[PATCH] Split the io_u io handling out of fio.c
[fio.git] / fio.c
diff --git a/fio.c b/fio.c
index 695d754976e8c82fa7ae4edc323a505b057ff9ad..818394a0859846690389408ffd15a75124b66fd0 100644 (file)
--- a/fio.c
+++ b/fio.c
@@ -2,6 +2,7 @@
  * fio - the flexible io tester
  *
  * Copyright (C) 2005 Jens Axboe <axboe@suse.de>
+ * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk>
  *
  *  This program is free software; you can redistribute it and/or modify
  *  it under the terms of the GNU General Public License as published by
  *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  *
  */
-#include <stdio.h>
-#include <stdlib.h>
 #include <unistd.h>
 #include <fcntl.h>
 #include <string.h>
-#include <errno.h>
 #include <signal.h>
 #include <time.h>
-#include <math.h>
 #include <assert.h>
-#include <dirent.h>
-#include <libgen.h>
-#include <sys/types.h>
 #include <sys/stat.h>
 #include <sys/wait.h>
 #include <sys/ipc.h>
 
 int groupid = 0;
 int thread_number = 0;
-static char run_str[MAX_JOBS + 1];
 int shm_id = 0;
-static LIST_HEAD(disk_list);
-static struct itimerval itimer;
-static struct timeval genesis;
-
-static void update_io_ticks(void);
-static void disk_util_timer_arm(void);
-static void print_thread_status(void);
-
-extern unsigned long long mlock_size;
-
-/*
- * thread life cycle
- */
-enum {
-       TD_NOT_CREATED = 0,
-       TD_CREATED,
-       TD_INITIALIZED,
-       TD_RUNNING,
-       TD_VERIFYING,
-       TD_EXITED,
-       TD_REAPED,
-};
+int temp_stall_ts;
+char *fio_inst_prefix = _INST_PREFIX;
 
 #define should_fsync(td)       ((td_write(td) || td_rw(td)) && (!(td)->odirect || (td)->override_sync))
 
-static sem_t startup_sem;
+static volatile int startup_sem;
 
 #define TERMINATE_ALL          (-1)
 #define JOB_START_TIMEOUT      (5 * 1000)
 
 static void terminate_threads(int group_id)
 {
+       struct thread_data *td;
        int i;
 
-       for (i = 0; i < thread_number; i++) {
-               struct thread_data *td = &threads[i];
-
+       for_each_td(td, i) {
                if (group_id == TERMINATE_ALL || groupid == td->groupid) {
                        td->terminate = 1;
                        td->start_delay = 0;
@@ -109,1739 +81,479 @@ static void sig_handler(int sig)
        }
 }
 
-static unsigned long utime_since(struct timeval *s, struct timeval *e)
+/*
+ * Check if we are above the minimum rate given.
+ */
+static int check_min_rate(struct thread_data *td, struct timeval *now)
 {
-       double sec, usec;
+       unsigned long spent;
+       unsigned long rate;
+       int ddir = td->ddir;
 
-       sec = e->tv_sec - s->tv_sec;
-       usec = e->tv_usec - s->tv_usec;
-       if (sec > 0 && usec < 0) {
-               sec--;
-               usec += 1000000;
-       }
+       /*
+        * allow a 2 second settle period in the beginning
+        */
+       if (mtime_since(&td->start, now) < 2000)
+               return 0;
+
+       /*
+        * if rate blocks is set, sample is running
+        */
+       if (td->rate_bytes) {
+               spent = mtime_since(&td->lastrate, now);
+               if (spent < td->ratecycle)
+                       return 0;
 
-       sec *= (double) 1000000;
+               rate = (td->this_io_bytes[ddir] - td->rate_bytes) / spent;
+               if (rate < td->ratemin) {
+                       fprintf(f_out, "%s: min rate %d not met, got %ldKiB/sec\n", td->name, td->ratemin, rate);
+                       return 1;
+               }
+       }
 
-       return sec + usec;
+       td->rate_bytes = td->this_io_bytes[ddir];
+       memcpy(&td->lastrate, now, sizeof(*now));
+       return 0;
 }
 
-static unsigned long utime_since_now(struct timeval *s)
+static inline int runtime_exceeded(struct thread_data *td, struct timeval *t)
 {
-       struct timeval t;
+       if (!td->timeout)
+               return 0;
+       if (mtime_since(&td->epoch, t) >= td->timeout * 1000)
+               return 1;
 
-       gettimeofday(&t, NULL);
-       return utime_since(s, &t);
+       return 0;
 }
 
-static unsigned long mtime_since(struct timeval *s, struct timeval *e)
+static inline void td_set_runstate(struct thread_data *td, int runstate)
 {
-       double sec, usec;
-
-       sec = e->tv_sec - s->tv_sec;
-       usec = e->tv_usec - s->tv_usec;
-       if (sec > 0 && usec < 0) {
-               sec--;
-               usec += 1000000;
-       }
-
-       sec *= (double) 1000;
-       usec /= (double) 1000;
-
-       return sec + usec;
+       td->runstate = runstate;
 }
 
-static unsigned long mtime_since_now(struct timeval *s)
+static struct fio_file *get_next_file(struct thread_data *td)
 {
-       struct timeval t;
+       unsigned int old_next_file = td->next_file;
+       struct fio_file *f;
 
-       gettimeofday(&t, NULL);
-       return mtime_since(s, &t);
-}
+       do {
+               f = &td->files[td->next_file];
 
-static inline unsigned long msec_now(struct timeval *s)
-{
-       return s->tv_sec * 1000 + s->tv_usec / 1000;
-}
+               td->next_file++;
+               if (td->next_file >= td->nr_files)
+                       td->next_file = 0;
 
-static unsigned long time_since_now(struct timeval *s)
-{
-       return mtime_since_now(s) / 1000;
-}
+               if (f->fd != -1)
+                       break;
 
-static int random_map_free(struct thread_data *td, unsigned long long block)
-{
-       unsigned int idx = RAND_MAP_IDX(td, block);
-       unsigned int bit = RAND_MAP_BIT(td, block);
+               f = NULL;
+       } while (td->next_file != old_next_file);
 
-       return (td->file_map[idx] & (1UL << bit)) == 0;
+       return f;
 }
 
-static int get_next_free_block(struct thread_data *td, unsigned long long *b)
+/*
+ * When job exits, we can cancel the in-flight IO if we are using async
+ * io. Attempt to do so.
+ */
+static void cleanup_pending_aio(struct thread_data *td)
 {
-       int i;
+       struct timespec ts = { .tv_sec = 0, .tv_nsec = 0};
+       struct list_head *entry, *n;
+       struct io_completion_data icd;
+       struct io_u *io_u;
+       int r;
 
-       *b = 0;
-       i = 0;
-       while ((*b) * td->min_bs < td->io_size) {
-               if (td->file_map[i] != -1UL) {
-                       *b += ffz(td->file_map[i]);
-                       return 0;
-               }
+       /*
+        * get immediately available events, if any
+        */
+       r = td_io_getevents(td, 0, td->cur_depth, &ts);
+       if (r > 0) {
+               icd.nr = r;
+               ios_completed(td, &icd);
+       }
 
-               *b += BLOCKS_PER_MAP;
-               i++;
+       /*
+        * now cancel remaining active events
+        */
+       if (td->io_ops->cancel) {
+               list_for_each_safe(entry, n, &td->io_u_busylist) {
+                       io_u = list_entry(entry, struct io_u, list);
+
+                       r = td->io_ops->cancel(td, io_u);
+                       if (!r)
+                               put_io_u(td, io_u);
+               }
        }
 
-       return 1;
+       if (td->cur_depth) {
+               r = td_io_getevents(td, td->cur_depth, td->cur_depth, NULL);
+               if (r > 0) {
+                       icd.nr = r;
+                       ios_completed(td, &icd);
+               }
+       }
 }
 
-static void mark_random_map(struct thread_data *td, struct io_u *io_u)
+/*
+ * The main verify engine. Runs over the writes we previusly submitted,
+ * reads the blocks back in, and checks the crc/md5 of the data.
+ */
+void do_verify(struct thread_data *td)
 {
-       unsigned long long block = io_u->offset / (unsigned long long) td->min_bs;
-       unsigned int blocks = 0;
+       struct timeval t;
+       struct io_u *io_u, *v_io_u = NULL;
+       struct io_completion_data icd;
+       struct fio_file *f;
+       int ret, i;
 
-       while (blocks < (io_u->buflen / td->min_bs)) {
-               unsigned int idx, bit;
+       /*
+        * sync io first and invalidate cache, to make sure we really
+        * read from disk.
+        */
+       for_each_file(td, f, i) {
+               td_io_sync(td, f);
+               file_invalidate_cache(td, f);
+       }
 
-               if (!random_map_free(td, block))
-                       break;
+       td_set_runstate(td, TD_VERIFYING);
 
-               idx = RAND_MAP_IDX(td, block);
-               bit = RAND_MAP_BIT(td, block);
+       do {
+               if (td->terminate)
+                       break;
 
-               assert(idx < td->num_maps);
+               gettimeofday(&t, NULL);
+               if (runtime_exceeded(td, &t))
+                       break;
 
-               td->file_map[idx] |= (1UL << bit);
-               block++;
-               blocks++;
-       }
+               io_u = __get_io_u(td);
+               if (!io_u)
+                       break;
 
-       if ((blocks * td->min_bs) < io_u->buflen)
-               io_u->buflen = blocks * td->min_bs;
-}
+               if (get_next_verify(td, io_u)) {
+                       put_io_u(td, io_u);
+                       break;
+               }
 
-static inline void add_stat_sample(struct io_stat *is, unsigned long val)
-{
-       if (val > is->max_val)
-               is->max_val = val;
-       if (val < is->min_val)
-               is->min_val = val;
-
-       is->val += val;
-       is->val_sq += val * val;
-       is->samples++;
-}
+               f = get_next_file(td);
+               if (!f)
+                       break;
 
-static void add_log_sample(struct thread_data *td, struct io_log *iolog,
-                          unsigned long val, int ddir)
-{
-       if (iolog->nr_samples == iolog->max_samples) {
-               int new_size = sizeof(struct io_sample) * iolog->max_samples*2;
+               io_u->file = f;
 
-               iolog->log = realloc(iolog->log, new_size);
-               iolog->max_samples <<= 1;
-       }
+               if (td_io_prep(td, io_u)) {
+                       put_io_u(td, io_u);
+                       break;
+               }
 
-       iolog->log[iolog->nr_samples].val = val;
-       iolog->log[iolog->nr_samples].time = mtime_since_now(&td->epoch);
-       iolog->log[iolog->nr_samples].ddir = ddir;
-       iolog->nr_samples++;
-}
+               ret = td_io_queue(td, io_u);
+               if (ret) {
+                       put_io_u(td, io_u);
+                       td_verror(td, ret);
+                       break;
+               }
 
-static void add_clat_sample(struct thread_data *td, int ddir,unsigned long msec)
-{
-       add_stat_sample(&td->clat_stat[ddir], msec);
+               /*
+                * we have one pending to verify, do that while
+                * we are doing io on the next one
+                */
+               if (do_io_u_verify(td, &v_io_u))
+                       break;
 
-       if (td->clat_log)
-               add_log_sample(td, td->clat_log, msec, ddir);
-}
+               ret = td_io_getevents(td, 1, 1, NULL);
+               if (ret != 1) {
+                       if (ret < 0)
+                               td_verror(td, ret);
+                       break;
+               }
 
-static void add_slat_sample(struct thread_data *td, int ddir,unsigned long msec)
-{
-       add_stat_sample(&td->slat_stat[ddir], msec);
+               v_io_u = td->io_ops->event(td, 0);
+               icd.nr = 1;
+               icd.error = 0;
+               io_completed(td, v_io_u, &icd);
 
-       if (td->slat_log)
-               add_log_sample(td, td->slat_log, msec, ddir);
-}
+               if (icd.error) {
+                       td_verror(td, icd.error);
+                       put_io_u(td, v_io_u);
+                       v_io_u = NULL;
+                       break;
+               }
 
-static void add_bw_sample(struct thread_data *td, int ddir)
-{
-       unsigned long spent = mtime_since_now(&td->stat_sample_time[ddir]);
-       unsigned long rate;
+               /*
+                * if we can't submit more io, we need to verify now
+                */
+               if (queue_full(td) && do_io_u_verify(td, &v_io_u))
+                       break;
 
-       if (spent < td->bw_avg_time)
-               return;
+       } while (1);
 
-       rate = (td->this_io_bytes[ddir] - td->stat_io_bytes[ddir]) / spent;
-       add_stat_sample(&td->bw_stat[ddir], rate);
+       do_io_u_verify(td, &v_io_u);
 
-       if (td->bw_log)
-               add_log_sample(td, td->bw_log, rate, ddir);
+       if (td->cur_depth)
+               cleanup_pending_aio(td);
 
-       gettimeofday(&td->stat_sample_time[ddir], NULL);
-       td->stat_io_bytes[ddir] = td->this_io_bytes[ddir];
+       td_set_runstate(td, TD_RUNNING);
 }
 
-static int get_next_offset(struct thread_data *td, unsigned long long *offset)
+/*
+ * Not really an io thread, all it does is burn CPU cycles in the specified
+ * manner.
+ */
+static void do_cpuio(struct thread_data *td)
 {
-       unsigned long long b, rb;
-       long r;
-
-       if (!td->sequential) {
-               unsigned long long max_blocks = td->io_size / td->min_bs;
-               int loops = 50;
-
-               do {
-                       lrand48_r(&td->random_state, &r);
-                       b = ((max_blocks - 1) * r / (unsigned long long) (RAND_MAX+1.0));
-                       rb = b + (td->file_offset / td->min_bs);
-                       loops--;
-               } while (!random_map_free(td, rb) && loops);
-
-               if (!loops) {
-                       if (get_next_free_block(td, &b))
-                               return 1;
-               }
-       } else
-               b = td->last_pos / td->min_bs;
-
-       *offset = (b * td->min_bs) + td->file_offset;
-       if (*offset > td->real_file_size)
-               return 1;
+       struct timeval e;
+       int split = 100 / td->cpuload;
+       int i = 0;
 
-       return 0;
-}
+       while (!td->terminate) {
+               gettimeofday(&e, NULL);
 
-static unsigned int get_next_buflen(struct thread_data *td)
-{
-       unsigned int buflen;
-       long r;
-
-       if (td->min_bs == td->max_bs)
-               buflen = td->min_bs;
-       else {
-               lrand48_r(&td->bsrange_state, &r);
-               buflen = (1 + (double) (td->max_bs - 1) * r / (RAND_MAX + 1.0));
-               buflen = (buflen + td->min_bs - 1) & ~(td->min_bs - 1);
-       }
+               if (runtime_exceeded(td, &e))
+                       break;
 
-       if (buflen > td->io_size - td->this_io_bytes[td->ddir])
-               buflen = td->io_size - td->this_io_bytes[td->ddir];
+               if (!(i % split))
+                       __usec_sleep(10000);
+               else
+                       usec_sleep(td, 10000);
 
-       return buflen;
+               i++;
+       }
 }
 
 /*
- * busy looping version for the last few usec
+ * Main IO worker function. It retrieves io_u's to process and queues
+ * and reaps them, checking for rate and errors along the way.
  */
-static void __usec_sleep(unsigned int usec)
+static void do_io(struct thread_data *td)
 {
-       struct timeval start;
+       struct io_completion_data icd;
+       struct timeval s, e;
+       unsigned long usec;
+       struct fio_file *f;
+       int i, ret = 0;
 
-       gettimeofday(&start, NULL);
-       while (utime_since_now(&start) < usec)
-               nop;
-}
+       td_set_runstate(td, TD_RUNNING);
 
-static void usec_sleep(struct thread_data *td, unsigned long usec)
-{
-       struct timespec req, rem;
+       while (td->this_io_bytes[td->ddir] < td->io_size) {
+               struct timespec ts = { .tv_sec = 0, .tv_nsec = 0};
+               struct timespec *timeout;
+               int min_evts = 0;
+               struct io_u *io_u;
 
-       req.tv_sec = usec / 1000000;
-       req.tv_nsec = usec * 1000 - req.tv_sec * 1000000;
+               if (td->terminate)
+                       break;
 
-       do {
-               if (usec < 5000) {
-                       __usec_sleep(usec);
+               f = get_next_file(td);
+               if (!f)
                        break;
-               }
 
-               rem.tv_sec = rem.tv_nsec = 0;
-               if (nanosleep(&req, &rem) < 0)
+               io_u = get_io_u(td, f);
+               if (!io_u)
                        break;
 
-               if ((rem.tv_sec + rem.tv_nsec) == 0)
+               memcpy(&s, &io_u->start_time, sizeof(s));
+
+               ret = td_io_queue(td, io_u);
+               if (ret) {
+                       put_io_u(td, io_u);
+                       td_verror(td, ret);
                        break;
+               }
 
-               req.tv_nsec = rem.tv_nsec;
-               req.tv_sec = rem.tv_sec;
+               add_slat_sample(td, io_u->ddir, mtime_since(&io_u->start_time, &io_u->issue_time));
 
-               usec = rem.tv_sec * 1000000 + rem.tv_nsec / 1000;
-       } while (!td->terminate);
-}
+               if (td->cur_depth < td->iodepth) {
+                       timeout = &ts;
+                       min_evts = 0;
+               } else {
+                       timeout = NULL;
+                       min_evts = 1;
+               }
 
-static void rate_throttle(struct thread_data *td, unsigned long time_spent,
-                         unsigned int bytes)
-{
-       unsigned long usec_cycle;
+               ret = td_io_getevents(td, min_evts, td->cur_depth, timeout);
+               if (ret < 0) {
+                       td_verror(td, -ret);
+                       break;
+               } else if (!ret)
+                       continue;
 
-       if (!td->rate)
-               return;
+               icd.nr = ret;
+               ios_completed(td, &icd);
+               if (icd.error) {
+                       td_verror(td, icd.error);
+                       break;
+               }
 
-       usec_cycle = td->rate_usec_cycle * (bytes / td->min_bs);
+               /*
+                * the rate is batched for now, it should work for batches
+                * of completions except the very first one which may look
+                * a little bursty
+                */
+               gettimeofday(&e, NULL);
+               usec = utime_since(&s, &e);
 
-       if (time_spent < usec_cycle) {
-               unsigned long s = usec_cycle - time_spent;
+               rate_throttle(td, usec, icd.bytes_done[td->ddir]);
 
-               td->rate_pending_usleep += s;
-               if (td->rate_pending_usleep >= 100000) {
-                       usec_sleep(td, td->rate_pending_usleep);
-                       td->rate_pending_usleep = 0;
+               if (check_min_rate(td, &e)) {
+                       if (rate_quit)
+                               terminate_threads(td->groupid);
+                       td_verror(td, ENOMEM);
+                       break;
                }
-       } else {
-               long overtime = time_spent - usec_cycle;
 
-               td->rate_pending_usleep -= overtime;
+               if (runtime_exceeded(td, &e))
+                       break;
+
+               if (td->thinktime)
+                       usec_sleep(td, td->thinktime);
+
+               if (should_fsync(td) && td->fsync_blocks &&
+                   (td->io_blocks[DDIR_WRITE] % td->fsync_blocks) == 0)
+                       td_io_sync(td, f);
+       }
+
+       if (!ret) {
+               if (td->cur_depth)
+                       cleanup_pending_aio(td);
+
+               if (should_fsync(td) && td->end_fsync) {
+                       td_set_runstate(td, TD_FSYNCING);
+                       for_each_file(td, f, i)
+                               td_io_sync(td, f);
+               }
        }
 }
 
-static int check_min_rate(struct thread_data *td, struct timeval *now)
+static int td_io_init(struct thread_data *td)
 {
-       unsigned long spent;
-       unsigned long rate;
-       int ddir = td->ddir;
-
-       /*
-        * allow a 2 second settle period in the beginning
-        */
-       if (mtime_since(&td->start, now) < 2000)
-               return 0;
-
-       /*
-        * if rate blocks is set, sample is running
-        */
-       if (td->rate_bytes) {
-               spent = mtime_since(&td->lastrate, now);
-               if (spent < td->ratecycle)
-                       return 0;
-
-               rate = (td->this_io_bytes[ddir] - td->rate_bytes) / spent;
-               if (rate < td->ratemin) {
-                       printf("Client%d: min rate %d not met, got %ldKiB/sec\n", td->thread_number, td->ratemin, rate);
-                       if (rate_quit)
-                               terminate_threads(td->groupid);
-                       return 1;
-               }
-       }
-
-       td->rate_bytes = td->this_io_bytes[ddir];
-       memcpy(&td->lastrate, now, sizeof(*now));
-       return 0;
-}
-
-static inline int runtime_exceeded(struct thread_data *td, struct timeval *t)
-{
-       if (!td->timeout)
-               return 0;
-       if (mtime_since(&td->epoch, t) >= td->timeout * 1000)
-               return 1;
-
-       return 0;
-}
-
-static void fill_random_bytes(struct thread_data *td,
-                             unsigned char *p, unsigned int len)
-{
-       unsigned int todo;
-       double r;
-
-       while (len) {
-               drand48_r(&td->verify_state, &r);
-
-               /*
-                * lrand48_r seems to be broken and only fill the bottom
-                * 32-bits, even on 64-bit archs with 64-bit longs
-                */
-               todo = sizeof(r);
-               if (todo > len)
-                       todo = len;
-
-               memcpy(p, &r, todo);
-
-               len -= todo;
-               p += todo;
-       }
-}
-
-static void hexdump(void *buffer, int len)
-{
-       unsigned char *p = buffer;
-       int i;
-
-       for (i = 0; i < len; i++)
-               printf("%02x", p[i]);
-       printf("\n");
-}
-
-static int verify_io_u_crc32(struct verify_header *hdr, struct io_u *io_u)
-{
-       unsigned char *p = (unsigned char *) io_u->buf;
-       unsigned long c;
-       int ret;
-
-       p += sizeof(*hdr);
-       c = crc32(p, hdr->len - sizeof(*hdr));
-       ret = c != hdr->crc32;
-
-       if (ret) {
-               fprintf(stderr, "crc32: verify failed at %llu/%u\n", io_u->offset, io_u->buflen);
-               fprintf(stderr, "crc32: wanted %lx, got %lx\n", hdr->crc32, c);
-       }
-
-       return ret;
-}
-
-static int verify_io_u_md5(struct verify_header *hdr, struct io_u *io_u)
-{
-       unsigned char *p = (unsigned char *) io_u->buf;
-       struct md5_ctx md5_ctx;
-       int ret;
-
-       memset(&md5_ctx, 0, sizeof(md5_ctx));
-       p += sizeof(*hdr);
-       md5_update(&md5_ctx, p, hdr->len - sizeof(*hdr));
-
-       ret = memcmp(hdr->md5_digest, md5_ctx.hash, sizeof(md5_ctx.hash));
-       if (ret) {
-               fprintf(stderr, "md5: verify failed at %llu/%u\n", io_u->offset, io_u->buflen);
-               hexdump(hdr->md5_digest, sizeof(hdr->md5_digest));
-               hexdump(md5_ctx.hash, sizeof(md5_ctx.hash));
-       }
-
-       return ret;
-}
-
-static int verify_io_u(struct io_u *io_u)
-{
-       struct verify_header *hdr = (struct verify_header *) io_u->buf;
-       int ret;
-
-       if (hdr->fio_magic != FIO_HDR_MAGIC)
-               return 1;
-
-       if (hdr->verify_type == VERIFY_MD5)
-               ret = verify_io_u_md5(hdr, io_u);
-       else if (hdr->verify_type == VERIFY_CRC32)
-               ret = verify_io_u_crc32(hdr, io_u);
-       else {
-               fprintf(stderr, "Bad verify type %d\n", hdr->verify_type);
-               ret = 1;
-       }
-
-       return ret;
-}
-
-static void fill_crc32(struct verify_header *hdr, void *p, unsigned int len)
-{
-       hdr->crc32 = crc32(p, len);
-}
-
-static void fill_md5(struct verify_header *hdr, void *p, unsigned int len)
-{
-       struct md5_ctx md5_ctx;
-
-       memset(&md5_ctx, 0, sizeof(md5_ctx));
-       md5_update(&md5_ctx, p, len);
-       memcpy(hdr->md5_digest, md5_ctx.hash, sizeof(md5_ctx.hash));
-}
-
-unsigned int hweight32(unsigned int w)
-{
-       unsigned int res = w - ((w >> 1) & 0x55555555);
-
-       res = (res & 0x33333333) + ((res >> 2) & 0x33333333);
-       res = (res + (res >> 4)) & 0x0F0F0F0F;
-       res = res + (res >> 8);
-
-       return (res + (res >> 16)) & 0x000000FF;
-}
-
-unsigned long hweight64(unsigned long long w)
-{
-#if __WORDSIZE == 32
-        return hweight32((unsigned int)(w >> 32)) + hweight32((unsigned int)w);
-#elif __WORDSIZE == 64
-       unsigned long long v = w - ((w >> 1) & 0x5555555555555555ul);
-
-       v = (v & 0x3333333333333333ul) + ((v >> 2) & 0x3333333333333333ul);
-       v = (v + (v >> 4)) & 0x0F0F0F0F0F0F0F0Ful;
-       v = v + (v >> 8);
-       v = v + (v >> 16);
-
-       return (v + (v >> 32)) & 0x00000000000000FFul;
-#else
-#error __WORDSIZE not defined
-#endif
-}
-
-static int get_rw_ddir(struct thread_data *td)
-{
-       /*
-        * perhaps cheasy, but use the hamming weight of the position
-        * as a randomizer for data direction.
-        */
-       if (td_rw(td))
-               return hweight64(td->last_pos) & 1;
-       else if (td_read(td))
-               return DDIR_READ;
-       else
-               return DDIR_WRITE;
-}
-
-/*
- * fill body of io_u->buf with random data and add a header with the
- * (eg) sha1sum of that data.
- */
-static void populate_io_u(struct thread_data *td, struct io_u *io_u)
-{
-       unsigned char *p = (unsigned char *) io_u->buf;
-       struct verify_header hdr;
-
-       hdr.fio_magic = FIO_HDR_MAGIC;
-       hdr.len = io_u->buflen;
-       p += sizeof(hdr);
-       fill_random_bytes(td, p, io_u->buflen - sizeof(hdr));
-
-       if (td->verify == VERIFY_MD5) {
-               fill_md5(&hdr, p, io_u->buflen - sizeof(hdr));
-               hdr.verify_type = VERIFY_MD5;
-       } else {
-               fill_crc32(&hdr, p, io_u->buflen - sizeof(hdr));
-               hdr.verify_type = VERIFY_CRC32;
-       }
-
-       memcpy(io_u->buf, &hdr, sizeof(hdr));
-}
-
-static int td_io_prep(struct thread_data *td, struct io_u *io_u)
-{
-       if (td->io_prep && td->io_prep(td, io_u))
-               return 1;
+       if (td->io_ops->init)
+               return td->io_ops->init(td);
 
        return 0;
 }
 
-void put_io_u(struct thread_data *td, struct io_u *io_u)
-{
-       list_del(&io_u->list);
-       list_add(&io_u->list, &td->io_u_freelist);
-       td->cur_depth--;
-}
-
-static void write_iolog_put(struct thread_data *td, struct io_u *io_u)
-{
-       fprintf(td->iolog_f, "%d,%llu,%u\n", io_u->ddir, io_u->offset, io_u->buflen);
-}
-
-static int read_iolog_get(struct thread_data *td, struct io_u *io_u)
-{
-       struct io_piece *ipo;
-
-       if (!list_empty(&td->io_log_list)) {
-               ipo = list_entry(td->io_log_list.next, struct io_piece, list);
-               list_del(&ipo->list);
-               io_u->offset = ipo->offset;
-               io_u->buflen = ipo->len;
-               io_u->ddir = ipo->ddir;
-               free(ipo);
-               return 0;
-       }
-
-       return 1;
-}
-
-static int fill_io_u(struct thread_data *td, struct io_u *io_u)
-{
-       /*
-        * If using an iolog, grab next piece if any available.
-        */
-       if (td->read_iolog)
-               return read_iolog_get(td, io_u);
-
-       /*
-        * No log, let the seq/rand engine retrieve the next position.
-        */
-       if (!get_next_offset(td, &io_u->offset)) {
-               io_u->buflen = get_next_buflen(td);
-
-               if (io_u->buflen) {
-                       io_u->ddir = get_rw_ddir(td);
-
-                       /*
-                        * If using a write iolog, store this entry.
-                        */
-                       if (td->write_iolog)
-                               write_iolog_put(td, io_u);
-
-                       return 0;
-               }
-       }
-
-       return 1;
-}
-
-#define queue_full(td) (list_empty(&(td)->io_u_freelist))
-
-struct io_u *__get_io_u(struct thread_data *td)
-{
-       struct io_u *io_u;
-
-       if (queue_full(td))
-               return NULL;
-
-       io_u = list_entry(td->io_u_freelist.next, struct io_u, list);
-       io_u->error = 0;
-       io_u->resid = 0;
-       list_del(&io_u->list);
-       list_add(&io_u->list, &td->io_u_busylist);
-       td->cur_depth++;
-       return io_u;
-}
-
-static struct io_u *get_io_u(struct thread_data *td)
+static void cleanup_io_u(struct thread_data *td)
 {
+       struct list_head *entry, *n;
        struct io_u *io_u;
 
-       io_u = __get_io_u(td);
-       if (!io_u)
-               return NULL;
-
-       if (td->zone_bytes >= td->zone_size) {
-               td->zone_bytes = 0;
-               td->last_pos += td->zone_skip;
-       }
-
-       if (fill_io_u(td, io_u)) {
-               put_io_u(td, io_u);
-               return NULL;
-       }
-
-       if (io_u->buflen + io_u->offset > td->real_file_size)
-               io_u->buflen = td->real_file_size - io_u->offset;
-
-       if (!io_u->buflen) {
-               put_io_u(td, io_u);
-               return NULL;
-       }
-
-       if (!td->read_iolog && !td->sequential)
-               mark_random_map(td, io_u);
-
-       td->last_pos += io_u->buflen;
-
-       if (td->verify != VERIFY_NONE)
-               populate_io_u(td, io_u);
-
-       if (td_io_prep(td, io_u)) {
-               put_io_u(td, io_u);
-               return NULL;
-       }
-
-       gettimeofday(&io_u->start_time, NULL);
-       return io_u;
-}
-
-static inline void td_set_runstate(struct thread_data *td, int runstate)
-{
-       td->old_runstate = td->runstate;
-       td->runstate = runstate;
-}
-
-static int get_next_verify(struct thread_data *td, struct io_u *io_u)
-{
-       struct io_piece *ipo;
-
-       if (list_empty(&td->io_hist_list))
-               return 1;
-
-       ipo = list_entry(td->io_hist_list.next, struct io_piece, list);
-       list_del(&ipo->list);
-
-       io_u->offset = ipo->offset;
-       io_u->buflen = ipo->len;
-       io_u->ddir = DDIR_READ;
-       free(ipo);
-       return 0;
-}
-
-static void prune_io_piece_log(struct thread_data *td)
-{
-       struct io_piece *ipo;
-
-       while (!list_empty(&td->io_hist_list)) {
-               ipo = list_entry(td->io_hist_list.next, struct io_piece, list);
-
-               list_del(&ipo->list);
-               free(ipo);
-       }
-}
-
-/*
- * log a succesful write, so we can unwind the log for verify
- */
-static void log_io_piece(struct thread_data *td, struct io_u *io_u)
-{
-       struct io_piece *ipo = malloc(sizeof(struct io_piece));
-       struct list_head *entry;
-
-       INIT_LIST_HEAD(&ipo->list);
-       ipo->offset = io_u->offset;
-       ipo->len = io_u->buflen;
-
-       /*
-        * for random io where the writes extend the file, it will typically
-        * be laid out with the block scattered as written. it's faster to
-        * read them in in that order again, so don't sort
-        */
-       if (td->sequential || !td->overwrite) {
-               list_add_tail(&ipo->list, &td->io_hist_list);
-               return;
-       }
-
-       /*
-        * for random io, sort the list so verify will run faster
-        */
-       entry = &td->io_hist_list;
-       while ((entry = entry->prev) != &td->io_hist_list) {
-               struct io_piece *__ipo = list_entry(entry, struct io_piece, list);
-
-               if (__ipo->offset < ipo->offset)
-                       break;
-       }
-
-       list_add(&ipo->list, entry);
-}
-
-static void write_iolog_close(struct thread_data *td)
-{
-       fflush(td->iolog_f);
-       fclose(td->iolog_f);
-       free(td->iolog_buf);
-}
-
-static int init_iolog(struct thread_data *td)
-{
-       unsigned long long offset;
-       unsigned int bytes;
-       char *str, *p;
-       FILE *f;
-       int rw, i, reads, writes;
-
-       if (!td->read_iolog && !td->write_iolog)
-               return 0;
-
-       if (td->read_iolog)
-               f = fopen(td->iolog_file, "r");
-       else
-               f = fopen(td->iolog_file, "w");
-
-       if (!f) {
-               perror("fopen iolog");
-               printf("file %s, %d/%d\n", td->iolog_file, td->read_iolog, td->write_iolog);
-               return 1;
-       }
-
-       /*
-        * That's it for writing, setup a log buffer and we're done.
-         */
-       if (td->write_iolog) {
-               td->iolog_f = f;
-               td->iolog_buf = malloc(8192);
-               setvbuf(f, td->iolog_buf, _IOFBF, 8192);
-               return 0;
-       }
-
-       /*
-        * Read in the read iolog and store it, reuse the infrastructure
-        * for doing verifications.
-        */
-       str = malloc(4096);
-       reads = writes = i = 0;
-       while ((p = fgets(str, 4096, f)) != NULL) {
-               struct io_piece *ipo;
-
-               if (sscanf(p, "%d,%llu,%u", &rw, &offset, &bytes) != 3) {
-                       fprintf(stderr, "bad iolog: %s\n", p);
-                       continue;
-               }
-               if (rw == DDIR_READ)
-                       reads++;
-               else if (rw == DDIR_WRITE)
-                       writes++;
-               else {
-                       fprintf(stderr, "bad ddir: %d\n", rw);
-                       continue;
-               }
-
-               ipo = malloc(sizeof(*ipo));
-               INIT_LIST_HEAD(&ipo->list);
-               ipo->offset = offset;
-               ipo->len = bytes;
-               if (bytes > td->max_bs)
-                       td->max_bs = bytes;
-               ipo->ddir = rw;
-               list_add_tail(&ipo->list, &td->io_log_list);
-               i++;
-       }
-
-       free(str);
-       fclose(f);
-
-       if (!i)
-               return 1;
-
-       if (reads && !writes)
-               td->ddir = DDIR_READ;
-       else if (!reads && writes)
-               td->ddir = DDIR_READ;
-       else
-               td->iomix = 1;
-
-       return 0;
-}
-
-static int sync_td(struct thread_data *td)
-{
-       if (td->io_sync)
-               return td->io_sync(td);
-
-       return 0;
-}
-
-static int io_u_getevents(struct thread_data *td, int min, int max,
-                         struct timespec *t)
-{
-       return td->io_getevents(td, min, max, t);
-}
-
-static int io_u_queue(struct thread_data *td, struct io_u *io_u)
-{
-       gettimeofday(&io_u->issue_time, NULL);
-
-       return td->io_queue(td, io_u);
-}
-
-#define iocb_time(iocb)        ((unsigned long) (iocb)->data)
-
-static void io_completed(struct thread_data *td, struct io_u *io_u,
-                        struct io_completion_data *icd)
-{
-       struct timeval e;
-       unsigned long msec;
-
-       gettimeofday(&e, NULL);
-
-       if (!io_u->error) {
-               unsigned int bytes = io_u->buflen - io_u->resid;
-               const int idx = io_u->ddir;
-
-               td->io_blocks[idx]++;
-               td->io_bytes[idx] += bytes;
-               td->zone_bytes += bytes;
-               td->this_io_bytes[idx] += bytes;
-
-               msec = mtime_since(&io_u->issue_time, &e);
-
-               add_clat_sample(td, idx, msec);
-               add_bw_sample(td, idx);
-
-               if ((td_rw(td) || td_write(td)) && idx == DDIR_WRITE)
-                       log_io_piece(td, io_u);
-
-               icd->bytes_done[idx] += bytes;
-       } else
-               icd->error = io_u->error;
-}
-
-static void ios_completed(struct thread_data *td,struct io_completion_data *icd)
-{
-       struct io_u *io_u;
-       int i;
-
-       icd->error = 0;
-       icd->bytes_done[0] = icd->bytes_done[1] = 0;
-
-       for (i = 0; i < icd->nr; i++) {
-               io_u = td->io_event(td, i);
-
-               io_completed(td, io_u, icd);
-               put_io_u(td, io_u);
-       }
-}
-
-static void cleanup_pending_aio(struct thread_data *td)
-{
-       struct timespec ts = { .tv_sec = 0, .tv_nsec = 0};
-       struct list_head *entry, *n;
-       struct io_completion_data icd;
-       struct io_u *io_u;
-       int r;
-
-       /*
-        * get immediately available events, if any
-        */
-       r = io_u_getevents(td, 0, td->cur_depth, &ts);
-       if (r > 0) {
-               icd.nr = r;
-               ios_completed(td, &icd);
-       }
-
-       /*
-        * now cancel remaining active events
-        */
-       if (td->io_cancel) {
-               list_for_each_safe(entry, n, &td->io_u_busylist) {
-                       io_u = list_entry(entry, struct io_u, list);
-
-                       r = td->io_cancel(td, io_u);
-                       if (!r)
-                               put_io_u(td, io_u);
-               }
-       }
-
-       if (td->cur_depth) {
-               r = io_u_getevents(td, td->cur_depth, td->cur_depth, NULL);
-               if (r > 0) {
-                       icd.nr = r;
-                       ios_completed(td, &icd);
-               }
-       }
-}
-
-static int do_io_u_verify(struct thread_data *td, struct io_u **io_u)
-{
-       struct io_u *v_io_u = *io_u;
-       int ret = 0;
-
-       if (v_io_u) {
-               ret = verify_io_u(v_io_u);
-               put_io_u(td, v_io_u);
-               *io_u = NULL;
-       }
-
-       return ret;
-}
-
-static void do_verify(struct thread_data *td)
-{
-       struct timeval t;
-       struct io_u *io_u, *v_io_u = NULL;
-       struct io_completion_data icd;
-       int ret;
-
-       td_set_runstate(td, TD_VERIFYING);
-
-       do {
-               if (td->terminate)
-                       break;
-
-               gettimeofday(&t, NULL);
-               if (runtime_exceeded(td, &t))
-                       break;
-
-               io_u = __get_io_u(td);
-               if (!io_u)
-                       break;
-
-               if (get_next_verify(td, io_u)) {
-                       put_io_u(td, io_u);
-                       break;
-               }
-
-               if (td_io_prep(td, io_u)) {
-                       put_io_u(td, io_u);
-                       break;
-               }
-
-               ret = io_u_queue(td, io_u);
-               if (ret) {
-                       put_io_u(td, io_u);
-                       td_verror(td, ret);
-                       break;
-               }
-
-               /*
-                * we have one pending to verify, do that while
-                * we are doing io on the next one
-                */
-               if (do_io_u_verify(td, &v_io_u))
-                       break;
-
-               ret = io_u_getevents(td, 1, 1, NULL);
-               if (ret != 1) {
-                       if (ret < 0)
-                               td_verror(td, ret);
-                       break;
-               }
-
-               v_io_u = td->io_event(td, 0);
-               icd.nr = 1;
-               icd.error = 0;
-               io_completed(td, v_io_u, &icd);
-
-               if (icd.error) {
-                       td_verror(td, icd.error);
-                       put_io_u(td, v_io_u);
-                       v_io_u = NULL;
-                       break;
-               }
-
-               /*
-                * if we can't submit more io, we need to verify now
-                */
-               if (queue_full(td) && do_io_u_verify(td, &v_io_u))
-                       break;
-
-       } while (1);
-
-       do_io_u_verify(td, &v_io_u);
-
-       if (td->cur_depth)
-               cleanup_pending_aio(td);
-
-       td_set_runstate(td, TD_RUNNING);
-}
-
-static void do_io(struct thread_data *td)
-{
-       struct io_completion_data icd;
-       struct timeval s, e;
-       unsigned long usec;
-
-       while (td->this_io_bytes[td->ddir] < td->io_size) {
-               struct timespec ts = { .tv_sec = 0, .tv_nsec = 0};
-               struct timespec *timeout;
-               int ret, min_evts = 0;
-               struct io_u *io_u;
-
-               if (td->terminate)
-                       break;
-
-               io_u = get_io_u(td);
-               if (!io_u)
-                       break;
-
-               memcpy(&s, &io_u->start_time, sizeof(s));
-
-               ret = io_u_queue(td, io_u);
-               if (ret) {
-                       put_io_u(td, io_u);
-                       td_verror(td, ret);
-                       break;
-               }
-
-               add_slat_sample(td, io_u->ddir, mtime_since(&io_u->start_time, &io_u->issue_time));
-
-               if (td->cur_depth < td->iodepth) {
-                       timeout = &ts;
-                       min_evts = 0;
-               } else {
-                       timeout = NULL;
-                       min_evts = 1;
-               }
-
-               ret = io_u_getevents(td, min_evts, td->cur_depth, timeout);
-               if (ret < 0) {
-                       td_verror(td, ret);
-                       break;
-               } else if (!ret)
-                       continue;
-
-               icd.nr = ret;
-               ios_completed(td, &icd);
-               if (icd.error) {
-                       td_verror(td, icd.error);
-                       break;
-               }
-
-               /*
-                * the rate is batched for now, it should work for batches
-                * of completions except the very first one which may look
-                * a little bursty
-                */
-               gettimeofday(&e, NULL);
-               usec = utime_since(&s, &e);
-
-               rate_throttle(td, usec, icd.bytes_done[td->ddir]);
-
-               if (check_min_rate(td, &e)) {
-                       td_verror(td, ENOMEM);
-                       break;
-               }
-
-               if (runtime_exceeded(td, &e))
-                       break;
-
-               if (td->thinktime)
-                       usec_sleep(td, td->thinktime);
-
-               if (should_fsync(td) && td->fsync_blocks &&
-                   (td->io_blocks[DDIR_WRITE] % td->fsync_blocks) == 0)
-                       sync_td(td);
-       }
-
-       if (td->cur_depth)
-               cleanup_pending_aio(td);
-
-       if (should_fsync(td) && td->end_fsync)
-               sync_td(td);
-}
-
-static void cleanup_io(struct thread_data *td)
-{
-       if (td->io_cleanup)
-               td->io_cleanup(td);
-}
-
-static int init_io(struct thread_data *td)
-{
-       if (td->io_engine == FIO_SYNCIO)
-               return fio_syncio_init(td);
-       else if (td->io_engine == FIO_MMAPIO)
-               return fio_mmapio_init(td);
-       else if (td->io_engine == FIO_LIBAIO)
-               return fio_libaio_init(td);
-       else if (td->io_engine == FIO_POSIXAIO)
-               return fio_posixaio_init(td);
-       else if (td->io_engine == FIO_SGIO)
-               return fio_sgio_init(td);
-       else if (td->io_engine == FIO_SPLICEIO)
-               return fio_spliceio_init(td);
-       else {
-               fprintf(stderr, "bad io_engine %d\n", td->io_engine);
-               return 1;
-       }
-}
-
-static void cleanup_io_u(struct thread_data *td)
-{
-       struct list_head *entry, *n;
-       struct io_u *io_u;
-
-       list_for_each_safe(entry, n, &td->io_u_freelist) {
-               io_u = list_entry(entry, struct io_u, list);
+       list_for_each_safe(entry, n, &td->io_u_freelist) {
+               io_u = list_entry(entry, struct io_u, list);
 
                list_del(&io_u->list);
                free(io_u);
        }
-
-       if (td->mem_type == MEM_MALLOC)
-               free(td->orig_buffer);
-       else if (td->mem_type == MEM_SHM) {
-               struct shmid_ds sbuf;
-
-               shmdt(td->orig_buffer);
-               shmctl(td->shm_id, IPC_RMID, &sbuf);
-       } else if (td->mem_type == MEM_MMAP)
-               munmap(td->orig_buffer, td->orig_buffer_size);
-       else
-               fprintf(stderr, "Bad memory type %d\n", td->mem_type);
-
-       td->orig_buffer = NULL;
-}
-
-static int init_io_u(struct thread_data *td)
-{
-       struct io_u *io_u;
-       int i, max_units;
-       char *p;
-
-       if (td->io_engine & FIO_SYNCIO)
-               max_units = 1;
-       else
-               max_units = td->iodepth;
-
-       td->orig_buffer_size = td->max_bs * max_units + MASK;
-
-       if (td->mem_type == MEM_MALLOC)
-               td->orig_buffer = malloc(td->orig_buffer_size);
-       else if (td->mem_type == MEM_SHM) {
-               td->shm_id = shmget(IPC_PRIVATE, td->orig_buffer_size, IPC_CREAT | 0600);
-               if (td->shm_id < 0) {
-                       td_verror(td, errno);
-                       perror("shmget");
-                       return 1;
-               }
-
-               td->orig_buffer = shmat(td->shm_id, NULL, 0);
-               if (td->orig_buffer == (void *) -1) {
-                       td_verror(td, errno);
-                       perror("shmat");
-                       td->orig_buffer = NULL;
-                       return 1;
-               }
-       } else if (td->mem_type == MEM_MMAP) {
-               td->orig_buffer = mmap(NULL, td->orig_buffer_size, PROT_READ | PROT_WRITE, MAP_PRIVATE | OS_MAP_ANON, 0, 0);
-               if (td->orig_buffer == MAP_FAILED) {
-                       td_verror(td, errno);
-                       perror("mmap");
-                       td->orig_buffer = NULL;
-                       return 1;
-               }
-       }
-
-       p = ALIGN(td->orig_buffer);
-       for (i = 0; i < max_units; i++) {
-               io_u = malloc(sizeof(*io_u));
-               memset(io_u, 0, sizeof(*io_u));
-               INIT_LIST_HEAD(&io_u->list);
-
-               io_u->buf = p + td->max_bs * i;
-               io_u->index = i;
-               list_add(&io_u->list, &td->io_u_freelist);
-       }
-
-       return 0;
-}
-
-static int create_file(struct thread_data *td, unsigned long long size,
-                      int extend)
-{
-       unsigned long long left;
-       unsigned int bs;
-       int r, oflags;
-       char *b;
-
-       /*
-        * unless specifically asked for overwrite, let normal io extend it
-        */
-       if (td_write(td) && !td->overwrite)
-               return 0;
-
-       if (!size) {
-               fprintf(stderr, "Need size for create\n");
-               td_verror(td, EINVAL);
-               return 1;
-       }
-
-       if (!extend) {
-               oflags = O_CREAT | O_TRUNC;
-               printf("Client%d: Laying out IO file (%LuMiB)\n", td->thread_number, size >> 20);
-       } else {
-               oflags = O_APPEND;
-               printf("Client%d: Extending IO file (%Lu -> %LuMiB)\n", td->thread_number, (td->file_size - size) >> 20, td->file_size >> 20);
-       }
-
-       td->fd = open(td->file_name, O_WRONLY | oflags, 0644);
-       if (td->fd < 0) {
-               td_verror(td, errno);
-               return 1;
-       }
-
-       if (!extend && ftruncate(td->fd, td->file_size) == -1) {
-               td_verror(td, errno);
-               return 1;
-       }
-
-       td->io_size = td->file_size;
-       b = malloc(td->max_bs);
-       memset(b, 0, td->max_bs);
-
-       left = size;
-       while (left && !td->terminate) {
-               bs = td->max_bs;
-               if (bs > left)
-                       bs = left;
-
-               r = write(td->fd, b, bs);
-
-               if (r == (int) bs) {
-                       left -= bs;
-                       continue;
-               } else {
-                       if (r < 0)
-                               td_verror(td, errno);
-                       else
-                               td_verror(td, EIO);
-
-                       break;
-               }
-       }
-
-       if (td->terminate)
-               unlink(td->file_name);
-       else if (td->create_fsync)
-               fsync(td->fd);
-
-       close(td->fd);
-       td->fd = -1;
-       free(b);
-       return 0;
-}
-
-static int file_size(struct thread_data *td)
-{
-       struct stat st;
-
-       if (fstat(td->fd, &st) == -1) {
-               td_verror(td, errno);
-               return 1;
-       }
-
-       td->real_file_size = st.st_size;
-
-       if (!td->file_size || td->file_size > td->real_file_size)
-               td->file_size = td->real_file_size;
-
-       td->file_size -= td->file_offset;
-       return 0;
-}
-
-static int bdev_size(struct thread_data *td)
-{
-       unsigned long long bytes;
-       int r;
-
-       r = blockdev_size(td->fd, &bytes);
-       if (r) {
-               td_verror(td, r);
-               return 1;
-       }
-
-       td->real_file_size = bytes;
-
-       /*
-        * no extend possibilities, so limit size to device size if too large
-        */
-       if (!td->file_size || td->file_size > td->real_file_size)
-               td->file_size = td->real_file_size;
-
-       td->file_size -= td->file_offset;
-       return 0;
-}
-
-static int get_file_size(struct thread_data *td)
-{
-       int ret = 0;
-
-       if (td->filetype == FIO_TYPE_FILE)
-               ret = file_size(td);
-       else if (td->filetype == FIO_TYPE_BD)
-               ret = bdev_size(td);
-       else
-               td->real_file_size = -1;
-
-       if (ret)
-               return ret;
-
-       if (td->file_offset > td->real_file_size) {
-               fprintf(stderr, "Client%d: offset extends end (%Lu > %Lu)\n", td->thread_number, td->file_offset, td->real_file_size);
-               return 1;
-       }
-
-       td->io_size = td->file_size;
-       if (td->io_size == 0) {
-               fprintf(stderr, "Client%d: no io blocks\n", td->thread_number);
-               td_verror(td, EINVAL);
-               return 1;
-       }
-
-       if (!td->zone_size)
-               td->zone_size = td->io_size;
-
-       td->total_io_size = td->io_size * td->loops;
-       return 0;
-}
-
-static int setup_file_mmap(struct thread_data *td)
-{
-       int flags;
-
-       if (td_rw(td))
-               flags = PROT_READ | PROT_WRITE;
-       else if (td_write(td)) {
-               flags = PROT_WRITE;
-
-               if (td->verify != VERIFY_NONE)
-                       flags |= PROT_READ;
-       } else
-               flags = PROT_READ;
-
-       td->mmap = mmap(NULL, td->file_size, flags, MAP_SHARED, td->fd, td->file_offset);
-       if (td->mmap == MAP_FAILED) {
-               td->mmap = NULL;
-               td_verror(td, errno);
-               return 1;
-       }
-
-       if (td->invalidate_cache) {
-               if (madvise(td->mmap, td->file_size, MADV_DONTNEED) < 0) {
-                       td_verror(td, errno);
-                       return 1;
-               }
-       }
-
-       if (td->sequential) {
-               if (madvise(td->mmap, td->file_size, MADV_SEQUENTIAL) < 0) {
-                       td_verror(td, errno);
-                       return 1;
-               }
-       } else {
-               if (madvise(td->mmap, td->file_size, MADV_RANDOM) < 0) {
-                       td_verror(td, errno);
-                       return 1;
-               }
-       }
-
-       return 0;
-}
-
-static int setup_file_plain(struct thread_data *td)
-{
-       if (td->invalidate_cache) {
-               if (fadvise(td->fd, td->file_offset, td->file_size, POSIX_FADV_DONTNEED) < 0) {
-                       td_verror(td, errno);
-                       return 1;
-               }
-       }
-
-       if (td->sequential) {
-               if (fadvise(td->fd, td->file_offset, td->file_size, POSIX_FADV_SEQUENTIAL) < 0) {
-                       td_verror(td, errno);
-                       return 1;
-               }
-       } else {
-               if (fadvise(td->fd, td->file_offset, td->file_size, POSIX_FADV_RANDOM) < 0) {
-                       td_verror(td, errno);
-                       return 1;
-               }
-       }
-
-       return 0;
-}
-
-static int setup_file(struct thread_data *td)
-{
-       struct stat st;
-       int flags = 0;
-
-       if (stat(td->file_name, &st) == -1) {
-               if (errno != ENOENT) {
-                       td_verror(td, errno);
-                       return 1;
-               }
-               if (!td->create_file) {
-                       td_verror(td, ENOENT);
-                       return 1;
-               }
-               if (create_file(td, td->file_size, 0))
-                       return 1;
-       } else if (td->filetype == FIO_TYPE_FILE) {
-               if (st.st_size < (off_t) td->file_size) {
-                       if (create_file(td, td->file_size - st.st_size, 1))
-                               return 1;
-               }
-       }
-
-       if (td->odirect)
-               flags |= O_DIRECT;
-
-       if (td_write(td) || td_rw(td)) {
-               if (td->filetype == FIO_TYPE_FILE) {
-                       if (!td->overwrite)
-                               flags |= O_TRUNC;
-
-                       flags |= O_CREAT;
-               }
-               if (td->sync_io)
-                       flags |= O_SYNC;
-
-               flags |= O_RDWR;
-
-               td->fd = open(td->file_name, flags, 0600);
-       } else {
-               if (td->filetype == FIO_TYPE_CHAR)
-                       flags |= O_RDWR;
-               else
-                       flags |= O_RDONLY;
-
-               td->fd = open(td->file_name, flags);
-       }
-
-       if (td->fd == -1) {
-               td_verror(td, errno);
-               return 1;
-       }
-
-       if (get_file_size(td))
-               return 1;
-
-       if (td->io_engine != FIO_MMAPIO)
-               return setup_file_plain(td);
-       else
-               return setup_file_mmap(td);
-}
-
-static int check_dev_match(dev_t dev, char *path)
-{
-       unsigned int major, minor;
-       char line[256], *p;
-       FILE *f;
-
-       f = fopen(path, "r");
-       if (!f) {
-               perror("open path");
-               return 1;
-       }
-
-       p = fgets(line, sizeof(line), f);
-       if (!p) {
-               fclose(f);
-               return 1;
-       }
-
-       if (sscanf(p, "%u:%u", &major, &minor) != 2) {
-               fclose(f);
-               return 1;
-       }
-
-       if (((major << 8) | minor) == dev) {
-               fclose(f);
-               return 0;
-       }
-
-       fclose(f);
-       return 1;
-}
-
-static int find_block_dir(dev_t dev, char *path)
-{
-       struct dirent *dir;
-       struct stat st;
-       int found = 0;
-       DIR *D;
-
-       D = opendir(path);
-       if (!D)
-               return 0;
-
-       while ((dir = readdir(D)) != NULL) {
-               char full_path[256];
-
-               if (!strcmp(dir->d_name, ".") || !strcmp(dir->d_name, ".."))
-                       continue;
-               if (!strcmp(dir->d_name, "device"))
-                       continue;
-
-               sprintf(full_path, "%s/%s", path, dir->d_name);
-
-               if (!strcmp(dir->d_name, "dev")) {
-                       if (!check_dev_match(dev, full_path)) {
-                               found = 1;
-                               break;
-                       }
-               }
-
-               if (stat(full_path, &st) == -1) {
-                       perror("stat");
-                       break;
-               }
-
-               if (!S_ISDIR(st.st_mode) || S_ISLNK(st.st_mode))
-                       continue;
-
-               found = find_block_dir(dev, full_path);
-               if (found) {
-                       strcpy(path, full_path);
-                       break;
-               }
-       }
-
-       closedir(D);
-       return found;
-}
-
-static int get_io_ticks(struct disk_util *du, struct disk_util_stat *dus)
-{
-       unsigned in_flight;
-       char line[256];
-       FILE *f;
-       char *p;
-
-       f = fopen(du->path, "r");
-       if (!f)
-               return 1;
-
-       p = fgets(line, sizeof(line), f);
-       if (!p) {
-               fclose(f);
-               return 1;
-       }
-
-       if (sscanf(p, "%u %u %llu %u %u %u %llu %u %u %u %u\n", &dus->ios[0], &dus->merges[0], &dus->sectors[0], &dus->ticks[0], &dus->ios[1], &dus->merges[1], &dus->sectors[1], &dus->ticks[1], &in_flight, &dus->io_ticks, &dus->time_in_queue) != 11) {
-               fclose(f);
-               return 1;
-       }
-
-       fclose(f);
-       return 0;
-}
-
-static void update_io_tick_disk(struct disk_util *du)
-{
-       struct disk_util_stat __dus, *dus, *ldus;
-       struct timeval t;
-
-       if (get_io_ticks(du, &__dus))
-               return;
-
-       dus = &du->dus;
-       ldus = &du->last_dus;
-
-       dus->sectors[0] += (__dus.sectors[0] - ldus->sectors[0]);
-       dus->sectors[1] += (__dus.sectors[1] - ldus->sectors[1]);
-       dus->ios[0] += (__dus.ios[0] - ldus->ios[0]);
-       dus->ios[1] += (__dus.ios[1] - ldus->ios[1]);
-       dus->merges[0] += (__dus.merges[0] - ldus->merges[0]);
-       dus->merges[1] += (__dus.merges[1] - ldus->merges[1]);
-       dus->ticks[0] += (__dus.ticks[0] - ldus->ticks[0]);
-       dus->ticks[1] += (__dus.ticks[1] - ldus->ticks[1]);
-       dus->io_ticks += (__dus.io_ticks - ldus->io_ticks);
-       dus->time_in_queue += (__dus.time_in_queue - ldus->time_in_queue);
-
-       gettimeofday(&t, NULL);
-       du->msec += mtime_since(&du->time, &t);
-       memcpy(&du->time, &t, sizeof(t));
-       memcpy(ldus, &__dus, sizeof(__dus));
-}
-
-static void update_io_ticks(void)
-{
-       struct list_head *entry;
-       struct disk_util *du;
-
-       list_for_each(entry, &disk_list) {
-               du = list_entry(entry, struct disk_util, list);
-               update_io_tick_disk(du);
-       }
+
+       free_io_mem(td);
 }
 
-static int disk_util_exists(dev_t dev)
+static int init_io_u(struct thread_data *td)
 {
-       struct list_head *entry;
-       struct disk_util *du;
+       struct io_u *io_u;
+       int i, max_units;
+       char *p;
 
-       list_for_each(entry, &disk_list) {
-               du = list_entry(entry, struct disk_util, list);
+       if (td->io_ops->flags & FIO_CPUIO)
+               return 0;
 
-               if (du->dev == dev)
-                       return 1;
-       }
+       if (td->io_ops->flags & FIO_SYNCIO)
+               max_units = 1;
+       else
+               max_units = td->iodepth;
 
-       return 0;
-}
+       td->orig_buffer_size = td->max_bs * max_units + MASK;
 
-static void disk_util_add(dev_t dev, char *path)
-{
-       struct disk_util *du = malloc(sizeof(*du));
+       if (allocate_io_mem(td))
+               return 1;
 
-       memset(du, 0, sizeof(*du));
-       INIT_LIST_HEAD(&du->list);
-       sprintf(du->path, "%s/stat", path);
-       du->name = strdup(basename(path));
-       du->dev = dev;
+       p = ALIGN(td->orig_buffer);
+       for (i = 0; i < max_units; i++) {
+               io_u = malloc(sizeof(*io_u));
+               memset(io_u, 0, sizeof(*io_u));
+               INIT_LIST_HEAD(&io_u->list);
 
-       gettimeofday(&du->time, NULL);
-       get_io_ticks(du, &du->last_dus);
+               io_u->buf = p + td->max_bs * i;
+               io_u->index = i;
+               list_add(&io_u->list, &td->io_u_freelist);
+       }
 
-       list_add_tail(&du->list, &disk_list);
+       return 0;
 }
 
-static void init_disk_util(struct thread_data *td)
+static int switch_ioscheduler(struct thread_data *td)
 {
-       struct stat st;
-       char foo[256], tmp[256];
-       dev_t dev;
-       char *p;
+       char tmp[256], tmp2[128];
+       FILE *f;
+       int ret;
 
-       if (!td->do_disk_util)
-               return;
+       sprintf(tmp, "%s/queue/scheduler", td->sysfs_root);
 
-       if (!stat(td->file_name, &st)) {
-               if (S_ISBLK(st.st_mode))
-                       dev = st.st_rdev;
-               else
-                       dev = st.st_dev;
-       } else {
-               /*
-                * must be a file, open "." in that path
-                */
-               strcpy(foo, td->file_name);
-               p = dirname(foo);
-               if (stat(p, &st)) {
-                       perror("disk util stat");
-                       return;
-               }
+       f = fopen(tmp, "r+");
+       if (!f) {
+               td_verror(td, errno);
+               return 1;
+       }
 
-               dev = st.st_dev;
+       /*
+        * Set io scheduler.
+        */
+       ret = fwrite(td->ioscheduler, strlen(td->ioscheduler), 1, f);
+       if (ferror(f) || ret != 1) {
+               td_verror(td, errno);
+               fclose(f);
+               return 1;
        }
 
-       if (disk_util_exists(dev))
-               return;
-               
-       sprintf(foo, "/sys/block");
-       if (!find_block_dir(dev, foo))
-               return;
+       rewind(f);
 
        /*
-        * If there's a ../queue/ directory there, we are inside a partition.
-        * Check if that is the case and jump back. For loop/md/dm etc we
-        * are already in the right spot.
+        * Read back and check that the selected scheduler is now the default.
         */
-       sprintf(tmp, "%s/../queue", foo);
-       if (!stat(tmp, &st)) {
-               p = dirname(foo);
-               sprintf(tmp, "%s/queue", p);
-               if (stat(tmp, &st)) {
-                       fprintf(stderr, "unknown sysfs layout\n");
-                       return;
-               }
-               sprintf(foo, "%s", p);
+       ret = fread(tmp, 1, sizeof(tmp), f);
+       if (ferror(f) || ret < 0) {
+               td_verror(td, errno);
+               fclose(f);
+               return 1;
        }
 
-       disk_util_add(dev, foo);
-}
+       sprintf(tmp2, "[%s]", td->ioscheduler);
+       if (!strstr(tmp, tmp2)) {
+               log_err("fio: io scheduler %s not found\n", td->ioscheduler);
+               td_verror(td, EINVAL);
+               fclose(f);
+               return 1;
+       }
 
-static void disk_util_timer_arm(void)
-{
-       itimer.it_value.tv_sec = 0;
-       itimer.it_value.tv_usec = DISK_UTIL_MSEC * 1000;
-       setitimer(ITIMER_REAL, &itimer, NULL);
+       fclose(f);
+       return 0;
 }
 
 static void clear_io_state(struct thread_data *td)
 {
-       if (td->io_engine == FIO_SYNCIO)
-               lseek(td->fd, SEEK_SET, 0);
+       struct fio_file *f;
+       int i;
 
-       td->last_pos = 0;
        td->stat_io_bytes[0] = td->stat_io_bytes[1] = 0;
        td->this_io_bytes[0] = td->this_io_bytes[1] = 0;
        td->zone_bytes = 0;
 
-       if (td->file_map)
-               memset(td->file_map, 0, td->num_maps * sizeof(long));
-}
-
-static void update_rusage_stat(struct thread_data *td)
-{
-       if (!(td->runtime[0] + td->runtime[1]))
-               return;
-
-       getrusage(RUSAGE_SELF, &td->ru_end);
-
-       td->usr_time += mtime_since(&td->ru_start.ru_utime, &td->ru_end.ru_utime);
-       td->sys_time += mtime_since(&td->ru_start.ru_stime, &td->ru_end.ru_stime);
-       td->ctx += td->ru_end.ru_nvcsw + td->ru_end.ru_nivcsw - (td->ru_start.ru_nvcsw + td->ru_start.ru_nivcsw);
+       for_each_file(td, f, i) {
+               f->last_pos = 0;
+               if (td->io_ops->flags & FIO_SYNCIO)
+                       lseek(f->fd, SEEK_SET, 0);
 
-       
-       memcpy(&td->ru_start, &td->ru_end, sizeof(td->ru_end));
+               if (f->file_map)
+                       memset(f->file_map, 0, f->num_maps * sizeof(long));
+       }
 }
 
+/*
+ * Entry point for the thread based jobs. The process based jobs end up
+ * here as well, after a little setup.
+ */
 static void *thread_main(void *data)
 {
        struct thread_data *td = data;
-       int ret = 1;
 
        if (!td->use_thread)
                setsid();
@@ -1861,7 +573,7 @@ static void *thread_main(void *data)
                goto err;
        }
 
-       if (init_io(td))
+       if (td_io_init(td))
                goto err;
 
        if (init_iolog(td))
@@ -1874,19 +586,29 @@ static void *thread_main(void *data)
                }
        }
 
+       if (nice(td->nice) == -1) {
+               td_verror(td, errno);
+               goto err;
+       }
+
        if (init_random_state(td))
                goto err;
 
+       if (td->ioscheduler && switch_ioscheduler(td))
+               goto err;
+
        td_set_runstate(td, TD_INITIALIZED);
-       sem_post(&startup_sem);
-       sem_wait(&td->mutex);
-       ret = 0;
+       fio_sem_up(&startup_sem);
+       fio_sem_down(&td->mutex);
 
-       if (!td->create_serialize && setup_file(td))
+       if (!td->create_serialize && setup_files(td))
                goto err;
 
        gettimeofday(&td->epoch, NULL);
 
+       if (td->exec_prerun)
+               system(td->exec_prerun);
+
        while (td->loops--) {
                getrusage(RUSAGE_SELF, &td->ru_start);
                gettimeofday(&td->start, NULL);
@@ -1898,7 +620,10 @@ static void *thread_main(void *data)
                clear_io_state(td);
                prune_io_piece_log(td);
 
-               do_io(td);
+               if (td->io_ops->flags & FIO_CPUIO)
+                       do_cpuio(td);
+               else
+                       do_io(td);
 
                td->runtime[td->ddir] += mtime_since_now(&td->start);
                if (td_rw(td) && td->io_bytes[td->ddir ^ 1])
@@ -1931,28 +656,25 @@ static void *thread_main(void *data)
                finish_log(td, td->clat_log, "clat");
        if (td->write_iolog)
                write_iolog_close(td);
+       if (td->exec_postrun)
+               system(td->exec_postrun);
 
        if (exitall_on_terminate)
                terminate_threads(td->groupid);
 
 err:
-       if (td->fd != -1) {
-               close(td->fd);
-               td->fd = -1;
-       }
-       if (td->mmap)
-               munmap(td->mmap, td->file_size);
-       cleanup_io(td);
+       close_files(td);
+       close_ioengine(td);
        cleanup_io_u(td);
-       if (ret) {
-               sem_post(&startup_sem);
-               sem_wait(&td->mutex);
-       }
        td_set_runstate(td, TD_EXITED);
        return NULL;
 
 }
 
+/*
+ * We cannot pass the td data into a forked process, so attach the td and
+ * pass it to the thread worker.
+ */
 static void *fork_main(int shmid, int offset)
 {
        struct thread_data *td;
@@ -1970,295 +692,25 @@ static void *fork_main(int shmid, int offset)
        return NULL;
 }
 
-static int calc_lat(struct io_stat *is, unsigned long *min, unsigned long *max,
-                   double *mean, double *dev)
-{
-       double n;
-
-       if (is->samples == 0)
-               return 0;
-
-       *min = is->min_val;
-       *max = is->max_val;
-
-       n = (double) is->samples;
-       *mean = (double) is->val / n;
-       *dev = sqrt(((double) is->val_sq - (*mean * *mean) / n) / (n - 1));
-       if (!(*min + *max) && !(*mean + *dev))
-               return 0;
-
-       return 1;
-}
-
-static void show_ddir_status(struct thread_data *td, struct group_run_stats *rs,
-                            int ddir)
-{
-       char *ddir_str[] = { "read ", "write" };
-       unsigned long min, max;
-       unsigned long long bw;
-       double mean, dev;
-
-       if (!td->runtime[ddir])
-               return;
-
-       bw = td->io_bytes[ddir] / td->runtime[ddir];
-       printf("  %s: io=%6lluMiB, bw=%6lluKiB/s, runt=%6lumsec\n", ddir_str[ddir], td->io_bytes[ddir] >> 20, bw, td->runtime[ddir]);
-
-       if (calc_lat(&td->slat_stat[ddir], &min, &max, &mean, &dev))
-               printf("    slat (msec): min=%5lu, max=%5lu, avg=%5.02f, dev=%5.02f\n", min, max, mean, dev);
-
-       if (calc_lat(&td->clat_stat[ddir], &min, &max, &mean, &dev))
-               printf("    clat (msec): min=%5lu, max=%5lu, avg=%5.02f, dev=%5.02f\n", min, max, mean, dev);
-
-       if (calc_lat(&td->bw_stat[ddir], &min, &max, &mean, &dev)) {
-               double p_of_agg;
-
-               p_of_agg = mean * 100 / (double) rs->agg[ddir];
-               printf("    bw (KiB/s) : min=%5lu, max=%5lu, per=%3.2f%%, avg=%5.02f, dev=%5.02f\n", min, max, p_of_agg, mean, dev);
-       }
-}
-
-static void show_thread_status(struct thread_data *td,
-                              struct group_run_stats *rs)
-{
-       double usr_cpu, sys_cpu;
-
-       if (!(td->io_bytes[0] + td->io_bytes[1]) && !td->error)
-               return;
-
-       printf("Client%d (groupid=%d): err=%2d:\n", td->thread_number, td->groupid, td->error);
-
-       show_ddir_status(td, rs, td->ddir);
-       if (td->io_bytes[td->ddir ^ 1])
-               show_ddir_status(td, rs, td->ddir ^ 1);
-
-       if (td->runtime[0] + td->runtime[1]) {
-               double runt = td->runtime[0] + td->runtime[1];
-
-               usr_cpu = (double) td->usr_time * 100 / runt;
-               sys_cpu = (double) td->sys_time * 100 / runt;
-       } else {
-               usr_cpu = 0;
-               sys_cpu = 0;
-       }
-
-       printf("  cpu          : usr=%3.2f%%, sys=%3.2f%%, ctx=%lu\n", usr_cpu, sys_cpu, td->ctx);
-}
-
-static void check_str_update(struct thread_data *td)
-{
-       char c = run_str[td->thread_number - 1];
-
-       if (td->runstate == td->old_runstate)
-               return;
-
-       switch (td->runstate) {
-               case TD_REAPED:
-                       c = '_';
-                       break;
-               case TD_EXITED:
-                       c = 'E';
-                       break;
-               case TD_RUNNING:
-                       if (td_rw(td)) {
-                               if (td->sequential)
-                                       c = 'M';
-                               else
-                                       c = 'm';
-                       } else if (td_read(td)) {
-                               if (td->sequential)
-                                       c = 'R';
-                               else
-                                       c = 'r';
-                       } else {
-                               if (td->sequential)
-                                       c = 'W';
-                               else
-                                       c = 'w';
-                       }
-                       break;
-               case TD_VERIFYING:
-                       c = 'V';
-                       break;
-               case TD_CREATED:
-                       c = 'C';
-                       break;
-               case TD_INITIALIZED:
-                       c = 'I';
-                       break;
-               case TD_NOT_CREATED:
-                       c = 'P';
-                       break;
-               default:
-                       printf("state %d\n", td->runstate);
-       }
-
-       run_str[td->thread_number - 1] = c;
-       td->old_runstate = td->runstate;
-}
-
-static void eta_to_str(char *str, int eta_sec)
-{
-       unsigned int d, h, m, s;
-       static int always_d, always_h;
-
-       d = h = m = s = 0;
-
-       s = eta_sec % 60;
-       eta_sec /= 60;
-       m = eta_sec % 60;
-       eta_sec /= 60;
-       h = eta_sec % 24;
-       eta_sec /= 24;
-       d = eta_sec;
-
-       if (d || always_d) {
-               always_d = 1;
-               str += sprintf(str, "%02dd:", d);
-       }
-       if (h || always_h) {
-               always_h = 1;
-               str += sprintf(str, "%02dh:", h);
-       }
-
-       str += sprintf(str, "%02dm:", m);
-       str += sprintf(str, "%02ds", s);
-}
-
-static int thread_eta(struct thread_data *td, unsigned long elapsed)
-{
-       unsigned long long bytes_total, bytes_done;
-       unsigned int eta_sec = 0;
-
-       bytes_total = td->total_io_size;
-
-       /*
-        * if writing, bytes_total will be twice the size. If mixing,
-        * assume a 50/50 split and thus bytes_total will be 50% larger.
-        */
-       if (td->verify) {
-               if (td_rw(td))
-                       bytes_total = bytes_total * 3 / 2;
-               else
-                       bytes_total <<= 1;
-       }
-       if (td->zone_size && td->zone_skip)
-               bytes_total /= (td->zone_skip / td->zone_size);
-
-       if (td->runstate == TD_RUNNING || td->runstate == TD_VERIFYING) {
-               double perc;
-
-               bytes_done = td->io_bytes[DDIR_READ] + td->io_bytes[DDIR_WRITE];
-               perc = (double) bytes_done / (double) bytes_total;
-               if (perc > 1.0)
-                       perc = 1.0;
-
-               eta_sec = (elapsed * (1.0 / perc)) - elapsed;
-
-               if (td->timeout && eta_sec > (td->timeout - elapsed))
-                       eta_sec = td->timeout - elapsed;
-       } else if (td->runstate == TD_NOT_CREATED || td->runstate == TD_CREATED
-                       || td->runstate == TD_INITIALIZED) {
-               int t_eta = 0, r_eta = 0;
-
-               /*
-                * We can only guess - assume it'll run the full timeout
-                * if given, otherwise assume it'll run at the specified rate.
-                */
-               if (td->timeout)
-                       t_eta = td->timeout + td->start_delay - elapsed;
-               if (td->rate) {
-                       r_eta = (bytes_total / 1024) / td->rate;
-                       r_eta += td->start_delay - elapsed;
-               }
-
-               if (r_eta && t_eta)
-                       eta_sec = min(r_eta, t_eta);
-               else if (r_eta)
-                       eta_sec = r_eta;
-               else if (t_eta)
-                       eta_sec = t_eta;
-               else
-                       eta_sec = INT_MAX;
-       } else {
-               /*
-                * thread is already done
-                */
-               eta_sec = 0;
-       }
-
-       return eta_sec;
-}
-
-static void print_thread_status(void)
-{
-       unsigned long elapsed = time_since_now(&genesis);
-       int i, nr_running, t_rate, m_rate, *eta_secs, eta_sec;
-       char eta_str[32];
-       double perc = 0.0;
-
-       eta_secs = malloc(thread_number * sizeof(int));
-       memset(eta_secs, 0, thread_number * sizeof(int));
-
-       nr_running = t_rate = m_rate = 0;
-       for (i = 0; i < thread_number; i++) {
-               struct thread_data *td = &threads[i];
-
-               if (td->runstate == TD_RUNNING || td->runstate == TD_VERIFYING){
-                       nr_running++;
-                       t_rate += td->rate;
-                       m_rate += td->ratemin;
-               }
-
-               if (elapsed >= 3)
-                       eta_secs[i] = thread_eta(td, elapsed);
-               else
-                       eta_secs[i] = INT_MAX;
-
-               check_str_update(td);
-       }
-
-       if (exitall_on_terminate)
-               eta_sec = INT_MAX;
-       else
-               eta_sec = 0;
-
-       for (i = 0; i < thread_number; i++) {
-               if (exitall_on_terminate) {
-                       if (eta_secs[i] < eta_sec)
-                               eta_sec = eta_secs[i];
-               } else {
-                       if (eta_secs[i] > eta_sec)
-                               eta_sec = eta_secs[i];
-               }
-       }
-
-       if (eta_sec != INT_MAX && elapsed) {
-               perc = (double) elapsed / (double) (elapsed + eta_sec);
-               eta_to_str(eta_str, eta_sec);
-       }
-
-       printf("Threads now running (%d)", nr_running);
-       if (m_rate || t_rate)
-               printf(", commitrate %d/%dKiB/sec", t_rate, m_rate);
-       if (eta_sec != INT_MAX) {
-               perc *= 100.0;
-               printf(": [%s] [%3.2f%% done] [eta %s]", run_str, perc,eta_str);
-       }
-       printf("\r");
-       fflush(stdout);
-       free(eta_secs);
-}
-
+/*
+ * Run over the job map and reap the threads that have exited, if any.
+ */
 static void reap_threads(int *nr_running, int *t_rate, int *m_rate)
 {
-       int i;
+       struct thread_data *td;
+       int i, cputhreads;
 
        /*
         * reap exited threads (TD_EXITED -> TD_REAPED)
         */
-       for (i = 0; i < thread_number; i++) {
-               struct thread_data *td = &threads[i];
+       cputhreads = 0;
+       for_each_td(td, i) {
+               /*
+                * ->io_ops is NULL for a thread that has closed its
+                * io engine
+                */
+               if (td->io_ops && td->io_ops->flags & FIO_CPUIO)
+                       cputhreads++;
 
                if (td->runstate != TD_EXITED)
                        continue;
@@ -2277,65 +729,27 @@ static void reap_threads(int *nr_running, int *t_rate, int *m_rate)
                (*m_rate) -= td->ratemin;
                (*t_rate) -= td->rate;
        }
-}
-
-static void fio_unpin_memory(void *pinned)
-{
-       if (pinned) {
-               if (munlock(pinned, mlock_size) < 0)
-                       perror("munlock");
-               munmap(pinned, mlock_size);
-       }
-}
-
-static void *fio_pin_memory(void)
-{
-       long pagesize, pages;
-       void *ptr;
-
-       if (!mlock_size)
-               return NULL;
-
-       /*
-        * Don't allow mlock of more than real_mem-128MB
-        */
-       pagesize = sysconf(_SC_PAGESIZE);
-       pages = sysconf(_SC_PHYS_PAGES);
-       if (pages != -1 && pagesize != -1) {
-               unsigned long long real_mem = pages * pagesize;
-
-               if ((mlock_size + 128 * 1024 * 1024) > real_mem) {
-                       mlock_size = real_mem - 128 * 1024 * 1024;
-                       printf("fio: limiting mlocked memory to %lluMiB\n",
-                                                       mlock_size >> 20);
-               }
-       }
-
-       ptr = mmap(NULL, mlock_size, PROT_READ | PROT_WRITE, MAP_PRIVATE | OS_MAP_ANON, 0, 0);
-       if (!ptr) {
-               perror("malloc locked mem");
-               return NULL;
-       }
-       if (mlock(ptr, mlock_size) < 0) {
-               munmap(ptr, mlock_size);
-               perror("mlock");
-               return NULL;
-       }
 
-       return ptr;
+       if (*nr_running == cputhreads)
+               terminate_threads(TERMINATE_ALL);
 }
 
+/*
+ * Main function for kicking off and reaping jobs, as needed.
+ */
 static void run_threads(void)
 {
        struct thread_data *td;
        unsigned long spent;
        int i, todo, nr_running, m_rate, t_rate, nr_started;
-       void *mlocked_mem;
 
-       mlocked_mem = fio_pin_memory();
+       if (fio_pin_memory())
+               return;
 
-       printf("Starting %d thread%s\n", thread_number, thread_number > 1 ? "s" : "");
-       fflush(stdout);
+       if (!terse_output) {
+               printf("Starting %d thread%s\n", thread_number, thread_number > 1 ? "s" : "");
+               fflush(stdout);
+       }
 
        signal(SIGINT, sig_handler);
        signal(SIGALRM, sig_handler);
@@ -2345,10 +759,8 @@ static void run_threads(void)
        nr_started = 0;
        m_rate = t_rate = 0;
 
-       for (i = 0; i < thread_number; i++) {
-               td = &threads[i];
-
-               run_str[td->thread_number - 1] = 'P';
+       for_each_td(td, i) {
+               print_status_init(td->thread_number - 1);
 
                init_disk_util(td);
 
@@ -2360,13 +772,13 @@ static void run_threads(void)
                 * we don't want X number of threads getting their
                 * client data interspersed on disk
                 */
-               if (setup_file(td)) {
+               if (setup_files(td)) {
                        td_set_runstate(td, TD_REAPED);
                        todo--;
                }
        }
 
-       gettimeofday(&genesis, NULL);
+       time_init();
 
        while (todo) {
                struct thread_data *map[MAX_JOBS];
@@ -2376,9 +788,7 @@ static void run_threads(void)
                /*
                 * create threads (TD_NOT_CREATED -> TD_CREATED)
                 */
-               for (i = 0; i < thread_number; i++) {
-                       td = &threads[i];
-
+               for_each_td(td, i) {
                        if (td->runstate != TD_NOT_CREATED)
                                continue;
 
@@ -2392,7 +802,7 @@ static void run_threads(void)
                        }
 
                        if (td->start_delay) {
-                               spent = mtime_since_now(&genesis);
+                               spent = mtime_since_genesis();
 
                                if (td->start_delay * 1000 > spent)
                                        continue;
@@ -2407,7 +817,7 @@ static void run_threads(void)
                         */
                        td_set_runstate(td, TD_CREATED);
                        map[this_jobs++] = td;
-                       sem_init(&startup_sem, 0, 1);
+                       fio_sem_init(&startup_sem, 1);
                        nr_started++;
 
                        if (td->use_thread) {
@@ -2417,7 +827,7 @@ static void run_threads(void)
                                }
                        } else {
                                if (fork())
-                                       sem_wait(&startup_sem);
+                                       fio_sem_down(&startup_sem);
                                else {
                                        fork_main(shm_id, i);
                                        exit(0);
@@ -2429,7 +839,6 @@ static void run_threads(void)
                 * Wait for the started threads to transition to
                 * TD_INITIALIZED.
                 */
-               printf("fio: Waiting for threads to initialize...\n");
                gettimeofday(&this_start, NULL);
                left = this_jobs;
                while (left) {
@@ -2442,17 +851,20 @@ static void run_threads(void)
                                td = map[i];
                                if (!td)
                                        continue;
-                               if (td->runstate == TD_INITIALIZED ||
-                                   td->runstate >= TD_EXITED) {
+                               if (td->runstate == TD_INITIALIZED) {
                                        map[i] = NULL;
                                        left--;
-                                       continue;
+                               } else if (td->runstate >= TD_EXITED) {
+                                       map[i] = NULL;
+                                       left--;
+                                       todo--;
+                                       nr_running++; /* work-around... */
                                }
                        }
                }
 
                if (left) {
-                       fprintf(stderr, "fio: %d jobs failed to start\n", left);
+                       log_err("fio: %d jobs failed to start\n", left);
                        for (i = 0; i < this_jobs; i++) {
                                td = map[i];
                                if (!td)
@@ -2463,12 +875,9 @@ static void run_threads(void)
                }
 
                /*
-                * start created threads (TD_INITIALIZED -> TD_RUNNING)
+                * start created threads (TD_INITIALIZED -> TD_RUNNING).
                 */
-               printf("fio: Go for launch\n");
-               for (i = 0; i < thread_number; i++) {
-                       td = &threads[i];
-
+               for_each_td(td, i) {
                        if (td->runstate != TD_INITIALIZED)
                                continue;
 
@@ -2478,7 +887,7 @@ static void run_threads(void)
                        m_rate += td->ratemin;
                        t_rate += td->rate;
                        todo--;
-                       sem_post(&td->mutex);
+                       fio_sem_up(&td->mutex);
                }
 
                reap_threads(&nr_running, &t_rate, &m_rate);
@@ -2493,121 +902,7 @@ static void run_threads(void)
        }
 
        update_io_ticks();
-       fio_unpin_memory(mlocked_mem);
-}
-
-static void show_group_stats(struct group_run_stats *rs, int id)
-{
-       printf("\nRun status group %d (all jobs):\n", id);
-
-       if (rs->max_run[DDIR_READ])
-               printf("   READ: io=%lluMiB, aggrb=%llu, minb=%llu, maxb=%llu, mint=%llumsec, maxt=%llumsec\n", rs->io_mb[0], rs->agg[0], rs->min_bw[0], rs->max_bw[0], rs->min_run[0], rs->max_run[0]);
-       if (rs->max_run[DDIR_WRITE])
-               printf("  WRITE: io=%lluMiB, aggrb=%llu, minb=%llu, maxb=%llu, mint=%llumsec, maxt=%llumsec\n", rs->io_mb[1], rs->agg[1], rs->min_bw[1], rs->max_bw[1], rs->min_run[1], rs->max_run[1]);
-}
-
-static void show_disk_util(void)
-{
-       struct disk_util_stat *dus;
-       struct list_head *entry;
-       struct disk_util *du;
-       double util;
-
-       printf("\nDisk stats (read/write):\n");
-
-       list_for_each(entry, &disk_list) {
-               du = list_entry(entry, struct disk_util, list);
-               dus = &du->dus;
-
-               util = (double) 100 * du->dus.io_ticks / (double) du->msec;
-               if (util > 100.0)
-                       util = 100.0;
-
-               printf("  %s: ios=%u/%u, merge=%u/%u, ticks=%u/%u, in_queue=%u, util=%3.2f%%\n", du->name, dus->ios[0], dus->ios[1], dus->merges[0], dus->merges[1], dus->ticks[0], dus->ticks[1], dus->time_in_queue, util);
-       }
-}
-
-static void show_run_stats(void)
-{
-       struct group_run_stats *runstats, *rs;
-       struct thread_data *td;
-       int i;
-
-       runstats = malloc(sizeof(struct group_run_stats) * (groupid + 1));
-
-       for (i = 0; i < groupid + 1; i++) {
-               rs = &runstats[i];
-
-               memset(rs, 0, sizeof(*rs));
-               rs->min_bw[0] = rs->min_run[0] = ~0UL;
-               rs->min_bw[1] = rs->min_run[1] = ~0UL;
-       }
-
-       for (i = 0; i < thread_number; i++) {
-               unsigned long long rbw, wbw;
-
-               td = &threads[i];
-
-               if (td->error) {
-                       printf("Client%d: %s\n", td->thread_number, td->verror);
-                       continue;
-               }
-
-               rs = &runstats[td->groupid];
-
-               if (td->runtime[0] < rs->min_run[0] || !rs->min_run[0])
-                       rs->min_run[0] = td->runtime[0];
-               if (td->runtime[0] > rs->max_run[0])
-                       rs->max_run[0] = td->runtime[0];
-               if (td->runtime[1] < rs->min_run[1] || !rs->min_run[1])
-                       rs->min_run[1] = td->runtime[1];
-               if (td->runtime[1] > rs->max_run[1])
-                       rs->max_run[1] = td->runtime[1];
-
-               rbw = wbw = 0;
-               if (td->runtime[0])
-                       rbw = td->io_bytes[0] / (unsigned long long) td->runtime[0];
-               if (td->runtime[1])
-                       wbw = td->io_bytes[1] / (unsigned long long) td->runtime[1];
-
-               if (rbw < rs->min_bw[0])
-                       rs->min_bw[0] = rbw;
-               if (wbw < rs->min_bw[1])
-                       rs->min_bw[1] = wbw;
-               if (rbw > rs->max_bw[0])
-                       rs->max_bw[0] = rbw;
-               if (wbw > rs->max_bw[1])
-                       rs->max_bw[1] = wbw;
-
-               rs->io_mb[0] += td->io_bytes[0] >> 20;
-               rs->io_mb[1] += td->io_bytes[1] >> 20;
-       }
-
-       for (i = 0; i < groupid + 1; i++) {
-               rs = &runstats[i];
-
-               if (rs->max_run[0])
-                       rs->agg[0] = (rs->io_mb[0]*1024*1000) / rs->max_run[0];
-               if (rs->max_run[1])
-                       rs->agg[1] = (rs->io_mb[1]*1024*1000) / rs->max_run[1];
-       }
-
-       /*
-        * don't overwrite last signal output
-        */
-       printf("\n");
-
-       for (i = 0; i < thread_number; i++) {
-               td = &threads[i];
-               rs = &runstats[td->groupid];
-
-               show_thread_status(td, rs);
-       }
-
-       for (i = 0; i < groupid + 1; i++)
-               show_group_stats(&runstats[i], i);
-
-       show_disk_util();
+       fio_unpin_memory();
 }
 
 int main(int argc, char *argv[])
@@ -2616,7 +911,7 @@ int main(int argc, char *argv[])
                return 1;
 
        if (!thread_number) {
-               printf("Nothing to do\n");
+               log_err("Nothing to do\n");
                return 1;
        }