cpuset.c 76.1 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 62
#include <linux/workqueue.h>
#include <linux/cgroup.h>
63
#include <linux/wait.h>
Linus Torvalds's avatar
Linus Torvalds committed
64

65
struct static_key cpusets_enabled_key __read_mostly = STATIC_KEY_INIT_FALSE;
66

67 68 69 70 71
/* See "Frequency meter" comments, below. */

struct fmeter {
	int cnt;		/* unprocessed events count */
	int val;		/* most recent output value */
72
	time64_t time;		/* clock (secs) when val computed */
73 74 75
	spinlock_t lock;	/* guards read or write of above */
};

Linus Torvalds's avatar
Linus Torvalds committed
76
struct cpuset {
77 78
	struct cgroup_subsys_state css;

Linus Torvalds's avatar
Linus Torvalds committed
79
	unsigned long flags;		/* "unsigned long" so bitops work */
80

81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100
	/*
	 * 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.
	 */

101 102 103 104 105 106 107
	/* 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
108

109 110 111 112 113 114 115 116 117 118 119 120
	/*
	 * 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;

121
	struct fmeter fmeter;		/* memory_pressure filter */
Paul Jackson's avatar
Paul Jackson committed
122

123 124 125 126 127 128
	/*
	 * 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
129 130
	/* partition number for rebuild_sched_domains() */
	int pn;
131

132 133
	/* for custom sched domain */
	int relax_domain_level;
Linus Torvalds's avatar
Linus Torvalds committed
134 135
};

136
static inline struct cpuset *css_cs(struct cgroup_subsys_state *css)
137
{
138
	return css ? container_of(css, struct cpuset, css) : NULL;
139 140 141 142 143
}

/* Retrieve the cpuset for a task */
static inline struct cpuset *task_cs(struct task_struct *task)
{
144
	return css_cs(task_css(task, cpuset_cgrp_id));
145 146
}

147
static inline struct cpuset *parent_cs(struct cpuset *cs)
Tejun Heo's avatar
Tejun Heo committed
148
{
Tejun Heo's avatar
Tejun Heo committed
149
	return css_cs(cs->css.parent);
Tejun Heo's avatar
Tejun Heo committed
150 151
}

152 153 154 155 156 157 158 159 160 161 162 163 164
#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
165 166
/* bits in struct cpuset flags field */
typedef enum {
Tejun Heo's avatar
Tejun Heo committed
167
	CS_ONLINE,
Linus Torvalds's avatar
Linus Torvalds committed
168 169
	CS_CPU_EXCLUSIVE,
	CS_MEM_EXCLUSIVE,
170
	CS_MEM_HARDWALL,
171
	CS_MEMORY_MIGRATE,
Paul Jackson's avatar
Paul Jackson committed
172
	CS_SCHED_LOAD_BALANCE,
173 174
	CS_SPREAD_PAGE,
	CS_SPREAD_SLAB,
Linus Torvalds's avatar
Linus Torvalds committed
175 176 177
} cpuset_flagbits_t;

/* convenient tests for these bits */
Tejun Heo's avatar
Tejun Heo committed
178 179 180 181 182
static inline bool is_cpuset_online(const struct cpuset *cs)
{
	return test_bit(CS_ONLINE, &cs->flags);
}

Linus Torvalds's avatar
Linus Torvalds committed
183 184
static inline int is_cpu_exclusive(const struct cpuset *cs)
{
185
	return test_bit(CS_CPU_EXCLUSIVE, &cs->flags);
Linus Torvalds's avatar
Linus Torvalds committed
186 187 188 189
}

static inline int is_mem_exclusive(const struct cpuset *cs)
{
190
	return test_bit(CS_MEM_EXCLUSIVE, &cs->flags);
Linus Torvalds's avatar
Linus Torvalds committed
191 192
}

193 194 195 196 197
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
198 199 200 201 202
static inline int is_sched_load_balance(const struct cpuset *cs)
{
	return test_bit(CS_SCHED_LOAD_BALANCE, &cs->flags);
}

203 204
static inline int is_memory_migrate(const struct cpuset *cs)
{
205
	return test_bit(CS_MEMORY_MIGRATE, &cs->flags);
206 207
}

208 209 210 211 212 213 214 215 216 217
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
218
static struct cpuset top_cpuset = {
Tejun Heo's avatar
Tejun Heo committed
219 220
	.flags = ((1 << CS_ONLINE) | (1 << CS_CPU_EXCLUSIVE) |
		  (1 << CS_MEM_EXCLUSIVE)),
Linus Torvalds's avatar
Linus Torvalds committed
221 222
};

223 224 225
/**
 * cpuset_for_each_child - traverse online children of a cpuset
 * @child_cs: loop cursor pointing to the current child
226
 * @pos_css: used for iteration
227 228 229 230 231
 * @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.
 */
232 233 234
#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)))))
235

236 237 238
/**
 * cpuset_for_each_descendant_pre - pre-order walk of a cpuset's descendants
 * @des_cs: loop cursor pointing to the current descendant
239
 * @pos_css: used for iteration
240 241 242
 * @root_cs: target cpuset to walk ancestor of
 *
 * Walk @des_cs through the online descendants of @root_cs.  Must be used
243
 * with RCU read locked.  The caller may modify @pos_css by calling
244 245
 * css_rightmost_descendant() to skip subtree.  @root_cs is included in the
 * iteration and the first node to be visited.
246
 */
247 248 249
#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)))))
250

Linus Torvalds's avatar
Linus Torvalds committed
251
/*
252 253 254 255
 * 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.
256
 *
257
 * A task must hold both locks to modify cpusets.  If a task holds
258
 * cpuset_mutex, then it blocks others wanting that mutex, ensuring that it
259
 * is the only task able to also acquire callback_lock and be able to
260 261 262
 * 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
263 264
 * callback routines can briefly acquire callback_lock to query cpusets.
 * Once it is ready to make the changes, it takes callback_lock, blocking
265
 * everyone else.
266 267
 *
 * Calls to the kernel memory allocator can not be made while holding
268
 * callback_lock, as that would risk double tripping on callback_lock
269 270 271
 * from one of the callbacks into the cpuset code from within
 * __alloc_pages().
 *
272
 * If a task is only holding callback_lock, then it has read-only
273 274
 * access to cpusets.
 *
275 276 277
 * 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.
278
 *
279
 * The cpuset_common_file_read() handlers only hold callback_lock across
280 281 282
 * small pieces of code, such as when reading out possibly multi-word
 * cpumasks and nodemasks.
 *
283 284
 * 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
285 286
 */

287
static DEFINE_MUTEX(cpuset_mutex);
288
static DEFINE_SPINLOCK(callback_lock);
289

290 291
static struct workqueue_struct *cpuset_migrate_mm_wq;

292 293 294 295 296 297
/*
 * CPU / memory hotplug is handled asynchronously.
 */
static void cpuset_hotplug_workfn(struct work_struct *work);
static DECLARE_WORK(cpuset_hotplug_work, cpuset_hotplug_workfn);

298 299
static DECLARE_WAIT_QUEUE_HEAD(cpuset_attach_wq);

300 301
/*
 * This is ugly, but preserves the userspace API for existing cpuset
302
 * users. If someone tries to mount the "cpuset" filesystem, we
303 304
 * silently switch it to mount "cgroup" instead
 */
Al Viro's avatar
Al Viro committed
305 306
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
307
{
308
	struct file_system_type *cgroup_fs = get_fs_type("cgroup");
Al Viro's avatar
Al Viro committed
309
	struct dentry *ret = ERR_PTR(-ENODEV);
310 311 312 313
	if (cgroup_fs) {
		char mountopts[] =
			"cpuset,noprefix,"
			"release_agent=/sbin/cpuset_release_agent";
Al Viro's avatar
Al Viro committed
314 315
		ret = cgroup_fs->mount(cgroup_fs, flags,
					   unused_dev_name, mountopts);
316 317 318
		put_filesystem(cgroup_fs);
	}
	return ret;
Linus Torvalds's avatar
Linus Torvalds committed
319 320 321 322
}

static struct file_system_type cpuset_fs_type = {
	.name = "cpuset",
Al Viro's avatar
Al Viro committed
323
	.mount = cpuset_mount,
Linus Torvalds's avatar
Linus Torvalds committed
324 325 326
};

/*
327
 * Return in pmask the portion of a cpusets's cpus_allowed that
Linus Torvalds's avatar
Linus Torvalds committed
328
 * are online.  If none are online, walk up the cpuset hierarchy
329 330
 * 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
331 332
 *
 * One way or another, we guarantee to return some non-empty subset
333
 * of cpu_online_mask.
Linus Torvalds's avatar
Linus Torvalds committed
334
 *
335
 * Call with callback_lock or cpuset_mutex held.
Linus Torvalds's avatar
Linus Torvalds committed
336
 */
337
static void guarantee_online_cpus(struct cpuset *cs, struct cpumask *pmask)
Linus Torvalds's avatar
Linus Torvalds committed
338
{
339
	while (!cpumask_intersects(cs->effective_cpus, cpu_online_mask))
Tejun Heo's avatar
Tejun Heo committed
340
		cs = parent_cs(cs);
341
	cpumask_and(pmask, cs->effective_cpus, cpu_online_mask);
Linus Torvalds's avatar
Linus Torvalds committed
342 343 344 345
}

/*
 * Return in *pmask the portion of a cpusets's mems_allowed that
346 347
 * are online, with memory.  If none are online with memory, walk
 * up the cpuset hierarchy until we find one that does have some
348
 * online mems.  The top cpuset always has some mems online.
Linus Torvalds's avatar
Linus Torvalds committed
349 350
 *
 * One way or another, we guarantee to return some non-empty subset
351
 * of node_states[N_MEMORY].
Linus Torvalds's avatar
Linus Torvalds committed
352
 *
353
 * Call with callback_lock or cpuset_mutex held.
Linus Torvalds's avatar
Linus Torvalds committed
354
 */
355
static void guarantee_online_mems(struct cpuset *cs, nodemask_t *pmask)
Linus Torvalds's avatar
Linus Torvalds committed
356
{
357
	while (!nodes_intersects(cs->effective_mems, node_states[N_MEMORY]))
Tejun Heo's avatar
Tejun Heo committed
358
		cs = parent_cs(cs);
359
	nodes_and(*pmask, cs->effective_mems, node_states[N_MEMORY]);
Linus Torvalds's avatar
Linus Torvalds committed
360 361
}

362 363 364
/*
 * update task's spread flag if cpuset's page/slab spread flag is set
 *
365
 * Call with callback_lock or cpuset_mutex held.
366 367 368 369 370
 */
static void cpuset_update_task_spread_flag(struct cpuset *cs,
					struct task_struct *tsk)
{
	if (is_spread_page(cs))
371
		task_set_spread_page(tsk);
372
	else
373 374
		task_clear_spread_page(tsk);

375
	if (is_spread_slab(cs))
376
		task_set_spread_slab(tsk);
377
	else
378
		task_clear_spread_slab(tsk);
379 380
}

Linus Torvalds's avatar
Linus Torvalds committed
381 382 383 384 385
/*
 * 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
386
 * are only set if the other's are set.  Call holding cpuset_mutex.
Linus Torvalds's avatar
Linus Torvalds committed
387 388 389 390
 */

static int is_cpuset_subset(const struct cpuset *p, const struct cpuset *q)
{
391
	return	cpumask_subset(p->cpus_allowed, q->cpus_allowed) &&
Linus Torvalds's avatar
Linus Torvalds committed
392 393 394 395 396
		nodes_subset(p->mems_allowed, q->mems_allowed) &&
		is_cpu_exclusive(p) <= is_cpu_exclusive(q) &&
		is_mem_exclusive(p) <= is_mem_exclusive(q);
}

397 398 399 400
/**
 * alloc_trial_cpuset - allocate a trial cpuset
 * @cs: the cpuset that the trial cpuset duplicates
 */
401
static struct cpuset *alloc_trial_cpuset(struct cpuset *cs)
402
{
403 404 405 406 407 408
	struct cpuset *trial;

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

409 410 411 412
	if (!alloc_cpumask_var(&trial->cpus_allowed, GFP_KERNEL))
		goto free_cs;
	if (!alloc_cpumask_var(&trial->effective_cpus, GFP_KERNEL))
		goto free_cpus;
413

414 415
	cpumask_copy(trial->cpus_allowed, cs->cpus_allowed);
	cpumask_copy(trial->effective_cpus, cs->effective_cpus);
416
	return trial;
417 418 419 420 421 422

free_cpus:
	free_cpumask_var(trial->cpus_allowed);
free_cs:
	kfree(trial);
	return NULL;
423 424 425 426 427 428 429 430
}

/**
 * free_trial_cpuset - free the trial cpuset
 * @trial: the trial cpuset to be freed
 */
static void free_trial_cpuset(struct cpuset *trial)
{
431
	free_cpumask_var(trial->effective_cpus);
432
	free_cpumask_var(trial->cpus_allowed);
433 434 435
	kfree(trial);
}

Linus Torvalds's avatar
Linus Torvalds committed
436 437 438 439 440 441 442
/*
 * 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
443
 * cpuset_mutex held.
Linus Torvalds's avatar
Linus Torvalds committed
444 445 446 447 448 449 450 451 452 453 454 455
 *
 * '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.
 */

456
static int validate_change(struct cpuset *cur, struct cpuset *trial)
Linus Torvalds's avatar
Linus Torvalds committed
457
{
458
	struct cgroup_subsys_state *css;
Linus Torvalds's avatar
Linus Torvalds committed
459
	struct cpuset *c, *par;
460 461 462
	int ret;

	rcu_read_lock();
Linus Torvalds's avatar
Linus Torvalds committed
463 464

	/* Each of our child cpusets must be a subset of us */
465
	ret = -EBUSY;
466
	cpuset_for_each_child(c, css, cur)
467 468
		if (!is_cpuset_subset(c, trial))
			goto out;
Linus Torvalds's avatar
Linus Torvalds committed
469 470

	/* Remaining checks don't apply to root cpuset */
471
	ret = 0;
472
	if (cur == &top_cpuset)
473
		goto out;
Linus Torvalds's avatar
Linus Torvalds committed
474

Tejun Heo's avatar
Tejun Heo committed
475
	par = parent_cs(cur);
476

477
	/* On legacy hiearchy, we must be a subset of our parent cpuset. */
478
	ret = -EACCES;
479 480
	if (!cgroup_subsys_on_dfl(cpuset_cgrp_subsys) &&
	    !is_cpuset_subset(trial, par))
481
		goto out;
Linus Torvalds's avatar
Linus Torvalds committed
482

483 484 485 486
	/*
	 * If either I or some sibling (!= me) is exclusive, we can't
	 * overlap
	 */
487
	ret = -EINVAL;
488
	cpuset_for_each_child(c, css, par) {
Linus Torvalds's avatar
Linus Torvalds committed
489 490
		if ((is_cpu_exclusive(trial) || is_cpu_exclusive(c)) &&
		    c != cur &&
491
		    cpumask_intersects(trial->cpus_allowed, c->cpus_allowed))
492
			goto out;
Linus Torvalds's avatar
Linus Torvalds committed
493 494 495
		if ((is_mem_exclusive(trial) || is_mem_exclusive(c)) &&
		    c != cur &&
		    nodes_intersects(trial->mems_allowed, c->mems_allowed))
496
			goto out;
Linus Torvalds's avatar
Linus Torvalds committed
497 498
	}

499 500
	/*
	 * Cpusets with tasks - existing or newly being attached - can't
501
	 * be changed to have empty cpus_allowed or mems_allowed.
502
	 */
503
	ret = -ENOSPC;
504
	if ((cgroup_is_populated(cur->css.cgroup) || cur->attach_in_progress)) {
505 506 507 508 509 510 511
		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;
	}
512

513 514 515 516 517 518 519 520 521 522
	/*
	 * 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;

523 524 525 526
	ret = 0;
out:
	rcu_read_unlock();
	return ret;
Linus Torvalds's avatar
Linus Torvalds committed
527 528
}

529
#ifdef CONFIG_SMP
Paul Jackson's avatar
Paul Jackson committed
530
/*
531
 * Helper routine for generate_sched_domains().
532
 * Do cpusets a, b have overlapping effective cpus_allowed masks?
Paul Jackson's avatar
Paul Jackson committed
533 534 535
 */
static int cpusets_overlap(struct cpuset *a, struct cpuset *b)
{
536
	return cpumask_intersects(a->effective_cpus, b->effective_cpus);
Paul Jackson's avatar
Paul Jackson committed
537 538
}

539 540 541 542 543 544 545 546
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;
}

547 548
static void update_domain_attr_tree(struct sched_domain_attr *dattr,
				    struct cpuset *root_cs)
549
{
550
	struct cpuset *cp;
551
	struct cgroup_subsys_state *pos_css;
552

553
	rcu_read_lock();
554
	cpuset_for_each_descendant_pre(cp, pos_css, root_cs) {
555 556
		/* skip the whole subtree if @cp doesn't have any CPU */
		if (cpumask_empty(cp->cpus_allowed)) {
557
			pos_css = css_rightmost_descendant(pos_css);
558
			continue;
559
		}
560 561 562 563

		if (is_sched_load_balance(cp))
			update_domain_attr(dattr, cp);
	}
564
	rcu_read_unlock();
565 566
}

Paul Jackson's avatar
Paul Jackson committed
567
/*
568 569 570 571 572
 * 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.
573
 * The output of this function needs to be passed to kernel/sched/core.c
574 575 576
 * 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
577
 *
578
 * See "What is sched_load_balance" in Documentation/cgroups/cpusets.txt
Paul Jackson's avatar
Paul Jackson committed
579 580 581 582 583 584 585
 * 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.
 *
586
 * Must be called with cpuset_mutex held.
Paul Jackson's avatar
Paul Jackson committed
587 588
 *
 * The three key local variables below are:
589
 *    q  - a linked-list queue of cpuset pointers, used to implement a
Paul Jackson's avatar
Paul Jackson committed
590 591 592 593 594 595 596 597 598 599 600 601
 *	   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
602
 *	   the kernel/sched/core.c routine partition_sched_domains() in a
Paul Jackson's avatar
Paul Jackson committed
603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620
 *	   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().
 */
621
static int generate_sched_domains(cpumask_var_t **domains,
622
			struct sched_domain_attr **attributes)
Paul Jackson's avatar
Paul Jackson committed
623 624 625 626 627
{
	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 */
628
	cpumask_var_t *doms;	/* resulting partition; i.e. sched domains */
629
	cpumask_var_t non_isolated_cpus;  /* load balanced CPUs */
630
	struct sched_domain_attr *dattr;  /* attributes for custom domains */
631
	int ndoms = 0;		/* number of sched domains in result */
632
	int nslot;		/* next empty doms[] struct cpumask slot */
633
	struct cgroup_subsys_state *pos_css;
Paul Jackson's avatar
Paul Jackson committed
634 635

	doms = NULL;
636
	dattr = NULL;
637
	csa = NULL;
Paul Jackson's avatar
Paul Jackson committed
638

639 640 641 642
	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
643 644
	/* Special case for the 99% of systems with one, full, sched domain */
	if (is_sched_load_balance(&top_cpuset)) {
645 646
		ndoms = 1;
		doms = alloc_sched_domains(ndoms);
Paul Jackson's avatar
Paul Jackson committed
647
		if (!doms)
648 649
			goto done;

650 651 652
		dattr = kmalloc(sizeof(struct sched_domain_attr), GFP_KERNEL);
		if (dattr) {
			*dattr = SD_ATTR_INIT;
653
			update_domain_attr_tree(dattr, &top_cpuset);
654
		}
655 656
		cpumask_and(doms[0], top_cpuset.effective_cpus,
				     non_isolated_cpus);
657 658

		goto done;
Paul Jackson's avatar
Paul Jackson committed
659 660
	}

661
	csa = kmalloc(nr_cpusets() * sizeof(cp), GFP_KERNEL);
Paul Jackson's avatar
Paul Jackson committed
662 663 664 665
	if (!csa)
		goto done;
	csn = 0;

666
	rcu_read_lock();
667
	cpuset_for_each_descendant_pre(cp, pos_css, &top_cpuset) {
668 669
		if (cp == &top_cpuset)
			continue;
670
		/*
671 672 673 674 675 676
		 * 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.
677
		 */
678
		if (!cpumask_empty(cp->cpus_allowed) &&
679 680
		    !(is_sched_load_balance(cp) &&
		      cpumask_intersects(cp->cpus_allowed, non_isolated_cpus)))
681
			continue;
682

683 684 685 686
		if (is_sched_load_balance(cp))
			csa[csn++] = cp;

		/* skip @cp's subtree */
687
		pos_css = css_rightmost_descendant(pos_css);
688 689
	}
	rcu_read_unlock();
Paul Jackson's avatar
Paul Jackson committed
690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717

	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;
			}
		}
	}

718 719 720 721
	/*
	 * Now we know how many domains to create.
	 * Convert <csn, csa> to <ndoms, doms> and populate cpu masks.
	 */
722
	doms = alloc_sched_domains(ndoms);
723
	if (!doms)
724 725 726 727 728 729
		goto done;

	/*
	 * The rest of the code, including the scheduler, can deal with
	 * dattr==NULL case. No need to abort if alloc fails.
	 */
730
	dattr = kmalloc(ndoms * sizeof(struct sched_domain_attr), GFP_KERNEL);
Paul Jackson's avatar
Paul Jackson committed
731 732 733

	for (nslot = 0, i = 0; i < csn; i++) {
		struct cpuset *a = csa[i];
734
		struct cpumask *dp;
Paul Jackson's avatar
Paul Jackson committed
735 736
		int apn = a->pn;

737 738 739 740 741
		if (apn < 0) {
			/* Skip completed partitions */
			continue;
		}

742
		dp = doms[nslot];
743 744 745 746

		if (nslot == ndoms) {
			static int warnings = 10;
			if (warnings) {
747 748
				pr_warn("rebuild_sched_domains confused: nslot %d, ndoms %d, csn %d, i %d, apn %d\n",
					nslot, ndoms, csn, i, apn);
749
				warnings--;
Paul Jackson's avatar
Paul Jackson committed
750
			}
751 752
			continue;
		}
Paul Jackson's avatar
Paul Jackson committed
753

754
		cpumask_clear(dp);
755 756 757 758 759 760
		if (dattr)
			*(dattr + nslot) = SD_ATTR_INIT;
		for (j = i; j < csn; j++) {
			struct cpuset *b = csa[j];

			if (apn == b->pn) {
761
				cpumask_or(dp, dp, b->effective_cpus);
762
				cpumask_and(dp, dp, non_isolated_cpus);
763 764 765 766 767
				if (dattr)
					update_domain_attr_tree(dattr + nslot, b);

				/* Done with this partition */
				b->pn = -1;
Paul Jackson's avatar
Paul Jackson committed
768 769
			}
		}
770
		nslot++;
Paul Jackson's avatar
Paul Jackson committed
771 772 773
	}
	BUG_ON(nslot != ndoms);

774
done:
775
	free_cpumask_var(non_isolated_cpus);
776 777
	kfree(csa);

778 779 780 781 782 783 784
	/*
	 * Fallback to the default domain if kmalloc() failed.
	 * See comments in partition_sched_domains().
	 */
	if (doms == NULL)
		ndoms = 1;

785 786 787 788 789 790 791 792
	*domains    = doms;
	*attributes = dattr;
	return ndoms;
}

/*
 * Rebuild scheduler domains.
 *
793 794 795 796 797
 * 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.
798
 *
799
 * Call with cpuset_mutex held.  Takes get_online_cpus().
800
 */
801
static void rebuild_sched_domains_locked(void)
802 803
{
	struct sched_domain_attr *attr;
804
	cpumask_var_t *doms;
805 806
	int ndoms;

807
	lockdep_assert_held(&cpuset_mutex);
808
	get_online_cpus();
809

810 811 812 813 814
	/*
	 * 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.
	 */
815
	if (!cpumask_equal(top_cpuset.effective_cpus, cpu_active_mask))
816 817
		goto out;

818 819 820 821 822
	/* Generate domain masks and attrs */
	ndoms = generate_sched_domains(&doms, &attr);

	/* Have scheduler rebuild the domains */
	partition_sched_domains(ndoms, doms, attr);
823
out:
824
	put_online_cpus();
825
}
826
#else /* !CONFIG_SMP */
827
static void rebuild_sched_domains_locked(void)
828 829 830
{
}
#endif /* CONFIG_SMP */
Paul Jackson's avatar
Paul Jackson committed
831

832 833
void rebuild_sched_domains(void)
{
834
	mutex_lock(&cpuset_mutex);
835
	rebuild_sched_domains_locked();
836
	mutex_unlock(&cpuset_mutex);
Paul Jackson's avatar
Paul Jackson committed
837 838
}

839 840 841 842
/**
 * 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
 *
843 844 845
 * 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.
846
 */
847
static void update_tasks_cpumask(struct cpuset *cs)
848
{
849 850 851 852 853
	struct css_task_iter it;
	struct task_struct *task;

	css_task_iter_start(&cs->css, &it);
	while ((task = css_task_iter_next(&it)))
854
		set_cpus_allowed_ptr(task, cs->effective_cpus);
855
	css_task_iter_end(&it);
856 857
}

858
/*
859 860 861 862 863 864
 * 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.
865
 *
866
 * On legacy hierachy, effective_cpus will be the same with cpu_allowed.
867 868 869
 *
 * Called with cpuset_mutex held
 */
870
static void update_cpumasks_hier(struct cpuset *cs, struct cpumask *new_cpus)
871 872
{
	struct cpuset *cp;
873
	struct cgroup_subsys_state *pos_css;
874
	bool need_rebuild_sched_domains = false;
875 876

	rcu_read_lock();
877 878 879 880 881
	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);

882 883 884 885
		/*
		 * If it becomes empty, inherit the effective mask of the
		 * parent, which is guaranteed to have some CPUs.
		 */
886 887
		if (cgroup_subsys_on_dfl(cpuset_cgrp_subsys) &&
		    cpumask_empty(new_cpus))
888 889
			cpumask_copy(new_cpus, parent->effective_cpus);

890 891 892 893
		/* 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;
894
		}
895

896
		if (!css_tryget_online(&cp->css))
897 898 899
			continue;
		rcu_read_unlock();

900
		spin_lock_irq(&callback_lock);
901
		cpumask_copy(cp->effective_cpus, new_cpus);
902
		spin_unlock_irq(&callback_lock);
903

904
		WARN_ON(!cgroup_subsys_on_dfl(cpuset_cgrp_subsys) &&
905 906
			!cpumask_equal(cp->cpus_allowed, cp->effective_cpus));

907
		update_tasks_cpumask(cp);
908

909 910 911 912 913 914 915 916
		/*
		 * 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;

917 918 919 920
		rcu_read_lock();
		css_put(&cp->css);
	}
	rcu_read_unlock();
921 922 923

	if (need_rebuild_sched_domains)
		rebuild_sched_domains_locked();
924 925
}

926 927 928
/**
 * update_cpumask - update the cpus_allowed mask of a cpuset and all tasks in it
 * @cs: the cpuset to consider
929
 * @trialcs: trial cpuset
930 931
 * @buf: buffer of cpu numbers written to this cpuset
 */
932 933
static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs,
			  const char *buf)
Linus Torvalds's avatar
Linus Torvalds committed
934
{
935
	int retval;
Linus Torvalds's avatar
Linus Torvalds committed
936

937
	/* top_cpuset.cpus_allowed tracks cpu_online_mask; it's read-only */
938 939 940
	if (cs == &top_cpuset)
		return -EACCES;

941
	/*
942
	 * An empty cpus_allowed is ok only if the cpuset has no tasks.
943 944 945
	 * Since cpulist_parse() fails on an empty mask, we special case
	 * that parsing.  The validate_change() call ensures that cpusets
	 * with tasks have cpus.
946
	 */
947
	if (!*buf) {
948
		cpumask_clear(trialcs->cpus_allowed);
949
	} else {
950
		retval = cpulist_parse(buf, trialcs->cpus_allowed);
951 952
		if (retval < 0)
			return retval;
953

954 955
		if (!cpumask_subset(trialcs->cpus_allowed,
				    top_cpuset.cpus_allowed))
956
			return -EINVAL;
957
	}
Paul Jackson's avatar
Paul Jackson committed
958

Paul Menage's avatar
Paul Menage committed
959
	/* Nothing to do if the cpus didn't change */
960
	if (cpumask_equal(cs->cpus_allowed, trialcs->cpus_allowed))
Paul Menage's avatar
Paul Menage committed
961
		return 0;
962

963 964 965 966
	retval = validate_change(cs, trialcs);
	if (retval < 0)
		return retval;

967
	spin_lock_irq(&callback_lock);
968
	cpumask_copy(cs->cpus_allowed, trialcs->cpus_allowed);
969
	spin_unlock_irq(&callback_lock);
Paul Jackson's avatar
Paul Jackson committed
970

971 972
	/* use trialcs->cpus_allowed as a temp variable */
	update_cpumasks_hier(cs, trialcs->cpus_allowed);
973
	return 0;
Linus Torvalds's avatar
Linus Torvalds committed
974 975
}

976
/*
977 978 979 980 981
 * 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.
982 983
 */

984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001
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);
}

1002 1003 1004
static void cpuset_migrate_mm(struct mm_struct *mm, const nodemask_t *from,
							const nodemask_t *to)
{
1005
	struct cpuset_migrate_mm_work *mwork;