]>
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 | |
202f72d5 PJ |
64 | /* |
65 | * Tracks how many cpusets are currently defined in system. | |
66 | * When there is only one cpuset (the root cpuset) we can | |
67 | * short circuit some hooks. | |
68 | */ | |
7edc5962 | 69 | int number_of_cpusets __read_mostly; |
202f72d5 | 70 | |
2df167a3 | 71 | /* Forward declare cgroup structures */ |
8793d854 PM |
72 | struct cgroup_subsys cpuset_subsys; |
73 | struct cpuset; | |
74 | ||
3e0d98b9 PJ |
75 | /* See "Frequency meter" comments, below. */ |
76 | ||
77 | struct fmeter { | |
78 | int cnt; /* unprocessed events count */ | |
79 | int val; /* most recent output value */ | |
80 | time_t time; /* clock (secs) when val computed */ | |
81 | spinlock_t lock; /* guards read or write of above */ | |
82 | }; | |
83 | ||
1da177e4 | 84 | struct cpuset { |
8793d854 PM |
85 | struct cgroup_subsys_state css; |
86 | ||
1da177e4 | 87 | unsigned long flags; /* "unsigned long" so bitops work */ |
300ed6cb | 88 | cpumask_var_t cpus_allowed; /* CPUs allowed to tasks in cpuset */ |
1da177e4 LT |
89 | nodemask_t mems_allowed; /* Memory Nodes allowed to tasks */ |
90 | ||
33ad801d LZ |
91 | /* |
92 | * This is old Memory Nodes tasks took on. | |
93 | * | |
94 | * - top_cpuset.old_mems_allowed is initialized to mems_allowed. | |
95 | * - A new cpuset's old_mems_allowed is initialized when some | |
96 | * task is moved into it. | |
97 | * - old_mems_allowed is used in cpuset_migrate_mm() when we change | |
98 | * cpuset.mems_allowed and have tasks' nodemask updated, and | |
99 | * then old_mems_allowed is updated to mems_allowed. | |
100 | */ | |
101 | nodemask_t old_mems_allowed; | |
102 | ||
3e0d98b9 | 103 | struct fmeter fmeter; /* memory_pressure filter */ |
029190c5 | 104 | |
452477fa TH |
105 | /* |
106 | * Tasks are being attached to this cpuset. Used to prevent | |
107 | * zeroing cpus/mems_allowed between ->can_attach() and ->attach(). | |
108 | */ | |
109 | int attach_in_progress; | |
110 | ||
029190c5 PJ |
111 | /* partition number for rebuild_sched_domains() */ |
112 | int pn; | |
956db3ca | 113 | |
1d3504fc HS |
114 | /* for custom sched domain */ |
115 | int relax_domain_level; | |
1da177e4 LT |
116 | }; |
117 | ||
8793d854 PM |
118 | /* Retrieve the cpuset for a cgroup */ |
119 | static inline struct cpuset *cgroup_cs(struct cgroup *cont) | |
120 | { | |
121 | return container_of(cgroup_subsys_state(cont, cpuset_subsys_id), | |
122 | struct cpuset, css); | |
123 | } | |
124 | ||
125 | /* Retrieve the cpuset for a task */ | |
126 | static inline struct cpuset *task_cs(struct task_struct *task) | |
127 | { | |
128 | return container_of(task_subsys_state(task, cpuset_subsys_id), | |
129 | struct cpuset, css); | |
130 | } | |
8793d854 | 131 | |
c431069f TH |
132 | static inline struct cpuset *parent_cs(const struct cpuset *cs) |
133 | { | |
134 | struct cgroup *pcgrp = cs->css.cgroup->parent; | |
135 | ||
136 | if (pcgrp) | |
137 | return cgroup_cs(pcgrp); | |
138 | return NULL; | |
139 | } | |
140 | ||
b246272e DR |
141 | #ifdef CONFIG_NUMA |
142 | static inline bool task_has_mempolicy(struct task_struct *task) | |
143 | { | |
144 | return task->mempolicy; | |
145 | } | |
146 | #else | |
147 | static inline bool task_has_mempolicy(struct task_struct *task) | |
148 | { | |
149 | return false; | |
150 | } | |
151 | #endif | |
152 | ||
153 | ||
1da177e4 LT |
154 | /* bits in struct cpuset flags field */ |
155 | typedef enum { | |
efeb77b2 | 156 | CS_ONLINE, |
1da177e4 LT |
157 | CS_CPU_EXCLUSIVE, |
158 | CS_MEM_EXCLUSIVE, | |
78608366 | 159 | CS_MEM_HARDWALL, |
45b07ef3 | 160 | CS_MEMORY_MIGRATE, |
029190c5 | 161 | CS_SCHED_LOAD_BALANCE, |
825a46af PJ |
162 | CS_SPREAD_PAGE, |
163 | CS_SPREAD_SLAB, | |
1da177e4 LT |
164 | } cpuset_flagbits_t; |
165 | ||
166 | /* convenient tests for these bits */ | |
efeb77b2 TH |
167 | static inline bool is_cpuset_online(const struct cpuset *cs) |
168 | { | |
169 | return test_bit(CS_ONLINE, &cs->flags); | |
170 | } | |
171 | ||
1da177e4 LT |
172 | static inline int is_cpu_exclusive(const struct cpuset *cs) |
173 | { | |
7b5b9ef0 | 174 | return test_bit(CS_CPU_EXCLUSIVE, &cs->flags); |
1da177e4 LT |
175 | } |
176 | ||
177 | static inline int is_mem_exclusive(const struct cpuset *cs) | |
178 | { | |
7b5b9ef0 | 179 | return test_bit(CS_MEM_EXCLUSIVE, &cs->flags); |
1da177e4 LT |
180 | } |
181 | ||
78608366 PM |
182 | static inline int is_mem_hardwall(const struct cpuset *cs) |
183 | { | |
184 | return test_bit(CS_MEM_HARDWALL, &cs->flags); | |
185 | } | |
186 | ||
029190c5 PJ |
187 | static inline int is_sched_load_balance(const struct cpuset *cs) |
188 | { | |
189 | return test_bit(CS_SCHED_LOAD_BALANCE, &cs->flags); | |
190 | } | |
191 | ||
45b07ef3 PJ |
192 | static inline int is_memory_migrate(const struct cpuset *cs) |
193 | { | |
7b5b9ef0 | 194 | return test_bit(CS_MEMORY_MIGRATE, &cs->flags); |
45b07ef3 PJ |
195 | } |
196 | ||
825a46af PJ |
197 | static inline int is_spread_page(const struct cpuset *cs) |
198 | { | |
199 | return test_bit(CS_SPREAD_PAGE, &cs->flags); | |
200 | } | |
201 | ||
202 | static inline int is_spread_slab(const struct cpuset *cs) | |
203 | { | |
204 | return test_bit(CS_SPREAD_SLAB, &cs->flags); | |
205 | } | |
206 | ||
1da177e4 | 207 | static struct cpuset top_cpuset = { |
efeb77b2 TH |
208 | .flags = ((1 << CS_ONLINE) | (1 << CS_CPU_EXCLUSIVE) | |
209 | (1 << CS_MEM_EXCLUSIVE)), | |
1da177e4 LT |
210 | }; |
211 | ||
ae8086ce TH |
212 | /** |
213 | * cpuset_for_each_child - traverse online children of a cpuset | |
214 | * @child_cs: loop cursor pointing to the current child | |
215 | * @pos_cgrp: used for iteration | |
216 | * @parent_cs: target cpuset to walk children of | |
217 | * | |
218 | * Walk @child_cs through the online children of @parent_cs. Must be used | |
219 | * with RCU read locked. | |
220 | */ | |
221 | #define cpuset_for_each_child(child_cs, pos_cgrp, parent_cs) \ | |
222 | cgroup_for_each_child((pos_cgrp), (parent_cs)->css.cgroup) \ | |
223 | if (is_cpuset_online(((child_cs) = cgroup_cs((pos_cgrp))))) | |
224 | ||
fc560a26 TH |
225 | /** |
226 | * cpuset_for_each_descendant_pre - pre-order walk of a cpuset's descendants | |
227 | * @des_cs: loop cursor pointing to the current descendant | |
228 | * @pos_cgrp: used for iteration | |
229 | * @root_cs: target cpuset to walk ancestor of | |
230 | * | |
231 | * Walk @des_cs through the online descendants of @root_cs. Must be used | |
232 | * with RCU read locked. The caller may modify @pos_cgrp by calling | |
233 | * cgroup_rightmost_descendant() to skip subtree. | |
234 | */ | |
235 | #define cpuset_for_each_descendant_pre(des_cs, pos_cgrp, root_cs) \ | |
236 | cgroup_for_each_descendant_pre((pos_cgrp), (root_cs)->css.cgroup) \ | |
237 | if (is_cpuset_online(((des_cs) = cgroup_cs((pos_cgrp))))) | |
238 | ||
1da177e4 | 239 | /* |
5d21cc2d TH |
240 | * There are two global mutexes guarding cpuset structures - cpuset_mutex |
241 | * and callback_mutex. The latter may nest inside the former. We also | |
242 | * require taking task_lock() when dereferencing a task's cpuset pointer. | |
243 | * See "The task_lock() exception", at the end of this comment. | |
244 | * | |
245 | * A task must hold both mutexes to modify cpusets. If a task holds | |
246 | * cpuset_mutex, then it blocks others wanting that mutex, ensuring that it | |
247 | * is the only task able to also acquire callback_mutex and be able to | |
248 | * modify cpusets. It can perform various checks on the cpuset structure | |
249 | * first, knowing nothing will change. It can also allocate memory while | |
250 | * just holding cpuset_mutex. While it is performing these checks, various | |
251 | * callback routines can briefly acquire callback_mutex to query cpusets. | |
252 | * Once it is ready to make the changes, it takes callback_mutex, blocking | |
253 | * everyone else. | |
053199ed PJ |
254 | * |
255 | * Calls to the kernel memory allocator can not be made while holding | |
3d3f26a7 | 256 | * callback_mutex, as that would risk double tripping on callback_mutex |
053199ed PJ |
257 | * from one of the callbacks into the cpuset code from within |
258 | * __alloc_pages(). | |
259 | * | |
3d3f26a7 | 260 | * If a task is only holding callback_mutex, then it has read-only |
053199ed PJ |
261 | * access to cpusets. |
262 | * | |
58568d2a MX |
263 | * Now, the task_struct fields mems_allowed and mempolicy may be changed |
264 | * by other task, we use alloc_lock in the task_struct fields to protect | |
265 | * them. | |
053199ed | 266 | * |
3d3f26a7 | 267 | * The cpuset_common_file_read() handlers only hold callback_mutex across |
053199ed PJ |
268 | * small pieces of code, such as when reading out possibly multi-word |
269 | * cpumasks and nodemasks. | |
270 | * | |
2df167a3 PM |
271 | * Accessing a task's cpuset should be done in accordance with the |
272 | * guidelines for accessing subsystem state in kernel/cgroup.c | |
1da177e4 LT |
273 | */ |
274 | ||
5d21cc2d | 275 | static DEFINE_MUTEX(cpuset_mutex); |
3d3f26a7 | 276 | static DEFINE_MUTEX(callback_mutex); |
4247bdc6 | 277 | |
3a5a6d0c TH |
278 | /* |
279 | * CPU / memory hotplug is handled asynchronously. | |
280 | */ | |
281 | static void cpuset_hotplug_workfn(struct work_struct *work); | |
3a5a6d0c TH |
282 | static DECLARE_WORK(cpuset_hotplug_work, cpuset_hotplug_workfn); |
283 | ||
e44193d3 LZ |
284 | static DECLARE_WAIT_QUEUE_HEAD(cpuset_attach_wq); |
285 | ||
cf417141 MK |
286 | /* |
287 | * This is ugly, but preserves the userspace API for existing cpuset | |
8793d854 | 288 | * users. If someone tries to mount the "cpuset" filesystem, we |
cf417141 MK |
289 | * silently switch it to mount "cgroup" instead |
290 | */ | |
f7e83571 AV |
291 | static struct dentry *cpuset_mount(struct file_system_type *fs_type, |
292 | int flags, const char *unused_dev_name, void *data) | |
1da177e4 | 293 | { |
8793d854 | 294 | struct file_system_type *cgroup_fs = get_fs_type("cgroup"); |
f7e83571 | 295 | struct dentry *ret = ERR_PTR(-ENODEV); |
8793d854 PM |
296 | if (cgroup_fs) { |
297 | char mountopts[] = | |
298 | "cpuset,noprefix," | |
299 | "release_agent=/sbin/cpuset_release_agent"; | |
f7e83571 AV |
300 | ret = cgroup_fs->mount(cgroup_fs, flags, |
301 | unused_dev_name, mountopts); | |
8793d854 PM |
302 | put_filesystem(cgroup_fs); |
303 | } | |
304 | return ret; | |
1da177e4 LT |
305 | } |
306 | ||
307 | static struct file_system_type cpuset_fs_type = { | |
308 | .name = "cpuset", | |
f7e83571 | 309 | .mount = cpuset_mount, |
1da177e4 LT |
310 | }; |
311 | ||
1da177e4 | 312 | /* |
300ed6cb | 313 | * Return in pmask the portion of a cpusets's cpus_allowed that |
1da177e4 | 314 | * are online. If none are online, walk up the cpuset hierarchy |
40df2deb LZ |
315 | * until we find one that does have some online cpus. The top |
316 | * cpuset always has some cpus online. | |
1da177e4 LT |
317 | * |
318 | * One way or another, we guarantee to return some non-empty subset | |
5f054e31 | 319 | * of cpu_online_mask. |
1da177e4 | 320 | * |
3d3f26a7 | 321 | * Call with callback_mutex held. |
1da177e4 | 322 | */ |
6af866af LZ |
323 | static void guarantee_online_cpus(const struct cpuset *cs, |
324 | struct cpumask *pmask) | |
1da177e4 | 325 | { |
40df2deb | 326 | while (!cpumask_intersects(cs->cpus_allowed, cpu_online_mask)) |
c431069f | 327 | cs = parent_cs(cs); |
40df2deb | 328 | cpumask_and(pmask, cs->cpus_allowed, cpu_online_mask); |
1da177e4 LT |
329 | } |
330 | ||
331 | /* | |
332 | * Return in *pmask the portion of a cpusets's mems_allowed that | |
0e1e7c7a CL |
333 | * are online, with memory. If none are online with memory, walk |
334 | * up the cpuset hierarchy until we find one that does have some | |
40df2deb | 335 | * online mems. The top cpuset always has some mems online. |
1da177e4 LT |
336 | * |
337 | * One way or another, we guarantee to return some non-empty subset | |
38d7bee9 | 338 | * of node_states[N_MEMORY]. |
1da177e4 | 339 | * |
3d3f26a7 | 340 | * Call with callback_mutex held. |
1da177e4 | 341 | */ |
1da177e4 LT |
342 | static void guarantee_online_mems(const struct cpuset *cs, nodemask_t *pmask) |
343 | { | |
40df2deb | 344 | while (!nodes_intersects(cs->mems_allowed, node_states[N_MEMORY])) |
c431069f | 345 | cs = parent_cs(cs); |
40df2deb | 346 | nodes_and(*pmask, cs->mems_allowed, node_states[N_MEMORY]); |
1da177e4 LT |
347 | } |
348 | ||
f3b39d47 MX |
349 | /* |
350 | * update task's spread flag if cpuset's page/slab spread flag is set | |
351 | * | |
5d21cc2d | 352 | * Called with callback_mutex/cpuset_mutex held |
f3b39d47 MX |
353 | */ |
354 | static void cpuset_update_task_spread_flag(struct cpuset *cs, | |
355 | struct task_struct *tsk) | |
356 | { | |
357 | if (is_spread_page(cs)) | |
358 | tsk->flags |= PF_SPREAD_PAGE; | |
359 | else | |
360 | tsk->flags &= ~PF_SPREAD_PAGE; | |
361 | if (is_spread_slab(cs)) | |
362 | tsk->flags |= PF_SPREAD_SLAB; | |
363 | else | |
364 | tsk->flags &= ~PF_SPREAD_SLAB; | |
365 | } | |
366 | ||
1da177e4 LT |
367 | /* |
368 | * is_cpuset_subset(p, q) - Is cpuset p a subset of cpuset q? | |
369 | * | |
370 | * One cpuset is a subset of another if all its allowed CPUs and | |
371 | * Memory Nodes are a subset of the other, and its exclusive flags | |
5d21cc2d | 372 | * are only set if the other's are set. Call holding cpuset_mutex. |
1da177e4 LT |
373 | */ |
374 | ||
375 | static int is_cpuset_subset(const struct cpuset *p, const struct cpuset *q) | |
376 | { | |
300ed6cb | 377 | return cpumask_subset(p->cpus_allowed, q->cpus_allowed) && |
1da177e4 LT |
378 | nodes_subset(p->mems_allowed, q->mems_allowed) && |
379 | is_cpu_exclusive(p) <= is_cpu_exclusive(q) && | |
380 | is_mem_exclusive(p) <= is_mem_exclusive(q); | |
381 | } | |
382 | ||
645fcc9d LZ |
383 | /** |
384 | * alloc_trial_cpuset - allocate a trial cpuset | |
385 | * @cs: the cpuset that the trial cpuset duplicates | |
386 | */ | |
387 | static struct cpuset *alloc_trial_cpuset(const struct cpuset *cs) | |
388 | { | |
300ed6cb LZ |
389 | struct cpuset *trial; |
390 | ||
391 | trial = kmemdup(cs, sizeof(*cs), GFP_KERNEL); | |
392 | if (!trial) | |
393 | return NULL; | |
394 | ||
395 | if (!alloc_cpumask_var(&trial->cpus_allowed, GFP_KERNEL)) { | |
396 | kfree(trial); | |
397 | return NULL; | |
398 | } | |
399 | cpumask_copy(trial->cpus_allowed, cs->cpus_allowed); | |
400 | ||
401 | return trial; | |
645fcc9d LZ |
402 | } |
403 | ||
404 | /** | |
405 | * free_trial_cpuset - free the trial cpuset | |
406 | * @trial: the trial cpuset to be freed | |
407 | */ | |
408 | static void free_trial_cpuset(struct cpuset *trial) | |
409 | { | |
300ed6cb | 410 | free_cpumask_var(trial->cpus_allowed); |
645fcc9d LZ |
411 | kfree(trial); |
412 | } | |
413 | ||
1da177e4 LT |
414 | /* |
415 | * validate_change() - Used to validate that any proposed cpuset change | |
416 | * follows the structural rules for cpusets. | |
417 | * | |
418 | * If we replaced the flag and mask values of the current cpuset | |
419 | * (cur) with those values in the trial cpuset (trial), would | |
420 | * our various subset and exclusive rules still be valid? Presumes | |
5d21cc2d | 421 | * cpuset_mutex held. |
1da177e4 LT |
422 | * |
423 | * 'cur' is the address of an actual, in-use cpuset. Operations | |
424 | * such as list traversal that depend on the actual address of the | |
425 | * cpuset in the list must use cur below, not trial. | |
426 | * | |
427 | * 'trial' is the address of bulk structure copy of cur, with | |
428 | * perhaps one or more of the fields cpus_allowed, mems_allowed, | |
429 | * or flags changed to new, trial values. | |
430 | * | |
431 | * Return 0 if valid, -errno if not. | |
432 | */ | |
433 | ||
434 | static int validate_change(const struct cpuset *cur, const struct cpuset *trial) | |
435 | { | |
8793d854 | 436 | struct cgroup *cont; |
1da177e4 | 437 | struct cpuset *c, *par; |
ae8086ce TH |
438 | int ret; |
439 | ||
440 | rcu_read_lock(); | |
1da177e4 LT |
441 | |
442 | /* Each of our child cpusets must be a subset of us */ | |
ae8086ce TH |
443 | ret = -EBUSY; |
444 | cpuset_for_each_child(c, cont, cur) | |
445 | if (!is_cpuset_subset(c, trial)) | |
446 | goto out; | |
1da177e4 LT |
447 | |
448 | /* Remaining checks don't apply to root cpuset */ | |
ae8086ce | 449 | ret = 0; |
69604067 | 450 | if (cur == &top_cpuset) |
ae8086ce | 451 | goto out; |
1da177e4 | 452 | |
c431069f | 453 | par = parent_cs(cur); |
69604067 | 454 | |
1da177e4 | 455 | /* We must be a subset of our parent cpuset */ |
ae8086ce | 456 | ret = -EACCES; |
1da177e4 | 457 | if (!is_cpuset_subset(trial, par)) |
ae8086ce | 458 | goto out; |
1da177e4 | 459 | |
2df167a3 PM |
460 | /* |
461 | * If either I or some sibling (!= me) is exclusive, we can't | |
462 | * overlap | |
463 | */ | |
ae8086ce TH |
464 | ret = -EINVAL; |
465 | cpuset_for_each_child(c, cont, par) { | |
1da177e4 LT |
466 | if ((is_cpu_exclusive(trial) || is_cpu_exclusive(c)) && |
467 | c != cur && | |
300ed6cb | 468 | cpumask_intersects(trial->cpus_allowed, c->cpus_allowed)) |
ae8086ce | 469 | goto out; |
1da177e4 LT |
470 | if ((is_mem_exclusive(trial) || is_mem_exclusive(c)) && |
471 | c != cur && | |
472 | nodes_intersects(trial->mems_allowed, c->mems_allowed)) | |
ae8086ce | 473 | goto out; |
1da177e4 LT |
474 | } |
475 | ||
452477fa TH |
476 | /* |
477 | * Cpusets with tasks - existing or newly being attached - can't | |
478 | * have empty cpus_allowed or mems_allowed. | |
479 | */ | |
ae8086ce | 480 | ret = -ENOSPC; |
452477fa | 481 | if ((cgroup_task_count(cur->css.cgroup) || cur->attach_in_progress) && |
ae8086ce TH |
482 | (cpumask_empty(trial->cpus_allowed) || |
483 | nodes_empty(trial->mems_allowed))) | |
484 | goto out; | |
020958b6 | 485 | |
ae8086ce TH |
486 | ret = 0; |
487 | out: | |
488 | rcu_read_unlock(); | |
489 | return ret; | |
1da177e4 LT |
490 | } |
491 | ||
db7f47cf | 492 | #ifdef CONFIG_SMP |
029190c5 | 493 | /* |
cf417141 | 494 | * Helper routine for generate_sched_domains(). |
029190c5 PJ |
495 | * Do cpusets a, b have overlapping cpus_allowed masks? |
496 | */ | |
029190c5 PJ |
497 | static int cpusets_overlap(struct cpuset *a, struct cpuset *b) |
498 | { | |
300ed6cb | 499 | return cpumask_intersects(a->cpus_allowed, b->cpus_allowed); |
029190c5 PJ |
500 | } |
501 | ||
1d3504fc HS |
502 | static void |
503 | update_domain_attr(struct sched_domain_attr *dattr, struct cpuset *c) | |
504 | { | |
1d3504fc HS |
505 | if (dattr->relax_domain_level < c->relax_domain_level) |
506 | dattr->relax_domain_level = c->relax_domain_level; | |
507 | return; | |
508 | } | |
509 | ||
fc560a26 TH |
510 | static void update_domain_attr_tree(struct sched_domain_attr *dattr, |
511 | struct cpuset *root_cs) | |
f5393693 | 512 | { |
fc560a26 TH |
513 | struct cpuset *cp; |
514 | struct cgroup *pos_cgrp; | |
f5393693 | 515 | |
fc560a26 TH |
516 | rcu_read_lock(); |
517 | cpuset_for_each_descendant_pre(cp, pos_cgrp, root_cs) { | |
518 | /* skip the whole subtree if @cp doesn't have any CPU */ | |
519 | if (cpumask_empty(cp->cpus_allowed)) { | |
520 | pos_cgrp = cgroup_rightmost_descendant(pos_cgrp); | |
f5393693 | 521 | continue; |
fc560a26 | 522 | } |
f5393693 LJ |
523 | |
524 | if (is_sched_load_balance(cp)) | |
525 | update_domain_attr(dattr, cp); | |
f5393693 | 526 | } |
fc560a26 | 527 | rcu_read_unlock(); |
f5393693 LJ |
528 | } |
529 | ||
029190c5 | 530 | /* |
cf417141 MK |
531 | * generate_sched_domains() |
532 | * | |
533 | * This function builds a partial partition of the systems CPUs | |
534 | * A 'partial partition' is a set of non-overlapping subsets whose | |
535 | * union is a subset of that set. | |
536 | * The output of this function needs to be passed to kernel/sched.c | |
537 | * partition_sched_domains() routine, which will rebuild the scheduler's | |
538 | * load balancing domains (sched domains) as specified by that partial | |
539 | * partition. | |
029190c5 | 540 | * |
45ce80fb | 541 | * See "What is sched_load_balance" in Documentation/cgroups/cpusets.txt |
029190c5 PJ |
542 | * for a background explanation of this. |
543 | * | |
544 | * Does not return errors, on the theory that the callers of this | |
545 | * routine would rather not worry about failures to rebuild sched | |
546 | * domains when operating in the severe memory shortage situations | |
547 | * that could cause allocation failures below. | |
548 | * | |
5d21cc2d | 549 | * Must be called with cpuset_mutex held. |
029190c5 PJ |
550 | * |
551 | * The three key local variables below are: | |
aeed6824 | 552 | * q - a linked-list queue of cpuset pointers, used to implement a |
029190c5 PJ |
553 | * top-down scan of all cpusets. This scan loads a pointer |
554 | * to each cpuset marked is_sched_load_balance into the | |
555 | * array 'csa'. For our purposes, rebuilding the schedulers | |
556 | * sched domains, we can ignore !is_sched_load_balance cpusets. | |
557 | * csa - (for CpuSet Array) Array of pointers to all the cpusets | |
558 | * that need to be load balanced, for convenient iterative | |
559 | * access by the subsequent code that finds the best partition, | |
560 | * i.e the set of domains (subsets) of CPUs such that the | |
561 | * cpus_allowed of every cpuset marked is_sched_load_balance | |
562 | * is a subset of one of these domains, while there are as | |
563 | * many such domains as possible, each as small as possible. | |
564 | * doms - Conversion of 'csa' to an array of cpumasks, for passing to | |
565 | * the kernel/sched.c routine partition_sched_domains() in a | |
566 | * convenient format, that can be easily compared to the prior | |
567 | * value to determine what partition elements (sched domains) | |
568 | * were changed (added or removed.) | |
569 | * | |
570 | * Finding the best partition (set of domains): | |
571 | * The triple nested loops below over i, j, k scan over the | |
572 | * load balanced cpusets (using the array of cpuset pointers in | |
573 | * csa[]) looking for pairs of cpusets that have overlapping | |
574 | * cpus_allowed, but which don't have the same 'pn' partition | |
575 | * number and gives them in the same partition number. It keeps | |
576 | * looping on the 'restart' label until it can no longer find | |
577 | * any such pairs. | |
578 | * | |
579 | * The union of the cpus_allowed masks from the set of | |
580 | * all cpusets having the same 'pn' value then form the one | |
581 | * element of the partition (one sched domain) to be passed to | |
582 | * partition_sched_domains(). | |
583 | */ | |
acc3f5d7 | 584 | static int generate_sched_domains(cpumask_var_t **domains, |
cf417141 | 585 | struct sched_domain_attr **attributes) |
029190c5 | 586 | { |
029190c5 PJ |
587 | struct cpuset *cp; /* scans q */ |
588 | struct cpuset **csa; /* array of all cpuset ptrs */ | |
589 | int csn; /* how many cpuset ptrs in csa so far */ | |
590 | int i, j, k; /* indices for partition finding loops */ | |
acc3f5d7 | 591 | cpumask_var_t *doms; /* resulting partition; i.e. sched domains */ |
1d3504fc | 592 | struct sched_domain_attr *dattr; /* attributes for custom domains */ |
1583715d | 593 | int ndoms = 0; /* number of sched domains in result */ |
6af866af | 594 | int nslot; /* next empty doms[] struct cpumask slot */ |
fc560a26 | 595 | struct cgroup *pos_cgrp; |
029190c5 | 596 | |
029190c5 | 597 | doms = NULL; |
1d3504fc | 598 | dattr = NULL; |
cf417141 | 599 | csa = NULL; |
029190c5 PJ |
600 | |
601 | /* Special case for the 99% of systems with one, full, sched domain */ | |
602 | if (is_sched_load_balance(&top_cpuset)) { | |
acc3f5d7 RR |
603 | ndoms = 1; |
604 | doms = alloc_sched_domains(ndoms); | |
029190c5 | 605 | if (!doms) |
cf417141 MK |
606 | goto done; |
607 | ||
1d3504fc HS |
608 | dattr = kmalloc(sizeof(struct sched_domain_attr), GFP_KERNEL); |
609 | if (dattr) { | |
610 | *dattr = SD_ATTR_INIT; | |
93a65575 | 611 | update_domain_attr_tree(dattr, &top_cpuset); |
1d3504fc | 612 | } |
acc3f5d7 | 613 | cpumask_copy(doms[0], top_cpuset.cpus_allowed); |
cf417141 | 614 | |
cf417141 | 615 | goto done; |
029190c5 PJ |
616 | } |
617 | ||
029190c5 PJ |
618 | csa = kmalloc(number_of_cpusets * sizeof(cp), GFP_KERNEL); |
619 | if (!csa) | |
620 | goto done; | |
621 | csn = 0; | |
622 | ||
fc560a26 TH |
623 | rcu_read_lock(); |
624 | cpuset_for_each_descendant_pre(cp, pos_cgrp, &top_cpuset) { | |
f5393693 | 625 | /* |
fc560a26 TH |
626 | * Continue traversing beyond @cp iff @cp has some CPUs and |
627 | * isn't load balancing. The former is obvious. The | |
628 | * latter: All child cpusets contain a subset of the | |
629 | * parent's cpus, so just skip them, and then we call | |
630 | * update_domain_attr_tree() to calc relax_domain_level of | |
631 | * the corresponding sched domain. | |
f5393693 | 632 | */ |
fc560a26 TH |
633 | if (!cpumask_empty(cp->cpus_allowed) && |
634 | !is_sched_load_balance(cp)) | |
f5393693 | 635 | continue; |
489a5393 | 636 | |
fc560a26 TH |
637 | if (is_sched_load_balance(cp)) |
638 | csa[csn++] = cp; | |
639 | ||
640 | /* skip @cp's subtree */ | |
641 | pos_cgrp = cgroup_rightmost_descendant(pos_cgrp); | |
642 | } | |
643 | rcu_read_unlock(); | |
029190c5 PJ |
644 | |
645 | for (i = 0; i < csn; i++) | |
646 | csa[i]->pn = i; | |
647 | ndoms = csn; | |
648 | ||
649 | restart: | |
650 | /* Find the best partition (set of sched domains) */ | |
651 | for (i = 0; i < csn; i++) { | |
652 | struct cpuset *a = csa[i]; | |
653 | int apn = a->pn; | |
654 | ||
655 | for (j = 0; j < csn; j++) { | |
656 | struct cpuset *b = csa[j]; | |
657 | int bpn = b->pn; | |
658 | ||
659 | if (apn != bpn && cpusets_overlap(a, b)) { | |
660 | for (k = 0; k < csn; k++) { | |
661 | struct cpuset *c = csa[k]; | |
662 | ||
663 | if (c->pn == bpn) | |
664 | c->pn = apn; | |
665 | } | |
666 | ndoms--; /* one less element */ | |
667 | goto restart; | |
668 | } | |
669 | } | |
670 | } | |
671 | ||
cf417141 MK |
672 | /* |
673 | * Now we know how many domains to create. | |
674 | * Convert <csn, csa> to <ndoms, doms> and populate cpu masks. | |
675 | */ | |
acc3f5d7 | 676 | doms = alloc_sched_domains(ndoms); |
700018e0 | 677 | if (!doms) |
cf417141 | 678 | goto done; |
cf417141 MK |
679 | |
680 | /* | |
681 | * The rest of the code, including the scheduler, can deal with | |
682 | * dattr==NULL case. No need to abort if alloc fails. | |
683 | */ | |
1d3504fc | 684 | dattr = kmalloc(ndoms * sizeof(struct sched_domain_attr), GFP_KERNEL); |
029190c5 PJ |
685 | |
686 | for (nslot = 0, i = 0; i < csn; i++) { | |
687 | struct cpuset *a = csa[i]; | |
6af866af | 688 | struct cpumask *dp; |
029190c5 PJ |
689 | int apn = a->pn; |
690 | ||
cf417141 MK |
691 | if (apn < 0) { |
692 | /* Skip completed partitions */ | |
693 | continue; | |
694 | } | |
695 | ||
acc3f5d7 | 696 | dp = doms[nslot]; |
cf417141 MK |
697 | |
698 | if (nslot == ndoms) { | |
699 | static int warnings = 10; | |
700 | if (warnings) { | |
701 | printk(KERN_WARNING | |
702 | "rebuild_sched_domains confused:" | |
703 | " nslot %d, ndoms %d, csn %d, i %d," | |
704 | " apn %d\n", | |
705 | nslot, ndoms, csn, i, apn); | |
706 | warnings--; | |
029190c5 | 707 | } |
cf417141 MK |
708 | continue; |
709 | } | |
029190c5 | 710 | |
6af866af | 711 | cpumask_clear(dp); |
cf417141 MK |
712 | if (dattr) |
713 | *(dattr + nslot) = SD_ATTR_INIT; | |
714 | for (j = i; j < csn; j++) { | |
715 | struct cpuset *b = csa[j]; | |
716 | ||
717 | if (apn == b->pn) { | |
300ed6cb | 718 | cpumask_or(dp, dp, b->cpus_allowed); |
cf417141 MK |
719 | if (dattr) |
720 | update_domain_attr_tree(dattr + nslot, b); | |
721 | ||
722 | /* Done with this partition */ | |
723 | b->pn = -1; | |
029190c5 | 724 | } |
029190c5 | 725 | } |
cf417141 | 726 | nslot++; |
029190c5 PJ |
727 | } |
728 | BUG_ON(nslot != ndoms); | |
729 | ||
cf417141 MK |
730 | done: |
731 | kfree(csa); | |
732 | ||
700018e0 LZ |
733 | /* |
734 | * Fallback to the default domain if kmalloc() failed. | |
735 | * See comments in partition_sched_domains(). | |
736 | */ | |
737 | if (doms == NULL) | |
738 | ndoms = 1; | |
739 | ||
cf417141 MK |
740 | *domains = doms; |
741 | *attributes = dattr; | |
742 | return ndoms; | |
743 | } | |
744 | ||
745 | /* | |
746 | * Rebuild scheduler domains. | |
747 | * | |
699140ba TH |
748 | * If the flag 'sched_load_balance' of any cpuset with non-empty |
749 | * 'cpus' changes, or if the 'cpus' allowed changes in any cpuset | |
750 | * which has that flag enabled, or if any cpuset with a non-empty | |
751 | * 'cpus' is removed, then call this routine to rebuild the | |
752 | * scheduler's dynamic sched domains. | |
cf417141 | 753 | * |
5d21cc2d | 754 | * Call with cpuset_mutex held. Takes get_online_cpus(). |
cf417141 | 755 | */ |
699140ba | 756 | static void rebuild_sched_domains_locked(void) |
cf417141 MK |
757 | { |
758 | struct sched_domain_attr *attr; | |
acc3f5d7 | 759 | cpumask_var_t *doms; |
cf417141 MK |
760 | int ndoms; |
761 | ||
5d21cc2d | 762 | lockdep_assert_held(&cpuset_mutex); |
86ef5c9a | 763 | get_online_cpus(); |
cf417141 | 764 | |
5b16c2a4 LZ |
765 | /* |
766 | * We have raced with CPU hotplug. Don't do anything to avoid | |
767 | * passing doms with offlined cpu to partition_sched_domains(). | |
768 | * Anyways, hotplug work item will rebuild sched domains. | |
769 | */ | |
770 | if (!cpumask_equal(top_cpuset.cpus_allowed, cpu_active_mask)) | |
771 | goto out; | |
772 | ||
cf417141 | 773 | /* Generate domain masks and attrs */ |
cf417141 | 774 | ndoms = generate_sched_domains(&doms, &attr); |
cf417141 MK |
775 | |
776 | /* Have scheduler rebuild the domains */ | |
777 | partition_sched_domains(ndoms, doms, attr); | |
5b16c2a4 | 778 | out: |
86ef5c9a | 779 | put_online_cpus(); |
cf417141 | 780 | } |
db7f47cf | 781 | #else /* !CONFIG_SMP */ |
699140ba | 782 | static void rebuild_sched_domains_locked(void) |
db7f47cf PM |
783 | { |
784 | } | |
db7f47cf | 785 | #endif /* CONFIG_SMP */ |
029190c5 | 786 | |
cf417141 MK |
787 | void rebuild_sched_domains(void) |
788 | { | |
5d21cc2d | 789 | mutex_lock(&cpuset_mutex); |
699140ba | 790 | rebuild_sched_domains_locked(); |
5d21cc2d | 791 | mutex_unlock(&cpuset_mutex); |
029190c5 PJ |
792 | } |
793 | ||
58f4790b CW |
794 | /** |
795 | * cpuset_change_cpumask - make a task's cpus_allowed the same as its cpuset's | |
796 | * @tsk: task to test | |
797 | * @scan: struct cgroup_scanner containing the cgroup of the task | |
798 | * | |
799 | * Called by cgroup_scan_tasks() for each task in a cgroup whose | |
800 | * cpus_allowed mask needs to be changed. | |
801 | * | |
802 | * We don't need to re-check for the cgroup/cpuset membership, since we're | |
5d21cc2d | 803 | * holding cpuset_mutex at this point. |
58f4790b | 804 | */ |
9e0c914c AB |
805 | static void cpuset_change_cpumask(struct task_struct *tsk, |
806 | struct cgroup_scanner *scan) | |
58f4790b | 807 | { |
300ed6cb | 808 | set_cpus_allowed_ptr(tsk, ((cgroup_cs(scan->cg))->cpus_allowed)); |
58f4790b CW |
809 | } |
810 | ||
0b2f630a MX |
811 | /** |
812 | * update_tasks_cpumask - Update the cpumasks of tasks in the cpuset. | |
813 | * @cs: the cpuset in which each task's cpus_allowed mask needs to be changed | |
4e74339a | 814 | * @heap: if NULL, defer allocating heap memory to cgroup_scan_tasks() |
0b2f630a | 815 | * |
5d21cc2d | 816 | * Called with cpuset_mutex held |
0b2f630a MX |
817 | * |
818 | * The cgroup_scan_tasks() function will scan all the tasks in a cgroup, | |
819 | * calling callback functions for each. | |
820 | * | |
4e74339a LZ |
821 | * No return value. It's guaranteed that cgroup_scan_tasks() always returns 0 |
822 | * if @heap != NULL. | |
0b2f630a | 823 | */ |
4e74339a | 824 | static void update_tasks_cpumask(struct cpuset *cs, struct ptr_heap *heap) |
0b2f630a MX |
825 | { |
826 | struct cgroup_scanner scan; | |
0b2f630a MX |
827 | |
828 | scan.cg = cs->css.cgroup; | |
249cc86d | 829 | scan.test_task = NULL; |
0b2f630a | 830 | scan.process_task = cpuset_change_cpumask; |
4e74339a LZ |
831 | scan.heap = heap; |
832 | cgroup_scan_tasks(&scan); | |
0b2f630a MX |
833 | } |
834 | ||
58f4790b CW |
835 | /** |
836 | * update_cpumask - update the cpus_allowed mask of a cpuset and all tasks in it | |
837 | * @cs: the cpuset to consider | |
838 | * @buf: buffer of cpu numbers written to this cpuset | |
839 | */ | |
645fcc9d LZ |
840 | static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs, |
841 | const char *buf) | |
1da177e4 | 842 | { |
4e74339a | 843 | struct ptr_heap heap; |
58f4790b CW |
844 | int retval; |
845 | int is_load_balanced; | |
1da177e4 | 846 | |
5f054e31 | 847 | /* top_cpuset.cpus_allowed tracks cpu_online_mask; it's read-only */ |
4c4d50f7 PJ |
848 | if (cs == &top_cpuset) |
849 | return -EACCES; | |
850 | ||
6f7f02e7 | 851 | /* |
c8d9c90c | 852 | * An empty cpus_allowed is ok only if the cpuset has no tasks. |
020958b6 PJ |
853 | * Since cpulist_parse() fails on an empty mask, we special case |
854 | * that parsing. The validate_change() call ensures that cpusets | |
855 | * with tasks have cpus. | |
6f7f02e7 | 856 | */ |
020958b6 | 857 | if (!*buf) { |
300ed6cb | 858 | cpumask_clear(trialcs->cpus_allowed); |
6f7f02e7 | 859 | } else { |
300ed6cb | 860 | retval = cpulist_parse(buf, trialcs->cpus_allowed); |
6f7f02e7 DR |
861 | if (retval < 0) |
862 | return retval; | |
37340746 | 863 | |
6ad4c188 | 864 | if (!cpumask_subset(trialcs->cpus_allowed, cpu_active_mask)) |
37340746 | 865 | return -EINVAL; |
6f7f02e7 | 866 | } |
029190c5 | 867 | |
8707d8b8 | 868 | /* Nothing to do if the cpus didn't change */ |
300ed6cb | 869 | if (cpumask_equal(cs->cpus_allowed, trialcs->cpus_allowed)) |
8707d8b8 | 870 | return 0; |
58f4790b | 871 | |
a73456f3 LZ |
872 | retval = validate_change(cs, trialcs); |
873 | if (retval < 0) | |
874 | return retval; | |
875 | ||
4e74339a LZ |
876 | retval = heap_init(&heap, PAGE_SIZE, GFP_KERNEL, NULL); |
877 | if (retval) | |
878 | return retval; | |
879 | ||
645fcc9d | 880 | is_load_balanced = is_sched_load_balance(trialcs); |
029190c5 | 881 | |
3d3f26a7 | 882 | mutex_lock(&callback_mutex); |
300ed6cb | 883 | cpumask_copy(cs->cpus_allowed, trialcs->cpus_allowed); |
3d3f26a7 | 884 | mutex_unlock(&callback_mutex); |
029190c5 | 885 | |
8707d8b8 PM |
886 | /* |
887 | * Scan tasks in the cpuset, and update the cpumasks of any | |
58f4790b | 888 | * that need an update. |
8707d8b8 | 889 | */ |
4e74339a LZ |
890 | update_tasks_cpumask(cs, &heap); |
891 | ||
892 | heap_free(&heap); | |
58f4790b | 893 | |
8707d8b8 | 894 | if (is_load_balanced) |
699140ba | 895 | rebuild_sched_domains_locked(); |
85d7b949 | 896 | return 0; |
1da177e4 LT |
897 | } |
898 | ||
e4e364e8 PJ |
899 | /* |
900 | * cpuset_migrate_mm | |
901 | * | |
902 | * Migrate memory region from one set of nodes to another. | |
903 | * | |
904 | * Temporarilly set tasks mems_allowed to target nodes of migration, | |
905 | * so that the migration code can allocate pages on these nodes. | |
906 | * | |
5d21cc2d | 907 | * Call holding cpuset_mutex, so current's cpuset won't change |
c8d9c90c | 908 | * during this call, as manage_mutex holds off any cpuset_attach() |
e4e364e8 PJ |
909 | * calls. Therefore we don't need to take task_lock around the |
910 | * call to guarantee_online_mems(), as we know no one is changing | |
2df167a3 | 911 | * our task's cpuset. |
e4e364e8 | 912 | * |
e4e364e8 PJ |
913 | * While the mm_struct we are migrating is typically from some |
914 | * other task, the task_struct mems_allowed that we are hacking | |
915 | * is for our current task, which must allocate new pages for that | |
916 | * migrating memory region. | |
e4e364e8 PJ |
917 | */ |
918 | ||
919 | static void cpuset_migrate_mm(struct mm_struct *mm, const nodemask_t *from, | |
920 | const nodemask_t *to) | |
921 | { | |
922 | struct task_struct *tsk = current; | |
923 | ||
e4e364e8 | 924 | tsk->mems_allowed = *to; |
e4e364e8 PJ |
925 | |
926 | do_migrate_pages(mm, from, to, MPOL_MF_MOVE_ALL); | |
927 | ||
8793d854 | 928 | guarantee_online_mems(task_cs(tsk),&tsk->mems_allowed); |
e4e364e8 PJ |
929 | } |
930 | ||
3b6766fe | 931 | /* |
58568d2a MX |
932 | * cpuset_change_task_nodemask - change task's mems_allowed and mempolicy |
933 | * @tsk: the task to change | |
934 | * @newmems: new nodes that the task will be set | |
935 | * | |
936 | * In order to avoid seeing no nodes if the old and new nodes are disjoint, | |
937 | * we structure updates as setting all new allowed nodes, then clearing newly | |
938 | * disallowed ones. | |
58568d2a MX |
939 | */ |
940 | static void cpuset_change_task_nodemask(struct task_struct *tsk, | |
941 | nodemask_t *newmems) | |
942 | { | |
b246272e | 943 | bool need_loop; |
89e8a244 | 944 | |
c0ff7453 MX |
945 | /* |
946 | * Allow tasks that have access to memory reserves because they have | |
947 | * been OOM killed to get memory anywhere. | |
948 | */ | |
949 | if (unlikely(test_thread_flag(TIF_MEMDIE))) | |
950 | return; | |
951 | if (current->flags & PF_EXITING) /* Let dying task have memory */ | |
952 | return; | |
953 | ||
954 | task_lock(tsk); | |
b246272e DR |
955 | /* |
956 | * Determine if a loop is necessary if another thread is doing | |
957 | * get_mems_allowed(). If at least one node remains unchanged and | |
958 | * tsk does not have a mempolicy, then an empty nodemask will not be | |
959 | * possible when mems_allowed is larger than a word. | |
960 | */ | |
961 | need_loop = task_has_mempolicy(tsk) || | |
962 | !nodes_intersects(*newmems, tsk->mems_allowed); | |
c0ff7453 | 963 | |
cc9a6c87 MG |
964 | if (need_loop) |
965 | write_seqcount_begin(&tsk->mems_allowed_seq); | |
c0ff7453 | 966 | |
cc9a6c87 MG |
967 | nodes_or(tsk->mems_allowed, tsk->mems_allowed, *newmems); |
968 | mpol_rebind_task(tsk, newmems, MPOL_REBIND_STEP1); | |
c0ff7453 MX |
969 | |
970 | mpol_rebind_task(tsk, newmems, MPOL_REBIND_STEP2); | |
58568d2a | 971 | tsk->mems_allowed = *newmems; |
cc9a6c87 MG |
972 | |
973 | if (need_loop) | |
974 | write_seqcount_end(&tsk->mems_allowed_seq); | |
975 | ||
c0ff7453 | 976 | task_unlock(tsk); |
58568d2a MX |
977 | } |
978 | ||
979 | /* | |
980 | * Update task's mems_allowed and rebind its mempolicy and vmas' mempolicy | |
981 | * of it to cpuset's new mems_allowed, and migrate pages to new nodes if | |
5d21cc2d | 982 | * memory_migrate flag is set. Called with cpuset_mutex held. |
3b6766fe LZ |
983 | */ |
984 | static void cpuset_change_nodemask(struct task_struct *p, | |
985 | struct cgroup_scanner *scan) | |
986 | { | |
33ad801d | 987 | struct cpuset *cs = cgroup_cs(scan->cg); |
3b6766fe | 988 | struct mm_struct *mm; |
3b6766fe | 989 | int migrate; |
33ad801d | 990 | nodemask_t *newmems = scan->data; |
58568d2a | 991 | |
33ad801d | 992 | cpuset_change_task_nodemask(p, newmems); |
53feb297 | 993 | |
3b6766fe LZ |
994 | mm = get_task_mm(p); |
995 | if (!mm) | |
996 | return; | |
997 | ||
3b6766fe LZ |
998 | migrate = is_memory_migrate(cs); |
999 | ||
1000 | mpol_rebind_mm(mm, &cs->mems_allowed); | |
1001 | if (migrate) | |
33ad801d | 1002 | cpuset_migrate_mm(mm, &cs->old_mems_allowed, newmems); |
3b6766fe LZ |
1003 | mmput(mm); |
1004 | } | |
1005 | ||
8793d854 PM |
1006 | static void *cpuset_being_rebound; |
1007 | ||
0b2f630a MX |
1008 | /** |
1009 | * update_tasks_nodemask - Update the nodemasks of tasks in the cpuset. | |
1010 | * @cs: the cpuset in which each task's mems_allowed mask needs to be changed | |
010cfac4 | 1011 | * @heap: if NULL, defer allocating heap memory to cgroup_scan_tasks() |
0b2f630a | 1012 | * |
5d21cc2d | 1013 | * Called with cpuset_mutex held |
010cfac4 LZ |
1014 | * No return value. It's guaranteed that cgroup_scan_tasks() always returns 0 |
1015 | * if @heap != NULL. | |
0b2f630a | 1016 | */ |
33ad801d | 1017 | static void update_tasks_nodemask(struct cpuset *cs, struct ptr_heap *heap) |
1da177e4 | 1018 | { |
33ad801d | 1019 | static nodemask_t newmems; /* protected by cpuset_mutex */ |
3b6766fe | 1020 | struct cgroup_scanner scan; |
59dac16f | 1021 | |
846a16bf | 1022 | cpuset_being_rebound = cs; /* causes mpol_dup() rebind */ |
4225399a | 1023 | |
33ad801d LZ |
1024 | guarantee_online_mems(cs, &newmems); |
1025 | ||
3b6766fe LZ |
1026 | scan.cg = cs->css.cgroup; |
1027 | scan.test_task = NULL; | |
1028 | scan.process_task = cpuset_change_nodemask; | |
010cfac4 | 1029 | scan.heap = heap; |
33ad801d | 1030 | scan.data = &newmems; |
4225399a PJ |
1031 | |
1032 | /* | |
3b6766fe LZ |
1033 | * The mpol_rebind_mm() call takes mmap_sem, which we couldn't |
1034 | * take while holding tasklist_lock. Forks can happen - the | |
1035 | * mpol_dup() cpuset_being_rebound check will catch such forks, | |
1036 | * and rebind their vma mempolicies too. Because we still hold | |
5d21cc2d | 1037 | * the global cpuset_mutex, we know that no other rebind effort |
3b6766fe | 1038 | * will be contending for the global variable cpuset_being_rebound. |
4225399a | 1039 | * It's ok if we rebind the same mm twice; mpol_rebind_mm() |
04c19fa6 | 1040 | * is idempotent. Also migrate pages in each mm to new nodes. |
4225399a | 1041 | */ |
010cfac4 | 1042 | cgroup_scan_tasks(&scan); |
4225399a | 1043 | |
33ad801d LZ |
1044 | /* |
1045 | * All the tasks' nodemasks have been updated, update | |
1046 | * cs->old_mems_allowed. | |
1047 | */ | |
1048 | cs->old_mems_allowed = newmems; | |
1049 | ||
2df167a3 | 1050 | /* We're done rebinding vmas to this cpuset's new mems_allowed. */ |
8793d854 | 1051 | cpuset_being_rebound = NULL; |
1da177e4 LT |
1052 | } |
1053 | ||
0b2f630a MX |
1054 | /* |
1055 | * Handle user request to change the 'mems' memory placement | |
1056 | * of a cpuset. Needs to validate the request, update the | |
58568d2a MX |
1057 | * cpusets mems_allowed, and for each task in the cpuset, |
1058 | * update mems_allowed and rebind task's mempolicy and any vma | |
1059 | * mempolicies and if the cpuset is marked 'memory_migrate', | |
1060 | * migrate the tasks pages to the new memory. | |
0b2f630a | 1061 | * |
5d21cc2d | 1062 | * Call with cpuset_mutex held. May take callback_mutex during call. |
0b2f630a MX |
1063 | * Will take tasklist_lock, scan tasklist for tasks in cpuset cs, |
1064 | * lock each such tasks mm->mmap_sem, scan its vma's and rebind | |
1065 | * their mempolicies to the cpusets new mems_allowed. | |
1066 | */ | |
645fcc9d LZ |
1067 | static int update_nodemask(struct cpuset *cs, struct cpuset *trialcs, |
1068 | const char *buf) | |
0b2f630a | 1069 | { |
0b2f630a | 1070 | int retval; |
010cfac4 | 1071 | struct ptr_heap heap; |
0b2f630a MX |
1072 | |
1073 | /* | |
38d7bee9 | 1074 | * top_cpuset.mems_allowed tracks node_stats[N_MEMORY]; |
0b2f630a MX |
1075 | * it's read-only |
1076 | */ | |
53feb297 MX |
1077 | if (cs == &top_cpuset) { |
1078 | retval = -EACCES; | |
1079 | goto done; | |
1080 | } | |
0b2f630a | 1081 | |
0b2f630a MX |
1082 | /* |
1083 | * An empty mems_allowed is ok iff there are no tasks in the cpuset. | |
1084 | * Since nodelist_parse() fails on an empty mask, we special case | |
1085 | * that parsing. The validate_change() call ensures that cpusets | |
1086 | * with tasks have memory. | |
1087 | */ | |
1088 | if (!*buf) { | |
645fcc9d | 1089 | nodes_clear(trialcs->mems_allowed); |
0b2f630a | 1090 | } else { |
645fcc9d | 1091 | retval = nodelist_parse(buf, trialcs->mems_allowed); |
0b2f630a MX |
1092 | if (retval < 0) |
1093 | goto done; | |
1094 | ||
645fcc9d | 1095 | if (!nodes_subset(trialcs->mems_allowed, |
38d7bee9 | 1096 | node_states[N_MEMORY])) { |
53feb297 MX |
1097 | retval = -EINVAL; |
1098 | goto done; | |
1099 | } | |
0b2f630a | 1100 | } |
33ad801d LZ |
1101 | |
1102 | if (nodes_equal(cs->mems_allowed, trialcs->mems_allowed)) { | |
0b2f630a MX |
1103 | retval = 0; /* Too easy - nothing to do */ |
1104 | goto done; | |
1105 | } | |
645fcc9d | 1106 | retval = validate_change(cs, trialcs); |
0b2f630a MX |
1107 | if (retval < 0) |
1108 | goto done; | |
1109 | ||
010cfac4 LZ |
1110 | retval = heap_init(&heap, PAGE_SIZE, GFP_KERNEL, NULL); |
1111 | if (retval < 0) | |
1112 | goto done; | |
1113 | ||
0b2f630a | 1114 | mutex_lock(&callback_mutex); |
645fcc9d | 1115 | cs->mems_allowed = trialcs->mems_allowed; |
0b2f630a MX |
1116 | mutex_unlock(&callback_mutex); |
1117 | ||
33ad801d | 1118 | update_tasks_nodemask(cs, &heap); |
010cfac4 LZ |
1119 | |
1120 | heap_free(&heap); | |
0b2f630a MX |
1121 | done: |
1122 | return retval; | |
1123 | } | |
1124 | ||
8793d854 PM |
1125 | int current_cpuset_is_being_rebound(void) |
1126 | { | |
1127 | return task_cs(current) == cpuset_being_rebound; | |
1128 | } | |
1129 | ||
5be7a479 | 1130 | static int update_relax_domain_level(struct cpuset *cs, s64 val) |
1d3504fc | 1131 | { |
db7f47cf | 1132 | #ifdef CONFIG_SMP |
60495e77 | 1133 | if (val < -1 || val >= sched_domain_level_max) |
30e0e178 | 1134 | return -EINVAL; |
db7f47cf | 1135 | #endif |
1d3504fc HS |
1136 | |
1137 | if (val != cs->relax_domain_level) { | |
1138 | cs->relax_domain_level = val; | |
300ed6cb LZ |
1139 | if (!cpumask_empty(cs->cpus_allowed) && |
1140 | is_sched_load_balance(cs)) | |
699140ba | 1141 | rebuild_sched_domains_locked(); |
1d3504fc HS |
1142 | } |
1143 | ||
1144 | return 0; | |
1145 | } | |
1146 | ||
950592f7 MX |
1147 | /* |
1148 | * cpuset_change_flag - make a task's spread flags the same as its cpuset's | |
1149 | * @tsk: task to be updated | |
1150 | * @scan: struct cgroup_scanner containing the cgroup of the task | |
1151 | * | |
1152 | * Called by cgroup_scan_tasks() for each task in a cgroup. | |
1153 | * | |
1154 | * We don't need to re-check for the cgroup/cpuset membership, since we're | |
5d21cc2d | 1155 | * holding cpuset_mutex at this point. |
950592f7 MX |
1156 | */ |
1157 | static void cpuset_change_flag(struct task_struct *tsk, | |
1158 | struct cgroup_scanner *scan) | |
1159 | { | |
1160 | cpuset_update_task_spread_flag(cgroup_cs(scan->cg), tsk); | |
1161 | } | |
1162 | ||
1163 | /* | |
1164 | * update_tasks_flags - update the spread flags of tasks in the cpuset. | |
1165 | * @cs: the cpuset in which each task's spread flags needs to be changed | |
1166 | * @heap: if NULL, defer allocating heap memory to cgroup_scan_tasks() | |
1167 | * | |
5d21cc2d | 1168 | * Called with cpuset_mutex held |
950592f7 MX |
1169 | * |
1170 | * The cgroup_scan_tasks() function will scan all the tasks in a cgroup, | |
1171 | * calling callback functions for each. | |
1172 | * | |
1173 | * No return value. It's guaranteed that cgroup_scan_tasks() always returns 0 | |
1174 | * if @heap != NULL. | |
1175 | */ | |
1176 | static void update_tasks_flags(struct cpuset *cs, struct ptr_heap *heap) | |
1177 | { | |
1178 | struct cgroup_scanner scan; | |
1179 | ||
1180 | scan.cg = cs->css.cgroup; | |
1181 | scan.test_task = NULL; | |
1182 | scan.process_task = cpuset_change_flag; | |
1183 | scan.heap = heap; | |
1184 | cgroup_scan_tasks(&scan); | |
1185 | } | |
1186 | ||
1da177e4 LT |
1187 | /* |
1188 | * update_flag - read a 0 or a 1 in a file and update associated flag | |
78608366 PM |
1189 | * bit: the bit to update (see cpuset_flagbits_t) |
1190 | * cs: the cpuset to update | |
1191 | * turning_on: whether the flag is being set or cleared | |
053199ed | 1192 | * |
5d21cc2d | 1193 | * Call with cpuset_mutex held. |
1da177e4 LT |
1194 | */ |
1195 | ||
700fe1ab PM |
1196 | static int update_flag(cpuset_flagbits_t bit, struct cpuset *cs, |
1197 | int turning_on) | |
1da177e4 | 1198 | { |
645fcc9d | 1199 | struct cpuset *trialcs; |
40b6a762 | 1200 | int balance_flag_changed; |
950592f7 MX |
1201 | int spread_flag_changed; |
1202 | struct ptr_heap heap; | |
1203 | int err; | |
1da177e4 | 1204 | |
645fcc9d LZ |
1205 | trialcs = alloc_trial_cpuset(cs); |
1206 | if (!trialcs) | |
1207 | return -ENOMEM; | |
1208 | ||
1da177e4 | 1209 | if (turning_on) |
645fcc9d | 1210 | set_bit(bit, &trialcs->flags); |
1da177e4 | 1211 | else |
645fcc9d | 1212 | clear_bit(bit, &trialcs->flags); |
1da177e4 | 1213 | |
645fcc9d | 1214 | err = validate_change(cs, trialcs); |
85d7b949 | 1215 | if (err < 0) |
645fcc9d | 1216 | goto out; |
029190c5 | 1217 | |
950592f7 MX |
1218 | err = heap_init(&heap, PAGE_SIZE, GFP_KERNEL, NULL); |
1219 | if (err < 0) | |
1220 | goto out; | |
1221 | ||
029190c5 | 1222 | balance_flag_changed = (is_sched_load_balance(cs) != |
645fcc9d | 1223 | is_sched_load_balance(trialcs)); |
029190c5 | 1224 | |
950592f7 MX |
1225 | spread_flag_changed = ((is_spread_slab(cs) != is_spread_slab(trialcs)) |
1226 | || (is_spread_page(cs) != is_spread_page(trialcs))); | |
1227 | ||
3d3f26a7 | 1228 | mutex_lock(&callback_mutex); |
645fcc9d | 1229 | cs->flags = trialcs->flags; |
3d3f26a7 | 1230 | mutex_unlock(&callback_mutex); |
85d7b949 | 1231 | |
300ed6cb | 1232 | if (!cpumask_empty(trialcs->cpus_allowed) && balance_flag_changed) |
699140ba | 1233 | rebuild_sched_domains_locked(); |
029190c5 | 1234 | |
950592f7 MX |
1235 | if (spread_flag_changed) |
1236 | update_tasks_flags(cs, &heap); | |
1237 | heap_free(&heap); | |
645fcc9d LZ |
1238 | out: |
1239 | free_trial_cpuset(trialcs); | |
1240 | return err; | |
1da177e4 LT |
1241 | } |
1242 | ||
3e0d98b9 | 1243 | /* |
80f7228b | 1244 | * Frequency meter - How fast is some event occurring? |
3e0d98b9 PJ |
1245 | * |
1246 | * These routines manage a digitally filtered, constant time based, | |
1247 | * event frequency meter. There are four routines: | |
1248 | * fmeter_init() - initialize a frequency meter. | |
1249 | * fmeter_markevent() - called each time the event happens. | |
1250 | * fmeter_getrate() - returns the recent rate of such events. | |
1251 | * fmeter_update() - internal routine used to update fmeter. | |
1252 | * | |
1253 | * A common data structure is passed to each of these routines, | |
1254 | * which is used to keep track of the state required to manage the | |
1255 | * frequency meter and its digital filter. | |
1256 | * | |
1257 | * The filter works on the number of events marked per unit time. | |
1258 | * The filter is single-pole low-pass recursive (IIR). The time unit | |
1259 | * is 1 second. Arithmetic is done using 32-bit integers scaled to | |
1260 | * simulate 3 decimal digits of precision (multiplied by 1000). | |
1261 | * | |
1262 | * With an FM_COEF of 933, and a time base of 1 second, the filter | |
1263 | * has a half-life of 10 seconds, meaning that if the events quit | |
1264 | * happening, then the rate returned from the fmeter_getrate() | |
1265 | * will be cut in half each 10 seconds, until it converges to zero. | |
1266 | * | |
1267 | * It is not worth doing a real infinitely recursive filter. If more | |
1268 | * than FM_MAXTICKS ticks have elapsed since the last filter event, | |
1269 | * just compute FM_MAXTICKS ticks worth, by which point the level | |
1270 | * will be stable. | |
1271 | * | |
1272 | * Limit the count of unprocessed events to FM_MAXCNT, so as to avoid | |
1273 | * arithmetic overflow in the fmeter_update() routine. | |
1274 | * | |
1275 | * Given the simple 32 bit integer arithmetic used, this meter works | |
1276 | * best for reporting rates between one per millisecond (msec) and | |
1277 | * one per 32 (approx) seconds. At constant rates faster than one | |
1278 | * per msec it maxes out at values just under 1,000,000. At constant | |
1279 | * rates between one per msec, and one per second it will stabilize | |
1280 | * to a value N*1000, where N is the rate of events per second. | |
1281 | * At constant rates between one per second and one per 32 seconds, | |
1282 | * it will be choppy, moving up on the seconds that have an event, | |
1283 | * and then decaying until the next event. At rates slower than | |
1284 | * about one in 32 seconds, it decays all the way back to zero between | |
1285 | * each event. | |
1286 | */ | |
1287 | ||
1288 | #define FM_COEF 933 /* coefficient for half-life of 10 secs */ | |
1289 | #define FM_MAXTICKS ((time_t)99) /* useless computing more ticks than this */ | |
1290 | #define FM_MAXCNT 1000000 /* limit cnt to avoid overflow */ | |
1291 | #define FM_SCALE 1000 /* faux fixed point scale */ | |
1292 | ||
1293 | /* Initialize a frequency meter */ | |
1294 | static void fmeter_init(struct fmeter *fmp) | |
1295 | { | |
1296 | fmp->cnt = 0; | |
1297 | fmp->val = 0; | |
1298 | fmp->time = 0; | |
1299 | spin_lock_init(&fmp->lock); | |
1300 | } | |
1301 | ||
1302 | /* Internal meter update - process cnt events and update value */ | |
1303 | static void fmeter_update(struct fmeter *fmp) | |
1304 | { | |
1305 | time_t now = get_seconds(); | |
1306 | time_t ticks = now - fmp->time; | |
1307 | ||
1308 | if (ticks == 0) | |
1309 | return; | |
1310 | ||
1311 | ticks = min(FM_MAXTICKS, ticks); | |
1312 | while (ticks-- > 0) | |
1313 | fmp->val = (FM_COEF * fmp->val) / FM_SCALE; | |
1314 | fmp->time = now; | |
1315 | ||
1316 | fmp->val += ((FM_SCALE - FM_COEF) * fmp->cnt) / FM_SCALE; | |
1317 | fmp->cnt = 0; | |
1318 | } | |
1319 | ||
1320 | /* Process any previous ticks, then bump cnt by one (times scale). */ | |
1321 | static void fmeter_markevent(struct fmeter *fmp) | |
1322 | { | |
1323 | spin_lock(&fmp->lock); | |
1324 | fmeter_update(fmp); | |
1325 | fmp->cnt = min(FM_MAXCNT, fmp->cnt + FM_SCALE); | |
1326 | spin_unlock(&fmp->lock); | |
1327 | } | |
1328 | ||
1329 | /* Process any previous ticks, then return current value. */ | |
1330 | static int fmeter_getrate(struct fmeter *fmp) | |
1331 | { | |
1332 | int val; | |
1333 | ||
1334 | spin_lock(&fmp->lock); | |
1335 | fmeter_update(fmp); | |
1336 | val = fmp->val; | |
1337 | spin_unlock(&fmp->lock); | |
1338 | return val; | |
1339 | } | |
1340 | ||
5d21cc2d | 1341 | /* Called by cgroups to determine if a cpuset is usable; cpuset_mutex held */ |
761b3ef5 | 1342 | static int cpuset_can_attach(struct cgroup *cgrp, struct cgroup_taskset *tset) |
f780bdb7 | 1343 | { |
2f7ee569 | 1344 | struct cpuset *cs = cgroup_cs(cgrp); |
bb9d97b6 TH |
1345 | struct task_struct *task; |
1346 | int ret; | |
1da177e4 | 1347 | |
5d21cc2d TH |
1348 | mutex_lock(&cpuset_mutex); |
1349 | ||
1350 | ret = -ENOSPC; | |
300ed6cb | 1351 | if (cpumask_empty(cs->cpus_allowed) || nodes_empty(cs->mems_allowed)) |
5d21cc2d | 1352 | goto out_unlock; |
9985b0ba | 1353 | |
bb9d97b6 TH |
1354 | cgroup_taskset_for_each(task, cgrp, tset) { |
1355 | /* | |
14a40ffc TH |
1356 | * Kthreads which disallow setaffinity shouldn't be moved |
1357 | * to a new cpuset; we don't want to change their cpu | |
1358 | * affinity and isolating such threads by their set of | |
1359 | * allowed nodes is unnecessary. Thus, cpusets are not | |
1360 | * applicable for such threads. This prevents checking for | |
1361 | * success of set_cpus_allowed_ptr() on all attached tasks | |
1362 | * before cpus_allowed may be changed. | |
bb9d97b6 | 1363 | */ |
5d21cc2d | 1364 | ret = -EINVAL; |
14a40ffc | 1365 | if (task->flags & PF_NO_SETAFFINITY) |
5d21cc2d TH |
1366 | goto out_unlock; |
1367 | ret = security_task_setscheduler(task); | |
1368 | if (ret) | |
1369 | goto out_unlock; | |
bb9d97b6 | 1370 | } |
f780bdb7 | 1371 | |
452477fa TH |
1372 | /* |
1373 | * Mark attach is in progress. This makes validate_change() fail | |
1374 | * changes which zero cpus/mems_allowed. | |
1375 | */ | |
1376 | cs->attach_in_progress++; | |
5d21cc2d TH |
1377 | ret = 0; |
1378 | out_unlock: | |
1379 | mutex_unlock(&cpuset_mutex); | |
1380 | return ret; | |
8793d854 | 1381 | } |
f780bdb7 | 1382 | |
452477fa TH |
1383 | static void cpuset_cancel_attach(struct cgroup *cgrp, |
1384 | struct cgroup_taskset *tset) | |
1385 | { | |
5d21cc2d | 1386 | mutex_lock(&cpuset_mutex); |
452477fa | 1387 | cgroup_cs(cgrp)->attach_in_progress--; |
5d21cc2d | 1388 | mutex_unlock(&cpuset_mutex); |
8793d854 | 1389 | } |
1da177e4 | 1390 | |
4e4c9a14 | 1391 | /* |
5d21cc2d | 1392 | * Protected by cpuset_mutex. cpus_attach is used only by cpuset_attach() |
4e4c9a14 TH |
1393 | * but we can't allocate it dynamically there. Define it global and |
1394 | * allocate from cpuset_init(). | |
1395 | */ | |
1396 | static cpumask_var_t cpus_attach; | |
1397 | ||
761b3ef5 | 1398 | static void cpuset_attach(struct cgroup *cgrp, struct cgroup_taskset *tset) |
8793d854 | 1399 | { |
67bd2c59 | 1400 | /* static buf protected by cpuset_mutex */ |
4e4c9a14 | 1401 | static nodemask_t cpuset_attach_nodemask_to; |
8793d854 | 1402 | struct mm_struct *mm; |
bb9d97b6 TH |
1403 | struct task_struct *task; |
1404 | struct task_struct *leader = cgroup_taskset_first(tset); | |
2f7ee569 TH |
1405 | struct cgroup *oldcgrp = cgroup_taskset_cur_cgroup(tset); |
1406 | struct cpuset *cs = cgroup_cs(cgrp); | |
1407 | struct cpuset *oldcs = cgroup_cs(oldcgrp); | |
22fb52dd | 1408 | |
5d21cc2d TH |
1409 | mutex_lock(&cpuset_mutex); |
1410 | ||
4e4c9a14 TH |
1411 | /* prepare for attach */ |
1412 | if (cs == &top_cpuset) | |
1413 | cpumask_copy(cpus_attach, cpu_possible_mask); | |
1414 | else | |
1415 | guarantee_online_cpus(cs, cpus_attach); | |
1416 | ||
1417 | guarantee_online_mems(cs, &cpuset_attach_nodemask_to); | |
1418 | ||
bb9d97b6 TH |
1419 | cgroup_taskset_for_each(task, cgrp, tset) { |
1420 | /* | |
1421 | * can_attach beforehand should guarantee that this doesn't | |
1422 | * fail. TODO: have a better way to handle failure here | |
1423 | */ | |
1424 | WARN_ON_ONCE(set_cpus_allowed_ptr(task, cpus_attach)); | |
1425 | ||
1426 | cpuset_change_task_nodemask(task, &cpuset_attach_nodemask_to); | |
1427 | cpuset_update_task_spread_flag(cs, task); | |
1428 | } | |
22fb52dd | 1429 | |
f780bdb7 BB |
1430 | /* |
1431 | * Change mm, possibly for multiple threads in a threadgroup. This is | |
1432 | * expensive and may sleep. | |
1433 | */ | |
f780bdb7 | 1434 | cpuset_attach_nodemask_to = cs->mems_allowed; |
bb9d97b6 | 1435 | mm = get_task_mm(leader); |
4225399a | 1436 | if (mm) { |
f780bdb7 | 1437 | mpol_rebind_mm(mm, &cpuset_attach_nodemask_to); |
2741a559 | 1438 | if (is_memory_migrate(cs)) |
67bd2c59 | 1439 | cpuset_migrate_mm(mm, &oldcs->mems_allowed, |
f780bdb7 | 1440 | &cpuset_attach_nodemask_to); |
4225399a PJ |
1441 | mmput(mm); |
1442 | } | |
452477fa | 1443 | |
33ad801d LZ |
1444 | cs->old_mems_allowed = cpuset_attach_nodemask_to; |
1445 | ||
452477fa | 1446 | cs->attach_in_progress--; |
e44193d3 LZ |
1447 | if (!cs->attach_in_progress) |
1448 | wake_up(&cpuset_attach_wq); | |
5d21cc2d TH |
1449 | |
1450 | mutex_unlock(&cpuset_mutex); | |
1da177e4 LT |
1451 | } |
1452 | ||
1453 | /* The various types of files and directories in a cpuset file system */ | |
1454 | ||
1455 | typedef enum { | |
45b07ef3 | 1456 | FILE_MEMORY_MIGRATE, |
1da177e4 LT |
1457 | FILE_CPULIST, |
1458 | FILE_MEMLIST, | |
1459 | FILE_CPU_EXCLUSIVE, | |
1460 | FILE_MEM_EXCLUSIVE, | |
78608366 | 1461 | FILE_MEM_HARDWALL, |
029190c5 | 1462 | FILE_SCHED_LOAD_BALANCE, |
1d3504fc | 1463 | FILE_SCHED_RELAX_DOMAIN_LEVEL, |
3e0d98b9 PJ |
1464 | FILE_MEMORY_PRESSURE_ENABLED, |
1465 | FILE_MEMORY_PRESSURE, | |
825a46af PJ |
1466 | FILE_SPREAD_PAGE, |
1467 | FILE_SPREAD_SLAB, | |
1da177e4 LT |
1468 | } cpuset_filetype_t; |
1469 | ||
700fe1ab PM |
1470 | static int cpuset_write_u64(struct cgroup *cgrp, struct cftype *cft, u64 val) |
1471 | { | |
700fe1ab PM |
1472 | struct cpuset *cs = cgroup_cs(cgrp); |
1473 | cpuset_filetype_t type = cft->private; | |
5d21cc2d | 1474 | int retval = -ENODEV; |
700fe1ab | 1475 | |
5d21cc2d TH |
1476 | mutex_lock(&cpuset_mutex); |
1477 | if (!is_cpuset_online(cs)) | |
1478 | goto out_unlock; | |
700fe1ab PM |
1479 | |
1480 | switch (type) { | |
1da177e4 | 1481 | case FILE_CPU_EXCLUSIVE: |
700fe1ab | 1482 | retval = update_flag(CS_CPU_EXCLUSIVE, cs, val); |
1da177e4 LT |
1483 | break; |
1484 | case FILE_MEM_EXCLUSIVE: | |
700fe1ab | 1485 | retval = update_flag(CS_MEM_EXCLUSIVE, cs, val); |
1da177e4 | 1486 | break; |
78608366 PM |
1487 | case FILE_MEM_HARDWALL: |
1488 | retval = update_flag(CS_MEM_HARDWALL, cs, val); | |
1489 | break; | |
029190c5 | 1490 | case FILE_SCHED_LOAD_BALANCE: |
700fe1ab | 1491 | retval = update_flag(CS_SCHED_LOAD_BALANCE, cs, val); |
1d3504fc | 1492 | break; |
45b07ef3 | 1493 | case FILE_MEMORY_MIGRATE: |
700fe1ab | 1494 | retval = update_flag(CS_MEMORY_MIGRATE, cs, val); |
45b07ef3 | 1495 | break; |
3e0d98b9 | 1496 | case FILE_MEMORY_PRESSURE_ENABLED: |
700fe1ab | 1497 | cpuset_memory_pressure_enabled = !!val; |
3e0d98b9 PJ |
1498 | break; |
1499 | case FILE_MEMORY_PRESSURE: | |
1500 | retval = -EACCES; | |
1501 | break; | |
825a46af | 1502 | case FILE_SPREAD_PAGE: |
700fe1ab | 1503 | retval = update_flag(CS_SPREAD_PAGE, cs, val); |
825a46af PJ |
1504 | break; |
1505 | case FILE_SPREAD_SLAB: | |
700fe1ab | 1506 | retval = update_flag(CS_SPREAD_SLAB, cs, val); |
825a46af | 1507 | break; |
1da177e4 LT |
1508 | default: |
1509 | retval = -EINVAL; | |
700fe1ab | 1510 | break; |
1da177e4 | 1511 | } |
5d21cc2d TH |
1512 | out_unlock: |
1513 | mutex_unlock(&cpuset_mutex); | |
1da177e4 LT |
1514 | return retval; |
1515 | } | |
1516 | ||
5be7a479 PM |
1517 | static int cpuset_write_s64(struct cgroup *cgrp, struct cftype *cft, s64 val) |
1518 | { | |
5be7a479 PM |
1519 | struct cpuset *cs = cgroup_cs(cgrp); |
1520 | cpuset_filetype_t type = cft->private; | |
5d21cc2d | 1521 | int retval = -ENODEV; |
5be7a479 | 1522 | |
5d21cc2d TH |
1523 | mutex_lock(&cpuset_mutex); |
1524 | if (!is_cpuset_online(cs)) | |
1525 | goto out_unlock; | |
e3712395 | 1526 | |
5be7a479 PM |
1527 | switch (type) { |
1528 | case FILE_SCHED_RELAX_DOMAIN_LEVEL: | |
1529 | retval = update_relax_domain_level(cs, val); | |
1530 | break; | |
1531 | default: | |
1532 | retval = -EINVAL; | |
1533 | break; | |
1534 | } | |
5d21cc2d TH |
1535 | out_unlock: |
1536 | mutex_unlock(&cpuset_mutex); | |
5be7a479 PM |
1537 | return retval; |
1538 | } | |
1539 | ||
e3712395 PM |
1540 | /* |
1541 | * Common handling for a write to a "cpus" or "mems" file. | |
1542 | */ | |
1543 | static int cpuset_write_resmask(struct cgroup *cgrp, struct cftype *cft, | |
1544 | const char *buf) | |
1545 | { | |
645fcc9d LZ |
1546 | struct cpuset *cs = cgroup_cs(cgrp); |
1547 | struct cpuset *trialcs; | |
5d21cc2d | 1548 | int retval = -ENODEV; |
e3712395 | 1549 | |
3a5a6d0c TH |
1550 | /* |
1551 | * CPU or memory hotunplug may leave @cs w/o any execution | |
1552 | * resources, in which case the hotplug code asynchronously updates | |
1553 | * configuration and transfers all tasks to the nearest ancestor | |
1554 | * which can execute. | |
1555 | * | |
1556 | * As writes to "cpus" or "mems" may restore @cs's execution | |
1557 | * resources, wait for the previously scheduled operations before | |
1558 | * proceeding, so that we don't end up keep removing tasks added | |
1559 | * after execution capability is restored. | |
1560 | */ | |
1561 | flush_work(&cpuset_hotplug_work); | |
1562 | ||
5d21cc2d TH |
1563 | mutex_lock(&cpuset_mutex); |
1564 | if (!is_cpuset_online(cs)) | |
1565 | goto out_unlock; | |
e3712395 | 1566 | |
645fcc9d | 1567 | trialcs = alloc_trial_cpuset(cs); |
b75f38d6 LZ |
1568 | if (!trialcs) { |
1569 | retval = -ENOMEM; | |
5d21cc2d | 1570 | goto out_unlock; |
b75f38d6 | 1571 | } |
645fcc9d | 1572 | |
e3712395 PM |
1573 | switch (cft->private) { |
1574 | case FILE_CPULIST: | |
645fcc9d | 1575 | retval = update_cpumask(cs, trialcs, buf); |
e3712395 PM |
1576 | break; |
1577 | case FILE_MEMLIST: | |
645fcc9d | 1578 | retval = update_nodemask(cs, trialcs, buf); |
e3712395 PM |
1579 | break; |
1580 | default: | |
1581 | retval = -EINVAL; | |
1582 | break; | |
1583 | } | |
645fcc9d LZ |
1584 | |
1585 | free_trial_cpuset(trialcs); | |
5d21cc2d TH |
1586 | out_unlock: |
1587 | mutex_unlock(&cpuset_mutex); | |
e3712395 PM |
1588 | return retval; |
1589 | } | |
1590 | ||
1da177e4 LT |
1591 | /* |
1592 | * These ascii lists should be read in a single call, by using a user | |
1593 | * buffer large enough to hold the entire map. If read in smaller | |
1594 | * chunks, there is no guarantee of atomicity. Since the display format | |
1595 | * used, list of ranges of sequential numbers, is variable length, | |
1596 | * and since these maps can change value dynamically, one could read | |
1597 | * gibberish by doing partial reads while a list was changing. | |
1598 | * A single large read to a buffer that crosses a page boundary is | |
1599 | * ok, because the result being copied to user land is not recomputed | |
1600 | * across a page fault. | |
1601 | */ | |
1602 | ||
9303e0c4 | 1603 | static size_t cpuset_sprintf_cpulist(char *page, struct cpuset *cs) |
1da177e4 | 1604 | { |
9303e0c4 | 1605 | size_t count; |
1da177e4 | 1606 | |
3d3f26a7 | 1607 | mutex_lock(&callback_mutex); |
9303e0c4 | 1608 | count = cpulist_scnprintf(page, PAGE_SIZE, cs->cpus_allowed); |
3d3f26a7 | 1609 | mutex_unlock(&callback_mutex); |
1da177e4 | 1610 | |
9303e0c4 | 1611 | return count; |
1da177e4 LT |
1612 | } |
1613 | ||
9303e0c4 | 1614 | static size_t cpuset_sprintf_memlist(char *page, struct cpuset *cs) |
1da177e4 | 1615 | { |
9303e0c4 | 1616 | size_t count; |
1da177e4 | 1617 | |
3d3f26a7 | 1618 | mutex_lock(&callback_mutex); |
9303e0c4 | 1619 | count = nodelist_scnprintf(page, PAGE_SIZE, cs->mems_allowed); |
3d3f26a7 | 1620 | mutex_unlock(&callback_mutex); |
1da177e4 | 1621 | |
9303e0c4 | 1622 | return count; |
1da177e4 LT |
1623 | } |
1624 | ||
8793d854 PM |
1625 | static ssize_t cpuset_common_file_read(struct cgroup *cont, |
1626 | struct cftype *cft, | |
1627 | struct file *file, | |
1628 | char __user *buf, | |
1629 | size_t nbytes, loff_t *ppos) | |
1da177e4 | 1630 | { |
8793d854 | 1631 | struct cpuset *cs = cgroup_cs(cont); |
1da177e4 LT |
1632 | cpuset_filetype_t type = cft->private; |
1633 | char *page; | |
1634 | ssize_t retval = 0; | |
1635 | char *s; | |
1da177e4 | 1636 | |
e12ba74d | 1637 | if (!(page = (char *)__get_free_page(GFP_TEMPORARY))) |
1da177e4 LT |
1638 | return -ENOMEM; |
1639 | ||
1640 | s = page; | |
1641 | ||
1642 | switch (type) { | |
1643 | case FILE_CPULIST: | |
1644 | s += cpuset_sprintf_cpulist(s, cs); | |
1645 | break; | |
1646 | case FILE_MEMLIST: | |
1647 | s += cpuset_sprintf_memlist(s, cs); | |
1648 | break; | |
1da177e4 LT |
1649 | default: |
1650 | retval = -EINVAL; | |
1651 | goto out; | |
1652 | } | |
1653 | *s++ = '\n'; | |
1da177e4 | 1654 | |
eacaa1f5 | 1655 | retval = simple_read_from_buffer(buf, nbytes, ppos, page, s - page); |
1da177e4 LT |
1656 | out: |
1657 | free_page((unsigned long)page); | |
1658 | return retval; | |
1659 | } | |
1660 | ||
700fe1ab PM |
1661 | static u64 cpuset_read_u64(struct cgroup *cont, struct cftype *cft) |
1662 | { | |
1663 | struct cpuset *cs = cgroup_cs(cont); | |
1664 | cpuset_filetype_t type = cft->private; | |
1665 | switch (type) { | |
1666 | case FILE_CPU_EXCLUSIVE: | |
1667 | return is_cpu_exclusive(cs); | |
1668 | case FILE_MEM_EXCLUSIVE: | |
1669 | return is_mem_exclusive(cs); | |
78608366 PM |
1670 | case FILE_MEM_HARDWALL: |
1671 | return is_mem_hardwall(cs); | |
700fe1ab PM |
1672 | case FILE_SCHED_LOAD_BALANCE: |
1673 | return is_sched_load_balance(cs); | |
1674 | case FILE_MEMORY_MIGRATE: | |
1675 | return is_memory_migrate(cs); | |
1676 | case FILE_MEMORY_PRESSURE_ENABLED: | |
1677 | return cpuset_memory_pressure_enabled; | |
1678 | case FILE_MEMORY_PRESSURE: | |
1679 | return fmeter_getrate(&cs->fmeter); | |
1680 | case FILE_SPREAD_PAGE: | |
1681 | return is_spread_page(cs); | |
1682 | case FILE_SPREAD_SLAB: | |
1683 | return is_spread_slab(cs); | |
1684 | default: | |
1685 | BUG(); | |
1686 | } | |
cf417141 MK |
1687 | |
1688 | /* Unreachable but makes gcc happy */ | |
1689 | return 0; | |
700fe1ab | 1690 | } |
1da177e4 | 1691 | |
5be7a479 PM |
1692 | static s64 cpuset_read_s64(struct cgroup *cont, struct cftype *cft) |
1693 | { | |
1694 | struct cpuset *cs = cgroup_cs(cont); | |
1695 | cpuset_filetype_t type = cft->private; | |
1696 | switch (type) { | |
1697 | case FILE_SCHED_RELAX_DOMAIN_LEVEL: | |
1698 | return cs->relax_domain_level; | |
1699 | default: | |
1700 | BUG(); | |
1701 | } | |
cf417141 MK |
1702 | |
1703 | /* Unrechable but makes gcc happy */ | |
1704 | return 0; | |
5be7a479 PM |
1705 | } |
1706 | ||
1da177e4 LT |
1707 | |
1708 | /* | |
1709 | * for the common functions, 'private' gives the type of file | |
1710 | */ | |
1711 | ||
addf2c73 PM |
1712 | static struct cftype files[] = { |
1713 | { | |
1714 | .name = "cpus", | |
1715 | .read = cpuset_common_file_read, | |
e3712395 PM |
1716 | .write_string = cpuset_write_resmask, |
1717 | .max_write_len = (100U + 6 * NR_CPUS), | |
addf2c73 PM |
1718 | .private = FILE_CPULIST, |
1719 | }, | |
1720 | ||
1721 | { | |
1722 | .name = "mems", | |
1723 | .read = cpuset_common_file_read, | |
e3712395 PM |
1724 | .write_string = cpuset_write_resmask, |
1725 | .max_write_len = (100U + 6 * MAX_NUMNODES), | |
addf2c73 PM |
1726 | .private = FILE_MEMLIST, |
1727 | }, | |
1728 | ||
1729 | { | |
1730 | .name = "cpu_exclusive", | |
1731 | .read_u64 = cpuset_read_u64, | |
1732 | .write_u64 = cpuset_write_u64, | |
1733 | .private = FILE_CPU_EXCLUSIVE, | |
1734 | }, | |
1735 | ||
1736 | { | |
1737 | .name = "mem_exclusive", | |
1738 | .read_u64 = cpuset_read_u64, | |
1739 | .write_u64 = cpuset_write_u64, | |
1740 | .private = FILE_MEM_EXCLUSIVE, | |
1741 | }, | |
1742 | ||
78608366 PM |
1743 | { |
1744 | .name = "mem_hardwall", | |
1745 | .read_u64 = cpuset_read_u64, | |
1746 | .write_u64 = cpuset_write_u64, | |
1747 | .private = FILE_MEM_HARDWALL, | |
1748 | }, | |
1749 | ||
addf2c73 PM |
1750 | { |
1751 | .name = "sched_load_balance", | |
1752 | .read_u64 = cpuset_read_u64, | |
1753 | .write_u64 = cpuset_write_u64, | |
1754 | .private = FILE_SCHED_LOAD_BALANCE, | |
1755 | }, | |
1756 | ||
1757 | { | |
1758 | .name = "sched_relax_domain_level", | |
5be7a479 PM |
1759 | .read_s64 = cpuset_read_s64, |
1760 | .write_s64 = cpuset_write_s64, | |
addf2c73 PM |
1761 | .private = FILE_SCHED_RELAX_DOMAIN_LEVEL, |
1762 | }, | |
1763 | ||
1764 | { | |
1765 | .name = "memory_migrate", | |
1766 | .read_u64 = cpuset_read_u64, | |
1767 | .write_u64 = cpuset_write_u64, | |
1768 | .private = FILE_MEMORY_MIGRATE, | |
1769 | }, | |
1770 | ||
1771 | { | |
1772 | .name = "memory_pressure", | |
1773 | .read_u64 = cpuset_read_u64, | |
1774 | .write_u64 = cpuset_write_u64, | |
1775 | .private = FILE_MEMORY_PRESSURE, | |
099fca32 | 1776 | .mode = S_IRUGO, |
addf2c73 PM |
1777 | }, |
1778 | ||
1779 | { | |
1780 | .name = "memory_spread_page", | |
1781 | .read_u64 = cpuset_read_u64, | |
1782 | .write_u64 = cpuset_write_u64, | |
1783 | .private = FILE_SPREAD_PAGE, | |
1784 | }, | |
1785 | ||
1786 | { | |
1787 | .name = "memory_spread_slab", | |
1788 | .read_u64 = cpuset_read_u64, | |
1789 | .write_u64 = cpuset_write_u64, | |
1790 | .private = FILE_SPREAD_SLAB, | |
1791 | }, | |
3e0d98b9 | 1792 | |
4baf6e33 TH |
1793 | { |
1794 | .name = "memory_pressure_enabled", | |
1795 | .flags = CFTYPE_ONLY_ON_ROOT, | |
1796 | .read_u64 = cpuset_read_u64, | |
1797 | .write_u64 = cpuset_write_u64, | |
1798 | .private = FILE_MEMORY_PRESSURE_ENABLED, | |
1799 | }, | |
1da177e4 | 1800 | |
4baf6e33 TH |
1801 | { } /* terminate */ |
1802 | }; | |
1da177e4 LT |
1803 | |
1804 | /* | |
92fb9748 | 1805 | * cpuset_css_alloc - allocate a cpuset css |
2df167a3 | 1806 | * cont: control group that the new cpuset will be part of |
1da177e4 LT |
1807 | */ |
1808 | ||
92fb9748 | 1809 | static struct cgroup_subsys_state *cpuset_css_alloc(struct cgroup *cont) |
1da177e4 | 1810 | { |
c8f699bb | 1811 | struct cpuset *cs; |
1da177e4 | 1812 | |
c8f699bb | 1813 | if (!cont->parent) |
8793d854 | 1814 | return &top_cpuset.css; |
033fa1c5 | 1815 | |
c8f699bb | 1816 | cs = kzalloc(sizeof(*cs), GFP_KERNEL); |
1da177e4 | 1817 | if (!cs) |
8793d854 | 1818 | return ERR_PTR(-ENOMEM); |
300ed6cb LZ |
1819 | if (!alloc_cpumask_var(&cs->cpus_allowed, GFP_KERNEL)) { |
1820 | kfree(cs); | |
1821 | return ERR_PTR(-ENOMEM); | |
1822 | } | |
1da177e4 | 1823 | |
029190c5 | 1824 | set_bit(CS_SCHED_LOAD_BALANCE, &cs->flags); |
300ed6cb | 1825 | cpumask_clear(cs->cpus_allowed); |
f9a86fcb | 1826 | nodes_clear(cs->mems_allowed); |
3e0d98b9 | 1827 | fmeter_init(&cs->fmeter); |
1d3504fc | 1828 | cs->relax_domain_level = -1; |
1da177e4 | 1829 | |
c8f699bb TH |
1830 | return &cs->css; |
1831 | } | |
1832 | ||
1833 | static int cpuset_css_online(struct cgroup *cgrp) | |
1834 | { | |
1835 | struct cpuset *cs = cgroup_cs(cgrp); | |
c431069f | 1836 | struct cpuset *parent = parent_cs(cs); |
ae8086ce TH |
1837 | struct cpuset *tmp_cs; |
1838 | struct cgroup *pos_cg; | |
c8f699bb TH |
1839 | |
1840 | if (!parent) | |
1841 | return 0; | |
1842 | ||
5d21cc2d TH |
1843 | mutex_lock(&cpuset_mutex); |
1844 | ||
efeb77b2 | 1845 | set_bit(CS_ONLINE, &cs->flags); |
c8f699bb TH |
1846 | if (is_spread_page(parent)) |
1847 | set_bit(CS_SPREAD_PAGE, &cs->flags); | |
1848 | if (is_spread_slab(parent)) | |
1849 | set_bit(CS_SPREAD_SLAB, &cs->flags); | |
1da177e4 | 1850 | |
202f72d5 | 1851 | number_of_cpusets++; |
033fa1c5 | 1852 | |
c8f699bb | 1853 | if (!test_bit(CGRP_CPUSET_CLONE_CHILDREN, &cgrp->flags)) |
5d21cc2d | 1854 | goto out_unlock; |
033fa1c5 TH |
1855 | |
1856 | /* | |
1857 | * Clone @parent's configuration if CGRP_CPUSET_CLONE_CHILDREN is | |
1858 | * set. This flag handling is implemented in cgroup core for | |
1859 | * histrical reasons - the flag may be specified during mount. | |
1860 | * | |
1861 | * Currently, if any sibling cpusets have exclusive cpus or mem, we | |
1862 | * refuse to clone the configuration - thereby refusing the task to | |
1863 | * be entered, and as a result refusing the sys_unshare() or | |
1864 | * clone() which initiated it. If this becomes a problem for some | |
1865 | * users who wish to allow that scenario, then this could be | |
1866 | * changed to grant parent->cpus_allowed-sibling_cpus_exclusive | |
1867 | * (and likewise for mems) to the new cgroup. | |
1868 | */ | |
ae8086ce TH |
1869 | rcu_read_lock(); |
1870 | cpuset_for_each_child(tmp_cs, pos_cg, parent) { | |
1871 | if (is_mem_exclusive(tmp_cs) || is_cpu_exclusive(tmp_cs)) { | |
1872 | rcu_read_unlock(); | |
5d21cc2d | 1873 | goto out_unlock; |
ae8086ce | 1874 | } |
033fa1c5 | 1875 | } |
ae8086ce | 1876 | rcu_read_unlock(); |
033fa1c5 TH |
1877 | |
1878 | mutex_lock(&callback_mutex); | |
1879 | cs->mems_allowed = parent->mems_allowed; | |
1880 | cpumask_copy(cs->cpus_allowed, parent->cpus_allowed); | |
1881 | mutex_unlock(&callback_mutex); | |
5d21cc2d TH |
1882 | out_unlock: |
1883 | mutex_unlock(&cpuset_mutex); | |
c8f699bb TH |
1884 | return 0; |
1885 | } | |
1886 | ||
1887 | static void cpuset_css_offline(struct cgroup *cgrp) | |
1888 | { | |
1889 | struct cpuset *cs = cgroup_cs(cgrp); | |
1890 | ||
5d21cc2d | 1891 | mutex_lock(&cpuset_mutex); |
c8f699bb TH |
1892 | |
1893 | if (is_sched_load_balance(cs)) | |
1894 | update_flag(CS_SCHED_LOAD_BALANCE, cs, 0); | |
1895 | ||
1896 | number_of_cpusets--; | |
efeb77b2 | 1897 | clear_bit(CS_ONLINE, &cs->flags); |
c8f699bb | 1898 | |
5d21cc2d | 1899 | mutex_unlock(&cpuset_mutex); |
1da177e4 LT |
1900 | } |
1901 | ||
029190c5 | 1902 | /* |
029190c5 PJ |
1903 | * If the cpuset being removed has its flag 'sched_load_balance' |
1904 | * enabled, then simulate turning sched_load_balance off, which | |
699140ba | 1905 | * will call rebuild_sched_domains_locked(). |
029190c5 PJ |
1906 | */ |
1907 | ||
92fb9748 | 1908 | static void cpuset_css_free(struct cgroup *cont) |
1da177e4 | 1909 | { |
8793d854 | 1910 | struct cpuset *cs = cgroup_cs(cont); |
1da177e4 | 1911 | |
300ed6cb | 1912 | free_cpumask_var(cs->cpus_allowed); |
8793d854 | 1913 | kfree(cs); |
1da177e4 LT |
1914 | } |
1915 | ||
8793d854 PM |
1916 | struct cgroup_subsys cpuset_subsys = { |
1917 | .name = "cpuset", | |
92fb9748 | 1918 | .css_alloc = cpuset_css_alloc, |
c8f699bb TH |
1919 | .css_online = cpuset_css_online, |
1920 | .css_offline = cpuset_css_offline, | |
92fb9748 | 1921 | .css_free = cpuset_css_free, |
8793d854 | 1922 | .can_attach = cpuset_can_attach, |
452477fa | 1923 | .cancel_attach = cpuset_cancel_attach, |
8793d854 | 1924 | .attach = cpuset_attach, |
8793d854 | 1925 | .subsys_id = cpuset_subsys_id, |
4baf6e33 | 1926 | .base_cftypes = files, |
8793d854 PM |
1927 | .early_init = 1, |
1928 | }; | |
1929 | ||
1da177e4 LT |
1930 | /** |
1931 | * cpuset_init - initialize cpusets at system boot | |
1932 | * | |
1933 | * Description: Initialize top_cpuset and the cpuset internal file system, | |
1934 | **/ | |
1935 | ||
1936 | int __init cpuset_init(void) | |
1937 | { | |
8793d854 | 1938 | int err = 0; |
1da177e4 | 1939 | |
58568d2a MX |
1940 | if (!alloc_cpumask_var(&top_cpuset.cpus_allowed, GFP_KERNEL)) |
1941 | BUG(); | |
1942 | ||
300ed6cb | 1943 | cpumask_setall(top_cpuset.cpus_allowed); |
f9a86fcb | 1944 | nodes_setall(top_cpuset.mems_allowed); |
1da177e4 | 1945 | |
3e0d98b9 | 1946 | fmeter_init(&top_cpuset.fmeter); |
029190c5 | 1947 | set_bit(CS_SCHED_LOAD_BALANCE, &top_cpuset.flags); |
1d3504fc | 1948 | top_cpuset.relax_domain_level = -1; |
1da177e4 | 1949 | |
1da177e4 LT |
1950 | err = register_filesystem(&cpuset_fs_type); |
1951 | if (err < 0) | |
8793d854 PM |
1952 | return err; |
1953 | ||
2341d1b6 LZ |
1954 | if (!alloc_cpumask_var(&cpus_attach, GFP_KERNEL)) |
1955 | BUG(); | |
1956 | ||
202f72d5 | 1957 | number_of_cpusets = 1; |
8793d854 | 1958 | return 0; |
1da177e4 LT |
1959 | } |
1960 | ||
b1aac8bb | 1961 | /* |
cf417141 | 1962 | * If CPU and/or memory hotplug handlers, below, unplug any CPUs |
b1aac8bb PJ |
1963 | * or memory nodes, we need to walk over the cpuset hierarchy, |
1964 | * removing that CPU or node from all cpusets. If this removes the | |
956db3ca CW |
1965 | * last CPU or node from a cpuset, then move the tasks in the empty |
1966 | * cpuset to its next-highest non-empty parent. | |
b1aac8bb | 1967 | */ |
956db3ca CW |
1968 | static void remove_tasks_in_empty_cpuset(struct cpuset *cs) |
1969 | { | |
1970 | struct cpuset *parent; | |
1971 | ||
956db3ca CW |
1972 | /* |
1973 | * Find its next-highest non-empty parent, (top cpuset | |
1974 | * has online cpus, so can't be empty). | |
1975 | */ | |
c431069f | 1976 | parent = parent_cs(cs); |
300ed6cb | 1977 | while (cpumask_empty(parent->cpus_allowed) || |
b4501295 | 1978 | nodes_empty(parent->mems_allowed)) |
c431069f | 1979 | parent = parent_cs(parent); |
956db3ca | 1980 | |
8cc99345 TH |
1981 | if (cgroup_transfer_tasks(parent->css.cgroup, cs->css.cgroup)) { |
1982 | rcu_read_lock(); | |
1983 | printk(KERN_ERR "cpuset: failed to transfer tasks out of empty cpuset %s\n", | |
1984 | cgroup_name(cs->css.cgroup)); | |
1985 | rcu_read_unlock(); | |
1986 | } | |
956db3ca CW |
1987 | } |
1988 | ||
deb7aa30 | 1989 | /** |
388afd85 | 1990 | * cpuset_hotplug_update_tasks - update tasks in a cpuset for hotunplug |
deb7aa30 | 1991 | * @cs: cpuset in interest |
956db3ca | 1992 | * |
deb7aa30 TH |
1993 | * Compare @cs's cpu and mem masks against top_cpuset and if some have gone |
1994 | * offline, update @cs accordingly. If @cs ends up with no CPU or memory, | |
1995 | * all its tasks are moved to the nearest ancestor with both resources. | |
80d1fa64 | 1996 | */ |
388afd85 | 1997 | static void cpuset_hotplug_update_tasks(struct cpuset *cs) |
80d1fa64 | 1998 | { |
deb7aa30 | 1999 | static cpumask_t off_cpus; |
33ad801d | 2000 | static nodemask_t off_mems; |
5d21cc2d | 2001 | bool is_empty; |
80d1fa64 | 2002 | |
e44193d3 LZ |
2003 | retry: |
2004 | wait_event(cpuset_attach_wq, cs->attach_in_progress == 0); | |
2005 | ||
5d21cc2d | 2006 | mutex_lock(&cpuset_mutex); |
7ddf96b0 | 2007 | |
e44193d3 LZ |
2008 | /* |
2009 | * We have raced with task attaching. We wait until attaching | |
2010 | * is finished, so we won't attach a task to an empty cpuset. | |
2011 | */ | |
2012 | if (cs->attach_in_progress) { | |
2013 | mutex_unlock(&cpuset_mutex); | |
2014 | goto retry; | |
2015 | } | |
2016 | ||
deb7aa30 TH |
2017 | cpumask_andnot(&off_cpus, cs->cpus_allowed, top_cpuset.cpus_allowed); |
2018 | nodes_andnot(off_mems, cs->mems_allowed, top_cpuset.mems_allowed); | |
80d1fa64 | 2019 | |
deb7aa30 TH |
2020 | /* remove offline cpus from @cs */ |
2021 | if (!cpumask_empty(&off_cpus)) { | |
2022 | mutex_lock(&callback_mutex); | |
2023 | cpumask_andnot(cs->cpus_allowed, cs->cpus_allowed, &off_cpus); | |
2024 | mutex_unlock(&callback_mutex); | |
2025 | update_tasks_cpumask(cs, NULL); | |
80d1fa64 SB |
2026 | } |
2027 | ||
deb7aa30 TH |
2028 | /* remove offline mems from @cs */ |
2029 | if (!nodes_empty(off_mems)) { | |
deb7aa30 TH |
2030 | mutex_lock(&callback_mutex); |
2031 | nodes_andnot(cs->mems_allowed, cs->mems_allowed, off_mems); | |
2032 | mutex_unlock(&callback_mutex); | |
33ad801d | 2033 | update_tasks_nodemask(cs, NULL); |
b1aac8bb | 2034 | } |
deb7aa30 | 2035 | |
5d21cc2d TH |
2036 | is_empty = cpumask_empty(cs->cpus_allowed) || |
2037 | nodes_empty(cs->mems_allowed); | |
8d033948 | 2038 | |
5d21cc2d TH |
2039 | mutex_unlock(&cpuset_mutex); |
2040 | ||
2041 | /* | |
2042 | * If @cs became empty, move tasks to the nearest ancestor with | |
2043 | * execution resources. This is full cgroup operation which will | |
2044 | * also call back into cpuset. Should be done outside any lock. | |
2045 | */ | |
2046 | if (is_empty) | |
2047 | remove_tasks_in_empty_cpuset(cs); | |
b1aac8bb PJ |
2048 | } |
2049 | ||
deb7aa30 | 2050 | /** |
3a5a6d0c | 2051 | * cpuset_hotplug_workfn - handle CPU/memory hotunplug for a cpuset |
956db3ca | 2052 | * |
deb7aa30 TH |
2053 | * This function is called after either CPU or memory configuration has |
2054 | * changed and updates cpuset accordingly. The top_cpuset is always | |
2055 | * synchronized to cpu_active_mask and N_MEMORY, which is necessary in | |
2056 | * order to make cpusets transparent (of no affect) on systems that are | |
2057 | * actively using CPU hotplug but making no active use of cpusets. | |
956db3ca | 2058 | * |
deb7aa30 | 2059 | * Non-root cpusets are only affected by offlining. If any CPUs or memory |
388afd85 LZ |
2060 | * nodes have been taken down, cpuset_hotplug_update_tasks() is invoked on |
2061 | * all descendants. | |
956db3ca | 2062 | * |
deb7aa30 TH |
2063 | * Note that CPU offlining during suspend is ignored. We don't modify |
2064 | * cpusets across suspend/resume cycles at all. | |
956db3ca | 2065 | */ |
3a5a6d0c | 2066 | static void cpuset_hotplug_workfn(struct work_struct *work) |
b1aac8bb | 2067 | { |
deb7aa30 TH |
2068 | static cpumask_t new_cpus, tmp_cpus; |
2069 | static nodemask_t new_mems, tmp_mems; | |
2070 | bool cpus_updated, mems_updated; | |
2071 | bool cpus_offlined, mems_offlined; | |
b1aac8bb | 2072 | |
5d21cc2d | 2073 | mutex_lock(&cpuset_mutex); |
956db3ca | 2074 | |
deb7aa30 TH |
2075 | /* fetch the available cpus/mems and find out which changed how */ |
2076 | cpumask_copy(&new_cpus, cpu_active_mask); | |
2077 | new_mems = node_states[N_MEMORY]; | |
7ddf96b0 | 2078 | |
deb7aa30 TH |
2079 | cpus_updated = !cpumask_equal(top_cpuset.cpus_allowed, &new_cpus); |
2080 | cpus_offlined = cpumask_andnot(&tmp_cpus, top_cpuset.cpus_allowed, | |
2081 | &new_cpus); | |
7ddf96b0 | 2082 | |
deb7aa30 TH |
2083 | mems_updated = !nodes_equal(top_cpuset.mems_allowed, new_mems); |
2084 | nodes_andnot(tmp_mems, top_cpuset.mems_allowed, new_mems); | |
2085 | mems_offlined = !nodes_empty(tmp_mems); | |
7ddf96b0 | 2086 | |
deb7aa30 TH |
2087 | /* synchronize cpus_allowed to cpu_active_mask */ |
2088 | if (cpus_updated) { | |
2089 | mutex_lock(&callback_mutex); | |
2090 | cpumask_copy(top_cpuset.cpus_allowed, &new_cpus); | |
2091 | mutex_unlock(&callback_mutex); | |
2092 | /* we don't mess with cpumasks of tasks in top_cpuset */ | |
2093 | } | |
b4501295 | 2094 | |
deb7aa30 TH |
2095 | /* synchronize mems_allowed to N_MEMORY */ |
2096 | if (mems_updated) { | |
deb7aa30 TH |
2097 | mutex_lock(&callback_mutex); |
2098 | top_cpuset.mems_allowed = new_mems; | |
2099 | mutex_unlock(&callback_mutex); | |
33ad801d | 2100 | update_tasks_nodemask(&top_cpuset, NULL); |
deb7aa30 | 2101 | } |
b4501295 | 2102 | |
388afd85 LZ |
2103 | mutex_unlock(&cpuset_mutex); |
2104 | ||
deb7aa30 TH |
2105 | /* if cpus or mems went down, we need to propagate to descendants */ |
2106 | if (cpus_offlined || mems_offlined) { | |
2107 | struct cpuset *cs; | |
fc560a26 | 2108 | struct cgroup *pos_cgrp; |
f9b4fb8d | 2109 | |
fc560a26 | 2110 | rcu_read_lock(); |
388afd85 LZ |
2111 | cpuset_for_each_descendant_pre(cs, pos_cgrp, &top_cpuset) { |
2112 | if (!css_tryget(&cs->css)) | |
2113 | continue; | |
2114 | rcu_read_unlock(); | |
7ddf96b0 | 2115 | |
388afd85 | 2116 | cpuset_hotplug_update_tasks(cs); |
b4501295 | 2117 | |
388afd85 LZ |
2118 | rcu_read_lock(); |
2119 | css_put(&cs->css); | |
2120 | } | |
2121 | rcu_read_unlock(); | |
2122 | } | |
8d033948 | 2123 | |
deb7aa30 | 2124 | /* rebuild sched domains if cpus_allowed has changed */ |
e0e80a02 LZ |
2125 | if (cpus_updated) |
2126 | rebuild_sched_domains(); | |
b1aac8bb PJ |
2127 | } |
2128 | ||
7ddf96b0 | 2129 | void cpuset_update_active_cpus(bool cpu_online) |
4c4d50f7 | 2130 | { |
3a5a6d0c TH |
2131 | /* |
2132 | * We're inside cpu hotplug critical region which usually nests | |
2133 | * inside cgroup synchronization. Bounce actual hotplug processing | |
2134 | * to a work item to avoid reverse locking order. | |
2135 | * | |
2136 | * We still need to do partition_sched_domains() synchronously; | |
2137 | * otherwise, the scheduler will get confused and put tasks to the | |
2138 | * dead CPU. Fall back to the default single domain. | |
2139 | * cpuset_hotplug_workfn() will rebuild it as necessary. | |
2140 | */ | |
2141 | partition_sched_domains(1, NULL, NULL); | |
2142 | schedule_work(&cpuset_hotplug_work); | |
4c4d50f7 | 2143 | } |
4c4d50f7 | 2144 | |
38837fc7 | 2145 | /* |
38d7bee9 LJ |
2146 | * Keep top_cpuset.mems_allowed tracking node_states[N_MEMORY]. |
2147 | * Call this routine anytime after node_states[N_MEMORY] changes. | |
a1cd2b13 | 2148 | * See cpuset_update_active_cpus() for CPU hotplug handling. |
38837fc7 | 2149 | */ |
f481891f MX |
2150 | static int cpuset_track_online_nodes(struct notifier_block *self, |
2151 | unsigned long action, void *arg) | |
38837fc7 | 2152 | { |
3a5a6d0c | 2153 | schedule_work(&cpuset_hotplug_work); |
f481891f | 2154 | return NOTIFY_OK; |
38837fc7 | 2155 | } |
d8f10cb3 AM |
2156 | |
2157 | static struct notifier_block cpuset_track_online_nodes_nb = { | |
2158 | .notifier_call = cpuset_track_online_nodes, | |
2159 | .priority = 10, /* ??! */ | |
2160 | }; | |
38837fc7 | 2161 | |
1da177e4 LT |
2162 | /** |
2163 | * cpuset_init_smp - initialize cpus_allowed | |
2164 | * | |
2165 | * Description: Finish top cpuset after cpu, node maps are initialized | |
d8f10cb3 | 2166 | */ |
1da177e4 LT |
2167 | void __init cpuset_init_smp(void) |
2168 | { | |
6ad4c188 | 2169 | cpumask_copy(top_cpuset.cpus_allowed, cpu_active_mask); |
38d7bee9 | 2170 | top_cpuset.mems_allowed = node_states[N_MEMORY]; |
33ad801d | 2171 | top_cpuset.old_mems_allowed = top_cpuset.mems_allowed; |
4c4d50f7 | 2172 | |
d8f10cb3 | 2173 | register_hotmemory_notifier(&cpuset_track_online_nodes_nb); |
1da177e4 LT |
2174 | } |
2175 | ||
2176 | /** | |
1da177e4 LT |
2177 | * cpuset_cpus_allowed - return cpus_allowed mask from a tasks cpuset. |
2178 | * @tsk: pointer to task_struct from which to obtain cpuset->cpus_allowed. | |
6af866af | 2179 | * @pmask: pointer to struct cpumask variable to receive cpus_allowed set. |
1da177e4 | 2180 | * |
300ed6cb | 2181 | * Description: Returns the cpumask_var_t cpus_allowed of the cpuset |
1da177e4 | 2182 | * attached to the specified @tsk. Guaranteed to return some non-empty |
5f054e31 | 2183 | * subset of cpu_online_mask, even if this means going outside the |
1da177e4 LT |
2184 | * tasks cpuset. |
2185 | **/ | |
2186 | ||
6af866af | 2187 | void cpuset_cpus_allowed(struct task_struct *tsk, struct cpumask *pmask) |
1da177e4 | 2188 | { |
3d3f26a7 | 2189 | mutex_lock(&callback_mutex); |
909d75a3 | 2190 | task_lock(tsk); |
f9a86fcb | 2191 | guarantee_online_cpus(task_cs(tsk), pmask); |
909d75a3 | 2192 | task_unlock(tsk); |
897f0b3c | 2193 | mutex_unlock(&callback_mutex); |
1da177e4 LT |
2194 | } |
2195 | ||
2baab4e9 | 2196 | void cpuset_cpus_allowed_fallback(struct task_struct *tsk) |
9084bb82 ON |
2197 | { |
2198 | const struct cpuset *cs; | |
9084bb82 ON |
2199 | |
2200 | rcu_read_lock(); | |
2201 | cs = task_cs(tsk); | |
06d6b3cb | 2202 | do_set_cpus_allowed(tsk, cs->cpus_allowed); |
9084bb82 ON |
2203 | rcu_read_unlock(); |
2204 | ||
2205 | /* | |
2206 | * We own tsk->cpus_allowed, nobody can change it under us. | |
2207 | * | |
2208 | * But we used cs && cs->cpus_allowed lockless and thus can | |
2209 | * race with cgroup_attach_task() or update_cpumask() and get | |
2210 | * the wrong tsk->cpus_allowed. However, both cases imply the | |
2211 | * subsequent cpuset_change_cpumask()->set_cpus_allowed_ptr() | |
2212 | * which takes task_rq_lock(). | |
2213 | * | |
2214 | * If we are called after it dropped the lock we must see all | |
2215 | * changes in tsk_cs()->cpus_allowed. Otherwise we can temporary | |
2216 | * set any mask even if it is not right from task_cs() pov, | |
2217 | * the pending set_cpus_allowed_ptr() will fix things. | |
2baab4e9 PZ |
2218 | * |
2219 | * select_fallback_rq() will fix things ups and set cpu_possible_mask | |
2220 | * if required. | |
9084bb82 | 2221 | */ |
9084bb82 ON |
2222 | } |
2223 | ||
1da177e4 LT |
2224 | void cpuset_init_current_mems_allowed(void) |
2225 | { | |
f9a86fcb | 2226 | nodes_setall(current->mems_allowed); |
1da177e4 LT |
2227 | } |
2228 | ||
909d75a3 PJ |
2229 | /** |
2230 | * cpuset_mems_allowed - return mems_allowed mask from a tasks cpuset. | |
2231 | * @tsk: pointer to task_struct from which to obtain cpuset->mems_allowed. | |
2232 | * | |
2233 | * Description: Returns the nodemask_t mems_allowed of the cpuset | |
2234 | * attached to the specified @tsk. Guaranteed to return some non-empty | |
38d7bee9 | 2235 | * subset of node_states[N_MEMORY], even if this means going outside the |
909d75a3 PJ |
2236 | * tasks cpuset. |
2237 | **/ | |
2238 | ||
2239 | nodemask_t cpuset_mems_allowed(struct task_struct *tsk) | |
2240 | { | |
2241 | nodemask_t mask; | |
2242 | ||
3d3f26a7 | 2243 | mutex_lock(&callback_mutex); |
909d75a3 | 2244 | task_lock(tsk); |
8793d854 | 2245 | guarantee_online_mems(task_cs(tsk), &mask); |
909d75a3 | 2246 | task_unlock(tsk); |
3d3f26a7 | 2247 | mutex_unlock(&callback_mutex); |
909d75a3 PJ |
2248 | |
2249 | return mask; | |
2250 | } | |
2251 | ||
d9fd8a6d | 2252 | /** |
19770b32 MG |
2253 | * cpuset_nodemask_valid_mems_allowed - check nodemask vs. curremt mems_allowed |
2254 | * @nodemask: the nodemask to be checked | |
d9fd8a6d | 2255 | * |
19770b32 | 2256 | * Are any of the nodes in the nodemask allowed in current->mems_allowed? |
1da177e4 | 2257 | */ |
19770b32 | 2258 | int cpuset_nodemask_valid_mems_allowed(nodemask_t *nodemask) |
1da177e4 | 2259 | { |
19770b32 | 2260 | return nodes_intersects(*nodemask, current->mems_allowed); |
1da177e4 LT |
2261 | } |
2262 | ||
9bf2229f | 2263 | /* |
78608366 PM |
2264 | * nearest_hardwall_ancestor() - Returns the nearest mem_exclusive or |
2265 | * mem_hardwall ancestor to the specified cpuset. Call holding | |
2266 | * callback_mutex. If no ancestor is mem_exclusive or mem_hardwall | |
2267 | * (an unusual configuration), then returns the root cpuset. | |
9bf2229f | 2268 | */ |
78608366 | 2269 | static const struct cpuset *nearest_hardwall_ancestor(const struct cpuset *cs) |
9bf2229f | 2270 | { |
c431069f TH |
2271 | while (!(is_mem_exclusive(cs) || is_mem_hardwall(cs)) && parent_cs(cs)) |
2272 | cs = parent_cs(cs); | |
9bf2229f PJ |
2273 | return cs; |
2274 | } | |
2275 | ||
d9fd8a6d | 2276 | /** |
a1bc5a4e DR |
2277 | * cpuset_node_allowed_softwall - Can we allocate on a memory node? |
2278 | * @node: is this an allowed node? | |
02a0e53d | 2279 | * @gfp_mask: memory allocation flags |
d9fd8a6d | 2280 | * |
a1bc5a4e DR |
2281 | * If we're in interrupt, yes, we can always allocate. If __GFP_THISNODE is |
2282 | * set, yes, we can always allocate. If node is in our task's mems_allowed, | |
2283 | * yes. If it's not a __GFP_HARDWALL request and this node is in the nearest | |
2284 | * hardwalled cpuset ancestor to this task's cpuset, yes. If the task has been | |
2285 | * OOM killed and has access to memory reserves as specified by the TIF_MEMDIE | |
2286 | * flag, yes. | |
9bf2229f PJ |
2287 | * Otherwise, no. |
2288 | * | |
a1bc5a4e DR |
2289 | * If __GFP_HARDWALL is set, cpuset_node_allowed_softwall() reduces to |
2290 | * cpuset_node_allowed_hardwall(). Otherwise, cpuset_node_allowed_softwall() | |
2291 | * might sleep, and might allow a node from an enclosing cpuset. | |
02a0e53d | 2292 | * |
a1bc5a4e DR |
2293 | * cpuset_node_allowed_hardwall() only handles the simpler case of hardwall |
2294 | * cpusets, and never sleeps. | |
02a0e53d PJ |
2295 | * |
2296 | * The __GFP_THISNODE placement logic is really handled elsewhere, | |
2297 | * by forcibly using a zonelist starting at a specified node, and by | |
2298 | * (in get_page_from_freelist()) refusing to consider the zones for | |
2299 | * any node on the zonelist except the first. By the time any such | |
2300 | * calls get to this routine, we should just shut up and say 'yes'. | |
2301 | * | |
9bf2229f | 2302 | * GFP_USER allocations are marked with the __GFP_HARDWALL bit, |
c596d9f3 DR |
2303 | * and do not allow allocations outside the current tasks cpuset |
2304 | * unless the task has been OOM killed as is marked TIF_MEMDIE. | |
9bf2229f | 2305 | * GFP_KERNEL allocations are not so marked, so can escape to the |
78608366 | 2306 | * nearest enclosing hardwalled ancestor cpuset. |
9bf2229f | 2307 | * |
02a0e53d PJ |
2308 | * Scanning up parent cpusets requires callback_mutex. The |
2309 | * __alloc_pages() routine only calls here with __GFP_HARDWALL bit | |
2310 | * _not_ set if it's a GFP_KERNEL allocation, and all nodes in the | |
2311 | * current tasks mems_allowed came up empty on the first pass over | |
2312 | * the zonelist. So only GFP_KERNEL allocations, if all nodes in the | |
2313 | * cpuset are short of memory, might require taking the callback_mutex | |
2314 | * mutex. | |
9bf2229f | 2315 | * |
36be57ff | 2316 | * The first call here from mm/page_alloc:get_page_from_freelist() |
02a0e53d PJ |
2317 | * has __GFP_HARDWALL set in gfp_mask, enforcing hardwall cpusets, |
2318 | * so no allocation on a node outside the cpuset is allowed (unless | |
2319 | * in interrupt, of course). | |
36be57ff PJ |
2320 | * |
2321 | * The second pass through get_page_from_freelist() doesn't even call | |
2322 | * here for GFP_ATOMIC calls. For those calls, the __alloc_pages() | |
2323 | * variable 'wait' is not set, and the bit ALLOC_CPUSET is not set | |
2324 | * in alloc_flags. That logic and the checks below have the combined | |
2325 | * affect that: | |
9bf2229f PJ |
2326 | * in_interrupt - any node ok (current task context irrelevant) |
2327 | * GFP_ATOMIC - any node ok | |
c596d9f3 | 2328 | * TIF_MEMDIE - any node ok |
78608366 | 2329 | * GFP_KERNEL - any node in enclosing hardwalled cpuset ok |
9bf2229f | 2330 | * GFP_USER - only nodes in current tasks mems allowed ok. |
36be57ff PJ |
2331 | * |
2332 | * Rule: | |
a1bc5a4e | 2333 | * Don't call cpuset_node_allowed_softwall if you can't sleep, unless you |
36be57ff PJ |
2334 | * pass in the __GFP_HARDWALL flag set in gfp_flag, which disables |
2335 | * the code that might scan up ancestor cpusets and sleep. | |
02a0e53d | 2336 | */ |
a1bc5a4e | 2337 | int __cpuset_node_allowed_softwall(int node, gfp_t gfp_mask) |
1da177e4 | 2338 | { |
9bf2229f | 2339 | const struct cpuset *cs; /* current cpuset ancestors */ |
29afd49b | 2340 | int allowed; /* is allocation in zone z allowed? */ |
9bf2229f | 2341 | |
9b819d20 | 2342 | if (in_interrupt() || (gfp_mask & __GFP_THISNODE)) |
9bf2229f | 2343 | return 1; |
92d1dbd2 | 2344 | might_sleep_if(!(gfp_mask & __GFP_HARDWALL)); |
9bf2229f PJ |
2345 | if (node_isset(node, current->mems_allowed)) |
2346 | return 1; | |
c596d9f3 DR |
2347 | /* |
2348 | * Allow tasks that have access to memory reserves because they have | |
2349 | * been OOM killed to get memory anywhere. | |
2350 | */ | |
2351 | if (unlikely(test_thread_flag(TIF_MEMDIE))) | |
2352 | return 1; | |
9bf2229f PJ |
2353 | if (gfp_mask & __GFP_HARDWALL) /* If hardwall request, stop here */ |
2354 | return 0; | |
2355 | ||
5563e770 BP |
2356 | if (current->flags & PF_EXITING) /* Let dying task have memory */ |
2357 | return 1; | |
2358 | ||
9bf2229f | 2359 | /* Not hardwall and node outside mems_allowed: scan up cpusets */ |
3d3f26a7 | 2360 | mutex_lock(&callback_mutex); |
053199ed | 2361 | |
053199ed | 2362 | task_lock(current); |
78608366 | 2363 | cs = nearest_hardwall_ancestor(task_cs(current)); |
053199ed PJ |
2364 | task_unlock(current); |
2365 | ||
9bf2229f | 2366 | allowed = node_isset(node, cs->mems_allowed); |
3d3f26a7 | 2367 | mutex_unlock(&callback_mutex); |
9bf2229f | 2368 | return allowed; |
1da177e4 LT |
2369 | } |
2370 | ||
02a0e53d | 2371 | /* |
a1bc5a4e DR |
2372 | * cpuset_node_allowed_hardwall - Can we allocate on a memory node? |
2373 | * @node: is this an allowed node? | |
02a0e53d PJ |
2374 | * @gfp_mask: memory allocation flags |
2375 | * | |
a1bc5a4e DR |
2376 | * If we're in interrupt, yes, we can always allocate. If __GFP_THISNODE is |
2377 | * set, yes, we can always allocate. If node is in our task's mems_allowed, | |
2378 | * yes. If the task has been OOM killed and has access to memory reserves as | |
2379 | * specified by the TIF_MEMDIE flag, yes. | |
2380 | * Otherwise, no. | |
02a0e53d PJ |
2381 | * |
2382 | * The __GFP_THISNODE placement logic is really handled elsewhere, | |
2383 | * by forcibly using a zonelist starting at a specified node, and by | |
2384 | * (in get_page_from_freelist()) refusing to consider the zones for | |
2385 | * any node on the zonelist except the first. By the time any such | |
2386 | * calls get to this routine, we should just shut up and say 'yes'. | |
2387 | * | |
a1bc5a4e DR |
2388 | * Unlike the cpuset_node_allowed_softwall() variant, above, |
2389 | * this variant requires that the node be in the current task's | |
02a0e53d PJ |
2390 | * mems_allowed or that we're in interrupt. It does not scan up the |
2391 | * cpuset hierarchy for the nearest enclosing mem_exclusive cpuset. | |
2392 | * It never sleeps. | |
2393 | */ | |
a1bc5a4e | 2394 | int __cpuset_node_allowed_hardwall(int node, gfp_t gfp_mask) |
02a0e53d | 2395 | { |
02a0e53d PJ |
2396 | if (in_interrupt() || (gfp_mask & __GFP_THISNODE)) |
2397 | return 1; | |
02a0e53d PJ |
2398 | if (node_isset(node, current->mems_allowed)) |
2399 | return 1; | |
dedf8b79 DW |
2400 | /* |
2401 | * Allow tasks that have access to memory reserves because they have | |
2402 | * been OOM killed to get memory anywhere. | |
2403 | */ | |
2404 | if (unlikely(test_thread_flag(TIF_MEMDIE))) | |
2405 | return 1; | |
02a0e53d PJ |
2406 | return 0; |
2407 | } | |
2408 | ||
825a46af | 2409 | /** |
6adef3eb JS |
2410 | * cpuset_mem_spread_node() - On which node to begin search for a file page |
2411 | * cpuset_slab_spread_node() - On which node to begin search for a slab page | |
825a46af PJ |
2412 | * |
2413 | * If a task is marked PF_SPREAD_PAGE or PF_SPREAD_SLAB (as for | |
2414 | * tasks in a cpuset with is_spread_page or is_spread_slab set), | |
2415 | * and if the memory allocation used cpuset_mem_spread_node() | |
2416 | * to determine on which node to start looking, as it will for | |
2417 | * certain page cache or slab cache pages such as used for file | |
2418 | * system buffers and inode caches, then instead of starting on the | |
2419 | * local node to look for a free page, rather spread the starting | |
2420 | * node around the tasks mems_allowed nodes. | |
2421 | * | |
2422 | * We don't have to worry about the returned node being offline | |
2423 | * because "it can't happen", and even if it did, it would be ok. | |
2424 | * | |
2425 | * The routines calling guarantee_online_mems() are careful to | |
2426 | * only set nodes in task->mems_allowed that are online. So it | |
2427 | * should not be possible for the following code to return an | |
2428 | * offline node. But if it did, that would be ok, as this routine | |
2429 | * is not returning the node where the allocation must be, only | |
2430 | * the node where the search should start. The zonelist passed to | |
2431 | * __alloc_pages() will include all nodes. If the slab allocator | |
2432 | * is passed an offline node, it will fall back to the local node. | |
2433 | * See kmem_cache_alloc_node(). | |
2434 | */ | |
2435 | ||
6adef3eb | 2436 | static int cpuset_spread_node(int *rotor) |
825a46af PJ |
2437 | { |
2438 | int node; | |
2439 | ||
6adef3eb | 2440 | node = next_node(*rotor, current->mems_allowed); |
825a46af PJ |
2441 | if (node == MAX_NUMNODES) |
2442 | node = first_node(current->mems_allowed); | |
6adef3eb | 2443 | *rotor = node; |
825a46af PJ |
2444 | return node; |
2445 | } | |
6adef3eb JS |
2446 | |
2447 | int cpuset_mem_spread_node(void) | |
2448 | { | |
778d3b0f MH |
2449 | if (current->cpuset_mem_spread_rotor == NUMA_NO_NODE) |
2450 | current->cpuset_mem_spread_rotor = | |
2451 | node_random(¤t->mems_allowed); | |
2452 | ||
6adef3eb JS |
2453 | return cpuset_spread_node(¤t->cpuset_mem_spread_rotor); |
2454 | } | |
2455 | ||
2456 | int cpuset_slab_spread_node(void) | |
2457 | { | |
778d3b0f MH |
2458 | if (current->cpuset_slab_spread_rotor == NUMA_NO_NODE) |
2459 | current->cpuset_slab_spread_rotor = | |
2460 | node_random(¤t->mems_allowed); | |
2461 | ||
6adef3eb JS |
2462 | return cpuset_spread_node(¤t->cpuset_slab_spread_rotor); |
2463 | } | |
2464 | ||
825a46af PJ |
2465 | EXPORT_SYMBOL_GPL(cpuset_mem_spread_node); |
2466 | ||
ef08e3b4 | 2467 | /** |
bbe373f2 DR |
2468 | * cpuset_mems_allowed_intersects - Does @tsk1's mems_allowed intersect @tsk2's? |
2469 | * @tsk1: pointer to task_struct of some task. | |
2470 | * @tsk2: pointer to task_struct of some other task. | |
2471 | * | |
2472 | * Description: Return true if @tsk1's mems_allowed intersects the | |
2473 | * mems_allowed of @tsk2. Used by the OOM killer to determine if | |
2474 | * one of the task's memory usage might impact the memory available | |
2475 | * to the other. | |
ef08e3b4 PJ |
2476 | **/ |
2477 | ||
bbe373f2 DR |
2478 | int cpuset_mems_allowed_intersects(const struct task_struct *tsk1, |
2479 | const struct task_struct *tsk2) | |
ef08e3b4 | 2480 | { |
bbe373f2 | 2481 | return nodes_intersects(tsk1->mems_allowed, tsk2->mems_allowed); |
ef08e3b4 PJ |
2482 | } |
2483 | ||
f440d98f LZ |
2484 | #define CPUSET_NODELIST_LEN (256) |
2485 | ||
75aa1994 DR |
2486 | /** |
2487 | * cpuset_print_task_mems_allowed - prints task's cpuset and mems_allowed | |
2488 | * @task: pointer to task_struct of some task. | |
2489 | * | |
2490 | * Description: Prints @task's name, cpuset name, and cached copy of its | |
2491 | * mems_allowed to the kernel log. Must hold task_lock(task) to allow | |
2492 | * dereferencing task_cs(task). | |
2493 | */ | |
2494 | void cpuset_print_task_mems_allowed(struct task_struct *tsk) | |
2495 | { | |
f440d98f LZ |
2496 | /* Statically allocated to prevent using excess stack. */ |
2497 | static char cpuset_nodelist[CPUSET_NODELIST_LEN]; | |
2498 | static DEFINE_SPINLOCK(cpuset_buffer_lock); | |
75aa1994 | 2499 | |
f440d98f | 2500 | struct cgroup *cgrp = task_cs(tsk)->css.cgroup; |
63f43f55 | 2501 | |
cfb5966b | 2502 | rcu_read_lock(); |
f440d98f | 2503 | spin_lock(&cpuset_buffer_lock); |
63f43f55 | 2504 | |
75aa1994 DR |
2505 | nodelist_scnprintf(cpuset_nodelist, CPUSET_NODELIST_LEN, |
2506 | tsk->mems_allowed); | |
2507 | printk(KERN_INFO "%s cpuset=%s mems_allowed=%s\n", | |
f440d98f LZ |
2508 | tsk->comm, cgroup_name(cgrp), cpuset_nodelist); |
2509 | ||
75aa1994 | 2510 | spin_unlock(&cpuset_buffer_lock); |
cfb5966b | 2511 | rcu_read_unlock(); |
75aa1994 DR |
2512 | } |
2513 | ||
3e0d98b9 PJ |
2514 | /* |
2515 | * Collection of memory_pressure is suppressed unless | |
2516 | * this flag is enabled by writing "1" to the special | |
2517 | * cpuset file 'memory_pressure_enabled' in the root cpuset. | |
2518 | */ | |
2519 | ||
c5b2aff8 | 2520 | int cpuset_memory_pressure_enabled __read_mostly; |
3e0d98b9 PJ |
2521 | |
2522 | /** | |
2523 | * cpuset_memory_pressure_bump - keep stats of per-cpuset reclaims. | |
2524 | * | |
2525 | * Keep a running average of the rate of synchronous (direct) | |
2526 | * page reclaim efforts initiated by tasks in each cpuset. | |
2527 | * | |
2528 | * This represents the rate at which some task in the cpuset | |
2529 | * ran low on memory on all nodes it was allowed to use, and | |
2530 | * had to enter the kernels page reclaim code in an effort to | |
2531 | * create more free memory by tossing clean pages or swapping | |
2532 | * or writing dirty pages. | |
2533 | * | |
2534 | * Display to user space in the per-cpuset read-only file | |
2535 | * "memory_pressure". Value displayed is an integer | |
2536 | * representing the recent rate of entry into the synchronous | |
2537 | * (direct) page reclaim by any task attached to the cpuset. | |
2538 | **/ | |
2539 | ||
2540 | void __cpuset_memory_pressure_bump(void) | |
2541 | { | |
3e0d98b9 | 2542 | task_lock(current); |
8793d854 | 2543 | fmeter_markevent(&task_cs(current)->fmeter); |
3e0d98b9 PJ |
2544 | task_unlock(current); |
2545 | } | |
2546 | ||
8793d854 | 2547 | #ifdef CONFIG_PROC_PID_CPUSET |
1da177e4 LT |
2548 | /* |
2549 | * proc_cpuset_show() | |
2550 | * - Print tasks cpuset path into seq_file. | |
2551 | * - Used for /proc/<pid>/cpuset. | |
053199ed PJ |
2552 | * - No need to task_lock(tsk) on this tsk->cpuset reference, as it |
2553 | * doesn't really matter if tsk->cpuset changes after we read it, | |
5d21cc2d | 2554 | * and we take cpuset_mutex, keeping cpuset_attach() from changing it |
2df167a3 | 2555 | * anyway. |
1da177e4 | 2556 | */ |
8d8b97ba | 2557 | int proc_cpuset_show(struct seq_file *m, void *unused_v) |
1da177e4 | 2558 | { |
13b41b09 | 2559 | struct pid *pid; |
1da177e4 LT |
2560 | struct task_struct *tsk; |
2561 | char *buf; | |
8793d854 | 2562 | struct cgroup_subsys_state *css; |
99f89551 | 2563 | int retval; |
1da177e4 | 2564 | |
99f89551 | 2565 | retval = -ENOMEM; |
1da177e4 LT |
2566 | buf = kmalloc(PAGE_SIZE, GFP_KERNEL); |
2567 | if (!buf) | |
99f89551 EB |
2568 | goto out; |
2569 | ||
2570 | retval = -ESRCH; | |
13b41b09 EB |
2571 | pid = m->private; |
2572 | tsk = get_pid_task(pid, PIDTYPE_PID); | |
99f89551 EB |
2573 | if (!tsk) |
2574 | goto out_free; | |
1da177e4 | 2575 | |
27e89ae5 | 2576 | rcu_read_lock(); |
8793d854 PM |
2577 | css = task_subsys_state(tsk, cpuset_subsys_id); |
2578 | retval = cgroup_path(css->cgroup, buf, PAGE_SIZE); | |
27e89ae5 | 2579 | rcu_read_unlock(); |
1da177e4 | 2580 | if (retval < 0) |
27e89ae5 | 2581 | goto out_put_task; |
1da177e4 LT |
2582 | seq_puts(m, buf); |
2583 | seq_putc(m, '\n'); | |
27e89ae5 | 2584 | out_put_task: |
99f89551 EB |
2585 | put_task_struct(tsk); |
2586 | out_free: | |
1da177e4 | 2587 | kfree(buf); |
99f89551 | 2588 | out: |
1da177e4 LT |
2589 | return retval; |
2590 | } | |
8793d854 | 2591 | #endif /* CONFIG_PROC_PID_CPUSET */ |
1da177e4 | 2592 | |
d01d4827 | 2593 | /* Display task mems_allowed in /proc/<pid>/status file. */ |
df5f8314 EB |
2594 | void cpuset_task_status_allowed(struct seq_file *m, struct task_struct *task) |
2595 | { | |
df5f8314 | 2596 | seq_printf(m, "Mems_allowed:\t"); |
30e8e136 | 2597 | seq_nodemask(m, &task->mems_allowed); |
df5f8314 | 2598 | seq_printf(m, "\n"); |
39106dcf | 2599 | seq_printf(m, "Mems_allowed_list:\t"); |
30e8e136 | 2600 | seq_nodemask_list(m, &task->mems_allowed); |
39106dcf | 2601 | seq_printf(m, "\n"); |
1da177e4 | 2602 | } |