static uint64_t alloc_reply(uint64_t tag, uint16_t opcode)
{
struct fio_net_cmd_reply *reply;
reply = calloc(1, sizeof(*reply));
INIT_FLIST_HEAD(&reply->list);
fio_gettime(&reply->tv, NULL);
reply->saved_tag = tag;
reply->opcode = opcode;
return (uintptr_t) reply;
}
static enum fio_ddir rate_ddir(struct thread_data *td, enum fio_ddir ddir)
{
enum fio_ddir odir = ddir ^ 1;
struct timeval t;
long usec;
assert(ddir_rw(ddir));
if (td->rate_pending_usleep[ddir] <= 0)
return ddir;
/*
* We have too much pending sleep 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
*/
if (td->rate_pending_usleep[odir] < 100000)
return odir;
/*
* Both directions have pending sleep. Sleep the minimum time
* and deduct from both.
*/
if (td->rate_pending_usleep[ddir] <=
td->rate_pending_usleep[odir]) {
usec = td->rate_pending_usleep[ddir];
} else {
usec = td->rate_pending_usleep[odir];
ddir = odir;
}
} else
usec = td->rate_pending_usleep[ddir];
io_u_quiesce(td);
fio_gettime(&t, NULL);
usec_sleep(td, usec);
usec = utime_since_now(&t);
td->rate_pending_usleep[ddir] -= usec;
odir = ddir ^ 1;
if (td_rw(td) && __should_check_rate(td, odir))
td->rate_pending_usleep[odir] -= usec;
if (ddir_trim(ddir))
return ddir;
return ddir;
}
void lat_target_init(struct thread_data *td)
{
if (td->o.latency_target) {
dprint(FD_RATE, "Latency target=%llu\n", td->o.latency_target);
fio_gettime(&td->latency_ts, NULL);
td->latency_qd = 1;
td->latency_qd_high = td->o.iodepth;
td->latency_qd_low = 1;
td->latency_ios = ddir_rw_sum(td->io_blocks);
} else
td->latency_qd = td->o.iodepth;
}
static void init_icd(struct thread_data *td, struct io_completion_data *icd,
int nr)
{
int ddir;
if (!td->o.disable_clat || !td->o.disable_bw)
fio_gettime(&icd->time, NULL);
icd->nr = nr;
icd->error = 0;
for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++)
icd->bytes_done[ddir] = 0;
}
static void init_icd(struct thread_data *td, struct io_completion_data *icd,
int nr)
{
int ddir;
if (!gtod_reduce(td))
fio_gettime(&icd->time, NULL);
icd->nr = nr;
icd->error = 0;
for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++)
icd->bytes_done[ddir] = 0;
}
int ramp_time_over(struct thread_data *td)
{
struct timeval tv;
if (!td->o.ramp_time || td->ramp_time_over)
return 1;
fio_gettime(&tv, NULL);
if (mtime_since(&td->epoch, &tv) >= td->o.ramp_time * 1000) {
td->ramp_time_over = 1;
reset_all_stats(td);
td_set_runstate(td, TD_RAMP);
return 1;
}
return 0;
}
static int init_submit_worker(struct submit_worker *sw)
{
struct thread_data *parent = sw->wq->td;
struct thread_data *td = &sw->td;
int fio_unused ret;
memcpy(&td->o, &parent->o, sizeof(td->o));
memcpy(&td->ts, &parent->ts, sizeof(td->ts));
td->o.uid = td->o.gid = -1U;
dup_files(td, parent);
td->eo = parent->eo;
fio_options_mem_dupe(td);
if (ioengine_load(td))
goto err;
if (td->o.odirect)
td->io_ops->flags |= FIO_RAWIO;
td->pid = gettid();
INIT_FLIST_HEAD(&td->io_log_list);
INIT_FLIST_HEAD(&td->io_hist_list);
INIT_FLIST_HEAD(&td->verify_list);
INIT_FLIST_HEAD(&td->trim_list);
INIT_FLIST_HEAD(&td->next_rand_list);
td->io_hist_tree = RB_ROOT;
td->o.iodepth = 1;
if (td_io_init(td))
goto err_io_init;
fio_gettime(&td->epoch, NULL);
fio_getrusage(&td->ru_start);
clear_io_state(td);
td_set_runstate(td, TD_RUNNING);
td->flags |= TD_F_CHILD;
td->parent = parent;
return 0;
err_io_init:
close_ioengine(td);
err:
return 1;
}
static void fio_guasi_queued(struct thread_data *td, struct io_u **io_us, int nr)
{
int i;
struct io_u *io_u;
struct timeval now;
if (!fio_fill_issue_time(td))
return;
io_u_mark_submit(td, nr);
fio_gettime(&now, NULL);
for (i = 0; i < nr; i++) {
io_u = io_us[i];
memcpy(&io_u->issue_time, &now, sizeof(now));
io_u_queued(td, io_u);
}
}
void fio_idle_prof_stop(void)
{
int i;
uint64_t runt;
struct timeval tp;
struct timespec ts;
struct idle_prof_thread *ipt;
if (ipc.opt == IDLE_PROF_OPT_NONE)
return;
if (ipc.opt == IDLE_PROF_OPT_CALI)
return;
ipc.status = IDLE_PROF_STATUS_PROF_STOP;
/* wait for all threads to exit from profiling */
for (i = 0; i < ipc.nr_cpus; i++) {
ipt = &ipc.ipts[i];
pthread_mutex_lock(&ipt->start_lock);
while ((ipt->state != TD_EXITED) &&
(ipt->state!=TD_NOT_CREATED)) {
fio_gettime(&tp, NULL);
ts.tv_sec = tp.tv_sec + 1;
ts.tv_nsec = tp.tv_usec * 1000;
/* timed wait in case a signal is not received */
pthread_cond_timedwait(&ipt->cond, &ipt->start_lock, &ts);
}
pthread_mutex_unlock(&ipt->start_lock);
/* calculate idleness */
if (ipc.cali_mean != 0.0) {
runt = utime_since(&ipt->tps, &ipt->tpe);
if (runt)
ipt->idleness = ipt->loops * ipc.cali_mean / runt;
else
ipt->idleness = 0.0;
} else
ipt->idleness = 0.0;
}
/*
* memory allocations are freed via explicit fio_idle_prof_cleanup
* after profiling stats are collected by apps.
*/
}
static uint64_t t_crc32(void)
{
struct timeval s;
uint64_t ret;
void *buf;
int i;
buf = malloc(CHUNK);
randomize_buf(buf, CHUNK, 0x8989);
fio_gettime(&s, NULL);
for (i = 0; i < NR_CHUNKS; i++)
fio_crc32(buf, CHUNK);
ret = utime_since_now(&s);
free(buf);
return ret;
}
void reset_all_stats(struct thread_data *td)
{
struct timeval tv;
int i;
reset_io_counters(td);
for (i = 0; i < DDIR_RWDIR_CNT; i++) {
td->io_bytes[i] = 0;
td->io_blocks[i] = 0;
td->io_issues[i] = 0;
td->ts.total_io_u[i] = 0;
td->ts.runtime[i] = 0;
td->rwmix_issues = 0;
}
fio_gettime(&tv, NULL);
memcpy(&td->epoch, &tv, sizeof(tv));
memcpy(&td->start, &tv, sizeof(tv));
}
static void fio_ioring_queued(struct thread_data *td, int start, int nr)
{
struct ioring_data *ld = td->io_ops_data;
struct timespec now;
if (!fio_fill_issue_time(td))
return;
fio_gettime(&now, NULL);
while (nr--) {
struct io_sq_ring *ring = &ld->sq_ring;
int index = ring->array[start & ld->sq_ring_mask];
struct io_u *io_u = ld->io_u_index[index];
memcpy(&io_u->issue_time, &now, sizeof(now));
io_u_queued(td, io_u);
start++;
}
}
static uint64_t t_sha512(void)
{
uint8_t sha[128];
struct fio_sha512_ctx ctx = { .buf = sha };
struct timeval s;
uint64_t ret;
void *buf;
int i;
fio_sha512_init(&ctx);
buf = malloc(CHUNK);
randomize_buf(buf, CHUNK, 0x8989);
fio_gettime(&s, NULL);
for (i = 0; i < NR_CHUNKS; i++)
fio_sha512_update(&ctx, buf, CHUNK);
ret = utime_since_now(&s);
free(buf);
return ret;
}
static uint64_t t_xxhash(void)
{
void *state;
struct timeval s;
uint64_t ret;
void *buf;
int i;
state = XXH32_init(0x8989);
buf = malloc(CHUNK);
randomize_buf(buf, CHUNK, 0x8989);
fio_gettime(&s, NULL);
for (i = 0; i < NR_CHUNKS; i++)
XXH32_update(state, buf, CHUNK);
XXH32_digest(state);
ret = utime_since_now(&s);
free(buf);
return ret;
}
static uint64_t t_md5(void)
{
uint32_t digest[4];
struct fio_md5_ctx ctx = { .hash = digest };
struct timeval s;
uint64_t ret;
void *buf;
int i;
fio_md5_init(&ctx);
buf = malloc(CHUNK);
randomize_buf(buf, CHUNK, 0x8989);
fio_gettime(&s, NULL);
for (i = 0; i < NR_CHUNKS; i++)
fio_md5_update(&ctx, buf, CHUNK);
ret = utime_since_now(&s);
free(buf);
return ret;
}
uint64_t usec_sleep(struct thread_data *td, unsigned long usec)
{
struct timespec req;
struct timeval tv;
uint64_t t = 0;
do {
unsigned long ts = usec;
if (usec < ns_granularity) {
t += usec_spin(usec);
break;
}
ts = usec - ns_granularity;
if (ts >= 1000000) {
req.tv_sec = ts / 1000000;
ts -= 1000000 * req.tv_sec;
} else
req.tv_sec = 0;
req.tv_nsec = ts * 1000;
fio_gettime(&tv, NULL);
if (nanosleep(&req, NULL) < 0)
break;
ts = utime_since_now(&tv);
t += ts;
if (ts >= usec)
break;
usec -= ts;
} while (!td->terminate);
return t;
}
static void fio_init time_init(void)
{
int i;
/*
* Check the granularity of the nanosleep function
*/
for (i = 0; i < 10; i++) {
struct timeval tv;
struct timespec ts;
unsigned long elapsed;
fio_gettime(&tv, NULL);
ts.tv_sec = 0;
ts.tv_nsec = 1000;
nanosleep(&ts, NULL);
elapsed = utime_since_now(&tv);
if (elapsed > ns_granularity)
ns_granularity = elapsed;
}
}
static void fio_rdmaio_queued(struct thread_data *td, struct io_u **io_us,
unsigned int nr)
{
struct rdmaio_data *rd = td->io_ops->data;
struct timeval now;
unsigned int i;
if (!fio_fill_issue_time(td))
return;
fio_gettime(&now, NULL);
for (i = 0; i < nr; i++) {
struct io_u *io_u = io_us[i];
/* queued -> flight */
rd->io_us_flight[rd->io_u_flight_nr] = io_u;
rd->io_u_flight_nr++;
memcpy(&io_u->issue_time, &now, sizeof(now));
io_u_queued(td, io_u);
}
}
void rate_throttle(struct thread_data *td, unsigned long time_spent,
unsigned int bytes)
{
unsigned long usec_cycle;
unsigned int bs;
if (!td->o.rate && !td->o.rate_iops)
return;
if (td_rw(td))
bs = td->o.rw_min_bs;
else if (td_read(td))
bs = td->o.min_bs[DDIR_READ];
else
bs = td->o.min_bs[DDIR_WRITE];
usec_cycle = td->rate_usec_cycle * (bytes / bs);
if (time_spent < usec_cycle) {
unsigned long s = usec_cycle - time_spent;
td->rate_pending_usleep += s;
if (td->rate_pending_usleep >= 100000) {
struct timeval t;
fio_gettime(&t, NULL);
usec_sleep(td, td->rate_pending_usleep);
td->rate_pending_usleep -= utime_since_now(&t);
}
} else {
long overtime = time_spent - usec_cycle;
td->rate_pending_usleep -= overtime;
}
}
/*
* Print status of the jobs we know about. This includes rate estimates,
* ETA, thread state, etc.
*/
int calc_thread_status(struct jobs_eta *je, int force)
{
struct thread_data *td;
int i;
unsigned long rate_time, disp_time, bw_avg_time, *eta_secs;
unsigned long long io_bytes[DDIR_RWDIR_CNT];
unsigned long long io_iops[DDIR_RWDIR_CNT];
struct timeval now;
static unsigned long long rate_io_bytes[DDIR_RWDIR_CNT];
static unsigned long long disp_io_bytes[DDIR_RWDIR_CNT];
static unsigned long long disp_io_iops[DDIR_RWDIR_CNT];
static struct timeval rate_prev_time, disp_prev_time;
if (!force) {
if (output_format != FIO_OUTPUT_NORMAL)
return 0;
if (temp_stall_ts || eta_print == FIO_ETA_NEVER)
return 0;
if (!isatty(STDOUT_FILENO) && (eta_print != FIO_ETA_ALWAYS))
return 0;
}
if (!ddir_rw_sum(rate_io_bytes))
fill_start_time(&rate_prev_time);
if (!ddir_rw_sum(disp_io_bytes))
fill_start_time(&disp_prev_time);
eta_secs = malloc(thread_number * sizeof(unsigned long));
memset(eta_secs, 0, thread_number * sizeof(unsigned long));
je->elapsed_sec = (mtime_since_genesis() + 999) / 1000;
io_bytes[DDIR_READ] = io_bytes[DDIR_WRITE] = io_bytes[DDIR_TRIM] = 0;
io_iops[DDIR_READ] = io_iops[DDIR_WRITE] = io_iops[DDIR_TRIM] = 0;
bw_avg_time = ULONG_MAX;
for_each_td(td, i) {
if (is_power_of_2(td->o.kb_base))
je->is_pow2 = 1;
if (td->o.bw_avg_time < bw_avg_time)
bw_avg_time = td->o.bw_avg_time;
if (td->runstate == TD_RUNNING || td->runstate == TD_VERIFYING
|| td->runstate == TD_FSYNCING
|| td->runstate == TD_PRE_READING) {
je->nr_running++;
if (td_read(td)) {
je->t_rate += td->o.rate[DDIR_READ];
je->t_iops += td->o.rate_iops[DDIR_READ];
je->m_rate += td->o.ratemin[DDIR_READ];
je->m_iops += td->o.rate_iops_min[DDIR_READ];
}
if (td_write(td)) {
je->t_rate += td->o.rate[DDIR_WRITE];
je->t_iops += td->o.rate_iops[DDIR_WRITE];
je->m_rate += td->o.ratemin[DDIR_WRITE];
je->m_iops += td->o.rate_iops_min[DDIR_WRITE];
}
if (td_trim(td)) {
je->t_rate += td->o.rate[DDIR_TRIM];
je->t_iops += td->o.rate_iops[DDIR_TRIM];
je->m_rate += td->o.ratemin[DDIR_TRIM];
je->m_iops += td->o.rate_iops_min[DDIR_TRIM];
}
je->files_open += td->nr_open_files;
} else if (td->runstate == TD_RAMP) {
je->nr_running++;
je->nr_ramp++;
} else if (td->runstate == TD_SETTING_UP)
je->nr_running++;
else if (td->runstate < TD_RUNNING)
je->nr_pending++;
if (je->elapsed_sec >= 3)
eta_secs[i] = thread_eta(td);
else
eta_secs[i] = INT_MAX;
check_str_update(td);
if (td->runstate > TD_RAMP) {
int ddir;
for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) {
io_bytes[ddir] += td->io_bytes[ddir];
io_iops[ddir] += td->io_blocks[ddir];
}
}
}
if (exitall_on_terminate)
je->eta_sec = INT_MAX;
else
je->eta_sec = 0;
for_each_td(td, i) {
if (exitall_on_terminate) {
if (eta_secs[i] < je->eta_sec)
je->eta_sec = eta_secs[i];
} else {
if (eta_secs[i] > je->eta_sec)
je->eta_sec = eta_secs[i];
}
}
free(eta_secs);
fio_gettime(&now, NULL);
rate_time = mtime_since(&rate_prev_time, &now);
if (write_bw_log && rate_time > bw_avg_time && !in_ramp_time(td)) {
calc_rate(rate_time, io_bytes, rate_io_bytes, je->rate);
memcpy(&rate_prev_time, &now, sizeof(now));
add_agg_sample(je->rate[DDIR_READ], DDIR_READ, 0);
add_agg_sample(je->rate[DDIR_WRITE], DDIR_WRITE, 0);
add_agg_sample(je->rate[DDIR_TRIM], DDIR_TRIM, 0);
}
disp_time = mtime_since(&disp_prev_time, &now);
/*
* Allow a little slack, the target is to print it every 1000 msecs
*/
if (!force && disp_time < 900)
return 0;
calc_rate(disp_time, io_bytes, disp_io_bytes, je->rate);
calc_iops(disp_time, io_iops, disp_io_iops, je->iops);
memcpy(&disp_prev_time, &now, sizeof(now));
if (!force && !je->nr_running && !je->nr_pending)
return 0;
je->nr_threads = thread_number;
memcpy(je->run_str, run_str, thread_number * sizeof(char));
return 1;
}
static void *idle_prof_thread_fn(void *data)
{
int retval;
unsigned long j, k;
struct idle_prof_thread *ipt = data;
/* wait for all threads are spawned */
pthread_mutex_lock(&ipt->init_lock);
/* exit if any other thread failed to start */
if (ipc.status == IDLE_PROF_STATUS_ABORT) {
pthread_mutex_unlock(&ipt->init_lock);
return NULL;
}
retval = set_cpu_affinity(ipt);
if (retval == -1) {
ipt->state = TD_EXITED;
pthread_mutex_unlock(&ipt->init_lock);
return NULL;
}
ipt->cali_time = calibrate_unit(ipt->data);
/* delay to set IDLE class till now for better calibration accuracy */
#if defined(CONFIG_SCHED_IDLE)
if ((retval = fio_set_sched_idle()))
log_err("fio: fio_set_sched_idle failed\n");
#else
retval = -1;
log_err("fio: fio_set_sched_idle not supported\n");
#endif
if (retval == -1) {
ipt->state = TD_EXITED;
pthread_mutex_unlock(&ipt->init_lock);
goto do_exit;
}
ipt->state = TD_INITIALIZED;
/* signal the main thread that calibration is done */
pthread_cond_signal(&ipt->cond);
pthread_mutex_unlock(&ipt->init_lock);
/* wait for other calibration to finish */
pthread_mutex_lock(&ipt->start_lock);
/* exit if other threads failed to initialize */
if (ipc.status == IDLE_PROF_STATUS_ABORT) {
pthread_mutex_unlock(&ipt->start_lock);
goto do_exit;
}
/* exit if we are doing calibration only */
if (ipc.status == IDLE_PROF_STATUS_CALI_STOP) {
pthread_mutex_unlock(&ipt->start_lock);
goto do_exit;
}
fio_gettime(&ipt->tps, NULL);
ipt->state = TD_RUNNING;
j = 0;
while (1) {
for (k = 0; k < page_size; k++) {
ipt->data[(k + j) % page_size] = k % 256;
if (ipc.status == IDLE_PROF_STATUS_PROF_STOP) {
fio_gettime(&ipt->tpe, NULL);
goto idle_prof_done;
}
}
j++;
}
idle_prof_done:
ipt->loops = j + (double) k / page_size;
ipt->state = TD_EXITED;
pthread_mutex_unlock(&ipt->start_lock);
do_exit:
free_cpu_affinity(ipt);
return NULL;
}
void fio_idle_prof_init(void)
{
int i, ret;
struct timeval tp;
struct timespec ts;
pthread_attr_t tattr;
struct idle_prof_thread *ipt;
ipc.nr_cpus = cpus_online();
ipc.status = IDLE_PROF_STATUS_OK;
if (ipc.opt == IDLE_PROF_OPT_NONE)
return;
if ((ret = pthread_attr_init(&tattr))) {
log_err("fio: pthread_attr_init %s\n", strerror(ret));
return;
}
if ((ret = pthread_attr_setscope(&tattr, PTHREAD_SCOPE_SYSTEM))) {
log_err("fio: pthread_attr_setscope %s\n", strerror(ret));
return;
}
ipc.ipts = malloc(ipc.nr_cpus * sizeof(struct idle_prof_thread));
if (!ipc.ipts) {
log_err("fio: malloc failed\n");
return;
}
ipc.buf = malloc(ipc.nr_cpus * page_size);
if (!ipc.buf) {
log_err("fio: malloc failed\n");
free(ipc.ipts);
return;
}
/*
* profiling aborts on any single thread failure since the
* result won't be accurate if any cpu is not used.
*/
for (i = 0; i < ipc.nr_cpus; i++) {
ipt = &ipc.ipts[i];
ipt->cpu = i;
ipt->state = TD_NOT_CREATED;
ipt->data = (unsigned char *)(ipc.buf + page_size * i);
if ((ret = pthread_mutex_init(&ipt->init_lock, NULL))) {
ipc.status = IDLE_PROF_STATUS_ABORT;
log_err("fio: pthread_mutex_init %s\n", strerror(ret));
break;
}
if ((ret = pthread_mutex_init(&ipt->start_lock, NULL))) {
ipc.status = IDLE_PROF_STATUS_ABORT;
log_err("fio: pthread_mutex_init %s\n", strerror(ret));
break;
}
if ((ret = pthread_cond_init(&ipt->cond, NULL))) {
ipc.status = IDLE_PROF_STATUS_ABORT;
log_err("fio: pthread_cond_init %s\n", strerror(ret));
break;
}
/* make sure all threads are spawned before they start */
pthread_mutex_lock(&ipt->init_lock);
/* make sure all threads finish init before profiling starts */
pthread_mutex_lock(&ipt->start_lock);
if ((ret = pthread_create(&ipt->thread, &tattr, idle_prof_thread_fn, ipt))) {
ipc.status = IDLE_PROF_STATUS_ABORT;
log_err("fio: pthread_create %s\n", strerror(ret));
break;
} else
ipt->state = TD_CREATED;
if ((ret = pthread_detach(ipt->thread))) {
/* log error and let the thread spin */
log_err("fio: pthread_detatch %s\n", strerror(ret));
}
}
/*
* let good threads continue so that they can exit
* if errors on other threads occurred previously.
*/
for (i = 0; i < ipc.nr_cpus; i++) {
ipt = &ipc.ipts[i];
pthread_mutex_unlock(&ipt->init_lock);
}
if (ipc.status == IDLE_PROF_STATUS_ABORT)
return;
/* wait for calibration to finish */
for (i = 0; i < ipc.nr_cpus; i++) {
ipt = &ipc.ipts[i];
pthread_mutex_lock(&ipt->init_lock);
while ((ipt->state != TD_EXITED) &&
(ipt->state!=TD_INITIALIZED)) {
fio_gettime(&tp, NULL);
ts.tv_sec = tp.tv_sec + 1;
ts.tv_nsec = tp.tv_usec * 1000;
pthread_cond_timedwait(&ipt->cond, &ipt->init_lock, &ts);
}
pthread_mutex_unlock(&ipt->init_lock);
/*
* any thread failed to initialize would abort other threads
* later after fio_idle_prof_start.
*/
if (ipt->state == TD_EXITED)
ipc.status = IDLE_PROF_STATUS_ABORT;
}
if (ipc.status != IDLE_PROF_STATUS_ABORT)
calibration_stats();
else
ipc.cali_mean = ipc.cali_stddev = 0.0;
if (ipc.opt == IDLE_PROF_OPT_CALI)
ipc.status = IDLE_PROF_STATUS_CALI_STOP;
}
void set_genesis_time(void)
{
fio_gettime(&genesis, NULL);
}
int main(int argc, char *argv[])
{
int r;
struct timeval start, end;
struct fio_lfsr *fl;
int verify = 0;
unsigned int spin = 0;
uint64_t seed = 0;
uint64_t numbers;
uint64_t v_size;
uint64_t i;
void *v = NULL, *v_start;
double total, mean;
/* Read arguments */
switch (argc) {
case 5: if (strncmp(argv[4], "verify", 7) == 0)
verify = 1;
case 4: spin = atoi(argv[3]);
case 3: seed = atol(argv[2]);
case 2: numbers = strtol(argv[1], NULL, 16);
break;
default: usage();
return 1;
}
/* Initialize LFSR */
fl = malloc(sizeof(struct fio_lfsr));
if (!fl) {
perror("malloc");
return 1;
}
r = lfsr_init(fl, numbers, seed, spin);
if (r) {
printf("Initialization failed.\n");
return r;
}
/* Print specs */
printf("LFSR specs\n");
printf("==========================\n");
printf("Size is %u\n", 64 - __builtin_clzl(fl->cached_bit));
printf("Max val is %lu\n", (unsigned long) fl->max_val);
printf("XOR-mask is 0x%lX\n", (unsigned long) fl->xormask);
printf("Seed is %lu\n", (unsigned long) fl->last_val);
printf("Spin is %u\n", fl->spin);
printf("Cycle length is %lu\n", (unsigned long) fl->cycle_length);
/* Create verification table */
if (verify) {
v_size = numbers * sizeof(uint8_t);
v = malloc(v_size);
memset(v, 0, v_size);
printf("\nVerification table is %lf KBs\n", (double)(v_size) / 1024);
}
v_start = v;
/*
* Iterate over a tight loop until we have produced all the requested
* numbers. Verifying the results should introduce some small yet not
* negligible overhead.
*/
fprintf(stderr, "\nTest initiated... ");
fio_gettime(&start, NULL);
while (!lfsr_next(fl, &i)) {
if (verify)
*(uint8_t *)(v + i) += 1;
}
fio_gettime(&end, NULL);
fprintf(stderr, "finished.\n");
/* Check if all expected numbers within range have been calculated */
r = 0;
if (verify) {
fprintf(stderr, "Verifying results... ");
for (i = 0; i < numbers; i++) {
if (*(uint8_t *)(v + i) != 1) {
fprintf(stderr, "failed (%lu = %d).\n",
(unsigned long) i,
*(uint8_t *)(v + i));
r = 1;
break;
}
}
if (!r)
fprintf(stderr, "OK!\n");
}
/* Calculate elapsed time and mean time per number */
total = utime_since(&start, &end);
mean = total / fl->num_vals;
printf("\nTime results ");
if (verify)
printf("(slower due to verification)");
printf("\n==============================\n");
printf("Elapsed: %lf s\n", total / pow(10,6));
printf("Mean: %lf us\n", mean);
free(v_start);
free(fl);
return r;
}
static enum fio_ddir rate_ddir(struct thread_data *td, enum fio_ddir ddir)
{
enum fio_ddir odir = ddir ^ 1;
struct timeval t;
long usec;
assert(ddir_rw(ddir));
if (td->rate_pending_usleep[ddir] <= 0)
return ddir;
/*
* We have too much pending sleep in this direction. See if we
* should switch.
*/
if (td_rw(td)) {
/*
* Other direction does not have too much pending, switch
*/
if (td->rate_pending_usleep[odir] < 100000)
return odir;
/*
* Both directions have pending sleep. Sleep the minimum time
* and deduct from both.
*/
if (td->rate_pending_usleep[ddir] <=
td->rate_pending_usleep[odir]) {
usec = td->rate_pending_usleep[ddir];
} else {
usec = td->rate_pending_usleep[odir];
ddir = odir;
}
} else
usec = td->rate_pending_usleep[ddir];
/*
* We are going to sleep, ensure that we flush anything pending as
* not to skew our latency numbers.
*
* Changed to only monitor 'in flight' requests here instead of the
* td->cur_depth, b/c td->cur_depth does not accurately represent
* io's that have been actually submitted to an async engine,
* and cur_depth is meaningless for sync engines.
*/
if (td->io_u_in_flight) {
int fio_unused ret;
ret = io_u_queued_complete(td, td->io_u_in_flight, NULL);
}
fio_gettime(&t, NULL);
usec_sleep(td, usec);
usec = utime_since_now(&t);
td->rate_pending_usleep[ddir] -= usec;
odir = ddir ^ 1;
if (td_rw(td) && __should_check_rate(td, odir))
td->rate_pending_usleep[odir] -= usec;
if (ddir_trim(ddir))
return ddir;
return ddir;
}
static void lat_new_cycle(struct thread_data *td)
{
fio_gettime(&td->latency_ts, NULL);
td->latency_ios = ddir_rw_sum(td->io_blocks);
td->latency_failed = 0;
}
/*
* Return an io_u to be processed. Gets a buflen and offset, sets direction,
* etc. The returned io_u is fully ready to be prepped and submitted.
*/
struct io_u *get_io_u(struct thread_data *td)
{
struct fio_file *f;
struct io_u *io_u;
int do_scramble = 0;
long ret = 0;
io_u = __get_io_u(td);
if (!io_u) {
dprint(FD_IO, "__get_io_u failed\n");
return NULL;
}
if (check_get_verify(td, io_u))
goto out;
if (check_get_trim(td, io_u))
goto out;
/*
* from a requeue, io_u already setup
*/
if (io_u->file)
goto out;
/*
* If using an iolog, grab next piece if any available.
*/
if (td->flags & TD_F_READ_IOLOG) {
if (read_iolog_get(td, io_u))
goto err_put;
} else if (set_io_u_file(td, io_u)) {
ret = -EBUSY;
dprint(FD_IO, "io_u %p, setting file failed\n", io_u);
goto err_put;
}
f = io_u->file;
if (!f) {
dprint(FD_IO, "io_u %p, setting file failed\n", io_u);
goto err_put;
}
assert(fio_file_open(f));
if (ddir_rw(io_u->ddir)) {
if (!io_u->buflen && !(td->io_ops->flags & FIO_NOIO)) {
dprint(FD_IO, "get_io_u: zero buflen on %p\n", io_u);
goto err_put;
}
f->last_start = io_u->offset;
f->last_pos = io_u->offset + io_u->buflen;
if (io_u->ddir == DDIR_WRITE) {
if (td->flags & TD_F_REFILL_BUFFERS) {
io_u_fill_buffer(td, io_u,
io_u->xfer_buflen, io_u->xfer_buflen);
} else if (td->flags & TD_F_SCRAMBLE_BUFFERS)
do_scramble = 1;
if (td->flags & TD_F_VER_NONE) {
populate_verify_io_u(td, io_u);
do_scramble = 0;
}
} else if (io_u->ddir == DDIR_READ) {
/*
* Reset the buf_filled parameters so next time if the
* buffer is used for writes it is refilled.
*/
io_u->buf_filled_len = 0;
}
}
/*
* Set io data pointers.
*/
io_u->xfer_buf = io_u->buf;
io_u->xfer_buflen = io_u->buflen;
out:
assert(io_u->file);
if (!td_io_prep(td, io_u)) {
if (!td->o.disable_slat)
fio_gettime(&io_u->start_time, NULL);
if (do_scramble)
small_content_scramble(io_u);
return io_u;
}
err_put:
dprint(FD_IO, "get_io_u failed\n");
put_io_u(td, io_u);
return ERR_PTR(ret);
}
/*
* Print status of the jobs we know about. This includes rate estimates,
* ETA, thread state, etc.
*/
void print_thread_status(void)
{
unsigned long elapsed = (mtime_since_genesis() + 999) / 1000;
int i, nr_ramp, nr_running, nr_pending, t_rate, m_rate;
int t_iops, m_iops, files_open;
struct thread_data *td;
char eta_str[128];
double perc = 0.0;
unsigned long long io_bytes[2], io_iops[2];
unsigned long rate_time, disp_time, bw_avg_time, *eta_secs, eta_sec;
struct timeval now;
static unsigned long long rate_io_bytes[2];
static unsigned long long disp_io_bytes[2];
static unsigned long long disp_io_iops[2];
static struct timeval rate_prev_time, disp_prev_time;
static unsigned int rate[2], iops[2];
static int linelen_last;
static int eta_good;
int i2p = 0;
if (temp_stall_ts || terse_output || eta_print == FIO_ETA_NEVER)
return;
if (!isatty(STDOUT_FILENO) && (eta_print != FIO_ETA_ALWAYS))
return;
if (!rate_io_bytes[0] && !rate_io_bytes[1])
fill_start_time(&rate_prev_time);
if (!disp_io_bytes[0] && !disp_io_bytes[1])
fill_start_time(&disp_prev_time);
eta_secs = malloc(thread_number * sizeof(unsigned long));
memset(eta_secs, 0, thread_number * sizeof(unsigned long));
io_bytes[0] = io_bytes[1] = 0;
io_iops[0] = io_iops[1] = 0;
nr_pending = nr_running = t_rate = m_rate = t_iops = m_iops = 0;
nr_ramp = 0;
bw_avg_time = ULONG_MAX;
files_open = 0;
for_each_td(td, i) {
if (td->o.bw_avg_time < bw_avg_time)
bw_avg_time = td->o.bw_avg_time;
if (td->runstate == TD_RUNNING || td->runstate == TD_VERIFYING
|| td->runstate == TD_FSYNCING
|| td->runstate == TD_PRE_READING) {
nr_running++;
t_rate += td->o.rate[0] + td->o.rate[1];
m_rate += td->o.ratemin[0] + td->o.ratemin[1];
t_iops += td->o.rate_iops[0] + td->o.rate_iops[1];
m_iops += td->o.rate_iops_min[0] + td->o.rate_iops_min[1];
files_open += td->nr_open_files;
} else if (td->runstate == TD_RAMP) {
nr_running++;
nr_ramp++;
} else if (td->runstate < TD_RUNNING)
nr_pending++;
if (elapsed >= 3)
eta_secs[i] = thread_eta(td);
else
eta_secs[i] = INT_MAX;
check_str_update(td);
if (td->runstate > TD_RAMP) {
io_bytes[0] += td->io_bytes[0];
io_bytes[1] += td->io_bytes[1];
io_iops[0] += td->io_blocks[0];
io_iops[1] += td->io_blocks[1];
}
}
if (exitall_on_terminate)
eta_sec = INT_MAX;
else
eta_sec = 0;
for_each_td(td, i) {
if (!i2p && is_power_of_2(td->o.kb_base))
i2p = 1;
if (exitall_on_terminate) {
if (eta_secs[i] < eta_sec)
eta_sec = eta_secs[i];
} else {
if (eta_secs[i] > eta_sec)
eta_sec = eta_secs[i];
}
}
free(eta_secs);
if (eta_sec != INT_MAX && elapsed) {
perc = (double) elapsed / (double) (elapsed + eta_sec);
eta_to_str(eta_str, eta_sec);
}
fio_gettime(&now, NULL);
rate_time = mtime_since(&rate_prev_time, &now);
if (write_bw_log && rate_time > bw_avg_time && !in_ramp_time(td)) {
calc_rate(rate_time, io_bytes, rate_io_bytes, rate);
memcpy(&rate_prev_time, &now, sizeof(now));
add_agg_sample(rate[DDIR_READ], DDIR_READ, 0);
add_agg_sample(rate[DDIR_WRITE], DDIR_WRITE, 0);
}
disp_time = mtime_since(&disp_prev_time, &now);
if (disp_time < 1000)
return;
calc_rate(disp_time, io_bytes, disp_io_bytes, rate);
calc_iops(disp_time, io_iops, disp_io_iops, iops);
memcpy(&disp_prev_time, &now, sizeof(now));
if (!nr_running && !nr_pending)
return;
printf("Jobs: %d (f=%d)", nr_running, files_open);
if (m_rate || t_rate) {
char *tr, *mr;
mr = num2str(m_rate, 4, 0, i2p);
tr = num2str(t_rate, 4, 0, i2p);
printf(", CR=%s/%s KB/s", tr, mr);
free(tr);
free(mr);
} else if (m_iops || t_iops)
printf(", CR=%d/%d IOPS", t_iops, m_iops);
if (eta_sec != INT_MAX && nr_running) {
char perc_str[32];
char *iops_str[2];
char *rate_str[2];
int l;
if ((!eta_sec && !eta_good) || nr_ramp == nr_running)
strcpy(perc_str, "-.-% done");
else {
eta_good = 1;
perc *= 100.0;
sprintf(perc_str, "%3.1f%% done", perc);
}
rate_str[0] = num2str(rate[0], 5, 10, i2p);
rate_str[1] = num2str(rate[1], 5, 10, i2p);
iops_str[0] = num2str(iops[0], 4, 1, 0);
iops_str[1] = num2str(iops[1], 4, 1, 0);
l = printf(": [%s] [%s] [%s/%s /s] [%s/%s iops] [eta %s]",
run_str, perc_str, rate_str[0], rate_str[1],
iops_str[0], iops_str[1], eta_str);
if (l >= 0 && l < linelen_last)
printf("%*s", linelen_last - l, "");
linelen_last = l;
free(rate_str[0]);
free(rate_str[1]);
free(iops_str[0]);
free(iops_str[1]);
}
printf("\r");
fflush(stdout);
}
