tree.c 132.1 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14
/*
 * Read-Copy Update mechanism for mutual exclusion
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2 of the License, or
 * (at your option) any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
15 16
 * along with this program; if not, you can access it online at
 * http://www.gnu.org/licenses/gpl-2.0.html.
17 18 19 20 21 22 23 24 25 26 27
 *
 * Copyright IBM Corporation, 2008
 *
 * Authors: Dipankar Sarma <dipankar@in.ibm.com>
 *	    Manfred Spraul <manfred@colorfullife.com>
 *	    Paul E. McKenney <paulmck@linux.vnet.ibm.com> Hierarchical version
 *
 * Based on the original work by Paul McKenney <paulmck@us.ibm.com>
 * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen.
 *
 * For detailed explanation of Read-Copy Update mechanism see -
28
 *	Documentation/RCU
29 30 31 32 33 34 35 36 37
 */
#include <linux/types.h>
#include <linux/kernel.h>
#include <linux/init.h>
#include <linux/spinlock.h>
#include <linux/smp.h>
#include <linux/rcupdate.h>
#include <linux/interrupt.h>
#include <linux/sched.h>
38
#include <linux/nmi.h>
39
#include <linux/atomic.h>
40
#include <linux/bitops.h>
41
#include <linux/export.h>
42 43
#include <linux/completion.h>
#include <linux/moduleparam.h>
44
#include <linux/module.h>
45 46 47 48 49
#include <linux/percpu.h>
#include <linux/notifier.h>
#include <linux/cpu.h>
#include <linux/mutex.h>
#include <linux/time.h>
50
#include <linux/kernel_stat.h>
51 52
#include <linux/wait.h>
#include <linux/kthread.h>
53
#include <linux/prefetch.h>
54 55
#include <linux/delay.h>
#include <linux/stop_machine.h>
56
#include <linux/random.h>
57
#include <linux/trace_events.h>
58
#include <linux/suspend.h>
59

60
#include "tree.h"
61
#include "rcu.h"
62

63 64 65 66 67 68
MODULE_ALIAS("rcutree");
#ifdef MODULE_PARAM_PREFIX
#undef MODULE_PARAM_PREFIX
#endif
#define MODULE_PARAM_PREFIX "rcutree."

69 70
/* Data structures. */

71
static struct lock_class_key rcu_node_class[RCU_NUM_LVLS];
72
static struct lock_class_key rcu_fqs_class[RCU_NUM_LVLS];
73

74 75 76 77 78 79 80 81
/*
 * In order to export the rcu_state name to the tracing tools, it
 * needs to be added in the __tracepoint_string section.
 * This requires defining a separate variable tp_<sname>_varname
 * that points to the string being used, and this will allow
 * the tracing userspace tools to be able to decipher the string
 * address to the matching string.
 */
82 83
#ifdef CONFIG_TRACING
# define DEFINE_RCU_TPS(sname) \
84
static char sname##_varname[] = #sname; \
85 86 87 88 89 90 91 92 93
static const char *tp_##sname##_varname __used __tracepoint_string = sname##_varname;
# define RCU_STATE_NAME(sname) sname##_varname
#else
# define DEFINE_RCU_TPS(sname)
# define RCU_STATE_NAME(sname) __stringify(sname)
#endif

#define RCU_STATE_INITIALIZER(sname, sabbr, cr) \
DEFINE_RCU_TPS(sname) \
94
static DEFINE_PER_CPU_SHARED_ALIGNED(struct rcu_data, sname##_data); \
95
struct rcu_state sname##_state = { \
96
	.level = { &sname##_state.node[0] }, \
97
	.rda = &sname##_data, \
98
	.call = cr, \
99
	.fqs_state = RCU_GP_IDLE, \
P
Paul E. McKenney 已提交
100 101
	.gpnum = 0UL - 300UL, \
	.completed = 0UL - 300UL, \
102
	.orphan_lock = __RAW_SPIN_LOCK_UNLOCKED(&sname##_state.orphan_lock), \
103 104
	.orphan_nxttail = &sname##_state.orphan_nxtlist, \
	.orphan_donetail = &sname##_state.orphan_donelist, \
105
	.barrier_mutex = __MUTEX_INITIALIZER(sname##_state.barrier_mutex), \
106
	.name = RCU_STATE_NAME(sname), \
107
	.abbr = sabbr, \
108
}
109

110 111
RCU_STATE_INITIALIZER(rcu_sched, 's', call_rcu_sched);
RCU_STATE_INITIALIZER(rcu_bh, 'b', call_rcu_bh);
112

113
static struct rcu_state *const rcu_state_p;
114
static struct rcu_data __percpu *const rcu_data_p;
115
LIST_HEAD(rcu_struct_flavors);
116

117 118 119
/* Dump rcu_node combining tree at boot to verify correct setup. */
static bool dump_tree;
module_param(dump_tree, bool, 0444);
120 121 122
/* Control rcu_node-tree auto-balancing at boot time. */
static bool rcu_fanout_exact;
module_param(rcu_fanout_exact, bool, 0444);
123 124
/* Increase (but not decrease) the RCU_FANOUT_LEAF at boot time. */
static int rcu_fanout_leaf = RCU_FANOUT_LEAF;
125
module_param(rcu_fanout_leaf, int, 0444);
126 127 128 129 130 131 132 133 134 135
int rcu_num_lvls __read_mostly = RCU_NUM_LVLS;
static int num_rcu_lvl[] = {  /* Number of rcu_nodes at specified level. */
	NUM_RCU_LVL_0,
	NUM_RCU_LVL_1,
	NUM_RCU_LVL_2,
	NUM_RCU_LVL_3,
	NUM_RCU_LVL_4,
};
int rcu_num_nodes __read_mostly = NUM_RCU_NODES; /* Total # rcu_nodes in use. */

136 137 138 139
/*
 * The rcu_scheduler_active variable transitions from zero to one just
 * before the first task is spawned.  So when this variable is zero, RCU
 * can assume that there is but one task, allowing RCU to (for example)
140
 * optimize synchronize_sched() to a simple barrier().  When this variable
141 142 143 144
 * is one, RCU must actually do all the hard work required to detect real
 * grace periods.  This variable is also used to suppress boot-time false
 * positives from lockdep-RCU error checking.
 */
145 146 147
int rcu_scheduler_active __read_mostly;
EXPORT_SYMBOL_GPL(rcu_scheduler_active);

148 149 150 151 152 153 154 155 156 157 158 159 160 161
/*
 * The rcu_scheduler_fully_active variable transitions from zero to one
 * during the early_initcall() processing, which is after the scheduler
 * is capable of creating new tasks.  So RCU processing (for example,
 * creating tasks for RCU priority boosting) must be delayed until after
 * rcu_scheduler_fully_active transitions from zero to one.  We also
 * currently delay invocation of any RCU callbacks until after this point.
 *
 * It might later prove better for people registering RCU callbacks during
 * early boot to take responsibility for these callbacks, but one step at
 * a time.
 */
static int rcu_scheduler_fully_active __read_mostly;

162 163
static void rcu_init_new_rnp(struct rcu_node *rnp_leaf);
static void rcu_cleanup_dead_rnp(struct rcu_node *rnp_leaf);
T
Thomas Gleixner 已提交
164
static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu);
165 166
static void invoke_rcu_core(void);
static void invoke_rcu_callbacks(struct rcu_state *rsp, struct rcu_data *rdp);
167

168
/* rcuc/rcub kthread realtime priority */
169
#ifdef CONFIG_RCU_KTHREAD_PRIO
170
static int kthread_prio = CONFIG_RCU_KTHREAD_PRIO;
171 172 173
#else /* #ifdef CONFIG_RCU_KTHREAD_PRIO */
static int kthread_prio = IS_ENABLED(CONFIG_RCU_BOOST) ? 1 : 0;
#endif /* #else #ifdef CONFIG_RCU_KTHREAD_PRIO */
174 175
module_param(kthread_prio, int, 0644);

176
/* Delay in jiffies for grace-period initialization delays, debug only. */
177 178 179 180 181 182 183 184

#ifdef CONFIG_RCU_TORTURE_TEST_SLOW_PREINIT
static int gp_preinit_delay = CONFIG_RCU_TORTURE_TEST_SLOW_PREINIT_DELAY;
module_param(gp_preinit_delay, int, 0644);
#else /* #ifdef CONFIG_RCU_TORTURE_TEST_SLOW_PREINIT */
static const int gp_preinit_delay;
#endif /* #else #ifdef CONFIG_RCU_TORTURE_TEST_SLOW_PREINIT */

185 186
#ifdef CONFIG_RCU_TORTURE_TEST_SLOW_INIT
static int gp_init_delay = CONFIG_RCU_TORTURE_TEST_SLOW_INIT_DELAY;
187
module_param(gp_init_delay, int, 0644);
188 189 190
#else /* #ifdef CONFIG_RCU_TORTURE_TEST_SLOW_INIT */
static const int gp_init_delay;
#endif /* #else #ifdef CONFIG_RCU_TORTURE_TEST_SLOW_INIT */
191

192 193 194 195 196 197 198
#ifdef CONFIG_RCU_TORTURE_TEST_SLOW_CLEANUP
static int gp_cleanup_delay = CONFIG_RCU_TORTURE_TEST_SLOW_CLEANUP_DELAY;
module_param(gp_cleanup_delay, int, 0644);
#else /* #ifdef CONFIG_RCU_TORTURE_TEST_SLOW_CLEANUP */
static const int gp_cleanup_delay;
#endif /* #else #ifdef CONFIG_RCU_TORTURE_TEST_SLOW_CLEANUP */

199 200 201 202 203 204 205 206 207 208
/*
 * Number of grace periods between delays, normalized by the duration of
 * the delay.  The longer the the delay, the more the grace periods between
 * each delay.  The reason for this normalization is that it means that,
 * for non-zero delays, the overall slowdown of grace periods is constant
 * regardless of the duration of the delay.  This arrangement balances
 * the need for long delays to increase some race probabilities with the
 * need for fast grace periods to increase other race probabilities.
 */
#define PER_RCU_NODE_PERIOD 3	/* Number of grace periods between delays. */
209

210 211 212 213 214 215 216 217 218 219 220 221
/*
 * Track the rcutorture test sequence number and the update version
 * number within a given test.  The rcutorture_testseq is incremented
 * on every rcutorture module load and unload, so has an odd value
 * when a test is running.  The rcutorture_vernum is set to zero
 * when rcutorture starts and is incremented on each rcutorture update.
 * These variables enable correlating rcutorture output with the
 * RCU tracing information.
 */
unsigned long rcutorture_testseq;
unsigned long rcutorture_vernum;

222 223 224 225 226 227 228 229
/*
 * Compute the mask of online CPUs for the specified rcu_node structure.
 * This will not be stable unless the rcu_node structure's ->lock is
 * held, but the bit corresponding to the current CPU will be stable
 * in most contexts.
 */
unsigned long rcu_rnp_online_cpus(struct rcu_node *rnp)
{
230
	return READ_ONCE(rnp->qsmaskinitnext);
231 232
}

233
/*
234
 * Return true if an RCU grace period is in progress.  The READ_ONCE()s
235 236 237 238 239
 * permit this function to be invoked without holding the root rcu_node
 * structure's ->lock, but of course results can be subject to change.
 */
static int rcu_gp_in_progress(struct rcu_state *rsp)
{
240
	return READ_ONCE(rsp->completed) != READ_ONCE(rsp->gpnum);
241 242
}

243
/*
244
 * Note a quiescent state.  Because we do not need to know
245
 * how many quiescent states passed, just if there was at least
246
 * one since the start of the grace period, this just sets a flag.
247
 * The caller must have disabled preemption.
248
 */
249
void rcu_sched_qs(void)
250
{
251 252 253 254 255 256
	if (!__this_cpu_read(rcu_sched_data.passed_quiesce)) {
		trace_rcu_grace_period(TPS("rcu_sched"),
				       __this_cpu_read(rcu_sched_data.gpnum),
				       TPS("cpuqs"));
		__this_cpu_write(rcu_sched_data.passed_quiesce, 1);
	}
257 258
}

259
void rcu_bh_qs(void)
260
{
261 262 263 264 265 266
	if (!__this_cpu_read(rcu_bh_data.passed_quiesce)) {
		trace_rcu_grace_period(TPS("rcu_bh"),
				       __this_cpu_read(rcu_bh_data.gpnum),
				       TPS("cpuqs"));
		__this_cpu_write(rcu_bh_data.passed_quiesce, 1);
	}
267
}
268

269 270 271 272 273 274 275 276 277 278 279
static DEFINE_PER_CPU(int, rcu_sched_qs_mask);

static DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks) = {
	.dynticks_nesting = DYNTICK_TASK_EXIT_IDLE,
	.dynticks = ATOMIC_INIT(1),
#ifdef CONFIG_NO_HZ_FULL_SYSIDLE
	.dynticks_idle_nesting = DYNTICK_TASK_NEST_VALUE,
	.dynticks_idle = ATOMIC_INIT(1),
#endif /* #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */
};

280 281 282
DEFINE_PER_CPU_SHARED_ALIGNED(unsigned long, rcu_qs_ctr);
EXPORT_PER_CPU_SYMBOL_GPL(rcu_qs_ctr);

283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316
/*
 * Let the RCU core know that this CPU has gone through the scheduler,
 * which is a quiescent state.  This is called when the need for a
 * quiescent state is urgent, so we burn an atomic operation and full
 * memory barriers to let the RCU core know about it, regardless of what
 * this CPU might (or might not) do in the near future.
 *
 * We inform the RCU core by emulating a zero-duration dyntick-idle
 * period, which we in turn do by incrementing the ->dynticks counter
 * by two.
 */
static void rcu_momentary_dyntick_idle(void)
{
	unsigned long flags;
	struct rcu_data *rdp;
	struct rcu_dynticks *rdtp;
	int resched_mask;
	struct rcu_state *rsp;

	local_irq_save(flags);

	/*
	 * Yes, we can lose flag-setting operations.  This is OK, because
	 * the flag will be set again after some delay.
	 */
	resched_mask = raw_cpu_read(rcu_sched_qs_mask);
	raw_cpu_write(rcu_sched_qs_mask, 0);

	/* Find the flavor that needs a quiescent state. */
	for_each_rcu_flavor(rsp) {
		rdp = raw_cpu_ptr(rsp->rda);
		if (!(resched_mask & rsp->flavor_mask))
			continue;
		smp_mb(); /* rcu_sched_qs_mask before cond_resched_completed. */
317 318
		if (READ_ONCE(rdp->mynode->completed) !=
		    READ_ONCE(rdp->cond_resched_completed))
319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335
			continue;

		/*
		 * Pretend to be momentarily idle for the quiescent state.
		 * This allows the grace-period kthread to record the
		 * quiescent state, with no need for this CPU to do anything
		 * further.
		 */
		rdtp = this_cpu_ptr(&rcu_dynticks);
		smp_mb__before_atomic(); /* Earlier stuff before QS. */
		atomic_add(2, &rdtp->dynticks);  /* QS. */
		smp_mb__after_atomic(); /* Later stuff after QS. */
		break;
	}
	local_irq_restore(flags);
}

336 337 338
/*
 * Note a context switch.  This is a quiescent state for RCU-sched,
 * and requires special handling for preemptible RCU.
339
 * The caller must have disabled preemption.
340
 */
341
void rcu_note_context_switch(void)
342
{
343
	trace_rcu_utilization(TPS("Start context switch"));
344
	rcu_sched_qs();
345
	rcu_preempt_note_context_switch();
346 347
	if (unlikely(raw_cpu_read(rcu_sched_qs_mask)))
		rcu_momentary_dyntick_idle();
348
	trace_rcu_utilization(TPS("End context switch"));
349
}
350
EXPORT_SYMBOL_GPL(rcu_note_context_switch);
351

352
/*
353
 * Register a quiescent state for all RCU flavors.  If there is an
354 355
 * emergency, invoke rcu_momentary_dyntick_idle() to do a heavy-weight
 * dyntick-idle quiescent state visible to other CPUs (but only for those
356
 * RCU flavors in desperate need of a quiescent state, which will normally
357 358 359 360 361 362 363 364 365 366 367
 * be none of them).  Either way, do a lightweight quiescent state for
 * all RCU flavors.
 */
void rcu_all_qs(void)
{
	if (unlikely(raw_cpu_read(rcu_sched_qs_mask)))
		rcu_momentary_dyntick_idle();
	this_cpu_inc(rcu_qs_ctr);
}
EXPORT_SYMBOL_GPL(rcu_all_qs);

E
Eric Dumazet 已提交
368 369 370
static long blimit = 10;	/* Maximum callbacks per rcu_do_batch. */
static long qhimark = 10000;	/* If this many pending, ignore blimit. */
static long qlowmark = 100;	/* Once only this many pending, use blimit. */
371

E
Eric Dumazet 已提交
372 373 374
module_param(blimit, long, 0444);
module_param(qhimark, long, 0444);
module_param(qlowmark, long, 0444);
375

376 377
static ulong jiffies_till_first_fqs = ULONG_MAX;
static ulong jiffies_till_next_fqs = ULONG_MAX;
378 379 380 381

module_param(jiffies_till_first_fqs, ulong, 0644);
module_param(jiffies_till_next_fqs, ulong, 0644);

382 383 384 385 386 387 388
/*
 * How long the grace period must be before we start recruiting
 * quiescent-state help from rcu_note_context_switch().
 */
static ulong jiffies_till_sched_qs = HZ / 20;
module_param(jiffies_till_sched_qs, ulong, 0644);

389
static bool rcu_start_gp_advanced(struct rcu_state *rsp, struct rcu_node *rnp,
390
				  struct rcu_data *rdp);
391 392 393 394
static void force_qs_rnp(struct rcu_state *rsp,
			 int (*f)(struct rcu_data *rsp, bool *isidle,
				  unsigned long *maxj),
			 bool *isidle, unsigned long *maxj);
395
static void force_quiescent_state(struct rcu_state *rsp);
396
static int rcu_pending(void);
397 398

/*
399
 * Return the number of RCU batches started thus far for debug & stats.
400
 */
401 402 403 404 405 406 407 408
unsigned long rcu_batches_started(void)
{
	return rcu_state_p->gpnum;
}
EXPORT_SYMBOL_GPL(rcu_batches_started);

/*
 * Return the number of RCU-sched batches started thus far for debug & stats.
409
 */
410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435
unsigned long rcu_batches_started_sched(void)
{
	return rcu_sched_state.gpnum;
}
EXPORT_SYMBOL_GPL(rcu_batches_started_sched);

/*
 * Return the number of RCU BH batches started thus far for debug & stats.
 */
unsigned long rcu_batches_started_bh(void)
{
	return rcu_bh_state.gpnum;
}
EXPORT_SYMBOL_GPL(rcu_batches_started_bh);

/*
 * Return the number of RCU batches completed thus far for debug & stats.
 */
unsigned long rcu_batches_completed(void)
{
	return rcu_state_p->completed;
}
EXPORT_SYMBOL_GPL(rcu_batches_completed);

/*
 * Return the number of RCU-sched batches completed thus far for debug & stats.
436
 */
437
unsigned long rcu_batches_completed_sched(void)
438
{
439
	return rcu_sched_state.completed;
440
}
441
EXPORT_SYMBOL_GPL(rcu_batches_completed_sched);
442 443

/*
444
 * Return the number of RCU BH batches completed thus far for debug & stats.
445
 */
446
unsigned long rcu_batches_completed_bh(void)
447 448 449 450 451
{
	return rcu_bh_state.completed;
}
EXPORT_SYMBOL_GPL(rcu_batches_completed_bh);

452 453 454 455 456
/*
 * Force a quiescent state.
 */
void rcu_force_quiescent_state(void)
{
457
	force_quiescent_state(rcu_state_p);
458 459 460
}
EXPORT_SYMBOL_GPL(rcu_force_quiescent_state);

461 462 463 464 465
/*
 * Force a quiescent state for RCU BH.
 */
void rcu_bh_force_quiescent_state(void)
{
466
	force_quiescent_state(&rcu_bh_state);
467 468 469
}
EXPORT_SYMBOL_GPL(rcu_bh_force_quiescent_state);

470 471 472 473 474 475 476 477 478
/*
 * Force a quiescent state for RCU-sched.
 */
void rcu_sched_force_quiescent_state(void)
{
	force_quiescent_state(&rcu_sched_state);
}
EXPORT_SYMBOL_GPL(rcu_sched_force_quiescent_state);

479 480 481 482 483 484 485 486 487 488 489 490 491 492 493
/*
 * Show the state of the grace-period kthreads.
 */
void show_rcu_gp_kthreads(void)
{
	struct rcu_state *rsp;

	for_each_rcu_flavor(rsp) {
		pr_info("%s: wait state: %d ->state: %#lx\n",
			rsp->name, rsp->gp_state, rsp->gp_kthread->state);
		/* sched_show_task(rsp->gp_kthread); */
	}
}
EXPORT_SYMBOL_GPL(show_rcu_gp_kthreads);

494 495 496 497 498 499 500 501 502 503 504 505 506 507
/*
 * Record the number of times rcutorture tests have been initiated and
 * terminated.  This information allows the debugfs tracing stats to be
 * correlated to the rcutorture messages, even when the rcutorture module
 * is being repeatedly loaded and unloaded.  In other words, we cannot
 * store this state in rcutorture itself.
 */
void rcutorture_record_test_transition(void)
{
	rcutorture_testseq++;
	rcutorture_vernum = 0;
}
EXPORT_SYMBOL_GPL(rcutorture_record_test_transition);

508 509 510 511 512 513 514 515 516 517
/*
 * Send along grace-period-related data for rcutorture diagnostics.
 */
void rcutorture_get_gp_data(enum rcutorture_type test_type, int *flags,
			    unsigned long *gpnum, unsigned long *completed)
{
	struct rcu_state *rsp = NULL;

	switch (test_type) {
	case RCU_FLAVOR:
518
		rsp = rcu_state_p;
519 520 521 522 523 524 525 526 527 528 529
		break;
	case RCU_BH_FLAVOR:
		rsp = &rcu_bh_state;
		break;
	case RCU_SCHED_FLAVOR:
		rsp = &rcu_sched_state;
		break;
	default:
		break;
	}
	if (rsp != NULL) {
530 531 532
		*flags = READ_ONCE(rsp->gp_flags);
		*gpnum = READ_ONCE(rsp->gpnum);
		*completed = READ_ONCE(rsp->completed);
533 534 535 536 537 538 539 540
		return;
	}
	*flags = 0;
	*gpnum = 0;
	*completed = 0;
}
EXPORT_SYMBOL_GPL(rcutorture_get_gp_data);

541 542 543 544 545 546 547 548 549 550 551
/*
 * Record the number of writer passes through the current rcutorture test.
 * This is also used to correlate debugfs tracing stats with the rcutorture
 * messages.
 */
void rcutorture_record_progress(unsigned long vernum)
{
	rcutorture_vernum++;
}
EXPORT_SYMBOL_GPL(rcutorture_record_progress);

552 553 554 555 556 557
/*
 * Does the CPU have callbacks ready to be invoked?
 */
static int
cpu_has_callbacks_ready_to_invoke(struct rcu_data *rdp)
{
P
Paul E. McKenney 已提交
558 559
	return &rdp->nxtlist != rdp->nxttail[RCU_DONE_TAIL] &&
	       rdp->nxttail[RCU_DONE_TAIL] != NULL;
560 561
}

562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577
/*
 * Return the root node of the specified rcu_state structure.
 */
static struct rcu_node *rcu_get_root(struct rcu_state *rsp)
{
	return &rsp->node[0];
}

/*
 * Is there any need for future grace periods?
 * Interrupts must be disabled.  If the caller does not hold the root
 * rnp_node structure's ->lock, the results are advisory only.
 */
static int rcu_future_needs_gp(struct rcu_state *rsp)
{
	struct rcu_node *rnp = rcu_get_root(rsp);
578
	int idx = (READ_ONCE(rnp->completed) + 1) & 0x1;
579 580
	int *fp = &rnp->need_future_gp[idx];

581
	return READ_ONCE(*fp);
582 583
}

584
/*
585 586 587
 * Does the current CPU require a not-yet-started grace period?
 * The caller must have disabled interrupts to prevent races with
 * normal callback registry.
588 589 590 591
 */
static int
cpu_needs_another_gp(struct rcu_state *rsp, struct rcu_data *rdp)
{
592
	int i;
P
Paul E. McKenney 已提交
593

594 595
	if (rcu_gp_in_progress(rsp))
		return 0;  /* No, a grace period is already in progress. */
596
	if (rcu_future_needs_gp(rsp))
597
		return 1;  /* Yes, a no-CBs CPU needs one. */
598 599 600 601 602 603
	if (!rdp->nxttail[RCU_NEXT_TAIL])
		return 0;  /* No, this is a no-CBs (or offline) CPU. */
	if (*rdp->nxttail[RCU_NEXT_READY_TAIL])
		return 1;  /* Yes, this CPU has newly registered callbacks. */
	for (i = RCU_WAIT_TAIL; i < RCU_NEXT_TAIL; i++)
		if (rdp->nxttail[i - 1] != rdp->nxttail[i] &&
604
		    ULONG_CMP_LT(READ_ONCE(rsp->completed),
605 606 607
				 rdp->nxtcompleted[i]))
			return 1;  /* Yes, CBs for future grace period. */
	return 0; /* No grace period needed. */
608 609
}

610
/*
611
 * rcu_eqs_enter_common - current CPU is moving towards extended quiescent state
612 613 614 615 616
 *
 * If the new value of the ->dynticks_nesting counter now is zero,
 * we really have entered idle, and must do the appropriate accounting.
 * The caller must have disabled interrupts.
 */
617
static void rcu_eqs_enter_common(long long oldval, bool user)
618
{
619 620
	struct rcu_state *rsp;
	struct rcu_data *rdp;
621
	struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks);
622

623
	trace_rcu_dyntick(TPS("Start"), oldval, rdtp->dynticks_nesting);
624 625
	if (IS_ENABLED(CONFIG_RCU_EQS_DEBUG) &&
	    !user && !is_idle_task(current)) {
626 627
		struct task_struct *idle __maybe_unused =
			idle_task(smp_processor_id());
628

629
		trace_rcu_dyntick(TPS("Error on entry: not idle task"), oldval, 0);
630
		ftrace_dump(DUMP_ORIG);
631 632 633
		WARN_ONCE(1, "Current pid: %d comm: %s / Idle pid: %d comm: %s",
			  current->pid, current->comm,
			  idle->pid, idle->comm); /* must be idle task! */
634
	}
635 636 637 638
	for_each_rcu_flavor(rsp) {
		rdp = this_cpu_ptr(rsp->rda);
		do_nocb_deferred_wakeup(rdp);
	}
639
	rcu_prepare_for_idle();
640
	/* CPUs seeing atomic_inc() must see prior RCU read-side crit sects */
641
	smp_mb__before_atomic();  /* See above. */
642
	atomic_inc(&rdtp->dynticks);
643
	smp_mb__after_atomic();  /* Force ordering with next sojourn. */
644 645
	WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) &&
		     atomic_read(&rdtp->dynticks) & 0x1);
646
	rcu_dynticks_task_enter();
647 648

	/*
649
	 * It is illegal to enter an extended quiescent state while
650 651 652 653 654 655 656 657
	 * in an RCU read-side critical section.
	 */
	rcu_lockdep_assert(!lock_is_held(&rcu_lock_map),
			   "Illegal idle entry in RCU read-side critical section.");
	rcu_lockdep_assert(!lock_is_held(&rcu_bh_lock_map),
			   "Illegal idle entry in RCU-bh read-side critical section.");
	rcu_lockdep_assert(!lock_is_held(&rcu_sched_lock_map),
			   "Illegal idle entry in RCU-sched read-side critical section.");
658
}
659

660 661 662
/*
 * Enter an RCU extended quiescent state, which can be either the
 * idle loop or adaptive-tickless usermode execution.
663
 */
664
static void rcu_eqs_enter(bool user)
665
{
666
	long long oldval;
667 668
	struct rcu_dynticks *rdtp;

669
	rdtp = this_cpu_ptr(&rcu_dynticks);
670
	oldval = rdtp->dynticks_nesting;
671 672
	WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) &&
		     (oldval & DYNTICK_TASK_NEST_MASK) == 0);
673
	if ((oldval & DYNTICK_TASK_NEST_MASK) == DYNTICK_TASK_NEST_VALUE) {
674
		rdtp->dynticks_nesting = 0;
675
		rcu_eqs_enter_common(oldval, user);
676
	} else {
677
		rdtp->dynticks_nesting -= DYNTICK_TASK_NEST_VALUE;
678
	}
679
}
680 681 682 683 684 685 686 687 688 689 690 691 692 693 694

/**
 * rcu_idle_enter - inform RCU that current CPU is entering idle
 *
 * Enter idle mode, in other words, -leave- the mode in which RCU
 * read-side critical sections can occur.  (Though RCU read-side
 * critical sections can occur in irq handlers in idle, a possibility
 * handled by irq_enter() and irq_exit().)
 *
 * We crowbar the ->dynticks_nesting field to zero to allow for
 * the possibility of usermode upcalls having messed up our count
 * of interrupt nesting level during the prior busy period.
 */
void rcu_idle_enter(void)
{
695 696 697
	unsigned long flags;

	local_irq_save(flags);
698
	rcu_eqs_enter(false);
699
	rcu_sysidle_enter(0);
700
	local_irq_restore(flags);
701
}
702
EXPORT_SYMBOL_GPL(rcu_idle_enter);
703

704
#ifdef CONFIG_NO_HZ_FULL
705 706 707 708 709 710 711 712 713 714
/**
 * rcu_user_enter - inform RCU that we are resuming userspace.
 *
 * Enter RCU idle mode right before resuming userspace.  No use of RCU
 * is permitted between this call and rcu_user_exit(). This way the
 * CPU doesn't need to maintain the tick for RCU maintenance purposes
 * when the CPU runs in userspace.
 */
void rcu_user_enter(void)
{
715
	rcu_eqs_enter(1);
716
}
717
#endif /* CONFIG_NO_HZ_FULL */
718

719 720 721 722 723 724
/**
 * rcu_irq_exit - inform RCU that current CPU is exiting irq towards idle
 *
 * Exit from an interrupt handler, which might possibly result in entering
 * idle mode, in other words, leaving the mode in which read-side critical
 * sections can occur.
725
 *
726 727 728 729 730 731 732 733
 * This code assumes that the idle loop never does anything that might
 * result in unbalanced calls to irq_enter() and irq_exit().  If your
 * architecture violates this assumption, RCU will give you what you
 * deserve, good and hard.  But very infrequently and irreproducibly.
 *
 * Use things like work queues to work around this limitation.
 *
 * You have been warned.
734
 */
735
void rcu_irq_exit(void)
736 737
{
	unsigned long flags;
738
	long long oldval;
739 740 741
	struct rcu_dynticks *rdtp;

	local_irq_save(flags);
742
	rdtp = this_cpu_ptr(&rcu_dynticks);
743
	oldval = rdtp->dynticks_nesting;
744
	rdtp->dynticks_nesting--;
745 746
	WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) &&
		     rdtp->dynticks_nesting < 0);
747
	if (rdtp->dynticks_nesting)
748
		trace_rcu_dyntick(TPS("--="), oldval, rdtp->dynticks_nesting);
749
	else
750 751
		rcu_eqs_enter_common(oldval, true);
	rcu_sysidle_enter(1);
752 753 754 755
	local_irq_restore(flags);
}

/*
756
 * rcu_eqs_exit_common - current CPU moving away from extended quiescent state
757 758 759 760 761
 *
 * If the new value of the ->dynticks_nesting counter was previously zero,
 * we really have exited idle, and must do the appropriate accounting.
 * The caller must have disabled interrupts.
 */
762
static void rcu_eqs_exit_common(long long oldval, int user)
763
{
764 765
	struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks);

766
	rcu_dynticks_task_exit();
767
	smp_mb__before_atomic();  /* Force ordering w/previous sojourn. */
768 769
	atomic_inc(&rdtp->dynticks);
	/* CPUs seeing atomic_inc() must see later RCU read-side crit sects */
770
	smp_mb__after_atomic();  /* See above. */
771 772
	WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) &&
		     !(atomic_read(&rdtp->dynticks) & 0x1));
773
	rcu_cleanup_after_idle();
774
	trace_rcu_dyntick(TPS("End"), oldval, rdtp->dynticks_nesting);
775 776
	if (IS_ENABLED(CONFIG_RCU_EQS_DEBUG) &&
	    !user && !is_idle_task(current)) {
777 778
		struct task_struct *idle __maybe_unused =
			idle_task(smp_processor_id());
779

780
		trace_rcu_dyntick(TPS("Error on exit: not idle task"),
781
				  oldval, rdtp->dynticks_nesting);
782
		ftrace_dump(DUMP_ORIG);
783 784 785
		WARN_ONCE(1, "Current pid: %d comm: %s / Idle pid: %d comm: %s",
			  current->pid, current->comm,
			  idle->pid, idle->comm); /* must be idle task! */
786 787 788
	}
}

789 790 791
/*
 * Exit an RCU extended quiescent state, which can be either the
 * idle loop or adaptive-tickless usermode execution.
792
 */
793
static void rcu_eqs_exit(bool user)
794 795 796 797
{
	struct rcu_dynticks *rdtp;
	long long oldval;

798
	rdtp = this_cpu_ptr(&rcu_dynticks);
799
	oldval = rdtp->dynticks_nesting;
800
	WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) && oldval < 0);
801
	if (oldval & DYNTICK_TASK_NEST_MASK) {
802
		rdtp->dynticks_nesting += DYNTICK_TASK_NEST_VALUE;
803
	} else {
804
		rdtp->dynticks_nesting = DYNTICK_TASK_EXIT_IDLE;
805
		rcu_eqs_exit_common(oldval, user);
806
	}
807
}
808 809 810 811 812 813 814 815 816 817 818 819 820 821

/**
 * rcu_idle_exit - inform RCU that current CPU is leaving idle
 *
 * Exit idle mode, in other words, -enter- the mode in which RCU
 * read-side critical sections can occur.
 *
 * We crowbar the ->dynticks_nesting field to DYNTICK_TASK_NEST to
 * allow for the possibility of usermode upcalls messing up our count
 * of interrupt nesting level during the busy period that is just
 * now starting.
 */
void rcu_idle_exit(void)
{
822 823 824
	unsigned long flags;

	local_irq_save(flags);
825
	rcu_eqs_exit(false);
826
	rcu_sysidle_exit(0);
827
	local_irq_restore(flags);
828
}
829
EXPORT_SYMBOL_GPL(rcu_idle_exit);
830

831
#ifdef CONFIG_NO_HZ_FULL
832 833 834 835 836 837 838 839
/**
 * rcu_user_exit - inform RCU that we are exiting userspace.
 *
 * Exit RCU idle mode while entering the kernel because it can
 * run a RCU read side critical section anytime.
 */
void rcu_user_exit(void)
{
840
	rcu_eqs_exit(1);
841
}
842
#endif /* CONFIG_NO_HZ_FULL */
843

844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869
/**
 * rcu_irq_enter - inform RCU that current CPU is entering irq away from idle
 *
 * Enter an interrupt handler, which might possibly result in exiting
 * idle mode, in other words, entering the mode in which read-side critical
 * sections can occur.
 *
 * Note that the Linux kernel is fully capable of entering an interrupt
 * handler that it never exits, for example when doing upcalls to
 * user mode!  This code assumes that the idle loop never does upcalls to
 * user mode.  If your architecture does do upcalls from the idle loop (or
 * does anything else that results in unbalanced calls to the irq_enter()
 * and irq_exit() functions), RCU will give you what you deserve, good
 * and hard.  But very infrequently and irreproducibly.
 *
 * Use things like work queues to work around this limitation.
 *
 * You have been warned.
 */
void rcu_irq_enter(void)
{
	unsigned long flags;
	struct rcu_dynticks *rdtp;
	long long oldval;

	local_irq_save(flags);
870
	rdtp = this_cpu_ptr(&rcu_dynticks);
871 872
	oldval = rdtp->dynticks_nesting;
	rdtp->dynticks_nesting++;
873 874
	WARN_ON_ONCE(IS_ENABLED(CONFIG_RCU_EQS_DEBUG) &&
		     rdtp->dynticks_nesting == 0);
875
	if (oldval)
876
		trace_rcu_dyntick(TPS("++="), oldval, rdtp->dynticks_nesting);
877
	else
878 879
		rcu_eqs_exit_common(oldval, true);
	rcu_sysidle_exit(1);
880 881 882 883 884 885
	local_irq_restore(flags);
}

/**
 * rcu_nmi_enter - inform RCU of entry to NMI context
 *
886 887 888 889 890
 * If the CPU was idle from RCU's viewpoint, update rdtp->dynticks and
 * rdtp->dynticks_nmi_nesting to let the RCU grace-period handling know
 * that the CPU is active.  This implementation permits nested NMIs, as
 * long as the nesting level does not overflow an int.  (You will probably
 * run out of stack space first.)
891 892 893
 */
void rcu_nmi_enter(void)
{
894
	struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks);
895
	int incby = 2;
896

897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917
	/* Complain about underflow. */
	WARN_ON_ONCE(rdtp->dynticks_nmi_nesting < 0);

	/*
	 * If idle from RCU viewpoint, atomically increment ->dynticks
	 * to mark non-idle and increment ->dynticks_nmi_nesting by one.
	 * Otherwise, increment ->dynticks_nmi_nesting by two.  This means
	 * if ->dynticks_nmi_nesting is equal to one, we are guaranteed
	 * to be in the outermost NMI handler that interrupted an RCU-idle
	 * period (observation due to Andy Lutomirski).
	 */
	if (!(atomic_read(&rdtp->dynticks) & 0x1)) {
		smp_mb__before_atomic();  /* Force delay from prior write. */
		atomic_inc(&rdtp->dynticks);
		/* atomic_inc() before later RCU read-side crit sects */
		smp_mb__after_atomic();  /* See above. */
		WARN_ON_ONCE(!(atomic_read(&rdtp->dynticks) & 0x1));
		incby = 1;
	}
	rdtp->dynticks_nmi_nesting += incby;
	barrier();
918 919 920 921 922
}

/**
 * rcu_nmi_exit - inform RCU of exit from NMI context
 *
923 924 925 926
 * If we are returning from the outermost NMI handler that interrupted an
 * RCU-idle period, update rdtp->dynticks and rdtp->dynticks_nmi_nesting
 * to let the RCU grace-period handling know that the CPU is back to
 * being RCU-idle.
927 928 929
 */
void rcu_nmi_exit(void)
{
930
	struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks);
931

932 933 934 935 936 937 938 939 940 941 942 943 944 945
	/*
	 * Check for ->dynticks_nmi_nesting underflow and bad ->dynticks.
	 * (We are exiting an NMI handler, so RCU better be paying attention
	 * to us!)
	 */
	WARN_ON_ONCE(rdtp->dynticks_nmi_nesting <= 0);
	WARN_ON_ONCE(!(atomic_read(&rdtp->dynticks) & 0x1));

	/*
	 * If the nesting level is not 1, the CPU wasn't RCU-idle, so
	 * leave it in non-RCU-idle state.
	 */
	if (rdtp->dynticks_nmi_nesting != 1) {
		rdtp->dynticks_nmi_nesting -= 2;
946
		return;
947 948 949 950
	}

	/* This NMI interrupted an RCU-idle CPU, restore RCU-idleness. */
	rdtp->dynticks_nmi_nesting = 0;
951
	/* CPUs seeing atomic_inc() must see prior RCU read-side crit sects */
952
	smp_mb__before_atomic();  /* See above. */
953
	atomic_inc(&rdtp->dynticks);
954
	smp_mb__after_atomic();  /* Force delay to next write. */
955
	WARN_ON_ONCE(atomic_read(&rdtp->dynticks) & 0x1);
956 957 958
}

/**
959 960 961 962 963 964 965
 * __rcu_is_watching - are RCU read-side critical sections safe?
 *
 * Return true if RCU is watching the running CPU, which means that
 * this CPU can safely enter RCU read-side critical sections.  Unlike
 * rcu_is_watching(), the caller of __rcu_is_watching() must have at
 * least disabled preemption.
 */
966
bool notrace __rcu_is_watching(void)
967 968 969 970 971 972
{
	return atomic_read(this_cpu_ptr(&rcu_dynticks.dynticks)) & 0x1;
}

/**
 * rcu_is_watching - see if RCU thinks that the current CPU is idle
973
 *
974
 * If the current CPU is in its idle loop and is neither in an interrupt
975
 * or NMI handler, return true.
976
 */
977
bool notrace rcu_is_watching(void)
978
{
979
	bool ret;
980 981

	preempt_disable();
982
	ret = __rcu_is_watching();
983 984
	preempt_enable();
	return ret;
985
}
986
EXPORT_SYMBOL_GPL(rcu_is_watching);
987

988
#if defined(CONFIG_PROVE_RCU) && defined(CONFIG_HOTPLUG_CPU)
989 990 991 992 993 994 995

/*
 * Is the current CPU online?  Disable preemption to avoid false positives
 * that could otherwise happen due to the current CPU number being sampled,
 * this task being preempted, its old CPU being taken offline, resuming
 * on some other CPU, then determining that its old CPU is now offline.
 * It is OK to use RCU on an offline processor during initial boot, hence
996 997 998 999 1000 1001 1002 1003 1004 1005 1006
 * the check for rcu_scheduler_fully_active.  Note also that it is OK
 * for a CPU coming online to use RCU for one jiffy prior to marking itself
 * online in the cpu_online_mask.  Similarly, it is OK for a CPU going
 * offline to continue to use RCU for one jiffy after marking itself
 * offline in the cpu_online_mask.  This leniency is necessary given the
 * non-atomic nature of the online and offline processing, for example,
 * the fact that a CPU enters the scheduler after completing the CPU_DYING
 * notifiers.
 *
 * This is also why RCU internally marks CPUs online during the
 * CPU_UP_PREPARE phase and offline during the CPU_DEAD phase.
1007 1008 1009 1010 1011 1012
 *
 * Disable checking if in an NMI handler because we cannot safely report
 * errors from NMI handlers anyway.
 */
bool rcu_lockdep_current_cpu_online(void)
{
1013 1014
	struct rcu_data *rdp;
	struct rcu_node *rnp;
1015 1016 1017
	bool ret;

	if (in_nmi())
F
Fengguang Wu 已提交
1018
		return true;
1019
	preempt_disable();
1020
	rdp = this_cpu_ptr(&rcu_sched_data);
1021
	rnp = rdp->mynode;
1022
	ret = (rdp->grpmask & rcu_rnp_online_cpus(rnp)) ||
1023 1024 1025 1026 1027 1028
	      !rcu_scheduler_fully_active;
	preempt_enable();
	return ret;
}
EXPORT_SYMBOL_GPL(rcu_lockdep_current_cpu_online);

1029
#endif /* #if defined(CONFIG_PROVE_RCU) && defined(CONFIG_HOTPLUG_CPU) */
1030

1031
/**
1032
 * rcu_is_cpu_rrupt_from_idle - see if idle or immediately interrupted from idle
1033
 *
1034 1035 1036
 * If the current CPU is idle or running at a first-level (not nested)
 * interrupt from idle, return true.  The caller must have at least
 * disabled preemption.
1037
 */
1038
static int rcu_is_cpu_rrupt_from_idle(void)
1039
{
1040
	return __this_cpu_read(rcu_dynticks.dynticks_nesting) <= 1;
1041 1042 1043 1044 1045
}

/*
 * Snapshot the specified CPU's dynticks counter so that we can later
 * credit them with an implicit quiescent state.  Return 1 if this CPU
1046
 * is in dynticks idle mode, which is an extended quiescent state.
1047
 */
1048 1049
static int dyntick_save_progress_counter(struct rcu_data *rdp,
					 bool *isidle, unsigned long *maxj)
1050
{
1051
	rdp->dynticks_snap = atomic_add_return(0, &rdp->dynticks->dynticks);
1052
	rcu_sysidle_check_cpu(rdp, isidle, maxj);
1053 1054 1055 1056
	if ((rdp->dynticks_snap & 0x1) == 0) {
		trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, TPS("dti"));
		return 1;
	} else {
1057
		if (ULONG_CMP_LT(READ_ONCE(rdp->gpnum) + ULONG_MAX / 4,
1058
				 rdp->mynode->gpnum))
1059
			WRITE_ONCE(rdp->gpwrap, true);
1060 1061
		return 0;
	}
1062 1063 1064 1065 1066 1067
}

/*
 * Return true if the specified CPU has passed through a quiescent
 * state by virtue of being in or having passed through an dynticks
 * idle state since the last call to dyntick_save_progress_counter()
1068
 * for this same CPU, or by virtue of having been offline.
1069
 */
1070 1071
static int rcu_implicit_dynticks_qs(struct rcu_data *rdp,
				    bool *isidle, unsigned long *maxj)
1072
{
1073
	unsigned int curr;
1074
	int *rcrmp;
1075
	unsigned int snap;
1076

1077 1078
	curr = (unsigned int)atomic_add_return(0, &rdp->dynticks->dynticks);
	snap = (unsigned int)rdp->dynticks_snap;
1079 1080 1081 1082 1083 1084 1085 1086 1087

	/*
	 * If the CPU passed through or entered a dynticks idle phase with
	 * no active irq/NMI handlers, then we can safely pretend that the CPU
	 * already acknowledged the request to pass through a quiescent
	 * state.  Either way, that CPU cannot possibly be in an RCU
	 * read-side critical section that started before the beginning
	 * of the current RCU grace period.
	 */
1088
	if ((curr & 0x1) == 0 || UINT_CMP_GE(curr, snap + 2)) {
1089
		trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, TPS("dti"));
1090 1091 1092 1093
		rdp->dynticks_fqs++;
		return 1;
	}

1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108
	/*
	 * Check for the CPU being offline, but only if the grace period
	 * is old enough.  We don't need to worry about the CPU changing
	 * state: If we see it offline even once, it has been through a
	 * quiescent state.
	 *
	 * The reason for insisting that the grace period be at least
	 * one jiffy old is that CPUs that are not quite online and that
	 * have just gone offline can still execute RCU read-side critical
	 * sections.
	 */
	if (ULONG_CMP_GE(rdp->rsp->gp_start + 2, jiffies))
		return 0;  /* Grace period is not old enough. */
	barrier();
	if (cpu_is_offline(rdp->cpu)) {
1109
		trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, TPS("ofl"));
1110 1111 1112
		rdp->offline_fqs++;
		return 1;
	}
1113 1114

	/*
1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133
	 * A CPU running for an extended time within the kernel can
	 * delay RCU grace periods.  When the CPU is in NO_HZ_FULL mode,
	 * even context-switching back and forth between a pair of
	 * in-kernel CPU-bound tasks cannot advance grace periods.
	 * So if the grace period is old enough, make the CPU pay attention.
	 * Note that the unsynchronized assignments to the per-CPU
	 * rcu_sched_qs_mask variable are safe.  Yes, setting of
	 * bits can be lost, but they will be set again on the next
	 * force-quiescent-state pass.  So lost bit sets do not result
	 * in incorrect behavior, merely in a grace period lasting
	 * a few jiffies longer than it might otherwise.  Because
	 * there are at most four threads involved, and because the
	 * updates are only once every few jiffies, the probability of
	 * lossage (and thus of slight grace-period extension) is
	 * quite low.
	 *
	 * Note that if the jiffies_till_sched_qs boot/sysfs parameter
	 * is set too high, we override with half of the RCU CPU stall
	 * warning delay.
1134
	 */
1135 1136 1137
	rcrmp = &per_cpu(rcu_sched_qs_mask, rdp->cpu);
	if (ULONG_CMP_GE(jiffies,
			 rdp->rsp->gp_start + jiffies_till_sched_qs) ||
1138
	    ULONG_CMP_GE(jiffies, rdp->rsp->jiffies_resched)) {
1139 1140 1141
		if (!(READ_ONCE(*rcrmp) & rdp->rsp->flavor_mask)) {
			WRITE_ONCE(rdp->cond_resched_completed,
				   READ_ONCE(rdp->mynode->completed));
1142
			smp_mb(); /* ->cond_resched_completed before *rcrmp. */
1143 1144
			WRITE_ONCE(*rcrmp,
				   READ_ONCE(*rcrmp) + rdp->rsp->flavor_mask);
1145 1146 1147 1148 1149 1150 1151
			resched_cpu(rdp->cpu);  /* Force CPU into scheduler. */
			rdp->rsp->jiffies_resched += 5; /* Enable beating. */
		} else if (ULONG_CMP_GE(jiffies, rdp->rsp->jiffies_resched)) {
			/* Time to beat on that CPU again! */
			resched_cpu(rdp->cpu);  /* Force CPU into scheduler. */
			rdp->rsp->jiffies_resched += 5; /* Re-enable beating. */
		}
1152 1153
	}

1154
	return 0;
1155 1156 1157 1158
}

static void record_gp_stall_check_time(struct rcu_state *rsp)
{
1159
	unsigned long j = jiffies;
1160
	unsigned long j1;
1161 1162 1163

	rsp->gp_start = j;
	smp_wmb(); /* Record start time before stall time. */
1164
	j1 = rcu_jiffies_till_stall_check();
1165
	WRITE_ONCE(rsp->jiffies_stall, j + j1);
1166
	rsp->jiffies_resched = j + j1 / 2;
1167
	rsp->n_force_qs_gpstart = READ_ONCE(rsp->n_force_qs);
1168 1169
}

1170 1171 1172 1173 1174 1175 1176 1177 1178
/*
 * Complain about starvation of grace-period kthread.
 */
static void rcu_check_gp_kthread_starvation(struct rcu_state *rsp)
{
	unsigned long gpa;
	unsigned long j;

	j = jiffies;
1179
	gpa = READ_ONCE(rsp->gp_activity);
1180
	if (j - gpa > 2 * HZ)
1181 1182 1183
		pr_err("%s kthread starved for %ld jiffies! g%lu c%lu f%#x\n",
		       rsp->name, j - gpa,
		       rsp->gpnum, rsp->completed, rsp->gp_flags);
1184 1185
}

1186
/*
1187
 * Dump stacks of all tasks running on stalled CPUs.
1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205
 */
static void rcu_dump_cpu_stacks(struct rcu_state *rsp)
{
	int cpu;
	unsigned long flags;
	struct rcu_node *rnp;

	rcu_for_each_leaf_node(rsp, rnp) {
		raw_spin_lock_irqsave(&rnp->lock, flags);
		if (rnp->qsmask != 0) {
			for (cpu = 0; cpu <= rnp->grphi - rnp->grplo; cpu++)
				if (rnp->qsmask & (1UL << cpu))
					dump_cpu_task(rnp->grplo + cpu);
		}
		raw_spin_unlock_irqrestore(&rnp->lock, flags);
	}
}

1206
static void print_other_cpu_stall(struct rcu_state *rsp, unsigned long gpnum)
1207 1208 1209 1210
{
	int cpu;
	long delta;
	unsigned long flags;
1211 1212
	unsigned long gpa;
	unsigned long j;
1213
	int ndetected = 0;
1214
	struct rcu_node *rnp = rcu_get_root(rsp);
1215
	long totqlen = 0;
1216 1217 1218

	/* Only let one CPU complain about others per time interval. */

P
Paul E. McKenney 已提交
1219
	raw_spin_lock_irqsave(&rnp->lock, flags);
1220
	delta = jiffies - READ_ONCE(rsp->jiffies_stall);
1221
	if (delta < RCU_STALL_RAT_DELAY || !rcu_gp_in_progress(rsp)) {
P
Paul E. McKenney 已提交
1222
		raw_spin_unlock_irqrestore(&rnp->lock, flags);
1223 1224
		return;
	}
1225 1226
	WRITE_ONCE(rsp->jiffies_stall,
		   jiffies + 3 * rcu_jiffies_till_stall_check() + 3);
P
Paul E. McKenney 已提交
1227
	raw_spin_unlock_irqrestore(&rnp->lock, flags);
1228

1229 1230 1231 1232 1233
	/*
	 * OK, time to rat on our buddy...
	 * See Documentation/RCU/stallwarn.txt for info on how to debug
	 * RCU CPU stall warnings.
	 */
1234
	pr_err("INFO: %s detected stalls on CPUs/tasks:",
1235
	       rsp->name);
1236
	print_cpu_stall_info_begin();
1237
	rcu_for_each_leaf_node(rsp, rnp) {
1238
		raw_spin_lock_irqsave(&rnp->lock, flags);
1239
		ndetected += rcu_print_task_stall(rnp);
1240 1241 1242 1243 1244 1245 1246 1247
		if (rnp->qsmask != 0) {
			for (cpu = 0; cpu <= rnp->grphi - rnp->grplo; cpu++)
				if (rnp->qsmask & (1UL << cpu)) {
					print_cpu_stall_info(rsp,
							     rnp->grplo + cpu);
					ndetected++;
				}
		}
1248
		raw_spin_unlock_irqrestore(&rnp->lock, flags);
1249
	}
1250 1251

	print_cpu_stall_info_end();
1252 1253
	for_each_possible_cpu(cpu)
		totqlen += per_cpu_ptr(rsp->rda, cpu)->qlen;
1254
	pr_cont("(detected by %d, t=%ld jiffies, g=%ld, c=%ld, q=%lu)\n",
1255
	       smp_processor_id(), (long)(jiffies - rsp->gp_start),
1256
	       (long)rsp->gpnum, (long)rsp->completed, totqlen);
1257
	if (ndetected) {
1258
		rcu_dump_cpu_stacks(rsp);
1259
	} else {
1260 1261
		if (READ_ONCE(rsp->gpnum) != gpnum ||
		    READ_ONCE(rsp->completed) == gpnum) {
1262 1263 1264
			pr_err("INFO: Stall ended before state dump start\n");
		} else {
			j = jiffies;
1265
			gpa = READ_ONCE(rsp->gp_activity);
1266
			pr_err("All QSes seen, last %s kthread activity %ld (%ld-%ld), jiffies_till_next_fqs=%ld, root ->qsmask %#lx\n",
1267
			       rsp->name, j - gpa, j, gpa,
1268 1269
			       jiffies_till_next_fqs,
			       rcu_get_root(rsp)->qsmask);
1270 1271 1272 1273
			/* In this case, the current CPU might be at fault. */
			sched_show_task(current);
		}
	}
1274

1275
	/* Complain about tasks blocking the grace period. */
1276 1277
	rcu_print_detail_task_stall(rsp);

1278 1279
	rcu_check_gp_kthread_starvation(rsp);

1280
	force_quiescent_state(rsp);  /* Kick them all. */
1281 1282 1283 1284
}

static void print_cpu_stall(struct rcu_state *rsp)
{
1285
	int cpu;
1286 1287
	unsigned long flags;
	struct rcu_node *rnp = rcu_get_root(rsp);
1288
	long totqlen = 0;
1289

1290 1291 1292 1293 1294
	/*
	 * OK, time to rat on ourselves...
	 * See Documentation/RCU/stallwarn.txt for info on how to debug
	 * RCU CPU stall warnings.
	 */
1295
	pr_err("INFO: %s self-detected stall on CPU", rsp->name);
1296 1297 1298
	print_cpu_stall_info_begin();
	print_cpu_stall_info(rsp, smp_processor_id());
	print_cpu_stall_info_end();
1299 1300
	for_each_possible_cpu(cpu)
		totqlen += per_cpu_ptr(rsp->rda, cpu)->qlen;
1301 1302 1303
	pr_cont(" (t=%lu jiffies g=%ld c=%ld q=%lu)\n",
		jiffies - rsp->gp_start,
		(long)rsp->gpnum, (long)rsp->completed, totqlen);
1304 1305 1306

	rcu_check_gp_kthread_starvation(rsp);

1307
	rcu_dump_cpu_stacks(rsp);
1308

P
Paul E. McKenney 已提交
1309
	raw_spin_lock_irqsave(&rnp->lock, flags);
1310 1311 1312
	if (ULONG_CMP_GE(jiffies, READ_ONCE(rsp->jiffies_stall)))
		WRITE_ONCE(rsp->jiffies_stall,
			   jiffies + 3 * rcu_jiffies_till_stall_check() + 3);
P
Paul E. McKenney 已提交
1313
	raw_spin_unlock_irqrestore(&rnp->lock, flags);
1314

1315 1316 1317 1318 1319 1320 1321 1322
	/*
	 * Attempt to revive the RCU machinery by forcing a context switch.
	 *
	 * A context switch would normally allow the RCU state machine to make
	 * progress and it could be we're stuck in kernel space without context
	 * switches for an entirely unreasonable amount of time.
	 */
	resched_cpu(smp_processor_id());
1323 1324 1325 1326
}

static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp)
{
1327 1328 1329
	unsigned long completed;
	unsigned long gpnum;
	unsigned long gps;
1330 1331
	unsigned long j;
	unsigned long js;
1332 1333
	struct rcu_node *rnp;

1334
	if (rcu_cpu_stall_suppress || !rcu_gp_in_progress(rsp))
1335
		return;
1336
	j = jiffies;
1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354

	/*
	 * Lots of memory barriers to reject false positives.
	 *
	 * The idea is to pick up rsp->gpnum, then rsp->jiffies_stall,
	 * then rsp->gp_start, and finally rsp->completed.  These values
	 * are updated in the opposite order with memory barriers (or
	 * equivalent) during grace-period initialization and cleanup.
	 * Now, a false positive can occur if we get an new value of
	 * rsp->gp_start and a old value of rsp->jiffies_stall.  But given
	 * the memory barriers, the only way that this can happen is if one
	 * grace period ends and another starts between these two fetches.
	 * Detect this by comparing rsp->completed with the previous fetch
	 * from rsp->gpnum.
	 *
	 * Given this check, comparisons of jiffies, rsp->jiffies_stall,
	 * and rsp->gp_start suffice to forestall false positives.
	 */
1355
	gpnum = READ_ONCE(rsp->gpnum);
1356
	smp_rmb(); /* Pick up ->gpnum first... */
1357
	js = READ_ONCE(rsp->jiffies_stall);
1358
	smp_rmb(); /* ...then ->jiffies_stall before the rest... */
1359
	gps = READ_ONCE(rsp->gp_start);
1360
	smp_rmb(); /* ...and finally ->gp_start before ->completed. */
1361
	completed = READ_ONCE(rsp->completed);
1362 1363 1364 1365
	if (ULONG_CMP_GE(completed, gpnum) ||
	    ULONG_CMP_LT(j, js) ||
	    ULONG_CMP_GE(gps, js))
		return; /* No stall or GP completed since entering function. */
1366
	rnp = rdp->mynode;
1367
	if (rcu_gp_in_progress(rsp) &&
1368
	    (READ_ONCE(rnp->qsmask) & rdp->grpmask)) {
1369 1370 1371 1372

		/* We haven't checked in, so go dump stack. */
		print_cpu_stall(rsp);

1373 1374
	} else if (rcu_gp_in_progress(rsp) &&
		   ULONG_CMP_GE(j, js + RCU_STALL_RAT_DELAY)) {
1375

1376
		/* They had a few time units to dump stack, so complain. */
1377
		print_other_cpu_stall(rsp, gpnum);
1378 1379 1380
	}
}

1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391
/**
 * rcu_cpu_stall_reset - prevent further stall warnings in current grace period
 *
 * Set the stall-warning timeout way off into the future, thus preventing
 * any RCU CPU stall-warning messages from appearing in the current set of
 * RCU grace periods.
 *
 * The caller must disable hard irqs.
 */
void rcu_cpu_stall_reset(void)
{
1392 1393 1394
	struct rcu_state *rsp;

	for_each_rcu_flavor(rsp)
1395
		WRITE_ONCE(rsp->jiffies_stall, jiffies + ULONG_MAX / 2);
1396 1397
}

1398
/*
1399 1400 1401
 * Initialize the specified rcu_data structure's default callback list
 * to empty.  The default callback list is the one that is not used by
 * no-callbacks CPUs.
1402
 */
1403
static void init_default_callback_list(struct rcu_data *rdp)
1404 1405 1406 1407 1408 1409 1410 1411
{
	int i;

	rdp->nxtlist = NULL;
	for (i = 0; i < RCU_NEXT_SIZE; i++)
		rdp->nxttail[i] = &rdp->nxtlist;
}

1412 1413 1414 1415 1416 1417 1418 1419 1420 1421
/*
 * Initialize the specified rcu_data structure's callback list to empty.
 */
static void init_callback_list(struct rcu_data *rdp)
{
	if (init_nocb_callback_list(rdp))
		return;
	init_default_callback_list(rdp);
}

1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450
/*
 * Determine the value that ->completed will have at the end of the
 * next subsequent grace period.  This is used to tag callbacks so that
 * a CPU can invoke callbacks in a timely fashion even if that CPU has
 * been dyntick-idle for an extended period with callbacks under the
 * influence of RCU_FAST_NO_HZ.
 *
 * The caller must hold rnp->lock with interrupts disabled.
 */
static unsigned long rcu_cbs_completed(struct rcu_state *rsp,
				       struct rcu_node *rnp)
{
	/*
	 * If RCU is idle, we just wait for the next grace period.
	 * But we can only be sure that RCU is idle if we are looking
	 * at the root rcu_node structure -- otherwise, a new grace
	 * period might have started, but just not yet gotten around
	 * to initializing the current non-root rcu_node structure.
	 */
	if (rcu_get_root(rsp) == rnp && rnp->gpnum == rnp->completed)
		return rnp->completed + 1;

	/*
	 * Otherwise, wait for a possible partial grace period and
	 * then the subsequent full grace period.
	 */
	return rnp->completed + 2;
}

1451 1452 1453 1454 1455
/*
 * Trace-event helper function for rcu_start_future_gp() and
 * rcu_nocb_wait_gp().
 */
static void trace_rcu_future_gp(struct rcu_node *rnp, struct rcu_data *rdp,
1456
				unsigned long c, const char *s)
1457 1458 1459 1460 1461 1462 1463 1464 1465
{
	trace_rcu_future_grace_period(rdp->rsp->name, rnp->gpnum,
				      rnp->completed, c, rnp->level,
				      rnp->grplo, rnp->grphi, s);
}

/*
 * Start some future grace period, as needed to handle newly arrived
 * callbacks.  The required future grace periods are recorded in each
1466 1467
 * rcu_node structure's ->need_future_gp field.  Returns true if there
 * is reason to awaken the grace-period kthread.
1468 1469 1470
 *
 * The caller must hold the specified rcu_node structure's ->lock.
 */
1471 1472 1473
static bool __maybe_unused
rcu_start_future_gp(struct rcu_node *rnp, struct rcu_data *rdp,
		    unsigned long *c_out)
1474 1475 1476
{
	unsigned long c;
	int i;
1477
	bool ret = false;
1478 1479 1480 1481 1482 1483 1484
	struct rcu_node *rnp_root = rcu_get_root(rdp->rsp);

	/*
	 * Pick up grace-period number for new callbacks.  If this
	 * grace period is already marked as needed, return to the caller.
	 */
	c = rcu_cbs_completed(rdp->rsp, rnp);
1485
	trace_rcu_future_gp(rnp, rdp, c, TPS("Startleaf"));
1486
	if (rnp->need_future_gp[c & 0x1]) {
1487
		trace_rcu_future_gp(rnp, rdp, c, TPS("Prestartleaf"));
1488
		goto out;
1489 1490 1491 1492 1493 1494 1495
	}

	/*
	 * If either this rcu_node structure or the root rcu_node structure
	 * believe that a grace period is in progress, then we must wait
	 * for the one following, which is in "c".  Because our request
	 * will be noticed at the end of the current grace period, we don't
1496 1497 1498 1499 1500 1501 1502
	 * need to explicitly start one.  We only do the lockless check
	 * of rnp_root's fields if the current rcu_node structure thinks
	 * there is no grace period in flight, and because we hold rnp->lock,
	 * the only possible change is when rnp_root's two fields are
	 * equal, in which case rnp_root->gpnum might be concurrently
	 * incremented.  But that is OK, as it will just result in our
	 * doing some extra useless work.
1503 1504
	 */
	if (rnp->gpnum != rnp->completed ||
1505
	    READ_ONCE(rnp_root->gpnum) != READ_ONCE(rnp_root->completed)) {
1506
		rnp->need_future_gp[c & 0x1]++;
1507
		trace_rcu_future_gp(rnp, rdp, c, TPS("Startedleaf"));
1508
		goto out;
1509 1510 1511 1512 1513 1514 1515
	}

	/*
	 * There might be no grace period in progress.  If we don't already
	 * hold it, acquire the root rcu_node structure's lock in order to
	 * start one (if needed).
	 */
1516
	if (rnp != rnp_root) {
1517
		raw_spin_lock(&rnp_root->lock);
1518 1519
		smp_mb__after_unlock_lock();
	}
1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536

	/*
	 * Get a new grace-period number.  If there really is no grace
	 * period in progress, it will be smaller than the one we obtained
	 * earlier.  Adjust callbacks as needed.  Note that even no-CBs
	 * CPUs have a ->nxtcompleted[] array, so no no-CBs checks needed.
	 */
	c = rcu_cbs_completed(rdp->rsp, rnp_root);
	for (i = RCU_DONE_TAIL; i < RCU_NEXT_TAIL; i++)
		if (ULONG_CMP_LT(c, rdp->nxtcompleted[i]))
			rdp->nxtcompleted[i] = c;

	/*
	 * If the needed for the required grace period is already
	 * recorded, trace and leave.
	 */
	if (rnp_root->need_future_gp[c & 0x1]) {
1537
		trace_rcu_future_gp(rnp, rdp, c, TPS("Prestartedroot"));
1538 1539 1540 1541 1542 1543 1544 1545
		goto unlock_out;
	}

	/* Record the need for the future grace period. */
	rnp_root->need_future_gp[c & 0x1]++;

	/* If a grace period is not already in progress, start one. */
	if (rnp_root->gpnum != rnp_root->completed) {
1546
		trace_rcu_future_gp(rnp, rdp, c, TPS("Startedleafroot"));
1547
	} else {
1548
		trace_rcu_future_gp(rnp, rdp, c, TPS("Startedroot"));
1549
		ret = rcu_start_gp_advanced(rdp->rsp, rnp_root, rdp);
1550 1551 1552 1553
	}
unlock_out:
	if (rnp != rnp_root)
		raw_spin_unlock(&rnp_root->lock);
1554 1555 1556 1557
out:
	if (c_out != NULL)
		*c_out = c;
	return ret;
1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574
}

/*
 * Clean up any old requests for the just-ended grace period.  Also return
 * whether any additional grace periods have been requested.  Also invoke
 * rcu_nocb_gp_cleanup() in order to wake up any no-callbacks kthreads
 * waiting for this grace period to complete.
 */
static int rcu_future_gp_cleanup(struct rcu_state *rsp, struct rcu_node *rnp)
{
	int c = rnp->completed;
	int needmore;
	struct rcu_data *rdp = this_cpu_ptr(rsp->rda);

	rcu_nocb_gp_cleanup(rsp, rnp);
	rnp->need_future_gp[c & 0x1] = 0;
	needmore = rnp->need_future_gp[(c + 1) & 0x1];
1575 1576
	trace_rcu_future_gp(rnp, rdp, c,
			    needmore ? TPS("CleanupMore") : TPS("Cleanup"));
1577 1578 1579
	return needmore;
}

1580 1581 1582 1583 1584 1585 1586 1587 1588 1589
/*
 * Awaken the grace-period kthread for the specified flavor of RCU.
 * Don't do a self-awaken, and don't bother awakening when there is
 * nothing for the grace-period kthread to do (as in several CPUs
 * raced to awaken, and we lost), and finally don't try to awaken
 * a kthread that has not yet been created.
 */
static void rcu_gp_kthread_wake(struct rcu_state *rsp)
{
	if (current == rsp->gp_kthread ||
1590
	    !READ_ONCE(rsp->gp_flags) ||
1591 1592 1593 1594 1595
	    !rsp->gp_kthread)
		return;
	wake_up(&rsp->gp_wq);
}

1596 1597 1598 1599 1600 1601 1602
/*
 * If there is room, assign a ->completed number to any callbacks on
 * this CPU that have not already been assigned.  Also accelerate any
 * callbacks that were previously assigned a ->completed number that has
 * since proven to be too conservative, which can happen if callbacks get
 * assigned a ->completed number while RCU is idle, but with reference to
 * a non-root rcu_node structure.  This function is idempotent, so it does
1603 1604
 * not hurt to call it repeatedly.  Returns an flag saying that we should
 * awaken the RCU grace-period kthread.
1605 1606 1607
 *
 * The caller must hold rnp->lock with interrupts disabled.
 */
1608
static bool rcu_accelerate_cbs(struct rcu_state *rsp, struct rcu_node *rnp,
1609 1610 1611 1612
			       struct rcu_data *rdp)
{
	unsigned long c;
	int i;
1613
	bool ret;
1614 1615 1616

	/* If the CPU has no callbacks, nothing to do. */
	if (!rdp->nxttail[RCU_NEXT_TAIL] || !*rdp->nxttail[RCU_DONE_TAIL])
1617
		return false;
1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645

	/*
	 * Starting from the sublist containing the callbacks most
	 * recently assigned a ->completed number and working down, find the
	 * first sublist that is not assignable to an upcoming grace period.
	 * Such a sublist has something in it (first two tests) and has
	 * a ->completed number assigned that will complete sooner than
	 * the ->completed number for newly arrived callbacks (last test).
	 *
	 * The key point is that any later sublist can be assigned the
	 * same ->completed number as the newly arrived callbacks, which
	 * means that the callbacks in any of these later sublist can be
	 * grouped into a single sublist, whether or not they have already
	 * been assigned a ->completed number.
	 */
	c = rcu_cbs_completed(rsp, rnp);
	for (i = RCU_NEXT_TAIL - 1; i > RCU_DONE_TAIL; i--)
		if (rdp->nxttail[i] != rdp->nxttail[i - 1] &&
		    !ULONG_CMP_GE(rdp->nxtcompleted[i], c))
			break;

	/*
	 * If there are no sublist for unassigned callbacks, leave.
	 * At the same time, advance "i" one sublist, so that "i" will
	 * index into the sublist where all the remaining callbacks should
	 * be grouped into.
	 */
	if (++i >= RCU_NEXT_TAIL)
1646
		return false;
1647 1648 1649 1650 1651 1652 1653 1654 1655 1656

	/*
	 * Assign all subsequent callbacks' ->completed number to the next
	 * full grace period and group them all in the sublist initially
	 * indexed by "i".
	 */
	for (; i <= RCU_NEXT_TAIL; i++) {
		rdp->nxttail[i] = rdp->nxttail[RCU_NEXT_TAIL];
		rdp->nxtcompleted[i] = c;
	}
1657
	/* Record any needed additional grace periods. */
1658
	ret = rcu_start_future_gp(rnp, rdp, NULL);
1659 1660 1661

	/* Trace depending on how much we were able to accelerate. */
	if (!*rdp->nxttail[RCU_WAIT_TAIL])
1662
		trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("AccWaitCB"));
1663
	else
1664
		trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("AccReadyCB"));
1665
	return ret;
1666 1667 1668 1669 1670 1671 1672 1673
}

/*
 * Move any callbacks whose grace period has completed to the
 * RCU_DONE_TAIL sublist, then compact the remaining sublists and
 * assign ->completed numbers to any callbacks in the RCU_NEXT_TAIL
 * sublist.  This function is idempotent, so it does not hurt to
 * invoke it repeatedly.  As long as it is not invoked -too- often...
1674
 * Returns true if the RCU grace-period kthread needs to be awakened.
1675 1676 1677
 *
 * The caller must hold rnp->lock with interrupts disabled.
 */
1678
static bool rcu_advance_cbs(struct rcu_state *rsp, struct rcu_node *rnp,
1679 1680 1681 1682 1683 1684
			    struct rcu_data *rdp)
{
	int i, j;

	/* If the CPU has no callbacks, nothing to do. */
	if (!rdp->nxttail[RCU_NEXT_TAIL] || !*rdp->nxttail[RCU_DONE_TAIL])
1685
		return false;
1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708

	/*
	 * Find all callbacks whose ->completed numbers indicate that they
	 * are ready to invoke, and put them into the RCU_DONE_TAIL sublist.
	 */
	for (i = RCU_WAIT_TAIL; i < RCU_NEXT_TAIL; i++) {
		if (ULONG_CMP_LT(rnp->completed, rdp->nxtcompleted[i]))
			break;
		rdp->nxttail[RCU_DONE_TAIL] = rdp->nxttail[i];
	}
	/* Clean up any sublist tail pointers that were misordered above. */
	for (j = RCU_WAIT_TAIL; j < i; j++)
		rdp->nxttail[j] = rdp->nxttail[RCU_DONE_TAIL];

	/* Copy down callbacks to fill in empty sublists. */
	for (j = RCU_WAIT_TAIL; i < RCU_NEXT_TAIL; i++, j++) {
		if (rdp->nxttail[j] == rdp->nxttail[RCU_NEXT_TAIL])
			break;
		rdp->nxttail[j] = rdp->nxttail[i];
		rdp->nxtcompleted[j] = rdp->nxtcompleted[i];
	}

	/* Classify any remaining callbacks. */
1709
	return rcu_accelerate_cbs(rsp, rnp, rdp);
1710 1711
}

1712
/*
1713 1714 1715
 * Update CPU-local rcu_data state to record the beginnings and ends of
 * grace periods.  The caller must hold the ->lock of the leaf rcu_node
 * structure corresponding to the current CPU, and must have irqs disabled.
1716
 * Returns true if the grace-period kthread needs to be awakened.
1717
 */
1718 1719
static bool __note_gp_changes(struct rcu_state *rsp, struct rcu_node *rnp,
			      struct rcu_data *rdp)
1720
{
1721 1722
	bool ret;

1723
	/* Handle the ends of any preceding grace periods first. */
1724
	if (rdp->completed == rnp->completed &&
1725
	    !unlikely(READ_ONCE(rdp->gpwrap))) {
1726

1727
		/* No grace period end, so just accelerate recent callbacks. */
1728
		ret = rcu_accelerate_cbs(rsp, rnp, rdp);
1729

1730 1731 1732
	} else {

		/* Advance callbacks. */
1733
		ret = rcu_advance_cbs(rsp, rnp, rdp);
1734 1735 1736

		/* Remember that we saw this grace-period completion. */
		rdp->completed = rnp->completed;
1737
		trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("cpuend"));
1738
	}
1739

1740
	if (rdp->gpnum != rnp->gpnum || unlikely(READ_ONCE(rdp->gpwrap))) {
1741 1742 1743 1744 1745 1746
		/*
		 * If the current grace period is waiting for this CPU,
		 * set up to detect a quiescent state, otherwise don't
		 * go looking for one.
		 */
		rdp->gpnum = rnp->gpnum;
1747
		trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("cpustart"));
1748
		rdp->passed_quiesce = 0;
1749
		rdp->rcu_qs_ctr_snap = __this_cpu_read(rcu_qs_ctr);
1750 1751
		rdp->qs_pending = !!(rnp->qsmask & rdp->grpmask);
		zero_cpu_stall_ticks(rdp);
1752
		WRITE_ONCE(rdp->gpwrap, false);
1753
	}
1754
	return ret;
1755 1756
}

1757
static void note_gp_changes(struct rcu_state *rsp, struct rcu_data *rdp)
1758 1759
{
	unsigned long flags;
1760
	bool needwake;
1761 1762 1763 1764
	struct rcu_node *rnp;

	local_irq_save(flags);
	rnp = rdp->mynode;
1765 1766 1767
	if ((rdp->gpnum == READ_ONCE(rnp->gpnum) &&
	     rdp->completed == READ_ONCE(rnp->completed) &&
	     !unlikely(READ_ONCE(rdp->gpwrap))) || /* w/out lock. */
1768 1769 1770 1771
	    !raw_spin_trylock(&rnp->lock)) { /* irqs already off, so later. */
		local_irq_restore(flags);
		return;
	}
1772
	smp_mb__after_unlock_lock();
1773
	needwake = __note_gp_changes(rsp, rnp, rdp);
1774
	raw_spin_unlock_irqrestore(&rnp->lock, flags);
1775 1776
	if (needwake)
		rcu_gp_kthread_wake(rsp);
1777 1778
}

1779 1780 1781 1782 1783 1784 1785
static void rcu_gp_slow(struct rcu_state *rsp, int delay)
{
	if (delay > 0 &&
	    !(rsp->gpnum % (rcu_num_nodes * PER_RCU_NODE_PERIOD * delay)))
		schedule_timeout_uninterruptible(delay);
}

1786
/*
1787
 * Initialize a new grace period.  Return 0 if no grace period required.
1788
 */
1789
static int rcu_gp_init(struct rcu_state *rsp)
1790
{
1791
	unsigned long oldmask;
1792
	struct rcu_data *rdp;
1793
	struct rcu_node *rnp = rcu_get_root(rsp);
1794

1795
	WRITE_ONCE(rsp->gp_activity, jiffies);
1796
	raw_spin_lock_irq(&rnp->lock);
1797
	smp_mb__after_unlock_lock();
1798
	if (!READ_ONCE(rsp->gp_flags)) {
1799 1800 1801 1802
		/* Spurious wakeup, tell caller to go back to sleep.  */
		raw_spin_unlock_irq(&rnp->lock);
		return 0;
	}
1803
	WRITE_ONCE(rsp->gp_flags, 0); /* Clear all flags: New grace period. */
1804

1805 1806 1807 1808 1809
	if (WARN_ON_ONCE(rcu_gp_in_progress(rsp))) {
		/*
		 * Grace period already in progress, don't start another.
		 * Not supposed to be able to happen.
		 */
1810 1811 1812 1813 1814
		raw_spin_unlock_irq(&rnp->lock);
		return 0;
	}

	/* Advance to a new grace period and initialize state. */
1815
	record_gp_stall_check_time(rsp);
1816 1817
	/* Record GP times before starting GP, hence smp_store_release(). */
	smp_store_release(&rsp->gpnum, rsp->gpnum + 1);
1818
	trace_rcu_grace_period(rsp->name, rsp->gpnum, TPS("start"));
1819 1820
	raw_spin_unlock_irq(&rnp->lock);

1821 1822 1823 1824 1825 1826 1827
	/*
	 * Apply per-leaf buffered online and offline operations to the
	 * rcu_node tree.  Note that this new grace period need not wait
	 * for subsequent online CPUs, and that quiescent-state forcing
	 * will handle subsequent offline CPUs.
	 */
	rcu_for_each_leaf_node(rsp, rnp) {
1828
		rcu_gp_slow(rsp, gp_preinit_delay);
1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869
		raw_spin_lock_irq(&rnp->lock);
		smp_mb__after_unlock_lock();
		if (rnp->qsmaskinit == rnp->qsmaskinitnext &&
		    !rnp->wait_blkd_tasks) {
			/* Nothing to do on this leaf rcu_node structure. */
			raw_spin_unlock_irq(&rnp->lock);
			continue;
		}

		/* Record old state, apply changes to ->qsmaskinit field. */
		oldmask = rnp->qsmaskinit;
		rnp->qsmaskinit = rnp->qsmaskinitnext;

		/* If zero-ness of ->qsmaskinit changed, propagate up tree. */
		if (!oldmask != !rnp->qsmaskinit) {
			if (!oldmask) /* First online CPU for this rcu_node. */
				rcu_init_new_rnp(rnp);
			else if (rcu_preempt_has_tasks(rnp)) /* blocked tasks */
				rnp->wait_blkd_tasks = true;
			else /* Last offline CPU and can propagate. */
				rcu_cleanup_dead_rnp(rnp);
		}

		/*
		 * If all waited-on tasks from prior grace period are
		 * done, and if all this rcu_node structure's CPUs are
		 * still offline, propagate up the rcu_node tree and
		 * clear ->wait_blkd_tasks.  Otherwise, if one of this
		 * rcu_node structure's CPUs has since come back online,
		 * simply clear ->wait_blkd_tasks (but rcu_cleanup_dead_rnp()
		 * checks for this, so just call it unconditionally).
		 */
		if (rnp->wait_blkd_tasks &&
		    (!rcu_preempt_has_tasks(rnp) ||
		     rnp->qsmaskinit)) {
			rnp->wait_blkd_tasks = false;
			rcu_cleanup_dead_rnp(rnp);
		}

		raw_spin_unlock_irq(&rnp->lock);
	}
1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884

	/*
	 * Set the quiescent-state-needed bits in all the rcu_node
	 * structures for all currently online CPUs in breadth-first order,
	 * starting from the root rcu_node structure, relying on the layout
	 * of the tree within the rsp->node[] array.  Note that other CPUs
	 * will access only the leaves of the hierarchy, thus seeing that no
	 * grace period is in progress, at least until the corresponding
	 * leaf node has been initialized.  In addition, we have excluded
	 * CPU-hotplug operations.
	 *
	 * The grace period cannot complete until the initialization
	 * process finishes, because this kthread handles both.
	 */
	rcu_for_each_node_breadth_first(rsp, rnp) {
1885
		rcu_gp_slow(rsp, gp_init_delay);
1886
		raw_spin_lock_irq(&rnp->lock);
1887
		smp_mb__after_unlock_lock();
1888
		rdp = this_cpu_ptr(rsp->rda);
1889 1890
		rcu_preempt_check_blocked_tasks(rnp);
		rnp->qsmask = rnp->qsmaskinit;
1891
		WRITE_ONCE(rnp->gpnum, rsp->gpnum);
1892
		if (WARN_ON_ONCE(rnp->completed != rsp->completed))
1893
			WRITE_ONCE(rnp->completed, rsp->completed);
1894
		if (rnp == rdp->mynode)
1895
			(void)__note_gp_changes(rsp, rnp, rdp);
1896 1897 1898 1899 1900
		rcu_preempt_boost_start_gp(rnp);
		trace_rcu_grace_period_init(rsp->name, rnp->gpnum,
					    rnp->level, rnp->grplo,
					    rnp->grphi, rnp->qsmask);
		raw_spin_unlock_irq(&rnp->lock);
1901
		cond_resched_rcu_qs();
1902
		WRITE_ONCE(rsp->gp_activity, jiffies);
1903
	}
1904

1905 1906
	return 1;
}
1907

1908 1909 1910
/*
 * Do one round of quiescent-state forcing.
 */
1911
static int rcu_gp_fqs(struct rcu_state *rsp, int fqs_state_in)
1912 1913
{
	int fqs_state = fqs_state_in;
1914 1915
	bool isidle = false;
	unsigned long maxj;
1916 1917
	struct rcu_node *rnp = rcu_get_root(rsp);

1918
	WRITE_ONCE(rsp->gp_activity, jiffies);
1919 1920 1921
	rsp->n_force_qs++;
	if (fqs_state == RCU_SAVE_DYNTICK) {
		/* Collect dyntick-idle snapshots. */
1922
		if (is_sysidle_rcu_state(rsp)) {
1923
			isidle = true;
1924 1925
			maxj = jiffies - ULONG_MAX / 4;
		}
1926 1927
		force_qs_rnp(rsp, dyntick_save_progress_counter,
			     &isidle, &maxj);
1928
		rcu_sysidle_report_gp(rsp, isidle, maxj);
1929 1930 1931
		fqs_state = RCU_FORCE_QS;
	} else {
		/* Handle dyntick-idle and offline CPUs. */
1932
		isidle = true;
1933
		force_qs_rnp(rsp, rcu_implicit_dynticks_qs, &isidle, &maxj);
1934 1935
	}
	/* Clear flag to prevent immediate re-entry. */
1936
	if (READ_ONCE(rsp->gp_flags) & RCU_GP_FLAG_FQS) {
1937
		raw_spin_lock_irq(&rnp->lock);
1938
		smp_mb__after_unlock_lock();
1939 1940
		WRITE_ONCE(rsp->gp_flags,
			   READ_ONCE(rsp->gp_flags) & ~RCU_GP_FLAG_FQS);
1941 1942 1943 1944 1945
		raw_spin_unlock_irq(&rnp->lock);
	}
	return fqs_state;
}

1946 1947 1948
/*
 * Clean up after the old grace period.
 */
1949
static void rcu_gp_cleanup(struct rcu_state *rsp)
1950 1951
{
	unsigned long gp_duration;
1952
	bool needgp = false;
1953
	int nocb = 0;
1954 1955
	struct rcu_data *rdp;
	struct rcu_node *rnp = rcu_get_root(rsp);
1956

1957
	WRITE_ONCE(rsp->gp_activity, jiffies);
1958
	raw_spin_lock_irq(&rnp->lock);
1959
	smp_mb__after_unlock_lock();
1960 1961 1962
	gp_duration = jiffies - rsp->gp_start;
	if (gp_duration > rsp->gp_max)
		rsp->gp_max = gp_duration;
1963

1964 1965 1966 1967 1968 1969 1970 1971
	/*
	 * We know the grace period is complete, but to everyone else
	 * it appears to still be ongoing.  But it is also the case
	 * that to everyone else it looks like there is nothing that
	 * they can do to advance the grace period.  It is therefore
	 * safe for us to drop the lock in order to mark the grace
	 * period as completed in all of the rcu_node structures.
	 */
1972
	raw_spin_unlock_irq(&rnp->lock);
1973

1974 1975 1976 1977 1978 1979 1980 1981 1982 1983
	/*
	 * Propagate new ->completed value to rcu_node structures so
	 * that other CPUs don't have to wait until the start of the next
	 * grace period to process their callbacks.  This also avoids
	 * some nasty RCU grace-period initialization races by forcing
	 * the end of the current grace period to be completely recorded in
	 * all of the rcu_node structures before the beginning of the next
	 * grace period is recorded in any of the rcu_node structures.
	 */
	rcu_for_each_node_breadth_first(rsp, rnp) {
1984
		raw_spin_lock_irq(&rnp->lock);
1985
		smp_mb__after_unlock_lock();
1986 1987
		WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp));
		WARN_ON_ONCE(rnp->qsmask);
1988
		WRITE_ONCE(rnp->completed, rsp->gpnum);
1989 1990
		rdp = this_cpu_ptr(rsp->rda);
		if (rnp == rdp->mynode)
1991
			needgp = __note_gp_changes(rsp, rnp, rdp) || needgp;
1992
		/* smp_mb() provided by prior unlock-lock pair. */
1993
		nocb += rcu_future_gp_cleanup(rsp, rnp);
1994
		raw_spin_unlock_irq(&rnp->lock);
1995
		cond_resched_rcu_qs();
1996
		WRITE_ONCE(rsp->gp_activity, jiffies);
1997
		rcu_gp_slow(rsp, gp_cleanup_delay);
1998
	}
1999 2000
	rnp = rcu_get_root(rsp);
	raw_spin_lock_irq(&rnp->lock);
2001
	smp_mb__after_unlock_lock(); /* Order GP before ->completed update. */
2002
	rcu_nocb_gp_set(rnp, nocb);
2003

2004
	/* Declare grace period done. */
2005
	WRITE_ONCE(rsp->completed, rsp->gpnum);
2006
	trace_rcu_grace_period(rsp->name, rsp->completed, TPS("end"));
2007
	rsp->fqs_state = RCU_GP_IDLE;
2008
	rdp = this_cpu_ptr(rsp->rda);
2009 2010 2011
	/* Advance CBs to reduce false positives below. */
	needgp = rcu_advance_cbs(rsp, rnp, rdp) || needgp;
	if (needgp || cpu_needs_another_gp(rsp, rdp)) {
2012
		WRITE_ONCE(rsp->gp_flags, RCU_GP_FLAG_INIT);
2013
		trace_rcu_grace_period(rsp->name,
2014
				       READ_ONCE(rsp->gpnum),
2015 2016
				       TPS("newreq"));
	}
2017 2018 2019 2020 2021 2022 2023 2024
	raw_spin_unlock_irq(&rnp->lock);
}

/*
 * Body of kthread that handles grace periods.
 */
static int __noreturn rcu_gp_kthread(void *arg)
{
2025
	int fqs_state;
2026
	int gf;
2027
	unsigned long j;
2028
	int ret;
2029 2030 2031
	struct rcu_state *rsp = arg;
	struct rcu_node *rnp = rcu_get_root(rsp);

2032
	rcu_bind_gp_kthread();
2033 2034 2035 2036
	for (;;) {

		/* Handle grace-period start. */
		for (;;) {
2037
			trace_rcu_grace_period(rsp->name,
2038
					       READ_ONCE(rsp->gpnum),
2039
					       TPS("reqwait"));
2040
			rsp->gp_state = RCU_GP_WAIT_GPS;
2041
			wait_event_interruptible(rsp->gp_wq,
2042
						 READ_ONCE(rsp->gp_flags) &
2043
						 RCU_GP_FLAG_INIT);
2044
			/* Locking provides needed memory barrier. */
2045
			if (rcu_gp_init(rsp))
2046
				break;
2047
			cond_resched_rcu_qs();
2048
			WRITE_ONCE(rsp->gp_activity, jiffies);
2049
			WARN_ON(signal_pending(current));
2050
			trace_rcu_grace_period(rsp->name,
2051
					       READ_ONCE(rsp->gpnum),
2052
					       TPS("reqwaitsig"));
2053
		}
2054

2055 2056
		/* Handle quiescent-state forcing. */
		fqs_state = RCU_SAVE_DYNTICK;
2057 2058 2059 2060 2061
		j = jiffies_till_first_fqs;
		if (j > HZ) {
			j = HZ;
			jiffies_till_first_fqs = HZ;
		}
2062
		ret = 0;
2063
		for (;;) {
2064 2065
			if (!ret)
				rsp->jiffies_force_qs = jiffies + j;
2066
			trace_rcu_grace_period(rsp->name,
2067
					       READ_ONCE(rsp->gpnum),
2068
					       TPS("fqswait"));
2069
			rsp->gp_state = RCU_GP_WAIT_FQS;
2070
			ret = wait_event_interruptible_timeout(rsp->gp_wq,
2071
					((gf = READ_ONCE(rsp->gp_flags)) &
2072
					 RCU_GP_FLAG_FQS) ||
2073
					(!READ_ONCE(rnp->qsmask) &&
2074
					 !rcu_preempt_blocked_readers_cgp(rnp)),
2075
					j);
2076
			/* Locking provides needed memory barriers. */
2077
			/* If grace period done, leave loop. */
2078
			if (!READ_ONCE(rnp->qsmask) &&
2079
			    !rcu_preempt_blocked_readers_cgp(rnp))
2080
				break;
2081
			/* If time for quiescent-state forcing, do it. */
2082 2083
			if (ULONG_CMP_GE(jiffies, rsp->jiffies_force_qs) ||
			    (gf & RCU_GP_FLAG_FQS)) {
2084
				trace_rcu_grace_period(rsp->name,
2085
						       READ_ONCE(rsp->gpnum),
2086
						       TPS("fqsstart"));
2087
				fqs_state = rcu_gp_fqs(rsp, fqs_state);
2088
				trace_rcu_grace_period(rsp->name,
2089
						       READ_ONCE(rsp->gpnum),
2090
						       TPS("fqsend"));
2091
				cond_resched_rcu_qs();
2092
				WRITE_ONCE(rsp->gp_activity, jiffies);
2093 2094
			} else {
				/* Deal with stray signal. */
2095
				cond_resched_rcu_qs();
2096
				WRITE_ONCE(rsp->gp_activity, jiffies);
2097
				WARN_ON(signal_pending(current));
2098
				trace_rcu_grace_period(rsp->name,
2099
						       READ_ONCE(rsp->gpnum),
2100
						       TPS("fqswaitsig"));
2101
			}
2102 2103 2104 2105 2106 2107 2108 2109
			j = jiffies_till_next_fqs;
			if (j > HZ) {
				j = HZ;
				jiffies_till_next_fqs = HZ;
			} else if (j < 1) {
				j = 1;
				jiffies_till_next_fqs = 1;
			}
2110
		}
2111 2112 2113

		/* Handle grace-period end. */
		rcu_gp_cleanup(rsp);
2114 2115 2116
	}
}

2117 2118 2119
/*
 * Start a new RCU grace period if warranted, re-initializing the hierarchy
 * in preparation for detecting the next grace period.  The caller must hold
2120
 * the root node's ->lock and hard irqs must be disabled.
2121 2122 2123 2124
 *
 * Note that it is legal for a dying CPU (which is marked as offline) to
 * invoke this function.  This can happen when the dying CPU reports its
 * quiescent state.
2125 2126
 *
 * Returns true if the grace-period kthread must be awakened.
2127
 */
2128
static bool
2129 2130
rcu_start_gp_advanced(struct rcu_state *rsp, struct rcu_node *rnp,
		      struct rcu_data *rdp)
2131
{
2132
	if (!rsp->gp_kthread || !cpu_needs_another_gp(rsp, rdp)) {
2133
		/*
2134
		 * Either we have not yet spawned the grace-period
2135 2136
		 * task, this CPU does not need another grace period,
		 * or a grace period is already in progress.
2137
		 * Either way, don't start a new grace period.
2138
		 */
2139
		return false;
2140
	}
2141 2142
	WRITE_ONCE(rsp->gp_flags, RCU_GP_FLAG_INIT);
	trace_rcu_grace_period(rsp->name, READ_ONCE(rsp->gpnum),
2143
			       TPS("newreq"));
2144

2145 2146
	/*
	 * We can't do wakeups while holding the rnp->lock, as that
2147
	 * could cause possible deadlocks with the rq->lock. Defer
2148
	 * the wakeup to our caller.
2149
	 */
2150
	return true;
2151 2152
}

2153 2154 2155 2156 2157 2158
/*
 * Similar to rcu_start_gp_advanced(), but also advance the calling CPU's
 * callbacks.  Note that rcu_start_gp_advanced() cannot do this because it
 * is invoked indirectly from rcu_advance_cbs(), which would result in
 * endless recursion -- or would do so if it wasn't for the self-deadlock
 * that is encountered beforehand.
2159 2160
 *
 * Returns true if the grace-period kthread needs to be awakened.
2161
 */
2162
static bool rcu_start_gp(struct rcu_state *rsp)
2163 2164 2165
{
	struct rcu_data *rdp = this_cpu_ptr(rsp->rda);
	struct rcu_node *rnp = rcu_get_root(rsp);
2166
	bool ret = false;
2167 2168 2169 2170 2171 2172 2173 2174 2175

	/*
	 * If there is no grace period in progress right now, any
	 * callbacks we have up to this point will be satisfied by the
	 * next grace period.  Also, advancing the callbacks reduces the
	 * probability of false positives from cpu_needs_another_gp()
	 * resulting in pointless grace periods.  So, advance callbacks
	 * then start the grace period!
	 */
2176 2177 2178
	ret = rcu_advance_cbs(rsp, rnp, rdp) || ret;
	ret = rcu_start_gp_advanced(rsp, rnp, rdp) || ret;
	return ret;
2179 2180
}

2181
/*
P
Paul E. McKenney 已提交
2182 2183 2184
 * Report a full set of quiescent states to the specified rcu_state
 * data structure.  This involves cleaning up after the prior grace
 * period and letting rcu_start_gp() start up the next grace period
2185 2186
 * if one is needed.  Note that the caller must hold rnp->lock, which
 * is released before return.
2187
 */
P
Paul E. McKenney 已提交
2188
static void rcu_report_qs_rsp(struct rcu_state *rsp, unsigned long flags)
2189
	__releases(rcu_get_root(rsp)->lock)
2190
{
2191
	WARN_ON_ONCE(!rcu_gp_in_progress(rsp));
2192
	WRITE_ONCE(rsp->gp_flags, READ_ONCE(rsp->gp_flags) | RCU_GP_FLAG_FQS);
2193
	raw_spin_unlock_irqrestore(&rcu_get_root(rsp)->lock, flags);
2194
	rcu_gp_kthread_wake(rsp);
2195 2196
}

2197
/*
P
Paul E. McKenney 已提交
2198 2199 2200
 * Similar to rcu_report_qs_rdp(), for which it is a helper function.
 * Allows quiescent states for a group of CPUs to be reported at one go
 * to the specified rcu_node structure, though all the CPUs in the group
2201 2202 2203 2204 2205
 * must be represented by the same rcu_node structure (which need not be a
 * leaf rcu_node structure, though it often will be).  The gps parameter
 * is the grace-period snapshot, which means that the quiescent states
 * are valid only if rnp->gpnum is equal to gps.  That structure's lock
 * must be held upon entry, and it is released before return.
2206 2207
 */
static void
P
Paul E. McKenney 已提交
2208
rcu_report_qs_rnp(unsigned long mask, struct rcu_state *rsp,
2209
		  struct rcu_node *rnp, unsigned long gps, unsigned long flags)
2210 2211
	__releases(rnp->lock)
{
2212
	unsigned long oldmask = 0;
2213 2214
	struct rcu_node *rnp_c;

2215 2216
	/* Walk up the rcu_node hierarchy. */
	for (;;) {
2217
		if (!(rnp->qsmask & mask) || rnp->gpnum != gps) {
2218

2219 2220 2221 2222
			/*
			 * Our bit has already been cleared, or the
			 * relevant grace period is already over, so done.
			 */
P
Paul E. McKenney 已提交
2223
			raw_spin_unlock_irqrestore(&rnp->lock, flags);
2224 2225
			return;
		}
2226
		WARN_ON_ONCE(oldmask); /* Any child must be all zeroed! */
2227
		rnp->qsmask &= ~mask;
2228 2229 2230 2231
		trace_rcu_quiescent_state_report(rsp->name, rnp->gpnum,
						 mask, rnp->qsmask, rnp->level,
						 rnp->grplo, rnp->grphi,
						 !!rnp->gp_tasks);
2232
		if (rnp->qsmask != 0 || rcu_preempt_blocked_readers_cgp(rnp)) {
2233 2234

			/* Other bits still set at this level, so done. */
P
Paul E. McKenney 已提交
2235
			raw_spin_unlock_irqrestore(&rnp->lock, flags);
2236 2237 2238 2239 2240 2241 2242 2243 2244
			return;
		}
		mask = rnp->grpmask;
		if (rnp->parent == NULL) {

			/* No more levels.  Exit loop holding root lock. */

			break;
		}
P
Paul E. McKenney 已提交
2245
		raw_spin_unlock_irqrestore(&rnp->lock, flags);
2246
		rnp_c = rnp;
2247
		rnp = rnp->parent;
P
Paul E. McKenney 已提交
2248
		raw_spin_lock_irqsave(&rnp->lock, flags);
2249
		smp_mb__after_unlock_lock();
2250
		oldmask = rnp_c->qsmask;
2251 2252 2253 2254
	}

	/*
	 * Get here if we are the last CPU to pass through a quiescent
P
Paul E. McKenney 已提交
2255
	 * state for this grace period.  Invoke rcu_report_qs_rsp()
2256
	 * to clean up and start the next grace period if one is needed.
2257
	 */
P
Paul E. McKenney 已提交
2258
	rcu_report_qs_rsp(rsp, flags); /* releases rnp->lock. */
2259 2260
}

2261 2262 2263 2264 2265 2266 2267
/*
 * Record a quiescent state for all tasks that were previously queued
 * on the specified rcu_node structure and that were blocking the current
 * RCU grace period.  The caller must hold the specified rnp->lock with
 * irqs disabled, and this lock is released upon return, but irqs remain
 * disabled.
 */
2268
static void rcu_report_unblock_qs_rnp(struct rcu_state *rsp,
2269 2270 2271
				      struct rcu_node *rnp, unsigned long flags)
	__releases(rnp->lock)
{
2272
	unsigned long gps;
2273 2274 2275
	unsigned long mask;
	struct rcu_node *rnp_p;

2276 2277
	if (rcu_state_p == &rcu_sched_state || rsp != rcu_state_p ||
	    rnp->qsmask != 0 || rcu_preempt_blocked_readers_cgp(rnp)) {
2278 2279 2280 2281 2282 2283 2284
		raw_spin_unlock_irqrestore(&rnp->lock, flags);
		return;  /* Still need more quiescent states! */
	}

	rnp_p = rnp->parent;
	if (rnp_p == NULL) {
		/*
2285 2286
		 * Only one rcu_node structure in the tree, so don't
		 * try to report up to its nonexistent parent!
2287 2288 2289 2290 2291
		 */
		rcu_report_qs_rsp(rsp, flags);
		return;
	}

2292 2293
	/* Report up the rest of the hierarchy, tracking current ->gpnum. */
	gps = rnp->gpnum;
2294 2295 2296 2297
	mask = rnp->grpmask;
	raw_spin_unlock(&rnp->lock);	/* irqs remain disabled. */
	raw_spin_lock(&rnp_p->lock);	/* irqs already disabled. */
	smp_mb__after_unlock_lock();
2298
	rcu_report_qs_rnp(mask, rsp, rnp_p, gps, flags);
2299 2300
}

2301
/*
P
Paul E. McKenney 已提交
2302 2303 2304 2305 2306 2307 2308
 * Record a quiescent state for the specified CPU to that CPU's rcu_data
 * structure.  This must be either called from the specified CPU, or
 * called when the specified CPU is known to be offline (and when it is
 * also known that no other CPU is concurrently trying to help the offline
 * CPU).  The lastcomp argument is used to make sure we are still in the
 * grace period of interest.  We don't want to end the current grace period
 * based on quiescent states detected in an earlier grace period!
2309 2310
 */
static void
2311
rcu_report_qs_rdp(int cpu, struct rcu_state *rsp, struct rcu_data *rdp)
2312 2313 2314
{
	unsigned long flags;
	unsigned long mask;
2315
	bool needwake;
2316 2317 2318
	struct rcu_node *rnp;

	rnp = rdp->mynode;
P
Paul E. McKenney 已提交
2319
	raw_spin_lock_irqsave(&rnp->lock, flags);
2320
	smp_mb__after_unlock_lock();
2321 2322 2323 2324
	if ((rdp->passed_quiesce == 0 &&
	     rdp->rcu_qs_ctr_snap == __this_cpu_read(rcu_qs_ctr)) ||
	    rdp->gpnum != rnp->gpnum || rnp->completed == rnp->gpnum ||
	    rdp->gpwrap) {
2325 2326

		/*
2327 2328 2329 2330
		 * The grace period in which this quiescent state was
		 * recorded has ended, so don't report it upwards.
		 * We will instead need a new quiescent state that lies
		 * within the current grace period.
2331
		 */
2332
		rdp->passed_quiesce = 0;	/* need qs for new gp. */
2333
		rdp->rcu_qs_ctr_snap = __this_cpu_read(rcu_qs_ctr);
P
Paul E. McKenney 已提交
2334
		raw_spin_unlock_irqrestore(&rnp->lock, flags);
2335 2336 2337 2338
		return;
	}
	mask = rdp->grpmask;
	if ((rnp->qsmask & mask) == 0) {
P
Paul E. McKenney 已提交
2339
		raw_spin_unlock_irqrestore(&rnp->lock, flags);
2340 2341 2342 2343 2344 2345 2346
	} else {
		rdp->qs_pending = 0;

		/*
		 * This GP can't end until cpu checks in, so all of our
		 * callbacks can be processed during the next GP.
		 */
2347
		needwake = rcu_accelerate_cbs(rsp, rnp, rdp);
2348

2349 2350
		rcu_report_qs_rnp(mask, rsp, rnp, rnp->gpnum, flags);
		/* ^^^ Released rnp->lock */
2351 2352
		if (needwake)
			rcu_gp_kthread_wake(rsp);
2353 2354 2355 2356 2357 2358 2359 2360 2361 2362 2363 2364
	}
}

/*
 * Check to see if there is a new grace period of which this CPU
 * is not yet aware, and if so, set up local rcu_data state for it.
 * Otherwise, see if this CPU has just passed through its first
 * quiescent state for this grace period, and record that fact if so.
 */
static void
rcu_check_quiescent_state(struct rcu_state *rsp, struct rcu_data *rdp)
{
2365 2366
	/* Check for grace-period ends and beginnings. */
	note_gp_changes(rsp, rdp);
2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378

	/*
	 * Does this CPU still need to do its part for current grace period?
	 * If no, return and let the other CPUs do their part as well.
	 */
	if (!rdp->qs_pending)
		return;

	/*
	 * Was there a quiescent state since the beginning of the grace
	 * period? If no, then exit and wait for the next call.
	 */
2379 2380
	if (!rdp->passed_quiesce &&
	    rdp->rcu_qs_ctr_snap == __this_cpu_read(rcu_qs_ctr))
2381 2382
		return;

P
Paul E. McKenney 已提交
2383 2384 2385 2386
	/*
	 * Tell RCU we are done (but rcu_report_qs_rdp() will be the
	 * judge of that).
	 */
2387
	rcu_report_qs_rdp(rdp->cpu, rsp, rdp);
2388 2389
}

2390
/*
2391 2392
 * Send the specified CPU's RCU callbacks to the orphanage.  The
 * specified CPU must be offline, and the caller must hold the
2393
 * ->orphan_lock.
2394
 */
2395 2396 2397
static void
rcu_send_cbs_to_orphanage(int cpu, struct rcu_state *rsp,
			  struct rcu_node *rnp, struct rcu_data *rdp)
2398
{
P
Paul E. McKenney 已提交
2399
	/* No-CBs CPUs do not have orphanable callbacks. */
2400
	if (!IS_ENABLED(CONFIG_HOTPLUG_CPU) || rcu_is_nocb_cpu(rdp->cpu))
P
Paul E. McKenney 已提交
2401 2402
		return;

2403 2404
	/*
	 * Orphan the callbacks.  First adjust the counts.  This is safe
2405 2406
	 * because _rcu_barrier() excludes CPU-hotplug operations, so it
	 * cannot be running now.  Thus no memory barrier is required.
2407
	 */
2408
	if (rdp->nxtlist != NULL) {
2409 2410 2411
		rsp->qlen_lazy += rdp->qlen_lazy;
		rsp->qlen += rdp->qlen;
		rdp->n_cbs_orphaned += rdp->qlen;
2412
		rdp->qlen_lazy = 0;
2413
		WRITE_ONCE(rdp->qlen, 0);
2414 2415 2416
	}

	/*
2417 2418 2419 2420 2421 2422 2423
	 * Next, move those callbacks still needing a grace period to
	 * the orphanage, where some other CPU will pick them up.
	 * Some of the callbacks might have gone partway through a grace
	 * period, but that is too bad.  They get to start over because we
	 * cannot assume that grace periods are synchronized across CPUs.
	 * We don't bother updating the ->nxttail[] array yet, instead
	 * we just reset the whole thing later on.
2424
	 */
2425 2426 2427 2428
	if (*rdp->nxttail[RCU_DONE_TAIL] != NULL) {
		*rsp->orphan_nxttail = *rdp->nxttail[RCU_DONE_TAIL];
		rsp->orphan_nxttail = rdp->nxttail[RCU_NEXT_TAIL];
		*rdp->nxttail[RCU_DONE_TAIL] = NULL;
2429 2430 2431
	}

	/*
2432 2433 2434
	 * Then move the ready-to-invoke callbacks to the orphanage,
	 * where some other CPU will pick them up.  These will not be
	 * required to pass though another grace period: They are done.
2435
	 */
2436
	if (rdp->nxtlist != NULL) {
2437 2438
		*rsp->orphan_donetail = rdp->nxtlist;
		rsp->orphan_donetail = rdp->nxttail[RCU_DONE_TAIL];
2439
	}
2440

2441 2442 2443 2444
	/*
	 * Finally, initialize the rcu_data structure's list to empty and
	 * disallow further callbacks on this CPU.
	 */
2445
	init_callback_list(rdp);
2446
	rdp->nxttail[RCU_NEXT_TAIL] = NULL;
2447 2448 2449 2450
}

/*
 * Adopt the RCU callbacks from the specified rcu_state structure's
2451
 * orphanage.  The caller must hold the ->orphan_lock.
2452
 */
2453
static void rcu_adopt_orphan_cbs(struct rcu_state *rsp, unsigned long flags)
2454 2455
{
	int i;
2456
	struct rcu_data *rdp = raw_cpu_ptr(rsp->rda);
2457

P
Paul E. McKenney 已提交
2458
	/* No-CBs CPUs are handled specially. */
2459 2460
	if (!IS_ENABLED(CONFIG_HOTPLUG_CPU) ||
	    rcu_nocb_adopt_orphan_cbs(rsp, rdp, flags))
P
Paul E. McKenney 已提交
2461 2462
		return;

2463 2464 2465 2466
	/* Do the accounting first. */
	rdp->qlen_lazy += rsp->qlen_lazy;
	rdp->qlen += rsp->qlen;
	rdp->n_cbs_adopted += rsp->qlen;
2467 2468
	if (rsp->qlen_lazy != rsp->qlen)
		rcu_idle_count_callbacks_posted();
2469 2470 2471 2472 2473 2474 2475 2476 2477 2478 2479 2480 2481 2482 2483 2484 2485 2486 2487 2488 2489 2490 2491 2492 2493 2494 2495 2496 2497 2498 2499 2500 2501 2502 2503 2504 2505 2506
	rsp->qlen_lazy = 0;
	rsp->qlen = 0;

	/*
	 * We do not need a memory barrier here because the only way we
	 * can get here if there is an rcu_barrier() in flight is if
	 * we are the task doing the rcu_barrier().
	 */

	/* First adopt the ready-to-invoke callbacks. */
	if (rsp->orphan_donelist != NULL) {
		*rsp->orphan_donetail = *rdp->nxttail[RCU_DONE_TAIL];
		*rdp->nxttail[RCU_DONE_TAIL] = rsp->orphan_donelist;
		for (i = RCU_NEXT_SIZE - 1; i >= RCU_DONE_TAIL; i--)
			if (rdp->nxttail[i] == rdp->nxttail[RCU_DONE_TAIL])
				rdp->nxttail[i] = rsp->orphan_donetail;
		rsp->orphan_donelist = NULL;
		rsp->orphan_donetail = &rsp->orphan_donelist;
	}

	/* And then adopt the callbacks that still need a grace period. */
	if (rsp->orphan_nxtlist != NULL) {
		*rdp->nxttail[RCU_NEXT_TAIL] = rsp->orphan_nxtlist;
		rdp->nxttail[RCU_NEXT_TAIL] = rsp->orphan_nxttail;
		rsp->orphan_nxtlist = NULL;
		rsp->orphan_nxttail = &rsp->orphan_nxtlist;
	}
}

/*
 * Trace the fact that this CPU is going offline.
 */
static void rcu_cleanup_dying_cpu(struct rcu_state *rsp)
{
	RCU_TRACE(unsigned long mask);
	RCU_TRACE(struct rcu_data *rdp = this_cpu_ptr(rsp->rda));
	RCU_TRACE(struct rcu_node *rnp = rdp->mynode);

2507 2508 2509
	if (!IS_ENABLED(CONFIG_HOTPLUG_CPU))
		return;

2510
	RCU_TRACE(mask = rdp->grpmask);
2511 2512
	trace_rcu_grace_period(rsp->name,
			       rnp->gpnum + 1 - !!(rnp->qsmask & mask),
2513
			       TPS("cpuofl"));
2514 2515
}

2516 2517 2518 2519 2520 2521 2522 2523 2524 2525 2526 2527 2528 2529 2530 2531 2532 2533 2534 2535 2536 2537
/*
 * All CPUs for the specified rcu_node structure have gone offline,
 * and all tasks that were preempted within an RCU read-side critical
 * section while running on one of those CPUs have since exited their RCU
 * read-side critical section.  Some other CPU is reporting this fact with
 * the specified rcu_node structure's ->lock held and interrupts disabled.
 * This function therefore goes up the tree of rcu_node structures,
 * clearing the corresponding bits in the ->qsmaskinit fields.  Note that
 * the leaf rcu_node structure's ->qsmaskinit field has already been
 * updated
 *
 * This function does check that the specified rcu_node structure has
 * all CPUs offline and no blocked tasks, so it is OK to invoke it
 * prematurely.  That said, invoking it after the fact will cost you
 * a needless lock acquisition.  So once it has done its work, don't
 * invoke it again.
 */
static void rcu_cleanup_dead_rnp(struct rcu_node *rnp_leaf)
{
	long mask;
	struct rcu_node *rnp = rnp_leaf;

2538 2539
	if (!IS_ENABLED(CONFIG_HOTPLUG_CPU) ||
	    rnp->qsmaskinit || rcu_preempt_has_tasks(rnp))
2540 2541 2542 2543 2544 2545 2546 2547 2548
		return;
	for (;;) {
		mask = rnp->grpmask;
		rnp = rnp->parent;
		if (!rnp)
			break;
		raw_spin_lock(&rnp->lock); /* irqs already disabled. */
		smp_mb__after_unlock_lock(); /* GP memory ordering. */
		rnp->qsmaskinit &= ~mask;
2549
		rnp->qsmask &= ~mask;
2550 2551 2552 2553 2554 2555 2556 2557
		if (rnp->qsmaskinit) {
			raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */
			return;
		}
		raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */
	}
}

2558 2559 2560 2561 2562 2563 2564 2565 2566 2567 2568 2569
/*
 * The CPU is exiting the idle loop into the arch_cpu_idle_dead()
 * function.  We now remove it from the rcu_node tree's ->qsmaskinit
 * bit masks.
 */
static void rcu_cleanup_dying_idle_cpu(int cpu, struct rcu_state *rsp)
{
	unsigned long flags;
	unsigned long mask;
	struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
	struct rcu_node *rnp = rdp->mynode;  /* Outgoing CPU's rdp & rnp. */

2570 2571 2572
	if (!IS_ENABLED(CONFIG_HOTPLUG_CPU))
		return;

2573 2574 2575 2576 2577 2578 2579 2580
	/* Remove outgoing CPU from mask in the leaf rcu_node structure. */
	mask = rdp->grpmask;
	raw_spin_lock_irqsave(&rnp->lock, flags);
	smp_mb__after_unlock_lock();	/* Enforce GP memory-order guarantee. */
	rnp->qsmaskinitnext &= ~mask;
	raw_spin_unlock_irqrestore(&rnp->lock, flags);
}

2581
/*
2582
 * The CPU has been completely removed, and some other CPU is reporting
2583 2584
 * this fact from process context.  Do the remainder of the cleanup,
 * including orphaning the outgoing CPU's RCU callbacks, and also
2585 2586
 * adopting them.  There can only be one CPU hotplug operation at a time,
 * so no other CPU can be attempting to update rcu_cpu_kthread_task.
2587
 */
2588
static void rcu_cleanup_dead_cpu(int cpu, struct rcu_state *rsp)
2589
{
2590
	unsigned long flags;
2591
	struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
2592
	struct rcu_node *rnp = rdp->mynode;  /* Outgoing CPU's rdp & rnp. */
2593

2594 2595 2596
	if (!IS_ENABLED(CONFIG_HOTPLUG_CPU))
		return;

2597
	/* Adjust any no-longer-needed kthreads. */
T
Thomas Gleixner 已提交
2598
	rcu_boost_kthread_setaffinity(rnp, -1);
2599

2600
	/* Orphan the dead CPU's callbacks, and adopt them if appropriate. */
2601
	raw_spin_lock_irqsave(&rsp->orphan_lock, flags);
2602
	rcu_send_cbs_to_orphanage(cpu, rsp, rnp, rdp);
2603
	rcu_adopt_orphan_cbs(rsp, flags);
2604
	raw_spin_unlock_irqrestore(&rsp->orphan_lock, flags);
2605

2606 2607 2608
	WARN_ONCE(rdp->qlen != 0 || rdp->nxtlist != NULL,
		  "rcu_cleanup_dead_cpu: Callbacks on offline CPU %d: qlen=%lu, nxtlist=%p\n",
		  cpu, rdp->qlen, rdp->nxtlist);
2609 2610 2611 2612 2613 2614
}

/*
 * Invoke any RCU callbacks that have made it to the end of their grace
 * period.  Thottle as specified by rdp->blimit.
 */
2615
static void rcu_do_batch(struct rcu_state *rsp, struct rcu_data *rdp)
2616 2617 2618
{
	unsigned long flags;
	struct rcu_head *next, *list, **tail;
E
Eric Dumazet 已提交
2619 2620
	long bl, count, count_lazy;
	int i;
2621

2622
	/* If no callbacks are ready, just return. */
2623
	if (!cpu_has_callbacks_ready_to_invoke(rdp)) {
2624
		trace_rcu_batch_start(rsp->name, rdp->qlen_lazy, rdp->qlen, 0);
2625
		trace_rcu_batch_end(rsp->name, 0, !!READ_ONCE(rdp->nxtlist),
2626 2627
				    need_resched(), is_idle_task(current),
				    rcu_is_callbacks_kthread());
2628
		return;
2629
	}
2630 2631 2632 2633 2634 2635

	/*
	 * Extract the list of ready callbacks, disabling to prevent
	 * races with call_rcu() from interrupt handlers.
	 */
	local_irq_save(flags);
2636
	WARN_ON_ONCE(cpu_is_offline(smp_processor_id()));
2637
	bl = rdp->blimit;
2638
	trace_rcu_batch_start(rsp->name, rdp->qlen_lazy, rdp->qlen, bl);
2639 2640 2641 2642
	list = rdp->nxtlist;
	rdp->nxtlist = *rdp->nxttail[RCU_DONE_TAIL];
	*rdp->nxttail[RCU_DONE_TAIL] = NULL;
	tail = rdp->nxttail[RCU_DONE_TAIL];
2643 2644 2645
	for (i = RCU_NEXT_SIZE - 1; i >= 0; i--)
		if (rdp->nxttail[i] == rdp->nxttail[RCU_DONE_TAIL])
			rdp->nxttail[i] = &rdp->nxtlist;
2646 2647 2648
	local_irq_restore(flags);

	/* Invoke callbacks. */
2649
	count = count_lazy = 0;
2650 2651 2652
	while (list) {
		next = list->next;
		prefetch(next);
2653
		debug_rcu_head_unqueue(list);
2654 2655
		if (__rcu_reclaim(rsp->name, list))
			count_lazy++;
2656
		list = next;
2657 2658 2659 2660
		/* Stop only if limit reached and CPU has something to do. */
		if (++count >= bl &&
		    (need_resched() ||
		     (!is_idle_task(current) && !rcu_is_callbacks_kthread())))
2661 2662 2663 2664
			break;
	}

	local_irq_save(flags);
2665 2666 2667
	trace_rcu_batch_end(rsp->name, count, !!list, need_resched(),
			    is_idle_task(current),
			    rcu_is_callbacks_kthread());
2668 2669 2670 2671 2672

	/* Update count, and requeue any remaining callbacks. */
	if (list != NULL) {
		*tail = rdp->nxtlist;
		rdp->nxtlist = list;
2673 2674 2675
		for (i = 0; i < RCU_NEXT_SIZE; i++)
			if (&rdp->nxtlist == rdp->nxttail[i])
				rdp->nxttail[i] = tail;
2676 2677 2678
			else
				break;
	}
2679 2680
	smp_mb(); /* List handling before counting for rcu_barrier(). */
	rdp->qlen_lazy -= count_lazy;
2681
	WRITE_ONCE(rdp->qlen, rdp->qlen - count);
2682
	rdp->n_cbs_invoked += count;
2683 2684 2685 2686 2687

	/* Reinstate batch limit if we have worked down the excess. */
	if (rdp->blimit == LONG_MAX && rdp->qlen <= qlowmark)
		rdp->blimit = blimit;

2688 2689 2690 2691 2692 2693
	/* Reset ->qlen_last_fqs_check trigger if enough CBs have drained. */
	if (rdp->qlen == 0 && rdp->qlen_last_fqs_check != 0) {
		rdp->qlen_last_fqs_check = 0;
		rdp->n_force_qs_snap = rsp->n_force_qs;
	} else if (rdp->qlen < rdp->qlen_last_fqs_check - qhimark)
		rdp->qlen_last_fqs_check = rdp->qlen;
2694
	WARN_ON_ONCE((rdp->nxtlist == NULL) != (rdp->qlen == 0));
2695

2696 2697
	local_irq_restore(flags);

2698
	/* Re-invoke RCU core processing if there are callbacks remaining. */
2699
	if (cpu_has_callbacks_ready_to_invoke(rdp))
2700
		invoke_rcu_core();
2701 2702 2703 2704 2705
}

/*
 * Check to see if this CPU is in a non-context-switch quiescent state
 * (user mode or idle loop for rcu, non-softirq execution for rcu_bh).
2706
 * Also schedule RCU core processing.
2707
 *
2708
 * This function must be called from hardirq context.  It is normally
2709 2710 2711
 * invoked from the scheduling-clock interrupt.  If rcu_pending returns
 * false, there is no point in invoking rcu_check_callbacks().
 */
2712
void rcu_check_callbacks(int user)
2713
{
2714
	trace_rcu_utilization(TPS("Start scheduler-tick"));
2715
	increment_cpu_stall_ticks();
2716
	if (user || rcu_is_cpu_rrupt_from_idle()) {
2717 2718 2719 2720 2721

		/*
		 * Get here if this CPU took its interrupt from user
		 * mode or from the idle loop, and if this is not a
		 * nested interrupt.  In this case, the CPU is in
2722
		 * a quiescent state, so note it.
2723 2724
		 *
		 * No memory barrier is required here because both
2725 2726 2727
		 * rcu_sched_qs() and rcu_bh_qs() reference only CPU-local
		 * variables that other CPUs neither access nor modify,
		 * at least not while the corresponding CPU is online.
2728 2729
		 */

2730 2731
		rcu_sched_qs();
		rcu_bh_qs();
2732 2733 2734 2735 2736 2737 2738

	} else if (!in_softirq()) {

		/*
		 * Get here if this CPU did not take its interrupt from
		 * softirq, in other words, if it is not interrupting
		 * a rcu_bh read-side critical section.  This is an _bh
2739
		 * critical section, so note it.
2740 2741
		 */

2742
		rcu_bh_qs();
2743
	}
2744
	rcu_preempt_check_callbacks();
2745
	if (rcu_pending())
2746
		invoke_rcu_core();
P
Paul E. McKenney 已提交
2747 2748
	if (user)
		rcu_note_voluntary_context_switch(current);
2749
	trace_rcu_utilization(TPS("End scheduler-tick"));
2750 2751 2752 2753 2754
}

/*
 * Scan the leaf rcu_node structures, processing dyntick state for any that
 * have not yet encountered a quiescent state, using the function specified.
2755 2756
 * Also initiate boosting for any threads blocked on the root rcu_node.
 *
2757
 * The caller must have suppressed start of new grace periods.
2758
 */
2759 2760 2761 2762
static void force_qs_rnp(struct rcu_state *rsp,
			 int (*f)(struct rcu_data *rsp, bool *isidle,
				  unsigned long *maxj),
			 bool *isidle, unsigned long *maxj)
2763 2764 2765 2766 2767
{
	unsigned long bit;
	int cpu;
	unsigned long flags;
	unsigned long mask;
2768
	struct rcu_node *rnp;
2769

2770
	rcu_for_each_leaf_node(rsp, rnp) {
2771
		cond_resched_rcu_qs();
2772
		mask = 0;
P
Paul E. McKenney 已提交
2773
		raw_spin_lock_irqsave(&rnp->lock, flags);
2774
		smp_mb__after_unlock_lock();
2775
		if (rnp->qsmask == 0) {
2776 2777 2778 2779 2780 2781 2782 2783 2784 2785 2786 2787 2788 2789 2790 2791 2792 2793 2794 2795 2796 2797 2798
			if (rcu_state_p == &rcu_sched_state ||
			    rsp != rcu_state_p ||
			    rcu_preempt_blocked_readers_cgp(rnp)) {
				/*
				 * No point in scanning bits because they
				 * are all zero.  But we might need to
				 * priority-boost blocked readers.
				 */
				rcu_initiate_boost(rnp, flags);
				/* rcu_initiate_boost() releases rnp->lock */
				continue;
			}
			if (rnp->parent &&
			    (rnp->parent->qsmask & rnp->grpmask)) {
				/*
				 * Race between grace-period
				 * initialization and task exiting RCU
				 * read-side critical section: Report.
				 */
				rcu_report_unblock_qs_rnp(rsp, rnp, flags);
				/* rcu_report_unblock_qs_rnp() rlses ->lock */
				continue;
			}
2799
		}
2800
		cpu = rnp->grplo;
2801
		bit = 1;
2802
		for (; cpu <= rnp->grphi; cpu++, bit <<= 1) {
2803 2804 2805 2806
			if ((rnp->qsmask & bit) != 0) {
				if (f(per_cpu_ptr(rsp->rda, cpu), isidle, maxj))
					mask |= bit;
			}
2807
		}
2808
		if (mask != 0) {
2809 2810
			/* Idle/offline CPUs, report (releases rnp->lock. */
			rcu_report_qs_rnp(mask, rsp, rnp, rnp->gpnum, flags);
2811 2812 2813
		} else {
			/* Nothing to do here, so just drop the lock. */
			raw_spin_unlock_irqrestore(&rnp->lock, flags);
2814 2815 2816 2817 2818 2819 2820 2821
		}
	}
}

/*
 * Force quiescent states on reluctant CPUs, and also detect which
 * CPUs are in dyntick-idle mode.
 */
2822
static void force_quiescent_state(struct rcu_state *rsp)
2823 2824
{
	unsigned long flags;
2825 2826 2827 2828 2829
	bool ret;
	struct rcu_node *rnp;
	struct rcu_node *rnp_old = NULL;

	/* Funnel through hierarchy to reduce memory contention. */
2830
	rnp = __this_cpu_read(rsp->rda->mynode);
2831
	for (; rnp != NULL; rnp = rnp->parent) {
2832
		ret = (READ_ONCE(rsp->gp_flags) & RCU_GP_FLAG_FQS) ||
2833 2834 2835 2836
		      !raw_spin_trylock(&rnp->fqslock);
		if (rnp_old != NULL)
			raw_spin_unlock(&rnp_old->fqslock);
		if (ret) {
2837
			rsp->n_force_qs_lh++;
2838 2839 2840 2841 2842
			return;
		}
		rnp_old = rnp;
	}
	/* rnp_old == rcu_get_root(rsp), rnp == NULL. */
2843

2844 2845
	/* Reached the root of the rcu_node tree, acquire lock. */
	raw_spin_lock_irqsave(&rnp_old->lock, flags);
2846
	smp_mb__after_unlock_lock();
2847
	raw_spin_unlock(&rnp_old->fqslock);
2848
	if (READ_ONCE(rsp->gp_flags) & RCU_GP_FLAG_FQS) {
2849
		rsp->n_force_qs_lh++;
2850
		raw_spin_unlock_irqrestore(&rnp_old->lock, flags);
2851
		return;  /* Someone beat us to it. */
2852
	}
2853
	WRITE_ONCE(rsp->gp_flags, READ_ONCE(rsp->gp_flags) | RCU_GP_FLAG_FQS);
2854
	raw_spin_unlock_irqrestore(&rnp_old->lock, flags);
2855
	rcu_gp_kthread_wake(rsp);
2856 2857 2858
}

/*
2859 2860 2861
 * This does the RCU core processing work for the specified rcu_state
 * and rcu_data structures.  This may be called only from the CPU to
 * whom the rdp belongs.
2862 2863
 */
static void
2864
__rcu_process_callbacks(struct rcu_state *rsp)
2865 2866
{
	unsigned long flags;
2867
	bool needwake;
2868
	struct rcu_data *rdp = raw_cpu_ptr(rsp->rda);
2869

2870 2871
	WARN_ON_ONCE(rdp->beenonline == 0);

2872 2873 2874 2875
	/* Update RCU state based on any recent quiescent states. */
	rcu_check_quiescent_state(rsp, rdp);

	/* Does this CPU require a not-yet-started grace period? */
2876
	local_irq_save(flags);
2877
	if (cpu_needs_another_gp(rsp, rdp)) {
2878
		raw_spin_lock(&rcu_get_root(rsp)->lock); /* irqs disabled. */
2879
		needwake = rcu_start_gp(rsp);
2880
		raw_spin_unlock_irqrestore(&rcu_get_root(rsp)->lock, flags);
2881 2882
		if (needwake)
			rcu_gp_kthread_wake(rsp);
2883 2884
	} else {
		local_irq_restore(flags);
2885 2886 2887
	}

	/* If there are callbacks ready, invoke them. */
2888
	if (cpu_has_callbacks_ready_to_invoke(rdp))
2889
		invoke_rcu_callbacks(rsp, rdp);
2890 2891 2892

	/* Do any needed deferred wakeups of rcuo kthreads. */
	do_nocb_deferred_wakeup(rdp);
2893 2894
}

2895
/*
2896
 * Do RCU core processing for the current CPU.
2897
 */
2898
static void rcu_process_callbacks(struct softirq_action *unused)
2899
{
2900 2901
	struct rcu_state *rsp;

2902 2903
	if (cpu_is_offline(smp_processor_id()))
		return;
2904
	trace_rcu_utilization(TPS("Start RCU core"));
2905 2906
	for_each_rcu_flavor(rsp)
		__rcu_process_callbacks(rsp);
2907
	trace_rcu_utilization(TPS("End RCU core"));
2908 2909
}

2910
/*
2911 2912 2913
 * Schedule RCU callback invocation.  If the specified type of RCU
 * does not support RCU priority boosting, just do a direct call,
 * otherwise wake up the per-CPU kernel kthread.  Note that because we
2914
 * are running on the current CPU with softirqs disabled, the
2915
 * rcu_cpu_kthread_task cannot disappear out from under us.
2916
 */
2917
static void invoke_rcu_callbacks(struct rcu_state *rsp, struct rcu_data *rdp)
2918
{
2919
	if (unlikely(!READ_ONCE(rcu_scheduler_fully_active)))
2920
		return;
2921 2922
	if (likely(!rsp->boost)) {
		rcu_do_batch(rsp, rdp);
2923 2924
		return;
	}
2925
	invoke_rcu_callbacks_kthread();
2926 2927
}

2928
static void invoke_rcu_core(void)
2929
{
2930 2931
	if (cpu_online(smp_processor_id()))
		raise_softirq(RCU_SOFTIRQ);
2932 2933
}

2934 2935 2936 2937 2938
/*
 * Handle any core-RCU processing required by a call_rcu() invocation.
 */
static void __call_rcu_core(struct rcu_state *rsp, struct rcu_data *rdp,
			    struct rcu_head *head, unsigned long flags)
2939
{
2940 2941
	bool needwake;

2942 2943 2944 2945
	/*
	 * If called from an extended quiescent state, invoke the RCU
	 * core in order to force a re-evaluation of RCU's idleness.
	 */
2946
	if (!rcu_is_watching())
2947 2948
		invoke_rcu_core();

2949
	/* If interrupts were disabled or CPU offline, don't invoke RCU core. */
2950
	if (irqs_disabled_flags(flags) || cpu_is_offline(smp_processor_id()))
2951
		return;
2952

2953 2954 2955 2956 2957 2958 2959
	/*
	 * Force the grace period if too many callbacks or too long waiting.
	 * Enforce hysteresis, and don't invoke force_quiescent_state()
	 * if some other CPU has recently done so.  Also, don't bother
	 * invoking force_quiescent_state() if the newly enqueued callback
	 * is the only one waiting for a grace period to complete.
	 */
2960
	if (unlikely(rdp->qlen > rdp->qlen_last_fqs_check + qhimark)) {
2961 2962

		/* Are we ignoring a completed grace period? */
2963
		note_gp_changes(rsp, rdp);
2964 2965 2966 2967 2968

		/* Start a new grace period if one not already started. */
		if (!rcu_gp_in_progress(rsp)) {
			struct rcu_node *rnp_root = rcu_get_root(rsp);

2969
			raw_spin_lock(&rnp_root->lock);
2970
			smp_mb__after_unlock_lock();
2971
			needwake = rcu_start_gp(rsp);
2972
			raw_spin_unlock(&rnp_root->lock);
2973 2974
			if (needwake)
				rcu_gp_kthread_wake(rsp);
2975 2976 2977 2978 2979
		} else {
			/* Give the grace period a kick. */
			rdp->blimit = LONG_MAX;
			if (rsp->n_force_qs == rdp->n_force_qs_snap &&
			    *rdp->nxttail[RCU_DONE_TAIL] != head)
2980
				force_quiescent_state(rsp);
2981 2982 2983
			rdp->n_force_qs_snap = rsp->n_force_qs;
			rdp->qlen_last_fqs_check = rdp->qlen;
		}
2984
	}
2985 2986
}

2987 2988 2989 2990 2991 2992 2993
/*
 * RCU callback function to leak a callback.
 */
static void rcu_leak_callback(struct rcu_head *rhp)
{
}

P
Paul E. McKenney 已提交
2994 2995 2996 2997 2998 2999
/*
 * Helper function for call_rcu() and friends.  The cpu argument will
 * normally be -1, indicating "currently running CPU".  It may specify
 * a CPU only if that CPU is a no-CBs CPU.  Currently, only _rcu_barrier()
 * is expected to specify a CPU.
 */
3000 3001
static void
__call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu),
P
Paul E. McKenney 已提交
3002
	   struct rcu_state *rsp, int cpu, bool lazy)
3003 3004 3005 3006
{
	unsigned long flags;
	struct rcu_data *rdp;

3007
	WARN_ON_ONCE((unsigned long)head & 0x1); /* Misaligned rcu_head! */
3008 3009
	if (debug_rcu_head_queue(head)) {
		/* Probable double call_rcu(), so leak the callback. */
3010
		WRITE_ONCE(head->func, rcu_leak_callback);
3011 3012 3013
		WARN_ONCE(1, "__call_rcu(): Leaked duplicate callback\n");
		return;
	}
3014 3015 3016 3017 3018 3019 3020 3021 3022 3023
	head->func = func;
	head->next = NULL;

	/*
	 * Opportunistically note grace-period endings and beginnings.
	 * Note that we might see a beginning right after we see an
	 * end, but never vice versa, since this CPU has to pass through
	 * a quiescent state betweentimes.
	 */
	local_irq_save(flags);
3024
	rdp = this_cpu_ptr(rsp->rda);
3025 3026

	/* Add the callback to our list. */
P
Paul E. McKenney 已提交
3027 3028 3029 3030 3031
	if (unlikely(rdp->nxttail[RCU_NEXT_TAIL] == NULL) || cpu != -1) {
		int offline;

		if (cpu != -1)
			rdp = per_cpu_ptr(rsp->rda, cpu);
3032 3033 3034 3035 3036 3037 3038 3039 3040 3041 3042 3043 3044
		if (likely(rdp->mynode)) {
			/* Post-boot, so this should be for a no-CBs CPU. */
			offline = !__call_rcu_nocb(rdp, head, lazy, flags);
			WARN_ON_ONCE(offline);
			/* Offline CPU, _call_rcu() illegal, leak callback.  */
			local_irq_restore(flags);
			return;
		}
		/*
		 * Very early boot, before rcu_init().  Initialize if needed
		 * and then drop through to queue the callback.
		 */
		BUG_ON(cpu != -1);
3045
		WARN_ON_ONCE(!rcu_is_watching());
3046 3047
		if (!likely(rdp->nxtlist))
			init_default_callback_list(rdp);
3048
	}
3049
	WRITE_ONCE(rdp->qlen, rdp->qlen + 1);
3050 3051
	if (lazy)
		rdp->qlen_lazy++;
3052 3053
	else
		rcu_idle_count_callbacks_posted();
3054 3055 3056
	smp_mb();  /* Count before adding callback for rcu_barrier(). */
	*rdp->nxttail[RCU_NEXT_TAIL] = head;
	rdp->nxttail[RCU_NEXT_TAIL] = &head->next;
3057

3058 3059
	if (__is_kfree_rcu_offset((unsigned long)func))
		trace_rcu_kfree_callback(rsp->name, head, (unsigned long)func,
3060
					 rdp->qlen_lazy, rdp->qlen);
3061
	else
3062
		trace_rcu_callback(rsp->name, head, rdp->qlen_lazy, rdp->qlen);
3063

3064 3065
	/* Go handle any RCU core processing required. */
	__call_rcu_core(rsp, rdp, head, flags);
3066 3067 3068 3069
	local_irq_restore(flags);
}

/*
3070
 * Queue an RCU-sched callback for invocation after a grace period.
3071
 */
3072
void call_rcu_sched(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
3073
{
P
Paul E. McKenney 已提交
3074
	__call_rcu(head, func, &rcu_sched_state, -1, 0);
3075
}
3076
EXPORT_SYMBOL_GPL(call_rcu_sched);
3077 3078

/*
3079
 * Queue an RCU callback for invocation after a quicker grace period.
3080 3081 3082
 */
void call_rcu_bh(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
{
P
Paul E. McKenney 已提交
3083
	__call_rcu(head, func, &rcu_bh_state, -1, 0);
3084 3085 3086
}
EXPORT_SYMBOL_GPL(call_rcu_bh);

3087 3088 3089 3090 3091 3092 3093 3094 3095 3096
/*
 * Queue an RCU callback for lazy invocation after a grace period.
 * This will likely be later named something like "call_rcu_lazy()",
 * but this change will require some way of tagging the lazy RCU
 * callbacks in the list of pending callbacks. Until then, this
 * function may only be called from __kfree_rcu().
 */
void kfree_call_rcu(struct rcu_head *head,
		    void (*func)(struct rcu_head *rcu))
{
3097
	__call_rcu(head, func, rcu_state_p, -1, 1);
3098 3099 3100
}
EXPORT_SYMBOL_GPL(kfree_call_rcu);

3101 3102 3103 3104 3105 3106 3107 3108 3109 3110 3111
/*
 * Because a context switch is a grace period for RCU-sched and RCU-bh,
 * any blocking grace-period wait automatically implies a grace period
 * if there is only one CPU online at any point time during execution
 * of either synchronize_sched() or synchronize_rcu_bh().  It is OK to
 * occasionally incorrectly indicate that there are multiple CPUs online
 * when there was in fact only one the whole time, as this just adds
 * some overhead: RCU still operates correctly.
 */
static inline int rcu_blocking_is_gp(void)
{
3112 3113
	int ret;

3114
	might_sleep();  /* Check for RCU read-side critical section. */
3115 3116 3117 3118
	preempt_disable();
	ret = num_online_cpus() <= 1;
	preempt_enable();
	return ret;
3119 3120
}

3121 3122 3123 3124 3125 3126 3127 3128 3129 3130 3131 3132
/**
 * synchronize_sched - wait until an rcu-sched grace period has elapsed.
 *
 * Control will return to the caller some time after a full rcu-sched
 * grace period has elapsed, in other words after all currently executing
 * rcu-sched read-side critical sections have completed.   These read-side
 * critical sections are delimited by rcu_read_lock_sched() and
 * rcu_read_unlock_sched(), and may be nested.  Note that preempt_disable(),
 * local_irq_disable(), and so on may be used in place of
 * rcu_read_lock_sched().
 *
 * This means that all preempt_disable code sequences, including NMI and
3133 3134 3135 3136 3137 3138 3139 3140 3141 3142 3143 3144 3145 3146 3147 3148 3149 3150 3151 3152 3153 3154
 * non-threaded hardware-interrupt handlers, in progress on entry will
 * have completed before this primitive returns.  However, this does not
 * guarantee that softirq handlers will have completed, since in some
 * kernels, these handlers can run in process context, and can block.
 *
 * Note that this guarantee implies further memory-ordering guarantees.
 * On systems with more than one CPU, when synchronize_sched() returns,
 * each CPU is guaranteed to have executed a full memory barrier since the
 * end of its last RCU-sched read-side critical section whose beginning
 * preceded the call to synchronize_sched().  In addition, each CPU having
 * an RCU read-side critical section that extends beyond the return from
 * synchronize_sched() is guaranteed to have executed a full memory barrier
 * after the beginning of synchronize_sched() and before the beginning of
 * that RCU read-side critical section.  Note that these guarantees include
 * CPUs that are offline, idle, or executing in user mode, as well as CPUs
 * that are executing in the kernel.
 *
 * Furthermore, if CPU A invoked synchronize_sched(), which returned
 * to its caller on CPU B, then both CPU A and CPU B are guaranteed
 * to have executed a full memory barrier during the execution of
 * synchronize_sched() -- even if CPU A and CPU B are the same CPU (but
 * again only if the system has more than one CPU).
3155 3156 3157 3158 3159 3160 3161 3162 3163
 *
 * This primitive provides the guarantees made by the (now removed)
 * synchronize_kernel() API.  In contrast, synchronize_rcu() only
 * guarantees that rcu_read_lock() sections will have completed.
 * In "classic RCU", these two guarantees happen to be one and
 * the same, but can differ in realtime RCU implementations.
 */
void synchronize_sched(void)
{
3164 3165 3166 3167
	rcu_lockdep_assert(!lock_is_held(&rcu_bh_lock_map) &&
			   !lock_is_held(&rcu_lock_map) &&
			   !lock_is_held(&rcu_sched_lock_map),
			   "Illegal synchronize_sched() in RCU-sched read-side critical section");
3168 3169
	if (rcu_blocking_is_gp())
		return;
3170
	if (rcu_gp_is_expedited())
3171 3172 3173
		synchronize_sched_expedited();
	else
		wait_rcu_gp(call_rcu_sched);
3174 3175 3176 3177 3178 3179 3180 3181 3182 3183 3184
}
EXPORT_SYMBOL_GPL(synchronize_sched);

/**
 * synchronize_rcu_bh - wait until an rcu_bh grace period has elapsed.
 *
 * Control will return to the caller some time after a full rcu_bh grace
 * period has elapsed, in other words after all currently executing rcu_bh
 * read-side critical sections have completed.  RCU read-side critical
 * sections are delimited by rcu_read_lock_bh() and rcu_read_unlock_bh(),
 * and may be nested.
3185 3186 3187
 *
 * See the description of synchronize_sched() for more detailed information
 * on memory ordering guarantees.
3188 3189 3190
 */
void synchronize_rcu_bh(void)
{
3191 3192 3193 3194
	rcu_lockdep_assert(!lock_is_held(&rcu_bh_lock_map) &&
			   !lock_is_held(&rcu_lock_map) &&
			   !lock_is_held(&rcu_sched_lock_map),
			   "Illegal synchronize_rcu_bh() in RCU-bh read-side critical section");
3195 3196
	if (rcu_blocking_is_gp())
		return;
3197
	if (rcu_gp_is_expedited())
3198 3199 3200
		synchronize_rcu_bh_expedited();
	else
		wait_rcu_gp(call_rcu_bh);
3201 3202 3203
}
EXPORT_SYMBOL_GPL(synchronize_rcu_bh);

3204 3205 3206 3207 3208 3209 3210 3211 3212 3213 3214 3215 3216 3217 3218 3219 3220 3221 3222 3223
/**
 * get_state_synchronize_rcu - Snapshot current RCU state
 *
 * Returns a cookie that is used by a later call to cond_synchronize_rcu()
 * to determine whether or not a full grace period has elapsed in the
 * meantime.
 */
unsigned long get_state_synchronize_rcu(void)
{
	/*
	 * Any prior manipulation of RCU-protected data must happen
	 * before the load from ->gpnum.
	 */
	smp_mb();  /* ^^^ */

	/*
	 * Make sure this load happens before the purportedly
	 * time-consuming work between get_state_synchronize_rcu()
	 * and cond_synchronize_rcu().
	 */
3224
	return smp_load_acquire(&rcu_state_p->gpnum);
3225 3226 3227 3228 3229 3230 3231 3232 3233 3234 3235 3236 3237 3238 3239 3240 3241 3242 3243 3244 3245 3246 3247 3248 3249
}
EXPORT_SYMBOL_GPL(get_state_synchronize_rcu);

/**
 * cond_synchronize_rcu - Conditionally wait for an RCU grace period
 *
 * @oldstate: return value from earlier call to get_state_synchronize_rcu()
 *
 * If a full RCU grace period has elapsed since the earlier call to
 * get_state_synchronize_rcu(), just return.  Otherwise, invoke
 * synchronize_rcu() to wait for a full grace period.
 *
 * Yes, this function does not take counter wrap into account.  But
 * counter wrap is harmless.  If the counter wraps, we have waited for
 * more than 2 billion grace periods (and way more on a 64-bit system!),
 * so waiting for one additional grace period should be just fine.
 */
void cond_synchronize_rcu(unsigned long oldstate)
{
	unsigned long newstate;

	/*
	 * Ensure that this load happens before any RCU-destructive
	 * actions the caller might carry out after we return.
	 */
3250
	newstate = smp_load_acquire(&rcu_state_p->completed);
3251 3252 3253 3254 3255
	if (ULONG_CMP_GE(oldstate, newstate))
		synchronize_rcu();
}
EXPORT_SYMBOL_GPL(cond_synchronize_rcu);

3256 3257 3258 3259 3260 3261 3262 3263 3264 3265 3266 3267 3268 3269 3270 3271 3272 3273 3274 3275 3276 3277 3278 3279 3280 3281 3282 3283 3284 3285 3286 3287 3288 3289 3290 3291 3292 3293 3294 3295 3296 3297 3298 3299 3300 3301 3302 3303 3304 3305 3306 3307
/**
 * get_state_synchronize_sched - Snapshot current RCU-sched state
 *
 * Returns a cookie that is used by a later call to cond_synchronize_sched()
 * to determine whether or not a full grace period has elapsed in the
 * meantime.
 */
unsigned long get_state_synchronize_sched(void)
{
	/*
	 * Any prior manipulation of RCU-protected data must happen
	 * before the load from ->gpnum.
	 */
	smp_mb();  /* ^^^ */

	/*
	 * Make sure this load happens before the purportedly
	 * time-consuming work between get_state_synchronize_sched()
	 * and cond_synchronize_sched().
	 */
	return smp_load_acquire(&rcu_sched_state.gpnum);
}
EXPORT_SYMBOL_GPL(get_state_synchronize_sched);

/**
 * cond_synchronize_sched - Conditionally wait for an RCU-sched grace period
 *
 * @oldstate: return value from earlier call to get_state_synchronize_sched()
 *
 * If a full RCU-sched grace period has elapsed since the earlier call to
 * get_state_synchronize_sched(), just return.  Otherwise, invoke
 * synchronize_sched() to wait for a full grace period.
 *
 * Yes, this function does not take counter wrap into account.  But
 * counter wrap is harmless.  If the counter wraps, we have waited for
 * more than 2 billion grace periods (and way more on a 64-bit system!),
 * so waiting for one additional grace period should be just fine.
 */
void cond_synchronize_sched(unsigned long oldstate)
{
	unsigned long newstate;

	/*
	 * Ensure that this load happens before any RCU-destructive
	 * actions the caller might carry out after we return.
	 */
	newstate = smp_load_acquire(&rcu_sched_state.completed);
	if (ULONG_CMP_GE(oldstate, newstate))
		synchronize_sched();
}
EXPORT_SYMBOL_GPL(cond_synchronize_sched);

3308 3309 3310 3311 3312 3313 3314 3315 3316 3317 3318 3319 3320 3321 3322 3323 3324
static int synchronize_sched_expedited_cpu_stop(void *data)
{
	/*
	 * There must be a full memory barrier on each affected CPU
	 * between the time that try_stop_cpus() is called and the
	 * time that it returns.
	 *
	 * In the current initial implementation of cpu_stop, the
	 * above condition is already met when the control reaches
	 * this point and the following smp_mb() is not strictly
	 * necessary.  Do smp_mb() anyway for documentation and
	 * robustness against future implementation changes.
	 */
	smp_mb(); /* See above comment block. */
	return 0;
}

3325 3326 3327 3328 3329 3330 3331 3332 3333 3334
/**
 * synchronize_sched_expedited - Brute-force RCU-sched grace period
 *
 * Wait for an RCU-sched grace period to elapse, but use a "big hammer"
 * approach to force the grace period to end quickly.  This consumes
 * significant time on all CPUs and is unfriendly to real-time workloads,
 * so is thus not recommended for any sort of common-case code.  In fact,
 * if you are using synchronize_sched_expedited() in a loop, please
 * restructure your code to batch your updates, and then use a single
 * synchronize_sched() instead.
3335 3336 3337 3338 3339 3340 3341 3342 3343 3344 3345 3346 3347 3348 3349 3350 3351 3352 3353 3354 3355 3356 3357 3358
 *
 * This implementation can be thought of as an application of ticket
 * locking to RCU, with sync_sched_expedited_started and
 * sync_sched_expedited_done taking on the roles of the halves
 * of the ticket-lock word.  Each task atomically increments
 * sync_sched_expedited_started upon entry, snapshotting the old value,
 * then attempts to stop all the CPUs.  If this succeeds, then each
 * CPU will have executed a context switch, resulting in an RCU-sched
 * grace period.  We are then done, so we use atomic_cmpxchg() to
 * update sync_sched_expedited_done to match our snapshot -- but
 * only if someone else has not already advanced past our snapshot.
 *
 * On the other hand, if try_stop_cpus() fails, we check the value
 * of sync_sched_expedited_done.  If it has advanced past our
 * initial snapshot, then someone else must have forced a grace period
 * some time after we took our snapshot.  In this case, our work is
 * done for us, and we can simply return.  Otherwise, we try again,
 * but keep our initial snapshot for purposes of checking for someone
 * doing our work for us.
 *
 * If we fail too many times in a row, we fall back to synchronize_sched().
 */
void synchronize_sched_expedited(void)
{
3359 3360 3361
	cpumask_var_t cm;
	bool cma = false;
	int cpu;
3362 3363
	long firstsnap, s, snap;
	int trycount = 0;
3364
	struct rcu_state *rsp = &rcu_sched_state;
3365

3366 3367 3368 3369 3370 3371 3372 3373
	/*
	 * If we are in danger of counter wrap, just do synchronize_sched().
	 * By allowing sync_sched_expedited_started to advance no more than
	 * ULONG_MAX/8 ahead of sync_sched_expedited_done, we are ensuring
	 * that more than 3.5 billion CPUs would be required to force a
	 * counter wrap on a 32-bit system.  Quite a few more CPUs would of
	 * course be required on a 64-bit system.
	 */
3374 3375
	if (ULONG_CMP_GE((ulong)atomic_long_read(&rsp->expedited_start),
			 (ulong)atomic_long_read(&rsp->expedited_done) +
3376
			 ULONG_MAX / 8)) {
3377
		wait_rcu_gp(call_rcu_sched);
3378
		atomic_long_inc(&rsp->expedited_wrap);
3379 3380
		return;
	}
3381

3382 3383 3384 3385
	/*
	 * Take a ticket.  Note that atomic_inc_return() implies a
	 * full memory barrier.
	 */
3386
	snap = atomic_long_inc_return(&rsp->expedited_start);
3387
	firstsnap = snap;
3388 3389 3390 3391 3392 3393
	if (!try_get_online_cpus()) {
		/* CPU hotplug operation in flight, fall back to normal GP. */
		wait_rcu_gp(call_rcu_sched);
		atomic_long_inc(&rsp->expedited_normal);
		return;
	}
3394
	WARN_ON_ONCE(cpu_is_offline(raw_smp_processor_id()));
3395

3396 3397 3398 3399 3400 3401 3402 3403 3404 3405 3406 3407 3408 3409 3410
	/* Offline CPUs, idle CPUs, and any CPU we run on are quiescent. */
	cma = zalloc_cpumask_var(&cm, GFP_KERNEL);
	if (cma) {
		cpumask_copy(cm, cpu_online_mask);
		cpumask_clear_cpu(raw_smp_processor_id(), cm);
		for_each_cpu(cpu, cm) {
			struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu);

			if (!(atomic_add_return(0, &rdtp->dynticks) & 0x1))
				cpumask_clear_cpu(cpu, cm);
		}
		if (cpumask_weight(cm) == 0)
			goto all_cpus_idle;
	}

3411 3412 3413 3414
	/*
	 * Each pass through the following loop attempts to force a
	 * context switch on each CPU.
	 */
3415
	while (try_stop_cpus(cma ? cm : cpu_online_mask,
3416 3417 3418
			     synchronize_sched_expedited_cpu_stop,
			     NULL) == -EAGAIN) {
		put_online_cpus();
3419
		atomic_long_inc(&rsp->expedited_tryfail);
3420

3421
		/* Check to see if someone else did our work for us. */
3422
		s = atomic_long_read(&rsp->expedited_done);
3423
		if (ULONG_CMP_GE((ulong)s, (ulong)firstsnap)) {
3424
			/* ensure test happens before caller kfree */
3425
			smp_mb__before_atomic(); /* ^^^ */
3426
			atomic_long_inc(&rsp->expedited_workdone1);
3427
			free_cpumask_var(cm);
3428 3429
			return;
		}
3430 3431

		/* No joy, try again later.  Or just synchronize_sched(). */
3432
		if (trycount++ < 10) {
3433
			udelay(trycount * num_online_cpus());
3434
		} else {
3435
			wait_rcu_gp(call_rcu_sched);
3436
			atomic_long_inc(&rsp->expedited_normal);
3437
			free_cpumask_var(cm);
3438 3439 3440
			return;
		}

3441
		/* Recheck to see if someone else did our work for us. */
3442
		s = atomic_long_read(&rsp->expedited_done);
3443
		if (ULONG_CMP_GE((ulong)s, (ulong)firstsnap)) {
3444
			/* ensure test happens before caller kfree */
3445
			smp_mb__before_atomic(); /* ^^^ */
3446
			atomic_long_inc(&rsp->expedited_workdone2);
3447
			free_cpumask_var(cm);
3448 3449 3450 3451 3452
			return;
		}

		/*
		 * Refetching sync_sched_expedited_started allows later
3453 3454 3455 3456
		 * callers to piggyback on our grace period.  We retry
		 * after they started, so our grace period works for them,
		 * and they started after our first try, so their grace
		 * period works for us.
3457
		 */
3458 3459 3460 3461
		if (!try_get_online_cpus()) {
			/* CPU hotplug operation in flight, use normal GP. */
			wait_rcu_gp(call_rcu_sched);
			atomic_long_inc(&rsp->expedited_normal);
3462
			free_cpumask_var(cm);
3463 3464
			return;
		}
3465
		snap = atomic_long_read(&rsp->expedited_start);
3466 3467
		smp_mb(); /* ensure read is before try_stop_cpus(). */
	}
3468
	atomic_long_inc(&rsp->expedited_stoppedcpus);
3469

3470 3471 3472
all_cpus_idle:
	free_cpumask_var(cm);

3473 3474 3475 3476
	/*
	 * Everyone up to our most recent fetch is covered by our grace
	 * period.  Update the counter, but only if our work is still
	 * relevant -- which it won't be if someone who started later
3477
	 * than we did already did their update.
3478 3479
	 */
	do {
3480
		atomic_long_inc(&rsp->expedited_done_tries);
3481
		s = atomic_long_read(&rsp->expedited_done);
3482
		if (ULONG_CMP_GE((ulong)s, (ulong)snap)) {
3483
			/* ensure test happens before caller kfree */
3484
			smp_mb__before_atomic(); /* ^^^ */
3485
			atomic_long_inc(&rsp->expedited_done_lost);
3486 3487
			break;
		}
3488
	} while (atomic_long_cmpxchg(&rsp->expedited_done, s, snap) != s);
3489
	atomic_long_inc(&rsp->expedited_done_exit);
3490 3491 3492 3493 3494

	put_online_cpus();
}
EXPORT_SYMBOL_GPL(synchronize_sched_expedited);

3495 3496 3497 3498 3499 3500 3501 3502 3503
/*
 * Check to see if there is any immediate RCU-related work to be done
 * by the current CPU, for the specified type of RCU, returning 1 if so.
 * The checks are in order of increasing expense: checks that can be
 * carried out against CPU-local state are performed first.  However,
 * we must check for CPU stalls first, else we might not get a chance.
 */
static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp)
{
3504 3505
	struct rcu_node *rnp = rdp->mynode;

3506 3507 3508 3509 3510
	rdp->n_rcu_pending++;

	/* Check for CPU stalls, if enabled. */
	check_cpu_stall(rsp, rdp);

3511 3512 3513 3514
	/* Is this CPU a NO_HZ_FULL CPU that should ignore RCU? */
	if (rcu_nohz_full_cpu(rsp))
		return 0;

3515
	/* Is the RCU core waiting for a quiescent state from this CPU? */
3516
	if (rcu_scheduler_fully_active &&
3517 3518
	    rdp->qs_pending && !rdp->passed_quiesce &&
	    rdp->rcu_qs_ctr_snap == __this_cpu_read(rcu_qs_ctr)) {
3519
		rdp->n_rp_qs_pending++;
3520 3521 3522
	} else if (rdp->qs_pending &&
		   (rdp->passed_quiesce ||
		    rdp->rcu_qs_ctr_snap != __this_cpu_read(rcu_qs_ctr))) {
3523
		rdp->n_rp_report_qs++;
3524
		return 1;
3525
	}
3526 3527

	/* Does this CPU have callbacks ready to invoke? */
3528 3529
	if (cpu_has_callbacks_ready_to_invoke(rdp)) {
		rdp->n_rp_cb_ready++;
3530
		return 1;
3531
	}
3532 3533

	/* Has RCU gone idle with this CPU needing another grace period? */
3534 3535
	if (cpu_needs_another_gp(rsp, rdp)) {
		rdp->n_rp_cpu_needs_gp++;
3536
		return 1;
3537
	}
3538 3539

	/* Has another RCU grace period completed?  */
3540
	if (READ_ONCE(rnp->completed) != rdp->completed) { /* outside lock */
3541
		rdp->n_rp_gp_completed++;
3542
		return 1;
3543
	}
3544 3545

	/* Has a new RCU grace period started? */
3546 3547
	if (READ_ONCE(rnp->gpnum) != rdp->gpnum ||
	    unlikely(READ_ONCE(rdp->gpwrap))) { /* outside lock */
3548
		rdp->n_rp_gp_started++;
3549
		return 1;
3550
	}
3551

3552 3553 3554 3555 3556 3557
	/* Does this CPU need a deferred NOCB wakeup? */
	if (rcu_nocb_need_deferred_wakeup(rdp)) {
		rdp->n_rp_nocb_defer_wakeup++;
		return 1;
	}

3558
	/* nothing to do */
3559
	rdp->n_rp_need_nothing++;
3560 3561 3562 3563 3564 3565 3566 3567
	return 0;
}

/*
 * Check to see if there is any immediate RCU-related work to be done
 * by the current CPU, returning 1 if so.  This function is part of the
 * RCU implementation; it is -not- an exported member of the RCU API.
 */
3568
static int rcu_pending(void)
3569
{
3570 3571 3572
	struct rcu_state *rsp;

	for_each_rcu_flavor(rsp)
3573
		if (__rcu_pending(rsp, this_cpu_ptr(rsp->rda)))
3574 3575
			return 1;
	return 0;
3576 3577 3578
}

/*
3579 3580 3581
 * Return true if the specified CPU has any callback.  If all_lazy is
 * non-NULL, store an indication of whether all callbacks are lazy.
 * (If there are no callbacks, all of them are deemed to be lazy.)
3582
 */
3583
static bool __maybe_unused rcu_cpu_has_callbacks(bool *all_lazy)
3584
{
3585 3586 3587
	bool al = true;
	bool hc = false;
	struct rcu_data *rdp;
3588 3589
	struct rcu_state *rsp;

3590
	for_each_rcu_flavor(rsp) {
3591
		rdp = this_cpu_ptr(rsp->rda);
3592 3593 3594 3595
		if (!rdp->nxtlist)
			continue;
		hc = true;
		if (rdp->qlen != rdp->qlen_lazy || !all_lazy) {
3596
			al = false;
3597 3598
			break;
		}
3599 3600 3601 3602
	}
	if (all_lazy)
		*all_lazy = al;
	return hc;
3603 3604
}

3605 3606 3607 3608
/*
 * Helper function for _rcu_barrier() tracing.  If tracing is disabled,
 * the compiler is expected to optimize this away.
 */
3609
static void _rcu_barrier_trace(struct rcu_state *rsp, const char *s,
3610 3611 3612 3613 3614 3615
			       int cpu, unsigned long done)
{
	trace_rcu_barrier(rsp->name, s, cpu,
			  atomic_read(&rsp->barrier_cpu_count), done);
}

3616 3617 3618 3619
/*
 * RCU callback function for _rcu_barrier().  If we are last, wake
 * up the task executing _rcu_barrier().
 */
3620
static void rcu_barrier_callback(struct rcu_head *rhp)
3621
{
3622 3623 3624
	struct rcu_data *rdp = container_of(rhp, struct rcu_data, barrier_head);
	struct rcu_state *rsp = rdp->rsp;

3625 3626
	if (atomic_dec_and_test(&rsp->barrier_cpu_count)) {
		_rcu_barrier_trace(rsp, "LastCB", -1, rsp->n_barrier_done);
3627
		complete(&rsp->barrier_completion);
3628 3629 3630
	} else {
		_rcu_barrier_trace(rsp, "CB", -1, rsp->n_barrier_done);
	}
3631 3632 3633 3634 3635 3636 3637
}

/*
 * Called with preemption disabled, and from cross-cpu IRQ context.
 */
static void rcu_barrier_func(void *type)
{
3638
	struct rcu_state *rsp = type;
3639
	struct rcu_data *rdp = raw_cpu_ptr(rsp->rda);
3640

3641
	_rcu_barrier_trace(rsp, "IRQ", -1, rsp->n_barrier_done);
3642
	atomic_inc(&rsp->barrier_cpu_count);
3643
	rsp->call(&rdp->barrier_head, rcu_barrier_callback);
3644 3645 3646 3647 3648 3649
}

/*
 * Orchestrate the specified type of RCU barrier, waiting for all
 * RCU callbacks of the specified type to complete.
 */
3650
static void _rcu_barrier(struct rcu_state *rsp)
3651
{
3652 3653
	int cpu;
	struct rcu_data *rdp;
3654
	unsigned long snap = READ_ONCE(rsp->n_barrier_done);
3655
	unsigned long snap_done;
3656

3657
	_rcu_barrier_trace(rsp, "Begin", -1, snap);
3658

3659
	/* Take mutex to serialize concurrent rcu_barrier() requests. */
3660
	mutex_lock(&rsp->barrier_mutex);
3661

3662 3663 3664 3665 3666 3667 3668 3669 3670 3671 3672 3673
	/*
	 * Ensure that all prior references, including to ->n_barrier_done,
	 * are ordered before the _rcu_barrier() machinery.
	 */
	smp_mb();  /* See above block comment. */

	/*
	 * Recheck ->n_barrier_done to see if others did our work for us.
	 * This means checking ->n_barrier_done for an even-to-odd-to-even
	 * transition.  The "if" expression below therefore rounds the old
	 * value up to the next even number and adds two before comparing.
	 */
3674
	snap_done = rsp->n_barrier_done;
3675
	_rcu_barrier_trace(rsp, "Check", -1, snap_done);
3676 3677 3678 3679 3680 3681 3682 3683 3684 3685 3686 3687

	/*
	 * If the value in snap is odd, we needed to wait for the current
	 * rcu_barrier() to complete, then wait for the next one, in other
	 * words, we need the value of snap_done to be three larger than
	 * the value of snap.  On the other hand, if the value in snap is
	 * even, we only had to wait for the next rcu_barrier() to complete,
	 * in other words, we need the value of snap_done to be only two
	 * greater than the value of snap.  The "(snap + 3) & ~0x1" computes
	 * this for us (thank you, Linus!).
	 */
	if (ULONG_CMP_GE(snap_done, (snap + 3) & ~0x1)) {
3688
		_rcu_barrier_trace(rsp, "EarlyExit", -1, snap_done);
3689 3690 3691 3692 3693 3694 3695
		smp_mb(); /* caller's subsequent code after above check. */
		mutex_unlock(&rsp->barrier_mutex);
		return;
	}

	/*
	 * Increment ->n_barrier_done to avoid duplicate work.  Use
3696
	 * WRITE_ONCE() to prevent the compiler from speculating
3697 3698
	 * the increment to precede the early-exit check.
	 */
3699
	WRITE_ONCE(rsp->n_barrier_done, rsp->n_barrier_done + 1);
3700
	WARN_ON_ONCE((rsp->n_barrier_done & 0x1) != 1);
3701
	_rcu_barrier_trace(rsp, "Inc1", -1, rsp->n_barrier_done);
3702
	smp_mb(); /* Order ->n_barrier_done increment with below mechanism. */
3703

3704
	/*
3705 3706
	 * Initialize the count to one rather than to zero in order to
	 * avoid a too-soon return to zero in case of a short grace period
3707 3708
	 * (or preemption of this task).  Exclude CPU-hotplug operations
	 * to ensure that no offline CPU has callbacks queued.
3709
	 */
3710
	init_completion(&rsp->barrier_completion);
3711
	atomic_set(&rsp->barrier_cpu_count, 1);
3712
	get_online_cpus();
3713 3714

	/*
3715 3716 3717
	 * Force each CPU with callbacks to register a new callback.
	 * When that callback is invoked, we will know that all of the
	 * corresponding CPU's preceding callbacks have been invoked.
3718
	 */
P
Paul E. McKenney 已提交
3719
	for_each_possible_cpu(cpu) {
3720
		if (!cpu_online(cpu) && !rcu_is_nocb_cpu(cpu))
P
Paul E. McKenney 已提交
3721
			continue;
3722
		rdp = per_cpu_ptr(rsp->rda, cpu);
3723
		if (rcu_is_nocb_cpu(cpu)) {
3724 3725 3726 3727 3728 3729
			if (!rcu_nocb_cpu_needs_barrier(rsp, cpu)) {
				_rcu_barrier_trace(rsp, "OfflineNoCB", cpu,
						   rsp->n_barrier_done);
			} else {
				_rcu_barrier_trace(rsp, "OnlineNoCB", cpu,
						   rsp->n_barrier_done);
3730
				smp_mb__before_atomic();
3731 3732 3733 3734
				atomic_inc(&rsp->barrier_cpu_count);
				__call_rcu(&rdp->barrier_head,
					   rcu_barrier_callback, rsp, cpu, 0);
			}
3735
		} else if (READ_ONCE(rdp->qlen)) {
3736 3737
			_rcu_barrier_trace(rsp, "OnlineQ", cpu,
					   rsp->n_barrier_done);
3738
			smp_call_function_single(cpu, rcu_barrier_func, rsp, 1);
3739
		} else {
3740 3741
			_rcu_barrier_trace(rsp, "OnlineNQ", cpu,
					   rsp->n_barrier_done);
3742 3743
		}
	}
3744
	put_online_cpus();
3745 3746 3747 3748 3749

	/*
	 * Now that we have an rcu_barrier_callback() callback on each
	 * CPU, and thus each counted, remove the initial count.
	 */
3750
	if (atomic_dec_and_test(&rsp->barrier_cpu_count))
3751
		complete(&rsp->barrier_completion);
3752

3753 3754
	/* Increment ->n_barrier_done to prevent duplicate work. */
	smp_mb(); /* Keep increment after above mechanism. */
3755
	WRITE_ONCE(rsp->n_barrier_done, rsp->n_barrier_done + 1);
3756
	WARN_ON_ONCE((rsp->n_barrier_done & 0x1) != 0);
3757
	_rcu_barrier_trace(rsp, "Inc2", -1, rsp->n_barrier_done);
3758 3759
	smp_mb(); /* Keep increment before caller's subsequent code. */

3760
	/* Wait for all rcu_barrier_callback() callbacks to be invoked. */
3761
	wait_for_completion(&rsp->barrier_completion);
3762 3763

	/* Other rcu_barrier() invocations can now safely proceed. */
3764
	mutex_unlock(&rsp->barrier_mutex);
3765 3766 3767 3768 3769 3770 3771
}

/**
 * rcu_barrier_bh - Wait until all in-flight call_rcu_bh() callbacks complete.
 */
void rcu_barrier_bh(void)
{
3772
	_rcu_barrier(&rcu_bh_state);
3773 3774 3775 3776 3777 3778 3779 3780
}
EXPORT_SYMBOL_GPL(rcu_barrier_bh);

/**
 * rcu_barrier_sched - Wait for in-flight call_rcu_sched() callbacks.
 */
void rcu_barrier_sched(void)
{
3781
	_rcu_barrier(&rcu_sched_state);
3782 3783 3784
}
EXPORT_SYMBOL_GPL(rcu_barrier_sched);

3785 3786 3787 3788 3789 3790 3791 3792 3793 3794 3795 3796 3797 3798 3799 3800 3801 3802 3803 3804 3805 3806
/*
 * Propagate ->qsinitmask bits up the rcu_node tree to account for the
 * first CPU in a given leaf rcu_node structure coming online.  The caller
 * must hold the corresponding leaf rcu_node ->lock with interrrupts
 * disabled.
 */
static void rcu_init_new_rnp(struct rcu_node *rnp_leaf)
{
	long mask;
	struct rcu_node *rnp = rnp_leaf;

	for (;;) {
		mask = rnp->grpmask;
		rnp = rnp->parent;
		if (rnp == NULL)
			return;
		raw_spin_lock(&rnp->lock); /* Interrupts already disabled. */
		rnp->qsmaskinit |= mask;
		raw_spin_unlock(&rnp->lock); /* Interrupts remain disabled. */
	}
}

3807
/*
3808
 * Do boot-time initialization of a CPU's per-CPU RCU data.
3809
 */
3810 3811
static void __init
rcu_boot_init_percpu_data(int cpu, struct rcu_state *rsp)
3812 3813
{
	unsigned long flags;
3814
	struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
3815 3816 3817
	struct rcu_node *rnp = rcu_get_root(rsp);

	/* Set up local state, ensuring consistent view of global state. */
P
Paul E. McKenney 已提交
3818
	raw_spin_lock_irqsave(&rnp->lock, flags);
3819 3820
	rdp->grpmask = 1UL << (cpu - rdp->mynode->grplo);
	rdp->dynticks = &per_cpu(rcu_dynticks, cpu);
3821
	WARN_ON_ONCE(rdp->dynticks->dynticks_nesting != DYNTICK_TASK_EXIT_IDLE);
3822
	WARN_ON_ONCE(atomic_read(&rdp->dynticks->dynticks) != 1);
3823
	rdp->cpu = cpu;
3824
	rdp->rsp = rsp;
P
Paul E. McKenney 已提交
3825
	rcu_boot_init_nocb_percpu_data(rdp);
P
Paul E. McKenney 已提交
3826
	raw_spin_unlock_irqrestore(&rnp->lock, flags);
3827 3828 3829 3830 3831 3832 3833
}

/*
 * Initialize a CPU's per-CPU RCU data.  Note that only one online or
 * offline event can be happening at a given time.  Note also that we
 * can accept some slop in the rsp->completed access due to the fact
 * that this CPU cannot possibly have any RCU callbacks in flight yet.
3834
 */
3835
static void
3836
rcu_init_percpu_data(int cpu, struct rcu_state *rsp)
3837 3838 3839
{
	unsigned long flags;
	unsigned long mask;
3840
	struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
3841 3842 3843
	struct rcu_node *rnp = rcu_get_root(rsp);

	/* Set up local state, ensuring consistent view of global state. */
P
Paul E. McKenney 已提交
3844
	raw_spin_lock_irqsave(&rnp->lock, flags);
3845
	rdp->beenonline = 1;	 /* We have now been online. */
3846 3847
	rdp->qlen_last_fqs_check = 0;
	rdp->n_force_qs_snap = rsp->n_force_qs;
3848
	rdp->blimit = blimit;
3849 3850
	if (!rdp->nxtlist)
		init_callback_list(rdp);  /* Re-enable callbacks on this CPU. */
3851
	rdp->dynticks->dynticks_nesting = DYNTICK_TASK_EXIT_IDLE;
3852
	rcu_sysidle_init_percpu_data(rdp->dynticks);
3853 3854
	atomic_set(&rdp->dynticks->dynticks,
		   (atomic_read(&rdp->dynticks->dynticks) & ~0x1) + 1);
P
Paul E. McKenney 已提交
3855
	raw_spin_unlock(&rnp->lock);		/* irqs remain disabled. */
3856

3857 3858 3859 3860 3861
	/*
	 * Add CPU to leaf rcu_node pending-online bitmask.  Any needed
	 * propagation up the rcu_node tree will happen at the beginning
	 * of the next grace period.
	 */
3862 3863
	rnp = rdp->mynode;
	mask = rdp->grpmask;
3864 3865 3866 3867 3868 3869
	raw_spin_lock(&rnp->lock);		/* irqs already disabled. */
	smp_mb__after_unlock_lock();
	rnp->qsmaskinitnext |= mask;
	rdp->gpnum = rnp->completed; /* Make CPU later note any new GP. */
	rdp->completed = rnp->completed;
	rdp->passed_quiesce = false;
3870
	rdp->rcu_qs_ctr_snap = per_cpu(rcu_qs_ctr, cpu);
3871 3872 3873
	rdp->qs_pending = false;
	trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("cpuonl"));
	raw_spin_unlock_irqrestore(&rnp->lock, flags);
3874 3875
}

3876
static void rcu_prepare_cpu(int cpu)
3877
{
3878 3879 3880
	struct rcu_state *rsp;

	for_each_rcu_flavor(rsp)
3881
		rcu_init_percpu_data(cpu, rsp);
3882 3883 3884
}

/*
3885
 * Handle CPU online/offline notification events.
3886
 */
3887 3888
int rcu_cpu_notify(struct notifier_block *self,
		   unsigned long action, void *hcpu)
3889 3890
{
	long cpu = (long)hcpu;
3891
	struct rcu_data *rdp = per_cpu_ptr(rcu_state_p->rda, cpu);
3892
	struct rcu_node *rnp = rdp->mynode;
3893
	struct rcu_state *rsp;
3894 3895 3896 3897

	switch (action) {
	case CPU_UP_PREPARE:
	case CPU_UP_PREPARE_FROZEN:
P
Peter Zijlstra 已提交
3898 3899
		rcu_prepare_cpu(cpu);
		rcu_prepare_kthreads(cpu);
3900
		rcu_spawn_all_nocb_kthreads(cpu);
3901 3902
		break;
	case CPU_ONLINE:
3903
	case CPU_DOWN_FAILED:
T
Thomas Gleixner 已提交
3904
		rcu_boost_kthread_setaffinity(rnp, -1);
3905 3906
		break;
	case CPU_DOWN_PREPARE:
3907
		rcu_boost_kthread_setaffinity(rnp, cpu);
3908
		break;
3909 3910
	case CPU_DYING:
	case CPU_DYING_FROZEN:
3911 3912
		for_each_rcu_flavor(rsp)
			rcu_cleanup_dying_cpu(rsp);
3913
		break;
3914 3915 3916 3917 3918
	case CPU_DYING_IDLE:
		for_each_rcu_flavor(rsp) {
			rcu_cleanup_dying_idle_cpu(cpu, rsp);
		}
		break;
3919 3920 3921 3922
	case CPU_DEAD:
	case CPU_DEAD_FROZEN:
	case CPU_UP_CANCELED:
	case CPU_UP_CANCELED_FROZEN:
3923
		for_each_rcu_flavor(rsp) {
3924
			rcu_cleanup_dead_cpu(cpu, rsp);
3925 3926
			do_nocb_deferred_wakeup(per_cpu_ptr(rsp->rda, cpu));
		}
3927 3928 3929 3930
		break;
	default:
		break;
	}
3931
	return NOTIFY_OK;
3932 3933
}

3934 3935 3936 3937 3938 3939 3940
static int rcu_pm_notify(struct notifier_block *self,
			 unsigned long action, void *hcpu)
{
	switch (action) {
	case PM_HIBERNATION_PREPARE:
	case PM_SUSPEND_PREPARE:
		if (nr_cpu_ids <= 256) /* Expediting bad for large systems. */
3941
			rcu_expedite_gp();
3942 3943 3944
		break;
	case PM_POST_HIBERNATION:
	case PM_POST_SUSPEND:
3945 3946
		if (nr_cpu_ids <= 256) /* Expediting bad for large systems. */
			rcu_unexpedite_gp();
3947 3948 3949 3950 3951 3952 3953
		break;
	default:
		break;
	}
	return NOTIFY_OK;
}

3954
/*
3955
 * Spawn the kthreads that handle each RCU flavor's grace periods.
3956 3957 3958 3959
 */
static int __init rcu_spawn_gp_kthread(void)
{
	unsigned long flags;
3960
	int kthread_prio_in = kthread_prio;
3961 3962
	struct rcu_node *rnp;
	struct rcu_state *rsp;
3963
	struct sched_param sp;
3964 3965
	struct task_struct *t;

3966 3967 3968 3969 3970 3971 3972 3973 3974 3975 3976
	/* Force priority into range. */
	if (IS_ENABLED(CONFIG_RCU_BOOST) && kthread_prio < 1)
		kthread_prio = 1;
	else if (kthread_prio < 0)
		kthread_prio = 0;
	else if (kthread_prio > 99)
		kthread_prio = 99;
	if (kthread_prio != kthread_prio_in)
		pr_alert("rcu_spawn_gp_kthread(): Limited prio to %d from %d\n",
			 kthread_prio, kthread_prio_in);

3977
	rcu_scheduler_fully_active = 1;
3978
	for_each_rcu_flavor(rsp) {
3979
		t = kthread_create(rcu_gp_kthread, rsp, "%s", rsp->name);
3980 3981 3982 3983
		BUG_ON(IS_ERR(t));
		rnp = rcu_get_root(rsp);
		raw_spin_lock_irqsave(&rnp->lock, flags);
		rsp->gp_kthread = t;
3984 3985 3986 3987 3988
		if (kthread_prio) {
			sp.sched_priority = kthread_prio;
			sched_setscheduler_nocheck(t, SCHED_FIFO, &sp);
		}
		wake_up_process(t);
3989 3990
		raw_spin_unlock_irqrestore(&rnp->lock, flags);
	}
3991
	rcu_spawn_nocb_kthreads();
3992
	rcu_spawn_boost_kthreads();
3993 3994 3995 3996
	return 0;
}
early_initcall(rcu_spawn_gp_kthread);

3997 3998 3999 4000 4001 4002 4003 4004 4005 4006 4007 4008 4009 4010 4011
/*
 * This function is invoked towards the end of the scheduler's initialization
 * process.  Before this is called, the idle task might contain
 * RCU read-side critical sections (during which time, this idle
 * task is booting the system).  After this function is called, the
 * idle tasks are prohibited from containing RCU read-side critical
 * sections.  This function also enables RCU lockdep checking.
 */
void rcu_scheduler_starting(void)
{
	WARN_ON(num_online_cpus() != 1);
	WARN_ON(nr_context_switches() > 0);
	rcu_scheduler_active = 1;
}

4012 4013
/*
 * Compute the per-level fanout, either using the exact fanout specified
4014
 * or balancing the tree, depending on the rcu_fanout_exact boot parameter.
4015 4016 4017 4018 4019
 */
static void __init rcu_init_levelspread(struct rcu_state *rsp)
{
	int i;

4020
	if (rcu_fanout_exact) {
4021 4022
		rsp->levelspread[rcu_num_lvls - 1] = rcu_fanout_leaf;
		for (i = rcu_num_lvls - 2; i >= 0; i--)
4023
			rsp->levelspread[i] = RCU_FANOUT;
4024 4025 4026 4027 4028 4029 4030 4031 4032 4033
	} else {
		int ccur;
		int cprv;

		cprv = nr_cpu_ids;
		for (i = rcu_num_lvls - 1; i >= 0; i--) {
			ccur = rsp->levelcnt[i];
			rsp->levelspread[i] = (cprv + ccur - 1) / ccur;
			cprv = ccur;
		}
4034 4035 4036 4037 4038 4039
	}
}

/*
 * Helper function for rcu_init() that initializes one rcu_state structure.
 */
4040 4041
static void __init rcu_init_one(struct rcu_state *rsp,
		struct rcu_data __percpu *rda)
4042
{
4043 4044 4045 4046 4047 4048 4049 4050 4051 4052
	static const char * const buf[] = {
		"rcu_node_0",
		"rcu_node_1",
		"rcu_node_2",
		"rcu_node_3" };  /* Match MAX_RCU_LVLS */
	static const char * const fqs[] = {
		"rcu_node_fqs_0",
		"rcu_node_fqs_1",
		"rcu_node_fqs_2",
		"rcu_node_fqs_3" };  /* Match MAX_RCU_LVLS */
4053
	static u8 fl_mask = 0x1;
4054 4055 4056 4057 4058
	int cpustride = 1;
	int i;
	int j;
	struct rcu_node *rnp;

4059 4060
	BUILD_BUG_ON(MAX_RCU_LVLS > ARRAY_SIZE(buf));  /* Fix buf[] init! */

4061 4062 4063
	/* Silence gcc 4.8 false positive about array index out of range. */
	if (rcu_num_lvls <= 0 || rcu_num_lvls > RCU_NUM_LVLS)
		panic("rcu_init_one: rcu_num_lvls out of range");
4064

4065 4066
	/* Initialize the level-tracking arrays. */

4067 4068 4069
	for (i = 0; i < rcu_num_lvls; i++)
		rsp->levelcnt[i] = num_rcu_lvl[i];
	for (i = 1; i < rcu_num_lvls; i++)
4070 4071
		rsp->level[i] = rsp->level[i - 1] + rsp->levelcnt[i - 1];
	rcu_init_levelspread(rsp);
4072 4073
	rsp->flavor_mask = fl_mask;
	fl_mask <<= 1;
4074 4075 4076

	/* Initialize the elements themselves, starting from the leaves. */

4077
	for (i = rcu_num_lvls - 1; i >= 0; i--) {
4078 4079 4080
		cpustride *= rsp->levelspread[i];
		rnp = rsp->level[i];
		for (j = 0; j < rsp->levelcnt[i]; j++, rnp++) {
P
Paul E. McKenney 已提交
4081
			raw_spin_lock_init(&rnp->lock);
4082 4083
			lockdep_set_class_and_name(&rnp->lock,
						   &rcu_node_class[i], buf[i]);
4084 4085 4086
			raw_spin_lock_init(&rnp->fqslock);
			lockdep_set_class_and_name(&rnp->fqslock,
						   &rcu_fqs_class[i], fqs[i]);
4087 4088
			rnp->gpnum = rsp->gpnum;
			rnp->completed = rsp->completed;
4089 4090 4091 4092
			rnp->qsmask = 0;
			rnp->qsmaskinit = 0;
			rnp->grplo = j * cpustride;
			rnp->grphi = (j + 1) * cpustride - 1;
4093 4094
			if (rnp->grphi >= nr_cpu_ids)
				rnp->grphi = nr_cpu_ids - 1;
4095 4096 4097 4098 4099 4100 4101 4102 4103 4104 4105
			if (i == 0) {
				rnp->grpnum = 0;
				rnp->grpmask = 0;
				rnp->parent = NULL;
			} else {
				rnp->grpnum = j % rsp->levelspread[i - 1];
				rnp->grpmask = 1UL << rnp->grpnum;
				rnp->parent = rsp->level[i - 1] +
					      j / rsp->levelspread[i - 1];
			}
			rnp->level = i;
4106
			INIT_LIST_HEAD(&rnp->blkd_tasks);
4107
			rcu_init_one_nocb(rnp);
4108 4109
		}
	}
4110

4111
	init_waitqueue_head(&rsp->gp_wq);
4112
	rnp = rsp->level[rcu_num_lvls - 1];
4113
	for_each_possible_cpu(i) {
4114
		while (i > rnp->grphi)
4115
			rnp++;
4116
		per_cpu_ptr(rsp->rda, i)->mynode = rnp;
4117 4118
		rcu_boot_init_percpu_data(i, rsp);
	}
4119
	list_add(&rsp->flavors, &rcu_struct_flavors);
4120 4121
}

4122 4123
/*
 * Compute the rcu_node tree geometry from kernel parameters.  This cannot
4124
 * replace the definitions in tree.h because those are needed to size
4125 4126 4127 4128
 * the ->node array in the rcu_state structure.
 */
static void __init rcu_init_geometry(void)
{
4129
	ulong d;
4130 4131
	int i;
	int j;
4132
	int n = nr_cpu_ids;
4133 4134
	int rcu_capacity[MAX_RCU_LVLS + 1];

4135 4136 4137 4138 4139 4140 4141 4142 4143 4144 4145 4146 4147
	/*
	 * Initialize any unspecified boot parameters.
	 * The default values of jiffies_till_first_fqs and
	 * jiffies_till_next_fqs are set to the RCU_JIFFIES_TILL_FORCE_QS
	 * value, which is a function of HZ, then adding one for each
	 * RCU_JIFFIES_FQS_DIV CPUs that might be on the system.
	 */
	d = RCU_JIFFIES_TILL_FORCE_QS + nr_cpu_ids / RCU_JIFFIES_FQS_DIV;
	if (jiffies_till_first_fqs == ULONG_MAX)
		jiffies_till_first_fqs = d;
	if (jiffies_till_next_fqs == ULONG_MAX)
		jiffies_till_next_fqs = d;

4148
	/* If the compile-time values are accurate, just leave. */
4149
	if (rcu_fanout_leaf == RCU_FANOUT_LEAF &&
4150
	    nr_cpu_ids == NR_CPUS)
4151
		return;
4152 4153
	pr_info("RCU: Adjusting geometry for rcu_fanout_leaf=%d, nr_cpu_ids=%d\n",
		rcu_fanout_leaf, nr_cpu_ids);
4154 4155 4156 4157 4158 4159 4160 4161 4162

	/*
	 * Compute number of nodes that can be handled an rcu_node tree
	 * with the given number of levels.  Setting rcu_capacity[0] makes
	 * some of the arithmetic easier.
	 */
	rcu_capacity[0] = 1;
	rcu_capacity[1] = rcu_fanout_leaf;
	for (i = 2; i <= MAX_RCU_LVLS; i++)
4163
		rcu_capacity[i] = rcu_capacity[i - 1] * RCU_FANOUT;
4164 4165 4166 4167 4168 4169 4170 4171 4172

	/*
	 * The boot-time rcu_fanout_leaf parameter is only permitted
	 * to increase the leaf-level fanout, not decrease it.  Of course,
	 * the leaf-level fanout cannot exceed the number of bits in
	 * the rcu_node masks.  Finally, the tree must be able to accommodate
	 * the configured number of CPUs.  Complain and fall back to the
	 * compile-time values if these limits are exceeded.
	 */
4173
	if (rcu_fanout_leaf < RCU_FANOUT_LEAF ||
4174 4175 4176 4177 4178 4179 4180 4181 4182 4183 4184 4185 4186 4187 4188 4189 4190 4191 4192 4193 4194 4195 4196 4197 4198
	    rcu_fanout_leaf > sizeof(unsigned long) * 8 ||
	    n > rcu_capacity[MAX_RCU_LVLS]) {
		WARN_ON(1);
		return;
	}

	/* Calculate the number of rcu_nodes at each level of the tree. */
	for (i = 1; i <= MAX_RCU_LVLS; i++)
		if (n <= rcu_capacity[i]) {
			for (j = 0; j <= i; j++)
				num_rcu_lvl[j] =
					DIV_ROUND_UP(n, rcu_capacity[i - j]);
			rcu_num_lvls = i;
			for (j = i + 1; j <= MAX_RCU_LVLS; j++)
				num_rcu_lvl[j] = 0;
			break;
		}

	/* Calculate the total number of rcu_node structures. */
	rcu_num_nodes = 0;
	for (i = 0; i <= MAX_RCU_LVLS; i++)
		rcu_num_nodes += num_rcu_lvl[i];
	rcu_num_nodes -= n;
}

4199 4200 4201 4202 4203 4204 4205 4206 4207 4208 4209 4210 4211 4212 4213 4214 4215 4216 4217 4218 4219 4220
/*
 * Dump out the structure of the rcu_node combining tree associated
 * with the rcu_state structure referenced by rsp.
 */
static void __init rcu_dump_rcu_node_tree(struct rcu_state *rsp)
{
	int level = 0;
	struct rcu_node *rnp;

	pr_info("rcu_node tree layout dump\n");
	pr_info(" ");
	rcu_for_each_node_breadth_first(rsp, rnp) {
		if (rnp->level != level) {
			pr_cont("\n");
			pr_info(" ");
			level = rnp->level;
		}
		pr_cont("%d:%d ^%d  ", rnp->grplo, rnp->grphi, rnp->grpnum);
	}
	pr_cont("\n");
}

4221
void __init rcu_init(void)
4222
{
P
Paul E. McKenney 已提交
4223
	int cpu;
4224

4225 4226
	rcu_early_boot_tests();

4227
	rcu_bootup_announce();
4228
	rcu_init_geometry();
4229
	rcu_init_one(&rcu_bh_state, &rcu_bh_data);
4230
	rcu_init_one(&rcu_sched_state, &rcu_sched_data);
4231 4232
	if (dump_tree)
		rcu_dump_rcu_node_tree(&rcu_sched_state);
4233
	__rcu_init_preempt();
J
Jiang Fang 已提交
4234
	open_softirq(RCU_SOFTIRQ, rcu_process_callbacks);
4235 4236 4237 4238 4239 4240 4241

	/*
	 * We don't need protection against CPU-hotplug here because
	 * this is called early in boot, before either interrupts
	 * or the scheduler are operational.
	 */
	cpu_notifier(rcu_cpu_notify, 0);
4242
	pm_notifier(rcu_pm_notify, 0);
P
Paul E. McKenney 已提交
4243 4244
	for_each_online_cpu(cpu)
		rcu_cpu_notify(NULL, CPU_UP_PREPARE, (void *)(long)cpu);
4245 4246
}

4247
#include "tree_plugin.h"