--- /dev/null
+/*
+ * test_ht.c
+ *
+ * Userspace RCU library - test program
+ *
+ * Copyright February 2009 - Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program 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 General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License along
+ * with this program; if not, write to the Free Software Foundation, Inc.,
+ * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+ */
+
+#define _GNU_SOURCE
+#include <stdio.h>
+#include <pthread.h>
+#include <stdlib.h>
+#include <string.h>
+#include <sys/types.h>
+#include <sys/wait.h>
+#include <unistd.h>
+#include <stdio.h>
+#include <assert.h>
+#include <sys/syscall.h>
+#include <sched.h>
+#include <urcu-ht.h>
+#include <urcu-defer.h>
+#include <errno.h>
+
+#include "../arch.h"
+
+#define HASH_SIZE 32
+#define RAND_POOL 1000
+
+/* Make this big enough to include the POWER5+ L3 cacheline size of 256B */
+#define CACHE_LINE_SIZE 4096
+
+/* hardcoded number of CPUs */
+#define NR_CPUS 16384
+
+#if defined(_syscall0)
+_syscall0(pid_t, gettid)
+#elif defined(__NR_gettid)
+static inline pid_t gettid(void)
+{
+ return syscall(__NR_gettid);
+}
+#else
+#warning "use pid as tid"
+static inline pid_t gettid(void)
+{
+ return getpid();
+}
+#endif
+
+#ifndef DYNAMIC_LINK_TEST
+#define _LGPL_SOURCE
+#else
+#define debug_yield_read()
+#endif
+#include "../urcu.h"
+
+static unsigned int __thread rand_lookup;
+static unsigned long __thread nr_add;
+static unsigned long __thread nr_addexist;
+static unsigned long __thread nr_del;
+static unsigned long __thread nr_delnoent;
+static unsigned long __thread lookup_fail;
+static unsigned long __thread lookup_ok;
+
+static struct rcu_ht *test_ht;
+
+struct test_data {
+ int a;
+ int b;
+};
+
+static volatile int test_go, test_stop;
+
+static unsigned long wdelay;
+
+static unsigned long duration;
+
+/* read-side C.S. duration, in loops */
+static unsigned long rduration;
+
+static inline void loop_sleep(unsigned long l)
+{
+ while(l-- != 0)
+ cpu_relax();
+}
+
+static int verbose_mode;
+
+#define printf_verbose(fmt, args...) \
+ do { \
+ if (verbose_mode) \
+ printf(fmt, args); \
+ } while (0)
+
+static unsigned int cpu_affinities[NR_CPUS];
+static unsigned int next_aff = 0;
+static int use_affinity = 0;
+
+pthread_mutex_t affinity_mutex = PTHREAD_MUTEX_INITIALIZER;
+
+static void set_affinity(void)
+{
+ cpu_set_t mask;
+ int cpu;
+ int ret;
+
+ if (!use_affinity)
+ return;
+
+ ret = pthread_mutex_lock(&affinity_mutex);
+ if (ret) {
+ perror("Error in pthread mutex lock");
+ exit(-1);
+ }
+ cpu = cpu_affinities[next_aff++];
+ ret = pthread_mutex_unlock(&affinity_mutex);
+ if (ret) {
+ perror("Error in pthread mutex unlock");
+ exit(-1);
+ }
+ CPU_ZERO(&mask);
+ CPU_SET(cpu, &mask);
+ sched_setaffinity(0, sizeof(mask), &mask);
+}
+
+/*
+ * returns 0 if test should end.
+ */
+static int test_duration_write(void)
+{
+ return !test_stop;
+}
+
+static int test_duration_read(void)
+{
+ return !test_stop;
+}
+
+static unsigned long long __thread nr_writes;
+static unsigned long long __thread nr_reads;
+
+static unsigned int nr_readers;
+static unsigned int nr_writers;
+
+pthread_mutex_t rcu_copy_mutex = PTHREAD_MUTEX_INITIALIZER;
+
+void rcu_copy_mutex_lock(void)
+{
+ int ret;
+ ret = pthread_mutex_lock(&rcu_copy_mutex);
+ if (ret) {
+ perror("Error in pthread mutex lock");
+ exit(-1);
+ }
+}
+
+void rcu_copy_mutex_unlock(void)
+{
+ int ret;
+
+ ret = pthread_mutex_unlock(&rcu_copy_mutex);
+ if (ret) {
+ perror("Error in pthread mutex unlock");
+ exit(-1);
+ }
+}
+
+#define ARRAY_POISON 0xDEADBEEF
+
+void *thr_reader(void *_count)
+{
+ unsigned long long *count = _count;
+ struct test_data *local_ptr;
+
+ printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
+ "reader", pthread_self(), (unsigned long)gettid());
+
+ set_affinity();
+
+ rcu_register_thread();
+
+ while (!test_go)
+ {
+ }
+ smp_mb();
+
+ for (;;) {
+ rcu_read_lock();
+ local_ptr = ht_lookup(test_ht,
+ (void *)(unsigned long)(rand_r(&rand_lookup) % RAND_POOL));
+ if (local_ptr == NULL)
+ lookup_fail++;
+ else
+ lookup_ok++;
+ debug_yield_read();
+ if (unlikely(rduration))
+ loop_sleep(rduration);
+ rcu_read_unlock();
+ nr_reads++;
+ if (unlikely(!test_duration_read()))
+ break;
+ }
+
+ rcu_unregister_thread();
+
+ *count = nr_reads;
+ printf_verbose("thread_end %s, thread id : %lx, tid %lu\n",
+ "reader", pthread_self(), (unsigned long)gettid());
+ printf_verbose("readid : %lx, lookupfail %lu, lookupok %lu\n",
+ pthread_self(), lookup_fail, lookup_ok);
+ return ((void*)1);
+
+}
+
+void *thr_writer(void *_count)
+{
+ unsigned long long *count = _count;
+ struct test_data *data;
+ int ret;
+
+ printf_verbose("thread_begin %s, thread id : %lx, tid %lu\n",
+ "writer", pthread_self(), (unsigned long)gettid());
+
+ set_affinity();
+
+ rcu_register_thread();
+ rcu_defer_register_thread();
+
+ while (!test_go)
+ {
+ }
+ smp_mb();
+
+ for (;;) {
+ if (rand_r(&rand_lookup) & 1) {
+ data = malloc(sizeof(struct test_data));
+ //rcu_copy_mutex_lock();
+ ret = ht_add(test_ht,
+ (void *)(unsigned long)(rand_r(&rand_lookup) % RAND_POOL),
+ data);
+ if (ret == -EEXIST) {
+ free(data);
+ nr_addexist++;
+ } else {
+ nr_add++;
+ }
+ //rcu_copy_mutex_unlock();
+ } else {
+ /* May delete */
+ //rcu_copy_mutex_lock();
+ ret = ht_delete(test_ht,
+ (void *)(unsigned long)(rand_r(&rand_lookup) % RAND_POOL));
+ if (ret == -ENOENT)
+ nr_delnoent++;
+ else
+ nr_del++;
+ //rcu_copy_mutex_unlock();
+ }
+ //if (nr_writes % 100000 == 0) {
+ if (nr_writes % 1000 == 0) {
+ if (rand_r(&rand_lookup) & 1) {
+ ht_resize(test_ht, 1);
+ } else {
+ ht_resize(test_ht, -1);
+ }
+ }
+ nr_writes++;
+ if (unlikely(!test_duration_write()))
+ break;
+ if (unlikely(wdelay))
+ loop_sleep(wdelay);
+ }
+
+ rcu_defer_unregister_thread();
+ rcu_unregister_thread();
+
+ printf_verbose("thread_end %s, thread id : %lx, tid %lu\n",
+ "writer", pthread_self(), (unsigned long)gettid());
+ printf_verbose("info id %lx: nr_add %lu, nr_addexist %lu, nr_del %lu, "
+ "nr_delnoent %lu\n", pthread_self(), nr_add,
+ nr_addexist, nr_del, nr_delnoent);
+ *count = nr_writes;
+ return ((void*)2);
+}
+
+void show_usage(int argc, char **argv)
+{
+ printf("Usage : %s nr_readers nr_writers duration (s)", argv[0]);
+#ifdef DEBUG_YIELD
+ printf(" [-r] [-w] (yield reader and/or writer)");
+#endif
+ printf(" [-d delay] (writer period (us))");
+ printf(" [-c duration] (reader C.S. duration (in loops))");
+ printf(" [-v] (verbose output)");
+ printf(" [-a cpu#] [-a cpu#]... (affinity)");
+ printf("\n");
+}
+
+int main(int argc, char **argv)
+{
+ int err;
+ pthread_t *tid_reader, *tid_writer;
+ void *tret;
+ unsigned long long *count_reader, *count_writer;
+ unsigned long long tot_reads = 0, tot_writes = 0;
+ int i, a, ret;
+
+ if (argc < 4) {
+ show_usage(argc, argv);
+ return -1;
+ }
+
+ err = sscanf(argv[1], "%u", &nr_readers);
+ if (err != 1) {
+ show_usage(argc, argv);
+ return -1;
+ }
+
+ err = sscanf(argv[2], "%u", &nr_writers);
+ if (err != 1) {
+ show_usage(argc, argv);
+ return -1;
+ }
+
+ err = sscanf(argv[3], "%lu", &duration);
+ if (err != 1) {
+ show_usage(argc, argv);
+ return -1;
+ }
+
+ for (i = 4; i < argc; i++) {
+ if (argv[i][0] != '-')
+ continue;
+ switch (argv[i][1]) {
+#ifdef DEBUG_YIELD
+ case 'r':
+ yield_active |= YIELD_READ;
+ break;
+ case 'w':
+ yield_active |= YIELD_WRITE;
+ break;
+#endif
+ case 'a':
+ if (argc < i + 2) {
+ show_usage(argc, argv);
+ return -1;
+ }
+ a = atoi(argv[++i]);
+ cpu_affinities[next_aff++] = a;
+ use_affinity = 1;
+ printf_verbose("Adding CPU %d affinity\n", a);
+ break;
+ case 'c':
+ if (argc < i + 2) {
+ show_usage(argc, argv);
+ return -1;
+ }
+ rduration = atol(argv[++i]);
+ break;
+ case 'd':
+ if (argc < i + 2) {
+ show_usage(argc, argv);
+ return -1;
+ }
+ wdelay = atol(argv[++i]);
+ break;
+ case 'v':
+ verbose_mode = 1;
+ break;
+ }
+ }
+
+ printf_verbose("running test for %lu seconds, %u readers, %u writers.\n",
+ duration, nr_readers, nr_writers);
+ printf_verbose("Writer delay : %lu loops.\n", wdelay);
+ printf_verbose("Reader duration : %lu loops.\n", rduration);
+ printf_verbose("thread %-6s, thread id : %lx, tid %lu\n",
+ "main", pthread_self(), (unsigned long)gettid());
+
+ tid_reader = malloc(sizeof(*tid_reader) * nr_readers);
+ tid_writer = malloc(sizeof(*tid_writer) * nr_writers);
+ count_reader = malloc(sizeof(*count_reader) * nr_readers);
+ count_writer = malloc(sizeof(*count_writer) * nr_writers);
+ test_ht = ht_new(ht_jhash, free, HASH_SIZE, sizeof(unsigned long),
+ 43223455);
+ next_aff = 0;
+
+ for (i = 0; i < nr_readers; i++) {
+ err = pthread_create(&tid_reader[i], NULL, thr_reader,
+ &count_reader[i]);
+ if (err != 0)
+ exit(1);
+ }
+ for (i = 0; i < nr_writers; i++) {
+ err = pthread_create(&tid_writer[i], NULL, thr_writer,
+ &count_writer[i]);
+ if (err != 0)
+ exit(1);
+ }
+
+ smp_mb();
+
+ test_go = 1;
+
+ sleep(duration);
+
+ test_stop = 1;
+
+ for (i = 0; i < nr_readers; i++) {
+ err = pthread_join(tid_reader[i], &tret);
+ if (err != 0)
+ exit(1);
+ tot_reads += count_reader[i];
+ }
+ for (i = 0; i < nr_writers; i++) {
+ err = pthread_join(tid_writer[i], &tret);
+ if (err != 0)
+ exit(1);
+ tot_writes += count_writer[i];
+ }
+ rcu_register_thread();
+ rcu_defer_register_thread();
+ ret = ht_destroy(test_ht);
+ rcu_defer_unregister_thread();
+ rcu_unregister_thread();
+
+ printf_verbose("final delete: %d items\n", ret);
+ printf_verbose("total number of reads : %llu, writes %llu\n", tot_reads,
+ tot_writes);
+ printf("SUMMARY %-25s testdur %4lu nr_readers %3u rdur %6lu "
+ "nr_writers %3u "
+ "wdelay %6lu nr_reads %12llu nr_writes %12llu nr_ops %12llu\n",
+ argv[0], duration, nr_readers, rduration,
+ nr_writers, wdelay, tot_reads, tot_writes,
+ tot_reads + tot_writes);
+ free(tid_reader);
+ free(tid_writer);
+ free(count_reader);
+ free(count_writer);
+ return 0;
+}
--- /dev/null
+
+/*
+ * TODO: keys are currently assumed <= sizeof(void *). Key target never freed.
+ */
+
+#define _LGPL_SOURCE
+#include <stdlib.h>
+#include <errno.h>
+#include <assert.h>
+#include <stdio.h>
+
+#include <urcu.h>
+#include <urcu-defer.h>
+#include <arch.h>
+#include <arch_atomic.h>
+#include <compiler.h>
+#include <urcu/jhash.h>
+#include <stdio.h>
+#include <pthread.h>
+#include <urcu-ht.h>
+
+/*
+ * Maximum number of hash table buckets: 256M on 64-bit.
+ * Should take about 512MB max if we assume 1 node per 4 buckets.
+ */
+#define MAX_HT_BUCKETS ((256 << 10) / sizeof(void *))
+
+/* node flags */
+#define NODE_STOLEN (1 << 0)
+
+struct rcu_ht_node;
+
+struct rcu_ht_node {
+ struct rcu_ht_node *next;
+ void *key;
+ void *data;
+ unsigned int flags;
+};
+
+struct rcu_table {
+ unsigned long size;
+ struct rcu_ht_node *tbl[0];
+};
+
+struct rcu_ht {
+ struct rcu_table *t; /* shared */
+ ht_hash_fct hash_fct;
+ void (*free_fct)(void *data); /* fct to free data */
+ uint32_t keylen;
+ uint32_t hashseed;
+ pthread_mutex_t resize_mutex; /* resize mutex: add/del mutex */
+ int resize_ongoing; /* fast-path resize check */
+};
+
+struct rcu_ht *ht_new(ht_hash_fct hash_fct, void (*free_fct)(void *data),
+ unsigned long init_size, uint32_t keylen,
+ uint32_t hashseed)
+{
+ struct rcu_ht *ht;
+
+ ht = calloc(1, sizeof(struct rcu_ht));
+ ht->hash_fct = hash_fct;
+ ht->free_fct = free_fct;
+ ht->keylen = keylen;
+ ht->hashseed = hashseed;
+ /* this mutex should not nest in read-side C.S. */
+ pthread_mutex_init(&ht->resize_mutex, NULL);
+ ht->resize_ongoing = 0; /* shared */
+ ht->t = calloc(1, sizeof(struct rcu_table)
+ + (init_size * sizeof(struct rcu_ht_node *)));
+ ht->t->size = init_size;
+ return ht;
+}
+
+void *ht_lookup(struct rcu_ht *ht, void *key)
+{
+ struct rcu_table *t;
+ unsigned long hash;
+ struct rcu_ht_node *node;
+ void *ret;
+
+ rcu_read_lock();
+ t = rcu_dereference(ht->t);
+ smp_read_barrier_depends(); /* read t before size and table */
+ hash = ht->hash_fct(key, ht->keylen, ht->hashseed) % t->size;
+ smp_read_barrier_depends(); /* read size before links */
+ node = rcu_dereference(t->tbl[hash]);
+ for (;;) {
+ if (likely(!node)) {
+ ret = NULL;
+ break;
+ }
+ if (node->key == key) {
+ ret = node->data;
+ break;
+ }
+ node = rcu_dereference(node->next);
+ }
+ rcu_read_unlock();
+
+ return ret;
+}
+
+/*
+ * Will re-try until either:
+ * - The key is already there (-EEXIST)
+ * - We successfully add the key at the head of a table bucket.
+ */
+int ht_add(struct rcu_ht *ht, void *key, void *data)
+{
+ struct rcu_ht_node *node, *old_head, *new_head;
+ struct rcu_table *t;
+ unsigned long hash;
+ int ret = 0;
+
+ new_head = calloc(1, sizeof(struct rcu_ht_node));
+ new_head->key = key;
+ new_head->data = data;
+ new_head->flags = 0;
+ /* here comes the fun and tricky part.
+ * Add at the beginning with a cmpxchg.
+ * Hold a read lock between the moment the first element is read
+ * and the nodes traversal (to find duplicates). This ensures
+ * the head pointer has not been reclaimed when cmpxchg is done.
+ * Always adding at the head ensures that we would have to
+ * re-try if a new item has been added concurrently. So we ensure that
+ * we never add duplicates. */
+retry:
+ rcu_read_lock();
+
+ if (unlikely(LOAD_SHARED(ht->resize_ongoing))) {
+ rcu_read_unlock();
+ /*
+ * Wait for resize to complete before continuing.
+ */
+ ret = pthread_mutex_lock(&ht->resize_mutex);
+ assert(!ret);
+ ret = pthread_mutex_unlock(&ht->resize_mutex);
+ assert(!ret);
+ goto retry;
+ }
+
+ t = rcu_dereference(ht->t);
+ /* no read barrier needed, because no concurrency with resize */
+ hash = ht->hash_fct(key, ht->keylen, ht->hashseed) % t->size;
+
+ old_head = node = rcu_dereference(t->tbl[hash]);
+ for (;;) {
+ if (likely(!node)) {
+ break;
+ }
+ if (node->key == key) {
+ ret = -EEXIST;
+ goto end;
+ }
+ node = rcu_dereference(node->next);
+ }
+ new_head->next = old_head;
+ if (rcu_cmpxchg_pointer(&t->tbl[hash], old_head, new_head) != old_head)
+ goto restart;
+end:
+ rcu_read_unlock();
+ return ret;
+
+ /* restart loop, release and re-take the read lock to be kind to GP */
+restart:
+ rcu_read_unlock();
+ goto retry;
+}
+
+/*
+ * Restart until we successfully remove the entry, or no entry is left
+ * ((void *)(unsigned long)-ENOENT).
+ * Deal with concurrent stealers by doing an extra verification pass to check
+ * that no element in the list are still pointing to the element stolen.
+ * This could happen if two concurrent steal for consecutive objects are
+ * executed. A pointer to an object being stolen could be saved by the
+ * concurrent stealer for the previous object.
+ * Also, given that in this precise scenario, another stealer can also want to
+ * delete the doubly-referenced object; use a "stolen" flag to let only one
+ * stealer delete the object.
+ */
+void *ht_steal(struct rcu_ht *ht, void *key)
+{
+ struct rcu_ht_node **prev, *node, *del_node = NULL;
+ struct rcu_table *t;
+ unsigned long hash;
+ void *data;
+ int ret;
+
+retry:
+ rcu_read_lock();
+
+ if (unlikely(LOAD_SHARED(ht->resize_ongoing))) {
+ rcu_read_unlock();
+ /*
+ * Wait for resize to complete before continuing.
+ */
+ ret = pthread_mutex_lock(&ht->resize_mutex);
+ assert(!ret);
+ ret = pthread_mutex_unlock(&ht->resize_mutex);
+ assert(!ret);
+ goto retry;
+ }
+
+ t = rcu_dereference(ht->t);
+ /* no read barrier needed, because no concurrency with resize */
+ hash = ht->hash_fct(key, ht->keylen, ht->hashseed) % t->size;
+
+ prev = &t->tbl[hash];
+ node = rcu_dereference(*prev);
+ for (;;) {
+ if (likely(!node)) {
+ if (del_node) {
+ goto end;
+ } else {
+ goto error;
+ }
+ }
+ if (node->key == key) {
+ break;
+ }
+ prev = &node->next;
+ node = rcu_dereference(*prev);
+ }
+
+ if (!del_node) {
+ /*
+ * Another concurrent thread stole it ? If so, let it deal with
+ * this. Assume NODE_STOLEN is the only flag. If this changes,
+ * read flags before cmpxchg.
+ */
+ if (cmpxchg(&node->flags, 0, NODE_STOLEN) != 0)
+ goto error;
+ }
+
+ /* Found it ! pointer to object is in "prev" */
+ if (rcu_cmpxchg_pointer(prev, node, node->next) == node)
+ del_node = node;
+ goto restart;
+
+end:
+ /*
+ * From that point, we own node. Note that there can still be concurrent
+ * RCU readers using it. We can free it outside of read lock after a GP.
+ */
+ rcu_read_unlock();
+
+ data = del_node->data;
+ call_rcu(free, del_node);
+ return data;
+
+error:
+ data = (void *)(unsigned long)-ENOENT;
+ rcu_read_unlock();
+ return data;
+
+ /* restart loop, release and re-take the read lock to be kind to GP */
+restart:
+ rcu_read_unlock();
+ goto retry;
+}
+
+int ht_delete(struct rcu_ht *ht, void *key)
+{
+ void *data;
+
+ data = ht_steal(ht, key);
+ if (data && data != (void *)(unsigned long)-ENOENT) {
+ if (ht->free_fct)
+ call_rcu(ht->free_fct, data);
+ return 0;
+ } else {
+ return -ENOENT;
+ }
+}
+
+/* Delete all old elements. Allow concurrent writer accesses. */
+int ht_delete_all(struct rcu_ht *ht)
+{
+ unsigned long i;
+ struct rcu_ht_node **prev, *node, *inext;
+ struct rcu_table *t;
+ int cnt = 0;
+ int ret;
+
+ /*
+ * Mutual exclusion with resize operations, but leave add/steal execute
+ * concurrently. This is OK because we operate only on the heads.
+ */
+ ret = pthread_mutex_lock(&ht->resize_mutex);
+ assert(!ret);
+
+ t = rcu_dereference(ht->t);
+ /* no read barrier needed, because no concurrency with resize */
+ for (i = 0; i < t->size; i++) {
+ rcu_read_lock();
+ prev = &t->tbl[i];
+ /*
+ * Cut the head. After that, we own the first element.
+ */
+ node = rcu_xchg_pointer(prev, NULL);
+ if (!node) {
+ rcu_read_unlock();
+ continue;
+ }
+ /*
+ * We manage a list shared with concurrent writers and readers.
+ * Note that a concurrent add may or may not be deleted by us,
+ * depending if it arrives before or after the head is cut.
+ * "node" points to our first node. Remove first elements
+ * iteratively.
+ */
+ for (;;) {
+ inext = NULL;
+ prev = &node->next;
+ if (prev)
+ inext = rcu_xchg_pointer(prev, NULL);
+ /*
+ * "node" is the first element of the list we have cut.
+ * We therefore own it, no concurrent writer may delete
+ * it. There can only be concurrent lookups. Concurrent
+ * add can only be done on a bucket head, but we've cut
+ * it already. inext is also owned by us, because we
+ * have exchanged it for "NULL". It will therefore be
+ * safe to use it after a G.P.
+ */
+ rcu_read_unlock();
+ if (node->data)
+ call_rcu(ht->free_fct, node->data);
+ call_rcu(free, node);
+ cnt++;
+ if (likely(!inext))
+ break;
+ rcu_read_lock();
+ node = inext;
+ }
+ }
+
+ ret = pthread_mutex_unlock(&ht->resize_mutex);
+ assert(!ret);
+ return cnt;
+}
+
+/*
+ * Should only be called when no more concurrent readers nor writers can
+ * possibly access the table.
+ */
+int ht_destroy(struct rcu_ht *ht)
+{
+ int ret;
+
+ ret = ht_delete_all(ht);
+ free(ht->t);
+ free(ht);
+ return ret;
+}
+
+static void ht_resize_grow(struct rcu_ht *ht)
+{
+ unsigned long i, new_size, old_size;
+ struct rcu_table *new_t, *old_t;
+ struct rcu_ht_node *node, *new_node, *tmp;
+ unsigned long hash;
+
+ old_t = ht->t;
+ old_size = old_t->size;
+
+ if (old_size == MAX_HT_BUCKETS)
+ return;
+
+ new_size = old_size << 1;
+ new_t = calloc(1, sizeof(struct rcu_table)
+ + (new_size * sizeof(struct rcu_ht_node *)));
+ new_t->size = new_size;
+
+ for (i = 0; i < old_size; i++) {
+ /*
+ * Re-hash each entry, insert in new table.
+ * It's important that a reader looking for a key _will_ find it
+ * if it's in the table.
+ * Copy each node. (just the node, not ->data)
+ */
+ node = old_t->tbl[i];
+ while (node) {
+ hash = ht->hash_fct(node->key, ht->keylen, ht->hashseed)
+ % new_size;
+ new_node = malloc(sizeof(struct rcu_ht_node));
+ new_node->key = node->key;
+ new_node->data = node->data;
+ new_node->flags = node->flags;
+ new_node->next = new_t->tbl[hash]; /* link to first */
+ new_t->tbl[hash] = new_node; /* add to head */
+ node = node->next;
+ }
+ }
+
+ /* Changing table and size atomically wrt lookups */
+ rcu_assign_pointer(ht->t, new_t);
+
+ /* Ensure all concurrent lookups use new size and table */
+ synchronize_rcu();
+
+ for (i = 0; i < old_size; i++) {
+ node = old_t->tbl[i];
+ while (node) {
+ tmp = node->next;
+ free(node);
+ node = tmp;
+ }
+ }
+ free(old_t);
+}
+
+static void ht_resize_shrink(struct rcu_ht *ht)
+{
+ unsigned long i, new_size;
+ struct rcu_table *new_t, *old_t;
+ struct rcu_ht_node **prev, *node;
+
+ old_t = ht->t;
+ if (old_t->size == 1)
+ return;
+
+ new_size = old_t->size >> 1;
+
+ for (i = 0; i < new_size; i++) {
+ /* Link end with first entry of i + new_size */
+ prev = &old_t->tbl[i];
+ node = *prev;
+ while (node) {
+ prev = &node->next;
+ node = *prev;
+ }
+ *prev = old_t->tbl[i + new_size];
+ }
+ smp_wmb(); /* write links before changing size */
+ STORE_SHARED(old_t->size, new_size);
+
+ /* Ensure all concurrent lookups use new size */
+ synchronize_rcu();
+
+ new_t = realloc(old_t, sizeof(struct rcu_table)
+ + (new_size * sizeof(struct rcu_ht_node *)));
+ /* shrinking, pointers should not move */
+ assert(new_t == old_t);
+}
+
+/*
+ * growth: >0: *2, <0: /2
+ */
+void ht_resize(struct rcu_ht *ht, int growth)
+{
+ int ret;
+
+ ret = pthread_mutex_lock(&ht->resize_mutex);
+ assert(!ret);
+ STORE_SHARED(ht->resize_ongoing, 1);
+ synchronize_rcu();
+ /* All add/remove are waiting on the mutex. */
+ if (growth > 0)
+ ht_resize_grow(ht);
+ else if (growth < 0)
+ ht_resize_shrink(ht);
+ smp_mb();
+ STORE_SHARED(ht->resize_ongoing, 0);
+ ret = pthread_mutex_unlock(&ht->resize_mutex);
+ assert(!ret);
+}
+
+/*
+ * Expects keys <= than pointer size to be encoded in the pointer itself.
+ */
+uint32_t ht_jhash(void *key, uint32_t length, uint32_t initval)
+{
+ uint32_t ret;
+ void *vkey;
+
+ if (length <= sizeof(void *))
+ vkey = &key;
+ else
+ vkey = key;
+ ret = jhash(vkey, length, initval);
+ return ret;
+}
--- /dev/null
+#ifndef _URCU_HT_H
+#define _URCU_HT_H
+
+#include <stdint.h>
+
+/*
+ * Caution !
+ * Ensure writer threads are registered as urcu readers and with with
+ * urcu-defer.
+ * Ensure reader threads are registered as urcu readers.
+ */
+
+typedef uint32_t (*ht_hash_fct)(void *key, uint32_t length, uint32_t initval);
+
+/*
+ * init_size must be power of two.
+ */
+struct rcu_ht *ht_new(ht_hash_fct hash_fct, void (*free_fct)(void *data),
+ unsigned long init_size, uint32_t keylen,
+ uint32_t hashseed);
+
+int ht_delete_all(struct rcu_ht *ht);
+
+int ht_destroy(struct rcu_ht *ht);
+
+void *ht_lookup(struct rcu_ht *ht, void *key);
+
+int ht_add(struct rcu_ht *ht, void *key, void *data);
+
+int ht_delete(struct rcu_ht *ht, void *key);
+
+void *ht_steal(struct rcu_ht *ht, void *key);
+
+void ht_resize(struct rcu_ht *ht, int growth);
+
+uint32_t ht_jhash(void *key, uint32_t length, uint32_t initval);
+
+#endif /* _URCU_HT_H */
--- /dev/null
+#ifndef _KCOMPAT_JHASH_H
+#define _KCOMPAT_JHASH_H
+
+/* jhash.h: Jenkins hash support.
+ *
+ * Copyright (C) 1996 Bob Jenkins (bob_jenkins@burtleburtle.net)
+ *
+ * http://burtleburtle.net/bob/hash/
+ *
+ * These are the credits from Bob's sources:
+ *
+ * lookup2.c, by Bob Jenkins, December 1996, Public Domain.
+ * hash(), hash2(), hash3, and mix() are externally useful functions.
+ * Routines to test the hash are included if SELF_TEST is defined.
+ * You can use this free for any purpose. It has no warranty.
+ *
+ * Copyright (C) 2003 David S. Miller (davem@redhat.com)
+ *
+ * I've modified Bob's hash to be useful in the Linux kernel, and
+ * any bugs present are surely my fault. -DaveM
+ */
+
+#include <stdint.h>
+
+typedef uint8_t u8;
+typedef uint32_t u32;
+
+/* NOTE: Arguments are modified. */
+#define __jhash_mix(a, b, c) \
+{ \
+ a -= b; a -= c; a ^= (c>>13); \
+ b -= c; b -= a; b ^= (a<<8); \
+ c -= a; c -= b; c ^= (b>>13); \
+ a -= b; a -= c; a ^= (c>>12); \
+ b -= c; b -= a; b ^= (a<<16); \
+ c -= a; c -= b; c ^= (b>>5); \
+ a -= b; a -= c; a ^= (c>>3); \
+ b -= c; b -= a; b ^= (a<<10); \
+ c -= a; c -= b; c ^= (b>>15); \
+}
+
+/* The golden ration: an arbitrary value */
+#define JHASH_GOLDEN_RATIO 0x9e3779b9
+
+/* The most generic version, hashes an arbitrary sequence
+ * of bytes. No alignment or length assumptions are made about
+ * the input key.
+ */
+static inline u32 jhash(const void *key, u32 length, u32 initval)
+{
+ u32 a, b, c, len;
+ const u8 *k = key;
+
+ len = length;
+ a = b = JHASH_GOLDEN_RATIO;
+ c = initval;
+
+ while (len >= 12) {
+ a += (k[0] +((u32)k[1]<<8) +((u32)k[2]<<16) +((u32)k[3]<<24));
+ b += (k[4] +((u32)k[5]<<8) +((u32)k[6]<<16) +((u32)k[7]<<24));
+ c += (k[8] +((u32)k[9]<<8) +((u32)k[10]<<16)+((u32)k[11]<<24));
+
+ __jhash_mix(a,b,c);
+
+ k += 12;
+ len -= 12;
+ }
+
+ c += length;
+ switch (len) {
+ case 11: c += ((u32)k[10]<<24);
+ case 10: c += ((u32)k[9]<<16);
+ case 9 : c += ((u32)k[8]<<8);
+ case 8 : b += ((u32)k[7]<<24);
+ case 7 : b += ((u32)k[6]<<16);
+ case 6 : b += ((u32)k[5]<<8);
+ case 5 : b += k[4];
+ case 4 : a += ((u32)k[3]<<24);
+ case 3 : a += ((u32)k[2]<<16);
+ case 2 : a += ((u32)k[1]<<8);
+ case 1 : a += k[0];
+ };
+
+ __jhash_mix(a,b,c);
+
+ return c;
+}
+
+#endif /* _KCOMPAT_JHASH_H */