#include <string.h>
#include <signal.h>
#include <time.h>
-#include <assert.h>
#include "fio.h"
#include "os.h"
static void mark_random_map(struct thread_data *td, struct fio_file *f,
struct io_u *io_u)
{
- unsigned long long block = io_u->offset / (unsigned long long) td->min_bs;
- unsigned int blocks = 0;
+ unsigned int min_bs = td->rw_min_bs;
+ unsigned long long block;
+ unsigned int blocks;
+ unsigned int nr_blocks;
- while (blocks < (io_u->buflen / td->min_bs)) {
+ block = io_u->offset / (unsigned long long) min_bs;
+ blocks = 0;
+ nr_blocks = (io_u->buflen + min_bs - 1) / min_bs;
+
+ while (blocks < nr_blocks) {
unsigned int idx, bit;
if (!random_map_free(td, f, block))
idx = RAND_MAP_IDX(td, f, block);
bit = RAND_MAP_BIT(td, f, block);
- assert(idx < f->num_maps);
+ fio_assert(td, idx < f->num_maps);
f->file_map[idx] |= (1UL << bit);
block++;
blocks++;
}
- if ((blocks * td->min_bs) < io_u->buflen)
- io_u->buflen = blocks * td->min_bs;
+ if ((blocks * min_bs) < io_u->buflen)
+ io_u->buflen = blocks * min_bs;
}
/*
{
int i;
- *b = 0;
- i = 0;
- while ((*b) * td->min_bs < f->file_size) {
+ i = f->last_free_lookup;
+ *b = (i * BLOCKS_PER_MAP);
+ while ((*b) * td->rw_min_bs < f->real_file_size) {
if (f->file_map[i] != -1UL) {
*b += ffz(f->file_map[i]);
+ f->last_free_lookup = i;
return 0;
}
* the last io issued.
*/
static int get_next_offset(struct thread_data *td, struct fio_file *f,
- unsigned long long *offset)
+ struct io_u *io_u)
{
+ const int ddir = io_u->ddir;
unsigned long long b, rb;
long r;
if (!td->sequential) {
- unsigned long long max_blocks = td->io_size / td->min_bs;
- int loops = 50;
+ unsigned long long max_blocks = f->file_size / td->min_bs[ddir];
+ int loops = 5;
do {
r = os_random_long(&td->random_state);
b = ((max_blocks - 1) * r / (unsigned long long) (RAND_MAX+1.0));
- rb = b + (f->file_offset / td->min_bs);
+ if (td->norandommap)
+ break;
+ rb = b + (f->file_offset / td->min_bs[ddir]);
loops--;
} while (!random_map_free(td, f, rb) && loops);
- if (!loops) {
- if (get_next_free_block(td, f, &b))
- return 1;
- }
+ /*
+ * if we failed to retrieve a truly random offset within
+ * the loops assigned, see if there are free ones left at all
+ */
+ if (!loops && get_next_free_block(td, f, &b))
+ return 1;
} else
- b = f->last_pos / td->min_bs;
+ b = f->last_pos / td->min_bs[ddir];
- *offset = (b * td->min_bs) + f->file_offset;
- if (*offset > f->file_size)
+ io_u->offset = (b * td->min_bs[ddir]) + f->file_offset;
+ if (io_u->offset >= f->real_file_size)
return 1;
return 0;
}
-static unsigned int get_next_buflen(struct thread_data *td)
+static unsigned int get_next_buflen(struct thread_data *td, struct fio_file *f,
+ struct io_u *io_u)
{
+ const int ddir = io_u->ddir;
unsigned int buflen;
long r;
- if (td->min_bs == td->max_bs)
- buflen = td->min_bs;
+ if (td->min_bs[ddir] == td->max_bs[ddir])
+ buflen = td->min_bs[ddir];
else {
r = os_random_long(&td->bsrange_state);
- buflen = (1 + (double) (td->max_bs - 1) * r / (RAND_MAX + 1.0));
- buflen = (buflen + td->min_bs - 1) & ~(td->min_bs - 1);
+ buflen = (unsigned int) (1 + (double) (td->max_bs[ddir] - 1) * r / (RAND_MAX + 1.0));
+ if (!td->bs_unaligned)
+ buflen = (buflen + td->min_bs[ddir] - 1) & ~(td->min_bs[ddir] - 1);
}
- if (buflen > td->io_size - td->this_io_bytes[td->ddir]) {
- /*
- * if using direct/raw io, we may not be able to
- * shrink the size. so just fail it.
- */
- if (td->io_ops->flags & FIO_RAWIO)
+ while (buflen + io_u->offset > f->real_file_size) {
+ if (buflen == td->min_bs[ddir])
return 0;
- buflen = td->io_size - td->this_io_bytes[td->ddir];
+ buflen = td->min_bs[ddir];
}
return buflen;
* mixed read/write workload, check the rwmix cycle and switch if
* necessary.
*/
-static int get_rw_ddir(struct thread_data *td)
+static enum fio_ddir get_rw_ddir(struct thread_data *td)
{
if (td_rw(td)) {
struct timeval now;
unsigned long elapsed;
- gettimeofday(&now, NULL);
+ fio_gettime(&now, NULL);
elapsed = mtime_since_now(&td->rwmix_switch);
/*
return read_iolog_get(td, io_u);
/*
- * No log, let the seq/rand engine retrieve the next position.
+ * see if it's time to sync
*/
- if (!get_next_offset(td, f, &io_u->offset)) {
- io_u->buflen = get_next_buflen(td);
+ if (td->fsync_blocks && !(td->io_blocks[DDIR_WRITE] % td->fsync_blocks)
+ && should_fsync(td)) {
+ io_u->ddir = DDIR_SYNC;
+ io_u->file = f;
+ return 0;
+ }
- if (io_u->buflen) {
- io_u->ddir = get_rw_ddir(td);
+ 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);
+ /*
+ * No log, let the seq/rand engine retrieve the next buflen and
+ * position.
+ */
+ if (get_next_offset(td, f, io_u))
+ return 1;
- io_u->file = f;
- return 0;
- }
+ io_u->buflen = get_next_buflen(td, f, io_u);
+ if (!io_u->buflen)
+ return 1;
+
+ /*
+ * mark entry before potentially trimming io_u
+ */
+ if (!td->read_iolog && !td->sequential && !td->norandommap)
+ mark_random_map(td, f, io_u);
+
+ /*
+ * If using a write iolog, store this entry.
+ */
+ if (td->write_iolog_file)
+ write_iolog_put(td, io_u);
+
+ io_u->file = f;
+ return 0;
+}
+
+static void io_u_mark_depth(struct thread_data *td)
+{
+ int index = 0;
+
+ switch (td->cur_depth) {
+ default:
+ index++;
+ case 32 ... 63:
+ index++;
+ case 16 ... 31:
+ index++;
+ case 8 ... 15:
+ index++;
+ case 4 ... 7:
+ index++;
+ case 2 ... 3:
+ index++;
+ case 1:
+ break;
}
- return 1;
+ td->io_u_map[index]++;
+ td->total_io_u++;
+}
+
+static void io_u_mark_latency(struct thread_data *td, unsigned long msec)
+{
+ int index = 0;
+
+ switch (msec) {
+ default:
+ index++;
+ case 1024 ... 2047:
+ index++;
+ case 512 ... 1023:
+ index++;
+ case 256 ... 511:
+ index++;
+ case 128 ... 255:
+ index++;
+ case 64 ... 127:
+ index++;
+ case 32 ... 63:
+ index++;
+ case 16 ... 31:
+ index++;
+ case 8 ... 15:
+ index++;
+ case 4 ... 7:
+ index++;
+ case 2 ... 3:
+ index++;
+ case 0 ... 1:
+ break;
+ }
+
+ td->io_u_lat[index]++;
}
struct io_u *__get_io_u(struct thread_data *td)
if (!queue_full(td)) {
io_u = list_entry(td->io_u_freelist.next, struct io_u, list);
+ io_u->buflen = 0;
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++;
+ io_u_mark_depth(td);
}
return io_u;
return NULL;
}
- if (io_u->buflen + io_u->offset > f->file_size) {
+ if (io_u->buflen + io_u->offset > f->real_file_size) {
if (td->io_ops->flags & FIO_RAWIO) {
put_io_u(td, io_u);
return NULL;
}
- io_u->buflen = f->file_size - io_u->offset;
+ io_u->buflen = f->real_file_size - io_u->offset;
}
- if (!io_u->buflen) {
- put_io_u(td, io_u);
- return NULL;
- }
+ if (io_u->ddir != DDIR_SYNC) {
+ if (!io_u->buflen) {
+ put_io_u(td, io_u);
+ return NULL;
+ }
- if (!td->read_iolog && !td->sequential)
- mark_random_map(td, f, io_u);
+ f->last_pos = io_u->offset + io_u->buflen;
- f->last_pos += io_u->buflen;
+ if (td->verify != VERIFY_NONE)
+ populate_verify_io_u(td, io_u);
+ }
- if (td->verify != VERIFY_NONE)
- populate_verify_io_u(td, io_u);
+ /*
+ * Set io data pointers.
+ */
+ io_u->xfer_buf = io_u->buf;
+ io_u->xfer_buflen = io_u->buflen;
if (td_io_prep(td, io_u)) {
put_io_u(td, io_u);
return NULL;
}
- gettimeofday(&io_u->start_time, NULL);
+ fio_gettime(&io_u->start_time, NULL);
return io_u;
}
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->ddir == DDIR_SYNC) {
+ td->last_was_sync = 1;
+ return;
+ }
+
+ td->last_was_sync = 0;
if (!io_u->error) {
unsigned int bytes = io_u->buflen - io_u->resid;
- const int idx = io_u->ddir;
+ const enum fio_ddir 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);
+ io_u->file->last_completed_pos = io_u->offset + io_u->buflen;
+
+ msec = mtime_since(&io_u->issue_time, &icd->time);
add_clat_sample(td, idx, msec);
- add_bw_sample(td, idx);
+ add_bw_sample(td, idx, &icd->time);
+ io_u_mark_latency(td, msec);
if ((td_rw(td) || td_write(td)) && idx == DDIR_WRITE)
log_io_piece(td, io_u);
icd->error = io_u->error;
}
+void init_icd(struct io_completion_data *icd)
+{
+ fio_gettime(&icd->time, NULL);
+
+ icd->error = 0;
+ icd->bytes_done[0] = icd->bytes_done[1] = 0;
+}
+
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;
+ init_icd(icd);
for (i = 0; i < icd->nr; i++) {
io_u = td->io_ops->event(td, i);
put_io_u(td, io_u);
}
}
-
-