X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=io_u.c;h=8d42d6541f7dbb9592639619874295153101bd91;hp=f9870e70bc8d408ab7bf72adcf46cddf6818b4da;hb=fe8d7feef3bf53e14d794975e8943f63e848fb36;hpb=b0a84f48509d76032455842f1482901e3ab12afd

diff --git a/io_u.c b/io_u.c
index f9870e70..8d42d654 100644
--- a/io_u.c
+++ b/io_u.c
@@ -20,7 +20,7 @@ struct io_completion_data {
 
 	int error;			/* output */
 	uint64_t bytes_done[DDIR_RWDIR_CNT];	/* output */
-	struct timeval time;		/* output */
+	struct timespec time;		/* output */
 };
 
 /*
@@ -62,6 +62,7 @@ static uint64_t last_block(struct thread_data *td, struct fio_file *f,
 
 	/*
 	 * Hmm, should we make sure that ->io_size <= ->real_file_size?
+	 * -> not for now since there is code assuming it could go either.
 	 */
 	max_size = f->io_size;
 	if (max_size > f->real_file_size)
@@ -328,7 +329,8 @@ static int get_next_rand_block(struct thread_data *td, struct fio_file *f,
 	if (!get_next_rand_offset(td, f, ddir, b))
 		return 0;
 
-	if (td->o.time_based) {
+	if (td->o.time_based ||
+	    (td->o.file_service_type & __FIO_FSERVICE_NONUNIFORM)) {
 		fio_file_reset(td, f);
 		if (!get_next_rand_offset(td, f, ddir, b))
 			return 0;
@@ -361,8 +363,12 @@ static int get_next_seq_offset(struct thread_data *td, struct fio_file *f,
 	if (f->last_pos[ddir] < f->real_file_size) {
 		uint64_t pos;
 
-		if (f->last_pos[ddir] == f->file_offset && o->ddir_seq_add < 0)
-			f->last_pos[ddir] = f->real_file_size;
+		if (f->last_pos[ddir] == f->file_offset && o->ddir_seq_add < 0) {
+			if (f->real_file_size > f->io_size)
+				f->last_pos[ddir] = f->io_size;
+			else
+				f->last_pos[ddir] = f->real_file_size;
+		}
 
 		pos = f->last_pos[ddir] - f->file_offset;
 		if (pos && o->ddir_seq_add) {
@@ -377,8 +383,14 @@ static int get_next_seq_offset(struct thread_data *td, struct fio_file *f,
 			if (pos >= f->real_file_size) {
 				if (o->ddir_seq_add > 0)
 					pos = f->file_offset;
-				else
-					pos = f->real_file_size + o->ddir_seq_add;
+				else {
+					if (f->real_file_size > f->io_size)
+						pos = f->io_size;
+					else
+						pos = f->real_file_size;
+
+					pos += o->ddir_seq_add;
+				}
 			}
 		}
 
@@ -408,7 +420,7 @@ static int get_next_block(struct thread_data *td, struct io_u *io_u,
 				*is_random = 1;
 			} else {
 				*is_random = 0;
-				io_u_set(io_u, IO_U_F_BUSY_OK);
+				io_u_set(td, io_u, IO_U_F_BUSY_OK);
 				ret = get_next_seq_offset(td, f, ddir, &offset);
 				if (ret)
 					ret = get_next_rand_block(td, f, ddir, &b);
@@ -418,7 +430,7 @@ static int get_next_block(struct thread_data *td, struct io_u *io_u,
 			ret = get_next_seq_offset(td, f, ddir, &offset);
 		}
 	} else {
-		io_u_set(io_u, IO_U_F_BUSY_OK);
+		io_u_set(td, io_u, IO_U_F_BUSY_OK);
 		*is_random = 0;
 
 		if (td->o.rw_seq == RW_SEQ_SEQ) {
@@ -520,8 +532,8 @@ static unsigned int __get_next_buflen(struct thread_data *td, struct io_u *io_u,
 	int ddir = io_u->ddir;
 	unsigned int buflen = 0;
 	unsigned int minbs, maxbs;
-	uint64_t frand_max;
-	unsigned long r;
+	uint64_t frand_max, r;
+	bool power_2;
 
 	assert(ddir_rw(ddir));
 
@@ -550,7 +562,7 @@ static unsigned int __get_next_buflen(struct thread_data *td, struct io_u *io_u,
 			if (buflen < minbs)
 				buflen = minbs;
 		} else {
-			long perc = 0;
+			long long perc = 0;
 			unsigned int i;
 
 			for (i = 0; i < td->o.bssplit_nr[ddir]; i++) {
@@ -558,19 +570,19 @@ static unsigned int __get_next_buflen(struct thread_data *td, struct io_u *io_u,
 
 				buflen = bsp->bs;
 				perc += bsp->perc;
-				if ((r * 100UL <= frand_max * perc) &&
+				if (!perc)
+					break;
+				if ((r / perc <= frand_max / 100ULL) &&
 				    io_u_fits(td, io_u, buflen))
 					break;
 			}
 		}
 
-		if (td->o.verify != VERIFY_NONE)
-			buflen = (buflen + td->o.verify_interval - 1) &
-				~(td->o.verify_interval - 1);
-
-		if (!td->o.bs_unaligned && is_power_of_2(minbs))
+		power_2 = is_power_of_2(minbs);
+		if (!td->o.bs_unaligned && power_2)
 			buflen &= ~(minbs - 1);
-
+		else if (!td->o.bs_unaligned && !power_2) 
+			buflen -= buflen % minbs; 
 	} while (!io_u_fits(td, io_u, buflen));
 
 	return buflen;
@@ -634,20 +646,24 @@ int io_u_quiesce(struct thread_data *td)
 	}
 
 	while (td->io_u_in_flight) {
-		int fio_unused ret;
+		int ret;
 
 		ret = io_u_queued_complete(td, 1);
 		if (ret > 0)
 			completed += ret;
 	}
 
+	if (td->flags & TD_F_REGROW_LOGS)
+		regrow_logs(td);
+
 	return completed;
 }
 
 static enum fio_ddir rate_ddir(struct thread_data *td, enum fio_ddir ddir)
 {
 	enum fio_ddir odir = ddir ^ 1;
-	long usec, now;
+	long usec;
+	uint64_t now;
 
 	assert(ddir_rw(ddir));
 	now = utime_since_now(&td->start);
@@ -701,28 +717,22 @@ static enum fio_ddir get_rw_ddir(struct thread_data *td)
 	enum fio_ddir ddir;
 
 	/*
-	 * see if it's time to fsync
-	 */
-	if (td->o.fsync_blocks &&
-	   !(td->io_issues[DDIR_WRITE] % td->o.fsync_blocks) &&
-	     td->io_issues[DDIR_WRITE] && should_fsync(td))
-		return DDIR_SYNC;
-
-	/*
-	 * see if it's time to fdatasync
-	 */
-	if (td->o.fdatasync_blocks &&
-	   !(td->io_issues[DDIR_WRITE] % td->o.fdatasync_blocks) &&
-	     td->io_issues[DDIR_WRITE] && should_fsync(td))
-		return DDIR_DATASYNC;
-
-	/*
-	 * see if it's time to sync_file_range
+	 * See if it's time to fsync/fdatasync/sync_file_range first,
+	 * and if not then move on to check regular I/Os.
 	 */
-	if (td->sync_file_range_nr &&
-	   !(td->io_issues[DDIR_WRITE] % td->sync_file_range_nr) &&
-	     td->io_issues[DDIR_WRITE] && should_fsync(td))
-		return DDIR_SYNC_FILE_RANGE;
+	if (should_fsync(td)) {
+		if (td->o.fsync_blocks && td->io_issues[DDIR_WRITE] &&
+		    !(td->io_issues[DDIR_WRITE] % td->o.fsync_blocks))
+			return DDIR_SYNC;
+
+		if (td->o.fdatasync_blocks && td->io_issues[DDIR_WRITE] &&
+		    !(td->io_issues[DDIR_WRITE] % td->o.fdatasync_blocks))
+			return DDIR_DATASYNC;
+
+		if (td->sync_file_range_nr && td->io_issues[DDIR_WRITE] &&
+		    !(td->io_issues[DDIR_WRITE] % td->sync_file_range_nr))
+			return DDIR_SYNC_FILE_RANGE;
+	}
 
 	if (td_rw(td)) {
 		/*
@@ -746,8 +756,10 @@ static enum fio_ddir get_rw_ddir(struct thread_data *td)
 		ddir = DDIR_READ;
 	else if (td_write(td))
 		ddir = DDIR_WRITE;
-	else
+	else if (td_trim(td))
 		ddir = DDIR_TRIM;
+	else
+		ddir = DDIR_INVAL;
 
 	td->rwmix_ddir = rate_ddir(td, ddir);
 	return td->rwmix_ddir;
@@ -767,11 +779,11 @@ static void set_rw_ddir(struct thread_data *td, struct io_u *io_u)
 
 	io_u->ddir = io_u->acct_ddir = ddir;
 
-	if (io_u->ddir == DDIR_WRITE && (td->io_ops->flags & FIO_BARRIER) &&
+	if (io_u->ddir == DDIR_WRITE && td_ioengine_flagged(td, FIO_BARRIER) &&
 	    td->o.barrier_blocks &&
 	   !(td->io_issues[DDIR_WRITE] % td->o.barrier_blocks) &&
 	     td->io_issues[DDIR_WRITE])
-		io_u_set(io_u, IO_U_F_BARRIER);
+		io_u_set(td, io_u, IO_U_F_BARRIER);
 }
 
 void put_file_log(struct thread_data *td, struct fio_file *f)
@@ -793,7 +805,7 @@ void put_io_u(struct thread_data *td, struct io_u *io_u)
 		put_file_log(td, io_u->file);
 
 	io_u->file = NULL;
-	io_u_set(io_u, IO_U_F_FREE);
+	io_u_set(td, io_u, IO_U_F_FREE);
 
 	if (io_u->flags & IO_U_F_IN_CUR_DEPTH) {
 		td->cur_depth--;
@@ -806,7 +818,7 @@ void put_io_u(struct thread_data *td, struct io_u *io_u)
 
 void clear_io_u(struct thread_data *td, struct io_u *io_u)
 {
-	io_u_clear(io_u, IO_U_F_FLIGHT);
+	io_u_clear(td, io_u, IO_U_F_FLIGHT);
 	put_io_u(td, io_u);
 }
 
@@ -822,11 +834,11 @@ void requeue_io_u(struct thread_data *td, struct io_u **io_u)
 
 	td_io_u_lock(td);
 
-	io_u_set(__io_u, IO_U_F_FREE);
+	io_u_set(td, __io_u, IO_U_F_FREE);
 	if ((__io_u->flags & IO_U_F_FLIGHT) && ddir_rw(ddir))
 		td->io_issues[ddir]--;
 
-	io_u_clear(__io_u, IO_U_F_FLIGHT);
+	io_u_clear(td, __io_u, IO_U_F_FLIGHT);
 	if (__io_u->flags & IO_U_F_IN_CUR_DEPTH) {
 		td->cur_depth--;
 		assert(!(td->flags & TD_F_CHILD));
@@ -842,7 +854,7 @@ static int fill_io_u(struct thread_data *td, struct io_u *io_u)
 {
 	unsigned int is_random;
 
-	if (td->io_ops->flags & FIO_NOIO)
+	if (td_ioengine_flagged(td, FIO_NOIO))
 		goto out;
 
 	set_rw_ddir(td, io_u);
@@ -887,8 +899,9 @@ static int fill_io_u(struct thread_data *td, struct io_u *io_u)
 	}
 
 	if (io_u->offset + io_u->buflen > io_u->file->real_file_size) {
-		dprint(FD_IO, "io_u %p, offset too large\n", io_u);
-		dprint(FD_IO, "  off=%llu/%lu > %llu\n",
+		dprint(FD_IO, "io_u %p, offset + buflen exceeds file size\n",
+			io_u);
+		dprint(FD_IO, "  offset=%llu/buflen=%lu > %llu\n",
 			(unsigned long long) io_u->offset, io_u->buflen,
 			(unsigned long long) io_u->file->real_file_size);
 		return 1;
@@ -976,11 +989,52 @@ void io_u_mark_depth(struct thread_data *td, unsigned int nr)
 	td->ts.io_u_map[idx] += nr;
 }
 
-static void io_u_mark_lat_usec(struct thread_data *td, unsigned long usec)
+static void io_u_mark_lat_nsec(struct thread_data *td, unsigned long long nsec)
 {
 	int idx = 0;
 
-	assert(usec < 1000);
+	assert(nsec < 1000);
+
+	switch (nsec) {
+	case 750 ... 999:
+		idx = 9;
+		break;
+	case 500 ... 749:
+		idx = 8;
+		break;
+	case 250 ... 499:
+		idx = 7;
+		break;
+	case 100 ... 249:
+		idx = 6;
+		break;
+	case 50 ... 99:
+		idx = 5;
+		break;
+	case 20 ... 49:
+		idx = 4;
+		break;
+	case 10 ... 19:
+		idx = 3;
+		break;
+	case 4 ... 9:
+		idx = 2;
+		break;
+	case 2 ... 3:
+		idx = 1;
+	case 0 ... 1:
+		break;
+	}
+
+	assert(idx < FIO_IO_U_LAT_N_NR);
+	td->ts.io_u_lat_n[idx]++;
+}
+
+static void io_u_mark_lat_usec(struct thread_data *td, unsigned long long usec)
+{
+	int idx = 0;
+
+	assert(usec < 1000 && usec >= 1);
 
 	switch (usec) {
 	case 750 ... 999:
@@ -1017,10 +1071,12 @@ static void io_u_mark_lat_usec(struct thread_data *td, unsigned long usec)
 	td->ts.io_u_lat_u[idx]++;
 }
 
-static void io_u_mark_lat_msec(struct thread_data *td, unsigned long msec)
+static void io_u_mark_lat_msec(struct thread_data *td, unsigned long long msec)
 {
 	int idx = 0;
 
+	assert(msec >= 1);
+
 	switch (msec) {
 	default:
 		idx = 11;
@@ -1062,12 +1118,42 @@ static void io_u_mark_lat_msec(struct thread_data *td, unsigned long msec)
 	td->ts.io_u_lat_m[idx]++;
 }
 
-static void io_u_mark_latency(struct thread_data *td, unsigned long usec)
+static void io_u_mark_latency(struct thread_data *td, unsigned long long nsec)
 {
-	if (usec < 1000)
-		io_u_mark_lat_usec(td, usec);
+	if (nsec < 1000)
+		io_u_mark_lat_nsec(td, nsec);
+	else if (nsec < 1000000)
+		io_u_mark_lat_usec(td, nsec / 1000);
 	else
-		io_u_mark_lat_msec(td, usec / 1000);
+		io_u_mark_lat_msec(td, nsec / 1000000);
+}
+
+static unsigned int __get_next_fileno_rand(struct thread_data *td)
+{
+	unsigned long fileno;
+
+	if (td->o.file_service_type == FIO_FSERVICE_RANDOM) {
+		uint64_t frand_max = rand_max(&td->next_file_state);
+		unsigned long r;
+
+		r = __rand(&td->next_file_state);
+		return (unsigned int) ((double) td->o.nr_files
+				* (r / (frand_max + 1.0)));
+	}
+
+	if (td->o.file_service_type == FIO_FSERVICE_ZIPF)
+		fileno = zipf_next(&td->next_file_zipf);
+	else if (td->o.file_service_type == FIO_FSERVICE_PARETO)
+		fileno = pareto_next(&td->next_file_zipf);
+	else if (td->o.file_service_type == FIO_FSERVICE_GAUSS)
+		fileno = gauss_next(&td->next_file_gauss);
+	else {
+		log_err("fio: bad file service type: %d\n", td->o.file_service_type);
+		assert(0);
+		return 0;
+	}
+
+	return fileno >> FIO_FSERVICE_SHIFT;
 }
 
 /*
@@ -1077,17 +1163,13 @@ static struct fio_file *get_next_file_rand(struct thread_data *td,
 					   enum fio_file_flags goodf,
 					   enum fio_file_flags badf)
 {
-	uint64_t frand_max = rand_max(&td->next_file_state);
 	struct fio_file *f;
 	int fno;
 
 	do {
 		int opened = 0;
-		unsigned long r;
 
-		r = __rand(&td->next_file_state);
-		fno = (unsigned int) ((double) td->o.nr_files
-				* (r / (frand_max + 1.0)));
+		fno = __get_next_fileno_rand(td);
 
 		f = td->files[fno];
 		if (fio_file_done(f))
@@ -1240,10 +1322,14 @@ static long set_io_u_file(struct thread_data *td, struct io_u *io_u)
 		put_file_log(td, f);
 		td_io_close_file(td, f);
 		io_u->file = NULL;
-		fio_file_set_done(f);
-		td->nr_done_files++;
-		dprint(FD_FILE, "%s: is done (%d of %d)\n", f->file_name,
+		if (td->o.file_service_type & __FIO_FSERVICE_NONUNIFORM)
+			fio_file_reset(td, f);
+		else {
+			fio_file_set_done(f);
+			td->nr_done_files++;
+			dprint(FD_FILE, "%s: is done (%d of %d)\n", f->file_name,
 					td->nr_done_files, td->o.nr_files);
+		}
 	} while (1);
 
 	return 0;
@@ -1428,7 +1514,7 @@ again:
 
 	if (io_u) {
 		assert(io_u->flags & IO_U_F_FREE);
-		io_u_clear(io_u, IO_U_F_FREE | IO_U_F_NO_FILE_PUT |
+		io_u_clear(td, io_u, IO_U_F_FREE | IO_U_F_NO_FILE_PUT |
 				 IO_U_F_TRIMMED | IO_U_F_BARRIER |
 				 IO_U_F_VER_LIST);
 
@@ -1436,7 +1522,7 @@ again:
 		io_u->acct_ddir = -1;
 		td->cur_depth++;
 		assert(!(td->flags & TD_F_CHILD));
-		io_u_set(io_u, IO_U_F_IN_CUR_DEPTH);
+		io_u_set(td, io_u, IO_U_F_IN_CUR_DEPTH);
 		io_u->ipo = NULL;
 	} else if (td_async_processing(td)) {
 		/*
@@ -1471,7 +1557,7 @@ static bool check_get_trim(struct thread_data *td, struct io_u *io_u)
 			get_trim = 1;
 		}
 
-		if (get_trim && !get_next_trim(td, io_u))
+		if (get_trim && get_next_trim(td, io_u))
 			return true;
 	}
 
@@ -1516,7 +1602,7 @@ static void small_content_scramble(struct io_u *io_u)
 	unsigned int i, nr_blocks = io_u->buflen / 512;
 	uint64_t boffset;
 	unsigned int offset;
-	void *p, *end;
+	char *p, *end;
 
 	if (!nr_blocks)
 		return;
@@ -1531,7 +1617,7 @@ static void small_content_scramble(struct io_u *io_u)
 		 * the buffer, given by the product of the usec time
 		 * and the actual offset.
 		 */
-		offset = (io_u->start_time.tv_usec ^ boffset) & 511;
+		offset = ((io_u->start_time.tv_nsec/1000) ^ boffset) & 511;
 		offset &= ~(sizeof(uint64_t) - 1);
 		if (offset >= 512 - sizeof(uint64_t))
 			offset -= sizeof(uint64_t);
@@ -1593,7 +1679,7 @@ struct io_u *get_io_u(struct thread_data *td)
 	assert(fio_file_open(f));
 
 	if (ddir_rw(io_u->ddir)) {
-		if (!io_u->buflen && !(td->io_ops->flags & FIO_NOIO)) {
+		if (!io_u->buflen && !td_ioengine_flagged(td, FIO_NOIO)) {
 			dprint(FD_IO, "get_io_u: zero buflen on %p\n", io_u);
 			goto err_put;
 		}
@@ -1633,8 +1719,10 @@ out:
 	if (!td_io_prep(td, io_u)) {
 		if (!td->o.disable_lat)
 			fio_gettime(&io_u->start_time, NULL);
+
 		if (do_scramble)
 			small_content_scramble(io_u);
+
 		return io_u;
 	}
 err_put:
@@ -1686,43 +1774,46 @@ static void account_io_completion(struct thread_data *td, struct io_u *io_u,
 				  const enum fio_ddir idx, unsigned int bytes)
 {
 	const int no_reduce = !gtod_reduce(td);
-	unsigned long lusec = 0;
+	unsigned long long llnsec = 0;
 
 	if (td->parent)
 		td = td->parent;
 
+	if (!td->o.stats)
+		return;
+
 	if (no_reduce)
-		lusec = utime_since(&io_u->issue_time, &icd->time);
+		llnsec = ntime_since(&io_u->issue_time, &icd->time);
 
 	if (!td->o.disable_lat) {
-		unsigned long tusec;
+		unsigned long long tnsec;
 
-		tusec = utime_since(&io_u->start_time, &icd->time);
-		add_lat_sample(td, idx, tusec, bytes, io_u->offset);
+		tnsec = ntime_since(&io_u->start_time, &icd->time);
+		add_lat_sample(td, idx, tnsec, bytes, io_u->offset);
 
 		if (td->flags & TD_F_PROFILE_OPS) {
 			struct prof_io_ops *ops = &td->prof_io_ops;
 
 			if (ops->io_u_lat)
-				icd->error = ops->io_u_lat(td, tusec);
+				icd->error = ops->io_u_lat(td, tnsec/1000);
 		}
 
-		if (td->o.max_latency && tusec > td->o.max_latency)
-			lat_fatal(td, icd, tusec, td->o.max_latency);
-		if (td->o.latency_target && tusec > td->o.latency_target) {
+		if (td->o.max_latency && tnsec/1000 > td->o.max_latency)
+			lat_fatal(td, icd, tnsec/1000, td->o.max_latency);
+		if (td->o.latency_target && tnsec/1000 > td->o.latency_target) {
 			if (lat_target_failed(td))
-				lat_fatal(td, icd, tusec, td->o.latency_target);
+				lat_fatal(td, icd, tnsec/1000, td->o.latency_target);
 		}
 	}
 
 	if (ddir_rw(idx)) {
 		if (!td->o.disable_clat) {
-			add_clat_sample(td, idx, lusec, bytes, io_u->offset);
-			io_u_mark_latency(td, lusec);
+			add_clat_sample(td, idx, llnsec, bytes, io_u->offset);
+			io_u_mark_latency(td, llnsec);
 		}
 
 		if (!td->o.disable_bw && per_unit_log(td->bw_log))
-			add_bw_sample(td, io_u, bytes, lusec);
+			add_bw_sample(td, io_u, bytes, llnsec);
 
 		if (no_reduce && per_unit_log(td->iops_log))
 			add_iops_sample(td, io_u, bytes);
@@ -1774,7 +1865,7 @@ static void io_completed(struct thread_data *td, struct io_u **io_u_ptr,
 	dprint_io_u(io_u, "io complete");
 
 	assert(io_u->flags & IO_U_F_FLIGHT);
-	io_u_clear(io_u, IO_U_F_FLIGHT | IO_U_F_BUSY_OK);
+	io_u_clear(td, io_u, IO_U_F_FLIGHT | IO_U_F_BUSY_OK);
 
 	/*
 	 * Mark IO ok to verify
@@ -1862,7 +1953,7 @@ static void init_icd(struct thread_data *td, struct io_completion_data *icd,
 	icd->nr = nr;
 
 	icd->error = 0;
-	for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++)
+	for (ddir = 0; ddir < DDIR_RWDIR_CNT; ddir++)
 		icd->bytes_done[ddir] = 0;
 }
 
@@ -1901,7 +1992,7 @@ int io_u_sync_complete(struct thread_data *td, struct io_u *io_u)
 		return -1;
 	}
 
-	for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++)
+	for (ddir = 0; ddir < DDIR_RWDIR_CNT; ddir++)
 		td->bytes_done[ddir] += icd.bytes_done[ddir];
 
 	return 0;
@@ -1917,7 +2008,7 @@ int io_u_queued_complete(struct thread_data *td, int min_evts)
 	int ret, ddir;
 	struct timespec ts = { .tv_sec = 0, .tv_nsec = 0, };
 
-	dprint(FD_IO, "io_u_queued_completed: min=%d\n", min_evts);
+	dprint(FD_IO, "io_u_queued_complete: min=%d\n", min_evts);
 
 	if (!min_evts)
 		tvp = &ts;
@@ -1940,7 +2031,7 @@ int io_u_queued_complete(struct thread_data *td, int min_evts)
 		return -1;
 	}
 
-	for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++)
+	for (ddir = 0; ddir < DDIR_RWDIR_CNT; ddir++)
 		td->bytes_done[ddir] += icd.bytes_done[ddir];
 
 	return ret;
@@ -1951,10 +2042,10 @@ int io_u_queued_complete(struct thread_data *td, int min_evts)
  */
 void io_u_queued(struct thread_data *td, struct io_u *io_u)
 {
-	if (!td->o.disable_slat) {
+	if (!td->o.disable_slat && ramp_time_over(td) && td->o.stats) {
 		unsigned long slat_time;
 
-		slat_time = utime_since(&io_u->start_time, &io_u->issue_time);
+		slat_time = ntime_since(&io_u->start_time, &io_u->issue_time);
 
 		if (td->parent)
 			td = td->parent;
@@ -1999,6 +2090,9 @@ void fill_io_buffer(struct thread_data *td, void *buf, unsigned int min_write,
 {
 	struct thread_options *o = &td->o;
 
+	if (o->mem_type == MEM_CUDA_MALLOC)
+		return;
+
 	if (o->compress_percentage || o->dedupe_percentage) {
 		unsigned int perc = td->o.compress_percentage;
 		struct frand_state *rs;
@@ -2044,3 +2138,61 @@ void io_u_fill_buffer(struct thread_data *td, struct io_u *io_u,
 	io_u->buf_filled_len = 0;
 	fill_io_buffer(td, io_u->buf, min_write, max_bs);
 }
+
+static int do_sync_file_range(const struct thread_data *td,
+			      struct fio_file *f)
+{
+	off64_t offset, nbytes;
+
+	offset = f->first_write;
+	nbytes = f->last_write - f->first_write;
+
+	if (!nbytes)
+		return 0;
+
+	return sync_file_range(f->fd, offset, nbytes, td->o.sync_file_range);
+}
+
+int do_io_u_sync(const struct thread_data *td, struct io_u *io_u)
+{
+	int ret;
+
+	if (io_u->ddir == DDIR_SYNC) {
+		ret = fsync(io_u->file->fd);
+	} else if (io_u->ddir == DDIR_DATASYNC) {
+#ifdef CONFIG_FDATASYNC
+		ret = fdatasync(io_u->file->fd);
+#else
+		ret = io_u->xfer_buflen;
+		io_u->error = EINVAL;
+#endif
+	} else if (io_u->ddir == DDIR_SYNC_FILE_RANGE)
+		ret = do_sync_file_range(td, io_u->file);
+	else {
+		ret = io_u->xfer_buflen;
+		io_u->error = EINVAL;
+	}
+
+	if (ret < 0)
+		io_u->error = errno;
+
+	return ret;
+}
+
+int do_io_u_trim(const struct thread_data *td, struct io_u *io_u)
+{
+#ifndef FIO_HAVE_TRIM
+	io_u->error = EINVAL;
+	return 0;
+#else
+	struct fio_file *f = io_u->file;
+	int ret;
+
+	ret = os_trim(f->fd, io_u->offset, io_u->xfer_buflen);
+	if (!ret)
+		return io_u->xfer_buflen;
+
+	io_u->error = ret;
+	return 0;
+#endif
+}