#include "lib/getrusage.h"
#include "idletime.h"
#include "err.h"
-#include "lib/tp.h"
#include "workqueue.h"
#include "lib/mountcheck.h"
-
-static pthread_t helper_thread;
-static pthread_mutex_t helper_lock;
-pthread_cond_t helper_cond;
-int helper_do_stat = 0;
+#include "rate-submit.h"
+#include "helper_thread.h"
static struct fio_mutex *startup_mutex;
static struct flist_head *cgroup_list;
int shm_id = 0;
int temp_stall_ts;
unsigned long done_secs = 0;
-volatile int helper_exit = 0;
#define PAGE_ALIGN(buf) \
(char *) (((uintptr_t) (buf) + page_mask) & ~page_mask)
put_io_u(td, io_u);
return true;
} else if (ret == FIO_Q_QUEUED) {
+ if (td_io_commit(td))
+ return true;
if (io_u_queued_complete(td, 1) < 0)
return true;
} else if (ret == FIO_Q_COMPLETED) {
int min_evts = 0;
int ret;
+ if (td->flags & TD_F_REGROW_LOGS) {
+ ret = io_u_quiesce(td);
+ regrow_logs(td);
+ return ret;
+ }
+
/*
* if the queue is full, we MUST reap at least 1 event
*/
if (*ret < 0)
break;
}
+
+ if (td->flags & TD_F_REGROW_LOGS)
+ regrow_logs(td);
+
+ /*
+ * when doing I/O (not when verifying),
+ * check for any errors that are to be ignored
+ */
+ if (!from_verify)
+ break;
+
return 0;
case FIO_Q_QUEUED:
/*
*
* Returns number of bytes written and trimmed.
*/
-static uint64_t do_io(struct thread_data *td)
+static void do_io(struct thread_data *td, uint64_t *bytes_done)
{
unsigned int i;
int ret = 0;
uint64_t total_bytes, bytes_issued = 0;
+ for (i = 0; i < DDIR_RWDIR_CNT; i++)
+ bytes_done[i] = td->bytes_done[i];
+
if (in_ramp_time(td))
td_set_runstate(td, TD_RAMP);
else
if (flow_threshold_exceeded(td))
continue;
- if (!td->o.time_based && bytes_issued >= total_bytes)
+ /*
+ * Break if we exceeded the bytes. The exception is time
+ * based runs, but we still need to break out of the loop
+ * for those to run verification, if enabled.
+ */
+ if (bytes_issued >= total_bytes &&
+ (!td->o.time_based ||
+ (td->o.time_based && td->o.verify != VERIFY_NONE)))
break;
io_u = get_io_u(td);
if (td->error)
break;
- ret = workqueue_enqueue(&td->io_wq, &io_u->work);
- if (ret)
- ret = FIO_Q_QUEUED;
- else
- ret = FIO_Q_BUSY;
+ workqueue_enqueue(&td->io_wq, &io_u->work);
+ ret = FIO_Q_QUEUED;
- if (ret == FIO_Q_QUEUED && ddir_rw(ddir)) {
+ if (ddir_rw(ddir)) {
td->io_issues[ddir]++;
td->io_issue_bytes[ddir] += blen;
td->rate_io_issue_bytes[ddir] += blen;
if (!in_ramp_time(td) && should_check_rate(td)) {
if (check_min_rate(td, &comp_time)) {
- if (exitall_on_terminate)
+ if (exitall_on_terminate || td->o.exitall_error)
fio_terminate_threads(td->groupid);
td_verror(td, EIO, "check_min_rate");
break;
if (!ddir_rw_sum(td->this_io_bytes))
td->done = 1;
- return td->bytes_done[DDIR_WRITE] + td->bytes_done[DDIR_TRIM];
+ for (i = 0; i < DDIR_RWDIR_CNT; i++)
+ bytes_done[i] = td->bytes_done[i] - bytes_done[i];
+}
+
+static void free_file_completion_logging(struct thread_data *td)
+{
+ struct fio_file *f;
+ unsigned int i;
+
+ for_each_file(td, f, i) {
+ if (!f->last_write_comp)
+ break;
+ sfree(f->last_write_comp);
+ }
+}
+
+static int init_file_completion_logging(struct thread_data *td,
+ unsigned int depth)
+{
+ struct fio_file *f;
+ unsigned int i;
+
+ if (td->o.verify == VERIFY_NONE || !td->o.verify_state_save)
+ return 0;
+
+ for_each_file(td, f, i) {
+ f->last_write_comp = scalloc(depth, sizeof(uint64_t));
+ if (!f->last_write_comp)
+ goto cleanup;
+ }
+
+ return 0;
+
+cleanup:
+ free_file_completion_logging(td);
+ log_err("fio: failed to alloc write comp data\n");
+ return 1;
}
static void cleanup_io_u(struct thread_data *td)
io_u_qexit(&td->io_u_freelist);
io_u_qexit(&td->io_u_all);
- if (td->last_write_comp)
- sfree(td->last_write_comp);
+ free_file_completion_logging(td);
}
static int init_io_u(struct thread_data *td)
p += max_bs;
}
- if (td->o.verify != VERIFY_NONE) {
- td->last_write_comp = scalloc(max_units, sizeof(uint64_t));
- if (!td->last_write_comp) {
- log_err("fio: failed to alloc write comp data\n");
- return 1;
- }
- }
+ if (init_file_completion_logging(td, max_units))
+ return 1;
return 0;
}
if (diff < td_max_bs(td))
return false;
- if (fio_files_done(td))
+ if (fio_files_done(td) && !td->o.io_limit)
return false;
return true;
return td->bytes_done[DDIR_WRITE] + td->bytes_done[DDIR_TRIM];
}
-static void io_workqueue_fn(struct submit_worker *sw,
- struct workqueue_work *work)
-{
- struct io_u *io_u = container_of(work, struct io_u, work);
- const enum fio_ddir ddir = io_u->ddir;
- struct thread_data *td = sw->private;
- int ret;
-
- dprint(FD_RATE, "io_u %p queued by %u\n", io_u, gettid());
-
- io_u_set(io_u, IO_U_F_NO_FILE_PUT);
-
- td->cur_depth++;
-
- do {
- ret = td_io_queue(td, io_u);
- if (ret != FIO_Q_BUSY)
- break;
- ret = io_u_queued_complete(td, 1);
- if (ret > 0)
- td->cur_depth -= ret;
- io_u_clear(io_u, IO_U_F_FLIGHT);
- } while (1);
-
- dprint(FD_RATE, "io_u %p ret %d by %u\n", io_u, ret, gettid());
-
- io_queue_event(td, io_u, &ret, ddir, NULL, 0, NULL);
-
- if (ret == FIO_Q_COMPLETED)
- td->cur_depth--;
- else if (ret == FIO_Q_QUEUED) {
- unsigned int min_evts;
-
- if (td->o.iodepth == 1)
- min_evts = 1;
- else
- min_evts = 0;
-
- ret = io_u_queued_complete(td, min_evts);
- if (ret > 0)
- td->cur_depth -= ret;
- } else if (ret == FIO_Q_BUSY) {
- ret = io_u_queued_complete(td, td->cur_depth);
- if (ret > 0)
- td->cur_depth -= ret;
- }
-}
-
-static bool io_workqueue_pre_sleep_flush_fn(struct submit_worker *sw)
-{
- struct thread_data *td = sw->private;
-
- if (td->io_u_queued || td->cur_depth || td->io_u_in_flight)
- return true;
-
- return false;
-}
-
-static void io_workqueue_pre_sleep_fn(struct submit_worker *sw)
-{
- struct thread_data *td = sw->private;
- int ret;
-
- ret = io_u_quiesce(td);
- if (ret > 0)
- td->cur_depth -= ret;
-}
-
-static int io_workqueue_alloc_fn(struct submit_worker *sw)
-{
+struct fork_data {
struct thread_data *td;
-
- td = calloc(1, sizeof(*td));
- sw->private = td;
- return 0;
-}
-
-static void io_workqueue_free_fn(struct submit_worker *sw)
-{
- free(sw->private);
- sw->private = NULL;
-}
-
-static int io_workqueue_init_worker_fn(struct submit_worker *sw)
-{
- struct thread_data *parent = sw->wq->td;
- struct thread_data *td = sw->private;
- 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, 1);
-
- 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 io_workqueue_exit_worker_fn(struct submit_worker *sw,
- unsigned int *sum_cnt)
-{
- struct thread_data *td = sw->private;
-
- (*sum_cnt)++;
- sum_thread_stats(&sw->wq->td->ts, &td->ts, *sum_cnt == 1);
-
- fio_options_free(td);
- close_and_free_files(td);
- if (td->io_ops)
- close_ioengine(td);
- td_set_runstate(td, TD_EXITED);
-}
-
-#ifdef CONFIG_SFAA
-static void sum_val(uint64_t *dst, uint64_t *src)
-{
- if (*src) {
- __sync_fetch_and_add(dst, *src);
- *src = 0;
- }
-}
-#else
-static void sum_val(uint64_t *dst, uint64_t *src)
-{
- if (*src) {
- *dst += *src;
- *src = 0;
- }
-}
-#endif
-
-static void pthread_double_unlock(pthread_mutex_t *lock1,
- pthread_mutex_t *lock2)
-{
-#ifndef CONFIG_SFAA
- pthread_mutex_unlock(lock1);
- pthread_mutex_unlock(lock2);
-#endif
-}
-
-static void pthread_double_lock(pthread_mutex_t *lock1, pthread_mutex_t *lock2)
-{
-#ifndef CONFIG_SFAA
- if (lock1 < lock2) {
- pthread_mutex_lock(lock1);
- pthread_mutex_lock(lock2);
- } else {
- pthread_mutex_lock(lock2);
- pthread_mutex_lock(lock1);
- }
-#endif
-}
-
-static void sum_ddir(struct thread_data *dst, struct thread_data *src,
- enum fio_ddir ddir)
-{
- pthread_double_lock(&dst->io_wq.stat_lock, &src->io_wq.stat_lock);
-
- sum_val(&dst->io_bytes[ddir], &src->io_bytes[ddir]);
- sum_val(&dst->io_blocks[ddir], &src->io_blocks[ddir]);
- sum_val(&dst->this_io_blocks[ddir], &src->this_io_blocks[ddir]);
- sum_val(&dst->this_io_bytes[ddir], &src->this_io_bytes[ddir]);
- sum_val(&dst->bytes_done[ddir], &src->bytes_done[ddir]);
-
- pthread_double_unlock(&dst->io_wq.stat_lock, &src->io_wq.stat_lock);
-}
-
-static void io_workqueue_update_acct_fn(struct submit_worker *sw)
-{
- struct thread_data *src = sw->private;
- struct thread_data *dst = sw->wq->td;
-
- if (td_read(src))
- sum_ddir(dst, src, DDIR_READ);
- if (td_write(src))
- sum_ddir(dst, src, DDIR_WRITE);
- if (td_trim(src))
- sum_ddir(dst, src, DDIR_TRIM);
-
-}
-
-struct workqueue_ops rated_wq_ops = {
- .fn = io_workqueue_fn,
- .pre_sleep_flush_fn = io_workqueue_pre_sleep_flush_fn,
- .pre_sleep_fn = io_workqueue_pre_sleep_fn,
- .update_acct_fn = io_workqueue_update_acct_fn,
- .alloc_worker_fn = io_workqueue_alloc_fn,
- .free_worker_fn = io_workqueue_free_fn,
- .init_worker_fn = io_workqueue_init_worker_fn,
- .exit_worker_fn = io_workqueue_exit_worker_fn,
+ struct sk_out *sk_out;
};
/*
*/
static void *thread_main(void *data)
{
+ struct fork_data *fd = data;
unsigned long long elapsed_us[DDIR_RWDIR_CNT] = { 0, };
- struct thread_data *td = data;
+ struct thread_data *td = fd->td;
struct thread_options *o = &td->o;
+ struct sk_out *sk_out = fd->sk_out;
pthread_condattr_t attr;
int clear_state;
int ret;
+ sk_out_assign(sk_out);
+ free(fd);
+
if (!o->use_thread) {
setsid();
td->pid = getpid();
goto err;
}
+ /*
+ * Do this early, we don't want the compress threads to be limited
+ * to the same CPUs as the IO workers. So do this before we set
+ * any potential CPU affinity
+ */
+ if (iolog_compress_init(td, sk_out))
+ goto err;
+
/*
* If we have a gettimeofday() thread, make sure we exclude that
* thread from this job
goto err;
}
- if (td->flags & TD_F_COMPRESS_LOG)
- tp_init(&td->tp_data);
-
fio_verify_init(td);
- if ((o->io_submit_mode == IO_MODE_OFFLOAD) &&
- workqueue_init(td, &td->io_wq, &rated_wq_ops, td->o.iodepth))
+ if (rate_submit_init(td, sk_out))
goto err;
fio_gettime(&td->epoch, NULL);
if (td->o.verify_only && (td_write(td) || td_rw(td)))
verify_bytes = do_dry_run(td);
- else
- verify_bytes = do_io(td);
+ else {
+ uint64_t bytes_done[DDIR_RWDIR_CNT];
+
+ do_io(td, bytes_done);
+
+ if (!ddir_rw_sum(bytes_done)) {
+ fio_mark_td_terminate(td);
+ verify_bytes = 0;
+ } else {
+ verify_bytes = bytes_done[DDIR_WRITE] +
+ bytes_done[DDIR_TRIM];
+ }
+ }
clear_state = 1;
break;
}
+ td_set_runstate(td, TD_FINISHING);
+
update_rusage_stat(td);
td->ts.total_run_time = mtime_since_now(&td->epoch);
td->ts.io_bytes[DDIR_READ] = td->io_bytes[DDIR_READ];
fio_unpin_memory(td);
- fio_writeout_logs(td);
+ td_writeout_logs(td, true);
- if (o->io_submit_mode == IO_MODE_OFFLOAD)
- workqueue_exit(&td->io_wq);
-
- if (td->flags & TD_F_COMPRESS_LOG)
- tp_exit(&td->tp_data);
+ iolog_compress_exit(td);
+ rate_submit_exit(td);
if (o->exec_postrun)
exec_string(o, o->exec_postrun, (const char *)"postrun");
- if (exitall_on_terminate)
+ if (exitall_on_terminate || (o->exitall_error && td->error))
fio_terminate_threads(td->groupid);
err:
cgroup_shutdown(td, &cgroup_mnt);
verify_free_state(td);
+ if (td->zone_state_index) {
+ int i;
+
+ for (i = 0; i < DDIR_RWDIR_CNT; i++)
+ free(td->zone_state_index[i]);
+ free(td->zone_state_index);
+ td->zone_state_index = NULL;
+ }
+
if (fio_option_is_set(o, cpumask)) {
ret = fio_cpuset_exit(&o->cpumask);
if (ret)
*/
check_update_rusage(td);
+ sk_out_drop();
return (void *) (uintptr_t) td->error;
}
-
-/*
- * We cannot pass the td data into a forked process, so attach the td and
- * pass it to the thread worker.
- */
-static int fork_main(int shmid, int offset)
-{
- struct thread_data *td;
- void *data, *ret;
-
-#if !defined(__hpux) && !defined(CONFIG_NO_SHM)
- data = shmat(shmid, NULL, 0);
- if (data == (void *) -1) {
- int __err = errno;
-
- perror("shmat");
- return __err;
- }
-#else
- /*
- * HP-UX inherits shm mappings?
- */
- data = threads;
-#endif
-
- td = data + offset * sizeof(struct thread_data);
- ret = thread_main(td);
- shmdt(data);
- return (int) (uintptr_t) ret;
-}
-
static void dump_td_info(struct thread_data *td)
{
- log_err("fio: job '%s' hasn't exited in %lu seconds, it appears to "
- "be stuck. Doing forceful exit of this job.\n", td->o.name,
+ log_err("fio: job '%s' (state=%d) hasn't exited in %lu seconds, it "
+ "appears to be stuck. Doing forceful exit of this job.\n",
+ td->o.name, td->runstate,
(unsigned long) time_since_now(&td->terminate_time));
}
* move on.
*/
if (td->terminate &&
+ td->runstate < TD_FSYNCING &&
time_since_now(&td->terminate_time) >= FIO_REAP_TIMEOUT) {
dump_td_info(td);
td_set_runstate(td, TD_REAPED);
if (is_backend) {
void *data;
- int ver;
ret = fio_server_get_verify_state(td->o.name,
- td->thread_number - 1, &data, &ver);
+ td->thread_number - 1, &data);
if (!ret)
- verify_convert_assign_state(td, data, ver);
+ verify_assign_state(td, data);
} else
ret = verify_load_state(td, "local");
return true;
}
+static bool waitee_running(struct thread_data *me)
+{
+ const char *waitee = me->o.wait_for;
+ const char *self = me->o.name;
+ struct thread_data *td;
+ int i;
+
+ if (!waitee)
+ return false;
+
+ for_each_td(td, i) {
+ if (!strcmp(td->o.name, self) || strcmp(td->o.name, waitee))
+ continue;
+
+ if (td->runstate < TD_EXITED) {
+ dprint(FD_PROCESS, "%s fenced by %s(%s)\n",
+ self, td->o.name,
+ runstate_to_name(td->runstate));
+ return true;
+ }
+ }
+
+ dprint(FD_PROCESS, "%s: %s completed, can run\n", self, waitee);
+ return false;
+}
+
/*
* Main function for kicking off and reaping jobs, as needed.
*/
-static void run_threads(void)
+static void run_threads(struct sk_out *sk_out)
{
struct thread_data *td;
unsigned int i, todo, nr_running, m_rate, t_rate, nr_started;
struct thread_data *map[REAL_MAX_JOBS];
struct timeval this_start;
int this_jobs = 0, left;
+ struct fork_data *fd;
/*
* create threads (TD_NOT_CREATED -> TD_CREATED)
break;
}
+ if (waitee_running(td)) {
+ dprint(FD_PROCESS, "%s: waiting for %s\n",
+ td->o.name, td->o.wait_for);
+ continue;
+ }
+
init_disk_util(td);
td->rusage_sem = fio_mutex_init(FIO_MUTEX_LOCKED);
map[this_jobs++] = td;
nr_started++;
+ fd = calloc(1, sizeof(*fd));
+ fd->td = td;
+ fd->sk_out = sk_out;
+
if (td->o.use_thread) {
int ret;
dprint(FD_PROCESS, "will pthread_create\n");
ret = pthread_create(&td->thread, NULL,
- thread_main, td);
+ thread_main, fd);
if (ret) {
log_err("pthread_create: %s\n",
strerror(ret));
+ free(fd);
nr_started--;
break;
}
dprint(FD_PROCESS, "will fork\n");
pid = fork();
if (!pid) {
- int ret = fork_main(shm_id, i);
+ int ret;
+ ret = (int)(uintptr_t)thread_main(fd);
_exit(ret);
} 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, 10)) {
+ if (fio_mutex_down_timeout(startup_mutex, 10000)) {
log_err("fio: job startup hung? exiting.\n");
fio_terminate_threads(TERMINATE_ALL);
fio_abort = 1;
update_io_ticks();
}
-static void wait_for_helper_thread_exit(void)
-{
- void *ret;
-
- helper_exit = 1;
- pthread_cond_signal(&helper_cond);
- pthread_join(helper_thread, &ret);
-}
-
static void free_disk_util(void)
{
disk_util_prune_entries();
-
- pthread_cond_destroy(&helper_cond);
-}
-
-static void *helper_thread_main(void *data)
-{
- int ret = 0;
-
- fio_mutex_up(startup_mutex);
-
- while (!ret) {
- uint64_t sec = DISK_UTIL_MSEC / 1000;
- uint64_t nsec = (DISK_UTIL_MSEC % 1000) * 1000000;
- struct timespec ts;
- struct timeval tv;
-
- gettimeofday(&tv, NULL);
- ts.tv_sec = tv.tv_sec + sec;
- ts.tv_nsec = (tv.tv_usec * 1000) + nsec;
-
- if (ts.tv_nsec >= 1000000000ULL) {
- ts.tv_nsec -= 1000000000ULL;
- ts.tv_sec++;
- }
-
- pthread_cond_timedwait(&helper_cond, &helper_lock, &ts);
-
- ret = update_io_ticks();
-
- if (helper_do_stat) {
- helper_do_stat = 0;
- __show_running_run_stats();
- }
-
- if (!is_backend)
- print_thread_status();
- }
-
- return NULL;
-}
-
-static int create_helper_thread(void)
-{
- int ret;
-
- setup_disk_util();
-
- pthread_cond_init(&helper_cond, NULL);
- pthread_mutex_init(&helper_lock, NULL);
-
- ret = pthread_create(&helper_thread, NULL, helper_thread_main, NULL);
- if (ret) {
- log_err("Can't create helper thread: %s\n", strerror(ret));
- return 1;
- }
-
- dprint(FD_MUTEX, "wait on startup_mutex\n");
- fio_mutex_down(startup_mutex);
- dprint(FD_MUTEX, "done waiting on startup_mutex\n");
- return 0;
+ helper_thread_destroy();
}
-int fio_backend(void)
+int fio_backend(struct sk_out *sk_out)
{
struct thread_data *td;
int i;
set_genesis_time();
stat_init();
- create_helper_thread();
+ helper_thread_create(startup_mutex, sk_out);
cgroup_list = smalloc(sizeof(*cgroup_list));
INIT_FLIST_HEAD(cgroup_list);
- run_threads();
+ run_threads(sk_out);
- wait_for_helper_thread_exit();
+ helper_thread_exit();
if (!fio_abort) {
__show_run_stats();
projects / fio.git / blobdiff