urcu/workqueue-fifo.h

   1 #ifndef _URCU_WORKQUEUE_FIFO_H
   2 #define _URCU_WORKQUEUE_FIFO_H
   3
   4 /*
   5  * urcu/workqueue-fifo.h
   6  *
   7  * Userspace RCU library - work queue scheme with FIFO semantic
   8  *
   9  * Copyright (c) 2014 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
  10  *
  11  * This library is free software; you can redistribute it and/or
  12  * modify it under the terms of the GNU Lesser General Public
  13  * License as published by the Free Software Foundation; either
  14  * version 2.1 of the License, or (at your option) any later version.
  15  *
  16  * This library is distributed in the hope that it will be useful,
  17  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  18  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  19  * Lesser General Public License for more details.
  20  *
  21  * You should have received a copy of the GNU Lesser General Public
  22  * License along with this library; if not, write to the Free Software
  23  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  24  */
  25
  26 #include <urcu/uatomic.h>
  27 #include <urcu/lfstack.h>
  28 #include <urcu/waitqueue-lifo.h>
  29 #include <urcu/wfcqueue.h>
  30 #include <urcu/rculist.h>
  31 #include <pthread.h>
  32 #include <assert.h>
  33
  34 enum urcu_accept_ret {
  35         URCU_ACCEPT_WORK        = 0,
  36         URCU_ACCEPT_SHUTDOWN    = 1,
  37 };
  38
  39 /*
  40  * We use RCU to steal work from siblings. Therefore, one of RCU flavors
  41  * need to be included before this header. All worker that participate
  42  * in stealing (initialized with the URCU_WORKER_STEAL flag) need to be
  43  * registered RCU readers threads.
  44  */
  45
  46 struct urcu_work {
  47         struct cds_wfcq_node node;
  48 };
  49
  50 struct urcu_workqueue {
  51         /* FIFO work queue */
  52         struct __cds_wfcq_head head;
  53         struct cds_wfcq_tail tail;
  54
  55         /* Associated wait queue for LIFO wait/wakeup */
  56         struct urcu_wait_queue waitqueue;
  57
  58         /* RCU linked list head of siblings for work stealing. */
  59         struct cds_list_head sibling_head;
  60         pthread_mutex_t sibling_lock;   /* Protect sibling list updates */
  61
  62         bool shutdown;                  /* Shutdown performed */
  63 };
  64
  65 struct urcu_worker {
  66         /* Workqueue which can be either used by worker, or stolen. */
  67         struct cds_wfcq_head head;
  68         struct cds_wfcq_tail tail;
  69
  70         /* Work belonging to worker. Cannot be stolen. */
  71         struct urcu_work *own;
  72
  73         struct urcu_wait_node wait_node;
  74         /* RCU linked list node of siblings for work stealing. */
  75         struct cds_list_head sibling_node;
  76         int flags;      /* enum urcu_worker_flags */
  77 };
  78
  79 enum urcu_worker_flags {
  80         URCU_WORKER_STEAL       = (1 << 0),
  81 };
  82
  83 static inline
  84 void urcu_workqueue_init(struct urcu_workqueue *queue)
  85 {
  86         __cds_wfcq_init(&queue->head, &queue->tail);
  87         urcu_wait_queue_init(&queue->waitqueue);
  88         CDS_INIT_LIST_HEAD(&queue->sibling_head);
  89         queue->shutdown = false;
  90 }
  91
  92 static inline
  93 void urcu_queue_work(struct urcu_workqueue *queue, struct urcu_work *work)
  94 {
  95         bool was_empty;
  96
  97         cds_wfcq_node_init(&work->node);
  98
  99         /* Enqueue work. */
 100         was_empty = !cds_wfcq_enqueue(&queue->head, &queue->tail,
 101                         &work->node);
 102         /*
 103          * If workqueue was previously empty, wakeup one worker thread.
 104          * It will eventually grab the entire content of the work-queue
 105          * (therefore grabbing a "work batch"). After having grabbed the
 106          * work batch, while that thread is running and taking care of
 107          * that work batch, when we enqueue more work, we will wake
 108          * another thread (if there is one waiting), which will
 109          * eventually grab the new batch, and so on. This scheme ensures
 110          * that contiguous batch of work are handled by the same thread
 111          * (for locality), and also ensures that we scale work to many
 112          * worker threads when threads are busy enough to still be
 113          * running when work is enqueued.
 114          */
 115         if (was_empty) {
 116                 rcu_read_lock();        /* Protect stack dequeue */
 117                 (void) urcu_dequeue_wake_single(&queue->waitqueue);
 118                 rcu_read_unlock();      /* Protect stack dequeue */
 119         }
 120 }
 121
 122 static inline
 123 void __urcu_workqueue_wakeup_all(struct urcu_workqueue *queue)
 124 {
 125         struct urcu_waiters waiters;
 126
 127         rcu_read_lock();        /* Protect stack dequeue */
 128         urcu_move_waiters(&waiters, &queue->waitqueue);
 129         rcu_read_unlock();      /* Protect stack dequeue */
 130
 131         (void) urcu_wake_all_waiters(&waiters);
 132 }
 133
 134 static inline
 135 void urcu_worker_init(struct urcu_worker *worker, int flags)
 136 {
 137         cds_wfcq_init(&worker->head, &worker->tail);
 138         worker->flags = flags;
 139         urcu_wait_node_init(&worker->wait_node, URCU_WAIT_RUNNING);
 140         worker->own = NULL;
 141 }
 142
 143 static inline
 144 void urcu_worker_register(struct urcu_workqueue *queue,
 145                 struct urcu_worker *worker)
 146 {
 147         if (worker->flags & URCU_WORKER_STEAL) {
 148                 pthread_mutex_lock(&queue->sibling_lock);
 149                 cds_list_add_rcu(&worker->sibling_node, &queue->sibling_head);
 150                 pthread_mutex_unlock(&queue->sibling_lock);
 151         }
 152 }
 153
 154 static inline
 155 void urcu_worker_unregister(struct urcu_workqueue *queue,
 156                 struct urcu_worker *worker)
 157 {
 158         enum cds_wfcq_ret wfcq_ret;
 159
 160         if (worker->flags & URCU_WORKER_STEAL) {
 161                 pthread_mutex_lock(&queue->sibling_lock);
 162                 cds_list_del_rcu(&worker->sibling_node);
 163                 pthread_mutex_unlock(&queue->sibling_lock);
 164         }
 165
 166         /*
 167          * Make sure we are removed from waitqueue.
 168          */
 169         if (CMM_LOAD_SHARED(worker->wait_node.node.next))
 170                 __urcu_workqueue_wakeup_all(queue);
 171
 172         /*
 173          * Put any local work we still have back into the workqueue.
 174          */
 175         wfcq_ret = __cds_wfcq_splice_blocking(&queue->head,
 176                         &queue->tail,
 177                         &worker->head,
 178                         &worker->tail);
 179         if (wfcq_ret != CDS_WFCQ_RET_SRC_EMPTY
 180                         && wfcq_ret == CDS_WFCQ_RET_DEST_EMPTY) {
 181                 /*
 182                  * Wakeup worker thread if we have put work back into
 183                  * workqueue that was previously empty.
 184                  */
 185                 rcu_read_lock();        /* Protect stack dequeue */
 186                 (void) urcu_dequeue_wake_single(&queue->waitqueue);
 187                 rcu_read_unlock();      /* Protect stack dequeue */
 188         }
 189
 190         /*
 191          * Wait for grace period before freeing or reusing
 192          * "worker" because used by RCU linked list.
 193          * Also prevents ABA for waitqueue stack dequeue: matches RCU
 194          * read-side critical sections around dequeue and move all
 195          * operations on waitqueue).
 196          */
 197         synchronize_rcu();
 198 }
 199
 200 static inline
 201 bool ___urcu_grab_work(struct urcu_worker *worker,
 202                 cds_wfcq_head_ptr_t src_head,
 203                 struct cds_wfcq_tail *src_tail,
 204                 bool steal)
 205 {
 206         enum cds_wfcq_ret splice_ret;
 207         struct cds_wfcq_head tmp_head;
 208         struct cds_wfcq_tail tmp_tail;
 209         struct cds_wfcq_node *node;
 210
 211         /*
 212          * Don't bother grabbing the src queue lock if it is empty.
 213          */
 214         if (cds_wfcq_empty(src_head, src_tail))
 215                 return false;
 216         cds_wfcq_init(&tmp_head, &tmp_tail);
 217
 218         /* Ensure that we preserve FIFO work order. */
 219         assert(!steal || worker->own == NULL);
 220
 221         /* Splice to temporary queue. */
 222         if (steal)
 223                 cds_wfcq_dequeue_lock(src_head.h, src_tail);
 224         splice_ret = __cds_wfcq_splice_blocking(&tmp_head,
 225                         &tmp_tail,
 226                         src_head,
 227                         src_tail);
 228         if (steal)
 229                 cds_wfcq_dequeue_unlock(src_head.h, src_tail);
 230         if (splice_ret == CDS_WFCQ_RET_SRC_EMPTY)
 231                 return false;
 232
 233         /*
 234          * Keep one work entry for ourself. This ensures forward
 235          * progress amongst stealing co-workers. This also ensures that
 236          * when a worker grab some work from the global workqueue, it
 237          * will have at least one work item to deal with.
 238          */
 239         if (worker->own == NULL) {
 240                 if (!steal) {
 241                         /*
 242                          * Try to grab own work from worker workqueue to
 243                          * preserve FIFO order.
 244                          */
 245                         node = cds_wfcq_dequeue_blocking(&worker->head,
 246                                 &worker->tail);
 247                         if (node)
 248                                 goto got_node;
 249                 }
 250                 node = __cds_wfcq_dequeue_blocking(&tmp_head, &tmp_tail);
 251                 assert(node != NULL);
 252 got_node:
 253                 worker->own = caa_container_of(node, struct urcu_work, node);
 254         }
 255
 256         /* Splice into worker workqueue. */
 257         splice_ret = cds_wfcq_splice_blocking(&worker->head,
 258                         &worker->tail,
 259                         &tmp_head,
 260                         &tmp_tail);
 261         /* Ensure that we preserve FIFO work order. */
 262         assert(!steal || splice_ret != CDS_WFCQ_RET_DEST_NON_EMPTY);
 263         return true;
 264 }
 265
 266 /*
 267  * Try stealing work from siblings when we have nothing to do.
 268  */
 269 static inline
 270 bool ___urcu_steal_work(struct urcu_worker *worker,
 271                 struct urcu_worker *sibling)
 272 {
 273         return ___urcu_grab_work(worker, &sibling->head, &sibling->tail, 1);
 274 }
 275
 276 static inline
 277 bool __urcu_steal_work(struct urcu_workqueue *queue,
 278                 struct urcu_worker *worker)
 279 {
 280         struct urcu_worker *sibling_prev, *sibling_next;
 281         struct cds_list_head *sibling_node;
 282         bool steal_performed = 0;
 283
 284         if (!(worker->flags & URCU_WORKER_STEAL))
 285                 return false;
 286
 287         rcu_read_lock();
 288
 289         sibling_node = rcu_dereference(worker->sibling_node.next);
 290         if (sibling_node == &queue->sibling_head)
 291                 sibling_node = rcu_dereference(sibling_node->next);
 292         sibling_next = caa_container_of(sibling_node, struct urcu_worker,
 293                         sibling_node);
 294         if (sibling_next != worker)
 295                 steal_performed = ___urcu_steal_work(worker, sibling_next);
 296         if (steal_performed)
 297                 goto end;
 298
 299         sibling_node = rcu_dereference(worker->sibling_node.prev);
 300         if (sibling_node == &queue->sibling_head)
 301                 sibling_node = rcu_dereference(sibling_node->prev);
 302         sibling_prev = caa_container_of(sibling_node, struct urcu_worker,
 303                         sibling_node);
 304         if (sibling_prev != worker && sibling_prev != sibling_next)
 305                 steal_performed =  ___urcu_steal_work(worker, sibling_prev);
 306 end:
 307         rcu_read_unlock();
 308
 309         return steal_performed;
 310 }
 311
 312 static inline
 313 bool ___urcu_wakeup_sibling(struct urcu_worker *sibling)
 314 {
 315         return urcu_adaptative_wake_up(&sibling->wait_node);
 316 }
 317
 318 static inline
 319 bool __urcu_wakeup_siblings(struct urcu_workqueue *queue,
 320                 struct urcu_worker *worker)
 321 {
 322         struct urcu_worker *sibling_prev, *sibling_next;
 323         struct cds_list_head *sibling_node;
 324         bool wakeup_performed = 0;
 325
 326         if (!(worker->flags & URCU_WORKER_STEAL))
 327                 return;
 328
 329         /* Only wakeup siblings if we have work in our own queue. */
 330         if (cds_wfcq_empty(&worker->head, &worker->tail))
 331                 return;
 332
 333         rcu_read_lock();
 334
 335         sibling_node = rcu_dereference(worker->sibling_node.next);
 336         if (sibling_node == &queue->sibling_head)
 337                 sibling_node = rcu_dereference(sibling_node->next);
 338         sibling_next = caa_container_of(sibling_node, struct urcu_worker,
 339                         sibling_node);
 340         if (sibling_next != worker)
 341                 wakeup_performed = ___urcu_wakeup_sibling(sibling_next);
 342         if (wakeup_performed)
 343                 goto end;
 344
 345         sibling_node = rcu_dereference(worker->sibling_node.prev);
 346         if (sibling_node == &queue->sibling_head)
 347                 sibling_node = rcu_dereference(sibling_node->prev);
 348         sibling_prev = caa_container_of(sibling_node, struct urcu_worker,
 349                         sibling_node);
 350         if (sibling_prev != worker && sibling_prev != sibling_next)
 351                 wakeup_performed = ___urcu_wakeup_sibling(sibling_prev);
 352 end:
 353         rcu_read_unlock();
 354
 355         return wakeup_performed;
 356 }
 357
 358 static inline
 359 enum urcu_accept_ret urcu_accept_work(struct urcu_workqueue *queue,
 360                 struct urcu_worker *worker)
 361 {
 362         enum cds_wfcq_ret wfcq_ret;
 363         bool has_work;
 364
 365         has_work = ___urcu_grab_work(worker, &queue->head, &queue->tail, 0);
 366         /* Don't wait if we have work to do. */
 367         if (has_work || !cds_wfcq_empty(&worker->head, &worker->tail))
 368                 goto do_work;
 369         /* Try to steal work from sibling instead of blocking */
 370         if (__urcu_steal_work(queue, worker))
 371                 goto do_work;
 372         /* No more work to do, check shutdown state */
 373         if (CMM_LOAD_SHARED(queue->shutdown))
 374                 return URCU_ACCEPT_SHUTDOWN;
 375         urcu_wait_set_state(&worker->wait_node,
 376                         URCU_WAIT_WAITING);
 377         if (!CMM_LOAD_SHARED(worker->wait_node.node.next)) {
 378                 int was_empty;
 379
 380                 /*
 381                  * NULL next pointer. We are therefore not in
 382                  * the queue.
 383                  */
 384                 cds_lfs_node_init(&worker->wait_node.node);
 385                 /* Protect stack dequeue against ABA */
 386                 synchronize_rcu();
 387                 was_empty = !urcu_wait_add(&queue->waitqueue,
 388                                 &worker->wait_node);
 389                 /*
 390                  * If the wait queue was empty, it means we are the
 391                  * first thread to be put back into an otherwise empty
 392                  * wait queue. Re-check if work queue is empty after
 393                  * adding ourself to wait queue, so we can wakeup the
 394                  * top of wait queue since new work have appeared, and
 395                  * work enqueuer may not have seen that it needed to do
 396                  * a wake up.
 397                  */
 398                 if (was_empty && !cds_wfcq_empty(&queue->head,
 399                                                 &queue->tail)) {
 400                         rcu_read_lock();        /* Protect stack dequeue */
 401                         (void) urcu_dequeue_wake_single(&queue->waitqueue);
 402                         rcu_read_unlock();      /* Protect stack dequeue */
 403                 }
 404         } else {
 405                 /*
 406                  * Non-NULL next pointer. We are therefore in
 407                  * the queue, or the dispatcher just removed us
 408                  * from it (after we read the next pointer), and
 409                  * is therefore awakening us. The state will
 410                  * therefore have been changed from WAITING to
 411                  * some other state, which will let the busy
 412                  * wait pass through.
 413                  */
 414         }
 415         urcu_adaptative_busy_wait(&worker->wait_node);
 416         return;
 417
 418 do_work:
 419         /*
 420          * We will be busy handling the work batch, awaken siblings so
 421          * they can steal from us.
 422          */
 423         (void) __urcu_wakeup_siblings(queue, worker);
 424         return URCU_ACCEPT_WORK;
 425 }
 426
 427 static inline
 428 struct urcu_work *urcu_dequeue_work(struct urcu_worker *worker)
 429 {
 430         struct cds_wfcq_node *node;
 431
 432         if (worker->own) {
 433                 struct urcu_work *work;
 434
 435                 /* Process our own work entry. */
 436                 work = worker->own;
 437                 worker->own = NULL;
 438                 return work;
 439         }
 440         /*
 441          * If we are registered for work stealing, we need to dequeue
 442          * safely against siblings.
 443          */
 444         if (worker->flags & URCU_WORKER_STEAL) {
 445                 /*
 446                  * Don't bother grabbing the worker queue lock if it is
 447                  * empty.
 448                  */
 449                 if (cds_wfcq_empty(&worker->head, &worker->tail))
 450                         return NULL;
 451                 node = cds_wfcq_dequeue_blocking(&worker->head,
 452                                 &worker->tail);
 453         } else {
 454                 node = ___cds_wfcq_dequeue_with_state(&worker->head,
 455                                 &worker->tail, NULL, 1, 0);
 456         }
 457         if (!node)
 458                 return NULL;
 459         return caa_container_of(node, struct urcu_work, node);
 460 }
 461
 462 static inline
 463 void urcu_workqueue_shutdown(struct urcu_workqueue *queue)
 464 {
 465         /* Set shutdown */
 466         CMM_STORE_SHARED(queue->shutdown, true);
 467         /* Wakeup all workers */
 468         __urcu_workqueue_wakeup_all(queue);
 469 }
 470
 471 #endif /* _URCU_WORKQUEUE_FIFO_H */