xref: /dragonfly/contrib/gdb-7/gdb/amd64-tdep.c (revision 8accc937)
1 /* Target-dependent code for AMD64.
2 
3    Copyright (C) 2001-2012 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 "gdbcmd.h"
33 #include "gdbcore.h"
34 #include "objfiles.h"
35 #include "regcache.h"
36 #include "regset.h"
37 #include "symfile.h"
38 #include "disasm.h"
39 #include "gdb_assert.h"
40 #include "exceptions.h"
41 #include "amd64-tdep.h"
42 #include "i387-tdep.h"
43 
44 #include "features/i386/amd64.c"
45 #include "features/i386/amd64-avx.c"
46 
47 #include "ax.h"
48 #include "ax-gdb.h"
49 
50 /* Note that the AMD64 architecture was previously known as x86-64.
51    The latter is (forever) engraved into the canonical system name as
52    returned by config.guess, and used as the name for the AMD64 port
53    of GNU/Linux.  The BSD's have renamed their ports to amd64; they
54    don't like to shout.  For GDB we prefer the amd64_-prefix over the
55    x86_64_-prefix since it's so much easier to type.  */
56 
57 /* Register information.  */
58 
59 static const char *amd64_register_names[] =
60 {
61   "rax", "rbx", "rcx", "rdx", "rsi", "rdi", "rbp", "rsp",
62 
63   /* %r8 is indeed register number 8.  */
64   "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15",
65   "rip", "eflags", "cs", "ss", "ds", "es", "fs", "gs",
66 
67   /* %st0 is register number 24.  */
68   "st0", "st1", "st2", "st3", "st4", "st5", "st6", "st7",
69   "fctrl", "fstat", "ftag", "fiseg", "fioff", "foseg", "fooff", "fop",
70 
71   /* %xmm0 is register number 40.  */
72   "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7",
73   "xmm8", "xmm9", "xmm10", "xmm11", "xmm12", "xmm13", "xmm14", "xmm15",
74   "mxcsr",
75 };
76 
77 static const char *amd64_ymm_names[] =
78 {
79   "ymm0", "ymm1", "ymm2", "ymm3",
80   "ymm4", "ymm5", "ymm6", "ymm7",
81   "ymm8", "ymm9", "ymm10", "ymm11",
82   "ymm12", "ymm13", "ymm14", "ymm15"
83 };
84 
85 static const char *amd64_ymmh_names[] =
86 {
87   "ymm0h", "ymm1h", "ymm2h", "ymm3h",
88   "ymm4h", "ymm5h", "ymm6h", "ymm7h",
89   "ymm8h", "ymm9h", "ymm10h", "ymm11h",
90   "ymm12h", "ymm13h", "ymm14h", "ymm15h"
91 };
92 
93 /* The registers used to pass integer arguments during a function call.  */
94 static int amd64_dummy_call_integer_regs[] =
95 {
96   AMD64_RDI_REGNUM,		/* %rdi */
97   AMD64_RSI_REGNUM,		/* %rsi */
98   AMD64_RDX_REGNUM,		/* %rdx */
99   AMD64_RCX_REGNUM,		/* %rcx */
100   8,				/* %r8 */
101   9				/* %r9 */
102 };
103 
104 /* DWARF Register Number Mapping as defined in the System V psABI,
105    section 3.6.  */
106 
107 static int amd64_dwarf_regmap[] =
108 {
109   /* General Purpose Registers RAX, RDX, RCX, RBX, RSI, RDI.  */
110   AMD64_RAX_REGNUM, AMD64_RDX_REGNUM,
111   AMD64_RCX_REGNUM, AMD64_RBX_REGNUM,
112   AMD64_RSI_REGNUM, AMD64_RDI_REGNUM,
113 
114   /* Frame Pointer Register RBP.  */
115   AMD64_RBP_REGNUM,
116 
117   /* Stack Pointer Register RSP.  */
118   AMD64_RSP_REGNUM,
119 
120   /* Extended Integer Registers 8 - 15.  */
121   8, 9, 10, 11, 12, 13, 14, 15,
122 
123   /* Return Address RA.  Mapped to RIP.  */
124   AMD64_RIP_REGNUM,
125 
126   /* SSE Registers 0 - 7.  */
127   AMD64_XMM0_REGNUM + 0, AMD64_XMM1_REGNUM,
128   AMD64_XMM0_REGNUM + 2, AMD64_XMM0_REGNUM + 3,
129   AMD64_XMM0_REGNUM + 4, AMD64_XMM0_REGNUM + 5,
130   AMD64_XMM0_REGNUM + 6, AMD64_XMM0_REGNUM + 7,
131 
132   /* Extended SSE Registers 8 - 15.  */
133   AMD64_XMM0_REGNUM + 8, AMD64_XMM0_REGNUM + 9,
134   AMD64_XMM0_REGNUM + 10, AMD64_XMM0_REGNUM + 11,
135   AMD64_XMM0_REGNUM + 12, AMD64_XMM0_REGNUM + 13,
136   AMD64_XMM0_REGNUM + 14, AMD64_XMM0_REGNUM + 15,
137 
138   /* Floating Point Registers 0-7.  */
139   AMD64_ST0_REGNUM + 0, AMD64_ST0_REGNUM + 1,
140   AMD64_ST0_REGNUM + 2, AMD64_ST0_REGNUM + 3,
141   AMD64_ST0_REGNUM + 4, AMD64_ST0_REGNUM + 5,
142   AMD64_ST0_REGNUM + 6, AMD64_ST0_REGNUM + 7,
143 
144   /* Control and Status Flags Register.  */
145   AMD64_EFLAGS_REGNUM,
146 
147   /* Selector Registers.  */
148   AMD64_ES_REGNUM,
149   AMD64_CS_REGNUM,
150   AMD64_SS_REGNUM,
151   AMD64_DS_REGNUM,
152   AMD64_FS_REGNUM,
153   AMD64_GS_REGNUM,
154   -1,
155   -1,
156 
157   /* Segment Base Address Registers.  */
158   -1,
159   -1,
160   -1,
161   -1,
162 
163   /* Special Selector Registers.  */
164   -1,
165   -1,
166 
167   /* Floating Point Control Registers.  */
168   AMD64_MXCSR_REGNUM,
169   AMD64_FCTRL_REGNUM,
170   AMD64_FSTAT_REGNUM
171 };
172 
173 static const int amd64_dwarf_regmap_len =
174   (sizeof (amd64_dwarf_regmap) / sizeof (amd64_dwarf_regmap[0]));
175 
176 /* Convert DWARF register number REG to the appropriate register
177    number used by GDB.  */
178 
179 static int
180 amd64_dwarf_reg_to_regnum (struct gdbarch *gdbarch, int reg)
181 {
182   struct gdbarch_tdep *tdep = gdbarch_tdep (gdbarch);
183   int ymm0_regnum = tdep->ymm0_regnum;
184   int regnum = -1;
185 
186   if (reg >= 0 && reg < amd64_dwarf_regmap_len)
187     regnum = amd64_dwarf_regmap[reg];
188 
189   if (regnum == -1)
190     warning (_("Unmapped DWARF Register #%d encountered."), reg);
191   else if (ymm0_regnum >= 0
192 	   && i386_xmm_regnum_p (gdbarch, regnum))
193     regnum += ymm0_regnum - I387_XMM0_REGNUM (tdep);
194 
195   return regnum;
196 }
197 
198 /* Map architectural register numbers to gdb register numbers.  */
199 
200 static const int amd64_arch_regmap[16] =
201 {
202   AMD64_RAX_REGNUM,	/* %rax */
203   AMD64_RCX_REGNUM,	/* %rcx */
204   AMD64_RDX_REGNUM,	/* %rdx */
205   AMD64_RBX_REGNUM,	/* %rbx */
206   AMD64_RSP_REGNUM,	/* %rsp */
207   AMD64_RBP_REGNUM,	/* %rbp */
208   AMD64_RSI_REGNUM,	/* %rsi */
209   AMD64_RDI_REGNUM,	/* %rdi */
210   AMD64_R8_REGNUM,	/* %r8 */
211   AMD64_R9_REGNUM,	/* %r9 */
212   AMD64_R10_REGNUM,	/* %r10 */
213   AMD64_R11_REGNUM,	/* %r11 */
214   AMD64_R12_REGNUM,	/* %r12 */
215   AMD64_R13_REGNUM,	/* %r13 */
216   AMD64_R14_REGNUM,	/* %r14 */
217   AMD64_R15_REGNUM	/* %r15 */
218 };
219 
220 static const int amd64_arch_regmap_len =
221   (sizeof (amd64_arch_regmap) / sizeof (amd64_arch_regmap[0]));
222 
223 /* Convert architectural register number REG to the appropriate register
224    number used by GDB.  */
225 
226 static int
227 amd64_arch_reg_to_regnum (int reg)
228 {
229   gdb_assert (reg >= 0 && reg < amd64_arch_regmap_len);
230 
231   return amd64_arch_regmap[reg];
232 }
233 
234 /* Register names for byte pseudo-registers.  */
235 
236 static const char *amd64_byte_names[] =
237 {
238   "al", "bl", "cl", "dl", "sil", "dil", "bpl", "spl",
239   "r8l", "r9l", "r10l", "r11l", "r12l", "r13l", "r14l", "r15l",
240   "ah", "bh", "ch", "dh"
241 };
242 
243 /* Number of lower byte registers.  */
244 #define AMD64_NUM_LOWER_BYTE_REGS 16
245 
246 /* Register names for word pseudo-registers.  */
247 
248 static const char *amd64_word_names[] =
249 {
250   "ax", "bx", "cx", "dx", "si", "di", "bp", "",
251   "r8w", "r9w", "r10w", "r11w", "r12w", "r13w", "r14w", "r15w"
252 };
253 
254 /* Register names for dword pseudo-registers.  */
255 
256 static const char *amd64_dword_names[] =
257 {
258   "eax", "ebx", "ecx", "edx", "esi", "edi", "ebp", "esp",
259   "r8d", "r9d", "r10d", "r11d", "r12d", "r13d", "r14d", "r15d"
260 };
261 
262 /* Return the name of register REGNUM.  */
263 
264 static const char *
265 amd64_pseudo_register_name (struct gdbarch *gdbarch, int regnum)
266 {
267   struct gdbarch_tdep *tdep = gdbarch_tdep (gdbarch);
268   if (i386_byte_regnum_p (gdbarch, regnum))
269     return amd64_byte_names[regnum - tdep->al_regnum];
270   else if (i386_ymm_regnum_p (gdbarch, regnum))
271     return amd64_ymm_names[regnum - tdep->ymm0_regnum];
272   else if (i386_word_regnum_p (gdbarch, regnum))
273     return amd64_word_names[regnum - tdep->ax_regnum];
274   else if (i386_dword_regnum_p (gdbarch, regnum))
275     return amd64_dword_names[regnum - tdep->eax_regnum];
276   else
277     return i386_pseudo_register_name (gdbarch, regnum);
278 }
279 
280 static struct value *
281 amd64_pseudo_register_read_value (struct gdbarch *gdbarch,
282 				  struct regcache *regcache,
283 				  int regnum)
284 {
285   gdb_byte raw_buf[MAX_REGISTER_SIZE];
286   struct gdbarch_tdep *tdep = gdbarch_tdep (gdbarch);
287   enum register_status status;
288   struct value *result_value;
289   gdb_byte *buf;
290 
291   result_value = allocate_value (register_type (gdbarch, regnum));
292   VALUE_LVAL (result_value) = lval_register;
293   VALUE_REGNUM (result_value) = regnum;
294   buf = value_contents_raw (result_value);
295 
296   if (i386_byte_regnum_p (gdbarch, regnum))
297     {
298       int gpnum = regnum - tdep->al_regnum;
299 
300       /* Extract (always little endian).  */
301       if (gpnum >= AMD64_NUM_LOWER_BYTE_REGS)
302 	{
303 	  /* Special handling for AH, BH, CH, DH.  */
304 	  status = regcache_raw_read (regcache,
305 				      gpnum - AMD64_NUM_LOWER_BYTE_REGS,
306 				      raw_buf);
307 	  if (status == REG_VALID)
308 	    memcpy (buf, raw_buf + 1, 1);
309 	  else
310 	    mark_value_bytes_unavailable (result_value, 0,
311 					  TYPE_LENGTH (value_type (result_value)));
312 	}
313       else
314 	{
315 	  status = regcache_raw_read (regcache, gpnum, raw_buf);
316 	  if (status == REG_VALID)
317 	    memcpy (buf, raw_buf, 1);
318 	  else
319 	    mark_value_bytes_unavailable (result_value, 0,
320 					  TYPE_LENGTH (value_type (result_value)));
321 	}
322     }
323   else if (i386_dword_regnum_p (gdbarch, regnum))
324     {
325       int gpnum = regnum - tdep->eax_regnum;
326       /* Extract (always little endian).  */
327       status = regcache_raw_read (regcache, gpnum, raw_buf);
328       if (status == REG_VALID)
329 	memcpy (buf, raw_buf, 4);
330       else
331 	mark_value_bytes_unavailable (result_value, 0,
332 				      TYPE_LENGTH (value_type (result_value)));
333     }
334   else
335     i386_pseudo_register_read_into_value (gdbarch, regcache, regnum,
336 					  result_value);
337 
338   return result_value;
339 }
340 
341 static void
342 amd64_pseudo_register_write (struct gdbarch *gdbarch,
343 			     struct regcache *regcache,
344 			     int regnum, const gdb_byte *buf)
345 {
346   gdb_byte raw_buf[MAX_REGISTER_SIZE];
347   struct gdbarch_tdep *tdep = gdbarch_tdep (gdbarch);
348 
349   if (i386_byte_regnum_p (gdbarch, regnum))
350     {
351       int gpnum = regnum - tdep->al_regnum;
352 
353       if (gpnum >= AMD64_NUM_LOWER_BYTE_REGS)
354 	{
355 	  /* Read ... AH, BH, CH, DH.  */
356 	  regcache_raw_read (regcache,
357 			     gpnum - AMD64_NUM_LOWER_BYTE_REGS, raw_buf);
358 	  /* ... Modify ... (always little endian).  */
359 	  memcpy (raw_buf + 1, buf, 1);
360 	  /* ... Write.  */
361 	  regcache_raw_write (regcache,
362 			      gpnum - AMD64_NUM_LOWER_BYTE_REGS, raw_buf);
363 	}
364       else
365 	{
366 	  /* Read ...  */
367 	  regcache_raw_read (regcache, gpnum, raw_buf);
368 	  /* ... Modify ... (always little endian).  */
369 	  memcpy (raw_buf, buf, 1);
370 	  /* ... Write.  */
371 	  regcache_raw_write (regcache, gpnum, raw_buf);
372 	}
373     }
374   else if (i386_dword_regnum_p (gdbarch, regnum))
375     {
376       int gpnum = regnum - tdep->eax_regnum;
377 
378       /* Read ...  */
379       regcache_raw_read (regcache, gpnum, raw_buf);
380       /* ... Modify ... (always little endian).  */
381       memcpy (raw_buf, buf, 4);
382       /* ... Write.  */
383       regcache_raw_write (regcache, gpnum, raw_buf);
384     }
385   else
386     i386_pseudo_register_write (gdbarch, regcache, regnum, buf);
387 }
388 
389 
390 
391 /* Return the union class of CLASS1 and CLASS2.  See the psABI for
392    details.  */
393 
394 static enum amd64_reg_class
395 amd64_merge_classes (enum amd64_reg_class class1, enum amd64_reg_class class2)
396 {
397   /* Rule (a): If both classes are equal, this is the resulting class.  */
398   if (class1 == class2)
399     return class1;
400 
401   /* Rule (b): If one of the classes is NO_CLASS, the resulting class
402      is the other class.  */
403   if (class1 == AMD64_NO_CLASS)
404     return class2;
405   if (class2 == AMD64_NO_CLASS)
406     return class1;
407 
408   /* Rule (c): If one of the classes is MEMORY, the result is MEMORY.  */
409   if (class1 == AMD64_MEMORY || class2 == AMD64_MEMORY)
410     return AMD64_MEMORY;
411 
412   /* Rule (d): If one of the classes is INTEGER, the result is INTEGER.  */
413   if (class1 == AMD64_INTEGER || class2 == AMD64_INTEGER)
414     return AMD64_INTEGER;
415 
416   /* Rule (e): If one of the classes is X87, X87UP, COMPLEX_X87 class,
417      MEMORY is used as class.  */
418   if (class1 == AMD64_X87 || class1 == AMD64_X87UP
419       || class1 == AMD64_COMPLEX_X87 || class2 == AMD64_X87
420       || class2 == AMD64_X87UP || class2 == AMD64_COMPLEX_X87)
421     return AMD64_MEMORY;
422 
423   /* Rule (f): Otherwise class SSE is used.  */
424   return AMD64_SSE;
425 }
426 
427 /* Return non-zero if TYPE is a non-POD structure or union type.  */
428 
429 static int
430 amd64_non_pod_p (struct type *type)
431 {
432   /* ??? A class with a base class certainly isn't POD, but does this
433      catch all non-POD structure types?  */
434   if (TYPE_CODE (type) == TYPE_CODE_STRUCT && TYPE_N_BASECLASSES (type) > 0)
435     return 1;
436 
437   return 0;
438 }
439 
440 /* Classify TYPE according to the rules for aggregate (structures and
441    arrays) and union types, and store the result in CLASS.  */
442 
443 static void
444 amd64_classify_aggregate (struct type *type, enum amd64_reg_class class[2])
445 {
446   int len = TYPE_LENGTH (type);
447 
448   /* 1. If the size of an object is larger than two eightbytes, or in
449         C++, is a non-POD structure or union type, or contains
450         unaligned fields, it has class memory.  */
451   if (len > 16 || amd64_non_pod_p (type))
452     {
453       class[0] = class[1] = AMD64_MEMORY;
454       return;
455     }
456 
457   /* 2. Both eightbytes get initialized to class NO_CLASS.  */
458   class[0] = class[1] = AMD64_NO_CLASS;
459 
460   /* 3. Each field of an object is classified recursively so that
461         always two fields are considered. The resulting class is
462         calculated according to the classes of the fields in the
463         eightbyte: */
464 
465   if (TYPE_CODE (type) == TYPE_CODE_ARRAY)
466     {
467       struct type *subtype = check_typedef (TYPE_TARGET_TYPE (type));
468 
469       /* All fields in an array have the same type.  */
470       amd64_classify (subtype, class);
471       if (len > 8 && class[1] == AMD64_NO_CLASS)
472 	class[1] = class[0];
473     }
474   else
475     {
476       int i;
477 
478       /* Structure or union.  */
479       gdb_assert (TYPE_CODE (type) == TYPE_CODE_STRUCT
480 		  || TYPE_CODE (type) == TYPE_CODE_UNION);
481 
482       for (i = 0; i < TYPE_NFIELDS (type); i++)
483 	{
484 	  struct type *subtype = check_typedef (TYPE_FIELD_TYPE (type, i));
485 	  int pos = TYPE_FIELD_BITPOS (type, i) / 64;
486 	  enum amd64_reg_class subclass[2];
487 	  int bitsize = TYPE_FIELD_BITSIZE (type, i);
488 	  int endpos;
489 
490 	  if (bitsize == 0)
491 	    bitsize = TYPE_LENGTH (subtype) * 8;
492 	  endpos = (TYPE_FIELD_BITPOS (type, i) + bitsize - 1) / 64;
493 
494 	  /* Ignore static fields.  */
495 	  if (field_is_static (&TYPE_FIELD (type, i)))
496 	    continue;
497 
498 	  gdb_assert (pos == 0 || pos == 1);
499 
500 	  amd64_classify (subtype, subclass);
501 	  class[pos] = amd64_merge_classes (class[pos], subclass[0]);
502 	  if (bitsize <= 64 && pos == 0 && endpos == 1)
503 	    /* This is a bit of an odd case:  We have a field that would
504 	       normally fit in one of the two eightbytes, except that
505 	       it is placed in a way that this field straddles them.
506 	       This has been seen with a structure containing an array.
507 
508 	       The ABI is a bit unclear in this case, but we assume that
509 	       this field's class (stored in subclass[0]) must also be merged
510 	       into class[1].  In other words, our field has a piece stored
511 	       in the second eight-byte, and thus its class applies to
512 	       the second eight-byte as well.
513 
514 	       In the case where the field length exceeds 8 bytes,
515 	       it should not be necessary to merge the field class
516 	       into class[1].  As LEN > 8, subclass[1] is necessarily
517 	       different from AMD64_NO_CLASS.  If subclass[1] is equal
518 	       to subclass[0], then the normal class[1]/subclass[1]
519 	       merging will take care of everything.  For subclass[1]
520 	       to be different from subclass[0], I can only see the case
521 	       where we have a SSE/SSEUP or X87/X87UP pair, which both
522 	       use up all 16 bytes of the aggregate, and are already
523 	       handled just fine (because each portion sits on its own
524 	       8-byte).  */
525 	    class[1] = amd64_merge_classes (class[1], subclass[0]);
526 	  if (pos == 0)
527 	    class[1] = amd64_merge_classes (class[1], subclass[1]);
528 	}
529     }
530 
531   /* 4. Then a post merger cleanup is done:  */
532 
533   /* Rule (a): If one of the classes is MEMORY, the whole argument is
534      passed in memory.  */
535   if (class[0] == AMD64_MEMORY || class[1] == AMD64_MEMORY)
536     class[0] = class[1] = AMD64_MEMORY;
537 
538   /* Rule (b): If SSEUP is not preceded by SSE, it is converted to
539      SSE.  */
540   if (class[0] == AMD64_SSEUP)
541     class[0] = AMD64_SSE;
542   if (class[1] == AMD64_SSEUP && class[0] != AMD64_SSE)
543     class[1] = AMD64_SSE;
544 }
545 
546 /* Classify TYPE, and store the result in CLASS.  */
547 
548 void
549 amd64_classify (struct type *type, enum amd64_reg_class class[2])
550 {
551   enum type_code code = TYPE_CODE (type);
552   int len = TYPE_LENGTH (type);
553 
554   class[0] = class[1] = AMD64_NO_CLASS;
555 
556   /* Arguments of types (signed and unsigned) _Bool, char, short, int,
557      long, long long, and pointers are in the INTEGER class.  Similarly,
558      range types, used by languages such as Ada, are also in the INTEGER
559      class.  */
560   if ((code == TYPE_CODE_INT || code == TYPE_CODE_ENUM
561        || code == TYPE_CODE_BOOL || code == TYPE_CODE_RANGE
562        || code == TYPE_CODE_CHAR
563        || code == TYPE_CODE_PTR || code == TYPE_CODE_REF)
564       && (len == 1 || len == 2 || len == 4 || len == 8))
565     class[0] = AMD64_INTEGER;
566 
567   /* Arguments of types float, double, _Decimal32, _Decimal64 and __m64
568      are in class SSE.  */
569   else if ((code == TYPE_CODE_FLT || code == TYPE_CODE_DECFLOAT)
570 	   && (len == 4 || len == 8))
571     /* FIXME: __m64 .  */
572     class[0] = AMD64_SSE;
573 
574   /* Arguments of types __float128, _Decimal128 and __m128 are split into
575      two halves.  The least significant ones belong to class SSE, the most
576      significant one to class SSEUP.  */
577   else if (code == TYPE_CODE_DECFLOAT && len == 16)
578     /* FIXME: __float128, __m128.  */
579     class[0] = AMD64_SSE, class[1] = AMD64_SSEUP;
580 
581   /* The 64-bit mantissa of arguments of type long double belongs to
582      class X87, the 16-bit exponent plus 6 bytes of padding belongs to
583      class X87UP.  */
584   else if (code == TYPE_CODE_FLT && len == 16)
585     /* Class X87 and X87UP.  */
586     class[0] = AMD64_X87, class[1] = AMD64_X87UP;
587 
588   /* Aggregates.  */
589   else if (code == TYPE_CODE_ARRAY || code == TYPE_CODE_STRUCT
590 	   || code == TYPE_CODE_UNION)
591     amd64_classify_aggregate (type, class);
592 }
593 
594 static enum return_value_convention
595 amd64_return_value (struct gdbarch *gdbarch, struct type *func_type,
596 		    struct type *type, struct regcache *regcache,
597 		    gdb_byte *readbuf, const gdb_byte *writebuf)
598 {
599   struct gdbarch_tdep *tdep = gdbarch_tdep (gdbarch);
600   enum amd64_reg_class class[2];
601   int len = TYPE_LENGTH (type);
602   static int integer_regnum[] = { AMD64_RAX_REGNUM, AMD64_RDX_REGNUM };
603   static int sse_regnum[] = { AMD64_XMM0_REGNUM, AMD64_XMM1_REGNUM };
604   int integer_reg = 0;
605   int sse_reg = 0;
606   int i;
607 
608   gdb_assert (!(readbuf && writebuf));
609   gdb_assert (tdep->classify);
610 
611   /* 1. Classify the return type with the classification algorithm.  */
612   tdep->classify (type, class);
613 
614   /* 2. If the type has class MEMORY, then the caller provides space
615      for the return value and passes the address of this storage in
616      %rdi as if it were the first argument to the function.  In effect,
617      this address becomes a hidden first argument.
618 
619      On return %rax will contain the address that has been passed in
620      by the caller in %rdi.  */
621   if (class[0] == AMD64_MEMORY)
622     {
623       /* As indicated by the comment above, the ABI guarantees that we
624          can always find the return value just after the function has
625          returned.  */
626 
627       if (readbuf)
628 	{
629 	  ULONGEST addr;
630 
631 	  regcache_raw_read_unsigned (regcache, AMD64_RAX_REGNUM, &addr);
632 	  read_memory (addr, readbuf, TYPE_LENGTH (type));
633 	}
634 
635       return RETURN_VALUE_ABI_RETURNS_ADDRESS;
636     }
637 
638   gdb_assert (class[1] != AMD64_MEMORY);
639   gdb_assert (len <= 16);
640 
641   for (i = 0; len > 0; i++, len -= 8)
642     {
643       int regnum = -1;
644       int offset = 0;
645 
646       switch (class[i])
647 	{
648 	case AMD64_INTEGER:
649 	  /* 3. If the class is INTEGER, the next available register
650 	     of the sequence %rax, %rdx is used.  */
651 	  regnum = integer_regnum[integer_reg++];
652 	  break;
653 
654 	case AMD64_SSE:
655 	  /* 4. If the class is SSE, the next available SSE register
656              of the sequence %xmm0, %xmm1 is used.  */
657 	  regnum = sse_regnum[sse_reg++];
658 	  break;
659 
660 	case AMD64_SSEUP:
661 	  /* 5. If the class is SSEUP, the eightbyte is passed in the
662 	     upper half of the last used SSE register.  */
663 	  gdb_assert (sse_reg > 0);
664 	  regnum = sse_regnum[sse_reg - 1];
665 	  offset = 8;
666 	  break;
667 
668 	case AMD64_X87:
669 	  /* 6. If the class is X87, the value is returned on the X87
670              stack in %st0 as 80-bit x87 number.  */
671 	  regnum = AMD64_ST0_REGNUM;
672 	  if (writebuf)
673 	    i387_return_value (gdbarch, regcache);
674 	  break;
675 
676 	case AMD64_X87UP:
677 	  /* 7. If the class is X87UP, the value is returned together
678              with the previous X87 value in %st0.  */
679 	  gdb_assert (i > 0 && class[0] == AMD64_X87);
680 	  regnum = AMD64_ST0_REGNUM;
681 	  offset = 8;
682 	  len = 2;
683 	  break;
684 
685 	case AMD64_NO_CLASS:
686 	  continue;
687 
688 	default:
689 	  gdb_assert (!"Unexpected register class.");
690 	}
691 
692       gdb_assert (regnum != -1);
693 
694       if (readbuf)
695 	regcache_raw_read_part (regcache, regnum, offset, min (len, 8),
696 				readbuf + i * 8);
697       if (writebuf)
698 	regcache_raw_write_part (regcache, regnum, offset, min (len, 8),
699 				 writebuf + i * 8);
700     }
701 
702   return RETURN_VALUE_REGISTER_CONVENTION;
703 }
704 
705 
706 static CORE_ADDR
707 amd64_push_arguments (struct regcache *regcache, int nargs,
708 		      struct value **args, CORE_ADDR sp, int struct_return)
709 {
710   struct gdbarch *gdbarch = get_regcache_arch (regcache);
711   struct gdbarch_tdep *tdep = gdbarch_tdep (gdbarch);
712   int *integer_regs = tdep->call_dummy_integer_regs;
713   int num_integer_regs = tdep->call_dummy_num_integer_regs;
714 
715   static int sse_regnum[] =
716   {
717     /* %xmm0 ... %xmm7 */
718     AMD64_XMM0_REGNUM + 0, AMD64_XMM1_REGNUM,
719     AMD64_XMM0_REGNUM + 2, AMD64_XMM0_REGNUM + 3,
720     AMD64_XMM0_REGNUM + 4, AMD64_XMM0_REGNUM + 5,
721     AMD64_XMM0_REGNUM + 6, AMD64_XMM0_REGNUM + 7,
722   };
723   struct value **stack_args = alloca (nargs * sizeof (struct value *));
724   /* An array that mirrors the stack_args array.  For all arguments
725      that are passed by MEMORY, if that argument's address also needs
726      to be stored in a register, the ARG_ADDR_REGNO array will contain
727      that register number (or a negative value otherwise).  */
728   int *arg_addr_regno = alloca (nargs * sizeof (int));
729   int num_stack_args = 0;
730   int num_elements = 0;
731   int element = 0;
732   int integer_reg = 0;
733   int sse_reg = 0;
734   int i;
735 
736   gdb_assert (tdep->classify);
737 
738   /* Reserve a register for the "hidden" argument.  */
739   if (struct_return)
740     integer_reg++;
741 
742   for (i = 0; i < nargs; i++)
743     {
744       struct type *type = value_type (args[i]);
745       int len = TYPE_LENGTH (type);
746       enum amd64_reg_class class[2];
747       int needed_integer_regs = 0;
748       int needed_sse_regs = 0;
749       int j;
750 
751       /* Classify argument.  */
752       tdep->classify (type, class);
753 
754       /* Calculate the number of integer and SSE registers needed for
755          this argument.  */
756       for (j = 0; j < 2; j++)
757 	{
758 	  if (class[j] == AMD64_INTEGER)
759 	    needed_integer_regs++;
760 	  else if (class[j] == AMD64_SSE)
761 	    needed_sse_regs++;
762 	}
763 
764       /* Check whether enough registers are available, and if the
765          argument should be passed in registers at all.  */
766       if (integer_reg + needed_integer_regs > num_integer_regs
767 	  || sse_reg + needed_sse_regs > ARRAY_SIZE (sse_regnum)
768 	  || (needed_integer_regs == 0 && needed_sse_regs == 0))
769 	{
770 	  /* The argument will be passed on the stack.  */
771 	  num_elements += ((len + 7) / 8);
772 	  stack_args[num_stack_args] = args[i];
773           /* If this is an AMD64_MEMORY argument whose address must also
774              be passed in one of the integer registers, reserve that
775              register and associate this value to that register so that
776              we can store the argument address as soon as we know it.  */
777           if (class[0] == AMD64_MEMORY
778               && tdep->memory_args_by_pointer
779               && integer_reg < tdep->call_dummy_num_integer_regs)
780             arg_addr_regno[num_stack_args] =
781               tdep->call_dummy_integer_regs[integer_reg++];
782           else
783             arg_addr_regno[num_stack_args] = -1;
784           num_stack_args++;
785 	}
786       else
787 	{
788 	  /* The argument will be passed in registers.  */
789 	  const gdb_byte *valbuf = value_contents (args[i]);
790 	  gdb_byte buf[8];
791 
792 	  gdb_assert (len <= 16);
793 
794 	  for (j = 0; len > 0; j++, len -= 8)
795 	    {
796 	      int regnum = -1;
797 	      int offset = 0;
798 
799 	      switch (class[j])
800 		{
801 		case AMD64_INTEGER:
802 		  regnum = integer_regs[integer_reg++];
803 		  break;
804 
805 		case AMD64_SSE:
806 		  regnum = sse_regnum[sse_reg++];
807 		  break;
808 
809 		case AMD64_SSEUP:
810 		  gdb_assert (sse_reg > 0);
811 		  regnum = sse_regnum[sse_reg - 1];
812 		  offset = 8;
813 		  break;
814 
815 		default:
816 		  gdb_assert (!"Unexpected register class.");
817 		}
818 
819 	      gdb_assert (regnum != -1);
820 	      memset (buf, 0, sizeof buf);
821 	      memcpy (buf, valbuf + j * 8, min (len, 8));
822 	      regcache_raw_write_part (regcache, regnum, offset, 8, buf);
823 	    }
824 	}
825     }
826 
827   /* Allocate space for the arguments on the stack.  */
828   sp -= num_elements * 8;
829 
830   /* The psABI says that "The end of the input argument area shall be
831      aligned on a 16 byte boundary."  */
832   sp &= ~0xf;
833 
834   /* Write out the arguments to the stack.  */
835   for (i = 0; i < num_stack_args; i++)
836     {
837       struct type *type = value_type (stack_args[i]);
838       const gdb_byte *valbuf = value_contents (stack_args[i]);
839       int len = TYPE_LENGTH (type);
840       CORE_ADDR arg_addr = sp + element * 8;
841 
842       write_memory (arg_addr, valbuf, len);
843       if (arg_addr_regno[i] >= 0)
844         {
845           /* We also need to store the address of that argument in
846              the given register.  */
847           gdb_byte buf[8];
848           enum bfd_endian byte_order = gdbarch_byte_order (gdbarch);
849 
850           store_unsigned_integer (buf, 8, byte_order, arg_addr);
851           regcache_cooked_write (regcache, arg_addr_regno[i], buf);
852         }
853       element += ((len + 7) / 8);
854     }
855 
856   /* The psABI says that "For calls that may call functions that use
857      varargs or stdargs (prototype-less calls or calls to functions
858      containing ellipsis (...) in the declaration) %al is used as
859      hidden argument to specify the number of SSE registers used.  */
860   regcache_raw_write_unsigned (regcache, AMD64_RAX_REGNUM, sse_reg);
861   return sp;
862 }
863 
864 static CORE_ADDR
865 amd64_push_dummy_call (struct gdbarch *gdbarch, struct value *function,
866 		       struct regcache *regcache, CORE_ADDR bp_addr,
867 		       int nargs, struct value **args,	CORE_ADDR sp,
868 		       int struct_return, CORE_ADDR struct_addr)
869 {
870   enum bfd_endian byte_order = gdbarch_byte_order (gdbarch);
871   struct gdbarch_tdep *tdep = gdbarch_tdep (gdbarch);
872   gdb_byte buf[8];
873 
874   /* Pass arguments.  */
875   sp = amd64_push_arguments (regcache, nargs, args, sp, struct_return);
876 
877   /* Pass "hidden" argument".  */
878   if (struct_return)
879     {
880       /* The "hidden" argument is passed throught the first argument
881          register.  */
882       const int arg_regnum = tdep->call_dummy_integer_regs[0];
883 
884       store_unsigned_integer (buf, 8, byte_order, struct_addr);
885       regcache_cooked_write (regcache, arg_regnum, buf);
886     }
887 
888   /* Reserve some memory on the stack for the integer-parameter registers,
889      if required by the ABI.  */
890   if (tdep->integer_param_regs_saved_in_caller_frame)
891     sp -= tdep->call_dummy_num_integer_regs * 8;
892 
893   /* Store return address.  */
894   sp -= 8;
895   store_unsigned_integer (buf, 8, byte_order, bp_addr);
896   write_memory (sp, buf, 8);
897 
898   /* Finally, update the stack pointer...  */
899   store_unsigned_integer (buf, 8, byte_order, sp);
900   regcache_cooked_write (regcache, AMD64_RSP_REGNUM, buf);
901 
902   /* ...and fake a frame pointer.  */
903   regcache_cooked_write (regcache, AMD64_RBP_REGNUM, buf);
904 
905   return sp + 16;
906 }
907 
908 /* Displaced instruction handling.  */
909 
910 /* A partially decoded instruction.
911    This contains enough details for displaced stepping purposes.  */
912 
913 struct amd64_insn
914 {
915   /* The number of opcode bytes.  */
916   int opcode_len;
917   /* The offset of the rex prefix or -1 if not present.  */
918   int rex_offset;
919   /* The offset to the first opcode byte.  */
920   int opcode_offset;
921   /* The offset to the modrm byte or -1 if not present.  */
922   int modrm_offset;
923 
924   /* The raw instruction.  */
925   gdb_byte *raw_insn;
926 };
927 
928 struct displaced_step_closure
929 {
930   /* For rip-relative insns, saved copy of the reg we use instead of %rip.  */
931   int tmp_used;
932   int tmp_regno;
933   ULONGEST tmp_save;
934 
935   /* Details of the instruction.  */
936   struct amd64_insn insn_details;
937 
938   /* Amount of space allocated to insn_buf.  */
939   int max_len;
940 
941   /* The possibly modified insn.
942      This is a variable-length field.  */
943   gdb_byte insn_buf[1];
944 };
945 
946 /* WARNING: Keep onebyte_has_modrm, twobyte_has_modrm in sync with
947    ../opcodes/i386-dis.c (until libopcodes exports them, or an alternative,
948    at which point delete these in favor of libopcodes' versions).  */
949 
950 static const unsigned char onebyte_has_modrm[256] = {
951   /*	   0 1 2 3 4 5 6 7 8 9 a b c d e f	  */
952   /*	   -------------------------------	  */
953   /* 00 */ 1,1,1,1,0,0,0,0,1,1,1,1,0,0,0,0, /* 00 */
954   /* 10 */ 1,1,1,1,0,0,0,0,1,1,1,1,0,0,0,0, /* 10 */
955   /* 20 */ 1,1,1,1,0,0,0,0,1,1,1,1,0,0,0,0, /* 20 */
956   /* 30 */ 1,1,1,1,0,0,0,0,1,1,1,1,0,0,0,0, /* 30 */
957   /* 40 */ 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /* 40 */
958   /* 50 */ 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /* 50 */
959   /* 60 */ 0,0,1,1,0,0,0,0,0,1,0,1,0,0,0,0, /* 60 */
960   /* 70 */ 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /* 70 */
961   /* 80 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* 80 */
962   /* 90 */ 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /* 90 */
963   /* a0 */ 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /* a0 */
964   /* b0 */ 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /* b0 */
965   /* c0 */ 1,1,0,0,1,1,1,1,0,0,0,0,0,0,0,0, /* c0 */
966   /* d0 */ 1,1,1,1,0,0,0,0,1,1,1,1,1,1,1,1, /* d0 */
967   /* e0 */ 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /* e0 */
968   /* f0 */ 0,0,0,0,0,0,1,1,0,0,0,0,0,0,1,1  /* f0 */
969   /*	   -------------------------------	  */
970   /*	   0 1 2 3 4 5 6 7 8 9 a b c d e f	  */
971 };
972 
973 static const unsigned char twobyte_has_modrm[256] = {
974   /*	   0 1 2 3 4 5 6 7 8 9 a b c d e f	  */
975   /*	   -------------------------------	  */
976   /* 00 */ 1,1,1,1,0,0,0,0,0,0,0,0,0,1,0,1, /* 0f */
977   /* 10 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* 1f */
978   /* 20 */ 1,1,1,1,1,1,1,0,1,1,1,1,1,1,1,1, /* 2f */
979   /* 30 */ 0,0,0,0,0,0,0,0,1,0,1,0,0,0,0,0, /* 3f */
980   /* 40 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* 4f */
981   /* 50 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* 5f */
982   /* 60 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* 6f */
983   /* 70 */ 1,1,1,1,1,1,1,0,1,1,1,1,1,1,1,1, /* 7f */
984   /* 80 */ 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /* 8f */
985   /* 90 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* 9f */
986   /* a0 */ 0,0,0,1,1,1,1,1,0,0,0,1,1,1,1,1, /* af */
987   /* b0 */ 1,1,1,1,1,1,1,1,1,0,1,1,1,1,1,1, /* bf */
988   /* c0 */ 1,1,1,1,1,1,1,1,0,0,0,0,0,0,0,0, /* cf */
989   /* d0 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* df */
990   /* e0 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* ef */
991   /* f0 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,0  /* ff */
992   /*	   -------------------------------	  */
993   /*	   0 1 2 3 4 5 6 7 8 9 a b c d e f	  */
994 };
995 
996 static int amd64_syscall_p (const struct amd64_insn *insn, int *lengthp);
997 
998 static int
999 rex_prefix_p (gdb_byte pfx)
1000 {
1001   return REX_PREFIX_P (pfx);
1002 }
1003 
1004 /* Skip the legacy instruction prefixes in INSN.
1005    We assume INSN is properly sentineled so we don't have to worry
1006    about falling off the end of the buffer.  */
1007 
1008 static gdb_byte *
1009 amd64_skip_prefixes (gdb_byte *insn)
1010 {
1011   while (1)
1012     {
1013       switch (*insn)
1014 	{
1015 	case DATA_PREFIX_OPCODE:
1016 	case ADDR_PREFIX_OPCODE:
1017 	case CS_PREFIX_OPCODE:
1018 	case DS_PREFIX_OPCODE:
1019 	case ES_PREFIX_OPCODE:
1020 	case FS_PREFIX_OPCODE:
1021 	case GS_PREFIX_OPCODE:
1022 	case SS_PREFIX_OPCODE:
1023 	case LOCK_PREFIX_OPCODE:
1024 	case REPE_PREFIX_OPCODE:
1025 	case REPNE_PREFIX_OPCODE:
1026 	  ++insn;
1027 	  continue;
1028 	default:
1029 	  break;
1030 	}
1031       break;
1032     }
1033 
1034   return insn;
1035 }
1036 
1037 /* Return an integer register (other than RSP) that is unused as an input
1038    operand in INSN.
1039    In order to not require adding a rex prefix if the insn doesn't already
1040    have one, the result is restricted to RAX ... RDI, sans RSP.
1041    The register numbering of the result follows architecture ordering,
1042    e.g. RDI = 7.  */
1043 
1044 static int
1045 amd64_get_unused_input_int_reg (const struct amd64_insn *details)
1046 {
1047   /* 1 bit for each reg */
1048   int used_regs_mask = 0;
1049 
1050   /* There can be at most 3 int regs used as inputs in an insn, and we have
1051      7 to choose from (RAX ... RDI, sans RSP).
1052      This allows us to take a conservative approach and keep things simple.
1053      E.g. By avoiding RAX, we don't have to specifically watch for opcodes
1054      that implicitly specify RAX.  */
1055 
1056   /* Avoid RAX.  */
1057   used_regs_mask |= 1 << EAX_REG_NUM;
1058   /* Similarily avoid RDX, implicit operand in divides.  */
1059   used_regs_mask |= 1 << EDX_REG_NUM;
1060   /* Avoid RSP.  */
1061   used_regs_mask |= 1 << ESP_REG_NUM;
1062 
1063   /* If the opcode is one byte long and there's no ModRM byte,
1064      assume the opcode specifies a register.  */
1065   if (details->opcode_len == 1 && details->modrm_offset == -1)
1066     used_regs_mask |= 1 << (details->raw_insn[details->opcode_offset] & 7);
1067 
1068   /* Mark used regs in the modrm/sib bytes.  */
1069   if (details->modrm_offset != -1)
1070     {
1071       int modrm = details->raw_insn[details->modrm_offset];
1072       int mod = MODRM_MOD_FIELD (modrm);
1073       int reg = MODRM_REG_FIELD (modrm);
1074       int rm = MODRM_RM_FIELD (modrm);
1075       int have_sib = mod != 3 && rm == 4;
1076 
1077       /* Assume the reg field of the modrm byte specifies a register.  */
1078       used_regs_mask |= 1 << reg;
1079 
1080       if (have_sib)
1081 	{
1082 	  int base = SIB_BASE_FIELD (details->raw_insn[details->modrm_offset + 1]);
1083 	  int idx = SIB_INDEX_FIELD (details->raw_insn[details->modrm_offset + 1]);
1084 	  used_regs_mask |= 1 << base;
1085 	  used_regs_mask |= 1 << idx;
1086 	}
1087       else
1088 	{
1089 	  used_regs_mask |= 1 << rm;
1090 	}
1091     }
1092 
1093   gdb_assert (used_regs_mask < 256);
1094   gdb_assert (used_regs_mask != 255);
1095 
1096   /* Finally, find a free reg.  */
1097   {
1098     int i;
1099 
1100     for (i = 0; i < 8; ++i)
1101       {
1102 	if (! (used_regs_mask & (1 << i)))
1103 	  return i;
1104       }
1105 
1106     /* We shouldn't get here.  */
1107     internal_error (__FILE__, __LINE__, _("unable to find free reg"));
1108   }
1109 }
1110 
1111 /* Extract the details of INSN that we need.  */
1112 
1113 static void
1114 amd64_get_insn_details (gdb_byte *insn, struct amd64_insn *details)
1115 {
1116   gdb_byte *start = insn;
1117   int need_modrm;
1118 
1119   details->raw_insn = insn;
1120 
1121   details->opcode_len = -1;
1122   details->rex_offset = -1;
1123   details->opcode_offset = -1;
1124   details->modrm_offset = -1;
1125 
1126   /* Skip legacy instruction prefixes.  */
1127   insn = amd64_skip_prefixes (insn);
1128 
1129   /* Skip REX instruction prefix.  */
1130   if (rex_prefix_p (*insn))
1131     {
1132       details->rex_offset = insn - start;
1133       ++insn;
1134     }
1135 
1136   details->opcode_offset = insn - start;
1137 
1138   if (*insn == TWO_BYTE_OPCODE_ESCAPE)
1139     {
1140       /* Two or three-byte opcode.  */
1141       ++insn;
1142       need_modrm = twobyte_has_modrm[*insn];
1143 
1144       /* Check for three-byte opcode.  */
1145       switch (*insn)
1146 	{
1147 	case 0x24:
1148 	case 0x25:
1149 	case 0x38:
1150 	case 0x3a:
1151 	case 0x7a:
1152 	case 0x7b:
1153 	  ++insn;
1154 	  details->opcode_len = 3;
1155 	  break;
1156 	default:
1157 	  details->opcode_len = 2;
1158 	  break;
1159 	}
1160     }
1161   else
1162     {
1163       /* One-byte opcode.  */
1164       need_modrm = onebyte_has_modrm[*insn];
1165       details->opcode_len = 1;
1166     }
1167 
1168   if (need_modrm)
1169     {
1170       ++insn;
1171       details->modrm_offset = insn - start;
1172     }
1173 }
1174 
1175 /* Update %rip-relative addressing in INSN.
1176 
1177    %rip-relative addressing only uses a 32-bit displacement.
1178    32 bits is not enough to be guaranteed to cover the distance between where
1179    the real instruction is and where its copy is.
1180    Convert the insn to use base+disp addressing.
1181    We set base = pc + insn_length so we can leave disp unchanged.  */
1182 
1183 static void
1184 fixup_riprel (struct gdbarch *gdbarch, struct displaced_step_closure *dsc,
1185 	      CORE_ADDR from, CORE_ADDR to, struct regcache *regs)
1186 {
1187   enum bfd_endian byte_order = gdbarch_byte_order (gdbarch);
1188   const struct amd64_insn *insn_details = &dsc->insn_details;
1189   int modrm_offset = insn_details->modrm_offset;
1190   gdb_byte *insn = insn_details->raw_insn + modrm_offset;
1191   CORE_ADDR rip_base;
1192   int32_t disp;
1193   int insn_length;
1194   int arch_tmp_regno, tmp_regno;
1195   ULONGEST orig_value;
1196 
1197   /* %rip+disp32 addressing mode, displacement follows ModRM byte.  */
1198   ++insn;
1199 
1200   /* Compute the rip-relative address.	*/
1201   disp = extract_signed_integer (insn, sizeof (int32_t), byte_order);
1202   insn_length = gdb_buffered_insn_length (gdbarch, dsc->insn_buf,
1203 					  dsc->max_len, from);
1204   rip_base = from + insn_length;
1205 
1206   /* We need a register to hold the address.
1207      Pick one not used in the insn.
1208      NOTE: arch_tmp_regno uses architecture ordering, e.g. RDI = 7.  */
1209   arch_tmp_regno = amd64_get_unused_input_int_reg (insn_details);
1210   tmp_regno = amd64_arch_reg_to_regnum (arch_tmp_regno);
1211 
1212   /* REX.B should be unset as we were using rip-relative addressing,
1213      but ensure it's unset anyway, tmp_regno is not r8-r15.  */
1214   if (insn_details->rex_offset != -1)
1215     dsc->insn_buf[insn_details->rex_offset] &= ~REX_B;
1216 
1217   regcache_cooked_read_unsigned (regs, tmp_regno, &orig_value);
1218   dsc->tmp_regno = tmp_regno;
1219   dsc->tmp_save = orig_value;
1220   dsc->tmp_used = 1;
1221 
1222   /* Convert the ModRM field to be base+disp.  */
1223   dsc->insn_buf[modrm_offset] &= ~0xc7;
1224   dsc->insn_buf[modrm_offset] |= 0x80 + arch_tmp_regno;
1225 
1226   regcache_cooked_write_unsigned (regs, tmp_regno, rip_base);
1227 
1228   if (debug_displaced)
1229     fprintf_unfiltered (gdb_stdlog, "displaced: %%rip-relative addressing used.\n"
1230 			"displaced: using temp reg %d, old value %s, new value %s\n",
1231 			dsc->tmp_regno, paddress (gdbarch, dsc->tmp_save),
1232 			paddress (gdbarch, rip_base));
1233 }
1234 
1235 static void
1236 fixup_displaced_copy (struct gdbarch *gdbarch,
1237 		      struct displaced_step_closure *dsc,
1238 		      CORE_ADDR from, CORE_ADDR to, struct regcache *regs)
1239 {
1240   const struct amd64_insn *details = &dsc->insn_details;
1241 
1242   if (details->modrm_offset != -1)
1243     {
1244       gdb_byte modrm = details->raw_insn[details->modrm_offset];
1245 
1246       if ((modrm & 0xc7) == 0x05)
1247 	{
1248 	  /* The insn uses rip-relative addressing.
1249 	     Deal with it.  */
1250 	  fixup_riprel (gdbarch, dsc, from, to, regs);
1251 	}
1252     }
1253 }
1254 
1255 struct displaced_step_closure *
1256 amd64_displaced_step_copy_insn (struct gdbarch *gdbarch,
1257 				CORE_ADDR from, CORE_ADDR to,
1258 				struct regcache *regs)
1259 {
1260   int len = gdbarch_max_insn_length (gdbarch);
1261   /* Extra space for sentinels so fixup_{riprel,displaced_copy don't have to
1262      continually watch for running off the end of the buffer.  */
1263   int fixup_sentinel_space = len;
1264   struct displaced_step_closure *dsc =
1265     xmalloc (sizeof (*dsc) + len + fixup_sentinel_space);
1266   gdb_byte *buf = &dsc->insn_buf[0];
1267   struct amd64_insn *details = &dsc->insn_details;
1268 
1269   dsc->tmp_used = 0;
1270   dsc->max_len = len + fixup_sentinel_space;
1271 
1272   read_memory (from, buf, len);
1273 
1274   /* Set up the sentinel space so we don't have to worry about running
1275      off the end of the buffer.  An excessive number of leading prefixes
1276      could otherwise cause this.  */
1277   memset (buf + len, 0, fixup_sentinel_space);
1278 
1279   amd64_get_insn_details (buf, details);
1280 
1281   /* GDB may get control back after the insn after the syscall.
1282      Presumably this is a kernel bug.
1283      If this is a syscall, make sure there's a nop afterwards.  */
1284   {
1285     int syscall_length;
1286 
1287     if (amd64_syscall_p (details, &syscall_length))
1288       buf[details->opcode_offset + syscall_length] = NOP_OPCODE;
1289   }
1290 
1291   /* Modify the insn to cope with the address where it will be executed from.
1292      In particular, handle any rip-relative addressing.	 */
1293   fixup_displaced_copy (gdbarch, dsc, from, to, regs);
1294 
1295   write_memory (to, buf, len);
1296 
1297   if (debug_displaced)
1298     {
1299       fprintf_unfiltered (gdb_stdlog, "displaced: copy %s->%s: ",
1300 			  paddress (gdbarch, from), paddress (gdbarch, to));
1301       displaced_step_dump_bytes (gdb_stdlog, buf, len);
1302     }
1303 
1304   return dsc;
1305 }
1306 
1307 static int
1308 amd64_absolute_jmp_p (const struct amd64_insn *details)
1309 {
1310   const gdb_byte *insn = &details->raw_insn[details->opcode_offset];
1311 
1312   if (insn[0] == 0xff)
1313     {
1314       /* jump near, absolute indirect (/4) */
1315       if ((insn[1] & 0x38) == 0x20)
1316 	return 1;
1317 
1318       /* jump far, absolute indirect (/5) */
1319       if ((insn[1] & 0x38) == 0x28)
1320 	return 1;
1321     }
1322 
1323   return 0;
1324 }
1325 
1326 static int
1327 amd64_absolute_call_p (const struct amd64_insn *details)
1328 {
1329   const gdb_byte *insn = &details->raw_insn[details->opcode_offset];
1330 
1331   if (insn[0] == 0xff)
1332     {
1333       /* Call near, absolute indirect (/2) */
1334       if ((insn[1] & 0x38) == 0x10)
1335 	return 1;
1336 
1337       /* Call far, absolute indirect (/3) */
1338       if ((insn[1] & 0x38) == 0x18)
1339 	return 1;
1340     }
1341 
1342   return 0;
1343 }
1344 
1345 static int
1346 amd64_ret_p (const struct amd64_insn *details)
1347 {
1348   /* NOTE: gcc can emit "repz ; ret".  */
1349   const gdb_byte *insn = &details->raw_insn[details->opcode_offset];
1350 
1351   switch (insn[0])
1352     {
1353     case 0xc2: /* ret near, pop N bytes */
1354     case 0xc3: /* ret near */
1355     case 0xca: /* ret far, pop N bytes */
1356     case 0xcb: /* ret far */
1357     case 0xcf: /* iret */
1358       return 1;
1359 
1360     default:
1361       return 0;
1362     }
1363 }
1364 
1365 static int
1366 amd64_call_p (const struct amd64_insn *details)
1367 {
1368   const gdb_byte *insn = &details->raw_insn[details->opcode_offset];
1369 
1370   if (amd64_absolute_call_p (details))
1371     return 1;
1372 
1373   /* call near, relative */
1374   if (insn[0] == 0xe8)
1375     return 1;
1376 
1377   return 0;
1378 }
1379 
1380 /* Return non-zero if INSN is a system call, and set *LENGTHP to its
1381    length in bytes.  Otherwise, return zero.  */
1382 
1383 static int
1384 amd64_syscall_p (const struct amd64_insn *details, int *lengthp)
1385 {
1386   const gdb_byte *insn = &details->raw_insn[details->opcode_offset];
1387 
1388   if (insn[0] == 0x0f && insn[1] == 0x05)
1389     {
1390       *lengthp = 2;
1391       return 1;
1392     }
1393 
1394   return 0;
1395 }
1396 
1397 /* Fix up the state of registers and memory after having single-stepped
1398    a displaced instruction.  */
1399 
1400 void
1401 amd64_displaced_step_fixup (struct gdbarch *gdbarch,
1402 			    struct displaced_step_closure *dsc,
1403 			    CORE_ADDR from, CORE_ADDR to,
1404 			    struct regcache *regs)
1405 {
1406   enum bfd_endian byte_order = gdbarch_byte_order (gdbarch);
1407   /* The offset we applied to the instruction's address.  */
1408   ULONGEST insn_offset = to - from;
1409   gdb_byte *insn = dsc->insn_buf;
1410   const struct amd64_insn *insn_details = &dsc->insn_details;
1411 
1412   if (debug_displaced)
1413     fprintf_unfiltered (gdb_stdlog,
1414 			"displaced: fixup (%s, %s), "
1415 			"insn = 0x%02x 0x%02x ...\n",
1416 			paddress (gdbarch, from), paddress (gdbarch, to),
1417 			insn[0], insn[1]);
1418 
1419   /* If we used a tmp reg, restore it.	*/
1420 
1421   if (dsc->tmp_used)
1422     {
1423       if (debug_displaced)
1424 	fprintf_unfiltered (gdb_stdlog, "displaced: restoring reg %d to %s\n",
1425 			    dsc->tmp_regno, paddress (gdbarch, dsc->tmp_save));
1426       regcache_cooked_write_unsigned (regs, dsc->tmp_regno, dsc->tmp_save);
1427     }
1428 
1429   /* The list of issues to contend with here is taken from
1430      resume_execution in arch/x86/kernel/kprobes.c, Linux 2.6.28.
1431      Yay for Free Software!  */
1432 
1433   /* Relocate the %rip back to the program's instruction stream,
1434      if necessary.  */
1435 
1436   /* Except in the case of absolute or indirect jump or call
1437      instructions, or a return instruction, the new rip is relative to
1438      the displaced instruction; make it relative to the original insn.
1439      Well, signal handler returns don't need relocation either, but we use the
1440      value of %rip to recognize those; see below.  */
1441   if (! amd64_absolute_jmp_p (insn_details)
1442       && ! amd64_absolute_call_p (insn_details)
1443       && ! amd64_ret_p (insn_details))
1444     {
1445       ULONGEST orig_rip;
1446       int insn_len;
1447 
1448       regcache_cooked_read_unsigned (regs, AMD64_RIP_REGNUM, &orig_rip);
1449 
1450       /* A signal trampoline system call changes the %rip, resuming
1451 	 execution of the main program after the signal handler has
1452 	 returned.  That makes them like 'return' instructions; we
1453 	 shouldn't relocate %rip.
1454 
1455 	 But most system calls don't, and we do need to relocate %rip.
1456 
1457 	 Our heuristic for distinguishing these cases: if stepping
1458 	 over the system call instruction left control directly after
1459 	 the instruction, the we relocate --- control almost certainly
1460 	 doesn't belong in the displaced copy.	Otherwise, we assume
1461 	 the instruction has put control where it belongs, and leave
1462 	 it unrelocated.  Goodness help us if there are PC-relative
1463 	 system calls.	*/
1464       if (amd64_syscall_p (insn_details, &insn_len)
1465 	  && orig_rip != to + insn_len
1466 	  /* GDB can get control back after the insn after the syscall.
1467 	     Presumably this is a kernel bug.
1468 	     Fixup ensures its a nop, we add one to the length for it.  */
1469 	  && orig_rip != to + insn_len + 1)
1470 	{
1471 	  if (debug_displaced)
1472 	    fprintf_unfiltered (gdb_stdlog,
1473 				"displaced: syscall changed %%rip; "
1474 				"not relocating\n");
1475 	}
1476       else
1477 	{
1478 	  ULONGEST rip = orig_rip - insn_offset;
1479 
1480 	  /* If we just stepped over a breakpoint insn, we don't backup
1481 	     the pc on purpose; this is to match behaviour without
1482 	     stepping.  */
1483 
1484 	  regcache_cooked_write_unsigned (regs, AMD64_RIP_REGNUM, rip);
1485 
1486 	  if (debug_displaced)
1487 	    fprintf_unfiltered (gdb_stdlog,
1488 				"displaced: "
1489 				"relocated %%rip from %s to %s\n",
1490 				paddress (gdbarch, orig_rip),
1491 				paddress (gdbarch, rip));
1492 	}
1493     }
1494 
1495   /* If the instruction was PUSHFL, then the TF bit will be set in the
1496      pushed value, and should be cleared.  We'll leave this for later,
1497      since GDB already messes up the TF flag when stepping over a
1498      pushfl.  */
1499 
1500   /* If the instruction was a call, the return address now atop the
1501      stack is the address following the copied instruction.  We need
1502      to make it the address following the original instruction.	 */
1503   if (amd64_call_p (insn_details))
1504     {
1505       ULONGEST rsp;
1506       ULONGEST retaddr;
1507       const ULONGEST retaddr_len = 8;
1508 
1509       regcache_cooked_read_unsigned (regs, AMD64_RSP_REGNUM, &rsp);
1510       retaddr = read_memory_unsigned_integer (rsp, retaddr_len, byte_order);
1511       retaddr = (retaddr - insn_offset) & 0xffffffffUL;
1512       write_memory_unsigned_integer (rsp, retaddr_len, byte_order, retaddr);
1513 
1514       if (debug_displaced)
1515 	fprintf_unfiltered (gdb_stdlog,
1516 			    "displaced: relocated return addr at %s "
1517 			    "to %s\n",
1518 			    paddress (gdbarch, rsp),
1519 			    paddress (gdbarch, retaddr));
1520     }
1521 }
1522 
1523 /* If the instruction INSN uses RIP-relative addressing, return the
1524    offset into the raw INSN where the displacement to be adjusted is
1525    found.  Returns 0 if the instruction doesn't use RIP-relative
1526    addressing.  */
1527 
1528 static int
1529 rip_relative_offset (struct amd64_insn *insn)
1530 {
1531   if (insn->modrm_offset != -1)
1532     {
1533       gdb_byte modrm = insn->raw_insn[insn->modrm_offset];
1534 
1535       if ((modrm & 0xc7) == 0x05)
1536 	{
1537 	  /* The displacement is found right after the ModRM byte.  */
1538 	  return insn->modrm_offset + 1;
1539 	}
1540     }
1541 
1542   return 0;
1543 }
1544 
1545 static void
1546 append_insns (CORE_ADDR *to, ULONGEST len, const gdb_byte *buf)
1547 {
1548   target_write_memory (*to, buf, len);
1549   *to += len;
1550 }
1551 
1552 static void
1553 amd64_relocate_instruction (struct gdbarch *gdbarch,
1554 			    CORE_ADDR *to, CORE_ADDR oldloc)
1555 {
1556   enum bfd_endian byte_order = gdbarch_byte_order (gdbarch);
1557   int len = gdbarch_max_insn_length (gdbarch);
1558   /* Extra space for sentinels.  */
1559   int fixup_sentinel_space = len;
1560   gdb_byte *buf = xmalloc (len + fixup_sentinel_space);
1561   struct amd64_insn insn_details;
1562   int offset = 0;
1563   LONGEST rel32, newrel;
1564   gdb_byte *insn;
1565   int insn_length;
1566 
1567   read_memory (oldloc, buf, len);
1568 
1569   /* Set up the sentinel space so we don't have to worry about running
1570      off the end of the buffer.  An excessive number of leading prefixes
1571      could otherwise cause this.  */
1572   memset (buf + len, 0, fixup_sentinel_space);
1573 
1574   insn = buf;
1575   amd64_get_insn_details (insn, &insn_details);
1576 
1577   insn_length = gdb_buffered_insn_length (gdbarch, insn, len, oldloc);
1578 
1579   /* Skip legacy instruction prefixes.  */
1580   insn = amd64_skip_prefixes (insn);
1581 
1582   /* Adjust calls with 32-bit relative addresses as push/jump, with
1583      the address pushed being the location where the original call in
1584      the user program would return to.  */
1585   if (insn[0] == 0xe8)
1586     {
1587       gdb_byte push_buf[16];
1588       unsigned int ret_addr;
1589 
1590       /* Where "ret" in the original code will return to.  */
1591       ret_addr = oldloc + insn_length;
1592       push_buf[0] = 0x68; /* pushq $...  */
1593       memcpy (&push_buf[1], &ret_addr, 4);
1594       /* Push the push.  */
1595       append_insns (to, 5, push_buf);
1596 
1597       /* Convert the relative call to a relative jump.  */
1598       insn[0] = 0xe9;
1599 
1600       /* Adjust the destination offset.  */
1601       rel32 = extract_signed_integer (insn + 1, 4, byte_order);
1602       newrel = (oldloc - *to) + rel32;
1603       store_signed_integer (insn + 1, 4, byte_order, newrel);
1604 
1605       if (debug_displaced)
1606 	fprintf_unfiltered (gdb_stdlog,
1607 			    "Adjusted insn rel32=%s at %s to"
1608 			    " rel32=%s at %s\n",
1609 			    hex_string (rel32), paddress (gdbarch, oldloc),
1610 			    hex_string (newrel), paddress (gdbarch, *to));
1611 
1612       /* Write the adjusted jump into its displaced location.  */
1613       append_insns (to, 5, insn);
1614       return;
1615     }
1616 
1617   offset = rip_relative_offset (&insn_details);
1618   if (!offset)
1619     {
1620       /* Adjust jumps with 32-bit relative addresses.  Calls are
1621 	 already handled above.  */
1622       if (insn[0] == 0xe9)
1623 	offset = 1;
1624       /* Adjust conditional jumps.  */
1625       else if (insn[0] == 0x0f && (insn[1] & 0xf0) == 0x80)
1626 	offset = 2;
1627     }
1628 
1629   if (offset)
1630     {
1631       rel32 = extract_signed_integer (insn + offset, 4, byte_order);
1632       newrel = (oldloc - *to) + rel32;
1633       store_signed_integer (insn + offset, 4, byte_order, newrel);
1634       if (debug_displaced)
1635 	fprintf_unfiltered (gdb_stdlog,
1636 			    "Adjusted insn rel32=%s at %s to"
1637 			    " rel32=%s at %s\n",
1638 			    hex_string (rel32), paddress (gdbarch, oldloc),
1639 			    hex_string (newrel), paddress (gdbarch, *to));
1640     }
1641 
1642   /* Write the adjusted instruction into its displaced location.  */
1643   append_insns (to, insn_length, buf);
1644 }
1645 
1646 
1647 /* The maximum number of saved registers.  This should include %rip.  */
1648 #define AMD64_NUM_SAVED_REGS	AMD64_NUM_GREGS
1649 
1650 struct amd64_frame_cache
1651 {
1652   /* Base address.  */
1653   CORE_ADDR base;
1654   int base_p;
1655   CORE_ADDR sp_offset;
1656   CORE_ADDR pc;
1657 
1658   /* Saved registers.  */
1659   CORE_ADDR saved_regs[AMD64_NUM_SAVED_REGS];
1660   CORE_ADDR saved_sp;
1661   int saved_sp_reg;
1662 
1663   /* Do we have a frame?  */
1664   int frameless_p;
1665 };
1666 
1667 /* Initialize a frame cache.  */
1668 
1669 static void
1670 amd64_init_frame_cache (struct amd64_frame_cache *cache)
1671 {
1672   int i;
1673 
1674   /* Base address.  */
1675   cache->base = 0;
1676   cache->base_p = 0;
1677   cache->sp_offset = -8;
1678   cache->pc = 0;
1679 
1680   /* Saved registers.  We initialize these to -1 since zero is a valid
1681      offset (that's where %rbp is supposed to be stored).
1682      The values start out as being offsets, and are later converted to
1683      addresses (at which point -1 is interpreted as an address, still meaning
1684      "invalid").  */
1685   for (i = 0; i < AMD64_NUM_SAVED_REGS; i++)
1686     cache->saved_regs[i] = -1;
1687   cache->saved_sp = 0;
1688   cache->saved_sp_reg = -1;
1689 
1690   /* Frameless until proven otherwise.  */
1691   cache->frameless_p = 1;
1692 }
1693 
1694 /* Allocate and initialize a frame cache.  */
1695 
1696 static struct amd64_frame_cache *
1697 amd64_alloc_frame_cache (void)
1698 {
1699   struct amd64_frame_cache *cache;
1700 
1701   cache = FRAME_OBSTACK_ZALLOC (struct amd64_frame_cache);
1702   amd64_init_frame_cache (cache);
1703   return cache;
1704 }
1705 
1706 /* GCC 4.4 and later, can put code in the prologue to realign the
1707    stack pointer.  Check whether PC points to such code, and update
1708    CACHE accordingly.  Return the first instruction after the code
1709    sequence or CURRENT_PC, whichever is smaller.  If we don't
1710    recognize the code, return PC.  */
1711 
1712 static CORE_ADDR
1713 amd64_analyze_stack_align (CORE_ADDR pc, CORE_ADDR current_pc,
1714 			   struct amd64_frame_cache *cache)
1715 {
1716   /* There are 2 code sequences to re-align stack before the frame
1717      gets set up:
1718 
1719 	1. Use a caller-saved saved register:
1720 
1721 		leaq  8(%rsp), %reg
1722 		andq  $-XXX, %rsp
1723 		pushq -8(%reg)
1724 
1725 	2. Use a callee-saved saved register:
1726 
1727 		pushq %reg
1728 		leaq  16(%rsp), %reg
1729 		andq  $-XXX, %rsp
1730 		pushq -8(%reg)
1731 
1732      "andq $-XXX, %rsp" can be either 4 bytes or 7 bytes:
1733 
1734      	0x48 0x83 0xe4 0xf0			andq $-16, %rsp
1735      	0x48 0x81 0xe4 0x00 0xff 0xff 0xff	andq $-256, %rsp
1736    */
1737 
1738   gdb_byte buf[18];
1739   int reg, r;
1740   int offset, offset_and;
1741 
1742   if (target_read_memory (pc, buf, sizeof buf))
1743     return pc;
1744 
1745   /* Check caller-saved saved register.  The first instruction has
1746      to be "leaq 8(%rsp), %reg".  */
1747   if ((buf[0] & 0xfb) == 0x48
1748       && buf[1] == 0x8d
1749       && buf[3] == 0x24
1750       && buf[4] == 0x8)
1751     {
1752       /* MOD must be binary 10 and R/M must be binary 100.  */
1753       if ((buf[2] & 0xc7) != 0x44)
1754 	return pc;
1755 
1756       /* REG has register number.  */
1757       reg = (buf[2] >> 3) & 7;
1758 
1759       /* Check the REX.R bit.  */
1760       if (buf[0] == 0x4c)
1761 	reg += 8;
1762 
1763       offset = 5;
1764     }
1765   else
1766     {
1767       /* Check callee-saved saved register.  The first instruction
1768 	 has to be "pushq %reg".  */
1769       reg = 0;
1770       if ((buf[0] & 0xf8) == 0x50)
1771 	offset = 0;
1772       else if ((buf[0] & 0xf6) == 0x40
1773 	       && (buf[1] & 0xf8) == 0x50)
1774 	{
1775 	  /* Check the REX.B bit.  */
1776 	  if ((buf[0] & 1) != 0)
1777 	    reg = 8;
1778 
1779 	  offset = 1;
1780 	}
1781       else
1782 	return pc;
1783 
1784       /* Get register.  */
1785       reg += buf[offset] & 0x7;
1786 
1787       offset++;
1788 
1789       /* The next instruction has to be "leaq 16(%rsp), %reg".  */
1790       if ((buf[offset] & 0xfb) != 0x48
1791 	  || buf[offset + 1] != 0x8d
1792 	  || buf[offset + 3] != 0x24
1793 	  || buf[offset + 4] != 0x10)
1794 	return pc;
1795 
1796       /* MOD must be binary 10 and R/M must be binary 100.  */
1797       if ((buf[offset + 2] & 0xc7) != 0x44)
1798 	return pc;
1799 
1800       /* REG has register number.  */
1801       r = (buf[offset + 2] >> 3) & 7;
1802 
1803       /* Check the REX.R bit.  */
1804       if (buf[offset] == 0x4c)
1805 	r += 8;
1806 
1807       /* Registers in pushq and leaq have to be the same.  */
1808       if (reg != r)
1809 	return pc;
1810 
1811       offset += 5;
1812     }
1813 
1814   /* Rigister can't be %rsp nor %rbp.  */
1815   if (reg == 4 || reg == 5)
1816     return pc;
1817 
1818   /* The next instruction has to be "andq $-XXX, %rsp".  */
1819   if (buf[offset] != 0x48
1820       || buf[offset + 2] != 0xe4
1821       || (buf[offset + 1] != 0x81 && buf[offset + 1] != 0x83))
1822     return pc;
1823 
1824   offset_and = offset;
1825   offset += buf[offset + 1] == 0x81 ? 7 : 4;
1826 
1827   /* The next instruction has to be "pushq -8(%reg)".  */
1828   r = 0;
1829   if (buf[offset] == 0xff)
1830     offset++;
1831   else if ((buf[offset] & 0xf6) == 0x40
1832 	   && buf[offset + 1] == 0xff)
1833     {
1834       /* Check the REX.B bit.  */
1835       if ((buf[offset] & 0x1) != 0)
1836 	r = 8;
1837       offset += 2;
1838     }
1839   else
1840     return pc;
1841 
1842   /* 8bit -8 is 0xf8.  REG must be binary 110 and MOD must be binary
1843      01.  */
1844   if (buf[offset + 1] != 0xf8
1845       || (buf[offset] & 0xf8) != 0x70)
1846     return pc;
1847 
1848   /* R/M has register.  */
1849   r += buf[offset] & 7;
1850 
1851   /* Registers in leaq and pushq have to be the same.  */
1852   if (reg != r)
1853     return pc;
1854 
1855   if (current_pc > pc + offset_and)
1856     cache->saved_sp_reg = amd64_arch_reg_to_regnum (reg);
1857 
1858   return min (pc + offset + 2, current_pc);
1859 }
1860 
1861 /* Do a limited analysis of the prologue at PC and update CACHE
1862    accordingly.  Bail out early if CURRENT_PC is reached.  Return the
1863    address where the analysis stopped.
1864 
1865    We will handle only functions beginning with:
1866 
1867       pushq %rbp        0x55
1868       movq %rsp, %rbp   0x48 0x89 0xe5
1869 
1870    Any function that doesn't start with this sequence will be assumed
1871    to have no prologue and thus no valid frame pointer in %rbp.  */
1872 
1873 static CORE_ADDR
1874 amd64_analyze_prologue (struct gdbarch *gdbarch,
1875 			CORE_ADDR pc, CORE_ADDR current_pc,
1876 			struct amd64_frame_cache *cache)
1877 {
1878   enum bfd_endian byte_order = gdbarch_byte_order (gdbarch);
1879   static gdb_byte proto[3] = { 0x48, 0x89, 0xe5 }; /* movq %rsp, %rbp */
1880   gdb_byte buf[3];
1881   gdb_byte op;
1882 
1883   if (current_pc <= pc)
1884     return current_pc;
1885 
1886   pc = amd64_analyze_stack_align (pc, current_pc, cache);
1887 
1888   op = read_memory_unsigned_integer (pc, 1, byte_order);
1889 
1890   if (op == 0x55)		/* pushq %rbp */
1891     {
1892       /* Take into account that we've executed the `pushq %rbp' that
1893          starts this instruction sequence.  */
1894       cache->saved_regs[AMD64_RBP_REGNUM] = 0;
1895       cache->sp_offset += 8;
1896 
1897       /* If that's all, return now.  */
1898       if (current_pc <= pc + 1)
1899         return current_pc;
1900 
1901       /* Check for `movq %rsp, %rbp'.  */
1902       read_memory (pc + 1, buf, 3);
1903       if (memcmp (buf, proto, 3) != 0)
1904 	return pc + 1;
1905 
1906       /* OK, we actually have a frame.  */
1907       cache->frameless_p = 0;
1908       return pc + 4;
1909     }
1910 
1911   return pc;
1912 }
1913 
1914 /* Work around false termination of prologue - GCC PR debug/48827.
1915 
1916    START_PC is the first instruction of a function, PC is its minimal already
1917    determined advanced address.  Function returns PC if it has nothing to do.
1918 
1919    84 c0                test   %al,%al
1920    74 23                je     after
1921    <-- here is 0 lines advance - the false prologue end marker.
1922    0f 29 85 70 ff ff ff movaps %xmm0,-0x90(%rbp)
1923    0f 29 4d 80          movaps %xmm1,-0x80(%rbp)
1924    0f 29 55 90          movaps %xmm2,-0x70(%rbp)
1925    0f 29 5d a0          movaps %xmm3,-0x60(%rbp)
1926    0f 29 65 b0          movaps %xmm4,-0x50(%rbp)
1927    0f 29 6d c0          movaps %xmm5,-0x40(%rbp)
1928    0f 29 75 d0          movaps %xmm6,-0x30(%rbp)
1929    0f 29 7d e0          movaps %xmm7,-0x20(%rbp)
1930    after:  */
1931 
1932 static CORE_ADDR
1933 amd64_skip_xmm_prologue (CORE_ADDR pc, CORE_ADDR start_pc)
1934 {
1935   struct symtab_and_line start_pc_sal, next_sal;
1936   gdb_byte buf[4 + 8 * 7];
1937   int offset, xmmreg;
1938 
1939   if (pc == start_pc)
1940     return pc;
1941 
1942   start_pc_sal = find_pc_sect_line (start_pc, NULL, 0);
1943   if (start_pc_sal.symtab == NULL
1944       || producer_is_gcc_ge_4 (start_pc_sal.symtab->producer) < 6
1945       || start_pc_sal.pc != start_pc || pc >= start_pc_sal.end)
1946     return pc;
1947 
1948   next_sal = find_pc_sect_line (start_pc_sal.end, NULL, 0);
1949   if (next_sal.line != start_pc_sal.line)
1950     return pc;
1951 
1952   /* START_PC can be from overlayed memory, ignored here.  */
1953   if (target_read_memory (next_sal.pc - 4, buf, sizeof (buf)) != 0)
1954     return pc;
1955 
1956   /* test %al,%al */
1957   if (buf[0] != 0x84 || buf[1] != 0xc0)
1958     return pc;
1959   /* je AFTER */
1960   if (buf[2] != 0x74)
1961     return pc;
1962 
1963   offset = 4;
1964   for (xmmreg = 0; xmmreg < 8; xmmreg++)
1965     {
1966       /* 0x0f 0x29 0b??000101 movaps %xmmreg?,-0x??(%rbp) */
1967       if (buf[offset] != 0x0f || buf[offset + 1] != 0x29
1968           || (buf[offset + 2] & 0x3f) != (xmmreg << 3 | 0x5))
1969 	return pc;
1970 
1971       /* 0b01?????? */
1972       if ((buf[offset + 2] & 0xc0) == 0x40)
1973 	{
1974 	  /* 8-bit displacement.  */
1975 	  offset += 4;
1976 	}
1977       /* 0b10?????? */
1978       else if ((buf[offset + 2] & 0xc0) == 0x80)
1979 	{
1980 	  /* 32-bit displacement.  */
1981 	  offset += 7;
1982 	}
1983       else
1984 	return pc;
1985     }
1986 
1987   /* je AFTER */
1988   if (offset - 4 != buf[3])
1989     return pc;
1990 
1991   return next_sal.end;
1992 }
1993 
1994 /* Return PC of first real instruction.  */
1995 
1996 static CORE_ADDR
1997 amd64_skip_prologue (struct gdbarch *gdbarch, CORE_ADDR start_pc)
1998 {
1999   struct amd64_frame_cache cache;
2000   CORE_ADDR pc;
2001 
2002   amd64_init_frame_cache (&cache);
2003   pc = amd64_analyze_prologue (gdbarch, start_pc, 0xffffffffffffffffLL,
2004 			       &cache);
2005   if (cache.frameless_p)
2006     return start_pc;
2007 
2008   return amd64_skip_xmm_prologue (pc, start_pc);
2009 }
2010 
2011 
2012 /* Normal frames.  */
2013 
2014 static void
2015 amd64_frame_cache_1 (struct frame_info *this_frame,
2016 		     struct amd64_frame_cache *cache)
2017 {
2018   struct gdbarch *gdbarch = get_frame_arch (this_frame);
2019   enum bfd_endian byte_order = gdbarch_byte_order (gdbarch);
2020   gdb_byte buf[8];
2021   int i;
2022 
2023   cache->pc = get_frame_func (this_frame);
2024   if (cache->pc != 0)
2025     amd64_analyze_prologue (gdbarch, cache->pc, get_frame_pc (this_frame),
2026 			    cache);
2027 
2028   if (cache->frameless_p)
2029     {
2030       /* We didn't find a valid frame.  If we're at the start of a
2031 	 function, or somewhere half-way its prologue, the function's
2032 	 frame probably hasn't been fully setup yet.  Try to
2033 	 reconstruct the base address for the stack frame by looking
2034 	 at the stack pointer.  For truly "frameless" functions this
2035 	 might work too.  */
2036 
2037       if (cache->saved_sp_reg != -1)
2038 	{
2039 	  /* Stack pointer has been saved.  */
2040 	  get_frame_register (this_frame, cache->saved_sp_reg, buf);
2041 	  cache->saved_sp = extract_unsigned_integer (buf, 8, byte_order);
2042 
2043 	  /* We're halfway aligning the stack.  */
2044 	  cache->base = ((cache->saved_sp - 8) & 0xfffffffffffffff0LL) - 8;
2045 	  cache->saved_regs[AMD64_RIP_REGNUM] = cache->saved_sp - 8;
2046 
2047 	  /* This will be added back below.  */
2048 	  cache->saved_regs[AMD64_RIP_REGNUM] -= cache->base;
2049 	}
2050       else
2051 	{
2052 	  get_frame_register (this_frame, AMD64_RSP_REGNUM, buf);
2053 	  cache->base = extract_unsigned_integer (buf, 8, byte_order)
2054 			+ cache->sp_offset;
2055 	}
2056     }
2057   else
2058     {
2059       get_frame_register (this_frame, AMD64_RBP_REGNUM, buf);
2060       cache->base = extract_unsigned_integer (buf, 8, byte_order);
2061     }
2062 
2063   /* Now that we have the base address for the stack frame we can
2064      calculate the value of %rsp in the calling frame.  */
2065   cache->saved_sp = cache->base + 16;
2066 
2067   /* For normal frames, %rip is stored at 8(%rbp).  If we don't have a
2068      frame we find it at the same offset from the reconstructed base
2069      address.  If we're halfway aligning the stack, %rip is handled
2070      differently (see above).  */
2071   if (!cache->frameless_p || cache->saved_sp_reg == -1)
2072     cache->saved_regs[AMD64_RIP_REGNUM] = 8;
2073 
2074   /* Adjust all the saved registers such that they contain addresses
2075      instead of offsets.  */
2076   for (i = 0; i < AMD64_NUM_SAVED_REGS; i++)
2077     if (cache->saved_regs[i] != -1)
2078       cache->saved_regs[i] += cache->base;
2079 
2080   cache->base_p = 1;
2081 }
2082 
2083 static struct amd64_frame_cache *
2084 amd64_frame_cache (struct frame_info *this_frame, void **this_cache)
2085 {
2086   volatile struct gdb_exception ex;
2087   struct amd64_frame_cache *cache;
2088 
2089   if (*this_cache)
2090     return *this_cache;
2091 
2092   cache = amd64_alloc_frame_cache ();
2093   *this_cache = cache;
2094 
2095   TRY_CATCH (ex, RETURN_MASK_ERROR)
2096     {
2097       amd64_frame_cache_1 (this_frame, cache);
2098     }
2099   if (ex.reason < 0 && ex.error != NOT_AVAILABLE_ERROR)
2100     throw_exception (ex);
2101 
2102   return cache;
2103 }
2104 
2105 static enum unwind_stop_reason
2106 amd64_frame_unwind_stop_reason (struct frame_info *this_frame,
2107 				void **this_cache)
2108 {
2109   struct amd64_frame_cache *cache =
2110     amd64_frame_cache (this_frame, this_cache);
2111 
2112   if (!cache->base_p)
2113     return UNWIND_UNAVAILABLE;
2114 
2115   /* This marks the outermost frame.  */
2116   if (cache->base == 0)
2117     return UNWIND_OUTERMOST;
2118 
2119   return UNWIND_NO_REASON;
2120 }
2121 
2122 static void
2123 amd64_frame_this_id (struct frame_info *this_frame, void **this_cache,
2124 		     struct frame_id *this_id)
2125 {
2126   struct amd64_frame_cache *cache =
2127     amd64_frame_cache (this_frame, this_cache);
2128 
2129   if (!cache->base_p)
2130     return;
2131 
2132   /* This marks the outermost frame.  */
2133   if (cache->base == 0)
2134     return;
2135 
2136   (*this_id) = frame_id_build (cache->base + 16, cache->pc);
2137 }
2138 
2139 static struct value *
2140 amd64_frame_prev_register (struct frame_info *this_frame, void **this_cache,
2141 			   int regnum)
2142 {
2143   struct gdbarch *gdbarch = get_frame_arch (this_frame);
2144   struct amd64_frame_cache *cache =
2145     amd64_frame_cache (this_frame, this_cache);
2146 
2147   gdb_assert (regnum >= 0);
2148 
2149   if (regnum == gdbarch_sp_regnum (gdbarch) && cache->saved_sp)
2150     return frame_unwind_got_constant (this_frame, regnum, cache->saved_sp);
2151 
2152   if (regnum < AMD64_NUM_SAVED_REGS && cache->saved_regs[regnum] != -1)
2153     return frame_unwind_got_memory (this_frame, regnum,
2154 				    cache->saved_regs[regnum]);
2155 
2156   return frame_unwind_got_register (this_frame, regnum, regnum);
2157 }
2158 
2159 static const struct frame_unwind amd64_frame_unwind =
2160 {
2161   NORMAL_FRAME,
2162   amd64_frame_unwind_stop_reason,
2163   amd64_frame_this_id,
2164   amd64_frame_prev_register,
2165   NULL,
2166   default_frame_sniffer
2167 };
2168 
2169 /* Generate a bytecode expression to get the value of the saved PC.  */
2170 
2171 static void
2172 amd64_gen_return_address (struct gdbarch *gdbarch,
2173 			  struct agent_expr *ax, struct axs_value *value,
2174 			  CORE_ADDR scope)
2175 {
2176   /* The following sequence assumes the traditional use of the base
2177      register.  */
2178   ax_reg (ax, AMD64_RBP_REGNUM);
2179   ax_const_l (ax, 8);
2180   ax_simple (ax, aop_add);
2181   value->type = register_type (gdbarch, AMD64_RIP_REGNUM);
2182   value->kind = axs_lvalue_memory;
2183 }
2184 
2185 
2186 /* Signal trampolines.  */
2187 
2188 /* FIXME: kettenis/20030419: Perhaps, we can unify the 32-bit and
2189    64-bit variants.  This would require using identical frame caches
2190    on both platforms.  */
2191 
2192 static struct amd64_frame_cache *
2193 amd64_sigtramp_frame_cache (struct frame_info *this_frame, void **this_cache)
2194 {
2195   struct gdbarch *gdbarch = get_frame_arch (this_frame);
2196   struct gdbarch_tdep *tdep = gdbarch_tdep (gdbarch);
2197   enum bfd_endian byte_order = gdbarch_byte_order (gdbarch);
2198   volatile struct gdb_exception ex;
2199   struct amd64_frame_cache *cache;
2200   CORE_ADDR addr;
2201   gdb_byte buf[8];
2202   int i;
2203 
2204   if (*this_cache)
2205     return *this_cache;
2206 
2207   cache = amd64_alloc_frame_cache ();
2208 
2209   TRY_CATCH (ex, RETURN_MASK_ERROR)
2210     {
2211       get_frame_register (this_frame, AMD64_RSP_REGNUM, buf);
2212       cache->base = extract_unsigned_integer (buf, 8, byte_order) - 8;
2213 
2214       addr = tdep->sigcontext_addr (this_frame);
2215       gdb_assert (tdep->sc_reg_offset);
2216       gdb_assert (tdep->sc_num_regs <= AMD64_NUM_SAVED_REGS);
2217       for (i = 0; i < tdep->sc_num_regs; i++)
2218 	if (tdep->sc_reg_offset[i] != -1)
2219 	  cache->saved_regs[i] = addr + tdep->sc_reg_offset[i];
2220 
2221       cache->base_p = 1;
2222     }
2223   if (ex.reason < 0 && ex.error != NOT_AVAILABLE_ERROR)
2224     throw_exception (ex);
2225 
2226   *this_cache = cache;
2227   return cache;
2228 }
2229 
2230 static enum unwind_stop_reason
2231 amd64_sigtramp_frame_unwind_stop_reason (struct frame_info *this_frame,
2232 					 void **this_cache)
2233 {
2234   struct amd64_frame_cache *cache =
2235     amd64_sigtramp_frame_cache (this_frame, this_cache);
2236 
2237   if (!cache->base_p)
2238     return UNWIND_UNAVAILABLE;
2239 
2240   return UNWIND_NO_REASON;
2241 }
2242 
2243 static void
2244 amd64_sigtramp_frame_this_id (struct frame_info *this_frame,
2245 			      void **this_cache, struct frame_id *this_id)
2246 {
2247   struct amd64_frame_cache *cache =
2248     amd64_sigtramp_frame_cache (this_frame, this_cache);
2249 
2250   if (!cache->base_p)
2251     return;
2252 
2253   (*this_id) = frame_id_build (cache->base + 16, get_frame_pc (this_frame));
2254 }
2255 
2256 static struct value *
2257 amd64_sigtramp_frame_prev_register (struct frame_info *this_frame,
2258 				    void **this_cache, int regnum)
2259 {
2260   /* Make sure we've initialized the cache.  */
2261   amd64_sigtramp_frame_cache (this_frame, this_cache);
2262 
2263   return amd64_frame_prev_register (this_frame, this_cache, regnum);
2264 }
2265 
2266 static int
2267 amd64_sigtramp_frame_sniffer (const struct frame_unwind *self,
2268 			      struct frame_info *this_frame,
2269 			      void **this_cache)
2270 {
2271   struct gdbarch_tdep *tdep = gdbarch_tdep (get_frame_arch (this_frame));
2272 
2273   /* We shouldn't even bother if we don't have a sigcontext_addr
2274      handler.  */
2275   if (tdep->sigcontext_addr == NULL)
2276     return 0;
2277 
2278   if (tdep->sigtramp_p != NULL)
2279     {
2280       if (tdep->sigtramp_p (this_frame))
2281 	return 1;
2282     }
2283 
2284   if (tdep->sigtramp_start != 0)
2285     {
2286       CORE_ADDR pc = get_frame_pc (this_frame);
2287 
2288       gdb_assert (tdep->sigtramp_end != 0);
2289       if (pc >= tdep->sigtramp_start && pc < tdep->sigtramp_end)
2290 	return 1;
2291     }
2292 
2293   return 0;
2294 }
2295 
2296 static const struct frame_unwind amd64_sigtramp_frame_unwind =
2297 {
2298   SIGTRAMP_FRAME,
2299   amd64_sigtramp_frame_unwind_stop_reason,
2300   amd64_sigtramp_frame_this_id,
2301   amd64_sigtramp_frame_prev_register,
2302   NULL,
2303   amd64_sigtramp_frame_sniffer
2304 };
2305 
2306 
2307 static CORE_ADDR
2308 amd64_frame_base_address (struct frame_info *this_frame, void **this_cache)
2309 {
2310   struct amd64_frame_cache *cache =
2311     amd64_frame_cache (this_frame, this_cache);
2312 
2313   return cache->base;
2314 }
2315 
2316 static const struct frame_base amd64_frame_base =
2317 {
2318   &amd64_frame_unwind,
2319   amd64_frame_base_address,
2320   amd64_frame_base_address,
2321   amd64_frame_base_address
2322 };
2323 
2324 /* Normal frames, but in a function epilogue.  */
2325 
2326 /* The epilogue is defined here as the 'ret' instruction, which will
2327    follow any instruction such as 'leave' or 'pop %ebp' that destroys
2328    the function's stack frame.  */
2329 
2330 static int
2331 amd64_in_function_epilogue_p (struct gdbarch *gdbarch, CORE_ADDR pc)
2332 {
2333   gdb_byte insn;
2334   struct symtab *symtab;
2335 
2336   symtab = find_pc_symtab (pc);
2337   if (symtab && symtab->epilogue_unwind_valid)
2338     return 0;
2339 
2340   if (target_read_memory (pc, &insn, 1))
2341     return 0;   /* Can't read memory at pc.  */
2342 
2343   if (insn != 0xc3)     /* 'ret' instruction.  */
2344     return 0;
2345 
2346   return 1;
2347 }
2348 
2349 static int
2350 amd64_epilogue_frame_sniffer (const struct frame_unwind *self,
2351 			      struct frame_info *this_frame,
2352 			      void **this_prologue_cache)
2353 {
2354   if (frame_relative_level (this_frame) == 0)
2355     return amd64_in_function_epilogue_p (get_frame_arch (this_frame),
2356 					 get_frame_pc (this_frame));
2357   else
2358     return 0;
2359 }
2360 
2361 static struct amd64_frame_cache *
2362 amd64_epilogue_frame_cache (struct frame_info *this_frame, void **this_cache)
2363 {
2364   struct gdbarch *gdbarch = get_frame_arch (this_frame);
2365   enum bfd_endian byte_order = gdbarch_byte_order (gdbarch);
2366   volatile struct gdb_exception ex;
2367   struct amd64_frame_cache *cache;
2368   gdb_byte buf[8];
2369 
2370   if (*this_cache)
2371     return *this_cache;
2372 
2373   cache = amd64_alloc_frame_cache ();
2374   *this_cache = cache;
2375 
2376   TRY_CATCH (ex, RETURN_MASK_ERROR)
2377     {
2378       /* Cache base will be %esp plus cache->sp_offset (-8).  */
2379       get_frame_register (this_frame, AMD64_RSP_REGNUM, buf);
2380       cache->base = extract_unsigned_integer (buf, 8,
2381 					      byte_order) + cache->sp_offset;
2382 
2383       /* Cache pc will be the frame func.  */
2384       cache->pc = get_frame_pc (this_frame);
2385 
2386       /* The saved %esp will be at cache->base plus 16.  */
2387       cache->saved_sp = cache->base + 16;
2388 
2389       /* The saved %eip will be at cache->base plus 8.  */
2390       cache->saved_regs[AMD64_RIP_REGNUM] = cache->base + 8;
2391 
2392       cache->base_p = 1;
2393     }
2394   if (ex.reason < 0 && ex.error != NOT_AVAILABLE_ERROR)
2395     throw_exception (ex);
2396 
2397   return cache;
2398 }
2399 
2400 static enum unwind_stop_reason
2401 amd64_epilogue_frame_unwind_stop_reason (struct frame_info *this_frame,
2402 					 void **this_cache)
2403 {
2404   struct amd64_frame_cache *cache
2405     = amd64_epilogue_frame_cache (this_frame, this_cache);
2406 
2407   if (!cache->base_p)
2408     return UNWIND_UNAVAILABLE;
2409 
2410   return UNWIND_NO_REASON;
2411 }
2412 
2413 static void
2414 amd64_epilogue_frame_this_id (struct frame_info *this_frame,
2415 			      void **this_cache,
2416 			      struct frame_id *this_id)
2417 {
2418   struct amd64_frame_cache *cache = amd64_epilogue_frame_cache (this_frame,
2419 							       this_cache);
2420 
2421   if (!cache->base_p)
2422     return;
2423 
2424   (*this_id) = frame_id_build (cache->base + 8, cache->pc);
2425 }
2426 
2427 static const struct frame_unwind amd64_epilogue_frame_unwind =
2428 {
2429   NORMAL_FRAME,
2430   amd64_epilogue_frame_unwind_stop_reason,
2431   amd64_epilogue_frame_this_id,
2432   amd64_frame_prev_register,
2433   NULL,
2434   amd64_epilogue_frame_sniffer
2435 };
2436 
2437 static struct frame_id
2438 amd64_dummy_id (struct gdbarch *gdbarch, struct frame_info *this_frame)
2439 {
2440   CORE_ADDR fp;
2441 
2442   fp = get_frame_register_unsigned (this_frame, AMD64_RBP_REGNUM);
2443 
2444   return frame_id_build (fp + 16, get_frame_pc (this_frame));
2445 }
2446 
2447 /* 16 byte align the SP per frame requirements.  */
2448 
2449 static CORE_ADDR
2450 amd64_frame_align (struct gdbarch *gdbarch, CORE_ADDR sp)
2451 {
2452   return sp & -(CORE_ADDR)16;
2453 }
2454 
2455 
2456 /* Supply register REGNUM from the buffer specified by FPREGS and LEN
2457    in the floating-point register set REGSET to register cache
2458    REGCACHE.  If REGNUM is -1, do this for all registers in REGSET.  */
2459 
2460 static void
2461 amd64_supply_fpregset (const struct regset *regset, struct regcache *regcache,
2462 		       int regnum, const void *fpregs, size_t len)
2463 {
2464   const struct gdbarch_tdep *tdep = gdbarch_tdep (regset->arch);
2465 
2466   gdb_assert (len == tdep->sizeof_fpregset);
2467   amd64_supply_fxsave (regcache, regnum, fpregs);
2468 }
2469 
2470 /* Collect register REGNUM from the register cache REGCACHE and store
2471    it in the buffer specified by FPREGS and LEN as described by the
2472    floating-point register set REGSET.  If REGNUM is -1, do this for
2473    all registers in REGSET.  */
2474 
2475 static void
2476 amd64_collect_fpregset (const struct regset *regset,
2477 			const struct regcache *regcache,
2478 			int regnum, void *fpregs, size_t len)
2479 {
2480   const struct gdbarch_tdep *tdep = gdbarch_tdep (regset->arch);
2481 
2482   gdb_assert (len == tdep->sizeof_fpregset);
2483   amd64_collect_fxsave (regcache, regnum, fpregs);
2484 }
2485 
2486 /* Similar to amd64_supply_fpregset, but use XSAVE extended state.  */
2487 
2488 static void
2489 amd64_supply_xstateregset (const struct regset *regset,
2490 			   struct regcache *regcache, int regnum,
2491 			   const void *xstateregs, size_t len)
2492 {
2493   amd64_supply_xsave (regcache, regnum, xstateregs);
2494 }
2495 
2496 /* Similar to amd64_collect_fpregset, but use XSAVE extended state.  */
2497 
2498 static void
2499 amd64_collect_xstateregset (const struct regset *regset,
2500 			    const struct regcache *regcache,
2501 			    int regnum, void *xstateregs, size_t len)
2502 {
2503   amd64_collect_xsave (regcache, regnum, xstateregs, 1);
2504 }
2505 
2506 /* Return the appropriate register set for the core section identified
2507    by SECT_NAME and SECT_SIZE.  */
2508 
2509 static const struct regset *
2510 amd64_regset_from_core_section (struct gdbarch *gdbarch,
2511 				const char *sect_name, size_t sect_size)
2512 {
2513   struct gdbarch_tdep *tdep = gdbarch_tdep (gdbarch);
2514 
2515   if (strcmp (sect_name, ".reg2") == 0 && sect_size == tdep->sizeof_fpregset)
2516     {
2517       if (tdep->fpregset == NULL)
2518 	tdep->fpregset = regset_alloc (gdbarch, amd64_supply_fpregset,
2519 				       amd64_collect_fpregset);
2520 
2521       return tdep->fpregset;
2522     }
2523 
2524   if (strcmp (sect_name, ".reg-xstate") == 0)
2525     {
2526       if (tdep->xstateregset == NULL)
2527 	tdep->xstateregset = regset_alloc (gdbarch,
2528 					   amd64_supply_xstateregset,
2529 					   amd64_collect_xstateregset);
2530 
2531       return tdep->xstateregset;
2532     }
2533 
2534   return i386_regset_from_core_section (gdbarch, sect_name, sect_size);
2535 }
2536 
2537 
2538 /* Figure out where the longjmp will land.  Slurp the jmp_buf out of
2539    %rdi.  We expect its value to be a pointer to the jmp_buf structure
2540    from which we extract the address that we will land at.  This
2541    address is copied into PC.  This routine returns non-zero on
2542    success.  */
2543 
2544 static int
2545 amd64_get_longjmp_target (struct frame_info *frame, CORE_ADDR *pc)
2546 {
2547   gdb_byte buf[8];
2548   CORE_ADDR jb_addr;
2549   struct gdbarch *gdbarch = get_frame_arch (frame);
2550   int jb_pc_offset = gdbarch_tdep (gdbarch)->jb_pc_offset;
2551   int len = TYPE_LENGTH (builtin_type (gdbarch)->builtin_func_ptr);
2552 
2553   /* If JB_PC_OFFSET is -1, we have no way to find out where the
2554      longjmp will land.	 */
2555   if (jb_pc_offset == -1)
2556     return 0;
2557 
2558   get_frame_register (frame, AMD64_RDI_REGNUM, buf);
2559   jb_addr= extract_typed_address
2560 	    (buf, builtin_type (gdbarch)->builtin_data_ptr);
2561   if (target_read_memory (jb_addr + jb_pc_offset, buf, len))
2562     return 0;
2563 
2564   *pc = extract_typed_address (buf, builtin_type (gdbarch)->builtin_func_ptr);
2565 
2566   return 1;
2567 }
2568 
2569 static const int amd64_record_regmap[] =
2570 {
2571   AMD64_RAX_REGNUM, AMD64_RCX_REGNUM, AMD64_RDX_REGNUM, AMD64_RBX_REGNUM,
2572   AMD64_RSP_REGNUM, AMD64_RBP_REGNUM, AMD64_RSI_REGNUM, AMD64_RDI_REGNUM,
2573   AMD64_R8_REGNUM, AMD64_R9_REGNUM, AMD64_R10_REGNUM, AMD64_R11_REGNUM,
2574   AMD64_R12_REGNUM, AMD64_R13_REGNUM, AMD64_R14_REGNUM, AMD64_R15_REGNUM,
2575   AMD64_RIP_REGNUM, AMD64_EFLAGS_REGNUM, AMD64_CS_REGNUM, AMD64_SS_REGNUM,
2576   AMD64_DS_REGNUM, AMD64_ES_REGNUM, AMD64_FS_REGNUM, AMD64_GS_REGNUM
2577 };
2578 
2579 void
2580 amd64_init_abi (struct gdbarch_info info, struct gdbarch *gdbarch)
2581 {
2582   struct gdbarch_tdep *tdep = gdbarch_tdep (gdbarch);
2583   const struct target_desc *tdesc = info.target_desc;
2584 
2585   /* AMD64 generally uses `fxsave' instead of `fsave' for saving its
2586      floating-point registers.  */
2587   tdep->sizeof_fpregset = I387_SIZEOF_FXSAVE;
2588 
2589   if (! tdesc_has_registers (tdesc))
2590     tdesc = tdesc_amd64;
2591   tdep->tdesc = tdesc;
2592 
2593   tdep->num_core_regs = AMD64_NUM_GREGS + I387_NUM_REGS;
2594   tdep->register_names = amd64_register_names;
2595 
2596   if (tdesc_find_feature (tdesc, "org.gnu.gdb.i386.avx") != NULL)
2597     {
2598       tdep->ymmh_register_names = amd64_ymmh_names;
2599       tdep->num_ymm_regs = 16;
2600       tdep->ymm0h_regnum = AMD64_YMM0H_REGNUM;
2601     }
2602 
2603   tdep->num_byte_regs = 20;
2604   tdep->num_word_regs = 16;
2605   tdep->num_dword_regs = 16;
2606   /* Avoid wiring in the MMX registers for now.  */
2607   tdep->num_mmx_regs = 0;
2608 
2609   set_gdbarch_pseudo_register_read_value (gdbarch,
2610 					  amd64_pseudo_register_read_value);
2611   set_gdbarch_pseudo_register_write (gdbarch,
2612 				     amd64_pseudo_register_write);
2613 
2614   set_tdesc_pseudo_register_name (gdbarch, amd64_pseudo_register_name);
2615 
2616   /* AMD64 has an FPU and 16 SSE registers.  */
2617   tdep->st0_regnum = AMD64_ST0_REGNUM;
2618   tdep->num_xmm_regs = 16;
2619 
2620   /* This is what all the fuss is about.  */
2621   set_gdbarch_long_bit (gdbarch, 64);
2622   set_gdbarch_long_long_bit (gdbarch, 64);
2623   set_gdbarch_ptr_bit (gdbarch, 64);
2624 
2625   /* In contrast to the i386, on AMD64 a `long double' actually takes
2626      up 128 bits, even though it's still based on the i387 extended
2627      floating-point format which has only 80 significant bits.  */
2628   set_gdbarch_long_double_bit (gdbarch, 128);
2629 
2630   set_gdbarch_num_regs (gdbarch, AMD64_NUM_REGS);
2631 
2632   /* Register numbers of various important registers.  */
2633   set_gdbarch_sp_regnum (gdbarch, AMD64_RSP_REGNUM); /* %rsp */
2634   set_gdbarch_pc_regnum (gdbarch, AMD64_RIP_REGNUM); /* %rip */
2635   set_gdbarch_ps_regnum (gdbarch, AMD64_EFLAGS_REGNUM); /* %eflags */
2636   set_gdbarch_fp0_regnum (gdbarch, AMD64_ST0_REGNUM); /* %st(0) */
2637 
2638   /* The "default" register numbering scheme for AMD64 is referred to
2639      as the "DWARF Register Number Mapping" in the System V psABI.
2640      The preferred debugging format for all known AMD64 targets is
2641      actually DWARF2, and GCC doesn't seem to support DWARF (that is
2642      DWARF-1), but we provide the same mapping just in case.  This
2643      mapping is also used for stabs, which GCC does support.  */
2644   set_gdbarch_stab_reg_to_regnum (gdbarch, amd64_dwarf_reg_to_regnum);
2645   set_gdbarch_dwarf2_reg_to_regnum (gdbarch, amd64_dwarf_reg_to_regnum);
2646 
2647   /* We don't override SDB_REG_RO_REGNUM, since COFF doesn't seem to
2648      be in use on any of the supported AMD64 targets.  */
2649 
2650   /* Call dummy code.  */
2651   set_gdbarch_push_dummy_call (gdbarch, amd64_push_dummy_call);
2652   set_gdbarch_frame_align (gdbarch, amd64_frame_align);
2653   set_gdbarch_frame_red_zone_size (gdbarch, 128);
2654   tdep->call_dummy_num_integer_regs =
2655     ARRAY_SIZE (amd64_dummy_call_integer_regs);
2656   tdep->call_dummy_integer_regs = amd64_dummy_call_integer_regs;
2657   tdep->classify = amd64_classify;
2658 
2659   set_gdbarch_convert_register_p (gdbarch, i387_convert_register_p);
2660   set_gdbarch_register_to_value (gdbarch, i387_register_to_value);
2661   set_gdbarch_value_to_register (gdbarch, i387_value_to_register);
2662 
2663   set_gdbarch_return_value (gdbarch, amd64_return_value);
2664 
2665   set_gdbarch_skip_prologue (gdbarch, amd64_skip_prologue);
2666 
2667   tdep->record_regmap = amd64_record_regmap;
2668 
2669   set_gdbarch_dummy_id (gdbarch, amd64_dummy_id);
2670 
2671   /* Hook the function epilogue frame unwinder.  This unwinder is
2672      appended to the list first, so that it supercedes the other
2673      unwinders in function epilogues.  */
2674   frame_unwind_prepend_unwinder (gdbarch, &amd64_epilogue_frame_unwind);
2675 
2676   /* Hook the prologue-based frame unwinders.  */
2677   frame_unwind_append_unwinder (gdbarch, &amd64_sigtramp_frame_unwind);
2678   frame_unwind_append_unwinder (gdbarch, &amd64_frame_unwind);
2679   frame_base_set_default (gdbarch, &amd64_frame_base);
2680 
2681   /* If we have a register mapping, enable the generic core file support.  */
2682   if (tdep->gregset_reg_offset)
2683     set_gdbarch_regset_from_core_section (gdbarch,
2684 					  amd64_regset_from_core_section);
2685 
2686   set_gdbarch_get_longjmp_target (gdbarch, amd64_get_longjmp_target);
2687 
2688   set_gdbarch_relocate_instruction (gdbarch, amd64_relocate_instruction);
2689 
2690   set_gdbarch_gen_return_address (gdbarch, amd64_gen_return_address);
2691 }
2692 
2693 /* Provide a prototype to silence -Wmissing-prototypes.  */
2694 void _initialize_amd64_tdep (void);
2695 
2696 void
2697 _initialize_amd64_tdep (void)
2698 {
2699   initialize_tdesc_amd64 ();
2700   initialize_tdesc_amd64_avx ();
2701 }
2702 
2703 
2704 /* The 64-bit FXSAVE format differs from the 32-bit format in the
2705    sense that the instruction pointer and data pointer are simply
2706    64-bit offsets into the code segment and the data segment instead
2707    of a selector offset pair.  The functions below store the upper 32
2708    bits of these pointers (instead of just the 16-bits of the segment
2709    selector).  */
2710 
2711 /* Fill register REGNUM in REGCACHE with the appropriate
2712    floating-point or SSE register value from *FXSAVE.  If REGNUM is
2713    -1, do this for all registers.  This function masks off any of the
2714    reserved bits in *FXSAVE.  */
2715 
2716 void
2717 amd64_supply_fxsave (struct regcache *regcache, int regnum,
2718 		     const void *fxsave)
2719 {
2720   struct gdbarch *gdbarch = get_regcache_arch (regcache);
2721   struct gdbarch_tdep *tdep = gdbarch_tdep (gdbarch);
2722 
2723   i387_supply_fxsave (regcache, regnum, fxsave);
2724 
2725   if (fxsave && gdbarch_ptr_bit (gdbarch) == 64)
2726     {
2727       const gdb_byte *regs = fxsave;
2728 
2729       if (regnum == -1 || regnum == I387_FISEG_REGNUM (tdep))
2730 	regcache_raw_supply (regcache, I387_FISEG_REGNUM (tdep), regs + 12);
2731       if (regnum == -1 || regnum == I387_FOSEG_REGNUM (tdep))
2732 	regcache_raw_supply (regcache, I387_FOSEG_REGNUM (tdep), regs + 20);
2733     }
2734 }
2735 
2736 /* Similar to amd64_supply_fxsave, but use XSAVE extended state.  */
2737 
2738 void
2739 amd64_supply_xsave (struct regcache *regcache, int regnum,
2740 		    const void *xsave)
2741 {
2742   struct gdbarch *gdbarch = get_regcache_arch (regcache);
2743   struct gdbarch_tdep *tdep = gdbarch_tdep (gdbarch);
2744 
2745   i387_supply_xsave (regcache, regnum, xsave);
2746 
2747   if (xsave && gdbarch_ptr_bit (gdbarch) == 64)
2748     {
2749       const gdb_byte *regs = xsave;
2750 
2751       if (regnum == -1 || regnum == I387_FISEG_REGNUM (tdep))
2752 	regcache_raw_supply (regcache, I387_FISEG_REGNUM (tdep),
2753 			     regs + 12);
2754       if (regnum == -1 || regnum == I387_FOSEG_REGNUM (tdep))
2755 	regcache_raw_supply (regcache, I387_FOSEG_REGNUM (tdep),
2756 			     regs + 20);
2757     }
2758 }
2759 
2760 /* Fill register REGNUM (if it is a floating-point or SSE register) in
2761    *FXSAVE with the value from REGCACHE.  If REGNUM is -1, do this for
2762    all registers.  This function doesn't touch any of the reserved
2763    bits in *FXSAVE.  */
2764 
2765 void
2766 amd64_collect_fxsave (const struct regcache *regcache, int regnum,
2767 		      void *fxsave)
2768 {
2769   struct gdbarch *gdbarch = get_regcache_arch (regcache);
2770   struct gdbarch_tdep *tdep = gdbarch_tdep (gdbarch);
2771   gdb_byte *regs = fxsave;
2772 
2773   i387_collect_fxsave (regcache, regnum, fxsave);
2774 
2775   if (gdbarch_ptr_bit (gdbarch) == 64)
2776     {
2777       if (regnum == -1 || regnum == I387_FISEG_REGNUM (tdep))
2778 	regcache_raw_collect (regcache, I387_FISEG_REGNUM (tdep), regs + 12);
2779       if (regnum == -1 || regnum == I387_FOSEG_REGNUM (tdep))
2780 	regcache_raw_collect (regcache, I387_FOSEG_REGNUM (tdep), regs + 20);
2781     }
2782 }
2783 
2784 /* Similar to amd64_collect_fxsave, but use XSAVE extended state.  */
2785 
2786 void
2787 amd64_collect_xsave (const struct regcache *regcache, int regnum,
2788 		     void *xsave, int gcore)
2789 {
2790   struct gdbarch *gdbarch = get_regcache_arch (regcache);
2791   struct gdbarch_tdep *tdep = gdbarch_tdep (gdbarch);
2792   gdb_byte *regs = xsave;
2793 
2794   i387_collect_xsave (regcache, regnum, xsave, gcore);
2795 
2796   if (gdbarch_ptr_bit (gdbarch) == 64)
2797     {
2798       if (regnum == -1 || regnum == I387_FISEG_REGNUM (tdep))
2799 	regcache_raw_collect (regcache, I387_FISEG_REGNUM (tdep),
2800 			      regs + 12);
2801       if (regnum == -1 || regnum == I387_FOSEG_REGNUM (tdep))
2802 	regcache_raw_collect (regcache, I387_FOSEG_REGNUM (tdep),
2803 			      regs + 20);
2804     }
2805 }
2806