workqueue: remove knowledge of io issue (and others) stats
[fio.git] / backend.c
index b180196e4c4ec5b16d61120efaf14c9a2abdbf1f..63d47a29bb63a5ddad78fe8618590db2db499145 100644 (file)
--- a/backend.c
+++ b/backend.c
@@ -35,6 +35,7 @@
 #include <sys/wait.h>
 #include <sys/ipc.h>
 #include <sys/mman.h>
+#include <math.h>
 
 #include "fio.h"
 #ifndef FIO_NO_HAVE_SHM_H
@@ -48,7 +49,7 @@
 #include "cgroup.h"
 #include "profile.h"
 #include "lib/rand.h"
-#include "memalign.h"
+#include "lib/memalign.h"
 #include "server.h"
 #include "lib/getrusage.h"
 #include "idletime.h"
@@ -143,8 +144,8 @@ static void set_sig_handlers(void)
 /*
  * Check if we are above the minimum rate given.
  */
-static int __check_min_rate(struct thread_data *td, struct timeval *now,
-                           enum fio_ddir ddir)
+static bool __check_min_rate(struct thread_data *td, struct timeval *now,
+                            enum fio_ddir ddir)
 {
        unsigned long long bytes = 0;
        unsigned long iops = 0;
@@ -157,13 +158,13 @@ static int __check_min_rate(struct thread_data *td, struct timeval *now,
        assert(ddir_rw(ddir));
 
        if (!td->o.ratemin[ddir] && !td->o.rate_iops_min[ddir])
-               return 0;
+               return false;
 
        /*
         * allow a 2 second settle period in the beginning
         */
        if (mtime_since(&td->start, now) < 2000)
-               return 0;
+               return false;
 
        iops += td->this_io_blocks[ddir];
        bytes += td->this_io_bytes[ddir];
@@ -177,7 +178,7 @@ static int __check_min_rate(struct thread_data *td, struct timeval *now,
        if (td->rate_bytes[ddir] || td->rate_blocks[ddir]) {
                spent = mtime_since(&td->lastrate[ddir], now);
                if (spent < td->o.ratecycle)
-                       return 0;
+                       return false;
 
                if (td->o.rate[ddir] || td->o.ratemin[ddir]) {
                        /*
@@ -186,7 +187,7 @@ static int __check_min_rate(struct thread_data *td, struct timeval *now,
                        if (bytes < td->rate_bytes[ddir]) {
                                log_err("%s: min rate %u not met\n", td->o.name,
                                                                ratemin);
-                               return 1;
+                               return true;
                        } else {
                                if (spent)
                                        rate = ((bytes - td->rate_bytes[ddir]) * 1000) / spent;
@@ -198,7 +199,7 @@ static int __check_min_rate(struct thread_data *td, struct timeval *now,
                                        log_err("%s: min rate %u not met, got"
                                                " %luKB/sec\n", td->o.name,
                                                        ratemin, rate);
-                                       return 1;
+                                       return true;
                                }
                        }
                } else {
@@ -208,7 +209,7 @@ static int __check_min_rate(struct thread_data *td, struct timeval *now,
                        if (iops < rate_iops) {
                                log_err("%s: min iops rate %u not met\n",
                                                td->o.name, rate_iops);
-                               return 1;
+                               return true;
                        } else {
                                if (spent)
                                        rate = ((iops - td->rate_blocks[ddir]) * 1000) / spent;
@@ -220,7 +221,7 @@ static int __check_min_rate(struct thread_data *td, struct timeval *now,
                                        log_err("%s: min iops rate %u not met,"
                                                " got %lu\n", td->o.name,
                                                        rate_iops_min, rate);
-                                       return 1;
+                                       return true;
                                }
                        }
                }
@@ -229,12 +230,12 @@ static int __check_min_rate(struct thread_data *td, struct timeval *now,
        td->rate_bytes[ddir] = bytes;
        td->rate_blocks[ddir] = iops;
        memcpy(&td->lastrate[ddir], now, sizeof(*now));
-       return 0;
+       return false;
 }
 
-static int check_min_rate(struct thread_data *td, struct timeval *now)
+static bool check_min_rate(struct thread_data *td, struct timeval *now)
 {
-       int ret = 0;
+       bool ret = false;
 
        if (td->bytes_done[DDIR_READ])
                ret |= __check_min_rate(td, now, DDIR_READ);
@@ -285,20 +286,20 @@ static void cleanup_pending_aio(struct thread_data *td)
  * Helper to handle the final sync of a file. Works just like the normal
  * io path, just does everything sync.
  */
-static int fio_io_sync(struct thread_data *td, struct fio_file *f)
+static bool fio_io_sync(struct thread_data *td, struct fio_file *f)
 {
        struct io_u *io_u = __get_io_u(td);
        int ret;
 
        if (!io_u)
-               return 1;
+               return true;
 
        io_u->ddir = DDIR_SYNC;
        io_u->file = f;
 
        if (td_io_prep(td, io_u)) {
                put_io_u(td, io_u);
-               return 1;
+               return true;
        }
 
 requeue:
@@ -306,25 +307,25 @@ requeue:
        if (ret < 0) {
                td_verror(td, io_u->error, "td_io_queue");
                put_io_u(td, io_u);
-               return 1;
+               return true;
        } else if (ret == FIO_Q_QUEUED) {
                if (io_u_queued_complete(td, 1) < 0)
-                       return 1;
+                       return true;
        } else if (ret == FIO_Q_COMPLETED) {
                if (io_u->error) {
                        td_verror(td, io_u->error, "td_io_queue");
-                       return 1;
+                       return true;
                }
 
                if (io_u_sync_complete(td, io_u) < 0)
-                       return 1;
+                       return true;
        } else if (ret == FIO_Q_BUSY) {
                if (td_io_commit(td))
-                       return 1;
+                       return true;
                goto requeue;
        }
 
-       return 0;
+       return false;
 }
 
 static int fio_file_fsync(struct thread_data *td, struct fio_file *f)
@@ -353,16 +354,16 @@ static inline void update_tv_cache(struct thread_data *td)
                __update_tv_cache(td);
 }
 
-static inline int runtime_exceeded(struct thread_data *td, struct timeval *t)
+static inline bool runtime_exceeded(struct thread_data *td, struct timeval *t)
 {
        if (in_ramp_time(td))
-               return 0;
+               return false;
        if (!td->o.timeout)
-               return 0;
+               return false;
        if (utime_since(&td->epoch, t) >= td->o.timeout)
-               return 1;
+               return true;
 
-       return 0;
+       return false;
 }
 
 /*
@@ -382,8 +383,8 @@ static inline void update_runtime(struct thread_data *td,
        td->ts.runtime[ddir] += (elapsed_us[ddir] + 999) / 1000;
 }
 
-static int break_on_this_error(struct thread_data *td, enum fio_ddir ddir,
-                              int *retptr)
+static bool break_on_this_error(struct thread_data *td, enum fio_ddir ddir,
+                               int *retptr)
 {
        int ret = *retptr;
 
@@ -396,7 +397,7 @@ static int break_on_this_error(struct thread_data *td, enum fio_ddir ddir,
 
                eb = td_error_type(ddir, err);
                if (!(td->o.continue_on_error & (1 << eb)))
-                       return 1;
+                       return true;
 
                if (td_non_fatal_error(td, eb, err)) {
                        /*
@@ -406,7 +407,7 @@ static int break_on_this_error(struct thread_data *td, enum fio_ddir ddir,
                        update_error_count(td, err);
                        td_clear_error(td);
                        *retptr = 0;
-                       return 0;
+                       return false;
                } else if (td->o.fill_device && err == ENOSPC) {
                        /*
                         * We expect to hit this error if
@@ -414,18 +415,18 @@ static int break_on_this_error(struct thread_data *td, enum fio_ddir ddir,
                         */
                        td_clear_error(td);
                        fio_mark_td_terminate(td);
-                       return 1;
+                       return true;
                } else {
                        /*
                         * Stop the I/O in case of a fatal
                         * error.
                         */
                        update_error_count(td, err);
-                       return 1;
+                       return true;
                }
        }
 
-       return 0;
+       return false;
 }
 
 static void check_update_rusage(struct thread_data *td)
@@ -447,7 +448,7 @@ static int wait_for_completions(struct thread_data *td, struct timeval *time)
         * if the queue is full, we MUST reap at least 1 event
         */
        min_evts = min(td->o.iodepth_batch_complete_min, td->cur_depth);
-    if ((full && !min_evts) || !td->o.iodepth_batch_complete_min)
+       if ((full && !min_evts) || !td->o.iodepth_batch_complete_min)
                min_evts = 1;
 
        if (time && (__should_check_rate(td, DDIR_READ) ||
@@ -551,7 +552,7 @@ sync_done:
        return 0;
 }
 
-static inline int io_in_polling(struct thread_data *td)
+static inline bool io_in_polling(struct thread_data *td)
 {
        return !td->o.iodepth_batch_complete_min &&
                   !td->o.iodepth_batch_complete_max;
@@ -712,12 +713,12 @@ reap:
        dprint(FD_VERIFY, "exiting loop\n");
 }
 
-static unsigned int exceeds_number_ios(struct thread_data *td)
+static bool exceeds_number_ios(struct thread_data *td)
 {
        unsigned long long number_ios;
 
        if (!td->o.number_ios)
-               return 0;
+               return false;
 
        number_ios = ddir_rw_sum(td->io_blocks);
        number_ios += td->io_u_queued + td->io_u_in_flight;
@@ -725,7 +726,7 @@ static unsigned int exceeds_number_ios(struct thread_data *td)
        return number_ios >= (td->o.number_ios * td->loops);
 }
 
-static int io_issue_bytes_exceeded(struct thread_data *td)
+static bool io_issue_bytes_exceeded(struct thread_data *td)
 {
        unsigned long long bytes, limit;
 
@@ -747,7 +748,7 @@ static int io_issue_bytes_exceeded(struct thread_data *td)
        return bytes >= limit || exceeds_number_ios(td);
 }
 
-static int io_complete_bytes_exceeded(struct thread_data *td)
+static bool io_complete_bytes_exceeded(struct thread_data *td)
 {
        unsigned long long bytes, limit;
 
@@ -775,17 +776,30 @@ static int io_complete_bytes_exceeded(struct thread_data *td)
  */
 static long long usec_for_io(struct thread_data *td, enum fio_ddir ddir)
 {
-       uint64_t secs, remainder, bps, bytes;
+       uint64_t secs, remainder, bps, bytes, iops;
 
        assert(!(td->flags & TD_F_CHILD));
        bytes = td->rate_io_issue_bytes[ddir];
        bps = td->rate_bps[ddir];
-       if (bps) {
+
+       if (td->o.rate_process == RATE_PROCESS_POISSON) {
+               uint64_t val;
+               iops = bps / td->o.bs[ddir];
+               val = (int64_t) (1000000 / iops) *
+                               -logf(__rand_0_1(&td->poisson_state));
+               if (val) {
+                       dprint(FD_RATE, "poisson rate iops=%llu\n",
+                                       (unsigned long long) 1000000 / val);
+               }
+               td->last_usec += val;
+               return td->last_usec;
+       } else if (bps) {
                secs = bytes / bps;
                remainder = bytes % bps;
                return remainder * 1000000 / bps + secs * 1000000;
-       } else
-               return 0;
+       }
+
+       return 0;
 }
 
 /*
@@ -854,7 +868,7 @@ static uint64_t do_io(struct thread_data *td)
                if (flow_threshold_exceeded(td))
                        continue;
 
-               if (bytes_issued >= total_bytes)
+               if (!td->o.time_based && bytes_issued >= total_bytes)
                        break;
 
                io_u = get_io_u(td);
@@ -914,10 +928,20 @@ static uint64_t do_io(struct thread_data *td)
                        log_io_piece(td, io_u);
 
                if (td->o.io_submit_mode == IO_MODE_OFFLOAD) {
+                       const unsigned long blen = io_u->xfer_buflen;
+                       const enum fio_ddir ddir = acct_ddir(io_u);
+
                        if (td->error)
                                break;
+
                        ret = workqueue_enqueue(&td->io_wq, io_u);
 
+                       if (ret == FIO_Q_QUEUED && ddir_rw(ddir)) {
+                               td->io_issues[ddir]++;
+                               td->io_issue_bytes[ddir] += blen;
+                               td->rate_io_issue_bytes[ddir] += blen;
+                       }
+
                        if (should_check_rate(td))
                                td->rate_next_io_time[ddir] = usec_for_io(td, ddir);
 
@@ -1233,20 +1257,20 @@ static int switch_ioscheduler(struct thread_data *td)
        return 0;
 }
 
-static int keep_running(struct thread_data *td)
+static bool keep_running(struct thread_data *td)
 {
        unsigned long long limit;
 
        if (td->done)
-               return 0;
+               return false;
        if (td->o.time_based)
-               return 1;
+               return true;
        if (td->o.loops) {
                td->o.loops--;
-               return 1;
+               return true;
        }
        if (exceeds_number_ios(td))
-               return 0;
+               return false;
 
        if (td->o.io_limit)
                limit = td->o.io_limit;
@@ -1262,15 +1286,15 @@ static int keep_running(struct thread_data *td)
                 */
                diff = limit - ddir_rw_sum(td->io_bytes);
                if (diff < td_max_bs(td))
-                       return 0;
+                       return false;
 
                if (fio_files_done(td))
-                       return 0;
+                       return false;
 
-               return 1;
+               return true;
        }
 
-       return 0;
+       return false;
 }
 
 static int exec_string(struct thread_options *o, const char *string, const char *mode)
@@ -1344,16 +1368,38 @@ static void io_workqueue_fn(struct thread_data *td, struct io_u *io_u)
 
        td->cur_depth++;
 
-       ret = td_io_queue(td, io_u);
+       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_QUEUED)
-               ret = io_u_queued_complete(td, 1);
+       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;
 
-       td->cur_depth--;
+               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;
+       }
 }
 
 /*
@@ -1559,27 +1605,28 @@ static void *thread_main(void *data)
 
        fio_gettime(&td->epoch, NULL);
        fio_getrusage(&td->ru_start);
+       memcpy(&td->bw_sample_time, &td->epoch, sizeof(td->epoch));
+       memcpy(&td->iops_sample_time, &td->epoch, sizeof(td->epoch));
+
+       if (o->ratemin[DDIR_READ] || o->ratemin[DDIR_WRITE] ||
+                       o->ratemin[DDIR_TRIM]) {
+               memcpy(&td->lastrate[DDIR_READ], &td->bw_sample_time,
+                                       sizeof(td->bw_sample_time));
+               memcpy(&td->lastrate[DDIR_WRITE], &td->bw_sample_time,
+                                       sizeof(td->bw_sample_time));
+               memcpy(&td->lastrate[DDIR_TRIM], &td->bw_sample_time,
+                                       sizeof(td->bw_sample_time));
+       }
+
        clear_state = 0;
        while (keep_running(td)) {
                uint64_t verify_bytes;
 
                fio_gettime(&td->start, NULL);
-               memcpy(&td->bw_sample_time, &td->start, sizeof(td->start));
-               memcpy(&td->iops_sample_time, &td->start, sizeof(td->start));
                memcpy(&td->tv_cache, &td->start, sizeof(td->start));
 
-               if (o->ratemin[DDIR_READ] || o->ratemin[DDIR_WRITE] ||
-                               o->ratemin[DDIR_TRIM]) {
-                       memcpy(&td->lastrate[DDIR_READ], &td->bw_sample_time,
-                                               sizeof(td->bw_sample_time));
-                       memcpy(&td->lastrate[DDIR_WRITE], &td->bw_sample_time,
-                                               sizeof(td->bw_sample_time));
-                       memcpy(&td->lastrate[DDIR_TRIM], &td->bw_sample_time,
-                                               sizeof(td->bw_sample_time));
-               }
-
                if (clear_state)
-                       clear_io_state(td);
+                       clear_io_state(td, 0);
 
                prune_io_piece_log(td);
 
@@ -1617,7 +1664,7 @@ static void *thread_main(void *data)
                    (td->io_ops->flags & FIO_UNIDIR))
                        continue;
 
-               clear_io_state(td);
+               clear_io_state(td, 0);
 
                fio_gettime(&td->start, NULL);
 
@@ -1852,29 +1899,29 @@ reaped:
                fio_terminate_threads(TERMINATE_ALL);
 }
 
-static int __check_trigger_file(void)
+static bool __check_trigger_file(void)
 {
        struct stat sb;
 
        if (!trigger_file)
-               return 0;
+               return false;
 
        if (stat(trigger_file, &sb))
-               return 0;
+               return false;
 
        if (unlink(trigger_file) < 0)
                log_err("fio: failed to unlink %s: %s\n", trigger_file,
                                                        strerror(errno));
 
-       return 1;
+       return true;
 }
 
-static int trigger_timedout(void)
+static bool trigger_timedout(void)
 {
        if (trigger_timeout)
                return time_since_genesis() >= trigger_timeout;
 
-       return 0;
+       return false;
 }
 
 void exec_trigger(const char *cmd)
@@ -1930,13 +1977,13 @@ static void do_usleep(unsigned int usecs)
        usleep(usecs);
 }
 
-static int check_mount_writes(struct thread_data *td)
+static bool check_mount_writes(struct thread_data *td)
 {
        struct fio_file *f;
        unsigned int i;
 
        if (!td_write(td) || td->o.allow_mounted_write)
-               return 0;
+               return false;
 
        for_each_file(td, f, i) {
                if (f->filetype != FIO_TYPE_BD)
@@ -1945,10 +1992,10 @@ static int check_mount_writes(struct thread_data *td)
                        goto mounted;
        }
 
-       return 0;
+       return false;
 mounted:
        log_err("fio: %s appears mounted, and 'allow_mounted_write' isn't set. Aborting.", f->file_name);
-       return 1;
+       return true;
 }
 
 /*