X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=io_u.c;h=6dda5790a7f1d494fd46758ecc9b3b6406e4a11a;hp=16065128c324e76e3cf9326ff1824fec42baad5b;hb=82407585a3b3b10281428e4ca9c2d3e7dfca7392;hpb=56d9fa4b8d4fa5166e3ec5dcdd37b5789b2cb01d

diff --git a/io_u.c b/io_u.c
index 16065128..6dda5790 100644
--- a/io_u.c
+++ b/io_u.c
@@ -12,6 +12,7 @@
 #include "lib/rand.h"
 #include "lib/axmap.h"
 #include "err.h"
+#include "lib/pow2.h"
 
 struct io_completion_data {
 	int nr;				/* input */
@@ -88,18 +89,20 @@ static int __get_next_rand_offset(struct thread_data *td, struct fio_file *f,
 {
 	uint64_t r;
 
-	if (td->o.random_generator == FIO_RAND_GEN_TAUSWORTHE) {
-		uint64_t lastb;
+	if (td->o.random_generator == FIO_RAND_GEN_TAUSWORTHE ||
+	    td->o.random_generator == FIO_RAND_GEN_TAUSWORTHE64) {
+		uint64_t frand_max, lastb;
 
 		lastb = last_block(td, f, ddir);
 		if (!lastb)
 			return 1;
 
+		frand_max = rand_max(&td->random_state);
 		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 / ((uint64_t) frand_max + 1.0));
 	} else {
 		uint64_t off = 0;
 
@@ -194,7 +197,8 @@ static inline int should_sort_io(struct thread_data *td)
 		return 0;
 	if (td->runstate != TD_VERIFYING)
 		return 0;
-	if (td->o.random_generator == FIO_RAND_GEN_TAUSWORTHE)
+	if (td->o.random_generator == FIO_RAND_GEN_TAUSWORTHE ||
+	    td->o.random_generator == FIO_RAND_GEN_TAUSWORTHE64)
 		return 0;
 
 	return 1;
@@ -202,14 +206,16 @@ static inline int should_sort_io(struct thread_data *td)
 
 static int should_do_random(struct thread_data *td, enum fio_ddir ddir)
 {
+	uint64_t frand_max;
 	unsigned int v;
 	unsigned long r;
 
 	if (td->o.perc_rand[ddir] == 100)
 		return 1;
 
+	frand_max = rand_max(&td->seq_rand_state[ddir]);
 	r = __rand(&td->seq_rand_state[ddir]);
-	v = 1 + (int) (100.0 * (r / (FRAND_MAX + 1.0)));
+	v = 1 + (int) (100.0 * (r / (frand_max + 1.0)));
 
 	return v <= td->o.perc_rand[ddir];
 }
@@ -326,7 +332,7 @@ static int get_next_block(struct thread_data *td, struct io_u *io_u,
 				*is_random = 1;
 			} else {
 				*is_random = 0;
-				io_u->flags |= IO_U_F_BUSY_OK;
+				io_u_set(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);
@@ -336,7 +342,7 @@ 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->flags |= IO_U_F_BUSY_OK;
+		io_u_set(io_u, IO_U_F_BUSY_OK);
 		*is_random = 0;
 
 		if (td->o.rw_seq == RW_SEQ_SEQ) {
@@ -438,6 +444,7 @@ 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 +464,13 @@ 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 {
@@ -474,7 +482,7 @@ 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 <= ((frand_max / 100L) * perc)) &&
 				    io_u_fits(td, io_u, buflen))
 					break;
 			}
@@ -485,7 +493,7 @@ static unsigned int __get_next_buflen(struct thread_data *td, struct io_u *io_u,
 				~(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));
 
@@ -520,11 +528,12 @@ static void set_rwmix_bytes(struct thread_data *td)
 
 static inline enum fio_ddir get_rand_ddir(struct thread_data *td)
 {
+	uint64_t frand_max = rand_max(&td->rwmix_state);
 	unsigned int v;
 	unsigned long r;
 
 	r = __rand(&td->rwmix_state);
-	v = 1 + (int) (100.0 * (r / (FRAND_MAX + 1.0)));
+	v = 1 + (int) (100.0 * (r / (frand_max + 1.0)));
 
 	if (v <= td->o.rwmix[DDIR_READ])
 		return DDIR_READ;
@@ -552,55 +561,54 @@ void io_u_quiesce(struct thread_data *td)
 	while (td->io_u_in_flight) {
 		int fio_unused ret;
 
-		ret = io_u_queued_complete(td, 1, NULL);
+		ret = io_u_queued_complete(td, 1);
 	}
 }
 
 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;
 
-	io_u_quiesce(td);
+	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;
 }
 
@@ -668,13 +676,23 @@ static enum fio_ddir get_rw_ddir(struct thread_data *td)
 
 static void set_rw_ddir(struct thread_data *td, struct io_u *io_u)
 {
-	io_u->ddir = io_u->acct_ddir = get_rw_ddir(td);
+	enum fio_ddir ddir = get_rw_ddir(td);
+
+	if (td_trimwrite(td)) {
+		struct fio_file *f = io_u->file;
+		if (f->last_pos[DDIR_WRITE] == f->last_pos[DDIR_TRIM])
+			ddir = DDIR_TRIM;
+		else
+			ddir = DDIR_WRITE;
+	}
+
+	io_u->ddir = io_u->acct_ddir = ddir;
 
 	if (io_u->ddir == DDIR_WRITE && (td->io_ops->flags & FIO_BARRIER) &&
 	    td->o.barrier_blocks &&
 	   !(td->io_issues[DDIR_WRITE] % td->o.barrier_blocks) &&
 	     td->io_issues[DDIR_WRITE])
-		io_u->flags |= IO_U_F_BARRIER;
+		io_u_set(io_u, IO_U_F_BARRIER);
 }
 
 void put_file_log(struct thread_data *td, struct fio_file *f)
@@ -687,16 +705,21 @@ void put_file_log(struct thread_data *td, struct fio_file *f)
 
 void put_io_u(struct thread_data *td, struct io_u *io_u)
 {
+	if (td->parent)
+		td = td->parent;
+
 	td_io_u_lock(td);
 
 	if (io_u->file && !(io_u->flags & IO_U_F_NO_FILE_PUT))
 		put_file_log(td, io_u->file);
 
 	io_u->file = NULL;
-	io_u->flags |= IO_U_F_FREE;
+	io_u_set(io_u, IO_U_F_FREE);
 
-	if (io_u->flags & IO_U_F_IN_CUR_DEPTH)
+	if (io_u->flags & IO_U_F_IN_CUR_DEPTH) {
 		td->cur_depth--;
+		assert(!(td->flags & TD_F_CHILD));
+	}
 	io_u_qpush(&td->io_u_freelist, io_u);
 	td_io_u_unlock(td);
 	td_io_u_free_notify(td);
@@ -704,7 +727,7 @@ 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->flags &= ~IO_U_F_FLIGHT;
+	io_u_clear(io_u, IO_U_F_FLIGHT);
 	put_io_u(td, io_u);
 }
 
@@ -715,18 +738,24 @@ void requeue_io_u(struct thread_data *td, struct io_u **io_u)
 
 	dprint(FD_IO, "requeue %p\n", __io_u);
 
+	if (td->parent)
+		td = td->parent;
+
 	td_io_u_lock(td);
 
-	__io_u->flags |= IO_U_F_FREE;
+	io_u_set(__io_u, IO_U_F_FREE);
 	if ((__io_u->flags & IO_U_F_FLIGHT) && ddir_rw(ddir))
 		td->io_issues[ddir]--;
 
-	__io_u->flags &= ~IO_U_F_FLIGHT;
-	if (__io_u->flags & IO_U_F_IN_CUR_DEPTH)
+	io_u_clear(__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));
+	}
 
 	io_u_rpush(&td->io_u_requeues, __io_u);
 	td_io_u_unlock(td);
+	td_io_u_free_notify(td);
 	*io_u = NULL;
 }
 
@@ -969,6 +998,7 @@ 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;
 
@@ -978,7 +1008,7 @@ 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))
@@ -1319,21 +1349,23 @@ again:
 
 	if (io_u) {
 		assert(io_u->flags & IO_U_F_FREE);
-		io_u->flags &= ~(IO_U_F_FREE | IO_U_F_NO_FILE_PUT |
+		io_u_clear(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);
 
 		io_u->error = 0;
 		io_u->acct_ddir = -1;
 		td->cur_depth++;
-		io_u->flags |= IO_U_F_IN_CUR_DEPTH;
+		assert(!(td->flags & TD_F_CHILD));
+		io_u_set(io_u, IO_U_F_IN_CUR_DEPTH);
 		io_u->ipo = NULL;
-	} else if (td->o.verify_async) {
+	} else if (td_async_processing(td)) {
 		/*
 		 * We ran out, wait for async verify threads to finish and
 		 * return one
 		 */
-		pthread_cond_wait(&td->free_cond, &td->io_u_lock);
+		assert(!(td->flags & TD_F_CHILD));
+		assert(!pthread_cond_wait(&td->free_cond, &td->io_u_lock));
 		goto again;
 	}
 
@@ -1532,7 +1564,7 @@ err_put:
 	return ERR_PTR(ret);
 }
 
-void io_u_log_error(struct thread_data *td, struct io_u *io_u)
+static void __io_u_log_error(struct thread_data *td, struct io_u *io_u)
 {
 	enum error_type_bit eb = td_error_type(io_u->ddir, io_u->error);
 
@@ -1546,10 +1578,24 @@ 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");
 }
 
+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);
+}
+
 static inline int gtod_reduce(struct thread_data *td)
 {
 	return td->o.disable_clat && td->o.disable_lat && td->o.disable_slat
@@ -1560,9 +1606,13 @@ static void account_io_completion(struct thread_data *td, struct io_u *io_u,
 				  struct io_completion_data *icd,
 				  const enum fio_ddir idx, unsigned int bytes)
 {
+	const int no_reduce = !gtod_reduce(td);
 	unsigned long lusec = 0;
 
-	if (!gtod_reduce(td))
+	if (td->parent)
+		td = td->parent;
+
+	if (no_reduce)
 		lusec = utime_since(&io_u->issue_time, &icd->time);
 
 	if (!td->o.disable_lat) {
@@ -1594,19 +1644,21 @@ static void account_io_completion(struct thread_data *td, struct io_u *io_u,
 	if (!td->o.disable_bw)
 		add_bw_sample(td, idx, bytes, &icd->time);
 
-	if (!gtod_reduce(td))
+	if (no_reduce)
 		add_iops_sample(td, idx, bytes, &icd->time);
-}
 
-static long long usec_for_io(struct thread_data *td, enum fio_ddir ddir)
-{
-	uint64_t secs, remainder, bps, bytes;
-
-	bytes = td->this_io_bytes[ddir];
-	bps = td->rate_bps[ddir];
-	secs = bytes / bps;
-	remainder = bytes % bps;
-	return remainder * 1000000 / bps + secs * 1000000;
+	if (td->ts.nr_block_infos && io_u->ddir == DDIR_TRIM) {
+		uint32_t *info = io_u_block_info(td, io_u);
+		if (BLOCK_INFO_STATE(*info) < BLOCK_STATE_TRIM_FAILURE) {
+			if (io_u->ddir == DDIR_TRIM) {
+				*info = BLOCK_INFO(BLOCK_STATE_TRIMMED,
+						BLOCK_INFO_TRIMS(*info) + 1);
+			} else if (io_u->ddir == DDIR_WRITE) {
+				*info = BLOCK_INFO_SET_STATE(BLOCK_STATE_WRITTEN,
+								*info);
+			}
+		}
+	}
 }
 
 static void io_completed(struct thread_data *td, struct io_u **io_u_ptr,
@@ -1618,9 +1670,8 @@ static void io_completed(struct thread_data *td, struct io_u **io_u_ptr,
 
 	dprint_io_u(io_u, "io complete");
 
-	td_io_u_lock(td);
 	assert(io_u->flags & IO_U_F_FLIGHT);
-	io_u->flags &= ~(IO_U_F_FLIGHT | IO_U_F_BUSY_OK);
+	io_u_clear(io_u, IO_U_F_FLIGHT | IO_U_F_BUSY_OK);
 
 	/*
 	 * Mark IO ok to verify
@@ -1637,8 +1688,6 @@ static void io_completed(struct thread_data *td, struct io_u **io_u_ptr,
 		}
 	}
 
-	td_io_u_unlock(td);
-
 	if (ddir_sync(ddir)) {
 		td->last_was_sync = 1;
 		if (f) {
@@ -1653,7 +1702,6 @@ 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]++;
@@ -1682,22 +1730,9 @@ static void io_completed(struct thread_data *td, struct io_u **io_u_ptr,
 		}
 
 		if (ramp_time_over(td) && (td->runstate == TD_RUNNING ||
-					   td->runstate == TD_VERIFYING)) {
+					   td->runstate == TD_VERIFYING))
 			account_io_completion(td, io_u, icd, ddir, bytes);
 
-			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) {
@@ -1762,10 +1797,10 @@ static void ios_completed(struct thread_data *td,
 /*
  * Complete a single io_u for the sync engines.
  */
-int io_u_sync_complete(struct thread_data *td, struct io_u *io_u,
-		       uint64_t *bytes)
+int io_u_sync_complete(struct thread_data *td, struct io_u *io_u)
 {
 	struct io_completion_data icd;
+	int ddir;
 
 	init_icd(td, &icd, 1);
 	io_completed(td, &io_u, &icd);
@@ -1778,12 +1813,8 @@ int io_u_sync_complete(struct thread_data *td, struct io_u *io_u,
 		return -1;
 	}
 
-	if (bytes) {
-		int ddir;
-
-		for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++)
-			bytes[ddir] += icd.bytes_done[ddir];
-	}
+	for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++)
+		td->bytes_done[ddir] += icd.bytes_done[ddir];
 
 	return 0;
 }
@@ -1791,12 +1822,11 @@ int io_u_sync_complete(struct thread_data *td, struct io_u *io_u,
 /*
  * Called to complete min_events number of io for the async engines.
  */
-int io_u_queued_complete(struct thread_data *td, int min_evts,
-			 uint64_t *bytes)
+int io_u_queued_complete(struct thread_data *td, int min_evts)
 {
 	struct io_completion_data icd;
 	struct timespec *tvp = NULL;
-	int ret;
+	int ret, ddir;
 	struct timespec ts = { .tv_sec = 0, .tv_nsec = 0, };
 
 	dprint(FD_IO, "io_u_queued_completed: min=%d\n", min_evts);
@@ -1806,7 +1836,9 @@ 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;
@@ -1820,12 +1852,8 @@ int io_u_queued_complete(struct thread_data *td, int min_evts,
 		return -1;
 	}
 
-	if (bytes) {
-		int ddir;
-
-		for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++)
-			bytes[ddir] += icd.bytes_done[ddir];
-	}
+	for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++)
+		td->bytes_done[ddir] += icd.bytes_done[ddir];
 
 	return 0;
 }
@@ -1839,6 +1867,10 @@ 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);
 	}
@@ -1849,16 +1881,20 @@ void io_u_queued(struct thread_data *td, struct io_u *io_u)
  */
 static struct frand_state *get_buf_state(struct thread_data *td)
 {
+	uint64_t frand_max;
 	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;
+	}
 
+	frand_max = rand_max(&td->dedupe_state);
 	r = __rand(&td->dedupe_state);
-	v = 1 + (int) (100.0 * (r / (FRAND_MAX + 1.0)));
+	v = 1 + (int) (100.0 * (r / (frand_max + 1.0)));
 
 	if (v <= td->o.dedupe_percentage)
 		return &td->buf_state_prev;
@@ -1868,7 +1904,9 @@ 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);
 }