1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
|
// SPDX-License-Identifier: GPL-2.0-only
/*
* Copyright 2013, Michael Ellerman, IBM Corp.
*/
#define _GNU_SOURCE
#include <stdio.h>
#include <stdbool.h>
#include <string.h>
#include <sys/prctl.h>
#include "event.h"
#include "utils.h"
#include "lib.h"
extern void thirty_two_instruction_loop(u64 loops);
static void setup_event(struct event *e, u64 config, char *name)
{
event_init_opts(e, config, PERF_TYPE_HARDWARE, name);
e->attr.disabled = 1;
e->attr.exclude_kernel = 1;
e->attr.exclude_hv = 1;
e->attr.exclude_idle = 1;
}
static int do_count_loop(struct event *events, u64 instructions,
u64 overhead, bool report)
{
s64 difference, expected;
double percentage;
prctl(PR_TASK_PERF_EVENTS_ENABLE);
/* Run for 1M instructions */
thirty_two_instruction_loop(instructions >> 5);
prctl(PR_TASK_PERF_EVENTS_DISABLE);
event_read(&events[0]);
event_read(&events[1]);
expected = instructions + overhead;
difference = events[0].result.value - expected;
percentage = (double)difference / events[0].result.value * 100;
if (report) {
event_report(&events[0]);
event_report(&events[1]);
printf("Looped for %llu instructions, overhead %llu\n", instructions, overhead);
printf("Expected %llu\n", expected);
printf("Actual %llu\n", events[0].result.value);
printf("Delta %lld, %f%%\n", difference, percentage);
}
event_reset(&events[0]);
event_reset(&events[1]);
if (difference < 0)
difference = -difference;
/* Tolerate a difference below 0.0001 % */
difference *= 10000 * 100;
if (difference / events[0].result.value)
return -1;
return 0;
}
/* Count how many instructions it takes to do a null loop */
static u64 determine_overhead(struct event *events)
{
u64 current, overhead;
int i;
do_count_loop(events, 0, 0, false);
overhead = events[0].result.value;
for (i = 0; i < 100; i++) {
do_count_loop(events, 0, 0, false);
current = events[0].result.value;
if (current < overhead) {
printf("Replacing overhead %llu with %llu\n", overhead, current);
overhead = current;
}
}
return overhead;
}
static int test_body(void)
{
struct event events[2];
u64 overhead;
setup_event(&events[0], PERF_COUNT_HW_INSTRUCTIONS, "instructions");
setup_event(&events[1], PERF_COUNT_HW_CPU_CYCLES, "cycles");
if (event_open(&events[0])) {
perror("perf_event_open");
return -1;
}
if (event_open_with_group(&events[1], events[0].fd)) {
perror("perf_event_open");
return -1;
}
overhead = determine_overhead(events);
printf("Overhead of null loop: %llu instructions\n", overhead);
/* Run for 1Mi instructions */
FAIL_IF(do_count_loop(events, 1000000, overhead, true));
/* Run for 10Mi instructions */
FAIL_IF(do_count_loop(events, 10000000, overhead, true));
/* Run for 100Mi instructions */
FAIL_IF(do_count_loop(events, 100000000, overhead, true));
/* Run for 1Bi instructions */
FAIL_IF(do_count_loop(events, 1000000000, overhead, true));
/* Run for 16Bi instructions */
FAIL_IF(do_count_loop(events, 16000000000, overhead, true));
/* Run for 64Bi instructions */
FAIL_IF(do_count_loop(events, 64000000000, overhead, true));
event_close(&events[0]);
event_close(&events[1]);
return 0;
}
static int count_instructions(void)
{
return eat_cpu(test_body);
}
int main(void)
{
return test_harness(count_instructions, "count_instructions");
}
|
