1 /*
2 * Copyright (c) 2008, 2025, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25
26 #include "asm/assembler.inline.hpp"
27 #include "asm/macroAssembler.hpp"
28 #include "ci/ciUtilities.hpp"
29 #include "classfile/javaClasses.hpp"
30 #include "code/codeCache.hpp"
31 #include "compiler/disassembler.hpp"
32 #include "gc/shared/cardTable.hpp"
33 #include "gc/shared/cardTableBarrierSet.hpp"
34 #include "gc/shared/collectedHeap.hpp"
35 #include "logging/log.hpp"
36 #include "memory/resourceArea.hpp"
37 #include "memory/universe.hpp"
38 #include "oops/oop.inline.hpp"
39 #include "runtime/handles.inline.hpp"
40 #include "runtime/java.hpp"
41 #include "runtime/os.hpp"
42 #include "runtime/stubCodeGenerator.hpp"
43 #include "runtime/stubRoutines.hpp"
44 #include "utilities/hashTable.hpp"
45
46 void* Disassembler::_library = nullptr;
47 bool Disassembler::_tried_to_load_library = false;
48 bool Disassembler::_library_usable = false;
49
50 // This routine is in the shared library:
51 Disassembler::decode_func_virtual Disassembler::_decode_instructions_virtual = nullptr;
52
53 static const char hsdis_library_name[] = "hsdis-" HOTSPOT_LIB_ARCH;
54 static const char decode_instructions_virtual_name[] = "decode_instructions_virtual";
55 #define COMMENT_COLUMN 52 LP64_ONLY(+8) /*could be an option*/
56 #define BYTES_COMMENT ";..." /* funky byte display comment */
57
58 class decode_env {
59 private:
60 outputStream* _output; // where the disassembly is directed to
61 CodeBlob* _codeBlob; // != nullptr only when decoding a CodeBlob
62 nmethod* _nm; // != nullptr only when decoding a nmethod
63
64 address _start; // != nullptr when decoding a range of unknown type
65 address _end; // != nullptr when decoding a range of unknown type
66
67 char _option_buf[512];
68 char _print_raw;
69 address _cur_insn; // address of instruction currently being decoded
70 int _bytes_per_line; // arch-specific formatting option
71 int _pre_decode_alignment;
72 int _post_decode_alignment;
73 bool _print_file_name;
74 bool _print_help;
75 bool _helpPrinted;
76 static bool _optionsParsed;
77 #ifndef PRODUCT
78 const AsmRemarks* _remarks; // Used with start/end range to provide code remarks.
79 ptrdiff_t _disp; // Adjustment to offset -> remark mapping.
80 #endif
81
82 enum {
83 tabspacing = 8
84 };
85
86 // Check if the event matches the expected tag
87 // The tag must be a substring of the event, and
88 // the tag must be a token in the event, i.e. separated by delimiters
89 static bool match(const char* event, const char* tag) {
90 size_t eventlen = strlen(event);
91 size_t taglen = strlen(tag);
92 if (eventlen < taglen) // size mismatch
93 return false;
94 if (strncmp(event, tag, taglen) != 0) // string mismatch
95 return false;
96 char delim = event[taglen];
97 return delim == '\0' || delim == ' ' || delim == '/' || delim == '=';
98 }
99
100 // Merge new option string with previously recorded options
101 void collect_options(const char* p) {
102 if (p == nullptr || p[0] == '\0') return;
103 size_t opt_so_far = strlen(_option_buf);
104 if (opt_so_far + 1 + strlen(p) + 1 > sizeof(_option_buf)) return;
105 char* fillp = &_option_buf[opt_so_far];
106 if (opt_so_far > 0) *fillp++ = ',';
107 strcat(fillp, p);
108 // replace white space by commas:
109 char* q = fillp;
110 while ((q = strpbrk(q, " \t\n")) != nullptr)
111 *q++ = ',';
112 }
113
114 void process_options(outputStream* ost);
115
116 void print_insn_labels();
117 void print_insn_prefix();
118 void print_address(address value);
119
120 // Properly initializes _start/_end. Overwritten too often if
121 // printing of instructions is called for each instruction.
122 void set_start(address s) { _start = s; }
123 void set_end (address e) { _end = e; }
124 void set_nm (nmethod* nm) { _nm = nm; }
125 void set_output(outputStream* st) { _output = st; }
126
127 #if defined(SUPPORT_ASSEMBLY) || defined(SUPPORT_ABSTRACT_ASSEMBLY)
128 // The disassembler library (sometimes) uses tabs to nicely align the instruction operands.
129 // Depending on the mnemonic length and the column position where the
130 // mnemonic is printed, alignment may turn out to be not so nice.
131 // To improve, we assume 8-character tab spacing and left-align the mnemonic on a tab position.
132 // Instruction comments are aligned 4 tab positions to the right of the mnemonic.
133 void calculate_alignment() {
134 _pre_decode_alignment = ((output()->position()+tabspacing-1)/tabspacing)*tabspacing;
135 _post_decode_alignment = _pre_decode_alignment + 4*tabspacing;
136 }
137
138 void start_insn(address pc) {
139 _cur_insn = pc;
140 output()->bol();
141 print_insn_labels();
142 print_insn_prefix();
143 }
144
145 void end_insn(address pc) {
146 address pc0 = cur_insn();
147 outputStream* st = output();
148
149 if (AbstractDisassembler::show_comment()) {
150 if ((_nm != nullptr) && _nm->has_code_comment(pc0, pc)) {
151 _nm->print_code_comment_on
152 (st,
153 _post_decode_alignment ? _post_decode_alignment : COMMENT_COLUMN,
154 pc0, pc);
155 // this calls reloc_string_for which calls oop::print_value_on
156 }
157 print_hook_comments(pc0, _nm != nullptr);
158 }
159 Disassembler::annotate(pc0, output());
160 // follow each complete insn by a nice newline
161 st->bol();
162 }
163 #endif
164
165 struct SourceFileInfo {
166 struct Link : public CHeapObj<mtCode> {
167 const char* file;
168 int line;
169 Link* next;
170 Link(const char* f, int l) : file(f), line(l), next(nullptr) {}
171 };
172 Link *head, *tail;
173
174 void append(const char* file, int line) {
175 if (tail != nullptr && tail->file == file && tail->line == line) {
176 // Don't print duplicated lines at the same address. This could happen with C
177 // macros that end up having multiple "__" tokens on the same __LINE__.
178 return;
179 }
180 Link *link = new Link(file, line);
181 if (head == nullptr) {
182 head = tail = link;
183 } else {
184 tail->next = link;
185 tail = link;
186 }
187 }
188 SourceFileInfo(const char* file, int line) : head(nullptr), tail(nullptr) {
189 append(file, line);
190 }
191 };
192
193 typedef HashTable<
194 address, SourceFileInfo,
195 15889, // prime number
196 AnyObj::C_HEAP> SourceFileInfoTable;
197
198 static SourceFileInfoTable* _src_table;
199 static const char* _cached_src;
200 static GrowableArray<const char*>* _cached_src_lines;
201
202 static SourceFileInfoTable& src_table() {
203 if (_src_table == nullptr) {
204 _src_table = new (mtCode)SourceFileInfoTable();
205 }
206 return *_src_table;
207 }
208
209 public:
210 decode_env(CodeBlob* code, outputStream* output);
211 decode_env(nmethod* code, outputStream* output);
212 // Constructor for a 'decode_env' to decode an arbitrary
213 // piece of memory, hopefully containing code.
214 decode_env(address start, address end, outputStream* output
215 NOT_PRODUCT(COMMA const AsmRemarks* remarks = nullptr COMMA ptrdiff_t disp = 0));
216
217 // Add 'original_start' argument which is the original address
218 // the instructions were located at (if this is not equal to 'start').
219 address decode_instructions(address start, address end, address original_start = nullptr);
220
221 address handle_event(const char* event, address arg);
222
223 outputStream* output() { return _output; }
224 address cur_insn() { return _cur_insn; }
225 const char* options() { return _option_buf; }
226 static void hook(const char* file, int line, address pc);
227 void print_hook_comments(address pc, bool newline);
228 };
229
230 bool decode_env::_optionsParsed = false;
231
232 decode_env::SourceFileInfoTable* decode_env::_src_table = nullptr;
233 const char* decode_env::_cached_src = nullptr;
234 GrowableArray<const char*>* decode_env::_cached_src_lines = nullptr;
235
236 void decode_env::hook(const char* file, int line, address pc) {
237 // For simplication, we never free from this table. It's really not
238 // necessary as we add to the table only when PrintInterpreter is true,
239 // which means we are debugging the VM and a little bit of extra
240 // memory usage doesn't matter.
241 SourceFileInfo* found = src_table().get(pc);
242 if (found != nullptr) {
243 found->append(file, line);
244 } else {
245 SourceFileInfo sfi(file, line);
246 src_table().put(pc, sfi); // sfi is copied by value
247 }
248 }
249
250 void decode_env::print_hook_comments(address pc, bool newline) {
251 SourceFileInfo* found = src_table().get(pc);
252 outputStream* st = output();
253 if (found != nullptr) {
254 for (SourceFileInfo::Link *link = found->head; link; link = link->next) {
255 const char* file = link->file;
256 int line = link->line;
257 if (_cached_src == nullptr || strcmp(_cached_src, file) != 0) {
258 FILE* fp;
259
260 // _cached_src_lines is a single cache of the lines of a source file, and we refill this cache
261 // every time we need to print a line from a different source file. It's not the fastest,
262 // but seems bearable.
263 if (_cached_src_lines != nullptr) {
264 for (int i=0; i<_cached_src_lines->length(); i++) {
265 os::free((void*)_cached_src_lines->at(i));
266 }
267 _cached_src_lines->clear();
268 } else {
269 _cached_src_lines = new (mtCode) GrowableArray<const char*>(0, mtCode);
270 }
271
272 if ((fp = os::fopen(file, "r")) == nullptr) {
273 _cached_src = nullptr;
274 return;
275 }
276 _cached_src = file;
277
278 char line[500]; // don't write lines that are too long in your source files!
279 while (fgets(line, sizeof(line), fp) != nullptr) {
280 size_t len = strlen(line);
281 if (len > 0 && line[len-1] == '\n') {
282 line[len-1] = '\0';
283 }
284 _cached_src_lines->append(os::strdup(line));
285 }
286 fclose(fp);
287 _print_file_name = true;
288 }
289
290 if (_print_file_name) {
291 // We print the file name whenever we switch to a new file, or when
292 // Disassembler::decode is called to disassemble a new block of code.
293 _print_file_name = false;
294 if (newline) {
295 st->cr();
296 }
297 st->move_to(COMMENT_COLUMN);
298 st->print(";;@FILE: %s", file);
299 newline = true;
300 }
301
302 int index = line - 1; // 1-based line number -> 0-based index.
303 if (index >= _cached_src_lines->length()) {
304 // This could happen if source file is mismatched.
305 } else {
306 const char* source_line = _cached_src_lines->at(index);
307 if (newline) {
308 st->cr();
309 }
310 st->move_to(COMMENT_COLUMN);
311 st->print(";;%5d: %s", line, source_line);
312 newline = true;
313 }
314 }
315 }
316 }
317
318 decode_env::decode_env(CodeBlob* code, outputStream* output) :
319 _output(output ? output : tty),
320 _codeBlob(code),
321 _nm(_codeBlob != nullptr && _codeBlob->is_nmethod() ? (nmethod*) code : nullptr),
322 _start(nullptr),
323 _end(nullptr),
324 _option_buf(),
325 _print_raw(0),
326 _cur_insn(nullptr),
327 _bytes_per_line(0),
328 _pre_decode_alignment(0),
329 _post_decode_alignment(0),
330 _print_file_name(false),
331 _print_help(false),
332 _helpPrinted(false)
333 NOT_PRODUCT(COMMA _remarks(nullptr))
334 NOT_PRODUCT(COMMA _disp(0))
335 {
336 memset(_option_buf, 0, sizeof(_option_buf));
337 process_options(_output);
338 }
339
340 decode_env::decode_env(nmethod* code, outputStream* output) :
341 _output(output ? output : tty),
342 _codeBlob(nullptr),
343 _nm(code),
344 _start(_nm->code_begin()),
345 _end(_nm->code_end()),
346 _option_buf(),
347 _print_raw(0),
348 _cur_insn(nullptr),
349 _bytes_per_line(0),
350 _pre_decode_alignment(0),
351 _post_decode_alignment(0),
352 _print_file_name(false),
353 _print_help(false),
354 _helpPrinted(false)
355 NOT_PRODUCT(COMMA _remarks(nullptr))
356 NOT_PRODUCT(COMMA _disp(0))
357 {
358 memset(_option_buf, 0, sizeof(_option_buf));
359 process_options(_output);
360 }
361
362 // Constructor for a 'decode_env' to decode a memory range [start, end)
363 // of unknown origin, assuming it contains code.
364 decode_env::decode_env(address start, address end, outputStream* output
365 NOT_PRODUCT(COMMA const AsmRemarks* remarks COMMA ptrdiff_t disp)) :
366 _output(output ? output : tty),
367 _codeBlob(nullptr),
368 _nm(nullptr),
369 _start(start),
370 _end(end),
371 _option_buf(),
372 _print_raw(0),
373 _cur_insn(nullptr),
374 _bytes_per_line(0),
375 _pre_decode_alignment(0),
376 _post_decode_alignment(0),
377 _print_file_name(false),
378 _print_help(false),
379 _helpPrinted(false)
380 NOT_PRODUCT(COMMA _remarks(remarks))
381 NOT_PRODUCT(COMMA _disp(disp))
382 {
383 assert(start < end, "Range must have a positive size, [" PTR_FORMAT ".." PTR_FORMAT ").", p2i(start), p2i(end));
384 memset(_option_buf, 0, sizeof(_option_buf));
385 process_options(_output);
386 }
387
388 void decode_env::process_options(outputStream* ost) {
389 // by default, output pc but not bytes:
390 _print_help = false;
391 _bytes_per_line = Disassembler::pd_instruction_alignment();
392 _print_file_name = true;
393
394 // parse the global option string
395 // We need to fill the options buffer for each newly created
396 // decode_env instance. The hsdis_* library looks for options
397 // in that buffer.
398 collect_options(Disassembler::pd_cpu_opts());
399 collect_options(PrintAssemblyOptions);
400
401 if (strstr(options(), "print-raw")) {
402 _print_raw = (strstr(options(), "xml") ? 2 : 1);
403 }
404
405 if (_optionsParsed) return; // parse only once
406
407 if (strstr(options(), "help")) {
408 _print_help = true;
409 }
410 if (strstr(options(), "align-instr")) {
411 AbstractDisassembler::toggle_align_instr();
412 }
413 if (strstr(options(), "show-pc")) {
414 AbstractDisassembler::toggle_show_pc();
415 }
416 if (strstr(options(), "show-offset")) {
417 AbstractDisassembler::toggle_show_offset();
418 }
419 if (strstr(options(), "show-bytes")) {
420 AbstractDisassembler::toggle_show_bytes();
421 }
422 if (strstr(options(), "show-data-hex")) {
423 AbstractDisassembler::toggle_show_data_hex();
424 }
425 if (strstr(options(), "show-data-int")) {
426 AbstractDisassembler::toggle_show_data_int();
427 }
428 if (strstr(options(), "show-data-float")) {
429 AbstractDisassembler::toggle_show_data_float();
430 }
431 if (strstr(options(), "show-structs")) {
432 AbstractDisassembler::toggle_show_structs();
433 }
434 if (strstr(options(), "show-comment")) {
435 AbstractDisassembler::toggle_show_comment();
436 }
437 if (strstr(options(), "show-block-comment")) {
438 AbstractDisassembler::toggle_show_block_comment();
439 }
440 _optionsParsed = true;
441
442 if (_print_help && ! _helpPrinted) {
443 _helpPrinted = true;
444 ost->print_cr("PrintAssemblyOptions help:");
445 ost->print_cr(" print-raw test plugin by requesting raw output");
446 ost->print_cr(" print-raw-xml test plugin by requesting raw xml");
447 ost->cr();
448 ost->print_cr(" show-pc toggle printing current pc, currently %s", AbstractDisassembler::show_pc() ? "ON" : "OFF");
449 ost->print_cr(" show-offset toggle printing current offset, currently %s", AbstractDisassembler::show_offset() ? "ON" : "OFF");
450 ost->print_cr(" show-bytes toggle printing instruction bytes, currently %s", AbstractDisassembler::show_bytes() ? "ON" : "OFF");
451 ost->print_cr(" show-data-hex toggle formatting data as hex, currently %s", AbstractDisassembler::show_data_hex() ? "ON" : "OFF");
452 ost->print_cr(" show-data-int toggle formatting data as int, currently %s", AbstractDisassembler::show_data_int() ? "ON" : "OFF");
453 ost->print_cr(" show-data-float toggle formatting data as float, currently %s", AbstractDisassembler::show_data_float() ? "ON" : "OFF");
454 ost->print_cr(" show-structs toggle compiler data structures, currently %s", AbstractDisassembler::show_structs() ? "ON" : "OFF");
455 ost->print_cr(" show-comment toggle instruction comments, currently %s", AbstractDisassembler::show_comment() ? "ON" : "OFF");
456 ost->print_cr(" show-block-comment toggle block comments, currently %s", AbstractDisassembler::show_block_comment() ? "ON" : "OFF");
457 ost->print_cr(" align-instr toggle instruction alignment, currently %s", AbstractDisassembler::align_instr() ? "ON" : "OFF");
458 ost->print_cr("combined options: %s", options());
459 }
460 }
461
462 // Disassembly Event Handler.
463 // This method receives events from the disassembler library hsdis
464 // via event_to_env for each decoding step (installed by
465 // Disassembler::decode_instructions(), replacing the default
466 // callback method). This enables dumping additional info
467 // and custom line formatting.
468 // In a future extension, calling a custom decode method will be
469 // supported. We can use such a method to decode instructions the
470 // binutils decoder does not handle to our liking (suboptimal
471 // formatting, incomplete information, ...).
472 // Returns:
473 // - nullptr for all standard invocations. The function result is not
474 // examined (as of now, 20190409) by the hsdis decoder loop.
475 // - next for 'insn0' invocations.
476 // next == arg: the custom decoder didn't do anything.
477 // next > arg: the custom decoder did decode the instruction.
478 // next points to the next undecoded instruction
479 // (continuation point for decoder loop).
480 //
481 // "Normal" sequence of events:
482 // insns - start of instruction stream decoding
483 // mach - display architecture
484 // format - display bytes-per-line
485 // for each instruction:
486 // insn - start of instruction decoding
487 // insn0 - custom decoder invocation (if any)
488 // addr - print address value
489 // /insn - end of instruction decoding
490 // /insns - premature end of instruction stream due to no progress
491 //
492 address decode_env::handle_event(const char* event, address arg) {
493
494 #if defined(SUPPORT_ASSEMBLY) || defined(SUPPORT_ABSTRACT_ASSEMBLY)
495
496 //---< Event: end decoding loop (error, no progress) >---
497 if (decode_env::match(event, "/insns")) {
498 // Nothing to be done here.
499 return nullptr;
500 }
501
502 //---< Event: start decoding loop >---
503 if (decode_env::match(event, "insns")) {
504 // Nothing to be done here.
505 return nullptr;
506 }
507
508 //---< Event: finish decoding an instruction >---
509 if (decode_env::match(event, "/insn")) {
510 output()->fill_to(_post_decode_alignment);
511 end_insn(arg);
512 return nullptr;
513 }
514
515 //---< Event: start decoding an instruction >---
516 if (decode_env::match(event, "insn")) {
517 start_insn(arg);
518 } else if (match(event, "/insn")) {
519 end_insn(arg);
520 } else if (match(event, "addr")) {
521 if (arg != nullptr) {
522 print_address(arg);
523 return arg;
524 }
525 calculate_alignment();
526 output()->fill_to(_pre_decode_alignment);
527 return nullptr;
528 }
529
530 //---< Event: call custom decoder (platform specific) >---
531 if (decode_env::match(event, "insn0")) {
532 return Disassembler::decode_instruction0(arg, output(), arg);
533 }
534
535 //---< Event: Print address >---
536 if (decode_env::match(event, "addr")) {
537 print_address(arg);
538 return arg;
539 }
540
541 //---< Event: mach (inform about machine architecture) >---
542 // This event is problematic because it messes up the output.
543 // The event is fired after the instruction address has already
544 // been printed. The decoded instruction (event "insn") is
545 // printed afterwards. That doesn't look nice.
546 if (decode_env::match(event, "mach")) {
547 guarantee(arg != nullptr, "event_to_env - arg must not be nullptr for event 'mach'");
548 static char buffer[64] = { 0, };
549 // Output suppressed because it messes up disassembly.
550 // Only print this when the mach changes.
551 if (false && (strcmp(buffer, (const char*)arg) != 0 ||
552 strlen((const char*)arg) > sizeof(buffer) - 1)) {
553 // Only print this when the mach changes
554 strncpy(buffer, (const char*)arg, sizeof(buffer) - 1);
555 buffer[sizeof(buffer) - 1] = '\0';
556 output()->print_cr("[Disassembling for mach='%s']", (const char*)arg);
557 }
558 return nullptr;
559 }
560
561 //---< Event: format bytes-per-line >---
562 if (decode_env::match(event, "format bytes-per-line")) {
563 _bytes_per_line = (int) (intptr_t) arg;
564 return nullptr;
565 }
566 #endif
567 return nullptr;
568 }
569
570 static void* event_to_env(void* env_pv, const char* event, void* arg) {
571 decode_env* env = (decode_env*) env_pv;
572 return env->handle_event(event, (address) arg);
573 }
574
575 // called by the disassembler to print out jump targets and data addresses
576 void decode_env::print_address(address adr) {
577 outputStream* st = output();
578
579 if (adr == nullptr) {
580 st->print("nullptr");
581 return;
582 }
583
584 int small_num = (int)(intptr_t)adr;
585 if ((intptr_t)adr == (intptr_t)small_num
586 && -1 <= small_num && small_num <= 9) {
587 st->print("%d", small_num);
588 return;
589 }
590
591 if (Universe::is_fully_initialized()) {
592 if (StubRoutines::contains(adr)) {
593 StubCodeDesc* desc = StubCodeDesc::desc_for(adr);
594 if (desc == nullptr) {
595 desc = StubCodeDesc::desc_for(adr + frame::pc_return_offset);
596 }
597 if (desc != nullptr) {
598 st->print("Stub::%s", desc->name());
599 if (desc->begin() != adr) {
600 st->print("%+zd " PTR_FORMAT, adr - desc->begin(), p2i(adr));
601 } else if (WizardMode) {
602 st->print(" " PTR_FORMAT, p2i(adr));
603 }
604 return;
605 }
606 st->print("Stub::<unknown> " PTR_FORMAT, p2i(adr));
607 return;
608 }
609
610 if (is_card_table_address(adr)) {
611 st->print("word_map_base");
612 if (WizardMode) st->print(" " INTPTR_FORMAT, p2i(adr));
613 return;
614 }
615 }
616
617 if (_nm == nullptr) {
618 // Don't do this for native methods, as the function name will be printed in
619 // nmethod::reloc_string_for().
620 // Allocate the buffer on the stack instead of as RESOURCE array.
621 // In case we do DecodeErrorFile, Thread will not be initialized,
622 // causing a "assert(current != __null) failed" failure.
623 const int buflen = 1024;
624 char buf[buflen];
625 int offset;
626 if (os::dll_address_to_function_name(adr, buf, buflen, &offset)) {
627 st->print(PTR_FORMAT " = %s", p2i(adr), buf);
628 if (offset != 0) {
629 st->print("+%d", offset);
630 }
631 return;
632 }
633 }
634
635 // Fall through to a simple (hexadecimal) numeral.
636 st->print(PTR_FORMAT, p2i(adr));
637 }
638
639 void decode_env::print_insn_labels() {
640 if (AbstractDisassembler::show_block_comment()) {
641 address p = cur_insn();
642 outputStream* st = output();
643
644 //---< Block comments for nmethod >---
645 // Outputs a bol() before and a cr() after, but only if a comment is printed.
646 // Prints nmethod_section_label as well.
647 if (_nm != nullptr) {
648 _nm->print_block_comment(st, p);
649 }
650 else if (_codeBlob != nullptr) {
651 _codeBlob->print_block_comment(st, p);
652 }
653 #ifndef PRODUCT
654 else if (_remarks != nullptr) {
655 _remarks->print((p - _start) + _disp, st);
656 }
657 #endif
658 }
659 }
660
661 void decode_env::print_insn_prefix() {
662 address p = cur_insn();
663 outputStream* st = output();
664 AbstractDisassembler::print_location(p, _start, _end, st, false, false);
665 AbstractDisassembler::print_instruction(p, Assembler::instr_len(p), Assembler::instr_maxlen(), st, true, false);
666 }
667
668 ATTRIBUTE_PRINTF(2, 3)
669 static int printf_to_env(void* env_pv, const char* format, ...) {
670 decode_env* env = (decode_env*) env_pv;
671 outputStream* st = env->output();
672 size_t flen = strlen(format);
673 const char* raw = nullptr;
674 if (flen == 0) return 0;
675 if (flen == 1 && format[0] == '\n') { st->bol(); return 1; }
676 if (flen < 2 ||
677 strchr(format, '%') == nullptr) {
678 raw = format;
679 } else if (format[0] == '%' && format[1] == '%' &&
680 strchr(format+2, '%') == nullptr) {
681 // happens a lot on machines with names like %foo
682 flen--;
683 raw = format+1;
684 }
685 if (raw != nullptr) {
686 st->print_raw(raw, flen);
687 return (int) flen;
688 }
689 va_list ap;
690 va_start(ap, format);
691 julong cnt0 = st->count();
692 st->vprint(format, ap);
693 julong cnt1 = st->count();
694 va_end(ap);
695 return (int)(cnt1 - cnt0);
696 }
697
698 // The 'original_start' argument holds the original address where
699 // the instructions were located in the originating system. If zero (nullptr)
700 // is passed in, there is no original address.
701 address decode_env::decode_instructions(address start, address end, address original_start /* = 0*/) {
702 // CodeComment in Stubs.
703 // Properly initialize _start/_end. Overwritten too often if
704 // printing of instructions is called for each instruction.
705 assert((_start == nullptr) || (start == nullptr) || (_start == start), "don't overwrite CTOR values");
706 assert((_end == nullptr) || (end == nullptr) || (_end == end ), "don't overwrite CTOR values");
707 if (start != nullptr) set_start(start);
708 if (end != nullptr) set_end(end);
709 if (original_start == nullptr) {
710 original_start = start;
711 }
712
713 //---< Check (and correct) alignment >---
714 // Don't check alignment of end, it is not aligned.
715 if (((uint64_t)start & ((uint64_t)Disassembler::pd_instruction_alignment() - 1)) != 0) {
716 output()->print_cr("Decode range start:" PTR_FORMAT ": ... (unaligned)", p2i(start));
717 start = (address)((uint64_t)start & ~((uint64_t)Disassembler::pd_instruction_alignment() - 1));
718 }
719
720 // Trying to decode instructions doesn't make sense if we
721 // couldn't load the disassembler library.
722 if (Disassembler::is_abstract()) {
723 return nullptr;
724 }
725
726 // decode a series of instructions and return the end of the last instruction
727
728 if (_print_raw) {
729 // Print whatever the library wants to print, w/o fancy callbacks.
730 // This is mainly for debugging the library itself.
731 FILE* out = stdout;
732 FILE* xmlout = (_print_raw > 1 ? out : nullptr);
733 return
734 (address)
735 (*Disassembler::_decode_instructions_virtual)((uintptr_t)start, (uintptr_t)end,
736 start, end - start,
737 nullptr, (void*) xmlout,
738 nullptr, (void*) out,
739 options(), 0/*nice new line*/);
740 }
741
742 return
743 (address)
744 (*Disassembler::_decode_instructions_virtual)((uintptr_t)start, (uintptr_t)end,
745 start, end - start,
746 &event_to_env, (void*) this,
747 &printf_to_env, (void*) this,
748 options(), 0/*nice new line*/);
749 }
750
751 // ----------------------------------------------------------------------------
752 // Disassembler
753 // Used as a static wrapper for decode_env.
754 // Each method will create a decode_env before decoding.
755 // You can call the decode_env methods directly if you already have one.
756
757 void* Disassembler::dll_load(char* buf, int buflen, int offset, char* ebuf, int ebuflen, outputStream* st) {
758 int sz = buflen - offset;
759 int written = jio_snprintf(&buf[offset], sz, "%s%s", hsdis_library_name, JNI_LIB_SUFFIX);
760 if (written < sz) { // written successfully, not truncated.
761 if (Verbose) st->print_cr("Trying to load: %s", buf);
762 return os::dll_load(buf, ebuf, ebuflen);
763 } else if (Verbose) {
764 st->print_cr("Try to load hsdis library failed: the length of path is beyond the OS limit");
765 }
766 return nullptr;
767 }
768
769 bool Disassembler::load_library(outputStream* st) {
770 // Do not try to load multiple times. Failed once -> fails always.
771 // To force retry in debugger: assign _tried_to_load_library=0
772 if (_tried_to_load_library) {
773 return _library_usable;
774 }
775
776 #if defined(SUPPORT_ASSEMBLY) || defined(SUPPORT_ABSTRACT_ASSEMBLY)
777 // Print to given stream, if any.
778 // Print to tty if Verbose is on and no stream given.
779 st = ((st == nullptr) && Verbose) ? tty : st;
780
781 // Compute fully qualified library name.
782 char ebuf[1024];
783 char buf[JVM_MAXPATHLEN];
784 os::jvm_path(buf, sizeof(buf));
785 int jvm_offset = -1;
786 int lib_offset = -1;
787
788 if (is_vm_statically_linked()) {
789 char* p = strrchr(buf, '/');
790 *p = '\0';
791 strcat(p, "/lib/");
792 lib_offset = jvm_offset = (int)strlen(buf);
793 } else {
794 // Match "libjvm" instead of "jvm" on *nix platforms. Creates better matches.
795 // Match "[lib]jvm[^/]*" in jvm_path.
796 const char* base = buf;
797 const char* p = strrchr(buf, *os::file_separator());
798 if (p != nullptr) lib_offset = p - base + 1; // this points to the first char after separator
799 #ifdef _WIN32
800 p = strstr(p ? p : base, "jvm");
801 if (p != nullptr) jvm_offset = p - base; // this points to 'j' in jvm.
802 #else
803 p = strstr(p ? p : base, "libjvm");
804 if (p != nullptr) jvm_offset = p - base + 3; // this points to 'j' in libjvm.
805 #endif
806 }
807
808 // Find the disassembler shared library.
809 // Search for several paths derived from libjvm, in this order:
810 // 1. <home>/lib/<vm>/libhsdis-<arch>.so (for compatibility)
811 // 2. <home>/lib/<vm>/hsdis-<arch>.so
812 // 3. <home>/lib/hsdis-<arch>.so
813 // 4. hsdis-<arch>.so (using LD_LIBRARY_PATH)
814 if (jvm_offset >= 0) {
815 // 1. <home>/lib/<vm>/libhsdis-<arch>.so
816 _library = dll_load(buf, sizeof buf, jvm_offset, ebuf, sizeof ebuf, st);
817 if (_library == nullptr && lib_offset >= 0) {
818 // 2. <home>/lib/<vm>/hsdis-<arch>.so
819 _library = dll_load(buf, sizeof buf, lib_offset, ebuf, sizeof ebuf, st);
820 }
821 if (_library == nullptr && lib_offset > 0) {
822 // 3. <home>/lib/hsdis-<arch>.so
823 buf[lib_offset - 1] = '\0';
824 const char* p = strrchr(buf, *os::file_separator());
825 if (p != nullptr) {
826 lib_offset = p - buf + 1;
827 _library = dll_load(buf, sizeof buf, lib_offset, ebuf, sizeof ebuf, st);
828 }
829 }
830 }
831 if (_library == nullptr) {
832 _library = dll_load(buf, sizeof buf, 0, ebuf, sizeof ebuf, st);
833 }
834
835 // load the decoder function to use.
836 if (_library != nullptr) {
837 _decode_instructions_virtual = CAST_TO_FN_PTR(Disassembler::decode_func_virtual,
838 os::dll_lookup(_library, decode_instructions_virtual_name));
839 } else {
840 log_warning(os)("Loading hsdis library failed");
841 }
842 _tried_to_load_library = true;
843 _library_usable = _decode_instructions_virtual != nullptr;
844
845 // Create a dummy environment to initialize PrintAssemblyOptions.
846 // The PrintAssemblyOptions must be known for abstract disassemblies as well.
847 decode_env dummy((unsigned char*)(&buf[0]), (unsigned char*)(&buf[1]), st);
848
849 // Report problems during dll_load or dll_lookup, if any.
850 if (st != nullptr) {
851 // Success.
852 if (_library_usable) {
853 st->print_cr("Loaded disassembler from %s", buf);
854 } else {
855 st->print_cr("Could not load %s; %s; %s",
856 buf,
857 ((_library != nullptr)
858 ? "entry point is missing"
859 : ((WizardMode || PrintMiscellaneous)
860 ? (const char*)ebuf
861 : "library not loadable")),
862 "PrintAssembly defaults to abstract disassembly.");
863 }
864 }
865 #endif
866 return _library_usable;
867 }
868
869
870 // Directly disassemble code blob.
871 void Disassembler::decode(CodeBlob* cb, outputStream* st) {
872 #if defined(SUPPORT_ASSEMBLY) || defined(SUPPORT_ABSTRACT_ASSEMBLY)
873 if (cb->is_nmethod()) {
874 // If we have an nmethod at hand,
875 // call the specialized decoder directly.
876 ((nmethod*)cb)->decode2(st);
877 return;
878 }
879
880 decode_env env(cb, st);
881 env.output()->print_cr("--------------------------------------------------------------------------------");
882 env.output()->print("Decoding CodeBlob");
883 if (cb->name() != nullptr) {
884 env.output()->print(", name: %s,", cb->name());
885 }
886 env.output()->print_cr(" at [" PTR_FORMAT ", " PTR_FORMAT "] " JLONG_FORMAT " bytes", p2i(cb->code_begin()), p2i(cb->code_end()), ((jlong)(cb->code_end() - cb->code_begin())));
887
888 if (is_abstract()) {
889 AbstractDisassembler::decode_abstract(cb->code_begin(), cb->code_end(), env.output(), Assembler::instr_maxlen());
890 } else {
891 env.decode_instructions(cb->code_begin(), cb->code_end());
892 }
893 env.output()->print_cr("--------------------------------------------------------------------------------");
894 #endif
895 }
896
897 // Decode a nmethod.
898 // This includes printing the constant pool and all code segments.
899 // The nmethod data structures (oop maps, relocations and the like) are not printed.
900 void Disassembler::decode(nmethod* nm, outputStream* st) {
901 #if defined(SUPPORT_ASSEMBLY) || defined(SUPPORT_ABSTRACT_ASSEMBLY)
902 ttyLocker ttyl;
903
904 decode_env env(nm, st);
905 env.output()->print_cr("--------------------------------------------------------------------------------");
906 nm->print_constant_pool(env.output());
907 env.output()->print_cr("--------------------------------------------------------------------------------");
908 env.output()->cr();
909 if (is_abstract()) {
910 AbstractDisassembler::decode_abstract(nm->code_begin(), nm->code_end(), env.output(), Assembler::instr_maxlen());
911 } else {
912 env.decode_instructions(nm->code_begin(), nm->code_end());
913 }
914 env.output()->print_cr("--------------------------------------------------------------------------------");
915 #endif
916 }
917
918 // Decode a range, given as [start address, end address)
919 void Disassembler::decode(address start, address end, outputStream* st
920 NOT_PRODUCT(COMMA const AsmRemarks* remarks COMMA ptrdiff_t disp)) {
921 #if defined(SUPPORT_ASSEMBLY) || defined(SUPPORT_ABSTRACT_ASSEMBLY)
922 //---< Test memory before decoding >---
923 if (!os::is_readable_range(start, end)) {
924 //---< Allow output suppression, but prevent writing to a nullptr stream. Could happen with +PrintStubCode. >---
925 if (st != nullptr) {
926 st->print("Memory range [" PTR_FORMAT ".." PTR_FORMAT "] not readable", p2i(start), p2i(end));
927 }
928 return;
929 }
930
931 if (is_abstract()) {
932 AbstractDisassembler::decode_abstract(start, end, st, Assembler::instr_maxlen());
933 } else {
934 // This seems to be just a chunk of memory.
935 decode_env env(start, end, st NOT_PRODUCT(COMMA remarks COMMA disp));
936 env.output()->print_cr("--------------------------------------------------------------------------------");
937 env.decode_instructions(start, end);
938 env.output()->print_cr("--------------------------------------------------------------------------------");
939 }
940 #endif
941 }
942
943 // To prevent excessive code expansion in the interpreter generator, we
944 // do not inline this function into Disassembler::hook().
945 void Disassembler::_hook(const char* file, int line, MacroAssembler* masm) {
946 decode_env::hook(file, line, masm->code_section()->end());
947 }