linux/kernel/rcutiny_plugin.h
<<
>>
Prefs
   1/*
   2 * Read-Copy Update mechanism for mutual exclusion, the Bloatwatch edition
   3 * Internal non-public definitions that provide either classic
   4 * or preemptible semantics.
   5 *
   6 * This program is free software; you can redistribute it and/or modify
   7 * it under the terms of the GNU General Public License as published by
   8 * the Free Software Foundation; either version 2 of the License, or
   9 * (at your option) any later version.
  10 *
  11 * This program is distributed in the hope that it will be useful,
  12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14 * GNU General Public License for more details.
  15 *
  16 * You should have received a copy of the GNU General Public License
  17 * along with this program; if not, write to the Free Software
  18 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
  19 *
  20 * Copyright (c) 2010 Linaro
  21 *
  22 * Author: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
  23 */
  24
  25#include <linux/kthread.h>
  26#include <linux/module.h>
  27#include <linux/debugfs.h>
  28#include <linux/seq_file.h>
  29
  30/* Global control variables for rcupdate callback mechanism. */
  31struct rcu_ctrlblk {
  32        struct rcu_head *rcucblist;     /* List of pending callbacks (CBs). */
  33        struct rcu_head **donetail;     /* ->next pointer of last "done" CB. */
  34        struct rcu_head **curtail;      /* ->next pointer of last CB. */
  35        RCU_TRACE(long qlen);           /* Number of pending CBs. */
  36        RCU_TRACE(char *name);          /* Name of RCU type. */
  37};
  38
  39/* Definition for rcupdate control block. */
  40static struct rcu_ctrlblk rcu_sched_ctrlblk = {
  41        .donetail       = &rcu_sched_ctrlblk.rcucblist,
  42        .curtail        = &rcu_sched_ctrlblk.rcucblist,
  43        RCU_TRACE(.name = "rcu_sched")
  44};
  45
  46static struct rcu_ctrlblk rcu_bh_ctrlblk = {
  47        .donetail       = &rcu_bh_ctrlblk.rcucblist,
  48        .curtail        = &rcu_bh_ctrlblk.rcucblist,
  49        RCU_TRACE(.name = "rcu_bh")
  50};
  51
  52#ifdef CONFIG_DEBUG_LOCK_ALLOC
  53int rcu_scheduler_active __read_mostly;
  54EXPORT_SYMBOL_GPL(rcu_scheduler_active);
  55#endif /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
  56
  57#ifdef CONFIG_TINY_PREEMPT_RCU
  58
  59#include <linux/delay.h>
  60
  61/* Global control variables for preemptible RCU. */
  62struct rcu_preempt_ctrlblk {
  63        struct rcu_ctrlblk rcb; /* curtail: ->next ptr of last CB for GP. */
  64        struct rcu_head **nexttail;
  65                                /* Tasks blocked in a preemptible RCU */
  66                                /*  read-side critical section while an */
  67                                /*  preemptible-RCU grace period is in */
  68                                /*  progress must wait for a later grace */
  69                                /*  period.  This pointer points to the */
  70                                /*  ->next pointer of the last task that */
  71                                /*  must wait for a later grace period, or */
  72                                /*  to &->rcb.rcucblist if there is no */
  73                                /*  such task. */
  74        struct list_head blkd_tasks;
  75                                /* Tasks blocked in RCU read-side critical */
  76                                /*  section.  Tasks are placed at the head */
  77                                /*  of this list and age towards the tail. */
  78        struct list_head *gp_tasks;
  79                                /* Pointer to the first task blocking the */
  80                                /*  current grace period, or NULL if there */
  81                                /*  is no such task. */
  82        struct list_head *exp_tasks;
  83                                /* Pointer to first task blocking the */
  84                                /*  current expedited grace period, or NULL */
  85                                /*  if there is no such task.  If there */
  86                                /*  is no current expedited grace period, */
  87                                /*  then there cannot be any such task. */
  88#ifdef CONFIG_RCU_BOOST
  89        struct list_head *boost_tasks;
  90                                /* Pointer to first task that needs to be */
  91                                /*  priority-boosted, or NULL if no priority */
  92                                /*  boosting is needed.  If there is no */
  93                                /*  current or expedited grace period, there */
  94                                /*  can be no such task. */
  95#endif /* #ifdef CONFIG_RCU_BOOST */
  96        u8 gpnum;               /* Current grace period. */
  97        u8 gpcpu;               /* Last grace period blocked by the CPU. */
  98        u8 completed;           /* Last grace period completed. */
  99                                /*  If all three are equal, RCU is idle. */
 100#ifdef CONFIG_RCU_BOOST
 101        unsigned long boost_time; /* When to start boosting (jiffies) */
 102#endif /* #ifdef CONFIG_RCU_BOOST */
 103#ifdef CONFIG_RCU_TRACE
 104        unsigned long n_grace_periods;
 105#ifdef CONFIG_RCU_BOOST
 106        unsigned long n_tasks_boosted;
 107                                /* Total number of tasks boosted. */
 108        unsigned long n_exp_boosts;
 109                                /* Number of tasks boosted for expedited GP. */
 110        unsigned long n_normal_boosts;
 111                                /* Number of tasks boosted for normal GP. */
 112        unsigned long n_balk_blkd_tasks;
 113                                /* Refused to boost: no blocked tasks. */
 114        unsigned long n_balk_exp_gp_tasks;
 115                                /* Refused to boost: nothing blocking GP. */
 116        unsigned long n_balk_boost_tasks;
 117                                /* Refused to boost: already boosting. */
 118        unsigned long n_balk_notyet;
 119                                /* Refused to boost: not yet time. */
 120        unsigned long n_balk_nos;
 121                                /* Refused to boost: not sure why, though. */
 122                                /*  This can happen due to race conditions. */
 123#endif /* #ifdef CONFIG_RCU_BOOST */
 124#endif /* #ifdef CONFIG_RCU_TRACE */
 125};
 126
 127static struct rcu_preempt_ctrlblk rcu_preempt_ctrlblk = {
 128        .rcb.donetail = &rcu_preempt_ctrlblk.rcb.rcucblist,
 129        .rcb.curtail = &rcu_preempt_ctrlblk.rcb.rcucblist,
 130        .nexttail = &rcu_preempt_ctrlblk.rcb.rcucblist,
 131        .blkd_tasks = LIST_HEAD_INIT(rcu_preempt_ctrlblk.blkd_tasks),
 132        RCU_TRACE(.rcb.name = "rcu_preempt")
 133};
 134
 135static void rcu_read_unlock_special(struct task_struct *t);
 136static int rcu_preempted_readers_exp(void);
 137static void rcu_report_exp_done(void);
 138
 139/*
 140 * Return true if the CPU has not yet responded to the current grace period.
 141 */
 142static int rcu_cpu_blocking_cur_gp(void)
 143{
 144        return rcu_preempt_ctrlblk.gpcpu != rcu_preempt_ctrlblk.gpnum;
 145}
 146
 147/*
 148 * Check for a running RCU reader.  Because there is only one CPU,
 149 * there can be but one running RCU reader at a time.  ;-)
 150 *
 151 * Returns zero if there are no running readers.  Returns a positive
 152 * number if there is at least one reader within its RCU read-side
 153 * critical section.  Returns a negative number if an outermost reader
 154 * is in the midst of exiting from its RCU read-side critical section
 155 *
 156 * Returns zero if there are no running readers.  Returns a positive
 157 * number if there is at least one reader within its RCU read-side
 158 * critical section.  Returns a negative number if an outermost reader
 159 * is in the midst of exiting from its RCU read-side critical section.
 160 */
 161static int rcu_preempt_running_reader(void)
 162{
 163        return current->rcu_read_lock_nesting;
 164}
 165
 166/*
 167 * Check for preempted RCU readers blocking any grace period.
 168 * If the caller needs a reliable answer, it must disable hard irqs.
 169 */
 170static int rcu_preempt_blocked_readers_any(void)
 171{
 172        return !list_empty(&rcu_preempt_ctrlblk.blkd_tasks);
 173}
 174
 175/*
 176 * Check for preempted RCU readers blocking the current grace period.
 177 * If the caller needs a reliable answer, it must disable hard irqs.
 178 */
 179static int rcu_preempt_blocked_readers_cgp(void)
 180{
 181        return rcu_preempt_ctrlblk.gp_tasks != NULL;
 182}
 183
 184/*
 185 * Return true if another preemptible-RCU grace period is needed.
 186 */
 187static int rcu_preempt_needs_another_gp(void)
 188{
 189        return *rcu_preempt_ctrlblk.rcb.curtail != NULL;
 190}
 191
 192/*
 193 * Return true if a preemptible-RCU grace period is in progress.
 194 * The caller must disable hardirqs.
 195 */
 196static int rcu_preempt_gp_in_progress(void)
 197{
 198        return rcu_preempt_ctrlblk.completed != rcu_preempt_ctrlblk.gpnum;
 199}
 200
 201/*
 202 * Advance a ->blkd_tasks-list pointer to the next entry, instead
 203 * returning NULL if at the end of the list.
 204 */
 205static struct list_head *rcu_next_node_entry(struct task_struct *t)
 206{
 207        struct list_head *np;
 208
 209        np = t->rcu_node_entry.next;
 210        if (np == &rcu_preempt_ctrlblk.blkd_tasks)
 211                np = NULL;
 212        return np;
 213}
 214
 215#ifdef CONFIG_RCU_TRACE
 216
 217#ifdef CONFIG_RCU_BOOST
 218static void rcu_initiate_boost_trace(void);
 219#endif /* #ifdef CONFIG_RCU_BOOST */
 220
 221/*
 222 * Dump additional statistice for TINY_PREEMPT_RCU.
 223 */
 224static void show_tiny_preempt_stats(struct seq_file *m)
 225{
 226        seq_printf(m, "rcu_preempt: qlen=%ld gp=%lu g%u/p%u/c%u tasks=%c%c%c\n",
 227                   rcu_preempt_ctrlblk.rcb.qlen,
 228                   rcu_preempt_ctrlblk.n_grace_periods,
 229                   rcu_preempt_ctrlblk.gpnum,
 230                   rcu_preempt_ctrlblk.gpcpu,
 231                   rcu_preempt_ctrlblk.completed,
 232                   "T."[list_empty(&rcu_preempt_ctrlblk.blkd_tasks)],
 233                   "N."[!rcu_preempt_ctrlblk.gp_tasks],
 234                   "E."[!rcu_preempt_ctrlblk.exp_tasks]);
 235#ifdef CONFIG_RCU_BOOST
 236        seq_printf(m, "%sttb=%c ntb=%lu neb=%lu nnb=%lu j=%04x bt=%04x\n",
 237                   "             ",
 238                   "B."[!rcu_preempt_ctrlblk.boost_tasks],
 239                   rcu_preempt_ctrlblk.n_tasks_boosted,
 240                   rcu_preempt_ctrlblk.n_exp_boosts,
 241                   rcu_preempt_ctrlblk.n_normal_boosts,
 242                   (int)(jiffies & 0xffff),
 243                   (int)(rcu_preempt_ctrlblk.boost_time & 0xffff));
 244        seq_printf(m, "%s: nt=%lu egt=%lu bt=%lu ny=%lu nos=%lu\n",
 245                   "             balk",
 246                   rcu_preempt_ctrlblk.n_balk_blkd_tasks,
 247                   rcu_preempt_ctrlblk.n_balk_exp_gp_tasks,
 248                   rcu_preempt_ctrlblk.n_balk_boost_tasks,
 249                   rcu_preempt_ctrlblk.n_balk_notyet,
 250                   rcu_preempt_ctrlblk.n_balk_nos);
 251#endif /* #ifdef CONFIG_RCU_BOOST */
 252}
 253
 254#endif /* #ifdef CONFIG_RCU_TRACE */
 255
 256#ifdef CONFIG_RCU_BOOST
 257
 258#include "rtmutex_common.h"
 259
 260#define RCU_BOOST_PRIO CONFIG_RCU_BOOST_PRIO
 261
 262/* Controls for rcu_kthread() kthread. */
 263static struct task_struct *rcu_kthread_task;
 264static DECLARE_WAIT_QUEUE_HEAD(rcu_kthread_wq);
 265static unsigned long have_rcu_kthread_work;
 266
 267/*
 268 * Carry out RCU priority boosting on the task indicated by ->boost_tasks,
 269 * and advance ->boost_tasks to the next task in the ->blkd_tasks list.
 270 */
 271static int rcu_boost(void)
 272{
 273        unsigned long flags;
 274        struct rt_mutex mtx;
 275        struct task_struct *t;
 276        struct list_head *tb;
 277
 278        if (rcu_preempt_ctrlblk.boost_tasks == NULL &&
 279            rcu_preempt_ctrlblk.exp_tasks == NULL)
 280                return 0;  /* Nothing to boost. */
 281
 282        raw_local_irq_save(flags);
 283
 284        /*
 285         * Recheck with irqs disabled: all tasks in need of boosting
 286         * might exit their RCU read-side critical sections on their own
 287         * if we are preempted just before disabling irqs.
 288         */
 289        if (rcu_preempt_ctrlblk.boost_tasks == NULL &&
 290            rcu_preempt_ctrlblk.exp_tasks == NULL) {
 291                raw_local_irq_restore(flags);
 292                return 0;
 293        }
 294
 295        /*
 296         * Preferentially boost tasks blocking expedited grace periods.
 297         * This cannot starve the normal grace periods because a second
 298         * expedited grace period must boost all blocked tasks, including
 299         * those blocking the pre-existing normal grace period.
 300         */
 301        if (rcu_preempt_ctrlblk.exp_tasks != NULL) {
 302                tb = rcu_preempt_ctrlblk.exp_tasks;
 303                RCU_TRACE(rcu_preempt_ctrlblk.n_exp_boosts++);
 304        } else {
 305                tb = rcu_preempt_ctrlblk.boost_tasks;
 306                RCU_TRACE(rcu_preempt_ctrlblk.n_normal_boosts++);
 307        }
 308        RCU_TRACE(rcu_preempt_ctrlblk.n_tasks_boosted++);
 309
 310        /*
 311         * We boost task t by manufacturing an rt_mutex that appears to
 312         * be held by task t.  We leave a pointer to that rt_mutex where
 313         * task t can find it, and task t will release the mutex when it
 314         * exits its outermost RCU read-side critical section.  Then
 315         * simply acquiring this artificial rt_mutex will boost task
 316         * t's priority.  (Thanks to tglx for suggesting this approach!)
 317         */
 318        t = container_of(tb, struct task_struct, rcu_node_entry);
 319        rt_mutex_init_proxy_locked(&mtx, t);
 320        t->rcu_boost_mutex = &mtx;
 321        raw_local_irq_restore(flags);
 322        rt_mutex_lock(&mtx);
 323        rt_mutex_unlock(&mtx);  /* Keep lockdep happy. */
 324
 325        return ACCESS_ONCE(rcu_preempt_ctrlblk.boost_tasks) != NULL ||
 326               ACCESS_ONCE(rcu_preempt_ctrlblk.exp_tasks) != NULL;
 327}
 328
 329/*
 330 * Check to see if it is now time to start boosting RCU readers blocking
 331 * the current grace period, and, if so, tell the rcu_kthread_task to
 332 * start boosting them.  If there is an expedited boost in progress,
 333 * we wait for it to complete.
 334 *
 335 * If there are no blocked readers blocking the current grace period,
 336 * return 0 to let the caller know, otherwise return 1.  Note that this
 337 * return value is independent of whether or not boosting was done.
 338 */
 339static int rcu_initiate_boost(void)
 340{
 341        if (!rcu_preempt_blocked_readers_cgp() &&
 342            rcu_preempt_ctrlblk.exp_tasks == NULL) {
 343                RCU_TRACE(rcu_preempt_ctrlblk.n_balk_exp_gp_tasks++);
 344                return 0;
 345        }
 346        if (rcu_preempt_ctrlblk.exp_tasks != NULL ||
 347            (rcu_preempt_ctrlblk.gp_tasks != NULL &&
 348             rcu_preempt_ctrlblk.boost_tasks == NULL &&
 349             ULONG_CMP_GE(jiffies, rcu_preempt_ctrlblk.boost_time))) {
 350                if (rcu_preempt_ctrlblk.exp_tasks == NULL)
 351                        rcu_preempt_ctrlblk.boost_tasks =
 352                                rcu_preempt_ctrlblk.gp_tasks;
 353                invoke_rcu_callbacks();
 354        } else
 355                RCU_TRACE(rcu_initiate_boost_trace());
 356        return 1;
 357}
 358
 359#define RCU_BOOST_DELAY_JIFFIES DIV_ROUND_UP(CONFIG_RCU_BOOST_DELAY * HZ, 1000)
 360
 361/*
 362 * Do priority-boost accounting for the start of a new grace period.
 363 */
 364static void rcu_preempt_boost_start_gp(void)
 365{
 366        rcu_preempt_ctrlblk.boost_time = jiffies + RCU_BOOST_DELAY_JIFFIES;
 367}
 368
 369#else /* #ifdef CONFIG_RCU_BOOST */
 370
 371/*
 372 * If there is no RCU priority boosting, we don't initiate boosting,
 373 * but we do indicate whether there are blocked readers blocking the
 374 * current grace period.
 375 */
 376static int rcu_initiate_boost(void)
 377{
 378        return rcu_preempt_blocked_readers_cgp();
 379}
 380
 381/*
 382 * If there is no RCU priority boosting, nothing to do at grace-period start.
 383 */
 384static void rcu_preempt_boost_start_gp(void)
 385{
 386}
 387
 388#endif /* else #ifdef CONFIG_RCU_BOOST */
 389
 390/*
 391 * Record a preemptible-RCU quiescent state for the specified CPU.  Note
 392 * that this just means that the task currently running on the CPU is
 393 * in a quiescent state.  There might be any number of tasks blocked
 394 * while in an RCU read-side critical section.
 395 *
 396 * Unlike the other rcu_*_qs() functions, callers to this function
 397 * must disable irqs in order to protect the assignment to
 398 * ->rcu_read_unlock_special.
 399 *
 400 * Because this is a single-CPU implementation, the only way a grace
 401 * period can end is if the CPU is in a quiescent state.  The reason is
 402 * that a blocked preemptible-RCU reader can exit its critical section
 403 * only if the CPU is running it at the time.  Therefore, when the
 404 * last task blocking the current grace period exits its RCU read-side
 405 * critical section, neither the CPU nor blocked tasks will be stopping
 406 * the current grace period.  (In contrast, SMP implementations
 407 * might have CPUs running in RCU read-side critical sections that
 408 * block later grace periods -- but this is not possible given only
 409 * one CPU.)
 410 */
 411static void rcu_preempt_cpu_qs(void)
 412{
 413        /* Record both CPU and task as having responded to current GP. */
 414        rcu_preempt_ctrlblk.gpcpu = rcu_preempt_ctrlblk.gpnum;
 415        current->rcu_read_unlock_special &= ~RCU_READ_UNLOCK_NEED_QS;
 416
 417        /* If there is no GP then there is nothing more to do.  */
 418        if (!rcu_preempt_gp_in_progress())
 419                return;
 420        /*
 421         * Check up on boosting.  If there are readers blocking the
 422         * current grace period, leave.
 423         */
 424        if (rcu_initiate_boost())
 425                return;
 426
 427        /* Advance callbacks. */
 428        rcu_preempt_ctrlblk.completed = rcu_preempt_ctrlblk.gpnum;
 429        rcu_preempt_ctrlblk.rcb.donetail = rcu_preempt_ctrlblk.rcb.curtail;
 430        rcu_preempt_ctrlblk.rcb.curtail = rcu_preempt_ctrlblk.nexttail;
 431
 432        /* If there are no blocked readers, next GP is done instantly. */
 433        if (!rcu_preempt_blocked_readers_any())
 434                rcu_preempt_ctrlblk.rcb.donetail = rcu_preempt_ctrlblk.nexttail;
 435
 436        /* If there are done callbacks, cause them to be invoked. */
 437        if (*rcu_preempt_ctrlblk.rcb.donetail != NULL)
 438                invoke_rcu_callbacks();
 439}
 440
 441/*
 442 * Start a new RCU grace period if warranted.  Hard irqs must be disabled.
 443 */
 444static void rcu_preempt_start_gp(void)
 445{
 446        if (!rcu_preempt_gp_in_progress() && rcu_preempt_needs_another_gp()) {
 447
 448                /* Official start of GP. */
 449                rcu_preempt_ctrlblk.gpnum++;
 450                RCU_TRACE(rcu_preempt_ctrlblk.n_grace_periods++);
 451
 452                /* Any blocked RCU readers block new GP. */
 453                if (rcu_preempt_blocked_readers_any())
 454                        rcu_preempt_ctrlblk.gp_tasks =
 455                                rcu_preempt_ctrlblk.blkd_tasks.next;
 456
 457                /* Set up for RCU priority boosting. */
 458                rcu_preempt_boost_start_gp();
 459
 460                /* If there is no running reader, CPU is done with GP. */
 461                if (!rcu_preempt_running_reader())
 462                        rcu_preempt_cpu_qs();
 463        }
 464}
 465
 466/*
 467 * We have entered the scheduler, and the current task might soon be
 468 * context-switched away from.  If this task is in an RCU read-side
 469 * critical section, we will no longer be able to rely on the CPU to
 470 * record that fact, so we enqueue the task on the blkd_tasks list.
 471 * If the task started after the current grace period began, as recorded
 472 * by ->gpcpu, we enqueue at the beginning of the list.  Otherwise
 473 * before the element referenced by ->gp_tasks (or at the tail if
 474 * ->gp_tasks is NULL) and point ->gp_tasks at the newly added element.
 475 * The task will dequeue itself when it exits the outermost enclosing
 476 * RCU read-side critical section.  Therefore, the current grace period
 477 * cannot be permitted to complete until the ->gp_tasks pointer becomes
 478 * NULL.
 479 *
 480 * Caller must disable preemption.
 481 */
 482void rcu_preempt_note_context_switch(void)
 483{
 484        struct task_struct *t = current;
 485        unsigned long flags;
 486
 487        local_irq_save(flags); /* must exclude scheduler_tick(). */
 488        if (rcu_preempt_running_reader() > 0 &&
 489            (t->rcu_read_unlock_special & RCU_READ_UNLOCK_BLOCKED) == 0) {
 490
 491                /* Possibly blocking in an RCU read-side critical section. */
 492                t->rcu_read_unlock_special |= RCU_READ_UNLOCK_BLOCKED;
 493
 494                /*
 495                 * If this CPU has already checked in, then this task
 496                 * will hold up the next grace period rather than the
 497                 * current grace period.  Queue the task accordingly.
 498                 * If the task is queued for the current grace period
 499                 * (i.e., this CPU has not yet passed through a quiescent
 500                 * state for the current grace period), then as long
 501                 * as that task remains queued, the current grace period
 502                 * cannot end.
 503                 */
 504                list_add(&t->rcu_node_entry, &rcu_preempt_ctrlblk.blkd_tasks);
 505                if (rcu_cpu_blocking_cur_gp())
 506                        rcu_preempt_ctrlblk.gp_tasks = &t->rcu_node_entry;
 507        } else if (rcu_preempt_running_reader() < 0 &&
 508                   t->rcu_read_unlock_special) {
 509                /*
 510                 * Complete exit from RCU read-side critical section on
 511                 * behalf of preempted instance of __rcu_read_unlock().
 512                 */
 513                rcu_read_unlock_special(t);
 514        }
 515
 516        /*
 517         * Either we were not in an RCU read-side critical section to
 518         * begin with, or we have now recorded that critical section
 519         * globally.  Either way, we can now note a quiescent state
 520         * for this CPU.  Again, if we were in an RCU read-side critical
 521         * section, and if that critical section was blocking the current
 522         * grace period, then the fact that the task has been enqueued
 523         * means that current grace period continues to be blocked.
 524         */
 525        rcu_preempt_cpu_qs();
 526        local_irq_restore(flags);
 527}
 528
 529/*
 530 * Tiny-preemptible RCU implementation for rcu_read_lock().
 531 * Just increment ->rcu_read_lock_nesting, shared state will be updated
 532 * if we block.
 533 */
 534void __rcu_read_lock(void)
 535{
 536        current->rcu_read_lock_nesting++;
 537        barrier();  /* needed if we ever invoke rcu_read_lock in rcutiny.c */
 538}
 539EXPORT_SYMBOL_GPL(__rcu_read_lock);
 540
 541/*
 542 * Handle special cases during rcu_read_unlock(), such as needing to
 543 * notify RCU core processing or task having blocked during the RCU
 544 * read-side critical section.
 545 */
 546static noinline void rcu_read_unlock_special(struct task_struct *t)
 547{
 548        int empty;
 549        int empty_exp;
 550        unsigned long flags;
 551        struct list_head *np;
 552#ifdef CONFIG_RCU_BOOST
 553        struct rt_mutex *rbmp = NULL;
 554#endif /* #ifdef CONFIG_RCU_BOOST */
 555        int special;
 556
 557        /*
 558         * NMI handlers cannot block and cannot safely manipulate state.
 559         * They therefore cannot possibly be special, so just leave.
 560         */
 561        if (in_nmi())
 562                return;
 563
 564        local_irq_save(flags);
 565
 566        /*
 567         * If RCU core is waiting for this CPU to exit critical section,
 568         * let it know that we have done so.
 569         */
 570        special = t->rcu_read_unlock_special;
 571        if (special & RCU_READ_UNLOCK_NEED_QS)
 572                rcu_preempt_cpu_qs();
 573
 574        /* Hardware IRQ handlers cannot block. */
 575        if (in_irq() || in_serving_softirq()) {
 576                local_irq_restore(flags);
 577                return;
 578        }
 579
 580        /* Clean up if blocked during RCU read-side critical section. */
 581        if (special & RCU_READ_UNLOCK_BLOCKED) {
 582                t->rcu_read_unlock_special &= ~RCU_READ_UNLOCK_BLOCKED;
 583
 584                /*
 585                 * Remove this task from the ->blkd_tasks list and adjust
 586                 * any pointers that might have been referencing it.
 587                 */
 588                empty = !rcu_preempt_blocked_readers_cgp();
 589                empty_exp = rcu_preempt_ctrlblk.exp_tasks == NULL;
 590                np = rcu_next_node_entry(t);
 591                list_del_init(&t->rcu_node_entry);
 592                if (&t->rcu_node_entry == rcu_preempt_ctrlblk.gp_tasks)
 593                        rcu_preempt_ctrlblk.gp_tasks = np;
 594                if (&t->rcu_node_entry == rcu_preempt_ctrlblk.exp_tasks)
 595                        rcu_preempt_ctrlblk.exp_tasks = np;
 596#ifdef CONFIG_RCU_BOOST
 597                if (&t->rcu_node_entry == rcu_preempt_ctrlblk.boost_tasks)
 598                        rcu_preempt_ctrlblk.boost_tasks = np;
 599#endif /* #ifdef CONFIG_RCU_BOOST */
 600
 601                /*
 602                 * If this was the last task on the current list, and if
 603                 * we aren't waiting on the CPU, report the quiescent state
 604                 * and start a new grace period if needed.
 605                 */
 606                if (!empty && !rcu_preempt_blocked_readers_cgp()) {
 607                        rcu_preempt_cpu_qs();
 608                        rcu_preempt_start_gp();
 609                }
 610
 611                /*
 612                 * If this was the last task on the expedited lists,
 613                 * then we need wake up the waiting task.
 614                 */
 615                if (!empty_exp && rcu_preempt_ctrlblk.exp_tasks == NULL)
 616                        rcu_report_exp_done();
 617        }
 618#ifdef CONFIG_RCU_BOOST
 619        /* Unboost self if was boosted. */
 620        if (t->rcu_boost_mutex != NULL) {
 621                rbmp = t->rcu_boost_mutex;
 622                t->rcu_boost_mutex = NULL;
 623                rt_mutex_unlock(rbmp);
 624        }
 625#endif /* #ifdef CONFIG_RCU_BOOST */
 626        local_irq_restore(flags);
 627}
 628
 629/*
 630 * Tiny-preemptible RCU implementation for rcu_read_unlock().
 631 * Decrement ->rcu_read_lock_nesting.  If the result is zero (outermost
 632 * rcu_read_unlock()) and ->rcu_read_unlock_special is non-zero, then
 633 * invoke rcu_read_unlock_special() to clean up after a context switch
 634 * in an RCU read-side critical section and other special cases.
 635 */
 636void __rcu_read_unlock(void)
 637{
 638        struct task_struct *t = current;
 639
 640        barrier();  /* needed if we ever invoke rcu_read_unlock in rcutiny.c */
 641        if (t->rcu_read_lock_nesting != 1)
 642                --t->rcu_read_lock_nesting;
 643        else {
 644                t->rcu_read_lock_nesting = INT_MIN;
 645                barrier();  /* assign before ->rcu_read_unlock_special load */
 646                if (unlikely(ACCESS_ONCE(t->rcu_read_unlock_special)))
 647                        rcu_read_unlock_special(t);
 648                barrier();  /* ->rcu_read_unlock_special load before assign */
 649                t->rcu_read_lock_nesting = 0;
 650        }
 651#ifdef CONFIG_PROVE_LOCKING
 652        {
 653                int rrln = ACCESS_ONCE(t->rcu_read_lock_nesting);
 654
 655                WARN_ON_ONCE(rrln < 0 && rrln > INT_MIN / 2);
 656        }
 657#endif /* #ifdef CONFIG_PROVE_LOCKING */
 658}
 659EXPORT_SYMBOL_GPL(__rcu_read_unlock);
 660
 661/*
 662 * Check for a quiescent state from the current CPU.  When a task blocks,
 663 * the task is recorded in the rcu_preempt_ctrlblk structure, which is
 664 * checked elsewhere.  This is called from the scheduling-clock interrupt.
 665 *
 666 * Caller must disable hard irqs.
 667 */
 668static void rcu_preempt_check_callbacks(void)
 669{
 670        struct task_struct *t = current;
 671
 672        if (rcu_preempt_gp_in_progress() &&
 673            (!rcu_preempt_running_reader() ||
 674             !rcu_cpu_blocking_cur_gp()))
 675                rcu_preempt_cpu_qs();
 676        if (&rcu_preempt_ctrlblk.rcb.rcucblist !=
 677            rcu_preempt_ctrlblk.rcb.donetail)
 678                invoke_rcu_callbacks();
 679        if (rcu_preempt_gp_in_progress() &&
 680            rcu_cpu_blocking_cur_gp() &&
 681            rcu_preempt_running_reader() > 0)
 682                t->rcu_read_unlock_special |= RCU_READ_UNLOCK_NEED_QS;
 683}
 684
 685/*
 686 * TINY_PREEMPT_RCU has an extra callback-list tail pointer to
 687 * update, so this is invoked from rcu_process_callbacks() to
 688 * handle that case.  Of course, it is invoked for all flavors of
 689 * RCU, but RCU callbacks can appear only on one of the lists, and
 690 * neither ->nexttail nor ->donetail can possibly be NULL, so there
 691 * is no need for an explicit check.
 692 */
 693static void rcu_preempt_remove_callbacks(struct rcu_ctrlblk *rcp)
 694{
 695        if (rcu_preempt_ctrlblk.nexttail == rcp->donetail)
 696                rcu_preempt_ctrlblk.nexttail = &rcp->rcucblist;
 697}
 698
 699/*
 700 * Process callbacks for preemptible RCU.
 701 */
 702static void rcu_preempt_process_callbacks(void)
 703{
 704        __rcu_process_callbacks(&rcu_preempt_ctrlblk.rcb);
 705}
 706
 707/*
 708 * Queue a preemptible -RCU callback for invocation after a grace period.
 709 */
 710void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
 711{
 712        unsigned long flags;
 713
 714        debug_rcu_head_queue(head);
 715        head->func = func;
 716        head->next = NULL;
 717
 718        local_irq_save(flags);
 719        *rcu_preempt_ctrlblk.nexttail = head;
 720        rcu_preempt_ctrlblk.nexttail = &head->next;
 721        RCU_TRACE(rcu_preempt_ctrlblk.rcb.qlen++);
 722        rcu_preempt_start_gp();  /* checks to see if GP needed. */
 723        local_irq_restore(flags);
 724}
 725EXPORT_SYMBOL_GPL(call_rcu);
 726
 727/*
 728 * synchronize_rcu - wait until a grace period has elapsed.
 729 *
 730 * Control will return to the caller some time after a full grace
 731 * period has elapsed, in other words after all currently executing RCU
 732 * read-side critical sections have completed.  RCU read-side critical
 733 * sections are delimited by rcu_read_lock() and rcu_read_unlock(),
 734 * and may be nested.
 735 */
 736void synchronize_rcu(void)
 737{
 738        rcu_lockdep_assert(!lock_is_held(&rcu_bh_lock_map) &&
 739                           !lock_is_held(&rcu_lock_map) &&
 740                           !lock_is_held(&rcu_sched_lock_map),
 741                           "Illegal synchronize_rcu() in RCU read-side critical section");
 742
 743#ifdef CONFIG_DEBUG_LOCK_ALLOC
 744        if (!rcu_scheduler_active)
 745                return;
 746#endif /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
 747
 748        WARN_ON_ONCE(rcu_preempt_running_reader());
 749        if (!rcu_preempt_blocked_readers_any())
 750                return;
 751
 752        /* Once we get past the fastpath checks, same code as rcu_barrier(). */
 753        rcu_barrier();
 754}
 755EXPORT_SYMBOL_GPL(synchronize_rcu);
 756
 757static DECLARE_WAIT_QUEUE_HEAD(sync_rcu_preempt_exp_wq);
 758static unsigned long sync_rcu_preempt_exp_count;
 759static DEFINE_MUTEX(sync_rcu_preempt_exp_mutex);
 760
 761/*
 762 * Return non-zero if there are any tasks in RCU read-side critical
 763 * sections blocking the current preemptible-RCU expedited grace period.
 764 * If there is no preemptible-RCU expedited grace period currently in
 765 * progress, returns zero unconditionally.
 766 */
 767static int rcu_preempted_readers_exp(void)
 768{
 769        return rcu_preempt_ctrlblk.exp_tasks != NULL;
 770}
 771
 772/*
 773 * Report the exit from RCU read-side critical section for the last task
 774 * that queued itself during or before the current expedited preemptible-RCU
 775 * grace period.
 776 */
 777static void rcu_report_exp_done(void)
 778{
 779        wake_up(&sync_rcu_preempt_exp_wq);
 780}
 781
 782/*
 783 * Wait for an rcu-preempt grace period, but expedite it.  The basic idea
 784 * is to rely in the fact that there is but one CPU, and that it is
 785 * illegal for a task to invoke synchronize_rcu_expedited() while in a
 786 * preemptible-RCU read-side critical section.  Therefore, any such
 787 * critical sections must correspond to blocked tasks, which must therefore
 788 * be on the ->blkd_tasks list.  So just record the current head of the
 789 * list in the ->exp_tasks pointer, and wait for all tasks including and
 790 * after the task pointed to by ->exp_tasks to drain.
 791 */
 792void synchronize_rcu_expedited(void)
 793{
 794        unsigned long flags;
 795        struct rcu_preempt_ctrlblk *rpcp = &rcu_preempt_ctrlblk;
 796        unsigned long snap;
 797
 798        barrier(); /* ensure prior action seen before grace period. */
 799
 800        WARN_ON_ONCE(rcu_preempt_running_reader());
 801
 802        /*
 803         * Acquire lock so that there is only one preemptible RCU grace
 804         * period in flight.  Of course, if someone does the expedited
 805         * grace period for us while we are acquiring the lock, just leave.
 806         */
 807        snap = sync_rcu_preempt_exp_count + 1;
 808        mutex_lock(&sync_rcu_preempt_exp_mutex);
 809        if (ULONG_CMP_LT(snap, sync_rcu_preempt_exp_count))
 810                goto unlock_mb_ret; /* Others did our work for us. */
 811
 812        local_irq_save(flags);
 813
 814        /*
 815         * All RCU readers have to already be on blkd_tasks because
 816         * we cannot legally be executing in an RCU read-side critical
 817         * section.
 818         */
 819
 820        /* Snapshot current head of ->blkd_tasks list. */
 821        rpcp->exp_tasks = rpcp->blkd_tasks.next;
 822        if (rpcp->exp_tasks == &rpcp->blkd_tasks)
 823                rpcp->exp_tasks = NULL;
 824
 825        /* Wait for tail of ->blkd_tasks list to drain. */
 826        if (!rcu_preempted_readers_exp())
 827                local_irq_restore(flags);
 828        else {
 829                rcu_initiate_boost();
 830                local_irq_restore(flags);
 831                wait_event(sync_rcu_preempt_exp_wq,
 832                           !rcu_preempted_readers_exp());
 833        }
 834
 835        /* Clean up and exit. */
 836        barrier(); /* ensure expedited GP seen before counter increment. */
 837        sync_rcu_preempt_exp_count++;
 838unlock_mb_ret:
 839        mutex_unlock(&sync_rcu_preempt_exp_mutex);
 840        barrier(); /* ensure subsequent action seen after grace period. */
 841}
 842EXPORT_SYMBOL_GPL(synchronize_rcu_expedited);
 843
 844/*
 845 * Does preemptible RCU need the CPU to stay out of dynticks mode?
 846 */
 847int rcu_preempt_needs_cpu(void)
 848{
 849        if (!rcu_preempt_running_reader())
 850                rcu_preempt_cpu_qs();
 851        return rcu_preempt_ctrlblk.rcb.rcucblist != NULL;
 852}
 853
 854#else /* #ifdef CONFIG_TINY_PREEMPT_RCU */
 855
 856#ifdef CONFIG_RCU_TRACE
 857
 858/*
 859 * Because preemptible RCU does not exist, it is not necessary to
 860 * dump out its statistics.
 861 */
 862static void show_tiny_preempt_stats(struct seq_file *m)
 863{
 864}
 865
 866#endif /* #ifdef CONFIG_RCU_TRACE */
 867
 868/*
 869 * Because preemptible RCU does not exist, it never has any callbacks
 870 * to check.
 871 */
 872static void rcu_preempt_check_callbacks(void)
 873{
 874}
 875
 876/*
 877 * Because preemptible RCU does not exist, it never has any callbacks
 878 * to remove.
 879 */
 880static void rcu_preempt_remove_callbacks(struct rcu_ctrlblk *rcp)
 881{
 882}
 883
 884/*
 885 * Because preemptible RCU does not exist, it never has any callbacks
 886 * to process.
 887 */
 888static void rcu_preempt_process_callbacks(void)
 889{
 890}
 891
 892#endif /* #else #ifdef CONFIG_TINY_PREEMPT_RCU */
 893
 894#ifdef CONFIG_RCU_BOOST
 895
 896/*
 897 * Wake up rcu_kthread() to process callbacks now eligible for invocation
 898 * or to boost readers.
 899 */
 900static void invoke_rcu_callbacks(void)
 901{
 902        have_rcu_kthread_work = 1;
 903        if (rcu_kthread_task != NULL)
 904                wake_up(&rcu_kthread_wq);
 905}
 906
 907#ifdef CONFIG_RCU_TRACE
 908
 909/*
 910 * Is the current CPU running the RCU-callbacks kthread?
 911 * Caller must have preemption disabled.
 912 */
 913static bool rcu_is_callbacks_kthread(void)
 914{
 915        return rcu_kthread_task == current;
 916}
 917
 918#endif /* #ifdef CONFIG_RCU_TRACE */
 919
 920/*
 921 * This kthread invokes RCU callbacks whose grace periods have
 922 * elapsed.  It is awakened as needed, and takes the place of the
 923 * RCU_SOFTIRQ that is used for this purpose when boosting is disabled.
 924 * This is a kthread, but it is never stopped, at least not until
 925 * the system goes down.
 926 */
 927static int rcu_kthread(void *arg)
 928{
 929        unsigned long work;
 930        unsigned long morework;
 931        unsigned long flags;
 932
 933        for (;;) {
 934                wait_event_interruptible(rcu_kthread_wq,
 935                                         have_rcu_kthread_work != 0);
 936                morework = rcu_boost();
 937                local_irq_save(flags);
 938                work = have_rcu_kthread_work;
 939                have_rcu_kthread_work = morework;
 940                local_irq_restore(flags);
 941                if (work)
 942                        rcu_process_callbacks(NULL);
 943                schedule_timeout_interruptible(1); /* Leave CPU for others. */
 944        }
 945
 946        return 0;  /* Not reached, but needed to shut gcc up. */
 947}
 948
 949/*
 950 * Spawn the kthread that invokes RCU callbacks.
 951 */
 952static int __init rcu_spawn_kthreads(void)
 953{
 954        struct sched_param sp;
 955
 956        rcu_kthread_task = kthread_run(rcu_kthread, NULL, "rcu_kthread");
 957        sp.sched_priority = RCU_BOOST_PRIO;
 958        sched_setscheduler_nocheck(rcu_kthread_task, SCHED_FIFO, &sp);
 959        return 0;
 960}
 961early_initcall(rcu_spawn_kthreads);
 962
 963#else /* #ifdef CONFIG_RCU_BOOST */
 964
 965/* Hold off callback invocation until early_initcall() time. */
 966static int rcu_scheduler_fully_active __read_mostly;
 967
 968/*
 969 * Start up softirq processing of callbacks.
 970 */
 971void invoke_rcu_callbacks(void)
 972{
 973        if (rcu_scheduler_fully_active)
 974                raise_softirq(RCU_SOFTIRQ);
 975}
 976
 977#ifdef CONFIG_RCU_TRACE
 978
 979/*
 980 * There is no callback kthread, so this thread is never it.
 981 */
 982static bool rcu_is_callbacks_kthread(void)
 983{
 984        return false;
 985}
 986
 987#endif /* #ifdef CONFIG_RCU_TRACE */
 988
 989static int __init rcu_scheduler_really_started(void)
 990{
 991        rcu_scheduler_fully_active = 1;
 992        open_softirq(RCU_SOFTIRQ, rcu_process_callbacks);
 993        raise_softirq(RCU_SOFTIRQ);  /* Invoke any callbacks from early boot. */
 994        return 0;
 995}
 996early_initcall(rcu_scheduler_really_started);
 997
 998#endif /* #else #ifdef CONFIG_RCU_BOOST */
 999
1000#ifdef CONFIG_DEBUG_LOCK_ALLOC
1001#include <linux/kernel_stat.h>
1002
1003/*
1004 * During boot, we forgive RCU lockdep issues.  After this function is
1005 * invoked, we start taking RCU lockdep issues seriously.
1006 */
1007void __init rcu_scheduler_starting(void)
1008{
1009        WARN_ON(nr_context_switches() > 0);
1010        rcu_scheduler_active = 1;
1011}
1012
1013#endif /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
1014
1015#ifdef CONFIG_RCU_TRACE
1016
1017#ifdef CONFIG_RCU_BOOST
1018
1019static void rcu_initiate_boost_trace(void)
1020{
1021        if (list_empty(&rcu_preempt_ctrlblk.blkd_tasks))
1022                rcu_preempt_ctrlblk.n_balk_blkd_tasks++;
1023        else if (rcu_preempt_ctrlblk.gp_tasks == NULL &&
1024                 rcu_preempt_ctrlblk.exp_tasks == NULL)
1025                rcu_preempt_ctrlblk.n_balk_exp_gp_tasks++;
1026        else if (rcu_preempt_ctrlblk.boost_tasks != NULL)
1027                rcu_preempt_ctrlblk.n_balk_boost_tasks++;
1028        else if (!ULONG_CMP_GE(jiffies, rcu_preempt_ctrlblk.boost_time))
1029                rcu_preempt_ctrlblk.n_balk_notyet++;
1030        else
1031                rcu_preempt_ctrlblk.n_balk_nos++;
1032}
1033
1034#endif /* #ifdef CONFIG_RCU_BOOST */
1035
1036static void rcu_trace_sub_qlen(struct rcu_ctrlblk *rcp, int n)
1037{
1038        unsigned long flags;
1039
1040        raw_local_irq_save(flags);
1041        rcp->qlen -= n;
1042        raw_local_irq_restore(flags);
1043}
1044
1045/*
1046 * Dump statistics for TINY_RCU, such as they are.
1047 */
1048static int show_tiny_stats(struct seq_file *m, void *unused)
1049{
1050        show_tiny_preempt_stats(m);
1051        seq_printf(m, "rcu_sched: qlen: %ld\n", rcu_sched_ctrlblk.qlen);
1052        seq_printf(m, "rcu_bh: qlen: %ld\n", rcu_bh_ctrlblk.qlen);
1053        return 0;
1054}
1055
1056static int show_tiny_stats_open(struct inode *inode, struct file *file)
1057{
1058        return single_open(file, show_tiny_stats, NULL);
1059}
1060
1061static const struct file_operations show_tiny_stats_fops = {
1062        .owner = THIS_MODULE,
1063        .open = show_tiny_stats_open,
1064        .read = seq_read,
1065        .llseek = seq_lseek,
1066        .release = single_release,
1067};
1068
1069static struct dentry *rcudir;
1070
1071static int __init rcutiny_trace_init(void)
1072{
1073        struct dentry *retval;
1074
1075        rcudir = debugfs_create_dir("rcu", NULL);
1076        if (!rcudir)
1077                goto free_out;
1078        retval = debugfs_create_file("rcudata", 0444, rcudir,
1079                                     NULL, &show_tiny_stats_fops);
1080        if (!retval)
1081                goto free_out;
1082        return 0;
1083free_out:
1084        debugfs_remove_recursive(rcudir);
1085        return 1;
1086}
1087
1088static void __exit rcutiny_trace_cleanup(void)
1089{
1090        debugfs_remove_recursive(rcudir);
1091}
1092
1093module_init(rcutiny_trace_init);
1094module_exit(rcutiny_trace_cleanup);
1095
1096MODULE_AUTHOR("Paul E. McKenney");
1097MODULE_DESCRIPTION("Read-Copy Update tracing for tiny implementation");
1098MODULE_LICENSE("GPL");
1099
1100#endif /* #ifdef CONFIG_RCU_TRACE */
1101
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.