cpuset.c 75.4 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

290 291 292 293 294 295
/*
 * CPU / memory hotplug is handled asynchronously.
 */
static void cpuset_hotplug_workfn(struct work_struct *work);
static DECLARE_WORK(cpuset_hotplug_work, cpuset_hotplug_workfn);

296 297
static DECLARE_WAIT_QUEUE_HEAD(cpuset_attach_wq);

298 299
/*
 * This is ugly, but preserves the userspace API for existing cpuset
300
 * users. If someone tries to mount the "cpuset" filesystem, we
301 302
 * silently switch it to mount "cgroup" instead
 */
Al Viro's avatar
Al Viro committed
303 304
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
305
{
306
	struct file_system_type *cgroup_fs = get_fs_type("cgroup");
Al Viro's avatar
Al Viro committed
307
	struct dentry *ret = ERR_PTR(-ENODEV);
308 309 310 311
	if (cgroup_fs) {
		char mountopts[] =
			"cpuset,noprefix,"
			"release_agent=/sbin/cpuset_release_agent";
Al Viro's avatar
Al Viro committed
312 313
		ret = cgroup_fs->mount(cgroup_fs, flags,
					   unused_dev_name, mountopts);
314 315 316
		put_filesystem(cgroup_fs);
	}
	return ret;
Linus Torvalds's avatar
Linus Torvalds committed
317 318 319 320
}

static struct file_system_type cpuset_fs_type = {
	.name = "cpuset",
Al Viro's avatar
Al Viro committed
321
	.mount = cpuset_mount,
Linus Torvalds's avatar
Linus Torvalds committed
322 323 324
};

/*
325
 * Return in pmask the portion of a cpusets's cpus_allowed that
Linus Torvalds's avatar
Linus Torvalds committed
326
 * are online.  If none are online, walk up the cpuset hierarchy
327 328
 * 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
329 330
 *
 * One way or another, we guarantee to return some non-empty subset
331
 * of cpu_online_mask.
Linus Torvalds's avatar
Linus Torvalds committed
332
 *
333
 * Call with callback_lock or cpuset_mutex held.
Linus Torvalds's avatar
Linus Torvalds committed
334
 */
335
static void guarantee_online_cpus(struct cpuset *cs, struct cpumask *pmask)
Linus Torvalds's avatar
Linus Torvalds committed
336
{
337
	while (!cpumask_intersects(cs->effective_cpus, cpu_online_mask))
Tejun Heo's avatar
Tejun Heo committed
338
		cs = parent_cs(cs);
339
	cpumask_and(pmask, cs->effective_cpus, cpu_online_mask);
Linus Torvalds's avatar
Linus Torvalds committed
340 341 342 343
}

/*
 * Return in *pmask the portion of a cpusets's mems_allowed that
344 345
 * are online, with memory.  If none are online with memory, walk
 * up the cpuset hierarchy until we find one that does have some
346
 * online mems.  The top cpuset always has some mems online.
Linus Torvalds's avatar
Linus Torvalds committed
347 348
 *
 * One way or another, we guarantee to return some non-empty subset
349
 * of node_states[N_MEMORY].
Linus Torvalds's avatar
Linus Torvalds committed
350
 *
351
 * Call with callback_lock or cpuset_mutex held.
Linus Torvalds's avatar
Linus Torvalds committed
352
 */
353
static void guarantee_online_mems(struct cpuset *cs, nodemask_t *pmask)
Linus Torvalds's avatar
Linus Torvalds committed
354
{
355
	while (!nodes_intersects(cs->effective_mems, node_states[N_MEMORY]))
Tejun Heo's avatar
Tejun Heo committed
356
		cs = parent_cs(cs);
357
	nodes_and(*pmask, cs->effective_mems, node_states[N_MEMORY]);
Linus Torvalds's avatar
Linus Torvalds committed
358 359
}

360 361 362
/*
 * update task's spread flag if cpuset's page/slab spread flag is set
 *
363
 * Call with callback_lock or cpuset_mutex held.
364 365 366 367 368
 */
static void cpuset_update_task_spread_flag(struct cpuset *cs,
					struct task_struct *tsk)
{
	if (is_spread_page(cs))
369
		task_set_spread_page(tsk);
370
	else
371 372
		task_clear_spread_page(tsk);

373
	if (is_spread_slab(cs))
374
		task_set_spread_slab(tsk);
375
	else
376
		task_clear_spread_slab(tsk);
377 378
}

Linus Torvalds's avatar
Linus Torvalds committed
379 380 381 382 383
/*
 * 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
384
 * are only set if the other's are set.  Call holding cpuset_mutex.
Linus Torvalds's avatar
Linus Torvalds committed
385 386 387 388
 */

static int is_cpuset_subset(const struct cpuset *p, const struct cpuset *q)
{
389
	return	cpumask_subset(p->cpus_allowed, q->cpus_allowed) &&
Linus Torvalds's avatar
Linus Torvalds committed
390 391 392 393 394
		nodes_subset(p->mems_allowed, q->mems_allowed) &&
		is_cpu_exclusive(p) <= is_cpu_exclusive(q) &&
		is_mem_exclusive(p) <= is_mem_exclusive(q);
}

395 396 397 398
/**
 * alloc_trial_cpuset - allocate a trial cpuset
 * @cs: the cpuset that the trial cpuset duplicates
 */
399
static struct cpuset *alloc_trial_cpuset(struct cpuset *cs)
400
{
401 402 403 404 405 406
	struct cpuset *trial;

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

407 408 409 410
	if (!alloc_cpumask_var(&trial->cpus_allowed, GFP_KERNEL))
		goto free_cs;
	if (!alloc_cpumask_var(&trial->effective_cpus, GFP_KERNEL))
		goto free_cpus;
411

412 413
	cpumask_copy(trial->cpus_allowed, cs->cpus_allowed);
	cpumask_copy(trial->effective_cpus, cs->effective_cpus);
414
	return trial;
415 416 417 418 419 420

free_cpus:
	free_cpumask_var(trial->cpus_allowed);
free_cs:
	kfree(trial);
	return NULL;
421 422 423 424 425 426 427 428
}

/**
 * free_trial_cpuset - free the trial cpuset
 * @trial: the trial cpuset to be freed
 */
static void free_trial_cpuset(struct cpuset *trial)
{
429
	free_cpumask_var(trial->effective_cpus);
430
	free_cpumask_var(trial->cpus_allowed);
431 432 433
	kfree(trial);
}

Linus Torvalds's avatar
Linus Torvalds committed
434 435 436 437 438 439 440
/*
 * 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
441
 * cpuset_mutex held.
Linus Torvalds's avatar
Linus Torvalds committed
442 443 444 445 446 447 448 449 450 451 452 453
 *
 * '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.
 */

454
static int validate_change(struct cpuset *cur, struct cpuset *trial)
Linus Torvalds's avatar
Linus Torvalds committed
455
{
456
	struct cgroup_subsys_state *css;
Linus Torvalds's avatar
Linus Torvalds committed
457
	struct cpuset *c, *par;
458 459 460
	int ret;

	rcu_read_lock();
Linus Torvalds's avatar
Linus Torvalds committed
461 462

	/* Each of our child cpusets must be a subset of us */
463
	ret = -EBUSY;
464
	cpuset_for_each_child(c, css, cur)
465 466
		if (!is_cpuset_subset(c, trial))
			goto out;
Linus Torvalds's avatar
Linus Torvalds committed
467 468

	/* Remaining checks don't apply to root cpuset */
469
	ret = 0;
470
	if (cur == &top_cpuset)
471
		goto out;
Linus Torvalds's avatar
Linus Torvalds committed
472

Tejun Heo's avatar
Tejun Heo committed
473
	par = parent_cs(cur);
474

475
	/* On legacy hiearchy, we must be a subset of our parent cpuset. */
476
	ret = -EACCES;
477 478
	if (!cgroup_subsys_on_dfl(cpuset_cgrp_subsys) &&
	    !is_cpuset_subset(trial, par))
479
		goto out;
Linus Torvalds's avatar
Linus Torvalds committed
480

481 482 483 484
	/*
	 * If either I or some sibling (!= me) is exclusive, we can't
	 * overlap
	 */
485
	ret = -EINVAL;
486
	cpuset_for_each_child(c, css, par) {
Linus Torvalds's avatar
Linus Torvalds committed
487 488
		if ((is_cpu_exclusive(trial) || is_cpu_exclusive(c)) &&
		    c != cur &&
489
		    cpumask_intersects(trial->cpus_allowed, c->cpus_allowed))
490
			goto out;
Linus Torvalds's avatar
Linus Torvalds committed
491 492 493
		if ((is_mem_exclusive(trial) || is_mem_exclusive(c)) &&
		    c != cur &&
		    nodes_intersects(trial->mems_allowed, c->mems_allowed))
494
			goto out;
Linus Torvalds's avatar
Linus Torvalds committed
495 496
	}

497 498
	/*
	 * Cpusets with tasks - existing or newly being attached - can't
499
	 * be changed to have empty cpus_allowed or mems_allowed.
500
	 */
501
	ret = -ENOSPC;
502
	if ((cgroup_is_populated(cur->css.cgroup) || cur->attach_in_progress)) {
503 504 505 506 507 508 509
		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;
	}
510

511 512 513 514 515 516 517 518 519 520
	/*
	 * 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;

521 522 523 524
	ret = 0;
out:
	rcu_read_unlock();
	return ret;
Linus Torvalds's avatar
Linus Torvalds committed
525 526
}

527
#ifdef CONFIG_SMP
Paul Jackson's avatar
Paul Jackson committed
528
/*
529
 * Helper routine for generate_sched_domains().
530
 * Do cpusets a, b have overlapping effective cpus_allowed masks?
Paul Jackson's avatar
Paul Jackson committed
531 532 533
 */
static int cpusets_overlap(struct cpuset *a, struct cpuset *b)
{
534
	return cpumask_intersects(a->effective_cpus, b->effective_cpus);
Paul Jackson's avatar
Paul Jackson committed
535 536
}

537 538 539 540 541 542 543 544
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;
}

545 546
static void update_domain_attr_tree(struct sched_domain_attr *dattr,
				    struct cpuset *root_cs)
547
{
548
	struct cpuset *cp;
549
	struct cgroup_subsys_state *pos_css;
550

551
	rcu_read_lock();
552
	cpuset_for_each_descendant_pre(cp, pos_css, root_cs) {
553 554
		/* skip the whole subtree if @cp doesn't have any CPU */
		if (cpumask_empty(cp->cpus_allowed)) {
555
			pos_css = css_rightmost_descendant(pos_css);
556
			continue;
557
		}
558 559 560 561

		if (is_sched_load_balance(cp))
			update_domain_attr(dattr, cp);
	}
562
	rcu_read_unlock();
563 564
}

Paul Jackson's avatar
Paul Jackson committed
565
/*
566 567 568 569 570
 * 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.
571
 * The output of this function needs to be passed to kernel/sched/core.c
572 573 574
 * 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
575
 *
576
 * See "What is sched_load_balance" in Documentation/cgroups/cpusets.txt
Paul Jackson's avatar
Paul Jackson committed
577 578 579 580 581 582 583
 * 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.
 *
584
 * Must be called with cpuset_mutex held.
Paul Jackson's avatar
Paul Jackson committed
585 586
 *
 * The three key local variables below are:
587
 *    q  - a linked-list queue of cpuset pointers, used to implement a
Paul Jackson's avatar
Paul Jackson committed
588 589 590 591 592 593 594 595 596 597 598 599
 *	   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
600
 *	   the kernel/sched/core.c routine partition_sched_domains() in a
Paul Jackson's avatar
Paul Jackson committed
601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618
 *	   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().
 */
619
static int generate_sched_domains(cpumask_var_t **domains,
620
			struct sched_domain_attr **attributes)
Paul Jackson's avatar
Paul Jackson committed
621 622 623 624 625
{
	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 */
626
	cpumask_var_t *doms;	/* resulting partition; i.e. sched domains */
627
	cpumask_var_t non_isolated_cpus;  /* load balanced CPUs */
628
	struct sched_domain_attr *dattr;  /* attributes for custom domains */
629
	int ndoms = 0;		/* number of sched domains in result */
630
	int nslot;		/* next empty doms[] struct cpumask slot */
631
	struct cgroup_subsys_state *pos_css;
Paul Jackson's avatar
Paul Jackson committed
632 633

	doms = NULL;
634
	dattr = NULL;
635
	csa = NULL;
Paul Jackson's avatar
Paul Jackson committed
636

637 638 639 640
	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
641 642
	/* Special case for the 99% of systems with one, full, sched domain */
	if (is_sched_load_balance(&top_cpuset)) {
643 644
		ndoms = 1;
		doms = alloc_sched_domains(ndoms);
Paul Jackson's avatar
Paul Jackson committed
645
		if (!doms)
646 647
			goto done;

648 649 650
		dattr = kmalloc(sizeof(struct sched_domain_attr), GFP_KERNEL);
		if (dattr) {
			*dattr = SD_ATTR_INIT;
651
			update_domain_attr_tree(dattr, &top_cpuset);
652
		}
653 654
		cpumask_and(doms[0], top_cpuset.effective_cpus,
				     non_isolated_cpus);
655 656

		goto done;
Paul Jackson's avatar
Paul Jackson committed
657 658
	}

659
	csa = kmalloc(nr_cpusets() * sizeof(cp), GFP_KERNEL);
Paul Jackson's avatar
Paul Jackson committed
660 661 662 663
	if (!csa)
		goto done;
	csn = 0;

664
	rcu_read_lock();
665
	cpuset_for_each_descendant_pre(cp, pos_css, &top_cpuset) {
666 667
		if (cp == &top_cpuset)
			continue;
668
		/*
669 670 671 672 673 674
		 * 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.
675
		 */
676
		if (!cpumask_empty(cp->cpus_allowed) &&
677 678
		    !(is_sched_load_balance(cp) &&
		      cpumask_intersects(cp->cpus_allowed, non_isolated_cpus)))
679
			continue;
680

681 682 683 684
		if (is_sched_load_balance(cp))
			csa[csn++] = cp;

		/* skip @cp's subtree */
685
		pos_css = css_rightmost_descendant(pos_css);
686 687
	}
	rcu_read_unlock();
Paul Jackson's avatar
Paul Jackson committed
688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715

	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;
			}
		}
	}

716 717 718 719
	/*
	 * Now we know how many domains to create.
	 * Convert <csn, csa> to <ndoms, doms> and populate cpu masks.
	 */
720
	doms = alloc_sched_domains(ndoms);
721
	if (!doms)
722 723 724 725 726 727
		goto done;

	/*
	 * The rest of the code, including the scheduler, can deal with
	 * dattr==NULL case. No need to abort if alloc fails.
	 */
728
	dattr = kmalloc(ndoms * sizeof(struct sched_domain_attr), GFP_KERNEL);
Paul Jackson's avatar
Paul Jackson committed
729 730 731

	for (nslot = 0, i = 0; i < csn; i++) {
		struct cpuset *a = csa[i];
732
		struct cpumask *dp;
Paul Jackson's avatar
Paul Jackson committed
733 734
		int apn = a->pn;

735 736 737 738 739
		if (apn < 0) {
			/* Skip completed partitions */
			continue;
		}

740
		dp = doms[nslot];
741 742 743 744

		if (nslot == ndoms) {
			static int warnings = 10;
			if (warnings) {
745 746
				pr_warn("rebuild_sched_domains confused: nslot %d, ndoms %d, csn %d, i %d, apn %d\n",
					nslot, ndoms, csn, i, apn);
747
				warnings--;
Paul Jackson's avatar
Paul Jackson committed
748
			}
749 750
			continue;
		}
Paul Jackson's avatar
Paul Jackson committed
751

752
		cpumask_clear(dp);
753 754 755 756 757 758
		if (dattr)
			*(dattr + nslot) = SD_ATTR_INIT;
		for (j = i; j < csn; j++) {
			struct cpuset *b = csa[j];

			if (apn == b->pn) {
759
				cpumask_or(dp, dp, b->effective_cpus);
760
				cpumask_and(dp, dp, non_isolated_cpus);
761 762 763 764 765
				if (dattr)
					update_domain_attr_tree(dattr + nslot, b);

				/* Done with this partition */
				b->pn = -1;
Paul Jackson's avatar
Paul Jackson committed
766 767
			}
		}
768
		nslot++;
Paul Jackson's avatar
Paul Jackson committed
769 770 771
	}
	BUG_ON(nslot != ndoms);

772
done:
773
	free_cpumask_var(non_isolated_cpus);
774 775
	kfree(csa);

776 777 778 779 780 781 782
	/*
	 * Fallback to the default domain if kmalloc() failed.
	 * See comments in partition_sched_domains().
	 */
	if (doms == NULL)
		ndoms = 1;

783 784 785 786 787 788 789 790
	*domains    = doms;
	*attributes = dattr;
	return ndoms;
}

/*
 * Rebuild scheduler domains.
 *
791 792 793 794 795
 * 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.
796
 *
797
 * Call with cpuset_mutex held.  Takes get_online_cpus().
798
 */
799
static void rebuild_sched_domains_locked(void)
800 801
{
	struct sched_domain_attr *attr;
802
	cpumask_var_t *doms;
803 804
	int ndoms;

805
	lockdep_assert_held(&cpuset_mutex);
806
	get_online_cpus();
807

808 809 810 811 812
	/*
	 * 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.
	 */
813
	if (!cpumask_equal(top_cpuset.effective_cpus, cpu_active_mask))
814 815
		goto out;

816 817 818 819 820
	/* Generate domain masks and attrs */
	ndoms = generate_sched_domains(&doms, &attr);

	/* Have scheduler rebuild the domains */
	partition_sched_domains(ndoms, doms, attr);
821
out:
822
	put_online_cpus();
823
}
824
#else /* !CONFIG_SMP */
825
static void rebuild_sched_domains_locked(void)
826 827 828
{
}
#endif /* CONFIG_SMP */
Paul Jackson's avatar
Paul Jackson committed
829

830 831
void rebuild_sched_domains(void)
{
832
	mutex_lock(&cpuset_mutex);
833
	rebuild_sched_domains_locked();
834
	mutex_unlock(&cpuset_mutex);
Paul Jackson's avatar
Paul Jackson committed
835 836
}

837 838 839 840
/**
 * 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
 *
841 842 843
 * 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.
844
 */
845
static void update_tasks_cpumask(struct cpuset *cs)
846
{
847 848 849 850 851
	struct css_task_iter it;
	struct task_struct *task;

	css_task_iter_start(&cs->css, &it);
	while ((task = css_task_iter_next(&it)))
852
		set_cpus_allowed_ptr(task, cs->effective_cpus);
853
	css_task_iter_end(&it);
854 855
}

856
/*
857 858 859 860 861 862
 * 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.
863
 *
864
 * On legacy hierachy, effective_cpus will be the same with cpu_allowed.
865 866 867
 *
 * Called with cpuset_mutex held
 */
868
static void update_cpumasks_hier(struct cpuset *cs, struct cpumask *new_cpus)
869 870
{
	struct cpuset *cp;
871
	struct cgroup_subsys_state *pos_css;
872
	bool need_rebuild_sched_domains = false;
873 874

	rcu_read_lock();
875 876 877 878 879
	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);

880 881 882 883
		/*
		 * If it becomes empty, inherit the effective mask of the
		 * parent, which is guaranteed to have some CPUs.
		 */
884 885
		if (cgroup_subsys_on_dfl(cpuset_cgrp_subsys) &&
		    cpumask_empty(new_cpus))
886 887
			cpumask_copy(new_cpus, parent->effective_cpus);

888 889 890 891
		/* 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;
892
		}
893

894
		if (!css_tryget_online(&cp->css))
895 896 897
			continue;
		rcu_read_unlock();

898
		spin_lock_irq(&callback_lock);
899
		cpumask_copy(cp->effective_cpus, new_cpus);
900
		spin_unlock_irq(&callback_lock);
901

902
		WARN_ON(!cgroup_subsys_on_dfl(cpuset_cgrp_subsys) &&
903 904
			!cpumask_equal(cp->cpus_allowed, cp->effective_cpus));

905
		update_tasks_cpumask(cp);
906

907 908 909 910 911 912 913 914
		/*
		 * 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;

915 916 917 918
		rcu_read_lock();
		css_put(&cp->css);
	}
	rcu_read_unlock();
919 920 921

	if (need_rebuild_sched_domains)
		rebuild_sched_domains_locked();
922 923
}

924 925 926
/**
 * update_cpumask - update the cpus_allowed mask of a cpuset and all tasks in it
 * @cs: the cpuset to consider
927
 * @trialcs: trial cpuset
928 929
 * @buf: buffer of cpu numbers written to this cpuset
 */
930 931
static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs,
			  const char *buf)
Linus Torvalds's avatar
Linus Torvalds committed
932
{
933
	int retval;
Linus Torvalds's avatar
Linus Torvalds committed
934

935
	/* top_cpuset.cpus_allowed tracks cpu_online_mask; it's read-only */
936 937 938
	if (cs == &top_cpuset)
		return -EACCES;

939
	/*
940
	 * An empty cpus_allowed is ok only if the cpuset has no tasks.
941 942 943
	 * Since cpulist_parse() fails on an empty mask, we special case
	 * that parsing.  The validate_change() call ensures that cpusets
	 * with tasks have cpus.
944
	 */
945
	if (!*buf) {
946
		cpumask_clear(trialcs->cpus_allowed);
947
	} else {
948
		retval = cpulist_parse(buf, trialcs->cpus_allowed);
949 950
		if (retval < 0)
			return retval;
951

952 953
		if (!cpumask_subset(trialcs->cpus_allowed,
				    top_cpuset.cpus_allowed))
954
			return -EINVAL;
955
	}
Paul Jackson's avatar
Paul Jackson committed
956

Paul Menage's avatar
Paul Menage committed
957
	/* Nothing to do if the cpus didn't change */
958
	if (cpumask_equal(cs->cpus_allowed, trialcs->cpus_allowed))
Paul Menage's avatar
Paul Menage committed
959
		return 0;
960

961 962 963 964
	retval = validate_change(cs, trialcs);
	if (retval < 0)
		return retval;

965
	spin_lock_irq(&callback_lock);
966
	cpumask_copy(cs->cpus_allowed, trialcs->cpus_allowed);
967
	spin_unlock_irq(&callback_lock);
Paul Jackson's avatar
Paul Jackson committed
968

969 970
	/* use trialcs->cpus_allowed as a temp variable */
	update_cpumasks_hier(cs, trialcs->cpus_allowed);
971
	return 0;
Linus Torvalds's avatar
Linus Torvalds committed
972 973
}

974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996
/*
 * cpuset_migrate_mm
 *
 *    Migrate memory region from one set of nodes to another.
 *
 *    Temporarilly set tasks mems_allowed to target nodes of migration,
 *    so that the migration code can allocate pages on these nodes.
 *
 *    While the mm_struct we are migrating is typically from some
 *    other task, the task_struct mems_allowed that we are hacking
 *    is for our current task, which must allocate new pages for that
 *    migrating memory region.
 */

static void cpuset_migrate_mm(struct mm_struct *mm, const nodemask_t *from,
							const nodemask_t *to)
{
	struct task_struct *tsk = current;

	tsk->mems_allowed = *to;

	do_migrate_pages(mm, from, to, MPOL_MF_MOVE_ALL);

997
	rcu_read_lock();
998
	guarantee_online_mems(task_cs(tsk), &tsk->mems_allowed);
999
	rcu_read_unlock();
1000 1001
}

1002
/*
1003 1004 1005 1006 1007 1008 1009 1010 1011 1012