Add missing urcu-qsbr
authorMathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Tue, 9 Jun 2009 21:50:23 +0000 (17:50 -0400)
committerMathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Tue, 9 Jun 2009 21:50:23 +0000 (17:50 -0400)
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
urcu-qsbr.c [new file with mode: 0644]

diff --git a/urcu-qsbr.c b/urcu-qsbr.c
new file mode 100644 (file)
index 0000000..fd0ebf6
--- /dev/null
@@ -0,0 +1,390 @@
+/*
+ * urcu.c
+ *
+ * Userspace RCU library
+ *
+ * Copyright (c) 2009 Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
+ * Copyright (c) 2009 Paul E. McKenney, IBM Corporation.
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ *
+ * IBM's contributions to this file may be relicensed under LGPLv2 or later.
+ */
+
+#include <stdio.h>
+#include <pthread.h>
+#include <signal.h>
+#include <assert.h>
+#include <stdlib.h>
+#include <string.h>
+#include <errno.h>
+#include <poll.h>
+
+#include "urcu-static.h"
+/* Do not #define _LGPL_SOURCE to ensure we can emit the wrapper symbols */
+#include "urcu.h"
+
+void __attribute__((constructor)) urcu_init(void);
+void __attribute__((destructor)) urcu_exit(void);
+
+int init_done;
+
+pthread_mutex_t urcu_mutex = PTHREAD_MUTEX_INITIALIZER;
+
+/*
+ * Global grace period counter.
+ */
+long urcu_gp_ctr = 0;
+
+/*
+ * Written to only by each individual reader. Read by both the reader and the
+ * writers.
+ */
+long __thread urcu_active_readers;
+
+/* Thread IDs of registered readers */
+#define INIT_NUM_THREADS 4
+
+struct reader_registry {
+       pthread_t tid;
+       long *urcu_active_readers;
+       char *need_mb;
+};
+
+#ifdef DEBUG_YIELD
+unsigned int yield_active;
+unsigned int __thread rand_yield;
+#endif
+
+static struct reader_registry *registry;
+static char __thread need_mb;
+static int num_readers, alloc_readers;
+
+void internal_urcu_lock(void)
+{
+       int ret;
+
+#ifndef DISTRUST_SIGNALS_EXTREME
+       ret = pthread_mutex_lock(&urcu_mutex);
+       if (ret) {
+               perror("Error in pthread mutex lock");
+               exit(-1);
+       }
+#else /* #ifndef DISTRUST_SIGNALS_EXTREME */
+       while ((ret = pthread_mutex_trylock(&urcu_mutex)) != 0) {
+               if (ret != EBUSY && ret != EINTR) {
+                       printf("ret = %d, errno = %d\n", ret, errno);
+                       perror("Error in pthread mutex lock");
+                       exit(-1);
+               }
+               if (need_mb) {
+                       smp_mb();
+                       need_mb = 0;
+                       smp_mb();
+               }
+               poll(NULL,0,10);
+       }
+#endif /* #else #ifndef DISTRUST_SIGNALS_EXTREME */
+}
+
+void internal_urcu_unlock(void)
+{
+       int ret;
+
+       ret = pthread_mutex_unlock(&urcu_mutex);
+       if (ret) {
+               perror("Error in pthread mutex unlock");
+               exit(-1);
+       }
+}
+
+#ifdef DEBUG_FULL_MB
+#ifdef HAS_INCOHERENT_CACHES
+static void force_mb_single_thread(struct reader_registry *index)
+{
+       smp_mb();
+}
+#endif /* #ifdef HAS_INCOHERENT_CACHES */
+
+static void force_mb_all_threads(void)
+{
+       smp_mb();
+}
+#else /* #ifdef DEBUG_FULL_MB */
+#ifdef HAS_INCOHERENT_CACHES
+static void force_mb_single_thread(struct reader_registry *index)
+{
+       assert(registry);
+       /*
+        * pthread_kill has a smp_mb(). But beware, we assume it performs
+        * a cache flush on architectures with non-coherent cache. Let's play
+        * safe and don't assume anything : we use smp_mc() to make sure the
+        * cache flush is enforced.
+        */
+       *index->need_mb = 1;
+       smp_mc();       /* write ->need_mb before sending the signals */
+       pthread_kill(index->tid, SIGURCU);
+       smp_mb();
+       /*
+        * Wait for sighandler (and thus mb()) to execute on every thread.
+        * BUSY-LOOP.
+        */
+       while (*index->need_mb) {
+               poll(NULL, 0, 1);
+       }
+       smp_mb();       /* read ->need_mb before ending the barrier */
+}
+#endif /* #ifdef HAS_INCOHERENT_CACHES */
+
+static void force_mb_all_threads(void)
+{
+       struct reader_registry *index;
+       /*
+        * Ask for each threads to execute a smp_mb() so we can consider the
+        * compiler barriers around rcu read lock as real memory barriers.
+        */
+       if (!registry)
+               return;
+       /*
+        * pthread_kill has a smp_mb(). But beware, we assume it performs
+        * a cache flush on architectures with non-coherent cache. Let's play
+        * safe and don't assume anything : we use smp_mc() to make sure the
+        * cache flush is enforced.
+        */
+       for (index = registry; index < registry + num_readers; index++) {
+               *index->need_mb = 1;
+               smp_mc();       /* write need_mb before sending the signal */
+               pthread_kill(index->tid, SIGURCU);
+       }
+       /*
+        * Wait for sighandler (and thus mb()) to execute on every thread.
+        *
+        * Note that the pthread_kill() will never be executed on systems
+        * that correctly deliver signals in a timely manner.  However, it
+        * is not uncommon for kernels to have bugs that can result in
+        * lost or unduly delayed signals.
+        *
+        * If you are seeing the below pthread_kill() executing much at
+        * all, we suggest testing the underlying kernel and filing the
+        * relevant bug report.  For Linux kernels, we recommend getting
+        * the Linux Test Project (LTP).
+        */
+       for (index = registry; index < registry + num_readers; index++) {
+               while (*index->need_mb) {
+                       pthread_kill(index->tid, SIGURCU);
+                       poll(NULL, 0, 1);
+               }
+       }
+       smp_mb();       /* read ->need_mb before ending the barrier */
+}
+#endif /* #else #ifdef DEBUG_FULL_MB */
+
+void wait_for_quiescent_state(void)
+{
+       struct reader_registry *index;
+
+       if (!registry)
+               return;
+       /*
+        * Wait for each thread urcu_active_readers count to become 0.
+        */
+       for (index = registry; index < registry + num_readers; index++) {
+#ifndef HAS_INCOHERENT_CACHES
+               while (rcu_old_gp_ongoing(index->urcu_active_readers))
+                       cpu_relax();
+#else /* #ifndef HAS_INCOHERENT_CACHES */
+               int wait_loops = 0;
+               /*
+                * BUSY-LOOP. Force the reader thread to commit its
+                * urcu_active_readers update to memory if we wait for too long.
+                */
+               while (rcu_old_gp_ongoing(index->urcu_active_readers)) {
+                       if (wait_loops++ == KICK_READER_LOOPS) {
+                               force_mb_single_thread(index);
+                               wait_loops = 0;
+                       } else {
+                               cpu_relax();
+                       }
+               }
+#endif /* #else #ifndef HAS_INCOHERENT_CACHES */
+       }
+}
+
+void synchronize_rcu(void)
+{
+       internal_urcu_lock();
+       force_mb_all_threads();
+       urcu_gp_ctr += 2;
+       wait_for_quiescent_state();
+       force_mb_all_threads();
+       internal_urcu_unlock();
+}
+
+/*
+ * library wrappers to be used by non-LGPL compatible source code.
+ */
+
+void rcu_read_lock(void)
+{
+       _rcu_read_lock();
+}
+
+void rcu_read_unlock(void)
+{
+       _rcu_read_unlock();
+}
+
+void *rcu_dereference(void *p)
+{
+       return _rcu_dereference(p);
+}
+
+void *rcu_assign_pointer_sym(void **p, void *v)
+{
+       wmb();
+       return STORE_SHARED(p, v);
+}
+
+void *rcu_xchg_pointer_sym(void **p, void *v)
+{
+       wmb();
+       return xchg(p, v);
+}
+
+void *rcu_publish_content_sym(void **p, void *v)
+{
+       void *oldptr;
+
+       oldptr = _rcu_xchg_pointer(p, v);
+       synchronize_rcu();
+       return oldptr;
+}
+
+static void rcu_add_reader(pthread_t id)
+{
+       struct reader_registry *oldarray;
+
+       if (!registry) {
+               alloc_readers = INIT_NUM_THREADS;
+               num_readers = 0;
+               registry =
+                       malloc(sizeof(struct reader_registry) * alloc_readers);
+       }
+       if (alloc_readers < num_readers + 1) {
+               oldarray = registry;
+               registry = malloc(sizeof(struct reader_registry)
+                               * (alloc_readers << 1));
+               memcpy(registry, oldarray,
+                       sizeof(struct reader_registry) * alloc_readers);
+               alloc_readers <<= 1;
+               free(oldarray);
+       }
+       registry[num_readers].tid = id;
+       /* reference to the TLS of _this_ reader thread. */
+       registry[num_readers].urcu_active_readers = &urcu_active_readers;
+       registry[num_readers].need_mb = &need_mb;
+       num_readers++;
+}
+
+/*
+ * Never shrink (implementation limitation).
+ * This is O(nb threads). Eventually use a hash table.
+ */
+static void rcu_remove_reader(pthread_t id)
+{
+       struct reader_registry *index;
+
+       assert(registry != NULL);
+       for (index = registry; index < registry + num_readers; index++) {
+               if (pthread_equal(index->tid, id)) {
+                       memcpy(index, &registry[num_readers - 1],
+                               sizeof(struct reader_registry));
+                       registry[num_readers - 1].tid = 0;
+                       registry[num_readers - 1].urcu_active_readers = NULL;
+                       num_readers--;
+                       return;
+               }
+       }
+       /* Hrm not found, forgot to register ? */
+       assert(0);
+}
+
+void rcu_register_thread(void)
+{
+       internal_urcu_lock();
+       urcu_init();    /* In case gcc does not support constructor attribute */
+       rcu_add_reader(pthread_self());
+       internal_urcu_unlock();
+}
+
+void rcu_unregister_thread(void)
+{
+       internal_urcu_lock();
+       rcu_remove_reader(pthread_self());
+       internal_urcu_unlock();
+}
+
+#ifndef DEBUG_FULL_MB
+static void sigurcu_handler(int signo, siginfo_t *siginfo, void *context)
+{
+       /*
+        * Executing this smp_mb() is the only purpose of this signal handler.
+        * It punctually promotes barrier() into smp_mb() on every thread it is
+        * executed on.
+        */
+       smp_mb();
+       need_mb = 0;
+       smp_mb();
+}
+
+/*
+ * urcu_init constructor. Called when the library is linked, but also when
+ * reader threads are calling rcu_register_thread().
+ * Should only be called by a single thread at a given time. This is ensured by
+ * holing the internal_urcu_lock() from rcu_register_thread() or by running at
+ * library load time, which should not be executed by multiple threads nor
+ * concurrently with rcu_register_thread() anyway.
+ */
+void urcu_init(void)
+{
+       struct sigaction act;
+       int ret;
+
+       if (init_done)
+               return;
+       init_done = 1;
+
+       act.sa_sigaction = sigurcu_handler;
+       ret = sigaction(SIGURCU, &act, NULL);
+       if (ret) {
+               perror("Error in sigaction");
+               exit(-1);
+       }
+}
+
+void urcu_exit(void)
+{
+       struct sigaction act;
+       int ret;
+
+       ret = sigaction(SIGURCU, NULL, &act);
+       if (ret) {
+               perror("Error in sigaction");
+               exit(-1);
+       }
+       assert(act.sa_sigaction == sigurcu_handler);
+       free(registry);
+}
+#endif /* #ifndef DEBUG_FULL_MB */
This page took 0.028487 seconds and 4 git commands to generate.