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 |
1da177e4 LT |
17 | * |
18 | * This file is subject to the terms and conditions of the GNU General Public | |
19 | * License. See the file COPYING in the main directory of the Linux | |
20 | * distribution for more details. | |
21 | */ | |
22 | ||
1da177e4 LT |
23 | #include <linux/cpu.h> |
24 | #include <linux/cpumask.h> | |
25 | #include <linux/cpuset.h> | |
26 | #include <linux/err.h> | |
27 | #include <linux/errno.h> | |
28 | #include <linux/file.h> | |
29 | #include <linux/fs.h> | |
30 | #include <linux/init.h> | |
31 | #include <linux/interrupt.h> | |
32 | #include <linux/kernel.h> | |
33 | #include <linux/kmod.h> | |
34 | #include <linux/list.h> | |
68860ec1 | 35 | #include <linux/mempolicy.h> |
1da177e4 LT |
36 | #include <linux/mm.h> |
37 | #include <linux/module.h> | |
38 | #include <linux/mount.h> | |
39 | #include <linux/namei.h> | |
40 | #include <linux/pagemap.h> | |
41 | #include <linux/proc_fs.h> | |
6b9c2603 | 42 | #include <linux/rcupdate.h> |
1da177e4 LT |
43 | #include <linux/sched.h> |
44 | #include <linux/seq_file.h> | |
22fb52dd | 45 | #include <linux/security.h> |
1da177e4 | 46 | #include <linux/slab.h> |
1da177e4 LT |
47 | #include <linux/spinlock.h> |
48 | #include <linux/stat.h> | |
49 | #include <linux/string.h> | |
50 | #include <linux/time.h> | |
51 | #include <linux/backing-dev.h> | |
52 | #include <linux/sort.h> | |
53 | ||
54 | #include <asm/uaccess.h> | |
55 | #include <asm/atomic.h> | |
3d3f26a7 | 56 | #include <linux/mutex.h> |
956db3ca CW |
57 | #include <linux/workqueue.h> |
58 | #include <linux/cgroup.h> | |
1da177e4 | 59 | |
202f72d5 PJ |
60 | /* |
61 | * Tracks how many cpusets are currently defined in system. | |
62 | * When there is only one cpuset (the root cpuset) we can | |
63 | * short circuit some hooks. | |
64 | */ | |
7edc5962 | 65 | int number_of_cpusets __read_mostly; |
202f72d5 | 66 | |
2df167a3 | 67 | /* Forward declare cgroup structures */ |
8793d854 PM |
68 | struct cgroup_subsys cpuset_subsys; |
69 | struct cpuset; | |
70 | ||
3e0d98b9 PJ |
71 | /* See "Frequency meter" comments, below. */ |
72 | ||
73 | struct fmeter { | |
74 | int cnt; /* unprocessed events count */ | |
75 | int val; /* most recent output value */ | |
76 | time_t time; /* clock (secs) when val computed */ | |
77 | spinlock_t lock; /* guards read or write of above */ | |
78 | }; | |
79 | ||
1da177e4 | 80 | struct cpuset { |
8793d854 PM |
81 | struct cgroup_subsys_state css; |
82 | ||
1da177e4 LT |
83 | unsigned long flags; /* "unsigned long" so bitops work */ |
84 | cpumask_t cpus_allowed; /* CPUs allowed to tasks in cpuset */ | |
85 | nodemask_t mems_allowed; /* Memory Nodes allowed to tasks */ | |
86 | ||
1da177e4 | 87 | struct cpuset *parent; /* my parent */ |
1da177e4 LT |
88 | |
89 | /* | |
90 | * Copy of global cpuset_mems_generation as of the most | |
91 | * recent time this cpuset changed its mems_allowed. | |
92 | */ | |
3e0d98b9 PJ |
93 | int mems_generation; |
94 | ||
95 | struct fmeter fmeter; /* memory_pressure filter */ | |
029190c5 PJ |
96 | |
97 | /* partition number for rebuild_sched_domains() */ | |
98 | int pn; | |
956db3ca | 99 | |
1d3504fc HS |
100 | /* for custom sched domain */ |
101 | int relax_domain_level; | |
102 | ||
956db3ca CW |
103 | /* used for walking a cpuset heirarchy */ |
104 | struct list_head stack_list; | |
1da177e4 LT |
105 | }; |
106 | ||
8793d854 PM |
107 | /* Retrieve the cpuset for a cgroup */ |
108 | static inline struct cpuset *cgroup_cs(struct cgroup *cont) | |
109 | { | |
110 | return container_of(cgroup_subsys_state(cont, cpuset_subsys_id), | |
111 | struct cpuset, css); | |
112 | } | |
113 | ||
114 | /* Retrieve the cpuset for a task */ | |
115 | static inline struct cpuset *task_cs(struct task_struct *task) | |
116 | { | |
117 | return container_of(task_subsys_state(task, cpuset_subsys_id), | |
118 | struct cpuset, css); | |
119 | } | |
956db3ca CW |
120 | struct cpuset_hotplug_scanner { |
121 | struct cgroup_scanner scan; | |
122 | struct cgroup *to; | |
123 | }; | |
8793d854 | 124 | |
1da177e4 LT |
125 | /* bits in struct cpuset flags field */ |
126 | typedef enum { | |
127 | CS_CPU_EXCLUSIVE, | |
128 | CS_MEM_EXCLUSIVE, | |
78608366 | 129 | CS_MEM_HARDWALL, |
45b07ef3 | 130 | CS_MEMORY_MIGRATE, |
029190c5 | 131 | CS_SCHED_LOAD_BALANCE, |
825a46af PJ |
132 | CS_SPREAD_PAGE, |
133 | CS_SPREAD_SLAB, | |
1da177e4 LT |
134 | } cpuset_flagbits_t; |
135 | ||
136 | /* convenient tests for these bits */ | |
137 | static inline int is_cpu_exclusive(const struct cpuset *cs) | |
138 | { | |
7b5b9ef0 | 139 | return test_bit(CS_CPU_EXCLUSIVE, &cs->flags); |
1da177e4 LT |
140 | } |
141 | ||
142 | static inline int is_mem_exclusive(const struct cpuset *cs) | |
143 | { | |
7b5b9ef0 | 144 | return test_bit(CS_MEM_EXCLUSIVE, &cs->flags); |
1da177e4 LT |
145 | } |
146 | ||
78608366 PM |
147 | static inline int is_mem_hardwall(const struct cpuset *cs) |
148 | { | |
149 | return test_bit(CS_MEM_HARDWALL, &cs->flags); | |
150 | } | |
151 | ||
029190c5 PJ |
152 | static inline int is_sched_load_balance(const struct cpuset *cs) |
153 | { | |
154 | return test_bit(CS_SCHED_LOAD_BALANCE, &cs->flags); | |
155 | } | |
156 | ||
45b07ef3 PJ |
157 | static inline int is_memory_migrate(const struct cpuset *cs) |
158 | { | |
7b5b9ef0 | 159 | return test_bit(CS_MEMORY_MIGRATE, &cs->flags); |
45b07ef3 PJ |
160 | } |
161 | ||
825a46af PJ |
162 | static inline int is_spread_page(const struct cpuset *cs) |
163 | { | |
164 | return test_bit(CS_SPREAD_PAGE, &cs->flags); | |
165 | } | |
166 | ||
167 | static inline int is_spread_slab(const struct cpuset *cs) | |
168 | { | |
169 | return test_bit(CS_SPREAD_SLAB, &cs->flags); | |
170 | } | |
171 | ||
1da177e4 | 172 | /* |
151a4420 | 173 | * Increment this integer everytime any cpuset changes its |
1da177e4 LT |
174 | * mems_allowed value. Users of cpusets can track this generation |
175 | * number, and avoid having to lock and reload mems_allowed unless | |
176 | * the cpuset they're using changes generation. | |
177 | * | |
2df167a3 | 178 | * A single, global generation is needed because cpuset_attach_task() could |
1da177e4 LT |
179 | * reattach a task to a different cpuset, which must not have its |
180 | * generation numbers aliased with those of that tasks previous cpuset. | |
181 | * | |
182 | * Generations are needed for mems_allowed because one task cannot | |
2df167a3 | 183 | * modify another's memory placement. So we must enable every task, |
1da177e4 LT |
184 | * on every visit to __alloc_pages(), to efficiently check whether |
185 | * its current->cpuset->mems_allowed has changed, requiring an update | |
186 | * of its current->mems_allowed. | |
151a4420 | 187 | * |
2df167a3 | 188 | * Since writes to cpuset_mems_generation are guarded by the cgroup lock |
151a4420 | 189 | * there is no need to mark it atomic. |
1da177e4 | 190 | */ |
151a4420 | 191 | static int cpuset_mems_generation; |
1da177e4 LT |
192 | |
193 | static struct cpuset top_cpuset = { | |
194 | .flags = ((1 << CS_CPU_EXCLUSIVE) | (1 << CS_MEM_EXCLUSIVE)), | |
195 | .cpus_allowed = CPU_MASK_ALL, | |
196 | .mems_allowed = NODE_MASK_ALL, | |
1da177e4 LT |
197 | }; |
198 | ||
1da177e4 | 199 | /* |
2df167a3 PM |
200 | * There are two global mutexes guarding cpuset structures. The first |
201 | * is the main control groups cgroup_mutex, accessed via | |
202 | * cgroup_lock()/cgroup_unlock(). The second is the cpuset-specific | |
203 | * callback_mutex, below. They can nest. It is ok to first take | |
204 | * cgroup_mutex, then nest callback_mutex. We also require taking | |
205 | * task_lock() when dereferencing a task's cpuset pointer. See "The | |
206 | * task_lock() exception", at the end of this comment. | |
053199ed | 207 | * |
3d3f26a7 | 208 | * A task must hold both mutexes to modify cpusets. If a task |
2df167a3 | 209 | * holds cgroup_mutex, then it blocks others wanting that mutex, |
3d3f26a7 | 210 | * ensuring that it is the only task able to also acquire callback_mutex |
053199ed PJ |
211 | * and be able to modify cpusets. It can perform various checks on |
212 | * the cpuset structure first, knowing nothing will change. It can | |
2df167a3 | 213 | * also allocate memory while just holding cgroup_mutex. While it is |
053199ed | 214 | * performing these checks, various callback routines can briefly |
3d3f26a7 IM |
215 | * acquire callback_mutex to query cpusets. Once it is ready to make |
216 | * the changes, it takes callback_mutex, blocking everyone else. | |
053199ed PJ |
217 | * |
218 | * Calls to the kernel memory allocator can not be made while holding | |
3d3f26a7 | 219 | * callback_mutex, as that would risk double tripping on callback_mutex |
053199ed PJ |
220 | * from one of the callbacks into the cpuset code from within |
221 | * __alloc_pages(). | |
222 | * | |
3d3f26a7 | 223 | * If a task is only holding callback_mutex, then it has read-only |
053199ed PJ |
224 | * access to cpusets. |
225 | * | |
226 | * The task_struct fields mems_allowed and mems_generation may only | |
227 | * be accessed in the context of that task, so require no locks. | |
228 | * | |
3d3f26a7 | 229 | * The cpuset_common_file_read() handlers only hold callback_mutex across |
053199ed PJ |
230 | * small pieces of code, such as when reading out possibly multi-word |
231 | * cpumasks and nodemasks. | |
232 | * | |
2df167a3 PM |
233 | * Accessing a task's cpuset should be done in accordance with the |
234 | * guidelines for accessing subsystem state in kernel/cgroup.c | |
1da177e4 LT |
235 | */ |
236 | ||
3d3f26a7 | 237 | static DEFINE_MUTEX(callback_mutex); |
4247bdc6 | 238 | |
8793d854 PM |
239 | /* This is ugly, but preserves the userspace API for existing cpuset |
240 | * users. If someone tries to mount the "cpuset" filesystem, we | |
241 | * silently switch it to mount "cgroup" instead */ | |
454e2398 DH |
242 | static int cpuset_get_sb(struct file_system_type *fs_type, |
243 | int flags, const char *unused_dev_name, | |
244 | void *data, struct vfsmount *mnt) | |
1da177e4 | 245 | { |
8793d854 PM |
246 | struct file_system_type *cgroup_fs = get_fs_type("cgroup"); |
247 | int ret = -ENODEV; | |
248 | if (cgroup_fs) { | |
249 | char mountopts[] = | |
250 | "cpuset,noprefix," | |
251 | "release_agent=/sbin/cpuset_release_agent"; | |
252 | ret = cgroup_fs->get_sb(cgroup_fs, flags, | |
253 | unused_dev_name, mountopts, mnt); | |
254 | put_filesystem(cgroup_fs); | |
255 | } | |
256 | return ret; | |
1da177e4 LT |
257 | } |
258 | ||
259 | static struct file_system_type cpuset_fs_type = { | |
260 | .name = "cpuset", | |
261 | .get_sb = cpuset_get_sb, | |
1da177e4 LT |
262 | }; |
263 | ||
1da177e4 LT |
264 | /* |
265 | * Return in *pmask the portion of a cpusets's cpus_allowed that | |
266 | * are online. If none are online, walk up the cpuset hierarchy | |
267 | * until we find one that does have some online cpus. If we get | |
268 | * all the way to the top and still haven't found any online cpus, | |
269 | * return cpu_online_map. Or if passed a NULL cs from an exit'ing | |
270 | * task, return cpu_online_map. | |
271 | * | |
272 | * One way or another, we guarantee to return some non-empty subset | |
273 | * of cpu_online_map. | |
274 | * | |
3d3f26a7 | 275 | * Call with callback_mutex held. |
1da177e4 LT |
276 | */ |
277 | ||
278 | static void guarantee_online_cpus(const struct cpuset *cs, cpumask_t *pmask) | |
279 | { | |
280 | while (cs && !cpus_intersects(cs->cpus_allowed, cpu_online_map)) | |
281 | cs = cs->parent; | |
282 | if (cs) | |
283 | cpus_and(*pmask, cs->cpus_allowed, cpu_online_map); | |
284 | else | |
285 | *pmask = cpu_online_map; | |
286 | BUG_ON(!cpus_intersects(*pmask, cpu_online_map)); | |
287 | } | |
288 | ||
289 | /* | |
290 | * Return in *pmask the portion of a cpusets's mems_allowed that | |
0e1e7c7a CL |
291 | * are online, with memory. If none are online with memory, walk |
292 | * up the cpuset hierarchy until we find one that does have some | |
293 | * online mems. If we get all the way to the top and still haven't | |
294 | * found any online mems, return node_states[N_HIGH_MEMORY]. | |
1da177e4 LT |
295 | * |
296 | * One way or another, we guarantee to return some non-empty subset | |
0e1e7c7a | 297 | * of node_states[N_HIGH_MEMORY]. |
1da177e4 | 298 | * |
3d3f26a7 | 299 | * Call with callback_mutex held. |
1da177e4 LT |
300 | */ |
301 | ||
302 | static void guarantee_online_mems(const struct cpuset *cs, nodemask_t *pmask) | |
303 | { | |
0e1e7c7a CL |
304 | while (cs && !nodes_intersects(cs->mems_allowed, |
305 | node_states[N_HIGH_MEMORY])) | |
1da177e4 LT |
306 | cs = cs->parent; |
307 | if (cs) | |
0e1e7c7a CL |
308 | nodes_and(*pmask, cs->mems_allowed, |
309 | node_states[N_HIGH_MEMORY]); | |
1da177e4 | 310 | else |
0e1e7c7a CL |
311 | *pmask = node_states[N_HIGH_MEMORY]; |
312 | BUG_ON(!nodes_intersects(*pmask, node_states[N_HIGH_MEMORY])); | |
1da177e4 LT |
313 | } |
314 | ||
cf2a473c PJ |
315 | /** |
316 | * cpuset_update_task_memory_state - update task memory placement | |
317 | * | |
318 | * If the current tasks cpusets mems_allowed changed behind our | |
319 | * backs, update current->mems_allowed, mems_generation and task NUMA | |
320 | * mempolicy to the new value. | |
053199ed | 321 | * |
cf2a473c PJ |
322 | * Task mempolicy is updated by rebinding it relative to the |
323 | * current->cpuset if a task has its memory placement changed. | |
324 | * Do not call this routine if in_interrupt(). | |
325 | * | |
4a01c8d5 | 326 | * Call without callback_mutex or task_lock() held. May be |
2df167a3 PM |
327 | * called with or without cgroup_mutex held. Thanks in part to |
328 | * 'the_top_cpuset_hack', the task's cpuset pointer will never | |
41f7f60d DR |
329 | * be NULL. This routine also might acquire callback_mutex during |
330 | * call. | |
053199ed | 331 | * |
6b9c2603 PJ |
332 | * Reading current->cpuset->mems_generation doesn't need task_lock |
333 | * to guard the current->cpuset derefence, because it is guarded | |
2df167a3 | 334 | * from concurrent freeing of current->cpuset using RCU. |
6b9c2603 PJ |
335 | * |
336 | * The rcu_dereference() is technically probably not needed, | |
337 | * as I don't actually mind if I see a new cpuset pointer but | |
338 | * an old value of mems_generation. However this really only | |
339 | * matters on alpha systems using cpusets heavily. If I dropped | |
340 | * that rcu_dereference(), it would save them a memory barrier. | |
341 | * For all other arch's, rcu_dereference is a no-op anyway, and for | |
342 | * alpha systems not using cpusets, another planned optimization, | |
343 | * avoiding the rcu critical section for tasks in the root cpuset | |
344 | * which is statically allocated, so can't vanish, will make this | |
345 | * irrelevant. Better to use RCU as intended, than to engage in | |
346 | * some cute trick to save a memory barrier that is impossible to | |
347 | * test, for alpha systems using cpusets heavily, which might not | |
348 | * even exist. | |
053199ed PJ |
349 | * |
350 | * This routine is needed to update the per-task mems_allowed data, | |
351 | * within the tasks context, when it is trying to allocate memory | |
352 | * (in various mm/mempolicy.c routines) and notices that some other | |
353 | * task has been modifying its cpuset. | |
1da177e4 LT |
354 | */ |
355 | ||
fe85a998 | 356 | void cpuset_update_task_memory_state(void) |
1da177e4 | 357 | { |
053199ed | 358 | int my_cpusets_mem_gen; |
cf2a473c | 359 | struct task_struct *tsk = current; |
6b9c2603 | 360 | struct cpuset *cs; |
053199ed | 361 | |
8793d854 | 362 | if (task_cs(tsk) == &top_cpuset) { |
03a285f5 PJ |
363 | /* Don't need rcu for top_cpuset. It's never freed. */ |
364 | my_cpusets_mem_gen = top_cpuset.mems_generation; | |
365 | } else { | |
366 | rcu_read_lock(); | |
da5ef6bb | 367 | my_cpusets_mem_gen = task_cs(tsk)->mems_generation; |
03a285f5 PJ |
368 | rcu_read_unlock(); |
369 | } | |
1da177e4 | 370 | |
cf2a473c | 371 | if (my_cpusets_mem_gen != tsk->cpuset_mems_generation) { |
3d3f26a7 | 372 | mutex_lock(&callback_mutex); |
cf2a473c | 373 | task_lock(tsk); |
8793d854 | 374 | cs = task_cs(tsk); /* Maybe changed when task not locked */ |
cf2a473c PJ |
375 | guarantee_online_mems(cs, &tsk->mems_allowed); |
376 | tsk->cpuset_mems_generation = cs->mems_generation; | |
825a46af PJ |
377 | if (is_spread_page(cs)) |
378 | tsk->flags |= PF_SPREAD_PAGE; | |
379 | else | |
380 | tsk->flags &= ~PF_SPREAD_PAGE; | |
381 | if (is_spread_slab(cs)) | |
382 | tsk->flags |= PF_SPREAD_SLAB; | |
383 | else | |
384 | tsk->flags &= ~PF_SPREAD_SLAB; | |
cf2a473c | 385 | task_unlock(tsk); |
3d3f26a7 | 386 | mutex_unlock(&callback_mutex); |
74cb2155 | 387 | mpol_rebind_task(tsk, &tsk->mems_allowed); |
1da177e4 LT |
388 | } |
389 | } | |
390 | ||
391 | /* | |
392 | * is_cpuset_subset(p, q) - Is cpuset p a subset of cpuset q? | |
393 | * | |
394 | * One cpuset is a subset of another if all its allowed CPUs and | |
395 | * Memory Nodes are a subset of the other, and its exclusive flags | |
2df167a3 | 396 | * are only set if the other's are set. Call holding cgroup_mutex. |
1da177e4 LT |
397 | */ |
398 | ||
399 | static int is_cpuset_subset(const struct cpuset *p, const struct cpuset *q) | |
400 | { | |
401 | return cpus_subset(p->cpus_allowed, q->cpus_allowed) && | |
402 | nodes_subset(p->mems_allowed, q->mems_allowed) && | |
403 | is_cpu_exclusive(p) <= is_cpu_exclusive(q) && | |
404 | is_mem_exclusive(p) <= is_mem_exclusive(q); | |
405 | } | |
406 | ||
407 | /* | |
408 | * validate_change() - Used to validate that any proposed cpuset change | |
409 | * follows the structural rules for cpusets. | |
410 | * | |
411 | * If we replaced the flag and mask values of the current cpuset | |
412 | * (cur) with those values in the trial cpuset (trial), would | |
413 | * our various subset and exclusive rules still be valid? Presumes | |
2df167a3 | 414 | * cgroup_mutex held. |
1da177e4 LT |
415 | * |
416 | * 'cur' is the address of an actual, in-use cpuset. Operations | |
417 | * such as list traversal that depend on the actual address of the | |
418 | * cpuset in the list must use cur below, not trial. | |
419 | * | |
420 | * 'trial' is the address of bulk structure copy of cur, with | |
421 | * perhaps one or more of the fields cpus_allowed, mems_allowed, | |
422 | * or flags changed to new, trial values. | |
423 | * | |
424 | * Return 0 if valid, -errno if not. | |
425 | */ | |
426 | ||
427 | static int validate_change(const struct cpuset *cur, const struct cpuset *trial) | |
428 | { | |
8793d854 | 429 | struct cgroup *cont; |
1da177e4 LT |
430 | struct cpuset *c, *par; |
431 | ||
432 | /* Each of our child cpusets must be a subset of us */ | |
8793d854 PM |
433 | list_for_each_entry(cont, &cur->css.cgroup->children, sibling) { |
434 | if (!is_cpuset_subset(cgroup_cs(cont), trial)) | |
1da177e4 LT |
435 | return -EBUSY; |
436 | } | |
437 | ||
438 | /* Remaining checks don't apply to root cpuset */ | |
69604067 | 439 | if (cur == &top_cpuset) |
1da177e4 LT |
440 | return 0; |
441 | ||
69604067 PJ |
442 | par = cur->parent; |
443 | ||
1da177e4 LT |
444 | /* We must be a subset of our parent cpuset */ |
445 | if (!is_cpuset_subset(trial, par)) | |
446 | return -EACCES; | |
447 | ||
2df167a3 PM |
448 | /* |
449 | * If either I or some sibling (!= me) is exclusive, we can't | |
450 | * overlap | |
451 | */ | |
8793d854 PM |
452 | list_for_each_entry(cont, &par->css.cgroup->children, sibling) { |
453 | c = cgroup_cs(cont); | |
1da177e4 LT |
454 | if ((is_cpu_exclusive(trial) || is_cpu_exclusive(c)) && |
455 | c != cur && | |
456 | cpus_intersects(trial->cpus_allowed, c->cpus_allowed)) | |
457 | return -EINVAL; | |
458 | if ((is_mem_exclusive(trial) || is_mem_exclusive(c)) && | |
459 | c != cur && | |
460 | nodes_intersects(trial->mems_allowed, c->mems_allowed)) | |
461 | return -EINVAL; | |
462 | } | |
463 | ||
020958b6 PJ |
464 | /* Cpusets with tasks can't have empty cpus_allowed or mems_allowed */ |
465 | if (cgroup_task_count(cur->css.cgroup)) { | |
466 | if (cpus_empty(trial->cpus_allowed) || | |
467 | nodes_empty(trial->mems_allowed)) { | |
468 | return -ENOSPC; | |
469 | } | |
470 | } | |
471 | ||
1da177e4 LT |
472 | return 0; |
473 | } | |
474 | ||
029190c5 PJ |
475 | /* |
476 | * Helper routine for rebuild_sched_domains(). | |
477 | * Do cpusets a, b have overlapping cpus_allowed masks? | |
478 | */ | |
479 | ||
480 | static int cpusets_overlap(struct cpuset *a, struct cpuset *b) | |
481 | { | |
482 | return cpus_intersects(a->cpus_allowed, b->cpus_allowed); | |
483 | } | |
484 | ||
1d3504fc HS |
485 | static void |
486 | update_domain_attr(struct sched_domain_attr *dattr, struct cpuset *c) | |
487 | { | |
1d3504fc HS |
488 | if (dattr->relax_domain_level < c->relax_domain_level) |
489 | dattr->relax_domain_level = c->relax_domain_level; | |
490 | return; | |
491 | } | |
492 | ||
f5393693 LJ |
493 | static void |
494 | update_domain_attr_tree(struct sched_domain_attr *dattr, struct cpuset *c) | |
495 | { | |
496 | LIST_HEAD(q); | |
497 | ||
498 | list_add(&c->stack_list, &q); | |
499 | while (!list_empty(&q)) { | |
500 | struct cpuset *cp; | |
501 | struct cgroup *cont; | |
502 | struct cpuset *child; | |
503 | ||
504 | cp = list_first_entry(&q, struct cpuset, stack_list); | |
505 | list_del(q.next); | |
506 | ||
507 | if (cpus_empty(cp->cpus_allowed)) | |
508 | continue; | |
509 | ||
510 | if (is_sched_load_balance(cp)) | |
511 | update_domain_attr(dattr, cp); | |
512 | ||
513 | list_for_each_entry(cont, &cp->css.cgroup->children, sibling) { | |
514 | child = cgroup_cs(cont); | |
515 | list_add_tail(&child->stack_list, &q); | |
516 | } | |
517 | } | |
518 | } | |
519 | ||
029190c5 PJ |
520 | /* |
521 | * rebuild_sched_domains() | |
522 | * | |
c372e817 LZ |
523 | * This routine will be called to rebuild the scheduler's dynamic |
524 | * sched domains: | |
525 | * - if the flag 'sched_load_balance' of any cpuset with non-empty | |
526 | * 'cpus' changes, | |
527 | * - or if the 'cpus' allowed changes in any cpuset which has that | |
528 | * flag enabled, | |
529 | * - or if the 'sched_relax_domain_level' of any cpuset which has | |
530 | * that flag enabled and with non-empty 'cpus' changes, | |
531 | * - or if any cpuset with non-empty 'cpus' is removed, | |
532 | * - or if a cpu gets offlined. | |
029190c5 PJ |
533 | * |
534 | * This routine builds a partial partition of the systems CPUs | |
535 | * (the set of non-overlappping cpumask_t's in the array 'part' | |
536 | * below), and passes that partial partition to the kernel/sched.c | |
537 | * partition_sched_domains() routine, which will rebuild the | |
538 | * schedulers load balancing domains (sched domains) as specified | |
539 | * by that partial partition. A 'partial partition' is a set of | |
540 | * non-overlapping subsets whose union is a subset of that set. | |
541 | * | |
542 | * See "What is sched_load_balance" in Documentation/cpusets.txt | |
543 | * for a background explanation of this. | |
544 | * | |
545 | * Does not return errors, on the theory that the callers of this | |
546 | * routine would rather not worry about failures to rebuild sched | |
547 | * domains when operating in the severe memory shortage situations | |
548 | * that could cause allocation failures below. | |
549 | * | |
550 | * Call with cgroup_mutex held. May take callback_mutex during | |
551 | * call due to the kfifo_alloc() and kmalloc() calls. May nest | |
86ef5c9a | 552 | * a call to the get_online_cpus()/put_online_cpus() pair. |
029190c5 | 553 | * Must not be called holding callback_mutex, because we must not |
86ef5c9a GS |
554 | * call get_online_cpus() while holding callback_mutex. Elsewhere |
555 | * the kernel nests callback_mutex inside get_online_cpus() calls. | |
029190c5 PJ |
556 | * So the reverse nesting would risk an ABBA deadlock. |
557 | * | |
558 | * The three key local variables below are: | |
aeed6824 | 559 | * q - a linked-list queue of cpuset pointers, used to implement a |
029190c5 PJ |
560 | * top-down scan of all cpusets. This scan loads a pointer |
561 | * to each cpuset marked is_sched_load_balance into the | |
562 | * array 'csa'. For our purposes, rebuilding the schedulers | |
563 | * sched domains, we can ignore !is_sched_load_balance cpusets. | |
564 | * csa - (for CpuSet Array) Array of pointers to all the cpusets | |
565 | * that need to be load balanced, for convenient iterative | |
566 | * access by the subsequent code that finds the best partition, | |
567 | * i.e the set of domains (subsets) of CPUs such that the | |
568 | * cpus_allowed of every cpuset marked is_sched_load_balance | |
569 | * is a subset of one of these domains, while there are as | |
570 | * many such domains as possible, each as small as possible. | |
571 | * doms - Conversion of 'csa' to an array of cpumasks, for passing to | |
572 | * the kernel/sched.c routine partition_sched_domains() in a | |
573 | * convenient format, that can be easily compared to the prior | |
574 | * value to determine what partition elements (sched domains) | |
575 | * were changed (added or removed.) | |
576 | * | |
577 | * Finding the best partition (set of domains): | |
578 | * The triple nested loops below over i, j, k scan over the | |
579 | * load balanced cpusets (using the array of cpuset pointers in | |
580 | * csa[]) looking for pairs of cpusets that have overlapping | |
581 | * cpus_allowed, but which don't have the same 'pn' partition | |
582 | * number and gives them in the same partition number. It keeps | |
583 | * looping on the 'restart' label until it can no longer find | |
584 | * any such pairs. | |
585 | * | |
586 | * The union of the cpus_allowed masks from the set of | |
587 | * all cpusets having the same 'pn' value then form the one | |
588 | * element of the partition (one sched domain) to be passed to | |
589 | * partition_sched_domains(). | |
590 | */ | |
591 | ||
e761b772 | 592 | void rebuild_sched_domains(void) |
029190c5 | 593 | { |
aeed6824 | 594 | LIST_HEAD(q); /* queue of cpusets to be scanned*/ |
029190c5 PJ |
595 | struct cpuset *cp; /* scans q */ |
596 | struct cpuset **csa; /* array of all cpuset ptrs */ | |
597 | int csn; /* how many cpuset ptrs in csa so far */ | |
598 | int i, j, k; /* indices for partition finding loops */ | |
599 | cpumask_t *doms; /* resulting partition; i.e. sched domains */ | |
1d3504fc | 600 | struct sched_domain_attr *dattr; /* attributes for custom domains */ |
029190c5 PJ |
601 | int ndoms; /* number of sched domains in result */ |
602 | int nslot; /* next empty doms[] cpumask_t slot */ | |
603 | ||
029190c5 PJ |
604 | csa = NULL; |
605 | doms = NULL; | |
1d3504fc | 606 | dattr = NULL; |
029190c5 PJ |
607 | |
608 | /* Special case for the 99% of systems with one, full, sched domain */ | |
609 | if (is_sched_load_balance(&top_cpuset)) { | |
610 | ndoms = 1; | |
611 | doms = kmalloc(sizeof(cpumask_t), GFP_KERNEL); | |
612 | if (!doms) | |
613 | goto rebuild; | |
1d3504fc HS |
614 | dattr = kmalloc(sizeof(struct sched_domain_attr), GFP_KERNEL); |
615 | if (dattr) { | |
616 | *dattr = SD_ATTR_INIT; | |
93a65575 | 617 | update_domain_attr_tree(dattr, &top_cpuset); |
1d3504fc | 618 | } |
029190c5 PJ |
619 | *doms = top_cpuset.cpus_allowed; |
620 | goto rebuild; | |
621 | } | |
622 | ||
029190c5 PJ |
623 | csa = kmalloc(number_of_cpusets * sizeof(cp), GFP_KERNEL); |
624 | if (!csa) | |
625 | goto done; | |
626 | csn = 0; | |
627 | ||
aeed6824 LZ |
628 | list_add(&top_cpuset.stack_list, &q); |
629 | while (!list_empty(&q)) { | |
029190c5 PJ |
630 | struct cgroup *cont; |
631 | struct cpuset *child; /* scans child cpusets of cp */ | |
489a5393 | 632 | |
aeed6824 LZ |
633 | cp = list_first_entry(&q, struct cpuset, stack_list); |
634 | list_del(q.next); | |
635 | ||
489a5393 LJ |
636 | if (cpus_empty(cp->cpus_allowed)) |
637 | continue; | |
638 | ||
f5393693 LJ |
639 | /* |
640 | * All child cpusets contain a subset of the parent's cpus, so | |
641 | * just skip them, and then we call update_domain_attr_tree() | |
642 | * to calc relax_domain_level of the corresponding sched | |
643 | * domain. | |
644 | */ | |
645 | if (is_sched_load_balance(cp)) { | |
029190c5 | 646 | csa[csn++] = cp; |
f5393693 LJ |
647 | continue; |
648 | } | |
489a5393 | 649 | |
029190c5 PJ |
650 | list_for_each_entry(cont, &cp->css.cgroup->children, sibling) { |
651 | child = cgroup_cs(cont); | |
aeed6824 | 652 | list_add_tail(&child->stack_list, &q); |
029190c5 PJ |
653 | } |
654 | } | |
655 | ||
656 | for (i = 0; i < csn; i++) | |
657 | csa[i]->pn = i; | |
658 | ndoms = csn; | |
659 | ||
660 | restart: | |
661 | /* Find the best partition (set of sched domains) */ | |
662 | for (i = 0; i < csn; i++) { | |
663 | struct cpuset *a = csa[i]; | |
664 | int apn = a->pn; | |
665 | ||
666 | for (j = 0; j < csn; j++) { | |
667 | struct cpuset *b = csa[j]; | |
668 | int bpn = b->pn; | |
669 | ||
670 | if (apn != bpn && cpusets_overlap(a, b)) { | |
671 | for (k = 0; k < csn; k++) { | |
672 | struct cpuset *c = csa[k]; | |
673 | ||
674 | if (c->pn == bpn) | |
675 | c->pn = apn; | |
676 | } | |
677 | ndoms--; /* one less element */ | |
678 | goto restart; | |
679 | } | |
680 | } | |
681 | } | |
682 | ||
683 | /* Convert <csn, csa> to <ndoms, doms> */ | |
684 | doms = kmalloc(ndoms * sizeof(cpumask_t), GFP_KERNEL); | |
685 | if (!doms) | |
686 | goto rebuild; | |
1d3504fc | 687 | dattr = kmalloc(ndoms * sizeof(struct sched_domain_attr), GFP_KERNEL); |
029190c5 PJ |
688 | |
689 | for (nslot = 0, i = 0; i < csn; i++) { | |
690 | struct cpuset *a = csa[i]; | |
691 | int apn = a->pn; | |
692 | ||
693 | if (apn >= 0) { | |
694 | cpumask_t *dp = doms + nslot; | |
695 | ||
696 | if (nslot == ndoms) { | |
697 | static int warnings = 10; | |
698 | if (warnings) { | |
699 | printk(KERN_WARNING | |
700 | "rebuild_sched_domains confused:" | |
701 | " nslot %d, ndoms %d, csn %d, i %d," | |
702 | " apn %d\n", | |
703 | nslot, ndoms, csn, i, apn); | |
704 | warnings--; | |
705 | } | |
706 | continue; | |
707 | } | |
708 | ||
709 | cpus_clear(*dp); | |
1d3504fc HS |
710 | if (dattr) |
711 | *(dattr + nslot) = SD_ATTR_INIT; | |
029190c5 PJ |
712 | for (j = i; j < csn; j++) { |
713 | struct cpuset *b = csa[j]; | |
714 | ||
715 | if (apn == b->pn) { | |
716 | cpus_or(*dp, *dp, b->cpus_allowed); | |
717 | b->pn = -1; | |
91cd4d6e | 718 | if (dattr) |
f5393693 | 719 | update_domain_attr_tree(dattr |
91cd4d6e | 720 | + nslot, b); |
029190c5 PJ |
721 | } |
722 | } | |
723 | nslot++; | |
724 | } | |
725 | } | |
726 | BUG_ON(nslot != ndoms); | |
727 | ||
728 | rebuild: | |
729 | /* Have scheduler rebuild sched domains */ | |
86ef5c9a | 730 | get_online_cpus(); |
1d3504fc | 731 | partition_sched_domains(ndoms, doms, dattr); |
86ef5c9a | 732 | put_online_cpus(); |
029190c5 PJ |
733 | |
734 | done: | |
029190c5 PJ |
735 | kfree(csa); |
736 | /* Don't kfree(doms) -- partition_sched_domains() does that. */ | |
1d3504fc | 737 | /* Don't kfree(dattr) -- partition_sched_domains() does that. */ |
029190c5 PJ |
738 | } |
739 | ||
58f4790b CW |
740 | /** |
741 | * cpuset_test_cpumask - test a task's cpus_allowed versus its cpuset's | |
742 | * @tsk: task to test | |
743 | * @scan: struct cgroup_scanner contained in its struct cpuset_hotplug_scanner | |
744 | * | |
2df167a3 | 745 | * Call with cgroup_mutex held. May take callback_mutex during call. |
58f4790b CW |
746 | * Called for each task in a cgroup by cgroup_scan_tasks(). |
747 | * Return nonzero if this tasks's cpus_allowed mask should be changed (in other | |
748 | * words, if its mask is not equal to its cpuset's mask). | |
053199ed | 749 | */ |
9e0c914c AB |
750 | static int cpuset_test_cpumask(struct task_struct *tsk, |
751 | struct cgroup_scanner *scan) | |
58f4790b CW |
752 | { |
753 | return !cpus_equal(tsk->cpus_allowed, | |
754 | (cgroup_cs(scan->cg))->cpus_allowed); | |
755 | } | |
053199ed | 756 | |
58f4790b CW |
757 | /** |
758 | * cpuset_change_cpumask - make a task's cpus_allowed the same as its cpuset's | |
759 | * @tsk: task to test | |
760 | * @scan: struct cgroup_scanner containing the cgroup of the task | |
761 | * | |
762 | * Called by cgroup_scan_tasks() for each task in a cgroup whose | |
763 | * cpus_allowed mask needs to be changed. | |
764 | * | |
765 | * We don't need to re-check for the cgroup/cpuset membership, since we're | |
766 | * holding cgroup_lock() at this point. | |
767 | */ | |
9e0c914c AB |
768 | static void cpuset_change_cpumask(struct task_struct *tsk, |
769 | struct cgroup_scanner *scan) | |
58f4790b | 770 | { |
f9a86fcb | 771 | set_cpus_allowed_ptr(tsk, &((cgroup_cs(scan->cg))->cpus_allowed)); |
58f4790b CW |
772 | } |
773 | ||
0b2f630a MX |
774 | /** |
775 | * update_tasks_cpumask - Update the cpumasks of tasks in the cpuset. | |
776 | * @cs: the cpuset in which each task's cpus_allowed mask needs to be changed | |
777 | * | |
778 | * Called with cgroup_mutex held | |
779 | * | |
780 | * The cgroup_scan_tasks() function will scan all the tasks in a cgroup, | |
781 | * calling callback functions for each. | |
782 | * | |
783 | * Return 0 if successful, -errno if not. | |
784 | */ | |
785 | static int update_tasks_cpumask(struct cpuset *cs) | |
786 | { | |
787 | struct cgroup_scanner scan; | |
788 | struct ptr_heap heap; | |
789 | int retval; | |
790 | ||
02412483 LJ |
791 | /* |
792 | * cgroup_scan_tasks() will initialize heap->gt for us. | |
793 | * heap_init() is still needed here for we should not change | |
794 | * cs->cpus_allowed when heap_init() fails. | |
795 | */ | |
796 | retval = heap_init(&heap, PAGE_SIZE, GFP_KERNEL, NULL); | |
0b2f630a MX |
797 | if (retval) |
798 | return retval; | |
799 | ||
800 | scan.cg = cs->css.cgroup; | |
801 | scan.test_task = cpuset_test_cpumask; | |
802 | scan.process_task = cpuset_change_cpumask; | |
803 | scan.heap = &heap; | |
804 | retval = cgroup_scan_tasks(&scan); | |
805 | ||
806 | heap_free(&heap); | |
807 | return retval; | |
808 | } | |
809 | ||
58f4790b CW |
810 | /** |
811 | * update_cpumask - update the cpus_allowed mask of a cpuset and all tasks in it | |
812 | * @cs: the cpuset to consider | |
813 | * @buf: buffer of cpu numbers written to this cpuset | |
814 | */ | |
e3712395 | 815 | static int update_cpumask(struct cpuset *cs, const char *buf) |
1da177e4 LT |
816 | { |
817 | struct cpuset trialcs; | |
58f4790b CW |
818 | int retval; |
819 | int is_load_balanced; | |
1da177e4 | 820 | |
4c4d50f7 PJ |
821 | /* top_cpuset.cpus_allowed tracks cpu_online_map; it's read-only */ |
822 | if (cs == &top_cpuset) | |
823 | return -EACCES; | |
824 | ||
1da177e4 | 825 | trialcs = *cs; |
6f7f02e7 DR |
826 | |
827 | /* | |
c8d9c90c | 828 | * An empty cpus_allowed is ok only if the cpuset has no tasks. |
020958b6 PJ |
829 | * Since cpulist_parse() fails on an empty mask, we special case |
830 | * that parsing. The validate_change() call ensures that cpusets | |
831 | * with tasks have cpus. | |
6f7f02e7 | 832 | */ |
020958b6 | 833 | if (!*buf) { |
6f7f02e7 DR |
834 | cpus_clear(trialcs.cpus_allowed); |
835 | } else { | |
836 | retval = cpulist_parse(buf, trialcs.cpus_allowed); | |
837 | if (retval < 0) | |
838 | return retval; | |
37340746 LJ |
839 | |
840 | if (!cpus_subset(trialcs.cpus_allowed, cpu_online_map)) | |
841 | return -EINVAL; | |
6f7f02e7 | 842 | } |
1da177e4 | 843 | retval = validate_change(cs, &trialcs); |
85d7b949 DG |
844 | if (retval < 0) |
845 | return retval; | |
029190c5 | 846 | |
8707d8b8 PM |
847 | /* Nothing to do if the cpus didn't change */ |
848 | if (cpus_equal(cs->cpus_allowed, trialcs.cpus_allowed)) | |
849 | return 0; | |
58f4790b | 850 | |
029190c5 PJ |
851 | is_load_balanced = is_sched_load_balance(&trialcs); |
852 | ||
3d3f26a7 | 853 | mutex_lock(&callback_mutex); |
85d7b949 | 854 | cs->cpus_allowed = trialcs.cpus_allowed; |
3d3f26a7 | 855 | mutex_unlock(&callback_mutex); |
029190c5 | 856 | |
8707d8b8 PM |
857 | /* |
858 | * Scan tasks in the cpuset, and update the cpumasks of any | |
58f4790b | 859 | * that need an update. |
8707d8b8 | 860 | */ |
0b2f630a MX |
861 | retval = update_tasks_cpumask(cs); |
862 | if (retval < 0) | |
863 | return retval; | |
58f4790b | 864 | |
8707d8b8 | 865 | if (is_load_balanced) |
029190c5 | 866 | rebuild_sched_domains(); |
85d7b949 | 867 | return 0; |
1da177e4 LT |
868 | } |
869 | ||
e4e364e8 PJ |
870 | /* |
871 | * cpuset_migrate_mm | |
872 | * | |
873 | * Migrate memory region from one set of nodes to another. | |
874 | * | |
875 | * Temporarilly set tasks mems_allowed to target nodes of migration, | |
876 | * so that the migration code can allocate pages on these nodes. | |
877 | * | |
2df167a3 | 878 | * Call holding cgroup_mutex, so current's cpuset won't change |
c8d9c90c | 879 | * during this call, as manage_mutex holds off any cpuset_attach() |
e4e364e8 PJ |
880 | * calls. Therefore we don't need to take task_lock around the |
881 | * call to guarantee_online_mems(), as we know no one is changing | |
2df167a3 | 882 | * our task's cpuset. |
e4e364e8 PJ |
883 | * |
884 | * Hold callback_mutex around the two modifications of our tasks | |
885 | * mems_allowed to synchronize with cpuset_mems_allowed(). | |
886 | * | |
887 | * While the mm_struct we are migrating is typically from some | |
888 | * other task, the task_struct mems_allowed that we are hacking | |
889 | * is for our current task, which must allocate new pages for that | |
890 | * migrating memory region. | |
891 | * | |
892 | * We call cpuset_update_task_memory_state() before hacking | |
893 | * our tasks mems_allowed, so that we are assured of being in | |
894 | * sync with our tasks cpuset, and in particular, callbacks to | |
895 | * cpuset_update_task_memory_state() from nested page allocations | |
896 | * won't see any mismatch of our cpuset and task mems_generation | |
897 | * values, so won't overwrite our hacked tasks mems_allowed | |
898 | * nodemask. | |
899 | */ | |
900 | ||
901 | static void cpuset_migrate_mm(struct mm_struct *mm, const nodemask_t *from, | |
902 | const nodemask_t *to) | |
903 | { | |
904 | struct task_struct *tsk = current; | |
905 | ||
906 | cpuset_update_task_memory_state(); | |
907 | ||
908 | mutex_lock(&callback_mutex); | |
909 | tsk->mems_allowed = *to; | |
910 | mutex_unlock(&callback_mutex); | |
911 | ||
912 | do_migrate_pages(mm, from, to, MPOL_MF_MOVE_ALL); | |
913 | ||
914 | mutex_lock(&callback_mutex); | |
8793d854 | 915 | guarantee_online_mems(task_cs(tsk),&tsk->mems_allowed); |
e4e364e8 PJ |
916 | mutex_unlock(&callback_mutex); |
917 | } | |
918 | ||
8793d854 PM |
919 | static void *cpuset_being_rebound; |
920 | ||
0b2f630a MX |
921 | /** |
922 | * update_tasks_nodemask - Update the nodemasks of tasks in the cpuset. | |
923 | * @cs: the cpuset in which each task's mems_allowed mask needs to be changed | |
924 | * @oldmem: old mems_allowed of cpuset cs | |
925 | * | |
926 | * Called with cgroup_mutex held | |
927 | * Return 0 if successful, -errno if not. | |
928 | */ | |
929 | static int update_tasks_nodemask(struct cpuset *cs, const nodemask_t *oldmem) | |
1da177e4 | 930 | { |
8793d854 | 931 | struct task_struct *p; |
4225399a PJ |
932 | struct mm_struct **mmarray; |
933 | int i, n, ntasks; | |
04c19fa6 | 934 | int migrate; |
4225399a | 935 | int fudge; |
8793d854 | 936 | struct cgroup_iter it; |
0b2f630a | 937 | int retval; |
59dac16f | 938 | |
846a16bf | 939 | cpuset_being_rebound = cs; /* causes mpol_dup() rebind */ |
4225399a PJ |
940 | |
941 | fudge = 10; /* spare mmarray[] slots */ | |
942 | fudge += cpus_weight(cs->cpus_allowed); /* imagine one fork-bomb/cpu */ | |
943 | retval = -ENOMEM; | |
944 | ||
945 | /* | |
946 | * Allocate mmarray[] to hold mm reference for each task | |
947 | * in cpuset cs. Can't kmalloc GFP_KERNEL while holding | |
948 | * tasklist_lock. We could use GFP_ATOMIC, but with a | |
949 | * few more lines of code, we can retry until we get a big | |
950 | * enough mmarray[] w/o using GFP_ATOMIC. | |
951 | */ | |
952 | while (1) { | |
8793d854 | 953 | ntasks = cgroup_task_count(cs->css.cgroup); /* guess */ |
4225399a PJ |
954 | ntasks += fudge; |
955 | mmarray = kmalloc(ntasks * sizeof(*mmarray), GFP_KERNEL); | |
956 | if (!mmarray) | |
957 | goto done; | |
c2aef333 | 958 | read_lock(&tasklist_lock); /* block fork */ |
8793d854 | 959 | if (cgroup_task_count(cs->css.cgroup) <= ntasks) |
4225399a | 960 | break; /* got enough */ |
c2aef333 | 961 | read_unlock(&tasklist_lock); /* try again */ |
4225399a PJ |
962 | kfree(mmarray); |
963 | } | |
964 | ||
965 | n = 0; | |
966 | ||
967 | /* Load up mmarray[] with mm reference for each task in cpuset. */ | |
8793d854 PM |
968 | cgroup_iter_start(cs->css.cgroup, &it); |
969 | while ((p = cgroup_iter_next(cs->css.cgroup, &it))) { | |
4225399a PJ |
970 | struct mm_struct *mm; |
971 | ||
972 | if (n >= ntasks) { | |
973 | printk(KERN_WARNING | |
974 | "Cpuset mempolicy rebind incomplete.\n"); | |
8793d854 | 975 | break; |
4225399a | 976 | } |
4225399a PJ |
977 | mm = get_task_mm(p); |
978 | if (!mm) | |
979 | continue; | |
980 | mmarray[n++] = mm; | |
8793d854 PM |
981 | } |
982 | cgroup_iter_end(cs->css.cgroup, &it); | |
c2aef333 | 983 | read_unlock(&tasklist_lock); |
4225399a PJ |
984 | |
985 | /* | |
986 | * Now that we've dropped the tasklist spinlock, we can | |
987 | * rebind the vma mempolicies of each mm in mmarray[] to their | |
988 | * new cpuset, and release that mm. The mpol_rebind_mm() | |
989 | * call takes mmap_sem, which we couldn't take while holding | |
846a16bf | 990 | * tasklist_lock. Forks can happen again now - the mpol_dup() |
4225399a PJ |
991 | * cpuset_being_rebound check will catch such forks, and rebind |
992 | * their vma mempolicies too. Because we still hold the global | |
2df167a3 | 993 | * cgroup_mutex, we know that no other rebind effort will |
4225399a PJ |
994 | * be contending for the global variable cpuset_being_rebound. |
995 | * It's ok if we rebind the same mm twice; mpol_rebind_mm() | |
04c19fa6 | 996 | * is idempotent. Also migrate pages in each mm to new nodes. |
4225399a | 997 | */ |
04c19fa6 | 998 | migrate = is_memory_migrate(cs); |
4225399a PJ |
999 | for (i = 0; i < n; i++) { |
1000 | struct mm_struct *mm = mmarray[i]; | |
1001 | ||
1002 | mpol_rebind_mm(mm, &cs->mems_allowed); | |
e4e364e8 | 1003 | if (migrate) |
0b2f630a | 1004 | cpuset_migrate_mm(mm, oldmem, &cs->mems_allowed); |
4225399a PJ |
1005 | mmput(mm); |
1006 | } | |
1007 | ||
2df167a3 | 1008 | /* We're done rebinding vmas to this cpuset's new mems_allowed. */ |
4225399a | 1009 | kfree(mmarray); |
8793d854 | 1010 | cpuset_being_rebound = NULL; |
4225399a | 1011 | retval = 0; |
59dac16f | 1012 | done: |
1da177e4 LT |
1013 | return retval; |
1014 | } | |
1015 | ||
0b2f630a MX |
1016 | /* |
1017 | * Handle user request to change the 'mems' memory placement | |
1018 | * of a cpuset. Needs to validate the request, update the | |
1019 | * cpusets mems_allowed and mems_generation, and for each | |
1020 | * task in the cpuset, rebind any vma mempolicies and if | |
1021 | * the cpuset is marked 'memory_migrate', migrate the tasks | |
1022 | * pages to the new memory. | |
1023 | * | |
1024 | * Call with cgroup_mutex held. May take callback_mutex during call. | |
1025 | * Will take tasklist_lock, scan tasklist for tasks in cpuset cs, | |
1026 | * lock each such tasks mm->mmap_sem, scan its vma's and rebind | |
1027 | * their mempolicies to the cpusets new mems_allowed. | |
1028 | */ | |
1029 | static int update_nodemask(struct cpuset *cs, const char *buf) | |
1030 | { | |
1031 | struct cpuset trialcs; | |
1032 | nodemask_t oldmem; | |
1033 | int retval; | |
1034 | ||
1035 | /* | |
1036 | * top_cpuset.mems_allowed tracks node_stats[N_HIGH_MEMORY]; | |
1037 | * it's read-only | |
1038 | */ | |
1039 | if (cs == &top_cpuset) | |
1040 | return -EACCES; | |
1041 | ||
1042 | trialcs = *cs; | |
1043 | ||
1044 | /* | |
1045 | * An empty mems_allowed is ok iff there are no tasks in the cpuset. | |
1046 | * Since nodelist_parse() fails on an empty mask, we special case | |
1047 | * that parsing. The validate_change() call ensures that cpusets | |
1048 | * with tasks have memory. | |
1049 | */ | |
1050 | if (!*buf) { | |
1051 | nodes_clear(trialcs.mems_allowed); | |
1052 | } else { | |
1053 | retval = nodelist_parse(buf, trialcs.mems_allowed); | |
1054 | if (retval < 0) | |
1055 | goto done; | |
1056 | ||
1057 | if (!nodes_subset(trialcs.mems_allowed, | |
1058 | node_states[N_HIGH_MEMORY])) | |
1059 | return -EINVAL; | |
1060 | } | |
1061 | oldmem = cs->mems_allowed; | |
1062 | if (nodes_equal(oldmem, trialcs.mems_allowed)) { | |
1063 | retval = 0; /* Too easy - nothing to do */ | |
1064 | goto done; | |
1065 | } | |
1066 | retval = validate_change(cs, &trialcs); | |
1067 | if (retval < 0) | |
1068 | goto done; | |
1069 | ||
1070 | mutex_lock(&callback_mutex); | |
1071 | cs->mems_allowed = trialcs.mems_allowed; | |
1072 | cs->mems_generation = cpuset_mems_generation++; | |
1073 | mutex_unlock(&callback_mutex); | |
1074 | ||
1075 | retval = update_tasks_nodemask(cs, &oldmem); | |
1076 | done: | |
1077 | return retval; | |
1078 | } | |
1079 | ||
8793d854 PM |
1080 | int current_cpuset_is_being_rebound(void) |
1081 | { | |
1082 | return task_cs(current) == cpuset_being_rebound; | |
1083 | } | |
1084 | ||
5be7a479 | 1085 | static int update_relax_domain_level(struct cpuset *cs, s64 val) |
1d3504fc | 1086 | { |
30e0e178 LZ |
1087 | if (val < -1 || val >= SD_LV_MAX) |
1088 | return -EINVAL; | |
1d3504fc HS |
1089 | |
1090 | if (val != cs->relax_domain_level) { | |
1091 | cs->relax_domain_level = val; | |
c372e817 LZ |
1092 | if (!cpus_empty(cs->cpus_allowed) && is_sched_load_balance(cs)) |
1093 | rebuild_sched_domains(); | |
1d3504fc HS |
1094 | } |
1095 | ||
1096 | return 0; | |
1097 | } | |
1098 | ||
1da177e4 LT |
1099 | /* |
1100 | * update_flag - read a 0 or a 1 in a file and update associated flag | |
78608366 PM |
1101 | * bit: the bit to update (see cpuset_flagbits_t) |
1102 | * cs: the cpuset to update | |
1103 | * turning_on: whether the flag is being set or cleared | |
053199ed | 1104 | * |
2df167a3 | 1105 | * Call with cgroup_mutex held. |
1da177e4 LT |
1106 | */ |
1107 | ||
700fe1ab PM |
1108 | static int update_flag(cpuset_flagbits_t bit, struct cpuset *cs, |
1109 | int turning_on) | |
1da177e4 | 1110 | { |
1da177e4 | 1111 | struct cpuset trialcs; |
607717a6 | 1112 | int err; |
029190c5 | 1113 | int cpus_nonempty, balance_flag_changed; |
1da177e4 | 1114 | |
1da177e4 LT |
1115 | trialcs = *cs; |
1116 | if (turning_on) | |
1117 | set_bit(bit, &trialcs.flags); | |
1118 | else | |
1119 | clear_bit(bit, &trialcs.flags); | |
1120 | ||
1121 | err = validate_change(cs, &trialcs); | |
85d7b949 DG |
1122 | if (err < 0) |
1123 | return err; | |
029190c5 PJ |
1124 | |
1125 | cpus_nonempty = !cpus_empty(trialcs.cpus_allowed); | |
1126 | balance_flag_changed = (is_sched_load_balance(cs) != | |
1127 | is_sched_load_balance(&trialcs)); | |
1128 | ||
3d3f26a7 | 1129 | mutex_lock(&callback_mutex); |
69604067 | 1130 | cs->flags = trialcs.flags; |
3d3f26a7 | 1131 | mutex_unlock(&callback_mutex); |
85d7b949 | 1132 | |
029190c5 PJ |
1133 | if (cpus_nonempty && balance_flag_changed) |
1134 | rebuild_sched_domains(); | |
1135 | ||
85d7b949 | 1136 | return 0; |
1da177e4 LT |
1137 | } |
1138 | ||
3e0d98b9 | 1139 | /* |
80f7228b | 1140 | * Frequency meter - How fast is some event occurring? |
3e0d98b9 PJ |
1141 | * |
1142 | * These routines manage a digitally filtered, constant time based, | |
1143 | * event frequency meter. There are four routines: | |
1144 | * fmeter_init() - initialize a frequency meter. | |
1145 | * fmeter_markevent() - called each time the event happens. | |
1146 | * fmeter_getrate() - returns the recent rate of such events. | |
1147 | * fmeter_update() - internal routine used to update fmeter. | |
1148 | * | |
1149 | * A common data structure is passed to each of these routines, | |
1150 | * which is used to keep track of the state required to manage the | |
1151 | * frequency meter and its digital filter. | |
1152 | * | |
1153 | * The filter works on the number of events marked per unit time. | |
1154 | * The filter is single-pole low-pass recursive (IIR). The time unit | |
1155 | * is 1 second. Arithmetic is done using 32-bit integers scaled to | |
1156 | * simulate 3 decimal digits of precision (multiplied by 1000). | |
1157 | * | |
1158 | * With an FM_COEF of 933, and a time base of 1 second, the filter | |
1159 | * has a half-life of 10 seconds, meaning that if the events quit | |
1160 | * happening, then the rate returned from the fmeter_getrate() | |
1161 | * will be cut in half each 10 seconds, until it converges to zero. | |
1162 | * | |
1163 | * It is not worth doing a real infinitely recursive filter. If more | |
1164 | * than FM_MAXTICKS ticks have elapsed since the last filter event, | |
1165 | * just compute FM_MAXTICKS ticks worth, by which point the level | |
1166 | * will be stable. | |
1167 | * | |
1168 | * Limit the count of unprocessed events to FM_MAXCNT, so as to avoid | |
1169 | * arithmetic overflow in the fmeter_update() routine. | |
1170 | * | |
1171 | * Given the simple 32 bit integer arithmetic used, this meter works | |
1172 | * best for reporting rates between one per millisecond (msec) and | |
1173 | * one per 32 (approx) seconds. At constant rates faster than one | |
1174 | * per msec it maxes out at values just under 1,000,000. At constant | |
1175 | * rates between one per msec, and one per second it will stabilize | |
1176 | * to a value N*1000, where N is the rate of events per second. | |
1177 | * At constant rates between one per second and one per 32 seconds, | |
1178 | * it will be choppy, moving up on the seconds that have an event, | |
1179 | * and then decaying until the next event. At rates slower than | |
1180 | * about one in 32 seconds, it decays all the way back to zero between | |
1181 | * each event. | |
1182 | */ | |
1183 | ||
1184 | #define FM_COEF 933 /* coefficient for half-life of 10 secs */ | |
1185 | #define FM_MAXTICKS ((time_t)99) /* useless computing more ticks than this */ | |
1186 | #define FM_MAXCNT 1000000 /* limit cnt to avoid overflow */ | |
1187 | #define FM_SCALE 1000 /* faux fixed point scale */ | |
1188 | ||
1189 | /* Initialize a frequency meter */ | |
1190 | static void fmeter_init(struct fmeter *fmp) | |
1191 | { | |
1192 | fmp->cnt = 0; | |
1193 | fmp->val = 0; | |
1194 | fmp->time = 0; | |
1195 | spin_lock_init(&fmp->lock); | |
1196 | } | |
1197 | ||
1198 | /* Internal meter update - process cnt events and update value */ | |
1199 | static void fmeter_update(struct fmeter *fmp) | |
1200 | { | |
1201 | time_t now = get_seconds(); | |
1202 | time_t ticks = now - fmp->time; | |
1203 | ||
1204 | if (ticks == 0) | |
1205 | return; | |
1206 | ||
1207 | ticks = min(FM_MAXTICKS, ticks); | |
1208 | while (ticks-- > 0) | |
1209 | fmp->val = (FM_COEF * fmp->val) / FM_SCALE; | |
1210 | fmp->time = now; | |
1211 | ||
1212 | fmp->val += ((FM_SCALE - FM_COEF) * fmp->cnt) / FM_SCALE; | |
1213 | fmp->cnt = 0; | |
1214 | } | |
1215 | ||
1216 | /* Process any previous ticks, then bump cnt by one (times scale). */ | |
1217 | static void fmeter_markevent(struct fmeter *fmp) | |
1218 | { | |
1219 | spin_lock(&fmp->lock); | |
1220 | fmeter_update(fmp); | |
1221 | fmp->cnt = min(FM_MAXCNT, fmp->cnt + FM_SCALE); | |
1222 | spin_unlock(&fmp->lock); | |
1223 | } | |
1224 | ||
1225 | /* Process any previous ticks, then return current value. */ | |
1226 | static int fmeter_getrate(struct fmeter *fmp) | |
1227 | { | |
1228 | int val; | |
1229 | ||
1230 | spin_lock(&fmp->lock); | |
1231 | fmeter_update(fmp); | |
1232 | val = fmp->val; | |
1233 | spin_unlock(&fmp->lock); | |
1234 | return val; | |
1235 | } | |
1236 | ||
2df167a3 | 1237 | /* Called by cgroups to determine if a cpuset is usable; cgroup_mutex held */ |
8793d854 PM |
1238 | static int cpuset_can_attach(struct cgroup_subsys *ss, |
1239 | struct cgroup *cont, struct task_struct *tsk) | |
1da177e4 | 1240 | { |
8793d854 | 1241 | struct cpuset *cs = cgroup_cs(cont); |
1da177e4 | 1242 | |
1da177e4 LT |
1243 | if (cpus_empty(cs->cpus_allowed) || nodes_empty(cs->mems_allowed)) |
1244 | return -ENOSPC; | |
9985b0ba DR |
1245 | if (tsk->flags & PF_THREAD_BOUND) { |
1246 | cpumask_t mask; | |
1247 | ||
1248 | mutex_lock(&callback_mutex); | |
1249 | mask = cs->cpus_allowed; | |
1250 | mutex_unlock(&callback_mutex); | |
1251 | if (!cpus_equal(tsk->cpus_allowed, mask)) | |
1252 | return -EINVAL; | |
1253 | } | |
1da177e4 | 1254 | |
8793d854 PM |
1255 | return security_task_setscheduler(tsk, 0, NULL); |
1256 | } | |
1da177e4 | 1257 | |
8793d854 PM |
1258 | static void cpuset_attach(struct cgroup_subsys *ss, |
1259 | struct cgroup *cont, struct cgroup *oldcont, | |
1260 | struct task_struct *tsk) | |
1261 | { | |
1262 | cpumask_t cpus; | |
1263 | nodemask_t from, to; | |
1264 | struct mm_struct *mm; | |
1265 | struct cpuset *cs = cgroup_cs(cont); | |
1266 | struct cpuset *oldcs = cgroup_cs(oldcont); | |
9985b0ba | 1267 | int err; |
22fb52dd | 1268 | |
3d3f26a7 | 1269 | mutex_lock(&callback_mutex); |
1da177e4 | 1270 | guarantee_online_cpus(cs, &cpus); |
9985b0ba | 1271 | err = set_cpus_allowed_ptr(tsk, &cpus); |
8793d854 | 1272 | mutex_unlock(&callback_mutex); |
9985b0ba DR |
1273 | if (err) |
1274 | return; | |
1da177e4 | 1275 | |
45b07ef3 PJ |
1276 | from = oldcs->mems_allowed; |
1277 | to = cs->mems_allowed; | |
4225399a PJ |
1278 | mm = get_task_mm(tsk); |
1279 | if (mm) { | |
1280 | mpol_rebind_mm(mm, &to); | |
2741a559 | 1281 | if (is_memory_migrate(cs)) |
e4e364e8 | 1282 | cpuset_migrate_mm(mm, &from, &to); |
4225399a PJ |
1283 | mmput(mm); |
1284 | } | |
1285 | ||
1da177e4 LT |
1286 | } |
1287 | ||
1288 | /* The various types of files and directories in a cpuset file system */ | |
1289 | ||
1290 | typedef enum { | |
45b07ef3 | 1291 | FILE_MEMORY_MIGRATE, |
1da177e4 LT |
1292 | FILE_CPULIST, |
1293 | FILE_MEMLIST, | |
1294 | FILE_CPU_EXCLUSIVE, | |
1295 | FILE_MEM_EXCLUSIVE, | |
78608366 | 1296 | FILE_MEM_HARDWALL, |
029190c5 | 1297 | FILE_SCHED_LOAD_BALANCE, |
1d3504fc | 1298 | FILE_SCHED_RELAX_DOMAIN_LEVEL, |
3e0d98b9 PJ |
1299 | FILE_MEMORY_PRESSURE_ENABLED, |
1300 | FILE_MEMORY_PRESSURE, | |
825a46af PJ |
1301 | FILE_SPREAD_PAGE, |
1302 | FILE_SPREAD_SLAB, | |
1da177e4 LT |
1303 | } cpuset_filetype_t; |
1304 | ||
700fe1ab PM |
1305 | static int cpuset_write_u64(struct cgroup *cgrp, struct cftype *cft, u64 val) |
1306 | { | |
1307 | int retval = 0; | |
1308 | struct cpuset *cs = cgroup_cs(cgrp); | |
1309 | cpuset_filetype_t type = cft->private; | |
1310 | ||
e3712395 | 1311 | if (!cgroup_lock_live_group(cgrp)) |
700fe1ab | 1312 | return -ENODEV; |
700fe1ab PM |
1313 | |
1314 | switch (type) { | |
1da177e4 | 1315 | case FILE_CPU_EXCLUSIVE: |
700fe1ab | 1316 | retval = update_flag(CS_CPU_EXCLUSIVE, cs, val); |
1da177e4 LT |
1317 | break; |
1318 | case FILE_MEM_EXCLUSIVE: | |
700fe1ab | 1319 | retval = update_flag(CS_MEM_EXCLUSIVE, cs, val); |
1da177e4 | 1320 | break; |
78608366 PM |
1321 | case FILE_MEM_HARDWALL: |
1322 | retval = update_flag(CS_MEM_HARDWALL, cs, val); | |
1323 | break; | |
029190c5 | 1324 | case FILE_SCHED_LOAD_BALANCE: |
700fe1ab | 1325 | retval = update_flag(CS_SCHED_LOAD_BALANCE, cs, val); |
1d3504fc | 1326 | break; |
45b07ef3 | 1327 | case FILE_MEMORY_MIGRATE: |
700fe1ab | 1328 | retval = update_flag(CS_MEMORY_MIGRATE, cs, val); |
45b07ef3 | 1329 | break; |
3e0d98b9 | 1330 | case FILE_MEMORY_PRESSURE_ENABLED: |
700fe1ab | 1331 | cpuset_memory_pressure_enabled = !!val; |
3e0d98b9 PJ |
1332 | break; |
1333 | case FILE_MEMORY_PRESSURE: | |
1334 | retval = -EACCES; | |
1335 | break; | |
825a46af | 1336 | case FILE_SPREAD_PAGE: |
700fe1ab | 1337 | retval = update_flag(CS_SPREAD_PAGE, cs, val); |
151a4420 | 1338 | cs->mems_generation = cpuset_mems_generation++; |
825a46af PJ |
1339 | break; |
1340 | case FILE_SPREAD_SLAB: | |
700fe1ab | 1341 | retval = update_flag(CS_SPREAD_SLAB, cs, val); |
151a4420 | 1342 | cs->mems_generation = cpuset_mems_generation++; |
825a46af | 1343 | break; |
1da177e4 LT |
1344 | default: |
1345 | retval = -EINVAL; | |
700fe1ab | 1346 | break; |
1da177e4 | 1347 | } |
8793d854 | 1348 | cgroup_unlock(); |
1da177e4 LT |
1349 | return retval; |
1350 | } | |
1351 | ||
5be7a479 PM |
1352 | static int cpuset_write_s64(struct cgroup *cgrp, struct cftype *cft, s64 val) |
1353 | { | |
1354 | int retval = 0; | |
1355 | struct cpuset *cs = cgroup_cs(cgrp); | |
1356 | cpuset_filetype_t type = cft->private; | |
1357 | ||
e3712395 | 1358 | if (!cgroup_lock_live_group(cgrp)) |
5be7a479 | 1359 | return -ENODEV; |
e3712395 | 1360 | |
5be7a479 PM |
1361 | switch (type) { |
1362 | case FILE_SCHED_RELAX_DOMAIN_LEVEL: | |
1363 | retval = update_relax_domain_level(cs, val); | |
1364 | break; | |
1365 | default: | |
1366 | retval = -EINVAL; | |
1367 | break; | |
1368 | } | |
1369 | cgroup_unlock(); | |
1370 | return retval; | |
1371 | } | |
1372 | ||
e3712395 PM |
1373 | /* |
1374 | * Common handling for a write to a "cpus" or "mems" file. | |
1375 | */ | |
1376 | static int cpuset_write_resmask(struct cgroup *cgrp, struct cftype *cft, | |
1377 | const char *buf) | |
1378 | { | |
1379 | int retval = 0; | |
1380 | ||
1381 | if (!cgroup_lock_live_group(cgrp)) | |
1382 | return -ENODEV; | |
1383 | ||
1384 | switch (cft->private) { | |
1385 | case FILE_CPULIST: | |
1386 | retval = update_cpumask(cgroup_cs(cgrp), buf); | |
1387 | break; | |
1388 | case FILE_MEMLIST: | |
1389 | retval = update_nodemask(cgroup_cs(cgrp), buf); | |
1390 | break; | |
1391 | default: | |
1392 | retval = -EINVAL; | |
1393 | break; | |
1394 | } | |
1395 | cgroup_unlock(); | |
1396 | return retval; | |
1397 | } | |
1398 | ||
1da177e4 LT |
1399 | /* |
1400 | * These ascii lists should be read in a single call, by using a user | |
1401 | * buffer large enough to hold the entire map. If read in smaller | |
1402 | * chunks, there is no guarantee of atomicity. Since the display format | |
1403 | * used, list of ranges of sequential numbers, is variable length, | |
1404 | * and since these maps can change value dynamically, one could read | |
1405 | * gibberish by doing partial reads while a list was changing. | |
1406 | * A single large read to a buffer that crosses a page boundary is | |
1407 | * ok, because the result being copied to user land is not recomputed | |
1408 | * across a page fault. | |
1409 | */ | |
1410 | ||
1411 | static int cpuset_sprintf_cpulist(char *page, struct cpuset *cs) | |
1412 | { | |
1413 | cpumask_t mask; | |
1414 | ||
3d3f26a7 | 1415 | mutex_lock(&callback_mutex); |
1da177e4 | 1416 | mask = cs->cpus_allowed; |
3d3f26a7 | 1417 | mutex_unlock(&callback_mutex); |
1da177e4 LT |
1418 | |
1419 | return cpulist_scnprintf(page, PAGE_SIZE, mask); | |
1420 | } | |
1421 | ||
1422 | static int cpuset_sprintf_memlist(char *page, struct cpuset *cs) | |
1423 | { | |
1424 | nodemask_t mask; | |
1425 | ||
3d3f26a7 | 1426 | mutex_lock(&callback_mutex); |
1da177e4 | 1427 | mask = cs->mems_allowed; |
3d3f26a7 | 1428 | mutex_unlock(&callback_mutex); |
1da177e4 LT |
1429 | |
1430 | return nodelist_scnprintf(page, PAGE_SIZE, mask); | |
1431 | } | |
1432 | ||
8793d854 PM |
1433 | static ssize_t cpuset_common_file_read(struct cgroup *cont, |
1434 | struct cftype *cft, | |
1435 | struct file *file, | |
1436 | char __user *buf, | |
1437 | size_t nbytes, loff_t *ppos) | |
1da177e4 | 1438 | { |
8793d854 | 1439 | struct cpuset *cs = cgroup_cs(cont); |
1da177e4 LT |
1440 | cpuset_filetype_t type = cft->private; |
1441 | char *page; | |
1442 | ssize_t retval = 0; | |
1443 | char *s; | |
1da177e4 | 1444 | |
e12ba74d | 1445 | if (!(page = (char *)__get_free_page(GFP_TEMPORARY))) |
1da177e4 LT |
1446 | return -ENOMEM; |
1447 | ||
1448 | s = page; | |
1449 | ||
1450 | switch (type) { | |
1451 | case FILE_CPULIST: | |
1452 | s += cpuset_sprintf_cpulist(s, cs); | |
1453 | break; | |
1454 | case FILE_MEMLIST: | |
1455 | s += cpuset_sprintf_memlist(s, cs); | |
1456 | break; | |
1da177e4 LT |
1457 | default: |
1458 | retval = -EINVAL; | |
1459 | goto out; | |
1460 | } | |
1461 | *s++ = '\n'; | |
1da177e4 | 1462 | |
eacaa1f5 | 1463 | retval = simple_read_from_buffer(buf, nbytes, ppos, page, s - page); |
1da177e4 LT |
1464 | out: |
1465 | free_page((unsigned long)page); | |
1466 | return retval; | |
1467 | } | |
1468 | ||
700fe1ab PM |
1469 | static u64 cpuset_read_u64(struct cgroup *cont, struct cftype *cft) |
1470 | { | |
1471 | struct cpuset *cs = cgroup_cs(cont); | |
1472 | cpuset_filetype_t type = cft->private; | |
1473 | switch (type) { | |
1474 | case FILE_CPU_EXCLUSIVE: | |
1475 | return is_cpu_exclusive(cs); | |
1476 | case FILE_MEM_EXCLUSIVE: | |
1477 | return is_mem_exclusive(cs); | |
78608366 PM |
1478 | case FILE_MEM_HARDWALL: |
1479 | return is_mem_hardwall(cs); | |
700fe1ab PM |
1480 | case FILE_SCHED_LOAD_BALANCE: |
1481 | return is_sched_load_balance(cs); | |
1482 | case FILE_MEMORY_MIGRATE: | |
1483 | return is_memory_migrate(cs); | |
1484 | case FILE_MEMORY_PRESSURE_ENABLED: | |
1485 | return cpuset_memory_pressure_enabled; | |
1486 | case FILE_MEMORY_PRESSURE: | |
1487 | return fmeter_getrate(&cs->fmeter); | |
1488 | case FILE_SPREAD_PAGE: | |
1489 | return is_spread_page(cs); | |
1490 | case FILE_SPREAD_SLAB: | |
1491 | return is_spread_slab(cs); | |
1492 | default: | |
1493 | BUG(); | |
1494 | } | |
1495 | } | |
1da177e4 | 1496 | |
5be7a479 PM |
1497 | static s64 cpuset_read_s64(struct cgroup *cont, struct cftype *cft) |
1498 | { | |
1499 | struct cpuset *cs = cgroup_cs(cont); | |
1500 | cpuset_filetype_t type = cft->private; | |
1501 | switch (type) { | |
1502 | case FILE_SCHED_RELAX_DOMAIN_LEVEL: | |
1503 | return cs->relax_domain_level; | |
1504 | default: | |
1505 | BUG(); | |
1506 | } | |
1507 | } | |
1508 | ||
1da177e4 LT |
1509 | |
1510 | /* | |
1511 | * for the common functions, 'private' gives the type of file | |
1512 | */ | |
1513 | ||
addf2c73 PM |
1514 | static struct cftype files[] = { |
1515 | { | |
1516 | .name = "cpus", | |
1517 | .read = cpuset_common_file_read, | |
e3712395 PM |
1518 | .write_string = cpuset_write_resmask, |
1519 | .max_write_len = (100U + 6 * NR_CPUS), | |
addf2c73 PM |
1520 | .private = FILE_CPULIST, |
1521 | }, | |
1522 | ||
1523 | { | |
1524 | .name = "mems", | |
1525 | .read = cpuset_common_file_read, | |
e3712395 PM |
1526 | .write_string = cpuset_write_resmask, |
1527 | .max_write_len = (100U + 6 * MAX_NUMNODES), | |
addf2c73 PM |
1528 | .private = FILE_MEMLIST, |
1529 | }, | |
1530 | ||
1531 | { | |
1532 | .name = "cpu_exclusive", | |
1533 | .read_u64 = cpuset_read_u64, | |
1534 | .write_u64 = cpuset_write_u64, | |
1535 | .private = FILE_CPU_EXCLUSIVE, | |
1536 | }, | |
1537 | ||
1538 | { | |
1539 | .name = "mem_exclusive", | |
1540 | .read_u64 = cpuset_read_u64, | |
1541 | .write_u64 = cpuset_write_u64, | |
1542 | .private = FILE_MEM_EXCLUSIVE, | |
1543 | }, | |
1544 | ||
78608366 PM |
1545 | { |
1546 | .name = "mem_hardwall", | |
1547 | .read_u64 = cpuset_read_u64, | |
1548 | .write_u64 = cpuset_write_u64, | |
1549 | .private = FILE_MEM_HARDWALL, | |
1550 | }, | |
1551 | ||
addf2c73 PM |
1552 | { |
1553 | .name = "sched_load_balance", | |
1554 | .read_u64 = cpuset_read_u64, | |
1555 | .write_u64 = cpuset_write_u64, | |
1556 | .private = FILE_SCHED_LOAD_BALANCE, | |
1557 | }, | |
1558 | ||
1559 | { | |
1560 | .name = "sched_relax_domain_level", | |
5be7a479 PM |
1561 | .read_s64 = cpuset_read_s64, |
1562 | .write_s64 = cpuset_write_s64, | |
addf2c73 PM |
1563 | .private = FILE_SCHED_RELAX_DOMAIN_LEVEL, |
1564 | }, | |
1565 | ||
1566 | { | |
1567 | .name = "memory_migrate", | |
1568 | .read_u64 = cpuset_read_u64, | |
1569 | .write_u64 = cpuset_write_u64, | |
1570 | .private = FILE_MEMORY_MIGRATE, | |
1571 | }, | |
1572 | ||
1573 | { | |
1574 | .name = "memory_pressure", | |
1575 | .read_u64 = cpuset_read_u64, | |
1576 | .write_u64 = cpuset_write_u64, | |
1577 | .private = FILE_MEMORY_PRESSURE, | |
1578 | }, | |
1579 | ||
1580 | { | |
1581 | .name = "memory_spread_page", | |
1582 | .read_u64 = cpuset_read_u64, | |
1583 | .write_u64 = cpuset_write_u64, | |
1584 | .private = FILE_SPREAD_PAGE, | |
1585 | }, | |
1586 | ||
1587 | { | |
1588 | .name = "memory_spread_slab", | |
1589 | .read_u64 = cpuset_read_u64, | |
1590 | .write_u64 = cpuset_write_u64, | |
1591 | .private = FILE_SPREAD_SLAB, | |
1592 | }, | |
45b07ef3 PJ |
1593 | }; |
1594 | ||
3e0d98b9 PJ |
1595 | static struct cftype cft_memory_pressure_enabled = { |
1596 | .name = "memory_pressure_enabled", | |
700fe1ab PM |
1597 | .read_u64 = cpuset_read_u64, |
1598 | .write_u64 = cpuset_write_u64, | |
3e0d98b9 PJ |
1599 | .private = FILE_MEMORY_PRESSURE_ENABLED, |
1600 | }; | |
1601 | ||
8793d854 | 1602 | static int cpuset_populate(struct cgroup_subsys *ss, struct cgroup *cont) |
1da177e4 LT |
1603 | { |
1604 | int err; | |
1605 | ||
addf2c73 PM |
1606 | err = cgroup_add_files(cont, ss, files, ARRAY_SIZE(files)); |
1607 | if (err) | |
1da177e4 | 1608 | return err; |
8793d854 | 1609 | /* memory_pressure_enabled is in root cpuset only */ |
addf2c73 | 1610 | if (!cont->parent) |
8793d854 | 1611 | err = cgroup_add_file(cont, ss, |
addf2c73 PM |
1612 | &cft_memory_pressure_enabled); |
1613 | return err; | |
1da177e4 LT |
1614 | } |
1615 | ||
8793d854 PM |
1616 | /* |
1617 | * post_clone() is called at the end of cgroup_clone(). | |
1618 | * 'cgroup' was just created automatically as a result of | |
1619 | * a cgroup_clone(), and the current task is about to | |
1620 | * be moved into 'cgroup'. | |
1621 | * | |
1622 | * Currently we refuse to set up the cgroup - thereby | |
1623 | * refusing the task to be entered, and as a result refusing | |
1624 | * the sys_unshare() or clone() which initiated it - if any | |
1625 | * sibling cpusets have exclusive cpus or mem. | |
1626 | * | |
1627 | * If this becomes a problem for some users who wish to | |
1628 | * allow that scenario, then cpuset_post_clone() could be | |
1629 | * changed to grant parent->cpus_allowed-sibling_cpus_exclusive | |
2df167a3 PM |
1630 | * (and likewise for mems) to the new cgroup. Called with cgroup_mutex |
1631 | * held. | |
8793d854 PM |
1632 | */ |
1633 | static void cpuset_post_clone(struct cgroup_subsys *ss, | |
1634 | struct cgroup *cgroup) | |
1635 | { | |
1636 | struct cgroup *parent, *child; | |
1637 | struct cpuset *cs, *parent_cs; | |
1638 | ||
1639 | parent = cgroup->parent; | |
1640 | list_for_each_entry(child, &parent->children, sibling) { | |
1641 | cs = cgroup_cs(child); | |
1642 | if (is_mem_exclusive(cs) || is_cpu_exclusive(cs)) | |
1643 | return; | |
1644 | } | |
1645 | cs = cgroup_cs(cgroup); | |
1646 | parent_cs = cgroup_cs(parent); | |
1647 | ||
1648 | cs->mems_allowed = parent_cs->mems_allowed; | |
1649 | cs->cpus_allowed = parent_cs->cpus_allowed; | |
1650 | return; | |
1651 | } | |
1652 | ||
1da177e4 LT |
1653 | /* |
1654 | * cpuset_create - create a cpuset | |
2df167a3 PM |
1655 | * ss: cpuset cgroup subsystem |
1656 | * cont: control group that the new cpuset will be part of | |
1da177e4 LT |
1657 | */ |
1658 | ||
8793d854 PM |
1659 | static struct cgroup_subsys_state *cpuset_create( |
1660 | struct cgroup_subsys *ss, | |
1661 | struct cgroup *cont) | |
1da177e4 LT |
1662 | { |
1663 | struct cpuset *cs; | |
8793d854 | 1664 | struct cpuset *parent; |
1da177e4 | 1665 | |
8793d854 PM |
1666 | if (!cont->parent) { |
1667 | /* This is early initialization for the top cgroup */ | |
1668 | top_cpuset.mems_generation = cpuset_mems_generation++; | |
1669 | return &top_cpuset.css; | |
1670 | } | |
1671 | parent = cgroup_cs(cont->parent); | |
1da177e4 LT |
1672 | cs = kmalloc(sizeof(*cs), GFP_KERNEL); |
1673 | if (!cs) | |
8793d854 | 1674 | return ERR_PTR(-ENOMEM); |
1da177e4 | 1675 | |
cf2a473c | 1676 | cpuset_update_task_memory_state(); |
1da177e4 | 1677 | cs->flags = 0; |
825a46af PJ |
1678 | if (is_spread_page(parent)) |
1679 | set_bit(CS_SPREAD_PAGE, &cs->flags); | |
1680 | if (is_spread_slab(parent)) | |
1681 | set_bit(CS_SPREAD_SLAB, &cs->flags); | |
029190c5 | 1682 | set_bit(CS_SCHED_LOAD_BALANCE, &cs->flags); |
f9a86fcb MT |
1683 | cpus_clear(cs->cpus_allowed); |
1684 | nodes_clear(cs->mems_allowed); | |
151a4420 | 1685 | cs->mems_generation = cpuset_mems_generation++; |
3e0d98b9 | 1686 | fmeter_init(&cs->fmeter); |
1d3504fc | 1687 | cs->relax_domain_level = -1; |
1da177e4 LT |
1688 | |
1689 | cs->parent = parent; | |
202f72d5 | 1690 | number_of_cpusets++; |
8793d854 | 1691 | return &cs->css ; |
1da177e4 LT |
1692 | } |
1693 | ||
029190c5 PJ |
1694 | /* |
1695 | * Locking note on the strange update_flag() call below: | |
1696 | * | |
1697 | * If the cpuset being removed has its flag 'sched_load_balance' | |
1698 | * enabled, then simulate turning sched_load_balance off, which | |
86ef5c9a | 1699 | * will call rebuild_sched_domains(). The get_online_cpus() |
029190c5 PJ |
1700 | * call in rebuild_sched_domains() must not be made while holding |
1701 | * callback_mutex. Elsewhere the kernel nests callback_mutex inside | |
86ef5c9a | 1702 | * get_online_cpus() calls. So the reverse nesting would risk an |
029190c5 PJ |
1703 | * ABBA deadlock. |
1704 | */ | |
1705 | ||
8793d854 | 1706 | static void cpuset_destroy(struct cgroup_subsys *ss, struct cgroup *cont) |
1da177e4 | 1707 | { |
8793d854 | 1708 | struct cpuset *cs = cgroup_cs(cont); |
1da177e4 | 1709 | |
cf2a473c | 1710 | cpuset_update_task_memory_state(); |
029190c5 PJ |
1711 | |
1712 | if (is_sched_load_balance(cs)) | |
700fe1ab | 1713 | update_flag(CS_SCHED_LOAD_BALANCE, cs, 0); |
029190c5 | 1714 | |
202f72d5 | 1715 | number_of_cpusets--; |
8793d854 | 1716 | kfree(cs); |
1da177e4 LT |
1717 | } |
1718 | ||
8793d854 PM |
1719 | struct cgroup_subsys cpuset_subsys = { |
1720 | .name = "cpuset", | |
1721 | .create = cpuset_create, | |
1722 | .destroy = cpuset_destroy, | |
1723 | .can_attach = cpuset_can_attach, | |
1724 | .attach = cpuset_attach, | |
1725 | .populate = cpuset_populate, | |
1726 | .post_clone = cpuset_post_clone, | |
1727 | .subsys_id = cpuset_subsys_id, | |
1728 | .early_init = 1, | |
1729 | }; | |
1730 | ||
c417f024 PJ |
1731 | /* |
1732 | * cpuset_init_early - just enough so that the calls to | |
1733 | * cpuset_update_task_memory_state() in early init code | |
1734 | * are harmless. | |
1735 | */ | |
1736 | ||
1737 | int __init cpuset_init_early(void) | |
1738 | { | |
8793d854 | 1739 | top_cpuset.mems_generation = cpuset_mems_generation++; |
c417f024 PJ |
1740 | return 0; |
1741 | } | |
1742 | ||
8793d854 | 1743 | |
1da177e4 LT |
1744 | /** |
1745 | * cpuset_init - initialize cpusets at system boot | |
1746 | * | |
1747 | * Description: Initialize top_cpuset and the cpuset internal file system, | |
1748 | **/ | |
1749 | ||
1750 | int __init cpuset_init(void) | |
1751 | { | |
8793d854 | 1752 | int err = 0; |
1da177e4 | 1753 | |
f9a86fcb MT |
1754 | cpus_setall(top_cpuset.cpus_allowed); |
1755 | nodes_setall(top_cpuset.mems_allowed); | |
1da177e4 | 1756 | |
3e0d98b9 | 1757 | fmeter_init(&top_cpuset.fmeter); |
151a4420 | 1758 | top_cpuset.mems_generation = cpuset_mems_generation++; |
029190c5 | 1759 | set_bit(CS_SCHED_LOAD_BALANCE, &top_cpuset.flags); |
1d3504fc | 1760 | top_cpuset.relax_domain_level = -1; |
1da177e4 | 1761 | |
1da177e4 LT |
1762 | err = register_filesystem(&cpuset_fs_type); |
1763 | if (err < 0) | |
8793d854 PM |
1764 | return err; |
1765 | ||
202f72d5 | 1766 | number_of_cpusets = 1; |
8793d854 | 1767 | return 0; |
1da177e4 LT |
1768 | } |
1769 | ||
956db3ca CW |
1770 | /** |
1771 | * cpuset_do_move_task - move a given task to another cpuset | |
1772 | * @tsk: pointer to task_struct the task to move | |
1773 | * @scan: struct cgroup_scanner contained in its struct cpuset_hotplug_scanner | |
1774 | * | |
1775 | * Called by cgroup_scan_tasks() for each task in a cgroup. | |
1776 | * Return nonzero to stop the walk through the tasks. | |
1777 | */ | |
9e0c914c AB |
1778 | static void cpuset_do_move_task(struct task_struct *tsk, |
1779 | struct cgroup_scanner *scan) | |
956db3ca CW |
1780 | { |
1781 | struct cpuset_hotplug_scanner *chsp; | |
1782 | ||
1783 | chsp = container_of(scan, struct cpuset_hotplug_scanner, scan); | |
1784 | cgroup_attach_task(chsp->to, tsk); | |
1785 | } | |
1786 | ||
1787 | /** | |
1788 | * move_member_tasks_to_cpuset - move tasks from one cpuset to another | |
1789 | * @from: cpuset in which the tasks currently reside | |
1790 | * @to: cpuset to which the tasks will be moved | |
1791 | * | |
c8d9c90c PJ |
1792 | * Called with cgroup_mutex held |
1793 | * callback_mutex must not be held, as cpuset_attach() will take it. | |
956db3ca CW |
1794 | * |
1795 | * The cgroup_scan_tasks() function will scan all the tasks in a cgroup, | |
1796 | * calling callback functions for each. | |
1797 | */ | |
1798 | static void move_member_tasks_to_cpuset(struct cpuset *from, struct cpuset *to) | |
1799 | { | |
1800 | struct cpuset_hotplug_scanner scan; | |
1801 | ||
1802 | scan.scan.cg = from->css.cgroup; | |
1803 | scan.scan.test_task = NULL; /* select all tasks in cgroup */ | |
1804 | scan.scan.process_task = cpuset_do_move_task; | |
1805 | scan.scan.heap = NULL; | |
1806 | scan.to = to->css.cgroup; | |
1807 | ||
da5ef6bb | 1808 | if (cgroup_scan_tasks(&scan.scan)) |
956db3ca CW |
1809 | printk(KERN_ERR "move_member_tasks_to_cpuset: " |
1810 | "cgroup_scan_tasks failed\n"); | |
1811 | } | |
1812 | ||
b1aac8bb PJ |
1813 | /* |
1814 | * If common_cpu_mem_hotplug_unplug(), below, unplugs any CPUs | |
1815 | * or memory nodes, we need to walk over the cpuset hierarchy, | |
1816 | * removing that CPU or node from all cpusets. If this removes the | |
956db3ca CW |
1817 | * last CPU or node from a cpuset, then move the tasks in the empty |
1818 | * cpuset to its next-highest non-empty parent. | |
b1aac8bb | 1819 | * |
c8d9c90c PJ |
1820 | * Called with cgroup_mutex held |
1821 | * callback_mutex must not be held, as cpuset_attach() will take it. | |
b1aac8bb | 1822 | */ |
956db3ca CW |
1823 | static void remove_tasks_in_empty_cpuset(struct cpuset *cs) |
1824 | { | |
1825 | struct cpuset *parent; | |
1826 | ||
c8d9c90c PJ |
1827 | /* |
1828 | * The cgroup's css_sets list is in use if there are tasks | |
1829 | * in the cpuset; the list is empty if there are none; | |
1830 | * the cs->css.refcnt seems always 0. | |
1831 | */ | |
956db3ca CW |
1832 | if (list_empty(&cs->css.cgroup->css_sets)) |
1833 | return; | |
b1aac8bb | 1834 | |
956db3ca CW |
1835 | /* |
1836 | * Find its next-highest non-empty parent, (top cpuset | |
1837 | * has online cpus, so can't be empty). | |
1838 | */ | |
1839 | parent = cs->parent; | |
b4501295 PJ |
1840 | while (cpus_empty(parent->cpus_allowed) || |
1841 | nodes_empty(parent->mems_allowed)) | |
956db3ca | 1842 | parent = parent->parent; |
956db3ca CW |
1843 | |
1844 | move_member_tasks_to_cpuset(cs, parent); | |
1845 | } | |
1846 | ||
1847 | /* | |
1848 | * Walk the specified cpuset subtree and look for empty cpusets. | |
1849 | * The tasks of such cpuset must be moved to a parent cpuset. | |
1850 | * | |
2df167a3 | 1851 | * Called with cgroup_mutex held. We take callback_mutex to modify |
956db3ca CW |
1852 | * cpus_allowed and mems_allowed. |
1853 | * | |
1854 | * This walk processes the tree from top to bottom, completing one layer | |
1855 | * before dropping down to the next. It always processes a node before | |
1856 | * any of its children. | |
1857 | * | |
1858 | * For now, since we lack memory hot unplug, we'll never see a cpuset | |
1859 | * that has tasks along with an empty 'mems'. But if we did see such | |
1860 | * a cpuset, we'd handle it just like we do if its 'cpus' was empty. | |
1861 | */ | |
1862 | static void scan_for_empty_cpusets(const struct cpuset *root) | |
b1aac8bb | 1863 | { |
8d1e6266 | 1864 | LIST_HEAD(queue); |
956db3ca CW |
1865 | struct cpuset *cp; /* scans cpusets being updated */ |
1866 | struct cpuset *child; /* scans child cpusets of cp */ | |
8793d854 | 1867 | struct cgroup *cont; |
f9b4fb8d | 1868 | nodemask_t oldmems; |
b1aac8bb | 1869 | |
956db3ca CW |
1870 | list_add_tail((struct list_head *)&root->stack_list, &queue); |
1871 | ||
956db3ca | 1872 | while (!list_empty(&queue)) { |
8d1e6266 | 1873 | cp = list_first_entry(&queue, struct cpuset, stack_list); |
956db3ca CW |
1874 | list_del(queue.next); |
1875 | list_for_each_entry(cont, &cp->css.cgroup->children, sibling) { | |
1876 | child = cgroup_cs(cont); | |
1877 | list_add_tail(&child->stack_list, &queue); | |
1878 | } | |
b4501295 PJ |
1879 | |
1880 | /* Continue past cpusets with all cpus, mems online */ | |
1881 | if (cpus_subset(cp->cpus_allowed, cpu_online_map) && | |
1882 | nodes_subset(cp->mems_allowed, node_states[N_HIGH_MEMORY])) | |
1883 | continue; | |
1884 | ||
f9b4fb8d MX |
1885 | oldmems = cp->mems_allowed; |
1886 | ||
956db3ca | 1887 | /* Remove offline cpus and mems from this cpuset. */ |
b4501295 | 1888 | mutex_lock(&callback_mutex); |
956db3ca CW |
1889 | cpus_and(cp->cpus_allowed, cp->cpus_allowed, cpu_online_map); |
1890 | nodes_and(cp->mems_allowed, cp->mems_allowed, | |
1891 | node_states[N_HIGH_MEMORY]); | |
b4501295 PJ |
1892 | mutex_unlock(&callback_mutex); |
1893 | ||
1894 | /* Move tasks from the empty cpuset to a parent */ | |
c8d9c90c | 1895 | if (cpus_empty(cp->cpus_allowed) || |
b4501295 | 1896 | nodes_empty(cp->mems_allowed)) |
956db3ca | 1897 | remove_tasks_in_empty_cpuset(cp); |
f9b4fb8d MX |
1898 | else { |
1899 | update_tasks_cpumask(cp); | |
1900 | update_tasks_nodemask(cp, &oldmems); | |
1901 | } | |
b1aac8bb PJ |
1902 | } |
1903 | } | |
1904 | ||
1905 | /* | |
1906 | * The cpus_allowed and mems_allowed nodemasks in the top_cpuset track | |
0e1e7c7a | 1907 | * cpu_online_map and node_states[N_HIGH_MEMORY]. Force the top cpuset to |
956db3ca | 1908 | * track what's online after any CPU or memory node hotplug or unplug event. |
b1aac8bb PJ |
1909 | * |
1910 | * Since there are two callers of this routine, one for CPU hotplug | |
1911 | * events and one for memory node hotplug events, we could have coded | |
1912 | * two separate routines here. We code it as a single common routine | |
1913 | * in order to minimize text size. | |
1914 | */ | |
1915 | ||
3e84050c | 1916 | static void common_cpu_mem_hotplug_unplug(int rebuild_sd) |
b1aac8bb | 1917 | { |
8793d854 | 1918 | cgroup_lock(); |
b1aac8bb | 1919 | |
b1aac8bb | 1920 | top_cpuset.cpus_allowed = cpu_online_map; |
0e1e7c7a | 1921 | top_cpuset.mems_allowed = node_states[N_HIGH_MEMORY]; |
956db3ca | 1922 | scan_for_empty_cpusets(&top_cpuset); |
b1aac8bb | 1923 | |
5c8e1ed1 MK |
1924 | /* |
1925 | * Scheduler destroys domains on hotplug events. | |
1926 | * Rebuild them based on the current settings. | |
1927 | */ | |
3e84050c DA |
1928 | if (rebuild_sd) |
1929 | rebuild_sched_domains(); | |
5c8e1ed1 | 1930 | |
8793d854 | 1931 | cgroup_unlock(); |
b1aac8bb | 1932 | } |
b1aac8bb | 1933 | |
4c4d50f7 PJ |
1934 | /* |
1935 | * The top_cpuset tracks what CPUs and Memory Nodes are online, | |
1936 | * period. This is necessary in order to make cpusets transparent | |
1937 | * (of no affect) on systems that are actively using CPU hotplug | |
1938 | * but making no active use of cpusets. | |
1939 | * | |
38837fc7 PJ |
1940 | * This routine ensures that top_cpuset.cpus_allowed tracks |
1941 | * cpu_online_map on each CPU hotplug (cpuhp) event. | |
4c4d50f7 PJ |
1942 | */ |
1943 | ||
029190c5 PJ |
1944 | static int cpuset_handle_cpuhp(struct notifier_block *unused_nb, |
1945 | unsigned long phase, void *unused_cpu) | |
4c4d50f7 | 1946 | { |
3e84050c DA |
1947 | switch (phase) { |
1948 | case CPU_UP_CANCELED: | |
1949 | case CPU_UP_CANCELED_FROZEN: | |
1950 | case CPU_DOWN_FAILED: | |
1951 | case CPU_DOWN_FAILED_FROZEN: | |
1952 | case CPU_ONLINE: | |
1953 | case CPU_ONLINE_FROZEN: | |
1954 | case CPU_DEAD: | |
1955 | case CPU_DEAD_FROZEN: | |
1956 | common_cpu_mem_hotplug_unplug(1); | |
1957 | break; | |
1958 | default: | |
ac076758 | 1959 | return NOTIFY_DONE; |
3e84050c | 1960 | } |
ac076758 | 1961 | |
3e84050c | 1962 | return NOTIFY_OK; |
4c4d50f7 | 1963 | } |
4c4d50f7 | 1964 | |
b1aac8bb | 1965 | #ifdef CONFIG_MEMORY_HOTPLUG |
38837fc7 | 1966 | /* |
0e1e7c7a CL |
1967 | * Keep top_cpuset.mems_allowed tracking node_states[N_HIGH_MEMORY]. |
1968 | * Call this routine anytime after you change | |
1969 | * node_states[N_HIGH_MEMORY]. | |
38837fc7 PJ |
1970 | * See also the previous routine cpuset_handle_cpuhp(). |
1971 | */ | |
1972 | ||
1af98928 | 1973 | void cpuset_track_online_nodes(void) |
38837fc7 | 1974 | { |
3e84050c | 1975 | common_cpu_mem_hotplug_unplug(0); |
38837fc7 PJ |
1976 | } |
1977 | #endif | |
1978 | ||
1da177e4 LT |
1979 | /** |
1980 | * cpuset_init_smp - initialize cpus_allowed | |
1981 | * | |
1982 | * Description: Finish top cpuset after cpu, node maps are initialized | |
1983 | **/ | |
1984 | ||
1985 | void __init cpuset_init_smp(void) | |
1986 | { | |
1987 | top_cpuset.cpus_allowed = cpu_online_map; | |
0e1e7c7a | 1988 | top_cpuset.mems_allowed = node_states[N_HIGH_MEMORY]; |
4c4d50f7 PJ |
1989 | |
1990 | hotcpu_notifier(cpuset_handle_cpuhp, 0); | |
1da177e4 LT |
1991 | } |
1992 | ||
1993 | /** | |
1da177e4 LT |
1994 | * cpuset_cpus_allowed - return cpus_allowed mask from a tasks cpuset. |
1995 | * @tsk: pointer to task_struct from which to obtain cpuset->cpus_allowed. | |
f9a86fcb | 1996 | * @pmask: pointer to cpumask_t variable to receive cpus_allowed set. |
1da177e4 LT |
1997 | * |
1998 | * Description: Returns the cpumask_t cpus_allowed of the cpuset | |
1999 | * attached to the specified @tsk. Guaranteed to return some non-empty | |
2000 | * subset of cpu_online_map, even if this means going outside the | |
2001 | * tasks cpuset. | |
2002 | **/ | |
2003 | ||
f9a86fcb | 2004 | void cpuset_cpus_allowed(struct task_struct *tsk, cpumask_t *pmask) |
1da177e4 | 2005 | { |
3d3f26a7 | 2006 | mutex_lock(&callback_mutex); |
f9a86fcb | 2007 | cpuset_cpus_allowed_locked(tsk, pmask); |
470fd646 | 2008 | mutex_unlock(&callback_mutex); |
470fd646 CW |
2009 | } |
2010 | ||
2011 | /** | |
2012 | * cpuset_cpus_allowed_locked - return cpus_allowed mask from a tasks cpuset. | |
2df167a3 | 2013 | * Must be called with callback_mutex held. |
470fd646 | 2014 | **/ |
f9a86fcb | 2015 | void cpuset_cpus_allowed_locked(struct task_struct *tsk, cpumask_t *pmask) |
470fd646 | 2016 | { |
909d75a3 | 2017 | task_lock(tsk); |
f9a86fcb | 2018 | guarantee_online_cpus(task_cs(tsk), pmask); |
909d75a3 | 2019 | task_unlock(tsk); |
1da177e4 LT |
2020 | } |
2021 | ||
2022 | void cpuset_init_current_mems_allowed(void) | |
2023 | { | |
f9a86fcb | 2024 | nodes_setall(current->mems_allowed); |
1da177e4 LT |
2025 | } |
2026 | ||
909d75a3 PJ |
2027 | /** |
2028 | * cpuset_mems_allowed - return mems_allowed mask from a tasks cpuset. | |
2029 | * @tsk: pointer to task_struct from which to obtain cpuset->mems_allowed. | |
2030 | * | |
2031 | * Description: Returns the nodemask_t mems_allowed of the cpuset | |
2032 | * attached to the specified @tsk. Guaranteed to return some non-empty | |
0e1e7c7a | 2033 | * subset of node_states[N_HIGH_MEMORY], even if this means going outside the |
909d75a3 PJ |
2034 | * tasks cpuset. |
2035 | **/ | |
2036 | ||
2037 | nodemask_t cpuset_mems_allowed(struct task_struct *tsk) | |
2038 | { | |
2039 | nodemask_t mask; | |
2040 | ||
3d3f26a7 | 2041 | mutex_lock(&callback_mutex); |
909d75a3 | 2042 | task_lock(tsk); |
8793d854 | 2043 | guarantee_online_mems(task_cs(tsk), &mask); |
909d75a3 | 2044 | task_unlock(tsk); |
3d3f26a7 | 2045 | mutex_unlock(&callback_mutex); |
909d75a3 PJ |
2046 | |
2047 | return mask; | |
2048 | } | |
2049 | ||
d9fd8a6d | 2050 | /** |
19770b32 MG |
2051 | * cpuset_nodemask_valid_mems_allowed - check nodemask vs. curremt mems_allowed |
2052 | * @nodemask: the nodemask to be checked | |
d9fd8a6d | 2053 | * |
19770b32 | 2054 | * Are any of the nodes in the nodemask allowed in current->mems_allowed? |
1da177e4 | 2055 | */ |
19770b32 | 2056 | int cpuset_nodemask_valid_mems_allowed(nodemask_t *nodemask) |
1da177e4 | 2057 | { |
19770b32 | 2058 | return nodes_intersects(*nodemask, current->mems_allowed); |
1da177e4 LT |
2059 | } |
2060 | ||
9bf2229f | 2061 | /* |
78608366 PM |
2062 | * nearest_hardwall_ancestor() - Returns the nearest mem_exclusive or |
2063 | * mem_hardwall ancestor to the specified cpuset. Call holding | |
2064 | * callback_mutex. If no ancestor is mem_exclusive or mem_hardwall | |
2065 | * (an unusual configuration), then returns the root cpuset. | |
9bf2229f | 2066 | */ |
78608366 | 2067 | static const struct cpuset *nearest_hardwall_ancestor(const struct cpuset *cs) |
9bf2229f | 2068 | { |
78608366 | 2069 | while (!(is_mem_exclusive(cs) || is_mem_hardwall(cs)) && cs->parent) |
9bf2229f PJ |
2070 | cs = cs->parent; |
2071 | return cs; | |
2072 | } | |
2073 | ||
d9fd8a6d | 2074 | /** |
02a0e53d | 2075 | * cpuset_zone_allowed_softwall - Can we allocate on zone z's memory node? |
9bf2229f | 2076 | * @z: is this zone on an allowed node? |
02a0e53d | 2077 | * @gfp_mask: memory allocation flags |
d9fd8a6d | 2078 | * |
02a0e53d PJ |
2079 | * If we're in interrupt, yes, we can always allocate. If |
2080 | * __GFP_THISNODE is set, yes, we can always allocate. If zone | |
9bf2229f PJ |
2081 | * z's node is in our tasks mems_allowed, yes. If it's not a |
2082 | * __GFP_HARDWALL request and this zone's nodes is in the nearest | |
78608366 | 2083 | * hardwalled cpuset ancestor to this tasks cpuset, yes. |
c596d9f3 DR |
2084 | * If the task has been OOM killed and has access to memory reserves |
2085 | * as specified by the TIF_MEMDIE flag, yes. | |
9bf2229f PJ |
2086 | * Otherwise, no. |
2087 | * | |
02a0e53d PJ |
2088 | * If __GFP_HARDWALL is set, cpuset_zone_allowed_softwall() |
2089 | * reduces to cpuset_zone_allowed_hardwall(). Otherwise, | |
2090 | * cpuset_zone_allowed_softwall() might sleep, and might allow a zone | |
2091 | * from an enclosing cpuset. | |
2092 | * | |
2093 | * cpuset_zone_allowed_hardwall() only handles the simpler case of | |
2094 | * hardwall cpusets, and never sleeps. | |
2095 | * | |
2096 | * The __GFP_THISNODE placement logic is really handled elsewhere, | |
2097 | * by forcibly using a zonelist starting at a specified node, and by | |
2098 | * (in get_page_from_freelist()) refusing to consider the zones for | |
2099 | * any node on the zonelist except the first. By the time any such | |
2100 | * calls get to this routine, we should just shut up and say 'yes'. | |
2101 | * | |
9bf2229f | 2102 | * GFP_USER allocations are marked with the __GFP_HARDWALL bit, |
c596d9f3 DR |
2103 | * and do not allow allocations outside the current tasks cpuset |
2104 | * unless the task has been OOM killed as is marked TIF_MEMDIE. | |
9bf2229f | 2105 | * GFP_KERNEL allocations are not so marked, so can escape to the |
78608366 | 2106 | * nearest enclosing hardwalled ancestor cpuset. |
9bf2229f | 2107 | * |
02a0e53d PJ |
2108 | * Scanning up parent cpusets requires callback_mutex. The |
2109 | * __alloc_pages() routine only calls here with __GFP_HARDWALL bit | |
2110 | * _not_ set if it's a GFP_KERNEL allocation, and all nodes in the | |
2111 | * current tasks mems_allowed came up empty on the first pass over | |
2112 | * the zonelist. So only GFP_KERNEL allocations, if all nodes in the | |
2113 | * cpuset are short of memory, might require taking the callback_mutex | |
2114 | * mutex. | |
9bf2229f | 2115 | * |
36be57ff | 2116 | * The first call here from mm/page_alloc:get_page_from_freelist() |
02a0e53d PJ |
2117 | * has __GFP_HARDWALL set in gfp_mask, enforcing hardwall cpusets, |
2118 | * so no allocation on a node outside the cpuset is allowed (unless | |
2119 | * in interrupt, of course). | |
36be57ff PJ |
2120 | * |
2121 | * The second pass through get_page_from_freelist() doesn't even call | |
2122 | * here for GFP_ATOMIC calls. For those calls, the __alloc_pages() | |
2123 | * variable 'wait' is not set, and the bit ALLOC_CPUSET is not set | |
2124 | * in alloc_flags. That logic and the checks below have the combined | |
2125 | * affect that: | |
9bf2229f PJ |
2126 | * in_interrupt - any node ok (current task context irrelevant) |
2127 | * GFP_ATOMIC - any node ok | |
c596d9f3 | 2128 | * TIF_MEMDIE - any node ok |
78608366 | 2129 | * GFP_KERNEL - any node in enclosing hardwalled cpuset ok |
9bf2229f | 2130 | * GFP_USER - only nodes in current tasks mems allowed ok. |
36be57ff PJ |
2131 | * |
2132 | * Rule: | |
02a0e53d | 2133 | * Don't call cpuset_zone_allowed_softwall if you can't sleep, unless you |
36be57ff PJ |
2134 | * pass in the __GFP_HARDWALL flag set in gfp_flag, which disables |
2135 | * the code that might scan up ancestor cpusets and sleep. | |
02a0e53d | 2136 | */ |
9bf2229f | 2137 | |
02a0e53d | 2138 | int __cpuset_zone_allowed_softwall(struct zone *z, gfp_t gfp_mask) |
1da177e4 | 2139 | { |
9bf2229f PJ |
2140 | int node; /* node that zone z is on */ |
2141 | const struct cpuset *cs; /* current cpuset ancestors */ | |
29afd49b | 2142 | int allowed; /* is allocation in zone z allowed? */ |
9bf2229f | 2143 | |
9b819d20 | 2144 | if (in_interrupt() || (gfp_mask & __GFP_THISNODE)) |
9bf2229f | 2145 | return 1; |
89fa3024 | 2146 | node = zone_to_nid(z); |
92d1dbd2 | 2147 | might_sleep_if(!(gfp_mask & __GFP_HARDWALL)); |
9bf2229f PJ |
2148 | if (node_isset(node, current->mems_allowed)) |
2149 | return 1; | |
c596d9f3 DR |
2150 | /* |
2151 | * Allow tasks that have access to memory reserves because they have | |
2152 | * been OOM killed to get memory anywhere. | |
2153 | */ | |
2154 | if (unlikely(test_thread_flag(TIF_MEMDIE))) | |
2155 | return 1; | |
9bf2229f PJ |
2156 | if (gfp_mask & __GFP_HARDWALL) /* If hardwall request, stop here */ |
2157 | return 0; | |
2158 | ||
5563e770 BP |
2159 | if (current->flags & PF_EXITING) /* Let dying task have memory */ |
2160 | return 1; | |
2161 | ||
9bf2229f | 2162 | /* Not hardwall and node outside mems_allowed: scan up cpusets */ |
3d3f26a7 | 2163 | mutex_lock(&callback_mutex); |
053199ed | 2164 | |
053199ed | 2165 | task_lock(current); |
78608366 | 2166 | cs = nearest_hardwall_ancestor(task_cs(current)); |
053199ed PJ |
2167 | task_unlock(current); |
2168 | ||
9bf2229f | 2169 | allowed = node_isset(node, cs->mems_allowed); |
3d3f26a7 | 2170 | mutex_unlock(&callback_mutex); |
9bf2229f | 2171 | return allowed; |
1da177e4 LT |
2172 | } |
2173 | ||
02a0e53d PJ |
2174 | /* |
2175 | * cpuset_zone_allowed_hardwall - Can we allocate on zone z's memory node? | |
2176 | * @z: is this zone on an allowed node? | |
2177 | * @gfp_mask: memory allocation flags | |
2178 | * | |
2179 | * If we're in interrupt, yes, we can always allocate. | |
2180 | * If __GFP_THISNODE is set, yes, we can always allocate. If zone | |
c596d9f3 DR |
2181 | * z's node is in our tasks mems_allowed, yes. If the task has been |
2182 | * OOM killed and has access to memory reserves as specified by the | |
2183 | * TIF_MEMDIE flag, yes. Otherwise, no. | |
02a0e53d PJ |
2184 | * |
2185 | * The __GFP_THISNODE placement logic is really handled elsewhere, | |
2186 | * by forcibly using a zonelist starting at a specified node, and by | |
2187 | * (in get_page_from_freelist()) refusing to consider the zones for | |
2188 | * any node on the zonelist except the first. By the time any such | |
2189 | * calls get to this routine, we should just shut up and say 'yes'. | |
2190 | * | |
2191 | * Unlike the cpuset_zone_allowed_softwall() variant, above, | |
2192 | * this variant requires that the zone be in the current tasks | |
2193 | * mems_allowed or that we're in interrupt. It does not scan up the | |
2194 | * cpuset hierarchy for the nearest enclosing mem_exclusive cpuset. | |
2195 | * It never sleeps. | |
2196 | */ | |
2197 | ||
2198 | int __cpuset_zone_allowed_hardwall(struct zone *z, gfp_t gfp_mask) | |
2199 | { | |
2200 | int node; /* node that zone z is on */ | |
2201 | ||
2202 | if (in_interrupt() || (gfp_mask & __GFP_THISNODE)) | |
2203 | return 1; | |
2204 | node = zone_to_nid(z); | |
2205 | if (node_isset(node, current->mems_allowed)) | |
2206 | return 1; | |
dedf8b79 DW |
2207 | /* |
2208 | * Allow tasks that have access to memory reserves because they have | |
2209 | * been OOM killed to get memory anywhere. | |
2210 | */ | |
2211 | if (unlikely(test_thread_flag(TIF_MEMDIE))) | |
2212 | return 1; | |
02a0e53d PJ |
2213 | return 0; |
2214 | } | |
2215 | ||
505970b9 PJ |
2216 | /** |
2217 | * cpuset_lock - lock out any changes to cpuset structures | |
2218 | * | |
3d3f26a7 | 2219 | * The out of memory (oom) code needs to mutex_lock cpusets |
505970b9 | 2220 | * from being changed while it scans the tasklist looking for a |
3d3f26a7 | 2221 | * task in an overlapping cpuset. Expose callback_mutex via this |
505970b9 PJ |
2222 | * cpuset_lock() routine, so the oom code can lock it, before |
2223 | * locking the task list. The tasklist_lock is a spinlock, so | |
3d3f26a7 | 2224 | * must be taken inside callback_mutex. |
505970b9 PJ |
2225 | */ |
2226 | ||
2227 | void cpuset_lock(void) | |
2228 | { | |
3d3f26a7 | 2229 | mutex_lock(&callback_mutex); |
505970b9 PJ |
2230 | } |
2231 | ||
2232 | /** | |
2233 | * cpuset_unlock - release lock on cpuset changes | |
2234 | * | |
2235 | * Undo the lock taken in a previous cpuset_lock() call. | |
2236 | */ | |
2237 | ||
2238 | void cpuset_unlock(void) | |
2239 | { | |
3d3f26a7 | 2240 | mutex_unlock(&callback_mutex); |
505970b9 PJ |
2241 | } |
2242 | ||
825a46af PJ |
2243 | /** |
2244 | * cpuset_mem_spread_node() - On which node to begin search for a page | |
2245 | * | |
2246 | * If a task is marked PF_SPREAD_PAGE or PF_SPREAD_SLAB (as for | |
2247 | * tasks in a cpuset with is_spread_page or is_spread_slab set), | |
2248 | * and if the memory allocation used cpuset_mem_spread_node() | |
2249 | * to determine on which node to start looking, as it will for | |
2250 | * certain page cache or slab cache pages such as used for file | |
2251 | * system buffers and inode caches, then instead of starting on the | |
2252 | * local node to look for a free page, rather spread the starting | |
2253 | * node around the tasks mems_allowed nodes. | |
2254 | * | |
2255 | * We don't have to worry about the returned node being offline | |
2256 | * because "it can't happen", and even if it did, it would be ok. | |
2257 | * | |
2258 | * The routines calling guarantee_online_mems() are careful to | |
2259 | * only set nodes in task->mems_allowed that are online. So it | |
2260 | * should not be possible for the following code to return an | |
2261 | * offline node. But if it did, that would be ok, as this routine | |
2262 | * is not returning the node where the allocation must be, only | |
2263 | * the node where the search should start. The zonelist passed to | |
2264 | * __alloc_pages() will include all nodes. If the slab allocator | |
2265 | * is passed an offline node, it will fall back to the local node. | |
2266 | * See kmem_cache_alloc_node(). | |
2267 | */ | |
2268 | ||
2269 | int cpuset_mem_spread_node(void) | |
2270 | { | |
2271 | int node; | |
2272 | ||
2273 | node = next_node(current->cpuset_mem_spread_rotor, current->mems_allowed); | |
2274 | if (node == MAX_NUMNODES) | |
2275 | node = first_node(current->mems_allowed); | |
2276 | current->cpuset_mem_spread_rotor = node; | |
2277 | return node; | |
2278 | } | |
2279 | EXPORT_SYMBOL_GPL(cpuset_mem_spread_node); | |
2280 | ||
ef08e3b4 | 2281 | /** |
bbe373f2 DR |
2282 | * cpuset_mems_allowed_intersects - Does @tsk1's mems_allowed intersect @tsk2's? |
2283 | * @tsk1: pointer to task_struct of some task. | |
2284 | * @tsk2: pointer to task_struct of some other task. | |
2285 | * | |
2286 | * Description: Return true if @tsk1's mems_allowed intersects the | |
2287 | * mems_allowed of @tsk2. Used by the OOM killer to determine if | |
2288 | * one of the task's memory usage might impact the memory available | |
2289 | * to the other. | |
ef08e3b4 PJ |
2290 | **/ |
2291 | ||
bbe373f2 DR |
2292 | int cpuset_mems_allowed_intersects(const struct task_struct *tsk1, |
2293 | const struct task_struct *tsk2) | |
ef08e3b4 | 2294 | { |
bbe373f2 | 2295 | return nodes_intersects(tsk1->mems_allowed, tsk2->mems_allowed); |
ef08e3b4 PJ |
2296 | } |
2297 | ||
3e0d98b9 PJ |
2298 | /* |
2299 | * Collection of memory_pressure is suppressed unless | |
2300 | * this flag is enabled by writing "1" to the special | |
2301 | * cpuset file 'memory_pressure_enabled' in the root cpuset. | |
2302 | */ | |
2303 | ||
c5b2aff8 | 2304 | int cpuset_memory_pressure_enabled __read_mostly; |
3e0d98b9 PJ |
2305 | |
2306 | /** | |
2307 | * cpuset_memory_pressure_bump - keep stats of per-cpuset reclaims. | |
2308 | * | |
2309 | * Keep a running average of the rate of synchronous (direct) | |
2310 | * page reclaim efforts initiated by tasks in each cpuset. | |
2311 | * | |
2312 | * This represents the rate at which some task in the cpuset | |
2313 | * ran low on memory on all nodes it was allowed to use, and | |
2314 | * had to enter the kernels page reclaim code in an effort to | |
2315 | * create more free memory by tossing clean pages or swapping | |
2316 | * or writing dirty pages. | |
2317 | * | |
2318 | * Display to user space in the per-cpuset read-only file | |
2319 | * "memory_pressure". Value displayed is an integer | |
2320 | * representing the recent rate of entry into the synchronous | |
2321 | * (direct) page reclaim by any task attached to the cpuset. | |
2322 | **/ | |
2323 | ||
2324 | void __cpuset_memory_pressure_bump(void) | |
2325 | { | |
3e0d98b9 | 2326 | task_lock(current); |
8793d854 | 2327 | fmeter_markevent(&task_cs(current)->fmeter); |
3e0d98b9 PJ |
2328 | task_unlock(current); |
2329 | } | |
2330 | ||
8793d854 | 2331 | #ifdef CONFIG_PROC_PID_CPUSET |
1da177e4 LT |
2332 | /* |
2333 | * proc_cpuset_show() | |
2334 | * - Print tasks cpuset path into seq_file. | |
2335 | * - Used for /proc/<pid>/cpuset. | |
053199ed PJ |
2336 | * - No need to task_lock(tsk) on this tsk->cpuset reference, as it |
2337 | * doesn't really matter if tsk->cpuset changes after we read it, | |
c8d9c90c | 2338 | * and we take cgroup_mutex, keeping cpuset_attach() from changing it |
2df167a3 | 2339 | * anyway. |
1da177e4 | 2340 | */ |
029190c5 | 2341 | static int proc_cpuset_show(struct seq_file *m, void *unused_v) |
1da177e4 | 2342 | { |
13b41b09 | 2343 | struct pid *pid; |
1da177e4 LT |
2344 | struct task_struct *tsk; |
2345 | char *buf; | |
8793d854 | 2346 | struct cgroup_subsys_state *css; |
99f89551 | 2347 | int retval; |
1da177e4 | 2348 | |
99f89551 | 2349 | retval = -ENOMEM; |
1da177e4 LT |
2350 | buf = kmalloc(PAGE_SIZE, GFP_KERNEL); |
2351 | if (!buf) | |
99f89551 EB |
2352 | goto out; |
2353 | ||
2354 | retval = -ESRCH; | |
13b41b09 EB |
2355 | pid = m->private; |
2356 | tsk = get_pid_task(pid, PIDTYPE_PID); | |
99f89551 EB |
2357 | if (!tsk) |
2358 | goto out_free; | |
1da177e4 | 2359 | |
99f89551 | 2360 | retval = -EINVAL; |
8793d854 PM |
2361 | cgroup_lock(); |
2362 | css = task_subsys_state(tsk, cpuset_subsys_id); | |
2363 | retval = cgroup_path(css->cgroup, buf, PAGE_SIZE); | |
1da177e4 | 2364 | if (retval < 0) |
99f89551 | 2365 | goto out_unlock; |
1da177e4 LT |
2366 | seq_puts(m, buf); |
2367 | seq_putc(m, '\n'); | |
99f89551 | 2368 | out_unlock: |
8793d854 | 2369 | cgroup_unlock(); |
99f89551 EB |
2370 | put_task_struct(tsk); |
2371 | out_free: | |
1da177e4 | 2372 | kfree(buf); |
99f89551 | 2373 | out: |
1da177e4 LT |
2374 | return retval; |
2375 | } | |
2376 | ||
2377 | static int cpuset_open(struct inode *inode, struct file *file) | |
2378 | { | |
13b41b09 EB |
2379 | struct pid *pid = PROC_I(inode)->pid; |
2380 | return single_open(file, proc_cpuset_show, pid); | |
1da177e4 LT |
2381 | } |
2382 | ||
9a32144e | 2383 | const struct file_operations proc_cpuset_operations = { |
1da177e4 LT |
2384 | .open = cpuset_open, |
2385 | .read = seq_read, | |
2386 | .llseek = seq_lseek, | |
2387 | .release = single_release, | |
2388 | }; | |
8793d854 | 2389 | #endif /* CONFIG_PROC_PID_CPUSET */ |
1da177e4 LT |
2390 | |
2391 | /* Display task cpus_allowed, mems_allowed in /proc/<pid>/status file. */ | |
df5f8314 EB |
2392 | void cpuset_task_status_allowed(struct seq_file *m, struct task_struct *task) |
2393 | { | |
2394 | seq_printf(m, "Cpus_allowed:\t"); | |
2395 | m->count += cpumask_scnprintf(m->buf + m->count, m->size - m->count, | |
2396 | task->cpus_allowed); | |
2397 | seq_printf(m, "\n"); | |
39106dcf MT |
2398 | seq_printf(m, "Cpus_allowed_list:\t"); |
2399 | m->count += cpulist_scnprintf(m->buf + m->count, m->size - m->count, | |
2400 | task->cpus_allowed); | |
2401 | seq_printf(m, "\n"); | |
df5f8314 EB |
2402 | seq_printf(m, "Mems_allowed:\t"); |
2403 | m->count += nodemask_scnprintf(m->buf + m->count, m->size - m->count, | |
2404 | task->mems_allowed); | |
2405 | seq_printf(m, "\n"); | |
39106dcf MT |
2406 | seq_printf(m, "Mems_allowed_list:\t"); |
2407 | m->count += nodelist_scnprintf(m->buf + m->count, m->size - m->count, | |
2408 | task->mems_allowed); | |
2409 | seq_printf(m, "\n"); | |
1da177e4 | 2410 | } |