X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=fio.c;h=2aa8b40156ce60abf93efd0a657286159f6a2a26;hp=7a420daf909afb84009792d63d2b146d85ce520d;hb=5f8f9726e7466419db8b8e9c2205430ad1d33aad;hpb=ccb0fa24f0ff37f9270754ea5a1b66de9fd7053e

diff --git a/fio.c b/fio.c
index 7a420daf..2aa8b401 100644
--- a/fio.c
+++ b/fio.c
@@ -61,7 +61,7 @@ struct io_log *agg_io_log[2];
 #define TERMINATE_ALL		(-1)
 #define JOB_START_TIMEOUT	(5 * 1000)
 
-static inline void td_set_runstate(struct thread_data *td, int runstate)
+void td_set_runstate(struct thread_data *td, int runstate)
 {
 	if (td->runstate == runstate)
 		return;
@@ -107,36 +107,78 @@ static void status_timer_arm(void)
 	setitimer(ITIMER_REAL, &itimer, NULL);
 }
 
+static void sig_alrm(int fio_unused sig)
+{
+	if (threads) {
+		update_io_ticks();
+		print_thread_status();
+		status_timer_arm();
+	}
+}
+
 /*
- * We need to rearm on BSD/solaris. Switch this to sigaction in the future...
+ * Happens on thread runs with ctrl-c, ignore our own SIGQUIT
  */
-static void set_sig_handlers(void (*sighandler)(int))
+static void sig_quit(int sig)
 {
-	signal(SIGINT, sighandler);
-	signal(SIGALRM, sighandler);
 }
 
-static void sig_handler(int sig)
+static void sig_int(int sig)
 {
-	set_sig_handlers(sig_handler);
-
-	if (!threads)
-		return;
-
-	switch (sig) {
-	case SIGALRM:
-		update_io_ticks();
-		print_thread_status();
-		status_timer_arm();
-		break;
-	default:
+	if (threads) {
 		printf("\nfio: terminating on signal %d\n", sig);
 		fflush(stdout);
 		terminate_threads(TERMINATE_ALL);
-		break;
 	}
 }
 
+static void sig_ill(int fio_unused sig)
+{
+	if (!threads)
+		return;
+
+	log_err("fio: illegal instruction. your cpu does not support "
+		"the sse4.2 instruction for crc32c\n");
+	terminate_threads(TERMINATE_ALL);
+	exit(4);
+}
+
+static void set_sig_handlers(void)
+{
+	struct sigaction act;
+
+	memset(&act, 0, sizeof(act));
+	act.sa_handler = sig_alrm;
+	act.sa_flags = SA_RESTART;
+	sigaction(SIGALRM, &act, NULL);
+
+	memset(&act, 0, sizeof(act));
+	act.sa_handler = sig_int;
+	act.sa_flags = SA_RESTART;
+	sigaction(SIGINT, &act, NULL);
+
+	memset(&act, 0, sizeof(act));
+	act.sa_handler = sig_ill;
+	act.sa_flags = SA_RESTART;
+	sigaction(SIGILL, &act, NULL);
+
+	memset(&act, 0, sizeof(act));
+	act.sa_handler = sig_quit;
+	act.sa_flags = SA_RESTART;
+	sigaction(SIGQUIT, &act, NULL);
+}
+
+static inline int should_check_rate(struct thread_data *td)
+{
+	/*
+	 * No minimum rate set, always ok
+	 */
+	if (!td->o.ratemin && !td->o.rate_iops_min)
+		return 0;
+
+	return 1;
+}
+
 /*
  * Check if we are above the minimum rate given.
  */
@@ -147,12 +189,6 @@ static int check_min_rate(struct thread_data *td, struct timeval *now)
 	unsigned long spent;
 	unsigned long rate;
 
-	/*
-	 * No minimum rate set, always ok
-	 */
-	if (!td->o.ratemin && !td->o.rate_iops_min)
-		return 0;
-
 	/*
 	 * allow a 2 second settle period in the beginning
 	 */
@@ -237,7 +273,7 @@ static inline int runtime_exceeded(struct thread_data *td, struct timeval *t)
  */
 static void cleanup_pending_aio(struct thread_data *td)
 {
-	struct list_head *entry, *n;
+	struct flist_head *entry, *n;
 	struct io_u *io_u;
 	int r;
 
@@ -252,8 +288,8 @@ static void cleanup_pending_aio(struct thread_data *td)
 	 * now cancel remaining active events
 	 */
 	if (td->io_ops->cancel) {
-		list_for_each_safe(entry, n, &td->io_u_busylist) {
-			io_u = list_entry(entry, struct io_u, list);
+		flist_for_each_safe(entry, n, &td->io_u_busylist) {
+			io_u = flist_entry(entry, struct io_u, list);
 
 			/*
 			 * if the io_u isn't in flight, then that generally
@@ -321,6 +357,12 @@ requeue:
 	return 0;
 }
 
+static inline void update_tv_cache(struct thread_data *td)
+{
+	if ((++td->tv_cache_nr & td->tv_cache_mask) == td->tv_cache_mask)
+		fio_gettime(&td->tv_cache, NULL);
+}
+
 /*
  * The main verify engine. Runs over the writes we previously submitted,
  * reads the blocks back in, and checks the crc/md5 of the data.
@@ -352,13 +394,15 @@ static void do_verify(struct thread_data *td)
 
 	io_u = NULL;
 	while (!td->terminate) {
-		int ret2;
+		int ret2, full;
 
 		io_u = __get_io_u(td);
 		if (!io_u)
 			break;
 
-		if (runtime_exceeded(td, &io_u->start_time)) {
+		update_tv_cache(td);
+
+		if (runtime_exceeded(td, &td->tv_cache)) {
 			put_io_u(td, io_u);
 			td->terminate = 1;
 			break;
@@ -389,7 +433,7 @@ static void do_verify(struct thread_data *td)
 				 * zero read, fail
 				 */
 				if (!bytes) {
-					td_verror(td, ENODATA, "full resid");
+					td_verror(td, EIO, "full resid");
 					put_io_u(td, io_u);
 					break;
 				}
@@ -432,19 +476,25 @@ sync_done:
 		 * if we can queue more, do so. but check if there are
 		 * completed io_u's first.
 		 */
-		min_events = 0;
-		if (queue_full(td) || ret == FIO_Q_BUSY) {
-			if (td->cur_depth >= td->o.iodepth_low)
-				min_events = td->cur_depth - td->o.iodepth_low;
-			if (!min_events)
+		full = queue_full(td) || ret == FIO_Q_BUSY;
+		if (full || !td->o.iodepth_batch_complete) {
+			min_events = td->o.iodepth_batch_complete;
+			if (full && !min_events)
 				min_events = 1;
-		}
 
-		/*
-		 * Reap required number of io units, if any, and do the
-		 * verification on them through the callback handler
-		 */
-		if (io_u_queued_complete(td, min_events) < 0)
+			do {
+				/*
+				 * Reap required number of io units, if any,
+				 * and do the verification on them through
+				 * the callback handler
+				 */
+				if (io_u_queued_complete(td, min_events) < 0) {
+					ret = -1;
+					break;
+				}
+			} while (full && (td->cur_depth > td->o.iodepth_low));
+		}
+		if (ret < 0)
 			break;
 	}
 
@@ -465,19 +515,21 @@ sync_done:
  */
 static void do_io(struct thread_data *td)
 {
-	struct timeval s;
 	unsigned long usec;
 	unsigned int i;
 	int ret = 0;
 
-	td_set_runstate(td, TD_RUNNING);
+	if (in_ramp_time(td))
+		td_set_runstate(td, TD_RAMP);
+	else
+		td_set_runstate(td, TD_RUNNING);
 
 	while ((td->this_io_bytes[0] + td->this_io_bytes[1]) < td->o.size) {
 		struct timeval comp_time;
 		long bytes_done = 0;
 		int min_evts = 0;
 		struct io_u *io_u;
-		int ret2;
+		int ret2, full;
 
 		if (td->terminate)
 			break;
@@ -486,9 +538,9 @@ static void do_io(struct thread_data *td)
 		if (!io_u)
 			break;
 
-		memcpy(&s, &io_u->start_time, sizeof(s));
+		update_tv_cache(td);
 
-		if (runtime_exceeded(td, &s)) {
+		if (runtime_exceeded(td, &td->tv_cache)) {
 			put_io_u(td, io_u);
 			td->terminate = 1;
 			break;
@@ -501,7 +553,9 @@ static void do_io(struct thread_data *td)
 		if (td->o.verify != VERIFY_NONE && io_u->ddir == DDIR_READ) {
 			io_u->end_io = verify_io_u;
 			td_set_runstate(td, TD_VERIFYING);
-		} else
+		} else if (in_ramp_time(td))
+			td_set_runstate(td, TD_RAMP);
+		else
 			td_set_runstate(td, TD_RUNNING);
 
 		ret = td_io_queue(td, io_u);
@@ -517,7 +571,7 @@ static void do_io(struct thread_data *td)
 				 * zero read, fail
 				 */
 				if (!bytes) {
-					td_verror(td, ENODATA, "full resid");
+					td_verror(td, EIO, "full resid");
 					put_io_u(td, io_u);
 					break;
 				}
@@ -534,7 +588,9 @@ static void do_io(struct thread_data *td)
 				requeue_io_u(td, &io_u);
 			} else {
 sync_done:
-				fio_gettime(&comp_time, NULL);
+				if (should_check_rate(td))
+					fio_gettime(&comp_time, NULL);
+
 				bytes_done = io_u_sync_complete(td, io_u);
 				if (bytes_done < 0)
 					ret = bytes_done;
@@ -567,18 +623,26 @@ sync_done:
 		/*
 		 * See if we need to complete some commands
 		 */
-		if (queue_full(td) || ret == FIO_Q_BUSY) {
-			min_evts = 0;
-			if (td->cur_depth >= td->o.iodepth_low)
-				min_evts = td->cur_depth - td->o.iodepth_low;
-			if (!min_evts)
+		full = queue_full(td) || ret == FIO_Q_BUSY;
+		if (full || !td->o.iodepth_batch_complete) {
+			min_evts = td->o.iodepth_batch_complete;
+			if (full && !min_evts)
 				min_evts = 1;
-			fio_gettime(&comp_time, NULL);
-			bytes_done = io_u_queued_complete(td, min_evts);
-			if (bytes_done < 0)
-				break;
+
+			if (should_check_rate(td))
+				fio_gettime(&comp_time, NULL);
+
+			do {
+				ret = io_u_queued_complete(td, min_evts);
+				if (ret <= 0)
+					break;
+
+				bytes_done += ret;
+			} while (full && (td->cur_depth > td->o.iodepth_low));
 		}
 
+		if (ret < 0)
+			break;
 		if (!bytes_done)
 			continue;
 
@@ -587,15 +651,17 @@ sync_done:
 		 * of completions except the very first one which may look
 		 * a little bursty
 		 */
-		usec = utime_since(&s, &comp_time);
+		if (!in_ramp_time(td) && should_check_rate(td)) {
+			usec = utime_since(&td->tv_cache, &comp_time);
 
-		rate_throttle(td, usec, bytes_done);
+			rate_throttle(td, usec, bytes_done);
 
-		if (check_min_rate(td, &comp_time)) {
-			if (exitall_on_terminate)
-				terminate_threads(td->groupid);
-			td_verror(td, ENODATA, "check_min_rate");
-			break;
+			if (check_min_rate(td, &comp_time)) {
+				if (exitall_on_terminate)
+					terminate_threads(td->groupid);
+				td_verror(td, EIO, "check_min_rate");
+				break;
+			}
 		}
 
 		if (td->o.thinktime) {
@@ -647,13 +713,13 @@ sync_done:
 
 static void cleanup_io_u(struct thread_data *td)
 {
-	struct list_head *entry, *n;
+	struct flist_head *entry, *n;
 	struct io_u *io_u;
 
-	list_for_each_safe(entry, n, &td->io_u_freelist) {
-		io_u = list_entry(entry, struct io_u, list);
+	flist_for_each_safe(entry, n, &td->io_u_freelist) {
+		io_u = flist_entry(entry, struct io_u, list);
 
-		list_del(&io_u->list);
+		flist_del(&io_u->list);
 		free(io_u);
 	}
 
@@ -697,7 +763,7 @@ static int init_io_u(struct thread_data *td)
 			return 1;
 		io_u = malloc(sizeof(*io_u));
 		memset(io_u, 0, sizeof(*io_u));
-		INIT_LIST_HEAD(&io_u->list);
+		INIT_FLIST_HEAD(&io_u->list);
 
 		if (!(td->io_ops->flags & FIO_NOIO)) {
 			io_u->buf = p + max_bs * i;
@@ -708,11 +774,9 @@ static int init_io_u(struct thread_data *td)
 
 		io_u->index = i;
 		io_u->flags = IO_U_F_FREE;
-		list_add(&io_u->list, &td->io_u_freelist);
+		flist_add(&io_u->list, &td->io_u_freelist);
 	}
 
-	io_u_init_timeout();
-
 	return 0;
 }
 
@@ -793,12 +857,8 @@ static int keep_running(struct thread_data *td)
 	return 0;
 }
 
-static int clear_io_state(struct thread_data *td)
+static void reset_io_counters(struct thread_data *td)
 {
-	struct fio_file *f;
-	unsigned int i;
-	int ret;
-
 	td->ts.stat_io_bytes[0] = td->ts.stat_io_bytes[1] = 0;
 	td->this_io_bytes[0] = td->this_io_bytes[1] = 0;
 	td->zone_bytes = 0;
@@ -813,6 +873,34 @@ static int clear_io_state(struct thread_data *td)
 	 */
 	if (td->o.time_based || td->o.loops)
 		td->nr_done_files = 0;
+}
+
+void reset_all_stats(struct thread_data *td)
+{
+	struct timeval tv;
+	int i;
+
+	reset_io_counters(td);
+
+	for (i = 0; i < 2; i++) {
+		td->io_bytes[i] = 0;
+		td->io_blocks[i] = 0;
+		td->io_issues[i] = 0;
+		td->ts.total_io_u[i] = 0;
+	}
+	
+	fio_gettime(&tv, NULL);
+	memcpy(&td->epoch, &tv, sizeof(tv));
+	memcpy(&td->start, &tv, sizeof(tv));
+}
+
+static int clear_io_state(struct thread_data *td)
+{
+	struct fio_file *f;
+	unsigned int i;
+	int ret;
+
+	reset_io_counters(td);
 
 	close_files(td);
 
@@ -844,11 +932,11 @@ static void *thread_main(void *data)
 
 	dprint(FD_PROCESS, "jobs pid=%d started\n", (int) td->pid);
 
-	INIT_LIST_HEAD(&td->io_u_freelist);
-	INIT_LIST_HEAD(&td->io_u_busylist);
-	INIT_LIST_HEAD(&td->io_u_requeues);
-	INIT_LIST_HEAD(&td->io_log_list);
-	INIT_LIST_HEAD(&td->io_hist_list);
+	INIT_FLIST_HEAD(&td->io_u_freelist);
+	INIT_FLIST_HEAD(&td->io_u_busylist);
+	INIT_FLIST_HEAD(&td->io_u_requeues);
+	INIT_FLIST_HEAD(&td->io_log_list);
+	INIT_FLIST_HEAD(&td->io_hist_list);
 	td->io_hist_tree = RB_ROOT;
 
 	td_set_runstate(td, TD_INITIALIZED);
@@ -909,7 +997,6 @@ static void *thread_main(void *data)
 	}
 
 	fio_gettime(&td->epoch, NULL);
-	memcpy(&td->timeout_end, &td->epoch, sizeof(td->epoch));
 	getrusage(RUSAGE_SELF, &td->ts.ru_start);
 
 	runtime[0] = runtime[1] = 0;
@@ -917,6 +1004,7 @@ static void *thread_main(void *data)
 	while (keep_running(td)) {
 		fio_gettime(&td->start, NULL);
 		memcpy(&td->ts.stat_sample_time, &td->start, sizeof(td->start));
+		memcpy(&td->tv_cache, &td->start, sizeof(td->start));
 
 		if (td->o.ratemin)
 			memcpy(&td->lastrate, &td->ts.stat_sample_time,
@@ -1156,7 +1244,7 @@ static void run_threads(void)
 		fflush(stdout);
 	}
 
-	set_sig_handlers(sig_handler);
+	set_sig_handlers();
 
 	todo = thread_number;
 	nr_running = 0;
@@ -1313,7 +1401,10 @@ static void run_threads(void)
 			if (td->runstate != TD_INITIALIZED)
 				continue;
 
-			td_set_runstate(td, TD_RUNNING);
+			if (in_ramp_time(td))
+				td_set_runstate(td, TD_RAMP);
+			else
+				td_set_runstate(td, TD_RUNNING);
 			nr_running++;
 			nr_started--;
 			m_rate += td->o.ratemin;