tree.c 116.5 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/ftrace_event.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
struct rcu_state sname##_state = { \
95
	.level = { &sname##_state.node[0] }, \
96
	.call = cr, \
97
	.fqs_state = RCU_GP_IDLE, \
P
Paul E. McKenney 已提交
98 99
	.gpnum = 0UL - 300UL, \
	.completed = 0UL - 300UL, \
100
	.orphan_lock = __RAW_SPIN_LOCK_UNLOCKED(&sname##_state.orphan_lock), \
101 102
	.orphan_nxttail = &sname##_state.orphan_nxtlist, \
	.orphan_donetail = &sname##_state.orphan_donelist, \
103
	.barrier_mutex = __MUTEX_INITIALIZER(sname##_state.barrier_mutex), \
104
	.onoff_mutex = __MUTEX_INITIALIZER(sname##_state.onoff_mutex), \
105
	.name = RCU_STATE_NAME(sname), \
106
	.abbr = sabbr, \
107 108
}; \
DEFINE_PER_CPU(struct rcu_data, sname##_data)
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 *rcu_state_p;
114
LIST_HEAD(rcu_struct_flavors);
115

116 117
/* Increase (but not decrease) the CONFIG_RCU_FANOUT_LEAF at boot time. */
static int rcu_fanout_leaf = CONFIG_RCU_FANOUT_LEAF;
118
module_param(rcu_fanout_leaf, int, 0444);
119 120 121 122 123 124 125 126 127 128
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. */

129 130 131 132
/*
 * 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)
133
 * optimize synchronize_sched() to a simple barrier().  When this variable
134 135 136 137
 * 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.
 */
138 139 140
int rcu_scheduler_active __read_mostly;
EXPORT_SYMBOL_GPL(rcu_scheduler_active);

141 142 143 144 145 146 147 148 149 150 151 152 153 154
/*
 * 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;

155 156
#ifdef CONFIG_RCU_BOOST

157 158 159 160 161
/*
 * Control variables for per-CPU and per-rcu_node kthreads.  These
 * handle all flavors of RCU.
 */
static DEFINE_PER_CPU(struct task_struct *, rcu_cpu_kthread_task);
162
DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_status);
163
DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_loops);
164
DEFINE_PER_CPU(char, rcu_cpu_has_work);
165

166 167
#endif /* #ifdef CONFIG_RCU_BOOST */

T
Thomas Gleixner 已提交
168
static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu);
169 170
static void invoke_rcu_core(void);
static void invoke_rcu_callbacks(struct rcu_state *rsp, struct rcu_data *rdp);
171

172 173 174 175 176 177 178 179 180 181 182 183
/*
 * 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;

184 185 186 187 188 189 190 191 192 193
/*
 * Return true if an RCU grace period is in progress.  The ACCESS_ONCE()s
 * 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)
{
	return ACCESS_ONCE(rsp->completed) != ACCESS_ONCE(rsp->gpnum);
}

194
/*
195
 * Note a quiescent state.  Because we do not need to know
196
 * how many quiescent states passed, just if there was at least
197
 * one since the start of the grace period, this just sets a flag.
198
 * The caller must have disabled preemption.
199
 */
200
void rcu_sched_qs(void)
201
{
202 203 204 205 206 207
	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);
	}
208 209
}

210
void rcu_bh_qs(void)
211
{
212 213 214 215 216 217
	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);
	}
218
}
219

220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283
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 */
};

/*
 * 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. */
		if (ACCESS_ONCE(rdp->mynode->completed) !=
		    ACCESS_ONCE(rdp->cond_resched_completed))
			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);
}

284 285 286
/*
 * Note a context switch.  This is a quiescent state for RCU-sched,
 * and requires special handling for preemptible RCU.
287
 * The caller must have disabled preemption.
288 289 290
 */
void rcu_note_context_switch(int cpu)
{
291
	trace_rcu_utilization(TPS("Start context switch"));
292
	rcu_sched_qs();
293
	rcu_preempt_note_context_switch(cpu);
294 295
	if (unlikely(raw_cpu_read(rcu_sched_qs_mask)))
		rcu_momentary_dyntick_idle();
296
	trace_rcu_utilization(TPS("End context switch"));
297
}
298
EXPORT_SYMBOL_GPL(rcu_note_context_switch);
299

E
Eric Dumazet 已提交
300 301 302
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. */
303

E
Eric Dumazet 已提交
304 305 306
module_param(blimit, long, 0444);
module_param(qhimark, long, 0444);
module_param(qlowmark, long, 0444);
307

308 309
static ulong jiffies_till_first_fqs = ULONG_MAX;
static ulong jiffies_till_next_fqs = ULONG_MAX;
310 311 312 313

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

314 315 316 317 318 319 320
/*
 * 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);

321
static bool rcu_start_gp_advanced(struct rcu_state *rsp, struct rcu_node *rnp,
322
				  struct rcu_data *rdp);
323 324 325 326
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);
327
static void force_quiescent_state(struct rcu_state *rsp);
328
static int rcu_pending(int cpu);
329 330

/*
331
 * Return the number of RCU-sched batches processed thus far for debug & stats.
332
 */
333
long rcu_batches_completed_sched(void)
334
{
335
	return rcu_sched_state.completed;
336
}
337
EXPORT_SYMBOL_GPL(rcu_batches_completed_sched);
338 339 340 341 342 343 344 345 346 347

/*
 * Return the number of RCU BH batches processed thus far for debug & stats.
 */
long rcu_batches_completed_bh(void)
{
	return rcu_bh_state.completed;
}
EXPORT_SYMBOL_GPL(rcu_batches_completed_bh);

348 349 350 351 352
/*
 * Force a quiescent state.
 */
void rcu_force_quiescent_state(void)
{
353
	force_quiescent_state(rcu_state_p);
354 355 356
}
EXPORT_SYMBOL_GPL(rcu_force_quiescent_state);

357 358 359 360 361
/*
 * Force a quiescent state for RCU BH.
 */
void rcu_bh_force_quiescent_state(void)
{
362
	force_quiescent_state(&rcu_bh_state);
363 364 365
}
EXPORT_SYMBOL_GPL(rcu_bh_force_quiescent_state);

366 367 368 369 370 371 372 373 374 375 376 377 378 379 380
/*
 * 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);

381 382 383 384 385 386 387 388 389 390 391 392 393 394
/*
 * 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);

395 396 397 398 399 400 401 402 403 404
/*
 * 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:
405
		rsp = rcu_state_p;
406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427
		break;
	case RCU_BH_FLAVOR:
		rsp = &rcu_bh_state;
		break;
	case RCU_SCHED_FLAVOR:
		rsp = &rcu_sched_state;
		break;
	default:
		break;
	}
	if (rsp != NULL) {
		*flags = ACCESS_ONCE(rsp->gp_flags);
		*gpnum = ACCESS_ONCE(rsp->gpnum);
		*completed = ACCESS_ONCE(rsp->completed);
		return;
	}
	*flags = 0;
	*gpnum = 0;
	*completed = 0;
}
EXPORT_SYMBOL_GPL(rcutorture_get_gp_data);

428 429 430 431 432 433 434 435 436 437 438
/*
 * 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);

439 440 441 442 443
/*
 * Force a quiescent state for RCU-sched.
 */
void rcu_sched_force_quiescent_state(void)
{
444
	force_quiescent_state(&rcu_sched_state);
445 446 447
}
EXPORT_SYMBOL_GPL(rcu_sched_force_quiescent_state);

448 449 450 451 452 453
/*
 * 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
454 455
	return &rdp->nxtlist != rdp->nxttail[RCU_DONE_TAIL] &&
	       rdp->nxttail[RCU_DONE_TAIL] != NULL;
456 457
}

458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479
/*
 * 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);
	int idx = (ACCESS_ONCE(rnp->completed) + 1) & 0x1;
	int *fp = &rnp->need_future_gp[idx];

	return ACCESS_ONCE(*fp);
}

480
/*
481 482 483
 * Does the current CPU require a not-yet-started grace period?
 * The caller must have disabled interrupts to prevent races with
 * normal callback registry.
484 485 486 487
 */
static int
cpu_needs_another_gp(struct rcu_state *rsp, struct rcu_data *rdp)
{
488
	int i;
P
Paul E. McKenney 已提交
489

490 491
	if (rcu_gp_in_progress(rsp))
		return 0;  /* No, a grace period is already in progress. */
492
	if (rcu_future_needs_gp(rsp))
493
		return 1;  /* Yes, a no-CBs CPU needs one. */
494 495 496 497 498 499 500 501 502 503
	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] &&
		    ULONG_CMP_LT(ACCESS_ONCE(rsp->completed),
				 rdp->nxtcompleted[i]))
			return 1;  /* Yes, CBs for future grace period. */
	return 0; /* No grace period needed. */
504 505
}

506
/*
507
 * rcu_eqs_enter_common - current CPU is moving towards extended quiescent state
508 509 510 511 512
 *
 * 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.
 */
513 514
static void rcu_eqs_enter_common(struct rcu_dynticks *rdtp, long long oldval,
				bool user)
515
{
516 517 518
	struct rcu_state *rsp;
	struct rcu_data *rdp;

519
	trace_rcu_dyntick(TPS("Start"), oldval, rdtp->dynticks_nesting);
520
	if (!user && !is_idle_task(current)) {
521 522
		struct task_struct *idle __maybe_unused =
			idle_task(smp_processor_id());
523

524
		trace_rcu_dyntick(TPS("Error on entry: not idle task"), oldval, 0);
525
		ftrace_dump(DUMP_ORIG);
526 527 528
		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! */
529
	}
530 531 532 533
	for_each_rcu_flavor(rsp) {
		rdp = this_cpu_ptr(rsp->rda);
		do_nocb_deferred_wakeup(rdp);
	}
534
	rcu_prepare_for_idle(smp_processor_id());
535
	/* CPUs seeing atomic_inc() must see prior RCU read-side crit sects */
536
	smp_mb__before_atomic();  /* See above. */
537
	atomic_inc(&rdtp->dynticks);
538
	smp_mb__after_atomic();  /* Force ordering with next sojourn. */
539
	WARN_ON_ONCE(atomic_read(&rdtp->dynticks) & 0x1);
540
	rcu_dynticks_task_enter();
541 542

	/*
543
	 * It is illegal to enter an extended quiescent state while
544 545 546 547 548 549 550 551
	 * 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.");
552
}
553

554 555 556
/*
 * Enter an RCU extended quiescent state, which can be either the
 * idle loop or adaptive-tickless usermode execution.
557
 */
558
static void rcu_eqs_enter(bool user)
559
{
560
	long long oldval;
561 562
	struct rcu_dynticks *rdtp;

563
	rdtp = this_cpu_ptr(&rcu_dynticks);
564
	oldval = rdtp->dynticks_nesting;
565
	WARN_ON_ONCE((oldval & DYNTICK_TASK_NEST_MASK) == 0);
566
	if ((oldval & DYNTICK_TASK_NEST_MASK) == DYNTICK_TASK_NEST_VALUE) {
567
		rdtp->dynticks_nesting = 0;
568 569
		rcu_eqs_enter_common(rdtp, oldval, user);
	} else {
570
		rdtp->dynticks_nesting -= DYNTICK_TASK_NEST_VALUE;
571
	}
572
}
573 574 575 576 577 578 579 580 581 582 583 584 585 586 587

/**
 * 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)
{
588 589 590
	unsigned long flags;

	local_irq_save(flags);
591
	rcu_eqs_enter(false);
592
	rcu_sysidle_enter(this_cpu_ptr(&rcu_dynticks), 0);
593
	local_irq_restore(flags);
594
}
595
EXPORT_SYMBOL_GPL(rcu_idle_enter);
596

597
#ifdef CONFIG_RCU_USER_QS
598 599 600 601 602 603 604 605 606 607
/**
 * 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)
{
608
	rcu_eqs_enter(1);
609
}
610
#endif /* CONFIG_RCU_USER_QS */
611

612 613 614 615 616 617
/**
 * 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.
618
 *
619 620 621 622 623 624 625 626
 * 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.
627
 */
628
void rcu_irq_exit(void)
629 630
{
	unsigned long flags;
631
	long long oldval;
632 633 634
	struct rcu_dynticks *rdtp;

	local_irq_save(flags);
635
	rdtp = this_cpu_ptr(&rcu_dynticks);
636
	oldval = rdtp->dynticks_nesting;
637 638
	rdtp->dynticks_nesting--;
	WARN_ON_ONCE(rdtp->dynticks_nesting < 0);
639
	if (rdtp->dynticks_nesting)
640
		trace_rcu_dyntick(TPS("--="), oldval, rdtp->dynticks_nesting);
641
	else
642
		rcu_eqs_enter_common(rdtp, oldval, true);
643
	rcu_sysidle_enter(rdtp, 1);
644 645 646 647
	local_irq_restore(flags);
}

/*
648
 * rcu_eqs_exit_common - current CPU moving away from extended quiescent state
649 650 651 652 653
 *
 * 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.
 */
654 655
static void rcu_eqs_exit_common(struct rcu_dynticks *rdtp, long long oldval,
			       int user)
656
{
657
	rcu_dynticks_task_exit();
658
	smp_mb__before_atomic();  /* Force ordering w/previous sojourn. */
659 660
	atomic_inc(&rdtp->dynticks);
	/* CPUs seeing atomic_inc() must see later RCU read-side crit sects */
661
	smp_mb__after_atomic();  /* See above. */
662
	WARN_ON_ONCE(!(atomic_read(&rdtp->dynticks) & 0x1));
663
	rcu_cleanup_after_idle(smp_processor_id());
664
	trace_rcu_dyntick(TPS("End"), oldval, rdtp->dynticks_nesting);
665
	if (!user && !is_idle_task(current)) {
666 667
		struct task_struct *idle __maybe_unused =
			idle_task(smp_processor_id());
668

669
		trace_rcu_dyntick(TPS("Error on exit: not idle task"),
670
				  oldval, rdtp->dynticks_nesting);
671
		ftrace_dump(DUMP_ORIG);
672 673 674
		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! */
675 676 677
	}
}

678 679 680
/*
 * Exit an RCU extended quiescent state, which can be either the
 * idle loop or adaptive-tickless usermode execution.
681
 */
682
static void rcu_eqs_exit(bool user)
683 684 685 686
{
	struct rcu_dynticks *rdtp;
	long long oldval;

687
	rdtp = this_cpu_ptr(&rcu_dynticks);
688
	oldval = rdtp->dynticks_nesting;
689
	WARN_ON_ONCE(oldval < 0);
690
	if (oldval & DYNTICK_TASK_NEST_MASK) {
691
		rdtp->dynticks_nesting += DYNTICK_TASK_NEST_VALUE;
692
	} else {
693
		rdtp->dynticks_nesting = DYNTICK_TASK_EXIT_IDLE;
694 695
		rcu_eqs_exit_common(rdtp, oldval, user);
	}
696
}
697 698 699 700 701 702 703 704 705 706 707 708 709 710

/**
 * 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)
{
711 712 713
	unsigned long flags;

	local_irq_save(flags);
714
	rcu_eqs_exit(false);
715
	rcu_sysidle_exit(this_cpu_ptr(&rcu_dynticks), 0);
716
	local_irq_restore(flags);
717
}
718
EXPORT_SYMBOL_GPL(rcu_idle_exit);
719

720
#ifdef CONFIG_RCU_USER_QS
721 722 723 724 725 726 727 728
/**
 * 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)
{
729
	rcu_eqs_exit(1);
730
}
731
#endif /* CONFIG_RCU_USER_QS */
732

733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758
/**
 * 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);
759
	rdtp = this_cpu_ptr(&rcu_dynticks);
760 761 762
	oldval = rdtp->dynticks_nesting;
	rdtp->dynticks_nesting++;
	WARN_ON_ONCE(rdtp->dynticks_nesting == 0);
763
	if (oldval)
764
		trace_rcu_dyntick(TPS("++="), oldval, rdtp->dynticks_nesting);
765
	else
766
		rcu_eqs_exit_common(rdtp, oldval, true);
767
	rcu_sysidle_exit(rdtp, 1);
768 769 770 771 772 773 774 775 776 777 778 779
	local_irq_restore(flags);
}

/**
 * rcu_nmi_enter - inform RCU of entry to NMI context
 *
 * If the CPU was idle with dynamic ticks active, and there is no
 * irq handler running, this updates rdtp->dynticks_nmi to let the
 * RCU grace-period handling know that the CPU is active.
 */
void rcu_nmi_enter(void)
{
780
	struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks);
781

782 783
	if (rdtp->dynticks_nmi_nesting == 0 &&
	    (atomic_read(&rdtp->dynticks) & 0x1))
784
		return;
785
	rdtp->dynticks_nmi_nesting++;
786
	smp_mb__before_atomic();  /* Force delay from prior write. */
787 788
	atomic_inc(&rdtp->dynticks);
	/* CPUs seeing atomic_inc() must see later RCU read-side crit sects */
789
	smp_mb__after_atomic();  /* See above. */
790
	WARN_ON_ONCE(!(atomic_read(&rdtp->dynticks) & 0x1));
791 792 793 794 795 796 797 798 799 800 801
}

/**
 * rcu_nmi_exit - inform RCU of exit from NMI context
 *
 * If the CPU was idle with dynamic ticks active, and there is no
 * irq handler running, this updates rdtp->dynticks_nmi to let the
 * RCU grace-period handling know that the CPU is no longer active.
 */
void rcu_nmi_exit(void)
{
802
	struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks);
803

804 805
	if (rdtp->dynticks_nmi_nesting == 0 ||
	    --rdtp->dynticks_nmi_nesting != 0)
806
		return;
807
	/* CPUs seeing atomic_inc() must see prior RCU read-side crit sects */
808
	smp_mb__before_atomic();  /* See above. */
809
	atomic_inc(&rdtp->dynticks);
810
	smp_mb__after_atomic();  /* Force delay to next write. */
811
	WARN_ON_ONCE(atomic_read(&rdtp->dynticks) & 0x1);
812 813 814
}

/**
815 816 817 818 819 820 821
 * __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.
 */
822
bool notrace __rcu_is_watching(void)
823 824 825 826 827 828
{
	return atomic_read(this_cpu_ptr(&rcu_dynticks.dynticks)) & 0x1;
}

/**
 * rcu_is_watching - see if RCU thinks that the current CPU is idle
829
 *
830
 * If the current CPU is in its idle loop and is neither in an interrupt
831
 * or NMI handler, return true.
832
 */
833
bool notrace rcu_is_watching(void)
834
{
835
	bool ret;
836 837

	preempt_disable();
838
	ret = __rcu_is_watching();
839 840
	preempt_enable();
	return ret;
841
}
842
EXPORT_SYMBOL_GPL(rcu_is_watching);
843

844
#if defined(CONFIG_PROVE_RCU) && defined(CONFIG_HOTPLUG_CPU)
845 846 847 848 849 850 851

/*
 * 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
852 853 854 855 856 857 858 859 860 861 862
 * 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.
863 864 865 866 867 868
 *
 * Disable checking if in an NMI handler because we cannot safely report
 * errors from NMI handlers anyway.
 */
bool rcu_lockdep_current_cpu_online(void)
{
869 870
	struct rcu_data *rdp;
	struct rcu_node *rnp;
871 872 873
	bool ret;

	if (in_nmi())
F
Fengguang Wu 已提交
874
		return true;
875
	preempt_disable();
876
	rdp = this_cpu_ptr(&rcu_sched_data);
877 878
	rnp = rdp->mynode;
	ret = (rdp->grpmask & rnp->qsmaskinit) ||
879 880 881 882 883 884
	      !rcu_scheduler_fully_active;
	preempt_enable();
	return ret;
}
EXPORT_SYMBOL_GPL(rcu_lockdep_current_cpu_online);

885
#endif /* #if defined(CONFIG_PROVE_RCU) && defined(CONFIG_HOTPLUG_CPU) */
886

887
/**
888
 * rcu_is_cpu_rrupt_from_idle - see if idle or immediately interrupted from idle
889
 *
890 891 892
 * 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.
893
 */
894
static int rcu_is_cpu_rrupt_from_idle(void)
895
{
896
	return __this_cpu_read(rcu_dynticks.dynticks_nesting) <= 1;
897 898 899 900 901
}

/*
 * Snapshot the specified CPU's dynticks counter so that we can later
 * credit them with an implicit quiescent state.  Return 1 if this CPU
902
 * is in dynticks idle mode, which is an extended quiescent state.
903
 */
904 905
static int dyntick_save_progress_counter(struct rcu_data *rdp,
					 bool *isidle, unsigned long *maxj)
906
{
907
	rdp->dynticks_snap = atomic_add_return(0, &rdp->dynticks->dynticks);
908
	rcu_sysidle_check_cpu(rdp, isidle, maxj);
909 910 911 912 913 914
	if ((rdp->dynticks_snap & 0x1) == 0) {
		trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, TPS("dti"));
		return 1;
	} else {
		return 0;
	}
915 916
}

917 918 919 920 921 922
/*
 * This function really isn't for public consumption, but RCU is special in
 * that context switches can allow the state machine to make progress.
 */
extern void resched_cpu(int cpu);

923 924 925 926
/*
 * 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()
927
 * for this same CPU, or by virtue of having been offline.
928
 */
929 930
static int rcu_implicit_dynticks_qs(struct rcu_data *rdp,
				    bool *isidle, unsigned long *maxj)
931
{
932
	unsigned int curr;
933
	int *rcrmp;
934
	unsigned int snap;
935

936 937
	curr = (unsigned int)atomic_add_return(0, &rdp->dynticks->dynticks);
	snap = (unsigned int)rdp->dynticks_snap;
938 939 940 941 942 943 944 945 946

	/*
	 * 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.
	 */
947
	if ((curr & 0x1) == 0 || UINT_CMP_GE(curr, snap + 2)) {
948
		trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, TPS("dti"));
949 950 951 952
		rdp->dynticks_fqs++;
		return 1;
	}

953 954 955 956 957 958 959 960 961 962 963 964 965 966 967
	/*
	 * 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)) {
968
		trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, TPS("ofl"));
969 970 971
		rdp->offline_fqs++;
		return 1;
	}
972 973

	/*
974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992
	 * 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.
993
	 */
994 995 996
	rcrmp = &per_cpu(rcu_sched_qs_mask, rdp->cpu);
	if (ULONG_CMP_GE(jiffies,
			 rdp->rsp->gp_start + jiffies_till_sched_qs) ||
997
	    ULONG_CMP_GE(jiffies, rdp->rsp->jiffies_resched)) {
998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010
		if (!(ACCESS_ONCE(*rcrmp) & rdp->rsp->flavor_mask)) {
			ACCESS_ONCE(rdp->cond_resched_completed) =
				ACCESS_ONCE(rdp->mynode->completed);
			smp_mb(); /* ->cond_resched_completed before *rcrmp. */
			ACCESS_ONCE(*rcrmp) =
				ACCESS_ONCE(*rcrmp) + rdp->rsp->flavor_mask;
			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. */
		}
1011 1012
	}

1013
	return 0;
1014 1015 1016 1017
}

static void record_gp_stall_check_time(struct rcu_state *rsp)
{
1018
	unsigned long j = jiffies;
1019
	unsigned long j1;
1020 1021 1022

	rsp->gp_start = j;
	smp_wmb(); /* Record start time before stall time. */
1023
	j1 = rcu_jiffies_till_stall_check();
1024
	ACCESS_ONCE(rsp->jiffies_stall) = j + j1;
1025
	rsp->jiffies_resched = j + j1 / 2;
1026 1027
}

1028
/*
1029
 * Dump stacks of all tasks running on stalled CPUs.
1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047
 */
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);
	}
}

1048 1049 1050 1051 1052
static void print_other_cpu_stall(struct rcu_state *rsp)
{
	int cpu;
	long delta;
	unsigned long flags;
1053
	int ndetected = 0;
1054
	struct rcu_node *rnp = rcu_get_root(rsp);
1055
	long totqlen = 0;
1056 1057 1058

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

P
Paul E. McKenney 已提交
1059
	raw_spin_lock_irqsave(&rnp->lock, flags);
1060
	delta = jiffies - ACCESS_ONCE(rsp->jiffies_stall);
1061
	if (delta < RCU_STALL_RAT_DELAY || !rcu_gp_in_progress(rsp)) {
P
Paul E. McKenney 已提交
1062
		raw_spin_unlock_irqrestore(&rnp->lock, flags);
1063 1064
		return;
	}
1065
	ACCESS_ONCE(rsp->jiffies_stall) = jiffies + 3 * rcu_jiffies_till_stall_check() + 3;
P
Paul E. McKenney 已提交
1066
	raw_spin_unlock_irqrestore(&rnp->lock, flags);
1067

1068 1069 1070 1071 1072
	/*
	 * OK, time to rat on our buddy...
	 * See Documentation/RCU/stallwarn.txt for info on how to debug
	 * RCU CPU stall warnings.
	 */
1073
	pr_err("INFO: %s detected stalls on CPUs/tasks:",
1074
	       rsp->name);
1075
	print_cpu_stall_info_begin();
1076
	rcu_for_each_leaf_node(rsp, rnp) {
1077
		raw_spin_lock_irqsave(&rnp->lock, flags);
1078
		ndetected += rcu_print_task_stall(rnp);
1079 1080 1081 1082 1083 1084 1085 1086
		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++;
				}
		}
1087
		raw_spin_unlock_irqrestore(&rnp->lock, flags);
1088
	}
1089 1090 1091 1092 1093 1094 1095

	/*
	 * Now rat on any tasks that got kicked up to the root rcu_node
	 * due to CPU offlining.
	 */
	rnp = rcu_get_root(rsp);
	raw_spin_lock_irqsave(&rnp->lock, flags);
1096
	ndetected += rcu_print_task_stall(rnp);
1097 1098 1099
	raw_spin_unlock_irqrestore(&rnp->lock, flags);

	print_cpu_stall_info_end();
1100 1101
	for_each_possible_cpu(cpu)
		totqlen += per_cpu_ptr(rsp->rda, cpu)->qlen;
1102
	pr_cont("(detected by %d, t=%ld jiffies, g=%ld, c=%ld, q=%lu)\n",
1103
	       smp_processor_id(), (long)(jiffies - rsp->gp_start),
1104
	       (long)rsp->gpnum, (long)rsp->completed, totqlen);
1105
	if (ndetected == 0)
1106
		pr_err("INFO: Stall ended before state dump start\n");
1107
	else
1108
		rcu_dump_cpu_stacks(rsp);
1109

1110
	/* Complain about tasks blocking the grace period. */
1111 1112 1113

	rcu_print_detail_task_stall(rsp);

1114
	force_quiescent_state(rsp);  /* Kick them all. */
1115 1116 1117 1118
}

static void print_cpu_stall(struct rcu_state *rsp)
{
1119
	int cpu;
1120 1121
	unsigned long flags;
	struct rcu_node *rnp = rcu_get_root(rsp);
1122
	long totqlen = 0;
1123

1124 1125 1126 1127 1128
	/*
	 * OK, time to rat on ourselves...
	 * See Documentation/RCU/stallwarn.txt for info on how to debug
	 * RCU CPU stall warnings.
	 */
1129
	pr_err("INFO: %s self-detected stall on CPU", rsp->name);
1130 1131 1132
	print_cpu_stall_info_begin();
	print_cpu_stall_info(rsp, smp_processor_id());
	print_cpu_stall_info_end();
1133 1134
	for_each_possible_cpu(cpu)
		totqlen += per_cpu_ptr(rsp->rda, cpu)->qlen;
1135 1136 1137
	pr_cont(" (t=%lu jiffies g=%ld c=%ld q=%lu)\n",
		jiffies - rsp->gp_start,
		(long)rsp->gpnum, (long)rsp->completed, totqlen);
1138
	rcu_dump_cpu_stacks(rsp);
1139

P
Paul E. McKenney 已提交
1140
	raw_spin_lock_irqsave(&rnp->lock, flags);
1141 1142
	if (ULONG_CMP_GE(jiffies, ACCESS_ONCE(rsp->jiffies_stall)))
		ACCESS_ONCE(rsp->jiffies_stall) = jiffies +
1143
				     3 * rcu_jiffies_till_stall_check() + 3;
P
Paul E. McKenney 已提交
1144
	raw_spin_unlock_irqrestore(&rnp->lock, flags);
1145

1146 1147 1148 1149 1150 1151 1152 1153
	/*
	 * 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());
1154 1155 1156 1157
}

static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp)
{
1158 1159 1160
	unsigned long completed;
	unsigned long gpnum;
	unsigned long gps;
1161 1162
	unsigned long j;
	unsigned long js;
1163 1164
	struct rcu_node *rnp;

1165
	if (rcu_cpu_stall_suppress || !rcu_gp_in_progress(rsp))
1166
		return;
1167
	j = jiffies;
1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187

	/*
	 * 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.
	 */
	gpnum = ACCESS_ONCE(rsp->gpnum);
	smp_rmb(); /* Pick up ->gpnum first... */
1188
	js = ACCESS_ONCE(rsp->jiffies_stall);
1189 1190 1191 1192 1193 1194 1195 1196
	smp_rmb(); /* ...then ->jiffies_stall before the rest... */
	gps = ACCESS_ONCE(rsp->gp_start);
	smp_rmb(); /* ...and finally ->gp_start before ->completed. */
	completed = ACCESS_ONCE(rsp->completed);
	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. */
1197
	rnp = rdp->mynode;
1198
	if (rcu_gp_in_progress(rsp) &&
1199
	    (ACCESS_ONCE(rnp->qsmask) & rdp->grpmask)) {
1200 1201 1202 1203

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

1204 1205
	} else if (rcu_gp_in_progress(rsp) &&
		   ULONG_CMP_GE(j, js + RCU_STALL_RAT_DELAY)) {
1206

1207
		/* They had a few time units to dump stack, so complain. */
1208 1209 1210 1211
		print_other_cpu_stall(rsp);
	}
}

1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222
/**
 * 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)
{
1223 1224 1225
	struct rcu_state *rsp;

	for_each_rcu_flavor(rsp)
1226
		ACCESS_ONCE(rsp->jiffies_stall) = jiffies + ULONG_MAX / 2;
1227 1228
}

1229 1230 1231 1232 1233 1234 1235
/*
 * Initialize the specified rcu_data structure's callback list to empty.
 */
static void init_callback_list(struct rcu_data *rdp)
{
	int i;

1236 1237
	if (init_nocb_callback_list(rdp))
		return;
1238 1239 1240 1241 1242
	rdp->nxtlist = NULL;
	for (i = 0; i < RCU_NEXT_SIZE; i++)
		rdp->nxttail[i] = &rdp->nxtlist;
}

1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271
/*
 * 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;
}

1272 1273 1274 1275 1276
/*
 * 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,
1277
				unsigned long c, const char *s)
1278 1279 1280 1281 1282 1283 1284 1285 1286
{
	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
1287 1288
 * rcu_node structure's ->need_future_gp field.  Returns true if there
 * is reason to awaken the grace-period kthread.
1289 1290 1291
 *
 * The caller must hold the specified rcu_node structure's ->lock.
 */
1292 1293 1294
static bool __maybe_unused
rcu_start_future_gp(struct rcu_node *rnp, struct rcu_data *rdp,
		    unsigned long *c_out)
1295 1296 1297
{
	unsigned long c;
	int i;
1298
	bool ret = false;
1299 1300 1301 1302 1303 1304 1305
	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);
1306
	trace_rcu_future_gp(rnp, rdp, c, TPS("Startleaf"));
1307
	if (rnp->need_future_gp[c & 0x1]) {
1308
		trace_rcu_future_gp(rnp, rdp, c, TPS("Prestartleaf"));
1309
		goto out;
1310 1311 1312 1313 1314 1315 1316
	}

	/*
	 * 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
1317 1318 1319 1320 1321 1322 1323
	 * 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.
1324 1325
	 */
	if (rnp->gpnum != rnp->completed ||
1326
	    ACCESS_ONCE(rnp_root->gpnum) != ACCESS_ONCE(rnp_root->completed)) {
1327
		rnp->need_future_gp[c & 0x1]++;
1328
		trace_rcu_future_gp(rnp, rdp, c, TPS("Startedleaf"));
1329
		goto out;
1330 1331 1332 1333 1334 1335 1336
	}

	/*
	 * 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).
	 */
1337
	if (rnp != rnp_root) {
1338
		raw_spin_lock(&rnp_root->lock);
1339 1340
		smp_mb__after_unlock_lock();
	}
1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357

	/*
	 * 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]) {
1358
		trace_rcu_future_gp(rnp, rdp, c, TPS("Prestartedroot"));
1359 1360 1361 1362 1363 1364 1365 1366
		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) {
1367
		trace_rcu_future_gp(rnp, rdp, c, TPS("Startedleafroot"));
1368
	} else {
1369
		trace_rcu_future_gp(rnp, rdp, c, TPS("Startedroot"));
1370
		ret = rcu_start_gp_advanced(rdp->rsp, rnp_root, rdp);
1371 1372 1373 1374
	}
unlock_out:
	if (rnp != rnp_root)
		raw_spin_unlock(&rnp_root->lock);
1375 1376 1377 1378
out:
	if (c_out != NULL)
		*c_out = c;
	return ret;
1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395
}

/*
 * 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];
1396 1397
	trace_rcu_future_gp(rnp, rdp, c,
			    needmore ? TPS("CleanupMore") : TPS("Cleanup"));
1398 1399 1400
	return needmore;
}

1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416
/*
 * 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 ||
	    !ACCESS_ONCE(rsp->gp_flags) ||
	    !rsp->gp_kthread)
		return;
	wake_up(&rsp->gp_wq);
}

1417 1418 1419 1420 1421 1422 1423
/*
 * 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
1424 1425
 * not hurt to call it repeatedly.  Returns an flag saying that we should
 * awaken the RCU grace-period kthread.
1426 1427 1428
 *
 * The caller must hold rnp->lock with interrupts disabled.
 */
1429
static bool rcu_accelerate_cbs(struct rcu_state *rsp, struct rcu_node *rnp,
1430 1431 1432 1433
			       struct rcu_data *rdp)
{
	unsigned long c;
	int i;
1434
	bool ret;
1435 1436 1437

	/* If the CPU has no callbacks, nothing to do. */
	if (!rdp->nxttail[RCU_NEXT_TAIL] || !*rdp->nxttail[RCU_DONE_TAIL])
1438
		return false;
1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466

	/*
	 * 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)
1467
		return false;
1468 1469 1470 1471 1472 1473 1474 1475 1476 1477

	/*
	 * 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;
	}
1478
	/* Record any needed additional grace periods. */
1479
	ret = rcu_start_future_gp(rnp, rdp, NULL);
1480 1481 1482

	/* Trace depending on how much we were able to accelerate. */
	if (!*rdp->nxttail[RCU_WAIT_TAIL])
1483
		trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("AccWaitCB"));
1484
	else
1485
		trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("AccReadyCB"));
1486
	return ret;
1487 1488 1489 1490 1491 1492 1493 1494
}

/*
 * 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...
1495
 * Returns true if the RCU grace-period kthread needs to be awakened.
1496 1497 1498
 *
 * The caller must hold rnp->lock with interrupts disabled.
 */
1499
static bool rcu_advance_cbs(struct rcu_state *rsp, struct rcu_node *rnp,
1500 1501 1502 1503 1504 1505
			    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])
1506
		return false;
1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529

	/*
	 * 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. */
1530
	return rcu_accelerate_cbs(rsp, rnp, rdp);
1531 1532
}

1533
/*
1534 1535 1536
 * 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.
1537
 * Returns true if the grace-period kthread needs to be awakened.
1538
 */
1539 1540
static bool __note_gp_changes(struct rcu_state *rsp, struct rcu_node *rnp,
			      struct rcu_data *rdp)
1541
{
1542 1543
	bool ret;

1544
	/* Handle the ends of any preceding grace periods first. */
1545
	if (rdp->completed == rnp->completed) {
1546

1547
		/* No grace period end, so just accelerate recent callbacks. */
1548
		ret = rcu_accelerate_cbs(rsp, rnp, rdp);
1549

1550 1551 1552
	} else {

		/* Advance callbacks. */
1553
		ret = rcu_advance_cbs(rsp, rnp, rdp);
1554 1555 1556

		/* Remember that we saw this grace-period completion. */
		rdp->completed = rnp->completed;
1557
		trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("cpuend"));
1558
	}
1559

1560 1561 1562 1563 1564 1565 1566
	if (rdp->gpnum != rnp->gpnum) {
		/*
		 * 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;
1567
		trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("cpustart"));
1568 1569 1570 1571
		rdp->passed_quiesce = 0;
		rdp->qs_pending = !!(rnp->qsmask & rdp->grpmask);
		zero_cpu_stall_ticks(rdp);
	}
1572
	return ret;
1573 1574
}

1575
static void note_gp_changes(struct rcu_state *rsp, struct rcu_data *rdp)
1576 1577
{
	unsigned long flags;
1578
	bool needwake;
1579 1580 1581 1582
	struct rcu_node *rnp;

	local_irq_save(flags);
	rnp = rdp->mynode;
1583 1584
	if ((rdp->gpnum == ACCESS_ONCE(rnp->gpnum) &&
	     rdp->completed == ACCESS_ONCE(rnp->completed)) || /* w/out lock. */
1585 1586 1587 1588
	    !raw_spin_trylock(&rnp->lock)) { /* irqs already off, so later. */
		local_irq_restore(flags);
		return;
	}
1589
	smp_mb__after_unlock_lock();
1590
	needwake = __note_gp_changes(rsp, rnp, rdp);
1591
	raw_spin_unlock_irqrestore(&rnp->lock, flags);
1592 1593
	if (needwake)
		rcu_gp_kthread_wake(rsp);
1594 1595
}

1596
/*
1597
 * Initialize a new grace period.  Return 0 if no grace period required.
1598
 */
1599
static int rcu_gp_init(struct rcu_state *rsp)
1600 1601
{
	struct rcu_data *rdp;
1602
	struct rcu_node *rnp = rcu_get_root(rsp);
1603

1604
	rcu_bind_gp_kthread();
1605
	raw_spin_lock_irq(&rnp->lock);
1606
	smp_mb__after_unlock_lock();
1607
	if (!ACCESS_ONCE(rsp->gp_flags)) {
1608 1609 1610 1611
		/* Spurious wakeup, tell caller to go back to sleep.  */
		raw_spin_unlock_irq(&rnp->lock);
		return 0;
	}
1612
	ACCESS_ONCE(rsp->gp_flags) = 0; /* Clear all flags: New grace period. */
1613

1614 1615 1616 1617 1618
	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.
		 */
1619 1620 1621 1622 1623
		raw_spin_unlock_irq(&rnp->lock);
		return 0;
	}

	/* Advance to a new grace period and initialize state. */
1624
	record_gp_stall_check_time(rsp);
1625 1626
	/* Record GP times before starting GP, hence smp_store_release(). */
	smp_store_release(&rsp->gpnum, rsp->gpnum + 1);
1627
	trace_rcu_grace_period(rsp->name, rsp->gpnum, TPS("start"));
1628 1629 1630
	raw_spin_unlock_irq(&rnp->lock);

	/* Exclude any concurrent CPU-hotplug operations. */
1631
	mutex_lock(&rsp->onoff_mutex);
1632
	smp_mb__after_unlock_lock(); /* ->gpnum increment before GP! */
1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647

	/*
	 * 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) {
1648
		raw_spin_lock_irq(&rnp->lock);
1649
		smp_mb__after_unlock_lock();
1650
		rdp = this_cpu_ptr(rsp->rda);
1651 1652
		rcu_preempt_check_blocked_tasks(rnp);
		rnp->qsmask = rnp->qsmaskinit;
1653
		ACCESS_ONCE(rnp->gpnum) = rsp->gpnum;
1654
		WARN_ON_ONCE(rnp->completed != rsp->completed);
1655
		ACCESS_ONCE(rnp->completed) = rsp->completed;
1656
		if (rnp == rdp->mynode)
1657
			(void)__note_gp_changes(rsp, rnp, rdp);
1658 1659 1660 1661 1662
		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);
1663
		cond_resched_rcu_qs();
1664
	}
1665

1666
	mutex_unlock(&rsp->onoff_mutex);
1667 1668
	return 1;
}
1669

1670 1671 1672
/*
 * Do one round of quiescent-state forcing.
 */
1673
static int rcu_gp_fqs(struct rcu_state *rsp, int fqs_state_in)
1674 1675
{
	int fqs_state = fqs_state_in;
1676 1677
	bool isidle = false;
	unsigned long maxj;
1678 1679 1680 1681 1682
	struct rcu_node *rnp = rcu_get_root(rsp);

	rsp->n_force_qs++;
	if (fqs_state == RCU_SAVE_DYNTICK) {
		/* Collect dyntick-idle snapshots. */
1683
		if (is_sysidle_rcu_state(rsp)) {
1684
			isidle = true;
1685 1686
			maxj = jiffies - ULONG_MAX / 4;
		}
1687 1688
		force_qs_rnp(rsp, dyntick_save_progress_counter,
			     &isidle, &maxj);
1689
		rcu_sysidle_report_gp(rsp, isidle, maxj);
1690 1691 1692
		fqs_state = RCU_FORCE_QS;
	} else {
		/* Handle dyntick-idle and offline CPUs. */
1693
		isidle = false;
1694
		force_qs_rnp(rsp, rcu_implicit_dynticks_qs, &isidle, &maxj);
1695 1696 1697 1698
	}
	/* Clear flag to prevent immediate re-entry. */
	if (ACCESS_ONCE(rsp->gp_flags) & RCU_GP_FLAG_FQS) {
		raw_spin_lock_irq(&rnp->lock);
1699
		smp_mb__after_unlock_lock();
1700 1701
		ACCESS_ONCE(rsp->gp_flags) =
			ACCESS_ONCE(rsp->gp_flags) & ~RCU_GP_FLAG_FQS;
1702 1703 1704 1705 1706
		raw_spin_unlock_irq(&rnp->lock);
	}
	return fqs_state;
}

1707 1708 1709
/*
 * Clean up after the old grace period.
 */
1710
static void rcu_gp_cleanup(struct rcu_state *rsp)
1711 1712
{
	unsigned long gp_duration;
1713
	bool needgp = false;
1714
	int nocb = 0;
1715 1716
	struct rcu_data *rdp;
	struct rcu_node *rnp = rcu_get_root(rsp);
1717

1718
	raw_spin_lock_irq(&rnp->lock);
1719
	smp_mb__after_unlock_lock();
1720 1721 1722
	gp_duration = jiffies - rsp->gp_start;
	if (gp_duration > rsp->gp_max)
		rsp->gp_max = gp_duration;
1723

1724 1725 1726 1727 1728 1729 1730 1731
	/*
	 * 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.
	 */
1732
	raw_spin_unlock_irq(&rnp->lock);
1733

1734 1735 1736 1737 1738 1739 1740 1741 1742 1743
	/*
	 * 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) {
1744
		raw_spin_lock_irq(&rnp->lock);
1745
		smp_mb__after_unlock_lock();
1746
		ACCESS_ONCE(rnp->completed) = rsp->gpnum;
1747 1748
		rdp = this_cpu_ptr(rsp->rda);
		if (rnp == rdp->mynode)
1749
			needgp = __note_gp_changes(rsp, rnp, rdp) || needgp;
1750
		/* smp_mb() provided by prior unlock-lock pair. */
1751
		nocb += rcu_future_gp_cleanup(rsp, rnp);
1752
		raw_spin_unlock_irq(&rnp->lock);
1753
		cond_resched_rcu_qs();
1754
	}
1755 1756
	rnp = rcu_get_root(rsp);
	raw_spin_lock_irq(&rnp->lock);
1757
	smp_mb__after_unlock_lock(); /* Order GP before ->completed update. */
1758
	rcu_nocb_gp_set(rnp, nocb);
1759

1760 1761
	/* Declare grace period done. */
	ACCESS_ONCE(rsp->completed) = rsp->gpnum;
1762
	trace_rcu_grace_period(rsp->name, rsp->completed, TPS("end"));
1763
	rsp->fqs_state = RCU_GP_IDLE;
1764
	rdp = this_cpu_ptr(rsp->rda);
1765 1766 1767
	/* Advance CBs to reduce false positives below. */
	needgp = rcu_advance_cbs(rsp, rnp, rdp) || needgp;
	if (needgp || cpu_needs_another_gp(rsp, rdp)) {
1768
		ACCESS_ONCE(rsp->gp_flags) = RCU_GP_FLAG_INIT;
1769 1770 1771 1772
		trace_rcu_grace_period(rsp->name,
				       ACCESS_ONCE(rsp->gpnum),
				       TPS("newreq"));
	}
1773 1774 1775 1776 1777 1778 1779 1780
	raw_spin_unlock_irq(&rnp->lock);
}

/*
 * Body of kthread that handles grace periods.
 */
static int __noreturn rcu_gp_kthread(void *arg)
{
1781
	int fqs_state;
1782
	int gf;
1783
	unsigned long j;
1784
	int ret;
1785 1786 1787 1788 1789 1790 1791
	struct rcu_state *rsp = arg;
	struct rcu_node *rnp = rcu_get_root(rsp);

	for (;;) {

		/* Handle grace-period start. */
		for (;;) {
1792 1793 1794
			trace_rcu_grace_period(rsp->name,
					       ACCESS_ONCE(rsp->gpnum),
					       TPS("reqwait"));
1795
			rsp->gp_state = RCU_GP_WAIT_GPS;
1796
			wait_event_interruptible(rsp->gp_wq,
1797
						 ACCESS_ONCE(rsp->gp_flags) &
1798
						 RCU_GP_FLAG_INIT);
1799
			/* Locking provides needed memory barrier. */
1800
			if (rcu_gp_init(rsp))
1801
				break;
1802
			cond_resched_rcu_qs();
1803
			WARN_ON(signal_pending(current));
1804 1805 1806
			trace_rcu_grace_period(rsp->name,
					       ACCESS_ONCE(rsp->gpnum),
					       TPS("reqwaitsig"));
1807
		}
1808

1809 1810
		/* Handle quiescent-state forcing. */
		fqs_state = RCU_SAVE_DYNTICK;
1811 1812 1813 1814 1815
		j = jiffies_till_first_fqs;
		if (j > HZ) {
			j = HZ;
			jiffies_till_first_fqs = HZ;
		}
1816
		ret = 0;
1817
		for (;;) {
1818 1819
			if (!ret)
				rsp->jiffies_force_qs = jiffies + j;
1820 1821 1822
			trace_rcu_grace_period(rsp->name,
					       ACCESS_ONCE(rsp->gpnum),
					       TPS("fqswait"));
1823
			rsp->gp_state = RCU_GP_WAIT_FQS;
1824
			ret = wait_event_interruptible_timeout(rsp->gp_wq,
1825 1826
					((gf = ACCESS_ONCE(rsp->gp_flags)) &
					 RCU_GP_FLAG_FQS) ||
1827 1828
					(!ACCESS_ONCE(rnp->qsmask) &&
					 !rcu_preempt_blocked_readers_cgp(rnp)),
1829
					j);
1830
			/* Locking provides needed memory barriers. */
1831
			/* If grace period done, leave loop. */
1832
			if (!ACCESS_ONCE(rnp->qsmask) &&
1833
			    !rcu_preempt_blocked_readers_cgp(rnp))
1834
				break;
1835
			/* If time for quiescent-state forcing, do it. */
1836 1837
			if (ULONG_CMP_GE(jiffies, rsp->jiffies_force_qs) ||
			    (gf & RCU_GP_FLAG_FQS)) {
1838 1839 1840
				trace_rcu_grace_period(rsp->name,
						       ACCESS_ONCE(rsp->gpnum),
						       TPS("fqsstart"));
1841
				fqs_state = rcu_gp_fqs(rsp, fqs_state);
1842 1843 1844
				trace_rcu_grace_period(rsp->name,
						       ACCESS_ONCE(rsp->gpnum),
						       TPS("fqsend"));
1845
				cond_resched_rcu_qs();
1846 1847
			} else {
				/* Deal with stray signal. */
1848
				cond_resched_rcu_qs();
1849
				WARN_ON(signal_pending(current));
1850 1851 1852
				trace_rcu_grace_period(rsp->name,
						       ACCESS_ONCE(rsp->gpnum),
						       TPS("fqswaitsig"));
1853
			}
1854 1855 1856 1857 1858 1859 1860 1861
			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;
			}
1862
		}
1863 1864 1865

		/* Handle grace-period end. */
		rcu_gp_cleanup(rsp);
1866 1867 1868
	}
}

1869 1870 1871
/*
 * Start a new RCU grace period if warranted, re-initializing the hierarchy
 * in preparation for detecting the next grace period.  The caller must hold
1872
 * the root node's ->lock and hard irqs must be disabled.
1873 1874 1875 1876
 *
 * 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.
1877 1878
 *
 * Returns true if the grace-period kthread must be awakened.
1879
 */
1880
static bool
1881 1882
rcu_start_gp_advanced(struct rcu_state *rsp, struct rcu_node *rnp,
		      struct rcu_data *rdp)
1883
{
1884
	if (!rsp->gp_kthread || !cpu_needs_another_gp(rsp, rdp)) {
1885
		/*
1886
		 * Either we have not yet spawned the grace-period
1887 1888
		 * task, this CPU does not need another grace period,
		 * or a grace period is already in progress.
1889
		 * Either way, don't start a new grace period.
1890
		 */
1891
		return false;
1892
	}
1893
	ACCESS_ONCE(rsp->gp_flags) = RCU_GP_FLAG_INIT;
1894 1895
	trace_rcu_grace_period(rsp->name, ACCESS_ONCE(rsp->gpnum),
			       TPS("newreq"));
1896

1897 1898
	/*
	 * We can't do wakeups while holding the rnp->lock, as that
1899
	 * could cause possible deadlocks with the rq->lock. Defer
1900
	 * the wakeup to our caller.
1901
	 */
1902
	return true;
1903 1904
}

1905 1906 1907 1908 1909 1910
/*
 * 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.
1911 1912
 *
 * Returns true if the grace-period kthread needs to be awakened.
1913
 */
1914
static bool rcu_start_gp(struct rcu_state *rsp)
1915 1916 1917
{
	struct rcu_data *rdp = this_cpu_ptr(rsp->rda);
	struct rcu_node *rnp = rcu_get_root(rsp);
1918
	bool ret = false;
1919 1920 1921 1922 1923 1924 1925 1926 1927

	/*
	 * 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!
	 */
1928 1929 1930
	ret = rcu_advance_cbs(rsp, rnp, rdp) || ret;
	ret = rcu_start_gp_advanced(rsp, rnp, rdp) || ret;
	return ret;
1931 1932
}

1933
/*
P
Paul E. McKenney 已提交
1934 1935 1936
 * 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
1937 1938
 * if one is needed.  Note that the caller must hold rnp->lock, which
 * is released before return.
1939
 */
P
Paul E. McKenney 已提交
1940
static void rcu_report_qs_rsp(struct rcu_state *rsp, unsigned long flags)
1941
	__releases(rcu_get_root(rsp)->lock)
1942
{
1943
	WARN_ON_ONCE(!rcu_gp_in_progress(rsp));
1944
	raw_spin_unlock_irqrestore(&rcu_get_root(rsp)->lock, flags);
1945
	rcu_gp_kthread_wake(rsp);
1946 1947
}

1948
/*
P
Paul E. McKenney 已提交
1949 1950 1951 1952 1953 1954
 * 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
 * must be represented by the same rcu_node structure (which need not be
 * a leaf rcu_node structure, though it often will be).  That structure's
 * lock must be held upon entry, and it is released before return.
1955 1956
 */
static void
P
Paul E. McKenney 已提交
1957 1958
rcu_report_qs_rnp(unsigned long mask, struct rcu_state *rsp,
		  struct rcu_node *rnp, unsigned long flags)
1959 1960
	__releases(rnp->lock)
{
1961 1962
	struct rcu_node *rnp_c;

1963 1964 1965 1966 1967
	/* Walk up the rcu_node hierarchy. */
	for (;;) {
		if (!(rnp->qsmask & mask)) {

			/* Our bit has already been cleared, so done. */
P
Paul E. McKenney 已提交
1968
			raw_spin_unlock_irqrestore(&rnp->lock, flags);
1969 1970 1971
			return;
		}
		rnp->qsmask &= ~mask;
1972 1973 1974 1975
		trace_rcu_quiescent_state_report(rsp->name, rnp->gpnum,
						 mask, rnp->qsmask, rnp->level,
						 rnp->grplo, rnp->grphi,
						 !!rnp->gp_tasks);
1976
		if (rnp->qsmask != 0 || rcu_preempt_blocked_readers_cgp(rnp)) {
1977 1978

			/* Other bits still set at this level, so done. */
P
Paul E. McKenney 已提交
1979
			raw_spin_unlock_irqrestore(&rnp->lock, flags);
1980 1981 1982 1983 1984 1985 1986 1987 1988
			return;
		}
		mask = rnp->grpmask;
		if (rnp->parent == NULL) {

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

			break;
		}
P
Paul E. McKenney 已提交
1989
		raw_spin_unlock_irqrestore(&rnp->lock, flags);
1990
		rnp_c = rnp;
1991
		rnp = rnp->parent;
P
Paul E. McKenney 已提交
1992
		raw_spin_lock_irqsave(&rnp->lock, flags);
1993
		smp_mb__after_unlock_lock();
1994
		WARN_ON_ONCE(rnp_c->qsmask);
1995 1996 1997 1998
	}

	/*
	 * Get here if we are the last CPU to pass through a quiescent
P
Paul E. McKenney 已提交
1999
	 * state for this grace period.  Invoke rcu_report_qs_rsp()
2000
	 * to clean up and start the next grace period if one is needed.
2001
	 */
P
Paul E. McKenney 已提交
2002
	rcu_report_qs_rsp(rsp, flags); /* releases rnp->lock. */
2003 2004 2005
}

/*
P
Paul E. McKenney 已提交
2006 2007 2008 2009 2010 2011 2012
 * 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!
2013 2014
 */
static void
2015
rcu_report_qs_rdp(int cpu, struct rcu_state *rsp, struct rcu_data *rdp)
2016 2017 2018
{
	unsigned long flags;
	unsigned long mask;
2019
	bool needwake;
2020 2021 2022
	struct rcu_node *rnp;

	rnp = rdp->mynode;
P
Paul E. McKenney 已提交
2023
	raw_spin_lock_irqsave(&rnp->lock, flags);
2024
	smp_mb__after_unlock_lock();
2025 2026
	if (rdp->passed_quiesce == 0 || rdp->gpnum != rnp->gpnum ||
	    rnp->completed == rnp->gpnum) {
2027 2028

		/*
2029 2030 2031 2032
		 * 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.
2033
		 */
2034
		rdp->passed_quiesce = 0;	/* need qs for new gp. */
P
Paul E. McKenney 已提交
2035
		raw_spin_unlock_irqrestore(&rnp->lock, flags);
2036 2037 2038 2039
		return;
	}
	mask = rdp->grpmask;
	if ((rnp->qsmask & mask) == 0) {
P
Paul E. McKenney 已提交
2040
		raw_spin_unlock_irqrestore(&rnp->lock, flags);
2041 2042 2043 2044 2045 2046 2047
	} 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.
		 */
2048
		needwake = rcu_accelerate_cbs(rsp, rnp, rdp);
2049

P
Paul E. McKenney 已提交
2050
		rcu_report_qs_rnp(mask, rsp, rnp, flags); /* rlses rnp->lock */
2051 2052
		if (needwake)
			rcu_gp_kthread_wake(rsp);
2053 2054 2055 2056 2057 2058 2059 2060 2061 2062 2063 2064
	}
}

/*
 * 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)
{
2065 2066
	/* Check for grace-period ends and beginnings. */
	note_gp_changes(rsp, rdp);
2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078

	/*
	 * 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.
	 */
2079
	if (!rdp->passed_quiesce)
2080 2081
		return;

P
Paul E. McKenney 已提交
2082 2083 2084 2085
	/*
	 * Tell RCU we are done (but rcu_report_qs_rdp() will be the
	 * judge of that).
	 */
2086
	rcu_report_qs_rdp(rdp->cpu, rsp, rdp);
2087 2088 2089 2090
}

#ifdef CONFIG_HOTPLUG_CPU

2091
/*
2092 2093
 * Send the specified CPU's RCU callbacks to the orphanage.  The
 * specified CPU must be offline, and the caller must hold the
2094
 * ->orphan_lock.
2095
 */
2096 2097 2098
static void
rcu_send_cbs_to_orphanage(int cpu, struct rcu_state *rsp,
			  struct rcu_node *rnp, struct rcu_data *rdp)
2099
{
P
Paul E. McKenney 已提交
2100
	/* No-CBs CPUs do not have orphanable callbacks. */
2101
	if (rcu_is_nocb_cpu(rdp->cpu))
P
Paul E. McKenney 已提交
2102 2103
		return;

2104 2105
	/*
	 * Orphan the callbacks.  First adjust the counts.  This is safe
2106 2107
	 * because _rcu_barrier() excludes CPU-hotplug operations, so it
	 * cannot be running now.  Thus no memory barrier is required.
2108
	 */
2109
	if (rdp->nxtlist != NULL) {
2110 2111 2112
		rsp->qlen_lazy += rdp->qlen_lazy;
		rsp->qlen += rdp->qlen;
		rdp->n_cbs_orphaned += rdp->qlen;
2113
		rdp->qlen_lazy = 0;
2114
		ACCESS_ONCE(rdp->qlen) = 0;
2115 2116 2117
	}

	/*
2118 2119 2120 2121 2122 2123 2124
	 * 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.
2125
	 */
2126 2127 2128 2129
	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;
2130 2131 2132
	}

	/*
2133 2134 2135
	 * 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.
2136
	 */
2137
	if (rdp->nxtlist != NULL) {
2138 2139
		*rsp->orphan_donetail = rdp->nxtlist;
		rsp->orphan_donetail = rdp->nxttail[RCU_DONE_TAIL];
2140
	}
2141

2142
	/* Finally, initialize the rcu_data structure's list to empty.  */
2143
	init_callback_list(rdp);
2144 2145 2146 2147
}

/*
 * Adopt the RCU callbacks from the specified rcu_state structure's
2148
 * orphanage.  The caller must hold the ->orphan_lock.
2149
 */
2150
static void rcu_adopt_orphan_cbs(struct rcu_state *rsp, unsigned long flags)
2151 2152
{
	int i;
2153
	struct rcu_data *rdp = raw_cpu_ptr(rsp->rda);
2154

P
Paul E. McKenney 已提交
2155
	/* No-CBs CPUs are handled specially. */
2156
	if (rcu_nocb_adopt_orphan_cbs(rsp, rdp, flags))
P
Paul E. McKenney 已提交
2157 2158
		return;

2159 2160 2161 2162
	/* Do the accounting first. */
	rdp->qlen_lazy += rsp->qlen_lazy;
	rdp->qlen += rsp->qlen;
	rdp->n_cbs_adopted += rsp->qlen;
2163 2164
	if (rsp->qlen_lazy != rsp->qlen)
		rcu_idle_count_callbacks_posted();
2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175 2176 2177 2178 2179 2180 2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197 2198 2199 2200 2201 2202 2203
	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);

	RCU_TRACE(mask = rdp->grpmask);
2204 2205
	trace_rcu_grace_period(rsp->name,
			       rnp->gpnum + 1 - !!(rnp->qsmask & mask),
2206
			       TPS("cpuofl"));
2207 2208 2209
}

/*
2210
 * The CPU has been completely removed, and some other CPU is reporting
2211 2212
 * this fact from process context.  Do the remainder of the cleanup,
 * including orphaning the outgoing CPU's RCU callbacks, and also
2213 2214
 * 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.
2215
 */
2216
static void rcu_cleanup_dead_cpu(int cpu, struct rcu_state *rsp)
2217
{
2218 2219 2220
	unsigned long flags;
	unsigned long mask;
	int need_report = 0;
2221
	struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
2222
	struct rcu_node *rnp = rdp->mynode;  /* Outgoing CPU's rdp & rnp. */
2223

2224
	/* Adjust any no-longer-needed kthreads. */
T
Thomas Gleixner 已提交
2225
	rcu_boost_kthread_setaffinity(rnp, -1);
2226 2227

	/* Exclude any attempts to start a new grace period. */
2228
	mutex_lock(&rsp->onoff_mutex);
2229
	raw_spin_lock_irqsave(&rsp->orphan_lock, flags);
2230

2231 2232
	/* Orphan the dead CPU's callbacks, and adopt them if appropriate. */
	rcu_send_cbs_to_orphanage(cpu, rsp, rnp, rdp);
2233
	rcu_adopt_orphan_cbs(rsp, flags);
2234

2235 2236 2237 2238
	/* Remove the outgoing CPU from the masks in the rcu_node hierarchy. */
	mask = rdp->grpmask;	/* rnp->grplo is constant. */
	do {
		raw_spin_lock(&rnp->lock);	/* irqs already disabled. */
2239
		smp_mb__after_unlock_lock();
2240 2241 2242 2243 2244 2245 2246 2247 2248 2249 2250 2251 2252 2253 2254 2255 2256
		rnp->qsmaskinit &= ~mask;
		if (rnp->qsmaskinit != 0) {
			if (rnp != rdp->mynode)
				raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */
			break;
		}
		if (rnp == rdp->mynode)
			need_report = rcu_preempt_offline_tasks(rsp, rnp, rdp);
		else
			raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */
		mask = rnp->grpmask;
		rnp = rnp->parent;
	} while (rnp != NULL);

	/*
	 * We still hold the leaf rcu_node structure lock here, and
	 * irqs are still disabled.  The reason for this subterfuge is
2257
	 * because invoking rcu_report_unblock_qs_rnp() with ->orphan_lock
2258 2259
	 * held leads to deadlock.
	 */
2260
	raw_spin_unlock(&rsp->orphan_lock); /* irqs remain disabled. */
2261 2262 2263 2264 2265 2266 2267
	rnp = rdp->mynode;
	if (need_report & RCU_OFL_TASKS_NORM_GP)
		rcu_report_unblock_qs_rnp(rnp, flags);
	else
		raw_spin_unlock_irqrestore(&rnp->lock, flags);
	if (need_report & RCU_OFL_TASKS_EXP_GP)
		rcu_report_exp_rnp(rsp, rnp, true);
2268 2269 2270
	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);
2271 2272 2273
	init_callback_list(rdp);
	/* Disallow further callbacks on this CPU. */
	rdp->nxttail[RCU_NEXT_TAIL] = NULL;
2274
	mutex_unlock(&rsp->onoff_mutex);
2275 2276 2277 2278
}

#else /* #ifdef CONFIG_HOTPLUG_CPU */

2279
static void rcu_cleanup_dying_cpu(struct rcu_state *rsp)
2280 2281 2282
{
}

2283
static void rcu_cleanup_dead_cpu(int cpu, struct rcu_state *rsp)
2284 2285 2286 2287 2288 2289 2290 2291 2292
{
}

#endif /* #else #ifdef CONFIG_HOTPLUG_CPU */

/*
 * Invoke any RCU callbacks that have made it to the end of their grace
 * period.  Thottle as specified by rdp->blimit.
 */
2293
static void rcu_do_batch(struct rcu_state *rsp, struct rcu_data *rdp)
2294 2295 2296
{
	unsigned long flags;
	struct rcu_head *next, *list, **tail;
E
Eric Dumazet 已提交
2297 2298
	long bl, count, count_lazy;
	int i;
2299

2300
	/* If no callbacks are ready, just return. */
2301
	if (!cpu_has_callbacks_ready_to_invoke(rdp)) {
2302
		trace_rcu_batch_start(rsp->name, rdp->qlen_lazy, rdp->qlen, 0);
2303 2304 2305
		trace_rcu_batch_end(rsp->name, 0, !!ACCESS_ONCE(rdp->nxtlist),
				    need_resched(), is_idle_task(current),
				    rcu_is_callbacks_kthread());
2306
		return;
2307
	}
2308 2309 2310 2311 2312 2313

	/*
	 * Extract the list of ready callbacks, disabling to prevent
	 * races with call_rcu() from interrupt handlers.
	 */
	local_irq_save(flags);
2314
	WARN_ON_ONCE(cpu_is_offline(smp_processor_id()));
2315
	bl = rdp->blimit;
2316
	trace_rcu_batch_start(rsp->name, rdp->qlen_lazy, rdp->qlen, bl);
2317 2318 2319 2320
	list = rdp->nxtlist;
	rdp->nxtlist = *rdp->nxttail[RCU_DONE_TAIL];
	*rdp->nxttail[RCU_DONE_TAIL] = NULL;
	tail = rdp->nxttail[RCU_DONE_TAIL];
2321 2322 2323
	for (i = RCU_NEXT_SIZE - 1; i >= 0; i--)
		if (rdp->nxttail[i] == rdp->nxttail[RCU_DONE_TAIL])
			rdp->nxttail[i] = &rdp->nxtlist;
2324 2325 2326
	local_irq_restore(flags);

	/* Invoke callbacks. */
2327
	count = count_lazy = 0;
2328 2329 2330
	while (list) {
		next = list->next;
		prefetch(next);
2331
		debug_rcu_head_unqueue(list);
2332 2333
		if (__rcu_reclaim(rsp->name, list))
			count_lazy++;
2334
		list = next;
2335 2336 2337 2338
		/* Stop only if limit reached and CPU has something to do. */
		if (++count >= bl &&
		    (need_resched() ||
		     (!is_idle_task(current) && !rcu_is_callbacks_kthread())))
2339 2340 2341 2342
			break;
	}

	local_irq_save(flags);
2343 2344 2345
	trace_rcu_batch_end(rsp->name, count, !!list, need_resched(),
			    is_idle_task(current),
			    rcu_is_callbacks_kthread());
2346 2347 2348 2349 2350

	/* Update count, and requeue any remaining callbacks. */
	if (list != NULL) {
		*tail = rdp->nxtlist;
		rdp->nxtlist = list;
2351 2352 2353
		for (i = 0; i < RCU_NEXT_SIZE; i++)
			if (&rdp->nxtlist == rdp->nxttail[i])
				rdp->nxttail[i] = tail;
2354 2355 2356
			else
				break;
	}
2357 2358
	smp_mb(); /* List handling before counting for rcu_barrier(). */
	rdp->qlen_lazy -= count_lazy;
2359
	ACCESS_ONCE(rdp->qlen) = rdp->qlen - count;
2360
	rdp->n_cbs_invoked += count;
2361 2362 2363 2364 2365

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

2366 2367 2368 2369 2370 2371
	/* 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;
2372
	WARN_ON_ONCE((rdp->nxtlist == NULL) != (rdp->qlen == 0));
2373

2374 2375
	local_irq_restore(flags);

2376
	/* Re-invoke RCU core processing if there are callbacks remaining. */
2377
	if (cpu_has_callbacks_ready_to_invoke(rdp))
2378
		invoke_rcu_core();
2379 2380 2381 2382 2383
}

/*
 * 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).
2384
 * Also schedule RCU core processing.
2385
 *
2386
 * This function must be called from hardirq context.  It is normally
2387 2388 2389 2390 2391
 * invoked from the scheduling-clock interrupt.  If rcu_pending returns
 * false, there is no point in invoking rcu_check_callbacks().
 */
void rcu_check_callbacks(int cpu, int user)
{
2392
	trace_rcu_utilization(TPS("Start scheduler-tick"));
2393
	increment_cpu_stall_ticks();
2394
	if (user || rcu_is_cpu_rrupt_from_idle()) {
2395 2396 2397 2398 2399

		/*
		 * 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
2400
		 * a quiescent state, so note it.
2401 2402
		 *
		 * No memory barrier is required here because both
2403 2404 2405
		 * 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.
2406 2407
		 */

2408 2409
		rcu_sched_qs();
		rcu_bh_qs();
2410 2411 2412 2413 2414 2415 2416

	} 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
2417
		 * critical section, so note it.
2418 2419
		 */

2420
		rcu_bh_qs();
2421
	}
2422
	rcu_preempt_check_callbacks(cpu);
2423
	if (rcu_pending(cpu))
2424
		invoke_rcu_core();
P
Paul E. McKenney 已提交
2425 2426
	if (user)
		rcu_note_voluntary_context_switch(current);
2427
	trace_rcu_utilization(TPS("End scheduler-tick"));
2428 2429 2430 2431 2432
}

/*
 * Scan the leaf rcu_node structures, processing dyntick state for any that
 * have not yet encountered a quiescent state, using the function specified.
2433 2434
 * Also initiate boosting for any threads blocked on the root rcu_node.
 *
2435
 * The caller must have suppressed start of new grace periods.
2436
 */
2437 2438 2439 2440
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)
2441 2442 2443 2444 2445
{
	unsigned long bit;
	int cpu;
	unsigned long flags;
	unsigned long mask;
2446
	struct rcu_node *rnp;
2447

2448
	rcu_for_each_leaf_node(rsp, rnp) {
2449
		cond_resched_rcu_qs();
2450
		mask = 0;
P
Paul E. McKenney 已提交
2451
		raw_spin_lock_irqsave(&rnp->lock, flags);
2452
		smp_mb__after_unlock_lock();
2453
		if (!rcu_gp_in_progress(rsp)) {
P
Paul E. McKenney 已提交
2454
			raw_spin_unlock_irqrestore(&rnp->lock, flags);
2455
			return;
2456
		}
2457
		if (rnp->qsmask == 0) {
2458
			rcu_initiate_boost(rnp, flags); /* releases rnp->lock */
2459 2460
			continue;
		}
2461
		cpu = rnp->grplo;
2462
		bit = 1;
2463
		for (; cpu <= rnp->grphi; cpu++, bit <<= 1) {
2464 2465
			if ((rnp->qsmask & bit) != 0) {
				if ((rnp->qsmaskinit & bit) != 0)
2466
					*isidle = false;
2467 2468 2469
				if (f(per_cpu_ptr(rsp->rda, cpu), isidle, maxj))
					mask |= bit;
			}
2470
		}
2471
		if (mask != 0) {
2472

P
Paul E. McKenney 已提交
2473 2474
			/* rcu_report_qs_rnp() releases rnp->lock. */
			rcu_report_qs_rnp(mask, rsp, rnp, flags);
2475 2476
			continue;
		}
P
Paul E. McKenney 已提交
2477
		raw_spin_unlock_irqrestore(&rnp->lock, flags);
2478
	}
2479
	rnp = rcu_get_root(rsp);
2480 2481
	if (rnp->qsmask == 0) {
		raw_spin_lock_irqsave(&rnp->lock, flags);
2482
		smp_mb__after_unlock_lock();
2483 2484
		rcu_initiate_boost(rnp, flags); /* releases rnp->lock. */
	}
2485 2486 2487 2488 2489 2490
}

/*
 * Force quiescent states on reluctant CPUs, and also detect which
 * CPUs are in dyntick-idle mode.
 */
2491
static void force_quiescent_state(struct rcu_state *rsp)
2492 2493
{
	unsigned long flags;
2494 2495 2496 2497 2498
	bool ret;
	struct rcu_node *rnp;
	struct rcu_node *rnp_old = NULL;

	/* Funnel through hierarchy to reduce memory contention. */
2499
	rnp = __this_cpu_read(rsp->rda->mynode);
2500 2501 2502 2503 2504 2505
	for (; rnp != NULL; rnp = rnp->parent) {
		ret = (ACCESS_ONCE(rsp->gp_flags) & RCU_GP_FLAG_FQS) ||
		      !raw_spin_trylock(&rnp->fqslock);
		if (rnp_old != NULL)
			raw_spin_unlock(&rnp_old->fqslock);
		if (ret) {
2506
			rsp->n_force_qs_lh++;
2507 2508 2509 2510 2511
			return;
		}
		rnp_old = rnp;
	}
	/* rnp_old == rcu_get_root(rsp), rnp == NULL. */
2512

2513 2514
	/* Reached the root of the rcu_node tree, acquire lock. */
	raw_spin_lock_irqsave(&rnp_old->lock, flags);
2515
	smp_mb__after_unlock_lock();
2516 2517
	raw_spin_unlock(&rnp_old->fqslock);
	if (ACCESS_ONCE(rsp->gp_flags) & RCU_GP_FLAG_FQS) {
2518
		rsp->n_force_qs_lh++;
2519
		raw_spin_unlock_irqrestore(&rnp_old->lock, flags);
2520
		return;  /* Someone beat us to it. */
2521
	}
2522 2523
	ACCESS_ONCE(rsp->gp_flags) =
		ACCESS_ONCE(rsp->gp_flags) | RCU_GP_FLAG_FQS;
2524
	raw_spin_unlock_irqrestore(&rnp_old->lock, flags);
2525
	rcu_gp_kthread_wake(rsp);
2526 2527 2528
}

/*
2529 2530 2531
 * 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.
2532 2533
 */
static void
2534
__rcu_process_callbacks(struct rcu_state *rsp)
2535 2536
{
	unsigned long flags;
2537
	bool needwake;
2538
	struct rcu_data *rdp = raw_cpu_ptr(rsp->rda);
2539

2540 2541
	WARN_ON_ONCE(rdp->beenonline == 0);

2542 2543 2544 2545
	/* 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? */
2546
	local_irq_save(flags);
2547
	if (cpu_needs_another_gp(rsp, rdp)) {
2548
		raw_spin_lock(&rcu_get_root(rsp)->lock); /* irqs disabled. */
2549
		needwake = rcu_start_gp(rsp);
2550
		raw_spin_unlock_irqrestore(&rcu_get_root(rsp)->lock, flags);
2551 2552
		if (needwake)
			rcu_gp_kthread_wake(rsp);
2553 2554
	} else {
		local_irq_restore(flags);
2555 2556 2557
	}

	/* If there are callbacks ready, invoke them. */
2558
	if (cpu_has_callbacks_ready_to_invoke(rdp))
2559
		invoke_rcu_callbacks(rsp, rdp);
2560 2561 2562

	/* Do any needed deferred wakeups of rcuo kthreads. */
	do_nocb_deferred_wakeup(rdp);
2563 2564
}

2565
/*
2566
 * Do RCU core processing for the current CPU.
2567
 */
2568
static void rcu_process_callbacks(struct softirq_action *unused)
2569
{
2570 2571
	struct rcu_state *rsp;

2572 2573
	if (cpu_is_offline(smp_processor_id()))
		return;
2574
	trace_rcu_utilization(TPS("Start RCU core"));
2575 2576
	for_each_rcu_flavor(rsp)
		__rcu_process_callbacks(rsp);
2577
	trace_rcu_utilization(TPS("End RCU core"));
2578 2579
}

2580
/*
2581 2582 2583 2584 2585
 * 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
 * are running on the current CPU with interrupts disabled, the
 * rcu_cpu_kthread_task cannot disappear out from under us.
2586
 */
2587
static void invoke_rcu_callbacks(struct rcu_state *rsp, struct rcu_data *rdp)
2588
{
2589 2590
	if (unlikely(!ACCESS_ONCE(rcu_scheduler_fully_active)))
		return;
2591 2592
	if (likely(!rsp->boost)) {
		rcu_do_batch(rsp, rdp);
2593 2594
		return;
	}
2595
	invoke_rcu_callbacks_kthread();
2596 2597
}

2598
static void invoke_rcu_core(void)
2599
{
2600 2601
	if (cpu_online(smp_processor_id()))
		raise_softirq(RCU_SOFTIRQ);
2602 2603
}

2604 2605 2606 2607 2608
/*
 * 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)
2609
{
2610 2611
	bool needwake;

2612 2613 2614 2615
	/*
	 * If called from an extended quiescent state, invoke the RCU
	 * core in order to force a re-evaluation of RCU's idleness.
	 */
2616
	if (!rcu_is_watching() && cpu_online(smp_processor_id()))
2617 2618
		invoke_rcu_core();

2619
	/* If interrupts were disabled or CPU offline, don't invoke RCU core. */
2620
	if (irqs_disabled_flags(flags) || cpu_is_offline(smp_processor_id()))
2621
		return;
2622

2623 2624 2625 2626 2627 2628 2629
	/*
	 * 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.
	 */
2630
	if (unlikely(rdp->qlen > rdp->qlen_last_fqs_check + qhimark)) {
2631 2632

		/* Are we ignoring a completed grace period? */
2633
		note_gp_changes(rsp, rdp);
2634 2635 2636 2637 2638

		/* 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);

2639
			raw_spin_lock(&rnp_root->lock);
2640
			smp_mb__after_unlock_lock();
2641
			needwake = rcu_start_gp(rsp);
2642
			raw_spin_unlock(&rnp_root->lock);
2643 2644
			if (needwake)
				rcu_gp_kthread_wake(rsp);
2645 2646 2647 2648 2649
		} 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)
2650
				force_quiescent_state(rsp);
2651 2652 2653
			rdp->n_force_qs_snap = rsp->n_force_qs;
			rdp->qlen_last_fqs_check = rdp->qlen;
		}
2654
	}
2655 2656
}

2657 2658 2659 2660 2661 2662 2663
/*
 * RCU callback function to leak a callback.
 */
static void rcu_leak_callback(struct rcu_head *rhp)
{
}

P
Paul E. McKenney 已提交
2664 2665 2666 2667 2668 2669
/*
 * 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.
 */
2670 2671
static void
__call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu),
P
Paul E. McKenney 已提交
2672
	   struct rcu_state *rsp, int cpu, bool lazy)
2673 2674 2675 2676
{
	unsigned long flags;
	struct rcu_data *rdp;

2677
	WARN_ON_ONCE((unsigned long)head & 0x1); /* Misaligned rcu_head! */
2678 2679 2680 2681 2682 2683
	if (debug_rcu_head_queue(head)) {
		/* Probable double call_rcu(), so leak the callback. */
		ACCESS_ONCE(head->func) = rcu_leak_callback;
		WARN_ONCE(1, "__call_rcu(): Leaked duplicate callback\n");
		return;
	}
2684 2685 2686 2687 2688 2689 2690 2691 2692 2693
	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);
2694
	rdp = this_cpu_ptr(rsp->rda);
2695 2696

	/* Add the callback to our list. */
P
Paul E. McKenney 已提交
2697 2698 2699 2700 2701
	if (unlikely(rdp->nxttail[RCU_NEXT_TAIL] == NULL) || cpu != -1) {
		int offline;

		if (cpu != -1)
			rdp = per_cpu_ptr(rsp->rda, cpu);
2702
		offline = !__call_rcu_nocb(rdp, head, lazy, flags);
P
Paul E. McKenney 已提交
2703
		WARN_ON_ONCE(offline);
2704 2705 2706 2707
		/* _call_rcu() is illegal on offline CPU; leak the callback. */
		local_irq_restore(flags);
		return;
	}
2708
	ACCESS_ONCE(rdp->qlen) = rdp->qlen + 1;
2709 2710
	if (lazy)
		rdp->qlen_lazy++;
2711 2712
	else
		rcu_idle_count_callbacks_posted();
2713 2714 2715
	smp_mb();  /* Count before adding callback for rcu_barrier(). */
	*rdp->nxttail[RCU_NEXT_TAIL] = head;
	rdp->nxttail[RCU_NEXT_TAIL] = &head->next;
2716

2717 2718
	if (__is_kfree_rcu_offset((unsigned long)func))
		trace_rcu_kfree_callback(rsp->name, head, (unsigned long)func,
2719
					 rdp->qlen_lazy, rdp->qlen);
2720
	else
2721
		trace_rcu_callback(rsp->name, head, rdp->qlen_lazy, rdp->qlen);
2722

2723 2724
	/* Go handle any RCU core processing required. */
	__call_rcu_core(rsp, rdp, head, flags);
2725 2726 2727 2728
	local_irq_restore(flags);
}

/*
2729
 * Queue an RCU-sched callback for invocation after a grace period.
2730
 */
2731
void call_rcu_sched(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
2732
{
P
Paul E. McKenney 已提交
2733
	__call_rcu(head, func, &rcu_sched_state, -1, 0);
2734
}
2735
EXPORT_SYMBOL_GPL(call_rcu_sched);
2736 2737

/*
2738
 * Queue an RCU callback for invocation after a quicker grace period.
2739 2740 2741
 */
void call_rcu_bh(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
{
P
Paul E. McKenney 已提交
2742
	__call_rcu(head, func, &rcu_bh_state, -1, 0);
2743 2744 2745
}
EXPORT_SYMBOL_GPL(call_rcu_bh);

2746 2747 2748 2749 2750 2751 2752 2753 2754 2755
/*
 * 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))
{
2756
	__call_rcu(head, func, rcu_state_p, -1, 1);
2757 2758 2759
}
EXPORT_SYMBOL_GPL(kfree_call_rcu);

2760 2761 2762 2763 2764 2765 2766 2767 2768 2769 2770
/*
 * 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)
{
2771 2772
	int ret;

2773
	might_sleep();  /* Check for RCU read-side critical section. */
2774 2775 2776 2777
	preempt_disable();
	ret = num_online_cpus() <= 1;
	preempt_enable();
	return ret;
2778 2779
}

2780 2781 2782 2783 2784 2785 2786 2787 2788 2789 2790 2791
/**
 * 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
2792 2793 2794 2795 2796 2797 2798 2799 2800 2801 2802 2803 2804 2805 2806 2807 2808 2809 2810 2811 2812 2813
 * 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).
2814 2815 2816 2817 2818 2819 2820 2821 2822
 *
 * 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)
{
2823 2824 2825 2826
	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");
2827 2828
	if (rcu_blocking_is_gp())
		return;
2829 2830 2831 2832
	if (rcu_expedited)
		synchronize_sched_expedited();
	else
		wait_rcu_gp(call_rcu_sched);
2833 2834 2835 2836 2837 2838 2839 2840 2841 2842 2843
}
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.
2844 2845 2846
 *
 * See the description of synchronize_sched() for more detailed information
 * on memory ordering guarantees.
2847 2848 2849
 */
void synchronize_rcu_bh(void)
{
2850 2851 2852 2853
	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");
2854 2855
	if (rcu_blocking_is_gp())
		return;
2856 2857 2858 2859
	if (rcu_expedited)
		synchronize_rcu_bh_expedited();
	else
		wait_rcu_gp(call_rcu_bh);
2860 2861 2862
}
EXPORT_SYMBOL_GPL(synchronize_rcu_bh);

2863 2864 2865 2866 2867 2868 2869 2870 2871 2872 2873 2874 2875 2876 2877 2878 2879 2880 2881 2882
/**
 * 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().
	 */
2883
	return smp_load_acquire(&rcu_state_p->gpnum);
2884 2885 2886 2887 2888 2889 2890 2891 2892 2893 2894 2895 2896 2897 2898 2899 2900 2901 2902 2903 2904 2905 2906 2907 2908
}
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.
	 */
2909
	newstate = smp_load_acquire(&rcu_state_p->completed);
2910 2911 2912 2913 2914
	if (ULONG_CMP_GE(oldstate, newstate))
		synchronize_rcu();
}
EXPORT_SYMBOL_GPL(cond_synchronize_rcu);

2915 2916 2917 2918 2919 2920 2921 2922 2923 2924 2925 2926 2927 2928 2929 2930 2931
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;
}

2932 2933 2934 2935 2936 2937 2938 2939 2940 2941
/**
 * 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.
2942
 *
2943 2944 2945 2946
 * Note that it is illegal to call this function while holding any lock
 * that is acquired by a CPU-hotplug notifier.  And yes, it is also illegal
 * to call this function from a CPU-hotplug notifier.  Failing to observe
 * these restriction will result in deadlock.
2947 2948 2949 2950 2951 2952 2953 2954 2955 2956 2957 2958 2959 2960 2961 2962 2963 2964 2965 2966 2967 2968 2969 2970
 *
 * 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)
{
2971 2972
	long firstsnap, s, snap;
	int trycount = 0;
2973
	struct rcu_state *rsp = &rcu_sched_state;
2974

2975 2976 2977 2978 2979 2980 2981 2982
	/*
	 * 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.
	 */
2983 2984
	if (ULONG_CMP_GE((ulong)atomic_long_read(&rsp->expedited_start),
			 (ulong)atomic_long_read(&rsp->expedited_done) +
2985 2986
			 ULONG_MAX / 8)) {
		synchronize_sched();
2987
		atomic_long_inc(&rsp->expedited_wrap);
2988 2989
		return;
	}
2990

2991 2992 2993 2994
	/*
	 * Take a ticket.  Note that atomic_inc_return() implies a
	 * full memory barrier.
	 */
2995
	snap = atomic_long_inc_return(&rsp->expedited_start);
2996
	firstsnap = snap;
2997
	get_online_cpus();
2998
	WARN_ON_ONCE(cpu_is_offline(raw_smp_processor_id()));
2999 3000 3001 3002 3003 3004 3005 3006 3007

	/*
	 * Each pass through the following loop attempts to force a
	 * context switch on each CPU.
	 */
	while (try_stop_cpus(cpu_online_mask,
			     synchronize_sched_expedited_cpu_stop,
			     NULL) == -EAGAIN) {
		put_online_cpus();
3008
		atomic_long_inc(&rsp->expedited_tryfail);
3009

3010
		/* Check to see if someone else did our work for us. */
3011
		s = atomic_long_read(&rsp->expedited_done);
3012
		if (ULONG_CMP_GE((ulong)s, (ulong)firstsnap)) {
3013
			/* ensure test happens before caller kfree */
3014
			smp_mb__before_atomic(); /* ^^^ */
3015
			atomic_long_inc(&rsp->expedited_workdone1);
3016 3017
			return;
		}
3018 3019

		/* No joy, try again later.  Or just synchronize_sched(). */
3020
		if (trycount++ < 10) {
3021
			udelay(trycount * num_online_cpus());
3022
		} else {
3023
			wait_rcu_gp(call_rcu_sched);
3024
			atomic_long_inc(&rsp->expedited_normal);
3025 3026 3027
			return;
		}

3028
		/* Recheck to see if someone else did our work for us. */
3029
		s = atomic_long_read(&rsp->expedited_done);
3030
		if (ULONG_CMP_GE((ulong)s, (ulong)firstsnap)) {
3031
			/* ensure test happens before caller kfree */
3032
			smp_mb__before_atomic(); /* ^^^ */
3033
			atomic_long_inc(&rsp->expedited_workdone2);
3034 3035 3036 3037 3038
			return;
		}

		/*
		 * Refetching sync_sched_expedited_started allows later
3039 3040 3041 3042
		 * 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.
3043 3044
		 */
		get_online_cpus();
3045
		snap = atomic_long_read(&rsp->expedited_start);
3046 3047
		smp_mb(); /* ensure read is before try_stop_cpus(). */
	}
3048
	atomic_long_inc(&rsp->expedited_stoppedcpus);
3049 3050 3051 3052 3053

	/*
	 * 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
3054
	 * than we did already did their update.
3055 3056
	 */
	do {
3057
		atomic_long_inc(&rsp->expedited_done_tries);
3058
		s = atomic_long_read(&rsp->expedited_done);
3059
		if (ULONG_CMP_GE((ulong)s, (ulong)snap)) {
3060
			/* ensure test happens before caller kfree */
3061
			smp_mb__before_atomic(); /* ^^^ */
3062
			atomic_long_inc(&rsp->expedited_done_lost);
3063 3064
			break;
		}
3065
	} while (atomic_long_cmpxchg(&rsp->expedited_done, s, snap) != s);
3066
	atomic_long_inc(&rsp->expedited_done_exit);
3067 3068 3069 3070 3071

	put_online_cpus();
}
EXPORT_SYMBOL_GPL(synchronize_sched_expedited);

3072 3073 3074 3075 3076 3077 3078 3079 3080
/*
 * 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)
{
3081 3082
	struct rcu_node *rnp = rdp->mynode;

3083 3084 3085 3086 3087
	rdp->n_rcu_pending++;

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

3088 3089 3090 3091
	/* Is this CPU a NO_HZ_FULL CPU that should ignore RCU? */
	if (rcu_nohz_full_cpu(rsp))
		return 0;

3092
	/* Is the RCU core waiting for a quiescent state from this CPU? */
3093 3094
	if (rcu_scheduler_fully_active &&
	    rdp->qs_pending && !rdp->passed_quiesce) {
3095
		rdp->n_rp_qs_pending++;
3096
	} else if (rdp->qs_pending && rdp->passed_quiesce) {
3097
		rdp->n_rp_report_qs++;
3098
		return 1;
3099
	}
3100 3101

	/* Does this CPU have callbacks ready to invoke? */
3102 3103
	if (cpu_has_callbacks_ready_to_invoke(rdp)) {
		rdp->n_rp_cb_ready++;
3104
		return 1;
3105
	}
3106 3107

	/* Has RCU gone idle with this CPU needing another grace period? */
3108 3109
	if (cpu_needs_another_gp(rsp, rdp)) {
		rdp->n_rp_cpu_needs_gp++;
3110
		return 1;
3111
	}
3112 3113

	/* Has another RCU grace period completed?  */
3114
	if (ACCESS_ONCE(rnp->completed) != rdp->completed) { /* outside lock */
3115
		rdp->n_rp_gp_completed++;
3116
		return 1;
3117
	}
3118 3119

	/* Has a new RCU grace period started? */
3120
	if (ACCESS_ONCE(rnp->gpnum) != rdp->gpnum) { /* outside lock */
3121
		rdp->n_rp_gp_started++;
3122
		return 1;
3123
	}
3124

3125 3126 3127 3128 3129 3130
	/* Does this CPU need a deferred NOCB wakeup? */
	if (rcu_nocb_need_deferred_wakeup(rdp)) {
		rdp->n_rp_nocb_defer_wakeup++;
		return 1;
	}

3131
	/* nothing to do */
3132
	rdp->n_rp_need_nothing++;
3133 3134 3135 3136 3137 3138 3139 3140
	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.
 */
3141
static int rcu_pending(int cpu)
3142
{
3143 3144 3145 3146 3147 3148
	struct rcu_state *rsp;

	for_each_rcu_flavor(rsp)
		if (__rcu_pending(rsp, per_cpu_ptr(rsp->rda, cpu)))
			return 1;
	return 0;
3149 3150 3151
}

/*
3152 3153 3154
 * 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.)
3155
 */
3156
static int __maybe_unused rcu_cpu_has_callbacks(int cpu, bool *all_lazy)
3157
{
3158 3159 3160
	bool al = true;
	bool hc = false;
	struct rcu_data *rdp;
3161 3162
	struct rcu_state *rsp;

3163 3164
	for_each_rcu_flavor(rsp) {
		rdp = per_cpu_ptr(rsp->rda, cpu);
3165 3166 3167 3168
		if (!rdp->nxtlist)
			continue;
		hc = true;
		if (rdp->qlen != rdp->qlen_lazy || !all_lazy) {
3169
			al = false;
3170 3171
			break;
		}
3172 3173 3174 3175
	}
	if (all_lazy)
		*all_lazy = al;
	return hc;
3176 3177
}

3178 3179 3180 3181
/*
 * Helper function for _rcu_barrier() tracing.  If tracing is disabled,
 * the compiler is expected to optimize this away.
 */
3182
static void _rcu_barrier_trace(struct rcu_state *rsp, const char *s,
3183 3184 3185 3186 3187 3188
			       int cpu, unsigned long done)
{
	trace_rcu_barrier(rsp->name, s, cpu,
			  atomic_read(&rsp->barrier_cpu_count), done);
}

3189 3190 3191 3192
/*
 * RCU callback function for _rcu_barrier().  If we are last, wake
 * up the task executing _rcu_barrier().
 */
3193
static void rcu_barrier_callback(struct rcu_head *rhp)
3194
{
3195 3196 3197
	struct rcu_data *rdp = container_of(rhp, struct rcu_data, barrier_head);
	struct rcu_state *rsp = rdp->rsp;

3198 3199
	if (atomic_dec_and_test(&rsp->barrier_cpu_count)) {
		_rcu_barrier_trace(rsp, "LastCB", -1, rsp->n_barrier_done);
3200
		complete(&rsp->barrier_completion);
3201 3202 3203
	} else {
		_rcu_barrier_trace(rsp, "CB", -1, rsp->n_barrier_done);
	}
3204 3205 3206 3207 3208 3209 3210
}

/*
 * Called with preemption disabled, and from cross-cpu IRQ context.
 */
static void rcu_barrier_func(void *type)
{
3211
	struct rcu_state *rsp = type;
3212
	struct rcu_data *rdp = raw_cpu_ptr(rsp->rda);
3213

3214
	_rcu_barrier_trace(rsp, "IRQ", -1, rsp->n_barrier_done);
3215
	atomic_inc(&rsp->barrier_cpu_count);
3216
	rsp->call(&rdp->barrier_head, rcu_barrier_callback);
3217 3218 3219 3220 3221 3222
}

/*
 * Orchestrate the specified type of RCU barrier, waiting for all
 * RCU callbacks of the specified type to complete.
 */
3223
static void _rcu_barrier(struct rcu_state *rsp)
3224
{
3225 3226
	int cpu;
	struct rcu_data *rdp;
3227 3228
	unsigned long snap = ACCESS_ONCE(rsp->n_barrier_done);
	unsigned long snap_done;
3229

3230
	_rcu_barrier_trace(rsp, "Begin", -1, snap);
3231

3232
	/* Take mutex to serialize concurrent rcu_barrier() requests. */
3233
	mutex_lock(&rsp->barrier_mutex);
3234

3235 3236 3237 3238 3239 3240 3241 3242 3243 3244 3245 3246
	/*
	 * 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.
	 */
3247
	snap_done = rsp->n_barrier_done;
3248
	_rcu_barrier_trace(rsp, "Check", -1, snap_done);
3249 3250 3251 3252 3253 3254 3255 3256 3257 3258 3259 3260

	/*
	 * 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)) {
3261
		_rcu_barrier_trace(rsp, "EarlyExit", -1, snap_done);
3262 3263 3264 3265 3266 3267 3268 3269 3270 3271
		smp_mb(); /* caller's subsequent code after above check. */
		mutex_unlock(&rsp->barrier_mutex);
		return;
	}

	/*
	 * Increment ->n_barrier_done to avoid duplicate work.  Use
	 * ACCESS_ONCE() to prevent the compiler from speculating
	 * the increment to precede the early-exit check.
	 */
3272
	ACCESS_ONCE(rsp->n_barrier_done) = rsp->n_barrier_done + 1;
3273
	WARN_ON_ONCE((rsp->n_barrier_done & 0x1) != 1);
3274
	_rcu_barrier_trace(rsp, "Inc1", -1, rsp->n_barrier_done);
3275
	smp_mb(); /* Order ->n_barrier_done increment with below mechanism. */
3276

3277
	/*
3278 3279
	 * 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
3280 3281
	 * (or preemption of this task).  Exclude CPU-hotplug operations
	 * to ensure that no offline CPU has callbacks queued.
3282
	 */
3283
	init_completion(&rsp->barrier_completion);
3284
	atomic_set(&rsp->barrier_cpu_count, 1);
3285
	get_online_cpus();
3286 3287

	/*
3288 3289 3290
	 * 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.
3291
	 */
P
Paul E. McKenney 已提交
3292
	for_each_possible_cpu(cpu) {
3293
		if (!cpu_online(cpu) && !rcu_is_nocb_cpu(cpu))
P
Paul E. McKenney 已提交
3294
			continue;
3295
		rdp = per_cpu_ptr(rsp->rda, cpu);
3296
		if (rcu_is_nocb_cpu(cpu)) {
P
Paul E. McKenney 已提交
3297 3298 3299 3300 3301 3302
			_rcu_barrier_trace(rsp, "OnlineNoCB", cpu,
					   rsp->n_barrier_done);
			atomic_inc(&rsp->barrier_cpu_count);
			__call_rcu(&rdp->barrier_head, rcu_barrier_callback,
				   rsp, cpu, 0);
		} else if (ACCESS_ONCE(rdp->qlen)) {
3303 3304
			_rcu_barrier_trace(rsp, "OnlineQ", cpu,
					   rsp->n_barrier_done);
3305
			smp_call_function_single(cpu, rcu_barrier_func, rsp, 1);
3306
		} else {
3307 3308
			_rcu_barrier_trace(rsp, "OnlineNQ", cpu,
					   rsp->n_barrier_done);
3309 3310
		}
	}
3311
	put_online_cpus();
3312 3313 3314 3315 3316

	/*
	 * Now that we have an rcu_barrier_callback() callback on each
	 * CPU, and thus each counted, remove the initial count.
	 */
3317
	if (atomic_dec_and_test(&rsp->barrier_cpu_count))
3318
		complete(&rsp->barrier_completion);
3319

3320 3321
	/* Increment ->n_barrier_done to prevent duplicate work. */
	smp_mb(); /* Keep increment after above mechanism. */
3322
	ACCESS_ONCE(rsp->n_barrier_done) = rsp->n_barrier_done + 1;
3323
	WARN_ON_ONCE((rsp->n_barrier_done & 0x1) != 0);
3324
	_rcu_barrier_trace(rsp, "Inc2", -1, rsp->n_barrier_done);
3325 3326
	smp_mb(); /* Keep increment before caller's subsequent code. */

3327
	/* Wait for all rcu_barrier_callback() callbacks to be invoked. */
3328
	wait_for_completion(&rsp->barrier_completion);
3329 3330

	/* Other rcu_barrier() invocations can now safely proceed. */
3331
	mutex_unlock(&rsp->barrier_mutex);
3332 3333 3334 3335 3336 3337 3338
}

/**
 * rcu_barrier_bh - Wait until all in-flight call_rcu_bh() callbacks complete.
 */
void rcu_barrier_bh(void)
{
3339
	_rcu_barrier(&rcu_bh_state);
3340 3341 3342 3343 3344 3345 3346 3347
}
EXPORT_SYMBOL_GPL(rcu_barrier_bh);

/**
 * rcu_barrier_sched - Wait for in-flight call_rcu_sched() callbacks.
 */
void rcu_barrier_sched(void)
{
3348
	_rcu_barrier(&rcu_sched_state);
3349 3350 3351
}
EXPORT_SYMBOL_GPL(rcu_barrier_sched);

3352
/*
3353
 * Do boot-time initialization of a CPU's per-CPU RCU data.
3354
 */
3355 3356
static void __init
rcu_boot_init_percpu_data(int cpu, struct rcu_state *rsp)
3357 3358
{
	unsigned long flags;
3359
	struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
3360 3361 3362
	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
3363
	raw_spin_lock_irqsave(&rnp->lock, flags);
3364
	rdp->grpmask = 1UL << (cpu - rdp->mynode->grplo);
3365
	init_callback_list(rdp);
3366
	rdp->qlen_lazy = 0;
3367
	ACCESS_ONCE(rdp->qlen) = 0;
3368
	rdp->dynticks = &per_cpu(rcu_dynticks, cpu);
3369
	WARN_ON_ONCE(rdp->dynticks->dynticks_nesting != DYNTICK_TASK_EXIT_IDLE);
3370
	WARN_ON_ONCE(atomic_read(&rdp->dynticks->dynticks) != 1);
3371
	rdp->cpu = cpu;
3372
	rdp->rsp = rsp;
P
Paul E. McKenney 已提交
3373
	rcu_boot_init_nocb_percpu_data(rdp);
P
Paul E. McKenney 已提交
3374
	raw_spin_unlock_irqrestore(&rnp->lock, flags);
3375 3376 3377 3378 3379 3380 3381
}

/*
 * 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.
3382
 */
3383
static void
3384
rcu_init_percpu_data(int cpu, struct rcu_state *rsp)
3385 3386 3387
{
	unsigned long flags;
	unsigned long mask;
3388
	struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
3389 3390
	struct rcu_node *rnp = rcu_get_root(rsp);

3391 3392 3393
	/* Exclude new grace periods. */
	mutex_lock(&rsp->onoff_mutex);

3394
	/* Set up local state, ensuring consistent view of global state. */
P
Paul E. McKenney 已提交
3395
	raw_spin_lock_irqsave(&rnp->lock, flags);
3396
	rdp->beenonline = 1;	 /* We have now been online. */
3397 3398
	rdp->qlen_last_fqs_check = 0;
	rdp->n_force_qs_snap = rsp->n_force_qs;
3399
	rdp->blimit = blimit;
3400
	init_callback_list(rdp);  /* Re-enable callbacks on this CPU. */
3401
	rdp->dynticks->dynticks_nesting = DYNTICK_TASK_EXIT_IDLE;
3402
	rcu_sysidle_init_percpu_data(rdp->dynticks);
3403 3404
	atomic_set(&rdp->dynticks->dynticks,
		   (atomic_read(&rdp->dynticks->dynticks) & ~0x1) + 1);
P
Paul E. McKenney 已提交
3405
	raw_spin_unlock(&rnp->lock);		/* irqs remain disabled. */
3406 3407 3408 3409 3410 3411

	/* Add CPU to rcu_node bitmasks. */
	rnp = rdp->mynode;
	mask = rdp->grpmask;
	do {
		/* Exclude any attempts to start a new GP on small systems. */
P
Paul E. McKenney 已提交
3412
		raw_spin_lock(&rnp->lock);	/* irqs already disabled. */
3413 3414
		rnp->qsmaskinit |= mask;
		mask = rnp->grpmask;
3415
		if (rnp == rdp->mynode) {
3416 3417 3418 3419 3420 3421
			/*
			 * If there is a grace period in progress, we will
			 * set up to wait for it next time we run the
			 * RCU core code.
			 */
			rdp->gpnum = rnp->completed;
3422
			rdp->completed = rnp->completed;
3423 3424
			rdp->passed_quiesce = 0;
			rdp->qs_pending = 0;
3425
			trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("cpuonl"));
3426
		}
P
Paul E. McKenney 已提交
3427
		raw_spin_unlock(&rnp->lock); /* irqs already disabled. */
3428 3429
		rnp = rnp->parent;
	} while (rnp != NULL && !(rnp->qsmaskinit & mask));
3430
	local_irq_restore(flags);
3431

3432
	mutex_unlock(&rsp->onoff_mutex);
3433 3434
}

3435
static void rcu_prepare_cpu(int cpu)
3436
{
3437 3438 3439
	struct rcu_state *rsp;

	for_each_rcu_flavor(rsp)
3440
		rcu_init_percpu_data(cpu, rsp);
3441 3442 3443
}

/*
3444
 * Handle CPU online/offline notification events.
3445
 */
3446
static int rcu_cpu_notify(struct notifier_block *self,
3447
				    unsigned long action, void *hcpu)
3448 3449
{
	long cpu = (long)hcpu;
3450
	struct rcu_data *rdp = per_cpu_ptr(rcu_state_p->rda, cpu);
3451
	struct rcu_node *rnp = rdp->mynode;
3452
	struct rcu_state *rsp;
3453

3454
	trace_rcu_utilization(TPS("Start CPU hotplug"));
3455 3456 3457
	switch (action) {
	case CPU_UP_PREPARE:
	case CPU_UP_PREPARE_FROZEN:
P
Peter Zijlstra 已提交
3458 3459
		rcu_prepare_cpu(cpu);
		rcu_prepare_kthreads(cpu);
3460
		rcu_spawn_all_nocb_kthreads(cpu);
3461 3462
		break;
	case CPU_ONLINE:
3463
	case CPU_DOWN_FAILED:
T
Thomas Gleixner 已提交
3464
		rcu_boost_kthread_setaffinity(rnp, -1);
3465 3466
		break;
	case CPU_DOWN_PREPARE:
3467
		rcu_boost_kthread_setaffinity(rnp, cpu);
3468
		break;
3469 3470
	case CPU_DYING:
	case CPU_DYING_FROZEN:
3471 3472
		for_each_rcu_flavor(rsp)
			rcu_cleanup_dying_cpu(rsp);
3473
		break;
3474 3475 3476 3477
	case CPU_DEAD:
	case CPU_DEAD_FROZEN:
	case CPU_UP_CANCELED:
	case CPU_UP_CANCELED_FROZEN:
3478 3479
		for_each_rcu_flavor(rsp)
			rcu_cleanup_dead_cpu(cpu, rsp);
3480 3481 3482 3483
		break;
	default:
		break;
	}
3484
	trace_rcu_utilization(TPS("End CPU hotplug"));
3485
	return NOTIFY_OK;
3486 3487
}

3488 3489 3490 3491 3492 3493 3494 3495 3496 3497 3498 3499 3500 3501 3502 3503 3504 3505 3506
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. */
			rcu_expedited = 1;
		break;
	case PM_POST_HIBERNATION:
	case PM_POST_SUSPEND:
		rcu_expedited = 0;
		break;
	default:
		break;
	}
	return NOTIFY_OK;
}

3507
/*
3508
 * Spawn the kthreads that handle each RCU flavor's grace periods.
3509 3510 3511 3512 3513 3514 3515 3516
 */
static int __init rcu_spawn_gp_kthread(void)
{
	unsigned long flags;
	struct rcu_node *rnp;
	struct rcu_state *rsp;
	struct task_struct *t;

3517
	rcu_scheduler_fully_active = 1;
3518
	for_each_rcu_flavor(rsp) {
3519
		t = kthread_run(rcu_gp_kthread, rsp, "%s", rsp->name);
3520 3521 3522 3523 3524 3525
		BUG_ON(IS_ERR(t));
		rnp = rcu_get_root(rsp);
		raw_spin_lock_irqsave(&rnp->lock, flags);
		rsp->gp_kthread = t;
		raw_spin_unlock_irqrestore(&rnp->lock, flags);
	}
3526
	rcu_spawn_nocb_kthreads();
3527
	rcu_spawn_boost_kthreads();
3528 3529 3530 3531
	return 0;
}
early_initcall(rcu_spawn_gp_kthread);

3532 3533 3534 3535 3536 3537 3538 3539 3540 3541 3542 3543 3544 3545 3546
/*
 * 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;
}

3547 3548 3549 3550 3551 3552 3553 3554 3555
/*
 * Compute the per-level fanout, either using the exact fanout specified
 * or balancing the tree, depending on CONFIG_RCU_FANOUT_EXACT.
 */
#ifdef CONFIG_RCU_FANOUT_EXACT
static void __init rcu_init_levelspread(struct rcu_state *rsp)
{
	int i;

3556 3557
	rsp->levelspread[rcu_num_lvls - 1] = rcu_fanout_leaf;
	for (i = rcu_num_lvls - 2; i >= 0; i--)
3558 3559 3560 3561 3562 3563 3564 3565 3566
		rsp->levelspread[i] = CONFIG_RCU_FANOUT;
}
#else /* #ifdef CONFIG_RCU_FANOUT_EXACT */
static void __init rcu_init_levelspread(struct rcu_state *rsp)
{
	int ccur;
	int cprv;
	int i;

3567
	cprv = nr_cpu_ids;
3568
	for (i = rcu_num_lvls - 1; i >= 0; i--) {
3569 3570 3571 3572 3573 3574 3575 3576 3577 3578
		ccur = rsp->levelcnt[i];
		rsp->levelspread[i] = (cprv + ccur - 1) / ccur;
		cprv = ccur;
	}
}
#endif /* #else #ifdef CONFIG_RCU_FANOUT_EXACT */

/*
 * Helper function for rcu_init() that initializes one rcu_state structure.
 */
3579 3580
static void __init rcu_init_one(struct rcu_state *rsp,
		struct rcu_data __percpu *rda)
3581
{
3582 3583 3584 3585 3586 3587 3588 3589 3590 3591
	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 */
3592
	static u8 fl_mask = 0x1;
3593 3594 3595 3596 3597
	int cpustride = 1;
	int i;
	int j;
	struct rcu_node *rnp;

3598 3599
	BUILD_BUG_ON(MAX_RCU_LVLS > ARRAY_SIZE(buf));  /* Fix buf[] init! */

3600 3601 3602 3603
	/* Silence gcc 4.8 warning about array index out of range. */
	if (rcu_num_lvls > RCU_NUM_LVLS)
		panic("rcu_init_one: rcu_num_lvls overflow");

3604 3605
	/* Initialize the level-tracking arrays. */

3606 3607 3608
	for (i = 0; i < rcu_num_lvls; i++)
		rsp->levelcnt[i] = num_rcu_lvl[i];
	for (i = 1; i < rcu_num_lvls; i++)
3609 3610
		rsp->level[i] = rsp->level[i - 1] + rsp->levelcnt[i - 1];
	rcu_init_levelspread(rsp);
3611 3612
	rsp->flavor_mask = fl_mask;
	fl_mask <<= 1;
3613 3614 3615

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

3616
	for (i = rcu_num_lvls - 1; i >= 0; i--) {
3617 3618 3619
		cpustride *= rsp->levelspread[i];
		rnp = rsp->level[i];
		for (j = 0; j < rsp->levelcnt[i]; j++, rnp++) {
P
Paul E. McKenney 已提交
3620
			raw_spin_lock_init(&rnp->lock);
3621 3622
			lockdep_set_class_and_name(&rnp->lock,
						   &rcu_node_class[i], buf[i]);
3623 3624 3625
			raw_spin_lock_init(&rnp->fqslock);
			lockdep_set_class_and_name(&rnp->fqslock,
						   &rcu_fqs_class[i], fqs[i]);
3626 3627
			rnp->gpnum = rsp->gpnum;
			rnp->completed = rsp->completed;
3628 3629 3630 3631
			rnp->qsmask = 0;
			rnp->qsmaskinit = 0;
			rnp->grplo = j * cpustride;
			rnp->grphi = (j + 1) * cpustride - 1;
3632 3633
			if (rnp->grphi >= nr_cpu_ids)
				rnp->grphi = nr_cpu_ids - 1;
3634 3635 3636 3637 3638 3639 3640 3641 3642 3643 3644
			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;
3645
			INIT_LIST_HEAD(&rnp->blkd_tasks);
3646
			rcu_init_one_nocb(rnp);
3647 3648
		}
	}
3649

3650
	rsp->rda = rda;
3651
	init_waitqueue_head(&rsp->gp_wq);
3652
	rnp = rsp->level[rcu_num_lvls - 1];
3653
	for_each_possible_cpu(i) {
3654
		while (i > rnp->grphi)
3655
			rnp++;
3656
		per_cpu_ptr(rsp->rda, i)->mynode = rnp;
3657 3658
		rcu_boot_init_percpu_data(i, rsp);
	}
3659
	list_add(&rsp->flavors, &rcu_struct_flavors);
3660 3661
}

3662 3663
/*
 * Compute the rcu_node tree geometry from kernel parameters.  This cannot
3664
 * replace the definitions in tree.h because those are needed to size
3665 3666 3667 3668
 * the ->node array in the rcu_state structure.
 */
static void __init rcu_init_geometry(void)
{
3669
	ulong d;
3670 3671
	int i;
	int j;
3672
	int n = nr_cpu_ids;
3673 3674
	int rcu_capacity[MAX_RCU_LVLS + 1];

3675 3676 3677 3678 3679 3680 3681 3682 3683 3684 3685 3686 3687
	/*
	 * 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;

3688
	/* If the compile-time values are accurate, just leave. */
3689 3690
	if (rcu_fanout_leaf == CONFIG_RCU_FANOUT_LEAF &&
	    nr_cpu_ids == NR_CPUS)
3691
		return;
3692 3693
	pr_info("RCU: Adjusting geometry for rcu_fanout_leaf=%d, nr_cpu_ids=%d\n",
		rcu_fanout_leaf, nr_cpu_ids);
3694 3695 3696 3697 3698 3699 3700 3701 3702 3703 3704 3705 3706 3707 3708 3709 3710 3711 3712 3713 3714 3715 3716 3717 3718 3719 3720 3721 3722 3723 3724 3725 3726 3727 3728 3729 3730 3731 3732 3733 3734 3735 3736 3737 3738

	/*
	 * 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++)
		rcu_capacity[i] = rcu_capacity[i - 1] * CONFIG_RCU_FANOUT;

	/*
	 * 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.
	 */
	if (rcu_fanout_leaf < CONFIG_RCU_FANOUT_LEAF ||
	    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;
}

3739
void __init rcu_init(void)
3740
{
P
Paul E. McKenney 已提交
3741
	int cpu;
3742

3743
	rcu_bootup_announce();
3744
	rcu_init_geometry();
3745
	rcu_init_one(&rcu_bh_state, &rcu_bh_data);
3746
	rcu_init_one(&rcu_sched_state, &rcu_sched_data);
3747
	__rcu_init_preempt();
J
Jiang Fang 已提交
3748
	open_softirq(RCU_SOFTIRQ, rcu_process_callbacks);
3749 3750 3751 3752 3753 3754 3755

	/*
	 * 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);
3756
	pm_notifier(rcu_pm_notify, 0);
P
Paul E. McKenney 已提交
3757 3758
	for_each_online_cpu(cpu)
		rcu_cpu_notify(NULL, CPU_UP_PREPARE, (void *)(long)cpu);
3759 3760
}

3761
#include "tree_plugin.h"