os: make fio_cpu_isset() return a bool
[fio.git] / stat.h
CommitLineData
a64e88da
JA
1#ifndef FIO_STAT_H
2#define FIO_STAT_H
3
ec41265e 4#include "iolog.h"
a666cab8 5#include "lib/output_buffer.h"
ec41265e 6
a64e88da 7struct group_run_stats {
6eaf09d6
SL
8 uint64_t max_run[DDIR_RWDIR_CNT], min_run[DDIR_RWDIR_CNT];
9 uint64_t max_bw[DDIR_RWDIR_CNT], min_bw[DDIR_RWDIR_CNT];
af7f87cb 10 uint64_t iobytes[DDIR_RWDIR_CNT];
6eaf09d6 11 uint64_t agg[DDIR_RWDIR_CNT];
a64e88da 12 uint32_t kb_base;
ad705bcb 13 uint32_t unit_base;
a64e88da 14 uint32_t groupid;
771e58be 15 uint32_t unified_rw_rep;
e5021b34 16} __attribute__((packed));
a64e88da
JA
17
18/*
19 * How many depth levels to log
20 */
21#define FIO_IO_U_MAP_NR 7
d6bb626e 22#define FIO_IO_U_LAT_N_NR 10
a64e88da
JA
23#define FIO_IO_U_LAT_U_NR 10
24#define FIO_IO_U_LAT_M_NR 12
25
37b08652
VF
26/*
27 * Constants for clat percentiles
28 */
29#define FIO_IO_U_PLAT_BITS 6
30#define FIO_IO_U_PLAT_VAL (1 << FIO_IO_U_PLAT_BITS)
31#define FIO_IO_U_PLAT_GROUP_NR 29
32#define FIO_IO_U_PLAT_NR (FIO_IO_U_PLAT_GROUP_NR * FIO_IO_U_PLAT_VAL)
33#define FIO_IO_U_LIST_MAX_LEN 20 /* The size of the default and user-specified
34 list of percentiles */
35
a64e88da
JA
36/*
37 * Aggregate clat samples to report percentile(s) of them.
38 *
39 * EXECUTIVE SUMMARY
40 *
41 * FIO_IO_U_PLAT_BITS determines the maximum statistical error on the
42 * value of resulting percentiles. The error will be approximately
43 * 1/2^(FIO_IO_U_PLAT_BITS+1) of the value.
44 *
45 * FIO_IO_U_PLAT_GROUP_NR and FIO_IO_U_PLAT_BITS determine the maximum
46 * range being tracked for latency samples. The maximum value tracked
37b08652 47 * accurately will be 2^(GROUP_NR + PLAT_BITS - 1) nanoseconds.
a64e88da
JA
48 *
49 * FIO_IO_U_PLAT_GROUP_NR and FIO_IO_U_PLAT_BITS determine the memory
50 * requirement of storing those aggregate counts. The memory used will
51 * be (FIO_IO_U_PLAT_GROUP_NR * 2^FIO_IO_U_PLAT_BITS) * sizeof(int)
52 * bytes.
53 *
54 * FIO_IO_U_PLAT_NR is the total number of buckets.
55 *
56 * DETAILS
57 *
58 * Suppose the clat varies from 0 to 999 (usec), the straightforward
59 * method is to keep an array of (999 + 1) buckets, in which a counter
60 * keeps the count of samples which fall in the bucket, e.g.,
61 * {[0],[1],...,[999]}. However this consumes a huge amount of space,
62 * and can be avoided if an approximation is acceptable.
63 *
64 * One such method is to let the range of the bucket to be greater
65 * than one. This method has low accuracy when the value is small. For
66 * example, let the buckets be {[0,99],[100,199],...,[900,999]}, and
67 * the represented value of each bucket be the mean of the range. Then
68 * a value 0 has an round-off error of 49.5. To improve on this, we
69 * use buckets with non-uniform ranges, while bounding the error of
70 * each bucket within a ratio of the sample value. A simple example
71 * would be when error_bound = 0.005, buckets are {
72 * {[0],[1],...,[99]}, {[100,101],[102,103],...,[198,199]},..,
73 * {[900,909],[910,919]...} }. The total range is partitioned into
74 * groups with different ranges, then buckets with uniform ranges. An
75 * upper bound of the error is (range_of_bucket/2)/value_of_bucket
76 *
77 * For better efficiency, we implement this using base two. We group
78 * samples by their Most Significant Bit (MSB), extract the next M bit
79 * of them as an index within the group, and discard the rest of the
80 * bits.
81 *
82 * E.g., assume a sample 'x' whose MSB is bit n (starting from bit 0),
83 * and use M bit for indexing
84 *
85 * | n | M bits | bit (n-M-1) ... bit 0 |
86 *
87 * Because x is at least 2^n, and bit 0 to bit (n-M-1) is at most
88 * (2^(n-M) - 1), discarding bit 0 to (n-M-1) makes the round-off
89 * error
90 *
91 * 2^(n-M)-1 2^(n-M) 1
92 * e <= --------- <= ------- = ---
93 * 2^n 2^n 2^M
94 *
95 * Furthermore, we use "mean" of the range to represent the bucket,
96 * the error e can be lowered by half to 1 / 2^(M+1). By using M bits
97 * as the index, each group must contains 2^M buckets.
98 *
99 * E.g. Let M (FIO_IO_U_PLAT_BITS) be 6
100 * Error bound is 1/2^(6+1) = 0.0078125 (< 1%)
101 *
102 * Group MSB #discarded range of #buckets
103 * error_bits value
104 * ----------------------------------------------------------------
105 * 0* 0~5 0 [0,63] 64
106 * 1* 6 0 [64,127] 64
107 * 2 7 1 [128,255] 64
108 * 3 8 2 [256,511] 64
109 * 4 9 3 [512,1023] 64
110 * ... ... ... [...,...] ...
37b08652 111 * 28 33 27 [8589934592,+inf]** 64
a64e88da
JA
112 *
113 * * Special cases: when n < (M-1) or when n == (M-1), in both cases,
114 * the value cannot be rounded off. Use all bits of the sample as
115 * index.
116 *
37b08652 117 * ** If a sample's MSB is greater than 33, it will be counted as 33.
a64e88da
JA
118 */
119
66347cfa
DE
120/*
121 * Trim cycle count measurements
122 */
123#define MAX_NR_BLOCK_INFOS 8192
124#define BLOCK_INFO_STATE_SHIFT 29
125#define BLOCK_INFO_TRIMS(block_info) \
126 ((block_info) & ((1 << BLOCK_INFO_STATE_SHIFT) - 1))
127#define BLOCK_INFO_STATE(block_info) \
128 ((block_info) >> BLOCK_INFO_STATE_SHIFT)
129#define BLOCK_INFO(state, trim_cycles) \
fc8d6d05 130 ((trim_cycles) | ((unsigned int) (state) << BLOCK_INFO_STATE_SHIFT))
66347cfa
DE
131#define BLOCK_INFO_SET_STATE(block_info, state) \
132 BLOCK_INFO(state, BLOCK_INFO_TRIMS(block_info))
133enum block_info_state {
134 BLOCK_STATE_UNINIT,
135 BLOCK_STATE_TRIMMED,
136 BLOCK_STATE_WRITTEN,
137 BLOCK_STATE_TRIM_FAILURE,
138 BLOCK_STATE_WRITE_FAILURE,
139 BLOCK_STATE_COUNT,
1d6d3455 140};
66347cfa 141
a64e88da
JA
142#define MAX_PATTERN_SIZE 512
143#define FIO_JOBNAME_SIZE 128
4e59d0f3 144#define FIO_JOBDESC_SIZE 256
a64e88da
JA
145#define FIO_VERROR_SIZE 128
146
147struct thread_stat {
148 char name[FIO_JOBNAME_SIZE];
149 char verror[FIO_VERROR_SIZE];
ddcc0b69 150 uint32_t error;
2f122b13 151 uint32_t thread_number;
ddcc0b69 152 uint32_t groupid;
a64e88da 153 uint32_t pid;
4e59d0f3 154 char description[FIO_JOBDESC_SIZE];
a64e88da 155 uint32_t members;
771e58be 156 uint32_t unified_rw_rep;
a64e88da
JA
157
158 /*
159 * bandwidth and latency stats
160 */
6eaf09d6
SL
161 struct io_stat clat_stat[DDIR_RWDIR_CNT]; /* completion latency */
162 struct io_stat slat_stat[DDIR_RWDIR_CNT]; /* submission latency */
163 struct io_stat lat_stat[DDIR_RWDIR_CNT]; /* total latency */
164 struct io_stat bw_stat[DDIR_RWDIR_CNT]; /* bandwidth stats */
165 struct io_stat iops_stat[DDIR_RWDIR_CNT]; /* IOPS stats */
a64e88da
JA
166
167 /*
168 * fio system usage accounting
169 */
170 uint64_t usr_time;
171 uint64_t sys_time;
172 uint64_t ctx;
173 uint64_t minf, majf;
174
175 /*
176 * IO depth and latency stats
177 */
b599759b
JA
178 uint32_t clat_percentiles;
179 uint32_t lat_percentiles;
435d195a 180 uint64_t percentile_precision;
802ad4a8 181 fio_fp64_t percentile_list[FIO_IO_U_LIST_MAX_LEN];
a64e88da
JA
182
183 uint32_t io_u_map[FIO_IO_U_MAP_NR];
184 uint32_t io_u_submit[FIO_IO_U_MAP_NR];
185 uint32_t io_u_complete[FIO_IO_U_MAP_NR];
d6bb626e 186 uint32_t io_u_lat_n[FIO_IO_U_LAT_N_NR];
a64e88da
JA
187 uint32_t io_u_lat_u[FIO_IO_U_LAT_U_NR];
188 uint32_t io_u_lat_m[FIO_IO_U_LAT_M_NR];
6eaf09d6 189 uint32_t io_u_plat[DDIR_RWDIR_CNT][FIO_IO_U_PLAT_NR];
3d0ebb30
GG
190 uint32_t pad;
191
de5cdfea
JA
192 uint64_t total_io_u[DDIR_RWDIR_CNT];
193 uint64_t short_io_u[DDIR_RWDIR_CNT];
194 uint64_t drop_io_u[DDIR_RWDIR_CNT];
a64e88da
JA
195 uint64_t total_submit;
196 uint64_t total_complete;
197
6eaf09d6
SL
198 uint64_t io_bytes[DDIR_RWDIR_CNT];
199 uint64_t runtime[DDIR_RWDIR_CNT];
a64e88da
JA
200 uint64_t total_run_time;
201
202 /*
203 * IO Error related stats
204 */
3d0ebb30
GG
205 union {
206 uint16_t continue_on_error;
11e955e3 207 uint32_t pad2;
3d0ebb30 208 };
ddcc0b69 209 uint32_t first_error;
11e955e3 210 uint64_t total_err_count;
a64e88da 211
66347cfa
DE
212 uint64_t nr_block_infos;
213 uint32_t block_infos[MAX_NR_BLOCK_INFOS];
214
a64e88da 215 uint32_t kb_base;
ad705bcb 216 uint32_t unit_base;
3e260a46
JA
217
218 uint32_t latency_depth;
11e955e3 219 uint32_t pad3;
3e260a46
JA
220 uint64_t latency_target;
221 fio_fp64_t latency_percentile;
222 uint64_t latency_window;
16e56d25 223
bb49c8bd
VF
224 uint64_t ss_dur;
225 uint32_t ss_state;
226 uint32_t ss_head;
bb49c8bd 227
bb49c8bd
VF
228 fio_fp64_t ss_limit;
229 fio_fp64_t ss_slope;
230 fio_fp64_t ss_deviation;
231 fio_fp64_t ss_criterion;
cb84f1fa
VF
232
233 union {
234 uint64_t *ss_iops_data;
11e955e3 235 uint64_t pad4;
cb84f1fa
VF
236 };
237
238 union {
239 uint64_t *ss_bw_data;
11e955e3 240 uint64_t pad5;
cb84f1fa 241 };
e5021b34 242} __attribute__((packed));
a64e88da 243
b75a394f
JA
244struct jobs_eta {
245 uint32_t nr_running;
246 uint32_t nr_ramp;
819db5f5 247
b75a394f 248 uint32_t nr_pending;
714e85f3 249 uint32_t nr_setting_up;
819db5f5 250
90eff1c9 251 uint64_t m_rate[DDIR_RWDIR_CNT], t_rate[DDIR_RWDIR_CNT];
90eff1c9 252 uint64_t rate[DDIR_RWDIR_CNT];
dbf285f7 253 uint32_t m_iops[DDIR_RWDIR_CNT], t_iops[DDIR_RWDIR_CNT];
6eaf09d6 254 uint32_t iops[DDIR_RWDIR_CNT];
b75a394f
JA
255 uint64_t elapsed_sec;
256 uint64_t eta_sec;
b7f05eb0 257 uint32_t is_pow2;
ad705bcb 258 uint32_t unit_base;
1d1f45ae 259
dbf285f7
SW
260 uint32_t files_open;
261
1d1f45ae
JA
262 /*
263 * Network 'copy' of run_str[]
264 */
265 uint32_t nr_threads;
372aecb9 266 uint8_t run_str[];
e5021b34 267} __attribute__((packed));
b75a394f 268
65a4d15c
KC
269struct io_u_plat_entry {
270 struct flist_head list;
271 unsigned int io_u_plat[FIO_IO_U_PLAT_NR];
272};
273
e5437a07
VT
274extern struct fio_mutex *stat_mutex;
275
c5103619 276extern struct jobs_eta *get_jobs_eta(bool force, size_t *size);
723297c9 277
cef9175e
JA
278extern void stat_init(void);
279extern void stat_exit(void);
280
0279b880 281extern struct json_object * show_thread_status(struct thread_stat *ts, struct group_run_stats *rs, struct flist_head *, struct buf_output *);
a666cab8 282extern void show_group_stats(struct group_run_stats *rs, struct buf_output *);
8aa89d70 283extern bool calc_thread_status(struct jobs_eta *je, int force);
cf451d1e 284extern void display_thread_status(struct jobs_eta *je);
5b9babb7 285extern void show_run_stats(void);
83f7b64e 286extern void __show_run_stats(void);
5ddc6707 287extern void __show_running_run_stats(void);
b852e7cf 288extern void show_running_run_stats(void);
06464907 289extern void check_for_running_stats(void);
fd595830 290extern void sum_thread_stats(struct thread_stat *dst, struct thread_stat *src, bool first);
37f0c1ae
JA
291extern void sum_group_stats(struct group_run_stats *dst, struct group_run_stats *src);
292extern void init_thread_stat(struct thread_stat *ts);
293extern void init_group_run_stat(struct group_run_stats *gs);
3e47bd25 294extern void eta_to_str(char *str, unsigned long eta_sec);
d6bb626e 295extern bool calc_lat(struct io_stat *is, unsigned long long *min, unsigned long long *max, double *mean, double *dev);
d7bb1620 296extern unsigned int calc_clat_percentiles(unsigned int *io_u_plat, unsigned long long nr, fio_fp64_t *plist, unsigned long long **output, unsigned long long *maxv, unsigned long long *minv);
fea76a0c 297extern void stat_calc_lat_n(struct thread_stat *ts, double *io_u_lat);
e5bd1347
JA
298extern void stat_calc_lat_m(struct thread_stat *ts, double *io_u_lat);
299extern void stat_calc_lat_u(struct thread_stat *ts, double *io_u_lat);
2e33101f 300extern void stat_calc_dist(unsigned int *map, unsigned long total, double *io_u_dist);
6bb58215 301extern void reset_io_stats(struct thread_data *);
210dd0fc
JA
302extern void update_rusage_stat(struct thread_data *);
303extern void clear_rusage_stat(struct thread_data *);
2e33101f 304
d6bb626e 305extern void add_lat_sample(struct thread_data *, enum fio_ddir, unsigned long long,
cf8f852a 306 unsigned int, uint64_t);
d6bb626e 307extern void add_clat_sample(struct thread_data *, enum fio_ddir, unsigned long long,
cf8f852a
JA
308 unsigned int, uint64_t);
309extern void add_slat_sample(struct thread_data *, enum fio_ddir, unsigned long,
310 unsigned int, uint64_t);
af2fde19 311extern void add_agg_sample(union io_sample_data, enum fio_ddir, unsigned int);
a47591e4
JA
312extern void add_iops_sample(struct thread_data *, struct io_u *,
313 unsigned int);
314extern void add_bw_sample(struct thread_data *, struct io_u *,
d6bb626e 315 unsigned int, unsigned long long);
a47591e4 316extern int calc_log_samples(void);
cf8f852a
JA
317
318extern struct io_log *agg_io_log[DDIR_RWDIR_CNT];
319extern int write_bw_log;
320
74558486
JA
321static inline bool nsec_to_usec(unsigned long long *min,
322 unsigned long long *max, double *mean,
323 double *dev)
b29ad562 324{
d6bb626e 325 if (*min > 2000 && *max > 99999 && *dev > 1000.0) {
b29ad562
JA
326 *min /= 1000;
327 *max /= 1000;
328 *mean /= 1000.0;
329 *dev /= 1000.0;
8aa89d70 330 return true;
b29ad562
JA
331 }
332
8aa89d70 333 return false;
b29ad562 334}
d6bb626e 335
74558486
JA
336static inline bool nsec_to_msec(unsigned long long *min,
337 unsigned long long *max, double *mean,
338 double *dev)
d6bb626e
VF
339{
340 if (*min > 2000000 && *max > 99999999ULL && *dev > 1000000.0) {
341 *min /= 1000000;
342 *max /= 1000000;
343 *mean /= 1000000.0;
344 *dev /= 1000000.0;
345 return true;
346 }
347
348 return false;
349}
74558486 350
723297c9
JA
351/*
352 * Worst level condensing would be 1:5, so allow enough room for that
353 */
354#define __THREAD_RUNSTR_SZ(nr) ((nr) * 5)
11155857
JA
355#define THREAD_RUNSTR_SZ __THREAD_RUNSTR_SZ(thread_number)
356
66347cfa
DE
357uint32_t *io_u_block_info(struct thread_data *td, struct io_u *io_u);
358
a64e88da 359#endif