return 0;
}
-static int check_min_rate(struct thread_data *td, struct timeval *now,
- uint64_t *bytes_done)
+static int check_min_rate(struct thread_data *td, struct timeval *now)
{
int ret = 0;
- if (bytes_done[DDIR_READ])
+ if (td->bytes_done[DDIR_READ])
ret |= __check_min_rate(td, now, DDIR_READ);
- if (bytes_done[DDIR_WRITE])
+ if (td->bytes_done[DDIR_WRITE])
ret |= __check_min_rate(td, now, DDIR_WRITE);
- if (bytes_done[DDIR_TRIM])
+ if (td->bytes_done[DDIR_TRIM])
ret |= __check_min_rate(td, now, DDIR_TRIM);
return ret;
/*
* get immediately available events, if any
*/
- r = io_u_queued_complete(td, 0, NULL);
+ r = io_u_queued_complete(td, 0);
if (r < 0)
return;
}
if (td->cur_depth)
- r = io_u_queued_complete(td, td->cur_depth, NULL);
+ r = io_u_queued_complete(td, td->cur_depth);
}
/*
put_io_u(td, io_u);
return 1;
} else if (ret == FIO_Q_QUEUED) {
- if (io_u_queued_complete(td, 1, NULL) < 0)
+ if (io_u_queued_complete(td, 1) < 0)
return 1;
} else if (ret == FIO_Q_COMPLETED) {
if (io_u->error) {
return 1;
}
- if (io_u_sync_complete(td, io_u, NULL) < 0)
+ if (io_u_sync_complete(td, io_u) < 0)
return 1;
} else if (ret == FIO_Q_BUSY) {
if (td_io_commit(td))
}
}
+static int wait_for_completions(struct thread_data *td, struct timeval *time)
+{
+ const int full = queue_full(td);
+ int min_evts = 0;
+ int ret;
+
+ /*
+ * if the queue is full, we MUST reap at least 1 event
+ */
+ min_evts = min(td->o.iodepth_batch_complete, td->cur_depth);
+ if (full && !min_evts)
+ min_evts = 1;
+
+ if (time && (__should_check_rate(td, DDIR_READ) ||
+ __should_check_rate(td, DDIR_WRITE) ||
+ __should_check_rate(td, DDIR_TRIM)))
+ fio_gettime(time, NULL);
+
+ do {
+ ret = io_u_queued_complete(td, min_evts);
+ if (ret < 0)
+ break;
+ } while (full && (td->cur_depth > td->o.iodepth_low));
+
+ return ret;
+}
+
+int io_queue_event(struct thread_data *td, struct io_u *io_u, int *ret,
+ enum fio_ddir ddir, uint64_t *bytes_issued, int from_verify,
+ struct timeval *comp_time)
+{
+ int ret2;
+
+ switch (*ret) {
+ case FIO_Q_COMPLETED:
+ if (io_u->error) {
+ *ret = -io_u->error;
+ clear_io_u(td, io_u);
+ } else if (io_u->resid) {
+ int bytes = io_u->xfer_buflen - io_u->resid;
+ struct fio_file *f = io_u->file;
+
+ if (bytes_issued)
+ *bytes_issued += bytes;
+
+ if (!from_verify)
+ trim_io_piece(td, io_u);
+
+ /*
+ * zero read, fail
+ */
+ if (!bytes) {
+ if (!from_verify)
+ unlog_io_piece(td, io_u);
+ td_verror(td, EIO, "full resid");
+ put_io_u(td, io_u);
+ break;
+ }
+
+ io_u->xfer_buflen = io_u->resid;
+ io_u->xfer_buf += bytes;
+ io_u->offset += bytes;
+
+ if (ddir_rw(io_u->ddir))
+ td->ts.short_io_u[io_u->ddir]++;
+
+ f = io_u->file;
+ if (io_u->offset == f->real_file_size)
+ goto sync_done;
+
+ requeue_io_u(td, &io_u);
+ } else {
+sync_done:
+ if (comp_time && (__should_check_rate(td, DDIR_READ) ||
+ __should_check_rate(td, DDIR_WRITE) ||
+ __should_check_rate(td, DDIR_TRIM)))
+ fio_gettime(comp_time, NULL);
+
+ *ret = io_u_sync_complete(td, io_u);
+ if (*ret < 0)
+ break;
+ }
+ return 0;
+ case FIO_Q_QUEUED:
+ /*
+ * if the engine doesn't have a commit hook,
+ * the io_u is really queued. if it does have such
+ * a hook, it has to call io_u_queued() itself.
+ */
+ if (td->io_ops->commit == NULL)
+ io_u_queued(td, io_u);
+ if (bytes_issued)
+ *bytes_issued += io_u->xfer_buflen;
+ break;
+ case FIO_Q_BUSY:
+ if (!from_verify)
+ unlog_io_piece(td, io_u);
+ requeue_io_u(td, &io_u);
+ ret2 = td_io_commit(td);
+ if (ret2 < 0)
+ *ret = ret2;
+ break;
+ default:
+ assert(ret < 0);
+ td_verror(td, -(*ret), "td_io_queue");
+ break;
+ }
+
+ if (break_on_this_error(td, ddir, ret))
+ return 1;
+
+ return 0;
+}
+
/*
* The main verify engine. Runs over the writes we previously submitted,
* reads the blocks back in, and checks the crc/md5 of the data.
*/
static void do_verify(struct thread_data *td, uint64_t verify_bytes)
{
- uint64_t bytes_done[DDIR_RWDIR_CNT] = { 0, 0, 0 };
struct fio_file *f;
struct io_u *io_u;
int ret, min_events;
io_u = NULL;
while (!td->terminate) {
enum fio_ddir ddir;
- int ret2, full;
+ int full;
update_tv_cache(td);
check_update_rusage(td);
break;
}
} else {
- if (ddir_rw_sum(bytes_done) + td->o.rw_min_bs > verify_bytes)
+ if (ddir_rw_sum(td->bytes_done) + td->o.rw_min_bs > verify_bytes)
break;
while ((io_u = get_io_u(td)) != NULL) {
io_u->end_io = verify_io_u;
ddir = io_u->ddir;
+ if (!td->o.disable_slat)
+ fio_gettime(&io_u->start_time, NULL);
ret = td_io_queue(td, io_u);
- switch (ret) {
- case FIO_Q_COMPLETED:
- if (io_u->error) {
- ret = -io_u->error;
- clear_io_u(td, io_u);
- } else if (io_u->resid) {
- int bytes = io_u->xfer_buflen - io_u->resid;
-
- /*
- * zero read, fail
- */
- if (!bytes) {
- td_verror(td, EIO, "full resid");
- put_io_u(td, io_u);
- break;
- }
-
- io_u->xfer_buflen = io_u->resid;
- io_u->xfer_buf += bytes;
- io_u->offset += bytes;
- if (ddir_rw(io_u->ddir))
- td->ts.short_io_u[io_u->ddir]++;
-
- f = io_u->file;
- if (io_u->offset == f->real_file_size)
- goto sync_done;
-
- requeue_io_u(td, &io_u);
- } else {
-sync_done:
- ret = io_u_sync_complete(td, io_u, bytes_done);
- if (ret < 0)
- break;
- }
- continue;
- case FIO_Q_QUEUED:
- break;
- case FIO_Q_BUSY:
- requeue_io_u(td, &io_u);
- ret2 = td_io_commit(td);
- if (ret2 < 0)
- ret = ret2;
- break;
- default:
- assert(ret < 0);
- td_verror(td, -ret, "td_io_queue");
- break;
- }
-
- if (break_on_this_error(td, ddir, &ret))
+ if (io_queue_event(td, io_u, &ret, ddir, NULL, 1, NULL))
break;
/*
*/
reap:
full = queue_full(td) || (ret == FIO_Q_BUSY && td->cur_depth);
- if (full || !td->o.iodepth_batch_complete) {
- min_events = min(td->o.iodepth_batch_complete,
- td->cur_depth);
- /*
- * if the queue is full, we MUST reap at least 1 event
- */
- if (full && !min_events)
- min_events = 1;
+ if (full || !td->o.iodepth_batch_complete)
+ ret = wait_for_completions(td, NULL);
- 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, bytes_done) < 0) {
- ret = -1;
- break;
- }
- } while (full && (td->cur_depth > td->o.iodepth_low));
- }
if (ret < 0)
break;
}
min_events = td->cur_depth;
if (min_events)
- ret = io_u_queued_complete(td, min_events, NULL);
+ ret = io_u_queued_complete(td, min_events);
} else
cleanup_pending_aio(td);
if (!td->o.number_ios)
return 0;
- number_ios = ddir_rw_sum(td->this_io_blocks);
+ number_ios = ddir_rw_sum(td->io_blocks);
number_ios += td->io_u_queued + td->io_u_in_flight;
- return number_ios >= td->o.number_ios;
+ return number_ios >= (td->o.number_ios * td->loops);
}
-static int io_bytes_exceeded(struct thread_data *td)
+static int io_issue_bytes_exceeded(struct thread_data *td)
+{
+ unsigned long long bytes, limit;
+
+ if (td_rw(td))
+ bytes = td->io_issue_bytes[DDIR_READ] + td->io_issue_bytes[DDIR_WRITE];
+ else if (td_write(td))
+ bytes = td->io_issue_bytes[DDIR_WRITE];
+ else if (td_read(td))
+ bytes = td->io_issue_bytes[DDIR_READ];
+ else
+ bytes = td->io_issue_bytes[DDIR_TRIM];
+
+ if (td->o.io_limit)
+ limit = td->o.io_limit;
+ else
+ limit = td->o.size;
+
+ limit *= td->loops;
+ return bytes >= limit || exceeds_number_ios(td);
+}
+
+static int io_complete_bytes_exceeded(struct thread_data *td)
{
unsigned long long bytes, limit;
else
limit = td->o.size;
+ limit *= td->loops;
return bytes >= limit || exceeds_number_ios(td);
}
*/
static uint64_t do_io(struct thread_data *td)
{
- uint64_t bytes_done[DDIR_RWDIR_CNT] = { 0, 0, 0 };
unsigned int i;
int ret = 0;
uint64_t total_bytes, bytes_issued = 0;
lat_target_init(td);
+ total_bytes = td->o.size;
+ /*
+ * Allow random overwrite workloads to write up to io_limit
+ * before starting verification phase as 'size' doesn't apply.
+ */
+ if (td_write(td) && td_random(td) && td->o.norandommap)
+ total_bytes = max(total_bytes, (uint64_t) td->o.io_limit);
/*
* If verify_backlog is enabled, we'll run the verify in this
* handler as well. For that case, we may need up to twice the
* amount of bytes.
*/
- total_bytes = td->o.size;
if (td->o.verify != VERIFY_NONE &&
(td_write(td) && td->o.verify_backlog))
total_bytes += td->o.size;
+ /* In trimwrite mode, each byte is trimmed and then written, so
+ * allow total_bytes to be twice as big */
+ if (td_trimwrite(td))
+ total_bytes += td->total_io_size;
+
while ((td->o.read_iolog_file && !flist_empty(&td->io_log_list)) ||
- (!flist_empty(&td->trim_list)) || !io_bytes_exceeded(td) ||
+ (!flist_empty(&td->trim_list)) || !io_issue_bytes_exceeded(td) ||
td->o.time_based) {
struct timeval comp_time;
- int min_evts = 0;
struct io_u *io_u;
- int ret2, full;
+ int full;
enum fio_ddir ddir;
check_update_rusage(td);
log_io_piece(td, io_u);
ret = td_io_queue(td, io_u);
- switch (ret) {
- case FIO_Q_COMPLETED:
- if (io_u->error) {
- ret = -io_u->error;
- unlog_io_piece(td, io_u);
- clear_io_u(td, io_u);
- } else if (io_u->resid) {
- int bytes = io_u->xfer_buflen - io_u->resid;
- struct fio_file *f = io_u->file;
-
- bytes_issued += bytes;
-
- trim_io_piece(td, io_u);
-
- /*
- * zero read, fail
- */
- if (!bytes) {
- unlog_io_piece(td, io_u);
- td_verror(td, EIO, "full resid");
- put_io_u(td, io_u);
- break;
- }
-
- io_u->xfer_buflen = io_u->resid;
- io_u->xfer_buf += bytes;
- io_u->offset += bytes;
-
- if (ddir_rw(io_u->ddir))
- td->ts.short_io_u[io_u->ddir]++;
-
- if (io_u->offset == f->real_file_size)
- goto sync_done;
-
- requeue_io_u(td, &io_u);
- } else {
-sync_done:
- if (__should_check_rate(td, DDIR_READ) ||
- __should_check_rate(td, DDIR_WRITE) ||
- __should_check_rate(td, DDIR_TRIM))
- fio_gettime(&comp_time, NULL);
- ret = io_u_sync_complete(td, io_u, bytes_done);
- if (ret < 0)
- break;
- bytes_issued += io_u->xfer_buflen;
- }
- break;
- case FIO_Q_QUEUED:
- /*
- * if the engine doesn't have a commit hook,
- * the io_u is really queued. if it does have such
- * a hook, it has to call io_u_queued() itself.
- */
- if (td->io_ops->commit == NULL)
- io_u_queued(td, io_u);
- bytes_issued += io_u->xfer_buflen;
- break;
- case FIO_Q_BUSY:
- unlog_io_piece(td, io_u);
- requeue_io_u(td, &io_u);
- ret2 = td_io_commit(td);
- if (ret2 < 0)
- ret = ret2;
- break;
- default:
- assert(ret < 0);
- put_io_u(td, io_u);
- break;
- }
-
- if (break_on_this_error(td, ddir, &ret))
+ if (io_queue_event(td, io_u, &ret, ddir, &bytes_issued, 1, &comp_time))
break;
/*
* resource starved.
*/
reap:
- full = queue_full(td) || (ret == FIO_Q_BUSY && td->cur_depth);
- if (full || !td->o.iodepth_batch_complete) {
- min_evts = min(td->o.iodepth_batch_complete,
- td->cur_depth);
- /*
- * if the queue is full, we MUST reap at least 1 event
- */
- if (full && !min_evts)
- min_evts = 1;
-
- if (__should_check_rate(td, DDIR_READ) ||
- __should_check_rate(td, DDIR_WRITE) ||
- __should_check_rate(td, DDIR_TRIM))
- fio_gettime(&comp_time, NULL);
-
- do {
- ret = io_u_queued_complete(td, min_evts, bytes_done);
- if (ret < 0)
- break;
-
- } while (full && (td->cur_depth > td->o.iodepth_low));
- }
-
+ full = queue_full(td) ||
+ (ret == FIO_Q_BUSY && td->cur_depth);
+ if (full || !td->o.iodepth_batch_complete)
+ ret = wait_for_completions(td, &comp_time);
if (ret < 0)
break;
- if (!ddir_rw_sum(bytes_done) && !(td->io_ops->flags & FIO_NOIO))
+ if (!ddir_rw_sum(td->bytes_done) &&
+ !(td->io_ops->flags & FIO_NOIO))
continue;
- if (!in_ramp_time(td) && should_check_rate(td, bytes_done)) {
- if (check_min_rate(td, &comp_time, bytes_done)) {
+ if (!in_ramp_time(td) && should_check_rate(td)) {
+ if (check_min_rate(td, &comp_time)) {
if (exitall_on_terminate)
fio_terminate_threads(td->groupid);
td_verror(td, EIO, "check_min_rate");
i = td->cur_depth;
if (i) {
- ret = io_u_queued_complete(td, i, bytes_done);
+ ret = io_u_queued_complete(td, i);
if (td->o.fill_device && td->error == ENOSPC)
td->error = 0;
}
if (!ddir_rw_sum(td->this_io_bytes))
td->done = 1;
- return bytes_done[DDIR_WRITE] + bytes_done[DDIR_TRIM];
+ return td->bytes_done[DDIR_WRITE] + td->bytes_done[DDIR_TRIM];
}
static void cleanup_io_u(struct thread_data *td)
*/
static uint64_t do_dry_run(struct thread_data *td)
{
- uint64_t bytes_done[DDIR_RWDIR_CNT] = { 0, 0, 0 };
-
td_set_runstate(td, TD_RUNNING);
while ((td->o.read_iolog_file && !flist_empty(&td->io_log_list)) ||
- (!flist_empty(&td->trim_list)) || !io_bytes_exceeded(td)) {
+ (!flist_empty(&td->trim_list)) || !io_complete_bytes_exceeded(td)) {
struct io_u *io_u;
int ret;
!td->o.experimental_verify)
log_io_piece(td, io_u);
- ret = io_u_sync_complete(td, io_u, bytes_done);
+ ret = io_u_sync_complete(td, io_u);
(void) ret;
}
- return bytes_done[DDIR_WRITE] + bytes_done[DDIR_TRIM];
+ return td->bytes_done[DDIR_WRITE] + td->bytes_done[DDIR_TRIM];
}
/*
* Set affinity first, in case it has an impact on the memory
* allocations.
*/
- if (o->cpumask_set) {
+ if (fio_option_is_set(o, cpumask)) {
if (o->cpus_allowed_policy == FIO_CPUS_SPLIT) {
ret = fio_cpus_split(&o->cpumask, td->thread_number - 1);
if (!ret) {
#ifdef CONFIG_LIBNUMA
/* numa node setup */
- if (o->numa_cpumask_set || o->numa_memmask_set) {
+ if (fio_option_is_set(o, numa_cpunodes) ||
+ fio_option_is_set(o, numa_memnodes)) {
struct bitmask *mask;
if (numa_available() < 0) {
goto err;
}
- if (o->numa_cpumask_set) {
+ if (fio_option_is_set(o, numa_cpunodes)) {
mask = numa_parse_nodestring(o->numa_cpunodes);
ret = numa_run_on_node_mask(mask);
numa_free_nodemask(mask);
}
}
- if (o->numa_memmask_set) {
-
+ if (fio_option_is_set(o, numa_memnodes)) {
mask = NULL;
if (o->numa_memnodes)
mask = numa_parse_nodestring(o->numa_memnodes);
if (o->verify_async && verify_async_init(td))
goto err;
- if (o->ioprio) {
+ if (fio_option_is_set(o, ioprio) ||
+ fio_option_is_set(o, ioprio_class)) {
ret = ioprio_set(IOPRIO_WHO_PROCESS, 0, o->ioprio_class, o->ioprio);
if (ret == -1) {
td_verror(td, errno, "ioprio_set");
clear_state = 1;
+ /*
+ * Make sure we've successfully updated the rusage stats
+ * before waiting on the stat mutex. Otherwise we could have
+ * the stat thread holding stat mutex and waiting for
+ * the rusage_sem, which would never get upped because
+ * this thread is waiting for the stat mutex.
+ */
+ check_update_rusage(td);
+
fio_mutex_down(stat_mutex);
if (td_read(td) && td->io_bytes[DDIR_READ]) {
elapsed = mtime_since_now(&td->start);
do_verify(td, verify_bytes);
+ /*
+ * See comment further up for why this is done here.
+ */
+ check_update_rusage(td);
+
fio_mutex_down(stat_mutex);
td->ts.runtime[DDIR_READ] += mtime_since_now(&td->start);
fio_gettime(&td->start, NULL);
cgroup_shutdown(td, &cgroup_mnt);
verify_free_state(td);
- if (o->cpumask_set) {
+ if (fio_option_is_set(o, cpumask)) {
ret = fio_cpuset_exit(&o->cpumask);
if (ret)
td_verror(td, ret, "fio_cpuset_exit");
void check_trigger_file(void)
{
if (__check_trigger_file() || trigger_timedout()) {
- if (nr_clients) {
- if (trigger_remote_cmd)
- fio_clients_send_trigger(trigger_remote_cmd);
- if (trigger_cmd)
- exec_trigger(trigger_cmd);
- } else {
+ if (nr_clients)
+ fio_clients_send_trigger(trigger_remote_cmd);
+ else {
verify_save_state();
fio_terminate_threads(TERMINATE_ALL);
exec_trigger(trigger_cmd);
gettimeofday(&tv, NULL);
ts.tv_sec = tv.tv_sec + sec;
ts.tv_nsec = (tv.tv_usec * 1000) + nsec;
- if (ts.tv_nsec > 1000000000ULL) {
+
+ if (ts.tv_nsec >= 1000000000ULL) {
ts.tv_nsec -= 1000000000ULL;
ts.tv_sec++;
}
projects / fio.git / blobdiff