb | 2113 | while (cpumask_empty(parent->cpus_allowed) || |
b4501295 | 2114 | nodes_empty(parent->mems_allowed)) |
c431069f | 2115 | parent = parent_cs(parent); |
956db3ca | 2116 | |
8cc99345 | 2117 | if (cgroup_transfer_tasks(parent->css.cgroup, cs->css.cgroup)) { |
12d3089c | 2118 | pr_err("cpuset: failed to transfer tasks out of empty cpuset "); |
e61734c5 TH |
2119 | pr_cont_cgroup_name(cs->css.cgroup); |
2120 | pr_cont("\n"); | |
8cc99345 | 2121 | } |
956db3ca CW |
2122 | } |
2123 | ||
be4c9dd7 LZ |
2124 | static void |
2125 | hotplug_update_tasks_legacy(struct cpuset *cs, | |
2126 | struct cpumask *new_cpus, nodemask_t *new_mems, | |
2127 | bool cpus_updated, bool mems_updated) | |
390a36aa LZ |
2128 | { |
2129 | bool is_empty; | |
2130 | ||
2131 | mutex_lock(&callback_mutex); | |
be4c9dd7 LZ |
2132 | cpumask_copy(cs->cpus_allowed, new_cpus); |
2133 | cpumask_copy(cs->effective_cpus, new_cpus); | |
2134 | cs->mems_allowed = *new_mems; | |
2135 | cs->effective_mems = *new_mems; | |
390a36aa LZ |
2136 | mutex_unlock(&callback_mutex); |
2137 | ||
2138 | /* | |
2139 | * Don't call update_tasks_cpumask() if the cpuset becomes empty, | |
2140 | * as the tasks will be migratecd to an ancestor. | |
2141 | */ | |
be4c9dd7 | 2142 | if (cpus_updated && !cpumask_empty(cs->cpus_allowed)) |
390a36aa | 2143 | update_tasks_cpumask(cs); |
be4c9dd7 | 2144 | if (mems_updated && !nodes_empty(cs->mems_allowed)) |
390a36aa LZ |
2145 | update_tasks_nodemask(cs); |
2146 | ||
2147 | is_empty = cpumask_empty(cs->cpus_allowed) || | |
2148 | nodes_empty(cs->mems_allowed); | |
2149 | ||
2150 | mutex_unlock(&cpuset_mutex); | |
2151 | ||
2152 | /* | |
2153 | * Move tasks to the nearest ancestor with execution resources, | |
2154 | * This is full cgroup operation which will also call back into | |
2155 | * cpuset. Should be done outside any lock. | |
2156 | */ | |
2157 | if (is_empty) | |
2158 | remove_tasks_in_empty_cpuset(cs); | |
2159 | ||
2160 | mutex_lock(&cpuset_mutex); | |
2161 | } | |
2162 | ||
be4c9dd7 LZ |
2163 | static void |
2164 | hotplug_update_tasks(struct cpuset *cs, | |
2165 | struct cpumask *new_cpus, nodemask_t *new_mems, | |
2166 | bool cpus_updated, bool mems_updated) | |
390a36aa | 2167 | { |
be4c9dd7 LZ |
2168 | if (cpumask_empty(new_cpus)) |
2169 | cpumask_copy(new_cpus, parent_cs(cs)->effective_cpus); | |
2170 | if (nodes_empty(*new_mems)) | |
2171 | *new_mems = parent_cs(cs)->effective_mems; | |
2172 | ||
390a36aa | 2173 | mutex_lock(&callback_mutex); |
be4c9dd7 LZ |
2174 | cpumask_copy(cs->effective_cpus, new_cpus); |
2175 | cs->effective_mems = *new_mems; | |
390a36aa LZ |
2176 | mutex_unlock(&callback_mutex); |
2177 | ||
be4c9dd7 | 2178 | if (cpus_updated) |
390a36aa | 2179 | update_tasks_cpumask(cs); |
be4c9dd7 | 2180 | if (mems_updated) |
390a36aa LZ |
2181 | update_tasks_nodemask(cs); |
2182 | } | |
2183 | ||
deb7aa30 | 2184 | /** |
388afd85 | 2185 | * cpuset_hotplug_update_tasks - update tasks in a cpuset for hotunplug |
deb7aa30 | 2186 | * @cs: cpuset in interest |
956db3ca | 2187 | * |
deb7aa30 TH |
2188 | * Compare @cs's cpu and mem masks against top_cpuset and if some have gone |
2189 | * offline, update @cs accordingly. If @cs ends up with no CPU or memory, | |
2190 | * all its tasks are moved to the nearest ancestor with both resources. | |
80d1fa64 | 2191 | */ |
388afd85 | 2192 | static void cpuset_hotplug_update_tasks(struct cpuset *cs) |
80d1fa64 | 2193 | { |
be4c9dd7 LZ |
2194 | static cpumask_t new_cpus; |
2195 | static nodemask_t new_mems; | |
2196 | bool cpus_updated; | |
2197 | bool mems_updated; | |
e44193d3 LZ |
2198 | retry: |
2199 | wait_event(cpuset_attach_wq, cs->attach_in_progress == 0); | |
80d1fa64 | 2200 | |
5d21cc2d | 2201 | mutex_lock(&cpuset_mutex); |
7ddf96b0 | 2202 | |
e44193d3 LZ |
2203 | /* |
2204 | * We have raced with task attaching. We wait until attaching | |
2205 | * is finished, so we won't attach a task to an empty cpuset. | |
2206 | */ | |
2207 | if (cs->attach_in_progress) { | |
2208 | mutex_unlock(&cpuset_mutex); | |
2209 | goto retry; | |
2210 | } | |
2211 | ||
be4c9dd7 LZ |
2212 | cpumask_and(&new_cpus, cs->cpus_allowed, parent_cs(cs)->effective_cpus); |
2213 | nodes_and(new_mems, cs->mems_allowed, parent_cs(cs)->effective_mems); | |
80d1fa64 | 2214 | |
be4c9dd7 LZ |
2215 | cpus_updated = !cpumask_equal(&new_cpus, cs->effective_cpus); |
2216 | mems_updated = !nodes_equal(new_mems, cs->effective_mems); | |
deb7aa30 | 2217 | |
390a36aa | 2218 | if (cgroup_on_dfl(cs->css.cgroup)) |
be4c9dd7 LZ |
2219 | hotplug_update_tasks(cs, &new_cpus, &new_mems, |
2220 | cpus_updated, mems_updated); | |
390a36aa | 2221 | else |
be4c9dd7 LZ |
2222 | hotplug_update_tasks_legacy(cs, &new_cpus, &new_mems, |
2223 | cpus_updated, mems_updated); | |
8d033948 | 2224 | |
5d21cc2d | 2225 | mutex_unlock(&cpuset_mutex); |
b1aac8bb PJ |
2226 | } |
2227 | ||
deb7aa30 | 2228 | /** |
3a5a6d0c | 2229 | * cpuset_hotplug_workfn - handle CPU/memory hotunplug for a cpuset |
956db3ca | 2230 | * |
deb7aa30 TH |
2231 | * This function is called after either CPU or memory configuration has |
2232 | * changed and updates cpuset accordingly. The top_cpuset is always | |
2233 | * synchronized to cpu_active_mask and N_MEMORY, which is necessary in | |
2234 | * order to make cpusets transparent (of no affect) on systems that are | |
2235 | * actively using CPU hotplug but making no active use of cpusets. | |
956db3ca | 2236 | * |
deb7aa30 | 2237 | * Non-root cpusets are only affected by offlining. If any CPUs or memory |
388afd85 LZ |
2238 | * nodes have been taken down, cpuset_hotplug_update_tasks() is invoked on |
2239 | * all descendants. | |
956db3ca | 2240 | * |
deb7aa30 TH |
2241 | * Note that CPU offlining during suspend is ignored. We don't modify |
2242 | * cpusets across suspend/resume cycles at all. | |
956db3ca | 2243 | */ |
3a5a6d0c | 2244 | static void cpuset_hotplug_workfn(struct work_struct *work) |
b1aac8bb | 2245 | { |
5c5cc623 LZ |
2246 | static cpumask_t new_cpus; |
2247 | static nodemask_t new_mems; | |
deb7aa30 | 2248 | bool cpus_updated, mems_updated; |
7e88291b | 2249 | bool on_dfl = cgroup_on_dfl(top_cpuset.css.cgroup); |
b1aac8bb | 2250 | |
5d21cc2d | 2251 | mutex_lock(&cpuset_mutex); |
956db3ca | 2252 | |
deb7aa30 TH |
2253 | /* fetch the available cpus/mems and find out which changed how */ |
2254 | cpumask_copy(&new_cpus, cpu_active_mask); | |
2255 | new_mems = node_states[N_MEMORY]; | |
7ddf96b0 | 2256 | |
7e88291b LZ |
2257 | cpus_updated = !cpumask_equal(top_cpuset.effective_cpus, &new_cpus); |
2258 | mems_updated = !nodes_equal(top_cpuset.effective_mems, new_mems); | |
7ddf96b0 | 2259 | |
deb7aa30 TH |
2260 | /* synchronize cpus_allowed to cpu_active_mask */ |
2261 | if (cpus_updated) { | |
2262 | mutex_lock(&callback_mutex); | |
7e88291b LZ |
2263 | if (!on_dfl) |
2264 | cpumask_copy(top_cpuset.cpus_allowed, &new_cpus); | |
1344ab9c | 2265 | cpumask_copy(top_cpuset.effective_cpus, &new_cpus); |
deb7aa30 TH |
2266 | mutex_unlock(&callback_mutex); |
2267 | /* we don't mess with cpumasks of tasks in top_cpuset */ | |
2268 | } | |
b4501295 | 2269 | |
deb7aa30 TH |
2270 | /* synchronize mems_allowed to N_MEMORY */ |
2271 | if (mems_updated) { | |
deb7aa30 | 2272 | mutex_lock(&callback_mutex); |
7e88291b LZ |
2273 | if (!on_dfl) |
2274 | top_cpuset.mems_allowed = new_mems; | |
1344ab9c | 2275 | top_cpuset.effective_mems = new_mems; |
deb7aa30 | 2276 | mutex_unlock(&callback_mutex); |
d66393e5 | 2277 | update_tasks_nodemask(&top_cpuset); |
deb7aa30 | 2278 | } |
b4501295 | 2279 | |
388afd85 LZ |
2280 | mutex_unlock(&cpuset_mutex); |
2281 | ||
5c5cc623 LZ |
2282 | /* if cpus or mems changed, we need to propagate to descendants */ |
2283 | if (cpus_updated || mems_updated) { | |
deb7aa30 | 2284 | struct cpuset *cs; |
492eb21b | 2285 | struct cgroup_subsys_state *pos_css; |
f9b4fb8d | 2286 | |
fc560a26 | 2287 | rcu_read_lock(); |
492eb21b | 2288 | cpuset_for_each_descendant_pre(cs, pos_css, &top_cpuset) { |
ec903c0c | 2289 | if (cs == &top_cpuset || !css_tryget_online(&cs->css)) |
388afd85 LZ |
2290 | continue; |
2291 | rcu_read_unlock(); | |
7ddf96b0 | 2292 | |
388afd85 | 2293 | cpuset_hotplug_update_tasks(cs); |
b4501295 | 2294 | |
388afd85 LZ |
2295 | rcu_read_lock(); |
2296 | css_put(&cs->css); | |
2297 | } | |
2298 | rcu_read_unlock(); | |
2299 | } | |
8d033948 | 2300 | |
deb7aa30 | 2301 | /* rebuild sched domains if cpus_allowed has changed */ |
e0e80a02 LZ |
2302 | if (cpus_updated) |
2303 | rebuild_sched_domains(); | |
b1aac8bb PJ |
2304 | } |
2305 | ||
7ddf96b0 | 2306 | void cpuset_update_active_cpus(bool cpu_online) |
4c4d50f7 | 2307 | { |
3a5a6d0c TH |
2308 | /* |
2309 | * We're inside cpu hotplug critical region which usually nests | |
2310 | * inside cgroup synchronization. Bounce actual hotplug processing | |
2311 | * to a work item to avoid reverse locking order. | |
2312 | * | |
2313 | * We still need to do partition_sched_domains() synchronously; | |
2314 | * otherwise, the scheduler will get confused and put tasks to the | |
2315 | * dead CPU. Fall back to the default single domain. | |
2316 | * cpuset_hotplug_workfn() will rebuild it as necessary. | |
2317 | */ | |
2318 | partition_sched_domains(1, NULL, NULL); | |
2319 | schedule_work(&cpuset_hotplug_work); | |
4c4d50f7 | 2320 | } |
4c4d50f7 | 2321 | |
38837fc7 | 2322 | /* |
38d7bee9 LJ |
2323 | * Keep top_cpuset.mems_allowed tracking node_states[N_MEMORY]. |
2324 | * Call this routine anytime after node_states[N_MEMORY] changes. | |
a1cd2b13 | 2325 | * See cpuset_update_active_cpus() for CPU hotplug handling. |
38837fc7 | 2326 | */ |
f481891f MX |
2327 | static int cpuset_track_online_nodes(struct notifier_block *self, |
2328 | unsigned long action, void *arg) | |
38837fc7 | 2329 | { |
3a5a6d0c | 2330 | schedule_work(&cpuset_hotplug_work); |
f481891f | 2331 | return NOTIFY_OK; |
38837fc7 | 2332 | } |
d8f10cb3 AM |
2333 | |
2334 | static struct notifier_block cpuset_track_online_nodes_nb = { | |
2335 | .notifier_call = cpuset_track_online_nodes, | |
2336 | .priority = 10, /* ??! */ | |
2337 | }; | |
38837fc7 | 2338 | |
1da177e4 LT |
2339 | /** |
2340 | * cpuset_init_smp - initialize cpus_allowed | |
2341 | * | |
2342 | * Description: Finish top cpuset after cpu, node maps are initialized | |
d8f10cb3 | 2343 | */ |
1da177e4 LT |
2344 | void __init cpuset_init_smp(void) |
2345 | { | |
6ad4c188 | 2346 | cpumask_copy(top_cpuset.cpus_allowed, cpu_active_mask); |
38d7bee9 | 2347 | top_cpuset.mems_allowed = node_states[N_MEMORY]; |
33ad801d | 2348 | top_cpuset.old_mems_allowed = top_cpuset.mems_allowed; |
4c4d50f7 | 2349 | |
e2b9a3d7 LZ |
2350 | cpumask_copy(top_cpuset.effective_cpus, cpu_active_mask); |
2351 | top_cpuset.effective_mems = node_states[N_MEMORY]; | |
2352 | ||
d8f10cb3 | 2353 | register_hotmemory_notifier(&cpuset_track_online_nodes_nb); |
1da177e4 LT |
2354 | } |
2355 | ||
2356 | /** | |
1da177e4 LT |
2357 | * cpuset_cpus_allowed - return cpus_allowed mask from a tasks cpuset. |
2358 | * @tsk: pointer to task_struct from which to obtain cpuset->cpus_allowed. | |
6af866af | 2359 | * @pmask: pointer to struct cpumask variable to receive cpus_allowed set. |
1da177e4 | 2360 | * |
300ed6cb | 2361 | * Description: Returns the cpumask_var_t cpus_allowed of the cpuset |
1da177e4 | 2362 | * attached to the specified @tsk. Guaranteed to return some non-empty |
5f054e31 | 2363 | * subset of cpu_online_mask, even if this means going outside the |
1da177e4 LT |
2364 | * tasks cpuset. |
2365 | **/ | |
2366 | ||
6af866af | 2367 | void cpuset_cpus_allowed(struct task_struct *tsk, struct cpumask *pmask) |
1da177e4 | 2368 | { |
3d3f26a7 | 2369 | mutex_lock(&callback_mutex); |
b8dadcb5 | 2370 | rcu_read_lock(); |
ae1c8023 | 2371 | guarantee_online_cpus(task_cs(tsk), pmask); |
b8dadcb5 | 2372 | rcu_read_unlock(); |
897f0b3c | 2373 | mutex_unlock(&callback_mutex); |
1da177e4 LT |
2374 | } |
2375 | ||
2baab4e9 | 2376 | void cpuset_cpus_allowed_fallback(struct task_struct *tsk) |
9084bb82 | 2377 | { |
9084bb82 | 2378 | rcu_read_lock(); |
ae1c8023 | 2379 | do_set_cpus_allowed(tsk, task_cs(tsk)->effective_cpus); |
9084bb82 ON |
2380 | rcu_read_unlock(); |
2381 | ||
2382 | /* | |
2383 | * We own tsk->cpus_allowed, nobody can change it under us. | |
2384 | * | |
2385 | * But we used cs && cs->cpus_allowed lockless and thus can | |
2386 | * race with cgroup_attach_task() or update_cpumask() and get | |
2387 | * the wrong tsk->cpus_allowed. However, both cases imply the | |
2388 | * subsequent cpuset_change_cpumask()->set_cpus_allowed_ptr() | |
2389 | * which takes task_rq_lock(). | |
2390 | * | |
2391 | * If we are called after it dropped the lock we must see all | |
2392 | * changes in tsk_cs()->cpus_allowed. Otherwise we can temporary | |
2393 | * set any mask even if it is not right from task_cs() pov, | |
2394 | * the pending set_cpus_allowed_ptr() will fix things. | |
2baab4e9 PZ |
2395 | * |
2396 | * select_fallback_rq() will fix things ups and set cpu_possible_mask | |
2397 | * if required. | |
9084bb82 | 2398 | */ |
9084bb82 ON |
2399 | } |
2400 | ||
1da177e4 LT |
2401 | void cpuset_init_current_mems_allowed(void) |
2402 | { | |
f9a86fcb | 2403 | nodes_setall(current->mems_allowed); |
1da177e4 LT |
2404 | } |
2405 | ||
909d75a3 PJ |
2406 | /** |
2407 | * cpuset_mems_allowed - return mems_allowed mask from a tasks cpuset. | |
2408 | * @tsk: pointer to task_struct from which to obtain cpuset->mems_allowed. | |
2409 | * | |
2410 | * Description: Returns the nodemask_t mems_allowed of the cpuset | |
2411 | * attached to the specified @tsk. Guaranteed to return some non-empty | |
38d7bee9 | 2412 | * subset of node_states[N_MEMORY], even if this means going outside the |
909d75a3 PJ |
2413 | * tasks cpuset. |
2414 | **/ | |
2415 | ||
2416 | nodemask_t cpuset_mems_allowed(struct task_struct *tsk) | |
2417 | { | |
2418 | nodemask_t mask; | |
2419 | ||
3d3f26a7 | 2420 | mutex_lock(&callback_mutex); |
b8dadcb5 | 2421 | rcu_read_lock(); |
ae1c8023 | 2422 | guarantee_online_mems(task_cs(tsk), &mask); |
b8dadcb5 | 2423 | rcu_read_unlock(); |
3d3f26a7 | 2424 | mutex_unlock(&callback_mutex); |
909d75a3 PJ |
2425 | |
2426 | return mask; | |
2427 | } | |
2428 | ||
d9fd8a6d | 2429 | /** |
19770b32 MG |
2430 | * cpuset_nodemask_valid_mems_allowed - check nodemask vs. curremt mems_allowed |
2431 | * @nodemask: the nodemask to be checked | |
d9fd8a6d | 2432 | * |
19770b32 | 2433 | * Are any of the nodes in the nodemask allowed in current->mems_allowed? |
1da177e4 | 2434 | */ |
19770b32 | 2435 | int cpuset_nodemask_valid_mems_allowed(nodemask_t *nodemask) |
1da177e4 | 2436 | { |
19770b32 | 2437 | return nodes_intersects(*nodemask, current->mems_allowed); |
1da177e4 LT |
2438 | } |
2439 | ||
9bf2229f | 2440 | /* |
78608366 PM |
2441 | * nearest_hardwall_ancestor() - Returns the nearest mem_exclusive or |
2442 | * mem_hardwall ancestor to the specified cpuset. Call holding | |
2443 | * callback_mutex. If no ancestor is mem_exclusive or mem_hardwall | |
2444 | * (an unusual configuration), then returns the root cpuset. | |
9bf2229f | 2445 | */ |
c9710d80 | 2446 | static struct cpuset *nearest_hardwall_ancestor(struct cpuset *cs) |
9bf2229f | 2447 | { |
c431069f TH |
2448 | while (!(is_mem_exclusive(cs) || is_mem_hardwall(cs)) && parent_cs(cs)) |
2449 | cs = parent_cs(cs); | |
9bf2229f PJ |
2450 | return cs; |
2451 | } | |
2452 | ||
d9fd8a6d | 2453 | /** |
a1bc5a4e DR |
2454 | * cpuset_node_allowed_softwall - Can we allocate on a memory node? |
2455 | * @node: is this an allowed node? | |
02a0e53d | 2456 | * @gfp_mask: memory allocation flags |
d9fd8a6d | 2457 | * |
a1bc5a4e DR |
2458 | * If we're in interrupt, yes, we can always allocate. If __GFP_THISNODE is |
2459 | * set, yes, we can always allocate. If node is in our task's mems_allowed, | |
2460 | * yes. If it's not a __GFP_HARDWALL request and this node is in the nearest | |
2461 | * hardwalled cpuset ancestor to this task's cpuset, yes. If the task has been | |
2462 | * OOM killed and has access to memory reserves as specified by the TIF_MEMDIE | |
2463 | * flag, yes. | |
9bf2229f PJ |
2464 | * Otherwise, no. |
2465 | * | |
a1bc5a4e DR |
2466 | * If __GFP_HARDWALL is set, cpuset_node_allowed_softwall() reduces to |
2467 | * cpuset_node_allowed_hardwall(). Otherwise, cpuset_node_allowed_softwall() | |
2468 | * might sleep, and might allow a node from an enclosing cpuset. | |
02a0e53d | 2469 | * |
a1bc5a4e DR |
2470 | * cpuset_node_allowed_hardwall() only handles the simpler case of hardwall |
2471 | * cpusets, and never sleeps. | |
02a0e53d PJ |
2472 | * |
2473 | * The __GFP_THISNODE placement logic is really handled elsewhere, | |
2474 | * by forcibly using a zonelist starting at a specified node, and by | |
2475 | * (in get_page_from_freelist()) refusing to consider the zones for | |
2476 | * any node on the zonelist except the first. By the time any such | |
2477 | * calls get to this routine, we should just shut up and say 'yes'. | |
2478 | * | |
9bf2229f | 2479 | * GFP_USER allocations are marked with the __GFP_HARDWALL bit, |
c596d9f3 DR |
2480 | * and do not allow allocations outside the current tasks cpuset |
2481 | * unless the task has been OOM killed as is marked TIF_MEMDIE. | |
9bf2229f | 2482 | * GFP_KERNEL allocations are not so marked, so can escape to the |
78608366 | 2483 | * nearest enclosing hardwalled ancestor cpuset. |
9bf2229f | 2484 | * |
02a0e53d PJ |
2485 | * Scanning up parent cpusets requires callback_mutex. The |
2486 | * __alloc_pages() routine only calls here with __GFP_HARDWALL bit | |
2487 | * _not_ set if it's a GFP_KERNEL allocation, and all nodes in the | |
2488 | * current tasks mems_allowed came up empty on the first pass over | |
2489 | * the zonelist. So only GFP_KERNEL allocations, if all nodes in the | |
2490 | * cpuset are short of memory, might require taking the callback_mutex | |
2491 | * mutex. | |
9bf2229f | 2492 | * |
36be57ff | 2493 | * The first call here from mm/page_alloc:get_page_from_freelist() |
02a0e53d PJ |
2494 | * has __GFP_HARDWALL set in gfp_mask, enforcing hardwall cpusets, |
2495 | * so no allocation on a node outside the cpuset is allowed (unless | |
2496 | * in interrupt, of course). | |
36be57ff PJ |
2497 | * |
2498 | * The second pass through get_page_from_freelist() doesn't even call | |
2499 | * here for GFP_ATOMIC calls. For those calls, the __alloc_pages() | |
2500 | * variable 'wait' is not set, and the bit ALLOC_CPUSET is not set | |
2501 | * in alloc_flags. That logic and the checks below have the combined | |
2502 | * affect that: | |
9bf2229f PJ |
2503 | * in_interrupt - any node ok (current task context irrelevant) |
2504 | * GFP_ATOMIC - any node ok | |
c596d9f3 | 2505 | * TIF_MEMDIE - any node ok |
78608366 | 2506 | * GFP_KERNEL - any node in enclosing hardwalled cpuset ok |
9bf2229f | 2507 | * GFP_USER - only nodes in current tasks mems allowed ok. |
36be57ff PJ |
2508 | * |
2509 | * Rule: | |
a1bc5a4e | 2510 | * Don't call cpuset_node_allowed_softwall if you can't sleep, unless you |
36be57ff PJ |
2511 | * pass in the __GFP_HARDWALL flag set in gfp_flag, which disables |
2512 | * the code that might scan up ancestor cpusets and sleep. | |
02a0e53d | 2513 | */ |
a1bc5a4e | 2514 | int __cpuset_node_allowed_softwall(int node, gfp_t gfp_mask) |
1da177e4 | 2515 | { |
c9710d80 | 2516 | struct cpuset *cs; /* current cpuset ancestors */ |
29afd49b | 2517 | int allowed; /* is allocation in zone z allowed? */ |
9bf2229f | 2518 | |
9b819d20 | 2519 | if (in_interrupt() || (gfp_mask & __GFP_THISNODE)) |
9bf2229f | 2520 | return 1; |
92d1dbd2 | 2521 | might_sleep_if(!(gfp_mask & __GFP_HARDWALL)); |
9bf2229f PJ |
2522 | if (node_isset(node, current->mems_allowed)) |
2523 | return 1; | |
c596d9f3 DR |
2524 | /* |
2525 | * Allow tasks that have access to memory reserves because they have | |
2526 | * been OOM killed to get memory anywhere. | |
2527 | */ | |
2528 | if (unlikely(test_thread_flag(TIF_MEMDIE))) | |
2529 | return 1; | |
9bf2229f PJ |
2530 | if (gfp_mask & __GFP_HARDWALL) /* If hardwall request, stop here */ |
2531 | return 0; | |
2532 | ||
5563e770 BP |
2533 | if (current->flags & PF_EXITING) /* Let dying task have memory */ |
2534 | return 1; | |
2535 | ||
9bf2229f | 2536 | /* Not hardwall and node outside mems_allowed: scan up cpusets */ |
3d3f26a7 | 2537 | mutex_lock(&callback_mutex); |
053199ed | 2538 | |
b8dadcb5 | 2539 | rcu_read_lock(); |
78608366 | 2540 | cs = nearest_hardwall_ancestor(task_cs(current)); |
99afb0fd | 2541 | allowed = node_isset(node, cs->mems_allowed); |
b8dadcb5 | 2542 | rcu_read_unlock(); |
053199ed | 2543 | |
3d3f26a7 | 2544 | mutex_unlock(&callback_mutex); |
9bf2229f | 2545 | return allowed; |
1da177e4 LT |
2546 | } |
2547 | ||
02a0e53d | 2548 | /* |
a1bc5a4e DR |
2549 | * cpuset_node_allowed_hardwall - Can we allocate on a memory node? |
2550 | * @node: is this an allowed node? | |
02a0e53d PJ |
2551 | * @gfp_mask: memory allocation flags |
2552 | * | |
a1bc5a4e DR |
2553 | * If we're in interrupt, yes, we can always allocate. If __GFP_THISNODE is |
2554 | * set, yes, we can always allocate. If node is in our task's mems_allowed, | |
2555 | * yes. If the task has been OOM killed and has access to memory reserves as | |
2556 | * specified by the TIF_MEMDIE flag, yes. | |
2557 | * Otherwise, no. | |
02a0e53d PJ |
2558 | * |
2559 | * The __GFP_THISNODE placement logic is really handled elsewhere, | |
2560 | * by forcibly using a zonelist starting at a specified node, and by | |
2561 | * (in get_page_from_freelist()) refusing to consider the zones for | |
2562 | * any node on the zonelist except the first. By the time any such | |
2563 | * calls get to this routine, we should just shut up and say 'yes'. | |
2564 | * | |
a1bc5a4e DR |
2565 | * Unlike the cpuset_node_allowed_softwall() variant, above, |
2566 | * this variant requires that the node be in the current task's | |
02a0e53d PJ |
2567 | * mems_allowed or that we're in interrupt. It does not scan up the |
2568 | * cpuset hierarchy for the nearest enclosing mem_exclusive cpuset. | |
2569 | * It never sleeps. | |
2570 | */ | |
a1bc5a4e | 2571 | int __cpuset_node_allowed_hardwall(int node, gfp_t gfp_mask) |
02a0e53d | 2572 | { |
02a0e53d PJ |
2573 | if (in_interrupt() || (gfp_mask & __GFP_THISNODE)) |
2574 | return 1; | |
02a0e53d PJ |
2575 | if (node_isset(node, current->mems_allowed)) |
2576 | return 1; | |
dedf8b79 DW |
2577 | /* |
2578 | * Allow tasks that have access to memory reserves because they have | |
2579 | * been OOM killed to get memory anywhere. | |
2580 | */ | |
2581 | if (unlikely(test_thread_flag(TIF_MEMDIE))) | |
2582 | return 1; | |
02a0e53d PJ |
2583 | return 0; |
2584 | } | |
2585 | ||
825a46af | 2586 | /** |
6adef3eb JS |
2587 | * cpuset_mem_spread_node() - On which node to begin search for a file page |
2588 | * cpuset_slab_spread_node() - On which node to begin search for a slab page | |
825a46af PJ |
2589 | * |
2590 | * If a task is marked PF_SPREAD_PAGE or PF_SPREAD_SLAB (as for | |
2591 | * tasks in a cpuset with is_spread_page or is_spread_slab set), | |
2592 | * and if the memory allocation used cpuset_mem_spread_node() | |
2593 | * to determine on which node to start looking, as it will for | |
2594 | * certain page cache or slab cache pages such as used for file | |
2595 | * system buffers and inode caches, then instead of starting on the | |
2596 | * local node to look for a free page, rather spread the starting | |
2597 | * node around the tasks mems_allowed nodes. | |
2598 | * | |
2599 | * We don't have to worry about the returned node being offline | |
2600 | * because "it can't happen", and even if it did, it would be ok. | |
2601 | * | |
2602 | * The routines calling guarantee_online_mems() are careful to | |
2603 | * only set nodes in task->mems_allowed that are online. So it | |
2604 | * should not be possible for the following code to return an | |
2605 | * offline node. But if it did, that would be ok, as this routine | |
2606 | * is not returning the node where the allocation must be, only | |
2607 | * the node where the search should start. The zonelist passed to | |
2608 | * __alloc_pages() will include all nodes. If the slab allocator | |
2609 | * is passed an offline node, it will fall back to the local node. | |
2610 | * See kmem_cache_alloc_node(). | |
2611 | */ | |
2612 | ||
6adef3eb | 2613 | static int cpuset_spread_node(int *rotor) |
825a46af PJ |
2614 | { |
2615 | int node; | |
2616 | ||
6adef3eb | 2617 | node = next_node(*rotor, current->mems_allowed); |
825a46af PJ |
2618 | if (node == MAX_NUMNODES) |
2619 | node = first_node(current->mems_allowed); | |
6adef3eb | 2620 | *rotor = node; |
825a46af PJ |
2621 | return node; |
2622 | } | |
6adef3eb JS |
2623 | |
2624 | int cpuset_mem_spread_node(void) | |
2625 | { | |
778d3b0f MH |
2626 | if (current->cpuset_mem_spread_rotor == NUMA_NO_NODE) |
2627 | current->cpuset_mem_spread_rotor = | |
2628 | node_random(¤t->mems_allowed); | |
2629 | ||
6adef3eb JS |
2630 | return cpuset_spread_node(¤t->cpuset_mem_spread_rotor); |
2631 | } | |
2632 | ||
2633 | int cpuset_slab_spread_node(void) | |
2634 | { | |
778d3b0f MH |
2635 | if (current->cpuset_slab_spread_rotor == NUMA_NO_NODE) |
2636 | current->cpuset_slab_spread_rotor = | |
2637 | node_random(¤t->mems_allowed); | |
2638 | ||
6adef3eb JS |
2639 | return cpuset_spread_node(¤t->cpuset_slab_spread_rotor); |
2640 | } | |
2641 | ||
825a46af PJ |
2642 | EXPORT_SYMBOL_GPL(cpuset_mem_spread_node); |
2643 | ||
ef08e3b4 | 2644 | /** |
bbe373f2 DR |
2645 | * cpuset_mems_allowed_intersects - Does @tsk1's mems_allowed intersect @tsk2's? |
2646 | * @tsk1: pointer to task_struct of some task. | |
2647 | * @tsk2: pointer to task_struct of some other task. | |
2648 | * | |
2649 | * Description: Return true if @tsk1's mems_allowed intersects the | |
2650 | * mems_allowed of @tsk2. Used by the OOM killer to determine if | |
2651 | * one of the task's memory usage might impact the memory available | |
2652 | * to the other. | |
ef08e3b4 PJ |
2653 | **/ |
2654 | ||
bbe373f2 DR |
2655 | int cpuset_mems_allowed_intersects(const struct task_struct *tsk1, |
2656 | const struct task_struct *tsk2) | |
ef08e3b4 | 2657 | { |
bbe373f2 | 2658 | return nodes_intersects(tsk1->mems_allowed, tsk2->mems_allowed); |
ef08e3b4 PJ |
2659 | } |
2660 | ||
f440d98f LZ |
2661 | #define CPUSET_NODELIST_LEN (256) |
2662 | ||
75aa1994 DR |
2663 | /** |
2664 | * cpuset_print_task_mems_allowed - prints task's cpuset and mems_allowed | |
fc34ac1d | 2665 | * @tsk: pointer to task_struct of some task. |
75aa1994 DR |
2666 | * |
2667 | * Description: Prints @task's name, cpuset name, and cached copy of its | |
b8dadcb5 | 2668 | * mems_allowed to the kernel log. |
75aa1994 DR |
2669 | */ |
2670 | void cpuset_print_task_mems_allowed(struct task_struct *tsk) | |
2671 | { | |
f440d98f LZ |
2672 | /* Statically allocated to prevent using excess stack. */ |
2673 | static char cpuset_nodelist[CPUSET_NODELIST_LEN]; | |
2674 | static DEFINE_SPINLOCK(cpuset_buffer_lock); | |
b8dadcb5 | 2675 | struct cgroup *cgrp; |
75aa1994 | 2676 | |
f440d98f | 2677 | spin_lock(&cpuset_buffer_lock); |
b8dadcb5 | 2678 | rcu_read_lock(); |
63f43f55 | 2679 | |
b8dadcb5 | 2680 | cgrp = task_cs(tsk)->css.cgroup; |
75aa1994 DR |
2681 | nodelist_scnprintf(cpuset_nodelist, CPUSET_NODELIST_LEN, |
2682 | tsk->mems_allowed); | |
12d3089c | 2683 | pr_info("%s cpuset=", tsk->comm); |
e61734c5 TH |
2684 | pr_cont_cgroup_name(cgrp); |
2685 | pr_cont(" mems_allowed=%s\n", cpuset_nodelist); | |
f440d98f | 2686 | |
cfb5966b | 2687 | rcu_read_unlock(); |
75aa1994 DR |
2688 | spin_unlock(&cpuset_buffer_lock); |
2689 | } | |
2690 | ||
3e0d98b9 PJ |
2691 | /* |
2692 | * Collection of memory_pressure is suppressed unless | |
2693 | * this flag is enabled by writing "1" to the special | |
2694 | * cpuset file 'memory_pressure_enabled' in the root cpuset. | |
2695 | */ | |
2696 | ||
c5b2aff8 | 2697 | int cpuset_memory_pressure_enabled __read_mostly; |
3e0d98b9 PJ |
2698 | |
2699 | /** | |
2700 | * cpuset_memory_pressure_bump - keep stats of per-cpuset reclaims. | |
2701 | * | |
2702 | * Keep a running average of the rate of synchronous (direct) | |
2703 | * page reclaim efforts initiated by tasks in each cpuset. | |
2704 | * | |
2705 | * This represents the rate at which some task in the cpuset | |
2706 | * ran low on memory on all nodes it was allowed to use, and | |
2707 | * had to enter the kernels page reclaim code in an effort to | |
2708 | * create more free memory by tossing clean pages or swapping | |
2709 | * or writing dirty pages. | |
2710 | * | |
2711 | * Display to user space in the per-cpuset read-only file | |
2712 | * "memory_pressure". Value displayed is an integer | |
2713 | * representing the recent rate of entry into the synchronous | |
2714 | * (direct) page reclaim by any task attached to the cpuset. | |
2715 | **/ | |
2716 | ||
2717 | void __cpuset_memory_pressure_bump(void) | |
2718 | { | |
b8dadcb5 | 2719 | rcu_read_lock(); |
8793d854 | 2720 | fmeter_markevent(&task_cs(current)->fmeter); |
b8dadcb5 | 2721 | rcu_read_unlock(); |
3e0d98b9 PJ |
2722 | } |
2723 | ||
8793d854 | 2724 | #ifdef CONFIG_PROC_PID_CPUSET |
1da177e4 LT |
2725 | /* |
2726 | * proc_cpuset_show() | |
2727 | * - Print tasks cpuset path into seq_file. | |
2728 | * - Used for /proc/<pid>/cpuset. | |
053199ed PJ |
2729 | * - No need to task_lock(tsk) on this tsk->cpuset reference, as it |
2730 | * doesn't really matter if tsk->cpuset changes after we read it, | |
5d21cc2d | 2731 | * and we take cpuset_mutex, keeping cpuset_attach() from changing it |
2df167a3 | 2732 | * anyway. |
1da177e4 | 2733 | */ |
52de4779 ZL |
2734 | int proc_cpuset_show(struct seq_file *m, struct pid_namespace *ns, |
2735 | struct pid *pid, struct task_struct *tsk) | |
1da177e4 | 2736 | { |
e61734c5 | 2737 | char *buf, *p; |
8793d854 | 2738 | struct cgroup_subsys_state *css; |
99f89551 | 2739 | int retval; |
1da177e4 | 2740 | |
99f89551 | 2741 | retval = -ENOMEM; |
e61734c5 | 2742 | buf = kmalloc(PATH_MAX, GFP_KERNEL); |
1da177e4 | 2743 | if (!buf) |
99f89551 EB |
2744 | goto out; |
2745 | ||
e61734c5 | 2746 | retval = -ENAMETOOLONG; |
27e89ae5 | 2747 | rcu_read_lock(); |
073219e9 | 2748 | css = task_css(tsk, cpuset_cgrp_id); |
e61734c5 | 2749 | p = cgroup_path(css->cgroup, buf, PATH_MAX); |
27e89ae5 | 2750 | rcu_read_unlock(); |
e61734c5 | 2751 | if (!p) |
52de4779 | 2752 | goto out_free; |
e61734c5 | 2753 | seq_puts(m, p); |
1da177e4 | 2754 | seq_putc(m, '\n'); |
e61734c5 | 2755 | retval = 0; |
99f89551 | 2756 | out_free: |
1da177e4 | 2757 | kfree(buf); |
99f89551 | 2758 | out: |
1da177e4 LT |
2759 | return retval; |
2760 | } | |
8793d854 | 2761 | #endif /* CONFIG_PROC_PID_CPUSET */ |
1da177e4 | 2762 | |
d01d4827 | 2763 | /* Display task mems_allowed in /proc/<pid>/status file. */ |
df5f8314 EB |
2764 | void cpuset_task_status_allowed(struct seq_file *m, struct task_struct *task) |
2765 | { | |
fc34ac1d | 2766 | seq_puts(m, "Mems_allowed:\t"); |
30e8e136 | 2767 | seq_nodemask(m, &task->mems_allowed); |
fc34ac1d FF |
2768 | seq_puts(m, "\n"); |
2769 | seq_puts(m, "Mems_allowed_list:\t"); | |
30e8e136 | 2770 | seq_nodemask_list(m, &task->mems_allowed); |
fc34ac1d | 2771 | seq_puts(m, "\n"); |
1da177e4 | 2772 | } |