[lttng-dev] [RFC PATCH urcu] Use urcu/tls-compat.h

Mathieu Desnoyers mathieu.desnoyers at efficios.com
Wed May 16 09:58:02 EDT 2012


Provides compatibility for OpenBSD, NetBSD and Darwin.

Suggested-by: Marek Vavruša <marek.vavrusa at nic.cz>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers at efficios.com>
---
diff --git a/tests/test_mutex.c b/tests/test_mutex.c
index 3f84bbf..ba184a4 100644
--- a/tests/test_mutex.c
+++ b/tests/test_mutex.c
@@ -35,6 +35,7 @@
 #include <errno.h>
 
 #include <urcu/arch.h>
+#include <urcu/tls-compat.h>
 
 #ifdef __linux__
 #include <syscall.h>
@@ -155,8 +156,8 @@ static int test_duration_read(void)
 	return !test_stop;
 }
 
-static unsigned long long __thread nr_writes;
-static unsigned long long __thread nr_reads;
+static DEFINE_URCU_TLS(unsigned long long, nr_writes);
+static DEFINE_URCU_TLS(unsigned long long, nr_reads);
 
 static
 unsigned long long __attribute__((aligned(CAA_CACHE_LINE_SIZE))) *tot_nr_writes;
@@ -208,12 +209,12 @@ void *thr_reader(void *data)
 		if (caa_unlikely(rduration))
 			loop_sleep(rduration);
 		pthread_mutex_unlock(&lock);
-		nr_reads++;
+		URCU_TLS(nr_reads)++;
 		if (caa_unlikely(!test_duration_read()))
 			break;
 	}
 
-	tot_nr_reads[tidx] = nr_reads;
+	tot_nr_reads[tidx] = URCU_TLS(nr_reads);
 	printf_verbose("thread_end %s, thread id : %lx, tid %lu\n",
 			"reader", pthread_self(), (unsigned long)gettid());
 	return ((void*)1);
@@ -241,7 +242,7 @@ void *thr_writer(void *data)
 		if (caa_unlikely(wduration))
 			loop_sleep(wduration);
 		pthread_mutex_unlock(&lock);
-		nr_writes++;
+		URCU_TLS(nr_writes)++;
 		if (caa_unlikely(!test_duration_write()))
 			break;
 		if (caa_unlikely(wdelay))
@@ -250,7 +251,7 @@ void *thr_writer(void *data)
 
 	printf_verbose("thread_end %s, thread id : %lx, tid %lu\n",
 			"writer", pthread_self(), (unsigned long)gettid());
-	tot_nr_writes[wtidx] = nr_writes;
+	tot_nr_writes[wtidx] = URCU_TLS(nr_writes);
 	return ((void*)2);
 }
 
diff --git a/tests/test_perthreadlock.c b/tests/test_perthreadlock.c
index fa9c89a..6ff0412 100644
--- a/tests/test_perthreadlock.c
+++ b/tests/test_perthreadlock.c
@@ -35,6 +35,7 @@
 #include <errno.h>
 
 #include <urcu/arch.h>
+#include <urcu/tls-compat.h>
 
 #ifdef __linux__
 #include <syscall.h>
@@ -159,8 +160,8 @@ static int test_duration_read(void)
 	return !test_stop;
 }
 
-static unsigned long long __thread nr_writes;
-static unsigned long long __thread nr_reads;
+static DEFINE_URCU_TLS(unsigned long long, nr_writes);
+static DEFINE_URCU_TLS(unsigned long long, nr_reads);
 
 static
 unsigned long long __attribute__((aligned(CAA_CACHE_LINE_SIZE))) *tot_nr_writes;
@@ -212,12 +213,12 @@ void *thr_reader(void *data)
 		if (caa_unlikely(rduration))
 			loop_sleep(rduration);
 		pthread_mutex_unlock(&per_thread_lock[tidx].lock);
-		nr_reads++;
+		URCU_TLS(nr_reads)++;
 		if (caa_unlikely(!test_duration_read()))
 			break;
 	}
 
-	tot_nr_reads[tidx] = nr_reads;
+	tot_nr_reads[tidx] = URCU_TLS(nr_reads);
 	printf_verbose("thread_end %s, thread id : %lx, tid %lu\n",
 			"reader", pthread_self(), (unsigned long)gettid());
 	return ((void*)1);
@@ -250,7 +251,7 @@ void *thr_writer(void *data)
 		for (tidx = (long)nr_readers - 1; tidx >= 0; tidx--) {
 			pthread_mutex_unlock(&per_thread_lock[tidx].lock);
 		}
-		nr_writes++;
+		URCU_TLS(nr_writes)++;
 		if (caa_unlikely(!test_duration_write()))
 			break;
 		if (caa_unlikely(wdelay))
@@ -259,7 +260,7 @@ void *thr_writer(void *data)
 
 	printf_verbose("thread_end %s, thread id : %lx, tid %lu\n",
 			"writer", pthread_self(), (unsigned long)gettid());
-	tot_nr_writes[wtidx] = nr_writes;
+	tot_nr_writes[wtidx] = URCU_TLS(nr_writes);
 	return ((void*)2);
 }
 
diff --git a/tests/test_rwlock.c b/tests/test_rwlock.c
index 34d8c07..ccbc9d1 100644
--- a/tests/test_rwlock.c
+++ b/tests/test_rwlock.c
@@ -35,6 +35,7 @@
 #include <errno.h>
 
 #include <urcu/arch.h>
+#include <urcu/tls-compat.h>
 
 #ifdef __linux__
 #include <syscall.h>
@@ -156,8 +157,8 @@ static int test_duration_read(void)
 	return !test_stop;
 }
 
-static unsigned long long __thread nr_writes;
-static unsigned long long __thread nr_reads;
+static DEFINE_URCU_TLS(unsigned long long, nr_writes);
+static DEFINE_URCU_TLS(unsigned long long, nr_reads);
 
 static unsigned int nr_readers;
 static unsigned int nr_writers;
@@ -204,12 +205,12 @@ void *thr_reader(void *_count)
 		if (caa_unlikely(rduration))
 			loop_sleep(rduration);
 		pthread_rwlock_unlock(&lock);
-		nr_reads++;
+		URCU_TLS(nr_reads)++;
 		if (caa_unlikely(!test_duration_read()))
 			break;
 	}
 
-	*count = nr_reads;
+	*count = URCU_TLS(nr_reads);
 	printf_verbose("thread_end %s, thread id : %lx, tid %lu\n",
 			"reader", pthread_self(), (unsigned long)gettid());
 	return ((void*)1);
@@ -237,7 +238,7 @@ void *thr_writer(void *_count)
 		if (caa_unlikely(wduration))
 			loop_sleep(wduration);
 		pthread_rwlock_unlock(&lock);
-		nr_writes++;
+		URCU_TLS(nr_writes)++;
 		if (caa_unlikely(!test_duration_write()))
 			break;
 		if (caa_unlikely(wdelay))
@@ -246,7 +247,7 @@ void *thr_writer(void *_count)
 
 	printf_verbose("thread_end %s, thread id : %lx, tid %lu\n",
 			"writer", pthread_self(), (unsigned long)gettid());
-	*count = nr_writes;
+	*count = URCU_TLS(nr_writes);
 	return ((void*)2);
 }
 
diff --git a/tests/test_urcu.c b/tests/test_urcu.c
index 870f133..1b1b94b 100644
--- a/tests/test_urcu.c
+++ b/tests/test_urcu.c
@@ -35,6 +35,7 @@
 #include <errno.h>
 
 #include <urcu/arch.h>
+#include <urcu/tls-compat.h>
 
 #ifdef __linux__
 #include <syscall.h>
@@ -154,8 +155,8 @@ static int test_duration_read(void)
 	return !test_stop;
 }
 
-static unsigned long long __thread nr_writes;
-static unsigned long long __thread nr_reads;
+static DEFINE_URCU_TLS(unsigned long long, nr_writes);
+static DEFINE_URCU_TLS(unsigned long long, nr_reads);
 
 static unsigned int nr_readers;
 static unsigned int nr_writers;
@@ -242,7 +243,7 @@ void *thr_reader(void *_count)
 		if (caa_unlikely(rduration))
 			loop_sleep(rduration);
 		rcu_read_unlock();
-		nr_reads++;
+		URCU_TLS(nr_reads)++;
 		if (caa_unlikely(!test_duration_read()))
 			break;
 	}
@@ -253,7 +254,7 @@ void *thr_reader(void *_count)
 	rcu_register_thread();
 	rcu_unregister_thread();
 
-	*count = nr_reads;
+	*count = URCU_TLS(nr_reads);
 	printf_verbose("thread_end %s, thread id : %lx, tid %lu\n",
 			"reader", pthread_self(), (unsigned long)gettid());
 	return ((void*)1);
@@ -287,7 +288,7 @@ void *thr_writer(void *_count)
 			old->a = 0;
 		test_array_free(old);
 		rcu_copy_mutex_unlock();
-		nr_writes++;
+		URCU_TLS(nr_writes)++;
 		if (caa_unlikely(!test_duration_write()))
 			break;
 		if (caa_unlikely(wdelay))
@@ -296,7 +297,7 @@ void *thr_writer(void *_count)
 
 	printf_verbose("thread_end %s, thread id : %lx, tid %lu\n",
 			"writer", pthread_self(), (unsigned long)gettid());
-	*count = nr_writes;
+	*count = URCU_TLS(nr_writes);
 	return ((void*)2);
 }
 
diff --git a/tests/test_urcu_assign.c b/tests/test_urcu_assign.c
index 42d70c2..31e22e5 100644
--- a/tests/test_urcu_assign.c
+++ b/tests/test_urcu_assign.c
@@ -35,6 +35,7 @@
 #include <errno.h>
 
 #include <urcu/arch.h>
+#include <urcu/tls-compat.h>
 
 #ifdef __linux__
 #include <syscall.h>
@@ -154,8 +155,8 @@ static int test_duration_read(void)
 	return !test_stop;
 }
 
-static unsigned long long __thread nr_writes;
-static unsigned long long __thread nr_reads;
+static DEFINE_URCU_TLS(unsigned long long, nr_writes);
+static DEFINE_URCU_TLS(unsigned long long, nr_reads);
 
 static unsigned int nr_readers;
 static unsigned int nr_writers;
@@ -242,14 +243,14 @@ void *thr_reader(void *_count)
 		if (caa_unlikely(rduration))
 			loop_sleep(rduration);
 		rcu_read_unlock();
-		nr_reads++;
+		URCU_TLS(nr_reads)++;
 		if (caa_unlikely(!test_duration_read()))
 			break;
 	}
 
 	rcu_unregister_thread();
 
-	*count = nr_reads;
+	*count = URCU_TLS(nr_reads);
 	printf_verbose("thread_end %s, thread id : %lx, tid %lu\n",
 			"reader", pthread_self(), (unsigned long)gettid());
 	return ((void*)1);
@@ -284,7 +285,7 @@ void *thr_writer(void *_count)
 			old->a = 0;
 		test_array_free(old);
 		rcu_copy_mutex_unlock();
-		nr_writes++;
+		URCU_TLS(nr_writes)++;
 		if (caa_unlikely(!test_duration_write()))
 			break;
 		if (caa_unlikely(wdelay))
@@ -293,7 +294,7 @@ void *thr_writer(void *_count)
 
 	printf_verbose("thread_end %s, thread id : %lx, tid %lu\n",
 			"writer", pthread_self(), (unsigned long)gettid());
-	*count = nr_writes;
+	*count = URCU_TLS(nr_writes);
 	return ((void*)2);
 }
 
diff --git a/tests/test_urcu_bp.c b/tests/test_urcu_bp.c
index 857913f..41caeb0 100644
--- a/tests/test_urcu_bp.c
+++ b/tests/test_urcu_bp.c
@@ -35,6 +35,7 @@
 #include <errno.h>
 
 #include <urcu/arch.h>
+#include <urcu/tls-compat.h>
 
 #ifdef __linux__
 #include <syscall.h>
@@ -154,8 +155,8 @@ static int test_duration_read(void)
 	return !test_stop;
 }
 
-static unsigned long long __thread nr_writes;
-static unsigned long long __thread nr_reads;
+static DEFINE_URCU_TLS(unsigned long long, nr_writes);
+static DEFINE_URCU_TLS(unsigned long long, nr_reads);
 
 static unsigned int nr_readers;
 static unsigned int nr_writers;
@@ -242,14 +243,14 @@ void *thr_reader(void *_count)
 		if (caa_unlikely(rduration))
 			loop_sleep(rduration);
 		rcu_read_unlock();
-		nr_reads++;
+		URCU_TLS(nr_reads)++;
 		if (caa_unlikely(!test_duration_read()))
 			break;
 	}
 
 	rcu_unregister_thread();
 
-	*count = nr_reads;
+	*count = URCU_TLS(nr_reads);
 	printf_verbose("thread_end %s, thread id : %lx, tid %lu\n",
 			"reader", pthread_self(), (unsigned long)gettid());
 	return ((void*)1);
@@ -283,7 +284,7 @@ void *thr_writer(void *_count)
 			old->a = 0;
 		test_array_free(old);
 		rcu_copy_mutex_unlock();
-		nr_writes++;
+		URCU_TLS(nr_writes)++;
 		if (caa_unlikely(!test_duration_write()))
 			break;
 		if (caa_unlikely(wdelay))
@@ -292,7 +293,7 @@ void *thr_writer(void *_count)
 
 	printf_verbose("thread_end %s, thread id : %lx, tid %lu\n",
 			"writer", pthread_self(), (unsigned long)gettid());
-	*count = nr_writes;
+	*count = URCU_TLS(nr_writes);
 	return ((void*)2);
 }
 
diff --git a/tests/test_urcu_defer.c b/tests/test_urcu_defer.c
index 1575e9c..cd9780b 100644
--- a/tests/test_urcu_defer.c
+++ b/tests/test_urcu_defer.c
@@ -35,6 +35,7 @@
 #include <errno.h>
 
 #include <urcu/arch.h>
+#include <urcu/tls-compat.h>
 
 #ifdef __linux__
 #include <syscall.h>
@@ -155,8 +156,8 @@ static int test_duration_read(void)
 	return !test_stop;
 }
 
-static unsigned long long __thread nr_writes;
-static unsigned long long __thread nr_reads;
+static DEFINE_URCU_TLS(unsigned long long, nr_writes);
+static DEFINE_URCU_TLS(unsigned long long, nr_reads);
 
 static
 unsigned long long __attribute__((aligned(CAA_CACHE_LINE_SIZE))) *tot_nr_writes;
@@ -213,14 +214,14 @@ void *thr_reader(void *_count)
 		if (caa_unlikely(rduration))
 			loop_sleep(rduration);
 		rcu_read_unlock();
-		nr_reads++;
+		URCU_TLS(nr_reads)++;
 		if (caa_unlikely(!test_duration_read()))
 			break;
 	}
 
 	rcu_unregister_thread();
 
-	*count = nr_reads;
+	*count = URCU_TLS(nr_reads);
 	printf_verbose("thread_end %s, thread id : %lx, tid %lu\n",
 			"reader", pthread_self(), (unsigned long)gettid());
 	return ((void*)1);
@@ -271,7 +272,7 @@ void *thr_writer(void *data)
 		defer_rcu(test_cb2, (void *)-2L);
 		defer_rcu(test_cb2, (void *)-4L);
 		defer_rcu(test_cb2, (void *)-2L);
-		nr_writes++;
+		URCU_TLS(nr_writes)++;
 		if (caa_unlikely(!test_duration_write()))
 			break;
 		if (caa_unlikely(wdelay))
@@ -282,7 +283,7 @@ void *thr_writer(void *data)
 
 	printf_verbose("thread_end %s, thread id : %lx, tid %lu\n",
 			"writer", pthread_self(), (unsigned long)gettid());
-	tot_nr_writes[wtidx] = nr_writes;
+	tot_nr_writes[wtidx] = URCU_TLS(nr_writes);
 	return ((void*)2);
 }
 
diff --git a/tests/test_urcu_gc.c b/tests/test_urcu_gc.c
index 21c5d56..3a42506 100644
--- a/tests/test_urcu_gc.c
+++ b/tests/test_urcu_gc.c
@@ -35,6 +35,7 @@
 #include <errno.h>
 
 #include <urcu/arch.h>
+#include <urcu/tls-compat.h>
 
 #ifdef __linux__
 #include <syscall.h>
@@ -163,8 +164,8 @@ static int test_duration_read(void)
 	return !test_stop;
 }
 
-static unsigned long long __thread nr_writes;
-static unsigned long long __thread nr_reads;
+static DEFINE_URCU_TLS(unsigned long long, nr_writes);
+static DEFINE_URCU_TLS(unsigned long long, nr_reads);
 
 static
 unsigned long long __attribute__((aligned(CAA_CACHE_LINE_SIZE))) *tot_nr_writes;
@@ -221,14 +222,14 @@ void *thr_reader(void *_count)
 		if (caa_unlikely(rduration))
 			loop_sleep(rduration);
 		rcu_read_unlock();
-		nr_reads++;
+		URCU_TLS(nr_reads)++;
 		if (caa_unlikely(!test_duration_read()))
 			break;
 	}
 
 	rcu_unregister_thread();
 
-	*count = nr_reads;
+	*count = URCU_TLS(nr_reads);
 	printf_verbose("thread_end %s, thread id : %lx, tid %lu\n",
 			"reader", pthread_self(), (unsigned long)gettid());
 	return ((void*)1);
@@ -294,7 +295,7 @@ void *thr_writer(void *data)
 		if (caa_unlikely(wduration))
 			loop_sleep(wduration);
 		rcu_gc_reclaim(wtidx, old);
-		nr_writes++;
+		URCU_TLS(nr_writes)++;
 		if (caa_unlikely(!test_duration_write()))
 			break;
 		if (caa_unlikely(wdelay))
@@ -303,7 +304,7 @@ void *thr_writer(void *data)
 
 	printf_verbose("thread_end %s, thread id : %lx, tid %lu\n",
 			"writer", pthread_self(), (unsigned long)gettid());
-	tot_nr_writes[wtidx] = nr_writes;
+	tot_nr_writes[wtidx] = URCU_TLS(nr_writes);
 	return ((void*)2);
 }
 
diff --git a/tests/test_urcu_hash.c b/tests/test_urcu_hash.c
index 2223413..5710de0 100644
--- a/tests/test_urcu_hash.c
+++ b/tests/test_urcu_hash.c
@@ -82,13 +82,13 @@ int (*get_populate_hash_cb(void))(void)
 	return test_hash_cb[test_choice].populate_hash;
 }
 
-unsigned int __thread rand_lookup;
-unsigned long __thread nr_add;
-unsigned long __thread nr_addexist;
-unsigned long __thread nr_del;
-unsigned long __thread nr_delnoent;
-unsigned long __thread lookup_fail;
-unsigned long __thread lookup_ok;
+DEFINE_URCU_TLS(unsigned int, rand_lookup);
+DEFINE_URCU_TLS(unsigned long, nr_add);
+DEFINE_URCU_TLS(unsigned long, nr_addexist);
+DEFINE_URCU_TLS(unsigned long, nr_del);
+DEFINE_URCU_TLS(unsigned long, nr_delnoent);
+DEFINE_URCU_TLS(unsigned long, lookup_fail);
+DEFINE_URCU_TLS(unsigned long, lookup_ok);
 
 struct cds_lfht *test_ht;
 
@@ -126,8 +126,8 @@ int use_affinity = 0;
 
 pthread_mutex_t affinity_mutex = PTHREAD_MUTEX_INITIALIZER;
 
-unsigned long long __thread nr_writes;
-unsigned long long __thread nr_reads;
+DEFINE_URCU_TLS(unsigned long long, nr_writes);
+DEFINE_URCU_TLS(unsigned long long, nr_reads);
 
 unsigned int nr_readers;
 unsigned int nr_writers;
diff --git a/tests/test_urcu_hash.h b/tests/test_urcu_hash.h
index 083e71c..fe13c36 100644
--- a/tests/test_urcu_hash.h
+++ b/tests/test_urcu_hash.h
@@ -37,6 +37,8 @@
 #include <errno.h>
 #include <signal.h>
 
+#include <urcu/tls-compat.h>
+
 #ifdef __linux__
 #include <syscall.h>
 #endif
@@ -102,13 +104,13 @@ struct wr_count {
 	unsigned long remove;
 };
 
-extern unsigned int __thread rand_lookup;
-extern unsigned long __thread nr_add;
-extern unsigned long __thread nr_addexist;
-extern unsigned long __thread nr_del;
-extern unsigned long __thread nr_delnoent;
-extern unsigned long __thread lookup_fail;
-extern unsigned long __thread lookup_ok;
+extern DECLARE_URCU_TLS(unsigned int, rand_lookup);
+extern DECLARE_URCU_TLS(unsigned long, nr_add);
+extern DECLARE_URCU_TLS(unsigned long, nr_addexist);
+extern DECLARE_URCU_TLS(unsigned long, nr_del);
+extern DECLARE_URCU_TLS(unsigned long, nr_delnoent);
+extern DECLARE_URCU_TLS(unsigned long, lookup_fail);
+extern DECLARE_URCU_TLS(unsigned long, lookup_ok);
 
 extern struct cds_lfht *test_ht;
 
@@ -214,8 +216,8 @@ static inline int test_duration_read(void)
 	return !test_stop;
 }
 
-extern unsigned long long __thread nr_writes;
-extern unsigned long long __thread nr_reads;
+extern DECLARE_URCU_TLS(unsigned long long, nr_writes);
+extern DECLARE_URCU_TLS(unsigned long long, nr_reads);
 
 extern unsigned int nr_readers;
 extern unsigned int nr_writers;
diff --git a/tests/test_urcu_hash_rw.c b/tests/test_urcu_hash_rw.c
index 7310344..a789236 100644
--- a/tests/test_urcu_hash_rw.c
+++ b/tests/test_urcu_hash_rw.c
@@ -80,7 +80,7 @@ void *test_hash_rw_thr_reader(void *_count)
 	for (;;) {
 		rcu_read_lock();
 		cds_lfht_test_lookup(test_ht,
-			(void *)(((unsigned long) rand_r(&rand_lookup) % lookup_pool_size) + lookup_pool_offset),
+			(void *)(((unsigned long) rand_r(&URCU_TLS(rand_lookup)) % lookup_pool_size) + lookup_pool_offset),
 			sizeof(void *), &iter);
 		node = cds_lfht_iter_get_test_node(&iter);
 		if (node == NULL) {
@@ -88,28 +88,29 @@ void *test_hash_rw_thr_reader(void *_count)
 				printf("[ERROR] Lookup cannot find initial node.\n");
 				exit(-1);
 			}
-			lookup_fail++;
+			URCU_TLS(lookup_fail)++;
 		} else {
-			lookup_ok++;
+			URCU_TLS(lookup_ok)++;
 		}
 		debug_yield_read();
 		if (caa_unlikely(rduration))
 			loop_sleep(rduration);
 		rcu_read_unlock();
-		nr_reads++;
+		URCU_TLS(nr_reads)++;
 		if (caa_unlikely(!test_duration_read()))
 			break;
-		if (caa_unlikely((nr_reads & ((1 << 10) - 1)) == 0))
+		if (caa_unlikely((URCU_TLS(nr_reads) & ((1 << 10) - 1)) == 0))
 			rcu_quiescent_state();
 	}
 
 	rcu_unregister_thread();
 
-	*count = nr_reads;
+	*count = URCU_TLS(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);
+			pthread_self(), URCU_TLS(lookup_fail),
+			URCU_TLS(lookup_ok));
 	return ((void*)1);
 
 }
@@ -136,10 +137,10 @@ void *test_hash_rw_thr_writer(void *_count)
 
 	for (;;) {
 		if ((addremove == AR_ADD || add_only)
-				|| (addremove == AR_RANDOM && rand_r(&rand_lookup) & 1)) {
+				|| (addremove == AR_RANDOM && rand_r(&URCU_TLS(rand_lookup)) & 1)) {
 			node = malloc(sizeof(struct lfht_test_node));
 			lfht_test_node_init(node,
-				(void *)(((unsigned long) rand_r(&rand_lookup) % write_pool_size) + write_pool_offset),
+				(void *)(((unsigned long) rand_r(&URCU_TLS(rand_lookup)) % write_pool_size) + write_pool_offset),
 				sizeof(void *));
 			rcu_read_lock();
 			if (add_unique) {
@@ -159,36 +160,36 @@ void *test_hash_rw_thr_writer(void *_count)
 			rcu_read_unlock();
 			if (add_unique && ret_node != &node->node) {
 				free(node);
-				nr_addexist++;
+				URCU_TLS(nr_addexist)++;
 			} else {
 				if (add_replace && ret_node) {
 					call_rcu(&to_test_node(ret_node)->head,
 							free_node_cb);
-					nr_addexist++;
+					URCU_TLS(nr_addexist)++;
 				} else {
-					nr_add++;
+					URCU_TLS(nr_add)++;
 				}
 			}
 		} else {
 			/* May delete */
 			rcu_read_lock();
 			cds_lfht_test_lookup(test_ht,
-				(void *)(((unsigned long) rand_r(&rand_lookup) % write_pool_size) + write_pool_offset),
+				(void *)(((unsigned long) rand_r(&URCU_TLS(rand_lookup)) % write_pool_size) + write_pool_offset),
 				sizeof(void *), &iter);
 			ret = cds_lfht_del(test_ht, cds_lfht_iter_get_node(&iter));
 			rcu_read_unlock();
 			if (ret == 0) {
 				node = cds_lfht_iter_get_test_node(&iter);
 				call_rcu(&node->head, free_node_cb);
-				nr_del++;
+				URCU_TLS(nr_del)++;
 			} else
-				nr_delnoent++;
+				URCU_TLS(nr_delnoent)++;
 		}
 #if 0
-		//if (nr_writes % 100000 == 0) {
-		if (nr_writes % 1000 == 0) {
+		//if (URCU_TLS(nr_writes) % 100000 == 0) {
+		if (URCU_TLS(nr_writes) % 1000 == 0) {
 			rcu_read_lock();
-			if (rand_r(&rand_lookup) & 1) {
+			if (rand_r(&URCU_TLS(rand_lookup)) & 1) {
 				ht_resize(test_ht, 1);
 			} else {
 				ht_resize(test_ht, -1);
@@ -196,12 +197,12 @@ void *test_hash_rw_thr_writer(void *_count)
 			rcu_read_unlock();
 		}
 #endif //0
-		nr_writes++;
+		URCU_TLS(nr_writes)++;
 		if (caa_unlikely(!test_duration_write()))
 			break;
 		if (caa_unlikely(wdelay))
 			loop_sleep(wdelay);
-		if (caa_unlikely((nr_writes & ((1 << 10) - 1)) == 0))
+		if (caa_unlikely((URCU_TLS(nr_writes) & ((1 << 10) - 1)) == 0))
 			rcu_quiescent_state();
 	}
 
@@ -210,12 +211,13 @@ void *test_hash_rw_thr_writer(void *_count)
 	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->update_ops = nr_writes;
-	count->add = nr_add;
-	count->add_exist = nr_addexist;
-	count->remove = nr_del;
+			"nr_delnoent %lu\n", pthread_self(), URCU_TLS(nr_add),
+			URCU_TLS(nr_addexist), URCU_TLS(nr_del),
+			URCU_TLS(nr_delnoent));
+	count->update_ops = URCU_TLS(nr_writes);
+	count->add = URCU_TLS(nr_add);
+	count->add_exist = URCU_TLS(nr_addexist);
+	count->remove = URCU_TLS(nr_del);
 	return ((void*)2);
 }
 
@@ -235,10 +237,10 @@ int test_hash_rw_populate_hash(void)
 "larger random pool (-p option). This may take a while...\n", init_populate, init_pool_size);
 	}
 
-	while (nr_add < init_populate) {
+	while (URCU_TLS(nr_add) < init_populate) {
 		node = malloc(sizeof(struct lfht_test_node));
 		lfht_test_node_init(node,
-			(void *)(((unsigned long) rand_r(&rand_lookup) % init_pool_size) + init_pool_offset),
+			(void *)(((unsigned long) rand_r(&URCU_TLS(rand_lookup)) % init_pool_size) + init_pool_offset),
 			sizeof(void *));
 		rcu_read_lock();
 		if (add_unique) {
@@ -258,16 +260,16 @@ int test_hash_rw_populate_hash(void)
 		rcu_read_unlock();
 		if (add_unique && ret_node != &node->node) {
 			free(node);
-			nr_addexist++;
+			URCU_TLS(nr_addexist)++;
 		} else {
 			if (add_replace && ret_node) {
 				call_rcu(&to_test_node(ret_node)->head, free_node_cb);
-				nr_addexist++;
+				URCU_TLS(nr_addexist)++;
 			} else {
-				nr_add++;
+				URCU_TLS(nr_add)++;
 			}
 		}
-		nr_writes++;
+		URCU_TLS(nr_writes)++;
 	}
 	return 0;
 }
diff --git a/tests/test_urcu_hash_unique.c b/tests/test_urcu_hash_unique.c
index c934feb..610f479 100644
--- a/tests/test_urcu_hash_unique.c
+++ b/tests/test_urcu_hash_unique.c
@@ -98,20 +98,21 @@ void *test_hash_unique_thr_reader(void *_count)
 		debug_yield_read();
 		if (caa_unlikely(rduration))
 			loop_sleep(rduration);
-		nr_reads++;
+		URCU_TLS(nr_reads)++;
 		if (caa_unlikely(!test_duration_read()))
 			break;
-		if (caa_unlikely((nr_reads & ((1 << 10) - 1)) == 0))
+		if (caa_unlikely((URCU_TLS(nr_reads) & ((1 << 10) - 1)) == 0))
 			rcu_quiescent_state();
 	}
 
 	rcu_unregister_thread();
 
-	*count = nr_reads;
+	*count = URCU_TLS(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);
+			pthread_self(), URCU_TLS(lookup_fail),
+			URCU_TLS(lookup_ok));
 	return ((void*)1);
 
 }
@@ -142,13 +143,13 @@ void *test_hash_unique_thr_writer(void *_count)
 		 * add unique/add replace with new node key from range.
 		 */
 		if (1 || (addremove == AR_ADD || add_only)
-				|| (addremove == AR_RANDOM && rand_r(&rand_lookup) & 1)) {
+				|| (addremove == AR_RANDOM && rand_r(&URCU_TLS(rand_lookup)) & 1)) {
 			node = malloc(sizeof(struct lfht_test_node));
 			lfht_test_node_init(node,
-				(void *)(((unsigned long) rand_r(&rand_lookup) % write_pool_size) + write_pool_offset),
+				(void *)(((unsigned long) rand_r(&URCU_TLS(rand_lookup)) % write_pool_size) + write_pool_offset),
 				sizeof(void *));
 			rcu_read_lock();
-			loc_add_unique = rand_r(&rand_lookup) & 1;
+			loc_add_unique = rand_r(&URCU_TLS(rand_lookup)) & 1;
 			if (loc_add_unique) {
 				ret_node = cds_lfht_add_unique(test_ht,
 					test_hash(node->key, node->key_len, TEST_HASH_SEED),
@@ -168,39 +169,39 @@ void *test_hash_unique_thr_writer(void *_count)
 			if (loc_add_unique) {
 				if (ret_node != &node->node) {
 					free(node);
-					nr_addexist++;
+					URCU_TLS(nr_addexist)++;
 				} else {
-					nr_add++;
+					URCU_TLS(nr_add)++;
 				}
 			} else {
 				if (ret_node) {
 					call_rcu(&to_test_node(ret_node)->head,
 							free_node_cb);
-					nr_addexist++;
+					URCU_TLS(nr_addexist)++;
 				} else {
-					nr_add++;
+					URCU_TLS(nr_add)++;
 				}
 			}
 		} else {
 			/* May delete */
 			rcu_read_lock();
 			cds_lfht_test_lookup(test_ht,
-				(void *)(((unsigned long) rand_r(&rand_lookup) % write_pool_size) + write_pool_offset),
+				(void *)(((unsigned long) rand_r(&URCU_TLS(rand_lookup)) % write_pool_size) + write_pool_offset),
 				sizeof(void *), &iter);
 			ret = cds_lfht_del(test_ht, cds_lfht_iter_get_node(&iter));
 			rcu_read_unlock();
 			if (ret == 0) {
 				node = cds_lfht_iter_get_test_node(&iter);
 				call_rcu(&node->head, free_node_cb);
-				nr_del++;
+				URCU_TLS(nr_del)++;
 			} else
-				nr_delnoent++;
+				URCU_TLS(nr_delnoent)++;
 		}
 #if 0
-		//if (nr_writes % 100000 == 0) {
-		if (nr_writes % 1000 == 0) {
+		//if (URCU_TLS(nr_writes) % 100000 == 0) {
+		if (URCU_TLS(nr_writes) % 1000 == 0) {
 			rcu_read_lock();
-			if (rand_r(&rand_lookup) & 1) {
+			if (rand_r(&URCU_TLS(rand_lookup)) & 1) {
 				ht_resize(test_ht, 1);
 			} else {
 				ht_resize(test_ht, -1);
@@ -208,12 +209,12 @@ void *test_hash_unique_thr_writer(void *_count)
 			rcu_read_unlock();
 		}
 #endif //0
-		nr_writes++;
+		URCU_TLS(nr_writes)++;
 		if (caa_unlikely(!test_duration_write()))
 			break;
 		if (caa_unlikely(wdelay))
 			loop_sleep(wdelay);
-		if (caa_unlikely((nr_writes & ((1 << 10) - 1)) == 0))
+		if (caa_unlikely((URCU_TLS(nr_writes) & ((1 << 10) - 1)) == 0))
 			rcu_quiescent_state();
 	}
 
@@ -222,12 +223,13 @@ void *test_hash_unique_thr_writer(void *_count)
 	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->update_ops = nr_writes;
-	count->add = nr_add;
-	count->add_exist = nr_addexist;
-	count->remove = nr_del;
+			"nr_delnoent %lu\n", pthread_self(), URCU_TLS(nr_add),
+			URCU_TLS(nr_addexist), URCU_TLS(nr_del),
+			URCU_TLS(nr_delnoent));
+	count->update_ops = URCU_TLS(nr_writes);
+	count->add = URCU_TLS(nr_add);
+	count->add_exist = URCU_TLS(nr_addexist);
+	count->remove = URCU_TLS(nr_del);
 	return ((void*)2);
 }
 
@@ -247,10 +249,10 @@ int test_hash_unique_populate_hash(void)
 "larger random pool (-p option). This may take a while...\n", init_populate, init_pool_size);
 	}
 
-	while (nr_add < init_populate) {
+	while (URCU_TLS(nr_add) < init_populate) {
 		node = malloc(sizeof(struct lfht_test_node));
 		lfht_test_node_init(node,
-			(void *)(((unsigned long) rand_r(&rand_lookup) % init_pool_size) + init_pool_offset),
+			(void *)(((unsigned long) rand_r(&URCU_TLS(rand_lookup)) % init_pool_size) + init_pool_offset),
 			sizeof(void *));
 		rcu_read_lock();
 		ret_node = cds_lfht_add_replace(test_ht,
@@ -259,11 +261,11 @@ int test_hash_unique_populate_hash(void)
 		rcu_read_unlock();
 		if (ret_node) {
 			call_rcu(&to_test_node(ret_node)->head, free_node_cb);
-			nr_addexist++;
+			URCU_TLS(nr_addexist)++;
 		} else {
-			nr_add++;
+			URCU_TLS(nr_add)++;
 		}
-		nr_writes++;
+		URCU_TLS(nr_writes)++;
 	}
 	return 0;
 }
diff --git a/tests/test_urcu_lfq.c b/tests/test_urcu_lfq.c
index 01a2781..66ddd41 100644
--- a/tests/test_urcu_lfq.c
+++ b/tests/test_urcu_lfq.c
@@ -38,6 +38,7 @@
 #include <errno.h>
 
 #include <urcu/arch.h>
+#include <urcu/tls-compat.h>
 
 #ifdef __linux__
 #include <syscall.h>
@@ -148,11 +149,11 @@ static int test_duration_enqueue(void)
 	return !test_stop;
 }
 
-static unsigned long long __thread nr_dequeues;
-static unsigned long long __thread nr_enqueues;
+static DEFINE_URCU_TLS(unsigned long long, nr_dequeues);
+static DEFINE_URCU_TLS(unsigned long long, nr_enqueues);
 
-static unsigned long long __thread nr_successful_dequeues;
-static unsigned long long __thread nr_successful_enqueues;
+static DEFINE_URCU_TLS(unsigned long long, nr_successful_dequeues);
+static DEFINE_URCU_TLS(unsigned long long, nr_successful_enqueues);
 
 static unsigned int nr_enqueuers;
 static unsigned int nr_dequeuers;
@@ -188,24 +189,24 @@ void *thr_enqueuer(void *_count)
 		rcu_read_lock();
 		cds_lfq_enqueue_rcu(&q, &node->list);
 		rcu_read_unlock();
-		nr_successful_enqueues++;
+		URCU_TLS(nr_successful_enqueues)++;
 
 		if (caa_unlikely(wdelay))
 			loop_sleep(wdelay);
 fail:
-		nr_enqueues++;
+		URCU_TLS(nr_enqueues)++;
 		if (caa_unlikely(!test_duration_enqueue()))
 			break;
 	}
 
 	rcu_unregister_thread();
 
-	count[0] = nr_enqueues;
-	count[1] = nr_successful_enqueues;
+	count[0] = URCU_TLS(nr_enqueues);
+	count[1] = URCU_TLS(nr_successful_enqueues);
 	printf_verbose("enqueuer thread_end, thread id : %lx, tid %lu, "
 		       "enqueues %llu successful_enqueues %llu\n",
-		       pthread_self(), (unsigned long)gettid(), nr_enqueues,
-		       nr_successful_enqueues);
+		       pthread_self(), (unsigned long)gettid(),
+		       URCU_TLS(nr_enqueues), URCU_TLS(nr_successful_enqueues));
 	return ((void*)1);
 
 }
@@ -251,10 +252,10 @@ void *thr_dequeuer(void *_count)
 
 		if (node) {
 			call_rcu(&node->rcu, free_node_cb);
-			nr_successful_dequeues++;
+			URCU_TLS(nr_successful_dequeues)++;
 		}
 
-		nr_dequeues++;
+		URCU_TLS(nr_dequeues)++;
 		if (caa_unlikely(!test_duration_dequeue()))
 			break;
 		if (caa_unlikely(rduration))
@@ -265,10 +266,10 @@ void *thr_dequeuer(void *_count)
 	rcu_defer_unregister_thread();
 	printf_verbose("dequeuer thread_end, thread id : %lx, tid %lu, "
 		       "dequeues %llu, successful_dequeues %llu\n",
-		       pthread_self(), (unsigned long)gettid(), nr_dequeues,
-		       nr_successful_dequeues);
-	count[0] = nr_dequeues;
-	count[1] = nr_successful_dequeues;
+		       pthread_self(), (unsigned long)gettid(),
+		       URCU_TLS(nr_dequeues), URCU_TLS(nr_successful_dequeues));
+	count[0] = URCU_TLS(nr_dequeues);
+	count[1] = URCU_TLS(nr_successful_dequeues);
 	return ((void*)2);
 }
 
diff --git a/tests/test_urcu_lfs.c b/tests/test_urcu_lfs.c
index b48a35f..88bf65d 100644
--- a/tests/test_urcu_lfs.c
+++ b/tests/test_urcu_lfs.c
@@ -38,6 +38,7 @@
 #include <errno.h>
 
 #include <urcu/arch.h>
+#include <urcu/tls-compat.h>
 
 #ifdef __linux__
 #include <syscall.h>
@@ -148,11 +149,11 @@ static int test_duration_enqueue(void)
 	return !test_stop;
 }
 
-static unsigned long long __thread nr_dequeues;
-static unsigned long long __thread nr_enqueues;
+static DEFINE_URCU_TLS(unsigned long long, nr_dequeues);
+static DEFINE_URCU_TLS(unsigned long long, nr_enqueues);
 
-static unsigned long long __thread nr_successful_dequeues;
-static unsigned long long __thread nr_successful_enqueues;
+static DEFINE_URCU_TLS(unsigned long long, nr_successful_dequeues);
+static DEFINE_URCU_TLS(unsigned long long, nr_successful_enqueues);
 
 static unsigned int nr_enqueuers;
 static unsigned int nr_dequeuers;
@@ -187,24 +188,24 @@ void *thr_enqueuer(void *_count)
 		cds_lfs_node_init_rcu(&node->list);
 		/* No rcu read-side is needed for push */
 		cds_lfs_push_rcu(&s, &node->list);
-		nr_successful_enqueues++;
+		URCU_TLS(nr_successful_enqueues)++;
 
 		if (caa_unlikely(wdelay))
 			loop_sleep(wdelay);
 fail:
-		nr_enqueues++;
+		URCU_TLS(nr_enqueues)++;
 		if (caa_unlikely(!test_duration_enqueue()))
 			break;
 	}
 
 	rcu_unregister_thread();
 
-	count[0] = nr_enqueues;
-	count[1] = nr_successful_enqueues;
+	count[0] = URCU_TLS(nr_enqueues);
+	count[1] = URCU_TLS(nr_successful_enqueues);
 	printf_verbose("enqueuer thread_end, thread id : %lx, tid %lu, "
 		       "enqueues %llu successful_enqueues %llu\n",
-		       pthread_self(), (unsigned long)gettid(), nr_enqueues,
-		       nr_successful_enqueues);
+		       pthread_self(), (unsigned long)gettid(),
+		       URCU_TLS(nr_enqueues), URCU_TLS(nr_successful_enqueues));
 	return ((void*)1);
 
 }
@@ -249,9 +250,9 @@ void *thr_dequeuer(void *_count)
 		rcu_read_unlock();
 		if (node) {
 			call_rcu(&node->rcu, free_node_cb);
-			nr_successful_dequeues++;
+			URCU_TLS(nr_successful_dequeues)++;
 		}
-		nr_dequeues++;
+		URCU_TLS(nr_dequeues)++;
 		if (caa_unlikely(!test_duration_dequeue()))
 			break;
 		if (caa_unlikely(rduration))
@@ -263,10 +264,10 @@ void *thr_dequeuer(void *_count)
 
 	printf_verbose("dequeuer thread_end, thread id : %lx, tid %lu, "
 		       "dequeues %llu, successful_dequeues %llu\n",
-		       pthread_self(), (unsigned long)gettid(), nr_dequeues,
-		       nr_successful_dequeues);
-	count[0] = nr_dequeues;
-	count[1] = nr_successful_dequeues;
+		       pthread_self(), (unsigned long)gettid(),
+		       URCU_TLS(nr_dequeues), URCU_TLS(nr_successful_dequeues));
+	count[0] = URCU_TLS(nr_dequeues);
+	count[1] = URCU_TLS(nr_successful_dequeues);
 	return ((void*)2);
 }
 
diff --git a/tests/test_urcu_qsbr.c b/tests/test_urcu_qsbr.c
index b986fd8..421fb07 100644
--- a/tests/test_urcu_qsbr.c
+++ b/tests/test_urcu_qsbr.c
@@ -35,6 +35,7 @@
 #include <errno.h>
 
 #include <urcu/arch.h>
+#include <urcu/tls-compat.h>
 
 #ifdef __linux__
 #include <syscall.h>
@@ -153,8 +154,8 @@ static int test_duration_read(void)
 	return !test_stop;
 }
 
-static unsigned long long __thread nr_writes;
-static unsigned long long __thread nr_reads;
+static DEFINE_URCU_TLS(unsigned long long, nr_writes);
+static DEFINE_URCU_TLS(unsigned long long, nr_reads);
 
 static unsigned int nr_readers;
 static unsigned int nr_writers;
@@ -241,9 +242,9 @@ void *thr_reader(void *_count)
 		if (caa_unlikely(rduration))
 			loop_sleep(rduration);
 		rcu_read_unlock();
-		nr_reads++;
+		URCU_TLS(nr_reads)++;
 		/* QS each 1024 reads */
-		if (caa_unlikely((nr_reads & ((1 << 10) - 1)) == 0))
+		if (caa_unlikely((URCU_TLS(nr_reads) & ((1 << 10) - 1)) == 0))
 			rcu_quiescent_state();
 		if (caa_unlikely(!test_duration_read()))
 			break;
@@ -255,7 +256,7 @@ void *thr_reader(void *_count)
 	rcu_register_thread();
 	rcu_unregister_thread();
 
-	*count = nr_reads;
+	*count = URCU_TLS(nr_reads);
 	printf_verbose("thread_end %s, thread id : %lx, tid %lu\n",
 			"reader", pthread_self(), (unsigned long)gettid());
 	return ((void*)1);
@@ -290,7 +291,7 @@ void *thr_writer(void *_count)
 			old->a = 0;
 		test_array_free(old);
 		rcu_copy_mutex_unlock();
-		nr_writes++;
+		URCU_TLS(nr_writes)++;
 		if (caa_unlikely(!test_duration_write()))
 			break;
 		if (caa_unlikely(wdelay))
@@ -299,7 +300,7 @@ void *thr_writer(void *_count)
 
 	printf_verbose("thread_end %s, thread id : %lx, tid %lu\n",
 			"writer", pthread_self(), (unsigned long)gettid());
-	*count = nr_writes;
+	*count = URCU_TLS(nr_writes);
 	return ((void*)2);
 }
 
diff --git a/tests/test_urcu_qsbr_gc.c b/tests/test_urcu_qsbr_gc.c
index 9deb0aa..561a475 100644
--- a/tests/test_urcu_qsbr_gc.c
+++ b/tests/test_urcu_qsbr_gc.c
@@ -35,6 +35,7 @@
 #include <errno.h>
 
 #include <urcu/arch.h>
+#include <urcu/tls-compat.h>
 
 #ifdef __linux__
 #include <syscall.h>
@@ -159,8 +160,8 @@ static int test_duration_read(void)
 	return !test_stop;
 }
 
-static unsigned long long __thread nr_writes;
-static unsigned long long __thread nr_reads;
+static DEFINE_URCU_TLS(unsigned long long, nr_writes);
+static DEFINE_URCU_TLS(unsigned long long, nr_reads);
 
 static unsigned int nr_readers;
 static unsigned int nr_writers;
@@ -217,9 +218,9 @@ void *thr_reader(void *_count)
 		if (caa_unlikely(rduration))
 			loop_sleep(rduration);
 		_rcu_read_unlock();
-		nr_reads++;
+		URCU_TLS(nr_reads)++;
 		/* QS each 1024 reads */
-		if (caa_unlikely((nr_reads & ((1 << 10) - 1)) == 0))
+		if (caa_unlikely((URCU_TLS(nr_reads) & ((1 << 10) - 1)) == 0))
 			_rcu_quiescent_state();
 		if (caa_unlikely(!test_duration_read()))
 			break;
@@ -227,7 +228,7 @@ void *thr_reader(void *_count)
 
 	rcu_unregister_thread();
 
-	*count = nr_reads;
+	*count = URCU_TLS(nr_reads);
 	printf_verbose("thread_end %s, thread id : %lx, tid %lu\n",
 			"reader", pthread_self(), (unsigned long)gettid());
 	return ((void*)1);
@@ -293,7 +294,7 @@ void *thr_writer(void *data)
 		if (caa_unlikely(wduration))
 			loop_sleep(wduration);
 		rcu_gc_reclaim(wtidx, old);
-		nr_writes++;
+		URCU_TLS(nr_writes)++;
 		if (caa_unlikely(!test_duration_write()))
 			break;
 		if (caa_unlikely(wdelay))
@@ -302,7 +303,7 @@ void *thr_writer(void *data)
 
 	printf_verbose("thread_end %s, thread id : %lx, tid %lu\n",
 			"writer", pthread_self(), (unsigned long)gettid());
-	tot_nr_writes[wtidx] = nr_writes;
+	tot_nr_writes[wtidx] = URCU_TLS(nr_writes);
 	return ((void*)2);
 }
 
diff --git a/tests/test_urcu_wfq.c b/tests/test_urcu_wfq.c
index 83ec635..75a9275 100644
--- a/tests/test_urcu_wfq.c
+++ b/tests/test_urcu_wfq.c
@@ -38,6 +38,7 @@
 #include <errno.h>
 
 #include <urcu/arch.h>
+#include <urcu/tls-compat.h>
 
 #ifdef __linux__
 #include <syscall.h>
@@ -147,11 +148,11 @@ static int test_duration_enqueue(void)
 	return !test_stop;
 }
 
-static unsigned long long __thread nr_dequeues;
-static unsigned long long __thread nr_enqueues;
+static DEFINE_URCU_TLS(unsigned long long, nr_dequeues);
+static DEFINE_URCU_TLS(unsigned long long, nr_enqueues);
 
-static unsigned long long __thread nr_successful_dequeues;
-static unsigned long long __thread nr_successful_enqueues;
+static DEFINE_URCU_TLS(unsigned long long, nr_successful_dequeues);
+static DEFINE_URCU_TLS(unsigned long long, nr_successful_enqueues);
 
 static unsigned int nr_enqueuers;
 static unsigned int nr_dequeuers;
@@ -178,22 +179,22 @@ void *thr_enqueuer(void *_count)
 			goto fail;
 		cds_wfq_node_init(node);
 		cds_wfq_enqueue(&q, node);
-		nr_successful_enqueues++;
+		URCU_TLS(nr_successful_enqueues)++;
 
 		if (caa_unlikely(wdelay))
 			loop_sleep(wdelay);
 fail:
-		nr_enqueues++;
+		URCU_TLS(nr_enqueues)++;
 		if (caa_unlikely(!test_duration_enqueue()))
 			break;
 	}
 
-	count[0] = nr_enqueues;
-	count[1] = nr_successful_enqueues;
+	count[0] = URCU_TLS(nr_enqueues);
+	count[1] = URCU_TLS(nr_successful_enqueues);
 	printf_verbose("enqueuer thread_end, thread id : %lx, tid %lu, "
 		       "enqueues %llu successful_enqueues %llu\n",
-		       pthread_self(), (unsigned long)gettid(), nr_enqueues,
-		       nr_successful_enqueues);
+		       pthread_self(), (unsigned long)gettid(),
+		       URCU_TLS(nr_enqueues), URCU_TLS(nr_successful_enqueues));
 	return ((void*)1);
 
 }
@@ -217,10 +218,10 @@ void *thr_dequeuer(void *_count)
 
 		if (node) {
 			free(node);
-			nr_successful_dequeues++;
+			URCU_TLS(nr_successful_dequeues)++;
 		}
 
-		nr_dequeues++;
+		URCU_TLS(nr_dequeues)++;
 		if (caa_unlikely(!test_duration_dequeue()))
 			break;
 		if (caa_unlikely(rduration))
@@ -229,10 +230,10 @@ void *thr_dequeuer(void *_count)
 
 	printf_verbose("dequeuer thread_end, thread id : %lx, tid %lu, "
 		       "dequeues %llu, successful_dequeues %llu\n",
-		       pthread_self(), (unsigned long)gettid(), nr_dequeues,
-		       nr_successful_dequeues);
-	count[0] = nr_dequeues;
-	count[1] = nr_successful_dequeues;
+		       pthread_self(), (unsigned long)gettid(),
+		       URCU_TLS(nr_dequeues), URCU_TLS(nr_successful_dequeues));
+	count[0] = URCU_TLS(nr_dequeues);
+	count[1] = URCU_TLS(nr_successful_dequeues);
 	return ((void*)2);
 }
 
diff --git a/tests/test_urcu_wfs.c b/tests/test_urcu_wfs.c
index 7746a1d..be09944 100644
--- a/tests/test_urcu_wfs.c
+++ b/tests/test_urcu_wfs.c
@@ -38,6 +38,7 @@
 #include <errno.h>
 
 #include <urcu/arch.h>
+#include <urcu/tls-compat.h>
 
 #ifdef __linux__
 #include <syscall.h>
@@ -147,11 +148,11 @@ static int test_duration_enqueue(void)
 	return !test_stop;
 }
 
-static unsigned long long __thread nr_dequeues;
-static unsigned long long __thread nr_enqueues;
+static DEFINE_URCU_TLS(unsigned long long, nr_dequeues);
+static DEFINE_URCU_TLS(unsigned long long, nr_enqueues);
 
-static unsigned long long __thread nr_successful_dequeues;
-static unsigned long long __thread nr_successful_enqueues;
+static DEFINE_URCU_TLS(unsigned long long, nr_successful_dequeues);
+static DEFINE_URCU_TLS(unsigned long long, nr_successful_enqueues);
 
 static unsigned int nr_enqueuers;
 static unsigned int nr_dequeuers;
@@ -178,22 +179,22 @@ void *thr_enqueuer(void *_count)
 			goto fail;
 		cds_wfs_node_init(node);
 		cds_wfs_push(&s, node);
-		nr_successful_enqueues++;
+		URCU_TLS(nr_successful_enqueues)++;
 
 		if (caa_unlikely(wdelay))
 			loop_sleep(wdelay);
 fail:
-		nr_enqueues++;
+		URCU_TLS(nr_enqueues)++;
 		if (caa_unlikely(!test_duration_enqueue()))
 			break;
 	}
 
-	count[0] = nr_enqueues;
-	count[1] = nr_successful_enqueues;
+	count[0] = URCU_TLS(nr_enqueues);
+	count[1] = URCU_TLS(nr_successful_enqueues);
 	printf_verbose("enqueuer thread_end, thread id : %lx, tid %lu, "
 		       "enqueues %llu successful_enqueues %llu\n",
-		       pthread_self(), (unsigned long)gettid(), nr_enqueues,
-		       nr_successful_enqueues);
+		       pthread_self(), (unsigned long)gettid(),
+		       URCU_TLS(nr_enqueues), URCU_TLS(nr_successful_enqueues));
 	return ((void*)1);
 
 }
@@ -217,10 +218,10 @@ void *thr_dequeuer(void *_count)
 
 		if (node) {
 			free(node);
-			nr_successful_dequeues++;
+			URCU_TLS(nr_successful_dequeues)++;
 		}
 
-		nr_dequeues++;
+		URCU_TLS(nr_dequeues)++;
 		if (caa_unlikely(!test_duration_dequeue()))
 			break;
 		if (caa_unlikely(rduration))
@@ -229,10 +230,10 @@ void *thr_dequeuer(void *_count)
 
 	printf_verbose("dequeuer thread_end, thread id : %lx, tid %lu, "
 		       "dequeues %llu, successful_dequeues %llu\n",
-		       pthread_self(), (unsigned long)gettid(), nr_dequeues,
-		       nr_successful_dequeues);
-	count[0] = nr_dequeues;
-	count[1] = nr_successful_dequeues;
+		       pthread_self(), (unsigned long)gettid(),
+		       URCU_TLS(nr_dequeues), URCU_TLS(nr_successful_dequeues));
+	count[0] = URCU_TLS(nr_dequeues);
+	count[1] = URCU_TLS(nr_successful_dequeues);
 	return ((void*)2);
 }
 
diff --git a/urcu-bp.c b/urcu-bp.c
index 08b4b55..bb657d7 100644
--- a/urcu-bp.c
+++ b/urcu-bp.c
@@ -40,6 +40,7 @@
 #include "urcu/map/urcu-bp.h"
 #include "urcu/static/urcu-bp.h"
 #include "urcu-pointer.h"
+#include "urcu/tls-compat.h"
 
 /* Do not #define _LGPL_SOURCE to ensure we can emit the wrapper symbols */
 #undef _LGPL_SOURCE
@@ -94,7 +95,7 @@ static pthread_mutex_t rcu_gp_lock = PTHREAD_MUTEX_INITIALIZER;
 
 #ifdef DEBUG_YIELD
 unsigned int yield_active;
-unsigned int __thread rand_yield;
+DEFINE_URCU_TLS(unsigned int, rand_yield);
 #endif
 
 /*
@@ -109,7 +110,7 @@ long rcu_gp_ctr = RCU_GP_COUNT;
  * Pointer to registry elements. Written to only by each individual reader. Read
  * by both the reader and the writers.
  */
-struct rcu_reader __thread *rcu_reader;
+DEFINE_URCU_TLS(struct rcu_reader *, rcu_reader);
 
 static CDS_LIST_HEAD(registry);
 
@@ -322,7 +323,7 @@ static void add_thread(void)
 	rcu_reader_reg->tid = pthread_self();
 	assert(rcu_reader_reg->ctr == 0);
 	cds_list_add(&rcu_reader_reg->node, &registry);
-	rcu_reader = rcu_reader_reg;
+	URCU_TLS(rcu_reader) = rcu_reader_reg;
 }
 
 /* Called with signals off and mutex locked */
@@ -363,7 +364,7 @@ void rcu_bp_register(void)
 	/*
 	 * Check if a signal concurrently registered our thread since
 	 * the check in rcu_read_lock(). */
-	if (rcu_reader)
+	if (URCU_TLS(rcu_reader))
 		goto end;
 
 	mutex_lock(&rcu_gp_lock);
diff --git a/urcu-call-rcu-impl.h b/urcu-call-rcu-impl.h
index 36e3cf4..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. */
 
@@ -232,7 +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 */
@@ -470,8 +471,8 @@ 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());
@@ -488,7 +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);
 }
 
 /*
@@ -504,7 +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;
 }
 
 /*
@@ -746,7 +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) {
diff --git a/urcu-defer-impl.h b/urcu-defer-impl.h
index 4d1ca5e..f65e410 100644
--- a/urcu-defer-impl.h
+++ b/urcu-defer-impl.h
@@ -48,6 +48,7 @@
 #include <urcu/uatomic.h>
 #include <urcu/list.h>
 #include <urcu/system.h>
+#include <urcu/tls-compat.h>
 
 /*
  * Number of entries in the per-thread defer queue. Must be power of 2.
@@ -130,7 +131,7 @@ static int32_t defer_thread_stop;
  * Written to only by each individual deferer. Read by both the deferer and
  * the reclamation tread.
  */
-static struct defer_queue __thread defer_queue;
+static DEFINE_URCU_TLS(struct defer_queue, defer_queue);
 static CDS_LIST_HEAD(registry_defer);
 static pthread_t tid_defer;
 
@@ -245,12 +246,12 @@ static void _rcu_defer_barrier_thread(void)
 {
 	unsigned long head, num_items;
 
-	head = defer_queue.head;
-	num_items = head - defer_queue.tail;
+	head = URCU_TLS(defer_queue).head;
+	num_items = head - URCU_TLS(defer_queue).tail;
 	if (caa_unlikely(!num_items))
 		return;
 	synchronize_rcu();
-	rcu_defer_barrier_queue(&defer_queue, head);
+	rcu_defer_barrier_queue(&URCU_TLS(defer_queue), head);
 }
 
 void rcu_defer_barrier_thread(void)
@@ -311,8 +312,8 @@ void _defer_rcu(void (*fct)(void *p), void *p)
 	 * Head is only modified by ourself. Tail can be modified by reclamation
 	 * thread.
 	 */
-	head = defer_queue.head;
-	tail = CMM_LOAD_SHARED(defer_queue.tail);
+	head = URCU_TLS(defer_queue).head;
+	tail = CMM_LOAD_SHARED(URCU_TLS(defer_queue).tail);
 
 	/*
 	 * If queue is full, or reached threshold. Empty queue ourself.
@@ -321,7 +322,7 @@ void _defer_rcu(void (*fct)(void *p), void *p)
 	if (caa_unlikely(head - tail >= DEFER_QUEUE_SIZE - 2)) {
 		assert(head - tail <= DEFER_QUEUE_SIZE);
 		rcu_defer_barrier_thread();
-		assert(head - CMM_LOAD_SHARED(defer_queue.tail) == 0);
+		assert(head - CMM_LOAD_SHARED(URCU_TLS(defer_queue).tail) == 0);
 	}
 
 	/*
@@ -340,25 +341,25 @@ void _defer_rcu(void (*fct)(void *p), void *p)
 	 * Decode: see the comments before 'struct defer_queue'
 	 *         or the code in rcu_defer_barrier_queue().
 	 */
-	if (caa_unlikely(defer_queue.last_fct_in != fct
+	if (caa_unlikely(URCU_TLS(defer_queue).last_fct_in != fct
 			|| DQ_IS_FCT_BIT(p)
 			|| p == DQ_FCT_MARK)) {
-		defer_queue.last_fct_in = fct;
+		URCU_TLS(defer_queue).last_fct_in = fct;
 		if (caa_unlikely(DQ_IS_FCT_BIT(fct) || fct == DQ_FCT_MARK)) {
-			_CMM_STORE_SHARED(defer_queue.q[head++ & DEFER_QUEUE_MASK],
+			_CMM_STORE_SHARED(URCU_TLS(defer_queue).q[head++ & DEFER_QUEUE_MASK],
 				      DQ_FCT_MARK);
-			_CMM_STORE_SHARED(defer_queue.q[head++ & DEFER_QUEUE_MASK],
+			_CMM_STORE_SHARED(URCU_TLS(defer_queue).q[head++ & DEFER_QUEUE_MASK],
 				      fct);
 		} else {
 			DQ_SET_FCT_BIT(fct);
-			_CMM_STORE_SHARED(defer_queue.q[head++ & DEFER_QUEUE_MASK],
+			_CMM_STORE_SHARED(URCU_TLS(defer_queue).q[head++ & DEFER_QUEUE_MASK],
 				      fct);
 		}
 	}
-	_CMM_STORE_SHARED(defer_queue.q[head++ & DEFER_QUEUE_MASK], p);
+	_CMM_STORE_SHARED(URCU_TLS(defer_queue).q[head++ & DEFER_QUEUE_MASK], p);
 	cmm_smp_wmb();	/* Publish new pointer before head */
 			/* Write q[] before head. */
-	CMM_STORE_SHARED(defer_queue.head, head);
+	CMM_STORE_SHARED(URCU_TLS(defer_queue).head, head);
 	cmm_smp_mb();	/* Write queue head before read futex */
 	/*
 	 * Wake-up any waiting defer thread.
@@ -422,16 +423,16 @@ int rcu_defer_register_thread(void)
 {
 	int was_empty;
 
-	assert(defer_queue.last_head == 0);
-	assert(defer_queue.q == NULL);
-	defer_queue.q = malloc(sizeof(void *) * DEFER_QUEUE_SIZE);
-	if (!defer_queue.q)
+	assert(URCU_TLS(defer_queue).last_head == 0);
+	assert(URCU_TLS(defer_queue).q == NULL);
+	URCU_TLS(defer_queue).q = malloc(sizeof(void *) * DEFER_QUEUE_SIZE);
+	if (!URCU_TLS(defer_queue).q)
 		return -ENOMEM;
 
 	mutex_lock_defer(&defer_thread_mutex);
 	mutex_lock_defer(&rcu_defer_mutex);
 	was_empty = cds_list_empty(&registry_defer);
-	cds_list_add(&defer_queue.list, &registry_defer);
+	cds_list_add(&URCU_TLS(defer_queue).list, &registry_defer);
 	mutex_unlock(&rcu_defer_mutex);
 
 	if (was_empty)
@@ -446,10 +447,10 @@ void rcu_defer_unregister_thread(void)
 
 	mutex_lock_defer(&defer_thread_mutex);
 	mutex_lock_defer(&rcu_defer_mutex);
-	cds_list_del(&defer_queue.list);
+	cds_list_del(&URCU_TLS(defer_queue).list);
 	_rcu_defer_barrier_thread();
-	free(defer_queue.q);
-	defer_queue.q = NULL;
+	free(URCU_TLS(defer_queue).q);
+	URCU_TLS(defer_queue).q = NULL;
 	is_empty = cds_list_empty(&registry_defer);
 	mutex_unlock(&rcu_defer_mutex);
 
diff --git a/urcu-qsbr.c b/urcu-qsbr.c
index 745676e..b20d564 100644
--- a/urcu-qsbr.c
+++ b/urcu-qsbr.c
@@ -40,6 +40,7 @@
 #define BUILD_QSBR_LIB
 #include "urcu/static/urcu-qsbr.h"
 #include "urcu-pointer.h"
+#include "urcu/tls-compat.h"
 
 /* Do not #define _LGPL_SOURCE to ensure we can emit the wrapper symbols */
 #undef _LGPL_SOURCE
@@ -66,11 +67,11 @@ unsigned long rcu_gp_ctr = RCU_GP_ONLINE;
  * Written to only by each individual reader. Read by both the reader and the
  * writers.
  */
-struct rcu_reader __thread rcu_reader;
+DEFINE_URCU_TLS(struct rcu_reader, rcu_reader);
 
 #ifdef DEBUG_YIELD
 unsigned int yield_active;
-unsigned int __thread rand_yield;
+DEFINE_URCU_TLS(unsigned int, rand_yield);
 #endif
 
 static CDS_LIST_HEAD(registry);
@@ -139,7 +140,7 @@ static void update_counter_and_wait(void)
 	 * quiescent state. Failure to do so could result in the writer
 	 * waiting forever while new readers are always accessing data
 	 * (no progress). Enforce compiler-order of store to rcu_gp_ctr
-	 * before load rcu_reader ctr.
+	 * before load URCU_TLS(rcu_reader).ctr.
 	 */
 	cmm_barrier();
 
@@ -206,7 +207,7 @@ void synchronize_rcu(void)
 {
 	unsigned long was_online;
 
-	was_online = rcu_reader.ctr;
+	was_online = URCU_TLS(rcu_reader).ctr;
 
 	/* All threads should read qparity before accessing data structure
 	 * where new ptr points to.  In the "then" case, rcu_thread_offline
@@ -236,7 +237,7 @@ void synchronize_rcu(void)
 	 * committing next rcu_gp_ctr update to memory. Failure to
 	 * do so could result in the writer waiting forever while new
 	 * readers are always accessing data (no progress).  Enforce
-	 * compiler-order of load rcu_reader ctr before store to
+	 * compiler-order of load URCU_TLS(rcu_reader).ctr before store to
 	 * rcu_gp_ctr.
 	 */
 	cmm_barrier();
@@ -269,7 +270,7 @@ void synchronize_rcu(void)
 {
 	unsigned long was_online;
 
-	was_online = rcu_reader.ctr;
+	was_online = URCU_TLS(rcu_reader).ctr;
 
 	/*
 	 * Mark the writer thread offline to make sure we don't wait for
@@ -326,11 +327,11 @@ void rcu_thread_online(void)
 
 void rcu_register_thread(void)
 {
-	rcu_reader.tid = pthread_self();
-	assert(rcu_reader.ctr == 0);
+	URCU_TLS(rcu_reader).tid = pthread_self();
+	assert(URCU_TLS(rcu_reader).ctr == 0);
 
 	mutex_lock(&rcu_gp_lock);
-	cds_list_add(&rcu_reader.node, &registry);
+	cds_list_add(&URCU_TLS(rcu_reader).node, &registry);
 	mutex_unlock(&rcu_gp_lock);
 	_rcu_thread_online();
 }
@@ -343,7 +344,7 @@ void rcu_unregister_thread(void)
 	 */
 	_rcu_thread_offline();
 	mutex_lock(&rcu_gp_lock);
-	cds_list_del(&rcu_reader.node);
+	cds_list_del(&URCU_TLS(rcu_reader).node);
 	mutex_unlock(&rcu_gp_lock);
 }
 
diff --git a/urcu.c b/urcu.c
index 3948629..5fb4db8 100644
--- a/urcu.c
+++ b/urcu.c
@@ -40,6 +40,7 @@
 #include "urcu/map/urcu.h"
 #include "urcu/static/urcu.h"
 #include "urcu-pointer.h"
+#include "urcu/tls-compat.h"
 
 /* Do not #define _LGPL_SOURCE to ensure we can emit the wrapper symbols */
 #undef _LGPL_SOURCE
@@ -94,11 +95,11 @@ unsigned long rcu_gp_ctr = RCU_GP_COUNT;
  * Written to only by each individual reader. Read by both the reader and the
  * writers.
  */
-struct rcu_reader __thread rcu_reader;
+DEFINE_URCU_TLS(struct rcu_reader, rcu_reader);
 
 #ifdef DEBUG_YIELD
 unsigned int yield_active;
-unsigned int __thread rand_yield;
+DEFINE_URCU_TLS(unsigned int, rand_yield);
 #endif
 
 static CDS_LIST_HEAD(registry);
@@ -120,9 +121,9 @@ static void mutex_lock(pthread_mutex_t *mutex)
 			perror("Error in pthread mutex lock");
 			exit(-1);
 		}
-		if (CMM_LOAD_SHARED(rcu_reader.need_mb)) {
+		if (CMM_LOAD_SHARED(URCU_TLS(rcu_reader).need_mb)) {
 			cmm_smp_mb();
-			_CMM_STORE_SHARED(rcu_reader.need_mb, 0);
+			_CMM_STORE_SHARED(URCU_TLS(rcu_reader).need_mb, 0);
 			cmm_smp_mb();
 		}
 		poll(NULL,0,10);
@@ -245,7 +246,7 @@ void update_counter_and_wait(void)
 	cmm_smp_mb();
 
 	/*
-	 * Wait for each thread rcu_reader.ctr count to become 0.
+	 * Wait for each thread URCU_TLS(rcu_reader).ctr count to become 0.
 	 */
 	for (;;) {
 		wait_loops++;
@@ -277,7 +278,8 @@ void update_counter_and_wait(void)
 #else /* #ifndef HAS_INCOHERENT_CACHES */
 		/*
 		 * BUSY-LOOP. Force the reader thread to commit its
-		 * rcu_reader.ctr update to memory if we wait for too long.
+		 * URCU_TLS(rcu_reader).ctr update to memory if we wait
+		 * for too long.
 		 */
 		if (cds_list_empty(&registry)) {
 			if (wait_loops == RCU_QS_ACTIVE_ATTEMPTS) {
@@ -328,7 +330,7 @@ void synchronize_rcu(void)
 	 * committing next rcu_gp_ctr update to memory. Failure to do so could
 	 * result in the writer waiting forever while new readers are always
 	 * accessing data (no progress).  Enforce compiler-order of load
-	 * rcu_reader ctr before store to rcu_gp_ctr.
+	 * URCU_TLS(rcu_reader).ctr before store to rcu_gp_ctr.
 	 */
 	cmm_barrier();
 
@@ -368,20 +370,20 @@ void rcu_read_unlock(void)
 
 void rcu_register_thread(void)
 {
-	rcu_reader.tid = pthread_self();
-	assert(rcu_reader.need_mb == 0);
-	assert(!(rcu_reader.ctr & RCU_GP_CTR_NEST_MASK));
+	URCU_TLS(rcu_reader).tid = pthread_self();
+	assert(URCU_TLS(rcu_reader).need_mb == 0);
+	assert(!(URCU_TLS(rcu_reader).ctr & RCU_GP_CTR_NEST_MASK));
 
 	mutex_lock(&rcu_gp_lock);
 	rcu_init();	/* In case gcc does not support constructor attribute */
-	cds_list_add(&rcu_reader.node, &registry);
+	cds_list_add(&URCU_TLS(rcu_reader).node, &registry);
 	mutex_unlock(&rcu_gp_lock);
 }
 
 void rcu_unregister_thread(void)
 {
 	mutex_lock(&rcu_gp_lock);
-	cds_list_del(&rcu_reader.node);
+	cds_list_del(&URCU_TLS(rcu_reader).node);
 	mutex_unlock(&rcu_gp_lock);
 }
 
@@ -405,7 +407,7 @@ static void sigrcu_handler(int signo, siginfo_t *siginfo, void *context)
 	 * executed on.
 	 */
 	cmm_smp_mb();
-	_CMM_STORE_SHARED(rcu_reader.need_mb, 0);
+	_CMM_STORE_SHARED(URCU_TLS(rcu_reader).need_mb, 0);
 	cmm_smp_mb();
 }
 
diff --git a/urcu/static/urcu-bp.h b/urcu/static/urcu-bp.h
index 8d22163..e7b2eda 100644
--- a/urcu/static/urcu-bp.h
+++ b/urcu/static/urcu-bp.h
@@ -38,6 +38,7 @@
 #include <urcu/system.h>
 #include <urcu/uatomic.h>
 #include <urcu/list.h>
+#include <urcu/tls-compat.h>
 
 /*
  * This code section can only be included in LGPL 2.1 compatible source code.
@@ -74,25 +75,25 @@ extern "C" {
 #define MAX_SLEEP 50
 
 extern unsigned int yield_active;
-extern unsigned int __thread rand_yield;
+extern DECLARE_URCU_TLS(unsigned int, rand_yield);
 
 static inline void debug_yield_read(void)
 {
 	if (yield_active & YIELD_READ)
-		if (rand_r(&rand_yield) & 0x1)
-			usleep(rand_r(&rand_yield) % MAX_SLEEP);
+		if (rand_r(&URCU_TLS(rand_yield)) & 0x1)
+			usleep(rand_r(&URCU_TLS(rand_yield)) % MAX_SLEEP);
 }
 
 static inline void debug_yield_write(void)
 {
 	if (yield_active & YIELD_WRITE)
-		if (rand_r(&rand_yield) & 0x1)
-			usleep(rand_r(&rand_yield) % MAX_SLEEP);
+		if (rand_r(&URCU_TLS(rand_yield)) & 0x1)
+			usleep(rand_r(&URCU_TLS(rand_yield)) % MAX_SLEEP);
 }
 
 static inline void debug_yield_init(void)
 {
-	rand_yield = time(NULL) ^ pthread_self();
+	URCU_TLS(rand_yield) = time(NULL) ^ pthread_self();
 }
 #else
 static inline void debug_yield_read(void)
@@ -144,7 +145,7 @@ struct rcu_reader {
  * Adds a pointer dereference on the read-side, but won't require to unregister
  * the reader thread.
  */
-extern struct rcu_reader __thread *rcu_reader;
+extern DECLARE_URCU_TLS(struct rcu_reader *, rcu_reader);
 
 static inline int rcu_old_gp_ongoing(long *value)
 {
@@ -166,24 +167,24 @@ static inline void _rcu_read_lock(void)
 	long tmp;
 
 	/* Check if registered */
-	if (caa_unlikely(!rcu_reader))
+	if (caa_unlikely(!URCU_TLS(rcu_reader)))
 		rcu_bp_register();
 
 	cmm_barrier();	/* Ensure the compiler does not reorder us with mutex */
-	tmp = rcu_reader->ctr;
+	tmp = URCU_TLS(rcu_reader)->ctr;
 	/*
 	 * rcu_gp_ctr is
 	 *   RCU_GP_COUNT | (~RCU_GP_CTR_PHASE or RCU_GP_CTR_PHASE)
 	 */
 	if (caa_likely(!(tmp & RCU_GP_CTR_NEST_MASK))) {
-		_CMM_STORE_SHARED(rcu_reader->ctr, _CMM_LOAD_SHARED(rcu_gp_ctr));
+		_CMM_STORE_SHARED(URCU_TLS(rcu_reader)->ctr, _CMM_LOAD_SHARED(rcu_gp_ctr));
 		/*
 		 * Set active readers count for outermost nesting level before
 		 * accessing the pointer.
 		 */
 		cmm_smp_mb();
 	} else {
-		_CMM_STORE_SHARED(rcu_reader->ctr, tmp + RCU_GP_COUNT);
+		_CMM_STORE_SHARED(URCU_TLS(rcu_reader)->ctr, tmp + RCU_GP_COUNT);
 	}
 }
 
@@ -193,7 +194,7 @@ static inline void _rcu_read_unlock(void)
 	 * Finish using rcu before decrementing the pointer.
 	 */
 	cmm_smp_mb();
-	_CMM_STORE_SHARED(rcu_reader->ctr, rcu_reader->ctr - RCU_GP_COUNT);
+	_CMM_STORE_SHARED(URCU_TLS(rcu_reader)->ctr, URCU_TLS(rcu_reader)->ctr - RCU_GP_COUNT);
 	cmm_barrier();	/* Ensure the compiler does not reorder us with mutex */
 }
 
diff --git a/urcu/static/urcu-qsbr.h b/urcu/static/urcu-qsbr.h
index 68bfc31..22908a4 100644
--- a/urcu/static/urcu-qsbr.h
+++ b/urcu/static/urcu-qsbr.h
@@ -42,6 +42,7 @@
 #include <urcu/uatomic.h>
 #include <urcu/list.h>
 #include <urcu/futex.h>
+#include <urcu/tls-compat.h>
 
 #ifdef __cplusplus
 extern "C" {
@@ -74,25 +75,25 @@ extern "C" {
 #define MAX_SLEEP 50
 
 extern unsigned int yield_active;
-extern unsigned int __thread rand_yield;
+extern DECLARE_URCU_TLS(unsigned int, rand_yield);
 
 static inline void debug_yield_read(void)
 {
 	if (yield_active & YIELD_READ)
-		if (rand_r(&rand_yield) & 0x1)
-			usleep(rand_r(&rand_yield) % MAX_SLEEP);
+		if (rand_r(&URCU_TLS(rand_yield)) & 0x1)
+			usleep(rand_r(&URCU_TLS(rand_yield)) % MAX_SLEEP);
 }
 
 static inline void debug_yield_write(void)
 {
 	if (yield_active & YIELD_WRITE)
-		if (rand_r(&rand_yield) & 0x1)
-			usleep(rand_r(&rand_yield) % MAX_SLEEP);
+		if (rand_r(&URCU_TLS(rand_yield)) & 0x1)
+			usleep(rand_r(&URCU_TLS(rand_yield)) % MAX_SLEEP);
 }
 
 static inline void debug_yield_init(void)
 {
-	rand_yield = time(NULL) ^ pthread_self();
+	URCU_TLS(rand_yield) = time(NULL) ^ pthread_self();
 }
 #else
 static inline void debug_yield_read(void)
@@ -128,7 +129,7 @@ struct rcu_reader {
 	pthread_t tid;
 };
 
-extern struct rcu_reader __thread rcu_reader;
+extern DECLARE_URCU_TLS(struct rcu_reader, rcu_reader);
 
 extern int32_t gp_futex;
 
@@ -137,8 +138,8 @@ extern int32_t gp_futex;
  */
 static inline void wake_up_gp(void)
 {
-	if (caa_unlikely(_CMM_LOAD_SHARED(rcu_reader.waiting))) {
-		_CMM_STORE_SHARED(rcu_reader.waiting, 0);
+	if (caa_unlikely(_CMM_LOAD_SHARED(URCU_TLS(rcu_reader).waiting))) {
+		_CMM_STORE_SHARED(URCU_TLS(rcu_reader).waiting, 0);
 		cmm_smp_mb();
 		if (uatomic_read(&gp_futex) != -1)
 			return;
@@ -158,7 +159,7 @@ static inline int rcu_gp_ongoing(unsigned long *ctr)
 
 static inline void _rcu_read_lock(void)
 {
-	rcu_assert(rcu_reader.ctr);
+	rcu_assert(URCU_TLS(rcu_reader).ctr);
 }
 
 static inline void _rcu_read_unlock(void)
@@ -168,8 +169,8 @@ static inline void _rcu_read_unlock(void)
 static inline void _rcu_quiescent_state(void)
 {
 	cmm_smp_mb();
-	_CMM_STORE_SHARED(rcu_reader.ctr, _CMM_LOAD_SHARED(rcu_gp_ctr));
-	cmm_smp_mb();	/* write rcu_reader.ctr before read futex */
+	_CMM_STORE_SHARED(URCU_TLS(rcu_reader).ctr, _CMM_LOAD_SHARED(rcu_gp_ctr));
+	cmm_smp_mb();	/* write URCU_TLS(rcu_reader).ctr before read futex */
 	wake_up_gp();
 	cmm_smp_mb();
 }
@@ -177,8 +178,8 @@ static inline void _rcu_quiescent_state(void)
 static inline void _rcu_thread_offline(void)
 {
 	cmm_smp_mb();
-	CMM_STORE_SHARED(rcu_reader.ctr, 0);
-	cmm_smp_mb();	/* write rcu_reader.ctr before read futex */
+	CMM_STORE_SHARED(URCU_TLS(rcu_reader).ctr, 0);
+	cmm_smp_mb();	/* write URCU_TLS(rcu_reader).ctr before read futex */
 	wake_up_gp();
 	cmm_barrier();	/* Ensure the compiler does not reorder us with mutex */
 }
@@ -186,7 +187,7 @@ static inline void _rcu_thread_offline(void)
 static inline void _rcu_thread_online(void)
 {
 	cmm_barrier();	/* Ensure the compiler does not reorder us with mutex */
-	_CMM_STORE_SHARED(rcu_reader.ctr, CMM_LOAD_SHARED(rcu_gp_ctr));
+	_CMM_STORE_SHARED(URCU_TLS(rcu_reader).ctr, CMM_LOAD_SHARED(rcu_gp_ctr));
 	cmm_smp_mb();
 }
 
diff --git a/urcu/static/urcu.h b/urcu/static/urcu.h
index 7ae0185..f27f8b6 100644
--- a/urcu/static/urcu.h
+++ b/urcu/static/urcu.h
@@ -40,6 +40,7 @@
 #include <urcu/uatomic.h>
 #include <urcu/list.h>
 #include <urcu/futex.h>
+#include <urcu/tls-compat.h>
 
 #ifdef __cplusplus
 extern "C" {
@@ -121,25 +122,25 @@ extern "C" {
 #endif
 
 extern unsigned int yield_active;
-extern unsigned int __thread rand_yield;
+extern DECLARE_URCU_TLS(unsigned int, rand_yield);
 
 static inline void debug_yield_read(void)
 {
 	if (yield_active & YIELD_READ)
-		if (rand_r(&rand_yield) & 0x1)
-			usleep(rand_r(&rand_yield) % MAX_SLEEP);
+		if (rand_r(&URCU_TLS(rand_yield)) & 0x1)
+			usleep(rand_r(&URCU_TLS(rand_yield)) % MAX_SLEEP);
 }
 
 static inline void debug_yield_write(void)
 {
 	if (yield_active & YIELD_WRITE)
-		if (rand_r(&rand_yield) & 0x1)
-			usleep(rand_r(&rand_yield) % MAX_SLEEP);
+		if (rand_r(&URCU_TLS(rand_yield)) & 0x1)
+			usleep(rand_r(&URCU_TLS(rand_yield)) % MAX_SLEEP);
 }
 
 static inline void debug_yield_init(void)
 {
-	rand_yield = time(NULL) ^ (unsigned long) pthread_self();
+	URCU_TLS(rand_yield) = time(NULL) ^ (unsigned long) pthread_self();
 }
 #else
 static inline void debug_yield_read(void)
@@ -222,7 +223,7 @@ struct rcu_reader {
 	pthread_t tid;
 };
 
-extern struct rcu_reader __thread rcu_reader;
+extern DECLARE_URCU_TLS(struct rcu_reader, rcu_reader);
 
 extern int32_t gp_futex;
 
@@ -256,20 +257,20 @@ static inline void _rcu_read_lock(void)
 	unsigned long tmp;
 
 	cmm_barrier();	/* Ensure the compiler does not reorder us with mutex */
-	tmp = rcu_reader.ctr;
+	tmp = URCU_TLS(rcu_reader).ctr;
 	/*
 	 * rcu_gp_ctr is
 	 *   RCU_GP_COUNT | (~RCU_GP_CTR_PHASE or RCU_GP_CTR_PHASE)
 	 */
 	if (caa_likely(!(tmp & RCU_GP_CTR_NEST_MASK))) {
-		_CMM_STORE_SHARED(rcu_reader.ctr, _CMM_LOAD_SHARED(rcu_gp_ctr));
+		_CMM_STORE_SHARED(URCU_TLS(rcu_reader).ctr, _CMM_LOAD_SHARED(rcu_gp_ctr));
 		/*
 		 * Set active readers count for outermost nesting level before
 		 * accessing the pointer. See smp_mb_master().
 		 */
 		smp_mb_slave(RCU_MB_GROUP);
 	} else {
-		_CMM_STORE_SHARED(rcu_reader.ctr, tmp + RCU_GP_COUNT);
+		_CMM_STORE_SHARED(URCU_TLS(rcu_reader).ctr, tmp + RCU_GP_COUNT);
 	}
 }
 
@@ -277,19 +278,19 @@ static inline void _rcu_read_unlock(void)
 {
 	unsigned long tmp;
 
-	tmp = rcu_reader.ctr;
+	tmp = URCU_TLS(rcu_reader).ctr;
 	/*
 	 * Finish using rcu before decrementing the pointer.
 	 * See smp_mb_master().
 	 */
 	if (caa_likely((tmp & RCU_GP_CTR_NEST_MASK) == RCU_GP_COUNT)) {
 		smp_mb_slave(RCU_MB_GROUP);
-		_CMM_STORE_SHARED(rcu_reader.ctr, rcu_reader.ctr - RCU_GP_COUNT);
-		/* write rcu_reader.ctr before read futex */
+		_CMM_STORE_SHARED(URCU_TLS(rcu_reader).ctr, URCU_TLS(rcu_reader).ctr - RCU_GP_COUNT);
+		/* write URCU_TLS(rcu_reader).ctr before read futex */
 		smp_mb_slave(RCU_MB_GROUP);
 		wake_up_gp();
 	} else {
-		_CMM_STORE_SHARED(rcu_reader.ctr, rcu_reader.ctr - RCU_GP_COUNT);
+		_CMM_STORE_SHARED(URCU_TLS(rcu_reader).ctr, URCU_TLS(rcu_reader).ctr - RCU_GP_COUNT);
 	}
 	cmm_barrier();	/* Ensure the compiler does not reorder us with mutex */
 }

-- 
Mathieu Desnoyers
Operating System Efficiency R&D Consultant
EfficiOS Inc.
http://www.efficios.com



More information about the lttng-dev mailing list