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                 phase = !(rnp->qsmask & rdp->grpmask) ^ (rnp->gpnum & 0x1);
 121                 list_add(&t->rcu_node_entry, &rnp->blocked_tasks[phase]);
 122                 smp_mb();  /* Ensure later ctxt swtch seen after above. */
 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         t->rcu_read_unlock_special &= ~RCU_READ_UNLOCK_NEED_QS;
 137 }
 138
 139 /*
 140  * Tree-preemptable RCU implementation for rcu_read_lock().
 141  * Just increment ->rcu_read_lock_nesting, shared state will be updated
 142  * if we block.
 143  */
 144 void __rcu_read_lock(void)
 145 {
 146         ACCESS_ONCE(current->rcu_read_lock_nesting)++;
 147         barrier();  /* needed if we ever invoke rcu_read_lock in rcutree.c */
 148 }
 149 EXPORT_SYMBOL_GPL(__rcu_read_lock);
 150
 151 static void rcu_read_unlock_special(struct task_struct *t)
 152 {
 153         int empty;
 154         unsigned long flags;
 155         unsigned long mask;
 156         struct rcu_node *rnp;
 157         int special;
 158
 159         /* NMI handlers cannot block and cannot safely manipulate state. */
 160         if (in_nmi())
 161                 return;
 162
 163         local_irq_save(flags);
 164
 165         /*
 166          * If RCU core is waiting for this CPU to exit critical section,
 167          * let it know that we have done so.
 168          */
 169         special = t->rcu_read_unlock_special;
 170         if (special & RCU_READ_UNLOCK_NEED_QS) {
 171                 t->rcu_read_unlock_special &= ~RCU_READ_UNLOCK_NEED_QS;
 172                 rcu_preempt_qs(smp_processor_id());
 173         }
 174
 175         /* Hardware IRQ handlers cannot block. */
 176         if (in_irq()) {
 177                 local_irq_restore(flags);
 178                 return;
 179         }
 180
 181         /* Clean up if blocked during RCU read-side critical section. */
 182         if (special & RCU_READ_UNLOCK_BLOCKED) {
 183                 t->rcu_read_unlock_special &= ~RCU_READ_UNLOCK_BLOCKED;
 184
 185                 /*
 186                  * Remove this task from the list it blocked on.  The
 187                  * task can migrate while we acquire the lock, but at
 188                  * most one time.  So at most two passes through loop.
 189                  */
 190                 for (;;) {
 191                         rnp = t->rcu_blocked_node;
 192                         spin_lock(&rnp->lock);
 193                         if (rnp == t->rcu_blocked_node)
 194                                 break;
 195                         spin_unlock(&rnp->lock);
 196                 }
 197                 empty = list_empty(&rnp->blocked_tasks[rnp->gpnum & 0x1]);
 198                 list_del_init(&t->rcu_node_entry);
 199                 t->rcu_blocked_node = NULL;
 200
 201                 /*
 202                  * If this was the last task on the current list, and if
 203                  * we aren't waiting on any CPUs, report the quiescent state.
 204                  * Note that both cpu_quiet_msk_finish() and cpu_quiet_msk()
 205                  * drop rnp->lock and restore irq.
 206                  */
 207                 if (!empty && rnp->qsmask == 0 &&
 208                     list_empty(&rnp->blocked_tasks[rnp->gpnum & 0x1])) {
 209                         struct rcu_node *rnp_p;
 210
 211                         if (rnp->parent == NULL) {
 212                                 /* Only one rcu_node in the tree. */
 213                                 cpu_quiet_msk_finish(&rcu_preempt_state, flags);
 214                                 return;
 215                         }
 216                         /* Report up the rest of the hierarchy. */
 217                         mask = rnp->grpmask;
 218                         spin_unlock_irqrestore(&rnp->lock, flags);
 219                         rnp_p = rnp->parent;
 220                         spin_lock_irqsave(&rnp_p->lock, flags);
 221                         WARN_ON_ONCE(rnp->qsmask);
 222                         cpu_quiet_msk(mask, &rcu_preempt_state, rnp_p, flags);
 223                         return;
 224                 }
 225                 spin_unlock(&rnp->lock);
 226         }
 227         local_irq_restore(flags);
 228 }
 229
 230 /*
 231  * Tree-preemptable RCU implementation for rcu_read_unlock().
 232  * Decrement ->rcu_read_lock_nesting.  If the result is zero (outermost
 233  * rcu_read_unlock()) and ->rcu_read_unlock_special is non-zero, then
 234  * invoke rcu_read_unlock_special() to clean up after a context switch
 235  * in an RCU read-side critical section and other special cases.
 236  */
 237 void __rcu_read_unlock(void)
 238 {
 239         struct task_struct *t = current;
 240
 241         barrier();  /* needed if we ever invoke rcu_read_unlock in rcutree.c */
 242         if (--ACCESS_ONCE(t->rcu_read_lock_nesting) == 0 &&
 243             unlikely(ACCESS_ONCE(t->rcu_read_unlock_special)))
 244                 rcu_read_unlock_special(t);
 245 }
 246 EXPORT_SYMBOL_GPL(__rcu_read_unlock);
 247
 248 #ifdef CONFIG_RCU_CPU_STALL_DETECTOR
 249
 250 /*
 251  * Scan the current list of tasks blocked within RCU read-side critical
 252  * sections, printing out the tid of each.
 253  */
 254 static void rcu_print_task_stall(struct rcu_node *rnp)
 255 {
 256         unsigned long flags;
 257         struct list_head *lp;
 258         int phase = rnp->gpnum & 0x1;
 259         struct task_struct *t;
 260
 261         if (!list_empty(&rnp->blocked_tasks[phase])) {
 262                 spin_lock_irqsave(&rnp->lock, flags);
 263                 phase = rnp->gpnum & 0x1; /* re-read under lock. */
 264                 lp = &rnp->blocked_tasks[phase];
 265                 list_for_each_entry(t, lp, rcu_node_entry)
 266                         printk(" P%d", t->pid);
 267                 spin_unlock_irqrestore(&rnp->lock, flags);
 268         }
 269 }
 270
 271 #endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
 272
 273 /*
 274  * Check that the list of blocked tasks for the newly completed grace
 275  * period is in fact empty.  It is a serious bug to complete a grace
 276  * period that still has RCU readers blocked!  This function must be
 277  * invoked -before- updating this rnp's ->gpnum, and the rnp's ->lock
 278  * must be held by the caller.
 279  */
 280 static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp)
 281 {
 282         WARN_ON_ONCE(!list_empty(&rnp->blocked_tasks[rnp->gpnum & 0x1]));
 283         WARN_ON_ONCE(rnp->qsmask);
 284 }
 285
 286 /*
 287  * Check for preempted RCU readers for the specified rcu_node structure.
 288  * If the caller needs a reliable answer, it must hold the rcu_node's
 289  * >lock.
 290  */
 291 static int rcu_preempted_readers(struct rcu_node *rnp)
 292 {
 293         return !list_empty(&rnp->blocked_tasks[rnp->gpnum & 0x1]);
 294 }
 295
 296 #ifdef CONFIG_HOTPLUG_CPU
 297
 298 /*
 299  * Handle tasklist migration for case in which all CPUs covered by the
 300  * specified rcu_node have gone offline.  Move them up to the root
 301  * rcu_node.  The reason for not just moving them to the immediate
 302  * parent is to remove the need for rcu_read_unlock_special() to
 303  * make more than two attempts to acquire the target rcu_node's lock.
 304  *
 305  * The caller must hold rnp->lock with irqs disabled.
 306  */
 307 static void rcu_preempt_offline_tasks(struct rcu_state *rsp,
 308                                       struct rcu_node *rnp,
 309                                       struct rcu_data *rdp)
 310 {
 311         int i;
 312         struct list_head *lp;
 313         struct list_head *lp_root;
 314         struct rcu_node *rnp_root = rcu_get_root(rsp);
 315         struct task_struct *tp;
 316
 317         if (rnp == rnp_root) {
 318                 WARN_ONCE(1, "Last CPU thought to be offlined?");
 319                 return;  /* Shouldn't happen: at least one CPU online. */
 320         }
 321         WARN_ON_ONCE(rnp != rdp->mynode &&
 322                      (!list_empty(&rnp->blocked_tasks[0]) ||
 323                       !list_empty(&rnp->blocked_tasks[1])));
 324
 325         /*
 326          * Move tasks up to root rcu_node.  Rely on the fact that the
 327          * root rcu_node can be at most one ahead of the rest of the
 328          * rcu_nodes in terms of gp_num value.  This fact allows us to
 329          * move the blocked_tasks[] array directly, element by element.
 330          */
 331         for (i = 0; i < 2; i++) {
 332                 lp = &rnp->blocked_tasks[i];
 333                 lp_root = &rnp_root->blocked_tasks[i];
 334                 while (!list_empty(lp)) {
 335                         tp = list_entry(lp->next, typeof(*tp), rcu_node_entry);
 336                         spin_lock(&rnp_root->lock); /* irqs already disabled */
 337                         list_del(&tp->rcu_node_entry);
 338                         tp->rcu_blocked_node = rnp_root;
 339                         list_add(&tp->rcu_node_entry, lp_root);
 340                         spin_unlock(&rnp_root->lock); /* irqs remain disabled */
 341                 }
 342         }
 343 }
 344
 345 /*
 346  * Do CPU-offline processing for preemptable RCU.
 347  */
 348 static void rcu_preempt_offline_cpu(int cpu)
 349 {
 350         __rcu_offline_cpu(cpu, &rcu_preempt_state);
 351 }
 352
 353 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
 354
 355 /*
 356  * Check for a quiescent state from the current CPU.  When a task blocks,
 357  * the task is recorded in the corresponding CPU's rcu_node structure,
 358  * which is checked elsewhere.
 359  *
 360  * Caller must disable hard irqs.
 361  */
 362 static void rcu_preempt_check_callbacks(int cpu)
 363 {
 364         struct task_struct *t = current;
 365
 366         if (t->rcu_read_lock_nesting == 0) {
 367                 t->rcu_read_unlock_special &= ~RCU_READ_UNLOCK_NEED_QS;
 368                 rcu_preempt_qs(cpu);
 369                 return;
 370         }
 371         if (per_cpu(rcu_preempt_data, cpu).qs_pending) {
 372                 t->rcu_read_unlock_special |= RCU_READ_UNLOCK_NEED_QS;
 373         }
 374 }
 375
 376 /*
 377  * Process callbacks for preemptable RCU.
 378  */
 379 static void rcu_preempt_process_callbacks(void)
 380 {
 381         __rcu_process_callbacks(&rcu_preempt_state,
 382                                 &__get_cpu_var(rcu_preempt_data));
 383 }
 384
 385 /*
 386  * Queue a preemptable-RCU callback for invocation after a grace period.
 387  */
 388 void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
 389 {
 390         __call_rcu(head, func, &rcu_preempt_state);
 391 }
 392 EXPORT_SYMBOL_GPL(call_rcu);
 393
 394 /*
 395  * Check to see if there is any immediate preemptable-RCU-related work
 396  * to be done.
 397  */
 398 static int rcu_preempt_pending(int cpu)
 399 {
 400         return __rcu_pending(&rcu_preempt_state,
 401                              &per_cpu(rcu_preempt_data, cpu));
 402 }
 403
 404 /*
 405  * Does preemptable RCU need the CPU to stay out of dynticks mode?
 406  */
 407 static int rcu_preempt_needs_cpu(int cpu)
 408 {
 409         return !!per_cpu(rcu_preempt_data, cpu).nxtlist;
 410 }
 411
 412 /*
 413  * Initialize preemptable RCU's per-CPU data.
 414  */
 415 static void __cpuinit rcu_preempt_init_percpu_data(int cpu)
 416 {
 417         rcu_init_percpu_data(cpu, &rcu_preempt_state, 1);
 418 }
 419
 420 /*
 421  * Check for a task exiting while in a preemptable-RCU read-side
 422  * critical section, clean up if so.  No need to issue warnings,
 423  * as debug_check_no_locks_held() already does this if lockdep
 424  * is enabled.
 425  */
 426 void exit_rcu(void)
 427 {
 428         struct task_struct *t = current;
 429
 430         if (t->rcu_read_lock_nesting == 0)
 431                 return;
 432         t->rcu_read_lock_nesting = 1;
 433         rcu_read_unlock();
 434 }
 435
 436 #else /* #ifdef CONFIG_TREE_PREEMPT_RCU */
 437
 438 /*
 439  * Tell them what RCU they are running.
 440  */
 441 static inline void rcu_bootup_announce(void)
 442 {
 443         printk(KERN_INFO "Hierarchical RCU implementation.\n");
 444 }
 445
 446 /*
 447  * Return the number of RCU batches processed thus far for debug & stats.
 448  */
 449 long rcu_batches_completed(void)
 450 {
 451         return rcu_batches_completed_sched();
 452 }
 453 EXPORT_SYMBOL_GPL(rcu_batches_completed);
 454
 455 /*
 456  * Because preemptable RCU does not exist, we never have to check for
 457  * CPUs being in quiescent states.
 458  */
 459 static void rcu_preempt_note_context_switch(int cpu)
 460 {
 461 }
 462
 463 #ifdef CONFIG_RCU_CPU_STALL_DETECTOR
 464
 465 /*
 466  * Because preemptable RCU does not exist, we never have to check for
 467  * tasks blocked within RCU read-side critical sections.
 468  */
 469 static void rcu_print_task_stall(struct rcu_node *rnp)
 470 {
 471 }
 472
 473 #endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
 474
 475 /*
 476  * Because there is no preemptable RCU, there can be no readers blocked,
 477  * so there is no need to check for blocked tasks.
 478  */
 479 static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp)
 480 {
 481 }
 482
 483 /*
 484  * Because preemptable RCU does not exist, there are never any preempted
 485  * RCU readers.
 486  */
 487 static int rcu_preempted_readers(struct rcu_node *rnp)
 488 {
 489         return 0;
 490 }
 491
 492 #ifdef CONFIG_HOTPLUG_CPU
 493
 494 /*
 495  * Because preemptable RCU does not exist, it never needs to migrate
 496  * tasks that were blocked within RCU read-side critical sections.
 497  */
 498 static void rcu_preempt_offline_tasks(struct rcu_state *rsp,
 499                                       struct rcu_node *rnp,
 500                                       struct rcu_data *rdp)
 501 {
 502 }
 503
 504 /*
 505  * Because preemptable RCU does not exist, it never needs CPU-offline
 506  * processing.
 507  */
 508 static void rcu_preempt_offline_cpu(int cpu)
 509 {
 510 }
 511
 512 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
 513
 514 /*
 515  * Because preemptable RCU does not exist, it never has any callbacks
 516  * to check.
 517  */
 518 void rcu_preempt_check_callbacks(int cpu)
 519 {
 520 }
 521
 522 /*
 523  * Because preemptable RCU does not exist, it never has any callbacks
 524  * to process.
 525  */
 526 void rcu_preempt_process_callbacks(void)
 527 {
 528 }
 529
 530 /*
 531  * In classic RCU, call_rcu() is just call_rcu_sched().
 532  */
 533 void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
 534 {
 535         call_rcu_sched(head, func);
 536 }
 537 EXPORT_SYMBOL_GPL(call_rcu);
 538
 539 /*
 540  * Because preemptable RCU does not exist, it never has any work to do.
 541  */
 542 static int rcu_preempt_pending(int cpu)
 543 {
 544         return 0;
 545 }
 546
 547 /*
 548  * Because preemptable RCU does not exist, it never needs any CPU.
 549  */
 550 static int rcu_preempt_needs_cpu(int cpu)
 551 {
 552         return 0;
 553 }
 554
 555 /*
 556  * Because preemptable RCU does not exist, there is no per-CPU
 557  * data to initialize.
 558  */
 559 static void __cpuinit rcu_preempt_init_percpu_data(int cpu)
 560 {
 561 }
 562
 563 #endif /* #else #ifdef CONFIG_TREE_PREEMPT_RCU */