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