kernel/rcutree_plugin.h

   1 /*
   2  * Read-Copy Update mechanism for mutual exclusion (tree-based version)
   3  * Internal non-public definitions that provide either classic
   4  * or preemptable 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 Red Hat, 2009
  21  * Copyright IBM Corporation, 2009
  22  *
  23  * Author: Ingo Molnar <mingo@elte.hu>
  24  *         Paul E. McKenney <paulmck@linux.vnet.ibm.com>
  25  */
  26
  27
  28 #ifdef CONFIG_TREE_PREEMPT_RCU
  29
  30 struct rcu_state rcu_preempt_state = RCU_STATE_INITIALIZER(rcu_preempt_state);
  31 DEFINE_PER_CPU(struct rcu_data, rcu_preempt_data);
  32
  33 /*
  34  * Tell them what RCU they are running.
  35  */
  36 static inline void rcu_bootup_announce(void)
  37 {
  38         printk(KERN_INFO
  39                "Experimental preemptable hierarchical RCU implementation.\n");
  40 }
  41
  42 /*
  43  * Return the number of RCU-preempt batches processed thus far
  44  * for debug and statistics.
  45  */
  46 long rcu_batches_completed_preempt(void)
  47 {
  48         return rcu_preempt_state.completed;
  49 }
  50 EXPORT_SYMBOL_GPL(rcu_batches_completed_preempt);
  51
  52 /*
  53  * Return the number of RCU batches processed thus far for debug & stats.
  54  */
  55 long rcu_batches_completed(void)
  56 {
  57         return rcu_batches_completed_preempt();
  58 }
  59 EXPORT_SYMBOL_GPL(rcu_batches_completed);
  60
  61 /*
  62  * Record a preemptable-RCU quiescent state for the specified CPU.  Note
  63  * that this just means that the task currently running on the CPU is
  64  * not in a quiescent state.  There might be any number of tasks blocked
  65  * while in an RCU read-side critical section.
  66  */
  67 static void rcu_preempt_qs(int cpu)
  68 {
  69         struct rcu_data *rdp = &per_cpu(rcu_preempt_data, cpu);
  70         rdp->passed_quiesc_completed = rdp->completed;
  71         barrier();
  72         rdp->passed_quiesc = 1;
  73 }
  74
  75 /*
  76  * We have entered the scheduler, and the current task might soon be
  77  * context-switched away from.  If this task is in an RCU read-side
  78  * critical section, we will no longer be able to rely on the CPU to
  79  * record that fact, so we enqueue the task on the appropriate entry
  80  * of the blocked_tasks[] array.  The task will dequeue itself when
  81  * it exits the outermost enclosing RCU read-side critical section.
  82  * Therefore, the current grace period cannot be permitted to complete
  83  * until the blocked_tasks[] entry indexed by the low-order bit of
  84  * rnp->gpnum empties.
  85  *
  86  * Caller must disable preemption.
  87  */
  88 static void rcu_preempt_note_context_switch(int cpu)
  89 {
  90         struct task_struct *t = current;
  91         unsigned long flags;
  92         int phase;
  93         struct rcu_data *rdp;
  94         struct rcu_node *rnp;
  95
  96         if (t->rcu_read_lock_nesting &&
  97             (t->rcu_read_unlock_special & RCU_READ_UNLOCK_BLOCKED) == 0) {
  98
  99                 /* Possibly blocking in an RCU read-side critical section. */
 100                 rdp = rcu_preempt_state.rda[cpu];
 101                 rnp = rdp->mynode;
 102                 spin_lock_irqsave(&rnp->lock, flags);
 103                 t->rcu_read_unlock_special |= RCU_READ_UNLOCK_BLOCKED;
 104                 t->rcu_blocked_node = rnp;
 105
 106                 /*
 107                  * If this CPU has already checked in, then this task
 108                  * will hold up the next grace period rather than the
 109                  * current grace period.  Queue the task accordingly.
 110                  * If the task is queued for the current grace period
 111                  * (i.e., this CPU has not yet passed through a quiescent
 112                  * state for the current grace period), then as long
 113                  * as that task remains queued, the current grace period
 114                  * cannot end.
 115                  *
 116                  * But first, note that the current CPU must still be
 117                  * on line!
 118                  */
 119                 WARN_ON_ONCE((rdp->grpmask & rnp->qsmaskinit) == 0);
 120                 WARN_ON_ONCE(!list_empty(&t->rcu_node_entry));
 121                 phase = (rnp->gpnum + !(rnp->qsmask & rdp->grpmask)) & 0x1;
 122                 list_add(&t->rcu_node_entry, &rnp->blocked_tasks[phase]);
 123                 spin_unlock_irqrestore(&rnp->lock, flags);
 124         }
 125
 126         /*
 127          * Either we were not in an RCU read-side critical section to
 128          * begin with, or we have now recorded that critical section
 129          * globally.  Either way, we can now note a quiescent state
 130          * for this CPU.  Again, if we were in an RCU read-side critical
 131          * section, and if that critical section was blocking the current
 132          * grace period, then the fact that the task has been enqueued
 133          * means that we continue to block the current grace period.
 134          */
 135         rcu_preempt_qs(cpu);
 136         local_irq_save(flags);
 137         t->rcu_read_unlock_special &= ~RCU_READ_UNLOCK_NEED_QS;
 138         local_irq_restore(flags);
 139 }
 140
 141 /*
 142  * Tree-preemptable RCU implementation for rcu_read_lock().
 143  * Just increment ->rcu_read_lock_nesting, shared state will be updated
 144  * if we block.
 145  */
 146 void __rcu_read_lock(void)
 147 {
 148         ACCESS_ONCE(current->rcu_read_lock_nesting)++;
 149         barrier();  /* needed if we ever invoke rcu_read_lock in rcutree.c */
 150 }
 151 EXPORT_SYMBOL_GPL(__rcu_read_lock);
 152
 153 /*
 154  * Check for preempted RCU readers blocking the current grace period
 155  * for the specified rcu_node structure.  If the caller needs a reliable
 156  * answer, it must hold the rcu_node's ->lock.
 157  */
 158 static int rcu_preempted_readers(struct rcu_node *rnp)
 159 {
 160         return !list_empty(&rnp->blocked_tasks[rnp->gpnum & 0x1]);
 161 }
 162
 163 static void rcu_read_unlock_special(struct task_struct *t)
 164 {
 165         int empty;
 166         unsigned long flags;
 167         unsigned long mask;
 168         struct rcu_node *rnp;
 169         int special;
 170
 171         /* NMI handlers cannot block and cannot safely manipulate state. */
 172         if (in_nmi())
 173                 return;
 174
 175         local_irq_save(flags);
 176
 177         /*
 178          * If RCU core is waiting for this CPU to exit critical section,
 179          * let it know that we have done so.
 180          */
 181         special = t->rcu_read_unlock_special;
 182         if (special & RCU_READ_UNLOCK_NEED_QS) {
 183                 t->rcu_read_unlock_special &= ~RCU_READ_UNLOCK_NEED_QS;
 184                 rcu_preempt_qs(smp_processor_id());
 185         }
 186
 187         /* Hardware IRQ handlers cannot block. */
 188         if (in_irq()) {
 189                 local_irq_restore(flags);
 190                 return;
 191         }
 192
 193         /* Clean up if blocked during RCU read-side critical section. */
 194         if (special & RCU_READ_UNLOCK_BLOCKED) {
 195                 t->rcu_read_unlock_special &= ~RCU_READ_UNLOCK_BLOCKED;
 196
 197                 /*
 198                  * Remove this task from the list it blocked on.  The
 199                  * task can migrate while we acquire the lock, but at
 200                  * most one time.  So at most two passes through loop.
 201                  */
 202                 for (;;) {
 203                         rnp = t->rcu_blocked_node;
 204                         spin_lock(&rnp->lock);  /* irqs already disabled. */
 205                         if (rnp == t->rcu_blocked_node)
 206                                 break;
 207                         spin_unlock(&rnp->lock);  /* irqs remain disabled. */
 208                 }
 209                 empty = !rcu_preempted_readers(rnp);
 210                 list_del_init(&t->rcu_node_entry);
 211                 t->rcu_blocked_node = NULL;
 212
 213                 /*
 214                  * If this was the last task on the current list, and if
 215                  * we aren't waiting on any CPUs, report the quiescent state.
 216                  * Note that both cpu_quiet_msk_finish() and cpu_quiet_msk()
 217                  * drop rnp->lock and restore irq.
 218                  */
 219                 if (!empty && rnp->qsmask == 0 &&
 220                     !rcu_preempted_readers(rnp)) {
 221                         struct rcu_node *rnp_p;
 222
 223                         if (rnp->parent == NULL) {
 224                                 /* Only one rcu_node in the tree. */
 225                                 cpu_quiet_msk_finish(&rcu_preempt_state, flags);
 226                                 return;
 227                         }
 228                         /* Report up the rest of the hierarchy. */
 229                         mask = rnp->grpmask;
 230                         spin_unlock_irqrestore(&rnp->lock, flags);
 231                         rnp_p = rnp->parent;
 232                         spin_lock_irqsave(&rnp_p->lock, flags);
 233                         WARN_ON_ONCE(rnp->qsmask);
 234                         cpu_quiet_msk(mask, &rcu_preempt_state, rnp_p, flags);
 235                         return;
 236                 }
 237                 spin_unlock(&rnp->lock);
 238         }
 239         local_irq_restore(flags);
 240 }
 241
 242 /*
 243  * Tree-preemptable RCU implementation for rcu_read_unlock().
 244  * Decrement ->rcu_read_lock_nesting.  If the result is zero (outermost
 245  * rcu_read_unlock()) and ->rcu_read_unlock_special is non-zero, then
 246  * invoke rcu_read_unlock_special() to clean up after a context switch
 247  * in an RCU read-side critical section and other special cases.
 248  */
 249 void __rcu_read_unlock(void)
 250 {
 251         struct task_struct *t = current;
 252
 253         barrier();  /* needed if we ever invoke rcu_read_unlock in rcutree.c */
 254         if (--ACCESS_ONCE(t->rcu_read_lock_nesting) == 0 &&
 255             unlikely(ACCESS_ONCE(t->rcu_read_unlock_special)))
 256                 rcu_read_unlock_special(t);
 257 }
 258 EXPORT_SYMBOL_GPL(__rcu_read_unlock);
 259
 260 #ifdef CONFIG_RCU_CPU_STALL_DETECTOR
 261
 262 /*
 263  * Scan the current list of tasks blocked within RCU read-side critical
 264  * sections, printing out the tid of each.
 265  */
 266 static void rcu_print_task_stall(struct rcu_node *rnp)
 267 {
 268         unsigned long flags;
 269         struct list_head *lp;
 270         int phase;
 271         struct task_struct *t;
 272
 273         if (rcu_preempted_readers(rnp)) {
 274                 spin_lock_irqsave(&rnp->lock, flags);
 275                 phase = rnp->gpnum & 0x1;
 276                 lp = &rnp->blocked_tasks[phase];
 277                 list_for_each_entry(t, lp, rcu_node_entry)
 278                         printk(" P%d", t->pid);
 279                 spin_unlock_irqrestore(&rnp->lock, flags);
 280         }
 281 }
 282
 283 #endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
 284
 285 /*
 286  * Check that the list of blocked tasks for the newly completed grace
 287  * period is in fact empty.  It is a serious bug to complete a grace
 288  * period that still has RCU readers blocked!  This function must be
 289  * invoked -before- updating this rnp's ->gpnum, and the rnp's ->lock
 290  * must be held by the caller.
 291  */
 292 static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp)
 293 {
 294         WARN_ON_ONCE(rcu_preempted_readers(rnp));
 295         WARN_ON_ONCE(rnp->qsmask);
 296 }
 297
 298 #ifdef CONFIG_HOTPLUG_CPU
 299
 300 /*
 301  * Handle tasklist migration for case in which all CPUs covered by the
 302  * specified rcu_node have gone offline.  Move them up to the root
 303  * rcu_node.  The reason for not just moving them to the immediate
 304  * parent is to remove the need for rcu_read_unlock_special() to
 305  * make more than two attempts to acquire the target rcu_node's lock.
 306  *
 307  * Returns 1 if there was previously a task blocking the current grace
 308  * period on the specified rcu_node structure.
 309  *
 310  * The caller must hold rnp->lock with irqs disabled.
 311  */
 312 static int rcu_preempt_offline_tasks(struct rcu_state *rsp,
 313                                      struct rcu_node *rnp,
 314                                      struct rcu_data *rdp)
 315 {
 316         int i;
 317         struct list_head *lp;
 318         struct list_head *lp_root;
 319         int retval = rcu_preempted_readers(rnp);
 320         struct rcu_node *rnp_root = rcu_get_root(rsp);
 321         struct task_struct *tp;
 322
 323         if (rnp == rnp_root) {
 324                 WARN_ONCE(1, "Last CPU thought to be offlined?");
 325                 return 0;  /* Shouldn't happen: at least one CPU online. */
 326         }
 327         WARN_ON_ONCE(rnp != rdp->mynode &&
 328                      (!list_empty(&rnp->blocked_tasks[0]) ||
 329                       !list_empty(&rnp->blocked_tasks[1])));
 330
 331         /*
 332          * Move tasks up to root rcu_node.  Rely on the fact that the
 333          * root rcu_node can be at most one ahead of the rest of the
 334          * rcu_nodes in terms of gp_num value.  This fact allows us to
 335          * move the blocked_tasks[] array directly, element by element.
 336          */
 337         for (i = 0; i < 2; i++) {
 338                 lp = &rnp->blocked_tasks[i];
 339                 lp_root = &rnp_root->blocked_tasks[i];
 340                 while (!list_empty(lp)) {
 341                         tp = list_entry(lp->next, typeof(*tp), rcu_node_entry);
 342                         spin_lock(&rnp_root->lock); /* irqs already disabled */
 343                         list_del(&tp->rcu_node_entry);
 344                         tp->rcu_blocked_node = rnp_root;
 345                         list_add(&tp->rcu_node_entry, lp_root);
 346                         spin_unlock(&rnp_root->lock); /* irqs remain disabled */
 347                 }
 348         }
 349
 350         return retval;
 351 }
 352
 353 /*
 354  * Do CPU-offline processing for preemptable RCU.
 355  */
 356 static void rcu_preempt_offline_cpu(int cpu)
 357 {
 358         __rcu_offline_cpu(cpu, &rcu_preempt_state);
 359 }
 360
 361 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
 362
 363 /*
 364  * Check for a quiescent state from the current CPU.  When a task blocks,
 365  * the task is recorded in the corresponding CPU's rcu_node structure,
 366  * which is checked elsewhere.
 367  *
 368  * Caller must disable hard irqs.
 369  */
 370 static void rcu_preempt_check_callbacks(int cpu)
 371 {
 372         struct task_struct *t = current;
 373
 374         if (t->rcu_read_lock_nesting == 0) {
 375                 t->rcu_read_unlock_special &= ~RCU_READ_UNLOCK_NEED_QS;
 376                 rcu_preempt_qs(cpu);
 377                 return;
 378         }
 379         if (per_cpu(rcu_preempt_data, cpu).qs_pending)
 380                 t->rcu_read_unlock_special |= RCU_READ_UNLOCK_NEED_QS;
 381 }
 382
 383 /*
 384  * Process callbacks for preemptable RCU.
 385  */
 386 static void rcu_preempt_process_callbacks(void)
 387 {
 388         __rcu_process_callbacks(&rcu_preempt_state,
 389                                 &__get_cpu_var(rcu_preempt_data));
 390 }
 391
 392 /*
 393  * Queue a preemptable-RCU callback for invocation after a grace period.
 394  */
 395 void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
 396 {
 397         __call_rcu(head, func, &rcu_preempt_state);
 398 }
 399 EXPORT_SYMBOL_GPL(call_rcu);
 400
 401 /*
 402  * Wait for an rcu-preempt grace period.  We are supposed to expedite the
 403  * grace period, but this is the crude slow compatability hack, so just
 404  * invoke synchronize_rcu().
 405  */
 406 void synchronize_rcu_expedited(void)
 407 {
 408         synchronize_rcu();
 409 }
 410 EXPORT_SYMBOL_GPL(synchronize_rcu_expedited);
 411
 412 /*
 413  * Check to see if there is any immediate preemptable-RCU-related work
 414  * to be done.
 415  */
 416 static int rcu_preempt_pending(int cpu)
 417 {
 418         return __rcu_pending(&rcu_preempt_state,
 419                              &per_cpu(rcu_preempt_data, cpu));
 420 }
 421
 422 /*
 423  * Does preemptable RCU need the CPU to stay out of dynticks mode?
 424  */
 425 static int rcu_preempt_needs_cpu(int cpu)
 426 {
 427         return !!per_cpu(rcu_preempt_data, cpu).nxtlist;
 428 }
 429
 430 /**
 431  * rcu_barrier - Wait until all in-flight call_rcu() callbacks complete.
 432  */
 433 void rcu_barrier(void)
 434 {
 435         _rcu_barrier(&rcu_preempt_state, call_rcu);
 436 }
 437 EXPORT_SYMBOL_GPL(rcu_barrier);
 438
 439 /*
 440  * Initialize preemptable RCU's per-CPU data.
 441  */
 442 static void __cpuinit rcu_preempt_init_percpu_data(int cpu)
 443 {
 444         rcu_init_percpu_data(cpu, &rcu_preempt_state, 1);
 445 }
 446
 447 /*
 448  * Move preemptable RCU's callbacks to ->orphan_cbs_list.
 449  */
 450 static void rcu_preempt_send_cbs_to_orphanage(void)
 451 {
 452         rcu_send_cbs_to_orphanage(&rcu_preempt_state);
 453 }
 454
 455 /*
 456  * Initialize preemptable RCU's state structures.
 457  */
 458 static void __init __rcu_init_preempt(void)
 459 {
 460         RCU_INIT_FLAVOR(&rcu_preempt_state, rcu_preempt_data);
 461 }
 462
 463 /*
 464  * Check for a task exiting while in a preemptable-RCU read-side
 465  * critical section, clean up if so.  No need to issue warnings,
 466  * as debug_check_no_locks_held() already does this if lockdep
 467  * is enabled.
 468  */
 469 void exit_rcu(void)
 470 {
 471         struct task_struct *t = current;
 472
 473         if (t->rcu_read_lock_nesting == 0)
 474                 return;
 475         t->rcu_read_lock_nesting = 1;
 476         rcu_read_unlock();
 477 }
 478
 479 #else /* #ifdef CONFIG_TREE_PREEMPT_RCU */
 480
 481 /*
 482  * Tell them what RCU they are running.
 483  */
 484 static inline void rcu_bootup_announce(void)
 485 {
 486         printk(KERN_INFO "Hierarchical RCU implementation.\n");
 487 }
 488
 489 /*
 490  * Return the number of RCU batches processed thus far for debug & stats.
 491  */
 492 long rcu_batches_completed(void)
 493 {
 494         return rcu_batches_completed_sched();
 495 }
 496 EXPORT_SYMBOL_GPL(rcu_batches_completed);
 497
 498 /*
 499  * Because preemptable RCU does not exist, we never have to check for
 500  * CPUs being in quiescent states.
 501  */
 502 static void rcu_preempt_note_context_switch(int cpu)
 503 {
 504 }
 505
 506 /*
 507  * Because preemptable RCU does not exist, there are never any preempted
 508  * RCU readers.
 509  */
 510 static int rcu_preempted_readers(struct rcu_node *rnp)
 511 {
 512         return 0;
 513 }
 514
 515 #ifdef CONFIG_RCU_CPU_STALL_DETECTOR
 516
 517 /*
 518  * Because preemptable RCU does not exist, we never have to check for
 519  * tasks blocked within RCU read-side critical sections.
 520  */
 521 static void rcu_print_task_stall(struct rcu_node *rnp)
 522 {
 523 }
 524
 525 #endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
 526
 527 /*
 528  * Because there is no preemptable RCU, there can be no readers blocked,
 529  * so there is no need to check for blocked tasks.  So check only for
 530  * bogus qsmask values.
 531  */
 532 static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp)
 533 {
 534         WARN_ON_ONCE(rnp->qsmask);
 535 }
 536
 537 #ifdef CONFIG_HOTPLUG_CPU
 538
 539 /*
 540  * Because preemptable RCU does not exist, it never needs to migrate
 541  * tasks that were blocked within RCU read-side critical sections, and
 542  * such non-existent tasks cannot possibly have been blocking the current
 543  * grace period.
 544  */
 545 static int rcu_preempt_offline_tasks(struct rcu_state *rsp,
 546                                      struct rcu_node *rnp,
 547                                      struct rcu_data *rdp)
 548 {
 549         return 0;
 550 }
 551
 552 /*
 553  * Because preemptable RCU does not exist, it never needs CPU-offline
 554  * processing.
 555  */
 556 static void rcu_preempt_offline_cpu(int cpu)
 557 {
 558 }
 559
 560 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
 561
 562 /*
 563  * Because preemptable RCU does not exist, it never has any callbacks
 564  * to check.
 565  */
 566 static void rcu_preempt_check_callbacks(int cpu)
 567 {
 568 }
 569
 570 /*
 571  * Because preemptable RCU does not exist, it never has any callbacks
 572  * to process.
 573  */
 574 static void rcu_preempt_process_callbacks(void)
 575 {
 576 }
 577
 578 /*
 579  * In classic RCU, call_rcu() is just call_rcu_sched().
 580  */
 581 void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
 582 {
 583         call_rcu_sched(head, func);
 584 }
 585 EXPORT_SYMBOL_GPL(call_rcu);
 586
 587 /*
 588  * Wait for an rcu-preempt grace period, but make it happen quickly.
 589  * But because preemptable RCU does not exist, map to rcu-sched.
 590  */
 591 void synchronize_rcu_expedited(void)
 592 {
 593         synchronize_sched_expedited();
 594 }
 595 EXPORT_SYMBOL_GPL(synchronize_rcu_expedited);
 596
 597 /*
 598  * Because preemptable RCU does not exist, it never has any work to do.
 599  */
 600 static int rcu_preempt_pending(int cpu)
 601 {
 602         return 0;
 603 }
 604
 605 /*
 606  * Because preemptable RCU does not exist, it never needs any CPU.
 607  */
 608 static int rcu_preempt_needs_cpu(int cpu)
 609 {
 610         return 0;
 611 }
 612
 613 /*
 614  * Because preemptable RCU does not exist, rcu_barrier() is just
 615  * another name for rcu_barrier_sched().
 616  */
 617 void rcu_barrier(void)
 618 {
 619         rcu_barrier_sched();
 620 }
 621 EXPORT_SYMBOL_GPL(rcu_barrier);
 622
 623 /*
 624  * Because preemptable RCU does not exist, there is no per-CPU
 625  * data to initialize.
 626  */
 627 static void __cpuinit rcu_preempt_init_percpu_data(int cpu)
 628 {
 629 }
 630
 631 /*
 632  * Because there is no preemptable RCU, there are no callbacks to move.
 633  */
 634 static void rcu_preempt_send_cbs_to_orphanage(void)
 635 {
 636 }
 637
 638 /*
 639  * Because preemptable RCU does not exist, it need not be initialized.
 640  */
 641 static void __init __rcu_init_preempt(void)
 642 {
 643 }
 644
 645 #endif /* #else #ifdef CONFIG_TREE_PREEMPT_RCU */