cpuset.c 74.1 KB
Newer Older
Linus Torvalds's avatar
Linus Torvalds committed
1 2 3 4 5 6
/*
 *  kernel/cpuset.c
 *
 *  Processor and Memory placement constraints for sets of tasks.
 *
 *  Copyright (C) 2003 BULL SA.
Paul Jackson's avatar
Paul Jackson committed
7
 *  Copyright (C) 2004-2007 Silicon Graphics, Inc.
8
 *  Copyright (C) 2006 Google, Inc
Linus Torvalds's avatar
Linus Torvalds committed
9 10 11 12
 *
 *  Portions derived from Patrick Mochel's sysfs code.
 *  sysfs is Copyright (c) 2001-3 Patrick Mochel
 *
13
 *  2003-10-10 Written by Simon Derr.
Linus Torvalds's avatar
Linus Torvalds committed
14
 *  2003-10-22 Updates by Stephen Hemminger.
15
 *  2004 May-July Rework by Paul Jackson.
16
 *  2006 Rework by Paul Menage to use generic cgroups
17 18
 *  2008 Rework of the scheduler domains and CPU hotplug handling
 *       by Max Krasnyansky
Linus Torvalds's avatar
Linus Torvalds committed
19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36
 *
 *  This file is subject to the terms and conditions of the GNU General Public
 *  License.  See the file COPYING in the main directory of the Linux
 *  distribution for more details.
 */

#include <linux/cpu.h>
#include <linux/cpumask.h>
#include <linux/cpuset.h>
#include <linux/err.h>
#include <linux/errno.h>
#include <linux/file.h>
#include <linux/fs.h>
#include <linux/init.h>
#include <linux/interrupt.h>
#include <linux/kernel.h>
#include <linux/kmod.h>
#include <linux/list.h>
37
#include <linux/mempolicy.h>
Linus Torvalds's avatar
Linus Torvalds committed
38
#include <linux/mm.h>
39
#include <linux/memory.h>
40
#include <linux/export.h>
Linus Torvalds's avatar
Linus Torvalds committed
41 42 43 44
#include <linux/mount.h>
#include <linux/namei.h>
#include <linux/pagemap.h>
#include <linux/proc_fs.h>
45
#include <linux/rcupdate.h>
Linus Torvalds's avatar
Linus Torvalds committed
46 47
#include <linux/sched.h>
#include <linux/seq_file.h>
48
#include <linux/security.h>
Linus Torvalds's avatar
Linus Torvalds committed
49 50 51 52 53 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>
Linus Torvalds's avatar
Linus Torvalds committed
62

63 64 65 66 67
/*
 * Tracks how many cpusets are currently defined in system.
 * When there is only one cpuset (the root cpuset) we can
 * short circuit some hooks.
 */
68
int number_of_cpusets __read_mostly;
69

70
/* Forward declare cgroup structures */
71 72 73
struct cgroup_subsys cpuset_subsys;
struct cpuset;

74 75 76 77 78 79 80 81 82
/* 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
83
struct cpuset {
84 85
	struct cgroup_subsys_state css;

Linus Torvalds's avatar
Linus Torvalds committed
86
	unsigned long flags;		/* "unsigned long" so bitops work */
87
	cpumask_var_t cpus_allowed;	/* CPUs allowed to tasks in cpuset */
Linus Torvalds's avatar
Linus Torvalds committed
88 89
	nodemask_t mems_allowed;	/* Memory Nodes allowed to tasks */

90
	struct fmeter fmeter;		/* memory_pressure filter */
Paul Jackson's avatar
Paul Jackson committed
91

92 93 94 95 96 97
	/*
	 * 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
98 99
	/* partition number for rebuild_sched_domains() */
	int pn;
100

101 102 103
	/* for custom sched domain */
	int relax_domain_level;

104
	struct work_struct hotplug_work;
Linus Torvalds's avatar
Linus Torvalds committed
105 106
};

107 108 109 110 111 112 113 114 115 116 117 118 119 120
/* 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
121 122 123 124 125 126 127 128 129
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;
}

130 131 132 133 134 135 136 137 138 139 140 141 142
#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
143 144
/* bits in struct cpuset flags field */
typedef enum {
Tejun Heo's avatar
Tejun Heo committed
145
	CS_ONLINE,
Linus Torvalds's avatar
Linus Torvalds committed
146 147
	CS_CPU_EXCLUSIVE,
	CS_MEM_EXCLUSIVE,
148
	CS_MEM_HARDWALL,
149
	CS_MEMORY_MIGRATE,
Paul Jackson's avatar
Paul Jackson committed
150
	CS_SCHED_LOAD_BALANCE,
151 152
	CS_SPREAD_PAGE,
	CS_SPREAD_SLAB,
Linus Torvalds's avatar
Linus Torvalds committed
153 154 155
} cpuset_flagbits_t;

/* convenient tests for these bits */
Tejun Heo's avatar
Tejun Heo committed
156 157 158 159 160
static inline bool is_cpuset_online(const struct cpuset *cs)
{
	return test_bit(CS_ONLINE, &cs->flags);
}

Linus Torvalds's avatar
Linus Torvalds committed
161 162
static inline int is_cpu_exclusive(const struct cpuset *cs)
{
163
	return test_bit(CS_CPU_EXCLUSIVE, &cs->flags);
Linus Torvalds's avatar
Linus Torvalds committed
164 165 166 167
}

static inline int is_mem_exclusive(const struct cpuset *cs)
{
168
	return test_bit(CS_MEM_EXCLUSIVE, &cs->flags);
Linus Torvalds's avatar
Linus Torvalds committed
169 170
}

171 172 173 174 175
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
176 177 178 179 180
static inline int is_sched_load_balance(const struct cpuset *cs)
{
	return test_bit(CS_SCHED_LOAD_BALANCE, &cs->flags);
}

181 182
static inline int is_memory_migrate(const struct cpuset *cs)
{
183
	return test_bit(CS_MEMORY_MIGRATE, &cs->flags);
184 185
}

186 187 188 189 190 191 192 193 194 195
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
196
static struct cpuset top_cpuset = {
Tejun Heo's avatar
Tejun Heo committed
197 198
	.flags = ((1 << CS_ONLINE) | (1 << CS_CPU_EXCLUSIVE) |
		  (1 << CS_MEM_EXCLUSIVE)),
Linus Torvalds's avatar
Linus Torvalds committed
199 200
};

201 202 203 204 205 206 207 208 209 210 211 212 213
/**
 * 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)))))

214 215 216 217 218 219 220 221 222 223 224 225 226 227
/**
 * 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
228
/*
229 230 231 232 233 234 235 236 237 238 239 240 241 242
 * 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.
243 244
 *
 * Calls to the kernel memory allocator can not be made while holding
245
 * callback_mutex, as that would risk double tripping on callback_mutex
246 247 248
 * from one of the callbacks into the cpuset code from within
 * __alloc_pages().
 *
249
 * If a task is only holding callback_mutex, then it has read-only
250 251
 * access to cpusets.
 *
252 253 254
 * 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.
255
 *
256
 * The cpuset_common_file_read() handlers only hold callback_mutex across
257 258 259
 * small pieces of code, such as when reading out possibly multi-word
 * cpumasks and nodemasks.
 *
260 261
 * 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
262 263
 */

264
static DEFINE_MUTEX(cpuset_mutex);
265
static DEFINE_MUTEX(callback_mutex);
266

267 268 269
/*
 * CPU / memory hotplug is handled asynchronously.
 */
270 271
static struct workqueue_struct *cpuset_propagate_hotplug_wq;

272
static void cpuset_hotplug_workfn(struct work_struct *work);
273
static void cpuset_propagate_hotplug_workfn(struct work_struct *work);
274
static void schedule_cpuset_propagate_hotplug(struct cpuset *cs);
275 276 277

static DECLARE_WORK(cpuset_hotplug_work, cpuset_hotplug_workfn);

278 279
/*
 * This is ugly, but preserves the userspace API for existing cpuset
280
 * users. If someone tries to mount the "cpuset" filesystem, we
281 282
 * silently switch it to mount "cgroup" instead
 */
Al Viro's avatar
Al Viro committed
283 284
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
285
{
286
	struct file_system_type *cgroup_fs = get_fs_type("cgroup");
Al Viro's avatar
Al Viro committed
287
	struct dentry *ret = ERR_PTR(-ENODEV);
288 289 290 291
	if (cgroup_fs) {
		char mountopts[] =
			"cpuset,noprefix,"
			"release_agent=/sbin/cpuset_release_agent";
Al Viro's avatar
Al Viro committed
292 293
		ret = cgroup_fs->mount(cgroup_fs, flags,
					   unused_dev_name, mountopts);
294 295 296
		put_filesystem(cgroup_fs);
	}
	return ret;
Linus Torvalds's avatar
Linus Torvalds committed
297 298 299 300
}

static struct file_system_type cpuset_fs_type = {
	.name = "cpuset",
Al Viro's avatar
Al Viro committed
301
	.mount = cpuset_mount,
Linus Torvalds's avatar
Linus Torvalds committed
302 303 304
};

/*
305
 * Return in pmask the portion of a cpusets's cpus_allowed that
Linus Torvalds's avatar
Linus Torvalds committed
306
 * are online.  If none are online, walk up the cpuset hierarchy
307 308
 * 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
309 310
 *
 * One way or another, we guarantee to return some non-empty subset
311
 * of cpu_online_mask.
Linus Torvalds's avatar
Linus Torvalds committed
312
 *
313
 * Call with callback_mutex held.
Linus Torvalds's avatar
Linus Torvalds committed
314
 */
315 316
static void guarantee_online_cpus(const struct cpuset *cs,
				  struct cpumask *pmask)
Linus Torvalds's avatar
Linus Torvalds committed
317
{
318
	while (!cpumask_intersects(cs->cpus_allowed, cpu_online_mask))
Tejun Heo's avatar
Tejun Heo committed
319
		cs = parent_cs(cs);
320
	cpumask_and(pmask, cs->cpus_allowed, cpu_online_mask);
Linus Torvalds's avatar
Linus Torvalds committed
321 322 323 324
}

/*
 * Return in *pmask the portion of a cpusets's mems_allowed that
325 326
 * are online, with memory.  If none are online with memory, walk
 * up the cpuset hierarchy until we find one that does have some
327
 * online mems.  The top cpuset always has some mems online.
Linus Torvalds's avatar
Linus Torvalds committed
328 329
 *
 * One way or another, we guarantee to return some non-empty subset
330
 * of node_states[N_MEMORY].
Linus Torvalds's avatar
Linus Torvalds committed
331
 *
332
 * Call with callback_mutex held.
Linus Torvalds's avatar
Linus Torvalds committed
333 334 335
 */
static void guarantee_online_mems(const struct cpuset *cs, nodemask_t *pmask)
{
336
	while (!nodes_intersects(cs->mems_allowed, node_states[N_MEMORY]))
Tejun Heo's avatar
Tejun Heo committed
337
		cs = parent_cs(cs);
338
	nodes_and(*pmask, cs->mems_allowed, node_states[N_MEMORY]);
Linus Torvalds's avatar
Linus Torvalds committed
339 340
}

341 342 343
/*
 * update task's spread flag if cpuset's page/slab spread flag is set
 *
344
 * Called with callback_mutex/cpuset_mutex held
345 346 347 348 349 350 351 352 353 354 355 356 357 358
 */
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
359 360 361 362 363
/*
 * 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
364
 * are only set if the other's are set.  Call holding cpuset_mutex.
Linus Torvalds's avatar
Linus Torvalds committed
365 366 367 368
 */

static int is_cpuset_subset(const struct cpuset *p, const struct cpuset *q)
{
369
	return	cpumask_subset(p->cpus_allowed, q->cpus_allowed) &&
Linus Torvalds's avatar
Linus Torvalds committed
370 371 372 373 374
		nodes_subset(p->mems_allowed, q->mems_allowed) &&
		is_cpu_exclusive(p) <= is_cpu_exclusive(q) &&
		is_mem_exclusive(p) <= is_mem_exclusive(q);
}

375 376 377 378 379 380
/**
 * 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)
{
381 382 383 384 385 386 387 388 389 390 391 392 393
	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;
394 395 396 397 398 399 400 401
}

/**
 * free_trial_cpuset - free the trial cpuset
 * @trial: the trial cpuset to be freed
 */
static void free_trial_cpuset(struct cpuset *trial)
{
402
	free_cpumask_var(trial->cpus_allowed);
403 404 405
	kfree(trial);
}

Linus Torvalds's avatar
Linus Torvalds committed
406 407 408 409 410 411 412
/*
 * 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
413
 * cpuset_mutex held.
Linus Torvalds's avatar
Linus Torvalds committed
414 415 416 417 418 419 420 421 422 423 424 425 426 427
 *
 * '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)
{
428
	struct cgroup *cont;
Linus Torvalds's avatar
Linus Torvalds committed
429
	struct cpuset *c, *par;
430 431 432
	int ret;

	rcu_read_lock();
Linus Torvalds's avatar
Linus Torvalds committed
433 434

	/* Each of our child cpusets must be a subset of us */
435 436 437 438
	ret = -EBUSY;
	cpuset_for_each_child(c, cont, cur)
		if (!is_cpuset_subset(c, trial))
			goto out;
Linus Torvalds's avatar
Linus Torvalds committed
439 440

	/* Remaining checks don't apply to root cpuset */
441
	ret = 0;
442
	if (cur == &top_cpuset)
443
		goto out;
Linus Torvalds's avatar
Linus Torvalds committed
444

Tejun Heo's avatar
Tejun Heo committed
445
	par = parent_cs(cur);
446

Linus Torvalds's avatar
Linus Torvalds committed
447
	/* We must be a subset of our parent cpuset */
448
	ret = -EACCES;
Linus Torvalds's avatar
Linus Torvalds committed
449
	if (!is_cpuset_subset(trial, par))
450
		goto out;
Linus Torvalds's avatar
Linus Torvalds committed
451

452 453 454 455
	/*
	 * If either I or some sibling (!= me) is exclusive, we can't
	 * overlap
	 */
456 457
	ret = -EINVAL;
	cpuset_for_each_child(c, cont, par) {
Linus Torvalds's avatar
Linus Torvalds committed
458 459
		if ((is_cpu_exclusive(trial) || is_cpu_exclusive(c)) &&
		    c != cur &&
460
		    cpumask_intersects(trial->cpus_allowed, c->cpus_allowed))
461
			goto out;
Linus Torvalds's avatar
Linus Torvalds committed
462 463 464
		if ((is_mem_exclusive(trial) || is_mem_exclusive(c)) &&
		    c != cur &&
		    nodes_intersects(trial->mems_allowed, c->mems_allowed))
465
			goto out;
Linus Torvalds's avatar
Linus Torvalds committed
466 467
	}

468 469 470 471
	/*
	 * Cpusets with tasks - existing or newly being attached - can't
	 * have empty cpus_allowed or mems_allowed.
	 */
472
	ret = -ENOSPC;
473
	if ((cgroup_task_count(cur->css.cgroup) || cur->attach_in_progress) &&
474 475 476
	    (cpumask_empty(trial->cpus_allowed) ||
	     nodes_empty(trial->mems_allowed)))
		goto out;
477

478 479 480 481
	ret = 0;
out:
	rcu_read_unlock();
	return ret;
Linus Torvalds's avatar
Linus Torvalds committed
482 483
}

484
#ifdef CONFIG_SMP
Paul Jackson's avatar
Paul Jackson committed
485
/*
486
 * Helper routine for generate_sched_domains().
Paul Jackson's avatar
Paul Jackson committed
487 488 489 490
 * Do cpusets a, b have overlapping cpus_allowed masks?
 */
static int cpusets_overlap(struct cpuset *a, struct cpuset *b)
{
491
	return cpumask_intersects(a->cpus_allowed, b->cpus_allowed);
Paul Jackson's avatar
Paul Jackson committed
492 493
}

494 495 496 497 498 499 500 501
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;
}

502 503
static void update_domain_attr_tree(struct sched_domain_attr *dattr,
				    struct cpuset *root_cs)
504
{
505 506
	struct cpuset *cp;
	struct cgroup *pos_cgrp;
507

508 509 510 511 512
	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);
513
			continue;
514
		}
515 516 517 518

		if (is_sched_load_balance(cp))
			update_domain_attr(dattr, cp);
	}
519
	rcu_read_unlock();
520 521
}

Paul Jackson's avatar
Paul Jackson committed
522
/*
523 524 525 526 527 528 529 530 531
 * 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
532
 *
533
 * See "What is sched_load_balance" in Documentation/cgroups/cpusets.txt
Paul Jackson's avatar
Paul Jackson committed
534 535 536 537 538 539 540
 * 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.
 *
541
 * Must be called with cpuset_mutex held.
Paul Jackson's avatar
Paul Jackson committed
542 543
 *
 * The three key local variables below are:
544
 *    q  - a linked-list queue of cpuset pointers, used to implement a
Paul Jackson's avatar
Paul Jackson committed
545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575
 *	   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().
 */
576
static int generate_sched_domains(cpumask_var_t **domains,
577
			struct sched_domain_attr **attributes)
Paul Jackson's avatar
Paul Jackson committed
578 579 580 581 582
{
	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 */
583
	cpumask_var_t *doms;	/* resulting partition; i.e. sched domains */
584
	struct sched_domain_attr *dattr;  /* attributes for custom domains */
585
	int ndoms = 0;		/* number of sched domains in result */
586
	int nslot;		/* next empty doms[] struct cpumask slot */
587
	struct cgroup *pos_cgrp;
Paul Jackson's avatar
Paul Jackson committed
588 589

	doms = NULL;
590
	dattr = NULL;
591
	csa = NULL;
Paul Jackson's avatar
Paul Jackson committed
592 593 594

	/* Special case for the 99% of systems with one, full, sched domain */
	if (is_sched_load_balance(&top_cpuset)) {
595 596
		ndoms = 1;
		doms = alloc_sched_domains(ndoms);
Paul Jackson's avatar
Paul Jackson committed
597
		if (!doms)
598 599
			goto done;

600 601 602
		dattr = kmalloc(sizeof(struct sched_domain_attr), GFP_KERNEL);
		if (dattr) {
			*dattr = SD_ATTR_INIT;
603
			update_domain_attr_tree(dattr, &top_cpuset);
604
		}
605
		cpumask_copy(doms[0], top_cpuset.cpus_allowed);
606 607

		goto done;
Paul Jackson's avatar
Paul Jackson committed
608 609 610 611 612 613 614
	}

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

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

629 630 631 632 633 634 635
		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
636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663

	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;
			}
		}
	}

664 665 666 667
	/*
	 * Now we know how many domains to create.
	 * Convert <csn, csa> to <ndoms, doms> and populate cpu masks.
	 */
668
	doms = alloc_sched_domains(ndoms);
669
	if (!doms)
670 671 672 673 674 675
		goto done;

	/*
	 * The rest of the code, including the scheduler, can deal with
	 * dattr==NULL case. No need to abort if alloc fails.
	 */
676
	dattr = kmalloc(ndoms * sizeof(struct sched_domain_attr), GFP_KERNEL);
Paul Jackson's avatar
Paul Jackson committed
677 678 679

	for (nslot = 0, i = 0; i < csn; i++) {
		struct cpuset *a = csa[i];
680
		struct cpumask *dp;
Paul Jackson's avatar
Paul Jackson committed
681 682
		int apn = a->pn;

683 684 685 686 687
		if (apn < 0) {
			/* Skip completed partitions */
			continue;
		}

688
		dp = doms[nslot];
689 690 691 692 693 694 695 696 697 698

		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
699
			}
700 701
			continue;
		}
Paul Jackson's avatar
Paul Jackson committed
702

703
		cpumask_clear(dp);
704 705 706 707 708 709
		if (dattr)
			*(dattr + nslot) = SD_ATTR_INIT;
		for (j = i; j < csn; j++) {
			struct cpuset *b = csa[j];

			if (apn == b->pn) {
710
				cpumask_or(dp, dp, b->cpus_allowed);
711 712 713 714 715
				if (dattr)
					update_domain_attr_tree(dattr + nslot, b);

				/* Done with this partition */
				b->pn = -1;
Paul Jackson's avatar
Paul Jackson committed
716 717
			}
		}
718
		nslot++;
Paul Jackson's avatar
Paul Jackson committed
719 720 721
	}
	BUG_ON(nslot != ndoms);

722 723 724
done:
	kfree(csa);

725 726 727 728 729 730 731
	/*
	 * Fallback to the default domain if kmalloc() failed.
	 * See comments in partition_sched_domains().
	 */
	if (doms == NULL)
		ndoms = 1;

732 733 734 735 736 737 738 739
	*domains    = doms;
	*attributes = dattr;
	return ndoms;
}

/*
 * Rebuild scheduler domains.
 *
740 741 742 743 744
 * 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.
745
 *
746
 * Call with cpuset_mutex held.  Takes get_online_cpus().
747
 */
748
static void rebuild_sched_domains_locked(void)
749 750
{
	struct sched_domain_attr *attr;
751
	cpumask_var_t *doms;
752 753
	int ndoms;

754
	lockdep_assert_held(&cpuset_mutex);
755
	get_online_cpus();
756

757 758 759 760 761 762 763 764
	/*
	 * 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;

765 766 767 768 769
	/* Generate domain masks and attrs */
	ndoms = generate_sched_domains(&doms, &attr);

	/* Have scheduler rebuild the domains */
	partition_sched_domains(ndoms, doms, attr);
770
out:
771
	put_online_cpus();
772
}
773
#else /* !CONFIG_SMP */
774
static void rebuild_sched_domains_locked(void)
775 776 777
{
}
#endif /* CONFIG_SMP */
Paul Jackson's avatar
Paul Jackson committed
778

779 780
void rebuild_sched_domains(void)
{
781
	mutex_lock(&cpuset_mutex);
782
	rebuild_sched_domains_locked();
783
	mutex_unlock(&cpuset_mutex);
Paul Jackson's avatar
Paul Jackson committed
784 785
}

786 787 788 789 790 791 792 793 794
/**
 * 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
795
 * holding cpuset_mutex at this point.
796
 */
797 798
static void cpuset_change_cpumask(struct task_struct *tsk,
				  struct cgroup_scanner *scan)
799
{
800
	set_cpus_allowed_ptr(tsk, ((cgroup_cs(scan->cg))->cpus_allowed));
801 802
}

803 804 805
/**
 * 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
806
 * @heap: if NULL, defer allocating heap memory to cgroup_scan_tasks()
807
 *
808
 * Called with cpuset_mutex held
809 810 811 812
 *
 * The cgroup_scan_tasks() function will scan all the tasks in a cgroup,
 * calling callback functions for each.
 *
813 814
 * No return value. It's guaranteed that cgroup_scan_tasks() always returns 0
 * if @heap != NULL.
815
 */
816
static void update_tasks_cpumask(struct cpuset *cs, struct ptr_heap *heap)
817 818 819 820
{
	struct cgroup_scanner scan;

	scan.cg = cs->css.cgroup;
821
	scan.test_task = NULL;
822
	scan.process_task = cpuset_change_cpumask;
823 824
	scan.heap = heap;
	cgroup_scan_tasks(&scan);
825 826
}

827 828 829 830 831
/**
 * 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
 */
832 833
static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs,
			  const char *buf)
Linus Torvalds's avatar
Linus Torvalds committed
834
{
835
	struct ptr_heap heap;
836 837
	int retval;
	int is_load_balanced;
Linus Torvalds's avatar
Linus Torvalds committed
838

839
	/* top_cpuset.cpus_allowed tracks cpu_online_mask; it's read-only */
840 841 842
	if (cs == &top_cpuset)
		return -EACCES;

843
	/*
844
	 * An empty cpus_allowed is ok only if the cpuset has no tasks.
845 846 847
	 * Since cpulist_parse() fails on an empty mask, we special case
	 * that parsing.  The validate_change() call ensures that cpusets
	 * with tasks have cpus.
848
	 */
849
	if (!*buf) {
850
		cpumask_clear(trialcs->cpus_allowed);
851
	} else {
852
		retval = cpulist_parse(buf, trialcs->cpus_allowed);
853 854
		if (retval < 0)
			return retval;
855

856
		if (!cpumask_subset(trialcs->cpus_allowed, cpu_active_mask))
857
			return -EINVAL;
858
	}
Paul Jackson's avatar
Paul Jackson committed
859

Paul Menage's avatar
Paul Menage committed
860
	/* Nothing to do if the cpus didn't change */
861
	if (cpumask_equal(cs->cpus_allowed, trialcs->cpus_allowed))
Paul Menage's avatar
Paul Menage committed
862
		return 0;
863

864 865 866 867
	retval = validate_change(cs, trialcs);
	if (retval < 0)
		return retval;

868 869 870 871
	retval = heap_init(&heap, PAGE_SIZE, GFP_KERNEL, NULL);
	if (retval)
		return retval;

872
	is_load_balanced = is_sched_load_balance(trialcs);
Paul Jackson's avatar
Paul Jackson committed
873

874
	mutex_lock(&callback_mutex);
875
	cpumask_copy(cs->cpus_allowed, trialcs->cpus_allowed);
876
	mutex_unlock(&callback_mutex);
Paul Jackson's avatar
Paul Jackson committed
877

Paul Menage's avatar
Paul Menage committed
878 879
	/*
	 * Scan tasks in the cpuset, and update the cpumasks of any
880
	 * that need an update.
Paul Menage's avatar
Paul Menage committed
881
	 */
882 883 884
	update_tasks_cpumask(cs, &heap);

	heap_free(&heap);
885

Paul Menage's avatar
Paul Menage committed
886
	if (is_load_balanced)
887
		rebuild_sched_domains_locked();
888
	return 0;
Linus Torvalds's avatar
Linus Torvalds committed
889 890
}

891 892 893 894 895 896 897 898
/*
 * 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.
 *
899
 *    Call holding cpuset_mutex, so current's cpuset won't change
900
 *    during this call, as manage_mutex holds off any cpuset_attach()
901 902
 *    calls.  Therefore we don't need to take task_lock around the
 *    call to guarantee_online_mems(), as we know no one is changing
903
 *    our task's cpuset.
904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919
 *
 *    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);

920
	guarantee_online_mems(task_cs(tsk),&tsk->mems_allowed);
921 922
}

923
/*
924 925 926 927 928 929 930 931 932 933 934
 * 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)
{
935
	bool need_loop;
936

937 938 939 940 941 942 943 944 945 946
	/*
	 * 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);
947 948 949 950 951 952 953 954
	/*
	 * 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);
955

956 957
	if (need_loop)
		write_seqcount_begin(&tsk->mems_allowed_seq);
958

959 960
	nodes_or(tsk->mems_allowed, tsk->mems_allowed, *newmems);
	mpol_rebind_task(tsk, newmems, MPOL_REBIND_STEP1);
961 962

	mpol_rebind_task(tsk, newmems, MPOL_REBIND_STEP2);
963
	tsk->mems_allowed = *newmems;
964 965 966 967

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

968
	task_unlock(tsk);
969 970 971 972 973
}

/*
 * 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
974
 * memory_migrate flag is set. Called with cpuset_mutex held.
975 976 977 978 979 980 981 982
 */
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;
983
	static nodemask_t newmems;	/* protected by cpuset_mutex */
984 985

	cs = cgroup_cs(scan->cg);
986
	guarantee_online_mems(cs, &newmems);
987

988
	cpuset_change_task_nodemask(p, &newmems);
989

990 991 992 993 994 995 996 997 998 999 1000 1001
	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);
}

1002 1003
static void *cpuset_being_rebound;

1004 1005 1006 1007
/**
 * 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
1008
 * @heap: if NULL, defer allocating heap memory to cgroup_scan_tasks()
1009
 *
1010
 * Called with cpuset_mutex held
1011 1012
 * No return value. It's guaranteed that cgroup_scan_tasks() always returns 0
 * if @heap != NULL.
1013
 */
1014 1015
static void update_tasks_nodemask(struct cpuset *cs, const nodemask_t *oldmem,
				 struct ptr_heap *heap)
Linus Torvalds's avatar
Linus Torvalds committed
1016
{
1017
	struct cgroup_scanner scan;
1018

1019
	cpuset_being_rebound = cs;		/* causes mpol_dup() rebind */
1020

1021 1022 1023
	scan.cg = cs->css.cgroup;
	scan.test_task = NULL;
	scan.process_task = cpuset_change_nodemask;
1024
	scan.heap = heap;
1025
	scan.data = (nodemask_t *)oldmem;
1026 1027

	/*
1028 1029 1030 1031
	 * 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
1032
	 * the global cpuset_mutex, we know that no other rebind effort
1033
	 * will be contending for the global variable cpuset_being_rebound.
1034
	 * It's ok if we rebind the same mm twice; mpol_rebind_mm()
1035
	 * is idempotent.  Also migrate pages in each mm to new nodes.
1036
	 */
1037
	cgroup_scan_tasks(&scan);
1038

1039
	/* We're done rebinding vmas to this cpuset's new mems_allowed. */
1040
	cpuset_being_rebound = NULL;
Linus Torvalds's avatar
Linus Torvalds committed
1041 1042
}

1043 1044 1045
/*
 * Handle user request to change the 'mems' memory placement
 * of a cpuset.  Needs to validate the request, update the
1046 1047 1048 1049
 * 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.
1050
 *
1051
 * Call with cpuset_mutex held.  May take callback_mutex during call.
1052 1053 1054 1055
 * 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.
 */
1056 1057
static int update_nodemask(struct cpuset *cs, struct cpuset *trialcs,
			   const char *buf)
1058
{
1059
	NODEMASK_ALLOC(nodemask_t, oldmem, GFP_KERNEL);
1060
	int retval;
1061
	struct ptr_heap heap;
1062

1063 1064 1065
	if (!oldmem)
		return -ENOMEM;

1066
	/*
1067
	 * top_cpuset.mems_allowed tracks node_stats[N_MEMORY];
1068 1069
	 * it's read-only
	 */
1070 1071 1072 1073
	if (cs == &top_cpuset) {
		retval = -EACCES;
		goto done;
	}
1074 1075 1076 1077 1078 1079 1080 1081

	/*
	 * 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) {
1082
		nodes_clear(trialcs->mems_allowed);
1083
	} else {
Li Zefan's avatar