cpuset.c 74.3 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
	cpumask_var_t cpus_allowed;	/* CPUs allowed to tasks in cpuset */
Linus Torvalds's avatar
Linus Torvalds committed
80 81
	nodemask_t mems_allowed;	/* Memory Nodes allowed to tasks */

82 83 84 85 86 87 88 89 90 91 92 93
	/*
	 * 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;

94
	struct fmeter fmeter;		/* memory_pressure filter */
Paul Jackson's avatar
Paul Jackson committed
95

96 97 98 99 100 101
	/*
	 * 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
102 103
	/* partition number for rebuild_sched_domains() */
	int pn;
104

105 106
	/* for custom sched domain */
	int relax_domain_level;
Linus Torvalds's avatar
Linus Torvalds committed
107 108
};

109
static inline struct cpuset *css_cs(struct cgroup_subsys_state *css)
110
{
111
	return css ? container_of(css, struct cpuset, css) : NULL;
112 113 114 115 116
}

/* Retrieve the cpuset for a task */
static inline struct cpuset *task_cs(struct task_struct *task)
{
117
	return css_cs(task_css(task, cpuset_cgrp_id));
118 119
}

120
static inline struct cpuset *parent_cs(struct cpuset *cs)
Tejun Heo's avatar
Tejun Heo committed
121
{
Tejun Heo's avatar
Tejun Heo committed
122
	return css_cs(cs->css.parent);
Tejun Heo's avatar
Tejun Heo committed
123 124
}

125 126 127 128 129 130 131 132 133 134 135 136 137
#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
138 139
/* bits in struct cpuset flags field */
typedef enum {
Tejun Heo's avatar
Tejun Heo committed
140
	CS_ONLINE,
Linus Torvalds's avatar
Linus Torvalds committed
141 142
	CS_CPU_EXCLUSIVE,
	CS_MEM_EXCLUSIVE,
143
	CS_MEM_HARDWALL,
144
	CS_MEMORY_MIGRATE,
Paul Jackson's avatar
Paul Jackson committed
145
	CS_SCHED_LOAD_BALANCE,
146 147
	CS_SPREAD_PAGE,
	CS_SPREAD_SLAB,
Linus Torvalds's avatar
Linus Torvalds committed
148 149 150
} cpuset_flagbits_t;

/* convenient tests for these bits */
Tejun Heo's avatar
Tejun Heo committed
151 152 153 154 155
static inline bool is_cpuset_online(const struct cpuset *cs)
{
	return test_bit(CS_ONLINE, &cs->flags);
}

Linus Torvalds's avatar
Linus Torvalds committed
156 157
static inline int is_cpu_exclusive(const struct cpuset *cs)
{
158
	return test_bit(CS_CPU_EXCLUSIVE, &cs->flags);
Linus Torvalds's avatar
Linus Torvalds committed
159 160 161 162
}

static inline int is_mem_exclusive(const struct cpuset *cs)
{
163
	return test_bit(CS_MEM_EXCLUSIVE, &cs->flags);
Linus Torvalds's avatar
Linus Torvalds committed
164 165
}

166 167 168 169 170
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
171 172 173 174 175
static inline int is_sched_load_balance(const struct cpuset *cs)
{
	return test_bit(CS_SCHED_LOAD_BALANCE, &cs->flags);
}

176 177
static inline int is_memory_migrate(const struct cpuset *cs)
{
178
	return test_bit(CS_MEMORY_MIGRATE, &cs->flags);
179 180
}

181 182 183 184 185 186 187 188 189 190
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
191
static struct cpuset top_cpuset = {
Tejun Heo's avatar
Tejun Heo committed
192 193
	.flags = ((1 << CS_ONLINE) | (1 << CS_CPU_EXCLUSIVE) |
		  (1 << CS_MEM_EXCLUSIVE)),
Linus Torvalds's avatar
Linus Torvalds committed
194 195
};

196 197 198
/**
 * cpuset_for_each_child - traverse online children of a cpuset
 * @child_cs: loop cursor pointing to the current child
199
 * @pos_css: used for iteration
200 201 202 203 204
 * @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.
 */
205 206 207
#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)))))
208

209 210 211
/**
 * cpuset_for_each_descendant_pre - pre-order walk of a cpuset's descendants
 * @des_cs: loop cursor pointing to the current descendant
212
 * @pos_css: used for iteration
213 214 215
 * @root_cs: target cpuset to walk ancestor of
 *
 * Walk @des_cs through the online descendants of @root_cs.  Must be used
216
 * with RCU read locked.  The caller may modify @pos_css by calling
217 218
 * css_rightmost_descendant() to skip subtree.  @root_cs is included in the
 * iteration and the first node to be visited.
219
 */
220 221 222
#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)))))
223

Linus Torvalds's avatar
Linus Torvalds committed
224
/*
225 226 227 228 229 230 231 232 233 234 235 236 237 238
 * 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.
239 240
 *
 * Calls to the kernel memory allocator can not be made while holding
241
 * callback_mutex, as that would risk double tripping on callback_mutex
242 243 244
 * from one of the callbacks into the cpuset code from within
 * __alloc_pages().
 *
245
 * If a task is only holding callback_mutex, then it has read-only
246 247
 * access to cpusets.
 *
248 249 250
 * 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.
251
 *
252
 * The cpuset_common_file_read() handlers only hold callback_mutex across
253 254 255
 * small pieces of code, such as when reading out possibly multi-word
 * cpumasks and nodemasks.
 *
256 257
 * 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
258 259
 */

260
static DEFINE_MUTEX(cpuset_mutex);
261
static DEFINE_MUTEX(callback_mutex);
262

263 264 265 266 267 268
/*
 * CPU / memory hotplug is handled asynchronously.
 */
static void cpuset_hotplug_workfn(struct work_struct *work);
static DECLARE_WORK(cpuset_hotplug_work, cpuset_hotplug_workfn);

269 270
static DECLARE_WAIT_QUEUE_HEAD(cpuset_attach_wq);

271 272
/*
 * This is ugly, but preserves the userspace API for existing cpuset
273
 * users. If someone tries to mount the "cpuset" filesystem, we
274 275
 * silently switch it to mount "cgroup" instead
 */
Al Viro's avatar
Al Viro committed
276 277
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
278
{
279
	struct file_system_type *cgroup_fs = get_fs_type("cgroup");
Al Viro's avatar
Al Viro committed
280
	struct dentry *ret = ERR_PTR(-ENODEV);
281 282 283 284
	if (cgroup_fs) {
		char mountopts[] =
			"cpuset,noprefix,"
			"release_agent=/sbin/cpuset_release_agent";
Al Viro's avatar
Al Viro committed
285 286
		ret = cgroup_fs->mount(cgroup_fs, flags,
					   unused_dev_name, mountopts);
287 288 289
		put_filesystem(cgroup_fs);
	}
	return ret;
Linus Torvalds's avatar
Linus Torvalds committed
290 291 292 293
}

static struct file_system_type cpuset_fs_type = {
	.name = "cpuset",
Al Viro's avatar
Al Viro committed
294
	.mount = cpuset_mount,
Linus Torvalds's avatar
Linus Torvalds committed
295 296 297
};

/*
298
 * Return in pmask the portion of a cpusets's cpus_allowed that
Linus Torvalds's avatar
Linus Torvalds committed
299
 * are online.  If none are online, walk up the cpuset hierarchy
300 301
 * 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
302 303
 *
 * One way or another, we guarantee to return some non-empty subset
304
 * of cpu_online_mask.
Linus Torvalds's avatar
Linus Torvalds committed
305
 *
306
 * Call with callback_mutex held.
Linus Torvalds's avatar
Linus Torvalds committed
307
 */
308
static void guarantee_online_cpus(struct cpuset *cs, struct cpumask *pmask)
Linus Torvalds's avatar
Linus Torvalds committed
309
{
310
	while (!cpumask_intersects(cs->cpus_allowed, cpu_online_mask))
Tejun Heo's avatar
Tejun Heo committed
311
		cs = parent_cs(cs);
312
	cpumask_and(pmask, cs->cpus_allowed, cpu_online_mask);
Linus Torvalds's avatar
Linus Torvalds committed
313 314 315 316
}

/*
 * Return in *pmask the portion of a cpusets's mems_allowed that
317 318
 * are online, with memory.  If none are online with memory, walk
 * up the cpuset hierarchy until we find one that does have some
319
 * online mems.  The top cpuset always has some mems online.
Linus Torvalds's avatar
Linus Torvalds committed
320 321
 *
 * One way or another, we guarantee to return some non-empty subset
322
 * of node_states[N_MEMORY].
Linus Torvalds's avatar
Linus Torvalds committed
323
 *
324
 * Call with callback_mutex held.
Linus Torvalds's avatar
Linus Torvalds committed
325
 */
326
static void guarantee_online_mems(struct cpuset *cs, nodemask_t *pmask)
Linus Torvalds's avatar
Linus Torvalds committed
327
{
328
	while (!nodes_intersects(cs->mems_allowed, node_states[N_MEMORY]))
Tejun Heo's avatar
Tejun Heo committed
329
		cs = parent_cs(cs);
330
	nodes_and(*pmask, cs->mems_allowed, node_states[N_MEMORY]);
Linus Torvalds's avatar
Linus Torvalds committed
331 332
}

333 334 335
/*
 * update task's spread flag if cpuset's page/slab spread flag is set
 *
336
 * Called with callback_mutex/cpuset_mutex held
337 338 339 340 341 342 343 344 345 346 347 348 349 350
 */
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
351 352 353 354 355
/*
 * 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
356
 * are only set if the other's are set.  Call holding cpuset_mutex.
Linus Torvalds's avatar
Linus Torvalds committed
357 358 359 360
 */

static int is_cpuset_subset(const struct cpuset *p, const struct cpuset *q)
{
361
	return	cpumask_subset(p->cpus_allowed, q->cpus_allowed) &&
Linus Torvalds's avatar
Linus Torvalds committed
362 363 364 365 366
		nodes_subset(p->mems_allowed, q->mems_allowed) &&
		is_cpu_exclusive(p) <= is_cpu_exclusive(q) &&
		is_mem_exclusive(p) <= is_mem_exclusive(q);
}

367 368 369 370
/**
 * alloc_trial_cpuset - allocate a trial cpuset
 * @cs: the cpuset that the trial cpuset duplicates
 */
371
static struct cpuset *alloc_trial_cpuset(struct cpuset *cs)
372
{
373 374 375 376 377 378 379 380 381 382 383 384 385
	struct cpuset *trial;

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

	if (!alloc_cpumask_var(&trial->cpus_allowed, GFP_KERNEL)) {
		kfree(trial);
		return NULL;
	}
	cpumask_copy(trial->cpus_allowed, cs->cpus_allowed);

	return trial;
386 387 388 389 390 391 392 393
}

/**
 * free_trial_cpuset - free the trial cpuset
 * @trial: the trial cpuset to be freed
 */
static void free_trial_cpuset(struct cpuset *trial)
{
394
	free_cpumask_var(trial->cpus_allowed);
395 396 397
	kfree(trial);
}

Linus Torvalds's avatar
Linus Torvalds committed
398 399 400 401 402 403 404
/*
 * 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
405
 * cpuset_mutex held.
Linus Torvalds's avatar
Linus Torvalds committed
406 407 408 409 410 411 412 413 414 415 416 417
 *
 * '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.
 */

418
static int validate_change(struct cpuset *cur, struct cpuset *trial)
Linus Torvalds's avatar
Linus Torvalds committed
419
{
420
	struct cgroup_subsys_state *css;
Linus Torvalds's avatar
Linus Torvalds committed
421
	struct cpuset *c, *par;
422 423 424
	int ret;

	rcu_read_lock();
Linus Torvalds's avatar
Linus Torvalds committed
425 426

	/* Each of our child cpusets must be a subset of us */
427
	ret = -EBUSY;
428
	cpuset_for_each_child(c, css, cur)
429 430
		if (!is_cpuset_subset(c, trial))
			goto out;
Linus Torvalds's avatar
Linus Torvalds committed
431 432

	/* Remaining checks don't apply to root cpuset */
433
	ret = 0;
434
	if (cur == &top_cpuset)
435
		goto out;
Linus Torvalds's avatar
Linus Torvalds committed
436

Tejun Heo's avatar
Tejun Heo committed
437
	par = parent_cs(cur);
438

Linus Torvalds's avatar
Linus Torvalds committed
439
	/* We must be a subset of our parent cpuset */
440
	ret = -EACCES;
Linus Torvalds's avatar
Linus Torvalds committed
441
	if (!is_cpuset_subset(trial, par))
442
		goto out;
Linus Torvalds's avatar
Linus Torvalds committed
443

444 445 446 447
	/*
	 * If either I or some sibling (!= me) is exclusive, we can't
	 * overlap
	 */
448
	ret = -EINVAL;
449
	cpuset_for_each_child(c, css, par) {
Linus Torvalds's avatar
Linus Torvalds committed
450 451
		if ((is_cpu_exclusive(trial) || is_cpu_exclusive(c)) &&
		    c != cur &&
452
		    cpumask_intersects(trial->cpus_allowed, c->cpus_allowed))
453
			goto out;
Linus Torvalds's avatar
Linus Torvalds committed
454 455 456
		if ((is_mem_exclusive(trial) || is_mem_exclusive(c)) &&
		    c != cur &&
		    nodes_intersects(trial->mems_allowed, c->mems_allowed))
457
			goto out;
Linus Torvalds's avatar
Linus Torvalds committed
458 459
	}

460 461
	/*
	 * Cpusets with tasks - existing or newly being attached - can't
462
	 * be changed to have empty cpus_allowed or mems_allowed.
463
	 */
464
	ret = -ENOSPC;
465
	if ((cgroup_has_tasks(cur->css.cgroup) || cur->attach_in_progress)) {
466 467 468 469 470 471 472
		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;
	}
473

474 475 476 477
	ret = 0;
out:
	rcu_read_unlock();
	return ret;
Linus Torvalds's avatar
Linus Torvalds committed
478 479
}

480
#ifdef CONFIG_SMP
Paul Jackson's avatar
Paul Jackson committed
481
/*
482
 * Helper routine for generate_sched_domains().
Paul Jackson's avatar
Paul Jackson committed
483 484 485 486
 * Do cpusets a, b have overlapping cpus_allowed masks?
 */
static int cpusets_overlap(struct cpuset *a, struct cpuset *b)
{
487
	return cpumask_intersects(a->cpus_allowed, b->cpus_allowed);
Paul Jackson's avatar
Paul Jackson committed
488 489
}

490 491 492 493 494 495 496 497
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;
}

498 499
static void update_domain_attr_tree(struct sched_domain_attr *dattr,
				    struct cpuset *root_cs)
500
{
501
	struct cpuset *cp;
502
	struct cgroup_subsys_state *pos_css;
503

504
	rcu_read_lock();
505
	cpuset_for_each_descendant_pre(cp, pos_css, root_cs) {
506 507 508
		if (cp == root_cs)
			continue;

509 510
		/* skip the whole subtree if @cp doesn't have any CPU */
		if (cpumask_empty(cp->cpus_allowed)) {
511
			pos_css = css_rightmost_descendant(pos_css);
512
			continue;
513
		}
514 515 516 517

		if (is_sched_load_balance(cp))
			update_domain_attr(dattr, cp);
	}
518
	rcu_read_unlock();
519 520
}

Paul Jackson's avatar
Paul Jackson committed
521
/*
522 523 524 525 526
 * 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.
527
 * The output of this function needs to be passed to kernel/sched/core.c
528 529 530
 * 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
531
 *
532
 * See "What is sched_load_balance" in Documentation/cgroups/cpusets.txt
Paul Jackson's avatar
Paul Jackson committed
533 534 535 536 537 538 539
 * 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.
 *
540
 * Must be called with cpuset_mutex held.
Paul Jackson's avatar
Paul Jackson committed
541 542
 *
 * The three key local variables below are:
543
 *    q  - a linked-list queue of cpuset pointers, used to implement a
Paul Jackson's avatar
Paul Jackson committed
544 545 546 547 548 549 550 551 552 553 554 555
 *	   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
556
 *	   the kernel/sched/core.c routine partition_sched_domains() in a
Paul Jackson's avatar
Paul Jackson committed
557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574
 *	   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().
 */
575
static int generate_sched_domains(cpumask_var_t **domains,
576
			struct sched_domain_attr **attributes)
Paul Jackson's avatar
Paul Jackson committed
577 578 579 580 581
{
	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 */
582
	cpumask_var_t *doms;	/* resulting partition; i.e. sched domains */
583
	struct sched_domain_attr *dattr;  /* attributes for custom domains */
584
	int ndoms = 0;		/* number of sched domains in result */
585
	int nslot;		/* next empty doms[] struct cpumask slot */
586
	struct cgroup_subsys_state *pos_css;
Paul Jackson's avatar
Paul Jackson committed
587 588

	doms = NULL;
589
	dattr = NULL;
590
	csa = NULL;
Paul Jackson's avatar
Paul Jackson committed
591 592 593

	/* Special case for the 99% of systems with one, full, sched domain */
	if (is_sched_load_balance(&top_cpuset)) {
594 595
		ndoms = 1;
		doms = alloc_sched_domains(ndoms);
Paul Jackson's avatar
Paul Jackson committed
596
		if (!doms)
597 598
			goto done;

599 600 601
		dattr = kmalloc(sizeof(struct sched_domain_attr), GFP_KERNEL);
		if (dattr) {
			*dattr = SD_ATTR_INIT;
602
			update_domain_attr_tree(dattr, &top_cpuset);
603
		}
604
		cpumask_copy(doms[0], top_cpuset.cpus_allowed);
605 606

		goto done;
Paul Jackson's avatar
Paul Jackson committed
607 608
	}

609
	csa = kmalloc(nr_cpusets() * sizeof(cp), GFP_KERNEL);
Paul Jackson's avatar
Paul Jackson committed
610 611 612 613
	if (!csa)
		goto done;
	csn = 0;

614
	rcu_read_lock();
615
	cpuset_for_each_descendant_pre(cp, pos_css, &top_cpuset) {
616 617
		if (cp == &top_cpuset)
			continue;
618
		/*
619 620 621 622 623 624
		 * 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.
625
		 */
626 627
		if (!cpumask_empty(cp->cpus_allowed) &&
		    !is_sched_load_balance(cp))
628
			continue;
629

630 631 632 633
		if (is_sched_load_balance(cp))
			csa[csn++] = cp;

		/* skip @cp's subtree */
634
		pos_css = css_rightmost_descendant(pos_css);
635 636
	}
	rcu_read_unlock();
Paul Jackson's avatar
Paul Jackson committed
637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664

	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;
			}
		}
	}

665 666 667 668
	/*
	 * Now we know how many domains to create.
	 * Convert <csn, csa> to <ndoms, doms> and populate cpu masks.
	 */
669
	doms = alloc_sched_domains(ndoms);
670
	if (!doms)
671 672 673 674 675 676
		goto done;

	/*
	 * The rest of the code, including the scheduler, can deal with
	 * dattr==NULL case. No need to abort if alloc fails.
	 */
677
	dattr = kmalloc(ndoms * sizeof(struct sched_domain_attr), GFP_KERNEL);
Paul Jackson's avatar
Paul Jackson committed
678 679 680

	for (nslot = 0, i = 0; i < csn; i++) {
		struct cpuset *a = csa[i];
681
		struct cpumask *dp;
Paul Jackson's avatar
Paul Jackson committed
682 683
		int apn = a->pn;

684 685 686 687 688
		if (apn < 0) {
			/* Skip completed partitions */
			continue;
		}

689
		dp = doms[nslot];
690 691 692 693

		if (nslot == ndoms) {
			static int warnings = 10;
			if (warnings) {
694 695
				pr_warn("rebuild_sched_domains confused: nslot %d, ndoms %d, csn %d, i %d, apn %d\n",
					nslot, ndoms, csn, i, apn);
696
				warnings--;
Paul Jackson's avatar
Paul Jackson committed
697
			}
698 699
			continue;
		}
Paul Jackson's avatar
Paul Jackson committed
700

701
		cpumask_clear(dp);
702 703 704 705 706 707
		if (dattr)
			*(dattr + nslot) = SD_ATTR_INIT;
		for (j = i; j < csn; j++) {
			struct cpuset *b = csa[j];

			if (apn == b->pn) {
708
				cpumask_or(dp, dp, b->cpus_allowed);
709 710 711 712 713
				if (dattr)
					update_domain_attr_tree(dattr + nslot, b);

				/* Done with this partition */
				b->pn = -1;
Paul Jackson's avatar
Paul Jackson committed
714 715
			}
		}
716
		nslot++;
Paul Jackson's avatar
Paul Jackson committed
717 718 719
	}
	BUG_ON(nslot != ndoms);

720 721 722
done:
	kfree(csa);

723 724 725 726 727 728 729
	/*
	 * Fallback to the default domain if kmalloc() failed.
	 * See comments in partition_sched_domains().
	 */
	if (doms == NULL)
		ndoms = 1;

730 731 732 733 734 735 736 737
	*domains    = doms;
	*attributes = dattr;
	return ndoms;
}

/*
 * Rebuild scheduler domains.
 *
738 739 740 741 742
 * 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.
743
 *
744
 * Call with cpuset_mutex held.  Takes get_online_cpus().
745
 */
746
static void rebuild_sched_domains_locked(void)
747 748
{
	struct sched_domain_attr *attr;
749
	cpumask_var_t *doms;
750 751
	int ndoms;

752
	lockdep_assert_held(&cpuset_mutex);
753
	get_online_cpus();
754

755 756 757 758 759 760 761 762
	/*
	 * 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;

763 764 765 766 767
	/* Generate domain masks and attrs */
	ndoms = generate_sched_domains(&doms, &attr);

	/* Have scheduler rebuild the domains */
	partition_sched_domains(ndoms, doms, attr);
768
out:
769
	put_online_cpus();
770
}
771
#else /* !CONFIG_SMP */
772
static void rebuild_sched_domains_locked(void)
773 774 775
{
}
#endif /* CONFIG_SMP */
Paul Jackson's avatar
Paul Jackson committed
776

777 778
void rebuild_sched_domains(void)
{
779
	mutex_lock(&cpuset_mutex);
780
	rebuild_sched_domains_locked();
781
	mutex_unlock(&cpuset_mutex);
Paul Jackson's avatar
Paul Jackson committed
782 783
}

784 785 786
/*
 * effective_cpumask_cpuset - return nearest ancestor with non-empty cpus
 * @cs: the cpuset in interest
787
 *
788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814
 * 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.
815
 */
816
static struct cpuset *effective_nodemask_cpuset(struct cpuset *cs)
817
{
818 819 820
	while (nodes_empty(cs->mems_allowed))
		cs = parent_cs(cs);
	return cs;
821
}
822

823 824 825 826
/**
 * 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
 *
827 828 829
 * 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.
830
 */
831
static void update_tasks_cpumask(struct cpuset *cs)
832
{
833 834 835 836 837 838 839 840
	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);
841 842
}

843 844 845 846 847 848 849 850 851 852
/*
 * 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
 */
853
static void update_tasks_cpumask_hier(struct cpuset *root_cs, bool update_root)
854 855
{
	struct cpuset *cp;
856
	struct cgroup_subsys_state *pos_css;
857 858

	rcu_read_lock();
859
	cpuset_for_each_descendant_pre(cp, pos_css, root_cs) {
860 861 862 863 864 865 866 867 868
		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;
			}
869
		}
870
		if (!css_tryget_online(&cp->css))
871 872 873
			continue;
		rcu_read_unlock();

874
		update_tasks_cpumask(cp);
875 876 877 878 879 880 881

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

882 883 884
/**
 * update_cpumask - update the cpus_allowed mask of a cpuset and all tasks in it
 * @cs: the cpuset to consider
885
 * @trialcs: trial cpuset
886 887
 * @buf: buffer of cpu numbers written to this cpuset
 */
888 889
static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs,
			  const char *buf)
Linus Torvalds's avatar
Linus Torvalds committed
890
{
891 892
	int retval;
	int is_load_balanced;
Linus Torvalds's avatar
Linus Torvalds committed
893

894
	/* top_cpuset.cpus_allowed tracks cpu_online_mask; it's read-only */
895 896 897
	if (cs == &top_cpuset)
		return -EACCES;

898
	/*
899
	 * An empty cpus_allowed is ok only if the cpuset has no tasks.
900 901 902
	 * Since cpulist_parse() fails on an empty mask, we special case
	 * that parsing.  The validate_change() call ensures that cpusets
	 * with tasks have cpus.
903
	 */
904
	if (!*buf) {
905
		cpumask_clear(trialcs->cpus_allowed);
906
	} else {
907
		retval = cpulist_parse(buf, trialcs->cpus_allowed);
908 909
		if (retval < 0)
			return retval;
910

911
		if (!cpumask_subset(trialcs->cpus_allowed, cpu_active_mask))
912
			return -EINVAL;
913
	}
Paul Jackson's avatar
Paul Jackson committed
914

Paul Menage's avatar
Paul Menage committed
915
	/* Nothing to do if the cpus didn't change */
916
	if (cpumask_equal(cs->cpus_allowed, trialcs->cpus_allowed))
Paul Menage's avatar
Paul Menage committed
917
		return 0;
918

919 920 921 922
	retval = validate_change(cs, trialcs);
	if (retval < 0)
		return retval;

923
	is_load_balanced = is_sched_load_balance(trialcs);
Paul Jackson's avatar
Paul Jackson committed
924

925
	mutex_lock(&callback_mutex);
926
	cpumask_copy(cs->cpus_allowed, trialcs->cpus_allowed);
927
	mutex_unlock(&callback_mutex);
Paul Jackson's avatar
Paul Jackson committed
928

929
	update_tasks_cpumask_hier(cs, true);
930

Paul Menage's avatar
Paul Menage committed
931
	if (is_load_balanced)
932
		rebuild_sched_domains_locked();
933
	return 0;
Linus Torvalds's avatar
Linus Torvalds committed
934 935
}

936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953
/*
 * 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;
954
	struct cpuset *mems_cs;
955 956 957 958 959

	tsk->mems_allowed = *to;

	do_migrate_pages(mm, from, to, MPOL_MF_MOVE_ALL);

960
	rcu_read_lock();
961 962
	mems_cs = effective_nodemask_cpuset(task_cs(tsk));
	guarantee_online_mems(mems_cs, &tsk->mems_allowed);
963
	rcu_read_unlock();
964 965
}

966
/*
967 968 969 970 971 972 973 974 975 976 977
 * 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)
{
978
	bool need_loop;
979

980 981 982 983 984 985 986 987 988 989
	/*
	 * 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);
990 991
	/*
	 * Determine if a loop is necessary if another thread is doing
992
	 * read_mems_allowed_begin().  If at least one node remains unchanged and
993 994 995 996 997
	 * 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);
998

999 1000
	if (need_loop) {
		local_irq_disable();
1001
		write_seqcount_begin(&tsk->mems_allowed_seq);
1002
	}
1003

1004 1005
	nodes_or(tsk->mems_allowed, tsk->mems_allowed, *newmems);
	mpol_rebind_task(tsk, newmems, MPOL_REBIND_STEP1);
1006 1007

	mpol_rebind_task(tsk, newmems, MPOL_REBIND_STEP2);
1008
	tsk->mems_allowed = *newmems;
1009

1010
	if (need_loop) {
1011
		write_seqcount_end(&tsk->mems_allowed_seq);
1012 1013
		local_irq_enable();
	}
1014

1015
	task_unlock(tsk);
1016 1017
}

1018 1019
static void *cpuset_being_rebound;

1020 1021 1022 1023
/**
 * 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
 *
1024 1025 1026
 * 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.
1027
 */
1028
static void update_tasks_nodemask(struct cpuset *cs)
Linus Torvalds's avatar
Linus Torvalds committed
1029
{
1030
	static nodemask_t newmems;	/* protected by cpuset_mutex */
1031
	struct cpuset *mems_cs = effective_nodemask_cpuset(cs);
1032 1033
	struct css_task_iter it;
	struct task_struct *task;
1034

1035
	cpuset_being_rebound = cs;		/* causes mpol_dup() rebind */
1036

1037
	guarantee_online_mems(mems_cs, &newmems);
1038