X-Git-Url: https://git.kernel.dk/?p=fio.git;a=blobdiff_plain;f=io_u.c;h=8d42d6541f7dbb9592639619874295153101bd91;hp=363bfe10283f5043aa8565cd0d5c1236adf5d532;hb=6fc820956c033fcb097d88760bb6124938779668;hpb=e2c75fc49d028736eccea411fcb34b1d1b9efd36

diff --git a/io_u.c b/io_u.c
index 363bfe10..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 */
 };
 
 /*
@@ -899,8 +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;
@@ -988,11 +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:
@@ -1029,10 +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;
@@ -1074,12 +1118,14 @@ 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)
@@ -1556,7 +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;
@@ -1571,7 +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);
@@ -1673,8 +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:
@@ -1726,7 +1774,7 @@ 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;
@@ -1735,37 +1783,37 @@ static void account_io_completion(struct thread_data *td, struct io_u *io_u,
 		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);
@@ -1997,7 +2045,7 @@ void io_u_queued(struct thread_data *td, struct io_u *io_u)
 	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;
@@ -2042,6 +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;