qsbr: micro optimization of the gp use.

[urcu.git] / urcu-qsbr.c

diff --git a/urcu-qsbr.c b/urcu-qsbr.c
index 14733ab21e1b61d10bf93ecc082967cdf9d717cd..1955277ac65d309e8b187f82b2eec242a7d211f2 100644 (file)
--- a/urcu-qsbr.c
+++ b/urcu-qsbr.c
@@ -32,6 +32,7 @@
 #include <errno.h>
 #include <poll.h>
 
+#define BUILD_QSBR_LIB
 #include "urcu-qsbr-static.h"
 /* Do not #define _LGPL_SOURCE to ensure we can emit the wrapper symbols */
 #include "urcu-qsbr.h"
@@ -41,7 +42,7 @@ pthread_mutex_t urcu_mutex = PTHREAD_MUTEX_INITIALIZER;
 /*
  * Global grace period counter.
  */
-long urcu_gp_ctr = 0;
+long urcu_gp_ctr = 1;
 
 /*
  * Written to only by each individual reader. Read by both the reader and the
@@ -105,11 +106,6 @@ static void force_mb_single_thread(struct reader_registry *index)
 }
 #endif /* #ifdef HAS_INCOHERENT_CACHES */
 
-static void force_mb_all_threads(void)
-{
-       smp_mb();
-}
-
 static void wait_for_quiescent_state(void)
 {
        struct reader_registry *index;
@@ -121,8 +117,7 @@ static void wait_for_quiescent_state(void)
         */
        for (index = registry; index < registry + num_readers; index++) {
 #ifndef HAS_INCOHERENT_CACHES
-               while (rcu_gp_ongoing(index->rcu_reader_qs_gp) &&
-                      (*index->rcu_reader_qs_gp - urcu_gp_ctr < 0))
+               while (rcu_gp_ongoing(index->rcu_reader_qs_gp))
                        cpu_relax();
 #else /* #ifndef HAS_INCOHERENT_CACHES */
                int wait_loops = 0;
@@ -130,8 +125,7 @@ static void wait_for_quiescent_state(void)
                 * BUSY-LOOP. Force the reader thread to commit its
                 * rcu_reader_qs_gp update to memory if we wait for too long.
                 */
-               while (rcu_gp_ongoing(index->rcu_reader_qs_gp) &&
-                      (*index->rcu_reader_qs_gp - urcu_gp_ctr < 0)) {
+               while (rcu_gp_ongoing(index->rcu_reader_qs_gp)) {
                        if (wait_loops++ == KICK_READER_LOOPS) {
                                force_mb_single_thread(index);
                                wait_loops = 0;
@@ -145,12 +139,27 @@ static void wait_for_quiescent_state(void)
 
 void synchronize_rcu(void)
 {
+       long was_online;
+
+       was_online = rcu_reader_qs_gp;
+
+       /*
+        * Mark the writer thread offline to make sure we don't wait for
+        * our own quiescent state. This allows using synchronize_rcu() in
+        * threads registered as readers.
+        */
+       smp_mb();
+       if (was_online)
+               STORE_SHARED(rcu_reader_qs_gp, 0);
+
        internal_urcu_lock();
-       force_mb_all_threads();
-       urcu_gp_ctr += 2;
+       STORE_SHARED(urcu_gp_ctr, urcu_gp_ctr + 2);
        wait_for_quiescent_state();
-       force_mb_all_threads();
        internal_urcu_unlock();
+
+       if (was_online)
+               _STORE_SHARED(rcu_reader_qs_gp, LOAD_SHARED(urcu_gp_ctr));
+       smp_mb();
 }
 
 /*