#include "fio.h"
#include "hash.h"
+#include "smalloc.h"
unsigned long page_mask;
unsigned long page_size;
int nr_thread = 0;
int shm_id = 0;
int temp_stall_ts;
+unsigned long done_secs = 0;
static struct fio_mutex *startup_mutex;
static volatile int fio_abort;
static int exit_value;
+static struct itimerval itimer;
struct io_log *agg_io_log[2];
static inline void td_set_runstate(struct thread_data *td, int runstate)
{
- dprint(FD_PROCESS, "pid=%d: runstate %d -> %d\n", td->pid, td->runstate,
- runstate);
+ if (td->runstate == runstate)
+ return;
+
+ dprint(FD_PROCESS, "pid=%d: runstate %d -> %d\n", (int) td->pid,
+ td->runstate, runstate);
td->runstate = runstate;
}
struct thread_data *td;
int i;
+ dprint(FD_PROCESS, "terminate group_id=%d\n", group_id);
+
for_each_td(td, i) {
if (group_id == TERMINATE_ALL || groupid == td->groupid) {
- dprint(FD_PROCESS, "setting terminate on %d\n",td->pid);
-
+ dprint(FD_PROCESS, "setting terminate on %s/%d\n",
+ td->o.name, (int) td->pid);
td->terminate = 1;
td->o.start_delay = 0;
}
}
-static void sig_handler(int sig)
+static void status_timer_arm(void)
{
- switch (sig) {
- case SIGALRM:
- update_io_ticks();
- disk_util_timer_arm();
- print_thread_status();
- break;
- default:
- printf("\nfio: terminating on signal %d\n", sig);
- fflush(stdout);
- terminate_threads(TERMINATE_ALL);
- break;
+ itimer.it_value.tv_sec = 0;
+ itimer.it_value.tv_usec = DISK_UTIL_MSEC * 1000;
+ setitimer(ITIMER_REAL, &itimer, NULL);
+}
+
+static void sig_alrm(int sig)
+{
+ if (threads) {
+ update_io_ticks();
+ print_thread_status();
+ status_timer_arm();
+ }
+}
+
+static void sig_int(int sig)
+{
+ if (threads) {
+ printf("\nfio: terminating on signal %d\n", sig);
+ fflush(stdout);
+ terminate_threads(TERMINATE_ALL);
}
}
+static void sig_ill(int sig)
+{
+ if (!threads)
+ return;
+
+ log_err("fio: illegal instruction. your cpu does not support "
+ "the sse4.2 instruction for crc32c\n");
+ terminate_threads(TERMINATE_ALL);
+ exit(4);
+}
+
+static void set_sig_handlers(void)
+{
+ struct sigaction act;
+
+ memset(&act, 0, sizeof(act));
+ act.sa_handler = sig_alrm;
+ act.sa_flags = SA_RESTART;
+ sigaction(SIGALRM, &act, NULL);
+
+ memset(&act, 0, sizeof(act));
+ act.sa_handler = sig_int;
+ act.sa_flags = SA_RESTART;
+ sigaction(SIGINT, &act, NULL);
+
+ memset(&act, 0, sizeof(act));
+ act.sa_handler = sig_ill;
+ act.sa_flags = SA_RESTART;
+ sigaction(SIGILL, &act, NULL);
+}
+
/*
* Check if we are above the minimum rate given.
*/
* check bandwidth specified rate
*/
if (bytes < td->rate_bytes) {
- log_err("%s: min rate %u not met\n", td->o.name, td->o.ratemin);
+ log_err("%s: min rate %u not met\n", td->o.name,
+ td->o.ratemin);
return 1;
} else {
rate = (bytes - td->rate_bytes) / spent;
- if (rate < td->o.ratemin || bytes < td->rate_bytes) {
- log_err("%s: min rate %u not met, got %luKiB/sec\n", td->o.name, td->o.ratemin, rate);
+ if (rate < td->o.ratemin ||
+ bytes < td->rate_bytes) {
+ log_err("%s: min rate %u not met, got"
+ " %luKiB/sec\n", td->o.name,
+ td->o.ratemin, rate);
return 1;
}
}
* checks iops specified rate
*/
if (iops < td->o.rate_iops) {
- log_err("%s: min iops rate %u not met\n", td->o.name, td->o.rate_iops);
+ log_err("%s: min iops rate %u not met\n",
+ td->o.name, td->o.rate_iops);
return 1;
} else {
rate = (iops - td->rate_blocks) / spent;
- if (rate < td->o.rate_iops_min || iops < td->rate_blocks) {
- log_err("%s: min iops rate %u not met, got %lu\n", td->o.name, td->o.rate_iops_min, rate);
+ if (rate < td->o.rate_iops_min ||
+ iops < td->rate_blocks) {
+ log_err("%s: min iops rate %u not met,"
+ " got %lu\n", td->o.name,
+ td->o.rate_iops_min,
+ rate);
}
}
}
*/
static void cleanup_pending_aio(struct thread_data *td)
{
- struct list_head *entry, *n;
+ struct flist_head *entry, *n;
struct io_u *io_u;
int r;
* now cancel remaining active events
*/
if (td->io_ops->cancel) {
- list_for_each_safe(entry, n, &td->io_u_busylist) {
- io_u = list_entry(entry, struct io_u, list);
+ flist_for_each_safe(entry, n, &td->io_u_busylist) {
+ io_u = flist_entry(entry, struct io_u, list);
/*
* if the io_u isn't in flight, then that generally
io_u = NULL;
while (!td->terminate) {
- int ret2;
+ int ret2, full;
io_u = __get_io_u(td);
if (!io_u)
* zero read, fail
*/
if (!bytes) {
- td_verror(td, ENODATA, "full resid");
+ td_verror(td, EIO, "full resid");
put_io_u(td, io_u);
break;
}
* if we can queue more, do so. but check if there are
* completed io_u's first.
*/
- min_events = 0;
- if (queue_full(td) || ret == FIO_Q_BUSY) {
- min_events = 1;
+ full = queue_full(td) || ret == FIO_Q_BUSY;
+ if (full || !td->o.iodepth_batch_complete) {
+ min_events = td->o.iodepth_batch_complete;
+ if (full && !min_events)
+ min_events = 1;
- if (td->cur_depth > td->o.iodepth_low)
- min_events = td->cur_depth - td->o.iodepth_low;
+ do {
+ /*
+ * Reap required number of io units, if any,
+ * and do the verification on them through
+ * the callback handler
+ */
+ if (io_u_queued_complete(td, min_events) < 0) {
+ ret = -1;
+ break;
+ }
+ } while (full && (td->cur_depth > td->o.iodepth_low));
}
-
- /*
- * Reap required number of io units, if any, and do the
- * verification on them through the callback handler
- */
- if (io_u_queued_complete(td, min_events) < 0)
+ if (ret < 0)
break;
}
long bytes_done = 0;
int min_evts = 0;
struct io_u *io_u;
- int ret2;
+ int ret2, full;
if (td->terminate)
break;
* Add verification end_io handler, if asked to verify
* a previously written file.
*/
- if (td->o.verify != VERIFY_NONE)
+ if (td->o.verify != VERIFY_NONE && io_u->ddir == DDIR_READ) {
io_u->end_io = verify_io_u;
+ td_set_runstate(td, TD_VERIFYING);
+ } else
+ td_set_runstate(td, TD_RUNNING);
ret = td_io_queue(td, io_u);
switch (ret) {
* zero read, fail
*/
if (!bytes) {
- td_verror(td, ENODATA, "full resid");
+ td_verror(td, EIO, "full resid");
put_io_u(td, io_u);
break;
}
/*
* See if we need to complete some commands
*/
- if (ret == FIO_Q_QUEUED || ret == FIO_Q_BUSY) {
- min_evts = 0;
- if (queue_full(td) || ret == FIO_Q_BUSY) {
+ full = queue_full(td) || ret == FIO_Q_BUSY;
+ if (full || !td->o.iodepth_batch_complete) {
+ min_evts = td->o.iodepth_batch_complete;
+ if (full && !min_evts)
min_evts = 1;
- if (td->cur_depth > td->o.iodepth_low)
- min_evts = td->cur_depth - td->o.iodepth_low;
- }
-
fio_gettime(&comp_time, NULL);
- bytes_done = io_u_queued_complete(td, min_evts);
- if (bytes_done < 0)
- break;
+
+ do {
+ ret = io_u_queued_complete(td, min_evts);
+ if (ret <= 0)
+ break;
+
+ bytes_done += ret;
+ } while (full && (td->cur_depth > td->o.iodepth_low));
}
+ if (ret < 0)
+ break;
if (!bytes_done)
continue;
* of completions except the very first one which may look
* a little bursty
*/
- usec = utime_since(&s, &comp_time);
+ if (ramp_time_over(td)) {
+ usec = utime_since(&s, &comp_time);
- rate_throttle(td, usec, bytes_done);
+ rate_throttle(td, usec, bytes_done);
- if (check_min_rate(td, &comp_time)) {
- if (exitall_on_terminate)
- terminate_threads(td->groupid);
- td_verror(td, ENODATA, "check_min_rate");
- break;
+ if (check_min_rate(td, &comp_time)) {
+ if (exitall_on_terminate)
+ terminate_threads(td->groupid);
+ td_verror(td, EIO, "check_min_rate");
+ break;
+ }
}
if (td->o.thinktime) {
static void cleanup_io_u(struct thread_data *td)
{
- struct list_head *entry, *n;
+ struct flist_head *entry, *n;
struct io_u *io_u;
- list_for_each_safe(entry, n, &td->io_u_freelist) {
- io_u = list_entry(entry, struct io_u, list);
+ flist_for_each_safe(entry, n, &td->io_u_freelist) {
+ io_u = flist_entry(entry, struct io_u, list);
- list_del(&io_u->list);
+ flist_del(&io_u->list);
free(io_u);
}
free_io_mem(td);
}
-/*
- * "randomly" fill the buffer contents
- */
-static void fill_io_buf(struct thread_data *td, struct io_u *io_u, int max_bs)
-{
- long *ptr = io_u->buf;
-
- if (!td->o.zero_buffers) {
- while ((void *) ptr - io_u->buf < max_bs) {
- *ptr = rand() * GOLDEN_RATIO_PRIME;
- ptr++;
- }
- } else
- memset(ptr, 0, max_bs);
-}
-
static int init_io_u(struct thread_data *td)
{
struct io_u *io_u;
max_units = td->o.iodepth;
max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
- td->orig_buffer_size = (unsigned long long) max_bs * (unsigned long long) max_units;
+ td->orig_buffer_size = (unsigned long long) max_bs
+ * (unsigned long long) max_units;
- if (td->o.mem_type == MEM_SHMHUGE || td->o.mem_type == MEM_MMAPHUGE)
- td->orig_buffer_size = (td->orig_buffer_size + td->o.hugepage_size - 1) & ~(td->o.hugepage_size - 1);
+ if (td->o.mem_type == MEM_SHMHUGE || td->o.mem_type == MEM_MMAPHUGE) {
+ unsigned long bs;
+
+ bs = td->orig_buffer_size + td->o.hugepage_size - 1;
+ td->orig_buffer_size = bs & ~(td->o.hugepage_size - 1);
+ }
if (td->orig_buffer_size != (size_t) td->orig_buffer_size) {
log_err("fio: IO memory too large. Reduce max_bs or iodepth\n");
return 1;
io_u = malloc(sizeof(*io_u));
memset(io_u, 0, sizeof(*io_u));
- INIT_LIST_HEAD(&io_u->list);
+ INIT_FLIST_HEAD(&io_u->list);
if (!(td->io_ops->flags & FIO_NOIO)) {
io_u->buf = p + max_bs * i;
- if (td_write(td))
- fill_io_buf(td, io_u, max_bs);
+ if (td_write(td) && !td->o.refill_buffers)
+ io_u_fill_buffer(td, io_u, max_bs);
}
io_u->index = i;
io_u->flags = IO_U_F_FREE;
- list_add(&io_u->list, &td->io_u_freelist);
+ flist_add(&io_u->list, &td->io_u_freelist);
}
io_u_init_timeout();
f = fopen(tmp, "r+");
if (!f) {
if (errno == ENOENT) {
- log_err("fio: os or kernel doesn't support IO scheduler switching\n");
+ log_err("fio: os or kernel doesn't support IO scheduler"
+ " switching\n");
return 0;
}
td_verror(td, errno, "fopen iosched");
return 1;
}
- io_done = td->io_bytes[DDIR_READ] + td->io_bytes[DDIR_WRITE] + td->io_skip_bytes;
+ io_done = td->io_bytes[DDIR_READ] + td->io_bytes[DDIR_WRITE]
+ + td->io_skip_bytes;
if (io_done < td->o.size)
return 1;
td->pid = getpid();
- dprint(FD_PROCESS, "jobs pid=%d started\n", td->pid);
+ dprint(FD_PROCESS, "jobs pid=%d started\n", (int) td->pid);
- INIT_LIST_HEAD(&td->io_u_freelist);
- INIT_LIST_HEAD(&td->io_u_busylist);
- INIT_LIST_HEAD(&td->io_u_requeues);
- INIT_LIST_HEAD(&td->io_log_list);
- INIT_LIST_HEAD(&td->io_hist_list);
+ INIT_FLIST_HEAD(&td->io_u_freelist);
+ INIT_FLIST_HEAD(&td->io_u_busylist);
+ INIT_FLIST_HEAD(&td->io_u_requeues);
+ INIT_FLIST_HEAD(&td->io_log_list);
+ INIT_FLIST_HEAD(&td->io_hist_list);
td->io_hist_tree = RB_ROOT;
td_set_runstate(td, TD_INITIALIZED);
memcpy(&td->ts.stat_sample_time, &td->start, sizeof(td->start));
if (td->o.ratemin)
- memcpy(&td->lastrate, &td->ts.stat_sample_time, sizeof(td->lastrate));
+ memcpy(&td->lastrate, &td->ts.stat_sample_time,
+ sizeof(td->lastrate));
if (clear_state && clear_io_state(td))
break;
if (td_read(td) && td->io_bytes[DDIR_READ]) {
if (td->rw_end_set[DDIR_READ])
- elapsed = utime_since(&td->start, &td->rw_end[DDIR_READ]);
+ elapsed = utime_since(&td->start,
+ &td->rw_end[DDIR_READ]);
else
elapsed = utime_since_now(&td->start);
}
if (td_write(td) && td->io_bytes[DDIR_WRITE]) {
if (td->rw_end_set[DDIR_WRITE])
- elapsed = utime_since(&td->start, &td->rw_end[DDIR_WRITE]);
+ elapsed = utime_since(&td->start,
+ &td->rw_end[DDIR_WRITE]);
else
elapsed = utime_since_now(&td->start);
runtime[DDIR_WRITE] += elapsed;
}
-
+
if (td->error || td->terminate)
break;
err:
if (td->error)
- printf("fio: pid=%d, err=%d/%s\n", td->pid, td->error, td->verror);
+ printf("fio: pid=%d, err=%d/%s\n", (int) td->pid, td->error,
+ td->verror);
close_and_free_files(td);
close_ioengine(td);
cleanup_io_u(td);
ret = waitpid(td->pid, &status, flags);
if (ret < 0) {
if (errno == ECHILD) {
- log_err("fio: pid=%d disappeared %d\n", td->pid, td->runstate);
+ log_err("fio: pid=%d disappeared %d\n",
+ (int) td->pid, td->runstate);
td_set_runstate(td, TD_REAPED);
goto reaped;
}
int sig = WTERMSIG(status);
if (sig != SIGQUIT)
- log_err("fio: pid=%d, got signal=%d\n", td->pid, sig);
+ log_err("fio: pid=%d, got signal=%d\n",
+ (int) td->pid, sig);
td_set_runstate(td, TD_REAPED);
goto reaped;
}
pending++;
continue;
reaped:
- if (td->o.use_thread) {
- long ret;
-
- dprint(FD_PROCESS, "joining tread %d\n", td->pid);
- if (pthread_join(td->thread, (void *) &ret)) {
- dprint(FD_PROCESS, "join failed %ld\n", ret);
- perror("pthread_join");
- }
- }
-
(*nr_running)--;
(*m_rate) -= td->o.ratemin;
(*t_rate) -= td->o.rate;
- pending--;
+ if (!td->pid)
+ pending--;
if (td->error)
exit_value++;
+
+ done_secs += mtime_since_now(&td->epoch) / 1000;
}
if (*nr_running == cputhreads && !pending && realthreads)
if (!terse_output) {
printf("Starting ");
if (nr_thread)
- printf("%d thread%s", nr_thread, nr_thread > 1 ? "s" : "");
+ printf("%d thread%s", nr_thread,
+ nr_thread > 1 ? "s" : "");
if (nr_process) {
if (nr_thread)
printf(" and ");
- printf("%d process%s", nr_process, nr_process > 1 ? "es" : "");
+ printf("%d process%s", nr_process,
+ nr_process > 1 ? "es" : "");
}
printf("\n");
fflush(stdout);
}
- signal(SIGINT, sig_handler);
- signal(SIGALRM, sig_handler);
+ set_sig_handlers();
todo = thread_number;
nr_running = 0;
if (setup_files(td)) {
exit_value++;
if (td->error)
- log_err("fio: pid=%d, err=%d/%s\n", td->pid, td->error, td->verror);
+ log_err("fio: pid=%d, err=%d/%s\n",
+ (int) td->pid, td->error, td->verror);
td_set_runstate(td, TD_REAPED);
todo--;
+ } else {
+ struct fio_file *f;
+ unsigned int i;
+
+ /*
+ * for sharing to work, each job must always open
+ * its own files. so close them, if we opened them
+ * for creation
+ */
+ for_each_file(td, f, i)
+ td_io_close_file(td, f);
}
init_disk_util(td);
continue;
}
- if (td->o.stonewall && (nr_started || nr_running))
+ if (td->o.stonewall && (nr_started || nr_running)) {
+ dprint(FD_PROCESS, "%s: stonewall wait\n",
+ td->o.name);
break;
+ }
/*
* Set state to created. Thread will transition
if (td->o.use_thread) {
dprint(FD_PROCESS, "will pthread_create\n");
- if (pthread_create(&td->thread, NULL, thread_main, td)) {
- perror("thread_create");
+ if (pthread_create(&td->thread, NULL,
+ thread_main, td)) {
+ perror("pthread_create");
nr_started--;
break;
}
+ if (pthread_detach(td->thread) < 0)
+ perror("pthread_detach");
} else {
+ pid_t pid;
dprint(FD_PROCESS, "will fork\n");
- if (!fork()) {
+ pid = fork();
+ if (!pid) {
int ret = fork_main(shm_id, i);
- exit(ret);
- }
+ _exit(ret);
+ } else if (i == fio_debug_jobno)
+ *fio_debug_jobp = pid;
}
fio_mutex_down(startup_mutex);
}
{
long ps;
+ sinit();
+
/*
* We need locale for number printing, if it isn't set then just
* go with the US format.
set_genesis_time();
- disk_util_timer_arm();
+ status_timer_arm();
run_threads();
if (!fio_abort) {
show_run_stats();
if (write_bw_log) {
- __finish_log(agg_io_log[DDIR_READ],"agg-read_bw.log");
- __finish_log(agg_io_log[DDIR_WRITE],"agg-write_bw.log");
+ __finish_log(agg_io_log[DDIR_READ], "agg-read_bw.log");
+ __finish_log(agg_io_log[DDIR_WRITE],
+ "agg-write_bw.log");
}
}
projects / fio.git / blobdiff