X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=io_u.c;h=f9870e70bc8d408ab7bf72adcf46cddf6818b4da;hp=ba3f7ca00fcd1db4f657f5dcf7321fc8c9f7ca0e;hb=b0a84f48509d76032455842f1482901e3ab12afd;hpb=36870c47801ca85d1a98c7ec47950b0aa360f865

diff --git a/io_u.c b/io_u.c
index ba3f7ca0..f9870e70 100644
--- a/io_u.c
+++ b/io_u.c
@@ -12,6 +12,8 @@
 #include "lib/rand.h"
 #include "lib/axmap.h"
 #include "err.h"
+#include "lib/pow2.h"
+#include "minmax.h"
 
 struct io_completion_data {
 	int nr;				/* input */
@@ -25,7 +27,7 @@ struct io_completion_data {
  * The ->io_axmap contains a map of blocks we have or have not done io
  * to yet. Used to make sure we cover the entire range in a fair fashion.
  */
-static int random_map_free(struct fio_file *f, const uint64_t block)
+static bool random_map_free(struct fio_file *f, const uint64_t block)
 {
 	return !axmap_isset(f->io_axmap, block);
 }
@@ -84,22 +86,19 @@ struct rand_off {
 };
 
 static int __get_next_rand_offset(struct thread_data *td, struct fio_file *f,
-				  enum fio_ddir ddir, uint64_t *b)
+				  enum fio_ddir ddir, uint64_t *b,
+				  uint64_t lastb)
 {
 	uint64_t r;
 
-	if (td->o.random_generator == FIO_RAND_GEN_TAUSWORTHE) {
-		uint64_t lastb;
-
-		lastb = last_block(td, f, ddir);
-		if (!lastb)
-			return 1;
+	if (td->o.random_generator == FIO_RAND_GEN_TAUSWORTHE ||
+	    td->o.random_generator == FIO_RAND_GEN_TAUSWORTHE64) {
 
 		r = __rand(&td->random_state);
 
 		dprint(FD_RANDOM, "off rand %llu\n", (unsigned long long) r);
 
-		*b = lastb * (r / ((uint64_t) FRAND_MAX + 1.0));
+		*b = lastb * (r / (rand_max(&td->random_state) + 1.0));
 	} else {
 		uint64_t off = 0;
 
@@ -157,6 +156,70 @@ static int __get_next_rand_offset_gauss(struct thread_data *td,
 	return 0;
 }
 
+static int __get_next_rand_offset_zoned(struct thread_data *td,
+					struct fio_file *f, enum fio_ddir ddir,
+					uint64_t *b)
+{
+	unsigned int v, send, stotal;
+	uint64_t offset, lastb;
+	static int warned;
+	struct zone_split_index *zsi;
+
+	lastb = last_block(td, f, ddir);
+	if (!lastb)
+		return 1;
+
+	if (!td->o.zone_split_nr[ddir]) {
+bail:
+		return __get_next_rand_offset(td, f, ddir, b, lastb);
+	}
+
+	/*
+	 * Generate a value, v, between 1 and 100, both inclusive
+	 */
+	v = rand32_between(&td->zone_state, 1, 100);
+
+	zsi = &td->zone_state_index[ddir][v - 1];
+	stotal = zsi->size_perc_prev;
+	send = zsi->size_perc;
+
+	/*
+	 * Should never happen
+	 */
+	if (send == -1U) {
+		if (!warned) {
+			log_err("fio: bug in zoned generation\n");
+			warned = 1;
+		}
+		goto bail;
+	}
+
+	/*
+	 * 'send' is some percentage below or equal to 100 that
+	 * marks the end of the current IO range. 'stotal' marks
+	 * the start, in percent.
+	 */
+	if (stotal)
+		offset = stotal * lastb / 100ULL;
+	else
+		offset = 0;
+
+	lastb = lastb * (send - stotal) / 100ULL;
+
+	/*
+	 * Generate index from 0..send-of-lastb
+	 */
+	if (__get_next_rand_offset(td, f, ddir, b, lastb) == 1)
+		return 1;
+
+	/*
+	 * Add our start offset, if any
+	 */
+	if (offset)
+		*b += offset;
+
+	return 0;
+}
 
 static int flist_cmp(void *data, struct flist_head *a, struct flist_head *b)
 {
@@ -169,14 +232,22 @@ static int flist_cmp(void *data, struct flist_head *a, struct flist_head *b)
 static int get_off_from_method(struct thread_data *td, struct fio_file *f,
 			       enum fio_ddir ddir, uint64_t *b)
 {
-	if (td->o.random_distribution == FIO_RAND_DIST_RANDOM)
-		return __get_next_rand_offset(td, f, ddir, b);
-	else if (td->o.random_distribution == FIO_RAND_DIST_ZIPF)
+	if (td->o.random_distribution == FIO_RAND_DIST_RANDOM) {
+		uint64_t lastb;
+
+		lastb = last_block(td, f, ddir);
+		if (!lastb)
+			return 1;
+
+		return __get_next_rand_offset(td, f, ddir, b, lastb);
+	} else if (td->o.random_distribution == FIO_RAND_DIST_ZIPF)
 		return __get_next_rand_offset_zipf(td, f, ddir, b);
 	else if (td->o.random_distribution == FIO_RAND_DIST_PARETO)
 		return __get_next_rand_offset_pareto(td, f, ddir, b);
 	else if (td->o.random_distribution == FIO_RAND_DIST_GAUSS)
 		return __get_next_rand_offset_gauss(td, f, ddir, b);
+	else if (td->o.random_distribution == FIO_RAND_DIST_ZONED)
+		return __get_next_rand_offset_zoned(td, f, ddir, b);
 
 	log_err("fio: unknown random distribution: %d\n", td->o.random_distribution);
 	return 1;
@@ -186,30 +257,29 @@ static int get_off_from_method(struct thread_data *td, struct fio_file *f,
  * Sort the reads for a verify phase in batches of verifysort_nr, if
  * specified.
  */
-static inline int should_sort_io(struct thread_data *td)
+static inline bool should_sort_io(struct thread_data *td)
 {
 	if (!td->o.verifysort_nr || !td->o.do_verify)
-		return 0;
+		return false;
 	if (!td_random(td))
-		return 0;
+		return false;
 	if (td->runstate != TD_VERIFYING)
-		return 0;
-	if (td->o.random_generator == FIO_RAND_GEN_TAUSWORTHE)
-		return 0;
+		return false;
+	if (td->o.random_generator == FIO_RAND_GEN_TAUSWORTHE ||
+	    td->o.random_generator == FIO_RAND_GEN_TAUSWORTHE64)
+		return false;
 
-	return 1;
+	return true;
 }
 
-static int should_do_random(struct thread_data *td, enum fio_ddir ddir)
+static bool should_do_random(struct thread_data *td, enum fio_ddir ddir)
 {
 	unsigned int v;
-	unsigned long r;
 
 	if (td->o.perc_rand[ddir] == 100)
-		return 1;
+		return true;
 
-	r = __rand(&td->seq_rand_state[ddir]);
-	v = 1 + (int) (100.0 * (r / (FRAND_MAX + 1.0)));
+	v = rand32_between(&td->seq_rand_state[ddir], 1, 100);
 
 	return v <= td->o.perc_rand[ddir];
 }
@@ -278,8 +348,15 @@ static int get_next_seq_offset(struct thread_data *td, struct fio_file *f,
 	assert(ddir_rw(ddir));
 
 	if (f->last_pos[ddir] >= f->io_size + get_start_offset(td, f) &&
-	    o->time_based)
-		f->last_pos[ddir] = f->last_pos[ddir] - f->io_size;
+	    o->time_based) {
+		struct thread_options *o = &td->o;
+		uint64_t io_size = f->io_size + (f->io_size % o->min_bs[ddir]);
+
+		if (io_size > f->last_pos[ddir])
+			f->last_pos[ddir] = 0;
+		else
+			f->last_pos[ddir] = f->last_pos[ddir] - io_size;
+	}
 
 	if (f->last_pos[ddir] < f->real_file_size) {
 		uint64_t pos;
@@ -294,10 +371,15 @@ static int get_next_seq_offset(struct thread_data *td, struct fio_file *f,
 			/*
 			 * If we reach beyond the end of the file
 			 * with holed IO, wrap around to the
-			 * beginning again.
+			 * beginning again. If we're doing backwards IO,
+			 * wrap to the end.
 			 */
-			if (pos >= f->real_file_size)
-				pos = f->file_offset;
+			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;
+			}
 		}
 
 		*offset = pos;
@@ -424,8 +506,8 @@ static int get_next_offset(struct thread_data *td, struct io_u *io_u,
 	return __get_next_offset(td, io_u, is_random);
 }
 
-static inline int io_u_fits(struct thread_data *td, struct io_u *io_u,
-			    unsigned int buflen)
+static inline bool io_u_fits(struct thread_data *td, struct io_u *io_u,
+			     unsigned int buflen)
 {
 	struct fio_file *f = io_u->file;
 
@@ -438,6 +520,7 @@ 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;
 
 	assert(ddir_rw(ddir));
@@ -457,12 +540,13 @@ static unsigned int __get_next_buflen(struct thread_data *td, struct io_u *io_u,
 	if (!io_u_fits(td, io_u, minbs))
 		return 0;
 
+	frand_max = rand_max(&td->bsrange_state);
 	do {
 		r = __rand(&td->bsrange_state);
 
 		if (!td->o.bssplit_nr[ddir]) {
 			buflen = 1 + (unsigned int) ((double) maxbs *
-					(r / (FRAND_MAX + 1.0)));
+					(r / (frand_max + 1.0)));
 			if (buflen < minbs)
 				buflen = minbs;
 		} else {
@@ -474,18 +558,18 @@ static unsigned int __get_next_buflen(struct thread_data *td, struct io_u *io_u,
 
 				buflen = bsp->bs;
 				perc += bsp->perc;
-				if ((r <= ((FRAND_MAX / 100L) * perc)) &&
+				if ((r * 100UL <= frand_max * perc) &&
 				    io_u_fits(td, io_u, buflen))
 					break;
 			}
 		}
 
-		if (td->o.do_verify && td->o.verify != VERIFY_NONE)
+		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))
-			buflen = (buflen + minbs - 1) & ~(minbs - 1);
+			buflen &= ~(minbs - 1);
 
 	} while (!io_u_fits(td, io_u, buflen));
 
@@ -521,10 +605,8 @@ static void set_rwmix_bytes(struct thread_data *td)
 static inline enum fio_ddir get_rand_ddir(struct thread_data *td)
 {
 	unsigned int v;
-	unsigned long r;
 
-	r = __rand(&td->rwmix_state);
-	v = 1 + (int) (100.0 * (r / (FRAND_MAX + 1.0)));
+	v = rand32_between(&td->rwmix_state, 1, 100);
 
 	if (v <= td->o.rwmix[DDIR_READ])
 		return DDIR_READ;
@@ -532,8 +614,10 @@ static inline enum fio_ddir get_rand_ddir(struct thread_data *td)
 	return DDIR_WRITE;
 }
 
-void io_u_quiesce(struct thread_data *td)
+int io_u_quiesce(struct thread_data *td)
 {
+	int completed = 0;
+
 	/*
 	 * We are going to sleep, ensure that we flush anything pending as
 	 * not to skew our latency numbers.
@@ -553,55 +637,57 @@ void io_u_quiesce(struct thread_data *td)
 		int fio_unused ret;
 
 		ret = io_u_queued_complete(td, 1);
+		if (ret > 0)
+			completed += ret;
 	}
+
+	return completed;
 }
 
 static enum fio_ddir rate_ddir(struct thread_data *td, enum fio_ddir ddir)
 {
 	enum fio_ddir odir = ddir ^ 1;
-	long usec;
+	long usec, now;
 
 	assert(ddir_rw(ddir));
+	now = utime_since_now(&td->start);
 
-	if (td->rate_pending_usleep[ddir] <= 0)
+	/*
+	 * if rate_next_io_time is in the past, need to catch up to rate
+	 */
+	if (td->rate_next_io_time[ddir] <= now)
 		return ddir;
 
 	/*
-	 * We have too much pending sleep in this direction. See if we
+	 * We are ahead of rate in this direction. See if we
 	 * should switch.
 	 */
 	if (td_rw(td) && td->o.rwmix[odir]) {
 		/*
-		 * Other direction does not have too much pending, switch
+		 * Other direction is behind rate, switch
 		 */
-		if (td->rate_pending_usleep[odir] < 100000)
+		if (td->rate_next_io_time[odir] <= now)
 			return odir;
 
 		/*
-		 * Both directions have pending sleep. Sleep the minimum time
-		 * and deduct from both.
+		 * Both directions are ahead of rate. sleep the min
+		 * switch if necissary
 		 */
-		if (td->rate_pending_usleep[ddir] <=
-			td->rate_pending_usleep[odir]) {
-			usec = td->rate_pending_usleep[ddir];
+		if (td->rate_next_io_time[ddir] <=
+			td->rate_next_io_time[odir]) {
+			usec = td->rate_next_io_time[ddir] - now;
 		} else {
-			usec = td->rate_pending_usleep[odir];
+			usec = td->rate_next_io_time[odir] - now;
 			ddir = odir;
 		}
 	} else
-		usec = td->rate_pending_usleep[ddir];
+		usec = td->rate_next_io_time[ddir] - now;
 
 	if (td->o.io_submit_mode == IO_MODE_INLINE)
 		io_u_quiesce(td);
 
 	usec = usec_sleep(td, usec);
 
-	td->rate_pending_usleep[ddir] -= usec;
-
-	odir = ddir ^ 1;
-	if (td_rw(td) && __should_check_rate(td, odir))
-		td->rate_pending_usleep[odir] -= usec;
-
 	return ddir;
 }
 
@@ -991,6 +1077,7 @@ 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;
 
@@ -1000,7 +1087,7 @@ static struct fio_file *get_next_file_rand(struct thread_data *td,
 
 		r = __rand(&td->next_file_state);
 		fno = (unsigned int) ((double) td->o.nr_files
-				* (r / (FRAND_MAX + 1.0)));
+				* (r / (frand_max + 1.0)));
 
 		f = td->files[fno];
 		if (fio_file_done(f))
@@ -1182,10 +1269,10 @@ static void lat_new_cycle(struct thread_data *td)
  * We had an IO outside the latency target. Reduce the queue depth. If we
  * are at QD=1, then it's time to give up.
  */
-static int __lat_target_failed(struct thread_data *td)
+static bool __lat_target_failed(struct thread_data *td)
 {
 	if (td->latency_qd == 1)
-		return 1;
+		return true;
 
 	td->latency_qd_high = td->latency_qd;
 
@@ -1202,16 +1289,16 @@ static int __lat_target_failed(struct thread_data *td)
 	 */
 	io_u_quiesce(td);
 	lat_new_cycle(td);
-	return 0;
+	return false;
 }
 
-static int lat_target_failed(struct thread_data *td)
+static bool lat_target_failed(struct thread_data *td)
 {
 	if (td->o.latency_percentile.u.f == 100.0)
 		return __lat_target_failed(td);
 
 	td->latency_failed++;
-	return 0;
+	return false;
 }
 
 void lat_target_init(struct thread_data *td)
@@ -1306,14 +1393,14 @@ void lat_target_check(struct thread_data *td)
  * If latency target is enabled, we might be ramping up or down and not
  * using the full queue depth available.
  */
-int queue_full(const struct thread_data *td)
+bool queue_full(const struct thread_data *td)
 {
 	const int qempty = io_u_qempty(&td->io_u_freelist);
 
 	if (qempty)
-		return 1;
+		return true;
 	if (!td->o.latency_target)
-		return 0;
+		return false;
 
 	return td->cur_depth >= td->latency_qd;
 }
@@ -1365,10 +1452,10 @@ again:
 	return io_u;
 }
 
-static int check_get_trim(struct thread_data *td, struct io_u *io_u)
+static bool check_get_trim(struct thread_data *td, struct io_u *io_u)
 {
 	if (!(td->flags & TD_F_TRIM_BACKLOG))
-		return 0;
+		return false;
 
 	if (td->trim_entries) {
 		int get_trim = 0;
@@ -1385,16 +1472,16 @@ static int check_get_trim(struct thread_data *td, struct io_u *io_u)
 		}
 
 		if (get_trim && !get_next_trim(td, io_u))
-			return 1;
+			return true;
 	}
 
-	return 0;
+	return false;
 }
 
-static int check_get_verify(struct thread_data *td, struct io_u *io_u)
+static bool check_get_verify(struct thread_data *td, struct io_u *io_u)
 {
 	if (!(td->flags & TD_F_VER_BACKLOG))
-		return 0;
+		return false;
 
 	if (td->io_hist_len) {
 		int get_verify = 0;
@@ -1411,11 +1498,11 @@ static int check_get_verify(struct thread_data *td, struct io_u *io_u)
 
 		if (get_verify && !get_next_verify(td, io_u)) {
 			td->verify_batch--;
-			return 1;
+			return true;
 		}
 	}
 
-	return 0;
+	return false;
 }
 
 /*
@@ -1518,7 +1605,7 @@ struct io_u *get_io_u(struct thread_data *td)
 			if (td->flags & TD_F_REFILL_BUFFERS) {
 				io_u_fill_buffer(td, io_u,
 					td->o.min_bs[DDIR_WRITE],
-					io_u->xfer_buflen);
+					io_u->buflen);
 			} else if ((td->flags & TD_F_SCRAMBLE_BUFFERS) &&
 				   !(td->flags & TD_F_COMPRESS))
 				do_scramble = 1;
@@ -1544,7 +1631,7 @@ struct io_u *get_io_u(struct thread_data *td)
 out:
 	assert(io_u->file);
 	if (!td_io_prep(td, io_u)) {
-		if (!td->o.disable_slat)
+		if (!td->o.disable_lat)
 			fio_gettime(&io_u->start_time, NULL);
 		if (do_scramble)
 			small_content_scramble(io_u);
@@ -1570,6 +1657,13 @@ static void __io_u_log_error(struct thread_data *td, struct io_u *io_u)
 		io_ddir_name(io_u->ddir),
 		io_u->offset, io_u->xfer_buflen);
 
+	if (td->io_ops->errdetails) {
+		char *err = td->io_ops->errdetails(io_u);
+
+		log_err("fio: %s\n", err);
+		free(err);
+	}
+
 	if (!td->error)
 		td_verror(td, io_u->error, "io_u error");
 }
@@ -1578,13 +1672,13 @@ void io_u_log_error(struct thread_data *td, struct io_u *io_u)
 {
 	__io_u_log_error(td, io_u);
 	if (td->parent)
-		__io_u_log_error(td, io_u);
+		__io_u_log_error(td->parent, io_u);
 }
 
-static inline int gtod_reduce(struct thread_data *td)
+static inline bool gtod_reduce(struct thread_data *td)
 {
-	return td->o.disable_clat && td->o.disable_lat && td->o.disable_slat
-		&& td->o.disable_bw;
+	return (td->o.disable_clat && td->o.disable_slat && td->o.disable_bw)
+			|| td->o.gtod_reduce;
 }
 
 static void account_io_completion(struct thread_data *td, struct io_u *io_u,
@@ -1594,6 +1688,9 @@ static void account_io_completion(struct thread_data *td, struct io_u *io_u,
 	const int no_reduce = !gtod_reduce(td);
 	unsigned long lusec = 0;
 
+	if (td->parent)
+		td = td->parent;
+
 	if (no_reduce)
 		lusec = utime_since(&io_u->issue_time, &icd->time);
 
@@ -1618,19 +1715,18 @@ static void account_io_completion(struct thread_data *td, struct io_u *io_u,
 		}
 	}
 
-	if (!td->o.disable_clat) {
-		add_clat_sample(td, idx, lusec, bytes, io_u->offset);
-		io_u_mark_latency(td, lusec);
-	}
-
-	if (td->parent)
-		td = td->parent;
+	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);
+		}
 
-	if (!td->o.disable_bw)
-		add_bw_sample(td, idx, bytes, &icd->time);
+		if (!td->o.disable_bw && per_unit_log(td->bw_log))
+			add_bw_sample(td, io_u, bytes, lusec);
 
-	if (no_reduce)
-		add_iops_sample(td, idx, bytes, &icd->time);
+		if (no_reduce && per_unit_log(td->iops_log))
+			add_iops_sample(td, io_u, bytes);
+	}
 
 	if (td->ts.nr_block_infos && io_u->ddir == DDIR_TRIM) {
 		uint32_t *info = io_u_block_info(td, io_u);
@@ -1646,16 +1742,26 @@ static void account_io_completion(struct thread_data *td, struct io_u *io_u,
 	}
 }
 
-static long long usec_for_io(struct thread_data *td, enum fio_ddir ddir)
+static void file_log_write_comp(const struct thread_data *td, struct fio_file *f,
+				uint64_t offset, unsigned int bytes)
 {
-	uint64_t secs, remainder, bps, bytes;
+	int idx;
+
+	if (!f)
+		return;
+
+	if (f->first_write == -1ULL || offset < f->first_write)
+		f->first_write = offset;
+	if (f->last_write == -1ULL || ((offset + bytes) > f->last_write))
+		f->last_write = offset + bytes;
 
-	assert(!(td->flags & TD_F_CHILD));
-	bytes = td->this_io_bytes[ddir];
-	bps = td->rate_bps[ddir];
-	secs = bytes / bps;
-	remainder = bytes % bps;
-	return remainder * 1000000 / bps + secs * 1000000;
+	if (!f->last_write_comp)
+		return;
+
+	idx = f->last_write_idx++;
+	f->last_write_comp[idx] = offset;
+	if (f->last_write_idx == td->o.iodepth)
+		f->last_write_idx = 0;
 }
 
 static void io_completed(struct thread_data *td, struct io_u **io_u_ptr,
@@ -1699,7 +1805,6 @@ static void io_completed(struct thread_data *td, struct io_u **io_u_ptr,
 
 	if (!io_u->error && ddir_rw(ddir)) {
 		unsigned int bytes = io_u->buflen - io_u->resid;
-		const enum fio_ddir oddir = ddir ^ 1;
 		int ret;
 
 		td->io_blocks[ddir]++;
@@ -1709,46 +1814,13 @@ static void io_completed(struct thread_data *td, struct io_u **io_u_ptr,
 		if (!(io_u->flags & IO_U_F_VER_LIST))
 			td->this_io_bytes[ddir] += bytes;
 
-		if (ddir == DDIR_WRITE) {
-			if (f) {
-				if (f->first_write == -1ULL ||
-				    io_u->offset < f->first_write)
-					f->first_write = io_u->offset;
-				if (f->last_write == -1ULL ||
-				    ((io_u->offset + bytes) > f->last_write))
-					f->last_write = io_u->offset + bytes;
-			}
-			if (td->last_write_comp) {
-				int idx = td->last_write_idx++;
-
-				td->last_write_comp[idx] = io_u->offset;
-				if (td->last_write_idx == td->o.iodepth)
-					td->last_write_idx = 0;
-			}
-		}
+		if (ddir == DDIR_WRITE)
+			file_log_write_comp(td, f, io_u->offset, bytes);
 
 		if (ramp_time_over(td) && (td->runstate == TD_RUNNING ||
-					   td->runstate == TD_VERIFYING)) {
-			struct thread_data *__td = td;
-
+					   td->runstate == TD_VERIFYING))
 			account_io_completion(td, io_u, icd, ddir, bytes);
 
-			if (td->parent)
-				__td = td->parent;
-
-			if (__should_check_rate(__td, ddir)) {
-				__td->rate_pending_usleep[ddir] =
-					(usec_for_io(__td, ddir) -
-					 utime_since_now(&__td->start));
-			}
-			if (ddir != DDIR_TRIM &&
-			    __should_check_rate(__td, oddir)) {
-				__td->rate_pending_usleep[oddir] =
-					(usec_for_io(__td, oddir) -
-					 utime_since_now(&__td->start));
-			}
-		}
-
 		icd->bytes_done[ddir] += bytes;
 
 		if (io_u->end_io) {
@@ -1852,7 +1924,9 @@ int io_u_queued_complete(struct thread_data *td, int min_evts)
 	else if (min_evts > td->cur_depth)
 		min_evts = td->cur_depth;
 
-	ret = td_io_getevents(td, min_evts, td->o.iodepth_batch_complete, tvp);
+	/* No worries, td_io_getevents fixes min and max if they are
+	 * set incorrectly */
+	ret = td_io_getevents(td, min_evts, td->o.iodepth_batch_complete_max, tvp);
 	if (ret < 0) {
 		td_verror(td, -ret, "td_io_getevents");
 		return ret;
@@ -1869,7 +1943,7 @@ int io_u_queued_complete(struct thread_data *td, int min_evts)
 	for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++)
 		td->bytes_done[ddir] += icd.bytes_done[ddir];
 
-	return 0;
+	return ret;
 }
 
 /*
@@ -1881,6 +1955,10 @@ void io_u_queued(struct thread_data *td, struct io_u *io_u)
 		unsigned long slat_time;
 
 		slat_time = utime_since(&io_u->start_time, &io_u->issue_time);
+
+		if (td->parent)
+			td = td->parent;
+
 		add_slat_sample(td, io_u->ddir, slat_time, io_u->xfer_buflen,
 				io_u->offset);
 	}
@@ -1892,15 +1970,15 @@ void io_u_queued(struct thread_data *td, struct io_u *io_u)
 static struct frand_state *get_buf_state(struct thread_data *td)
 {
 	unsigned int v;
-	unsigned long r;
 
 	if (!td->o.dedupe_percentage)
 		return &td->buf_state;
-	else if (td->o.dedupe_percentage == 100)
-		return &td->buf_state_prev;
+	else if (td->o.dedupe_percentage == 100) {
+		frand_copy(&td->buf_state_prev, &td->buf_state);
+		return &td->buf_state;
+	}
 
-	r = __rand(&td->dedupe_state);
-	v = 1 + (int) (100.0 * (r / (FRAND_MAX + 1.0)));
+	v = rand32_between(&td->dedupe_state, 1, 100);
 
 	if (v <= td->o.dedupe_percentage)
 		return &td->buf_state_prev;
@@ -1910,7 +1988,9 @@ static struct frand_state *get_buf_state(struct thread_data *td)
 
 static void save_buf_state(struct thread_data *td, struct frand_state *rs)
 {
-	if (rs == &td->buf_state)
+	if (td->o.dedupe_percentage == 100)
+		frand_copy(rs, &td->buf_state_prev);
+	else if (rs == &td->buf_state)
 		frand_copy(&td->buf_state_prev, rs);
 }
 
@@ -1923,6 +2003,7 @@ void fill_io_buffer(struct thread_data *td, void *buf, unsigned int min_write,
 		unsigned int perc = td->o.compress_percentage;
 		struct frand_state *rs;
 		unsigned int left = max_bs;
+		unsigned int this_write;
 
 		do {
 			rs = get_buf_state(td);
@@ -1930,20 +2011,20 @@ void fill_io_buffer(struct thread_data *td, void *buf, unsigned int min_write,
 			min_write = min(min_write, left);
 
 			if (perc) {
-				unsigned int seg = min_write;
+				this_write = min_not_zero(min_write,
+							td->o.compress_chunk);
 
-				seg = min(min_write, td->o.compress_chunk);
-				if (!seg)
-					seg = min_write;
-
-				fill_random_buf_percentage(rs, buf, perc, seg,
-					min_write, o->buffer_pattern,
-						   o->buffer_pattern_bytes);
-			} else
+				fill_random_buf_percentage(rs, buf, perc,
+					this_write, this_write,
+					o->buffer_pattern,
+					o->buffer_pattern_bytes);
+			} else {
 				fill_random_buf(rs, buf, min_write);
+				this_write = min_write;
+			}
 
-			buf += min_write;
-			left -= min_write;
+			buf += this_write;
+			left -= this_write;
 			save_buf_state(td, rs);
 		} while (left);
 	} else if (o->buffer_pattern_bytes)