stat: report clat stats on a per priority granularity
[fio.git] / stat.c
CommitLineData
3c39a379
JA
1#include <stdio.h>
2#include <string.h>
3#include <sys/time.h>
5c4e1dbc 4#include <sys/stat.h>
3c39a379
JA
5#include <math.h>
6
7#include "fio.h"
7c9b1bce 8#include "diskutil.h"
c7c6cb4c 9#include "lib/ieee754.h"
cc372b17 10#include "json.h"
44404c5a 11#include "lib/getrusage.h"
f2a2ce0e 12#include "idletime.h"
0f38bbef 13#include "lib/pow2.h"
a666cab8 14#include "lib/output_buffer.h"
a39fb9ea 15#include "helper_thread.h"
90d2d53f 16#include "smalloc.h"
fd5d733f 17#include "zbd.h"
f5bff36e 18#include "oslib/asprintf.h"
3c39a379 19
3d0d549a
BP
20#ifdef WIN32
21#define LOG_MSEC_SLACK 2
22#else
674456bf 23#define LOG_MSEC_SLACK 1
3d0d549a 24#endif
d454a205 25
971caeb1 26struct fio_sem *stat_sem;
cef9175e 27
210dd0fc
JA
28void clear_rusage_stat(struct thread_data *td)
29{
30 struct thread_stat *ts = &td->ts;
31
32 fio_getrusage(&td->ru_start);
33 ts->usr_time = ts->sys_time = 0;
34 ts->ctx = 0;
35 ts->minf = ts->majf = 0;
36}
37
3c39a379
JA
38void update_rusage_stat(struct thread_data *td)
39{
756867bd 40 struct thread_stat *ts = &td->ts;
3c39a379 41
44404c5a 42 fio_getrusage(&td->ru_end);
8b6a404c 43 ts->usr_time += mtime_since_tv(&td->ru_start.ru_utime,
c8aaba19 44 &td->ru_end.ru_utime);
8b6a404c 45 ts->sys_time += mtime_since_tv(&td->ru_start.ru_stime,
c8aaba19
JA
46 &td->ru_end.ru_stime);
47 ts->ctx += td->ru_end.ru_nvcsw + td->ru_end.ru_nivcsw
48 - (td->ru_start.ru_nvcsw + td->ru_start.ru_nivcsw);
49 ts->minf += td->ru_end.ru_minflt - td->ru_start.ru_minflt;
50 ts->majf += td->ru_end.ru_majflt - td->ru_start.ru_majflt;
5ec10eaa 51
c8aaba19 52 memcpy(&td->ru_start, &td->ru_end, sizeof(td->ru_end));
3c39a379
JA
53}
54
83349190
YH
55/*
56 * Given a latency, return the index of the corresponding bucket in
57 * the structure tracking percentiles.
58 *
59 * (1) find the group (and error bits) that the value (latency)
60 * belongs to by looking at its MSB. (2) find the bucket number in the
61 * group by looking at the index bits.
62 *
63 */
d6bb626e 64static unsigned int plat_val_to_idx(unsigned long long val)
83349190
YH
65{
66 unsigned int msb, error_bits, base, offset, idx;
67
68 /* Find MSB starting from bit 0 */
69 if (val == 0)
70 msb = 0;
71 else
d6bb626e 72 msb = (sizeof(val)*8) - __builtin_clzll(val) - 1;
83349190 73
716050f2
JA
74 /*
75 * MSB <= (FIO_IO_U_PLAT_BITS-1), cannot be rounded off. Use
76 * all bits of the sample as index
77 */
83349190
YH
78 if (msb <= FIO_IO_U_PLAT_BITS)
79 return val;
80
81 /* Compute the number of error bits to discard*/
82 error_bits = msb - FIO_IO_U_PLAT_BITS;
83
84 /* Compute the number of buckets before the group */
85 base = (error_bits + 1) << FIO_IO_U_PLAT_BITS;
86
716050f2
JA
87 /*
88 * Discard the error bits and apply the mask to find the
3c3ed070 89 * index for the buckets in the group
716050f2 90 */
83349190
YH
91 offset = (FIO_IO_U_PLAT_VAL - 1) & (val >> error_bits);
92
93 /* Make sure the index does not exceed (array size - 1) */
3c3ed070 94 idx = (base + offset) < (FIO_IO_U_PLAT_NR - 1) ?
83349190
YH
95 (base + offset) : (FIO_IO_U_PLAT_NR - 1);
96
97 return idx;
98}
99
100/*
101 * Convert the given index of the bucket array to the value
102 * represented by the bucket
103 */
129e4193 104static unsigned long long plat_idx_to_val(unsigned int idx)
83349190 105{
c8b44cfa
RL
106 unsigned int error_bits;
107 unsigned long long k, base;
83349190
YH
108
109 assert(idx < FIO_IO_U_PLAT_NR);
110
111 /* MSB <= (FIO_IO_U_PLAT_BITS-1), cannot be rounded off. Use
112 * all bits of the sample as index */
3c3ed070 113 if (idx < (FIO_IO_U_PLAT_VAL << 1))
83349190
YH
114 return idx;
115
116 /* Find the group and compute the minimum value of that group */
3c3ed070 117 error_bits = (idx >> FIO_IO_U_PLAT_BITS) - 1;
c8b44cfa 118 base = ((unsigned long long) 1) << (error_bits + FIO_IO_U_PLAT_BITS);
83349190
YH
119
120 /* Find its bucket number of the group */
121 k = idx % FIO_IO_U_PLAT_VAL;
122
123 /* Return the mean of the range of the bucket */
124 return base + ((k + 0.5) * (1 << error_bits));
125}
126
127static int double_cmp(const void *a, const void *b)
128{
802ad4a8
JA
129 const fio_fp64_t fa = *(const fio_fp64_t *) a;
130 const fio_fp64_t fb = *(const fio_fp64_t *) b;
83349190
YH
131 int cmp = 0;
132
802ad4a8 133 if (fa.u.f > fb.u.f)
83349190 134 cmp = 1;
802ad4a8 135 else if (fa.u.f < fb.u.f)
83349190
YH
136 cmp = -1;
137
138 return cmp;
139}
140
6cc0e5aa 141unsigned int calc_clat_percentiles(uint64_t *io_u_plat, unsigned long long nr,
d6bb626e
VF
142 fio_fp64_t *plist, unsigned long long **output,
143 unsigned long long *maxv, unsigned long long *minv)
83349190 144{
447b94e1 145 unsigned long long sum = 0;
1db92cb6 146 unsigned int len, i, j = 0;
d6bb626e 147 unsigned long long *ovals = NULL;
8985b491 148 bool is_last;
1db92cb6 149
d6bb626e 150 *minv = -1ULL;
1db92cb6 151 *maxv = 0;
83349190 152
802ad4a8
JA
153 len = 0;
154 while (len < FIO_IO_U_LIST_MAX_LEN && plist[len].u.f != 0.0)
155 len++;
716050f2 156
351de8de 157 if (!len)
1db92cb6 158 return 0;
351de8de 159
716050f2 160 /*
802ad4a8
JA
161 * Sort the percentile list. Note that it may already be sorted if
162 * we are using the default values, but since it's a short list this
163 * isn't a worry. Also note that this does not work for NaN values.
716050f2 164 */
802ad4a8 165 if (len > 1)
68e0dc1a 166 qsort(plist, len, sizeof(plist[0]), double_cmp);
83349190 167
5d89c625
VF
168 ovals = malloc(len * sizeof(*ovals));
169 if (!ovals)
170 return 0;
171
4f6f8298
JA
172 /*
173 * Calculate bucket values, note down max and min values
174 */
8985b491 175 is_last = false;
07511a63 176 for (i = 0; i < FIO_IO_U_PLAT_NR && !is_last; i++) {
83349190 177 sum += io_u_plat[i];
298d751e 178 while (sum >= ((long double) plist[j].u.f / 100.0 * nr)) {
802ad4a8 179 assert(plist[j].u.f <= 100.0);
83349190 180
4f6f8298 181 ovals[j] = plat_idx_to_val(i);
1db92cb6
JA
182 if (ovals[j] < *minv)
183 *minv = ovals[j];
184 if (ovals[j] > *maxv)
185 *maxv = ovals[j];
07511a63 186
8985b491 187 is_last = (j == len - 1) != 0;
07511a63
JA
188 if (is_last)
189 break;
190
4f6f8298
JA
191 j++;
192 }
193 }
83349190 194
298d751e
VF
195 if (!is_last)
196 log_err("fio: error calculating latency percentiles\n");
197
1db92cb6
JA
198 *output = ovals;
199 return len;
200}
201
202/*
203 * Find and display the p-th percentile of clat
204 */
6cc0e5aa 205static void show_clat_percentiles(uint64_t *io_u_plat, unsigned long long nr,
a666cab8 206 fio_fp64_t *plist, unsigned int precision,
b2b3eefe 207 const char *pre, struct buf_output *out)
1db92cb6 208{
d6bb626e
VF
209 unsigned int divisor, len, i, j = 0;
210 unsigned long long minv, maxv;
211 unsigned long long *ovals;
8985b491 212 int per_line, scale_down, time_width;
8985b491 213 bool is_last;
eef02441 214 char fmt[32];
1db92cb6
JA
215
216 len = calc_clat_percentiles(io_u_plat, nr, plist, &ovals, &maxv, &minv);
b2b3eefe 217 if (!len || !ovals)
824d8186 218 return;
1db92cb6 219
4f6f8298 220 /*
d6bb626e
VF
221 * We default to nsecs, but if the value range is such that we
222 * should scale down to usecs or msecs, do that.
4f6f8298 223 */
d6bb626e
VF
224 if (minv > 2000000 && maxv > 99999999ULL) {
225 scale_down = 2;
226 divisor = 1000000;
b599759b 227 log_buf(out, " %s percentiles (msec):\n |", pre);
d6bb626e
VF
228 } else if (minv > 2000 && maxv > 99999) {
229 scale_down = 1;
230 divisor = 1000;
b599759b 231 log_buf(out, " %s percentiles (usec):\n |", pre);
4f6f8298
JA
232 } else {
233 scale_down = 0;
d6bb626e 234 divisor = 1;
b599759b 235 log_buf(out, " %s percentiles (nsec):\n |", pre);
4f6f8298 236 }
83349190 237
619adf9c 238
d6bb626e 239 time_width = max(5, (int) (log10(maxv / divisor) + 1));
74558486
JA
240 snprintf(fmt, sizeof(fmt), " %%%u.%ufth=[%%%dllu]%%c", precision + 3,
241 precision, time_width);
242 /* fmt will be something like " %5.2fth=[%4llu]%c" */
d6bb626e 243 per_line = (80 - 7) / (precision + 10 + time_width);
81ab0b3a 244
d6bb626e 245 for (j = 0; j < len; j++) {
4f6f8298 246 /* for formatting */
eef02441 247 if (j != 0 && (j % per_line) == 0)
a666cab8 248 log_buf(out, " |");
83349190 249
4f6f8298 250 /* end of the list */
8985b491 251 is_last = (j == len - 1) != 0;
83349190 252
d6bb626e 253 for (i = 0; i < scale_down; i++)
4f6f8298
JA
254 ovals[j] = (ovals[j] + 999) / 1000;
255
d6bb626e 256 log_buf(out, fmt, plist[j].u.f, ovals[j], is_last ? '\n' : ',');
4f6f8298
JA
257
258 if (is_last)
259 break;
260
eef02441 261 if ((j % per_line) == per_line - 1) /* for formatting */
a666cab8 262 log_buf(out, "\n");
83349190 263 }
4f6f8298 264
283feb79 265 free(ovals);
83349190
YH
266}
267
692dec0c
NC
268static int get_nr_prios_with_samples(struct thread_stat *ts, enum fio_ddir ddir)
269{
270 int i, nr_prios_with_samples = 0;
271
272 for (i = 0; i < ts->nr_clat_prio[ddir]; i++) {
273 if (ts->clat_prio[ddir][i].clat_stat.samples)
274 nr_prios_with_samples++;
275 }
276
277 return nr_prios_with_samples;
278}
279
74558486
JA
280bool calc_lat(struct io_stat *is, unsigned long long *min,
281 unsigned long long *max, double *mean, double *dev)
3c39a379 282{
ee0ccb79 283 double n = (double) is->samples;
3c39a379 284
ee0ccb79 285 if (n == 0)
8aa89d70 286 return false;
3c39a379
JA
287
288 *min = is->min_val;
289 *max = is->max_val;
802ad4a8 290 *mean = is->mean.u.f;
e6d276f2 291
68704084 292 if (n > 1.0)
802ad4a8 293 *dev = sqrt(is->S.u.f / (n - 1.0));
ef9c5c40 294 else
4b43f54e 295 *dev = 0;
ef9c5c40 296
8aa89d70 297 return true;
3c39a379
JA
298}
299
5cb8a8cd
BP
300void show_mixed_group_stats(struct group_run_stats *rs, struct buf_output *out)
301{
302 char *io, *agg, *min, *max;
303 char *ioalt, *aggalt, *minalt, *maxalt;
6dc75062
JA
304 uint64_t io_mix = 0, agg_mix = 0, min_mix = -1, max_mix = 0;
305 uint64_t min_run = -1, max_run = 0;
5cb8a8cd 306 const int i2p = is_power_of_2(rs->kb_base);
6dc75062 307 int i;
5cb8a8cd
BP
308
309 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
310 if (!rs->max_run[i])
311 continue;
312 io_mix += rs->iobytes[i];
313 agg_mix += rs->agg[i];
314 min_mix = min_mix < rs->min_bw[i] ? min_mix : rs->min_bw[i];
315 max_mix = max_mix > rs->max_bw[i] ? max_mix : rs->max_bw[i];
316 min_run = min_run < rs->min_run[i] ? min_run : rs->min_run[i];
317 max_run = max_run > rs->max_run[i] ? max_run : rs->max_run[i];
318 }
319 io = num2str(io_mix, rs->sig_figs, 1, i2p, N2S_BYTE);
320 ioalt = num2str(io_mix, rs->sig_figs, 1, !i2p, N2S_BYTE);
321 agg = num2str(agg_mix, rs->sig_figs, 1, i2p, rs->unit_base);
322 aggalt = num2str(agg_mix, rs->sig_figs, 1, !i2p, rs->unit_base);
323 min = num2str(min_mix, rs->sig_figs, 1, i2p, rs->unit_base);
324 minalt = num2str(min_mix, rs->sig_figs, 1, !i2p, rs->unit_base);
325 max = num2str(max_mix, rs->sig_figs, 1, i2p, rs->unit_base);
326 maxalt = num2str(max_mix, rs->sig_figs, 1, !i2p, rs->unit_base);
327 log_buf(out, " MIXED: bw=%s (%s), %s-%s (%s-%s), io=%s (%s), run=%llu-%llumsec\n",
328 agg, aggalt, min, max, minalt, maxalt, io, ioalt,
329 (unsigned long long) min_run,
330 (unsigned long long) max_run);
331 free(io);
332 free(agg);
333 free(min);
334 free(max);
335 free(ioalt);
336 free(aggalt);
337 free(minalt);
338 free(maxalt);
339}
340
a666cab8 341void show_group_stats(struct group_run_stats *rs, struct buf_output *out)
3c39a379 342{
d694a6a7
RE
343 char *io, *agg, *min, *max;
344 char *ioalt, *aggalt, *minalt, *maxalt;
42da5c8b 345 const char *str[] = { " READ", " WRITE" , " TRIM"};
dbe1125e
JA
346 int i;
347
a666cab8 348 log_buf(out, "\nRun status group %d (all jobs):\n", rs->groupid);
3c39a379 349
6eaf09d6 350 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
90fef2d1
JA
351 const int i2p = is_power_of_2(rs->kb_base);
352
dbe1125e
JA
353 if (!rs->max_run[i])
354 continue;
355
e883cb35
JF
356 io = num2str(rs->iobytes[i], rs->sig_figs, 1, i2p, N2S_BYTE);
357 ioalt = num2str(rs->iobytes[i], rs->sig_figs, 1, !i2p, N2S_BYTE);
358 agg = num2str(rs->agg[i], rs->sig_figs, 1, i2p, rs->unit_base);
359 aggalt = num2str(rs->agg[i], rs->sig_figs, 1, !i2p, rs->unit_base);
360 min = num2str(rs->min_bw[i], rs->sig_figs, 1, i2p, rs->unit_base);
361 minalt = num2str(rs->min_bw[i], rs->sig_figs, 1, !i2p, rs->unit_base);
362 max = num2str(rs->max_bw[i], rs->sig_figs, 1, i2p, rs->unit_base);
363 maxalt = num2str(rs->max_bw[i], rs->sig_figs, 1, !i2p, rs->unit_base);
d694a6a7 364 log_buf(out, "%s: bw=%s (%s), %s-%s (%s-%s), io=%s (%s), run=%llu-%llumsec\n",
5cb8a8cd 365 (rs->unified_rw_rep == UNIFIED_MIXED) ? " MIXED" : str[i],
d694a6a7 366 agg, aggalt, min, max, minalt, maxalt, io, ioalt,
4e0a8fa2
JA
367 (unsigned long long) rs->min_run[i],
368 (unsigned long long) rs->max_run[i]);
dbe1125e 369
d694a6a7
RE
370 free(io);
371 free(agg);
372 free(min);
373 free(max);
374 free(ioalt);
375 free(aggalt);
376 free(minalt);
377 free(maxalt);
dbe1125e 378 }
6dc75062 379
5cb8a8cd 380 /* Need to aggregate statisitics to show mixed values */
6dc75062 381 if (rs->unified_rw_rep == UNIFIED_BOTH)
5cb8a8cd 382 show_mixed_group_stats(rs, out);
3c39a379
JA
383}
384
6cc0e5aa 385void stat_calc_dist(uint64_t *map, unsigned long total, double *io_u_dist)
2270890c
JA
386{
387 int i;
388
389 /*
390 * Do depth distribution calculations
391 */
392 for (i = 0; i < FIO_IO_U_MAP_NR; i++) {
838bc709
JA
393 if (total) {
394 io_u_dist[i] = (double) map[i] / (double) total;
395 io_u_dist[i] *= 100.0;
396 if (io_u_dist[i] < 0.1 && map[i])
397 io_u_dist[i] = 0.1;
398 } else
399 io_u_dist[i] = 0.0;
2270890c
JA
400 }
401}
402
04a0feae 403static void stat_calc_lat(struct thread_stat *ts, double *dst,
6cc0e5aa 404 uint64_t *src, int nr)
2270890c 405{
d79db122 406 unsigned long total = ddir_rw_sum(ts->total_io_u);
2270890c
JA
407 int i;
408
409 /*
410 * Do latency distribution calculations
411 */
04a0feae 412 for (i = 0; i < nr; i++) {
838bc709
JA
413 if (total) {
414 dst[i] = (double) src[i] / (double) total;
415 dst[i] *= 100.0;
416 if (dst[i] < 0.01 && src[i])
417 dst[i] = 0.01;
418 } else
419 dst[i] = 0.0;
2270890c
JA
420 }
421}
422
247823cc
VF
423/*
424 * To keep the terse format unaltered, add all of the ns latency
425 * buckets to the first us latency bucket
426 */
52e4c651 427static void stat_calc_lat_nu(struct thread_stat *ts, double *io_u_lat_u)
247823cc
VF
428{
429 unsigned long ntotal = 0, total = ddir_rw_sum(ts->total_io_u);
430 int i;
431
432 stat_calc_lat(ts, io_u_lat_u, ts->io_u_lat_u, FIO_IO_U_LAT_U_NR);
433
434 for (i = 0; i < FIO_IO_U_LAT_N_NR; i++)
435 ntotal += ts->io_u_lat_n[i];
436
437 io_u_lat_u[0] += 100.0 * (double) ntotal / (double) total;
438}
439
d6bb626e
VF
440void stat_calc_lat_n(struct thread_stat *ts, double *io_u_lat)
441{
442 stat_calc_lat(ts, io_u_lat, ts->io_u_lat_n, FIO_IO_U_LAT_N_NR);
443}
444
e5bd1347 445void stat_calc_lat_u(struct thread_stat *ts, double *io_u_lat)
04a0feae
JA
446{
447 stat_calc_lat(ts, io_u_lat, ts->io_u_lat_u, FIO_IO_U_LAT_U_NR);
448}
449
e5bd1347 450void stat_calc_lat_m(struct thread_stat *ts, double *io_u_lat)
04a0feae
JA
451{
452 stat_calc_lat(ts, io_u_lat, ts->io_u_lat_m, FIO_IO_U_LAT_M_NR);
453}
454
74558486
JA
455static void display_lat(const char *name, unsigned long long min,
456 unsigned long long max, double mean, double dev,
457 struct buf_output *out)
ea2accc5 458{
d6bb626e 459 const char *base = "(nsec)";
b29ad562 460 char *minp, *maxp;
ea2accc5 461
d6bb626e 462 if (nsec_to_msec(&min, &max, &mean, &dev))
b29ad562 463 base = "(msec)";
d6bb626e
VF
464 else if (nsec_to_usec(&min, &max, &mean, &dev))
465 base = "(usec)";
b29ad562 466
d694a6a7
RE
467 minp = num2str(min, 6, 1, 0, N2S_NONE);
468 maxp = num2str(max, 6, 1, 0, N2S_NONE);
b29ad562 469
a666cab8 470 log_buf(out, " %s %s: min=%s, max=%s, avg=%5.02f,"
b29ad562
JA
471 " stdev=%5.02f\n", name, base, minp, maxp, mean, dev);
472
473 free(minp);
474 free(maxp);
ea2accc5
JA
475}
476
b182f077
NC
477static struct thread_stat *gen_mixed_ddir_stats_from_ts(struct thread_stat *ts)
478{
479 struct thread_stat *ts_lcl;
480
481 /*
482 * Handle aggregation of Reads (ddir = 0), Writes (ddir = 1), and
483 * Trims (ddir = 2)
484 */
485 ts_lcl = malloc(sizeof(struct thread_stat));
486 if (!ts_lcl) {
487 log_err("fio: failed to allocate local thread stat\n");
488 return NULL;
489 }
490
491 init_thread_stat(ts_lcl);
492
493 /* calculate mixed stats */
494 ts_lcl->unified_rw_rep = UNIFIED_MIXED;
495 ts_lcl->lat_percentiles = ts->lat_percentiles;
496 ts_lcl->clat_percentiles = ts->clat_percentiles;
497 ts_lcl->slat_percentiles = ts->slat_percentiles;
498 ts_lcl->percentile_precision = ts->percentile_precision;
499 memcpy(ts_lcl->percentile_list, ts->percentile_list, sizeof(ts->percentile_list));
500
501 sum_thread_stats(ts_lcl, ts);
502
503 return ts_lcl;
504}
505
2e545522
NC
506static double convert_agg_kbytes_percent(struct group_run_stats *rs,
507 enum fio_ddir ddir, int mean)
5c0abd5e 508{
509 double p_of_agg = 100.0;
510 if (rs && rs->agg[ddir] > 1024) {
106e14ce 511 p_of_agg = mean * 100.0 / (double) (rs->agg[ddir] / 1024.0);
5c0abd5e 512
513 if (p_of_agg > 100.0)
514 p_of_agg = 100.0;
515 }
516 return p_of_agg;
517}
518
756867bd 519static void show_ddir_status(struct group_run_stats *rs, struct thread_stat *ts,
2e545522 520 enum fio_ddir ddir, struct buf_output *out)
3c39a379 521{
d6bb626e
VF
522 unsigned long runt;
523 unsigned long long min, max, bw, iops;
3c39a379 524 double mean, dev;
96563db9 525 char *io_p, *bw_p, *bw_p_alt, *iops_p, *post_st = NULL;
692dec0c
NC
526 int i2p, i;
527 const char *clat_type = ts->lat_percentiles ? "lat" : "clat";
3c39a379 528
b2b3eefe
JA
529 if (ddir_sync(ddir)) {
530 if (calc_lat(&ts->sync_stat, &min, &max, &mean, &dev)) {
531 log_buf(out, " %s:\n", "fsync/fdatasync/sync_file_range");
425d3e0e 532 display_lat(io_ddir_name(ddir), min, max, mean, dev, out);
b2b3eefe
JA
533 show_clat_percentiles(ts->io_u_sync_plat,
534 ts->sync_stat.samples,
535 ts->percentile_list,
536 ts->percentile_precision,
425d3e0e 537 io_ddir_name(ddir), out);
b2b3eefe
JA
538 }
539 return;
540 }
541
ff58fced
JA
542 assert(ddir_rw(ddir));
543
756867bd 544 if (!ts->runtime[ddir])
3c39a379
JA
545 return;
546
90fef2d1 547 i2p = is_power_of_2(rs->kb_base);
8879fd15
JA
548 runt = ts->runtime[ddir];
549
550 bw = (1000 * ts->io_bytes[ddir]) / runt;
e883cb35
JF
551 io_p = num2str(ts->io_bytes[ddir], ts->sig_figs, 1, i2p, N2S_BYTE);
552 bw_p = num2str(bw, ts->sig_figs, 1, i2p, ts->unit_base);
553 bw_p_alt = num2str(bw, ts->sig_figs, 1, !i2p, ts->unit_base);
8879fd15 554
0aacc50c 555 iops = (1000 * (uint64_t)ts->total_io_u[ddir]) / runt;
e883cb35 556 iops_p = num2str(iops, ts->sig_figs, 1, 0, N2S_NONE);
fd5d733f 557 if (ddir == DDIR_WRITE)
96563db9
JA
558 post_st = zbd_write_status(ts);
559 else if (ddir == DDIR_READ && ts->cachehit && ts->cachemiss) {
560 uint64_t total;
561 double hit;
562
563 total = ts->cachehit + ts->cachemiss;
564 hit = (double) ts->cachehit / (double) total;
565 hit *= 100.0;
566 if (asprintf(&post_st, "; Cachehit=%0.2f%%", hit) < 0)
567 post_st = NULL;
568 }
dbe1125e 569
fd5d733f 570 log_buf(out, " %s: IOPS=%s, BW=%s (%s)(%s/%llumsec)%s\n",
5cb8a8cd 571 (ts->unified_rw_rep == UNIFIED_MIXED) ? "mixed" : io_ddir_name(ddir),
d694a6a7 572 iops_p, bw_p, bw_p_alt, io_p,
fd5d733f 573 (unsigned long long) ts->runtime[ddir],
96563db9 574 post_st ? : "");
dbe1125e 575
96563db9 576 free(post_st);
dbe1125e
JA
577 free(io_p);
578 free(bw_p);
d694a6a7 579 free(bw_p_alt);
b3605062 580 free(iops_p);
3c39a379 581
b29ad562 582 if (calc_lat(&ts->slat_stat[ddir], &min, &max, &mean, &dev))
a666cab8 583 display_lat("slat", min, max, mean, dev, out);
b29ad562 584 if (calc_lat(&ts->clat_stat[ddir], &min, &max, &mean, &dev))
a666cab8 585 display_lat("clat", min, max, mean, dev, out);
b29ad562 586 if (calc_lat(&ts->lat_stat[ddir], &min, &max, &mean, &dev))
a666cab8 587 display_lat(" lat", min, max, mean, dev, out);
692dec0c
NC
588
589 /* Only print per prio stats if there are >= 2 prios with samples */
590 if (get_nr_prios_with_samples(ts, ddir) >= 2) {
591 for (i = 0; i < ts->nr_clat_prio[ddir]; i++) {
592 if (calc_lat(&ts->clat_prio[ddir][i].clat_stat, &min,
593 &max, &mean, &dev)) {
594 char buf[64];
595
596 snprintf(buf, sizeof(buf),
597 "%s prio %u/%u",
598 clat_type,
599 ts->clat_prio[ddir][i].ioprio >> 13,
600 ts->clat_prio[ddir][i].ioprio & 7);
601 display_lat(buf, min, max, mean, dev, out);
602 }
603 }
2e86a90c 604 }
02af0988 605
56440e63
VF
606 if (ts->slat_percentiles && ts->slat_stat[ddir].samples > 0)
607 show_clat_percentiles(ts->io_u_plat[FIO_SLAT][ddir],
608 ts->slat_stat[ddir].samples,
609 ts->percentile_list,
610 ts->percentile_precision, "slat", out);
611 if (ts->clat_percentiles && ts->clat_stat[ddir].samples > 0)
612 show_clat_percentiles(ts->io_u_plat[FIO_CLAT][ddir],
613 ts->clat_stat[ddir].samples,
614 ts->percentile_list,
615 ts->percentile_precision, "clat", out);
616 if (ts->lat_percentiles && ts->lat_stat[ddir].samples > 0)
617 show_clat_percentiles(ts->io_u_plat[FIO_LAT][ddir],
618 ts->lat_stat[ddir].samples,
619 ts->percentile_list,
620 ts->percentile_precision, "lat", out);
621
b599759b 622 if (ts->clat_percentiles || ts->lat_percentiles) {
692dec0c 623 char prio_name[64];
af1600c1
SW
624 uint64_t samples;
625
56440e63 626 if (ts->lat_percentiles)
af1600c1 627 samples = ts->lat_stat[ddir].samples;
56440e63
VF
628 else
629 samples = ts->clat_stat[ddir].samples;
b2a432bf 630
692dec0c
NC
631 /* Only print per prio stats if there are >= 2 prios with samples */
632 if (get_nr_prios_with_samples(ts, ddir) >= 2) {
633 for (i = 0; i < ts->nr_clat_prio[ddir]; i++) {
634 uint64_t prio_samples = ts->clat_prio[ddir][i].clat_stat.samples;
635
636 if (prio_samples > 0) {
637 snprintf(prio_name, sizeof(prio_name),
638 "%s prio %u/%u (%.2f%% of IOs)",
639 clat_type,
640 ts->clat_prio[ddir][i].ioprio >> 13,
641 ts->clat_prio[ddir][i].ioprio & 7,
642 100. * (double) prio_samples / (double) samples);
643 show_clat_percentiles(ts->clat_prio[ddir][i].io_u_plat,
644 prio_samples, ts->percentile_list,
645 ts->percentile_precision,
646 prio_name, out);
647 }
648 }
b2a432bf 649 }
83349190 650 }
56440e63 651
079ad09b 652 if (calc_lat(&ts->bw_stat[ddir], &min, &max, &mean, &dev)) {
142c7f2d 653 double p_of_agg = 100.0, fkb_base = (double)rs->kb_base;
d694a6a7
RE
654 const char *bw_str;
655
656 if ((rs->unit_base == 1) && i2p)
657 bw_str = "Kibit";
658 else if (rs->unit_base == 1)
659 bw_str = "kbit";
660 else if (i2p)
661 bw_str = "KiB";
662 else
663 bw_str = "kB";
d686990a 664
5c0abd5e 665 p_of_agg = convert_agg_kbytes_percent(rs, ddir, mean);
d1707474 666
d686990a
SN
667 if (rs->unit_base == 1) {
668 min *= 8.0;
669 max *= 8.0;
670 mean *= 8.0;
671 dev *= 8.0;
672 }
3c39a379 673
142c7f2d
SN
674 if (mean > fkb_base * fkb_base) {
675 min /= fkb_base;
676 max /= fkb_base;
677 mean /= fkb_base;
678 dev /= fkb_base;
d694a6a7 679 bw_str = (rs->unit_base == 1 ? "Mibit" : "MiB");
b7017e32
JA
680 }
681
54c05828 682 log_buf(out, " bw (%5s/s): min=%5llu, max=%5llu, per=%3.2f%%, "
29eb371b 683 "avg=%5.02f, stdev=%5.02f, samples=%" PRIu64 "\n",
54c05828
AH
684 bw_str, min, max, p_of_agg, mean, dev,
685 (&ts->bw_stat[ddir])->samples);
3c39a379 686 }
188b6016 687 if (calc_lat(&ts->iops_stat[ddir], &min, &max, &mean, &dev)) {
21ba6606 688 log_buf(out, " iops : min=%5llu, max=%5llu, "
29eb371b 689 "avg=%5.02f, stdev=%5.02f, samples=%" PRIu64 "\n",
54c05828 690 min, max, mean, dev, (&ts->iops_stat[ddir])->samples);
188b6016 691 }
3c39a379
JA
692}
693
53d966f8
NC
694static void show_mixed_ddir_status(struct group_run_stats *rs,
695 struct thread_stat *ts,
696 struct buf_output *out)
697{
b182f077 698 struct thread_stat *ts_lcl = gen_mixed_ddir_stats_from_ts(ts);
53d966f8 699
b182f077
NC
700 if (ts_lcl)
701 show_ddir_status(rs, ts_lcl, DDIR_READ, out);
53d966f8 702
692dec0c 703 free_clat_prio_stats(ts_lcl);
53d966f8
NC
704 free(ts_lcl);
705}
706
8985b491
JA
707static bool show_lat(double *io_u_lat, int nr, const char **ranges,
708 const char *msg, struct buf_output *out)
04a0feae 709{
8985b491
JA
710 bool new_line = true, shown = false;
711 int i, line = 0;
04a0feae
JA
712
713 for (i = 0; i < nr; i++) {
714 if (io_u_lat[i] <= 0.0)
715 continue;
8985b491 716 shown = true;
04a0feae 717 if (new_line) {
4539ed73 718 if (line)
a666cab8 719 log_buf(out, "\n");
4a19fcaa 720 log_buf(out, " lat (%s) : ", msg);
8985b491 721 new_line = false;
04a0feae
JA
722 line = 0;
723 }
724 if (line)
a666cab8
JA
725 log_buf(out, ", ");
726 log_buf(out, "%s%3.2f%%", ranges[i], io_u_lat[i]);
04a0feae
JA
727 line++;
728 if (line == 5)
8985b491 729 new_line = true;
04a0feae 730 }
7e1773ba
JA
731
732 if (shown)
a666cab8 733 log_buf(out, "\n");
7e1773ba 734
8985b491 735 return true;
04a0feae
JA
736}
737
d6bb626e
VF
738static void show_lat_n(double *io_u_lat_n, struct buf_output *out)
739{
740 const char *ranges[] = { "2=", "4=", "10=", "20=", "50=", "100=",
741 "250=", "500=", "750=", "1000=", };
742
743 show_lat(io_u_lat_n, FIO_IO_U_LAT_N_NR, ranges, "nsec", out);
744}
745
a666cab8 746static void show_lat_u(double *io_u_lat_u, struct buf_output *out)
04a0feae
JA
747{
748 const char *ranges[] = { "2=", "4=", "10=", "20=", "50=", "100=",
749 "250=", "500=", "750=", "1000=", };
750
a666cab8 751 show_lat(io_u_lat_u, FIO_IO_U_LAT_U_NR, ranges, "usec", out);
04a0feae
JA
752}
753
a666cab8 754static void show_lat_m(double *io_u_lat_m, struct buf_output *out)
04a0feae
JA
755{
756 const char *ranges[] = { "2=", "4=", "10=", "20=", "50=", "100=",
757 "250=", "500=", "750=", "1000=", "2000=",
758 ">=2000=", };
759
a666cab8 760 show_lat(io_u_lat_m, FIO_IO_U_LAT_M_NR, ranges, "msec", out);
04a0feae
JA
761}
762
a666cab8 763static void show_latencies(struct thread_stat *ts, struct buf_output *out)
04a0feae 764{
d6bb626e 765 double io_u_lat_n[FIO_IO_U_LAT_N_NR];
c551f65a
JA
766 double io_u_lat_u[FIO_IO_U_LAT_U_NR];
767 double io_u_lat_m[FIO_IO_U_LAT_M_NR];
768
d6bb626e 769 stat_calc_lat_n(ts, io_u_lat_n);
5a18988e
JA
770 stat_calc_lat_u(ts, io_u_lat_u);
771 stat_calc_lat_m(ts, io_u_lat_m);
772
d6bb626e 773 show_lat_n(io_u_lat_n, out);
a666cab8
JA
774 show_lat_u(io_u_lat_u, out);
775 show_lat_m(io_u_lat_m, out);
04a0feae
JA
776}
777
66347cfa
DE
778static int block_state_category(int block_state)
779{
780 switch (block_state) {
781 case BLOCK_STATE_UNINIT:
782 return 0;
783 case BLOCK_STATE_TRIMMED:
784 case BLOCK_STATE_WRITTEN:
785 return 1;
786 case BLOCK_STATE_WRITE_FAILURE:
787 case BLOCK_STATE_TRIM_FAILURE:
788 return 2;
789 default:
081c2dc3 790 /* Silence compile warning on some BSDs and have a return */
66347cfa 791 assert(0);
07ff5841 792 return -1;
66347cfa
DE
793 }
794}
795
796static int compare_block_infos(const void *bs1, const void *bs2)
797{
5fff9543
JF
798 uint64_t block1 = *(uint64_t *)bs1;
799 uint64_t block2 = *(uint64_t *)bs2;
66347cfa
DE
800 int state1 = BLOCK_INFO_STATE(block1);
801 int state2 = BLOCK_INFO_STATE(block2);
802 int bscat1 = block_state_category(state1);
803 int bscat2 = block_state_category(state2);
804 int cycles1 = BLOCK_INFO_TRIMS(block1);
805 int cycles2 = BLOCK_INFO_TRIMS(block2);
806
807 if (bscat1 < bscat2)
808 return -1;
809 if (bscat1 > bscat2)
810 return 1;
811
812 if (cycles1 < cycles2)
813 return -1;
814 if (cycles1 > cycles2)
815 return 1;
816
817 if (state1 < state2)
818 return -1;
819 if (state1 > state2)
820 return 1;
821
822 assert(block1 == block2);
823 return 0;
824}
825
826static int calc_block_percentiles(int nr_block_infos, uint32_t *block_infos,
827 fio_fp64_t *plist, unsigned int **percentiles,
828 unsigned int *types)
829{
830 int len = 0;
831 int i, nr_uninit;
832
833 qsort(block_infos, nr_block_infos, sizeof(uint32_t), compare_block_infos);
834
835 while (len < FIO_IO_U_LIST_MAX_LEN && plist[len].u.f != 0.0)
836 len++;
837
838 if (!len)
839 return 0;
840
841 /*
842 * Sort the percentile list. Note that it may already be sorted if
843 * we are using the default values, but since it's a short list this
844 * isn't a worry. Also note that this does not work for NaN values.
845 */
846 if (len > 1)
68e0dc1a 847 qsort(plist, len, sizeof(plist[0]), double_cmp);
66347cfa 848
66347cfa
DE
849 /* Start only after the uninit entries end */
850 for (nr_uninit = 0;
851 nr_uninit < nr_block_infos
852 && BLOCK_INFO_STATE(block_infos[nr_uninit]) == BLOCK_STATE_UNINIT;
853 nr_uninit ++)
854 ;
855
856 if (nr_uninit == nr_block_infos)
857 return 0;
858
859 *percentiles = calloc(len, sizeof(**percentiles));
860
861 for (i = 0; i < len; i++) {
862 int idx = (plist[i].u.f * (nr_block_infos - nr_uninit) / 100)
863 + nr_uninit;
864 (*percentiles)[i] = BLOCK_INFO_TRIMS(block_infos[idx]);
865 }
866
867 memset(types, 0, sizeof(*types) * BLOCK_STATE_COUNT);
868 for (i = 0; i < nr_block_infos; i++)
869 types[BLOCK_INFO_STATE(block_infos[i])]++;
870
871 return len;
872}
873
874static const char *block_state_names[] = {
875 [BLOCK_STATE_UNINIT] = "unwritten",
876 [BLOCK_STATE_TRIMMED] = "trimmed",
877 [BLOCK_STATE_WRITTEN] = "written",
878 [BLOCK_STATE_TRIM_FAILURE] = "trim failure",
879 [BLOCK_STATE_WRITE_FAILURE] = "write failure",
880};
881
882static void show_block_infos(int nr_block_infos, uint32_t *block_infos,
a666cab8 883 fio_fp64_t *plist, struct buf_output *out)
66347cfa
DE
884{
885 int len, pos, i;
886 unsigned int *percentiles = NULL;
887 unsigned int block_state_counts[BLOCK_STATE_COUNT];
888
889 len = calc_block_percentiles(nr_block_infos, block_infos, plist,
890 &percentiles, block_state_counts);
891
a666cab8 892 log_buf(out, " block lifetime percentiles :\n |");
66347cfa
DE
893 pos = 0;
894 for (i = 0; i < len; i++) {
895 uint32_t block_info = percentiles[i];
896#define LINE_LENGTH 75
897 char str[LINE_LENGTH];
898 int strln = snprintf(str, LINE_LENGTH, " %3.2fth=%u%c",
899 plist[i].u.f, block_info,
900 i == len - 1 ? '\n' : ',');
901 assert(strln < LINE_LENGTH);
902 if (pos + strln > LINE_LENGTH) {
903 pos = 0;
a666cab8 904 log_buf(out, "\n |");
66347cfa 905 }
a666cab8 906 log_buf(out, "%s", str);
66347cfa
DE
907 pos += strln;
908#undef LINE_LENGTH
909 }
910 if (percentiles)
911 free(percentiles);
912
a666cab8 913 log_buf(out, " states :");
66347cfa 914 for (i = 0; i < BLOCK_STATE_COUNT; i++)
a666cab8 915 log_buf(out, " %s=%u%c",
66347cfa
DE
916 block_state_names[i], block_state_counts[i],
917 i == BLOCK_STATE_COUNT - 1 ? '\n' : ',');
918}
919
d685adfb
VF
920static void show_ss_normal(struct thread_stat *ts, struct buf_output *out)
921{
d694a6a7 922 char *p1, *p1alt, *p2;
d685adfb
VF
923 unsigned long long bw_mean, iops_mean;
924 const int i2p = is_power_of_2(ts->kb_base);
925
0c13c969 926 if (!ts->ss_dur)
d685adfb
VF
927 return;
928
bb49c8bd
VF
929 bw_mean = steadystate_bw_mean(ts);
930 iops_mean = steadystate_iops_mean(ts);
d685adfb 931
e883cb35
JF
932 p1 = num2str(bw_mean / ts->kb_base, ts->sig_figs, ts->kb_base, i2p, ts->unit_base);
933 p1alt = num2str(bw_mean / ts->kb_base, ts->sig_figs, ts->kb_base, !i2p, ts->unit_base);
934 p2 = num2str(iops_mean, ts->sig_figs, 1, 0, N2S_NONE);
d685adfb 935
d694a6a7 936 log_buf(out, " steadystate : attained=%s, bw=%s (%s), iops=%s, %s%s=%.3f%s\n",
c8caba48 937 ts->ss_state & FIO_SS_ATTAINED ? "yes" : "no",
d694a6a7 938 p1, p1alt, p2,
c8caba48
JA
939 ts->ss_state & FIO_SS_IOPS ? "iops" : "bw",
940 ts->ss_state & FIO_SS_SLOPE ? " slope": " mean dev",
bb49c8bd 941 ts->ss_criterion.u.f,
c8caba48 942 ts->ss_state & FIO_SS_PCT ? "%" : "");
d685adfb
VF
943
944 free(p1);
d694a6a7 945 free(p1alt);
d685adfb
VF
946 free(p2);
947}
948
2a2fdab1
L
949static void show_agg_stats(struct disk_util_agg *agg, int terse,
950 struct buf_output *out)
951{
952 if (!agg->slavecount)
953 return;
954
955 if (!terse) {
956 log_buf(out, ", aggrios=%llu/%llu, aggrmerge=%llu/%llu, "
957 "aggrticks=%llu/%llu, aggrin_queue=%llu, "
958 "aggrutil=%3.2f%%",
959 (unsigned long long) agg->ios[0] / agg->slavecount,
960 (unsigned long long) agg->ios[1] / agg->slavecount,
961 (unsigned long long) agg->merges[0] / agg->slavecount,
962 (unsigned long long) agg->merges[1] / agg->slavecount,
963 (unsigned long long) agg->ticks[0] / agg->slavecount,
964 (unsigned long long) agg->ticks[1] / agg->slavecount,
965 (unsigned long long) agg->time_in_queue / agg->slavecount,
966 agg->max_util.u.f);
967 } else {
968 log_buf(out, ";slaves;%llu;%llu;%llu;%llu;%llu;%llu;%llu;%3.2f%%",
969 (unsigned long long) agg->ios[0] / agg->slavecount,
970 (unsigned long long) agg->ios[1] / agg->slavecount,
971 (unsigned long long) agg->merges[0] / agg->slavecount,
972 (unsigned long long) agg->merges[1] / agg->slavecount,
973 (unsigned long long) agg->ticks[0] / agg->slavecount,
974 (unsigned long long) agg->ticks[1] / agg->slavecount,
975 (unsigned long long) agg->time_in_queue / agg->slavecount,
976 agg->max_util.u.f);
977 }
978}
979
980static void aggregate_slaves_stats(struct disk_util *masterdu)
981{
982 struct disk_util_agg *agg = &masterdu->agg;
983 struct disk_util_stat *dus;
984 struct flist_head *entry;
985 struct disk_util *slavedu;
986 double util;
987
988 flist_for_each(entry, &masterdu->slaves) {
989 slavedu = flist_entry(entry, struct disk_util, slavelist);
990 dus = &slavedu->dus;
991 agg->ios[0] += dus->s.ios[0];
992 agg->ios[1] += dus->s.ios[1];
993 agg->merges[0] += dus->s.merges[0];
994 agg->merges[1] += dus->s.merges[1];
995 agg->sectors[0] += dus->s.sectors[0];
996 agg->sectors[1] += dus->s.sectors[1];
997 agg->ticks[0] += dus->s.ticks[0];
998 agg->ticks[1] += dus->s.ticks[1];
999 agg->time_in_queue += dus->s.time_in_queue;
1000 agg->slavecount++;
1001
1002 util = (double) (100 * dus->s.io_ticks / (double) slavedu->dus.s.msec);
1003 /* System utilization is the utilization of the
1004 * component with the highest utilization.
1005 */
1006 if (util > agg->max_util.u.f)
1007 agg->max_util.u.f = util;
1008
1009 }
1010
1011 if (agg->max_util.u.f > 100.0)
1012 agg->max_util.u.f = 100.0;
1013}
1014
1015void print_disk_util(struct disk_util_stat *dus, struct disk_util_agg *agg,
1016 int terse, struct buf_output *out)
1017{
1018 double util = 0;
1019
1020 if (dus->s.msec)
1021 util = (double) 100 * dus->s.io_ticks / (double) dus->s.msec;
1022 if (util > 100.0)
1023 util = 100.0;
1024
1025 if (!terse) {
1026 if (agg->slavecount)
1027 log_buf(out, " ");
1028
1029 log_buf(out, " %s: ios=%llu/%llu, merge=%llu/%llu, "
1030 "ticks=%llu/%llu, in_queue=%llu, util=%3.2f%%",
1031 dus->name,
1032 (unsigned long long) dus->s.ios[0],
1033 (unsigned long long) dus->s.ios[1],
1034 (unsigned long long) dus->s.merges[0],
1035 (unsigned long long) dus->s.merges[1],
1036 (unsigned long long) dus->s.ticks[0],
1037 (unsigned long long) dus->s.ticks[1],
1038 (unsigned long long) dus->s.time_in_queue,
1039 util);
1040 } else {
1041 log_buf(out, ";%s;%llu;%llu;%llu;%llu;%llu;%llu;%llu;%3.2f%%",
1042 dus->name,
1043 (unsigned long long) dus->s.ios[0],
1044 (unsigned long long) dus->s.ios[1],
1045 (unsigned long long) dus->s.merges[0],
1046 (unsigned long long) dus->s.merges[1],
1047 (unsigned long long) dus->s.ticks[0],
1048 (unsigned long long) dus->s.ticks[1],
1049 (unsigned long long) dus->s.time_in_queue,
1050 util);
1051 }
1052
1053 /*
1054 * If the device has slaves, aggregate the stats for
1055 * those slave devices also.
1056 */
1057 show_agg_stats(agg, terse, out);
1058
1059 if (!terse)
1060 log_buf(out, "\n");
1061}
1062
1063void json_array_add_disk_util(struct disk_util_stat *dus,
1064 struct disk_util_agg *agg, struct json_array *array)
1065{
1066 struct json_object *obj;
1067 double util = 0;
1068
1069 if (dus->s.msec)
1070 util = (double) 100 * dus->s.io_ticks / (double) dus->s.msec;
1071 if (util > 100.0)
1072 util = 100.0;
1073
1074 obj = json_create_object();
1075 json_array_add_value_object(array, obj);
1076
eb2f29b7 1077 json_object_add_value_string(obj, "name", (const char *)dus->name);
2a2fdab1
L
1078 json_object_add_value_int(obj, "read_ios", dus->s.ios[0]);
1079 json_object_add_value_int(obj, "write_ios", dus->s.ios[1]);
1080 json_object_add_value_int(obj, "read_merges", dus->s.merges[0]);
1081 json_object_add_value_int(obj, "write_merges", dus->s.merges[1]);
1082 json_object_add_value_int(obj, "read_ticks", dus->s.ticks[0]);
1083 json_object_add_value_int(obj, "write_ticks", dus->s.ticks[1]);
1084 json_object_add_value_int(obj, "in_queue", dus->s.time_in_queue);
1085 json_object_add_value_float(obj, "util", util);
1086
1087 /*
1088 * If the device has slaves, aggregate the stats for
1089 * those slave devices also.
1090 */
1091 if (!agg->slavecount)
1092 return;
1093 json_object_add_value_int(obj, "aggr_read_ios",
1094 agg->ios[0] / agg->slavecount);
1095 json_object_add_value_int(obj, "aggr_write_ios",
1096 agg->ios[1] / agg->slavecount);
1097 json_object_add_value_int(obj, "aggr_read_merges",
1098 agg->merges[0] / agg->slavecount);
1099 json_object_add_value_int(obj, "aggr_write_merge",
1100 agg->merges[1] / agg->slavecount);
1101 json_object_add_value_int(obj, "aggr_read_ticks",
1102 agg->ticks[0] / agg->slavecount);
1103 json_object_add_value_int(obj, "aggr_write_ticks",
1104 agg->ticks[1] / agg->slavecount);
1105 json_object_add_value_int(obj, "aggr_in_queue",
1106 agg->time_in_queue / agg->slavecount);
1107 json_object_add_value_float(obj, "aggr_util", agg->max_util.u.f);
1108}
1109
1110static void json_object_add_disk_utils(struct json_object *obj,
1111 struct flist_head *head)
1112{
1113 struct json_array *array = json_create_array();
1114 struct flist_head *entry;
1115 struct disk_util *du;
1116
1117 json_object_add_value_array(obj, "disk_util", array);
1118
1119 flist_for_each(entry, head) {
1120 du = flist_entry(entry, struct disk_util, list);
1121
1122 aggregate_slaves_stats(du);
1123 json_array_add_disk_util(&du->dus, &du->agg, array);
1124 }
1125}
1126
1127void show_disk_util(int terse, struct json_object *parent,
1128 struct buf_output *out)
1129{
1130 struct flist_head *entry;
1131 struct disk_util *du;
1132 bool do_json;
1133
1134 if (!is_running_backend())
1135 return;
1136
6dc75062 1137 if (flist_empty(&disk_list))
2a2fdab1 1138 return;
2a2fdab1
L
1139
1140 if ((output_format & FIO_OUTPUT_JSON) && parent)
1141 do_json = true;
1142 else
1143 do_json = false;
1144
1145 if (!terse && !do_json)
1146 log_buf(out, "\nDisk stats (read/write):\n");
1147
6dc75062 1148 if (do_json) {
2a2fdab1 1149 json_object_add_disk_utils(parent, &disk_list);
6dc75062 1150 } else if (output_format & ~(FIO_OUTPUT_JSON | FIO_OUTPUT_JSON_PLUS)) {
2a2fdab1
L
1151 flist_for_each(entry, &disk_list) {
1152 du = flist_entry(entry, struct disk_util, list);
1153
1154 aggregate_slaves_stats(du);
1155 print_disk_util(&du->dus, &du->agg, terse, out);
1156 }
1157 }
1158}
1159
10aa136b 1160static void show_thread_status_normal(struct thread_stat *ts,
a666cab8
JA
1161 struct group_run_stats *rs,
1162 struct buf_output *out)
3c39a379
JA
1163{
1164 double usr_cpu, sys_cpu;
69008999 1165 unsigned long runtime;
71619dc2 1166 double io_u_dist[FIO_IO_U_MAP_NR];
57a64324 1167 time_t time_p;
35326842 1168 char time_buf[32];
3c39a379 1169
9966af7b 1170 if (!ddir_rw_sum(ts->io_bytes) && !ddir_rw_sum(ts->total_io_u))
3c39a379 1171 return;
7a4e480d 1172
33477b4b 1173 memset(time_buf, 0, sizeof(time_buf));
3c39a379 1174
57a64324 1175 time(&time_p);
45054cbe 1176 os_ctime_r((const time_t *) &time_p, time_buf, sizeof(time_buf));
57a64324 1177
5ec10eaa 1178 if (!ts->error) {
a666cab8 1179 log_buf(out, "%s: (groupid=%d, jobs=%d): err=%2d: pid=%d: %s",
5ec10eaa 1180 ts->name, ts->groupid, ts->members,
57a64324 1181 ts->error, (int) ts->pid, time_buf);
5ec10eaa 1182 } else {
a666cab8 1183 log_buf(out, "%s: (groupid=%d, jobs=%d): err=%2d (%s): pid=%d: %s",
5ec10eaa 1184 ts->name, ts->groupid, ts->members,
57a64324
JA
1185 ts->error, ts->verror, (int) ts->pid,
1186 time_buf);
5ec10eaa 1187 }
3c39a379 1188
259e47de 1189 if (strlen(ts->description))
a666cab8 1190 log_buf(out, " Description : [%s]\n", ts->description);
7bdce1bd 1191
dde39f8c
AD
1192 for_each_rw_ddir(ddir) {
1193 if (ts->io_bytes[ddir])
1194 show_ddir_status(rs, ts, ddir, out);
1195 }
3c39a379 1196
5cb8a8cd
BP
1197 if (ts->unified_rw_rep == UNIFIED_BOTH)
1198 show_mixed_ddir_status(rs, ts, out);
1199
a666cab8 1200 show_latencies(ts, out);
7e1773ba 1201
b2b3eefe
JA
1202 if (ts->sync_stat.samples)
1203 show_ddir_status(rs, ts, DDIR_SYNC, out);
1204
756867bd 1205 runtime = ts->total_run_time;
69008999 1206 if (runtime) {
1e97cce9 1207 double runt = (double) runtime;
3c39a379 1208
756867bd
JA
1209 usr_cpu = (double) ts->usr_time * 100 / runt;
1210 sys_cpu = (double) ts->sys_time * 100 / runt;
3c39a379
JA
1211 } else {
1212 usr_cpu = 0;
1213 sys_cpu = 0;
1214 }
1215
a666cab8 1216 log_buf(out, " cpu : usr=%3.2f%%, sys=%3.2f%%, ctx=%llu,"
4e0a8fa2
JA
1217 " majf=%llu, minf=%llu\n", usr_cpu, sys_cpu,
1218 (unsigned long long) ts->ctx,
1219 (unsigned long long) ts->majf,
1220 (unsigned long long) ts->minf);
71619dc2 1221
d79db122 1222 stat_calc_dist(ts->io_u_map, ddir_rw_sum(ts->total_io_u), io_u_dist);
a666cab8 1223 log_buf(out, " IO depths : 1=%3.1f%%, 2=%3.1f%%, 4=%3.1f%%, 8=%3.1f%%,"
5ec10eaa
JA
1224 " 16=%3.1f%%, 32=%3.1f%%, >=64=%3.1f%%\n", io_u_dist[0],
1225 io_u_dist[1], io_u_dist[2],
1226 io_u_dist[3], io_u_dist[4],
1227 io_u_dist[5], io_u_dist[6]);
838bc709
JA
1228
1229 stat_calc_dist(ts->io_u_submit, ts->total_submit, io_u_dist);
a666cab8 1230 log_buf(out, " submit : 0=%3.1f%%, 4=%3.1f%%, 8=%3.1f%%, 16=%3.1f%%,"
838bc709
JA
1231 " 32=%3.1f%%, 64=%3.1f%%, >=64=%3.1f%%\n", io_u_dist[0],
1232 io_u_dist[1], io_u_dist[2],
1233 io_u_dist[3], io_u_dist[4],
1234 io_u_dist[5], io_u_dist[6]);
1235 stat_calc_dist(ts->io_u_complete, ts->total_complete, io_u_dist);
a666cab8 1236 log_buf(out, " complete : 0=%3.1f%%, 4=%3.1f%%, 8=%3.1f%%, 16=%3.1f%%,"
838bc709
JA
1237 " 32=%3.1f%%, 64=%3.1f%%, >=64=%3.1f%%\n", io_u_dist[0],
1238 io_u_dist[1], io_u_dist[2],
1239 io_u_dist[3], io_u_dist[4],
1240 io_u_dist[5], io_u_dist[6]);
7f3ecee2
JA
1241 log_buf(out, " issued rwts: total=%llu,%llu,%llu,%llu"
1242 " short=%llu,%llu,%llu,0"
1243 " dropped=%llu,%llu,%llu,0\n",
4e0a8fa2
JA
1244 (unsigned long long) ts->total_io_u[0],
1245 (unsigned long long) ts->total_io_u[1],
1246 (unsigned long long) ts->total_io_u[2],
7f3ecee2 1247 (unsigned long long) ts->total_io_u[3],
4e0a8fa2
JA
1248 (unsigned long long) ts->short_io_u[0],
1249 (unsigned long long) ts->short_io_u[1],
3bcb9d94
JA
1250 (unsigned long long) ts->short_io_u[2],
1251 (unsigned long long) ts->drop_io_u[0],
1252 (unsigned long long) ts->drop_io_u[1],
1253 (unsigned long long) ts->drop_io_u[2]);
f2bba182 1254 if (ts->continue_on_error) {
a666cab8 1255 log_buf(out, " errors : total=%llu, first_error=%d/<%s>\n",
4e0a8fa2 1256 (unsigned long long)ts->total_err_count,
1ec99eea
JA
1257 ts->first_error,
1258 strerror(ts->first_error));
f2bba182 1259 }
3e260a46 1260 if (ts->latency_depth) {
a666cab8 1261 log_buf(out, " latency : target=%llu, window=%llu, percentile=%.2f%%, depth=%u\n",
3e260a46
JA
1262 (unsigned long long)ts->latency_target,
1263 (unsigned long long)ts->latency_window,
1264 ts->latency_percentile.u.f,
1265 ts->latency_depth);
1266 }
66347cfa
DE
1267
1268 if (ts->nr_block_infos)
1269 show_block_infos(ts->nr_block_infos, ts->block_infos,
a666cab8 1270 ts->percentile_list, out);
d685adfb 1271
bb49c8bd 1272 if (ts->ss_dur)
d685adfb 1273 show_ss_normal(ts, out);
3c39a379
JA
1274}
1275
756867bd 1276static void show_ddir_status_terse(struct thread_stat *ts,
2e545522
NC
1277 struct group_run_stats *rs,
1278 enum fio_ddir ddir, int ver,
1279 struct buf_output *out)
c6ae0a5b 1280{
d6bb626e
VF
1281 unsigned long long min, max, minv, maxv, bw, iops;
1282 unsigned long long *ovals = NULL;
c6ae0a5b 1283 double mean, dev;
d6bb626e 1284 unsigned int len;
a2c95580 1285 int i, bw_stat;
c6ae0a5b 1286
ff58fced
JA
1287 assert(ddir_rw(ddir));
1288
312b4af2
JA
1289 iops = bw = 0;
1290 if (ts->runtime[ddir]) {
1291 uint64_t runt = ts->runtime[ddir];
1292
420b104a 1293 bw = ((1000 * ts->io_bytes[ddir]) / runt) / 1024; /* KiB/s */
312b4af2
JA
1294 iops = (1000 * (uint64_t) ts->total_io_u[ddir]) / runt;
1295 }
c6ae0a5b 1296
a666cab8 1297 log_buf(out, ";%llu;%llu;%llu;%llu",
4e0a8fa2
JA
1298 (unsigned long long) ts->io_bytes[ddir] >> 10, bw, iops,
1299 (unsigned long long) ts->runtime[ddir]);
c6ae0a5b 1300
079ad09b 1301 if (calc_lat(&ts->slat_stat[ddir], &min, &max, &mean, &dev))
247823cc 1302 log_buf(out, ";%llu;%llu;%f;%f", min/1000, max/1000, mean/1000, dev/1000);
c6ae0a5b 1303 else
d6bb626e 1304 log_buf(out, ";%llu;%llu;%f;%f", 0ULL, 0ULL, 0.0, 0.0);
c6ae0a5b 1305
079ad09b 1306 if (calc_lat(&ts->clat_stat[ddir], &min, &max, &mean, &dev))
247823cc 1307 log_buf(out, ";%llu;%llu;%f;%f", min/1000, max/1000, mean/1000, dev/1000);
c6ae0a5b 1308 else
d6bb626e 1309 log_buf(out, ";%llu;%llu;%f;%f", 0ULL, 0ULL, 0.0, 0.0);
c6ae0a5b 1310
6dc75062 1311 if (ts->lat_percentiles) {
56440e63
VF
1312 len = calc_clat_percentiles(ts->io_u_plat[FIO_LAT][ddir],
1313 ts->lat_stat[ddir].samples,
1314 ts->percentile_list, &ovals, &maxv,
1315 &minv);
6dc75062 1316 } else if (ts->clat_percentiles) {
df8781b6 1317 len = calc_clat_percentiles(ts->io_u_plat[FIO_CLAT][ddir],
1db92cb6
JA
1318 ts->clat_stat[ddir].samples,
1319 ts->percentile_list, &ovals, &maxv,
1320 &minv);
6dc75062 1321 } else {
1db92cb6 1322 len = 0;
6dc75062
JA
1323 }
1324
1db92cb6
JA
1325 for (i = 0; i < FIO_IO_U_LIST_MAX_LEN; i++) {
1326 if (i >= len) {
a666cab8 1327 log_buf(out, ";0%%=0");
1db92cb6
JA
1328 continue;
1329 }
247823cc 1330 log_buf(out, ";%f%%=%llu", ts->percentile_list[i].u.f, ovals[i]/1000);
1db92cb6 1331 }
2341a37a
K
1332
1333 if (calc_lat(&ts->lat_stat[ddir], &min, &max, &mean, &dev))
247823cc 1334 log_buf(out, ";%llu;%llu;%f;%f", min/1000, max/1000, mean/1000, dev/1000);
2341a37a 1335 else
d6bb626e 1336 log_buf(out, ";%llu;%llu;%f;%f", 0ULL, 0ULL, 0.0, 0.0);
2341a37a 1337
283feb79 1338 free(ovals);
1db92cb6 1339
a2c95580
AH
1340 bw_stat = calc_lat(&ts->bw_stat[ddir], &min, &max, &mean, &dev);
1341 if (bw_stat) {
19d3e967
JA
1342 double p_of_agg = 100.0;
1343
1344 if (rs->agg[ddir]) {
d1707474 1345 p_of_agg = mean * 100 / (double) (rs->agg[ddir] / 1024);
19d3e967
JA
1346 if (p_of_agg > 100.0)
1347 p_of_agg = 100.0;
1348 }
c6ae0a5b 1349
d6bb626e 1350 log_buf(out, ";%llu;%llu;%f%%;%f;%f", min, max, p_of_agg, mean, dev);
6dc75062 1351 } else {
d6bb626e 1352 log_buf(out, ";%llu;%llu;%f%%;%f;%f", 0ULL, 0ULL, 0.0, 0.0, 0.0);
6dc75062 1353 }
a2c95580
AH
1354
1355 if (ver == 5) {
1356 if (bw_stat)
29eb371b 1357 log_buf(out, ";%" PRIu64, (&ts->bw_stat[ddir])->samples);
a2c95580
AH
1358 else
1359 log_buf(out, ";%lu", 0UL);
1360
1361 if (calc_lat(&ts->iops_stat[ddir], &min, &max, &mean, &dev))
29eb371b 1362 log_buf(out, ";%llu;%llu;%f;%f;%" PRIu64, min, max,
a2c95580
AH
1363 mean, dev, (&ts->iops_stat[ddir])->samples);
1364 else
1365 log_buf(out, ";%llu;%llu;%f;%f;%lu", 0ULL, 0ULL, 0.0, 0.0, 0UL);
1366 }
c6ae0a5b
JA
1367}
1368
5cb8a8cd
BP
1369static void show_mixed_ddir_status_terse(struct thread_stat *ts,
1370 struct group_run_stats *rs,
1371 int ver, struct buf_output *out)
1372{
b182f077 1373 struct thread_stat *ts_lcl = gen_mixed_ddir_stats_from_ts(ts);
5cb8a8cd 1374
b182f077
NC
1375 if (ts_lcl)
1376 show_ddir_status_terse(ts_lcl, rs, DDIR_READ, ver, out);
5cb8a8cd 1377
692dec0c 1378 free_clat_prio_stats(ts_lcl);
5cb8a8cd
BP
1379 free(ts_lcl);
1380}
1381
6dc75062
JA
1382static struct json_object *add_ddir_lat_json(struct thread_stat *ts,
1383 uint32_t percentiles,
1384 struct io_stat *lat_stat,
1385 uint64_t *io_u_plat)
56440e63
VF
1386{
1387 char buf[120];
1388 double mean, dev;
1389 unsigned int i, len;
1390 struct json_object *lat_object, *percentile_object, *clat_bins_object;
1391 unsigned long long min, max, maxv, minv, *ovals = NULL;
1392
1393 if (!calc_lat(lat_stat, &min, &max, &mean, &dev)) {
1394 min = max = 0;
1395 mean = dev = 0.0;
1396 }
1397 lat_object = json_create_object();
1398 json_object_add_value_int(lat_object, "min", min);
1399 json_object_add_value_int(lat_object, "max", max);
1400 json_object_add_value_float(lat_object, "mean", mean);
1401 json_object_add_value_float(lat_object, "stddev", dev);
1402 json_object_add_value_int(lat_object, "N", lat_stat->samples);
1403
1404 if (percentiles && lat_stat->samples) {
1405 len = calc_clat_percentiles(io_u_plat, lat_stat->samples,
1406 ts->percentile_list, &ovals, &maxv, &minv);
1407
1408 if (len > FIO_IO_U_LIST_MAX_LEN)
1409 len = FIO_IO_U_LIST_MAX_LEN;
1410
1411 percentile_object = json_create_object();
1412 json_object_add_value_object(lat_object, "percentile", percentile_object);
1413 for (i = 0; i < len; i++) {
1414 snprintf(buf, sizeof(buf), "%f", ts->percentile_list[i].u.f);
1415 json_object_add_value_int(percentile_object, buf, ovals[i]);
1416 }
1417 free(ovals);
1418
1419 if (output_format & FIO_OUTPUT_JSON_PLUS) {
1420 clat_bins_object = json_create_object();
1421 json_object_add_value_object(lat_object, "bins", clat_bins_object);
1422
1423 for(i = 0; i < FIO_IO_U_PLAT_NR; i++)
1424 if (io_u_plat[i]) {
1425 snprintf(buf, sizeof(buf), "%llu", plat_idx_to_val(i));
1426 json_object_add_value_int(clat_bins_object, buf, io_u_plat[i]);
1427 }
1428 }
1429 }
1430
1431 return lat_object;
1432}
1433
cc372b17 1434static void add_ddir_status_json(struct thread_stat *ts,
2e545522
NC
1435 struct group_run_stats *rs, enum fio_ddir ddir,
1436 struct json_object *parent)
cc372b17 1437{
56440e63 1438 unsigned long long min, max;
aedd021d 1439 unsigned long long bw_bytes, bw;
9f68fe3a 1440 double mean, dev, iops;
56440e63 1441 struct json_object *dir_object, *tmp_object;
cc372b17
SL
1442 double p_of_agg = 100.0;
1443
b2b3eefe 1444 assert(ddir_rw(ddir) || ddir_sync(ddir));
cc372b17 1445
5cb8a8cd 1446 if ((ts->unified_rw_rep == UNIFIED_MIXED) && ddir != DDIR_READ)
771e58be
JA
1447 return;
1448
cc372b17 1449 dir_object = json_create_object();
771e58be 1450 json_object_add_value_object(parent,
5cb8a8cd 1451 (ts->unified_rw_rep == UNIFIED_MIXED) ? "mixed" : io_ddir_name(ddir), dir_object);
cc372b17 1452
b2b3eefe
JA
1453 if (ddir_rw(ddir)) {
1454 bw_bytes = 0;
1455 bw = 0;
1456 iops = 0.0;
1457 if (ts->runtime[ddir]) {
1458 uint64_t runt = ts->runtime[ddir];
cc372b17 1459
b2b3eefe
JA
1460 bw_bytes = ((1000 * ts->io_bytes[ddir]) / runt); /* Bytes/s */
1461 bw = bw_bytes / 1024; /* KiB/s */
1462 iops = (1000.0 * (uint64_t) ts->total_io_u[ddir]) / runt;
1463 }
cc372b17 1464
b2b3eefe
JA
1465 json_object_add_value_int(dir_object, "io_bytes", ts->io_bytes[ddir]);
1466 json_object_add_value_int(dir_object, "io_kbytes", ts->io_bytes[ddir] >> 10);
1467 json_object_add_value_int(dir_object, "bw_bytes", bw_bytes);
1468 json_object_add_value_int(dir_object, "bw", bw);
1469 json_object_add_value_float(dir_object, "iops", iops);
1470 json_object_add_value_int(dir_object, "runtime", ts->runtime[ddir]);
1471 json_object_add_value_int(dir_object, "total_ios", ts->total_io_u[ddir]);
1472 json_object_add_value_int(dir_object, "short_ios", ts->short_io_u[ddir]);
1473 json_object_add_value_int(dir_object, "drop_ios", ts->drop_io_u[ddir]);
1474
56440e63
VF
1475 tmp_object = add_ddir_lat_json(ts, ts->slat_percentiles,
1476 &ts->slat_stat[ddir], ts->io_u_plat[FIO_SLAT][ddir]);
b2b3eefe 1477 json_object_add_value_object(dir_object, "slat_ns", tmp_object);
56440e63
VF
1478
1479 tmp_object = add_ddir_lat_json(ts, ts->clat_percentiles,
1480 &ts->clat_stat[ddir], ts->io_u_plat[FIO_CLAT][ddir]);
b2b3eefe 1481 json_object_add_value_object(dir_object, "clat_ns", tmp_object);
cc372b17 1482
56440e63
VF
1483 tmp_object = add_ddir_lat_json(ts, ts->lat_percentiles,
1484 &ts->lat_stat[ddir], ts->io_u_plat[FIO_LAT][ddir]);
b2b3eefe 1485 json_object_add_value_object(dir_object, "lat_ns", tmp_object);
56440e63 1486 } else {
f5ec8123 1487 json_object_add_value_int(dir_object, "total_ios", ts->total_io_u[DDIR_SYNC]);
56440e63
VF
1488 tmp_object = add_ddir_lat_json(ts, ts->lat_percentiles | ts->clat_percentiles,
1489 &ts->sync_stat, ts->io_u_sync_plat);
1490 json_object_add_value_object(dir_object, "lat_ns", tmp_object);
513e37ee
VF
1491 }
1492
56440e63
VF
1493 if (!ddir_rw(ddir))
1494 return;
b2a432bf
PC
1495
1496 /* Only print PRIO latencies if some high priority samples were gathered */
1497 if (ts->clat_high_prio_stat[ddir].samples > 0) {
56440e63 1498 const char *high, *low;
b2a432bf 1499
56440e63 1500 if (ts->lat_percentiles) {
efd78265 1501 high = "lat_high_prio";
56440e63
VF
1502 low = "lat_low_prio";
1503 } else {
efd78265 1504 high = "clat_high_prio";
56440e63 1505 low = "clat_low_prio";
b2a432bf 1506 }
b2a432bf 1507
56440e63
VF
1508 tmp_object = add_ddir_lat_json(ts, ts->clat_percentiles | ts->lat_percentiles,
1509 &ts->clat_high_prio_stat[ddir], ts->io_u_plat_high_prio[ddir]);
1510 json_object_add_value_object(dir_object, high, tmp_object);
b2b3eefe 1511
56440e63 1512 tmp_object = add_ddir_lat_json(ts, ts->clat_percentiles | ts->lat_percentiles,
efd78265 1513 &ts->clat_low_prio_stat[ddir], ts->io_u_plat_low_prio[ddir]);
56440e63 1514 json_object_add_value_object(dir_object, low, tmp_object);
cc372b17 1515 }
24cab44e 1516
b9e1b491 1517 if (calc_lat(&ts->bw_stat[ddir], &min, &max, &mean, &dev)) {
5c0abd5e 1518 p_of_agg = convert_agg_kbytes_percent(rs, ddir, mean);
cc372b17
SL
1519 } else {
1520 min = max = 0;
1521 p_of_agg = mean = dev = 0.0;
1522 }
56440e63 1523
cc372b17
SL
1524 json_object_add_value_int(dir_object, "bw_min", min);
1525 json_object_add_value_int(dir_object, "bw_max", max);
a806bf2e 1526 json_object_add_value_float(dir_object, "bw_agg", p_of_agg);
cc372b17
SL
1527 json_object_add_value_float(dir_object, "bw_mean", mean);
1528 json_object_add_value_float(dir_object, "bw_dev", dev);
54c05828
AH
1529 json_object_add_value_int(dir_object, "bw_samples",
1530 (&ts->bw_stat[ddir])->samples);
188b6016
AH
1531
1532 if (!calc_lat(&ts->iops_stat[ddir], &min, &max, &mean, &dev)) {
1533 min = max = 0;
1534 mean = dev = 0.0;
1535 }
1536 json_object_add_value_int(dir_object, "iops_min", min);
1537 json_object_add_value_int(dir_object, "iops_max", max);
1538 json_object_add_value_float(dir_object, "iops_mean", mean);
1539 json_object_add_value_float(dir_object, "iops_stddev", dev);
54c05828
AH
1540 json_object_add_value_int(dir_object, "iops_samples",
1541 (&ts->iops_stat[ddir])->samples);
96563db9
JA
1542
1543 if (ts->cachehit + ts->cachemiss) {
1544 uint64_t total;
1545 double hit;
1546
1547 total = ts->cachehit + ts->cachemiss;
1548 hit = (double) ts->cachehit / (double) total;
1549 hit *= 100.0;
1550 json_object_add_value_float(dir_object, "cachehit", hit);
1551 }
cc372b17
SL
1552}
1553
5cb8a8cd
BP
1554static void add_mixed_ddir_status_json(struct thread_stat *ts,
1555 struct group_run_stats *rs, struct json_object *parent)
1556{
b182f077 1557 struct thread_stat *ts_lcl = gen_mixed_ddir_stats_from_ts(ts);
5cb8a8cd
BP
1558
1559 /* add the aggregated stats to json parent */
b182f077
NC
1560 if (ts_lcl)
1561 add_ddir_status_json(ts_lcl, rs, DDIR_READ, parent);
1562
692dec0c 1563 free_clat_prio_stats(ts_lcl);
5cb8a8cd
BP
1564 free(ts_lcl);
1565}
1566
bef2112b
AH
1567static void show_thread_status_terse_all(struct thread_stat *ts,
1568 struct group_run_stats *rs, int ver,
1569 struct buf_output *out)
4d658652
JA
1570{
1571 double io_u_dist[FIO_IO_U_MAP_NR];
1572 double io_u_lat_u[FIO_IO_U_LAT_U_NR];
1573 double io_u_lat_m[FIO_IO_U_LAT_M_NR];
1574 double usr_cpu, sys_cpu;
1575 int i;
1576
1577 /* General Info */
bef2112b
AH
1578 if (ver == 2)
1579 log_buf(out, "2;%s;%d;%d", ts->name, ts->groupid, ts->error);
1580 else
1581 log_buf(out, "%d;%s;%s;%d;%d", ver, fio_version_string,
1582 ts->name, ts->groupid, ts->error);
c6ae0a5b 1583
5cb8a8cd 1584 /* Log Read Status, or mixed if unified_rw_rep = 1 */
a2c95580 1585 show_ddir_status_terse(ts, rs, DDIR_READ, ver, out);
5cb8a8cd
BP
1586 if (ts->unified_rw_rep != UNIFIED_MIXED) {
1587 /* Log Write Status */
1588 show_ddir_status_terse(ts, rs, DDIR_WRITE, ver, out);
1589 /* Log Trim Status */
1590 if (ver == 2 || ver == 4 || ver == 5)
1591 show_ddir_status_terse(ts, rs, DDIR_TRIM, ver, out);
1592 }
1593 if (ts->unified_rw_rep == UNIFIED_BOTH)
1594 show_mixed_ddir_status_terse(ts, rs, ver, out);
562c2d2f 1595 /* CPU Usage */
756867bd
JA
1596 if (ts->total_run_time) {
1597 double runt = (double) ts->total_run_time;
c6ae0a5b 1598
756867bd
JA
1599 usr_cpu = (double) ts->usr_time * 100 / runt;
1600 sys_cpu = (double) ts->sys_time * 100 / runt;
c6ae0a5b
JA
1601 } else {
1602 usr_cpu = 0;
1603 sys_cpu = 0;
1604 }
1605
a666cab8 1606 log_buf(out, ";%f%%;%f%%;%llu;%llu;%llu", usr_cpu, sys_cpu,
4e0a8fa2
JA
1607 (unsigned long long) ts->ctx,
1608 (unsigned long long) ts->majf,
1609 (unsigned long long) ts->minf);
2270890c 1610
562c2d2f 1611 /* Calc % distribution of IO depths, usecond, msecond latency */
d79db122 1612 stat_calc_dist(ts->io_u_map, ddir_rw_sum(ts->total_io_u), io_u_dist);
247823cc 1613 stat_calc_lat_nu(ts, io_u_lat_u);
04a0feae 1614 stat_calc_lat_m(ts, io_u_lat_m);
2270890c 1615
562c2d2f 1616 /* Only show fixed 7 I/O depth levels*/
a666cab8 1617 log_buf(out, ";%3.1f%%;%3.1f%%;%3.1f%%;%3.1f%%;%3.1f%%;%3.1f%%;%3.1f%%",
5ec10eaa
JA
1618 io_u_dist[0], io_u_dist[1], io_u_dist[2], io_u_dist[3],
1619 io_u_dist[4], io_u_dist[5], io_u_dist[6]);
2270890c 1620
562c2d2f 1621 /* Microsecond latency */
04a0feae 1622 for (i = 0; i < FIO_IO_U_LAT_U_NR; i++)
a666cab8 1623 log_buf(out, ";%3.2f%%", io_u_lat_u[i]);
562c2d2f 1624 /* Millisecond latency */
04a0feae 1625 for (i = 0; i < FIO_IO_U_LAT_M_NR; i++)
a666cab8 1626 log_buf(out, ";%3.2f%%", io_u_lat_m[i]);
f2f788dd
JA
1627
1628 /* disk util stats, if any */
24a97130 1629 if (ver >= 3 && is_running_backend())
bef2112b 1630 show_disk_util(1, NULL, out);
f2f788dd 1631
562c2d2f 1632 /* Additional output if continue_on_error set - default off*/
f2bba182 1633 if (ts->continue_on_error)
a666cab8 1634 log_buf(out, ";%llu;%d", (unsigned long long) ts->total_err_count, ts->first_error);
2270890c 1635
562c2d2f 1636 /* Additional output if description is set */
d5495f0b
VF
1637 if (strlen(ts->description)) {
1638 if (ver == 2)
1639 log_buf(out, "\n");
a666cab8 1640 log_buf(out, ";%s", ts->description);
d5495f0b 1641 }
946e4276 1642
a666cab8 1643 log_buf(out, "\n");
756867bd
JA
1644}
1645
a89ba4b1 1646static void json_add_job_opts(struct json_object *root, const char *name,
0cf542af 1647 struct flist_head *opt_list)
66e19a38
JA
1648{
1649 struct json_object *dir_object;
1650 struct flist_head *entry;
1651 struct print_option *p;
1652
1653 if (flist_empty(opt_list))
1654 return;
1655
1656 dir_object = json_create_object();
1657 json_object_add_value_object(root, name, dir_object);
1658
1659 flist_for_each(entry, opt_list) {
66e19a38 1660 p = flist_entry(entry, struct print_option, list);
c42b48fe 1661 json_object_add_value_string(dir_object, p->name, p->value);
66e19a38
JA
1662 }
1663}
1664
cc372b17 1665static struct json_object *show_thread_status_json(struct thread_stat *ts,
66e19a38
JA
1666 struct group_run_stats *rs,
1667 struct flist_head *opt_list)
cc372b17
SL
1668{
1669 struct json_object *root, *tmp;
b01af66b 1670 struct jobs_eta *je;
cc372b17 1671 double io_u_dist[FIO_IO_U_MAP_NR];
d6bb626e 1672 double io_u_lat_n[FIO_IO_U_LAT_N_NR];
cc372b17
SL
1673 double io_u_lat_u[FIO_IO_U_LAT_U_NR];
1674 double io_u_lat_m[FIO_IO_U_LAT_M_NR];
1675 double usr_cpu, sys_cpu;
1676 int i;
b01af66b
CJ
1677 size_t size;
1678
cc372b17
SL
1679 root = json_create_object();
1680 json_object_add_value_string(root, "jobname", ts->name);
1681 json_object_add_value_int(root, "groupid", ts->groupid);
1682 json_object_add_value_int(root, "error", ts->error);
1683
b01af66b 1684 /* ETA Info */
c5103619 1685 je = get_jobs_eta(true, &size);
2de615ad
JA
1686 if (je) {
1687 json_object_add_value_int(root, "eta", je->eta_sec);
1688 json_object_add_value_int(root, "elapsed", je->elapsed_sec);
1689 }
b01af66b 1690
66e19a38 1691 if (opt_list)
0cf542af 1692 json_add_job_opts(root, "job options", opt_list);
66e19a38 1693
cc372b17
SL
1694 add_ddir_status_json(ts, rs, DDIR_READ, root);
1695 add_ddir_status_json(ts, rs, DDIR_WRITE, root);
f3afa57e 1696 add_ddir_status_json(ts, rs, DDIR_TRIM, root);
b2b3eefe 1697 add_ddir_status_json(ts, rs, DDIR_SYNC, root);
cc372b17 1698
5cb8a8cd
BP
1699 if (ts->unified_rw_rep == UNIFIED_BOTH)
1700 add_mixed_ddir_status_json(ts, rs, root);
1701
cc372b17
SL
1702 /* CPU Usage */
1703 if (ts->total_run_time) {
1704 double runt = (double) ts->total_run_time;
1705
1706 usr_cpu = (double) ts->usr_time * 100 / runt;
1707 sys_cpu = (double) ts->sys_time * 100 / runt;
1708 } else {
1709 usr_cpu = 0;
1710 sys_cpu = 0;
1711 }
bcbb4c6c 1712 json_object_add_value_int(root, "job_runtime", ts->total_run_time);
cc372b17
SL
1713 json_object_add_value_float(root, "usr_cpu", usr_cpu);
1714 json_object_add_value_float(root, "sys_cpu", sys_cpu);
1715 json_object_add_value_int(root, "ctx", ts->ctx);
1716 json_object_add_value_int(root, "majf", ts->majf);
1717 json_object_add_value_int(root, "minf", ts->minf);
1718
ec3e3648 1719 /* Calc % distribution of IO depths */
d79db122 1720 stat_calc_dist(ts->io_u_map, ddir_rw_sum(ts->total_io_u), io_u_dist);
cc372b17
SL
1721 tmp = json_create_object();
1722 json_object_add_value_object(root, "iodepth_level", tmp);
1723 /* Only show fixed 7 I/O depth levels*/
1724 for (i = 0; i < 7; i++) {
1725 char name[20];
1726 if (i < 6)
98ffb8f3 1727 snprintf(name, 20, "%d", 1 << i);
cc372b17 1728 else
98ffb8f3 1729 snprintf(name, 20, ">=%d", 1 << i);
cc372b17
SL
1730 json_object_add_value_float(tmp, (const char *)name, io_u_dist[i]);
1731 }
1732
ec3e3648
VF
1733 /* Calc % distribution of submit IO depths */
1734 stat_calc_dist(ts->io_u_submit, ts->total_submit, io_u_dist);
1735 tmp = json_create_object();
1736 json_object_add_value_object(root, "iodepth_submit", tmp);
1737 /* Only show fixed 7 I/O depth levels*/
1738 for (i = 0; i < 7; i++) {
1739 char name[20];
1740 if (i == 0)
1741 snprintf(name, 20, "0");
1742 else if (i < 6)
1743 snprintf(name, 20, "%d", 1 << (i+1));
1744 else
1745 snprintf(name, 20, ">=%d", 1 << i);
1746 json_object_add_value_float(tmp, (const char *)name, io_u_dist[i]);
1747 }
1748
1749 /* Calc % distribution of completion IO depths */
1750 stat_calc_dist(ts->io_u_complete, ts->total_complete, io_u_dist);
1751 tmp = json_create_object();
1752 json_object_add_value_object(root, "iodepth_complete", tmp);
1753 /* Only show fixed 7 I/O depth levels*/
1754 for (i = 0; i < 7; i++) {
1755 char name[20];
1756 if (i == 0)
1757 snprintf(name, 20, "0");
1758 else if (i < 6)
1759 snprintf(name, 20, "%d", 1 << (i+1));
1760 else
1761 snprintf(name, 20, ">=%d", 1 << i);
1762 json_object_add_value_float(tmp, (const char *)name, io_u_dist[i]);
1763 }
1764
1765 /* Calc % distribution of nsecond, usecond, msecond latency */
1766 stat_calc_dist(ts->io_u_map, ddir_rw_sum(ts->total_io_u), io_u_dist);
1767 stat_calc_lat_n(ts, io_u_lat_n);
1768 stat_calc_lat_u(ts, io_u_lat_u);
1769 stat_calc_lat_m(ts, io_u_lat_m);
1770
d6bb626e 1771 /* Nanosecond latency */
cc372b17 1772 tmp = json_create_object();
d6bb626e
VF
1773 json_object_add_value_object(root, "latency_ns", tmp);
1774 for (i = 0; i < FIO_IO_U_LAT_N_NR; i++) {
1775 const char *ranges[] = { "2", "4", "10", "20", "50", "100",
1776 "250", "500", "750", "1000", };
1777 json_object_add_value_float(tmp, ranges[i], io_u_lat_n[i]);
1778 }
cc372b17 1779 /* Microsecond latency */
d6bb626e
VF
1780 tmp = json_create_object();
1781 json_object_add_value_object(root, "latency_us", tmp);
cc372b17
SL
1782 for (i = 0; i < FIO_IO_U_LAT_U_NR; i++) {
1783 const char *ranges[] = { "2", "4", "10", "20", "50", "100",
1784 "250", "500", "750", "1000", };
1785 json_object_add_value_float(tmp, ranges[i], io_u_lat_u[i]);
1786 }
1787 /* Millisecond latency */
1788 tmp = json_create_object();
1789 json_object_add_value_object(root, "latency_ms", tmp);
1790 for (i = 0; i < FIO_IO_U_LAT_M_NR; i++) {
1791 const char *ranges[] = { "2", "4", "10", "20", "50", "100",
1792 "250", "500", "750", "1000", "2000",
1793 ">=2000", };
1794 json_object_add_value_float(tmp, ranges[i], io_u_lat_m[i]);
1795 }
1796
1797 /* Additional output if continue_on_error set - default off*/
1798 if (ts->continue_on_error) {
1799 json_object_add_value_int(root, "total_err", ts->total_err_count);
952b05e0 1800 json_object_add_value_int(root, "first_error", ts->first_error);
cc372b17
SL
1801 }
1802
3e260a46
JA
1803 if (ts->latency_depth) {
1804 json_object_add_value_int(root, "latency_depth", ts->latency_depth);
1805 json_object_add_value_int(root, "latency_target", ts->latency_target);
1806 json_object_add_value_float(root, "latency_percentile", ts->latency_percentile.u.f);
1807 json_object_add_value_int(root, "latency_window", ts->latency_window);
1808 }
1809
cc372b17
SL
1810 /* Additional output if description is set */
1811 if (strlen(ts->description))
1812 json_object_add_value_string(root, "desc", ts->description);
1813
66347cfa
DE
1814 if (ts->nr_block_infos) {
1815 /* Block error histogram and types */
1816 int len;
1817 unsigned int *percentiles = NULL;
1818 unsigned int block_state_counts[BLOCK_STATE_COUNT];
1819
1820 len = calc_block_percentiles(ts->nr_block_infos, ts->block_infos,
1821 ts->percentile_list,
1822 &percentiles, block_state_counts);
1823
1824 if (len) {
1825 struct json_object *block, *percentile_object, *states;
8a68c41c 1826 int state;
66347cfa
DE
1827 block = json_create_object();
1828 json_object_add_value_object(root, "block", block);
1829
1830 percentile_object = json_create_object();
1831 json_object_add_value_object(block, "percentiles",
1832 percentile_object);
1833 for (i = 0; i < len; i++) {
1834 char buf[20];
1835 snprintf(buf, sizeof(buf), "%f",
1836 ts->percentile_list[i].u.f);
1837 json_object_add_value_int(percentile_object,
c0a6be6c 1838 buf,
66347cfa
DE
1839 percentiles[i]);
1840 }
1841
1842 states = json_create_object();
1843 json_object_add_value_object(block, "states", states);
1844 for (state = 0; state < BLOCK_STATE_COUNT; state++) {
1845 json_object_add_value_int(states,
1846 block_state_names[state],
1847 block_state_counts[state]);
1848 }
1849 free(percentiles);
1850 }
1851 }
1852
bb49c8bd 1853 if (ts->ss_dur) {
ba8fb6f6
VF
1854 struct json_object *data;
1855 struct json_array *iops, *bw;
a43f4461 1856 int j, k, l;
6da94b07 1857 char ss_buf[64];
16e56d25 1858
6da94b07 1859 snprintf(ss_buf, sizeof(ss_buf), "%s%s:%f%s",
c8caba48
JA
1860 ts->ss_state & FIO_SS_IOPS ? "iops" : "bw",
1861 ts->ss_state & FIO_SS_SLOPE ? "_slope" : "",
bb49c8bd 1862 (float) ts->ss_limit.u.f,
c8caba48 1863 ts->ss_state & FIO_SS_PCT ? "%" : "");
16e56d25
VF
1864
1865 tmp = json_create_object();
1866 json_object_add_value_object(root, "steadystate", tmp);
6da94b07 1867 json_object_add_value_string(tmp, "ss", ss_buf);
bb49c8bd 1868 json_object_add_value_int(tmp, "duration", (int)ts->ss_dur);
c8caba48 1869 json_object_add_value_int(tmp, "attained", (ts->ss_state & FIO_SS_ATTAINED) > 0);
6da94b07 1870
bb49c8bd 1871 snprintf(ss_buf, sizeof(ss_buf), "%f%s", (float) ts->ss_criterion.u.f,
c8caba48 1872 ts->ss_state & FIO_SS_PCT ? "%" : "");
6da94b07 1873 json_object_add_value_string(tmp, "criterion", ss_buf);
bb49c8bd
VF
1874 json_object_add_value_float(tmp, "max_deviation", ts->ss_deviation.u.f);
1875 json_object_add_value_float(tmp, "slope", ts->ss_slope.u.f);
ba8fb6f6
VF
1876
1877 data = json_create_object();
1878 json_object_add_value_object(tmp, "data", data);
1879 bw = json_create_array();
1880 iops = json_create_array();
412c7d91
VF
1881
1882 /*
1883 ** if ss was attained or the buffer is not full,
1884 ** ss->head points to the first element in the list.
1885 ** otherwise it actually points to the second element
1886 ** in the list
1887 */
c8caba48 1888 if ((ts->ss_state & FIO_SS_ATTAINED) || !(ts->ss_state & FIO_SS_BUFFER_FULL))
bb49c8bd 1889 j = ts->ss_head;
412c7d91 1890 else
bb49c8bd 1891 j = ts->ss_head == 0 ? ts->ss_dur - 1 : ts->ss_head - 1;
a43f4461
JA
1892 for (l = 0; l < ts->ss_dur; l++) {
1893 k = (j + l) % ts->ss_dur;
bb49c8bd
VF
1894 json_array_add_value_int(bw, ts->ss_bw_data[k]);
1895 json_array_add_value_int(iops, ts->ss_iops_data[k]);
16e56d25 1896 }
bb49c8bd
VF
1897 json_object_add_value_int(data, "bw_mean", steadystate_bw_mean(ts));
1898 json_object_add_value_int(data, "iops_mean", steadystate_iops_mean(ts));
6da94b07
VF
1899 json_object_add_value_array(data, "iops", iops);
1900 json_object_add_value_array(data, "bw", bw);
16e56d25
VF
1901 }
1902
cc372b17
SL
1903 return root;
1904}
1905
4d658652 1906static void show_thread_status_terse(struct thread_stat *ts,
a666cab8
JA
1907 struct group_run_stats *rs,
1908 struct buf_output *out)
4d658652 1909{
a2c95580 1910 if (terse_version >= 2 && terse_version <= 5)
bef2112b 1911 show_thread_status_terse_all(ts, rs, terse_version, out);
4d658652
JA
1912 else
1913 log_err("fio: bad terse version!? %d\n", terse_version);
1914}
1915
952b05e0 1916struct json_object *show_thread_status(struct thread_stat *ts,
a666cab8 1917 struct group_run_stats *rs,
0279b880 1918 struct flist_head *opt_list,
a666cab8 1919 struct buf_output *out)
952b05e0 1920{
129fb2d4
JA
1921 struct json_object *ret = NULL;
1922
1923 if (output_format & FIO_OUTPUT_TERSE)
a666cab8 1924 show_thread_status_terse(ts, rs, out);
129fb2d4 1925 if (output_format & FIO_OUTPUT_JSON)
0279b880 1926 ret = show_thread_status_json(ts, rs, opt_list);
129fb2d4 1927 if (output_format & FIO_OUTPUT_NORMAL)
a666cab8 1928 show_thread_status_normal(ts, rs, out);
129fb2d4
JA
1929
1930 return ret;
952b05e0
CF
1931}
1932
70750d6a 1933static void __sum_stat(struct io_stat *dst, struct io_stat *src, bool first)
756867bd
JA
1934{
1935 double mean, S;
1936
1937 dst->min_val = min(dst->min_val, src->min_val);
1938 dst->max_val = max(dst->max_val, src->max_val);
756867bd
JA
1939
1940 /*
cdcac5cf
YH
1941 * Compute new mean and S after the merge
1942 * <http://en.wikipedia.org/wiki/Algorithms_for_calculating_variance
1943 * #Parallel_algorithm>
756867bd 1944 */
fd595830 1945 if (first) {
802ad4a8
JA
1946 mean = src->mean.u.f;
1947 S = src->S.u.f;
756867bd 1948 } else {
802ad4a8 1949 double delta = src->mean.u.f - dst->mean.u.f;
cdcac5cf 1950
802ad4a8
JA
1951 mean = ((src->mean.u.f * src->samples) +
1952 (dst->mean.u.f * dst->samples)) /
cdcac5cf
YH
1953 (dst->samples + src->samples);
1954
802ad4a8 1955 S = src->S.u.f + dst->S.u.f + pow(delta, 2.0) *
cdcac5cf
YH
1956 (dst->samples * src->samples) /
1957 (dst->samples + src->samples);
756867bd
JA
1958 }
1959
cdcac5cf 1960 dst->samples += src->samples;
802ad4a8
JA
1961 dst->mean.u.f = mean;
1962 dst->S.u.f = S;
70750d6a
JA
1963
1964}
1965
1966/*
1967 * We sum two kinds of stats - one that is time based, in which case we
1968 * apply the proper summing technique, and then one that is iops/bw
1969 * numbers. For group_reporting, we should just add those up, not make
1970 * them the mean of everything.
1971 */
016869be 1972static void sum_stat(struct io_stat *dst, struct io_stat *src, bool pure_sum)
70750d6a 1973{
016869be
NC
1974 bool first = dst->samples == 0;
1975
70750d6a
JA
1976 if (src->samples == 0)
1977 return;
1978
1979 if (!pure_sum) {
1980 __sum_stat(dst, src, first);
1981 return;
1982 }
1983
68afa5b5
JA
1984 if (first) {
1985 dst->min_val = src->min_val;
1986 dst->max_val = src->max_val;
1987 dst->samples = src->samples;
1988 dst->mean.u.f = src->mean.u.f;
1989 dst->S.u.f = src->S.u.f;
1990 } else {
1991 dst->min_val += src->min_val;
1992 dst->max_val += src->max_val;
1993 dst->samples += src->samples;
1994 dst->mean.u.f += src->mean.u.f;
1995 dst->S.u.f += src->S.u.f;
1996 }
756867bd
JA
1997}
1998
37f0c1ae
JA
1999void sum_group_stats(struct group_run_stats *dst, struct group_run_stats *src)
2000{
2001 int i;
2002
6eaf09d6 2003 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
37f0c1ae
JA
2004 if (dst->max_run[i] < src->max_run[i])
2005 dst->max_run[i] = src->max_run[i];
2006 if (dst->min_run[i] && dst->min_run[i] > src->min_run[i])
2007 dst->min_run[i] = src->min_run[i];
2008 if (dst->max_bw[i] < src->max_bw[i])
2009 dst->max_bw[i] = src->max_bw[i];
2010 if (dst->min_bw[i] && dst->min_bw[i] > src->min_bw[i])
2011 dst->min_bw[i] = src->min_bw[i];
2012
af7f87cb 2013 dst->iobytes[i] += src->iobytes[i];
37f0c1ae
JA
2014 dst->agg[i] += src->agg[i];
2015 }
2016
dbae1bd6
JA
2017 if (!dst->kb_base)
2018 dst->kb_base = src->kb_base;
2019 if (!dst->unit_base)
2020 dst->unit_base = src->unit_base;
18aa1998
JF
2021 if (!dst->sig_figs)
2022 dst->sig_figs = src->sig_figs;
37f0c1ae
JA
2023}
2024
4ad85649
NC
2025/*
2026 * Free the clat_prio_stat arrays allocated by alloc_clat_prio_stat_ddir().
2027 */
2028void free_clat_prio_stats(struct thread_stat *ts)
2029{
2030 enum fio_ddir ddir;
2031
2032 for (ddir = 0; ddir < DDIR_RWDIR_CNT; ddir++) {
2033 sfree(ts->clat_prio[ddir]);
2034 ts->clat_prio[ddir] = NULL;
2035 ts->nr_clat_prio[ddir] = 0;
2036 }
2037}
2038
2039/*
2040 * Allocate a clat_prio_stat array. The array has to be allocated/freed using
2041 * smalloc/sfree, so that it is accessible by the process/thread summing the
2042 * thread_stats.
2043 */
2044int alloc_clat_prio_stat_ddir(struct thread_stat *ts, enum fio_ddir ddir,
2045 int nr_prios)
2046{
2047 struct clat_prio_stat *clat_prio;
2048 int i;
2049
2050 clat_prio = scalloc(nr_prios, sizeof(*ts->clat_prio[ddir]));
2051 if (!clat_prio) {
2052 log_err("fio: failed to allocate ts clat data\n");
2053 return 1;
2054 }
2055
2056 for (i = 0; i < nr_prios; i++)
2057 clat_prio[i].clat_stat.min_val = ULONG_MAX;
2058
2059 ts->clat_prio[ddir] = clat_prio;
2060 ts->nr_clat_prio[ddir] = nr_prios;
2061
2062 return 0;
2063}
2064
692dec0c
NC
2065static int grow_clat_prio_stat(struct thread_stat *dst, enum fio_ddir ddir)
2066{
2067 int curr_len = dst->nr_clat_prio[ddir];
2068 void *new_arr;
2069
2070 new_arr = scalloc(curr_len + 1, sizeof(*dst->clat_prio[ddir]));
2071 if (!new_arr) {
2072 log_err("fio: failed to grow clat prio array\n");
2073 return 1;
2074 }
2075
2076 memcpy(new_arr, dst->clat_prio[ddir],
2077 curr_len * sizeof(*dst->clat_prio[ddir]));
2078 sfree(dst->clat_prio[ddir]);
2079
2080 dst->clat_prio[ddir] = new_arr;
2081 dst->clat_prio[ddir][curr_len].clat_stat.min_val = ULONG_MAX;
2082 dst->nr_clat_prio[ddir]++;
2083
2084 return 0;
2085}
2086
2087static int find_clat_prio_index(struct thread_stat *dst, enum fio_ddir ddir,
2088 uint32_t ioprio)
2089{
2090 int i, nr_prios = dst->nr_clat_prio[ddir];
2091
2092 for (i = 0; i < nr_prios; i++) {
2093 if (dst->clat_prio[ddir][i].ioprio == ioprio)
2094 return i;
2095 }
2096
2097 return -1;
2098}
2099
2100static int alloc_or_get_clat_prio_index(struct thread_stat *dst,
2101 enum fio_ddir ddir, uint32_t ioprio,
2102 int *idx)
2103{
2104 int index = find_clat_prio_index(dst, ddir, ioprio);
2105
2106 if (index == -1) {
2107 index = dst->nr_clat_prio[ddir];
2108
2109 if (grow_clat_prio_stat(dst, ddir))
2110 return 1;
2111
2112 dst->clat_prio[ddir][index].ioprio = ioprio;
2113 }
2114
2115 *idx = index;
2116
2117 return 0;
2118}
2119
2120static int clat_prio_stats_copy(struct thread_stat *dst, struct thread_stat *src,
2121 enum fio_ddir dst_ddir, enum fio_ddir src_ddir)
2122{
2123 size_t sz = sizeof(*src->clat_prio[src_ddir]) *
2124 src->nr_clat_prio[src_ddir];
2125
2126 dst->clat_prio[dst_ddir] = smalloc(sz);
2127 if (!dst->clat_prio[dst_ddir]) {
2128 log_err("fio: failed to alloc clat prio array\n");
2129 return 1;
2130 }
2131
2132 memcpy(dst->clat_prio[dst_ddir], src->clat_prio[src_ddir], sz);
2133 dst->nr_clat_prio[dst_ddir] = src->nr_clat_prio[src_ddir];
2134
2135 return 0;
2136}
2137
2138static int clat_prio_stat_add_samples(struct thread_stat *dst,
2139 enum fio_ddir dst_ddir, uint32_t ioprio,
2140 struct io_stat *io_stat,
2141 uint64_t *io_u_plat)
2142{
2143 int i, dst_index;
2144
2145 if (!io_stat->samples)
2146 return 0;
2147
2148 if (alloc_or_get_clat_prio_index(dst, dst_ddir, ioprio, &dst_index))
2149 return 1;
2150
2151 sum_stat(&dst->clat_prio[dst_ddir][dst_index].clat_stat, io_stat,
2152 false);
2153
2154 for (i = 0; i < FIO_IO_U_PLAT_NR; i++)
2155 dst->clat_prio[dst_ddir][dst_index].io_u_plat[i] += io_u_plat[i];
2156
2157 return 0;
2158}
2159
2160static int sum_clat_prio_stats_src_single_prio(struct thread_stat *dst,
2161 struct thread_stat *src,
2162 enum fio_ddir dst_ddir,
2163 enum fio_ddir src_ddir)
2164{
2165 struct io_stat *io_stat;
2166 uint64_t *io_u_plat;
2167
2168 /*
2169 * If src ts has no clat_prio_stat array, then all I/Os were submitted
2170 * using src->ioprio. Thus, the global samples in src->clat_stat (or
2171 * src->lat_stat) can be used as the 'per prio' samples for src->ioprio.
2172 */
2173 assert(!src->clat_prio[src_ddir]);
2174 assert(src->nr_clat_prio[src_ddir] == 0);
2175
2176 if (src->lat_percentiles) {
2177 io_u_plat = src->io_u_plat[FIO_LAT][src_ddir];
2178 io_stat = &src->lat_stat[src_ddir];
2179 } else {
2180 io_u_plat = src->io_u_plat[FIO_CLAT][src_ddir];
2181 io_stat = &src->clat_stat[src_ddir];
2182 }
2183
2184 return clat_prio_stat_add_samples(dst, dst_ddir, src->ioprio, io_stat,
2185 io_u_plat);
2186}
2187
2188static int sum_clat_prio_stats_src_multi_prio(struct thread_stat *dst,
2189 struct thread_stat *src,
2190 enum fio_ddir dst_ddir,
2191 enum fio_ddir src_ddir)
2192{
2193 int i;
2194
2195 /*
2196 * If src ts has a clat_prio_stat array, then there are multiple prios
2197 * in use (i.e. src ts had cmdprio_percentage or cmdprio_bssplit set).
2198 * The samples for the default prio will exist in the src->clat_prio
2199 * array, just like the samples for any other prio.
2200 */
2201 assert(src->clat_prio[src_ddir]);
2202 assert(src->nr_clat_prio[src_ddir]);
2203
2204 /* If the dst ts doesn't yet have a clat_prio array, simply memcpy. */
2205 if (!dst->clat_prio[dst_ddir])
2206 return clat_prio_stats_copy(dst, src, dst_ddir, src_ddir);
2207
2208 /* The dst ts already has a clat_prio_array, add src stats into it. */
2209 for (i = 0; i < src->nr_clat_prio[src_ddir]; i++) {
2210 struct io_stat *io_stat = &src->clat_prio[src_ddir][i].clat_stat;
2211 uint64_t *io_u_plat = src->clat_prio[src_ddir][i].io_u_plat;
2212 uint32_t ioprio = src->clat_prio[src_ddir][i].ioprio;
2213
2214 if (clat_prio_stat_add_samples(dst, dst_ddir, ioprio, io_stat, io_u_plat))
2215 return 1;
2216 }
2217
2218 return 0;
2219}
2220
2221static int sum_clat_prio_stats(struct thread_stat *dst, struct thread_stat *src,
2222 enum fio_ddir dst_ddir, enum fio_ddir src_ddir)
2223{
2224 if (dst->disable_prio_stat)
2225 return 0;
2226
2227 if (!src->clat_prio[src_ddir])
2228 return sum_clat_prio_stats_src_single_prio(dst, src, dst_ddir,
2229 src_ddir);
2230
2231 return sum_clat_prio_stats_src_multi_prio(dst, src, dst_ddir, src_ddir);
2232}
2233
016869be 2234void sum_thread_stats(struct thread_stat *dst, struct thread_stat *src)
5b9babb7 2235{
df8781b6 2236 int k, l, m;
5b9babb7 2237
6eaf09d6 2238 for (l = 0; l < DDIR_RWDIR_CNT; l++) {
6dc75062 2239 if (dst->unified_rw_rep != UNIFIED_MIXED) {
016869be 2240 sum_stat(&dst->clat_stat[l], &src->clat_stat[l], false);
016869be
NC
2241 sum_stat(&dst->slat_stat[l], &src->slat_stat[l], false);
2242 sum_stat(&dst->lat_stat[l], &src->lat_stat[l], false);
2243 sum_stat(&dst->bw_stat[l], &src->bw_stat[l], true);
2244 sum_stat(&dst->iops_stat[l], &src->iops_stat[l], true);
692dec0c 2245 sum_clat_prio_stats(dst, src, l, l);
771e58be
JA
2246
2247 dst->io_bytes[l] += src->io_bytes[l];
2248
2249 if (dst->runtime[l] < src->runtime[l])
2250 dst->runtime[l] = src->runtime[l];
2251 } else {
016869be 2252 sum_stat(&dst->clat_stat[0], &src->clat_stat[l], false);
016869be
NC
2253 sum_stat(&dst->slat_stat[0], &src->slat_stat[l], false);
2254 sum_stat(&dst->lat_stat[0], &src->lat_stat[l], false);
2255 sum_stat(&dst->bw_stat[0], &src->bw_stat[l], true);
2256 sum_stat(&dst->iops_stat[0], &src->iops_stat[l], true);
692dec0c 2257 sum_clat_prio_stats(dst, src, 0, l);
771e58be
JA
2258
2259 dst->io_bytes[0] += src->io_bytes[l];
2260
2261 if (dst->runtime[0] < src->runtime[l])
2262 dst->runtime[0] = src->runtime[l];
2263 }
5b9babb7
JA
2264 }
2265
016869be 2266 sum_stat(&dst->sync_stat, &src->sync_stat, false);
5b9babb7
JA
2267 dst->usr_time += src->usr_time;
2268 dst->sys_time += src->sys_time;
2269 dst->ctx += src->ctx;
2270 dst->majf += src->majf;
2271 dst->minf += src->minf;
2272
b2b3eefe 2273 for (k = 0; k < FIO_IO_U_MAP_NR; k++) {
5b9babb7 2274 dst->io_u_map[k] += src->io_u_map[k];
5b9babb7 2275 dst->io_u_submit[k] += src->io_u_submit[k];
5b9babb7 2276 dst->io_u_complete[k] += src->io_u_complete[k];
b2b3eefe 2277 }
bf14b39e
VF
2278
2279 for (k = 0; k < FIO_IO_U_LAT_N_NR; k++)
d6bb626e 2280 dst->io_u_lat_n[k] += src->io_u_lat_n[k];
bf14b39e 2281 for (k = 0; k < FIO_IO_U_LAT_U_NR; k++)
5b9babb7 2282 dst->io_u_lat_u[k] += src->io_u_lat_u[k];
bf14b39e 2283 for (k = 0; k < FIO_IO_U_LAT_M_NR; k++)
5b9babb7 2284 dst->io_u_lat_m[k] += src->io_u_lat_m[k];
bf14b39e 2285
6eaf09d6 2286 for (k = 0; k < DDIR_RWDIR_CNT; k++) {
6dc75062 2287 if (dst->unified_rw_rep != UNIFIED_MIXED) {
771e58be
JA
2288 dst->total_io_u[k] += src->total_io_u[k];
2289 dst->short_io_u[k] += src->short_io_u[k];
3bcb9d94 2290 dst->drop_io_u[k] += src->drop_io_u[k];
771e58be
JA
2291 } else {
2292 dst->total_io_u[0] += src->total_io_u[k];
2293 dst->short_io_u[0] += src->short_io_u[k];
3bcb9d94 2294 dst->drop_io_u[0] += src->drop_io_u[k];
771e58be 2295 }
5b9babb7
JA
2296 }
2297
7f3ecee2
JA
2298 dst->total_io_u[DDIR_SYNC] += src->total_io_u[DDIR_SYNC];
2299
df8781b6
VF
2300 for (k = 0; k < FIO_LAT_CNT; k++)
2301 for (l = 0; l < DDIR_RWDIR_CNT; l++)
2302 for (m = 0; m < FIO_IO_U_PLAT_NR; m++)
6dc75062 2303 if (dst->unified_rw_rep != UNIFIED_MIXED)
df8781b6
VF
2304 dst->io_u_plat[k][l][m] += src->io_u_plat[k][l][m];
2305 else
2306 dst->io_u_plat[k][0][m] += src->io_u_plat[k][l][m];
771e58be 2307
df8781b6
VF
2308 for (k = 0; k < FIO_IO_U_PLAT_NR; k++)
2309 dst->io_u_sync_plat[k] += src->io_u_sync_plat[k];
2310
5b9babb7
JA
2311 dst->total_run_time += src->total_run_time;
2312 dst->total_submit += src->total_submit;
2313 dst->total_complete += src->total_complete;
fd5d733f 2314 dst->nr_zone_resets += src->nr_zone_resets;
96563db9
JA
2315 dst->cachehit += src->cachehit;
2316 dst->cachemiss += src->cachemiss;
5b9babb7
JA
2317}
2318
37f0c1ae
JA
2319void init_group_run_stat(struct group_run_stats *gs)
2320{
6eaf09d6 2321 int i;
37f0c1ae 2322 memset(gs, 0, sizeof(*gs));
6eaf09d6
SL
2323
2324 for (i = 0; i < DDIR_RWDIR_CNT; i++)
2325 gs->min_bw[i] = gs->min_run[i] = ~0UL;
37f0c1ae
JA
2326}
2327
6619fc32 2328void init_thread_stat_min_vals(struct thread_stat *ts)
37f0c1ae 2329{
6619fc32
NC
2330 int i;
2331
2332 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
2333 ts->clat_stat[i].min_val = ULONG_MAX;
2334 ts->slat_stat[i].min_val = ULONG_MAX;
2335 ts->lat_stat[i].min_val = ULONG_MAX;
2336 ts->bw_stat[i].min_val = ULONG_MAX;
2337 ts->iops_stat[i].min_val = ULONG_MAX;
6619fc32
NC
2338 }
2339 ts->sync_stat.min_val = ULONG_MAX;
2340}
37f0c1ae 2341
6619fc32
NC
2342void init_thread_stat(struct thread_stat *ts)
2343{
37f0c1ae
JA
2344 memset(ts, 0, sizeof(*ts));
2345
6619fc32 2346 init_thread_stat_min_vals(ts);
37f0c1ae
JA
2347 ts->groupid = -1;
2348}
2349
691310e2
NC
2350static void init_per_prio_stats(struct thread_stat *threadstats, int nr_ts)
2351{
2352 struct thread_data *td;
2353 struct thread_stat *ts;
2354 int i, j, last_ts, idx;
2355 enum fio_ddir ddir;
2356
2357 j = 0;
2358 last_ts = -1;
2359 idx = 0;
2360
2361 /*
2362 * Loop through all tds, if a td requires per prio stats, temporarily
2363 * store a 1 in ts->disable_prio_stat, and then do an additional
2364 * loop at the end where we invert the ts->disable_prio_stat values.
2365 */
2366 for_each_td(td, i) {
2367 if (!td->o.stats)
2368 continue;
2369 if (idx &&
2370 (!td->o.group_reporting ||
2371 (td->o.group_reporting && last_ts != td->groupid))) {
2372 idx = 0;
2373 j++;
2374 }
2375
2376 last_ts = td->groupid;
2377 ts = &threadstats[j];
2378
2379 /* idx == 0 means first td in group, or td is not in a group. */
2380 if (idx == 0)
2381 ts->ioprio = td->ioprio;
2382 else if (td->ioprio != ts->ioprio)
2383 ts->disable_prio_stat = 1;
2384
2385 for (ddir = 0; ddir < DDIR_RWDIR_CNT; ddir++) {
2386 if (td->ts.clat_prio[ddir]) {
2387 ts->disable_prio_stat = 1;
2388 break;
2389 }
2390 }
2391
2392 idx++;
2393 }
2394
2395 /* Loop through all dst threadstats and fixup the values. */
2396 for (i = 0; i < nr_ts; i++) {
2397 ts = &threadstats[i];
2398 ts->disable_prio_stat = !ts->disable_prio_stat;
2399 }
2400}
2401
83f7b64e 2402void __show_run_stats(void)
3c39a379
JA
2403{
2404 struct group_run_stats *runstats, *rs;
2405 struct thread_data *td;
756867bd 2406 struct thread_stat *threadstats, *ts;
4da24b69 2407 int i, j, k, nr_ts, last_ts, idx;
8985b491
JA
2408 bool kb_base_warned = false;
2409 bool unit_base_warned = false;
cc372b17
SL
2410 struct json_object *root = NULL;
2411 struct json_array *array = NULL;
a666cab8 2412 struct buf_output output[FIO_OUTPUT_NR];
66e19a38 2413 struct flist_head **opt_lists;
a666cab8 2414
3c39a379
JA
2415 runstats = malloc(sizeof(struct group_run_stats) * (groupid + 1));
2416
37f0c1ae
JA
2417 for (i = 0; i < groupid + 1; i++)
2418 init_group_run_stat(&runstats[i]);
3c39a379 2419
756867bd
JA
2420 /*
2421 * find out how many threads stats we need. if group reporting isn't
2422 * enabled, it's one-per-td.
2423 */
2424 nr_ts = 0;
2425 last_ts = -1;
2426 for_each_td(td, i) {
2dc1bbeb 2427 if (!td->o.group_reporting) {
756867bd
JA
2428 nr_ts++;
2429 continue;
2430 }
2431 if (last_ts == td->groupid)
2432 continue;
8243be59
JA
2433 if (!td->o.stats)
2434 continue;
756867bd
JA
2435
2436 last_ts = td->groupid;
2437 nr_ts++;
2438 }
2439
2440 threadstats = malloc(nr_ts * sizeof(struct thread_stat));
66e19a38 2441 opt_lists = malloc(nr_ts * sizeof(struct flist_head *));
756867bd 2442
66e19a38 2443 for (i = 0; i < nr_ts; i++) {
37f0c1ae 2444 init_thread_stat(&threadstats[i]);
66e19a38
JA
2445 opt_lists[i] = NULL;
2446 }
756867bd 2447
691310e2
NC
2448 init_per_prio_stats(threadstats, nr_ts);
2449
756867bd
JA
2450 j = 0;
2451 last_ts = -1;
197574e4 2452 idx = 0;
34572e28 2453 for_each_td(td, i) {
8243be59
JA
2454 if (!td->o.stats)
2455 continue;
2dc1bbeb
JA
2456 if (idx && (!td->o.group_reporting ||
2457 (td->o.group_reporting && last_ts != td->groupid))) {
7abd0e3a
JA
2458 idx = 0;
2459 j++;
2460 }
2461
2462 last_ts = td->groupid;
2463
756867bd
JA
2464 ts = &threadstats[j];
2465
83349190 2466 ts->clat_percentiles = td->o.clat_percentiles;
b599759b 2467 ts->lat_percentiles = td->o.lat_percentiles;
56440e63 2468 ts->slat_percentiles = td->o.slat_percentiles;
435d195a 2469 ts->percentile_precision = td->o.percentile_precision;
fd112d34 2470 memcpy(ts->percentile_list, td->o.percentile_list, sizeof(td->o.percentile_list));
66e19a38 2471 opt_lists[j] = &td->opt_list;
83349190 2472
197574e4 2473 idx++;
756867bd 2474
7abd0e3a 2475 if (ts->groupid == -1) {
2dc84ba7
JA
2476 /*
2477 * These are per-group shared already
2478 */
36833fb0 2479 snprintf(ts->name, sizeof(ts->name), "%s", td->o.name);
a64e88da 2480 if (td->o.description)
36833fb0
BVA
2481 snprintf(ts->description,
2482 sizeof(ts->description), "%s",
2483 td->o.description);
a64e88da 2484 else
4e59d0f3 2485 memset(ts->description, 0, FIO_JOBDESC_SIZE);
a64e88da 2486
2f122b13
JA
2487 /*
2488 * If multiple entries in this group, this is
2489 * the first member.
2490 */
2491 ts->thread_number = td->thread_number;
756867bd 2492 ts->groupid = td->groupid;
2dc84ba7
JA
2493
2494 /*
2495 * first pid in group, not very useful...
2496 */
756867bd 2497 ts->pid = td->pid;
90fef2d1
JA
2498
2499 ts->kb_base = td->o.kb_base;
ad705bcb 2500 ts->unit_base = td->o.unit_base;
e883cb35 2501 ts->sig_figs = td->o.sig_figs;
771e58be 2502 ts->unified_rw_rep = td->o.unified_rw_rep;
90fef2d1
JA
2503 } else if (ts->kb_base != td->o.kb_base && !kb_base_warned) {
2504 log_info("fio: kb_base differs for jobs in group, using"
2505 " %u as the base\n", ts->kb_base);
8985b491 2506 kb_base_warned = true;
ad705bcb
SN
2507 } else if (ts->unit_base != td->o.unit_base && !unit_base_warned) {
2508 log_info("fio: unit_base differs for jobs in group, using"
2509 " %u as the base\n", ts->unit_base);
8985b491 2510 unit_base_warned = true;
2dc84ba7
JA
2511 }
2512
f2bba182
RR
2513 ts->continue_on_error = td->o.continue_on_error;
2514 ts->total_err_count += td->total_err_count;
2515 ts->first_error = td->first_error;
2516 if (!ts->error) {
2517 if (!td->error && td->o.continue_on_error &&
2518 td->first_error) {
2519 ts->error = td->first_error;
36833fb0
BVA
2520 snprintf(ts->verror, sizeof(ts->verror), "%s",
2521 td->verror);
f2bba182
RR
2522 } else if (td->error) {
2523 ts->error = td->error;
36833fb0
BVA
2524 snprintf(ts->verror, sizeof(ts->verror), "%s",
2525 td->verror);
f2bba182 2526 }
756867bd
JA
2527 }
2528
3e260a46
JA
2529 ts->latency_depth = td->latency_qd;
2530 ts->latency_target = td->o.latency_target;
2531 ts->latency_percentile = td->o.latency_percentile;
2532 ts->latency_window = td->o.latency_window;
2533
0e4dd95c 2534 ts->nr_block_infos = td->ts.nr_block_infos;
4da24b69
JA
2535 for (k = 0; k < ts->nr_block_infos; k++)
2536 ts->block_infos[k] = td->ts.block_infos[k];
0e4dd95c 2537
016869be 2538 sum_thread_stats(ts, &td->ts);
16e56d25 2539
a381c018
NC
2540 ts->members++;
2541
bb49c8bd
VF
2542 if (td->o.ss_dur) {
2543 ts->ss_state = td->ss.state;
2544 ts->ss_dur = td->ss.dur;
2545 ts->ss_head = td->ss.head;
bb49c8bd
VF
2546 ts->ss_bw_data = td->ss.bw_data;
2547 ts->ss_iops_data = td->ss.iops_data;
2548 ts->ss_limit.u.f = td->ss.limit;
2549 ts->ss_slope.u.f = td->ss.slope;
2550 ts->ss_deviation.u.f = td->ss.deviation;
2551 ts->ss_criterion.u.f = td->ss.criterion;
2552 }
16e56d25 2553 else
bb49c8bd 2554 ts->ss_dur = ts->ss_state = 0;
756867bd
JA
2555 }
2556
2557 for (i = 0; i < nr_ts; i++) {
94370ac4 2558 unsigned long long bw;
3c39a379 2559
756867bd 2560 ts = &threadstats[i];
dedb88eb
JA
2561 if (ts->groupid == -1)
2562 continue;
756867bd 2563 rs = &runstats[ts->groupid];
90fef2d1 2564 rs->kb_base = ts->kb_base;
ad705bcb 2565 rs->unit_base = ts->unit_base;
e883cb35 2566 rs->sig_figs = ts->sig_figs;
5cb8a8cd 2567 rs->unified_rw_rep |= ts->unified_rw_rep;
3c39a379 2568
6eaf09d6 2569 for (j = 0; j < DDIR_RWDIR_CNT; j++) {
94370ac4
JA
2570 if (!ts->runtime[j])
2571 continue;
2572 if (ts->runtime[j] < rs->min_run[j] || !rs->min_run[j])
2573 rs->min_run[j] = ts->runtime[j];
2574 if (ts->runtime[j] > rs->max_run[j])
2575 rs->max_run[j] = ts->runtime[j];
2576
2577 bw = 0;
af7f87cb
RE
2578 if (ts->runtime[j])
2579 bw = ts->io_bytes[j] * 1000 / ts->runtime[j];
94370ac4
JA
2580 if (bw < rs->min_bw[j])
2581 rs->min_bw[j] = bw;
2582 if (bw > rs->max_bw[j])
2583 rs->max_bw[j] = bw;
2584
af7f87cb 2585 rs->iobytes[j] += ts->io_bytes[j];
94370ac4 2586 }
3c39a379
JA
2587 }
2588
2589 for (i = 0; i < groupid + 1; i++) {
2e545522 2590 enum fio_ddir ddir;
6eaf09d6 2591
3c39a379
JA
2592 rs = &runstats[i];
2593
6eaf09d6
SL
2594 for (ddir = 0; ddir < DDIR_RWDIR_CNT; ddir++) {
2595 if (rs->max_run[ddir])
af7f87cb 2596 rs->agg[ddir] = (rs->iobytes[ddir] * 1000) /
6eaf09d6
SL
2597 rs->max_run[ddir];
2598 }
3c39a379
JA
2599 }
2600
a666cab8 2601 for (i = 0; i < FIO_OUTPUT_NR; i++)
e250c0a9 2602 buf_output_init(&output[i]);
a666cab8 2603
3c39a379
JA
2604 /*
2605 * don't overwrite last signal output
2606 */
129fb2d4 2607 if (output_format & FIO_OUTPUT_NORMAL)
a666cab8 2608 log_buf(&output[__FIO_OUTPUT_NORMAL], "\n");
129fb2d4 2609 if (output_format & FIO_OUTPUT_JSON) {
66e19a38 2610 struct thread_data *global;
35326842 2611 char time_buf[32];
aa7d2ef0
RH
2612 struct timeval now;
2613 unsigned long long ms_since_epoch;
afd74e2a 2614 time_t tv_sec;
91e53870 2615
aa7d2ef0
RH
2616 gettimeofday(&now, NULL);
2617 ms_since_epoch = (unsigned long long)(now.tv_sec) * 1000 +
2618 (unsigned long long)(now.tv_usec) / 1000;
2619
afd74e2a
BVA
2620 tv_sec = now.tv_sec;
2621 os_ctime_r(&tv_sec, time_buf, sizeof(time_buf));
1d272416
JA
2622 if (time_buf[strlen(time_buf) - 1] == '\n')
2623 time_buf[strlen(time_buf) - 1] = '\0';
91e53870 2624
cc372b17
SL
2625 root = json_create_object();
2626 json_object_add_value_string(root, "fio version", fio_version_string);
aa7d2ef0
RH
2627 json_object_add_value_int(root, "timestamp", now.tv_sec);
2628 json_object_add_value_int(root, "timestamp_ms", ms_since_epoch);
91e53870 2629 json_object_add_value_string(root, "time", time_buf);
66e19a38 2630 global = get_global_options();
0cf542af 2631 json_add_job_opts(root, "global options", &global->opt_list);
cc372b17
SL
2632 array = json_create_array();
2633 json_object_add_value_array(root, "jobs", array);
2634 }
3c39a379 2635
0279b880
JA
2636 if (is_backend)
2637 fio_server_send_job_options(&get_global_options()->opt_list, -1U);
2638
756867bd
JA
2639 for (i = 0; i < nr_ts; i++) {
2640 ts = &threadstats[i];
2641 rs = &runstats[ts->groupid];
3c39a379 2642
0279b880
JA
2643 if (is_backend) {
2644 fio_server_send_job_options(opt_lists[i], i);
a64e88da 2645 fio_server_send_ts(ts, rs);
0279b880 2646 } else {
129fb2d4 2647 if (output_format & FIO_OUTPUT_TERSE)
a666cab8 2648 show_thread_status_terse(ts, rs, &output[__FIO_OUTPUT_TERSE]);
129fb2d4 2649 if (output_format & FIO_OUTPUT_JSON) {
66e19a38 2650 struct json_object *tmp = show_thread_status_json(ts, rs, opt_lists[i]);
129fb2d4
JA
2651 json_array_add_value_object(array, tmp);
2652 }
2653 if (output_format & FIO_OUTPUT_NORMAL)
a666cab8 2654 show_thread_status_normal(ts, rs, &output[__FIO_OUTPUT_NORMAL]);
129fb2d4 2655 }
3c39a379 2656 }
ab34ddd1 2657 if (!is_backend && (output_format & FIO_OUTPUT_JSON)) {
cc372b17 2658 /* disk util stats, if any */
a666cab8 2659 show_disk_util(1, root, &output[__FIO_OUTPUT_JSON]);
cc372b17 2660
a666cab8 2661 show_idle_prof_stats(FIO_OUTPUT_JSON, root, &output[__FIO_OUTPUT_JSON]);
f2a2ce0e 2662
a666cab8
JA
2663 json_print_object(root, &output[__FIO_OUTPUT_JSON]);
2664 log_buf(&output[__FIO_OUTPUT_JSON], "\n");
cc372b17
SL
2665 json_free_object(root);
2666 }
3c39a379 2667
72c27ff8
JA
2668 for (i = 0; i < groupid + 1; i++) {
2669 rs = &runstats[i];
3c39a379 2670
72c27ff8 2671 rs->groupid = i;
d09a64a0 2672 if (is_backend)
72c27ff8 2673 fio_server_send_gs(rs);
129fb2d4 2674 else if (output_format & FIO_OUTPUT_NORMAL)
a666cab8 2675 show_group_stats(rs, &output[__FIO_OUTPUT_NORMAL]);
c6ae0a5b 2676 }
eecf272f 2677
72c27ff8
JA
2678 if (is_backend)
2679 fio_server_send_du();
129fb2d4 2680 else if (output_format & FIO_OUTPUT_NORMAL) {
a666cab8
JA
2681 show_disk_util(0, NULL, &output[__FIO_OUTPUT_NORMAL]);
2682 show_idle_prof_stats(FIO_OUTPUT_NORMAL, NULL, &output[__FIO_OUTPUT_NORMAL]);
60904003 2683 }
f2a2ce0e 2684
3d57c0e9 2685 for (i = 0; i < FIO_OUTPUT_NR; i++) {
3dc3aa41 2686 struct buf_output *out = &output[i];
8dd0eca3 2687
3dc3aa41 2688 log_info_buf(out->buf, out->buflen);
3dc3aa41 2689 buf_output_free(out);
3d57c0e9 2690 }
2b8c71b0 2691
7ceefc09
FS
2692 fio_idle_prof_cleanup();
2693
fdd5f15f 2694 log_info_flush();
eecf272f 2695 free(runstats);
692dec0c
NC
2696
2697 /* free arrays allocated by sum_thread_stats(), if any */
2698 for (i = 0; i < nr_ts; i++) {
2699 ts = &threadstats[i];
2700 free_clat_prio_stats(ts);
2701 }
756867bd 2702 free(threadstats);
66e19a38 2703 free(opt_lists);
3c39a379
JA
2704}
2705
966f8ef9 2706int __show_running_run_stats(void)
b852e7cf
JA
2707{
2708 struct thread_data *td;
2709 unsigned long long *rt;
8b6a404c 2710 struct timespec ts;
b852e7cf
JA
2711 int i;
2712
971caeb1 2713 fio_sem_down(stat_sem);
90811930 2714
b852e7cf 2715 rt = malloc(thread_number * sizeof(unsigned long long));
8b6a404c 2716 fio_gettime(&ts, NULL);
b852e7cf
JA
2717
2718 for_each_td(td, i) {
c97f1ad6 2719 td->update_rusage = 1;
dde39f8c
AD
2720 for_each_rw_ddir(ddir) {
2721 td->ts.io_bytes[ddir] = td->io_bytes[ddir];
2722 }
8b6a404c 2723 td->ts.total_run_time = mtime_since(&td->epoch, &ts);
6c041a88 2724
8b6a404c 2725 rt[i] = mtime_since(&td->start, &ts);
6c041a88 2726 if (td_read(td) && td->ts.io_bytes[DDIR_READ])
2727 td->ts.runtime[DDIR_READ] += rt[i];
2728 if (td_write(td) && td->ts.io_bytes[DDIR_WRITE])
2729 td->ts.runtime[DDIR_WRITE] += rt[i];
2730 if (td_trim(td) && td->ts.io_bytes[DDIR_TRIM])
2731 td->ts.runtime[DDIR_TRIM] += rt[i];
b852e7cf
JA
2732 }
2733
c97f1ad6 2734 for_each_td(td, i) {
fda2cfac
JA
2735 if (td->runstate >= TD_EXITED)
2736 continue;
c97f1ad6
JA
2737 if (td->rusage_sem) {
2738 td->update_rusage = 1;
971caeb1 2739 fio_sem_down(td->rusage_sem);
c97f1ad6
JA
2740 }
2741 td->update_rusage = 0;
2742 }
2743
cef9175e 2744 __show_run_stats();
b852e7cf
JA
2745
2746 for_each_td(td, i) {
6c041a88 2747 if (td_read(td) && td->ts.io_bytes[DDIR_READ])
b852e7cf 2748 td->ts.runtime[DDIR_READ] -= rt[i];
6c041a88 2749 if (td_write(td) && td->ts.io_bytes[DDIR_WRITE])
b852e7cf 2750 td->ts.runtime[DDIR_WRITE] -= rt[i];
6c041a88 2751 if (td_trim(td) && td->ts.io_bytes[DDIR_TRIM])
6eaf09d6 2752 td->ts.runtime[DDIR_TRIM] -= rt[i];
b852e7cf
JA
2753 }
2754
2755 free(rt);
971caeb1 2756 fio_sem_up(stat_sem);
966f8ef9
BVA
2757
2758 return 0;
b852e7cf
JA
2759}
2760
8985b491 2761static bool status_file_disabled;
06464907 2762
dac45f23 2763#define FIO_STATUS_FILE "fio-dump-status"
06464907
JA
2764
2765static int check_status_file(void)
2766{
2767 struct stat sb;
a0bafb7d
BC
2768 const char *temp_dir;
2769 char fio_status_file_path[PATH_MAX];
06464907 2770
77d99675
JA
2771 if (status_file_disabled)
2772 return 0;
2773
a0bafb7d 2774 temp_dir = getenv("TMPDIR");
48b16e27 2775 if (temp_dir == NULL) {
a0bafb7d 2776 temp_dir = getenv("TEMP");
48b16e27
JA
2777 if (temp_dir && strlen(temp_dir) >= PATH_MAX)
2778 temp_dir = NULL;
2779 }
a0bafb7d
BC
2780 if (temp_dir == NULL)
2781 temp_dir = "/tmp";
3544731c
BVA
2782#ifdef __COVERITY__
2783 __coverity_tainted_data_sanitize__(temp_dir);
2784#endif
a0bafb7d
BC
2785
2786 snprintf(fio_status_file_path, sizeof(fio_status_file_path), "%s/%s", temp_dir, FIO_STATUS_FILE);
2787
2788 if (stat(fio_status_file_path, &sb))
06464907
JA
2789 return 0;
2790
77d99675
JA
2791 if (unlink(fio_status_file_path) < 0) {
2792 log_err("fio: failed to unlink %s: %s\n", fio_status_file_path,
2793 strerror(errno));
2794 log_err("fio: disabling status file updates\n");
8985b491 2795 status_file_disabled = true;
77d99675
JA
2796 }
2797
06464907
JA
2798 return 1;
2799}
2800
2801void check_for_running_stats(void)
2802{
06464907
JA
2803 if (check_status_file()) {
2804 show_running_run_stats();
2805 return;
2806 }
2807}
2808
d6bb626e 2809static inline void add_stat_sample(struct io_stat *is, unsigned long long data)
3c39a379 2810{
68704084 2811 double val = data;
6660cc67 2812 double delta;
68704084
JA
2813
2814 if (data > is->max_val)
2815 is->max_val = data;
2816 if (data < is->min_val)
2817 is->min_val = data;
2818
802ad4a8 2819 delta = val - is->mean.u.f;
ef11d737 2820 if (delta) {
802ad4a8
JA
2821 is->mean.u.f += delta / (is->samples + 1.0);
2822 is->S.u.f += delta * (val - is->mean.u.f);
ef11d737 2823 }
3c39a379 2824
3c39a379
JA
2825 is->samples++;
2826}
2827
692dec0c
NC
2828static inline void add_stat_prio_sample(struct clat_prio_stat *clat_prio,
2829 unsigned short clat_prio_index,
2830 unsigned long long nsec)
2831{
2832 if (clat_prio)
2833 add_stat_sample(&clat_prio[clat_prio_index].clat_stat, nsec);
2834}
2835
7e419452
JA
2836/*
2837 * Return a struct io_logs, which is added to the tail of the log
2838 * list for 'iolog'.
2839 */
2840static struct io_logs *get_new_log(struct io_log *iolog)
2841{
0a852a50 2842 size_t new_samples;
7e419452
JA
2843 struct io_logs *cur_log;
2844
2845 /*
2846 * Cap the size at MAX_LOG_ENTRIES, so we don't keep doubling
2847 * forever
2848 */
0a852a50
DLM
2849 if (!iolog->cur_log_max) {
2850 new_samples = iolog->td->o.log_entries;
2851 } else {
7e419452
JA
2852 new_samples = iolog->cur_log_max * 2;
2853 if (new_samples > MAX_LOG_ENTRIES)
2854 new_samples = MAX_LOG_ENTRIES;
2855 }
2856
90d2d53f 2857 cur_log = smalloc(sizeof(*cur_log));
7e419452
JA
2858 if (cur_log) {
2859 INIT_FLIST_HEAD(&cur_log->list);
0a852a50 2860 cur_log->log = calloc(new_samples, log_entry_sz(iolog));
7e419452
JA
2861 if (cur_log->log) {
2862 cur_log->nr_samples = 0;
2863 cur_log->max_samples = new_samples;
2864 flist_add_tail(&cur_log->list, &iolog->io_logs);
2865 iolog->cur_log_max = new_samples;
2866 return cur_log;
2867 }
90d2d53f 2868 sfree(cur_log);
7e419452
JA
2869 }
2870
2871 return NULL;
2872}
2873
1fed2080
JA
2874/*
2875 * Add and return a new log chunk, or return current log if big enough
2876 */
2877static struct io_logs *regrow_log(struct io_log *iolog)
7e419452
JA
2878{
2879 struct io_logs *cur_log;
1fed2080 2880 int i;
7e419452 2881
1fed2080 2882 if (!iolog || iolog->disabled)
2ab71dc4 2883 goto disable;
7e419452 2884
1fed2080 2885 cur_log = iolog_cur_log(iolog);
a9afa45a
JA
2886 if (!cur_log) {
2887 cur_log = get_new_log(iolog);
2888 if (!cur_log)
2889 return NULL;
2890 }
2891
7e419452
JA
2892 if (cur_log->nr_samples < cur_log->max_samples)
2893 return cur_log;
2894
2895 /*
2896 * No room for a new sample. If we're compressing on the fly, flush
2897 * out the current chunk
2898 */
2899 if (iolog->log_gz) {
2900 if (iolog_cur_flush(iolog, cur_log)) {
2901 log_err("fio: failed flushing iolog! Will stop logging.\n");
2902 return NULL;
2903 }
2904 }
2905
2906 /*
2907 * Get a new log array, and add to our list
2908 */
2909 cur_log = get_new_log(iolog);
1fed2080
JA
2910 if (!cur_log) {
2911 log_err("fio: failed extending iolog! Will stop logging.\n");
2912 return NULL;
2913 }
2914
2915 if (!iolog->pending || !iolog->pending->nr_samples)
80a24ba9 2916 return cur_log;
7e419452 2917
1fed2080
JA
2918 /*
2919 * Flush pending items to new log
2920 */
2921 for (i = 0; i < iolog->pending->nr_samples; i++) {
2922 struct io_sample *src, *dst;
2923
2924 src = get_sample(iolog, iolog->pending, i);
2925 dst = get_sample(iolog, cur_log, i);
2926 memcpy(dst, src, log_entry_sz(iolog));
2927 }
0b2eef49 2928 cur_log->nr_samples = iolog->pending->nr_samples;
1fed2080
JA
2929
2930 iolog->pending->nr_samples = 0;
2931 return cur_log;
2ab71dc4
JA
2932disable:
2933 if (iolog)
2934 iolog->disabled = true;
2935 return NULL;
1fed2080
JA
2936}
2937
2938void regrow_logs(struct thread_data *td)
2939{
2ab71dc4
JA
2940 regrow_log(td->slat_log);
2941 regrow_log(td->clat_log);
1e613c9c 2942 regrow_log(td->clat_hist_log);
2ab71dc4
JA
2943 regrow_log(td->lat_log);
2944 regrow_log(td->bw_log);
2945 regrow_log(td->iops_log);
1fed2080
JA
2946 td->flags &= ~TD_F_REGROW_LOGS;
2947}
2948
76204de3
PM
2949void regrow_agg_logs(void)
2950{
2951 enum fio_ddir ddir;
2952
2953 for (ddir = 0; ddir < DDIR_RWDIR_CNT; ddir++)
2954 regrow_log(agg_io_log[ddir]);
2955}
2956
1fed2080
JA
2957static struct io_logs *get_cur_log(struct io_log *iolog)
2958{
2959 struct io_logs *cur_log;
2960
2961 cur_log = iolog_cur_log(iolog);
2962 if (!cur_log) {
2963 cur_log = get_new_log(iolog);
2964 if (!cur_log)
2965 return NULL;
2966 }
2967
2968 if (cur_log->nr_samples < cur_log->max_samples)
2969 return cur_log;
2970
2971 /*
1eb467fb
JA
2972 * Out of space. If we're in IO offload mode, or we're not doing
2973 * per unit logging (hence logging happens outside of the IO thread
2974 * as well), add a new log chunk inline. If we're doing inline
2975 * submissions, flag 'td' as needing a log regrow and we'll take
2976 * care of it on the submission side.
1fed2080 2977 */
ab5643cb 2978 if ((iolog->td && iolog->td->o.io_submit_mode == IO_MODE_OFFLOAD) ||
1eb467fb 2979 !per_unit_log(iolog))
1fed2080
JA
2980 return regrow_log(iolog);
2981
ab5643cb
IK
2982 if (iolog->td)
2983 iolog->td->flags |= TD_F_REGROW_LOGS;
2984 if (iolog->pending)
2985 assert(iolog->pending->nr_samples < iolog->pending->max_samples);
1fed2080 2986 return iolog->pending;
7e419452
JA
2987}
2988
af2fde19 2989static void __add_log_sample(struct io_log *iolog, union io_sample_data data,
5fff9543 2990 enum fio_ddir ddir, unsigned long long bs,
03ec570f
DLM
2991 unsigned long t, uint64_t offset,
2992 unsigned int priority)
3c39a379 2993{
7e419452 2994 struct io_logs *cur_log;
306ddc97 2995
3c568239
JA
2996 if (iolog->disabled)
2997 return;
7e419452 2998 if (flist_empty(&iolog->io_logs))
8355119b 2999 iolog->avg_last[ddir] = t;
b8bc8cba 3000
7e419452
JA
3001 cur_log = get_cur_log(iolog);
3002 if (cur_log) {
3003 struct io_sample *s;
3c39a379 3004
7e419452 3005 s = get_sample(iolog, cur_log, cur_log->nr_samples);
3c39a379 3006
af2fde19 3007 s->data = data;
a9179eeb 3008 s->time = t + (iolog->td ? iolog->td->unix_epoch : 0);
7e419452
JA
3009 io_sample_set_ddir(iolog, s, ddir);
3010 s->bs = bs;
03ec570f 3011 s->priority = priority;
ae588852 3012
7e419452
JA
3013 if (iolog->log_offset) {
3014 struct io_sample_offset *so = (void *) s;
ae588852 3015
7e419452
JA
3016 so->offset = offset;
3017 }
ae588852 3018
7e419452
JA
3019 cur_log->nr_samples++;
3020 return;
ae588852
JA
3021 }
3022
7e419452 3023 iolog->disabled = true;
3c39a379
JA
3024}
3025
7fb28d36
JA
3026static inline void reset_io_stat(struct io_stat *ios)
3027{
5b8f19b7
JA
3028 ios->min_val = -1ULL;
3029 ios->max_val = ios->samples = 0;
7fb28d36
JA
3030 ios->mean.u.f = ios->S.u.f = 0;
3031}
3032
2f045d2e
NC
3033static inline void reset_io_u_plat(uint64_t *io_u_plat)
3034{
3035 int i;
3036
3037 for (i = 0; i < FIO_IO_U_PLAT_NR; i++)
3038 io_u_plat[i] = 0;
3039}
3040
692dec0c
NC
3041static inline void reset_clat_prio_stats(struct thread_stat *ts)
3042{
3043 enum fio_ddir ddir;
3044 int i;
3045
3046 for (ddir = 0; ddir < DDIR_RWDIR_CNT; ddir++) {
3047 if (!ts->clat_prio[ddir])
3048 continue;
3049
3050 for (i = 0; i < ts->nr_clat_prio[ddir]; i++) {
3051 reset_io_stat(&ts->clat_prio[ddir][i].clat_stat);
3052 reset_io_u_plat(ts->clat_prio[ddir][i].io_u_plat);
3053 }
3054 }
3055}
3056
6bb58215
JA
3057void reset_io_stats(struct thread_data *td)
3058{
3059 struct thread_stat *ts = &td->ts;
2f045d2e 3060 int i, j;
6bb58215
JA
3061
3062 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
3063 reset_io_stat(&ts->clat_stat[i]);
3064 reset_io_stat(&ts->slat_stat[i]);
3065 reset_io_stat(&ts->lat_stat[i]);
3066 reset_io_stat(&ts->bw_stat[i]);
3067 reset_io_stat(&ts->iops_stat[i]);
3068
3069 ts->io_bytes[i] = 0;
3070 ts->runtime[i] = 0;
71cb78c1
VF
3071 ts->total_io_u[i] = 0;
3072 ts->short_io_u[i] = 0;
3073 ts->drop_io_u[i] = 0;
6bb58215
JA
3074 }
3075
df8781b6
VF
3076 for (i = 0; i < FIO_LAT_CNT; i++)
3077 for (j = 0; j < DDIR_RWDIR_CNT; j++)
2f045d2e 3078 reset_io_u_plat(ts->io_u_plat[i][j]);
df8781b6 3079
692dec0c
NC
3080 reset_clat_prio_stats(ts);
3081
7f3ecee2 3082 ts->total_io_u[DDIR_SYNC] = 0;
2f045d2e 3083 reset_io_u_plat(ts->io_u_sync_plat);
7f3ecee2 3084
6bb58215
JA
3085 for (i = 0; i < FIO_IO_U_MAP_NR; i++) {
3086 ts->io_u_map[i] = 0;
3087 ts->io_u_submit[i] = 0;
3088 ts->io_u_complete[i] = 0;
71cb78c1
VF
3089 }
3090
d6bb626e
VF
3091 for (i = 0; i < FIO_IO_U_LAT_N_NR; i++)
3092 ts->io_u_lat_n[i] = 0;
71cb78c1 3093 for (i = 0; i < FIO_IO_U_LAT_U_NR; i++)
6bb58215 3094 ts->io_u_lat_u[i] = 0;
71cb78c1 3095 for (i = 0; i < FIO_IO_U_LAT_M_NR; i++)
6bb58215 3096 ts->io_u_lat_m[i] = 0;
6bb58215 3097
71cb78c1
VF
3098 ts->total_submit = 0;
3099 ts->total_complete = 0;
fd5d733f 3100 ts->nr_zone_resets = 0;
96563db9 3101 ts->cachehit = ts->cachemiss = 0;
6bb58215
JA
3102}
3103
d96d3bb3 3104static void __add_stat_to_log(struct io_log *iolog, enum fio_ddir ddir,
03ec570f 3105 unsigned long elapsed, bool log_max)
99007068
PO
3106{
3107 /*
3108 * Note an entry in the log. Use the mean from the logged samples,
3109 * making sure to properly round up. Only write a log entry if we
3110 * had actual samples done.
3111 */
d96d3bb3 3112 if (iolog->avg_window[ddir].samples) {
af2fde19 3113 union io_sample_data data;
99007068 3114
e6989e10 3115 if (log_max)
af2fde19 3116 data.val = iolog->avg_window[ddir].max_val;
e6989e10 3117 else
af2fde19 3118 data.val = iolog->avg_window[ddir].mean.u.f + 0.50;
e6989e10 3119
03ec570f 3120 __add_log_sample(iolog, data, ddir, 0, elapsed, 0, 0);
99007068 3121 }
99007068 3122
d96d3bb3
JA
3123 reset_io_stat(&iolog->avg_window[ddir]);
3124}
99007068 3125
e6989e10 3126static void _add_stat_to_log(struct io_log *iolog, unsigned long elapsed,
03ec570f 3127 bool log_max)
d96d3bb3 3128{
2e545522 3129 enum fio_ddir ddir;
99007068 3130
d96d3bb3 3131 for (ddir = 0; ddir < DDIR_RWDIR_CNT; ddir++)
03ec570f 3132 __add_stat_to_log(iolog, ddir, elapsed, log_max);
99007068
PO
3133}
3134
674456bf
JF
3135static unsigned long add_log_sample(struct thread_data *td,
3136 struct io_log *iolog,
3137 union io_sample_data data,
5fff9543 3138 enum fio_ddir ddir, unsigned long long bs,
03ec570f 3139 uint64_t offset, unsigned int ioprio)
bb3884d8 3140{
7fb28d36 3141 unsigned long elapsed, this_window;
b8bc8cba 3142
ff58fced 3143 if (!ddir_rw(ddir))
d454a205 3144 return 0;
ff58fced 3145
b8bc8cba
JA
3146 elapsed = mtime_since_now(&td->epoch);
3147
3148 /*
3149 * If no time averaging, just add the log sample.
3150 */
3151 if (!iolog->avg_msec) {
03ec570f
DLM
3152 __add_log_sample(iolog, data, ddir, bs, elapsed, offset,
3153 ioprio);
d454a205 3154 return 0;
b8bc8cba
JA
3155 }
3156
3157 /*
3158 * Add the sample. If the time period has passed, then
3159 * add that entry to the log and clear.
3160 */
af2fde19 3161 add_stat_sample(&iolog->avg_window[ddir], data.val);
b8bc8cba 3162
7fb28d36
JA
3163 /*
3164 * If period hasn't passed, adding the above sample is all we
3165 * need to do.
3166 */
8355119b
SW
3167 this_window = elapsed - iolog->avg_last[ddir];
3168 if (elapsed < iolog->avg_last[ddir])
3169 return iolog->avg_last[ddir] - elapsed;
f5a568cf 3170 else if (this_window < iolog->avg_msec) {
674456bf 3171 unsigned long diff = iolog->avg_msec - this_window;
d454a205 3172
b392f36d 3173 if (inline_log(iolog) || diff > LOG_MSEC_SLACK)
d454a205
JA
3174 return diff;
3175 }
b8bc8cba 3176
03ec570f 3177 __add_stat_to_log(iolog, ddir, elapsed, td->o.log_max != 0);
b8bc8cba 3178
58a2d29d
JL
3179 iolog->avg_last[ddir] = elapsed - (elapsed % iolog->avg_msec);
3180
d454a205 3181 return iolog->avg_msec;
99007068 3182}
6eaf09d6 3183
a47591e4 3184void finalize_logs(struct thread_data *td, bool unit_logs)
99007068
PO
3185{
3186 unsigned long elapsed;
6eaf09d6 3187
99007068 3188 elapsed = mtime_since_now(&td->epoch);
b8bc8cba 3189
a47591e4 3190 if (td->clat_log && unit_logs)
03ec570f 3191 _add_stat_to_log(td->clat_log, elapsed, td->o.log_max != 0);
a47591e4 3192 if (td->slat_log && unit_logs)
03ec570f 3193 _add_stat_to_log(td->slat_log, elapsed, td->o.log_max != 0);
a47591e4 3194 if (td->lat_log && unit_logs)
03ec570f 3195 _add_stat_to_log(td->lat_log, elapsed, td->o.log_max != 0);
a47591e4 3196 if (td->bw_log && (unit_logs == per_unit_log(td->bw_log)))
03ec570f 3197 _add_stat_to_log(td->bw_log, elapsed, td->o.log_max != 0);
a47591e4 3198 if (td->iops_log && (unit_logs == per_unit_log(td->iops_log)))
03ec570f 3199 _add_stat_to_log(td->iops_log, elapsed, td->o.log_max != 0);
bb3884d8
JA
3200}
3201
03ec570f
DLM
3202void add_agg_sample(union io_sample_data data, enum fio_ddir ddir,
3203 unsigned long long bs)
bb3884d8 3204{
ff58fced 3205 struct io_log *iolog;
bb3884d8 3206
ff58fced
JA
3207 if (!ddir_rw(ddir))
3208 return;
3209
3210 iolog = agg_io_log[ddir];
03ec570f 3211 __add_log_sample(iolog, data, ddir, bs, mtime_since_genesis(), 0, 0);
bb3884d8
JA
3212}
3213
b2b3eefe
JA
3214void add_sync_clat_sample(struct thread_stat *ts, unsigned long long nsec)
3215{
3216 unsigned int idx = plat_val_to_idx(nsec);
3217 assert(idx < FIO_IO_U_PLAT_NR);
3218
3219 ts->io_u_sync_plat[idx]++;
3220 add_stat_sample(&ts->sync_stat, nsec);
3221}
3222
ed7f3a07
NC
3223static inline void add_lat_percentile_sample(struct thread_stat *ts,
3224 unsigned long long nsec,
3225 enum fio_ddir ddir,
3226 enum fio_lat lat)
83349190 3227{
d6bb626e 3228 unsigned int idx = plat_val_to_idx(nsec);
83349190
YH
3229 assert(idx < FIO_IO_U_PLAT_NR);
3230
56440e63
VF
3231 ts->io_u_plat[lat][ddir][idx]++;
3232}
b2a432bf 3233
692dec0c
NC
3234static inline void
3235add_lat_percentile_prio_sample(struct thread_stat *ts, unsigned long long nsec,
3236 enum fio_ddir ddir,
3237 unsigned short clat_prio_index)
56440e63
VF
3238{
3239 unsigned int idx = plat_val_to_idx(nsec);
3240
692dec0c
NC
3241 if (ts->clat_prio[ddir])
3242 ts->clat_prio[ddir][clat_prio_index].io_u_plat[idx]++;
83349190
YH
3243}
3244
1e97cce9 3245void add_clat_sample(struct thread_data *td, enum fio_ddir ddir,
5fff9543 3246 unsigned long long nsec, unsigned long long bs,
692dec0c
NC
3247 uint64_t offset, unsigned int ioprio,
3248 unsigned short clat_prio_index)
3c39a379 3249{
26b3a188 3250 const bool needs_lock = td_async_processing(td);
1e613c9c 3251 unsigned long elapsed, this_window;
756867bd 3252 struct thread_stat *ts = &td->ts;
1e613c9c 3253 struct io_log *iolog = td->clat_hist_log;
079ad09b 3254
26b3a188
JA
3255 if (needs_lock)
3256 __td_io_u_lock(td);
75dc383e 3257
d6bb626e 3258 add_stat_sample(&ts->clat_stat[ddir], nsec);
3c39a379 3259
717569b1 3260 /*
692dec0c 3261 * When lat_percentiles=1 (default 0), the reported per priority
717569b1
NC
3262 * percentiles and stats are used for describing total latency values,
3263 * even though the variable names themselves start with clat_.
3264 *
3265 * Because of the above definition, add a prio stat sample only when
3266 * lat_percentiles=0. add_lat_sample() will add the prio stat sample
3267 * when lat_percentiles=1.
3268 */
692dec0c
NC
3269 if (!ts->lat_percentiles)
3270 add_stat_prio_sample(ts->clat_prio[ddir], clat_prio_index,
3271 nsec);
b2a432bf 3272
7b9f733a 3273 if (td->clat_log)
d6bb626e 3274 add_log_sample(td, td->clat_log, sample_val(nsec), ddir, bs,
03ec570f 3275 offset, ioprio);
83349190 3276
b2a432bf 3277 if (ts->clat_percentiles) {
717569b1
NC
3278 /*
3279 * Because of the above definition, add a prio lat percentile
3280 * sample only when lat_percentiles=0. add_lat_sample() will add
3281 * the prio lat percentile sample when lat_percentiles=1.
3282 */
bf65f7c8
NC
3283 add_lat_percentile_sample(ts, nsec, ddir, FIO_CLAT);
3284 if (!ts->lat_percentiles)
3285 add_lat_percentile_prio_sample(ts, nsec, ddir,
692dec0c 3286 clat_prio_index);
b2a432bf 3287 }
75dc383e 3288
1e613c9c 3289 if (iolog && iolog->hist_msec) {
93168285
JA
3290 struct io_hist *hw = &iolog->hist_window[ddir];
3291
3292 hw->samples++;
1e613c9c 3293 elapsed = mtime_since_now(&td->epoch);
93168285 3294 if (!hw->hist_last)
1e613c9c
KC
3295 hw->hist_last = elapsed;
3296 this_window = elapsed - hw->hist_last;
7a4e480d 3297
1e613c9c 3298 if (this_window >= iolog->hist_msec) {
6cc0e5aa 3299 uint64_t *io_u_plat;
65a4d15c 3300 struct io_u_plat_entry *dst;
93168285 3301
1e613c9c 3302 /*
93168285
JA
3303 * Make a byte-for-byte copy of the latency histogram
3304 * stored in td->ts.io_u_plat[ddir], recording it in a
3305 * log sample. Note that the matching call to free() is
3306 * located in iolog.c after printing this sample to the
3307 * log file.
1e613c9c 3308 */
df8781b6 3309 io_u_plat = (uint64_t *) td->ts.io_u_plat[FIO_CLAT][ddir];
65a4d15c
KC
3310 dst = malloc(sizeof(struct io_u_plat_entry));
3311 memcpy(&(dst->io_u_plat), io_u_plat,
1fb9250b 3312 FIO_IO_U_PLAT_NR * sizeof(uint64_t));
d730bc58 3313 flist_add(&dst->list, &hw->list);
af2fde19 3314 __add_log_sample(iolog, sample_plat(dst), ddir, bs,
03ec570f 3315 elapsed, offset, ioprio);
1e613c9c
KC
3316
3317 /*
93168285
JA
3318 * Update the last time we recorded as being now, minus
3319 * any drift in time we encountered before actually
3320 * making the record.
1e613c9c
KC
3321 */
3322 hw->hist_last = elapsed - (this_window - iolog->hist_msec);
3323 hw->samples = 0;
3324 }
3325 }
3326
26b3a188
JA
3327 if (needs_lock)
3328 __td_io_u_unlock(td);
3c39a379
JA
3329}
3330
1e97cce9 3331void add_slat_sample(struct thread_data *td, enum fio_ddir ddir,
03ec570f
DLM
3332 unsigned long long nsec, unsigned long long bs,
3333 uint64_t offset, unsigned int ioprio)
3c39a379 3334{
26b3a188 3335 const bool needs_lock = td_async_processing(td);
756867bd 3336 struct thread_stat *ts = &td->ts;
079ad09b 3337
ff58fced
JA
3338 if (!ddir_rw(ddir))
3339 return;
3340
26b3a188
JA
3341 if (needs_lock)
3342 __td_io_u_lock(td);
75dc383e 3343
56440e63 3344 add_stat_sample(&ts->slat_stat[ddir], nsec);
3c39a379 3345
7b9f733a 3346 if (td->slat_log)
03ec570f
DLM
3347 add_log_sample(td, td->slat_log, sample_val(nsec), ddir, bs,
3348 offset, ioprio);
75dc383e 3349
56440e63 3350 if (ts->slat_percentiles)
3dd92a75 3351 add_lat_percentile_sample(ts, nsec, ddir, FIO_SLAT);
56440e63 3352
26b3a188
JA
3353 if (needs_lock)
3354 __td_io_u_unlock(td);
3c39a379
JA
3355}
3356
02af0988 3357void add_lat_sample(struct thread_data *td, enum fio_ddir ddir,
5fff9543 3358 unsigned long long nsec, unsigned long long bs,
692dec0c
NC
3359 uint64_t offset, unsigned int ioprio,
3360 unsigned short clat_prio_index)
02af0988 3361{
26b3a188 3362 const bool needs_lock = td_async_processing(td);
02af0988
JA
3363 struct thread_stat *ts = &td->ts;
3364
ff58fced
JA
3365 if (!ddir_rw(ddir))
3366 return;
3367
26b3a188
JA
3368 if (needs_lock)
3369 __td_io_u_lock(td);
75dc383e 3370
d6bb626e 3371 add_stat_sample(&ts->lat_stat[ddir], nsec);
02af0988 3372
7b9f733a 3373 if (td->lat_log)
d6bb626e 3374 add_log_sample(td, td->lat_log, sample_val(nsec), ddir, bs,
03ec570f 3375 offset, ioprio);
75dc383e 3376
717569b1 3377 /*
692dec0c 3378 * When lat_percentiles=1 (default 0), the reported per priority
717569b1
NC
3379 * percentiles and stats are used for describing total latency values,
3380 * even though the variable names themselves start with clat_.
3381 *
3382 * Because of the above definition, add a prio stat and prio lat
3383 * percentile sample only when lat_percentiles=1. add_clat_sample() will
3384 * add the prio stat and prio lat percentile sample when
3385 * lat_percentiles=0.
3386 */
38ec5c51 3387 if (ts->lat_percentiles) {
bf65f7c8 3388 add_lat_percentile_sample(ts, nsec, ddir, FIO_LAT);
692dec0c
NC
3389 add_lat_percentile_prio_sample(ts, nsec, ddir, clat_prio_index);
3390 add_stat_prio_sample(ts->clat_prio[ddir], clat_prio_index,
3391 nsec);
38ec5c51 3392 }
26b3a188
JA
3393 if (needs_lock)
3394 __td_io_u_unlock(td);
02af0988
JA
3395}
3396
a47591e4 3397void add_bw_sample(struct thread_data *td, struct io_u *io_u,
d6bb626e 3398 unsigned int bytes, unsigned long long spent)
a47591e4 3399{
26b3a188 3400 const bool needs_lock = td_async_processing(td);
a47591e4
JA
3401 struct thread_stat *ts = &td->ts;
3402 unsigned long rate;
3403
3404 if (spent)
d6bb626e 3405 rate = (unsigned long) (bytes * 1000000ULL / spent);
a47591e4
JA
3406 else
3407 rate = 0;
3408
26b3a188
JA
3409 if (needs_lock)
3410 __td_io_u_lock(td);
a47591e4
JA
3411
3412 add_stat_sample(&ts->bw_stat[io_u->ddir], rate);
3413
3414 if (td->bw_log)
af2fde19 3415 add_log_sample(td, td->bw_log, sample_val(rate), io_u->ddir,
03ec570f 3416 bytes, io_u->offset, io_u->ioprio);
a47591e4
JA
3417
3418 td->stat_io_bytes[io_u->ddir] = td->this_io_bytes[io_u->ddir];
26b3a188
JA
3419
3420 if (needs_lock)
3421 __td_io_u_unlock(td);
a47591e4
JA
3422}
3423
8b6a404c
VF
3424static int __add_samples(struct thread_data *td, struct timespec *parent_tv,
3425 struct timespec *t, unsigned int avg_time,
4843277a
JA
3426 uint64_t *this_io_bytes, uint64_t *stat_io_bytes,
3427 struct io_stat *stat, struct io_log *log,
3428 bool is_kb)
3c39a379 3429{
26b3a188 3430 const bool needs_lock = td_async_processing(td);
ff58fced 3431 unsigned long spent, rate;
a47591e4 3432 enum fio_ddir ddir;
674456bf 3433 unsigned long next, next_log;
d454a205 3434
4843277a 3435 next_log = avg_time;
ff58fced 3436
4843277a 3437 spent = mtime_since(parent_tv, t);
58a2d29d 3438 if (spent < avg_time && avg_time - spent > LOG_MSEC_SLACK)
4843277a 3439 return avg_time - spent;
9602d8df 3440
26b3a188
JA
3441 if (needs_lock)
3442 __td_io_u_lock(td);
a9da8ab2 3443
9602d8df 3444 /*
5daa4ebe
JC
3445 * Compute both read and write rates for the interval.
3446 */
c1f50f76 3447 for (ddir = 0; ddir < DDIR_RWDIR_CNT; ddir++) {
5daa4ebe
JC
3448 uint64_t delta;
3449
4843277a 3450 delta = this_io_bytes[ddir] - stat_io_bytes[ddir];
5daa4ebe
JC
3451 if (!delta)
3452 continue; /* No entries for interval */
3c39a379 3453
4843277a
JA
3454 if (spent) {
3455 if (is_kb)
3456 rate = delta * 1000 / spent / 1024; /* KiB/s */
3457 else
3458 rate = (delta * 1000) / spent;
3459 } else
0956264f
JA
3460 rate = 0;
3461
4843277a 3462 add_stat_sample(&stat[ddir], rate);
3c39a379 3463
37181709 3464 if (log) {
5fff9543 3465 unsigned long long bs = 0;
66b98c9f
JA
3466
3467 if (td->o.min_bs[ddir] == td->o.max_bs[ddir])
3468 bs = td->o.min_bs[ddir];
3469
03ec570f
DLM
3470 next = add_log_sample(td, log, sample_val(rate), ddir,
3471 bs, 0, 0);
d454a205 3472 next_log = min(next_log, next);
66b98c9f 3473 }
5daa4ebe 3474
4843277a 3475 stat_io_bytes[ddir] = this_io_bytes[ddir];
5daa4ebe 3476 }
3c39a379 3477
cdb8308d 3478 *parent_tv = *t;
a47591e4 3479
26b3a188
JA
3480 if (needs_lock)
3481 __td_io_u_unlock(td);
a47591e4 3482
4843277a
JA
3483 if (spent <= avg_time)
3484 next = avg_time;
306fea38 3485 else
4843277a 3486 next = avg_time - (1 + spent - avg_time);
a47591e4 3487
d454a205 3488 return min(next, next_log);
a47591e4
JA
3489}
3490
8b6a404c 3491static int add_bw_samples(struct thread_data *td, struct timespec *t)
4843277a
JA
3492{
3493 return __add_samples(td, &td->bw_sample_time, t, td->o.bw_avg_time,
3494 td->this_io_bytes, td->stat_io_bytes,
3495 td->ts.bw_stat, td->bw_log, true);
3496}
3497
a47591e4
JA
3498void add_iops_sample(struct thread_data *td, struct io_u *io_u,
3499 unsigned int bytes)
3500{
26b3a188 3501 const bool needs_lock = td_async_processing(td);
a47591e4
JA
3502 struct thread_stat *ts = &td->ts;
3503
26b3a188
JA
3504 if (needs_lock)
3505 __td_io_u_lock(td);
a47591e4
JA
3506
3507 add_stat_sample(&ts->iops_stat[io_u->ddir], 1);
3508
3509 if (td->iops_log)
af2fde19 3510 add_log_sample(td, td->iops_log, sample_val(1), io_u->ddir,
03ec570f 3511 bytes, io_u->offset, io_u->ioprio);
a47591e4
JA
3512
3513 td->stat_io_blocks[io_u->ddir] = td->this_io_blocks[io_u->ddir];
26b3a188
JA
3514
3515 if (needs_lock)
3516 __td_io_u_unlock(td);
3c39a379 3517}
c8eeb9df 3518
8b6a404c 3519static int add_iops_samples(struct thread_data *td, struct timespec *t)
c8eeb9df 3520{
4843277a
JA
3521 return __add_samples(td, &td->iops_sample_time, t, td->o.iops_avg_time,
3522 td->this_io_blocks, td->stat_io_blocks,
3523 td->ts.iops_stat, td->iops_log, false);
a47591e4
JA
3524}
3525
3526/*
3527 * Returns msecs to next event
3528 */
3529int calc_log_samples(void)
3530{
3531 struct thread_data *td;
58a2d29d 3532 unsigned int next = ~0U, tmp = 0, next_mod = 0, log_avg_msec_min = -1U;
8b6a404c 3533 struct timespec now;
a47591e4 3534 int i;
58a2d29d 3535 long elapsed_time = 0;
a47591e4
JA
3536
3537 fio_gettime(&now, NULL);
3538
3539 for_each_td(td, i) {
58a2d29d
JL
3540 elapsed_time = mtime_since_now(&td->epoch);
3541
8243be59
JA
3542 if (!td->o.stats)
3543 continue;
6a89b401 3544 if (in_ramp_time(td) ||
a47591e4
JA
3545 !(td->runstate == TD_RUNNING || td->runstate == TD_VERIFYING)) {
3546 next = min(td->o.iops_avg_time, td->o.bw_avg_time);
3547 continue;
3548 }
37181709
AH
3549 if (!td->bw_log ||
3550 (td->bw_log && !per_unit_log(td->bw_log))) {
a47591e4 3551 tmp = add_bw_samples(td, &now);
58a2d29d
JL
3552
3553 if (td->bw_log)
3554 log_avg_msec_min = min(log_avg_msec_min, (unsigned int)td->bw_log->avg_msec);
a47591e4 3555 }
37181709
AH
3556 if (!td->iops_log ||
3557 (td->iops_log && !per_unit_log(td->iops_log))) {
a47591e4 3558 tmp = add_iops_samples(td, &now);
58a2d29d
JL
3559
3560 if (td->iops_log)
3561 log_avg_msec_min = min(log_avg_msec_min, (unsigned int)td->iops_log->avg_msec);
a47591e4 3562 }
58a2d29d
JL
3563
3564 if (tmp < next)
3565 next = tmp;
a47591e4
JA
3566 }
3567
58a2d29d
JL
3568 /* if log_avg_msec_min has not been changed, set it to 0 */
3569 if (log_avg_msec_min == -1U)
3570 log_avg_msec_min = 0;
3571
3572 if (log_avg_msec_min == 0)
3573 next_mod = elapsed_time;
3574 else
3575 next_mod = elapsed_time % log_avg_msec_min;
3576
3577 /* correction to keep the time on the log avg msec boundary */
3578 next = min(next, (log_avg_msec_min - next_mod));
3579
a47591e4 3580 return next == ~0U ? 0 : next;
c8eeb9df 3581}
cef9175e
JA
3582
3583void stat_init(void)
3584{
971caeb1 3585 stat_sem = fio_sem_init(FIO_SEM_UNLOCKED);
cef9175e
JA
3586}
3587
3588void stat_exit(void)
3589{
3590 /*
3591 * When we have the mutex, we know out-of-band access to it
3592 * have ended.
3593 */
971caeb1
BVA
3594 fio_sem_down(stat_sem);
3595 fio_sem_remove(stat_sem);
cef9175e 3596}
b2ee7647 3597
b2ee7647
JA
3598/*
3599 * Called from signal handler. Wake up status thread.
3600 */
3601void show_running_run_stats(void)
3602{
a47591e4 3603 helper_do_stat();
b2ee7647 3604}
66347cfa
DE
3605
3606uint32_t *io_u_block_info(struct thread_data *td, struct io_u *io_u)
3607{
3608 /* Ignore io_u's which span multiple blocks--they will just get
3609 * inaccurate counts. */
3610 int idx = (io_u->offset - io_u->file->file_offset)
3611 / td->o.bs[DDIR_TRIM];
3612 uint32_t *info = &td->ts.block_infos[idx];
3613 assert(idx < td->ts.nr_block_infos);
3614 return info;
3615}
5c0abd5e 3616