/* i386.c -- Assemble code for the Intel 80386
- Copyright (C) 1989, 1991, 1992 Free Software Foundation.
+ Copyright (C) 1989, 91, 92, 93, 94, 95, 96, 97, 98, 99, 2000
+ Free Software Foundation, Inc.
This file is part of GAS, the GNU Assembler.
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
- along with GAS; see the file COPYING. If not, write to
- the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA. */
+ along with GAS; see the file COPYING. If not, write to the Free
+ Software Foundation, 59 Temple Place - Suite 330, Boston, MA
+ 02111-1307, USA. */
-/*
- Intel 80386 machine specific gas.
- Written by Eliot Dresselhaus (eliot@mgm.mit.edu).
- Bugs & suggestions are completely welcome. This is free software.
- Please help us make it better.
- */
+/* Intel 80386 machine specific gas.
+ Written by Eliot Dresselhaus (eliot@mgm.mit.edu).
+ Bugs & suggestions are completely welcome. This is free software.
+ Please help us make it better. */
#include <ctype.h>
#include "as.h"
-
-#include "obstack.h"
+#include "subsegs.h"
+#include "dwarf2dbg.h"
#include "opcode/i386.h"
+#ifndef REGISTER_WARNINGS
+#define REGISTER_WARNINGS 1
+#endif
+
+#ifndef INFER_ADDR_PREFIX
+#define INFER_ADDR_PREFIX 1
+#endif
+
+#ifndef SCALE1_WHEN_NO_INDEX
+/* Specifying a scale factor besides 1 when there is no index is
+ futile. eg. `mov (%ebx,2),%al' does exactly the same as
+ `mov (%ebx),%al'. To slavishly follow what the programmer
+ specified, set SCALE1_WHEN_NO_INDEX to 0. */
+#define SCALE1_WHEN_NO_INDEX 1
+#endif
+
+#define true 1
+#define false 0
+
+static unsigned int mode_from_disp_size PARAMS ((unsigned int));
+static int fits_in_signed_byte PARAMS ((offsetT));
+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 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_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));
+#endif
+
/* 'md_assemble ()' gathers together information and puts it into a
- i386_insn. */
+ i386_insn. */
+
+union i386_op
+ {
+ expressionS *disps;
+ expressionS *imms;
+ const reg_entry *regs;
+ };
struct _i386_insn
{
- /* TM holds the template for the insn were currently assembling. */
+ /* TM holds the template for the insn were currently assembling. */
template tm;
- /* SUFFIX holds the opcode suffix (e.g. 'l' for 'movl') if given. */
+
+ /* SUFFIX holds the instruction mnemonic suffix if given.
+ (e.g. 'l' for 'movl') */
char suffix;
- /* Operands are coded with OPERANDS, TYPES, DISPS, IMMS, and REGS. */
- /* OPERANDS gives the number of given operands. */
+ /* OPERANDS gives the number of given operands. */
unsigned int operands;
/* REG_OPERANDS, DISP_OPERANDS, MEM_OPERANDS, IMM_OPERANDS give the number
of given register, displacement, memory operands and immediate
- operands. */
+ operands. */
unsigned int reg_operands, disp_operands, mem_operands, imm_operands;
/* TYPES [i] is the type (see above #defines) which tells us how to
- search through DISPS [i] & IMMS [i] & REGS [i] for the required
- operand. */
+ use OP[i] for the corresponding operand. */
unsigned int types[MAX_OPERANDS];
- /* Displacements (if given) for each operand. */
- expressionS *disps[MAX_OPERANDS];
-
- /* Immediate operands (if given) for each operand. */
- expressionS *imms[MAX_OPERANDS];
+ /* Displacement expression, immediate expression, or register for each
+ operand. */
+ union i386_op op[MAX_OPERANDS];
- /* Register operands (if given) for each operand. */
- reg_entry *regs[MAX_OPERANDS];
+ /* Relocation type for operand */
+#ifdef BFD_ASSEMBLER
+ enum bfd_reloc_code_real disp_reloc[MAX_OPERANDS];
+#else
+ int disp_reloc[MAX_OPERANDS];
+#endif
/* BASE_REG, INDEX_REG, and LOG2_SCALE_FACTOR are used to encode
the base index byte below. */
- reg_entry *base_reg;
- reg_entry *index_reg;
+ const reg_entry *base_reg;
+ const reg_entry *index_reg;
unsigned int log2_scale_factor;
- /* SEG gives the seg_entry of this insn. It is equal to zero unless
- an explicit segment override is given. */
- const seg_entry *seg; /* segment for memory operands (if given) */
+ /* SEG gives the seg_entries of this insn. They are zero unless
+ explicit segment overrides are given. */
+ const seg_entry *seg[2];
/* PREFIX holds all the given prefix opcodes (usually null).
- PREFIXES is the size of PREFIX. */
- /* richfix: really unsigned? */
- unsigned char prefix[MAX_PREFIXES];
+ PREFIXES is the number of prefix opcodes. */
unsigned int prefixes;
+ unsigned char prefix[MAX_PREFIXES];
- /* RM and IB are the modrm byte and the base index byte where the
- addressing modes of this insn are encoded. */
+ /* RM and SIB are the modrm byte and the sib byte where the
+ addressing modes of this insn are encoded. */
modrm_byte rm;
- base_index_byte bi;
+ sib_byte sib;
};
typedef struct _i386_insn i386_insn;
+/* List of chars besides those in app.c:symbol_chars that can start an
+ operand. Used to prevent the scrubber eating vital white-space. */
+#ifdef LEX_AT
+const char extra_symbol_chars[] = "*%-(@";
+#else
+const char extra_symbol_chars[] = "*%-(";
+#endif
+
/* This array holds the chars that always start a comment. If the
- pre-processor is disabled, these aren't very useful */
-#ifdef TE_I386AIX
+ 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. */
const char comment_chars[] = "#/";
+#define PREFIX_SEPARATOR '\\'
#else
const char comment_chars[] = "#";
+#define PREFIX_SEPARATOR '/'
#endif
/* 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
- '/' isn't otherwise defined. */
-const char line_comment_chars[] = "/"; /* removed '#' xoxorich. */
-const char line_separator_chars[] = "";
+ #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[] = "";
+#else
+const char line_comment_chars[] = "/";
+#endif
-/* Chars that can be used to separate mant from exp in floating point nums */
+const char line_separator_chars[] = ";";
+
+/* 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 */
-static char opcode_chars[256];
+/* Tables for lexical analysis. */
+static char mnemonic_chars[256];
static char register_chars[256];
static char operand_chars[256];
-static char space_chars[256];
static char identifier_chars[256];
static char digit_chars[256];
-/* lexical macros */
-#define is_opcode_char(x) (opcode_chars[(unsigned char) x])
+/* 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_space_char(x) (space_chars[(unsigned char) x])
+#define is_space_char(x) ((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 */
-static char operand_special_chars[] = "%$-+(,)*._~/<>|&^!:";
-
-static char *ordinal_names[] = {"first", "second", "third"}; /* for printfs */
+/* All non-digit non-letter charcters that may occur in an operand. */
+static char operand_special_chars[] = "%$-+(,)*._~/<>|&^!:[@]";
/* md_assemble() always leaves the strings it's passed unaltered. To
effect this we maintain a stack of saved characters that we've smashed
with '\0's (indicating end of strings for various sub-fields of the
- assembler instruction). */
+ assembler instruction). */
static char save_stack[32];
-static char *save_stack_p; /* stack pointer */
-#define END_STRING_AND_SAVE(s) *save_stack_p++ = *s; *s = '\0'
-#define RESTORE_END_STRING(s) *s = *--save_stack_p
+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) \
+ do { *(s) = *--save_stack_p; } while (0)
-/* The instruction we're assembling. */
+/* The instruction we're assembling. */
static i386_insn i;
-/* Per instruction expressionS buffers: 2 displacements & 2 immediate max. */
+/* Possible templates for current insn. */
+static const templates *current_templates;
+
+/* Per instruction expressionS buffers: 2 displacements & 2 immediate max. */
static expressionS disp_expressions[2], im_expressions[2];
-/* pointers to ebp & esp entries in reg_hash hash table */
-static reg_entry *ebp, *esp;
+/* 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;
-static int this_operand; /* current operand we are working on */
+/* 1 for intel syntax,
+ 0 if att syntax. */
+static int intel_syntax = 0;
+
+/* 1 if register prefix % not required. */
+static int allow_naked_reg = 0;
+
+/* Used in 16 bit gcc mode to add an l suffix to call, ret, enter,
+ leave, push, and pop instructions so that gcc has the same stack
+ frame as in 32 bit mode. */
+static char stackop_size = '\0';
+
+/* Non-zero to quieten some warnings. */
+static int quiet_warnings = 0;
+
+/* CPU name. */
+static const char *cpu_arch_name = NULL;
+
+/* CPU feature flags. */
+static unsigned int cpu_arch_flags = 0;
/* Interface to relax_segment.
There are 2 relax states for 386 jump insns: one for conditional &
- one for unconditional jumps. This is because the these two types
- of jumps add different sizes to frags when we're figuring out what
+ one for unconditional jumps. This is because these two types of
+ jumps add different sizes to frags when we're figuring out what
sort of jump to choose to reach a given label. */
-/* types */
-#define COND_JUMP 1 /* conditional jump */
-#define UNCOND_JUMP 2 /* unconditional jump */
-/* sizes */
-#define BYTE 0
-#define WORD 1
-#define DWORD 2
-#define UNKNOWN_SIZE 3
+/* Types. */
+#define COND_JUMP 1
+#define UNCOND_JUMP 2
+/* Sizes. */
+#define CODE16 1
+#define SMALL 0
+#define SMALL16 (SMALL|CODE16)
+#define BIG 2
+#define BIG16 (BIG|CODE16)
#ifndef INLINE
#ifdef __GNUC__
#endif
#endif
-#define ENCODE_RELAX_STATE(type,size) ((type<<2) | (size))
+#define ENCODE_RELAX_STATE(type,size) \
+ ((relax_substateT) ((type<<2) | (size)))
#define SIZE_FROM_RELAX_STATE(s) \
- ( (((s) & 0x3) == BYTE ? 1 : (((s) & 0x3) == WORD ? 2 : 4)) )
+ ( (((s) & 0x3) == BIG ? 4 : (((s) & 0x3) == BIG16 ? 2 : 1)) )
+
+/* This table is used by relax_frag to promote short jumps to long
+ ones where necessary. SMALL (short) jumps may be promoted to BIG
+ (32 bit long) ones, and SMALL16 jumps to BIG16 (16 bit long). We
+ don't allow a short jump in a 32 bit code segment to be promoted to
+ a 16 bit offset jump because it's slower (requires data size
+ prefix), and doesn't work, unless the destination is in the bottom
+ 64k of the code segment (The top 16 bits of eip are zeroed). */
const relax_typeS md_relax_table[] =
{
-/* The fields are:
- 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.
- */
+ /* The fields are:
+ 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. */
{1, 1, 0, 0},
{1, 1, 0, 0},
{1, 1, 0, 0},
{1, 1, 0, 0},
- /* For now we don't use word displacement jumps; they may be
- untrustworthy. */
- {127 + 1, -128 + 1, 0, ENCODE_RELAX_STATE (COND_JUMP, DWORD)},
- /* word conditionals add 3 bytes to frag:
- 2 opcode prefix; 1 displacement bytes */
- {32767 + 2, -32768 + 2, 3, ENCODE_RELAX_STATE (COND_JUMP, DWORD)},
+ {127 + 1, -128 + 1, 0, ENCODE_RELAX_STATE (COND_JUMP, BIG)},
+ {127 + 1, -128 + 1, 0, ENCODE_RELAX_STATE (COND_JUMP, BIG16)},
/* dword conditionals adds 4 bytes to frag:
- 1 opcode prefix; 3 displacement bytes */
+ 1 extra opcode byte, 3 extra displacement bytes. */
{0, 0, 4, 0},
- {1, 1, 0, 0},
+ /* word conditionals add 2 bytes to frag:
+ 1 extra opcode byte, 1 extra displacement byte. */
+ {0, 0, 2, 0},
- {127 + 1, -128 + 1, 0, ENCODE_RELAX_STATE (UNCOND_JUMP, DWORD)},
- /* word jmp adds 2 bytes to frag:
- 1 opcode prefix; 1 displacement bytes */
- {32767 + 2, -32768 + 2, 2, ENCODE_RELAX_STATE (UNCOND_JUMP, DWORD)},
+ {127 + 1, -128 + 1, 0, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG)},
+ {127 + 1, -128 + 1, 0, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16)},
/* dword jmp adds 3 bytes to frag:
- 0 opcode prefix; 3 displacement bytes */
+ 0 extra opcode bytes, 3 extra displacement bytes. */
{0, 0, 3, 0},
- {1, 1, 0, 0},
+ /* word jmp adds 1 byte to frag:
+ 0 extra opcode bytes, 1 extra displacement byte. */
+ {0, 0, 1, 0}
};
+static const arch_entry cpu_arch[] = {
+ {"i8086", Cpu086 },
+ {"i186", Cpu086|Cpu186 },
+ {"i286", Cpu086|Cpu186|Cpu286 },
+ {"i386", Cpu086|Cpu186|Cpu286|Cpu386 },
+ {"i486", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486 },
+ {"i586", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuMMX },
+ {"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 },
+ {NULL, 0 }
+};
+
+void
+i386_align_code (fragP, count)
+ 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. */
+ static const char f32_1[] =
+ {0x90}; /* nop */
+ static const char f32_2[] =
+ {0x89,0xf6}; /* movl %esi,%esi */
+ static const char f32_3[] =
+ {0x8d,0x76,0x00}; /* leal 0(%esi),%esi */
+ static const char f32_4[] =
+ {0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
+ static const char f32_5[] =
+ {0x90, /* nop */
+ 0x8d,0x74,0x26,0x00}; /* leal 0(%esi,1),%esi */
+ static const char f32_6[] =
+ {0x8d,0xb6,0x00,0x00,0x00,0x00}; /* leal 0L(%esi),%esi */
+ static const char f32_7[] =
+ {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
+ static const char f32_8[] =
+ {0x90, /* nop */
+ 0x8d,0xb4,0x26,0x00,0x00,0x00,0x00}; /* leal 0L(%esi,1),%esi */
+ static const char f32_9[] =
+ {0x89,0xf6, /* movl %esi,%esi */
+ 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
+ static const char f32_10[] =
+ {0x8d,0x76,0x00, /* leal 0(%esi),%esi */
+ 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
+ static const char f32_11[] =
+ {0x8d,0x74,0x26,0x00, /* leal 0(%esi,1),%esi */
+ 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
+ static const char f32_12[] =
+ {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
+ 0x8d,0xbf,0x00,0x00,0x00,0x00}; /* leal 0L(%edi),%edi */
+ static const char f32_13[] =
+ {0x8d,0xb6,0x00,0x00,0x00,0x00, /* leal 0L(%esi),%esi */
+ 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
+ static const char f32_14[] =
+ {0x8d,0xb4,0x26,0x00,0x00,0x00,0x00, /* leal 0L(%esi,1),%esi */
+ 0x8d,0xbc,0x27,0x00,0x00,0x00,0x00}; /* leal 0L(%edi,1),%edi */
+ static const char f32_15[] =
+ {0xeb,0x0d,0x90,0x90,0x90,0x90,0x90, /* jmp .+15; lotsa nops */
+ 0x90,0x90,0x90,0x90,0x90,0x90,0x90,0x90};
+ static const char f16_3[] =
+ {0x8d,0x74,0x00}; /* lea 0(%esi),%esi */
+ static const char f16_4[] =
+ {0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
+ static const char f16_5[] =
+ {0x90, /* nop */
+ 0x8d,0xb4,0x00,0x00}; /* lea 0w(%si),%si */
+ static const char f16_6[] =
+ {0x89,0xf6, /* mov %si,%si */
+ 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
+ static const char f16_7[] =
+ {0x8d,0x74,0x00, /* lea 0(%si),%si */
+ 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
+ static const char f16_8[] =
+ {0x8d,0xb4,0x00,0x00, /* lea 0w(%si),%si */
+ 0x8d,0xbd,0x00,0x00}; /* lea 0w(%di),%di */
+ static const char *const f32_patt[] = {
+ f32_1, f32_2, f32_3, f32_4, f32_5, f32_6, f32_7, f32_8,
+ f32_9, f32_10, f32_11, f32_12, f32_13, f32_14, f32_15
+ };
+ static const char *const f16_patt[] = {
+ f32_1, f32_2, f16_3, f16_4, f16_5, f16_6, f16_7, f16_8,
+ f32_15, f32_15, f32_15, f32_15, f32_15, f32_15, f32_15
+ };
+
+ if (count > 0 && count <= 15)
+ {
+ if (flag_16bit_code)
+ {
+ memcpy (fragP->fr_literal + fragP->fr_fix,
+ f16_patt[count - 1], count);
+ if (count > 8)
+ /* Adjust jump offset. */
+ fragP->fr_literal[fragP->fr_fix + 1] = count - 2;
+ }
+ else
+ memcpy (fragP->fr_literal + fragP->fr_fix,
+ f32_patt[count - 1], count);
+ fragP->fr_var = count;
+ }
+}
+
static char *output_invalid PARAMS ((int c));
static int i386_operand PARAMS ((char *operand_string));
-static reg_entry *parse_register PARAMS ((char *reg_string));
+static int i386_intel_operand PARAMS ((char *operand_string, int got_a_float));
+static const reg_entry *parse_register PARAMS ((char *reg_string,
+ char **end_op));
+
#ifndef I386COFF
-static void s_bss PARAMS ((void));
+static void s_bss PARAMS ((int));
#endif
-static INLINE unsigned long
+symbolS *GOT_symbol; /* Pre-defined "_GLOBAL_OFFSET_TABLE_". */
+
+static INLINE unsigned int
mode_from_disp_size (t)
- unsigned long t;
+ unsigned int t;
{
- return (t & Disp8) ? 1 : (t & Disp32) ? 2 : 0;
+ return (t & Disp8) ? 1 : (t & (Disp16 | Disp32)) ? 2 : 0;
}
-/* convert opcode suffix ('b' 'w' 'l' typically) into type specifier */
-
-static INLINE unsigned long
-opcode_suffix_to_type (s)
- unsigned long s;
-{
- return (s == BYTE_OPCODE_SUFFIX
- ? Byte : (s == WORD_OPCODE_SUFFIX
- ? Word : DWord));
-} /* opcode_suffix_to_type() */
-
static INLINE int
fits_in_signed_byte (num)
- long num;
+ offsetT num;
{
return (num >= -128) && (num <= 127);
-} /* fits_in_signed_byte() */
+}
static INLINE int
fits_in_unsigned_byte (num)
- long num;
+ offsetT num;
{
return (num & 0xff) == num;
-} /* fits_in_unsigned_byte() */
+}
static INLINE int
fits_in_unsigned_word (num)
- long num;
+ offsetT num;
{
return (num & 0xffff) == num;
-} /* fits_in_unsigned_word() */
+}
static INLINE int
fits_in_signed_word (num)
- long num;
+ offsetT num;
{
return (-32768 <= num) && (num <= 32767);
-} /* fits_in_signed_word() */
+}
static int
smallest_imm_type (num)
- long num;
+ offsetT num;
{
- return ((num == 1)
- ? (Imm1 | Imm8 | Imm8S | Imm16 | Imm32)
- : fits_in_signed_byte (num)
+ if (cpu_arch_flags != 0
+ && cpu_arch_flags != (Cpu086 | Cpu186 | Cpu286 | Cpu386 | Cpu486))
+ {
+ /* This code is disabled on the 486 because all the Imm1 forms
+ in the opcode table are slower on the i486. They're the
+ versions with the implicitly specified single-position
+ displacement, which has another syntax if you really want to
+ use that form. */
+ if (num == 1)
+ return Imm1 | Imm8 | Imm8S | Imm16 | Imm32;
+ }
+ return (fits_in_signed_byte (num)
? (Imm8S | Imm8 | Imm16 | Imm32)
: fits_in_unsigned_byte (num)
? (Imm8 | Imm16 | Imm32)
: (fits_in_signed_word (num) || fits_in_unsigned_word (num))
? (Imm16 | Imm32)
: (Imm32));
-} /* smallest_imm_type() */
+}
-/* Ignore certain directives generated by gcc. This probably should
- not be here. */
-void
-dummy ()
+static offsetT
+offset_in_range (val, size)
+ offsetT val;
+ int size;
{
- while (*input_line_pointer && *input_line_pointer != '\n')
- input_line_pointer++;
+ addressT mask;
+
+ switch (size)
+ {
+ case 1: mask = ((addressT) 1 << 8) - 1; break;
+ case 2: mask = ((addressT) 1 << 16) - 1; break;
+ case 4: mask = ((addressT) 2 << 31) - 1; break;
+ default: abort ();
+ }
+
+ /* If BFD64, sign extend val. */
+ if ((val & ~(((addressT) 2 << 31) - 1)) == 0)
+ val = (val ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
+
+ if ((val & ~mask) != 0 && (val & ~mask) != ~mask)
+ {
+ char buf1[40], buf2[40];
+
+ sprint_value (buf1, val);
+ sprint_value (buf2, val & mask);
+ as_warn (_("%s shortened to %s"), buf1, buf2);
+ }
+ return val & mask;
}
-const pseudo_typeS md_pseudo_table[] =
+/* Returns 0 if attempting to add a prefix where one from the same
+ class already exists, 1 if non rep/repne added, 2 if rep/repne
+ added. */
+static int
+add_prefix (prefix)
+ unsigned int prefix;
{
-#ifndef I386COFF
- {"bss", s_bss, 0},
+ 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;
+
+ 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 ADDR_PREFIX_OPCODE:
+ q = ADDR_PREFIX;
+ break;
+
+ case DATA_PREFIX_OPCODE:
+ q = DATA_PREFIX;
+ break;
+ }
+
+ if (i.prefix[q])
+ {
+ as_bad (_("same type of prefix used twice"));
+ return 0;
+ }
+
+ i.prefixes += 1;
+ i.prefix[q] = prefix;
+ return ret;
+}
+
+static void
+set_16bit_code_flag (new_16bit_code_flag)
+ int new_16bit_code_flag;
+{
+ flag_16bit_code = new_16bit_code_flag;
+ stackop_size = '\0';
+}
+
+static void
+set_16bit_gcc_code_flag (new_16bit_code_flag)
+ int new_16bit_code_flag;
+{
+ flag_16bit_code = new_16bit_code_flag;
+ stackop_size = new_16bit_code_flag ? 'l' : '\0';
+}
+
+static void
+set_intel_syntax (syntax_flag)
+ int syntax_flag;
+{
+ /* Find out if register prefixing is specified. */
+ int ask_naked_reg = 0;
+
+ SKIP_WHITESPACE ();
+ if (! is_end_of_line[(unsigned char) *input_line_pointer])
+ {
+ char *string = input_line_pointer;
+ int e = get_symbol_end ();
+
+ if (strcmp (string, "prefix") == 0)
+ ask_naked_reg = 1;
+ else if (strcmp (string, "noprefix") == 0)
+ ask_naked_reg = -1;
+ else
+ as_bad (_("bad argument to syntax directive."));
+ *input_line_pointer = e;
+ }
+ demand_empty_rest_of_line ();
+
+ intel_syntax = syntax_flag;
+
+ if (ask_naked_reg == 0)
+ {
+#ifdef BFD_ASSEMBLER
+ allow_naked_reg = (intel_syntax
+ && (bfd_get_symbol_leading_char (stdoutput) != '\0'));
+#else
+ /* Conservative default. */
+ allow_naked_reg = 0;
#endif
-#ifndef OBJ_AOUT
+ }
+ else
+ allow_naked_reg = (ask_naked_reg < 0);
+}
+
+static void
+set_cpu_arch (dummy)
+ int dummy ATTRIBUTE_UNUSED;
+{
+ SKIP_WHITESPACE ();
+
+ if (! is_end_of_line[(unsigned char) *input_line_pointer])
+ {
+ char *string = input_line_pointer;
+ int e = get_symbol_end ();
+ int i;
+
+ for (i = 0; cpu_arch[i].name; i++)
+ {
+ if (strcmp (string, cpu_arch[i].name) == 0)
+ {
+ cpu_arch_name = cpu_arch[i].name;
+ cpu_arch_flags = cpu_arch[i].flags;
+ break;
+ }
+ }
+ if (!cpu_arch[i].name)
+ as_bad (_("no such architecture: `%s'"), string);
+
+ *input_line_pointer = e;
+ }
+ else
+ as_bad (_("missing cpu architecture"));
+
+ demand_empty_rest_of_line ();
+}
+
+const pseudo_typeS md_pseudo_table[] =
+{
+#if !defined(OBJ_AOUT) && !defined(USE_ALIGN_PTWO)
{"align", s_align_bytes, 0},
#else
{"align", s_align_ptwo, 0},
+#endif
+ {"arch", set_cpu_arch, 0},
+#ifndef I386COFF
+ {"bss", s_bss, 0},
#endif
{"ffloat", float_cons, 'f'},
{"dfloat", float_cons, 'd'},
{"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},
+ {"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}
};
-/* for interface with expression () */
+/* For interface with expression (). */
extern char *input_line_pointer;
-/* obstack for constructing various things in md_begin */
-struct obstack o;
+/* Hash table for instruction mnemonic lookup. */
+static struct hash_control *op_hash;
-/* hash table for opcode lookup */
-static struct hash_control *op_hash = (struct hash_control *) 0;
-/* hash table for register lookup */
-static struct hash_control *reg_hash = (struct hash_control *) 0;
-/* hash table for prefix lookup */
-static struct hash_control *prefix_hash = (struct hash_control *) 0;
+/* Hash table for register lookup. */
+static struct hash_control *reg_hash;
\f
-
void
md_begin ()
{
- char *hash_err;
+ const char *hash_err;
- obstack_begin (&o, 4096);
-
- /* initialize op_hash hash table */
- op_hash = hash_new (); /* xmalloc handles error */
+ /* Initialize op_hash hash table. */
+ op_hash = hash_new ();
{
register const template *optab;
register templates *core_optab;
- char *prev_name;
- optab = i386_optab; /* setup for loop */
- prev_name = optab->name;
- obstack_grow (&o, optab, sizeof (template));
+ /* Setup for loop. */
+ optab = i386_optab;
core_optab = (templates *) xmalloc (sizeof (templates));
+ core_optab->start = optab;
- for (optab++; optab < i386_optab_end; optab++)
+ while (1)
{
- if (!strcmp (optab->name, prev_name))
- {
- /* same name as before --> append to current template list */
- obstack_grow (&o, optab, sizeof (template));
- }
- else
+ ++optab;
+ if (optab->name == NULL
+ || strcmp (optab->name, (optab - 1)->name) != 0)
{
/* different name --> ship out current template list;
- add to hash table; & begin anew */
- /* Note: end must be set before start! since obstack_next_free
- changes upon opstack_finish */
- core_optab->end = (template *) obstack_next_free (&o);
- core_optab->start = (template *) obstack_finish (&o);
- hash_err = hash_insert (op_hash, prev_name, (char *) core_optab);
- if (hash_err && *hash_err)
+ add to hash table; & begin anew. */
+ core_optab->end = optab;
+ hash_err = hash_insert (op_hash,
+ (optab - 1)->name,
+ (PTR) core_optab);
+ if (hash_err)
{
hash_error:
- as_fatal ("Internal Error: Can't hash %s: %s", prev_name,
+ as_fatal (_("Internal Error: Can't hash %s: %s"),
+ (optab - 1)->name,
hash_err);
}
- prev_name = optab->name;
+ if (optab->name == NULL)
+ break;
core_optab = (templates *) xmalloc (sizeof (templates));
- obstack_grow (&o, optab, sizeof (template));
+ core_optab->start = optab;
}
}
}
- /* initialize reg_hash hash table */
+ /* Initialize reg_hash hash table. */
reg_hash = hash_new ();
{
register const reg_entry *regtab;
- for (regtab = i386_regtab; regtab < i386_regtab_end; regtab++)
- {
- hash_err = hash_insert (reg_hash, regtab->reg_name, regtab);
- if (hash_err && *hash_err)
- goto hash_error;
- }
- }
-
- esp = (reg_entry *) hash_find (reg_hash, "esp");
- ebp = (reg_entry *) hash_find (reg_hash, "ebp");
-
- /* initialize reg_hash hash table */
- prefix_hash = hash_new ();
- {
- register const prefix_entry *prefixtab;
-
- for (prefixtab = i386_prefixtab;
- prefixtab < i386_prefixtab_end; prefixtab++)
+ for (regtab = i386_regtab;
+ regtab < i386_regtab + sizeof (i386_regtab) / sizeof (i386_regtab[0]);
+ regtab++)
{
- hash_err = hash_insert (prefix_hash, prefixtab->prefix_name, prefixtab);
- if (hash_err && *hash_err)
+ hash_err = hash_insert (reg_hash, regtab->reg_name, (PTR) regtab);
+ if (hash_err)
goto hash_error;
}
}
- /* fill in lexical tables: opcode_chars, operand_chars, space_chars */
+ /* Fill in lexical tables: mnemonic_chars, operand_chars. */
{
- register unsigned int c;
-
- memset (opcode_chars, '\0', sizeof (opcode_chars));
- memset (operand_chars, '\0', sizeof (operand_chars));
- memset (space_chars, '\0', sizeof (space_chars));
- memset (identifier_chars, '\0', sizeof (identifier_chars));
- memset (digit_chars, '\0', sizeof (digit_chars));
+ register int c;
+ register char *p;
for (c = 0; c < 256; c++)
{
- if (islower (c) || isdigit (c))
+ if (isdigit (c))
{
- opcode_chars[c] = c;
+ digit_chars[c] = c;
+ mnemonic_chars[c] = c;
register_chars[c] = c;
+ operand_chars[c] = c;
}
- else if (isupper (c))
+ else if (islower (c))
{
- opcode_chars[c] = tolower (c);
- register_chars[c] = opcode_chars[c];
+ mnemonic_chars[c] = c;
+ register_chars[c] = c;
+ operand_chars[c] = c;
}
- else if (c == PREFIX_SEPERATOR)
+ else if (isupper (c))
{
- opcode_chars[c] = c;
+ mnemonic_chars[c] = tolower (c);
+ register_chars[c] = mnemonic_chars[c];
+ operand_chars[c] = c;
}
- else if (c == ')' || c == '(')
+
+ if (isalpha (c) || isdigit (c))
+ identifier_chars[c] = c;
+ else if (c >= 128)
{
- register_chars[c] = c;
+ identifier_chars[c] = c;
+ operand_chars[c] = c;
}
+ }
- if (isupper (c) || islower (c) || isdigit (c))
- operand_chars[c] = c;
- else if (c && strchr (operand_special_chars, c))
- operand_chars[c] = c;
-
- if (isdigit (c) || c == '-')
- digit_chars[c] = c;
-
- if (isalpha (c) || c == '_' || c == '.' || isdigit (c))
- identifier_chars[c] = c;
+#ifdef LEX_AT
+ identifier_chars['@'] = '@';
+#endif
+ digit_chars['-'] = '-';
+ identifier_chars['_'] = '_';
+ identifier_chars['.'] = '.';
- if (c == ' ' || c == '\t')
- space_chars[c] = c;
- }
+ for (p = operand_special_chars; *p != '\0'; p++)
+ operand_chars[(unsigned char) *p] = *p;
}
+
+#if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
+ if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
+ {
+ record_alignment (text_section, 2);
+ record_alignment (data_section, 2);
+ record_alignment (bss_section, 2);
+ }
+#endif
}
void
-md_end ()
+i386_print_statistics (file)
+ FILE *file;
{
-} /* not much to do here. */
+ hash_print_statistics (file, "i386 opcode", op_hash);
+ hash_print_statistics (file, "i386 register", reg_hash);
+}
\f
-
#ifdef DEBUG386
-/* debugging routines for md_assemble */
-/* static void pi (), pte (), pt (), pe (), ps (); */
+/* Debugging routines for md_assemble. */
+static void pi PARAMS ((char *, i386_insn *));
+static void pte PARAMS ((template *));
+static void pt PARAMS ((unsigned int));
+static void pe PARAMS ((expressionS *));
+static void ps PARAMS ((symbolS *));
static void
pi (line, x)
fprintf (stdout, " #%d: ", i + 1);
pt (x->types[i]);
fprintf (stdout, "\n");
- if (x->types[i] & Reg)
- fprintf (stdout, "%s\n", x->regs[i]->reg_name);
+ if (x->types[i]
+ & (Reg | SReg2 | SReg3 | Control | Debug | Test | RegMMX | RegXMM))
+ fprintf (stdout, "%s\n", x->op[i].regs->reg_name);
if (x->types[i] & Imm)
- pe (x->imms[i]);
- if (x->types[i] & (Disp | Abs))
- pe (x->disps[i]);
+ pe (x->op[i].imms);
+ if (x->types[i] & Disp)
+ pe (x->op[i].disps);
}
}
{
int i;
fprintf (stdout, " %d operands ", t->operands);
- fprintf (stdout, "opcode %x ",
- t->base_opcode);
+ fprintf (stdout, "opcode %x ", t->base_opcode);
if (t->extension_opcode != None)
fprintf (stdout, "ext %x ", t->extension_opcode);
if (t->opcode_modifier & D)
pe (e)
expressionS *e;
{
- fprintf (stdout, " segment %s\n", segment_name (e->X_seg));
- fprintf (stdout, " add_number %d (%x)\n",
- e->X_add_number, e->X_add_number);
+ fprintf (stdout, " operation %d\n", e->X_op);
+ fprintf (stdout, " add_number %ld (%lx)\n",
+ (long) e->X_add_number, (long) e->X_add_number);
if (e->X_add_symbol)
{
fprintf (stdout, " add_symbol ");
ps (e->X_add_symbol);
fprintf (stdout, "\n");
}
- if (e->X_subtract_symbol)
+ if (e->X_op_symbol)
{
- fprintf (stdout, " sub_symbol ");
- ps (e->X_subtract_symbol);
+ fprintf (stdout, " op_symbol ");
+ ps (e->X_op_symbol);
fprintf (stdout, "\n");
}
}
{ Imm8S, "i8s" },
{ Imm16, "i16" },
{ Imm32, "i32" },
- { Mem8, "Mem8" },
- { Mem16, "Mem16" },
- { Mem32, "Mem32" },
+ { Imm1, "i1" },
{ BaseIndex, "BaseIndex" },
- { Abs8, "Abs8" },
- { Abs16, "Abs16" },
- { Abs32, "Abs32" },
{ Disp8, "d8" },
{ Disp16, "d16" },
{ Disp32, "d32" },
- { SReg2, "SReg2" },
- { SReg3, "SReg3" },
- { Acc, "Acc" },
{ InOutPortReg, "InOutPortReg" },
{ ShiftCount, "ShiftCount" },
- { Imm1, "i1" },
{ Control, "control reg" },
{ Test, "test reg" },
+ { Debug, "debug reg" },
{ FloatReg, "FReg" },
{ FloatAcc, "FAcc" },
+ { SReg2, "SReg2" },
+ { SReg3, "SReg3" },
+ { Acc, "Acc" },
{ JumpAbsolute, "Jump Absolute" },
+ { RegMMX, "rMMX" },
+ { RegXMM, "rXMM" },
+ { EsSeg, "es" },
{ 0, "" }
};
if (t == Unknown)
{
- fprintf (stdout, "Unknown");
+ fprintf (stdout, _("Unknown"));
}
else
{
#endif /* DEBUG386 */
\f
+int
+tc_i386_force_relocation (fixp)
+ struct fix *fixp;
+{
+#ifdef BFD_ASSEMBLER
+ if (fixp->fx_r_type == BFD_RELOC_VTABLE_INHERIT
+ || fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
+ return 1;
+ return 0;
+#else
+ /* 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)
+reloc (size, pcrel, other)
int size;
int pcrel;
+ bfd_reloc_code_real_type other;
{
+ if (other != NO_RELOC)
+ return other;
+
if (pcrel)
- switch (size)
- {
- case 1: return BFD_RELOC_8_PCREL;
- case 2: return BFD_RELOC_16_PCREL;
- case 4: return BFD_RELOC_32_PCREL;
- }
+ {
+ switch (size)
+ {
+ case 1: return BFD_RELOC_8_PCREL;
+ case 2: return BFD_RELOC_16_PCREL;
+ case 4: return BFD_RELOC_32_PCREL;
+ }
+ as_bad (_("can not do %d byte pc-relative relocation"), size);
+ }
else
- switch (size)
- {
- case 1: return BFD_RELOC_8;
- case 2: return BFD_RELOC_16;
- case 4: return BFD_RELOC_32;
- }
- abort ();
+ {
+ 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);
+ }
+
+ return BFD_RELOC_NONE;
+}
+
+/* Here we decide which fixups can be adjusted to make them relative to
+ the beginning of the section instead of the symbol. Basically we need
+ to make sure that the dynamic relocations are done correctly, so in
+ some cases we force the original symbol to be used. */
+
+int
+tc_i386_fix_adjustable (fixP)
+ fixS *fixP;
+{
+#if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
+ /* Prevent all adjustments to global symbols, or else dynamic
+ linking will not work correctly. */
+ if (S_IS_EXTERNAL (fixP->fx_addsy)
+ || S_IS_WEAK (fixP->fx_addsy))
+ return 0;
+#endif
+ /* 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_VTABLE_INHERIT
+ || fixP->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
+ return 0;
+ return 1;
}
+#else
+#define reloc(SIZE,PCREL,OTHER) 0
+#define BFD_RELOC_16 0
+#define BFD_RELOC_32 0
+#define BFD_RELOC_16_PCREL 0
+#define BFD_RELOC_32_PCREL 0
+#define BFD_RELOC_386_PLT32 0
+#define BFD_RELOC_386_GOT32 0
+#define BFD_RELOC_386_GOTOFF 0
#endif
+static int intel_float_operand PARAMS ((char *mnemonic));
+
+static int
+intel_float_operand (mnemonic)
+ char *mnemonic;
+{
+ if (mnemonic[0] == 'f' && mnemonic[1] == 'i')
+ return 2;
+
+ if (mnemonic[0] == 'f')
+ return 1;
+
+ return 0;
+}
+
/* This is the guts of the machine-dependent assembler. LINE points to a
machine dependent instruction. This function is supposed to emit
the frags/bytes it assembles to. */
+
void
md_assemble (line)
char *line;
{
- /* Holds temlate once we've found it. */
- register template *t;
+ /* Points to template once we've found it. */
+ const template *t;
- /* Possible templates for current insn */
- templates *current_templates = (templates *) 0;
+ /* Count the size of the instruction generated. */
+ int insn_size = 0;
- /* Initialize globals. */
+ int j;
+
+ char mnemonic[MAX_MNEM_SIZE];
+
+ /* Initialize globals. */
memset (&i, '\0', sizeof (i));
+ for (j = 0; j < MAX_OPERANDS; j++)
+ 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;
- /* Fist parse an opcode & call i386_operand for the operands.
+ /* 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
- start of a (possibly prefixed) opcode. */
+ start of a (possibly prefixed) mnemonic. */
{
- register char *l = line; /* Fast place to put LINE. */
-
- /* 1 if operand is pending after ','. */
- unsigned int expecting_operand = 0;
- /* 1 if we found a prefix only acceptable with string insns. */
- unsigned int expecting_string_instruction = 0;
- /* Non-zero if operand parens not balenced. */
- unsigned int paren_not_balenced;
+ char *l = line;
char *token_start = l;
+ char *mnem_p;
- while (!is_space_char (*l) && *l != END_OF_INSN)
+ /* Non-zero if we found a prefix only acceptable with string insns. */
+ const char *expecting_string_instruction = NULL;
+
+ while (1)
{
- if (!is_opcode_char (*l))
+ mnem_p = mnemonic;
+ while ((*mnem_p = mnemonic_chars[(unsigned char) *l]) != 0)
+ {
+ mnem_p++;
+ if (mnem_p >= mnemonic + sizeof (mnemonic))
+ {
+ as_bad (_("no such instruction: `%s'"), token_start);
+ return;
+ }
+ l++;
+ }
+ if (!is_space_char (*l)
+ && *l != END_OF_INSN
+ && *l != PREFIX_SEPARATOR)
{
- as_bad ("invalid character %s in opcode", output_invalid (*l));
+ as_bad (_("invalid character %s in mnemonic"),
+ output_invalid (*l));
return;
}
- else if (*l != PREFIX_SEPERATOR)
+ if (token_start == l)
{
- *l = opcode_chars[(unsigned char) *l]; /* fold case of opcodes */
- l++;
+ if (*l == PREFIX_SEPARATOR)
+ as_bad (_("expecting prefix; got nothing"));
+ else
+ as_bad (_("expecting mnemonic; got nothing"));
+ return;
}
- else
- { /* this opcode's got a prefix */
- register unsigned int q;
- register prefix_entry *prefix;
- if (l == token_start)
+ /* Look up instruction (or prefix) via hash table. */
+ current_templates = hash_find (op_hash, mnemonic);
+
+ if (*l != END_OF_INSN
+ && (! is_space_char (*l) || l[1] != END_OF_INSN)
+ && current_templates
+ && (current_templates->start->opcode_modifier & IsPrefix))
+ {
+ /* If we are in 16-bit mode, do not allow addr16 or data16.
+ 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))
{
- as_bad ("expecting prefix; got nothing");
+ as_bad (_("redundant %s prefix"),
+ current_templates->start->name);
return;
}
- END_STRING_AND_SAVE (l);
- prefix = (prefix_entry *) hash_find (prefix_hash, token_start);
- if (!prefix)
+ /* Add prefix, checking for repeated prefixes. */
+ switch (add_prefix (current_templates->start->base_opcode))
{
- as_bad ("no such opcode prefix ('%s')", token_start);
+ case 0:
return;
+ case 2:
+ expecting_string_instruction = current_templates->start->name;
+ break;
}
- RESTORE_END_STRING (l);
- /* check for repeated prefix */
- for (q = 0; q < i.prefixes; q++)
- if (i.prefix[q] == prefix->prefix_code)
- {
- as_bad ("same prefix used twice; you don't really want this!");
- return;
- }
- if (i.prefixes == MAX_PREFIXES)
- {
- as_bad ("too many opcode prefixes");
- return;
- }
- i.prefix[i.prefixes++] = prefix->prefix_code;
- if (prefix->prefix_code == REPE || prefix->prefix_code == REPNE)
- expecting_string_instruction = 1;
- /* skip past PREFIX_SEPERATOR and reset token_start */
+ /* Skip past PREFIX_SEPARATOR and reset token_start. */
token_start = ++l;
}
- }
- END_STRING_AND_SAVE (l);
- if (token_start == l)
- {
- as_bad ("expecting opcode; got nothing");
- return;
+ else
+ break;
}
- /* Lookup insn in hash; try intel & att naming conventions if appropriate;
- that is: we only use the opcode suffix 'b' 'w' or 'l' if we need to. */
- current_templates = (templates *) hash_find (op_hash, token_start);
if (!current_templates)
{
- int last_index = strlen (token_start) - 1;
- char last_char = token_start[last_index];
- switch (last_char)
+ /* See if we can get a match by trimming off a suffix. */
+ switch (mnem_p[-1])
{
- case DWORD_OPCODE_SUFFIX:
- case WORD_OPCODE_SUFFIX:
- case BYTE_OPCODE_SUFFIX:
- token_start[last_index] = '\0';
- current_templates = (templates *) hash_find (op_hash, token_start);
- token_start[last_index] = last_char;
- i.suffix = last_char;
+ case WORD_MNEM_SUFFIX:
+ case BYTE_MNEM_SUFFIX:
+ case SHORT_MNEM_SUFFIX:
+ case LONG_MNEM_SUFFIX:
+ i.suffix = mnem_p[-1];
+ mnem_p[-1] = '\0';
+ current_templates = hash_find (op_hash, mnemonic);
+ break;
+
+ /* Intel Syntax. */
+ case DWORD_MNEM_SUFFIX:
+ if (intel_syntax)
+ {
+ i.suffix = mnem_p[-1];
+ mnem_p[-1] = '\0';
+ current_templates = hash_find (op_hash, mnemonic);
+ break;
+ }
}
if (!current_templates)
{
- as_bad ("no such 386 instruction: `%s'", token_start);
+ as_bad (_("no such instruction: `%s'"), token_start);
return;
}
}
- RESTORE_END_STRING (l);
- /* check for rep/repne without a string instruction */
- if (expecting_string_instruction &&
- !IS_STRING_INSTRUCTION (current_templates->
- start->base_opcode))
+ /* Check if instruction is supported on specified architecture. */
+ if (cpu_arch_flags != 0)
+ {
+ if (current_templates->start->cpu_flags & ~cpu_arch_flags)
+ {
+ as_warn (_("`%s' is not supported on `%s'"),
+ current_templates->start->name, cpu_arch_name);
+ }
+ else if ((Cpu386 & ~cpu_arch_flags) && !flag_16bit_code)
+ {
+ as_warn (_("use .code16 to ensure correct addressing mode"));
+ }
+ }
+
+ /* Check for rep/repne without a string instruction. */
+ if (expecting_string_instruction
+ && !(current_templates->start->opcode_modifier & IsString))
{
- as_bad ("expecting string instruction after rep/repne");
+ as_bad (_("expecting string instruction after `%s'"),
+ expecting_string_instruction);
return;
}
- /* There may be operands to parse. */
- if (*l != END_OF_INSN &&
- /* For string instructions, we ignore any operands if given. This
- kludges, for example, 'rep/movsb %ds:(%esi), %es:(%edi)' where
- the operands are always going to be the same, and are not really
- encoded in machine code. */
- !IS_STRING_INSTRUCTION (current_templates->
- start->base_opcode))
+ /* There may be operands to parse. */
+ if (*l != END_OF_INSN)
{
- /* parse operands */
+ /* 1 if operand is pending after ','. */
+ unsigned int expecting_operand = 0;
+
+ /* Non-zero if operand parens not balanced. */
+ unsigned int paren_not_balanced;
+
do
{
- /* skip optional white space before operand */
- while (!is_operand_char (*l) && *l != END_OF_INSN)
+ /* Skip optional white space before operand. */
+ if (is_space_char (*l))
+ ++l;
+ if (!is_operand_char (*l) && *l != END_OF_INSN)
{
- if (!is_space_char (*l))
- {
- as_bad ("invalid character %s before %s operand",
- output_invalid (*l),
- ordinal_names[i.operands]);
- return;
- }
- l++;
+ as_bad (_("invalid character %s before operand %d"),
+ output_invalid (*l),
+ i.operands + 1);
+ return;
}
token_start = l; /* after white space */
- paren_not_balenced = 0;
- while (paren_not_balenced || *l != ',')
+ paren_not_balanced = 0;
+ while (paren_not_balanced || *l != ',')
{
if (*l == END_OF_INSN)
{
- if (paren_not_balenced)
+ if (paren_not_balanced)
{
- as_bad ("unbalenced parenthesis in %s operand.",
- ordinal_names[i.operands]);
+ if (!intel_syntax)
+ as_bad (_("unbalanced parenthesis in operand %d."),
+ i.operands + 1);
+ else
+ as_bad (_("unbalanced brackets in operand %d."),
+ i.operands + 1);
return;
}
else
break; /* we are done */
}
- else if (!is_operand_char (*l))
+ else if (!is_operand_char (*l) && !is_space_char (*l))
{
- as_bad ("invalid character %s in %s operand",
+ as_bad (_("invalid character %s in operand %d"),
output_invalid (*l),
- ordinal_names[i.operands]);
+ i.operands + 1);
return;
}
- if (*l == '(')
- ++paren_not_balenced;
- if (*l == ')')
- --paren_not_balenced;
+ if (!intel_syntax)
+ {
+ if (*l == '(')
+ ++paren_not_balanced;
+ if (*l == ')')
+ --paren_not_balanced;
+ }
+ else
+ {
+ if (*l == '[')
+ ++paren_not_balanced;
+ if (*l == ']')
+ --paren_not_balanced;
+ }
l++;
}
if (l != token_start)
- { /* yes, we've read in another operand */
+ { /* Yes, we've read in another operand. */
unsigned int operand_ok;
this_operand = i.operands++;
if (i.operands > MAX_OPERANDS)
{
- as_bad ("spurious operands; (%d operands/instruction max)",
+ as_bad (_("spurious operands; (%d operands/instruction max)"),
MAX_OPERANDS);
return;
}
- /* now parse operand adding info to 'i' as we go along */
+ /* Now parse operand adding info to 'i' as we go along. */
END_STRING_AND_SAVE (l);
- operand_ok = i386_operand (token_start);
- RESTORE_END_STRING (l); /* restore old contents */
+
+ if (intel_syntax)
+ operand_ok =
+ i386_intel_operand (token_start,
+ intel_float_operand (mnemonic));
+ else
+ operand_ok = i386_operand (token_start);
+
+ RESTORE_END_STRING (l);
if (!operand_ok)
return;
}
if (expecting_operand)
{
expecting_operand_after_comma:
- as_bad ("expecting operand after ','; got nothing");
+ as_bad (_("expecting operand after ','; got nothing"));
return;
}
if (*l == ',')
{
- as_bad ("expecting operand before ','; got nothing");
+ as_bad (_("expecting operand before ','; got nothing"));
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);
}
}
- /* Now we've parsed the opcode into a set of templates, and have the
+ /* Now we've parsed the mnemonic into a set of templates, and have the
operands at hand.
Next, we find a template that matches the given insn,
making sure the overlap of the given operands types is consistent
- with the template operand types. */
-
-#define MATCH(overlap,given_type) \
- (overlap && \
- (((overlap & (JumpAbsolute|BaseIndex|Mem8)) \
- == (given_type & (JumpAbsolute|BaseIndex|Mem8))) \
- || (overlap == InOutPortReg)))
-
-
- /* If m0 and m1 are register matches they must be consistent
- with the expected operand types t0 and t1.
- That is, if both m0 & m1 are register matches
- i.e. ( ((m0 & (Reg)) && (m1 & (Reg)) ) ?
- then, either 1. or 2. must be true:
- 1. the expected operand type register overlap is null:
- (t0 & t1 & Reg) == 0
- AND
- the given register overlap is null:
- (m0 & m1 & Reg) == 0
- 2. the expected operand type register overlap == the given
- operand type overlap: (t0 & t1 & m0 & m1 & Reg).
- */
-#define CONSISTENT_REGISTER_MATCH(m0, m1, t0, t1) \
- ( ((m0 & (Reg)) && (m1 & (Reg))) ? \
- ( ((t0 & t1 & (Reg)) == 0 && (m0 & m1 & (Reg)) == 0) || \
- ((t0 & t1) & (m0 & m1) & (Reg)) \
- ) : 1)
+ with the template operand types. */
+
+#define MATCH(overlap, given, template) \
+ ((overlap & ~JumpAbsolute) \
+ && ((given) & (BaseIndex|JumpAbsolute)) == ((overlap) & (BaseIndex|JumpAbsolute)))
+
+ /* If given types r0 and r1 are registers they must be of the same type
+ unless the expected operand type register overlap is null.
+ Note that Acc in a template matches every size of reg. */
+#define CONSISTENT_REGISTER_MATCH(m0, g0, t0, m1, g1, t1) \
+ ( ((g0) & Reg) == 0 || ((g1) & Reg) == 0 || \
+ ((g0) & Reg) == ((g1) & Reg) || \
+ ((((m0) & Acc) ? Reg : (t0)) & (((m1) & Acc) ? Reg : (t1)) & Reg) == 0 )
+
{
register unsigned int overlap0, overlap1;
- expressionS *exp;
unsigned int overlap2;
unsigned int found_reverse_match;
+ int suffix_check;
+
+ /* All intel opcodes have reversed operands except for "bound" and
+ "enter". We also don't reverse intersegment "jmp" and "call"
+ instructions with 2 immediate operands so that the immediate segment
+ precedes the offset, as it does when in AT&T mode. "enter" and the
+ intersegment "jmp" and "call" instructions are the only ones that
+ have two immediate operands. */
+ if (intel_syntax && i.operands > 1
+ && (strcmp (mnemonic, "bound") != 0)
+ && !((i.types[0] & Imm) && (i.types[1] & Imm)))
+ {
+ union i386_op temp_op;
+ unsigned int temp_type;
+#ifdef BFD_ASSEMBLER
+ enum bfd_reloc_code_real temp_disp_reloc;
+#else
+ int temp_disp_reloc;
+#endif
+ int xchg1 = 0;
+ int xchg2 = 0;
+
+ if (i.operands == 2)
+ {
+ xchg1 = 0;
+ xchg2 = 1;
+ }
+ else if (i.operands == 3)
+ {
+ xchg1 = 0;
+ xchg2 = 2;
+ }
+ temp_type = i.types[xchg2];
+ i.types[xchg2] = i.types[xchg1];
+ i.types[xchg1] = temp_type;
+ temp_op = i.op[xchg2];
+ i.op[xchg2] = i.op[xchg1];
+ i.op[xchg1] = temp_op;
+ temp_disp_reloc = i.disp_reloc[xchg2];
+ i.disp_reloc[xchg2] = i.disp_reloc[xchg1];
+ i.disp_reloc[xchg1] = temp_disp_reloc;
+
+ if (i.mem_operands == 2)
+ {
+ const seg_entry *temp_seg;
+ temp_seg = i.seg[0];
+ i.seg[0] = i.seg[1];
+ i.seg[1] = temp_seg;
+ }
+ }
+
+ if (i.imm_operands)
+ {
+ /* Try to ensure constant immediates are represented in the smallest
+ opcode possible. */
+ char guess_suffix = 0;
+ int op;
+
+ if (i.suffix)
+ guess_suffix = i.suffix;
+ else if (i.reg_operands)
+ {
+ /* Figure out a suffix from the last register operand specified.
+ We can't do this properly yet, ie. excluding InOutPortReg,
+ but the following works for instructions with immediates.
+ In any case, we can't set i.suffix yet. */
+ for (op = i.operands; --op >= 0;)
+ if (i.types[op] & Reg)
+ {
+ if (i.types[op] & Reg8)
+ guess_suffix = BYTE_MNEM_SUFFIX;
+ else if (i.types[op] & Reg16)
+ guess_suffix = WORD_MNEM_SUFFIX;
+ break;
+ }
+ }
+ else if (flag_16bit_code ^ (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 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;
+ }
+
+ /* 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);
+ }
+ i.types[op] |= smallest_imm_type ((long) i.op[op].imms->X_add_number);
+ }
+ }
+
+ if (i.disp_operands)
+ {
+ /* Try to use the smallest displacement type too. */
+ int op;
+
+ for (op = i.operands; --op >= 0;)
+ if ((i.types[op] & Disp)
+ && i.op[op].imms->X_op == O_constant)
+ {
+ offsetT disp = i.op[op].disps->X_add_number;
+
+ if (i.types[op] & Disp16)
+ {
+ /* We know this operand is at most 16 bits, so
+ convert to a signed 16 bit number before trying
+ to see whether it will fit in an even smaller
+ size. */
+
+ disp = (((disp & 0xffff) ^ 0x8000) - 0x8000);
+ }
+ if (fits_in_signed_byte (disp))
+ i.types[op] |= Disp8;
+ }
+ }
+
+ overlap0 = 0;
+ overlap1 = 0;
+ overlap2 = 0;
+ found_reverse_match = 0;
+ suffix_check = (i.suffix == BYTE_MNEM_SUFFIX
+ ? No_bSuf
+ : (i.suffix == WORD_MNEM_SUFFIX
+ ? No_wSuf
+ : (i.suffix == SHORT_MNEM_SUFFIX
+ ? No_sSuf
+ : (i.suffix == LONG_MNEM_SUFFIX
+ ? No_lSuf
+ : (i.suffix == DWORD_MNEM_SUFFIX
+ ? No_dSuf
+ : (i.suffix == LONG_DOUBLE_MNEM_SUFFIX ? No_xSuf : 0))))));
- overlap0 = overlap1 = overlap2 = found_reverse_match = 0;
for (t = current_templates->start;
t < current_templates->end;
t++)
{
-
- /* must have right number of operands */
+ /* Must have right number of operands. */
if (i.operands != t->operands)
continue;
+
+ /* 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" */
+ continue;
+
else if (!t->operands)
- break; /* 0 operands always matches */
+ /* 0 operands always matches. */
+ break;
overlap0 = i.types[0] & t->operand_types[0];
switch (t->operands)
{
case 1:
- if (!MATCH (overlap0, i.types[0]))
+ if (!MATCH (overlap0, i.types[0], t->operand_types[0]))
continue;
break;
case 2:
case 3:
overlap1 = i.types[1] & t->operand_types[1];
- if (!MATCH (overlap0, i.types[0]) ||
- !MATCH (overlap1, i.types[1]) ||
- !CONSISTENT_REGISTER_MATCH (overlap0, overlap1,
- t->operand_types[0],
- t->operand_types[1]))
+ if (!MATCH (overlap0, i.types[0], t->operand_types[0])
+ || !MATCH (overlap1, i.types[1], t->operand_types[1])
+ || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
+ t->operand_types[0],
+ overlap1, i.types[1],
+ t->operand_types[1]))
{
-
- /* check if other direction is valid ... */
- if (!(t->opcode_modifier & COMES_IN_BOTH_DIRECTIONS))
+ /* Check if other direction is valid ... */
+ if ((t->opcode_modifier & (D|FloatD)) == 0)
continue;
- /* try reversing direction of operands */
+ /* Try reversing direction of operands. */
overlap0 = i.types[0] & t->operand_types[1];
overlap1 = i.types[1] & t->operand_types[0];
- if (!MATCH (overlap0, i.types[0]) ||
- !MATCH (overlap1, i.types[1]) ||
- !CONSISTENT_REGISTER_MATCH (overlap0, overlap1,
- t->operand_types[0],
- t->operand_types[1]))
+ if (!MATCH (overlap0, i.types[0], t->operand_types[1])
+ || !MATCH (overlap1, i.types[1], t->operand_types[0])
+ || !CONSISTENT_REGISTER_MATCH (overlap0, i.types[0],
+ t->operand_types[1],
+ overlap1, i.types[1],
+ t->operand_types[0]))
{
- /* does not match either direction */
+ /* Does not match either direction. */
continue;
}
- /* found a reverse match here -- slip through */
- /* found_reverse_match holds which of D or FloatD we've found */
- found_reverse_match = t->opcode_modifier & COMES_IN_BOTH_DIRECTIONS;
- } /* endif: not forward match */
- /* found either forward/reverse 2 operand match here */
+ /* 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)
{
+ /* Here we make use of the fact that there are no
+ reverse match 3 operand instructions, and all 3
+ operand instructions only need to be checked for
+ register consistency between operands 2 and 3. */
overlap2 = i.types[2] & t->operand_types[2];
- if (!MATCH (overlap2, i.types[2]) ||
- !CONSISTENT_REGISTER_MATCH (overlap0, overlap2,
- t->operand_types[0],
- t->operand_types[2]) ||
- !CONSISTENT_REGISTER_MATCH (overlap1, overlap2,
- t->operand_types[1],
- t->operand_types[2]))
+ if (!MATCH (overlap2, i.types[2], t->operand_types[2])
+ || !CONSISTENT_REGISTER_MATCH (overlap1, i.types[1],
+ t->operand_types[1],
+ overlap2, i.types[2],
+ t->operand_types[2]))
+
continue;
}
- /* found either forward/reverse 2 or 3 operand match here:
- slip through to break */
+ /* Found either forward/reverse 2 or 3 operand match here:
+ slip through to break. */
}
- break; /* we've found a match; break out of loop */
- } /* for (t = ... */
+ /* We've found a match; break out of loop. */
+ break;
+ }
if (t == current_templates->end)
- { /* we found no match */
- as_bad ("operands given don't match any known 386 instruction");
+ {
+ /* We found no match. */
+ as_bad (_("suffix or operands invalid for `%s'"),
+ current_templates->start->name);
return;
}
- /* Copy the template we found (we may change it!). */
- memcpy (&i.tm, t, sizeof (template));
- t = &i.tm; /* alter new copy of template */
+ if (!quiet_warnings)
+ {
+ if (!intel_syntax
+ && ((i.types[0] & JumpAbsolute)
+ != (t->operand_types[0] & JumpAbsolute)))
+ {
+ as_warn (_("indirect %s without `*'"), t->name);
+ }
+
+ if ((t->opcode_modifier & (IsPrefix|IgnoreSize))
+ == (IsPrefix|IgnoreSize))
+ {
+ /* Warn them that a data or address size prefix doesn't
+ affect assembly of the next line of code. */
+ as_warn (_("stand-alone `%s' prefix"), t->name);
+ }
+ }
+
+ /* Copy the template we found. */
+ i.tm = *t;
+ if (found_reverse_match)
+ {
+ /* If we found a reverse match we must alter the opcode
+ direction bit. found_reverse_match holds bits to change
+ (different for int & float insns). */
+
+ i.tm.base_opcode ^= found_reverse_match;
+
+ i.tm.operand_types[0] = t->operand_types[1];
+ i.tm.operand_types[1] = t->operand_types[0];
+ }
+
+ /* Undo SYSV386_COMPAT brokenness when in Intel mode. See i386.h */
+ if (SYSV386_COMPAT
+ && intel_syntax
+ && (i.tm.base_opcode & 0xfffffde0) == 0xdce0)
+ i.tm.base_opcode ^= FloatR;
+
+ if (i.tm.opcode_modifier & FWait)
+ if (! add_prefix (FWAIT_OPCODE))
+ return;
- /* If there's no opcode suffix we try to invent one based on register
- operands. */
- if (!i.suffix && i.reg_operands)
+ /* Check string instruction segment overrides. */
+ if ((i.tm.opcode_modifier & IsString) != 0 && i.mem_operands != 0)
{
- /* We take i.suffix from the LAST register operand specified. This
- assumes that the last register operands is the destination register
- operand. */
- int o;
- for (o = 0; o < MAX_OPERANDS; o++)
- if (i.types[o] & Reg)
- {
- i.suffix = (i.types[o] == Reg8) ? BYTE_OPCODE_SUFFIX :
- (i.types[o] == Reg16) ? WORD_OPCODE_SUFFIX :
- DWORD_OPCODE_SUFFIX;
- }
+ int mem_op = (i.types[0] & AnyMem) ? 0 : 1;
+ if ((i.tm.operand_types[mem_op] & EsSeg) != 0)
+ {
+ if (i.seg[0] != NULL && i.seg[0] != &es)
+ {
+ as_bad (_("`%s' operand %d must use `%%es' segment"),
+ i.tm.name,
+ mem_op + 1);
+ return;
+ }
+ /* There's only ever one segment override allowed per instruction.
+ This instruction possibly has a legal segment override on the
+ second operand, so copy the segment to where non-string
+ instructions store it, allowing common code. */
+ i.seg[0] = i.seg[1];
+ }
+ else if ((i.tm.operand_types[mem_op + 1] & EsSeg) != 0)
+ {
+ if (i.seg[1] != NULL && i.seg[1] != &es)
+ {
+ as_bad (_("`%s' operand %d must use `%%es' segment"),
+ i.tm.name,
+ mem_op + 2);
+ return;
+ }
+ }
+ }
+
+ /* If matched instruction specifies an explicit instruction mnemonic
+ suffix, use it. */
+ if (i.tm.opcode_modifier & (Size16 | Size32))
+ {
+ if (i.tm.opcode_modifier & Size16)
+ i.suffix = WORD_MNEM_SUFFIX;
+ else
+ i.suffix = LONG_MNEM_SUFFIX;
+ }
+ else if (i.reg_operands)
+ {
+ /* If there's no instruction mnemonic suffix we try to invent one
+ based on register operands. */
+ if (!i.suffix)
+ {
+ /* We take i.suffix from the last register operand specified,
+ Destination register type is more significant than source
+ register type. */
+ int op;
+ for (op = i.operands; --op >= 0;)
+ if ((i.types[op] & Reg)
+ && !(i.tm.operand_types[op] & InOutPortReg))
+ {
+ i.suffix = ((i.types[op] & Reg8) ? BYTE_MNEM_SUFFIX :
+ (i.types[op] & Reg16) ? WORD_MNEM_SUFFIX :
+ LONG_MNEM_SUFFIX);
+ break;
+ }
+ }
+ else if (i.suffix == BYTE_MNEM_SUFFIX)
+ {
+ int op;
+ for (op = i.operands; --op >= 0;)
+ {
+ /* If this is an eight bit register, it's OK. If it's
+ the 16 or 32 bit version of an eight bit register,
+ we will just use the low portion, and that's OK too. */
+ if (i.types[op] & Reg8)
+ continue;
+
+ /* movzx and movsx should not generate this warning. */
+ if (intel_syntax
+ && (i.tm.base_opcode == 0xfb7
+ || i.tm.base_opcode == 0xfb6
+ || i.tm.base_opcode == 0xfbe
+ || i.tm.base_opcode == 0xfbf))
+ continue;
+
+ if ((i.types[op] & WordReg) && i.op[op].regs->reg_num < 4
+#if 0
+ /* Check that the template allows eight bit regs
+ This kills insns such as `orb $1,%edx', which
+ maybe should be allowed. */
+ && (i.tm.operand_types[op] & (Reg8|InOutPortReg))
+#endif
+ )
+ {
+#if REGISTER_WARNINGS
+ if (!quiet_warnings
+ && (i.tm.operand_types[op] & InOutPortReg) == 0)
+ as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
+ (i.op[op].regs - (i.types[op] & Reg16 ? 8 : 16))->reg_name,
+ i.op[op].regs->reg_name,
+ i.suffix);
+#endif
+ continue;
+ }
+ /* Any other register is bad. */
+ if (i.types[op] & (Reg | RegMMX | RegXMM
+ | SReg2 | SReg3
+ | Control | Debug | Test
+ | FloatReg | FloatAcc))
+ {
+ as_bad (_("`%%%s' not allowed with `%s%c'"),
+ i.op[op].regs->reg_name,
+ i.tm.name,
+ i.suffix);
+ return;
+ }
+ }
+ }
+ else if (i.suffix == LONG_MNEM_SUFFIX)
+ {
+ int op;
+
+ 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;
+ }
+#if REGISTER_WARNINGS
+ /* Warn if the e prefix on a general reg is missing. */
+ else if (!quiet_warnings
+ && (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);
+ }
+#endif
+ }
+ else if (i.suffix == WORD_MNEM_SUFFIX)
+ {
+ int op;
+ 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;
+ }
+#if REGISTER_WARNINGS
+ /* Warn if the e prefix on a general reg is present. */
+ else if (!quiet_warnings
+ && (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);
+ }
+#endif
+ }
+ else if (intel_syntax && (i.tm.opcode_modifier & IgnoreSize))
+ /* Do nothing if the instruction is going to ignore the prefix. */
+ ;
+ else
+ abort ();
+ }
+ else if ((i.tm.opcode_modifier & DefaultSize) && !i.suffix)
+ {
+ i.suffix = stackop_size;
}
/* Make still unresolved immediate matches conform to size of immediate
- given in i.suffix. Note: overlap2 cannot be an immediate!
- We assume this. */
+ given in i.suffix. Note: overlap2 cannot be an immediate! */
if ((overlap0 & (Imm8 | Imm8S | Imm16 | Imm32))
&& overlap0 != Imm8 && overlap0 != Imm8S
&& overlap0 != Imm16 && overlap0 != Imm32)
{
- if (!i.suffix)
+ if (i.suffix)
+ {
+ overlap0 &= (i.suffix == BYTE_MNEM_SUFFIX ? (Imm8 | Imm8S) :
+ (i.suffix == WORD_MNEM_SUFFIX ? Imm16 : Imm32));
+ }
+ else if (overlap0 == (Imm16 | Imm32))
{
- as_bad ("no opcode suffix given; can't determine immediate size");
+ overlap0 =
+ (flag_16bit_code ^ (i.prefix[DATA_PREFIX] != 0)) ? Imm16 : Imm32;
+ }
+ else
+ {
+ as_bad (_("no instruction mnemonic suffix given; can't determine immediate size"));
return;
}
- overlap0 &= (i.suffix == BYTE_OPCODE_SUFFIX ? (Imm8 | Imm8S) :
- (i.suffix == WORD_OPCODE_SUFFIX ? Imm16 : Imm32));
}
if ((overlap1 & (Imm8 | Imm8S | Imm16 | Imm32))
&& overlap1 != Imm8 && overlap1 != Imm8S
&& overlap1 != Imm16 && overlap1 != Imm32)
{
- if (!i.suffix)
+ if (i.suffix)
+ {
+ overlap1 &= (i.suffix == BYTE_MNEM_SUFFIX ? (Imm8 | Imm8S) :
+ (i.suffix == WORD_MNEM_SUFFIX ? Imm16 : Imm32));
+ }
+ else if (overlap1 == (Imm16 | Imm32))
+ {
+ overlap1 =
+ (flag_16bit_code ^ (i.prefix[DATA_PREFIX] != 0)) ? Imm16 : Imm32;
+ }
+ else
{
- as_bad ("no opcode suffix given; can't determine immediate size");
+ as_bad (_("no instruction mnemonic suffix given; can't determine immediate size"));
return;
}
- overlap1 &= (i.suffix == BYTE_OPCODE_SUFFIX ? (Imm8 | Imm8S) :
- (i.suffix == WORD_OPCODE_SUFFIX ? Imm16 : Imm32));
}
+ assert ((overlap2 & Imm) == 0);
i.types[0] = overlap0;
- i.types[1] = overlap1;
- i.types[2] = overlap2;
-
if (overlap0 & ImplicitRegister)
i.reg_operands--;
+ if (overlap0 & Imm1)
+ i.imm_operands = 0; /* kludge for shift insns. */
+
+ i.types[1] = overlap1;
if (overlap1 & ImplicitRegister)
i.reg_operands--;
+
+ i.types[2] = overlap2;
if (overlap2 & ImplicitRegister)
i.reg_operands--;
- if (overlap0 & Imm1)
- i.imm_operands = 0; /* kludge for shift insns */
-
- if (found_reverse_match)
- {
- unsigned int save;
- save = t->operand_types[0];
- t->operand_types[0] = t->operand_types[1];
- t->operand_types[1] = save;
- }
/* Finalize opcode. First, we change the opcode based on the operand
- size given by i.suffix: we never have to change things for byte insns,
- or when no opcode suffix is need to size the operands. */
+ size given by i.suffix: We need not change things for byte insns. */
- if (!i.suffix && (t->opcode_modifier & W))
+ if (!i.suffix && (i.tm.opcode_modifier & W))
{
- as_bad ("no opcode suffix given and no register operands; can't size instruction");
+ as_bad (_("no instruction mnemonic suffix given and no register operands; can't size instruction"));
return;
}
- if (i.suffix && i.suffix != BYTE_OPCODE_SUFFIX)
+ /* 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)
+ {
+ unsigned int prefix = DATA_PREFIX_OPCODE;
+
+ if ((i.op[1].regs->reg_type & Reg16) != 0)
+ if (!add_prefix (prefix))
+ return;
+ }
+
+ if (i.suffix && i.suffix != BYTE_MNEM_SUFFIX)
{
- /* Select between byte and word/dword operations. */
- if (t->opcode_modifier & W)
- t->base_opcode |= W;
- /* Now select between word & dword operations via the
- operand size prefix. */
- if (i.suffix == WORD_OPCODE_SUFFIX)
+ /* It's not a byte, select word/dword operation. */
+ if (i.tm.opcode_modifier & W)
{
- if (i.prefixes == MAX_PREFIXES)
- {
- as_bad ("%d prefixes given and 'w' opcode suffix gives too many prefixes",
- MAX_PREFIXES);
- return;
- }
- i.prefix[i.prefixes++] = WORD_PREFIX_OPCODE;
+ if (i.tm.opcode_modifier & ShortForm)
+ i.tm.base_opcode |= 8;
+ else
+ i.tm.base_opcode |= 1;
+ }
+ /* 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
+ && !(i.tm.opcode_modifier & IgnoreSize))
+ {
+ unsigned int prefix = DATA_PREFIX_OPCODE;
+ if (i.tm.opcode_modifier & JumpByte) /* jcxz, loop */
+ prefix = ADDR_PREFIX_OPCODE;
+
+ if (! add_prefix (prefix))
+ return;
+ }
+ /* Size floating point instruction. */
+ if (i.suffix == LONG_MNEM_SUFFIX
+ || (intel_syntax && i.suffix == DWORD_MNEM_SUFFIX))
+ {
+ if (i.tm.opcode_modifier & FloatMF)
+ i.tm.base_opcode ^= 4;
}
}
- /* For insns with operands there are more diddles to do to the opcode. */
- if (i.operands)
+ if (i.tm.opcode_modifier & ImmExt)
{
- /* If we found a reverse match we must alter the opcode direction bit
- found_reverse_match holds bit to set (different for int &
- float insns). */
+ /* These AMD 3DNow! and Intel Katmai New Instructions have an
+ opcode suffix which is coded in the same place as an 8-bit
+ immediate field would be. Here we fake an 8-bit immediate
+ operand from the opcode suffix stored in tm.extension_opcode. */
- if (found_reverse_match)
- {
- t->base_opcode |= found_reverse_match;
- }
+ expressionS *exp;
+
+ assert (i.imm_operands == 0 && i.operands <= 2 && 2 < MAX_OPERANDS);
+
+ exp = &im_expressions[i.imm_operands++];
+ i.op[i.operands].imms = exp;
+ i.types[i.operands++] = Imm8;
+ exp->X_op = O_constant;
+ exp->X_add_number = i.tm.extension_opcode;
+ i.tm.extension_opcode = None;
+ }
+
+ /* For insns with operands there are more diddles to do to the opcode. */
+ if (i.operands)
+ {
+ /* Default segment register this instruction will use
+ for memory accesses. 0 means unknown.
+ This is only for optimizing out unnecessary segment overrides. */
+ const seg_entry *default_seg = 0;
/* The imul $imm, %reg instruction is converted into
- imul $imm, %reg, %reg. */
- if (t->opcode_modifier & imulKludge)
+ imul $imm, %reg, %reg, and the clr %reg instruction
+ is converted into xor %reg, %reg. */
+ if (i.tm.opcode_modifier & regKludge)
{
- /* Pretend we saw the 3 operand case. */
- i.regs[2] = i.regs[1];
+ unsigned int first_reg_op = (i.types[0] & Reg) ? 0 : 1;
+ /* Pretend we saw the extra register operand. */
+ assert (i.op[first_reg_op + 1].regs == 0);
+ i.op[first_reg_op + 1].regs = i.op[first_reg_op].regs;
+ i.types[first_reg_op + 1] = i.types[first_reg_op];
i.reg_operands = 2;
}
- /* Certain instructions expect the destination to be in the i.rm.reg
- field. This is by far the exceptional case. For these
- instructions, if the source operand is a register, we must reverse
- the i.rm.reg and i.rm.regmem fields. We accomplish this by faking
- that the two register operands were given in the reverse order. */
- if ((t->opcode_modifier & ReverseRegRegmem) && i.reg_operands == 2)
- {
- unsigned int first_reg_operand = (i.types[0] & Reg) ? 0 : 1;
- unsigned int second_reg_operand = first_reg_operand + 1;
- reg_entry *tmp = i.regs[first_reg_operand];
- i.regs[first_reg_operand] = i.regs[second_reg_operand];
- i.regs[second_reg_operand] = tmp;
- }
-
- if (t->opcode_modifier & ShortForm)
- {
- /* The register or float register operand is in operand 0 or 1. */
- unsigned int o = (i.types[0] & (Reg | FloatReg)) ? 0 : 1;
- /* Register goes in low 3 bits of opcode. */
- t->base_opcode |= i.regs[o]->reg_num;
- }
- else if (t->opcode_modifier & ShortFormW)
- {
- /* Short form with 0x8 width bit. Register is always dest. operand */
- t->base_opcode |= i.regs[1]->reg_num;
- if (i.suffix == WORD_OPCODE_SUFFIX ||
- i.suffix == DWORD_OPCODE_SUFFIX)
- t->base_opcode |= 0x8;
- }
- else if (t->opcode_modifier & Seg2ShortForm)
+ if (i.tm.opcode_modifier & ShortForm)
{
- if (t->base_opcode == POP_SEG_SHORT && i.regs[0]->reg_num == 1)
+ /* The register or float register operand is in operand 0 or 1. */
+ 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 (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
{
- as_bad ("you can't 'pop cs' on the 386.");
- return;
+ /* Warn about some common errors, but press on regardless.
+ The first case can be generated by gcc (<= 2.8.1). */
+ if (i.operands == 2)
+ {
+ /* Reversed arguments on faddp, fsubp, etc. */
+ as_warn (_("translating to `%s %%%s,%%%s'"), i.tm.name,
+ i.op[1].regs->reg_name,
+ i.op[0].regs->reg_name);
+ }
+ else
+ {
+ /* Extraneous `l' suffix on fp insn. */
+ as_warn (_("translating to `%s %%%s'"), i.tm.name,
+ i.op[0].regs->reg_name);
+ }
}
- t->base_opcode |= (i.regs[0]->reg_num << 3);
}
- else if (t->opcode_modifier & Seg3ShortForm)
- {
- /* 'push %fs' is 0x0fa0; 'pop %fs' is 0x0fa1.
- 'push %gs' is 0x0fa8; 'pop %fs' is 0x0fa9.
- So, only if i.regs[0]->reg_num == 5 (%gs) do we need
- to change the opcode. */
- if (i.regs[0]->reg_num == 5)
- t->base_opcode |= 0x08;
- }
- else if (t->opcode_modifier & Modrm)
+ else if (i.tm.opcode_modifier & Modrm)
{
- /* The opcode is completed (modulo t->extension_opcode which must
- be put into the modrm byte.
- Now, we make the modrm & index base bytes based on all the info
- we've collected. */
+ /* The opcode is completed (modulo i.tm.extension_opcode which
+ must be put into the modrm byte).
+ Now, we make the modrm & index base bytes based on all the
+ info we've collected. */
/* i.reg_operands MUST be the number of real register operands;
- implicit registers do not count. */
+ implicit registers do not count. */
if (i.reg_operands == 2)
{
unsigned int source, dest;
- source = (i.types[0] & (Reg | SReg2 | SReg3 | Control | Debug | Test)) ? 0 : 1;
+ source = ((i.types[0]
+ & (Reg | RegMMX | RegXMM
+ | SReg2 | SReg3
+ | Control | Debug | Test))
+ ? 0 : 1);
dest = source + 1;
+
i.rm.mode = 3;
- /* We must be careful to make sure that all
- segment/control/test/debug registers go into the i.rm.reg
- field (despite the whether they are source or destination
- operands). */
- if (i.regs[dest]->reg_type & (SReg2 | SReg3 | Control | Debug | Test))
+ /* One of the register operands will be encoded in the
+ i.tm.reg field, the other in the combined i.tm.mode
+ and i.tm.regmem fields. If no form of this
+ instruction supports a memory destination operand,
+ then we assume the source operand may sometimes be
+ a memory operand and so we need to store the
+ destination in the i.rm.reg field. */
+ if ((i.tm.operand_types[dest] & AnyMem) == 0)
{
- i.rm.reg = i.regs[dest]->reg_num;
- i.rm.regmem = i.regs[source]->reg_num;
+ i.rm.reg = i.op[dest].regs->reg_num;
+ i.rm.regmem = i.op[source].regs->reg_num;
}
else
{
- i.rm.reg = i.regs[source]->reg_num;
- i.rm.regmem = i.regs[dest]->reg_num;
+ i.rm.reg = i.op[source].regs->reg_num;
+ i.rm.regmem = i.op[dest].regs->reg_num;
}
}
else
- { /* if it's not 2 reg operands... */
+ { /* If it's not 2 reg operands... */
if (i.mem_operands)
{
unsigned int fake_zero_displacement = 0;
- unsigned int o = (i.types[0] & Mem) ? 0 : ((i.types[1] & Mem) ? 1 : 2);
+ unsigned int op = ((i.types[0] & AnyMem)
+ ? 0
+ : (i.types[1] & AnyMem) ? 1 : 2);
- /* Encode memory operand into modrm byte and base index
- byte. */
+ default_seg = &ds;
- if (i.base_reg == esp && !i.index_reg)
- {
- /* <disp>(%esp) becomes two byte modrm with no index
- register. */
- i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
- i.rm.mode = mode_from_disp_size (i.types[o]);
- i.bi.base = ESP_REG_NUM;
- i.bi.index = NO_INDEX_REGISTER;
- i.bi.scale = 0; /* Must be zero! */
- }
- else if (i.base_reg == ebp && !i.index_reg)
+ if (! i.base_reg)
{
- if (!(i.types[o] & Disp))
+ i.rm.mode = 0;
+ if (! i.disp_operands)
+ fake_zero_displacement = 1;
+ if (! i.index_reg)
{
- /* Must fake a zero byte displacement. There is
- no direct way to code '(%ebp)' directly. */
- fake_zero_displacement = 1;
- /* fake_zero_displacement code does not set this. */
- i.types[o] |= Disp8;
- }
- i.rm.mode = mode_from_disp_size (i.types[o]);
- i.rm.regmem = EBP_REG_NUM;
- }
- else if (!i.base_reg && (i.types[o] & BaseIndex))
- {
- /* There are three cases here.
- Case 1: '<32bit disp>(,1)' -- indirect absolute.
- (Same as cases 2 & 3 with NO index register)
- Case 2: <32bit disp> (,<index>) -- no base register with disp
- Case 3: (, <index>) --- no base register;
- no disp (must add 32bit 0 disp). */
- i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
- i.rm.mode = 0; /* 32bit mode */
- i.bi.base = NO_BASE_REGISTER;
- i.types[o] &= ~Disp;
- i.types[o] |= Disp32; /* Must be 32bit! */
- if (i.index_reg)
- { /* case 2 or case 3 */
- i.bi.index = i.index_reg->reg_num;
- i.bi.scale = i.log2_scale_factor;
- if (i.disp_operands == 0)
- fake_zero_displacement = 1; /* case 3 */
+ /* Operand is just <disp> */
+ if (flag_16bit_code ^ (i.prefix[ADDR_PREFIX] != 0))
+ {
+ i.rm.regmem = NO_BASE_REGISTER_16;
+ i.types[op] &= ~Disp;
+ i.types[op] |= Disp16;
+ }
+ else
+ {
+ i.rm.regmem = NO_BASE_REGISTER;
+ i.types[op] &= ~Disp;
+ i.types[op] |= Disp32;
+ }
}
- else
+ else /* ! i.base_reg && i.index_reg */
{
- i.bi.index = NO_INDEX_REGISTER;
- i.bi.scale = 0;
+ i.sib.index = i.index_reg->reg_num;
+ i.sib.base = NO_BASE_REGISTER;
+ 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. */
}
}
- else if (i.disp_operands && !i.base_reg && !i.index_reg)
+ else if (i.base_reg->reg_type & Reg16)
{
- /* Operand is just <32bit disp> */
- i.rm.regmem = EBP_REG_NUM;
- i.rm.mode = 0;
- i.types[o] &= ~Disp;
- i.types[o] |= Disp32;
+ switch (i.base_reg->reg_num)
+ {
+ case 3: /* (%bx) */
+ if (! i.index_reg)
+ i.rm.regmem = 7;
+ else /* (%bx,%si) -> 0, or (%bx,%di) -> 1 */
+ i.rm.regmem = i.index_reg->reg_num - 6;
+ break;
+ case 5: /* (%bp) */
+ default_seg = &ss;
+ if (! i.index_reg)
+ {
+ i.rm.regmem = 6;
+ if ((i.types[op] & Disp) == 0)
+ {
+ /* fake (%bp) into 0(%bp) */
+ i.types[op] |= Disp8;
+ fake_zero_displacement = 1;
+ }
+ }
+ else /* (%bp,%si) -> 2, or (%bp,%di) -> 3 */
+ i.rm.regmem = i.index_reg->reg_num - 6 + 2;
+ break;
+ default: /* (%si) -> 4 or (%di) -> 5 */
+ i.rm.regmem = i.base_reg->reg_num - 6 + 4;
+ }
+ i.rm.mode = mode_from_disp_size (i.types[op]);
}
- else
+ else /* i.base_reg and 32 bit mode */
{
- /* It's not a special case; rev'em up. */
i.rm.regmem = i.base_reg->reg_num;
- i.rm.mode = mode_from_disp_size (i.types[o]);
- if (i.index_reg)
+ i.sib.base = i.base_reg->reg_num;
+ if (i.base_reg->reg_num == EBP_REG_NUM)
{
- i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
- i.bi.base = i.base_reg->reg_num;
- i.bi.index = i.index_reg->reg_num;
- i.bi.scale = i.log2_scale_factor;
- if (i.base_reg == ebp && i.disp_operands == 0)
- { /* pace */
+ default_seg = &ss;
+ if (i.disp_operands == 0)
+ {
fake_zero_displacement = 1;
- i.types[o] |= Disp8;
- i.rm.mode = mode_from_disp_size (i.types[o]);
+ i.types[op] |= Disp8;
}
}
- }
- if (fake_zero_displacement)
- {
- /* Fakes a zero displacement assuming that i.types[o]
- holds the correct displacement size. */
- exp = &disp_expressions[i.disp_operands++];
- i.disps[o] = exp;
- exp->X_seg = absolute_section;
- exp->X_add_number = 0;
- exp->X_add_symbol = (symbolS *) 0;
- exp->X_subtract_symbol = (symbolS *) 0;
- }
-
- /* Select the correct segment for the memory operand. */
- if (i.seg)
- {
- unsigned int seg_index;
- const seg_entry *default_seg;
-
- if (i.rm.regmem == ESCAPE_TO_TWO_BYTE_ADDRESSING)
+ else if (i.base_reg->reg_num == ESP_REG_NUM)
{
- seg_index = (i.rm.mode << 3) | i.bi.base;
- default_seg = two_byte_segment_defaults[seg_index];
+ default_seg = &ss;
}
- else
+ i.sib.scale = i.log2_scale_factor;
+ if (! i.index_reg)
{
- seg_index = (i.rm.mode << 3) | i.rm.regmem;
- default_seg = one_byte_segment_defaults[seg_index];
+ /* <disp>(%esp) becomes two byte modrm
+ with no index register. We've already
+ stored the code for esp in i.rm.regmem
+ ie. ESCAPE_TO_TWO_BYTE_ADDRESSING. Any
+ base register besides %esp will not use
+ the extra modrm byte. */
+ i.sib.index = NO_INDEX_REGISTER;
+#if ! SCALE1_WHEN_NO_INDEX
+ /* Another case where we force the second
+ modrm byte. */
+ if (i.log2_scale_factor)
+ i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
+#endif
}
- /* If the specified segment is not the default, use an
- opcode prefix to select it */
- if (i.seg != default_seg)
+ else
{
- if (i.prefixes == MAX_PREFIXES)
- {
- as_bad ("%d prefixes given and %s segment override gives too many prefixes",
- MAX_PREFIXES, i.seg->seg_name);
- return;
- }
- i.prefix[i.prefixes++] = i.seg->seg_prefix;
+ i.sib.index = i.index_reg->reg_num;
+ i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
}
+ i.rm.mode = mode_from_disp_size (i.types[op]);
+ }
+
+ if (fake_zero_displacement)
+ {
+ /* Fakes a zero displacement assuming that i.types[op]
+ holds the correct displacement size. */
+ expressionS *exp;
+
+ assert (i.op[op].disps == 0);
+ exp = &disp_expressions[i.disp_operands++];
+ i.op[op].disps = exp;
+ exp->X_op = O_constant;
+ exp->X_add_number = 0;
+ exp->X_add_symbol = (symbolS *) 0;
+ exp->X_op_symbol = (symbolS *) 0;
}
}
- /* Fill in i.rm.reg or i.rm.regmem field with register operand
- (if any) based on t->extension_opcode. Again, we must be
- careful to make sure that segment/control/debug/test
- registers are coded into the i.rm.reg field. */
+ /* Fill in i.rm.reg or i.rm.regmem field with register
+ operand (if any) based on i.tm.extension_opcode.
+ Again, we must be careful to make sure that
+ segment/control/debug/test/MMX registers are coded
+ into the i.rm.reg field. */
if (i.reg_operands)
{
- unsigned int o =
- (i.types[0] & (Reg | SReg2 | SReg3 | Control | Debug | Test)) ? 0 :
- (i.types[1] & (Reg | SReg2 | SReg3 | Control | Debug | Test)) ? 1 : 2;
+ unsigned int op =
+ ((i.types[0]
+ & (Reg | RegMMX | RegXMM
+ | SReg2 | SReg3
+ | Control | Debug | Test))
+ ? 0
+ : ((i.types[1]
+ & (Reg | RegMMX | RegXMM
+ | SReg2 | SReg3
+ | Control | Debug | Test))
+ ? 1
+ : 2));
/* If there is an extension opcode to put here, the
- register number must be put into the regmem field. */
- if (t->extension_opcode != None)
- i.rm.regmem = i.regs[o]->reg_num;
+ register number must be put into the regmem field. */
+ if (i.tm.extension_opcode != None)
+ i.rm.regmem = i.op[op].regs->reg_num;
else
- i.rm.reg = i.regs[o]->reg_num;
+ i.rm.reg = i.op[op].regs->reg_num;
/* 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
- operand int the regmem field */
+ operand in the regmem field. */
if (!i.mem_operands)
i.rm.mode = 3;
}
- /* Fill in i.rm.reg field with extension opcode (if any). */
- if (t->extension_opcode != None)
- i.rm.reg = t->extension_opcode;
+ /* Fill in i.rm.reg field with extension opcode (if any). */
+ if (i.tm.extension_opcode != None)
+ i.rm.reg = i.tm.extension_opcode;
+ }
+ }
+ else if (i.tm.opcode_modifier & (Seg2ShortForm | Seg3ShortForm))
+ {
+ if (i.tm.base_opcode == POP_SEG_SHORT
+ && i.op[0].regs->reg_num == 1)
+ {
+ as_bad (_("you can't `pop %%cs'"));
+ return;
}
+ i.tm.base_opcode |= (i.op[0].regs->reg_num << 3);
+ }
+ else if ((i.tm.base_opcode & ~(D|W)) == MOV_AX_DISP32)
+ {
+ default_seg = &ds;
}
+ else if ((i.tm.opcode_modifier & IsString) != 0)
+ {
+ /* For the string instructions that allow a segment override
+ on one of their operands, the default segment is ds. */
+ default_seg = &ds;
+ }
+
+ /* If a segment was explicitly specified,
+ and the specified segment is not the default,
+ use an opcode prefix to select it.
+ If we never figured out what the default segment is,
+ then default_seg will be zero at this point,
+ and the specified segment prefix will always be used. */
+ if ((i.seg[0]) && (i.seg[0] != default_seg))
+ {
+ if (! add_prefix (i.seg[0]->seg_prefix))
+ return;
+ }
+ }
+ else if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
+ {
+ /* UnixWare fsub no args is alias for fsubp, fadd -> faddp, etc. */
+ as_warn (_("translating to `%sp'"), i.tm.name);
}
}
- /* Handle conversion of 'int $3' --> special int3 insn. */
- if (t->base_opcode == INT_OPCODE && i.imms[0]->X_add_number == 3)
+ /* Handle conversion of 'int $3' --> special int3 insn. */
+ if (i.tm.base_opcode == INT_OPCODE && i.op[0].imms->X_add_number == 3)
{
- t->base_opcode = INT3_OPCODE;
+ i.tm.base_opcode = INT3_OPCODE;
i.imm_operands = 0;
}
- /* We are ready to output the insn. */
+ if ((i.tm.opcode_modifier & (Jump | JumpByte | JumpDword))
+ && i.op[0].disps->X_op == O_constant)
+ {
+ /* Convert "jmp constant" (and "call constant") to a jump (call) to
+ the absolute address given by the constant. Since ix86 jumps and
+ calls are pc relative, we need to generate a reloc. */
+ i.op[0].disps->X_add_symbol = &abs_symbol;
+ i.op[0].disps->X_op = O_symbol;
+ }
+
+ /* We are ready to output the insn. */
{
register char *p;
- /* Output jumps. */
- if (t->opcode_modifier & Jump)
+ /* Output jumps. */
+ if (i.tm.opcode_modifier & Jump)
{
- int n = i.disps[0]->X_add_number;
- segT seg;
+ int size;
+ int code16;
+ int prefix;
- seg = i.disps[0]->X_seg;
+ code16 = 0;
+ if (flag_16bit_code)
+ code16 = CODE16;
- if (seg == absolute_section)
+ prefix = 0;
+ if (i.prefix[DATA_PREFIX])
{
- if (fits_in_signed_byte (n))
- {
- p = frag_more (2);
- p[0] = t->base_opcode;
- p[1] = n;
- }
-#if 0 /* leave out 16 bit jumps - pace */
- else if (fits_in_signed_word (n))
+ prefix = 1;
+ i.prefixes -= 1;
+ code16 ^= CODE16;
+ }
+
+ size = 4;
+ if (code16)
+ size = 2;
+
+ if (i.prefixes != 0 && !intel_syntax)
+ as_warn (_("skipping prefixes on this instruction"));
+
+ /* It's always a symbol; End frag & setup for relax.
+ Make sure there is enough room in this frag for the largest
+ instruction we may generate in md_convert_frag. This is 2
+ bytes for the opcode and room for the prefix and largest
+ displacement. */
+ frag_grow (prefix + 2 + size);
+ insn_size += prefix + 1;
+ /* Prefix and 1 opcode byte go in fr_fix. */
+ p = frag_more (prefix + 1);
+ if (prefix)
+ *p++ = DATA_PREFIX_OPCODE;
+ *p = i.tm.base_opcode;
+ /* 1 possible extra opcode + displacement go in var part.
+ Pass reloc in fr_var. */
+ frag_var (rs_machine_dependent,
+ 1 + size,
+ i.disp_reloc[0],
+ ((unsigned char) *p == JUMP_PC_RELATIVE
+ ? ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL) | code16
+ : ENCODE_RELAX_STATE (COND_JUMP, SMALL) | code16),
+ i.op[0].disps->X_add_symbol,
+ i.op[0].disps->X_add_number,
+ p);
+ }
+ else if (i.tm.opcode_modifier & (JumpByte | JumpDword))
+ {
+ int size;
+
+ if (i.tm.opcode_modifier & JumpByte)
+ {
+ /* This is a loop or jecxz type instruction. */
+ size = 1;
+ if (i.prefix[ADDR_PREFIX])
{
- p = frag_more (4);
- p[0] = WORD_PREFIX_OPCODE;
- p[1] = t->base_opcode;
- md_number_to_chars (&p[2], n, 2);
- }
-#endif
- else
- { /* It's an absolute dword displacement. */
- if (t->base_opcode == JUMP_PC_RELATIVE)
- { /* pace */
- /* unconditional jump */
- p = frag_more (5);
- p[0] = 0xe9;
- md_number_to_chars (&p[1], n, 4);
- }
- else
- {
- /* conditional jump */
- p = frag_more (6);
- p[0] = TWO_BYTE_OPCODE_ESCAPE;
- p[1] = t->base_opcode + 0x10;
- md_number_to_chars (&p[2], n, 4);
- }
+ insn_size += 1;
+ FRAG_APPEND_1_CHAR (ADDR_PREFIX_OPCODE);
+ i.prefixes -= 1;
}
}
else
{
- /* It's a symbol; end frag & setup for relax.
- Make sure there are more than 6 chars left in the current frag;
- if not we'll have to start a new one. */
- if (obstack_room (&frags) <= 6)
+ int code16;
+
+ code16 = 0;
+ if (flag_16bit_code)
+ code16 = CODE16;
+
+ if (i.prefix[DATA_PREFIX])
{
- frag_wane (frag_now);
- frag_new (0);
+ insn_size += 1;
+ FRAG_APPEND_1_CHAR (DATA_PREFIX_OPCODE);
+ i.prefixes -= 1;
+ code16 ^= CODE16;
}
- p = frag_more (1);
- p[0] = t->base_opcode;
- frag_var (rs_machine_dependent,
- 6, /* 2 opcode/prefix + 4 displacement */
- 1,
- ((unsigned char) *p == JUMP_PC_RELATIVE
- ? ENCODE_RELAX_STATE (UNCOND_JUMP, BYTE)
- : ENCODE_RELAX_STATE (COND_JUMP, BYTE)),
- i.disps[0]->X_add_symbol,
- n, p);
+
+ size = 4;
+ if (code16)
+ size = 2;
}
- }
- else if (t->opcode_modifier & (JumpByte | JumpDword))
- {
- int size = (t->opcode_modifier & JumpByte) ? 1 : 4;
- int n = i.disps[0]->X_add_number;
- if (fits_in_unsigned_byte (t->base_opcode))
+ if (i.prefixes != 0 && !intel_syntax)
+ as_warn (_("skipping prefixes on this instruction"));
+
+ if (fits_in_unsigned_byte (i.tm.base_opcode))
{
- FRAG_APPEND_1_CHAR (t->base_opcode);
+ insn_size += 1 + size;
+ p = frag_more (1 + size);
}
else
{
- p = frag_more (2); /* opcode can be at most two bytes */
- /* put out high byte first: can't use md_number_to_chars! */
- *p++ = (t->base_opcode >> 8) & 0xff;
- *p = t->base_opcode & 0xff;
+ /* Opcode can be at most two bytes. */
+ insn_size += 2 + size;
+ p = frag_more (2 + size);
+ *p++ = (i.tm.base_opcode >> 8) & 0xff;
}
+ *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]));
+ }
+ else if (i.tm.opcode_modifier & JumpInterSegment)
+ {
+ int size;
+ int prefix;
+ int code16;
- p = frag_more (size);
- if (i.disps[0]->X_seg == absolute_section)
+ code16 = 0;
+ if (flag_16bit_code)
+ code16 = CODE16;
+
+ prefix = 0;
+ if (i.prefix[DATA_PREFIX])
{
- md_number_to_chars (p, n, size);
- if (size == 1 && !fits_in_signed_byte (n))
- {
- as_bad ("loop/jecx only takes byte displacement; %d shortened to %d",
- n, *p);
- }
+ prefix = 1;
+ i.prefixes -= 1;
+ code16 ^= CODE16;
}
- else
+
+ size = 4;
+ if (code16)
+ size = 2;
+
+ if (i.prefixes != 0 && !intel_syntax)
+ as_warn (_("skipping prefixes on this instruction"));
+
+ /* 1 opcode; 2 segment; offset */
+ insn_size += prefix + 1 + 2 + size;
+ p = frag_more (prefix + 1 + 2 + size);
+ if (prefix)
+ *p++ = DATA_PREFIX_OPCODE;
+ *p++ = i.tm.base_opcode;
+ if (i.op[1].imms->X_op == O_constant)
{
- fix_new (frag_now, p - frag_now->fr_literal, size,
- i.disps[0]->X_add_symbol, i.disps[0]->X_subtract_symbol,
- i.disps[0]->X_add_number, 1, NO_RELOC);
+ offsetT n = i.op[1].imms->X_add_number;
+
+ if (size == 2
+ && !fits_in_unsigned_word (n)
+ && !fits_in_signed_word (n))
+ {
+ as_bad (_("16-bit jump out of range"));
+ return;
+ }
+ md_number_to_chars (p, n, size);
}
- }
- else if (t->opcode_modifier & JumpInterSegment)
- {
- p = frag_more (1 + 2 + 4); /* 1 opcode; 2 segment; 4 offset */
- p[0] = t->base_opcode;
- if (i.imms[1]->X_seg == absolute_section)
- md_number_to_chars (p + 1, i.imms[1]->X_add_number, 4);
else
- fix_new (frag_now, p + 1 - frag_now->fr_literal, 4,
- i.imms[1]->X_add_symbol,
- i.imms[1]->X_subtract_symbol,
- i.imms[1]->X_add_number, 0, NO_RELOC);
- if (i.imms[0]->X_seg != absolute_section)
- as_bad ("can't handle non absolute segment in long call/jmp");
- md_number_to_chars (p + 5, i.imms[0]->X_add_number, 2);
+ fix_new_exp (frag_now, p - frag_now->fr_literal, size,
+ i.op[1].imms, 0, reloc (size, 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);
+ md_number_to_chars (p + size, (valueT) i.op[0].imms->X_add_number, 2);
}
else
{
- /* Output normal instructions here. */
+ /* Output normal instructions here. */
unsigned char *q;
- /* First the prefix bytes. */
- for (q = i.prefix; q < i.prefix + i.prefixes; q++)
+ /* The prefix bytes. */
+ for (q = i.prefix;
+ q < i.prefix + sizeof (i.prefix) / sizeof (i.prefix[0]);
+ q++)
{
- p = frag_more (1);
- md_number_to_chars (p, (unsigned int) *q, 1);
+ if (*q)
+ {
+ insn_size += 1;
+ p = frag_more (1);
+ md_number_to_chars (p, (valueT) *q, 1);
+ }
}
- /* Now the opcode; be careful about word order here! */
- if (fits_in_unsigned_byte (t->base_opcode))
+ /* Now the opcode; be careful about word order here! */
+ if (fits_in_unsigned_byte (i.tm.base_opcode))
{
- FRAG_APPEND_1_CHAR (t->base_opcode);
+ insn_size += 1;
+ FRAG_APPEND_1_CHAR (i.tm.base_opcode);
}
- else if (fits_in_unsigned_word (t->base_opcode))
+ else if (fits_in_unsigned_word (i.tm.base_opcode))
{
+ insn_size += 2;
p = frag_more (2);
- /* put out high byte first: can't use md_number_to_chars! */
- *p++ = (t->base_opcode >> 8) & 0xff;
- *p = t->base_opcode & 0xff;
+ /* Put out high byte first: can't use md_number_to_chars! */
+ *p++ = (i.tm.base_opcode >> 8) & 0xff;
+ *p = i.tm.base_opcode & 0xff;
}
else
- { /* opcode is either 3 or 4 bytes */
- if (t->base_opcode & 0xff000000)
+ { /* Opcode is either 3 or 4 bytes. */
+ if (i.tm.base_opcode & 0xff000000)
{
+ insn_size += 4;
p = frag_more (4);
- *p++ = (t->base_opcode >> 24) & 0xff;
+ *p++ = (i.tm.base_opcode >> 24) & 0xff;
}
else
- p = frag_more (3);
- *p++ = (t->base_opcode >> 16) & 0xff;
- *p++ = (t->base_opcode >> 8) & 0xff;
- *p = (t->base_opcode) & 0xff;
+ {
+ insn_size += 3;
+ p = frag_more (3);
+ }
+ *p++ = (i.tm.base_opcode >> 16) & 0xff;
+ *p++ = (i.tm.base_opcode >> 8) & 0xff;
+ *p = (i.tm.base_opcode) & 0xff;
}
- /* Now the modrm byte and base index byte (if present). */
- if (t->opcode_modifier & Modrm)
+ /* Now the modrm byte and sib byte (if present). */
+ if (i.tm.opcode_modifier & Modrm)
{
+ insn_size += 1;
p = frag_more (1);
- /* md_number_to_chars (p, i.rm, 1); */
- md_number_to_chars (p, (i.rm.regmem << 0 | i.rm.reg << 3 | i.rm.mode << 6), 1);
- /* If i.rm.regmem == ESP (4) && i.rm.mode != Mode 3 (Register mode)
- ==> need second modrm byte. */
- if (i.rm.regmem == ESCAPE_TO_TWO_BYTE_ADDRESSING && i.rm.mode != 3)
+ md_number_to_chars (p,
+ (valueT) (i.rm.regmem << 0
+ | i.rm.reg << 3
+ | i.rm.mode << 6),
+ 1);
+ /* If i.rm.regmem == ESP (4)
+ && i.rm.mode != (Register mode)
+ && not 16 bit
+ ==> need second modrm byte. */
+ if (i.rm.regmem == ESCAPE_TO_TWO_BYTE_ADDRESSING
+ && i.rm.mode != 3
+ && !(i.base_reg && (i.base_reg->reg_type & Reg16) != 0))
{
+ insn_size += 1;
p = frag_more (1);
- /* md_number_to_chars (p, i.bi, 1); */
- md_number_to_chars (p, (i.bi.base << 0 | i.bi.index << 3 | i.bi.scale << 6), 1);
+ md_number_to_chars (p,
+ (valueT) (i.sib.base << 0
+ | i.sib.index << 3
+ | i.sib.scale << 6),
+ 1);
}
}
for (n = 0; n < i.operands; n++)
{
- if (i.disps[n])
+ if (i.types[n] & Disp)
{
- if (i.disps[n]->X_seg == absolute_section)
+ if (i.op[n].disps->X_op == O_constant)
{
- if (i.types[n] & (Disp8 | Abs8))
- {
- p = frag_more (1);
- md_number_to_chars (p, i.disps[n]->X_add_number, 1);
- }
- else if (i.types[n] & (Disp16 | Abs16))
+ int size;
+ offsetT val;
+
+ size = 4;
+ if (i.types[n] & (Disp8 | Disp16))
{
- p = frag_more (2);
- md_number_to_chars (p, i.disps[n]->X_add_number, 2);
- }
- else
- { /* Disp32|Abs32 */
- p = frag_more (4);
- md_number_to_chars (p, i.disps[n]->X_add_number, 4);
+ size = 2;
+ if (i.types[n] & Disp8)
+ size = 1;
}
+ val = offset_in_range (i.op[n].disps->X_add_number,
+ size);
+ insn_size += size;
+ p = frag_more (size);
+ md_number_to_chars (p, val, size);
}
else
- { /* not absolute_section */
- /* need a 32-bit fixup (don't support 8bit non-absolute disps) */
- p = frag_more (4);
- fix_new (frag_now, p - frag_now->fr_literal, 4,
- i.disps[n]->X_add_symbol, i.disps[n]->X_subtract_symbol,
- i.disps[n]->X_add_number, 0, NO_RELOC);
+ {
+ int size = 4;
+
+ if (i.types[n] & Disp16)
+ size = 2;
+
+ 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]));
}
}
}
- } /* end displacement output */
+ }
- /* output immediate */
+ /* Output immediate. */
if (i.imm_operands)
{
register unsigned int n;
for (n = 0; n < i.operands; n++)
{
- if (i.imms[n])
+ if (i.types[n] & Imm)
{
- if (i.imms[n]->X_seg == absolute_section)
+ if (i.op[n].imms->X_op == O_constant)
{
- if (i.types[n] & (Imm8 | Imm8S))
- {
- p = frag_more (1);
- md_number_to_chars (p, i.imms[n]->X_add_number, 1);
- }
- else if (i.types[n] & Imm16)
- {
- p = frag_more (2);
- md_number_to_chars (p, i.imms[n]->X_add_number, 2);
- }
- else
+ int size;
+ offsetT val;
+
+ size = 4;
+ if (i.types[n] & (Imm8 | Imm8S | Imm16))
{
- p = frag_more (4);
- md_number_to_chars (p, i.imms[n]->X_add_number, 4);
+ size = 2;
+ if (i.types[n] & (Imm8 | Imm8S))
+ size = 1;
}
+ val = offset_in_range (i.op[n].imms->X_add_number,
+ size);
+ insn_size += size;
+ p = frag_more (size);
+ md_number_to_chars (p, val, size);
}
else
- { /* not absolute_section */
- /* need a 32-bit fixup (don't support 8bit non-absolute ims) */
- /* try to support other sizes ... */
- int size;
- if (i.types[n] & (Imm8 | Imm8S))
- size = 1;
- else if (i.types[n] & Imm16)
+ {
+ /* Not absolute_section.
+ Need a 32-bit fixup (don't support 8bit
+ non-absolute imms). Try to support other
+ sizes ... */
+#ifdef BFD_ASSEMBLER
+ enum bfd_reloc_code_real reloc_type;
+#else
+ int reloc_type;
+#endif
+ int size = 4;
+
+ if (i.types[n] & Imm16)
size = 2;
- else
- size = 4;
+ else if (i.types[n] & (Imm8 | Imm8S))
+ size = 1;
+
+ insn_size += size;
p = frag_more (size);
- fix_new (frag_now, p - frag_now->fr_literal, size,
- i.imms[n]->X_add_symbol, i.imms[n]->X_subtract_symbol,
- i.imms[n]->X_add_number, 0, NO_RELOC);
+ reloc_type = reloc (size, 0, i.disp_reloc[0]);
+#ifdef BFD_ASSEMBLER
+ if (reloc_type == BFD_RELOC_32
+ && GOT_symbol
+ && GOT_symbol == i.op[n].imms->X_add_symbol
+ && (i.op[n].imms->X_op == O_symbol
+ || (i.op[n].imms->X_op == O_add
+ && ((symbol_get_value_expression
+ (i.op[n].imms->X_op_symbol)->X_op)
+ == O_subtract))))
+ {
+ reloc_type = BFD_RELOC_386_GOTPC;
+ i.op[n].imms->X_add_number += 3;
+ }
+#endif
+ fix_new_exp (frag_now, p - frag_now->fr_literal, size,
+ i.op[n].imms, 0, reloc_type);
}
}
}
- } /* end immediate output */
+ }
}
+ dwarf2_emit_insn (insn_size);
+
#ifdef DEBUG386
- if (flagseen['D'])
+ if (flag_debug)
{
pi (line, &i);
}
-#endif /* DEBUG386 */
-
+#endif /* DEBUG386 */
}
- return;
}
\f
-/* Parse OPERAND_STRING into the i386_insn structure I. Returns non-zero
- on error. */
+static int i386_immediate PARAMS ((char *));
static int
-i386_operand (operand_string)
- char *operand_string;
+i386_immediate (imm_start)
+ char *imm_start;
{
- register char *op_string = operand_string;
+ char *save_input_line_pointer;
+ segT exp_seg = 0;
+ expressionS *exp;
- /* Address of '\0' at end of operand_string. */
- char *end_of_operand_string = operand_string + strlen (operand_string);
+ if (i.imm_operands == MAX_IMMEDIATE_OPERANDS)
+ {
+ as_bad (_("only 1 or 2 immediate operands are allowed"));
+ return 0;
+ }
- /* Start and end of displacement string expression (if found). */
- char *displacement_string_start = NULL;
- char *displacement_string_end = NULL;
+ exp = &im_expressions[i.imm_operands++];
+ i.op[this_operand].imms = exp;
- /* We check for an absolute prefix (differentiating,
- for example, 'jmp pc_relative_label' from 'jmp *absolute_label'. */
- if (*op_string == ABSOLUTE_PREFIX)
+ if (is_space_char (*imm_start))
+ ++imm_start;
+
+ save_input_line_pointer = input_line_pointer;
+ input_line_pointer = imm_start;
+
+#ifndef LEX_AT
+ {
+ /* We can have operands of the form
+ <symbol>@GOTOFF+<nnn>
+ Take the easy way out here and copy everything
+ into a temporary buffer... */
+ register char *cp;
+
+ cp = strchr (input_line_pointer, '@');
+ if (cp != NULL)
+ {
+ char *tmpbuf;
+ int len = 0;
+ int first;
+
+ /* GOT relocations are not supported in 16 bit mode. */
+ if (flag_16bit_code)
+ as_bad (_("GOT relocations not supported in 16 bit mode"));
+
+ if (GOT_symbol == NULL)
+ GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
+
+ if (strncmp (cp + 1, "PLT", 3) == 0)
+ {
+ i.disp_reloc[this_operand] = BFD_RELOC_386_PLT32;
+ len = 3;
+ }
+ else if (strncmp (cp + 1, "GOTOFF", 6) == 0)
+ {
+ 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;
+ len = 3;
+ }
+ else
+ as_bad (_("bad reloc specifier in expression"));
+
+ /* Replace the relocation token with ' ', so that errors like
+ foo@GOTOFF1 will be detected. */
+ first = cp - input_line_pointer;
+ tmpbuf = (char *) alloca (strlen (input_line_pointer));
+ memcpy (tmpbuf, input_line_pointer, first);
+ tmpbuf[first] = ' ';
+ strcpy (tmpbuf + first + 1, cp + 1 + len);
+ input_line_pointer = tmpbuf;
+ }
+ }
+#endif
+
+ exp_seg = expression (exp);
+
+ SKIP_WHITESPACE ();
+ if (*input_line_pointer)
+ as_bad (_("ignoring junk `%s' after expression"), input_line_pointer);
+
+ input_line_pointer = save_input_line_pointer;
+
+ if (exp->X_op == O_absent || exp->X_op == O_big)
{
- op_string++;
- i.types[this_operand] |= JumpAbsolute;
+ /* Missing or bad expr becomes absolute 0. */
+ as_bad (_("missing or invalid immediate expression `%s' taken as 0"),
+ imm_start);
+ exp->X_op = O_constant;
+ exp->X_add_number = 0;
+ exp->X_add_symbol = (symbolS *) 0;
+ exp->X_op_symbol = (symbolS *) 0;
}
- /* Check if operand is a register. */
- if (*op_string == REGISTER_PREFIX)
+ if (exp->X_op == O_constant)
{
- register reg_entry *r;
- if (!(r = parse_register (op_string)))
- {
- as_bad ("bad register name ('%s')", op_string);
- return 0;
- }
- /* Check for segment override, rather than segment register by
- searching for ':' after %<x>s where <x> = s, c, d, e, f, g. */
- if ((r->reg_type & (SReg2 | SReg3)) && op_string[3] == ':')
- {
- switch (r->reg_num)
- {
- case 0:
- i.seg = (seg_entry *) & es;
- break;
- case 1:
- i.seg = (seg_entry *) & cs;
- break;
- case 2:
- i.seg = (seg_entry *) & ss;
- break;
- case 3:
- i.seg = (seg_entry *) & ds;
- break;
- case 4:
- i.seg = (seg_entry *) & fs;
- break;
- case 5:
- i.seg = (seg_entry *) & gs;
- break;
- }
- op_string += 4; /* skip % <x> s : */
- operand_string = op_string; /* Pretend given string starts here. */
- if (!is_digit_char (*op_string) && !is_identifier_char (*op_string)
- && *op_string != '(' && *op_string != ABSOLUTE_PREFIX)
- {
- as_bad ("bad memory operand after segment override");
- return 0;
- }
- /* Handle case of %es:*foo. */
- if (*op_string == ABSOLUTE_PREFIX)
- {
- op_string++;
- i.types[this_operand] |= JumpAbsolute;
- }
- goto do_memory_reference;
- }
- i.types[this_operand] |= r->reg_type;
- i.regs[this_operand] = r;
- i.reg_operands++;
+ /* Size it properly later. */
+ i.types[this_operand] |= Imm32;
}
- else if (*op_string == IMMEDIATE_PREFIX)
- { /* ... or an immediate */
- char *save_input_line_pointer;
- segT exp_seg = 0;
- expressionS *exp;
-
- if (i.imm_operands == MAX_IMMEDIATE_OPERANDS)
- {
- as_bad ("only 1 or 2 immediate operands are allowed");
- return 0;
- }
-
- exp = &im_expressions[i.imm_operands++];
- i.imms[this_operand] = exp;
- save_input_line_pointer = input_line_pointer;
- input_line_pointer = ++op_string; /* must advance op_string! */
- exp_seg = expression (exp);
- input_line_pointer = save_input_line_pointer;
-
- if (exp_seg == absent_section)
- {
- /* missing or bad expr becomes absolute 0 */
- as_bad ("missing or invalid immediate expression '%s' taken as 0",
- operand_string);
- exp->X_seg = absolute_section;
- exp->X_add_number = 0;
- exp->X_add_symbol = (symbolS *) 0;
- exp->X_subtract_symbol = (symbolS *) 0;
- i.types[this_operand] |= Imm;
- }
- else if (exp_seg == absolute_section)
- {
- i.types[this_operand] |= smallest_imm_type (exp->X_add_number);
- }
-#ifndef I386COFF
- else if (exp_seg != text_section
- && exp_seg != data_section
- && exp_seg != bss_section
- && exp_seg != undefined_section
+#if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
+ else if (1
#ifdef BFD_ASSEMBLER
- && ! bfd_is_com_section (exp_seg)
+ && OUTPUT_FLAVOR == bfd_target_aout_flavour
#endif
- )
- {
- seg_unimplemented:
- as_bad ("Unimplemented segment type %d in parse_operand", exp_seg);
- return 0;
- }
+ && exp_seg != text_section
+ && exp_seg != data_section
+ && exp_seg != bss_section
+ && exp_seg != undefined_section
+#ifdef BFD_ASSEMBLER
+ && !bfd_is_com_section (exp_seg)
#endif
- else
- {
- /* this is an address ==> 32bit */
- i.types[this_operand] |= Imm32;
- }
- /* shorten this type of this operand if the instruction wants
- * fewer bits than are present in the immediate. The bit field
- * code can put out 'andb $0xffffff, %al', for example. pace
- * also 'movw $foo,(%eax)'
- */
- switch (i.suffix)
- {
- case WORD_OPCODE_SUFFIX:
- i.types[this_operand] |= Imm16;
- break;
- case BYTE_OPCODE_SUFFIX:
- i.types[this_operand] |= Imm16 | Imm8 | Imm8S;
- break;
- }
+ )
+ {
+#ifdef BFD_ASSEMBLER
+ as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
+#else
+ as_bad (_("unimplemented segment type %d in operand"), exp_seg);
+#endif
+ return 0;
}
- else if (is_digit_char (*op_string) || is_identifier_char (*op_string)
- || *op_string == '(')
+#endif
+ else
{
- /* This is a memory reference of some sort. */
- register char *base_string;
- unsigned int found_base_index_form;
+ /* 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);
+ }
- do_memory_reference:
- if (i.mem_operands == MAX_MEMORY_OPERANDS)
- {
- as_bad ("more than 1 memory reference in instruction");
- return 0;
- }
- i.mem_operands++;
+ return 1;
+}
- /* Determine type of memory operand from opcode_suffix;
- no opcode suffix implies general memory references. */
- switch (i.suffix)
- {
- case BYTE_OPCODE_SUFFIX:
- i.types[this_operand] |= Mem8;
- break;
- case WORD_OPCODE_SUFFIX:
- i.types[this_operand] |= Mem16;
- break;
- case DWORD_OPCODE_SUFFIX:
- default:
- i.types[this_operand] |= Mem32;
- }
+static int i386_scale PARAMS ((char *));
- /* Check for base index form. We detect the base index form by
- looking for an ')' at the end of the operand, searching
- for the '(' matching it, and finding a REGISTER_PREFIX or ','
- after it. */
- base_string = end_of_operand_string - 1;
- found_base_index_form = 0;
- if (*base_string == ')')
- {
- unsigned int parens_balenced = 1;
- /* We've already checked that the number of left & right ()'s are
- equal, so this loop will not be infinite. */
- do
- {
- base_string--;
- if (*base_string == ')')
- parens_balenced++;
- if (*base_string == '(')
- parens_balenced--;
- }
- while (parens_balenced);
- base_string++; /* Skip past '('. */
- if (*base_string == REGISTER_PREFIX || *base_string == ',')
- found_base_index_form = 1;
- }
+static int
+i386_scale (scale)
+ char *scale;
+{
+ if (!isdigit (*scale))
+ goto bad_scale;
- /* If we can't parse a base index register expression, we've found
- a pure displacement expression. We set up displacement_string_start
- and displacement_string_end for the code below. */
- if (!found_base_index_form)
- {
- displacement_string_start = op_string;
- displacement_string_end = end_of_operand_string;
- }
- else
- {
- char *base_reg_name, *index_reg_name, *num_string;
- int num;
+ switch (*scale)
+ {
+ case '0':
+ case '1':
+ i.log2_scale_factor = 0;
+ break;
+ case '2':
+ i.log2_scale_factor = 1;
+ break;
+ case '4':
+ i.log2_scale_factor = 2;
+ break;
+ case '8':
+ i.log2_scale_factor = 3;
+ break;
+ default:
+ bad_scale:
+ as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
+ scale);
+ return 0;
+ }
+ if (i.log2_scale_factor != 0 && ! i.index_reg)
+ {
+ as_warn (_("scale factor of %d without an index register"),
+ 1 << i.log2_scale_factor);
+#if SCALE1_WHEN_NO_INDEX
+ i.log2_scale_factor = 0;
+#endif
+ }
+ return 1;
+}
- i.types[this_operand] |= BaseIndex;
+static int i386_displacement PARAMS ((char *, char *));
- /* If there is a displacement set-up for it to be parsed later. */
- if (base_string != op_string + 1)
- {
- displacement_string_start = op_string;
- displacement_string_end = base_string - 1;
- }
+static int
+i386_displacement (disp_start, disp_end)
+ char *disp_start;
+ char *disp_end;
+{
+ register expressionS *exp;
+ segT exp_seg = 0;
+ char *save_input_line_pointer;
+ int bigdisp = Disp32;
+
+ if (flag_16bit_code ^ (i.prefix[ADDR_PREFIX] != 0))
+ bigdisp = Disp16;
+ i.types[this_operand] |= bigdisp;
+
+ exp = &disp_expressions[i.disp_operands];
+ i.op[this_operand].disps = exp;
+ i.disp_operands++;
+ save_input_line_pointer = input_line_pointer;
+ input_line_pointer = disp_start;
+ END_STRING_AND_SAVE (disp_end);
+
+#ifndef GCC_ASM_O_HACK
+#define GCC_ASM_O_HACK 0
+#endif
+#if GCC_ASM_O_HACK
+ END_STRING_AND_SAVE (disp_end + 1);
+ if ((i.types[this_operand] & BaseIndex) != 0
+ && displacement_string_end[-1] == '+')
+ {
+ /* This hack is to avoid a warning when using the "o"
+ constraint within gcc asm statements.
+ For instance:
+
+ #define _set_tssldt_desc(n,addr,limit,type) \
+ __asm__ __volatile__ ( \
+ "movw %w2,%0\n\t" \
+ "movw %w1,2+%0\n\t" \
+ "rorl $16,%1\n\t" \
+ "movb %b1,4+%0\n\t" \
+ "movb %4,5+%0\n\t" \
+ "movb $0,6+%0\n\t" \
+ "movb %h1,7+%0\n\t" \
+ "rorl $16,%1" \
+ : "=o"(*(n)) : "q" (addr), "ri"(limit), "i"(type))
+
+ This works great except that the output assembler ends
+ up looking a bit weird if it turns out that there is
+ no offset. You end up producing code that looks like:
+
+ #APP
+ movw $235,(%eax)
+ movw %dx,2+(%eax)
+ rorl $16,%edx
+ movb %dl,4+(%eax)
+ movb $137,5+(%eax)
+ movb $0,6+(%eax)
+ movb %dh,7+(%eax)
+ rorl $16,%edx
+ #NO_APP
+
+ So here we provide the missing zero. */
+
+ *displacement_string_end = '0';
+ }
+#endif
+#ifndef LEX_AT
+ {
+ /* We can have operands of the form
+ <symbol>@GOTOFF+<nnn>
+ Take the easy way out here and copy everything
+ into a temporary buffer... */
+ register char *cp;
+
+ cp = strchr (input_line_pointer, '@');
+ if (cp != NULL)
+ {
+ char *tmpbuf;
+ int len = 0;
+ int first;
+
+ /* GOT relocations are not supported in 16 bit mode. */
+ if (flag_16bit_code)
+ as_bad (_("GOT relocations not supported in 16 bit mode"));
+
+ if (GOT_symbol == NULL)
+ GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
+
+ if (strncmp (cp + 1, "PLT", 3) == 0)
+ {
+ i.disp_reloc[this_operand] = BFD_RELOC_386_PLT32;
+ len = 3;
+ }
+ else if (strncmp (cp + 1, "GOTOFF", 6) == 0)
+ {
+ 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;
+ len = 3;
+ }
+ else
+ as_bad (_("bad reloc specifier in expression"));
+
+ /* Replace the relocation token with ' ', so that errors like
+ foo@GOTOFF1 will be detected. */
+ first = cp - input_line_pointer;
+ tmpbuf = (char *) alloca (strlen (input_line_pointer));
+ memcpy (tmpbuf, input_line_pointer, first);
+ tmpbuf[first] = ' ';
+ strcpy (tmpbuf + first + 1, cp + 1 + len);
+ input_line_pointer = tmpbuf;
+ }
+ }
+#endif
+
+ exp_seg = expression (exp);
+
+#ifdef BFD_ASSEMBLER
+ /* 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 (S_IS_LOCAL(exp->X_add_symbol)
+ && S_GET_SEGMENT (exp->X_add_symbol) != undefined_section)
+ section_symbol (S_GET_SEGMENT (exp->X_add_symbol));
+ assert (exp->X_op == O_symbol);
+ exp->X_op = O_subtract;
+ exp->X_op_symbol = GOT_symbol;
+ i.disp_reloc[this_operand] = BFD_RELOC_32;
+ }
+#endif
+
+ SKIP_WHITESPACE ();
+ if (*input_line_pointer)
+ as_bad (_("ignoring junk `%s' after expression"),
+ input_line_pointer);
+#if GCC_ASM_O_HACK
+ RESTORE_END_STRING (disp_end + 1);
+#endif
+ RESTORE_END_STRING (disp_end);
+ input_line_pointer = save_input_line_pointer;
+
+ if (exp->X_op == O_absent || exp->X_op == O_big)
+ {
+ /* Missing or bad expr becomes absolute 0. */
+ as_bad (_("missing or invalid displacement expression `%s' taken as 0"),
+ disp_start);
+ exp->X_op = O_constant;
+ exp->X_add_number = 0;
+ exp->X_add_symbol = (symbolS *) 0;
+ exp->X_op_symbol = (symbolS *) 0;
+ }
+
+#if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
+ if (exp->X_op != O_constant
+#ifdef BFD_ASSEMBLER
+ && OUTPUT_FLAVOR == bfd_target_aout_flavour
+#endif
+ && exp_seg != text_section
+ && exp_seg != data_section
+ && exp_seg != bss_section
+ && exp_seg != undefined_section)
+ {
+#ifdef BFD_ASSEMBLER
+ as_bad (_("unimplemented segment %s in operand"), exp_seg->name);
+#else
+ as_bad (_("unimplemented segment type %d in operand"), exp_seg);
+#endif
+ return 0;
+ }
+#endif
+ 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;
+}
+
+/* Parse OPERAND_STRING into the i386_insn structure I. Returns non-zero
+ on error. */
+
+static int
+i386_operand (operand_string)
+ char *operand_string;
+{
+ const reg_entry *r;
+ char *end_op;
+ char *op_string = operand_string;
+
+ if (is_space_char (*op_string))
+ ++op_string;
+
+ /* We check for an absolute prefix (differentiating,
+ for example, 'jmp pc_relative_label' from 'jmp *absolute_label'. */
+ if (*op_string == ABSOLUTE_PREFIX)
+ {
+ ++op_string;
+ if (is_space_char (*op_string))
+ ++op_string;
+ i.types[this_operand] |= JumpAbsolute;
+ }
- /* Find base register (if any). */
- if (*base_string != ',')
+ /* Check if operand is a register. */
+ 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)
{
- base_reg_name = base_string++;
- /* skip past register name & parse it */
- while (isalpha (*base_string))
- base_string++;
- if (base_string == base_reg_name + 1)
- {
- as_bad ("can't find base register name after '(%c'",
- REGISTER_PREFIX);
- return 0;
- }
- END_STRING_AND_SAVE (base_string);
- if (!(i.base_reg = parse_register (base_reg_name)))
- {
- as_bad ("bad base register name ('%s')", base_reg_name);
- return 0;
- }
- RESTORE_END_STRING (base_string);
+ 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;
}
- /* Now check seperator; must be ',' ==> index reg
- OR num ==> no index reg. just scale factor
- OR ')' ==> end. (scale factor = 1) */
- if (*base_string != ',' && *base_string != ')')
+ /* Skip the ':' and whitespace. */
+ ++op_string;
+ if (is_space_char (*op_string))
+ ++op_string;
+
+ if (!is_digit_char (*op_string)
+ && !is_identifier_char (*op_string)
+ && *op_string != '('
+ && *op_string != ABSOLUTE_PREFIX)
{
- as_bad ("expecting ',' or ')' after base register in `%s'",
- operand_string);
+ as_bad (_("bad memory operand `%s'"), op_string);
return 0;
}
+ /* Handle case of %es:*foo. */
+ if (*op_string == ABSOLUTE_PREFIX)
+ {
+ ++op_string;
+ if (is_space_char (*op_string))
+ ++op_string;
+ i.types[this_operand] |= JumpAbsolute;
+ }
+ goto do_memory_reference;
+ }
+ if (*op_string)
+ {
+ as_bad (_("junk `%s' after register"), op_string);
+ return 0;
+ }
+ 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 (*op_string == IMMEDIATE_PREFIX)
+ {
+ ++op_string;
+ if (i.types[this_operand] & JumpAbsolute)
+ {
+ as_bad (_("immediate operand illegal with absolute jump"));
+ return 0;
+ }
+ if (!i386_immediate (op_string))
+ return 0;
+ }
+ else if (is_digit_char (*op_string)
+ || is_identifier_char (*op_string)
+ || *op_string == '(' )
+ {
+ /* This is a memory reference of some sort. */
+ char *base_string;
+
+ /* Start and end of displacement string expression (if found). */
+ char *displacement_string_start;
+ char *displacement_string_end;
+
+ do_memory_reference:
+ 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;
+ }
+
+ /* Check for base index form. We detect the base index form by
+ looking for an ')' at the end of the operand, searching
+ for the '(' matching it, and finding a REGISTER_PREFIX or ','
+ after the '('. */
+ base_string = op_string + strlen (op_string);
- /* There may index reg here; and there may be a scale factor. */
- if (*base_string == ',' && *(base_string + 1) == REGISTER_PREFIX)
+ --base_string;
+ if (is_space_char (*base_string))
+ --base_string;
+
+ /* If we only have a displacement, set-up for it to be parsed later. */
+ displacement_string_start = op_string;
+ displacement_string_end = base_string + 1;
+
+ if (*base_string == ')')
+ {
+ char *temp_string;
+ unsigned int parens_balanced = 1;
+ /* We've already checked that the number of left & right ()'s are
+ equal, so this loop will not be infinite. */
+ do
{
- index_reg_name = ++base_string;
- while (isalpha (*++base_string));
- END_STRING_AND_SAVE (base_string);
- if (!(i.index_reg = parse_register (index_reg_name)))
- {
- as_bad ("bad index register name ('%s')", index_reg_name);
- return 0;
- }
- RESTORE_END_STRING (base_string);
+ base_string--;
+ if (*base_string == ')')
+ parens_balanced++;
+ if (*base_string == '(')
+ parens_balanced--;
}
+ while (parens_balanced);
- /* Check for scale factor. */
- if (*base_string == ',' && isdigit (*(base_string + 1)))
+ temp_string = base_string;
+
+ /* Skip past '(' and whitespace. */
+ ++base_string;
+ if (is_space_char (*base_string))
+ ++base_string;
+
+ if (*base_string == ','
+ || ((*base_string == REGISTER_PREFIX || allow_naked_reg)
+ && (i.base_reg = parse_register (base_string, &end_op)) != NULL))
{
- num_string = ++base_string;
- while (is_digit_char (*base_string))
- base_string++;
- if (base_string == num_string)
+ displacement_string_end = temp_string;
+
+ i.types[this_operand] |= BaseIndex;
+
+ if (i.base_reg)
{
- as_bad ("can't find a scale factor after ','");
- return 0;
+ base_string = end_op;
+ if (is_space_char (*base_string))
+ ++base_string;
}
- END_STRING_AND_SAVE (base_string);
- /* We've got a scale factor. */
- if (!sscanf (num_string, "%d", &num))
+
+ /* There may be an index reg or scale factor here. */
+ if (*base_string == ',')
{
- as_bad ("can't parse scale factor from '%s'", num_string);
- return 0;
- }
- RESTORE_END_STRING (base_string);
- switch (num)
- { /* must be 1 digit scale */
- case 1:
- i.log2_scale_factor = 0;
- break;
- case 2:
- i.log2_scale_factor = 1;
- break;
- case 4:
- i.log2_scale_factor = 2;
- break;
- case 8:
- i.log2_scale_factor = 3;
- break;
- default:
- as_bad ("expecting scale factor of 1, 2, 4, 8; got %d", num);
- return 0;
+ ++base_string;
+ if (is_space_char (*base_string))
+ ++base_string;
+
+ if ((*base_string == REGISTER_PREFIX || allow_naked_reg)
+ && (i.index_reg = parse_register (base_string, &end_op)) != NULL)
+ {
+ base_string = end_op;
+ if (is_space_char (*base_string))
+ ++base_string;
+ if (*base_string == ',')
+ {
+ ++base_string;
+ if (is_space_char (*base_string))
+ ++base_string;
+ }
+ else if (*base_string != ')' )
+ {
+ as_bad (_("expecting `,' or `)' after index register in `%s'"),
+ operand_string);
+ return 0;
+ }
+ }
+ else if (*base_string == REGISTER_PREFIX)
+ {
+ as_bad (_("bad register name `%s'"), base_string);
+ return 0;
+ }
+
+ /* Check for scale factor. */
+ if (isdigit ((unsigned char) *base_string))
+ {
+ if (!i386_scale (base_string))
+ return 0;
+
+ ++base_string;
+ if (is_space_char (*base_string))
+ ++base_string;
+ if (*base_string != ')')
+ {
+ as_bad (_("expecting `)' after scale factor in `%s'"),
+ operand_string);
+ return 0;
+ }
+ }
+ else if (!i.index_reg)
+ {
+ as_bad (_("expecting index register or scale factor after `,'; got '%c'"),
+ *base_string);
+ return 0;
+ }
}
- }
- else
- {
- if (!i.index_reg && *base_string == ',')
+ else if (*base_string != ')')
{
- as_bad ("expecting index register or scale factor after ','; got '%c'",
- *(base_string + 1));
+ as_bad (_("expecting `,' or `)' after base register in `%s'"),
+ operand_string);
return 0;
}
}
- }
-
- /* If there's an expression begining the operand, parse it,
- assuming displacement_string_start and displacement_string_end
- are meaningful. */
- if (displacement_string_start)
- {
- register expressionS *exp;
- segT exp_seg = 0;
- char *save_input_line_pointer;
- exp = &disp_expressions[i.disp_operands];
- i.disps[this_operand] = exp;
- i.disp_operands++;
- save_input_line_pointer = input_line_pointer;
- input_line_pointer = displacement_string_start;
- END_STRING_AND_SAVE (displacement_string_end);
- exp_seg = expression (exp);
- if (*input_line_pointer)
- as_bad ("Ignoring junk '%s' after expression", input_line_pointer);
- RESTORE_END_STRING (displacement_string_end);
- input_line_pointer = save_input_line_pointer;
- if (exp_seg == absent_section)
- {
- /* missing expr becomes absolute 0 */
- as_bad ("missing or invalid displacement '%s' taken as 0",
- operand_string);
- i.types[this_operand] |= (Disp | Abs);
- exp->X_seg = absolute_section;
- exp->X_add_number = 0;
- exp->X_add_symbol = (symbolS *) 0;
- exp->X_subtract_symbol = (symbolS *) 0;
- }
- else if (exp_seg == absolute_section)
- {
- i.types[this_operand] |= SMALLEST_DISP_TYPE (exp->X_add_number);
- }
- else if (exp_seg == text_section
- || exp_seg == data_section
- || exp_seg == bss_section
- || exp_seg == undefined_section)
- {
- i.types[this_operand] |= Disp32;
- }
- else
+ else if (*base_string == REGISTER_PREFIX)
{
-#ifdef I386COFF
- i.types[this_operand] |= Disp32;
-#else
- goto seg_unimplemented;
-#endif
+ as_bad (_("bad register name `%s'"), base_string);
+ return 0;
}
}
- /* Make sure the memory operand we've been dealt is valid. */
- if (i.base_reg && i.index_reg &&
- !(i.base_reg->reg_type & i.index_reg->reg_type & Reg))
+ /* If there's an expression beginning the operand, parse it,
+ assuming displacement_string_start and
+ displacement_string_end are meaningful. */
+ if (displacement_string_start != displacement_string_end)
{
- as_bad ("register size mismatch in (base,index,scale) expression");
- return 0;
+ if (!i386_displacement (displacement_string_start,
+ displacement_string_end))
+ return 0;
}
- /*
- * special case for (%dx) while doing input/output op
- */
- if ((i.base_reg &&
- (i.base_reg->reg_type == (Reg16 | InOutPortReg)) &&
- (i.index_reg == 0)))
+
+ /* Special case for (%dx) while doing input/output op. */
+ if (i.base_reg
+ && i.base_reg->reg_type == (Reg16 | InOutPortReg)
+ && i.index_reg == 0
+ && i.log2_scale_factor == 0
+ && i.seg[i.mem_operands] == 0
+ && (i.types[this_operand] & Disp) == 0)
{
- i.types[this_operand] |= InOutPortReg;
+ i.types[this_operand] = InOutPortReg;
return 1;
}
- if ((i.base_reg && (i.base_reg->reg_type & Reg32) == 0) ||
- (i.index_reg && (i.index_reg->reg_type & Reg32) == 0))
- {
- as_bad ("base/index register must be 32 bit register");
- return 0;
- }
- if (i.index_reg && i.index_reg == esp)
- {
- as_bad ("%s may not be used as an index register", esp->reg_name);
- return 0;
- }
+
+ if (i386_index_check (operand_string) == 0)
+ return 0;
+ i.mem_operands++;
}
else
- { /* it's not a memory operand; argh! */
- as_bad ("invalid char %s begining %s operand '%s'",
- output_invalid (*op_string), ordinal_names[this_operand],
+ {
+ /* It's not a memory operand; argh! */
+ as_bad (_("invalid char %s beginning operand %d `%s'"),
+ output_invalid (*op_string),
+ this_operand + 1,
op_string);
return 0;
}
- return 1; /* normal return */
+ return 1; /* Normal return. */
}
\f
-/*
- * md_estimate_size_before_relax()
- *
- * Called just before relax().
- * Any symbol that is now undefined will not become defined.
- * Return the correct fr_subtype in the frag.
- * Return the initial "guess for fr_var" to caller.
- * The guess for fr_var is ACTUALLY the growth beyond fr_fix.
- * Whatever we do to grow fr_fix or fr_var contributes to our returned value.
- * Although it may not be explicit in the frag, pretend fr_var starts with a
- * 0 value.
- */
+/* md_estimate_size_before_relax()
+
+ Called just before relax() for rs_machine_dependent frags. The x86
+ assembler uses these frags to handle variable size jump
+ instructions.
+
+ Any symbol that is now undefined will not become defined.
+ Return the correct fr_subtype in the frag.
+ Return the initial "guess for variable size of frag" to caller.
+ The guess is actually the growth beyond the fixed part. Whatever
+ we do to grow the fixed or variable part contributes to our
+ returned value. */
+
int
md_estimate_size_before_relax (fragP, segment)
register fragS *fragP;
register segT segment;
{
- register unsigned char *opcode;
- register int old_fr_fix;
-
- old_fr_fix = fragP->fr_fix;
- opcode = (unsigned char *) fragP->fr_opcode;
- /* We've already got fragP->fr_subtype right; all we have to do is check
- for un-relaxable symbols. */
- if (S_GET_SEGMENT (fragP->fr_symbol) != segment)
+ /* We've already got fragP->fr_subtype right; all we have to do is
+ check for un-relaxable symbols. On an ELF system, we can't relax
+ an externally visible symbol, because it may be overridden by a
+ shared library. */
+ if (S_GET_SEGMENT (fragP->fr_symbol) != segment
+#if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
+ || S_IS_EXTERNAL (fragP->fr_symbol)
+ || S_IS_WEAK (fragP->fr_symbol)
+#endif
+ )
{
- /* symbol is undefined in this segment */
+ /* Symbol is undefined in this segment, or we need to keep a
+ reloc so that weak symbols can be overridden. */
+ int size = (fragP->fr_subtype & CODE16) ? 2 : 4;
+#ifdef BFD_ASSEMBLER
+ enum bfd_reloc_code_real reloc_type;
+#else
+ int reloc_type;
+#endif
+ unsigned char *opcode;
+ int old_fr_fix;
+
+ if (fragP->fr_var != NO_RELOC)
+ reloc_type = fragP->fr_var;
+ else if (size == 2)
+ reloc_type = BFD_RELOC_16_PCREL;
+ else
+ reloc_type = BFD_RELOC_32_PCREL;
+
+ old_fr_fix = fragP->fr_fix;
+ opcode = (unsigned char *) fragP->fr_opcode;
+
switch (opcode[0])
{
- case JUMP_PC_RELATIVE: /* make jmp (0xeb) a dword displacement jump */
- opcode[0] = 0xe9; /* dword disp jmp */
- fragP->fr_fix += 4;
- fix_new (fragP, old_fr_fix, 4,
+ case JUMP_PC_RELATIVE:
+ /* Make jmp (0xeb) a dword displacement jump. */
+ opcode[0] = 0xe9;
+ fragP->fr_fix += size;
+ fix_new (fragP, old_fr_fix, size,
fragP->fr_symbol,
- (symbolS *) 0,
- fragP->fr_offset, 1, NO_RELOC);
+ fragP->fr_offset, 1,
+ reloc_type);
break;
default:
- /* This changes the byte-displacement jump 0x7N -->
- the dword-displacement jump 0x0f8N */
+ /* This changes the byte-displacement jump 0x7N
+ to the dword-displacement jump 0x0f,0x8N. */
opcode[1] = opcode[0] + 0x10;
- opcode[0] = TWO_BYTE_OPCODE_ESCAPE; /* two-byte escape */
- fragP->fr_fix += 1 + 4; /* we've added an opcode byte */
- fix_new (fragP, old_fr_fix + 1, 4,
+ opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
+ /* We've added an opcode byte. */
+ fragP->fr_fix += 1 + size;
+ fix_new (fragP, old_fr_fix + 1, size,
fragP->fr_symbol,
- (symbolS *) 0,
- fragP->fr_offset, 1, NO_RELOC);
+ fragP->fr_offset, 1,
+ reloc_type);
break;
}
frag_wane (fragP);
+ return fragP->fr_fix - old_fr_fix;
}
- return (fragP->fr_var + fragP->fr_fix - old_fr_fix);
-} /* md_estimate_size_before_relax() */
-\f
-/*
- * md_convert_frag();
- *
- * Called after relax() is finished.
- * In: Address of frag.
- * fr_type == rs_machine_dependent.
- * fr_subtype is what the address relaxed to.
- *
- * Out: Any fixSs and constants are set up.
- * Caller will turn frag into a ".space 0".
- */
+ /* Guess a short jump. */
+ return 1;
+}
+
+/* Called after relax() is finished.
+
+ In: Address of frag.
+ fr_type == rs_machine_dependent.
+ fr_subtype is what the address relaxed to.
+
+ Out: Any fixSs and constants are set up.
+ Caller will turn frag into a ".space 0". */
+
#ifndef BFD_ASSEMBLER
void
-md_convert_frag (headers, fragP)
- object_headers *headers;
+md_convert_frag (headers, sec, fragP)
+ object_headers *headers ATTRIBUTE_UNUSED;
+ segT sec ATTRIBUTE_UNUSED;
register fragS *fragP;
#else
void
md_convert_frag (abfd, sec, fragP)
- bfd *abfd;
- segT sec;
+ bfd *abfd ATTRIBUTE_UNUSED;
+ segT sec ATTRIBUTE_UNUSED;
register fragS *fragP;
#endif
{
register unsigned char *opcode;
unsigned char *where_to_put_displacement = NULL;
- unsigned int target_address;
- unsigned int opcode_address;
+ offsetT target_address;
+ offsetT opcode_address;
unsigned int extension = 0;
- int displacement_from_opcode_start;
+ offsetT displacement_from_opcode_start;
opcode = (unsigned char *) fragP->fr_opcode;
- /* Address we want to reach in file space. */
+ /* Address we want to reach in file space. */
target_address = S_GET_VALUE (fragP->fr_symbol) + fragP->fr_offset;
+#ifdef BFD_ASSEMBLER
+ /* Not needed otherwise? */
+ target_address += symbol_get_frag (fragP->fr_symbol)->fr_address;
+#endif
- /* Address opcode resides at in file space. */
+ /* Address opcode resides at in file space. */
opcode_address = fragP->fr_address + fragP->fr_fix;
- /* Displacement from opcode start to fill into instruction. */
+ /* Displacement from opcode start to fill into instruction. */
displacement_from_opcode_start = target_address - opcode_address;
switch (fragP->fr_subtype)
{
- case ENCODE_RELAX_STATE (COND_JUMP, BYTE):
- case ENCODE_RELAX_STATE (UNCOND_JUMP, BYTE):
- /* don't have to change opcode */
- extension = 1; /* 1 opcode + 1 displacement */
+ case ENCODE_RELAX_STATE (COND_JUMP, SMALL):
+ case ENCODE_RELAX_STATE (COND_JUMP, SMALL16):
+ case ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL):
+ case ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL16):
+ /* Don't have to change opcode. */
+ extension = 1; /* 1 opcode + 1 displacement */
where_to_put_displacement = &opcode[1];
break;
- case ENCODE_RELAX_STATE (COND_JUMP, WORD):
- opcode[1] = TWO_BYTE_OPCODE_ESCAPE;
- opcode[2] = opcode[0] + 0x10;
- opcode[0] = WORD_PREFIX_OPCODE;
- extension = 4; /* 3 opcode + 2 displacement */
- where_to_put_displacement = &opcode[3];
+ case ENCODE_RELAX_STATE (COND_JUMP, BIG):
+ extension = 5; /* 2 opcode + 4 displacement */
+ opcode[1] = opcode[0] + 0x10;
+ opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
+ where_to_put_displacement = &opcode[2];
break;
- case ENCODE_RELAX_STATE (UNCOND_JUMP, WORD):
- opcode[1] = 0xe9;
- opcode[0] = WORD_PREFIX_OPCODE;
- extension = 3; /* 2 opcode + 2 displacement */
- where_to_put_displacement = &opcode[2];
+ case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG):
+ extension = 4; /* 1 opcode + 4 displacement */
+ opcode[0] = 0xe9;
+ where_to_put_displacement = &opcode[1];
break;
- case ENCODE_RELAX_STATE (COND_JUMP, DWORD):
+ case ENCODE_RELAX_STATE (COND_JUMP, BIG16):
+ extension = 3; /* 2 opcode + 2 displacement */
opcode[1] = opcode[0] + 0x10;
opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
- extension = 5; /* 2 opcode + 4 displacement */
where_to_put_displacement = &opcode[2];
break;
- case ENCODE_RELAX_STATE (UNCOND_JUMP, DWORD):
+ case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16):
+ extension = 2; /* 1 opcode + 2 displacement */
opcode[0] = 0xe9;
- extension = 4; /* 1 opcode + 4 displacement */
where_to_put_displacement = &opcode[1];
break;
BAD_CASE (fragP->fr_subtype);
break;
}
- /* now put displacement after opcode */
+ /* Now put displacement after opcode. */
md_number_to_chars ((char *) where_to_put_displacement,
- displacement_from_opcode_start - extension,
+ (valueT) (displacement_from_opcode_start - extension),
SIZE_FROM_RELAX_STATE (fragP->fr_subtype));
fragP->fr_fix += extension;
}
\f
+/* Size of byte displacement jmp. */
+int md_short_jump_size = 2;
+
+/* Size of dword displacement jmp. */
+int md_long_jump_size = 5;
-int md_short_jump_size = 2; /* size of byte displacement jmp */
-int md_long_jump_size = 5; /* size of dword displacement jmp */
-const int md_reloc_size = 8; /* Size of relocation record */
+/* Size of relocation record. */
+const int md_reloc_size = 8;
void
md_create_short_jump (ptr, from_addr, to_addr, frag, to_symbol)
char *ptr;
addressT from_addr, to_addr;
- fragS *frag;
- symbolS *to_symbol;
+ fragS *frag ATTRIBUTE_UNUSED;
+ symbolS *to_symbol ATTRIBUTE_UNUSED;
{
- long offset;
+ offsetT offset;
offset = to_addr - (from_addr + 2);
- md_number_to_chars (ptr, (long) 0xeb, 1); /* opcode for byte-disp jump */
- md_number_to_chars (ptr + 1, offset, 1);
+ /* Opcode for byte-disp jump. */
+ md_number_to_chars (ptr, (valueT) 0xeb, 1);
+ md_number_to_chars (ptr + 1, (valueT) offset, 1);
}
void
md_create_long_jump (ptr, from_addr, to_addr, frag, to_symbol)
char *ptr;
addressT from_addr, to_addr;
- fragS *frag;
- symbolS *to_symbol;
+ fragS *frag ATTRIBUTE_UNUSED;
+ symbolS *to_symbol ATTRIBUTE_UNUSED;
{
- long offset;
+ offsetT offset;
- if (flagseen['m'])
- {
- offset = to_addr - S_GET_VALUE (to_symbol);
- md_number_to_chars (ptr, 0xe9, 1); /* opcode for long jmp */
- md_number_to_chars (ptr + 1, offset, 4);
- fix_new (frag, (ptr + 1) - frag->fr_literal, 4,
- to_symbol, (symbolS *) 0, (long) 0, 0, NO_RELOC);
- }
- else
- {
- offset = to_addr - (from_addr + 5);
- md_number_to_chars (ptr, (long) 0xe9, 1);
- md_number_to_chars (ptr + 1, offset, 4);
- }
-}
-\f
-int
-md_parse_option (argP, cntP, vecP)
- char **argP;
- int *cntP;
- char ***vecP;
-{
- return 1;
+ offset = to_addr - (from_addr + 5);
+ md_number_to_chars (ptr, (valueT) 0xe9, 1);
+ md_number_to_chars (ptr + 1, (valueT) offset, 4);
}
\f
-void /* Knows about order of bytes in address. */
-md_number_to_chars (con, value, nbytes)
- char con[]; /* Return 'nbytes' of chars here. */
- valueT value; /* The value of the bits. */
- int nbytes; /* Number of bytes in the output. */
-{
- register char *p = con;
-
- switch (nbytes)
- {
- case 1:
- p[0] = value & 0xff;
- break;
- case 2:
- p[0] = value & 0xff;
- p[1] = (value >> 8) & 0xff;
- break;
- case 4:
- p[0] = value & 0xff;
- p[1] = (value >> 8) & 0xff;
- p[2] = (value >> 16) & 0xff;
- p[3] = (value >> 24) & 0xff;
- break;
- default:
- BAD_CASE (nbytes);
- }
-}
-
-
/* Apply a fixup (fixS) to segment data, once it has been determined
by our caller that we have all the info we need to fix it up.
the same (little-endian) format, so we don't need to care about which
we are handling. */
-static void
-md_apply_fix_1 (fixP, value)
- fixS *fixP; /* The fix we're to put in */
- long value; /* The value of the bits. */
+int
+md_apply_fix3 (fixP, valp, seg)
+ /* The fix we're to put in. */
+ fixS *fixP;
+
+ /* Pointer to the value of the bits. */
+ valueT *valp;
+
+ /* Segment fix is from. */
+ segT seg ATTRIBUTE_UNUSED;
{
register char *p = fixP->fx_where + fixP->fx_frag->fr_literal;
+ valueT value = *valp;
- switch (fixP->fx_size)
+#if defined (BFD_ASSEMBLER) && !defined (TE_Mach)
+ if (fixP->fx_pcrel)
{
- case 1:
- *p = value;
- break;
- case 2:
- *p++ = value;
- *p = (value >> 8);
- break;
- case 4:
- *p++ = value;
- *p++ = (value >> 8);
- *p++ = (value >> 16);
- *p = (value >> 24);
- break;
- default:
- BAD_CASE (fixP->fx_size);
- }
-}
+ switch (fixP->fx_r_type)
+ {
+ default:
+ break;
-#ifdef BFD_ASSEMBLER
-int
-md_apply_fix (fixP, valp)
- fixS *fixP;
- long *valp;
-{
- md_apply_fix_1 (fixP, *valp);
- return 1;
-}
-#else
-void
-md_apply_fix (fixP, val)
- fixS *fixP;
- long val;
-{
- md_apply_fix_1 (fixP, val);
-}
-#endif
+ case BFD_RELOC_32:
+ fixP->fx_r_type = BFD_RELOC_32_PCREL;
+ break;
+ case BFD_RELOC_16:
+ fixP->fx_r_type = BFD_RELOC_16_PCREL;
+ break;
+ case BFD_RELOC_8:
+ fixP->fx_r_type = BFD_RELOC_8_PCREL;
+ break;
+ }
+ }
-long /* Knows about the byte order in a word. */
-md_chars_to_number (con, nbytes)
- unsigned char con[]; /* Low order byte 1st. */
- int nbytes; /* Number of bytes in the input. */
-{
- long retval;
- for (retval = 0, con += nbytes - 1; nbytes--; con--)
+ /* This is a hack. There should be a better way to handle this.
+ This covers for the fact that bfd_install_relocation will
+ subtract the current location (for partial_inplace, PC relative
+ relocations); see more below. */
+ if ((fixP->fx_r_type == BFD_RELOC_32_PCREL
+ || fixP->fx_r_type == BFD_RELOC_16_PCREL
+ || fixP->fx_r_type == BFD_RELOC_8_PCREL)
+ && fixP->fx_addsy)
{
- retval <<= BITS_PER_CHAR;
- retval |= *con;
+#ifndef OBJ_AOUT
+ if (OUTPUT_FLAVOR == bfd_target_elf_flavour
+#ifdef TE_PE
+ || OUTPUT_FLAVOR == bfd_target_coff_flavour
+#endif
+ )
+ value += fixP->fx_where + fixP->fx_frag->fr_address;
+#endif
+#if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
+ if (OUTPUT_FLAVOR == bfd_target_elf_flavour)
+ {
+ segT fseg = S_GET_SEGMENT (fixP->fx_addsy);
+
+ if ((fseg == seg
+ || (symbol_section_p (fixP->fx_addsy)
+ && fseg != absolute_section))
+ && ! S_IS_EXTERNAL (fixP->fx_addsy)
+ && ! S_IS_WEAK (fixP->fx_addsy)
+ && S_IS_DEFINED (fixP->fx_addsy)
+ && ! S_IS_COMMON (fixP->fx_addsy))
+ {
+ /* Yes, we add the values in twice. This is because
+ bfd_perform_relocation subtracts them out again. I think
+ bfd_perform_relocation is broken, but I don't dare change
+ it. FIXME. */
+ value += fixP->fx_where + fixP->fx_frag->fr_address;
+ }
+ }
+#endif
+#if defined (OBJ_COFF) && defined (TE_PE)
+ /* For some reason, the PE format does not store a section
+ address offset for a PC relative symbol. */
+ if (S_GET_SEGMENT (fixP->fx_addsy) != seg)
+ value += md_pcrel_from (fixP);
+#endif
}
- return retval;
-}
-\f
+ /* Fix a few things - the dynamic linker expects certain values here,
+ and we must not dissappoint it. */
+#if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
+ if (OUTPUT_FLAVOR == bfd_target_elf_flavour
+ && fixP->fx_addsy)
+ switch (fixP->fx_r_type)
+ {
+ case BFD_RELOC_386_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;
+ break;
+ case BFD_RELOC_386_GOTPC:
+
+/* This is tough to explain. We end up with this one if we have
+ * operands that look like "_GLOBAL_OFFSET_TABLE_+[.-.L284]". The goal
+ * here is to obtain the absolute address of the GOT, and it is strongly
+ * preferable from a performance point of view to avoid using a runtime
+ * relocation for this. The actual sequence of instructions often look
+ * something like:
+ *
+ * call .L66
+ * .L66:
+ * popl %ebx
+ * addl $_GLOBAL_OFFSET_TABLE_+[.-.L66],%ebx
+ *
+ * The call and pop essentially return the absolute address of
+ * the label .L66 and store it in %ebx. The linker itself will
+ * ultimately change the first operand of the addl so that %ebx points to
+ * the GOT, but to keep things simple, the .o file must have this operand
+ * set so that it generates not the absolute address of .L66, but the
+ * absolute address of itself. This allows the linker itself simply
+ * treat a GOTPC relocation as asking for a pcrel offset to the GOT to be
+ * added in, and the addend of the relocation is stored in the operand
+ * field for the instruction itself.
+ *
+ * Our job here is to fix the operand so that it would add the correct
+ * offset so that %ebx would point to itself. The thing that is tricky is
+ * that .-.L66 will point to the beginning of the instruction, so we need
+ * to further modify the operand so that it will point to itself.
+ * There are other cases where you have something like:
+ *
+ * .long $_GLOBAL_OFFSET_TABLE_+[.-.L66]
+ *
+ * and here no correction would be required. Internally in the assembler
+ * we treat operands of this form as not being pcrel since the '.' is
+ * explicitly mentioned, and I wonder whether it would simplify matters
+ * to do it this way. Who knows. In earlier versions of the PIC patches,
+ * the pcrel_adjust field was used to store the correction, but since the
+ * expression is not pcrel, I felt it would be confusing to do it this
+ * way. */
+
+ value -= 1;
+ break;
+ case BFD_RELOC_386_GOT32:
+ value = 0; /* Fully resolved at runtime. No addend. */
+ break;
+ case BFD_RELOC_386_GOTOFF:
+ break;
+
+ case BFD_RELOC_VTABLE_INHERIT:
+ case BFD_RELOC_VTABLE_ENTRY:
+ fixP->fx_done = 0;
+ return 1;
+
+ default:
+ break;
+ }
+#endif /* defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) */
+ *valp = value;
+#endif /* defined (BFD_ASSEMBLER) && !defined (TE_Mach) */
+ md_number_to_chars (p, value, fixP->fx_size);
+ return 1;
+}
+\f
#define MAX_LITTLENUMS 6
-/* Turn the string pointed to by litP into a floating point constant of type
- type, and emit the appropriate bytes. The number of LITTLENUMS emitted
- is stored in *sizeP . An error message is returned, or NULL on OK. */
+/* Turn the string pointed to by litP into a floating point constant
+ of type TYPE, and emit the appropriate bytes. The number of
+ LITTLENUMS emitted is stored in *SIZEP. An error message is
+ returned, or NULL on OK. */
+
char *
md_atof (type, litP, sizeP)
- char type;
+ int type;
char *litP;
int *sizeP;
{
default:
*sizeP = 0;
- return "Bad call to md_atof ()";
+ return _("Bad call to md_atof ()");
}
t = atof_ieee (input_line_pointer, type, words);
if (t)
the bigendian 386. */
for (wordP = words + prec - 1; prec--;)
{
- md_number_to_chars (litP, (long) (*wordP--), sizeof (LITTLENUM_TYPE));
+ md_number_to_chars (litP, (valueT) (*wordP--), sizeof (LITTLENUM_TYPE));
litP += sizeof (LITTLENUM_TYPE);
}
return 0;
static char *
output_invalid (c)
- char c;
+ int c;
{
if (isprint (c))
sprintf (output_invalid_buf, "'%c'", c);
return output_invalid_buf;
}
-static reg_entry *
-parse_register (reg_string)
- /* reg_string starts *before* REGISTER_PREFIX */
+/* REG_STRING starts *before* REGISTER_PREFIX. */
+
+static const reg_entry *
+parse_register (reg_string, end_op)
char *reg_string;
+ char **end_op;
{
- register char *s = reg_string;
- register char *p;
- char reg_name_given[MAX_REG_NAME_SIZE];
+ char *s = reg_string;
+ char *p;
+ char reg_name_given[MAX_REG_NAME_SIZE + 1];
+ const reg_entry *r;
- s++; /* skip REGISTER_PREFIX */
- for (p = reg_name_given; is_register_char (*s); p++, s++)
+ /* Skip possible REGISTER_PREFIX and possible whitespace. */
+ if (*s == REGISTER_PREFIX)
+ ++s;
+
+ if (is_space_char (*s))
+ ++s;
+
+ p = reg_name_given;
+ while ((*p++ = register_chars[(unsigned char) *s]) != '\0')
{
- *p = register_chars[*s];
if (p >= reg_name_given + MAX_REG_NAME_SIZE)
- return (reg_entry *) 0;
+ return (const reg_entry *) NULL;
+ 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);
+
+ /* Handle floating point regs, allowing spaces in the (i) part. */
+ if (r == i386_regtab /* %st is first entry of table */)
+ {
+ if (is_space_char (*s))
+ ++s;
+ if (*s == '(')
+ {
+ ++s;
+ if (is_space_char (*s))
+ ++s;
+ if (*s >= '0' && *s <= '7')
+ {
+ r = &i386_float_regtab[*s - '0'];
+ ++s;
+ if (is_space_char (*s))
+ ++s;
+ if (*s == ')')
+ {
+ *end_op = s + 1;
+ return r;
+ }
+ }
+ /* We have "%st(" then garbage. */
+ return (const reg_entry *) NULL;
+ }
}
- *p = '\0';
- return (reg_entry *) hash_find (reg_hash, reg_name_given);
+
+ return r;
}
+\f
+#if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
+const char *md_shortopts = "kVQ:sq";
+#else
+const char *md_shortopts = "q";
+#endif
+struct option md_longopts[] = {
+ {NULL, no_argument, NULL, 0}
+};
+size_t md_longopts_size = sizeof (md_longopts);
+int
+md_parse_option (c, arg)
+ int c;
+ char *arg ATTRIBUTE_UNUSED;
+{
+ switch (c)
+ {
+ case 'q':
+ quiet_warnings = 1;
+ break;
-/* We have no need to default values of symbols. */
+#if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
+ /* -Qy, -Qn: SVR4 arguments controlling whether a .comment section
+ should be emitted or not. FIXME: Not implemented. */
+ case 'Q':
+ break;
-/* ARGSUSED */
-symbolS *
-md_undefined_symbol (name)
- char *name;
+ /* -V: SVR4 argument to print version ID. */
+ case 'V':
+ print_version_id ();
+ break;
+
+ /* -k: Ignore for FreeBSD compatibility. */
+ case 'k':
+ break;
+
+ case 's':
+ /* -s: On i386 Solaris, this tells the native assembler to use
+ .stab instead of .stab.excl. We always use .stab anyhow. */
+ break;
+#endif
+
+ default:
+ return 0;
+ }
+ return 1;
+}
+
+void
+md_show_usage (stream)
+ FILE *stream;
{
- return 0;
+#if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
+ fprintf (stream, _("\
+ -Q ignored\n\
+ -V print assembler version number\n\
+ -k ignored\n\
+ -q quieten some warnings\n\
+ -s ignored\n"));
+#else
+ fprintf (stream, _("\
+ -q quieten some warnings\n"));
+#endif
}
-/* Parse an operand that is machine-specific.
- We just return without modifying the expression if we have nothing
- to do. */
+#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)))
-/* ARGSUSED */
-void
-md_operand (expressionP)
- expressionS *expressionP;
+/* Pick the target format to use. */
+
+const char *
+i386_target_format ()
{
+ switch (OUTPUT_FLAVOR)
+ {
+#ifdef OBJ_MAYBE_AOUT
+ case bfd_target_aout_flavour:
+ return AOUT_TARGET_FORMAT;
+#endif
+#ifdef OBJ_MAYBE_COFF
+ case bfd_target_coff_flavour:
+ return "coff-i386";
+#endif
+#ifdef OBJ_MAYBE_ELF
+ case bfd_target_elf_flavour:
+ return "elf32-i386";
+#endif
+ default:
+ abort ();
+ return NULL;
+ }
+}
+
+#endif /* OBJ_MAYBE_ more than one */
+#endif /* BFD_ASSEMBLER */
+\f
+symbolS *
+md_undefined_symbol (name)
+ char *name;
+{
+ if (name[0] == GLOBAL_OFFSET_TABLE_NAME[0]
+ && name[1] == GLOBAL_OFFSET_TABLE_NAME[1]
+ && name[2] == GLOBAL_OFFSET_TABLE_NAME[2]
+ && strcmp (name, GLOBAL_OFFSET_TABLE_NAME) == 0)
+ {
+ if (!GOT_symbol)
+ {
+ if (symbol_find (name))
+ as_bad (_("GOT already in symbol table"));
+ GOT_symbol = symbol_new (name, undefined_section,
+ (valueT) 0, &zero_address_frag);
+ };
+ return GOT_symbol;
+ }
+ return 0;
}
/* Round up a section size to the appropriate boundary. */
+
valueT
md_section_align (segment, size)
- segT segment;
+ segT segment ATTRIBUTE_UNUSED;
valueT size;
{
- return size; /* Byte alignment is fine */
+#ifdef BFD_ASSEMBLER
+#if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
+ if (OUTPUT_FLAVOR == bfd_target_aout_flavour)
+ {
+ /* For a.out, force the section size to be aligned. If we don't do
+ this, BFD will align it for us, but it will not write out the
+ final bytes of the section. This may be a bug in BFD, but it is
+ easier to fix it here since that is how the other a.out targets
+ work. */
+ int align;
+
+ align = bfd_get_section_alignment (stdoutput, segment);
+ size = ((size + (1 << align) - 1) & ((valueT) -1 << align));
+ }
+#endif
+#endif
+
+ return size;
}
-/* Exactly what point is a PC-relative offset relative TO?
- On the i386, they're relative to the address of the offset, plus
- its size. (??? Is this right? FIXME-SOON!) */
+/* On the i386, PC-relative offsets are relative to the start of the
+ next instruction. That is, the address of the offset, plus its
+ size, since the offset is always the last part of the insn. */
+
long
md_pcrel_from (fixP)
fixS *fixP;
#ifndef I386COFF
static void
-s_bss ()
+s_bss (ignore)
+ int ignore ATTRIBUTE_UNUSED;
{
register int temp;
temp = get_absolute_expression ();
-#ifdef BFD_ASSEMBLER
subseg_set (bss_section, (subsegT) temp);
-#else
- subseg_new (bss_section, (subsegT) temp);
-#endif
demand_empty_rest_of_line ();
}
#endif
-
#ifdef BFD_ASSEMBLER
+void
+i386_validate_fix (fixp)
+ fixS *fixp;
+{
+ if (fixp->fx_subsy && fixp->fx_subsy == GOT_symbol)
+ {
+ fixp->fx_r_type = BFD_RELOC_386_GOTOFF;
+ fixp->fx_subsy = 0;
+ }
+}
+
arelent *
tc_gen_reloc (section, fixp)
- asection *section;
+ asection *section ATTRIBUTE_UNUSED;
fixS *fixp;
{
- arelent *reloc;
+ arelent *rel;
bfd_reloc_code_real_type code;
-#define F(SZ,PCREL) (((SZ) << 1) + (PCREL))
- switch (F (fixp->fx_size, fixp->fx_pcrel))
+ switch (fixp->fx_r_type)
{
-#define MAP(SZ,PCREL,TYPE) case F(SZ,PCREL): code = (TYPE); break
- MAP (1, 0, BFD_RELOC_8);
- MAP (2, 0, BFD_RELOC_16);
- MAP (4, 0, BFD_RELOC_32);
- MAP (1, 1, BFD_RELOC_8_PCREL);
- MAP (2, 1, BFD_RELOC_16_PCREL);
- MAP (4, 1, BFD_RELOC_32_PCREL);
+ case BFD_RELOC_386_PLT32:
+ case BFD_RELOC_386_GOT32:
+ case BFD_RELOC_386_GOTOFF:
+ case BFD_RELOC_386_GOTPC:
+ case BFD_RELOC_RVA:
+ case BFD_RELOC_VTABLE_ENTRY:
+ case BFD_RELOC_VTABLE_INHERIT:
+ code = fixp->fx_r_type;
+ break;
default:
- abort ();
+ if (fixp->fx_pcrel)
+ {
+ switch (fixp->fx_size)
+ {
+ default:
+ as_bad (_("can not do %d byte pc-relative relocation"),
+ fixp->fx_size);
+ code = BFD_RELOC_32_PCREL;
+ break;
+ case 1: code = BFD_RELOC_8_PCREL; break;
+ case 2: code = BFD_RELOC_16_PCREL; break;
+ case 4: code = BFD_RELOC_32_PCREL; break;
+ }
+ }
+ else
+ {
+ switch (fixp->fx_size)
+ {
+ default:
+ as_bad (_("can not do %d byte relocation"), fixp->fx_size);
+ code = BFD_RELOC_32;
+ break;
+ case 1: code = BFD_RELOC_8; break;
+ case 2: code = BFD_RELOC_16; break;
+ case 4: code = BFD_RELOC_32; break;
+ }
+ }
+ break;
}
-#undef MAP
-#undef F
- reloc = (arelent *) bfd_alloc_by_size_t (stdoutput, sizeof (arelent));
- assert (reloc != 0);
- reloc->sym_ptr_ptr = &fixp->fx_addsy->bsym;
- reloc->address = fixp->fx_frag->fr_address + fixp->fx_where;
+ if (code == BFD_RELOC_32
+ && GOT_symbol
+ && fixp->fx_addsy == GOT_symbol)
+ 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 (fixp->fx_pcrel)
- reloc->addend = fixp->fx_addnumber;
+ rel->addend = fixp->fx_addnumber;
else
- reloc->addend = 0;
+ rel->addend = 0;
- reloc->howto = bfd_reloc_type_lookup (stdoutput, code);
- assert (reloc->howto != 0);
+ rel->howto = bfd_reloc_type_lookup (stdoutput, code);
+ if (rel->howto == NULL)
+ {
+ as_bad_where (fixp->fx_file, fixp->fx_line,
+ _("cannot represent relocation type %s"),
+ bfd_get_reloc_code_name (code));
+ /* Set howto to a garbage value so that we can keep going. */
+ rel->howto = bfd_reloc_type_lookup (stdoutput, BFD_RELOC_32);
+ assert (rel->howto != NULL);
+ }
- return reloc;
+ return rel;
}
-#else /* ! BFD_ASSEMBLER */
+#else /* ! BFD_ASSEMBLER */
#if (defined(OBJ_AOUT) | defined(OBJ_BOUT))
void
fixS *fixP;
relax_addressT segment_address_in_file;
{
- /*
- * In: length of relocation (or of address) in chars: 1, 2 or 4.
- * Out: GNU LD relocation length code: 0, 1, or 2.
- */
+ /* In: length of relocation (or of address) in chars: 1, 2 or 4.
+ Out: GNU LD relocation length code: 0, 1, or 2. */
- static const unsigned char nbytes_r_length[] = {42, 0, 1, 42, 2};
+ static const unsigned char nbytes_r_length[] = { 42, 0, 1, 42, 2 };
long r_symbolnum;
know (fixP->fx_addsy != NULL);
md_number_to_chars (where,
- fixP->fx_frag->fr_address + fixP->fx_where - segment_address_in_file,
+ (valueT) (fixP->fx_frag->fr_address
+ + fixP->fx_where - segment_address_in_file),
4);
r_symbolnum = (S_IS_DEFINED (fixP->fx_addsy)
| (((fixP->fx_pcrel << 0) & 0x01) & 0x0f));
}
-#endif /* OBJ_AOUT or OBJ_BOUT */
+#endif /* OBJ_AOUT or OBJ_BOUT. */
#if defined (I386COFF)
tc_coff_fix2rtype (fixP)
fixS *fixP;
{
+ if (fixP->fx_r_type == R_IMAGEBASE)
+ return R_IMAGEBASE;
+
return (fixP->fx_pcrel ?
(fixP->fx_size == 1 ? R_PCRBYTE :
fixP->fx_size == 2 ? R_PCRWORD :
(fixP->fx_size == 1 ? R_RELBYTE :
fixP->fx_size == 2 ? R_RELWORD :
R_DIR32));
-
-
}
int
return 0;
}
-#endif /* I386COFF */
+#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'
-#endif /* BFD_ASSEMBLER? */
+ 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. */
+#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
+#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);
-/* end of tc-i386.c */
+ 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)
+ i.suffix = DWORD_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 == ':')
+ {
+ 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 ('[');
+ intel_parser.is_mem = 1;
+
+ /* Add a '+' to the displacement string if necessary. */
+ if (*intel_parser.disp != '\0')
+ strcat (intel_parser.disp, "+");
+
+ return (intel_expr () && intel_match_token (']') && intel_e10_1 ());
+ }
+
+ /* 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 ('[');
+ intel_parser.is_mem = 1;
+
+ /* 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')
+ strcat (intel_parser.disp, "+");
+
+ return (intel_expr () && intel_match_token (']'));
+ }
+
+ /* 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);
+ 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.op_modifier != 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;
+}