Loading...
1// SPDX-License-Identifier: GPL-2.0
2#include "evlist.h"
3#include "evsel.h"
4#include "cpumap.h"
5#include "parse-events.h"
6#include <errno.h>
7#include <api/fs/fs.h>
8#include <subcmd/parse-options.h>
9#include "util.h"
10#include "cloexec.h"
11
12typedef void (*setup_probe_fn_t)(struct perf_evsel *evsel);
13
14static int perf_do_probe_api(setup_probe_fn_t fn, int cpu, const char *str)
15{
16 struct perf_evlist *evlist;
17 struct perf_evsel *evsel;
18 unsigned long flags = perf_event_open_cloexec_flag();
19 int err = -EAGAIN, fd;
20 static pid_t pid = -1;
21
22 evlist = perf_evlist__new();
23 if (!evlist)
24 return -ENOMEM;
25
26 if (parse_events(evlist, str, NULL))
27 goto out_delete;
28
29 evsel = perf_evlist__first(evlist);
30
31 while (1) {
32 fd = sys_perf_event_open(&evsel->attr, pid, cpu, -1, flags);
33 if (fd < 0) {
34 if (pid == -1 && errno == EACCES) {
35 pid = 0;
36 continue;
37 }
38 goto out_delete;
39 }
40 break;
41 }
42 close(fd);
43
44 fn(evsel);
45
46 fd = sys_perf_event_open(&evsel->attr, pid, cpu, -1, flags);
47 if (fd < 0) {
48 if (errno == EINVAL)
49 err = -EINVAL;
50 goto out_delete;
51 }
52 close(fd);
53 err = 0;
54
55out_delete:
56 perf_evlist__delete(evlist);
57 return err;
58}
59
60static bool perf_probe_api(setup_probe_fn_t fn)
61{
62 const char *try[] = {"cycles:u", "instructions:u", "cpu-clock:u", NULL};
63 struct cpu_map *cpus;
64 int cpu, ret, i = 0;
65
66 cpus = cpu_map__new(NULL);
67 if (!cpus)
68 return false;
69 cpu = cpus->map[0];
70 cpu_map__put(cpus);
71
72 do {
73 ret = perf_do_probe_api(fn, cpu, try[i++]);
74 if (!ret)
75 return true;
76 } while (ret == -EAGAIN && try[i]);
77
78 return false;
79}
80
81static void perf_probe_sample_identifier(struct perf_evsel *evsel)
82{
83 evsel->attr.sample_type |= PERF_SAMPLE_IDENTIFIER;
84}
85
86static void perf_probe_comm_exec(struct perf_evsel *evsel)
87{
88 evsel->attr.comm_exec = 1;
89}
90
91static void perf_probe_context_switch(struct perf_evsel *evsel)
92{
93 evsel->attr.context_switch = 1;
94}
95
96bool perf_can_sample_identifier(void)
97{
98 return perf_probe_api(perf_probe_sample_identifier);
99}
100
101static bool perf_can_comm_exec(void)
102{
103 return perf_probe_api(perf_probe_comm_exec);
104}
105
106bool perf_can_record_switch_events(void)
107{
108 return perf_probe_api(perf_probe_context_switch);
109}
110
111bool perf_can_record_cpu_wide(void)
112{
113 struct perf_event_attr attr = {
114 .type = PERF_TYPE_SOFTWARE,
115 .config = PERF_COUNT_SW_CPU_CLOCK,
116 .exclude_kernel = 1,
117 };
118 struct cpu_map *cpus;
119 int cpu, fd;
120
121 cpus = cpu_map__new(NULL);
122 if (!cpus)
123 return false;
124 cpu = cpus->map[0];
125 cpu_map__put(cpus);
126
127 fd = sys_perf_event_open(&attr, -1, cpu, -1, 0);
128 if (fd < 0)
129 return false;
130 close(fd);
131
132 return true;
133}
134
135void perf_evlist__config(struct perf_evlist *evlist, struct record_opts *opts,
136 struct callchain_param *callchain)
137{
138 struct perf_evsel *evsel;
139 bool use_sample_identifier = false;
140 bool use_comm_exec;
141 bool sample_id = opts->sample_id;
142
143 /*
144 * Set the evsel leader links before we configure attributes,
145 * since some might depend on this info.
146 */
147 if (opts->group)
148 perf_evlist__set_leader(evlist);
149
150 if (evlist->cpus->map[0] < 0)
151 opts->no_inherit = true;
152
153 use_comm_exec = perf_can_comm_exec();
154
155 evlist__for_each_entry(evlist, evsel) {
156 perf_evsel__config(evsel, opts, callchain);
157 if (evsel->tracking && use_comm_exec)
158 evsel->attr.comm_exec = 1;
159 }
160
161 if (opts->full_auxtrace) {
162 /*
163 * Need to be able to synthesize and parse selected events with
164 * arbitrary sample types, which requires always being able to
165 * match the id.
166 */
167 use_sample_identifier = perf_can_sample_identifier();
168 sample_id = true;
169 } else if (evlist->nr_entries > 1) {
170 struct perf_evsel *first = perf_evlist__first(evlist);
171
172 evlist__for_each_entry(evlist, evsel) {
173 if (evsel->attr.sample_type == first->attr.sample_type)
174 continue;
175 use_sample_identifier = perf_can_sample_identifier();
176 break;
177 }
178 sample_id = true;
179 }
180
181 if (sample_id) {
182 evlist__for_each_entry(evlist, evsel)
183 perf_evsel__set_sample_id(evsel, use_sample_identifier);
184 }
185
186 perf_evlist__set_id_pos(evlist);
187}
188
189static int get_max_rate(unsigned int *rate)
190{
191 return sysctl__read_int("kernel/perf_event_max_sample_rate", (int *)rate);
192}
193
194static int record_opts__config_freq(struct record_opts *opts)
195{
196 bool user_freq = opts->user_freq != UINT_MAX;
197 unsigned int max_rate;
198
199 if (opts->user_interval != ULLONG_MAX)
200 opts->default_interval = opts->user_interval;
201 if (user_freq)
202 opts->freq = opts->user_freq;
203
204 /*
205 * User specified count overrides default frequency.
206 */
207 if (opts->default_interval)
208 opts->freq = 0;
209 else if (opts->freq) {
210 opts->default_interval = opts->freq;
211 } else {
212 pr_err("frequency and count are zero, aborting\n");
213 return -1;
214 }
215
216 if (get_max_rate(&max_rate))
217 return 0;
218
219 /*
220 * User specified frequency is over current maximum.
221 */
222 if (user_freq && (max_rate < opts->freq)) {
223 if (opts->strict_freq) {
224 pr_err("error: Maximum frequency rate (%'u Hz) exceeded.\n"
225 " Please use -F freq option with a lower value or consider\n"
226 " tweaking /proc/sys/kernel/perf_event_max_sample_rate.\n",
227 max_rate);
228 return -1;
229 } else {
230 pr_warning("warning: Maximum frequency rate (%'u Hz) exceeded, throttling from %'u Hz to %'u Hz.\n"
231 " The limit can be raised via /proc/sys/kernel/perf_event_max_sample_rate.\n"
232 " The kernel will lower it when perf's interrupts take too long.\n"
233 " Use --strict-freq to disable this throttling, refusing to record.\n",
234 max_rate, opts->freq, max_rate);
235
236 opts->freq = max_rate;
237 }
238 }
239
240 /*
241 * Default frequency is over current maximum.
242 */
243 if (max_rate < opts->freq) {
244 pr_warning("Lowering default frequency rate to %u.\n"
245 "Please consider tweaking "
246 "/proc/sys/kernel/perf_event_max_sample_rate.\n",
247 max_rate);
248 opts->freq = max_rate;
249 }
250
251 return 0;
252}
253
254int record_opts__config(struct record_opts *opts)
255{
256 return record_opts__config_freq(opts);
257}
258
259bool perf_evlist__can_select_event(struct perf_evlist *evlist, const char *str)
260{
261 struct perf_evlist *temp_evlist;
262 struct perf_evsel *evsel;
263 int err, fd, cpu;
264 bool ret = false;
265 pid_t pid = -1;
266
267 temp_evlist = perf_evlist__new();
268 if (!temp_evlist)
269 return false;
270
271 err = parse_events(temp_evlist, str, NULL);
272 if (err)
273 goto out_delete;
274
275 evsel = perf_evlist__last(temp_evlist);
276
277 if (!evlist || cpu_map__empty(evlist->cpus)) {
278 struct cpu_map *cpus = cpu_map__new(NULL);
279
280 cpu = cpus ? cpus->map[0] : 0;
281 cpu_map__put(cpus);
282 } else {
283 cpu = evlist->cpus->map[0];
284 }
285
286 while (1) {
287 fd = sys_perf_event_open(&evsel->attr, pid, cpu, -1,
288 perf_event_open_cloexec_flag());
289 if (fd < 0) {
290 if (pid == -1 && errno == EACCES) {
291 pid = 0;
292 continue;
293 }
294 goto out_delete;
295 }
296 break;
297 }
298 close(fd);
299 ret = true;
300
301out_delete:
302 perf_evlist__delete(temp_evlist);
303 return ret;
304}
305
306int record__parse_freq(const struct option *opt, const char *str, int unset __maybe_unused)
307{
308 unsigned int freq;
309 struct record_opts *opts = opt->value;
310
311 if (!str)
312 return -EINVAL;
313
314 if (strcasecmp(str, "max") == 0) {
315 if (get_max_rate(&freq)) {
316 pr_err("couldn't read /proc/sys/kernel/perf_event_max_sample_rate\n");
317 return -1;
318 }
319 pr_info("info: Using a maximum frequency rate of %'d Hz\n", freq);
320 } else {
321 freq = atoi(str);
322 }
323
324 opts->user_freq = freq;
325 return 0;
326}
1#include "evlist.h"
2#include "evsel.h"
3#include "cpumap.h"
4#include "parse-events.h"
5#include <api/fs/fs.h>
6#include "util.h"
7#include "cloexec.h"
8
9typedef void (*setup_probe_fn_t)(struct perf_evsel *evsel);
10
11static int perf_do_probe_api(setup_probe_fn_t fn, int cpu, const char *str)
12{
13 struct perf_evlist *evlist;
14 struct perf_evsel *evsel;
15 unsigned long flags = perf_event_open_cloexec_flag();
16 int err = -EAGAIN, fd;
17 static pid_t pid = -1;
18
19 evlist = perf_evlist__new();
20 if (!evlist)
21 return -ENOMEM;
22
23 if (parse_events(evlist, str, NULL))
24 goto out_delete;
25
26 evsel = perf_evlist__first(evlist);
27
28 while (1) {
29 fd = sys_perf_event_open(&evsel->attr, pid, cpu, -1, flags);
30 if (fd < 0) {
31 if (pid == -1 && errno == EACCES) {
32 pid = 0;
33 continue;
34 }
35 goto out_delete;
36 }
37 break;
38 }
39 close(fd);
40
41 fn(evsel);
42
43 fd = sys_perf_event_open(&evsel->attr, pid, cpu, -1, flags);
44 if (fd < 0) {
45 if (errno == EINVAL)
46 err = -EINVAL;
47 goto out_delete;
48 }
49 close(fd);
50 err = 0;
51
52out_delete:
53 perf_evlist__delete(evlist);
54 return err;
55}
56
57static bool perf_probe_api(setup_probe_fn_t fn)
58{
59 const char *try[] = {"cycles:u", "instructions:u", "cpu-clock:u", NULL};
60 struct cpu_map *cpus;
61 int cpu, ret, i = 0;
62
63 cpus = cpu_map__new(NULL);
64 if (!cpus)
65 return false;
66 cpu = cpus->map[0];
67 cpu_map__put(cpus);
68
69 do {
70 ret = perf_do_probe_api(fn, cpu, try[i++]);
71 if (!ret)
72 return true;
73 } while (ret == -EAGAIN && try[i]);
74
75 return false;
76}
77
78static void perf_probe_sample_identifier(struct perf_evsel *evsel)
79{
80 evsel->attr.sample_type |= PERF_SAMPLE_IDENTIFIER;
81}
82
83static void perf_probe_comm_exec(struct perf_evsel *evsel)
84{
85 evsel->attr.comm_exec = 1;
86}
87
88static void perf_probe_context_switch(struct perf_evsel *evsel)
89{
90 evsel->attr.context_switch = 1;
91}
92
93bool perf_can_sample_identifier(void)
94{
95 return perf_probe_api(perf_probe_sample_identifier);
96}
97
98static bool perf_can_comm_exec(void)
99{
100 return perf_probe_api(perf_probe_comm_exec);
101}
102
103bool perf_can_record_switch_events(void)
104{
105 return perf_probe_api(perf_probe_context_switch);
106}
107
108bool perf_can_record_cpu_wide(void)
109{
110 struct perf_event_attr attr = {
111 .type = PERF_TYPE_SOFTWARE,
112 .config = PERF_COUNT_SW_CPU_CLOCK,
113 .exclude_kernel = 1,
114 };
115 struct cpu_map *cpus;
116 int cpu, fd;
117
118 cpus = cpu_map__new(NULL);
119 if (!cpus)
120 return false;
121 cpu = cpus->map[0];
122 cpu_map__put(cpus);
123
124 fd = sys_perf_event_open(&attr, -1, cpu, -1, 0);
125 if (fd < 0)
126 return false;
127 close(fd);
128
129 return true;
130}
131
132void perf_evlist__config(struct perf_evlist *evlist, struct record_opts *opts,
133 struct callchain_param *callchain)
134{
135 struct perf_evsel *evsel;
136 bool use_sample_identifier = false;
137 bool use_comm_exec;
138
139 /*
140 * Set the evsel leader links before we configure attributes,
141 * since some might depend on this info.
142 */
143 if (opts->group)
144 perf_evlist__set_leader(evlist);
145
146 if (evlist->cpus->map[0] < 0)
147 opts->no_inherit = true;
148
149 use_comm_exec = perf_can_comm_exec();
150
151 evlist__for_each_entry(evlist, evsel) {
152 perf_evsel__config(evsel, opts, callchain);
153 if (evsel->tracking && use_comm_exec)
154 evsel->attr.comm_exec = 1;
155 }
156
157 if (opts->full_auxtrace) {
158 /*
159 * Need to be able to synthesize and parse selected events with
160 * arbitrary sample types, which requires always being able to
161 * match the id.
162 */
163 use_sample_identifier = perf_can_sample_identifier();
164 evlist__for_each_entry(evlist, evsel)
165 perf_evsel__set_sample_id(evsel, use_sample_identifier);
166 } else if (evlist->nr_entries > 1) {
167 struct perf_evsel *first = perf_evlist__first(evlist);
168
169 evlist__for_each_entry(evlist, evsel) {
170 if (evsel->attr.sample_type == first->attr.sample_type)
171 continue;
172 use_sample_identifier = perf_can_sample_identifier();
173 break;
174 }
175 evlist__for_each_entry(evlist, evsel)
176 perf_evsel__set_sample_id(evsel, use_sample_identifier);
177 }
178
179 perf_evlist__set_id_pos(evlist);
180}
181
182static int get_max_rate(unsigned int *rate)
183{
184 return sysctl__read_int("kernel/perf_event_max_sample_rate", (int *)rate);
185}
186
187static int record_opts__config_freq(struct record_opts *opts)
188{
189 bool user_freq = opts->user_freq != UINT_MAX;
190 unsigned int max_rate;
191
192 if (opts->user_interval != ULLONG_MAX)
193 opts->default_interval = opts->user_interval;
194 if (user_freq)
195 opts->freq = opts->user_freq;
196
197 /*
198 * User specified count overrides default frequency.
199 */
200 if (opts->default_interval)
201 opts->freq = 0;
202 else if (opts->freq) {
203 opts->default_interval = opts->freq;
204 } else {
205 pr_err("frequency and count are zero, aborting\n");
206 return -1;
207 }
208
209 if (get_max_rate(&max_rate))
210 return 0;
211
212 /*
213 * User specified frequency is over current maximum.
214 */
215 if (user_freq && (max_rate < opts->freq)) {
216 pr_err("Maximum frequency rate (%u) reached.\n"
217 "Please use -F freq option with lower value or consider\n"
218 "tweaking /proc/sys/kernel/perf_event_max_sample_rate.\n",
219 max_rate);
220 return -1;
221 }
222
223 /*
224 * Default frequency is over current maximum.
225 */
226 if (max_rate < opts->freq) {
227 pr_warning("Lowering default frequency rate to %u.\n"
228 "Please consider tweaking "
229 "/proc/sys/kernel/perf_event_max_sample_rate.\n",
230 max_rate);
231 opts->freq = max_rate;
232 }
233
234 return 0;
235}
236
237int record_opts__config(struct record_opts *opts)
238{
239 return record_opts__config_freq(opts);
240}
241
242bool perf_evlist__can_select_event(struct perf_evlist *evlist, const char *str)
243{
244 struct perf_evlist *temp_evlist;
245 struct perf_evsel *evsel;
246 int err, fd, cpu;
247 bool ret = false;
248 pid_t pid = -1;
249
250 temp_evlist = perf_evlist__new();
251 if (!temp_evlist)
252 return false;
253
254 err = parse_events(temp_evlist, str, NULL);
255 if (err)
256 goto out_delete;
257
258 evsel = perf_evlist__last(temp_evlist);
259
260 if (!evlist || cpu_map__empty(evlist->cpus)) {
261 struct cpu_map *cpus = cpu_map__new(NULL);
262
263 cpu = cpus ? cpus->map[0] : 0;
264 cpu_map__put(cpus);
265 } else {
266 cpu = evlist->cpus->map[0];
267 }
268
269 while (1) {
270 fd = sys_perf_event_open(&evsel->attr, pid, cpu, -1,
271 perf_event_open_cloexec_flag());
272 if (fd < 0) {
273 if (pid == -1 && errno == EACCES) {
274 pid = 0;
275 continue;
276 }
277 goto out_delete;
278 }
279 break;
280 }
281 close(fd);
282 ret = true;
283
284out_delete:
285 perf_evlist__delete(temp_evlist);
286 return ret;
287}