cpuset.c 75.6 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 54 55 56 57
#include <linux/slab.h>
#include <linux/spinlock.h>
#include <linux/stat.h>
#include <linux/string.h>
#include <linux/time.h>
#include <linux/backing-dev.h>
#include <linux/sort.h>

#include <asm/uaccess.h>
Arun Sharma's avatar
Arun Sharma committed
58
#include <linux/atomic.h>
59
#include <linux/mutex.h>
60 61
#include <linux/workqueue.h>
#include <linux/cgroup.h>
62
#include <linux/wait.h>
Linus Torvalds's avatar
Linus Torvalds committed
63

64
struct static_key cpusets_enabled_key __read_mostly = STATIC_KEY_INIT_FALSE;
65

66 67 68 69 70 71 72 73 74
/* See "Frequency meter" comments, below. */

struct fmeter {
	int cnt;		/* unprocessed events count */
	int val;		/* most recent output value */
	time_t time;		/* clock (secs) when val computed */
	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

	/* 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
87

88 89 90 91 92 93 94 95 96 97 98 99
	/*
	 * 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;

100
	struct fmeter fmeter;		/* memory_pressure filter */
Paul Jackson's avatar
Paul Jackson committed
101

102 103 104 105 106 107
	/*
	 * 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
108 109
	/* partition number for rebuild_sched_domains() */
	int pn;
110

111 112
	/* for custom sched domain */
	int relax_domain_level;
Linus Torvalds's avatar
Linus Torvalds committed
113 114
};

115
static inline struct cpuset *css_cs(struct cgroup_subsys_state *css)
116
{
117
	return css ? container_of(css, struct cpuset, css) : NULL;
118 119 120 121 122
}

/* Retrieve the cpuset for a task */
static inline struct cpuset *task_cs(struct task_struct *task)
{
123
	return css_cs(task_css(task, cpuset_cgrp_id));
124 125
}

126
static inline struct cpuset *parent_cs(struct cpuset *cs)
Tejun Heo's avatar
Tejun Heo committed
127
{
Tejun Heo's avatar
Tejun Heo committed
128
	return css_cs(cs->css.parent);
Tejun Heo's avatar
Tejun Heo committed
129 130
}

131 132 133 134 135 136 137 138 139 140 141 142 143
#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
144 145
/* bits in struct cpuset flags field */
typedef enum {
Tejun Heo's avatar
Tejun Heo committed
146
	CS_ONLINE,
Linus Torvalds's avatar
Linus Torvalds committed
147 148
	CS_CPU_EXCLUSIVE,
	CS_MEM_EXCLUSIVE,
149
	CS_MEM_HARDWALL,
150
	CS_MEMORY_MIGRATE,
Paul Jackson's avatar
Paul Jackson committed
151
	CS_SCHED_LOAD_BALANCE,
152 153
	CS_SPREAD_PAGE,
	CS_SPREAD_SLAB,
Linus Torvalds's avatar
Linus Torvalds committed
154 155 156
} cpuset_flagbits_t;

/* convenient tests for these bits */
Tejun Heo's avatar
Tejun Heo committed
157 158 159 160 161
static inline bool is_cpuset_online(const struct cpuset *cs)
{
	return test_bit(CS_ONLINE, &cs->flags);
}

Linus Torvalds's avatar
Linus Torvalds committed
162 163
static inline int is_cpu_exclusive(const struct cpuset *cs)
{
164
	return test_bit(CS_CPU_EXCLUSIVE, &cs->flags);
Linus Torvalds's avatar
Linus Torvalds committed
165 166 167 168
}

static inline int is_mem_exclusive(const struct cpuset *cs)
{
169
	return test_bit(CS_MEM_EXCLUSIVE, &cs->flags);
Linus Torvalds's avatar
Linus Torvalds committed
170 171
}

172 173 174 175 176
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
177 178 179 180 181
static inline int is_sched_load_balance(const struct cpuset *cs)
{
	return test_bit(CS_SCHED_LOAD_BALANCE, &cs->flags);
}

182 183
static inline int is_memory_migrate(const struct cpuset *cs)
{
184
	return test_bit(CS_MEMORY_MIGRATE, &cs->flags);
185 186
}

187 188 189 190 191 192 193 194 195 196
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
197
static struct cpuset top_cpuset = {
Tejun Heo's avatar
Tejun Heo committed
198 199
	.flags = ((1 << CS_ONLINE) | (1 << CS_CPU_EXCLUSIVE) |
		  (1 << CS_MEM_EXCLUSIVE)),
Linus Torvalds's avatar
Linus Torvalds committed
200 201
};

202 203 204
/**
 * cpuset_for_each_child - traverse online children of a cpuset
 * @child_cs: loop cursor pointing to the current child
205
 * @pos_css: used for iteration
206 207 208 209 210
 * @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.
 */
211 212 213
#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)))))
214

215 216 217
/**
 * cpuset_for_each_descendant_pre - pre-order walk of a cpuset's descendants
 * @des_cs: loop cursor pointing to the current descendant
218
 * @pos_css: used for iteration
219 220 221
 * @root_cs: target cpuset to walk ancestor of
 *
 * Walk @des_cs through the online descendants of @root_cs.  Must be used
222
 * with RCU read locked.  The caller may modify @pos_css by calling
223 224
 * css_rightmost_descendant() to skip subtree.  @root_cs is included in the
 * iteration and the first node to be visited.
225
 */
226 227 228
#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)))))
229

Linus Torvalds's avatar
Linus Torvalds committed
230
/*
231 232 233 234 235 236 237 238 239 240 241 242 243 244
 * There are two global mutexes guarding cpuset structures - cpuset_mutex
 * and callback_mutex.  The latter may nest inside the former.  We also
 * require taking task_lock() when dereferencing a task's cpuset pointer.
 * See "The task_lock() exception", at the end of this comment.
 *
 * A task must hold both mutexes to modify cpusets.  If a task holds
 * cpuset_mutex, then it blocks others wanting that mutex, ensuring that it
 * is the only task able to also acquire callback_mutex and be able to
 * 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
 * callback routines can briefly acquire callback_mutex to query cpusets.
 * Once it is ready to make the changes, it takes callback_mutex, blocking
 * everyone else.
245 246
 *
 * Calls to the kernel memory allocator can not be made while holding
247
 * callback_mutex, as that would risk double tripping on callback_mutex
248 249 250
 * from one of the callbacks into the cpuset code from within
 * __alloc_pages().
 *
251
 * If a task is only holding callback_mutex, then it has read-only
252 253
 * access to cpusets.
 *
254 255 256
 * 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.
257
 *
258
 * The cpuset_common_file_read() handlers only hold callback_mutex across
259 260 261
 * small pieces of code, such as when reading out possibly multi-word
 * cpumasks and nodemasks.
 *
262 263
 * 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
264 265
 */

266
static DEFINE_MUTEX(cpuset_mutex);
267
static DEFINE_MUTEX(callback_mutex);
268

269 270 271 272 273 274
/*
 * CPU / memory hotplug is handled asynchronously.
 */
static void cpuset_hotplug_workfn(struct work_struct *work);
static DECLARE_WORK(cpuset_hotplug_work, cpuset_hotplug_workfn);

275 276
static DECLARE_WAIT_QUEUE_HEAD(cpuset_attach_wq);

277 278
/*
 * This is ugly, but preserves the userspace API for existing cpuset
279
 * users. If someone tries to mount the "cpuset" filesystem, we
280 281
 * silently switch it to mount "cgroup" instead
 */
Al Viro's avatar
Al Viro committed
282 283
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
284
{
285
	struct file_system_type *cgroup_fs = get_fs_type("cgroup");
Al Viro's avatar
Al Viro committed
286
	struct dentry *ret = ERR_PTR(-ENODEV);
287 288 289 290
	if (cgroup_fs) {
		char mountopts[] =
			"cpuset,noprefix,"
			"release_agent=/sbin/cpuset_release_agent";
Al Viro's avatar
Al Viro committed
291 292
		ret = cgroup_fs->mount(cgroup_fs, flags,
					   unused_dev_name, mountopts);
293 294 295
		put_filesystem(cgroup_fs);
	}
	return ret;
Linus Torvalds's avatar
Linus Torvalds committed
296 297 298 299
}

static struct file_system_type cpuset_fs_type = {
	.name = "cpuset",
Al Viro's avatar
Al Viro committed
300
	.mount = cpuset_mount,
Linus Torvalds's avatar
Linus Torvalds committed
301 302 303
};

/*
304
 * Return in pmask the portion of a cpusets's cpus_allowed that
Linus Torvalds's avatar
Linus Torvalds committed
305
 * are online.  If none are online, walk up the cpuset hierarchy
306 307
 * 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
308 309
 *
 * One way or another, we guarantee to return some non-empty subset
310
 * of cpu_online_mask.
Linus Torvalds's avatar
Linus Torvalds committed
311
 *
312
 * Call with callback_mutex held.
Linus Torvalds's avatar
Linus Torvalds committed
313
 */
314
static void guarantee_online_cpus(struct cpuset *cs, struct cpumask *pmask)
Linus Torvalds's avatar
Linus Torvalds committed
315
{
316
	while (!cpumask_intersects(cs->cpus_allowed, cpu_online_mask))
Tejun Heo's avatar
Tejun Heo committed
317
		cs = parent_cs(cs);
318
	cpumask_and(pmask, cs->cpus_allowed, cpu_online_mask);
Linus Torvalds's avatar
Linus Torvalds committed
319 320 321 322
}

/*
 * Return in *pmask the portion of a cpusets's mems_allowed that
323 324
 * are online, with memory.  If none are online with memory, walk
 * up the cpuset hierarchy until we find one that does have some
325
 * online mems.  The top cpuset always has some mems online.
Linus Torvalds's avatar
Linus Torvalds committed
326 327
 *
 * One way or another, we guarantee to return some non-empty subset
328
 * of node_states[N_MEMORY].
Linus Torvalds's avatar
Linus Torvalds committed
329
 *
330
 * Call with callback_mutex held.
Linus Torvalds's avatar
Linus Torvalds committed
331
 */
332
static void guarantee_online_mems(struct cpuset *cs, nodemask_t *pmask)
Linus Torvalds's avatar
Linus Torvalds committed
333
{
334
	while (!nodes_intersects(cs->mems_allowed, node_states[N_MEMORY]))
Tejun Heo's avatar
Tejun Heo committed
335
		cs = parent_cs(cs);
336
	nodes_and(*pmask, cs->mems_allowed, node_states[N_MEMORY]);
Linus Torvalds's avatar
Linus Torvalds committed
337 338
}

339 340 341
/*
 * update task's spread flag if cpuset's page/slab spread flag is set
 *
342
 * Called with callback_mutex/cpuset_mutex held
343 344 345 346 347 348 349 350 351 352 353 354 355 356
 */
static void cpuset_update_task_spread_flag(struct cpuset *cs,
					struct task_struct *tsk)
{
	if (is_spread_page(cs))
		tsk->flags |= PF_SPREAD_PAGE;
	else
		tsk->flags &= ~PF_SPREAD_PAGE;
	if (is_spread_slab(cs))
		tsk->flags |= PF_SPREAD_SLAB;
	else
		tsk->flags &= ~PF_SPREAD_SLAB;
}

Linus Torvalds's avatar
Linus Torvalds committed
357 358 359 360 361
/*
 * 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
362
 * are only set if the other's are set.  Call holding cpuset_mutex.
Linus Torvalds's avatar
Linus Torvalds committed
363 364 365 366
 */

static int is_cpuset_subset(const struct cpuset *p, const struct cpuset *q)
{
367
	return	cpumask_subset(p->cpus_allowed, q->cpus_allowed) &&
Linus Torvalds's avatar
Linus Torvalds committed
368 369 370 371 372
		nodes_subset(p->mems_allowed, q->mems_allowed) &&
		is_cpu_exclusive(p) <= is_cpu_exclusive(q) &&
		is_mem_exclusive(p) <= is_mem_exclusive(q);
}

373 374 375 376
/**
 * alloc_trial_cpuset - allocate a trial cpuset
 * @cs: the cpuset that the trial cpuset duplicates
 */
377
static struct cpuset *alloc_trial_cpuset(struct cpuset *cs)
378
{
379 380 381 382 383 384
	struct cpuset *trial;

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

385 386 387 388
	if (!alloc_cpumask_var(&trial->cpus_allowed, GFP_KERNEL))
		goto free_cs;
	if (!alloc_cpumask_var(&trial->effective_cpus, GFP_KERNEL))
		goto free_cpus;
389

390 391
	cpumask_copy(trial->cpus_allowed, cs->cpus_allowed);
	cpumask_copy(trial->effective_cpus, cs->effective_cpus);
392
	return trial;
393 394 395 396 397 398

free_cpus:
	free_cpumask_var(trial->cpus_allowed);
free_cs:
	kfree(trial);
	return NULL;
399 400 401 402 403 404 405 406
}

/**
 * free_trial_cpuset - free the trial cpuset
 * @trial: the trial cpuset to be freed
 */
static void free_trial_cpuset(struct cpuset *trial)
{
407
	free_cpumask_var(trial->effective_cpus);
408
	free_cpumask_var(trial->cpus_allowed);
409 410 411
	kfree(trial);
}

Linus Torvalds's avatar
Linus Torvalds committed
412 413 414 415 416 417 418
/*
 * 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
419
 * cpuset_mutex held.
Linus Torvalds's avatar
Linus Torvalds committed
420 421 422 423 424 425 426 427 428 429 430 431
 *
 * '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.
 */

432
static int validate_change(struct cpuset *cur, struct cpuset *trial)
Linus Torvalds's avatar
Linus Torvalds committed
433
{
434
	struct cgroup_subsys_state *css;
Linus Torvalds's avatar
Linus Torvalds committed
435
	struct cpuset *c, *par;
436 437 438
	int ret;

	rcu_read_lock();
Linus Torvalds's avatar
Linus Torvalds committed
439 440

	/* Each of our child cpusets must be a subset of us */
441
	ret = -EBUSY;
442
	cpuset_for_each_child(c, css, cur)
443 444
		if (!is_cpuset_subset(c, trial))
			goto out;
Linus Torvalds's avatar
Linus Torvalds committed
445 446

	/* Remaining checks don't apply to root cpuset */
447
	ret = 0;
448
	if (cur == &top_cpuset)
449
		goto out;
Linus Torvalds's avatar
Linus Torvalds committed
450

Tejun Heo's avatar
Tejun Heo committed
451
	par = parent_cs(cur);
452

Linus Torvalds's avatar
Linus Torvalds committed
453
	/* We must be a subset of our parent cpuset */
454
	ret = -EACCES;
Linus Torvalds's avatar
Linus Torvalds committed
455
	if (!is_cpuset_subset(trial, par))
456
		goto out;
Linus Torvalds's avatar
Linus Torvalds committed
457

458 459 460 461
	/*
	 * If either I or some sibling (!= me) is exclusive, we can't
	 * overlap
	 */
462
	ret = -EINVAL;
463
	cpuset_for_each_child(c, css, par) {
Linus Torvalds's avatar
Linus Torvalds committed
464 465
		if ((is_cpu_exclusive(trial) || is_cpu_exclusive(c)) &&
		    c != cur &&
466
		    cpumask_intersects(trial->cpus_allowed, c->cpus_allowed))
467
			goto out;
Linus Torvalds's avatar
Linus Torvalds committed
468 469 470
		if ((is_mem_exclusive(trial) || is_mem_exclusive(c)) &&
		    c != cur &&
		    nodes_intersects(trial->mems_allowed, c->mems_allowed))
471
			goto out;
Linus Torvalds's avatar
Linus Torvalds committed
472 473
	}

474 475
	/*
	 * Cpusets with tasks - existing or newly being attached - can't
476
	 * be changed to have empty cpus_allowed or mems_allowed.
477
	 */
478
	ret = -ENOSPC;
479
	if ((cgroup_has_tasks(cur->css.cgroup) || cur->attach_in_progress)) {
480 481 482 483 484 485 486
		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;
	}
487

488 489 490 491
	ret = 0;
out:
	rcu_read_unlock();
	return ret;
Linus Torvalds's avatar
Linus Torvalds committed
492 493
}

494
#ifdef CONFIG_SMP
Paul Jackson's avatar
Paul Jackson committed
495
/*
496
 * Helper routine for generate_sched_domains().
Paul Jackson's avatar
Paul Jackson committed
497 498 499 500
 * Do cpusets a, b have overlapping cpus_allowed masks?
 */
static int cpusets_overlap(struct cpuset *a, struct cpuset *b)
{
501
	return cpumask_intersects(a->cpus_allowed, b->cpus_allowed);
Paul Jackson's avatar
Paul Jackson committed
502 503
}

504 505 506 507 508 509 510 511
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;
}

512 513
static void update_domain_attr_tree(struct sched_domain_attr *dattr,
				    struct cpuset *root_cs)
514
{
515
	struct cpuset *cp;
516
	struct cgroup_subsys_state *pos_css;
517

518
	rcu_read_lock();
519
	cpuset_for_each_descendant_pre(cp, pos_css, root_cs) {
520 521 522
		if (cp == root_cs)
			continue;

523 524
		/* skip the whole subtree if @cp doesn't have any CPU */
		if (cpumask_empty(cp->cpus_allowed)) {
525
			pos_css = css_rightmost_descendant(pos_css);
526
			continue;
527
		}
528 529 530 531

		if (is_sched_load_balance(cp))
			update_domain_attr(dattr, cp);
	}
532
	rcu_read_unlock();
533 534
}

Paul Jackson's avatar
Paul Jackson committed
535
/*
536 537 538 539 540
 * 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.
541
 * The output of this function needs to be passed to kernel/sched/core.c
542 543 544
 * 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
545
 *
546
 * See "What is sched_load_balance" in Documentation/cgroups/cpusets.txt
Paul Jackson's avatar
Paul Jackson committed
547 548 549 550 551 552 553
 * 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.
 *
554
 * Must be called with cpuset_mutex held.
Paul Jackson's avatar
Paul Jackson committed
555 556
 *
 * The three key local variables below are:
557
 *    q  - a linked-list queue of cpuset pointers, used to implement a
Paul Jackson's avatar
Paul Jackson committed
558 559 560 561 562 563 564 565 566 567 568 569
 *	   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
570
 *	   the kernel/sched/core.c routine partition_sched_domains() in a
Paul Jackson's avatar
Paul Jackson committed
571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588
 *	   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().
 */
589
static int generate_sched_domains(cpumask_var_t **domains,
590
			struct sched_domain_attr **attributes)
Paul Jackson's avatar
Paul Jackson committed
591 592 593 594 595
{
	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 */
596
	cpumask_var_t *doms;	/* resulting partition; i.e. sched domains */
597
	struct sched_domain_attr *dattr;  /* attributes for custom domains */
598
	int ndoms = 0;		/* number of sched domains in result */
599
	int nslot;		/* next empty doms[] struct cpumask slot */
600
	struct cgroup_subsys_state *pos_css;
Paul Jackson's avatar
Paul Jackson committed
601 602

	doms = NULL;
603
	dattr = NULL;
604
	csa = NULL;
Paul Jackson's avatar
Paul Jackson committed
605 606 607

	/* Special case for the 99% of systems with one, full, sched domain */
	if (is_sched_load_balance(&top_cpuset)) {
608 609
		ndoms = 1;
		doms = alloc_sched_domains(ndoms);
Paul Jackson's avatar
Paul Jackson committed
610
		if (!doms)
611 612
			goto done;

613 614 615
		dattr = kmalloc(sizeof(struct sched_domain_attr), GFP_KERNEL);
		if (dattr) {
			*dattr = SD_ATTR_INIT;
616
			update_domain_attr_tree(dattr, &top_cpuset);
617
		}
618
		cpumask_copy(doms[0], top_cpuset.cpus_allowed);
619 620

		goto done;
Paul Jackson's avatar
Paul Jackson committed
621 622
	}

623
	csa = kmalloc(nr_cpusets() * sizeof(cp), GFP_KERNEL);
Paul Jackson's avatar
Paul Jackson committed
624 625 626 627
	if (!csa)
		goto done;
	csn = 0;

628
	rcu_read_lock();
629
	cpuset_for_each_descendant_pre(cp, pos_css, &top_cpuset) {
630 631
		if (cp == &top_cpuset)
			continue;
632
		/*
633 634 635 636 637 638
		 * 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.
639
		 */
640 641
		if (!cpumask_empty(cp->cpus_allowed) &&
		    !is_sched_load_balance(cp))
642
			continue;
643

644 645 646 647
		if (is_sched_load_balance(cp))
			csa[csn++] = cp;

		/* skip @cp's subtree */
648
		pos_css = css_rightmost_descendant(pos_css);
649 650
	}
	rcu_read_unlock();
Paul Jackson's avatar
Paul Jackson committed
651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678

	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;
			}
		}
	}

679 680 681 682
	/*
	 * Now we know how many domains to create.
	 * Convert <csn, csa> to <ndoms, doms> and populate cpu masks.
	 */
683
	doms = alloc_sched_domains(ndoms);
684
	if (!doms)
685 686 687 688 689 690
		goto done;

	/*
	 * The rest of the code, including the scheduler, can deal with
	 * dattr==NULL case. No need to abort if alloc fails.
	 */
691
	dattr = kmalloc(ndoms * sizeof(struct sched_domain_attr), GFP_KERNEL);
Paul Jackson's avatar
Paul Jackson committed
692 693 694

	for (nslot = 0, i = 0; i < csn; i++) {
		struct cpuset *a = csa[i];
695
		struct cpumask *dp;
Paul Jackson's avatar
Paul Jackson committed
696 697
		int apn = a->pn;

698 699 700 701 702
		if (apn < 0) {
			/* Skip completed partitions */
			continue;
		}

703
		dp = doms[nslot];
704 705 706 707

		if (nslot == ndoms) {
			static int warnings = 10;
			if (warnings) {
708 709
				pr_warn("rebuild_sched_domains confused: nslot %d, ndoms %d, csn %d, i %d, apn %d\n",
					nslot, ndoms, csn, i, apn);
710
				warnings--;
Paul Jackson's avatar
Paul Jackson committed
711
			}
712 713
			continue;
		}
Paul Jackson's avatar
Paul Jackson committed
714

715
		cpumask_clear(dp);
716 717 718 719 720 721
		if (dattr)
			*(dattr + nslot) = SD_ATTR_INIT;
		for (j = i; j < csn; j++) {
			struct cpuset *b = csa[j];

			if (apn == b->pn) {
722
				cpumask_or(dp, dp, b->cpus_allowed);
723 724 725 726 727
				if (dattr)
					update_domain_attr_tree(dattr + nslot, b);

				/* Done with this partition */
				b->pn = -1;
Paul Jackson's avatar
Paul Jackson committed
728 729
			}
		}
730
		nslot++;
Paul Jackson's avatar
Paul Jackson committed
731 732 733
	}
	BUG_ON(nslot != ndoms);

734 735 736
done:
	kfree(csa);

737 738 739 740 741 742 743
	/*
	 * Fallback to the default domain if kmalloc() failed.
	 * See comments in partition_sched_domains().
	 */
	if (doms == NULL)
		ndoms = 1;

744 745 746 747 748 749 750 751
	*domains    = doms;
	*attributes = dattr;
	return ndoms;
}

/*
 * Rebuild scheduler domains.
 *
752 753 754 755 756
 * 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.
757
 *
758
 * Call with cpuset_mutex held.  Takes get_online_cpus().
759
 */
760
static void rebuild_sched_domains_locked(void)
761 762
{
	struct sched_domain_attr *attr;
763
	cpumask_var_t *doms;
764 765
	int ndoms;

766
	lockdep_assert_held(&cpuset_mutex);
767
	get_online_cpus();
768

769 770 771 772 773 774 775 776
	/*
	 * 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.
	 */
	if (!cpumask_equal(top_cpuset.cpus_allowed, cpu_active_mask))
		goto out;

777 778 779 780 781
	/* Generate domain masks and attrs */
	ndoms = generate_sched_domains(&doms, &attr);

	/* Have scheduler rebuild the domains */
	partition_sched_domains(ndoms, doms, attr);
782
out:
783
	put_online_cpus();
784
}
785
#else /* !CONFIG_SMP */
786
static void rebuild_sched_domains_locked(void)
787 788 789
{
}
#endif /* CONFIG_SMP */
Paul Jackson's avatar
Paul Jackson committed
790

791 792
void rebuild_sched_domains(void)
{
793
	mutex_lock(&cpuset_mutex);
794
	rebuild_sched_domains_locked();
795
	mutex_unlock(&cpuset_mutex);
Paul Jackson's avatar
Paul Jackson committed
796 797
}

798 799 800
/*
 * effective_cpumask_cpuset - return nearest ancestor with non-empty cpus
 * @cs: the cpuset in interest
801
 *
802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828
 * A cpuset's effective cpumask is the cpumask of the nearest ancestor
 * with non-empty cpus. We use effective cpumask whenever:
 * - we update tasks' cpus_allowed. (they take on the ancestor's cpumask
 *   if the cpuset they reside in has no cpus)
 * - we want to retrieve task_cs(tsk)'s cpus_allowed.
 *
 * Called with cpuset_mutex held. cpuset_cpus_allowed_fallback() is an
 * exception. See comments there.
 */
static struct cpuset *effective_cpumask_cpuset(struct cpuset *cs)
{
	while (cpumask_empty(cs->cpus_allowed))
		cs = parent_cs(cs);
	return cs;
}

/*
 * effective_nodemask_cpuset - return nearest ancestor with non-empty mems
 * @cs: the cpuset in interest
 *
 * A cpuset's effective nodemask is the nodemask of the nearest ancestor
 * with non-empty memss. We use effective nodemask whenever:
 * - we update tasks' mems_allowed. (they take on the ancestor's nodemask
 *   if the cpuset they reside in has no mems)
 * - we want to retrieve task_cs(tsk)'s mems_allowed.
 *
 * Called with cpuset_mutex held.
829
 */
830
static struct cpuset *effective_nodemask_cpuset(struct cpuset *cs)
831
{
832 833 834
	while (nodes_empty(cs->mems_allowed))
		cs = parent_cs(cs);
	return cs;
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
{
847 848 849 850 851 852 853 854
	struct cpuset *cpus_cs = effective_cpumask_cpuset(cs);
	struct css_task_iter it;
	struct task_struct *task;

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

857 858 859 860 861 862 863 864 865 866
/*
 * update_tasks_cpumask_hier - Update the cpumasks of tasks in the hierarchy.
 * @root_cs: the root cpuset of the hierarchy
 * @update_root: update root cpuset or not?
 *
 * This will update cpumasks of tasks in @root_cs and all other empty cpusets
 * which take on cpumask of @root_cs.
 *
 * Called with cpuset_mutex held
 */
867
static void update_tasks_cpumask_hier(struct cpuset *root_cs, bool update_root)
868 869
{
	struct cpuset *cp;
870
	struct cgroup_subsys_state *pos_css;
871 872

	rcu_read_lock();
873
	cpuset_for_each_descendant_pre(cp, pos_css, root_cs) {
874 875 876 877 878 879 880 881 882
		if (cp == root_cs) {
			if (!update_root)
				continue;
		} else {
			/* skip the whole subtree if @cp have some CPU */
			if (!cpumask_empty(cp->cpus_allowed)) {
				pos_css = css_rightmost_descendant(pos_css);
				continue;
			}
883
		}
884
		if (!css_tryget_online(&cp->css))
885 886 887
			continue;
		rcu_read_unlock();

888
		update_tasks_cpumask(cp);
889 890 891 892 893 894 895

		rcu_read_lock();
		css_put(&cp->css);
	}
	rcu_read_unlock();
}

896 897 898
/**
 * update_cpumask - update the cpus_allowed mask of a cpuset and all tasks in it
 * @cs: the cpuset to consider
899
 * @trialcs: trial cpuset
900 901
 * @buf: buffer of cpu numbers written to this cpuset
 */
902 903
static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs,
			  const char *buf)
Linus Torvalds's avatar
Linus Torvalds committed
904
{
905 906
	int retval;
	int is_load_balanced;
Linus Torvalds's avatar
Linus Torvalds committed
907

908
	/* top_cpuset.cpus_allowed tracks cpu_online_mask; it's read-only */
909 910 911
	if (cs == &top_cpuset)
		return -EACCES;

912
	/*
913
	 * An empty cpus_allowed is ok only if the cpuset has no tasks.
914 915 916
	 * Since cpulist_parse() fails on an empty mask, we special case
	 * that parsing.  The validate_change() call ensures that cpusets
	 * with tasks have cpus.
917
	 */
918
	if (!*buf) {
919
		cpumask_clear(trialcs->cpus_allowed);
920
	} else {
921
		retval = cpulist_parse(buf, trialcs->cpus_allowed);
922 923
		if (retval < 0)
			return retval;
924

925
		if (!cpumask_subset(trialcs->cpus_allowed, cpu_active_mask))
926
			return -EINVAL;
927
	}
Paul Jackson's avatar
Paul Jackson committed
928

Paul Menage's avatar
Paul Menage committed
929
	/* Nothing to do if the cpus didn't change */
930
	if (cpumask_equal(cs->cpus_allowed, trialcs->cpus_allowed))
Paul Menage's avatar
Paul Menage committed
931
		return 0;
932

933 934 935 936
	retval = validate_change(cs, trialcs);
	if (retval < 0)
		return retval;

937
	is_load_balanced = is_sched_load_balance(trialcs);
Paul Jackson's avatar
Paul Jackson committed
938

939
	mutex_lock(&callback_mutex);
940
	cpumask_copy(cs->cpus_allowed, trialcs->cpus_allowed);
941
	mutex_unlock(&callback_mutex);
Paul Jackson's avatar
Paul Jackson committed
942

943
	update_tasks_cpumask_hier(cs, true);
944

Paul Menage's avatar
Paul Menage committed
945
	if (is_load_balanced)
946
		rebuild_sched_domains_locked();
947
	return 0;
Linus Torvalds's avatar
Linus Torvalds committed
948 949
}

950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967
/*
 * 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;
968
	struct cpuset *mems_cs;
969 970 971 972 973

	tsk->mems_allowed = *to;

	do_migrate_pages(mm, from, to, MPOL_MF_MOVE_ALL);

974
	rcu_read_lock();
975 976
	mems_cs = effective_nodemask_cpuset(task_cs(tsk));
	guarantee_online_mems(mems_cs, &tsk->mems_allowed);
977
	rcu_read_unlock();
978 979
}

980
/*
981 982 983 984 985 986 987 988 989 990 991
 * cpuset_change_task_nodemask - change task's mems_allowed and mempolicy
 * @tsk: the task to change
 * @newmems: new nodes that the task will be set
 *
 * In order to avoid seeing no nodes if the old and new nodes are disjoint,
 * we structure updates as setting all new allowed nodes, then clearing newly
 * disallowed ones.
 */
static void cpuset_change_task_nodemask(struct task_struct *tsk,
					nodemask_t *newmems)
{
992
	bool need_loop;
993

994 995 996 997 998 999 1000 1001 1002 1003
	/*
	 * Allow tasks that have access to memory reserves because they have
	 * been OOM killed to get memory anywhere.
	 */
	if (unlikely(test_thread_flag(TIF_MEMDIE)))
		return;
	if (current->flags & PF_EXITING) /* Let dying task have memory */
		return;

	task_lock(tsk);
1004 1005
	/*
	 * Determine if a loop is necessary if another thread is doing
1006
	 * read_mems_allowed_begin().  If at least one node remains unchanged and
1007 1008 1009 1010 1011
	 * tsk does not have a mempolicy, then an empty nodemask will not be
	 * possible when mems_allowed is larger than a word.
	 */
	need_loop = task_has_mempolicy(tsk) ||
			!nodes_intersects(*newmems, tsk->mems_allowed);
1012

1013 1014
	if (need_loop) {
		local_irq_disable();
1015
		write_seqcount_begin(&tsk->mems_allowed_seq);
1016
	}
1017

1018 1019
	nodes_or(tsk->mems_allowed, tsk->mems_allowed, *newmems);
	mpol_rebind_task(tsk, newmems, MPOL_REBIND_STEP1);
1020 1021

	mpol_rebind_task(tsk, newmems, MPOL_REBIND_STEP2);
1022
	tsk->mems_allowed = *newmems;
1023

1024
	if (need_loop) {
1025
		write_seqcount_end(&tsk->mems_allowed_seq);
1026 1027
		local_irq_enable();
	}
1028

1029
	task_unlock(tsk);
1030 1031
}

1032 1033
static void *cpuset_being_rebound;

1034 1035 1036 1037
/**
 * update_tasks_nodemask - Update the nodemasks of tasks in the cpuset.
 * @cs: the cpuset in which each task's mems_allowed mask needs to be changed
 *
1038 1039 1040
 * Iterate through each task of @cs updating its mems_allowed to the
 * effective cpuset's.  As this function is called with cpuset_mutex held,
 * cpuset membership stays stable.
1041
 */
1042
static void update_tasks_nodemask(struct cpuset *cs)
Linus Torvalds's avatar
Linus Torvalds committed
1043
{
1044
	static nodemask_t newmems;	/* protected by cpuset_mutex */