qsbr: only mark reader thread as being waited for in contended case

[userspace-rcu.git] / urcu.c

diff --git a/urcu.c b/urcu.c
index 07661a3cb448b34505f24a8552a421741bb282bd..f74304ca51fa3c2c0aed789585192b06646eb5c9 100644 (file)
--- a/urcu.c
+++ b/urcu.c
@@ -49,8 +49,6 @@ void urcu_init(void)
 
 static pthread_mutex_t urcu_mutex = PTHREAD_MUTEX_INITIALIZER;
 
-int gp_futex;
-
 /*
  * Global grace period counter.
  * Contains the current RCU_GP_CTR_BIT.
@@ -63,14 +61,14 @@ long urcu_gp_ctr = RCU_GP_COUNT;
  * Written to only by each individual reader. Read by both the reader and the
  * writers.
  */
-long __thread urcu_active_readers;
+struct urcu_reader_status __thread urcu_reader_status;
 
 /* Thread IDs of registered readers */
 #define INIT_NUM_THREADS 4
 
 struct reader_registry {
        pthread_t tid;
-       long *urcu_active_readers;
+       struct urcu_reader_status *urcu_reader_status;
        char *need_mb;
 };
 
@@ -130,16 +128,19 @@ static void switch_next_urcu_qparity(void)
 }
 
 #ifdef URCU_MB
+#ifdef HAS_INCOHERENT_CACHES
 static void force_mb_single_thread(struct reader_registry *index)
 {
        smp_mb();
 }
+#endif /* #ifdef HAS_INCOHERENT_CACHES */
 
 static void force_mb_all_threads(void)
 {
        smp_mb();
 }
 #else /* #ifdef URCU_MB */
+#ifdef HAS_INCOHERENT_CACHES
 static void force_mb_single_thread(struct reader_registry *index)
 {
        assert(registry);
@@ -162,6 +163,7 @@ static void force_mb_single_thread(struct reader_registry *index)
        }
        smp_mb();       /* read ->need_mb before ending the barrier */
 }
+#endif /* #ifdef HAS_INCOHERENT_CACHES */
 
 static void force_mb_all_threads(void)
 {
@@ -206,27 +208,6 @@ static void force_mb_all_threads(void)
 }
 #endif /* #else #ifdef URCU_MB */
 
-/*
- * synchronize_rcu() waiting. Single thread.
- */
-static void wait_gp(struct reader_registry *index)
-{
-       atomic_dec(&gp_futex);
-       force_mb_single_thread(index); /* Write futex before read reader_gp */
-       if (!rcu_old_gp_ongoing(index->urcu_active_readers)) {
-               /* Read reader_gp before write futex */
-               force_mb_single_thread(index);
-               /* Callbacks are queued, don't wait. */
-               atomic_set(&gp_futex, 0);
-       } else {
-               /* Read reader_gp before read futex */
-               force_mb_single_thread(index);
-               if (atomic_read(&gp_futex) == -1)
-                       futex(&gp_futex, FUTEX_WAIT, -1,
-                             NULL, NULL, 0);
-       }
-}
-
 void wait_for_quiescent_state(void)
 {
        struct reader_registry *index;
@@ -234,14 +215,21 @@ void wait_for_quiescent_state(void)
        if (!registry)
                return;
        /*
-        * Wait for each thread urcu_active_readers count to become 0.
+        * Wait for each thread active_readers count to become 0.
         */
        for (index = registry; index < registry + num_readers; index++) {
                int wait_loops = 0;
+
+               if (likely(!rcu_old_gp_ongoing(
+                               &index->urcu_reader_status->active_readers)))
+                       continue;
+
+               index->urcu_reader_status->gp_waiting = 1;
 #ifndef HAS_INCOHERENT_CACHES
-               while (rcu_old_gp_ongoing(index->urcu_active_readers)) {
+               while (rcu_old_gp_ongoing(
+                               &index->urcu_reader_status->active_readers)) {
                        if (wait_loops++ == RCU_QS_ACTIVE_ATTEMPTS) {
-                               wait_gp(index);
+                               sched_yield();  /* ideally sched_yield_to() */
                        } else {
                                cpu_relax();
                        }
@@ -249,12 +237,13 @@ void wait_for_quiescent_state(void)
 #else /* #ifndef HAS_INCOHERENT_CACHES */
                /*
                 * BUSY-LOOP. Force the reader thread to commit its
-                * urcu_active_readers update to memory if we wait for too long.
+                * active_readers update to memory if we wait for too long.
                 */
-               while (rcu_old_gp_ongoing(index->urcu_active_readers)) {
+               while (rcu_old_gp_ongoing(
+                               &index->urcu_reader_status->active_readers)) {
                        switch (wait_loops++) {
                        case RCU_QS_ACTIVE_ATTEMPTS:
-                               wait_gp(index);
+                               sched_yield();  /* ideally sched_yield_to() */
                                break;
                        case KICK_READER_LOOPS:
                                force_mb_single_thread(index);
@@ -265,6 +254,7 @@ void wait_for_quiescent_state(void)
                        }
                }
 #endif /* #else #ifndef HAS_INCOHERENT_CACHES */
+               index->urcu_reader_status->gp_waiting = 0;
        }
 }
 
@@ -278,6 +268,8 @@ void synchronize_rcu(void)
        /* Write new ptr before changing the qparity */
        force_mb_all_threads();
 
+       STORE_SHARED(urcu_gp_ctr, urcu_gp_ctr ^ RCU_GP_ONGOING);
+
        switch_next_urcu_qparity();     /* 0 -> 1 */
 
        /*
@@ -337,6 +329,8 @@ void synchronize_rcu(void)
         */
        wait_for_quiescent_state();     /* Wait readers in parity 1 */
 
+       STORE_SHARED(urcu_gp_ctr, urcu_gp_ctr ^ RCU_GP_ONGOING);
+
        /* Finish waiting for reader threads before letting the old ptr being
         * freed. Must be done within internal_urcu_lock because it iterates on
         * reader threads. */
@@ -412,7 +406,7 @@ static void rcu_add_reader(pthread_t id)
        }
        registry[num_readers].tid = id;
        /* reference to the TLS of _this_ reader thread. */
-       registry[num_readers].urcu_active_readers = &urcu_active_readers;
+       registry[num_readers].urcu_reader_status = &urcu_reader_status;
        registry[num_readers].need_mb = &need_mb;
        num_readers++;
 }
@@ -431,7 +425,7 @@ static void rcu_remove_reader(pthread_t id)
                        memcpy(index, &registry[num_readers - 1],
                                sizeof(struct reader_registry));
                        registry[num_readers - 1].tid = 0;
-                       registry[num_readers - 1].urcu_active_readers = NULL;
+                       registry[num_readers - 1].urcu_reader_status = NULL;
                        num_readers--;
                        return;
                }