1 // SPDX-License-Identifier: GPL-2.0
2 #include <dirent.h>
3 #include <errno.h>
4 #include <fcntl.h>
5 #include <inttypes.h>
6 #include <linux/kernel.h>
7 #include <linux/types.h>
8 #include <sys/types.h>
9 #include <sys/stat.h>
10 #include <unistd.h>
11 #include <uapi/linux/mman.h> /* To get things like MAP_HUGETLB even on older libc headers */
12 #include <api/fs/fs.h>
13 #include <linux/perf_event.h>
14 #include "event.h"
15 #include "debug.h"
16 #include "hist.h"
17 #include "machine.h"
18 #include "sort.h"
19 #include "string2.h"
20 #include "strlist.h"
21 #include "thread.h"
22 #include "thread_map.h"
23 #include "sane_ctype.h"
24 #include "symbol/kallsyms.h"
25 #include "asm/bug.h"
26 #include "stat.h"
27 
28 static const char *perf_event__names[] = {
29 	[0]					= "TOTAL",
30 	[PERF_RECORD_MMAP]			= "MMAP",
31 	[PERF_RECORD_MMAP2]			= "MMAP2",
32 	[PERF_RECORD_LOST]			= "LOST",
33 	[PERF_RECORD_COMM]			= "COMM",
34 	[PERF_RECORD_EXIT]			= "EXIT",
35 	[PERF_RECORD_THROTTLE]			= "THROTTLE",
36 	[PERF_RECORD_UNTHROTTLE]		= "UNTHROTTLE",
37 	[PERF_RECORD_FORK]			= "FORK",
38 	[PERF_RECORD_READ]			= "READ",
39 	[PERF_RECORD_SAMPLE]			= "SAMPLE",
40 	[PERF_RECORD_AUX]			= "AUX",
41 	[PERF_RECORD_ITRACE_START]		= "ITRACE_START",
42 	[PERF_RECORD_LOST_SAMPLES]		= "LOST_SAMPLES",
43 	[PERF_RECORD_SWITCH]			= "SWITCH",
44 	[PERF_RECORD_SWITCH_CPU_WIDE]		= "SWITCH_CPU_WIDE",
45 	[PERF_RECORD_NAMESPACES]		= "NAMESPACES",
46 	[PERF_RECORD_HEADER_ATTR]		= "ATTR",
47 	[PERF_RECORD_HEADER_EVENT_TYPE]		= "EVENT_TYPE",
48 	[PERF_RECORD_HEADER_TRACING_DATA]	= "TRACING_DATA",
49 	[PERF_RECORD_HEADER_BUILD_ID]		= "BUILD_ID",
50 	[PERF_RECORD_FINISHED_ROUND]		= "FINISHED_ROUND",
51 	[PERF_RECORD_ID_INDEX]			= "ID_INDEX",
52 	[PERF_RECORD_AUXTRACE_INFO]		= "AUXTRACE_INFO",
53 	[PERF_RECORD_AUXTRACE]			= "AUXTRACE",
54 	[PERF_RECORD_AUXTRACE_ERROR]		= "AUXTRACE_ERROR",
55 	[PERF_RECORD_THREAD_MAP]		= "THREAD_MAP",
56 	[PERF_RECORD_CPU_MAP]			= "CPU_MAP",
57 	[PERF_RECORD_STAT_CONFIG]		= "STAT_CONFIG",
58 	[PERF_RECORD_STAT]			= "STAT",
59 	[PERF_RECORD_STAT_ROUND]		= "STAT_ROUND",
60 	[PERF_RECORD_EVENT_UPDATE]		= "EVENT_UPDATE",
61 	[PERF_RECORD_TIME_CONV]			= "TIME_CONV",
62 	[PERF_RECORD_HEADER_FEATURE]		= "FEATURE",
63 };
64 
65 static const char *perf_ns__names[] = {
66 	[NET_NS_INDEX]		= "net",
67 	[UTS_NS_INDEX]		= "uts",
68 	[IPC_NS_INDEX]		= "ipc",
69 	[PID_NS_INDEX]		= "pid",
70 	[USER_NS_INDEX]		= "user",
71 	[MNT_NS_INDEX]		= "mnt",
72 	[CGROUP_NS_INDEX]	= "cgroup",
73 };
74 
perf_event__name(unsigned int id)75 const char *perf_event__name(unsigned int id)
76 {
77 	if (id >= ARRAY_SIZE(perf_event__names))
78 		return "INVALID";
79 	if (!perf_event__names[id])
80 		return "UNKNOWN";
81 	return perf_event__names[id];
82 }
83 
perf_ns__name(unsigned int id)84 static const char *perf_ns__name(unsigned int id)
85 {
86 	if (id >= ARRAY_SIZE(perf_ns__names))
87 		return "UNKNOWN";
88 	return perf_ns__names[id];
89 }
90 
perf_tool__process_synth_event(struct perf_tool * tool,union perf_event * event,struct machine * machine,perf_event__handler_t process)91 int perf_tool__process_synth_event(struct perf_tool *tool,
92 				   union perf_event *event,
93 				   struct machine *machine,
94 				   perf_event__handler_t process)
95 {
96 	struct perf_sample synth_sample = {
97 	.pid	   = -1,
98 	.tid	   = -1,
99 	.time	   = -1,
100 	.stream_id = -1,
101 	.cpu	   = -1,
102 	.period	   = 1,
103 	.cpumode   = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK,
104 	};
105 
106 	return process(tool, event, &synth_sample, machine);
107 };
108 
109 /*
110  * Assumes that the first 4095 bytes of /proc/pid/stat contains
111  * the comm, tgid and ppid.
112  */
perf_event__get_comm_ids(pid_t pid,char * comm,size_t len,pid_t * tgid,pid_t * ppid)113 static int perf_event__get_comm_ids(pid_t pid, char *comm, size_t len,
114 				    pid_t *tgid, pid_t *ppid)
115 {
116 	char filename[PATH_MAX];
117 	char bf[4096];
118 	int fd;
119 	size_t size = 0;
120 	ssize_t n;
121 	char *name, *tgids, *ppids;
122 
123 	*tgid = -1;
124 	*ppid = -1;
125 
126 	snprintf(filename, sizeof(filename), "/proc/%d/status", pid);
127 
128 	fd = open(filename, O_RDONLY);
129 	if (fd < 0) {
130 		pr_debug("couldn't open %s\n", filename);
131 		return -1;
132 	}
133 
134 	n = read(fd, bf, sizeof(bf) - 1);
135 	close(fd);
136 	if (n <= 0) {
137 		pr_warning("Couldn't get COMM, tigd and ppid for pid %d\n",
138 			   pid);
139 		return -1;
140 	}
141 	bf[n] = '\0';
142 
143 	name = strstr(bf, "Name:");
144 	tgids = strstr(bf, "Tgid:");
145 	ppids = strstr(bf, "PPid:");
146 
147 	if (name) {
148 		char *nl;
149 
150 		name += 5;  /* strlen("Name:") */
151 		name = ltrim(name);
152 
153 		nl = strchr(name, '\n');
154 		if (nl)
155 			*nl = '\0';
156 
157 		size = strlen(name);
158 		if (size >= len)
159 			size = len - 1;
160 		memcpy(comm, name, size);
161 		comm[size] = '\0';
162 	} else {
163 		pr_debug("Name: string not found for pid %d\n", pid);
164 	}
165 
166 	if (tgids) {
167 		tgids += 5;  /* strlen("Tgid:") */
168 		*tgid = atoi(tgids);
169 	} else {
170 		pr_debug("Tgid: string not found for pid %d\n", pid);
171 	}
172 
173 	if (ppids) {
174 		ppids += 5;  /* strlen("PPid:") */
175 		*ppid = atoi(ppids);
176 	} else {
177 		pr_debug("PPid: string not found for pid %d\n", pid);
178 	}
179 
180 	return 0;
181 }
182 
perf_event__prepare_comm(union perf_event * event,pid_t pid,struct machine * machine,pid_t * tgid,pid_t * ppid)183 static int perf_event__prepare_comm(union perf_event *event, pid_t pid,
184 				    struct machine *machine,
185 				    pid_t *tgid, pid_t *ppid)
186 {
187 	size_t size;
188 
189 	*ppid = -1;
190 
191 	memset(&event->comm, 0, sizeof(event->comm));
192 
193 	if (machine__is_host(machine)) {
194 		if (perf_event__get_comm_ids(pid, event->comm.comm,
195 					     sizeof(event->comm.comm),
196 					     tgid, ppid) != 0) {
197 			return -1;
198 		}
199 	} else {
200 		*tgid = machine->pid;
201 	}
202 
203 	if (*tgid < 0)
204 		return -1;
205 
206 	event->comm.pid = *tgid;
207 	event->comm.header.type = PERF_RECORD_COMM;
208 
209 	size = strlen(event->comm.comm) + 1;
210 	size = PERF_ALIGN(size, sizeof(u64));
211 	memset(event->comm.comm + size, 0, machine->id_hdr_size);
212 	event->comm.header.size = (sizeof(event->comm) -
213 				(sizeof(event->comm.comm) - size) +
214 				machine->id_hdr_size);
215 	event->comm.tid = pid;
216 
217 	return 0;
218 }
219 
perf_event__synthesize_comm(struct perf_tool * tool,union perf_event * event,pid_t pid,perf_event__handler_t process,struct machine * machine)220 pid_t perf_event__synthesize_comm(struct perf_tool *tool,
221 					 union perf_event *event, pid_t pid,
222 					 perf_event__handler_t process,
223 					 struct machine *machine)
224 {
225 	pid_t tgid, ppid;
226 
227 	if (perf_event__prepare_comm(event, pid, machine, &tgid, &ppid) != 0)
228 		return -1;
229 
230 	if (perf_tool__process_synth_event(tool, event, machine, process) != 0)
231 		return -1;
232 
233 	return tgid;
234 }
235 
perf_event__get_ns_link_info(pid_t pid,const char * ns,struct perf_ns_link_info * ns_link_info)236 static void perf_event__get_ns_link_info(pid_t pid, const char *ns,
237 					 struct perf_ns_link_info *ns_link_info)
238 {
239 	struct stat64 st;
240 	char proc_ns[128];
241 
242 	sprintf(proc_ns, "/proc/%u/ns/%s", pid, ns);
243 	if (stat64(proc_ns, &st) == 0) {
244 		ns_link_info->dev = st.st_dev;
245 		ns_link_info->ino = st.st_ino;
246 	}
247 }
248 
perf_event__synthesize_namespaces(struct perf_tool * tool,union perf_event * event,pid_t pid,pid_t tgid,perf_event__handler_t process,struct machine * machine)249 int perf_event__synthesize_namespaces(struct perf_tool *tool,
250 				      union perf_event *event,
251 				      pid_t pid, pid_t tgid,
252 				      perf_event__handler_t process,
253 				      struct machine *machine)
254 {
255 	u32 idx;
256 	struct perf_ns_link_info *ns_link_info;
257 
258 	if (!tool || !tool->namespace_events)
259 		return 0;
260 
261 	memset(&event->namespaces, 0, (sizeof(event->namespaces) +
262 	       (NR_NAMESPACES * sizeof(struct perf_ns_link_info)) +
263 	       machine->id_hdr_size));
264 
265 	event->namespaces.pid = tgid;
266 	event->namespaces.tid = pid;
267 
268 	event->namespaces.nr_namespaces = NR_NAMESPACES;
269 
270 	ns_link_info = event->namespaces.link_info;
271 
272 	for (idx = 0; idx < event->namespaces.nr_namespaces; idx++)
273 		perf_event__get_ns_link_info(pid, perf_ns__name(idx),
274 					     &ns_link_info[idx]);
275 
276 	event->namespaces.header.type = PERF_RECORD_NAMESPACES;
277 
278 	event->namespaces.header.size = (sizeof(event->namespaces) +
279 			(NR_NAMESPACES * sizeof(struct perf_ns_link_info)) +
280 			machine->id_hdr_size);
281 
282 	if (perf_tool__process_synth_event(tool, event, machine, process) != 0)
283 		return -1;
284 
285 	return 0;
286 }
287 
perf_event__synthesize_fork(struct perf_tool * tool,union perf_event * event,pid_t pid,pid_t tgid,pid_t ppid,perf_event__handler_t process,struct machine * machine)288 static int perf_event__synthesize_fork(struct perf_tool *tool,
289 				       union perf_event *event,
290 				       pid_t pid, pid_t tgid, pid_t ppid,
291 				       perf_event__handler_t process,
292 				       struct machine *machine)
293 {
294 	memset(&event->fork, 0, sizeof(event->fork) + machine->id_hdr_size);
295 
296 	/*
297 	 * for main thread set parent to ppid from status file. For other
298 	 * threads set parent pid to main thread. ie., assume main thread
299 	 * spawns all threads in a process
300 	*/
301 	if (tgid == pid) {
302 		event->fork.ppid = ppid;
303 		event->fork.ptid = ppid;
304 	} else {
305 		event->fork.ppid = tgid;
306 		event->fork.ptid = tgid;
307 	}
308 	event->fork.pid  = tgid;
309 	event->fork.tid  = pid;
310 	event->fork.header.type = PERF_RECORD_FORK;
311 
312 	event->fork.header.size = (sizeof(event->fork) + machine->id_hdr_size);
313 
314 	if (perf_tool__process_synth_event(tool, event, machine, process) != 0)
315 		return -1;
316 
317 	return 0;
318 }
319 
perf_event__synthesize_mmap_events(struct perf_tool * tool,union perf_event * event,pid_t pid,pid_t tgid,perf_event__handler_t process,struct machine * machine,bool mmap_data,unsigned int proc_map_timeout)320 int perf_event__synthesize_mmap_events(struct perf_tool *tool,
321 				       union perf_event *event,
322 				       pid_t pid, pid_t tgid,
323 				       perf_event__handler_t process,
324 				       struct machine *machine,
325 				       bool mmap_data,
326 				       unsigned int proc_map_timeout)
327 {
328 	char filename[PATH_MAX];
329 	FILE *fp;
330 	unsigned long long t;
331 	bool truncation = false;
332 	unsigned long long timeout = proc_map_timeout * 1000000ULL;
333 	int rc = 0;
334 	const char *hugetlbfs_mnt = hugetlbfs__mountpoint();
335 	int hugetlbfs_mnt_len = hugetlbfs_mnt ? strlen(hugetlbfs_mnt) : 0;
336 
337 	if (machine__is_default_guest(machine))
338 		return 0;
339 
340 	snprintf(filename, sizeof(filename), "%s/proc/%d/task/%d/maps",
341 		 machine->root_dir, pid, pid);
342 
343 	fp = fopen(filename, "r");
344 	if (fp == NULL) {
345 		/*
346 		 * We raced with a task exiting - just return:
347 		 */
348 		pr_debug("couldn't open %s\n", filename);
349 		return -1;
350 	}
351 
352 	event->header.type = PERF_RECORD_MMAP2;
353 	t = rdclock();
354 
355 	while (1) {
356 		char bf[BUFSIZ];
357 		char prot[5];
358 		char execname[PATH_MAX];
359 		char anonstr[] = "//anon";
360 		unsigned int ino;
361 		size_t size;
362 		ssize_t n;
363 
364 		if (fgets(bf, sizeof(bf), fp) == NULL)
365 			break;
366 
367 		if ((rdclock() - t) > timeout) {
368 			pr_warning("Reading %s time out. "
369 				   "You may want to increase "
370 				   "the time limit by --proc-map-timeout\n",
371 				   filename);
372 			truncation = true;
373 			goto out;
374 		}
375 
376 		/* ensure null termination since stack will be reused. */
377 		strcpy(execname, "");
378 
379 		/* 00400000-0040c000 r-xp 00000000 fd:01 41038  /bin/cat */
380 		n = sscanf(bf, "%"PRIx64"-%"PRIx64" %s %"PRIx64" %x:%x %u %[^\n]\n",
381 		       &event->mmap2.start, &event->mmap2.len, prot,
382 		       &event->mmap2.pgoff, &event->mmap2.maj,
383 		       &event->mmap2.min,
384 		       &ino, execname);
385 
386 		/*
387  		 * Anon maps don't have the execname.
388  		 */
389 		if (n < 7)
390 			continue;
391 
392 		event->mmap2.ino = (u64)ino;
393 
394 		/*
395 		 * Just like the kernel, see __perf_event_mmap in kernel/perf_event.c
396 		 */
397 		if (machine__is_host(machine))
398 			event->header.misc = PERF_RECORD_MISC_USER;
399 		else
400 			event->header.misc = PERF_RECORD_MISC_GUEST_USER;
401 
402 		/* map protection and flags bits */
403 		event->mmap2.prot = 0;
404 		event->mmap2.flags = 0;
405 		if (prot[0] == 'r')
406 			event->mmap2.prot |= PROT_READ;
407 		if (prot[1] == 'w')
408 			event->mmap2.prot |= PROT_WRITE;
409 		if (prot[2] == 'x')
410 			event->mmap2.prot |= PROT_EXEC;
411 
412 		if (prot[3] == 's')
413 			event->mmap2.flags |= MAP_SHARED;
414 		else
415 			event->mmap2.flags |= MAP_PRIVATE;
416 
417 		if (prot[2] != 'x') {
418 			if (!mmap_data || prot[0] != 'r')
419 				continue;
420 
421 			event->header.misc |= PERF_RECORD_MISC_MMAP_DATA;
422 		}
423 
424 out:
425 		if (truncation)
426 			event->header.misc |= PERF_RECORD_MISC_PROC_MAP_PARSE_TIMEOUT;
427 
428 		if (!strcmp(execname, ""))
429 			strcpy(execname, anonstr);
430 
431 		if (hugetlbfs_mnt_len &&
432 		    !strncmp(execname, hugetlbfs_mnt, hugetlbfs_mnt_len)) {
433 			strcpy(execname, anonstr);
434 			event->mmap2.flags |= MAP_HUGETLB;
435 		}
436 
437 		size = strlen(execname) + 1;
438 		memcpy(event->mmap2.filename, execname, size);
439 		size = PERF_ALIGN(size, sizeof(u64));
440 		event->mmap2.len -= event->mmap.start;
441 		event->mmap2.header.size = (sizeof(event->mmap2) -
442 					(sizeof(event->mmap2.filename) - size));
443 		memset(event->mmap2.filename + size, 0, machine->id_hdr_size);
444 		event->mmap2.header.size += machine->id_hdr_size;
445 		event->mmap2.pid = tgid;
446 		event->mmap2.tid = pid;
447 
448 		if (perf_tool__process_synth_event(tool, event, machine, process) != 0) {
449 			rc = -1;
450 			break;
451 		}
452 
453 		if (truncation)
454 			break;
455 	}
456 
457 	fclose(fp);
458 	return rc;
459 }
460 
perf_event__synthesize_modules(struct perf_tool * tool,perf_event__handler_t process,struct machine * machine)461 int perf_event__synthesize_modules(struct perf_tool *tool,
462 				   perf_event__handler_t process,
463 				   struct machine *machine)
464 {
465 	int rc = 0;
466 	struct map *pos;
467 	struct maps *maps = machine__kernel_maps(machine);
468 	union perf_event *event = zalloc((sizeof(event->mmap) +
469 					  machine->id_hdr_size));
470 	if (event == NULL) {
471 		pr_debug("Not enough memory synthesizing mmap event "
472 			 "for kernel modules\n");
473 		return -1;
474 	}
475 
476 	event->header.type = PERF_RECORD_MMAP;
477 
478 	/*
479 	 * kernel uses 0 for user space maps, see kernel/perf_event.c
480 	 * __perf_event_mmap
481 	 */
482 	if (machine__is_host(machine))
483 		event->header.misc = PERF_RECORD_MISC_KERNEL;
484 	else
485 		event->header.misc = PERF_RECORD_MISC_GUEST_KERNEL;
486 
487 	for (pos = maps__first(maps); pos; pos = map__next(pos)) {
488 		size_t size;
489 
490 		if (!__map__is_kmodule(pos))
491 			continue;
492 
493 		size = PERF_ALIGN(pos->dso->long_name_len + 1, sizeof(u64));
494 		event->mmap.header.type = PERF_RECORD_MMAP;
495 		event->mmap.header.size = (sizeof(event->mmap) -
496 				        (sizeof(event->mmap.filename) - size));
497 		memset(event->mmap.filename + size, 0, machine->id_hdr_size);
498 		event->mmap.header.size += machine->id_hdr_size;
499 		event->mmap.start = pos->start;
500 		event->mmap.len   = pos->end - pos->start;
501 		event->mmap.pid   = machine->pid;
502 
503 		memcpy(event->mmap.filename, pos->dso->long_name,
504 		       pos->dso->long_name_len + 1);
505 		if (perf_tool__process_synth_event(tool, event, machine, process) != 0) {
506 			rc = -1;
507 			break;
508 		}
509 	}
510 
511 	free(event);
512 	return rc;
513 }
514 
__event__synthesize_thread(union perf_event * comm_event,union perf_event * mmap_event,union perf_event * fork_event,union perf_event * namespaces_event,pid_t pid,int full,perf_event__handler_t process,struct perf_tool * tool,struct machine * machine,bool mmap_data,unsigned int proc_map_timeout)515 static int __event__synthesize_thread(union perf_event *comm_event,
516 				      union perf_event *mmap_event,
517 				      union perf_event *fork_event,
518 				      union perf_event *namespaces_event,
519 				      pid_t pid, int full,
520 				      perf_event__handler_t process,
521 				      struct perf_tool *tool,
522 				      struct machine *machine,
523 				      bool mmap_data,
524 				      unsigned int proc_map_timeout)
525 {
526 	char filename[PATH_MAX];
527 	DIR *tasks;
528 	struct dirent *dirent;
529 	pid_t tgid, ppid;
530 	int rc = 0;
531 
532 	/* special case: only send one comm event using passed in pid */
533 	if (!full) {
534 		tgid = perf_event__synthesize_comm(tool, comm_event, pid,
535 						   process, machine);
536 
537 		if (tgid == -1)
538 			return -1;
539 
540 		if (perf_event__synthesize_namespaces(tool, namespaces_event, pid,
541 						      tgid, process, machine) < 0)
542 			return -1;
543 
544 		/*
545 		 * send mmap only for thread group leader
546 		 * see thread__init_map_groups
547 		 */
548 		if (pid == tgid &&
549 		    perf_event__synthesize_mmap_events(tool, mmap_event, pid, tgid,
550 						       process, machine, mmap_data,
551 						       proc_map_timeout))
552 			return -1;
553 
554 		return 0;
555 	}
556 
557 	if (machine__is_default_guest(machine))
558 		return 0;
559 
560 	snprintf(filename, sizeof(filename), "%s/proc/%d/task",
561 		 machine->root_dir, pid);
562 
563 	tasks = opendir(filename);
564 	if (tasks == NULL) {
565 		pr_debug("couldn't open %s\n", filename);
566 		return 0;
567 	}
568 
569 	while ((dirent = readdir(tasks)) != NULL) {
570 		char *end;
571 		pid_t _pid;
572 
573 		_pid = strtol(dirent->d_name, &end, 10);
574 		if (*end)
575 			continue;
576 
577 		rc = -1;
578 		if (perf_event__prepare_comm(comm_event, _pid, machine,
579 					     &tgid, &ppid) != 0)
580 			break;
581 
582 		if (perf_event__synthesize_fork(tool, fork_event, _pid, tgid,
583 						ppid, process, machine) < 0)
584 			break;
585 
586 		if (perf_event__synthesize_namespaces(tool, namespaces_event, _pid,
587 						      tgid, process, machine) < 0)
588 			break;
589 
590 		/*
591 		 * Send the prepared comm event
592 		 */
593 		if (perf_tool__process_synth_event(tool, comm_event, machine, process) != 0)
594 			break;
595 
596 		rc = 0;
597 		if (_pid == pid) {
598 			/* process the parent's maps too */
599 			rc = perf_event__synthesize_mmap_events(tool, mmap_event, pid, tgid,
600 						process, machine, mmap_data, proc_map_timeout);
601 			if (rc)
602 				break;
603 		}
604 	}
605 
606 	closedir(tasks);
607 	return rc;
608 }
609 
perf_event__synthesize_thread_map(struct perf_tool * tool,struct thread_map * threads,perf_event__handler_t process,struct machine * machine,bool mmap_data,unsigned int proc_map_timeout)610 int perf_event__synthesize_thread_map(struct perf_tool *tool,
611 				      struct thread_map *threads,
612 				      perf_event__handler_t process,
613 				      struct machine *machine,
614 				      bool mmap_data,
615 				      unsigned int proc_map_timeout)
616 {
617 	union perf_event *comm_event, *mmap_event, *fork_event;
618 	union perf_event *namespaces_event;
619 	int err = -1, thread, j;
620 
621 	comm_event = malloc(sizeof(comm_event->comm) + machine->id_hdr_size);
622 	if (comm_event == NULL)
623 		goto out;
624 
625 	mmap_event = malloc(sizeof(mmap_event->mmap2) + machine->id_hdr_size);
626 	if (mmap_event == NULL)
627 		goto out_free_comm;
628 
629 	fork_event = malloc(sizeof(fork_event->fork) + machine->id_hdr_size);
630 	if (fork_event == NULL)
631 		goto out_free_mmap;
632 
633 	namespaces_event = malloc(sizeof(namespaces_event->namespaces) +
634 				  (NR_NAMESPACES * sizeof(struct perf_ns_link_info)) +
635 				  machine->id_hdr_size);
636 	if (namespaces_event == NULL)
637 		goto out_free_fork;
638 
639 	err = 0;
640 	for (thread = 0; thread < threads->nr; ++thread) {
641 		if (__event__synthesize_thread(comm_event, mmap_event,
642 					       fork_event, namespaces_event,
643 					       thread_map__pid(threads, thread), 0,
644 					       process, tool, machine,
645 					       mmap_data, proc_map_timeout)) {
646 			err = -1;
647 			break;
648 		}
649 
650 		/*
651 		 * comm.pid is set to thread group id by
652 		 * perf_event__synthesize_comm
653 		 */
654 		if ((int) comm_event->comm.pid != thread_map__pid(threads, thread)) {
655 			bool need_leader = true;
656 
657 			/* is thread group leader in thread_map? */
658 			for (j = 0; j < threads->nr; ++j) {
659 				if ((int) comm_event->comm.pid == thread_map__pid(threads, j)) {
660 					need_leader = false;
661 					break;
662 				}
663 			}
664 
665 			/* if not, generate events for it */
666 			if (need_leader &&
667 			    __event__synthesize_thread(comm_event, mmap_event,
668 						       fork_event, namespaces_event,
669 						       comm_event->comm.pid, 0,
670 						       process, tool, machine,
671 						       mmap_data, proc_map_timeout)) {
672 				err = -1;
673 				break;
674 			}
675 		}
676 	}
677 	free(namespaces_event);
678 out_free_fork:
679 	free(fork_event);
680 out_free_mmap:
681 	free(mmap_event);
682 out_free_comm:
683 	free(comm_event);
684 out:
685 	return err;
686 }
687 
__perf_event__synthesize_threads(struct perf_tool * tool,perf_event__handler_t process,struct machine * machine,bool mmap_data,unsigned int proc_map_timeout,struct dirent ** dirent,int start,int num)688 static int __perf_event__synthesize_threads(struct perf_tool *tool,
689 					    perf_event__handler_t process,
690 					    struct machine *machine,
691 					    bool mmap_data,
692 					    unsigned int proc_map_timeout,
693 					    struct dirent **dirent,
694 					    int start,
695 					    int num)
696 {
697 	union perf_event *comm_event, *mmap_event, *fork_event;
698 	union perf_event *namespaces_event;
699 	int err = -1;
700 	char *end;
701 	pid_t pid;
702 	int i;
703 
704 	comm_event = malloc(sizeof(comm_event->comm) + machine->id_hdr_size);
705 	if (comm_event == NULL)
706 		goto out;
707 
708 	mmap_event = malloc(sizeof(mmap_event->mmap2) + machine->id_hdr_size);
709 	if (mmap_event == NULL)
710 		goto out_free_comm;
711 
712 	fork_event = malloc(sizeof(fork_event->fork) + machine->id_hdr_size);
713 	if (fork_event == NULL)
714 		goto out_free_mmap;
715 
716 	namespaces_event = malloc(sizeof(namespaces_event->namespaces) +
717 				  (NR_NAMESPACES * sizeof(struct perf_ns_link_info)) +
718 				  machine->id_hdr_size);
719 	if (namespaces_event == NULL)
720 		goto out_free_fork;
721 
722 	for (i = start; i < start + num; i++) {
723 		if (!isdigit(dirent[i]->d_name[0]))
724 			continue;
725 
726 		pid = (pid_t)strtol(dirent[i]->d_name, &end, 10);
727 		/* only interested in proper numerical dirents */
728 		if (*end)
729 			continue;
730 		/*
731 		 * We may race with exiting thread, so don't stop just because
732 		 * one thread couldn't be synthesized.
733 		 */
734 		__event__synthesize_thread(comm_event, mmap_event, fork_event,
735 					   namespaces_event, pid, 1, process,
736 					   tool, machine, mmap_data,
737 					   proc_map_timeout);
738 	}
739 	err = 0;
740 
741 	free(namespaces_event);
742 out_free_fork:
743 	free(fork_event);
744 out_free_mmap:
745 	free(mmap_event);
746 out_free_comm:
747 	free(comm_event);
748 out:
749 	return err;
750 }
751 
752 struct synthesize_threads_arg {
753 	struct perf_tool *tool;
754 	perf_event__handler_t process;
755 	struct machine *machine;
756 	bool mmap_data;
757 	unsigned int proc_map_timeout;
758 	struct dirent **dirent;
759 	int num;
760 	int start;
761 };
762 
synthesize_threads_worker(void * arg)763 static void *synthesize_threads_worker(void *arg)
764 {
765 	struct synthesize_threads_arg *args = arg;
766 
767 	__perf_event__synthesize_threads(args->tool, args->process,
768 					 args->machine, args->mmap_data,
769 					 args->proc_map_timeout, args->dirent,
770 					 args->start, args->num);
771 	return NULL;
772 }
773 
perf_event__synthesize_threads(struct perf_tool * tool,perf_event__handler_t process,struct machine * machine,bool mmap_data,unsigned int proc_map_timeout,unsigned int nr_threads_synthesize)774 int perf_event__synthesize_threads(struct perf_tool *tool,
775 				   perf_event__handler_t process,
776 				   struct machine *machine,
777 				   bool mmap_data,
778 				   unsigned int proc_map_timeout,
779 				   unsigned int nr_threads_synthesize)
780 {
781 	struct synthesize_threads_arg *args = NULL;
782 	pthread_t *synthesize_threads = NULL;
783 	char proc_path[PATH_MAX];
784 	struct dirent **dirent;
785 	int num_per_thread;
786 	int m, n, i, j;
787 	int thread_nr;
788 	int base = 0;
789 	int err = -1;
790 
791 
792 	if (machine__is_default_guest(machine))
793 		return 0;
794 
795 	snprintf(proc_path, sizeof(proc_path), "%s/proc", machine->root_dir);
796 	n = scandir(proc_path, &dirent, 0, alphasort);
797 	if (n < 0)
798 		return err;
799 
800 	if (nr_threads_synthesize == UINT_MAX)
801 		thread_nr = sysconf(_SC_NPROCESSORS_ONLN);
802 	else
803 		thread_nr = nr_threads_synthesize;
804 
805 	if (thread_nr <= 1) {
806 		err = __perf_event__synthesize_threads(tool, process,
807 						       machine, mmap_data,
808 						       proc_map_timeout,
809 						       dirent, base, n);
810 		goto free_dirent;
811 	}
812 	if (thread_nr > n)
813 		thread_nr = n;
814 
815 	synthesize_threads = calloc(sizeof(pthread_t), thread_nr);
816 	if (synthesize_threads == NULL)
817 		goto free_dirent;
818 
819 	args = calloc(sizeof(*args), thread_nr);
820 	if (args == NULL)
821 		goto free_threads;
822 
823 	num_per_thread = n / thread_nr;
824 	m = n % thread_nr;
825 	for (i = 0; i < thread_nr; i++) {
826 		args[i].tool = tool;
827 		args[i].process = process;
828 		args[i].machine = machine;
829 		args[i].mmap_data = mmap_data;
830 		args[i].proc_map_timeout = proc_map_timeout;
831 		args[i].dirent = dirent;
832 	}
833 	for (i = 0; i < m; i++) {
834 		args[i].num = num_per_thread + 1;
835 		args[i].start = i * args[i].num;
836 	}
837 	if (i != 0)
838 		base = args[i-1].start + args[i-1].num;
839 	for (j = i; j < thread_nr; j++) {
840 		args[j].num = num_per_thread;
841 		args[j].start = base + (j - i) * args[i].num;
842 	}
843 
844 	for (i = 0; i < thread_nr; i++) {
845 		if (pthread_create(&synthesize_threads[i], NULL,
846 				   synthesize_threads_worker, &args[i]))
847 			goto out_join;
848 	}
849 	err = 0;
850 out_join:
851 	for (i = 0; i < thread_nr; i++)
852 		pthread_join(synthesize_threads[i], NULL);
853 	free(args);
854 free_threads:
855 	free(synthesize_threads);
856 free_dirent:
857 	for (i = 0; i < n; i++)
858 		free(dirent[i]);
859 	free(dirent);
860 
861 	return err;
862 }
863 
864 struct process_symbol_args {
865 	const char *name;
866 	u64	   start;
867 };
868 
find_symbol_cb(void * arg,const char * name,char type,u64 start)869 static int find_symbol_cb(void *arg, const char *name, char type,
870 			  u64 start)
871 {
872 	struct process_symbol_args *args = arg;
873 
874 	/*
875 	 * Must be a function or at least an alias, as in PARISC64, where "_text" is
876 	 * an 'A' to the same address as "_stext".
877 	 */
878 	if (!(kallsyms__is_function(type) ||
879 	      type == 'A') || strcmp(name, args->name))
880 		return 0;
881 
882 	args->start = start;
883 	return 1;
884 }
885 
kallsyms__get_function_start(const char * kallsyms_filename,const char * symbol_name,u64 * addr)886 int kallsyms__get_function_start(const char *kallsyms_filename,
887 				 const char *symbol_name, u64 *addr)
888 {
889 	struct process_symbol_args args = { .name = symbol_name, };
890 
891 	if (kallsyms__parse(kallsyms_filename, &args, find_symbol_cb) <= 0)
892 		return -1;
893 
894 	*addr = args.start;
895 	return 0;
896 }
897 
perf_event__synthesize_extra_kmaps(struct perf_tool * tool __maybe_unused,perf_event__handler_t process __maybe_unused,struct machine * machine __maybe_unused)898 int __weak perf_event__synthesize_extra_kmaps(struct perf_tool *tool __maybe_unused,
899 					      perf_event__handler_t process __maybe_unused,
900 					      struct machine *machine __maybe_unused)
901 {
902 	return 0;
903 }
904 
__perf_event__synthesize_kernel_mmap(struct perf_tool * tool,perf_event__handler_t process,struct machine * machine)905 static int __perf_event__synthesize_kernel_mmap(struct perf_tool *tool,
906 						perf_event__handler_t process,
907 						struct machine *machine)
908 {
909 	size_t size;
910 	struct map *map = machine__kernel_map(machine);
911 	struct kmap *kmap;
912 	int err;
913 	union perf_event *event;
914 
915 	if (symbol_conf.kptr_restrict)
916 		return -1;
917 	if (map == NULL)
918 		return -1;
919 
920 	/*
921 	 * We should get this from /sys/kernel/sections/.text, but till that is
922 	 * available use this, and after it is use this as a fallback for older
923 	 * kernels.
924 	 */
925 	event = zalloc((sizeof(event->mmap) + machine->id_hdr_size));
926 	if (event == NULL) {
927 		pr_debug("Not enough memory synthesizing mmap event "
928 			 "for kernel modules\n");
929 		return -1;
930 	}
931 
932 	if (machine__is_host(machine)) {
933 		/*
934 		 * kernel uses PERF_RECORD_MISC_USER for user space maps,
935 		 * see kernel/perf_event.c __perf_event_mmap
936 		 */
937 		event->header.misc = PERF_RECORD_MISC_KERNEL;
938 	} else {
939 		event->header.misc = PERF_RECORD_MISC_GUEST_KERNEL;
940 	}
941 
942 	kmap = map__kmap(map);
943 	size = snprintf(event->mmap.filename, sizeof(event->mmap.filename),
944 			"%s%s", machine->mmap_name, kmap->ref_reloc_sym->name) + 1;
945 	size = PERF_ALIGN(size, sizeof(u64));
946 	event->mmap.header.type = PERF_RECORD_MMAP;
947 	event->mmap.header.size = (sizeof(event->mmap) -
948 			(sizeof(event->mmap.filename) - size) + machine->id_hdr_size);
949 	event->mmap.pgoff = kmap->ref_reloc_sym->addr;
950 	event->mmap.start = map->start;
951 	event->mmap.len   = map->end - event->mmap.start;
952 	event->mmap.pid   = machine->pid;
953 
954 	err = perf_tool__process_synth_event(tool, event, machine, process);
955 	free(event);
956 
957 	return err;
958 }
959 
perf_event__synthesize_kernel_mmap(struct perf_tool * tool,perf_event__handler_t process,struct machine * machine)960 int perf_event__synthesize_kernel_mmap(struct perf_tool *tool,
961 				       perf_event__handler_t process,
962 				       struct machine *machine)
963 {
964 	int err;
965 
966 	err = __perf_event__synthesize_kernel_mmap(tool, process, machine);
967 	if (err < 0)
968 		return err;
969 
970 	return perf_event__synthesize_extra_kmaps(tool, process, machine);
971 }
972 
perf_event__synthesize_thread_map2(struct perf_tool * tool,struct thread_map * threads,perf_event__handler_t process,struct machine * machine)973 int perf_event__synthesize_thread_map2(struct perf_tool *tool,
974 				      struct thread_map *threads,
975 				      perf_event__handler_t process,
976 				      struct machine *machine)
977 {
978 	union perf_event *event;
979 	int i, err, size;
980 
981 	size  = sizeof(event->thread_map);
982 	size +=	threads->nr * sizeof(event->thread_map.entries[0]);
983 
984 	event = zalloc(size);
985 	if (!event)
986 		return -ENOMEM;
987 
988 	event->header.type = PERF_RECORD_THREAD_MAP;
989 	event->header.size = size;
990 	event->thread_map.nr = threads->nr;
991 
992 	for (i = 0; i < threads->nr; i++) {
993 		struct thread_map_event_entry *entry = &event->thread_map.entries[i];
994 		char *comm = thread_map__comm(threads, i);
995 
996 		if (!comm)
997 			comm = (char *) "";
998 
999 		entry->pid = thread_map__pid(threads, i);
1000 		strncpy((char *) &entry->comm, comm, sizeof(entry->comm));
1001 	}
1002 
1003 	err = process(tool, event, NULL, machine);
1004 
1005 	free(event);
1006 	return err;
1007 }
1008 
synthesize_cpus(struct cpu_map_entries * cpus,struct cpu_map * map)1009 static void synthesize_cpus(struct cpu_map_entries *cpus,
1010 			    struct cpu_map *map)
1011 {
1012 	int i;
1013 
1014 	cpus->nr = map->nr;
1015 
1016 	for (i = 0; i < map->nr; i++)
1017 		cpus->cpu[i] = map->map[i];
1018 }
1019 
synthesize_mask(struct cpu_map_mask * mask,struct cpu_map * map,int max)1020 static void synthesize_mask(struct cpu_map_mask *mask,
1021 			    struct cpu_map *map, int max)
1022 {
1023 	int i;
1024 
1025 	mask->nr = BITS_TO_LONGS(max);
1026 	mask->long_size = sizeof(long);
1027 
1028 	for (i = 0; i < map->nr; i++)
1029 		set_bit(map->map[i], mask->mask);
1030 }
1031 
cpus_size(struct cpu_map * map)1032 static size_t cpus_size(struct cpu_map *map)
1033 {
1034 	return sizeof(struct cpu_map_entries) + map->nr * sizeof(u16);
1035 }
1036 
mask_size(struct cpu_map * map,int * max)1037 static size_t mask_size(struct cpu_map *map, int *max)
1038 {
1039 	int i;
1040 
1041 	*max = 0;
1042 
1043 	for (i = 0; i < map->nr; i++) {
1044 		/* bit possition of the cpu is + 1 */
1045 		int bit = map->map[i] + 1;
1046 
1047 		if (bit > *max)
1048 			*max = bit;
1049 	}
1050 
1051 	return sizeof(struct cpu_map_mask) + BITS_TO_LONGS(*max) * sizeof(long);
1052 }
1053 
cpu_map_data__alloc(struct cpu_map * map,size_t * size,u16 * type,int * max)1054 void *cpu_map_data__alloc(struct cpu_map *map, size_t *size, u16 *type, int *max)
1055 {
1056 	size_t size_cpus, size_mask;
1057 	bool is_dummy = cpu_map__empty(map);
1058 
1059 	/*
1060 	 * Both array and mask data have variable size based
1061 	 * on the number of cpus and their actual values.
1062 	 * The size of the 'struct cpu_map_data' is:
1063 	 *
1064 	 *   array = size of 'struct cpu_map_entries' +
1065 	 *           number of cpus * sizeof(u64)
1066 	 *
1067 	 *   mask  = size of 'struct cpu_map_mask' +
1068 	 *           maximum cpu bit converted to size of longs
1069 	 *
1070 	 * and finaly + the size of 'struct cpu_map_data'.
1071 	 */
1072 	size_cpus = cpus_size(map);
1073 	size_mask = mask_size(map, max);
1074 
1075 	if (is_dummy || (size_cpus < size_mask)) {
1076 		*size += size_cpus;
1077 		*type  = PERF_CPU_MAP__CPUS;
1078 	} else {
1079 		*size += size_mask;
1080 		*type  = PERF_CPU_MAP__MASK;
1081 	}
1082 
1083 	*size += sizeof(struct cpu_map_data);
1084 	*size = PERF_ALIGN(*size, sizeof(u64));
1085 	return zalloc(*size);
1086 }
1087 
cpu_map_data__synthesize(struct cpu_map_data * data,struct cpu_map * map,u16 type,int max)1088 void cpu_map_data__synthesize(struct cpu_map_data *data, struct cpu_map *map,
1089 			      u16 type, int max)
1090 {
1091 	data->type = type;
1092 
1093 	switch (type) {
1094 	case PERF_CPU_MAP__CPUS:
1095 		synthesize_cpus((struct cpu_map_entries *) data->data, map);
1096 		break;
1097 	case PERF_CPU_MAP__MASK:
1098 		synthesize_mask((struct cpu_map_mask *) data->data, map, max);
1099 	default:
1100 		break;
1101 	};
1102 }
1103 
cpu_map_event__new(struct cpu_map * map)1104 static struct cpu_map_event* cpu_map_event__new(struct cpu_map *map)
1105 {
1106 	size_t size = sizeof(struct cpu_map_event);
1107 	struct cpu_map_event *event;
1108 	int max;
1109 	u16 type;
1110 
1111 	event = cpu_map_data__alloc(map, &size, &type, &max);
1112 	if (!event)
1113 		return NULL;
1114 
1115 	event->header.type = PERF_RECORD_CPU_MAP;
1116 	event->header.size = size;
1117 	event->data.type   = type;
1118 
1119 	cpu_map_data__synthesize(&event->data, map, type, max);
1120 	return event;
1121 }
1122 
perf_event__synthesize_cpu_map(struct perf_tool * tool,struct cpu_map * map,perf_event__handler_t process,struct machine * machine)1123 int perf_event__synthesize_cpu_map(struct perf_tool *tool,
1124 				   struct cpu_map *map,
1125 				   perf_event__handler_t process,
1126 				   struct machine *machine)
1127 {
1128 	struct cpu_map_event *event;
1129 	int err;
1130 
1131 	event = cpu_map_event__new(map);
1132 	if (!event)
1133 		return -ENOMEM;
1134 
1135 	err = process(tool, (union perf_event *) event, NULL, machine);
1136 
1137 	free(event);
1138 	return err;
1139 }
1140 
perf_event__synthesize_stat_config(struct perf_tool * tool,struct perf_stat_config * config,perf_event__handler_t process,struct machine * machine)1141 int perf_event__synthesize_stat_config(struct perf_tool *tool,
1142 				       struct perf_stat_config *config,
1143 				       perf_event__handler_t process,
1144 				       struct machine *machine)
1145 {
1146 	struct stat_config_event *event;
1147 	int size, i = 0, err;
1148 
1149 	size  = sizeof(*event);
1150 	size += (PERF_STAT_CONFIG_TERM__MAX * sizeof(event->data[0]));
1151 
1152 	event = zalloc(size);
1153 	if (!event)
1154 		return -ENOMEM;
1155 
1156 	event->header.type = PERF_RECORD_STAT_CONFIG;
1157 	event->header.size = size;
1158 	event->nr          = PERF_STAT_CONFIG_TERM__MAX;
1159 
1160 #define ADD(__term, __val)					\
1161 	event->data[i].tag = PERF_STAT_CONFIG_TERM__##__term;	\
1162 	event->data[i].val = __val;				\
1163 	i++;
1164 
1165 	ADD(AGGR_MODE,	config->aggr_mode)
1166 	ADD(INTERVAL,	config->interval)
1167 	ADD(SCALE,	config->scale)
1168 
1169 	WARN_ONCE(i != PERF_STAT_CONFIG_TERM__MAX,
1170 		  "stat config terms unbalanced\n");
1171 #undef ADD
1172 
1173 	err = process(tool, (union perf_event *) event, NULL, machine);
1174 
1175 	free(event);
1176 	return err;
1177 }
1178 
perf_event__synthesize_stat(struct perf_tool * tool,u32 cpu,u32 thread,u64 id,struct perf_counts_values * count,perf_event__handler_t process,struct machine * machine)1179 int perf_event__synthesize_stat(struct perf_tool *tool,
1180 				u32 cpu, u32 thread, u64 id,
1181 				struct perf_counts_values *count,
1182 				perf_event__handler_t process,
1183 				struct machine *machine)
1184 {
1185 	struct stat_event event;
1186 
1187 	event.header.type = PERF_RECORD_STAT;
1188 	event.header.size = sizeof(event);
1189 	event.header.misc = 0;
1190 
1191 	event.id        = id;
1192 	event.cpu       = cpu;
1193 	event.thread    = thread;
1194 	event.val       = count->val;
1195 	event.ena       = count->ena;
1196 	event.run       = count->run;
1197 
1198 	return process(tool, (union perf_event *) &event, NULL, machine);
1199 }
1200 
perf_event__synthesize_stat_round(struct perf_tool * tool,u64 evtime,u64 type,perf_event__handler_t process,struct machine * machine)1201 int perf_event__synthesize_stat_round(struct perf_tool *tool,
1202 				      u64 evtime, u64 type,
1203 				      perf_event__handler_t process,
1204 				      struct machine *machine)
1205 {
1206 	struct stat_round_event event;
1207 
1208 	event.header.type = PERF_RECORD_STAT_ROUND;
1209 	event.header.size = sizeof(event);
1210 	event.header.misc = 0;
1211 
1212 	event.time = evtime;
1213 	event.type = type;
1214 
1215 	return process(tool, (union perf_event *) &event, NULL, machine);
1216 }
1217 
perf_event__read_stat_config(struct perf_stat_config * config,struct stat_config_event * event)1218 void perf_event__read_stat_config(struct perf_stat_config *config,
1219 				  struct stat_config_event *event)
1220 {
1221 	unsigned i;
1222 
1223 	for (i = 0; i < event->nr; i++) {
1224 
1225 		switch (event->data[i].tag) {
1226 #define CASE(__term, __val)					\
1227 		case PERF_STAT_CONFIG_TERM__##__term:		\
1228 			config->__val = event->data[i].val;	\
1229 			break;
1230 
1231 		CASE(AGGR_MODE, aggr_mode)
1232 		CASE(SCALE,     scale)
1233 		CASE(INTERVAL,  interval)
1234 #undef CASE
1235 		default:
1236 			pr_warning("unknown stat config term %" PRIu64 "\n",
1237 				   event->data[i].tag);
1238 		}
1239 	}
1240 }
1241 
perf_event__fprintf_comm(union perf_event * event,FILE * fp)1242 size_t perf_event__fprintf_comm(union perf_event *event, FILE *fp)
1243 {
1244 	const char *s;
1245 
1246 	if (event->header.misc & PERF_RECORD_MISC_COMM_EXEC)
1247 		s = " exec";
1248 	else
1249 		s = "";
1250 
1251 	return fprintf(fp, "%s: %s:%d/%d\n", s, event->comm.comm, event->comm.pid, event->comm.tid);
1252 }
1253 
perf_event__fprintf_namespaces(union perf_event * event,FILE * fp)1254 size_t perf_event__fprintf_namespaces(union perf_event *event, FILE *fp)
1255 {
1256 	size_t ret = 0;
1257 	struct perf_ns_link_info *ns_link_info;
1258 	u32 nr_namespaces, idx;
1259 
1260 	ns_link_info = event->namespaces.link_info;
1261 	nr_namespaces = event->namespaces.nr_namespaces;
1262 
1263 	ret += fprintf(fp, " %d/%d - nr_namespaces: %u\n\t\t[",
1264 		       event->namespaces.pid,
1265 		       event->namespaces.tid,
1266 		       nr_namespaces);
1267 
1268 	for (idx = 0; idx < nr_namespaces; idx++) {
1269 		if (idx && (idx % 4 == 0))
1270 			ret += fprintf(fp, "\n\t\t ");
1271 
1272 		ret  += fprintf(fp, "%u/%s: %" PRIu64 "/%#" PRIx64 "%s", idx,
1273 				perf_ns__name(idx), (u64)ns_link_info[idx].dev,
1274 				(u64)ns_link_info[idx].ino,
1275 				((idx + 1) != nr_namespaces) ? ", " : "]\n");
1276 	}
1277 
1278 	return ret;
1279 }
1280 
perf_event__process_comm(struct perf_tool * tool __maybe_unused,union perf_event * event,struct perf_sample * sample,struct machine * machine)1281 int perf_event__process_comm(struct perf_tool *tool __maybe_unused,
1282 			     union perf_event *event,
1283 			     struct perf_sample *sample,
1284 			     struct machine *machine)
1285 {
1286 	return machine__process_comm_event(machine, event, sample);
1287 }
1288 
perf_event__process_namespaces(struct perf_tool * tool __maybe_unused,union perf_event * event,struct perf_sample * sample,struct machine * machine)1289 int perf_event__process_namespaces(struct perf_tool *tool __maybe_unused,
1290 				   union perf_event *event,
1291 				   struct perf_sample *sample,
1292 				   struct machine *machine)
1293 {
1294 	return machine__process_namespaces_event(machine, event, sample);
1295 }
1296 
perf_event__process_lost(struct perf_tool * tool __maybe_unused,union perf_event * event,struct perf_sample * sample,struct machine * machine)1297 int perf_event__process_lost(struct perf_tool *tool __maybe_unused,
1298 			     union perf_event *event,
1299 			     struct perf_sample *sample,
1300 			     struct machine *machine)
1301 {
1302 	return machine__process_lost_event(machine, event, sample);
1303 }
1304 
perf_event__process_aux(struct perf_tool * tool __maybe_unused,union perf_event * event,struct perf_sample * sample __maybe_unused,struct machine * machine)1305 int perf_event__process_aux(struct perf_tool *tool __maybe_unused,
1306 			    union perf_event *event,
1307 			    struct perf_sample *sample __maybe_unused,
1308 			    struct machine *machine)
1309 {
1310 	return machine__process_aux_event(machine, event);
1311 }
1312 
perf_event__process_itrace_start(struct perf_tool * tool __maybe_unused,union perf_event * event,struct perf_sample * sample __maybe_unused,struct machine * machine)1313 int perf_event__process_itrace_start(struct perf_tool *tool __maybe_unused,
1314 				     union perf_event *event,
1315 				     struct perf_sample *sample __maybe_unused,
1316 				     struct machine *machine)
1317 {
1318 	return machine__process_itrace_start_event(machine, event);
1319 }
1320 
perf_event__process_lost_samples(struct perf_tool * tool __maybe_unused,union perf_event * event,struct perf_sample * sample,struct machine * machine)1321 int perf_event__process_lost_samples(struct perf_tool *tool __maybe_unused,
1322 				     union perf_event *event,
1323 				     struct perf_sample *sample,
1324 				     struct machine *machine)
1325 {
1326 	return machine__process_lost_samples_event(machine, event, sample);
1327 }
1328 
perf_event__process_switch(struct perf_tool * tool __maybe_unused,union perf_event * event,struct perf_sample * sample __maybe_unused,struct machine * machine)1329 int perf_event__process_switch(struct perf_tool *tool __maybe_unused,
1330 			       union perf_event *event,
1331 			       struct perf_sample *sample __maybe_unused,
1332 			       struct machine *machine)
1333 {
1334 	return machine__process_switch_event(machine, event);
1335 }
1336 
perf_event__fprintf_mmap(union perf_event * event,FILE * fp)1337 size_t perf_event__fprintf_mmap(union perf_event *event, FILE *fp)
1338 {
1339 	return fprintf(fp, " %d/%d: [%#" PRIx64 "(%#" PRIx64 ") @ %#" PRIx64 "]: %c %s\n",
1340 		       event->mmap.pid, event->mmap.tid, event->mmap.start,
1341 		       event->mmap.len, event->mmap.pgoff,
1342 		       (event->header.misc & PERF_RECORD_MISC_MMAP_DATA) ? 'r' : 'x',
1343 		       event->mmap.filename);
1344 }
1345 
perf_event__fprintf_mmap2(union perf_event * event,FILE * fp)1346 size_t perf_event__fprintf_mmap2(union perf_event *event, FILE *fp)
1347 {
1348 	return fprintf(fp, " %d/%d: [%#" PRIx64 "(%#" PRIx64 ") @ %#" PRIx64
1349 			   " %02x:%02x %"PRIu64" %"PRIu64"]: %c%c%c%c %s\n",
1350 		       event->mmap2.pid, event->mmap2.tid, event->mmap2.start,
1351 		       event->mmap2.len, event->mmap2.pgoff, event->mmap2.maj,
1352 		       event->mmap2.min, event->mmap2.ino,
1353 		       event->mmap2.ino_generation,
1354 		       (event->mmap2.prot & PROT_READ) ? 'r' : '-',
1355 		       (event->mmap2.prot & PROT_WRITE) ? 'w' : '-',
1356 		       (event->mmap2.prot & PROT_EXEC) ? 'x' : '-',
1357 		       (event->mmap2.flags & MAP_SHARED) ? 's' : 'p',
1358 		       event->mmap2.filename);
1359 }
1360 
perf_event__fprintf_thread_map(union perf_event * event,FILE * fp)1361 size_t perf_event__fprintf_thread_map(union perf_event *event, FILE *fp)
1362 {
1363 	struct thread_map *threads = thread_map__new_event(&event->thread_map);
1364 	size_t ret;
1365 
1366 	ret = fprintf(fp, " nr: ");
1367 
1368 	if (threads)
1369 		ret += thread_map__fprintf(threads, fp);
1370 	else
1371 		ret += fprintf(fp, "failed to get threads from event\n");
1372 
1373 	thread_map__put(threads);
1374 	return ret;
1375 }
1376 
perf_event__fprintf_cpu_map(union perf_event * event,FILE * fp)1377 size_t perf_event__fprintf_cpu_map(union perf_event *event, FILE *fp)
1378 {
1379 	struct cpu_map *cpus = cpu_map__new_data(&event->cpu_map.data);
1380 	size_t ret;
1381 
1382 	ret = fprintf(fp, ": ");
1383 
1384 	if (cpus)
1385 		ret += cpu_map__fprintf(cpus, fp);
1386 	else
1387 		ret += fprintf(fp, "failed to get cpumap from event\n");
1388 
1389 	cpu_map__put(cpus);
1390 	return ret;
1391 }
1392 
perf_event__process_mmap(struct perf_tool * tool __maybe_unused,union perf_event * event,struct perf_sample * sample,struct machine * machine)1393 int perf_event__process_mmap(struct perf_tool *tool __maybe_unused,
1394 			     union perf_event *event,
1395 			     struct perf_sample *sample,
1396 			     struct machine *machine)
1397 {
1398 	return machine__process_mmap_event(machine, event, sample);
1399 }
1400 
perf_event__process_mmap2(struct perf_tool * tool __maybe_unused,union perf_event * event,struct perf_sample * sample,struct machine * machine)1401 int perf_event__process_mmap2(struct perf_tool *tool __maybe_unused,
1402 			     union perf_event *event,
1403 			     struct perf_sample *sample,
1404 			     struct machine *machine)
1405 {
1406 	return machine__process_mmap2_event(machine, event, sample);
1407 }
1408 
perf_event__fprintf_task(union perf_event * event,FILE * fp)1409 size_t perf_event__fprintf_task(union perf_event *event, FILE *fp)
1410 {
1411 	return fprintf(fp, "(%d:%d):(%d:%d)\n",
1412 		       event->fork.pid, event->fork.tid,
1413 		       event->fork.ppid, event->fork.ptid);
1414 }
1415 
perf_event__process_fork(struct perf_tool * tool __maybe_unused,union perf_event * event,struct perf_sample * sample,struct machine * machine)1416 int perf_event__process_fork(struct perf_tool *tool __maybe_unused,
1417 			     union perf_event *event,
1418 			     struct perf_sample *sample,
1419 			     struct machine *machine)
1420 {
1421 	return machine__process_fork_event(machine, event, sample);
1422 }
1423 
perf_event__process_exit(struct perf_tool * tool __maybe_unused,union perf_event * event,struct perf_sample * sample,struct machine * machine)1424 int perf_event__process_exit(struct perf_tool *tool __maybe_unused,
1425 			     union perf_event *event,
1426 			     struct perf_sample *sample,
1427 			     struct machine *machine)
1428 {
1429 	return machine__process_exit_event(machine, event, sample);
1430 }
1431 
perf_event__fprintf_aux(union perf_event * event,FILE * fp)1432 size_t perf_event__fprintf_aux(union perf_event *event, FILE *fp)
1433 {
1434 	return fprintf(fp, " offset: %#"PRIx64" size: %#"PRIx64" flags: %#"PRIx64" [%s%s%s]\n",
1435 		       event->aux.aux_offset, event->aux.aux_size,
1436 		       event->aux.flags,
1437 		       event->aux.flags & PERF_AUX_FLAG_TRUNCATED ? "T" : "",
1438 		       event->aux.flags & PERF_AUX_FLAG_OVERWRITE ? "O" : "",
1439 		       event->aux.flags & PERF_AUX_FLAG_PARTIAL   ? "P" : "");
1440 }
1441 
perf_event__fprintf_itrace_start(union perf_event * event,FILE * fp)1442 size_t perf_event__fprintf_itrace_start(union perf_event *event, FILE *fp)
1443 {
1444 	return fprintf(fp, " pid: %u tid: %u\n",
1445 		       event->itrace_start.pid, event->itrace_start.tid);
1446 }
1447 
perf_event__fprintf_switch(union perf_event * event,FILE * fp)1448 size_t perf_event__fprintf_switch(union perf_event *event, FILE *fp)
1449 {
1450 	bool out = event->header.misc & PERF_RECORD_MISC_SWITCH_OUT;
1451 	const char *in_out = !out ? "IN         " :
1452 		!(event->header.misc & PERF_RECORD_MISC_SWITCH_OUT_PREEMPT) ?
1453 				    "OUT        " : "OUT preempt";
1454 
1455 	if (event->header.type == PERF_RECORD_SWITCH)
1456 		return fprintf(fp, " %s\n", in_out);
1457 
1458 	return fprintf(fp, " %s  %s pid/tid: %5u/%-5u\n",
1459 		       in_out, out ? "next" : "prev",
1460 		       event->context_switch.next_prev_pid,
1461 		       event->context_switch.next_prev_tid);
1462 }
1463 
perf_event__fprintf_lost(union perf_event * event,FILE * fp)1464 static size_t perf_event__fprintf_lost(union perf_event *event, FILE *fp)
1465 {
1466 	return fprintf(fp, " lost %" PRIu64 "\n", event->lost.lost);
1467 }
1468 
perf_event__fprintf(union perf_event * event,FILE * fp)1469 size_t perf_event__fprintf(union perf_event *event, FILE *fp)
1470 {
1471 	size_t ret = fprintf(fp, "PERF_RECORD_%s",
1472 			     perf_event__name(event->header.type));
1473 
1474 	switch (event->header.type) {
1475 	case PERF_RECORD_COMM:
1476 		ret += perf_event__fprintf_comm(event, fp);
1477 		break;
1478 	case PERF_RECORD_FORK:
1479 	case PERF_RECORD_EXIT:
1480 		ret += perf_event__fprintf_task(event, fp);
1481 		break;
1482 	case PERF_RECORD_MMAP:
1483 		ret += perf_event__fprintf_mmap(event, fp);
1484 		break;
1485 	case PERF_RECORD_NAMESPACES:
1486 		ret += perf_event__fprintf_namespaces(event, fp);
1487 		break;
1488 	case PERF_RECORD_MMAP2:
1489 		ret += perf_event__fprintf_mmap2(event, fp);
1490 		break;
1491 	case PERF_RECORD_AUX:
1492 		ret += perf_event__fprintf_aux(event, fp);
1493 		break;
1494 	case PERF_RECORD_ITRACE_START:
1495 		ret += perf_event__fprintf_itrace_start(event, fp);
1496 		break;
1497 	case PERF_RECORD_SWITCH:
1498 	case PERF_RECORD_SWITCH_CPU_WIDE:
1499 		ret += perf_event__fprintf_switch(event, fp);
1500 		break;
1501 	case PERF_RECORD_LOST:
1502 		ret += perf_event__fprintf_lost(event, fp);
1503 		break;
1504 	default:
1505 		ret += fprintf(fp, "\n");
1506 	}
1507 
1508 	return ret;
1509 }
1510 
perf_event__process(struct perf_tool * tool __maybe_unused,union perf_event * event,struct perf_sample * sample,struct machine * machine)1511 int perf_event__process(struct perf_tool *tool __maybe_unused,
1512 			union perf_event *event,
1513 			struct perf_sample *sample,
1514 			struct machine *machine)
1515 {
1516 	return machine__process_event(machine, event, sample);
1517 }
1518 
thread__find_map(struct thread * thread,u8 cpumode,u64 addr,struct addr_location * al)1519 struct map *thread__find_map(struct thread *thread, u8 cpumode, u64 addr,
1520 			     struct addr_location *al)
1521 {
1522 	struct map_groups *mg = thread->mg;
1523 	struct machine *machine = mg->machine;
1524 	bool load_map = false;
1525 
1526 	al->machine = machine;
1527 	al->thread = thread;
1528 	al->addr = addr;
1529 	al->cpumode = cpumode;
1530 	al->filtered = 0;
1531 
1532 	if (machine == NULL) {
1533 		al->map = NULL;
1534 		return NULL;
1535 	}
1536 
1537 	if (cpumode == PERF_RECORD_MISC_KERNEL && perf_host) {
1538 		al->level = 'k';
1539 		mg = &machine->kmaps;
1540 		load_map = true;
1541 	} else if (cpumode == PERF_RECORD_MISC_USER && perf_host) {
1542 		al->level = '.';
1543 	} else if (cpumode == PERF_RECORD_MISC_GUEST_KERNEL && perf_guest) {
1544 		al->level = 'g';
1545 		mg = &machine->kmaps;
1546 		load_map = true;
1547 	} else if (cpumode == PERF_RECORD_MISC_GUEST_USER && perf_guest) {
1548 		al->level = 'u';
1549 	} else {
1550 		al->level = 'H';
1551 		al->map = NULL;
1552 
1553 		if ((cpumode == PERF_RECORD_MISC_GUEST_USER ||
1554 			cpumode == PERF_RECORD_MISC_GUEST_KERNEL) &&
1555 			!perf_guest)
1556 			al->filtered |= (1 << HIST_FILTER__GUEST);
1557 		if ((cpumode == PERF_RECORD_MISC_USER ||
1558 			cpumode == PERF_RECORD_MISC_KERNEL) &&
1559 			!perf_host)
1560 			al->filtered |= (1 << HIST_FILTER__HOST);
1561 
1562 		return NULL;
1563 	}
1564 
1565 	al->map = map_groups__find(mg, al->addr);
1566 	if (al->map != NULL) {
1567 		/*
1568 		 * Kernel maps might be changed when loading symbols so loading
1569 		 * must be done prior to using kernel maps.
1570 		 */
1571 		if (load_map)
1572 			map__load(al->map);
1573 		al->addr = al->map->map_ip(al->map, al->addr);
1574 	}
1575 
1576 	return al->map;
1577 }
1578 
thread__find_symbol(struct thread * thread,u8 cpumode,u64 addr,struct addr_location * al)1579 struct symbol *thread__find_symbol(struct thread *thread, u8 cpumode,
1580 				   u64 addr, struct addr_location *al)
1581 {
1582 	al->sym = NULL;
1583 	if (thread__find_map(thread, cpumode, addr, al))
1584 		al->sym = map__find_symbol(al->map, al->addr);
1585 	return al->sym;
1586 }
1587 
1588 /*
1589  * Callers need to drop the reference to al->thread, obtained in
1590  * machine__findnew_thread()
1591  */
machine__resolve(struct machine * machine,struct addr_location * al,struct perf_sample * sample)1592 int machine__resolve(struct machine *machine, struct addr_location *al,
1593 		     struct perf_sample *sample)
1594 {
1595 	struct thread *thread = machine__findnew_thread(machine, sample->pid,
1596 							sample->tid);
1597 
1598 	if (thread == NULL)
1599 		return -1;
1600 
1601 	dump_printf(" ... thread: %s:%d\n", thread__comm_str(thread), thread->tid);
1602 	thread__find_map(thread, sample->cpumode, sample->ip, al);
1603 	dump_printf(" ...... dso: %s\n",
1604 		    al->map ? al->map->dso->long_name :
1605 			al->level == 'H' ? "[hypervisor]" : "<not found>");
1606 
1607 	if (thread__is_filtered(thread))
1608 		al->filtered |= (1 << HIST_FILTER__THREAD);
1609 
1610 	al->sym = NULL;
1611 	al->cpu = sample->cpu;
1612 	al->socket = -1;
1613 	al->srcline = NULL;
1614 
1615 	if (al->cpu >= 0) {
1616 		struct perf_env *env = machine->env;
1617 
1618 		if (env && env->cpu)
1619 			al->socket = env->cpu[al->cpu].socket_id;
1620 	}
1621 
1622 	if (al->map) {
1623 		struct dso *dso = al->map->dso;
1624 
1625 		if (symbol_conf.dso_list &&
1626 		    (!dso || !(strlist__has_entry(symbol_conf.dso_list,
1627 						  dso->short_name) ||
1628 			       (dso->short_name != dso->long_name &&
1629 				strlist__has_entry(symbol_conf.dso_list,
1630 						   dso->long_name))))) {
1631 			al->filtered |= (1 << HIST_FILTER__DSO);
1632 		}
1633 
1634 		al->sym = map__find_symbol(al->map, al->addr);
1635 	}
1636 
1637 	if (symbol_conf.sym_list &&
1638 		(!al->sym || !strlist__has_entry(symbol_conf.sym_list,
1639 						al->sym->name))) {
1640 		al->filtered |= (1 << HIST_FILTER__SYMBOL);
1641 	}
1642 
1643 	return 0;
1644 }
1645 
1646 /*
1647  * The preprocess_sample method will return with reference counts for the
1648  * in it, when done using (and perhaps getting ref counts if needing to
1649  * keep a pointer to one of those entries) it must be paired with
1650  * addr_location__put(), so that the refcounts can be decremented.
1651  */
addr_location__put(struct addr_location * al)1652 void addr_location__put(struct addr_location *al)
1653 {
1654 	thread__zput(al->thread);
1655 }
1656 
is_bts_event(struct perf_event_attr * attr)1657 bool is_bts_event(struct perf_event_attr *attr)
1658 {
1659 	return attr->type == PERF_TYPE_HARDWARE &&
1660 	       (attr->config & PERF_COUNT_HW_BRANCH_INSTRUCTIONS) &&
1661 	       attr->sample_period == 1;
1662 }
1663 
sample_addr_correlates_sym(struct perf_event_attr * attr)1664 bool sample_addr_correlates_sym(struct perf_event_attr *attr)
1665 {
1666 	if (attr->type == PERF_TYPE_SOFTWARE &&
1667 	    (attr->config == PERF_COUNT_SW_PAGE_FAULTS ||
1668 	     attr->config == PERF_COUNT_SW_PAGE_FAULTS_MIN ||
1669 	     attr->config == PERF_COUNT_SW_PAGE_FAULTS_MAJ))
1670 		return true;
1671 
1672 	if (is_bts_event(attr))
1673 		return true;
1674 
1675 	return false;
1676 }
1677 
thread__resolve(struct thread * thread,struct addr_location * al,struct perf_sample * sample)1678 void thread__resolve(struct thread *thread, struct addr_location *al,
1679 		     struct perf_sample *sample)
1680 {
1681 	thread__find_map(thread, sample->cpumode, sample->addr, al);
1682 
1683 	al->cpu = sample->cpu;
1684 	al->sym = NULL;
1685 
1686 	if (al->map)
1687 		al->sym = map__find_symbol(al->map, al->addr);
1688 }
1689