* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
*
*/
-#include <stdio.h>
-#include <stdlib.h>
#include <unistd.h>
#include <fcntl.h>
#include <string.h>
-#include <errno.h>
#include <signal.h>
#include <time.h>
-#include <math.h>
#include <assert.h>
-#include <dirent.h>
-#include <libgen.h>
-#include <sys/types.h>
#include <sys/stat.h>
#include <sys/wait.h>
#include <sys/ipc.h>
int thread_number = 0;
static char run_str[MAX_JOBS + 1];
int shm_id = 0;
-static LIST_HEAD(disk_list);
-static struct itimerval itimer;
static struct timeval genesis;
-static void update_io_ticks(void);
-static void disk_util_timer_arm(void);
static void print_thread_status(void);
+extern unsigned long long mlock_size;
+
/*
* thread life cycle
*/
enum {
TD_NOT_CREATED = 0,
TD_CREATED,
+ TD_INITIALIZED,
TD_RUNNING,
TD_VERIFYING,
TD_EXITED,
static sem_t startup_sem;
#define TERMINATE_ALL (-1)
+#define JOB_START_TIMEOUT (5 * 1000)
static void terminate_threads(int group_id)
{
}
}
-static unsigned long utime_since(struct timeval *s, struct timeval *e)
-{
- double sec, usec;
-
- sec = e->tv_sec - s->tv_sec;
- usec = e->tv_usec - s->tv_usec;
- if (sec > 0 && usec < 0) {
- sec--;
- usec += 1000000;
- }
-
- sec *= (double) 1000000;
-
- return sec + usec;
-}
-
-static unsigned long utime_since_now(struct timeval *s)
-{
- struct timeval t;
-
- gettimeofday(&t, NULL);
- return utime_since(s, &t);
-}
-
-static unsigned long mtime_since(struct timeval *s, struct timeval *e)
-{
- double sec, usec;
-
- sec = e->tv_sec - s->tv_sec;
- usec = e->tv_usec - s->tv_usec;
- if (sec > 0 && usec < 0) {
- sec--;
- usec += 1000000;
- }
-
- sec *= (double) 1000;
- usec /= (double) 1000;
-
- return sec + usec;
-}
-
-static unsigned long mtime_since_now(struct timeval *s)
-{
- struct timeval t;
-
- gettimeofday(&t, NULL);
- return mtime_since(s, &t);
-}
-
-static inline unsigned long msec_now(struct timeval *s)
-{
- return s->tv_sec * 1000 + s->tv_usec / 1000;
-}
-
-static unsigned long time_since_now(struct timeval *s)
-{
- return mtime_since_now(s) / 1000;
-}
-
static int random_map_free(struct thread_data *td, unsigned long long block)
{
unsigned int idx = RAND_MAP_IDX(td, block);
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)) {
io_u->buflen = blocks * td->min_bs;
}
-static inline void add_stat_sample(struct io_stat *is, unsigned long val)
-{
- if (val > is->max_val)
- is->max_val = val;
- if (val < is->min_val)
- is->min_val = val;
-
- is->val += val;
- is->val_sq += val * val;
- is->samples++;
-}
-
-static void add_log_sample(struct thread_data *td, struct io_log *iolog,
- unsigned long val, int ddir)
-{
- if (iolog->nr_samples == iolog->max_samples) {
- int new_size = sizeof(struct io_sample) * iolog->max_samples*2;
-
- iolog->log = realloc(iolog->log, new_size);
- iolog->max_samples <<= 1;
- }
-
- iolog->log[iolog->nr_samples].val = val;
- iolog->log[iolog->nr_samples].time = mtime_since_now(&td->epoch);
- iolog->log[iolog->nr_samples].ddir = ddir;
- iolog->nr_samples++;
-}
-
-static void add_clat_sample(struct thread_data *td, int ddir,unsigned long msec)
-{
- add_stat_sample(&td->clat_stat[ddir], msec);
-
- if (td->clat_log)
- add_log_sample(td, td->clat_log, msec, ddir);
-}
-
-static void add_slat_sample(struct thread_data *td, int ddir,unsigned long msec)
-{
- add_stat_sample(&td->slat_stat[ddir], msec);
-
- if (td->slat_log)
- add_log_sample(td, td->slat_log, msec, ddir);
-}
-
-static void add_bw_sample(struct thread_data *td, int ddir)
-{
- unsigned long spent = mtime_since_now(&td->stat_sample_time[ddir]);
- unsigned long rate;
-
- if (spent < td->bw_avg_time)
- return;
-
- rate = (td->this_io_bytes[ddir] - td->stat_io_bytes[ddir]) / spent;
- add_stat_sample(&td->bw_stat[ddir], rate);
-
- if (td->bw_log)
- add_log_sample(td, td->bw_log, rate, ddir);
-
- gettimeofday(&td->stat_sample_time[ddir], NULL);
- td->stat_io_bytes[ddir] = td->this_io_bytes[ddir];
-}
-
static int get_next_offset(struct thread_data *td, unsigned long long *offset)
{
unsigned long long b, rb;
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);
return buflen;
}
-/*
- * busy looping version for the last few usec
- */
-static void __usec_sleep(unsigned int usec)
-{
- struct timeval start;
-
- gettimeofday(&start, NULL);
- while (utime_since_now(&start) < usec)
- nop;
-}
-
-static void usec_sleep(struct thread_data *td, unsigned long usec)
-{
- struct timespec req, rem;
-
- req.tv_sec = usec / 1000000;
- req.tv_nsec = usec * 1000 - req.tv_sec * 1000000;
-
- do {
- if (usec < 5000) {
- __usec_sleep(usec);
- break;
- }
-
- rem.tv_sec = rem.tv_nsec = 0;
- if (nanosleep(&req, &rem) < 0)
- break;
-
- if ((rem.tv_sec + rem.tv_nsec) == 0)
- break;
-
- req.tv_nsec = rem.tv_nsec;
- req.tv_sec = rem.tv_sec;
-
- usec = rem.tv_sec * 1000000 + rem.tv_nsec / 1000;
- } while (!td->terminate);
-}
-
-static void rate_throttle(struct thread_data *td, unsigned long time_spent,
- unsigned int bytes)
-{
- unsigned long usec_cycle;
-
- if (!td->rate)
- return;
-
- usec_cycle = td->rate_usec_cycle * (bytes / td->min_bs);
-
- if (time_spent < usec_cycle) {
- unsigned long s = usec_cycle - time_spent;
-
- td->rate_pending_usleep += s;
- if (td->rate_pending_usleep >= 100000) {
- usec_sleep(td, td->rate_pending_usleep);
- td->rate_pending_usleep = 0;
- }
- } else {
- long overtime = time_spent - usec_cycle;
-
- td->rate_pending_usleep -= overtime;
- }
-}
-
static int check_min_rate(struct thread_data *td, struct timeval *now)
{
unsigned long spent;
rate = (td->this_io_bytes[ddir] - td->rate_bytes) / spent;
if (rate < td->ratemin) {
- printf("Client%d: min rate %d not met, got %ldKiB/sec\n", td->thread_number, td->ratemin, rate);
+ printf("%s: min rate %d not met, got %ldKiB/sec\n", td->name, td->ratemin, rate);
if (rate_quit)
terminate_threads(td->groupid);
return 1;
{
unsigned char *p = (unsigned char *) io_u->buf;
unsigned long c;
- int ret;
p += sizeof(*hdr);
c = crc32(p, hdr->len - sizeof(*hdr));
- ret = c != hdr->crc32;
- if (ret) {
+ if (c != hdr->crc32) {
fprintf(stderr, "crc32: verify failed at %llu/%u\n", io_u->offset, io_u->buflen);
fprintf(stderr, "crc32: wanted %lx, got %lx\n", hdr->crc32, c);
+ return 1;
}
- return ret;
+ return 0;
}
static int verify_io_u_md5(struct verify_header *hdr, struct io_u *io_u)
{
unsigned char *p = (unsigned char *) io_u->buf;
struct md5_ctx md5_ctx;
- int ret;
memset(&md5_ctx, 0, sizeof(md5_ctx));
p += sizeof(*hdr);
md5_update(&md5_ctx, p, hdr->len - sizeof(*hdr));
- ret = memcmp(hdr->md5_digest, md5_ctx.hash, sizeof(md5_ctx.hash));
- if (ret) {
+ if (memcmp(hdr->md5_digest, md5_ctx.hash, sizeof(md5_ctx.hash))) {
fprintf(stderr, "md5: verify failed at %llu/%u\n", io_u->offset, io_u->buflen);
hexdump(hdr->md5_digest, sizeof(hdr->md5_digest));
hexdump(md5_ctx.hash, sizeof(md5_ctx.hash));
+ return 1;
}
- return ret;
+ return 0;
}
static int verify_io_u(struct io_u *io_u)
memcpy(hdr->md5_digest, md5_ctx.hash, sizeof(md5_ctx.hash));
}
-unsigned int hweight32(unsigned int w)
+static int get_rw_ddir(struct thread_data *td)
{
- unsigned int res = w - ((w >> 1) & 0x55555555);
+ if (td_rw(td)) {
+ struct timeval now;
+ unsigned long elapsed;
- res = (res & 0x33333333) + ((res >> 2) & 0x33333333);
- res = (res + (res >> 4)) & 0x0F0F0F0F;
- res = res + (res >> 8);
-
- return (res + (res >> 16)) & 0x000000FF;
-}
+ gettimeofday(&now, NULL);
+ elapsed = mtime_since_now(&td->rwmix_switch);
-unsigned long hweight64(unsigned long long w)
-{
-#if __WORDSIZE == 32
- return hweight32((unsigned int)(w >> 32)) + hweight32((unsigned int)w);
-#elif __WORDSIZE == 64
- unsigned long long v = w - ((w >> 1) & 0x5555555555555555ul);
-
- v = (v & 0x3333333333333333ul) + ((v >> 2) & 0x3333333333333333ul);
- v = (v + (v >> 4)) & 0x0F0F0F0F0F0F0F0Ful;
- v = v + (v >> 8);
- v = v + (v >> 16);
-
- 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;
/*
* fill body of io_u->buf with random data and add a header with the
- * (eg) sha1sum of that data.
+ * crc32 or md5 sum of that data.
*/
static void populate_io_u(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;
-
- if (list_empty(&td->io_log_list))
- return 1;
-
- ipo = list_entry(td->io_log_list.next, struct io_piece, list);
- list_del(&ipo->list);
- io_u->offset = ipo->offset;
- io_u->buflen = ipo->len;
- io_u->ddir = ipo->ddir;
- free(ipo);
- return 0;
- }
+ if (td->read_iolog)
+ return read_iolog_get(td, io_u);
/*
* No log, let the seq/rand engine retrieve the next position.
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;
}
}
return 1;
}
-#define queue_full(td) (list_empty(&(td)->io_u_freelist))
+#define queue_full(td) list_empty(&(td)->io_u_freelist)
struct io_u *__get_io_u(struct thread_data *td)
{
- struct io_u *io_u;
+ struct io_u *io_u = NULL;
- if (queue_full(td))
- return NULL;
+ if (!queue_full(td)) {
+ io_u = list_entry(td->io_u_freelist.next, struct io_u, list);
+
+ io_u->error = 0;
+ io_u->resid = 0;
+ list_del(&io_u->list);
+ list_add(&io_u->list, &td->io_u_busylist);
+ td->cur_depth++;
+ }
- io_u = list_entry(td->io_u_freelist.next, struct io_u, list);
- io_u->error = 0;
- io_u->resid = 0;
- list_del(&io_u->list);
- list_add(&io_u->list, &td->io_u_busylist);
- td->cur_depth++;
return io_u;
}
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;
{
struct io_piece *ipo;
- if (list_empty(&td->io_hist_list))
- return 1;
-
- ipo = list_entry(td->io_hist_list.next, struct io_piece, list);
- list_del(&ipo->list);
-
- io_u->offset = ipo->offset;
- io_u->buflen = ipo->len;
- io_u->ddir = DDIR_READ;
- free(ipo);
- return 0;
-}
-
-static void prune_io_piece_log(struct thread_data *td)
-{
- struct io_piece *ipo;
-
- while (!list_empty(&td->io_hist_list)) {
+ if (!list_empty(&td->io_hist_list)) {
ipo = list_entry(td->io_hist_list.next, struct io_piece, list);
list_del(&ipo->list);
- free(ipo);
- }
-}
-
-/*
- * log a succesful write, so we can unwind the log for verify
- */
-static void log_io_piece(struct thread_data *td, struct io_u *io_u)
-{
- struct io_piece *ipo = malloc(sizeof(struct io_piece));
- struct list_head *entry;
-
- INIT_LIST_HEAD(&ipo->list);
- ipo->offset = io_u->offset;
- ipo->len = io_u->buflen;
-
- /*
- * for random io where the writes extend the file, it will typically
- * be laid out with the block scattered as written. it's faster to
- * read them in in that order again, so don't sort
- */
- if (td->sequential || !td->overwrite) {
- list_add_tail(&ipo->list, &td->io_hist_list);
- return;
- }
-
- /*
- * for random io, sort the list so verify will run faster
- */
- entry = &td->io_hist_list;
- while ((entry = entry->prev) != &td->io_hist_list) {
- struct io_piece *__ipo = list_entry(entry, struct io_piece, list);
-
- if (__ipo->offset < ipo->offset)
- break;
- }
- list_add(&ipo->list, entry);
-}
-
-static int init_iolog(struct thread_data *td)
-{
- unsigned long long offset;
- unsigned int bytes;
- char *str, *p;
- FILE *f;
- int rw, i, reads, writes;
-
- if (!td->iolog)
+ io_u->offset = ipo->offset;
+ io_u->buflen = ipo->len;
+ io_u->ddir = DDIR_READ;
+ free(ipo);
return 0;
-
- f = fopen(td->iolog_file, "r");
- if (!f) {
- perror("fopen iolog");
- return 1;
- }
-
- str = malloc(4096);
- reads = writes = i = 0;
- while ((p = fgets(str, 4096, f)) != NULL) {
- struct io_piece *ipo;
-
- if (sscanf(p, "%d,%llu,%u", &rw, &offset, &bytes) != 3) {
- fprintf(stderr, "bad iolog: %s\n", p);
- continue;
- }
- if (rw == DDIR_READ)
- reads++;
- else if (rw == DDIR_WRITE)
- writes++;
- else {
- fprintf(stderr, "bad ddir: %d\n", rw);
- continue;
- }
-
- ipo = malloc(sizeof(*ipo));
- INIT_LIST_HEAD(&ipo->list);
- ipo->offset = offset;
- ipo->len = bytes;
- if (bytes > td->max_bs)
- td->max_bs = bytes;
- ipo->ddir = rw;
- list_add_tail(&ipo->list, &td->io_log_list);
- i++;
}
- free(str);
- fclose(f);
-
- if (!i)
- return 1;
-
- if (reads && !writes)
- td->ddir = DDIR_READ;
- else if (!reads && writes)
- td->ddir = DDIR_READ;
- else
- td->iomix = 1;
-
- return 0;
+ return 1;
}
static int sync_td(struct thread_data *td)
td_set_runstate(td, TD_RUNNING);
}
+/*
+ * Main IO worker functions. It retrieves io_u's to process and queues
+ * and reaps them, checking for rate and errors along the way.
+ */
static void do_io(struct thread_data *td)
{
struct io_completion_data icd;
if (td->cur_depth)
cleanup_pending_aio(td);
- if (should_fsync(td) && td->fsync_blocks)
+ if (should_fsync(td) && td->end_fsync)
sync_td(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)
{
if (!extend) {
oflags = O_CREAT | O_TRUNC;
- printf("Client%d: Laying out IO file (%LuMiB)\n", td->thread_number, size >> 20);
+ printf("%s: Laying out IO file (%LuMiB)\n", td->name, size >> 20);
} else {
oflags = O_APPEND;
- printf("Client%d: Extending IO file (%Lu -> %LuMiB)\n", td->thread_number, (td->file_size - size) >> 20, td->file_size >> 20);
+ printf("%s: Extending IO file (%Lu -> %LuMiB)\n", td->name, (td->file_size - size) >> 20, td->file_size >> 20);
}
td->fd = open(td->file_name, O_WRONLY | oflags, 0644);
return ret;
if (td->file_offset > td->real_file_size) {
- fprintf(stderr, "Client%d: offset extends end (%Lu > %Lu)\n", td->thread_number, td->file_offset, td->real_file_size);
+ fprintf(stderr, "%s: offset extends end (%Lu > %Lu)\n", td->name, td->file_offset, td->real_file_size);
return 1;
}
td->io_size = td->file_size;
if (td->io_size == 0) {
- fprintf(stderr, "Client%d: no io blocks\n", td->thread_number);
+ fprintf(stderr, "%s: no io blocks\n", td->name);
td_verror(td, EINVAL);
return 1;
}
return setup_file_mmap(td);
}
-static int check_dev_match(dev_t dev, char *path)
+static int switch_ioscheduler(struct thread_data *td)
{
- unsigned int major, minor;
- char line[256], *p;
+ char tmp[256], tmp2[128];
FILE *f;
+ int ret;
- f = fopen(path, "r");
- if (!f) {
- perror("open path");
- return 1;
- }
+ sprintf(tmp, "%s/queue/scheduler", td->sysfs_root);
- p = fgets(line, sizeof(line), f);
- if (!p) {
- fclose(f);
+ f = fopen(tmp, "r+");
+ if (!f) {
+ td_verror(td, errno);
return 1;
}
- if (sscanf(p, "%u:%u", &major, &minor) != 2) {
+ /*
+ * 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;
}
- if (((major << 8) | minor) == dev) {
- fclose(f);
- return 0;
- }
-
- fclose(f);
- return 1;
-}
-
-static int find_block_dir(dev_t dev, char *path)
-{
- struct dirent *dir;
- struct stat st;
- int found = 0;
- DIR *D;
-
- D = opendir(path);
- if (!D)
- return 0;
-
- while ((dir = readdir(D)) != NULL) {
- char full_path[256];
-
- if (!strcmp(dir->d_name, ".") || !strcmp(dir->d_name, ".."))
- continue;
- if (!strcmp(dir->d_name, "device"))
- continue;
-
- sprintf(full_path, "%s/%s", path, dir->d_name);
-
- if (!strcmp(dir->d_name, "dev")) {
- if (!check_dev_match(dev, full_path)) {
- found = 1;
- break;
- }
- }
-
- if (stat(full_path, &st) == -1) {
- perror("stat");
- break;
- }
-
- if (!S_ISDIR(st.st_mode) || S_ISLNK(st.st_mode))
- continue;
-
- found = find_block_dir(dev, full_path);
- if (found) {
- strcpy(path, full_path);
- break;
- }
- }
-
- closedir(D);
- return found;
-}
+ rewind(f);
-static int get_io_ticks(struct disk_util *du, struct disk_util_stat *dus)
-{
- unsigned in_flight;
- char line[256];
- FILE *f;
- char *p;
-
- f = fopen(du->path, "r");
- if (!f)
- return 1;
-
- p = fgets(line, sizeof(line), f);
- if (!p) {
+ /*
+ * 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;
}
- if (sscanf(p, "%u %u %llu %u %u %u %llu %u %u %u %u\n", &dus->ios[0], &dus->merges[0], &dus->sectors[0], &dus->ticks[0], &dus->ios[1], &dus->merges[1], &dus->sectors[1], &dus->ticks[1], &in_flight, &dus->io_ticks, &dus->time_in_queue) != 11) {
+ 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;
}
return 0;
}
-static void update_io_tick_disk(struct disk_util *du)
-{
- struct disk_util_stat __dus, *dus, *ldus;
- struct timeval t;
-
- if (get_io_ticks(du, &__dus))
- return;
-
- dus = &du->dus;
- ldus = &du->last_dus;
-
- dus->sectors[0] += (__dus.sectors[0] - ldus->sectors[0]);
- dus->sectors[1] += (__dus.sectors[1] - ldus->sectors[1]);
- dus->ios[0] += (__dus.ios[0] - ldus->ios[0]);
- dus->ios[1] += (__dus.ios[1] - ldus->ios[1]);
- dus->merges[0] += (__dus.merges[0] - ldus->merges[0]);
- dus->merges[1] += (__dus.merges[1] - ldus->merges[1]);
- dus->ticks[0] += (__dus.ticks[0] - ldus->ticks[0]);
- dus->ticks[1] += (__dus.ticks[1] - ldus->ticks[1]);
- dus->io_ticks += (__dus.io_ticks - ldus->io_ticks);
- dus->time_in_queue += (__dus.time_in_queue - ldus->time_in_queue);
-
- gettimeofday(&t, NULL);
- du->msec += mtime_since(&du->time, &t);
- memcpy(&du->time, &t, sizeof(t));
- memcpy(ldus, &__dus, sizeof(__dus));
-}
-
-static void update_io_ticks(void)
-{
- struct list_head *entry;
- struct disk_util *du;
-
- list_for_each(entry, &disk_list) {
- du = list_entry(entry, struct disk_util, list);
- update_io_tick_disk(du);
- }
-}
-
-static int disk_util_exists(dev_t dev)
-{
- struct list_head *entry;
- struct disk_util *du;
-
- list_for_each(entry, &disk_list) {
- du = list_entry(entry, struct disk_util, list);
-
- if (du->dev == dev)
- return 1;
- }
-
- return 0;
-}
-
-static void disk_util_add(dev_t dev, char *path)
-{
- struct disk_util *du = malloc(sizeof(*du));
-
- memset(du, 0, sizeof(*du));
- INIT_LIST_HEAD(&du->list);
- sprintf(du->path, "%s/stat", path);
- du->name = strdup(basename(path));
- du->dev = dev;
-
- gettimeofday(&du->time, NULL);
- get_io_ticks(du, &du->last_dus);
-
- list_add_tail(&du->list, &disk_list);
-}
-
-static void init_disk_util(struct thread_data *td)
-{
- struct stat st;
- char foo[256], tmp[256];
- dev_t dev;
- char *p;
-
- if (!td->do_disk_util)
- return;
-
- if (!stat(td->file_name, &st)) {
- if (S_ISBLK(st.st_mode))
- dev = st.st_rdev;
- else
- dev = st.st_dev;
- } else {
- /*
- * must be a file, open "." in that path
- */
- strcpy(foo, td->file_name);
- p = dirname(foo);
- if (stat(p, &st)) {
- perror("disk util stat");
- return;
- }
-
- dev = st.st_dev;
- }
-
- if (disk_util_exists(dev))
- return;
-
- sprintf(foo, "/sys/block");
- if (!find_block_dir(dev, foo))
- return;
-
- /*
- * for md/dm, there's no queue dir. we already have the right place
- */
- sprintf(tmp, "%s/stat", foo);
- if (stat(tmp, &st)) {
- /*
- * if this is inside a partition dir, jump back to parent
- */
- sprintf(tmp, "%s/queue", foo);
- if (stat(tmp, &st)) {
- p = dirname(foo);
- sprintf(tmp, "%s/queue", p);
- if (stat(tmp, &st)) {
- fprintf(stderr, "unknown sysfs layout\n");
- return;
- }
- sprintf(foo, "%s", p);
- }
- }
-
- disk_util_add(dev, foo);
-}
-
-static void disk_util_timer_arm(void)
-{
- itimer.it_value.tv_sec = 0;
- itimer.it_value.tv_usec = DISK_UTIL_MSEC * 1000;
- setitimer(ITIMER_REAL, &itimer, NULL);
-}
-
static void clear_io_state(struct thread_data *td)
{
if (td->io_engine == FIO_SYNCIO)
memset(td->file_map, 0, td->num_maps * sizeof(long));
}
-static void update_rusage_stat(struct thread_data *td)
-{
- if (!(td->runtime[0] + td->runtime[1]))
- return;
-
- getrusage(RUSAGE_SELF, &td->ru_end);
-
- td->usr_time += mtime_since(&td->ru_start.ru_utime, &td->ru_end.ru_utime);
- td->sys_time += mtime_since(&td->ru_start.ru_stime, &td->ru_end.ru_stime);
- td->ctx += td->ru_end.ru_nvcsw + td->ru_end.ru_nivcsw - (td->ru_start.ru_nvcsw + td->ru_start.ru_nivcsw);
-
-
- memcpy(&td->ru_start, &td->ru_end, sizeof(td->ru_end));
-}
-
static void *thread_main(void *data)
{
struct thread_data *td = data;
- int ret = 1;
if (!td->use_thread)
setsid();
}
}
+ 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);
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);
}
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;
return NULL;
}
-static int calc_lat(struct io_stat *is, unsigned long *min, unsigned long *max,
- double *mean, double *dev)
-{
- double n;
-
- if (is->samples == 0)
- return 0;
-
- *min = is->min_val;
- *max = is->max_val;
-
- n = (double) is->samples;
- *mean = (double) is->val / n;
- *dev = sqrt(((double) is->val_sq - (*mean * *mean) / n) / (n - 1));
- if (!(*min + *max) && !(*mean + *dev))
- return 0;
-
- return 1;
-}
-
-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;
- 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]);
-
- 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);
-
- if (calc_lat(&td->clat_stat[ddir], &min, &max, &mean, &dev))
- printf(" clat (msec): min=%5lu, max=%5lu, avg=%5.02f, dev=%5.02f\n", min, max, mean, dev);
-
- if (calc_lat(&td->bw_stat[ddir], &min, &max, &mean, &dev)) {
- double p_of_agg;
-
- p_of_agg = mean * 100 / (double) rs->agg[ddir];
- printf(" bw (KiB/s) : min=%5lu, max=%5lu, per=%3.2f%%, avg=%5.02f, dev=%5.02f\n", min, max, p_of_agg, mean, dev);
- }
-}
-
-static void show_thread_status(struct thread_data *td,
- struct group_run_stats *rs)
-{
- double usr_cpu, sys_cpu;
-
- if (!(td->io_bytes[0] + td->io_bytes[1]) && !td->error)
- return;
-
- printf("Client%d (groupid=%d): err=%2d:\n", td->thread_number, td->groupid, td->error);
-
- show_ddir_status(td, rs, td->ddir);
- if (td->io_bytes[td->ddir ^ 1])
- show_ddir_status(td, rs, td->ddir ^ 1);
-
- if (td->runtime[0] + td->runtime[1]) {
- double runt = td->runtime[0] + td->runtime[1];
-
- usr_cpu = (double) td->usr_time * 100 / runt;
- sys_cpu = (double) td->sys_time * 100 / runt;
- } else {
- usr_cpu = 0;
- sys_cpu = 0;
- }
-
- printf(" cpu : usr=%3.2f%%, sys=%3.2f%%, ctx=%lu\n", usr_cpu, sys_cpu, td->ctx);
-}
-
static void check_str_update(struct thread_data *td)
{
char c = run_str[td->thread_number - 1];
case TD_CREATED:
c = 'C';
break;
+ case TD_INITIALIZED:
+ c = 'I';
+ break;
case TD_NOT_CREATED:
c = 'P';
break;
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;
/*
}
}
+static void fio_unpin_memory(void *pinned)
+{
+ if (pinned) {
+ if (munlock(pinned, mlock_size) < 0)
+ perror("munlock");
+ munmap(pinned, mlock_size);
+ }
+}
+
+static void *fio_pin_memory(void)
+{
+ unsigned long long phys_mem;
+ void *ptr;
+
+ if (!mlock_size)
+ return NULL;
+
+ /*
+ * Don't allow mlock of more than real_mem-128MB
+ */
+ phys_mem = os_phys_mem();
+ if (phys_mem) {
+ if ((mlock_size + 128 * 1024 * 1024) > phys_mem) {
+ mlock_size = phys_mem - 128 * 1024 * 1024;
+ printf("fio: limiting mlocked memory to %lluMiB\n",
+ mlock_size >> 20);
+ }
+ }
+
+ ptr = mmap(NULL, mlock_size, PROT_READ | PROT_WRITE, MAP_PRIVATE | OS_MAP_ANON, 0, 0);
+ if (!ptr) {
+ perror("malloc locked mem");
+ return NULL;
+ }
+ if (mlock(ptr, mlock_size) < 0) {
+ munmap(ptr, mlock_size);
+ perror("mlock");
+ return NULL;
+ }
+
+ return ptr;
+}
+
static void run_threads(void)
{
struct thread_data *td;
unsigned long spent;
int i, todo, nr_running, m_rate, t_rate, nr_started;
+ void *mlocked_mem;
+
+ mlocked_mem = fio_pin_memory();
printf("Starting %d thread%s\n", thread_number, thread_number > 1 ? "s" : "");
fflush(stdout);
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)
*/
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) {
}
/*
- * 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);
nr_started--;
m_rate += td->ratemin;
t_rate += td->rate;
+ todo--;
sem_post(&td->mutex);
}
}
update_io_ticks();
-}
-
-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]);
- 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]);
-}
-
-static void show_disk_util(void)
-{
- struct disk_util_stat *dus;
- struct list_head *entry;
- struct disk_util *du;
- double util;
-
- printf("\nDisk stats (read/write):\n");
-
- list_for_each(entry, &disk_list) {
- du = list_entry(entry, struct disk_util, list);
- dus = &du->dus;
-
- util = (double) 100 * du->dus.io_ticks / (double) du->msec;
- if (util > 100.0)
- util = 100.0;
-
- printf(" %s: ios=%u/%u, merge=%u/%u, ticks=%u/%u, in_queue=%u, util=%3.2f%%\n", du->name, dus->ios[0], dus->ios[1], dus->merges[0], dus->merges[1], dus->ticks[0], dus->ticks[1], dus->time_in_queue, util);
- }
-}
-
-static void show_run_stats(void)
-{
- struct group_run_stats *runstats, *rs;
- struct thread_data *td;
- int i;
-
- runstats = malloc(sizeof(struct group_run_stats) * (groupid + 1));
-
- for (i = 0; i < groupid + 1; i++) {
- rs = &runstats[i];
-
- memset(rs, 0, sizeof(*rs));
- rs->min_bw[0] = rs->min_run[0] = ~0UL;
- rs->min_bw[1] = rs->min_run[1] = ~0UL;
- }
-
- for (i = 0; i < thread_number; i++) {
- unsigned long rbw, wbw;
-
- td = &threads[i];
-
- if (td->error) {
- printf("Client%d: %s\n", td->thread_number, td->verror);
- continue;
- }
-
- rs = &runstats[td->groupid];
-
- if (td->runtime[0] < rs->min_run[0] || !rs->min_run[0])
- rs->min_run[0] = td->runtime[0];
- if (td->runtime[0] > rs->max_run[0])
- rs->max_run[0] = td->runtime[0];
- if (td->runtime[1] < rs->min_run[1] || !rs->min_run[1])
- rs->min_run[1] = td->runtime[1];
- if (td->runtime[1] > rs->max_run[1])
- rs->max_run[1] = td->runtime[1];
-
- rbw = wbw = 0;
- if (td->runtime[0])
- rbw = td->io_bytes[0] / td->runtime[0];
- if (td->runtime[1])
- wbw = td->io_bytes[1] / td->runtime[1];
-
- if (rbw < rs->min_bw[0])
- rs->min_bw[0] = rbw;
- if (wbw < rs->min_bw[1])
- rs->min_bw[1] = wbw;
- if (rbw > rs->max_bw[0])
- rs->max_bw[0] = rbw;
- 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;
- }
-
- 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];
- if (rs->max_run[1])
- rs->agg[1] = (rs->io_mb[1]*1024*1000) / rs->max_run[1];
- }
-
- /*
- * don't overwrite last signal output
- */
- printf("\n");
-
- for (i = 0; i < thread_number; i++) {
- td = &threads[i];
- rs = &runstats[td->groupid];
-
- show_thread_status(td, rs);
- }
-
- for (i = 0; i < groupid + 1; i++)
- show_group_stats(&runstats[i], i);
-
- show_disk_util();
+ fio_unpin_memory(mlocked_mem);
}
int main(int argc, char *argv[])