X-Git-Url: https://git.liburcu.org/?p=urcu.git;a=blobdiff_plain;f=urcu-call-rcu-impl.h;h=8ed2ab384f81c0806d7c40a74430eb527ad825de;hp=3e947af2fd84bfd6c6843788d9f31a5ff720b876;hb=450b97095e27646fcd1e4b83c99477d7253b987b;hpb=698d07788379df492ad55767bdb84bc24d93c20f

diff --git a/urcu-call-rcu-impl.h b/urcu-call-rcu-impl.h
index 3e947af..8ed2ab3 100644
--- a/urcu-call-rcu-impl.h
+++ b/urcu-call-rcu-impl.h
@@ -40,6 +40,7 @@
 #include "urcu-pointer.h"
 #include "urcu/list.h"
 #include "urcu/futex.h"
+#include "urcu/tls-compat.h"
 
 /* Data structure that identifies a call_rcu thread. */
 
@@ -62,7 +63,7 @@ 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. */
 
@@ -82,12 +83,20 @@ static struct call_rcu_data *default_call_rcu_data;
 
 /*
  * 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)
@@ -104,7 +113,7 @@ 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");
@@ -123,6 +132,10 @@ static void alloc_cpu_call_rcu_data(void)
 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)
 {
 }
@@ -192,7 +205,7 @@ 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);
@@ -220,7 +233,7 @@ static void *call_rcu_thread(void *arg)
 	 */
 	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 */
@@ -321,13 +334,18 @@ static void call_rcu_data_init(struct call_rcu_data **crdpp,
  * 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");
@@ -335,7 +353,7 @@ struct call_rcu_data *get_cpu_call_rcu_data(int cpu)
 	}
 	if (cpu < 0 || maxcpus <= cpu)
 		return NULL;
-	return per_cpu_call_rcu_data[cpu];
+	return rcu_dereference(pcpu_crdp[cpu]);
 }
 
 /*
@@ -379,6 +397,10 @@ 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)
@@ -409,7 +431,7 @@ int set_cpu_call_rcu_data(int cpu, struct call_rcu_data *crdp)
 		return -EEXIST;
 	}
 
-	per_cpu_call_rcu_data[cpu] = crdp;
+	rcu_set_pointer(&per_cpu_call_rcu_data[cpu], crdp);
 	call_rcu_unlock(&call_rcu_mutex);
 	return 0;
 }
@@ -441,13 +463,16 @@ struct call_rcu_data *get_default_call_rcu_data(void)
  * 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)
 {
 	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());
@@ -464,7 +489,7 @@ 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);
 }
 
 /*
@@ -480,7 +505,7 @@ 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;
 }
 
 /*
@@ -555,6 +580,8 @@ static void wake_call_rcu_thread(struct call_rcu_data *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,
@@ -564,10 +591,13 @@ 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();
 }
 
 /*
@@ -587,6 +617,10 @@ 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.
  */
 void call_rcu_data_free(struct call_rcu_data *crdp)
 {
@@ -619,7 +653,10 @@ void call_rcu_data_free(struct call_rcu_data *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);
 }
 
@@ -629,17 +666,38 @@ void call_rcu_data_free(struct call_rcu_data *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);
 }
 
 /*
@@ -674,6 +732,10 @@ void call_rcu_after_fork_child(void)
 	/* 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.
@@ -681,6 +743,12 @@ void call_rcu_after_fork_child(void)
 	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. */
 	cds_list_for_each_entry_safe(crdp, next, &call_rcu_data_list, list) {
 		if (crdp == default_call_rcu_data)