X-Git-Url: http://git.liburcu.org/?p=urcu.git;a=blobdiff_plain;f=urcu-call-rcu-impl.h;h=61aeee611a04da3c78083437bd6af48cbde8d366;hp=29c4e28e334a0654f98725392229dbe98b07645a;hb=1fe734e1914993dfa395e2b81e5c9ee0115cc56c;hpb=276c2fc88c222b5a7c534d21a271564c0bb57935

diff --git a/urcu-call-rcu-impl.h b/urcu-call-rcu-impl.h
index 29c4e28..61aeee6 100644
--- a/urcu-call-rcu-impl.h
+++ b/urcu-call-rcu-impl.h
@@ -21,6 +21,7 @@
  */
 
 #define _GNU_SOURCE
+#define _LGPL_SOURCE
 #include <stdio.h>
 #include <pthread.h>
 #include <signal.h>
@@ -35,16 +36,26 @@
 #include <sched.h>
 
 #include "config.h"
-#include "urcu/wfqueue.h"
+#include "urcu/wfcqueue.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;
+	/*
+	 * We do not align head on a different cache-line than tail
+	 * mainly because call_rcu callback-invocation threads use
+	 * batching ("splice") to get an entire list of callbacks, which
+	 * effectively empties the queue, and requires to touch the tail
+	 * anyway.
+	 */
+	struct cds_wfcq_tail cbs_tail;
+	struct cds_wfcq_head cbs_head;
 	unsigned long flags;
 	int32_t futex;
 	unsigned long qlen; /* maintained for debugging. */
@@ -62,7 +73,7 @@ CDS_LIST_HEAD(call_rcu_data_list);
 
 /* 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. */
 
@@ -150,20 +161,22 @@ static int sched_getcpu(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
@@ -204,7 +217,7 @@ static void call_rcu_wake_up(struct call_rcu_data *crdp)
 {
 	/* Write to call_rcu list before reading/writing futex */
 	cmm_smp_mb();
-	if (unlikely(uatomic_read(&crdp->futex) == -1)) {
+	if (caa_unlikely(uatomic_read(&crdp->futex) == -1)) {
 		uatomic_set(&crdp->futex, 0);
 		futex_async(&crdp->futex, FUTEX_WAKE, 1,
 		      NULL, NULL, 0);
@@ -216,58 +229,53 @@ static void call_rcu_wake_up(struct call_rcu_data *crdp)
 static void *call_rcu_thread(void *arg)
 {
 	unsigned long cbcount;
-	struct cds_wfq_node *cbs;
-	struct cds_wfq_node **cbs_tail;
-	struct call_rcu_data *crdp = (struct call_rcu_data *)arg;
-	struct rcu_head *rhp;
+	struct call_rcu_data *crdp = (struct call_rcu_data *) arg;
 	int rt = !!(uatomic_read(&crdp->flags) & URCU_CALL_RCU_RT);
+	int ret;
 
-	if (set_thread_cpu_affinity(crdp) != 0) {
-		perror("pthread_setaffinity_np");
-		exit(-1);
-	}
+	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();
 
-	thread_call_rcu_data = crdp;
+	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)
-				poll(NULL, 0, 1);
-			_CMM_STORE_SHARED(crdp->cbs.head, NULL);
-			cbs_tail = (struct cds_wfq_node **)
-				uatomic_xchg(&crdp->cbs.tail, &crdp->cbs.head);
+		struct cds_wfcq_head cbs_tmp_head;
+		struct cds_wfcq_tail cbs_tmp_tail;
+		struct cds_wfcq_node *cbs, *cbs_tmp_n;
+
+		cds_wfcq_init(&cbs_tmp_head, &cbs_tmp_tail);
+		__cds_wfcq_splice_blocking(&cbs_tmp_head, &cbs_tmp_tail,
+			&crdp->cbs_head, &crdp->cbs_tail);
+		if (!cds_wfcq_empty(&cbs_tmp_head, &cbs_tmp_tail)) {
 			synchronize_rcu();
 			cbcount = 0;
-			do {
-				while (cbs->next == NULL &&
-				       &cbs->next != cbs_tail)
-				       	poll(NULL, 0, 1);
-				if (cbs == &crdp->cbs.dummy) {
-					cbs = cbs->next;
-					continue;
-				}
-				rhp = (struct rcu_head *)cbs;
-				cbs = cbs->next;
+			__cds_wfcq_for_each_blocking_safe(&cbs_tmp_head,
+					&cbs_tmp_tail, cbs, cbs_tmp_n) {
+				struct rcu_head *rhp;
+
+				rhp = caa_container_of(cbs,
+					struct rcu_head, next);
 				rhp->func(rhp);
 				cbcount++;
-			} while (cbs != NULL);
+			}
 			uatomic_sub(&crdp->qlen, cbcount);
 		}
 		if (uatomic_read(&crdp->flags) & URCU_CALL_RCU_STOP)
 			break;
 		rcu_thread_offline();
 		if (!rt) {
-			if (&crdp->cbs.head
-			    == _CMM_LOAD_SHARED(crdp->cbs.tail)) {
+			if (cds_wfcq_empty(&crdp->cbs_head,
+					&crdp->cbs_tail)) {
 				call_rcu_wait(crdp);
 				poll(NULL, 0, 10);
 				uatomic_dec(&crdp->futex);
@@ -307,14 +315,13 @@ static void call_rcu_data_init(struct call_rcu_data **crdpp,
 			       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);
+	cds_wfcq_init(&crdp->cbs_head, &crdp->cbs_tail);
 	crdp->qlen = 0;
 	crdp->futex = 0;
 	crdp->flags = flags;
@@ -322,10 +329,9 @@ static void call_rcu_data_init(struct call_rcu_data **crdpp,
 	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);
 }
 
 /*
@@ -396,6 +402,10 @@ struct call_rcu_data *create_call_rcu_data(unsigned long flags,
  * 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)
@@ -466,8 +476,8 @@ struct call_rcu_data *get_call_rcu_data(void)
 {
 	struct call_rcu_data *crd;
 
-	if (thread_call_rcu_data != NULL)
-		return thread_call_rcu_data;
+	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());
@@ -484,7 +494,7 @@ struct call_rcu_data *get_call_rcu_data(void)
 
 struct call_rcu_data *get_thread_call_rcu_data(void)
 {
-	return thread_call_rcu_data;
+	return URCU_TLS(thread_call_rcu_data);
 }
 
 /*
@@ -500,7 +510,7 @@ struct call_rcu_data *get_thread_call_rcu_data(void)
 
 void set_thread_call_rcu_data(struct call_rcu_data *crdp)
 {
-	thread_call_rcu_data = crdp;
+	URCU_TLS(thread_call_rcu_data) = crdp;
 }
 
 /*
@@ -584,12 +594,12 @@ void call_rcu(struct rcu_head *head,
 {
 	struct call_rcu_data *crdp;
 
-	cds_wfq_node_init(&head->next);
+	cds_wfcq_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);
+	cds_wfcq_enqueue(&crdp->cbs_head, &crdp->cbs_tail, &head->next);
 	uatomic_inc(&crdp->qlen);
 	wake_call_rcu_thread(crdp);
 	rcu_read_unlock();
@@ -612,13 +622,17 @@ void call_rcu(struct rcu_head *head,
  *
  * 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.
+ *
+ * Note: introducing __cds_wfcq_splice_blocking() in this function fixed
+ * a list corruption bug in the 0.7.x series. The equivalent fix
+ * appeared in 0.6.8 for the stable-0.6 branch.
  */
 void call_rcu_data_free(struct call_rcu_data *crdp)
 {
-	struct cds_wfq_node *cbs;
-	struct cds_wfq_node **cbs_tail;
-	struct cds_wfq_node **cbs_endprev;
-
 	if (crdp == NULL || crdp == default_call_rcu_data) {
 		return;
 	}
@@ -628,17 +642,12 @@ void call_rcu_data_free(struct call_rcu_data *crdp)
 		while ((uatomic_read(&crdp->flags) & URCU_CALL_RCU_STOPPED) == 0)
 			poll(NULL, 0, 1);
 	}
-	if (&crdp->cbs.head != _CMM_LOAD_SHARED(crdp->cbs.tail)) {
-		while ((cbs = _CMM_LOAD_SHARED(crdp->cbs.head)) == NULL)
-			poll(NULL, 0, 1);
-		_CMM_STORE_SHARED(crdp->cbs.head, NULL);
-		cbs_tail = (struct cds_wfq_node **)
-			uatomic_xchg(&crdp->cbs.tail, &crdp->cbs.head);
+	if (!cds_wfcq_empty(&crdp->cbs_head, &crdp->cbs_tail)) {
 		/* 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;
+		__cds_wfcq_splice_blocking(&default_call_rcu_data->cbs_head,
+			&default_call_rcu_data->cbs_tail,
+			&crdp->cbs_head, &crdp->cbs_tail);
 		uatomic_add(&default_call_rcu_data->qlen,
 			    uatomic_read(&crdp->qlen));
 		wake_call_rcu_thread(default_call_rcu_data);
@@ -738,7 +747,7 @@ void call_rcu_after_fork_child(void)
 	maxcpus_reset();
 	free(per_cpu_call_rcu_data);
 	rcu_set_pointer(&per_cpu_call_rcu_data, NULL);
-	thread_call_rcu_data = NULL;
+	URCU_TLS(thread_call_rcu_data) = NULL;
 
 	/* Dispose of all of the rest of the call_rcu_data structures. */
 	cds_list_for_each_entry_safe(crdp, next, &call_rcu_data_list, list) {