X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=backend.c;h=4437d196c0d5cdf2b5a9ce34c1982d353e7ea480;hp=68540abb9a2c1bd80c346d311f94646b464e2ff9;hb=b2ee7647ee05604397672f6b7c1a804b07466270;hpb=78d55e721268e2fb9bea707529db89bacf9964ef

diff --git a/backend.c b/backend.c
index 68540abb..4437d196 100644
--- a/backend.c
+++ b/backend.c
@@ -56,7 +56,9 @@
 #include "lib/tp.h"
 
 static pthread_t disk_util_thread;
-static struct fio_mutex *disk_thread_mutex;
+static pthread_cond_t du_cond;
+static pthread_mutex_t du_lock;
+
 static struct fio_mutex *startup_mutex;
 static struct flist_head *cgroup_list;
 static char *cgroup_mnt;
@@ -87,7 +89,7 @@ static void sig_int(int sig)
 			fio_server_got_signal(sig);
 		else {
 			log_info("\nfio: terminating on signal %d\n", sig);
-			fflush(stdout);
+			log_info_flush();
 			exit_value = 128;
 		}
 
@@ -391,7 +393,7 @@ static int break_on_this_error(struct thread_data *td, enum fio_ddir ddir,
 			 * fill_device option is set.
 			 */
 			td_clear_error(td);
-			td->terminate = 1;
+			fio_mark_td_terminate(td);
 			return 1;
 		} else {
 			/*
@@ -460,7 +462,7 @@ static void do_verify(struct thread_data *td, uint64_t verify_bytes)
 		if (runtime_exceeded(td, &td->tv_cache)) {
 			__update_tv_cache(td);
 			if (runtime_exceeded(td, &td->tv_cache)) {
-				td->terminate = 1;
+				fio_mark_td_terminate(td);
 				break;
 			}
 		}
@@ -714,7 +716,7 @@ static uint64_t do_io(struct thread_data *td)
 		if (runtime_exceeded(td, &td->tv_cache)) {
 			__update_tv_cache(td);
 			if (runtime_exceeded(td, &td->tv_cache)) {
-				td->terminate = 1;
+				fio_mark_td_terminate(td);
 				break;
 			}
 		}
@@ -922,7 +924,7 @@ reap:
 
 	if (td->o.fill_device && td->error == ENOSPC) {
 		td->error = 0;
-		td->terminate = 1;
+		fio_mark_td_terminate(td);
 	}
 	if (!td->error) {
 		struct fio_file *f;
@@ -1603,6 +1605,13 @@ static int fork_main(int shmid, int offset)
 	return (int) (uintptr_t) ret;
 }
 
+static void dump_td_info(struct thread_data *td)
+{
+	log_err("fio: job '%s' hasn't exited in %lu seconds, it appears to "
+		"be stuck. Doing forceful exit of this job.\n", td->o.name,
+			(unsigned long) time_since_now(&td->terminate_time));
+}
+
 /*
  * Run over the job map and reap the threads that have exited, if any.
  */
@@ -1680,6 +1689,17 @@ static void reap_threads(unsigned int *nr_running, unsigned int *t_rate,
 			}
 		}
 
+		/*
+		 * If the job is stuck, do a forceful timeout of it and
+		 * move on.
+		 */
+		if (td->terminate &&
+		    time_since_now(&td->terminate_time) >= FIO_REAP_TIMEOUT) {
+			dump_td_info(td);
+			td_set_runstate(td, TD_REAPED);
+			goto reaped;
+		}
+
 		/*
 		 * thread is not dead, continue
 		 */
@@ -1745,7 +1765,7 @@ static void run_threads(void)
 						nr_process > 1 ? "es" : "");
 		}
 		log_info("\n");
-		fflush(stdout);
+		log_info_flush();
 	}
 
 	todo = thread_number;
@@ -1952,16 +1972,20 @@ static void run_threads(void)
 	update_io_ticks();
 }
 
-void wait_for_disk_thread_exit(void)
+static void wait_for_disk_thread_exit(void)
 {
-	fio_mutex_down(disk_thread_mutex);
+	void *ret;
+
+	disk_util_start_exit();
+	pthread_cond_signal(&du_cond);
+	pthread_join(disk_util_thread, &ret);
 }
 
 static void free_disk_util(void)
 {
-	disk_util_start_exit();
-	wait_for_disk_thread_exit();
 	disk_util_prune_entries();
+
+	pthread_cond_destroy(&du_cond);
 }
 
 static void *disk_thread_main(void *data)
@@ -1970,17 +1994,30 @@ static void *disk_thread_main(void *data)
 
 	fio_mutex_up(startup_mutex);
 
-	while (threads && !ret) {
-		usleep(DISK_UTIL_MSEC * 1000);
-		if (!threads)
+	while (!ret) {
+		uint64_t sec = DISK_UTIL_MSEC / 1000;
+		uint64_t nsec = (DISK_UTIL_MSEC % 1000) * 1000000;
+		struct timespec ts;
+		struct timeval tv;
+
+		gettimeofday(&tv, NULL);
+		ts.tv_sec = tv.tv_sec + sec;
+		ts.tv_nsec = (tv.tv_usec * 1000) + nsec;
+		if (ts.tv_nsec > 1000000000ULL) {
+			ts.tv_nsec -= 1000000000ULL;
+			ts.tv_sec++;
+		}
+
+		ret = pthread_cond_timedwait(&du_cond, &du_lock, &ts);
+		if (ret != ETIMEDOUT)
 			break;
+
 		ret = update_io_ticks();
 
 		if (!is_backend)
 			print_thread_status();
 	}
 
-	fio_mutex_up(disk_thread_mutex);
 	return NULL;
 }
 
@@ -1990,22 +2027,15 @@ static int create_disk_util_thread(void)
 
 	setup_disk_util();
 
-	disk_thread_mutex = fio_mutex_init(FIO_MUTEX_LOCKED);
+	pthread_cond_init(&du_cond, NULL);
+	pthread_mutex_init(&du_lock, NULL);
 
 	ret = pthread_create(&disk_util_thread, NULL, disk_thread_main, NULL);
 	if (ret) {
-		fio_mutex_remove(disk_thread_mutex);
 		log_err("Can't create disk util thread: %s\n", strerror(ret));
 		return 1;
 	}
 
-	ret = pthread_detach(disk_util_thread);
-	if (ret) {
-		fio_mutex_remove(disk_thread_mutex);
-		log_err("Can't detatch disk util thread: %s\n", strerror(ret));
-		return 1;
-	}
-
 	dprint(FD_MUTEX, "wait on startup_mutex\n");
 	fio_mutex_down(startup_mutex);
 	dprint(FD_MUTEX, "done waiting on startup_mutex\n");
@@ -2043,14 +2073,18 @@ int fio_backend(void)
 	set_genesis_time();
 	stat_init();
 	create_disk_util_thread();
+	create_status_interval_thread();
 
 	cgroup_list = smalloc(sizeof(*cgroup_list));
 	INIT_FLIST_HEAD(cgroup_list);
 
 	run_threads();
 
+	wait_for_disk_thread_exit();
+	wait_for_status_interval_thread_exit();
+
 	if (!fio_abort) {
-		show_run_stats();
+		__show_run_stats();
 		if (write_bw_log) {
 			int i;
 
@@ -2072,7 +2106,6 @@ int fio_backend(void)
 	sfree(cgroup_mnt);
 
 	fio_mutex_remove(startup_mutex);
-	fio_mutex_remove(disk_thread_mutex);
 	stat_exit();
 	return exit_value;
 }