4 * Processor and Memory placement constraints for sets of tasks.
6 * Copyright (C) 2003 BULL SA.
7 * Copyright (C) 2004-2007 Silicon Graphics, Inc.
8 * Copyright (C) 2006 Google, Inc
10 * Portions derived from Patrick Mochel's sysfs code.
11 * sysfs is Copyright (c) 2001-3 Patrick Mochel
13 * 2003-10-10 Written by Simon Derr.
14 * 2003-10-22 Updates by Stephen Hemminger.
15 * 2004 May-July Rework by Paul Jackson.
16 * 2006 Rework by Paul Menage to use generic cgroups
18 * This file is subject to the terms and conditions of the GNU General Public
19 * License. See the file COPYING in the main directory of the Linux
20 * distribution for more details.
23 #include <linux/cpu.h>
24 #include <linux/cpumask.h>
25 #include <linux/cpuset.h>
26 #include <linux/err.h>
27 #include <linux/errno.h>
28 #include <linux/file.h>
30 #include <linux/init.h>
31 #include <linux/interrupt.h>
32 #include <linux/kernel.h>
33 #include <linux/kmod.h>
34 #include <linux/list.h>
35 #include <linux/mempolicy.h>
37 #include <linux/module.h>
38 #include <linux/mount.h>
39 #include <linux/namei.h>
40 #include <linux/pagemap.h>
41 #include <linux/proc_fs.h>
42 #include <linux/rcupdate.h>
43 #include <linux/sched.h>
44 #include <linux/seq_file.h>
45 #include <linux/security.h>
46 #include <linux/slab.h>
47 #include <linux/spinlock.h>
48 #include <linux/stat.h>
49 #include <linux/string.h>
50 #include <linux/time.h>
51 #include <linux/backing-dev.h>
52 #include <linux/sort.h>
54 #include <asm/uaccess.h>
55 #include <asm/atomic.h>
56 #include <linux/mutex.h>
57 #include <linux/kfifo.h>
58 #include <linux/workqueue.h>
59 #include <linux/cgroup.h>
62 * Tracks how many cpusets are currently defined in system.
63 * When there is only one cpuset (the root cpuset) we can
64 * short circuit some hooks.
66 int number_of_cpusets __read_mostly;
68 /* Forward declare cgroup structures */
69 struct cgroup_subsys cpuset_subsys;
72 /* See "Frequency meter" comments, below. */
75 int cnt; /* unprocessed events count */
76 int val; /* most recent output value */
77 time_t time; /* clock (secs) when val computed */
78 spinlock_t lock; /* guards read or write of above */
82 struct cgroup_subsys_state css;
84 unsigned long flags; /* "unsigned long" so bitops work */
85 cpumask_t cpus_allowed; /* CPUs allowed to tasks in cpuset */
86 nodemask_t mems_allowed; /* Memory Nodes allowed to tasks */
88 struct cpuset *parent; /* my parent */
91 * Copy of global cpuset_mems_generation as of the most
92 * recent time this cpuset changed its mems_allowed.
96 struct fmeter fmeter; /* memory_pressure filter */
98 /* partition number for rebuild_sched_domains() */
101 /* for custom sched domain */
102 int relax_domain_level;
104 /* used for walking a cpuset heirarchy */
105 struct list_head stack_list;
108 /* Retrieve the cpuset for a cgroup */
109 static inline struct cpuset *cgroup_cs(struct cgroup *cont)
111 return container_of(cgroup_subsys_state(cont, cpuset_subsys_id),
115 /* Retrieve the cpuset for a task */
116 static inline struct cpuset *task_cs(struct task_struct *task)
118 return container_of(task_subsys_state(task, cpuset_subsys_id),
121 struct cpuset_hotplug_scanner {
122 struct cgroup_scanner scan;
126 /* bits in struct cpuset flags field */
132 CS_SCHED_LOAD_BALANCE,
137 /* convenient tests for these bits */
138 static inline int is_cpu_exclusive(const struct cpuset *cs)
140 return test_bit(CS_CPU_EXCLUSIVE, &cs->flags);
143 static inline int is_mem_exclusive(const struct cpuset *cs)
145 return test_bit(CS_MEM_EXCLUSIVE, &cs->flags);
148 static inline int is_mem_hardwall(const struct cpuset *cs)
150 return test_bit(CS_MEM_HARDWALL, &cs->flags);
153 static inline int is_sched_load_balance(const struct cpuset *cs)
155 return test_bit(CS_SCHED_LOAD_BALANCE, &cs->flags);
158 static inline int is_memory_migrate(const struct cpuset *cs)
160 return test_bit(CS_MEMORY_MIGRATE, &cs->flags);
163 static inline int is_spread_page(const struct cpuset *cs)
165 return test_bit(CS_SPREAD_PAGE, &cs->flags);
168 static inline int is_spread_slab(const struct cpuset *cs)
170 return test_bit(CS_SPREAD_SLAB, &cs->flags);
174 * Increment this integer everytime any cpuset changes its
175 * mems_allowed value. Users of cpusets can track this generation
176 * number, and avoid having to lock and reload mems_allowed unless
177 * the cpuset they're using changes generation.
179 * A single, global generation is needed because cpuset_attach_task() could
180 * reattach a task to a different cpuset, which must not have its
181 * generation numbers aliased with those of that tasks previous cpuset.
183 * Generations are needed for mems_allowed because one task cannot
184 * modify another's memory placement. So we must enable every task,
185 * on every visit to __alloc_pages(), to efficiently check whether
186 * its current->cpuset->mems_allowed has changed, requiring an update
187 * of its current->mems_allowed.
189 * Since writes to cpuset_mems_generation are guarded by the cgroup lock
190 * there is no need to mark it atomic.
192 static int cpuset_mems_generation;
194 static struct cpuset top_cpuset = {
195 .flags = ((1 << CS_CPU_EXCLUSIVE) | (1 << CS_MEM_EXCLUSIVE)),
196 .cpus_allowed = CPU_MASK_ALL,
197 .mems_allowed = NODE_MASK_ALL,
201 * There are two global mutexes guarding cpuset structures. The first
202 * is the main control groups cgroup_mutex, accessed via
203 * cgroup_lock()/cgroup_unlock(). The second is the cpuset-specific
204 * callback_mutex, below. They can nest. It is ok to first take
205 * cgroup_mutex, then nest callback_mutex. We also require taking
206 * task_lock() when dereferencing a task's cpuset pointer. See "The
207 * task_lock() exception", at the end of this comment.
209 * A task must hold both mutexes to modify cpusets. If a task
210 * holds cgroup_mutex, then it blocks others wanting that mutex,
211 * ensuring that it is the only task able to also acquire callback_mutex
212 * and be able to modify cpusets. It can perform various checks on
213 * the cpuset structure first, knowing nothing will change. It can
214 * also allocate memory while just holding cgroup_mutex. While it is
215 * performing these checks, various callback routines can briefly
216 * acquire callback_mutex to query cpusets. Once it is ready to make
217 * the changes, it takes callback_mutex, blocking everyone else.
219 * Calls to the kernel memory allocator can not be made while holding
220 * callback_mutex, as that would risk double tripping on callback_mutex
221 * from one of the callbacks into the cpuset code from within
224 * If a task is only holding callback_mutex, then it has read-only
227 * The task_struct fields mems_allowed and mems_generation may only
228 * be accessed in the context of that task, so require no locks.
230 * The cpuset_common_file_read() handlers only hold callback_mutex across
231 * small pieces of code, such as when reading out possibly multi-word
232 * cpumasks and nodemasks.
234 * Accessing a task's cpuset should be done in accordance with the
235 * guidelines for accessing subsystem state in kernel/cgroup.c
238 static DEFINE_MUTEX(callback_mutex);
240 /* This is ugly, but preserves the userspace API for existing cpuset
241 * users. If someone tries to mount the "cpuset" filesystem, we
242 * silently switch it to mount "cgroup" instead */
243 static int cpuset_get_sb(struct file_system_type *fs_type,
244 int flags, const char *unused_dev_name,
245 void *data, struct vfsmount *mnt)
247 struct file_system_type *cgroup_fs = get_fs_type("cgroup");
252 "release_agent=/sbin/cpuset_release_agent";
253 ret = cgroup_fs->get_sb(cgroup_fs, flags,
254 unused_dev_name, mountopts, mnt);
255 put_filesystem(cgroup_fs);
260 static struct file_system_type cpuset_fs_type = {
262 .get_sb = cpuset_get_sb,
266 * Return in *pmask the portion of a cpusets's cpus_allowed that
267 * are online. If none are online, walk up the cpuset hierarchy
268 * until we find one that does have some online cpus. If we get
269 * all the way to the top and still haven't found any online cpus,
270 * return cpu_online_map. Or if passed a NULL cs from an exit'ing
271 * task, return cpu_online_map.
273 * One way or another, we guarantee to return some non-empty subset
276 * Call with callback_mutex held.
279 static void guarantee_online_cpus(const struct cpuset *cs, cpumask_t *pmask)
281 while (cs && !cpus_intersects(cs->cpus_allowed, cpu_online_map))
284 cpus_and(*pmask, cs->cpus_allowed, cpu_online_map);
286 *pmask = cpu_online_map;
287 BUG_ON(!cpus_intersects(*pmask, cpu_online_map));
291 * Return in *pmask the portion of a cpusets's mems_allowed that
292 * are online, with memory. If none are online with memory, walk
293 * up the cpuset hierarchy until we find one that does have some
294 * online mems. If we get all the way to the top and still haven't
295 * found any online mems, return node_states[N_HIGH_MEMORY].
297 * One way or another, we guarantee to return some non-empty subset
298 * of node_states[N_HIGH_MEMORY].
300 * Call with callback_mutex held.
303 static void guarantee_online_mems(const struct cpuset *cs, nodemask_t *pmask)
305 while (cs && !nodes_intersects(cs->mems_allowed,
306 node_states[N_HIGH_MEMORY]))
309 nodes_and(*pmask, cs->mems_allowed,
310 node_states[N_HIGH_MEMORY]);
312 *pmask = node_states[N_HIGH_MEMORY];
313 BUG_ON(!nodes_intersects(*pmask, node_states[N_HIGH_MEMORY]));
317 * cpuset_update_task_memory_state - update task memory placement
319 * If the current tasks cpusets mems_allowed changed behind our
320 * backs, update current->mems_allowed, mems_generation and task NUMA
321 * mempolicy to the new value.
323 * Task mempolicy is updated by rebinding it relative to the
324 * current->cpuset if a task has its memory placement changed.
325 * Do not call this routine if in_interrupt().
327 * Call without callback_mutex or task_lock() held. May be
328 * called with or without cgroup_mutex held. Thanks in part to
329 * 'the_top_cpuset_hack', the task's cpuset pointer will never
330 * be NULL. This routine also might acquire callback_mutex during
333 * Reading current->cpuset->mems_generation doesn't need task_lock
334 * to guard the current->cpuset derefence, because it is guarded
335 * from concurrent freeing of current->cpuset using RCU.
337 * The rcu_dereference() is technically probably not needed,
338 * as I don't actually mind if I see a new cpuset pointer but
339 * an old value of mems_generation. However this really only
340 * matters on alpha systems using cpusets heavily. If I dropped
341 * that rcu_dereference(), it would save them a memory barrier.
342 * For all other arch's, rcu_dereference is a no-op anyway, and for
343 * alpha systems not using cpusets, another planned optimization,
344 * avoiding the rcu critical section for tasks in the root cpuset
345 * which is statically allocated, so can't vanish, will make this
346 * irrelevant. Better to use RCU as intended, than to engage in
347 * some cute trick to save a memory barrier that is impossible to
348 * test, for alpha systems using cpusets heavily, which might not
351 * This routine is needed to update the per-task mems_allowed data,
352 * within the tasks context, when it is trying to allocate memory
353 * (in various mm/mempolicy.c routines) and notices that some other
354 * task has been modifying its cpuset.
357 void cpuset_update_task_memory_state(void)
359 int my_cpusets_mem_gen;
360 struct task_struct *tsk = current;
363 if (task_cs(tsk) == &top_cpuset) {
364 /* Don't need rcu for top_cpuset. It's never freed. */
365 my_cpusets_mem_gen = top_cpuset.mems_generation;
368 my_cpusets_mem_gen = task_cs(current)->mems_generation;
372 if (my_cpusets_mem_gen != tsk->cpuset_mems_generation) {
373 mutex_lock(&callback_mutex);
375 cs = task_cs(tsk); /* Maybe changed when task not locked */
376 guarantee_online_mems(cs, &tsk->mems_allowed);
377 tsk->cpuset_mems_generation = cs->mems_generation;
378 if (is_spread_page(cs))
379 tsk->flags |= PF_SPREAD_PAGE;
381 tsk->flags &= ~PF_SPREAD_PAGE;
382 if (is_spread_slab(cs))
383 tsk->flags |= PF_SPREAD_SLAB;
385 tsk->flags &= ~PF_SPREAD_SLAB;
387 mutex_unlock(&callback_mutex);
388 mpol_rebind_task(tsk, &tsk->mems_allowed);
393 * is_cpuset_subset(p, q) - Is cpuset p a subset of cpuset q?
395 * One cpuset is a subset of another if all its allowed CPUs and
396 * Memory Nodes are a subset of the other, and its exclusive flags
397 * are only set if the other's are set. Call holding cgroup_mutex.
400 static int is_cpuset_subset(const struct cpuset *p, const struct cpuset *q)
402 return cpus_subset(p->cpus_allowed, q->cpus_allowed) &&
403 nodes_subset(p->mems_allowed, q->mems_allowed) &&
404 is_cpu_exclusive(p) <= is_cpu_exclusive(q) &&
405 is_mem_exclusive(p) <= is_mem_exclusive(q);
409 * validate_change() - Used to validate that any proposed cpuset change
410 * follows the structural rules for cpusets.
412 * If we replaced the flag and mask values of the current cpuset
413 * (cur) with those values in the trial cpuset (trial), would
414 * our various subset and exclusive rules still be valid? Presumes
417 * 'cur' is the address of an actual, in-use cpuset. Operations
418 * such as list traversal that depend on the actual address of the
419 * cpuset in the list must use cur below, not trial.
421 * 'trial' is the address of bulk structure copy of cur, with
422 * perhaps one or more of the fields cpus_allowed, mems_allowed,
423 * or flags changed to new, trial values.
425 * Return 0 if valid, -errno if not.
428 static int validate_change(const struct cpuset *cur, const struct cpuset *trial)
431 struct cpuset *c, *par;
433 /* Each of our child cpusets must be a subset of us */
434 list_for_each_entry(cont, &cur->css.cgroup->children, sibling) {
435 if (!is_cpuset_subset(cgroup_cs(cont), trial))
439 /* Remaining checks don't apply to root cpuset */
440 if (cur == &top_cpuset)
445 /* We must be a subset of our parent cpuset */
446 if (!is_cpuset_subset(trial, par))
450 * If either I or some sibling (!= me) is exclusive, we can't
453 list_for_each_entry(cont, &par->css.cgroup->children, sibling) {
455 if ((is_cpu_exclusive(trial) || is_cpu_exclusive(c)) &&
457 cpus_intersects(trial->cpus_allowed, c->cpus_allowed))
459 if ((is_mem_exclusive(trial) || is_mem_exclusive(c)) &&
461 nodes_intersects(trial->mems_allowed, c->mems_allowed))
465 /* Cpusets with tasks can't have empty cpus_allowed or mems_allowed */
466 if (cgroup_task_count(cur->css.cgroup)) {
467 if (cpus_empty(trial->cpus_allowed) ||
468 nodes_empty(trial->mems_allowed)) {
477 * Helper routine for rebuild_sched_domains().
478 * Do cpusets a, b have overlapping cpus_allowed masks?
481 static int cpusets_overlap(struct cpuset *a, struct cpuset *b)
483 return cpus_intersects(a->cpus_allowed, b->cpus_allowed);
487 update_domain_attr(struct sched_domain_attr *dattr, struct cpuset *c)
491 if (dattr->relax_domain_level < c->relax_domain_level)
492 dattr->relax_domain_level = c->relax_domain_level;
497 * rebuild_sched_domains()
499 * This routine will be called to rebuild the scheduler's dynamic
501 * - if the flag 'sched_load_balance' of any cpuset with non-empty
503 * - or if the 'cpus' allowed changes in any cpuset which has that
505 * - or if the 'sched_relax_domain_level' of any cpuset which has
506 * that flag enabled and with non-empty 'cpus' changes,
507 * - or if any cpuset with non-empty 'cpus' is removed,
508 * - or if a cpu gets offlined.
510 * This routine builds a partial partition of the systems CPUs
511 * (the set of non-overlappping cpumask_t's in the array 'part'
512 * below), and passes that partial partition to the kernel/sched.c
513 * partition_sched_domains() routine, which will rebuild the
514 * schedulers load balancing domains (sched domains) as specified
515 * by that partial partition. A 'partial partition' is a set of
516 * non-overlapping subsets whose union is a subset of that set.
518 * See "What is sched_load_balance" in Documentation/cpusets.txt
519 * for a background explanation of this.
521 * Does not return errors, on the theory that the callers of this
522 * routine would rather not worry about failures to rebuild sched
523 * domains when operating in the severe memory shortage situations
524 * that could cause allocation failures below.
526 * Call with cgroup_mutex held. May take callback_mutex during
527 * call due to the kfifo_alloc() and kmalloc() calls. May nest
528 * a call to the get_online_cpus()/put_online_cpus() pair.
529 * Must not be called holding callback_mutex, because we must not
530 * call get_online_cpus() while holding callback_mutex. Elsewhere
531 * the kernel nests callback_mutex inside get_online_cpus() calls.
532 * So the reverse nesting would risk an ABBA deadlock.
534 * The three key local variables below are:
535 * q - a kfifo queue of cpuset pointers, used to implement a
536 * top-down scan of all cpusets. This scan loads a pointer
537 * to each cpuset marked is_sched_load_balance into the
538 * array 'csa'. For our purposes, rebuilding the schedulers
539 * sched domains, we can ignore !is_sched_load_balance cpusets.
540 * csa - (for CpuSet Array) Array of pointers to all the cpusets
541 * that need to be load balanced, for convenient iterative
542 * access by the subsequent code that finds the best partition,
543 * i.e the set of domains (subsets) of CPUs such that the
544 * cpus_allowed of every cpuset marked is_sched_load_balance
545 * is a subset of one of these domains, while there are as
546 * many such domains as possible, each as small as possible.
547 * doms - Conversion of 'csa' to an array of cpumasks, for passing to
548 * the kernel/sched.c routine partition_sched_domains() in a
549 * convenient format, that can be easily compared to the prior
550 * value to determine what partition elements (sched domains)
551 * were changed (added or removed.)
553 * Finding the best partition (set of domains):
554 * The triple nested loops below over i, j, k scan over the
555 * load balanced cpusets (using the array of cpuset pointers in
556 * csa[]) looking for pairs of cpusets that have overlapping
557 * cpus_allowed, but which don't have the same 'pn' partition
558 * number and gives them in the same partition number. It keeps
559 * looping on the 'restart' label until it can no longer find
562 * The union of the cpus_allowed masks from the set of
563 * all cpusets having the same 'pn' value then form the one
564 * element of the partition (one sched domain) to be passed to
565 * partition_sched_domains().
568 void rebuild_sched_domains(void)
570 struct kfifo *q; /* queue of cpusets to be scanned */
571 struct cpuset *cp; /* scans q */
572 struct cpuset **csa; /* array of all cpuset ptrs */
573 int csn; /* how many cpuset ptrs in csa so far */
574 int i, j, k; /* indices for partition finding loops */
575 cpumask_t *doms; /* resulting partition; i.e. sched domains */
576 struct sched_domain_attr *dattr; /* attributes for custom domains */
577 int ndoms; /* number of sched domains in result */
578 int nslot; /* next empty doms[] cpumask_t slot */
585 /* Special case for the 99% of systems with one, full, sched domain */
586 if (is_sched_load_balance(&top_cpuset)) {
588 doms = kmalloc(sizeof(cpumask_t), GFP_KERNEL);
591 dattr = kmalloc(sizeof(struct sched_domain_attr), GFP_KERNEL);
593 *dattr = SD_ATTR_INIT;
594 update_domain_attr(dattr, &top_cpuset);
596 *doms = top_cpuset.cpus_allowed;
600 q = kfifo_alloc(number_of_cpusets * sizeof(cp), GFP_KERNEL, NULL);
603 csa = kmalloc(number_of_cpusets * sizeof(cp), GFP_KERNEL);
609 __kfifo_put(q, (void *)&cp, sizeof(cp));
610 while (__kfifo_get(q, (void *)&cp, sizeof(cp))) {
612 struct cpuset *child; /* scans child cpusets of cp */
614 if (cpus_empty(cp->cpus_allowed))
617 if (is_sched_load_balance(cp))
620 list_for_each_entry(cont, &cp->css.cgroup->children, sibling) {
621 child = cgroup_cs(cont);
622 __kfifo_put(q, (void *)&child, sizeof(cp));
626 for (i = 0; i < csn; i++)
631 /* Find the best partition (set of sched domains) */
632 for (i = 0; i < csn; i++) {
633 struct cpuset *a = csa[i];
636 for (j = 0; j < csn; j++) {
637 struct cpuset *b = csa[j];
640 if (apn != bpn && cpusets_overlap(a, b)) {
641 for (k = 0; k < csn; k++) {
642 struct cpuset *c = csa[k];
647 ndoms--; /* one less element */
653 /* Convert <csn, csa> to <ndoms, doms> */
654 doms = kmalloc(ndoms * sizeof(cpumask_t), GFP_KERNEL);
657 dattr = kmalloc(ndoms * sizeof(struct sched_domain_attr), GFP_KERNEL);
659 for (nslot = 0, i = 0; i < csn; i++) {
660 struct cpuset *a = csa[i];
664 cpumask_t *dp = doms + nslot;
666 if (nslot == ndoms) {
667 static int warnings = 10;
670 "rebuild_sched_domains confused:"
671 " nslot %d, ndoms %d, csn %d, i %d,"
673 nslot, ndoms, csn, i, apn);
681 *(dattr + nslot) = SD_ATTR_INIT;
682 for (j = i; j < csn; j++) {
683 struct cpuset *b = csa[j];
686 cpus_or(*dp, *dp, b->cpus_allowed);
689 update_domain_attr(dattr
696 BUG_ON(nslot != ndoms);
699 /* Have scheduler rebuild sched domains */
701 partition_sched_domains(ndoms, doms, dattr);
708 /* Don't kfree(doms) -- partition_sched_domains() does that. */
709 /* Don't kfree(dattr) -- partition_sched_domains() does that. */
712 static inline int started_after_time(struct task_struct *t1,
713 struct timespec *time,
714 struct task_struct *t2)
716 int start_diff = timespec_compare(&t1->start_time, time);
717 if (start_diff > 0) {
719 } else if (start_diff < 0) {
723 * Arbitrarily, if two processes started at the same
724 * time, we'll say that the lower pointer value
725 * started first. Note that t2 may have exited by now
726 * so this may not be a valid pointer any longer, but
727 * that's fine - it still serves to distinguish
728 * between two tasks started (effectively)
735 static inline int started_after(void *p1, void *p2)
737 struct task_struct *t1 = p1;
738 struct task_struct *t2 = p2;
739 return started_after_time(t1, &t2->start_time, t2);
743 * cpuset_test_cpumask - test a task's cpus_allowed versus its cpuset's
745 * @scan: struct cgroup_scanner contained in its struct cpuset_hotplug_scanner
747 * Call with cgroup_mutex held. May take callback_mutex during call.
748 * Called for each task in a cgroup by cgroup_scan_tasks().
749 * Return nonzero if this tasks's cpus_allowed mask should be changed (in other
750 * words, if its mask is not equal to its cpuset's mask).
752 static int cpuset_test_cpumask(struct task_struct *tsk,
753 struct cgroup_scanner *scan)
755 return !cpus_equal(tsk->cpus_allowed,
756 (cgroup_cs(scan->cg))->cpus_allowed);
760 * cpuset_change_cpumask - make a task's cpus_allowed the same as its cpuset's
762 * @scan: struct cgroup_scanner containing the cgroup of the task
764 * Called by cgroup_scan_tasks() for each task in a cgroup whose
765 * cpus_allowed mask needs to be changed.
767 * We don't need to re-check for the cgroup/cpuset membership, since we're
768 * holding cgroup_lock() at this point.
770 static void cpuset_change_cpumask(struct task_struct *tsk,
771 struct cgroup_scanner *scan)
773 set_cpus_allowed_ptr(tsk, &((cgroup_cs(scan->cg))->cpus_allowed));
777 * update_tasks_cpumask - Update the cpumasks of tasks in the cpuset.
778 * @cs: the cpuset in which each task's cpus_allowed mask needs to be changed
780 * Called with cgroup_mutex held
782 * The cgroup_scan_tasks() function will scan all the tasks in a cgroup,
783 * calling callback functions for each.
785 * Return 0 if successful, -errno if not.
787 static int update_tasks_cpumask(struct cpuset *cs)
789 struct cgroup_scanner scan;
790 struct ptr_heap heap;
793 retval = heap_init(&heap, PAGE_SIZE, GFP_KERNEL, &started_after);
797 scan.cg = cs->css.cgroup;
798 scan.test_task = cpuset_test_cpumask;
799 scan.process_task = cpuset_change_cpumask;
801 retval = cgroup_scan_tasks(&scan);
808 * update_cpumask - update the cpus_allowed mask of a cpuset and all tasks in it
809 * @cs: the cpuset to consider
810 * @buf: buffer of cpu numbers written to this cpuset
812 static int update_cpumask(struct cpuset *cs, const char *buf)
814 struct cpuset trialcs;
816 int is_load_balanced;
818 /* top_cpuset.cpus_allowed tracks cpu_online_map; it's read-only */
819 if (cs == &top_cpuset)
825 * An empty cpus_allowed is ok only if the cpuset has no tasks.
826 * Since cpulist_parse() fails on an empty mask, we special case
827 * that parsing. The validate_change() call ensures that cpusets
828 * with tasks have cpus.
831 cpus_clear(trialcs.cpus_allowed);
833 retval = cpulist_parse(buf, trialcs.cpus_allowed);
837 if (!cpus_subset(trialcs.cpus_allowed, cpu_online_map))
840 retval = validate_change(cs, &trialcs);
844 /* Nothing to do if the cpus didn't change */
845 if (cpus_equal(cs->cpus_allowed, trialcs.cpus_allowed))
848 is_load_balanced = is_sched_load_balance(&trialcs);
850 mutex_lock(&callback_mutex);
851 cs->cpus_allowed = trialcs.cpus_allowed;
852 mutex_unlock(&callback_mutex);
855 * Scan tasks in the cpuset, and update the cpumasks of any
856 * that need an update.
858 retval = update_tasks_cpumask(cs);
862 if (is_load_balanced)
863 rebuild_sched_domains();
870 * Migrate memory region from one set of nodes to another.
872 * Temporarilly set tasks mems_allowed to target nodes of migration,
873 * so that the migration code can allocate pages on these nodes.
875 * Call holding cgroup_mutex, so current's cpuset won't change
876 * during this call, as manage_mutex holds off any cpuset_attach()
877 * calls. Therefore we don't need to take task_lock around the
878 * call to guarantee_online_mems(), as we know no one is changing
881 * Hold callback_mutex around the two modifications of our tasks
882 * mems_allowed to synchronize with cpuset_mems_allowed().
884 * While the mm_struct we are migrating is typically from some
885 * other task, the task_struct mems_allowed that we are hacking
886 * is for our current task, which must allocate new pages for that
887 * migrating memory region.
889 * We call cpuset_update_task_memory_state() before hacking
890 * our tasks mems_allowed, so that we are assured of being in
891 * sync with our tasks cpuset, and in particular, callbacks to
892 * cpuset_update_task_memory_state() from nested page allocations
893 * won't see any mismatch of our cpuset and task mems_generation
894 * values, so won't overwrite our hacked tasks mems_allowed
898 static void cpuset_migrate_mm(struct mm_struct *mm, const nodemask_t *from,
899 const nodemask_t *to)
901 struct task_struct *tsk = current;
903 cpuset_update_task_memory_state();
905 mutex_lock(&callback_mutex);
906 tsk->mems_allowed = *to;
907 mutex_unlock(&callback_mutex);
909 do_migrate_pages(mm, from, to, MPOL_MF_MOVE_ALL);
911 mutex_lock(&callback_mutex);
912 guarantee_online_mems(task_cs(tsk),&tsk->mems_allowed);
913 mutex_unlock(&callback_mutex);
916 static void *cpuset_being_rebound;
919 * update_tasks_nodemask - Update the nodemasks of tasks in the cpuset.
920 * @cs: the cpuset in which each task's mems_allowed mask needs to be changed
921 * @oldmem: old mems_allowed of cpuset cs
923 * Called with cgroup_mutex held
924 * Return 0 if successful, -errno if not.
926 static int update_tasks_nodemask(struct cpuset *cs, const nodemask_t *oldmem)
928 struct task_struct *p;
929 struct mm_struct **mmarray;
933 struct cgroup_iter it;
936 cpuset_being_rebound = cs; /* causes mpol_dup() rebind */
938 fudge = 10; /* spare mmarray[] slots */
939 fudge += cpus_weight(cs->cpus_allowed); /* imagine one fork-bomb/cpu */
943 * Allocate mmarray[] to hold mm reference for each task
944 * in cpuset cs. Can't kmalloc GFP_KERNEL while holding
945 * tasklist_lock. We could use GFP_ATOMIC, but with a
946 * few more lines of code, we can retry until we get a big
947 * enough mmarray[] w/o using GFP_ATOMIC.
950 ntasks = cgroup_task_count(cs->css.cgroup); /* guess */
952 mmarray = kmalloc(ntasks * sizeof(*mmarray), GFP_KERNEL);
955 read_lock(&tasklist_lock); /* block fork */
956 if (cgroup_task_count(cs->css.cgroup) <= ntasks)
957 break; /* got enough */
958 read_unlock(&tasklist_lock); /* try again */
964 /* Load up mmarray[] with mm reference for each task in cpuset. */
965 cgroup_iter_start(cs->css.cgroup, &it);
966 while ((p = cgroup_iter_next(cs->css.cgroup, &it))) {
967 struct mm_struct *mm;
971 "Cpuset mempolicy rebind incomplete.\n");
979 cgroup_iter_end(cs->css.cgroup, &it);
980 read_unlock(&tasklist_lock);
983 * Now that we've dropped the tasklist spinlock, we can
984 * rebind the vma mempolicies of each mm in mmarray[] to their
985 * new cpuset, and release that mm. The mpol_rebind_mm()
986 * call takes mmap_sem, which we couldn't take while holding
987 * tasklist_lock. Forks can happen again now - the mpol_dup()
988 * cpuset_being_rebound check will catch such forks, and rebind
989 * their vma mempolicies too. Because we still hold the global
990 * cgroup_mutex, we know that no other rebind effort will
991 * be contending for the global variable cpuset_being_rebound.
992 * It's ok if we rebind the same mm twice; mpol_rebind_mm()
993 * is idempotent. Also migrate pages in each mm to new nodes.
995 migrate = is_memory_migrate(cs);
996 for (i = 0; i < n; i++) {
997 struct mm_struct *mm = mmarray[i];
999 mpol_rebind_mm(mm, &cs->mems_allowed);
1001 cpuset_migrate_mm(mm, oldmem, &cs->mems_allowed);
1005 /* We're done rebinding vmas to this cpuset's new mems_allowed. */
1007 cpuset_being_rebound = NULL;
1014 * Handle user request to change the 'mems' memory placement
1015 * of a cpuset. Needs to validate the request, update the
1016 * cpusets mems_allowed and mems_generation, and for each
1017 * task in the cpuset, rebind any vma mempolicies and if
1018 * the cpuset is marked 'memory_migrate', migrate the tasks
1019 * pages to the new memory.
1021 * Call with cgroup_mutex held. May take callback_mutex during call.
1022 * Will take tasklist_lock, scan tasklist for tasks in cpuset cs,
1023 * lock each such tasks mm->mmap_sem, scan its vma's and rebind
1024 * their mempolicies to the cpusets new mems_allowed.
1026 static int update_nodemask(struct cpuset *cs, const char *buf)
1028 struct cpuset trialcs;
1033 * top_cpuset.mems_allowed tracks node_stats[N_HIGH_MEMORY];
1036 if (cs == &top_cpuset)
1042 * An empty mems_allowed is ok iff there are no tasks in the cpuset.
1043 * Since nodelist_parse() fails on an empty mask, we special case
1044 * that parsing. The validate_change() call ensures that cpusets
1045 * with tasks have memory.
1048 nodes_clear(trialcs.mems_allowed);
1050 retval = nodelist_parse(buf, trialcs.mems_allowed);
1054 if (!nodes_subset(trialcs.mems_allowed,
1055 node_states[N_HIGH_MEMORY]))
1058 oldmem = cs->mems_allowed;
1059 if (nodes_equal(oldmem, trialcs.mems_allowed)) {
1060 retval = 0; /* Too easy - nothing to do */
1063 retval = validate_change(cs, &trialcs);
1067 mutex_lock(&callback_mutex);
1068 cs->mems_allowed = trialcs.mems_allowed;
1069 cs->mems_generation = cpuset_mems_generation++;
1070 mutex_unlock(&callback_mutex);
1072 retval = update_tasks_nodemask(cs, &oldmem);
1077 int current_cpuset_is_being_rebound(void)
1079 return task_cs(current) == cpuset_being_rebound;
1082 static int update_relax_domain_level(struct cpuset *cs, s64 val)
1084 if (val < -1 || val >= SD_LV_MAX)
1087 if (val != cs->relax_domain_level) {
1088 cs->relax_domain_level = val;
1089 if (!cpus_empty(cs->cpus_allowed) && is_sched_load_balance(cs))
1090 rebuild_sched_domains();
1097 * update_flag - read a 0 or a 1 in a file and update associated flag
1098 * bit: the bit to update (see cpuset_flagbits_t)
1099 * cs: the cpuset to update
1100 * turning_on: whether the flag is being set or cleared
1102 * Call with cgroup_mutex held.
1105 static int update_flag(cpuset_flagbits_t bit, struct cpuset *cs,
1108 struct cpuset trialcs;
1110 int cpus_nonempty, balance_flag_changed;
1114 set_bit(bit, &trialcs.flags);
1116 clear_bit(bit, &trialcs.flags);
1118 err = validate_change(cs, &trialcs);
1122 cpus_nonempty = !cpus_empty(trialcs.cpus_allowed);
1123 balance_flag_changed = (is_sched_load_balance(cs) !=
1124 is_sched_load_balance(&trialcs));
1126 mutex_lock(&callback_mutex);
1127 cs->flags = trialcs.flags;
1128 mutex_unlock(&callback_mutex);
1130 if (cpus_nonempty && balance_flag_changed)
1131 rebuild_sched_domains();
1137 * Frequency meter - How fast is some event occurring?
1139 * These routines manage a digitally filtered, constant time based,
1140 * event frequency meter. There are four routines:
1141 * fmeter_init() - initialize a frequency meter.
1142 * fmeter_markevent() - called each time the event happens.
1143 * fmeter_getrate() - returns the recent rate of such events.
1144 * fmeter_update() - internal routine used to update fmeter.
1146 * A common data structure is passed to each of these routines,
1147 * which is used to keep track of the state required to manage the
1148 * frequency meter and its digital filter.
1150 * The filter works on the number of events marked per unit time.
1151 * The filter is single-pole low-pass recursive (IIR). The time unit
1152 * is 1 second. Arithmetic is done using 32-bit integers scaled to
1153 * simulate 3 decimal digits of precision (multiplied by 1000).
1155 * With an FM_COEF of 933, and a time base of 1 second, the filter
1156 * has a half-life of 10 seconds, meaning that if the events quit
1157 * happening, then the rate returned from the fmeter_getrate()
1158 * will be cut in half each 10 seconds, until it converges to zero.
1160 * It is not worth doing a real infinitely recursive filter. If more
1161 * than FM_MAXTICKS ticks have elapsed since the last filter event,
1162 * just compute FM_MAXTICKS ticks worth, by which point the level
1165 * Limit the count of unprocessed events to FM_MAXCNT, so as to avoid
1166 * arithmetic overflow in the fmeter_update() routine.
1168 * Given the simple 32 bit integer arithmetic used, this meter works
1169 * best for reporting rates between one per millisecond (msec) and
1170 * one per 32 (approx) seconds. At constant rates faster than one
1171 * per msec it maxes out at values just under 1,000,000. At constant
1172 * rates between one per msec, and one per second it will stabilize
1173 * to a value N*1000, where N is the rate of events per second.
1174 * At constant rates between one per second and one per 32 seconds,
1175 * it will be choppy, moving up on the seconds that have an event,
1176 * and then decaying until the next event. At rates slower than
1177 * about one in 32 seconds, it decays all the way back to zero between
1181 #define FM_COEF 933 /* coefficient for half-life of 10 secs */
1182 #define FM_MAXTICKS ((time_t)99) /* useless computing more ticks than this */
1183 #define FM_MAXCNT 1000000 /* limit cnt to avoid overflow */
1184 #define FM_SCALE 1000 /* faux fixed point scale */
1186 /* Initialize a frequency meter */
1187 static void fmeter_init(struct fmeter *fmp)
1192 spin_lock_init(&fmp->lock);
1195 /* Internal meter update - process cnt events and update value */
1196 static void fmeter_update(struct fmeter *fmp)
1198 time_t now = get_seconds();
1199 time_t ticks = now - fmp->time;
1204 ticks = min(FM_MAXTICKS, ticks);
1206 fmp->val = (FM_COEF * fmp->val) / FM_SCALE;
1209 fmp->val += ((FM_SCALE - FM_COEF) * fmp->cnt) / FM_SCALE;
1213 /* Process any previous ticks, then bump cnt by one (times scale). */
1214 static void fmeter_markevent(struct fmeter *fmp)
1216 spin_lock(&fmp->lock);
1218 fmp->cnt = min(FM_MAXCNT, fmp->cnt + FM_SCALE);
1219 spin_unlock(&fmp->lock);
1222 /* Process any previous ticks, then return current value. */
1223 static int fmeter_getrate(struct fmeter *fmp)
1227 spin_lock(&fmp->lock);
1230 spin_unlock(&fmp->lock);
1234 /* Called by cgroups to determine if a cpuset is usable; cgroup_mutex held */
1235 static int cpuset_can_attach(struct cgroup_subsys *ss,
1236 struct cgroup *cont, struct task_struct *tsk)
1238 struct cpuset *cs = cgroup_cs(cont);
1240 if (cpus_empty(cs->cpus_allowed) || nodes_empty(cs->mems_allowed))
1242 if (tsk->flags & PF_THREAD_BOUND) {
1245 mutex_lock(&callback_mutex);
1246 mask = cs->cpus_allowed;
1247 mutex_unlock(&callback_mutex);
1248 if (!cpus_equal(tsk->cpus_allowed, mask))
1252 return security_task_setscheduler(tsk, 0, NULL);
1255 static void cpuset_attach(struct cgroup_subsys *ss,
1256 struct cgroup *cont, struct cgroup *oldcont,
1257 struct task_struct *tsk)
1260 nodemask_t from, to;
1261 struct mm_struct *mm;
1262 struct cpuset *cs = cgroup_cs(cont);
1263 struct cpuset *oldcs = cgroup_cs(oldcont);
1266 mutex_lock(&callback_mutex);
1267 guarantee_online_cpus(cs, &cpus);
1268 err = set_cpus_allowed_ptr(tsk, &cpus);
1269 mutex_unlock(&callback_mutex);
1273 from = oldcs->mems_allowed;
1274 to = cs->mems_allowed;
1275 mm = get_task_mm(tsk);
1277 mpol_rebind_mm(mm, &to);
1278 if (is_memory_migrate(cs))
1279 cpuset_migrate_mm(mm, &from, &to);
1285 /* The various types of files and directories in a cpuset file system */
1288 FILE_MEMORY_MIGRATE,
1294 FILE_SCHED_LOAD_BALANCE,
1295 FILE_SCHED_RELAX_DOMAIN_LEVEL,
1296 FILE_MEMORY_PRESSURE_ENABLED,
1297 FILE_MEMORY_PRESSURE,
1300 } cpuset_filetype_t;
1302 static int cpuset_write_u64(struct cgroup *cgrp, struct cftype *cft, u64 val)
1305 struct cpuset *cs = cgroup_cs(cgrp);
1306 cpuset_filetype_t type = cft->private;
1308 if (!cgroup_lock_live_group(cgrp))
1312 case FILE_CPU_EXCLUSIVE:
1313 retval = update_flag(CS_CPU_EXCLUSIVE, cs, val);
1315 case FILE_MEM_EXCLUSIVE:
1316 retval = update_flag(CS_MEM_EXCLUSIVE, cs, val);
1318 case FILE_MEM_HARDWALL:
1319 retval = update_flag(CS_MEM_HARDWALL, cs, val);
1321 case FILE_SCHED_LOAD_BALANCE:
1322 retval = update_flag(CS_SCHED_LOAD_BALANCE, cs, val);
1324 case FILE_MEMORY_MIGRATE:
1325 retval = update_flag(CS_MEMORY_MIGRATE, cs, val);
1327 case FILE_MEMORY_PRESSURE_ENABLED:
1328 cpuset_memory_pressure_enabled = !!val;
1330 case FILE_MEMORY_PRESSURE:
1333 case FILE_SPREAD_PAGE:
1334 retval = update_flag(CS_SPREAD_PAGE, cs, val);
1335 cs->mems_generation = cpuset_mems_generation++;
1337 case FILE_SPREAD_SLAB:
1338 retval = update_flag(CS_SPREAD_SLAB, cs, val);
1339 cs->mems_generation = cpuset_mems_generation++;
1349 static int cpuset_write_s64(struct cgroup *cgrp, struct cftype *cft, s64 val)
1352 struct cpuset *cs = cgroup_cs(cgrp);
1353 cpuset_filetype_t type = cft->private;
1355 if (!cgroup_lock_live_group(cgrp))
1359 case FILE_SCHED_RELAX_DOMAIN_LEVEL:
1360 retval = update_relax_domain_level(cs, val);
1371 * Common handling for a write to a "cpus" or "mems" file.
1373 static int cpuset_write_resmask(struct cgroup *cgrp, struct cftype *cft,
1378 if (!cgroup_lock_live_group(cgrp))
1381 switch (cft->private) {
1383 retval = update_cpumask(cgroup_cs(cgrp), buf);
1386 retval = update_nodemask(cgroup_cs(cgrp), buf);
1397 * These ascii lists should be read in a single call, by using a user
1398 * buffer large enough to hold the entire map. If read in smaller
1399 * chunks, there is no guarantee of atomicity. Since the display format
1400 * used, list of ranges of sequential numbers, is variable length,
1401 * and since these maps can change value dynamically, one could read
1402 * gibberish by doing partial reads while a list was changing.
1403 * A single large read to a buffer that crosses a page boundary is
1404 * ok, because the result being copied to user land is not recomputed
1405 * across a page fault.
1408 static int cpuset_sprintf_cpulist(char *page, struct cpuset *cs)
1412 mutex_lock(&callback_mutex);
1413 mask = cs->cpus_allowed;
1414 mutex_unlock(&callback_mutex);
1416 return cpulist_scnprintf(page, PAGE_SIZE, mask);
1419 static int cpuset_sprintf_memlist(char *page, struct cpuset *cs)
1423 mutex_lock(&callback_mutex);
1424 mask = cs->mems_allowed;
1425 mutex_unlock(&callback_mutex);
1427 return nodelist_scnprintf(page, PAGE_SIZE, mask);
1430 static ssize_t cpuset_common_file_read(struct cgroup *cont,
1434 size_t nbytes, loff_t *ppos)
1436 struct cpuset *cs = cgroup_cs(cont);
1437 cpuset_filetype_t type = cft->private;
1442 if (!(page = (char *)__get_free_page(GFP_TEMPORARY)))
1449 s += cpuset_sprintf_cpulist(s, cs);
1452 s += cpuset_sprintf_memlist(s, cs);
1460 retval = simple_read_from_buffer(buf, nbytes, ppos, page, s - page);
1462 free_page((unsigned long)page);
1466 static u64 cpuset_read_u64(struct cgroup *cont, struct cftype *cft)
1468 struct cpuset *cs = cgroup_cs(cont);
1469 cpuset_filetype_t type = cft->private;
1471 case FILE_CPU_EXCLUSIVE:
1472 return is_cpu_exclusive(cs);
1473 case FILE_MEM_EXCLUSIVE:
1474 return is_mem_exclusive(cs);
1475 case FILE_MEM_HARDWALL:
1476 return is_mem_hardwall(cs);
1477 case FILE_SCHED_LOAD_BALANCE:
1478 return is_sched_load_balance(cs);
1479 case FILE_MEMORY_MIGRATE:
1480 return is_memory_migrate(cs);
1481 case FILE_MEMORY_PRESSURE_ENABLED:
1482 return cpuset_memory_pressure_enabled;
1483 case FILE_MEMORY_PRESSURE:
1484 return fmeter_getrate(&cs->fmeter);
1485 case FILE_SPREAD_PAGE:
1486 return is_spread_page(cs);
1487 case FILE_SPREAD_SLAB:
1488 return is_spread_slab(cs);
1494 static s64 cpuset_read_s64(struct cgroup *cont, struct cftype *cft)
1496 struct cpuset *cs = cgroup_cs(cont);
1497 cpuset_filetype_t type = cft->private;
1499 case FILE_SCHED_RELAX_DOMAIN_LEVEL:
1500 return cs->relax_domain_level;
1508 * for the common functions, 'private' gives the type of file
1511 static struct cftype files[] = {
1514 .read = cpuset_common_file_read,
1515 .write_string = cpuset_write_resmask,
1516 .max_write_len = (100U + 6 * NR_CPUS),
1517 .private = FILE_CPULIST,
1522 .read = cpuset_common_file_read,
1523 .write_string = cpuset_write_resmask,
1524 .max_write_len = (100U + 6 * MAX_NUMNODES),
1525 .private = FILE_MEMLIST,
1529 .name = "cpu_exclusive",
1530 .read_u64 = cpuset_read_u64,
1531 .write_u64 = cpuset_write_u64,
1532 .private = FILE_CPU_EXCLUSIVE,
1536 .name = "mem_exclusive",
1537 .read_u64 = cpuset_read_u64,
1538 .write_u64 = cpuset_write_u64,
1539 .private = FILE_MEM_EXCLUSIVE,
1543 .name = "mem_hardwall",
1544 .read_u64 = cpuset_read_u64,
1545 .write_u64 = cpuset_write_u64,
1546 .private = FILE_MEM_HARDWALL,
1550 .name = "sched_load_balance",
1551 .read_u64 = cpuset_read_u64,
1552 .write_u64 = cpuset_write_u64,
1553 .private = FILE_SCHED_LOAD_BALANCE,
1557 .name = "sched_relax_domain_level",
1558 .read_s64 = cpuset_read_s64,
1559 .write_s64 = cpuset_write_s64,
1560 .private = FILE_SCHED_RELAX_DOMAIN_LEVEL,
1564 .name = "memory_migrate",
1565 .read_u64 = cpuset_read_u64,
1566 .write_u64 = cpuset_write_u64,
1567 .private = FILE_MEMORY_MIGRATE,
1571 .name = "memory_pressure",
1572 .read_u64 = cpuset_read_u64,
1573 .write_u64 = cpuset_write_u64,
1574 .private = FILE_MEMORY_PRESSURE,
1578 .name = "memory_spread_page",
1579 .read_u64 = cpuset_read_u64,
1580 .write_u64 = cpuset_write_u64,
1581 .private = FILE_SPREAD_PAGE,
1585 .name = "memory_spread_slab",
1586 .read_u64 = cpuset_read_u64,
1587 .write_u64 = cpuset_write_u64,
1588 .private = FILE_SPREAD_SLAB,
1592 static struct cftype cft_memory_pressure_enabled = {
1593 .name = "memory_pressure_enabled",
1594 .read_u64 = cpuset_read_u64,
1595 .write_u64 = cpuset_write_u64,
1596 .private = FILE_MEMORY_PRESSURE_ENABLED,
1599 static int cpuset_populate(struct cgroup_subsys *ss, struct cgroup *cont)
1603 err = cgroup_add_files(cont, ss, files, ARRAY_SIZE(files));
1606 /* memory_pressure_enabled is in root cpuset only */
1608 err = cgroup_add_file(cont, ss,
1609 &cft_memory_pressure_enabled);
1614 * post_clone() is called at the end of cgroup_clone().
1615 * 'cgroup' was just created automatically as a result of
1616 * a cgroup_clone(), and the current task is about to
1617 * be moved into 'cgroup'.
1619 * Currently we refuse to set up the cgroup - thereby
1620 * refusing the task to be entered, and as a result refusing
1621 * the sys_unshare() or clone() which initiated it - if any
1622 * sibling cpusets have exclusive cpus or mem.
1624 * If this becomes a problem for some users who wish to
1625 * allow that scenario, then cpuset_post_clone() could be
1626 * changed to grant parent->cpus_allowed-sibling_cpus_exclusive
1627 * (and likewise for mems) to the new cgroup. Called with cgroup_mutex
1630 static void cpuset_post_clone(struct cgroup_subsys *ss,
1631 struct cgroup *cgroup)
1633 struct cgroup *parent, *child;
1634 struct cpuset *cs, *parent_cs;
1636 parent = cgroup->parent;
1637 list_for_each_entry(child, &parent->children, sibling) {
1638 cs = cgroup_cs(child);
1639 if (is_mem_exclusive(cs) || is_cpu_exclusive(cs))
1642 cs = cgroup_cs(cgroup);
1643 parent_cs = cgroup_cs(parent);
1645 cs->mems_allowed = parent_cs->mems_allowed;
1646 cs->cpus_allowed = parent_cs->cpus_allowed;
1651 * cpuset_create - create a cpuset
1652 * ss: cpuset cgroup subsystem
1653 * cont: control group that the new cpuset will be part of
1656 static struct cgroup_subsys_state *cpuset_create(
1657 struct cgroup_subsys *ss,
1658 struct cgroup *cont)
1661 struct cpuset *parent;
1663 if (!cont->parent) {
1664 /* This is early initialization for the top cgroup */
1665 top_cpuset.mems_generation = cpuset_mems_generation++;
1666 return &top_cpuset.css;
1668 parent = cgroup_cs(cont->parent);
1669 cs = kmalloc(sizeof(*cs), GFP_KERNEL);
1671 return ERR_PTR(-ENOMEM);
1673 cpuset_update_task_memory_state();
1675 if (is_spread_page(parent))
1676 set_bit(CS_SPREAD_PAGE, &cs->flags);
1677 if (is_spread_slab(parent))
1678 set_bit(CS_SPREAD_SLAB, &cs->flags);
1679 set_bit(CS_SCHED_LOAD_BALANCE, &cs->flags);
1680 cpus_clear(cs->cpus_allowed);
1681 nodes_clear(cs->mems_allowed);
1682 cs->mems_generation = cpuset_mems_generation++;
1683 fmeter_init(&cs->fmeter);
1684 cs->relax_domain_level = -1;
1686 cs->parent = parent;
1687 number_of_cpusets++;
1692 * Locking note on the strange update_flag() call below:
1694 * If the cpuset being removed has its flag 'sched_load_balance'
1695 * enabled, then simulate turning sched_load_balance off, which
1696 * will call rebuild_sched_domains(). The get_online_cpus()
1697 * call in rebuild_sched_domains() must not be made while holding
1698 * callback_mutex. Elsewhere the kernel nests callback_mutex inside
1699 * get_online_cpus() calls. So the reverse nesting would risk an
1703 static void cpuset_destroy(struct cgroup_subsys *ss, struct cgroup *cont)
1705 struct cpuset *cs = cgroup_cs(cont);
1707 cpuset_update_task_memory_state();
1709 if (is_sched_load_balance(cs))
1710 update_flag(CS_SCHED_LOAD_BALANCE, cs, 0);
1712 number_of_cpusets--;
1716 struct cgroup_subsys cpuset_subsys = {
1718 .create = cpuset_create,
1719 .destroy = cpuset_destroy,
1720 .can_attach = cpuset_can_attach,
1721 .attach = cpuset_attach,
1722 .populate = cpuset_populate,
1723 .post_clone = cpuset_post_clone,
1724 .subsys_id = cpuset_subsys_id,
1729 * cpuset_init_early - just enough so that the calls to
1730 * cpuset_update_task_memory_state() in early init code
1734 int __init cpuset_init_early(void)
1736 top_cpuset.mems_generation = cpuset_mems_generation++;
1742 * cpuset_init - initialize cpusets at system boot
1744 * Description: Initialize top_cpuset and the cpuset internal file system,
1747 int __init cpuset_init(void)
1751 cpus_setall(top_cpuset.cpus_allowed);
1752 nodes_setall(top_cpuset.mems_allowed);
1754 fmeter_init(&top_cpuset.fmeter);
1755 top_cpuset.mems_generation = cpuset_mems_generation++;
1756 set_bit(CS_SCHED_LOAD_BALANCE, &top_cpuset.flags);
1757 top_cpuset.relax_domain_level = -1;
1759 err = register_filesystem(&cpuset_fs_type);
1763 number_of_cpusets = 1;
1768 * cpuset_do_move_task - move a given task to another cpuset
1769 * @tsk: pointer to task_struct the task to move
1770 * @scan: struct cgroup_scanner contained in its struct cpuset_hotplug_scanner
1772 * Called by cgroup_scan_tasks() for each task in a cgroup.
1773 * Return nonzero to stop the walk through the tasks.
1775 static void cpuset_do_move_task(struct task_struct *tsk,
1776 struct cgroup_scanner *scan)
1778 struct cpuset_hotplug_scanner *chsp;
1780 chsp = container_of(scan, struct cpuset_hotplug_scanner, scan);
1781 cgroup_attach_task(chsp->to, tsk);
1785 * move_member_tasks_to_cpuset - move tasks from one cpuset to another
1786 * @from: cpuset in which the tasks currently reside
1787 * @to: cpuset to which the tasks will be moved
1789 * Called with cgroup_mutex held
1790 * callback_mutex must not be held, as cpuset_attach() will take it.
1792 * The cgroup_scan_tasks() function will scan all the tasks in a cgroup,
1793 * calling callback functions for each.
1795 static void move_member_tasks_to_cpuset(struct cpuset *from, struct cpuset *to)
1797 struct cpuset_hotplug_scanner scan;
1799 scan.scan.cg = from->css.cgroup;
1800 scan.scan.test_task = NULL; /* select all tasks in cgroup */
1801 scan.scan.process_task = cpuset_do_move_task;
1802 scan.scan.heap = NULL;
1803 scan.to = to->css.cgroup;
1805 if (cgroup_scan_tasks((struct cgroup_scanner *)&scan))
1806 printk(KERN_ERR "move_member_tasks_to_cpuset: "
1807 "cgroup_scan_tasks failed\n");
1811 * If common_cpu_mem_hotplug_unplug(), below, unplugs any CPUs
1812 * or memory nodes, we need to walk over the cpuset hierarchy,
1813 * removing that CPU or node from all cpusets. If this removes the
1814 * last CPU or node from a cpuset, then move the tasks in the empty
1815 * cpuset to its next-highest non-empty parent.
1817 * Called with cgroup_mutex held
1818 * callback_mutex must not be held, as cpuset_attach() will take it.
1820 static void remove_tasks_in_empty_cpuset(struct cpuset *cs)
1822 struct cpuset *parent;
1825 * The cgroup's css_sets list is in use if there are tasks
1826 * in the cpuset; the list is empty if there are none;
1827 * the cs->css.refcnt seems always 0.
1829 if (list_empty(&cs->css.cgroup->css_sets))
1833 * Find its next-highest non-empty parent, (top cpuset
1834 * has online cpus, so can't be empty).
1836 parent = cs->parent;
1837 while (cpus_empty(parent->cpus_allowed) ||
1838 nodes_empty(parent->mems_allowed))
1839 parent = parent->parent;
1841 move_member_tasks_to_cpuset(cs, parent);
1845 * Walk the specified cpuset subtree and look for empty cpusets.
1846 * The tasks of such cpuset must be moved to a parent cpuset.
1848 * Called with cgroup_mutex held. We take callback_mutex to modify
1849 * cpus_allowed and mems_allowed.
1851 * This walk processes the tree from top to bottom, completing one layer
1852 * before dropping down to the next. It always processes a node before
1853 * any of its children.
1855 * For now, since we lack memory hot unplug, we'll never see a cpuset
1856 * that has tasks along with an empty 'mems'. But if we did see such
1857 * a cpuset, we'd handle it just like we do if its 'cpus' was empty.
1859 static void scan_for_empty_cpusets(const struct cpuset *root)
1861 struct cpuset *cp; /* scans cpusets being updated */
1862 struct cpuset *child; /* scans child cpusets of cp */
1863 struct list_head queue;
1864 struct cgroup *cont;
1867 INIT_LIST_HEAD(&queue);
1869 list_add_tail((struct list_head *)&root->stack_list, &queue);
1871 while (!list_empty(&queue)) {
1872 cp = container_of(queue.next, struct cpuset, stack_list);
1873 list_del(queue.next);
1874 list_for_each_entry(cont, &cp->css.cgroup->children, sibling) {
1875 child = cgroup_cs(cont);
1876 list_add_tail(&child->stack_list, &queue);
1878 cont = cp->css.cgroup;
1880 /* Continue past cpusets with all cpus, mems online */
1881 if (cpus_subset(cp->cpus_allowed, cpu_online_map) &&
1882 nodes_subset(cp->mems_allowed, node_states[N_HIGH_MEMORY]))
1885 oldmems = cp->mems_allowed;
1887 /* Remove offline cpus and mems from this cpuset. */
1888 mutex_lock(&callback_mutex);
1889 cpus_and(cp->cpus_allowed, cp->cpus_allowed, cpu_online_map);
1890 nodes_and(cp->mems_allowed, cp->mems_allowed,
1891 node_states[N_HIGH_MEMORY]);
1892 mutex_unlock(&callback_mutex);
1894 /* Move tasks from the empty cpuset to a parent */
1895 if (cpus_empty(cp->cpus_allowed) ||
1896 nodes_empty(cp->mems_allowed))
1897 remove_tasks_in_empty_cpuset(cp);
1899 update_tasks_cpumask(cp);
1900 update_tasks_nodemask(cp, &oldmems);
1906 * The cpus_allowed and mems_allowed nodemasks in the top_cpuset track
1907 * cpu_online_map and node_states[N_HIGH_MEMORY]. Force the top cpuset to
1908 * track what's online after any CPU or memory node hotplug or unplug event.
1910 * Since there are two callers of this routine, one for CPU hotplug
1911 * events and one for memory node hotplug events, we could have coded
1912 * two separate routines here. We code it as a single common routine
1913 * in order to minimize text size.
1916 static void common_cpu_mem_hotplug_unplug(int rebuild_sd)
1920 top_cpuset.cpus_allowed = cpu_online_map;
1921 top_cpuset.mems_allowed = node_states[N_HIGH_MEMORY];
1922 scan_for_empty_cpusets(&top_cpuset);
1925 * Scheduler destroys domains on hotplug events.
1926 * Rebuild them based on the current settings.
1929 rebuild_sched_domains();
1935 * The top_cpuset tracks what CPUs and Memory Nodes are online,
1936 * period. This is necessary in order to make cpusets transparent
1937 * (of no affect) on systems that are actively using CPU hotplug
1938 * but making no active use of cpusets.
1940 * This routine ensures that top_cpuset.cpus_allowed tracks
1941 * cpu_online_map on each CPU hotplug (cpuhp) event.
1944 static int cpuset_handle_cpuhp(struct notifier_block *unused_nb,
1945 unsigned long phase, void *unused_cpu)
1948 case CPU_UP_CANCELED:
1949 case CPU_UP_CANCELED_FROZEN:
1950 case CPU_DOWN_FAILED:
1951 case CPU_DOWN_FAILED_FROZEN:
1953 case CPU_ONLINE_FROZEN:
1955 case CPU_DEAD_FROZEN:
1956 common_cpu_mem_hotplug_unplug(1);
1965 #ifdef CONFIG_MEMORY_HOTPLUG
1967 * Keep top_cpuset.mems_allowed tracking node_states[N_HIGH_MEMORY].
1968 * Call this routine anytime after you change
1969 * node_states[N_HIGH_MEMORY].
1970 * See also the previous routine cpuset_handle_cpuhp().
1973 void cpuset_track_online_nodes(void)
1975 common_cpu_mem_hotplug_unplug(0);
1980 * cpuset_init_smp - initialize cpus_allowed
1982 * Description: Finish top cpuset after cpu, node maps are initialized
1985 void __init cpuset_init_smp(void)
1987 top_cpuset.cpus_allowed = cpu_online_map;
1988 top_cpuset.mems_allowed = node_states[N_HIGH_MEMORY];
1990 hotcpu_notifier(cpuset_handle_cpuhp, 0);
1994 * cpuset_cpus_allowed - return cpus_allowed mask from a tasks cpuset.
1995 * @tsk: pointer to task_struct from which to obtain cpuset->cpus_allowed.
1996 * @pmask: pointer to cpumask_t variable to receive cpus_allowed set.
1998 * Description: Returns the cpumask_t cpus_allowed of the cpuset
1999 * attached to the specified @tsk. Guaranteed to return some non-empty
2000 * subset of cpu_online_map, even if this means going outside the
2004 void cpuset_cpus_allowed(struct task_struct *tsk, cpumask_t *pmask)
2006 mutex_lock(&callback_mutex);
2007 cpuset_cpus_allowed_locked(tsk, pmask);
2008 mutex_unlock(&callback_mutex);
2012 * cpuset_cpus_allowed_locked - return cpus_allowed mask from a tasks cpuset.
2013 * Must be called with callback_mutex held.
2015 void cpuset_cpus_allowed_locked(struct task_struct *tsk, cpumask_t *pmask)
2018 guarantee_online_cpus(task_cs(tsk), pmask);
2022 void cpuset_init_current_mems_allowed(void)
2024 nodes_setall(current->mems_allowed);
2028 * cpuset_mems_allowed - return mems_allowed mask from a tasks cpuset.
2029 * @tsk: pointer to task_struct from which to obtain cpuset->mems_allowed.
2031 * Description: Returns the nodemask_t mems_allowed of the cpuset
2032 * attached to the specified @tsk. Guaranteed to return some non-empty
2033 * subset of node_states[N_HIGH_MEMORY], even if this means going outside the
2037 nodemask_t cpuset_mems_allowed(struct task_struct *tsk)
2041 mutex_lock(&callback_mutex);
2043 guarantee_online_mems(task_cs(tsk), &mask);
2045 mutex_unlock(&callback_mutex);
2051 * cpuset_nodemask_valid_mems_allowed - check nodemask vs. curremt mems_allowed
2052 * @nodemask: the nodemask to be checked
2054 * Are any of the nodes in the nodemask allowed in current->mems_allowed?
2056 int cpuset_nodemask_valid_mems_allowed(nodemask_t *nodemask)
2058 return nodes_intersects(*nodemask, current->mems_allowed);
2062 * nearest_hardwall_ancestor() - Returns the nearest mem_exclusive or
2063 * mem_hardwall ancestor to the specified cpuset. Call holding
2064 * callback_mutex. If no ancestor is mem_exclusive or mem_hardwall
2065 * (an unusual configuration), then returns the root cpuset.
2067 static const struct cpuset *nearest_hardwall_ancestor(const struct cpuset *cs)
2069 while (!(is_mem_exclusive(cs) || is_mem_hardwall(cs)) && cs->parent)
2075 * cpuset_zone_allowed_softwall - Can we allocate on zone z's memory node?
2076 * @z: is this zone on an allowed node?
2077 * @gfp_mask: memory allocation flags
2079 * If we're in interrupt, yes, we can always allocate. If
2080 * __GFP_THISNODE is set, yes, we can always allocate. If zone
2081 * z's node is in our tasks mems_allowed, yes. If it's not a
2082 * __GFP_HARDWALL request and this zone's nodes is in the nearest
2083 * hardwalled cpuset ancestor to this tasks cpuset, yes.
2084 * If the task has been OOM killed and has access to memory reserves
2085 * as specified by the TIF_MEMDIE flag, yes.
2088 * If __GFP_HARDWALL is set, cpuset_zone_allowed_softwall()
2089 * reduces to cpuset_zone_allowed_hardwall(). Otherwise,
2090 * cpuset_zone_allowed_softwall() might sleep, and might allow a zone
2091 * from an enclosing cpuset.
2093 * cpuset_zone_allowed_hardwall() only handles the simpler case of
2094 * hardwall cpusets, and never sleeps.
2096 * The __GFP_THISNODE placement logic is really handled elsewhere,
2097 * by forcibly using a zonelist starting at a specified node, and by
2098 * (in get_page_from_freelist()) refusing to consider the zones for
2099 * any node on the zonelist except the first. By the time any such
2100 * calls get to this routine, we should just shut up and say 'yes'.
2102 * GFP_USER allocations are marked with the __GFP_HARDWALL bit,
2103 * and do not allow allocations outside the current tasks cpuset
2104 * unless the task has been OOM killed as is marked TIF_MEMDIE.
2105 * GFP_KERNEL allocations are not so marked, so can escape to the
2106 * nearest enclosing hardwalled ancestor cpuset.
2108 * Scanning up parent cpusets requires callback_mutex. The
2109 * __alloc_pages() routine only calls here with __GFP_HARDWALL bit
2110 * _not_ set if it's a GFP_KERNEL allocation, and all nodes in the
2111 * current tasks mems_allowed came up empty on the first pass over
2112 * the zonelist. So only GFP_KERNEL allocations, if all nodes in the
2113 * cpuset are short of memory, might require taking the callback_mutex
2116 * The first call here from mm/page_alloc:get_page_from_freelist()
2117 * has __GFP_HARDWALL set in gfp_mask, enforcing hardwall cpusets,
2118 * so no allocation on a node outside the cpuset is allowed (unless
2119 * in interrupt, of course).
2121 * The second pass through get_page_from_freelist() doesn't even call
2122 * here for GFP_ATOMIC calls. For those calls, the __alloc_pages()
2123 * variable 'wait' is not set, and the bit ALLOC_CPUSET is not set
2124 * in alloc_flags. That logic and the checks below have the combined
2126 * in_interrupt - any node ok (current task context irrelevant)
2127 * GFP_ATOMIC - any node ok
2128 * TIF_MEMDIE - any node ok
2129 * GFP_KERNEL - any node in enclosing hardwalled cpuset ok
2130 * GFP_USER - only nodes in current tasks mems allowed ok.
2133 * Don't call cpuset_zone_allowed_softwall if you can't sleep, unless you
2134 * pass in the __GFP_HARDWALL flag set in gfp_flag, which disables
2135 * the code that might scan up ancestor cpusets and sleep.
2138 int __cpuset_zone_allowed_softwall(struct zone *z, gfp_t gfp_mask)
2140 int node; /* node that zone z is on */
2141 const struct cpuset *cs; /* current cpuset ancestors */
2142 int allowed; /* is allocation in zone z allowed? */
2144 if (in_interrupt() || (gfp_mask & __GFP_THISNODE))
2146 node = zone_to_nid(z);
2147 might_sleep_if(!(gfp_mask & __GFP_HARDWALL));
2148 if (node_isset(node, current->mems_allowed))
2151 * Allow tasks that have access to memory reserves because they have
2152 * been OOM killed to get memory anywhere.
2154 if (unlikely(test_thread_flag(TIF_MEMDIE)))
2156 if (gfp_mask & __GFP_HARDWALL) /* If hardwall request, stop here */
2159 if (current->flags & PF_EXITING) /* Let dying task have memory */
2162 /* Not hardwall and node outside mems_allowed: scan up cpusets */
2163 mutex_lock(&callback_mutex);
2166 cs = nearest_hardwall_ancestor(task_cs(current));
2167 task_unlock(current);
2169 allowed = node_isset(node, cs->mems_allowed);
2170 mutex_unlock(&callback_mutex);
2175 * cpuset_zone_allowed_hardwall - Can we allocate on zone z's memory node?
2176 * @z: is this zone on an allowed node?
2177 * @gfp_mask: memory allocation flags
2179 * If we're in interrupt, yes, we can always allocate.
2180 * If __GFP_THISNODE is set, yes, we can always allocate. If zone
2181 * z's node is in our tasks mems_allowed, yes. If the task has been
2182 * OOM killed and has access to memory reserves as specified by the
2183 * TIF_MEMDIE flag, yes. Otherwise, no.
2185 * The __GFP_THISNODE placement logic is really handled elsewhere,
2186 * by forcibly using a zonelist starting at a specified node, and by
2187 * (in get_page_from_freelist()) refusing to consider the zones for
2188 * any node on the zonelist except the first. By the time any such
2189 * calls get to this routine, we should just shut up and say 'yes'.
2191 * Unlike the cpuset_zone_allowed_softwall() variant, above,
2192 * this variant requires that the zone be in the current tasks
2193 * mems_allowed or that we're in interrupt. It does not scan up the
2194 * cpuset hierarchy for the nearest enclosing mem_exclusive cpuset.
2198 int __cpuset_zone_allowed_hardwall(struct zone *z, gfp_t gfp_mask)
2200 int node; /* node that zone z is on */
2202 if (in_interrupt() || (gfp_mask & __GFP_THISNODE))
2204 node = zone_to_nid(z);
2205 if (node_isset(node, current->mems_allowed))
2208 * Allow tasks that have access to memory reserves because they have
2209 * been OOM killed to get memory anywhere.
2211 if (unlikely(test_thread_flag(TIF_MEMDIE)))
2217 * cpuset_lock - lock out any changes to cpuset structures
2219 * The out of memory (oom) code needs to mutex_lock cpusets
2220 * from being changed while it scans the tasklist looking for a
2221 * task in an overlapping cpuset. Expose callback_mutex via this
2222 * cpuset_lock() routine, so the oom code can lock it, before
2223 * locking the task list. The tasklist_lock is a spinlock, so
2224 * must be taken inside callback_mutex.
2227 void cpuset_lock(void)
2229 mutex_lock(&callback_mutex);
2233 * cpuset_unlock - release lock on cpuset changes
2235 * Undo the lock taken in a previous cpuset_lock() call.
2238 void cpuset_unlock(void)
2240 mutex_unlock(&callback_mutex);
2244 * cpuset_mem_spread_node() - On which node to begin search for a page
2246 * If a task is marked PF_SPREAD_PAGE or PF_SPREAD_SLAB (as for
2247 * tasks in a cpuset with is_spread_page or is_spread_slab set),
2248 * and if the memory allocation used cpuset_mem_spread_node()
2249 * to determine on which node to start looking, as it will for
2250 * certain page cache or slab cache pages such as used for file
2251 * system buffers and inode caches, then instead of starting on the
2252 * local node to look for a free page, rather spread the starting
2253 * node around the tasks mems_allowed nodes.
2255 * We don't have to worry about the returned node being offline
2256 * because "it can't happen", and even if it did, it would be ok.
2258 * The routines calling guarantee_online_mems() are careful to
2259 * only set nodes in task->mems_allowed that are online. So it
2260 * should not be possible for the following code to return an
2261 * offline node. But if it did, that would be ok, as this routine
2262 * is not returning the node where the allocation must be, only
2263 * the node where the search should start. The zonelist passed to
2264 * __alloc_pages() will include all nodes. If the slab allocator
2265 * is passed an offline node, it will fall back to the local node.
2266 * See kmem_cache_alloc_node().
2269 int cpuset_mem_spread_node(void)
2273 node = next_node(current->cpuset_mem_spread_rotor, current->mems_allowed);
2274 if (node == MAX_NUMNODES)
2275 node = first_node(current->mems_allowed);
2276 current->cpuset_mem_spread_rotor = node;
2279 EXPORT_SYMBOL_GPL(cpuset_mem_spread_node);
2282 * cpuset_mems_allowed_intersects - Does @tsk1's mems_allowed intersect @tsk2's?
2283 * @tsk1: pointer to task_struct of some task.
2284 * @tsk2: pointer to task_struct of some other task.
2286 * Description: Return true if @tsk1's mems_allowed intersects the
2287 * mems_allowed of @tsk2. Used by the OOM killer to determine if
2288 * one of the task's memory usage might impact the memory available
2292 int cpuset_mems_allowed_intersects(const struct task_struct *tsk1,
2293 const struct task_struct *tsk2)
2295 return nodes_intersects(tsk1->mems_allowed, tsk2->mems_allowed);
2299 * Collection of memory_pressure is suppressed unless
2300 * this flag is enabled by writing "1" to the special
2301 * cpuset file 'memory_pressure_enabled' in the root cpuset.
2304 int cpuset_memory_pressure_enabled __read_mostly;
2307 * cpuset_memory_pressure_bump - keep stats of per-cpuset reclaims.
2309 * Keep a running average of the rate of synchronous (direct)
2310 * page reclaim efforts initiated by tasks in each cpuset.
2312 * This represents the rate at which some task in the cpuset
2313 * ran low on memory on all nodes it was allowed to use, and
2314 * had to enter the kernels page reclaim code in an effort to
2315 * create more free memory by tossing clean pages or swapping
2316 * or writing dirty pages.
2318 * Display to user space in the per-cpuset read-only file
2319 * "memory_pressure". Value displayed is an integer
2320 * representing the recent rate of entry into the synchronous
2321 * (direct) page reclaim by any task attached to the cpuset.
2324 void __cpuset_memory_pressure_bump(void)
2327 fmeter_markevent(&task_cs(current)->fmeter);
2328 task_unlock(current);
2331 #ifdef CONFIG_PROC_PID_CPUSET
2333 * proc_cpuset_show()
2334 * - Print tasks cpuset path into seq_file.
2335 * - Used for /proc/<pid>/cpuset.
2336 * - No need to task_lock(tsk) on this tsk->cpuset reference, as it
2337 * doesn't really matter if tsk->cpuset changes after we read it,
2338 * and we take cgroup_mutex, keeping cpuset_attach() from changing it
2341 static int proc_cpuset_show(struct seq_file *m, void *unused_v)
2344 struct task_struct *tsk;
2346 struct cgroup_subsys_state *css;
2350 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
2356 tsk = get_pid_task(pid, PIDTYPE_PID);
2362 css = task_subsys_state(tsk, cpuset_subsys_id);
2363 retval = cgroup_path(css->cgroup, buf, PAGE_SIZE);
2370 put_task_struct(tsk);
2377 static int cpuset_open(struct inode *inode, struct file *file)
2379 struct pid *pid = PROC_I(inode)->pid;
2380 return single_open(file, proc_cpuset_show, pid);
2383 const struct file_operations proc_cpuset_operations = {
2384 .open = cpuset_open,
2386 .llseek = seq_lseek,
2387 .release = single_release,
2389 #endif /* CONFIG_PROC_PID_CPUSET */
2391 /* Display task cpus_allowed, mems_allowed in /proc/<pid>/status file. */
2392 void cpuset_task_status_allowed(struct seq_file *m, struct task_struct *task)
2394 seq_printf(m, "Cpus_allowed:\t");
2395 m->count += cpumask_scnprintf(m->buf + m->count, m->size - m->count,
2396 task->cpus_allowed);
2397 seq_printf(m, "\n");
2398 seq_printf(m, "Cpus_allowed_list:\t");
2399 m->count += cpulist_scnprintf(m->buf + m->count, m->size - m->count,
2400 task->cpus_allowed);
2401 seq_printf(m, "\n");
2402 seq_printf(m, "Mems_allowed:\t");
2403 m->count += nodemask_scnprintf(m->buf + m->count, m->size - m->count,
2404 task->mems_allowed);
2405 seq_printf(m, "\n");
2406 seq_printf(m, "Mems_allowed_list:\t");
2407 m->count += nodelist_scnprintf(m->buf + m->count, m->size - m->count,
2408 task->mems_allowed);
2409 seq_printf(m, "\n");