1 // SPDX-License-Identifier: GPL-2.0-only
2 #include <ctype.h>
3 #include <elf.h>
4 #ifndef EF_CSKY_ABIMASK
5 #define EF_CSKY_ABIMASK 0XF0000000
6 #endif
7 #include <errno.h>
8 #include <fcntl.h>
9 #include <inttypes.h>
10 #include <libgen.h>
11 #include <regex.h>
12 #include <stdlib.h>
13 #include <unistd.h>
14
15 #include <linux/string.h>
16 #include <linux/zalloc.h>
17 #include <subcmd/run-command.h>
18
19 #include "annotate.h"
20 #include "annotate-data.h"
21 #include "build-id.h"
22 #include "capstone.h"
23 #include "debug.h"
24 #include "disasm.h"
25 #include "dso.h"
26 #include "dwarf-regs.h"
27 #include "env.h"
28 #include "evsel.h"
29 #include "libbfd.h"
30 #include "llvm.h"
31 #include "map.h"
32 #include "maps.h"
33 #include "namespaces.h"
34 #include "srcline.h"
35 #include "symbol.h"
36 #include "thread.h"
37 #include "util.h"
38
39 static regex_t file_lineno;
40
41 /* These can be referred from the arch-dependent code */
42 const struct ins_ops call_ops;
43 const struct ins_ops dec_ops;
44 const struct ins_ops jump_ops;
45 const struct ins_ops mov_ops;
46 const struct ins_ops nop_ops;
47 const struct ins_ops lock_ops;
48 const struct ins_ops ret_ops;
49 const struct ins_ops load_store_ops;
50 const struct ins_ops arithmetic_ops;
51
52 static void ins__sort(struct arch *arch);
53 static int disasm_line__parse(char *line, const char **namep, char **rawp);
54 static int disasm_line__parse_powerpc(struct disasm_line *dl, struct annotate_args *args);
55
symbol__init_regexpr(void)56 static __attribute__((constructor)) void symbol__init_regexpr(void)
57 {
58 regcomp(&file_lineno, "^/[^:]+:([0-9]+)", REG_EXTENDED);
59 }
60
arch__grow_instructions(struct arch * arch)61 static int arch__grow_instructions(struct arch *arch)
62 {
63 struct ins *new_instructions;
64 size_t new_nr_allocated;
65
66 if (arch->nr_instructions_allocated == 0 && arch->instructions)
67 goto grow_from_non_allocated_table;
68
69 new_nr_allocated = arch->nr_instructions_allocated + 128;
70 new_instructions = realloc((void *)arch->instructions,
71 new_nr_allocated * sizeof(struct ins));
72 if (new_instructions == NULL)
73 return -1;
74
75 out_update_instructions:
76 arch->instructions = new_instructions;
77 arch->nr_instructions_allocated = new_nr_allocated;
78 return 0;
79
80 grow_from_non_allocated_table:
81 new_nr_allocated = arch->nr_instructions + 128;
82 new_instructions = calloc(new_nr_allocated, sizeof(struct ins));
83 if (new_instructions == NULL)
84 return -1;
85
86 memcpy(new_instructions, arch->instructions, arch->nr_instructions * sizeof(struct ins));
87 goto out_update_instructions;
88 }
89
arch__associate_ins_ops(struct arch * arch,const char * name,const struct ins_ops * ops)90 int arch__associate_ins_ops(struct arch *arch, const char *name, const struct ins_ops *ops)
91 {
92 struct ins *ins;
93
94 if (arch->nr_instructions == arch->nr_instructions_allocated &&
95 arch__grow_instructions(arch))
96 return -1;
97
98 ins = (struct ins *)&arch->instructions[arch->nr_instructions];
99 ins->name = strdup(name);
100 if (!ins->name)
101 return -1;
102
103 ins->ops = ops;
104 arch->nr_instructions++;
105
106 ins__sort(arch);
107 return 0;
108 }
109
e_machine_and_eflags__cmp(const struct e_machine_and_e_flags * val1,const struct e_machine_and_e_flags * val2)110 static int e_machine_and_eflags__cmp(const struct e_machine_and_e_flags *val1,
111 const struct e_machine_and_e_flags *val2)
112 {
113 if (val1->e_machine == val2->e_machine) {
114 if (val1->e_machine != EM_CSKY)
115 return 0;
116 if ((val1->e_flags & EF_CSKY_ABIMASK) < (val2->e_flags & EF_CSKY_ABIMASK))
117 return -1;
118 return (val1->e_flags & EF_CSKY_ABIMASK) > (val2->e_flags & EF_CSKY_ABIMASK);
119 }
120 return val1->e_machine < val2->e_machine ? -1 : 1;
121 }
122
arch__key_cmp(const void * key,const void * archp)123 static int arch__key_cmp(const void *key, const void *archp)
124 {
125 const struct arch *const *arch = archp;
126
127 return e_machine_and_eflags__cmp(key, &(*arch)->id);
128 }
129
arch__cmp(const void * a,const void * b)130 static int arch__cmp(const void *a, const void *b)
131 {
132 const struct arch *const *aa = a;
133 const struct arch *const *ab = b;
134
135 return e_machine_and_eflags__cmp(&(*aa)->id, &(*ab)->id);
136 }
137
arch__find(uint16_t e_machine,uint32_t e_flags,const char * cpuid)138 const struct arch *arch__find(uint16_t e_machine, uint32_t e_flags, const char *cpuid)
139 {
140 static const struct arch *(*const arch_new_fn[])(const struct e_machine_and_e_flags *id,
141 const char *cpuid) = {
142 [EM_386] = arch__new_x86,
143 [EM_ARC] = arch__new_arc,
144 [EM_ARM] = arch__new_arm,
145 [EM_AARCH64] = arch__new_arm64,
146 [EM_CSKY] = arch__new_csky,
147 [EM_LOONGARCH] = arch__new_loongarch,
148 [EM_MIPS] = arch__new_mips,
149 [EM_PPC64] = arch__new_powerpc,
150 [EM_PPC] = arch__new_powerpc,
151 [EM_RISCV] = arch__new_riscv64,
152 [EM_S390] = arch__new_s390,
153 [EM_SPARC] = arch__new_sparc,
154 [EM_SPARCV9] = arch__new_sparc,
155 [EM_X86_64] = arch__new_x86,
156 };
157 static const struct arch **archs;
158 static size_t num_archs;
159 struct e_machine_and_e_flags key = {
160 .e_machine = e_machine,
161 .e_flags = e_flags,
162 };
163 const struct arch *result = NULL, **tmp;
164
165 if (num_archs > 0) {
166 tmp = bsearch(&key, archs, num_archs, sizeof(*archs), arch__key_cmp);
167 if (tmp)
168 result = *tmp;
169 }
170
171 if (result)
172 return result;
173
174 if (e_machine >= ARRAY_SIZE(arch_new_fn) || arch_new_fn[e_machine] == NULL) {
175 errno = ENOTSUP;
176 return NULL;
177 }
178
179 tmp = reallocarray(archs, num_archs + 1, sizeof(*archs));
180 if (!tmp)
181 return NULL;
182
183 result = arch_new_fn[e_machine](&key, cpuid);
184 if (!result) {
185 pr_err("%s: failed to initialize %s (%u) arch priv area\n",
186 __func__, result->name, e_machine);
187 free(tmp);
188 return NULL;
189 }
190 archs = tmp;
191 archs[num_archs++] = result;
192 qsort(archs, num_archs, sizeof(*archs), arch__cmp);
193 return result;
194 }
195
arch__is_x86(const struct arch * arch)196 bool arch__is_x86(const struct arch *arch)
197 {
198 return arch->id.e_machine == EM_386 || arch->id.e_machine == EM_X86_64;
199 }
200
arch__is_powerpc(const struct arch * arch)201 bool arch__is_powerpc(const struct arch *arch)
202 {
203 return arch->id.e_machine == EM_PPC || arch->id.e_machine == EM_PPC64;
204 }
205
ins_ops__delete(struct ins_operands * ops)206 static void ins_ops__delete(struct ins_operands *ops)
207 {
208 if (ops == NULL)
209 return;
210 zfree(&ops->source.raw);
211 zfree(&ops->source.name);
212 zfree(&ops->target.raw);
213 zfree(&ops->target.name);
214 }
215
ins__raw_scnprintf(const struct ins * ins,char * bf,size_t size,struct ins_operands * ops,int max_ins_name)216 int ins__raw_scnprintf(const struct ins *ins, char *bf, size_t size,
217 struct ins_operands *ops, int max_ins_name)
218 {
219 return scnprintf(bf, size, "%-*s %s", max_ins_name, ins->name, ops->raw);
220 }
221
ins__scnprintf(const struct ins * ins,char * bf,size_t size,struct ins_operands * ops,int max_ins_name)222 int ins__scnprintf(const struct ins *ins, char *bf, size_t size,
223 struct ins_operands *ops, int max_ins_name)
224 {
225 if (ins->ops->scnprintf)
226 return ins->ops->scnprintf(ins, bf, size, ops, max_ins_name);
227
228 return ins__raw_scnprintf(ins, bf, size, ops, max_ins_name);
229 }
230
ins__is_fused(const struct arch * arch,const char * ins1,const char * ins2)231 bool ins__is_fused(const struct arch *arch, const char *ins1, const char *ins2)
232 {
233 if (!arch || !arch->ins_is_fused)
234 return false;
235
236 return arch->ins_is_fused(arch, ins1, ins2);
237 }
238
call__parse(const struct arch * arch,struct ins_operands * ops,struct map_symbol * ms,struct disasm_line * dl __maybe_unused)239 static int call__parse(const struct arch *arch, struct ins_operands *ops, struct map_symbol *ms,
240 struct disasm_line *dl __maybe_unused)
241 {
242 char *endptr, *tok, *name;
243 struct map *map = ms->map;
244 struct addr_map_symbol target;
245
246 ops->target.addr = strtoull(ops->raw, &endptr, 16);
247
248 name = strchr(endptr, '<');
249 if (name == NULL)
250 goto indirect_call;
251
252 name++;
253
254 if (arch->objdump.skip_functions_char &&
255 strchr(name, arch->objdump.skip_functions_char))
256 return -1;
257
258 tok = strchr(name, '>');
259 if (tok == NULL)
260 return -1;
261
262 *tok = '\0';
263 ops->target.name = strdup(name);
264 *tok = '>';
265
266 if (ops->target.name == NULL)
267 return -1;
268 find_target:
269 target = (struct addr_map_symbol) {
270 .ms = { .map = map__get(map), },
271 .addr = map__objdump_2mem(map, ops->target.addr),
272 };
273
274 if (maps__find_ams(thread__maps(ms->thread), &target) == 0 &&
275 map__rip_2objdump(target.ms.map, map__map_ip(target.ms.map, target.addr)) == ops->target.addr)
276 ops->target.sym = target.ms.sym;
277
278 addr_map_symbol__exit(&target);
279 return 0;
280
281 indirect_call:
282 tok = strchr(endptr, '*');
283 if (tok != NULL) {
284 endptr++;
285
286 /* Indirect call can use a non-rip register and offset: callq *0x8(%rbx).
287 * Do not parse such instruction. */
288 if (strstr(endptr, "(%r") == NULL)
289 ops->target.addr = strtoull(endptr, NULL, 16);
290 }
291 goto find_target;
292 }
293
call__scnprintf(const struct ins * ins,char * bf,size_t size,struct ins_operands * ops,int max_ins_name)294 int call__scnprintf(const struct ins *ins, char *bf, size_t size,
295 struct ins_operands *ops, int max_ins_name)
296 {
297 if (ops->target.sym)
298 return scnprintf(bf, size, "%-*s %s", max_ins_name, ins->name, ops->target.sym->name);
299
300 if (ops->target.addr == 0)
301 return ins__raw_scnprintf(ins, bf, size, ops, max_ins_name);
302
303 if (ops->target.name)
304 return scnprintf(bf, size, "%-*s %s", max_ins_name, ins->name, ops->target.name);
305
306 return scnprintf(bf, size, "%-*s *%" PRIx64, max_ins_name, ins->name, ops->target.addr);
307 }
308
309 const struct ins_ops call_ops = {
310 .parse = call__parse,
311 .scnprintf = call__scnprintf,
312 .is_call = true,
313 };
314
ins__is_call(const struct ins * ins)315 bool ins__is_call(const struct ins *ins)
316 {
317 return ins->ops && ins->ops->is_call;
318 }
319
320 /*
321 * Prevents from matching commas in the comment section, e.g.:
322 * ffff200008446e70: b.cs ffff2000084470f4 <generic_exec_single+0x314> // b.hs, b.nlast
323 *
324 * and skip comma as part of function arguments, e.g.:
325 * 1d8b4ac <linemap_lookup(line_maps const*, unsigned int)+0xcc>
326 */
validate_comma(const char * c,struct ins_operands * ops)327 static inline const char *validate_comma(const char *c, struct ins_operands *ops)
328 {
329 if (ops->jump.raw_comment && c > ops->jump.raw_comment)
330 return NULL;
331
332 if (ops->jump.raw_func_start && c > ops->jump.raw_func_start)
333 return NULL;
334
335 return c;
336 }
337
jump__parse(const struct arch * arch,struct ins_operands * ops,struct map_symbol * ms,struct disasm_line * dl __maybe_unused)338 static int jump__parse(const struct arch *arch, struct ins_operands *ops, struct map_symbol *ms,
339 struct disasm_line *dl __maybe_unused)
340 {
341 struct map *map = ms->map;
342 struct symbol *sym = ms->sym;
343 struct addr_map_symbol target = {
344 .ms = { .map = map__get(map), },
345 };
346 const char *c = strchr(ops->raw, ',');
347 u64 start, end;
348
349 ops->jump.raw_comment = strchr(ops->raw, arch->objdump.comment_char);
350 ops->jump.raw_func_start = strchr(ops->raw, '<');
351
352 c = validate_comma(c, ops);
353
354 /*
355 * Examples of lines to parse for the _cpp_lex_token@@Base
356 * function:
357 *
358 * 1159e6c: jne 115aa32 <_cpp_lex_token@@Base+0xf92>
359 * 1159e8b: jne c469be <cpp_named_operator2name@@Base+0xa72>
360 *
361 * The first is a jump to an offset inside the same function,
362 * the second is to another function, i.e. that 0xa72 is an
363 * offset in the cpp_named_operator2name@@base function.
364 */
365 /*
366 * skip over possible up to 2 operands to get to address, e.g.:
367 * tbnz w0, #26, ffff0000083cd190 <security_file_permission+0xd0>
368 */
369 if (c != NULL) {
370 c++;
371 ops->target.addr = strtoull(c, NULL, 16);
372 if (!ops->target.addr) {
373 c = strchr(c, ',');
374 c = validate_comma(c, ops);
375 if (c != NULL) {
376 c++;
377 ops->target.addr = strtoull(c, NULL, 16);
378 }
379 }
380 } else {
381 ops->target.addr = strtoull(ops->raw, NULL, 16);
382 }
383
384 target.addr = map__objdump_2mem(map, ops->target.addr);
385 start = map__unmap_ip(map, sym->start);
386 end = map__unmap_ip(map, sym->end);
387
388 ops->target.outside = target.addr < start || target.addr >= end;
389
390 /*
391 * FIXME: things like this in _cpp_lex_token (gcc's cc1 program):
392
393 cpp_named_operator2name@@Base+0xa72
394
395 * Point to a place that is after the cpp_named_operator2name
396 * boundaries, i.e. in the ELF symbol table for cc1
397 * cpp_named_operator2name is marked as being 32-bytes long, but it in
398 * fact is much larger than that, so we seem to need a symbols__find()
399 * routine that looks for >= current->start and < next_symbol->start,
400 * possibly just for C++ objects?
401 *
402 * For now lets just make some progress by marking jumps to outside the
403 * current function as call like.
404 *
405 * Actual navigation will come next, with further understanding of how
406 * the symbol searching and disassembly should be done.
407 */
408 if (maps__find_ams(thread__maps(ms->thread), &target) == 0 &&
409 map__rip_2objdump(target.ms.map, map__map_ip(target.ms.map, target.addr)) == ops->target.addr)
410 ops->target.sym = target.ms.sym;
411
412 if (!ops->target.outside) {
413 ops->target.offset = target.addr - start;
414 ops->target.offset_avail = true;
415 } else {
416 ops->target.offset_avail = false;
417 }
418 addr_map_symbol__exit(&target);
419 return 0;
420 }
421
jump__scnprintf(const struct ins * ins,char * bf,size_t size,struct ins_operands * ops,int max_ins_name)422 int jump__scnprintf(const struct ins *ins, char *bf, size_t size,
423 struct ins_operands *ops, int max_ins_name)
424 {
425 const char *c;
426
427 if (!ops->target.addr || ops->target.offset < 0)
428 return ins__raw_scnprintf(ins, bf, size, ops, max_ins_name);
429
430 if (ops->target.outside && ops->target.sym != NULL)
431 return scnprintf(bf, size, "%-*s %s", max_ins_name, ins->name, ops->target.sym->name);
432
433 c = strchr(ops->raw, ',');
434 c = validate_comma(c, ops);
435
436 if (c != NULL) {
437 const char *c2 = strchr(c + 1, ',');
438
439 c2 = validate_comma(c2, ops);
440 /* check for 3-op insn */
441 if (c2 != NULL)
442 c = c2;
443 c++;
444
445 /* mirror arch objdump's space-after-comma style */
446 if (*c == ' ')
447 c++;
448 }
449
450 return scnprintf(bf, size, "%-*s %.*s%" PRIx64, max_ins_name,
451 ins->name, c ? c - ops->raw : 0, ops->raw,
452 ops->target.offset);
453 }
454
jump__delete(struct ins_operands * ops __maybe_unused)455 void jump__delete(struct ins_operands *ops __maybe_unused)
456 {
457 /*
458 * The ops->jump.raw_comment and ops->jump.raw_func_start belong to the
459 * raw string, don't free them.
460 */
461 }
462
463 const struct ins_ops jump_ops = {
464 .free = jump__delete,
465 .parse = jump__parse,
466 .scnprintf = jump__scnprintf,
467 .is_jump = true,
468 };
469
ins__is_jump(const struct ins * ins)470 bool ins__is_jump(const struct ins *ins)
471 {
472 return ins->ops && ins->ops->is_jump;
473 }
474
comment__symbol(char * raw,char * comment,u64 * addrp,char ** namep)475 static int comment__symbol(char *raw, char *comment, u64 *addrp, char **namep)
476 {
477 char *endptr, *name, *t;
478
479 if (strstr(raw, "(%rip)") == NULL)
480 return 0;
481
482 *addrp = strtoull(comment, &endptr, 16);
483 if (endptr == comment)
484 return 0;
485 name = strchr(endptr, '<');
486 if (name == NULL)
487 return -1;
488
489 name++;
490
491 t = strchr(name, '>');
492 if (t == NULL)
493 return 0;
494
495 *t = '\0';
496 *namep = strdup(name);
497 *t = '>';
498
499 return 0;
500 }
501
lock__parse(const struct arch * arch,struct ins_operands * ops,struct map_symbol * ms,struct disasm_line * dl __maybe_unused)502 static int lock__parse(const struct arch *arch, struct ins_operands *ops, struct map_symbol *ms,
503 struct disasm_line *dl __maybe_unused)
504 {
505 ops->locked.ops = zalloc(sizeof(*ops->locked.ops));
506 if (ops->locked.ops == NULL)
507 return 0;
508
509 if (disasm_line__parse(ops->raw, &ops->locked.ins.name, &ops->locked.ops->raw) < 0)
510 goto out_free_ops;
511
512 ops->locked.ins.ops = ins__find(arch, ops->locked.ins.name, 0);
513
514 if (ops->locked.ins.ops == NULL)
515 goto out_free_ops;
516
517 if (ops->locked.ins.ops->parse &&
518 ops->locked.ins.ops->parse(arch, ops->locked.ops, ms, NULL) < 0)
519 goto out_free_ops;
520
521 return 0;
522
523 out_free_ops:
524 zfree(&ops->locked.ops);
525 return 0;
526 }
527
lock__scnprintf(const struct ins * ins,char * bf,size_t size,struct ins_operands * ops,int max_ins_name)528 static int lock__scnprintf(const struct ins *ins, char *bf, size_t size,
529 struct ins_operands *ops, int max_ins_name)
530 {
531 int printed;
532
533 if (ops->locked.ins.ops == NULL)
534 return ins__raw_scnprintf(ins, bf, size, ops, max_ins_name);
535
536 printed = scnprintf(bf, size, "%-*s ", max_ins_name, ins->name);
537 return printed + ins__scnprintf(&ops->locked.ins, bf + printed,
538 size - printed, ops->locked.ops, max_ins_name);
539 }
540
lock__delete(struct ins_operands * ops)541 static void lock__delete(struct ins_operands *ops)
542 {
543 struct ins *ins = &ops->locked.ins;
544
545 if (ins->ops && ins->ops->free)
546 ins->ops->free(ops->locked.ops);
547 else
548 ins_ops__delete(ops->locked.ops);
549
550 zfree(&ops->locked.ops);
551 zfree(&ops->locked.ins.name);
552 zfree(&ops->target.raw);
553 zfree(&ops->target.name);
554 }
555
556 const struct ins_ops lock_ops = {
557 .free = lock__delete,
558 .parse = lock__parse,
559 .scnprintf = lock__scnprintf,
560 };
561
562 /*
563 * Check if the operand has more than one registers like x86 SIB addressing:
564 * 0x1234(%rax, %rbx, 8)
565 *
566 * But it doesn't care segment selectors like %gs:0x5678(%rcx), so just check
567 * the input string after 'memory_ref_char' if exists.
568 */
check_multi_regs(const struct arch * arch,const char * op)569 static bool check_multi_regs(const struct arch *arch, const char *op)
570 {
571 int count = 0;
572
573 if (arch->objdump.register_char == 0)
574 return false;
575
576 if (arch->objdump.memory_ref_char) {
577 op = strchr(op, arch->objdump.memory_ref_char);
578 if (op == NULL)
579 return false;
580 }
581
582 while ((op = strchr(op, arch->objdump.register_char)) != NULL) {
583 count++;
584 op++;
585 }
586
587 return count > 1;
588 }
589
mov__parse(const struct arch * arch,struct ins_operands * ops,struct map_symbol * ms __maybe_unused,struct disasm_line * dl __maybe_unused)590 static int mov__parse(const struct arch *arch, struct ins_operands *ops,
591 struct map_symbol *ms __maybe_unused,
592 struct disasm_line *dl __maybe_unused)
593 {
594 char *s = strchr(ops->raw, ','), *target, *comment, prev;
595
596 if (s == NULL)
597 return -1;
598
599 *s = '\0';
600
601 /*
602 * x86 SIB addressing has something like 0x8(%rax, %rcx, 1)
603 * then it needs to have the closing parenthesis.
604 */
605 if (strchr(ops->raw, '(')) {
606 *s = ',';
607 s = strchr(ops->raw, ')');
608 if (s == NULL || s[1] != ',')
609 return -1;
610 *++s = '\0';
611 }
612
613 ops->source.raw = strdup(ops->raw);
614 *s = ',';
615
616 if (ops->source.raw == NULL)
617 return -1;
618
619 ops->source.multi_regs = check_multi_regs(arch, ops->source.raw);
620
621 target = skip_spaces(++s);
622 comment = strchr(s, arch->objdump.comment_char);
623
624 if (comment != NULL)
625 s = comment - 1;
626 else
627 s = strchr(s, '\0') - 1;
628
629 while (s > target && isspace(s[0]))
630 --s;
631 s++;
632 prev = *s;
633 *s = '\0';
634
635 ops->target.raw = strdup(target);
636 *s = prev;
637
638 if (ops->target.raw == NULL)
639 goto out_free_source;
640
641 ops->target.multi_regs = check_multi_regs(arch, ops->target.raw);
642
643 if (comment == NULL)
644 return 0;
645
646 comment = skip_spaces(comment);
647 comment__symbol(ops->source.raw, comment + 1, &ops->source.addr, &ops->source.name);
648 comment__symbol(ops->target.raw, comment + 1, &ops->target.addr, &ops->target.name);
649
650 return 0;
651
652 out_free_source:
653 zfree(&ops->source.raw);
654 return -1;
655 }
656
mov__scnprintf(const struct ins * ins,char * bf,size_t size,struct ins_operands * ops,int max_ins_name)657 int mov__scnprintf(const struct ins *ins, char *bf, size_t size,
658 struct ins_operands *ops, int max_ins_name)
659 {
660 return scnprintf(bf, size, "%-*s %s,%s", max_ins_name, ins->name,
661 ops->source.name ?: ops->source.raw,
662 ops->target.name ?: ops->target.raw);
663 }
664
665 const struct ins_ops mov_ops = {
666 .parse = mov__parse,
667 .scnprintf = mov__scnprintf,
668 };
669
dec__parse(const struct arch * arch __maybe_unused,struct ins_operands * ops,struct map_symbol * ms __maybe_unused,struct disasm_line * dl __maybe_unused)670 static int dec__parse(const struct arch *arch __maybe_unused, struct ins_operands *ops,
671 struct map_symbol *ms __maybe_unused,
672 struct disasm_line *dl __maybe_unused)
673 {
674 char *target, *comment, *s, prev;
675
676 target = s = ops->raw;
677
678 while (s[0] != '\0' && !isspace(s[0]))
679 ++s;
680 prev = *s;
681 *s = '\0';
682
683 ops->target.raw = strdup(target);
684 *s = prev;
685
686 if (ops->target.raw == NULL)
687 return -1;
688
689 comment = strchr(s, arch->objdump.comment_char);
690 if (comment == NULL)
691 return 0;
692
693 comment = skip_spaces(comment);
694 comment__symbol(ops->target.raw, comment + 1, &ops->target.addr, &ops->target.name);
695
696 return 0;
697 }
698
dec__scnprintf(const struct ins * ins,char * bf,size_t size,struct ins_operands * ops,int max_ins_name)699 static int dec__scnprintf(const struct ins *ins, char *bf, size_t size,
700 struct ins_operands *ops, int max_ins_name)
701 {
702 return scnprintf(bf, size, "%-*s %s", max_ins_name, ins->name,
703 ops->target.name ?: ops->target.raw);
704 }
705
706 const struct ins_ops dec_ops = {
707 .parse = dec__parse,
708 .scnprintf = dec__scnprintf,
709 };
710
nop__scnprintf(const struct ins * ins __maybe_unused,char * bf,size_t size,struct ins_operands * ops __maybe_unused,int max_ins_name)711 static int nop__scnprintf(const struct ins *ins __maybe_unused, char *bf, size_t size,
712 struct ins_operands *ops __maybe_unused, int max_ins_name)
713 {
714 return scnprintf(bf, size, "%-*s", max_ins_name, "nop");
715 }
716
717 const struct ins_ops nop_ops = {
718 .scnprintf = nop__scnprintf,
719 };
720
721 const struct ins_ops ret_ops = {
722 .scnprintf = ins__raw_scnprintf,
723 };
724
ins__is_nop(const struct ins * ins)725 static bool ins__is_nop(const struct ins *ins)
726 {
727 return ins->ops == &nop_ops;
728 }
729
ins__is_ret(const struct ins * ins)730 bool ins__is_ret(const struct ins *ins)
731 {
732 return ins->ops == &ret_ops;
733 }
734
ins__is_lock(const struct ins * ins)735 bool ins__is_lock(const struct ins *ins)
736 {
737 return ins->ops == &lock_ops;
738 }
739
ins__key_cmp(const void * name,const void * insp)740 static int ins__key_cmp(const void *name, const void *insp)
741 {
742 const struct ins *ins = insp;
743
744 return strcmp(name, ins->name);
745 }
746
ins__cmp(const void * a,const void * b)747 static int ins__cmp(const void *a, const void *b)
748 {
749 const struct ins *ia = a;
750 const struct ins *ib = b;
751
752 return strcmp(ia->name, ib->name);
753 }
754
ins__sort(struct arch * arch)755 static void ins__sort(struct arch *arch)
756 {
757 const int nmemb = arch->nr_instructions;
758
759 qsort((void *)arch->instructions, nmemb, sizeof(struct ins), ins__cmp);
760 }
761
__ins__find(const struct arch * arch,const char * name,struct disasm_line * dl)762 static const struct ins_ops *__ins__find(const struct arch *arch, const char *name,
763 struct disasm_line *dl)
764 {
765 const struct ins *ins;
766 const int nmemb = arch->nr_instructions;
767
768 if (arch__is_powerpc(arch)) {
769 /*
770 * For powerpc, identify the instruction ops
771 * from the opcode using raw_insn.
772 */
773 const struct ins_ops *ops;
774
775 ops = check_ppc_insn(dl);
776 if (ops)
777 return ops;
778 }
779
780 if (!arch->sorted_instructions) {
781 ins__sort((struct arch *)arch);
782 ((struct arch *)arch)->sorted_instructions = true;
783 }
784
785 ins = bsearch(name, arch->instructions, nmemb, sizeof(struct ins), ins__key_cmp);
786 if (ins)
787 return ins->ops;
788
789 if (arch->insn_suffix) {
790 char tmp[32];
791 char suffix;
792 size_t len = strlen(name);
793
794 if (len == 0 || len >= sizeof(tmp))
795 return NULL;
796
797 suffix = name[len - 1];
798 if (strchr(arch->insn_suffix, suffix) == NULL)
799 return NULL;
800
801 strcpy(tmp, name);
802 tmp[len - 1] = '\0'; /* remove the suffix and check again */
803
804 ins = bsearch(tmp, arch->instructions, nmemb, sizeof(struct ins), ins__key_cmp);
805 }
806 return ins ? ins->ops : NULL;
807 }
808
ins__find(const struct arch * arch,const char * name,struct disasm_line * dl)809 const struct ins_ops *ins__find(const struct arch *arch, const char *name, struct disasm_line *dl)
810 {
811 const struct ins_ops *ops = __ins__find(arch, name, dl);
812
813 if (!ops && arch->associate_instruction_ops)
814 ops = arch->associate_instruction_ops((struct arch *)arch, name);
815
816 return ops;
817 }
818
disasm_line__init_ins(struct disasm_line * dl,const struct arch * arch,struct map_symbol * ms)819 static void disasm_line__init_ins(struct disasm_line *dl, const struct arch *arch,
820 struct map_symbol *ms)
821 {
822 dl->ins.ops = ins__find(arch, dl->ins.name, dl);
823
824 if (!dl->ins.ops)
825 return;
826
827 if (dl->ins.ops->parse && dl->ins.ops->parse(arch, &dl->ops, ms, dl) < 0)
828 dl->ins.ops = NULL;
829 }
830
disasm_line__parse(char * line,const char ** namep,char ** rawp)831 static int disasm_line__parse(char *line, const char **namep, char **rawp)
832 {
833 char tmp, *name = skip_spaces(line);
834
835 if (name[0] == '\0')
836 return -1;
837
838 *rawp = name + 1;
839
840 while ((*rawp)[0] != '\0' && !isspace((*rawp)[0]))
841 ++*rawp;
842
843 tmp = (*rawp)[0];
844 (*rawp)[0] = '\0';
845 *namep = strdup(name);
846
847 if (*namep == NULL)
848 goto out;
849
850 (*rawp)[0] = tmp;
851 *rawp = strim(*rawp);
852
853 return 0;
854
855 out:
856 return -1;
857 }
858
859 /*
860 * Parses the result captured from symbol__disassemble_*
861 * Example, line read from DSO file in powerpc:
862 * line: 38 01 81 e8
863 * opcode: fetched from arch specific get_opcode_insn
864 * rawp_insn: e8810138
865 *
866 * rawp_insn is used later to extract the reg/offset fields
867 */
868 #define PPC_OP(op) (((op) >> 26) & 0x3F)
869 #define RAW_BYTES 11
870
disasm_line__parse_powerpc(struct disasm_line * dl,struct annotate_args * args)871 static int disasm_line__parse_powerpc(struct disasm_line *dl, struct annotate_args *args)
872 {
873 char *line = dl->al.line;
874 const char **namep = &dl->ins.name;
875 char **rawp = &dl->ops.raw;
876 char *tmp_raw_insn, *name_raw_insn = skip_spaces(line);
877 char *name = skip_spaces(name_raw_insn + RAW_BYTES);
878 int disasm = 0;
879 int ret = 0;
880
881 if (args->options->disassembler_used)
882 disasm = 1;
883
884 if (name_raw_insn[0] == '\0')
885 return -1;
886
887 if (disasm)
888 ret = disasm_line__parse(name, namep, rawp);
889 else
890 *namep = "";
891
892 tmp_raw_insn = strndup(name_raw_insn, 11);
893 if (tmp_raw_insn == NULL)
894 return -1;
895
896 remove_spaces(tmp_raw_insn);
897
898 sscanf(tmp_raw_insn, "%x", &dl->raw.raw_insn);
899 if (disasm)
900 dl->raw.raw_insn = be32_to_cpu(dl->raw.raw_insn);
901
902 return ret;
903 }
904
annotation_line__init(struct annotation_line * al,struct annotate_args * args,int nr)905 static void annotation_line__init(struct annotation_line *al,
906 struct annotate_args *args,
907 int nr)
908 {
909 al->offset = args->offset;
910 al->line = strdup(args->line);
911 al->line_nr = args->line_nr;
912 al->fileloc = args->fileloc ? strdup(args->fileloc) : NULL;
913 al->data_nr = nr;
914 }
915
annotation_line__exit(struct annotation_line * al)916 static void annotation_line__exit(struct annotation_line *al)
917 {
918 zfree_srcline(&al->path);
919 zfree(&al->fileloc);
920 zfree(&al->line);
921 zfree(&al->cycles);
922 zfree(&al->br_cntr);
923 }
924
disasm_line_size(int nr)925 static size_t disasm_line_size(int nr)
926 {
927 struct annotation_line *al;
928
929 return (sizeof(struct disasm_line) + (sizeof(al->data[0]) * nr));
930 }
931
932 /*
933 * Allocating the disasm annotation line data with
934 * following structure:
935 *
936 * -------------------------------------------
937 * struct disasm_line | struct annotation_line
938 * -------------------------------------------
939 *
940 * We have 'struct annotation_line' member as last member
941 * of 'struct disasm_line' to have an easy access.
942 */
disasm_line__new(struct annotate_args * args)943 struct disasm_line *disasm_line__new(struct annotate_args *args)
944 {
945 struct disasm_line *dl = NULL;
946 struct annotation *notes = symbol__annotation(args->ms->sym);
947 int nr = notes->src->nr_events;
948
949 dl = zalloc(disasm_line_size(nr));
950 if (!dl)
951 return NULL;
952
953 annotation_line__init(&dl->al, args, nr);
954 if (dl->al.line == NULL)
955 goto out_free_line;
956
957 if (args->offset != -1) {
958 if (arch__is_powerpc(args->arch)) {
959 if (disasm_line__parse_powerpc(dl, args) < 0)
960 goto out_free_line;
961 } else if (disasm_line__parse(dl->al.line, &dl->ins.name, &dl->ops.raw) < 0)
962 goto out_free_line;
963
964 disasm_line__init_ins(dl, args->arch, args->ms);
965 }
966
967 return dl;
968
969 out_free_line:
970 annotation_line__exit(&dl->al);
971 free(dl);
972 return NULL;
973 }
974
disasm_line__free(struct disasm_line * dl)975 void disasm_line__free(struct disasm_line *dl)
976 {
977 if (dl->ins.ops && dl->ins.ops->free)
978 dl->ins.ops->free(&dl->ops);
979 else
980 ins_ops__delete(&dl->ops);
981 zfree(&dl->ins.name);
982 annotation_line__exit(&dl->al);
983 free(dl);
984 }
985
disasm_line__scnprintf(struct disasm_line * dl,char * bf,size_t size,bool raw,int max_ins_name)986 int disasm_line__scnprintf(struct disasm_line *dl, char *bf, size_t size, bool raw, int max_ins_name)
987 {
988 if (raw || !dl->ins.ops)
989 return scnprintf(bf, size, "%-*s %s", max_ins_name, dl->ins.name, dl->ops.raw);
990
991 return ins__scnprintf(&dl->ins, bf, size, &dl->ops, max_ins_name);
992 }
993
994 /*
995 * symbol__parse_objdump_line() parses objdump output (with -d --no-show-raw)
996 * which looks like following
997 *
998 * 0000000000415500 <_init>:
999 * 415500: sub $0x8,%rsp
1000 * 415504: mov 0x2f5ad5(%rip),%rax # 70afe0 <_DYNAMIC+0x2f8>
1001 * 41550b: test %rax,%rax
1002 * 41550e: je 415515 <_init+0x15>
1003 * 415510: callq 416e70 <__gmon_start__@plt>
1004 * 415515: add $0x8,%rsp
1005 * 415519: retq
1006 *
1007 * it will be parsed and saved into struct disasm_line as
1008 * <offset> <name> <ops.raw>
1009 *
1010 * The offset will be a relative offset from the start of the symbol and -1
1011 * means that it's not a disassembly line so should be treated differently.
1012 * The ops.raw part will be parsed further according to type of the instruction.
1013 */
symbol__parse_objdump_line(struct symbol * sym,struct annotate_args * args,char * parsed_line,int * line_nr,char ** fileloc)1014 static int symbol__parse_objdump_line(struct symbol *sym,
1015 struct annotate_args *args,
1016 char *parsed_line, int *line_nr, char **fileloc)
1017 {
1018 struct map *map = args->ms->map;
1019 struct annotation *notes = symbol__annotation(sym);
1020 struct disasm_line *dl;
1021 char *tmp;
1022 s64 line_ip, offset = -1;
1023 regmatch_t match[2];
1024
1025 /* /filename:linenr ? Save line number and ignore. */
1026 if (regexec(&file_lineno, parsed_line, 2, match, 0) == 0) {
1027 *line_nr = atoi(parsed_line + match[1].rm_so);
1028 free(*fileloc);
1029 *fileloc = strdup(parsed_line);
1030 return 0;
1031 }
1032
1033 /* Process hex address followed by ':'. */
1034 line_ip = strtoull(parsed_line, &tmp, 16);
1035 if (parsed_line != tmp && tmp[0] == ':' && tmp[1] != '\0') {
1036 u64 start = map__rip_2objdump(map, sym->start),
1037 end = map__rip_2objdump(map, sym->end);
1038
1039 offset = line_ip - start;
1040 if ((u64)line_ip < start || (u64)line_ip >= end)
1041 offset = -1;
1042 else
1043 parsed_line = tmp + 1;
1044 }
1045
1046 args->offset = offset;
1047 args->line = parsed_line;
1048 args->line_nr = *line_nr;
1049 args->fileloc = *fileloc;
1050 args->ms->sym = sym;
1051
1052 dl = disasm_line__new(args);
1053 (*line_nr)++;
1054
1055 if (dl == NULL)
1056 return -1;
1057
1058 if (!disasm_line__has_local_offset(dl)) {
1059 dl->ops.target.offset = dl->ops.target.addr -
1060 map__rip_2objdump(map, sym->start);
1061 dl->ops.target.offset_avail = true;
1062 }
1063
1064 /* kcore has no symbols, so add the call target symbol */
1065 if (dl->ins.ops && ins__is_call(&dl->ins) && !dl->ops.target.sym) {
1066 struct addr_map_symbol target = {
1067 .addr = dl->ops.target.addr,
1068 .ms = { .map = map__get(map), },
1069 };
1070
1071 if (!maps__find_ams(thread__maps(args->ms->thread), &target) &&
1072 target.ms.sym->start == target.al_addr)
1073 dl->ops.target.sym = target.ms.sym;
1074
1075 addr_map_symbol__exit(&target);
1076 }
1077
1078 annotation_line__add(&dl->al, ¬es->src->source);
1079 return 0;
1080 }
1081
delete_last_nop(struct symbol * sym)1082 static void delete_last_nop(struct symbol *sym)
1083 {
1084 struct annotation *notes = symbol__annotation(sym);
1085 struct list_head *list = ¬es->src->source;
1086 struct disasm_line *dl;
1087
1088 while (!list_empty(list)) {
1089 dl = list_entry(list->prev, struct disasm_line, al.node);
1090
1091 if (dl->ins.ops) {
1092 if (!ins__is_nop(&dl->ins))
1093 return;
1094 } else {
1095 if (!strstr(dl->al.line, " nop ") &&
1096 !strstr(dl->al.line, " nopl ") &&
1097 !strstr(dl->al.line, " nopw "))
1098 return;
1099 }
1100
1101 list_del_init(&dl->al.node);
1102 disasm_line__free(dl);
1103 }
1104 }
1105
symbol__strerror_disassemble(struct map_symbol * ms,int errnum,char * buf,size_t buflen)1106 int symbol__strerror_disassemble(struct map_symbol *ms, int errnum, char *buf, size_t buflen)
1107 {
1108 struct dso *dso = map__dso(ms->map);
1109
1110 BUG_ON(buflen == 0);
1111
1112 if (errnum >= 0) {
1113 str_error_r(errnum, buf, buflen);
1114 return 0;
1115 }
1116
1117 switch (errnum) {
1118 case SYMBOL_ANNOTATE_ERRNO__NO_VMLINUX: {
1119 char bf[SBUILD_ID_SIZE + 15] = " with build id ";
1120 char *build_id_msg = NULL;
1121
1122 if (dso__has_build_id(dso)) {
1123 build_id__snprintf(dso__bid(dso), bf + 15, sizeof(bf) - 15);
1124 build_id_msg = bf;
1125 }
1126 scnprintf(buf, buflen,
1127 "No vmlinux file%s\nwas found in the path.\n\n"
1128 "Note that annotation using /proc/kcore requires CAP_SYS_RAWIO capability.\n\n"
1129 "Please use:\n\n"
1130 " perf buildid-cache -vu vmlinux\n\n"
1131 "or:\n\n"
1132 " --vmlinux vmlinux\n", build_id_msg ?: "");
1133 }
1134 break;
1135 case SYMBOL_ANNOTATE_ERRNO__NO_LIBOPCODES_FOR_BPF:
1136 scnprintf(buf, buflen, "Please link with binutils's libopcode to enable BPF annotation");
1137 break;
1138 case SYMBOL_ANNOTATE_ERRNO__ARCH_INIT_REGEXP:
1139 scnprintf(buf, buflen, "Problems with arch specific instruction name regular expressions.");
1140 break;
1141 case SYMBOL_ANNOTATE_ERRNO__ARCH_INIT_CPUID_PARSING:
1142 scnprintf(buf, buflen, "Problems while parsing the CPUID in the arch specific initialization.");
1143 break;
1144 case SYMBOL_ANNOTATE_ERRNO__BPF_INVALID_FILE:
1145 scnprintf(buf, buflen, "Invalid BPF file: %s.", dso__long_name(dso));
1146 break;
1147 case SYMBOL_ANNOTATE_ERRNO__BPF_MISSING_BTF:
1148 scnprintf(buf, buflen, "The %s BPF file has no BTF section, compile with -g or use pahole -J.",
1149 dso__long_name(dso));
1150 break;
1151 case SYMBOL_ANNOTATE_ERRNO__COULDNT_DETERMINE_FILE_TYPE:
1152 scnprintf(buf, buflen, "Couldn't determine the file %s type.", dso__long_name(dso));
1153 break;
1154 default:
1155 scnprintf(buf, buflen, "Internal error: Invalid %d error code\n", errnum);
1156 break;
1157 }
1158
1159 return 0;
1160 }
1161
dso__disassemble_filename(struct dso * dso,char * filename,size_t filename_size)1162 static int dso__disassemble_filename(struct dso *dso, char *filename, size_t filename_size)
1163 {
1164 char linkname[PATH_MAX];
1165 char *build_id_filename;
1166 char *build_id_path = NULL;
1167 char *pos;
1168 int len;
1169
1170 if (dso__symtab_type(dso) == DSO_BINARY_TYPE__KALLSYMS &&
1171 !dso__is_kcore(dso))
1172 return SYMBOL_ANNOTATE_ERRNO__NO_VMLINUX;
1173
1174 build_id_filename = dso__build_id_filename(dso, NULL, 0, false);
1175 if (build_id_filename) {
1176 __symbol__join_symfs(filename, filename_size, build_id_filename);
1177 free(build_id_filename);
1178 } else {
1179 if (dso__has_build_id(dso))
1180 return ENOMEM;
1181 goto fallback;
1182 }
1183
1184 build_id_path = strdup(filename);
1185 if (!build_id_path)
1186 return ENOMEM;
1187
1188 /*
1189 * old style build-id cache has name of XX/XXXXXXX.. while
1190 * new style has XX/XXXXXXX../{elf,kallsyms,vdso}.
1191 * extract the build-id part of dirname in the new style only.
1192 */
1193 pos = strrchr(build_id_path, '/');
1194 if (pos && strlen(pos) < SBUILD_ID_SIZE - 2)
1195 dirname(build_id_path);
1196
1197 if (dso__is_kcore(dso))
1198 goto fallback;
1199
1200 len = readlink(build_id_path, linkname, sizeof(linkname) - 1);
1201 if (len < 0)
1202 goto fallback;
1203
1204 linkname[len] = '\0';
1205 if (strstr(linkname, DSO__NAME_KALLSYMS) ||
1206 access(filename, R_OK)) {
1207 fallback:
1208 /*
1209 * If we don't have build-ids or the build-id file isn't in the
1210 * cache, or is just a kallsyms file, well, lets hope that this
1211 * DSO is the same as when 'perf record' ran.
1212 */
1213 if (dso__kernel(dso) && dso__long_name(dso)[0] == '/')
1214 snprintf(filename, filename_size, "%s", dso__long_name(dso));
1215 else
1216 __symbol__join_symfs(filename, filename_size, dso__long_name(dso));
1217
1218 mutex_lock(dso__lock(dso));
1219 if (access(filename, R_OK) && errno == ENOENT && dso__nsinfo(dso)) {
1220 char *new_name = dso__filename_with_chroot(dso, filename);
1221 if (new_name) {
1222 strlcpy(filename, new_name, filename_size);
1223 free(new_name);
1224 }
1225 }
1226 mutex_unlock(dso__lock(dso));
1227 } else if (dso__binary_type(dso) == DSO_BINARY_TYPE__NOT_FOUND) {
1228 dso__set_binary_type(dso, DSO_BINARY_TYPE__BUILD_ID_CACHE);
1229 }
1230
1231 free(build_id_path);
1232 return 0;
1233 }
1234
symbol__disassemble_raw(char * filename,struct symbol * sym,struct annotate_args * args)1235 static int symbol__disassemble_raw(char *filename, struct symbol *sym,
1236 struct annotate_args *args)
1237 {
1238 struct annotation *notes = symbol__annotation(sym);
1239 struct map *map = args->ms->map;
1240 struct dso *dso = map__dso(map);
1241 u64 start = map__rip_2objdump(map, sym->start);
1242 u64 end = map__rip_2objdump(map, sym->end);
1243 u64 len = end - start;
1244 u64 offset;
1245 int i, count;
1246 u8 *buf = NULL;
1247 char disasm_buf[512];
1248 struct disasm_line *dl;
1249 u32 *line;
1250
1251 /* Return if objdump is specified explicitly */
1252 if (args->options->objdump_path)
1253 return -1;
1254
1255 pr_debug("Reading raw instruction from : %s using dso__data_read_offset\n", filename);
1256
1257 buf = malloc(len);
1258 if (buf == NULL)
1259 goto err;
1260
1261 count = dso__data_read_offset(dso, NULL, sym->start, buf, len);
1262
1263 line = (u32 *)buf;
1264
1265 if ((u64)count != len)
1266 goto err;
1267
1268 /* add the function address and name */
1269 scnprintf(disasm_buf, sizeof(disasm_buf), "%#"PRIx64" <%s>:",
1270 start, sym->name);
1271
1272 args->offset = -1;
1273 args->line = disasm_buf;
1274 args->line_nr = 0;
1275 args->fileloc = NULL;
1276 args->ms->sym = sym;
1277
1278 dl = disasm_line__new(args);
1279 if (dl == NULL)
1280 goto err;
1281
1282 annotation_line__add(&dl->al, ¬es->src->source);
1283
1284 /* Each raw instruction is 4 byte */
1285 count = len/4;
1286
1287 for (i = 0, offset = 0; i < count; i++) {
1288 args->offset = offset;
1289 sprintf(args->line, "%x", line[i]);
1290 dl = disasm_line__new(args);
1291 if (dl == NULL)
1292 break;
1293
1294 annotation_line__add(&dl->al, ¬es->src->source);
1295 offset += 4;
1296 }
1297
1298 /* It failed in the middle */
1299 if (offset != len) {
1300 struct list_head *list = ¬es->src->source;
1301
1302 /* Discard all lines and fallback to objdump */
1303 while (!list_empty(list)) {
1304 dl = list_first_entry(list, struct disasm_line, al.node);
1305
1306 list_del_init(&dl->al.node);
1307 disasm_line__free(dl);
1308 }
1309 count = -1;
1310 }
1311
1312 out:
1313 free(buf);
1314 return count < 0 ? count : 0;
1315
1316 err:
1317 count = -1;
1318 goto out;
1319 }
1320
1321 /*
1322 * Possibly create a new version of line with tabs expanded. Returns the
1323 * existing or new line, storage is updated if a new line is allocated. If
1324 * allocation fails then NULL is returned.
1325 */
expand_tabs(char * line,char ** storage,size_t * storage_len)1326 char *expand_tabs(char *line, char **storage, size_t *storage_len)
1327 {
1328 size_t i, src, dst, len, new_storage_len, num_tabs;
1329 char *new_line;
1330 size_t line_len = strlen(line);
1331
1332 for (num_tabs = 0, i = 0; i < line_len; i++)
1333 if (line[i] == '\t')
1334 num_tabs++;
1335
1336 if (num_tabs == 0)
1337 return line;
1338
1339 /*
1340 * Space for the line and '\0', less the leading and trailing
1341 * spaces. Each tab may introduce 7 additional spaces.
1342 */
1343 new_storage_len = line_len + 1 + (num_tabs * 7);
1344
1345 new_line = malloc(new_storage_len);
1346 if (new_line == NULL) {
1347 pr_err("Failure allocating memory for tab expansion\n");
1348 return NULL;
1349 }
1350
1351 /*
1352 * Copy regions starting at src and expand tabs. If there are two
1353 * adjacent tabs then 'src == i', the memcpy is of size 0 and the spaces
1354 * are inserted.
1355 */
1356 for (i = 0, src = 0, dst = 0; i < line_len && num_tabs; i++) {
1357 if (line[i] == '\t') {
1358 len = i - src;
1359 memcpy(&new_line[dst], &line[src], len);
1360 dst += len;
1361 new_line[dst++] = ' ';
1362 while (dst % 8 != 0)
1363 new_line[dst++] = ' ';
1364 src = i + 1;
1365 num_tabs--;
1366 }
1367 }
1368
1369 /* Expand the last region. */
1370 len = line_len - src;
1371 memcpy(&new_line[dst], &line[src], len);
1372 dst += len;
1373 new_line[dst] = '\0';
1374
1375 free(*storage);
1376 *storage = new_line;
1377 *storage_len = new_storage_len;
1378 return new_line;
1379 }
1380
symbol__disassemble_bpf_image(struct symbol * sym,struct annotate_args * args)1381 static int symbol__disassemble_bpf_image(struct symbol *sym, struct annotate_args *args)
1382 {
1383 struct annotation *notes = symbol__annotation(sym);
1384 struct disasm_line *dl;
1385
1386 args->offset = -1;
1387 args->line = strdup("to be implemented");
1388 args->line_nr = 0;
1389 args->fileloc = NULL;
1390 dl = disasm_line__new(args);
1391 if (dl)
1392 annotation_line__add(&dl->al, ¬es->src->source);
1393
1394 zfree(&args->line);
1395 return 0;
1396 }
1397
symbol__disassemble_objdump(const char * filename,struct symbol * sym,struct annotate_args * args)1398 static int symbol__disassemble_objdump(const char *filename, struct symbol *sym,
1399 struct annotate_args *args)
1400 {
1401 struct annotation_options *opts = &annotate_opts;
1402 struct map *map = args->ms->map;
1403 struct dso *dso = map__dso(map);
1404 char *command;
1405 FILE *file;
1406 int lineno = 0;
1407 char *fileloc = NULL;
1408 int nline;
1409 char *line;
1410 size_t line_len;
1411 const char *objdump_argv[] = {
1412 "/bin/sh",
1413 "-c",
1414 NULL, /* Will be the objdump command to run. */
1415 "--",
1416 NULL, /* Will be the symfs path. */
1417 NULL,
1418 };
1419 struct child_process objdump_process;
1420 int err;
1421
1422 if (dso__binary_type(dso) == DSO_BINARY_TYPE__BPF_PROG_INFO)
1423 return symbol__disassemble_bpf_libbfd(sym, args);
1424
1425 if (dso__binary_type(dso) == DSO_BINARY_TYPE__BPF_IMAGE)
1426 return symbol__disassemble_bpf_image(sym, args);
1427
1428 err = asprintf(&command,
1429 "%s %s%s --start-address=0x%016" PRIx64
1430 " --stop-address=0x%016" PRIx64
1431 " %s -d %s %s %s %c%s%c %s%s -C \"$1\"",
1432 opts->objdump_path ?: "objdump",
1433 opts->disassembler_style ? "-M " : "",
1434 opts->disassembler_style ?: "",
1435 map__rip_2objdump(map, sym->start),
1436 map__rip_2objdump(map, sym->end),
1437 opts->show_linenr ? "-l" : "",
1438 opts->show_asm_raw ? "" : "--no-show-raw-insn",
1439 opts->annotate_src ? "-S" : "",
1440 opts->prefix ? "--prefix " : "",
1441 opts->prefix ? '"' : ' ',
1442 opts->prefix ?: "",
1443 opts->prefix ? '"' : ' ',
1444 opts->prefix_strip ? "--prefix-strip=" : "",
1445 opts->prefix_strip ?: "");
1446
1447 if (err < 0) {
1448 pr_err("Failure allocating memory for the command to run\n");
1449 return err;
1450 }
1451
1452 pr_debug("Executing: %s\n", command);
1453
1454 objdump_argv[2] = command;
1455 objdump_argv[4] = filename;
1456
1457 /* Create a pipe to read from for stdout */
1458 memset(&objdump_process, 0, sizeof(objdump_process));
1459 objdump_process.argv = objdump_argv;
1460 objdump_process.out = -1;
1461 objdump_process.err = -1;
1462 objdump_process.no_stderr = 1;
1463 if (start_command(&objdump_process)) {
1464 pr_err("Failure starting to run %s\n", command);
1465 err = -1;
1466 goto out_free_command;
1467 }
1468
1469 file = fdopen(objdump_process.out, "r");
1470 if (!file) {
1471 pr_err("Failure creating FILE stream for %s\n", command);
1472 /*
1473 * If we were using debug info should retry with
1474 * original binary.
1475 */
1476 err = -1;
1477 goto out_close_stdout;
1478 }
1479
1480 /* Storage for getline. */
1481 line = NULL;
1482 line_len = 0;
1483
1484 nline = 0;
1485 while (!feof(file)) {
1486 const char *match;
1487 char *expanded_line;
1488
1489 if (getline(&line, &line_len, file) < 0 || !line)
1490 break;
1491
1492 /* Skip lines containing "filename:" */
1493 match = strstr(line, filename);
1494 if (match && match[strlen(filename)] == ':')
1495 continue;
1496
1497 expanded_line = strim(line);
1498 expanded_line = expand_tabs(expanded_line, &line, &line_len);
1499 if (!expanded_line)
1500 break;
1501
1502 /*
1503 * The source code line number (lineno) needs to be kept in
1504 * across calls to symbol__parse_objdump_line(), so that it
1505 * can associate it with the instructions till the next one.
1506 * See disasm_line__new() and struct disasm_line::line_nr.
1507 */
1508 if (symbol__parse_objdump_line(sym, args, expanded_line,
1509 &lineno, &fileloc) < 0)
1510 break;
1511 nline++;
1512 }
1513 free(line);
1514 free(fileloc);
1515
1516 err = finish_command(&objdump_process);
1517 if (err)
1518 pr_err("Error running %s\n", command);
1519
1520 if (nline == 0) {
1521 err = -1;
1522 pr_err("No output from %s\n", command);
1523 }
1524
1525 /*
1526 * kallsyms does not have symbol sizes so there may a nop at the end.
1527 * Remove it.
1528 */
1529 if (dso__is_kcore(dso))
1530 delete_last_nop(sym);
1531
1532 fclose(file);
1533
1534 out_close_stdout:
1535 close(objdump_process.out);
1536
1537 out_free_command:
1538 free(command);
1539 return err;
1540 }
1541
symbol__disassemble(struct symbol * sym,struct annotate_args * args)1542 int symbol__disassemble(struct symbol *sym, struct annotate_args *args)
1543 {
1544 struct annotation_options *options = args->options;
1545 struct map *map = args->ms->map;
1546 struct dso *dso = map__dso(map);
1547 char symfs_filename[PATH_MAX];
1548 bool delete_extract = false;
1549 struct kcore_extract kce;
1550 bool decomp = false;
1551 int err = dso__disassemble_filename(dso, symfs_filename, sizeof(symfs_filename));
1552
1553 if (err)
1554 return err;
1555
1556 pr_debug("%s: filename=%s, sym=%s, start=%#" PRIx64 ", end=%#" PRIx64 "\n", __func__,
1557 symfs_filename, sym->name, map__unmap_ip(map, sym->start),
1558 map__unmap_ip(map, sym->end));
1559
1560 pr_debug("annotating [%p] %30s : [%p] %30s\n", dso, dso__long_name(dso), sym, sym->name);
1561
1562 if (dso__binary_type(dso) == DSO_BINARY_TYPE__NOT_FOUND) {
1563 return SYMBOL_ANNOTATE_ERRNO__COULDNT_DETERMINE_FILE_TYPE;
1564 } else if (dso__is_kcore(dso)) {
1565 kce.addr = map__rip_2objdump(map, sym->start);
1566 kce.kcore_filename = symfs_filename;
1567 kce.len = sym->end - sym->start;
1568 kce.offs = sym->start;
1569
1570 if (!kcore_extract__create(&kce)) {
1571 delete_extract = true;
1572 strlcpy(symfs_filename, kce.extract_filename, sizeof(symfs_filename));
1573 }
1574 } else if (dso__needs_decompress(dso)) {
1575 char tmp[KMOD_DECOMP_LEN];
1576
1577 if (dso__decompress_kmodule_path(dso, symfs_filename, tmp, sizeof(tmp)) < 0)
1578 return -1;
1579
1580 decomp = true;
1581 strcpy(symfs_filename, tmp);
1582 }
1583
1584 /*
1585 * For powerpc data type profiling, use the dso__data_read_offset to
1586 * read raw instruction directly and interpret the binary code to
1587 * understand instructions and register fields. For sort keys as type
1588 * and typeoff, disassemble to mnemonic notation is not required in
1589 * case of powerpc.
1590 */
1591 if (arch__is_powerpc(args->arch)) {
1592 extern const char *sort_order;
1593
1594 if (sort_order && !strstr(sort_order, "sym")) {
1595 err = symbol__disassemble_raw(symfs_filename, sym, args);
1596 if (err == 0)
1597 goto out_remove_tmp;
1598
1599 err = symbol__disassemble_capstone_powerpc(symfs_filename, sym, args);
1600 if (err == 0)
1601 goto out_remove_tmp;
1602 }
1603 }
1604
1605 /* FIXME: LLVM and CAPSTONE should support source code */
1606 if (options->annotate_src && !options->hide_src_code) {
1607 err = symbol__disassemble_objdump(symfs_filename, sym, args);
1608 if (err == 0)
1609 goto out_remove_tmp;
1610 }
1611
1612 err = -1;
1613 for (u8 i = 0; i < ARRAY_SIZE(options->disassemblers) && err != 0; i++) {
1614 enum perf_disassembler dis = options->disassemblers[i];
1615
1616 switch (dis) {
1617 case PERF_DISASM_LLVM:
1618 args->options->disassembler_used = PERF_DISASM_LLVM;
1619 err = symbol__disassemble_llvm(symfs_filename, sym, args);
1620 break;
1621 case PERF_DISASM_CAPSTONE:
1622 args->options->disassembler_used = PERF_DISASM_CAPSTONE;
1623 err = symbol__disassemble_capstone(symfs_filename, sym, args);
1624 break;
1625 case PERF_DISASM_OBJDUMP:
1626 args->options->disassembler_used = PERF_DISASM_OBJDUMP;
1627 err = symbol__disassemble_objdump(symfs_filename, sym, args);
1628 break;
1629 case PERF_DISASM_UNKNOWN: /* End of disassemblers. */
1630 default:
1631 args->options->disassembler_used = PERF_DISASM_UNKNOWN;
1632 goto out_remove_tmp;
1633 }
1634 if (err == 0)
1635 pr_debug("Disassembled with %s\n", perf_disassembler__strs[dis]);
1636 }
1637 out_remove_tmp:
1638 if (decomp)
1639 unlink(symfs_filename);
1640
1641 if (delete_extract)
1642 kcore_extract__delete(&kce);
1643
1644 return err;
1645 }
1646