* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
+#define _GNU_SOURCE
#include <stdio.h>
#include <pthread.h>
#include <signal.h>
#include <assert.h>
#include <stdlib.h>
+#include <stdint.h>
#include <string.h>
#include <errno.h>
#include <poll.h>
#include <sys/time.h>
-#include <syscall.h>
#include <unistd.h>
+#include <sched.h>
#include "config.h"
#include "urcu/wfqueue.h"
#include "urcu-call-rcu.h"
#include "urcu-pointer.h"
#include "urcu/list.h"
+#include "urcu/futex.h"
+#include "urcu/tls-compat.h"
+#include "urcu-die.h"
/* Data structure that identifies a call_rcu thread. */
struct call_rcu_data {
struct cds_wfq_queue cbs;
unsigned long flags;
- pthread_mutex_t mtx;
- pthread_cond_t cond;
- unsigned long qlen;
+ int32_t futex;
+ unsigned long qlen; /* maintained for debugging. */
pthread_t tid;
+ int cpu_affinity;
struct cds_list_head list;
} __attribute__((aligned(CAA_CACHE_LINE_SIZE)));
/* Link a thread using call_rcu() to its call_rcu thread. */
-static __thread struct call_rcu_data *thread_call_rcu_data;
+static DEFINE_URCU_TLS(struct call_rcu_data *, thread_call_rcu_data);
/* Guard call_rcu thread creation. */
/*
* Pointer to array of pointers to per-CPU call_rcu_data structures
- * and # CPUs.
+ * and # CPUs. per_cpu_call_rcu_data is a RCU-protected pointer to an
+ * array of RCU-protected pointers to call_rcu_data. call_rcu acts as a
+ * RCU read-side and reads per_cpu_call_rcu_data and the per-cpu pointer
+ * without mutex. The call_rcu_mutex protects updates.
*/
static struct call_rcu_data **per_cpu_call_rcu_data;
static long maxcpus;
+static void maxcpus_reset(void)
+{
+ maxcpus = 0;
+}
+
/* Allocate the array if it has not already been allocated. */
static void alloc_cpu_call_rcu_data(void)
p = malloc(maxcpus * sizeof(*per_cpu_call_rcu_data));
if (p != NULL) {
memset(p, '\0', maxcpus * sizeof(*per_cpu_call_rcu_data));
- per_cpu_call_rcu_data = p;
+ rcu_set_pointer(&per_cpu_call_rcu_data, p);
} else {
if (!warned) {
fprintf(stderr, "[error] liburcu: unable to allocate per-CPU pointer array\n");
#else /* #if defined(HAVE_SCHED_GETCPU) && defined(HAVE_SYSCONF) */
-static const struct call_rcu_data **per_cpu_call_rcu_data = NULL;
+/*
+ * per_cpu_call_rcu_data should be constant, but some functions below, used both
+ * for cases where cpu number is available and not available, assume it it not
+ * constant.
+ */
+static struct call_rcu_data **per_cpu_call_rcu_data = NULL;
static const long maxcpus = -1;
+static void maxcpus_reset(void)
+{
+}
+
static void alloc_cpu_call_rcu_data(void)
{
}
static void call_rcu_lock(pthread_mutex_t *pmp)
{
- if (pthread_mutex_lock(pmp) != 0) {
- perror("pthread_mutex_lock");
- exit(-1);
- }
+ int ret;
+
+ ret = pthread_mutex_lock(pmp);
+ if (ret)
+ urcu_die(ret);
}
/* Release the specified pthread mutex. */
static void call_rcu_unlock(pthread_mutex_t *pmp)
{
- if (pthread_mutex_unlock(pmp) != 0) {
- perror("pthread_mutex_unlock");
- exit(-1);
+ int ret;
+
+ ret = pthread_mutex_unlock(pmp);
+ if (ret)
+ urcu_die(ret);
+}
+
+#if HAVE_SCHED_SETAFFINITY
+static
+int set_thread_cpu_affinity(struct call_rcu_data *crdp)
+{
+ cpu_set_t mask;
+
+ if (crdp->cpu_affinity < 0)
+ return 0;
+
+ CPU_ZERO(&mask);
+ CPU_SET(crdp->cpu_affinity, &mask);
+#if SCHED_SETAFFINITY_ARGS == 2
+ return sched_setaffinity(0, &mask);
+#else
+ return sched_setaffinity(0, sizeof(mask), &mask);
+#endif
+}
+#else
+static
+int set_thread_cpu_affinity(struct call_rcu_data *crdp)
+{
+ return 0;
+}
+#endif
+
+static void call_rcu_wait(struct call_rcu_data *crdp)
+{
+ /* Read call_rcu list before read futex */
+ cmm_smp_mb();
+ if (uatomic_read(&crdp->futex) == -1)
+ futex_async(&crdp->futex, FUTEX_WAIT, -1,
+ NULL, NULL, 0);
+}
+
+static void call_rcu_wake_up(struct call_rcu_data *crdp)
+{
+ /* Write to call_rcu list before reading/writing futex */
+ cmm_smp_mb();
+ if (caa_unlikely(uatomic_read(&crdp->futex) == -1)) {
+ uatomic_set(&crdp->futex, 0);
+ futex_async(&crdp->futex, FUTEX_WAKE, 1,
+ NULL, NULL, 0);
}
}
struct cds_wfq_node **cbs_tail;
struct call_rcu_data *crdp = (struct call_rcu_data *)arg;
struct rcu_head *rhp;
+ int rt = !!(uatomic_read(&crdp->flags) & URCU_CALL_RCU_RT);
+ int ret;
- thread_call_rcu_data = crdp;
+ ret = set_thread_cpu_affinity(crdp);
+ if (ret)
+ urcu_die(errno);
+
+ /*
+ * If callbacks take a read-side lock, we need to be registered.
+ */
+ rcu_register_thread();
+
+ URCU_TLS(thread_call_rcu_data) = crdp;
+ if (!rt) {
+ uatomic_dec(&crdp->futex);
+ /* Decrement futex before reading call_rcu list */
+ cmm_smp_mb();
+ }
for (;;) {
if (&crdp->cbs.head != _CMM_LOAD_SHARED(crdp->cbs.tail)) {
while ((cbs = _CMM_LOAD_SHARED(crdp->cbs.head)) == NULL)
} while (cbs != NULL);
uatomic_sub(&crdp->qlen, cbcount);
}
- if (crdp->flags & URCU_CALL_RCU_STOP)
+ if (uatomic_read(&crdp->flags) & URCU_CALL_RCU_STOP)
break;
- if (crdp->flags & URCU_CALL_RCU_RT)
- poll(NULL, 0, 10);
- else {
- call_rcu_lock(&crdp->mtx);
- _CMM_STORE_SHARED(crdp->flags,
- crdp->flags & ~URCU_CALL_RCU_RUNNING);
- if (&crdp->cbs.head ==
- _CMM_LOAD_SHARED(crdp->cbs.tail) &&
- pthread_cond_wait(&crdp->cond, &crdp->mtx) != 0) {
- perror("pthread_cond_wait");
- exit(-1);
+ rcu_thread_offline();
+ if (!rt) {
+ if (&crdp->cbs.head
+ == _CMM_LOAD_SHARED(crdp->cbs.tail)) {
+ call_rcu_wait(crdp);
+ poll(NULL, 0, 10);
+ uatomic_dec(&crdp->futex);
+ /*
+ * Decrement futex before reading
+ * call_rcu list.
+ */
+ cmm_smp_mb();
+ } else {
+ poll(NULL, 0, 10);
}
- _CMM_STORE_SHARED(crdp->flags,
- crdp->flags | URCU_CALL_RCU_RUNNING);
+ } else {
poll(NULL, 0, 10);
- call_rcu_unlock(&crdp->mtx);
}
+ rcu_thread_online();
}
- call_rcu_lock(&crdp->mtx);
- crdp->flags |= URCU_CALL_RCU_STOPPED;
- call_rcu_unlock(&crdp->mtx);
+ if (!rt) {
+ /*
+ * Read call_rcu list before write futex.
+ */
+ cmm_smp_mb();
+ uatomic_set(&crdp->futex, 0);
+ }
+ uatomic_or(&crdp->flags, URCU_CALL_RCU_STOPPED);
+ rcu_unregister_thread();
return NULL;
}
*/
static void call_rcu_data_init(struct call_rcu_data **crdpp,
- unsigned long flags)
+ unsigned long flags,
+ int cpu_affinity)
{
struct call_rcu_data *crdp;
+ int ret;
crdp = malloc(sizeof(*crdp));
- if (crdp == NULL) {
- fprintf(stderr, "Out of memory.\n");
- exit(-1);
- }
+ if (crdp == NULL)
+ urcu_die(errno);
memset(crdp, '\0', sizeof(*crdp));
cds_wfq_init(&crdp->cbs);
crdp->qlen = 0;
- if (pthread_mutex_init(&crdp->mtx, NULL) != 0) {
- perror("pthread_mutex_init");
- exit(-1);
- }
- if (pthread_cond_init(&crdp->cond, NULL) != 0) {
- perror("pthread_cond_init");
- exit(-1);
- }
- crdp->flags = flags | URCU_CALL_RCU_RUNNING;
+ crdp->futex = 0;
+ crdp->flags = flags;
cds_list_add(&crdp->list, &call_rcu_data_list);
+ crdp->cpu_affinity = cpu_affinity;
cmm_smp_mb(); /* Structure initialized before pointer is planted. */
*crdpp = crdp;
- if (pthread_create(&crdp->tid, NULL, call_rcu_thread, crdp) != 0) {
- perror("pthread_create");
- exit(-1);
- }
+ ret = pthread_create(&crdp->tid, NULL, call_rcu_thread, crdp);
+ if (ret)
+ urcu_die(ret);
}
/*
* CPU, returning NULL if there is none. We cannot automatically
* created it because the platform we are running on might not define
* sched_getcpu().
+ *
+ * The call to this function and use of the returned call_rcu_data
+ * should be protected by RCU read-side lock.
*/
struct call_rcu_data *get_cpu_call_rcu_data(int cpu)
{
static int warned = 0;
+ struct call_rcu_data **pcpu_crdp;
- if (per_cpu_call_rcu_data == NULL)
+ pcpu_crdp = rcu_dereference(per_cpu_call_rcu_data);
+ if (pcpu_crdp == NULL)
return NULL;
if (!warned && maxcpus > 0 && (cpu < 0 || maxcpus <= cpu)) {
fprintf(stderr, "[error] liburcu: get CPU # out of range\n");
}
if (cpu < 0 || maxcpus <= cpu)
return NULL;
- return per_cpu_call_rcu_data[cpu];
+ return rcu_dereference(pcpu_crdp[cpu]);
}
/*
* Create a call_rcu_data structure (with thread) and return a pointer.
*/
-static struct call_rcu_data *__create_call_rcu_data(unsigned long flags)
+static struct call_rcu_data *__create_call_rcu_data(unsigned long flags,
+ int cpu_affinity)
{
struct call_rcu_data *crdp;
- call_rcu_data_init(&crdp, flags);
+ call_rcu_data_init(&crdp, flags, cpu_affinity);
return crdp;
}
-struct call_rcu_data *create_call_rcu_data(unsigned long flags)
+struct call_rcu_data *create_call_rcu_data(unsigned long flags,
+ int cpu_affinity)
{
struct call_rcu_data *crdp;
call_rcu_lock(&call_rcu_mutex);
- crdp = __create_call_rcu_data(flags);
+ crdp = __create_call_rcu_data(flags, cpu_affinity);
call_rcu_unlock(&call_rcu_mutex);
return crdp;
}
* the caller's responsibility to dispose of the removed structure.
* Use get_cpu_call_rcu_data() to obtain a pointer to the old structure
* (prior to NULLing it out, of course).
+ *
+ * The caller must wait for a grace-period to pass between return from
+ * set_cpu_call_rcu_data() and call to call_rcu_data_free() passing the
+ * previous call rcu data as argument.
*/
int set_cpu_call_rcu_data(int cpu, struct call_rcu_data *crdp)
{
- int warned = 0;
+ static int warned = 0;
call_rcu_lock(&call_rcu_mutex);
+ alloc_cpu_call_rcu_data();
if (cpu < 0 || maxcpus <= cpu) {
if (!warned) {
fprintf(stderr, "[error] liburcu: set CPU # out of range\n");
errno = EINVAL;
return -EINVAL;
}
- alloc_cpu_call_rcu_data();
- call_rcu_unlock(&call_rcu_mutex);
+
if (per_cpu_call_rcu_data == NULL) {
+ call_rcu_unlock(&call_rcu_mutex);
errno = ENOMEM;
return -ENOMEM;
}
- per_cpu_call_rcu_data[cpu] = crdp;
+
+ if (per_cpu_call_rcu_data[cpu] != NULL && crdp != NULL) {
+ call_rcu_unlock(&call_rcu_mutex);
+ errno = EEXIST;
+ return -EEXIST;
+ }
+
+ rcu_set_pointer(&per_cpu_call_rcu_data[cpu], crdp);
+ call_rcu_unlock(&call_rcu_mutex);
return 0;
}
call_rcu_unlock(&call_rcu_mutex);
return default_call_rcu_data;
}
- call_rcu_data_init(&default_call_rcu_data, 0);
+ call_rcu_data_init(&default_call_rcu_data, 0, -1);
call_rcu_unlock(&call_rcu_mutex);
return default_call_rcu_data;
}
* structure assigned to the CPU on which the thread is running,
* followed by the default call_rcu_data structure. If there is not
* yet a default call_rcu_data structure, one will be created.
+ *
+ * Calls to this function and use of the returned call_rcu_data should
+ * be protected by RCU read-side lock.
*/
struct call_rcu_data *get_call_rcu_data(void)
{
- int curcpu;
- static int warned = 0;
+ struct call_rcu_data *crd;
- if (thread_call_rcu_data != NULL)
- return thread_call_rcu_data;
- if (maxcpus <= 0)
- return get_default_call_rcu_data();
- curcpu = sched_getcpu();
- if (!warned && (curcpu < 0 || maxcpus <= curcpu)) {
- fprintf(stderr, "[error] liburcu: gcrd CPU # out of range\n");
- warned = 1;
+ if (URCU_TLS(thread_call_rcu_data) != NULL)
+ return URCU_TLS(thread_call_rcu_data);
+
+ if (maxcpus > 0) {
+ crd = get_cpu_call_rcu_data(sched_getcpu());
+ if (crd)
+ return crd;
}
- if (curcpu >= 0 && maxcpus > curcpu &&
- per_cpu_call_rcu_data != NULL &&
- per_cpu_call_rcu_data[curcpu] != NULL)
- return per_cpu_call_rcu_data[curcpu];
+
return get_default_call_rcu_data();
}
struct call_rcu_data *get_thread_call_rcu_data(void)
{
- return thread_call_rcu_data;
+ return URCU_TLS(thread_call_rcu_data);
}
/*
void set_thread_call_rcu_data(struct call_rcu_data *crdp)
{
- thread_call_rcu_data = crdp;
+ URCU_TLS(thread_call_rcu_data) = crdp;
}
/*
* Create a separate call_rcu thread for each CPU. This does not
* replace a pre-existing call_rcu thread -- use the set_cpu_call_rcu_data()
- * function if you want that behavior.
+ * function if you want that behavior. Should be paired with
+ * free_all_cpu_call_rcu_data() to teardown these call_rcu worker
+ * threads.
*/
int create_all_cpu_call_rcu_data(unsigned long flags)
call_rcu_unlock(&call_rcu_mutex);
continue;
}
- crdp = __create_call_rcu_data(flags);
+ crdp = __create_call_rcu_data(flags, i);
if (crdp == NULL) {
call_rcu_unlock(&call_rcu_mutex);
errno = ENOMEM;
}
call_rcu_unlock(&call_rcu_mutex);
if ((ret = set_cpu_call_rcu_data(i, crdp)) != 0) {
- /* FIXME: Leaks crdp for now. */
- return ret; /* Can happen on race. */
+ call_rcu_data_free(crdp);
+
+ /* it has been created by other thread */
+ if (ret == -EEXIST)
+ continue;
+
+ return ret;
}
}
return 0;
*/
static void wake_call_rcu_thread(struct call_rcu_data *crdp)
{
- if (!(_CMM_LOAD_SHARED(crdp->flags) & URCU_CALL_RCU_RT)) {
- call_rcu_lock(&crdp->mtx);
- if (!(_CMM_LOAD_SHARED(crdp->flags) & URCU_CALL_RCU_RUNNING)) {
- if (pthread_cond_signal(&crdp->cond) != 0) {
- perror("pthread_cond_signal");
- exit(-1);
- }
- }
- call_rcu_unlock(&crdp->mtx);
- }
+ if (!(_CMM_LOAD_SHARED(crdp->flags) & URCU_CALL_RCU_RT))
+ call_rcu_wake_up(crdp);
}
/*
* need the first invocation of call_rcu() to be fast, make sure
* to create a call_rcu thread first. One way to accomplish this is
* "get_call_rcu_data();", and another is create_all_cpu_call_rcu_data().
+ *
+ * call_rcu must be called by registered RCU read-side threads.
*/
void call_rcu(struct rcu_head *head,
cds_wfq_node_init(&head->next);
head->func = func;
+ /* Holding rcu read-side lock across use of per-cpu crdp */
+ rcu_read_lock();
crdp = get_call_rcu_data();
cds_wfq_enqueue(&crdp->cbs, &head->next);
uatomic_inc(&crdp->qlen);
wake_call_rcu_thread(crdp);
+ rcu_read_unlock();
}
/*
*
* We also silently refuse to free NULL pointers. This simplifies
* the calling code.
+ *
+ * The caller must wait for a grace-period to pass between return from
+ * set_cpu_call_rcu_data() and call to call_rcu_data_free() passing the
+ * previous call rcu data as argument.
*/
void call_rcu_data_free(struct call_rcu_data *crdp)
{
if (crdp == NULL || crdp == default_call_rcu_data) {
return;
}
- if ((crdp->flags & URCU_CALL_RCU_STOPPED) == 0) {
- call_rcu_lock(&crdp->mtx);
- crdp->flags |= URCU_CALL_RCU_STOP;
- call_rcu_unlock(&crdp->mtx);
+ if ((uatomic_read(&crdp->flags) & URCU_CALL_RCU_STOPPED) == 0) {
+ uatomic_or(&crdp->flags, URCU_CALL_RCU_STOP);
wake_call_rcu_thread(crdp);
- while ((crdp->flags & URCU_CALL_RCU_STOPPED) == 0)
+ while ((uatomic_read(&crdp->flags) & URCU_CALL_RCU_STOPPED) == 0)
poll(NULL, 0, 1);
}
if (&crdp->cbs.head != _CMM_LOAD_SHARED(crdp->cbs.tail)) {
_CMM_STORE_SHARED(crdp->cbs.head, NULL);
cbs_tail = (struct cds_wfq_node **)
uatomic_xchg(&crdp->cbs.tail, &crdp->cbs.head);
+ /* Create default call rcu data if need be */
+ (void) get_default_call_rcu_data();
cbs_endprev = (struct cds_wfq_node **)
uatomic_xchg(&default_call_rcu_data, cbs_tail);
*cbs_endprev = cbs;
uatomic_add(&default_call_rcu_data->qlen,
uatomic_read(&crdp->qlen));
- cds_list_del(&crdp->list);
- free(crdp);
+ wake_call_rcu_thread(default_call_rcu_data);
}
+
+ call_rcu_lock(&call_rcu_mutex);
+ cds_list_del(&crdp->list);
+ call_rcu_unlock(&call_rcu_mutex);
+
+ free(crdp);
}
/*
void free_all_cpu_call_rcu_data(void)
{
int cpu;
- struct call_rcu_data *crdp;
+ struct call_rcu_data **crdp;
+ static int warned = 0;
if (maxcpus <= 0)
return;
+
+ crdp = malloc(sizeof(*crdp) * maxcpus);
+ if (!crdp) {
+ if (!warned) {
+ fprintf(stderr, "[error] liburcu: unable to allocate per-CPU pointer array\n");
+ }
+ warned = 1;
+ return;
+ }
+
for (cpu = 0; cpu < maxcpus; cpu++) {
- crdp = get_cpu_call_rcu_data(cpu);
- if (crdp == NULL)
+ crdp[cpu] = get_cpu_call_rcu_data(cpu);
+ if (crdp[cpu] == NULL)
continue;
set_cpu_call_rcu_data(cpu, NULL);
- call_rcu_data_free(crdp);
}
+ /*
+ * Wait for call_rcu sites acting as RCU readers of the
+ * call_rcu_data to become quiescent.
+ */
+ synchronize_rcu();
+ for (cpu = 0; cpu < maxcpus; cpu++) {
+ if (crdp[cpu] == NULL)
+ continue;
+ call_rcu_data_free(crdp[cpu]);
+ }
+ free(crdp);
}
/*
*/
void call_rcu_after_fork_child(void)
{
- struct call_rcu_data *crdp;
+ struct call_rcu_data *crdp, *next;
/* Release the mutex. */
call_rcu_unlock(&call_rcu_mutex);
+ /* Do nothing when call_rcu() has not been used */
+ if (cds_list_empty(&call_rcu_data_list))
+ return;
+
/*
* Allocate a new default call_rcu_data structure in order
* to get a working call_rcu thread to go with it.
default_call_rcu_data = NULL;
(void)get_default_call_rcu_data();
+ /* Cleanup call_rcu_data pointers before use */
+ maxcpus_reset();
+ free(per_cpu_call_rcu_data);
+ rcu_set_pointer(&per_cpu_call_rcu_data, NULL);
+ URCU_TLS(thread_call_rcu_data) = NULL;
+
/* Dispose of all of the rest of the call_rcu_data structures. */
- while (call_rcu_data_list.next != call_rcu_data_list.prev) {
- crdp = cds_list_entry(call_rcu_data_list.prev,
- struct call_rcu_data, list);
+ cds_list_for_each_entry_safe(crdp, next, &call_rcu_data_list, list) {
if (crdp == default_call_rcu_data)
- crdp = cds_list_entry(crdp->list.prev,
- struct call_rcu_data, list);
- crdp->flags = URCU_CALL_RCU_STOPPED;
+ continue;
+ uatomic_set(&crdp->flags, URCU_CALL_RCU_STOPPED);
call_rcu_data_free(crdp);
}
}