]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * kernel/cpuset.c | |
3 | * | |
4 | * Processor and Memory placement constraints for sets of tasks. | |
5 | * | |
6 | * Copyright (C) 2003 BULL SA. | |
029190c5 | 7 | * Copyright (C) 2004-2007 Silicon Graphics, Inc. |
8793d854 | 8 | * Copyright (C) 2006 Google, Inc |
1da177e4 LT |
9 | * |
10 | * Portions derived from Patrick Mochel's sysfs code. | |
11 | * sysfs is Copyright (c) 2001-3 Patrick Mochel | |
1da177e4 | 12 | * |
825a46af | 13 | * 2003-10-10 Written by Simon Derr. |
1da177e4 | 14 | * 2003-10-22 Updates by Stephen Hemminger. |
825a46af | 15 | * 2004 May-July Rework by Paul Jackson. |
8793d854 | 16 | * 2006 Rework by Paul Menage to use generic cgroups |
cf417141 MK |
17 | * 2008 Rework of the scheduler domains and CPU hotplug handling |
18 | * by Max Krasnyansky | |
1da177e4 LT |
19 | * |
20 | * This file is subject to the terms and conditions of the GNU General Public | |
21 | * License. See the file COPYING in the main directory of the Linux | |
22 | * distribution for more details. | |
23 | */ | |
24 | ||
1da177e4 LT |
25 | #include <linux/cpu.h> |
26 | #include <linux/cpumask.h> | |
27 | #include <linux/cpuset.h> | |
28 | #include <linux/err.h> | |
29 | #include <linux/errno.h> | |
30 | #include <linux/file.h> | |
31 | #include <linux/fs.h> | |
32 | #include <linux/init.h> | |
33 | #include <linux/interrupt.h> | |
34 | #include <linux/kernel.h> | |
35 | #include <linux/kmod.h> | |
36 | #include <linux/list.h> | |
68860ec1 | 37 | #include <linux/mempolicy.h> |
1da177e4 | 38 | #include <linux/mm.h> |
f481891f | 39 | #include <linux/memory.h> |
1da177e4 LT |
40 | #include <linux/module.h> |
41 | #include <linux/mount.h> | |
42 | #include <linux/namei.h> | |
43 | #include <linux/pagemap.h> | |
44 | #include <linux/proc_fs.h> | |
6b9c2603 | 45 | #include <linux/rcupdate.h> |
1da177e4 LT |
46 | #include <linux/sched.h> |
47 | #include <linux/seq_file.h> | |
22fb52dd | 48 | #include <linux/security.h> |
1da177e4 | 49 | #include <linux/slab.h> |
1da177e4 LT |
50 | #include <linux/spinlock.h> |
51 | #include <linux/stat.h> | |
52 | #include <linux/string.h> | |
53 | #include <linux/time.h> | |
54 | #include <linux/backing-dev.h> | |
55 | #include <linux/sort.h> | |
56 | ||
57 | #include <asm/uaccess.h> | |
58 | #include <asm/atomic.h> | |
3d3f26a7 | 59 | #include <linux/mutex.h> |
956db3ca CW |
60 | #include <linux/workqueue.h> |
61 | #include <linux/cgroup.h> | |
1da177e4 | 62 | |
202f72d5 PJ |
63 | /* |
64 | * Tracks how many cpusets are currently defined in system. | |
65 | * When there is only one cpuset (the root cpuset) we can | |
66 | * short circuit some hooks. | |
67 | */ | |
7edc5962 | 68 | int number_of_cpusets __read_mostly; |
202f72d5 | 69 | |
2df167a3 | 70 | /* Forward declare cgroup structures */ |
8793d854 PM |
71 | struct cgroup_subsys cpuset_subsys; |
72 | struct cpuset; | |
73 | ||
3e0d98b9 PJ |
74 | /* See "Frequency meter" comments, below. */ |
75 | ||
76 | struct fmeter { | |
77 | int cnt; /* unprocessed events count */ | |
78 | int val; /* most recent output value */ | |
79 | time_t time; /* clock (secs) when val computed */ | |
80 | spinlock_t lock; /* guards read or write of above */ | |
81 | }; | |
82 | ||
1da177e4 | 83 | struct cpuset { |
8793d854 PM |
84 | struct cgroup_subsys_state css; |
85 | ||
1da177e4 LT |
86 | unsigned long flags; /* "unsigned long" so bitops work */ |
87 | cpumask_t cpus_allowed; /* CPUs allowed to tasks in cpuset */ | |
88 | nodemask_t mems_allowed; /* Memory Nodes allowed to tasks */ | |
89 | ||
1da177e4 | 90 | struct cpuset *parent; /* my parent */ |
1da177e4 LT |
91 | |
92 | /* | |
93 | * Copy of global cpuset_mems_generation as of the most | |
94 | * recent time this cpuset changed its mems_allowed. | |
95 | */ | |
3e0d98b9 PJ |
96 | int mems_generation; |
97 | ||
98 | struct fmeter fmeter; /* memory_pressure filter */ | |
029190c5 PJ |
99 | |
100 | /* partition number for rebuild_sched_domains() */ | |
101 | int pn; | |
956db3ca | 102 | |
1d3504fc HS |
103 | /* for custom sched domain */ |
104 | int relax_domain_level; | |
105 | ||
956db3ca CW |
106 | /* used for walking a cpuset heirarchy */ |
107 | struct list_head stack_list; | |
1da177e4 LT |
108 | }; |
109 | ||
8793d854 PM |
110 | /* Retrieve the cpuset for a cgroup */ |
111 | static inline struct cpuset *cgroup_cs(struct cgroup *cont) | |
112 | { | |
113 | return container_of(cgroup_subsys_state(cont, cpuset_subsys_id), | |
114 | struct cpuset, css); | |
115 | } | |
116 | ||
117 | /* Retrieve the cpuset for a task */ | |
118 | static inline struct cpuset *task_cs(struct task_struct *task) | |
119 | { | |
120 | return container_of(task_subsys_state(task, cpuset_subsys_id), | |
121 | struct cpuset, css); | |
122 | } | |
956db3ca CW |
123 | struct cpuset_hotplug_scanner { |
124 | struct cgroup_scanner scan; | |
125 | struct cgroup *to; | |
126 | }; | |
8793d854 | 127 | |
1da177e4 LT |
128 | /* bits in struct cpuset flags field */ |
129 | typedef enum { | |
130 | CS_CPU_EXCLUSIVE, | |
131 | CS_MEM_EXCLUSIVE, | |
78608366 | 132 | CS_MEM_HARDWALL, |
45b07ef3 | 133 | CS_MEMORY_MIGRATE, |
029190c5 | 134 | CS_SCHED_LOAD_BALANCE, |
825a46af PJ |
135 | CS_SPREAD_PAGE, |
136 | CS_SPREAD_SLAB, | |
1da177e4 LT |
137 | } cpuset_flagbits_t; |
138 | ||
139 | /* convenient tests for these bits */ | |
140 | static inline int is_cpu_exclusive(const struct cpuset *cs) | |
141 | { | |
7b5b9ef0 | 142 | return test_bit(CS_CPU_EXCLUSIVE, &cs->flags); |
1da177e4 LT |
143 | } |
144 | ||
145 | static inline int is_mem_exclusive(const struct cpuset *cs) | |
146 | { | |
7b5b9ef0 | 147 | return test_bit(CS_MEM_EXCLUSIVE, &cs->flags); |
1da177e4 LT |
148 | } |
149 | ||
78608366 PM |
150 | static inline int is_mem_hardwall(const struct cpuset *cs) |
151 | { | |
152 | return test_bit(CS_MEM_HARDWALL, &cs->flags); | |
153 | } | |
154 | ||
029190c5 PJ |
155 | static inline int is_sched_load_balance(const struct cpuset *cs) |
156 | { | |
157 | return test_bit(CS_SCHED_LOAD_BALANCE, &cs->flags); | |
158 | } | |
159 | ||
45b07ef3 PJ |
160 | static inline int is_memory_migrate(const struct cpuset *cs) |
161 | { | |
7b5b9ef0 | 162 | return test_bit(CS_MEMORY_MIGRATE, &cs->flags); |
45b07ef3 PJ |
163 | } |
164 | ||
825a46af PJ |
165 | static inline int is_spread_page(const struct cpuset *cs) |
166 | { | |
167 | return test_bit(CS_SPREAD_PAGE, &cs->flags); | |
168 | } | |
169 | ||
170 | static inline int is_spread_slab(const struct cpuset *cs) | |
171 | { | |
172 | return test_bit(CS_SPREAD_SLAB, &cs->flags); | |
173 | } | |
174 | ||
1da177e4 | 175 | /* |
151a4420 | 176 | * Increment this integer everytime any cpuset changes its |
1da177e4 LT |
177 | * mems_allowed value. Users of cpusets can track this generation |
178 | * number, and avoid having to lock and reload mems_allowed unless | |
179 | * the cpuset they're using changes generation. | |
180 | * | |
2df167a3 | 181 | * A single, global generation is needed because cpuset_attach_task() could |
1da177e4 LT |
182 | * reattach a task to a different cpuset, which must not have its |
183 | * generation numbers aliased with those of that tasks previous cpuset. | |
184 | * | |
185 | * Generations are needed for mems_allowed because one task cannot | |
2df167a3 | 186 | * modify another's memory placement. So we must enable every task, |
1da177e4 LT |
187 | * on every visit to __alloc_pages(), to efficiently check whether |
188 | * its current->cpuset->mems_allowed has changed, requiring an update | |
189 | * of its current->mems_allowed. | |
151a4420 | 190 | * |
2df167a3 | 191 | * Since writes to cpuset_mems_generation are guarded by the cgroup lock |
151a4420 | 192 | * there is no need to mark it atomic. |
1da177e4 | 193 | */ |
151a4420 | 194 | static int cpuset_mems_generation; |
1da177e4 LT |
195 | |
196 | static struct cpuset top_cpuset = { | |
197 | .flags = ((1 << CS_CPU_EXCLUSIVE) | (1 << CS_MEM_EXCLUSIVE)), | |
198 | .cpus_allowed = CPU_MASK_ALL, | |
199 | .mems_allowed = NODE_MASK_ALL, | |
1da177e4 LT |
200 | }; |
201 | ||
1da177e4 | 202 | /* |
2df167a3 PM |
203 | * There are two global mutexes guarding cpuset structures. The first |
204 | * is the main control groups cgroup_mutex, accessed via | |
205 | * cgroup_lock()/cgroup_unlock(). The second is the cpuset-specific | |
206 | * callback_mutex, below. They can nest. It is ok to first take | |
207 | * cgroup_mutex, then nest callback_mutex. We also require taking | |
208 | * task_lock() when dereferencing a task's cpuset pointer. See "The | |
209 | * task_lock() exception", at the end of this comment. | |
053199ed | 210 | * |
3d3f26a7 | 211 | * A task must hold both mutexes to modify cpusets. If a task |
2df167a3 | 212 | * holds cgroup_mutex, then it blocks others wanting that mutex, |
3d3f26a7 | 213 | * ensuring that it is the only task able to also acquire callback_mutex |
053199ed PJ |
214 | * and be able to modify cpusets. It can perform various checks on |
215 | * the cpuset structure first, knowing nothing will change. It can | |
2df167a3 | 216 | * also allocate memory while just holding cgroup_mutex. While it is |
053199ed | 217 | * performing these checks, various callback routines can briefly |
3d3f26a7 IM |
218 | * acquire callback_mutex to query cpusets. Once it is ready to make |
219 | * the changes, it takes callback_mutex, blocking everyone else. | |
053199ed PJ |
220 | * |
221 | * Calls to the kernel memory allocator can not be made while holding | |
3d3f26a7 | 222 | * callback_mutex, as that would risk double tripping on callback_mutex |
053199ed PJ |
223 | * from one of the callbacks into the cpuset code from within |
224 | * __alloc_pages(). | |
225 | * | |
3d3f26a7 | 226 | * If a task is only holding callback_mutex, then it has read-only |
053199ed PJ |
227 | * access to cpusets. |
228 | * | |
229 | * The task_struct fields mems_allowed and mems_generation may only | |
230 | * be accessed in the context of that task, so require no locks. | |
231 | * | |
3d3f26a7 | 232 | * The cpuset_common_file_read() handlers only hold callback_mutex across |
053199ed PJ |
233 | * small pieces of code, such as when reading out possibly multi-word |
234 | * cpumasks and nodemasks. | |
235 | * | |
2df167a3 PM |
236 | * Accessing a task's cpuset should be done in accordance with the |
237 | * guidelines for accessing subsystem state in kernel/cgroup.c | |
1da177e4 LT |
238 | */ |
239 | ||
3d3f26a7 | 240 | static DEFINE_MUTEX(callback_mutex); |
4247bdc6 | 241 | |
75aa1994 DR |
242 | /* |
243 | * cpuset_buffer_lock protects both the cpuset_name and cpuset_nodelist | |
244 | * buffers. They are statically allocated to prevent using excess stack | |
245 | * when calling cpuset_print_task_mems_allowed(). | |
246 | */ | |
247 | #define CPUSET_NAME_LEN (128) | |
248 | #define CPUSET_NODELIST_LEN (256) | |
249 | static char cpuset_name[CPUSET_NAME_LEN]; | |
250 | static char cpuset_nodelist[CPUSET_NODELIST_LEN]; | |
251 | static DEFINE_SPINLOCK(cpuset_buffer_lock); | |
252 | ||
cf417141 MK |
253 | /* |
254 | * This is ugly, but preserves the userspace API for existing cpuset | |
8793d854 | 255 | * users. If someone tries to mount the "cpuset" filesystem, we |
cf417141 MK |
256 | * silently switch it to mount "cgroup" instead |
257 | */ | |
454e2398 DH |
258 | static int cpuset_get_sb(struct file_system_type *fs_type, |
259 | int flags, const char *unused_dev_name, | |
260 | void *data, struct vfsmount *mnt) | |
1da177e4 | 261 | { |
8793d854 PM |
262 | struct file_system_type *cgroup_fs = get_fs_type("cgroup"); |
263 | int ret = -ENODEV; | |
264 | if (cgroup_fs) { | |
265 | char mountopts[] = | |
266 | "cpuset,noprefix," | |
267 | "release_agent=/sbin/cpuset_release_agent"; | |
268 | ret = cgroup_fs->get_sb(cgroup_fs, flags, | |
269 | unused_dev_name, mountopts, mnt); | |
270 | put_filesystem(cgroup_fs); | |
271 | } | |
272 | return ret; | |
1da177e4 LT |
273 | } |
274 | ||
275 | static struct file_system_type cpuset_fs_type = { | |
276 | .name = "cpuset", | |
277 | .get_sb = cpuset_get_sb, | |
1da177e4 LT |
278 | }; |
279 | ||
1da177e4 LT |
280 | /* |
281 | * Return in *pmask the portion of a cpusets's cpus_allowed that | |
282 | * are online. If none are online, walk up the cpuset hierarchy | |
283 | * until we find one that does have some online cpus. If we get | |
284 | * all the way to the top and still haven't found any online cpus, | |
285 | * return cpu_online_map. Or if passed a NULL cs from an exit'ing | |
286 | * task, return cpu_online_map. | |
287 | * | |
288 | * One way or another, we guarantee to return some non-empty subset | |
289 | * of cpu_online_map. | |
290 | * | |
3d3f26a7 | 291 | * Call with callback_mutex held. |
1da177e4 LT |
292 | */ |
293 | ||
294 | static void guarantee_online_cpus(const struct cpuset *cs, cpumask_t *pmask) | |
295 | { | |
296 | while (cs && !cpus_intersects(cs->cpus_allowed, cpu_online_map)) | |
297 | cs = cs->parent; | |
298 | if (cs) | |
299 | cpus_and(*pmask, cs->cpus_allowed, cpu_online_map); | |
300 | else | |
301 | *pmask = cpu_online_map; | |
302 | BUG_ON(!cpus_intersects(*pmask, cpu_online_map)); | |
303 | } | |
304 | ||
305 | /* | |
306 | * Return in *pmask the portion of a cpusets's mems_allowed that | |
0e1e7c7a CL |
307 | * are online, with memory. If none are online with memory, walk |
308 | * up the cpuset hierarchy until we find one that does have some | |
309 | * online mems. If we get all the way to the top and still haven't | |
310 | * found any online mems, return node_states[N_HIGH_MEMORY]. | |
1da177e4 LT |
311 | * |
312 | * One way or another, we guarantee to return some non-empty subset | |
0e1e7c7a | 313 | * of node_states[N_HIGH_MEMORY]. |
1da177e4 | 314 | * |
3d3f26a7 | 315 | * Call with callback_mutex held. |
1da177e4 LT |
316 | */ |
317 | ||
318 | static void guarantee_online_mems(const struct cpuset *cs, nodemask_t *pmask) | |
319 | { | |
0e1e7c7a CL |
320 | while (cs && !nodes_intersects(cs->mems_allowed, |
321 | node_states[N_HIGH_MEMORY])) | |
1da177e4 LT |
322 | cs = cs->parent; |
323 | if (cs) | |
0e1e7c7a CL |
324 | nodes_and(*pmask, cs->mems_allowed, |
325 | node_states[N_HIGH_MEMORY]); | |
1da177e4 | 326 | else |
0e1e7c7a CL |
327 | *pmask = node_states[N_HIGH_MEMORY]; |
328 | BUG_ON(!nodes_intersects(*pmask, node_states[N_HIGH_MEMORY])); | |
1da177e4 LT |
329 | } |
330 | ||
cf2a473c PJ |
331 | /** |
332 | * cpuset_update_task_memory_state - update task memory placement | |
333 | * | |
334 | * If the current tasks cpusets mems_allowed changed behind our | |
335 | * backs, update current->mems_allowed, mems_generation and task NUMA | |
336 | * mempolicy to the new value. | |
053199ed | 337 | * |
cf2a473c PJ |
338 | * Task mempolicy is updated by rebinding it relative to the |
339 | * current->cpuset if a task has its memory placement changed. | |
340 | * Do not call this routine if in_interrupt(). | |
341 | * | |
4a01c8d5 | 342 | * Call without callback_mutex or task_lock() held. May be |
2df167a3 PM |
343 | * called with or without cgroup_mutex held. Thanks in part to |
344 | * 'the_top_cpuset_hack', the task's cpuset pointer will never | |
41f7f60d DR |
345 | * be NULL. This routine also might acquire callback_mutex during |
346 | * call. | |
053199ed | 347 | * |
6b9c2603 PJ |
348 | * Reading current->cpuset->mems_generation doesn't need task_lock |
349 | * to guard the current->cpuset derefence, because it is guarded | |
2df167a3 | 350 | * from concurrent freeing of current->cpuset using RCU. |
6b9c2603 PJ |
351 | * |
352 | * The rcu_dereference() is technically probably not needed, | |
353 | * as I don't actually mind if I see a new cpuset pointer but | |
354 | * an old value of mems_generation. However this really only | |
355 | * matters on alpha systems using cpusets heavily. If I dropped | |
356 | * that rcu_dereference(), it would save them a memory barrier. | |
357 | * For all other arch's, rcu_dereference is a no-op anyway, and for | |
358 | * alpha systems not using cpusets, another planned optimization, | |
359 | * avoiding the rcu critical section for tasks in the root cpuset | |
360 | * which is statically allocated, so can't vanish, will make this | |
361 | * irrelevant. Better to use RCU as intended, than to engage in | |
362 | * some cute trick to save a memory barrier that is impossible to | |
363 | * test, for alpha systems using cpusets heavily, which might not | |
364 | * even exist. | |
053199ed PJ |
365 | * |
366 | * This routine is needed to update the per-task mems_allowed data, | |
367 | * within the tasks context, when it is trying to allocate memory | |
368 | * (in various mm/mempolicy.c routines) and notices that some other | |
369 | * task has been modifying its cpuset. | |
1da177e4 LT |
370 | */ |
371 | ||
fe85a998 | 372 | void cpuset_update_task_memory_state(void) |
1da177e4 | 373 | { |
053199ed | 374 | int my_cpusets_mem_gen; |
cf2a473c | 375 | struct task_struct *tsk = current; |
6b9c2603 | 376 | struct cpuset *cs; |
053199ed | 377 | |
13337714 LJ |
378 | rcu_read_lock(); |
379 | my_cpusets_mem_gen = task_cs(tsk)->mems_generation; | |
380 | rcu_read_unlock(); | |
1da177e4 | 381 | |
cf2a473c | 382 | if (my_cpusets_mem_gen != tsk->cpuset_mems_generation) { |
3d3f26a7 | 383 | mutex_lock(&callback_mutex); |
cf2a473c | 384 | task_lock(tsk); |
8793d854 | 385 | cs = task_cs(tsk); /* Maybe changed when task not locked */ |
cf2a473c PJ |
386 | guarantee_online_mems(cs, &tsk->mems_allowed); |
387 | tsk->cpuset_mems_generation = cs->mems_generation; | |
825a46af PJ |
388 | if (is_spread_page(cs)) |
389 | tsk->flags |= PF_SPREAD_PAGE; | |
390 | else | |
391 | tsk->flags &= ~PF_SPREAD_PAGE; | |
392 | if (is_spread_slab(cs)) | |
393 | tsk->flags |= PF_SPREAD_SLAB; | |
394 | else | |
395 | tsk->flags &= ~PF_SPREAD_SLAB; | |
cf2a473c | 396 | task_unlock(tsk); |
3d3f26a7 | 397 | mutex_unlock(&callback_mutex); |
74cb2155 | 398 | mpol_rebind_task(tsk, &tsk->mems_allowed); |
1da177e4 LT |
399 | } |
400 | } | |
401 | ||
402 | /* | |
403 | * is_cpuset_subset(p, q) - Is cpuset p a subset of cpuset q? | |
404 | * | |
405 | * One cpuset is a subset of another if all its allowed CPUs and | |
406 | * Memory Nodes are a subset of the other, and its exclusive flags | |
2df167a3 | 407 | * are only set if the other's are set. Call holding cgroup_mutex. |
1da177e4 LT |
408 | */ |
409 | ||
410 | static int is_cpuset_subset(const struct cpuset *p, const struct cpuset *q) | |
411 | { | |
412 | return cpus_subset(p->cpus_allowed, q->cpus_allowed) && | |
413 | nodes_subset(p->mems_allowed, q->mems_allowed) && | |
414 | is_cpu_exclusive(p) <= is_cpu_exclusive(q) && | |
415 | is_mem_exclusive(p) <= is_mem_exclusive(q); | |
416 | } | |
417 | ||
645fcc9d LZ |
418 | /** |
419 | * alloc_trial_cpuset - allocate a trial cpuset | |
420 | * @cs: the cpuset that the trial cpuset duplicates | |
421 | */ | |
422 | static struct cpuset *alloc_trial_cpuset(const struct cpuset *cs) | |
423 | { | |
424 | return kmemdup(cs, sizeof(*cs), GFP_KERNEL); | |
425 | } | |
426 | ||
427 | /** | |
428 | * free_trial_cpuset - free the trial cpuset | |
429 | * @trial: the trial cpuset to be freed | |
430 | */ | |
431 | static void free_trial_cpuset(struct cpuset *trial) | |
432 | { | |
433 | kfree(trial); | |
434 | } | |
435 | ||
1da177e4 LT |
436 | /* |
437 | * validate_change() - Used to validate that any proposed cpuset change | |
438 | * follows the structural rules for cpusets. | |
439 | * | |
440 | * If we replaced the flag and mask values of the current cpuset | |
441 | * (cur) with those values in the trial cpuset (trial), would | |
442 | * our various subset and exclusive rules still be valid? Presumes | |
2df167a3 | 443 | * cgroup_mutex held. |
1da177e4 LT |
444 | * |
445 | * 'cur' is the address of an actual, in-use cpuset. Operations | |
446 | * such as list traversal that depend on the actual address of the | |
447 | * cpuset in the list must use cur below, not trial. | |
448 | * | |
449 | * 'trial' is the address of bulk structure copy of cur, with | |
450 | * perhaps one or more of the fields cpus_allowed, mems_allowed, | |
451 | * or flags changed to new, trial values. | |
452 | * | |
453 | * Return 0 if valid, -errno if not. | |
454 | */ | |
455 | ||
456 | static int validate_change(const struct cpuset *cur, const struct cpuset *trial) | |
457 | { | |
8793d854 | 458 | struct cgroup *cont; |
1da177e4 LT |
459 | struct cpuset *c, *par; |
460 | ||
461 | /* Each of our child cpusets must be a subset of us */ | |
8793d854 PM |
462 | list_for_each_entry(cont, &cur->css.cgroup->children, sibling) { |
463 | if (!is_cpuset_subset(cgroup_cs(cont), trial)) | |
1da177e4 LT |
464 | return -EBUSY; |
465 | } | |
466 | ||
467 | /* Remaining checks don't apply to root cpuset */ | |
69604067 | 468 | if (cur == &top_cpuset) |
1da177e4 LT |
469 | return 0; |
470 | ||
69604067 PJ |
471 | par = cur->parent; |
472 | ||
1da177e4 LT |
473 | /* We must be a subset of our parent cpuset */ |
474 | if (!is_cpuset_subset(trial, par)) | |
475 | return -EACCES; | |
476 | ||
2df167a3 PM |
477 | /* |
478 | * If either I or some sibling (!= me) is exclusive, we can't | |
479 | * overlap | |
480 | */ | |
8793d854 PM |
481 | list_for_each_entry(cont, &par->css.cgroup->children, sibling) { |
482 | c = cgroup_cs(cont); | |
1da177e4 LT |
483 | if ((is_cpu_exclusive(trial) || is_cpu_exclusive(c)) && |
484 | c != cur && | |
485 | cpus_intersects(trial->cpus_allowed, c->cpus_allowed)) | |
486 | return -EINVAL; | |
487 | if ((is_mem_exclusive(trial) || is_mem_exclusive(c)) && | |
488 | c != cur && | |
489 | nodes_intersects(trial->mems_allowed, c->mems_allowed)) | |
490 | return -EINVAL; | |
491 | } | |
492 | ||
020958b6 PJ |
493 | /* Cpusets with tasks can't have empty cpus_allowed or mems_allowed */ |
494 | if (cgroup_task_count(cur->css.cgroup)) { | |
495 | if (cpus_empty(trial->cpus_allowed) || | |
496 | nodes_empty(trial->mems_allowed)) { | |
497 | return -ENOSPC; | |
498 | } | |
499 | } | |
500 | ||
1da177e4 LT |
501 | return 0; |
502 | } | |
503 | ||
029190c5 | 504 | /* |
cf417141 | 505 | * Helper routine for generate_sched_domains(). |
029190c5 PJ |
506 | * Do cpusets a, b have overlapping cpus_allowed masks? |
507 | */ | |
029190c5 PJ |
508 | static int cpusets_overlap(struct cpuset *a, struct cpuset *b) |
509 | { | |
510 | return cpus_intersects(a->cpus_allowed, b->cpus_allowed); | |
511 | } | |
512 | ||
1d3504fc HS |
513 | static void |
514 | update_domain_attr(struct sched_domain_attr *dattr, struct cpuset *c) | |
515 | { | |
1d3504fc HS |
516 | if (dattr->relax_domain_level < c->relax_domain_level) |
517 | dattr->relax_domain_level = c->relax_domain_level; | |
518 | return; | |
519 | } | |
520 | ||
f5393693 LJ |
521 | static void |
522 | update_domain_attr_tree(struct sched_domain_attr *dattr, struct cpuset *c) | |
523 | { | |
524 | LIST_HEAD(q); | |
525 | ||
526 | list_add(&c->stack_list, &q); | |
527 | while (!list_empty(&q)) { | |
528 | struct cpuset *cp; | |
529 | struct cgroup *cont; | |
530 | struct cpuset *child; | |
531 | ||
532 | cp = list_first_entry(&q, struct cpuset, stack_list); | |
533 | list_del(q.next); | |
534 | ||
535 | if (cpus_empty(cp->cpus_allowed)) | |
536 | continue; | |
537 | ||
538 | if (is_sched_load_balance(cp)) | |
539 | update_domain_attr(dattr, cp); | |
540 | ||
541 | list_for_each_entry(cont, &cp->css.cgroup->children, sibling) { | |
542 | child = cgroup_cs(cont); | |
543 | list_add_tail(&child->stack_list, &q); | |
544 | } | |
545 | } | |
546 | } | |
547 | ||
029190c5 | 548 | /* |
cf417141 MK |
549 | * generate_sched_domains() |
550 | * | |
551 | * This function builds a partial partition of the systems CPUs | |
552 | * A 'partial partition' is a set of non-overlapping subsets whose | |
553 | * union is a subset of that set. | |
554 | * The output of this function needs to be passed to kernel/sched.c | |
555 | * partition_sched_domains() routine, which will rebuild the scheduler's | |
556 | * load balancing domains (sched domains) as specified by that partial | |
557 | * partition. | |
029190c5 PJ |
558 | * |
559 | * See "What is sched_load_balance" in Documentation/cpusets.txt | |
560 | * for a background explanation of this. | |
561 | * | |
562 | * Does not return errors, on the theory that the callers of this | |
563 | * routine would rather not worry about failures to rebuild sched | |
564 | * domains when operating in the severe memory shortage situations | |
565 | * that could cause allocation failures below. | |
566 | * | |
cf417141 | 567 | * Must be called with cgroup_lock held. |
029190c5 PJ |
568 | * |
569 | * The three key local variables below are: | |
aeed6824 | 570 | * q - a linked-list queue of cpuset pointers, used to implement a |
029190c5 PJ |
571 | * top-down scan of all cpusets. This scan loads a pointer |
572 | * to each cpuset marked is_sched_load_balance into the | |
573 | * array 'csa'. For our purposes, rebuilding the schedulers | |
574 | * sched domains, we can ignore !is_sched_load_balance cpusets. | |
575 | * csa - (for CpuSet Array) Array of pointers to all the cpusets | |
576 | * that need to be load balanced, for convenient iterative | |
577 | * access by the subsequent code that finds the best partition, | |
578 | * i.e the set of domains (subsets) of CPUs such that the | |
579 | * cpus_allowed of every cpuset marked is_sched_load_balance | |
580 | * is a subset of one of these domains, while there are as | |
581 | * many such domains as possible, each as small as possible. | |
582 | * doms - Conversion of 'csa' to an array of cpumasks, for passing to | |
583 | * the kernel/sched.c routine partition_sched_domains() in a | |
584 | * convenient format, that can be easily compared to the prior | |
585 | * value to determine what partition elements (sched domains) | |
586 | * were changed (added or removed.) | |
587 | * | |
588 | * Finding the best partition (set of domains): | |
589 | * The triple nested loops below over i, j, k scan over the | |
590 | * load balanced cpusets (using the array of cpuset pointers in | |
591 | * csa[]) looking for pairs of cpusets that have overlapping | |
592 | * cpus_allowed, but which don't have the same 'pn' partition | |
593 | * number and gives them in the same partition number. It keeps | |
594 | * looping on the 'restart' label until it can no longer find | |
595 | * any such pairs. | |
596 | * | |
597 | * The union of the cpus_allowed masks from the set of | |
598 | * all cpusets having the same 'pn' value then form the one | |
599 | * element of the partition (one sched domain) to be passed to | |
600 | * partition_sched_domains(). | |
601 | */ | |
cf417141 MK |
602 | static int generate_sched_domains(cpumask_t **domains, |
603 | struct sched_domain_attr **attributes) | |
029190c5 | 604 | { |
cf417141 | 605 | LIST_HEAD(q); /* queue of cpusets to be scanned */ |
029190c5 PJ |
606 | struct cpuset *cp; /* scans q */ |
607 | struct cpuset **csa; /* array of all cpuset ptrs */ | |
608 | int csn; /* how many cpuset ptrs in csa so far */ | |
609 | int i, j, k; /* indices for partition finding loops */ | |
610 | cpumask_t *doms; /* resulting partition; i.e. sched domains */ | |
1d3504fc | 611 | struct sched_domain_attr *dattr; /* attributes for custom domains */ |
1583715d | 612 | int ndoms = 0; /* number of sched domains in result */ |
029190c5 PJ |
613 | int nslot; /* next empty doms[] cpumask_t slot */ |
614 | ||
029190c5 | 615 | doms = NULL; |
1d3504fc | 616 | dattr = NULL; |
cf417141 | 617 | csa = NULL; |
029190c5 PJ |
618 | |
619 | /* Special case for the 99% of systems with one, full, sched domain */ | |
620 | if (is_sched_load_balance(&top_cpuset)) { | |
029190c5 PJ |
621 | doms = kmalloc(sizeof(cpumask_t), GFP_KERNEL); |
622 | if (!doms) | |
cf417141 MK |
623 | goto done; |
624 | ||
1d3504fc HS |
625 | dattr = kmalloc(sizeof(struct sched_domain_attr), GFP_KERNEL); |
626 | if (dattr) { | |
627 | *dattr = SD_ATTR_INIT; | |
93a65575 | 628 | update_domain_attr_tree(dattr, &top_cpuset); |
1d3504fc | 629 | } |
029190c5 | 630 | *doms = top_cpuset.cpus_allowed; |
cf417141 MK |
631 | |
632 | ndoms = 1; | |
633 | goto done; | |
029190c5 PJ |
634 | } |
635 | ||
029190c5 PJ |
636 | csa = kmalloc(number_of_cpusets * sizeof(cp), GFP_KERNEL); |
637 | if (!csa) | |
638 | goto done; | |
639 | csn = 0; | |
640 | ||
aeed6824 LZ |
641 | list_add(&top_cpuset.stack_list, &q); |
642 | while (!list_empty(&q)) { | |
029190c5 PJ |
643 | struct cgroup *cont; |
644 | struct cpuset *child; /* scans child cpusets of cp */ | |
489a5393 | 645 | |
aeed6824 LZ |
646 | cp = list_first_entry(&q, struct cpuset, stack_list); |
647 | list_del(q.next); | |
648 | ||
489a5393 LJ |
649 | if (cpus_empty(cp->cpus_allowed)) |
650 | continue; | |
651 | ||
f5393693 LJ |
652 | /* |
653 | * All child cpusets contain a subset of the parent's cpus, so | |
654 | * just skip them, and then we call update_domain_attr_tree() | |
655 | * to calc relax_domain_level of the corresponding sched | |
656 | * domain. | |
657 | */ | |
658 | if (is_sched_load_balance(cp)) { | |
029190c5 | 659 | csa[csn++] = cp; |
f5393693 LJ |
660 | continue; |
661 | } | |
489a5393 | 662 | |
029190c5 PJ |
663 | list_for_each_entry(cont, &cp->css.cgroup->children, sibling) { |
664 | child = cgroup_cs(cont); | |
aeed6824 | 665 | list_add_tail(&child->stack_list, &q); |
029190c5 PJ |
666 | } |
667 | } | |
668 | ||
669 | for (i = 0; i < csn; i++) | |
670 | csa[i]->pn = i; | |
671 | ndoms = csn; | |
672 | ||
673 | restart: | |
674 | /* Find the best partition (set of sched domains) */ | |
675 | for (i = 0; i < csn; i++) { | |
676 | struct cpuset *a = csa[i]; | |
677 | int apn = a->pn; | |
678 | ||
679 | for (j = 0; j < csn; j++) { | |
680 | struct cpuset *b = csa[j]; | |
681 | int bpn = b->pn; | |
682 | ||
683 | if (apn != bpn && cpusets_overlap(a, b)) { | |
684 | for (k = 0; k < csn; k++) { | |
685 | struct cpuset *c = csa[k]; | |
686 | ||
687 | if (c->pn == bpn) | |
688 | c->pn = apn; | |
689 | } | |
690 | ndoms--; /* one less element */ | |
691 | goto restart; | |
692 | } | |
693 | } | |
694 | } | |
695 | ||
cf417141 MK |
696 | /* |
697 | * Now we know how many domains to create. | |
698 | * Convert <csn, csa> to <ndoms, doms> and populate cpu masks. | |
699 | */ | |
029190c5 | 700 | doms = kmalloc(ndoms * sizeof(cpumask_t), GFP_KERNEL); |
700018e0 | 701 | if (!doms) |
cf417141 | 702 | goto done; |
cf417141 MK |
703 | |
704 | /* | |
705 | * The rest of the code, including the scheduler, can deal with | |
706 | * dattr==NULL case. No need to abort if alloc fails. | |
707 | */ | |
1d3504fc | 708 | dattr = kmalloc(ndoms * sizeof(struct sched_domain_attr), GFP_KERNEL); |
029190c5 PJ |
709 | |
710 | for (nslot = 0, i = 0; i < csn; i++) { | |
711 | struct cpuset *a = csa[i]; | |
cf417141 | 712 | cpumask_t *dp; |
029190c5 PJ |
713 | int apn = a->pn; |
714 | ||
cf417141 MK |
715 | if (apn < 0) { |
716 | /* Skip completed partitions */ | |
717 | continue; | |
718 | } | |
719 | ||
720 | dp = doms + nslot; | |
721 | ||
722 | if (nslot == ndoms) { | |
723 | static int warnings = 10; | |
724 | if (warnings) { | |
725 | printk(KERN_WARNING | |
726 | "rebuild_sched_domains confused:" | |
727 | " nslot %d, ndoms %d, csn %d, i %d," | |
728 | " apn %d\n", | |
729 | nslot, ndoms, csn, i, apn); | |
730 | warnings--; | |
029190c5 | 731 | } |
cf417141 MK |
732 | continue; |
733 | } | |
029190c5 | 734 | |
cf417141 MK |
735 | cpus_clear(*dp); |
736 | if (dattr) | |
737 | *(dattr + nslot) = SD_ATTR_INIT; | |
738 | for (j = i; j < csn; j++) { | |
739 | struct cpuset *b = csa[j]; | |
740 | ||
741 | if (apn == b->pn) { | |
742 | cpus_or(*dp, *dp, b->cpus_allowed); | |
743 | if (dattr) | |
744 | update_domain_attr_tree(dattr + nslot, b); | |
745 | ||
746 | /* Done with this partition */ | |
747 | b->pn = -1; | |
029190c5 | 748 | } |
029190c5 | 749 | } |
cf417141 | 750 | nslot++; |
029190c5 PJ |
751 | } |
752 | BUG_ON(nslot != ndoms); | |
753 | ||
cf417141 MK |
754 | done: |
755 | kfree(csa); | |
756 | ||
700018e0 LZ |
757 | /* |
758 | * Fallback to the default domain if kmalloc() failed. | |
759 | * See comments in partition_sched_domains(). | |
760 | */ | |
761 | if (doms == NULL) | |
762 | ndoms = 1; | |
763 | ||
cf417141 MK |
764 | *domains = doms; |
765 | *attributes = dattr; | |
766 | return ndoms; | |
767 | } | |
768 | ||
769 | /* | |
770 | * Rebuild scheduler domains. | |
771 | * | |
772 | * Call with neither cgroup_mutex held nor within get_online_cpus(). | |
773 | * Takes both cgroup_mutex and get_online_cpus(). | |
774 | * | |
775 | * Cannot be directly called from cpuset code handling changes | |
776 | * to the cpuset pseudo-filesystem, because it cannot be called | |
777 | * from code that already holds cgroup_mutex. | |
778 | */ | |
779 | static void do_rebuild_sched_domains(struct work_struct *unused) | |
780 | { | |
781 | struct sched_domain_attr *attr; | |
782 | cpumask_t *doms; | |
783 | int ndoms; | |
784 | ||
86ef5c9a | 785 | get_online_cpus(); |
cf417141 MK |
786 | |
787 | /* Generate domain masks and attrs */ | |
788 | cgroup_lock(); | |
789 | ndoms = generate_sched_domains(&doms, &attr); | |
790 | cgroup_unlock(); | |
791 | ||
792 | /* Have scheduler rebuild the domains */ | |
793 | partition_sched_domains(ndoms, doms, attr); | |
794 | ||
86ef5c9a | 795 | put_online_cpus(); |
cf417141 | 796 | } |
029190c5 | 797 | |
cf417141 MK |
798 | static DECLARE_WORK(rebuild_sched_domains_work, do_rebuild_sched_domains); |
799 | ||
800 | /* | |
801 | * Rebuild scheduler domains, asynchronously via workqueue. | |
802 | * | |
803 | * If the flag 'sched_load_balance' of any cpuset with non-empty | |
804 | * 'cpus' changes, or if the 'cpus' allowed changes in any cpuset | |
805 | * which has that flag enabled, or if any cpuset with a non-empty | |
806 | * 'cpus' is removed, then call this routine to rebuild the | |
807 | * scheduler's dynamic sched domains. | |
808 | * | |
809 | * The rebuild_sched_domains() and partition_sched_domains() | |
810 | * routines must nest cgroup_lock() inside get_online_cpus(), | |
811 | * but such cpuset changes as these must nest that locking the | |
812 | * other way, holding cgroup_lock() for much of the code. | |
813 | * | |
814 | * So in order to avoid an ABBA deadlock, the cpuset code handling | |
815 | * these user changes delegates the actual sched domain rebuilding | |
816 | * to a separate workqueue thread, which ends up processing the | |
817 | * above do_rebuild_sched_domains() function. | |
818 | */ | |
819 | static void async_rebuild_sched_domains(void) | |
820 | { | |
821 | schedule_work(&rebuild_sched_domains_work); | |
822 | } | |
823 | ||
824 | /* | |
825 | * Accomplishes the same scheduler domain rebuild as the above | |
826 | * async_rebuild_sched_domains(), however it directly calls the | |
827 | * rebuild routine synchronously rather than calling it via an | |
828 | * asynchronous work thread. | |
829 | * | |
830 | * This can only be called from code that is not holding | |
831 | * cgroup_mutex (not nested in a cgroup_lock() call.) | |
832 | */ | |
833 | void rebuild_sched_domains(void) | |
834 | { | |
835 | do_rebuild_sched_domains(NULL); | |
029190c5 PJ |
836 | } |
837 | ||
58f4790b CW |
838 | /** |
839 | * cpuset_test_cpumask - test a task's cpus_allowed versus its cpuset's | |
840 | * @tsk: task to test | |
841 | * @scan: struct cgroup_scanner contained in its struct cpuset_hotplug_scanner | |
842 | * | |
2df167a3 | 843 | * Call with cgroup_mutex held. May take callback_mutex during call. |
58f4790b CW |
844 | * Called for each task in a cgroup by cgroup_scan_tasks(). |
845 | * Return nonzero if this tasks's cpus_allowed mask should be changed (in other | |
846 | * words, if its mask is not equal to its cpuset's mask). | |
053199ed | 847 | */ |
9e0c914c AB |
848 | static int cpuset_test_cpumask(struct task_struct *tsk, |
849 | struct cgroup_scanner *scan) | |
58f4790b CW |
850 | { |
851 | return !cpus_equal(tsk->cpus_allowed, | |
852 | (cgroup_cs(scan->cg))->cpus_allowed); | |
853 | } | |
053199ed | 854 | |
58f4790b CW |
855 | /** |
856 | * cpuset_change_cpumask - make a task's cpus_allowed the same as its cpuset's | |
857 | * @tsk: task to test | |
858 | * @scan: struct cgroup_scanner containing the cgroup of the task | |
859 | * | |
860 | * Called by cgroup_scan_tasks() for each task in a cgroup whose | |
861 | * cpus_allowed mask needs to be changed. | |
862 | * | |
863 | * We don't need to re-check for the cgroup/cpuset membership, since we're | |
864 | * holding cgroup_lock() at this point. | |
865 | */ | |
9e0c914c AB |
866 | static void cpuset_change_cpumask(struct task_struct *tsk, |
867 | struct cgroup_scanner *scan) | |
58f4790b | 868 | { |
f9a86fcb | 869 | set_cpus_allowed_ptr(tsk, &((cgroup_cs(scan->cg))->cpus_allowed)); |
58f4790b CW |
870 | } |
871 | ||
0b2f630a MX |
872 | /** |
873 | * update_tasks_cpumask - Update the cpumasks of tasks in the cpuset. | |
874 | * @cs: the cpuset in which each task's cpus_allowed mask needs to be changed | |
4e74339a | 875 | * @heap: if NULL, defer allocating heap memory to cgroup_scan_tasks() |
0b2f630a MX |
876 | * |
877 | * Called with cgroup_mutex held | |
878 | * | |
879 | * The cgroup_scan_tasks() function will scan all the tasks in a cgroup, | |
880 | * calling callback functions for each. | |
881 | * | |
4e74339a LZ |
882 | * No return value. It's guaranteed that cgroup_scan_tasks() always returns 0 |
883 | * if @heap != NULL. | |
0b2f630a | 884 | */ |
4e74339a | 885 | static void update_tasks_cpumask(struct cpuset *cs, struct ptr_heap *heap) |
0b2f630a MX |
886 | { |
887 | struct cgroup_scanner scan; | |
0b2f630a MX |
888 | |
889 | scan.cg = cs->css.cgroup; | |
890 | scan.test_task = cpuset_test_cpumask; | |
891 | scan.process_task = cpuset_change_cpumask; | |
4e74339a LZ |
892 | scan.heap = heap; |
893 | cgroup_scan_tasks(&scan); | |
0b2f630a MX |
894 | } |
895 | ||
58f4790b CW |
896 | /** |
897 | * update_cpumask - update the cpus_allowed mask of a cpuset and all tasks in it | |
898 | * @cs: the cpuset to consider | |
899 | * @buf: buffer of cpu numbers written to this cpuset | |
900 | */ | |
645fcc9d LZ |
901 | static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs, |
902 | const char *buf) | |
1da177e4 | 903 | { |
4e74339a | 904 | struct ptr_heap heap; |
58f4790b CW |
905 | int retval; |
906 | int is_load_balanced; | |
1da177e4 | 907 | |
4c4d50f7 PJ |
908 | /* top_cpuset.cpus_allowed tracks cpu_online_map; it's read-only */ |
909 | if (cs == &top_cpuset) | |
910 | return -EACCES; | |
911 | ||
6f7f02e7 | 912 | /* |
c8d9c90c | 913 | * An empty cpus_allowed is ok only if the cpuset has no tasks. |
020958b6 PJ |
914 | * Since cpulist_parse() fails on an empty mask, we special case |
915 | * that parsing. The validate_change() call ensures that cpusets | |
916 | * with tasks have cpus. | |
6f7f02e7 | 917 | */ |
020958b6 | 918 | if (!*buf) { |
645fcc9d | 919 | cpus_clear(trialcs->cpus_allowed); |
6f7f02e7 | 920 | } else { |
645fcc9d | 921 | retval = cpulist_parse(buf, &trialcs->cpus_allowed); |
6f7f02e7 DR |
922 | if (retval < 0) |
923 | return retval; | |
37340746 | 924 | |
645fcc9d | 925 | if (!cpus_subset(trialcs->cpus_allowed, cpu_online_map)) |
37340746 | 926 | return -EINVAL; |
6f7f02e7 | 927 | } |
645fcc9d | 928 | retval = validate_change(cs, trialcs); |
85d7b949 DG |
929 | if (retval < 0) |
930 | return retval; | |
029190c5 | 931 | |
8707d8b8 | 932 | /* Nothing to do if the cpus didn't change */ |
645fcc9d | 933 | if (cpus_equal(cs->cpus_allowed, trialcs->cpus_allowed)) |
8707d8b8 | 934 | return 0; |
58f4790b | 935 | |
4e74339a LZ |
936 | retval = heap_init(&heap, PAGE_SIZE, GFP_KERNEL, NULL); |
937 | if (retval) | |
938 | return retval; | |
939 | ||
645fcc9d | 940 | is_load_balanced = is_sched_load_balance(trialcs); |
029190c5 | 941 | |
3d3f26a7 | 942 | mutex_lock(&callback_mutex); |
645fcc9d | 943 | cs->cpus_allowed = trialcs->cpus_allowed; |
3d3f26a7 | 944 | mutex_unlock(&callback_mutex); |
029190c5 | 945 | |
8707d8b8 PM |
946 | /* |
947 | * Scan tasks in the cpuset, and update the cpumasks of any | |
58f4790b | 948 | * that need an update. |
8707d8b8 | 949 | */ |
4e74339a LZ |
950 | update_tasks_cpumask(cs, &heap); |
951 | ||
952 | heap_free(&heap); | |
58f4790b | 953 | |
8707d8b8 | 954 | if (is_load_balanced) |
cf417141 | 955 | async_rebuild_sched_domains(); |
85d7b949 | 956 | return 0; |
1da177e4 LT |
957 | } |
958 | ||
e4e364e8 PJ |
959 | /* |
960 | * cpuset_migrate_mm | |
961 | * | |
962 | * Migrate memory region from one set of nodes to another. | |
963 | * | |
964 | * Temporarilly set tasks mems_allowed to target nodes of migration, | |
965 | * so that the migration code can allocate pages on these nodes. | |
966 | * | |
2df167a3 | 967 | * Call holding cgroup_mutex, so current's cpuset won't change |
c8d9c90c | 968 | * during this call, as manage_mutex holds off any cpuset_attach() |
e4e364e8 PJ |
969 | * calls. Therefore we don't need to take task_lock around the |
970 | * call to guarantee_online_mems(), as we know no one is changing | |
2df167a3 | 971 | * our task's cpuset. |
e4e364e8 PJ |
972 | * |
973 | * Hold callback_mutex around the two modifications of our tasks | |
974 | * mems_allowed to synchronize with cpuset_mems_allowed(). | |
975 | * | |
976 | * While the mm_struct we are migrating is typically from some | |
977 | * other task, the task_struct mems_allowed that we are hacking | |
978 | * is for our current task, which must allocate new pages for that | |
979 | * migrating memory region. | |
980 | * | |
981 | * We call cpuset_update_task_memory_state() before hacking | |
982 | * our tasks mems_allowed, so that we are assured of being in | |
983 | * sync with our tasks cpuset, and in particular, callbacks to | |
984 | * cpuset_update_task_memory_state() from nested page allocations | |
985 | * won't see any mismatch of our cpuset and task mems_generation | |
986 | * values, so won't overwrite our hacked tasks mems_allowed | |
987 | * nodemask. | |
988 | */ | |
989 | ||
990 | static void cpuset_migrate_mm(struct mm_struct *mm, const nodemask_t *from, | |
991 | const nodemask_t *to) | |
992 | { | |
993 | struct task_struct *tsk = current; | |
994 | ||
995 | cpuset_update_task_memory_state(); | |
996 | ||
997 | mutex_lock(&callback_mutex); | |
998 | tsk->mems_allowed = *to; | |
999 | mutex_unlock(&callback_mutex); | |
1000 | ||
1001 | do_migrate_pages(mm, from, to, MPOL_MF_MOVE_ALL); | |
1002 | ||
1003 | mutex_lock(&callback_mutex); | |
8793d854 | 1004 | guarantee_online_mems(task_cs(tsk),&tsk->mems_allowed); |
e4e364e8 PJ |
1005 | mutex_unlock(&callback_mutex); |
1006 | } | |
1007 | ||
8793d854 PM |
1008 | static void *cpuset_being_rebound; |
1009 | ||
0b2f630a MX |
1010 | /** |
1011 | * update_tasks_nodemask - Update the nodemasks of tasks in the cpuset. | |
1012 | * @cs: the cpuset in which each task's mems_allowed mask needs to be changed | |
1013 | * @oldmem: old mems_allowed of cpuset cs | |
1014 | * | |
1015 | * Called with cgroup_mutex held | |
1016 | * Return 0 if successful, -errno if not. | |
1017 | */ | |
1018 | static int update_tasks_nodemask(struct cpuset *cs, const nodemask_t *oldmem) | |
1da177e4 | 1019 | { |
8793d854 | 1020 | struct task_struct *p; |
4225399a PJ |
1021 | struct mm_struct **mmarray; |
1022 | int i, n, ntasks; | |
04c19fa6 | 1023 | int migrate; |
4225399a | 1024 | int fudge; |
8793d854 | 1025 | struct cgroup_iter it; |
0b2f630a | 1026 | int retval; |
59dac16f | 1027 | |
846a16bf | 1028 | cpuset_being_rebound = cs; /* causes mpol_dup() rebind */ |
4225399a PJ |
1029 | |
1030 | fudge = 10; /* spare mmarray[] slots */ | |
1031 | fudge += cpus_weight(cs->cpus_allowed); /* imagine one fork-bomb/cpu */ | |
1032 | retval = -ENOMEM; | |
1033 | ||
1034 | /* | |
1035 | * Allocate mmarray[] to hold mm reference for each task | |
1036 | * in cpuset cs. Can't kmalloc GFP_KERNEL while holding | |
1037 | * tasklist_lock. We could use GFP_ATOMIC, but with a | |
1038 | * few more lines of code, we can retry until we get a big | |
1039 | * enough mmarray[] w/o using GFP_ATOMIC. | |
1040 | */ | |
1041 | while (1) { | |
8793d854 | 1042 | ntasks = cgroup_task_count(cs->css.cgroup); /* guess */ |
4225399a PJ |
1043 | ntasks += fudge; |
1044 | mmarray = kmalloc(ntasks * sizeof(*mmarray), GFP_KERNEL); | |
1045 | if (!mmarray) | |
1046 | goto done; | |
c2aef333 | 1047 | read_lock(&tasklist_lock); /* block fork */ |
8793d854 | 1048 | if (cgroup_task_count(cs->css.cgroup) <= ntasks) |
4225399a | 1049 | break; /* got enough */ |
c2aef333 | 1050 | read_unlock(&tasklist_lock); /* try again */ |
4225399a PJ |
1051 | kfree(mmarray); |
1052 | } | |
1053 | ||
1054 | n = 0; | |
1055 | ||
1056 | /* Load up mmarray[] with mm reference for each task in cpuset. */ | |
8793d854 PM |
1057 | cgroup_iter_start(cs->css.cgroup, &it); |
1058 | while ((p = cgroup_iter_next(cs->css.cgroup, &it))) { | |
4225399a PJ |
1059 | struct mm_struct *mm; |
1060 | ||
1061 | if (n >= ntasks) { | |
1062 | printk(KERN_WARNING | |
1063 | "Cpuset mempolicy rebind incomplete.\n"); | |
8793d854 | 1064 | break; |
4225399a | 1065 | } |
4225399a PJ |
1066 | mm = get_task_mm(p); |
1067 | if (!mm) | |
1068 | continue; | |
1069 | mmarray[n++] = mm; | |
8793d854 PM |
1070 | } |
1071 | cgroup_iter_end(cs->css.cgroup, &it); | |
c2aef333 | 1072 | read_unlock(&tasklist_lock); |
4225399a PJ |
1073 | |
1074 | /* | |
1075 | * Now that we've dropped the tasklist spinlock, we can | |
1076 | * rebind the vma mempolicies of each mm in mmarray[] to their | |
1077 | * new cpuset, and release that mm. The mpol_rebind_mm() | |
1078 | * call takes mmap_sem, which we couldn't take while holding | |
846a16bf | 1079 | * tasklist_lock. Forks can happen again now - the mpol_dup() |
4225399a PJ |
1080 | * cpuset_being_rebound check will catch such forks, and rebind |
1081 | * their vma mempolicies too. Because we still hold the global | |
2df167a3 | 1082 | * cgroup_mutex, we know that no other rebind effort will |
4225399a PJ |
1083 | * be contending for the global variable cpuset_being_rebound. |
1084 | * It's ok if we rebind the same mm twice; mpol_rebind_mm() | |
04c19fa6 | 1085 | * is idempotent. Also migrate pages in each mm to new nodes. |
4225399a | 1086 | */ |
04c19fa6 | 1087 | migrate = is_memory_migrate(cs); |
4225399a PJ |
1088 | for (i = 0; i < n; i++) { |
1089 | struct mm_struct *mm = mmarray[i]; | |
1090 | ||
1091 | mpol_rebind_mm(mm, &cs->mems_allowed); | |
e4e364e8 | 1092 | if (migrate) |
0b2f630a | 1093 | cpuset_migrate_mm(mm, oldmem, &cs->mems_allowed); |
4225399a PJ |
1094 | mmput(mm); |
1095 | } | |
1096 | ||
2df167a3 | 1097 | /* We're done rebinding vmas to this cpuset's new mems_allowed. */ |
4225399a | 1098 | kfree(mmarray); |
8793d854 | 1099 | cpuset_being_rebound = NULL; |
4225399a | 1100 | retval = 0; |
59dac16f | 1101 | done: |
1da177e4 LT |
1102 | return retval; |
1103 | } | |
1104 | ||
0b2f630a MX |
1105 | /* |
1106 | * Handle user request to change the 'mems' memory placement | |
1107 | * of a cpuset. Needs to validate the request, update the | |
1108 | * cpusets mems_allowed and mems_generation, and for each | |
1109 | * task in the cpuset, rebind any vma mempolicies and if | |
1110 | * the cpuset is marked 'memory_migrate', migrate the tasks | |
1111 | * pages to the new memory. | |
1112 | * | |
1113 | * Call with cgroup_mutex held. May take callback_mutex during call. | |
1114 | * Will take tasklist_lock, scan tasklist for tasks in cpuset cs, | |
1115 | * lock each such tasks mm->mmap_sem, scan its vma's and rebind | |
1116 | * their mempolicies to the cpusets new mems_allowed. | |
1117 | */ | |
645fcc9d LZ |
1118 | static int update_nodemask(struct cpuset *cs, struct cpuset *trialcs, |
1119 | const char *buf) | |
0b2f630a | 1120 | { |
0b2f630a MX |
1121 | nodemask_t oldmem; |
1122 | int retval; | |
1123 | ||
1124 | /* | |
1125 | * top_cpuset.mems_allowed tracks node_stats[N_HIGH_MEMORY]; | |
1126 | * it's read-only | |
1127 | */ | |
1128 | if (cs == &top_cpuset) | |
1129 | return -EACCES; | |
1130 | ||
0b2f630a MX |
1131 | /* |
1132 | * An empty mems_allowed is ok iff there are no tasks in the cpuset. | |
1133 | * Since nodelist_parse() fails on an empty mask, we special case | |
1134 | * that parsing. The validate_change() call ensures that cpusets | |
1135 | * with tasks have memory. | |
1136 | */ | |
1137 | if (!*buf) { | |
645fcc9d | 1138 | nodes_clear(trialcs->mems_allowed); |
0b2f630a | 1139 | } else { |
645fcc9d | 1140 | retval = nodelist_parse(buf, trialcs->mems_allowed); |
0b2f630a MX |
1141 | if (retval < 0) |
1142 | goto done; | |
1143 | ||
645fcc9d | 1144 | if (!nodes_subset(trialcs->mems_allowed, |
0b2f630a MX |
1145 | node_states[N_HIGH_MEMORY])) |
1146 | return -EINVAL; | |
1147 | } | |
1148 | oldmem = cs->mems_allowed; | |
645fcc9d | 1149 | if (nodes_equal(oldmem, trialcs->mems_allowed)) { |
0b2f630a MX |
1150 | retval = 0; /* Too easy - nothing to do */ |
1151 | goto done; | |
1152 | } | |
645fcc9d | 1153 | retval = validate_change(cs, trialcs); |
0b2f630a MX |
1154 | if (retval < 0) |
1155 | goto done; | |
1156 | ||
1157 | mutex_lock(&callback_mutex); | |
645fcc9d | 1158 | cs->mems_allowed = trialcs->mems_allowed; |
0b2f630a MX |
1159 | cs->mems_generation = cpuset_mems_generation++; |
1160 | mutex_unlock(&callback_mutex); | |
1161 | ||
1162 | retval = update_tasks_nodemask(cs, &oldmem); | |
1163 | done: | |
1164 | return retval; | |
1165 | } | |
1166 | ||
8793d854 PM |
1167 | int current_cpuset_is_being_rebound(void) |
1168 | { | |
1169 | return task_cs(current) == cpuset_being_rebound; | |
1170 | } | |
1171 | ||
5be7a479 | 1172 | static int update_relax_domain_level(struct cpuset *cs, s64 val) |
1d3504fc | 1173 | { |
30e0e178 LZ |
1174 | if (val < -1 || val >= SD_LV_MAX) |
1175 | return -EINVAL; | |
1d3504fc HS |
1176 | |
1177 | if (val != cs->relax_domain_level) { | |
1178 | cs->relax_domain_level = val; | |
c372e817 | 1179 | if (!cpus_empty(cs->cpus_allowed) && is_sched_load_balance(cs)) |
cf417141 | 1180 | async_rebuild_sched_domains(); |
1d3504fc HS |
1181 | } |
1182 | ||
1183 | return 0; | |
1184 | } | |
1185 | ||
1da177e4 LT |
1186 | /* |
1187 | * update_flag - read a 0 or a 1 in a file and update associated flag | |
78608366 PM |
1188 | * bit: the bit to update (see cpuset_flagbits_t) |
1189 | * cs: the cpuset to update | |
1190 | * turning_on: whether the flag is being set or cleared | |
053199ed | 1191 | * |
2df167a3 | 1192 | * Call with cgroup_mutex held. |
1da177e4 LT |
1193 | */ |
1194 | ||
700fe1ab PM |
1195 | static int update_flag(cpuset_flagbits_t bit, struct cpuset *cs, |
1196 | int turning_on) | |
1da177e4 | 1197 | { |
645fcc9d | 1198 | struct cpuset *trialcs; |
607717a6 | 1199 | int err; |
40b6a762 | 1200 | int balance_flag_changed; |
1da177e4 | 1201 | |
645fcc9d LZ |
1202 | trialcs = alloc_trial_cpuset(cs); |
1203 | if (!trialcs) | |
1204 | return -ENOMEM; | |
1205 | ||
1da177e4 | 1206 | if (turning_on) |
645fcc9d | 1207 | set_bit(bit, &trialcs->flags); |
1da177e4 | 1208 | else |
645fcc9d | 1209 | clear_bit(bit, &trialcs->flags); |
1da177e4 | 1210 | |
645fcc9d | 1211 | err = validate_change(cs, trialcs); |
85d7b949 | 1212 | if (err < 0) |
645fcc9d | 1213 | goto out; |
029190c5 | 1214 | |
029190c5 | 1215 | balance_flag_changed = (is_sched_load_balance(cs) != |
645fcc9d | 1216 | is_sched_load_balance(trialcs)); |
029190c5 | 1217 | |
3d3f26a7 | 1218 | mutex_lock(&callback_mutex); |
645fcc9d | 1219 | cs->flags = trialcs->flags; |
3d3f26a7 | 1220 | mutex_unlock(&callback_mutex); |
85d7b949 | 1221 | |
645fcc9d | 1222 | if (!cpus_empty(trialcs->cpus_allowed) && balance_flag_changed) |
cf417141 | 1223 | async_rebuild_sched_domains(); |
029190c5 | 1224 | |
645fcc9d LZ |
1225 | out: |
1226 | free_trial_cpuset(trialcs); | |
1227 | return err; | |
1da177e4 LT |
1228 | } |
1229 | ||
3e0d98b9 | 1230 | /* |
80f7228b | 1231 | * Frequency meter - How fast is some event occurring? |
3e0d98b9 PJ |
1232 | * |
1233 | * These routines manage a digitally filtered, constant time based, | |
1234 | * event frequency meter. There are four routines: | |
1235 | * fmeter_init() - initialize a frequency meter. | |
1236 | * fmeter_markevent() - called each time the event happens. | |
1237 | * fmeter_getrate() - returns the recent rate of such events. | |
1238 | * fmeter_update() - internal routine used to update fmeter. | |
1239 | * | |
1240 | * A common data structure is passed to each of these routines, | |
1241 | * which is used to keep track of the state required to manage the | |
1242 | * frequency meter and its digital filter. | |
1243 | * | |
1244 | * The filter works on the number of events marked per unit time. | |
1245 | * The filter is single-pole low-pass recursive (IIR). The time unit | |
1246 | * is 1 second. Arithmetic is done using 32-bit integers scaled to | |
1247 | * simulate 3 decimal digits of precision (multiplied by 1000). | |
1248 | * | |
1249 | * With an FM_COEF of 933, and a time base of 1 second, the filter | |
1250 | * has a half-life of 10 seconds, meaning that if the events quit | |
1251 | * happening, then the rate returned from the fmeter_getrate() | |
1252 | * will be cut in half each 10 seconds, until it converges to zero. | |
1253 | * | |
1254 | * It is not worth doing a real infinitely recursive filter. If more | |
1255 | * than FM_MAXTICKS ticks have elapsed since the last filter event, | |
1256 | * just compute FM_MAXTICKS ticks worth, by which point the level | |
1257 | * will be stable. | |
1258 | * | |
1259 | * Limit the count of unprocessed events to FM_MAXCNT, so as to avoid | |
1260 | * arithmetic overflow in the fmeter_update() routine. | |
1261 | * | |
1262 | * Given the simple 32 bit integer arithmetic used, this meter works | |
1263 | * best for reporting rates between one per millisecond (msec) and | |
1264 | * one per 32 (approx) seconds. At constant rates faster than one | |
1265 | * per msec it maxes out at values just under 1,000,000. At constant | |
1266 | * rates between one per msec, and one per second it will stabilize | |
1267 | * to a value N*1000, where N is the rate of events per second. | |
1268 | * At constant rates between one per second and one per 32 seconds, | |
1269 | * it will be choppy, moving up on the seconds that have an event, | |
1270 | * and then decaying until the next event. At rates slower than | |
1271 | * about one in 32 seconds, it decays all the way back to zero between | |
1272 | * each event. | |
1273 | */ | |
1274 | ||
1275 | #define FM_COEF 933 /* coefficient for half-life of 10 secs */ | |
1276 | #define FM_MAXTICKS ((time_t)99) /* useless computing more ticks than this */ | |
1277 | #define FM_MAXCNT 1000000 /* limit cnt to avoid overflow */ | |
1278 | #define FM_SCALE 1000 /* faux fixed point scale */ | |
1279 | ||
1280 | /* Initialize a frequency meter */ | |
1281 | static void fmeter_init(struct fmeter *fmp) | |
1282 | { | |
1283 | fmp->cnt = 0; | |
1284 | fmp->val = 0; | |
1285 | fmp->time = 0; | |
1286 | spin_lock_init(&fmp->lock); | |
1287 | } | |
1288 | ||
1289 | /* Internal meter update - process cnt events and update value */ | |
1290 | static void fmeter_update(struct fmeter *fmp) | |
1291 | { | |
1292 | time_t now = get_seconds(); | |
1293 | time_t ticks = now - fmp->time; | |
1294 | ||
1295 | if (ticks == 0) | |
1296 | return; | |
1297 | ||
1298 | ticks = min(FM_MAXTICKS, ticks); | |
1299 | while (ticks-- > 0) | |
1300 | fmp->val = (FM_COEF * fmp->val) / FM_SCALE; | |
1301 | fmp->time = now; | |
1302 | ||
1303 | fmp->val += ((FM_SCALE - FM_COEF) * fmp->cnt) / FM_SCALE; | |
1304 | fmp->cnt = 0; | |
1305 | } | |
1306 | ||
1307 | /* Process any previous ticks, then bump cnt by one (times scale). */ | |
1308 | static void fmeter_markevent(struct fmeter *fmp) | |
1309 | { | |
1310 | spin_lock(&fmp->lock); | |
1311 | fmeter_update(fmp); | |
1312 | fmp->cnt = min(FM_MAXCNT, fmp->cnt + FM_SCALE); | |
1313 | spin_unlock(&fmp->lock); | |
1314 | } | |
1315 | ||
1316 | /* Process any previous ticks, then return current value. */ | |
1317 | static int fmeter_getrate(struct fmeter *fmp) | |
1318 | { | |
1319 | int val; | |
1320 | ||
1321 | spin_lock(&fmp->lock); | |
1322 | fmeter_update(fmp); | |
1323 | val = fmp->val; | |
1324 | spin_unlock(&fmp->lock); | |
1325 | return val; | |
1326 | } | |
1327 | ||
2341d1b6 LZ |
1328 | /* Protected by cgroup_lock */ |
1329 | static cpumask_var_t cpus_attach; | |
1330 | ||
2df167a3 | 1331 | /* Called by cgroups to determine if a cpuset is usable; cgroup_mutex held */ |
8793d854 PM |
1332 | static int cpuset_can_attach(struct cgroup_subsys *ss, |
1333 | struct cgroup *cont, struct task_struct *tsk) | |
1da177e4 | 1334 | { |
8793d854 | 1335 | struct cpuset *cs = cgroup_cs(cont); |
5771f0a2 | 1336 | int ret = 0; |
1da177e4 | 1337 | |
1da177e4 LT |
1338 | if (cpus_empty(cs->cpus_allowed) || nodes_empty(cs->mems_allowed)) |
1339 | return -ENOSPC; | |
9985b0ba | 1340 | |
5771f0a2 | 1341 | if (tsk->flags & PF_THREAD_BOUND) { |
9985b0ba | 1342 | mutex_lock(&callback_mutex); |
5771f0a2 LZ |
1343 | if (!cpus_equal(tsk->cpus_allowed, cs->cpus_allowed)) |
1344 | ret = -EINVAL; | |
9985b0ba | 1345 | mutex_unlock(&callback_mutex); |
9985b0ba | 1346 | } |
1da177e4 | 1347 | |
5771f0a2 | 1348 | return ret < 0 ? ret : security_task_setscheduler(tsk, 0, NULL); |
8793d854 | 1349 | } |
1da177e4 | 1350 | |
8793d854 PM |
1351 | static void cpuset_attach(struct cgroup_subsys *ss, |
1352 | struct cgroup *cont, struct cgroup *oldcont, | |
1353 | struct task_struct *tsk) | |
1354 | { | |
8793d854 PM |
1355 | nodemask_t from, to; |
1356 | struct mm_struct *mm; | |
1357 | struct cpuset *cs = cgroup_cs(cont); | |
1358 | struct cpuset *oldcs = cgroup_cs(oldcont); | |
9985b0ba | 1359 | int err; |
22fb52dd | 1360 | |
f5813d94 | 1361 | if (cs == &top_cpuset) { |
2341d1b6 | 1362 | cpumask_copy(cpus_attach, cpu_possible_mask); |
f5813d94 MX |
1363 | } else { |
1364 | mutex_lock(&callback_mutex); | |
2341d1b6 | 1365 | guarantee_online_cpus(cs, cpus_attach); |
f5813d94 MX |
1366 | mutex_unlock(&callback_mutex); |
1367 | } | |
2341d1b6 | 1368 | err = set_cpus_allowed_ptr(tsk, cpus_attach); |
9985b0ba DR |
1369 | if (err) |
1370 | return; | |
1da177e4 | 1371 | |
45b07ef3 PJ |
1372 | from = oldcs->mems_allowed; |
1373 | to = cs->mems_allowed; | |
4225399a PJ |
1374 | mm = get_task_mm(tsk); |
1375 | if (mm) { | |
1376 | mpol_rebind_mm(mm, &to); | |
2741a559 | 1377 | if (is_memory_migrate(cs)) |
e4e364e8 | 1378 | cpuset_migrate_mm(mm, &from, &to); |
4225399a PJ |
1379 | mmput(mm); |
1380 | } | |
1da177e4 LT |
1381 | } |
1382 | ||
1383 | /* The various types of files and directories in a cpuset file system */ | |
1384 | ||
1385 | typedef enum { | |
45b07ef3 | 1386 | FILE_MEMORY_MIGRATE, |
1da177e4 LT |
1387 | FILE_CPULIST, |
1388 | FILE_MEMLIST, | |
1389 | FILE_CPU_EXCLUSIVE, | |
1390 | FILE_MEM_EXCLUSIVE, | |
78608366 | 1391 | FILE_MEM_HARDWALL, |
029190c5 | 1392 | FILE_SCHED_LOAD_BALANCE, |
1d3504fc | 1393 | FILE_SCHED_RELAX_DOMAIN_LEVEL, |
3e0d98b9 PJ |
1394 | FILE_MEMORY_PRESSURE_ENABLED, |
1395 | FILE_MEMORY_PRESSURE, | |
825a46af PJ |
1396 | FILE_SPREAD_PAGE, |
1397 | FILE_SPREAD_SLAB, | |
1da177e4 LT |
1398 | } cpuset_filetype_t; |
1399 | ||
700fe1ab PM |
1400 | static int cpuset_write_u64(struct cgroup *cgrp, struct cftype *cft, u64 val) |
1401 | { | |
1402 | int retval = 0; | |
1403 | struct cpuset *cs = cgroup_cs(cgrp); | |
1404 | cpuset_filetype_t type = cft->private; | |
1405 | ||
e3712395 | 1406 | if (!cgroup_lock_live_group(cgrp)) |
700fe1ab | 1407 | return -ENODEV; |
700fe1ab PM |
1408 | |
1409 | switch (type) { | |
1da177e4 | 1410 | case FILE_CPU_EXCLUSIVE: |
700fe1ab | 1411 | retval = update_flag(CS_CPU_EXCLUSIVE, cs, val); |
1da177e4 LT |
1412 | break; |
1413 | case FILE_MEM_EXCLUSIVE: | |
700fe1ab | 1414 | retval = update_flag(CS_MEM_EXCLUSIVE, cs, val); |
1da177e4 | 1415 | break; |
78608366 PM |
1416 | case FILE_MEM_HARDWALL: |
1417 | retval = update_flag(CS_MEM_HARDWALL, cs, val); | |
1418 | break; | |
029190c5 | 1419 | case FILE_SCHED_LOAD_BALANCE: |
700fe1ab | 1420 | retval = update_flag(CS_SCHED_LOAD_BALANCE, cs, val); |
1d3504fc | 1421 | break; |
45b07ef3 | 1422 | case FILE_MEMORY_MIGRATE: |
700fe1ab | 1423 | retval = update_flag(CS_MEMORY_MIGRATE, cs, val); |
45b07ef3 | 1424 | break; |
3e0d98b9 | 1425 | case FILE_MEMORY_PRESSURE_ENABLED: |
700fe1ab | 1426 | cpuset_memory_pressure_enabled = !!val; |
3e0d98b9 PJ |
1427 | break; |
1428 | case FILE_MEMORY_PRESSURE: | |
1429 | retval = -EACCES; | |
1430 | break; | |
825a46af | 1431 | case FILE_SPREAD_PAGE: |
700fe1ab | 1432 | retval = update_flag(CS_SPREAD_PAGE, cs, val); |
151a4420 | 1433 | cs->mems_generation = cpuset_mems_generation++; |
825a46af PJ |
1434 | break; |
1435 | case FILE_SPREAD_SLAB: | |
700fe1ab | 1436 | retval = update_flag(CS_SPREAD_SLAB, cs, val); |
151a4420 | 1437 | cs->mems_generation = cpuset_mems_generation++; |
825a46af | 1438 | break; |
1da177e4 LT |
1439 | default: |
1440 | retval = -EINVAL; | |
700fe1ab | 1441 | break; |
1da177e4 | 1442 | } |
8793d854 | 1443 | cgroup_unlock(); |
1da177e4 LT |
1444 | return retval; |
1445 | } | |
1446 | ||
5be7a479 PM |
1447 | static int cpuset_write_s64(struct cgroup *cgrp, struct cftype *cft, s64 val) |
1448 | { | |
1449 | int retval = 0; | |
1450 | struct cpuset *cs = cgroup_cs(cgrp); | |
1451 | cpuset_filetype_t type = cft->private; | |
1452 | ||
e3712395 | 1453 | if (!cgroup_lock_live_group(cgrp)) |
5be7a479 | 1454 | return -ENODEV; |
e3712395 | 1455 | |
5be7a479 PM |
1456 | switch (type) { |
1457 | case FILE_SCHED_RELAX_DOMAIN_LEVEL: | |
1458 | retval = update_relax_domain_level(cs, val); | |
1459 | break; | |
1460 | default: | |
1461 | retval = -EINVAL; | |
1462 | break; | |
1463 | } | |
1464 | cgroup_unlock(); | |
1465 | return retval; | |
1466 | } | |
1467 | ||
e3712395 PM |
1468 | /* |
1469 | * Common handling for a write to a "cpus" or "mems" file. | |
1470 | */ | |
1471 | static int cpuset_write_resmask(struct cgroup *cgrp, struct cftype *cft, | |
1472 | const char *buf) | |
1473 | { | |
1474 | int retval = 0; | |
645fcc9d LZ |
1475 | struct cpuset *cs = cgroup_cs(cgrp); |
1476 | struct cpuset *trialcs; | |
e3712395 PM |
1477 | |
1478 | if (!cgroup_lock_live_group(cgrp)) | |
1479 | return -ENODEV; | |
1480 | ||
645fcc9d LZ |
1481 | trialcs = alloc_trial_cpuset(cs); |
1482 | if (!trialcs) | |
1483 | return -ENOMEM; | |
1484 | ||
e3712395 PM |
1485 | switch (cft->private) { |
1486 | case FILE_CPULIST: | |
645fcc9d | 1487 | retval = update_cpumask(cs, trialcs, buf); |
e3712395 PM |
1488 | break; |
1489 | case FILE_MEMLIST: | |
645fcc9d | 1490 | retval = update_nodemask(cs, trialcs, buf); |
e3712395 PM |
1491 | break; |
1492 | default: | |
1493 | retval = -EINVAL; | |
1494 | break; | |
1495 | } | |
645fcc9d LZ |
1496 | |
1497 | free_trial_cpuset(trialcs); | |
e3712395 PM |
1498 | cgroup_unlock(); |
1499 | return retval; | |
1500 | } | |
1501 | ||
1da177e4 LT |
1502 | /* |
1503 | * These ascii lists should be read in a single call, by using a user | |
1504 | * buffer large enough to hold the entire map. If read in smaller | |
1505 | * chunks, there is no guarantee of atomicity. Since the display format | |
1506 | * used, list of ranges of sequential numbers, is variable length, | |
1507 | * and since these maps can change value dynamically, one could read | |
1508 | * gibberish by doing partial reads while a list was changing. | |
1509 | * A single large read to a buffer that crosses a page boundary is | |
1510 | * ok, because the result being copied to user land is not recomputed | |
1511 | * across a page fault. | |
1512 | */ | |
1513 | ||
1514 | static int cpuset_sprintf_cpulist(char *page, struct cpuset *cs) | |
1515 | { | |
5a7625df | 1516 | int ret; |
1da177e4 | 1517 | |
3d3f26a7 | 1518 | mutex_lock(&callback_mutex); |
5a7625df | 1519 | ret = cpulist_scnprintf(page, PAGE_SIZE, &cs->cpus_allowed); |
3d3f26a7 | 1520 | mutex_unlock(&callback_mutex); |
1da177e4 | 1521 | |
5a7625df | 1522 | return ret; |
1da177e4 LT |
1523 | } |
1524 | ||
1525 | static int cpuset_sprintf_memlist(char *page, struct cpuset *cs) | |
1526 | { | |
1527 | nodemask_t mask; | |
1528 | ||
3d3f26a7 | 1529 | mutex_lock(&callback_mutex); |
1da177e4 | 1530 | mask = cs->mems_allowed; |
3d3f26a7 | 1531 | mutex_unlock(&callback_mutex); |
1da177e4 LT |
1532 | |
1533 | return nodelist_scnprintf(page, PAGE_SIZE, mask); | |
1534 | } | |
1535 | ||
8793d854 PM |
1536 | static ssize_t cpuset_common_file_read(struct cgroup *cont, |
1537 | struct cftype *cft, | |
1538 | struct file *file, | |
1539 | char __user *buf, | |
1540 | size_t nbytes, loff_t *ppos) | |
1da177e4 | 1541 | { |
8793d854 | 1542 | struct cpuset *cs = cgroup_cs(cont); |
1da177e4 LT |
1543 | cpuset_filetype_t type = cft->private; |
1544 | char *page; | |
1545 | ssize_t retval = 0; | |
1546 | char *s; | |
1da177e4 | 1547 | |
e12ba74d | 1548 | if (!(page = (char *)__get_free_page(GFP_TEMPORARY))) |
1da177e4 LT |
1549 | return -ENOMEM; |
1550 | ||
1551 | s = page; | |
1552 | ||
1553 | switch (type) { | |
1554 | case FILE_CPULIST: | |
1555 | s += cpuset_sprintf_cpulist(s, cs); | |
1556 | break; | |
1557 | case FILE_MEMLIST: | |
1558 | s += cpuset_sprintf_memlist(s, cs); | |
1559 | break; | |
1da177e4 LT |
1560 | default: |
1561 | retval = -EINVAL; | |
1562 | goto out; | |
1563 | } | |
1564 | *s++ = '\n'; | |
1da177e4 | 1565 | |
eacaa1f5 | 1566 | retval = simple_read_from_buffer(buf, nbytes, ppos, page, s - page); |
1da177e4 LT |
1567 | out: |
1568 | free_page((unsigned long)page); | |
1569 | return retval; | |
1570 | } | |
1571 | ||
700fe1ab PM |
1572 | static u64 cpuset_read_u64(struct cgroup *cont, struct cftype *cft) |
1573 | { | |
1574 | struct cpuset *cs = cgroup_cs(cont); | |
1575 | cpuset_filetype_t type = cft->private; | |
1576 | switch (type) { | |
1577 | case FILE_CPU_EXCLUSIVE: | |
1578 | return is_cpu_exclusive(cs); | |
1579 | case FILE_MEM_EXCLUSIVE: | |
1580 | return is_mem_exclusive(cs); | |
78608366 PM |
1581 | case FILE_MEM_HARDWALL: |
1582 | return is_mem_hardwall(cs); | |
700fe1ab PM |
1583 | case FILE_SCHED_LOAD_BALANCE: |
1584 | return is_sched_load_balance(cs); | |
1585 | case FILE_MEMORY_MIGRATE: | |
1586 | return is_memory_migrate(cs); | |
1587 | case FILE_MEMORY_PRESSURE_ENABLED: | |
1588 | return cpuset_memory_pressure_enabled; | |
1589 | case FILE_MEMORY_PRESSURE: | |
1590 | return fmeter_getrate(&cs->fmeter); | |
1591 | case FILE_SPREAD_PAGE: | |
1592 | return is_spread_page(cs); | |
1593 | case FILE_SPREAD_SLAB: | |
1594 | return is_spread_slab(cs); | |
1595 | default: | |
1596 | BUG(); | |
1597 | } | |
cf417141 MK |
1598 | |
1599 | /* Unreachable but makes gcc happy */ | |
1600 | return 0; | |
700fe1ab | 1601 | } |
1da177e4 | 1602 | |
5be7a479 PM |
1603 | static s64 cpuset_read_s64(struct cgroup *cont, struct cftype *cft) |
1604 | { | |
1605 | struct cpuset *cs = cgroup_cs(cont); | |
1606 | cpuset_filetype_t type = cft->private; | |
1607 | switch (type) { | |
1608 | case FILE_SCHED_RELAX_DOMAIN_LEVEL: | |
1609 | return cs->relax_domain_level; | |
1610 | default: | |
1611 | BUG(); | |
1612 | } | |
cf417141 MK |
1613 | |
1614 | /* Unrechable but makes gcc happy */ | |
1615 | return 0; | |
5be7a479 PM |
1616 | } |
1617 | ||
1da177e4 LT |
1618 | |
1619 | /* | |
1620 | * for the common functions, 'private' gives the type of file | |
1621 | */ | |
1622 | ||
addf2c73 PM |
1623 | static struct cftype files[] = { |
1624 | { | |
1625 | .name = "cpus", | |
1626 | .read = cpuset_common_file_read, | |
e3712395 PM |
1627 | .write_string = cpuset_write_resmask, |
1628 | .max_write_len = (100U + 6 * NR_CPUS), | |
addf2c73 PM |
1629 | .private = FILE_CPULIST, |
1630 | }, | |
1631 | ||
1632 | { | |
1633 | .name = "mems", | |
1634 | .read = cpuset_common_file_read, | |
e3712395 PM |
1635 | .write_string = cpuset_write_resmask, |
1636 | .max_write_len = (100U + 6 * MAX_NUMNODES), | |
addf2c73 PM |
1637 | .private = FILE_MEMLIST, |
1638 | }, | |
1639 | ||
1640 | { | |
1641 | .name = "cpu_exclusive", | |
1642 | .read_u64 = cpuset_read_u64, | |
1643 | .write_u64 = cpuset_write_u64, | |
1644 | .private = FILE_CPU_EXCLUSIVE, | |
1645 | }, | |
1646 | ||
1647 | { | |
1648 | .name = "mem_exclusive", | |
1649 | .read_u64 = cpuset_read_u64, | |
1650 | .write_u64 = cpuset_write_u64, | |
1651 | .private = FILE_MEM_EXCLUSIVE, | |
1652 | }, | |
1653 | ||
78608366 PM |
1654 | { |
1655 | .name = "mem_hardwall", | |
1656 | .read_u64 = cpuset_read_u64, | |
1657 | .write_u64 = cpuset_write_u64, | |
1658 | .private = FILE_MEM_HARDWALL, | |
1659 | }, | |
1660 | ||
addf2c73 PM |
1661 | { |
1662 | .name = "sched_load_balance", | |
1663 | .read_u64 = cpuset_read_u64, | |
1664 | .write_u64 = cpuset_write_u64, | |
1665 | .private = FILE_SCHED_LOAD_BALANCE, | |
1666 | }, | |
1667 | ||
1668 | { | |
1669 | .name = "sched_relax_domain_level", | |
5be7a479 PM |
1670 | .read_s64 = cpuset_read_s64, |
1671 | .write_s64 = cpuset_write_s64, | |
addf2c73 PM |
1672 | .private = FILE_SCHED_RELAX_DOMAIN_LEVEL, |
1673 | }, | |
1674 | ||
1675 | { | |
1676 | .name = "memory_migrate", | |
1677 | .read_u64 = cpuset_read_u64, | |
1678 | .write_u64 = cpuset_write_u64, | |
1679 | .private = FILE_MEMORY_MIGRATE, | |
1680 | }, | |
1681 | ||
1682 | { | |
1683 | .name = "memory_pressure", | |
1684 | .read_u64 = cpuset_read_u64, | |
1685 | .write_u64 = cpuset_write_u64, | |
1686 | .private = FILE_MEMORY_PRESSURE, | |
1687 | }, | |
1688 | ||
1689 | { | |
1690 | .name = "memory_spread_page", | |
1691 | .read_u64 = cpuset_read_u64, | |
1692 | .write_u64 = cpuset_write_u64, | |
1693 | .private = FILE_SPREAD_PAGE, | |
1694 | }, | |
1695 | ||
1696 | { | |
1697 | .name = "memory_spread_slab", | |
1698 | .read_u64 = cpuset_read_u64, | |
1699 | .write_u64 = cpuset_write_u64, | |
1700 | .private = FILE_SPREAD_SLAB, | |
1701 | }, | |
45b07ef3 PJ |
1702 | }; |
1703 | ||
3e0d98b9 PJ |
1704 | static struct cftype cft_memory_pressure_enabled = { |
1705 | .name = "memory_pressure_enabled", | |
700fe1ab PM |
1706 | .read_u64 = cpuset_read_u64, |
1707 | .write_u64 = cpuset_write_u64, | |
3e0d98b9 PJ |
1708 | .private = FILE_MEMORY_PRESSURE_ENABLED, |
1709 | }; | |
1710 | ||
8793d854 | 1711 | static int cpuset_populate(struct cgroup_subsys *ss, struct cgroup *cont) |
1da177e4 LT |
1712 | { |
1713 | int err; | |
1714 | ||
addf2c73 PM |
1715 | err = cgroup_add_files(cont, ss, files, ARRAY_SIZE(files)); |
1716 | if (err) | |
1da177e4 | 1717 | return err; |
8793d854 | 1718 | /* memory_pressure_enabled is in root cpuset only */ |
addf2c73 | 1719 | if (!cont->parent) |
8793d854 | 1720 | err = cgroup_add_file(cont, ss, |
addf2c73 PM |
1721 | &cft_memory_pressure_enabled); |
1722 | return err; | |
1da177e4 LT |
1723 | } |
1724 | ||
8793d854 PM |
1725 | /* |
1726 | * post_clone() is called at the end of cgroup_clone(). | |
1727 | * 'cgroup' was just created automatically as a result of | |
1728 | * a cgroup_clone(), and the current task is about to | |
1729 | * be moved into 'cgroup'. | |
1730 | * | |
1731 | * Currently we refuse to set up the cgroup - thereby | |
1732 | * refusing the task to be entered, and as a result refusing | |
1733 | * the sys_unshare() or clone() which initiated it - if any | |
1734 | * sibling cpusets have exclusive cpus or mem. | |
1735 | * | |
1736 | * If this becomes a problem for some users who wish to | |
1737 | * allow that scenario, then cpuset_post_clone() could be | |
1738 | * changed to grant parent->cpus_allowed-sibling_cpus_exclusive | |
2df167a3 PM |
1739 | * (and likewise for mems) to the new cgroup. Called with cgroup_mutex |
1740 | * held. | |
8793d854 PM |
1741 | */ |
1742 | static void cpuset_post_clone(struct cgroup_subsys *ss, | |
1743 | struct cgroup *cgroup) | |
1744 | { | |
1745 | struct cgroup *parent, *child; | |
1746 | struct cpuset *cs, *parent_cs; | |
1747 | ||
1748 | parent = cgroup->parent; | |
1749 | list_for_each_entry(child, &parent->children, sibling) { | |
1750 | cs = cgroup_cs(child); | |
1751 | if (is_mem_exclusive(cs) || is_cpu_exclusive(cs)) | |
1752 | return; | |
1753 | } | |
1754 | cs = cgroup_cs(cgroup); | |
1755 | parent_cs = cgroup_cs(parent); | |
1756 | ||
1757 | cs->mems_allowed = parent_cs->mems_allowed; | |
1758 | cs->cpus_allowed = parent_cs->cpus_allowed; | |
1759 | return; | |
1760 | } | |
1761 | ||
1da177e4 LT |
1762 | /* |
1763 | * cpuset_create - create a cpuset | |
2df167a3 PM |
1764 | * ss: cpuset cgroup subsystem |
1765 | * cont: control group that the new cpuset will be part of | |
1da177e4 LT |
1766 | */ |
1767 | ||
8793d854 PM |
1768 | static struct cgroup_subsys_state *cpuset_create( |
1769 | struct cgroup_subsys *ss, | |
1770 | struct cgroup *cont) | |
1da177e4 LT |
1771 | { |
1772 | struct cpuset *cs; | |
8793d854 | 1773 | struct cpuset *parent; |
1da177e4 | 1774 | |
8793d854 PM |
1775 | if (!cont->parent) { |
1776 | /* This is early initialization for the top cgroup */ | |
1777 | top_cpuset.mems_generation = cpuset_mems_generation++; | |
1778 | return &top_cpuset.css; | |
1779 | } | |
1780 | parent = cgroup_cs(cont->parent); | |
1da177e4 LT |
1781 | cs = kmalloc(sizeof(*cs), GFP_KERNEL); |
1782 | if (!cs) | |
8793d854 | 1783 | return ERR_PTR(-ENOMEM); |
1da177e4 | 1784 | |
cf2a473c | 1785 | cpuset_update_task_memory_state(); |
1da177e4 | 1786 | cs->flags = 0; |
825a46af PJ |
1787 | if (is_spread_page(parent)) |
1788 | set_bit(CS_SPREAD_PAGE, &cs->flags); | |
1789 | if (is_spread_slab(parent)) | |
1790 | set_bit(CS_SPREAD_SLAB, &cs->flags); | |
029190c5 | 1791 | set_bit(CS_SCHED_LOAD_BALANCE, &cs->flags); |
f9a86fcb MT |
1792 | cpus_clear(cs->cpus_allowed); |
1793 | nodes_clear(cs->mems_allowed); | |
151a4420 | 1794 | cs->mems_generation = cpuset_mems_generation++; |
3e0d98b9 | 1795 | fmeter_init(&cs->fmeter); |
1d3504fc | 1796 | cs->relax_domain_level = -1; |
1da177e4 LT |
1797 | |
1798 | cs->parent = parent; | |
202f72d5 | 1799 | number_of_cpusets++; |
8793d854 | 1800 | return &cs->css ; |
1da177e4 LT |
1801 | } |
1802 | ||
029190c5 | 1803 | /* |
029190c5 PJ |
1804 | * If the cpuset being removed has its flag 'sched_load_balance' |
1805 | * enabled, then simulate turning sched_load_balance off, which | |
cf417141 | 1806 | * will call async_rebuild_sched_domains(). |
029190c5 PJ |
1807 | */ |
1808 | ||
8793d854 | 1809 | static void cpuset_destroy(struct cgroup_subsys *ss, struct cgroup *cont) |
1da177e4 | 1810 | { |
8793d854 | 1811 | struct cpuset *cs = cgroup_cs(cont); |
1da177e4 | 1812 | |
cf2a473c | 1813 | cpuset_update_task_memory_state(); |
029190c5 PJ |
1814 | |
1815 | if (is_sched_load_balance(cs)) | |
700fe1ab | 1816 | update_flag(CS_SCHED_LOAD_BALANCE, cs, 0); |
029190c5 | 1817 | |
202f72d5 | 1818 | number_of_cpusets--; |
8793d854 | 1819 | kfree(cs); |
1da177e4 LT |
1820 | } |
1821 | ||
8793d854 PM |
1822 | struct cgroup_subsys cpuset_subsys = { |
1823 | .name = "cpuset", | |
1824 | .create = cpuset_create, | |
cf417141 | 1825 | .destroy = cpuset_destroy, |
8793d854 PM |
1826 | .can_attach = cpuset_can_attach, |
1827 | .attach = cpuset_attach, | |
1828 | .populate = cpuset_populate, | |
1829 | .post_clone = cpuset_post_clone, | |
1830 | .subsys_id = cpuset_subsys_id, | |
1831 | .early_init = 1, | |
1832 | }; | |
1833 | ||
c417f024 PJ |
1834 | /* |
1835 | * cpuset_init_early - just enough so that the calls to | |
1836 | * cpuset_update_task_memory_state() in early init code | |
1837 | * are harmless. | |
1838 | */ | |
1839 | ||
1840 | int __init cpuset_init_early(void) | |
1841 | { | |
8793d854 | 1842 | top_cpuset.mems_generation = cpuset_mems_generation++; |
c417f024 PJ |
1843 | return 0; |
1844 | } | |
1845 | ||
8793d854 | 1846 | |
1da177e4 LT |
1847 | /** |
1848 | * cpuset_init - initialize cpusets at system boot | |
1849 | * | |
1850 | * Description: Initialize top_cpuset and the cpuset internal file system, | |
1851 | **/ | |
1852 | ||
1853 | int __init cpuset_init(void) | |
1854 | { | |
8793d854 | 1855 | int err = 0; |
1da177e4 | 1856 | |
f9a86fcb MT |
1857 | cpus_setall(top_cpuset.cpus_allowed); |
1858 | nodes_setall(top_cpuset.mems_allowed); | |
1da177e4 | 1859 | |
3e0d98b9 | 1860 | fmeter_init(&top_cpuset.fmeter); |
151a4420 | 1861 | top_cpuset.mems_generation = cpuset_mems_generation++; |
029190c5 | 1862 | set_bit(CS_SCHED_LOAD_BALANCE, &top_cpuset.flags); |
1d3504fc | 1863 | top_cpuset.relax_domain_level = -1; |
1da177e4 | 1864 | |
1da177e4 LT |
1865 | err = register_filesystem(&cpuset_fs_type); |
1866 | if (err < 0) | |
8793d854 PM |
1867 | return err; |
1868 | ||
2341d1b6 LZ |
1869 | if (!alloc_cpumask_var(&cpus_attach, GFP_KERNEL)) |
1870 | BUG(); | |
1871 | ||
202f72d5 | 1872 | number_of_cpusets = 1; |
8793d854 | 1873 | return 0; |
1da177e4 LT |
1874 | } |
1875 | ||
956db3ca CW |
1876 | /** |
1877 | * cpuset_do_move_task - move a given task to another cpuset | |
1878 | * @tsk: pointer to task_struct the task to move | |
1879 | * @scan: struct cgroup_scanner contained in its struct cpuset_hotplug_scanner | |
1880 | * | |
1881 | * Called by cgroup_scan_tasks() for each task in a cgroup. | |
1882 | * Return nonzero to stop the walk through the tasks. | |
1883 | */ | |
9e0c914c AB |
1884 | static void cpuset_do_move_task(struct task_struct *tsk, |
1885 | struct cgroup_scanner *scan) | |
956db3ca CW |
1886 | { |
1887 | struct cpuset_hotplug_scanner *chsp; | |
1888 | ||
1889 | chsp = container_of(scan, struct cpuset_hotplug_scanner, scan); | |
1890 | cgroup_attach_task(chsp->to, tsk); | |
1891 | } | |
1892 | ||
1893 | /** | |
1894 | * move_member_tasks_to_cpuset - move tasks from one cpuset to another | |
1895 | * @from: cpuset in which the tasks currently reside | |
1896 | * @to: cpuset to which the tasks will be moved | |
1897 | * | |
c8d9c90c PJ |
1898 | * Called with cgroup_mutex held |
1899 | * callback_mutex must not be held, as cpuset_attach() will take it. | |
956db3ca CW |
1900 | * |
1901 | * The cgroup_scan_tasks() function will scan all the tasks in a cgroup, | |
1902 | * calling callback functions for each. | |
1903 | */ | |
1904 | static void move_member_tasks_to_cpuset(struct cpuset *from, struct cpuset *to) | |
1905 | { | |
1906 | struct cpuset_hotplug_scanner scan; | |
1907 | ||
1908 | scan.scan.cg = from->css.cgroup; | |
1909 | scan.scan.test_task = NULL; /* select all tasks in cgroup */ | |
1910 | scan.scan.process_task = cpuset_do_move_task; | |
1911 | scan.scan.heap = NULL; | |
1912 | scan.to = to->css.cgroup; | |
1913 | ||
da5ef6bb | 1914 | if (cgroup_scan_tasks(&scan.scan)) |
956db3ca CW |
1915 | printk(KERN_ERR "move_member_tasks_to_cpuset: " |
1916 | "cgroup_scan_tasks failed\n"); | |
1917 | } | |
1918 | ||
b1aac8bb | 1919 | /* |
cf417141 | 1920 | * If CPU and/or memory hotplug handlers, below, unplug any CPUs |
b1aac8bb PJ |
1921 | * or memory nodes, we need to walk over the cpuset hierarchy, |
1922 | * removing that CPU or node from all cpusets. If this removes the | |
956db3ca CW |
1923 | * last CPU or node from a cpuset, then move the tasks in the empty |
1924 | * cpuset to its next-highest non-empty parent. | |
b1aac8bb | 1925 | * |
c8d9c90c PJ |
1926 | * Called with cgroup_mutex held |
1927 | * callback_mutex must not be held, as cpuset_attach() will take it. | |
b1aac8bb | 1928 | */ |
956db3ca CW |
1929 | static void remove_tasks_in_empty_cpuset(struct cpuset *cs) |
1930 | { | |
1931 | struct cpuset *parent; | |
1932 | ||
c8d9c90c PJ |
1933 | /* |
1934 | * The cgroup's css_sets list is in use if there are tasks | |
1935 | * in the cpuset; the list is empty if there are none; | |
1936 | * the cs->css.refcnt seems always 0. | |
1937 | */ | |
956db3ca CW |
1938 | if (list_empty(&cs->css.cgroup->css_sets)) |
1939 | return; | |
b1aac8bb | 1940 | |
956db3ca CW |
1941 | /* |
1942 | * Find its next-highest non-empty parent, (top cpuset | |
1943 | * has online cpus, so can't be empty). | |
1944 | */ | |
1945 | parent = cs->parent; | |
b4501295 PJ |
1946 | while (cpus_empty(parent->cpus_allowed) || |
1947 | nodes_empty(parent->mems_allowed)) | |
956db3ca | 1948 | parent = parent->parent; |
956db3ca CW |
1949 | |
1950 | move_member_tasks_to_cpuset(cs, parent); | |
1951 | } | |
1952 | ||
1953 | /* | |
1954 | * Walk the specified cpuset subtree and look for empty cpusets. | |
1955 | * The tasks of such cpuset must be moved to a parent cpuset. | |
1956 | * | |
2df167a3 | 1957 | * Called with cgroup_mutex held. We take callback_mutex to modify |
956db3ca CW |
1958 | * cpus_allowed and mems_allowed. |
1959 | * | |
1960 | * This walk processes the tree from top to bottom, completing one layer | |
1961 | * before dropping down to the next. It always processes a node before | |
1962 | * any of its children. | |
1963 | * | |
1964 | * For now, since we lack memory hot unplug, we'll never see a cpuset | |
1965 | * that has tasks along with an empty 'mems'. But if we did see such | |
1966 | * a cpuset, we'd handle it just like we do if its 'cpus' was empty. | |
1967 | */ | |
d294eb83 | 1968 | static void scan_for_empty_cpusets(struct cpuset *root) |
b1aac8bb | 1969 | { |
8d1e6266 | 1970 | LIST_HEAD(queue); |
956db3ca CW |
1971 | struct cpuset *cp; /* scans cpusets being updated */ |
1972 | struct cpuset *child; /* scans child cpusets of cp */ | |
8793d854 | 1973 | struct cgroup *cont; |
f9b4fb8d | 1974 | nodemask_t oldmems; |
b1aac8bb | 1975 | |
956db3ca CW |
1976 | list_add_tail((struct list_head *)&root->stack_list, &queue); |
1977 | ||
956db3ca | 1978 | while (!list_empty(&queue)) { |
8d1e6266 | 1979 | cp = list_first_entry(&queue, struct cpuset, stack_list); |
956db3ca CW |
1980 | list_del(queue.next); |
1981 | list_for_each_entry(cont, &cp->css.cgroup->children, sibling) { | |
1982 | child = cgroup_cs(cont); | |
1983 | list_add_tail(&child->stack_list, &queue); | |
1984 | } | |
b4501295 PJ |
1985 | |
1986 | /* Continue past cpusets with all cpus, mems online */ | |
1987 | if (cpus_subset(cp->cpus_allowed, cpu_online_map) && | |
1988 | nodes_subset(cp->mems_allowed, node_states[N_HIGH_MEMORY])) | |
1989 | continue; | |
1990 | ||
f9b4fb8d MX |
1991 | oldmems = cp->mems_allowed; |
1992 | ||
956db3ca | 1993 | /* Remove offline cpus and mems from this cpuset. */ |
b4501295 | 1994 | mutex_lock(&callback_mutex); |
956db3ca CW |
1995 | cpus_and(cp->cpus_allowed, cp->cpus_allowed, cpu_online_map); |
1996 | nodes_and(cp->mems_allowed, cp->mems_allowed, | |
1997 | node_states[N_HIGH_MEMORY]); | |
b4501295 PJ |
1998 | mutex_unlock(&callback_mutex); |
1999 | ||
2000 | /* Move tasks from the empty cpuset to a parent */ | |
c8d9c90c | 2001 | if (cpus_empty(cp->cpus_allowed) || |
b4501295 | 2002 | nodes_empty(cp->mems_allowed)) |
956db3ca | 2003 | remove_tasks_in_empty_cpuset(cp); |
f9b4fb8d | 2004 | else { |
4e74339a | 2005 | update_tasks_cpumask(cp, NULL); |
f9b4fb8d MX |
2006 | update_tasks_nodemask(cp, &oldmems); |
2007 | } | |
b1aac8bb PJ |
2008 | } |
2009 | } | |
2010 | ||
4c4d50f7 PJ |
2011 | /* |
2012 | * The top_cpuset tracks what CPUs and Memory Nodes are online, | |
2013 | * period. This is necessary in order to make cpusets transparent | |
2014 | * (of no affect) on systems that are actively using CPU hotplug | |
2015 | * but making no active use of cpusets. | |
2016 | * | |
38837fc7 PJ |
2017 | * This routine ensures that top_cpuset.cpus_allowed tracks |
2018 | * cpu_online_map on each CPU hotplug (cpuhp) event. | |
cf417141 MK |
2019 | * |
2020 | * Called within get_online_cpus(). Needs to call cgroup_lock() | |
2021 | * before calling generate_sched_domains(). | |
4c4d50f7 | 2022 | */ |
cf417141 | 2023 | static int cpuset_track_online_cpus(struct notifier_block *unused_nb, |
029190c5 | 2024 | unsigned long phase, void *unused_cpu) |
4c4d50f7 | 2025 | { |
cf417141 MK |
2026 | struct sched_domain_attr *attr; |
2027 | cpumask_t *doms; | |
2028 | int ndoms; | |
2029 | ||
3e84050c | 2030 | switch (phase) { |
3e84050c DA |
2031 | case CPU_ONLINE: |
2032 | case CPU_ONLINE_FROZEN: | |
2033 | case CPU_DEAD: | |
2034 | case CPU_DEAD_FROZEN: | |
3e84050c | 2035 | break; |
cf417141 | 2036 | |
3e84050c | 2037 | default: |
ac076758 | 2038 | return NOTIFY_DONE; |
3e84050c | 2039 | } |
ac076758 | 2040 | |
cf417141 MK |
2041 | cgroup_lock(); |
2042 | top_cpuset.cpus_allowed = cpu_online_map; | |
2043 | scan_for_empty_cpusets(&top_cpuset); | |
2044 | ndoms = generate_sched_domains(&doms, &attr); | |
2045 | cgroup_unlock(); | |
2046 | ||
2047 | /* Have scheduler rebuild the domains */ | |
2048 | partition_sched_domains(ndoms, doms, attr); | |
2049 | ||
3e84050c | 2050 | return NOTIFY_OK; |
4c4d50f7 | 2051 | } |
4c4d50f7 | 2052 | |
b1aac8bb | 2053 | #ifdef CONFIG_MEMORY_HOTPLUG |
38837fc7 | 2054 | /* |
0e1e7c7a | 2055 | * Keep top_cpuset.mems_allowed tracking node_states[N_HIGH_MEMORY]. |
cf417141 MK |
2056 | * Call this routine anytime after node_states[N_HIGH_MEMORY] changes. |
2057 | * See also the previous routine cpuset_track_online_cpus(). | |
38837fc7 | 2058 | */ |
f481891f MX |
2059 | static int cpuset_track_online_nodes(struct notifier_block *self, |
2060 | unsigned long action, void *arg) | |
38837fc7 | 2061 | { |
cf417141 | 2062 | cgroup_lock(); |
f481891f MX |
2063 | switch (action) { |
2064 | case MEM_ONLINE: | |
2065 | top_cpuset.mems_allowed = node_states[N_HIGH_MEMORY]; | |
2066 | break; | |
2067 | case MEM_OFFLINE: | |
2068 | top_cpuset.mems_allowed = node_states[N_HIGH_MEMORY]; | |
2069 | scan_for_empty_cpusets(&top_cpuset); | |
2070 | break; | |
2071 | default: | |
2072 | break; | |
2073 | } | |
cf417141 | 2074 | cgroup_unlock(); |
f481891f | 2075 | return NOTIFY_OK; |
38837fc7 PJ |
2076 | } |
2077 | #endif | |
2078 | ||
1da177e4 LT |
2079 | /** |
2080 | * cpuset_init_smp - initialize cpus_allowed | |
2081 | * | |
2082 | * Description: Finish top cpuset after cpu, node maps are initialized | |
2083 | **/ | |
2084 | ||
2085 | void __init cpuset_init_smp(void) | |
2086 | { | |
2087 | top_cpuset.cpus_allowed = cpu_online_map; | |
0e1e7c7a | 2088 | top_cpuset.mems_allowed = node_states[N_HIGH_MEMORY]; |
4c4d50f7 | 2089 | |
cf417141 | 2090 | hotcpu_notifier(cpuset_track_online_cpus, 0); |
f481891f | 2091 | hotplug_memory_notifier(cpuset_track_online_nodes, 10); |
1da177e4 LT |
2092 | } |
2093 | ||
2094 | /** | |
1da177e4 LT |
2095 | * cpuset_cpus_allowed - return cpus_allowed mask from a tasks cpuset. |
2096 | * @tsk: pointer to task_struct from which to obtain cpuset->cpus_allowed. | |
f9a86fcb | 2097 | * @pmask: pointer to cpumask_t variable to receive cpus_allowed set. |
1da177e4 LT |
2098 | * |
2099 | * Description: Returns the cpumask_t cpus_allowed of the cpuset | |
2100 | * attached to the specified @tsk. Guaranteed to return some non-empty | |
2101 | * subset of cpu_online_map, even if this means going outside the | |
2102 | * tasks cpuset. | |
2103 | **/ | |
2104 | ||
f9a86fcb | 2105 | void cpuset_cpus_allowed(struct task_struct *tsk, cpumask_t *pmask) |
1da177e4 | 2106 | { |
3d3f26a7 | 2107 | mutex_lock(&callback_mutex); |
f9a86fcb | 2108 | cpuset_cpus_allowed_locked(tsk, pmask); |
470fd646 | 2109 | mutex_unlock(&callback_mutex); |
470fd646 CW |
2110 | } |
2111 | ||
2112 | /** | |
2113 | * cpuset_cpus_allowed_locked - return cpus_allowed mask from a tasks cpuset. | |
2df167a3 | 2114 | * Must be called with callback_mutex held. |
470fd646 | 2115 | **/ |
f9a86fcb | 2116 | void cpuset_cpus_allowed_locked(struct task_struct *tsk, cpumask_t *pmask) |
470fd646 | 2117 | { |
909d75a3 | 2118 | task_lock(tsk); |
f9a86fcb | 2119 | guarantee_online_cpus(task_cs(tsk), pmask); |
909d75a3 | 2120 | task_unlock(tsk); |
1da177e4 LT |
2121 | } |
2122 | ||
2123 | void cpuset_init_current_mems_allowed(void) | |
2124 | { | |
f9a86fcb | 2125 | nodes_setall(current->mems_allowed); |
1da177e4 LT |
2126 | } |
2127 | ||
909d75a3 PJ |
2128 | /** |
2129 | * cpuset_mems_allowed - return mems_allowed mask from a tasks cpuset. | |
2130 | * @tsk: pointer to task_struct from which to obtain cpuset->mems_allowed. | |
2131 | * | |
2132 | * Description: Returns the nodemask_t mems_allowed of the cpuset | |
2133 | * attached to the specified @tsk. Guaranteed to return some non-empty | |
0e1e7c7a | 2134 | * subset of node_states[N_HIGH_MEMORY], even if this means going outside the |
909d75a3 PJ |
2135 | * tasks cpuset. |
2136 | **/ | |
2137 | ||
2138 | nodemask_t cpuset_mems_allowed(struct task_struct *tsk) | |
2139 | { | |
2140 | nodemask_t mask; | |
2141 | ||
3d3f26a7 | 2142 | mutex_lock(&callback_mutex); |
909d75a3 | 2143 | task_lock(tsk); |
8793d854 | 2144 | guarantee_online_mems(task_cs(tsk), &mask); |
909d75a3 | 2145 | task_unlock(tsk); |
3d3f26a7 | 2146 | mutex_unlock(&callback_mutex); |
909d75a3 PJ |
2147 | |
2148 | return mask; | |
2149 | } | |
2150 | ||
d9fd8a6d | 2151 | /** |
19770b32 MG |
2152 | * cpuset_nodemask_valid_mems_allowed - check nodemask vs. curremt mems_allowed |
2153 | * @nodemask: the nodemask to be checked | |
d9fd8a6d | 2154 | * |
19770b32 | 2155 | * Are any of the nodes in the nodemask allowed in current->mems_allowed? |
1da177e4 | 2156 | */ |
19770b32 | 2157 | int cpuset_nodemask_valid_mems_allowed(nodemask_t *nodemask) |
1da177e4 | 2158 | { |
19770b32 | 2159 | return nodes_intersects(*nodemask, current->mems_allowed); |
1da177e4 LT |
2160 | } |
2161 | ||
9bf2229f | 2162 | /* |
78608366 PM |
2163 | * nearest_hardwall_ancestor() - Returns the nearest mem_exclusive or |
2164 | * mem_hardwall ancestor to the specified cpuset. Call holding | |
2165 | * callback_mutex. If no ancestor is mem_exclusive or mem_hardwall | |
2166 | * (an unusual configuration), then returns the root cpuset. | |
9bf2229f | 2167 | */ |
78608366 | 2168 | static const struct cpuset *nearest_hardwall_ancestor(const struct cpuset *cs) |
9bf2229f | 2169 | { |
78608366 | 2170 | while (!(is_mem_exclusive(cs) || is_mem_hardwall(cs)) && cs->parent) |
9bf2229f PJ |
2171 | cs = cs->parent; |
2172 | return cs; | |
2173 | } | |
2174 | ||
d9fd8a6d | 2175 | /** |
02a0e53d | 2176 | * cpuset_zone_allowed_softwall - Can we allocate on zone z's memory node? |
9bf2229f | 2177 | * @z: is this zone on an allowed node? |
02a0e53d | 2178 | * @gfp_mask: memory allocation flags |
d9fd8a6d | 2179 | * |
02a0e53d PJ |
2180 | * If we're in interrupt, yes, we can always allocate. If |
2181 | * __GFP_THISNODE is set, yes, we can always allocate. If zone | |
9bf2229f PJ |
2182 | * z's node is in our tasks mems_allowed, yes. If it's not a |
2183 | * __GFP_HARDWALL request and this zone's nodes is in the nearest | |
78608366 | 2184 | * hardwalled cpuset ancestor to this tasks cpuset, yes. |
c596d9f3 DR |
2185 | * If the task has been OOM killed and has access to memory reserves |
2186 | * as specified by the TIF_MEMDIE flag, yes. | |
9bf2229f PJ |
2187 | * Otherwise, no. |
2188 | * | |
02a0e53d PJ |
2189 | * If __GFP_HARDWALL is set, cpuset_zone_allowed_softwall() |
2190 | * reduces to cpuset_zone_allowed_hardwall(). Otherwise, | |
2191 | * cpuset_zone_allowed_softwall() might sleep, and might allow a zone | |
2192 | * from an enclosing cpuset. | |
2193 | * | |
2194 | * cpuset_zone_allowed_hardwall() only handles the simpler case of | |
2195 | * hardwall cpusets, and never sleeps. | |
2196 | * | |
2197 | * The __GFP_THISNODE placement logic is really handled elsewhere, | |
2198 | * by forcibly using a zonelist starting at a specified node, and by | |
2199 | * (in get_page_from_freelist()) refusing to consider the zones for | |
2200 | * any node on the zonelist except the first. By the time any such | |
2201 | * calls get to this routine, we should just shut up and say 'yes'. | |
2202 | * | |
9bf2229f | 2203 | * GFP_USER allocations are marked with the __GFP_HARDWALL bit, |
c596d9f3 DR |
2204 | * and do not allow allocations outside the current tasks cpuset |
2205 | * unless the task has been OOM killed as is marked TIF_MEMDIE. | |
9bf2229f | 2206 | * GFP_KERNEL allocations are not so marked, so can escape to the |
78608366 | 2207 | * nearest enclosing hardwalled ancestor cpuset. |
9bf2229f | 2208 | * |
02a0e53d PJ |
2209 | * Scanning up parent cpusets requires callback_mutex. The |
2210 | * __alloc_pages() routine only calls here with __GFP_HARDWALL bit | |
2211 | * _not_ set if it's a GFP_KERNEL allocation, and all nodes in the | |
2212 | * current tasks mems_allowed came up empty on the first pass over | |
2213 | * the zonelist. So only GFP_KERNEL allocations, if all nodes in the | |
2214 | * cpuset are short of memory, might require taking the callback_mutex | |
2215 | * mutex. | |
9bf2229f | 2216 | * |
36be57ff | 2217 | * The first call here from mm/page_alloc:get_page_from_freelist() |
02a0e53d PJ |
2218 | * has __GFP_HARDWALL set in gfp_mask, enforcing hardwall cpusets, |
2219 | * so no allocation on a node outside the cpuset is allowed (unless | |
2220 | * in interrupt, of course). | |
36be57ff PJ |
2221 | * |
2222 | * The second pass through get_page_from_freelist() doesn't even call | |
2223 | * here for GFP_ATOMIC calls. For those calls, the __alloc_pages() | |
2224 | * variable 'wait' is not set, and the bit ALLOC_CPUSET is not set | |
2225 | * in alloc_flags. That logic and the checks below have the combined | |
2226 | * affect that: | |
9bf2229f PJ |
2227 | * in_interrupt - any node ok (current task context irrelevant) |
2228 | * GFP_ATOMIC - any node ok | |
c596d9f3 | 2229 | * TIF_MEMDIE - any node ok |
78608366 | 2230 | * GFP_KERNEL - any node in enclosing hardwalled cpuset ok |
9bf2229f | 2231 | * GFP_USER - only nodes in current tasks mems allowed ok. |
36be57ff PJ |
2232 | * |
2233 | * Rule: | |
02a0e53d | 2234 | * Don't call cpuset_zone_allowed_softwall if you can't sleep, unless you |
36be57ff PJ |
2235 | * pass in the __GFP_HARDWALL flag set in gfp_flag, which disables |
2236 | * the code that might scan up ancestor cpusets and sleep. | |
02a0e53d | 2237 | */ |
9bf2229f | 2238 | |
02a0e53d | 2239 | int __cpuset_zone_allowed_softwall(struct zone *z, gfp_t gfp_mask) |
1da177e4 | 2240 | { |
9bf2229f PJ |
2241 | int node; /* node that zone z is on */ |
2242 | const struct cpuset *cs; /* current cpuset ancestors */ | |
29afd49b | 2243 | int allowed; /* is allocation in zone z allowed? */ |
9bf2229f | 2244 | |
9b819d20 | 2245 | if (in_interrupt() || (gfp_mask & __GFP_THISNODE)) |
9bf2229f | 2246 | return 1; |
89fa3024 | 2247 | node = zone_to_nid(z); |
92d1dbd2 | 2248 | might_sleep_if(!(gfp_mask & __GFP_HARDWALL)); |
9bf2229f PJ |
2249 | if (node_isset(node, current->mems_allowed)) |
2250 | return 1; | |
c596d9f3 DR |
2251 | /* |
2252 | * Allow tasks that have access to memory reserves because they have | |
2253 | * been OOM killed to get memory anywhere. | |
2254 | */ | |
2255 | if (unlikely(test_thread_flag(TIF_MEMDIE))) | |
2256 | return 1; | |
9bf2229f PJ |
2257 | if (gfp_mask & __GFP_HARDWALL) /* If hardwall request, stop here */ |
2258 | return 0; | |
2259 | ||
5563e770 BP |
2260 | if (current->flags & PF_EXITING) /* Let dying task have memory */ |
2261 | return 1; | |
2262 | ||
9bf2229f | 2263 | /* Not hardwall and node outside mems_allowed: scan up cpusets */ |
3d3f26a7 | 2264 | mutex_lock(&callback_mutex); |
053199ed | 2265 | |
053199ed | 2266 | task_lock(current); |
78608366 | 2267 | cs = nearest_hardwall_ancestor(task_cs(current)); |
053199ed PJ |
2268 | task_unlock(current); |
2269 | ||
9bf2229f | 2270 | allowed = node_isset(node, cs->mems_allowed); |
3d3f26a7 | 2271 | mutex_unlock(&callback_mutex); |
9bf2229f | 2272 | return allowed; |
1da177e4 LT |
2273 | } |
2274 | ||
02a0e53d PJ |
2275 | /* |
2276 | * cpuset_zone_allowed_hardwall - Can we allocate on zone z's memory node? | |
2277 | * @z: is this zone on an allowed node? | |
2278 | * @gfp_mask: memory allocation flags | |
2279 | * | |
2280 | * If we're in interrupt, yes, we can always allocate. | |
2281 | * If __GFP_THISNODE is set, yes, we can always allocate. If zone | |
c596d9f3 DR |
2282 | * z's node is in our tasks mems_allowed, yes. If the task has been |
2283 | * OOM killed and has access to memory reserves as specified by the | |
2284 | * TIF_MEMDIE flag, yes. Otherwise, no. | |
02a0e53d PJ |
2285 | * |
2286 | * The __GFP_THISNODE placement logic is really handled elsewhere, | |
2287 | * by forcibly using a zonelist starting at a specified node, and by | |
2288 | * (in get_page_from_freelist()) refusing to consider the zones for | |
2289 | * any node on the zonelist except the first. By the time any such | |
2290 | * calls get to this routine, we should just shut up and say 'yes'. | |
2291 | * | |
2292 | * Unlike the cpuset_zone_allowed_softwall() variant, above, | |
2293 | * this variant requires that the zone be in the current tasks | |
2294 | * mems_allowed or that we're in interrupt. It does not scan up the | |
2295 | * cpuset hierarchy for the nearest enclosing mem_exclusive cpuset. | |
2296 | * It never sleeps. | |
2297 | */ | |
2298 | ||
2299 | int __cpuset_zone_allowed_hardwall(struct zone *z, gfp_t gfp_mask) | |
2300 | { | |
2301 | int node; /* node that zone z is on */ | |
2302 | ||
2303 | if (in_interrupt() || (gfp_mask & __GFP_THISNODE)) | |
2304 | return 1; | |
2305 | node = zone_to_nid(z); | |
2306 | if (node_isset(node, current->mems_allowed)) | |
2307 | return 1; | |
dedf8b79 DW |
2308 | /* |
2309 | * Allow tasks that have access to memory reserves because they have | |
2310 | * been OOM killed to get memory anywhere. | |
2311 | */ | |
2312 | if (unlikely(test_thread_flag(TIF_MEMDIE))) | |
2313 | return 1; | |
02a0e53d PJ |
2314 | return 0; |
2315 | } | |
2316 | ||
505970b9 PJ |
2317 | /** |
2318 | * cpuset_lock - lock out any changes to cpuset structures | |
2319 | * | |
3d3f26a7 | 2320 | * The out of memory (oom) code needs to mutex_lock cpusets |
505970b9 | 2321 | * from being changed while it scans the tasklist looking for a |
3d3f26a7 | 2322 | * task in an overlapping cpuset. Expose callback_mutex via this |
505970b9 PJ |
2323 | * cpuset_lock() routine, so the oom code can lock it, before |
2324 | * locking the task list. The tasklist_lock is a spinlock, so | |
3d3f26a7 | 2325 | * must be taken inside callback_mutex. |
505970b9 PJ |
2326 | */ |
2327 | ||
2328 | void cpuset_lock(void) | |
2329 | { | |
3d3f26a7 | 2330 | mutex_lock(&callback_mutex); |
505970b9 PJ |
2331 | } |
2332 | ||
2333 | /** | |
2334 | * cpuset_unlock - release lock on cpuset changes | |
2335 | * | |
2336 | * Undo the lock taken in a previous cpuset_lock() call. | |
2337 | */ | |
2338 | ||
2339 | void cpuset_unlock(void) | |
2340 | { | |
3d3f26a7 | 2341 | mutex_unlock(&callback_mutex); |
505970b9 PJ |
2342 | } |
2343 | ||
825a46af PJ |
2344 | /** |
2345 | * cpuset_mem_spread_node() - On which node to begin search for a page | |
2346 | * | |
2347 | * If a task is marked PF_SPREAD_PAGE or PF_SPREAD_SLAB (as for | |
2348 | * tasks in a cpuset with is_spread_page or is_spread_slab set), | |
2349 | * and if the memory allocation used cpuset_mem_spread_node() | |
2350 | * to determine on which node to start looking, as it will for | |
2351 | * certain page cache or slab cache pages such as used for file | |
2352 | * system buffers and inode caches, then instead of starting on the | |
2353 | * local node to look for a free page, rather spread the starting | |
2354 | * node around the tasks mems_allowed nodes. | |
2355 | * | |
2356 | * We don't have to worry about the returned node being offline | |
2357 | * because "it can't happen", and even if it did, it would be ok. | |
2358 | * | |
2359 | * The routines calling guarantee_online_mems() are careful to | |
2360 | * only set nodes in task->mems_allowed that are online. So it | |
2361 | * should not be possible for the following code to return an | |
2362 | * offline node. But if it did, that would be ok, as this routine | |
2363 | * is not returning the node where the allocation must be, only | |
2364 | * the node where the search should start. The zonelist passed to | |
2365 | * __alloc_pages() will include all nodes. If the slab allocator | |
2366 | * is passed an offline node, it will fall back to the local node. | |
2367 | * See kmem_cache_alloc_node(). | |
2368 | */ | |
2369 | ||
2370 | int cpuset_mem_spread_node(void) | |
2371 | { | |
2372 | int node; | |
2373 | ||
2374 | node = next_node(current->cpuset_mem_spread_rotor, current->mems_allowed); | |
2375 | if (node == MAX_NUMNODES) | |
2376 | node = first_node(current->mems_allowed); | |
2377 | current->cpuset_mem_spread_rotor = node; | |
2378 | return node; | |
2379 | } | |
2380 | EXPORT_SYMBOL_GPL(cpuset_mem_spread_node); | |
2381 | ||
ef08e3b4 | 2382 | /** |
bbe373f2 DR |
2383 | * cpuset_mems_allowed_intersects - Does @tsk1's mems_allowed intersect @tsk2's? |
2384 | * @tsk1: pointer to task_struct of some task. | |
2385 | * @tsk2: pointer to task_struct of some other task. | |
2386 | * | |
2387 | * Description: Return true if @tsk1's mems_allowed intersects the | |
2388 | * mems_allowed of @tsk2. Used by the OOM killer to determine if | |
2389 | * one of the task's memory usage might impact the memory available | |
2390 | * to the other. | |
ef08e3b4 PJ |
2391 | **/ |
2392 | ||
bbe373f2 DR |
2393 | int cpuset_mems_allowed_intersects(const struct task_struct *tsk1, |
2394 | const struct task_struct *tsk2) | |
ef08e3b4 | 2395 | { |
bbe373f2 | 2396 | return nodes_intersects(tsk1->mems_allowed, tsk2->mems_allowed); |
ef08e3b4 PJ |
2397 | } |
2398 | ||
75aa1994 DR |
2399 | /** |
2400 | * cpuset_print_task_mems_allowed - prints task's cpuset and mems_allowed | |
2401 | * @task: pointer to task_struct of some task. | |
2402 | * | |
2403 | * Description: Prints @task's name, cpuset name, and cached copy of its | |
2404 | * mems_allowed to the kernel log. Must hold task_lock(task) to allow | |
2405 | * dereferencing task_cs(task). | |
2406 | */ | |
2407 | void cpuset_print_task_mems_allowed(struct task_struct *tsk) | |
2408 | { | |
2409 | struct dentry *dentry; | |
2410 | ||
2411 | dentry = task_cs(tsk)->css.cgroup->dentry; | |
2412 | spin_lock(&cpuset_buffer_lock); | |
2413 | snprintf(cpuset_name, CPUSET_NAME_LEN, | |
2414 | dentry ? (const char *)dentry->d_name.name : "/"); | |
2415 | nodelist_scnprintf(cpuset_nodelist, CPUSET_NODELIST_LEN, | |
2416 | tsk->mems_allowed); | |
2417 | printk(KERN_INFO "%s cpuset=%s mems_allowed=%s\n", | |
2418 | tsk->comm, cpuset_name, cpuset_nodelist); | |
2419 | spin_unlock(&cpuset_buffer_lock); | |
2420 | } | |
2421 | ||
3e0d98b9 PJ |
2422 | /* |
2423 | * Collection of memory_pressure is suppressed unless | |
2424 | * this flag is enabled by writing "1" to the special | |
2425 | * cpuset file 'memory_pressure_enabled' in the root cpuset. | |
2426 | */ | |
2427 | ||
c5b2aff8 | 2428 | int cpuset_memory_pressure_enabled __read_mostly; |
3e0d98b9 PJ |
2429 | |
2430 | /** | |
2431 | * cpuset_memory_pressure_bump - keep stats of per-cpuset reclaims. | |
2432 | * | |
2433 | * Keep a running average of the rate of synchronous (direct) | |
2434 | * page reclaim efforts initiated by tasks in each cpuset. | |
2435 | * | |
2436 | * This represents the rate at which some task in the cpuset | |
2437 | * ran low on memory on all nodes it was allowed to use, and | |
2438 | * had to enter the kernels page reclaim code in an effort to | |
2439 | * create more free memory by tossing clean pages or swapping | |
2440 | * or writing dirty pages. | |
2441 | * | |
2442 | * Display to user space in the per-cpuset read-only file | |
2443 | * "memory_pressure". Value displayed is an integer | |
2444 | * representing the recent rate of entry into the synchronous | |
2445 | * (direct) page reclaim by any task attached to the cpuset. | |
2446 | **/ | |
2447 | ||
2448 | void __cpuset_memory_pressure_bump(void) | |
2449 | { | |
3e0d98b9 | 2450 | task_lock(current); |
8793d854 | 2451 | fmeter_markevent(&task_cs(current)->fmeter); |
3e0d98b9 PJ |
2452 | task_unlock(current); |
2453 | } | |
2454 | ||
8793d854 | 2455 | #ifdef CONFIG_PROC_PID_CPUSET |
1da177e4 LT |
2456 | /* |
2457 | * proc_cpuset_show() | |
2458 | * - Print tasks cpuset path into seq_file. | |
2459 | * - Used for /proc/<pid>/cpuset. | |
053199ed PJ |
2460 | * - No need to task_lock(tsk) on this tsk->cpuset reference, as it |
2461 | * doesn't really matter if tsk->cpuset changes after we read it, | |
c8d9c90c | 2462 | * and we take cgroup_mutex, keeping cpuset_attach() from changing it |
2df167a3 | 2463 | * anyway. |
1da177e4 | 2464 | */ |
029190c5 | 2465 | static int proc_cpuset_show(struct seq_file *m, void *unused_v) |
1da177e4 | 2466 | { |
13b41b09 | 2467 | struct pid *pid; |
1da177e4 LT |
2468 | struct task_struct *tsk; |
2469 | char *buf; | |
8793d854 | 2470 | struct cgroup_subsys_state *css; |
99f89551 | 2471 | int retval; |
1da177e4 | 2472 | |
99f89551 | 2473 | retval = -ENOMEM; |
1da177e4 LT |
2474 | buf = kmalloc(PAGE_SIZE, GFP_KERNEL); |
2475 | if (!buf) | |
99f89551 EB |
2476 | goto out; |
2477 | ||
2478 | retval = -ESRCH; | |
13b41b09 EB |
2479 | pid = m->private; |
2480 | tsk = get_pid_task(pid, PIDTYPE_PID); | |
99f89551 EB |
2481 | if (!tsk) |
2482 | goto out_free; | |
1da177e4 | 2483 | |
99f89551 | 2484 | retval = -EINVAL; |
8793d854 PM |
2485 | cgroup_lock(); |
2486 | css = task_subsys_state(tsk, cpuset_subsys_id); | |
2487 | retval = cgroup_path(css->cgroup, buf, PAGE_SIZE); | |
1da177e4 | 2488 | if (retval < 0) |
99f89551 | 2489 | goto out_unlock; |
1da177e4 LT |
2490 | seq_puts(m, buf); |
2491 | seq_putc(m, '\n'); | |
99f89551 | 2492 | out_unlock: |
8793d854 | 2493 | cgroup_unlock(); |
99f89551 EB |
2494 | put_task_struct(tsk); |
2495 | out_free: | |
1da177e4 | 2496 | kfree(buf); |
99f89551 | 2497 | out: |
1da177e4 LT |
2498 | return retval; |
2499 | } | |
2500 | ||
2501 | static int cpuset_open(struct inode *inode, struct file *file) | |
2502 | { | |
13b41b09 EB |
2503 | struct pid *pid = PROC_I(inode)->pid; |
2504 | return single_open(file, proc_cpuset_show, pid); | |
1da177e4 LT |
2505 | } |
2506 | ||
9a32144e | 2507 | const struct file_operations proc_cpuset_operations = { |
1da177e4 LT |
2508 | .open = cpuset_open, |
2509 | .read = seq_read, | |
2510 | .llseek = seq_lseek, | |
2511 | .release = single_release, | |
2512 | }; | |
8793d854 | 2513 | #endif /* CONFIG_PROC_PID_CPUSET */ |
1da177e4 LT |
2514 | |
2515 | /* Display task cpus_allowed, mems_allowed in /proc/<pid>/status file. */ | |
df5f8314 EB |
2516 | void cpuset_task_status_allowed(struct seq_file *m, struct task_struct *task) |
2517 | { | |
2518 | seq_printf(m, "Cpus_allowed:\t"); | |
30e8e136 | 2519 | seq_cpumask(m, &task->cpus_allowed); |
df5f8314 | 2520 | seq_printf(m, "\n"); |
39106dcf | 2521 | seq_printf(m, "Cpus_allowed_list:\t"); |
30e8e136 | 2522 | seq_cpumask_list(m, &task->cpus_allowed); |
39106dcf | 2523 | seq_printf(m, "\n"); |
df5f8314 | 2524 | seq_printf(m, "Mems_allowed:\t"); |
30e8e136 | 2525 | seq_nodemask(m, &task->mems_allowed); |
df5f8314 | 2526 | seq_printf(m, "\n"); |
39106dcf | 2527 | seq_printf(m, "Mems_allowed_list:\t"); |
30e8e136 | 2528 | seq_nodemask_list(m, &task->mems_allowed); |
39106dcf | 2529 | seq_printf(m, "\n"); |
1da177e4 | 2530 | } |