Add total latency log
[fio.git] / io_u.c
diff --git a/io_u.c b/io_u.c
index a4bf0c08581861076ccd31a8f720dcd861fbfe01..b2b72302d58ee8b0790fb123dd882c508bbd9360 100644 (file)
--- a/io_u.c
+++ b/io_u.c
@@ -8,6 +8,7 @@
 #include "fio.h"
 #include "hash.h"
 #include "verify.h"
 #include "fio.h"
 #include "hash.h"
 #include "verify.h"
+#include "lib/rand.h"
 
 struct io_completion_data {
        int nr;                         /* input */
 
 struct io_completion_data {
        int nr;                         /* input */
@@ -53,11 +54,8 @@ static void mark_random_map(struct thread_data *td, struct io_u *io_u)
                 * If we have a mixed random workload, we may
                 * encounter blocks we already did IO to.
                 */
                 * If we have a mixed random workload, we may
                 * encounter blocks we already did IO to.
                 */
-               if ((td->o.ddir_nr == 1) && !random_map_free(f, block)) {
-                       if (!blocks)
-                               blocks = 1;
+               if ((td->o.ddir_nr == 1) && !random_map_free(f, block))
                        break;
                        break;
-               }
 
                idx = RAND_MAP_IDX(f, block);
                bit = RAND_MAP_BIT(f, block);
 
                idx = RAND_MAP_IDX(f, block);
                bit = RAND_MAP_BIT(f, block);
@@ -68,10 +66,20 @@ static void mark_random_map(struct thread_data *td, struct io_u *io_u)
                if (this_blocks + bit > BLOCKS_PER_MAP)
                        this_blocks = BLOCKS_PER_MAP - bit;
 
                if (this_blocks + bit > BLOCKS_PER_MAP)
                        this_blocks = BLOCKS_PER_MAP - bit;
 
-               if (this_blocks == BLOCKS_PER_MAP)
-                       mask = -1U;
-               else
-                       mask = ((1U << this_blocks) - 1) << bit;
+               do {
+                       if (this_blocks == BLOCKS_PER_MAP)
+                               mask = -1U;
+                       else
+                               mask = ((1U << this_blocks) - 1) << bit;
+       
+                       if (!(f->file_map[idx] & mask))
+                               break;
+
+                       this_blocks--;
+               } while (this_blocks);
+
+               if (!this_blocks)
+                       break;
 
                f->file_map[idx] |= mask;
                nr_blocks -= this_blocks;
 
                f->file_map[idx] |= mask;
                nr_blocks -= this_blocks;
@@ -888,14 +896,6 @@ again:
                io_u->end_io = NULL;
        }
 
                io_u->end_io = NULL;
        }
 
-       /*
-        * We ran out, wait for async verify threads to finish and return one
-        */
-       if (!io_u && td->o.verify_async) {
-               pthread_cond_wait(&td->free_cond, &td->io_u_lock);
-               goto again;
-       }
-
        if (io_u) {
                assert(io_u->flags & IO_U_F_FREE);
                io_u->flags &= ~(IO_U_F_FREE | IO_U_F_FREE_DEF);
        if (io_u) {
                assert(io_u->flags & IO_U_F_FREE);
                io_u->flags &= ~(IO_U_F_FREE | IO_U_F_FREE_DEF);
@@ -905,6 +905,13 @@ again:
                flist_add(&io_u->list, &td->io_u_busylist);
                td->cur_depth++;
                io_u->flags |= IO_U_F_IN_CUR_DEPTH;
                flist_add(&io_u->list, &td->io_u_busylist);
                td->cur_depth++;
                io_u->flags |= IO_U_F_IN_CUR_DEPTH;
+       } else if (td->o.verify_async) {
+               /*
+                * We ran out, wait for async verify threads to finish and
+                * return one
+                */
+               pthread_cond_wait(&td->free_cond, &td->io_u_lock);
+               goto again;
        }
 
        td_io_u_unlock(td);
        }
 
        td_io_u_unlock(td);
@@ -926,6 +933,24 @@ struct io_u *get_io_u(struct thread_data *td)
                return NULL;
        }
 
                return NULL;
        }
 
+       if (td->o.verify_backlog && td->io_hist_len) {
+               int get_verify = 0;
+
+               if (td->verify_batch) {
+                       td->verify_batch--;
+                       get_verify = 1;
+               } else if (!(td->io_hist_len % td->o.verify_backlog) &&
+                        td->last_ddir != DDIR_READ) {
+                       td->verify_batch = td->o.verify_batch;
+                       if (!td->verify_batch)
+                               td->verify_batch = td->o.verify_backlog;
+                       get_verify = 1;
+               }
+
+               if (get_verify && !get_next_verify(td, io_u))
+                       goto out;
+       }
+
        /*
         * from a requeue, io_u already setup
         */
        /*
         * from a requeue, io_u already setup
         */
@@ -1024,6 +1049,7 @@ static void io_completed(struct thread_data *td, struct io_u *io_u,
        }
 
        td->last_was_sync = 0;
        }
 
        td->last_was_sync = 0;
+       td->last_ddir = io_u->ddir;
 
        if (!io_u->error) {
                unsigned int bytes = io_u->buflen - io_u->resid;
 
        if (!io_u->error) {
                unsigned int bytes = io_u->buflen - io_u->resid;
@@ -1053,7 +1079,13 @@ static void io_completed(struct thread_data *td, struct io_u *io_u,
                        if (!td->o.disable_clat || !td->o.disable_bw)
                                lusec = utime_since(&io_u->issue_time,
                                                        &icd->time);
                        if (!td->o.disable_clat || !td->o.disable_bw)
                                lusec = utime_since(&io_u->issue_time,
                                                        &icd->time);
+                       if (!td->o.disable_lat) {
+                               unsigned long tusec;
 
 
+                               tusec = utime_since(&io_u->start_time,
+                                                       &icd->time);
+                               add_lat_sample(td, idx, tusec, bytes);
+                       }
                        if (!td->o.disable_clat) {
                                add_clat_sample(td, idx, lusec, bytes);
                                io_u_mark_latency(td, lusec);
                        if (!td->o.disable_clat) {
                                add_clat_sample(td, idx, lusec, bytes);
                                io_u_mark_latency(td, lusec);
@@ -1214,13 +1246,8 @@ void io_u_queued(struct thread_data *td, struct io_u *io_u)
 void io_u_fill_buffer(struct thread_data *td, struct io_u *io_u,
                      unsigned int max_bs)
 {
 void io_u_fill_buffer(struct thread_data *td, struct io_u *io_u,
                      unsigned int max_bs)
 {
-       long *ptr = io_u->buf;
-
-       if (!td->o.zero_buffers) {
-               while ((void *) ptr - io_u->buf < max_bs) {
-                       *ptr = rand() * GOLDEN_RATIO_PRIME;
-                       ptr++;
-               }
-       } else
-               memset(ptr, 0, max_bs);
+       if (!td->o.zero_buffers)
+               fill_random_buf(io_u->buf, max_bs);
+       else
+               memset(io_u->buf, 0, max_bs);
 }
 }