Get rid of (now unused) disk_util_mutex
[fio.git] / backend.c
index 5ac365925fdd469a948643bd9ed14bc9759b3f94..dd7dc67ea6bdae550a41d11a8cb6d9c844dcaf33 100644 (file)
--- a/backend.c
+++ b/backend.c
 #include "lib/getrusage.h"
 #include "idletime.h"
 #include "err.h"
+#include "lib/tp.h"
 
 static pthread_t disk_util_thread;
-static struct fio_mutex *disk_thread_mutex;
+static pthread_cond_t du_cond;
+static pthread_mutex_t du_lock;
+
 static struct fio_mutex *startup_mutex;
 static struct flist_head *cgroup_list;
 static char *cgroup_mnt;
@@ -86,7 +89,7 @@ static void sig_int(int sig)
                        fio_server_got_signal(sig);
                else {
                        log_info("\nfio: terminating on signal %d\n", sig);
-                       fflush(stdout);
+                       log_info_flush();
                        exit_value = 128;
                }
 
@@ -182,7 +185,11 @@ static int __check_min_rate(struct thread_data *td, struct timeval *now,
                                                                ratemin);
                                return 1;
                        } else {
-                               rate = ((bytes - td->rate_bytes[ddir]) * 1000) / spent;
+                               if (spent)
+                                       rate = ((bytes - td->rate_bytes[ddir]) * 1000) / spent;
+                               else
+                                       rate = 0;
+
                                if (rate < ratemin ||
                                    bytes < td->rate_bytes[ddir]) {
                                        log_err("%s: min rate %u not met, got"
@@ -200,7 +207,11 @@ static int __check_min_rate(struct thread_data *td, struct timeval *now,
                                                td->o.name, rate_iops);
                                return 1;
                        } else {
-                               rate = ((iops - td->rate_blocks[ddir]) * 1000) / spent;
+                               if (spent)
+                                       rate = ((iops - td->rate_blocks[ddir]) * 1000) / spent;
+                               else
+                                       rate = 0;
+
                                if (rate < rate_iops_min ||
                                    iops < td->rate_blocks[ddir]) {
                                        log_err("%s: min iops rate %u not met,"
@@ -382,7 +393,7 @@ static int break_on_this_error(struct thread_data *td, enum fio_ddir ddir,
                         * fill_device option is set.
                         */
                        td_clear_error(td);
-                       td->terminate = 1;
+                       fio_mark_td_terminate(td);
                        return 1;
                } else {
                        /*
@@ -451,7 +462,7 @@ static void do_verify(struct thread_data *td, uint64_t verify_bytes)
                if (runtime_exceeded(td, &td->tv_cache)) {
                        __update_tv_cache(td);
                        if (runtime_exceeded(td, &td->tv_cache)) {
-                               td->terminate = 1;
+                               fio_mark_td_terminate(td);
                                break;
                        }
                }
@@ -637,7 +648,7 @@ static unsigned int exceeds_number_ios(struct thread_data *td)
 
 static int io_bytes_exceeded(struct thread_data *td)
 {
-       unsigned long long bytes;
+       unsigned long long bytes, limit;
 
        if (td_rw(td))
                bytes = td->this_io_bytes[DDIR_READ] + td->this_io_bytes[DDIR_WRITE];
@@ -648,7 +659,12 @@ static int io_bytes_exceeded(struct thread_data *td)
        else
                bytes = td->this_io_bytes[DDIR_TRIM];
 
-       return bytes >= td->o.size || exceeds_number_ios(td);
+       if (td->o.io_limit)
+               limit = td->o.io_limit;
+       else
+               limit = td->o.size;
+
+       return bytes >= limit || exceeds_number_ios(td);
 }
 
 /*
@@ -700,7 +716,7 @@ static uint64_t do_io(struct thread_data *td)
                if (runtime_exceeded(td, &td->tv_cache)) {
                        __update_tv_cache(td);
                        if (runtime_exceeded(td, &td->tv_cache)) {
-                               td->terminate = 1;
+                               fio_mark_td_terminate(td);
                                break;
                        }
                }
@@ -767,16 +783,21 @@ static uint64_t do_io(struct thread_data *td)
                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;
@@ -817,6 +838,7 @@ sync_done:
                        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)
@@ -902,7 +924,7 @@ reap:
 
        if (td->o.fill_device && td->error == ENOSPC) {
                td->error = 0;
-               td->terminate = 1;
+               fio_mark_td_terminate(td);
        }
        if (!td->error) {
                struct fio_file *f;
@@ -1110,12 +1132,14 @@ static int switch_ioscheduler(struct thread_data *td)
        /*
         * Read back and check that the selected scheduler is now the default.
         */
-       ret = fread(tmp, 1, sizeof(tmp), f);
+       ret = fread(tmp, sizeof(tmp), 1, f);
        if (ferror(f) || ret < 0) {
                td_verror(td, errno, "fread");
                fclose(f);
                return 1;
        }
+       tmp[sizeof(tmp) - 1] = '\0';
+
 
        sprintf(tmp2, "[%s]", td->o.ioscheduler);
        if (!strstr(tmp, tmp2)) {
@@ -1131,6 +1155,8 @@ static int switch_ioscheduler(struct thread_data *td)
 
 static int keep_running(struct thread_data *td)
 {
+       unsigned long long limit;
+
        if (td->done)
                return 0;
        if (td->o.time_based)
@@ -1142,14 +1168,19 @@ static int keep_running(struct thread_data *td)
        if (exceeds_number_ios(td))
                return 0;
 
-       if (td->o.size != -1ULL && ddir_rw_sum(td->io_bytes) < td->o.size) {
+       if (td->o.io_limit)
+               limit = td->o.io_limit;
+       else
+               limit = td->o.size;
+
+       if (limit != -1ULL && ddir_rw_sum(td->io_bytes) < limit) {
                uint64_t diff;
 
                /*
                 * If the difference is less than the minimum IO size, we
                 * are done.
                 */
-               diff = td->o.size - ddir_rw_sum(td->io_bytes);
+               diff = limit - ddir_rw_sum(td->io_bytes);
                if (diff < td_max_bs(td))
                        return 0;
 
@@ -1242,11 +1273,6 @@ static void *thread_main(void *data)
        } else
                td->pid = gettid();
 
-       /*
-        * fio_time_init() may not have been called yet if running as a server
-        */
-       fio_time_init();
-
        fio_local_clock_init(o->use_thread);
 
        dprint(FD_PROCESS, "jobs pid=%d started\n", (int) td->pid);
@@ -1317,6 +1343,7 @@ static void *thread_main(void *data)
 #ifdef CONFIG_LIBNUMA
        /* numa node setup */
        if (o->numa_cpumask_set || o->numa_memmask_set) {
+               struct bitmask *mask;
                int ret;
 
                if (numa_available() < 0) {
@@ -1325,7 +1352,9 @@ static void *thread_main(void *data)
                }
 
                if (o->numa_cpumask_set) {
-                       ret = numa_run_on_node_mask(o->numa_cpunodesmask);
+                       mask = numa_parse_nodestring(o->numa_cpunodes);
+                       ret = numa_run_on_node_mask(mask);
+                       numa_free_nodemask(mask);
                        if (ret == -1) {
                                td_verror(td, errno, \
                                        "numa_run_on_node_mask failed\n");
@@ -1335,12 +1364,16 @@ static void *thread_main(void *data)
 
                if (o->numa_memmask_set) {
 
+                       mask = NULL;
+                       if (o->numa_memnodes)
+                               mask = numa_parse_nodestring(o->numa_memnodes);
+
                        switch (o->numa_mem_mode) {
                        case MPOL_INTERLEAVE:
-                               numa_set_interleave_mask(o->numa_memnodesmask);
+                               numa_set_interleave_mask(mask);
                                break;
                        case MPOL_BIND:
-                               numa_set_membind(o->numa_memnodesmask);
+                               numa_set_membind(mask);
                                break;
                        case MPOL_LOCAL:
                                numa_set_localalloc();
@@ -1353,6 +1386,9 @@ static void *thread_main(void *data)
                                break;
                        }
 
+                       if (mask)
+                               numa_free_nodemask(mask);
+
                }
        }
 #endif
@@ -1410,6 +1446,9 @@ static void *thread_main(void *data)
                        goto err;
        }
 
+       if (td->flags & TD_F_COMPRESS_LOG)
+               tp_init(&td->tp_data);
+
        fio_verify_init(td);
 
        fio_gettime(&td->epoch, NULL);
@@ -1491,6 +1530,9 @@ static void *thread_main(void *data)
 
        fio_writeout_logs(td);
 
+       if (td->flags & TD_F_COMPRESS_LOG)
+               tp_exit(&td->tp_data);
+
        if (o->exec_postrun)
                exec_string(o, o->exec_postrun, (const char *)"postrun");
 
@@ -1542,7 +1584,7 @@ static int fork_main(int shmid, int offset)
        struct thread_data *td;
        void *data, *ret;
 
-#ifndef __hpux
+#if !defined(__hpux) && !defined(CONFIG_NO_SHM)
        data = shmat(shmid, NULL, 0);
        if (data == (void *) -1) {
                int __err = errno;
@@ -1563,6 +1605,13 @@ static int fork_main(int shmid, int offset)
        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,
+                       (unsigned long) time_since_now(&td->terminate_time));
+}
+
 /*
  * Run over the job map and reap the threads that have exited, if any.
  */
@@ -1640,6 +1689,17 @@ static void reap_threads(unsigned int *nr_running, unsigned int *t_rate,
                        }
                }
 
+               /*
+                * If the job is stuck, do a forceful timeout of it and
+                * move on.
+                */
+               if (td->terminate &&
+                   time_since_now(&td->terminate_time) >= FIO_REAP_TIMEOUT) {
+                       dump_td_info(td);
+                       td_set_runstate(td, TD_REAPED);
+                       goto reaped;
+               }
+
                /*
                 * thread is not dead, continue
                 */
@@ -1705,7 +1765,7 @@ static void run_threads(void)
                                                nr_process > 1 ? "es" : "");
                }
                log_info("\n");
-               fflush(stdout);
+               log_info_flush();
        }
 
        todo = thread_number;
@@ -1912,16 +1972,20 @@ static void run_threads(void)
        update_io_ticks();
 }
 
-void wait_for_disk_thread_exit(void)
+static void wait_for_disk_thread_exit(void)
 {
-       fio_mutex_down(disk_thread_mutex);
+       void *ret;
+
+       disk_util_start_exit();
+       pthread_cond_signal(&du_cond);
+       pthread_join(disk_util_thread, &ret);
 }
 
 static void free_disk_util(void)
 {
-       disk_util_start_exit();
-       wait_for_disk_thread_exit();
        disk_util_prune_entries();
+
+       pthread_cond_destroy(&du_cond);
 }
 
 static void *disk_thread_main(void *data)
@@ -1930,17 +1994,32 @@ static void *disk_thread_main(void *data)
 
        fio_mutex_up(startup_mutex);
 
-       while (threads && !ret) {
-               usleep(DISK_UTIL_MSEC * 1000);
-               if (!threads)
+       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++;
+               }
+
+               ret = pthread_cond_timedwait(&du_cond, &du_lock, &ts);
+               if (ret != ETIMEDOUT) {
+                       printf("disk thread should exit %d\n", ret);
                        break;
+               }
+
                ret = update_io_ticks();
 
                if (!is_backend)
                        print_thread_status();
        }
 
-       fio_mutex_up(disk_thread_mutex);
        return NULL;
 }
 
@@ -1950,22 +2029,15 @@ static int create_disk_util_thread(void)
 
        setup_disk_util();
 
-       disk_thread_mutex = fio_mutex_init(FIO_MUTEX_LOCKED);
+       pthread_cond_init(&du_cond, NULL);
+       pthread_mutex_init(&du_lock, NULL);
 
        ret = pthread_create(&disk_util_thread, NULL, disk_thread_main, NULL);
        if (ret) {
-               fio_mutex_remove(disk_thread_mutex);
                log_err("Can't create disk util thread: %s\n", strerror(ret));
                return 1;
        }
 
-       ret = pthread_detach(disk_util_thread);
-       if (ret) {
-               fio_mutex_remove(disk_thread_mutex);
-               log_err("Can't detatch disk util 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");
@@ -1987,9 +2059,13 @@ int fio_backend(void)
                return 0;
 
        if (write_bw_log) {
-               setup_log(&agg_io_log[DDIR_READ], 0, IO_LOG_TYPE_BW);
-               setup_log(&agg_io_log[DDIR_WRITE], 0, IO_LOG_TYPE_BW);
-               setup_log(&agg_io_log[DDIR_TRIM], 0, IO_LOG_TYPE_BW);
+               struct log_params p = {
+                       .log_type = IO_LOG_TYPE_BW,
+               };
+
+               setup_log(&agg_io_log[DDIR_READ], &p, "agg-read_bw.log");
+               setup_log(&agg_io_log[DDIR_WRITE], &p, "agg-write_bw.log");
+               setup_log(&agg_io_log[DDIR_TRIM], &p, "agg-trim_bw.log");
        }
 
        startup_mutex = fio_mutex_init(FIO_MUTEX_LOCKED);
@@ -2005,14 +2081,19 @@ int fio_backend(void)
 
        run_threads();
 
+       wait_for_disk_thread_exit();
+
        if (!fio_abort) {
-               show_run_stats();
+               __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_TRIM],
-                                       "agg-write_bw.log");
+                       int i;
+
+                       for (i = 0; i < DDIR_RWDIR_CNT; i++) {
+                               struct io_log *log = agg_io_log[i];
+
+                               flush_log(log);
+                               free_log(log);
+                       }
                }
        }
 
@@ -2025,7 +2106,6 @@ int fio_backend(void)
        sfree(cgroup_mnt);
 
        fio_mutex_remove(startup_mutex);
-       fio_mutex_remove(disk_thread_mutex);
        stat_exit();
        return exit_value;
 }