urcu/static/lfstack.h

   1 #ifndef _URCU_STATIC_LFSTACK_H
   2 #define _URCU_STATIC_LFSTACK_H
   3
   4 /*
   5  * urcu/static/lfstack.h
   6  *
   7  * Userspace RCU library - Lock-Free Stack
   8  *
   9  * Copyright 2010-2012 - Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
  10  *
  11  * TO BE INCLUDED ONLY IN LGPL-COMPATIBLE CODE. See rculfstack.h for linking
  12  * dynamically with the userspace rcu library.
  13  *
  14  * This library is free software; you can redistribute it and/or
  15  * modify it under the terms of the GNU Lesser General Public
  16  * License as published by the Free Software Foundation; either
  17  * version 2.1 of the License, or (at your option) any later version.
  18  *
  19  * This library is distributed in the hope that it will be useful,
  20  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  21  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  22  * Lesser General Public License for more details.
  23  *
  24  * You should have received a copy of the GNU Lesser General Public
  25  * License along with this library; if not, write to the Free Software
  26  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  27  */
  28
  29 #include <stdbool.h>
  30 #include <pthread.h>
  31 #include <assert.h>
  32 #include <urcu/uatomic.h>
  33 #include <urcu-pointer.h>
  34
  35 #ifdef __cplusplus
  36 extern "C" {
  37 #endif
  38
  39 /*
  40  * Lock-free stack.
  41  *
  42  * Stack implementing push, pop, pop_all operations, as well as iterator
  43  * on the stack head returned by pop_all.
  44  *
  45  * Synchronization table:
  46  *
  47  * External synchronization techniques described in the API below is
  48  * required between pairs marked with "X". No external synchronization
  49  * required between pairs marked with "-".
  50  *
  51  *                      cds_lfs_push  __cds_lfs_pop  __cds_lfs_pop_all
  52  * cds_lfs_push               -              -                  -
  53  * __cds_lfs_pop              -              X                  X
  54  * __cds_lfs_pop_all          -              X                  -
  55  *
  56  * cds_lfs_pop_blocking and cds_lfs_pop_all_blocking use an internal
  57  * mutex to provide synchronization.
  58  */
  59
  60 /*
  61  * cds_lfs_node_init: initialize lock-free stack node.
  62  */
  63 static inline
  64 void _cds_lfs_node_init(struct cds_lfs_node *node)
  65 {
  66 }
  67
  68 /*
  69  * cds_lfs_init: initialize lock-free stack.
  70  */
  71 static inline
  72 void _cds_lfs_init(struct cds_lfs_stack *s)
  73 {
  74         int ret;
  75
  76         s->head = NULL;
  77         ret = pthread_mutex_init(&s->lock, NULL);
  78         assert(!ret);
  79 }
  80
  81 static inline
  82 bool ___cds_lfs_empty_head(struct cds_lfs_head *head)
  83 {
  84         return head == NULL;
  85 }
  86
  87 /*
  88  * cds_lfs_empty: return whether lock-free stack is empty.
  89  *
  90  * No memory barrier is issued. No mutual exclusion is required.
  91  */
  92 static inline
  93 bool _cds_lfs_empty(struct cds_lfs_stack *s)
  94 {
  95         return ___cds_lfs_empty_head(CMM_LOAD_SHARED(s->head));
  96 }
  97
  98 /*
  99  * cds_lfs_push: push a node into the stack.
 100  *
 101  * Does not require any synchronization with other push nor pop.
 102  *
 103  * Lock-free stack push is not subject to ABA problem, so no need to
 104  * take the RCU read-side lock. Even if "head" changes between two
 105  * uatomic_cmpxchg() invocations here (being popped, and then pushed
 106  * again by one or more concurrent threads), the second
 107  * uatomic_cmpxchg() invocation only cares about pushing a new entry at
 108  * the head of the stack, ensuring consistency by making sure the new
 109  * node->next is the same pointer value as the value replaced as head.
 110  * It does not care about the content of the actual next node, so it can
 111  * very well be reallocated between the two uatomic_cmpxchg().
 112  *
 113  * We take the approach of expecting the stack to be usually empty, so
 114  * we first try an initial uatomic_cmpxchg() on a NULL old_head, and
 115  * retry if the old head was non-NULL (the value read by the first
 116  * uatomic_cmpxchg() is used as old head for the following loop). The
 117  * upside of this scheme is to minimize the amount of cacheline traffic,
 118  * always performing an exclusive cacheline access, rather than doing
 119  * non-exclusive followed by exclusive cacheline access (which would be
 120  * required if we first read the old head value). This design decision
 121  * might be revisited after more throrough benchmarking on various
 122  * platforms.
 123  *
 124  * Returns 0 if the stack was empty prior to adding the node.
 125  * Returns non-zero otherwise.
 126  */
 127 static inline
 128 bool _cds_lfs_push(struct cds_lfs_stack *s,
 129                   struct cds_lfs_node *node)
 130 {
 131         struct cds_lfs_head *head = NULL;
 132         struct cds_lfs_head *new_head =
 133                 caa_container_of(node, struct cds_lfs_head, node);
 134
 135         for (;;) {
 136                 struct cds_lfs_head *old_head = head;
 137
 138                 /*
 139                  * node->next is still private at this point, no need to
 140                  * perform a _CMM_STORE_SHARED().
 141                  */
 142                 node->next = &head->node;
 143                 /*
 144                  * uatomic_cmpxchg() implicit memory barrier orders earlier
 145                  * stores to node before publication.
 146                  */
 147                 head = uatomic_cmpxchg(&s->head, old_head, new_head);
 148                 if (old_head == head)
 149                         break;
 150         }
 151         return ___cds_lfs_empty_head(head);
 152 }
 153
 154 /*
 155  * __cds_lfs_pop: pop a node from the stack.
 156  *
 157  * Returns NULL if stack is empty.
 158  *
 159  * __cds_lfs_pop needs to be synchronized using one of the following
 160  * techniques:
 161  *
 162  * 1) Calling __cds_lfs_pop under rcu read lock critical section. The
 163  *    caller must wait for a grace period to pass before freeing the
 164  *    returned node or modifying the cds_lfs_node structure.
 165  * 2) Using mutual exclusion (e.g. mutexes) to protect __cds_lfs_pop
 166  *    and __cds_lfs_pop_all callers.
 167  * 3) Ensuring that only ONE thread can call __cds_lfs_pop() and
 168  *    __cds_lfs_pop_all(). (multi-provider/single-consumer scheme).
 169  */
 170 static inline
 171 struct cds_lfs_node *___cds_lfs_pop(struct cds_lfs_stack *s)
 172 {
 173         for (;;) {
 174                 struct cds_lfs_head *head, *next_head;
 175                 struct cds_lfs_node *next;
 176
 177                 head = _CMM_LOAD_SHARED(s->head);
 178                 if (___cds_lfs_empty_head(head))
 179                         return NULL;    /* Empty stack */
 180
 181                 /*
 182                  * Read head before head->next. Matches the implicit
 183                  * memory barrier before uatomic_cmpxchg() in
 184                  * cds_lfs_push.
 185                  */
 186                 cmm_smp_read_barrier_depends();
 187                 next = _CMM_LOAD_SHARED(head->node.next);
 188                 next_head = caa_container_of(next,
 189                                 struct cds_lfs_head, node);
 190                 if (uatomic_cmpxchg(&s->head, head, next_head) == head)
 191                         return &head->node;
 192                 /* busy-loop if head changed under us */
 193         }
 194 }
 195
 196 /*
 197  * __cds_lfs_pop_all: pop all nodes from a stack.
 198  *
 199  * __cds_lfs_pop_all does not require any synchronization with other
 200  * push, nor with other __cds_lfs_pop_all, but requires synchronization
 201  * matching the technique used to synchronize __cds_lfs_pop:
 202  *
 203  * 1) If __cds_lfs_pop is called under rcu read lock critical section,
 204  *    both __cds_lfs_pop and cds_lfs_pop_all callers must wait for a
 205  *    grace period to pass before freeing the returned node or modifying
 206  *    the cds_lfs_node structure. However, no RCU read-side critical
 207  *    section is needed around __cds_lfs_pop_all.
 208  * 2) Using mutual exclusion (e.g. mutexes) to protect __cds_lfs_pop and
 209  *    __cds_lfs_pop_all callers.
 210  * 3) Ensuring that only ONE thread can call __cds_lfs_pop() and
 211  *    __cds_lfs_pop_all(). (multi-provider/single-consumer scheme).
 212  */
 213 static inline
 214 struct cds_lfs_head *___cds_lfs_pop_all(struct cds_lfs_stack *s)
 215 {
 216         /*
 217          * Implicit memory barrier after uatomic_xchg() matches implicit
 218          * memory barrier before uatomic_cmpxchg() in cds_lfs_push. It
 219          * ensures that all nodes of the returned list are consistent.
 220          * There is no need to issue memory barriers when iterating on
 221          * the returned list, because the full memory barrier issued
 222          * prior to each uatomic_cmpxchg, which each write to head, are
 223          * taking care to order writes to each node prior to the full
 224          * memory barrier after this uatomic_xchg().
 225          */
 226         return uatomic_xchg(&s->head, NULL);
 227 }
 228
 229 /*
 230  * cds_lfs_pop_lock: lock stack pop-protection mutex.
 231  */
 232 static inline void _cds_lfs_pop_lock(struct cds_lfs_stack *s)
 233 {
 234         int ret;
 235
 236         ret = pthread_mutex_lock(&s->lock);
 237         assert(!ret);
 238 }
 239
 240 /*
 241  * cds_lfs_pop_unlock: unlock stack pop-protection mutex.
 242  */
 243 static inline void _cds_lfs_pop_unlock(struct cds_lfs_stack *s)
 244 {
 245         int ret;
 246
 247         ret = pthread_mutex_unlock(&s->lock);
 248         assert(!ret);
 249 }
 250
 251 /*
 252  * Call __cds_lfs_pop with an internal pop mutex held.
 253  */
 254 static inline
 255 struct cds_lfs_node *
 256 _cds_lfs_pop_blocking(struct cds_lfs_stack *s)
 257 {
 258         struct cds_lfs_node *retnode;
 259
 260         _cds_lfs_pop_lock(s);
 261         retnode = ___cds_lfs_pop(s);
 262         _cds_lfs_pop_unlock(s);
 263         return retnode;
 264 }
 265
 266 /*
 267  * Call __cds_lfs_pop_all with an internal pop mutex held.
 268  */
 269 static inline
 270 struct cds_lfs_head *
 271 _cds_lfs_pop_all_blocking(struct cds_lfs_stack *s)
 272 {
 273         struct cds_lfs_head *rethead;
 274
 275         _cds_lfs_pop_lock(s);
 276         rethead = ___cds_lfs_pop_all(s);
 277         _cds_lfs_pop_unlock(s);
 278         return rethead;
 279 }
 280
 281 #ifdef __cplusplus
 282 }
 283 #endif
 284
 285 #endif /* _URCU_STATIC_LFSTACK_H */