*
*/
#include <unistd.h>
-#include <fcntl.h>
#include <string.h>
-#include <limits.h>
#include <signal.h>
-#include <time.h>
-#include <locale.h>
#include <assert.h>
-#include <time.h>
#include <inttypes.h>
#include <sys/stat.h>
#include <sys/wait.h>
-#include <sys/ipc.h>
-#include <sys/mman.h>
#include <math.h>
#include "fio.h"
-#ifndef FIO_NO_HAVE_SHM_H
-#include <sys/shm.h>
-#endif
-#include "hash.h"
#include "smalloc.h"
#include "verify.h"
-#include "trim.h"
#include "diskutil.h"
#include "cgroup.h"
#include "profile.h"
#include "lib/mountcheck.h"
#include "rate-submit.h"
#include "helper_thread.h"
+#include "pshared.h"
-static struct fio_mutex *startup_mutex;
+static struct fio_sem *startup_sem;
static struct flist_head *cgroup_list;
static char *cgroup_mnt;
static int exit_value;
if (td->update_rusage) {
td->update_rusage = 0;
update_rusage_stat(td);
- fio_mutex_up(td->rusage_sem);
+ fio_sem_up(td->rusage_sem);
}
}
break;
} else if (io_u->ddir == DDIR_WRITE) {
io_u->ddir = DDIR_READ;
+ populate_verify_io_u(td, io_u);
break;
} else {
put_io_u(td, io_u);
*/
static long long usec_for_io(struct thread_data *td, enum fio_ddir ddir)
{
- uint64_t secs, remainder, bps, bytes, iops;
+ uint64_t bps = td->rate_bps[ddir];
assert(!(td->flags & TD_F_CHILD));
- bytes = td->rate_io_issue_bytes[ddir];
- bps = td->rate_bps[ddir];
if (td->o.rate_process == RATE_PROCESS_POISSON) {
- uint64_t val;
+ uint64_t val, iops;
+
iops = bps / td->o.bs[ddir];
val = (int64_t) (1000000 / iops) *
-logf(__rand_0_1(&td->poisson_state[ddir]));
td->last_usec[ddir] += val;
return td->last_usec[ddir];
} else if (bps) {
- secs = bytes / bps;
- remainder = bytes % bps;
+ uint64_t bytes = td->rate_io_issue_bytes[ddir];
+ uint64_t secs = bytes / bps;
+ uint64_t remainder = bytes % bps;
+
return remainder * 1000000 / bps + secs * 1000000;
}
return 0;
}
+static void handle_thinktime(struct thread_data *td, enum fio_ddir ddir)
+{
+ unsigned long long b;
+ uint64_t total;
+ int left;
+
+ b = ddir_rw_sum(td->io_blocks);
+ if (b % td->o.thinktime_blocks)
+ return;
+
+ io_u_quiesce(td);
+
+ total = 0;
+ if (td->o.thinktime_spin)
+ total = usec_spin(td->o.thinktime_spin);
+
+ left = td->o.thinktime - total;
+ if (left)
+ total += usec_sleep(td, left);
+
+ /*
+ * If we're ignoring thinktime for the rate, add the number of bytes
+ * we would have done while sleeping, minus one block to ensure we
+ * start issuing immediately after the sleep.
+ */
+ if (total && td->rate_bps[ddir] && td->o.rate_ign_think) {
+ uint64_t missed = (td->rate_bps[ddir] * total) / 1000000ULL;
+ uint64_t bs = td->o.min_bs[ddir];
+ uint64_t usperop = bs * 1000000ULL / td->rate_bps[ddir];
+ uint64_t over;
+
+ if (usperop <= total)
+ over = bs;
+ else
+ over = (usperop - total) / usperop * -bs;
+
+ td->rate_io_issue_bytes[ddir] += (missed - over);
+ }
+}
+
/*
* Main IO worker function. It retrieves io_u's to process and queues
* and reaps them, checking for rate and errors along the way.
int err = PTR_ERR(io_u);
io_u = NULL;
+ ddir = DDIR_INVAL;
if (err == -EBUSY) {
ret = FIO_Q_BUSY;
goto reap;
break;
}
+ if (io_u->ddir == DDIR_WRITE && td->flags & TD_F_DO_VERIFY)
+ populate_verify_io_u(td, io_u);
+
ddir = io_u->ddir;
/*
if (!in_ramp_time(td) && td->o.latency_target)
lat_target_check(td);
- if (td->o.thinktime) {
- unsigned long long b;
-
- b = ddir_rw_sum(td->io_blocks);
- if (!(b % td->o.thinktime_blocks)) {
- int left;
-
- io_u_quiesce(td);
-
- if (td->o.thinktime_spin)
- usec_spin(td->o.thinktime_spin);
-
- left = td->o.thinktime - td->o.thinktime_spin;
- if (left)
- usec_sleep(td, left);
- }
- }
+ if (ddir_rw(ddir) && td->o.thinktime)
+ handle_thinktime(td, ddir);
}
check_update_rusage(td);
static int switch_ioscheduler(struct thread_data *td)
{
#ifdef FIO_HAVE_IOSCHED_SWITCH
- char tmp[256], tmp2[128];
+ char tmp[256], tmp2[128], *p;
FILE *f;
int ret;
/*
* Read back and check that the selected scheduler is now the default.
*/
- memset(tmp, 0, sizeof(tmp));
- ret = fread(tmp, sizeof(tmp), 1, f);
+ ret = fread(tmp, 1, sizeof(tmp) - 1, f);
if (ferror(f) || ret < 0) {
td_verror(td, errno, "fread");
fclose(f);
return 1;
}
+ tmp[ret] = '\0';
/*
- * either a list of io schedulers or "none\n" is expected.
+ * either a list of io schedulers or "none\n" is expected. Strip the
+ * trailing newline.
*/
- tmp[strlen(tmp) - 1] = '\0';
+ p = tmp;
+ strsep(&p, "\n");
/*
* Write to "none" entry doesn't fail, so check the result here.
}
td_set_runstate(td, TD_INITIALIZED);
- dprint(FD_MUTEX, "up startup_mutex\n");
- fio_mutex_up(startup_mutex);
- dprint(FD_MUTEX, "wait on td->mutex\n");
- fio_mutex_down(td->mutex);
- dprint(FD_MUTEX, "done waiting on td->mutex\n");
+ dprint(FD_MUTEX, "up startup_sem\n");
+ fio_sem_up(startup_sem);
+ dprint(FD_MUTEX, "wait on td->sem\n");
+ fio_sem_down(td->sem);
+ dprint(FD_MUTEX, "done waiting on td->sem\n");
/*
* A new gid requires privilege, so we need to do this before setting
deadlock_loop_cnt = 0;
do {
check_update_rusage(td);
- if (!fio_mutex_down_trylock(stat_mutex))
+ if (!fio_sem_down_trylock(stat_sem))
break;
usleep(1000);
if (deadlock_loop_cnt++ > 5000) {
- log_err("fio seems to be stuck grabbing stat_mutex, forcibly exiting\n");
+ log_err("fio seems to be stuck grabbing stat_sem, forcibly exiting\n");
td->error = EDEADLK;
goto err;
}
if (td_trim(td) && td->io_bytes[DDIR_TRIM])
update_runtime(td, elapsed_us, DDIR_TRIM);
fio_gettime(&td->start, NULL);
- fio_mutex_up(stat_mutex);
+ fio_sem_up(stat_sem);
if (td->error || td->terminate)
break;
*/
check_update_rusage(td);
- fio_mutex_down(stat_mutex);
+ fio_sem_down(stat_sem);
update_runtime(td, elapsed_us, DDIR_READ);
fio_gettime(&td->start, NULL);
- fio_mutex_up(stat_mutex);
+ fio_sem_up(stat_sem);
if (td->error || td->terminate)
break;
init_disk_util(td);
- td->rusage_sem = fio_mutex_init(FIO_MUTEX_LOCKED);
+ td->rusage_sem = fio_sem_init(FIO_SEM_LOCKED);
td->update_rusage = 0;
/*
} else if (i == fio_debug_jobno)
*fio_debug_jobp = pid;
}
- dprint(FD_MUTEX, "wait on startup_mutex\n");
- if (fio_mutex_down_timeout(startup_mutex, 10000)) {
+ dprint(FD_MUTEX, "wait on startup_sem\n");
+ if (fio_sem_down_timeout(startup_sem, 10000)) {
log_err("fio: job startup hung? exiting.\n");
fio_terminate_threads(TERMINATE_ALL);
fio_abort = 1;
free(fd);
break;
}
- dprint(FD_MUTEX, "done waiting on startup_mutex\n");
+ dprint(FD_MUTEX, "done waiting on startup_sem\n");
}
/*
m_rate += ddir_rw_sum(td->o.ratemin);
t_rate += ddir_rw_sum(td->o.rate);
todo--;
- fio_mutex_up(td->mutex);
+ fio_sem_up(td->sem);
}
reap_threads(&nr_running, &t_rate, &m_rate);
setup_log(&agg_io_log[DDIR_TRIM], &p, "agg-trim_bw.log");
}
- startup_mutex = fio_mutex_init(FIO_MUTEX_LOCKED);
- if (startup_mutex == NULL)
+ startup_sem = fio_sem_init(FIO_SEM_LOCKED);
+ if (startup_sem == NULL)
return 1;
set_genesis_time();
stat_init();
- helper_thread_create(startup_mutex, sk_out);
+ helper_thread_create(startup_sem, sk_out);
cgroup_list = smalloc(sizeof(*cgroup_list));
- INIT_FLIST_HEAD(cgroup_list);
+ if (cgroup_list)
+ INIT_FLIST_HEAD(cgroup_list);
run_threads(sk_out);
steadystate_free(td);
fio_options_free(td);
if (td->rusage_sem) {
- fio_mutex_remove(td->rusage_sem);
+ fio_sem_remove(td->rusage_sem);
td->rusage_sem = NULL;
}
- fio_mutex_remove(td->mutex);
- td->mutex = NULL;
+ fio_sem_remove(td->sem);
+ td->sem = NULL;
}
free_disk_util();
- cgroup_kill(cgroup_list);
- sfree(cgroup_list);
+ if (cgroup_list) {
+ cgroup_kill(cgroup_list);
+ sfree(cgroup_list);
+ }
sfree(cgroup_mnt);
- fio_mutex_remove(startup_mutex);
+ fio_sem_remove(startup_sem);
stat_exit();
return exit_value;
}