/* Intel 80386 machine specific gas.
Written by Eliot Dresselhaus (eliot@mgm.mit.edu).
+ x86_64 support by Jan Hubicka (jh@suse.cz)
Bugs & suggestions are completely welcome. This is free software.
Please help us make it better. */
#include "as.h"
#include "subsegs.h"
+#include "dwarf2dbg.h"
#include "opcode/i386.h"
#ifndef REGISTER_WARNINGS
static int fits_in_unsigned_byte PARAMS ((offsetT));
static int fits_in_unsigned_word PARAMS ((offsetT));
static int fits_in_signed_word PARAMS ((offsetT));
+static int fits_in_unsigned_long PARAMS ((offsetT));
+static int fits_in_signed_long PARAMS ((offsetT));
static int smallest_imm_type PARAMS ((offsetT));
static offsetT offset_in_range PARAMS ((offsetT, int));
static int add_prefix PARAMS ((unsigned int));
-static void set_16bit_code_flag PARAMS ((int));
+static void set_code_flag PARAMS ((int));
static void set_16bit_gcc_code_flag PARAMS ((int));
static void set_intel_syntax PARAMS ((int));
static void set_cpu_arch PARAMS ((int));
#ifdef BFD_ASSEMBLER
static bfd_reloc_code_real_type reloc
- PARAMS ((int, int, bfd_reloc_code_real_type));
+ PARAMS ((int, int, int, bfd_reloc_code_real_type));
#endif
+#ifndef DEFAULT_ARCH
+#define DEFAULT_ARCH "i386"
+#endif
+static char *default_arch = DEFAULT_ARCH;
+
/* 'md_assemble ()' gathers together information and puts it into a
i386_insn. */
operand. */
union i386_op op[MAX_OPERANDS];
+ /* Flags for operands. */
+ unsigned int flags[MAX_OPERANDS];
+#define Operand_PCrel 1
+
/* Relocation type for operand */
#ifdef BFD_ASSEMBLER
enum bfd_reloc_code_real disp_reloc[MAX_OPERANDS];
/* SEG gives the seg_entries of this insn. They are zero unless
explicit segment overrides are given. */
- const seg_entry *seg[2]; /* segments for memory operands (if given) */
+ const seg_entry *seg[2];
/* PREFIX holds all the given prefix opcodes (usually null).
PREFIXES is the number of prefix opcodes. */
addressing modes of this insn are encoded. */
modrm_byte rm;
+ rex_byte rex;
sib_byte sib;
};
#endif
/* This array holds the chars that always start a comment. If the
- pre-processor is disabled, these aren't very useful */
+ pre-processor is disabled, these aren't very useful. */
#if defined (TE_I386AIX) || ((defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)) && ! defined (TE_LINUX) && !defined(TE_FreeBSD))
/* Putting '/' here makes it impossible to use the divide operator.
However, we need it for compatibility with SVR4 systems. */
/* This array holds the chars that only start a comment at the beginning of
a line. If the line seems to have the form '# 123 filename'
- .line and .file directives will appear in the pre-processed output */
-/* Note that input_file.c hand checks for '#' at the beginning of the
+ .line and .file directives will appear in the pre-processed output.
+ Note that input_file.c hand checks for '#' at the beginning of the
first line of the input file. This is because the compiler outputs
- #NO_APP at the beginning of its output. */
-/* Also note that comments started like this one will always work if
+ #NO_APP at the beginning of its output.
+ Also note that comments started like this one will always work if
'/' isn't otherwise defined. */
#if defined (TE_I386AIX) || ((defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)) && ! defined (TE_LINUX) && !defined(TE_FreeBSD))
const char line_comment_chars[] = "";
const char line_separator_chars[] = ";";
-/* Chars that can be used to separate mant from exp in floating point nums */
+/* Chars that can be used to separate mant from exp in floating point
+ nums. */
const char EXP_CHARS[] = "eE";
-/* Chars that mean this number is a floating point constant */
-/* As in 0f12.456 */
-/* or 0d1.2345e12 */
+/* Chars that mean this number is a floating point constant
+ As in 0f12.456
+ or 0d1.2345e12. */
const char FLT_CHARS[] = "fFdDxX";
-/* tables for lexical analysis */
+/* Tables for lexical analysis. */
static char mnemonic_chars[256];
static char register_chars[256];
static char operand_chars[256];
static char identifier_chars[256];
static char digit_chars[256];
-/* lexical macros */
+/* Lexical macros. */
#define is_mnemonic_char(x) (mnemonic_chars[(unsigned char) x])
#define is_operand_char(x) (operand_chars[(unsigned char) x])
#define is_register_char(x) (register_chars[(unsigned char) x])
#define is_identifier_char(x) (identifier_chars[(unsigned char) x])
#define is_digit_char(x) (digit_chars[(unsigned char) x])
-/* put here all non-digit non-letter charcters that may occur in an operand */
+/* All non-digit non-letter charcters that may occur in an operand. */
static char operand_special_chars[] = "%$-+(,)*._~/<>|&^!:[@]";
/* md_assemble() always leaves the strings it's passed unaltered. To
with '\0's (indicating end of strings for various sub-fields of the
assembler instruction). */
static char save_stack[32];
-static char *save_stack_p; /* stack pointer */
+static char *save_stack_p;
#define END_STRING_AND_SAVE(s) \
do { *save_stack_p++ = *(s); *(s) = '\0'; } while (0)
#define RESTORE_END_STRING(s) \
/* Current operand we are working on. */
static int this_operand;
-/* 1 if we're writing 16-bit code,
- 0 if 32-bit. */
-static int flag_16bit_code;
+/* We support four different modes. FLAG_CODE variable is used to distinguish
+ these. */
+
+enum flag_code {
+ CODE_32BIT,
+ CODE_16BIT,
+ CODE_64BIT };
+
+static enum flag_code flag_code;
+static int use_rela_relocations = 0;
+
+/* The names used to print error messages. */
+static const char *flag_code_names[] =
+ {
+ "32",
+ "16",
+ "64"
+ };
/* 1 for intel syntax,
0 if att syntax. */
static const char *cpu_arch_name = NULL;
/* CPU feature flags. */
-static unsigned int cpu_arch_flags = 0;
+static unsigned int cpu_arch_flags = CpuUnknownFlags|CpuNo64;
/* Interface to relax_segment.
There are 2 relax states for 386 jump insns: one for conditional &
sort of jump to choose to reach a given label. */
/* Types. */
-#define COND_JUMP 1 /* Conditional jump. */
-#define UNCOND_JUMP 2 /* Unconditional jump. */
+#define COND_JUMP 1
+#define UNCOND_JUMP 2
/* Sizes. */
#define CODE16 1
#define SMALL 0
#endif
#define ENCODE_RELAX_STATE(type,size) \
- ((relax_substateT)((type<<2) | (size)))
+ ((relax_substateT) ((type<<2) | (size)))
#define SIZE_FROM_RELAX_STATE(s) \
( (((s) & 0x3) == BIG ? 4 : (((s) & 0x3) == BIG16 ? 2 : 1)) )
1) most positive reach of this state,
2) most negative reach of this state,
3) how many bytes this mode will add to the size of the current frag
- 4) which index into the table to try if we can't fit into this one.
- */
+ 4) which index into the table to try if we can't fit into this one. */
{1, 1, 0, 0},
{1, 1, 0, 0},
{1, 1, 0, 0},
{"i686", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuSSE },
{"pentium", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuMMX },
{"pentiumpro",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|CpuSSE },
- {"k6", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuMMX|Cpu3dnow },
- {"athlon", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuMMX|Cpu3dnow },
+ {"k6", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX|Cpu3dnow },
+ {"athlon", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuMMX|Cpu3dnow },
+ {"sledgehammer",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuSledgehammer|CpuMMX|Cpu3dnow|CpuSSE },
{NULL, 0 }
};
fragS *fragP;
int count;
{
- /* Various efficient no-op patterns for aligning code labels. */
- /* Note: Don't try to assemble the instructions in the comments. */
- /* 0L and 0w are not legal */
+ /* Various efficient no-op patterns for aligning code labels.
+ Note: Don't try to assemble the instructions in the comments.
+ 0L and 0w are not legal. */
static const char f32_1[] =
{0x90}; /* nop */
static const char f32_2[] =
f32_15, f32_15, f32_15, f32_15, f32_15, f32_15, f32_15
};
+ /* ??? We can't use these fillers for x86_64, since they often kills the
+ upper halves. Solve later. */
+ if (flag_code == CODE_64BIT)
+ count = 1;
+
if (count > 0 && count <= 15)
{
- if (flag_16bit_code)
+ if (flag_code == CODE_16BIT)
{
memcpy (fragP->fr_literal + fragP->fr_fix,
f16_patt[count - 1], count);
static void s_bss PARAMS ((int));
#endif
-symbolS *GOT_symbol; /* Pre-defined "_GLOBAL_OFFSET_TABLE_" */
+symbolS *GOT_symbol; /* Pre-defined "_GLOBAL_OFFSET_TABLE_". */
static INLINE unsigned int
mode_from_disp_size (t)
unsigned int t;
{
- return (t & Disp8) ? 1 : (t & (Disp16 | Disp32)) ? 2 : 0;
+ return (t & Disp8) ? 1 : (t & (Disp16 | Disp32 | Disp32S)) ? 2 : 0;
}
static INLINE int
{
return (-32768 <= num) && (num <= 32767);
}
+static INLINE int
+fits_in_signed_long (num)
+ offsetT num ATTRIBUTE_UNUSED;
+{
+#ifndef BFD64
+ return 1;
+#else
+ return (!(((offsetT) -1 << 31) & num)
+ || (((offsetT) -1 << 31) & num) == ((offsetT) -1 << 31));
+#endif
+} /* fits_in_signed_long() */
+static INLINE int
+fits_in_unsigned_long (num)
+ offsetT num ATTRIBUTE_UNUSED;
+{
+#ifndef BFD64
+ return 1;
+#else
+ return (num & (((offsetT) 2 << 31) - 1)) == num;
+#endif
+} /* fits_in_unsigned_long() */
static int
smallest_imm_type (num)
offsetT num;
{
- if (cpu_arch_flags != 0
- && cpu_arch_flags != (Cpu086 | Cpu186 | Cpu286 | Cpu386 | Cpu486))
+ if (cpu_arch_flags != (Cpu086 | Cpu186 | Cpu286 | Cpu386 | Cpu486 | CpuNo64)
+ && !(cpu_arch_flags & (CpuUnknown)))
{
/* This code is disabled on the 486 because all the Imm1 forms
in the opcode table are slower on the i486. They're the
displacement, which has another syntax if you really want to
use that form. */
if (num == 1)
- return Imm1 | Imm8 | Imm8S | Imm16 | Imm32;
+ return Imm1 | Imm8 | Imm8S | Imm16 | Imm32 | Imm32S | Imm64;
}
return (fits_in_signed_byte (num)
- ? (Imm8S | Imm8 | Imm16 | Imm32)
+ ? (Imm8S | Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
: fits_in_unsigned_byte (num)
- ? (Imm8 | Imm16 | Imm32)
+ ? (Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
: (fits_in_signed_word (num) || fits_in_unsigned_word (num))
- ? (Imm16 | Imm32)
- : (Imm32));
+ ? (Imm16 | Imm32 | Imm32S | Imm64)
+ : fits_in_signed_long (num)
+ ? (Imm32 | Imm32S | Imm64)
+ : fits_in_unsigned_long (num)
+ ? (Imm32 | Imm64)
+ : Imm64);
}
static offsetT
case 1: mask = ((addressT) 1 << 8) - 1; break;
case 2: mask = ((addressT) 1 << 16) - 1; break;
case 4: mask = ((addressT) 2 << 31) - 1; break;
+#ifdef BFD64
+ case 8: mask = ((addressT) 2 << 63) - 1; break;
+#endif
default: abort ();
}
/* If BFD64, sign extend val. */
- if ((val & ~(((addressT) 2 << 31) - 1)) == 0)
- val = (val ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
+ if (!use_rela_relocations)
+ if ((val & ~(((addressT) 2 << 31) - 1)) == 0)
+ val = (val ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
if ((val & ~mask) != 0 && (val & ~mask) != ~mask)
{
int ret = 1;
int q;
- switch (prefix)
- {
- default:
- abort ();
-
- case CS_PREFIX_OPCODE:
- case DS_PREFIX_OPCODE:
- case ES_PREFIX_OPCODE:
- case FS_PREFIX_OPCODE:
- case GS_PREFIX_OPCODE:
- case SS_PREFIX_OPCODE:
- q = SEG_PREFIX;
- break;
+ if (prefix >= 0x40 && prefix < 0x50 && flag_code == CODE_64BIT)
+ q = REX_PREFIX;
+ else
+ switch (prefix)
+ {
+ default:
+ abort ();
+
+ case CS_PREFIX_OPCODE:
+ case DS_PREFIX_OPCODE:
+ case ES_PREFIX_OPCODE:
+ case FS_PREFIX_OPCODE:
+ case GS_PREFIX_OPCODE:
+ case SS_PREFIX_OPCODE:
+ q = SEG_PREFIX;
+ break;
- case REPNE_PREFIX_OPCODE:
- case REPE_PREFIX_OPCODE:
- ret = 2;
- /* fall thru */
- case LOCK_PREFIX_OPCODE:
- q = LOCKREP_PREFIX;
- break;
+ case REPNE_PREFIX_OPCODE:
+ case REPE_PREFIX_OPCODE:
+ ret = 2;
+ /* fall thru */
+ case LOCK_PREFIX_OPCODE:
+ q = LOCKREP_PREFIX;
+ break;
- case FWAIT_OPCODE:
- q = WAIT_PREFIX;
- break;
+ case FWAIT_OPCODE:
+ q = WAIT_PREFIX;
+ break;
- case ADDR_PREFIX_OPCODE:
- q = ADDR_PREFIX;
- break;
+ case ADDR_PREFIX_OPCODE:
+ q = ADDR_PREFIX;
+ break;
- case DATA_PREFIX_OPCODE:
- q = DATA_PREFIX;
- break;
- }
+ case DATA_PREFIX_OPCODE:
+ q = DATA_PREFIX;
+ break;
+ }
if (i.prefix[q])
{
}
static void
-set_16bit_code_flag (new_16bit_code_flag)
- int new_16bit_code_flag;
+set_code_flag (value)
+ int value;
{
- flag_16bit_code = new_16bit_code_flag;
+ flag_code = value;
+ cpu_arch_flags &= ~(Cpu64 | CpuNo64);
+ cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
+ if (value == CODE_64BIT && !(cpu_arch_flags & CpuSledgehammer))
+ {
+ as_bad (_("64bit mode not supported on this CPU."));
+ }
+ if (value == CODE_32BIT && !(cpu_arch_flags & Cpu386))
+ {
+ as_bad (_("32bit mode not supported on this CPU."));
+ }
stackop_size = '\0';
}
static void
-set_16bit_gcc_code_flag (new_16bit_code_flag)
- int new_16bit_code_flag;
+set_16bit_gcc_code_flag (new_code_flag)
+ int new_code_flag;
{
- flag_16bit_code = new_16bit_code_flag;
- stackop_size = new_16bit_code_flag ? 'l' : '\0';
+ flag_code = new_code_flag;
+ cpu_arch_flags &= ~(Cpu64 | CpuNo64);
+ cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
+ stackop_size = 'l';
}
static void
if (strcmp (string, cpu_arch[i].name) == 0)
{
cpu_arch_name = cpu_arch[i].name;
- cpu_arch_flags = cpu_arch[i].flags;
+ cpu_arch_flags = cpu_arch[i].flags | (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
break;
}
}
{"value", cons, 2},
{"noopt", s_ignore, 0},
{"optim", s_ignore, 0},
- {"code16gcc", set_16bit_gcc_code_flag, 1},
- {"code16", set_16bit_code_flag, 1},
- {"code32", set_16bit_code_flag, 0},
+ {"code16gcc", set_16bit_gcc_code_flag, CODE_16BIT},
+ {"code16", set_code_flag, CODE_16BIT},
+ {"code32", set_code_flag, CODE_32BIT},
+ {"code64", set_code_flag, CODE_64BIT},
{"intel_syntax", set_intel_syntax, 1},
{"att_syntax", set_intel_syntax, 0},
+ {"file", dwarf2_directive_file, 0},
+ {"loc", dwarf2_directive_loc, 0},
{0, 0, 0}
};
(PTR) core_optab);
if (hash_err)
{
- hash_error:
as_fatal (_("Internal Error: Can't hash %s: %s"),
(optab - 1)->name,
hash_err);
{
hash_err = hash_insert (reg_hash, regtab->reg_name, (PTR) regtab);
if (hash_err)
- goto hash_error;
+ as_fatal (_("Internal Error: Can't hash %s: %s"),
+ regtab->reg_name,
+ hash_err);
}
}
\f
#ifdef DEBUG386
-/* debugging routines for md_assemble */
+/* Debugging routines for md_assemble. */
static void pi PARAMS ((char *, i386_insn *));
static void pte PARAMS ((template *));
static void pt PARAMS ((unsigned int));
return 1;
return 0;
#else
- /* For COFF */
+ /* For COFF. */
return fixp->fx_r_type == 7;
#endif
}
#ifdef BFD_ASSEMBLER
-static bfd_reloc_code_real_type reloc
- PARAMS ((int, int, bfd_reloc_code_real_type));
static bfd_reloc_code_real_type
-reloc (size, pcrel, other)
+reloc (size, pcrel, sign, other)
int size;
int pcrel;
+ int sign;
bfd_reloc_code_real_type other;
{
if (other != NO_RELOC)
if (pcrel)
{
+ if (!sign)
+ as_bad(_("There are no unsigned pc-relative relocations"));
switch (size)
{
case 1: return BFD_RELOC_8_PCREL;
}
else
{
- switch (size)
- {
- case 1: return BFD_RELOC_8;
- case 2: return BFD_RELOC_16;
- case 4: return BFD_RELOC_32;
- }
- as_bad (_("can not do %d byte relocation"), size);
+ if (sign)
+ switch (size)
+ {
+ case 4: return BFD_RELOC_X86_64_32S;
+ }
+ else
+ switch (size)
+ {
+ case 1: return BFD_RELOC_8;
+ case 2: return BFD_RELOC_16;
+ case 4: return BFD_RELOC_32;
+ case 8: return BFD_RELOC_64;
+ }
+ as_bad (_("can not do %s %d byte relocation"),
+ sign ? "signed" : "unsigned", size);
}
+ abort();
return BFD_RELOC_NONE;
}
|| S_IS_WEAK (fixP->fx_addsy))
return 0;
#endif
- /* adjust_reloc_syms doesn't know about the GOT */
+ /* adjust_reloc_syms doesn't know about the GOT. */
if (fixP->fx_r_type == BFD_RELOC_386_GOTOFF
|| fixP->fx_r_type == BFD_RELOC_386_PLT32
|| fixP->fx_r_type == BFD_RELOC_386_GOT32
+ || fixP->fx_r_type == BFD_RELOC_X86_64_PLT32
+ || fixP->fx_r_type == BFD_RELOC_X86_64_GOT32
|| fixP->fx_r_type == BFD_RELOC_VTABLE_INHERIT
|| fixP->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
return 0;
#define BFD_RELOC_386_PLT32 0
#define BFD_RELOC_386_GOT32 0
#define BFD_RELOC_386_GOTOFF 0
+#define BFD_RELOC_X86_64_PLT32 0
+#define BFD_RELOC_X86_64_GOT32 0
#endif
static int intel_float_operand PARAMS ((char *mnemonic));
i.disp_reloc[j] = NO_RELOC;
memset (disp_expressions, '\0', sizeof (disp_expressions));
memset (im_expressions, '\0', sizeof (im_expressions));
- save_stack_p = save_stack; /* reset stack pointer */
+ save_stack_p = save_stack;
/* First parse an instruction mnemonic & call i386_operand for the operands.
We assume that the scrubber has arranged it so that line[0] is the valid
Similarly, in 32-bit mode, do not allow addr32 or data32. */
if ((current_templates->start->opcode_modifier & (Size16 | Size32))
&& (((current_templates->start->opcode_modifier & Size32) != 0)
- ^ flag_16bit_code))
+ ^ (flag_code == CODE_16BIT)))
{
as_bad (_("redundant %s prefix"),
current_templates->start->name);
{
case WORD_MNEM_SUFFIX:
case BYTE_MNEM_SUFFIX:
- case SHORT_MNEM_SUFFIX:
- case LONG_MNEM_SUFFIX:
+ case QWORD_MNEM_SUFFIX:
i.suffix = mnem_p[-1];
mnem_p[-1] = '\0';
current_templates = hash_find (op_hash, mnemonic);
break;
+ case SHORT_MNEM_SUFFIX:
+ case LONG_MNEM_SUFFIX:
+ if (!intel_syntax)
+ {
+ i.suffix = mnem_p[-1];
+ mnem_p[-1] = '\0';
+ current_templates = hash_find (op_hash, mnemonic);
+ }
+ break;
- /* Intel Syntax */
- case DWORD_MNEM_SUFFIX:
+ /* Intel Syntax. */
+ case 'd':
if (intel_syntax)
{
- i.suffix = mnem_p[-1];
+ if (intel_float_operand (mnemonic))
+ i.suffix = SHORT_MNEM_SUFFIX;
+ else
+ i.suffix = LONG_MNEM_SUFFIX;
mnem_p[-1] = '\0';
current_templates = hash_find (op_hash, mnemonic);
- break;
}
+ break;
}
if (!current_templates)
{
/* Check if instruction is supported on specified architecture. */
if (cpu_arch_flags != 0)
{
- if (current_templates->start->cpu_flags & ~cpu_arch_flags)
+ if ((current_templates->start->cpu_flags & ~(Cpu64 | CpuNo64))
+ & ~(cpu_arch_flags & ~(Cpu64 | CpuNo64)))
{
as_warn (_("`%s' is not supported on `%s'"),
current_templates->start->name, cpu_arch_name);
}
- else if ((Cpu386 & ~cpu_arch_flags) && !flag_16bit_code)
+ else if ((Cpu386 & ~cpu_arch_flags) && (flag_code != CODE_16BIT))
{
as_warn (_("use .code16 to ensure correct addressing mode"));
}
}
- /* check for rep/repne without a string instruction */
+ /* Check for rep/repne without a string instruction. */
if (expecting_string_instruction
&& !(current_templates->start->opcode_modifier & IsString))
{
/* There may be operands to parse. */
if (*l != END_OF_INSN)
{
- /* parse operands */
-
/* 1 if operand is pending after ','. */
unsigned int expecting_operand = 0;
do
{
- /* skip optional white space before operand */
+ /* Skip optional white space before operand. */
if (is_space_char (*l))
++l;
if (!is_operand_char (*l) && *l != END_OF_INSN)
else
operand_ok = i386_operand (token_start);
- RESTORE_END_STRING (l); /* restore old contents */
+ RESTORE_END_STRING (l);
if (!operand_ok)
return;
}
}
}
- /* now *l must be either ',' or END_OF_INSN */
+ /* Now *l must be either ',' or END_OF_INSN. */
if (*l == ',')
{
if (*++l == END_OF_INSN)
- { /* just skip it, if it's \n complain */
+ {
+ /* Just skip it, if it's \n complain. */
goto expecting_operand_after_comma;
}
expecting_operand = 1;
}
}
- while (*l != END_OF_INSN); /* until we get end of insn */
+ while (*l != END_OF_INSN);
}
}
{
union i386_op temp_op;
unsigned int temp_type;
+#ifdef BFD_ASSEMBLER
+ enum bfd_reloc_code_real temp_reloc;
+#else
+ int temp_reloc;
+#endif
int xchg1 = 0;
int xchg2 = 0;
temp_op = i.op[xchg2];
i.op[xchg2] = i.op[xchg1];
i.op[xchg1] = temp_op;
+ temp_reloc = i.disp_reloc[xchg2];
+ i.disp_reloc[xchg2] = i.disp_reloc[xchg1];
+ i.disp_reloc[xchg1] = temp_reloc;
if (i.mem_operands == 2)
{
guess_suffix = BYTE_MNEM_SUFFIX;
else if (i.types[op] & Reg16)
guess_suffix = WORD_MNEM_SUFFIX;
+ else if (i.types[op] & Reg32)
+ guess_suffix = LONG_MNEM_SUFFIX;
+ else if (i.types[op] & Reg64)
+ guess_suffix = QWORD_MNEM_SUFFIX;
break;
}
}
- else if (flag_16bit_code ^ (i.prefix[DATA_PREFIX] != 0))
+ else if ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0))
guess_suffix = WORD_MNEM_SUFFIX;
for (op = i.operands; --op >= 0;)
- if ((i.types[op] & Imm)
- && i.op[op].imms->X_op == O_constant)
+ if (i.types[op] & Imm)
{
- /* If a suffix is given, this operand may be shortened. */
- switch (guess_suffix)
- {
- case WORD_MNEM_SUFFIX:
- i.types[op] |= Imm16;
- break;
- case BYTE_MNEM_SUFFIX:
- i.types[op] |= Imm16 | Imm8 | Imm8S;
- break;
- }
+ switch (i.op[op].imms->X_op)
+ {
+ case O_constant:
+ /* If a suffix is given, this operand may be shortened. */
+ switch (guess_suffix)
+ {
+ case LONG_MNEM_SUFFIX:
+ i.types[op] |= Imm32 | Imm64;
+ break;
+ case WORD_MNEM_SUFFIX:
+ i.types[op] |= Imm16 | Imm32S | Imm32 | Imm64;
+ break;
+ case BYTE_MNEM_SUFFIX:
+ i.types[op] |= Imm16 | Imm8 | Imm8S | Imm32S | Imm32 | Imm64;
+ break;
+ }
- /* If this operand is at most 16 bits, convert it to a
- signed 16 bit number before trying to see whether it will
- fit in an even smaller size. This allows a 16-bit operand
- such as $0xffe0 to be recognised as within Imm8S range. */
- if ((i.types[op] & Imm16)
- && (i.op[op].imms->X_add_number & ~(offsetT)0xffff) == 0)
- {
- i.op[op].imms->X_add_number =
- (((i.op[op].imms->X_add_number & 0xffff) ^ 0x8000) - 0x8000);
+ /* If this operand is at most 16 bits, convert it to a
+ signed 16 bit number before trying to see whether it will
+ fit in an even smaller size. This allows a 16-bit operand
+ such as $0xffe0 to be recognised as within Imm8S range. */
+ if ((i.types[op] & Imm16)
+ && (i.op[op].imms->X_add_number & ~(offsetT)0xffff) == 0)
+ {
+ i.op[op].imms->X_add_number =
+ (((i.op[op].imms->X_add_number & 0xffff) ^ 0x8000) - 0x8000);
+ }
+ if ((i.types[op] & Imm32)
+ && (i.op[op].imms->X_add_number & ~(((offsetT) 2 << 31) - 1)) == 0)
+ {
+ i.op[op].imms->X_add_number =
+ (i.op[op].imms->X_add_number ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
+ }
+ i.types[op] |= smallest_imm_type (i.op[op].imms->X_add_number);
+ /* We must avoid matching of Imm32 templates when 64bit only immediate is available. */
+ if (guess_suffix == QWORD_MNEM_SUFFIX)
+ i.types[op] &= ~Imm32;
+ break;
+ case O_absent:
+ case O_register:
+ abort();
+ /* Symbols and expressions. */
+ default:
+ /* Convert symbolic operand to proper sizes for matching. */
+ switch (guess_suffix)
+ {
+ case QWORD_MNEM_SUFFIX:
+ i.types[op] = Imm64 | Imm32S;
+ break;
+ case LONG_MNEM_SUFFIX:
+ i.types[op] = Imm32 | Imm64;
+ break;
+ case WORD_MNEM_SUFFIX:
+ i.types[op] = Imm16 | Imm32 | Imm64;
+ break;
+ break;
+ case BYTE_MNEM_SUFFIX:
+ i.types[op] = Imm8 | Imm8S | Imm16 | Imm32S | Imm32;
+ break;
+ break;
+ }
+ break;
}
- i.types[op] |= smallest_imm_type ((long) i.op[op].imms->X_add_number);
}
}
disp = (((disp & 0xffff) ^ 0x8000) - 0x8000);
}
- if (fits_in_signed_byte (disp))
+ else if (i.types[op] & Disp32)
+ {
+ /* We know this operand is at most 32 bits, so convert to a
+ signed 32 bit number before trying to see whether it will
+ fit in an even smaller size. */
+ disp &= (((offsetT) 2 << 31) - 1);
+ disp = (disp ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
+ }
+ if (flag_code == CODE_64BIT)
+ {
+ if (fits_in_signed_long (disp))
+ i.types[op] |= Disp32S;
+ if (fits_in_unsigned_long (disp))
+ i.types[op] |= Disp32;
+ }
+ if ((i.types[op] & (Disp32 | Disp32S | Disp16))
+ && fits_in_signed_byte (disp))
i.types[op] |= Disp8;
}
}
? No_sSuf
: (i.suffix == LONG_MNEM_SUFFIX
? No_lSuf
- : (i.suffix == DWORD_MNEM_SUFFIX
- ? No_dSuf
+ : (i.suffix == QWORD_MNEM_SUFFIX
+ ? No_qSuf
: (i.suffix == LONG_DOUBLE_MNEM_SUFFIX ? No_xSuf : 0))))));
for (t = current_templates->start;
/* Check the suffix, except for some instructions in intel mode. */
if ((t->opcode_modifier & suffix_check)
+ && !(intel_syntax
+ && (t->opcode_modifier & IgnoreSize))
&& !(intel_syntax
&& t->base_opcode == 0xd9
- && (t->extension_opcode == 5 /* 0xd9,5 "fldcw" */
- || t->extension_opcode == 7))) /* 0xd9,7 "f{n}stcw" */
+ && (t->extension_opcode == 5 /* 0xd9,5 "fldcw" */
+ || t->extension_opcode == 7))) /* 0xd9,7 "f{n}stcw" */
continue;
else if (!t->operands)
/* found_reverse_match holds which of D or FloatDR
we've found. */
found_reverse_match = t->opcode_modifier & (D|FloatDR);
- break;
}
/* Found a forward 2 operand match here. */
- if (t->operands == 3)
+ else if (t->operands == 3)
{
/* Here we make use of the fact that there are no
reverse match 3 operand instructions, and all 3
continue;
}
/* Found either forward/reverse 2 or 3 operand match here:
- slip through to break */
+ slip through to break. */
+ }
+ if (t->cpu_flags & ~cpu_arch_flags)
+ {
+ found_reverse_match = 0;
+ continue;
}
/* We've found a match; break out of loop. */
break;
- } /* for (t = ... */
+ }
if (t == current_templates->end)
{
/* We found no match. */
if (! add_prefix (FWAIT_OPCODE))
return;
- /* Check string instruction segment overrides */
+ /* Check string instruction segment overrides. */
if ((i.tm.opcode_modifier & IsString) != 0 && i.mem_operands != 0)
{
int mem_op = (i.types[0] & AnyMem) ? 0 : 1;
}
}
+ if (i.reg_operands && flag_code < CODE_64BIT)
+ {
+ int op;
+ for (op = i.operands; --op >= 0; )
+ if ((i.types[op] & Reg)
+ && (i.op[op].regs->reg_flags & (RegRex64|RegRex)))
+ as_bad (_("Extended register `%%%s' available only in 64bit mode."),
+ i.op[op].regs->reg_name);
+ }
+
/* If matched instruction specifies an explicit instruction mnemonic
suffix, use it. */
- if (i.tm.opcode_modifier & (Size16 | Size32))
+ if (i.tm.opcode_modifier & (Size16 | Size32 | Size64))
{
if (i.tm.opcode_modifier & Size16)
i.suffix = WORD_MNEM_SUFFIX;
+ else if (i.tm.opcode_modifier & Size64)
+ i.suffix = QWORD_MNEM_SUFFIX;
else
i.suffix = LONG_MNEM_SUFFIX;
}
{
i.suffix = ((i.types[op] & Reg8) ? BYTE_MNEM_SUFFIX :
(i.types[op] & Reg16) ? WORD_MNEM_SUFFIX :
+ (i.types[op] & Reg64) ? QWORD_MNEM_SUFFIX :
LONG_MNEM_SUFFIX);
break;
}
if (intel_syntax
&& (i.tm.base_opcode == 0xfb7
|| i.tm.base_opcode == 0xfb6
+ || i.tm.base_opcode == 0x63
|| i.tm.base_opcode == 0xfbe
|| i.tm.base_opcode == 0xfbf))
continue;
#endif
)
{
+ /* Prohibit these changes in the 64bit mode, since
+ the lowering is more complicated. */
+ if (flag_code == CODE_64BIT
+ && (i.tm.operand_types[op] & InOutPortReg) == 0)
+ as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
+ i.op[op].regs->reg_name,
+ i.suffix);
#if REGISTER_WARNINGS
if (!quiet_warnings
&& (i.tm.operand_types[op] & InOutPortReg) == 0)
#endif
continue;
}
- /* Any other register is bad */
+ /* Any other register is bad. */
if (i.types[op] & (Reg | RegMMX | RegXMM
| SReg2 | SReg3
| Control | Debug | Test
i.suffix);
return;
}
-#if REGISTER_WARNINGS
/* Warn if the e prefix on a general reg is missing. */
- else if (!quiet_warnings
+ else if ((!quiet_warnings || flag_code == CODE_64BIT)
&& (i.types[op] & Reg16) != 0
&& (i.tm.operand_types[op] & (Reg32|Acc)) != 0)
{
- as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
- (i.op[op].regs + 8)->reg_name,
- i.op[op].regs->reg_name,
- i.suffix);
- }
+ /* Prohibit these changes in the 64bit mode, since
+ the lowering is more complicated. */
+ if (flag_code == CODE_64BIT)
+ as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
+ i.op[op].regs->reg_name,
+ i.suffix);
+#if REGISTER_WARNINGS
+ else
+ as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
+ (i.op[op].regs + 8)->reg_name,
+ i.op[op].regs->reg_name,
+ i.suffix);
#endif
+ }
+ /* Warn if the r prefix on a general reg is missing. */
+ else if ((i.types[op] & Reg64) != 0
+ && (i.tm.operand_types[op] & (Reg32|Acc)) != 0)
+ {
+ as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
+ i.op[op].regs->reg_name,
+ i.suffix);
+ }
+ }
+ else if (i.suffix == QWORD_MNEM_SUFFIX)
+ {
+ int op;
+ if (flag_code < CODE_64BIT)
+ as_bad (_("64bit operations available only in 64bit modes."));
+
+ for (op = i.operands; --op >= 0; )
+ /* Reject eight bit registers, except where the template
+ requires them. (eg. movzb) */
+ if ((i.types[op] & Reg8) != 0
+ && (i.tm.operand_types[op] & (Reg16|Reg32|Acc)) != 0)
+ {
+ as_bad (_("`%%%s' not allowed with `%s%c'"),
+ i.op[op].regs->reg_name,
+ i.tm.name,
+ i.suffix);
+ return;
+ }
+ /* Warn if the e prefix on a general reg is missing. */
+ else if (((i.types[op] & Reg16) != 0
+ || (i.types[op] & Reg32) != 0)
+ && (i.tm.operand_types[op] & (Reg32|Acc)) != 0)
+ {
+ /* Prohibit these changes in the 64bit mode, since
+ the lowering is more complicated. */
+ as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
+ i.op[op].regs->reg_name,
+ i.suffix);
+ }
}
else if (i.suffix == WORD_MNEM_SUFFIX)
{
i.suffix);
return;
}
-#if REGISTER_WARNINGS
/* Warn if the e prefix on a general reg is present. */
- else if (!quiet_warnings
+ else if ((!quiet_warnings || flag_code == CODE_64BIT)
&& (i.types[op] & Reg32) != 0
&& (i.tm.operand_types[op] & (Reg16|Acc)) != 0)
{
- as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
- (i.op[op].regs - 8)->reg_name,
- i.op[op].regs->reg_name,
- i.suffix);
- }
+ /* Prohibit these changes in the 64bit mode, since
+ the lowering is more complicated. */
+ if (flag_code == CODE_64BIT)
+ as_bad (_("Incorrect register `%%%s' used with`%c' suffix"),
+ i.op[op].regs->reg_name,
+ i.suffix);
+ else
+#if REGISTER_WARNINGS
+ as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
+ (i.op[op].regs - 8)->reg_name,
+ i.op[op].regs->reg_name,
+ i.suffix);
#endif
+ }
}
+ else if (intel_syntax && (i.tm.opcode_modifier & IgnoreSize))
+ /* Do nothing if the instruction is going to ignore the prefix. */
+ ;
else
abort ();
}
{
i.suffix = stackop_size;
}
-
/* Make still unresolved immediate matches conform to size of immediate
given in i.suffix. Note: overlap2 cannot be an immediate! */
- if ((overlap0 & (Imm8 | Imm8S | Imm16 | Imm32))
+ if ((overlap0 & (Imm8 | Imm8S | Imm16 | Imm32 | Imm32S))
&& overlap0 != Imm8 && overlap0 != Imm8S
- && overlap0 != Imm16 && overlap0 != Imm32)
+ && overlap0 != Imm16 && overlap0 != Imm32S
+ && overlap0 != Imm32 && overlap0 != Imm64)
{
if (i.suffix)
{
overlap0 &= (i.suffix == BYTE_MNEM_SUFFIX ? (Imm8 | Imm8S) :
- (i.suffix == WORD_MNEM_SUFFIX ? Imm16 : Imm32));
+ (i.suffix == WORD_MNEM_SUFFIX ? Imm16 :
+ (i.suffix == QWORD_MNEM_SUFFIX ? Imm64 | Imm32S : Imm32)));
}
- else if (overlap0 == (Imm16 | Imm32))
+ else if (overlap0 == (Imm16 | Imm32S | Imm32)
+ || overlap0 == (Imm16 | Imm32)
+ || overlap0 == (Imm16 | Imm32S))
{
overlap0 =
- (flag_16bit_code ^ (i.prefix[DATA_PREFIX] != 0)) ? Imm16 : Imm32;
+ ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)) ? Imm16 : Imm32S;
}
- else
+ if (overlap0 != Imm8 && overlap0 != Imm8S
+ && overlap0 != Imm16 && overlap0 != Imm32S
+ && overlap0 != Imm32 && overlap0 != Imm64)
{
as_bad (_("no instruction mnemonic suffix given; can't determine immediate size"));
return;
}
}
- if ((overlap1 & (Imm8 | Imm8S | Imm16 | Imm32))
+ if ((overlap1 & (Imm8 | Imm8S | Imm16 | Imm32S | Imm32))
&& overlap1 != Imm8 && overlap1 != Imm8S
- && overlap1 != Imm16 && overlap1 != Imm32)
+ && overlap1 != Imm16 && overlap1 != Imm32S
+ && overlap1 != Imm32 && overlap1 != Imm64)
{
if (i.suffix)
{
overlap1 &= (i.suffix == BYTE_MNEM_SUFFIX ? (Imm8 | Imm8S) :
- (i.suffix == WORD_MNEM_SUFFIX ? Imm16 : Imm32));
+ (i.suffix == WORD_MNEM_SUFFIX ? Imm16 :
+ (i.suffix == QWORD_MNEM_SUFFIX ? Imm64 | Imm32S : Imm32)));
}
- else if (overlap1 == (Imm16 | Imm32))
+ else if (overlap1 == (Imm16 | Imm32 | Imm32S)
+ || overlap1 == (Imm16 | Imm32)
+ || overlap1 == (Imm16 | Imm32S))
{
overlap1 =
- (flag_16bit_code ^ (i.prefix[DATA_PREFIX] != 0)) ? Imm16 : Imm32;
+ ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)) ? Imm16 : Imm32S;
}
- else
+ if (overlap1 != Imm8 && overlap1 != Imm8S
+ && overlap1 != Imm16 && overlap1 != Imm32S
+ && overlap1 != Imm32 && overlap1 != Imm64)
{
- as_bad (_("no instruction mnemonic suffix given; can't determine immediate size"));
+ as_bad (_("no instruction mnemonic suffix given; can't determine immediate size %x %c"),overlap1, i.suffix);
return;
}
}
if (overlap0 & ImplicitRegister)
i.reg_operands--;
if (overlap0 & Imm1)
- i.imm_operands = 0; /* kludge for shift insns */
+ i.imm_operands = 0; /* kludge for shift insns. */
i.types[1] = overlap1;
if (overlap1 & ImplicitRegister)
return;
}
- /* For movzx and movsx, need to check the register type */
+ /* For movzx and movsx, need to check the register type. */
if (intel_syntax
&& (i.tm.base_opcode == 0xfb6 || i.tm.base_opcode == 0xfbe))
if (i.suffix && i.suffix == BYTE_MNEM_SUFFIX)
/* Now select between word & dword operations via the operand
size prefix, except for instructions that will ignore this
prefix anyway. */
- if (((intel_syntax && (i.suffix == DWORD_MNEM_SUFFIX))
- || i.suffix == LONG_MNEM_SUFFIX) == flag_16bit_code
+ if (i.suffix != QWORD_MNEM_SUFFIX
+ && (i.suffix == LONG_MNEM_SUFFIX) == (flag_code == CODE_16BIT)
&& !(i.tm.opcode_modifier & IgnoreSize))
{
unsigned int prefix = DATA_PREFIX_OPCODE;
if (! add_prefix (prefix))
return;
}
+
+ /* Set mode64 for an operand. */
+ if (i.suffix == QWORD_MNEM_SUFFIX
+ && !(i.tm.opcode_modifier & NoRex64))
+ i.rex.mode64 = 1;
+
/* Size floating point instruction. */
- if (i.suffix == LONG_MNEM_SUFFIX
- || (intel_syntax && i.suffix == DWORD_MNEM_SUFFIX))
+ if (i.suffix == LONG_MNEM_SUFFIX)
{
if (i.tm.opcode_modifier & FloatMF)
i.tm.base_opcode ^= 4;
unsigned int op = (i.types[0] & (Reg | FloatReg)) ? 0 : 1;
/* Register goes in low 3 bits of opcode. */
i.tm.base_opcode |= i.op[op].regs->reg_num;
+ if (i.op[op].regs->reg_flags & RegRex)
+ i.rex.extZ=1;
if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
{
/* Warn about some common errors, but press on regardless.
{
i.rm.reg = i.op[dest].regs->reg_num;
i.rm.regmem = i.op[source].regs->reg_num;
+ if (i.op[dest].regs->reg_flags & RegRex)
+ i.rex.extX=1;
+ if (i.op[source].regs->reg_flags & RegRex)
+ i.rex.extZ=1;
}
else
{
i.rm.reg = i.op[source].regs->reg_num;
i.rm.regmem = i.op[dest].regs->reg_num;
+ if (i.op[dest].regs->reg_flags & RegRex)
+ i.rex.extZ=1;
+ if (i.op[source].regs->reg_flags & RegRex)
+ i.rex.extX=1;
}
}
else
if (! i.index_reg)
{
/* Operand is just <disp> */
- if (flag_16bit_code ^ (i.prefix[ADDR_PREFIX] != 0))
+ if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
{
i.rm.regmem = NO_BASE_REGISTER_16;
i.types[op] &= ~Disp;
i.types[op] |= Disp16;
}
- else
+ else if (flag_code != CODE_64BIT)
{
i.rm.regmem = NO_BASE_REGISTER;
i.types[op] &= ~Disp;
i.types[op] |= Disp32;
}
+ else
+ {
+ /* 64bit mode overwrites the 32bit absolute addressing
+ by RIP relative addressing and absolute addressing
+ is encoded by one of the redundant SIB forms. */
+
+ i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
+ i.sib.base = NO_BASE_REGISTER;
+ i.sib.index = NO_INDEX_REGISTER;
+ i.types[op] &= ~Disp;
+ i.types[op] |= Disp32S;
+ }
}
else /* ! i.base_reg && i.index_reg */
{
i.sib.scale = i.log2_scale_factor;
i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
i.types[op] &= ~Disp;
- i.types[op] |= Disp32; /* Must be 32 bit. */
+ if (flag_code != CODE_64BIT)
+ i.types[op] |= Disp32; /* Must be 32 bit */
+ else
+ i.types[op] |= Disp32S;
+ if (i.index_reg->reg_flags & RegRex)
+ i.rex.extY=1;
}
}
+ /* RIP addressing for 64bit mode. */
+ else if (i.base_reg->reg_type == BaseIndex)
+ {
+ i.rm.regmem = NO_BASE_REGISTER;
+ i.types[op] &= ~Disp;
+ i.types[op] |= Disp32S;
+ i.flags[op] = Operand_PCrel;
+ }
else if (i.base_reg->reg_type & Reg16)
{
switch (i.base_reg->reg_num)
}
i.rm.mode = mode_from_disp_size (i.types[op]);
}
- else /* i.base_reg and 32 bit mode */
+ else /* i.base_reg and 32/64 bit mode */
{
+ if (flag_code == CODE_64BIT
+ && (i.types[op] & Disp))
+ {
+ if (i.types[op] & Disp8)
+ i.types[op] = Disp8 | Disp32S;
+ else
+ i.types[op] = Disp32S;
+ }
i.rm.regmem = i.base_reg->reg_num;
+ if (i.base_reg->reg_flags & RegRex)
+ i.rex.extZ=1;
i.sib.base = i.base_reg->reg_num;
- if (i.base_reg->reg_num == EBP_REG_NUM)
+ /* x86-64 ignores REX prefix bit here to avoid
+ decoder complications. */
+ if ((i.base_reg->reg_num & 7) == EBP_REG_NUM)
{
default_seg = &ss;
if (i.disp_operands == 0)
{
i.sib.index = i.index_reg->reg_num;
i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
+ if (i.index_reg->reg_flags & RegRex)
+ i.rex.extY=1;
}
i.rm.mode = mode_from_disp_size (i.types[op]);
}
/* If there is an extension opcode to put here, the
register number must be put into the regmem field. */
if (i.tm.extension_opcode != None)
- i.rm.regmem = i.op[op].regs->reg_num;
+ {
+ i.rm.regmem = i.op[op].regs->reg_num;
+ if (i.op[op].regs->reg_flags & RegRex)
+ i.rex.extZ=1;
+ }
else
- i.rm.reg = i.op[op].regs->reg_num;
+ {
+ i.rm.reg = i.op[op].regs->reg_num;
+ if (i.op[op].regs->reg_flags & RegRex)
+ i.rex.extX=1;
+ }
/* Now, if no memory operand has set i.rm.mode = 0, 1, 2
we must set it to 3 to indicate this is a register
return;
}
i.tm.base_opcode |= (i.op[0].regs->reg_num << 3);
+ if (i.op[0].regs->reg_flags & RegRex)
+ i.rex.extZ = 1;
}
else if ((i.tm.base_opcode & ~(D|W)) == MOV_AX_DISP32)
{
i.op[0].disps->X_op = O_symbol;
}
+ if (i.tm.opcode_modifier & Rex64)
+ i.rex.mode64 = 1;
+
+ /* For 8bit registers we would need an empty rex prefix.
+ Also in the case instruction is already having prefix,
+ we need to convert old registers to new ones. */
+
+ if (((i.types[0] & Reg8) && (i.op[0].regs->reg_flags & RegRex64))
+ || ((i.types[1] & Reg8) && (i.op[1].regs->reg_flags & RegRex64))
+ || ((i.rex.mode64 || i.rex.extX || i.rex.extY || i.rex.extZ || i.rex.empty)
+ && ((i.types[0] & Reg8) || (i.types[1] & Reg8))))
+ {
+ int x;
+ i.rex.empty=1;
+ for (x = 0; x < 2; x++)
+ {
+ /* Look for 8bit operand that does use old registers. */
+ if (i.types[x] & Reg8
+ && !(i.op[x].regs->reg_flags & RegRex64))
+ {
+ /* In case it is "hi" register, give up. */
+ if (i.op[x].regs->reg_num > 3)
+ as_bad (_("Can't encode registers '%%%s' in the instruction requiring REX prefix.\n"),
+ i.op[x].regs->reg_name);
+
+ /* Otherwise it is equivalent to the extended register.
+ Since the encoding don't change this is merely cosmetical
+ cleanup for debug output. */
+
+ i.op[x].regs = i.op[x].regs + 8;
+ }
+ }
+ }
+
+ if (i.rex.mode64 || i.rex.extX || i.rex.extY || i.rex.extZ || i.rex.empty)
+ add_prefix (0x40
+ | (i.rex.mode64 ? 8 : 0)
+ | (i.rex.extX ? 4 : 0)
+ | (i.rex.extY ? 2 : 0)
+ | (i.rex.extZ ? 1 : 0));
+
/* We are ready to output the insn. */
{
register char *p;
int prefix;
code16 = 0;
- if (flag_16bit_code)
+ if (flag_code == CODE_16BIT)
code16 = CODE16;
prefix = 0;
i.prefixes -= 1;
code16 ^= CODE16;
}
+ if (i.prefix[REX_PREFIX])
+ {
+ prefix++;
+ i.prefixes --;
+ }
size = 4;
if (code16)
insn_size += prefix + 1;
/* Prefix and 1 opcode byte go in fr_fix. */
p = frag_more (prefix + 1);
- if (prefix)
+ if (i.prefix[DATA_PREFIX])
*p++ = DATA_PREFIX_OPCODE;
+ if (i.prefix[REX_PREFIX])
+ *p++ = i.prefix[REX_PREFIX];
*p = i.tm.base_opcode;
/* 1 possible extra opcode + displacement go in var part.
Pass reloc in fr_var. */
int code16;
code16 = 0;
- if (flag_16bit_code)
+ if (flag_code == CODE_16BIT)
code16 = CODE16;
if (i.prefix[DATA_PREFIX])
size = 2;
}
+ if (i.prefix[REX_PREFIX])
+ {
+ FRAG_APPEND_1_CHAR (i.prefix[REX_PREFIX]);
+ insn_size++;
+ i.prefixes -= 1;
+ }
+
if (i.prefixes != 0 && !intel_syntax)
as_warn (_("skipping prefixes on this instruction"));
*p++ = i.tm.base_opcode & 0xff;
fix_new_exp (frag_now, p - frag_now->fr_literal, size,
- i.op[0].disps, 1, reloc (size, 1, i.disp_reloc[0]));
+ i.op[0].disps, 1, reloc (size, 1, 1, i.disp_reloc[0]));
}
else if (i.tm.opcode_modifier & JumpInterSegment)
{
int code16;
code16 = 0;
- if (flag_16bit_code)
+ if (flag_code == CODE_16BIT)
code16 = CODE16;
prefix = 0;
i.prefixes -= 1;
code16 ^= CODE16;
}
+ if (i.prefix[REX_PREFIX])
+ {
+ prefix++;
+ i.prefixes -= 1;
+ }
size = 4;
if (code16)
/* 1 opcode; 2 segment; offset */
insn_size += prefix + 1 + 2 + size;
p = frag_more (prefix + 1 + 2 + size);
- if (prefix)
+
+ if (i.prefix[DATA_PREFIX])
*p++ = DATA_PREFIX_OPCODE;
+
+ if (i.prefix[REX_PREFIX])
+ *p++ = i.prefix[REX_PREFIX];
+
*p++ = i.tm.base_opcode;
if (i.op[1].imms->X_op == O_constant)
{
}
else
fix_new_exp (frag_now, p - frag_now->fr_literal, size,
- i.op[1].imms, 0, reloc (size, 0, i.disp_reloc[0]));
+ i.op[1].imms, 0, reloc (size, 0, 0, i.disp_reloc[0]));
if (i.op[0].imms->X_op != O_constant)
as_bad (_("can't handle non absolute segment in `%s'"),
i.tm.name);
offsetT val;
size = 4;
- if (i.types[n] & (Disp8 | Disp16))
+ if (i.types[n] & (Disp8 | Disp16 | Disp64))
{
size = 2;
if (i.types[n] & Disp8)
size = 1;
+ if (i.types[n] & Disp64)
+ size = 8;
}
val = offset_in_range (i.op[n].disps->X_add_number,
size);
else
{
int size = 4;
+ int sign = 0;
+ int pcrel = (i.flags[n] & Operand_PCrel) != 0;
+
+ /* The PC relative address is computed relative
+ to the instruction boundary, so in case immediate
+ fields follows, we need to adjust the value. */
+ if (pcrel && i.imm_operands)
+ {
+ int imm_size = 4;
+ register unsigned int n1;
+
+ for (n1 = 0; n1 < i.operands; n1++)
+ if (i.types[n1] & Imm)
+ {
+ if (i.types[n1] & (Imm8 | Imm8S | Imm16 | Imm64))
+ {
+ imm_size = 2;
+ if (i.types[n1] & (Imm8 | Imm8S))
+ imm_size = 1;
+ if (i.types[n1] & Imm64)
+ imm_size = 8;
+ }
+ break;
+ }
+ /* We should find the immediate. */
+ if (n1 == i.operands)
+ abort();
+ i.op[n].disps->X_add_number -= imm_size;
+ }
- if (i.types[n] & Disp16)
- size = 2;
+ if (i.types[n] & Disp32S)
+ sign = 1;
+
+ if (i.types[n] & (Disp16 | Disp64))
+ {
+ size = 2;
+ if (i.types[n] & Disp64)
+ size = 8;
+ }
insn_size += size;
p = frag_more (size);
fix_new_exp (frag_now, p - frag_now->fr_literal, size,
- i.op[n].disps, 0,
- reloc (size, 0, i.disp_reloc[n]));
+ i.op[n].disps, pcrel,
+ reloc (size, pcrel, sign, i.disp_reloc[n]));
}
}
}
- } /* End displacement output. */
+ }
/* Output immediate. */
if (i.imm_operands)
offsetT val;
size = 4;
- if (i.types[n] & (Imm8 | Imm8S | Imm16))
+ if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
{
size = 2;
if (i.types[n] & (Imm8 | Imm8S))
size = 1;
+ else if (i.types[n] & Imm64)
+ size = 8;
}
val = offset_in_range (i.op[n].imms->X_add_number,
size);
md_number_to_chars (p, val, size);
}
else
- { /* Not absolute_section. */
- /* Need a 32-bit fixup (don't support 8bit
+ {
+ /* Not absolute_section.
+ Need a 32-bit fixup (don't support 8bit
non-absolute imms). Try to support other
sizes ... */
#ifdef BFD_ASSEMBLER
int reloc_type;
#endif
int size = 4;
+ int sign = 0;
- if (i.types[n] & Imm16)
- size = 2;
- else if (i.types[n] & (Imm8 | Imm8S))
- size = 1;
+ if ((i.types[n] & (Imm32S))
+ && i.suffix == QWORD_MNEM_SUFFIX)
+ sign = 1;
+ if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
+ {
+ size = 2;
+ if (i.types[n] & (Imm8 | Imm8S))
+ size = 1;
+ if (i.types[n] & Imm64)
+ size = 8;
+ }
insn_size += size;
p = frag_more (size);
- reloc_type = reloc (size, 0, i.disp_reloc[0]);
+ reloc_type = reloc (size, 0, sign, i.disp_reloc[0]);
#ifdef BFD_ASSEMBLER
if (reloc_type == BFD_RELOC_32
&& GOT_symbol
(i.op[n].imms->X_op_symbol)->X_op)
== O_subtract))))
{
+ /* We don't support dynamic linking on x86-64 yet. */
+ if (flag_code == CODE_64BIT)
+ abort();
reloc_type = BFD_RELOC_386_GOTPC;
i.op[n].imms->X_add_number += 3;
}
}
}
}
- } /* end immediate output */
+ }
}
+ dwarf2_emit_insn (insn_size);
+
#ifdef DEBUG386
if (flag_debug)
{
int first;
/* GOT relocations are not supported in 16 bit mode. */
- if (flag_16bit_code)
+ if (flag_code == CODE_16BIT)
as_bad (_("GOT relocations not supported in 16 bit mode"));
if (GOT_symbol == NULL)
if (strncmp (cp + 1, "PLT", 3) == 0)
{
- i.disp_reloc[this_operand] = BFD_RELOC_386_PLT32;
+ if (flag_code == CODE_64BIT)
+ i.disp_reloc[this_operand] = BFD_RELOC_X86_64_PLT32;
+ else
+ i.disp_reloc[this_operand] = BFD_RELOC_386_PLT32;
len = 3;
}
else if (strncmp (cp + 1, "GOTOFF", 6) == 0)
{
+ if (flag_code == CODE_64BIT)
+ as_bad ("GOTOFF relocations are unsupported in 64bit mode.");
i.disp_reloc[this_operand] = BFD_RELOC_386_GOTOFF;
len = 6;
}
else if (strncmp (cp + 1, "GOT", 3) == 0)
{
- i.disp_reloc[this_operand] = BFD_RELOC_386_GOT32;
+ if (flag_code == CODE_64BIT)
+ i.disp_reloc[this_operand] = BFD_RELOC_X86_64_GOT32;
+ else
+ i.disp_reloc[this_operand] = BFD_RELOC_386_GOT32;
+ len = 3;
+ }
+ else if (strncmp (cp + 1, "GOTPCREL", 3) == 0)
+ {
+ if (flag_code == CODE_64BIT)
+ i.disp_reloc[this_operand] = BFD_RELOC_X86_64_GOTPCREL;
+ else
+ as_bad ("GOTPCREL relocations are supported only in 64bit mode.");
len = 3;
}
else
exp->X_add_symbol = (symbolS *) 0;
exp->X_op_symbol = (symbolS *) 0;
}
-
- if (exp->X_op == O_constant)
+ else if (exp->X_op == O_constant)
{
/* Size it properly later. */
- i.types[this_operand] |= Imm32;
+ i.types[this_operand] |= Imm64;
+ /* If BFD64, sign extend val. */
+ if (!use_rela_relocations)
+ if ((exp->X_add_number & ~(((addressT) 2 << 31) - 1)) == 0)
+ exp->X_add_number = (exp->X_add_number ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
}
#if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
else if (1
{
/* This is an address. The size of the address will be
determined later, depending on destination register,
- suffix, or the default for the section. We exclude
- Imm8S here so that `push $foo' and other instructions
- with an Imm8S form will use Imm16 or Imm32. */
- i.types[this_operand] |= (Imm8 | Imm16 | Imm32);
+ suffix, or the default for the section. */
+ i.types[this_operand] |= Imm8 | Imm16 | Imm32 | Imm32S | Imm64;
}
return 1;
char *save_input_line_pointer;
int bigdisp = Disp32;
- if (flag_16bit_code ^ (i.prefix[ADDR_PREFIX] != 0))
+ if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
bigdisp = Disp16;
+ if (flag_code == CODE_64BIT)
+ bigdisp = Disp64;
i.types[this_operand] |= bigdisp;
exp = &disp_expressions[i.disp_operands];
int first;
/* GOT relocations are not supported in 16 bit mode. */
- if (flag_16bit_code)
+ if (flag_code == CODE_16BIT)
as_bad (_("GOT relocations not supported in 16 bit mode"));
if (GOT_symbol == NULL)
if (strncmp (cp + 1, "PLT", 3) == 0)
{
- i.disp_reloc[this_operand] = BFD_RELOC_386_PLT32;
+ if (flag_code == CODE_64BIT)
+ i.disp_reloc[this_operand] = BFD_RELOC_X86_64_PLT32;
+ else
+ i.disp_reloc[this_operand] = BFD_RELOC_386_PLT32;
len = 3;
}
else if (strncmp (cp + 1, "GOTOFF", 6) == 0)
{
+ if (flag_code == CODE_64BIT)
+ as_bad ("GOTOFF relocation is not supported in 64bit mode.");
i.disp_reloc[this_operand] = BFD_RELOC_386_GOTOFF;
len = 6;
}
else if (strncmp (cp + 1, "GOT", 3) == 0)
{
- i.disp_reloc[this_operand] = BFD_RELOC_386_GOT32;
+ if (flag_code == CODE_64BIT)
+ i.disp_reloc[this_operand] = BFD_RELOC_X86_64_GOT32;
+ else
+ i.disp_reloc[this_operand] = BFD_RELOC_386_GOT32;
+ len = 3;
+ }
+ else if (strncmp (cp + 1, "GOTPCREL", 3) == 0)
+ {
+ if (flag_code != CODE_64BIT)
+ as_bad ("GOTPCREL relocation is supported only in 64bit mode.");
+ i.disp_reloc[this_operand] = BFD_RELOC_X86_64_GOTPCREL;
len = 3;
}
else
/* We do this to make sure that the section symbol is in
the symbol table. We will ultimately change the relocation
to be relative to the beginning of the section. */
- if (i.disp_reloc[this_operand] == BFD_RELOC_386_GOTOFF)
+ if (i.disp_reloc[this_operand] == BFD_RELOC_386_GOTOFF
+ || i.disp_reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
{
if (S_IS_LOCAL(exp->X_add_symbol)
&& S_GET_SEGMENT (exp->X_add_symbol) != undefined_section)
return 0;
}
#endif
+ else if (flag_code == CODE_64BIT)
+ i.types[this_operand] |= Disp32S | Disp32;
return 1;
}
-static int i386_operand_modifier PARAMS ((char **, int));
+static int i386_index_check PARAMS((const char *));
+
+/* Make sure the memory operand we've been dealt is valid.
+ Return 1 on success, 0 on a failure. */
static int
-i386_operand_modifier (op_string, got_a_float)
- char **op_string;
- int got_a_float;
+i386_index_check (operand_string)
+ const char *operand_string;
{
- if (!strncasecmp (*op_string, "BYTE PTR", 8))
- {
- i.suffix = BYTE_MNEM_SUFFIX;
- *op_string += 8;
- return BYTE_PTR;
+ int ok;
+#if INFER_ADDR_PREFIX
+ int fudged = 0;
+ tryprefix:
+#endif
+ ok = 1;
+ if (flag_code == CODE_64BIT)
+ {
+ /* 64bit checks. */
+ if ((i.base_reg
+ && ((i.base_reg->reg_type & Reg64) == 0)
+ && (i.base_reg->reg_type != BaseIndex
+ || i.index_reg))
+ || (i.index_reg
+ && ((i.index_reg->reg_type & (Reg64|BaseIndex))
+ != (Reg64|BaseIndex))))
+ ok = 0;
}
- else if (!strncasecmp (*op_string, "WORD PTR", 8))
+ else
{
- if (got_a_float == 2) /* "fi..." */
- i.suffix = SHORT_MNEM_SUFFIX;
+ if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
+ {
+ /* 16bit checks. */
+ if ((i.base_reg
+ && ((i.base_reg->reg_type & (Reg16|BaseIndex|RegRex))
+ != (Reg16|BaseIndex)))
+ || (i.index_reg
+ && (((i.index_reg->reg_type & (Reg16|BaseIndex))
+ != (Reg16|BaseIndex))
+ || ! (i.base_reg
+ && i.base_reg->reg_num < 6
+ && i.index_reg->reg_num >= 6
+ && i.log2_scale_factor == 0))))
+ ok = 0;
+ }
else
- i.suffix = WORD_MNEM_SUFFIX;
- *op_string += 8;
- return WORD_PTR;
+ {
+ /* 32bit checks. */
+ if ((i.base_reg
+ && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
+ || (i.index_reg
+ && ((i.index_reg->reg_type & (Reg32|BaseIndex|RegRex))
+ != (Reg32|BaseIndex))))
+ ok = 0;
+ }
}
-
- else if (!strncasecmp (*op_string, "DWORD PTR", 9))
+ if (!ok)
{
- if (got_a_float == 1) /* "f..." */
- i.suffix = SHORT_MNEM_SUFFIX;
+#if INFER_ADDR_PREFIX
+ if (flag_code != CODE_64BIT
+ && i.prefix[ADDR_PREFIX] == 0 && stackop_size != '\0')
+ {
+ i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE;
+ i.prefixes += 1;
+ /* Change the size of any displacement too. At most one of
+ Disp16 or Disp32 is set.
+ FIXME. There doesn't seem to be any real need for separate
+ Disp16 and Disp32 flags. The same goes for Imm16 and Imm32.
+ Removing them would probably clean up the code quite a lot. */
+ if (i.types[this_operand] & (Disp16|Disp32))
+ i.types[this_operand] ^= (Disp16|Disp32);
+ fudged = 1;
+ goto tryprefix;
+ }
+ if (fudged)
+ as_bad (_("`%s' is not a valid base/index expression"),
+ operand_string);
else
- i.suffix = LONG_MNEM_SUFFIX;
- *op_string += 9;
- return DWORD_PTR;
+#endif
+ as_bad (_("`%s' is not a valid %s bit base/index expression"),
+ operand_string,
+ flag_code_names[flag_code]);
+ return 0;
}
-
- else if (!strncasecmp (*op_string, "QWORD PTR", 9))
- {
- i.suffix = DWORD_MNEM_SUFFIX;
- *op_string += 9;
- return QWORD_PTR;
- }
-
- else if (!strncasecmp (*op_string, "XWORD PTR", 9))
- {
- i.suffix = LONG_DOUBLE_MNEM_SUFFIX;
- *op_string += 9;
- return XWORD_PTR;
- }
-
- else if (!strncasecmp (*op_string, "SHORT", 5))
- {
- *op_string += 5;
- return SHORT;
- }
-
- else if (!strncasecmp (*op_string, "OFFSET FLAT:", 12))
- {
- *op_string += 12;
- return OFFSET_FLAT;
- }
-
- else if (!strncasecmp (*op_string, "FLAT", 4))
- {
- *op_string += 4;
- return FLAT;
- }
-
- else return NONE_FOUND;
-}
-
-static char * build_displacement_string PARAMS ((int, char *));
-
-static char *
-build_displacement_string (initial_disp, op_string)
- int initial_disp;
- char *op_string;
-{
- char *temp_string = (char *) malloc (strlen (op_string) + 1);
- char *end_of_operand_string;
- char *tc;
- char *temp_disp;
-
- temp_string[0] = '\0';
- tc = end_of_operand_string = strchr (op_string, '[');
- if (initial_disp && !end_of_operand_string)
- {
- strcpy (temp_string, op_string);
- return temp_string;
- }
-
- /* Build the whole displacement string. */
- if (initial_disp)
- {
- strncpy (temp_string, op_string, end_of_operand_string - op_string);
- temp_string[end_of_operand_string - op_string] = '\0';
- temp_disp = tc;
- }
- else
- temp_disp = op_string;
-
- while (*temp_disp != '\0')
- {
- char *end_op;
- int add_minus = (*temp_disp == '-');
-
- if (*temp_disp == '+' || *temp_disp == '-' || *temp_disp == '[')
- temp_disp++;
-
- if (is_space_char (*temp_disp))
- temp_disp++;
-
- /* Don't consider registers. */
- if ( !((*temp_disp == REGISTER_PREFIX || allow_naked_reg)
- && parse_register (temp_disp, &end_op)) )
- {
- char *string_start = temp_disp;
-
- while (*temp_disp != ']'
- && *temp_disp != '+'
- && *temp_disp != '-'
- && *temp_disp != '*')
- ++temp_disp;
-
- if (add_minus)
- strcat (temp_string, "-");
- else
- strcat (temp_string, "+");
-
- strncat (temp_string, string_start, temp_disp - string_start);
- if (*temp_disp == '+' || *temp_disp == '-')
- --temp_disp;
- }
-
- while (*temp_disp != '\0'
- && *temp_disp != '+'
- && *temp_disp != '-')
- ++temp_disp;
- }
-
- return temp_string;
-}
-
-static int i386_parse_seg PARAMS ((char *));
-
-static int
-i386_parse_seg (op_string)
- char *op_string;
-{
- if (is_space_char (*op_string))
- ++op_string;
-
- /* Should be one of es, cs, ss, ds fs or gs. */
- switch (*op_string++)
- {
- case 'e':
- i.seg[i.mem_operands] = &es;
- break;
- case 'c':
- i.seg[i.mem_operands] = &cs;
- break;
- case 's':
- i.seg[i.mem_operands] = &ss;
- break;
- case 'd':
- i.seg[i.mem_operands] = &ds;
- break;
- case 'f':
- i.seg[i.mem_operands] = &fs;
- break;
- case 'g':
- i.seg[i.mem_operands] = &gs;
- break;
- default:
- as_bad (_("bad segment name `%s'"), op_string);
- return 0;
- }
-
- if (*op_string++ != 's')
- {
- as_bad (_("bad segment name `%s'"), op_string);
- return 0;
- }
-
- if (is_space_char (*op_string))
- ++op_string;
-
- if (*op_string != ':')
- {
- as_bad (_("bad segment name `%s'"), op_string);
- return 0;
- }
-
- return 1;
-
-}
-
-static int i386_index_check PARAMS((const char *));
-
-/* Make sure the memory operand we've been dealt is valid.
- Return 1 on success, 0 on a failure. */
-
-static int
-i386_index_check (operand_string)
- const char *operand_string;
-{
-#if INFER_ADDR_PREFIX
- int fudged = 0;
-
- tryprefix:
-#endif
- if (flag_16bit_code ^ (i.prefix[ADDR_PREFIX] != 0)
- /* 16 bit mode checks. */
- ? ((i.base_reg
- && ((i.base_reg->reg_type & (Reg16|BaseIndex))
- != (Reg16|BaseIndex)))
- || (i.index_reg
- && (((i.index_reg->reg_type & (Reg16|BaseIndex))
- != (Reg16|BaseIndex))
- || ! (i.base_reg
- && i.base_reg->reg_num < 6
- && i.index_reg->reg_num >= 6
- && i.log2_scale_factor == 0))))
- /* 32 bit mode checks. */
- : ((i.base_reg
- && (i.base_reg->reg_type & Reg32) == 0)
- || (i.index_reg
- && ((i.index_reg->reg_type & (Reg32|BaseIndex))
- != (Reg32|BaseIndex)))))
- {
-#if INFER_ADDR_PREFIX
- if (i.prefix[ADDR_PREFIX] == 0 && stackop_size != '\0')
- {
- i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE;
- i.prefixes += 1;
- /* Change the size of any displacement too. At most one of
- Disp16 or Disp32 is set.
- FIXME. There doesn't seem to be any real need for separate
- Disp16 and Disp32 flags. The same goes for Imm16 and Imm32.
- Removing them would probably clean up the code quite a lot. */
- if (i.types[this_operand] & (Disp16|Disp32))
- i.types[this_operand] ^= (Disp16|Disp32);
- fudged = 1;
- goto tryprefix;
- }
- if (fudged)
- as_bad (_("`%s' is not a valid base/index expression"),
- operand_string);
- else
-#endif
- as_bad (_("`%s' is not a valid %s bit base/index expression"),
- operand_string,
- flag_16bit_code ^ (i.prefix[ADDR_PREFIX] != 0) ? "16" : "32");
- return 0;
- }
- return 1;
-}
-
-static int i386_intel_memory_operand PARAMS ((char *));
-
-static int
-i386_intel_memory_operand (operand_string)
- char *operand_string;
-{
- char *op_string = operand_string;
- char *end_of_operand_string;
-
- if ((i.mem_operands == 1
- && (current_templates->start->opcode_modifier & IsString) == 0)
- || i.mem_operands == 2)
- {
- as_bad (_("too many memory references for `%s'"),
- current_templates->start->name);
- return 0;
- }
-
- /* First check for a segment override. */
- if (*op_string != '[')
- {
- char *end_seg;
-
- end_seg = strchr (op_string, ':');
- if (end_seg)
- {
- if (!i386_parse_seg (op_string))
- return 0;
- op_string = end_seg + 1;
- }
- }
-
- /* Look for displacement preceding open bracket. */
- if (*op_string != '[')
- {
- char *temp_string;
-
- if (i.disp_operands)
- return 0;
-
- temp_string = build_displacement_string (true, op_string);
-
- if (!i386_displacement (temp_string, temp_string + strlen (temp_string)))
- {
- free (temp_string);
- return 0;
- }
- free (temp_string);
-
- end_of_operand_string = strchr (op_string, '[');
- if (!end_of_operand_string)
- end_of_operand_string = op_string + strlen (op_string);
-
- if (is_space_char (*end_of_operand_string))
- --end_of_operand_string;
-
- op_string = end_of_operand_string;
- }
-
- if (*op_string == '[')
- {
- ++op_string;
-
- /* Pick off each component and figure out where it belongs */
-
- end_of_operand_string = op_string;
-
- while (*op_string != ']')
- {
- const reg_entry *temp_reg;
- char *end_op;
- char *temp_string;
-
- while (*end_of_operand_string != '+'
- && *end_of_operand_string != '-'
- && *end_of_operand_string != '*'
- && *end_of_operand_string != ']')
- end_of_operand_string++;
-
- temp_string = op_string;
- if (*temp_string == '+')
- {
- ++temp_string;
- if (is_space_char (*temp_string))
- ++temp_string;
- }
-
- if ((*temp_string == REGISTER_PREFIX || allow_naked_reg)
- && (temp_reg = parse_register (temp_string, &end_op)) != NULL)
- {
- if (i.base_reg == NULL)
- i.base_reg = temp_reg;
- else
- i.index_reg = temp_reg;
-
- i.types[this_operand] |= BaseIndex;
- }
- else if (*temp_string == REGISTER_PREFIX)
- {
- as_bad (_("bad register name `%s'"), temp_string);
- return 0;
- }
- else if (is_digit_char (*op_string)
- || *op_string == '+' || *op_string == '-')
- {
- char *temp_str;
-
- if (i.disp_operands != 0)
- return 0;
-
- temp_string = build_displacement_string (false, op_string);
-
- temp_str = temp_string;
- if (*temp_str == '+')
- ++temp_str;
-
- if (!i386_displacement (temp_str, temp_str + strlen (temp_str)))
- {
- free (temp_string);
- return 0;
- }
- free (temp_string);
-
- ++op_string;
- end_of_operand_string = op_string;
- while (*end_of_operand_string != ']'
- && *end_of_operand_string != '+'
- && *end_of_operand_string != '-'
- && *end_of_operand_string != '*')
- ++end_of_operand_string;
- }
- else if (*op_string == '*')
- {
- ++op_string;
-
- if (i.base_reg && !i.index_reg)
- {
- i.index_reg = i.base_reg;
- i.base_reg = 0;
- }
-
- if (!i386_scale (op_string))
- return 0;
- }
- op_string = end_of_operand_string;
- ++end_of_operand_string;
- }
- }
-
- if (i386_index_check (operand_string) == 0)
- return 0;
-
- i.mem_operands++;
- return 1;
-}
-
-static int
-i386_intel_operand (operand_string, got_a_float)
- char *operand_string;
- int got_a_float;
-{
- const reg_entry *r;
- char *end_op;
- char *op_string = operand_string;
-
- int operand_modifier = i386_operand_modifier (&op_string, got_a_float);
- if (is_space_char (*op_string))
- ++op_string;
-
- switch (operand_modifier)
- {
- case BYTE_PTR:
- case WORD_PTR:
- case DWORD_PTR:
- case QWORD_PTR:
- case XWORD_PTR:
- if (!i386_intel_memory_operand (op_string))
- return 0;
- break;
-
- case FLAT:
- case OFFSET_FLAT:
- if (!i386_immediate (op_string))
- return 0;
- break;
-
- case SHORT:
- case NONE_FOUND:
- /* Should be register or immediate */
- if (is_digit_char (*op_string)
- && strchr (op_string, '[') == 0)
- {
- if (!i386_immediate (op_string))
- return 0;
- }
- else if ((*op_string == REGISTER_PREFIX || allow_naked_reg)
- && (r = parse_register (op_string, &end_op)) != NULL)
- {
- /* Check for a segment override by searching for ':' after a
- segment register. */
- op_string = end_op;
- if (is_space_char (*op_string))
- ++op_string;
- if (*op_string == ':' && (r->reg_type & (SReg2 | SReg3)))
- {
- switch (r->reg_num)
- {
- case 0:
- i.seg[i.mem_operands] = &es;
- break;
- case 1:
- i.seg[i.mem_operands] = &cs;
- break;
- case 2:
- i.seg[i.mem_operands] = &ss;
- break;
- case 3:
- i.seg[i.mem_operands] = &ds;
- break;
- case 4:
- i.seg[i.mem_operands] = &fs;
- break;
- case 5:
- i.seg[i.mem_operands] = &gs;
- break;
- }
-
- }
- i.types[this_operand] |= r->reg_type & ~BaseIndex;
- i.op[this_operand].regs = r;
- i.reg_operands++;
- }
- else if (*op_string == REGISTER_PREFIX)
- {
- as_bad (_("bad register name `%s'"), op_string);
- return 0;
- }
- else if (!i386_intel_memory_operand (op_string))
- return 0;
-
- break;
- } /* end switch */
-
- return 1;
-}
+ return 1;
+}
/* Parse OPERAND_STRING into the i386_insn structure I. Returns non-zero
on error. */
return 0;
}
else if (*op_string == IMMEDIATE_PREFIX)
- { /* ... or an immediate */
+ {
++op_string;
if (i.types[this_operand] & JumpAbsolute)
{
i.mem_operands++;
}
else
- { /* It's not a memory operand; argh! */
+ {
+ /* It's not a memory operand; argh! */
as_bad (_("invalid char %s beginning operand %d `%s'"),
output_invalid (*op_string),
this_operand + 1,
{
case JUMP_PC_RELATIVE:
/* Make jmp (0xeb) a dword displacement jump. */
- /* dword disp jmp */
opcode[0] = 0xe9;
fragP->fr_fix += size;
fix_new (fragP, old_fr_fix, size,
switch (fixP->fx_r_type)
{
case BFD_RELOC_386_PLT32:
+ case BFD_RELOC_X86_64_PLT32:
/* Make the jump instruction point to the address of the operand. At
runtime we merely add the offset to the actual PLT entry. */
value = -4;
value -= 1;
break;
case BFD_RELOC_386_GOT32:
+ case BFD_RELOC_X86_64_GOT32:
value = 0; /* Fully resolved at runtime. No addend. */
break;
case BFD_RELOC_386_GOTOFF:
+ case BFD_RELOC_X86_64_GOTPCREL:
break;
case BFD_RELOC_VTABLE_INHERIT:
#endif /* defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) */
*valp = value;
#endif /* defined (BFD_ASSEMBLER) && !defined (TE_Mach) */
+
+#ifndef BFD_ASSEMBLER
+ md_number_to_chars (p, value, fixP->fx_size);
+#else
+ /* Are we finished with this relocation now? */
+ if (fixP->fx_addsy == 0 && fixP->fx_pcrel == 0)
+ fixP->fx_done = 1;
+ else if (use_rela_relocations)
+ {
+ fixP->fx_no_overflow = 1;
+ value = 0;
+ }
md_number_to_chars (p, value, fixP->fx_size);
+#endif
return 1;
}
s++;
}
+ /* For naked regs, make sure that we are not dealing with an identifier.
+ This prevents confusing an identifier like `eax_var' with register
+ `eax'. */
+ if (allow_naked_reg && identifier_chars[(unsigned char) *s])
+ return (const reg_entry *) NULL;
+
*end_op = s;
r = (const reg_entry *) hash_find (reg_hash, reg_name_given);
const char *md_shortopts = "q";
#endif
struct option md_longopts[] = {
+#define OPTION_32 (OPTION_MD_BASE + 0)
+ {"32", no_argument, NULL, OPTION_32},
+#define OPTION_64 (OPTION_MD_BASE + 1)
+ {"64", no_argument, NULL, OPTION_64},
{NULL, no_argument, NULL, 0}
};
size_t md_longopts_size = sizeof (md_longopts);
.stab instead of .stab.excl. We always use .stab anyhow. */
break;
#endif
+#ifdef OBJ_ELF
+ case OPTION_32:
+ case OPTION_64:
+ {
+ const char **list, **l;
+
+ default_arch = c == OPTION_32 ? "i386" : "x86_64";
+ list = bfd_target_list ();
+ for (l = list; *l != NULL; l++)
+ {
+ if (c == OPTION_32)
+ {
+ if (strcmp (*l, "elf32-i386") == 0)
+ break;
+ }
+ else
+ {
+ if (strcmp (*l, "elf64-x86-64") == 0)
+ break;
+ }
+ }
+ if (*l == NULL)
+ as_fatal (_("No compiled in support for %d bit object file format"),
+ c == OPTION_32 ? 32 : 64);
+ free (list);
+ }
+ break;
+#endif
default:
return 0;
}
#ifdef BFD_ASSEMBLER
-#if ((defined (OBJ_MAYBE_ELF) && defined (OBJ_MAYBE_COFF)) \
- || (defined (OBJ_MAYBE_ELF) && defined (OBJ_MAYBE_AOUT)) \
- || (defined (OBJ_MAYBE_COFF) && defined (OBJ_MAYBE_AOUT)))
+#if ((defined (OBJ_MAYBE_COFF) && defined (OBJ_MAYBE_AOUT)) \
+ || defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
/* Pick the target format to use. */
const char *
i386_target_format ()
{
+ if (!strcmp (default_arch, "x86_64"))
+ set_code_flag (CODE_64BIT);
+ else if (!strcmp (default_arch, "i386"))
+ set_code_flag (CODE_32BIT);
+ else
+ as_fatal (_("Unknown architecture"));
switch (OUTPUT_FLAVOR)
{
#ifdef OBJ_MAYBE_AOUT
case bfd_target_coff_flavour:
return "coff-i386";
#endif
-#ifdef OBJ_MAYBE_ELF
+#if defined (OBJ_MAYBE_ELF) || defined (OBJ_ELF)
case bfd_target_elf_flavour:
- return "elf32-i386";
+ {
+ if (flag_code == CODE_64BIT)
+ use_rela_relocations = 1;
+ return flag_code == CODE_64BIT ? "elf64-x86-64" : "elf32-i386";
+ }
#endif
default:
abort ();
{
if (fixp->fx_subsy && fixp->fx_subsy == GOT_symbol)
{
+ /* GOTOFF relocation are nonsense in 64bit mode. */
+ if (flag_code == CODE_64BIT)
+ abort();
fixp->fx_r_type = BFD_RELOC_386_GOTOFF;
fixp->fx_subsy = 0;
}
switch (fixp->fx_r_type)
{
+ case BFD_RELOC_X86_64_PLT32:
+ case BFD_RELOC_X86_64_GOT32:
+ case BFD_RELOC_X86_64_GOTPCREL:
case BFD_RELOC_386_PLT32:
case BFD_RELOC_386_GOT32:
case BFD_RELOC_386_GOTOFF:
case BFD_RELOC_386_GOTPC:
+ case BFD_RELOC_X86_64_32S:
case BFD_RELOC_RVA:
case BFD_RELOC_VTABLE_ENTRY:
case BFD_RELOC_VTABLE_INHERIT:
case 1: code = BFD_RELOC_8; break;
case 2: code = BFD_RELOC_16; break;
case 4: code = BFD_RELOC_32; break;
+ case 8: code = BFD_RELOC_64; break;
}
}
break;
if (code == BFD_RELOC_32
&& GOT_symbol
&& fixp->fx_addsy == GOT_symbol)
- code = BFD_RELOC_386_GOTPC;
+ {
+ /* We don't support GOTPC on 64bit targets. */
+ if (flag_code == CODE_64BIT)
+ abort();
+ code = BFD_RELOC_386_GOTPC;
+ }
rel = (arelent *) xmalloc (sizeof (arelent));
rel->sym_ptr_ptr = (asymbol **) xmalloc (sizeof (asymbol *));
*rel->sym_ptr_ptr = symbol_get_bfdsym (fixp->fx_addsy);
rel->address = fixp->fx_frag->fr_address + fixp->fx_where;
- /* HACK: Since i386 ELF uses Rel instead of Rela, encode the
- vtable entry to be used in the relocation's section offset. */
- if (fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
- rel->address = fixp->fx_offset;
+ if (!use_rela_relocations)
+ {
+ /* HACK: Since i386 ELF uses Rel instead of Rela, encode the
+ vtable entry to be used in the relocation's section offset. */
+ if (fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
+ rel->address = fixp->fx_offset;
- if (fixp->fx_pcrel)
- rel->addend = fixp->fx_addnumber;
+ if (fixp->fx_pcrel)
+ rel->addend = fixp->fx_addnumber;
+ else
+ rel->addend = 0;
+ }
+ /* Use the rela in 64bit mode. */
else
- rel->addend = 0;
+ {
+ rel->addend = fixp->fx_offset;
+#ifdef OBJ_ELF
+ /* Ohhh, this is ugly. The problem is that if this is a local global
+ symbol, the relocation will entirely be performed at link time, not
+ at assembly time. bfd_perform_reloc doesn't know about this sort
+ of thing, and as a result we need to fake it out here. */
+ if ((S_IS_EXTERN (fixp->fx_addsy) || S_IS_WEAK (fixp->fx_addsy))
+ && !S_IS_COMMON(fixp->fx_addsy))
+ rel->addend -= symbol_get_bfdsym (fixp->fx_addsy)->value;
+#endif
+ if (fixp->fx_pcrel)
+ rel->addend -= fixp->fx_size;
+ }
+
rel->howto = bfd_reloc_type_lookup (stdoutput, code);
if (rel->howto == NULL)
#endif /* I386COFF */
#endif /* ! BFD_ASSEMBLER */
+\f
+/* Parse operands using Intel syntax. This implements a recursive descent
+ parser based on the BNF grammar published in Appendix B of the MASM 6.1
+ Programmer's Guide.
+
+ FIXME: We do not recognize the full operand grammar defined in the MASM
+ documentation. In particular, all the structure/union and
+ high-level macro operands are missing.
+
+ Uppercase words are terminals, lower case words are non-terminals.
+ Objects surrounded by double brackets '[[' ']]' are optional. Vertical
+ bars '|' denote choices. Most grammar productions are implemented in
+ functions called 'intel_<production>'.
+
+ Initial production is 'expr'.
+
+ addOp + | -
+
+ alpha [a-zA-Z]
+
+ byteRegister AL | AH | BL | BH | CL | CH | DL | DH
+
+ constant digits [[ radixOverride ]]
+
+ dataType BYTE | WORD | DWORD | QWORD | XWORD
+
+ digits decdigit
+ | digits decdigit
+ | digits hexdigit
+
+ decdigit [0-9]
+
+ e05 e05 addOp e06
+ | e06
+
+ e06 e06 mulOp e09
+ | e09
+
+ e09 OFFSET e10
+ | e09 PTR e10
+ | e09 : e10
+ | e10
+
+ e10 e10 [ expr ]
+ | e11
+
+ e11 ( expr )
+ | [ expr ]
+ | constant
+ | dataType
+ | id
+ | $
+ | register
+
+ => expr SHORT e05
+ | e05
+
+ gpRegister AX | EAX | BX | EBX | CX | ECX | DX | EDX
+ | BP | EBP | SP | ESP | DI | EDI | SI | ESI
+
+ hexdigit a | b | c | d | e | f
+ | A | B | C | D | E | F
+
+ id alpha
+ | id alpha
+ | id decdigit
+
+ mulOp * | / | MOD
+
+ quote " | '
+
+ register specialRegister
+ | gpRegister
+ | byteRegister
+
+ segmentRegister CS | DS | ES | FS | GS | SS
+
+ specialRegister CR0 | CR2 | CR3
+ | DR0 | DR1 | DR2 | DR3 | DR6 | DR7
+ | TR3 | TR4 | TR5 | TR6 | TR7
+
+ We simplify the grammar in obvious places (e.g., register parsing is
+ done by calling parse_register) and eliminate immediate left recursion
+ to implement a recursive-descent parser.
+
+ expr SHORT e05
+ | e05
+
+ e05 e06 e05'
+
+ e05' addOp e06 e05'
+ | Empty
+
+ e06 e09 e06'
+
+ e06' mulOp e09 e06'
+ | Empty
+
+ e09 OFFSET e10 e09'
+ | e10 e09'
+
+ e09' PTR e10 e09'
+ | : e10 e09'
+ | Empty
+
+ e10 e11 e10'
+
+ e10' [ expr ] e10'
+ | Empty
+
+ e11 ( expr )
+ | [ expr ]
+ | BYTE
+ | WORD
+ | DWORD
+ | QWORD
+ | XWORD
+ | .
+ | $
+ | register
+ | id
+ | constant */
+
+/* Parsing structure for the intel syntax parser. Used to implement the
+ semantic actions for the operand grammar. */
+struct intel_parser_s
+ {
+ char *op_string; /* The string being parsed. */
+ int got_a_float; /* Whether the operand is a float. */
+ int op_modifier; /* Operand modifier. */
+ int is_mem; /* 1 if operand is memory reference. */
+ const reg_entry *reg; /* Last register reference found. */
+ char *disp; /* Displacement string being built. */
+ };
+
+static struct intel_parser_s intel_parser;
+
+/* Token structure for parsing intel syntax. */
+struct intel_token
+ {
+ int code; /* Token code. */
+ const reg_entry *reg; /* Register entry for register tokens. */
+ char *str; /* String representation. */
+ };
+
+static struct intel_token cur_token, prev_token;
+
+
+/* Token codes for the intel parser. Since T_SHORT is already used
+ by COFF, undefine it first to prevent a warning. */
+#define T_NIL -1
+#define T_CONST 1
+#define T_REG 2
+#define T_BYTE 3
+#define T_WORD 4
+#define T_DWORD 5
+#define T_QWORD 6
+#define T_XWORD 7
+#undef T_SHORT
+#define T_SHORT 8
+#define T_OFFSET 9
+#define T_PTR 10
+#define T_ID 11
+
+/* Prototypes for intel parser functions. */
+static int intel_match_token PARAMS ((int code));
+static void intel_get_token PARAMS ((void));
+static void intel_putback_token PARAMS ((void));
+static int intel_expr PARAMS ((void));
+static int intel_e05 PARAMS ((void));
+static int intel_e05_1 PARAMS ((void));
+static int intel_e06 PARAMS ((void));
+static int intel_e06_1 PARAMS ((void));
+static int intel_e09 PARAMS ((void));
+static int intel_e09_1 PARAMS ((void));
+static int intel_e10 PARAMS ((void));
+static int intel_e10_1 PARAMS ((void));
+static int intel_e11 PARAMS ((void));
+
+static int
+i386_intel_operand (operand_string, got_a_float)
+ char *operand_string;
+ int got_a_float;
+{
+ int ret;
+ char *p;
+
+ /* Initialize token holders. */
+ cur_token.code = prev_token.code = T_NIL;
+ cur_token.reg = prev_token.reg = NULL;
+ cur_token.str = prev_token.str = NULL;
+
+ /* Initialize parser structure. */
+ p = intel_parser.op_string = (char *)malloc (strlen (operand_string) + 1);
+ if (p == NULL)
+ abort ();
+ strcpy (intel_parser.op_string, operand_string);
+ intel_parser.got_a_float = got_a_float;
+ intel_parser.op_modifier = -1;
+ intel_parser.is_mem = 0;
+ intel_parser.reg = NULL;
+ intel_parser.disp = (char *)malloc (strlen (operand_string) + 1);
+ if (intel_parser.disp == NULL)
+ abort ();
+ intel_parser.disp[0] = '\0';
+
+ /* Read the first token and start the parser. */
+ intel_get_token ();
+ ret = intel_expr ();
+
+ if (ret)
+ {
+ /* If we found a memory reference, hand it over to i386_displacement
+ to fill in the rest of the operand fields. */
+ if (intel_parser.is_mem)
+ {
+ if ((i.mem_operands == 1
+ && (current_templates->start->opcode_modifier & IsString) == 0)
+ || i.mem_operands == 2)
+ {
+ as_bad (_("too many memory references for '%s'"),
+ current_templates->start->name);
+ ret = 0;
+ }
+ else
+ {
+ char *s = intel_parser.disp;
+ i.mem_operands++;
+
+ /* Add the displacement expression. */
+ if (*s != '\0')
+ ret = i386_displacement (s, s + strlen (s))
+ && i386_index_check (s);
+ }
+ }
+
+ /* Constant and OFFSET expressions are handled by i386_immediate. */
+ else if (intel_parser.op_modifier == OFFSET_FLAT
+ || intel_parser.reg == NULL)
+ ret = i386_immediate (intel_parser.disp);
+ }
+
+ free (p);
+ free (intel_parser.disp);
+
+ return ret;
+}
+
+/* expr SHORT e05
+ | e05 */
+static int
+intel_expr ()
+{
+ /* expr SHORT e05 */
+ if (cur_token.code == T_SHORT)
+ {
+ intel_parser.op_modifier = SHORT;
+ intel_match_token (T_SHORT);
+
+ return (intel_e05 ());
+ }
+
+ /* expr e05 */
+ else
+ return intel_e05 ();
+}
+
+/* e05 e06 e05'
+
+ e05' addOp e06 e05'
+ | Empty */
+static int
+intel_e05 ()
+{
+ return (intel_e06 () && intel_e05_1 ());
+}
+
+static int
+intel_e05_1 ()
+{
+ /* e05' addOp e06 e05' */
+ if (cur_token.code == '+' || cur_token.code == '-')
+ {
+ strcat (intel_parser.disp, cur_token.str);
+ intel_match_token (cur_token.code);
+
+ return (intel_e06 () && intel_e05_1 ());
+ }
+
+ /* e05' Empty */
+ else
+ return 1;
+}
+
+/* e06 e09 e06'
+
+ e06' mulOp e09 e06'
+ | Empty */
+static int
+intel_e06 ()
+{
+ return (intel_e09 () && intel_e06_1 ());
+}
+
+static int
+intel_e06_1 ()
+{
+ /* e06' mulOp e09 e06' */
+ if (cur_token.code == '*' || cur_token.code == '/')
+ {
+ strcat (intel_parser.disp, cur_token.str);
+ intel_match_token (cur_token.code);
+
+ return (intel_e09 () && intel_e06_1 ());
+ }
+
+ /* e06' Empty */
+ else
+ return 1;
+}
+
+/* e09 OFFSET e10 e09'
+ | e10 e09'
+
+ e09' PTR e10 e09'
+ | : e10 e09'
+ | Empty */
+static int
+intel_e09 ()
+{
+ /* e09 OFFSET e10 e09' */
+ if (cur_token.code == T_OFFSET)
+ {
+ intel_parser.is_mem = 0;
+ intel_parser.op_modifier = OFFSET_FLAT;
+ intel_match_token (T_OFFSET);
+
+ return (intel_e10 () && intel_e09_1 ());
+ }
+
+ /* e09 e10 e09' */
+ else
+ return (intel_e10 () && intel_e09_1 ());
+}
+
+static int
+intel_e09_1 ()
+{
+ /* e09' PTR e10 e09' */
+ if (cur_token.code == T_PTR)
+ {
+ if (prev_token.code == T_BYTE)
+ i.suffix = BYTE_MNEM_SUFFIX;
+
+ else if (prev_token.code == T_WORD)
+ {
+ if (intel_parser.got_a_float == 2) /* "fi..." */
+ i.suffix = SHORT_MNEM_SUFFIX;
+ else
+ i.suffix = WORD_MNEM_SUFFIX;
+ }
+
+ else if (prev_token.code == T_DWORD)
+ {
+ if (intel_parser.got_a_float == 1) /* "f..." */
+ i.suffix = SHORT_MNEM_SUFFIX;
+ else
+ i.suffix = LONG_MNEM_SUFFIX;
+ }
+
+ else if (prev_token.code == T_QWORD)
+ {
+ if (intel_parser.got_a_float == 1) /* "f..." */
+ i.suffix = LONG_MNEM_SUFFIX;
+ else
+ i.suffix = QWORD_MNEM_SUFFIX;
+ }
+
+ else if (prev_token.code == T_XWORD)
+ i.suffix = LONG_DOUBLE_MNEM_SUFFIX;
+
+ else
+ {
+ as_bad (_("Unknown operand modifier `%s'\n"), prev_token.str);
+ return 0;
+ }
+
+ intel_match_token (T_PTR);
+
+ return (intel_e10 () && intel_e09_1 ());
+ }
+
+ /* e09 : e10 e09' */
+ else if (cur_token.code == ':')
+ {
+ /* Mark as a memory operand only if it's not already known to be an
+ offset expression. */
+ if (intel_parser.op_modifier != OFFSET_FLAT)
+ intel_parser.is_mem = 1;
+
+ return (intel_match_token (':') && intel_e10 () && intel_e09_1 ());
+ }
+
+ /* e09' Empty */
+ else
+ return 1;
+}
+
+/* e10 e11 e10'
+
+ e10' [ expr ] e10'
+ | Empty */
+static int
+intel_e10 ()
+{
+ return (intel_e11 () && intel_e10_1 ());
+}
+
+static int
+intel_e10_1 ()
+{
+ /* e10' [ expr ] e10' */
+ if (cur_token.code == '[')
+ {
+ intel_match_token ('[');
+
+ /* Mark as a memory operand only if it's not already known to be an
+ offset expression. If it's an offset expression, we need to keep
+ the brace in. */
+ if (intel_parser.op_modifier != OFFSET_FLAT)
+ intel_parser.is_mem = 1;
+ else
+ strcat (intel_parser.disp, "[");
+
+ /* Add a '+' to the displacement string if necessary. */
+ if (*intel_parser.disp != '\0'
+ && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
+ strcat (intel_parser.disp, "+");
+
+ if (intel_expr () && intel_match_token (']'))
+ {
+ /* Preserve brackets when the operand is an offset expression. */
+ if (intel_parser.op_modifier == OFFSET_FLAT)
+ strcat (intel_parser.disp, "]");
+
+ return intel_e10_1 ();
+ }
+ else
+ return 0;
+ }
+
+ /* e10' Empty */
+ else
+ return 1;
+}
+
+/* e11 ( expr )
+ | [ expr ]
+ | BYTE
+ | WORD
+ | DWORD
+ | QWORD
+ | XWORD
+ | $
+ | .
+ | register
+ | id
+ | constant */
+static int
+intel_e11 ()
+{
+ /* e11 ( expr ) */
+ if (cur_token.code == '(')
+ {
+ intel_match_token ('(');
+ strcat (intel_parser.disp, "(");
+
+ if (intel_expr () && intel_match_token (')'))
+ {
+ strcat (intel_parser.disp, ")");
+ return 1;
+ }
+ else
+ return 0;
+ }
+
+ /* e11 [ expr ] */
+ else if (cur_token.code == '[')
+ {
+ intel_match_token ('[');
+
+ /* Mark as a memory operand only if it's not already known to be an
+ offset expression. If it's an offset expression, we need to keep
+ the brace in. */
+ if (intel_parser.op_modifier != OFFSET_FLAT)
+ intel_parser.is_mem = 1;
+ else
+ strcat (intel_parser.disp, "[");
+
+ /* Operands for jump/call inside brackets denote absolute addresses. */
+ if (current_templates->start->opcode_modifier & Jump
+ || current_templates->start->opcode_modifier & JumpDword
+ || current_templates->start->opcode_modifier & JumpByte
+ || current_templates->start->opcode_modifier & JumpInterSegment)
+ i.types[this_operand] |= JumpAbsolute;
+
+ /* Add a '+' to the displacement string if necessary. */
+ if (*intel_parser.disp != '\0'
+ && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
+ strcat (intel_parser.disp, "+");
+
+ if (intel_expr () && intel_match_token (']'))
+ {
+ /* Preserve brackets when the operand is an offset expression. */
+ if (intel_parser.op_modifier == OFFSET_FLAT)
+ strcat (intel_parser.disp, "]");
+
+ return 1;
+ }
+ else
+ return 0;
+ }
+
+ /* e11 BYTE
+ | WORD
+ | DWORD
+ | QWORD
+ | XWORD */
+ else if (cur_token.code == T_BYTE
+ || cur_token.code == T_WORD
+ || cur_token.code == T_DWORD
+ || cur_token.code == T_QWORD
+ || cur_token.code == T_XWORD)
+ {
+ intel_match_token (cur_token.code);
+
+ return 1;
+ }
+
+ /* e11 $
+ | . */
+ else if (cur_token.code == '$' || cur_token.code == '.')
+ {
+ strcat (intel_parser.disp, cur_token.str);
+ intel_match_token (cur_token.code);
+
+ /* Mark as a memory operand only if it's not already known to be an
+ offset expression. */
+ if (intel_parser.op_modifier != OFFSET_FLAT)
+ intel_parser.is_mem = 1;
+
+ return 1;
+ }
+
+ /* e11 register */
+ else if (cur_token.code == T_REG)
+ {
+ const reg_entry *reg = intel_parser.reg = cur_token.reg;
+
+ intel_match_token (T_REG);
+
+ /* Check for segment change. */
+ if (cur_token.code == ':')
+ {
+ if (reg->reg_type & (SReg2 | SReg3))
+ {
+ switch (reg->reg_num)
+ {
+ case 0:
+ i.seg[i.mem_operands] = &es;
+ break;
+ case 1:
+ i.seg[i.mem_operands] = &cs;
+ break;
+ case 2:
+ i.seg[i.mem_operands] = &ss;
+ break;
+ case 3:
+ i.seg[i.mem_operands] = &ds;
+ break;
+ case 4:
+ i.seg[i.mem_operands] = &fs;
+ break;
+ case 5:
+ i.seg[i.mem_operands] = &gs;
+ break;
+ }
+ }
+ else
+ {
+ as_bad (_("`%s' is not a valid segment register"), reg->reg_name);
+ return 0;
+ }
+ }
+
+ /* Not a segment register. Check for register scaling. */
+ else if (cur_token.code == '*')
+ {
+ if (!intel_parser.is_mem)
+ {
+ as_bad (_("Register scaling only allowed in memory operands."));
+ return 0;
+ }
+
+ /* What follows must be a valid scale. */
+ if (intel_match_token ('*')
+ && strchr ("01248", *cur_token.str))
+ {
+ i.index_reg = reg;
+ i.types[this_operand] |= BaseIndex;
+
+ /* Set the scale after setting the register (otherwise,
+ i386_scale will complain) */
+ i386_scale (cur_token.str);
+ intel_match_token (T_CONST);
+ }
+ else
+ {
+ as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
+ cur_token.str);
+ return 0;
+ }
+ }
+
+ /* No scaling. If this is a memory operand, the register is either a
+ base register (first occurrence) or an index register (second
+ occurrence). */
+ else if (intel_parser.is_mem && !(reg->reg_type & (SReg2 | SReg3)))
+ {
+ if (i.base_reg && i.index_reg)
+ {
+ as_bad (_("Too many register references in memory operand.\n"));
+ return 0;
+ }
+
+ if (i.base_reg == NULL)
+ i.base_reg = reg;
+ else
+ i.index_reg = reg;
+
+ i.types[this_operand] |= BaseIndex;
+ }
+
+ /* Offset modifier. Add the register to the displacement string to be
+ parsed as an immediate expression after we're done. */
+ else if (intel_parser.op_modifier == OFFSET_FLAT)
+ strcat (intel_parser.disp, reg->reg_name);
+
+ /* It's neither base nor index nor offset. */
+ else
+ {
+ i.types[this_operand] |= reg->reg_type & ~BaseIndex;
+ i.op[this_operand].regs = reg;
+ i.reg_operands++;
+ }
+
+ /* Since registers are not part of the displacement string (except
+ when we're parsing offset operands), we may need to remove any
+ preceding '+' from the displacement string. */
+ if (*intel_parser.disp != '\0'
+ && intel_parser.op_modifier != OFFSET_FLAT)
+ {
+ char *s = intel_parser.disp;
+ s += strlen (s) - 1;
+ if (*s == '+')
+ *s = '\0';
+ }
+
+ return 1;
+ }
+
+ /* e11 id */
+ else if (cur_token.code == T_ID)
+ {
+ /* Add the identifier to the displacement string. */
+ strcat (intel_parser.disp, cur_token.str);
+ intel_match_token (T_ID);
+
+ /* The identifier represents a memory reference only if it's not
+ preceded by an offset modifier. */
+ if (intel_parser.op_modifier != OFFSET_FLAT)
+ intel_parser.is_mem = 1;
+
+ return 1;
+ }
+
+ /* e11 constant */
+ else if (cur_token.code == T_CONST
+ || cur_token.code == '-'
+ || cur_token.code == '+')
+ {
+ char *save_str;
+
+ /* Allow constants that start with `+' or `-'. */
+ if (cur_token.code == '-' || cur_token.code == '+')
+ {
+ strcat (intel_parser.disp, cur_token.str);
+ intel_match_token (cur_token.code);
+ if (cur_token.code != T_CONST)
+ {
+ as_bad (_("Syntax error. Expecting a constant. Got `%s'.\n"),
+ cur_token.str);
+ return 0;
+ }
+ }
+
+ save_str = (char *)malloc (strlen (cur_token.str) + 1);
+ if (save_str == NULL)
+ abort ();
+ strcpy (save_str, cur_token.str);
+
+ /* Get the next token to check for register scaling. */
+ intel_match_token (cur_token.code);
+
+ /* Check if this constant is a scaling factor for an index register. */
+ if (cur_token.code == '*')
+ {
+ if (intel_match_token ('*') && cur_token.code == T_REG)
+ {
+ if (!intel_parser.is_mem)
+ {
+ as_bad (_("Register scaling only allowed in memory operands."));
+ return 0;
+ }
+
+ /* The constant is followed by `* reg', so it must be
+ a valid scale. */
+ if (strchr ("01248", *save_str))
+ {
+ i.index_reg = cur_token.reg;
+ i.types[this_operand] |= BaseIndex;
+
+ /* Set the scale after setting the register (otherwise,
+ i386_scale will complain) */
+ i386_scale (save_str);
+ intel_match_token (T_REG);
+
+ /* Since registers are not part of the displacement
+ string, we may need to remove any preceding '+' from
+ the displacement string. */
+ if (*intel_parser.disp != '\0')
+ {
+ char *s = intel_parser.disp;
+ s += strlen (s) - 1;
+ if (*s == '+')
+ *s = '\0';
+ }
+
+ free (save_str);
+
+ return 1;
+ }
+ else
+ return 0;
+ }
+
+ /* The constant was not used for register scaling. Since we have
+ already consumed the token following `*' we now need to put it
+ back in the stream. */
+ else
+ intel_putback_token ();
+ }
+
+ /* Add the constant to the displacement string. */
+ strcat (intel_parser.disp, save_str);
+ free (save_str);
+
+ return 1;
+ }
+
+ as_bad (_("Unrecognized token '%s'"), cur_token.str);
+ return 0;
+}
+
+/* Match the given token against cur_token. If they match, read the next
+ token from the operand string. */
+static int
+intel_match_token (code)
+ int code;
+{
+ if (cur_token.code == code)
+ {
+ intel_get_token ();
+ return 1;
+ }
+ else
+ {
+ as_bad (_("Unexpected token `%s'\n"), cur_token.str);
+ return 0;
+ }
+}
+
+/* Read a new token from intel_parser.op_string and store it in cur_token. */
+static void
+intel_get_token ()
+{
+ char *end_op;
+ const reg_entry *reg;
+ struct intel_token new_token;
+
+ new_token.code = T_NIL;
+ new_token.reg = NULL;
+ new_token.str = NULL;
+
+ /* Free the memory allocated to the previous token and move
+ cur_token to prev_token. */
+ if (prev_token.str)
+ free (prev_token.str);
+
+ prev_token = cur_token;
+
+ /* Skip whitespace. */
+ while (is_space_char (*intel_parser.op_string))
+ intel_parser.op_string++;
+
+ /* Return an empty token if we find nothing else on the line. */
+ if (*intel_parser.op_string == '\0')
+ {
+ cur_token = new_token;
+ return;
+ }
+
+ /* The new token cannot be larger than the remainder of the operand
+ string. */
+ new_token.str = (char *)malloc (strlen (intel_parser.op_string) + 1);
+ if (new_token.str == NULL)
+ abort ();
+ new_token.str[0] = '\0';
+
+ if (strchr ("0123456789", *intel_parser.op_string))
+ {
+ char *p = new_token.str;
+ char *q = intel_parser.op_string;
+ new_token.code = T_CONST;
+
+ /* Allow any kind of identifier char to encompass floating point and
+ hexadecimal numbers. */
+ while (is_identifier_char (*q))
+ *p++ = *q++;
+ *p = '\0';
+
+ /* Recognize special symbol names [0-9][bf]. */
+ if (strlen (intel_parser.op_string) == 2
+ && (intel_parser.op_string[1] == 'b'
+ || intel_parser.op_string[1] == 'f'))
+ new_token.code = T_ID;
+ }
+
+ else if (strchr ("+-/*:[]()", *intel_parser.op_string))
+ {
+ new_token.code = *intel_parser.op_string;
+ new_token.str[0] = *intel_parser.op_string;
+ new_token.str[1] = '\0';
+ }
+
+ else if ((*intel_parser.op_string == REGISTER_PREFIX || allow_naked_reg)
+ && ((reg = parse_register (intel_parser.op_string, &end_op)) != NULL))
+ {
+ new_token.code = T_REG;
+ new_token.reg = reg;
+
+ if (*intel_parser.op_string == REGISTER_PREFIX)
+ {
+ new_token.str[0] = REGISTER_PREFIX;
+ new_token.str[1] = '\0';
+ }
+
+ strcat (new_token.str, reg->reg_name);
+ }
+
+ else if (is_identifier_char (*intel_parser.op_string))
+ {
+ char *p = new_token.str;
+ char *q = intel_parser.op_string;
+
+ /* A '.' or '$' followed by an identifier char is an identifier.
+ Otherwise, it's operator '.' followed by an expression. */
+ if ((*q == '.' || *q == '$') && !is_identifier_char (*(q + 1)))
+ {
+ new_token.code = *q;
+ new_token.str[0] = *q;
+ new_token.str[1] = '\0';
+ }
+ else
+ {
+ while (is_identifier_char (*q) || *q == '@')
+ *p++ = *q++;
+ *p = '\0';
+
+ if (strcasecmp (new_token.str, "BYTE") == 0)
+ new_token.code = T_BYTE;
+
+ else if (strcasecmp (new_token.str, "WORD") == 0)
+ new_token.code = T_WORD;
+
+ else if (strcasecmp (new_token.str, "DWORD") == 0)
+ new_token.code = T_DWORD;
+
+ else if (strcasecmp (new_token.str, "QWORD") == 0)
+ new_token.code = T_QWORD;
+
+ else if (strcasecmp (new_token.str, "XWORD") == 0)
+ new_token.code = T_XWORD;
+
+ else if (strcasecmp (new_token.str, "PTR") == 0)
+ new_token.code = T_PTR;
+
+ else if (strcasecmp (new_token.str, "SHORT") == 0)
+ new_token.code = T_SHORT;
+
+ else if (strcasecmp (new_token.str, "OFFSET") == 0)
+ {
+ new_token.code = T_OFFSET;
+
+ /* ??? This is not mentioned in the MASM grammar but gcc
+ makes use of it with -mintel-syntax. OFFSET may be
+ followed by FLAT: */
+ if (strncasecmp (q, " FLAT:", 6) == 0)
+ strcat (new_token.str, " FLAT:");
+ }
+
+ /* ??? This is not mentioned in the MASM grammar. */
+ else if (strcasecmp (new_token.str, "FLAT") == 0)
+ new_token.code = T_OFFSET;
+
+ else
+ new_token.code = T_ID;
+ }
+ }
+
+ else
+ as_bad (_("Unrecognized token `%s'\n"), intel_parser.op_string);
+
+ intel_parser.op_string += strlen (new_token.str);
+ cur_token = new_token;
+}
+
+/* Put cur_token back into the token stream and make cur_token point to
+ prev_token. */
+static void
+intel_putback_token ()
+{
+ intel_parser.op_string -= strlen (cur_token.str);
+ free (cur_token.str);
+ cur_token = prev_token;
+
+ /* Forget prev_token. */
+ prev_token.code = T_NIL;
+ prev_token.reg = NULL;
+ prev_token.str = NULL;
+}
projects / deliverable / binutils-gdb.git / blobdiff