cpuset.c 72.7 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 65 66 67 68
/*
 * Tracks how many cpusets are currently defined in system.
 * When there is only one cpuset (the root cpuset) we can
 * short circuit some hooks.
 */
69
int number_of_cpusets __read_mostly;
70

71
/* Forward declare cgroup structures */
72 73 74
struct cgroup_subsys cpuset_subsys;
struct cpuset;

75 76 77 78 79 80 81 82 83
/* 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
84
struct cpuset {
85 86
	struct cgroup_subsys_state css;

Linus Torvalds's avatar
Linus Torvalds committed
87
	unsigned long flags;		/* "unsigned long" so bitops work */
88
	cpumask_var_t cpus_allowed;	/* CPUs allowed to tasks in cpuset */
Linus Torvalds's avatar
Linus Torvalds committed
89 90
	nodemask_t mems_allowed;	/* Memory Nodes allowed to tasks */

91
	struct fmeter fmeter;		/* memory_pressure filter */
Paul Jackson's avatar
Paul Jackson committed
92

93 94 95 96 97 98
	/*
	 * 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
99 100
	/* partition number for rebuild_sched_domains() */
	int pn;
101

102 103
	/* for custom sched domain */
	int relax_domain_level;
Linus Torvalds's avatar
Linus Torvalds committed
104 105
};

106 107 108 109 110 111 112 113 114 115 116 117 118 119
/* Retrieve the cpuset for a cgroup */
static inline struct cpuset *cgroup_cs(struct cgroup *cont)
{
	return container_of(cgroup_subsys_state(cont, cpuset_subsys_id),
			    struct cpuset, css);
}

/* Retrieve the cpuset for a task */
static inline struct cpuset *task_cs(struct task_struct *task)
{
	return container_of(task_subsys_state(task, cpuset_subsys_id),
			    struct cpuset, css);
}

Tejun Heo's avatar
Tejun Heo committed
120 121 122 123 124 125 126 127 128
static inline struct cpuset *parent_cs(const struct cpuset *cs)
{
	struct cgroup *pcgrp = cs->css.cgroup->parent;

	if (pcgrp)
		return cgroup_cs(pcgrp);
	return NULL;
}

129 130 131 132 133 134 135 136 137 138 139 140 141
#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
142 143
/* bits in struct cpuset flags field */
typedef enum {
Tejun Heo's avatar
Tejun Heo committed
144
	CS_ONLINE,
Linus Torvalds's avatar
Linus Torvalds committed
145 146
	CS_CPU_EXCLUSIVE,
	CS_MEM_EXCLUSIVE,
147
	CS_MEM_HARDWALL,
148
	CS_MEMORY_MIGRATE,
Paul Jackson's avatar
Paul Jackson committed
149
	CS_SCHED_LOAD_BALANCE,
150 151
	CS_SPREAD_PAGE,
	CS_SPREAD_SLAB,
Linus Torvalds's avatar
Linus Torvalds committed
152 153 154
} cpuset_flagbits_t;

/* convenient tests for these bits */
Tejun Heo's avatar
Tejun Heo committed
155 156 157 158 159
static inline bool is_cpuset_online(const struct cpuset *cs)
{
	return test_bit(CS_ONLINE, &cs->flags);
}

Linus Torvalds's avatar
Linus Torvalds committed
160 161
static inline int is_cpu_exclusive(const struct cpuset *cs)
{
162
	return test_bit(CS_CPU_EXCLUSIVE, &cs->flags);
Linus Torvalds's avatar
Linus Torvalds committed
163 164 165 166
}

static inline int is_mem_exclusive(const struct cpuset *cs)
{
167
	return test_bit(CS_MEM_EXCLUSIVE, &cs->flags);
Linus Torvalds's avatar
Linus Torvalds committed
168 169
}

170 171 172 173 174
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
175 176 177 178 179
static inline int is_sched_load_balance(const struct cpuset *cs)
{
	return test_bit(CS_SCHED_LOAD_BALANCE, &cs->flags);
}

180 181
static inline int is_memory_migrate(const struct cpuset *cs)
{
182
	return test_bit(CS_MEMORY_MIGRATE, &cs->flags);
183 184
}

185 186 187 188 189 190 191 192 193 194
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
195
static struct cpuset top_cpuset = {
Tejun Heo's avatar
Tejun Heo committed
196 197
	.flags = ((1 << CS_ONLINE) | (1 << CS_CPU_EXCLUSIVE) |
		  (1 << CS_MEM_EXCLUSIVE)),
Linus Torvalds's avatar
Linus Torvalds committed
198 199
};

200 201 202 203 204 205 206 207 208 209 210 211 212
/**
 * cpuset_for_each_child - traverse online children of a cpuset
 * @child_cs: loop cursor pointing to the current child
 * @pos_cgrp: used for iteration
 * @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.
 */
#define cpuset_for_each_child(child_cs, pos_cgrp, parent_cs)		\
	cgroup_for_each_child((pos_cgrp), (parent_cs)->css.cgroup)	\
		if (is_cpuset_online(((child_cs) = cgroup_cs((pos_cgrp)))))

213 214 215 216 217 218 219 220 221 222 223 224 225 226
/**
 * cpuset_for_each_descendant_pre - pre-order walk of a cpuset's descendants
 * @des_cs: loop cursor pointing to the current descendant
 * @pos_cgrp: used for iteration
 * @root_cs: target cpuset to walk ancestor of
 *
 * Walk @des_cs through the online descendants of @root_cs.  Must be used
 * with RCU read locked.  The caller may modify @pos_cgrp by calling
 * cgroup_rightmost_descendant() to skip subtree.
 */
#define cpuset_for_each_descendant_pre(des_cs, pos_cgrp, root_cs)	\
	cgroup_for_each_descendant_pre((pos_cgrp), (root_cs)->css.cgroup) \
		if (is_cpuset_online(((des_cs) = cgroup_cs((pos_cgrp)))))

Linus Torvalds's avatar
Linus Torvalds committed
227
/*
228 229 230 231 232 233 234 235 236 237 238 239 240 241
 * 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.
242 243
 *
 * Calls to the kernel memory allocator can not be made while holding
244
 * callback_mutex, as that would risk double tripping on callback_mutex
245 246 247
 * from one of the callbacks into the cpuset code from within
 * __alloc_pages().
 *
248
 * If a task is only holding callback_mutex, then it has read-only
249 250
 * access to cpusets.
 *
251 252 253
 * 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.
254
 *
255
 * The cpuset_common_file_read() handlers only hold callback_mutex across
256 257 258
 * small pieces of code, such as when reading out possibly multi-word
 * cpumasks and nodemasks.
 *
259 260
 * 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
261 262
 */

263
static DEFINE_MUTEX(cpuset_mutex);
264
static DEFINE_MUTEX(callback_mutex);
265

266 267 268 269 270 271
/*
 * CPU / memory hotplug is handled asynchronously.
 */
static void cpuset_hotplug_workfn(struct work_struct *work);
static DECLARE_WORK(cpuset_hotplug_work, cpuset_hotplug_workfn);

272 273
static DECLARE_WAIT_QUEUE_HEAD(cpuset_attach_wq);

274 275
/*
 * This is ugly, but preserves the userspace API for existing cpuset
276
 * users. If someone tries to mount the "cpuset" filesystem, we
277 278
 * silently switch it to mount "cgroup" instead
 */
Al Viro's avatar
Al Viro committed
279 280
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
281
{
282
	struct file_system_type *cgroup_fs = get_fs_type("cgroup");
Al Viro's avatar
Al Viro committed
283
	struct dentry *ret = ERR_PTR(-ENODEV);
284 285 286 287
	if (cgroup_fs) {
		char mountopts[] =
			"cpuset,noprefix,"
			"release_agent=/sbin/cpuset_release_agent";
Al Viro's avatar
Al Viro committed
288 289
		ret = cgroup_fs->mount(cgroup_fs, flags,
					   unused_dev_name, mountopts);
290 291 292
		put_filesystem(cgroup_fs);
	}
	return ret;
Linus Torvalds's avatar
Linus Torvalds committed
293 294 295 296
}

static struct file_system_type cpuset_fs_type = {
	.name = "cpuset",
Al Viro's avatar
Al Viro committed
297
	.mount = cpuset_mount,
Linus Torvalds's avatar
Linus Torvalds committed
298 299 300
};

/*
301
 * Return in pmask the portion of a cpusets's cpus_allowed that
Linus Torvalds's avatar
Linus Torvalds committed
302
 * are online.  If none are online, walk up the cpuset hierarchy
303 304
 * 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
305 306
 *
 * One way or another, we guarantee to return some non-empty subset
307
 * of cpu_online_mask.
Linus Torvalds's avatar
Linus Torvalds committed
308
 *
309
 * Call with callback_mutex held.
Linus Torvalds's avatar
Linus Torvalds committed
310
 */
311 312
static void guarantee_online_cpus(const struct cpuset *cs,
				  struct cpumask *pmask)
Linus Torvalds's avatar
Linus Torvalds committed
313
{
314
	while (!cpumask_intersects(cs->cpus_allowed, cpu_online_mask))
Tejun Heo's avatar
Tejun Heo committed
315
		cs = parent_cs(cs);
316
	cpumask_and(pmask, cs->cpus_allowed, cpu_online_mask);
Linus Torvalds's avatar
Linus Torvalds committed
317 318 319 320
}

/*
 * Return in *pmask the portion of a cpusets's mems_allowed that
321 322
 * are online, with memory.  If none are online with memory, walk
 * up the cpuset hierarchy until we find one that does have some
323
 * online mems.  The top cpuset always has some mems online.
Linus Torvalds's avatar
Linus Torvalds committed
324 325
 *
 * One way or another, we guarantee to return some non-empty subset
326
 * of node_states[N_MEMORY].
Linus Torvalds's avatar
Linus Torvalds committed
327
 *
328
 * Call with callback_mutex held.
Linus Torvalds's avatar
Linus Torvalds committed
329 330 331
 */
static void guarantee_online_mems(const struct cpuset *cs, nodemask_t *pmask)
{
332
	while (!nodes_intersects(cs->mems_allowed, node_states[N_MEMORY]))
Tejun Heo's avatar
Tejun Heo committed
333
		cs = parent_cs(cs);
334
	nodes_and(*pmask, cs->mems_allowed, node_states[N_MEMORY]);
Linus Torvalds's avatar
Linus Torvalds committed
335 336
}

337 338 339
/*
 * update task's spread flag if cpuset's page/slab spread flag is set
 *
340
 * Called with callback_mutex/cpuset_mutex held
341 342 343 344 345 346 347 348 349 350 351 352 353 354
 */
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
355 356 357 358 359
/*
 * 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
360
 * are only set if the other's are set.  Call holding cpuset_mutex.
Linus Torvalds's avatar
Linus Torvalds committed
361 362 363 364
 */

static int is_cpuset_subset(const struct cpuset *p, const struct cpuset *q)
{
365
	return	cpumask_subset(p->cpus_allowed, q->cpus_allowed) &&
Linus Torvalds's avatar
Linus Torvalds committed
366 367 368 369 370
		nodes_subset(p->mems_allowed, q->mems_allowed) &&
		is_cpu_exclusive(p) <= is_cpu_exclusive(q) &&
		is_mem_exclusive(p) <= is_mem_exclusive(q);
}

371 372 373 374 375 376
/**
 * alloc_trial_cpuset - allocate a trial cpuset
 * @cs: the cpuset that the trial cpuset duplicates
 */
static struct cpuset *alloc_trial_cpuset(const struct cpuset *cs)
{
377 378 379 380 381 382 383 384 385 386 387 388 389
	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;
390 391 392 393 394 395 396 397
}

/**
 * free_trial_cpuset - free the trial cpuset
 * @trial: the trial cpuset to be freed
 */
static void free_trial_cpuset(struct cpuset *trial)
{
398
	free_cpumask_var(trial->cpus_allowed);
399 400 401
	kfree(trial);
}

Linus Torvalds's avatar
Linus Torvalds committed
402 403 404 405 406 407 408
/*
 * 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
409
 * cpuset_mutex held.
Linus Torvalds's avatar
Linus Torvalds committed
410 411 412 413 414 415 416 417 418 419 420 421 422 423
 *
 * '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.
 */

static int validate_change(const struct cpuset *cur, const struct cpuset *trial)
{
424
	struct cgroup *cont;
Linus Torvalds's avatar
Linus Torvalds committed
425
	struct cpuset *c, *par;
426 427 428
	int ret;

	rcu_read_lock();
Linus Torvalds's avatar
Linus Torvalds committed
429 430

	/* Each of our child cpusets must be a subset of us */
431 432 433 434
	ret = -EBUSY;
	cpuset_for_each_child(c, cont, cur)
		if (!is_cpuset_subset(c, trial))
			goto out;
Linus Torvalds's avatar
Linus Torvalds committed
435 436

	/* Remaining checks don't apply to root cpuset */
437
	ret = 0;
438
	if (cur == &top_cpuset)
439
		goto out;
Linus Torvalds's avatar
Linus Torvalds committed
440

Tejun Heo's avatar
Tejun Heo committed
441
	par = parent_cs(cur);
442

Linus Torvalds's avatar
Linus Torvalds committed
443
	/* We must be a subset of our parent cpuset */
444
	ret = -EACCES;
Linus Torvalds's avatar
Linus Torvalds committed
445
	if (!is_cpuset_subset(trial, par))
446
		goto out;
Linus Torvalds's avatar
Linus Torvalds committed
447

448 449 450 451
	/*
	 * If either I or some sibling (!= me) is exclusive, we can't
	 * overlap
	 */
452 453
	ret = -EINVAL;
	cpuset_for_each_child(c, cont, par) {
Linus Torvalds's avatar
Linus Torvalds committed
454 455
		if ((is_cpu_exclusive(trial) || is_cpu_exclusive(c)) &&
		    c != cur &&
456
		    cpumask_intersects(trial->cpus_allowed, c->cpus_allowed))
457
			goto out;
Linus Torvalds's avatar
Linus Torvalds committed
458 459 460
		if ((is_mem_exclusive(trial) || is_mem_exclusive(c)) &&
		    c != cur &&
		    nodes_intersects(trial->mems_allowed, c->mems_allowed))
461
			goto out;
Linus Torvalds's avatar
Linus Torvalds committed
462 463
	}

464 465 466 467
	/*
	 * Cpusets with tasks - existing or newly being attached - can't
	 * have empty cpus_allowed or mems_allowed.
	 */
468
	ret = -ENOSPC;
469
	if ((cgroup_task_count(cur->css.cgroup) || cur->attach_in_progress) &&
470 471 472
	    (cpumask_empty(trial->cpus_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
{
501 502
	struct cpuset *cp;
	struct cgroup *pos_cgrp;
503

504 505 506 507 508
	rcu_read_lock();
	cpuset_for_each_descendant_pre(cp, pos_cgrp, root_cs) {
		/* skip the whole subtree if @cp doesn't have any CPU */
		if (cpumask_empty(cp->cpus_allowed)) {
			pos_cgrp = cgroup_rightmost_descendant(pos_cgrp);
509
			continue;
510
		}
511 512 513 514

		if (is_sched_load_balance(cp))
			update_domain_attr(dattr, cp);
	}
515
	rcu_read_unlock();
516 517
}

Paul Jackson's avatar
Paul Jackson committed
518
/*
519 520 521 522 523 524 525 526 527
 * 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.
 * The output of this function needs to be passed to kernel/sched.c
 * 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
528
 *
529
 * See "What is sched_load_balance" in Documentation/cgroups/cpusets.txt
Paul Jackson's avatar
Paul Jackson committed
530 531 532 533 534 535 536
 * 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.
 *
537
 * Must be called with cpuset_mutex held.
Paul Jackson's avatar
Paul Jackson committed
538 539
 *
 * The three key local variables below are:
540
 *    q  - a linked-list queue of cpuset pointers, used to implement a
Paul Jackson's avatar
Paul Jackson committed
541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571
 *	   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
 *	   the kernel/sched.c routine partition_sched_domains() in a
 *	   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().
 */
572
static int generate_sched_domains(cpumask_var_t **domains,
573
			struct sched_domain_attr **attributes)
Paul Jackson's avatar
Paul Jackson committed
574 575 576 577 578
{
	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 */
579
	cpumask_var_t *doms;	/* resulting partition; i.e. sched domains */
580
	struct sched_domain_attr *dattr;  /* attributes for custom domains */
581
	int ndoms = 0;		/* number of sched domains in result */
582
	int nslot;		/* next empty doms[] struct cpumask slot */
583
	struct cgroup *pos_cgrp;
Paul Jackson's avatar
Paul Jackson committed
584 585

	doms = NULL;
586
	dattr = NULL;
587
	csa = NULL;
Paul Jackson's avatar
Paul Jackson committed
588 589 590

	/* Special case for the 99% of systems with one, full, sched domain */
	if (is_sched_load_balance(&top_cpuset)) {
591 592
		ndoms = 1;
		doms = alloc_sched_domains(ndoms);
Paul Jackson's avatar
Paul Jackson committed
593
		if (!doms)
594 595
			goto done;

596 597 598
		dattr = kmalloc(sizeof(struct sched_domain_attr), GFP_KERNEL);
		if (dattr) {
			*dattr = SD_ATTR_INIT;
599
			update_domain_attr_tree(dattr, &top_cpuset);
600
		}
601
		cpumask_copy(doms[0], top_cpuset.cpus_allowed);
602 603

		goto done;
Paul Jackson's avatar
Paul Jackson committed
604 605 606 607 608 609 610
	}

	csa = kmalloc(number_of_cpusets * sizeof(cp), GFP_KERNEL);
	if (!csa)
		goto done;
	csn = 0;

611 612
	rcu_read_lock();
	cpuset_for_each_descendant_pre(cp, pos_cgrp, &top_cpuset) {
613
		/*
614 615 616 617 618 619
		 * 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.
620
		 */
621 622
		if (!cpumask_empty(cp->cpus_allowed) &&
		    !is_sched_load_balance(cp))
623
			continue;
624

625 626 627 628 629 630 631
		if (is_sched_load_balance(cp))
			csa[csn++] = cp;

		/* skip @cp's subtree */
		pos_cgrp = cgroup_rightmost_descendant(pos_cgrp);
	}
	rcu_read_unlock();
Paul Jackson's avatar
Paul Jackson committed
632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659

	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;
			}
		}
	}

660 661 662 663
	/*
	 * Now we know how many domains to create.
	 * Convert <csn, csa> to <ndoms, doms> and populate cpu masks.
	 */
664
	doms = alloc_sched_domains(ndoms);
665
	if (!doms)
666 667 668 669 670 671
		goto done;

	/*
	 * The rest of the code, including the scheduler, can deal with
	 * dattr==NULL case. No need to abort if alloc fails.
	 */
672
	dattr = kmalloc(ndoms * sizeof(struct sched_domain_attr), GFP_KERNEL);
Paul Jackson's avatar
Paul Jackson committed
673 674 675

	for (nslot = 0, i = 0; i < csn; i++) {
		struct cpuset *a = csa[i];
676
		struct cpumask *dp;
Paul Jackson's avatar
Paul Jackson committed
677 678
		int apn = a->pn;

679 680 681 682 683
		if (apn < 0) {
			/* Skip completed partitions */
			continue;
		}

684
		dp = doms[nslot];
685 686 687 688 689 690 691 692 693 694

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

699
		cpumask_clear(dp);
700 701 702 703 704 705
		if (dattr)
			*(dattr + nslot) = SD_ATTR_INIT;
		for (j = i; j < csn; j++) {
			struct cpuset *b = csa[j];

			if (apn == b->pn) {
706
				cpumask_or(dp, dp, b->cpus_allowed);
707 708 709 710 711
				if (dattr)
					update_domain_attr_tree(dattr + nslot, b);

				/* Done with this partition */
				b->pn = -1;
Paul Jackson's avatar
Paul Jackson committed
712 713
			}
		}
714
		nslot++;
Paul Jackson's avatar
Paul Jackson committed
715 716 717
	}
	BUG_ON(nslot != ndoms);

718 719 720
done:
	kfree(csa);

721 722 723 724 725 726 727
	/*
	 * Fallback to the default domain if kmalloc() failed.
	 * See comments in partition_sched_domains().
	 */
	if (doms == NULL)
		ndoms = 1;

728 729 730 731 732 733 734 735
	*domains    = doms;
	*attributes = dattr;
	return ndoms;
}

/*
 * Rebuild scheduler domains.
 *
736 737 738 739 740
 * 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.
741
 *
742
 * Call with cpuset_mutex held.  Takes get_online_cpus().
743
 */
744
static void rebuild_sched_domains_locked(void)
745 746
{
	struct sched_domain_attr *attr;
747
	cpumask_var_t *doms;
748 749
	int ndoms;

750
	lockdep_assert_held(&cpuset_mutex);
751
	get_online_cpus();
752

753 754 755 756 757 758 759 760
	/*
	 * 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;

761 762 763 764 765
	/* Generate domain masks and attrs */
	ndoms = generate_sched_domains(&doms, &attr);

	/* Have scheduler rebuild the domains */
	partition_sched_domains(ndoms, doms, attr);
766
out:
767
	put_online_cpus();
768
}
769
#else /* !CONFIG_SMP */
770
static void rebuild_sched_domains_locked(void)
771 772 773
{
}
#endif /* CONFIG_SMP */
Paul Jackson's avatar
Paul Jackson committed
774

775 776
void rebuild_sched_domains(void)
{
777
	mutex_lock(&cpuset_mutex);
778
	rebuild_sched_domains_locked();
779
	mutex_unlock(&cpuset_mutex);
Paul Jackson's avatar
Paul Jackson committed
780 781
}

782 783 784 785 786 787 788 789 790
/**
 * cpuset_change_cpumask - make a task's cpus_allowed the same as its cpuset's
 * @tsk: task to test
 * @scan: struct cgroup_scanner containing the cgroup of the task
 *
 * Called by cgroup_scan_tasks() for each task in a cgroup whose
 * cpus_allowed mask needs to be changed.
 *
 * We don't need to re-check for the cgroup/cpuset membership, since we're
791
 * holding cpuset_mutex at this point.
792
 */
793 794
static void cpuset_change_cpumask(struct task_struct *tsk,
				  struct cgroup_scanner *scan)
795
{
796
	set_cpus_allowed_ptr(tsk, ((cgroup_cs(scan->cg))->cpus_allowed));
797 798
}

799 800 801
/**
 * 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
802
 * @heap: if NULL, defer allocating heap memory to cgroup_scan_tasks()
803
 *
804
 * Called with cpuset_mutex held
805 806 807 808
 *
 * The cgroup_scan_tasks() function will scan all the tasks in a cgroup,
 * calling callback functions for each.
 *
809 810
 * No return value. It's guaranteed that cgroup_scan_tasks() always returns 0
 * if @heap != NULL.
811
 */
812
static void update_tasks_cpumask(struct cpuset *cs, struct ptr_heap *heap)
813 814 815 816
{
	struct cgroup_scanner scan;

	scan.cg = cs->css.cgroup;
817
	scan.test_task = NULL;
818
	scan.process_task = cpuset_change_cpumask;
819 820
	scan.heap = heap;
	cgroup_scan_tasks(&scan);
821 822
}

823 824 825 826 827
/**
 * update_cpumask - update the cpus_allowed mask of a cpuset and all tasks in it
 * @cs: the cpuset to consider
 * @buf: buffer of cpu numbers written to this cpuset
 */
828 829
static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs,
			  const char *buf)
Linus Torvalds's avatar
Linus Torvalds committed
830
{
831
	struct ptr_heap heap;
832 833
	int retval;
	int is_load_balanced;
Linus Torvalds's avatar
Linus Torvalds committed
834

835
	/* top_cpuset.cpus_allowed tracks cpu_online_mask; it's read-only */
836 837 838
	if (cs == &top_cpuset)
		return -EACCES;

839
	/*
840
	 * An empty cpus_allowed is ok only if the cpuset has no tasks.
841 842 843
	 * Since cpulist_parse() fails on an empty mask, we special case
	 * that parsing.  The validate_change() call ensures that cpusets
	 * with tasks have cpus.
844
	 */
845
	if (!*buf) {
846
		cpumask_clear(trialcs->cpus_allowed);
847
	} else {
848
		retval = cpulist_parse(buf, trialcs->cpus_allowed);
849 850
		if (retval < 0)
			return retval;
851

852
		if (!cpumask_subset(trialcs->cpus_allowed, cpu_active_mask))
853
			return -EINVAL;
854
	}
Paul Jackson's avatar
Paul Jackson committed
855

Paul Menage's avatar
Paul Menage committed
856
	/* Nothing to do if the cpus didn't change */
857
	if (cpumask_equal(cs->cpus_allowed, trialcs->cpus_allowed))
Paul Menage's avatar
Paul Menage committed
858
		return 0;
859

860 861 862 863
	retval = validate_change(cs, trialcs);
	if (retval < 0)
		return retval;

864 865 866 867
	retval = heap_init(&heap, PAGE_SIZE, GFP_KERNEL, NULL);
	if (retval)
		return retval;

868
	is_load_balanced = is_sched_load_balance(trialcs);
Paul Jackson's avatar
Paul Jackson committed
869

870
	mutex_lock(&callback_mutex);
871
	cpumask_copy(cs->cpus_allowed, trialcs->cpus_allowed);
872
	mutex_unlock(&callback_mutex);
Paul Jackson's avatar
Paul Jackson committed
873

Paul Menage's avatar
Paul Menage committed
874 875
	/*
	 * Scan tasks in the cpuset, and update the cpumasks of any
876
	 * that need an update.
Paul Menage's avatar
Paul Menage committed
877
	 */
878 879 880
	update_tasks_cpumask(cs, &heap);

	heap_free(&heap);
881

Paul Menage's avatar
Paul Menage committed
882
	if (is_load_balanced)
883
		rebuild_sched_domains_locked();
884
	return 0;
Linus Torvalds's avatar
Linus Torvalds committed
885 886
}

887 888 889 890 891 892 893 894
/*
 * 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.
 *
895
 *    Call holding cpuset_mutex, so current's cpuset won't change
896
 *    during this call, as manage_mutex holds off any cpuset_attach()
897 898
 *    calls.  Therefore we don't need to take task_lock around the
 *    call to guarantee_online_mems(), as we know no one is changing
899
 *    our task's cpuset.
900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915
 *
 *    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);

916
	guarantee_online_mems(task_cs(tsk),&tsk->mems_allowed);
917 918
}

919
/*
920 921 922 923 924 925 926 927 928 929 930
 * 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)
{
931
	bool need_loop;
932

933 934 935 936 937 938 939 940 941 942
	/*
	 * 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);
943 944 945 946 947 948 949 950
	/*
	 * Determine if a loop is necessary if another thread is doing
	 * get_mems_allowed().  If at least one node remains unchanged and
	 * 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);
951

952 953
	if (need_loop)
		write_seqcount_begin(&tsk->mems_allowed_seq);
954

955 956
	nodes_or(tsk->mems_allowed, tsk->mems_allowed, *newmems);
	mpol_rebind_task(tsk, newmems, MPOL_REBIND_STEP1);
957 958

	mpol_rebind_task(tsk, newmems, MPOL_REBIND_STEP2);
959
	tsk->mems_allowed = *newmems;
960 961 962 963

	if (need_loop)
		write_seqcount_end(&tsk->mems_allowed_seq);

964
	task_unlock(tsk);
965 966 967 968 969
}

/*
 * Update task's mems_allowed and rebind its mempolicy and vmas' mempolicy
 * of it to cpuset's new mems_allowed, and migrate pages to new nodes if
970
 * memory_migrate flag is set. Called with cpuset_mutex held.
971 972 973 974 975 976 977 978
 */
static void cpuset_change_nodemask(struct task_struct *p,
				   struct cgroup_scanner *scan)
{
	struct mm_struct *mm;
	struct cpuset *cs;
	int migrate;
	const nodemask_t *oldmem = scan->data;
979
	static nodemask_t newmems;	/* protected by cpuset_mutex */
980 981

	cs = cgroup_cs(scan->cg);
982
	guarantee_online_mems(cs, &newmems);
983

984
	cpuset_change_task_nodemask(p, &newmems);
985

986 987 988 989 990 991 992 993 994 995 996 997
	mm = get_task_mm(p);
	if (!mm)
		return;

	migrate = is_memory_migrate(cs);

	mpol_rebind_mm(mm, &cs->mems_allowed);
	if (migrate)
		cpuset_migrate_mm(mm, oldmem, &cs->mems_allowed);
	mmput(mm);
}

998 999
static void *cpuset_being_rebound;

1000 1001 1002 1003
/**
 * 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
 * @oldmem: old mems_allowed of cpuset cs
1004
 * @heap: if NULL, defer allocating heap memory to cgroup_scan_tasks()
1005
 *
1006
 * Called with cpuset_mutex held
1007 1008
 * No return value. It's guaranteed that cgroup_scan_tasks() always returns 0
 * if @heap != NULL.
1009
 */
1010 1011
static void update_tasks_nodemask(struct cpuset *cs, const nodemask_t *oldmem,
				 struct ptr_heap *heap)
Linus Torvalds's avatar
Linus Torvalds committed
1012
{
1013
	struct cgroup_scanner scan;
1014

1015
	cpuset_being_rebound = cs;		/* causes mpol_dup() rebind */
1016

1017 1018 1019
	scan.cg = cs->css.cgroup;
	scan.test_task = NULL;
	scan.process_task = cpuset_change_nodemask;
1020
	scan.heap = heap;
1021
	scan.data = (nodemask_t *)oldmem;
1022 1023

	/*
1024 1025 1026 1027
	 * The mpol_rebind_mm() call takes mmap_sem, which we couldn't
	 * take while holding tasklist_lock.  Forks can happen - the
	 * mpol_dup() cpuset_being_rebound check will catch such forks,
	 * and rebind their vma mempolicies too.  Because we still hold
1028
	 * the global cpuset_mutex, we know that no other rebind effort
1029
	 * will be contending for the global variable cpuset_being_rebound.
1030
	 * It's ok if we rebind the same mm twice; mpol_rebind_mm()
1031
	 * is idempotent.  Also migrate pages in each mm to new nodes.
1032
	 */
1033
	cgroup_scan_tasks(&scan);
1034

1035
	/* We're done rebinding vmas to this cpuset's new mems_allowed. */
1036
	cpuset_being_rebound = NULL;
Linus Torvalds's avatar
Linus Torvalds committed
1037 1038
}

1039 1040 1041
/*
 * Handle user request to change the 'mems' memory placement
 * of a cpuset.  Needs to validate the request, update the
1042 1043 1044 1045
 * cpusets mems_allowed, and for each task in the cpuset,
 * update mems_allowed and rebind task's mempolicy and any vma
 * mempolicies and if the cpuset is marked 'memory_migrate',
 * migrate the tasks pages to the new memory.
1046
 *
1047
 * Call with cpuset_mutex held.  May take callback_mutex during call.
1048 1049 1050 1051
 * Will take tasklist_lock, scan tasklist for tasks in cpuset cs,
 * lock each such tasks mm->mmap_sem, scan its vma's and rebind
 * their mempolicies to the cpusets new mems_allowed.
 */
1052 1053
static int update_nodemask(struct cpuset *cs, struct cpuset *trialcs,
			   const char *buf)
1054
{
1055
	NODEMASK_ALLOC(nodemask_t, oldmem, GFP_KERNEL);
1056
	int retval;
1057
	struct ptr_heap heap;
1058

1059 1060 1061
	if (!oldmem)
		return -ENOMEM;

1062
	/*
1063
	 * top_cpuset.mems_allowed tracks node_stats[N_MEMORY];
1064 1065
	 * it's read-only
	 */
1066 1067 1068 1069
	if (cs == &top_cpuset) {
		retval = -EACCES;
		goto done;
	}
1070 1071 1072 1073 1074 1075 1076 1077

	/*
	 * An empty mems_allowed is ok iff there are no tasks in the cpuset.
	 * Since nodelist_parse() fails on an empty mask, we special case
	 * that parsing.  The validate_change() call ensures that cpusets
	 * with tasks have memory.
	 */
	if (!*buf) {
1078
		nodes_clear(trialcs->mems_allowed);
1079
	} else {
1080
		retval = nodelist_parse(buf, trialcs->mems_allowed);
1081 1082 1083
		if (retval < 0)
			goto done;

1084
		if (!nodes_subset(trialcs->mems_allowed,