tree.c 128.1 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14
/*
 * Read-Copy Update mechanism for mutual exclusion
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2 of the License, or
 * (at your option) any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License
15 16
 * along with this program; if not, you can access it online at
 * http://www.gnu.org/licenses/gpl-2.0.html.
17 18 19 20 21 22 23 24 25 26 27
 *
 * Copyright IBM Corporation, 2008
 *
 * Authors: Dipankar Sarma <dipankar@in.ibm.com>
 *	    Manfred Spraul <manfred@colorfullife.com>
 *	    Paul E. McKenney <paulmck@linux.vnet.ibm.com> Hierarchical version
 *
 * Based on the original work by Paul McKenney <paulmck@us.ibm.com>
 * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen.
 *
 * For detailed explanation of Read-Copy Update mechanism see -
28
 *	Documentation/RCU
29 30 31 32 33 34 35 36 37
 */
#include <linux/types.h>
#include <linux/kernel.h>
#include <linux/init.h>
#include <linux/spinlock.h>
#include <linux/smp.h>
#include <linux/rcupdate.h>
#include <linux/interrupt.h>
#include <linux/sched.h>
38
#include <linux/nmi.h>
39
#include <linux/atomic.h>
40
#include <linux/bitops.h>
41
#include <linux/export.h>
42 43
#include <linux/completion.h>
#include <linux/moduleparam.h>
44
#include <linux/module.h>
45 46 47 48 49
#include <linux/percpu.h>
#include <linux/notifier.h>
#include <linux/cpu.h>
#include <linux/mutex.h>
#include <linux/time.h>
50
#include <linux/kernel_stat.h>
51 52
#include <linux/wait.h>
#include <linux/kthread.h>
53
#include <linux/prefetch.h>
54 55
#include <linux/delay.h>
#include <linux/stop_machine.h>
56
#include <linux/random.h>
57
#include <linux/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
DEFINE_PER_CPU_SHARED_ALIGNED(struct rcu_data, sname##_data); \
95
struct rcu_state sname##_state = { \
96
	.level = { &sname##_state.node[0] }, \
97
	.rda = &sname##_data, \
98
	.call = cr, \
99
	.fqs_state = RCU_GP_IDLE, \
P
Paul E. McKenney 已提交
100 101
	.gpnum = 0UL - 300UL, \
	.completed = 0UL - 300UL, \
102
	.orphan_lock = __RAW_SPIN_LOCK_UNLOCKED(&sname##_state.orphan_lock), \
103 104
	.orphan_nxttail = &sname##_state.orphan_nxtlist, \
	.orphan_donetail = &sname##_state.orphan_donelist, \
105
	.barrier_mutex = __MUTEX_INITIALIZER(sname##_state.barrier_mutex), \
106
	.name = RCU_STATE_NAME(sname), \
107
	.abbr = sabbr, \
108
}
109

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

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

117 118
/* Increase (but not decrease) the CONFIG_RCU_FANOUT_LEAF at boot time. */
static int rcu_fanout_leaf = CONFIG_RCU_FANOUT_LEAF;
119
module_param(rcu_fanout_leaf, int, 0444);
120 121 122 123 124 125 126 127 128 129
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. */

130 131 132 133
/*
 * 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)
134
 * optimize synchronize_sched() to a simple barrier().  When this variable
135 136 137 138
 * 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.
 */
139 140 141
int rcu_scheduler_active __read_mostly;
EXPORT_SYMBOL_GPL(rcu_scheduler_active);

142 143 144 145 146 147 148 149 150 151 152 153 154 155
/*
 * 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;

156 157
static void rcu_init_new_rnp(struct rcu_node *rnp_leaf);
static void rcu_cleanup_dead_rnp(struct rcu_node *rnp_leaf);
T
Thomas Gleixner 已提交
158
static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu);
159 160
static void invoke_rcu_core(void);
static void invoke_rcu_callbacks(struct rcu_state *rsp, struct rcu_data *rdp);
161

162 163 164 165
/* rcuc/rcub kthread realtime priority */
static int kthread_prio = CONFIG_RCU_KTHREAD_PRIO;
module_param(kthread_prio, int, 0644);

166 167 168
/* Delay in jiffies for grace-period initialization delays, debug only. */
#ifdef CONFIG_RCU_TORTURE_TEST_SLOW_INIT
static int gp_init_delay = CONFIG_RCU_TORTURE_TEST_SLOW_INIT_DELAY;
169
module_param(gp_init_delay, int, 0644);
170 171 172 173
#else /* #ifdef CONFIG_RCU_TORTURE_TEST_SLOW_INIT */
static const int gp_init_delay;
#endif /* #else #ifdef CONFIG_RCU_TORTURE_TEST_SLOW_INIT */
#define PER_RCU_NODE_PERIOD 10	/* Number of grace periods between delays. */
174

175 176 177 178 179 180 181 182 183 184 185 186
/*
 * 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;

187 188 189 190 191 192 193 194
/*
 * Compute the mask of online CPUs for the specified rcu_node structure.
 * This will not be stable unless the rcu_node structure's ->lock is
 * held, but the bit corresponding to the current CPU will be stable
 * in most contexts.
 */
unsigned long rcu_rnp_online_cpus(struct rcu_node *rnp)
{
195
	return READ_ONCE(rnp->qsmaskinitnext);
196 197
}

198
/*
199
 * Return true if an RCU grace period is in progress.  The READ_ONCE()s
200 201 202 203 204
 * 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)
{
205
	return READ_ONCE(rsp->completed) != READ_ONCE(rsp->gpnum);
206 207
}

208
/*
209
 * Note a quiescent state.  Because we do not need to know
210
 * how many quiescent states passed, just if there was at least
211
 * one since the start of the grace period, this just sets a flag.
212
 * The caller must have disabled preemption.
213
 */
214
void rcu_sched_qs(void)
215
{
216 217 218 219 220 221
	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);
	}
222 223
}

224
void rcu_bh_qs(void)
225
{
226 227 228 229 230 231
	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);
	}
232
}
233

234 235 236 237 238 239 240 241 242 243 244
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 */
};

245 246 247
DEFINE_PER_CPU_SHARED_ALIGNED(unsigned long, rcu_qs_ctr);
EXPORT_PER_CPU_SYMBOL_GPL(rcu_qs_ctr);

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
/*
 * 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. */
282 283
		if (READ_ONCE(rdp->mynode->completed) !=
		    READ_ONCE(rdp->cond_resched_completed))
284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300
			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);
}

301 302 303
/*
 * Note a context switch.  This is a quiescent state for RCU-sched,
 * and requires special handling for preemptible RCU.
304
 * The caller must have disabled preemption.
305
 */
306
void rcu_note_context_switch(void)
307
{
308
	trace_rcu_utilization(TPS("Start context switch"));
309
	rcu_sched_qs();
310
	rcu_preempt_note_context_switch();
311 312
	if (unlikely(raw_cpu_read(rcu_sched_qs_mask)))
		rcu_momentary_dyntick_idle();
313
	trace_rcu_utilization(TPS("End context switch"));
314
}
315
EXPORT_SYMBOL_GPL(rcu_note_context_switch);
316

317
/*
318
 * Register a quiescent state for all RCU flavors.  If there is an
319 320
 * emergency, invoke rcu_momentary_dyntick_idle() to do a heavy-weight
 * dyntick-idle quiescent state visible to other CPUs (but only for those
321
 * RCU flavors in desperate need of a quiescent state, which will normally
322 323 324 325 326 327 328 329 330 331 332
 * be none of them).  Either way, do a lightweight quiescent state for
 * all RCU flavors.
 */
void rcu_all_qs(void)
{
	if (unlikely(raw_cpu_read(rcu_sched_qs_mask)))
		rcu_momentary_dyntick_idle();
	this_cpu_inc(rcu_qs_ctr);
}
EXPORT_SYMBOL_GPL(rcu_all_qs);

E
Eric Dumazet 已提交
333 334 335
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. */
336

E
Eric Dumazet 已提交
337 338 339
module_param(blimit, long, 0444);
module_param(qhimark, long, 0444);
module_param(qlowmark, long, 0444);
340

341 342
static ulong jiffies_till_first_fqs = ULONG_MAX;
static ulong jiffies_till_next_fqs = ULONG_MAX;
343 344 345 346

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

347 348 349 350 351 352 353
/*
 * 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);

354
static bool rcu_start_gp_advanced(struct rcu_state *rsp, struct rcu_node *rnp,
355
				  struct rcu_data *rdp);
356 357 358 359
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);
360
static void force_quiescent_state(struct rcu_state *rsp);
361
static int rcu_pending(void);
362 363

/*
364
 * Return the number of RCU batches started thus far for debug & stats.
365
 */
366 367 368 369 370 371 372 373
unsigned long rcu_batches_started(void)
{
	return rcu_state_p->gpnum;
}
EXPORT_SYMBOL_GPL(rcu_batches_started);

/*
 * Return the number of RCU-sched batches started thus far for debug & stats.
374
 */
375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400
unsigned long rcu_batches_started_sched(void)
{
	return rcu_sched_state.gpnum;
}
EXPORT_SYMBOL_GPL(rcu_batches_started_sched);

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

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

/*
 * Return the number of RCU-sched batches completed thus far for debug & stats.
401
 */
402
unsigned long rcu_batches_completed_sched(void)
403
{
404
	return rcu_sched_state.completed;
405
}
406
EXPORT_SYMBOL_GPL(rcu_batches_completed_sched);
407 408

/*
409
 * Return the number of RCU BH batches completed thus far for debug & stats.
410
 */
411
unsigned long rcu_batches_completed_bh(void)
412 413 414 415 416
{
	return rcu_bh_state.completed;
}
EXPORT_SYMBOL_GPL(rcu_batches_completed_bh);

417 418 419 420 421
/*
 * Force a quiescent state.
 */
void rcu_force_quiescent_state(void)
{
422
	force_quiescent_state(rcu_state_p);
423 424 425
}
EXPORT_SYMBOL_GPL(rcu_force_quiescent_state);

426 427 428 429 430
/*
 * Force a quiescent state for RCU BH.
 */
void rcu_bh_force_quiescent_state(void)
{
431
	force_quiescent_state(&rcu_bh_state);
432 433 434
}
EXPORT_SYMBOL_GPL(rcu_bh_force_quiescent_state);

435 436 437 438 439 440 441 442 443
/*
 * Force a quiescent state for RCU-sched.
 */
void rcu_sched_force_quiescent_state(void)
{
	force_quiescent_state(&rcu_sched_state);
}
EXPORT_SYMBOL_GPL(rcu_sched_force_quiescent_state);

444 445 446 447 448 449 450 451 452 453 454 455 456 457 458
/*
 * 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);

459 460 461 462 463 464 465 466 467 468 469 470 471 472
/*
 * 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);

473 474 475 476 477 478 479 480 481 482
/*
 * 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:
483
		rsp = rcu_state_p;
484 485 486 487 488 489 490 491 492 493 494
		break;
	case RCU_BH_FLAVOR:
		rsp = &rcu_bh_state;
		break;
	case RCU_SCHED_FLAVOR:
		rsp = &rcu_sched_state;
		break;
	default:
		break;
	}
	if (rsp != NULL) {
495 496 497
		*flags = READ_ONCE(rsp->gp_flags);
		*gpnum = READ_ONCE(rsp->gpnum);
		*completed = READ_ONCE(rsp->completed);
498 499 500 501 502 503 504 505
		return;
	}
	*flags = 0;
	*gpnum = 0;
	*completed = 0;
}
EXPORT_SYMBOL_GPL(rcutorture_get_gp_data);

506 507 508 509 510 511 512 513 514 515 516
/*
 * 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);

517 518 519 520 521 522
/*
 * 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
523 524
	return &rdp->nxtlist != rdp->nxttail[RCU_DONE_TAIL] &&
	       rdp->nxttail[RCU_DONE_TAIL] != NULL;
525 526
}

527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542
/*
 * 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);
543
	int idx = (READ_ONCE(rnp->completed) + 1) & 0x1;
544 545
	int *fp = &rnp->need_future_gp[idx];

546
	return READ_ONCE(*fp);
547 548
}

549
/*
550 551 552
 * Does the current CPU require a not-yet-started grace period?
 * The caller must have disabled interrupts to prevent races with
 * normal callback registry.
553 554 555 556
 */
static int
cpu_needs_another_gp(struct rcu_state *rsp, struct rcu_data *rdp)
{
557
	int i;
P
Paul E. McKenney 已提交
558

559 560
	if (rcu_gp_in_progress(rsp))
		return 0;  /* No, a grace period is already in progress. */
561
	if (rcu_future_needs_gp(rsp))
562
		return 1;  /* Yes, a no-CBs CPU needs one. */
563 564 565 566 567 568
	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] &&
569
		    ULONG_CMP_LT(READ_ONCE(rsp->completed),
570 571 572
				 rdp->nxtcompleted[i]))
			return 1;  /* Yes, CBs for future grace period. */
	return 0; /* No grace period needed. */
573 574
}

575
/*
576
 * rcu_eqs_enter_common - current CPU is moving towards extended quiescent state
577 578 579 580 581
 *
 * 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.
 */
582
static void rcu_eqs_enter_common(long long oldval, bool user)
583
{
584 585
	struct rcu_state *rsp;
	struct rcu_data *rdp;
586
	struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks);
587

588
	trace_rcu_dyntick(TPS("Start"), oldval, rdtp->dynticks_nesting);
589
	if (!user && !is_idle_task(current)) {
590 591
		struct task_struct *idle __maybe_unused =
			idle_task(smp_processor_id());
592

593
		trace_rcu_dyntick(TPS("Error on entry: not idle task"), oldval, 0);
594
		ftrace_dump(DUMP_ORIG);
595 596 597
		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! */
598
	}
599 600 601 602
	for_each_rcu_flavor(rsp) {
		rdp = this_cpu_ptr(rsp->rda);
		do_nocb_deferred_wakeup(rdp);
	}
603
	rcu_prepare_for_idle();
604
	/* CPUs seeing atomic_inc() must see prior RCU read-side crit sects */
605
	smp_mb__before_atomic();  /* See above. */
606
	atomic_inc(&rdtp->dynticks);
607
	smp_mb__after_atomic();  /* Force ordering with next sojourn. */
608
	WARN_ON_ONCE(atomic_read(&rdtp->dynticks) & 0x1);
609
	rcu_dynticks_task_enter();
610 611

	/*
612
	 * It is illegal to enter an extended quiescent state while
613 614 615 616 617 618 619 620
	 * 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.");
621
}
622

623 624 625
/*
 * Enter an RCU extended quiescent state, which can be either the
 * idle loop or adaptive-tickless usermode execution.
626
 */
627
static void rcu_eqs_enter(bool user)
628
{
629
	long long oldval;
630 631
	struct rcu_dynticks *rdtp;

632
	rdtp = this_cpu_ptr(&rcu_dynticks);
633
	oldval = rdtp->dynticks_nesting;
634
	WARN_ON_ONCE((oldval & DYNTICK_TASK_NEST_MASK) == 0);
635
	if ((oldval & DYNTICK_TASK_NEST_MASK) == DYNTICK_TASK_NEST_VALUE) {
636
		rdtp->dynticks_nesting = 0;
637
		rcu_eqs_enter_common(oldval, user);
638
	} else {
639
		rdtp->dynticks_nesting -= DYNTICK_TASK_NEST_VALUE;
640
	}
641
}
642 643 644 645 646 647 648 649 650 651 652 653 654 655 656

/**
 * 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)
{
657 658 659
	unsigned long flags;

	local_irq_save(flags);
660
	rcu_eqs_enter(false);
661
	rcu_sysidle_enter(0);
662
	local_irq_restore(flags);
663
}
664
EXPORT_SYMBOL_GPL(rcu_idle_enter);
665

666
#ifdef CONFIG_RCU_USER_QS
667 668 669 670 671 672 673 674 675 676
/**
 * 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)
{
677
	rcu_eqs_enter(1);
678
}
679
#endif /* CONFIG_RCU_USER_QS */
680

681 682 683 684 685 686
/**
 * 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.
687
 *
688 689 690 691 692 693 694 695
 * 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.
696
 */
697
void rcu_irq_exit(void)
698 699
{
	unsigned long flags;
700
	long long oldval;
701 702 703
	struct rcu_dynticks *rdtp;

	local_irq_save(flags);
704
	rdtp = this_cpu_ptr(&rcu_dynticks);
705
	oldval = rdtp->dynticks_nesting;
706 707
	rdtp->dynticks_nesting--;
	WARN_ON_ONCE(rdtp->dynticks_nesting < 0);
708
	if (rdtp->dynticks_nesting)
709
		trace_rcu_dyntick(TPS("--="), oldval, rdtp->dynticks_nesting);
710
	else
711 712
		rcu_eqs_enter_common(oldval, true);
	rcu_sysidle_enter(1);
713 714 715 716
	local_irq_restore(flags);
}

/*
717
 * rcu_eqs_exit_common - current CPU moving away from extended quiescent state
718 719 720 721 722
 *
 * 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.
 */
723
static void rcu_eqs_exit_common(long long oldval, int user)
724
{
725 726
	struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks);

727
	rcu_dynticks_task_exit();
728
	smp_mb__before_atomic();  /* Force ordering w/previous sojourn. */
729 730
	atomic_inc(&rdtp->dynticks);
	/* CPUs seeing atomic_inc() must see later RCU read-side crit sects */
731
	smp_mb__after_atomic();  /* See above. */
732
	WARN_ON_ONCE(!(atomic_read(&rdtp->dynticks) & 0x1));
733
	rcu_cleanup_after_idle();
734
	trace_rcu_dyntick(TPS("End"), oldval, rdtp->dynticks_nesting);
735
	if (!user && !is_idle_task(current)) {
736 737
		struct task_struct *idle __maybe_unused =
			idle_task(smp_processor_id());
738

739
		trace_rcu_dyntick(TPS("Error on exit: not idle task"),
740
				  oldval, rdtp->dynticks_nesting);
741
		ftrace_dump(DUMP_ORIG);
742 743 744
		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! */
745 746 747
	}
}

748 749 750
/*
 * Exit an RCU extended quiescent state, which can be either the
 * idle loop or adaptive-tickless usermode execution.
751
 */
752
static void rcu_eqs_exit(bool user)
753 754 755 756
{
	struct rcu_dynticks *rdtp;
	long long oldval;

757
	rdtp = this_cpu_ptr(&rcu_dynticks);
758
	oldval = rdtp->dynticks_nesting;
759
	WARN_ON_ONCE(oldval < 0);
760
	if (oldval & DYNTICK_TASK_NEST_MASK) {
761
		rdtp->dynticks_nesting += DYNTICK_TASK_NEST_VALUE;
762
	} else {
763
		rdtp->dynticks_nesting = DYNTICK_TASK_EXIT_IDLE;
764
		rcu_eqs_exit_common(oldval, user);
765
	}
766
}
767 768 769 770 771 772 773 774 775 776 777 778 779 780

/**
 * 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)
{
781 782 783
	unsigned long flags;

	local_irq_save(flags);
784
	rcu_eqs_exit(false);
785
	rcu_sysidle_exit(0);
786
	local_irq_restore(flags);
787
}
788
EXPORT_SYMBOL_GPL(rcu_idle_exit);
789

790
#ifdef CONFIG_RCU_USER_QS
791 792 793 794 795 796 797 798
/**
 * 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)
{
799
	rcu_eqs_exit(1);
800
}
801
#endif /* CONFIG_RCU_USER_QS */
802

803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828
/**
 * 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);
829
	rdtp = this_cpu_ptr(&rcu_dynticks);
830 831 832
	oldval = rdtp->dynticks_nesting;
	rdtp->dynticks_nesting++;
	WARN_ON_ONCE(rdtp->dynticks_nesting == 0);
833
	if (oldval)
834
		trace_rcu_dyntick(TPS("++="), oldval, rdtp->dynticks_nesting);
835
	else
836 837
		rcu_eqs_exit_common(oldval, true);
	rcu_sysidle_exit(1);
838 839 840 841 842 843
	local_irq_restore(flags);
}

/**
 * rcu_nmi_enter - inform RCU of entry to NMI context
 *
844 845 846 847 848
 * If the CPU was idle from RCU's viewpoint, update rdtp->dynticks and
 * rdtp->dynticks_nmi_nesting to let the RCU grace-period handling know
 * that the CPU is active.  This implementation permits nested NMIs, as
 * long as the nesting level does not overflow an int.  (You will probably
 * run out of stack space first.)
849 850 851
 */
void rcu_nmi_enter(void)
{
852
	struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks);
853
	int incby = 2;
854

855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875
	/* Complain about underflow. */
	WARN_ON_ONCE(rdtp->dynticks_nmi_nesting < 0);

	/*
	 * If idle from RCU viewpoint, atomically increment ->dynticks
	 * to mark non-idle and increment ->dynticks_nmi_nesting by one.
	 * Otherwise, increment ->dynticks_nmi_nesting by two.  This means
	 * if ->dynticks_nmi_nesting is equal to one, we are guaranteed
	 * to be in the outermost NMI handler that interrupted an RCU-idle
	 * period (observation due to Andy Lutomirski).
	 */
	if (!(atomic_read(&rdtp->dynticks) & 0x1)) {
		smp_mb__before_atomic();  /* Force delay from prior write. */
		atomic_inc(&rdtp->dynticks);
		/* atomic_inc() before later RCU read-side crit sects */
		smp_mb__after_atomic();  /* See above. */
		WARN_ON_ONCE(!(atomic_read(&rdtp->dynticks) & 0x1));
		incby = 1;
	}
	rdtp->dynticks_nmi_nesting += incby;
	barrier();
876 877 878 879 880
}

/**
 * rcu_nmi_exit - inform RCU of exit from NMI context
 *
881 882 883 884
 * If we are returning from the outermost NMI handler that interrupted an
 * RCU-idle period, update rdtp->dynticks and rdtp->dynticks_nmi_nesting
 * to let the RCU grace-period handling know that the CPU is back to
 * being RCU-idle.
885 886 887
 */
void rcu_nmi_exit(void)
{
888
	struct rcu_dynticks *rdtp = this_cpu_ptr(&rcu_dynticks);
889

890 891 892 893 894 895 896 897 898 899 900 901 902 903
	/*
	 * Check for ->dynticks_nmi_nesting underflow and bad ->dynticks.
	 * (We are exiting an NMI handler, so RCU better be paying attention
	 * to us!)
	 */
	WARN_ON_ONCE(rdtp->dynticks_nmi_nesting <= 0);
	WARN_ON_ONCE(!(atomic_read(&rdtp->dynticks) & 0x1));

	/*
	 * If the nesting level is not 1, the CPU wasn't RCU-idle, so
	 * leave it in non-RCU-idle state.
	 */
	if (rdtp->dynticks_nmi_nesting != 1) {
		rdtp->dynticks_nmi_nesting -= 2;
904
		return;
905 906 907 908
	}

	/* This NMI interrupted an RCU-idle CPU, restore RCU-idleness. */
	rdtp->dynticks_nmi_nesting = 0;
909
	/* CPUs seeing atomic_inc() must see prior RCU read-side crit sects */
910
	smp_mb__before_atomic();  /* See above. */
911
	atomic_inc(&rdtp->dynticks);
912
	smp_mb__after_atomic();  /* Force delay to next write. */
913
	WARN_ON_ONCE(atomic_read(&rdtp->dynticks) & 0x1);
914 915 916
}

/**
917 918 919 920 921 922 923
 * __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.
 */
924
bool notrace __rcu_is_watching(void)
925 926 927 928 929 930
{
	return atomic_read(this_cpu_ptr(&rcu_dynticks.dynticks)) & 0x1;
}

/**
 * rcu_is_watching - see if RCU thinks that the current CPU is idle
931
 *
932
 * If the current CPU is in its idle loop and is neither in an interrupt
933
 * or NMI handler, return true.
934
 */
935
bool notrace rcu_is_watching(void)
936
{
937
	bool ret;
938 939

	preempt_disable();
940
	ret = __rcu_is_watching();
941 942
	preempt_enable();
	return ret;
943
}
944
EXPORT_SYMBOL_GPL(rcu_is_watching);
945

946
#if defined(CONFIG_PROVE_RCU) && defined(CONFIG_HOTPLUG_CPU)
947 948 949 950 951 952 953

/*
 * 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
954 955 956 957 958 959 960 961 962 963 964
 * 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.
965 966 967 968 969 970
 *
 * Disable checking if in an NMI handler because we cannot safely report
 * errors from NMI handlers anyway.
 */
bool rcu_lockdep_current_cpu_online(void)
{
971 972
	struct rcu_data *rdp;
	struct rcu_node *rnp;
973 974 975
	bool ret;

	if (in_nmi())
F
Fengguang Wu 已提交
976
		return true;
977
	preempt_disable();
978
	rdp = this_cpu_ptr(&rcu_sched_data);
979
	rnp = rdp->mynode;
980
	ret = (rdp->grpmask & rcu_rnp_online_cpus(rnp)) ||
981 982 983 984 985 986
	      !rcu_scheduler_fully_active;
	preempt_enable();
	return ret;
}
EXPORT_SYMBOL_GPL(rcu_lockdep_current_cpu_online);

987
#endif /* #if defined(CONFIG_PROVE_RCU) && defined(CONFIG_HOTPLUG_CPU) */
988

989
/**
990
 * rcu_is_cpu_rrupt_from_idle - see if idle or immediately interrupted from idle
991
 *
992 993 994
 * 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.
995
 */
996
static int rcu_is_cpu_rrupt_from_idle(void)
997
{
998
	return __this_cpu_read(rcu_dynticks.dynticks_nesting) <= 1;
999 1000 1001 1002 1003
}

/*
 * Snapshot the specified CPU's dynticks counter so that we can later
 * credit them with an implicit quiescent state.  Return 1 if this CPU
1004
 * is in dynticks idle mode, which is an extended quiescent state.
1005
 */
1006 1007
static int dyntick_save_progress_counter(struct rcu_data *rdp,
					 bool *isidle, unsigned long *maxj)
1008
{
1009
	rdp->dynticks_snap = atomic_add_return(0, &rdp->dynticks->dynticks);
1010
	rcu_sysidle_check_cpu(rdp, isidle, maxj);
1011 1012 1013 1014
	if ((rdp->dynticks_snap & 0x1) == 0) {
		trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, TPS("dti"));
		return 1;
	} else {
1015
		if (ULONG_CMP_LT(READ_ONCE(rdp->gpnum) + ULONG_MAX / 4,
1016
				 rdp->mynode->gpnum))
1017
			WRITE_ONCE(rdp->gpwrap, true);
1018 1019
		return 0;
	}
1020 1021 1022 1023 1024 1025
}

/*
 * 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()
1026
 * for this same CPU, or by virtue of having been offline.
1027
 */
1028 1029
static int rcu_implicit_dynticks_qs(struct rcu_data *rdp,
				    bool *isidle, unsigned long *maxj)
1030
{
1031
	unsigned int curr;
1032
	int *rcrmp;
1033
	unsigned int snap;
1034

1035 1036
	curr = (unsigned int)atomic_add_return(0, &rdp->dynticks->dynticks);
	snap = (unsigned int)rdp->dynticks_snap;
1037 1038 1039 1040 1041 1042 1043 1044 1045

	/*
	 * 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.
	 */
1046
	if ((curr & 0x1) == 0 || UINT_CMP_GE(curr, snap + 2)) {
1047
		trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, TPS("dti"));
1048 1049 1050 1051
		rdp->dynticks_fqs++;
		return 1;
	}

1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066
	/*
	 * 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)) {
1067
		trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, TPS("ofl"));
1068 1069 1070
		rdp->offline_fqs++;
		return 1;
	}
1071 1072

	/*
1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091
	 * 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.
1092
	 */
1093 1094 1095
	rcrmp = &per_cpu(rcu_sched_qs_mask, rdp->cpu);
	if (ULONG_CMP_GE(jiffies,
			 rdp->rsp->gp_start + jiffies_till_sched_qs) ||
1096
	    ULONG_CMP_GE(jiffies, rdp->rsp->jiffies_resched)) {
1097 1098 1099
		if (!(READ_ONCE(*rcrmp) & rdp->rsp->flavor_mask)) {
			WRITE_ONCE(rdp->cond_resched_completed,
				   READ_ONCE(rdp->mynode->completed));
1100
			smp_mb(); /* ->cond_resched_completed before *rcrmp. */
1101 1102
			WRITE_ONCE(*rcrmp,
				   READ_ONCE(*rcrmp) + rdp->rsp->flavor_mask);
1103 1104 1105 1106 1107 1108 1109
			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. */
		}
1110 1111
	}

1112
	return 0;
1113 1114 1115 1116
}

static void record_gp_stall_check_time(struct rcu_state *rsp)
{
1117
	unsigned long j = jiffies;
1118
	unsigned long j1;
1119 1120 1121

	rsp->gp_start = j;
	smp_wmb(); /* Record start time before stall time. */
1122
	j1 = rcu_jiffies_till_stall_check();
1123
	WRITE_ONCE(rsp->jiffies_stall, j + j1);
1124
	rsp->jiffies_resched = j + j1 / 2;
1125
	rsp->n_force_qs_gpstart = READ_ONCE(rsp->n_force_qs);
1126 1127
}

1128 1129 1130 1131 1132 1133 1134 1135 1136
/*
 * Complain about starvation of grace-period kthread.
 */
static void rcu_check_gp_kthread_starvation(struct rcu_state *rsp)
{
	unsigned long gpa;
	unsigned long j;

	j = jiffies;
1137
	gpa = READ_ONCE(rsp->gp_activity);
1138
	if (j - gpa > 2 * HZ)
1139 1140 1141
		pr_err("%s kthread starved for %ld jiffies! g%lu c%lu f%#x\n",
		       rsp->name, j - gpa,
		       rsp->gpnum, rsp->completed, rsp->gp_flags);
1142 1143
}

1144
/*
1145
 * Dump stacks of all tasks running on stalled CPUs.
1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163
 */
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);
	}
}

1164
static void print_other_cpu_stall(struct rcu_state *rsp, unsigned long gpnum)
1165 1166 1167 1168
{
	int cpu;
	long delta;
	unsigned long flags;
1169 1170
	unsigned long gpa;
	unsigned long j;
1171
	int ndetected = 0;
1172
	struct rcu_node *rnp = rcu_get_root(rsp);
1173
	long totqlen = 0;
1174 1175 1176

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

P
Paul E. McKenney 已提交
1177
	raw_spin_lock_irqsave(&rnp->lock, flags);
1178
	delta = jiffies - READ_ONCE(rsp->jiffies_stall);
1179
	if (delta < RCU_STALL_RAT_DELAY || !rcu_gp_in_progress(rsp)) {
P
Paul E. McKenney 已提交
1180
		raw_spin_unlock_irqrestore(&rnp->lock, flags);
1181 1182
		return;
	}
1183 1184
	WRITE_ONCE(rsp->jiffies_stall,
		   jiffies + 3 * rcu_jiffies_till_stall_check() + 3);
P
Paul E. McKenney 已提交
1185
	raw_spin_unlock_irqrestore(&rnp->lock, flags);
1186

1187 1188 1189 1190 1191
	/*
	 * OK, time to rat on our buddy...
	 * See Documentation/RCU/stallwarn.txt for info on how to debug
	 * RCU CPU stall warnings.
	 */
1192
	pr_err("INFO: %s detected stalls on CPUs/tasks:",
1193
	       rsp->name);
1194
	print_cpu_stall_info_begin();
1195
	rcu_for_each_leaf_node(rsp, rnp) {
1196
		raw_spin_lock_irqsave(&rnp->lock, flags);
1197
		ndetected += rcu_print_task_stall(rnp);
1198 1199 1200 1201 1202 1203 1204 1205
		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++;
				}
		}
1206
		raw_spin_unlock_irqrestore(&rnp->lock, flags);
1207
	}
1208 1209

	print_cpu_stall_info_end();
1210 1211
	for_each_possible_cpu(cpu)
		totqlen += per_cpu_ptr(rsp->rda, cpu)->qlen;
1212
	pr_cont("(detected by %d, t=%ld jiffies, g=%ld, c=%ld, q=%lu)\n",
1213
	       smp_processor_id(), (long)(jiffies - rsp->gp_start),
1214
	       (long)rsp->gpnum, (long)rsp->completed, totqlen);
1215
	if (ndetected) {
1216
		rcu_dump_cpu_stacks(rsp);
1217
	} else {
1218 1219
		if (READ_ONCE(rsp->gpnum) != gpnum ||
		    READ_ONCE(rsp->completed) == gpnum) {
1220 1221 1222
			pr_err("INFO: Stall ended before state dump start\n");
		} else {
			j = jiffies;
1223
			gpa = READ_ONCE(rsp->gp_activity);
1224
			pr_err("All QSes seen, last %s kthread activity %ld (%ld-%ld), jiffies_till_next_fqs=%ld, root ->qsmask %#lx\n",
1225
			       rsp->name, j - gpa, j, gpa,
1226 1227
			       jiffies_till_next_fqs,
			       rcu_get_root(rsp)->qsmask);
1228 1229 1230 1231
			/* In this case, the current CPU might be at fault. */
			sched_show_task(current);
		}
	}
1232

1233
	/* Complain about tasks blocking the grace period. */
1234 1235
	rcu_print_detail_task_stall(rsp);

1236 1237
	rcu_check_gp_kthread_starvation(rsp);

1238
	force_quiescent_state(rsp);  /* Kick them all. */
1239 1240 1241 1242
}

static void print_cpu_stall(struct rcu_state *rsp)
{
1243
	int cpu;
1244 1245
	unsigned long flags;
	struct rcu_node *rnp = rcu_get_root(rsp);
1246
	long totqlen = 0;
1247

1248 1249 1250 1251 1252
	/*
	 * OK, time to rat on ourselves...
	 * See Documentation/RCU/stallwarn.txt for info on how to debug
	 * RCU CPU stall warnings.
	 */
1253
	pr_err("INFO: %s self-detected stall on CPU", rsp->name);
1254 1255 1256
	print_cpu_stall_info_begin();
	print_cpu_stall_info(rsp, smp_processor_id());
	print_cpu_stall_info_end();
1257 1258
	for_each_possible_cpu(cpu)
		totqlen += per_cpu_ptr(rsp->rda, cpu)->qlen;
1259 1260 1261
	pr_cont(" (t=%lu jiffies g=%ld c=%ld q=%lu)\n",
		jiffies - rsp->gp_start,
		(long)rsp->gpnum, (long)rsp->completed, totqlen);
1262 1263 1264

	rcu_check_gp_kthread_starvation(rsp);

1265
	rcu_dump_cpu_stacks(rsp);
1266

P
Paul E. McKenney 已提交
1267
	raw_spin_lock_irqsave(&rnp->lock, flags);
1268 1269 1270
	if (ULONG_CMP_GE(jiffies, READ_ONCE(rsp->jiffies_stall)))
		WRITE_ONCE(rsp->jiffies_stall,
			   jiffies + 3 * rcu_jiffies_till_stall_check() + 3);
P
Paul E. McKenney 已提交
1271
	raw_spin_unlock_irqrestore(&rnp->lock, flags);
1272

1273 1274 1275 1276 1277 1278 1279 1280
	/*
	 * 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());
1281 1282 1283 1284
}

static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp)
{
1285 1286 1287
	unsigned long completed;
	unsigned long gpnum;
	unsigned long gps;
1288 1289
	unsigned long j;
	unsigned long js;
1290 1291
	struct rcu_node *rnp;

1292
	if (rcu_cpu_stall_suppress || !rcu_gp_in_progress(rsp))
1293
		return;
1294
	j = jiffies;
1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312

	/*
	 * 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.
	 */
1313
	gpnum = READ_ONCE(rsp->gpnum);
1314
	smp_rmb(); /* Pick up ->gpnum first... */
1315
	js = READ_ONCE(rsp->jiffies_stall);
1316
	smp_rmb(); /* ...then ->jiffies_stall before the rest... */
1317
	gps = READ_ONCE(rsp->gp_start);
1318
	smp_rmb(); /* ...and finally ->gp_start before ->completed. */
1319
	completed = READ_ONCE(rsp->completed);
1320 1321 1322 1323
	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. */
1324
	rnp = rdp->mynode;
1325
	if (rcu_gp_in_progress(rsp) &&
1326
	    (READ_ONCE(rnp->qsmask) & rdp->grpmask)) {
1327 1328 1329 1330

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

1331 1332
	} else if (rcu_gp_in_progress(rsp) &&
		   ULONG_CMP_GE(j, js + RCU_STALL_RAT_DELAY)) {
1333

1334
		/* They had a few time units to dump stack, so complain. */
1335
		print_other_cpu_stall(rsp, gpnum);
1336 1337 1338
	}
}

1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349
/**
 * 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)
{
1350 1351 1352
	struct rcu_state *rsp;

	for_each_rcu_flavor(rsp)
1353
		WRITE_ONCE(rsp->jiffies_stall, jiffies + ULONG_MAX / 2);
1354 1355
}

1356
/*
1357 1358 1359
 * Initialize the specified rcu_data structure's default callback list
 * to empty.  The default callback list is the one that is not used by
 * no-callbacks CPUs.
1360
 */
1361
static void init_default_callback_list(struct rcu_data *rdp)
1362 1363 1364 1365 1366 1367 1368 1369
{
	int i;

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

1370 1371 1372 1373 1374 1375 1376 1377 1378 1379
/*
 * Initialize the specified rcu_data structure's callback list to empty.
 */
static void init_callback_list(struct rcu_data *rdp)
{
	if (init_nocb_callback_list(rdp))
		return;
	init_default_callback_list(rdp);
}

1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408
/*
 * 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;
}

1409 1410 1411 1412 1413
/*
 * 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,
1414
				unsigned long c, const char *s)
1415 1416 1417 1418 1419 1420 1421 1422 1423
{
	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
1424 1425
 * rcu_node structure's ->need_future_gp field.  Returns true if there
 * is reason to awaken the grace-period kthread.
1426 1427 1428
 *
 * The caller must hold the specified rcu_node structure's ->lock.
 */
1429 1430 1431
static bool __maybe_unused
rcu_start_future_gp(struct rcu_node *rnp, struct rcu_data *rdp,
		    unsigned long *c_out)
1432 1433 1434
{
	unsigned long c;
	int i;
1435
	bool ret = false;
1436 1437 1438 1439 1440 1441 1442
	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);
1443
	trace_rcu_future_gp(rnp, rdp, c, TPS("Startleaf"));
1444
	if (rnp->need_future_gp[c & 0x1]) {
1445
		trace_rcu_future_gp(rnp, rdp, c, TPS("Prestartleaf"));
1446
		goto out;
1447 1448 1449 1450 1451 1452 1453
	}

	/*
	 * 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
1454 1455 1456 1457 1458 1459 1460
	 * 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.
1461 1462
	 */
	if (rnp->gpnum != rnp->completed ||
1463
	    READ_ONCE(rnp_root->gpnum) != READ_ONCE(rnp_root->completed)) {
1464
		rnp->need_future_gp[c & 0x1]++;
1465
		trace_rcu_future_gp(rnp, rdp, c, TPS("Startedleaf"));
1466
		goto out;
1467 1468 1469 1470 1471 1472 1473
	}

	/*
	 * 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).
	 */
1474
	if (rnp != rnp_root) {
1475
		raw_spin_lock(&rnp_root->lock);
1476 1477
		smp_mb__after_unlock_lock();
	}
1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494

	/*
	 * 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]) {
1495
		trace_rcu_future_gp(rnp, rdp, c, TPS("Prestartedroot"));
1496 1497 1498 1499 1500 1501 1502 1503
		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) {
1504
		trace_rcu_future_gp(rnp, rdp, c, TPS("Startedleafroot"));
1505
	} else {
1506
		trace_rcu_future_gp(rnp, rdp, c, TPS("Startedroot"));
1507
		ret = rcu_start_gp_advanced(rdp->rsp, rnp_root, rdp);
1508 1509 1510 1511
	}
unlock_out:
	if (rnp != rnp_root)
		raw_spin_unlock(&rnp_root->lock);
1512 1513 1514 1515
out:
	if (c_out != NULL)
		*c_out = c;
	return ret;
1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532
}

/*
 * 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];
1533 1534
	trace_rcu_future_gp(rnp, rdp, c,
			    needmore ? TPS("CleanupMore") : TPS("Cleanup"));
1535 1536 1537
	return needmore;
}

1538 1539 1540 1541 1542 1543 1544 1545 1546 1547
/*
 * 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 ||
1548
	    !READ_ONCE(rsp->gp_flags) ||
1549 1550 1551 1552 1553
	    !rsp->gp_kthread)
		return;
	wake_up(&rsp->gp_wq);
}

1554 1555 1556 1557 1558 1559 1560
/*
 * 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
1561 1562
 * not hurt to call it repeatedly.  Returns an flag saying that we should
 * awaken the RCU grace-period kthread.
1563 1564 1565
 *
 * The caller must hold rnp->lock with interrupts disabled.
 */
1566
static bool rcu_accelerate_cbs(struct rcu_state *rsp, struct rcu_node *rnp,
1567 1568 1569 1570
			       struct rcu_data *rdp)
{
	unsigned long c;
	int i;
1571
	bool ret;
1572 1573 1574

	/* If the CPU has no callbacks, nothing to do. */
	if (!rdp->nxttail[RCU_NEXT_TAIL] || !*rdp->nxttail[RCU_DONE_TAIL])
1575
		return false;
1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603

	/*
	 * 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)
1604
		return false;
1605 1606 1607 1608 1609 1610 1611 1612 1613 1614

	/*
	 * 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;
	}
1615
	/* Record any needed additional grace periods. */
1616
	ret = rcu_start_future_gp(rnp, rdp, NULL);
1617 1618 1619

	/* Trace depending on how much we were able to accelerate. */
	if (!*rdp->nxttail[RCU_WAIT_TAIL])
1620
		trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("AccWaitCB"));
1621
	else
1622
		trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("AccReadyCB"));
1623
	return ret;
1624 1625 1626 1627 1628 1629 1630 1631
}

/*
 * 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...
1632
 * Returns true if the RCU grace-period kthread needs to be awakened.
1633 1634 1635
 *
 * The caller must hold rnp->lock with interrupts disabled.
 */
1636
static bool rcu_advance_cbs(struct rcu_state *rsp, struct rcu_node *rnp,
1637 1638 1639 1640 1641 1642
			    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])
1643
		return false;
1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666

	/*
	 * 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. */
1667
	return rcu_accelerate_cbs(rsp, rnp, rdp);
1668 1669
}

1670
/*
1671 1672 1673
 * 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.
1674
 * Returns true if the grace-period kthread needs to be awakened.
1675
 */
1676 1677
static bool __note_gp_changes(struct rcu_state *rsp, struct rcu_node *rnp,
			      struct rcu_data *rdp)
1678
{
1679 1680
	bool ret;

1681
	/* Handle the ends of any preceding grace periods first. */
1682
	if (rdp->completed == rnp->completed &&
1683
	    !unlikely(READ_ONCE(rdp->gpwrap))) {
1684

1685
		/* No grace period end, so just accelerate recent callbacks. */
1686
		ret = rcu_accelerate_cbs(rsp, rnp, rdp);
1687

1688 1689 1690
	} else {

		/* Advance callbacks. */
1691
		ret = rcu_advance_cbs(rsp, rnp, rdp);
1692 1693 1694

		/* Remember that we saw this grace-period completion. */
		rdp->completed = rnp->completed;
1695
		trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("cpuend"));
1696
	}
1697

1698
	if (rdp->gpnum != rnp->gpnum || unlikely(READ_ONCE(rdp->gpwrap))) {
1699 1700 1701 1702 1703 1704
		/*
		 * 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;
1705
		trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("cpustart"));
1706
		rdp->passed_quiesce = 0;
1707
		rdp->rcu_qs_ctr_snap = __this_cpu_read(rcu_qs_ctr);
1708 1709
		rdp->qs_pending = !!(rnp->qsmask & rdp->grpmask);
		zero_cpu_stall_ticks(rdp);
1710
		WRITE_ONCE(rdp->gpwrap, false);
1711
	}
1712
	return ret;
1713 1714
}

1715
static void note_gp_changes(struct rcu_state *rsp, struct rcu_data *rdp)
1716 1717
{
	unsigned long flags;
1718
	bool needwake;
1719 1720 1721 1722
	struct rcu_node *rnp;

	local_irq_save(flags);
	rnp = rdp->mynode;
1723 1724 1725
	if ((rdp->gpnum == READ_ONCE(rnp->gpnum) &&
	     rdp->completed == READ_ONCE(rnp->completed) &&
	     !unlikely(READ_ONCE(rdp->gpwrap))) || /* w/out lock. */
1726 1727 1728 1729
	    !raw_spin_trylock(&rnp->lock)) { /* irqs already off, so later. */
		local_irq_restore(flags);
		return;
	}
1730
	smp_mb__after_unlock_lock();
1731
	needwake = __note_gp_changes(rsp, rnp, rdp);
1732
	raw_spin_unlock_irqrestore(&rnp->lock, flags);
1733 1734
	if (needwake)
		rcu_gp_kthread_wake(rsp);
1735 1736
}

1737
/*
1738
 * Initialize a new grace period.  Return 0 if no grace period required.
1739
 */
1740
static int rcu_gp_init(struct rcu_state *rsp)
1741
{
1742
	unsigned long oldmask;
1743
	struct rcu_data *rdp;
1744
	struct rcu_node *rnp = rcu_get_root(rsp);
1745

1746
	WRITE_ONCE(rsp->gp_activity, jiffies);
1747
	raw_spin_lock_irq(&rnp->lock);
1748
	smp_mb__after_unlock_lock();
1749
	if (!READ_ONCE(rsp->gp_flags)) {
1750 1751 1752 1753
		/* Spurious wakeup, tell caller to go back to sleep.  */
		raw_spin_unlock_irq(&rnp->lock);
		return 0;
	}
1754
	WRITE_ONCE(rsp->gp_flags, 0); /* Clear all flags: New grace period. */
1755

1756 1757 1758 1759 1760
	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.
		 */
1761 1762 1763 1764 1765
		raw_spin_unlock_irq(&rnp->lock);
		return 0;
	}

	/* Advance to a new grace period and initialize state. */
1766
	record_gp_stall_check_time(rsp);
1767 1768
	/* Record GP times before starting GP, hence smp_store_release(). */
	smp_store_release(&rsp->gpnum, rsp->gpnum + 1);
1769
	trace_rcu_grace_period(rsp->name, rsp->gpnum, TPS("start"));
1770 1771
	raw_spin_unlock_irq(&rnp->lock);

1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819
	/*
	 * Apply per-leaf buffered online and offline operations to the
	 * rcu_node tree.  Note that this new grace period need not wait
	 * for subsequent online CPUs, and that quiescent-state forcing
	 * will handle subsequent offline CPUs.
	 */
	rcu_for_each_leaf_node(rsp, rnp) {
		raw_spin_lock_irq(&rnp->lock);
		smp_mb__after_unlock_lock();
		if (rnp->qsmaskinit == rnp->qsmaskinitnext &&
		    !rnp->wait_blkd_tasks) {
			/* Nothing to do on this leaf rcu_node structure. */
			raw_spin_unlock_irq(&rnp->lock);
			continue;
		}

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

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

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

		raw_spin_unlock_irq(&rnp->lock);
	}
1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834

	/*
	 * 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) {
1835
		raw_spin_lock_irq(&rnp->lock);
1836
		smp_mb__after_unlock_lock();
1837
		rdp = this_cpu_ptr(rsp->rda);
1838 1839
		rcu_preempt_check_blocked_tasks(rnp);
		rnp->qsmask = rnp->qsmaskinit;
1840
		WRITE_ONCE(rnp->gpnum, rsp->gpnum);
1841
		if (WARN_ON_ONCE(rnp->completed != rsp->completed))
1842
			WRITE_ONCE(rnp->completed, rsp->completed);
1843
		if (rnp == rdp->mynode)
1844
			(void)__note_gp_changes(rsp, rnp, rdp);
1845 1846 1847 1848 1849
		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);
1850
		cond_resched_rcu_qs();
1851
		WRITE_ONCE(rsp->gp_activity, jiffies);
1852 1853
		if (gp_init_delay > 0 &&
		    !(rsp->gpnum % (rcu_num_nodes * PER_RCU_NODE_PERIOD)))
1854
			schedule_timeout_uninterruptible(gp_init_delay);
1855
	}
1856

1857 1858
	return 1;
}
1859

1860 1861 1862
/*
 * Do one round of quiescent-state forcing.
 */
1863
static int rcu_gp_fqs(struct rcu_state *rsp, int fqs_state_in)
1864 1865
{
	int fqs_state = fqs_state_in;
1866 1867
	bool isidle = false;
	unsigned long maxj;
1868 1869
	struct rcu_node *rnp = rcu_get_root(rsp);

1870
	WRITE_ONCE(rsp->gp_activity, jiffies);
1871 1872 1873
	rsp->n_force_qs++;
	if (fqs_state == RCU_SAVE_DYNTICK) {
		/* Collect dyntick-idle snapshots. */
1874
		if (is_sysidle_rcu_state(rsp)) {
1875
			isidle = true;
1876 1877
			maxj = jiffies - ULONG_MAX / 4;
		}
1878 1879
		force_qs_rnp(rsp, dyntick_save_progress_counter,
			     &isidle, &maxj);
1880
		rcu_sysidle_report_gp(rsp, isidle, maxj);
1881 1882 1883
		fqs_state = RCU_FORCE_QS;
	} else {
		/* Handle dyntick-idle and offline CPUs. */
1884
		isidle = true;
1885
		force_qs_rnp(rsp, rcu_implicit_dynticks_qs, &isidle, &maxj);
1886 1887
	}
	/* Clear flag to prevent immediate re-entry. */
1888
	if (READ_ONCE(rsp->gp_flags) & RCU_GP_FLAG_FQS) {
1889
		raw_spin_lock_irq(&rnp->lock);
1890
		smp_mb__after_unlock_lock();
1891 1892
		WRITE_ONCE(rsp->gp_flags,
			   READ_ONCE(rsp->gp_flags) & ~RCU_GP_FLAG_FQS);
1893 1894 1895 1896 1897
		raw_spin_unlock_irq(&rnp->lock);
	}
	return fqs_state;
}

1898 1899 1900
/*
 * Clean up after the old grace period.
 */
1901
static void rcu_gp_cleanup(struct rcu_state *rsp)
1902 1903
{
	unsigned long gp_duration;
1904
	bool needgp = false;
1905
	int nocb = 0;
1906 1907
	struct rcu_data *rdp;
	struct rcu_node *rnp = rcu_get_root(rsp);
1908

1909
	WRITE_ONCE(rsp->gp_activity, jiffies);
1910
	raw_spin_lock_irq(&rnp->lock);
1911
	smp_mb__after_unlock_lock();
1912 1913 1914
	gp_duration = jiffies - rsp->gp_start;
	if (gp_duration > rsp->gp_max)
		rsp->gp_max = gp_duration;
1915

1916 1917 1918 1919 1920 1921 1922 1923
	/*
	 * 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.
	 */
1924
	raw_spin_unlock_irq(&rnp->lock);
1925

1926 1927 1928 1929 1930 1931 1932 1933 1934 1935
	/*
	 * 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) {
1936
		raw_spin_lock_irq(&rnp->lock);
1937
		smp_mb__after_unlock_lock();
1938 1939
		WARN_ON_ONCE(rcu_preempt_blocked_readers_cgp(rnp));
		WARN_ON_ONCE(rnp->qsmask);
1940
		WRITE_ONCE(rnp->completed, rsp->gpnum);
1941 1942
		rdp = this_cpu_ptr(rsp->rda);
		if (rnp == rdp->mynode)
1943
			needgp = __note_gp_changes(rsp, rnp, rdp) || needgp;
1944
		/* smp_mb() provided by prior unlock-lock pair. */
1945
		nocb += rcu_future_gp_cleanup(rsp, rnp);
1946
		raw_spin_unlock_irq(&rnp->lock);
1947
		cond_resched_rcu_qs();
1948
		WRITE_ONCE(rsp->gp_activity, jiffies);
1949
	}
1950 1951
	rnp = rcu_get_root(rsp);
	raw_spin_lock_irq(&rnp->lock);
1952
	smp_mb__after_unlock_lock(); /* Order GP before ->completed update. */
1953
	rcu_nocb_gp_set(rnp, nocb);
1954

1955
	/* Declare grace period done. */
1956
	WRITE_ONCE(rsp->completed, rsp->gpnum);
1957
	trace_rcu_grace_period(rsp->name, rsp->completed, TPS("end"));
1958
	rsp->fqs_state = RCU_GP_IDLE;
1959
	rdp = this_cpu_ptr(rsp->rda);
1960 1961 1962
	/* Advance CBs to reduce false positives below. */
	needgp = rcu_advance_cbs(rsp, rnp, rdp) || needgp;
	if (needgp || cpu_needs_another_gp(rsp, rdp)) {
1963
		WRITE_ONCE(rsp->gp_flags, RCU_GP_FLAG_INIT);
1964
		trace_rcu_grace_period(rsp->name,
1965
				       READ_ONCE(rsp->gpnum),
1966 1967
				       TPS("newreq"));
	}
1968 1969 1970 1971 1972 1973 1974 1975
	raw_spin_unlock_irq(&rnp->lock);
}

/*
 * Body of kthread that handles grace periods.
 */
static int __noreturn rcu_gp_kthread(void *arg)
{
1976
	int fqs_state;
1977
	int gf;
1978
	unsigned long j;
1979
	int ret;
1980 1981 1982
	struct rcu_state *rsp = arg;
	struct rcu_node *rnp = rcu_get_root(rsp);

1983
	rcu_bind_gp_kthread();
1984 1985 1986 1987
	for (;;) {

		/* Handle grace-period start. */
		for (;;) {
1988
			trace_rcu_grace_period(rsp->name,
1989
					       READ_ONCE(rsp->gpnum),
1990
					       TPS("reqwait"));
1991
			rsp->gp_state = RCU_GP_WAIT_GPS;
1992
			wait_event_interruptible(rsp->gp_wq,
1993
						 READ_ONCE(rsp->gp_flags) &
1994
						 RCU_GP_FLAG_INIT);
1995
			/* Locking provides needed memory barrier. */
1996
			if (rcu_gp_init(rsp))
1997
				break;
1998
			cond_resched_rcu_qs();
1999
			WRITE_ONCE(rsp->gp_activity, jiffies);
2000
			WARN_ON(signal_pending(current));
2001
			trace_rcu_grace_period(rsp->name,
2002
					       READ_ONCE(rsp->gpnum),
2003
					       TPS("reqwaitsig"));
2004
		}
2005

2006 2007
		/* Handle quiescent-state forcing. */
		fqs_state = RCU_SAVE_DYNTICK;
2008 2009 2010 2011 2012
		j = jiffies_till_first_fqs;
		if (j > HZ) {
			j = HZ;
			jiffies_till_first_fqs = HZ;
		}
2013
		ret = 0;
2014
		for (;;) {
2015 2016
			if (!ret)
				rsp->jiffies_force_qs = jiffies + j;
2017
			trace_rcu_grace_period(rsp->name,
2018
					       READ_ONCE(rsp->gpnum),
2019
					       TPS("fqswait"));
2020
			rsp->gp_state = RCU_GP_WAIT_FQS;
2021
			ret = wait_event_interruptible_timeout(rsp->gp_wq,
2022
					((gf = READ_ONCE(rsp->gp_flags)) &
2023
					 RCU_GP_FLAG_FQS) ||
2024
					(!READ_ONCE(rnp->qsmask) &&
2025
					 !rcu_preempt_blocked_readers_cgp(rnp)),
2026
					j);
2027
			/* Locking provides needed memory barriers. */
2028
			/* If grace period done, leave loop. */
2029
			if (!READ_ONCE(rnp->qsmask) &&
2030
			    !rcu_preempt_blocked_readers_cgp(rnp))
2031
				break;
2032
			/* If time for quiescent-state forcing, do it. */
2033 2034
			if (ULONG_CMP_GE(jiffies, rsp->jiffies_force_qs) ||
			    (gf & RCU_GP_FLAG_FQS)) {
2035
				trace_rcu_grace_period(rsp->name,
2036
						       READ_ONCE(rsp->gpnum),
2037
						       TPS("fqsstart"));
2038
				fqs_state = rcu_gp_fqs(rsp, fqs_state);
2039
				trace_rcu_grace_period(rsp->name,
2040
						       READ_ONCE(rsp->gpnum),
2041
						       TPS("fqsend"));
2042
				cond_resched_rcu_qs();
2043
				WRITE_ONCE(rsp->gp_activity, jiffies);
2044 2045
			} else {
				/* Deal with stray signal. */
2046
				cond_resched_rcu_qs();
2047
				WRITE_ONCE(rsp->gp_activity, jiffies);
2048
				WARN_ON(signal_pending(current));
2049
				trace_rcu_grace_period(rsp->name,
2050
						       READ_ONCE(rsp->gpnum),
2051
						       TPS("fqswaitsig"));
2052
			}
2053 2054 2055 2056 2057 2058 2059 2060
			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;
			}
2061
		}
2062 2063 2064

		/* Handle grace-period end. */
		rcu_gp_cleanup(rsp);
2065 2066 2067
	}
}

2068 2069 2070
/*
 * Start a new RCU grace period if warranted, re-initializing the hierarchy
 * in preparation for detecting the next grace period.  The caller must hold
2071
 * the root node's ->lock and hard irqs must be disabled.
2072 2073 2074 2075
 *
 * 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.
2076 2077
 *
 * Returns true if the grace-period kthread must be awakened.
2078
 */
2079
static bool
2080 2081
rcu_start_gp_advanced(struct rcu_state *rsp, struct rcu_node *rnp,
		      struct rcu_data *rdp)
2082
{
2083
	if (!rsp->gp_kthread || !cpu_needs_another_gp(rsp, rdp)) {
2084
		/*
2085
		 * Either we have not yet spawned the grace-period
2086 2087
		 * task, this CPU does not need another grace period,
		 * or a grace period is already in progress.
2088
		 * Either way, don't start a new grace period.
2089
		 */
2090
		return false;
2091
	}
2092 2093
	WRITE_ONCE(rsp->gp_flags, RCU_GP_FLAG_INIT);
	trace_rcu_grace_period(rsp->name, READ_ONCE(rsp->gpnum),
2094
			       TPS("newreq"));
2095

2096 2097
	/*
	 * We can't do wakeups while holding the rnp->lock, as that
2098
	 * could cause possible deadlocks with the rq->lock. Defer
2099
	 * the wakeup to our caller.
2100
	 */
2101
	return true;
2102 2103
}

2104 2105 2106 2107 2108 2109
/*
 * 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.
2110 2111
 *
 * Returns true if the grace-period kthread needs to be awakened.
2112
 */
2113
static bool rcu_start_gp(struct rcu_state *rsp)
2114 2115 2116
{
	struct rcu_data *rdp = this_cpu_ptr(rsp->rda);
	struct rcu_node *rnp = rcu_get_root(rsp);
2117
	bool ret = false;
2118 2119 2120 2121 2122 2123 2124 2125 2126

	/*
	 * 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!
	 */
2127 2128 2129
	ret = rcu_advance_cbs(rsp, rnp, rdp) || ret;
	ret = rcu_start_gp_advanced(rsp, rnp, rdp) || ret;
	return ret;
2130 2131
}

2132
/*
P
Paul E. McKenney 已提交
2133 2134 2135
 * 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
2136 2137
 * if one is needed.  Note that the caller must hold rnp->lock, which
 * is released before return.
2138
 */
P
Paul E. McKenney 已提交
2139
static void rcu_report_qs_rsp(struct rcu_state *rsp, unsigned long flags)
2140
	__releases(rcu_get_root(rsp)->lock)
2141
{
2142
	WARN_ON_ONCE(!rcu_gp_in_progress(rsp));
2143
	WRITE_ONCE(rsp->gp_flags, READ_ONCE(rsp->gp_flags) | RCU_GP_FLAG_FQS);
2144
	raw_spin_unlock_irqrestore(&rcu_get_root(rsp)->lock, flags);
2145
	rcu_gp_kthread_wake(rsp);
2146 2147
}

2148
/*
P
Paul E. McKenney 已提交
2149 2150 2151
 * 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
2152 2153 2154 2155 2156
 * must be represented by the same rcu_node structure (which need not be a
 * leaf rcu_node structure, though it often will be).  The gps parameter
 * is the grace-period snapshot, which means that the quiescent states
 * are valid only if rnp->gpnum is equal to gps.  That structure's lock
 * must be held upon entry, and it is released before return.
2157 2158
 */
static void
P
Paul E. McKenney 已提交
2159
rcu_report_qs_rnp(unsigned long mask, struct rcu_state *rsp,
2160
		  struct rcu_node *rnp, unsigned long gps, unsigned long flags)
2161 2162
	__releases(rnp->lock)
{
2163
	unsigned long oldmask = 0;
2164 2165
	struct rcu_node *rnp_c;

2166 2167
	/* Walk up the rcu_node hierarchy. */
	for (;;) {
2168
		if (!(rnp->qsmask & mask) || rnp->gpnum != gps) {
2169

2170 2171 2172 2173
			/*
			 * Our bit has already been cleared, or the
			 * relevant grace period is already over, so done.
			 */
P
Paul E. McKenney 已提交
2174
			raw_spin_unlock_irqrestore(&rnp->lock, flags);
2175 2176
			return;
		}
2177
		WARN_ON_ONCE(oldmask); /* Any child must be all zeroed! */
2178
		rnp->qsmask &= ~mask;
2179 2180 2181 2182
		trace_rcu_quiescent_state_report(rsp->name, rnp->gpnum,
						 mask, rnp->qsmask, rnp->level,
						 rnp->grplo, rnp->grphi,
						 !!rnp->gp_tasks);
2183
		if (rnp->qsmask != 0 || rcu_preempt_blocked_readers_cgp(rnp)) {
2184 2185

			/* Other bits still set at this level, so done. */
P
Paul E. McKenney 已提交
2186
			raw_spin_unlock_irqrestore(&rnp->lock, flags);
2187 2188 2189 2190 2191 2192 2193 2194 2195
			return;
		}
		mask = rnp->grpmask;
		if (rnp->parent == NULL) {

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

			break;
		}
P
Paul E. McKenney 已提交
2196
		raw_spin_unlock_irqrestore(&rnp->lock, flags);
2197
		rnp_c = rnp;
2198
		rnp = rnp->parent;
P
Paul E. McKenney 已提交
2199
		raw_spin_lock_irqsave(&rnp->lock, flags);
2200
		smp_mb__after_unlock_lock();
2201
		oldmask = rnp_c->qsmask;
2202 2203 2204 2205
	}

	/*
	 * Get here if we are the last CPU to pass through a quiescent
P
Paul E. McKenney 已提交
2206
	 * state for this grace period.  Invoke rcu_report_qs_rsp()
2207
	 * to clean up and start the next grace period if one is needed.
2208
	 */
P
Paul E. McKenney 已提交
2209
	rcu_report_qs_rsp(rsp, flags); /* releases rnp->lock. */
2210 2211
}

2212 2213 2214 2215 2216 2217 2218
/*
 * Record a quiescent state for all tasks that were previously queued
 * on the specified rcu_node structure and that were blocking the current
 * RCU grace period.  The caller must hold the specified rnp->lock with
 * irqs disabled, and this lock is released upon return, but irqs remain
 * disabled.
 */
2219
static void rcu_report_unblock_qs_rnp(struct rcu_state *rsp,
2220 2221 2222
				      struct rcu_node *rnp, unsigned long flags)
	__releases(rnp->lock)
{
2223
	unsigned long gps;
2224 2225 2226
	unsigned long mask;
	struct rcu_node *rnp_p;

2227 2228
	if (rcu_state_p == &rcu_sched_state || rsp != rcu_state_p ||
	    rnp->qsmask != 0 || rcu_preempt_blocked_readers_cgp(rnp)) {
2229 2230 2231 2232 2233 2234 2235
		raw_spin_unlock_irqrestore(&rnp->lock, flags);
		return;  /* Still need more quiescent states! */
	}

	rnp_p = rnp->parent;
	if (rnp_p == NULL) {
		/*
2236 2237
		 * Only one rcu_node structure in the tree, so don't
		 * try to report up to its nonexistent parent!
2238 2239 2240 2241 2242
		 */
		rcu_report_qs_rsp(rsp, flags);
		return;
	}

2243 2244
	/* Report up the rest of the hierarchy, tracking current ->gpnum. */
	gps = rnp->gpnum;
2245 2246 2247 2248
	mask = rnp->grpmask;
	raw_spin_unlock(&rnp->lock);	/* irqs remain disabled. */
	raw_spin_lock(&rnp_p->lock);	/* irqs already disabled. */
	smp_mb__after_unlock_lock();
2249
	rcu_report_qs_rnp(mask, rsp, rnp_p, gps, flags);
2250 2251
}

2252
/*
P
Paul E. McKenney 已提交
2253 2254 2255 2256 2257 2258 2259
 * 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!
2260 2261
 */
static void
2262
rcu_report_qs_rdp(int cpu, struct rcu_state *rsp, struct rcu_data *rdp)
2263 2264 2265
{
	unsigned long flags;
	unsigned long mask;
2266
	bool needwake;
2267 2268 2269
	struct rcu_node *rnp;

	rnp = rdp->mynode;
P
Paul E. McKenney 已提交
2270
	raw_spin_lock_irqsave(&rnp->lock, flags);
2271
	smp_mb__after_unlock_lock();
2272 2273 2274 2275
	if ((rdp->passed_quiesce == 0 &&
	     rdp->rcu_qs_ctr_snap == __this_cpu_read(rcu_qs_ctr)) ||
	    rdp->gpnum != rnp->gpnum || rnp->completed == rnp->gpnum ||
	    rdp->gpwrap) {
2276 2277

		/*
2278 2279 2280 2281
		 * 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.
2282
		 */
2283
		rdp->passed_quiesce = 0;	/* need qs for new gp. */
2284
		rdp->rcu_qs_ctr_snap = __this_cpu_read(rcu_qs_ctr);
P
Paul E. McKenney 已提交
2285
		raw_spin_unlock_irqrestore(&rnp->lock, flags);
2286 2287 2288 2289
		return;
	}
	mask = rdp->grpmask;
	if ((rnp->qsmask & mask) == 0) {
P
Paul E. McKenney 已提交
2290
		raw_spin_unlock_irqrestore(&rnp->lock, flags);
2291 2292 2293 2294 2295 2296 2297
	} 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.
		 */
2298
		needwake = rcu_accelerate_cbs(rsp, rnp, rdp);
2299

2300 2301
		rcu_report_qs_rnp(mask, rsp, rnp, rnp->gpnum, flags);
		/* ^^^ Released rnp->lock */
2302 2303
		if (needwake)
			rcu_gp_kthread_wake(rsp);
2304 2305 2306 2307 2308 2309 2310 2311 2312 2313 2314 2315
	}
}

/*
 * 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)
{
2316 2317
	/* Check for grace-period ends and beginnings. */
	note_gp_changes(rsp, rdp);
2318 2319 2320 2321 2322 2323 2324 2325 2326 2327 2328 2329

	/*
	 * 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.
	 */
2330 2331
	if (!rdp->passed_quiesce &&
	    rdp->rcu_qs_ctr_snap == __this_cpu_read(rcu_qs_ctr))
2332 2333
		return;

P
Paul E. McKenney 已提交
2334 2335 2336 2337
	/*
	 * Tell RCU we are done (but rcu_report_qs_rdp() will be the
	 * judge of that).
	 */
2338
	rcu_report_qs_rdp(rdp->cpu, rsp, rdp);
2339 2340 2341 2342
}

#ifdef CONFIG_HOTPLUG_CPU

2343
/*
2344 2345
 * Send the specified CPU's RCU callbacks to the orphanage.  The
 * specified CPU must be offline, and the caller must hold the
2346
 * ->orphan_lock.
2347
 */
2348 2349 2350
static void
rcu_send_cbs_to_orphanage(int cpu, struct rcu_state *rsp,
			  struct rcu_node *rnp, struct rcu_data *rdp)
2351
{
P
Paul E. McKenney 已提交
2352
	/* No-CBs CPUs do not have orphanable callbacks. */
2353
	if (rcu_is_nocb_cpu(rdp->cpu))
P
Paul E. McKenney 已提交
2354 2355
		return;

2356 2357
	/*
	 * Orphan the callbacks.  First adjust the counts.  This is safe
2358 2359
	 * because _rcu_barrier() excludes CPU-hotplug operations, so it
	 * cannot be running now.  Thus no memory barrier is required.
2360
	 */
2361
	if (rdp->nxtlist != NULL) {
2362 2363 2364
		rsp->qlen_lazy += rdp->qlen_lazy;
		rsp->qlen += rdp->qlen;
		rdp->n_cbs_orphaned += rdp->qlen;
2365
		rdp->qlen_lazy = 0;
2366
		WRITE_ONCE(rdp->qlen, 0);
2367 2368 2369
	}

	/*
2370 2371 2372 2373 2374 2375 2376
	 * 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.
2377
	 */
2378 2379 2380 2381
	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;
2382 2383 2384
	}

	/*
2385 2386 2387
	 * 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.
2388
	 */
2389
	if (rdp->nxtlist != NULL) {
2390 2391
		*rsp->orphan_donetail = rdp->nxtlist;
		rsp->orphan_donetail = rdp->nxttail[RCU_DONE_TAIL];
2392
	}
2393

2394 2395 2396 2397
	/*
	 * Finally, initialize the rcu_data structure's list to empty and
	 * disallow further callbacks on this CPU.
	 */
2398
	init_callback_list(rdp);
2399
	rdp->nxttail[RCU_NEXT_TAIL] = NULL;
2400 2401 2402 2403
}

/*
 * Adopt the RCU callbacks from the specified rcu_state structure's
2404
 * orphanage.  The caller must hold the ->orphan_lock.
2405
 */
2406
static void rcu_adopt_orphan_cbs(struct rcu_state *rsp, unsigned long flags)
2407 2408
{
	int i;
2409
	struct rcu_data *rdp = raw_cpu_ptr(rsp->rda);
2410

P
Paul E. McKenney 已提交
2411
	/* No-CBs CPUs are handled specially. */
2412
	if (rcu_nocb_adopt_orphan_cbs(rsp, rdp, flags))
P
Paul E. McKenney 已提交
2413 2414
		return;

2415 2416 2417 2418
	/* Do the accounting first. */
	rdp->qlen_lazy += rsp->qlen_lazy;
	rdp->qlen += rsp->qlen;
	rdp->n_cbs_adopted += rsp->qlen;
2419 2420
	if (rsp->qlen_lazy != rsp->qlen)
		rcu_idle_count_callbacks_posted();
2421 2422 2423 2424 2425 2426 2427 2428 2429 2430 2431 2432 2433 2434 2435 2436 2437 2438 2439 2440 2441 2442 2443 2444 2445 2446 2447 2448 2449 2450 2451 2452 2453 2454 2455 2456 2457 2458 2459
	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);
2460 2461
	trace_rcu_grace_period(rsp->name,
			       rnp->gpnum + 1 - !!(rnp->qsmask & mask),
2462
			       TPS("cpuofl"));
2463 2464
}

2465 2466 2467 2468 2469 2470 2471 2472 2473 2474 2475 2476 2477 2478 2479 2480 2481 2482 2483 2484 2485 2486 2487 2488 2489 2490 2491 2492 2493 2494 2495 2496
/*
 * All CPUs for the specified rcu_node structure have gone offline,
 * and all tasks that were preempted within an RCU read-side critical
 * section while running on one of those CPUs have since exited their RCU
 * read-side critical section.  Some other CPU is reporting this fact with
 * the specified rcu_node structure's ->lock held and interrupts disabled.
 * This function therefore goes up the tree of rcu_node structures,
 * clearing the corresponding bits in the ->qsmaskinit fields.  Note that
 * the leaf rcu_node structure's ->qsmaskinit field has already been
 * updated
 *
 * This function does check that the specified rcu_node structure has
 * all CPUs offline and no blocked tasks, so it is OK to invoke it
 * prematurely.  That said, invoking it after the fact will cost you
 * a needless lock acquisition.  So once it has done its work, don't
 * invoke it again.
 */
static void rcu_cleanup_dead_rnp(struct rcu_node *rnp_leaf)
{
	long mask;
	struct rcu_node *rnp = rnp_leaf;

	if (rnp->qsmaskinit || rcu_preempt_has_tasks(rnp))
		return;
	for (;;) {
		mask = rnp->grpmask;
		rnp = rnp->parent;
		if (!rnp)
			break;
		raw_spin_lock(&rnp->lock); /* irqs already disabled. */
		smp_mb__after_unlock_lock(); /* GP memory ordering. */
		rnp->qsmaskinit &= ~mask;
2497
		rnp->qsmask &= ~mask;
2498 2499 2500 2501 2502 2503 2504 2505
		if (rnp->qsmaskinit) {
			raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */
			return;
		}
		raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */
	}
}

2506 2507 2508 2509 2510 2511 2512 2513 2514 2515 2516 2517 2518 2519 2520 2521 2522 2523 2524 2525
/*
 * The CPU is exiting the idle loop into the arch_cpu_idle_dead()
 * function.  We now remove it from the rcu_node tree's ->qsmaskinit
 * bit masks.
 */
static void rcu_cleanup_dying_idle_cpu(int cpu, struct rcu_state *rsp)
{
	unsigned long flags;
	unsigned long mask;
	struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
	struct rcu_node *rnp = rdp->mynode;  /* Outgoing CPU's rdp & rnp. */

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

2526
/*
2527
 * The CPU has been completely removed, and some other CPU is reporting
2528 2529
 * this fact from process context.  Do the remainder of the cleanup,
 * including orphaning the outgoing CPU's RCU callbacks, and also
2530 2531
 * 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.
2532
 */
2533
static void rcu_cleanup_dead_cpu(int cpu, struct rcu_state *rsp)
2534
{
2535
	unsigned long flags;
2536
	struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
2537
	struct rcu_node *rnp = rdp->mynode;  /* Outgoing CPU's rdp & rnp. */
2538

2539
	/* Adjust any no-longer-needed kthreads. */
T
Thomas Gleixner 已提交
2540
	rcu_boost_kthread_setaffinity(rnp, -1);
2541

2542
	/* Orphan the dead CPU's callbacks, and adopt them if appropriate. */
2543
	raw_spin_lock_irqsave(&rsp->orphan_lock, flags);
2544
	rcu_send_cbs_to_orphanage(cpu, rsp, rnp, rdp);
2545
	rcu_adopt_orphan_cbs(rsp, flags);
2546
	raw_spin_unlock_irqrestore(&rsp->orphan_lock, flags);
2547

2548 2549 2550
	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);
2551 2552 2553 2554
}

#else /* #ifdef CONFIG_HOTPLUG_CPU */

2555
static void rcu_cleanup_dying_cpu(struct rcu_state *rsp)
2556 2557 2558
{
}

2559 2560 2561 2562
static void __maybe_unused rcu_cleanup_dead_rnp(struct rcu_node *rnp_leaf)
{
}

2563 2564 2565 2566
static void rcu_cleanup_dying_idle_cpu(int cpu, struct rcu_state *rsp)
{
}

2567
static void rcu_cleanup_dead_cpu(int cpu, struct rcu_state *rsp)
2568 2569 2570 2571 2572 2573 2574 2575 2576
{
}

#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.
 */
2577
static void rcu_do_batch(struct rcu_state *rsp, struct rcu_data *rdp)
2578 2579 2580
{
	unsigned long flags;
	struct rcu_head *next, *list, **tail;
E
Eric Dumazet 已提交
2581 2582
	long bl, count, count_lazy;
	int i;
2583

2584
	/* If no callbacks are ready, just return. */
2585
	if (!cpu_has_callbacks_ready_to_invoke(rdp)) {
2586
		trace_rcu_batch_start(rsp->name, rdp->qlen_lazy, rdp->qlen, 0);
2587
		trace_rcu_batch_end(rsp->name, 0, !!READ_ONCE(rdp->nxtlist),
2588 2589
				    need_resched(), is_idle_task(current),
				    rcu_is_callbacks_kthread());
2590
		return;
2591
	}
2592 2593 2594 2595 2596 2597

	/*
	 * Extract the list of ready callbacks, disabling to prevent
	 * races with call_rcu() from interrupt handlers.
	 */
	local_irq_save(flags);
2598
	WARN_ON_ONCE(cpu_is_offline(smp_processor_id()));
2599
	bl = rdp->blimit;
2600
	trace_rcu_batch_start(rsp->name, rdp->qlen_lazy, rdp->qlen, bl);
2601 2602 2603 2604
	list = rdp->nxtlist;
	rdp->nxtlist = *rdp->nxttail[RCU_DONE_TAIL];
	*rdp->nxttail[RCU_DONE_TAIL] = NULL;
	tail = rdp->nxttail[RCU_DONE_TAIL];
2605 2606 2607
	for (i = RCU_NEXT_SIZE - 1; i >= 0; i--)
		if (rdp->nxttail[i] == rdp->nxttail[RCU_DONE_TAIL])
			rdp->nxttail[i] = &rdp->nxtlist;
2608 2609 2610
	local_irq_restore(flags);

	/* Invoke callbacks. */
2611
	count = count_lazy = 0;
2612 2613 2614
	while (list) {
		next = list->next;
		prefetch(next);
2615
		debug_rcu_head_unqueue(list);
2616 2617
		if (__rcu_reclaim(rsp->name, list))
			count_lazy++;
2618
		list = next;
2619 2620 2621 2622
		/* Stop only if limit reached and CPU has something to do. */
		if (++count >= bl &&
		    (need_resched() ||
		     (!is_idle_task(current) && !rcu_is_callbacks_kthread())))
2623 2624 2625 2626
			break;
	}

	local_irq_save(flags);
2627 2628 2629
	trace_rcu_batch_end(rsp->name, count, !!list, need_resched(),
			    is_idle_task(current),
			    rcu_is_callbacks_kthread());
2630 2631 2632 2633 2634

	/* Update count, and requeue any remaining callbacks. */
	if (list != NULL) {
		*tail = rdp->nxtlist;
		rdp->nxtlist = list;
2635 2636 2637
		for (i = 0; i < RCU_NEXT_SIZE; i++)
			if (&rdp->nxtlist == rdp->nxttail[i])
				rdp->nxttail[i] = tail;
2638 2639 2640
			else
				break;
	}
2641 2642
	smp_mb(); /* List handling before counting for rcu_barrier(). */
	rdp->qlen_lazy -= count_lazy;
2643
	WRITE_ONCE(rdp->qlen, rdp->qlen - count);
2644
	rdp->n_cbs_invoked += count;
2645 2646 2647 2648 2649

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

2650 2651 2652 2653 2654 2655
	/* 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;
2656
	WARN_ON_ONCE((rdp->nxtlist == NULL) != (rdp->qlen == 0));
2657

2658 2659
	local_irq_restore(flags);

2660
	/* Re-invoke RCU core processing if there are callbacks remaining. */
2661
	if (cpu_has_callbacks_ready_to_invoke(rdp))
2662
		invoke_rcu_core();
2663 2664 2665 2666 2667
}

/*
 * 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).
2668
 * Also schedule RCU core processing.
2669
 *
2670
 * This function must be called from hardirq context.  It is normally
2671 2672 2673
 * invoked from the scheduling-clock interrupt.  If rcu_pending returns
 * false, there is no point in invoking rcu_check_callbacks().
 */
2674
void rcu_check_callbacks(int user)
2675
{
2676
	trace_rcu_utilization(TPS("Start scheduler-tick"));
2677
	increment_cpu_stall_ticks();
2678
	if (user || rcu_is_cpu_rrupt_from_idle()) {
2679 2680 2681 2682 2683

		/*
		 * 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
2684
		 * a quiescent state, so note it.
2685 2686
		 *
		 * No memory barrier is required here because both
2687 2688 2689
		 * 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.
2690 2691
		 */

2692 2693
		rcu_sched_qs();
		rcu_bh_qs();
2694 2695 2696 2697 2698 2699 2700

	} 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
2701
		 * critical section, so note it.
2702 2703
		 */

2704
		rcu_bh_qs();
2705
	}
2706
	rcu_preempt_check_callbacks();
2707
	if (rcu_pending())
2708
		invoke_rcu_core();
P
Paul E. McKenney 已提交
2709 2710
	if (user)
		rcu_note_voluntary_context_switch(current);
2711
	trace_rcu_utilization(TPS("End scheduler-tick"));
2712 2713 2714 2715 2716
}

/*
 * Scan the leaf rcu_node structures, processing dyntick state for any that
 * have not yet encountered a quiescent state, using the function specified.
2717 2718
 * Also initiate boosting for any threads blocked on the root rcu_node.
 *
2719
 * The caller must have suppressed start of new grace periods.
2720
 */
2721 2722 2723 2724
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)
2725 2726 2727 2728 2729
{
	unsigned long bit;
	int cpu;
	unsigned long flags;
	unsigned long mask;
2730
	struct rcu_node *rnp;
2731

2732
	rcu_for_each_leaf_node(rsp, rnp) {
2733
		cond_resched_rcu_qs();
2734
		mask = 0;
P
Paul E. McKenney 已提交
2735
		raw_spin_lock_irqsave(&rnp->lock, flags);
2736
		smp_mb__after_unlock_lock();
2737
		if (!rcu_gp_in_progress(rsp)) {
P
Paul E. McKenney 已提交
2738
			raw_spin_unlock_irqrestore(&rnp->lock, flags);
2739
			return;
2740
		}
2741
		if (rnp->qsmask == 0) {
2742 2743 2744 2745 2746 2747 2748 2749 2750 2751 2752 2753 2754 2755 2756 2757 2758 2759 2760 2761 2762 2763 2764
			if (rcu_state_p == &rcu_sched_state ||
			    rsp != rcu_state_p ||
			    rcu_preempt_blocked_readers_cgp(rnp)) {
				/*
				 * No point in scanning bits because they
				 * are all zero.  But we might need to
				 * priority-boost blocked readers.
				 */
				rcu_initiate_boost(rnp, flags);
				/* rcu_initiate_boost() releases rnp->lock */
				continue;
			}
			if (rnp->parent &&
			    (rnp->parent->qsmask & rnp->grpmask)) {
				/*
				 * Race between grace-period
				 * initialization and task exiting RCU
				 * read-side critical section: Report.
				 */
				rcu_report_unblock_qs_rnp(rsp, rnp, flags);
				/* rcu_report_unblock_qs_rnp() rlses ->lock */
				continue;
			}
2765
		}
2766
		cpu = rnp->grplo;
2767
		bit = 1;
2768
		for (; cpu <= rnp->grphi; cpu++, bit <<= 1) {
2769
			if ((rnp->qsmask & bit) != 0) {
2770 2771
				if ((rnp->qsmaskinit & bit) == 0)
					*isidle = false; /* Pending hotplug. */
2772 2773 2774
				if (f(per_cpu_ptr(rsp->rda, cpu), isidle, maxj))
					mask |= bit;
			}
2775
		}
2776
		if (mask != 0) {
2777 2778
			/* Idle/offline CPUs, report (releases rnp->lock. */
			rcu_report_qs_rnp(mask, rsp, rnp, rnp->gpnum, flags);
2779 2780 2781
		} else {
			/* Nothing to do here, so just drop the lock. */
			raw_spin_unlock_irqrestore(&rnp->lock, flags);
2782 2783 2784 2785 2786 2787 2788 2789
		}
	}
}

/*
 * Force quiescent states on reluctant CPUs, and also detect which
 * CPUs are in dyntick-idle mode.
 */
2790
static void force_quiescent_state(struct rcu_state *rsp)
2791 2792
{
	unsigned long flags;
2793 2794 2795 2796 2797
	bool ret;
	struct rcu_node *rnp;
	struct rcu_node *rnp_old = NULL;

	/* Funnel through hierarchy to reduce memory contention. */
2798
	rnp = __this_cpu_read(rsp->rda->mynode);
2799
	for (; rnp != NULL; rnp = rnp->parent) {
2800
		ret = (READ_ONCE(rsp->gp_flags) & RCU_GP_FLAG_FQS) ||
2801 2802 2803 2804
		      !raw_spin_trylock(&rnp->fqslock);
		if (rnp_old != NULL)
			raw_spin_unlock(&rnp_old->fqslock);
		if (ret) {
2805
			rsp->n_force_qs_lh++;
2806 2807 2808 2809 2810
			return;
		}
		rnp_old = rnp;
	}
	/* rnp_old == rcu_get_root(rsp), rnp == NULL. */
2811

2812 2813
	/* Reached the root of the rcu_node tree, acquire lock. */
	raw_spin_lock_irqsave(&rnp_old->lock, flags);
2814
	smp_mb__after_unlock_lock();
2815
	raw_spin_unlock(&rnp_old->fqslock);
2816
	if (READ_ONCE(rsp->gp_flags) & RCU_GP_FLAG_FQS) {
2817
		rsp->n_force_qs_lh++;
2818
		raw_spin_unlock_irqrestore(&rnp_old->lock, flags);
2819
		return;  /* Someone beat us to it. */
2820
	}
2821
	WRITE_ONCE(rsp->gp_flags, READ_ONCE(rsp->gp_flags) | RCU_GP_FLAG_FQS);
2822
	raw_spin_unlock_irqrestore(&rnp_old->lock, flags);
2823
	rcu_gp_kthread_wake(rsp);
2824 2825 2826
}

/*
2827 2828 2829
 * 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.
2830 2831
 */
static void
2832
__rcu_process_callbacks(struct rcu_state *rsp)
2833 2834
{
	unsigned long flags;
2835
	bool needwake;
2836
	struct rcu_data *rdp = raw_cpu_ptr(rsp->rda);
2837

2838 2839
	WARN_ON_ONCE(rdp->beenonline == 0);

2840 2841 2842 2843
	/* 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? */
2844
	local_irq_save(flags);
2845
	if (cpu_needs_another_gp(rsp, rdp)) {
2846
		raw_spin_lock(&rcu_get_root(rsp)->lock); /* irqs disabled. */
2847
		needwake = rcu_start_gp(rsp);
2848
		raw_spin_unlock_irqrestore(&rcu_get_root(rsp)->lock, flags);
2849 2850
		if (needwake)
			rcu_gp_kthread_wake(rsp);
2851 2852
	} else {
		local_irq_restore(flags);
2853 2854 2855
	}

	/* If there are callbacks ready, invoke them. */
2856
	if (cpu_has_callbacks_ready_to_invoke(rdp))
2857
		invoke_rcu_callbacks(rsp, rdp);
2858 2859 2860

	/* Do any needed deferred wakeups of rcuo kthreads. */
	do_nocb_deferred_wakeup(rdp);
2861 2862
}

2863
/*
2864
 * Do RCU core processing for the current CPU.
2865
 */
2866
static void rcu_process_callbacks(struct softirq_action *unused)
2867
{
2868 2869
	struct rcu_state *rsp;

2870 2871
	if (cpu_is_offline(smp_processor_id()))
		return;
2872
	trace_rcu_utilization(TPS("Start RCU core"));
2873 2874
	for_each_rcu_flavor(rsp)
		__rcu_process_callbacks(rsp);
2875
	trace_rcu_utilization(TPS("End RCU core"));
2876 2877
}

2878
/*
2879 2880 2881
 * 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
2882
 * are running on the current CPU with softirqs disabled, the
2883
 * rcu_cpu_kthread_task cannot disappear out from under us.
2884
 */
2885
static void invoke_rcu_callbacks(struct rcu_state *rsp, struct rcu_data *rdp)
2886
{
2887
	if (unlikely(!READ_ONCE(rcu_scheduler_fully_active)))
2888
		return;
2889 2890
	if (likely(!rsp->boost)) {
		rcu_do_batch(rsp, rdp);
2891 2892
		return;
	}
2893
	invoke_rcu_callbacks_kthread();
2894 2895
}

2896
static void invoke_rcu_core(void)
2897
{
2898 2899
	if (cpu_online(smp_processor_id()))
		raise_softirq(RCU_SOFTIRQ);
2900 2901
}

2902 2903 2904 2905 2906
/*
 * 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)
2907
{
2908 2909
	bool needwake;

2910 2911 2912 2913
	/*
	 * If called from an extended quiescent state, invoke the RCU
	 * core in order to force a re-evaluation of RCU's idleness.
	 */
2914
	if (!rcu_is_watching())
2915 2916
		invoke_rcu_core();

2917
	/* If interrupts were disabled or CPU offline, don't invoke RCU core. */
2918
	if (irqs_disabled_flags(flags) || cpu_is_offline(smp_processor_id()))
2919
		return;
2920

2921 2922 2923 2924 2925 2926 2927
	/*
	 * 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.
	 */
2928
	if (unlikely(rdp->qlen > rdp->qlen_last_fqs_check + qhimark)) {
2929 2930

		/* Are we ignoring a completed grace period? */
2931
		note_gp_changes(rsp, rdp);
2932 2933 2934 2935 2936

		/* 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);

2937
			raw_spin_lock(&rnp_root->lock);
2938
			smp_mb__after_unlock_lock();
2939
			needwake = rcu_start_gp(rsp);
2940
			raw_spin_unlock(&rnp_root->lock);
2941 2942
			if (needwake)
				rcu_gp_kthread_wake(rsp);
2943 2944 2945 2946 2947
		} 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)
2948
				force_quiescent_state(rsp);
2949 2950 2951
			rdp->n_force_qs_snap = rsp->n_force_qs;
			rdp->qlen_last_fqs_check = rdp->qlen;
		}
2952
	}
2953 2954
}

2955 2956 2957 2958 2959 2960 2961
/*
 * RCU callback function to leak a callback.
 */
static void rcu_leak_callback(struct rcu_head *rhp)
{
}

P
Paul E. McKenney 已提交
2962 2963 2964 2965 2966 2967
/*
 * 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.
 */
2968 2969
static void
__call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu),
P
Paul E. McKenney 已提交
2970
	   struct rcu_state *rsp, int cpu, bool lazy)
2971 2972 2973 2974
{
	unsigned long flags;
	struct rcu_data *rdp;

2975
	WARN_ON_ONCE((unsigned long)head & 0x1); /* Misaligned rcu_head! */
2976 2977
	if (debug_rcu_head_queue(head)) {
		/* Probable double call_rcu(), so leak the callback. */
2978
		WRITE_ONCE(head->func, rcu_leak_callback);
2979 2980 2981
		WARN_ONCE(1, "__call_rcu(): Leaked duplicate callback\n");
		return;
	}
2982 2983 2984 2985 2986 2987 2988 2989 2990 2991
	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);
2992
	rdp = this_cpu_ptr(rsp->rda);
2993 2994

	/* Add the callback to our list. */
P
Paul E. McKenney 已提交
2995 2996 2997 2998 2999
	if (unlikely(rdp->nxttail[RCU_NEXT_TAIL] == NULL) || cpu != -1) {
		int offline;

		if (cpu != -1)
			rdp = per_cpu_ptr(rsp->rda, cpu);
3000 3001 3002 3003 3004 3005 3006 3007 3008 3009 3010 3011 3012
		if (likely(rdp->mynode)) {
			/* Post-boot, so this should be for a no-CBs CPU. */
			offline = !__call_rcu_nocb(rdp, head, lazy, flags);
			WARN_ON_ONCE(offline);
			/* Offline CPU, _call_rcu() illegal, leak callback.  */
			local_irq_restore(flags);
			return;
		}
		/*
		 * Very early boot, before rcu_init().  Initialize if needed
		 * and then drop through to queue the callback.
		 */
		BUG_ON(cpu != -1);
3013
		WARN_ON_ONCE(!rcu_is_watching());
3014 3015
		if (!likely(rdp->nxtlist))
			init_default_callback_list(rdp);
3016
	}
3017
	WRITE_ONCE(rdp->qlen, rdp->qlen + 1);
3018 3019
	if (lazy)
		rdp->qlen_lazy++;
3020 3021
	else
		rcu_idle_count_callbacks_posted();
3022 3023 3024
	smp_mb();  /* Count before adding callback for rcu_barrier(). */
	*rdp->nxttail[RCU_NEXT_TAIL] = head;
	rdp->nxttail[RCU_NEXT_TAIL] = &head->next;
3025

3026 3027
	if (__is_kfree_rcu_offset((unsigned long)func))
		trace_rcu_kfree_callback(rsp->name, head, (unsigned long)func,
3028
					 rdp->qlen_lazy, rdp->qlen);
3029
	else
3030
		trace_rcu_callback(rsp->name, head, rdp->qlen_lazy, rdp->qlen);
3031

3032 3033
	/* Go handle any RCU core processing required. */
	__call_rcu_core(rsp, rdp, head, flags);
3034 3035 3036 3037
	local_irq_restore(flags);
}

/*
3038
 * Queue an RCU-sched callback for invocation after a grace period.
3039
 */
3040
void call_rcu_sched(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
3041
{
P
Paul E. McKenney 已提交
3042
	__call_rcu(head, func, &rcu_sched_state, -1, 0);
3043
}
3044
EXPORT_SYMBOL_GPL(call_rcu_sched);
3045 3046

/*
3047
 * Queue an RCU callback for invocation after a quicker grace period.
3048 3049 3050
 */
void call_rcu_bh(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
{
P
Paul E. McKenney 已提交
3051
	__call_rcu(head, func, &rcu_bh_state, -1, 0);
3052 3053 3054
}
EXPORT_SYMBOL_GPL(call_rcu_bh);

3055 3056 3057 3058 3059 3060 3061 3062 3063 3064
/*
 * 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))
{
3065
	__call_rcu(head, func, rcu_state_p, -1, 1);
3066 3067 3068
}
EXPORT_SYMBOL_GPL(kfree_call_rcu);

3069 3070 3071 3072 3073 3074 3075 3076 3077 3078 3079
/*
 * 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)
{
3080 3081
	int ret;

3082
	might_sleep();  /* Check for RCU read-side critical section. */
3083 3084 3085 3086
	preempt_disable();
	ret = num_online_cpus() <= 1;
	preempt_enable();
	return ret;
3087 3088
}

3089 3090 3091 3092 3093 3094 3095 3096 3097 3098 3099 3100
/**
 * 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
3101 3102 3103 3104 3105 3106 3107 3108 3109 3110 3111 3112 3113 3114 3115 3116 3117 3118 3119 3120 3121 3122
 * 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).
3123 3124 3125 3126 3127 3128 3129 3130 3131
 *
 * 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)
{
3132 3133 3134 3135
	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");
3136 3137
	if (rcu_blocking_is_gp())
		return;
3138
	if (rcu_gp_is_expedited())
3139 3140 3141
		synchronize_sched_expedited();
	else
		wait_rcu_gp(call_rcu_sched);
3142 3143 3144 3145 3146 3147 3148 3149 3150 3151 3152
}
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.
3153 3154 3155
 *
 * See the description of synchronize_sched() for more detailed information
 * on memory ordering guarantees.
3156 3157 3158
 */
void synchronize_rcu_bh(void)
{
3159 3160 3161 3162
	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");
3163 3164
	if (rcu_blocking_is_gp())
		return;
3165
	if (rcu_gp_is_expedited())
3166 3167 3168
		synchronize_rcu_bh_expedited();
	else
		wait_rcu_gp(call_rcu_bh);
3169 3170 3171
}
EXPORT_SYMBOL_GPL(synchronize_rcu_bh);

3172 3173 3174 3175 3176 3177 3178 3179 3180 3181 3182 3183 3184 3185 3186 3187 3188 3189 3190 3191
/**
 * 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().
	 */
3192
	return smp_load_acquire(&rcu_state_p->gpnum);
3193 3194 3195 3196 3197 3198 3199 3200 3201 3202 3203 3204 3205 3206 3207 3208 3209 3210 3211 3212 3213 3214 3215 3216 3217
}
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.
	 */
3218
	newstate = smp_load_acquire(&rcu_state_p->completed);
3219 3220 3221 3222 3223
	if (ULONG_CMP_GE(oldstate, newstate))
		synchronize_rcu();
}
EXPORT_SYMBOL_GPL(cond_synchronize_rcu);

3224 3225 3226 3227 3228 3229 3230 3231 3232 3233 3234 3235 3236 3237 3238 3239 3240
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;
}

3241 3242 3243 3244 3245 3246 3247 3248 3249 3250
/**
 * 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.
3251 3252 3253 3254 3255 3256 3257 3258 3259 3260 3261 3262 3263 3264 3265 3266 3267 3268 3269 3270 3271 3272 3273 3274
 *
 * 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)
{
3275 3276 3277
	cpumask_var_t cm;
	bool cma = false;
	int cpu;
3278 3279
	long firstsnap, s, snap;
	int trycount = 0;
3280
	struct rcu_state *rsp = &rcu_sched_state;
3281

3282 3283 3284 3285 3286 3287 3288 3289
	/*
	 * 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.
	 */
3290 3291
	if (ULONG_CMP_GE((ulong)atomic_long_read(&rsp->expedited_start),
			 (ulong)atomic_long_read(&rsp->expedited_done) +
3292 3293
			 ULONG_MAX / 8)) {
		synchronize_sched();
3294
		atomic_long_inc(&rsp->expedited_wrap);
3295 3296
		return;
	}
3297

3298 3299 3300 3301
	/*
	 * Take a ticket.  Note that atomic_inc_return() implies a
	 * full memory barrier.
	 */
3302
	snap = atomic_long_inc_return(&rsp->expedited_start);
3303
	firstsnap = snap;
3304 3305 3306 3307 3308 3309
	if (!try_get_online_cpus()) {
		/* CPU hotplug operation in flight, fall back to normal GP. */
		wait_rcu_gp(call_rcu_sched);
		atomic_long_inc(&rsp->expedited_normal);
		return;
	}
3310
	WARN_ON_ONCE(cpu_is_offline(raw_smp_processor_id()));
3311

3312 3313 3314 3315 3316 3317 3318 3319 3320 3321 3322 3323 3324 3325 3326
	/* Offline CPUs, idle CPUs, and any CPU we run on are quiescent. */
	cma = zalloc_cpumask_var(&cm, GFP_KERNEL);
	if (cma) {
		cpumask_copy(cm, cpu_online_mask);
		cpumask_clear_cpu(raw_smp_processor_id(), cm);
		for_each_cpu(cpu, cm) {
			struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu);

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

3327 3328 3329 3330
	/*
	 * Each pass through the following loop attempts to force a
	 * context switch on each CPU.
	 */
3331
	while (try_stop_cpus(cma ? cm : cpu_online_mask,
3332 3333 3334
			     synchronize_sched_expedited_cpu_stop,
			     NULL) == -EAGAIN) {
		put_online_cpus();
3335
		atomic_long_inc(&rsp->expedited_tryfail);
3336

3337
		/* Check to see if someone else did our work for us. */
3338
		s = atomic_long_read(&rsp->expedited_done);
3339
		if (ULONG_CMP_GE((ulong)s, (ulong)firstsnap)) {
3340
			/* ensure test happens before caller kfree */
3341
			smp_mb__before_atomic(); /* ^^^ */
3342
			atomic_long_inc(&rsp->expedited_workdone1);
3343
			free_cpumask_var(cm);
3344 3345
			return;
		}
3346 3347

		/* No joy, try again later.  Or just synchronize_sched(). */
3348
		if (trycount++ < 10) {
3349
			udelay(trycount * num_online_cpus());
3350
		} else {
3351
			wait_rcu_gp(call_rcu_sched);
3352
			atomic_long_inc(&rsp->expedited_normal);
3353
			free_cpumask_var(cm);
3354 3355 3356
			return;
		}

3357
		/* Recheck to see if someone else did our work for us. */
3358
		s = atomic_long_read(&rsp->expedited_done);
3359
		if (ULONG_CMP_GE((ulong)s, (ulong)firstsnap)) {
3360
			/* ensure test happens before caller kfree */
3361
			smp_mb__before_atomic(); /* ^^^ */
3362
			atomic_long_inc(&rsp->expedited_workdone2);
3363
			free_cpumask_var(cm);
3364 3365 3366 3367 3368
			return;
		}

		/*
		 * Refetching sync_sched_expedited_started allows later
3369 3370 3371 3372
		 * 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.
3373
		 */
3374 3375 3376 3377
		if (!try_get_online_cpus()) {
			/* CPU hotplug operation in flight, use normal GP. */
			wait_rcu_gp(call_rcu_sched);
			atomic_long_inc(&rsp->expedited_normal);
3378
			free_cpumask_var(cm);
3379 3380
			return;
		}
3381
		snap = atomic_long_read(&rsp->expedited_start);
3382 3383
		smp_mb(); /* ensure read is before try_stop_cpus(). */
	}
3384
	atomic_long_inc(&rsp->expedited_stoppedcpus);
3385

3386 3387 3388
all_cpus_idle:
	free_cpumask_var(cm);

3389 3390 3391 3392
	/*
	 * 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
3393
	 * than we did already did their update.
3394 3395
	 */
	do {
3396
		atomic_long_inc(&rsp->expedited_done_tries);
3397
		s = atomic_long_read(&rsp->expedited_done);
3398
		if (ULONG_CMP_GE((ulong)s, (ulong)snap)) {
3399
			/* ensure test happens before caller kfree */
3400
			smp_mb__before_atomic(); /* ^^^ */
3401
			atomic_long_inc(&rsp->expedited_done_lost);
3402 3403
			break;
		}
3404
	} while (atomic_long_cmpxchg(&rsp->expedited_done, s, snap) != s);
3405
	atomic_long_inc(&rsp->expedited_done_exit);
3406 3407 3408 3409 3410

	put_online_cpus();
}
EXPORT_SYMBOL_GPL(synchronize_sched_expedited);

3411 3412 3413 3414 3415 3416 3417 3418 3419
/*
 * 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)
{
3420 3421
	struct rcu_node *rnp = rdp->mynode;

3422 3423 3424 3425 3426
	rdp->n_rcu_pending++;

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

3427 3428 3429 3430
	/* Is this CPU a NO_HZ_FULL CPU that should ignore RCU? */
	if (rcu_nohz_full_cpu(rsp))
		return 0;

3431
	/* Is the RCU core waiting for a quiescent state from this CPU? */
3432
	if (rcu_scheduler_fully_active &&
3433 3434
	    rdp->qs_pending && !rdp->passed_quiesce &&
	    rdp->rcu_qs_ctr_snap == __this_cpu_read(rcu_qs_ctr)) {
3435
		rdp->n_rp_qs_pending++;
3436 3437 3438
	} else if (rdp->qs_pending &&
		   (rdp->passed_quiesce ||
		    rdp->rcu_qs_ctr_snap != __this_cpu_read(rcu_qs_ctr))) {
3439
		rdp->n_rp_report_qs++;
3440
		return 1;
3441
	}
3442 3443

	/* Does this CPU have callbacks ready to invoke? */
3444 3445
	if (cpu_has_callbacks_ready_to_invoke(rdp)) {
		rdp->n_rp_cb_ready++;
3446
		return 1;
3447
	}
3448 3449

	/* Has RCU gone idle with this CPU needing another grace period? */
3450 3451
	if (cpu_needs_another_gp(rsp, rdp)) {
		rdp->n_rp_cpu_needs_gp++;
3452
		return 1;
3453
	}
3454 3455

	/* Has another RCU grace period completed?  */
3456
	if (READ_ONCE(rnp->completed) != rdp->completed) { /* outside lock */
3457
		rdp->n_rp_gp_completed++;
3458
		return 1;
3459
	}
3460 3461

	/* Has a new RCU grace period started? */
3462 3463
	if (READ_ONCE(rnp->gpnum) != rdp->gpnum ||
	    unlikely(READ_ONCE(rdp->gpwrap))) { /* outside lock */
3464
		rdp->n_rp_gp_started++;
3465
		return 1;
3466
	}
3467

3468 3469 3470 3471 3472 3473
	/* Does this CPU need a deferred NOCB wakeup? */
	if (rcu_nocb_need_deferred_wakeup(rdp)) {
		rdp->n_rp_nocb_defer_wakeup++;
		return 1;
	}

3474
	/* nothing to do */
3475
	rdp->n_rp_need_nothing++;
3476 3477 3478 3479 3480 3481 3482 3483
	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.
 */
3484
static int rcu_pending(void)
3485
{
3486 3487 3488
	struct rcu_state *rsp;

	for_each_rcu_flavor(rsp)
3489
		if (__rcu_pending(rsp, this_cpu_ptr(rsp->rda)))
3490 3491
			return 1;
	return 0;
3492 3493 3494
}

/*
3495 3496 3497
 * 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.)
3498
 */
3499
static int __maybe_unused rcu_cpu_has_callbacks(bool *all_lazy)
3500
{
3501 3502 3503
	bool al = true;
	bool hc = false;
	struct rcu_data *rdp;
3504 3505
	struct rcu_state *rsp;

3506
	for_each_rcu_flavor(rsp) {
3507
		rdp = this_cpu_ptr(rsp->rda);
3508 3509 3510 3511
		if (!rdp->nxtlist)
			continue;
		hc = true;
		if (rdp->qlen != rdp->qlen_lazy || !all_lazy) {
3512
			al = false;
3513 3514
			break;
		}
3515 3516 3517 3518
	}
	if (all_lazy)
		*all_lazy = al;
	return hc;
3519 3520
}

3521 3522 3523 3524
/*
 * Helper function for _rcu_barrier() tracing.  If tracing is disabled,
 * the compiler is expected to optimize this away.
 */
3525
static void _rcu_barrier_trace(struct rcu_state *rsp, const char *s,
3526 3527 3528 3529 3530 3531
			       int cpu, unsigned long done)
{
	trace_rcu_barrier(rsp->name, s, cpu,
			  atomic_read(&rsp->barrier_cpu_count), done);
}

3532 3533 3534 3535
/*
 * RCU callback function for _rcu_barrier().  If we are last, wake
 * up the task executing _rcu_barrier().
 */
3536
static void rcu_barrier_callback(struct rcu_head *rhp)
3537
{
3538 3539 3540
	struct rcu_data *rdp = container_of(rhp, struct rcu_data, barrier_head);
	struct rcu_state *rsp = rdp->rsp;

3541 3542
	if (atomic_dec_and_test(&rsp->barrier_cpu_count)) {
		_rcu_barrier_trace(rsp, "LastCB", -1, rsp->n_barrier_done);
3543
		complete(&rsp->barrier_completion);
3544 3545 3546
	} else {
		_rcu_barrier_trace(rsp, "CB", -1, rsp->n_barrier_done);
	}
3547 3548 3549 3550 3551 3552 3553
}

/*
 * Called with preemption disabled, and from cross-cpu IRQ context.
 */
static void rcu_barrier_func(void *type)
{
3554
	struct rcu_state *rsp = type;
3555
	struct rcu_data *rdp = raw_cpu_ptr(rsp->rda);
3556

3557
	_rcu_barrier_trace(rsp, "IRQ", -1, rsp->n_barrier_done);
3558
	atomic_inc(&rsp->barrier_cpu_count);
3559
	rsp->call(&rdp->barrier_head, rcu_barrier_callback);
3560 3561 3562 3563 3564 3565
}

/*
 * Orchestrate the specified type of RCU barrier, waiting for all
 * RCU callbacks of the specified type to complete.
 */
3566
static void _rcu_barrier(struct rcu_state *rsp)
3567
{
3568 3569
	int cpu;
	struct rcu_data *rdp;
3570
	unsigned long snap = READ_ONCE(rsp->n_barrier_done);
3571
	unsigned long snap_done;
3572

3573
	_rcu_barrier_trace(rsp, "Begin", -1, snap);
3574

3575
	/* Take mutex to serialize concurrent rcu_barrier() requests. */
3576
	mutex_lock(&rsp->barrier_mutex);
3577

3578 3579 3580 3581 3582 3583 3584 3585 3586 3587 3588 3589
	/*
	 * 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.
	 */
3590
	snap_done = rsp->n_barrier_done;
3591
	_rcu_barrier_trace(rsp, "Check", -1, snap_done);
3592 3593 3594 3595 3596 3597 3598 3599 3600 3601 3602 3603

	/*
	 * 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)) {
3604
		_rcu_barrier_trace(rsp, "EarlyExit", -1, snap_done);
3605 3606 3607 3608 3609 3610 3611
		smp_mb(); /* caller's subsequent code after above check. */
		mutex_unlock(&rsp->barrier_mutex);
		return;
	}

	/*
	 * Increment ->n_barrier_done to avoid duplicate work.  Use
3612
	 * WRITE_ONCE() to prevent the compiler from speculating
3613 3614
	 * the increment to precede the early-exit check.
	 */
3615
	WRITE_ONCE(rsp->n_barrier_done, rsp->n_barrier_done + 1);
3616
	WARN_ON_ONCE((rsp->n_barrier_done & 0x1) != 1);
3617
	_rcu_barrier_trace(rsp, "Inc1", -1, rsp->n_barrier_done);
3618
	smp_mb(); /* Order ->n_barrier_done increment with below mechanism. */
3619

3620
	/*
3621 3622
	 * 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
3623 3624
	 * (or preemption of this task).  Exclude CPU-hotplug operations
	 * to ensure that no offline CPU has callbacks queued.
3625
	 */
3626
	init_completion(&rsp->barrier_completion);
3627
	atomic_set(&rsp->barrier_cpu_count, 1);
3628
	get_online_cpus();
3629 3630

	/*
3631 3632 3633
	 * 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.
3634
	 */
P
Paul E. McKenney 已提交
3635
	for_each_possible_cpu(cpu) {
3636
		if (!cpu_online(cpu) && !rcu_is_nocb_cpu(cpu))
P
Paul E. McKenney 已提交
3637
			continue;
3638
		rdp = per_cpu_ptr(rsp->rda, cpu);
3639
		if (rcu_is_nocb_cpu(cpu)) {
3640 3641 3642 3643 3644 3645
			if (!rcu_nocb_cpu_needs_barrier(rsp, cpu)) {
				_rcu_barrier_trace(rsp, "OfflineNoCB", cpu,
						   rsp->n_barrier_done);
			} else {
				_rcu_barrier_trace(rsp, "OnlineNoCB", cpu,
						   rsp->n_barrier_done);
3646
				smp_mb__before_atomic();
3647 3648 3649 3650
				atomic_inc(&rsp->barrier_cpu_count);
				__call_rcu(&rdp->barrier_head,
					   rcu_barrier_callback, rsp, cpu, 0);
			}
3651
		} else if (READ_ONCE(rdp->qlen)) {
3652 3653
			_rcu_barrier_trace(rsp, "OnlineQ", cpu,
					   rsp->n_barrier_done);
3654
			smp_call_function_single(cpu, rcu_barrier_func, rsp, 1);
3655
		} else {
3656 3657
			_rcu_barrier_trace(rsp, "OnlineNQ", cpu,
					   rsp->n_barrier_done);
3658 3659
		}
	}
3660
	put_online_cpus();
3661 3662 3663 3664 3665

	/*
	 * Now that we have an rcu_barrier_callback() callback on each
	 * CPU, and thus each counted, remove the initial count.
	 */
3666
	if (atomic_dec_and_test(&rsp->barrier_cpu_count))
3667
		complete(&rsp->barrier_completion);
3668

3669 3670
	/* Increment ->n_barrier_done to prevent duplicate work. */
	smp_mb(); /* Keep increment after above mechanism. */
3671
	WRITE_ONCE(rsp->n_barrier_done, rsp->n_barrier_done + 1);
3672
	WARN_ON_ONCE((rsp->n_barrier_done & 0x1) != 0);
3673
	_rcu_barrier_trace(rsp, "Inc2", -1, rsp->n_barrier_done);
3674 3675
	smp_mb(); /* Keep increment before caller's subsequent code. */

3676
	/* Wait for all rcu_barrier_callback() callbacks to be invoked. */
3677
	wait_for_completion(&rsp->barrier_completion);
3678 3679

	/* Other rcu_barrier() invocations can now safely proceed. */
3680
	mutex_unlock(&rsp->barrier_mutex);
3681 3682 3683 3684 3685 3686 3687
}

/**
 * rcu_barrier_bh - Wait until all in-flight call_rcu_bh() callbacks complete.
 */
void rcu_barrier_bh(void)
{
3688
	_rcu_barrier(&rcu_bh_state);
3689 3690 3691 3692 3693 3694 3695 3696
}
EXPORT_SYMBOL_GPL(rcu_barrier_bh);

/**
 * rcu_barrier_sched - Wait for in-flight call_rcu_sched() callbacks.
 */
void rcu_barrier_sched(void)
{
3697
	_rcu_barrier(&rcu_sched_state);
3698 3699 3700
}
EXPORT_SYMBOL_GPL(rcu_barrier_sched);

3701 3702 3703 3704 3705 3706 3707 3708 3709 3710 3711 3712 3713 3714 3715 3716 3717 3718 3719 3720 3721 3722
/*
 * Propagate ->qsinitmask bits up the rcu_node tree to account for the
 * first CPU in a given leaf rcu_node structure coming online.  The caller
 * must hold the corresponding leaf rcu_node ->lock with interrrupts
 * disabled.
 */
static void rcu_init_new_rnp(struct rcu_node *rnp_leaf)
{
	long mask;
	struct rcu_node *rnp = rnp_leaf;

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

3723
/*
3724
 * Do boot-time initialization of a CPU's per-CPU RCU data.
3725
 */
3726 3727
static void __init
rcu_boot_init_percpu_data(int cpu, struct rcu_state *rsp)
3728 3729
{
	unsigned long flags;
3730
	struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
3731 3732 3733
	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
3734
	raw_spin_lock_irqsave(&rnp->lock, flags);
3735 3736
	rdp->grpmask = 1UL << (cpu - rdp->mynode->grplo);
	rdp->dynticks = &per_cpu(rcu_dynticks, cpu);
3737
	WARN_ON_ONCE(rdp->dynticks->dynticks_nesting != DYNTICK_TASK_EXIT_IDLE);
3738
	WARN_ON_ONCE(atomic_read(&rdp->dynticks->dynticks) != 1);
3739
	rdp->cpu = cpu;
3740
	rdp->rsp = rsp;
P
Paul E. McKenney 已提交
3741
	rcu_boot_init_nocb_percpu_data(rdp);
P
Paul E. McKenney 已提交
3742
	raw_spin_unlock_irqrestore(&rnp->lock, flags);
3743 3744 3745 3746 3747 3748 3749
}

/*
 * 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.
3750
 */
3751
static void
3752
rcu_init_percpu_data(int cpu, struct rcu_state *rsp)
3753 3754 3755
{
	unsigned long flags;
	unsigned long mask;
3756
	struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
3757 3758 3759
	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
3760
	raw_spin_lock_irqsave(&rnp->lock, flags);
3761
	rdp->beenonline = 1;	 /* We have now been online. */
3762 3763
	rdp->qlen_last_fqs_check = 0;
	rdp->n_force_qs_snap = rsp->n_force_qs;
3764
	rdp->blimit = blimit;
3765 3766
	if (!rdp->nxtlist)
		init_callback_list(rdp);  /* Re-enable callbacks on this CPU. */
3767
	rdp->dynticks->dynticks_nesting = DYNTICK_TASK_EXIT_IDLE;
3768
	rcu_sysidle_init_percpu_data(rdp->dynticks);
3769 3770
	atomic_set(&rdp->dynticks->dynticks,
		   (atomic_read(&rdp->dynticks->dynticks) & ~0x1) + 1);
P
Paul E. McKenney 已提交
3771
	raw_spin_unlock(&rnp->lock);		/* irqs remain disabled. */
3772

3773 3774 3775 3776 3777
	/*
	 * Add CPU to leaf rcu_node pending-online bitmask.  Any needed
	 * propagation up the rcu_node tree will happen at the beginning
	 * of the next grace period.
	 */
3778 3779
	rnp = rdp->mynode;
	mask = rdp->grpmask;
3780 3781 3782 3783 3784 3785 3786 3787 3788 3789
	raw_spin_lock(&rnp->lock);		/* irqs already disabled. */
	smp_mb__after_unlock_lock();
	rnp->qsmaskinitnext |= mask;
	rdp->gpnum = rnp->completed; /* Make CPU later note any new GP. */
	rdp->completed = rnp->completed;
	rdp->passed_quiesce = false;
	rdp->rcu_qs_ctr_snap = __this_cpu_read(rcu_qs_ctr);
	rdp->qs_pending = false;
	trace_rcu_grace_period(rsp->name, rdp->gpnum, TPS("cpuonl"));
	raw_spin_unlock_irqrestore(&rnp->lock, flags);
3790 3791
}

3792
static void rcu_prepare_cpu(int cpu)
3793
{
3794 3795 3796
	struct rcu_state *rsp;

	for_each_rcu_flavor(rsp)
3797
		rcu_init_percpu_data(cpu, rsp);
3798 3799 3800
}

/*
3801
 * Handle CPU online/offline notification events.
3802
 */
3803 3804
int rcu_cpu_notify(struct notifier_block *self,
		   unsigned long action, void *hcpu)
3805 3806
{
	long cpu = (long)hcpu;
3807
	struct rcu_data *rdp = per_cpu_ptr(rcu_state_p->rda, cpu);
3808
	struct rcu_node *rnp = rdp->mynode;
3809
	struct rcu_state *rsp;
3810 3811 3812 3813

	switch (action) {
	case CPU_UP_PREPARE:
	case CPU_UP_PREPARE_FROZEN:
P
Peter Zijlstra 已提交
3814 3815
		rcu_prepare_cpu(cpu);
		rcu_prepare_kthreads(cpu);
3816
		rcu_spawn_all_nocb_kthreads(cpu);
3817 3818
		break;
	case CPU_ONLINE:
3819
	case CPU_DOWN_FAILED:
T
Thomas Gleixner 已提交
3820
		rcu_boost_kthread_setaffinity(rnp, -1);
3821 3822
		break;
	case CPU_DOWN_PREPARE:
3823
		rcu_boost_kthread_setaffinity(rnp, cpu);
3824
		break;
3825 3826
	case CPU_DYING:
	case CPU_DYING_FROZEN:
3827 3828
		for_each_rcu_flavor(rsp)
			rcu_cleanup_dying_cpu(rsp);
3829
		break;
3830 3831 3832 3833 3834
	case CPU_DYING_IDLE:
		for_each_rcu_flavor(rsp) {
			rcu_cleanup_dying_idle_cpu(cpu, rsp);
		}
		break;
3835 3836 3837 3838
	case CPU_DEAD:
	case CPU_DEAD_FROZEN:
	case CPU_UP_CANCELED:
	case CPU_UP_CANCELED_FROZEN:
3839
		for_each_rcu_flavor(rsp) {
3840
			rcu_cleanup_dead_cpu(cpu, rsp);
3841 3842
			do_nocb_deferred_wakeup(per_cpu_ptr(rsp->rda, cpu));
		}
3843 3844 3845 3846
		break;
	default:
		break;
	}
3847
	return NOTIFY_OK;
3848 3849
}

3850 3851 3852 3853 3854 3855 3856
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. */
3857
			rcu_expedite_gp();
3858 3859 3860
		break;
	case PM_POST_HIBERNATION:
	case PM_POST_SUSPEND:
3861 3862
		if (nr_cpu_ids <= 256) /* Expediting bad for large systems. */
			rcu_unexpedite_gp();
3863 3864 3865 3866 3867 3868 3869
		break;
	default:
		break;
	}
	return NOTIFY_OK;
}

3870
/*
3871
 * Spawn the kthreads that handle each RCU flavor's grace periods.
3872 3873 3874 3875
 */
static int __init rcu_spawn_gp_kthread(void)
{
	unsigned long flags;
3876
	int kthread_prio_in = kthread_prio;
3877 3878
	struct rcu_node *rnp;
	struct rcu_state *rsp;
3879
	struct sched_param sp;
3880 3881
	struct task_struct *t;

3882 3883 3884 3885 3886 3887 3888 3889 3890 3891 3892
	/* Force priority into range. */
	if (IS_ENABLED(CONFIG_RCU_BOOST) && kthread_prio < 1)
		kthread_prio = 1;
	else if (kthread_prio < 0)
		kthread_prio = 0;
	else if (kthread_prio > 99)
		kthread_prio = 99;
	if (kthread_prio != kthread_prio_in)
		pr_alert("rcu_spawn_gp_kthread(): Limited prio to %d from %d\n",
			 kthread_prio, kthread_prio_in);

3893
	rcu_scheduler_fully_active = 1;
3894
	for_each_rcu_flavor(rsp) {
3895
		t = kthread_create(rcu_gp_kthread, rsp, "%s", rsp->name);
3896 3897 3898 3899
		BUG_ON(IS_ERR(t));
		rnp = rcu_get_root(rsp);
		raw_spin_lock_irqsave(&rnp->lock, flags);
		rsp->gp_kthread = t;
3900 3901 3902 3903 3904
		if (kthread_prio) {
			sp.sched_priority = kthread_prio;
			sched_setscheduler_nocheck(t, SCHED_FIFO, &sp);
		}
		wake_up_process(t);
3905 3906
		raw_spin_unlock_irqrestore(&rnp->lock, flags);
	}
3907
	rcu_spawn_nocb_kthreads();
3908
	rcu_spawn_boost_kthreads();
3909 3910 3911 3912
	return 0;
}
early_initcall(rcu_spawn_gp_kthread);

3913 3914 3915 3916 3917 3918 3919 3920 3921 3922 3923 3924 3925 3926 3927
/*
 * 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;
}

3928 3929 3930 3931 3932 3933 3934 3935
/*
 * Compute the per-level fanout, either using the exact fanout specified
 * or balancing the tree, depending on CONFIG_RCU_FANOUT_EXACT.
 */
static void __init rcu_init_levelspread(struct rcu_state *rsp)
{
	int i;

3936 3937 3938 3939 3940 3941 3942 3943 3944 3945 3946 3947 3948 3949
	if (IS_ENABLED(CONFIG_RCU_FANOUT_EXACT)) {
		rsp->levelspread[rcu_num_lvls - 1] = rcu_fanout_leaf;
		for (i = rcu_num_lvls - 2; i >= 0; i--)
			rsp->levelspread[i] = CONFIG_RCU_FANOUT;
	} else {
		int ccur;
		int cprv;

		cprv = nr_cpu_ids;
		for (i = rcu_num_lvls - 1; i >= 0; i--) {
			ccur = rsp->levelcnt[i];
			rsp->levelspread[i] = (cprv + ccur - 1) / ccur;
			cprv = ccur;
		}
3950 3951 3952 3953 3954 3955
	}
}

/*
 * Helper function for rcu_init() that initializes one rcu_state structure.
 */
3956 3957
static void __init rcu_init_one(struct rcu_state *rsp,
		struct rcu_data __percpu *rda)
3958
{
3959 3960 3961 3962 3963 3964 3965 3966 3967 3968
	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 */
3969
	static u8 fl_mask = 0x1;
3970 3971 3972 3973 3974
	int cpustride = 1;
	int i;
	int j;
	struct rcu_node *rnp;

3975 3976
	BUILD_BUG_ON(MAX_RCU_LVLS > ARRAY_SIZE(buf));  /* Fix buf[] init! */

3977 3978 3979 3980
	/* 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");

3981 3982
	/* Initialize the level-tracking arrays. */

3983 3984 3985
	for (i = 0; i < rcu_num_lvls; i++)
		rsp->levelcnt[i] = num_rcu_lvl[i];
	for (i = 1; i < rcu_num_lvls; i++)
3986 3987
		rsp->level[i] = rsp->level[i - 1] + rsp->levelcnt[i - 1];
	rcu_init_levelspread(rsp);
3988 3989
	rsp->flavor_mask = fl_mask;
	fl_mask <<= 1;
3990 3991 3992

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

3993
	for (i = rcu_num_lvls - 1; i >= 0; i--) {
3994 3995 3996
		cpustride *= rsp->levelspread[i];
		rnp = rsp->level[i];
		for (j = 0; j < rsp->levelcnt[i]; j++, rnp++) {
P
Paul E. McKenney 已提交
3997
			raw_spin_lock_init(&rnp->lock);
3998 3999
			lockdep_set_class_and_name(&rnp->lock,
						   &rcu_node_class[i], buf[i]);
4000 4001 4002
			raw_spin_lock_init(&rnp->fqslock);
			lockdep_set_class_and_name(&rnp->fqslock,
						   &rcu_fqs_class[i], fqs[i]);
4003 4004
			rnp->gpnum = rsp->gpnum;
			rnp->completed = rsp->completed;
4005 4006 4007 4008
			rnp->qsmask = 0;
			rnp->qsmaskinit = 0;
			rnp->grplo = j * cpustride;
			rnp->grphi = (j + 1) * cpustride - 1;
4009 4010
			if (rnp->grphi >= nr_cpu_ids)
				rnp->grphi = nr_cpu_ids - 1;
4011 4012 4013 4014 4015 4016 4017 4018 4019 4020 4021
			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;
4022
			INIT_LIST_HEAD(&rnp->blkd_tasks);
4023
			rcu_init_one_nocb(rnp);
4024 4025
		}
	}
4026

4027
	init_waitqueue_head(&rsp->gp_wq);
4028
	rnp = rsp->level[rcu_num_lvls - 1];
4029
	for_each_possible_cpu(i) {
4030
		while (i > rnp->grphi)
4031
			rnp++;
4032
		per_cpu_ptr(rsp->rda, i)->mynode = rnp;
4033 4034
		rcu_boot_init_percpu_data(i, rsp);
	}
4035
	list_add(&rsp->flavors, &rcu_struct_flavors);
4036 4037
}

4038 4039
/*
 * Compute the rcu_node tree geometry from kernel parameters.  This cannot
4040
 * replace the definitions in tree.h because those are needed to size
4041 4042 4043 4044
 * the ->node array in the rcu_state structure.
 */
static void __init rcu_init_geometry(void)
{
4045
	ulong d;
4046 4047
	int i;
	int j;
4048
	int n = nr_cpu_ids;
4049 4050
	int rcu_capacity[MAX_RCU_LVLS + 1];

4051 4052 4053 4054 4055 4056 4057 4058 4059 4060 4061 4062 4063
	/*
	 * 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;

4064
	/* If the compile-time values are accurate, just leave. */
4065 4066
	if (rcu_fanout_leaf == CONFIG_RCU_FANOUT_LEAF &&
	    nr_cpu_ids == NR_CPUS)
4067
		return;
4068 4069
	pr_info("RCU: Adjusting geometry for rcu_fanout_leaf=%d, nr_cpu_ids=%d\n",
		rcu_fanout_leaf, nr_cpu_ids);
4070 4071 4072 4073 4074 4075 4076 4077 4078 4079 4080 4081 4082 4083 4084 4085 4086 4087 4088 4089 4090 4091 4092 4093 4094 4095 4096 4097 4098 4099 4100 4101 4102 4103 4104 4105 4106 4107 4108 4109 4110 4111 4112 4113 4114

	/*
	 * 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;
}

4115
void __init rcu_init(void)
4116
{
P
Paul E. McKenney 已提交
4117
	int cpu;
4118

4119 4120
	rcu_early_boot_tests();

4121
	rcu_bootup_announce();
4122
	rcu_init_geometry();
4123
	rcu_init_one(&rcu_bh_state, &rcu_bh_data);
4124
	rcu_init_one(&rcu_sched_state, &rcu_sched_data);
4125
	__rcu_init_preempt();
J
Jiang Fang 已提交
4126
	open_softirq(RCU_SOFTIRQ, rcu_process_callbacks);
4127 4128 4129 4130 4131 4132 4133

	/*
	 * 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);
4134
	pm_notifier(rcu_pm_notify, 0);
P
Paul E. McKenney 已提交
4135 4136
	for_each_online_cpu(cpu)
		rcu_cpu_notify(NULL, CPU_UP_PREPARE, (void *)(long)cpu);
4137 4138
}

4139
#include "tree_plugin.h"