gdb: remove SYMTAB_COMPUNIT macro, add getter/setter
[binutils-gdb.git] / gdb / amd64-tdep.c
1 /* Target-dependent code for AMD64.
2
3 Copyright (C) 2001-2022 Free Software Foundation, Inc.
4
5 Contributed by Jiri Smid, SuSE Labs.
6
7 This file is part of GDB.
8
9 This program is free software; you can redistribute it and/or modify
10 it under the terms of the GNU General Public License as published by
11 the Free Software Foundation; either version 3 of the License, or
12 (at your option) any later version.
13
14 This program is distributed in the hope that it will be useful,
15 but WITHOUT ANY WARRANTY; without even the implied warranty of
16 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 GNU General Public License for more details.
18
19 You should have received a copy of the GNU General Public License
20 along with this program. If not, see <http://www.gnu.org/licenses/>. */
21
22 #include "defs.h"
23 #include "opcode/i386.h"
24 #include "dis-asm.h"
25 #include "arch-utils.h"
26 #include "block.h"
27 #include "dummy-frame.h"
28 #include "frame.h"
29 #include "frame-base.h"
30 #include "frame-unwind.h"
31 #include "inferior.h"
32 #include "infrun.h"
33 #include "gdbcmd.h"
34 #include "gdbcore.h"
35 #include "objfiles.h"
36 #include "regcache.h"
37 #include "regset.h"
38 #include "symfile.h"
39 #include "disasm.h"
40 #include "amd64-tdep.h"
41 #include "i387-tdep.h"
42 #include "gdbsupport/x86-xstate.h"
43 #include <algorithm>
44 #include "target-descriptions.h"
45 #include "arch/amd64.h"
46 #include "producer.h"
47 #include "ax.h"
48 #include "ax-gdb.h"
49 #include "gdbsupport/byte-vector.h"
50 #include "osabi.h"
51 #include "x86-tdep.h"
52 #include "amd64-ravenscar-thread.h"
53
54 /* Note that the AMD64 architecture was previously known as x86-64.
55 The latter is (forever) engraved into the canonical system name as
56 returned by config.guess, and used as the name for the AMD64 port
57 of GNU/Linux. The BSD's have renamed their ports to amd64; they
58 don't like to shout. For GDB we prefer the amd64_-prefix over the
59 x86_64_-prefix since it's so much easier to type. */
60
61 /* Register information. */
62
63 static const char * const amd64_register_names[] =
64 {
65 "rax", "rbx", "rcx", "rdx", "rsi", "rdi", "rbp", "rsp",
66
67 /* %r8 is indeed register number 8. */
68 "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15",
69 "rip", "eflags", "cs", "ss", "ds", "es", "fs", "gs",
70
71 /* %st0 is register number 24. */
72 "st0", "st1", "st2", "st3", "st4", "st5", "st6", "st7",
73 "fctrl", "fstat", "ftag", "fiseg", "fioff", "foseg", "fooff", "fop",
74
75 /* %xmm0 is register number 40. */
76 "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7",
77 "xmm8", "xmm9", "xmm10", "xmm11", "xmm12", "xmm13", "xmm14", "xmm15",
78 "mxcsr",
79 };
80
81 static const char * const amd64_ymm_names[] =
82 {
83 "ymm0", "ymm1", "ymm2", "ymm3",
84 "ymm4", "ymm5", "ymm6", "ymm7",
85 "ymm8", "ymm9", "ymm10", "ymm11",
86 "ymm12", "ymm13", "ymm14", "ymm15"
87 };
88
89 static const char * const amd64_ymm_avx512_names[] =
90 {
91 "ymm16", "ymm17", "ymm18", "ymm19",
92 "ymm20", "ymm21", "ymm22", "ymm23",
93 "ymm24", "ymm25", "ymm26", "ymm27",
94 "ymm28", "ymm29", "ymm30", "ymm31"
95 };
96
97 static const char * const amd64_ymmh_names[] =
98 {
99 "ymm0h", "ymm1h", "ymm2h", "ymm3h",
100 "ymm4h", "ymm5h", "ymm6h", "ymm7h",
101 "ymm8h", "ymm9h", "ymm10h", "ymm11h",
102 "ymm12h", "ymm13h", "ymm14h", "ymm15h"
103 };
104
105 static const char * const amd64_ymmh_avx512_names[] =
106 {
107 "ymm16h", "ymm17h", "ymm18h", "ymm19h",
108 "ymm20h", "ymm21h", "ymm22h", "ymm23h",
109 "ymm24h", "ymm25h", "ymm26h", "ymm27h",
110 "ymm28h", "ymm29h", "ymm30h", "ymm31h"
111 };
112
113 static const char * const amd64_mpx_names[] =
114 {
115 "bnd0raw", "bnd1raw", "bnd2raw", "bnd3raw", "bndcfgu", "bndstatus"
116 };
117
118 static const char * const amd64_k_names[] =
119 {
120 "k0", "k1", "k2", "k3",
121 "k4", "k5", "k6", "k7"
122 };
123
124 static const char * const amd64_zmmh_names[] =
125 {
126 "zmm0h", "zmm1h", "zmm2h", "zmm3h",
127 "zmm4h", "zmm5h", "zmm6h", "zmm7h",
128 "zmm8h", "zmm9h", "zmm10h", "zmm11h",
129 "zmm12h", "zmm13h", "zmm14h", "zmm15h",
130 "zmm16h", "zmm17h", "zmm18h", "zmm19h",
131 "zmm20h", "zmm21h", "zmm22h", "zmm23h",
132 "zmm24h", "zmm25h", "zmm26h", "zmm27h",
133 "zmm28h", "zmm29h", "zmm30h", "zmm31h"
134 };
135
136 static const char * const amd64_zmm_names[] =
137 {
138 "zmm0", "zmm1", "zmm2", "zmm3",
139 "zmm4", "zmm5", "zmm6", "zmm7",
140 "zmm8", "zmm9", "zmm10", "zmm11",
141 "zmm12", "zmm13", "zmm14", "zmm15",
142 "zmm16", "zmm17", "zmm18", "zmm19",
143 "zmm20", "zmm21", "zmm22", "zmm23",
144 "zmm24", "zmm25", "zmm26", "zmm27",
145 "zmm28", "zmm29", "zmm30", "zmm31"
146 };
147
148 static const char * const amd64_xmm_avx512_names[] = {
149 "xmm16", "xmm17", "xmm18", "xmm19",
150 "xmm20", "xmm21", "xmm22", "xmm23",
151 "xmm24", "xmm25", "xmm26", "xmm27",
152 "xmm28", "xmm29", "xmm30", "xmm31"
153 };
154
155 static const char * const amd64_pkeys_names[] = {
156 "pkru"
157 };
158
159 /* DWARF Register Number Mapping as defined in the System V psABI,
160 section 3.6. */
161
162 static int amd64_dwarf_regmap[] =
163 {
164 /* General Purpose Registers RAX, RDX, RCX, RBX, RSI, RDI. */
165 AMD64_RAX_REGNUM, AMD64_RDX_REGNUM,
166 AMD64_RCX_REGNUM, AMD64_RBX_REGNUM,
167 AMD64_RSI_REGNUM, AMD64_RDI_REGNUM,
168
169 /* Frame Pointer Register RBP. */
170 AMD64_RBP_REGNUM,
171
172 /* Stack Pointer Register RSP. */
173 AMD64_RSP_REGNUM,
174
175 /* Extended Integer Registers 8 - 15. */
176 AMD64_R8_REGNUM, /* %r8 */
177 AMD64_R9_REGNUM, /* %r9 */
178 AMD64_R10_REGNUM, /* %r10 */
179 AMD64_R11_REGNUM, /* %r11 */
180 AMD64_R12_REGNUM, /* %r12 */
181 AMD64_R13_REGNUM, /* %r13 */
182 AMD64_R14_REGNUM, /* %r14 */
183 AMD64_R15_REGNUM, /* %r15 */
184
185 /* Return Address RA. Mapped to RIP. */
186 AMD64_RIP_REGNUM,
187
188 /* SSE Registers 0 - 7. */
189 AMD64_XMM0_REGNUM + 0, AMD64_XMM1_REGNUM,
190 AMD64_XMM0_REGNUM + 2, AMD64_XMM0_REGNUM + 3,
191 AMD64_XMM0_REGNUM + 4, AMD64_XMM0_REGNUM + 5,
192 AMD64_XMM0_REGNUM + 6, AMD64_XMM0_REGNUM + 7,
193
194 /* Extended SSE Registers 8 - 15. */
195 AMD64_XMM0_REGNUM + 8, AMD64_XMM0_REGNUM + 9,
196 AMD64_XMM0_REGNUM + 10, AMD64_XMM0_REGNUM + 11,
197 AMD64_XMM0_REGNUM + 12, AMD64_XMM0_REGNUM + 13,
198 AMD64_XMM0_REGNUM + 14, AMD64_XMM0_REGNUM + 15,
199
200 /* Floating Point Registers 0-7. */
201 AMD64_ST0_REGNUM + 0, AMD64_ST0_REGNUM + 1,
202 AMD64_ST0_REGNUM + 2, AMD64_ST0_REGNUM + 3,
203 AMD64_ST0_REGNUM + 4, AMD64_ST0_REGNUM + 5,
204 AMD64_ST0_REGNUM + 6, AMD64_ST0_REGNUM + 7,
205
206 /* MMX Registers 0 - 7.
207 We have to handle those registers specifically, as their register
208 number within GDB depends on the target (or they may even not be
209 available at all). */
210 -1, -1, -1, -1, -1, -1, -1, -1,
211
212 /* Control and Status Flags Register. */
213 AMD64_EFLAGS_REGNUM,
214
215 /* Selector Registers. */
216 AMD64_ES_REGNUM,
217 AMD64_CS_REGNUM,
218 AMD64_SS_REGNUM,
219 AMD64_DS_REGNUM,
220 AMD64_FS_REGNUM,
221 AMD64_GS_REGNUM,
222 -1,
223 -1,
224
225 /* Segment Base Address Registers. */
226 -1,
227 -1,
228 -1,
229 -1,
230
231 /* Special Selector Registers. */
232 -1,
233 -1,
234
235 /* Floating Point Control Registers. */
236 AMD64_MXCSR_REGNUM,
237 AMD64_FCTRL_REGNUM,
238 AMD64_FSTAT_REGNUM
239 };
240
241 static const int amd64_dwarf_regmap_len =
242 (sizeof (amd64_dwarf_regmap) / sizeof (amd64_dwarf_regmap[0]));
243
244 /* Convert DWARF register number REG to the appropriate register
245 number used by GDB. */
246
247 static int
248 amd64_dwarf_reg_to_regnum (struct gdbarch *gdbarch, int reg)
249 {
250 i386_gdbarch_tdep *tdep = (i386_gdbarch_tdep *) gdbarch_tdep (gdbarch);
251 int ymm0_regnum = tdep->ymm0_regnum;
252 int regnum = -1;
253
254 if (reg >= 0 && reg < amd64_dwarf_regmap_len)
255 regnum = amd64_dwarf_regmap[reg];
256
257 if (ymm0_regnum >= 0
258 && i386_xmm_regnum_p (gdbarch, regnum))
259 regnum += ymm0_regnum - I387_XMM0_REGNUM (tdep);
260
261 return regnum;
262 }
263
264 /* Map architectural register numbers to gdb register numbers. */
265
266 static const int amd64_arch_regmap[16] =
267 {
268 AMD64_RAX_REGNUM, /* %rax */
269 AMD64_RCX_REGNUM, /* %rcx */
270 AMD64_RDX_REGNUM, /* %rdx */
271 AMD64_RBX_REGNUM, /* %rbx */
272 AMD64_RSP_REGNUM, /* %rsp */
273 AMD64_RBP_REGNUM, /* %rbp */
274 AMD64_RSI_REGNUM, /* %rsi */
275 AMD64_RDI_REGNUM, /* %rdi */
276 AMD64_R8_REGNUM, /* %r8 */
277 AMD64_R9_REGNUM, /* %r9 */
278 AMD64_R10_REGNUM, /* %r10 */
279 AMD64_R11_REGNUM, /* %r11 */
280 AMD64_R12_REGNUM, /* %r12 */
281 AMD64_R13_REGNUM, /* %r13 */
282 AMD64_R14_REGNUM, /* %r14 */
283 AMD64_R15_REGNUM /* %r15 */
284 };
285
286 static const int amd64_arch_regmap_len =
287 (sizeof (amd64_arch_regmap) / sizeof (amd64_arch_regmap[0]));
288
289 /* Convert architectural register number REG to the appropriate register
290 number used by GDB. */
291
292 static int
293 amd64_arch_reg_to_regnum (int reg)
294 {
295 gdb_assert (reg >= 0 && reg < amd64_arch_regmap_len);
296
297 return amd64_arch_regmap[reg];
298 }
299
300 /* Register names for byte pseudo-registers. */
301
302 static const char * const amd64_byte_names[] =
303 {
304 "al", "bl", "cl", "dl", "sil", "dil", "bpl", "spl",
305 "r8l", "r9l", "r10l", "r11l", "r12l", "r13l", "r14l", "r15l",
306 "ah", "bh", "ch", "dh"
307 };
308
309 /* Number of lower byte registers. */
310 #define AMD64_NUM_LOWER_BYTE_REGS 16
311
312 /* Register names for word pseudo-registers. */
313
314 static const char * const amd64_word_names[] =
315 {
316 "ax", "bx", "cx", "dx", "si", "di", "bp", "",
317 "r8w", "r9w", "r10w", "r11w", "r12w", "r13w", "r14w", "r15w"
318 };
319
320 /* Register names for dword pseudo-registers. */
321
322 static const char * const amd64_dword_names[] =
323 {
324 "eax", "ebx", "ecx", "edx", "esi", "edi", "ebp", "esp",
325 "r8d", "r9d", "r10d", "r11d", "r12d", "r13d", "r14d", "r15d",
326 "eip"
327 };
328
329 /* Return the name of register REGNUM. */
330
331 static const char *
332 amd64_pseudo_register_name (struct gdbarch *gdbarch, int regnum)
333 {
334 i386_gdbarch_tdep *tdep = (i386_gdbarch_tdep *) gdbarch_tdep (gdbarch);
335 if (i386_byte_regnum_p (gdbarch, regnum))
336 return amd64_byte_names[regnum - tdep->al_regnum];
337 else if (i386_zmm_regnum_p (gdbarch, regnum))
338 return amd64_zmm_names[regnum - tdep->zmm0_regnum];
339 else if (i386_ymm_regnum_p (gdbarch, regnum))
340 return amd64_ymm_names[regnum - tdep->ymm0_regnum];
341 else if (i386_ymm_avx512_regnum_p (gdbarch, regnum))
342 return amd64_ymm_avx512_names[regnum - tdep->ymm16_regnum];
343 else if (i386_word_regnum_p (gdbarch, regnum))
344 return amd64_word_names[regnum - tdep->ax_regnum];
345 else if (i386_dword_regnum_p (gdbarch, regnum))
346 return amd64_dword_names[regnum - tdep->eax_regnum];
347 else
348 return i386_pseudo_register_name (gdbarch, regnum);
349 }
350
351 static struct value *
352 amd64_pseudo_register_read_value (struct gdbarch *gdbarch,
353 readable_regcache *regcache,
354 int regnum)
355 {
356 i386_gdbarch_tdep *tdep = (i386_gdbarch_tdep *) gdbarch_tdep (gdbarch);
357
358 value *result_value = allocate_value (register_type (gdbarch, regnum));
359 VALUE_LVAL (result_value) = lval_register;
360 VALUE_REGNUM (result_value) = regnum;
361 gdb_byte *buf = value_contents_raw (result_value).data ();
362
363 if (i386_byte_regnum_p (gdbarch, regnum))
364 {
365 int gpnum = regnum - tdep->al_regnum;
366
367 /* Extract (always little endian). */
368 if (gpnum >= AMD64_NUM_LOWER_BYTE_REGS)
369 {
370 gpnum -= AMD64_NUM_LOWER_BYTE_REGS;
371 gdb_byte raw_buf[register_size (gdbarch, gpnum)];
372
373 /* Special handling for AH, BH, CH, DH. */
374 register_status status = regcache->raw_read (gpnum, raw_buf);
375 if (status == REG_VALID)
376 memcpy (buf, raw_buf + 1, 1);
377 else
378 mark_value_bytes_unavailable (result_value, 0,
379 TYPE_LENGTH (value_type (result_value)));
380 }
381 else
382 {
383 gdb_byte raw_buf[register_size (gdbarch, gpnum)];
384 register_status status = regcache->raw_read (gpnum, raw_buf);
385 if (status == REG_VALID)
386 memcpy (buf, raw_buf, 1);
387 else
388 mark_value_bytes_unavailable (result_value, 0,
389 TYPE_LENGTH (value_type (result_value)));
390 }
391 }
392 else if (i386_dword_regnum_p (gdbarch, regnum))
393 {
394 int gpnum = regnum - tdep->eax_regnum;
395 gdb_byte raw_buf[register_size (gdbarch, gpnum)];
396 /* Extract (always little endian). */
397 register_status status = regcache->raw_read (gpnum, raw_buf);
398 if (status == REG_VALID)
399 memcpy (buf, raw_buf, 4);
400 else
401 mark_value_bytes_unavailable (result_value, 0,
402 TYPE_LENGTH (value_type (result_value)));
403 }
404 else
405 i386_pseudo_register_read_into_value (gdbarch, regcache, regnum,
406 result_value);
407
408 return result_value;
409 }
410
411 static void
412 amd64_pseudo_register_write (struct gdbarch *gdbarch,
413 struct regcache *regcache,
414 int regnum, const gdb_byte *buf)
415 {
416 i386_gdbarch_tdep *tdep = (i386_gdbarch_tdep *) gdbarch_tdep (gdbarch);
417
418 if (i386_byte_regnum_p (gdbarch, regnum))
419 {
420 int gpnum = regnum - tdep->al_regnum;
421
422 if (gpnum >= AMD64_NUM_LOWER_BYTE_REGS)
423 {
424 gpnum -= AMD64_NUM_LOWER_BYTE_REGS;
425 gdb_byte raw_buf[register_size (gdbarch, gpnum)];
426
427 /* Read ... AH, BH, CH, DH. */
428 regcache->raw_read (gpnum, raw_buf);
429 /* ... Modify ... (always little endian). */
430 memcpy (raw_buf + 1, buf, 1);
431 /* ... Write. */
432 regcache->raw_write (gpnum, raw_buf);
433 }
434 else
435 {
436 gdb_byte raw_buf[register_size (gdbarch, gpnum)];
437
438 /* Read ... */
439 regcache->raw_read (gpnum, raw_buf);
440 /* ... Modify ... (always little endian). */
441 memcpy (raw_buf, buf, 1);
442 /* ... Write. */
443 regcache->raw_write (gpnum, raw_buf);
444 }
445 }
446 else if (i386_dword_regnum_p (gdbarch, regnum))
447 {
448 int gpnum = regnum - tdep->eax_regnum;
449 gdb_byte raw_buf[register_size (gdbarch, gpnum)];
450
451 /* Read ... */
452 regcache->raw_read (gpnum, raw_buf);
453 /* ... Modify ... (always little endian). */
454 memcpy (raw_buf, buf, 4);
455 /* ... Write. */
456 regcache->raw_write (gpnum, raw_buf);
457 }
458 else
459 i386_pseudo_register_write (gdbarch, regcache, regnum, buf);
460 }
461
462 /* Implement the 'ax_pseudo_register_collect' gdbarch method. */
463
464 static int
465 amd64_ax_pseudo_register_collect (struct gdbarch *gdbarch,
466 struct agent_expr *ax, int regnum)
467 {
468 i386_gdbarch_tdep *tdep = (i386_gdbarch_tdep *) gdbarch_tdep (gdbarch);
469
470 if (i386_byte_regnum_p (gdbarch, regnum))
471 {
472 int gpnum = regnum - tdep->al_regnum;
473
474 if (gpnum >= AMD64_NUM_LOWER_BYTE_REGS)
475 ax_reg_mask (ax, gpnum - AMD64_NUM_LOWER_BYTE_REGS);
476 else
477 ax_reg_mask (ax, gpnum);
478 return 0;
479 }
480 else if (i386_dword_regnum_p (gdbarch, regnum))
481 {
482 int gpnum = regnum - tdep->eax_regnum;
483
484 ax_reg_mask (ax, gpnum);
485 return 0;
486 }
487 else
488 return i386_ax_pseudo_register_collect (gdbarch, ax, regnum);
489 }
490
491 \f
492
493 /* Register classes as defined in the psABI. */
494
495 enum amd64_reg_class
496 {
497 AMD64_INTEGER,
498 AMD64_SSE,
499 AMD64_SSEUP,
500 AMD64_X87,
501 AMD64_X87UP,
502 AMD64_COMPLEX_X87,
503 AMD64_NO_CLASS,
504 AMD64_MEMORY
505 };
506
507 /* Return the union class of CLASS1 and CLASS2. See the psABI for
508 details. */
509
510 static enum amd64_reg_class
511 amd64_merge_classes (enum amd64_reg_class class1, enum amd64_reg_class class2)
512 {
513 /* Rule (a): If both classes are equal, this is the resulting class. */
514 if (class1 == class2)
515 return class1;
516
517 /* Rule (b): If one of the classes is NO_CLASS, the resulting class
518 is the other class. */
519 if (class1 == AMD64_NO_CLASS)
520 return class2;
521 if (class2 == AMD64_NO_CLASS)
522 return class1;
523
524 /* Rule (c): If one of the classes is MEMORY, the result is MEMORY. */
525 if (class1 == AMD64_MEMORY || class2 == AMD64_MEMORY)
526 return AMD64_MEMORY;
527
528 /* Rule (d): If one of the classes is INTEGER, the result is INTEGER. */
529 if (class1 == AMD64_INTEGER || class2 == AMD64_INTEGER)
530 return AMD64_INTEGER;
531
532 /* Rule (e): If one of the classes is X87, X87UP, COMPLEX_X87 class,
533 MEMORY is used as class. */
534 if (class1 == AMD64_X87 || class1 == AMD64_X87UP
535 || class1 == AMD64_COMPLEX_X87 || class2 == AMD64_X87
536 || class2 == AMD64_X87UP || class2 == AMD64_COMPLEX_X87)
537 return AMD64_MEMORY;
538
539 /* Rule (f): Otherwise class SSE is used. */
540 return AMD64_SSE;
541 }
542
543 static void amd64_classify (struct type *type, enum amd64_reg_class theclass[2]);
544
545 /* Return true if TYPE is a structure or union with unaligned fields. */
546
547 static bool
548 amd64_has_unaligned_fields (struct type *type)
549 {
550 if (type->code () == TYPE_CODE_STRUCT
551 || type->code () == TYPE_CODE_UNION)
552 {
553 for (int i = 0; i < type->num_fields (); i++)
554 {
555 struct type *subtype = check_typedef (type->field (i).type ());
556
557 /* Ignore static fields, empty fields (for example nested
558 empty structures), and bitfields (these are handled by
559 the caller). */
560 if (field_is_static (&type->field (i))
561 || (TYPE_FIELD_BITSIZE (type, i) == 0
562 && TYPE_LENGTH (subtype) == 0)
563 || TYPE_FIELD_PACKED (type, i))
564 continue;
565
566 int bitpos = type->field (i).loc_bitpos ();
567
568 if (bitpos % 8 != 0)
569 return true;
570
571 int align = type_align (subtype);
572 if (align == 0)
573 error (_("could not determine alignment of type"));
574
575 int bytepos = bitpos / 8;
576 if (bytepos % align != 0)
577 return true;
578
579 if (amd64_has_unaligned_fields (subtype))
580 return true;
581 }
582 }
583
584 return false;
585 }
586
587 /* Classify field I of TYPE starting at BITOFFSET according to the rules for
588 structures and union types, and store the result in THECLASS. */
589
590 static void
591 amd64_classify_aggregate_field (struct type *type, int i,
592 enum amd64_reg_class theclass[2],
593 unsigned int bitoffset)
594 {
595 struct type *subtype = check_typedef (type->field (i).type ());
596 enum amd64_reg_class subclass[2];
597 int bitsize = TYPE_FIELD_BITSIZE (type, i);
598
599 if (bitsize == 0)
600 bitsize = TYPE_LENGTH (subtype) * 8;
601
602 /* Ignore static fields, or empty fields, for example nested
603 empty structures.*/
604 if (field_is_static (&type->field (i)) || bitsize == 0)
605 return;
606
607 int bitpos = bitoffset + type->field (i).loc_bitpos ();
608 int pos = bitpos / 64;
609 int endpos = (bitpos + bitsize - 1) / 64;
610
611 if (subtype->code () == TYPE_CODE_STRUCT
612 || subtype->code () == TYPE_CODE_UNION)
613 {
614 /* Each field of an object is classified recursively. */
615 int j;
616 for (j = 0; j < subtype->num_fields (); j++)
617 amd64_classify_aggregate_field (subtype, j, theclass, bitpos);
618 return;
619 }
620
621 gdb_assert (pos == 0 || pos == 1);
622
623 amd64_classify (subtype, subclass);
624 theclass[pos] = amd64_merge_classes (theclass[pos], subclass[0]);
625 if (bitsize <= 64 && pos == 0 && endpos == 1)
626 /* This is a bit of an odd case: We have a field that would
627 normally fit in one of the two eightbytes, except that
628 it is placed in a way that this field straddles them.
629 This has been seen with a structure containing an array.
630
631 The ABI is a bit unclear in this case, but we assume that
632 this field's class (stored in subclass[0]) must also be merged
633 into class[1]. In other words, our field has a piece stored
634 in the second eight-byte, and thus its class applies to
635 the second eight-byte as well.
636
637 In the case where the field length exceeds 8 bytes,
638 it should not be necessary to merge the field class
639 into class[1]. As LEN > 8, subclass[1] is necessarily
640 different from AMD64_NO_CLASS. If subclass[1] is equal
641 to subclass[0], then the normal class[1]/subclass[1]
642 merging will take care of everything. For subclass[1]
643 to be different from subclass[0], I can only see the case
644 where we have a SSE/SSEUP or X87/X87UP pair, which both
645 use up all 16 bytes of the aggregate, and are already
646 handled just fine (because each portion sits on its own
647 8-byte). */
648 theclass[1] = amd64_merge_classes (theclass[1], subclass[0]);
649 if (pos == 0)
650 theclass[1] = amd64_merge_classes (theclass[1], subclass[1]);
651 }
652
653 /* Classify TYPE according to the rules for aggregate (structures and
654 arrays) and union types, and store the result in CLASS. */
655
656 static void
657 amd64_classify_aggregate (struct type *type, enum amd64_reg_class theclass[2])
658 {
659 /* 1. If the size of an object is larger than two times eight bytes, or
660 it is a non-trivial C++ object, or it has unaligned fields, then it
661 has class memory.
662
663 It is important that the trivially_copyable check is before the
664 unaligned fields check, as C++ classes with virtual base classes
665 will have fields (for the virtual base classes) with non-constant
666 loc_bitpos attributes, which will cause an assert to trigger within
667 the unaligned field check. As classes with virtual bases are not
668 trivially copyable, checking that first avoids this problem. */
669 if (TYPE_LENGTH (type) > 16
670 || !language_pass_by_reference (type).trivially_copyable
671 || amd64_has_unaligned_fields (type))
672 {
673 theclass[0] = theclass[1] = AMD64_MEMORY;
674 return;
675 }
676
677 /* 2. Both eightbytes get initialized to class NO_CLASS. */
678 theclass[0] = theclass[1] = AMD64_NO_CLASS;
679
680 /* 3. Each field of an object is classified recursively so that
681 always two fields are considered. The resulting class is
682 calculated according to the classes of the fields in the
683 eightbyte: */
684
685 if (type->code () == TYPE_CODE_ARRAY)
686 {
687 struct type *subtype = check_typedef (TYPE_TARGET_TYPE (type));
688
689 /* All fields in an array have the same type. */
690 amd64_classify (subtype, theclass);
691 if (TYPE_LENGTH (type) > 8 && theclass[1] == AMD64_NO_CLASS)
692 theclass[1] = theclass[0];
693 }
694 else
695 {
696 int i;
697
698 /* Structure or union. */
699 gdb_assert (type->code () == TYPE_CODE_STRUCT
700 || type->code () == TYPE_CODE_UNION);
701
702 for (i = 0; i < type->num_fields (); i++)
703 amd64_classify_aggregate_field (type, i, theclass, 0);
704 }
705
706 /* 4. Then a post merger cleanup is done: */
707
708 /* Rule (a): If one of the classes is MEMORY, the whole argument is
709 passed in memory. */
710 if (theclass[0] == AMD64_MEMORY || theclass[1] == AMD64_MEMORY)
711 theclass[0] = theclass[1] = AMD64_MEMORY;
712
713 /* Rule (b): If SSEUP is not preceded by SSE, it is converted to
714 SSE. */
715 if (theclass[0] == AMD64_SSEUP)
716 theclass[0] = AMD64_SSE;
717 if (theclass[1] == AMD64_SSEUP && theclass[0] != AMD64_SSE)
718 theclass[1] = AMD64_SSE;
719 }
720
721 /* Classify TYPE, and store the result in CLASS. */
722
723 static void
724 amd64_classify (struct type *type, enum amd64_reg_class theclass[2])
725 {
726 enum type_code code = type->code ();
727 int len = TYPE_LENGTH (type);
728
729 theclass[0] = theclass[1] = AMD64_NO_CLASS;
730
731 /* Arguments of types (signed and unsigned) _Bool, char, short, int,
732 long, long long, and pointers are in the INTEGER class. Similarly,
733 range types, used by languages such as Ada, are also in the INTEGER
734 class. */
735 if ((code == TYPE_CODE_INT || code == TYPE_CODE_ENUM
736 || code == TYPE_CODE_BOOL || code == TYPE_CODE_RANGE
737 || code == TYPE_CODE_CHAR
738 || code == TYPE_CODE_PTR || TYPE_IS_REFERENCE (type))
739 && (len == 1 || len == 2 || len == 4 || len == 8))
740 theclass[0] = AMD64_INTEGER;
741
742 /* Arguments of types _Float16, float, double, _Decimal32, _Decimal64 and
743 __m64 are in class SSE. */
744 else if ((code == TYPE_CODE_FLT || code == TYPE_CODE_DECFLOAT)
745 && (len == 2 || len == 4 || len == 8))
746 /* FIXME: __m64 . */
747 theclass[0] = AMD64_SSE;
748
749 /* Arguments of types __float128, _Decimal128 and __m128 are split into
750 two halves. The least significant ones belong to class SSE, the most
751 significant one to class SSEUP. */
752 else if (code == TYPE_CODE_DECFLOAT && len == 16)
753 /* FIXME: __float128, __m128. */
754 theclass[0] = AMD64_SSE, theclass[1] = AMD64_SSEUP;
755
756 /* The 64-bit mantissa of arguments of type long double belongs to
757 class X87, the 16-bit exponent plus 6 bytes of padding belongs to
758 class X87UP. */
759 else if (code == TYPE_CODE_FLT && len == 16)
760 /* Class X87 and X87UP. */
761 theclass[0] = AMD64_X87, theclass[1] = AMD64_X87UP;
762
763 /* Arguments of complex T - where T is one of the types _Float16, float or
764 double - get treated as if they are implemented as:
765
766 struct complexT {
767 T real;
768 T imag;
769 };
770
771 */
772 else if (code == TYPE_CODE_COMPLEX && (len == 8 || len == 4))
773 theclass[0] = AMD64_SSE;
774 else if (code == TYPE_CODE_COMPLEX && len == 16)
775 theclass[0] = theclass[1] = AMD64_SSE;
776
777 /* A variable of type complex long double is classified as type
778 COMPLEX_X87. */
779 else if (code == TYPE_CODE_COMPLEX && len == 32)
780 theclass[0] = AMD64_COMPLEX_X87;
781
782 /* Aggregates. */
783 else if (code == TYPE_CODE_ARRAY || code == TYPE_CODE_STRUCT
784 || code == TYPE_CODE_UNION)
785 amd64_classify_aggregate (type, theclass);
786 }
787
788 static enum return_value_convention
789 amd64_return_value (struct gdbarch *gdbarch, struct value *function,
790 struct type *type, struct regcache *regcache,
791 gdb_byte *readbuf, const gdb_byte *writebuf)
792 {
793 enum amd64_reg_class theclass[2];
794 int len = TYPE_LENGTH (type);
795 static int integer_regnum[] = { AMD64_RAX_REGNUM, AMD64_RDX_REGNUM };
796 static int sse_regnum[] = { AMD64_XMM0_REGNUM, AMD64_XMM1_REGNUM };
797 int integer_reg = 0;
798 int sse_reg = 0;
799 int i;
800
801 gdb_assert (!(readbuf && writebuf));
802
803 /* 1. Classify the return type with the classification algorithm. */
804 amd64_classify (type, theclass);
805
806 /* 2. If the type has class MEMORY, then the caller provides space
807 for the return value and passes the address of this storage in
808 %rdi as if it were the first argument to the function. In effect,
809 this address becomes a hidden first argument.
810
811 On return %rax will contain the address that has been passed in
812 by the caller in %rdi. */
813 if (theclass[0] == AMD64_MEMORY)
814 {
815 /* As indicated by the comment above, the ABI guarantees that we
816 can always find the return value just after the function has
817 returned. */
818
819 if (readbuf)
820 {
821 ULONGEST addr;
822
823 regcache_raw_read_unsigned (regcache, AMD64_RAX_REGNUM, &addr);
824 read_memory (addr, readbuf, TYPE_LENGTH (type));
825 }
826
827 return RETURN_VALUE_ABI_RETURNS_ADDRESS;
828 }
829
830 /* 8. If the class is COMPLEX_X87, the real part of the value is
831 returned in %st0 and the imaginary part in %st1. */
832 if (theclass[0] == AMD64_COMPLEX_X87)
833 {
834 if (readbuf)
835 {
836 regcache->raw_read (AMD64_ST0_REGNUM, readbuf);
837 regcache->raw_read (AMD64_ST1_REGNUM, readbuf + 16);
838 }
839
840 if (writebuf)
841 {
842 i387_return_value (gdbarch, regcache);
843 regcache->raw_write (AMD64_ST0_REGNUM, writebuf);
844 regcache->raw_write (AMD64_ST1_REGNUM, writebuf + 16);
845
846 /* Fix up the tag word such that both %st(0) and %st(1) are
847 marked as valid. */
848 regcache_raw_write_unsigned (regcache, AMD64_FTAG_REGNUM, 0xfff);
849 }
850
851 return RETURN_VALUE_REGISTER_CONVENTION;
852 }
853
854 gdb_assert (theclass[1] != AMD64_MEMORY);
855 gdb_assert (len <= 16);
856
857 for (i = 0; len > 0; i++, len -= 8)
858 {
859 int regnum = -1;
860 int offset = 0;
861
862 switch (theclass[i])
863 {
864 case AMD64_INTEGER:
865 /* 3. If the class is INTEGER, the next available register
866 of the sequence %rax, %rdx is used. */
867 regnum = integer_regnum[integer_reg++];
868 break;
869
870 case AMD64_SSE:
871 /* 4. If the class is SSE, the next available SSE register
872 of the sequence %xmm0, %xmm1 is used. */
873 regnum = sse_regnum[sse_reg++];
874 break;
875
876 case AMD64_SSEUP:
877 /* 5. If the class is SSEUP, the eightbyte is passed in the
878 upper half of the last used SSE register. */
879 gdb_assert (sse_reg > 0);
880 regnum = sse_regnum[sse_reg - 1];
881 offset = 8;
882 break;
883
884 case AMD64_X87:
885 /* 6. If the class is X87, the value is returned on the X87
886 stack in %st0 as 80-bit x87 number. */
887 regnum = AMD64_ST0_REGNUM;
888 if (writebuf)
889 i387_return_value (gdbarch, regcache);
890 break;
891
892 case AMD64_X87UP:
893 /* 7. If the class is X87UP, the value is returned together
894 with the previous X87 value in %st0. */
895 gdb_assert (i > 0 && theclass[0] == AMD64_X87);
896 regnum = AMD64_ST0_REGNUM;
897 offset = 8;
898 len = 2;
899 break;
900
901 case AMD64_NO_CLASS:
902 continue;
903
904 default:
905 gdb_assert (!"Unexpected register class.");
906 }
907
908 gdb_assert (regnum != -1);
909
910 if (readbuf)
911 regcache->raw_read_part (regnum, offset, std::min (len, 8),
912 readbuf + i * 8);
913 if (writebuf)
914 regcache->raw_write_part (regnum, offset, std::min (len, 8),
915 writebuf + i * 8);
916 }
917
918 return RETURN_VALUE_REGISTER_CONVENTION;
919 }
920 \f
921
922 static CORE_ADDR
923 amd64_push_arguments (struct regcache *regcache, int nargs, struct value **args,
924 CORE_ADDR sp, function_call_return_method return_method)
925 {
926 static int integer_regnum[] =
927 {
928 AMD64_RDI_REGNUM, /* %rdi */
929 AMD64_RSI_REGNUM, /* %rsi */
930 AMD64_RDX_REGNUM, /* %rdx */
931 AMD64_RCX_REGNUM, /* %rcx */
932 AMD64_R8_REGNUM, /* %r8 */
933 AMD64_R9_REGNUM /* %r9 */
934 };
935 static int sse_regnum[] =
936 {
937 /* %xmm0 ... %xmm7 */
938 AMD64_XMM0_REGNUM + 0, AMD64_XMM1_REGNUM,
939 AMD64_XMM0_REGNUM + 2, AMD64_XMM0_REGNUM + 3,
940 AMD64_XMM0_REGNUM + 4, AMD64_XMM0_REGNUM + 5,
941 AMD64_XMM0_REGNUM + 6, AMD64_XMM0_REGNUM + 7,
942 };
943 struct value **stack_args = XALLOCAVEC (struct value *, nargs);
944 int num_stack_args = 0;
945 int num_elements = 0;
946 int element = 0;
947 int integer_reg = 0;
948 int sse_reg = 0;
949 int i;
950
951 /* Reserve a register for the "hidden" argument. */
952 if (return_method == return_method_struct)
953 integer_reg++;
954
955 for (i = 0; i < nargs; i++)
956 {
957 struct type *type = value_type (args[i]);
958 int len = TYPE_LENGTH (type);
959 enum amd64_reg_class theclass[2];
960 int needed_integer_regs = 0;
961 int needed_sse_regs = 0;
962 int j;
963
964 /* Classify argument. */
965 amd64_classify (type, theclass);
966
967 /* Calculate the number of integer and SSE registers needed for
968 this argument. */
969 for (j = 0; j < 2; j++)
970 {
971 if (theclass[j] == AMD64_INTEGER)
972 needed_integer_regs++;
973 else if (theclass[j] == AMD64_SSE)
974 needed_sse_regs++;
975 }
976
977 /* Check whether enough registers are available, and if the
978 argument should be passed in registers at all. */
979 if (integer_reg + needed_integer_regs > ARRAY_SIZE (integer_regnum)
980 || sse_reg + needed_sse_regs > ARRAY_SIZE (sse_regnum)
981 || (needed_integer_regs == 0 && needed_sse_regs == 0))
982 {
983 /* The argument will be passed on the stack. */
984 num_elements += ((len + 7) / 8);
985 stack_args[num_stack_args++] = args[i];
986 }
987 else
988 {
989 /* The argument will be passed in registers. */
990 const gdb_byte *valbuf = value_contents (args[i]).data ();
991 gdb_byte buf[8];
992
993 gdb_assert (len <= 16);
994
995 for (j = 0; len > 0; j++, len -= 8)
996 {
997 int regnum = -1;
998 int offset = 0;
999
1000 switch (theclass[j])
1001 {
1002 case AMD64_INTEGER:
1003 regnum = integer_regnum[integer_reg++];
1004 break;
1005
1006 case AMD64_SSE:
1007 regnum = sse_regnum[sse_reg++];
1008 break;
1009
1010 case AMD64_SSEUP:
1011 gdb_assert (sse_reg > 0);
1012 regnum = sse_regnum[sse_reg - 1];
1013 offset = 8;
1014 break;
1015
1016 case AMD64_NO_CLASS:
1017 continue;
1018
1019 default:
1020 gdb_assert (!"Unexpected register class.");
1021 }
1022
1023 gdb_assert (regnum != -1);
1024 memset (buf, 0, sizeof buf);
1025 memcpy (buf, valbuf + j * 8, std::min (len, 8));
1026 regcache->raw_write_part (regnum, offset, 8, buf);
1027 }
1028 }
1029 }
1030
1031 /* Allocate space for the arguments on the stack. */
1032 sp -= num_elements * 8;
1033
1034 /* The psABI says that "The end of the input argument area shall be
1035 aligned on a 16 byte boundary." */
1036 sp &= ~0xf;
1037
1038 /* Write out the arguments to the stack. */
1039 for (i = 0; i < num_stack_args; i++)
1040 {
1041 struct type *type = value_type (stack_args[i]);
1042 const gdb_byte *valbuf = value_contents (stack_args[i]).data ();
1043 int len = TYPE_LENGTH (type);
1044
1045 write_memory (sp + element * 8, valbuf, len);
1046 element += ((len + 7) / 8);
1047 }
1048
1049 /* The psABI says that "For calls that may call functions that use
1050 varargs or stdargs (prototype-less calls or calls to functions
1051 containing ellipsis (...) in the declaration) %al is used as
1052 hidden argument to specify the number of SSE registers used. */
1053 regcache_raw_write_unsigned (regcache, AMD64_RAX_REGNUM, sse_reg);
1054 return sp;
1055 }
1056
1057 static CORE_ADDR
1058 amd64_push_dummy_call (struct gdbarch *gdbarch, struct value *function,
1059 struct regcache *regcache, CORE_ADDR bp_addr,
1060 int nargs, struct value **args, CORE_ADDR sp,
1061 function_call_return_method return_method,
1062 CORE_ADDR struct_addr)
1063 {
1064 enum bfd_endian byte_order = gdbarch_byte_order (gdbarch);
1065 gdb_byte buf[8];
1066
1067 /* BND registers can be in arbitrary values at the moment of the
1068 inferior call. This can cause boundary violations that are not
1069 due to a real bug or even desired by the user. The best to be done
1070 is set the BND registers to allow access to the whole memory, INIT
1071 state, before pushing the inferior call. */
1072 i387_reset_bnd_regs (gdbarch, regcache);
1073
1074 /* Pass arguments. */
1075 sp = amd64_push_arguments (regcache, nargs, args, sp, return_method);
1076
1077 /* Pass "hidden" argument". */
1078 if (return_method == return_method_struct)
1079 {
1080 store_unsigned_integer (buf, 8, byte_order, struct_addr);
1081 regcache->cooked_write (AMD64_RDI_REGNUM, buf);
1082 }
1083
1084 /* Store return address. */
1085 sp -= 8;
1086 store_unsigned_integer (buf, 8, byte_order, bp_addr);
1087 write_memory (sp, buf, 8);
1088
1089 /* Finally, update the stack pointer... */
1090 store_unsigned_integer (buf, 8, byte_order, sp);
1091 regcache->cooked_write (AMD64_RSP_REGNUM, buf);
1092
1093 /* ...and fake a frame pointer. */
1094 regcache->cooked_write (AMD64_RBP_REGNUM, buf);
1095
1096 return sp + 16;
1097 }
1098 \f
1099 /* Displaced instruction handling. */
1100
1101 /* A partially decoded instruction.
1102 This contains enough details for displaced stepping purposes. */
1103
1104 struct amd64_insn
1105 {
1106 /* The number of opcode bytes. */
1107 int opcode_len;
1108 /* The offset of the REX/VEX instruction encoding prefix or -1 if
1109 not present. */
1110 int enc_prefix_offset;
1111 /* The offset to the first opcode byte. */
1112 int opcode_offset;
1113 /* The offset to the modrm byte or -1 if not present. */
1114 int modrm_offset;
1115
1116 /* The raw instruction. */
1117 gdb_byte *raw_insn;
1118 };
1119
1120 struct amd64_displaced_step_copy_insn_closure
1121 : public displaced_step_copy_insn_closure
1122 {
1123 amd64_displaced_step_copy_insn_closure (int insn_buf_len)
1124 : insn_buf (insn_buf_len, 0)
1125 {}
1126
1127 /* For rip-relative insns, saved copy of the reg we use instead of %rip. */
1128 int tmp_used = 0;
1129 int tmp_regno;
1130 ULONGEST tmp_save;
1131
1132 /* Details of the instruction. */
1133 struct amd64_insn insn_details;
1134
1135 /* The possibly modified insn. */
1136 gdb::byte_vector insn_buf;
1137 };
1138
1139 /* WARNING: Keep onebyte_has_modrm, twobyte_has_modrm in sync with
1140 ../opcodes/i386-dis.c (until libopcodes exports them, or an alternative,
1141 at which point delete these in favor of libopcodes' versions). */
1142
1143 static const unsigned char onebyte_has_modrm[256] = {
1144 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
1145 /* ------------------------------- */
1146 /* 00 */ 1,1,1,1,0,0,0,0,1,1,1,1,0,0,0,0, /* 00 */
1147 /* 10 */ 1,1,1,1,0,0,0,0,1,1,1,1,0,0,0,0, /* 10 */
1148 /* 20 */ 1,1,1,1,0,0,0,0,1,1,1,1,0,0,0,0, /* 20 */
1149 /* 30 */ 1,1,1,1,0,0,0,0,1,1,1,1,0,0,0,0, /* 30 */
1150 /* 40 */ 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /* 40 */
1151 /* 50 */ 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /* 50 */
1152 /* 60 */ 0,0,1,1,0,0,0,0,0,1,0,1,0,0,0,0, /* 60 */
1153 /* 70 */ 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /* 70 */
1154 /* 80 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* 80 */
1155 /* 90 */ 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /* 90 */
1156 /* a0 */ 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /* a0 */
1157 /* b0 */ 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /* b0 */
1158 /* c0 */ 1,1,0,0,1,1,1,1,0,0,0,0,0,0,0,0, /* c0 */
1159 /* d0 */ 1,1,1,1,0,0,0,0,1,1,1,1,1,1,1,1, /* d0 */
1160 /* e0 */ 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /* e0 */
1161 /* f0 */ 0,0,0,0,0,0,1,1,0,0,0,0,0,0,1,1 /* f0 */
1162 /* ------------------------------- */
1163 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
1164 };
1165
1166 static const unsigned char twobyte_has_modrm[256] = {
1167 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
1168 /* ------------------------------- */
1169 /* 00 */ 1,1,1,1,0,0,0,0,0,0,0,0,0,1,0,1, /* 0f */
1170 /* 10 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* 1f */
1171 /* 20 */ 1,1,1,1,1,1,1,0,1,1,1,1,1,1,1,1, /* 2f */
1172 /* 30 */ 0,0,0,0,0,0,0,0,1,0,1,0,0,0,0,0, /* 3f */
1173 /* 40 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* 4f */
1174 /* 50 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* 5f */
1175 /* 60 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* 6f */
1176 /* 70 */ 1,1,1,1,1,1,1,0,1,1,1,1,1,1,1,1, /* 7f */
1177 /* 80 */ 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /* 8f */
1178 /* 90 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* 9f */
1179 /* a0 */ 0,0,0,1,1,1,1,1,0,0,0,1,1,1,1,1, /* af */
1180 /* b0 */ 1,1,1,1,1,1,1,1,1,0,1,1,1,1,1,1, /* bf */
1181 /* c0 */ 1,1,1,1,1,1,1,1,0,0,0,0,0,0,0,0, /* cf */
1182 /* d0 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* df */
1183 /* e0 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* ef */
1184 /* f0 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0 /* ff */
1185 /* ------------------------------- */
1186 /* 0 1 2 3 4 5 6 7 8 9 a b c d e f */
1187 };
1188
1189 static int amd64_syscall_p (const struct amd64_insn *insn, int *lengthp);
1190
1191 static int
1192 rex_prefix_p (gdb_byte pfx)
1193 {
1194 return REX_PREFIX_P (pfx);
1195 }
1196
1197 /* True if PFX is the start of the 2-byte VEX prefix. */
1198
1199 static bool
1200 vex2_prefix_p (gdb_byte pfx)
1201 {
1202 return pfx == 0xc5;
1203 }
1204
1205 /* True if PFX is the start of the 3-byte VEX prefix. */
1206
1207 static bool
1208 vex3_prefix_p (gdb_byte pfx)
1209 {
1210 return pfx == 0xc4;
1211 }
1212
1213 /* Skip the legacy instruction prefixes in INSN.
1214 We assume INSN is properly sentineled so we don't have to worry
1215 about falling off the end of the buffer. */
1216
1217 static gdb_byte *
1218 amd64_skip_prefixes (gdb_byte *insn)
1219 {
1220 while (1)
1221 {
1222 switch (*insn)
1223 {
1224 case DATA_PREFIX_OPCODE:
1225 case ADDR_PREFIX_OPCODE:
1226 case CS_PREFIX_OPCODE:
1227 case DS_PREFIX_OPCODE:
1228 case ES_PREFIX_OPCODE:
1229 case FS_PREFIX_OPCODE:
1230 case GS_PREFIX_OPCODE:
1231 case SS_PREFIX_OPCODE:
1232 case LOCK_PREFIX_OPCODE:
1233 case REPE_PREFIX_OPCODE:
1234 case REPNE_PREFIX_OPCODE:
1235 ++insn;
1236 continue;
1237 default:
1238 break;
1239 }
1240 break;
1241 }
1242
1243 return insn;
1244 }
1245
1246 /* Return an integer register (other than RSP) that is unused as an input
1247 operand in INSN.
1248 In order to not require adding a rex prefix if the insn doesn't already
1249 have one, the result is restricted to RAX ... RDI, sans RSP.
1250 The register numbering of the result follows architecture ordering,
1251 e.g. RDI = 7. */
1252
1253 static int
1254 amd64_get_unused_input_int_reg (const struct amd64_insn *details)
1255 {
1256 /* 1 bit for each reg */
1257 int used_regs_mask = 0;
1258
1259 /* There can be at most 3 int regs used as inputs in an insn, and we have
1260 7 to choose from (RAX ... RDI, sans RSP).
1261 This allows us to take a conservative approach and keep things simple.
1262 E.g. By avoiding RAX, we don't have to specifically watch for opcodes
1263 that implicitly specify RAX. */
1264
1265 /* Avoid RAX. */
1266 used_regs_mask |= 1 << EAX_REG_NUM;
1267 /* Similarily avoid RDX, implicit operand in divides. */
1268 used_regs_mask |= 1 << EDX_REG_NUM;
1269 /* Avoid RSP. */
1270 used_regs_mask |= 1 << ESP_REG_NUM;
1271
1272 /* If the opcode is one byte long and there's no ModRM byte,
1273 assume the opcode specifies a register. */
1274 if (details->opcode_len == 1 && details->modrm_offset == -1)
1275 used_regs_mask |= 1 << (details->raw_insn[details->opcode_offset] & 7);
1276
1277 /* Mark used regs in the modrm/sib bytes. */
1278 if (details->modrm_offset != -1)
1279 {
1280 int modrm = details->raw_insn[details->modrm_offset];
1281 int mod = MODRM_MOD_FIELD (modrm);
1282 int reg = MODRM_REG_FIELD (modrm);
1283 int rm = MODRM_RM_FIELD (modrm);
1284 int have_sib = mod != 3 && rm == 4;
1285
1286 /* Assume the reg field of the modrm byte specifies a register. */
1287 used_regs_mask |= 1 << reg;
1288
1289 if (have_sib)
1290 {
1291 int base = SIB_BASE_FIELD (details->raw_insn[details->modrm_offset + 1]);
1292 int idx = SIB_INDEX_FIELD (details->raw_insn[details->modrm_offset + 1]);
1293 used_regs_mask |= 1 << base;
1294 used_regs_mask |= 1 << idx;
1295 }
1296 else
1297 {
1298 used_regs_mask |= 1 << rm;
1299 }
1300 }
1301
1302 gdb_assert (used_regs_mask < 256);
1303 gdb_assert (used_regs_mask != 255);
1304
1305 /* Finally, find a free reg. */
1306 {
1307 int i;
1308
1309 for (i = 0; i < 8; ++i)
1310 {
1311 if (! (used_regs_mask & (1 << i)))
1312 return i;
1313 }
1314
1315 /* We shouldn't get here. */
1316 internal_error (__FILE__, __LINE__, _("unable to find free reg"));
1317 }
1318 }
1319
1320 /* Extract the details of INSN that we need. */
1321
1322 static void
1323 amd64_get_insn_details (gdb_byte *insn, struct amd64_insn *details)
1324 {
1325 gdb_byte *start = insn;
1326 int need_modrm;
1327
1328 details->raw_insn = insn;
1329
1330 details->opcode_len = -1;
1331 details->enc_prefix_offset = -1;
1332 details->opcode_offset = -1;
1333 details->modrm_offset = -1;
1334
1335 /* Skip legacy instruction prefixes. */
1336 insn = amd64_skip_prefixes (insn);
1337
1338 /* Skip REX/VEX instruction encoding prefixes. */
1339 if (rex_prefix_p (*insn))
1340 {
1341 details->enc_prefix_offset = insn - start;
1342 ++insn;
1343 }
1344 else if (vex2_prefix_p (*insn))
1345 {
1346 /* Don't record the offset in this case because this prefix has
1347 no REX.B equivalent. */
1348 insn += 2;
1349 }
1350 else if (vex3_prefix_p (*insn))
1351 {
1352 details->enc_prefix_offset = insn - start;
1353 insn += 3;
1354 }
1355
1356 details->opcode_offset = insn - start;
1357
1358 if (*insn == TWO_BYTE_OPCODE_ESCAPE)
1359 {
1360 /* Two or three-byte opcode. */
1361 ++insn;
1362 need_modrm = twobyte_has_modrm[*insn];
1363
1364 /* Check for three-byte opcode. */
1365 switch (*insn)
1366 {
1367 case 0x24:
1368 case 0x25:
1369 case 0x38:
1370 case 0x3a:
1371 case 0x7a:
1372 case 0x7b:
1373 ++insn;
1374 details->opcode_len = 3;
1375 break;
1376 default:
1377 details->opcode_len = 2;
1378 break;
1379 }
1380 }
1381 else
1382 {
1383 /* One-byte opcode. */
1384 need_modrm = onebyte_has_modrm[*insn];
1385 details->opcode_len = 1;
1386 }
1387
1388 if (need_modrm)
1389 {
1390 ++insn;
1391 details->modrm_offset = insn - start;
1392 }
1393 }
1394
1395 /* Update %rip-relative addressing in INSN.
1396
1397 %rip-relative addressing only uses a 32-bit displacement.
1398 32 bits is not enough to be guaranteed to cover the distance between where
1399 the real instruction is and where its copy is.
1400 Convert the insn to use base+disp addressing.
1401 We set base = pc + insn_length so we can leave disp unchanged. */
1402
1403 static void
1404 fixup_riprel (struct gdbarch *gdbarch,
1405 amd64_displaced_step_copy_insn_closure *dsc,
1406 CORE_ADDR from, CORE_ADDR to, struct regcache *regs)
1407 {
1408 const struct amd64_insn *insn_details = &dsc->insn_details;
1409 int modrm_offset = insn_details->modrm_offset;
1410 gdb_byte *insn = insn_details->raw_insn + modrm_offset;
1411 CORE_ADDR rip_base;
1412 int insn_length;
1413 int arch_tmp_regno, tmp_regno;
1414 ULONGEST orig_value;
1415
1416 /* %rip+disp32 addressing mode, displacement follows ModRM byte. */
1417 ++insn;
1418
1419 /* Compute the rip-relative address. */
1420 insn_length = gdb_buffered_insn_length (gdbarch, dsc->insn_buf.data (),
1421 dsc->insn_buf.size (), from);
1422 rip_base = from + insn_length;
1423
1424 /* We need a register to hold the address.
1425 Pick one not used in the insn.
1426 NOTE: arch_tmp_regno uses architecture ordering, e.g. RDI = 7. */
1427 arch_tmp_regno = amd64_get_unused_input_int_reg (insn_details);
1428 tmp_regno = amd64_arch_reg_to_regnum (arch_tmp_regno);
1429
1430 /* Position of the not-B bit in the 3-byte VEX prefix (in byte 1). */
1431 static constexpr gdb_byte VEX3_NOT_B = 0x20;
1432
1433 /* REX.B should be unset (VEX.!B set) as we were using rip-relative
1434 addressing, but ensure it's unset (set for VEX) anyway, tmp_regno
1435 is not r8-r15. */
1436 if (insn_details->enc_prefix_offset != -1)
1437 {
1438 gdb_byte *pfx = &dsc->insn_buf[insn_details->enc_prefix_offset];
1439 if (rex_prefix_p (pfx[0]))
1440 pfx[0] &= ~REX_B;
1441 else if (vex3_prefix_p (pfx[0]))
1442 pfx[1] |= VEX3_NOT_B;
1443 else
1444 gdb_assert_not_reached ("unhandled prefix");
1445 }
1446
1447 regcache_cooked_read_unsigned (regs, tmp_regno, &orig_value);
1448 dsc->tmp_regno = tmp_regno;
1449 dsc->tmp_save = orig_value;
1450 dsc->tmp_used = 1;
1451
1452 /* Convert the ModRM field to be base+disp. */
1453 dsc->insn_buf[modrm_offset] &= ~0xc7;
1454 dsc->insn_buf[modrm_offset] |= 0x80 + arch_tmp_regno;
1455
1456 regcache_cooked_write_unsigned (regs, tmp_regno, rip_base);
1457
1458 displaced_debug_printf ("%%rip-relative addressing used.");
1459 displaced_debug_printf ("using temp reg %d, old value %s, new value %s",
1460 dsc->tmp_regno, paddress (gdbarch, dsc->tmp_save),
1461 paddress (gdbarch, rip_base));
1462 }
1463
1464 static void
1465 fixup_displaced_copy (struct gdbarch *gdbarch,
1466 amd64_displaced_step_copy_insn_closure *dsc,
1467 CORE_ADDR from, CORE_ADDR to, struct regcache *regs)
1468 {
1469 const struct amd64_insn *details = &dsc->insn_details;
1470
1471 if (details->modrm_offset != -1)
1472 {
1473 gdb_byte modrm = details->raw_insn[details->modrm_offset];
1474
1475 if ((modrm & 0xc7) == 0x05)
1476 {
1477 /* The insn uses rip-relative addressing.
1478 Deal with it. */
1479 fixup_riprel (gdbarch, dsc, from, to, regs);
1480 }
1481 }
1482 }
1483
1484 displaced_step_copy_insn_closure_up
1485 amd64_displaced_step_copy_insn (struct gdbarch *gdbarch,
1486 CORE_ADDR from, CORE_ADDR to,
1487 struct regcache *regs)
1488 {
1489 int len = gdbarch_max_insn_length (gdbarch);
1490 /* Extra space for sentinels so fixup_{riprel,displaced_copy} don't have to
1491 continually watch for running off the end of the buffer. */
1492 int fixup_sentinel_space = len;
1493 std::unique_ptr<amd64_displaced_step_copy_insn_closure> dsc
1494 (new amd64_displaced_step_copy_insn_closure (len + fixup_sentinel_space));
1495 gdb_byte *buf = &dsc->insn_buf[0];
1496 struct amd64_insn *details = &dsc->insn_details;
1497
1498 read_memory (from, buf, len);
1499
1500 /* Set up the sentinel space so we don't have to worry about running
1501 off the end of the buffer. An excessive number of leading prefixes
1502 could otherwise cause this. */
1503 memset (buf + len, 0, fixup_sentinel_space);
1504
1505 amd64_get_insn_details (buf, details);
1506
1507 /* GDB may get control back after the insn after the syscall.
1508 Presumably this is a kernel bug.
1509 If this is a syscall, make sure there's a nop afterwards. */
1510 {
1511 int syscall_length;
1512
1513 if (amd64_syscall_p (details, &syscall_length))
1514 buf[details->opcode_offset + syscall_length] = NOP_OPCODE;
1515 }
1516
1517 /* Modify the insn to cope with the address where it will be executed from.
1518 In particular, handle any rip-relative addressing. */
1519 fixup_displaced_copy (gdbarch, dsc.get (), from, to, regs);
1520
1521 write_memory (to, buf, len);
1522
1523 displaced_debug_printf ("copy %s->%s: %s",
1524 paddress (gdbarch, from), paddress (gdbarch, to),
1525 displaced_step_dump_bytes (buf, len).c_str ());
1526
1527 /* This is a work around for a problem with g++ 4.8. */
1528 return displaced_step_copy_insn_closure_up (dsc.release ());
1529 }
1530
1531 static int
1532 amd64_absolute_jmp_p (const struct amd64_insn *details)
1533 {
1534 const gdb_byte *insn = &details->raw_insn[details->opcode_offset];
1535
1536 if (insn[0] == 0xff)
1537 {
1538 /* jump near, absolute indirect (/4) */
1539 if ((insn[1] & 0x38) == 0x20)
1540 return 1;
1541
1542 /* jump far, absolute indirect (/5) */
1543 if ((insn[1] & 0x38) == 0x28)
1544 return 1;
1545 }
1546
1547 return 0;
1548 }
1549
1550 /* Return non-zero if the instruction DETAILS is a jump, zero otherwise. */
1551
1552 static int
1553 amd64_jmp_p (const struct amd64_insn *details)
1554 {
1555 const gdb_byte *insn = &details->raw_insn[details->opcode_offset];
1556
1557 /* jump short, relative. */
1558 if (insn[0] == 0xeb)
1559 return 1;
1560
1561 /* jump near, relative. */
1562 if (insn[0] == 0xe9)
1563 return 1;
1564
1565 return amd64_absolute_jmp_p (details);
1566 }
1567
1568 static int
1569 amd64_absolute_call_p (const struct amd64_insn *details)
1570 {
1571 const gdb_byte *insn = &details->raw_insn[details->opcode_offset];
1572
1573 if (insn[0] == 0xff)
1574 {
1575 /* Call near, absolute indirect (/2) */
1576 if ((insn[1] & 0x38) == 0x10)
1577 return 1;
1578
1579 /* Call far, absolute indirect (/3) */
1580 if ((insn[1] & 0x38) == 0x18)
1581 return 1;
1582 }
1583
1584 return 0;
1585 }
1586
1587 static int
1588 amd64_ret_p (const struct amd64_insn *details)
1589 {
1590 /* NOTE: gcc can emit "repz ; ret". */
1591 const gdb_byte *insn = &details->raw_insn[details->opcode_offset];
1592
1593 switch (insn[0])
1594 {
1595 case 0xc2: /* ret near, pop N bytes */
1596 case 0xc3: /* ret near */
1597 case 0xca: /* ret far, pop N bytes */
1598 case 0xcb: /* ret far */
1599 case 0xcf: /* iret */
1600 return 1;
1601
1602 default:
1603 return 0;
1604 }
1605 }
1606
1607 static int
1608 amd64_call_p (const struct amd64_insn *details)
1609 {
1610 const gdb_byte *insn = &details->raw_insn[details->opcode_offset];
1611
1612 if (amd64_absolute_call_p (details))
1613 return 1;
1614
1615 /* call near, relative */
1616 if (insn[0] == 0xe8)
1617 return 1;
1618
1619 return 0;
1620 }
1621
1622 /* Return non-zero if INSN is a system call, and set *LENGTHP to its
1623 length in bytes. Otherwise, return zero. */
1624
1625 static int
1626 amd64_syscall_p (const struct amd64_insn *details, int *lengthp)
1627 {
1628 const gdb_byte *insn = &details->raw_insn[details->opcode_offset];
1629
1630 if (insn[0] == 0x0f && insn[1] == 0x05)
1631 {
1632 *lengthp = 2;
1633 return 1;
1634 }
1635
1636 return 0;
1637 }
1638
1639 /* Classify the instruction at ADDR using PRED.
1640 Throw an error if the memory can't be read. */
1641
1642 static int
1643 amd64_classify_insn_at (struct gdbarch *gdbarch, CORE_ADDR addr,
1644 int (*pred) (const struct amd64_insn *))
1645 {
1646 struct amd64_insn details;
1647 gdb_byte *buf;
1648 int len, classification;
1649
1650 len = gdbarch_max_insn_length (gdbarch);
1651 buf = (gdb_byte *) alloca (len);
1652
1653 read_code (addr, buf, len);
1654 amd64_get_insn_details (buf, &details);
1655
1656 classification = pred (&details);
1657
1658 return classification;
1659 }
1660
1661 /* The gdbarch insn_is_call method. */
1662
1663 static int
1664 amd64_insn_is_call (struct gdbarch *gdbarch, CORE_ADDR addr)
1665 {
1666 return amd64_classify_insn_at (gdbarch, addr, amd64_call_p);
1667 }
1668
1669 /* The gdbarch insn_is_ret method. */
1670
1671 static int
1672 amd64_insn_is_ret (struct gdbarch *gdbarch, CORE_ADDR addr)
1673 {
1674 return amd64_classify_insn_at (gdbarch, addr, amd64_ret_p);
1675 }
1676
1677 /* The gdbarch insn_is_jump method. */
1678
1679 static int
1680 amd64_insn_is_jump (struct gdbarch *gdbarch, CORE_ADDR addr)
1681 {
1682 return amd64_classify_insn_at (gdbarch, addr, amd64_jmp_p);
1683 }
1684
1685 /* Fix up the state of registers and memory after having single-stepped
1686 a displaced instruction. */
1687
1688 void
1689 amd64_displaced_step_fixup (struct gdbarch *gdbarch,
1690 struct displaced_step_copy_insn_closure *dsc_,
1691 CORE_ADDR from, CORE_ADDR to,
1692 struct regcache *regs)
1693 {
1694 amd64_displaced_step_copy_insn_closure *dsc
1695 = (amd64_displaced_step_copy_insn_closure *) dsc_;
1696 enum bfd_endian byte_order = gdbarch_byte_order (gdbarch);
1697 /* The offset we applied to the instruction's address. */
1698 ULONGEST insn_offset = to - from;
1699 gdb_byte *insn = dsc->insn_buf.data ();
1700 const struct amd64_insn *insn_details = &dsc->insn_details;
1701
1702 displaced_debug_printf ("fixup (%s, %s), insn = 0x%02x 0x%02x ...",
1703 paddress (gdbarch, from), paddress (gdbarch, to),
1704 insn[0], insn[1]);
1705
1706 /* If we used a tmp reg, restore it. */
1707
1708 if (dsc->tmp_used)
1709 {
1710 displaced_debug_printf ("restoring reg %d to %s",
1711 dsc->tmp_regno, paddress (gdbarch, dsc->tmp_save));
1712 regcache_cooked_write_unsigned (regs, dsc->tmp_regno, dsc->tmp_save);
1713 }
1714
1715 /* The list of issues to contend with here is taken from
1716 resume_execution in arch/x86/kernel/kprobes.c, Linux 2.6.28.
1717 Yay for Free Software! */
1718
1719 /* Relocate the %rip back to the program's instruction stream,
1720 if necessary. */
1721
1722 /* Except in the case of absolute or indirect jump or call
1723 instructions, or a return instruction, the new rip is relative to
1724 the displaced instruction; make it relative to the original insn.
1725 Well, signal handler returns don't need relocation either, but we use the
1726 value of %rip to recognize those; see below. */
1727 if (! amd64_absolute_jmp_p (insn_details)
1728 && ! amd64_absolute_call_p (insn_details)
1729 && ! amd64_ret_p (insn_details))
1730 {
1731 ULONGEST orig_rip;
1732 int insn_len;
1733
1734 regcache_cooked_read_unsigned (regs, AMD64_RIP_REGNUM, &orig_rip);
1735
1736 /* A signal trampoline system call changes the %rip, resuming
1737 execution of the main program after the signal handler has
1738 returned. That makes them like 'return' instructions; we
1739 shouldn't relocate %rip.
1740
1741 But most system calls don't, and we do need to relocate %rip.
1742
1743 Our heuristic for distinguishing these cases: if stepping
1744 over the system call instruction left control directly after
1745 the instruction, the we relocate --- control almost certainly
1746 doesn't belong in the displaced copy. Otherwise, we assume
1747 the instruction has put control where it belongs, and leave
1748 it unrelocated. Goodness help us if there are PC-relative
1749 system calls. */
1750 if (amd64_syscall_p (insn_details, &insn_len)
1751 && orig_rip != to + insn_len
1752 /* GDB can get control back after the insn after the syscall.
1753 Presumably this is a kernel bug.
1754 Fixup ensures its a nop, we add one to the length for it. */
1755 && orig_rip != to + insn_len + 1)
1756 displaced_debug_printf ("syscall changed %%rip; not relocating");
1757 else
1758 {
1759 ULONGEST rip = orig_rip - insn_offset;
1760
1761 /* If we just stepped over a breakpoint insn, we don't backup
1762 the pc on purpose; this is to match behaviour without
1763 stepping. */
1764
1765 regcache_cooked_write_unsigned (regs, AMD64_RIP_REGNUM, rip);
1766
1767 displaced_debug_printf ("relocated %%rip from %s to %s",
1768 paddress (gdbarch, orig_rip),
1769 paddress (gdbarch, rip));
1770 }
1771 }
1772
1773 /* If the instruction was PUSHFL, then the TF bit will be set in the
1774 pushed value, and should be cleared. We'll leave this for later,
1775 since GDB already messes up the TF flag when stepping over a
1776 pushfl. */
1777
1778 /* If the instruction was a call, the return address now atop the
1779 stack is the address following the copied instruction. We need
1780 to make it the address following the original instruction. */
1781 if (amd64_call_p (insn_details))
1782 {
1783 ULONGEST rsp;
1784 ULONGEST retaddr;
1785 const ULONGEST retaddr_len = 8;
1786
1787 regcache_cooked_read_unsigned (regs, AMD64_RSP_REGNUM, &rsp);
1788 retaddr = read_memory_unsigned_integer (rsp, retaddr_len, byte_order);
1789 retaddr = (retaddr - insn_offset) & 0xffffffffffffffffULL;
1790 write_memory_unsigned_integer (rsp, retaddr_len, byte_order, retaddr);
1791
1792 displaced_debug_printf ("relocated return addr at %s to %s",
1793 paddress (gdbarch, rsp),
1794 paddress (gdbarch, retaddr));
1795 }
1796 }
1797
1798 /* If the instruction INSN uses RIP-relative addressing, return the
1799 offset into the raw INSN where the displacement to be adjusted is
1800 found. Returns 0 if the instruction doesn't use RIP-relative
1801 addressing. */
1802
1803 static int
1804 rip_relative_offset (struct amd64_insn *insn)
1805 {
1806 if (insn->modrm_offset != -1)
1807 {
1808 gdb_byte modrm = insn->raw_insn[insn->modrm_offset];
1809
1810 if ((modrm & 0xc7) == 0x05)
1811 {
1812 /* The displacement is found right after the ModRM byte. */
1813 return insn->modrm_offset + 1;
1814 }
1815 }
1816
1817 return 0;
1818 }
1819
1820 static void
1821 append_insns (CORE_ADDR *to, ULONGEST len, const gdb_byte *buf)
1822 {
1823 target_write_memory (*to, buf, len);
1824 *to += len;
1825 }
1826
1827 static void
1828 amd64_relocate_instruction (struct gdbarch *gdbarch,
1829 CORE_ADDR *to, CORE_ADDR oldloc)
1830 {
1831 enum bfd_endian byte_order = gdbarch_byte_order (gdbarch);
1832 int len = gdbarch_max_insn_length (gdbarch);
1833 /* Extra space for sentinels. */
1834 int fixup_sentinel_space = len;
1835 gdb_byte *buf = (gdb_byte *) xmalloc (len + fixup_sentinel_space);
1836 struct amd64_insn insn_details;
1837 int offset = 0;
1838 LONGEST rel32, newrel;
1839 gdb_byte *insn;
1840 int insn_length;
1841
1842 read_memory (oldloc, buf, len);
1843
1844 /* Set up the sentinel space so we don't have to worry about running
1845 off the end of the buffer. An excessive number of leading prefixes
1846 could otherwise cause this. */
1847 memset (buf + len, 0, fixup_sentinel_space);
1848
1849 insn = buf;
1850 amd64_get_insn_details (insn, &insn_details);
1851
1852 insn_length = gdb_buffered_insn_length (gdbarch, insn, len, oldloc);
1853
1854 /* Skip legacy instruction prefixes. */
1855 insn = amd64_skip_prefixes (insn);
1856
1857 /* Adjust calls with 32-bit relative addresses as push/jump, with
1858 the address pushed being the location where the original call in
1859 the user program would return to. */
1860 if (insn[0] == 0xe8)
1861 {
1862 gdb_byte push_buf[32];
1863 CORE_ADDR ret_addr;
1864 int i = 0;
1865
1866 /* Where "ret" in the original code will return to. */
1867 ret_addr = oldloc + insn_length;
1868
1869 /* If pushing an address higher than or equal to 0x80000000,
1870 avoid 'pushq', as that sign extends its 32-bit operand, which
1871 would be incorrect. */
1872 if (ret_addr <= 0x7fffffff)
1873 {
1874 push_buf[0] = 0x68; /* pushq $... */
1875 store_unsigned_integer (&push_buf[1], 4, byte_order, ret_addr);
1876 i = 5;
1877 }
1878 else
1879 {
1880 push_buf[i++] = 0x48; /* sub $0x8,%rsp */
1881 push_buf[i++] = 0x83;
1882 push_buf[i++] = 0xec;
1883 push_buf[i++] = 0x08;
1884
1885 push_buf[i++] = 0xc7; /* movl $imm,(%rsp) */
1886 push_buf[i++] = 0x04;
1887 push_buf[i++] = 0x24;
1888 store_unsigned_integer (&push_buf[i], 4, byte_order,
1889 ret_addr & 0xffffffff);
1890 i += 4;
1891
1892 push_buf[i++] = 0xc7; /* movl $imm,4(%rsp) */
1893 push_buf[i++] = 0x44;
1894 push_buf[i++] = 0x24;
1895 push_buf[i++] = 0x04;
1896 store_unsigned_integer (&push_buf[i], 4, byte_order,
1897 ret_addr >> 32);
1898 i += 4;
1899 }
1900 gdb_assert (i <= sizeof (push_buf));
1901 /* Push the push. */
1902 append_insns (to, i, push_buf);
1903
1904 /* Convert the relative call to a relative jump. */
1905 insn[0] = 0xe9;
1906
1907 /* Adjust the destination offset. */
1908 rel32 = extract_signed_integer (insn + 1, 4, byte_order);
1909 newrel = (oldloc - *to) + rel32;
1910 store_signed_integer (insn + 1, 4, byte_order, newrel);
1911
1912 displaced_debug_printf ("adjusted insn rel32=%s at %s to rel32=%s at %s",
1913 hex_string (rel32), paddress (gdbarch, oldloc),
1914 hex_string (newrel), paddress (gdbarch, *to));
1915
1916 /* Write the adjusted jump into its displaced location. */
1917 append_insns (to, 5, insn);
1918 return;
1919 }
1920
1921 offset = rip_relative_offset (&insn_details);
1922 if (!offset)
1923 {
1924 /* Adjust jumps with 32-bit relative addresses. Calls are
1925 already handled above. */
1926 if (insn[0] == 0xe9)
1927 offset = 1;
1928 /* Adjust conditional jumps. */
1929 else if (insn[0] == 0x0f && (insn[1] & 0xf0) == 0x80)
1930 offset = 2;
1931 }
1932
1933 if (offset)
1934 {
1935 rel32 = extract_signed_integer (insn + offset, 4, byte_order);
1936 newrel = (oldloc - *to) + rel32;
1937 store_signed_integer (insn + offset, 4, byte_order, newrel);
1938 displaced_debug_printf ("adjusted insn rel32=%s at %s to rel32=%s at %s",
1939 hex_string (rel32), paddress (gdbarch, oldloc),
1940 hex_string (newrel), paddress (gdbarch, *to));
1941 }
1942
1943 /* Write the adjusted instruction into its displaced location. */
1944 append_insns (to, insn_length, buf);
1945 }
1946
1947 \f
1948 /* The maximum number of saved registers. This should include %rip. */
1949 #define AMD64_NUM_SAVED_REGS AMD64_NUM_GREGS
1950
1951 struct amd64_frame_cache
1952 {
1953 /* Base address. */
1954 CORE_ADDR base;
1955 int base_p;
1956 CORE_ADDR sp_offset;
1957 CORE_ADDR pc;
1958
1959 /* Saved registers. */
1960 CORE_ADDR saved_regs[AMD64_NUM_SAVED_REGS];
1961 CORE_ADDR saved_sp;
1962 int saved_sp_reg;
1963
1964 /* Do we have a frame? */
1965 int frameless_p;
1966 };
1967
1968 /* Initialize a frame cache. */
1969
1970 static void
1971 amd64_init_frame_cache (struct amd64_frame_cache *cache)
1972 {
1973 int i;
1974
1975 /* Base address. */
1976 cache->base = 0;
1977 cache->base_p = 0;
1978 cache->sp_offset = -8;
1979 cache->pc = 0;
1980
1981 /* Saved registers. We initialize these to -1 since zero is a valid
1982 offset (that's where %rbp is supposed to be stored).
1983 The values start out as being offsets, and are later converted to
1984 addresses (at which point -1 is interpreted as an address, still meaning
1985 "invalid"). */
1986 for (i = 0; i < AMD64_NUM_SAVED_REGS; i++)
1987 cache->saved_regs[i] = -1;
1988 cache->saved_sp = 0;
1989 cache->saved_sp_reg = -1;
1990
1991 /* Frameless until proven otherwise. */
1992 cache->frameless_p = 1;
1993 }
1994
1995 /* Allocate and initialize a frame cache. */
1996
1997 static struct amd64_frame_cache *
1998 amd64_alloc_frame_cache (void)
1999 {
2000 struct amd64_frame_cache *cache;
2001
2002 cache = FRAME_OBSTACK_ZALLOC (struct amd64_frame_cache);
2003 amd64_init_frame_cache (cache);
2004 return cache;
2005 }
2006
2007 /* GCC 4.4 and later, can put code in the prologue to realign the
2008 stack pointer. Check whether PC points to such code, and update
2009 CACHE accordingly. Return the first instruction after the code
2010 sequence or CURRENT_PC, whichever is smaller. If we don't
2011 recognize the code, return PC. */
2012
2013 static CORE_ADDR
2014 amd64_analyze_stack_align (CORE_ADDR pc, CORE_ADDR current_pc,
2015 struct amd64_frame_cache *cache)
2016 {
2017 /* There are 2 code sequences to re-align stack before the frame
2018 gets set up:
2019
2020 1. Use a caller-saved saved register:
2021
2022 leaq 8(%rsp), %reg
2023 andq $-XXX, %rsp
2024 pushq -8(%reg)
2025
2026 2. Use a callee-saved saved register:
2027
2028 pushq %reg
2029 leaq 16(%rsp), %reg
2030 andq $-XXX, %rsp
2031 pushq -8(%reg)
2032
2033 "andq $-XXX, %rsp" can be either 4 bytes or 7 bytes:
2034
2035 0x48 0x83 0xe4 0xf0 andq $-16, %rsp
2036 0x48 0x81 0xe4 0x00 0xff 0xff 0xff andq $-256, %rsp
2037 */
2038
2039 gdb_byte buf[18];
2040 int reg, r;
2041 int offset, offset_and;
2042
2043 if (target_read_code (pc, buf, sizeof buf))
2044 return pc;
2045
2046 /* Check caller-saved saved register. The first instruction has
2047 to be "leaq 8(%rsp), %reg". */
2048 if ((buf[0] & 0xfb) == 0x48
2049 && buf[1] == 0x8d
2050 && buf[3] == 0x24
2051 && buf[4] == 0x8)
2052 {
2053 /* MOD must be binary 10 and R/M must be binary 100. */
2054 if ((buf[2] & 0xc7) != 0x44)
2055 return pc;
2056
2057 /* REG has register number. */
2058 reg = (buf[2] >> 3) & 7;
2059
2060 /* Check the REX.R bit. */
2061 if (buf[0] == 0x4c)
2062 reg += 8;
2063
2064 offset = 5;
2065 }
2066 else
2067 {
2068 /* Check callee-saved saved register. The first instruction
2069 has to be "pushq %reg". */
2070 reg = 0;
2071 if ((buf[0] & 0xf8) == 0x50)
2072 offset = 0;
2073 else if ((buf[0] & 0xf6) == 0x40
2074 && (buf[1] & 0xf8) == 0x50)
2075 {
2076 /* Check the REX.B bit. */
2077 if ((buf[0] & 1) != 0)
2078 reg = 8;
2079
2080 offset = 1;
2081 }
2082 else
2083 return pc;
2084
2085 /* Get register. */
2086 reg += buf[offset] & 0x7;
2087
2088 offset++;
2089
2090 /* The next instruction has to be "leaq 16(%rsp), %reg". */
2091 if ((buf[offset] & 0xfb) != 0x48
2092 || buf[offset + 1] != 0x8d
2093 || buf[offset + 3] != 0x24
2094 || buf[offset + 4] != 0x10)
2095 return pc;
2096
2097 /* MOD must be binary 10 and R/M must be binary 100. */
2098 if ((buf[offset + 2] & 0xc7) != 0x44)
2099 return pc;
2100
2101 /* REG has register number. */
2102 r = (buf[offset + 2] >> 3) & 7;
2103
2104 /* Check the REX.R bit. */
2105 if (buf[offset] == 0x4c)
2106 r += 8;
2107
2108 /* Registers in pushq and leaq have to be the same. */
2109 if (reg != r)
2110 return pc;
2111
2112 offset += 5;
2113 }
2114
2115 /* Rigister can't be %rsp nor %rbp. */
2116 if (reg == 4 || reg == 5)
2117 return pc;
2118
2119 /* The next instruction has to be "andq $-XXX, %rsp". */
2120 if (buf[offset] != 0x48
2121 || buf[offset + 2] != 0xe4
2122 || (buf[offset + 1] != 0x81 && buf[offset + 1] != 0x83))
2123 return pc;
2124
2125 offset_and = offset;
2126 offset += buf[offset + 1] == 0x81 ? 7 : 4;
2127
2128 /* The next instruction has to be "pushq -8(%reg)". */
2129 r = 0;
2130 if (buf[offset] == 0xff)
2131 offset++;
2132 else if ((buf[offset] & 0xf6) == 0x40
2133 && buf[offset + 1] == 0xff)
2134 {
2135 /* Check the REX.B bit. */
2136 if ((buf[offset] & 0x1) != 0)
2137 r = 8;
2138 offset += 2;
2139 }
2140 else
2141 return pc;
2142
2143 /* 8bit -8 is 0xf8. REG must be binary 110 and MOD must be binary
2144 01. */
2145 if (buf[offset + 1] != 0xf8
2146 || (buf[offset] & 0xf8) != 0x70)
2147 return pc;
2148
2149 /* R/M has register. */
2150 r += buf[offset] & 7;
2151
2152 /* Registers in leaq and pushq have to be the same. */
2153 if (reg != r)
2154 return pc;
2155
2156 if (current_pc > pc + offset_and)
2157 cache->saved_sp_reg = amd64_arch_reg_to_regnum (reg);
2158
2159 return std::min (pc + offset + 2, current_pc);
2160 }
2161
2162 /* Similar to amd64_analyze_stack_align for x32. */
2163
2164 static CORE_ADDR
2165 amd64_x32_analyze_stack_align (CORE_ADDR pc, CORE_ADDR current_pc,
2166 struct amd64_frame_cache *cache)
2167 {
2168 /* There are 2 code sequences to re-align stack before the frame
2169 gets set up:
2170
2171 1. Use a caller-saved saved register:
2172
2173 leaq 8(%rsp), %reg
2174 andq $-XXX, %rsp
2175 pushq -8(%reg)
2176
2177 or
2178
2179 [addr32] leal 8(%rsp), %reg
2180 andl $-XXX, %esp
2181 [addr32] pushq -8(%reg)
2182
2183 2. Use a callee-saved saved register:
2184
2185 pushq %reg
2186 leaq 16(%rsp), %reg
2187 andq $-XXX, %rsp
2188 pushq -8(%reg)
2189
2190 or
2191
2192 pushq %reg
2193 [addr32] leal 16(%rsp), %reg
2194 andl $-XXX, %esp
2195 [addr32] pushq -8(%reg)
2196
2197 "andq $-XXX, %rsp" can be either 4 bytes or 7 bytes:
2198
2199 0x48 0x83 0xe4 0xf0 andq $-16, %rsp
2200 0x48 0x81 0xe4 0x00 0xff 0xff 0xff andq $-256, %rsp
2201
2202 "andl $-XXX, %esp" can be either 3 bytes or 6 bytes:
2203
2204 0x83 0xe4 0xf0 andl $-16, %esp
2205 0x81 0xe4 0x00 0xff 0xff 0xff andl $-256, %esp
2206 */
2207
2208 gdb_byte buf[19];
2209 int reg, r;
2210 int offset, offset_and;
2211
2212 if (target_read_memory (pc, buf, sizeof buf))
2213 return pc;
2214
2215 /* Skip optional addr32 prefix. */
2216 offset = buf[0] == 0x67 ? 1 : 0;
2217
2218 /* Check caller-saved saved register. The first instruction has
2219 to be "leaq 8(%rsp), %reg" or "leal 8(%rsp), %reg". */
2220 if (((buf[offset] & 0xfb) == 0x48 || (buf[offset] & 0xfb) == 0x40)
2221 && buf[offset + 1] == 0x8d
2222 && buf[offset + 3] == 0x24
2223 && buf[offset + 4] == 0x8)
2224 {
2225 /* MOD must be binary 10 and R/M must be binary 100. */
2226 if ((buf[offset + 2] & 0xc7) != 0x44)
2227 return pc;
2228
2229 /* REG has register number. */
2230 reg = (buf[offset + 2] >> 3) & 7;
2231
2232 /* Check the REX.R bit. */
2233 if ((buf[offset] & 0x4) != 0)
2234 reg += 8;
2235
2236 offset += 5;
2237 }
2238 else
2239 {
2240 /* Check callee-saved saved register. The first instruction
2241 has to be "pushq %reg". */
2242 reg = 0;
2243 if ((buf[offset] & 0xf6) == 0x40
2244 && (buf[offset + 1] & 0xf8) == 0x50)
2245 {
2246 /* Check the REX.B bit. */
2247 if ((buf[offset] & 1) != 0)
2248 reg = 8;
2249
2250 offset += 1;
2251 }
2252 else if ((buf[offset] & 0xf8) != 0x50)
2253 return pc;
2254
2255 /* Get register. */
2256 reg += buf[offset] & 0x7;
2257
2258 offset++;
2259
2260 /* Skip optional addr32 prefix. */
2261 if (buf[offset] == 0x67)
2262 offset++;
2263
2264 /* The next instruction has to be "leaq 16(%rsp), %reg" or
2265 "leal 16(%rsp), %reg". */
2266 if (((buf[offset] & 0xfb) != 0x48 && (buf[offset] & 0xfb) != 0x40)
2267 || buf[offset + 1] != 0x8d
2268 || buf[offset + 3] != 0x24
2269 || buf[offset + 4] != 0x10)
2270 return pc;
2271
2272 /* MOD must be binary 10 and R/M must be binary 100. */
2273 if ((buf[offset + 2] & 0xc7) != 0x44)
2274 return pc;
2275
2276 /* REG has register number. */
2277 r = (buf[offset + 2] >> 3) & 7;
2278
2279 /* Check the REX.R bit. */
2280 if ((buf[offset] & 0x4) != 0)
2281 r += 8;
2282
2283 /* Registers in pushq and leaq have to be the same. */
2284 if (reg != r)
2285 return pc;
2286
2287 offset += 5;
2288 }
2289
2290 /* Rigister can't be %rsp nor %rbp. */
2291 if (reg == 4 || reg == 5)
2292 return pc;
2293
2294 /* The next instruction may be "andq $-XXX, %rsp" or
2295 "andl $-XXX, %esp". */
2296 if (buf[offset] != 0x48)
2297 offset--;
2298
2299 if (buf[offset + 2] != 0xe4
2300 || (buf[offset + 1] != 0x81 && buf[offset + 1] != 0x83))
2301 return pc;
2302
2303 offset_and = offset;
2304 offset += buf[offset + 1] == 0x81 ? 7 : 4;
2305
2306 /* Skip optional addr32 prefix. */
2307 if (buf[offset] == 0x67)
2308 offset++;
2309
2310 /* The next instruction has to be "pushq -8(%reg)". */
2311 r = 0;
2312 if (buf[offset] == 0xff)
2313 offset++;
2314 else if ((buf[offset] & 0xf6) == 0x40
2315 && buf[offset + 1] == 0xff)
2316 {
2317 /* Check the REX.B bit. */
2318 if ((buf[offset] & 0x1) != 0)
2319 r = 8;
2320 offset += 2;
2321 }
2322 else
2323 return pc;
2324
2325 /* 8bit -8 is 0xf8. REG must be binary 110 and MOD must be binary
2326 01. */
2327 if (buf[offset + 1] != 0xf8
2328 || (buf[offset] & 0xf8) != 0x70)
2329 return pc;
2330
2331 /* R/M has register. */
2332 r += buf[offset] & 7;
2333
2334 /* Registers in leaq and pushq have to be the same. */
2335 if (reg != r)
2336 return pc;
2337
2338 if (current_pc > pc + offset_and)
2339 cache->saved_sp_reg = amd64_arch_reg_to_regnum (reg);
2340
2341 return std::min (pc + offset + 2, current_pc);
2342 }
2343
2344 /* Do a limited analysis of the prologue at PC and update CACHE
2345 accordingly. Bail out early if CURRENT_PC is reached. Return the
2346 address where the analysis stopped.
2347
2348 We will handle only functions beginning with:
2349
2350 pushq %rbp 0x55
2351 movq %rsp, %rbp 0x48 0x89 0xe5 (or 0x48 0x8b 0xec)
2352
2353 or (for the X32 ABI):
2354
2355 pushq %rbp 0x55
2356 movl %esp, %ebp 0x89 0xe5 (or 0x8b 0xec)
2357
2358 The `endbr64` instruction can be found before these sequences, and will be
2359 skipped if found.
2360
2361 Any function that doesn't start with one of these sequences will be
2362 assumed to have no prologue and thus no valid frame pointer in
2363 %rbp. */
2364
2365 static CORE_ADDR
2366 amd64_analyze_prologue (struct gdbarch *gdbarch,
2367 CORE_ADDR pc, CORE_ADDR current_pc,
2368 struct amd64_frame_cache *cache)
2369 {
2370 enum bfd_endian byte_order = gdbarch_byte_order (gdbarch);
2371 /* The `endbr64` instruction. */
2372 static const gdb_byte endbr64[4] = { 0xf3, 0x0f, 0x1e, 0xfa };
2373 /* There are two variations of movq %rsp, %rbp. */
2374 static const gdb_byte mov_rsp_rbp_1[3] = { 0x48, 0x89, 0xe5 };
2375 static const gdb_byte mov_rsp_rbp_2[3] = { 0x48, 0x8b, 0xec };
2376 /* Ditto for movl %esp, %ebp. */
2377 static const gdb_byte mov_esp_ebp_1[2] = { 0x89, 0xe5 };
2378 static const gdb_byte mov_esp_ebp_2[2] = { 0x8b, 0xec };
2379
2380 gdb_byte buf[3];
2381 gdb_byte op;
2382
2383 if (current_pc <= pc)
2384 return current_pc;
2385
2386 if (gdbarch_ptr_bit (gdbarch) == 32)
2387 pc = amd64_x32_analyze_stack_align (pc, current_pc, cache);
2388 else
2389 pc = amd64_analyze_stack_align (pc, current_pc, cache);
2390
2391 op = read_code_unsigned_integer (pc, 1, byte_order);
2392
2393 /* Check for the `endbr64` instruction, skip it if found. */
2394 if (op == endbr64[0])
2395 {
2396 read_code (pc + 1, buf, 3);
2397
2398 if (memcmp (buf, &endbr64[1], 3) == 0)
2399 pc += 4;
2400
2401 op = read_code_unsigned_integer (pc, 1, byte_order);
2402 }
2403
2404 if (current_pc <= pc)
2405 return current_pc;
2406
2407 if (op == 0x55) /* pushq %rbp */
2408 {
2409 /* Take into account that we've executed the `pushq %rbp' that
2410 starts this instruction sequence. */
2411 cache->saved_regs[AMD64_RBP_REGNUM] = 0;
2412 cache->sp_offset += 8;
2413
2414 /* If that's all, return now. */
2415 if (current_pc <= pc + 1)
2416 return current_pc;
2417
2418 read_code (pc + 1, buf, 3);
2419
2420 /* Check for `movq %rsp, %rbp'. */
2421 if (memcmp (buf, mov_rsp_rbp_1, 3) == 0
2422 || memcmp (buf, mov_rsp_rbp_2, 3) == 0)
2423 {
2424 /* OK, we actually have a frame. */
2425 cache->frameless_p = 0;
2426 return pc + 4;
2427 }
2428
2429 /* For X32, also check for `movl %esp, %ebp'. */
2430 if (gdbarch_ptr_bit (gdbarch) == 32)
2431 {
2432 if (memcmp (buf, mov_esp_ebp_1, 2) == 0
2433 || memcmp (buf, mov_esp_ebp_2, 2) == 0)
2434 {
2435 /* OK, we actually have a frame. */
2436 cache->frameless_p = 0;
2437 return pc + 3;
2438 }
2439 }
2440
2441 return pc + 1;
2442 }
2443
2444 return pc;
2445 }
2446
2447 /* Work around false termination of prologue - GCC PR debug/48827.
2448
2449 START_PC is the first instruction of a function, PC is its minimal already
2450 determined advanced address. Function returns PC if it has nothing to do.
2451
2452 84 c0 test %al,%al
2453 74 23 je after
2454 <-- here is 0 lines advance - the false prologue end marker.
2455 0f 29 85 70 ff ff ff movaps %xmm0,-0x90(%rbp)
2456 0f 29 4d 80 movaps %xmm1,-0x80(%rbp)
2457 0f 29 55 90 movaps %xmm2,-0x70(%rbp)
2458 0f 29 5d a0 movaps %xmm3,-0x60(%rbp)
2459 0f 29 65 b0 movaps %xmm4,-0x50(%rbp)
2460 0f 29 6d c0 movaps %xmm5,-0x40(%rbp)
2461 0f 29 75 d0 movaps %xmm6,-0x30(%rbp)
2462 0f 29 7d e0 movaps %xmm7,-0x20(%rbp)
2463 after: */
2464
2465 static CORE_ADDR
2466 amd64_skip_xmm_prologue (CORE_ADDR pc, CORE_ADDR start_pc)
2467 {
2468 struct symtab_and_line start_pc_sal, next_sal;
2469 gdb_byte buf[4 + 8 * 7];
2470 int offset, xmmreg;
2471
2472 if (pc == start_pc)
2473 return pc;
2474
2475 start_pc_sal = find_pc_sect_line (start_pc, NULL, 0);
2476 if (start_pc_sal.symtab == NULL
2477 || producer_is_gcc_ge_4 (start_pc_sal.symtab->compunit ()
2478 ->producer ()) < 6
2479 || start_pc_sal.pc != start_pc || pc >= start_pc_sal.end)
2480 return pc;
2481
2482 next_sal = find_pc_sect_line (start_pc_sal.end, NULL, 0);
2483 if (next_sal.line != start_pc_sal.line)
2484 return pc;
2485
2486 /* START_PC can be from overlayed memory, ignored here. */
2487 if (target_read_code (next_sal.pc - 4, buf, sizeof (buf)) != 0)
2488 return pc;
2489
2490 /* test %al,%al */
2491 if (buf[0] != 0x84 || buf[1] != 0xc0)
2492 return pc;
2493 /* je AFTER */
2494 if (buf[2] != 0x74)
2495 return pc;
2496
2497 offset = 4;
2498 for (xmmreg = 0; xmmreg < 8; xmmreg++)
2499 {
2500 /* 0x0f 0x29 0b??000101 movaps %xmmreg?,-0x??(%rbp) */
2501 if (buf[offset] != 0x0f || buf[offset + 1] != 0x29
2502 || (buf[offset + 2] & 0x3f) != (xmmreg << 3 | 0x5))
2503 return pc;
2504
2505 /* 0b01?????? */
2506 if ((buf[offset + 2] & 0xc0) == 0x40)
2507 {
2508 /* 8-bit displacement. */
2509 offset += 4;
2510 }
2511 /* 0b10?????? */
2512 else if ((buf[offset + 2] & 0xc0) == 0x80)
2513 {
2514 /* 32-bit displacement. */
2515 offset += 7;
2516 }
2517 else
2518 return pc;
2519 }
2520
2521 /* je AFTER */
2522 if (offset - 4 != buf[3])
2523 return pc;
2524
2525 return next_sal.end;
2526 }
2527
2528 /* Return PC of first real instruction. */
2529
2530 static CORE_ADDR
2531 amd64_skip_prologue (struct gdbarch *gdbarch, CORE_ADDR start_pc)
2532 {
2533 struct amd64_frame_cache cache;
2534 CORE_ADDR pc;
2535 CORE_ADDR func_addr;
2536
2537 if (find_pc_partial_function (start_pc, NULL, &func_addr, NULL))
2538 {
2539 CORE_ADDR post_prologue_pc
2540 = skip_prologue_using_sal (gdbarch, func_addr);
2541 struct compunit_symtab *cust = find_pc_compunit_symtab (func_addr);
2542
2543 /* LLVM backend (Clang/Flang) always emits a line note before the
2544 prologue and another one after. We trust clang and newer Intel
2545 compilers to emit usable line notes. */
2546 if (post_prologue_pc
2547 && (cust != NULL
2548 && cust->producer () != nullptr
2549 && (producer_is_llvm (cust->producer ())
2550 || producer_is_icc_ge_19 (cust->producer ()))))
2551 return std::max (start_pc, post_prologue_pc);
2552 }
2553
2554 amd64_init_frame_cache (&cache);
2555 pc = amd64_analyze_prologue (gdbarch, start_pc, 0xffffffffffffffffLL,
2556 &cache);
2557 if (cache.frameless_p)
2558 return start_pc;
2559
2560 return amd64_skip_xmm_prologue (pc, start_pc);
2561 }
2562 \f
2563
2564 /* Normal frames. */
2565
2566 static void
2567 amd64_frame_cache_1 (struct frame_info *this_frame,
2568 struct amd64_frame_cache *cache)
2569 {
2570 struct gdbarch *gdbarch = get_frame_arch (this_frame);
2571 enum bfd_endian byte_order = gdbarch_byte_order (gdbarch);
2572 gdb_byte buf[8];
2573 int i;
2574
2575 cache->pc = get_frame_func (this_frame);
2576 if (cache->pc != 0)
2577 amd64_analyze_prologue (gdbarch, cache->pc, get_frame_pc (this_frame),
2578 cache);
2579
2580 if (cache->frameless_p)
2581 {
2582 /* We didn't find a valid frame. If we're at the start of a
2583 function, or somewhere half-way its prologue, the function's
2584 frame probably hasn't been fully setup yet. Try to
2585 reconstruct the base address for the stack frame by looking
2586 at the stack pointer. For truly "frameless" functions this
2587 might work too. */
2588
2589 if (cache->saved_sp_reg != -1)
2590 {
2591 /* Stack pointer has been saved. */
2592 get_frame_register (this_frame, cache->saved_sp_reg, buf);
2593 cache->saved_sp = extract_unsigned_integer (buf, 8, byte_order);
2594
2595 /* We're halfway aligning the stack. */
2596 cache->base = ((cache->saved_sp - 8) & 0xfffffffffffffff0LL) - 8;
2597 cache->saved_regs[AMD64_RIP_REGNUM] = cache->saved_sp - 8;
2598
2599 /* This will be added back below. */
2600 cache->saved_regs[AMD64_RIP_REGNUM] -= cache->base;
2601 }
2602 else
2603 {
2604 get_frame_register (this_frame, AMD64_RSP_REGNUM, buf);
2605 cache->base = extract_unsigned_integer (buf, 8, byte_order)
2606 + cache->sp_offset;
2607 }
2608 }
2609 else
2610 {
2611 get_frame_register (this_frame, AMD64_RBP_REGNUM, buf);
2612 cache->base = extract_unsigned_integer (buf, 8, byte_order);
2613 }
2614
2615 /* Now that we have the base address for the stack frame we can
2616 calculate the value of %rsp in the calling frame. */
2617 cache->saved_sp = cache->base + 16;
2618
2619 /* For normal frames, %rip is stored at 8(%rbp). If we don't have a
2620 frame we find it at the same offset from the reconstructed base
2621 address. If we're halfway aligning the stack, %rip is handled
2622 differently (see above). */
2623 if (!cache->frameless_p || cache->saved_sp_reg == -1)
2624 cache->saved_regs[AMD64_RIP_REGNUM] = 8;
2625
2626 /* Adjust all the saved registers such that they contain addresses
2627 instead of offsets. */
2628 for (i = 0; i < AMD64_NUM_SAVED_REGS; i++)
2629 if (cache->saved_regs[i] != -1)
2630 cache->saved_regs[i] += cache->base;
2631
2632 cache->base_p = 1;
2633 }
2634
2635 static struct amd64_frame_cache *
2636 amd64_frame_cache (struct frame_info *this_frame, void **this_cache)
2637 {
2638 struct amd64_frame_cache *cache;
2639
2640 if (*this_cache)
2641 return (struct amd64_frame_cache *) *this_cache;
2642
2643 cache = amd64_alloc_frame_cache ();
2644 *this_cache = cache;
2645
2646 try
2647 {
2648 amd64_frame_cache_1 (this_frame, cache);
2649 }
2650 catch (const gdb_exception_error &ex)
2651 {
2652 if (ex.error != NOT_AVAILABLE_ERROR)
2653 throw;
2654 }
2655
2656 return cache;
2657 }
2658
2659 static enum unwind_stop_reason
2660 amd64_frame_unwind_stop_reason (struct frame_info *this_frame,
2661 void **this_cache)
2662 {
2663 struct amd64_frame_cache *cache =
2664 amd64_frame_cache (this_frame, this_cache);
2665
2666 if (!cache->base_p)
2667 return UNWIND_UNAVAILABLE;
2668
2669 /* This marks the outermost frame. */
2670 if (cache->base == 0)
2671 return UNWIND_OUTERMOST;
2672
2673 return UNWIND_NO_REASON;
2674 }
2675
2676 static void
2677 amd64_frame_this_id (struct frame_info *this_frame, void **this_cache,
2678 struct frame_id *this_id)
2679 {
2680 struct amd64_frame_cache *cache =
2681 amd64_frame_cache (this_frame, this_cache);
2682
2683 if (!cache->base_p)
2684 (*this_id) = frame_id_build_unavailable_stack (cache->pc);
2685 else if (cache->base == 0)
2686 {
2687 /* This marks the outermost frame. */
2688 return;
2689 }
2690 else
2691 (*this_id) = frame_id_build (cache->base + 16, cache->pc);
2692 }
2693
2694 static struct value *
2695 amd64_frame_prev_register (struct frame_info *this_frame, void **this_cache,
2696 int regnum)
2697 {
2698 struct gdbarch *gdbarch = get_frame_arch (this_frame);
2699 struct amd64_frame_cache *cache =
2700 amd64_frame_cache (this_frame, this_cache);
2701
2702 gdb_assert (regnum >= 0);
2703
2704 if (regnum == gdbarch_sp_regnum (gdbarch) && cache->saved_sp)
2705 return frame_unwind_got_constant (this_frame, regnum, cache->saved_sp);
2706
2707 if (regnum < AMD64_NUM_SAVED_REGS && cache->saved_regs[regnum] != -1)
2708 return frame_unwind_got_memory (this_frame, regnum,
2709 cache->saved_regs[regnum]);
2710
2711 return frame_unwind_got_register (this_frame, regnum, regnum);
2712 }
2713
2714 static const struct frame_unwind amd64_frame_unwind =
2715 {
2716 "amd64 prologue",
2717 NORMAL_FRAME,
2718 amd64_frame_unwind_stop_reason,
2719 amd64_frame_this_id,
2720 amd64_frame_prev_register,
2721 NULL,
2722 default_frame_sniffer
2723 };
2724 \f
2725 /* Generate a bytecode expression to get the value of the saved PC. */
2726
2727 static void
2728 amd64_gen_return_address (struct gdbarch *gdbarch,
2729 struct agent_expr *ax, struct axs_value *value,
2730 CORE_ADDR scope)
2731 {
2732 /* The following sequence assumes the traditional use of the base
2733 register. */
2734 ax_reg (ax, AMD64_RBP_REGNUM);
2735 ax_const_l (ax, 8);
2736 ax_simple (ax, aop_add);
2737 value->type = register_type (gdbarch, AMD64_RIP_REGNUM);
2738 value->kind = axs_lvalue_memory;
2739 }
2740 \f
2741
2742 /* Signal trampolines. */
2743
2744 /* FIXME: kettenis/20030419: Perhaps, we can unify the 32-bit and
2745 64-bit variants. This would require using identical frame caches
2746 on both platforms. */
2747
2748 static struct amd64_frame_cache *
2749 amd64_sigtramp_frame_cache (struct frame_info *this_frame, void **this_cache)
2750 {
2751 struct gdbarch *gdbarch = get_frame_arch (this_frame);
2752 i386_gdbarch_tdep *tdep = (i386_gdbarch_tdep *) gdbarch_tdep (gdbarch);
2753 enum bfd_endian byte_order = gdbarch_byte_order (gdbarch);
2754 struct amd64_frame_cache *cache;
2755 CORE_ADDR addr;
2756 gdb_byte buf[8];
2757 int i;
2758
2759 if (*this_cache)
2760 return (struct amd64_frame_cache *) *this_cache;
2761
2762 cache = amd64_alloc_frame_cache ();
2763
2764 try
2765 {
2766 get_frame_register (this_frame, AMD64_RSP_REGNUM, buf);
2767 cache->base = extract_unsigned_integer (buf, 8, byte_order) - 8;
2768
2769 addr = tdep->sigcontext_addr (this_frame);
2770 gdb_assert (tdep->sc_reg_offset);
2771 gdb_assert (tdep->sc_num_regs <= AMD64_NUM_SAVED_REGS);
2772 for (i = 0; i < tdep->sc_num_regs; i++)
2773 if (tdep->sc_reg_offset[i] != -1)
2774 cache->saved_regs[i] = addr + tdep->sc_reg_offset[i];
2775
2776 cache->base_p = 1;
2777 }
2778 catch (const gdb_exception_error &ex)
2779 {
2780 if (ex.error != NOT_AVAILABLE_ERROR)
2781 throw;
2782 }
2783
2784 *this_cache = cache;
2785 return cache;
2786 }
2787
2788 static enum unwind_stop_reason
2789 amd64_sigtramp_frame_unwind_stop_reason (struct frame_info *this_frame,
2790 void **this_cache)
2791 {
2792 struct amd64_frame_cache *cache =
2793 amd64_sigtramp_frame_cache (this_frame, this_cache);
2794
2795 if (!cache->base_p)
2796 return UNWIND_UNAVAILABLE;
2797
2798 return UNWIND_NO_REASON;
2799 }
2800
2801 static void
2802 amd64_sigtramp_frame_this_id (struct frame_info *this_frame,
2803 void **this_cache, struct frame_id *this_id)
2804 {
2805 struct amd64_frame_cache *cache =
2806 amd64_sigtramp_frame_cache (this_frame, this_cache);
2807
2808 if (!cache->base_p)
2809 (*this_id) = frame_id_build_unavailable_stack (get_frame_pc (this_frame));
2810 else if (cache->base == 0)
2811 {
2812 /* This marks the outermost frame. */
2813 return;
2814 }
2815 else
2816 (*this_id) = frame_id_build (cache->base + 16, get_frame_pc (this_frame));
2817 }
2818
2819 static struct value *
2820 amd64_sigtramp_frame_prev_register (struct frame_info *this_frame,
2821 void **this_cache, int regnum)
2822 {
2823 /* Make sure we've initialized the cache. */
2824 amd64_sigtramp_frame_cache (this_frame, this_cache);
2825
2826 return amd64_frame_prev_register (this_frame, this_cache, regnum);
2827 }
2828
2829 static int
2830 amd64_sigtramp_frame_sniffer (const struct frame_unwind *self,
2831 struct frame_info *this_frame,
2832 void **this_cache)
2833 {
2834 gdbarch *arch = get_frame_arch (this_frame);
2835 i386_gdbarch_tdep *tdep = (i386_gdbarch_tdep *) gdbarch_tdep (arch);
2836
2837 /* We shouldn't even bother if we don't have a sigcontext_addr
2838 handler. */
2839 if (tdep->sigcontext_addr == NULL)
2840 return 0;
2841
2842 if (tdep->sigtramp_p != NULL)
2843 {
2844 if (tdep->sigtramp_p (this_frame))
2845 return 1;
2846 }
2847
2848 if (tdep->sigtramp_start != 0)
2849 {
2850 CORE_ADDR pc = get_frame_pc (this_frame);
2851
2852 gdb_assert (tdep->sigtramp_end != 0);
2853 if (pc >= tdep->sigtramp_start && pc < tdep->sigtramp_end)
2854 return 1;
2855 }
2856
2857 return 0;
2858 }
2859
2860 static const struct frame_unwind amd64_sigtramp_frame_unwind =
2861 {
2862 "amd64 sigtramp",
2863 SIGTRAMP_FRAME,
2864 amd64_sigtramp_frame_unwind_stop_reason,
2865 amd64_sigtramp_frame_this_id,
2866 amd64_sigtramp_frame_prev_register,
2867 NULL,
2868 amd64_sigtramp_frame_sniffer
2869 };
2870 \f
2871
2872 static CORE_ADDR
2873 amd64_frame_base_address (struct frame_info *this_frame, void **this_cache)
2874 {
2875 struct amd64_frame_cache *cache =
2876 amd64_frame_cache (this_frame, this_cache);
2877
2878 return cache->base;
2879 }
2880
2881 static const struct frame_base amd64_frame_base =
2882 {
2883 &amd64_frame_unwind,
2884 amd64_frame_base_address,
2885 amd64_frame_base_address,
2886 amd64_frame_base_address
2887 };
2888
2889 /* Normal frames, but in a function epilogue. */
2890
2891 /* Implement the stack_frame_destroyed_p gdbarch method.
2892
2893 The epilogue is defined here as the 'ret' instruction, which will
2894 follow any instruction such as 'leave' or 'pop %ebp' that destroys
2895 the function's stack frame. */
2896
2897 static int
2898 amd64_stack_frame_destroyed_p (struct gdbarch *gdbarch, CORE_ADDR pc)
2899 {
2900 gdb_byte insn;
2901 struct compunit_symtab *cust;
2902
2903 cust = find_pc_compunit_symtab (pc);
2904 if (cust != NULL && cust->epilogue_unwind_valid ())
2905 return 0;
2906
2907 if (target_read_memory (pc, &insn, 1))
2908 return 0; /* Can't read memory at pc. */
2909
2910 if (insn != 0xc3) /* 'ret' instruction. */
2911 return 0;
2912
2913 return 1;
2914 }
2915
2916 static int
2917 amd64_epilogue_frame_sniffer (const struct frame_unwind *self,
2918 struct frame_info *this_frame,
2919 void **this_prologue_cache)
2920 {
2921 if (frame_relative_level (this_frame) == 0)
2922 return amd64_stack_frame_destroyed_p (get_frame_arch (this_frame),
2923 get_frame_pc (this_frame));
2924 else
2925 return 0;
2926 }
2927
2928 static struct amd64_frame_cache *
2929 amd64_epilogue_frame_cache (struct frame_info *this_frame, void **this_cache)
2930 {
2931 struct gdbarch *gdbarch = get_frame_arch (this_frame);
2932 enum bfd_endian byte_order = gdbarch_byte_order (gdbarch);
2933 struct amd64_frame_cache *cache;
2934 gdb_byte buf[8];
2935
2936 if (*this_cache)
2937 return (struct amd64_frame_cache *) *this_cache;
2938
2939 cache = amd64_alloc_frame_cache ();
2940 *this_cache = cache;
2941
2942 try
2943 {
2944 /* Cache base will be %esp plus cache->sp_offset (-8). */
2945 get_frame_register (this_frame, AMD64_RSP_REGNUM, buf);
2946 cache->base = extract_unsigned_integer (buf, 8,
2947 byte_order) + cache->sp_offset;
2948
2949 /* Cache pc will be the frame func. */
2950 cache->pc = get_frame_pc (this_frame);
2951
2952 /* The saved %esp will be at cache->base plus 16. */
2953 cache->saved_sp = cache->base + 16;
2954
2955 /* The saved %eip will be at cache->base plus 8. */
2956 cache->saved_regs[AMD64_RIP_REGNUM] = cache->base + 8;
2957
2958 cache->base_p = 1;
2959 }
2960 catch (const gdb_exception_error &ex)
2961 {
2962 if (ex.error != NOT_AVAILABLE_ERROR)
2963 throw;
2964 }
2965
2966 return cache;
2967 }
2968
2969 static enum unwind_stop_reason
2970 amd64_epilogue_frame_unwind_stop_reason (struct frame_info *this_frame,
2971 void **this_cache)
2972 {
2973 struct amd64_frame_cache *cache
2974 = amd64_epilogue_frame_cache (this_frame, this_cache);
2975
2976 if (!cache->base_p)
2977 return UNWIND_UNAVAILABLE;
2978
2979 return UNWIND_NO_REASON;
2980 }
2981
2982 static void
2983 amd64_epilogue_frame_this_id (struct frame_info *this_frame,
2984 void **this_cache,
2985 struct frame_id *this_id)
2986 {
2987 struct amd64_frame_cache *cache = amd64_epilogue_frame_cache (this_frame,
2988 this_cache);
2989
2990 if (!cache->base_p)
2991 (*this_id) = frame_id_build_unavailable_stack (cache->pc);
2992 else
2993 (*this_id) = frame_id_build (cache->base + 8, cache->pc);
2994 }
2995
2996 static const struct frame_unwind amd64_epilogue_frame_unwind =
2997 {
2998 "amd64 epilogue",
2999 NORMAL_FRAME,
3000 amd64_epilogue_frame_unwind_stop_reason,
3001 amd64_epilogue_frame_this_id,
3002 amd64_frame_prev_register,
3003 NULL,
3004 amd64_epilogue_frame_sniffer
3005 };
3006
3007 static struct frame_id
3008 amd64_dummy_id (struct gdbarch *gdbarch, struct frame_info *this_frame)
3009 {
3010 CORE_ADDR fp;
3011
3012 fp = get_frame_register_unsigned (this_frame, AMD64_RBP_REGNUM);
3013
3014 return frame_id_build (fp + 16, get_frame_pc (this_frame));
3015 }
3016
3017 /* 16 byte align the SP per frame requirements. */
3018
3019 static CORE_ADDR
3020 amd64_frame_align (struct gdbarch *gdbarch, CORE_ADDR sp)
3021 {
3022 return sp & -(CORE_ADDR)16;
3023 }
3024 \f
3025
3026 /* Supply register REGNUM from the buffer specified by FPREGS and LEN
3027 in the floating-point register set REGSET to register cache
3028 REGCACHE. If REGNUM is -1, do this for all registers in REGSET. */
3029
3030 static void
3031 amd64_supply_fpregset (const struct regset *regset, struct regcache *regcache,
3032 int regnum, const void *fpregs, size_t len)
3033 {
3034 struct gdbarch *gdbarch = regcache->arch ();
3035 const i386_gdbarch_tdep *tdep = (i386_gdbarch_tdep *) gdbarch_tdep (gdbarch);
3036
3037 gdb_assert (len >= tdep->sizeof_fpregset);
3038 amd64_supply_fxsave (regcache, regnum, fpregs);
3039 }
3040
3041 /* Collect register REGNUM from the register cache REGCACHE and store
3042 it in the buffer specified by FPREGS and LEN as described by the
3043 floating-point register set REGSET. If REGNUM is -1, do this for
3044 all registers in REGSET. */
3045
3046 static void
3047 amd64_collect_fpregset (const struct regset *regset,
3048 const struct regcache *regcache,
3049 int regnum, void *fpregs, size_t len)
3050 {
3051 struct gdbarch *gdbarch = regcache->arch ();
3052 const i386_gdbarch_tdep *tdep = (i386_gdbarch_tdep *) gdbarch_tdep (gdbarch);
3053
3054 gdb_assert (len >= tdep->sizeof_fpregset);
3055 amd64_collect_fxsave (regcache, regnum, fpregs);
3056 }
3057
3058 const struct regset amd64_fpregset =
3059 {
3060 NULL, amd64_supply_fpregset, amd64_collect_fpregset
3061 };
3062 \f
3063
3064 /* Figure out where the longjmp will land. Slurp the jmp_buf out of
3065 %rdi. We expect its value to be a pointer to the jmp_buf structure
3066 from which we extract the address that we will land at. This
3067 address is copied into PC. This routine returns non-zero on
3068 success. */
3069
3070 static int
3071 amd64_get_longjmp_target (struct frame_info *frame, CORE_ADDR *pc)
3072 {
3073 gdb_byte buf[8];
3074 CORE_ADDR jb_addr;
3075 struct gdbarch *gdbarch = get_frame_arch (frame);
3076 i386_gdbarch_tdep *tdep = (i386_gdbarch_tdep *) gdbarch_tdep (gdbarch);
3077 int jb_pc_offset = tdep->jb_pc_offset;
3078 int len = TYPE_LENGTH (builtin_type (gdbarch)->builtin_func_ptr);
3079
3080 /* If JB_PC_OFFSET is -1, we have no way to find out where the
3081 longjmp will land. */
3082 if (jb_pc_offset == -1)
3083 return 0;
3084
3085 get_frame_register (frame, AMD64_RDI_REGNUM, buf);
3086 jb_addr= extract_typed_address
3087 (buf, builtin_type (gdbarch)->builtin_data_ptr);
3088 if (target_read_memory (jb_addr + jb_pc_offset, buf, len))
3089 return 0;
3090
3091 *pc = extract_typed_address (buf, builtin_type (gdbarch)->builtin_func_ptr);
3092
3093 return 1;
3094 }
3095
3096 static const int amd64_record_regmap[] =
3097 {
3098 AMD64_RAX_REGNUM, AMD64_RCX_REGNUM, AMD64_RDX_REGNUM, AMD64_RBX_REGNUM,
3099 AMD64_RSP_REGNUM, AMD64_RBP_REGNUM, AMD64_RSI_REGNUM, AMD64_RDI_REGNUM,
3100 AMD64_R8_REGNUM, AMD64_R9_REGNUM, AMD64_R10_REGNUM, AMD64_R11_REGNUM,
3101 AMD64_R12_REGNUM, AMD64_R13_REGNUM, AMD64_R14_REGNUM, AMD64_R15_REGNUM,
3102 AMD64_RIP_REGNUM, AMD64_EFLAGS_REGNUM, AMD64_CS_REGNUM, AMD64_SS_REGNUM,
3103 AMD64_DS_REGNUM, AMD64_ES_REGNUM, AMD64_FS_REGNUM, AMD64_GS_REGNUM
3104 };
3105
3106 /* Implement the "in_indirect_branch_thunk" gdbarch function. */
3107
3108 static bool
3109 amd64_in_indirect_branch_thunk (struct gdbarch *gdbarch, CORE_ADDR pc)
3110 {
3111 return x86_in_indirect_branch_thunk (pc, amd64_register_names,
3112 AMD64_RAX_REGNUM,
3113 AMD64_RIP_REGNUM);
3114 }
3115
3116 void
3117 amd64_init_abi (struct gdbarch_info info, struct gdbarch *gdbarch,
3118 const target_desc *default_tdesc)
3119 {
3120 i386_gdbarch_tdep *tdep = (i386_gdbarch_tdep *) gdbarch_tdep (gdbarch);
3121 const struct target_desc *tdesc = info.target_desc;
3122 static const char *const stap_integer_prefixes[] = { "$", NULL };
3123 static const char *const stap_register_prefixes[] = { "%", NULL };
3124 static const char *const stap_register_indirection_prefixes[] = { "(",
3125 NULL };
3126 static const char *const stap_register_indirection_suffixes[] = { ")",
3127 NULL };
3128
3129 /* AMD64 generally uses `fxsave' instead of `fsave' for saving its
3130 floating-point registers. */
3131 tdep->sizeof_fpregset = I387_SIZEOF_FXSAVE;
3132 tdep->fpregset = &amd64_fpregset;
3133
3134 if (! tdesc_has_registers (tdesc))
3135 tdesc = default_tdesc;
3136 tdep->tdesc = tdesc;
3137
3138 tdep->num_core_regs = AMD64_NUM_GREGS + I387_NUM_REGS;
3139 tdep->register_names = amd64_register_names;
3140
3141 if (tdesc_find_feature (tdesc, "org.gnu.gdb.i386.avx512") != NULL)
3142 {
3143 tdep->zmmh_register_names = amd64_zmmh_names;
3144 tdep->k_register_names = amd64_k_names;
3145 tdep->xmm_avx512_register_names = amd64_xmm_avx512_names;
3146 tdep->ymm16h_register_names = amd64_ymmh_avx512_names;
3147
3148 tdep->num_zmm_regs = 32;
3149 tdep->num_xmm_avx512_regs = 16;
3150 tdep->num_ymm_avx512_regs = 16;
3151
3152 tdep->zmm0h_regnum = AMD64_ZMM0H_REGNUM;
3153 tdep->k0_regnum = AMD64_K0_REGNUM;
3154 tdep->xmm16_regnum = AMD64_XMM16_REGNUM;
3155 tdep->ymm16h_regnum = AMD64_YMM16H_REGNUM;
3156 }
3157
3158 if (tdesc_find_feature (tdesc, "org.gnu.gdb.i386.avx") != NULL)
3159 {
3160 tdep->ymmh_register_names = amd64_ymmh_names;
3161 tdep->num_ymm_regs = 16;
3162 tdep->ymm0h_regnum = AMD64_YMM0H_REGNUM;
3163 }
3164
3165 if (tdesc_find_feature (tdesc, "org.gnu.gdb.i386.mpx") != NULL)
3166 {
3167 tdep->mpx_register_names = amd64_mpx_names;
3168 tdep->bndcfgu_regnum = AMD64_BNDCFGU_REGNUM;
3169 tdep->bnd0r_regnum = AMD64_BND0R_REGNUM;
3170 }
3171
3172 if (tdesc_find_feature (tdesc, "org.gnu.gdb.i386.segments") != NULL)
3173 {
3174 tdep->fsbase_regnum = AMD64_FSBASE_REGNUM;
3175 }
3176
3177 if (tdesc_find_feature (tdesc, "org.gnu.gdb.i386.pkeys") != NULL)
3178 {
3179 tdep->pkeys_register_names = amd64_pkeys_names;
3180 tdep->pkru_regnum = AMD64_PKRU_REGNUM;
3181 tdep->num_pkeys_regs = 1;
3182 }
3183
3184 tdep->num_byte_regs = 20;
3185 tdep->num_word_regs = 16;
3186 tdep->num_dword_regs = 16;
3187 /* Avoid wiring in the MMX registers for now. */
3188 tdep->num_mmx_regs = 0;
3189
3190 set_gdbarch_pseudo_register_read_value (gdbarch,
3191 amd64_pseudo_register_read_value);
3192 set_gdbarch_pseudo_register_write (gdbarch,
3193 amd64_pseudo_register_write);
3194 set_gdbarch_ax_pseudo_register_collect (gdbarch,
3195 amd64_ax_pseudo_register_collect);
3196
3197 set_tdesc_pseudo_register_name (gdbarch, amd64_pseudo_register_name);
3198
3199 /* AMD64 has an FPU and 16 SSE registers. */
3200 tdep->st0_regnum = AMD64_ST0_REGNUM;
3201 tdep->num_xmm_regs = 16;
3202
3203 /* This is what all the fuss is about. */
3204 set_gdbarch_long_bit (gdbarch, 64);
3205 set_gdbarch_long_long_bit (gdbarch, 64);
3206 set_gdbarch_ptr_bit (gdbarch, 64);
3207
3208 /* In contrast to the i386, on AMD64 a `long double' actually takes
3209 up 128 bits, even though it's still based on the i387 extended
3210 floating-point format which has only 80 significant bits. */
3211 set_gdbarch_long_double_bit (gdbarch, 128);
3212
3213 set_gdbarch_num_regs (gdbarch, AMD64_NUM_REGS);
3214
3215 /* Register numbers of various important registers. */
3216 set_gdbarch_sp_regnum (gdbarch, AMD64_RSP_REGNUM); /* %rsp */
3217 set_gdbarch_pc_regnum (gdbarch, AMD64_RIP_REGNUM); /* %rip */
3218 set_gdbarch_ps_regnum (gdbarch, AMD64_EFLAGS_REGNUM); /* %eflags */
3219 set_gdbarch_fp0_regnum (gdbarch, AMD64_ST0_REGNUM); /* %st(0) */
3220
3221 /* The "default" register numbering scheme for AMD64 is referred to
3222 as the "DWARF Register Number Mapping" in the System V psABI.
3223 The preferred debugging format for all known AMD64 targets is
3224 actually DWARF2, and GCC doesn't seem to support DWARF (that is
3225 DWARF-1), but we provide the same mapping just in case. This
3226 mapping is also used for stabs, which GCC does support. */
3227 set_gdbarch_stab_reg_to_regnum (gdbarch, amd64_dwarf_reg_to_regnum);
3228 set_gdbarch_dwarf2_reg_to_regnum (gdbarch, amd64_dwarf_reg_to_regnum);
3229
3230 /* We don't override SDB_REG_RO_REGNUM, since COFF doesn't seem to
3231 be in use on any of the supported AMD64 targets. */
3232
3233 /* Call dummy code. */
3234 set_gdbarch_push_dummy_call (gdbarch, amd64_push_dummy_call);
3235 set_gdbarch_frame_align (gdbarch, amd64_frame_align);
3236 set_gdbarch_frame_red_zone_size (gdbarch, 128);
3237
3238 set_gdbarch_convert_register_p (gdbarch, i387_convert_register_p);
3239 set_gdbarch_register_to_value (gdbarch, i387_register_to_value);
3240 set_gdbarch_value_to_register (gdbarch, i387_value_to_register);
3241
3242 set_gdbarch_return_value (gdbarch, amd64_return_value);
3243
3244 set_gdbarch_skip_prologue (gdbarch, amd64_skip_prologue);
3245
3246 tdep->record_regmap = amd64_record_regmap;
3247
3248 set_gdbarch_dummy_id (gdbarch, amd64_dummy_id);
3249
3250 /* Hook the function epilogue frame unwinder. This unwinder is
3251 appended to the list first, so that it supercedes the other
3252 unwinders in function epilogues. */
3253 frame_unwind_prepend_unwinder (gdbarch, &amd64_epilogue_frame_unwind);
3254
3255 /* Hook the prologue-based frame unwinders. */
3256 frame_unwind_append_unwinder (gdbarch, &amd64_sigtramp_frame_unwind);
3257 frame_unwind_append_unwinder (gdbarch, &amd64_frame_unwind);
3258 frame_base_set_default (gdbarch, &amd64_frame_base);
3259
3260 set_gdbarch_get_longjmp_target (gdbarch, amd64_get_longjmp_target);
3261
3262 set_gdbarch_relocate_instruction (gdbarch, amd64_relocate_instruction);
3263
3264 set_gdbarch_gen_return_address (gdbarch, amd64_gen_return_address);
3265
3266 /* SystemTap variables and functions. */
3267 set_gdbarch_stap_integer_prefixes (gdbarch, stap_integer_prefixes);
3268 set_gdbarch_stap_register_prefixes (gdbarch, stap_register_prefixes);
3269 set_gdbarch_stap_register_indirection_prefixes (gdbarch,
3270 stap_register_indirection_prefixes);
3271 set_gdbarch_stap_register_indirection_suffixes (gdbarch,
3272 stap_register_indirection_suffixes);
3273 set_gdbarch_stap_is_single_operand (gdbarch,
3274 i386_stap_is_single_operand);
3275 set_gdbarch_stap_parse_special_token (gdbarch,
3276 i386_stap_parse_special_token);
3277 set_gdbarch_insn_is_call (gdbarch, amd64_insn_is_call);
3278 set_gdbarch_insn_is_ret (gdbarch, amd64_insn_is_ret);
3279 set_gdbarch_insn_is_jump (gdbarch, amd64_insn_is_jump);
3280
3281 set_gdbarch_in_indirect_branch_thunk (gdbarch,
3282 amd64_in_indirect_branch_thunk);
3283
3284 register_amd64_ravenscar_ops (gdbarch);
3285 }
3286
3287 /* Initialize ARCH for x86-64, no osabi. */
3288
3289 static void
3290 amd64_none_init_abi (gdbarch_info info, gdbarch *arch)
3291 {
3292 amd64_init_abi (info, arch, amd64_target_description (X86_XSTATE_SSE_MASK,
3293 true));
3294 }
3295
3296 static struct type *
3297 amd64_x32_pseudo_register_type (struct gdbarch *gdbarch, int regnum)
3298 {
3299 i386_gdbarch_tdep *tdep = (i386_gdbarch_tdep *) gdbarch_tdep (gdbarch);
3300
3301 switch (regnum - tdep->eax_regnum)
3302 {
3303 case AMD64_RBP_REGNUM: /* %ebp */
3304 case AMD64_RSP_REGNUM: /* %esp */
3305 return builtin_type (gdbarch)->builtin_data_ptr;
3306 case AMD64_RIP_REGNUM: /* %eip */
3307 return builtin_type (gdbarch)->builtin_func_ptr;
3308 }
3309
3310 return i386_pseudo_register_type (gdbarch, regnum);
3311 }
3312
3313 void
3314 amd64_x32_init_abi (struct gdbarch_info info, struct gdbarch *gdbarch,
3315 const target_desc *default_tdesc)
3316 {
3317 i386_gdbarch_tdep *tdep = (i386_gdbarch_tdep *) gdbarch_tdep (gdbarch);
3318
3319 amd64_init_abi (info, gdbarch, default_tdesc);
3320
3321 tdep->num_dword_regs = 17;
3322 set_tdesc_pseudo_register_type (gdbarch, amd64_x32_pseudo_register_type);
3323
3324 set_gdbarch_long_bit (gdbarch, 32);
3325 set_gdbarch_ptr_bit (gdbarch, 32);
3326 }
3327
3328 /* Initialize ARCH for x64-32, no osabi. */
3329
3330 static void
3331 amd64_x32_none_init_abi (gdbarch_info info, gdbarch *arch)
3332 {
3333 amd64_x32_init_abi (info, arch,
3334 amd64_target_description (X86_XSTATE_SSE_MASK, true));
3335 }
3336
3337 /* Return the target description for a specified XSAVE feature mask. */
3338
3339 const struct target_desc *
3340 amd64_target_description (uint64_t xcr0, bool segments)
3341 {
3342 static target_desc *amd64_tdescs \
3343 [2/*AVX*/][2/*MPX*/][2/*AVX512*/][2/*PKRU*/][2/*segments*/] = {};
3344 target_desc **tdesc;
3345
3346 tdesc = &amd64_tdescs[(xcr0 & X86_XSTATE_AVX) ? 1 : 0]
3347 [(xcr0 & X86_XSTATE_MPX) ? 1 : 0]
3348 [(xcr0 & X86_XSTATE_AVX512) ? 1 : 0]
3349 [(xcr0 & X86_XSTATE_PKRU) ? 1 : 0]
3350 [segments ? 1 : 0];
3351
3352 if (*tdesc == NULL)
3353 *tdesc = amd64_create_target_description (xcr0, false, false,
3354 segments);
3355
3356 return *tdesc;
3357 }
3358
3359 void _initialize_amd64_tdep ();
3360 void
3361 _initialize_amd64_tdep ()
3362 {
3363 gdbarch_register_osabi (bfd_arch_i386, bfd_mach_x86_64, GDB_OSABI_NONE,
3364 amd64_none_init_abi);
3365 gdbarch_register_osabi (bfd_arch_i386, bfd_mach_x64_32, GDB_OSABI_NONE,
3366 amd64_x32_none_init_abi);
3367 }
3368 \f
3369
3370 /* The 64-bit FXSAVE format differs from the 32-bit format in the
3371 sense that the instruction pointer and data pointer are simply
3372 64-bit offsets into the code segment and the data segment instead
3373 of a selector offset pair. The functions below store the upper 32
3374 bits of these pointers (instead of just the 16-bits of the segment
3375 selector). */
3376
3377 /* Fill register REGNUM in REGCACHE with the appropriate
3378 floating-point or SSE register value from *FXSAVE. If REGNUM is
3379 -1, do this for all registers. This function masks off any of the
3380 reserved bits in *FXSAVE. */
3381
3382 void
3383 amd64_supply_fxsave (struct regcache *regcache, int regnum,
3384 const void *fxsave)
3385 {
3386 struct gdbarch *gdbarch = regcache->arch ();
3387 i386_gdbarch_tdep *tdep = (i386_gdbarch_tdep *) gdbarch_tdep (gdbarch);
3388
3389 i387_supply_fxsave (regcache, regnum, fxsave);
3390
3391 if (fxsave
3392 && gdbarch_bfd_arch_info (gdbarch)->bits_per_word == 64)
3393 {
3394 const gdb_byte *regs = (const gdb_byte *) fxsave;
3395
3396 if (regnum == -1 || regnum == I387_FISEG_REGNUM (tdep))
3397 regcache->raw_supply (I387_FISEG_REGNUM (tdep), regs + 12);
3398 if (regnum == -1 || regnum == I387_FOSEG_REGNUM (tdep))
3399 regcache->raw_supply (I387_FOSEG_REGNUM (tdep), regs + 20);
3400 }
3401 }
3402
3403 /* Similar to amd64_supply_fxsave, but use XSAVE extended state. */
3404
3405 void
3406 amd64_supply_xsave (struct regcache *regcache, int regnum,
3407 const void *xsave)
3408 {
3409 struct gdbarch *gdbarch = regcache->arch ();
3410 i386_gdbarch_tdep *tdep = (i386_gdbarch_tdep *) gdbarch_tdep (gdbarch);
3411
3412 i387_supply_xsave (regcache, regnum, xsave);
3413
3414 if (xsave
3415 && gdbarch_bfd_arch_info (gdbarch)->bits_per_word == 64)
3416 {
3417 const gdb_byte *regs = (const gdb_byte *) xsave;
3418 ULONGEST clear_bv;
3419
3420 clear_bv = i387_xsave_get_clear_bv (gdbarch, xsave);
3421
3422 /* If the FISEG and FOSEG registers have not been initialised yet
3423 (their CLEAR_BV bit is set) then their default values of zero will
3424 have already been setup by I387_SUPPLY_XSAVE. */
3425 if (!(clear_bv & X86_XSTATE_X87))
3426 {
3427 if (regnum == -1 || regnum == I387_FISEG_REGNUM (tdep))
3428 regcache->raw_supply (I387_FISEG_REGNUM (tdep), regs + 12);
3429 if (regnum == -1 || regnum == I387_FOSEG_REGNUM (tdep))
3430 regcache->raw_supply (I387_FOSEG_REGNUM (tdep), regs + 20);
3431 }
3432 }
3433 }
3434
3435 /* Fill register REGNUM (if it is a floating-point or SSE register) in
3436 *FXSAVE with the value from REGCACHE. If REGNUM is -1, do this for
3437 all registers. This function doesn't touch any of the reserved
3438 bits in *FXSAVE. */
3439
3440 void
3441 amd64_collect_fxsave (const struct regcache *regcache, int regnum,
3442 void *fxsave)
3443 {
3444 struct gdbarch *gdbarch = regcache->arch ();
3445 i386_gdbarch_tdep *tdep = (i386_gdbarch_tdep *) gdbarch_tdep (gdbarch);
3446 gdb_byte *regs = (gdb_byte *) fxsave;
3447
3448 i387_collect_fxsave (regcache, regnum, fxsave);
3449
3450 if (gdbarch_bfd_arch_info (gdbarch)->bits_per_word == 64)
3451 {
3452 if (regnum == -1 || regnum == I387_FISEG_REGNUM (tdep))
3453 regcache->raw_collect (I387_FISEG_REGNUM (tdep), regs + 12);
3454 if (regnum == -1 || regnum == I387_FOSEG_REGNUM (tdep))
3455 regcache->raw_collect (I387_FOSEG_REGNUM (tdep), regs + 20);
3456 }
3457 }
3458
3459 /* Similar to amd64_collect_fxsave, but use XSAVE extended state. */
3460
3461 void
3462 amd64_collect_xsave (const struct regcache *regcache, int regnum,
3463 void *xsave, int gcore)
3464 {
3465 struct gdbarch *gdbarch = regcache->arch ();
3466 i386_gdbarch_tdep *tdep = (i386_gdbarch_tdep *) gdbarch_tdep (gdbarch);
3467 gdb_byte *regs = (gdb_byte *) xsave;
3468
3469 i387_collect_xsave (regcache, regnum, xsave, gcore);
3470
3471 if (gdbarch_bfd_arch_info (gdbarch)->bits_per_word == 64)
3472 {
3473 if (regnum == -1 || regnum == I387_FISEG_REGNUM (tdep))
3474 regcache->raw_collect (I387_FISEG_REGNUM (tdep),
3475 regs + 12);
3476 if (regnum == -1 || regnum == I387_FOSEG_REGNUM (tdep))
3477 regcache->raw_collect (I387_FOSEG_REGNUM (tdep),
3478 regs + 20);
3479 }
3480 }