linux/kernel/rcutree.c
<<
>>
Prefs
   1/*
   2 * Read-Copy Update mechanism for mutual exclusion
   3 *
   4 * This program is free software; you can redistribute it and/or modify
   5 * it under the terms of the GNU General Public License as published by
   6 * the Free Software Foundation; either version 2 of the License, or
   7 * (at your option) any later version.
   8 *
   9 * This program is distributed in the hope that it will be useful,
  10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12 * GNU General Public License for more details.
  13 *
  14 * You should have received a copy of the GNU General Public License
  15 * along with this program; if not, write to the Free Software
  16 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
  17 *
  18 * Copyright IBM Corporation, 2008
  19 *
  20 * Authors: Dipankar Sarma <dipankar@in.ibm.com>
  21 *          Manfred Spraul <manfred@colorfullife.com>
  22 *          Paul E. McKenney <paulmck@linux.vnet.ibm.com> Hierarchical version
  23 *
  24 * Based on the original work by Paul McKenney <paulmck@us.ibm.com>
  25 * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen.
  26 *
  27 * For detailed explanation of Read-Copy Update mechanism see -
  28 *      Documentation/RCU
  29 */
  30#include <linux/types.h>
  31#include <linux/kernel.h>
  32#include <linux/init.h>
  33#include <linux/spinlock.h>
  34#include <linux/smp.h>
  35#include <linux/rcupdate.h>
  36#include <linux/interrupt.h>
  37#include <linux/sched.h>
  38#include <linux/nmi.h>
  39#include <asm/atomic.h>
  40#include <linux/bitops.h>
  41#include <linux/module.h>
  42#include <linux/completion.h>
  43#include <linux/moduleparam.h>
  44#include <linux/percpu.h>
  45#include <linux/notifier.h>
  46#include <linux/cpu.h>
  47#include <linux/mutex.h>
  48#include <linux/time.h>
  49#include <linux/kernel_stat.h>
  50
  51#include "rcutree.h"
  52
  53/* Data structures. */
  54
  55static struct lock_class_key rcu_node_class[NUM_RCU_LVLS];
  56
  57#define RCU_STATE_INITIALIZER(structname) { \
  58        .level = { &structname.node[0] }, \
  59        .levelcnt = { \
  60                NUM_RCU_LVL_0,  /* root of hierarchy. */ \
  61                NUM_RCU_LVL_1, \
  62                NUM_RCU_LVL_2, \
  63                NUM_RCU_LVL_3, \
  64                NUM_RCU_LVL_4, /* == MAX_RCU_LVLS */ \
  65        }, \
  66        .signaled = RCU_GP_IDLE, \
  67        .gpnum = -300, \
  68        .completed = -300, \
  69        .onofflock = __RAW_SPIN_LOCK_UNLOCKED(&structname.onofflock), \
  70        .orphan_cbs_list = NULL, \
  71        .orphan_cbs_tail = &structname.orphan_cbs_list, \
  72        .orphan_qlen = 0, \
  73        .fqslock = __RAW_SPIN_LOCK_UNLOCKED(&structname.fqslock), \
  74        .n_force_qs = 0, \
  75        .n_force_qs_ngp = 0, \
  76        .name = #structname, \
  77}
  78
  79struct rcu_state rcu_sched_state = RCU_STATE_INITIALIZER(rcu_sched_state);
  80DEFINE_PER_CPU(struct rcu_data, rcu_sched_data);
  81
  82struct rcu_state rcu_bh_state = RCU_STATE_INITIALIZER(rcu_bh_state);
  83DEFINE_PER_CPU(struct rcu_data, rcu_bh_data);
  84
  85int rcu_scheduler_active __read_mostly;
  86EXPORT_SYMBOL_GPL(rcu_scheduler_active);
  87
  88/*
  89 * Return true if an RCU grace period is in progress.  The ACCESS_ONCE()s
  90 * permit this function to be invoked without holding the root rcu_node
  91 * structure's ->lock, but of course results can be subject to change.
  92 */
  93static int rcu_gp_in_progress(struct rcu_state *rsp)
  94{
  95        return ACCESS_ONCE(rsp->completed) != ACCESS_ONCE(rsp->gpnum);
  96}
  97
  98/*
  99 * Note a quiescent state.  Because we do not need to know
 100 * how many quiescent states passed, just if there was at least
 101 * one since the start of the grace period, this just sets a flag.
 102 */
 103void rcu_sched_qs(int cpu)
 104{
 105        struct rcu_data *rdp = &per_cpu(rcu_sched_data, cpu);
 106
 107        rdp->passed_quiesc_completed = rdp->gpnum - 1;
 108        barrier();
 109        rdp->passed_quiesc = 1;
 110}
 111
 112void rcu_bh_qs(int cpu)
 113{
 114        struct rcu_data *rdp = &per_cpu(rcu_bh_data, cpu);
 115
 116        rdp->passed_quiesc_completed = rdp->gpnum - 1;
 117        barrier();
 118        rdp->passed_quiesc = 1;
 119}
 120
 121/*
 122 * Note a context switch.  This is a quiescent state for RCU-sched,
 123 * and requires special handling for preemptible RCU.
 124 */
 125void rcu_note_context_switch(int cpu)
 126{
 127        rcu_sched_qs(cpu);
 128        rcu_preempt_note_context_switch(cpu);
 129}
 130
 131#ifdef CONFIG_NO_HZ
 132DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks) = {
 133        .dynticks_nesting = 1,
 134        .dynticks = 1,
 135};
 136#endif /* #ifdef CONFIG_NO_HZ */
 137
 138static int blimit = 10;         /* Maximum callbacks per softirq. */
 139static int qhimark = 10000;     /* If this many pending, ignore blimit. */
 140static int qlowmark = 100;      /* Once only this many pending, use blimit. */
 141
 142module_param(blimit, int, 0);
 143module_param(qhimark, int, 0);
 144module_param(qlowmark, int, 0);
 145
 146#ifdef CONFIG_RCU_CPU_STALL_DETECTOR
 147int rcu_cpu_stall_suppress __read_mostly = RCU_CPU_STALL_SUPPRESS_INIT;
 148module_param(rcu_cpu_stall_suppress, int, 0644);
 149#endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
 150
 151static void force_quiescent_state(struct rcu_state *rsp, int relaxed);
 152static int rcu_pending(int cpu);
 153
 154/*
 155 * Return the number of RCU-sched batches processed thus far for debug & stats.
 156 */
 157long rcu_batches_completed_sched(void)
 158{
 159        return rcu_sched_state.completed;
 160}
 161EXPORT_SYMBOL_GPL(rcu_batches_completed_sched);
 162
 163/*
 164 * Return the number of RCU BH batches processed thus far for debug & stats.
 165 */
 166long rcu_batches_completed_bh(void)
 167{
 168        return rcu_bh_state.completed;
 169}
 170EXPORT_SYMBOL_GPL(rcu_batches_completed_bh);
 171
 172/*
 173 * Force a quiescent state for RCU BH.
 174 */
 175void rcu_bh_force_quiescent_state(void)
 176{
 177        force_quiescent_state(&rcu_bh_state, 0);
 178}
 179EXPORT_SYMBOL_GPL(rcu_bh_force_quiescent_state);
 180
 181/*
 182 * Force a quiescent state for RCU-sched.
 183 */
 184void rcu_sched_force_quiescent_state(void)
 185{
 186        force_quiescent_state(&rcu_sched_state, 0);
 187}
 188EXPORT_SYMBOL_GPL(rcu_sched_force_quiescent_state);
 189
 190/*
 191 * Does the CPU have callbacks ready to be invoked?
 192 */
 193static int
 194cpu_has_callbacks_ready_to_invoke(struct rcu_data *rdp)
 195{
 196        return &rdp->nxtlist != rdp->nxttail[RCU_DONE_TAIL];
 197}
 198
 199/*
 200 * Does the current CPU require a yet-as-unscheduled grace period?
 201 */
 202static int
 203cpu_needs_another_gp(struct rcu_state *rsp, struct rcu_data *rdp)
 204{
 205        return *rdp->nxttail[RCU_DONE_TAIL] && !rcu_gp_in_progress(rsp);
 206}
 207
 208/*
 209 * Return the root node of the specified rcu_state structure.
 210 */
 211static struct rcu_node *rcu_get_root(struct rcu_state *rsp)
 212{
 213        return &rsp->node[0];
 214}
 215
 216#ifdef CONFIG_SMP
 217
 218/*
 219 * If the specified CPU is offline, tell the caller that it is in
 220 * a quiescent state.  Otherwise, whack it with a reschedule IPI.
 221 * Grace periods can end up waiting on an offline CPU when that
 222 * CPU is in the process of coming online -- it will be added to the
 223 * rcu_node bitmasks before it actually makes it online.  The same thing
 224 * can happen while a CPU is in the process of coming online.  Because this
 225 * race is quite rare, we check for it after detecting that the grace
 226 * period has been delayed rather than checking each and every CPU
 227 * each and every time we start a new grace period.
 228 */
 229static int rcu_implicit_offline_qs(struct rcu_data *rdp)
 230{
 231        /*
 232         * If the CPU is offline, it is in a quiescent state.  We can
 233         * trust its state not to change because interrupts are disabled.
 234         */
 235        if (cpu_is_offline(rdp->cpu)) {
 236                rdp->offline_fqs++;
 237                return 1;
 238        }
 239
 240        /* If preemptable RCU, no point in sending reschedule IPI. */
 241        if (rdp->preemptable)
 242                return 0;
 243
 244        /* The CPU is online, so send it a reschedule IPI. */
 245        if (rdp->cpu != smp_processor_id())
 246                smp_send_reschedule(rdp->cpu);
 247        else
 248                set_need_resched();
 249        rdp->resched_ipi++;
 250        return 0;
 251}
 252
 253#endif /* #ifdef CONFIG_SMP */
 254
 255#ifdef CONFIG_NO_HZ
 256
 257/**
 258 * rcu_enter_nohz - inform RCU that current CPU is entering nohz
 259 *
 260 * Enter nohz mode, in other words, -leave- the mode in which RCU
 261 * read-side critical sections can occur.  (Though RCU read-side
 262 * critical sections can occur in irq handlers in nohz mode, a possibility
 263 * handled by rcu_irq_enter() and rcu_irq_exit()).
 264 */
 265void rcu_enter_nohz(void)
 266{
 267        unsigned long flags;
 268        struct rcu_dynticks *rdtp;
 269
 270        smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
 271        local_irq_save(flags);
 272        rdtp = &__get_cpu_var(rcu_dynticks);
 273        rdtp->dynticks++;
 274        rdtp->dynticks_nesting--;
 275        WARN_ON_ONCE(rdtp->dynticks & 0x1);
 276        local_irq_restore(flags);
 277}
 278
 279/*
 280 * rcu_exit_nohz - inform RCU that current CPU is leaving nohz
 281 *
 282 * Exit nohz mode, in other words, -enter- the mode in which RCU
 283 * read-side critical sections normally occur.
 284 */
 285void rcu_exit_nohz(void)
 286{
 287        unsigned long flags;
 288        struct rcu_dynticks *rdtp;
 289
 290        local_irq_save(flags);
 291        rdtp = &__get_cpu_var(rcu_dynticks);
 292        rdtp->dynticks++;
 293        rdtp->dynticks_nesting++;
 294        WARN_ON_ONCE(!(rdtp->dynticks & 0x1));
 295        local_irq_restore(flags);
 296        smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
 297}
 298
 299/**
 300 * rcu_nmi_enter - inform RCU of entry to NMI context
 301 *
 302 * If the CPU was idle with dynamic ticks active, and there is no
 303 * irq handler running, this updates rdtp->dynticks_nmi to let the
 304 * RCU grace-period handling know that the CPU is active.
 305 */
 306void rcu_nmi_enter(void)
 307{
 308        struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
 309
 310        if (rdtp->dynticks & 0x1)
 311                return;
 312        rdtp->dynticks_nmi++;
 313        WARN_ON_ONCE(!(rdtp->dynticks_nmi & 0x1));
 314        smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
 315}
 316
 317/**
 318 * rcu_nmi_exit - inform RCU of exit from NMI context
 319 *
 320 * If the CPU was idle with dynamic ticks active, and there is no
 321 * irq handler running, this updates rdtp->dynticks_nmi to let the
 322 * RCU grace-period handling know that the CPU is no longer active.
 323 */
 324void rcu_nmi_exit(void)
 325{
 326        struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
 327
 328        if (rdtp->dynticks & 0x1)
 329                return;
 330        smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
 331        rdtp->dynticks_nmi++;
 332        WARN_ON_ONCE(rdtp->dynticks_nmi & 0x1);
 333}
 334
 335/**
 336 * rcu_irq_enter - inform RCU of entry to hard irq context
 337 *
 338 * If the CPU was idle with dynamic ticks active, this updates the
 339 * rdtp->dynticks to let the RCU handling know that the CPU is active.
 340 */
 341void rcu_irq_enter(void)
 342{
 343        struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
 344
 345        if (rdtp->dynticks_nesting++)
 346                return;
 347        rdtp->dynticks++;
 348        WARN_ON_ONCE(!(rdtp->dynticks & 0x1));
 349        smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
 350}
 351
 352/**
 353 * rcu_irq_exit - inform RCU of exit from hard irq context
 354 *
 355 * If the CPU was idle with dynamic ticks active, update the rdp->dynticks
 356 * to put let the RCU handling be aware that the CPU is going back to idle
 357 * with no ticks.
 358 */
 359void rcu_irq_exit(void)
 360{
 361        struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks);
 362
 363        if (--rdtp->dynticks_nesting)
 364                return;
 365        smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
 366        rdtp->dynticks++;
 367        WARN_ON_ONCE(rdtp->dynticks & 0x1);
 368
 369        /* If the interrupt queued a callback, get out of dyntick mode. */
 370        if (__get_cpu_var(rcu_sched_data).nxtlist ||
 371            __get_cpu_var(rcu_bh_data).nxtlist)
 372                set_need_resched();
 373}
 374
 375#ifdef CONFIG_SMP
 376
 377/*
 378 * Snapshot the specified CPU's dynticks counter so that we can later
 379 * credit them with an implicit quiescent state.  Return 1 if this CPU
 380 * is in dynticks idle mode, which is an extended quiescent state.
 381 */
 382static int dyntick_save_progress_counter(struct rcu_data *rdp)
 383{
 384        int ret;
 385        int snap;
 386        int snap_nmi;
 387
 388        snap = rdp->dynticks->dynticks;
 389        snap_nmi = rdp->dynticks->dynticks_nmi;
 390        smp_mb();       /* Order sampling of snap with end of grace period. */
 391        rdp->dynticks_snap = snap;
 392        rdp->dynticks_nmi_snap = snap_nmi;
 393        ret = ((snap & 0x1) == 0) && ((snap_nmi & 0x1) == 0);
 394        if (ret)
 395                rdp->dynticks_fqs++;
 396        return ret;
 397}
 398
 399/*
 400 * Return true if the specified CPU has passed through a quiescent
 401 * state by virtue of being in or having passed through an dynticks
 402 * idle state since the last call to dyntick_save_progress_counter()
 403 * for this same CPU.
 404 */
 405static int rcu_implicit_dynticks_qs(struct rcu_data *rdp)
 406{
 407        long curr;
 408        long curr_nmi;
 409        long snap;
 410        long snap_nmi;
 411
 412        curr = rdp->dynticks->dynticks;
 413        snap = rdp->dynticks_snap;
 414        curr_nmi = rdp->dynticks->dynticks_nmi;
 415        snap_nmi = rdp->dynticks_nmi_snap;
 416        smp_mb(); /* force ordering with cpu entering/leaving dynticks. */
 417
 418        /*
 419         * If the CPU passed through or entered a dynticks idle phase with
 420         * no active irq/NMI handlers, then we can safely pretend that the CPU
 421         * already acknowledged the request to pass through a quiescent
 422         * state.  Either way, that CPU cannot possibly be in an RCU
 423         * read-side critical section that started before the beginning
 424         * of the current RCU grace period.
 425         */
 426        if ((curr != snap || (curr & 0x1) == 0) &&
 427            (curr_nmi != snap_nmi || (curr_nmi & 0x1) == 0)) {
 428                rdp->dynticks_fqs++;
 429                return 1;
 430        }
 431
 432        /* Go check for the CPU being offline. */
 433        return rcu_implicit_offline_qs(rdp);
 434}
 435
 436#endif /* #ifdef CONFIG_SMP */
 437
 438#else /* #ifdef CONFIG_NO_HZ */
 439
 440#ifdef CONFIG_SMP
 441
 442static int dyntick_save_progress_counter(struct rcu_data *rdp)
 443{
 444        return 0;
 445}
 446
 447static int rcu_implicit_dynticks_qs(struct rcu_data *rdp)
 448{
 449        return rcu_implicit_offline_qs(rdp);
 450}
 451
 452#endif /* #ifdef CONFIG_SMP */
 453
 454#endif /* #else #ifdef CONFIG_NO_HZ */
 455
 456#ifdef CONFIG_RCU_CPU_STALL_DETECTOR
 457
 458int rcu_cpu_stall_suppress __read_mostly;
 459
 460static void record_gp_stall_check_time(struct rcu_state *rsp)
 461{
 462        rsp->gp_start = jiffies;
 463        rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_CHECK;
 464}
 465
 466static void print_other_cpu_stall(struct rcu_state *rsp)
 467{
 468        int cpu;
 469        long delta;
 470        unsigned long flags;
 471        struct rcu_node *rnp = rcu_get_root(rsp);
 472
 473        /* Only let one CPU complain about others per time interval. */
 474
 475        raw_spin_lock_irqsave(&rnp->lock, flags);
 476        delta = jiffies - rsp->jiffies_stall;
 477        if (delta < RCU_STALL_RAT_DELAY || !rcu_gp_in_progress(rsp)) {
 478                raw_spin_unlock_irqrestore(&rnp->lock, flags);
 479                return;
 480        }
 481        rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_RECHECK;
 482
 483        /*
 484         * Now rat on any tasks that got kicked up to the root rcu_node
 485         * due to CPU offlining.
 486         */
 487        rcu_print_task_stall(rnp);
 488        raw_spin_unlock_irqrestore(&rnp->lock, flags);
 489
 490        /*
 491         * OK, time to rat on our buddy...
 492         * See Documentation/RCU/stallwarn.txt for info on how to debug
 493         * RCU CPU stall warnings.
 494         */
 495        printk(KERN_ERR "INFO: %s detected stalls on CPUs/tasks: {",
 496               rsp->name);
 497        rcu_for_each_leaf_node(rsp, rnp) {
 498                raw_spin_lock_irqsave(&rnp->lock, flags);
 499                rcu_print_task_stall(rnp);
 500                raw_spin_unlock_irqrestore(&rnp->lock, flags);
 501                if (rnp->qsmask == 0)
 502                        continue;
 503                for (cpu = 0; cpu <= rnp->grphi - rnp->grplo; cpu++)
 504                        if (rnp->qsmask & (1UL << cpu))
 505                                printk(" %d", rnp->grplo + cpu);
 506        }
 507        printk("} (detected by %d, t=%ld jiffies)\n",
 508               smp_processor_id(), (long)(jiffies - rsp->gp_start));
 509        trigger_all_cpu_backtrace();
 510
 511        /* If so configured, complain about tasks blocking the grace period. */
 512
 513        rcu_print_detail_task_stall(rsp);
 514
 515        force_quiescent_state(rsp, 0);  /* Kick them all. */
 516}
 517
 518static void print_cpu_stall(struct rcu_state *rsp)
 519{
 520        unsigned long flags;
 521        struct rcu_node *rnp = rcu_get_root(rsp);
 522
 523        /*
 524         * OK, time to rat on ourselves...
 525         * See Documentation/RCU/stallwarn.txt for info on how to debug
 526         * RCU CPU stall warnings.
 527         */
 528        printk(KERN_ERR "INFO: %s detected stall on CPU %d (t=%lu jiffies)\n",
 529               rsp->name, smp_processor_id(), jiffies - rsp->gp_start);
 530        trigger_all_cpu_backtrace();
 531
 532        raw_spin_lock_irqsave(&rnp->lock, flags);
 533        if (ULONG_CMP_GE(jiffies, rsp->jiffies_stall))
 534                rsp->jiffies_stall =
 535                        jiffies + RCU_SECONDS_TILL_STALL_RECHECK;
 536        raw_spin_unlock_irqrestore(&rnp->lock, flags);
 537
 538        set_need_resched();  /* kick ourselves to get things going. */
 539}
 540
 541static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp)
 542{
 543        long delta;
 544        struct rcu_node *rnp;
 545
 546        if (rcu_cpu_stall_suppress)
 547                return;
 548        delta = jiffies - ACCESS_ONCE(rsp->jiffies_stall);
 549        rnp = rdp->mynode;
 550        if ((ACCESS_ONCE(rnp->qsmask) & rdp->grpmask) && delta >= 0) {
 551
 552                /* We haven't checked in, so go dump stack. */
 553                print_cpu_stall(rsp);
 554
 555        } else if (rcu_gp_in_progress(rsp) && delta >= RCU_STALL_RAT_DELAY) {
 556
 557                /* They had two time units to dump stack, so complain. */
 558                print_other_cpu_stall(rsp);
 559        }
 560}
 561
 562static int rcu_panic(struct notifier_block *this, unsigned long ev, void *ptr)
 563{
 564        rcu_cpu_stall_suppress = 1;
 565        return NOTIFY_DONE;
 566}
 567
 568/**
 569 * rcu_cpu_stall_reset - prevent further stall warnings in current grace period
 570 *
 571 * Set the stall-warning timeout way off into the future, thus preventing
 572 * any RCU CPU stall-warning messages from appearing in the current set of
 573 * RCU grace periods.
 574 *
 575 * The caller must disable hard irqs.
 576 */
 577void rcu_cpu_stall_reset(void)
 578{
 579        rcu_sched_state.jiffies_stall = jiffies + ULONG_MAX / 2;
 580        rcu_bh_state.jiffies_stall = jiffies + ULONG_MAX / 2;
 581        rcu_preempt_stall_reset();
 582}
 583
 584static struct notifier_block rcu_panic_block = {
 585        .notifier_call = rcu_panic,
 586};
 587
 588static void __init check_cpu_stall_init(void)
 589{
 590        atomic_notifier_chain_register(&panic_notifier_list, &rcu_panic_block);
 591}
 592
 593#else /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
 594
 595static void record_gp_stall_check_time(struct rcu_state *rsp)
 596{
 597}
 598
 599static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp)
 600{
 601}
 602
 603void rcu_cpu_stall_reset(void)
 604{
 605}
 606
 607static void __init check_cpu_stall_init(void)
 608{
 609}
 610
 611#endif /* #else #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
 612
 613/*
 614 * Update CPU-local rcu_data state to record the newly noticed grace period.
 615 * This is used both when we started the grace period and when we notice
 616 * that someone else started the grace period.  The caller must hold the
 617 * ->lock of the leaf rcu_node structure corresponding to the current CPU,
 618 *  and must have irqs disabled.
 619 */
 620static void __note_new_gpnum(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp)
 621{
 622        if (rdp->gpnum != rnp->gpnum) {
 623                rdp->qs_pending = 1;
 624                rdp->passed_quiesc = 0;
 625                rdp->gpnum = rnp->gpnum;
 626        }
 627}
 628
 629static void note_new_gpnum(struct rcu_state *rsp, struct rcu_data *rdp)
 630{
 631        unsigned long flags;
 632        struct rcu_node *rnp;
 633
 634        local_irq_save(flags);
 635        rnp = rdp->mynode;
 636        if (rdp->gpnum == ACCESS_ONCE(rnp->gpnum) || /* outside lock. */
 637            !raw_spin_trylock(&rnp->lock)) { /* irqs already off, so later. */
 638                local_irq_restore(flags);
 639                return;
 640        }
 641        __note_new_gpnum(rsp, rnp, rdp);
 642        raw_spin_unlock_irqrestore(&rnp->lock, flags);
 643}
 644
 645/*
 646 * Did someone else start a new RCU grace period start since we last
 647 * checked?  Update local state appropriately if so.  Must be called
 648 * on the CPU corresponding to rdp.
 649 */
 650static int
 651check_for_new_grace_period(struct rcu_state *rsp, struct rcu_data *rdp)
 652{
 653        unsigned long flags;
 654        int ret = 0;
 655
 656        local_irq_save(flags);
 657        if (rdp->gpnum != rsp->gpnum) {
 658                note_new_gpnum(rsp, rdp);
 659                ret = 1;
 660        }
 661        local_irq_restore(flags);
 662        return ret;
 663}
 664
 665/*
 666 * Advance this CPU's callbacks, but only if the current grace period
 667 * has ended.  This may be called only from the CPU to whom the rdp
 668 * belongs.  In addition, the corresponding leaf rcu_node structure's
 669 * ->lock must be held by the caller, with irqs disabled.
 670 */
 671static void
 672__rcu_process_gp_end(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp)
 673{
 674        /* Did another grace period end? */
 675        if (rdp->completed != rnp->completed) {
 676
 677                /* Advance callbacks.  No harm if list empty. */
 678                rdp->nxttail[RCU_DONE_TAIL] = rdp->nxttail[RCU_WAIT_TAIL];
 679                rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_READY_TAIL];
 680                rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
 681
 682                /* Remember that we saw this grace-period completion. */
 683                rdp->completed = rnp->completed;
 684        }
 685}
 686
 687/*
 688 * Advance this CPU's callbacks, but only if the current grace period
 689 * has ended.  This may be called only from the CPU to whom the rdp
 690 * belongs.
 691 */
 692static void
 693rcu_process_gp_end(struct rcu_state *rsp, struct rcu_data *rdp)
 694{
 695        unsigned long flags;
 696        struct rcu_node *rnp;
 697
 698        local_irq_save(flags);
 699        rnp = rdp->mynode;
 700        if (rdp->completed == ACCESS_ONCE(rnp->completed) || /* outside lock. */
 701            !raw_spin_trylock(&rnp->lock)) { /* irqs already off, so later. */
 702                local_irq_restore(flags);
 703                return;
 704        }
 705        __rcu_process_gp_end(rsp, rnp, rdp);
 706        raw_spin_unlock_irqrestore(&rnp->lock, flags);
 707}
 708
 709/*
 710 * Do per-CPU grace-period initialization for running CPU.  The caller
 711 * must hold the lock of the leaf rcu_node structure corresponding to
 712 * this CPU.
 713 */
 714static void
 715rcu_start_gp_per_cpu(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_data *rdp)
 716{
 717        /* Prior grace period ended, so advance callbacks for current CPU. */
 718        __rcu_process_gp_end(rsp, rnp, rdp);
 719
 720        /*
 721         * Because this CPU just now started the new grace period, we know
 722         * that all of its callbacks will be covered by this upcoming grace
 723         * period, even the ones that were registered arbitrarily recently.
 724         * Therefore, advance all outstanding callbacks to RCU_WAIT_TAIL.
 725         *
 726         * Other CPUs cannot be sure exactly when the grace period started.
 727         * Therefore, their recently registered callbacks must pass through
 728         * an additional RCU_NEXT_READY stage, so that they will be handled
 729         * by the next RCU grace period.
 730         */
 731        rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
 732        rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
 733
 734        /* Set state so that this CPU will detect the next quiescent state. */
 735        __note_new_gpnum(rsp, rnp, rdp);
 736}
 737
 738/*
 739 * Start a new RCU grace period if warranted, re-initializing the hierarchy
 740 * in preparation for detecting the next grace period.  The caller must hold
 741 * the root node's ->lock, which is released before return.  Hard irqs must
 742 * be disabled.
 743 */
 744static void
 745rcu_start_gp(struct rcu_state *rsp, unsigned long flags)
 746        __releases(rcu_get_root(rsp)->lock)
 747{
 748        struct rcu_data *rdp = this_cpu_ptr(rsp->rda);
 749        struct rcu_node *rnp = rcu_get_root(rsp);
 750
 751        if (!cpu_needs_another_gp(rsp, rdp) || rsp->fqs_active) {
 752                if (cpu_needs_another_gp(rsp, rdp))
 753                        rsp->fqs_need_gp = 1;
 754                if (rnp->completed == rsp->completed) {
 755                        raw_spin_unlock_irqrestore(&rnp->lock, flags);
 756                        return;
 757                }
 758                raw_spin_unlock(&rnp->lock);     /* irqs remain disabled. */
 759
 760                /*
 761                 * Propagate new ->completed value to rcu_node structures
 762                 * so that other CPUs don't have to wait until the start
 763                 * of the next grace period to process their callbacks.
 764                 */
 765                rcu_for_each_node_breadth_first(rsp, rnp) {
 766                        raw_spin_lock(&rnp->lock); /* irqs already disabled. */
 767                        rnp->completed = rsp->completed;
 768                        raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */
 769                }
 770                local_irq_restore(flags);
 771                return;
 772        }
 773
 774        /* Advance to a new grace period and initialize state. */
 775        rsp->gpnum++;
 776        WARN_ON_ONCE(rsp->signaled == RCU_GP_INIT);
 777        rsp->signaled = RCU_GP_INIT; /* Hold off force_quiescent_state. */
 778        rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS;
 779        record_gp_stall_check_time(rsp);
 780
 781        /* Special-case the common single-level case. */
 782        if (NUM_RCU_NODES == 1) {
 783                rcu_preempt_check_blocked_tasks(rnp);
 784                rnp->qsmask = rnp->qsmaskinit;
 785                rnp->gpnum = rsp->gpnum;
 786                rnp->completed = rsp->completed;
 787                rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state OK. */
 788                rcu_start_gp_per_cpu(rsp, rnp, rdp);
 789                raw_spin_unlock_irqrestore(&rnp->lock, flags);
 790                return;
 791        }
 792
 793        raw_spin_unlock(&rnp->lock);  /* leave irqs disabled. */
 794
 795
 796        /* Exclude any concurrent CPU-hotplug operations. */
 797        raw_spin_lock(&rsp->onofflock);  /* irqs already disabled. */
 798
 799        /*
 800         * Set the quiescent-state-needed bits in all the rcu_node
 801         * structures for all currently online CPUs in breadth-first
 802         * order, starting from the root rcu_node structure.  This
 803         * operation relies on the layout of the hierarchy within the
 804         * rsp->node[] array.  Note that other CPUs will access only
 805         * the leaves of the hierarchy, which still indicate that no
 806         * grace period is in progress, at least until the corresponding
 807         * leaf node has been initialized.  In addition, we have excluded
 808         * CPU-hotplug operations.
 809         *
 810         * Note that the grace period cannot complete until we finish
 811         * the initialization process, as there will be at least one
 812         * qsmask bit set in the root node until that time, namely the
 813         * one corresponding to this CPU, due to the fact that we have
 814         * irqs disabled.
 815         */
 816        rcu_for_each_node_breadth_first(rsp, rnp) {
 817                raw_spin_lock(&rnp->lock);      /* irqs already disabled. */
 818                rcu_preempt_check_blocked_tasks(rnp);
 819                rnp->qsmask = rnp->qsmaskinit;
 820                rnp->gpnum = rsp->gpnum;
 821                rnp->completed = rsp->completed;
 822                if (rnp == rdp->mynode)
 823                        rcu_start_gp_per_cpu(rsp, rnp, rdp);
 824                raw_spin_unlock(&rnp->lock);    /* irqs remain disabled. */
 825        }
 826
 827        rnp = rcu_get_root(rsp);
 828        raw_spin_lock(&rnp->lock);              /* irqs already disabled. */
 829        rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state now OK. */
 830        raw_spin_unlock(&rnp->lock);            /* irqs remain disabled. */
 831        raw_spin_unlock_irqrestore(&rsp->onofflock, flags);
 832}
 833
 834/*
 835 * Report a full set of quiescent states to the specified rcu_state
 836 * data structure.  This involves cleaning up after the prior grace
 837 * period and letting rcu_start_gp() start up the next grace period
 838 * if one is needed.  Note that the caller must hold rnp->lock, as
 839 * required by rcu_start_gp(), which will release it.
 840 */
 841static void rcu_report_qs_rsp(struct rcu_state *rsp, unsigned long flags)
 842        __releases(rcu_get_root(rsp)->lock)
 843{
 844        WARN_ON_ONCE(!rcu_gp_in_progress(rsp));
 845        rsp->completed = rsp->gpnum;
 846        rsp->signaled = RCU_GP_IDLE;
 847        rcu_start_gp(rsp, flags);  /* releases root node's rnp->lock. */
 848}
 849
 850/*
 851 * Similar to rcu_report_qs_rdp(), for which it is a helper function.
 852 * Allows quiescent states for a group of CPUs to be reported at one go
 853 * to the specified rcu_node structure, though all the CPUs in the group
 854 * must be represented by the same rcu_node structure (which need not be
 855 * a leaf rcu_node structure, though it often will be).  That structure's
 856 * lock must be held upon entry, and it is released before return.
 857 */
 858static void
 859rcu_report_qs_rnp(unsigned long mask, struct rcu_state *rsp,
 860                  struct rcu_node *rnp, unsigned long flags)
 861        __releases(rnp->lock)
 862{
 863        struct rcu_node *rnp_c;
 864
 865        /* Walk up the rcu_node hierarchy. */
 866        for (;;) {
 867                if (!(rnp->qsmask & mask)) {
 868
 869                        /* Our bit has already been cleared, so done. */
 870                        raw_spin_unlock_irqrestore(&rnp->lock, flags);
 871                        return;
 872                }
 873                rnp->qsmask &= ~mask;
 874                if (rnp->qsmask != 0 || rcu_preempted_readers(rnp)) {
 875
 876                        /* Other bits still set at this level, so done. */
 877                        raw_spin_unlock_irqrestore(&rnp->lock, flags);
 878                        return;
 879                }
 880                mask = rnp->grpmask;
 881                if (rnp->parent == NULL) {
 882
 883                        /* No more levels.  Exit loop holding root lock. */
 884
 885                        break;
 886                }
 887                raw_spin_unlock_irqrestore(&rnp->lock, flags);
 888                rnp_c = rnp;
 889                rnp = rnp->parent;
 890                raw_spin_lock_irqsave(&rnp->lock, flags);
 891                WARN_ON_ONCE(rnp_c->qsmask);
 892        }
 893
 894        /*
 895         * Get here if we are the last CPU to pass through a quiescent
 896         * state for this grace period.  Invoke rcu_report_qs_rsp()
 897         * to clean up and start the next grace period if one is needed.
 898         */
 899        rcu_report_qs_rsp(rsp, flags); /* releases rnp->lock. */
 900}
 901
 902/*
 903 * Record a quiescent state for the specified CPU to that CPU's rcu_data
 904 * structure.  This must be either called from the specified CPU, or
 905 * called when the specified CPU is known to be offline (and when it is
 906 * also known that no other CPU is concurrently trying to help the offline
 907 * CPU).  The lastcomp argument is used to make sure we are still in the
 908 * grace period of interest.  We don't want to end the current grace period
 909 * based on quiescent states detected in an earlier grace period!
 910 */
 911static void
 912rcu_report_qs_rdp(int cpu, struct rcu_state *rsp, struct rcu_data *rdp, long lastcomp)
 913{
 914        unsigned long flags;
 915        unsigned long mask;
 916        struct rcu_node *rnp;
 917
 918        rnp = rdp->mynode;
 919        raw_spin_lock_irqsave(&rnp->lock, flags);
 920        if (lastcomp != rnp->completed) {
 921
 922                /*
 923                 * Someone beat us to it for this grace period, so leave.
 924                 * The race with GP start is resolved by the fact that we
 925                 * hold the leaf rcu_node lock, so that the per-CPU bits
 926                 * cannot yet be initialized -- so we would simply find our
 927                 * CPU's bit already cleared in rcu_report_qs_rnp() if this
 928                 * race occurred.
 929                 */
 930                rdp->passed_quiesc = 0; /* try again later! */
 931                raw_spin_unlock_irqrestore(&rnp->lock, flags);
 932                return;
 933        }
 934        mask = rdp->grpmask;
 935        if ((rnp->qsmask & mask) == 0) {
 936                raw_spin_unlock_irqrestore(&rnp->lock, flags);
 937        } else {
 938                rdp->qs_pending = 0;
 939
 940                /*
 941                 * This GP can't end until cpu checks in, so all of our
 942                 * callbacks can be processed during the next GP.
 943                 */
 944                rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
 945
 946                rcu_report_qs_rnp(mask, rsp, rnp, flags); /* rlses rnp->lock */
 947        }
 948}
 949
 950/*
 951 * Check to see if there is a new grace period of which this CPU
 952 * is not yet aware, and if so, set up local rcu_data state for it.
 953 * Otherwise, see if this CPU has just passed through its first
 954 * quiescent state for this grace period, and record that fact if so.
 955 */
 956static void
 957rcu_check_quiescent_state(struct rcu_state *rsp, struct rcu_data *rdp)
 958{
 959        /* If there is now a new grace period, record and return. */
 960        if (check_for_new_grace_period(rsp, rdp))
 961                return;
 962
 963        /*
 964         * Does this CPU still need to do its part for current grace period?
 965         * If no, return and let the other CPUs do their part as well.
 966         */
 967        if (!rdp->qs_pending)
 968                return;
 969
 970        /*
 971         * Was there a quiescent state since the beginning of the grace
 972         * period? If no, then exit and wait for the next call.
 973         */
 974        if (!rdp->passed_quiesc)
 975                return;
 976
 977        /*
 978         * Tell RCU we are done (but rcu_report_qs_rdp() will be the
 979         * judge of that).
 980         */
 981        rcu_report_qs_rdp(rdp->cpu, rsp, rdp, rdp->passed_quiesc_completed);
 982}
 983
 984#ifdef CONFIG_HOTPLUG_CPU
 985
 986/*
 987 * Move a dying CPU's RCU callbacks to the ->orphan_cbs_list for the
 988 * specified flavor of RCU.  The callbacks will be adopted by the next
 989 * _rcu_barrier() invocation or by the CPU_DEAD notifier, whichever
 990 * comes first.  Because this is invoked from the CPU_DYING notifier,
 991 * irqs are already disabled.
 992 */
 993static void rcu_send_cbs_to_orphanage(struct rcu_state *rsp)
 994{
 995        int i;
 996        struct rcu_data *rdp = this_cpu_ptr(rsp->rda);
 997
 998        if (rdp->nxtlist == NULL)
 999                return;  /* irqs disabled, so comparison is stable. */
1000        raw_spin_lock(&rsp->onofflock);  /* irqs already disabled. */
1001        *rsp->orphan_cbs_tail = rdp->nxtlist;
1002        rsp->orphan_cbs_tail = rdp->nxttail[RCU_NEXT_TAIL];
1003        rdp->nxtlist = NULL;
1004        for (i = 0; i < RCU_NEXT_SIZE; i++)
1005                rdp->nxttail[i] = &rdp->nxtlist;
1006        rsp->orphan_qlen += rdp->qlen;
1007        rdp->n_cbs_orphaned += rdp->qlen;
1008        rdp->qlen = 0;
1009        raw_spin_unlock(&rsp->onofflock);  /* irqs remain disabled. */
1010}
1011
1012/*
1013 * Adopt previously orphaned RCU callbacks.
1014 */
1015static void rcu_adopt_orphan_cbs(struct rcu_state *rsp)
1016{
1017        unsigned long flags;
1018        struct rcu_data *rdp;
1019
1020        raw_spin_lock_irqsave(&rsp->onofflock, flags);
1021        rdp = this_cpu_ptr(rsp->rda);
1022        if (rsp->orphan_cbs_list == NULL) {
1023                raw_spin_unlock_irqrestore(&rsp->onofflock, flags);
1024                return;
1025        }
1026        *rdp->nxttail[RCU_NEXT_TAIL] = rsp->orphan_cbs_list;
1027        rdp->nxttail[RCU_NEXT_TAIL] = rsp->orphan_cbs_tail;
1028        rdp->qlen += rsp->orphan_qlen;
1029        rdp->n_cbs_adopted += rsp->orphan_qlen;
1030        rsp->orphan_cbs_list = NULL;
1031        rsp->orphan_cbs_tail = &rsp->orphan_cbs_list;
1032        rsp->orphan_qlen = 0;
1033        raw_spin_unlock_irqrestore(&rsp->onofflock, flags);
1034}
1035
1036/*
1037 * Remove the outgoing CPU from the bitmasks in the rcu_node hierarchy
1038 * and move all callbacks from the outgoing CPU to the current one.
1039 */
1040static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp)
1041{
1042        unsigned long flags;
1043        unsigned long mask;
1044        int need_report = 0;
1045        struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
1046        struct rcu_node *rnp;
1047
1048        /* Exclude any attempts to start a new grace period. */
1049        raw_spin_lock_irqsave(&rsp->onofflock, flags);
1050
1051        /* Remove the outgoing CPU from the masks in the rcu_node hierarchy. */
1052        rnp = rdp->mynode;      /* this is the outgoing CPU's rnp. */
1053        mask = rdp->grpmask;    /* rnp->grplo is constant. */
1054        do {
1055                raw_spin_lock(&rnp->lock);      /* irqs already disabled. */
1056                rnp->qsmaskinit &= ~mask;
1057                if (rnp->qsmaskinit != 0) {
1058                        if (rnp != rdp->mynode)
1059                                raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */
1060                        break;
1061                }
1062                if (rnp == rdp->mynode)
1063                        need_report = rcu_preempt_offline_tasks(rsp, rnp, rdp);
1064                else
1065                        raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */
1066                mask = rnp->grpmask;
1067                rnp = rnp->parent;
1068        } while (rnp != NULL);
1069
1070        /*
1071         * We still hold the leaf rcu_node structure lock here, and
1072         * irqs are still disabled.  The reason for this subterfuge is
1073         * because invoking rcu_report_unblock_qs_rnp() with ->onofflock
1074         * held leads to deadlock.
1075         */
1076        raw_spin_unlock(&rsp->onofflock); /* irqs remain disabled. */
1077        rnp = rdp->mynode;
1078        if (need_report & RCU_OFL_TASKS_NORM_GP)
1079                rcu_report_unblock_qs_rnp(rnp, flags);
1080        else
1081                raw_spin_unlock_irqrestore(&rnp->lock, flags);
1082        if (need_report & RCU_OFL_TASKS_EXP_GP)
1083                rcu_report_exp_rnp(rsp, rnp);
1084
1085        rcu_adopt_orphan_cbs(rsp);
1086}
1087
1088/*
1089 * Remove the specified CPU from the RCU hierarchy and move any pending
1090 * callbacks that it might have to the current CPU.  This code assumes
1091 * that at least one CPU in the system will remain running at all times.
1092 * Any attempt to offline -all- CPUs is likely to strand RCU callbacks.
1093 */
1094static void rcu_offline_cpu(int cpu)
1095{
1096        __rcu_offline_cpu(cpu, &rcu_sched_state);
1097        __rcu_offline_cpu(cpu, &rcu_bh_state);
1098        rcu_preempt_offline_cpu(cpu);
1099}
1100
1101#else /* #ifdef CONFIG_HOTPLUG_CPU */
1102
1103static void rcu_send_cbs_to_orphanage(struct rcu_state *rsp)
1104{
1105}
1106
1107static void rcu_adopt_orphan_cbs(struct rcu_state *rsp)
1108{
1109}
1110
1111static void rcu_offline_cpu(int cpu)
1112{
1113}
1114
1115#endif /* #else #ifdef CONFIG_HOTPLUG_CPU */
1116
1117/*
1118 * Invoke any RCU callbacks that have made it to the end of their grace
1119 * period.  Thottle as specified by rdp->blimit.
1120 */
1121static void rcu_do_batch(struct rcu_state *rsp, struct rcu_data *rdp)
1122{
1123        unsigned long flags;
1124        struct rcu_head *next, *list, **tail;
1125        int count;
1126
1127        /* If no callbacks are ready, just return.*/
1128        if (!cpu_has_callbacks_ready_to_invoke(rdp))
1129                return;
1130
1131        /*
1132         * Extract the list of ready callbacks, disabling to prevent
1133         * races with call_rcu() from interrupt handlers.
1134         */
1135        local_irq_save(flags);
1136        list = rdp->nxtlist;
1137        rdp->nxtlist = *rdp->nxttail[RCU_DONE_TAIL];
1138        *rdp->nxttail[RCU_DONE_TAIL] = NULL;
1139        tail = rdp->nxttail[RCU_DONE_TAIL];
1140        for (count = RCU_NEXT_SIZE - 1; count >= 0; count--)
1141                if (rdp->nxttail[count] == rdp->nxttail[RCU_DONE_TAIL])
1142                        rdp->nxttail[count] = &rdp->nxtlist;
1143        local_irq_restore(flags);
1144
1145        /* Invoke callbacks. */
1146        count = 0;
1147        while (list) {
1148                next = list->next;
1149                prefetch(next);
1150                debug_rcu_head_unqueue(list);
1151                list->func(list);
1152                list = next;
1153                if (++count >= rdp->blimit)
1154                        break;
1155        }
1156
1157        local_irq_save(flags);
1158
1159        /* Update count, and requeue any remaining callbacks. */
1160        rdp->qlen -= count;
1161        rdp->n_cbs_invoked += count;
1162        if (list != NULL) {
1163                *tail = rdp->nxtlist;
1164                rdp->nxtlist = list;
1165                for (count = 0; count < RCU_NEXT_SIZE; count++)
1166                        if (&rdp->nxtlist == rdp->nxttail[count])
1167                                rdp->nxttail[count] = tail;
1168                        else
1169                                break;
1170        }
1171
1172        /* Reinstate batch limit if we have worked down the excess. */
1173        if (rdp->blimit == LONG_MAX && rdp->qlen <= qlowmark)
1174                rdp->blimit = blimit;
1175
1176        /* Reset ->qlen_last_fqs_check trigger if enough CBs have drained. */
1177        if (rdp->qlen == 0 && rdp->qlen_last_fqs_check != 0) {
1178                rdp->qlen_last_fqs_check = 0;
1179                rdp->n_force_qs_snap = rsp->n_force_qs;
1180        } else if (rdp->qlen < rdp->qlen_last_fqs_check - qhimark)
1181                rdp->qlen_last_fqs_check = rdp->qlen;
1182
1183        local_irq_restore(flags);
1184
1185        /* Re-raise the RCU softirq if there are callbacks remaining. */
1186        if (cpu_has_callbacks_ready_to_invoke(rdp))
1187                raise_softirq(RCU_SOFTIRQ);
1188}
1189
1190/*
1191 * Check to see if this CPU is in a non-context-switch quiescent state
1192 * (user mode or idle loop for rcu, non-softirq execution for rcu_bh).
1193 * Also schedule the RCU softirq handler.
1194 *
1195 * This function must be called with hardirqs disabled.  It is normally
1196 * invoked from the scheduling-clock interrupt.  If rcu_pending returns
1197 * false, there is no point in invoking rcu_check_callbacks().
1198 */
1199void rcu_check_callbacks(int cpu, int user)
1200{
1201        if (user ||
1202            (idle_cpu(cpu) && rcu_scheduler_active &&
1203             !in_softirq() && hardirq_count() <= (1 << HARDIRQ_SHIFT))) {
1204
1205                /*
1206                 * Get here if this CPU took its interrupt from user
1207                 * mode or from the idle loop, and if this is not a
1208                 * nested interrupt.  In this case, the CPU is in
1209                 * a quiescent state, so note it.
1210                 *
1211                 * No memory barrier is required here because both
1212                 * rcu_sched_qs() and rcu_bh_qs() reference only CPU-local
1213                 * variables that other CPUs neither access nor modify,
1214                 * at least not while the corresponding CPU is online.
1215                 */
1216
1217                rcu_sched_qs(cpu);
1218                rcu_bh_qs(cpu);
1219
1220        } else if (!in_softirq()) {
1221
1222                /*
1223                 * Get here if this CPU did not take its interrupt from
1224                 * softirq, in other words, if it is not interrupting
1225                 * a rcu_bh read-side critical section.  This is an _bh
1226                 * critical section, so note it.
1227                 */
1228
1229                rcu_bh_qs(cpu);
1230        }
1231        rcu_preempt_check_callbacks(cpu);
1232        if (rcu_pending(cpu))
1233                raise_softirq(RCU_SOFTIRQ);
1234}
1235
1236#ifdef CONFIG_SMP
1237
1238/*
1239 * Scan the leaf rcu_node structures, processing dyntick state for any that
1240 * have not yet encountered a quiescent state, using the function specified.
1241 * The caller must have suppressed start of new grace periods.
1242 */
1243static void force_qs_rnp(struct rcu_state *rsp, int (*f)(struct rcu_data *))
1244{
1245        unsigned long bit;
1246        int cpu;
1247        unsigned long flags;
1248        unsigned long mask;
1249        struct rcu_node *rnp;
1250
1251        rcu_for_each_leaf_node(rsp, rnp) {
1252                mask = 0;
1253                raw_spin_lock_irqsave(&rnp->lock, flags);
1254                if (!rcu_gp_in_progress(rsp)) {
1255                        raw_spin_unlock_irqrestore(&rnp->lock, flags);
1256                        return;
1257                }
1258                if (rnp->qsmask == 0) {
1259                        raw_spin_unlock_irqrestore(&rnp->lock, flags);
1260                        continue;
1261                }
1262                cpu = rnp->grplo;
1263                bit = 1;
1264                for (; cpu <= rnp->grphi; cpu++, bit <<= 1) {
1265                        if ((rnp->qsmask & bit) != 0 &&
1266                            f(per_cpu_ptr(rsp->rda, cpu)))
1267                                mask |= bit;
1268                }
1269                if (mask != 0) {
1270
1271                        /* rcu_report_qs_rnp() releases rnp->lock. */
1272                        rcu_report_qs_rnp(mask, rsp, rnp, flags);
1273                        continue;
1274                }
1275                raw_spin_unlock_irqrestore(&rnp->lock, flags);
1276        }
1277}
1278
1279/*
1280 * Force quiescent states on reluctant CPUs, and also detect which
1281 * CPUs are in dyntick-idle mode.
1282 */
1283static void force_quiescent_state(struct rcu_state *rsp, int relaxed)
1284{
1285        unsigned long flags;
1286        struct rcu_node *rnp = rcu_get_root(rsp);
1287
1288        if (!rcu_gp_in_progress(rsp))
1289                return;  /* No grace period in progress, nothing to force. */
1290        if (!raw_spin_trylock_irqsave(&rsp->fqslock, flags)) {
1291                rsp->n_force_qs_lh++; /* Inexact, can lose counts.  Tough! */
1292                return; /* Someone else is already on the job. */
1293        }
1294        if (relaxed && ULONG_CMP_GE(rsp->jiffies_force_qs, jiffies))
1295                goto unlock_fqs_ret; /* no emergency and done recently. */
1296        rsp->n_force_qs++;
1297        raw_spin_lock(&rnp->lock);  /* irqs already disabled */
1298        rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS;
1299        if(!rcu_gp_in_progress(rsp)) {
1300                rsp->n_force_qs_ngp++;
1301                raw_spin_unlock(&rnp->lock);  /* irqs remain disabled */
1302                goto unlock_fqs_ret;  /* no GP in progress, time updated. */
1303        }
1304        rsp->fqs_active = 1;
1305        switch (rsp->signaled) {
1306        case RCU_GP_IDLE:
1307        case RCU_GP_INIT:
1308
1309                break; /* grace period idle or initializing, ignore. */
1310
1311        case RCU_SAVE_DYNTICK:
1312                if (RCU_SIGNAL_INIT != RCU_SAVE_DYNTICK)
1313                        break; /* So gcc recognizes the dead code. */
1314
1315                raw_spin_unlock(&rnp->lock);  /* irqs remain disabled */
1316
1317                /* Record dyntick-idle state. */
1318                force_qs_rnp(rsp, dyntick_save_progress_counter);
1319                raw_spin_lock(&rnp->lock);  /* irqs already disabled */
1320                if (rcu_gp_in_progress(rsp))
1321                        rsp->signaled = RCU_FORCE_QS;
1322                break;
1323
1324        case RCU_FORCE_QS:
1325
1326                /* Check dyntick-idle state, send IPI to laggarts. */
1327                raw_spin_unlock(&rnp->lock);  /* irqs remain disabled */
1328                force_qs_rnp(rsp, rcu_implicit_dynticks_qs);
1329
1330                /* Leave state in case more forcing is required. */
1331
1332                raw_spin_lock(&rnp->lock);  /* irqs already disabled */
1333                break;
1334        }
1335        rsp->fqs_active = 0;
1336        if (rsp->fqs_need_gp) {
1337                raw_spin_unlock(&rsp->fqslock); /* irqs remain disabled */
1338                rsp->fqs_need_gp = 0;
1339                rcu_start_gp(rsp, flags); /* releases rnp->lock */
1340                return;
1341        }
1342        raw_spin_unlock(&rnp->lock);  /* irqs remain disabled */
1343unlock_fqs_ret:
1344        raw_spin_unlock_irqrestore(&rsp->fqslock, flags);
1345}
1346
1347#else /* #ifdef CONFIG_SMP */
1348
1349static void force_quiescent_state(struct rcu_state *rsp, int relaxed)
1350{
1351        set_need_resched();
1352}
1353
1354#endif /* #else #ifdef CONFIG_SMP */
1355
1356/*
1357 * This does the RCU processing work from softirq context for the
1358 * specified rcu_state and rcu_data structures.  This may be called
1359 * only from the CPU to whom the rdp belongs.
1360 */
1361static void
1362__rcu_process_callbacks(struct rcu_state *rsp, struct rcu_data *rdp)
1363{
1364        unsigned long flags;
1365
1366        WARN_ON_ONCE(rdp->beenonline == 0);
1367
1368        /*
1369         * If an RCU GP has gone long enough, go check for dyntick
1370         * idle CPUs and, if needed, send resched IPIs.
1371         */
1372        if (ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs), jiffies))
1373                force_quiescent_state(rsp, 1);
1374
1375        /*
1376         * Advance callbacks in response to end of earlier grace
1377         * period that some other CPU ended.
1378         */
1379        rcu_process_gp_end(rsp, rdp);
1380
1381        /* Update RCU state based on any recent quiescent states. */
1382        rcu_check_quiescent_state(rsp, rdp);
1383
1384        /* Does this CPU require a not-yet-started grace period? */
1385        if (cpu_needs_another_gp(rsp, rdp)) {
1386                raw_spin_lock_irqsave(&rcu_get_root(rsp)->lock, flags);
1387                rcu_start_gp(rsp, flags);  /* releases above lock */
1388        }
1389
1390        /* If there are callbacks ready, invoke them. */
1391        rcu_do_batch(rsp, rdp);
1392}
1393
1394/*
1395 * Do softirq processing for the current CPU.
1396 */
1397static void rcu_process_callbacks(struct softirq_action *unused)
1398{
1399        /*
1400         * Memory references from any prior RCU read-side critical sections
1401         * executed by the interrupted code must be seen before any RCU
1402         * grace-period manipulations below.
1403         */
1404        smp_mb(); /* See above block comment. */
1405
1406        __rcu_process_callbacks(&rcu_sched_state,
1407                                &__get_cpu_var(rcu_sched_data));
1408        __rcu_process_callbacks(&rcu_bh_state, &__get_cpu_var(rcu_bh_data));
1409        rcu_preempt_process_callbacks();
1410
1411        /*
1412         * Memory references from any later RCU read-side critical sections
1413         * executed by the interrupted code must be seen after any RCU
1414         * grace-period manipulations above.
1415         */
1416        smp_mb(); /* See above block comment. */
1417
1418        /* If we are last CPU on way to dyntick-idle mode, accelerate it. */
1419        rcu_needs_cpu_flush();
1420}
1421
1422static void
1423__call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu),
1424           struct rcu_state *rsp)
1425{
1426        unsigned long flags;
1427        struct rcu_data *rdp;
1428
1429        debug_rcu_head_queue(head);
1430        head->func = func;
1431        head->next = NULL;
1432
1433        smp_mb(); /* Ensure RCU update seen before callback registry. */
1434
1435        /*
1436         * Opportunistically note grace-period endings and beginnings.
1437         * Note that we might see a beginning right after we see an
1438         * end, but never vice versa, since this CPU has to pass through
1439         * a quiescent state betweentimes.
1440         */
1441        local_irq_save(flags);
1442        rdp = this_cpu_ptr(rsp->rda);
1443        rcu_process_gp_end(rsp, rdp);
1444        check_for_new_grace_period(rsp, rdp);
1445
1446        /* Add the callback to our list. */
1447        *rdp->nxttail[RCU_NEXT_TAIL] = head;
1448        rdp->nxttail[RCU_NEXT_TAIL] = &head->next;
1449
1450        /* Start a new grace period if one not already started. */
1451        if (!rcu_gp_in_progress(rsp)) {
1452                unsigned long nestflag;
1453                struct rcu_node *rnp_root = rcu_get_root(rsp);
1454
1455                raw_spin_lock_irqsave(&rnp_root->lock, nestflag);
1456                rcu_start_gp(rsp, nestflag);  /* releases rnp_root->lock. */
1457        }
1458
1459        /*
1460         * Force the grace period if too many callbacks or too long waiting.
1461         * Enforce hysteresis, and don't invoke force_quiescent_state()
1462         * if some other CPU has recently done so.  Also, don't bother
1463         * invoking force_quiescent_state() if the newly enqueued callback
1464         * is the only one waiting for a grace period to complete.
1465         */
1466        if (unlikely(++rdp->qlen > rdp->qlen_last_fqs_check + qhimark)) {
1467                rdp->blimit = LONG_MAX;
1468                if (rsp->n_force_qs == rdp->n_force_qs_snap &&
1469                    *rdp->nxttail[RCU_DONE_TAIL] != head)
1470                        force_quiescent_state(rsp, 0);
1471                rdp->n_force_qs_snap = rsp->n_force_qs;
1472                rdp->qlen_last_fqs_check = rdp->qlen;
1473        } else if (ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs), jiffies))
1474                force_quiescent_state(rsp, 1);
1475        local_irq_restore(flags);
1476}
1477
1478/*
1479 * Queue an RCU-sched callback for invocation after a grace period.
1480 */
1481void call_rcu_sched(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
1482{
1483        __call_rcu(head, func, &rcu_sched_state);
1484}
1485EXPORT_SYMBOL_GPL(call_rcu_sched);
1486
1487/*
1488 * Queue an RCU for invocation after a quicker grace period.
1489 */
1490void call_rcu_bh(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
1491{
1492        __call_rcu(head, func, &rcu_bh_state);
1493}
1494EXPORT_SYMBOL_GPL(call_rcu_bh);
1495
1496/**
1497 * synchronize_sched - wait until an rcu-sched grace period has elapsed.
1498 *
1499 * Control will return to the caller some time after a full rcu-sched
1500 * grace period has elapsed, in other words after all currently executing
1501 * rcu-sched read-side critical sections have completed.   These read-side
1502 * critical sections are delimited by rcu_read_lock_sched() and
1503 * rcu_read_unlock_sched(), and may be nested.  Note that preempt_disable(),
1504 * local_irq_disable(), and so on may be used in place of
1505 * rcu_read_lock_sched().
1506 *
1507 * This means that all preempt_disable code sequences, including NMI and
1508 * hardware-interrupt handlers, in progress on entry will have completed
1509 * before this primitive returns.  However, this does not guarantee that
1510 * softirq handlers will have completed, since in some kernels, these
1511 * handlers can run in process context, and can block.
1512 *
1513 * This primitive provides the guarantees made by the (now removed)
1514 * synchronize_kernel() API.  In contrast, synchronize_rcu() only
1515 * guarantees that rcu_read_lock() sections will have completed.
1516 * In "classic RCU", these two guarantees happen to be one and
1517 * the same, but can differ in realtime RCU implementations.
1518 */
1519void synchronize_sched(void)
1520{
1521        struct rcu_synchronize rcu;
1522
1523        if (rcu_blocking_is_gp())
1524                return;
1525
1526        init_rcu_head_on_stack(&rcu.head);
1527        init_completion(&rcu.completion);
1528        /* Will wake me after RCU finished. */
1529        call_rcu_sched(&rcu.head, wakeme_after_rcu);
1530        /* Wait for it. */
1531        wait_for_completion(&rcu.completion);
1532        destroy_rcu_head_on_stack(&rcu.head);
1533}
1534EXPORT_SYMBOL_GPL(synchronize_sched);
1535
1536/**
1537 * synchronize_rcu_bh - wait until an rcu_bh grace period has elapsed.
1538 *
1539 * Control will return to the caller some time after a full rcu_bh grace
1540 * period has elapsed, in other words after all currently executing rcu_bh
1541 * read-side critical sections have completed.  RCU read-side critical
1542 * sections are delimited by rcu_read_lock_bh() and rcu_read_unlock_bh(),
1543 * and may be nested.
1544 */
1545void synchronize_rcu_bh(void)
1546{
1547        struct rcu_synchronize rcu;
1548
1549        if (rcu_blocking_is_gp())
1550                return;
1551
1552        init_rcu_head_on_stack(&rcu.head);
1553        init_completion(&rcu.completion);
1554        /* Will wake me after RCU finished. */
1555        call_rcu_bh(&rcu.head, wakeme_after_rcu);
1556        /* Wait for it. */
1557        wait_for_completion(&rcu.completion);
1558        destroy_rcu_head_on_stack(&rcu.head);
1559}
1560EXPORT_SYMBOL_GPL(synchronize_rcu_bh);
1561
1562/*
1563 * Check to see if there is any immediate RCU-related work to be done
1564 * by the current CPU, for the specified type of RCU, returning 1 if so.
1565 * The checks are in order of increasing expense: checks that can be
1566 * carried out against CPU-local state are performed first.  However,
1567 * we must check for CPU stalls first, else we might not get a chance.
1568 */
1569static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp)
1570{
1571        struct rcu_node *rnp = rdp->mynode;
1572
1573        rdp->n_rcu_pending++;
1574
1575        /* Check for CPU stalls, if enabled. */
1576        check_cpu_stall(rsp, rdp);
1577
1578        /* Is the RCU core waiting for a quiescent state from this CPU? */
1579        if (rdp->qs_pending && !rdp->passed_quiesc) {
1580
1581                /*
1582                 * If force_quiescent_state() coming soon and this CPU
1583                 * needs a quiescent state, and this is either RCU-sched
1584                 * or RCU-bh, force a local reschedule.
1585                 */
1586                rdp->n_rp_qs_pending++;
1587                if (!rdp->preemptable &&
1588                    ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs) - 1,
1589                                 jiffies))
1590                        set_need_resched();
1591        } else if (rdp->qs_pending && rdp->passed_quiesc) {
1592                rdp->n_rp_report_qs++;
1593                return 1;
1594        }
1595
1596        /* Does this CPU have callbacks ready to invoke? */
1597        if (cpu_has_callbacks_ready_to_invoke(rdp)) {
1598                rdp->n_rp_cb_ready++;
1599                return 1;
1600        }
1601
1602        /* Has RCU gone idle with this CPU needing another grace period? */
1603        if (cpu_needs_another_gp(rsp, rdp)) {
1604                rdp->n_rp_cpu_needs_gp++;
1605                return 1;
1606        }
1607
1608        /* Has another RCU grace period completed?  */
1609        if (ACCESS_ONCE(rnp->completed) != rdp->completed) { /* outside lock */
1610                rdp->n_rp_gp_completed++;
1611                return 1;
1612        }
1613
1614        /* Has a new RCU grace period started? */
1615        if (ACCESS_ONCE(rnp->gpnum) != rdp->gpnum) { /* outside lock */
1616                rdp->n_rp_gp_started++;
1617                return 1;
1618        }
1619
1620        /* Has an RCU GP gone long enough to send resched IPIs &c? */
1621        if (rcu_gp_in_progress(rsp) &&
1622            ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs), jiffies)) {
1623                rdp->n_rp_need_fqs++;
1624                return 1;
1625        }
1626
1627        /* nothing to do */
1628        rdp->n_rp_need_nothing++;
1629        return 0;
1630}
1631
1632/*
1633 * Check to see if there is any immediate RCU-related work to be done
1634 * by the current CPU, returning 1 if so.  This function is part of the
1635 * RCU implementation; it is -not- an exported member of the RCU API.
1636 */
1637static int rcu_pending(int cpu)
1638{
1639        return __rcu_pending(&rcu_sched_state, &per_cpu(rcu_sched_data, cpu)) ||
1640               __rcu_pending(&rcu_bh_state, &per_cpu(rcu_bh_data, cpu)) ||
1641               rcu_preempt_pending(cpu);
1642}
1643
1644/*
1645 * Check to see if any future RCU-related work will need to be done
1646 * by the current CPU, even if none need be done immediately, returning
1647 * 1 if so.
1648 */
1649static int rcu_needs_cpu_quick_check(int cpu)
1650{
1651        /* RCU callbacks either ready or pending? */
1652        return per_cpu(rcu_sched_data, cpu).nxtlist ||
1653               per_cpu(rcu_bh_data, cpu).nxtlist ||
1654               rcu_preempt_needs_cpu(cpu);
1655}
1656
1657static DEFINE_PER_CPU(struct rcu_head, rcu_barrier_head) = {NULL};
1658static atomic_t rcu_barrier_cpu_count;
1659static DEFINE_MUTEX(rcu_barrier_mutex);
1660static struct completion rcu_barrier_completion;
1661
1662static void rcu_barrier_callback(struct rcu_head *notused)
1663{
1664        if (atomic_dec_and_test(&rcu_barrier_cpu_count))
1665                complete(&rcu_barrier_completion);
1666}
1667
1668/*
1669 * Called with preemption disabled, and from cross-cpu IRQ context.
1670 */
1671static void rcu_barrier_func(void *type)
1672{
1673        int cpu = smp_processor_id();
1674        struct rcu_head *head = &per_cpu(rcu_barrier_head, cpu);
1675        void (*call_rcu_func)(struct rcu_head *head,
1676                              void (*func)(struct rcu_head *head));
1677
1678        atomic_inc(&rcu_barrier_cpu_count);
1679        call_rcu_func = type;
1680        call_rcu_func(head, rcu_barrier_callback);
1681}
1682
1683/*
1684 * Orchestrate the specified type of RCU barrier, waiting for all
1685 * RCU callbacks of the specified type to complete.
1686 */
1687static void _rcu_barrier(struct rcu_state *rsp,
1688                         void (*call_rcu_func)(struct rcu_head *head,
1689                                               void (*func)(struct rcu_head *head)))
1690{
1691        BUG_ON(in_interrupt());
1692        /* Take mutex to serialize concurrent rcu_barrier() requests. */
1693        mutex_lock(&rcu_barrier_mutex);
1694        init_completion(&rcu_barrier_completion);
1695        /*
1696         * Initialize rcu_barrier_cpu_count to 1, then invoke
1697         * rcu_barrier_func() on each CPU, so that each CPU also has
1698         * incremented rcu_barrier_cpu_count.  Only then is it safe to
1699         * decrement rcu_barrier_cpu_count -- otherwise the first CPU
1700         * might complete its grace period before all of the other CPUs
1701         * did their increment, causing this function to return too
1702         * early.
1703         */
1704        atomic_set(&rcu_barrier_cpu_count, 1);
1705        preempt_disable(); /* stop CPU_DYING from filling orphan_cbs_list */
1706        rcu_adopt_orphan_cbs(rsp);
1707        on_each_cpu(rcu_barrier_func, (void *)call_rcu_func, 1);
1708        preempt_enable(); /* CPU_DYING can again fill orphan_cbs_list */
1709        if (atomic_dec_and_test(&rcu_barrier_cpu_count))
1710                complete(&rcu_barrier_completion);
1711        wait_for_completion(&rcu_barrier_completion);
1712        mutex_unlock(&rcu_barrier_mutex);
1713}
1714
1715/**
1716 * rcu_barrier_bh - Wait until all in-flight call_rcu_bh() callbacks complete.
1717 */
1718void rcu_barrier_bh(void)
1719{
1720        _rcu_barrier(&rcu_bh_state, call_rcu_bh);
1721}
1722EXPORT_SYMBOL_GPL(rcu_barrier_bh);
1723
1724/**
1725 * rcu_barrier_sched - Wait for in-flight call_rcu_sched() callbacks.
1726 */
1727void rcu_barrier_sched(void)
1728{
1729        _rcu_barrier(&rcu_sched_state, call_rcu_sched);
1730}
1731EXPORT_SYMBOL_GPL(rcu_barrier_sched);
1732
1733/*
1734 * Do boot-time initialization of a CPU's per-CPU RCU data.
1735 */
1736static void __init
1737rcu_boot_init_percpu_data(int cpu, struct rcu_state *rsp)
1738{
1739        unsigned long flags;
1740        int i;
1741        struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
1742        struct rcu_node *rnp = rcu_get_root(rsp);
1743
1744        /* Set up local state, ensuring consistent view of global state. */
1745        raw_spin_lock_irqsave(&rnp->lock, flags);
1746        rdp->grpmask = 1UL << (cpu - rdp->mynode->grplo);
1747        rdp->nxtlist = NULL;
1748        for (i = 0; i < RCU_NEXT_SIZE; i++)
1749                rdp->nxttail[i] = &rdp->nxtlist;
1750        rdp->qlen = 0;
1751#ifdef CONFIG_NO_HZ
1752        rdp->dynticks = &per_cpu(rcu_dynticks, cpu);
1753#endif /* #ifdef CONFIG_NO_HZ */
1754        rdp->cpu = cpu;
1755        raw_spin_unlock_irqrestore(&rnp->lock, flags);
1756}
1757
1758/*
1759 * Initialize a CPU's per-CPU RCU data.  Note that only one online or
1760 * offline event can be happening at a given time.  Note also that we
1761 * can accept some slop in the rsp->completed access due to the fact
1762 * that this CPU cannot possibly have any RCU callbacks in flight yet.
1763 */
1764static void __cpuinit
1765rcu_init_percpu_data(int cpu, struct rcu_state *rsp, int preemptable)
1766{
1767        unsigned long flags;
1768        unsigned long mask;
1769        struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
1770        struct rcu_node *rnp = rcu_get_root(rsp);
1771
1772        /* Set up local state, ensuring consistent view of global state. */
1773        raw_spin_lock_irqsave(&rnp->lock, flags);
1774        rdp->passed_quiesc = 0;  /* We could be racing with new GP, */
1775        rdp->qs_pending = 1;     /*  so set up to respond to current GP. */
1776        rdp->beenonline = 1;     /* We have now been online. */
1777        rdp->preemptable = preemptable;
1778        rdp->qlen_last_fqs_check = 0;
1779        rdp->n_force_qs_snap = rsp->n_force_qs;
1780        rdp->blimit = blimit;
1781        raw_spin_unlock(&rnp->lock);            /* irqs remain disabled. */
1782
1783        /*
1784         * A new grace period might start here.  If so, we won't be part
1785         * of it, but that is OK, as we are currently in a quiescent state.
1786         */
1787
1788        /* Exclude any attempts to start a new GP on large systems. */
1789        raw_spin_lock(&rsp->onofflock);         /* irqs already disabled. */
1790
1791        /* Add CPU to rcu_node bitmasks. */
1792        rnp = rdp->mynode;
1793        mask = rdp->grpmask;
1794        do {
1795                /* Exclude any attempts to start a new GP on small systems. */
1796                raw_spin_lock(&rnp->lock);      /* irqs already disabled. */
1797                rnp->qsmaskinit |= mask;
1798                mask = rnp->grpmask;
1799                if (rnp == rdp->mynode) {
1800                        rdp->gpnum = rnp->completed; /* if GP in progress... */
1801                        rdp->completed = rnp->completed;
1802                        rdp->passed_quiesc_completed = rnp->completed - 1;
1803                }
1804                raw_spin_unlock(&rnp->lock); /* irqs already disabled. */
1805                rnp = rnp->parent;
1806        } while (rnp != NULL && !(rnp->qsmaskinit & mask));
1807
1808        raw_spin_unlock_irqrestore(&rsp->onofflock, flags);
1809}
1810
1811static void __cpuinit rcu_online_cpu(int cpu)
1812{
1813        rcu_init_percpu_data(cpu, &rcu_sched_state, 0);
1814        rcu_init_percpu_data(cpu, &rcu_bh_state, 0);
1815        rcu_preempt_init_percpu_data(cpu);
1816}
1817
1818/*
1819 * Handle CPU online/offline notification events.
1820 */
1821static int __cpuinit rcu_cpu_notify(struct notifier_block *self,
1822                                    unsigned long action, void *hcpu)
1823{
1824        long cpu = (long)hcpu;
1825
1826        switch (action) {
1827        case CPU_UP_PREPARE:
1828        case CPU_UP_PREPARE_FROZEN:
1829                rcu_online_cpu(cpu);
1830                break;
1831        case CPU_DYING:
1832        case CPU_DYING_FROZEN:
1833                /*
1834                 * preempt_disable() in _rcu_barrier() prevents stop_machine(),
1835                 * so when "on_each_cpu(rcu_barrier_func, (void *)type, 1);"
1836                 * returns, all online cpus have queued rcu_barrier_func().
1837                 * The dying CPU clears its cpu_online_mask bit and
1838                 * moves all of its RCU callbacks to ->orphan_cbs_list
1839                 * in the context of stop_machine(), so subsequent calls
1840                 * to _rcu_barrier() will adopt these callbacks and only
1841                 * then queue rcu_barrier_func() on all remaining CPUs.
1842                 */
1843                rcu_send_cbs_to_orphanage(&rcu_bh_state);
1844                rcu_send_cbs_to_orphanage(&rcu_sched_state);
1845                rcu_preempt_send_cbs_to_orphanage();
1846                break;
1847        case CPU_DEAD:
1848        case CPU_DEAD_FROZEN:
1849        case CPU_UP_CANCELED:
1850        case CPU_UP_CANCELED_FROZEN:
1851                rcu_offline_cpu(cpu);
1852                break;
1853        default:
1854                break;
1855        }
1856        return NOTIFY_OK;
1857}
1858
1859/*
1860 * This function is invoked towards the end of the scheduler's initialization
1861 * process.  Before this is called, the idle task might contain
1862 * RCU read-side critical sections (during which time, this idle
1863 * task is booting the system).  After this function is called, the
1864 * idle tasks are prohibited from containing RCU read-side critical
1865 * sections.  This function also enables RCU lockdep checking.
1866 */
1867void rcu_scheduler_starting(void)
1868{
1869        WARN_ON(num_online_cpus() != 1);
1870        WARN_ON(nr_context_switches() > 0);
1871        rcu_scheduler_active = 1;
1872}
1873
1874/*
1875 * Compute the per-level fanout, either using the exact fanout specified
1876 * or balancing the tree, depending on CONFIG_RCU_FANOUT_EXACT.
1877 */
1878#ifdef CONFIG_RCU_FANOUT_EXACT
1879static void __init rcu_init_levelspread(struct rcu_state *rsp)
1880{
1881        int i;
1882
1883        for (i = NUM_RCU_LVLS - 1; i >= 0; i--)
1884                rsp->levelspread[i] = CONFIG_RCU_FANOUT;
1885}
1886#else /* #ifdef CONFIG_RCU_FANOUT_EXACT */
1887static void __init rcu_init_levelspread(struct rcu_state *rsp)
1888{
1889        int ccur;
1890        int cprv;
1891        int i;
1892
1893        cprv = NR_CPUS;
1894        for (i = NUM_RCU_LVLS - 1; i >= 0; i--) {
1895                ccur = rsp->levelcnt[i];
1896                rsp->levelspread[i] = (cprv + ccur - 1) / ccur;
1897                cprv = ccur;
1898        }
1899}
1900#endif /* #else #ifdef CONFIG_RCU_FANOUT_EXACT */
1901
1902/*
1903 * Helper function for rcu_init() that initializes one rcu_state structure.
1904 */
1905static void __init rcu_init_one(struct rcu_state *rsp,
1906                struct rcu_data __percpu *rda)
1907{
1908        static char *buf[] = { "rcu_node_level_0",
1909                               "rcu_node_level_1",
1910                               "rcu_node_level_2",
1911                               "rcu_node_level_3" };  /* Match MAX_RCU_LVLS */
1912        int cpustride = 1;
1913        int i;
1914        int j;
1915        struct rcu_node *rnp;
1916
1917        BUILD_BUG_ON(MAX_RCU_LVLS > ARRAY_SIZE(buf));  /* Fix buf[] init! */
1918
1919        /* Initialize the level-tracking arrays. */
1920
1921        for (i = 1; i < NUM_RCU_LVLS; i++)
1922                rsp->level[i] = rsp->level[i - 1] + rsp->levelcnt[i - 1];
1923        rcu_init_levelspread(rsp);
1924
1925        /* Initialize the elements themselves, starting from the leaves. */
1926
1927        for (i = NUM_RCU_LVLS - 1; i >= 0; i--) {
1928                cpustride *= rsp->levelspread[i];
1929                rnp = rsp->level[i];
1930                for (j = 0; j < rsp->levelcnt[i]; j++, rnp++) {
1931                        raw_spin_lock_init(&rnp->lock);
1932                        lockdep_set_class_and_name(&rnp->lock,
1933                                                   &rcu_node_class[i], buf[i]);
1934                        rnp->gpnum = 0;
1935                        rnp->qsmask = 0;
1936                        rnp->qsmaskinit = 0;
1937                        rnp->grplo = j * cpustride;
1938                        rnp->grphi = (j + 1) * cpustride - 1;
1939                        if (rnp->grphi >= NR_CPUS)
1940                                rnp->grphi = NR_CPUS - 1;
1941                        if (i == 0) {
1942                                rnp->grpnum = 0;
1943                                rnp->grpmask = 0;
1944                                rnp->parent = NULL;
1945                        } else {
1946                                rnp->grpnum = j % rsp->levelspread[i - 1];
1947                                rnp->grpmask = 1UL << rnp->grpnum;
1948                                rnp->parent = rsp->level[i - 1] +
1949                                              j / rsp->levelspread[i - 1];
1950                        }
1951                        rnp->level = i;
1952                        INIT_LIST_HEAD(&rnp->blocked_tasks[0]);
1953                        INIT_LIST_HEAD(&rnp->blocked_tasks[1]);
1954                        INIT_LIST_HEAD(&rnp->blocked_tasks[2]);
1955                        INIT_LIST_HEAD(&rnp->blocked_tasks[3]);
1956                }
1957        }
1958
1959        rsp->rda = rda;
1960        rnp = rsp->level[NUM_RCU_LVLS - 1];
1961        for_each_possible_cpu(i) {
1962                while (i > rnp->grphi)
1963                        rnp++;
1964                per_cpu_ptr(rsp->rda, i)->mynode = rnp;
1965                rcu_boot_init_percpu_data(i, rsp);
1966        }
1967}
1968
1969void __init rcu_init(void)
1970{
1971        int cpu;
1972
1973        rcu_bootup_announce();
1974        rcu_init_one(&rcu_sched_state, &rcu_sched_data);
1975        rcu_init_one(&rcu_bh_state, &rcu_bh_data);
1976        __rcu_init_preempt();
1977        open_softirq(RCU_SOFTIRQ, rcu_process_callbacks);
1978
1979        /*
1980         * We don't need protection against CPU-hotplug here because
1981         * this is called early in boot, before either interrupts
1982         * or the scheduler are operational.
1983         */
1984        cpu_notifier(rcu_cpu_notify, 0);
1985        for_each_online_cpu(cpu)
1986                rcu_cpu_notify(NULL, CPU_UP_PREPARE, (void *)(long)cpu);
1987        check_cpu_stall_init();
1988}
1989
1990#include "rcutree_plugin.h"
1991