1/*
  2 * builtin-ftrace.c
  3 *
  4 * Copyright (c) 2013  LG Electronics,  Namhyung Kim <namhyung@kernel.org>
  5 *
  6 * Released under the GPL v2.
  7 */
  8
  9#include "builtin.h"
 10#include "perf.h"
 11
 12#include <errno.h>
 13#include <unistd.h>
 14#include <signal.h>
 15#include <fcntl.h>
 16#include <poll.h>
 17
 18#include "debug.h"
 19#include <subcmd/parse-options.h>
 20#include <api/fs/tracing_path.h>
 21#include "evlist.h"
 22#include "target.h"
 23#include "cpumap.h"
 24#include "thread_map.h"
 25#include "util/config.h"
 26
 27
 28#define DEFAULT_TRACER  "function_graph"
 29
 30struct perf_ftrace {
 31	struct perf_evlist	*evlist;
 32	struct target		target;
 33	const char		*tracer;
 34	struct list_head	filters;
 35	struct list_head	notrace;
 36	struct list_head	graph_funcs;
 37	struct list_head	nograph_funcs;
 38	int			graph_depth;
 39};
 40
 41struct filter_entry {
 42	struct list_head	list;
 43	char			name[];
 44};
 45
 46static bool done;
 47
 48static void sig_handler(int sig __maybe_unused)
 49{
 50	done = true;
 51}
 52
 53/*
 54 * perf_evlist__prepare_workload will send a SIGUSR1 if the fork fails, since
 55 * we asked by setting its exec_error to the function below,
 56 * ftrace__workload_exec_failed_signal.
 57 *
 58 * XXX We need to handle this more appropriately, emitting an error, etc.
 59 */
 60static void ftrace__workload_exec_failed_signal(int signo __maybe_unused,
 61						siginfo_t *info __maybe_unused,
 62						void *ucontext __maybe_unused)
 63{
 64	/* workload_exec_errno = info->si_value.sival_int; */
 65	done = true;
 66}
 67
 68static int __write_tracing_file(const char *name, const char *val, bool append)
 69{
 70	char *file;
 71	int fd, ret = -1;
 72	ssize_t size = strlen(val);
 73	int flags = O_WRONLY;
 74	char errbuf[512];
 75	char *val_copy;
 76
 77	file = get_tracing_file(name);
 78	if (!file) {
 79		pr_debug("cannot get tracing file: %s\n", name);
 80		return -1;
 81	}
 82
 83	if (append)
 84		flags |= O_APPEND;
 85	else
 86		flags |= O_TRUNC;
 87
 88	fd = open(file, flags);
 89	if (fd < 0) {
 90		pr_debug("cannot open tracing file: %s: %s\n",
 91			 name, str_error_r(errno, errbuf, sizeof(errbuf)));
 92		goto out;
 93	}
 94
 95	/*
 96	 * Copy the original value and append a '\n'. Without this,
 97	 * the kernel can hide possible errors.
 98	 */
 99	val_copy = strdup(val);
100	if (!val_copy)
101		goto out_close;
102	val_copy[size] = '\n';
103
104	if (write(fd, val_copy, size + 1) == size + 1)
105		ret = 0;
106	else
107		pr_debug("write '%s' to tracing/%s failed: %s\n",
108			 val, name, str_error_r(errno, errbuf, sizeof(errbuf)));
109
110	free(val_copy);
111out_close:
112	close(fd);
113out:
114	put_tracing_file(file);
115	return ret;
116}
117
118static int write_tracing_file(const char *name, const char *val)
119{
120	return __write_tracing_file(name, val, false);
121}
122
123static int append_tracing_file(const char *name, const char *val)
124{
125	return __write_tracing_file(name, val, true);
126}
127
128static int reset_tracing_cpu(void);
129static void reset_tracing_filters(void);
130
131static int reset_tracing_files(struct perf_ftrace *ftrace __maybe_unused)
132{
133	if (write_tracing_file("tracing_on", "0") < 0)
134		return -1;
135
136	if (write_tracing_file("current_tracer", "nop") < 0)
137		return -1;
138
139	if (write_tracing_file("set_ftrace_pid", " ") < 0)
140		return -1;
141
142	if (reset_tracing_cpu() < 0)
143		return -1;
144
145	if (write_tracing_file("max_graph_depth", "0") < 0)
146		return -1;
147
148	reset_tracing_filters();
149	return 0;
150}
151
152static int set_tracing_pid(struct perf_ftrace *ftrace)
153{
154	int i;
155	char buf[16];
156
157	if (target__has_cpu(&ftrace->target))
158		return 0;
159
160	for (i = 0; i < thread_map__nr(ftrace->evlist->threads); i++) {
161		scnprintf(buf, sizeof(buf), "%d",
162			  ftrace->evlist->threads->map[i]);
163		if (append_tracing_file("set_ftrace_pid", buf) < 0)
164			return -1;
165	}
166	return 0;
167}
168
169static int set_tracing_cpumask(struct cpu_map *cpumap)
170{
171	char *cpumask;
172	size_t mask_size;
173	int ret;
174	int last_cpu;
175
176	last_cpu = cpu_map__cpu(cpumap, cpumap->nr - 1);
177	mask_size = (last_cpu + 3) / 4 + 1;
178	mask_size += last_cpu / 32; /* ',' is needed for every 32th cpus */
179
180	cpumask = malloc(mask_size);
181	if (cpumask == NULL) {
182		pr_debug("failed to allocate cpu mask\n");
183		return -1;
184	}
185
186	cpu_map__snprint_mask(cpumap, cpumask, mask_size);
187
188	ret = write_tracing_file("tracing_cpumask", cpumask);
189
190	free(cpumask);
191	return ret;
192}
193
194static int set_tracing_cpu(struct perf_ftrace *ftrace)
195{
196	struct cpu_map *cpumap = ftrace->evlist->cpus;
197
198	if (!target__has_cpu(&ftrace->target))
199		return 0;
200
201	return set_tracing_cpumask(cpumap);
202}
203
204static int reset_tracing_cpu(void)
205{
206	struct cpu_map *cpumap = cpu_map__new(NULL);
207	int ret;
208
209	ret = set_tracing_cpumask(cpumap);
210	cpu_map__put(cpumap);
211	return ret;
212}
213
214static int __set_tracing_filter(const char *filter_file, struct list_head *funcs)
215{
216	struct filter_entry *pos;
217
218	list_for_each_entry(pos, funcs, list) {
219		if (append_tracing_file(filter_file, pos->name) < 0)
220			return -1;
221	}
222
223	return 0;
224}
225
226static int set_tracing_filters(struct perf_ftrace *ftrace)
227{
228	int ret;
229
230	ret = __set_tracing_filter("set_ftrace_filter", &ftrace->filters);
231	if (ret < 0)
232		return ret;
233
234	ret = __set_tracing_filter("set_ftrace_notrace", &ftrace->notrace);
235	if (ret < 0)
236		return ret;
237
238	ret = __set_tracing_filter("set_graph_function", &ftrace->graph_funcs);
239	if (ret < 0)
240		return ret;
241
242	/* old kernels do not have this filter */
243	__set_tracing_filter("set_graph_notrace", &ftrace->nograph_funcs);
244
245	return ret;
246}
247
248static void reset_tracing_filters(void)
249{
250	write_tracing_file("set_ftrace_filter", " ");
251	write_tracing_file("set_ftrace_notrace", " ");
252	write_tracing_file("set_graph_function", " ");
253	write_tracing_file("set_graph_notrace", " ");
254}
255
256static int set_tracing_depth(struct perf_ftrace *ftrace)
257{
258	char buf[16];
259
260	if (ftrace->graph_depth == 0)
261		return 0;
262
263	if (ftrace->graph_depth < 0) {
264		pr_err("invalid graph depth: %d\n", ftrace->graph_depth);
265		return -1;
266	}
267
268	snprintf(buf, sizeof(buf), "%d", ftrace->graph_depth);
269
270	if (write_tracing_file("max_graph_depth", buf) < 0)
271		return -1;
272
273	return 0;
274}
275
276static int __cmd_ftrace(struct perf_ftrace *ftrace, int argc, const char **argv)
277{
278	char *trace_file;
279	int trace_fd;
280	char buf[4096];
281	struct pollfd pollfd = {
282		.events = POLLIN,
283	};
284
285	if (geteuid() != 0) {
286		pr_err("ftrace only works for root!\n");
287		return -1;
288	}
289
290	signal(SIGINT, sig_handler);
291	signal(SIGUSR1, sig_handler);
292	signal(SIGCHLD, sig_handler);
293	signal(SIGPIPE, sig_handler);
294
295	if (reset_tracing_files(ftrace) < 0) {
296		pr_err("failed to reset ftrace\n");
297		goto out;
298	}
299
300	/* reset ftrace buffer */
301	if (write_tracing_file("trace", "0") < 0)
302		goto out;
303
304	if (argc && perf_evlist__prepare_workload(ftrace->evlist,
305				&ftrace->target, argv, false,
306				ftrace__workload_exec_failed_signal) < 0) {
307		goto out;
308	}
309
310	if (set_tracing_pid(ftrace) < 0) {
311		pr_err("failed to set ftrace pid\n");
312		goto out_reset;
313	}
314
315	if (set_tracing_cpu(ftrace) < 0) {
316		pr_err("failed to set tracing cpumask\n");
317		goto out_reset;
318	}
319
320	if (set_tracing_filters(ftrace) < 0) {
321		pr_err("failed to set tracing filters\n");
322		goto out_reset;
323	}
324
325	if (set_tracing_depth(ftrace) < 0) {
326		pr_err("failed to set graph depth\n");
327		goto out_reset;
328	}
329
330	if (write_tracing_file("current_tracer", ftrace->tracer) < 0) {
331		pr_err("failed to set current_tracer to %s\n", ftrace->tracer);
332		goto out_reset;
333	}
334
335	setup_pager();
336
337	trace_file = get_tracing_file("trace_pipe");
338	if (!trace_file) {
339		pr_err("failed to open trace_pipe\n");
340		goto out_reset;
341	}
342
343	trace_fd = open(trace_file, O_RDONLY);
344
345	put_tracing_file(trace_file);
346
347	if (trace_fd < 0) {
348		pr_err("failed to open trace_pipe\n");
349		goto out_reset;
350	}
351
352	fcntl(trace_fd, F_SETFL, O_NONBLOCK);
353	pollfd.fd = trace_fd;
354
355	if (write_tracing_file("tracing_on", "1") < 0) {
356		pr_err("can't enable tracing\n");
357		goto out_close_fd;
358	}
359
360	perf_evlist__start_workload(ftrace->evlist);
361
362	while (!done) {
363		if (poll(&pollfd, 1, -1) < 0)
364			break;
365
366		if (pollfd.revents & POLLIN) {
367			int n = read(trace_fd, buf, sizeof(buf));
368			if (n < 0)
369				break;
370			if (fwrite(buf, n, 1, stdout) != 1)
371				break;
372		}
373	}
374
375	write_tracing_file("tracing_on", "0");
376
377	/* read remaining buffer contents */
378	while (true) {
379		int n = read(trace_fd, buf, sizeof(buf));
380		if (n <= 0)
381			break;
382		if (fwrite(buf, n, 1, stdout) != 1)
383			break;
384	}
385
386out_close_fd:
387	close(trace_fd);
388out_reset:
389	reset_tracing_files(ftrace);
390out:
391	return done ? 0 : -1;
392}
393
394static int perf_ftrace_config(const char *var, const char *value, void *cb)
395{
396	struct perf_ftrace *ftrace = cb;
397
398	if (!strstarts(var, "ftrace."))
399		return 0;
400
401	if (strcmp(var, "ftrace.tracer"))
402		return -1;
403
404	if (!strcmp(value, "function_graph") ||
405	    !strcmp(value, "function")) {
406		ftrace->tracer = value;
407		return 0;
408	}
409
410	pr_err("Please select \"function_graph\" (default) or \"function\"\n");
411	return -1;
412}
413
414static int parse_filter_func(const struct option *opt, const char *str,
415			     int unset __maybe_unused)
416{
417	struct list_head *head = opt->value;
418	struct filter_entry *entry;
419
420	entry = malloc(sizeof(*entry) + strlen(str) + 1);
421	if (entry == NULL)
422		return -ENOMEM;
423
424	strcpy(entry->name, str);
425	list_add_tail(&entry->list, head);
426
427	return 0;
428}
429
430static void delete_filter_func(struct list_head *head)
431{
432	struct filter_entry *pos, *tmp;
433
434	list_for_each_entry_safe(pos, tmp, head, list) {
435		list_del(&pos->list);
436		free(pos);
437	}
438}
439
440int cmd_ftrace(int argc, const char **argv)
441{
442	int ret;
443	struct perf_ftrace ftrace = {
444		.tracer = DEFAULT_TRACER,
445		.target = { .uid = UINT_MAX, },
446	};
447	const char * const ftrace_usage[] = {
448		"perf ftrace [<options>] [<command>]",
449		"perf ftrace [<options>] -- <command> [<options>]",
450		NULL
451	};
452	const struct option ftrace_options[] = {
453	OPT_STRING('t', "tracer", &ftrace.tracer, "tracer",
454		   "tracer to use: function_graph(default) or function"),
455	OPT_STRING('p', "pid", &ftrace.target.pid, "pid",
456		   "trace on existing process id"),
457	OPT_INCR('v', "verbose", &verbose,
458		 "be more verbose"),
459	OPT_BOOLEAN('a', "all-cpus", &ftrace.target.system_wide,
460		    "system-wide collection from all CPUs"),
461	OPT_STRING('C', "cpu", &ftrace.target.cpu_list, "cpu",
462		    "list of cpus to monitor"),
463	OPT_CALLBACK('T', "trace-funcs", &ftrace.filters, "func",
464		     "trace given functions only", parse_filter_func),
465	OPT_CALLBACK('N', "notrace-funcs", &ftrace.notrace, "func",
466		     "do not trace given functions", parse_filter_func),
467	OPT_CALLBACK('G', "graph-funcs", &ftrace.graph_funcs, "func",
468		     "Set graph filter on given functions", parse_filter_func),
469	OPT_CALLBACK('g', "nograph-funcs", &ftrace.nograph_funcs, "func",
470		     "Set nograph filter on given functions", parse_filter_func),
471	OPT_INTEGER('D', "graph-depth", &ftrace.graph_depth,
472		    "Max depth for function graph tracer"),
473	OPT_END()
474	};
475
476	INIT_LIST_HEAD(&ftrace.filters);
477	INIT_LIST_HEAD(&ftrace.notrace);
478	INIT_LIST_HEAD(&ftrace.graph_funcs);
479	INIT_LIST_HEAD(&ftrace.nograph_funcs);
480
481	ret = perf_config(perf_ftrace_config, &ftrace);
482	if (ret < 0)
483		return -1;
484
485	argc = parse_options(argc, argv, ftrace_options, ftrace_usage,
486			    PARSE_OPT_STOP_AT_NON_OPTION);
487	if (!argc && target__none(&ftrace.target))
488		usage_with_options(ftrace_usage, ftrace_options);
489
490	ret = target__validate(&ftrace.target);
491	if (ret) {
492		char errbuf[512];
493
494		target__strerror(&ftrace.target, ret, errbuf, 512);
495		pr_err("%s\n", errbuf);
496		goto out_delete_filters;
497	}
498
499	ftrace.evlist = perf_evlist__new();
500	if (ftrace.evlist == NULL) {
501		ret = -ENOMEM;
502		goto out_delete_filters;
503	}
504
505	ret = perf_evlist__create_maps(ftrace.evlist, &ftrace.target);
506	if (ret < 0)
507		goto out_delete_evlist;
508
509	ret = __cmd_ftrace(&ftrace, argc, argv);
510
511out_delete_evlist:
512	perf_evlist__delete(ftrace.evlist);
513
514out_delete_filters:
515	delete_filter_func(&ftrace.filters);
516	delete_filter_func(&ftrace.notrace);
517	delete_filter_func(&ftrace.graph_funcs);
518	delete_filter_func(&ftrace.nograph_funcs);
519
520	return ret;
521}