cpuset.c 76.2 KB
Newer Older
Linus Torvalds's avatar
Linus Torvalds committed
1 2 3 4 5 6
/*
 *  kernel/cpuset.c
 *
 *  Processor and Memory placement constraints for sets of tasks.
 *
 *  Copyright (C) 2003 BULL SA.
Paul Jackson's avatar
Paul Jackson committed
7
 *  Copyright (C) 2004-2007 Silicon Graphics, Inc.
8
 *  Copyright (C) 2006 Google, Inc
Linus Torvalds's avatar
Linus Torvalds committed
9 10 11 12
 *
 *  Portions derived from Patrick Mochel's sysfs code.
 *  sysfs is Copyright (c) 2001-3 Patrick Mochel
 *
13
 *  2003-10-10 Written by Simon Derr.
Linus Torvalds's avatar
Linus Torvalds committed
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
17 18
 *  2008 Rework of the scheduler domains and CPU hotplug handling
 *       by Max Krasnyansky
Linus Torvalds's avatar
Linus Torvalds committed
19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36
 *
 *  This file is subject to the terms and conditions of the GNU General Public
 *  License.  See the file COPYING in the main directory of the Linux
 *  distribution for more details.
 */

#include <linux/cpu.h>
#include <linux/cpumask.h>
#include <linux/cpuset.h>
#include <linux/err.h>
#include <linux/errno.h>
#include <linux/file.h>
#include <linux/fs.h>
#include <linux/init.h>
#include <linux/interrupt.h>
#include <linux/kernel.h>
#include <linux/kmod.h>
#include <linux/list.h>
37
#include <linux/mempolicy.h>
Linus Torvalds's avatar
Linus Torvalds committed
38
#include <linux/mm.h>
39
#include <linux/memory.h>
40
#include <linux/export.h>
Linus Torvalds's avatar
Linus Torvalds committed
41 42 43 44
#include <linux/mount.h>
#include <linux/namei.h>
#include <linux/pagemap.h>
#include <linux/proc_fs.h>
45
#include <linux/rcupdate.h>
Linus Torvalds's avatar
Linus Torvalds committed
46 47
#include <linux/sched.h>
#include <linux/seq_file.h>
48
#include <linux/security.h>
Linus Torvalds's avatar
Linus Torvalds committed
49 50 51 52 53
#include <linux/slab.h>
#include <linux/spinlock.h>
#include <linux/stat.h>
#include <linux/string.h>
#include <linux/time.h>
54
#include <linux/time64.h>
Linus Torvalds's avatar
Linus Torvalds committed
55 56 57 58
#include <linux/backing-dev.h>
#include <linux/sort.h>

#include <asm/uaccess.h>
Arun Sharma's avatar
Arun Sharma committed
59
#include <linux/atomic.h>
60
#include <linux/mutex.h>
61
#include <linux/cgroup.h>
62
#include <linux/wait.h>
Linus Torvalds's avatar
Linus Torvalds committed
63

64
DEFINE_STATIC_KEY_FALSE(cpusets_enabled_key);
65

66 67 68 69 70
/* See "Frequency meter" comments, below. */

struct fmeter {
	int cnt;		/* unprocessed events count */
	int val;		/* most recent output value */
71
	time64_t time;		/* clock (secs) when val computed */
72 73 74
	spinlock_t lock;	/* guards read or write of above */
};

Linus Torvalds's avatar
Linus Torvalds committed
75
struct cpuset {
76 77
	struct cgroup_subsys_state css;

Linus Torvalds's avatar
Linus Torvalds committed
78
	unsigned long flags;		/* "unsigned long" so bitops work */
79

80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99
	/*
	 * On default hierarchy:
	 *
	 * The user-configured masks can only be changed by writing to
	 * cpuset.cpus and cpuset.mems, and won't be limited by the
	 * parent masks.
	 *
	 * The effective masks is the real masks that apply to the tasks
	 * in the cpuset. They may be changed if the configured masks are
	 * changed or hotplug happens.
	 *
	 * effective_mask == configured_mask & parent's effective_mask,
	 * and if it ends up empty, it will inherit the parent's mask.
	 *
	 *
	 * On legacy hierachy:
	 *
	 * The user-configured masks are always the same with effective masks.
	 */

100 101 102 103 104 105 106
	/* user-configured CPUs and Memory Nodes allow to tasks */
	cpumask_var_t cpus_allowed;
	nodemask_t mems_allowed;

	/* effective CPUs and Memory Nodes allow to tasks */
	cpumask_var_t effective_cpus;
	nodemask_t effective_mems;
Linus Torvalds's avatar
Linus Torvalds committed
107

108 109 110 111 112 113 114 115 116 117 118 119
	/*
	 * This is old Memory Nodes tasks took on.
	 *
	 * - top_cpuset.old_mems_allowed is initialized to mems_allowed.
	 * - A new cpuset's old_mems_allowed is initialized when some
	 *   task is moved into it.
	 * - old_mems_allowed is used in cpuset_migrate_mm() when we change
	 *   cpuset.mems_allowed and have tasks' nodemask updated, and
	 *   then old_mems_allowed is updated to mems_allowed.
	 */
	nodemask_t old_mems_allowed;

120
	struct fmeter fmeter;		/* memory_pressure filter */
Paul Jackson's avatar
Paul Jackson committed
121

122 123 124 125 126 127
	/*
	 * Tasks are being attached to this cpuset.  Used to prevent
	 * zeroing cpus/mems_allowed between ->can_attach() and ->attach().
	 */
	int attach_in_progress;

Paul Jackson's avatar
Paul Jackson committed
128 129
	/* partition number for rebuild_sched_domains() */
	int pn;
130

131 132
	/* for custom sched domain */
	int relax_domain_level;
Linus Torvalds's avatar
Linus Torvalds committed
133 134
};

135
static inline struct cpuset *css_cs(struct cgroup_subsys_state *css)
136
{
137
	return css ? container_of(css, struct cpuset, css) : NULL;
138 139 140 141 142
}

/* Retrieve the cpuset for a task */
static inline struct cpuset *task_cs(struct task_struct *task)
{
143
	return css_cs(task_css(task, cpuset_cgrp_id));
144 145
}

146
static inline struct cpuset *parent_cs(struct cpuset *cs)
Tejun Heo's avatar
Tejun Heo committed
147
{
Tejun Heo's avatar
Tejun Heo committed
148
	return css_cs(cs->css.parent);
Tejun Heo's avatar
Tejun Heo committed
149 150
}

151 152 153 154 155 156 157 158 159 160 161 162 163
#ifdef CONFIG_NUMA
static inline bool task_has_mempolicy(struct task_struct *task)
{
	return task->mempolicy;
}
#else
static inline bool task_has_mempolicy(struct task_struct *task)
{
	return false;
}
#endif


Linus Torvalds's avatar
Linus Torvalds committed
164 165
/* bits in struct cpuset flags field */
typedef enum {
Tejun Heo's avatar
Tejun Heo committed
166
	CS_ONLINE,
Linus Torvalds's avatar
Linus Torvalds committed
167 168
	CS_CPU_EXCLUSIVE,
	CS_MEM_EXCLUSIVE,
169
	CS_MEM_HARDWALL,
170
	CS_MEMORY_MIGRATE,
Paul Jackson's avatar
Paul Jackson committed
171
	CS_SCHED_LOAD_BALANCE,
172 173
	CS_SPREAD_PAGE,
	CS_SPREAD_SLAB,
Linus Torvalds's avatar
Linus Torvalds committed
174 175 176
} cpuset_flagbits_t;

/* convenient tests for these bits */
Tejun Heo's avatar
Tejun Heo committed
177 178 179 180 181
static inline bool is_cpuset_online(const struct cpuset *cs)
{
	return test_bit(CS_ONLINE, &cs->flags);
}

Linus Torvalds's avatar
Linus Torvalds committed
182 183
static inline int is_cpu_exclusive(const struct cpuset *cs)
{
184
	return test_bit(CS_CPU_EXCLUSIVE, &cs->flags);
Linus Torvalds's avatar
Linus Torvalds committed
185 186 187 188
}

static inline int is_mem_exclusive(const struct cpuset *cs)
{
189
	return test_bit(CS_MEM_EXCLUSIVE, &cs->flags);
Linus Torvalds's avatar
Linus Torvalds committed
190 191
}

192 193 194 195 196
static inline int is_mem_hardwall(const struct cpuset *cs)
{
	return test_bit(CS_MEM_HARDWALL, &cs->flags);
}

Paul Jackson's avatar
Paul Jackson committed
197 198 199 200 201
static inline int is_sched_load_balance(const struct cpuset *cs)
{
	return test_bit(CS_SCHED_LOAD_BALANCE, &cs->flags);
}

202 203
static inline int is_memory_migrate(const struct cpuset *cs)
{
204
	return test_bit(CS_MEMORY_MIGRATE, &cs->flags);
205 206
}

207 208 209 210 211 212 213 214 215 216
static inline int is_spread_page(const struct cpuset *cs)
{
	return test_bit(CS_SPREAD_PAGE, &cs->flags);
}

static inline int is_spread_slab(const struct cpuset *cs)
{
	return test_bit(CS_SPREAD_SLAB, &cs->flags);
}

Linus Torvalds's avatar
Linus Torvalds committed
217
static struct cpuset top_cpuset = {
Tejun Heo's avatar
Tejun Heo committed
218 219
	.flags = ((1 << CS_ONLINE) | (1 << CS_CPU_EXCLUSIVE) |
		  (1 << CS_MEM_EXCLUSIVE)),
Linus Torvalds's avatar
Linus Torvalds committed
220 221
};

222 223 224
/**
 * cpuset_for_each_child - traverse online children of a cpuset
 * @child_cs: loop cursor pointing to the current child
225
 * @pos_css: used for iteration
226 227 228 229 230
 * @parent_cs: target cpuset to walk children of
 *
 * Walk @child_cs through the online children of @parent_cs.  Must be used
 * with RCU read locked.
 */
231 232 233
#define cpuset_for_each_child(child_cs, pos_css, parent_cs)		\
	css_for_each_child((pos_css), &(parent_cs)->css)		\
		if (is_cpuset_online(((child_cs) = css_cs((pos_css)))))
234

235 236 237
/**
 * cpuset_for_each_descendant_pre - pre-order walk of a cpuset's descendants
 * @des_cs: loop cursor pointing to the current descendant
238
 * @pos_css: used for iteration
239 240 241
 * @root_cs: target cpuset to walk ancestor of
 *
 * Walk @des_cs through the online descendants of @root_cs.  Must be used
242
 * with RCU read locked.  The caller may modify @pos_css by calling
243 244
 * css_rightmost_descendant() to skip subtree.  @root_cs is included in the
 * iteration and the first node to be visited.
245
 */
246 247 248
#define cpuset_for_each_descendant_pre(des_cs, pos_css, root_cs)	\
	css_for_each_descendant_pre((pos_css), &(root_cs)->css)		\
		if (is_cpuset_online(((des_cs) = css_cs((pos_css)))))
249

Linus Torvalds's avatar
Linus Torvalds committed
250
/*
251 252 253 254
 * There are two global locks guarding cpuset structures - cpuset_mutex and
 * callback_lock. We also require taking task_lock() when dereferencing a
 * task's cpuset pointer. See "The task_lock() exception", at the end of this
 * comment.
255
 *
256
 * A task must hold both locks to modify cpusets.  If a task holds
257
 * cpuset_mutex, then it blocks others wanting that mutex, ensuring that it
258
 * is the only task able to also acquire callback_lock and be able to
259 260 261
 * modify cpusets.  It can perform various checks on the cpuset structure
 * first, knowing nothing will change.  It can also allocate memory while
 * just holding cpuset_mutex.  While it is performing these checks, various
262 263
 * callback routines can briefly acquire callback_lock to query cpusets.
 * Once it is ready to make the changes, it takes callback_lock, blocking
264
 * everyone else.
265 266
 *
 * Calls to the kernel memory allocator can not be made while holding
267
 * callback_lock, as that would risk double tripping on callback_lock
268 269 270
 * from one of the callbacks into the cpuset code from within
 * __alloc_pages().
 *
271
 * If a task is only holding callback_lock, then it has read-only
272 273
 * access to cpusets.
 *
274 275 276
 * Now, the task_struct fields mems_allowed and mempolicy may be changed
 * by other task, we use alloc_lock in the task_struct fields to protect
 * them.
277
 *
278
 * The cpuset_common_file_read() handlers only hold callback_lock across
279 280 281
 * small pieces of code, such as when reading out possibly multi-word
 * cpumasks and nodemasks.
 *
282 283
 * Accessing a task's cpuset should be done in accordance with the
 * guidelines for accessing subsystem state in kernel/cgroup.c
Linus Torvalds's avatar
Linus Torvalds committed
284 285
 */

286
static DEFINE_MUTEX(cpuset_mutex);
287
static DEFINE_SPINLOCK(callback_lock);
288

289 290
static struct workqueue_struct *cpuset_migrate_mm_wq;

291 292 293 294 295 296
/*
 * CPU / memory hotplug is handled asynchronously.
 */
static void cpuset_hotplug_workfn(struct work_struct *work);
static DECLARE_WORK(cpuset_hotplug_work, cpuset_hotplug_workfn);

297 298
static DECLARE_WAIT_QUEUE_HEAD(cpuset_attach_wq);

299 300
/*
 * This is ugly, but preserves the userspace API for existing cpuset
301
 * users. If someone tries to mount the "cpuset" filesystem, we
302 303
 * silently switch it to mount "cgroup" instead
 */
Al Viro's avatar
Al Viro committed
304 305
static struct dentry *cpuset_mount(struct file_system_type *fs_type,
			 int flags, const char *unused_dev_name, void *data)
Linus Torvalds's avatar
Linus Torvalds committed
306
{
307
	struct file_system_type *cgroup_fs = get_fs_type("cgroup");
Al Viro's avatar
Al Viro committed
308
	struct dentry *ret = ERR_PTR(-ENODEV);
309 310 311 312
	if (cgroup_fs) {
		char mountopts[] =
			"cpuset,noprefix,"
			"release_agent=/sbin/cpuset_release_agent";
Al Viro's avatar
Al Viro committed
313 314
		ret = cgroup_fs->mount(cgroup_fs, flags,
					   unused_dev_name, mountopts);
315 316 317
		put_filesystem(cgroup_fs);
	}
	return ret;
Linus Torvalds's avatar
Linus Torvalds committed
318 319 320 321
}

static struct file_system_type cpuset_fs_type = {
	.name = "cpuset",
Al Viro's avatar
Al Viro committed
322
	.mount = cpuset_mount,
Linus Torvalds's avatar
Linus Torvalds committed
323 324 325
};

/*
326
 * Return in pmask the portion of a cpusets's cpus_allowed that
Linus Torvalds's avatar
Linus Torvalds committed
327
 * are online.  If none are online, walk up the cpuset hierarchy
328 329
 * until we find one that does have some online cpus.  The top
 * cpuset always has some cpus online.
Linus Torvalds's avatar
Linus Torvalds committed
330 331
 *
 * One way or another, we guarantee to return some non-empty subset
332
 * of cpu_online_mask.
Linus Torvalds's avatar
Linus Torvalds committed
333
 *
334
 * Call with callback_lock or cpuset_mutex held.
Linus Torvalds's avatar
Linus Torvalds committed
335
 */
336
static void guarantee_online_cpus(struct cpuset *cs, struct cpumask *pmask)
Linus Torvalds's avatar
Linus Torvalds committed
337
{
338
	while (!cpumask_intersects(cs->effective_cpus, cpu_online_mask))
Tejun Heo's avatar
Tejun Heo committed
339
		cs = parent_cs(cs);
340
	cpumask_and(pmask, cs->effective_cpus, cpu_online_mask);
Linus Torvalds's avatar
Linus Torvalds committed
341 342 343 344
}

/*
 * Return in *pmask the portion of a cpusets's mems_allowed that
345 346
 * are online, with memory.  If none are online with memory, walk
 * up the cpuset hierarchy until we find one that does have some
347
 * online mems.  The top cpuset always has some mems online.
Linus Torvalds's avatar
Linus Torvalds committed
348 349
 *
 * One way or another, we guarantee to return some non-empty subset
350
 * of node_states[N_MEMORY].
Linus Torvalds's avatar
Linus Torvalds committed
351
 *
352
 * Call with callback_lock or cpuset_mutex held.
Linus Torvalds's avatar
Linus Torvalds committed
353
 */
354
static void guarantee_online_mems(struct cpuset *cs, nodemask_t *pmask)
Linus Torvalds's avatar
Linus Torvalds committed
355
{
356
	while (!nodes_intersects(cs->effective_mems, node_states[N_MEMORY]))
Tejun Heo's avatar
Tejun Heo committed
357
		cs = parent_cs(cs);
358
	nodes_and(*pmask, cs->effective_mems, node_states[N_MEMORY]);
Linus Torvalds's avatar
Linus Torvalds committed
359 360
}

361 362 363
/*
 * update task's spread flag if cpuset's page/slab spread flag is set
 *
364
 * Call with callback_lock or cpuset_mutex held.
365 366 367 368 369
 */
static void cpuset_update_task_spread_flag(struct cpuset *cs,
					struct task_struct *tsk)
{
	if (is_spread_page(cs))
370
		task_set_spread_page(tsk);
371
	else
372 373
		task_clear_spread_page(tsk);

374
	if (is_spread_slab(cs))
375
		task_set_spread_slab(tsk);
376
	else
377
		task_clear_spread_slab(tsk);
378 379
}

Linus Torvalds's avatar
Linus Torvalds committed
380 381 382 383 384
/*
 * is_cpuset_subset(p, q) - Is cpuset p a subset of cpuset q?
 *
 * One cpuset is a subset of another if all its allowed CPUs and
 * Memory Nodes are a subset of the other, and its exclusive flags
385
 * are only set if the other's are set.  Call holding cpuset_mutex.
Linus Torvalds's avatar
Linus Torvalds committed
386 387 388 389
 */

static int is_cpuset_subset(const struct cpuset *p, const struct cpuset *q)
{
390
	return	cpumask_subset(p->cpus_allowed, q->cpus_allowed) &&
Linus Torvalds's avatar
Linus Torvalds committed
391 392 393 394 395
		nodes_subset(p->mems_allowed, q->mems_allowed) &&
		is_cpu_exclusive(p) <= is_cpu_exclusive(q) &&
		is_mem_exclusive(p) <= is_mem_exclusive(q);
}

396 397 398 399
/**
 * alloc_trial_cpuset - allocate a trial cpuset
 * @cs: the cpuset that the trial cpuset duplicates
 */
400
static struct cpuset *alloc_trial_cpuset(struct cpuset *cs)
401
{
402 403 404 405 406 407
	struct cpuset *trial;

	trial = kmemdup(cs, sizeof(*cs), GFP_KERNEL);
	if (!trial)
		return NULL;

408 409 410 411
	if (!alloc_cpumask_var(&trial->cpus_allowed, GFP_KERNEL))
		goto free_cs;
	if (!alloc_cpumask_var(&trial->effective_cpus, GFP_KERNEL))
		goto free_cpus;
412

413 414
	cpumask_copy(trial->cpus_allowed, cs->cpus_allowed);
	cpumask_copy(trial->effective_cpus, cs->effective_cpus);
415
	return trial;
416 417 418 419 420 421

free_cpus:
	free_cpumask_var(trial->cpus_allowed);
free_cs:
	kfree(trial);
	return NULL;
422 423 424 425 426 427 428 429
}

/**
 * free_trial_cpuset - free the trial cpuset
 * @trial: the trial cpuset to be freed
 */
static void free_trial_cpuset(struct cpuset *trial)
{
430
	free_cpumask_var(trial->effective_cpus);
431
	free_cpumask_var(trial->cpus_allowed);
432 433 434
	kfree(trial);
}

Linus Torvalds's avatar
Linus Torvalds committed
435 436 437 438 439 440 441
/*
 * validate_change() - Used to validate that any proposed cpuset change
 *		       follows the structural rules for cpusets.
 *
 * If we replaced the flag and mask values of the current cpuset
 * (cur) with those values in the trial cpuset (trial), would
 * our various subset and exclusive rules still be valid?  Presumes
442
 * cpuset_mutex held.
Linus Torvalds's avatar
Linus Torvalds committed
443 444 445 446 447 448 449 450 451 452 453 454
 *
 * 'cur' is the address of an actual, in-use cpuset.  Operations
 * such as list traversal that depend on the actual address of the
 * cpuset in the list must use cur below, not trial.
 *
 * 'trial' is the address of bulk structure copy of cur, with
 * perhaps one or more of the fields cpus_allowed, mems_allowed,
 * or flags changed to new, trial values.
 *
 * Return 0 if valid, -errno if not.
 */

455
static int validate_change(struct cpuset *cur, struct cpuset *trial)
Linus Torvalds's avatar
Linus Torvalds committed
456
{
457
	struct cgroup_subsys_state *css;
Linus Torvalds's avatar
Linus Torvalds committed
458
	struct cpuset *c, *par;
459 460 461
	int ret;

	rcu_read_lock();
Linus Torvalds's avatar
Linus Torvalds committed
462 463

	/* Each of our child cpusets must be a subset of us */
464
	ret = -EBUSY;
465
	cpuset_for_each_child(c, css, cur)
466 467
		if (!is_cpuset_subset(c, trial))
			goto out;
Linus Torvalds's avatar
Linus Torvalds committed
468 469

	/* Remaining checks don't apply to root cpuset */
470
	ret = 0;
471
	if (cur == &top_cpuset)
472
		goto out;
Linus Torvalds's avatar
Linus Torvalds committed
473

Tejun Heo's avatar
Tejun Heo committed
474
	par = parent_cs(cur);
475

476
	/* On legacy hiearchy, we must be a subset of our parent cpuset. */
477
	ret = -EACCES;
478 479
	if (!cgroup_subsys_on_dfl(cpuset_cgrp_subsys) &&
	    !is_cpuset_subset(trial, par))
480
		goto out;
Linus Torvalds's avatar
Linus Torvalds committed
481

482 483 484 485
	/*
	 * If either I or some sibling (!= me) is exclusive, we can't
	 * overlap
	 */
486
	ret = -EINVAL;
487
	cpuset_for_each_child(c, css, par) {
Linus Torvalds's avatar
Linus Torvalds committed
488 489
		if ((is_cpu_exclusive(trial) || is_cpu_exclusive(c)) &&
		    c != cur &&
490
		    cpumask_intersects(trial->cpus_allowed, c->cpus_allowed))
491
			goto out;
Linus Torvalds's avatar
Linus Torvalds committed
492 493 494
		if ((is_mem_exclusive(trial) || is_mem_exclusive(c)) &&
		    c != cur &&
		    nodes_intersects(trial->mems_allowed, c->mems_allowed))
495
			goto out;
Linus Torvalds's avatar
Linus Torvalds committed
496 497
	}

498 499
	/*
	 * Cpusets with tasks - existing or newly being attached - can't
500
	 * be changed to have empty cpus_allowed or mems_allowed.
501
	 */
502
	ret = -ENOSPC;
503
	if ((cgroup_is_populated(cur->css.cgroup) || cur->attach_in_progress)) {
504 505 506 507 508 509 510
		if (!cpumask_empty(cur->cpus_allowed) &&
		    cpumask_empty(trial->cpus_allowed))
			goto out;
		if (!nodes_empty(cur->mems_allowed) &&
		    nodes_empty(trial->mems_allowed))
			goto out;
	}
511

512 513 514 515 516 517 518 519 520 521
	/*
	 * We can't shrink if we won't have enough room for SCHED_DEADLINE
	 * tasks.
	 */
	ret = -EBUSY;
	if (is_cpu_exclusive(cur) &&
	    !cpuset_cpumask_can_shrink(cur->cpus_allowed,
				       trial->cpus_allowed))
		goto out;

522 523 524 525
	ret = 0;
out:
	rcu_read_unlock();
	return ret;
Linus Torvalds's avatar
Linus Torvalds committed
526 527
}

528
#ifdef CONFIG_SMP
Paul Jackson's avatar
Paul Jackson committed
529
/*
530
 * Helper routine for generate_sched_domains().
531
 * Do cpusets a, b have overlapping effective cpus_allowed masks?
Paul Jackson's avatar
Paul Jackson committed
532 533 534
 */
static int cpusets_overlap(struct cpuset *a, struct cpuset *b)
{
535
	return cpumask_intersects(a->effective_cpus, b->effective_cpus);
Paul Jackson's avatar
Paul Jackson committed
536 537
}

538 539 540 541 542 543 544 545
static void
update_domain_attr(struct sched_domain_attr *dattr, struct cpuset *c)
{
	if (dattr->relax_domain_level < c->relax_domain_level)
		dattr->relax_domain_level = c->relax_domain_level;
	return;
}

546 547
static void update_domain_attr_tree(struct sched_domain_attr *dattr,
				    struct cpuset *root_cs)
548
{
549
	struct cpuset *cp;
550
	struct cgroup_subsys_state *pos_css;
551

552
	rcu_read_lock();
553
	cpuset_for_each_descendant_pre(cp, pos_css, root_cs) {
554 555
		/* skip the whole subtree if @cp doesn't have any CPU */
		if (cpumask_empty(cp->cpus_allowed)) {
556
			pos_css = css_rightmost_descendant(pos_css);
557
			continue;
558
		}
559 560 561 562

		if (is_sched_load_balance(cp))
			update_domain_attr(dattr, cp);
	}
563
	rcu_read_unlock();
564 565
}

Paul Jackson's avatar
Paul Jackson committed
566
/*
567 568 569 570 571
 * generate_sched_domains()
 *
 * This function builds a partial partition of the systems CPUs
 * A 'partial partition' is a set of non-overlapping subsets whose
 * union is a subset of that set.
572
 * The output of this function needs to be passed to kernel/sched/core.c
573 574 575
 * partition_sched_domains() routine, which will rebuild the scheduler's
 * load balancing domains (sched domains) as specified by that partial
 * partition.
Paul Jackson's avatar
Paul Jackson committed
576
 *
577
 * See "What is sched_load_balance" in Documentation/cgroups/cpusets.txt
Paul Jackson's avatar
Paul Jackson committed
578 579 580 581 582 583 584
 * for a background explanation of this.
 *
 * Does not return errors, on the theory that the callers of this
 * routine would rather not worry about failures to rebuild sched
 * domains when operating in the severe memory shortage situations
 * that could cause allocation failures below.
 *
585
 * Must be called with cpuset_mutex held.
Paul Jackson's avatar
Paul Jackson committed
586 587
 *
 * The three key local variables below are:
588
 *    q  - a linked-list queue of cpuset pointers, used to implement a
Paul Jackson's avatar
Paul Jackson committed
589 590 591 592 593 594 595 596 597 598 599 600
 *	   top-down scan of all cpusets.  This scan loads a pointer
 *	   to each cpuset marked is_sched_load_balance into the
 *	   array 'csa'.  For our purposes, rebuilding the schedulers
 *	   sched domains, we can ignore !is_sched_load_balance cpusets.
 *  csa  - (for CpuSet Array) Array of pointers to all the cpusets
 *	   that need to be load balanced, for convenient iterative
 *	   access by the subsequent code that finds the best partition,
 *	   i.e the set of domains (subsets) of CPUs such that the
 *	   cpus_allowed of every cpuset marked is_sched_load_balance
 *	   is a subset of one of these domains, while there are as
 *	   many such domains as possible, each as small as possible.
 * doms  - Conversion of 'csa' to an array of cpumasks, for passing to
601
 *	   the kernel/sched/core.c routine partition_sched_domains() in a
Paul Jackson's avatar
Paul Jackson committed
602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619
 *	   convenient format, that can be easily compared to the prior
 *	   value to determine what partition elements (sched domains)
 *	   were changed (added or removed.)
 *
 * Finding the best partition (set of domains):
 *	The triple nested loops below over i, j, k scan over the
 *	load balanced cpusets (using the array of cpuset pointers in
 *	csa[]) looking for pairs of cpusets that have overlapping
 *	cpus_allowed, but which don't have the same 'pn' partition
 *	number and gives them in the same partition number.  It keeps
 *	looping on the 'restart' label until it can no longer find
 *	any such pairs.
 *
 *	The union of the cpus_allowed masks from the set of
 *	all cpusets having the same 'pn' value then form the one
 *	element of the partition (one sched domain) to be passed to
 *	partition_sched_domains().
 */
620
static int generate_sched_domains(cpumask_var_t **domains,
621
			struct sched_domain_attr **attributes)
Paul Jackson's avatar
Paul Jackson committed
622 623 624 625 626
{
	struct cpuset *cp;	/* scans q */
	struct cpuset **csa;	/* array of all cpuset ptrs */
	int csn;		/* how many cpuset ptrs in csa so far */
	int i, j, k;		/* indices for partition finding loops */
627
	cpumask_var_t *doms;	/* resulting partition; i.e. sched domains */
628
	cpumask_var_t non_isolated_cpus;  /* load balanced CPUs */
629
	struct sched_domain_attr *dattr;  /* attributes for custom domains */
630
	int ndoms = 0;		/* number of sched domains in result */
631
	int nslot;		/* next empty doms[] struct cpumask slot */
632
	struct cgroup_subsys_state *pos_css;
Paul Jackson's avatar
Paul Jackson committed
633 634

	doms = NULL;
635
	dattr = NULL;
636
	csa = NULL;
Paul Jackson's avatar
Paul Jackson committed
637

638 639 640 641
	if (!alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL))
		goto done;
	cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map);

Paul Jackson's avatar
Paul Jackson committed
642 643
	/* Special case for the 99% of systems with one, full, sched domain */
	if (is_sched_load_balance(&top_cpuset)) {
644 645
		ndoms = 1;
		doms = alloc_sched_domains(ndoms);
Paul Jackson's avatar
Paul Jackson committed
646
		if (!doms)
647 648
			goto done;

649 650 651
		dattr = kmalloc(sizeof(struct sched_domain_attr), GFP_KERNEL);
		if (dattr) {
			*dattr = SD_ATTR_INIT;
652
			update_domain_attr_tree(dattr, &top_cpuset);
653
		}
654 655
		cpumask_and(doms[0], top_cpuset.effective_cpus,
				     non_isolated_cpus);
656 657

		goto done;
Paul Jackson's avatar
Paul Jackson committed
658 659
	}

660
	csa = kmalloc(nr_cpusets() * sizeof(cp), GFP_KERNEL);
Paul Jackson's avatar
Paul Jackson committed
661 662 663 664
	if (!csa)
		goto done;
	csn = 0;

665
	rcu_read_lock();
666
	cpuset_for_each_descendant_pre(cp, pos_css, &top_cpuset) {
667 668
		if (cp == &top_cpuset)
			continue;
669
		/*
670 671 672 673 674 675
		 * Continue traversing beyond @cp iff @cp has some CPUs and
		 * isn't load balancing.  The former is obvious.  The
		 * latter: All child cpusets contain a subset of the
		 * parent's cpus, so just skip them, and then we call
		 * update_domain_attr_tree() to calc relax_domain_level of
		 * the corresponding sched domain.
676
		 */
677
		if (!cpumask_empty(cp->cpus_allowed) &&
678 679
		    !(is_sched_load_balance(cp) &&
		      cpumask_intersects(cp->cpus_allowed, non_isolated_cpus)))
680
			continue;
681

682 683 684 685
		if (is_sched_load_balance(cp))
			csa[csn++] = cp;

		/* skip @cp's subtree */
686
		pos_css = css_rightmost_descendant(pos_css);
687 688
	}
	rcu_read_unlock();
Paul Jackson's avatar
Paul Jackson committed
689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716

	for (i = 0; i < csn; i++)
		csa[i]->pn = i;
	ndoms = csn;

restart:
	/* Find the best partition (set of sched domains) */
	for (i = 0; i < csn; i++) {
		struct cpuset *a = csa[i];
		int apn = a->pn;

		for (j = 0; j < csn; j++) {
			struct cpuset *b = csa[j];
			int bpn = b->pn;

			if (apn != bpn && cpusets_overlap(a, b)) {
				for (k = 0; k < csn; k++) {
					struct cpuset *c = csa[k];

					if (c->pn == bpn)
						c->pn = apn;
				}
				ndoms--;	/* one less element */
				goto restart;
			}
		}
	}

717 718 719 720
	/*
	 * Now we know how many domains to create.
	 * Convert <csn, csa> to <ndoms, doms> and populate cpu masks.
	 */
721
	doms = alloc_sched_domains(ndoms);
722
	if (!doms)
723 724 725 726 727 728
		goto done;

	/*
	 * The rest of the code, including the scheduler, can deal with
	 * dattr==NULL case. No need to abort if alloc fails.
	 */
729
	dattr = kmalloc(ndoms * sizeof(struct sched_domain_attr), GFP_KERNEL);
Paul Jackson's avatar
Paul Jackson committed
730 731 732

	for (nslot = 0, i = 0; i < csn; i++) {
		struct cpuset *a = csa[i];
733
		struct cpumask *dp;
Paul Jackson's avatar
Paul Jackson committed
734 735
		int apn = a->pn;

736 737 738 739 740
		if (apn < 0) {
			/* Skip completed partitions */
			continue;
		}

741
		dp = doms[nslot];
742 743 744 745

		if (nslot == ndoms) {
			static int warnings = 10;
			if (warnings) {
746 747
				pr_warn("rebuild_sched_domains confused: nslot %d, ndoms %d, csn %d, i %d, apn %d\n",
					nslot, ndoms, csn, i, apn);
748
				warnings--;
Paul Jackson's avatar
Paul Jackson committed
749
			}
750 751
			continue;
		}
Paul Jackson's avatar
Paul Jackson committed
752

753
		cpumask_clear(dp);
754 755 756 757 758 759
		if (dattr)
			*(dattr + nslot) = SD_ATTR_INIT;
		for (j = i; j < csn; j++) {
			struct cpuset *b = csa[j];

			if (apn == b->pn) {
760
				cpumask_or(dp, dp, b->effective_cpus);
761
				cpumask_and(dp, dp, non_isolated_cpus);
762 763 764 765 766
				if (dattr)
					update_domain_attr_tree(dattr + nslot, b);

				/* Done with this partition */
				b->pn = -1;
Paul Jackson's avatar
Paul Jackson committed
767 768
			}
		}
769
		nslot++;
Paul Jackson's avatar
Paul Jackson committed
770 771 772
	}
	BUG_ON(nslot != ndoms);

773
done:
774
	free_cpumask_var(non_isolated_cpus);
775 776
	kfree(csa);

777 778 779 780 781 782 783
	/*
	 * Fallback to the default domain if kmalloc() failed.
	 * See comments in partition_sched_domains().
	 */
	if (doms == NULL)
		ndoms = 1;

784 785 786 787 788 789 790 791
	*domains    = doms;
	*attributes = dattr;
	return ndoms;
}

/*
 * Rebuild scheduler domains.
 *
792 793 794 795 796
 * If the flag 'sched_load_balance' of any cpuset with non-empty
 * 'cpus' changes, or if the 'cpus' allowed changes in any cpuset
 * which has that flag enabled, or if any cpuset with a non-empty
 * 'cpus' is removed, then call this routine to rebuild the
 * scheduler's dynamic sched domains.
797
 *
798
 * Call with cpuset_mutex held.  Takes get_online_cpus().
799
 */
800
static void rebuild_sched_domains_locked(void)
801 802
{
	struct sched_domain_attr *attr;
803
	cpumask_var_t *doms;
804 805
	int ndoms;

806
	lockdep_assert_held(&cpuset_mutex);
807
	get_online_cpus();
808

809 810 811 812 813
	/*
	 * We have raced with CPU hotplug. Don't do anything to avoid
	 * passing doms with offlined cpu to partition_sched_domains().
	 * Anyways, hotplug work item will rebuild sched domains.
	 */
814
	if (!cpumask_equal(top_cpuset.effective_cpus, cpu_active_mask))
815 816
		goto out;

817 818 819 820 821
	/* Generate domain masks and attrs */
	ndoms = generate_sched_domains(&doms, &attr);

	/* Have scheduler rebuild the domains */
	partition_sched_domains(ndoms, doms, attr);
822
out:
823
	put_online_cpus();
824
}
825
#else /* !CONFIG_SMP */
826
static void rebuild_sched_domains_locked(void)
827 828 829
{
}
#endif /* CONFIG_SMP */
Paul Jackson's avatar
Paul Jackson committed
830

831 832
void rebuild_sched_domains(void)
{
833
	mutex_lock(&cpuset_mutex);
834
	rebuild_sched_domains_locked();
835
	mutex_unlock(&cpuset_mutex);
Paul Jackson's avatar
Paul Jackson committed
836 837
}

838 839 840 841
/**
 * update_tasks_cpumask - Update the cpumasks of tasks in the cpuset.
 * @cs: the cpuset in which each task's cpus_allowed mask needs to be changed
 *
842 843 844
 * Iterate through each task of @cs updating its cpus_allowed to the
 * effective cpuset's.  As this function is called with cpuset_mutex held,
 * cpuset membership stays stable.
845
 */
846
static void update_tasks_cpumask(struct cpuset *cs)
847
{
848 849 850 851 852
	struct css_task_iter it;
	struct task_struct *task;

	css_task_iter_start(&cs->css, &it);
	while ((task = css_task_iter_next(&it)))
853
		set_cpus_allowed_ptr(task, cs->effective_cpus);
854
	css_task_iter_end(&it);
855 856
}

857
/*
858 859 860 861 862 863
 * update_cpumasks_hier - Update effective cpumasks and tasks in the subtree
 * @cs: the cpuset to consider
 * @new_cpus: temp variable for calculating new effective_cpus
 *
 * When congifured cpumask is changed, the effective cpumasks of this cpuset
 * and all its descendants need to be updated.
864
 *
865
 * On legacy hierachy, effective_cpus will be the same with cpu_allowed.
866 867 868
 *
 * Called with cpuset_mutex held
 */
869
static void update_cpumasks_hier(struct cpuset *cs, struct cpumask *new_cpus)
870 871
{
	struct cpuset *cp;
872
	struct cgroup_subsys_state *pos_css;
873
	bool need_rebuild_sched_domains = false;
874 875

	rcu_read_lock();
876 877 878 879 880
	cpuset_for_each_descendant_pre(cp, pos_css, cs) {
		struct cpuset *parent = parent_cs(cp);

		cpumask_and(new_cpus, cp->cpus_allowed, parent->effective_cpus);

881 882 883 884
		/*
		 * If it becomes empty, inherit the effective mask of the
		 * parent, which is guaranteed to have some CPUs.
		 */
885 886
		if (cgroup_subsys_on_dfl(cpuset_cgrp_subsys) &&
		    cpumask_empty(new_cpus))
887 888
			cpumask_copy(new_cpus, parent->effective_cpus);

889 890 891 892
		/* Skip the whole subtree if the cpumask remains the same. */
		if (cpumask_equal(new_cpus, cp->effective_cpus)) {
			pos_css = css_rightmost_descendant(pos_css);
			continue;
893
		}
894

895
		if (!css_tryget_online(&cp->css))
896 897 898
			continue;
		rcu_read_unlock();

899
		spin_lock_irq(&callback_lock);
900
		cpumask_copy(cp->effective_cpus, new_cpus);
901
		spin_unlock_irq(&callback_lock);
902

903
		WARN_ON(!cgroup_subsys_on_dfl(cpuset_cgrp_subsys) &&
904 905
			!cpumask_equal(cp->cpus_allowed, cp->effective_cpus));

906
		update_tasks_cpumask(cp);
907

908 909 910 911 912 913 914 915
		/*
		 * If the effective cpumask of any non-empty cpuset is changed,
		 * we need to rebuild sched domains.
		 */
		if (!cpumask_empty(cp->cpus_allowed) &&
		    is_sched_load_balance(cp))
			need_rebuild_sched_domains = true;

916 917 918 919
		rcu_read_lock();
		css_put(&cp->css);
	}
	rcu_read_unlock();
920 921 922

	if (need_rebuild_sched_domains)
		rebuild_sched_domains_locked();
923 924
}

925 926 927
/**
 * update_cpumask - update the cpus_allowed mask of a cpuset and all tasks in it
 * @cs: the cpuset to consider
928
 * @trialcs: trial cpuset
929 930
 * @buf: buffer of cpu numbers written to this cpuset
 */
931 932
static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs,
			  const char *buf)
Linus Torvalds's avatar
Linus Torvalds committed
933
{
934
	int retval;
Linus Torvalds's avatar
Linus Torvalds committed
935

936
	/* top_cpuset.cpus_allowed tracks cpu_online_mask; it's read-only */
937 938 939
	if (cs == &top_cpuset)
		return -EACCES;

940
	/*
941
	 * An empty cpus_allowed is ok only if the cpuset has no tasks.
942 943 944
	 * Since cpulist_parse() fails on an empty mask, we special case
	 * that parsing.  The validate_change() call ensures that cpusets
	 * with tasks have cpus.
945
	 */
946
	if (!*buf) {
947
		cpumask_clear(trialcs->cpus_allowed);
948
	} else {
949
		retval = cpulist_parse(buf, trialcs->cpus_allowed);
950 951
		if (retval < 0)
			return retval;
952

953 954
		if (!cpumask_subset(trialcs->cpus_allowed,
				    top_cpuset.cpus_allowed))
955
			return -EINVAL;
956
	}
Paul Jackson's avatar
Paul Jackson committed
957

Paul Menage's avatar
Paul Menage committed
958
	/* Nothing to do if the cpus didn't change */
959
	if (cpumask_equal(cs->cpus_allowed, trialcs->cpus_allowed))
Paul Menage's avatar
Paul Menage committed
960
		return 0;
961

962 963 964 965
	retval = validate_change(cs, trialcs);
	if (retval < 0)
		return retval;

966
	spin_lock_irq(&callback_lock);
967
	cpumask_copy(cs->cpus_allowed, trialcs->cpus_allowed);
968
	spin_unlock_irq(&callback_lock);
Paul Jackson's avatar
Paul Jackson committed
969

970 971
	/* use trialcs->cpus_allowed as a temp variable */
	update_cpumasks_hier(cs, trialcs->cpus_allowed);
972
	return 0;
Linus Torvalds's avatar
Linus Torvalds committed
973 974
}

975
/*
976 977 978 979 980
 * Migrate memory region from one set of nodes to another.  This is
 * performed asynchronously as it can be called from process migration path
 * holding locks involved in process management.  All mm migrations are
 * performed in the queued order and can be waited for by flushing
 * cpuset_migrate_mm_wq.
981 982
 */

983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000
struct cpuset_migrate_mm_work {
	struct work_struct	work;
	struct mm_struct	*mm;
	nodemask_t		from;
	nodemask_t		to;
};

static void cpuset_migrate_mm_workfn(struct work_struct *work)
{
	struct cpuset_migrate_mm_work *mwork =
		container_of(work, struct cpuset_migrate_mm_work, work);

	/* on a wq worker, no need to worry about %current's mems_allowed */
	do_migrate_pages(mwork->mm, &mwork->from, &mwork->to, MPOL_MF_MOVE_ALL);
	mmput(mwork->mm);
	kfree(mwork);
}

1001 1002 1003
static void cpuset_migrate_mm(struct mm_struct *mm, const nodemask_t *from,
							const nodemask_t *to)
{
1004
	struct cpuset_migrate_mm_work *mwork;