xref: /openbmc/linux/tools/perf/util/record.c (revision 8bbecfb4)
1 // SPDX-License-Identifier: GPL-2.0
2 #include "debug.h"
3 #include "evlist.h"
4 #include "evsel.h"
5 #include "evsel_config.h"
6 #include "parse-events.h"
7 #include <errno.h>
8 #include <limits.h>
9 #include <stdlib.h>
10 #include <api/fs/fs.h>
11 #include <subcmd/parse-options.h>
12 #include <perf/cpumap.h>
13 #include "cloexec.h"
14 #include "util/perf_api_probe.h"
15 #include "record.h"
16 #include "../perf-sys.h"
17 #include "topdown.h"
18 #include "map_symbol.h"
19 #include "mem-events.h"
20 
21 /*
22  * evsel__config_leader_sampling() uses special rules for leader sampling.
23  * However, if the leader is an AUX area event, then assume the event to sample
24  * is the next event.
25  */
26 static struct evsel *evsel__read_sampler(struct evsel *evsel, struct evlist *evlist)
27 {
28 	struct evsel *leader = evsel->leader;
29 
30 	if (evsel__is_aux_event(leader) || arch_topdown_sample_read(leader) ||
31 	    is_mem_loads_aux_event(leader)) {
32 		evlist__for_each_entry(evlist, evsel) {
33 			if (evsel->leader == leader && evsel != evsel->leader)
34 				return evsel;
35 		}
36 	}
37 
38 	return leader;
39 }
40 
41 static u64 evsel__config_term_mask(struct evsel *evsel)
42 {
43 	struct evsel_config_term *term;
44 	struct list_head *config_terms = &evsel->config_terms;
45 	u64 term_types = 0;
46 
47 	list_for_each_entry(term, config_terms, list) {
48 		term_types |= 1 << term->type;
49 	}
50 	return term_types;
51 }
52 
53 static void evsel__config_leader_sampling(struct evsel *evsel, struct evlist *evlist)
54 {
55 	struct perf_event_attr *attr = &evsel->core.attr;
56 	struct evsel *leader = evsel->leader;
57 	struct evsel *read_sampler;
58 	u64 term_types, freq_mask;
59 
60 	if (!leader->sample_read)
61 		return;
62 
63 	read_sampler = evsel__read_sampler(evsel, evlist);
64 
65 	if (evsel == read_sampler)
66 		return;
67 
68 	term_types = evsel__config_term_mask(evsel);
69 	/*
70 	 * Disable sampling for all group members except those with explicit
71 	 * config terms or the leader. In the case of an AUX area event, the 2nd
72 	 * event in the group is the one that 'leads' the sampling.
73 	 */
74 	freq_mask = (1 << EVSEL__CONFIG_TERM_FREQ) | (1 << EVSEL__CONFIG_TERM_PERIOD);
75 	if ((term_types & freq_mask) == 0) {
76 		attr->freq           = 0;
77 		attr->sample_freq    = 0;
78 		attr->sample_period  = 0;
79 	}
80 	if ((term_types & (1 << EVSEL__CONFIG_TERM_OVERWRITE)) == 0)
81 		attr->write_backward = 0;
82 
83 	/*
84 	 * We don't get a sample for slave events, we make them when delivering
85 	 * the group leader sample. Set the slave event to follow the master
86 	 * sample_type to ease up reporting.
87 	 * An AUX area event also has sample_type requirements, so also include
88 	 * the sample type bits from the leader's sample_type to cover that
89 	 * case.
90 	 */
91 	attr->sample_type = read_sampler->core.attr.sample_type |
92 			    leader->core.attr.sample_type;
93 }
94 
95 void evlist__config(struct evlist *evlist, struct record_opts *opts, struct callchain_param *callchain)
96 {
97 	struct evsel *evsel;
98 	bool use_sample_identifier = false;
99 	bool use_comm_exec;
100 	bool sample_id = opts->sample_id;
101 
102 	/*
103 	 * Set the evsel leader links before we configure attributes,
104 	 * since some might depend on this info.
105 	 */
106 	if (opts->group)
107 		evlist__set_leader(evlist);
108 
109 	if (evlist->core.cpus->map[0] < 0)
110 		opts->no_inherit = true;
111 
112 	use_comm_exec = perf_can_comm_exec();
113 
114 	evlist__for_each_entry(evlist, evsel) {
115 		evsel__config(evsel, opts, callchain);
116 		if (evsel->tracking && use_comm_exec)
117 			evsel->core.attr.comm_exec = 1;
118 	}
119 
120 	/* Configure leader sampling here now that the sample type is known */
121 	evlist__for_each_entry(evlist, evsel)
122 		evsel__config_leader_sampling(evsel, evlist);
123 
124 	if (opts->full_auxtrace) {
125 		/*
126 		 * Need to be able to synthesize and parse selected events with
127 		 * arbitrary sample types, which requires always being able to
128 		 * match the id.
129 		 */
130 		use_sample_identifier = perf_can_sample_identifier();
131 		sample_id = true;
132 	} else if (evlist->core.nr_entries > 1) {
133 		struct evsel *first = evlist__first(evlist);
134 
135 		evlist__for_each_entry(evlist, evsel) {
136 			if (evsel->core.attr.sample_type == first->core.attr.sample_type)
137 				continue;
138 			use_sample_identifier = perf_can_sample_identifier();
139 			break;
140 		}
141 		sample_id = true;
142 	}
143 
144 	if (sample_id) {
145 		evlist__for_each_entry(evlist, evsel)
146 			evsel__set_sample_id(evsel, use_sample_identifier);
147 	}
148 
149 	evlist__set_id_pos(evlist);
150 }
151 
152 static int get_max_rate(unsigned int *rate)
153 {
154 	return sysctl__read_int("kernel/perf_event_max_sample_rate", (int *)rate);
155 }
156 
157 static int record_opts__config_freq(struct record_opts *opts)
158 {
159 	bool user_freq = opts->user_freq != UINT_MAX;
160 	unsigned int max_rate;
161 
162 	if (opts->user_interval != ULLONG_MAX)
163 		opts->default_interval = opts->user_interval;
164 	if (user_freq)
165 		opts->freq = opts->user_freq;
166 
167 	/*
168 	 * User specified count overrides default frequency.
169 	 */
170 	if (opts->default_interval)
171 		opts->freq = 0;
172 	else if (opts->freq) {
173 		opts->default_interval = opts->freq;
174 	} else {
175 		pr_err("frequency and count are zero, aborting\n");
176 		return -1;
177 	}
178 
179 	if (get_max_rate(&max_rate))
180 		return 0;
181 
182 	/*
183 	 * User specified frequency is over current maximum.
184 	 */
185 	if (user_freq && (max_rate < opts->freq)) {
186 		if (opts->strict_freq) {
187 			pr_err("error: Maximum frequency rate (%'u Hz) exceeded.\n"
188 			       "       Please use -F freq option with a lower value or consider\n"
189 			       "       tweaking /proc/sys/kernel/perf_event_max_sample_rate.\n",
190 			       max_rate);
191 			return -1;
192 		} else {
193 			pr_warning("warning: Maximum frequency rate (%'u Hz) exceeded, throttling from %'u Hz to %'u Hz.\n"
194 				   "         The limit can be raised via /proc/sys/kernel/perf_event_max_sample_rate.\n"
195 				   "         The kernel will lower it when perf's interrupts take too long.\n"
196 				   "         Use --strict-freq to disable this throttling, refusing to record.\n",
197 				   max_rate, opts->freq, max_rate);
198 
199 			opts->freq = max_rate;
200 		}
201 	}
202 
203 	/*
204 	 * Default frequency is over current maximum.
205 	 */
206 	if (max_rate < opts->freq) {
207 		pr_warning("Lowering default frequency rate from %u to %u.\n"
208 			   "Please consider tweaking "
209 			   "/proc/sys/kernel/perf_event_max_sample_rate.\n",
210 			   opts->freq, max_rate);
211 		opts->freq = max_rate;
212 	}
213 
214 	return 0;
215 }
216 
217 int record_opts__config(struct record_opts *opts)
218 {
219 	return record_opts__config_freq(opts);
220 }
221 
222 bool evlist__can_select_event(struct evlist *evlist, const char *str)
223 {
224 	struct evlist *temp_evlist;
225 	struct evsel *evsel;
226 	int err, fd, cpu;
227 	bool ret = false;
228 	pid_t pid = -1;
229 
230 	temp_evlist = evlist__new();
231 	if (!temp_evlist)
232 		return false;
233 
234 	err = parse_events(temp_evlist, str, NULL);
235 	if (err)
236 		goto out_delete;
237 
238 	evsel = evlist__last(temp_evlist);
239 
240 	if (!evlist || perf_cpu_map__empty(evlist->core.cpus)) {
241 		struct perf_cpu_map *cpus = perf_cpu_map__new(NULL);
242 
243 		cpu =  cpus ? cpus->map[0] : 0;
244 		perf_cpu_map__put(cpus);
245 	} else {
246 		cpu = evlist->core.cpus->map[0];
247 	}
248 
249 	while (1) {
250 		fd = sys_perf_event_open(&evsel->core.attr, pid, cpu, -1,
251 					 perf_event_open_cloexec_flag());
252 		if (fd < 0) {
253 			if (pid == -1 && errno == EACCES) {
254 				pid = 0;
255 				continue;
256 			}
257 			goto out_delete;
258 		}
259 		break;
260 	}
261 	close(fd);
262 	ret = true;
263 
264 out_delete:
265 	evlist__delete(temp_evlist);
266 	return ret;
267 }
268 
269 int record__parse_freq(const struct option *opt, const char *str, int unset __maybe_unused)
270 {
271 	unsigned int freq;
272 	struct record_opts *opts = opt->value;
273 
274 	if (!str)
275 		return -EINVAL;
276 
277 	if (strcasecmp(str, "max") == 0) {
278 		if (get_max_rate(&freq)) {
279 			pr_err("couldn't read /proc/sys/kernel/perf_event_max_sample_rate\n");
280 			return -1;
281 		}
282 		pr_info("info: Using a maximum frequency rate of %'d Hz\n", freq);
283 	} else {
284 		freq = atoi(str);
285 	}
286 
287 	opts->user_freq = freq;
288 	return 0;
289 }
290