1libperf-counting(7) 2=================== 3 4NAME 5---- 6libperf-counting - counting interface 7 8DESCRIPTION 9----------- 10The counting interface provides API to meassure and get count for specific perf events. 11 12The following test tries to explain count on `counting.c` example. 13 14It is by no means complete guide to counting, but shows libperf basic API for counting. 15 16The `counting.c` comes with libbperf package and can be compiled and run like: 17 18[source,bash] 19-- 20$ gcc -o counting counting.c -lperf 21$ sudo ./counting 22count 176792, enabled 176944, run 176944 23count 176242, enabled 176242, run 176242 24-- 25 26It requires root access, because of the `PERF_COUNT_SW_CPU_CLOCK` event, 27which is available only for root. 28 29The `counting.c` example monitors two events on the current process and displays their count, in a nutshel it: 30 31* creates events 32* adds them to the event list 33* opens and enables events through the event list 34* does some workload 35* disables events 36* reads and displays event counts 37* destroys the event list 38 39The first thing you need to do before using libperf is to call init function: 40 41[source,c] 42-- 43 8 static int libperf_print(enum libperf_print_level level, 44 9 const char *fmt, va_list ap) 45 10 { 46 11 return vfprintf(stderr, fmt, ap); 47 12 } 48 49 14 int main(int argc, char **argv) 50 15 { 51 ... 52 35 libperf_init(libperf_print); 53-- 54 55It will setup the library and sets function for debug output from library. 56 57The `libperf_print` callback will receive any message with its debug level, 58defined as: 59 60[source,c] 61-- 62enum libperf_print_level { 63 LIBPERF_ERR, 64 LIBPERF_WARN, 65 LIBPERF_INFO, 66 LIBPERF_DEBUG, 67 LIBPERF_DEBUG2, 68 LIBPERF_DEBUG3, 69}; 70-- 71 72Once the setup is complete we start by defining specific events using the `struct perf_event_attr`. 73 74We create software events for cpu and task: 75 76[source,c] 77-- 78 20 struct perf_event_attr attr1 = { 79 21 .type = PERF_TYPE_SOFTWARE, 80 22 .config = PERF_COUNT_SW_CPU_CLOCK, 81 23 .read_format = PERF_FORMAT_TOTAL_TIME_ENABLED|PERF_FORMAT_TOTAL_TIME_RUNNING, 82 24 .disabled = 1, 83 25 }; 84 26 struct perf_event_attr attr2 = { 85 27 .type = PERF_TYPE_SOFTWARE, 86 28 .config = PERF_COUNT_SW_TASK_CLOCK, 87 29 .read_format = PERF_FORMAT_TOTAL_TIME_ENABLED|PERF_FORMAT_TOTAL_TIME_RUNNING, 88 30 .disabled = 1, 89 31 }; 90-- 91 92The `read_format` setup tells perf to include timing details together with each count. 93 94Next step is to prepare threads map. 95 96In this case we will monitor current process, so we create threads map with single pid (0): 97 98[source,c] 99-- 100 37 threads = perf_thread_map__new_dummy(); 101 38 if (!threads) { 102 39 fprintf(stderr, "failed to create threads\n"); 103 40 return -1; 104 41 } 105 42 106 43 perf_thread_map__set_pid(threads, 0, 0); 107-- 108 109Now we create libperf's event list, which will serve as holder for the events we want: 110 111[source,c] 112-- 113 45 evlist = perf_evlist__new(); 114 46 if (!evlist) { 115 47 fprintf(stderr, "failed to create evlist\n"); 116 48 goto out_threads; 117 49 } 118-- 119 120We create libperf's events for the attributes we defined earlier and add them to the list: 121 122[source,c] 123-- 124 51 evsel = perf_evsel__new(&attr1); 125 52 if (!evsel) { 126 53 fprintf(stderr, "failed to create evsel1\n"); 127 54 goto out_evlist; 128 55 } 129 56 130 57 perf_evlist__add(evlist, evsel); 131 58 132 59 evsel = perf_evsel__new(&attr2); 133 60 if (!evsel) { 134 61 fprintf(stderr, "failed to create evsel2\n"); 135 62 goto out_evlist; 136 63 } 137 64 138 65 perf_evlist__add(evlist, evsel); 139-- 140 141Configure event list with the thread map and open events: 142 143[source,c] 144-- 145 67 perf_evlist__set_maps(evlist, NULL, threads); 146 68 147 69 err = perf_evlist__open(evlist); 148 70 if (err) { 149 71 fprintf(stderr, "failed to open evsel\n"); 150 72 goto out_evlist; 151 73 } 152-- 153 154Both events are created as disabled (note the `disabled = 1` assignment above), 155so we need to enable the whole list explicitely (both events). 156 157From this moment events are counting and we can do our workload. 158 159When we are done we disable the events list. 160 161[source,c] 162-- 163 75 perf_evlist__enable(evlist); 164 76 165 77 while (count--); 166 78 167 79 perf_evlist__disable(evlist); 168-- 169 170Now we need to get the counts from events, following code iterates throught the events list and read counts: 171 172[source,c] 173-- 174 81 perf_evlist__for_each_evsel(evlist, evsel) { 175 82 perf_evsel__read(evsel, 0, 0, &counts); 176 83 fprintf(stdout, "count %llu, enabled %llu, run %llu\n", 177 84 counts.val, counts.ena, counts.run); 178 85 } 179-- 180 181And finaly cleanup. 182 183We close the whole events list (both events) and remove it together with the threads map: 184 185[source,c] 186-- 187 87 perf_evlist__close(evlist); 188 88 189 89 out_evlist: 190 90 perf_evlist__delete(evlist); 191 91 out_threads: 192 92 perf_thread_map__put(threads); 193 93 return err; 194 94 } 195-- 196 197REPORTING BUGS 198-------------- 199Report bugs to <linux-perf-users@vger.kernel.org>. 200 201LICENSE 202------- 203libperf is Free Software licensed under the GNU LGPL 2.1 204 205RESOURCES 206--------- 207https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git 208 209SEE ALSO 210-------- 211libperf(3), libperf-sampling(7) 212