X-Git-Url: https://git.kernel.dk/?a=blobdiff_plain;f=backend.c;h=2be71496f1ddf3076e9bf36b19539168824b7b92;hb=56d9fa4b8d4fa5166e3ec5dcdd37b5789b2cb01d;hp=e07ded86886ed5c836324d8fff704d48be0cb046;hpb=b2a9e6496494373f97f78dcdb3663fbb7b7a4ddb;p=fio.git

diff --git a/backend.c b/backend.c
index e07ded86..2be71496 100644
--- a/backend.c
+++ b/backend.c
@@ -418,6 +418,34 @@ static void check_update_rusage(struct thread_data *td)
 	}
 }
 
+static int wait_for_completions(struct thread_data *td, struct timeval *time,
+				uint64_t *bytes_done)
+{
+	const int full = queue_full(td);
+	int min_evts = 0;
+	int ret;
+
+	/*
+	 * if the queue is full, we MUST reap at least 1 event
+	 */
+	min_evts = min(td->o.iodepth_batch_complete, td->cur_depth);
+	if (full && !min_evts)
+		min_evts = 1;
+
+	if (time && (__should_check_rate(td, DDIR_READ) ||
+	    __should_check_rate(td, DDIR_WRITE) ||
+	    __should_check_rate(td, DDIR_TRIM)))
+		fio_gettime(time, NULL);
+
+	do {
+		ret = io_u_queued_complete(td, min_evts, bytes_done);
+		if (ret < 0)
+			break;
+	} while (full && (td->cur_depth > td->o.iodepth_low));
+
+	return ret;
+}
+
 /*
  * The main verify engine. Runs over the writes we previously submitted,
  * reads the blocks back in, and checks the crc/md5 of the data.
@@ -537,6 +565,8 @@ static void do_verify(struct thread_data *td, uint64_t verify_bytes)
 			io_u->end_io = verify_io_u;
 
 		ddir = io_u->ddir;
+		if (!td->o.disable_slat)
+			fio_gettime(&io_u->start_time, NULL);
 
 		ret = td_io_queue(td, io_u);
 		switch (ret) {
@@ -599,27 +629,9 @@ sync_done:
 		 */
 reap:
 		full = queue_full(td) || (ret == FIO_Q_BUSY && td->cur_depth);
-		if (full || !td->o.iodepth_batch_complete) {
-			min_events = min(td->o.iodepth_batch_complete,
-					 td->cur_depth);
-			/*
-			 * if the queue is full, we MUST reap at least 1 event
-			 */
-			if (full && !min_events)
-				min_events = 1;
+		if (full || !td->o.iodepth_batch_complete)
+			ret = wait_for_completions(td, NULL, bytes_done);
 
-			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, bytes_done) < 0) {
-					ret = -1;
-					break;
-				}
-			} while (full && (td->cur_depth > td->o.iodepth_low));
-		}
 		if (ret < 0)
 			break;
 	}
@@ -646,13 +658,35 @@ static unsigned int exceeds_number_ios(struct thread_data *td)
 	if (!td->o.number_ios)
 		return 0;
 
-	number_ios = ddir_rw_sum(td->this_io_blocks);
+	number_ios = ddir_rw_sum(td->io_blocks);
 	number_ios += td->io_u_queued + td->io_u_in_flight;
 
-	return number_ios >= td->o.number_ios;
+	return number_ios >= (td->o.number_ios * td->loops);
 }
 
-static int io_bytes_exceeded(struct thread_data *td)
+static int io_issue_bytes_exceeded(struct thread_data *td)
+{
+	unsigned long long bytes, limit;
+
+	if (td_rw(td))
+		bytes = td->io_issue_bytes[DDIR_READ] + td->io_issue_bytes[DDIR_WRITE];
+	else if (td_write(td))
+		bytes = td->io_issue_bytes[DDIR_WRITE];
+	else if (td_read(td))
+		bytes = td->io_issue_bytes[DDIR_READ];
+	else
+		bytes = td->io_issue_bytes[DDIR_TRIM];
+
+	if (td->o.io_limit)
+		limit = td->o.io_limit;
+	else
+		limit = td->o.size;
+
+	limit *= td->loops;
+	return bytes >= limit || exceeds_number_ios(td);
+}
+
+static int io_complete_bytes_exceeded(struct thread_data *td)
 {
 	unsigned long long bytes, limit;
 
@@ -670,6 +704,7 @@ static int io_bytes_exceeded(struct thread_data *td)
 	else
 		limit = td->o.size;
 
+	limit *= td->loops;
 	return bytes >= limit || exceeds_number_ios(td);
 }
 
@@ -693,21 +728,26 @@ static uint64_t do_io(struct thread_data *td)
 
 	lat_target_init(td);
 
+	total_bytes = td->o.size;
+	/*
+	* Allow random overwrite workloads to write up to io_limit
+	* before starting verification phase as 'size' doesn't apply.
+	*/
+	if (td_write(td) && td_random(td) && td->o.norandommap)
+		total_bytes = max(total_bytes, (uint64_t) td->o.io_limit);
 	/*
 	 * If verify_backlog is enabled, we'll run the verify in this
 	 * handler as well. For that case, we may need up to twice the
 	 * amount of bytes.
 	 */
-	total_bytes = td->o.size;
 	if (td->o.verify != VERIFY_NONE &&
 	   (td_write(td) && td->o.verify_backlog))
 		total_bytes += td->o.size;
 
 	while ((td->o.read_iolog_file && !flist_empty(&td->io_log_list)) ||
-		(!flist_empty(&td->trim_list)) || !io_bytes_exceeded(td) ||
+		(!flist_empty(&td->trim_list)) || !io_issue_bytes_exceeded(td) ||
 		td->o.time_based) {
 		struct timeval comp_time;
-		int min_evts = 0;
 		struct io_u *io_u;
 		int ret2, full;
 		enum fio_ddir ddir;
@@ -871,28 +911,8 @@ sync_done:
 		 */
 reap:
 		full = queue_full(td) || (ret == FIO_Q_BUSY && td->cur_depth);
-		if (full || !td->o.iodepth_batch_complete) {
-			min_evts = min(td->o.iodepth_batch_complete,
-					td->cur_depth);
-			/*
-			 * if the queue is full, we MUST reap at least 1 event
-			 */
-			if (full && !min_evts)
-				min_evts = 1;
-
-			if (__should_check_rate(td, DDIR_READ) ||
-			    __should_check_rate(td, DDIR_WRITE) ||
-			    __should_check_rate(td, DDIR_TRIM))
-				fio_gettime(&comp_time, NULL);
-
-			do {
-				ret = io_u_queued_complete(td, min_evts, bytes_done);
-				if (ret < 0)
-					break;
-
-			} while (full && (td->cur_depth > td->o.iodepth_low));
-		}
-
+		if (full || !td->o.iodepth_batch_complete)
+			ret = wait_for_completions(td, &comp_time, bytes_done);
 		if (ret < 0)
 			break;
 		if (!ddir_rw_sum(bytes_done) && !(td->io_ops->flags & FIO_NOIO))
@@ -1242,7 +1262,7 @@ static uint64_t do_dry_run(struct thread_data *td)
 	td_set_runstate(td, TD_RUNNING);
 
 	while ((td->o.read_iolog_file && !flist_empty(&td->io_log_list)) ||
-		(!flist_empty(&td->trim_list)) || !io_bytes_exceeded(td)) {
+		(!flist_empty(&td->trim_list)) || !io_complete_bytes_exceeded(td)) {
 		struct io_u *io_u;
 		int ret;
 
@@ -1430,7 +1450,8 @@ static void *thread_main(void *data)
 	if (o->verify_async && verify_async_init(td))
 		goto err;
 
-	if (o->ioprio) {
+	if (fio_option_is_set(o, ioprio) ||
+	    fio_option_is_set(o, ioprio_class)) {
 		ret = ioprio_set(IOPRIO_WHO_PROCESS, 0, o->ioprio_class, o->ioprio);
 		if (ret == -1) {
 			td_verror(td, errno, "ioprio_set");
@@ -1505,6 +1526,15 @@ static void *thread_main(void *data)
 
 		clear_state = 1;
 
+		/*
+		 * Make sure we've successfully updated the rusage stats
+		 * before waiting on the stat mutex. Otherwise we could have
+		 * the stat thread holding stat mutex and waiting for
+		 * the rusage_sem, which would never get upped because
+		 * this thread is waiting for the stat mutex.
+		 */
+		check_update_rusage(td);
+
 		fio_mutex_down(stat_mutex);
 		if (td_read(td) && td->io_bytes[DDIR_READ]) {
 			elapsed = mtime_since_now(&td->start);
@@ -1535,6 +1565,11 @@ static void *thread_main(void *data)
 
 		do_verify(td, verify_bytes);
 
+		/*
+		 * See comment further up for why this is done here.
+		 */
+		check_update_rusage(td);
+
 		fio_mutex_down(stat_mutex);
 		td->ts.runtime[DDIR_READ] += mtime_since_now(&td->start);
 		fio_gettime(&td->start, NULL);
@@ -2119,7 +2154,8 @@ static void *helper_thread_main(void *data)
 		gettimeofday(&tv, NULL);
 		ts.tv_sec = tv.tv_sec + sec;
 		ts.tv_nsec = (tv.tv_usec * 1000) + nsec;
-		if (ts.tv_nsec > 1000000000ULL) {
+
+		if (ts.tv_nsec >= 1000000000ULL) {
 			ts.tv_nsec -= 1000000000ULL;
 			ts.tv_sec++;
 		}