X-Git-Url: https://git.kernel.dk/?a=blobdiff_plain;f=fio.c;h=e5d8205598f64428be23a8813203b232a28ab13e;hb=e9b2a3fa90451c8390f8690f93d55fa0f52bf9d7;hp=df284581e0434f3e9eaeac327ee7952c88374000;hpb=d07f439ead57caecbcf4e049d322c17be92bc99b;p=fio.git

diff --git a/fio.c b/fio.c
index df284581..e5d82055 100644
--- a/fio.c
+++ b/fio.c
@@ -65,6 +65,7 @@ extern unsigned long long mlock_size;
 enum {
 	TD_NOT_CREATED = 0,
 	TD_CREATED,
+	TD_INITIALIZED,
 	TD_RUNNING,
 	TD_VERIFYING,
 	TD_EXITED,
@@ -76,6 +77,7 @@ enum {
 static sem_t startup_sem;
 
 #define TERMINATE_ALL		(-1)
+#define JOB_START_TIMEOUT	(5 * 1000)
 
 static void terminate_threads(int group_id)
 {
@@ -195,7 +197,7 @@ static int get_next_free_block(struct thread_data *td, unsigned long long *b)
 
 static void mark_random_map(struct thread_data *td, struct io_u *io_u)
 {
-	unsigned long block = io_u->offset / td->min_bs;
+	unsigned long long block = io_u->offset / (unsigned long long) td->min_bs;
 	unsigned int blocks = 0;
 
 	while (blocks < (io_u->buflen / td->min_bs)) {
@@ -286,12 +288,12 @@ static int get_next_offset(struct thread_data *td, unsigned long long *offset)
 	long r;
 
 	if (!td->sequential) {
-		unsigned long max_blocks = td->io_size / td->min_bs;
+		unsigned long long max_blocks = td->io_size / td->min_bs;
 		int loops = 50;
 
 		do {
 			lrand48_r(&td->random_state, &r);
-			b = ((max_blocks - 1) * r / (RAND_MAX+1.0));
+			b = ((max_blocks - 1) * r / (unsigned long long) (RAND_MAX+1.0));
 			rb = b + (td->file_offset / td->min_bs);
 			loops--;
 		} while (!random_map_free(td, rb) && loops);
@@ -543,44 +545,32 @@ static void fill_md5(struct verify_header *hdr, void *p, unsigned int len)
 	memcpy(hdr->md5_digest, md5_ctx.hash, sizeof(md5_ctx.hash));
 }
 
-unsigned int hweight32(unsigned int w)
-{
-	unsigned int res = w - ((w >> 1) & 0x55555555);
-
-	res = (res & 0x33333333) + ((res >> 2) & 0x33333333);
-	res = (res + (res >> 4)) & 0x0F0F0F0F;
-	res = res + (res >> 8);
-
-	return (res + (res >> 16)) & 0x000000FF;
-}
-
-unsigned long hweight64(unsigned long long w)
+static int get_rw_ddir(struct thread_data *td)
 {
-#if __WORDSIZE == 32
-        return hweight32((unsigned int)(w >> 32)) + hweight32((unsigned int)w);
-#elif __WORDSIZE == 64
-	unsigned long long v = w - ((w >> 1) & 0x5555555555555555ul);
+	if (td_rw(td)) {
+		struct timeval now;
+		unsigned long elapsed;
 
-	v = (v & 0x3333333333333333ul) + ((v >> 2) & 0x3333333333333333ul);
-	v = (v + (v >> 4)) & 0x0F0F0F0F0F0F0F0Ful;
-	v = v + (v >> 8);
-	v = v + (v >> 16);
+		gettimeofday(&now, NULL);
+	 	elapsed = mtime_since_now(&td->rwmix_switch);
 
-	return (v + (v >> 32)) & 0x00000000000000FFul;
-#else
-#error __WORDSIZE not defined
-#endif
-}
+		/*
+		 * Check if it's time to seed a new data direction.
+		 */
+		if (elapsed >= td->rwmixcycle) {
+			unsigned long v;
+			long r;
 
-static int get_rw_ddir(struct thread_data *td)
-{
-	/*
-	 * perhaps cheasy, but use the hamming weight of the position
-	 * as a randomizer for data direction.
-	 */
-	if (td_rw(td))
-		return hweight64(td->last_pos) & 1;
-	else if (td_read(td))
+			lrand48_r(&td->random_state, &r);
+			v = 100UL * r / (unsigned long) (RAND_MAX + 1.0);
+			if (v < td->rwmixread)
+				td->rwmix_ddir = DDIR_READ;
+			else
+				td->rwmix_ddir = DDIR_WRITE;
+			memcpy(&td->rwmix_switch, &now, sizeof(now));
+		}
+		return td->rwmix_ddir;
+	} else if (td_read(td))
 		return DDIR_READ;
 	else
 		return DDIR_WRITE;
@@ -626,17 +616,16 @@ void put_io_u(struct thread_data *td, struct io_u *io_u)
 	td->cur_depth--;
 }
 
-static int fill_io_u(struct thread_data *td, struct io_u *io_u)
+static void write_iolog_put(struct thread_data *td, struct io_u *io_u)
 {
-	/*
-	 * If using an iolog, grab next piece if any available.
-	 */
-	if (td->iolog) {
-		struct io_piece *ipo;
+	fprintf(td->iolog_f, "%d,%llu,%u\n", io_u->ddir, io_u->offset, io_u->buflen);
+}
 
-		if (list_empty(&td->io_log_list))
-			return 1;
+static int read_iolog_get(struct thread_data *td, struct io_u *io_u)
+{
+	struct io_piece *ipo;
 
+	if (!list_empty(&td->io_log_list)) {
 		ipo = list_entry(td->io_log_list.next, struct io_piece, list);
 		list_del(&ipo->list);
 		io_u->offset = ipo->offset;
@@ -646,6 +635,17 @@ static int fill_io_u(struct thread_data *td, struct io_u *io_u)
 		return 0;
 	}
 
+	return 1;
+}
+
+static int fill_io_u(struct thread_data *td, struct io_u *io_u)
+{
+	/*
+	 * If using an iolog, grab next piece if any available.
+	 */
+	if (td->read_iolog)
+		return read_iolog_get(td, io_u);
+
 	/*
 	 * No log, let the seq/rand engine retrieve the next position.
 	 */
@@ -654,6 +654,13 @@ static int fill_io_u(struct thread_data *td, struct io_u *io_u)
 
 		if (io_u->buflen) {
 			io_u->ddir = get_rw_ddir(td);
+
+			/*
+			 * If using a write iolog, store this entry.
+			 */
+			if (td->write_iolog)
+				write_iolog_put(td, io_u);
+
 			return 0;
 		}
 	}
@@ -705,7 +712,7 @@ static struct io_u *get_io_u(struct thread_data *td)
 		return NULL;
 	}
 
-	if (!td->iolog && !td->sequential)
+	if (!td->read_iolog && !td->sequential)
 		mark_random_map(td, io_u);
 
 	td->last_pos += io_u->buflen;
@@ -793,6 +800,13 @@ static void log_io_piece(struct thread_data *td, struct io_u *io_u)
 	list_add(&ipo->list, entry);
 }
 
+static void write_iolog_close(struct thread_data *td)
+{
+	fflush(td->iolog_f);
+	fclose(td->iolog_f);
+	free(td->iolog_buf);
+}
+
 static int init_iolog(struct thread_data *td)
 {
 	unsigned long long offset;
@@ -801,15 +815,34 @@ static int init_iolog(struct thread_data *td)
 	FILE *f;
 	int rw, i, reads, writes;
 
-	if (!td->iolog)
+	if (!td->read_iolog && !td->write_iolog)
 		return 0;
 
-	f = fopen(td->iolog_file, "r");
+	if (td->read_iolog)
+		f = fopen(td->iolog_file, "r");
+	else
+		f = fopen(td->iolog_file, "w");
+
 	if (!f) {
 		perror("fopen iolog");
+		printf("file %s, %d/%d\n", td->iolog_file, td->read_iolog, td->write_iolog);
 		return 1;
 	}
 
+	/*
+	 * That's it for writing, setup a log buffer and we're done.
+	  */
+	if (td->write_iolog) {
+		td->iolog_f = f;
+		td->iolog_buf = malloc(8192);
+		setvbuf(f, td->iolog_buf, _IOFBF, 8192);
+		return 0;
+	}
+
+	/*
+	 * Read in the read iolog and store it, reuse the infrastructure
+	 * for doing verifications.
+	 */
 	str = malloc(4096);
 	reads = writes = i = 0;
 	while ((p = fgets(str, 4096, f)) != NULL) {
@@ -1249,6 +1282,22 @@ static int init_io_u(struct thread_data *td)
 	return 0;
 }
 
+static void cleanup_allocs(struct thread_data *td)
+{
+	if (td->directory)
+		free(td->directory);
+	if (td->iolog_file)
+		free(td->iolog_file);
+	if (td->exec_prerun)
+		free(td->exec_prerun);
+	if (td->exec_postrun)
+		free(td->exec_postrun);
+	if (td->ioscheduler)
+		free(td->ioscheduler);
+	if (td->sysfs_root)
+		free(td->sysfs_root);
+}
+
 static int create_file(struct thread_data *td, unsigned long long size,
 		       int extend)
 {
@@ -1754,9 +1803,58 @@ static void init_disk_util(struct thread_data *td)
 		sprintf(foo, "%s", p);
 	}
 
+	td->sysfs_root = strdup(foo);
 	disk_util_add(dev, foo);
 }
 
+static int switch_ioscheduler(struct thread_data *td)
+{
+	char tmp[256], tmp2[128];
+	FILE *f;
+	int ret;
+
+	sprintf(tmp, "%s/queue/scheduler", td->sysfs_root);
+
+	f = fopen(tmp, "r+");
+	if (!f) {
+		td_verror(td, errno);
+		return 1;
+	}
+
+	/*
+	 * Set io scheduler.
+	 */
+	ret = fwrite(td->ioscheduler, strlen(td->ioscheduler), 1, f);
+	if (ferror(f) || ret != 1) {
+		td_verror(td, errno);
+		fclose(f);
+		return 1;
+	}
+
+	rewind(f);
+
+	/*
+	 * Read back and check that the selected scheduler is now the default.
+	 */
+	ret = fread(tmp, 1, sizeof(tmp), f);
+	if (ferror(f) || ret < 0) {
+		td_verror(td, errno);
+		fclose(f);
+		return 1;
+	}
+
+	sprintf(tmp2, "[%s]", td->ioscheduler);
+	if (!strstr(tmp, tmp2)) {
+		fprintf(stderr, "fio: io scheduler %s not found\n", td->ioscheduler);
+		td_verror(td, EINVAL);
+		fclose(f);
+		return 1;
+	}
+
+	fclose(f);
+	return 0;
+}
+
 static void disk_util_timer_arm(void)
 {
 	itimer.it_value.tv_sec = 0;
@@ -1796,7 +1894,6 @@ static void update_rusage_stat(struct thread_data *td)
 static void *thread_main(void *data)
 {
 	struct thread_data *td = data;
-	int ret = 1;
 
 	if (!td->use_thread)
 		setsid();
@@ -1829,17 +1926,29 @@ static void *thread_main(void *data)
 		}
 	}
 
+	if (nice(td->nice) < 0) {
+		td_verror(td, errno);
+		goto err;
+	}
+
+	if (init_random_state(td))
+		goto err;
+
+	if (td->ioscheduler && switch_ioscheduler(td))
+		goto err;
+
+	td_set_runstate(td, TD_INITIALIZED);
 	sem_post(&startup_sem);
 	sem_wait(&td->mutex);
 
 	if (!td->create_serialize && setup_file(td))
 		goto err;
 
-	if (init_random_state(td))
-		goto err;
-
 	gettimeofday(&td->epoch, NULL);
 
+	if (td->exec_prerun)
+		system(td->exec_prerun);
+
 	while (td->loops--) {
 		getrusage(RUSAGE_SELF, &td->ru_start);
 		gettimeofday(&td->start, NULL);
@@ -1876,14 +1985,16 @@ static void *thread_main(void *data)
 			break;
 	}
 
-	ret = 0;
-
 	if (td->bw_log)
 		finish_log(td, td->bw_log, "bw");
 	if (td->slat_log)
 		finish_log(td, td->slat_log, "slat");
 	if (td->clat_log)
 		finish_log(td, td->clat_log, "clat");
+	if (td->write_iolog)
+		write_iolog_close(td);
+	if (td->exec_postrun)
+		system(td->exec_postrun);
 
 	if (exitall_on_terminate)
 		terminate_threads(td->groupid);
@@ -1895,12 +2006,9 @@ err:
 	}
 	if (td->mmap)
 		munmap(td->mmap, td->file_size);
+	cleanup_allocs(td);
 	cleanup_io(td);
 	cleanup_io_u(td);
-	if (ret) {
-		sem_post(&startup_sem);
-		sem_wait(&td->mutex);
-	}
 	td_set_runstate(td, TD_EXITED);
 	return NULL;
 
@@ -1947,14 +2055,15 @@ static void show_ddir_status(struct thread_data *td, struct group_run_stats *rs,
 			     int ddir)
 {
 	char *ddir_str[] = { "read ", "write" };
-	unsigned long min, max, bw;
+	unsigned long min, max;
+	unsigned long long bw;
 	double mean, dev;
 
 	if (!td->runtime[ddir])
 		return;
 
 	bw = td->io_bytes[ddir] / td->runtime[ddir];
-	printf("  %s: io=%6lluMiB, bw=%6luKiB/s, runt=%6lumsec\n", ddir_str[ddir], td->io_bytes[ddir] >> 20, bw, td->runtime[ddir]);
+	printf("  %s: io=%6lluMiB, bw=%6lluKiB/s, runt=%6lumsec\n", ddir_str[ddir], td->io_bytes[ddir] >> 20, bw, td->runtime[ddir]);
 
 	if (calc_lat(&td->slat_stat[ddir], &min, &max, &mean, &dev))
 		printf("    slat (msec): min=%5lu, max=%5lu, avg=%5.02f, dev=%5.02f\n", min, max, mean, dev);
@@ -2035,6 +2144,9 @@ static void check_str_update(struct thread_data *td)
 		case TD_CREATED:
 			c = 'C';
 			break;
+		case TD_INITIALIZED:
+			c = 'I';
+			break;
 		case TD_NOT_CREATED:
 			c = 'P';
 			break;
@@ -2106,7 +2218,8 @@ static int thread_eta(struct thread_data *td, unsigned long elapsed)
 
 		if (td->timeout && eta_sec > (td->timeout - elapsed))
 			eta_sec = td->timeout - elapsed;
-	} else if (td->runstate == TD_NOT_CREATED || td->runstate == TD_CREATED) {
+	} else if (td->runstate == TD_NOT_CREATED || td->runstate == TD_CREATED
+			|| td->runstate == TD_INITIALIZED) {
 		int t_eta = 0, r_eta = 0;
 
 		/*
@@ -2317,6 +2430,10 @@ static void run_threads(void)
 	gettimeofday(&genesis, NULL);
 
 	while (todo) {
+		struct thread_data *map[MAX_JOBS];
+		struct timeval this_start;
+		int this_jobs = 0, left;
+
 		/*
 		 * create threads (TD_NOT_CREATED -> TD_CREATED)
 		 */
@@ -2345,9 +2462,13 @@ static void run_threads(void)
 			if (td->stonewall && (nr_started || nr_running))
 				break;
 
+			/*
+			 * Set state to created. Thread will transition
+			 * to TD_INITIALIZED when it's done setting up.
+			 */
 			td_set_runstate(td, TD_CREATED);
+			map[this_jobs++] = td;
 			sem_init(&startup_sem, 0, 1);
-			todo--;
 			nr_started++;
 
 			if (td->use_thread) {
@@ -2366,12 +2487,53 @@ static void run_threads(void)
 		}
 
 		/*
-		 * start created threads (TD_CREATED -> TD_RUNNING)
+		 * Wait for the started threads to transition to
+		 * TD_INITIALIZED.
 		 */
+		printf("fio: Waiting for threads to initialize...\n");
+		gettimeofday(&this_start, NULL);
+		left = this_jobs;
+		while (left) {
+			if (mtime_since_now(&this_start) > JOB_START_TIMEOUT)
+				break;
+
+			usleep(100000);
+
+			for (i = 0; i < this_jobs; i++) {
+				td = map[i];
+				if (!td)
+					continue;
+				if (td->runstate == TD_INITIALIZED) {
+					map[i] = NULL;
+					left--;
+				} else if (td->runstate >= TD_EXITED) {
+					map[i] = NULL;
+					left--;
+					todo--;
+					nr_running++; /* work-around... */
+				}
+			}
+		}
+
+		if (left) {
+			fprintf(stderr, "fio: %d jobs failed to start\n", left);
+			for (i = 0; i < this_jobs; i++) {
+				td = map[i];
+				if (!td)
+					continue;
+				kill(td->pid, SIGTERM);
+			}
+			break;
+		}
+
+		/*
+		 * start created threads (TD_INITIALIZED -> TD_RUNNING).
+		 */
+		printf("fio: Go for launch\n");
 		for (i = 0; i < thread_number; i++) {
 			td = &threads[i];
 
-			if (td->runstate != TD_CREATED)
+			if (td->runstate != TD_INITIALIZED)
 				continue;
 
 			td_set_runstate(td, TD_RUNNING);
@@ -2379,6 +2541,7 @@ static void run_threads(void)
 			nr_started--;
 			m_rate += td->ratemin;
 			t_rate += td->rate;
+			todo--;
 			sem_post(&td->mutex);
 		}
 
@@ -2402,9 +2565,9 @@ static void show_group_stats(struct group_run_stats *rs, int id)
 	printf("\nRun status group %d (all jobs):\n", id);
 
 	if (rs->max_run[DDIR_READ])
-		printf("   READ: io=%lluMiB, aggrb=%llu, minb=%llu, maxb=%llu, mint=%llumsec, maxt=%llumsec\n", rs->io_mb[0], rs->agg[0], rs->min_bw[0], rs->max_bw[0], rs->min_run[0], rs->max_run[0]);
+		printf("   READ: io=%lluMiB, aggrb=%llu, minb=%llu, maxb=%llu, mint=%llumsec, maxt=%llumsec\n", rs->io_kb[0] >> 10, rs->agg[0], rs->min_bw[0], rs->max_bw[0], rs->min_run[0], rs->max_run[0]);
 	if (rs->max_run[DDIR_WRITE])
-		printf("  WRITE: io=%lluMiB, aggrb=%llu, minb=%llu, maxb=%llu, mint=%llumsec, maxt=%llumsec\n", rs->io_mb[1], rs->agg[1], rs->min_bw[1], rs->max_bw[1], rs->min_run[1], rs->max_run[1]);
+		printf("  WRITE: io=%lluMiB, aggrb=%llu, minb=%llu, maxb=%llu, mint=%llumsec, maxt=%llumsec\n", rs->io_kb[1] >> 10, rs->agg[1], rs->min_bw[1], rs->max_bw[1], rs->min_run[1], rs->max_run[1]);
 }
 
 static void show_disk_util(void)
@@ -2445,7 +2608,7 @@ static void show_run_stats(void)
 	}
 
 	for (i = 0; i < thread_number; i++) {
-		unsigned long rbw, wbw;
+		unsigned long long rbw, wbw;
 
 		td = &threads[i];
 
@@ -2467,9 +2630,9 @@ static void show_run_stats(void)
 
 		rbw = wbw = 0;
 		if (td->runtime[0])
-			rbw = td->io_bytes[0] / td->runtime[0];
+			rbw = td->io_bytes[0] / (unsigned long long) td->runtime[0];
 		if (td->runtime[1])
-			wbw = td->io_bytes[1] / td->runtime[1];
+			wbw = td->io_bytes[1] / (unsigned long long) td->runtime[1];
 
 		if (rbw < rs->min_bw[0])
 			rs->min_bw[0] = rbw;
@@ -2480,17 +2643,17 @@ static void show_run_stats(void)
 		if (wbw > rs->max_bw[1])
 			rs->max_bw[1] = wbw;
 
-		rs->io_mb[0] += td->io_bytes[0] >> 20;
-		rs->io_mb[1] += td->io_bytes[1] >> 20;
+		rs->io_kb[0] += td->io_bytes[0] >> 10;
+		rs->io_kb[1] += td->io_bytes[1] >> 10;
 	}
 
 	for (i = 0; i < groupid + 1; i++) {
 		rs = &runstats[i];
 
 		if (rs->max_run[0])
-			rs->agg[0] = (rs->io_mb[0]*1024*1000) / rs->max_run[0];
+			rs->agg[0] = (rs->io_kb[0]*1024) / rs->max_run[0];
 		if (rs->max_run[1])
-			rs->agg[1] = (rs->io_mb[1]*1024*1000) / rs->max_run[1];
+			rs->agg[1] = (rs->io_kb[1]*1024) / rs->max_run[1];
 	}
 
 	/*