Confusion between mutex and semaphore naming
[fio.git] / fio.c
diff --git a/fio.c b/fio.c
index 75b448d5e56ad684dd561a54b8b0fc033cb1d4c5..1594bf64e5976c8a4d107aa2768d50bde0ca0404 100644 (file)
--- a/fio.c
+++ b/fio.c
@@ -49,7 +49,7 @@ int nr_thread = 0;
 int shm_id = 0;
 int temp_stall_ts;
 
-static struct fio_sem *startup_sem;
+static struct fio_mutex *startup_mutex;
 static volatile int fio_abort;
 static int exit_value;
 
@@ -60,6 +60,8 @@ 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);
        td->runstate = runstate;
 }
 
@@ -70,13 +72,18 @@ static void terminate_threads(int 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);
+
+                       td->terminate = 1;
+                       td->o.start_delay = 0;
+
                        /*
                         * if the thread is running, just let it exit
                         */
                        if (td->runstate < TD_RUNNING)
                                kill(td->pid, SIGQUIT);
-                       td->terminate = 1;
-                       td->o.start_delay = 0;
+                       else if (td->io_ops->flags & FIO_SIGQUIT)
+                               kill(td->pid, SIGQUIT);
                }
        }
 }
@@ -446,6 +453,13 @@ static void do_io(struct thread_data *td)
                        break;
                }
 
+               /*
+                * Add verification end_io handler, if asked to verify
+                * a previously written file.
+                */
+               if (td->o.verify != VERIFY_NONE)
+                       io_u->end_io = verify_io_u;
+
                ret = td_io_queue(td, io_u);
                switch (ret) {
                case FIO_Q_COMPLETED:
@@ -561,6 +575,10 @@ sync_done:
                }
        }
 
+       if (td->o.fill_device && td->error == ENOSPC) {
+               td->error = 0;
+               td->terminate = 1;
+       }
        if (!td->error) {
                struct fio_file *f;
 
@@ -579,6 +597,12 @@ sync_done:
                }
        } else
                cleanup_pending_aio(td);
+
+       /*
+        * stop job if we failed doing any IO
+        */
+       if ((td->this_io_bytes[0] + td->this_io_bytes[1]) == 0)
+               td->done = 1;
 }
 
 static void cleanup_io_u(struct thread_data *td)
@@ -614,36 +638,31 @@ static void fill_io_buf(struct thread_data *td, struct io_u *io_u, int max_bs)
 
 static int init_io_u(struct thread_data *td)
 {
-       unsigned long long buf_size;
        struct io_u *io_u;
        unsigned int max_bs;
        int i, max_units;
        char *p;
 
-       if (td->io_ops->flags & FIO_SYNCIO)
-               max_units = 1;
-       else
-               max_units = td->o.iodepth;
-
+       max_units = td->o.iodepth;
        max_bs = max(td->o.max_bs[DDIR_READ], td->o.max_bs[DDIR_WRITE]);
-       buf_size = (unsigned long long) max_bs * (unsigned long long) max_units;
-       buf_size += page_mask;
-       if (buf_size != (size_t) buf_size) {
-               log_err("fio: IO memory too large. Reduce max_bs or iodepth\n");
-               return 1;
-       }
-
-       td->orig_buffer_size = buf_size;
+       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);
-       else if (td->orig_buffer_size & page_mask)
-               td->orig_buffer_size = (td->orig_buffer_size + page_mask) & ~page_mask;
+
+       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;
+       }
 
        if (allocate_io_mem(td))
                return 1;
 
-       p = ALIGN(td->orig_buffer);
+       if (td->o.odirect)
+               p = ALIGN(td->orig_buffer);
+       else
+               p = td->orig_buffer;
+
        for (i = 0; i < max_units; i++) {
                if (td->terminate)
                        return 1;
@@ -651,10 +670,12 @@ static int init_io_u(struct thread_data *td)
                memset(io_u, 0, sizeof(*io_u));
                INIT_LIST_HEAD(&io_u->list);
 
-               io_u->buf = p + max_bs * i;
+               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))
+                               fill_io_buf(td, io_u, max_bs);
+               }
 
                io_u->index = i;
                io_u->flags = IO_U_F_FREE;
@@ -721,6 +742,26 @@ static int switch_ioscheduler(struct thread_data *td)
        return 0;
 }
 
+static int keep_running(struct thread_data *td)
+{
+       unsigned long long io_done;
+
+       if (td->done)
+               return 0;
+       if (td->o.time_based)
+               return 1;
+       if (td->o.loops) {
+               td->o.loops--;
+               return 1;
+       }
+
+       io_done = td->io_bytes[DDIR_READ] + td->io_bytes[DDIR_WRITE] + td->io_skip_bytes;
+       if (io_done < td->o.size)
+               return 1;
+
+       return 0;
+}
+
 static int clear_io_state(struct thread_data *td)
 {
        struct fio_file *f;
@@ -736,11 +777,18 @@ static int clear_io_state(struct thread_data *td)
 
        td->last_was_sync = 0;
 
+       /*
+        * reset file done count if we are to start over
+        */
+       if (td->o.time_based || td->o.loops)
+               td->nr_done_files = 0;
+
        for_each_file(td, f, i)
                td_io_close_file(td, f);
 
        ret = 0;
        for_each_file(td, f, i) {
+               f->flags &= ~FIO_FILE_DONE;
                ret = td_io_open_file(td, f);
                if (ret)
                        break;
@@ -755,9 +803,8 @@ static int clear_io_state(struct thread_data *td)
  */
 static void *thread_main(void *data)
 {
-       unsigned long long runtime[2];
+       unsigned long long runtime[2], elapsed;
        struct thread_data *td = data;
-       unsigned long elapsed;
        int clear_state;
 
        if (!td->o.use_thread)
@@ -765,6 +812,8 @@ static void *thread_main(void *data)
 
        td->pid = getpid();
 
+       dprint(FD_PROCESS, "jobs pid=%d started\n", td->pid);
+
        INIT_LIST_HEAD(&td->io_u_freelist);
        INIT_LIST_HEAD(&td->io_u_busylist);
        INIT_LIST_HEAD(&td->io_u_requeues);
@@ -773,27 +822,31 @@ static void *thread_main(void *data)
        td->io_hist_tree = RB_ROOT;
 
        td_set_runstate(td, TD_INITIALIZED);
-       fio_sem_up(startup_sem);
-       fio_sem_down(td->mutex);
+       fio_mutex_up(startup_mutex);
+       fio_mutex_down(td->mutex);
 
        /*
-        * the ->mutex semaphore is now no longer used, close it to avoid
+        * the ->mutex mutex is now no longer used, close it to avoid
         * eating a file descriptor
         */
-       fio_sem_remove(td->mutex);
+       fio_mutex_remove(td->mutex);
+
+       /*
+        * May alter parameters that init_io_u() will use, so we need to
+        * do this first.
+        */
+       if (init_iolog(td))
+               goto err;
 
        if (init_io_u(td))
                goto err;
 
-       if (fio_setaffinity(td) == -1) {
+       if (td->o.cpumask_set && fio_setaffinity(td) == -1) {
                td_verror(td, errno, "cpu_set_affinity");
                goto err;
        }
 
-       if (init_iolog(td))
-               goto err;
-
-       if (td->ioprio) {
+       if (td->ioprio_set) {
                if (ioprio_set(IOPRIO_WHO_PROCESS, 0, td->ioprio) == -1) {
                        td_verror(td, errno, "ioprio_set");
                        goto err;
@@ -831,7 +884,7 @@ static void *thread_main(void *data)
 
        runtime[0] = runtime[1] = 0;
        clear_state = 0;
-       while (td->o.time_based || td->o.loops--) {
+       while (keep_running(td)) {
                fio_gettime(&td->start, NULL);
                memcpy(&td->ts.stat_sample_time, &td->start, sizeof(td->start));
 
@@ -867,7 +920,9 @@ static void *thread_main(void *data)
                if (td->error || td->terminate)
                        break;
 
-               if (td->o.verify == VERIFY_NONE)
+               if (!td->o.do_verify ||
+                   td->o.verify == VERIFY_NONE ||
+                   (td->io_ops->flags & FIO_UNIDIR))
                        continue;
 
                if (clear_io_state(td))
@@ -884,8 +939,8 @@ static void *thread_main(void *data)
        }
 
        update_rusage_stat(td);
-       td->ts.runtime[0] = runtime[0] / 1000;
-       td->ts.runtime[1] = runtime[1] / 1000;
+       td->ts.runtime[0] = (runtime[0] + 999) / 1000;
+       td->ts.runtime[1] = (runtime[1] + 999) / 1000;
        td->ts.total_run_time = mtime_since_now(&td->epoch);
        td->ts.io_bytes[0] = td->io_bytes[0];
        td->ts.io_bytes[1] = td->io_bytes[1];
@@ -896,8 +951,6 @@ static void *thread_main(void *data)
                finish_log(td, td->ts.slat_log, "slat");
        if (td->ts.clat_log)
                finish_log(td, td->ts.clat_log, "clat");
-       if (td->o.write_iolog_file)
-               write_iolog_close(td);
        if (td->o.exec_postrun) {
                if (system(td->o.exec_postrun) < 0)
                        log_err("fio: postrun %s failed\n", td->o.exec_postrun);
@@ -912,6 +965,13 @@ err:
        close_files(td);
        close_ioengine(td);
        cleanup_io_u(td);
+
+       /*
+        * do this very late, it will log file closing as well
+        */
+       if (td->o.write_iolog_file)
+               write_iolog_close(td);
+
        options_mem_free(td);
        td_set_runstate(td, TD_EXITED);
        return (void *) (unsigned long) td->error;
@@ -946,12 +1006,12 @@ static int fork_main(int shmid, int offset)
 static void reap_threads(int *nr_running, int *t_rate, int *m_rate)
 {
        struct thread_data *td;
-       int i, cputhreads, pending, status, ret;
+       int i, cputhreads, realthreads, pending, status, ret;
 
        /*
         * reap exited threads (TD_EXITED -> TD_REAPED)
         */
-       pending = cputhreads = 0;
+       realthreads = pending = cputhreads = 0;
        for_each_td(td, i) {
                int flags = 0;
 
@@ -961,8 +1021,14 @@ static void reap_threads(int *nr_running, int *t_rate, int *m_rate)
                 */
                if (td->io_ops && !strcmp(td->io_ops->name, "cpuio"))
                        cputhreads++;
+               else
+                       realthreads++;
 
-               if (!td->pid || td->runstate == TD_REAPED)
+               if (!td->pid) {
+                       pending++;
+                       continue;
+               }
+               if (td->runstate == TD_REAPED)
                        continue;
                if (td->o.use_thread) {
                        if (td->runstate == TD_EXITED) {
@@ -1014,8 +1080,11 @@ reaped:
                if (td->o.use_thread) {
                        long ret;
 
-                       if (pthread_join(td->thread, (void *) &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)--;
@@ -1027,7 +1096,7 @@ reaped:
                        exit_value++;
        }
 
-       if (*nr_running == cputhreads && !pending)
+       if (*nr_running == cputhreads && !pending && realthreads)
                terminate_threads(TERMINATE_ALL);
 }
 
@@ -1130,19 +1199,21 @@ static void run_threads(void)
                        nr_started++;
 
                        if (td->o.use_thread) {
+                               dprint(FD_PROCESS, "will pthread_create\n");
                                if (pthread_create(&td->thread, NULL, thread_main, td)) {
                                        perror("thread_create");
                                        nr_started--;
                                        break;
                                }
                        } else {
+                               dprint(FD_PROCESS, "will fork\n");
                                if (!fork()) {
                                        int ret = fork_main(shm_id, i);
 
                                        exit(ret);
                                }
                        }
-                       fio_sem_down(startup_sem);
+                       fio_mutex_down(startup_mutex);
                }
 
                /*
@@ -1197,7 +1268,7 @@ static void run_threads(void)
                        m_rate += td->o.ratemin;
                        t_rate += td->o.rate;
                        todo--;
-                       fio_sem_up(td->mutex);
+                       fio_mutex_up(td->mutex);
                }
 
                reap_threads(&nr_running, &t_rate, &m_rate);
@@ -1246,7 +1317,7 @@ int main(int argc, char *argv[])
                setup_log(&agg_io_log[DDIR_WRITE]);
        }
 
-       startup_sem = fio_sem_init(0);
+       startup_mutex = fio_mutex_init(0);
 
        set_genesis_time();
 
@@ -1262,6 +1333,6 @@ int main(int argc, char *argv[])
                }
        }
 
-       fio_sem_remove(startup_sem);
+       fio_mutex_remove(startup_mutex);
        return exit_value;
 }