1/*
2 * Copyright (C) 2011, Red Hat Inc, Arnaldo Carvalho de Melo <acme@redhat.com>
3 *
4 * Parts came from builtin-annotate.c, see those files for further
5 * copyright notes.
6 *
7 * Released under the GPL v2. (and only v2, not any later version)
8 */
9
10#include "util.h"
11#include "ui/ui.h"
12#include "sort.h"
13#include "build-id.h"
14#include "color.h"
15#include "cache.h"
16#include "symbol.h"
17#include "debug.h"
18#include "annotate.h"
19#include "evsel.h"
20#include <regex.h>
21#include <pthread.h>
22#include <linux/bitops.h>
23
24const char 	*disassembler_style;
25const char	*objdump_path;
26static regex_t	 file_lineno;
27
28static struct ins *ins__find(const char *name);
29static int disasm_line__parse(char *line, char **namep, char **rawp);
30
31static void ins__delete(struct ins_operands *ops)
32{
33	if (ops == NULL)
34		return;
35	zfree(&ops->source.raw);
36	zfree(&ops->source.name);
37	zfree(&ops->target.raw);
38	zfree(&ops->target.name);
39}
40
41static int ins__raw_scnprintf(struct ins *ins, char *bf, size_t size,
42			      struct ins_operands *ops)
43{
44	return scnprintf(bf, size, "%-6.6s %s", ins->name, ops->raw);
45}
46
47int ins__scnprintf(struct ins *ins, char *bf, size_t size,
48		  struct ins_operands *ops)
49{
50	if (ins->ops->scnprintf)
51		return ins->ops->scnprintf(ins, bf, size, ops);
52
53	return ins__raw_scnprintf(ins, bf, size, ops);
54}
55
56static int call__parse(struct ins_operands *ops)
57{
58	char *endptr, *tok, *name;
59
60	ops->target.addr = strtoull(ops->raw, &endptr, 16);
61
62	name = strchr(endptr, '<');
63	if (name == NULL)
64		goto indirect_call;
65
66	name++;
67
68	tok = strchr(name, '>');
69	if (tok == NULL)
70		return -1;
71
72	*tok = '\0';
73	ops->target.name = strdup(name);
74	*tok = '>';
75
76	return ops->target.name == NULL ? -1 : 0;
77
78indirect_call:
79	tok = strchr(endptr, '(');
80	if (tok != NULL) {
81		ops->target.addr = 0;
82		return 0;
83	}
84
85	tok = strchr(endptr, '*');
86	if (tok == NULL)
87		return -1;
88
89	ops->target.addr = strtoull(tok + 1, NULL, 16);
90	return 0;
91}
92
93static int call__scnprintf(struct ins *ins, char *bf, size_t size,
94			   struct ins_operands *ops)
95{
96	if (ops->target.name)
97		return scnprintf(bf, size, "%-6.6s %s", ins->name, ops->target.name);
98
99	if (ops->target.addr == 0)
100		return ins__raw_scnprintf(ins, bf, size, ops);
101
102	return scnprintf(bf, size, "%-6.6s *%" PRIx64, ins->name, ops->target.addr);
103}
104
105static struct ins_ops call_ops = {
106	.parse	   = call__parse,
107	.scnprintf = call__scnprintf,
108};
109
110bool ins__is_call(const struct ins *ins)
111{
112	return ins->ops == &call_ops;
113}
114
115static int jump__parse(struct ins_operands *ops)
116{
117	const char *s = strchr(ops->raw, '+');
118
119	ops->target.addr = strtoull(ops->raw, NULL, 16);
120
121	if (s++ != NULL)
122		ops->target.offset = strtoull(s, NULL, 16);
123	else
124		ops->target.offset = UINT64_MAX;
125
126	return 0;
127}
128
129static int jump__scnprintf(struct ins *ins, char *bf, size_t size,
130			   struct ins_operands *ops)
131{
132	return scnprintf(bf, size, "%-6.6s %" PRIx64, ins->name, ops->target.offset);
133}
134
135static struct ins_ops jump_ops = {
136	.parse	   = jump__parse,
137	.scnprintf = jump__scnprintf,
138};
139
140bool ins__is_jump(const struct ins *ins)
141{
142	return ins->ops == &jump_ops;
143}
144
145static int comment__symbol(char *raw, char *comment, u64 *addrp, char **namep)
146{
147	char *endptr, *name, *t;
148
149	if (strstr(raw, "(%rip)") == NULL)
150		return 0;
151
152	*addrp = strtoull(comment, &endptr, 16);
153	name = strchr(endptr, '<');
154	if (name == NULL)
155		return -1;
156
157	name++;
158
159	t = strchr(name, '>');
160	if (t == NULL)
161		return 0;
162
163	*t = '\0';
164	*namep = strdup(name);
165	*t = '>';
166
167	return 0;
168}
169
170static int lock__parse(struct ins_operands *ops)
171{
172	char *name;
173
174	ops->locked.ops = zalloc(sizeof(*ops->locked.ops));
175	if (ops->locked.ops == NULL)
176		return 0;
177
178	if (disasm_line__parse(ops->raw, &name, &ops->locked.ops->raw) < 0)
179		goto out_free_ops;
180
181	ops->locked.ins = ins__find(name);
182	free(name);
183
184	if (ops->locked.ins == NULL)
185		goto out_free_ops;
186
187	if (!ops->locked.ins->ops)
188		return 0;
189
190	if (ops->locked.ins->ops->parse &&
191	    ops->locked.ins->ops->parse(ops->locked.ops) < 0)
192		goto out_free_ops;
193
194	return 0;
195
196out_free_ops:
197	zfree(&ops->locked.ops);
198	return 0;
199}
200
201static int lock__scnprintf(struct ins *ins, char *bf, size_t size,
202			   struct ins_operands *ops)
203{
204	int printed;
205
206	if (ops->locked.ins == NULL)
207		return ins__raw_scnprintf(ins, bf, size, ops);
208
209	printed = scnprintf(bf, size, "%-6.6s ", ins->name);
210	return printed + ins__scnprintf(ops->locked.ins, bf + printed,
211					size - printed, ops->locked.ops);
212}
213
214static void lock__delete(struct ins_operands *ops)
215{
216	struct ins *ins = ops->locked.ins;
217
218	if (ins && ins->ops->free)
219		ins->ops->free(ops->locked.ops);
220	else
221		ins__delete(ops->locked.ops);
222
223	zfree(&ops->locked.ops);
224	zfree(&ops->target.raw);
225	zfree(&ops->target.name);
226}
227
228static struct ins_ops lock_ops = {
229	.free	   = lock__delete,
230	.parse	   = lock__parse,
231	.scnprintf = lock__scnprintf,
232};
233
234static int mov__parse(struct ins_operands *ops)
235{
236	char *s = strchr(ops->raw, ','), *target, *comment, prev;
237
238	if (s == NULL)
239		return -1;
240
241	*s = '\0';
242	ops->source.raw = strdup(ops->raw);
243	*s = ',';
244
245	if (ops->source.raw == NULL)
246		return -1;
247
248	target = ++s;
249	comment = strchr(s, '#');
250
251	if (comment != NULL)
252		s = comment - 1;
253	else
254		s = strchr(s, '\0') - 1;
255
256	while (s > target && isspace(s[0]))
257		--s;
258	s++;
259	prev = *s;
260	*s = '\0';
261
262	ops->target.raw = strdup(target);
263	*s = prev;
264
265	if (ops->target.raw == NULL)
266		goto out_free_source;
267
268	if (comment == NULL)
269		return 0;
270
271	while (comment[0] != '\0' && isspace(comment[0]))
272		++comment;
273
274	comment__symbol(ops->source.raw, comment, &ops->source.addr, &ops->source.name);
275	comment__symbol(ops->target.raw, comment, &ops->target.addr, &ops->target.name);
276
277	return 0;
278
279out_free_source:
280	zfree(&ops->source.raw);
281	return -1;
282}
283
284static int mov__scnprintf(struct ins *ins, char *bf, size_t size,
285			   struct ins_operands *ops)
286{
287	return scnprintf(bf, size, "%-6.6s %s,%s", ins->name,
288			 ops->source.name ?: ops->source.raw,
289			 ops->target.name ?: ops->target.raw);
290}
291
292static struct ins_ops mov_ops = {
293	.parse	   = mov__parse,
294	.scnprintf = mov__scnprintf,
295};
296
297static int dec__parse(struct ins_operands *ops)
298{
299	char *target, *comment, *s, prev;
300
301	target = s = ops->raw;
302
303	while (s[0] != '\0' && !isspace(s[0]))
304		++s;
305	prev = *s;
306	*s = '\0';
307
308	ops->target.raw = strdup(target);
309	*s = prev;
310
311	if (ops->target.raw == NULL)
312		return -1;
313
314	comment = strchr(s, '#');
315	if (comment == NULL)
316		return 0;
317
318	while (comment[0] != '\0' && isspace(comment[0]))
319		++comment;
320
321	comment__symbol(ops->target.raw, comment, &ops->target.addr, &ops->target.name);
322
323	return 0;
324}
325
326static int dec__scnprintf(struct ins *ins, char *bf, size_t size,
327			   struct ins_operands *ops)
328{
329	return scnprintf(bf, size, "%-6.6s %s", ins->name,
330			 ops->target.name ?: ops->target.raw);
331}
332
333static struct ins_ops dec_ops = {
334	.parse	   = dec__parse,
335	.scnprintf = dec__scnprintf,
336};
337
338static int nop__scnprintf(struct ins *ins __maybe_unused, char *bf, size_t size,
339			  struct ins_operands *ops __maybe_unused)
340{
341	return scnprintf(bf, size, "%-6.6s", "nop");
342}
343
344static struct ins_ops nop_ops = {
345	.scnprintf = nop__scnprintf,
346};
347
348/*
349 * Must be sorted by name!
350 */
351static struct ins instructions[] = {
352	{ .name = "add",   .ops  = &mov_ops, },
353	{ .name = "addl",  .ops  = &mov_ops, },
354	{ .name = "addq",  .ops  = &mov_ops, },
355	{ .name = "addw",  .ops  = &mov_ops, },
356	{ .name = "and",   .ops  = &mov_ops, },
357	{ .name = "bts",   .ops  = &mov_ops, },
358	{ .name = "call",  .ops  = &call_ops, },
359	{ .name = "callq", .ops  = &call_ops, },
360	{ .name = "cmp",   .ops  = &mov_ops, },
361	{ .name = "cmpb",  .ops  = &mov_ops, },
362	{ .name = "cmpl",  .ops  = &mov_ops, },
363	{ .name = "cmpq",  .ops  = &mov_ops, },
364	{ .name = "cmpw",  .ops  = &mov_ops, },
365	{ .name = "cmpxch", .ops  = &mov_ops, },
366	{ .name = "dec",   .ops  = &dec_ops, },
367	{ .name = "decl",  .ops  = &dec_ops, },
368	{ .name = "imul",  .ops  = &mov_ops, },
369	{ .name = "inc",   .ops  = &dec_ops, },
370	{ .name = "incl",  .ops  = &dec_ops, },
371	{ .name = "ja",	   .ops  = &jump_ops, },
372	{ .name = "jae",   .ops  = &jump_ops, },
373	{ .name = "jb",	   .ops  = &jump_ops, },
374	{ .name = "jbe",   .ops  = &jump_ops, },
375	{ .name = "jc",	   .ops  = &jump_ops, },
376	{ .name = "jcxz",  .ops  = &jump_ops, },
377	{ .name = "je",	   .ops  = &jump_ops, },
378	{ .name = "jecxz", .ops  = &jump_ops, },
379	{ .name = "jg",	   .ops  = &jump_ops, },
380	{ .name = "jge",   .ops  = &jump_ops, },
381	{ .name = "jl",    .ops  = &jump_ops, },
382	{ .name = "jle",   .ops  = &jump_ops, },
383	{ .name = "jmp",   .ops  = &jump_ops, },
384	{ .name = "jmpq",  .ops  = &jump_ops, },
385	{ .name = "jna",   .ops  = &jump_ops, },
386	{ .name = "jnae",  .ops  = &jump_ops, },
387	{ .name = "jnb",   .ops  = &jump_ops, },
388	{ .name = "jnbe",  .ops  = &jump_ops, },
389	{ .name = "jnc",   .ops  = &jump_ops, },
390	{ .name = "jne",   .ops  = &jump_ops, },
391	{ .name = "jng",   .ops  = &jump_ops, },
392	{ .name = "jnge",  .ops  = &jump_ops, },
393	{ .name = "jnl",   .ops  = &jump_ops, },
394	{ .name = "jnle",  .ops  = &jump_ops, },
395	{ .name = "jno",   .ops  = &jump_ops, },
396	{ .name = "jnp",   .ops  = &jump_ops, },
397	{ .name = "jns",   .ops  = &jump_ops, },
398	{ .name = "jnz",   .ops  = &jump_ops, },
399	{ .name = "jo",	   .ops  = &jump_ops, },
400	{ .name = "jp",	   .ops  = &jump_ops, },
401	{ .name = "jpe",   .ops  = &jump_ops, },
402	{ .name = "jpo",   .ops  = &jump_ops, },
403	{ .name = "jrcxz", .ops  = &jump_ops, },
404	{ .name = "js",	   .ops  = &jump_ops, },
405	{ .name = "jz",	   .ops  = &jump_ops, },
406	{ .name = "lea",   .ops  = &mov_ops, },
407	{ .name = "lock",  .ops  = &lock_ops, },
408	{ .name = "mov",   .ops  = &mov_ops, },
409	{ .name = "movb",  .ops  = &mov_ops, },
410	{ .name = "movdqa",.ops  = &mov_ops, },
411	{ .name = "movl",  .ops  = &mov_ops, },
412	{ .name = "movq",  .ops  = &mov_ops, },
413	{ .name = "movslq", .ops  = &mov_ops, },
414	{ .name = "movzbl", .ops  = &mov_ops, },
415	{ .name = "movzwl", .ops  = &mov_ops, },
416	{ .name = "nop",   .ops  = &nop_ops, },
417	{ .name = "nopl",  .ops  = &nop_ops, },
418	{ .name = "nopw",  .ops  = &nop_ops, },
419	{ .name = "or",    .ops  = &mov_ops, },
420	{ .name = "orl",   .ops  = &mov_ops, },
421	{ .name = "test",  .ops  = &mov_ops, },
422	{ .name = "testb", .ops  = &mov_ops, },
423	{ .name = "testl", .ops  = &mov_ops, },
424	{ .name = "xadd",  .ops  = &mov_ops, },
425	{ .name = "xbeginl", .ops  = &jump_ops, },
426	{ .name = "xbeginq", .ops  = &jump_ops, },
427};
428
429static int ins__cmp(const void *name, const void *insp)
430{
431	const struct ins *ins = insp;
432
433	return strcmp(name, ins->name);
434}
435
436static struct ins *ins__find(const char *name)
437{
438	const int nmemb = ARRAY_SIZE(instructions);
439
440	return bsearch(name, instructions, nmemb, sizeof(struct ins), ins__cmp);
441}
442
443int symbol__annotate_init(struct map *map __maybe_unused, struct symbol *sym)
444{
445	struct annotation *notes = symbol__annotation(sym);
446	pthread_mutex_init(&notes->lock, NULL);
447	return 0;
448}
449
450int symbol__alloc_hist(struct symbol *sym)
451{
452	struct annotation *notes = symbol__annotation(sym);
453	const size_t size = symbol__size(sym);
454	size_t sizeof_sym_hist;
455
456	/* Check for overflow when calculating sizeof_sym_hist */
457	if (size > (SIZE_MAX - sizeof(struct sym_hist)) / sizeof(u64))
458		return -1;
459
460	sizeof_sym_hist = (sizeof(struct sym_hist) + size * sizeof(u64));
461
462	/* Check for overflow in zalloc argument */
463	if (sizeof_sym_hist > (SIZE_MAX - sizeof(*notes->src))
464				/ symbol_conf.nr_events)
465		return -1;
466
467	notes->src = zalloc(sizeof(*notes->src) + symbol_conf.nr_events * sizeof_sym_hist);
468	if (notes->src == NULL)
469		return -1;
470	notes->src->sizeof_sym_hist = sizeof_sym_hist;
471	notes->src->nr_histograms   = symbol_conf.nr_events;
472	INIT_LIST_HEAD(&notes->src->source);
473	return 0;
474}
475
476void symbol__annotate_zero_histograms(struct symbol *sym)
477{
478	struct annotation *notes = symbol__annotation(sym);
479
480	pthread_mutex_lock(&notes->lock);
481	if (notes->src != NULL)
482		memset(notes->src->histograms, 0,
483		       notes->src->nr_histograms * notes->src->sizeof_sym_hist);
484	pthread_mutex_unlock(&notes->lock);
485}
486
487static int __symbol__inc_addr_samples(struct symbol *sym, struct map *map,
488				      struct annotation *notes, int evidx, u64 addr)
489{
490	unsigned offset;
491	struct sym_hist *h;
492
493	pr_debug3("%s: addr=%#" PRIx64 "\n", __func__, map->unmap_ip(map, addr));
494
495	if (addr < sym->start || addr >= sym->end)
496		return -ERANGE;
497
498	offset = addr - sym->start;
499	h = annotation__histogram(notes, evidx);
500	h->sum++;
501	h->addr[offset]++;
502
503	pr_debug3("%#" PRIx64 " %s: period++ [addr: %#" PRIx64 ", %#" PRIx64
504		  ", evidx=%d] => %" PRIu64 "\n", sym->start, sym->name,
505		  addr, addr - sym->start, evidx, h->addr[offset]);
506	return 0;
507}
508
509static int symbol__inc_addr_samples(struct symbol *sym, struct map *map,
510				    int evidx, u64 addr)
511{
512	struct annotation *notes;
513
514	if (sym == NULL)
515		return 0;
516
517	notes = symbol__annotation(sym);
518	if (notes->src == NULL) {
519		if (symbol__alloc_hist(sym) < 0)
520			return -ENOMEM;
521	}
522
523	return __symbol__inc_addr_samples(sym, map, notes, evidx, addr);
524}
525
526int addr_map_symbol__inc_samples(struct addr_map_symbol *ams, int evidx)
527{
528	return symbol__inc_addr_samples(ams->sym, ams->map, evidx, ams->al_addr);
529}
530
531int hist_entry__inc_addr_samples(struct hist_entry *he, int evidx, u64 ip)
532{
533	return symbol__inc_addr_samples(he->ms.sym, he->ms.map, evidx, ip);
534}
535
536static void disasm_line__init_ins(struct disasm_line *dl)
537{
538	dl->ins = ins__find(dl->name);
539
540	if (dl->ins == NULL)
541		return;
542
543	if (!dl->ins->ops)
544		return;
545
546	if (dl->ins->ops->parse && dl->ins->ops->parse(&dl->ops) < 0)
547		dl->ins = NULL;
548}
549
550static int disasm_line__parse(char *line, char **namep, char **rawp)
551{
552	char *name = line, tmp;
553
554	while (isspace(name[0]))
555		++name;
556
557	if (name[0] == '\0')
558		return -1;
559
560	*rawp = name + 1;
561
562	while ((*rawp)[0] != '\0' && !isspace((*rawp)[0]))
563		++*rawp;
564
565	tmp = (*rawp)[0];
566	(*rawp)[0] = '\0';
567	*namep = strdup(name);
568
569	if (*namep == NULL)
570		goto out_free_name;
571
572	(*rawp)[0] = tmp;
573
574	if ((*rawp)[0] != '\0') {
575		(*rawp)++;
576		while (isspace((*rawp)[0]))
577			++(*rawp);
578	}
579
580	return 0;
581
582out_free_name:
583	zfree(namep);
584	return -1;
585}
586
587static struct disasm_line *disasm_line__new(s64 offset, char *line,
588					size_t privsize, int line_nr)
589{
590	struct disasm_line *dl = zalloc(sizeof(*dl) + privsize);
591
592	if (dl != NULL) {
593		dl->offset = offset;
594		dl->line = strdup(line);
595		dl->line_nr = line_nr;
596		if (dl->line == NULL)
597			goto out_delete;
598
599		if (offset != -1) {
600			if (disasm_line__parse(dl->line, &dl->name, &dl->ops.raw) < 0)
601				goto out_free_line;
602
603			disasm_line__init_ins(dl);
604		}
605	}
606
607	return dl;
608
609out_free_line:
610	zfree(&dl->line);
611out_delete:
612	free(dl);
613	return NULL;
614}
615
616void disasm_line__free(struct disasm_line *dl)
617{
618	zfree(&dl->line);
619	zfree(&dl->name);
620	if (dl->ins && dl->ins->ops->free)
621		dl->ins->ops->free(&dl->ops);
622	else
623		ins__delete(&dl->ops);
624	free(dl);
625}
626
627int disasm_line__scnprintf(struct disasm_line *dl, char *bf, size_t size, bool raw)
628{
629	if (raw || !dl->ins)
630		return scnprintf(bf, size, "%-6.6s %s", dl->name, dl->ops.raw);
631
632	return ins__scnprintf(dl->ins, bf, size, &dl->ops);
633}
634
635static void disasm__add(struct list_head *head, struct disasm_line *line)
636{
637	list_add_tail(&line->node, head);
638}
639
640struct disasm_line *disasm__get_next_ip_line(struct list_head *head, struct disasm_line *pos)
641{
642	list_for_each_entry_continue(pos, head, node)
643		if (pos->offset >= 0)
644			return pos;
645
646	return NULL;
647}
648
649double disasm__calc_percent(struct annotation *notes, int evidx, s64 offset,
650			    s64 end, const char **path)
651{
652	struct source_line *src_line = notes->src->lines;
653	double percent = 0.0;
654
655	if (src_line) {
656		size_t sizeof_src_line = sizeof(*src_line) +
657				sizeof(src_line->p) * (src_line->nr_pcnt - 1);
658
659		while (offset < end) {
660			src_line = (void *)notes->src->lines +
661					(sizeof_src_line * offset);
662
663			if (*path == NULL)
664				*path = src_line->path;
665
666			percent += src_line->p[evidx].percent;
667			offset++;
668		}
669	} else {
670		struct sym_hist *h = annotation__histogram(notes, evidx);
671		unsigned int hits = 0;
672
673		while (offset < end)
674			hits += h->addr[offset++];
675
676		if (h->sum)
677			percent = 100.0 * hits / h->sum;
678	}
679
680	return percent;
681}
682
683static int disasm_line__print(struct disasm_line *dl, struct symbol *sym, u64 start,
684		      struct perf_evsel *evsel, u64 len, int min_pcnt, int printed,
685		      int max_lines, struct disasm_line *queue)
686{
687	static const char *prev_line;
688	static const char *prev_color;
689
690	if (dl->offset != -1) {
691		const char *path = NULL;
692		double percent, max_percent = 0.0;
693		double *ppercents = &percent;
694		int i, nr_percent = 1;
695		const char *color;
696		struct annotation *notes = symbol__annotation(sym);
697		s64 offset = dl->offset;
698		const u64 addr = start + offset;
699		struct disasm_line *next;
700
701		next = disasm__get_next_ip_line(&notes->src->source, dl);
702
703		if (perf_evsel__is_group_event(evsel)) {
704			nr_percent = evsel->nr_members;
705			ppercents = calloc(nr_percent, sizeof(double));
706			if (ppercents == NULL)
707				return -1;
708		}
709
710		for (i = 0; i < nr_percent; i++) {
711			percent = disasm__calc_percent(notes,
712					notes->src->lines ? i : evsel->idx + i,
713					offset,
714					next ? next->offset : (s64) len,
715					&path);
716
717			ppercents[i] = percent;
718			if (percent > max_percent)
719				max_percent = percent;
720		}
721
722		if (max_percent < min_pcnt)
723			return -1;
724
725		if (max_lines && printed >= max_lines)
726			return 1;
727
728		if (queue != NULL) {
729			list_for_each_entry_from(queue, &notes->src->source, node) {
730				if (queue == dl)
731					break;
732				disasm_line__print(queue, sym, start, evsel, len,
733						    0, 0, 1, NULL);
734			}
735		}
736
737		color = get_percent_color(max_percent);
738
739		/*
740		 * Also color the filename and line if needed, with
741		 * the same color than the percentage. Don't print it
742		 * twice for close colored addr with the same filename:line
743		 */
744		if (path) {
745			if (!prev_line || strcmp(prev_line, path)
746				       || color != prev_color) {
747				color_fprintf(stdout, color, " %s", path);
748				prev_line = path;
749				prev_color = color;
750			}
751		}
752
753		for (i = 0; i < nr_percent; i++) {
754			percent = ppercents[i];
755			color = get_percent_color(percent);
756			color_fprintf(stdout, color, " %7.2f", percent);
757		}
758
759		printf(" :	");
760		color_fprintf(stdout, PERF_COLOR_MAGENTA, "  %" PRIx64 ":", addr);
761		color_fprintf(stdout, PERF_COLOR_BLUE, "%s\n", dl->line);
762
763		if (ppercents != &percent)
764			free(ppercents);
765
766	} else if (max_lines && printed >= max_lines)
767		return 1;
768	else {
769		int width = 8;
770
771		if (queue)
772			return -1;
773
774		if (perf_evsel__is_group_event(evsel))
775			width *= evsel->nr_members;
776
777		if (!*dl->line)
778			printf(" %*s:\n", width, " ");
779		else
780			printf(" %*s:	%s\n", width, " ", dl->line);
781	}
782
783	return 0;
784}
785
786/*
787 * symbol__parse_objdump_line() parses objdump output (with -d --no-show-raw)
788 * which looks like following
789 *
790 *  0000000000415500 <_init>:
791 *    415500:       sub    $0x8,%rsp
792 *    415504:       mov    0x2f5ad5(%rip),%rax        # 70afe0 <_DYNAMIC+0x2f8>
793 *    41550b:       test   %rax,%rax
794 *    41550e:       je     415515 <_init+0x15>
795 *    415510:       callq  416e70 <__gmon_start__@plt>
796 *    415515:       add    $0x8,%rsp
797 *    415519:       retq
798 *
799 * it will be parsed and saved into struct disasm_line as
800 *  <offset>       <name>  <ops.raw>
801 *
802 * The offset will be a relative offset from the start of the symbol and -1
803 * means that it's not a disassembly line so should be treated differently.
804 * The ops.raw part will be parsed further according to type of the instruction.
805 */
806static int symbol__parse_objdump_line(struct symbol *sym, struct map *map,
807				      FILE *file, size_t privsize,
808				      int *line_nr)
809{
810	struct annotation *notes = symbol__annotation(sym);
811	struct disasm_line *dl;
812	char *line = NULL, *parsed_line, *tmp, *tmp2, *c;
813	size_t line_len;
814	s64 line_ip, offset = -1;
815	regmatch_t match[2];
816
817	if (getline(&line, &line_len, file) < 0)
818		return -1;
819
820	if (!line)
821		return -1;
822
823	while (line_len != 0 && isspace(line[line_len - 1]))
824		line[--line_len] = '\0';
825
826	c = strchr(line, '\n');
827	if (c)
828		*c = 0;
829
830	line_ip = -1;
831	parsed_line = line;
832
833	/* /filename:linenr ? Save line number and ignore. */
834	if (regexec(&file_lineno, line, 2, match, 0) == 0) {
835		*line_nr = atoi(line + match[1].rm_so);
836		return 0;
837	}
838
839	/*
840	 * Strip leading spaces:
841	 */
842	tmp = line;
843	while (*tmp) {
844		if (*tmp != ' ')
845			break;
846		tmp++;
847	}
848
849	if (*tmp) {
850		/*
851		 * Parse hexa addresses followed by ':'
852		 */
853		line_ip = strtoull(tmp, &tmp2, 16);
854		if (*tmp2 != ':' || tmp == tmp2 || tmp2[1] == '\0')
855			line_ip = -1;
856	}
857
858	if (line_ip != -1) {
859		u64 start = map__rip_2objdump(map, sym->start),
860		    end = map__rip_2objdump(map, sym->end);
861
862		offset = line_ip - start;
863		if ((u64)line_ip < start || (u64)line_ip >= end)
864			offset = -1;
865		else
866			parsed_line = tmp2 + 1;
867	}
868
869	dl = disasm_line__new(offset, parsed_line, privsize, *line_nr);
870	free(line);
871	(*line_nr)++;
872
873	if (dl == NULL)
874		return -1;
875
876	if (dl->ops.target.offset == UINT64_MAX)
877		dl->ops.target.offset = dl->ops.target.addr -
878					map__rip_2objdump(map, sym->start);
879
880	/* kcore has no symbols, so add the call target name */
881	if (dl->ins && ins__is_call(dl->ins) && !dl->ops.target.name) {
882		struct addr_map_symbol target = {
883			.map = map,
884			.addr = dl->ops.target.addr,
885		};
886
887		if (!map_groups__find_ams(&target, NULL) &&
888		    target.sym->start == target.al_addr)
889			dl->ops.target.name = strdup(target.sym->name);
890	}
891
892	disasm__add(&notes->src->source, dl);
893
894	return 0;
895}
896
897static __attribute__((constructor)) void symbol__init_regexpr(void)
898{
899	regcomp(&file_lineno, "^/[^:]+:([0-9]+)", REG_EXTENDED);
900}
901
902static void delete_last_nop(struct symbol *sym)
903{
904	struct annotation *notes = symbol__annotation(sym);
905	struct list_head *list = &notes->src->source;
906	struct disasm_line *dl;
907
908	while (!list_empty(list)) {
909		dl = list_entry(list->prev, struct disasm_line, node);
910
911		if (dl->ins && dl->ins->ops) {
912			if (dl->ins->ops != &nop_ops)
913				return;
914		} else {
915			if (!strstr(dl->line, " nop ") &&
916			    !strstr(dl->line, " nopl ") &&
917			    !strstr(dl->line, " nopw "))
918				return;
919		}
920
921		list_del(&dl->node);
922		disasm_line__free(dl);
923	}
924}
925
926int symbol__annotate(struct symbol *sym, struct map *map, size_t privsize)
927{
928	struct dso *dso = map->dso;
929	char *filename = dso__build_id_filename(dso, NULL, 0);
930	bool free_filename = true;
931	char command[PATH_MAX * 2];
932	FILE *file;
933	int err = 0;
934	char symfs_filename[PATH_MAX];
935	struct kcore_extract kce;
936	bool delete_extract = false;
937	int lineno = 0;
938
939	if (filename)
940		symbol__join_symfs(symfs_filename, filename);
941
942	if (filename == NULL) {
943		if (dso->has_build_id) {
944			pr_err("Can't annotate %s: not enough memory\n",
945			       sym->name);
946			return -ENOMEM;
947		}
948		goto fallback;
949	} else if (dso__is_kcore(dso)) {
950		goto fallback;
951	} else if (readlink(symfs_filename, command, sizeof(command)) < 0 ||
952		   strstr(command, "[kernel.kallsyms]") ||
953		   access(symfs_filename, R_OK)) {
954		free(filename);
955fallback:
956		/*
957		 * If we don't have build-ids or the build-id file isn't in the
958		 * cache, or is just a kallsyms file, well, lets hope that this
959		 * DSO is the same as when 'perf record' ran.
960		 */
961		filename = (char *)dso->long_name;
962		symbol__join_symfs(symfs_filename, filename);
963		free_filename = false;
964	}
965
966	if (dso->symtab_type == DSO_BINARY_TYPE__KALLSYMS &&
967	    !dso__is_kcore(dso)) {
968		char bf[BUILD_ID_SIZE * 2 + 16] = " with build id ";
969		char *build_id_msg = NULL;
970
971		if (dso->annotate_warned)
972			goto out_free_filename;
973
974		if (dso->has_build_id) {
975			build_id__sprintf(dso->build_id,
976					  sizeof(dso->build_id), bf + 15);
977			build_id_msg = bf;
978		}
979		err = -ENOENT;
980		dso->annotate_warned = 1;
981		pr_err("Can't annotate %s:\n\n"
982		       "No vmlinux file%s\nwas found in the path.\n\n"
983		       "Please use:\n\n"
984		       "  perf buildid-cache -vu vmlinux\n\n"
985		       "or:\n\n"
986		       "  --vmlinux vmlinux\n",
987		       sym->name, build_id_msg ?: "");
988		goto out_free_filename;
989	}
990
991	pr_debug("%s: filename=%s, sym=%s, start=%#" PRIx64 ", end=%#" PRIx64 "\n", __func__,
992		 filename, sym->name, map->unmap_ip(map, sym->start),
993		 map->unmap_ip(map, sym->end));
994
995	pr_debug("annotating [%p] %30s : [%p] %30s\n",
996		 dso, dso->long_name, sym, sym->name);
997
998	if (dso__is_kcore(dso)) {
999		kce.kcore_filename = symfs_filename;
1000		kce.addr = map__rip_2objdump(map, sym->start);
1001		kce.offs = sym->start;
1002		kce.len = sym->end - sym->start;
1003		if (!kcore_extract__create(&kce)) {
1004			delete_extract = true;
1005			strlcpy(symfs_filename, kce.extract_filename,
1006				sizeof(symfs_filename));
1007			if (free_filename) {
1008				free(filename);
1009				free_filename = false;
1010			}
1011			filename = symfs_filename;
1012		}
1013	} else if (dso__needs_decompress(dso)) {
1014		char tmp[PATH_MAX];
1015		struct kmod_path m;
1016		int fd;
1017		bool ret;
1018
1019		if (kmod_path__parse_ext(&m, symfs_filename))
1020			goto out_free_filename;
1021
1022		snprintf(tmp, PATH_MAX, "/tmp/perf-kmod-XXXXXX");
1023
1024		fd = mkstemp(tmp);
1025		if (fd < 0) {
1026			free(m.ext);
1027			goto out_free_filename;
1028		}
1029
1030		ret = decompress_to_file(m.ext, symfs_filename, fd);
1031
1032		free(m.ext);
1033		close(fd);
1034
1035		if (!ret)
1036			goto out_free_filename;
1037
1038		strcpy(symfs_filename, tmp);
1039	}
1040
1041	snprintf(command, sizeof(command),
1042		 "%s %s%s --start-address=0x%016" PRIx64
1043		 " --stop-address=0x%016" PRIx64
1044		 " -l -d %s %s -C %s 2>/dev/null|grep -v %s|expand",
1045		 objdump_path ? objdump_path : "objdump",
1046		 disassembler_style ? "-M " : "",
1047		 disassembler_style ? disassembler_style : "",
1048		 map__rip_2objdump(map, sym->start),
1049		 map__rip_2objdump(map, sym->end),
1050		 symbol_conf.annotate_asm_raw ? "" : "--no-show-raw",
1051		 symbol_conf.annotate_src ? "-S" : "",
1052		 symfs_filename, filename);
1053
1054	pr_debug("Executing: %s\n", command);
1055
1056	file = popen(command, "r");
1057	if (!file)
1058		goto out_remove_tmp;
1059
1060	while (!feof(file))
1061		if (symbol__parse_objdump_line(sym, map, file, privsize,
1062			    &lineno) < 0)
1063			break;
1064
1065	/*
1066	 * kallsyms does not have symbol sizes so there may a nop at the end.
1067	 * Remove it.
1068	 */
1069	if (dso__is_kcore(dso))
1070		delete_last_nop(sym);
1071
1072	pclose(file);
1073
1074out_remove_tmp:
1075	if (dso__needs_decompress(dso))
1076		unlink(symfs_filename);
1077out_free_filename:
1078	if (delete_extract)
1079		kcore_extract__delete(&kce);
1080	if (free_filename)
1081		free(filename);
1082	return err;
1083}
1084
1085static void insert_source_line(struct rb_root *root, struct source_line *src_line)
1086{
1087	struct source_line *iter;
1088	struct rb_node **p = &root->rb_node;
1089	struct rb_node *parent = NULL;
1090	int i, ret;
1091
1092	while (*p != NULL) {
1093		parent = *p;
1094		iter = rb_entry(parent, struct source_line, node);
1095
1096		ret = strcmp(iter->path, src_line->path);
1097		if (ret == 0) {
1098			for (i = 0; i < src_line->nr_pcnt; i++)
1099				iter->p[i].percent_sum += src_line->p[i].percent;
1100			return;
1101		}
1102
1103		if (ret < 0)
1104			p = &(*p)->rb_left;
1105		else
1106			p = &(*p)->rb_right;
1107	}
1108
1109	for (i = 0; i < src_line->nr_pcnt; i++)
1110		src_line->p[i].percent_sum = src_line->p[i].percent;
1111
1112	rb_link_node(&src_line->node, parent, p);
1113	rb_insert_color(&src_line->node, root);
1114}
1115
1116static int cmp_source_line(struct source_line *a, struct source_line *b)
1117{
1118	int i;
1119
1120	for (i = 0; i < a->nr_pcnt; i++) {
1121		if (a->p[i].percent_sum == b->p[i].percent_sum)
1122			continue;
1123		return a->p[i].percent_sum > b->p[i].percent_sum;
1124	}
1125
1126	return 0;
1127}
1128
1129static void __resort_source_line(struct rb_root *root, struct source_line *src_line)
1130{
1131	struct source_line *iter;
1132	struct rb_node **p = &root->rb_node;
1133	struct rb_node *parent = NULL;
1134
1135	while (*p != NULL) {
1136		parent = *p;
1137		iter = rb_entry(parent, struct source_line, node);
1138
1139		if (cmp_source_line(src_line, iter))
1140			p = &(*p)->rb_left;
1141		else
1142			p = &(*p)->rb_right;
1143	}
1144
1145	rb_link_node(&src_line->node, parent, p);
1146	rb_insert_color(&src_line->node, root);
1147}
1148
1149static void resort_source_line(struct rb_root *dest_root, struct rb_root *src_root)
1150{
1151	struct source_line *src_line;
1152	struct rb_node *node;
1153
1154	node = rb_first(src_root);
1155	while (node) {
1156		struct rb_node *next;
1157
1158		src_line = rb_entry(node, struct source_line, node);
1159		next = rb_next(node);
1160		rb_erase(node, src_root);
1161
1162		__resort_source_line(dest_root, src_line);
1163		node = next;
1164	}
1165}
1166
1167static void symbol__free_source_line(struct symbol *sym, int len)
1168{
1169	struct annotation *notes = symbol__annotation(sym);
1170	struct source_line *src_line = notes->src->lines;
1171	size_t sizeof_src_line;
1172	int i;
1173
1174	sizeof_src_line = sizeof(*src_line) +
1175			  (sizeof(src_line->p) * (src_line->nr_pcnt - 1));
1176
1177	for (i = 0; i < len; i++) {
1178		free_srcline(src_line->path);
1179		src_line = (void *)src_line + sizeof_src_line;
1180	}
1181
1182	zfree(&notes->src->lines);
1183}
1184
1185/* Get the filename:line for the colored entries */
1186static int symbol__get_source_line(struct symbol *sym, struct map *map,
1187				   struct perf_evsel *evsel,
1188				   struct rb_root *root, int len)
1189{
1190	u64 start;
1191	int i, k;
1192	int evidx = evsel->idx;
1193	struct source_line *src_line;
1194	struct annotation *notes = symbol__annotation(sym);
1195	struct sym_hist *h = annotation__histogram(notes, evidx);
1196	struct rb_root tmp_root = RB_ROOT;
1197	int nr_pcnt = 1;
1198	u64 h_sum = h->sum;
1199	size_t sizeof_src_line = sizeof(struct source_line);
1200
1201	if (perf_evsel__is_group_event(evsel)) {
1202		for (i = 1; i < evsel->nr_members; i++) {
1203			h = annotation__histogram(notes, evidx + i);
1204			h_sum += h->sum;
1205		}
1206		nr_pcnt = evsel->nr_members;
1207		sizeof_src_line += (nr_pcnt - 1) * sizeof(src_line->p);
1208	}
1209
1210	if (!h_sum)
1211		return 0;
1212
1213	src_line = notes->src->lines = calloc(len, sizeof_src_line);
1214	if (!notes->src->lines)
1215		return -1;
1216
1217	start = map__rip_2objdump(map, sym->start);
1218
1219	for (i = 0; i < len; i++) {
1220		u64 offset;
1221		double percent_max = 0.0;
1222
1223		src_line->nr_pcnt = nr_pcnt;
1224
1225		for (k = 0; k < nr_pcnt; k++) {
1226			h = annotation__histogram(notes, evidx + k);
1227			src_line->p[k].percent = 100.0 * h->addr[i] / h->sum;
1228
1229			if (src_line->p[k].percent > percent_max)
1230				percent_max = src_line->p[k].percent;
1231		}
1232
1233		if (percent_max <= 0.5)
1234			goto next;
1235
1236		offset = start + i;
1237		src_line->path = get_srcline(map->dso, offset, NULL, false);
1238		insert_source_line(&tmp_root, src_line);
1239
1240	next:
1241		src_line = (void *)src_line + sizeof_src_line;
1242	}
1243
1244	resort_source_line(root, &tmp_root);
1245	return 0;
1246}
1247
1248static void print_summary(struct rb_root *root, const char *filename)
1249{
1250	struct source_line *src_line;
1251	struct rb_node *node;
1252
1253	printf("\nSorted summary for file %s\n", filename);
1254	printf("----------------------------------------------\n\n");
1255
1256	if (RB_EMPTY_ROOT(root)) {
1257		printf(" Nothing higher than %1.1f%%\n", MIN_GREEN);
1258		return;
1259	}
1260
1261	node = rb_first(root);
1262	while (node) {
1263		double percent, percent_max = 0.0;
1264		const char *color;
1265		char *path;
1266		int i;
1267
1268		src_line = rb_entry(node, struct source_line, node);
1269		for (i = 0; i < src_line->nr_pcnt; i++) {
1270			percent = src_line->p[i].percent_sum;
1271			color = get_percent_color(percent);
1272			color_fprintf(stdout, color, " %7.2f", percent);
1273
1274			if (percent > percent_max)
1275				percent_max = percent;
1276		}
1277
1278		path = src_line->path;
1279		color = get_percent_color(percent_max);
1280		color_fprintf(stdout, color, " %s\n", path);
1281
1282		node = rb_next(node);
1283	}
1284}
1285
1286static void symbol__annotate_hits(struct symbol *sym, struct perf_evsel *evsel)
1287{
1288	struct annotation *notes = symbol__annotation(sym);
1289	struct sym_hist *h = annotation__histogram(notes, evsel->idx);
1290	u64 len = symbol__size(sym), offset;
1291
1292	for (offset = 0; offset < len; ++offset)
1293		if (h->addr[offset] != 0)
1294			printf("%*" PRIx64 ": %" PRIu64 "\n", BITS_PER_LONG / 2,
1295			       sym->start + offset, h->addr[offset]);
1296	printf("%*s: %" PRIu64 "\n", BITS_PER_LONG / 2, "h->sum", h->sum);
1297}
1298
1299int symbol__annotate_printf(struct symbol *sym, struct map *map,
1300			    struct perf_evsel *evsel, bool full_paths,
1301			    int min_pcnt, int max_lines, int context)
1302{
1303	struct dso *dso = map->dso;
1304	char *filename;
1305	const char *d_filename;
1306	const char *evsel_name = perf_evsel__name(evsel);
1307	struct annotation *notes = symbol__annotation(sym);
1308	struct disasm_line *pos, *queue = NULL;
1309	u64 start = map__rip_2objdump(map, sym->start);
1310	int printed = 2, queue_len = 0;
1311	int more = 0;
1312	u64 len;
1313	int width = 8;
1314	int namelen, evsel_name_len, graph_dotted_len;
1315
1316	filename = strdup(dso->long_name);
1317	if (!filename)
1318		return -ENOMEM;
1319
1320	if (full_paths)
1321		d_filename = filename;
1322	else
1323		d_filename = basename(filename);
1324
1325	len = symbol__size(sym);
1326	namelen = strlen(d_filename);
1327	evsel_name_len = strlen(evsel_name);
1328
1329	if (perf_evsel__is_group_event(evsel))
1330		width *= evsel->nr_members;
1331
1332	printf(" %-*.*s|	Source code & Disassembly of %s for %s\n",
1333	       width, width, "Percent", d_filename, evsel_name);
1334
1335	graph_dotted_len = width + namelen + evsel_name_len;
1336	printf("-%-*.*s-----------------------------------------\n",
1337	       graph_dotted_len, graph_dotted_len, graph_dotted_line);
1338
1339	if (verbose)
1340		symbol__annotate_hits(sym, evsel);
1341
1342	list_for_each_entry(pos, &notes->src->source, node) {
1343		if (context && queue == NULL) {
1344			queue = pos;
1345			queue_len = 0;
1346		}
1347
1348		switch (disasm_line__print(pos, sym, start, evsel, len,
1349					    min_pcnt, printed, max_lines,
1350					    queue)) {
1351		case 0:
1352			++printed;
1353			if (context) {
1354				printed += queue_len;
1355				queue = NULL;
1356				queue_len = 0;
1357			}
1358			break;
1359		case 1:
1360			/* filtered by max_lines */
1361			++more;
1362			break;
1363		case -1:
1364		default:
1365			/*
1366			 * Filtered by min_pcnt or non IP lines when
1367			 * context != 0
1368			 */
1369			if (!context)
1370				break;
1371			if (queue_len == context)
1372				queue = list_entry(queue->node.next, typeof(*queue), node);
1373			else
1374				++queue_len;
1375			break;
1376		}
1377	}
1378
1379	free(filename);
1380
1381	return more;
1382}
1383
1384void symbol__annotate_zero_histogram(struct symbol *sym, int evidx)
1385{
1386	struct annotation *notes = symbol__annotation(sym);
1387	struct sym_hist *h = annotation__histogram(notes, evidx);
1388
1389	memset(h, 0, notes->src->sizeof_sym_hist);
1390}
1391
1392void symbol__annotate_decay_histogram(struct symbol *sym, int evidx)
1393{
1394	struct annotation *notes = symbol__annotation(sym);
1395	struct sym_hist *h = annotation__histogram(notes, evidx);
1396	int len = symbol__size(sym), offset;
1397
1398	h->sum = 0;
1399	for (offset = 0; offset < len; ++offset) {
1400		h->addr[offset] = h->addr[offset] * 7 / 8;
1401		h->sum += h->addr[offset];
1402	}
1403}
1404
1405void disasm__purge(struct list_head *head)
1406{
1407	struct disasm_line *pos, *n;
1408
1409	list_for_each_entry_safe(pos, n, head, node) {
1410		list_del(&pos->node);
1411		disasm_line__free(pos);
1412	}
1413}
1414
1415static size_t disasm_line__fprintf(struct disasm_line *dl, FILE *fp)
1416{
1417	size_t printed;
1418
1419	if (dl->offset == -1)
1420		return fprintf(fp, "%s\n", dl->line);
1421
1422	printed = fprintf(fp, "%#" PRIx64 " %s", dl->offset, dl->name);
1423
1424	if (dl->ops.raw[0] != '\0') {
1425		printed += fprintf(fp, "%.*s %s\n", 6 - (int)printed, " ",
1426				   dl->ops.raw);
1427	}
1428
1429	return printed + fprintf(fp, "\n");
1430}
1431
1432size_t disasm__fprintf(struct list_head *head, FILE *fp)
1433{
1434	struct disasm_line *pos;
1435	size_t printed = 0;
1436
1437	list_for_each_entry(pos, head, node)
1438		printed += disasm_line__fprintf(pos, fp);
1439
1440	return printed;
1441}
1442
1443int symbol__tty_annotate(struct symbol *sym, struct map *map,
1444			 struct perf_evsel *evsel, bool print_lines,
1445			 bool full_paths, int min_pcnt, int max_lines)
1446{
1447	struct dso *dso = map->dso;
1448	struct rb_root source_line = RB_ROOT;
1449	u64 len;
1450
1451	if (symbol__annotate(sym, map, 0) < 0)
1452		return -1;
1453
1454	len = symbol__size(sym);
1455
1456	if (print_lines) {
1457		symbol__get_source_line(sym, map, evsel, &source_line, len);
1458		print_summary(&source_line, dso->long_name);
1459	}
1460
1461	symbol__annotate_printf(sym, map, evsel, full_paths,
1462				min_pcnt, max_lines, 0);
1463	if (print_lines)
1464		symbol__free_source_line(sym, len);
1465
1466	disasm__purge(&symbol__annotation(sym)->src->source);
1467
1468	return 0;
1469}
1470
1471int hist_entry__annotate(struct hist_entry *he, size_t privsize)
1472{
1473	return symbol__annotate(he->ms.sym, he->ms.map, privsize);
1474}
1475
1476bool ui__has_annotation(void)
1477{
1478	return use_browser == 1 && sort__has_sym;
1479}
1480