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
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
|
libperf-counting(7)
===================
NAME
----
libperf-counting - counting interface
DESCRIPTION
-----------
The counting interface provides API to measure and get count for specific perf events.
The following test tries to explain count on `counting.c` example.
It is by no means complete guide to counting, but shows libperf basic API for counting.
The `counting.c` comes with libperf package and can be compiled and run like:
[source,bash]
--
$ gcc -o counting counting.c -lperf
$ sudo ./counting
count 176792, enabled 176944, run 176944
count 176242, enabled 176242, run 176242
--
It requires root access, because of the `PERF_COUNT_SW_CPU_CLOCK` event,
which is available only for root.
The `counting.c` example monitors two events on the current process and displays
their count, in a nutshell it:
* creates events
* adds them to the event list
* opens and enables events through the event list
* does some workload
* disables events
* reads and displays event counts
* destroys the event list
The first thing you need to do before using libperf is to call init function:
[source,c]
--
8 static int libperf_print(enum libperf_print_level level,
9 const char *fmt, va_list ap)
10 {
11 return vfprintf(stderr, fmt, ap);
12 }
14 int main(int argc, char **argv)
15 {
...
35 libperf_init(libperf_print);
--
It will setup the library and sets function for debug output from library.
The `libperf_print` callback will receive any message with its debug level,
defined as:
[source,c]
--
enum libperf_print_level {
LIBPERF_ERR,
LIBPERF_WARN,
LIBPERF_INFO,
LIBPERF_DEBUG,
LIBPERF_DEBUG2,
LIBPERF_DEBUG3,
};
--
Once the setup is complete we start by defining specific events using the `struct perf_event_attr`.
We create software events for cpu and task:
[source,c]
--
20 struct perf_event_attr attr1 = {
21 .type = PERF_TYPE_SOFTWARE,
22 .config = PERF_COUNT_SW_CPU_CLOCK,
23 .read_format = PERF_FORMAT_TOTAL_TIME_ENABLED|PERF_FORMAT_TOTAL_TIME_RUNNING,
24 .disabled = 1,
25 };
26 struct perf_event_attr attr2 = {
27 .type = PERF_TYPE_SOFTWARE,
28 .config = PERF_COUNT_SW_TASK_CLOCK,
29 .read_format = PERF_FORMAT_TOTAL_TIME_ENABLED|PERF_FORMAT_TOTAL_TIME_RUNNING,
30 .disabled = 1,
31 };
--
The `read_format` setup tells perf to include timing details together with each count.
Next step is to prepare threads map.
In this case we will monitor current process, so we create threads map with single pid (0):
[source,c]
--
37 threads = perf_thread_map__new_dummy();
38 if (!threads) {
39 fprintf(stderr, "failed to create threads\n");
40 return -1;
41 }
42
43 perf_thread_map__set_pid(threads, 0, 0);
--
Now we create libperf's event list, which will serve as holder for the events we want:
[source,c]
--
45 evlist = perf_evlist__new();
46 if (!evlist) {
47 fprintf(stderr, "failed to create evlist\n");
48 goto out_threads;
49 }
--
We create libperf's events for the attributes we defined earlier and add them to the list:
[source,c]
--
51 evsel = perf_evsel__new(&attr1);
52 if (!evsel) {
53 fprintf(stderr, "failed to create evsel1\n");
54 goto out_evlist;
55 }
56
57 perf_evlist__add(evlist, evsel);
58
59 evsel = perf_evsel__new(&attr2);
60 if (!evsel) {
61 fprintf(stderr, "failed to create evsel2\n");
62 goto out_evlist;
63 }
64
65 perf_evlist__add(evlist, evsel);
--
Configure event list with the thread map and open events:
[source,c]
--
67 perf_evlist__set_maps(evlist, NULL, threads);
68
69 err = perf_evlist__open(evlist);
70 if (err) {
71 fprintf(stderr, "failed to open evsel\n");
72 goto out_evlist;
73 }
--
Both events are created as disabled (note the `disabled = 1` assignment above),
so we need to enable the whole list explicitly (both events).
From this moment events are counting and we can do our workload.
When we are done we disable the events list.
[source,c]
--
75 perf_evlist__enable(evlist);
76
77 while (count--);
78
79 perf_evlist__disable(evlist);
--
Now we need to get the counts from events, following code iterates through the
events list and read counts:
[source,c]
--
81 perf_evlist__for_each_evsel(evlist, evsel) {
82 perf_evsel__read(evsel, 0, 0, &counts);
83 fprintf(stdout, "count %llu, enabled %llu, run %llu\n",
84 counts.val, counts.ena, counts.run);
85 }
--
And finally cleanup.
We close the whole events list (both events) and remove it together with the threads map:
[source,c]
--
87 perf_evlist__close(evlist);
88
89 out_evlist:
90 perf_evlist__delete(evlist);
91 out_threads:
92 perf_thread_map__put(threads);
93 return err;
94 }
--
REPORTING BUGS
--------------
Report bugs to <linux-perf-users@vger.kernel.org>.
LICENSE
-------
libperf is Free Software licensed under the GNU LGPL 2.1
RESOURCES
---------
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
SEE ALSO
--------
libperf(3), libperf-sampling(7)
|
