X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=io_u.c;h=44933a185f7a780906e06edb3b0f422c28fe05e8;hp=69bec4bf4bcf83d57d1edf40e63c0a34f912bd02;hb=e1c325d25dd977c28c9489c542a51ee05dfc620e;hpb=5c8e84cab3bca39de54a69092473f000f8a57f40

diff --git a/io_u.c b/io_u.c
index 69bec4bf..44933a18 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 */
 };
 
 /*
@@ -37,7 +37,7 @@ static bool random_map_free(struct fio_file *f, const uint64_t block)
  */
 static void mark_random_map(struct thread_data *td, struct io_u *io_u)
 {
-	unsigned int min_bs = td->o.rw_min_bs;
+	unsigned int min_bs = td->o.min_bs[io_u->ddir];
 	struct fio_file *f = io_u->file;
 	unsigned int nr_blocks;
 	uint64_t block;
@@ -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)
@@ -156,6 +157,66 @@ static int __get_next_rand_offset_gauss(struct thread_data *td,
 	return 0;
 }
 
+static int __get_next_rand_offset_zoned_abs(struct thread_data *td,
+					    struct fio_file *f,
+					    enum fio_ddir ddir, uint64_t *b)
+{
+	struct zone_split_index *zsi;
+	uint64_t lastb, send, stotal;
+	static int warned;
+	unsigned int v;
+
+	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);
+
+	/*
+	 * Find our generated table. 'send' is the end block of this zone,
+	 * 'stotal' is our start offset.
+	 */
+	zsi = &td->zone_state_index[ddir][v - 1];
+	stotal = zsi->size_prev / td->o.ba[ddir];
+	send = zsi->size / td->o.ba[ddir];
+
+	/*
+	 * Should never happen
+	 */
+	if (send == -1U) {
+		if (!warned) {
+			log_err("fio: bug in zoned generation\n");
+			warned = 1;
+		}
+		goto bail;
+	} else if (send > lastb) {
+		/*
+		 * This happens if the user specifies ranges that exceed
+		 * the file/device size. We can't handle that gracefully,
+		 * so error and exit.
+		 */
+		log_err("fio: zoned_abs sizes exceed file size\n");
+		return 1;
+	}
+
+	/*
+	 * Generate index from 0..send-stotal
+	 */
+	if (__get_next_rand_offset(td, f, ddir, b, send - stotal) == 1)
+		return 1;
+
+	*b += stotal;
+	return 0;
+}
+
 static int __get_next_rand_offset_zoned(struct thread_data *td,
 					struct fio_file *f, enum fio_ddir ddir,
 					uint64_t *b)
@@ -248,6 +309,8 @@ static int get_off_from_method(struct thread_data *td, struct fio_file *f,
 		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);
+	else if (td->o.random_distribution == FIO_RAND_DIST_ZONED_ABS)
+		return __get_next_rand_offset_zoned_abs(td, f, ddir, b);
 
 	log_err("fio: unknown random distribution: %d\n", td->o.random_distribution);
 	return 1;
@@ -322,6 +385,17 @@ fetch:
 	goto fetch;
 }
 
+static void loop_cache_invalidate(struct thread_data *td, struct fio_file *f)
+{
+	struct thread_options *o = &td->o;
+
+	if (o->invalidate_cache && !o->odirect) {
+		int fio_unused ret;
+
+		ret = file_invalidate_cache(td, f);
+	}
+}
+
 static int get_next_rand_block(struct thread_data *td, struct fio_file *f,
 			       enum fio_ddir ddir, uint64_t *b)
 {
@@ -333,6 +407,7 @@ static int get_next_rand_block(struct thread_data *td, struct fio_file *f,
 		fio_file_reset(td, f);
 		if (!get_next_rand_offset(td, f, ddir, b))
 			return 0;
+		loop_cache_invalidate(td, f);
 	}
 
 	dprint(FD_IO, "%s: rand offset failed, last=%llu, size=%llu\n",
@@ -348,15 +423,14 @@ static int get_next_seq_offset(struct thread_data *td, struct fio_file *f,
 
 	assert(ddir_rw(ddir));
 
+	/*
+	 * If we reach the end for a time based run, reset us back to 0
+	 * and invalidate the cache, if we need to.
+	 */
 	if (f->last_pos[ddir] >= f->io_size + get_start_offset(td, f) &&
 	    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;
+		f->last_pos[ddir] = f->file_offset;
+		loop_cache_invalidate(td, f);
 	}
 
 	if (f->last_pos[ddir] < f->real_file_size) {
@@ -532,6 +606,7 @@ static unsigned int __get_next_buflen(struct thread_data *td, struct io_u *io_u,
 	unsigned int buflen = 0;
 	unsigned int minbs, maxbs;
 	uint64_t frand_max, r;
+	bool power_2;
 
 	assert(ddir_rw(ddir));
 
@@ -550,9 +625,9 @@ 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);
+	frand_max = rand_max(&td->bsrange_state[ddir]);
 	do {
-		r = __rand(&td->bsrange_state);
+		r = __rand(&td->bsrange_state[ddir]);
 
 		if (!td->o.bssplit_nr[ddir]) {
 			buflen = 1 + (unsigned int) ((double) maxbs *
@@ -576,9 +651,11 @@ static unsigned int __get_next_buflen(struct thread_data *td, struct io_u *io_u,
 			}
 		}
 
-		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;
@@ -642,7 +719,7 @@ 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)
@@ -658,7 +735,7 @@ int io_u_quiesce(struct thread_data *td)
 static enum fio_ddir rate_ddir(struct thread_data *td, enum fio_ddir ddir)
 {
 	enum fio_ddir odir = ddir ^ 1;
-	long usec;
+	uint64_t usec;
 	uint64_t now;
 
 	assert(ddir_rw(ddir));
@@ -713,28 +790,22 @@ 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
+	 * 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->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
-	 */
-	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)) {
 		/*
@@ -901,8 +972,9 @@ 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;
@@ -990,11 +1062,52 @@ 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:
@@ -1031,10 +1144,12 @@ 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;
@@ -1076,12 +1191,14 @@ 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)
@@ -1292,10 +1409,10 @@ static long set_io_u_file(struct thread_data *td, struct io_u *io_u)
 }
 
 static void lat_fatal(struct thread_data *td, struct io_completion_data *icd,
-		      unsigned long tusec, unsigned long max_usec)
+		      unsigned long long tnsec, unsigned long long max_nsec)
 {
 	if (!td->error)
-		log_err("fio: latency of %lu usec exceeds specified max (%lu usec)\n", tusec, max_usec);
+		log_err("fio: latency of %llu nsec exceeds specified max (%llu nsec)\n", tnsec, max_nsec);
 	td_verror(td, ETIMEDOUT, "max latency exceeded");
 	icd->error = ETIMEDOUT;
 }
@@ -1498,22 +1615,19 @@ static bool check_get_trim(struct thread_data *td, struct io_u *io_u)
 {
 	if (!(td->flags & TD_F_TRIM_BACKLOG))
 		return false;
+	if (!td->trim_entries)
+		return false;
 
-	if (td->trim_entries) {
-		int get_trim = 0;
-
-		if (td->trim_batch) {
-			td->trim_batch--;
-			get_trim = 1;
-		} else if (!(td->io_hist_len % td->o.trim_backlog) &&
-			 td->last_ddir != DDIR_READ) {
-			td->trim_batch = td->o.trim_batch;
-			if (!td->trim_batch)
-				td->trim_batch = td->o.trim_backlog;
-			get_trim = 1;
-		}
-
-		if (get_trim && get_next_trim(td, io_u))
+	if (td->trim_batch) {
+		td->trim_batch--;
+		if (get_next_trim(td, io_u))
+			return true;
+	} else if (!(td->io_hist_len % td->o.trim_backlog) &&
+		     td->last_ddir != DDIR_READ) {
+		td->trim_batch = td->o.trim_batch;
+		if (!td->trim_batch)
+			td->trim_batch = td->o.trim_backlog;
+		if (get_next_trim(td, io_u))
 			return true;
 	}
 
@@ -1555,32 +1669,40 @@ static bool check_get_verify(struct thread_data *td, struct io_u *io_u)
  */
 static void small_content_scramble(struct io_u *io_u)
 {
-	unsigned int i, nr_blocks = io_u->buflen / 512;
-	uint64_t boffset;
+	unsigned int i, nr_blocks = io_u->buflen >> 9;
 	unsigned int offset;
-	void *p, *end;
+	uint64_t boffset, *iptr;
+	char *p;
 
 	if (!nr_blocks)
 		return;
 
 	p = io_u->xfer_buf;
 	boffset = io_u->offset;
-	io_u->buf_filled_len = 0;
+
+	if (io_u->buf_filled_len)
+		io_u->buf_filled_len = 0;
+
+	/*
+	 * Generate random index between 0..7. We do chunks of 512b, if
+	 * we assume a cacheline is 64 bytes, then we have 8 of those.
+	 * Scramble content within the blocks in the same cacheline to
+	 * speed things up.
+	 */
+	offset = (io_u->start_time.tv_nsec ^ boffset) & 7;
 
 	for (i = 0; i < nr_blocks; i++) {
 		/*
-		 * Fill the byte offset into a "random" start offset of
-		 * the buffer, given by the product of the usec time
-		 * and the actual offset.
+		 * Fill offset into start of cacheline, time into end
+		 * of cacheline
 		 */
-		offset = (io_u->start_time.tv_usec ^ boffset) & 511;
-		offset &= ~(sizeof(uint64_t) - 1);
-		if (offset >= 512 - sizeof(uint64_t))
-			offset -= sizeof(uint64_t);
-		memcpy(p + offset, &boffset, sizeof(boffset));
-
-		end = p + 512 - sizeof(io_u->start_time);
-		memcpy(end, &io_u->start_time, sizeof(io_u->start_time));
+		iptr = (void *) p + (offset << 6);
+		*iptr = boffset;
+
+		iptr = (void *) p + 64 - 2 * sizeof(uint64_t);
+		iptr[0] = io_u->start_time.tv_sec;
+		iptr[1] = io_u->start_time.tv_nsec;
+
 		p += 512;
 		boffset += 512;
 	}
@@ -1675,8 +1797,10 @@ 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:
@@ -1728,43 +1852,46 @@ 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 || td_ioengine_flagged(td, FIO_NOSTATS))
+		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);
 		}
 
-		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 > td->o.max_latency)
+			lat_fatal(td, icd, tnsec, td->o.max_latency);
+		if (td->o.latency_target && tnsec > td->o.latency_target) {
 			if (lat_target_failed(td))
-				lat_fatal(td, icd, tusec, td->o.latency_target);
+				lat_fatal(td, icd, tnsec, 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);
@@ -1850,11 +1977,12 @@ static void io_completed(struct thread_data *td, struct io_u **io_u_ptr,
 		int ret;
 
 		td->io_blocks[ddir]++;
-		td->this_io_blocks[ddir]++;
 		td->io_bytes[ddir] += bytes;
 
-		if (!(io_u->flags & IO_U_F_VER_LIST))
+		if (!(io_u->flags & IO_U_F_VER_LIST)) {
+			td->this_io_blocks[ddir]++;
 			td->this_io_bytes[ddir] += bytes;
+		}
 
 		if (ddir == DDIR_WRITE)
 			file_log_write_comp(td, f, io_u->offset, bytes);
@@ -1959,7 +2087,7 @@ 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;
@@ -1993,10 +2121,10 @@ 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;
@@ -2041,6 +2169,9 @@ 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;
@@ -2086,3 +2217,61 @@ 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, io_u->offset, io_u->xfer_buflen);
+	if (!ret)
+		return io_u->xfer_buflen;
+
+	io_u->error = ret;
+	return 0;
+#endif
+}