Revert "rcu: Unify force quiescent state"

This reverts commit 55d98e3edeeb17dd8445db27605d2b34f4c3ba85.

The commit introduced a regression in the replay functional test
on alpha (tests/functional/alpha/test_replay.py), that causes CI
failures regularly. Thus revert this change until someone has
figured out what is going wrong here.

Buglink: https://gitlab.com/qemu-project/qemu/-/issues/3197
Reviewed-by: Akihiko Odaki <odaki@rsg.ci.i.u-tokyo.ac.jp>
Signed-off-by: Thomas Huth <thuth@redhat.com>
Message-ID: <20260209120336.41454-1-thuth@redhat.com>

diff --git a/util/rcu.c b/util/rcu.c
index acac9446ea980a9caf980fef38ba058c4dbd0578..b703c86f15a36fc95d5e34f1f8513bf8ede9c0d5 100644 (file)
--- a/util/rcu.c
+++ b/util/rcu.c
@@ -43,14 +43,10 @@
 #define RCU_GP_LOCKED           (1UL << 0)
 #define RCU_GP_CTR              (1UL << 1)
 
-
-#define RCU_CALL_MIN_SIZE        30
-
 unsigned long rcu_gp_ctr = RCU_GP_LOCKED;
 
 QemuEvent rcu_gp_event;
 static int in_drain_call_rcu;
-static int rcu_call_count;
 static QemuMutex rcu_registry_lock;
 static QemuMutex rcu_sync_lock;
 
@@ -80,29 +76,15 @@ static void wait_for_readers(void)
 {
     ThreadList qsreaders = QLIST_HEAD_INITIALIZER(qsreaders);
     struct rcu_reader_data *index, *tmp;
-    int sleeps = 0;
-    bool forced = false;
 
     for (;;) {
-        /*
-         * Force the grace period to end and wait for it if any of the
-         * following heuristical conditions are satisfied:
-         * - A decent number of callbacks piled up.
-         * - It timed out.
-         * - It is in a drain_call_rcu() call.
-         *
-         * Otherwise, periodically poll the grace period, hoping it ends
-         * promptly.
+        /* We want to be notified of changes made to rcu_gp_ongoing
+         * while we walk the list.
          */
-        if (!forced &&
-            (qatomic_read(&rcu_call_count) >= RCU_CALL_MIN_SIZE ||
-             sleeps >= 5 || qatomic_read(&in_drain_call_rcu))) {
-            forced = true;
+        qemu_event_reset(&rcu_gp_event);
 
-            QLIST_FOREACH(index, &registry, node) {
-                notifier_list_notify(&index->force_rcu, NULL);
-                qatomic_set(&index->waiting, true);
-            }
+        QLIST_FOREACH(index, &registry, node) {
+            qatomic_set(&index->waiting, true);
         }
 
         /* Here, order the stores to index->waiting before the loads of
@@ -124,6 +106,8 @@ static void wait_for_readers(void)
                  * get some extra futex wakeups.
                  */
                 qatomic_set(&index->waiting, false);
+            } else if (qatomic_read(&in_drain_call_rcu)) {
+                notifier_list_notify(&index->force_rcu, NULL);
             }
         }
 
@@ -131,8 +115,7 @@ static void wait_for_readers(void)
             break;
         }
 
-        /*
-         * Sleep for a while and try again.
+        /* Wait for one thread to report a quiescent state and try again.
          * Release rcu_registry_lock, so rcu_(un)register_thread() doesn't
          * wait too much time.
          *
@@ -150,20 +133,7 @@ static void wait_for_readers(void)
          * rcu_registry_lock is released.
          */
         qemu_mutex_unlock(&rcu_registry_lock);
-
-        if (forced) {
-            qemu_event_wait(&rcu_gp_event);
-
-            /*
-             * We want to be notified of changes made to rcu_gp_ongoing
-             * while we walk the list.
-             */
-            qemu_event_reset(&rcu_gp_event);
-        } else {
-            g_usleep(10000);
-            sleeps++;
-        }
-
+        qemu_event_wait(&rcu_gp_event);
         qemu_mutex_lock(&rcu_registry_lock);
     }
 
@@ -203,11 +173,15 @@ void synchronize_rcu(void)
     }
 }
 
+
+#define RCU_CALL_MIN_SIZE        30
+
 /* Multi-producer, single-consumer queue based on urcu/static/wfqueue.h
  * from liburcu.  Note that head is only used by the consumer.
  */
 static struct rcu_head dummy;
 static struct rcu_head *head = &dummy, **tail = &dummy.next;
+static int rcu_call_count;
 static QemuEvent rcu_call_ready_event;
 
 static void enqueue(struct rcu_head *node)
@@ -285,27 +259,30 @@ static void *call_rcu_thread(void *opaque)
     rcu_register_thread();
 
     for (;;) {
-        int n;
+        int tries = 0;
+        int n = qatomic_read(&rcu_call_count);
 
-        /*
+        /* Heuristically wait for a decent number of callbacks to pile up.
          * Fetch rcu_call_count now, we only must process elements that were
          * added before synchronize_rcu() starts.
          */
-        for (;;) {
-            qemu_event_reset(&rcu_call_ready_event);
-            n = qatomic_read(&rcu_call_count);
-            if (n) {
-                break;
-            }
-
+        while (n == 0 || (n < RCU_CALL_MIN_SIZE && ++tries <= 5)) {
+            g_usleep(10000);
+            if (n == 0) {
+                qemu_event_reset(&rcu_call_ready_event);
+                n = qatomic_read(&rcu_call_count);
+                if (n == 0) {
 #if defined(CONFIG_MALLOC_TRIM)
-            malloc_trim(4 * 1024 * 1024);
+                    malloc_trim(4 * 1024 * 1024);
 #endif
-            qemu_event_wait(&rcu_call_ready_event);
+                    qemu_event_wait(&rcu_call_ready_event);
+                }
+            }
+            n = qatomic_read(&rcu_call_count);
         }
 
-        synchronize_rcu();
         qatomic_sub(&rcu_call_count, n);
+        synchronize_rcu();
         bql_lock();
         while (n > 0) {
             node = try_dequeue();