]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - kernel/cpuset.c
printk: clean up recursion check related static variables
[mirror_ubuntu-bionic-kernel.git] / kernel / cpuset.c
CommitLineData
1da177e4
LT
1/*
2 * kernel/cpuset.c
3 *
4 * Processor and Memory placement constraints for sets of tasks.
5 *
6 * Copyright (C) 2003 BULL SA.
029190c5 7 * Copyright (C) 2004-2007 Silicon Graphics, Inc.
8793d854 8 * Copyright (C) 2006 Google, Inc
1da177e4
LT
9 *
10 * Portions derived from Patrick Mochel's sysfs code.
11 * sysfs is Copyright (c) 2001-3 Patrick Mochel
1da177e4 12 *
825a46af 13 * 2003-10-10 Written by Simon Derr.
1da177e4 14 * 2003-10-22 Updates by Stephen Hemminger.
825a46af 15 * 2004 May-July Rework by Paul Jackson.
8793d854 16 * 2006 Rework by Paul Menage to use generic cgroups
1da177e4
LT
17 *
18 * This file is subject to the terms and conditions of the GNU General Public
19 * License. See the file COPYING in the main directory of the Linux
20 * distribution for more details.
21 */
22
1da177e4
LT
23#include <linux/cpu.h>
24#include <linux/cpumask.h>
25#include <linux/cpuset.h>
26#include <linux/err.h>
27#include <linux/errno.h>
28#include <linux/file.h>
29#include <linux/fs.h>
30#include <linux/init.h>
31#include <linux/interrupt.h>
32#include <linux/kernel.h>
33#include <linux/kmod.h>
34#include <linux/list.h>
68860ec1 35#include <linux/mempolicy.h>
1da177e4
LT
36#include <linux/mm.h>
37#include <linux/module.h>
38#include <linux/mount.h>
39#include <linux/namei.h>
40#include <linux/pagemap.h>
41#include <linux/proc_fs.h>
6b9c2603 42#include <linux/rcupdate.h>
1da177e4
LT
43#include <linux/sched.h>
44#include <linux/seq_file.h>
22fb52dd 45#include <linux/security.h>
1da177e4 46#include <linux/slab.h>
1da177e4
LT
47#include <linux/spinlock.h>
48#include <linux/stat.h>
49#include <linux/string.h>
50#include <linux/time.h>
51#include <linux/backing-dev.h>
52#include <linux/sort.h>
53
54#include <asm/uaccess.h>
55#include <asm/atomic.h>
3d3f26a7 56#include <linux/mutex.h>
029190c5 57#include <linux/kfifo.h>
956db3ca
CW
58#include <linux/workqueue.h>
59#include <linux/cgroup.h>
1da177e4 60
202f72d5
PJ
61/*
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.
65 */
7edc5962 66int number_of_cpusets __read_mostly;
202f72d5 67
2df167a3 68/* Forward declare cgroup structures */
8793d854
PM
69struct cgroup_subsys cpuset_subsys;
70struct cpuset;
71
3e0d98b9
PJ
72/* See "Frequency meter" comments, below. */
73
74struct fmeter {
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 */
79};
80
1da177e4 81struct cpuset {
8793d854
PM
82 struct cgroup_subsys_state css;
83
1da177e4
LT
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 */
87
1da177e4 88 struct cpuset *parent; /* my parent */
1da177e4
LT
89
90 /*
91 * Copy of global cpuset_mems_generation as of the most
92 * recent time this cpuset changed its mems_allowed.
93 */
3e0d98b9
PJ
94 int mems_generation;
95
96 struct fmeter fmeter; /* memory_pressure filter */
029190c5
PJ
97
98 /* partition number for rebuild_sched_domains() */
99 int pn;
956db3ca 100
1d3504fc
HS
101 /* for custom sched domain */
102 int relax_domain_level;
103
956db3ca
CW
104 /* used for walking a cpuset heirarchy */
105 struct list_head stack_list;
1da177e4
LT
106};
107
8793d854
PM
108/* Retrieve the cpuset for a cgroup */
109static inline struct cpuset *cgroup_cs(struct cgroup *cont)
110{
111 return container_of(cgroup_subsys_state(cont, cpuset_subsys_id),
112 struct cpuset, css);
113}
114
115/* Retrieve the cpuset for a task */
116static inline struct cpuset *task_cs(struct task_struct *task)
117{
118 return container_of(task_subsys_state(task, cpuset_subsys_id),
119 struct cpuset, css);
120}
956db3ca
CW
121struct cpuset_hotplug_scanner {
122 struct cgroup_scanner scan;
123 struct cgroup *to;
124};
8793d854 125
1da177e4
LT
126/* bits in struct cpuset flags field */
127typedef enum {
128 CS_CPU_EXCLUSIVE,
129 CS_MEM_EXCLUSIVE,
78608366 130 CS_MEM_HARDWALL,
45b07ef3 131 CS_MEMORY_MIGRATE,
029190c5 132 CS_SCHED_LOAD_BALANCE,
825a46af
PJ
133 CS_SPREAD_PAGE,
134 CS_SPREAD_SLAB,
1da177e4
LT
135} cpuset_flagbits_t;
136
137/* convenient tests for these bits */
138static inline int is_cpu_exclusive(const struct cpuset *cs)
139{
7b5b9ef0 140 return test_bit(CS_CPU_EXCLUSIVE, &cs->flags);
1da177e4
LT
141}
142
143static inline int is_mem_exclusive(const struct cpuset *cs)
144{
7b5b9ef0 145 return test_bit(CS_MEM_EXCLUSIVE, &cs->flags);
1da177e4
LT
146}
147
78608366
PM
148static inline int is_mem_hardwall(const struct cpuset *cs)
149{
150 return test_bit(CS_MEM_HARDWALL, &cs->flags);
151}
152
029190c5
PJ
153static inline int is_sched_load_balance(const struct cpuset *cs)
154{
155 return test_bit(CS_SCHED_LOAD_BALANCE, &cs->flags);
156}
157
45b07ef3
PJ
158static inline int is_memory_migrate(const struct cpuset *cs)
159{
7b5b9ef0 160 return test_bit(CS_MEMORY_MIGRATE, &cs->flags);
45b07ef3
PJ
161}
162
825a46af
PJ
163static inline int is_spread_page(const struct cpuset *cs)
164{
165 return test_bit(CS_SPREAD_PAGE, &cs->flags);
166}
167
168static inline int is_spread_slab(const struct cpuset *cs)
169{
170 return test_bit(CS_SPREAD_SLAB, &cs->flags);
171}
172
1da177e4 173/*
151a4420 174 * Increment this integer everytime any cpuset changes its
1da177e4
LT
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.
178 *
2df167a3 179 * A single, global generation is needed because cpuset_attach_task() could
1da177e4
LT
180 * reattach a task to a different cpuset, which must not have its
181 * generation numbers aliased with those of that tasks previous cpuset.
182 *
183 * Generations are needed for mems_allowed because one task cannot
2df167a3 184 * modify another's memory placement. So we must enable every task,
1da177e4
LT
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.
151a4420 188 *
2df167a3 189 * Since writes to cpuset_mems_generation are guarded by the cgroup lock
151a4420 190 * there is no need to mark it atomic.
1da177e4 191 */
151a4420 192static int cpuset_mems_generation;
1da177e4
LT
193
194static 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,
1da177e4
LT
198};
199
1da177e4 200/*
2df167a3
PM
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.
053199ed 208 *
3d3f26a7 209 * A task must hold both mutexes to modify cpusets. If a task
2df167a3 210 * holds cgroup_mutex, then it blocks others wanting that mutex,
3d3f26a7 211 * ensuring that it is the only task able to also acquire callback_mutex
053199ed
PJ
212 * and be able to modify cpusets. It can perform various checks on
213 * the cpuset structure first, knowing nothing will change. It can
2df167a3 214 * also allocate memory while just holding cgroup_mutex. While it is
053199ed 215 * performing these checks, various callback routines can briefly
3d3f26a7
IM
216 * acquire callback_mutex to query cpusets. Once it is ready to make
217 * the changes, it takes callback_mutex, blocking everyone else.
053199ed
PJ
218 *
219 * Calls to the kernel memory allocator can not be made while holding
3d3f26a7 220 * callback_mutex, as that would risk double tripping on callback_mutex
053199ed
PJ
221 * from one of the callbacks into the cpuset code from within
222 * __alloc_pages().
223 *
3d3f26a7 224 * If a task is only holding callback_mutex, then it has read-only
053199ed
PJ
225 * access to cpusets.
226 *
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.
229 *
053199ed 230 * The cpuset_common_file_write handler for operations that modify
2df167a3 231 * the cpuset hierarchy holds cgroup_mutex across the entire operation,
053199ed
PJ
232 * single threading all such cpuset modifications across the system.
233 *
3d3f26a7 234 * The cpuset_common_file_read() handlers only hold callback_mutex across
053199ed
PJ
235 * small pieces of code, such as when reading out possibly multi-word
236 * cpumasks and nodemasks.
237 *
2df167a3
PM
238 * Accessing a task's cpuset should be done in accordance with the
239 * guidelines for accessing subsystem state in kernel/cgroup.c
1da177e4
LT
240 */
241
3d3f26a7 242static DEFINE_MUTEX(callback_mutex);
4247bdc6 243
8793d854
PM
244/* This is ugly, but preserves the userspace API for existing cpuset
245 * users. If someone tries to mount the "cpuset" filesystem, we
246 * silently switch it to mount "cgroup" instead */
454e2398
DH
247static int cpuset_get_sb(struct file_system_type *fs_type,
248 int flags, const char *unused_dev_name,
249 void *data, struct vfsmount *mnt)
1da177e4 250{
8793d854
PM
251 struct file_system_type *cgroup_fs = get_fs_type("cgroup");
252 int ret = -ENODEV;
253 if (cgroup_fs) {
254 char mountopts[] =
255 "cpuset,noprefix,"
256 "release_agent=/sbin/cpuset_release_agent";
257 ret = cgroup_fs->get_sb(cgroup_fs, flags,
258 unused_dev_name, mountopts, mnt);
259 put_filesystem(cgroup_fs);
260 }
261 return ret;
1da177e4
LT
262}
263
264static struct file_system_type cpuset_fs_type = {
265 .name = "cpuset",
266 .get_sb = cpuset_get_sb,
1da177e4
LT
267};
268
1da177e4
LT
269/*
270 * Return in *pmask the portion of a cpusets's cpus_allowed that
271 * are online. If none are online, walk up the cpuset hierarchy
272 * until we find one that does have some online cpus. If we get
273 * all the way to the top and still haven't found any online cpus,
274 * return cpu_online_map. Or if passed a NULL cs from an exit'ing
275 * task, return cpu_online_map.
276 *
277 * One way or another, we guarantee to return some non-empty subset
278 * of cpu_online_map.
279 *
3d3f26a7 280 * Call with callback_mutex held.
1da177e4
LT
281 */
282
283static void guarantee_online_cpus(const struct cpuset *cs, cpumask_t *pmask)
284{
285 while (cs && !cpus_intersects(cs->cpus_allowed, cpu_online_map))
286 cs = cs->parent;
287 if (cs)
288 cpus_and(*pmask, cs->cpus_allowed, cpu_online_map);
289 else
290 *pmask = cpu_online_map;
291 BUG_ON(!cpus_intersects(*pmask, cpu_online_map));
292}
293
294/*
295 * Return in *pmask the portion of a cpusets's mems_allowed that
0e1e7c7a
CL
296 * are online, with memory. If none are online with memory, walk
297 * up the cpuset hierarchy until we find one that does have some
298 * online mems. If we get all the way to the top and still haven't
299 * found any online mems, return node_states[N_HIGH_MEMORY].
1da177e4
LT
300 *
301 * One way or another, we guarantee to return some non-empty subset
0e1e7c7a 302 * of node_states[N_HIGH_MEMORY].
1da177e4 303 *
3d3f26a7 304 * Call with callback_mutex held.
1da177e4
LT
305 */
306
307static void guarantee_online_mems(const struct cpuset *cs, nodemask_t *pmask)
308{
0e1e7c7a
CL
309 while (cs && !nodes_intersects(cs->mems_allowed,
310 node_states[N_HIGH_MEMORY]))
1da177e4
LT
311 cs = cs->parent;
312 if (cs)
0e1e7c7a
CL
313 nodes_and(*pmask, cs->mems_allowed,
314 node_states[N_HIGH_MEMORY]);
1da177e4 315 else
0e1e7c7a
CL
316 *pmask = node_states[N_HIGH_MEMORY];
317 BUG_ON(!nodes_intersects(*pmask, node_states[N_HIGH_MEMORY]));
1da177e4
LT
318}
319
cf2a473c
PJ
320/**
321 * cpuset_update_task_memory_state - update task memory placement
322 *
323 * If the current tasks cpusets mems_allowed changed behind our
324 * backs, update current->mems_allowed, mems_generation and task NUMA
325 * mempolicy to the new value.
053199ed 326 *
cf2a473c
PJ
327 * Task mempolicy is updated by rebinding it relative to the
328 * current->cpuset if a task has its memory placement changed.
329 * Do not call this routine if in_interrupt().
330 *
4a01c8d5 331 * Call without callback_mutex or task_lock() held. May be
2df167a3
PM
332 * called with or without cgroup_mutex held. Thanks in part to
333 * 'the_top_cpuset_hack', the task's cpuset pointer will never
41f7f60d
DR
334 * be NULL. This routine also might acquire callback_mutex during
335 * call.
053199ed 336 *
6b9c2603
PJ
337 * Reading current->cpuset->mems_generation doesn't need task_lock
338 * to guard the current->cpuset derefence, because it is guarded
2df167a3 339 * from concurrent freeing of current->cpuset using RCU.
6b9c2603
PJ
340 *
341 * The rcu_dereference() is technically probably not needed,
342 * as I don't actually mind if I see a new cpuset pointer but
343 * an old value of mems_generation. However this really only
344 * matters on alpha systems using cpusets heavily. If I dropped
345 * that rcu_dereference(), it would save them a memory barrier.
346 * For all other arch's, rcu_dereference is a no-op anyway, and for
347 * alpha systems not using cpusets, another planned optimization,
348 * avoiding the rcu critical section for tasks in the root cpuset
349 * which is statically allocated, so can't vanish, will make this
350 * irrelevant. Better to use RCU as intended, than to engage in
351 * some cute trick to save a memory barrier that is impossible to
352 * test, for alpha systems using cpusets heavily, which might not
353 * even exist.
053199ed
PJ
354 *
355 * This routine is needed to update the per-task mems_allowed data,
356 * within the tasks context, when it is trying to allocate memory
357 * (in various mm/mempolicy.c routines) and notices that some other
358 * task has been modifying its cpuset.
1da177e4
LT
359 */
360
fe85a998 361void cpuset_update_task_memory_state(void)
1da177e4 362{
053199ed 363 int my_cpusets_mem_gen;
cf2a473c 364 struct task_struct *tsk = current;
6b9c2603 365 struct cpuset *cs;
053199ed 366
8793d854 367 if (task_cs(tsk) == &top_cpuset) {
03a285f5
PJ
368 /* Don't need rcu for top_cpuset. It's never freed. */
369 my_cpusets_mem_gen = top_cpuset.mems_generation;
370 } else {
371 rcu_read_lock();
8793d854 372 my_cpusets_mem_gen = task_cs(current)->mems_generation;
03a285f5
PJ
373 rcu_read_unlock();
374 }
1da177e4 375
cf2a473c 376 if (my_cpusets_mem_gen != tsk->cpuset_mems_generation) {
3d3f26a7 377 mutex_lock(&callback_mutex);
cf2a473c 378 task_lock(tsk);
8793d854 379 cs = task_cs(tsk); /* Maybe changed when task not locked */
cf2a473c
PJ
380 guarantee_online_mems(cs, &tsk->mems_allowed);
381 tsk->cpuset_mems_generation = cs->mems_generation;
825a46af
PJ
382 if (is_spread_page(cs))
383 tsk->flags |= PF_SPREAD_PAGE;
384 else
385 tsk->flags &= ~PF_SPREAD_PAGE;
386 if (is_spread_slab(cs))
387 tsk->flags |= PF_SPREAD_SLAB;
388 else
389 tsk->flags &= ~PF_SPREAD_SLAB;
cf2a473c 390 task_unlock(tsk);
3d3f26a7 391 mutex_unlock(&callback_mutex);
74cb2155 392 mpol_rebind_task(tsk, &tsk->mems_allowed);
1da177e4
LT
393 }
394}
395
396/*
397 * is_cpuset_subset(p, q) - Is cpuset p a subset of cpuset q?
398 *
399 * One cpuset is a subset of another if all its allowed CPUs and
400 * Memory Nodes are a subset of the other, and its exclusive flags
2df167a3 401 * are only set if the other's are set. Call holding cgroup_mutex.
1da177e4
LT
402 */
403
404static int is_cpuset_subset(const struct cpuset *p, const struct cpuset *q)
405{
406 return cpus_subset(p->cpus_allowed, q->cpus_allowed) &&
407 nodes_subset(p->mems_allowed, q->mems_allowed) &&
408 is_cpu_exclusive(p) <= is_cpu_exclusive(q) &&
409 is_mem_exclusive(p) <= is_mem_exclusive(q);
410}
411
412/*
413 * validate_change() - Used to validate that any proposed cpuset change
414 * follows the structural rules for cpusets.
415 *
416 * If we replaced the flag and mask values of the current cpuset
417 * (cur) with those values in the trial cpuset (trial), would
418 * our various subset and exclusive rules still be valid? Presumes
2df167a3 419 * cgroup_mutex held.
1da177e4
LT
420 *
421 * 'cur' is the address of an actual, in-use cpuset. Operations
422 * such as list traversal that depend on the actual address of the
423 * cpuset in the list must use cur below, not trial.
424 *
425 * 'trial' is the address of bulk structure copy of cur, with
426 * perhaps one or more of the fields cpus_allowed, mems_allowed,
427 * or flags changed to new, trial values.
428 *
429 * Return 0 if valid, -errno if not.
430 */
431
432static int validate_change(const struct cpuset *cur, const struct cpuset *trial)
433{
8793d854 434 struct cgroup *cont;
1da177e4
LT
435 struct cpuset *c, *par;
436
437 /* Each of our child cpusets must be a subset of us */
8793d854
PM
438 list_for_each_entry(cont, &cur->css.cgroup->children, sibling) {
439 if (!is_cpuset_subset(cgroup_cs(cont), trial))
1da177e4
LT
440 return -EBUSY;
441 }
442
443 /* Remaining checks don't apply to root cpuset */
69604067 444 if (cur == &top_cpuset)
1da177e4
LT
445 return 0;
446
69604067
PJ
447 par = cur->parent;
448
1da177e4
LT
449 /* We must be a subset of our parent cpuset */
450 if (!is_cpuset_subset(trial, par))
451 return -EACCES;
452
2df167a3
PM
453 /*
454 * If either I or some sibling (!= me) is exclusive, we can't
455 * overlap
456 */
8793d854
PM
457 list_for_each_entry(cont, &par->css.cgroup->children, sibling) {
458 c = cgroup_cs(cont);
1da177e4
LT
459 if ((is_cpu_exclusive(trial) || is_cpu_exclusive(c)) &&
460 c != cur &&
461 cpus_intersects(trial->cpus_allowed, c->cpus_allowed))
462 return -EINVAL;
463 if ((is_mem_exclusive(trial) || is_mem_exclusive(c)) &&
464 c != cur &&
465 nodes_intersects(trial->mems_allowed, c->mems_allowed))
466 return -EINVAL;
467 }
468
020958b6
PJ
469 /* Cpusets with tasks can't have empty cpus_allowed or mems_allowed */
470 if (cgroup_task_count(cur->css.cgroup)) {
471 if (cpus_empty(trial->cpus_allowed) ||
472 nodes_empty(trial->mems_allowed)) {
473 return -ENOSPC;
474 }
475 }
476
1da177e4
LT
477 return 0;
478}
479
029190c5
PJ
480/*
481 * Helper routine for rebuild_sched_domains().
482 * Do cpusets a, b have overlapping cpus_allowed masks?
483 */
484
485static int cpusets_overlap(struct cpuset *a, struct cpuset *b)
486{
487 return cpus_intersects(a->cpus_allowed, b->cpus_allowed);
488}
489
1d3504fc
HS
490static void
491update_domain_attr(struct sched_domain_attr *dattr, struct cpuset *c)
492{
493 if (!dattr)
494 return;
495 if (dattr->relax_domain_level < c->relax_domain_level)
496 dattr->relax_domain_level = c->relax_domain_level;
497 return;
498}
499
029190c5
PJ
500/*
501 * rebuild_sched_domains()
502 *
503 * If the flag 'sched_load_balance' of any cpuset with non-empty
504 * 'cpus' changes, or if the 'cpus' allowed changes in any cpuset
505 * which has that flag enabled, or if any cpuset with a non-empty
506 * 'cpus' is removed, then call this routine to rebuild the
507 * scheduler's dynamic sched domains.
508 *
509 * This routine builds a partial partition of the systems CPUs
510 * (the set of non-overlappping cpumask_t's in the array 'part'
511 * below), and passes that partial partition to the kernel/sched.c
512 * partition_sched_domains() routine, which will rebuild the
513 * schedulers load balancing domains (sched domains) as specified
514 * by that partial partition. A 'partial partition' is a set of
515 * non-overlapping subsets whose union is a subset of that set.
516 *
517 * See "What is sched_load_balance" in Documentation/cpusets.txt
518 * for a background explanation of this.
519 *
520 * Does not return errors, on the theory that the callers of this
521 * routine would rather not worry about failures to rebuild sched
522 * domains when operating in the severe memory shortage situations
523 * that could cause allocation failures below.
524 *
525 * Call with cgroup_mutex held. May take callback_mutex during
526 * call due to the kfifo_alloc() and kmalloc() calls. May nest
86ef5c9a 527 * a call to the get_online_cpus()/put_online_cpus() pair.
029190c5 528 * Must not be called holding callback_mutex, because we must not
86ef5c9a
GS
529 * call get_online_cpus() while holding callback_mutex. Elsewhere
530 * the kernel nests callback_mutex inside get_online_cpus() calls.
029190c5
PJ
531 * So the reverse nesting would risk an ABBA deadlock.
532 *
533 * The three key local variables below are:
534 * q - a kfifo queue of cpuset pointers, used to implement a
535 * top-down scan of all cpusets. This scan loads a pointer
536 * to each cpuset marked is_sched_load_balance into the
537 * array 'csa'. For our purposes, rebuilding the schedulers
538 * sched domains, we can ignore !is_sched_load_balance cpusets.
539 * csa - (for CpuSet Array) Array of pointers to all the cpusets
540 * that need to be load balanced, for convenient iterative
541 * access by the subsequent code that finds the best partition,
542 * i.e the set of domains (subsets) of CPUs such that the
543 * cpus_allowed of every cpuset marked is_sched_load_balance
544 * is a subset of one of these domains, while there are as
545 * many such domains as possible, each as small as possible.
546 * doms - Conversion of 'csa' to an array of cpumasks, for passing to
547 * the kernel/sched.c routine partition_sched_domains() in a
548 * convenient format, that can be easily compared to the prior
549 * value to determine what partition elements (sched domains)
550 * were changed (added or removed.)
551 *
552 * Finding the best partition (set of domains):
553 * The triple nested loops below over i, j, k scan over the
554 * load balanced cpusets (using the array of cpuset pointers in
555 * csa[]) looking for pairs of cpusets that have overlapping
556 * cpus_allowed, but which don't have the same 'pn' partition
557 * number and gives them in the same partition number. It keeps
558 * looping on the 'restart' label until it can no longer find
559 * any such pairs.
560 *
561 * The union of the cpus_allowed masks from the set of
562 * all cpusets having the same 'pn' value then form the one
563 * element of the partition (one sched domain) to be passed to
564 * partition_sched_domains().
565 */
566
567static void rebuild_sched_domains(void)
568{
569 struct kfifo *q; /* queue of cpusets to be scanned */
570 struct cpuset *cp; /* scans q */
571 struct cpuset **csa; /* array of all cpuset ptrs */
572 int csn; /* how many cpuset ptrs in csa so far */
573 int i, j, k; /* indices for partition finding loops */
574 cpumask_t *doms; /* resulting partition; i.e. sched domains */
1d3504fc 575 struct sched_domain_attr *dattr; /* attributes for custom domains */
029190c5
PJ
576 int ndoms; /* number of sched domains in result */
577 int nslot; /* next empty doms[] cpumask_t slot */
578
579 q = NULL;
580 csa = NULL;
581 doms = NULL;
1d3504fc 582 dattr = NULL;
029190c5
PJ
583
584 /* Special case for the 99% of systems with one, full, sched domain */
585 if (is_sched_load_balance(&top_cpuset)) {
586 ndoms = 1;
587 doms = kmalloc(sizeof(cpumask_t), GFP_KERNEL);
588 if (!doms)
589 goto rebuild;
1d3504fc
HS
590 dattr = kmalloc(sizeof(struct sched_domain_attr), GFP_KERNEL);
591 if (dattr) {
592 *dattr = SD_ATTR_INIT;
593 update_domain_attr(dattr, &top_cpuset);
594 }
029190c5
PJ
595 *doms = top_cpuset.cpus_allowed;
596 goto rebuild;
597 }
598
599 q = kfifo_alloc(number_of_cpusets * sizeof(cp), GFP_KERNEL, NULL);
600 if (IS_ERR(q))
601 goto done;
602 csa = kmalloc(number_of_cpusets * sizeof(cp), GFP_KERNEL);
603 if (!csa)
604 goto done;
605 csn = 0;
606
607 cp = &top_cpuset;
608 __kfifo_put(q, (void *)&cp, sizeof(cp));
609 while (__kfifo_get(q, (void *)&cp, sizeof(cp))) {
610 struct cgroup *cont;
611 struct cpuset *child; /* scans child cpusets of cp */
612 if (is_sched_load_balance(cp))
613 csa[csn++] = cp;
614 list_for_each_entry(cont, &cp->css.cgroup->children, sibling) {
615 child = cgroup_cs(cont);
616 __kfifo_put(q, (void *)&child, sizeof(cp));
617 }
618 }
619
620 for (i = 0; i < csn; i++)
621 csa[i]->pn = i;
622 ndoms = csn;
623
624restart:
625 /* Find the best partition (set of sched domains) */
626 for (i = 0; i < csn; i++) {
627 struct cpuset *a = csa[i];
628 int apn = a->pn;
629
630 for (j = 0; j < csn; j++) {
631 struct cpuset *b = csa[j];
632 int bpn = b->pn;
633
634 if (apn != bpn && cpusets_overlap(a, b)) {
635 for (k = 0; k < csn; k++) {
636 struct cpuset *c = csa[k];
637
638 if (c->pn == bpn)
639 c->pn = apn;
640 }
641 ndoms--; /* one less element */
642 goto restart;
643 }
644 }
645 }
646
647 /* Convert <csn, csa> to <ndoms, doms> */
648 doms = kmalloc(ndoms * sizeof(cpumask_t), GFP_KERNEL);
649 if (!doms)
650 goto rebuild;
1d3504fc 651 dattr = kmalloc(ndoms * sizeof(struct sched_domain_attr), GFP_KERNEL);
029190c5
PJ
652
653 for (nslot = 0, i = 0; i < csn; i++) {
654 struct cpuset *a = csa[i];
655 int apn = a->pn;
656
657 if (apn >= 0) {
658 cpumask_t *dp = doms + nslot;
659
660 if (nslot == ndoms) {
661 static int warnings = 10;
662 if (warnings) {
663 printk(KERN_WARNING
664 "rebuild_sched_domains confused:"
665 " nslot %d, ndoms %d, csn %d, i %d,"
666 " apn %d\n",
667 nslot, ndoms, csn, i, apn);
668 warnings--;
669 }
670 continue;
671 }
672
673 cpus_clear(*dp);
1d3504fc
HS
674 if (dattr)
675 *(dattr + nslot) = SD_ATTR_INIT;
029190c5
PJ
676 for (j = i; j < csn; j++) {
677 struct cpuset *b = csa[j];
678
679 if (apn == b->pn) {
680 cpus_or(*dp, *dp, b->cpus_allowed);
681 b->pn = -1;
1d3504fc 682 update_domain_attr(dattr, b);
029190c5
PJ
683 }
684 }
685 nslot++;
686 }
687 }
688 BUG_ON(nslot != ndoms);
689
690rebuild:
691 /* Have scheduler rebuild sched domains */
86ef5c9a 692 get_online_cpus();
1d3504fc 693 partition_sched_domains(ndoms, doms, dattr);
86ef5c9a 694 put_online_cpus();
029190c5
PJ
695
696done:
697 if (q && !IS_ERR(q))
698 kfifo_free(q);
699 kfree(csa);
700 /* Don't kfree(doms) -- partition_sched_domains() does that. */
1d3504fc 701 /* Don't kfree(dattr) -- partition_sched_domains() does that. */
029190c5
PJ
702}
703
8707d8b8
PM
704static inline int started_after_time(struct task_struct *t1,
705 struct timespec *time,
706 struct task_struct *t2)
707{
708 int start_diff = timespec_compare(&t1->start_time, time);
709 if (start_diff > 0) {
710 return 1;
711 } else if (start_diff < 0) {
712 return 0;
713 } else {
714 /*
715 * Arbitrarily, if two processes started at the same
716 * time, we'll say that the lower pointer value
717 * started first. Note that t2 may have exited by now
718 * so this may not be a valid pointer any longer, but
719 * that's fine - it still serves to distinguish
720 * between two tasks started (effectively)
721 * simultaneously.
722 */
723 return t1 > t2;
724 }
725}
726
727static inline int started_after(void *p1, void *p2)
728{
729 struct task_struct *t1 = p1;
730 struct task_struct *t2 = p2;
731 return started_after_time(t1, &t2->start_time, t2);
732}
733
58f4790b
CW
734/**
735 * cpuset_test_cpumask - test a task's cpus_allowed versus its cpuset's
736 * @tsk: task to test
737 * @scan: struct cgroup_scanner contained in its struct cpuset_hotplug_scanner
738 *
2df167a3 739 * Call with cgroup_mutex held. May take callback_mutex during call.
58f4790b
CW
740 * Called for each task in a cgroup by cgroup_scan_tasks().
741 * Return nonzero if this tasks's cpus_allowed mask should be changed (in other
742 * words, if its mask is not equal to its cpuset's mask).
053199ed 743 */
9e0c914c
AB
744static int cpuset_test_cpumask(struct task_struct *tsk,
745 struct cgroup_scanner *scan)
58f4790b
CW
746{
747 return !cpus_equal(tsk->cpus_allowed,
748 (cgroup_cs(scan->cg))->cpus_allowed);
749}
053199ed 750
58f4790b
CW
751/**
752 * cpuset_change_cpumask - make a task's cpus_allowed the same as its cpuset's
753 * @tsk: task to test
754 * @scan: struct cgroup_scanner containing the cgroup of the task
755 *
756 * Called by cgroup_scan_tasks() for each task in a cgroup whose
757 * cpus_allowed mask needs to be changed.
758 *
759 * We don't need to re-check for the cgroup/cpuset membership, since we're
760 * holding cgroup_lock() at this point.
761 */
9e0c914c
AB
762static void cpuset_change_cpumask(struct task_struct *tsk,
763 struct cgroup_scanner *scan)
58f4790b 764{
f9a86fcb 765 set_cpus_allowed_ptr(tsk, &((cgroup_cs(scan->cg))->cpus_allowed));
58f4790b
CW
766}
767
768/**
769 * update_cpumask - update the cpus_allowed mask of a cpuset and all tasks in it
770 * @cs: the cpuset to consider
771 * @buf: buffer of cpu numbers written to this cpuset
772 */
1da177e4
LT
773static int update_cpumask(struct cpuset *cs, char *buf)
774{
775 struct cpuset trialcs;
58f4790b 776 struct cgroup_scanner scan;
8707d8b8 777 struct ptr_heap heap;
58f4790b
CW
778 int retval;
779 int is_load_balanced;
1da177e4 780
4c4d50f7
PJ
781 /* top_cpuset.cpus_allowed tracks cpu_online_map; it's read-only */
782 if (cs == &top_cpuset)
783 return -EACCES;
784
1da177e4 785 trialcs = *cs;
6f7f02e7
DR
786
787 /*
c8d9c90c 788 * An empty cpus_allowed is ok only if the cpuset has no tasks.
020958b6
PJ
789 * Since cpulist_parse() fails on an empty mask, we special case
790 * that parsing. The validate_change() call ensures that cpusets
791 * with tasks have cpus.
6f7f02e7 792 */
020958b6
PJ
793 buf = strstrip(buf);
794 if (!*buf) {
6f7f02e7
DR
795 cpus_clear(trialcs.cpus_allowed);
796 } else {
797 retval = cpulist_parse(buf, trialcs.cpus_allowed);
798 if (retval < 0)
799 return retval;
800 }
1da177e4 801 cpus_and(trialcs.cpus_allowed, trialcs.cpus_allowed, cpu_online_map);
1da177e4 802 retval = validate_change(cs, &trialcs);
85d7b949
DG
803 if (retval < 0)
804 return retval;
029190c5 805
8707d8b8
PM
806 /* Nothing to do if the cpus didn't change */
807 if (cpus_equal(cs->cpus_allowed, trialcs.cpus_allowed))
808 return 0;
58f4790b 809
8707d8b8
PM
810 retval = heap_init(&heap, PAGE_SIZE, GFP_KERNEL, &started_after);
811 if (retval)
812 return retval;
813
029190c5
PJ
814 is_load_balanced = is_sched_load_balance(&trialcs);
815
3d3f26a7 816 mutex_lock(&callback_mutex);
85d7b949 817 cs->cpus_allowed = trialcs.cpus_allowed;
3d3f26a7 818 mutex_unlock(&callback_mutex);
029190c5 819
8707d8b8
PM
820 /*
821 * Scan tasks in the cpuset, and update the cpumasks of any
58f4790b 822 * that need an update.
8707d8b8 823 */
58f4790b
CW
824 scan.cg = cs->css.cgroup;
825 scan.test_task = cpuset_test_cpumask;
826 scan.process_task = cpuset_change_cpumask;
827 scan.heap = &heap;
828 cgroup_scan_tasks(&scan);
8707d8b8 829 heap_free(&heap);
58f4790b 830
8707d8b8 831 if (is_load_balanced)
029190c5 832 rebuild_sched_domains();
85d7b949 833 return 0;
1da177e4
LT
834}
835
e4e364e8
PJ
836/*
837 * cpuset_migrate_mm
838 *
839 * Migrate memory region from one set of nodes to another.
840 *
841 * Temporarilly set tasks mems_allowed to target nodes of migration,
842 * so that the migration code can allocate pages on these nodes.
843 *
2df167a3 844 * Call holding cgroup_mutex, so current's cpuset won't change
c8d9c90c 845 * during this call, as manage_mutex holds off any cpuset_attach()
e4e364e8
PJ
846 * calls. Therefore we don't need to take task_lock around the
847 * call to guarantee_online_mems(), as we know no one is changing
2df167a3 848 * our task's cpuset.
e4e364e8
PJ
849 *
850 * Hold callback_mutex around the two modifications of our tasks
851 * mems_allowed to synchronize with cpuset_mems_allowed().
852 *
853 * While the mm_struct we are migrating is typically from some
854 * other task, the task_struct mems_allowed that we are hacking
855 * is for our current task, which must allocate new pages for that
856 * migrating memory region.
857 *
858 * We call cpuset_update_task_memory_state() before hacking
859 * our tasks mems_allowed, so that we are assured of being in
860 * sync with our tasks cpuset, and in particular, callbacks to
861 * cpuset_update_task_memory_state() from nested page allocations
862 * won't see any mismatch of our cpuset and task mems_generation
863 * values, so won't overwrite our hacked tasks mems_allowed
864 * nodemask.
865 */
866
867static void cpuset_migrate_mm(struct mm_struct *mm, const nodemask_t *from,
868 const nodemask_t *to)
869{
870 struct task_struct *tsk = current;
871
872 cpuset_update_task_memory_state();
873
874 mutex_lock(&callback_mutex);
875 tsk->mems_allowed = *to;
876 mutex_unlock(&callback_mutex);
877
878 do_migrate_pages(mm, from, to, MPOL_MF_MOVE_ALL);
879
880 mutex_lock(&callback_mutex);
8793d854 881 guarantee_online_mems(task_cs(tsk),&tsk->mems_allowed);
e4e364e8
PJ
882 mutex_unlock(&callback_mutex);
883}
884
053199ed 885/*
4225399a
PJ
886 * Handle user request to change the 'mems' memory placement
887 * of a cpuset. Needs to validate the request, update the
888 * cpusets mems_allowed and mems_generation, and for each
04c19fa6
PJ
889 * task in the cpuset, rebind any vma mempolicies and if
890 * the cpuset is marked 'memory_migrate', migrate the tasks
891 * pages to the new memory.
4225399a 892 *
2df167a3 893 * Call with cgroup_mutex held. May take callback_mutex during call.
4225399a
PJ
894 * Will take tasklist_lock, scan tasklist for tasks in cpuset cs,
895 * lock each such tasks mm->mmap_sem, scan its vma's and rebind
896 * their mempolicies to the cpusets new mems_allowed.
053199ed
PJ
897 */
898
8793d854
PM
899static void *cpuset_being_rebound;
900
1da177e4
LT
901static int update_nodemask(struct cpuset *cs, char *buf)
902{
903 struct cpuset trialcs;
04c19fa6 904 nodemask_t oldmem;
8793d854 905 struct task_struct *p;
4225399a
PJ
906 struct mm_struct **mmarray;
907 int i, n, ntasks;
04c19fa6 908 int migrate;
4225399a 909 int fudge;
1da177e4 910 int retval;
8793d854 911 struct cgroup_iter it;
1da177e4 912
0e1e7c7a
CL
913 /*
914 * top_cpuset.mems_allowed tracks node_stats[N_HIGH_MEMORY];
915 * it's read-only
916 */
38837fc7
PJ
917 if (cs == &top_cpuset)
918 return -EACCES;
919
1da177e4 920 trialcs = *cs;
6f7f02e7
DR
921
922 /*
020958b6
PJ
923 * An empty mems_allowed is ok iff there are no tasks in the cpuset.
924 * Since nodelist_parse() fails on an empty mask, we special case
925 * that parsing. The validate_change() call ensures that cpusets
926 * with tasks have memory.
6f7f02e7 927 */
020958b6
PJ
928 buf = strstrip(buf);
929 if (!*buf) {
6f7f02e7
DR
930 nodes_clear(trialcs.mems_allowed);
931 } else {
932 retval = nodelist_parse(buf, trialcs.mems_allowed);
933 if (retval < 0)
934 goto done;
935 }
0e1e7c7a
CL
936 nodes_and(trialcs.mems_allowed, trialcs.mems_allowed,
937 node_states[N_HIGH_MEMORY]);
04c19fa6
PJ
938 oldmem = cs->mems_allowed;
939 if (nodes_equal(oldmem, trialcs.mems_allowed)) {
940 retval = 0; /* Too easy - nothing to do */
941 goto done;
942 }
59dac16f
PJ
943 retval = validate_change(cs, &trialcs);
944 if (retval < 0)
945 goto done;
946
3d3f26a7 947 mutex_lock(&callback_mutex);
59dac16f 948 cs->mems_allowed = trialcs.mems_allowed;
151a4420 949 cs->mems_generation = cpuset_mems_generation++;
3d3f26a7 950 mutex_unlock(&callback_mutex);
59dac16f 951
846a16bf 952 cpuset_being_rebound = cs; /* causes mpol_dup() rebind */
4225399a
PJ
953
954 fudge = 10; /* spare mmarray[] slots */
955 fudge += cpus_weight(cs->cpus_allowed); /* imagine one fork-bomb/cpu */
956 retval = -ENOMEM;
957
958 /*
959 * Allocate mmarray[] to hold mm reference for each task
960 * in cpuset cs. Can't kmalloc GFP_KERNEL while holding
961 * tasklist_lock. We could use GFP_ATOMIC, but with a
962 * few more lines of code, we can retry until we get a big
963 * enough mmarray[] w/o using GFP_ATOMIC.
964 */
965 while (1) {
8793d854 966 ntasks = cgroup_task_count(cs->css.cgroup); /* guess */
4225399a
PJ
967 ntasks += fudge;
968 mmarray = kmalloc(ntasks * sizeof(*mmarray), GFP_KERNEL);
969 if (!mmarray)
970 goto done;
c2aef333 971 read_lock(&tasklist_lock); /* block fork */
8793d854 972 if (cgroup_task_count(cs->css.cgroup) <= ntasks)
4225399a 973 break; /* got enough */
c2aef333 974 read_unlock(&tasklist_lock); /* try again */
4225399a
PJ
975 kfree(mmarray);
976 }
977
978 n = 0;
979
980 /* Load up mmarray[] with mm reference for each task in cpuset. */
8793d854
PM
981 cgroup_iter_start(cs->css.cgroup, &it);
982 while ((p = cgroup_iter_next(cs->css.cgroup, &it))) {
4225399a
PJ
983 struct mm_struct *mm;
984
985 if (n >= ntasks) {
986 printk(KERN_WARNING
987 "Cpuset mempolicy rebind incomplete.\n");
8793d854 988 break;
4225399a 989 }
4225399a
PJ
990 mm = get_task_mm(p);
991 if (!mm)
992 continue;
993 mmarray[n++] = mm;
8793d854
PM
994 }
995 cgroup_iter_end(cs->css.cgroup, &it);
c2aef333 996 read_unlock(&tasklist_lock);
4225399a
PJ
997
998 /*
999 * Now that we've dropped the tasklist spinlock, we can
1000 * rebind the vma mempolicies of each mm in mmarray[] to their
1001 * new cpuset, and release that mm. The mpol_rebind_mm()
1002 * call takes mmap_sem, which we couldn't take while holding
846a16bf 1003 * tasklist_lock. Forks can happen again now - the mpol_dup()
4225399a
PJ
1004 * cpuset_being_rebound check will catch such forks, and rebind
1005 * their vma mempolicies too. Because we still hold the global
2df167a3 1006 * cgroup_mutex, we know that no other rebind effort will
4225399a
PJ
1007 * be contending for the global variable cpuset_being_rebound.
1008 * It's ok if we rebind the same mm twice; mpol_rebind_mm()
04c19fa6 1009 * is idempotent. Also migrate pages in each mm to new nodes.
4225399a 1010 */
04c19fa6 1011 migrate = is_memory_migrate(cs);
4225399a
PJ
1012 for (i = 0; i < n; i++) {
1013 struct mm_struct *mm = mmarray[i];
1014
1015 mpol_rebind_mm(mm, &cs->mems_allowed);
e4e364e8
PJ
1016 if (migrate)
1017 cpuset_migrate_mm(mm, &oldmem, &cs->mems_allowed);
4225399a
PJ
1018 mmput(mm);
1019 }
1020
2df167a3 1021 /* We're done rebinding vmas to this cpuset's new mems_allowed. */
4225399a 1022 kfree(mmarray);
8793d854 1023 cpuset_being_rebound = NULL;
4225399a 1024 retval = 0;
59dac16f 1025done:
1da177e4
LT
1026 return retval;
1027}
1028
8793d854
PM
1029int current_cpuset_is_being_rebound(void)
1030{
1031 return task_cs(current) == cpuset_being_rebound;
1032}
1033
5be7a479 1034static int update_relax_domain_level(struct cpuset *cs, s64 val)
1d3504fc 1035{
5be7a479 1036 if ((int)val < 0)
1d3504fc
HS
1037 val = -1;
1038
1039 if (val != cs->relax_domain_level) {
1040 cs->relax_domain_level = val;
1041 rebuild_sched_domains();
1042 }
1043
1044 return 0;
1045}
1046
1da177e4
LT
1047/*
1048 * update_flag - read a 0 or a 1 in a file and update associated flag
78608366
PM
1049 * bit: the bit to update (see cpuset_flagbits_t)
1050 * cs: the cpuset to update
1051 * turning_on: whether the flag is being set or cleared
053199ed 1052 *
2df167a3 1053 * Call with cgroup_mutex held.
1da177e4
LT
1054 */
1055
700fe1ab
PM
1056static int update_flag(cpuset_flagbits_t bit, struct cpuset *cs,
1057 int turning_on)
1da177e4 1058{
1da177e4 1059 struct cpuset trialcs;
607717a6 1060 int err;
029190c5 1061 int cpus_nonempty, balance_flag_changed;
1da177e4 1062
1da177e4
LT
1063 trialcs = *cs;
1064 if (turning_on)
1065 set_bit(bit, &trialcs.flags);
1066 else
1067 clear_bit(bit, &trialcs.flags);
1068
1069 err = validate_change(cs, &trialcs);
85d7b949
DG
1070 if (err < 0)
1071 return err;
029190c5
PJ
1072
1073 cpus_nonempty = !cpus_empty(trialcs.cpus_allowed);
1074 balance_flag_changed = (is_sched_load_balance(cs) !=
1075 is_sched_load_balance(&trialcs));
1076
3d3f26a7 1077 mutex_lock(&callback_mutex);
69604067 1078 cs->flags = trialcs.flags;
3d3f26a7 1079 mutex_unlock(&callback_mutex);
85d7b949 1080
029190c5
PJ
1081 if (cpus_nonempty && balance_flag_changed)
1082 rebuild_sched_domains();
1083
85d7b949 1084 return 0;
1da177e4
LT
1085}
1086
3e0d98b9 1087/*
80f7228b 1088 * Frequency meter - How fast is some event occurring?
3e0d98b9
PJ
1089 *
1090 * These routines manage a digitally filtered, constant time based,
1091 * event frequency meter. There are four routines:
1092 * fmeter_init() - initialize a frequency meter.
1093 * fmeter_markevent() - called each time the event happens.
1094 * fmeter_getrate() - returns the recent rate of such events.
1095 * fmeter_update() - internal routine used to update fmeter.
1096 *
1097 * A common data structure is passed to each of these routines,
1098 * which is used to keep track of the state required to manage the
1099 * frequency meter and its digital filter.
1100 *
1101 * The filter works on the number of events marked per unit time.
1102 * The filter is single-pole low-pass recursive (IIR). The time unit
1103 * is 1 second. Arithmetic is done using 32-bit integers scaled to
1104 * simulate 3 decimal digits of precision (multiplied by 1000).
1105 *
1106 * With an FM_COEF of 933, and a time base of 1 second, the filter
1107 * has a half-life of 10 seconds, meaning that if the events quit
1108 * happening, then the rate returned from the fmeter_getrate()
1109 * will be cut in half each 10 seconds, until it converges to zero.
1110 *
1111 * It is not worth doing a real infinitely recursive filter. If more
1112 * than FM_MAXTICKS ticks have elapsed since the last filter event,
1113 * just compute FM_MAXTICKS ticks worth, by which point the level
1114 * will be stable.
1115 *
1116 * Limit the count of unprocessed events to FM_MAXCNT, so as to avoid
1117 * arithmetic overflow in the fmeter_update() routine.
1118 *
1119 * Given the simple 32 bit integer arithmetic used, this meter works
1120 * best for reporting rates between one per millisecond (msec) and
1121 * one per 32 (approx) seconds. At constant rates faster than one
1122 * per msec it maxes out at values just under 1,000,000. At constant
1123 * rates between one per msec, and one per second it will stabilize
1124 * to a value N*1000, where N is the rate of events per second.
1125 * At constant rates between one per second and one per 32 seconds,
1126 * it will be choppy, moving up on the seconds that have an event,
1127 * and then decaying until the next event. At rates slower than
1128 * about one in 32 seconds, it decays all the way back to zero between
1129 * each event.
1130 */
1131
1132#define FM_COEF 933 /* coefficient for half-life of 10 secs */
1133#define FM_MAXTICKS ((time_t)99) /* useless computing more ticks than this */
1134#define FM_MAXCNT 1000000 /* limit cnt to avoid overflow */
1135#define FM_SCALE 1000 /* faux fixed point scale */
1136
1137/* Initialize a frequency meter */
1138static void fmeter_init(struct fmeter *fmp)
1139{
1140 fmp->cnt = 0;
1141 fmp->val = 0;
1142 fmp->time = 0;
1143 spin_lock_init(&fmp->lock);
1144}
1145
1146/* Internal meter update - process cnt events and update value */
1147static void fmeter_update(struct fmeter *fmp)
1148{
1149 time_t now = get_seconds();
1150 time_t ticks = now - fmp->time;
1151
1152 if (ticks == 0)
1153 return;
1154
1155 ticks = min(FM_MAXTICKS, ticks);
1156 while (ticks-- > 0)
1157 fmp->val = (FM_COEF * fmp->val) / FM_SCALE;
1158 fmp->time = now;
1159
1160 fmp->val += ((FM_SCALE - FM_COEF) * fmp->cnt) / FM_SCALE;
1161 fmp->cnt = 0;
1162}
1163
1164/* Process any previous ticks, then bump cnt by one (times scale). */
1165static void fmeter_markevent(struct fmeter *fmp)
1166{
1167 spin_lock(&fmp->lock);
1168 fmeter_update(fmp);
1169 fmp->cnt = min(FM_MAXCNT, fmp->cnt + FM_SCALE);
1170 spin_unlock(&fmp->lock);
1171}
1172
1173/* Process any previous ticks, then return current value. */
1174static int fmeter_getrate(struct fmeter *fmp)
1175{
1176 int val;
1177
1178 spin_lock(&fmp->lock);
1179 fmeter_update(fmp);
1180 val = fmp->val;
1181 spin_unlock(&fmp->lock);
1182 return val;
1183}
1184
2df167a3 1185/* Called by cgroups to determine if a cpuset is usable; cgroup_mutex held */
8793d854
PM
1186static int cpuset_can_attach(struct cgroup_subsys *ss,
1187 struct cgroup *cont, struct task_struct *tsk)
1da177e4 1188{
8793d854 1189 struct cpuset *cs = cgroup_cs(cont);
1da177e4 1190
1da177e4
LT
1191 if (cpus_empty(cs->cpus_allowed) || nodes_empty(cs->mems_allowed))
1192 return -ENOSPC;
1193
8793d854
PM
1194 return security_task_setscheduler(tsk, 0, NULL);
1195}
1da177e4 1196
8793d854
PM
1197static void cpuset_attach(struct cgroup_subsys *ss,
1198 struct cgroup *cont, struct cgroup *oldcont,
1199 struct task_struct *tsk)
1200{
1201 cpumask_t cpus;
1202 nodemask_t from, to;
1203 struct mm_struct *mm;
1204 struct cpuset *cs = cgroup_cs(cont);
1205 struct cpuset *oldcs = cgroup_cs(oldcont);
22fb52dd 1206
3d3f26a7 1207 mutex_lock(&callback_mutex);
1da177e4 1208 guarantee_online_cpus(cs, &cpus);
f9a86fcb 1209 set_cpus_allowed_ptr(tsk, &cpus);
8793d854 1210 mutex_unlock(&callback_mutex);
1da177e4 1211
45b07ef3
PJ
1212 from = oldcs->mems_allowed;
1213 to = cs->mems_allowed;
4225399a
PJ
1214 mm = get_task_mm(tsk);
1215 if (mm) {
1216 mpol_rebind_mm(mm, &to);
2741a559 1217 if (is_memory_migrate(cs))
e4e364e8 1218 cpuset_migrate_mm(mm, &from, &to);
4225399a
PJ
1219 mmput(mm);
1220 }
1221
1da177e4
LT
1222}
1223
1224/* The various types of files and directories in a cpuset file system */
1225
1226typedef enum {
45b07ef3 1227 FILE_MEMORY_MIGRATE,
1da177e4
LT
1228 FILE_CPULIST,
1229 FILE_MEMLIST,
1230 FILE_CPU_EXCLUSIVE,
1231 FILE_MEM_EXCLUSIVE,
78608366 1232 FILE_MEM_HARDWALL,
029190c5 1233 FILE_SCHED_LOAD_BALANCE,
1d3504fc 1234 FILE_SCHED_RELAX_DOMAIN_LEVEL,
3e0d98b9
PJ
1235 FILE_MEMORY_PRESSURE_ENABLED,
1236 FILE_MEMORY_PRESSURE,
825a46af
PJ
1237 FILE_SPREAD_PAGE,
1238 FILE_SPREAD_SLAB,
1da177e4
LT
1239} cpuset_filetype_t;
1240
8793d854
PM
1241static ssize_t cpuset_common_file_write(struct cgroup *cont,
1242 struct cftype *cft,
1243 struct file *file,
d3ed11c3 1244 const char __user *userbuf,
1da177e4
LT
1245 size_t nbytes, loff_t *unused_ppos)
1246{
8793d854 1247 struct cpuset *cs = cgroup_cs(cont);
1da177e4
LT
1248 cpuset_filetype_t type = cft->private;
1249 char *buffer;
1250 int retval = 0;
1251
1252 /* Crude upper limit on largest legitimate cpulist user might write. */
029190c5 1253 if (nbytes > 100U + 6 * max(NR_CPUS, MAX_NUMNODES))
1da177e4
LT
1254 return -E2BIG;
1255
1256 /* +1 for nul-terminator */
b331d259
HH
1257 buffer = kmalloc(nbytes + 1, GFP_KERNEL);
1258 if (!buffer)
1da177e4
LT
1259 return -ENOMEM;
1260
1261 if (copy_from_user(buffer, userbuf, nbytes)) {
1262 retval = -EFAULT;
1263 goto out1;
1264 }
1265 buffer[nbytes] = 0; /* nul-terminate */
1266
8793d854 1267 cgroup_lock();
1da177e4 1268
8793d854 1269 if (cgroup_is_removed(cont)) {
1da177e4
LT
1270 retval = -ENODEV;
1271 goto out2;
1272 }
1273
1274 switch (type) {
1275 case FILE_CPULIST:
1276 retval = update_cpumask(cs, buffer);
1277 break;
1278 case FILE_MEMLIST:
1279 retval = update_nodemask(cs, buffer);
1280 break;
700fe1ab
PM
1281 default:
1282 retval = -EINVAL;
1283 goto out2;
1284 }
1285
1286 if (retval == 0)
1287 retval = nbytes;
1288out2:
1289 cgroup_unlock();
1290out1:
1291 kfree(buffer);
1292 return retval;
1293}
1294
1295static int cpuset_write_u64(struct cgroup *cgrp, struct cftype *cft, u64 val)
1296{
1297 int retval = 0;
1298 struct cpuset *cs = cgroup_cs(cgrp);
1299 cpuset_filetype_t type = cft->private;
1300
1301 cgroup_lock();
1302
1303 if (cgroup_is_removed(cgrp)) {
1304 cgroup_unlock();
1305 return -ENODEV;
1306 }
1307
1308 switch (type) {
1da177e4 1309 case FILE_CPU_EXCLUSIVE:
700fe1ab 1310 retval = update_flag(CS_CPU_EXCLUSIVE, cs, val);
1da177e4
LT
1311 break;
1312 case FILE_MEM_EXCLUSIVE:
700fe1ab 1313 retval = update_flag(CS_MEM_EXCLUSIVE, cs, val);
1da177e4 1314 break;
78608366
PM
1315 case FILE_MEM_HARDWALL:
1316 retval = update_flag(CS_MEM_HARDWALL, cs, val);
1317 break;
029190c5 1318 case FILE_SCHED_LOAD_BALANCE:
700fe1ab 1319 retval = update_flag(CS_SCHED_LOAD_BALANCE, cs, val);
1d3504fc 1320 break;
45b07ef3 1321 case FILE_MEMORY_MIGRATE:
700fe1ab 1322 retval = update_flag(CS_MEMORY_MIGRATE, cs, val);
45b07ef3 1323 break;
3e0d98b9 1324 case FILE_MEMORY_PRESSURE_ENABLED:
700fe1ab 1325 cpuset_memory_pressure_enabled = !!val;
3e0d98b9
PJ
1326 break;
1327 case FILE_MEMORY_PRESSURE:
1328 retval = -EACCES;
1329 break;
825a46af 1330 case FILE_SPREAD_PAGE:
700fe1ab 1331 retval = update_flag(CS_SPREAD_PAGE, cs, val);
151a4420 1332 cs->mems_generation = cpuset_mems_generation++;
825a46af
PJ
1333 break;
1334 case FILE_SPREAD_SLAB:
700fe1ab 1335 retval = update_flag(CS_SPREAD_SLAB, cs, val);
151a4420 1336 cs->mems_generation = cpuset_mems_generation++;
825a46af 1337 break;
1da177e4
LT
1338 default:
1339 retval = -EINVAL;
700fe1ab 1340 break;
1da177e4 1341 }
8793d854 1342 cgroup_unlock();
1da177e4
LT
1343 return retval;
1344}
1345
5be7a479
PM
1346static int cpuset_write_s64(struct cgroup *cgrp, struct cftype *cft, s64 val)
1347{
1348 int retval = 0;
1349 struct cpuset *cs = cgroup_cs(cgrp);
1350 cpuset_filetype_t type = cft->private;
1351
1352 cgroup_lock();
1353
1354 if (cgroup_is_removed(cgrp)) {
1355 cgroup_unlock();
1356 return -ENODEV;
1357 }
1358 switch (type) {
1359 case FILE_SCHED_RELAX_DOMAIN_LEVEL:
1360 retval = update_relax_domain_level(cs, val);
1361 break;
1362 default:
1363 retval = -EINVAL;
1364 break;
1365 }
1366 cgroup_unlock();
1367 return retval;
1368}
1369
1da177e4
LT
1370/*
1371 * These ascii lists should be read in a single call, by using a user
1372 * buffer large enough to hold the entire map. If read in smaller
1373 * chunks, there is no guarantee of atomicity. Since the display format
1374 * used, list of ranges of sequential numbers, is variable length,
1375 * and since these maps can change value dynamically, one could read
1376 * gibberish by doing partial reads while a list was changing.
1377 * A single large read to a buffer that crosses a page boundary is
1378 * ok, because the result being copied to user land is not recomputed
1379 * across a page fault.
1380 */
1381
1382static int cpuset_sprintf_cpulist(char *page, struct cpuset *cs)
1383{
1384 cpumask_t mask;
1385
3d3f26a7 1386 mutex_lock(&callback_mutex);
1da177e4 1387 mask = cs->cpus_allowed;
3d3f26a7 1388 mutex_unlock(&callback_mutex);
1da177e4
LT
1389
1390 return cpulist_scnprintf(page, PAGE_SIZE, mask);
1391}
1392
1393static int cpuset_sprintf_memlist(char *page, struct cpuset *cs)
1394{
1395 nodemask_t mask;
1396
3d3f26a7 1397 mutex_lock(&callback_mutex);
1da177e4 1398 mask = cs->mems_allowed;
3d3f26a7 1399 mutex_unlock(&callback_mutex);
1da177e4
LT
1400
1401 return nodelist_scnprintf(page, PAGE_SIZE, mask);
1402}
1403
8793d854
PM
1404static ssize_t cpuset_common_file_read(struct cgroup *cont,
1405 struct cftype *cft,
1406 struct file *file,
1407 char __user *buf,
1408 size_t nbytes, loff_t *ppos)
1da177e4 1409{
8793d854 1410 struct cpuset *cs = cgroup_cs(cont);
1da177e4
LT
1411 cpuset_filetype_t type = cft->private;
1412 char *page;
1413 ssize_t retval = 0;
1414 char *s;
1da177e4 1415
e12ba74d 1416 if (!(page = (char *)__get_free_page(GFP_TEMPORARY)))
1da177e4
LT
1417 return -ENOMEM;
1418
1419 s = page;
1420
1421 switch (type) {
1422 case FILE_CPULIST:
1423 s += cpuset_sprintf_cpulist(s, cs);
1424 break;
1425 case FILE_MEMLIST:
1426 s += cpuset_sprintf_memlist(s, cs);
1427 break;
1da177e4
LT
1428 default:
1429 retval = -EINVAL;
1430 goto out;
1431 }
1432 *s++ = '\n';
1da177e4 1433
eacaa1f5 1434 retval = simple_read_from_buffer(buf, nbytes, ppos, page, s - page);
1da177e4
LT
1435out:
1436 free_page((unsigned long)page);
1437 return retval;
1438}
1439
700fe1ab
PM
1440static u64 cpuset_read_u64(struct cgroup *cont, struct cftype *cft)
1441{
1442 struct cpuset *cs = cgroup_cs(cont);
1443 cpuset_filetype_t type = cft->private;
1444 switch (type) {
1445 case FILE_CPU_EXCLUSIVE:
1446 return is_cpu_exclusive(cs);
1447 case FILE_MEM_EXCLUSIVE:
1448 return is_mem_exclusive(cs);
78608366
PM
1449 case FILE_MEM_HARDWALL:
1450 return is_mem_hardwall(cs);
700fe1ab
PM
1451 case FILE_SCHED_LOAD_BALANCE:
1452 return is_sched_load_balance(cs);
1453 case FILE_MEMORY_MIGRATE:
1454 return is_memory_migrate(cs);
1455 case FILE_MEMORY_PRESSURE_ENABLED:
1456 return cpuset_memory_pressure_enabled;
1457 case FILE_MEMORY_PRESSURE:
1458 return fmeter_getrate(&cs->fmeter);
1459 case FILE_SPREAD_PAGE:
1460 return is_spread_page(cs);
1461 case FILE_SPREAD_SLAB:
1462 return is_spread_slab(cs);
1463 default:
1464 BUG();
1465 }
1466}
1da177e4 1467
5be7a479
PM
1468static s64 cpuset_read_s64(struct cgroup *cont, struct cftype *cft)
1469{
1470 struct cpuset *cs = cgroup_cs(cont);
1471 cpuset_filetype_t type = cft->private;
1472 switch (type) {
1473 case FILE_SCHED_RELAX_DOMAIN_LEVEL:
1474 return cs->relax_domain_level;
1475 default:
1476 BUG();
1477 }
1478}
1479
1da177e4
LT
1480
1481/*
1482 * for the common functions, 'private' gives the type of file
1483 */
1484
addf2c73
PM
1485static struct cftype files[] = {
1486 {
1487 .name = "cpus",
1488 .read = cpuset_common_file_read,
1489 .write = cpuset_common_file_write,
1490 .private = FILE_CPULIST,
1491 },
1492
1493 {
1494 .name = "mems",
1495 .read = cpuset_common_file_read,
1496 .write = cpuset_common_file_write,
1497 .private = FILE_MEMLIST,
1498 },
1499
1500 {
1501 .name = "cpu_exclusive",
1502 .read_u64 = cpuset_read_u64,
1503 .write_u64 = cpuset_write_u64,
1504 .private = FILE_CPU_EXCLUSIVE,
1505 },
1506
1507 {
1508 .name = "mem_exclusive",
1509 .read_u64 = cpuset_read_u64,
1510 .write_u64 = cpuset_write_u64,
1511 .private = FILE_MEM_EXCLUSIVE,
1512 },
1513
78608366
PM
1514 {
1515 .name = "mem_hardwall",
1516 .read_u64 = cpuset_read_u64,
1517 .write_u64 = cpuset_write_u64,
1518 .private = FILE_MEM_HARDWALL,
1519 },
1520
addf2c73
PM
1521 {
1522 .name = "sched_load_balance",
1523 .read_u64 = cpuset_read_u64,
1524 .write_u64 = cpuset_write_u64,
1525 .private = FILE_SCHED_LOAD_BALANCE,
1526 },
1527
1528 {
1529 .name = "sched_relax_domain_level",
5be7a479
PM
1530 .read_s64 = cpuset_read_s64,
1531 .write_s64 = cpuset_write_s64,
addf2c73
PM
1532 .private = FILE_SCHED_RELAX_DOMAIN_LEVEL,
1533 },
1534
1535 {
1536 .name = "memory_migrate",
1537 .read_u64 = cpuset_read_u64,
1538 .write_u64 = cpuset_write_u64,
1539 .private = FILE_MEMORY_MIGRATE,
1540 },
1541
1542 {
1543 .name = "memory_pressure",
1544 .read_u64 = cpuset_read_u64,
1545 .write_u64 = cpuset_write_u64,
1546 .private = FILE_MEMORY_PRESSURE,
1547 },
1548
1549 {
1550 .name = "memory_spread_page",
1551 .read_u64 = cpuset_read_u64,
1552 .write_u64 = cpuset_write_u64,
1553 .private = FILE_SPREAD_PAGE,
1554 },
1555
1556 {
1557 .name = "memory_spread_slab",
1558 .read_u64 = cpuset_read_u64,
1559 .write_u64 = cpuset_write_u64,
1560 .private = FILE_SPREAD_SLAB,
1561 },
45b07ef3
PJ
1562};
1563
3e0d98b9
PJ
1564static struct cftype cft_memory_pressure_enabled = {
1565 .name = "memory_pressure_enabled",
700fe1ab
PM
1566 .read_u64 = cpuset_read_u64,
1567 .write_u64 = cpuset_write_u64,
3e0d98b9
PJ
1568 .private = FILE_MEMORY_PRESSURE_ENABLED,
1569};
1570
8793d854 1571static int cpuset_populate(struct cgroup_subsys *ss, struct cgroup *cont)
1da177e4
LT
1572{
1573 int err;
1574
addf2c73
PM
1575 err = cgroup_add_files(cont, ss, files, ARRAY_SIZE(files));
1576 if (err)
1da177e4 1577 return err;
8793d854 1578 /* memory_pressure_enabled is in root cpuset only */
addf2c73 1579 if (!cont->parent)
8793d854 1580 err = cgroup_add_file(cont, ss,
addf2c73
PM
1581 &cft_memory_pressure_enabled);
1582 return err;
1da177e4
LT
1583}
1584
8793d854
PM
1585/*
1586 * post_clone() is called at the end of cgroup_clone().
1587 * 'cgroup' was just created automatically as a result of
1588 * a cgroup_clone(), and the current task is about to
1589 * be moved into 'cgroup'.
1590 *
1591 * Currently we refuse to set up the cgroup - thereby
1592 * refusing the task to be entered, and as a result refusing
1593 * the sys_unshare() or clone() which initiated it - if any
1594 * sibling cpusets have exclusive cpus or mem.
1595 *
1596 * If this becomes a problem for some users who wish to
1597 * allow that scenario, then cpuset_post_clone() could be
1598 * changed to grant parent->cpus_allowed-sibling_cpus_exclusive
2df167a3
PM
1599 * (and likewise for mems) to the new cgroup. Called with cgroup_mutex
1600 * held.
8793d854
PM
1601 */
1602static void cpuset_post_clone(struct cgroup_subsys *ss,
1603 struct cgroup *cgroup)
1604{
1605 struct cgroup *parent, *child;
1606 struct cpuset *cs, *parent_cs;
1607
1608 parent = cgroup->parent;
1609 list_for_each_entry(child, &parent->children, sibling) {
1610 cs = cgroup_cs(child);
1611 if (is_mem_exclusive(cs) || is_cpu_exclusive(cs))
1612 return;
1613 }
1614 cs = cgroup_cs(cgroup);
1615 parent_cs = cgroup_cs(parent);
1616
1617 cs->mems_allowed = parent_cs->mems_allowed;
1618 cs->cpus_allowed = parent_cs->cpus_allowed;
1619 return;
1620}
1621
1da177e4
LT
1622/*
1623 * cpuset_create - create a cpuset
2df167a3
PM
1624 * ss: cpuset cgroup subsystem
1625 * cont: control group that the new cpuset will be part of
1da177e4
LT
1626 */
1627
8793d854
PM
1628static struct cgroup_subsys_state *cpuset_create(
1629 struct cgroup_subsys *ss,
1630 struct cgroup *cont)
1da177e4
LT
1631{
1632 struct cpuset *cs;
8793d854 1633 struct cpuset *parent;
1da177e4 1634
8793d854
PM
1635 if (!cont->parent) {
1636 /* This is early initialization for the top cgroup */
1637 top_cpuset.mems_generation = cpuset_mems_generation++;
1638 return &top_cpuset.css;
1639 }
1640 parent = cgroup_cs(cont->parent);
1da177e4
LT
1641 cs = kmalloc(sizeof(*cs), GFP_KERNEL);
1642 if (!cs)
8793d854 1643 return ERR_PTR(-ENOMEM);
1da177e4 1644
cf2a473c 1645 cpuset_update_task_memory_state();
1da177e4 1646 cs->flags = 0;
825a46af
PJ
1647 if (is_spread_page(parent))
1648 set_bit(CS_SPREAD_PAGE, &cs->flags);
1649 if (is_spread_slab(parent))
1650 set_bit(CS_SPREAD_SLAB, &cs->flags);
029190c5 1651 set_bit(CS_SCHED_LOAD_BALANCE, &cs->flags);
f9a86fcb
MT
1652 cpus_clear(cs->cpus_allowed);
1653 nodes_clear(cs->mems_allowed);
151a4420 1654 cs->mems_generation = cpuset_mems_generation++;
3e0d98b9 1655 fmeter_init(&cs->fmeter);
1d3504fc 1656 cs->relax_domain_level = -1;
1da177e4
LT
1657
1658 cs->parent = parent;
202f72d5 1659 number_of_cpusets++;
8793d854 1660 return &cs->css ;
1da177e4
LT
1661}
1662
029190c5
PJ
1663/*
1664 * Locking note on the strange update_flag() call below:
1665 *
1666 * If the cpuset being removed has its flag 'sched_load_balance'
1667 * enabled, then simulate turning sched_load_balance off, which
86ef5c9a 1668 * will call rebuild_sched_domains(). The get_online_cpus()
029190c5
PJ
1669 * call in rebuild_sched_domains() must not be made while holding
1670 * callback_mutex. Elsewhere the kernel nests callback_mutex inside
86ef5c9a 1671 * get_online_cpus() calls. So the reverse nesting would risk an
029190c5
PJ
1672 * ABBA deadlock.
1673 */
1674
8793d854 1675static void cpuset_destroy(struct cgroup_subsys *ss, struct cgroup *cont)
1da177e4 1676{
8793d854 1677 struct cpuset *cs = cgroup_cs(cont);
1da177e4 1678
cf2a473c 1679 cpuset_update_task_memory_state();
029190c5
PJ
1680
1681 if (is_sched_load_balance(cs))
700fe1ab 1682 update_flag(CS_SCHED_LOAD_BALANCE, cs, 0);
029190c5 1683
202f72d5 1684 number_of_cpusets--;
8793d854 1685 kfree(cs);
1da177e4
LT
1686}
1687
8793d854
PM
1688struct cgroup_subsys cpuset_subsys = {
1689 .name = "cpuset",
1690 .create = cpuset_create,
1691 .destroy = cpuset_destroy,
1692 .can_attach = cpuset_can_attach,
1693 .attach = cpuset_attach,
1694 .populate = cpuset_populate,
1695 .post_clone = cpuset_post_clone,
1696 .subsys_id = cpuset_subsys_id,
1697 .early_init = 1,
1698};
1699
c417f024
PJ
1700/*
1701 * cpuset_init_early - just enough so that the calls to
1702 * cpuset_update_task_memory_state() in early init code
1703 * are harmless.
1704 */
1705
1706int __init cpuset_init_early(void)
1707{
8793d854 1708 top_cpuset.mems_generation = cpuset_mems_generation++;
c417f024
PJ
1709 return 0;
1710}
1711
8793d854 1712
1da177e4
LT
1713/**
1714 * cpuset_init - initialize cpusets at system boot
1715 *
1716 * Description: Initialize top_cpuset and the cpuset internal file system,
1717 **/
1718
1719int __init cpuset_init(void)
1720{
8793d854 1721 int err = 0;
1da177e4 1722
f9a86fcb
MT
1723 cpus_setall(top_cpuset.cpus_allowed);
1724 nodes_setall(top_cpuset.mems_allowed);
1da177e4 1725
3e0d98b9 1726 fmeter_init(&top_cpuset.fmeter);
151a4420 1727 top_cpuset.mems_generation = cpuset_mems_generation++;
029190c5 1728 set_bit(CS_SCHED_LOAD_BALANCE, &top_cpuset.flags);
1d3504fc 1729 top_cpuset.relax_domain_level = -1;
1da177e4 1730
1da177e4
LT
1731 err = register_filesystem(&cpuset_fs_type);
1732 if (err < 0)
8793d854
PM
1733 return err;
1734
202f72d5 1735 number_of_cpusets = 1;
8793d854 1736 return 0;
1da177e4
LT
1737}
1738
956db3ca
CW
1739/**
1740 * cpuset_do_move_task - move a given task to another cpuset
1741 * @tsk: pointer to task_struct the task to move
1742 * @scan: struct cgroup_scanner contained in its struct cpuset_hotplug_scanner
1743 *
1744 * Called by cgroup_scan_tasks() for each task in a cgroup.
1745 * Return nonzero to stop the walk through the tasks.
1746 */
9e0c914c
AB
1747static void cpuset_do_move_task(struct task_struct *tsk,
1748 struct cgroup_scanner *scan)
956db3ca
CW
1749{
1750 struct cpuset_hotplug_scanner *chsp;
1751
1752 chsp = container_of(scan, struct cpuset_hotplug_scanner, scan);
1753 cgroup_attach_task(chsp->to, tsk);
1754}
1755
1756/**
1757 * move_member_tasks_to_cpuset - move tasks from one cpuset to another
1758 * @from: cpuset in which the tasks currently reside
1759 * @to: cpuset to which the tasks will be moved
1760 *
c8d9c90c
PJ
1761 * Called with cgroup_mutex held
1762 * callback_mutex must not be held, as cpuset_attach() will take it.
956db3ca
CW
1763 *
1764 * The cgroup_scan_tasks() function will scan all the tasks in a cgroup,
1765 * calling callback functions for each.
1766 */
1767static void move_member_tasks_to_cpuset(struct cpuset *from, struct cpuset *to)
1768{
1769 struct cpuset_hotplug_scanner scan;
1770
1771 scan.scan.cg = from->css.cgroup;
1772 scan.scan.test_task = NULL; /* select all tasks in cgroup */
1773 scan.scan.process_task = cpuset_do_move_task;
1774 scan.scan.heap = NULL;
1775 scan.to = to->css.cgroup;
1776
1777 if (cgroup_scan_tasks((struct cgroup_scanner *)&scan))
1778 printk(KERN_ERR "move_member_tasks_to_cpuset: "
1779 "cgroup_scan_tasks failed\n");
1780}
1781
b1aac8bb
PJ
1782/*
1783 * If common_cpu_mem_hotplug_unplug(), below, unplugs any CPUs
1784 * or memory nodes, we need to walk over the cpuset hierarchy,
1785 * removing that CPU or node from all cpusets. If this removes the
956db3ca
CW
1786 * last CPU or node from a cpuset, then move the tasks in the empty
1787 * cpuset to its next-highest non-empty parent.
b1aac8bb 1788 *
c8d9c90c
PJ
1789 * Called with cgroup_mutex held
1790 * callback_mutex must not be held, as cpuset_attach() will take it.
b1aac8bb 1791 */
956db3ca
CW
1792static void remove_tasks_in_empty_cpuset(struct cpuset *cs)
1793{
1794 struct cpuset *parent;
1795
c8d9c90c
PJ
1796 /*
1797 * The cgroup's css_sets list is in use if there are tasks
1798 * in the cpuset; the list is empty if there are none;
1799 * the cs->css.refcnt seems always 0.
1800 */
956db3ca
CW
1801 if (list_empty(&cs->css.cgroup->css_sets))
1802 return;
b1aac8bb 1803
956db3ca
CW
1804 /*
1805 * Find its next-highest non-empty parent, (top cpuset
1806 * has online cpus, so can't be empty).
1807 */
1808 parent = cs->parent;
b4501295
PJ
1809 while (cpus_empty(parent->cpus_allowed) ||
1810 nodes_empty(parent->mems_allowed))
956db3ca 1811 parent = parent->parent;
956db3ca
CW
1812
1813 move_member_tasks_to_cpuset(cs, parent);
1814}
1815
1816/*
1817 * Walk the specified cpuset subtree and look for empty cpusets.
1818 * The tasks of such cpuset must be moved to a parent cpuset.
1819 *
2df167a3 1820 * Called with cgroup_mutex held. We take callback_mutex to modify
956db3ca
CW
1821 * cpus_allowed and mems_allowed.
1822 *
1823 * This walk processes the tree from top to bottom, completing one layer
1824 * before dropping down to the next. It always processes a node before
1825 * any of its children.
1826 *
1827 * For now, since we lack memory hot unplug, we'll never see a cpuset
1828 * that has tasks along with an empty 'mems'. But if we did see such
1829 * a cpuset, we'd handle it just like we do if its 'cpus' was empty.
1830 */
1831static void scan_for_empty_cpusets(const struct cpuset *root)
b1aac8bb 1832{
956db3ca
CW
1833 struct cpuset *cp; /* scans cpusets being updated */
1834 struct cpuset *child; /* scans child cpusets of cp */
1835 struct list_head queue;
8793d854 1836 struct cgroup *cont;
b1aac8bb 1837
956db3ca
CW
1838 INIT_LIST_HEAD(&queue);
1839
1840 list_add_tail((struct list_head *)&root->stack_list, &queue);
1841
956db3ca
CW
1842 while (!list_empty(&queue)) {
1843 cp = container_of(queue.next, struct cpuset, stack_list);
1844 list_del(queue.next);
1845 list_for_each_entry(cont, &cp->css.cgroup->children, sibling) {
1846 child = cgroup_cs(cont);
1847 list_add_tail(&child->stack_list, &queue);
1848 }
1849 cont = cp->css.cgroup;
b4501295
PJ
1850
1851 /* Continue past cpusets with all cpus, mems online */
1852 if (cpus_subset(cp->cpus_allowed, cpu_online_map) &&
1853 nodes_subset(cp->mems_allowed, node_states[N_HIGH_MEMORY]))
1854 continue;
1855
956db3ca 1856 /* Remove offline cpus and mems from this cpuset. */
b4501295 1857 mutex_lock(&callback_mutex);
956db3ca
CW
1858 cpus_and(cp->cpus_allowed, cp->cpus_allowed, cpu_online_map);
1859 nodes_and(cp->mems_allowed, cp->mems_allowed,
1860 node_states[N_HIGH_MEMORY]);
b4501295
PJ
1861 mutex_unlock(&callback_mutex);
1862
1863 /* Move tasks from the empty cpuset to a parent */
c8d9c90c 1864 if (cpus_empty(cp->cpus_allowed) ||
b4501295 1865 nodes_empty(cp->mems_allowed))
956db3ca 1866 remove_tasks_in_empty_cpuset(cp);
b1aac8bb
PJ
1867 }
1868}
1869
1870/*
1871 * The cpus_allowed and mems_allowed nodemasks in the top_cpuset track
0e1e7c7a 1872 * cpu_online_map and node_states[N_HIGH_MEMORY]. Force the top cpuset to
956db3ca 1873 * track what's online after any CPU or memory node hotplug or unplug event.
b1aac8bb
PJ
1874 *
1875 * Since there are two callers of this routine, one for CPU hotplug
1876 * events and one for memory node hotplug events, we could have coded
1877 * two separate routines here. We code it as a single common routine
1878 * in order to minimize text size.
1879 */
1880
1881static void common_cpu_mem_hotplug_unplug(void)
1882{
8793d854 1883 cgroup_lock();
b1aac8bb 1884
b1aac8bb 1885 top_cpuset.cpus_allowed = cpu_online_map;
0e1e7c7a 1886 top_cpuset.mems_allowed = node_states[N_HIGH_MEMORY];
956db3ca 1887 scan_for_empty_cpusets(&top_cpuset);
b1aac8bb 1888
8793d854 1889 cgroup_unlock();
b1aac8bb 1890}
b1aac8bb 1891
4c4d50f7
PJ
1892/*
1893 * The top_cpuset tracks what CPUs and Memory Nodes are online,
1894 * period. This is necessary in order to make cpusets transparent
1895 * (of no affect) on systems that are actively using CPU hotplug
1896 * but making no active use of cpusets.
1897 *
38837fc7
PJ
1898 * This routine ensures that top_cpuset.cpus_allowed tracks
1899 * cpu_online_map on each CPU hotplug (cpuhp) event.
4c4d50f7
PJ
1900 */
1901
029190c5
PJ
1902static int cpuset_handle_cpuhp(struct notifier_block *unused_nb,
1903 unsigned long phase, void *unused_cpu)
4c4d50f7 1904{
ac076758
AK
1905 if (phase == CPU_DYING || phase == CPU_DYING_FROZEN)
1906 return NOTIFY_DONE;
1907
b1aac8bb 1908 common_cpu_mem_hotplug_unplug();
4c4d50f7
PJ
1909 return 0;
1910}
4c4d50f7 1911
b1aac8bb 1912#ifdef CONFIG_MEMORY_HOTPLUG
38837fc7 1913/*
0e1e7c7a
CL
1914 * Keep top_cpuset.mems_allowed tracking node_states[N_HIGH_MEMORY].
1915 * Call this routine anytime after you change
1916 * node_states[N_HIGH_MEMORY].
38837fc7
PJ
1917 * See also the previous routine cpuset_handle_cpuhp().
1918 */
1919
1af98928 1920void cpuset_track_online_nodes(void)
38837fc7 1921{
b1aac8bb 1922 common_cpu_mem_hotplug_unplug();
38837fc7
PJ
1923}
1924#endif
1925
1da177e4
LT
1926/**
1927 * cpuset_init_smp - initialize cpus_allowed
1928 *
1929 * Description: Finish top cpuset after cpu, node maps are initialized
1930 **/
1931
1932void __init cpuset_init_smp(void)
1933{
1934 top_cpuset.cpus_allowed = cpu_online_map;
0e1e7c7a 1935 top_cpuset.mems_allowed = node_states[N_HIGH_MEMORY];
4c4d50f7
PJ
1936
1937 hotcpu_notifier(cpuset_handle_cpuhp, 0);
1da177e4
LT
1938}
1939
1940/**
3077a260 1941
1da177e4
LT
1942 * cpuset_cpus_allowed - return cpus_allowed mask from a tasks cpuset.
1943 * @tsk: pointer to task_struct from which to obtain cpuset->cpus_allowed.
f9a86fcb 1944 * @pmask: pointer to cpumask_t variable to receive cpus_allowed set.
1da177e4
LT
1945 *
1946 * Description: Returns the cpumask_t cpus_allowed of the cpuset
1947 * attached to the specified @tsk. Guaranteed to return some non-empty
1948 * subset of cpu_online_map, even if this means going outside the
1949 * tasks cpuset.
1950 **/
1951
f9a86fcb 1952void cpuset_cpus_allowed(struct task_struct *tsk, cpumask_t *pmask)
1da177e4 1953{
3d3f26a7 1954 mutex_lock(&callback_mutex);
f9a86fcb 1955 cpuset_cpus_allowed_locked(tsk, pmask);
470fd646 1956 mutex_unlock(&callback_mutex);
470fd646
CW
1957}
1958
1959/**
1960 * cpuset_cpus_allowed_locked - return cpus_allowed mask from a tasks cpuset.
2df167a3 1961 * Must be called with callback_mutex held.
470fd646 1962 **/
f9a86fcb 1963void cpuset_cpus_allowed_locked(struct task_struct *tsk, cpumask_t *pmask)
470fd646 1964{
909d75a3 1965 task_lock(tsk);
f9a86fcb 1966 guarantee_online_cpus(task_cs(tsk), pmask);
909d75a3 1967 task_unlock(tsk);
1da177e4
LT
1968}
1969
1970void cpuset_init_current_mems_allowed(void)
1971{
f9a86fcb 1972 nodes_setall(current->mems_allowed);
1da177e4
LT
1973}
1974
909d75a3
PJ
1975/**
1976 * cpuset_mems_allowed - return mems_allowed mask from a tasks cpuset.
1977 * @tsk: pointer to task_struct from which to obtain cpuset->mems_allowed.
1978 *
1979 * Description: Returns the nodemask_t mems_allowed of the cpuset
1980 * attached to the specified @tsk. Guaranteed to return some non-empty
0e1e7c7a 1981 * subset of node_states[N_HIGH_MEMORY], even if this means going outside the
909d75a3
PJ
1982 * tasks cpuset.
1983 **/
1984
1985nodemask_t cpuset_mems_allowed(struct task_struct *tsk)
1986{
1987 nodemask_t mask;
1988
3d3f26a7 1989 mutex_lock(&callback_mutex);
909d75a3 1990 task_lock(tsk);
8793d854 1991 guarantee_online_mems(task_cs(tsk), &mask);
909d75a3 1992 task_unlock(tsk);
3d3f26a7 1993 mutex_unlock(&callback_mutex);
909d75a3
PJ
1994
1995 return mask;
1996}
1997
d9fd8a6d 1998/**
19770b32
MG
1999 * cpuset_nodemask_valid_mems_allowed - check nodemask vs. curremt mems_allowed
2000 * @nodemask: the nodemask to be checked
d9fd8a6d 2001 *
19770b32 2002 * Are any of the nodes in the nodemask allowed in current->mems_allowed?
1da177e4 2003 */
19770b32 2004int cpuset_nodemask_valid_mems_allowed(nodemask_t *nodemask)
1da177e4 2005{
19770b32 2006 return nodes_intersects(*nodemask, current->mems_allowed);
1da177e4
LT
2007}
2008
9bf2229f 2009/*
78608366
PM
2010 * nearest_hardwall_ancestor() - Returns the nearest mem_exclusive or
2011 * mem_hardwall ancestor to the specified cpuset. Call holding
2012 * callback_mutex. If no ancestor is mem_exclusive or mem_hardwall
2013 * (an unusual configuration), then returns the root cpuset.
9bf2229f 2014 */
78608366 2015static const struct cpuset *nearest_hardwall_ancestor(const struct cpuset *cs)
9bf2229f 2016{
78608366 2017 while (!(is_mem_exclusive(cs) || is_mem_hardwall(cs)) && cs->parent)
9bf2229f
PJ
2018 cs = cs->parent;
2019 return cs;
2020}
2021
d9fd8a6d 2022/**
02a0e53d 2023 * cpuset_zone_allowed_softwall - Can we allocate on zone z's memory node?
9bf2229f 2024 * @z: is this zone on an allowed node?
02a0e53d 2025 * @gfp_mask: memory allocation flags
d9fd8a6d 2026 *
02a0e53d
PJ
2027 * If we're in interrupt, yes, we can always allocate. If
2028 * __GFP_THISNODE is set, yes, we can always allocate. If zone
9bf2229f
PJ
2029 * z's node is in our tasks mems_allowed, yes. If it's not a
2030 * __GFP_HARDWALL request and this zone's nodes is in the nearest
78608366 2031 * hardwalled cpuset ancestor to this tasks cpuset, yes.
c596d9f3
DR
2032 * If the task has been OOM killed and has access to memory reserves
2033 * as specified by the TIF_MEMDIE flag, yes.
9bf2229f
PJ
2034 * Otherwise, no.
2035 *
02a0e53d
PJ
2036 * If __GFP_HARDWALL is set, cpuset_zone_allowed_softwall()
2037 * reduces to cpuset_zone_allowed_hardwall(). Otherwise,
2038 * cpuset_zone_allowed_softwall() might sleep, and might allow a zone
2039 * from an enclosing cpuset.
2040 *
2041 * cpuset_zone_allowed_hardwall() only handles the simpler case of
2042 * hardwall cpusets, and never sleeps.
2043 *
2044 * The __GFP_THISNODE placement logic is really handled elsewhere,
2045 * by forcibly using a zonelist starting at a specified node, and by
2046 * (in get_page_from_freelist()) refusing to consider the zones for
2047 * any node on the zonelist except the first. By the time any such
2048 * calls get to this routine, we should just shut up and say 'yes'.
2049 *
9bf2229f 2050 * GFP_USER allocations are marked with the __GFP_HARDWALL bit,
c596d9f3
DR
2051 * and do not allow allocations outside the current tasks cpuset
2052 * unless the task has been OOM killed as is marked TIF_MEMDIE.
9bf2229f 2053 * GFP_KERNEL allocations are not so marked, so can escape to the
78608366 2054 * nearest enclosing hardwalled ancestor cpuset.
9bf2229f 2055 *
02a0e53d
PJ
2056 * Scanning up parent cpusets requires callback_mutex. The
2057 * __alloc_pages() routine only calls here with __GFP_HARDWALL bit
2058 * _not_ set if it's a GFP_KERNEL allocation, and all nodes in the
2059 * current tasks mems_allowed came up empty on the first pass over
2060 * the zonelist. So only GFP_KERNEL allocations, if all nodes in the
2061 * cpuset are short of memory, might require taking the callback_mutex
2062 * mutex.
9bf2229f 2063 *
36be57ff 2064 * The first call here from mm/page_alloc:get_page_from_freelist()
02a0e53d
PJ
2065 * has __GFP_HARDWALL set in gfp_mask, enforcing hardwall cpusets,
2066 * so no allocation on a node outside the cpuset is allowed (unless
2067 * in interrupt, of course).
36be57ff
PJ
2068 *
2069 * The second pass through get_page_from_freelist() doesn't even call
2070 * here for GFP_ATOMIC calls. For those calls, the __alloc_pages()
2071 * variable 'wait' is not set, and the bit ALLOC_CPUSET is not set
2072 * in alloc_flags. That logic and the checks below have the combined
2073 * affect that:
9bf2229f
PJ
2074 * in_interrupt - any node ok (current task context irrelevant)
2075 * GFP_ATOMIC - any node ok
c596d9f3 2076 * TIF_MEMDIE - any node ok
78608366 2077 * GFP_KERNEL - any node in enclosing hardwalled cpuset ok
9bf2229f 2078 * GFP_USER - only nodes in current tasks mems allowed ok.
36be57ff
PJ
2079 *
2080 * Rule:
02a0e53d 2081 * Don't call cpuset_zone_allowed_softwall if you can't sleep, unless you
36be57ff
PJ
2082 * pass in the __GFP_HARDWALL flag set in gfp_flag, which disables
2083 * the code that might scan up ancestor cpusets and sleep.
02a0e53d 2084 */
9bf2229f 2085
02a0e53d 2086int __cpuset_zone_allowed_softwall(struct zone *z, gfp_t gfp_mask)
1da177e4 2087{
9bf2229f
PJ
2088 int node; /* node that zone z is on */
2089 const struct cpuset *cs; /* current cpuset ancestors */
29afd49b 2090 int allowed; /* is allocation in zone z allowed? */
9bf2229f 2091
9b819d20 2092 if (in_interrupt() || (gfp_mask & __GFP_THISNODE))
9bf2229f 2093 return 1;
89fa3024 2094 node = zone_to_nid(z);
92d1dbd2 2095 might_sleep_if(!(gfp_mask & __GFP_HARDWALL));
9bf2229f
PJ
2096 if (node_isset(node, current->mems_allowed))
2097 return 1;
c596d9f3
DR
2098 /*
2099 * Allow tasks that have access to memory reserves because they have
2100 * been OOM killed to get memory anywhere.
2101 */
2102 if (unlikely(test_thread_flag(TIF_MEMDIE)))
2103 return 1;
9bf2229f
PJ
2104 if (gfp_mask & __GFP_HARDWALL) /* If hardwall request, stop here */
2105 return 0;
2106
5563e770
BP
2107 if (current->flags & PF_EXITING) /* Let dying task have memory */
2108 return 1;
2109
9bf2229f 2110 /* Not hardwall and node outside mems_allowed: scan up cpusets */
3d3f26a7 2111 mutex_lock(&callback_mutex);
053199ed 2112
053199ed 2113 task_lock(current);
78608366 2114 cs = nearest_hardwall_ancestor(task_cs(current));
053199ed
PJ
2115 task_unlock(current);
2116
9bf2229f 2117 allowed = node_isset(node, cs->mems_allowed);
3d3f26a7 2118 mutex_unlock(&callback_mutex);
9bf2229f 2119 return allowed;
1da177e4
LT
2120}
2121
02a0e53d
PJ
2122/*
2123 * cpuset_zone_allowed_hardwall - Can we allocate on zone z's memory node?
2124 * @z: is this zone on an allowed node?
2125 * @gfp_mask: memory allocation flags
2126 *
2127 * If we're in interrupt, yes, we can always allocate.
2128 * If __GFP_THISNODE is set, yes, we can always allocate. If zone
c596d9f3
DR
2129 * z's node is in our tasks mems_allowed, yes. If the task has been
2130 * OOM killed and has access to memory reserves as specified by the
2131 * TIF_MEMDIE flag, yes. Otherwise, no.
02a0e53d
PJ
2132 *
2133 * The __GFP_THISNODE placement logic is really handled elsewhere,
2134 * by forcibly using a zonelist starting at a specified node, and by
2135 * (in get_page_from_freelist()) refusing to consider the zones for
2136 * any node on the zonelist except the first. By the time any such
2137 * calls get to this routine, we should just shut up and say 'yes'.
2138 *
2139 * Unlike the cpuset_zone_allowed_softwall() variant, above,
2140 * this variant requires that the zone be in the current tasks
2141 * mems_allowed or that we're in interrupt. It does not scan up the
2142 * cpuset hierarchy for the nearest enclosing mem_exclusive cpuset.
2143 * It never sleeps.
2144 */
2145
2146int __cpuset_zone_allowed_hardwall(struct zone *z, gfp_t gfp_mask)
2147{
2148 int node; /* node that zone z is on */
2149
2150 if (in_interrupt() || (gfp_mask & __GFP_THISNODE))
2151 return 1;
2152 node = zone_to_nid(z);
2153 if (node_isset(node, current->mems_allowed))
2154 return 1;
dedf8b79
DW
2155 /*
2156 * Allow tasks that have access to memory reserves because they have
2157 * been OOM killed to get memory anywhere.
2158 */
2159 if (unlikely(test_thread_flag(TIF_MEMDIE)))
2160 return 1;
02a0e53d
PJ
2161 return 0;
2162}
2163
505970b9
PJ
2164/**
2165 * cpuset_lock - lock out any changes to cpuset structures
2166 *
3d3f26a7 2167 * The out of memory (oom) code needs to mutex_lock cpusets
505970b9 2168 * from being changed while it scans the tasklist looking for a
3d3f26a7 2169 * task in an overlapping cpuset. Expose callback_mutex via this
505970b9
PJ
2170 * cpuset_lock() routine, so the oom code can lock it, before
2171 * locking the task list. The tasklist_lock is a spinlock, so
3d3f26a7 2172 * must be taken inside callback_mutex.
505970b9
PJ
2173 */
2174
2175void cpuset_lock(void)
2176{
3d3f26a7 2177 mutex_lock(&callback_mutex);
505970b9
PJ
2178}
2179
2180/**
2181 * cpuset_unlock - release lock on cpuset changes
2182 *
2183 * Undo the lock taken in a previous cpuset_lock() call.
2184 */
2185
2186void cpuset_unlock(void)
2187{
3d3f26a7 2188 mutex_unlock(&callback_mutex);
505970b9
PJ
2189}
2190
825a46af
PJ
2191/**
2192 * cpuset_mem_spread_node() - On which node to begin search for a page
2193 *
2194 * If a task is marked PF_SPREAD_PAGE or PF_SPREAD_SLAB (as for
2195 * tasks in a cpuset with is_spread_page or is_spread_slab set),
2196 * and if the memory allocation used cpuset_mem_spread_node()
2197 * to determine on which node to start looking, as it will for
2198 * certain page cache or slab cache pages such as used for file
2199 * system buffers and inode caches, then instead of starting on the
2200 * local node to look for a free page, rather spread the starting
2201 * node around the tasks mems_allowed nodes.
2202 *
2203 * We don't have to worry about the returned node being offline
2204 * because "it can't happen", and even if it did, it would be ok.
2205 *
2206 * The routines calling guarantee_online_mems() are careful to
2207 * only set nodes in task->mems_allowed that are online. So it
2208 * should not be possible for the following code to return an
2209 * offline node. But if it did, that would be ok, as this routine
2210 * is not returning the node where the allocation must be, only
2211 * the node where the search should start. The zonelist passed to
2212 * __alloc_pages() will include all nodes. If the slab allocator
2213 * is passed an offline node, it will fall back to the local node.
2214 * See kmem_cache_alloc_node().
2215 */
2216
2217int cpuset_mem_spread_node(void)
2218{
2219 int node;
2220
2221 node = next_node(current->cpuset_mem_spread_rotor, current->mems_allowed);
2222 if (node == MAX_NUMNODES)
2223 node = first_node(current->mems_allowed);
2224 current->cpuset_mem_spread_rotor = node;
2225 return node;
2226}
2227EXPORT_SYMBOL_GPL(cpuset_mem_spread_node);
2228
ef08e3b4 2229/**
bbe373f2
DR
2230 * cpuset_mems_allowed_intersects - Does @tsk1's mems_allowed intersect @tsk2's?
2231 * @tsk1: pointer to task_struct of some task.
2232 * @tsk2: pointer to task_struct of some other task.
2233 *
2234 * Description: Return true if @tsk1's mems_allowed intersects the
2235 * mems_allowed of @tsk2. Used by the OOM killer to determine if
2236 * one of the task's memory usage might impact the memory available
2237 * to the other.
ef08e3b4
PJ
2238 **/
2239
bbe373f2
DR
2240int cpuset_mems_allowed_intersects(const struct task_struct *tsk1,
2241 const struct task_struct *tsk2)
ef08e3b4 2242{
bbe373f2 2243 return nodes_intersects(tsk1->mems_allowed, tsk2->mems_allowed);
ef08e3b4
PJ
2244}
2245
3e0d98b9
PJ
2246/*
2247 * Collection of memory_pressure is suppressed unless
2248 * this flag is enabled by writing "1" to the special
2249 * cpuset file 'memory_pressure_enabled' in the root cpuset.
2250 */
2251
c5b2aff8 2252int cpuset_memory_pressure_enabled __read_mostly;
3e0d98b9
PJ
2253
2254/**
2255 * cpuset_memory_pressure_bump - keep stats of per-cpuset reclaims.
2256 *
2257 * Keep a running average of the rate of synchronous (direct)
2258 * page reclaim efforts initiated by tasks in each cpuset.
2259 *
2260 * This represents the rate at which some task in the cpuset
2261 * ran low on memory on all nodes it was allowed to use, and
2262 * had to enter the kernels page reclaim code in an effort to
2263 * create more free memory by tossing clean pages or swapping
2264 * or writing dirty pages.
2265 *
2266 * Display to user space in the per-cpuset read-only file
2267 * "memory_pressure". Value displayed is an integer
2268 * representing the recent rate of entry into the synchronous
2269 * (direct) page reclaim by any task attached to the cpuset.
2270 **/
2271
2272void __cpuset_memory_pressure_bump(void)
2273{
3e0d98b9 2274 task_lock(current);
8793d854 2275 fmeter_markevent(&task_cs(current)->fmeter);
3e0d98b9
PJ
2276 task_unlock(current);
2277}
2278
8793d854 2279#ifdef CONFIG_PROC_PID_CPUSET
1da177e4
LT
2280/*
2281 * proc_cpuset_show()
2282 * - Print tasks cpuset path into seq_file.
2283 * - Used for /proc/<pid>/cpuset.
053199ed
PJ
2284 * - No need to task_lock(tsk) on this tsk->cpuset reference, as it
2285 * doesn't really matter if tsk->cpuset changes after we read it,
c8d9c90c 2286 * and we take cgroup_mutex, keeping cpuset_attach() from changing it
2df167a3 2287 * anyway.
1da177e4 2288 */
029190c5 2289static int proc_cpuset_show(struct seq_file *m, void *unused_v)
1da177e4 2290{
13b41b09 2291 struct pid *pid;
1da177e4
LT
2292 struct task_struct *tsk;
2293 char *buf;
8793d854 2294 struct cgroup_subsys_state *css;
99f89551 2295 int retval;
1da177e4 2296
99f89551 2297 retval = -ENOMEM;
1da177e4
LT
2298 buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
2299 if (!buf)
99f89551
EB
2300 goto out;
2301
2302 retval = -ESRCH;
13b41b09
EB
2303 pid = m->private;
2304 tsk = get_pid_task(pid, PIDTYPE_PID);
99f89551
EB
2305 if (!tsk)
2306 goto out_free;
1da177e4 2307
99f89551 2308 retval = -EINVAL;
8793d854
PM
2309 cgroup_lock();
2310 css = task_subsys_state(tsk, cpuset_subsys_id);
2311 retval = cgroup_path(css->cgroup, buf, PAGE_SIZE);
1da177e4 2312 if (retval < 0)
99f89551 2313 goto out_unlock;
1da177e4
LT
2314 seq_puts(m, buf);
2315 seq_putc(m, '\n');
99f89551 2316out_unlock:
8793d854 2317 cgroup_unlock();
99f89551
EB
2318 put_task_struct(tsk);
2319out_free:
1da177e4 2320 kfree(buf);
99f89551 2321out:
1da177e4
LT
2322 return retval;
2323}
2324
2325static int cpuset_open(struct inode *inode, struct file *file)
2326{
13b41b09
EB
2327 struct pid *pid = PROC_I(inode)->pid;
2328 return single_open(file, proc_cpuset_show, pid);
1da177e4
LT
2329}
2330
9a32144e 2331const struct file_operations proc_cpuset_operations = {
1da177e4
LT
2332 .open = cpuset_open,
2333 .read = seq_read,
2334 .llseek = seq_lseek,
2335 .release = single_release,
2336};
8793d854 2337#endif /* CONFIG_PROC_PID_CPUSET */
1da177e4
LT
2338
2339/* Display task cpus_allowed, mems_allowed in /proc/<pid>/status file. */
df5f8314
EB
2340void cpuset_task_status_allowed(struct seq_file *m, struct task_struct *task)
2341{
2342 seq_printf(m, "Cpus_allowed:\t");
2343 m->count += cpumask_scnprintf(m->buf + m->count, m->size - m->count,
2344 task->cpus_allowed);
2345 seq_printf(m, "\n");
39106dcf
MT
2346 seq_printf(m, "Cpus_allowed_list:\t");
2347 m->count += cpulist_scnprintf(m->buf + m->count, m->size - m->count,
2348 task->cpus_allowed);
2349 seq_printf(m, "\n");
df5f8314
EB
2350 seq_printf(m, "Mems_allowed:\t");
2351 m->count += nodemask_scnprintf(m->buf + m->count, m->size - m->count,
2352 task->mems_allowed);
2353 seq_printf(m, "\n");
39106dcf
MT
2354 seq_printf(m, "Mems_allowed_list:\t");
2355 m->count += nodelist_scnprintf(m->buf + m->count, m->size - m->count,
2356 task->mems_allowed);
2357 seq_printf(m, "\n");
1da177e4 2358}