1 /*
2 * Copyright (c) 2008, 2018, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * The Universal Permissive License (UPL), Version 1.0
6 *
7 * Subject to the condition set forth below, permission is hereby granted to
8 * any person obtaining a copy of this software, associated documentation
9 * and/or data (collectively the "Software"), free of charge and under any
10 * and all copyright rights in the Software, and any and all patent rights
11 * owned or freely licensable by each licensor hereunder covering either (i)
12 * the unmodified Software as contributed to or provided by such licensor,
13 * or (ii) the Larger Works (as defined below), to deal in both
14 *
15 * (a) the Software, and
16 *
17 * (b) any piece of software and/or hardware listed in the lrgrwrks.txt file
18 * if one is included with the Software (each a "Larger Work" to which the
19 * Software is contributed by such licensors),
20 *
21 * without restriction, including without limitation the rights to copy,
22 * create derivative works of, display, perform, and distribute the Software
23 * and make, use, sell, offer for sale, import, export, have made, and have
24 * sold the Software and the Larger Work(s), and to sublicense the foregoing
25 * rights on either these or other terms.
26 *
27 * This license is subject to the following condition:
28 *
29 * The above copyright notice and either this complete permission notice or
30 * at a minimum a reference to the UPL must be included in all copies or
31 * substantial portions of the Software.
32 *
33 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
34 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
35 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN
36 * NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM,
37 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
38 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
39 * USE OR OTHER DEALINGS IN THE SOFTWARE.
40 *
41 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
42 * or visit www.oracle.com if you need additional information or have any
43 * questions.
44 *
45 */
46
47 /* hsdis.c -- dump a range of addresses as native instructions
48 This implements the plugin protocol required by the
49 HotSpot PrintAssembly option.
50 */
51
52 #include <config.h> /* required by bfd.h */
53 #include <errno.h>
54 #include <inttypes.h>
55 #include <string.h>
56
57 #include <libiberty.h>
58 #include <bfd.h>
59 #include <bfdver.h>
60 #include <dis-asm.h>
61
62 #include "hsdis.h"
63
64 #ifndef bool
65 #define bool int
66 #define true 1
67 #define false 0
68 #endif /*bool*/
69
70 /* short names for stuff in hsdis.h */
71 typedef decode_instructions_event_callback_ftype event_callback_t;
72 typedef decode_instructions_printf_callback_ftype printf_callback_t;
73
74 /* disassemble_info.application_data object */
75 struct hsdis_app_data {
76 /* virtual address of data */
77 uintptr_t start_va, end_va;
78 /* the instructions to be decoded */
79 unsigned char* buffer;
80 uintptr_t length;
81 event_callback_t event_callback; void* event_stream;
82 printf_callback_t printf_callback; void* printf_stream;
83 bool losing;
84 bool do_newline;
85
86 /* the architecture being disassembled */
87 const char* arch_name;
88 const bfd_arch_info_type* arch_info;
89
90 /* the disassembler we are going to use: */
91 disassembler_ftype dfn;
92 struct disassemble_info dinfo; /* the actual struct! */
93
94 char mach_option[64];
95 char insn_options[256];
96 };
97
98 static void* decode(struct hsdis_app_data* app_data, const char* options);
99
100 #define DECL_APP_DATA(dinfo) \
101 struct hsdis_app_data* app_data = (struct hsdis_app_data*) (dinfo)->application_data
102
103 #define DECL_EVENT_CALLBACK(app_data) \
104 event_callback_t event_callback = (app_data)->event_callback; \
105 void* event_stream = (app_data)->event_stream
106
107 #define DECL_PRINTF_CALLBACK(app_data) \
108 printf_callback_t printf_callback = (app_data)->printf_callback; \
109 void* printf_stream = (app_data)->printf_stream
110
111
112 static void print_help(struct hsdis_app_data* app_data,
113 const char* msg, const char* arg);
114 static void setup_app_data(struct hsdis_app_data* app_data,
115 const char* options);
116 static const char* format_insn_close(const char* close,
117 disassemble_info* dinfo,
118 char* buf, size_t bufsize);
119
120 void*
121 #ifdef DLL_ENTRY
122 DLL_ENTRY
123 #endif
124 decode_instructions_virtual(uintptr_t start_va, uintptr_t end_va,
125 unsigned char* buffer, uintptr_t length,
126 event_callback_t event_callback_arg, void* event_stream_arg,
127 printf_callback_t printf_callback_arg, void* printf_stream_arg,
128 const char* options, int newline) {
129 struct hsdis_app_data app_data;
130 memset(&app_data, 0, sizeof(app_data));
131 app_data.start_va = start_va;
132 app_data.end_va = end_va;
133 app_data.buffer = buffer;
134 app_data.length = length;
135 app_data.event_callback = event_callback_arg;
136 app_data.event_stream = event_stream_arg;
137 app_data.printf_callback = printf_callback_arg;
138 app_data.printf_stream = printf_stream_arg;
139 app_data.do_newline = newline == 0 ? false : true;
140
141 return decode(&app_data, options);
142 }
143
144 /* This is the compatability interface for older version of hotspot */
145 void*
146 #ifdef DLL_ENTRY
147 DLL_ENTRY
148 #endif
149 decode_instructions(void* start_pv, void* end_pv,
150 event_callback_t event_callback_arg, void* event_stream_arg,
151 printf_callback_t printf_callback_arg, void* printf_stream_arg,
152 const char* options) {
153 decode_instructions_virtual((uintptr_t)start_pv,
154 (uintptr_t)end_pv,
155 (unsigned char*)start_pv,
156 (uintptr_t)end_pv - (uintptr_t)start_pv,
157 event_callback_arg,
158 event_stream_arg,
159 printf_callback_arg,
160 printf_stream_arg,
161 options, false);
162 }
163
164 static void* decode(struct hsdis_app_data* app_data, const char* options) {
165 setup_app_data(app_data, options);
166 char buf[128];
167
168 {
169 /* now reload everything from app_data: */
170 DECL_EVENT_CALLBACK(app_data);
171 DECL_PRINTF_CALLBACK(app_data);
172 uintptr_t start = app_data->start_va;
173 uintptr_t end = app_data->end_va;
174 uintptr_t p = start;
175
176 (*event_callback)(event_stream, "insns", (void*)start);
177
178 (*event_callback)(event_stream, "mach name='%s'",
179 (void*) app_data->arch_info->printable_name);
180 if (app_data->dinfo.bytes_per_line != 0) {
181 (*event_callback)(event_stream, "format bytes-per-line='%p'/",
182 (void*)(intptr_t) app_data->dinfo.bytes_per_line);
183 }
184
185 while (p < end && !app_data->losing) {
186 (*event_callback)(event_stream, "insn", (void*) p);
187
188 /* reset certain state, so we can read it with confidence */
189 app_data->dinfo.insn_info_valid = 0;
190 app_data->dinfo.branch_delay_insns = 0;
191 app_data->dinfo.data_size = 0;
192 app_data->dinfo.insn_type = 0;
193
194 int size = (*app_data->dfn)((bfd_vma) p, &app_data->dinfo);
195
196 if (size > 0) p += size;
197 else app_data->losing = true;
198
199 if (!app_data->losing) {
200 const char* insn_close = format_insn_close("/insn", &app_data->dinfo,
201 buf, sizeof(buf));
202 (*event_callback)(event_stream, insn_close, (void*) p);
203
204 if (app_data->do_newline) {
205 /* follow each complete insn by a nice newline */
206 (*printf_callback)(printf_stream, "\n");
207 }
208 }
209 }
210
211 if (app_data->losing) (*event_callback)(event_stream, "/insns", (void*) p);
212 return (void*) p;
213 }
214 }
215
216 /* take the address of the function, for luck, and also test the typedef: */
217 const decode_func_vtype decode_func_virtual_address = &decode_instructions_virtual;
218 const decode_func_stype decode_func_address = &decode_instructions;
219
220 static const char* format_insn_close(const char* close,
221 disassemble_info* dinfo,
222 char* buf, size_t bufsize) {
223 if (!dinfo->insn_info_valid)
224 return close;
225 enum dis_insn_type itype = dinfo->insn_type;
226 int dsize = dinfo->data_size, delays = dinfo->branch_delay_insns;
227 if ((itype == dis_nonbranch && (dsize | delays) == 0)
228 || (strlen(close) + 3*20 > bufsize))
229 return close;
230
231 const char* type = "unknown";
232 switch (itype) {
233 case dis_nonbranch: type = NULL; break;
234 case dis_branch: type = "branch"; break;
235 case dis_condbranch: type = "condbranch"; break;
236 case dis_jsr: type = "jsr"; break;
237 case dis_condjsr: type = "condjsr"; break;
238 case dis_dref: type = "dref"; break;
239 case dis_dref2: type = "dref2"; break;
240 }
241
242 strcpy(buf, close);
243 char* p = buf;
244 if (type) sprintf(p += strlen(p), " type='%s'", type);
245 if (dsize) sprintf(p += strlen(p), " dsize='%d'", dsize);
246 if (delays) sprintf(p += strlen(p), " delay='%d'", delays);
247 return buf;
248 }
249
250 /* handler functions */
251
252 static int
253 hsdis_read_memory_func(bfd_vma memaddr,
254 bfd_byte* myaddr,
255 unsigned int length,
256 struct disassemble_info* dinfo) {
257 DECL_APP_DATA(dinfo);
258 /* convert the virtual address memaddr into an address within memory buffer */
259 uintptr_t offset = ((uintptr_t) memaddr) - app_data->start_va;
260 if (offset + length > app_data->length) {
261 /* read is out of bounds */
262 return EIO;
263 } else {
264 memcpy(myaddr, (bfd_byte*) (app_data->buffer + offset), length);
265 return 0;
266 }
267 }
268
269 static void
270 hsdis_print_address_func(bfd_vma vma, struct disassemble_info* dinfo) {
271 /* the actual value to print: */
272 void* addr_value = (void*) (uintptr_t) vma;
273 DECL_APP_DATA(dinfo);
274 DECL_EVENT_CALLBACK(app_data);
275
276 /* issue the event: */
277 void* result =
278 (*event_callback)(event_stream, "addr/", addr_value);
279 if (result == NULL) {
280 /* event declined */
281 generic_print_address(vma, dinfo);
282 }
283 }
284
285
286 /* configuration */
287
288 static void set_optional_callbacks(struct hsdis_app_data* app_data);
289 static void parse_caller_options(struct hsdis_app_data* app_data,
290 const char* caller_options);
291 static const char* native_arch_name();
292 static enum bfd_endian native_endian();
293 static const bfd_arch_info_type* find_arch_info(const char* arch_nane);
294 static bfd* get_native_bfd(const bfd_arch_info_type* arch_info,
295 /* to avoid malloc: */
296 bfd* empty_bfd, bfd_target* empty_xvec);
297 static void init_disassemble_info_from_bfd(struct disassemble_info* dinfo,
298 void *stream,
299 fprintf_ftype fprintf_func,
300 bfd* bfd,
301 char* disassembler_options);
302 static void parse_fake_insn(disassembler_ftype dfn,
303 struct disassemble_info* dinfo);
304
305 static void setup_app_data(struct hsdis_app_data* app_data,
306 const char* caller_options) {
307 /* Make reasonable defaults for null callbacks.
308 A non-null stream for a null callback is assumed to be a FILE* for output.
309 Events are rendered as XML.
310 */
311 set_optional_callbacks(app_data);
312
313 /* Look into caller_options for anything interesting. */
314 if (caller_options != NULL)
315 parse_caller_options(app_data, caller_options);
316
317 /* Discover which architecture we are going to disassemble. */
318 app_data->arch_name = &app_data->mach_option[0];
319 if (app_data->arch_name[0] == '\0')
320 app_data->arch_name = native_arch_name();
321 app_data->arch_info = find_arch_info(app_data->arch_name);
322
323 /* Make a fake bfd to hold the arch. and byteorder info. */
324 struct {
325 bfd_target empty_xvec;
326 bfd empty_bfd;
327 } buf;
328 bfd* native_bfd = get_native_bfd(app_data->arch_info,
329 /* to avoid malloc: */
330 &buf.empty_bfd, &buf.empty_xvec);
331 init_disassemble_info_from_bfd(&app_data->dinfo,
332 app_data->printf_stream,
333 app_data->printf_callback,
334 native_bfd,
335 /* On PowerPC we get warnings, if we pass empty options */
336 (caller_options == NULL) ? NULL : app_data->insn_options);
337
338 /* Finish linking together the various callback blocks. */
339 app_data->dinfo.application_data = (void*) app_data;
340 app_data->dfn = disassembler(bfd_get_arch(native_bfd),
341 bfd_big_endian(native_bfd),
342 bfd_get_mach(native_bfd),
343 native_bfd);
344 app_data->dinfo.print_address_func = hsdis_print_address_func;
345 app_data->dinfo.read_memory_func = hsdis_read_memory_func;
346
347 if (app_data->dfn == NULL) {
348 const char* bad = app_data->arch_name;
349 static bool complained;
350 if (bad == &app_data->mach_option[0])
351 print_help(app_data, "bad mach=%s", bad);
352 else if (!complained)
353 print_help(app_data, "bad native mach=%s; please port hsdis to this platform", bad);
354 complained = true;
355 /* must bail out */
356 app_data->losing = true;
357 return;
358 }
359
360 parse_fake_insn(app_data->dfn, &app_data->dinfo);
361 }
362
363
364 /* ignore all events, return a null */
365 static void* null_event_callback(void* ignore_stream, const char* ignore_event, void* arg) {
366 return NULL;
367 }
368
369 /* print all events as XML markup */
370 static void* xml_event_callback(void* stream, const char* event, void* arg) {
371 FILE* fp = (FILE*) stream;
372 #define NS_PFX "dis:"
373 if (event[0] != '/') {
374 /* issue the tag, with or without a formatted argument */
375 fprintf(fp, "<"NS_PFX);
376 fprintf(fp, event, arg);
377 fprintf(fp, ">");
378 } else {
379 ++event; /* skip slash */
380 const char* argp = strchr(event, ' ');
381 if (argp == NULL) {
382 /* no arguments; just issue the closing tag */
383 fprintf(fp, "</"NS_PFX"%s>", event);
384 } else {
385 /* split out the closing attributes as <dis:foo_done attr='val'/> */
386 int event_prefix = (argp - event);
387 fprintf(fp, "<"NS_PFX"%.*s_done", event_prefix, event);
388 fprintf(fp, argp, arg);
389 fprintf(fp, "/></"NS_PFX"%.*s>", event_prefix, event);
390 }
391 }
392 return NULL;
393 }
394
395 static void set_optional_callbacks(struct hsdis_app_data* app_data) {
396 if (app_data->printf_callback == NULL) {
397 int (*fprintf_callback)(FILE*, const char*, ...) = &fprintf;
398 FILE* fprintf_stream = stdout;
399 app_data->printf_callback = (printf_callback_t) fprintf_callback;
400 if (app_data->printf_stream == NULL)
401 app_data->printf_stream = (void*) fprintf_stream;
402 }
403 if (app_data->event_callback == NULL) {
404 if (app_data->event_stream == NULL)
405 app_data->event_callback = &null_event_callback;
406 else
407 app_data->event_callback = &xml_event_callback;
408 }
409
410 }
411
412 static void parse_caller_options(struct hsdis_app_data* app_data, const char* caller_options) {
413 char* iop_base = app_data->insn_options;
414 char* iop_limit = iop_base + sizeof(app_data->insn_options) - 1;
415 char* iop = iop_base;
416 const char* p;
417 for (p = caller_options; p != NULL; ) {
418 const char* q = strchr(p, ',');
419 size_t plen = (q == NULL) ? strlen(p) : ((q++) - p);
420 if (plen == 4 && strncmp(p, "help", plen) == 0) {
421 print_help(app_data, NULL, NULL);
422 } else if (plen >= 5 && strncmp(p, "mach=", 5) == 0) {
423 char* mach_option = app_data->mach_option;
424 size_t mach_size = sizeof(app_data->mach_option);
425 mach_size -= 1; /*leave room for the null*/
426 if (plen > mach_size) plen = mach_size;
427 strncpy(mach_option, p, plen);
428 mach_option[plen] = '\0';
429 } else if (plen > 6 && strncmp(p, "hsdis-", 6) == 0) {
430 // do not pass these to the next level
431 } else {
432 /* just copy it; {i386,sparc}-dis.c might like to see it */
433 if (iop > iop_base && iop < iop_limit) (*iop++) = ',';
434 if (iop + plen > iop_limit)
435 plen = iop_limit - iop;
436 strncpy(iop, p, plen);
437 iop += plen;
438 }
439 p = q;
440 }
441 *iop = '\0';
442 }
443
444 static void print_help(struct hsdis_app_data* app_data,
445 const char* msg, const char* arg) {
446 DECL_PRINTF_CALLBACK(app_data);
447 if (msg != NULL) {
448 (*printf_callback)(printf_stream, "hsdis: ");
449 (*printf_callback)(printf_stream, msg, arg);
450 (*printf_callback)(printf_stream, "\n");
451 }
452 (*printf_callback)(printf_stream, "hsdis output options:\n");
453 if (printf_callback == (printf_callback_t) &fprintf)
454 disassembler_usage((FILE*) printf_stream);
455 else
456 disassembler_usage(stderr); /* better than nothing */
457 (*printf_callback)(printf_stream, " mach=<arch> select disassembly mode\n");
458 #if defined(LIBARCH_i386) || defined(LIBARCH_amd64)
459 (*printf_callback)(printf_stream, " mach=i386 select 32-bit mode\n");
460 (*printf_callback)(printf_stream, " mach=x86-64 select 64-bit mode\n");
461 (*printf_callback)(printf_stream, " suffix always print instruction suffix\n");
462 #endif
463 (*printf_callback)(printf_stream, " help print this message\n");
464 }
465
466
467 /* low-level bfd and arch stuff that binutils doesn't do for us */
468
469 static const bfd_arch_info_type* find_arch_info(const char* arch_name) {
470 const bfd_arch_info_type* arch_info = bfd_scan_arch(arch_name);
471 if (arch_info == NULL) {
472 extern const bfd_arch_info_type bfd_default_arch_struct;
473 arch_info = &bfd_default_arch_struct;
474 }
475 return arch_info;
476 }
477
478 static const char* native_arch_name() {
479 const char* res = NULL;
480 #ifdef LIBARCH_i386
481 res = "i386";
482 #endif
483 #ifdef LIBARCH_amd64
484 res = "i386:x86-64";
485 #endif
486 #ifdef LIBARCH_sparc
487 res = "sparc:v8plusb";
488 #endif
489 #ifdef LIBARCH_sparcv9
490 res = "sparc:v9b";
491 #endif
492 #if defined(LIBARCH_ppc64) || defined(LIBARCH_ppc64le)
493 res = "powerpc:common64";
494 #endif
495 if (res == NULL)
496 res = "architecture not set in Makefile!";
497 return res;
498 }
499
500 static enum bfd_endian native_endian() {
501 int32_t endian_test = 'x';
502 if (*(const char*) &endian_test == 'x')
503 return BFD_ENDIAN_LITTLE;
504 else
505 return BFD_ENDIAN_BIG;
506 }
507
508 static bfd* get_native_bfd(const bfd_arch_info_type* arch_info,
509 bfd* empty_bfd, bfd_target* empty_xvec) {
510 memset(empty_bfd, 0, sizeof(*empty_bfd));
511 memset(empty_xvec, 0, sizeof(*empty_xvec));
512 empty_xvec->flavour = bfd_target_unknown_flavour;
513 empty_xvec->byteorder = native_endian();
514 empty_bfd->xvec = empty_xvec;
515 empty_bfd->arch_info = arch_info;
516 return empty_bfd;
517 }
518
519 static int read_zero_data_only(bfd_vma ignore_p,
520 bfd_byte* myaddr, unsigned int length,
521 struct disassemble_info *ignore_info) {
522 memset(myaddr, 0, length);
523 return 0;
524 }
525 static int print_to_dev_null(void* ignore_stream, const char* ignore_format, ...) {
526 return 0;
527 }
528
529 /* Prime the pump by running the selected disassembler on a null input.
530 This forces the machine-specific disassembler to divulge invariant
531 information like bytes_per_line.
532 */
533 static void parse_fake_insn(disassembler_ftype dfn,
534 struct disassemble_info* dinfo) {
535 typedef int (*read_memory_ftype)
536 (bfd_vma memaddr, bfd_byte *myaddr, unsigned int length,
537 struct disassemble_info *info);
538 read_memory_ftype read_memory_func = dinfo->read_memory_func;
539 fprintf_ftype fprintf_func = dinfo->fprintf_func;
540
541 dinfo->read_memory_func = &read_zero_data_only;
542 dinfo->fprintf_func = &print_to_dev_null;
543 (*dfn)(0, dinfo);
544
545 /* put it back */
546 dinfo->read_memory_func = read_memory_func;
547 dinfo->fprintf_func = fprintf_func;
548 }
549
550 static void init_disassemble_info_from_bfd(struct disassemble_info* dinfo,
551 void *stream,
552 fprintf_ftype fprintf_func,
553 bfd* abfd,
554 char* disassembler_options) {
555 init_disassemble_info(dinfo, stream, fprintf_func);
556
557 dinfo->flavour = bfd_get_flavour(abfd);
558 dinfo->arch = bfd_get_arch(abfd);
559 dinfo->mach = bfd_get_mach(abfd);
560 dinfo->disassembler_options = disassembler_options;
561 dinfo->octets_per_byte = bfd_octets_per_byte (abfd);
562 dinfo->skip_zeroes = sizeof(void*) * 2;
563 dinfo->skip_zeroes_at_end = sizeof(void*)-1;
564 dinfo->disassembler_needs_relocs = FALSE;
565
566 if (bfd_big_endian(abfd))
567 dinfo->display_endian = dinfo->endian = BFD_ENDIAN_BIG;
568 else if (bfd_little_endian(abfd))
569 dinfo->display_endian = dinfo->endian = BFD_ENDIAN_LITTLE;
570 else
571 dinfo->endian = native_endian();
572
573 disassemble_init_for_target(dinfo);
574 }