util/rcu.c

   1 /*
   2  * urcu-mb.c
   3  *
   4  * Userspace RCU library with explicit memory barriers
   5  *
   6  * Copyright (c) 2009 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
   7  * Copyright (c) 2009 Paul E. McKenney, IBM Corporation.
   8  * Copyright 2015 Red Hat, Inc.
   9  *
  10  * Ported to QEMU by Paolo Bonzini  <pbonzini@redhat.com>
  11  *
  12  * This library is free software; you can redistribute it and/or
  13  * modify it under the terms of the GNU Lesser General Public
  14  * License as published by the Free Software Foundation; either
  15  * version 2.1 of the License, or (at your option) any later version.
  16  *
  17  * This library is distributed in the hope that it will be useful,
  18  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  19  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  20  * Lesser General Public License for more details.
  21  *
  22  * You should have received a copy of the GNU Lesser General Public
  23  * License along with this library; if not, write to the Free Software
  24  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  25  *
  26  * IBM's contributions to this file may be relicensed under LGPLv2 or later.
  27  */
  28
  29 #include "qemu/osdep.h"
  30 #include "qemu/rcu.h"
  31 #include "qemu/atomic.h"
  32 #include "qemu/thread.h"
  33 #include "qemu/main-loop.h"
  34 #include "qemu/lockable.h"
  35 #if defined(CONFIG_MALLOC_TRIM)
  36 #include <malloc.h>
  37 #endif
  38
  39 /*
  40  * Global grace period counter.  Bit 0 is always one in rcu_gp_ctr.
  41  * Bits 1 and above are defined in synchronize_rcu.
  42  */
  43 #define RCU_GP_LOCKED           (1UL << 0)
  44 #define RCU_GP_CTR              (1UL << 1)
  45
  46 unsigned long rcu_gp_ctr = RCU_GP_LOCKED;
  47
  48 QemuEvent rcu_gp_event;
  49 static int in_drain_call_rcu;
  50 static QemuMutex rcu_registry_lock;
  51 static QemuMutex rcu_sync_lock;
  52
  53 /*
  54  * Check whether a quiescent state was crossed between the beginning of
  55  * update_counter_and_wait and now.
  56  */
  57 static inline int rcu_gp_ongoing(unsigned long *ctr)
  58 {
  59     unsigned long v;
  60
  61     v = qatomic_read(ctr);
  62     return v && (v != rcu_gp_ctr);
  63 }
  64
  65 /* Written to only by each individual reader. Read by both the reader and the
  66  * writers.
  67  */
  68 QEMU_DEFINE_CO_TLS(struct rcu_reader_data, rcu_reader)
  69
  70 /* Protected by rcu_registry_lock.  */
  71 typedef QLIST_HEAD(, rcu_reader_data) ThreadList;
  72 static ThreadList registry = QLIST_HEAD_INITIALIZER(registry);
  73
  74 /* Wait for previous parity/grace period to be empty of readers.  */
  75 static void wait_for_readers(void)
  76 {
  77     ThreadList qsreaders = QLIST_HEAD_INITIALIZER(qsreaders);
  78     struct rcu_reader_data *index, *tmp;
  79
  80     for (;;) {
  81         /* We want to be notified of changes made to rcu_gp_ongoing
  82          * while we walk the list.
  83          */
  84         qemu_event_reset(&rcu_gp_event);
  85
  86         QLIST_FOREACH(index, &registry, node) {
  87             qatomic_set(&index->waiting, true);
  88         }
  89
  90         /* Here, order the stores to index->waiting before the loads of
  91          * index->ctr.  Pairs with smp_mb_placeholder() in rcu_read_unlock(),
  92          * ensuring that the loads of index->ctr are sequentially consistent.
  93          *
  94          * If this is the last iteration, this barrier also prevents
  95          * frees from seeping upwards, and orders the two wait phases
  96          * on architectures with 32-bit longs; see synchronize_rcu().
  97          */
  98         smp_mb_global();
  99
 100         QLIST_FOREACH_SAFE(index, &registry, node, tmp) {
 101             if (!rcu_gp_ongoing(&index->ctr)) {
 102                 QLIST_REMOVE(index, node);
 103                 QLIST_INSERT_HEAD(&qsreaders, index, node);
 104
 105                 /* No need for memory barriers here, worst of all we
 106                  * get some extra futex wakeups.
 107                  */
 108                 qatomic_set(&index->waiting, false);
 109             } else if (qatomic_read(&in_drain_call_rcu)) {
 110                 notifier_list_notify(&index->force_rcu, NULL);
 111             }
 112         }
 113
 114         if (QLIST_EMPTY(&registry)) {
 115             break;
 116         }
 117
 118         /* Wait for one thread to report a quiescent state and try again.
 119          * Release rcu_registry_lock, so rcu_(un)register_thread() doesn't
 120          * wait too much time.
 121          *
 122          * rcu_register_thread() may add nodes to &registry; it will not
 123          * wake up synchronize_rcu, but that is okay because at least another
 124          * thread must exit its RCU read-side critical section before
 125          * synchronize_rcu is done.  The next iteration of the loop will
 126          * move the new thread's rcu_reader from &registry to &qsreaders,
 127          * because rcu_gp_ongoing() will return false.
 128          *
 129          * rcu_unregister_thread() may remove nodes from &qsreaders instead
 130          * of &registry if it runs during qemu_event_wait.  That's okay;
 131          * the node then will not be added back to &registry by QLIST_SWAP
 132          * below.  The invariant is that the node is part of one list when
 133          * rcu_registry_lock is released.
 134          */
 135         qemu_mutex_unlock(&rcu_registry_lock);
 136         qemu_event_wait(&rcu_gp_event);
 137         qemu_mutex_lock(&rcu_registry_lock);
 138     }
 139
 140     /* put back the reader list in the registry */
 141     QLIST_SWAP(&registry, &qsreaders, node);
 142 }
 143
 144 void synchronize_rcu(void)
 145 {
 146     QEMU_LOCK_GUARD(&rcu_sync_lock);
 147
 148     /* Write RCU-protected pointers before reading p_rcu_reader->ctr.
 149      * Pairs with smp_mb_placeholder() in rcu_read_lock().
 150      *
 151      * Also orders write to RCU-protected pointers before
 152      * write to rcu_gp_ctr.
 153      */
 154     smp_mb_global();
 155
 156     QEMU_LOCK_GUARD(&rcu_registry_lock);
 157     if (!QLIST_EMPTY(&registry)) {
 158         if (sizeof(rcu_gp_ctr) < 8) {
 159             /* For architectures with 32-bit longs, a two-subphases algorithm
 160              * ensures we do not encounter overflow bugs.
 161              *
 162              * Switch parity: 0 -> 1, 1 -> 0.
 163              */
 164             qatomic_set(&rcu_gp_ctr, rcu_gp_ctr ^ RCU_GP_CTR);
 165             wait_for_readers();
 166             qatomic_set(&rcu_gp_ctr, rcu_gp_ctr ^ RCU_GP_CTR);
 167         } else {
 168             /* Increment current grace period.  */
 169             qatomic_set(&rcu_gp_ctr, rcu_gp_ctr + RCU_GP_CTR);
 170         }
 171
 172         wait_for_readers();
 173     }
 174 }
 175
 176
 177 #define RCU_CALL_MIN_SIZE        30
 178
 179 /* Multi-producer, single-consumer queue based on urcu/static/wfqueue.h
 180  * from liburcu.  Note that head is only used by the consumer.
 181  */
 182 static struct rcu_head dummy;
 183 static struct rcu_head *head = &dummy, **tail = &dummy.next;
 184 static int rcu_call_count;
 185 static QemuEvent rcu_call_ready_event;
 186
 187 static void enqueue(struct rcu_head *node)
 188 {
 189     struct rcu_head **old_tail;
 190
 191     node->next = NULL;
 192
 193     /*
 194      * Make this node the tail of the list.  The node will be
 195      * used by further enqueue operations, but it will not
 196      * be dequeued yet...
 197      */
 198     old_tail = qatomic_xchg(&tail, &node->next);
 199
 200     /*
 201      * ... until it is pointed to from another item in the list.
 202      * In the meantime, try_dequeue() will find a NULL next pointer
 203      * and loop.
 204      *
 205      * Synchronizes with qatomic_load_acquire() in try_dequeue().
 206      */
 207     qatomic_store_release(old_tail, node);
 208 }
 209
 210 static struct rcu_head *try_dequeue(void)
 211 {
 212     struct rcu_head *node, *next;
 213
 214 retry:
 215     /* Head is only written by this thread, so no need for barriers.  */
 216     node = head;
 217
 218     /*
 219      * If the head node has NULL in its next pointer, the value is
 220      * wrong and we need to wait until its enqueuer finishes the update.
 221      */
 222     next = qatomic_load_acquire(&node->next);
 223     if (!next) {
 224         return NULL;
 225     }
 226
 227     /*
 228      * Test for an empty list, which we do not expect.  Note that for
 229      * the consumer head and tail are always consistent.  The head
 230      * is consistent because only the consumer reads/writes it.
 231      * The tail, because it is the first step in the enqueuing.
 232      * It is only the next pointers that might be inconsistent.
 233      */
 234     if (head == &dummy && qatomic_read(&tail) == &dummy.next) {
 235         abort();
 236     }
 237
 238     /*
 239      * Since we are the sole consumer, and we excluded the empty case
 240      * above, the queue will always have at least two nodes: the
 241      * dummy node, and the one being removed.  So we do not need to update
 242      * the tail pointer.
 243      */
 244     head = next;
 245
 246     /* If we dequeued the dummy node, add it back at the end and retry.  */
 247     if (node == &dummy) {
 248         enqueue(node);
 249         goto retry;
 250     }
 251
 252     return node;
 253 }
 254
 255 static void *call_rcu_thread(void *opaque)
 256 {
 257     struct rcu_head *node;
 258
 259     rcu_register_thread();
 260
 261     for (;;) {
 262         int tries = 0;
 263         int n = qatomic_read(&rcu_call_count);
 264
 265         /* Heuristically wait for a decent number of callbacks to pile up.
 266          * Fetch rcu_call_count now, we only must process elements that were
 267          * added before synchronize_rcu() starts.
 268          */
 269         while (n == 0 || (n < RCU_CALL_MIN_SIZE && ++tries <= 5)) {
 270             g_usleep(10000);
 271             if (n == 0) {
 272                 qemu_event_reset(&rcu_call_ready_event);
 273                 n = qatomic_read(&rcu_call_count);
 274                 if (n == 0) {
 275 #if defined(CONFIG_MALLOC_TRIM)
 276                     malloc_trim(4 * 1024 * 1024);
 277 #endif
 278                     qemu_event_wait(&rcu_call_ready_event);
 279                 }
 280             }
 281             n = qatomic_read(&rcu_call_count);
 282         }
 283
 284         qatomic_sub(&rcu_call_count, n);
 285         synchronize_rcu();
 286         qemu_mutex_lock_iothread();
 287         while (n > 0) {
 288             node = try_dequeue();
 289             while (!node) {
 290                 qemu_mutex_unlock_iothread();
 291                 qemu_event_reset(&rcu_call_ready_event);
 292                 node = try_dequeue();
 293                 if (!node) {
 294                     qemu_event_wait(&rcu_call_ready_event);
 295                     node = try_dequeue();
 296                 }
 297                 qemu_mutex_lock_iothread();
 298             }
 299
 300             n--;
 301             node->func(node);
 302         }
 303         qemu_mutex_unlock_iothread();
 304     }
 305     abort();
 306 }
 307
 308 void call_rcu1(struct rcu_head *node, void (*func)(struct rcu_head *node))
 309 {
 310     node->func = func;
 311     enqueue(node);
 312     qatomic_inc(&rcu_call_count);
 313     qemu_event_set(&rcu_call_ready_event);
 314 }
 315
 316
 317 struct rcu_drain {
 318     struct rcu_head rcu;
 319     QemuEvent drain_complete_event;
 320 };
 321
 322 static void drain_rcu_callback(struct rcu_head *node)
 323 {
 324     struct rcu_drain *event = (struct rcu_drain *)node;
 325     qemu_event_set(&event->drain_complete_event);
 326 }
 327
 328 /*
 329  * This function ensures that all pending RCU callbacks
 330  * on the current thread are done executing
 331
 332  * drops big qemu lock during the wait to allow RCU thread
 333  * to process the callbacks
 334  *
 335  */
 336
 337 void drain_call_rcu(void)
 338 {
 339     struct rcu_drain rcu_drain;
 340     bool locked = qemu_mutex_iothread_locked();
 341
 342     memset(&rcu_drain, 0, sizeof(struct rcu_drain));
 343     qemu_event_init(&rcu_drain.drain_complete_event, false);
 344
 345     if (locked) {
 346         qemu_mutex_unlock_iothread();
 347     }
 348
 349
 350     /*
 351      * RCU callbacks are invoked in the same order as in which they
 352      * are registered, thus we can be sure that when 'drain_rcu_callback'
 353      * is called, all RCU callbacks that were registered on this thread
 354      * prior to calling this function are completed.
 355      *
 356      * Note that since we have only one global queue of the RCU callbacks,
 357      * we also end up waiting for most of RCU callbacks that were registered
 358      * on the other threads, but this is a side effect that shoudn't be
 359      * assumed.
 360      */
 361
 362     qatomic_inc(&in_drain_call_rcu);
 363     call_rcu1(&rcu_drain.rcu, drain_rcu_callback);
 364     qemu_event_wait(&rcu_drain.drain_complete_event);
 365     qatomic_dec(&in_drain_call_rcu);
 366
 367     if (locked) {
 368         qemu_mutex_lock_iothread();
 369     }
 370
 371 }
 372
 373 void rcu_register_thread(void)
 374 {
 375     assert(get_ptr_rcu_reader()->ctr == 0);
 376     qemu_mutex_lock(&rcu_registry_lock);
 377     QLIST_INSERT_HEAD(&registry, get_ptr_rcu_reader(), node);
 378     qemu_mutex_unlock(&rcu_registry_lock);
 379 }
 380
 381 void rcu_unregister_thread(void)
 382 {
 383     qemu_mutex_lock(&rcu_registry_lock);
 384     QLIST_REMOVE(get_ptr_rcu_reader(), node);
 385     qemu_mutex_unlock(&rcu_registry_lock);
 386 }
 387
 388 void rcu_add_force_rcu_notifier(Notifier *n)
 389 {
 390     qemu_mutex_lock(&rcu_registry_lock);
 391     notifier_list_add(&get_ptr_rcu_reader()->force_rcu, n);
 392     qemu_mutex_unlock(&rcu_registry_lock);
 393 }
 394
 395 void rcu_remove_force_rcu_notifier(Notifier *n)
 396 {
 397     qemu_mutex_lock(&rcu_registry_lock);
 398     notifier_remove(n);
 399     qemu_mutex_unlock(&rcu_registry_lock);
 400 }
 401
 402 static void rcu_init_complete(void)
 403 {
 404     QemuThread thread;
 405
 406     qemu_mutex_init(&rcu_registry_lock);
 407     qemu_mutex_init(&rcu_sync_lock);
 408     qemu_event_init(&rcu_gp_event, true);
 409
 410     qemu_event_init(&rcu_call_ready_event, false);
 411
 412     /* The caller is assumed to have iothread lock, so the call_rcu thread
 413      * must have been quiescent even after forking, just recreate it.
 414      */
 415     qemu_thread_create(&thread, "call_rcu", call_rcu_thread,
 416                        NULL, QEMU_THREAD_DETACHED);
 417
 418     rcu_register_thread();
 419 }
 420
 421 static int atfork_depth = 1;
 422
 423 void rcu_enable_atfork(void)
 424 {
 425     atfork_depth++;
 426 }
 427
 428 void rcu_disable_atfork(void)
 429 {
 430     atfork_depth--;
 431 }
 432
 433 #ifdef CONFIG_POSIX
 434 static void rcu_init_lock(void)
 435 {
 436     if (atfork_depth < 1) {
 437         return;
 438     }
 439
 440     qemu_mutex_lock(&rcu_sync_lock);
 441     qemu_mutex_lock(&rcu_registry_lock);
 442 }
 443
 444 static void rcu_init_unlock(void)
 445 {
 446     if (atfork_depth < 1) {
 447         return;
 448     }
 449
 450     qemu_mutex_unlock(&rcu_registry_lock);
 451     qemu_mutex_unlock(&rcu_sync_lock);
 452 }
 453
 454 static void rcu_init_child(void)
 455 {
 456     if (atfork_depth < 1) {
 457         return;
 458     }
 459
 460     memset(&registry, 0, sizeof(registry));
 461     rcu_init_complete();
 462 }
 463 #endif
 464
 465 static void __attribute__((__constructor__)) rcu_init(void)
 466 {
 467     smp_mb_global_init();
 468 #ifdef CONFIG_POSIX
 469     pthread_atfork(rcu_init_lock, rcu_init_unlock, rcu_init_child);
 470 #endif
 471     rcu_init_complete();
 472 }