From: Michael Stapelberg <michael@stapelberg.de>
Date: Sat, 6 Feb 2010 15:10:33 +0000 (+0100)
Subject: Bugfix: correctly handle spurious wakeups in pthread_cond_timedwait
X-Git-Tag: collectd-4.8.4~25
X-Git-Url: https://git.tokkee.org/?a=commitdiff_plain;h=dbe1a7d;p=collectd.git

Bugfix: correctly handle spurious wakeups in pthread_cond_timedwait

This fixes problems storing data in rrd files on NetBSD 5.0 with more
than one CPU.

Signed-off-by: Florian Forster <octo@huhu.verplant.org>
---

diff --git a/src/plugin.c b/src/plugin.c
index 11a0ef6e..c163ca7d 100644
--- a/src/plugin.c
+++ b/src/plugin.c
@@ -302,6 +302,13 @@ static int plugin_load_file (char *file)
 	return (0);
 }
 
+static _Bool timeout_reached(struct timespec timeout)
+{
+	struct timeval now;
+	gettimeofday(&now, NULL);
+	return (now.tv_sec >= timeout.tv_sec && now.tv_usec >= (timeout.tv_nsec / 1000));
+}
+
 static void *plugin_read_thread (void __attribute__((unused)) *args)
 {
 	while (read_loop != 0)
@@ -310,6 +317,7 @@ static void *plugin_read_thread (void __attribute__((unused)) *args)
 		struct timeval now;
 		int status;
 		int rf_type;
+		int rc;
 
 		/* Get the read function that needs to be read next. */
 		rf = c_head_get_root (read_heap);
@@ -345,8 +353,16 @@ static void *plugin_read_thread (void __attribute__((unused)) *args)
 		/* sleep until this entry is due,
 		 * using pthread_cond_timedwait */
 		pthread_mutex_lock (&read_lock);
-		pthread_cond_timedwait (&read_cond, &read_lock,
+		/* In pthread_cond_timedwait, spurious wakeups are possible
+		 * (and really happen, at least on NetBSD with > 1 CPU), thus
+		 * we need to re-evaluate the condition every time
+		 * pthread_cond_timedwait returns. */
+		rc = 0;
+		while (!timeout_reached(rf->rf_next_read) && rc == 0) {
+			rc = pthread_cond_timedwait (&read_cond, &read_lock,
 				&rf->rf_next_read);
+		}
+
 		/* Must hold `real_lock' when accessing `rf->rf_type'. */
 		rf_type = rf->rf_type;
 		pthread_mutex_unlock (&read_lock);