/* i386.c -- Assemble code for the Intel 80386
- Copyright (C) 1989, 91, 92, 93, 94, 95, 96, 97, 98, 99, 2000
- Free Software Foundation.
+ Copyright 1989, 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999,
+ 2000, 2001, 2002
+ Free Software Foundation, Inc.
This file is part of GAS, the GNU Assembler.
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.
- */
-
-#include <ctype.h>
+/* Intel 80386 machine specific gas.
+ Written by Eliot Dresselhaus (eliot@mgm.mit.edu).
+ x86_64 support by Jan Hubicka (jh@suse.cz)
+ Bugs & suggestions are completely welcome. This is free software.
+ Please help us make it better. */
#include "as.h"
+#include "safe-ctype.h"
#include "subsegs.h"
+#include "dwarf2dbg.h"
#include "opcode/i386.h"
-#ifndef TC_RELOC
-#define TC_RELOC(X,Y) (Y)
-#endif
-
#ifndef REGISTER_WARNINGS
#define REGISTER_WARNINGS 1
#endif
#define SCALE1_WHEN_NO_INDEX 1
#endif
-#define true 1
-#define false 0
+#ifdef BFD_ASSEMBLER
+#define RELOC_ENUM enum bfd_reloc_code_real
+#else
+#define RELOC_ENUM int
+#endif
+
+#ifndef DEFAULT_ARCH
+#define DEFAULT_ARCH "i386"
+#endif
+
+#ifndef INLINE
+#if __GNUC__ >= 2
+#define INLINE __inline__
+#else
+#define INLINE
+#endif
+#endif
-static unsigned int mode_from_disp_size PARAMS ((unsigned int));
-static int fits_in_signed_byte PARAMS ((long));
-static int fits_in_unsigned_byte PARAMS ((long));
-static int fits_in_unsigned_word PARAMS ((long));
-static int fits_in_signed_word PARAMS ((long));
-static int smallest_imm_type PARAMS ((long));
+static INLINE unsigned int mode_from_disp_size PARAMS ((unsigned int));
+static INLINE int fits_in_signed_byte PARAMS ((offsetT));
+static INLINE int fits_in_unsigned_byte PARAMS ((offsetT));
+static INLINE int fits_in_unsigned_word PARAMS ((offsetT));
+static INLINE int fits_in_signed_word PARAMS ((offsetT));
+static INLINE int fits_in_unsigned_long PARAMS ((offsetT));
+static INLINE int fits_in_signed_long PARAMS ((offsetT));
+static int smallest_imm_type PARAMS ((offsetT));
+static offsetT offset_in_range PARAMS ((offsetT, int));
static int add_prefix PARAMS ((unsigned int));
-static void set_16bit_code_flag PARAMS ((int));
-static void set_16bit_gcc_code_flag PARAMS((int));
+static void set_code_flag PARAMS ((int));
+static void set_16bit_gcc_code_flag PARAMS ((int));
static void set_intel_syntax PARAMS ((int));
-
-#ifdef BFD_ASSEMBLER
-static bfd_reloc_code_real_type reloc
- PARAMS ((int, int, bfd_reloc_code_real_type));
+static void set_cpu_arch PARAMS ((int));
+static char *output_invalid PARAMS ((int c));
+static int i386_operand PARAMS ((char *operand_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));
+static char *parse_insn PARAMS ((char *, char *));
+static char *parse_operands PARAMS ((char *, const char *));
+static void swap_operands PARAMS ((void));
+static void optimize_imm PARAMS ((void));
+static void optimize_disp PARAMS ((void));
+static int match_template PARAMS ((void));
+static int check_string PARAMS ((void));
+static int process_suffix PARAMS ((void));
+static int check_byte_reg PARAMS ((void));
+static int check_long_reg PARAMS ((void));
+static int check_qword_reg PARAMS ((void));
+static int check_word_reg PARAMS ((void));
+static int finalize_imm PARAMS ((void));
+static int process_operands PARAMS ((void));
+static const seg_entry *build_modrm_byte PARAMS ((void));
+static void output_insn PARAMS ((void));
+static void output_branch PARAMS ((void));
+static void output_jump PARAMS ((void));
+static void output_interseg_jump PARAMS ((void));
+static void output_imm PARAMS ((void));
+static void output_disp PARAMS ((void));
+#ifndef I386COFF
+static void s_bss PARAMS ((int));
#endif
+static const char *default_arch = DEFAULT_ARCH;
+
/* '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 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];
-
- /* Relocation type for operand */
-#ifdef BFD_ASSEMBLER
- enum bfd_reloc_code_real disp_reloc[MAX_OPERANDS];
-#else
- int disp_reloc[MAX_OPERANDS];
-#endif
+ /* Displacement expression, immediate expression, or register for each
+ operand. */
+ union i386_op op[MAX_OPERANDS];
- /* Immediate operands (if given) for each operand. */
- expressionS *imms[MAX_OPERANDS];
+ /* Flags for operands. */
+ unsigned int flags[MAX_OPERANDS];
+#define Operand_PCrel 1
- /* Register operands (if given) for each operand. */
- const reg_entry *regs[MAX_OPERANDS];
+ /* Relocation type for operand */
+ RELOC_ENUM reloc[MAX_OPERANDS];
/* BASE_REG, INDEX_REG, and LOG2_SCALE_FACTOR are used to encode
the base index byte below. */
unsigned int log2_scale_factor;
/* SEG gives the seg_entries of this insn. They are zero unless
- explicit segment overrides are given. */
- const seg_entry *seg[2]; /* segments for memory operands (if given) */
+ explicit segment overrides are given. */
+ const seg_entry *seg[2];
/* PREFIX holds all the given prefix opcodes (usually null).
PREFIXES is the number of prefix opcodes. */
addressing modes of this insn are encoded. */
modrm_byte rm;
+ rex_byte rex;
sib_byte sib;
};
/* 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[] = "*%-(@";
+const char extra_symbol_chars[] = "*%-(@[";
#else
-const char extra_symbol_chars[] = "*%-(";
+const char extra_symbol_chars[] = "*%-([";
#endif
+#if (defined (TE_I386AIX) \
+ || ((defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)) \
+ && !defined (TE_LINUX) \
+ && !defined (TE_FreeBSD) \
+ && !defined (TE_NetBSD)))
/* This array holds the chars that always start a comment. If the
- pre-processor is disabled, these aren't very useful */
-#if defined (TE_I386AIX) || ((defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)) && ! defined (TE_LINUX))
-/* Putting '/' here makes it impossible to use the divide operator.
- However, we need it for compatibility with SVR4 systems. */
+ pre-processor is disabled, these aren't very useful. */
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
+ #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))
const char line_comment_chars[] = "";
+
#else
+/* 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 '/'
+
const char line_comment_chars[] = "/";
#endif
-const char line_separator_chars[] = "";
+const char line_separator_chars[] = ";";
-/* Chars that can be used to separate mant from exp in floating point nums */
+/* Chars that can be used to separate mant from exp in floating point
+ nums. */
const char EXP_CHARS[] = "eE";
-/* Chars that mean this number is a floating point constant */
-/* As in 0f12.456 */
-/* or 0d1.2345e12 */
+/* Chars that mean this number is a floating point constant
+ As in 0f12.456
+ or 0d1.2345e12. */
const char FLT_CHARS[] = "fFdDxX";
-/* tables for lexical analysis */
+/* Tables for lexical analysis. */
static char mnemonic_chars[256];
static char register_chars[256];
static char operand_chars[256];
static char identifier_chars[256];
static char digit_chars[256];
-/* lexical macros */
+/* Lexical macros. */
#define is_mnemonic_char(x) (mnemonic_chars[(unsigned char) x])
#define is_operand_char(x) (operand_chars[(unsigned char) x])
#define is_register_char(x) (register_chars[(unsigned char) x])
#define is_identifier_char(x) (identifier_chars[(unsigned char) x])
#define is_digit_char(x) (digit_chars[(unsigned char) x])
-/* put here all non-digit non-letter charcters that may occur in an operand */
+/* All non-digit non-letter charcters that may occur in an operand. */
static char operand_special_chars[] = "%$-+(,)*._~/<>|&^!:[@]";
/* md_assemble() always leaves the strings it's passed unaltered. To
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 */
+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;
/* Possible templates for current insn. */
static const templates *current_templates;
-/* Per instruction expressionS buffers: 2 displacements & 2 immediate max. */
+/* Per instruction expressionS buffers: 2 displacements & 2 immediate max. */
static expressionS disp_expressions[2], im_expressions[2];
-static int this_operand; /* current operand we are working on */
+/* Current operand we are working on. */
+static int this_operand;
+
+/* We support four different modes. FLAG_CODE variable is used to distinguish
+ these. */
+
+enum flag_code {
+ CODE_32BIT,
+ CODE_16BIT,
+ CODE_64BIT };
+#define NUM_FLAG_CODE ((int) CODE_64BIT + 1)
+
+static enum flag_code flag_code;
+static int use_rela_relocations = 0;
+
+/* The names used to print error messages. */
+static const char *flag_code_names[] =
+ {
+ "32",
+ "16",
+ "64"
+ };
+
+/* 1 for intel syntax,
+ 0 if att syntax. */
+static int intel_syntax = 0;
+
+/* 1 if register prefix % not required. */
+static int allow_naked_reg = 0;
-static int flag_do_long_jump; /* FIXME what does this do? */
+/* 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';
-static int flag_16bit_code; /* 1 if we're writing 16-bit code, 0 if 32-bit */
+/* Non-zero to quieten some warnings. */
+static int quiet_warnings = 0;
-static int intel_syntax = 0; /* 1 for intel syntax, 0 if att syntax */
+/* CPU name. */
+static const char *cpu_arch_name = NULL;
-static int allow_naked_reg = 0; /* 1 if register prefix % not required */
+/* CPU feature flags. */
+static unsigned int cpu_arch_flags = CpuUnknownFlags | CpuNo64;
-static char stackop_size = '\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. */
+/* If set, conditional jumps are not automatically promoted to handle
+ larger than a byte offset. */
+static unsigned int no_cond_jump_promotion = 0;
+
+/* Pre-defined "_GLOBAL_OFFSET_TABLE_". */
+symbolS *GOT_symbol;
/* Interface to relax_segment.
- There are 2 relax states for 386 jump insns: one for conditional &
- 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 */
+ There are 3 major relax states for 386 jump insns because the
+ different 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 UNCOND_JUMP 0
+#define COND_JUMP 1
+#define COND_JUMP86 2
+
+/* Sizes. */
#define CODE16 1
#define SMALL 0
-#define SMALL16 (SMALL|CODE16)
+#define SMALL16 (SMALL | CODE16)
#define BIG 2
-#define BIG16 (BIG|CODE16)
+#define BIG16 (BIG | CODE16)
#ifndef INLINE
#ifdef __GNUC__
#endif
#endif
-#define ENCODE_RELAX_STATE(type,size) \
- ((relax_substateT)((type<<2) | (size)))
-#define SIZE_FROM_RELAX_STATE(s) \
- ( (((s) & 0x3) == BIG ? 4 : (((s) & 0x3) == BIG16 ? 2 : 1)) )
+#define ENCODE_RELAX_STATE(type, size) \
+ ((relax_substateT) (((type) << 2) | (size)))
+#define TYPE_FROM_RELAX_STATE(s) \
+ ((s) >> 2)
+#define DISP_SIZE_FROM_RELAX_STATE(s) \
+ ((((s) & 3) == BIG ? 4 : (((s) & 3) == 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
/* 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},
-
- {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 extra opcode byte, 3 extra displacement bytes. */
+ 3) how many bytes this mode will have in the variable part of the frag
+ 4) which index into the table to try if we can't fit into this one. */
+
+ /* UNCOND_JUMP states. */
+ {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG)},
+ {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16)},
+ /* dword jmp adds 4 bytes to frag:
+ 0 extra opcode bytes, 4 displacement bytes. */
{0, 0, 4, 0},
- /* word conditionals add 2 bytes to frag:
- 1 extra opcode byte, 1 extra displacement byte. */
+ /* word jmp adds 2 byte2 to frag:
+ 0 extra opcode bytes, 2 displacement bytes. */
{0, 0, 2, 0},
- {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 extra opcode bytes, 3 extra displacement bytes. */
+ /* COND_JUMP states. */
+ {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG)},
+ {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP, BIG16)},
+ /* dword conditionals adds 5 bytes to frag:
+ 1 extra opcode byte, 4 displacement bytes. */
+ {0, 0, 5, 0},
+ /* word conditionals add 3 bytes to frag:
+ 1 extra opcode byte, 2 displacement bytes. */
{0, 0, 3, 0},
- /* word jmp adds 1 byte to frag:
- 0 extra opcode bytes, 1 extra displacement byte. */
- {0, 0, 1, 0}
+ /* COND_JUMP86 states. */
+ {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG)},
+ {127 + 1, -128 + 1, 1, ENCODE_RELAX_STATE (COND_JUMP86, BIG16)},
+ /* dword conditionals adds 5 bytes to frag:
+ 1 extra opcode byte, 4 displacement bytes. */
+ {0, 0, 5, 0},
+ /* word conditionals add 4 bytes to frag:
+ 1 displacement byte and a 3 byte long branch insn. */
+ {0, 0, 4, 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 },
+ {"pentium4", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuP4|CpuMMX|CpuSSE|CpuSSE2 },
+ {"k6", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|CpuK6|CpuMMX|Cpu3dnow },
+ {"athlon", Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuMMX|Cpu3dnow },
+ {"sledgehammer",Cpu086|Cpu186|Cpu286|Cpu386|Cpu486|Cpu586|Cpu686|CpuK6|CpuAthlon|CpuSledgehammer|CpuMMX|Cpu3dnow|CpuSSE|CpuSSE2 },
+ {NULL, 0 }
+};
+
+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'},
+ {"tfloat", float_cons, 'x'},
+ {"value", cons, 2},
+ {"noopt", s_ignore, 0},
+ {"optim", s_ignore, 0},
+ {"code16gcc", set_16bit_gcc_code_flag, CODE_16BIT},
+ {"code16", set_code_flag, CODE_16BIT},
+ {"code32", set_code_flag, CODE_32BIT},
+ {"code64", set_code_flag, CODE_64BIT},
+ {"intel_syntax", set_intel_syntax, 1},
+ {"att_syntax", set_intel_syntax, 0},
+ {"file", dwarf2_directive_file, 0},
+ {"loc", dwarf2_directive_loc, 0},
+ {0, 0, 0}
};
+/* For interface with expression (). */
+extern char *input_line_pointer;
+
+/* Hash table for instruction mnemonic lookup. */
+static struct hash_control *op_hash;
+/* Hash table for register lookup. */
+static struct hash_control *reg_hash;
+\f
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 */
+ /* Various efficient no-op patterns for aligning code labels.
+ Note: Don't try to assemble the instructions in the comments.
+ 0L and 0w are not legal. */
static const char f32_1[] =
{0x90}; /* nop */
static const char f32_2[] =
f32_15, f32_15, f32_15, f32_15, f32_15, f32_15, f32_15
};
+ /* ??? We can't use these fillers for x86_64, since they often kills the
+ upper halves. Solve later. */
+ if (flag_code == CODE_64BIT)
+ count = 1;
+
if (count > 0 && count <= 15)
{
- if (flag_16bit_code)
+ if (flag_code == CODE_16BIT)
{
- memcpy(fragP->fr_literal + fragP->fr_fix,
- f16_patt[count - 1], count);
- if (count > 8) /* adjust jump offset */
+ 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);
+ 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 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 ((int));
-#endif
-
-symbolS *GOT_symbol; /* Pre-defined "_GLOBAL_OFFSET_TABLE_" */
-
static INLINE unsigned int
mode_from_disp_size (t)
unsigned int t;
{
- return (t & Disp8) ? 1 : (t & (Disp16|Disp32)) ? 2 : 0;
+ return (t & Disp8) ? 1 : (t & (Disp16 | Disp32 | Disp32S)) ? 2 : 0;
}
static INLINE int
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 INLINE int
+fits_in_signed_long (num)
+ offsetT num ATTRIBUTE_UNUSED;
+{
+#ifndef BFD64
+ return 1;
+#else
+ return (!(((offsetT) -1 << 31) & num)
+ || (((offsetT) -1 << 31) & num) == ((offsetT) -1 << 31));
+#endif
+} /* fits_in_signed_long() */
+static INLINE int
+fits_in_unsigned_long (num)
+ offsetT num ATTRIBUTE_UNUSED;
+{
+#ifndef BFD64
+ return 1;
+#else
+ return (num & (((offsetT) 2 << 31) - 1)) == num;
+#endif
+} /* fits_in_unsigned_long() */
static int
smallest_imm_type (num)
- long num;
+ offsetT num;
{
-#if 0
- /* This code is disabled because all the Imm1 forms in the opcode table
- are slower on the i486, and they're the versions with the implicitly
- specified single-position displacement, which has another syntax if
- you really want to use that form. If you really prefer to have the
- one-byte-shorter Imm1 form despite these problems, re-enable this
- code. */
- if (num == 1)
- return Imm1 | Imm8 | Imm8S | Imm16 | Imm32;
-#endif
+ if (cpu_arch_flags != (Cpu086 | Cpu186 | Cpu286 | Cpu386 | Cpu486 | CpuNo64))
+ {
+ /* 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 | Imm32S | Imm64;
+ }
return (fits_in_signed_byte (num)
- ? (Imm8S | Imm8 | Imm16 | Imm32)
+ ? (Imm8S | Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
: fits_in_unsigned_byte (num)
- ? (Imm8 | Imm16 | Imm32)
+ ? (Imm8 | Imm16 | Imm32 | Imm32S | Imm64)
: (fits_in_signed_word (num) || fits_in_unsigned_word (num))
- ? (Imm16 | Imm32)
- : (Imm32));
-} /* smallest_imm_type() */
+ ? (Imm16 | Imm32 | Imm32S | Imm64)
+ : fits_in_signed_long (num)
+ ? (Imm32 | Imm32S | Imm64)
+ : fits_in_unsigned_long (num)
+ ? (Imm32 | Imm64)
+ : Imm64);
+}
+
+static offsetT
+offset_in_range (val, size)
+ offsetT val;
+ int size;
+{
+ 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;
+#ifdef BFD64
+ case 8: mask = ((addressT) 2 << 63) - 1; break;
+#endif
+ default: abort ();
+ }
+
+ /* If BFD64, sign extend val. */
+ if (!use_rela_relocations)
+ if ((val & ~(((addressT) 2 << 31) - 1)) == 0)
+ val = (val ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
+
+ if ((val & ~mask) != 0 && (val & ~mask) != ~mask)
+ {
+ 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;
+}
/* 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
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 (prefix >= REX_OPCODE && prefix < REX_OPCODE + 16
+ && flag_code == CODE_64BIT)
+ q = REX_PREFIX;
+ else
+ switch (prefix)
+ {
+ default:
+ abort ();
+
+ case CS_PREFIX_OPCODE:
+ case DS_PREFIX_OPCODE:
+ case ES_PREFIX_OPCODE:
+ case FS_PREFIX_OPCODE:
+ case GS_PREFIX_OPCODE:
+ case SS_PREFIX_OPCODE:
+ q = SEG_PREFIX;
+ break;
+
+ case REPNE_PREFIX_OPCODE:
+ case REPE_PREFIX_OPCODE:
+ ret = 2;
+ /* fall thru */
+ case LOCK_PREFIX_OPCODE:
+ q = LOCKREP_PREFIX;
+ break;
+
+ case 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])
+ if (i.prefix[q] != 0)
{
as_bad (_("same type of prefix used twice"));
return 0;
}
static void
-set_16bit_code_flag (new_16bit_code_flag)
- int new_16bit_code_flag;
+set_code_flag (value)
+ int value;
{
- flag_16bit_code = new_16bit_code_flag;
+ flag_code = value;
+ cpu_arch_flags &= ~(Cpu64 | CpuNo64);
+ cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
+ if (value == CODE_64BIT && !(cpu_arch_flags & CpuSledgehammer))
+ {
+ as_bad (_("64bit mode not supported on this CPU."));
+ }
+ if (value == CODE_32BIT && !(cpu_arch_flags & Cpu386))
+ {
+ as_bad (_("32bit mode not supported on this CPU."));
+ }
stackop_size = '\0';
}
static void
-set_16bit_gcc_code_flag (new_16bit_code_flag)
- int new_16bit_code_flag;
+set_16bit_gcc_code_flag (new_code_flag)
+ int new_code_flag;
{
- flag_16bit_code = new_16bit_code_flag;
- stackop_size = new_16bit_code_flag ? 'l' : '\0';
+ flag_code = new_code_flag;
+ cpu_arch_flags &= ~(Cpu64 | CpuNo64);
+ cpu_arch_flags |= (flag_code == CODE_64BIT ? Cpu64 : CpuNo64);
+ stackop_size = 'l';
}
static void
int ask_naked_reg = 0;
SKIP_WHITESPACE ();
- if (! is_end_of_line[(unsigned char) *input_line_pointer])
+ 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)
+ if (strcmp (string, "prefix") == 0)
ask_naked_reg = 1;
- else if (strcmp(string, "noprefix") == 0)
+ else if (strcmp (string, "noprefix") == 0)
ask_naked_reg = -1;
else
- as_bad (_("Bad argument to syntax directive."));
+ as_bad (_("bad argument to syntax directive."));
*input_line_pointer = e;
}
demand_empty_rest_of_line ();
allow_naked_reg = (intel_syntax
&& (bfd_get_symbol_leading_char (stdoutput) != '\0'));
#else
- allow_naked_reg = 0; /* conservative default */
+ /* Conservative default. */
+ allow_naked_reg = 0;
#endif
}
else
allow_naked_reg = (ask_naked_reg < 0);
}
-const pseudo_typeS md_pseudo_table[] =
+static void
+set_cpu_arch (dummy)
+ int dummy ATTRIBUTE_UNUSED;
{
-#ifndef I386COFF
- {"bss", s_bss, 0},
-#endif
-#if !defined(OBJ_AOUT) && !defined(USE_ALIGN_PTWO)
- {"align", s_align_bytes, 0},
-#else
- {"align", s_align_ptwo, 0},
-#endif
- {"ffloat", float_cons, 'f'},
- {"dfloat", float_cons, 'd'},
- {"tfloat", float_cons, 'x'},
- {"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},
- {0, 0, 0}
-};
+ SKIP_WHITESPACE ();
-/* for interface with expression () */
-extern char *input_line_pointer;
+ if (!is_end_of_line[(unsigned char) *input_line_pointer])
+ {
+ char *string = input_line_pointer;
+ int e = get_symbol_end ();
+ int i;
-/* hash table for instruction mnemonic lookup */
-static struct hash_control *op_hash;
-/* hash table for register lookup */
-static struct hash_control *reg_hash;
-\f
+ 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
+ | (flag_code == CODE_64BIT ? Cpu64 : CpuNo64));
+ break;
+ }
+ }
+ if (!cpu_arch[i].name)
+ as_bad (_("no such architecture: `%s'"), string);
+
+ *input_line_pointer = e;
+ }
+ else
+ as_bad (_("missing cpu architecture"));
+
+ no_cond_jump_promotion = 0;
+ if (*input_line_pointer == ','
+ && !is_end_of_line[(unsigned char) input_line_pointer[1]])
+ {
+ char *string = ++input_line_pointer;
+ int e = get_symbol_end ();
+
+ if (strcmp (string, "nojumps") == 0)
+ no_cond_jump_promotion = 1;
+ else if (strcmp (string, "jumps") == 0)
+ ;
+ else
+ as_bad (_("no such architecture modifier: `%s'"), string);
+
+ *input_line_pointer = e;
+ }
+
+ demand_empty_rest_of_line ();
+}
+#ifdef BFD_ASSEMBLER
+unsigned long
+i386_mach ()
+{
+ if (!strcmp (default_arch, "x86_64"))
+ return bfd_mach_x86_64;
+ else if (!strcmp (default_arch, "i386"))
+ return bfd_mach_i386_i386;
+ else
+ as_fatal (_("Unknown architecture"));
+}
+#endif
+\f
void
md_begin ()
{
const char *hash_err;
- /* initialize op_hash hash table */
+ /* Initialize op_hash hash table. */
op_hash = hash_new ();
{
- register const template *optab;
- register templates *core_optab;
+ const template *optab;
+ templates *core_optab;
- optab = i386_optab; /* setup for loop */
+ /* Setup for loop. */
+ optab = i386_optab;
core_optab = (templates *) xmalloc (sizeof (templates));
core_optab->start = optab;
|| strcmp (optab->name, (optab - 1)->name) != 0)
{
/* different name --> ship out current template list;
- add to hash table; & begin anew */
+ 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"),
(optab - 1)->name,
hash_err);
}
}
- /* initialize reg_hash hash table */
+ /* Initialize reg_hash hash table. */
reg_hash = hash_new ();
{
- register const reg_entry *regtab;
+ const reg_entry *regtab;
for (regtab = i386_regtab;
regtab < i386_regtab + sizeof (i386_regtab) / sizeof (i386_regtab[0]);
{
hash_err = hash_insert (reg_hash, regtab->reg_name, (PTR) regtab);
if (hash_err)
- goto hash_error;
+ as_fatal (_("Internal Error: Can't hash %s: %s"),
+ regtab->reg_name,
+ hash_err);
}
}
- /* fill in lexical tables: mnemonic_chars, operand_chars. */
+ /* Fill in lexical tables: mnemonic_chars, operand_chars. */
{
- register int c;
- register char *p;
+ int c;
+ char *p;
for (c = 0; c < 256; c++)
{
- if (isdigit (c))
+ if (ISDIGIT (c))
{
digit_chars[c] = c;
mnemonic_chars[c] = c;
register_chars[c] = c;
operand_chars[c] = c;
}
- else if (islower (c))
+ else if (ISLOWER (c))
{
mnemonic_chars[c] = c;
register_chars[c] = c;
operand_chars[c] = c;
}
- else if (isupper (c))
+ else if (ISUPPER (c))
{
- mnemonic_chars[c] = tolower (c);
+ mnemonic_chars[c] = TOLOWER (c);
register_chars[c] = mnemonic_chars[c];
operand_chars[c] = c;
}
- if (isalpha (c) || isdigit (c))
+ if (ISALPHA (c) || ISDIGIT (c))
identifier_chars[c] = c;
else if (c >= 128)
{
hash_print_statistics (file, "i386 register", reg_hash);
}
\f
-
#ifdef DEBUG386
-/* debugging routines for md_assemble */
+/* Debugging routines for md_assemble. */
static void pi PARAMS ((char *, i386_insn *));
static void pte PARAMS ((template *));
static void pt PARAMS ((unsigned int));
char *line;
i386_insn *x;
{
- register template *p;
- int i;
+ unsigned int i;
fprintf (stdout, "%s: template ", line);
pte (&x->tm);
- fprintf (stdout, " modrm: mode %x reg %x reg/mem %x",
+ fprintf (stdout, " address: base %s index %s scale %x\n",
+ x->base_reg ? x->base_reg->reg_name : "none",
+ x->index_reg ? x->index_reg->reg_name : "none",
+ x->log2_scale_factor);
+ fprintf (stdout, " modrm: mode %x reg %x reg/mem %x\n",
x->rm.mode, x->rm.reg, x->rm.regmem);
- fprintf (stdout, " base %x index %x scale %x\n",
- x->bi.base, x->bi.index, x->bi.scale);
+ fprintf (stdout, " sib: base %x index %x scale %x\n",
+ x->sib.base, x->sib.index, x->sib.scale);
+ fprintf (stdout, " rex: 64bit %x extX %x extY %x extZ %x\n",
+ (x->rex & REX_MODE64) != 0,
+ (x->rex & REX_EXTX) != 0,
+ (x->rex & REX_EXTY) != 0,
+ (x->rex & REX_EXTZ) != 0);
for (i = 0; i < x->operands; i++)
{
fprintf (stdout, " #%d: ", i + 1);
fprintf (stdout, "\n");
if (x->types[i]
& (Reg | SReg2 | SReg3 | Control | Debug | Test | RegMMX | RegXMM))
- fprintf (stdout, "%s\n", x->regs[i]->reg_name);
+ fprintf (stdout, "%s\n", x->op[i].regs->reg_name);
if (x->types[i] & Imm)
- pe (x->imms[i]);
+ pe (x->op[i].imms);
if (x->types[i] & Disp)
- pe (x->disps[i]);
+ pe (x->op[i].disps);
}
}
pte (t)
template *t;
{
- int i;
+ unsigned 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)
char *tname;
}
-type_names[] =
+static const type_names[] =
{
{ Reg8, "r8" },
{ Reg16, "r16" },
{ Reg32, "r32" },
+ { Reg64, "r64" },
{ Imm8, "i8" },
{ Imm8S, "i8s" },
{ Imm16, "i16" },
{ Imm32, "i32" },
+ { Imm32S, "i32s" },
+ { Imm64, "i64" },
{ Imm1, "i1" },
{ BaseIndex, "BaseIndex" },
{ Disp8, "d8" },
{ Disp16, "d16" },
{ Disp32, "d32" },
+ { Disp32S, "d32s" },
+ { Disp64, "d64" },
{ InOutPortReg, "InOutPortReg" },
{ ShiftCount, "ShiftCount" },
{ Control, "control reg" },
pt (t)
unsigned int t;
{
- register struct type_name *ty;
+ const struct type_name *ty;
- if (t == Unknown)
- {
- fprintf (stdout, _("Unknown"));
- }
- else
- {
- for (ty = type_names; ty->mask; ty++)
- if (t & ty->mask)
- fprintf (stdout, "%s, ", ty->tname);
- }
+ for (ty = type_names; ty->mask; ty++)
+ if (t & ty->mask)
+ fprintf (stdout, "%s, ", ty->tname);
fflush (stdout);
}
return 1;
return 0;
#else
- /* For COFF */
+ /* For COFF. */
return fixp->fx_r_type == 7;
#endif
}
#ifdef BFD_ASSEMBLER
static bfd_reloc_code_real_type reloc
- PARAMS ((int, int, bfd_reloc_code_real_type));
+ PARAMS ((int, int, int, bfd_reloc_code_real_type));
static bfd_reloc_code_real_type
-reloc (size, pcrel, other)
+reloc (size, pcrel, sign, other)
int size;
int pcrel;
+ int sign;
bfd_reloc_code_real_type other;
{
- if (other != NO_RELOC) return other;
+ if (other != NO_RELOC)
+ return other;
if (pcrel)
{
+ if (!sign)
+ as_bad (_("There are no unsigned pc-relative relocations"));
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);
+ 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;
- }
- as_bad (_("Can not do %d byte relocation"), size);
+ if (sign)
+ switch (size)
+ {
+ case 4: return BFD_RELOC_X86_64_32S;
+ }
+ else
+ switch (size)
+ {
+ case 1: return BFD_RELOC_8;
+ case 2: return BFD_RELOC_16;
+ case 4: return BFD_RELOC_32;
+ case 8: return BFD_RELOC_64;
+ }
+ as_bad (_("can not do %s %d byte relocation"),
+ sign ? "signed" : "unsigned", size);
}
+ abort ();
return BFD_RELOC_NONE;
}
-/*
- * 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.
- */
+/* 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 (TE_PE)
+#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_EXTERN (fixP->fx_addsy))
- return 0;
- if (S_IS_WEAK (fixP->fx_addsy))
+ if (S_IS_EXTERNAL (fixP->fx_addsy)
+ || S_IS_WEAK (fixP->fx_addsy)
+ /* Don't adjust pc-relative references to merge sections in 64-bit
+ mode. */
+ || (use_rela_relocations
+ && (S_GET_SEGMENT (fixP->fx_addsy)->flags & SEC_MERGE) != 0
+ && fixP->fx_pcrel))
return 0;
#endif
- /* adjust_reloc_syms doesn't know about the GOT */
+ /* adjust_reloc_syms doesn't know about the GOT. */
if (fixP->fx_r_type == BFD_RELOC_386_GOTOFF
|| fixP->fx_r_type == BFD_RELOC_386_PLT32
|| fixP->fx_r_type == BFD_RELOC_386_GOT32
- || fixP->fx_r_type == BFD_RELOC_RVA
+ || fixP->fx_r_type == BFD_RELOC_386_TLS_GD
+ || fixP->fx_r_type == BFD_RELOC_386_TLS_LDM
+ || fixP->fx_r_type == BFD_RELOC_386_TLS_LDO_32
+ || fixP->fx_r_type == BFD_RELOC_386_TLS_IE_32
+ || fixP->fx_r_type == BFD_RELOC_386_TLS_LE_32
+ || fixP->fx_r_type == BFD_RELOC_386_TLS_LE
+ || fixP->fx_r_type == BFD_RELOC_X86_64_PLT32
+ || fixP->fx_r_type == BFD_RELOC_X86_64_GOT32
+ || fixP->fx_r_type == BFD_RELOC_X86_64_GOTPCREL
|| 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
+#define reloc(SIZE,PCREL,SIGN,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
+#define BFD_RELOC_386_TLS_GD 0
+#define BFD_RELOC_386_TLS_LDM 0
+#define BFD_RELOC_386_TLS_LDO_32 0
+#define BFD_RELOC_386_TLS_IE_32 0
+#define BFD_RELOC_386_TLS_LE_32 0
+#define BFD_RELOC_386_TLS_LE 0
+#define BFD_RELOC_X86_64_PLT32 0
+#define BFD_RELOC_X86_64_GOT32 0
+#define BFD_RELOC_X86_64_GOTPCREL 0
#endif
-static int
-intel_float_operand PARAMS ((char *mnemonic));
+static int intel_float_operand PARAMS ((const char *mnemonic));
static int
intel_float_operand (mnemonic)
- char *mnemonic;
+ const char *mnemonic;
{
- if (mnemonic[0] == 'f' && mnemonic[1] =='i')
- return 0;
+ if (mnemonic[0] == 'f' && mnemonic[1] == 'i')
+ return 2;
if (mnemonic[0] == 'f')
return 1;
md_assemble (line)
char *line;
{
- /* Points to template once we've found it. */
- const template *t;
-
- /* Count the size of the instruction generated. */
- int insn_size = 0;
-
int j;
-
char mnemonic[MAX_MNEM_SIZE];
- /* Initialize globals. */
+ /* Initialize globals. */
memset (&i, '\0', sizeof (i));
for (j = 0; j < MAX_OPERANDS; j++)
- i.disp_reloc[j] = NO_RELOC;
+ i.reloc[j] = NO_RELOC;
memset (disp_expressions, '\0', sizeof (disp_expressions));
memset (im_expressions, '\0', sizeof (im_expressions));
- save_stack_p = save_stack; /* reset stack pointer */
+ save_stack_p = save_stack;
/* First parse an instruction mnemonic & call i386_operand for the operands.
We assume that the scrubber has arranged it so that line[0] is the valid
- start of a (possibly prefixed) mnemonic. */
- {
- char *l = line;
- char *token_start = l;
- char *mnem_p;
+ start of a (possibly prefixed) mnemonic. */
- /* Non-zero if we found a prefix only acceptable with string insns. */
- const char *expecting_string_instruction = NULL;
+ line = parse_insn (line, mnemonic);
+ if (line == NULL)
+ return;
- while (1)
- {
- mnem_p = mnemonic;
- while ((*mnem_p = mnemonic_chars[(unsigned char) *l]) != 0)
- {
- mnem_p++;
- if (mnem_p >= mnemonic + sizeof (mnemonic))
- {
- as_bad (_("no such 386 instruction: `%s'"), token_start);
- return;
- }
- l++;
- }
- if (!is_space_char (*l)
- && *l != END_OF_INSN
- && *l != PREFIX_SEPARATOR)
- {
- as_bad (_("invalid character %s in mnemonic"),
- output_invalid (*l));
- return;
- }
- if (token_start == l)
- {
- if (*l == PREFIX_SEPARATOR)
- as_bad (_("expecting prefix; got nothing"));
- else
- as_bad (_("expecting mnemonic; got nothing"));
- return;
- }
+ line = parse_operands (line, mnemonic);
+ if (line == NULL)
+ return;
- /* Look up instruction (or prefix) via hash table. */
- current_templates = hash_find (op_hash, mnemonic);
+ /* Now we've parsed the mnemonic into a set of templates, and have the
+ operands at hand. */
+
+ /* 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)))
+ swap_operands ();
+
+ if (i.imm_operands)
+ optimize_imm ();
+
+ if (i.disp_operands)
+ optimize_disp ();
+
+ /* 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. */
- 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 (_("redundant %s prefix"),
- current_templates->start->name);
- return;
- }
- /* Add prefix, checking for repeated prefixes. */
- switch (add_prefix (current_templates->start->base_opcode))
- {
- case 0:
- return;
- case 2:
- expecting_string_instruction =
- current_templates->start->name;
- break;
- }
- /* Skip past PREFIX_SEPARATOR and reset token_start. */
- token_start = ++l;
- }
- else
- break;
- }
+ if (!match_template ())
+ return;
- if (!current_templates)
- {
- /* See if we can get a match by trimming off a suffix. */
- switch (mnem_p[-1])
- {
- 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;
+ /* 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;
- /* 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);
- return;
- }
- }
+ if (i.tm.opcode_modifier & FWait)
+ if (!add_prefix (FWAIT_OPCODE))
+ return;
- /* check for rep/repne without a string instruction */
- if (expecting_string_instruction
- && !(current_templates->start->opcode_modifier & IsString))
- {
- as_bad (_("expecting string instruction after `%s'"),
- expecting_string_instruction);
+ /* Check string instruction segment overrides. */
+ if ((i.tm.opcode_modifier & IsString) != 0 && i.mem_operands != 0)
+ {
+ if (!check_string ())
return;
- }
+ }
- /* There may be operands to parse. */
- if (*l != END_OF_INSN)
- {
- /* parse operands */
+ if (!process_suffix ())
+ return;
- /* 1 if operand is pending after ','. */
- unsigned int expecting_operand = 0;
+ /* Make still unresolved immediate matches conform to size of immediate
+ given in i.suffix. */
+ if (!finalize_imm ())
+ return;
- /* Non-zero if operand parens not balanced. */
- unsigned int paren_not_balanced;
+ if (i.types[0] & Imm1)
+ i.imm_operands = 0; /* kludge for shift insns. */
+ if (i.types[0] & ImplicitRegister)
+ i.reg_operands--;
+ if (i.types[1] & ImplicitRegister)
+ i.reg_operands--;
+ if (i.types[2] & ImplicitRegister)
+ i.reg_operands--;
- do
- {
- /* skip optional white space before operand */
- if (is_space_char (*l))
- ++l;
- if (!is_operand_char (*l) && *l != END_OF_INSN)
- {
- as_bad (_("invalid character %s before operand %d"),
- output_invalid (*l),
- i.operands + 1);
- return;
- }
- token_start = l; /* after white space */
- paren_not_balanced = 0;
- while (paren_not_balanced || *l != ',')
- {
- if (*l == END_OF_INSN)
- {
- if (paren_not_balanced)
- {
- 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) && !is_space_char (*l))
- {
- as_bad (_("invalid character %s in operand %d"),
- output_invalid (*l),
- i.operands + 1);
- return;
- }
- 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 */
- unsigned int operand_ok;
- this_operand = i.operands++;
- if (i.operands > MAX_OPERANDS)
- {
- as_bad (_("spurious operands; (%d operands/instruction max)"),
- MAX_OPERANDS);
- return;
- }
- /* now parse operand adding info to 'i' as we go along */
- END_STRING_AND_SAVE (l);
-
- 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); /* restore old contents */
- if (!operand_ok)
- return;
- }
- else
- {
- if (expecting_operand)
- {
- expecting_operand_after_comma:
- as_bad (_("expecting operand after ','; got nothing"));
- return;
- }
- if (*l == ',')
- {
- as_bad (_("expecting operand before ','; got nothing"));
- return;
- }
- }
+ if (i.tm.opcode_modifier & ImmExt)
+ {
+ /* 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. */
- /* now *l must be either ',' or END_OF_INSN */
- if (*l == ',')
- {
- if (*++l == END_OF_INSN)
- { /* 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 */
- }
- }
+ expressionS *exp;
- /* Now we've parsed the mnemonic into a set of templates, and have the
- operands at hand.
+ assert (i.imm_operands == 0 && i.operands <= 2 && 2 < MAX_OPERANDS);
- 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. */
+ 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;
+ }
-#define MATCH(overlap, given, template) \
- ((overlap & ~JumpAbsolute) \
- && ((given) & (BaseIndex|JumpAbsolute)) == ((overlap) & (BaseIndex|JumpAbsolute)))
+ /* For insns with operands there are more diddles to do to the opcode. */
+ if (i.operands)
+ {
+ if (!process_operands ())
+ 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);
+ }
- /* 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 )
+ /* Handle conversion of 'int $3' --> special int3 insn. */
+ if (i.tm.base_opcode == INT_OPCODE && i.op[0].imms->X_add_number == 3)
+ {
+ i.tm.base_opcode = INT3_OPCODE;
+ i.imm_operands = 0;
+ }
- {
- register unsigned int overlap0, overlap1;
- unsigned int overlap2;
- unsigned int found_reverse_match;
- int suffix_check;
-
- /* All intel opcodes have reversed operands except for BOUND and ENTER */
- if (intel_syntax
- && (strcmp (mnemonic, "enter") != 0)
- && (strcmp (mnemonic, "bound") != 0)
- && (strncmp (mnemonic, "fsub", 4) !=0)
- && (strncmp (mnemonic, "fdiv", 4) !=0))
- {
- const reg_entry *temp_reg = NULL;
- expressionS *temp_disp = NULL;
- expressionS *temp_imm = NULL;
- unsigned int temp_type;
- int xchg1 = 0;
- int xchg2 = 0;
-
- if (i.operands == 2)
- {
- xchg1 = 0;
- xchg2 = 1;
- }
- else if (i.operands == 3)
- {
- xchg1 = 0;
- xchg2 = 2;
- }
+ 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;
+ }
- if (i.operands > 1)
- {
- temp_type = i.types[xchg2];
- if (temp_type & (Reg | FloatReg))
- temp_reg = i.regs[xchg2];
- else if (temp_type & Imm)
- temp_imm = i.imms[xchg2];
- else if (temp_type & Disp)
- temp_disp = i.disps[xchg2];
+ if ((i.tm.opcode_modifier & Rex64) != 0)
+ i.rex |= REX_MODE64;
- i.types[xchg2] = i.types[xchg1];
+ /* For 8 bit registers we need an empty rex prefix. Also if the
+ instruction already has a prefix, we need to convert old
+ registers to new ones. */
- if (i.types[xchg1] & (Reg | FloatReg))
- {
- i.regs[xchg2] = i.regs[xchg1];
- i.regs[xchg1] = NULL;
- }
- else if (i.types[xchg2] & Imm)
- {
- i.imms[xchg2] = i.imms[xchg1];
- i.imms[xchg1] = NULL;
- }
- else if (i.types[xchg2] & Disp)
- {
- i.disps[xchg2] = i.disps[xchg1];
- i.disps[xchg1] = NULL;
- }
+ if (((i.types[0] & Reg8) != 0
+ && (i.op[0].regs->reg_flags & RegRex64) != 0)
+ || ((i.types[1] & Reg8) != 0
+ && (i.op[1].regs->reg_flags & RegRex64) != 0)
+ || (((i.types[0] & Reg8) != 0 || (i.types[1] & Reg8) != 0)
+ && i.rex != 0))
+ {
+ int x;
- if (temp_type & (Reg | FloatReg))
- {
- i.regs[xchg1] = temp_reg;
- if (! (i.types[xchg1] & (Reg | FloatReg)))
- i.regs[xchg2] = NULL;
- }
- else if (temp_type & Imm)
- {
- i.imms[xchg1] = temp_imm;
- if (! (i.types[xchg1] & Imm))
- i.imms[xchg2] = NULL;
- }
- else if (temp_type & Disp)
- {
- i.disps[xchg1] = temp_disp;
- if (! (i.types[xchg1] & Disp))
- i.disps[xchg2] = NULL;
- }
+ i.rex |= REX_OPCODE;
+ for (x = 0; x < 2; x++)
+ {
+ /* Look for 8 bit operand that uses old registers. */
+ if ((i.types[x] & Reg8) != 0
+ && (i.op[x].regs->reg_flags & RegRex64) == 0)
+ {
+ /* In case it is "hi" register, give up. */
+ if (i.op[x].regs->reg_num > 3)
+ as_bad (_("can't encode register '%%%s' in an instruction requiring REX prefix.\n"),
+ i.op[x].regs->reg_name);
- i.types[xchg1] = temp_type;
- }
- }
- 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))))));
-
- for (t = current_templates->start;
- t < current_templates->end;
- t++)
- {
- /* Must have right number of operands. */
- if (i.operands != t->operands)
- continue;
+ /* Otherwise it is equivalent to the extended register.
+ Since the encoding doesn't change this is merely
+ cosmetic cleanup for debug output. */
- /* For some opcodes, don't check the suffix */
- if (intel_syntax)
- {
- if (strcmp (t->name, "fnstcw")
- && strcmp (t->name, "fldcw")
- && (t->opcode_modifier & suffix_check))
- continue;
- }
- /* Must not have disallowed suffix. */
- else if ((t->opcode_modifier & suffix_check))
- continue;
+ i.op[x].regs = i.op[x].regs + 8;
+ }
+ }
+ }
- else if (!t->operands)
- break; /* 0 operands always matches */
+ if (i.rex != 0)
+ add_prefix (REX_OPCODE | i.rex);
- overlap0 = i.types[0] & t->operand_types[0];
- switch (t->operands)
- {
- case 1:
- 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], 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]))
- {
+ /* We are ready to output the insn. */
+ output_insn ();
+}
- /* check if other direction is valid ... */
- if ((t->opcode_modifier & (D|FloatD)) == 0)
- continue;
+static char *
+parse_insn (line, mnemonic)
+ char *line;
+ char *mnemonic;
+{
+ char *l = line;
+ char *token_start = l;
+ char *mnem_p;
- /* 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], 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 */
- continue;
- }
- /* 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], t->operand_types[2])
- || !CONSISTENT_REGISTER_MATCH (overlap1, i.types[1],
- t->operand_types[1],
- overlap2, i.types[2],
- t->operand_types[2]))
+ /* Non-zero if we found a prefix only acceptable with string insns. */
+ const char *expecting_string_instruction = NULL;
- continue;
- }
- /* 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 = ... */
- if (t == current_templates->end)
- { /* we found no match */
- as_bad (_("suffix or operands invalid for `%s'"),
- current_templates->start->name);
- return;
- }
+ while (1)
+ {
+ mnem_p = mnemonic;
+ while ((*mnem_p = mnemonic_chars[(unsigned char) *l]) != 0)
+ {
+ mnem_p++;
+ if (mnem_p >= mnemonic + MAX_MNEM_SIZE)
+ {
+ as_bad (_("no such instruction: `%s'"), token_start);
+ return NULL;
+ }
+ l++;
+ }
+ if (!is_space_char (*l)
+ && *l != END_OF_INSN
+ && *l != PREFIX_SEPARATOR
+ && *l != ',')
+ {
+ as_bad (_("invalid character %s in mnemonic"),
+ output_invalid (*l));
+ return NULL;
+ }
+ if (token_start == l)
+ {
+ if (*l == PREFIX_SEPARATOR)
+ as_bad (_("expecting prefix; got nothing"));
+ else
+ as_bad (_("expecting mnemonic; got nothing"));
+ return NULL;
+ }
- if (!intel_syntax
- && (i.types[0] & JumpAbsolute) != (t->operand_types[0] & JumpAbsolute))
- {
- as_warn (_("Indirect %s without `*'"), t->name);
- }
+ /* Look up instruction (or prefix) via hash table. */
+ current_templates = hash_find (op_hash, mnemonic);
- 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);
- }
+ 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))
+ && flag_code != CODE_64BIT
+ && (((current_templates->start->opcode_modifier & Size32) != 0)
+ ^ (flag_code == CODE_16BIT)))
+ {
+ as_bad (_("redundant %s prefix"),
+ current_templates->start->name);
+ return NULL;
+ }
+ /* Add prefix, checking for repeated prefixes. */
+ switch (add_prefix (current_templates->start->base_opcode))
+ {
+ case 0:
+ return NULL;
+ case 2:
+ expecting_string_instruction = current_templates->start->name;
+ break;
+ }
+ /* Skip past PREFIX_SEPARATOR and reset token_start. */
+ token_start = ++l;
+ }
+ else
+ break;
+ }
- /* Copy the template we found. */
- i.tm = *t;
- if (found_reverse_match)
- {
- i.tm.operand_types[0] = t->operand_types[1];
- i.tm.operand_types[1] = t->operand_types[0];
- }
+ if (!current_templates)
+ {
+ /* See if we can get a match by trimming off a suffix. */
+ switch (mnem_p[-1])
+ {
+ case WORD_MNEM_SUFFIX:
+ case BYTE_MNEM_SUFFIX:
+ case QWORD_MNEM_SUFFIX:
+ i.suffix = mnem_p[-1];
+ mnem_p[-1] = '\0';
+ current_templates = hash_find (op_hash, mnemonic);
+ break;
+ case SHORT_MNEM_SUFFIX:
+ case LONG_MNEM_SUFFIX:
+ if (!intel_syntax)
+ {
+ i.suffix = mnem_p[-1];
+ mnem_p[-1] = '\0';
+ current_templates = hash_find (op_hash, mnemonic);
+ }
+ break;
+ /* Intel Syntax. */
+ case 'd':
+ if (intel_syntax)
+ {
+ if (intel_float_operand (mnemonic))
+ i.suffix = SHORT_MNEM_SUFFIX;
+ else
+ i.suffix = LONG_MNEM_SUFFIX;
+ mnem_p[-1] = '\0';
+ current_templates = hash_find (op_hash, mnemonic);
+ }
+ break;
+ }
+ if (!current_templates)
+ {
+ as_bad (_("no such instruction: `%s'"), token_start);
+ return NULL;
+ }
+ }
- if (i.tm.opcode_modifier & FWait)
- if (! add_prefix (FWAIT_OPCODE))
- return;
+ if (current_templates->start->opcode_modifier & (Jump | JumpByte))
+ {
+ /* Check for a branch hint. We allow ",pt" and ",pn" for
+ predict taken and predict not taken respectively.
+ I'm not sure that branch hints actually do anything on loop
+ and jcxz insns (JumpByte) for current Pentium4 chips. They
+ may work in the future and it doesn't hurt to accept them
+ now. */
+ if (l[0] == ',' && l[1] == 'p')
+ {
+ if (l[2] == 't')
+ {
+ if (!add_prefix (DS_PREFIX_OPCODE))
+ return NULL;
+ l += 3;
+ }
+ else if (l[2] == 'n')
+ {
+ if (!add_prefix (CS_PREFIX_OPCODE))
+ return NULL;
+ l += 3;
+ }
+ }
+ }
+ /* Any other comma loses. */
+ if (*l == ',')
+ {
+ as_bad (_("invalid character %s in mnemonic"),
+ output_invalid (*l));
+ return NULL;
+ }
- /* Check string instruction segment overrides */
- if ((i.tm.opcode_modifier & IsString) != 0 && i.mem_operands != 0)
- {
- int mem_op = (i.types[0] & AnyMem) ? 0 : 1;
- if ((i.tm.operand_types[mem_op] & EsSeg) != 0)
+ /* Check if instruction is supported on specified architecture. */
+ if ((current_templates->start->cpu_flags & ~(Cpu64 | CpuNo64))
+ & ~(cpu_arch_flags & ~(Cpu64 | CpuNo64)))
+ {
+ as_warn (_("`%s' is not supported on `%s'"),
+ current_templates->start->name, cpu_arch_name);
+ }
+ else if ((Cpu386 & ~cpu_arch_flags) && (flag_code != CODE_16BIT))
+ {
+ 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 `%s'"),
+ expecting_string_instruction);
+ return NULL;
+ }
+
+ return l;
+}
+
+static char *
+parse_operands (l, mnemonic)
+ char *l;
+ const char *mnemonic;
+{
+ char *token_start;
+
+ /* 1 if operand is pending after ','. */
+ unsigned int expecting_operand = 0;
+
+ /* Non-zero if operand parens not balanced. */
+ unsigned int paren_not_balanced;
+
+ while (*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)
+ {
+ as_bad (_("invalid character %s before operand %d"),
+ output_invalid (*l),
+ i.operands + 1);
+ return NULL;
+ }
+ token_start = l; /* after white space */
+ paren_not_balanced = 0;
+ while (paren_not_balanced || *l != ',')
+ {
+ if (*l == END_OF_INSN)
+ {
+ if (paren_not_balanced)
+ {
+ 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 NULL;
+ }
+ else
+ break; /* we are done */
+ }
+ else if (!is_operand_char (*l) && !is_space_char (*l))
+ {
+ as_bad (_("invalid character %s in operand %d"),
+ output_invalid (*l),
+ i.operands + 1);
+ return NULL;
+ }
+ 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. */
+ unsigned int operand_ok;
+ this_operand = i.operands++;
+ if (i.operands > MAX_OPERANDS)
+ {
+ as_bad (_("spurious operands; (%d operands/instruction max)"),
+ MAX_OPERANDS);
+ return NULL;
+ }
+ /* Now parse operand adding info to 'i' as we go along. */
+ END_STRING_AND_SAVE (l);
+
+ 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 NULL;
+ }
+ else
+ {
+ if (expecting_operand)
+ {
+ expecting_operand_after_comma:
+ as_bad (_("expecting operand after ','; got nothing"));
+ return NULL;
+ }
+ if (*l == ',')
+ {
+ as_bad (_("expecting operand before ','; got nothing"));
+ return NULL;
+ }
+ }
+
+ /* Now *l must be either ',' or END_OF_INSN. */
+ if (*l == ',')
+ {
+ if (*++l == END_OF_INSN)
+ {
+ /* Just skip it, if it's \n complain. */
+ goto expecting_operand_after_comma;
+ }
+ expecting_operand = 1;
+ }
+ }
+ return l;
+}
+
+static void
+swap_operands ()
+{
+ union i386_op temp_op;
+ unsigned int temp_type;
+ RELOC_ENUM temp_reloc;
+ 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_reloc = i.reloc[xchg2];
+ i.reloc[xchg2] = i.reloc[xchg1];
+ i.reloc[xchg1] = temp_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;
+ }
+}
+
+/* Try to ensure constant immediates are represented in the smallest
+ opcode possible. */
+static void
+optimize_imm ()
+{
+ 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.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];
+ if (i.types[op] & Reg8)
+ guess_suffix = BYTE_MNEM_SUFFIX;
+ else if (i.types[op] & Reg16)
+ guess_suffix = WORD_MNEM_SUFFIX;
+ else if (i.types[op] & Reg32)
+ guess_suffix = LONG_MNEM_SUFFIX;
+ else if (i.types[op] & Reg64)
+ guess_suffix = QWORD_MNEM_SUFFIX;
+ break;
}
- else if ((i.tm.operand_types[mem_op + 1] & EsSeg) != 0)
+ }
+ else if ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0))
+ guess_suffix = WORD_MNEM_SUFFIX;
+
+ for (op = i.operands; --op >= 0;)
+ if (i.types[op] & Imm)
+ {
+ switch (i.op[op].imms->X_op)
{
- if (i.seg[1] != NULL && i.seg[1] != &es)
+ case O_constant:
+ /* If a suffix is given, this operand may be shortened. */
+ switch (guess_suffix)
{
- as_bad (_("`%s' operand %d must use `%%es' segment"),
- i.tm.name,
- mem_op + 2);
- return;
+ case LONG_MNEM_SUFFIX:
+ i.types[op] |= Imm32 | Imm64;
+ break;
+ case WORD_MNEM_SUFFIX:
+ i.types[op] |= Imm16 | Imm32S | Imm32 | Imm64;
+ break;
+ case BYTE_MNEM_SUFFIX:
+ i.types[op] |= Imm16 | Imm8 | Imm8S | Imm32S | Imm32 | Imm64;
+ break;
}
- }
- }
- /* If 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.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 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)
{
- /* 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;
+ i.op[op].imms->X_add_number =
+ (((i.op[op].imms->X_add_number & 0xffff) ^ 0x8000) - 0x8000);
+ }
+ if ((i.types[op] & Imm32)
+ && ((i.op[op].imms->X_add_number & ~(((offsetT) 2 << 31) - 1))
+ == 0))
+ {
+ i.op[op].imms->X_add_number = ((i.op[op].imms->X_add_number
+ ^ ((offsetT) 1 << 31))
+ - ((offsetT) 1 << 31));
+ }
+ i.types[op] |= smallest_imm_type (i.op[op].imms->X_add_number);
- /* 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;
+ /* We must avoid matching of Imm32 templates when 64bit
+ only immediate is available. */
+ if (guess_suffix == QWORD_MNEM_SUFFIX)
+ i.types[op] &= ~Imm32;
+ break;
- if ((i.types[op] & WordReg) && i.regs[op]->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 ((i.tm.operand_types[op] & InOutPortReg) == 0)
- as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
- (i.regs[op] - (i.types[op] & Reg16 ? 8 : 16))->reg_name,
- i.regs[op]->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.regs[op]->reg_name,
- i.tm.name,
- i.suffix);
- return;
- }
+ case O_absent:
+ case O_register:
+ abort ();
+
+ /* Symbols and expressions. */
+ default:
+ /* Convert symbolic operand to proper sizes for matching. */
+ switch (guess_suffix)
+ {
+ case QWORD_MNEM_SUFFIX:
+ i.types[op] = Imm64 | Imm32S;
+ break;
+ case LONG_MNEM_SUFFIX:
+ i.types[op] = Imm32 | Imm64;
+ break;
+ case WORD_MNEM_SUFFIX:
+ i.types[op] = Imm16 | Imm32 | Imm64;
+ break;
+ break;
+ case BYTE_MNEM_SUFFIX:
+ i.types[op] = Imm8 | Imm8S | Imm16 | Imm32S | Imm32;
+ break;
+ break;
}
+ break;
}
- 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.regs[op]->reg_name,
- i.tm.name,
- i.suffix);
- return;
- }
-#if REGISTER_WARNINGS
- /* Warn if the e prefix on a general reg is missing. */
- else if ((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.regs[op] + 8)->reg_name,
- i.regs[op]->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.regs[op]->reg_name,
- i.tm.name,
- i.suffix);
- return;
- }
-#if REGISTER_WARNINGS
- /* Warn if the e prefix on a general reg is present. */
- else if ((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.regs[op] - 8)->reg_name,
- i.regs[op]->reg_name,
- i.suffix);
- }
-#endif
- }
- 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! */
- if ((overlap0 & (Imm8 | Imm8S | Imm16 | Imm32))
- && overlap0 != Imm8 && overlap0 != Imm8S
- && overlap0 != Imm16 && overlap0 != Imm32)
- {
- if (i.suffix)
- {
- overlap0 &= (i.suffix == BYTE_MNEM_SUFFIX ? (Imm8 | Imm8S) :
- (i.suffix == WORD_MNEM_SUFFIX ? Imm16 : Imm32));
- }
- else if (overlap0 == (Imm16 | Imm32))
- {
- 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;
- }
- }
- if ((overlap1 & (Imm8 | Imm8S | Imm16 | Imm32))
- && overlap1 != Imm8 && overlap1 != Imm8S
- && overlap1 != Imm16 && overlap1 != Imm32)
+/* Try to use the smallest displacement type too. */
+static void
+optimize_disp ()
+{
+ int op;
+
+ for (op = i.operands; --op >= 0;)
+ if ((i.types[op] & Disp) && i.op[op].disps->X_op == O_constant)
{
- if (i.suffix)
+ offsetT disp = i.op[op].disps->X_add_number;
+
+ if (i.types[op] & Disp16)
{
- overlap1 &= (i.suffix == BYTE_MNEM_SUFFIX ? (Imm8 | Imm8S) :
- (i.suffix == WORD_MNEM_SUFFIX ? Imm16 : Imm32));
+ /* 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);
}
- else if (overlap1 == (Imm16 | Imm32))
+ else if (i.types[op] & Disp32)
{
- overlap1 =
- (flag_16bit_code ^ (i.prefix[DATA_PREFIX] != 0)) ? Imm16 : Imm32;
+ /* We know this operand is at most 32 bits, so convert to a
+ signed 32 bit number before trying to see whether it will
+ fit in an even smaller size. */
+ disp &= (((offsetT) 2 << 31) - 1);
+ disp = (disp ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31);
}
- else
+ if (flag_code == CODE_64BIT)
{
- as_bad (_("no instruction mnemonic suffix given; can't determine immediate size"));
- return;
+ if (fits_in_signed_long (disp))
+ i.types[op] |= Disp32S;
+ if (fits_in_unsigned_long (disp))
+ i.types[op] |= Disp32;
}
+ if ((i.types[op] & (Disp32 | Disp32S | Disp16))
+ && fits_in_signed_byte (disp))
+ i.types[op] |= Disp8;
}
- assert ((overlap2 & Imm) == 0);
+}
- i.types[0] = overlap0;
- if (overlap0 & ImplicitRegister)
- i.reg_operands--;
- if (overlap0 & Imm1)
- i.imm_operands = 0; /* kludge for shift insns */
+static int
+match_template ()
+{
+ /* Points to template once we've found it. */
+ const template *t;
+ unsigned int overlap0, overlap1, overlap2;
+ unsigned int found_reverse_match;
+ int suffix_check;
- i.types[1] = overlap1;
- if (overlap1 & ImplicitRegister)
- i.reg_operands--;
+#define MATCH(overlap, given, template) \
+ ((overlap & ~JumpAbsolute) \
+ && (((given) & (BaseIndex | JumpAbsolute)) \
+ == ((overlap) & (BaseIndex | JumpAbsolute))))
- i.types[2] = overlap2;
- if (overlap2 & ImplicitRegister)
- i.reg_operands--;
+ /* 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 )
+
+ 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 == QWORD_MNEM_SUFFIX
+ ? No_qSuf
+ : (i.suffix == LONG_DOUBLE_MNEM_SUFFIX
+ ? No_xSuf : 0))))));
+
+ for (t = current_templates->start;
+ t < current_templates->end;
+ t++)
+ {
+ /* 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;
+
+ /* Do not verify operands when there are none. */
+ else if (!t->operands)
+ {
+ if (t->cpu_flags & ~cpu_arch_flags)
+ continue;
+ /* We've found a match; break out of loop. */
+ break;
+ }
- /* Finalize opcode. First, we change the opcode based on the operand
- size given by i.suffix: We need not change things for byte insns. */
+ overlap0 = i.types[0] & t->operand_types[0];
+ switch (t->operands)
+ {
+ case 1:
+ 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], 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 & (D | FloatD)) == 0)
+ continue;
+
+ /* 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], 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. */
+ continue;
+ }
+ /* found_reverse_match holds which of D or FloatDR
+ we've found. */
+ found_reverse_match = t->opcode_modifier & (D | FloatDR);
+ }
+ /* Found a forward 2 operand match here. */
+ else 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], 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. */
+ }
+ if (t->cpu_flags & ~cpu_arch_flags)
+ {
+ found_reverse_match = 0;
+ continue;
+ }
+ /* We've found a match; break out of loop. */
+ break;
+ }
- if (!i.suffix && (i.tm.opcode_modifier & W))
- {
- as_bad (_("no instruction mnemonic suffix given and no register operands; can't size instruction"));
- return;
- }
+ if (t == current_templates->end)
+ {
+ /* We found no match. */
+ as_bad (_("suffix or operands invalid for `%s'"),
+ current_templates->start->name);
+ return 0;
+ }
- /* 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)
+ if (!quiet_warnings)
+ {
+ if (!intel_syntax
+ && ((i.types[0] & JumpAbsolute)
+ != (t->operand_types[0] & JumpAbsolute)))
{
- unsigned int prefix = DATA_PREFIX_OPCODE;
+ as_warn (_("indirect %s without `*'"), t->name);
+ }
- if ((i.regs[1]->reg_type & Reg16) != 0)
- if (!add_prefix (prefix))
- return;
+ 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);
}
+ }
- if (i.suffix && i.suffix != BYTE_MNEM_SUFFIX)
- {
- /* It's not a byte, select word/dword operation. */
- if (i.tm.opcode_modifier & W)
- {
- 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;
+ /* 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). */
- 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;
- }
- }
+ i.tm.base_opcode ^= found_reverse_match;
- if (i.tm.opcode_modifier & ImmExt)
- {
- /* 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. */
+ i.tm.operand_types[0] = t->operand_types[1];
+ i.tm.operand_types[1] = t->operand_types[0];
+ }
- expressionS *exp;
+ return 1;
+}
- assert(i.imm_operands == 0 && i.operands <= 2);
+static int
+check_string ()
+{
+ 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 0;
+ }
+ /* 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 0;
+ }
+ }
+ return 1;
+}
- exp = &im_expressions[i.imm_operands++];
- i.imms[i.operands] = exp;
- i.types[i.operands++] = Imm8;
- exp->X_op = O_constant;
- exp->X_add_number = i.tm.extension_opcode;
- i.tm.extension_opcode = None;
- }
+static int
+process_suffix ()
+{
+ /* If matched instruction specifies an explicit instruction mnemonic
+ suffix, use it. */
+ if (i.tm.opcode_modifier & (Size16 | Size32 | Size64))
+ {
+ if (i.tm.opcode_modifier & Size16)
+ i.suffix = WORD_MNEM_SUFFIX;
+ else if (i.tm.opcode_modifier & Size64)
+ i.suffix = QWORD_MNEM_SUFFIX;
+ else
+ i.suffix = LONG_MNEM_SUFFIX;
+ }
+ 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 :
+ (i.types[op] & Reg64) ? QWORD_MNEM_SUFFIX :
+ LONG_MNEM_SUFFIX);
+ break;
+ }
+ }
+ else if (i.suffix == BYTE_MNEM_SUFFIX)
+ {
+ if (!check_byte_reg ())
+ return 0;
+ }
+ else if (i.suffix == LONG_MNEM_SUFFIX)
+ {
+ if (!check_long_reg ())
+ return 0;
+ }
+ else if (i.suffix == QWORD_MNEM_SUFFIX)
+ {
+ if (!check_qword_reg ())
+ return 0;
+ }
+ else if (i.suffix == WORD_MNEM_SUFFIX)
+ {
+ if (!check_word_reg ())
+ return 0;
+ }
+ 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;
+ }
- /* 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;
+ /* Change the opcode based on the operand size given by i.suffix;
+ We need not change things for byte insns. */
- /* 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). */
+ if (!i.suffix && (i.tm.opcode_modifier & W))
+ {
+ as_bad (_("no instruction mnemonic suffix given and no register operands; can't size instruction"));
+ return 0;
+ }
- i.tm.base_opcode ^= found_reverse_match;
+ /* For movzx and movsx, need to check the register type. */
+ if (intel_syntax
+ && (i.tm.base_opcode == 0xfb6 || i.tm.base_opcode == 0xfbe)
+ && i.suffix == BYTE_MNEM_SUFFIX)
+ {
+ unsigned int prefix = DATA_PREFIX_OPCODE;
- /* The imul $imm, %reg instruction is converted into
- imul $imm, %reg, %reg, and the clr %reg instruction
- is converted into xor %reg, %reg. */
- if (i.tm.opcode_modifier & regKludge)
- {
- unsigned int first_reg_op = (i.types[0] & Reg) ? 0 : 1;
- /* Pretend we saw the extra register operand. */
- i.regs[first_reg_op+1] = i.regs[first_reg_op];
- i.reg_operands = 2;
- }
+ if ((i.op[1].regs->reg_type & Reg16) != 0)
+ if (!add_prefix (prefix))
+ return 0;
+ }
- if (i.tm.opcode_modifier & ShortForm)
- {
- /* 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.regs[op]->reg_num;
- if ((i.tm.opcode_modifier & Ugh) != 0)
- {
- /* 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.regs[1]->reg_name,
- i.regs[0]->reg_name);
- }
- else
- {
- /* extraneous `l' suffix on fp insn */
- as_warn (_("translating to `%s %%%s'"), i.tm.name,
- i.regs[0]->reg_name);
- }
- }
- }
- else if (i.tm.opcode_modifier & Modrm)
- {
- /* 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. */
- if (i.reg_operands == 2)
- {
- unsigned int source, dest;
- source = ((i.types[0]
- & (Reg | RegMMX | RegXMM
- | SReg2 | SReg3
- | Control | Debug | Test))
- ? 0 : 1);
- dest = source + 1;
-
- i.rm.mode = 3;
- /* 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;
- }
- else
- {
- i.rm.reg = i.regs[source]->reg_num;
- i.rm.regmem = i.regs[dest]->reg_num;
- }
- }
- else
- { /* if it's not 2 reg operands... */
- if (i.mem_operands)
- {
- unsigned int fake_zero_displacement = 0;
- unsigned int op = ((i.types[0] & AnyMem)
- ? 0
- : (i.types[1] & AnyMem) ? 1 : 2);
-
- default_seg = &ds;
-
- if (! i.base_reg)
- {
- i.rm.mode = 0;
- if (! i.disp_operands)
- fake_zero_displacement = 1;
- if (! i.index_reg)
- {
- /* 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 /* ! i.base_reg && i.index_reg */
- {
- 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.base_reg->reg_type & Reg16)
- {
- 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 /* i.base_reg and 32 bit mode */
- {
- i.rm.regmem = i.base_reg->reg_num;
- i.sib.base = i.base_reg->reg_num;
- if (i.base_reg->reg_num == EBP_REG_NUM)
- {
- default_seg = &ss;
- if (i.disp_operands == 0)
- {
- fake_zero_displacement = 1;
- i.types[op] |= Disp8;
- }
- }
- else if (i.base_reg->reg_num == ESP_REG_NUM)
- {
- default_seg = &ss;
- }
- i.sib.scale = i.log2_scale_factor;
- if (! i.index_reg)
- {
- /* <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
- }
- else
- {
- 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 (i.suffix && i.suffix != BYTE_MNEM_SUFFIX)
+ {
+ /* It's not a byte, select word/dword operation. */
+ if (i.tm.opcode_modifier & W)
+ {
+ if (i.tm.opcode_modifier & ShortForm)
+ i.tm.base_opcode |= 8;
+ else
+ i.tm.base_opcode |= 1;
+ }
- if (fake_zero_displacement)
- {
- /* Fakes a zero displacement assuming that i.types[op]
- holds the correct displacement size. */
- expressionS *exp;
-
- exp = &disp_expressions[i.disp_operands++];
- i.disps[op] = 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 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 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 (i.tm.extension_opcode != None)
- i.rm.regmem = i.regs[op]->reg_num;
- else
- i.rm.reg = i.regs[op]->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 in the regmem field. */
- if (!i.mem_operands)
- i.rm.mode = 3;
- }
-
- /* 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.regs[0]->reg_num == 1)
- {
- as_bad (_("you can't `pop %%cs'"));
- return;
- }
- i.tm.base_opcode |= (i.regs[0]->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;
- }
+ /* Now select between word & dword operations via the operand
+ size prefix, except for instructions that will ignore this
+ prefix anyway. */
+ if (i.suffix != QWORD_MNEM_SUFFIX
+ && (i.suffix == LONG_MNEM_SUFFIX) == (flag_code == CODE_16BIT)
+ && !(i.tm.opcode_modifier & IgnoreSize))
+ {
+ unsigned int prefix = DATA_PREFIX_OPCODE;
+ if (i.tm.opcode_modifier & JumpByte) /* jcxz, loop */
+ prefix = ADDR_PREFIX_OPCODE;
- /* 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 ((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);
- }
- }
+ if (!add_prefix (prefix))
+ return 0;
+ }
- /* Handle conversion of 'int $3' --> special int3 insn. */
- if (i.tm.base_opcode == INT_OPCODE && i.imms[0]->X_add_number == 3)
- {
- i.tm.base_opcode = INT3_OPCODE;
- i.imm_operands = 0;
+ if (i.suffix != QWORD_MNEM_SUFFIX && (flag_code == CODE_64BIT)
+ && !(i.tm.opcode_modifier & IgnoreSize)
+ && (i.tm.opcode_modifier & JumpByte))
+ {
+ if (!add_prefix (ADDR_PREFIX_OPCODE))
+ return 0;
+ }
+
+ /* Set mode64 for an operand. */
+ if (i.suffix == QWORD_MNEM_SUFFIX
+ && (i.tm.opcode_modifier & NoRex64) == 0)
+ {
+ i.rex |= REX_MODE64;
+ if (flag_code < CODE_64BIT)
+ {
+ as_bad (_("64bit operations available only in 64bit modes."));
+ return 0;
+ }
+ }
+
+ /* Size floating point instruction. */
+ if (i.suffix == LONG_MNEM_SUFFIX)
+ {
+ if (i.tm.opcode_modifier & FloatMF)
+ i.tm.base_opcode ^= 4;
+ }
}
- if ((i.tm.opcode_modifier & (Jump | JumpByte | JumpDword))
- && i.disps[0]->X_op == O_constant)
+ return 1;
+}
+
+static int
+check_byte_reg ()
+{
+ int op;
+ for (op = i.operands; --op >= 0;)
{
- /* 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.disps[0]->X_add_symbol = &abs_symbol;
- i.disps[0]->X_op = O_symbol;
+ /* 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 == 0x63
+ || 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
+ )
+ {
+ /* Prohibit these changes in the 64bit mode, since the
+ lowering is more complicated. */
+ if (flag_code == CODE_64BIT
+ && (i.tm.operand_types[op] & InOutPortReg) == 0)
+ {
+ as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
+ i.op[op].regs->reg_name,
+ i.suffix);
+ return 0;
+ }
+#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
+ ? REGNAM_AL - REGNAM_AX
+ : REGNAM_AL - REGNAM_EAX))->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 0;
+ }
}
+ return 1;
+}
- /* We are ready to output the insn. */
- {
- register char *p;
+static int
+check_long_reg ()
+{
+ int op;
- /* Output jumps. */
- if (i.tm.opcode_modifier & Jump)
+ 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)
{
- int size;
- int code16;
- int prefix;
-
- code16 = 0;
- if (flag_16bit_code)
- code16 = CODE16;
-
- prefix = 0;
- if (i.prefix[DATA_PREFIX])
- {
- 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 fr_var. */
- frag_var (rs_machine_dependent,
- 1 + size,
- 1,
- ((unsigned char) *p == JUMP_PC_RELATIVE
- ? ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL) | code16
- : ENCODE_RELAX_STATE (COND_JUMP, SMALL) | code16),
- i.disps[0]->X_add_symbol,
- i.disps[0]->X_add_number,
- p);
+ as_bad (_("`%%%s' not allowed with `%s%c'"),
+ i.op[op].regs->reg_name,
+ i.tm.name,
+ i.suffix);
+ return 0;
}
- else if (i.tm.opcode_modifier & (JumpByte | JumpDword))
+ /* Warn if the e prefix on a general reg is missing. */
+ else if ((!quiet_warnings || flag_code == CODE_64BIT)
+ && (i.types[op] & Reg16) != 0
+ && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
{
- int size;
-
- if (i.tm.opcode_modifier & JumpByte)
+ /* Prohibit these changes in the 64bit mode, since the
+ lowering is more complicated. */
+ if (flag_code == CODE_64BIT)
{
- /* This is a loop or jecxz type instruction. */
- size = 1;
- if (i.prefix[ADDR_PREFIX])
- {
- insn_size += 1;
- FRAG_APPEND_1_CHAR (ADDR_PREFIX_OPCODE);
- i.prefixes -= 1;
- }
- }
- else
- {
- int code16;
-
- code16 = 0;
- if (flag_16bit_code)
- code16 = CODE16;
-
- if (i.prefix[DATA_PREFIX])
- {
- insn_size += 1;
- FRAG_APPEND_1_CHAR (DATA_PREFIX_OPCODE);
- i.prefixes -= 1;
- code16 ^= CODE16;
- }
-
- size = 4;
- if (code16)
- size = 2;
- }
-
- if (i.prefixes != 0 && !intel_syntax)
- as_warn (_("skipping prefixes on this instruction"));
-
- if (fits_in_unsigned_byte (i.tm.base_opcode))
- {
- insn_size += 1 + size;
- p = frag_more (1 + size);
+ as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
+ i.op[op].regs->reg_name,
+ i.suffix);
+ return 0;
}
+#if REGISTER_WARNINGS
else
- {
- /* 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.disps[0], 1, reloc (size, 1, i.disp_reloc[0]));
+ as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
+ (i.op[op].regs + REGNAM_EAX - REGNAM_AX)->reg_name,
+ i.op[op].regs->reg_name,
+ i.suffix);
+#endif
}
- else if (i.tm.opcode_modifier & JumpInterSegment)
+ /* Warn if the r prefix on a general reg is missing. */
+ else if ((i.types[op] & Reg64) != 0
+ && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
{
- int size;
- int prefix;
- int code16;
-
- code16 = 0;
- if (flag_16bit_code)
- code16 = CODE16;
-
- prefix = 0;
- if (i.prefix[DATA_PREFIX])
- {
- 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"));
+ as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
+ i.op[op].regs->reg_name,
+ i.suffix);
+ return 0;
+ }
+ return 1;
+}
- insn_size += prefix + 1 + 2 + size; /* 1 opcode; 2 segment; offset */
- p = frag_more (prefix + 1 + 2 + size);
- if (prefix)
- *p++ = DATA_PREFIX_OPCODE;
- *p++ = i.tm.base_opcode;
- if (i.imms[1]->X_op == O_constant)
- {
- long n = (long) i.imms[1]->X_add_number;
+static int
+check_qword_reg ()
+{
+ int op;
- if (size == 2 && !fits_in_unsigned_word (n))
- {
- as_bad (_("16-bit jump out of range"));
- return;
- }
- md_number_to_chars (p, (valueT) n, size);
- }
- else
- fix_new_exp (frag_now, p - frag_now->fr_literal, size,
- i.imms[1], 0, reloc (size, 0, i.disp_reloc[0]));
- if (i.imms[0]->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.imms[0]->X_add_number, 2);
+ 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 0;
}
- else
+ /* Warn if the e prefix on a general reg is missing. */
+ else if (((i.types[op] & Reg16) != 0
+ || (i.types[op] & Reg32) != 0)
+ && (i.tm.operand_types[op] & (Reg32 | Acc)) != 0)
{
- /* Output normal instructions here. */
- unsigned char *q;
-
- /* The prefix bytes. */
- for (q = i.prefix;
- q < i.prefix + sizeof (i.prefix) / sizeof (i.prefix[0]);
- q++)
- {
- if (*q)
- {
- insn_size += 1;
- p = frag_more (1);
- md_number_to_chars (p, (valueT) *q, 1);
- }
- }
+ /* Prohibit these changes in the 64bit mode, since the
+ lowering is more complicated. */
+ as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
+ i.op[op].regs->reg_name,
+ i.suffix);
+ return 0;
+ }
+ return 1;
+}
- /* Now the opcode; be careful about word order here! */
- if (fits_in_unsigned_byte (i.tm.base_opcode))
- {
- insn_size += 1;
- FRAG_APPEND_1_CHAR (i.tm.base_opcode);
- }
- else if (fits_in_unsigned_word (i.tm.base_opcode))
+static int
+check_word_reg ()
+{
+ 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 0;
+ }
+ /* Warn if the e prefix on a general reg is present. */
+ else if ((!quiet_warnings || flag_code == CODE_64BIT)
+ && (i.types[op] & Reg32) != 0
+ && (i.tm.operand_types[op] & (Reg16 | Acc)) != 0)
+ {
+ /* Prohibit these changes in the 64bit mode, since the
+ lowering is more complicated. */
+ if (flag_code == CODE_64BIT)
{
- insn_size += 2;
- p = frag_more (2);
- /* 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;
+ as_bad (_("Incorrect register `%%%s' used with `%c' suffix"),
+ i.op[op].regs->reg_name,
+ i.suffix);
+ return 0;
}
else
- { /* opcode is either 3 or 4 bytes */
- if (i.tm.base_opcode & 0xff000000)
- {
- insn_size += 4;
- p = frag_more (4);
- *p++ = (i.tm.base_opcode >> 24) & 0xff;
- }
- else
- {
- 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 sib byte (if present). */
- if (i.tm.opcode_modifier & Modrm)
- {
- insn_size += 1;
- p = frag_more (1);
- 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,
- (valueT) (i.sib.base << 0
- | i.sib.index << 3
- | i.sib.scale << 6),
- 1);
- }
- }
-
- if (i.disp_operands)
- {
- register unsigned int n;
-
- for (n = 0; n < i.operands; n++)
- {
- if (i.disps[n])
- {
- if (i.disps[n]->X_op == O_constant)
- {
- int size = 4;
- long val = (long) i.disps[n]->X_add_number;
-
- if (i.types[n] & (Disp8 | Disp16))
- {
- long mask;
-
- size = 2;
- mask = ~ (long) 0xffff;
- if (i.types[n] & Disp8)
- {
- size = 1;
- mask = ~ (long) 0xff;
- }
-
- if ((val & mask) != 0 && (val & mask) != mask)
- as_warn (_("%ld shortened to %ld"),
- val, val & ~mask);
- }
- insn_size += size;
- p = frag_more (size);
- md_number_to_chars (p, (valueT) val, size);
- }
- else if (i.types[n] & Disp32)
- {
- insn_size += 4;
- p = frag_more (4);
- fix_new_exp (frag_now, p - frag_now->fr_literal, 4,
- i.disps[n], 0,
- TC_RELOC (i.disp_reloc[n], BFD_RELOC_32));
- }
- else
- { /* must be Disp16 */
- insn_size += 2;
- p = frag_more (2);
- fix_new_exp (frag_now, p - frag_now->fr_literal, 2,
- i.disps[n], 0,
- TC_RELOC (i.disp_reloc[n], BFD_RELOC_16));
- }
- }
- }
- } /* end displacement output */
-
- /* output immediate */
- if (i.imm_operands)
- {
- register unsigned int n;
-
- for (n = 0; n < i.operands; n++)
- {
- if (i.imms[n])
- {
- if (i.imms[n]->X_op == O_constant)
- {
- int size = 4;
- long val = (long) i.imms[n]->X_add_number;
-
- if (i.types[n] & (Imm8 | Imm8S | Imm16))
- {
- long mask;
-
- size = 2;
- mask = ~ (long) 0xffff;
- if (i.types[n] & (Imm8 | Imm8S))
- {
- size = 1;
- mask = ~ (long) 0xff;
- }
- if ((val & mask) != 0 && (val & mask) != mask)
- as_warn (_("%ld shortened to %ld"),
- val, val & ~mask);
- }
- insn_size += size;
- p = frag_more (size);
- md_number_to_chars (p, (valueT) val, size);
- }
- else
- { /* not absolute_section */
- /* Need a 32-bit fixup (don't support 8bit
- non-absolute ims). Try to support other
- sizes ... */
-#ifdef BFD_ASSEMBLER
- enum bfd_reloc_code_real reloc_type;
-#else
- int reloc_type;
-#endif
- int size;
- int pcrel = 0;
-
- if (i.types[n] & (Imm8 | Imm8S))
- size = 1;
- else if (i.types[n] & Imm16)
- size = 2;
- else
- size = 4;
- insn_size += size;
- p = frag_more (size);
- reloc_type = reloc (size, 0, i.disp_reloc[0]);
-#ifdef BFD_ASSEMBLER
- if (reloc_type == BFD_RELOC_32
- && GOT_symbol
- && GOT_symbol == i.imms[n]->X_add_symbol
- && (i.imms[n]->X_op == O_symbol
- || (i.imms[n]->X_op == O_add
- && ((symbol_get_value_expression
- (i.imms[n]->X_op_symbol)->X_op)
- == O_subtract))))
- {
- reloc_type = BFD_RELOC_386_GOTPC;
- i.imms[n]->X_add_number += 3;
- }
+#if REGISTER_WARNINGS
+ as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"),
+ (i.op[op].regs + REGNAM_AX - REGNAM_EAX)->reg_name,
+ i.op[op].regs->reg_name,
+ i.suffix);
#endif
- fix_new_exp (frag_now, p - frag_now->fr_literal, size,
- i.imms[n], pcrel, reloc_type);
- }
- }
- }
- } /* end immediate output */
- }
-
-#ifdef DEBUG386
- if (flag_debug)
- {
- pi (line, &i);
}
-#endif /* DEBUG386 */
- }
+ return 1;
}
-\f
-static int i386_immediate PARAMS ((char *));
static int
-i386_immediate (imm_start)
- char *imm_start;
+finalize_imm ()
{
- char *save_input_line_pointer;
- segT exp_seg = 0;
- expressionS * exp;
+ unsigned int overlap0, overlap1, overlap2;
- if (i.imm_operands == MAX_IMMEDIATE_OPERANDS)
+ overlap0 = i.types[0] & i.tm.operand_types[0];
+ if ((overlap0 & (Imm8 | Imm8S | Imm16 | Imm32 | Imm32S))
+ && overlap0 != Imm8 && overlap0 != Imm8S
+ && overlap0 != Imm16 && overlap0 != Imm32S
+ && overlap0 != Imm32 && overlap0 != Imm64)
{
- as_bad (_("Only 1 or 2 immediate operands are allowed"));
- return 0;
+ if (i.suffix)
+ {
+ overlap0 &= (i.suffix == BYTE_MNEM_SUFFIX
+ ? Imm8 | Imm8S
+ : (i.suffix == WORD_MNEM_SUFFIX
+ ? Imm16
+ : (i.suffix == QWORD_MNEM_SUFFIX
+ ? Imm64 | Imm32S
+ : Imm32)));
+ }
+ else if (overlap0 == (Imm16 | Imm32S | Imm32)
+ || overlap0 == (Imm16 | Imm32)
+ || overlap0 == (Imm16 | Imm32S))
+ {
+ overlap0 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
+ ? Imm16 : Imm32S);
+ }
+ if (overlap0 != Imm8 && overlap0 != Imm8S
+ && overlap0 != Imm16 && overlap0 != Imm32S
+ && overlap0 != Imm32 && overlap0 != Imm64)
+ {
+ as_bad (_("no instruction mnemonic suffix given; can't determine immediate size"));
+ return 0;
+ }
}
+ i.types[0] = overlap0;
- exp = &im_expressions[i.imm_operands++];
- i.imms[this_operand] = exp;
-
- 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);
+ overlap1 = i.types[1] & i.tm.operand_types[1];
+ if ((overlap1 & (Imm8 | Imm8S | Imm16 | Imm32S | Imm32))
+ && overlap1 != Imm8 && overlap1 != Imm8S
+ && overlap1 != Imm16 && overlap1 != Imm32S
+ && overlap1 != Imm32 && overlap1 != Imm64)
+ {
+ if (i.suffix)
+ {
+ overlap1 &= (i.suffix == BYTE_MNEM_SUFFIX
+ ? Imm8 | Imm8S
+ : (i.suffix == WORD_MNEM_SUFFIX
+ ? Imm16
+ : (i.suffix == QWORD_MNEM_SUFFIX
+ ? Imm64 | Imm32S
+ : Imm32)));
+ }
+ else if (overlap1 == (Imm16 | Imm32 | Imm32S)
+ || overlap1 == (Imm16 | Imm32)
+ || overlap1 == (Imm16 | Imm32S))
+ {
+ overlap1 = ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)
+ ? Imm16 : Imm32S);
+ }
+ if (overlap1 != Imm8 && overlap1 != Imm8S
+ && overlap1 != Imm16 && overlap1 != Imm32S
+ && overlap1 != Imm32 && overlap1 != Imm64)
+ {
+ as_bad (_("no instruction mnemonic suffix given; can't determine immediate size %x %c"),overlap1, i.suffix);
+ return 0;
+ }
+ }
+ i.types[1] = overlap1;
- SKIP_WHITESPACE ();
- if (*input_line_pointer)
- as_bad (_("Ignoring junk `%s' after expression"), input_line_pointer);
+ overlap2 = i.types[2] & i.tm.operand_types[2];
+ assert ((overlap2 & Imm) == 0);
+ i.types[2] = overlap2;
- input_line_pointer = save_input_line_pointer;
+ return 1;
+}
- if (exp->X_op == O_absent || exp->X_op == O_big)
+static int
+process_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, and the clr %reg instruction
+ is converted into xor %reg, %reg. */
+ if (i.tm.opcode_modifier & regKludge)
{
- /* 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;
+ 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;
}
- if (exp->X_op == O_constant)
+ if (i.tm.opcode_modifier & ShortForm)
{
- int bigimm = Imm32;
- if (flag_16bit_code ^ (i.prefix[DATA_PREFIX] != 0))
- bigimm = Imm16;
-
- i.types[this_operand] |=
- (bigimm | smallest_imm_type ((long) exp->X_add_number));
+ /* 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 ((i.op[op].regs->reg_flags & RegRex) != 0)
+ i.rex |= REX_EXTZ;
+ if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0)
+ {
+ /* 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);
+ }
+ }
+ }
+ else if (i.tm.opcode_modifier & Modrm)
+ {
+ /* 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. */
- /* If a suffix is given, this operand may be shortened. */
- switch (i.suffix)
+ default_seg = build_modrm_byte ();
+ }
+ else if (i.tm.opcode_modifier & (Seg2ShortForm | Seg3ShortForm))
+ {
+ if (i.tm.base_opcode == POP_SEG_SHORT
+ && i.op[0].regs->reg_num == 1)
{
- case WORD_MNEM_SUFFIX:
- i.types[this_operand] |= Imm16;
- break;
- case BYTE_MNEM_SUFFIX:
- i.types[this_operand] |= Imm16 | Imm8 | Imm8S;
- break;
+ as_bad (_("you can't `pop %%cs'"));
+ return 0;
}
+ i.tm.base_opcode |= (i.op[0].regs->reg_num << 3);
+ if ((i.op[0].regs->reg_flags & RegRex) != 0)
+ i.rex |= REX_EXTZ;
}
-#if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
- else if (
-#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
- && !bfd_is_com_section (exp_seg)
-#endif
- )
+ else if ((i.tm.base_opcode & ~(D | W)) == MOV_AX_DISP32)
{
-#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;
+ default_seg = &ds;
}
-#endif
- else
+ else if ((i.tm.opcode_modifier & IsString) != 0)
{
- /* 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);
+ /* 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 0;
+ }
return 1;
}
-static int i386_scale PARAMS ((char *));
-
-static int
-i386_scale (scale)
- char *scale;
+static const seg_entry *
+build_modrm_byte ()
{
- if (!isdigit (*scale))
- goto bad_scale;
+ const seg_entry *default_seg = 0;
- 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)
+ /* i.reg_operands MUST be the number of real register operands;
+ implicit registers do not count. */
+ if (i.reg_operands == 2)
{
- 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
+ unsigned int source, dest;
+ source = ((i.types[0]
+ & (Reg | RegMMX | RegXMM
+ | SReg2 | SReg3
+ | Control | Debug | Test))
+ ? 0 : 1);
+ dest = source + 1;
+
+ i.rm.mode = 3;
+ /* 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.op[dest].regs->reg_num;
+ i.rm.regmem = i.op[source].regs->reg_num;
+ if ((i.op[dest].regs->reg_flags & RegRex) != 0)
+ i.rex |= REX_EXTX;
+ if ((i.op[source].regs->reg_flags & RegRex) != 0)
+ i.rex |= REX_EXTZ;
+ }
+ else
+ {
+ i.rm.reg = i.op[source].regs->reg_num;
+ i.rm.regmem = i.op[dest].regs->reg_num;
+ if ((i.op[dest].regs->reg_flags & RegRex) != 0)
+ i.rex |= REX_EXTZ;
+ if ((i.op[source].regs->reg_flags & RegRex) != 0)
+ i.rex |= REX_EXTX;
+ }
}
- return 1;
-}
-
-static int i386_displacement PARAMS ((char *, char *));
-
-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;
+ else
+ { /* If it's not 2 reg operands... */
+ if (i.mem_operands)
+ {
+ unsigned int fake_zero_displacement = 0;
+ unsigned int op = ((i.types[0] & AnyMem)
+ ? 0
+ : (i.types[1] & AnyMem) ? 1 : 2);
- if (flag_16bit_code ^ (i.prefix[ADDR_PREFIX] != 0))
- bigdisp = Disp16;
- i.types[this_operand] |= bigdisp;
+ default_seg = &ds;
- exp = &disp_expressions[i.disp_operands];
- i.disps[this_operand] = exp;
- i.disp_reloc[this_operand] = NO_RELOC;
- 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
+ if (i.base_reg == 0)
+ {
+ i.rm.mode = 0;
+ if (!i.disp_operands)
+ fake_zero_displacement = 1;
+ if (i.index_reg == 0)
+ {
+ /* Operand is just <disp> */
+ if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0)
+ && (flag_code != CODE_64BIT))
+ {
+ i.rm.regmem = NO_BASE_REGISTER_16;
+ i.types[op] &= ~Disp;
+ i.types[op] |= Disp16;
+ }
+ else if (flag_code != CODE_64BIT
+ || (i.prefix[ADDR_PREFIX] != 0))
+ {
+ i.rm.regmem = NO_BASE_REGISTER;
+ i.types[op] &= ~Disp;
+ i.types[op] |= Disp32;
+ }
+ else
+ {
+ /* 64bit mode overwrites the 32bit absolute
+ addressing by RIP relative addressing and
+ absolute addressing is encoded by one of the
+ redundant SIB forms. */
+ i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
+ i.sib.base = NO_BASE_REGISTER;
+ i.sib.index = NO_INDEX_REGISTER;
+ i.types[op] &= ~Disp;
+ i.types[op] |= Disp32S;
+ }
+ }
+ else /* !i.base_reg && i.index_reg */
+ {
+ i.sib.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;
+ if (flag_code != CODE_64BIT)
+ i.types[op] |= Disp32; /* Must be 32 bit */
+ else
+ i.types[op] |= Disp32S;
+ if ((i.index_reg->reg_flags & RegRex) != 0)
+ i.rex |= REX_EXTY;
+ }
+ }
+ /* RIP addressing for 64bit mode. */
+ else if (i.base_reg->reg_type == BaseIndex)
+ {
+ i.rm.regmem = NO_BASE_REGISTER;
+ i.types[op] &= ~Disp;
+ i.types[op] |= Disp32S;
+ i.flags[op] = Operand_PCrel;
+ }
+ else if (i.base_reg->reg_type & Reg16)
+ {
+ switch (i.base_reg->reg_num)
+ {
+ case 3: /* (%bx) */
+ if (i.index_reg == 0)
+ 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 == 0)
+ {
+ 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 /* i.base_reg and 32/64 bit mode */
+ {
+ if (flag_code == CODE_64BIT
+ && (i.types[op] & Disp))
+ {
+ if (i.types[op] & Disp8)
+ i.types[op] = Disp8 | Disp32S;
+ else
+ i.types[op] = Disp32S;
+ }
+ i.rm.regmem = i.base_reg->reg_num;
+ if ((i.base_reg->reg_flags & RegRex) != 0)
+ i.rex |= REX_EXTZ;
+ i.sib.base = i.base_reg->reg_num;
+ /* x86-64 ignores REX prefix bit here to avoid decoder
+ complications. */
+ if ((i.base_reg->reg_num & 7) == EBP_REG_NUM)
+ {
+ default_seg = &ss;
+ if (i.disp_operands == 0)
+ {
+ fake_zero_displacement = 1;
+ i.types[op] |= Disp8;
+ }
+ }
+ else if (i.base_reg->reg_num == ESP_REG_NUM)
+ {
+ default_seg = &ss;
+ }
+ i.sib.scale = i.log2_scale_factor;
+ if (i.index_reg == 0)
+ {
+ /* <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 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))
+ }
+ else
+ {
+ i.sib.index = i.index_reg->reg_num;
+ i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING;
+ if ((i.index_reg->reg_flags & RegRex) != 0)
+ i.rex |= REX_EXTY;
+ }
+ i.rm.mode = mode_from_disp_size (i.types[op]);
+ }
- 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:
+ 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;
+ }
+ }
- #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
+ /* 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 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 (i.tm.extension_opcode != None)
+ {
+ i.rm.regmem = i.op[op].regs->reg_num;
+ if ((i.op[op].regs->reg_flags & RegRex) != 0)
+ i.rex |= REX_EXTZ;
+ }
+ else
+ {
+ i.rm.reg = i.op[op].regs->reg_num;
+ if ((i.op[op].regs->reg_flags & RegRex) != 0)
+ i.rex |= REX_EXTX;
+ }
- So here we provide the missing zero.
- */
+ /* 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
+ in the regmem field. */
+ if (!i.mem_operands)
+ i.rm.mode = 3;
+ }
- *displacement_string_end = '0';
+ /* Fill in i.rm.reg field with extension opcode (if any). */
+ if (i.tm.extension_opcode != None)
+ i.rm.reg = i.tm.extension_opcode;
}
-#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);
+ return default_seg;
+}
-#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)
+static void
+output_branch ()
+{
+ char *p;
+ int code16;
+ int prefix;
+ relax_substateT subtype;
+ symbolS *sym;
+ offsetT off;
+
+ code16 = 0;
+ if (flag_code == CODE_16BIT)
+ code16 = CODE16;
+
+ prefix = 0;
+ if (i.prefix[DATA_PREFIX] != 0)
{
- 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;
+ prefix = 1;
+ i.prefixes -= 1;
+ code16 ^= CODE16;
}
-#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)
+ /* Pentium4 branch hints. */
+ if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
+ || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
{
- /* 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;
+ prefix++;
+ i.prefixes--;
}
-
- if (exp->X_op == O_constant)
+ if (i.prefix[REX_PREFIX] != 0)
{
- if (fits_in_signed_byte (exp->X_add_number))
- i.types[this_operand] |= Disp8;
+ prefix++;
+ i.prefixes--;
}
-#if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
- else if (
-#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)
+
+ 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 + 4);
+ /* Prefix and 1 opcode byte go in fr_fix. */
+ p = frag_more (prefix + 1);
+ if (i.prefix[DATA_PREFIX] != 0)
+ *p++ = DATA_PREFIX_OPCODE;
+ if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE
+ || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE)
+ *p++ = i.prefix[SEG_PREFIX];
+ if (i.prefix[REX_PREFIX] != 0)
+ *p++ = i.prefix[REX_PREFIX];
+ *p = i.tm.base_opcode;
+
+ if ((unsigned char) *p == JUMP_PC_RELATIVE)
+ subtype = ENCODE_RELAX_STATE (UNCOND_JUMP, SMALL);
+ else if ((cpu_arch_flags & Cpu386) != 0)
+ subtype = ENCODE_RELAX_STATE (COND_JUMP, SMALL);
+ else
+ subtype = ENCODE_RELAX_STATE (COND_JUMP86, SMALL);
+ subtype |= code16;
+
+ sym = i.op[0].disps->X_add_symbol;
+ off = i.op[0].disps->X_add_number;
+
+ if (i.op[0].disps->X_op != O_constant
+ && i.op[0].disps->X_op != O_symbol)
{
-#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;
+ /* Handle complex expressions. */
+ sym = make_expr_symbol (i.op[0].disps);
+ off = 0;
}
-#endif
- return 1;
-}
-static int i386_operand_modifier PARAMS ((char **, int));
+ /* 1 possible extra opcode + 4 byte displacement go in var part.
+ Pass reloc in fr_var. */
+ frag_var (rs_machine_dependent, 5, i.reloc[0], subtype, sym, off, p);
+}
-static int
-i386_operand_modifier (op_string, got_a_float)
- char **op_string;
- int got_a_float;
+static void
+output_jump ()
{
- if (!strncasecmp (*op_string, "BYTE PTR", 8))
- {
- i.suffix = BYTE_MNEM_SUFFIX;
- *op_string += 8;
- return BYTE_PTR;
+ char *p;
+ int size;
- }
- else if (!strncasecmp (*op_string, "WORD PTR", 8))
+ if (i.tm.opcode_modifier & JumpByte)
{
- i.suffix = WORD_MNEM_SUFFIX;
- *op_string += 8;
- return WORD_PTR;
+ /* This is a loop or jecxz type instruction. */
+ size = 1;
+ if (i.prefix[ADDR_PREFIX] != 0)
+ {
+ FRAG_APPEND_1_CHAR (ADDR_PREFIX_OPCODE);
+ i.prefixes -= 1;
+ }
+ /* Pentium4 branch hints. */
+ if (i.prefix[SEG_PREFIX] == CS_PREFIX_OPCODE /* not taken */
+ || i.prefix[SEG_PREFIX] == DS_PREFIX_OPCODE /* taken */)
+ {
+ FRAG_APPEND_1_CHAR (i.prefix[SEG_PREFIX]);
+ i.prefixes--;
+ }
}
-
- else if (!strncasecmp (*op_string, "DWORD PTR", 9))
+ else
{
- if (got_a_float)
- i.suffix = SHORT_MNEM_SUFFIX;
- else
- i.suffix = LONG_MNEM_SUFFIX;
- *op_string += 9;
- return DWORD_PTR;
- }
+ int code16;
- else if (!strncasecmp (*op_string, "QWORD PTR", 9))
- {
- i.suffix = DWORD_MNEM_SUFFIX;
- *op_string += 9;
- return QWORD_PTR;
- }
+ code16 = 0;
+ if (flag_code == CODE_16BIT)
+ code16 = CODE16;
- else if (!strncasecmp (*op_string, "XWORD PTR", 9))
- {
- i.suffix = LONG_DOUBLE_MNEM_SUFFIX;
- *op_string += 9;
- return XWORD_PTR;
- }
+ if (i.prefix[DATA_PREFIX] != 0)
+ {
+ FRAG_APPEND_1_CHAR (DATA_PREFIX_OPCODE);
+ i.prefixes -= 1;
+ code16 ^= CODE16;
+ }
- else if (!strncasecmp (*op_string, "SHORT", 5))
- {
- *op_string += 5;
- return SHORT;
+ size = 4;
+ if (code16)
+ size = 2;
}
- else if (!strncasecmp (*op_string, "OFFSET FLAT:", 12))
+ if (i.prefix[REX_PREFIX] != 0)
{
- *op_string += 12;
- return OFFSET_FLAT;
+ FRAG_APPEND_1_CHAR (i.prefix[REX_PREFIX]);
+ i.prefixes -= 1;
}
- else if (!strncasecmp (*op_string, "FLAT", 4))
- {
- *op_string += 4;
- return FLAT;
- }
+ if (i.prefixes != 0 && !intel_syntax)
+ as_warn (_("skipping prefixes on this instruction"));
- else return NONE_FOUND;
-}
+ p = frag_more (1 + size);
+ *p++ = i.tm.base_opcode;
-static char * build_displacement_string PARAMS ((int, char *));
+ fix_new_exp (frag_now, p - frag_now->fr_literal, size,
+ i.op[0].disps, 1, reloc (size, 1, 1, i.reloc[0]));
+}
-static char *
-build_displacement_string (initial_disp, op_string)
- int initial_disp;
- char *op_string;
+static void
+output_interseg_jump ()
{
- char *temp_string = (char *) malloc (strlen (op_string) + 1);
- char *end_of_operand_string;
- char *tc;
- char *temp_disp;
+ char *p;
+ int size;
+ int prefix;
+ int code16;
- temp_string[0] = '\0';
- tc = end_of_operand_string = strchr (op_string, '[');
- if ( initial_disp && !end_of_operand_string)
+ code16 = 0;
+ if (flag_code == CODE_16BIT)
+ code16 = CODE16;
+
+ prefix = 0;
+ if (i.prefix[DATA_PREFIX] != 0)
{
- strcpy (temp_string, op_string);
- return (temp_string);
+ prefix = 1;
+ i.prefixes -= 1;
+ code16 ^= CODE16;
}
-
- /* Build the whole displacement string */
- if (initial_disp)
+ if (i.prefix[REX_PREFIX] != 0)
{
- strncpy (temp_string, op_string, end_of_operand_string - op_string);
- temp_string[end_of_operand_string - op_string] = '\0';
- temp_disp = tc;
+ prefix++;
+ i.prefixes -= 1;
}
- else
- temp_disp = op_string;
- while (*temp_disp != '\0')
- {
- char *end_op;
- int add_minus = (*temp_disp == '-');
+ size = 4;
+ if (code16)
+ size = 2;
- if (*temp_disp == '+' || *temp_disp == '-' || *temp_disp == '[')
- temp_disp++;
+ if (i.prefixes != 0 && !intel_syntax)
+ as_warn (_("skipping prefixes on this instruction"));
- if (is_space_char (*temp_disp))
- temp_disp++;
+ /* 1 opcode; 2 segment; offset */
+ p = frag_more (prefix + 1 + 2 + size);
- /* Don't consider registers */
- if ( !((*temp_disp == REGISTER_PREFIX || allow_naked_reg)
- && parse_register (temp_disp, &end_op)) )
- {
- char *string_start = temp_disp;
+ if (i.prefix[DATA_PREFIX] != 0)
+ *p++ = DATA_PREFIX_OPCODE;
- while (*temp_disp != ']'
- && *temp_disp != '+'
- && *temp_disp != '-'
- && *temp_disp != '*')
- ++temp_disp;
+ if (i.prefix[REX_PREFIX] != 0)
+ *p++ = i.prefix[REX_PREFIX];
- if (add_minus)
- strcat (temp_string, "-");
- else
- strcat (temp_string, "+");
+ *p++ = i.tm.base_opcode;
+ if (i.op[1].imms->X_op == O_constant)
+ {
+ offsetT n = i.op[1].imms->X_add_number;
- strncat (temp_string, string_start, temp_disp - string_start);
- if (*temp_disp == '+' || *temp_disp == '-')
- --temp_disp;
+ if (size == 2
+ && !fits_in_unsigned_word (n)
+ && !fits_in_signed_word (n))
+ {
+ as_bad (_("16-bit jump out of range"));
+ return;
}
-
- while (*temp_disp != '\0'
- && *temp_disp != '+'
- && *temp_disp != '-')
- ++temp_disp;
+ md_number_to_chars (p, n, size);
}
-
- return temp_string;
+ else
+ fix_new_exp (frag_now, p - frag_now->fr_literal, size,
+ i.op[1].imms, 0, reloc (size, 0, 0, i.reloc[1]));
+ 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);
}
-static int i386_parse_seg PARAMS ((char *));
-
-static int
-i386_parse_seg (op_string)
- char *op_string;
+static void
+output_insn ()
{
- if (is_space_char (*op_string))
- ++op_string;
-
- /* Should be one of es, cs, ss, ds fs or gs */
- switch (*op_string++)
+ /* Tie dwarf2 debug info to the address at the start of the insn.
+ We can't do this after the insn has been output as the current
+ frag may have been closed off. eg. by frag_var. */
+ dwarf2_emit_insn (0);
+
+ /* Output jumps. */
+ if (i.tm.opcode_modifier & Jump)
+ output_branch ();
+ else if (i.tm.opcode_modifier & (JumpByte | JumpDword))
+ output_jump ();
+ else if (i.tm.opcode_modifier & JumpInterSegment)
+ output_interseg_jump ();
+ else
{
- case 'e':
- i.seg[i.mem_operands] = &es;
- break;
- case 'c':
- i.seg[i.mem_operands] = &cs;
- break;
- case 's':
- i.seg[i.mem_operands] = &ss;
- break;
- case 'd':
- i.seg[i.mem_operands] = &ds;
- break;
- case 'f':
- i.seg[i.mem_operands] = &fs;
- break;
- case 'g':
- i.seg[i.mem_operands] = &gs;
- break;
- default:
- as_bad (_("bad segment name `%s'"), op_string);
- return 0;
- }
+ /* Output normal instructions here. */
+ char *p;
+ unsigned char *q;
+
+ /* All opcodes on i386 have either 1 or 2 bytes. We may use third
+ byte for the SSE instructions to specify a prefix they require. */
+ if (i.tm.base_opcode & 0xff0000)
+ add_prefix ((i.tm.base_opcode >> 16) & 0xff);
+
+ /* The prefix bytes. */
+ for (q = i.prefix;
+ q < i.prefix + sizeof (i.prefix) / sizeof (i.prefix[0]);
+ q++)
+ {
+ if (*q)
+ {
+ p = frag_more (1);
+ md_number_to_chars (p, (valueT) *q, 1);
+ }
+ }
- if (*op_string++ != 's')
- {
- as_bad (_("bad segment name `%s'"), op_string);
- return 0;
- }
+ /* Now the opcode; be careful about word order here! */
+ if (fits_in_unsigned_byte (i.tm.base_opcode))
+ {
+ FRAG_APPEND_1_CHAR (i.tm.base_opcode);
+ }
+ else
+ {
+ p = frag_more (2);
+ /* 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;
+ }
- if (is_space_char (*op_string))
- ++op_string;
+ /* Now the modrm byte and sib byte (if present). */
+ if (i.tm.opcode_modifier & Modrm)
+ {
+ p = frag_more (1);
+ 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))
+ {
+ p = frag_more (1);
+ md_number_to_chars (p,
+ (valueT) (i.sib.base << 0
+ | i.sib.index << 3
+ | i.sib.scale << 6),
+ 1);
+ }
+ }
- if (*op_string != ':')
- {
- as_bad (_("bad segment name `%s'"), op_string);
- return 0;
- }
+ if (i.disp_operands)
+ output_disp ();
- return 1;
+ if (i.imm_operands)
+ output_imm ();
+ }
+#ifdef DEBUG386
+ if (flag_debug)
+ {
+ pi (line, &i);
+ }
+#endif /* DEBUG386 */
}
-static int i386_index_check PARAMS((const char *));
-
-/* Make sure the memory operand we've been dealt is valid.
- Returns 1 on success, 0 on a failure.
-*/
-static int
-i386_index_check (operand_string)
- const char *operand_string;
+static void
+output_disp ()
{
-#if INFER_ADDR_PREFIX
- int fudged = 0;
+ char *p;
+ unsigned int n;
- 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)))))
+ for (n = 0; n < i.operands; n++)
{
-#if INFER_ADDR_PREFIX
- if (i.prefix[ADDR_PREFIX] == 0 && stackop_size != '\0')
+ if (i.types[n] & Disp)
{
- 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;
-}
+ if (i.op[n].disps->X_op == O_constant)
+ {
+ int size;
+ offsetT val;
-static int i386_intel_memory_operand PARAMS ((char *));
+ size = 4;
+ if (i.types[n] & (Disp8 | Disp16 | Disp64))
+ {
+ size = 2;
+ if (i.types[n] & Disp8)
+ size = 1;
+ if (i.types[n] & Disp64)
+ size = 8;
+ }
+ val = offset_in_range (i.op[n].disps->X_add_number,
+ size);
+ p = frag_more (size);
+ md_number_to_chars (p, val, size);
+ }
+ else
+ {
+ int size = 4;
+ int sign = 0;
+ int pcrel = (i.flags[n] & Operand_PCrel) != 0;
+
+ /* The PC relative address is computed relative
+ to the instruction boundary, so in case immediate
+ fields follows, we need to adjust the value. */
+ if (pcrel && i.imm_operands)
+ {
+ int imm_size = 4;
+ unsigned int n1;
-static int
-i386_intel_memory_operand (operand_string)
- char *operand_string;
-{
- char *op_string = operand_string;
- char *end_of_operand_string;
+ for (n1 = 0; n1 < i.operands; n1++)
+ if (i.types[n1] & Imm)
+ {
+ if (i.types[n1] & (Imm8 | Imm8S | Imm16 | Imm64))
+ {
+ imm_size = 2;
+ if (i.types[n1] & (Imm8 | Imm8S))
+ imm_size = 1;
+ if (i.types[n1] & Imm64)
+ imm_size = 8;
+ }
+ break;
+ }
+ /* We should find the immediate. */
+ if (n1 == i.operands)
+ abort ();
+ i.op[n].disps->X_add_number -= imm_size;
+ }
- if ((i.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;
- }
+ if (i.types[n] & Disp32S)
+ sign = 1;
- /* Look for displacement preceding open bracket */
- if (*op_string != '[')
- {
- char *end_seg;
- char *temp_string;
+ if (i.types[n] & (Disp16 | Disp64))
+ {
+ size = 2;
+ if (i.types[n] & Disp64)
+ size = 8;
+ }
- end_seg = strchr (op_string, ':');
- if (end_seg)
- {
- if (!i386_parse_seg (op_string))
- return 0;
- op_string = end_seg + 1;
+ p = frag_more (size);
+ fix_new_exp (frag_now, p - frag_now->fr_literal, size,
+ i.op[n].disps, pcrel,
+ reloc (size, pcrel, sign, i.reloc[n]));
+ }
}
-
- temp_string = build_displacement_string (true, op_string);
-
- if (i.disp_operands == 0 &&
- !i386_displacement (temp_string, temp_string + strlen (temp_string)))
- return 0;
-
- end_of_operand_string = strchr (op_string, '[');
- if (!end_of_operand_string)
- end_of_operand_string = op_string + strlen (op_string);
-
- if (is_space_char (*end_of_operand_string))
- --end_of_operand_string;
-
- op_string = end_of_operand_string;
}
+}
- if (*op_string == '[')
- {
- ++op_string;
-
- /* Pick off each component and figure out where it belongs */
-
- end_of_operand_string = op_string;
+static void
+output_imm ()
+{
+ char *p;
+ unsigned int n;
- while (*op_string != ']')
+ for (n = 0; n < i.operands; n++)
+ {
+ if (i.types[n] & Imm)
{
- const reg_entry *temp_reg;
- char *end_op;
- char *temp_string;
-
- while (*end_of_operand_string != '+'
- && *end_of_operand_string != '-'
- && *end_of_operand_string != '*'
- && *end_of_operand_string != ']')
- end_of_operand_string++;
-
- temp_string = op_string;
- if (*temp_string == '+')
- {
- ++temp_string;
- if (is_space_char (*temp_string))
- ++temp_string;
- }
-
- if ((*temp_string == REGISTER_PREFIX || allow_naked_reg)
- && (temp_reg = parse_register (temp_string, &end_op)) != NULL)
- {
- if (i.base_reg == NULL)
- i.base_reg = temp_reg;
- else
- i.index_reg = temp_reg;
-
- i.types[this_operand] |= BaseIndex;
- }
- else if (*temp_string == REGISTER_PREFIX)
+ if (i.op[n].imms->X_op == O_constant)
{
- as_bad (_("bad register name `%s'"), temp_string);
- return 0;
- }
- else if (is_digit_char (*op_string)
- || *op_string == '+' || *op_string == '-')
- {
- temp_string = build_displacement_string (false, op_string);
-
- if (*temp_string == '+')
- ++temp_string;
+ int size;
+ offsetT val;
- if (i.disp_operands == 0 &&
- !i386_displacement (temp_string, temp_string + strlen (temp_string)))
- return 0;
-
- ++op_string;
- end_of_operand_string = op_string;
- while (*end_of_operand_string != ']'
- && *end_of_operand_string != '+'
- && *end_of_operand_string != '-'
- && *end_of_operand_string != '*')
- ++end_of_operand_string;
+ size = 4;
+ if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
+ {
+ size = 2;
+ if (i.types[n] & (Imm8 | Imm8S))
+ size = 1;
+ else if (i.types[n] & Imm64)
+ size = 8;
+ }
+ val = offset_in_range (i.op[n].imms->X_add_number,
+ size);
+ p = frag_more (size);
+ md_number_to_chars (p, val, size);
}
- else if (*op_string == '*')
+ else
{
- ++op_string;
-
- if (i.base_reg && !i.index_reg)
+ /* Not absolute_section.
+ Need a 32-bit fixup (don't support 8bit
+ non-absolute imms). Try to support other
+ sizes ... */
+ RELOC_ENUM reloc_type;
+ int size = 4;
+ int sign = 0;
+
+ if ((i.types[n] & (Imm32S))
+ && i.suffix == QWORD_MNEM_SUFFIX)
+ sign = 1;
+ if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64))
{
- i.index_reg = i.base_reg;
- i.base_reg = 0;
+ size = 2;
+ if (i.types[n] & (Imm8 | Imm8S))
+ size = 1;
+ if (i.types[n] & Imm64)
+ size = 8;
}
- if (!i386_scale (op_string))
- return 0;
+ p = frag_more (size);
+ reloc_type = reloc (size, 0, sign, i.reloc[n]);
+#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))))
+ {
+ /* We don't support dynamic linking on x86-64 yet. */
+ if (flag_code == CODE_64BIT)
+ abort ();
+ reloc_type = BFD_RELOC_386_GOTPC;
+ i.op[n].imms->X_add_number += 3;
+ }
+#endif
+ fix_new_exp (frag_now, p - frag_now->fr_literal, size,
+ i.op[n].imms, 0, reloc_type);
}
- op_string = end_of_operand_string;
- ++end_of_operand_string;
}
}
-
- if (i386_index_check (operand_string) == 0)
- return 0;
-
- i.mem_operands++;
- return 1;
}
+\f
+#ifndef LEX_AT
+static char *lex_got PARAMS ((RELOC_ENUM *, int *));
-static int
-i386_intel_operand (operand_string, got_a_float)
- char *operand_string;
- int got_a_float;
+/* Parse operands of the form
+ <symbol>@GOTOFF+<nnn>
+ and similar .plt or .got references.
+
+ If we find one, set up the correct relocation in RELOC and copy the
+ input string, minus the `@GOTOFF' into a malloc'd buffer for
+ parsing by the calling routine. Return this buffer, and if ADJUST
+ is non-null set it to the length of the string we removed from the
+ input line. Otherwise return NULL. */
+static char *
+lex_got (reloc, adjust)
+ RELOC_ENUM *reloc;
+ int *adjust;
{
- const reg_entry * r;
- char *end_op;
- char *op_string = operand_string;
+ static const char * const mode_name[NUM_FLAG_CODE] = { "32", "16", "64" };
+ static const struct {
+ const char *str;
+ const RELOC_ENUM rel[NUM_FLAG_CODE];
+ } gotrel[] = {
+ { "PLT", { BFD_RELOC_386_PLT32, 0, BFD_RELOC_X86_64_PLT32 } },
+ { "GOTOFF", { BFD_RELOC_386_GOTOFF, 0, 0 } },
+ { "GOTPCREL", { 0, 0, BFD_RELOC_X86_64_GOTPCREL } },
+ { "TLSGD", { BFD_RELOC_386_TLS_GD, 0, 0 } },
+ { "TLSLDM", { BFD_RELOC_386_TLS_LDM, 0, 0 } },
+ { "GOTTPOFF", { BFD_RELOC_386_TLS_IE_32, 0, 0 } },
+ { "TPOFF", { BFD_RELOC_386_TLS_LE_32, 0, 0 } },
+ { "NTPOFF", { BFD_RELOC_386_TLS_LE, 0, 0 } },
+ { "DTPOFF", { BFD_RELOC_386_TLS_LDO_32, 0, 0 } },
+ { "GOT", { BFD_RELOC_386_GOT32, 0, BFD_RELOC_X86_64_GOT32 } }
+ };
+ char *cp;
+ unsigned int j;
- int operand_modifier = i386_operand_modifier (&op_string, got_a_float);
- if (is_space_char (*op_string))
- ++op_string;
+ for (cp = input_line_pointer; *cp != '@'; cp++)
+ if (is_end_of_line[(unsigned char) *cp])
+ return NULL;
- switch (operand_modifier)
+ for (j = 0; j < sizeof (gotrel) / sizeof (gotrel[0]); j++)
{
- case BYTE_PTR:
- case WORD_PTR:
- case DWORD_PTR:
- case QWORD_PTR:
- case XWORD_PTR:
- if (!i386_intel_memory_operand (op_string))
- return 0;
- break;
-
- case FLAT:
- case OFFSET_FLAT:
- if (!i386_immediate (op_string))
- return 0;
- break;
+ int len;
- case SHORT:
- case NONE_FOUND:
- /* Should be register or immediate */
- if (is_digit_char (*op_string)
- && strchr (op_string, '[') == 0)
- {
- if (!i386_immediate (op_string))
- return 0;
- }
- else if ((*op_string == REGISTER_PREFIX || allow_naked_reg)
- && (r = parse_register (op_string, &end_op)) != NULL)
+ len = strlen (gotrel[j].str);
+ if (strncasecmp (cp + 1, gotrel[j].str, len) == 0)
{
- /* 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)))
+ if (gotrel[j].rel[(unsigned int) flag_code] != 0)
{
- switch (r->reg_num)
- {
- case 0:
- i.seg[i.mem_operands] = &es;
- break;
- case 1:
- i.seg[i.mem_operands] = &cs;
- break;
- case 2:
- i.seg[i.mem_operands] = &ss;
- break;
- case 3:
- i.seg[i.mem_operands] = &ds;
- break;
- case 4:
- i.seg[i.mem_operands] = &fs;
- break;
- case 5:
- i.seg[i.mem_operands] = &gs;
- break;
- }
-
+ int first, second;
+ char *tmpbuf, *past_reloc;
+
+ *reloc = gotrel[j].rel[(unsigned int) flag_code];
+ if (adjust)
+ *adjust = len;
+
+ if (GOT_symbol == NULL)
+ GOT_symbol = symbol_find_or_make (GLOBAL_OFFSET_TABLE_NAME);
+
+ /* Replace the relocation token with ' ', so that
+ errors like foo@GOTOFF1 will be detected. */
+
+ /* The length of the first part of our input line. */
+ first = cp - input_line_pointer;
+
+ /* The second part goes from after the reloc token until
+ (and including) an end_of_line char. Don't use strlen
+ here as the end_of_line char may not be a NUL. */
+ past_reloc = cp + 1 + len;
+ for (cp = past_reloc; !is_end_of_line[(unsigned char) *cp++]; )
+ ;
+ second = cp - past_reloc;
+
+ /* Allocate and copy string. The trailing NUL shouldn't
+ be necessary, but be safe. */
+ tmpbuf = xmalloc (first + second + 2);
+ memcpy (tmpbuf, input_line_pointer, first);
+ tmpbuf[first] = ' ';
+ memcpy (tmpbuf + first + 1, past_reloc, second);
+ tmpbuf[first + second + 1] = '\0';
+ return tmpbuf;
}
- i.types[this_operand] |= r->reg_type & ~BaseIndex;
- i.regs[this_operand] = r;
- i.reg_operands++;
- }
- else if (*op_string == REGISTER_PREFIX)
- {
- as_bad (_("bad register name `%s'"), op_string);
- return 0;
- }
- else if (!i386_intel_memory_operand (op_string))
- return 0;
- break;
- } /* end switch */
+ as_bad (_("@%s reloc is not supported in %s bit mode"),
+ gotrel[j].str, mode_name[(unsigned int) flag_code]);
+ return NULL;
+ }
+ }
- return 1;
+ /* Might be a symbol version string. Don't as_bad here. */
+ return NULL;
}
-/* Parse OPERAND_STRING into the i386_insn structure I. Returns non-zero
- on error. */
+/* x86_cons_fix_new is called via the expression parsing code when a
+ reloc is needed. We use this hook to get the correct .got reloc. */
+static RELOC_ENUM got_reloc = NO_RELOC;
-static int
-i386_operand (operand_string)
- char *operand_string;
+void
+x86_cons_fix_new (frag, off, len, exp)
+ fragS *frag;
+ unsigned int off;
+ unsigned int len;
+ expressionS *exp;
{
- const reg_entry *r;
- char *end_op;
- char *op_string = operand_string;
-
- if (is_space_char (*op_string))
- ++op_string;
+ RELOC_ENUM r = reloc (len, 0, 0, got_reloc);
+ got_reloc = NO_RELOC;
+ fix_new_exp (frag, off, len, exp, 0, r);
+}
- /* We check for an absolute prefix (differentiating,
- for example, 'jmp pc_relative_label' from 'jmp *absolute_label'. */
- if (*op_string == ABSOLUTE_PREFIX)
+void
+x86_cons (exp, size)
+ expressionS *exp;
+ int size;
+{
+ if (size == 4)
{
- ++op_string;
- if (is_space_char (*op_string))
- ++op_string;
- i.types[this_operand] |= JumpAbsolute;
- }
+ /* Handle @GOTOFF and the like in an expression. */
+ char *save;
+ char *gotfree_input_line;
+ int adjust;
- /* 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)
- {
- 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;
- }
+ save = input_line_pointer;
+ gotfree_input_line = lex_got (&got_reloc, &adjust);
+ if (gotfree_input_line)
+ input_line_pointer = gotfree_input_line;
- /* Skip the ':' and whitespace. */
- ++op_string;
- if (is_space_char (*op_string))
- ++op_string;
+ expression (exp);
- if (!is_digit_char (*op_string)
- && !is_identifier_char (*op_string)
- && *op_string != '('
- && *op_string != ABSOLUTE_PREFIX)
- {
- 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)
+ if (gotfree_input_line)
{
- as_bad (_("Junk `%s' after register"), op_string);
- return 0;
+ /* expression () has merrily parsed up to the end of line,
+ or a comma - in the wrong buffer. Transfer how far
+ input_line_pointer has moved to the right buffer. */
+ input_line_pointer = (save
+ + (input_line_pointer - gotfree_input_line)
+ + adjust);
+ free (gotfree_input_line);
}
- i.types[this_operand] |= r->reg_type & ~BaseIndex;
- i.regs[this_operand] = r;
- i.reg_operands++;
}
- else if (*op_string == REGISTER_PREFIX)
+ else
+ expression (exp);
+}
+#endif
+
+static int i386_immediate PARAMS ((char *));
+
+static int
+i386_immediate (imm_start)
+ char *imm_start;
+{
+ char *save_input_line_pointer;
+#ifndef LEX_AT
+ char *gotfree_input_line;
+#endif
+ segT exp_seg = 0;
+ expressionS *exp;
+
+ if (i.imm_operands == MAX_IMMEDIATE_OPERANDS)
{
- as_bad (_("bad register name `%s'"), op_string);
+ as_bad (_("only 1 or 2 immediate operands are allowed"));
return 0;
}
- else if (*op_string == IMMEDIATE_PREFIX)
- { /* ... or an immediate */
- ++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;
+ exp = &im_expressions[i.imm_operands++];
+ i.op[this_operand].imms = exp;
- do_memory_reference:
- if ((i.mem_operands == 1
+ if (is_space_char (*imm_start))
+ ++imm_start;
+
+ save_input_line_pointer = input_line_pointer;
+ input_line_pointer = imm_start;
+
+#ifndef LEX_AT
+ gotfree_input_line = lex_got (&i.reloc[this_operand], NULL);
+ if (gotfree_input_line)
+ input_line_pointer = gotfree_input_line;
+#endif
+
+ exp_seg = expression (exp);
+
+ SKIP_WHITESPACE ();
+ if (*input_line_pointer)
+ as_bad (_("junk `%s' after expression"), input_line_pointer);
+
+ input_line_pointer = save_input_line_pointer;
+#ifndef LEX_AT
+ if (gotfree_input_line)
+ free (gotfree_input_line);
+#endif
+
+ if (exp->X_op == O_absent || exp->X_op == O_big)
+ {
+ /* 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;
+ }
+ else if (exp->X_op == O_constant)
+ {
+ /* Size it properly later. */
+ i.types[this_operand] |= Imm64;
+ /* If BFD64, sign extend val. */
+ if (!use_rela_relocations)
+ if ((exp->X_add_number & ~(((addressT) 2 << 31) - 1)) == 0)
+ exp->X_add_number = (exp->X_add_number ^ ((addressT) 1 << 31)) - ((addressT) 1 << 31);
+ }
+#if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT))
+ else if (1
+#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
+ && !bfd_is_com_section (exp_seg)
+#endif
+ )
+ {
+#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
+ else
+ {
+ /* This is an address. The size of the address will be
+ determined later, depending on destination register,
+ suffix, or the default for the section. */
+ i.types[this_operand] |= Imm8 | Imm16 | Imm32 | Imm32S | Imm64;
+ }
+
+ return 1;
+}
+
+static char *i386_scale PARAMS ((char *));
+
+static char *
+i386_scale (scale)
+ char *scale;
+{
+ offsetT val;
+ char *save = input_line_pointer;
+
+ input_line_pointer = scale;
+ val = get_absolute_expression ();
+
+ switch (val)
+ {
+ 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:
+ as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"),
+ scale);
+ input_line_pointer = save;
+ return NULL;
+ }
+ if (i.log2_scale_factor != 0 && i.index_reg == 0)
+ {
+ 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
+ }
+ scale = input_line_pointer;
+ input_line_pointer = save;
+ return scale;
+}
+
+static int i386_displacement PARAMS ((char *, char *));
+
+static int
+i386_displacement (disp_start, disp_end)
+ char *disp_start;
+ char *disp_end;
+{
+ expressionS *exp;
+ segT exp_seg = 0;
+ char *save_input_line_pointer;
+#ifndef LEX_AT
+ char *gotfree_input_line;
+#endif
+ int bigdisp = Disp32;
+
+ if (flag_code == CODE_64BIT)
+ {
+ if (i.prefix[ADDR_PREFIX] == 0)
+ bigdisp = Disp64;
+ }
+ else if ((flag_code == CODE_16BIT) ^ (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
+ gotfree_input_line = lex_got (&i.reloc[this_operand], NULL);
+ if (gotfree_input_line)
+ input_line_pointer = gotfree_input_line;
+#endif
+
+ exp_seg = expression (exp);
+
+ SKIP_WHITESPACE ();
+ if (*input_line_pointer)
+ as_bad (_("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;
+#ifndef LEX_AT
+ if (gotfree_input_line)
+ free (gotfree_input_line);
+#endif
+
+#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.reloc[this_operand] == BFD_RELOC_386_GOTOFF
+ || i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
+ {
+ if (exp->X_op != O_symbol)
+ {
+ as_bad (_("bad expression used with @%s"),
+ (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL
+ ? "GOTPCREL"
+ : "GOTOFF"));
+ return 0;
+ }
+
+ 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));
+ exp->X_op = O_subtract;
+ exp->X_op_symbol = GOT_symbol;
+ if (i.reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL)
+ i.reloc[this_operand] = BFD_RELOC_32_PCREL;
+ else
+ i.reloc[this_operand] = BFD_RELOC_32;
+ }
+#endif
+
+ 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
+ else if (flag_code == CODE_64BIT)
+ i.types[this_operand] |= Disp32S | Disp32;
+ 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;
+{
+ int ok;
+#if INFER_ADDR_PREFIX
+ int fudged = 0;
+
+ tryprefix:
+#endif
+ ok = 1;
+ if (flag_code == CODE_64BIT)
+ {
+ if (i.prefix[ADDR_PREFIX] == 0)
+ {
+ /* 64bit checks. */
+ if ((i.base_reg
+ && ((i.base_reg->reg_type & Reg64) == 0)
+ && (i.base_reg->reg_type != BaseIndex
+ || i.index_reg))
+ || (i.index_reg
+ && ((i.index_reg->reg_type & (Reg64 | BaseIndex))
+ != (Reg64 | BaseIndex))))
+ ok = 0;
+ }
+ else
+ {
+ /* 32bit checks. */
+ if ((i.base_reg
+ && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
+ || (i.index_reg
+ && ((i.index_reg->reg_type & (Reg32 | BaseIndex | RegRex))
+ != (Reg32 | BaseIndex))))
+ ok = 0;
+ }
+ }
+ else
+ {
+ if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0))
+ {
+ /* 16bit checks. */
+ if ((i.base_reg
+ && ((i.base_reg->reg_type & (Reg16 | BaseIndex | RegRex))
+ != (Reg16 | BaseIndex)))
+ || (i.index_reg
+ && (((i.index_reg->reg_type & (Reg16 | BaseIndex))
+ != (Reg16 | BaseIndex))
+ || !(i.base_reg
+ && i.base_reg->reg_num < 6
+ && i.index_reg->reg_num >= 6
+ && i.log2_scale_factor == 0))))
+ ok = 0;
+ }
+ else
+ {
+ /* 32bit checks. */
+ if ((i.base_reg
+ && (i.base_reg->reg_type & (Reg32 | RegRex)) != Reg32)
+ || (i.index_reg
+ && ((i.index_reg->reg_type & (Reg32 | BaseIndex | RegRex))
+ != (Reg32 | BaseIndex))))
+ ok = 0;
+ }
+ }
+ if (!ok)
+ {
+#if INFER_ADDR_PREFIX
+ if (flag_code != CODE_64BIT
+ && i.prefix[ADDR_PREFIX] == 0 && stackop_size != '\0')
+ {
+ i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE;
+ i.prefixes += 1;
+ /* Change the size of any displacement too. At most one of
+ Disp16 or Disp32 is set.
+ FIXME. There doesn't seem to be any real need for separate
+ Disp16 and Disp32 flags. The same goes for Imm16 and Imm32.
+ Removing them would probably clean up the code quite a lot. */
+ if (i.types[this_operand] & (Disp16 | Disp32))
+ i.types[this_operand] ^= (Disp16 | Disp32);
+ fudged = 1;
+ goto tryprefix;
+ }
+ if (fudged)
+ as_bad (_("`%s' is not a valid base/index expression"),
+ operand_string);
+ else
+#endif
+ as_bad (_("`%s' is not a valid %s bit base/index expression"),
+ operand_string,
+ flag_code_names[flag_code]);
+ 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;
+ }
+
+ /* 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)
+ {
+ 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;
+ }
+
+ /* 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 (_("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;
+ 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);
+
+ --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
+ {
+ base_string--;
+ if (*base_string == ')')
+ parens_balanced++;
+ if (*base_string == '(')
+ parens_balanced--;
+ }
+ while (parens_balanced);
+
+ 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))
+ {
+ displacement_string_end = temp_string;
+
+ i.types[this_operand] |= BaseIndex;
+
+ if (i.base_reg)
+ {
+ base_string = end_op;
+ if (is_space_char (*base_string))
+ ++base_string;
+ }
+
+ /* There may be an index reg or scale factor here. */
+ if (*base_string == ',')
+ {
+ ++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 (*base_string != ')')
+ {
+ char *end_scale = i386_scale (base_string);
+
+ if (!end_scale)
+ return 0;
+
+ base_string = end_scale;
+ 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 (*base_string != ')')
+ {
+ as_bad (_("expecting `,' or `)' after base register in `%s'"),
+ operand_string);
+ return 0;
+ }
+ }
+ else if (*base_string == REGISTER_PREFIX)
+ {
+ as_bad (_("bad register name `%s'"), base_string);
+ return 0;
+ }
+ }
+
+ /* 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)
+ {
+ 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
+ && i.log2_scale_factor == 0
+ && i.seg[i.mem_operands] == 0
+ && (i.types[this_operand] & Disp) == 0)
+ {
+ i.types[this_operand] = InOutPortReg;
+ return 1;
+ }
+
+ 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 beginning operand %d `%s'"),
+ output_invalid (*op_string),
+ this_operand + 1,
+ op_string);
+ return 0;
+ }
+ return 1; /* Normal return. */
+}
+\f
+/* 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)
+ fragS *fragP;
+ segT 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, or we need to keep a
+ reloc so that weak symbols can be overridden. */
+ int size = (fragP->fr_subtype & CODE16) ? 2 : 4;
+ RELOC_ENUM reloc_type;
+ 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 (TYPE_FROM_RELAX_STATE (fragP->fr_subtype))
+ {
+ case UNCOND_JUMP:
+ /* Make jmp (0xeb) a (d)word displacement jump. */
+ opcode[0] = 0xe9;
+ fragP->fr_fix += size;
+ fix_new (fragP, old_fr_fix, size,
+ fragP->fr_symbol,
+ fragP->fr_offset, 1,
+ reloc_type);
+ break;
+
+ case COND_JUMP86:
+ if (size == 2
+ && (!no_cond_jump_promotion || fragP->fr_var != NO_RELOC))
+ {
+ /* Negate the condition, and branch past an
+ unconditional jump. */
+ opcode[0] ^= 1;
+ opcode[1] = 3;
+ /* Insert an unconditional jump. */
+ opcode[2] = 0xe9;
+ /* We added two extra opcode bytes, and have a two byte
+ offset. */
+ fragP->fr_fix += 2 + 2;
+ fix_new (fragP, old_fr_fix + 2, 2,
+ fragP->fr_symbol,
+ fragP->fr_offset, 1,
+ reloc_type);
+ break;
+ }
+ /* Fall through. */
+
+ case COND_JUMP:
+ if (no_cond_jump_promotion && fragP->fr_var == NO_RELOC)
+ {
+ fragP->fr_fix += 1;
+ fix_new (fragP, old_fr_fix, 1,
+ fragP->fr_symbol,
+ fragP->fr_offset, 1,
+ BFD_RELOC_8_PCREL);
+ break;
+ }
+
+ /* This changes the byte-displacement jump 0x7N
+ to the (d)word-displacement jump 0x0f,0x8N. */
+ opcode[1] = opcode[0] + 0x10;
+ 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,
+ fragP->fr_offset, 1,
+ reloc_type);
+ break;
+
+ default:
+ BAD_CASE (fragP->fr_subtype);
+ break;
+ }
+ frag_wane (fragP);
+ return fragP->fr_fix - old_fr_fix;
+ }
+
+ /* Guess size depending on current relax state. Initially the relax
+ state will correspond to a short jump and we return 1, because
+ the variable part of the frag (the branch offset) is one byte
+ long. However, we can relax a section more than once and in that
+ case we must either set fr_subtype back to the unrelaxed state,
+ or return the value for the appropriate branch. */
+ return md_relax_table[fragP->fr_subtype].rlx_length;
+}
+
+/* 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, sec, fragP)
+ object_headers *headers ATTRIBUTE_UNUSED;
+ segT sec ATTRIBUTE_UNUSED;
+ fragS *fragP;
+#else
+void
+md_convert_frag (abfd, sec, fragP)
+ bfd *abfd ATTRIBUTE_UNUSED;
+ segT sec ATTRIBUTE_UNUSED;
+ fragS *fragP;
+#endif
+{
+ unsigned char *opcode;
+ unsigned char *where_to_put_displacement = NULL;
+ offsetT target_address;
+ offsetT opcode_address;
+ unsigned int extension = 0;
+ offsetT displacement_from_opcode_start;
+
+ opcode = (unsigned char *) fragP->fr_opcode;
+
+ /* Address we want to reach in file space. */
+ target_address = S_GET_VALUE (fragP->fr_symbol) + fragP->fr_offset;
+
+ /* 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 = target_address - opcode_address;
+
+ if ((fragP->fr_subtype & BIG) == 0)
+ {
+ /* Don't have to change opcode. */
+ extension = 1; /* 1 opcode + 1 displacement */
+ where_to_put_displacement = &opcode[1];
+ }
+ else
+ {
+ if (no_cond_jump_promotion
+ && TYPE_FROM_RELAX_STATE (fragP->fr_subtype) != UNCOND_JUMP)
+ as_warn_where (fragP->fr_file, fragP->fr_line, _("long jump required"));
+
+ switch (fragP->fr_subtype)
+ {
+ 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 (UNCOND_JUMP, BIG16):
+ extension = 2; /* 1 opcode + 2 displacement */
+ opcode[0] = 0xe9;
+ where_to_put_displacement = &opcode[1];
+ break;
+
+ case ENCODE_RELAX_STATE (COND_JUMP, BIG):
+ case ENCODE_RELAX_STATE (COND_JUMP86, 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 (COND_JUMP, BIG16):
+ extension = 3; /* 2 opcode + 2 displacement */
+ opcode[1] = opcode[0] + 0x10;
+ opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
+ where_to_put_displacement = &opcode[2];
+ break;
+
+ case ENCODE_RELAX_STATE (COND_JUMP86, BIG16):
+ extension = 4;
+ opcode[0] ^= 1;
+ opcode[1] = 3;
+ opcode[2] = 0xe9;
+ where_to_put_displacement = &opcode[3];
+ break;
+
+ default:
+ BAD_CASE (fragP->fr_subtype);
+ break;
+ }
+ }
+
+ /* Now put displacement after opcode. */
+ md_number_to_chars ((char *) where_to_put_displacement,
+ (valueT) (displacement_from_opcode_start - extension),
+ DISP_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;
+
+/* 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 ATTRIBUTE_UNUSED;
+ symbolS *to_symbol ATTRIBUTE_UNUSED;
+{
+ offsetT offset;
+
+ offset = to_addr - (from_addr + 2);
+ /* 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 ATTRIBUTE_UNUSED;
+ symbolS *to_symbol ATTRIBUTE_UNUSED;
+{
+ offsetT offset;
+
+ offset = to_addr - (from_addr + 5);
+ md_number_to_chars (ptr, (valueT) 0xe9, 1);
+ md_number_to_chars (ptr + 1, (valueT) offset, 4);
+}
+\f
+/* 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.
+
+ On the 386, immediates, displacements, and data pointers are all in
+ the same (little-endian) format, so we don't need to care about which
+ we are handling. */
+
+void
+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;
+{
+ char *p = fixP->fx_where + fixP->fx_frag->fr_literal;
+ valueT value = * valP;
+
+#if defined (BFD_ASSEMBLER) && !defined (TE_Mach)
+ if (fixP->fx_pcrel)
+ {
+ switch (fixP->fx_r_type)
+ {
+ default:
+ break;
+
+ 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;
+ }
+ }
+
+ /* 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 && !use_rela_relocations)
+ {
+#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
+ }
+
+ /* 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:
+ case BFD_RELOC_X86_64_PLT32:
+ /* Make the jump instruction point to the address of the operand. At
+ runtime we merely add the offset to the actual PLT entry. */
+ value = -4;
+ 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:
+ case BFD_RELOC_386_TLS_GD:
+ case BFD_RELOC_386_TLS_LDM:
+ case BFD_RELOC_386_TLS_LDO_32:
+ case BFD_RELOC_386_TLS_IE_32:
+ case BFD_RELOC_386_TLS_LE_32:
+ case BFD_RELOC_386_TLS_LE:
+ case BFD_RELOC_X86_64_GOT32:
+ value = 0; /* Fully resolved at runtime. No addend. */
+ break;
+ case BFD_RELOC_386_GOTOFF:
+ case BFD_RELOC_X86_64_GOTPCREL:
+ break;
+
+ case BFD_RELOC_VTABLE_INHERIT:
+ case BFD_RELOC_VTABLE_ENTRY:
+ fixP->fx_done = 0;
+ return;
+
+ default:
+ break;
+ }
+#endif /* defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) */
+ * valP = value;
+#endif /* defined (BFD_ASSEMBLER) && !defined (TE_Mach) */
+
+ /* Are we finished with this relocation now? */
+ if (fixP->fx_addsy == NULL && fixP->fx_pcrel == 0)
+ fixP->fx_done = 1;
+#ifdef BFD_ASSEMBLER
+ else if (use_rela_relocations)
+ {
+ fixP->fx_no_overflow = 1;
+ /* Remember value for tc_gen_reloc. */
+ fixP->fx_addnumber = value;
+ value = 0;
+ }
+#endif
+ md_number_to_chars (p, value, fixP->fx_size);
+}
+\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. */
+
+char *
+md_atof (type, litP, sizeP)
+ int type;
+ char *litP;
+ int *sizeP;
+{
+ int prec;
+ LITTLENUM_TYPE words[MAX_LITTLENUMS];
+ LITTLENUM_TYPE *wordP;
+ char *t;
+
+ switch (type)
+ {
+ case 'f':
+ case 'F':
+ prec = 2;
+ break;
+
+ case 'd':
+ case 'D':
+ prec = 4;
+ break;
+
+ case 'x':
+ case 'X':
+ prec = 5;
+ break;
+
+ default:
+ *sizeP = 0;
+ return _("Bad call to md_atof ()");
+ }
+ t = atof_ieee (input_line_pointer, type, words);
+ if (t)
+ input_line_pointer = t;
+
+ *sizeP = prec * sizeof (LITTLENUM_TYPE);
+ /* This loops outputs the LITTLENUMs in REVERSE order; in accord with
+ the bigendian 386. */
+ for (wordP = words + prec - 1; prec--;)
+ {
+ md_number_to_chars (litP, (valueT) (*wordP--), sizeof (LITTLENUM_TYPE));
+ litP += sizeof (LITTLENUM_TYPE);
+ }
+ return 0;
+}
+\f
+char output_invalid_buf[8];
+
+static char *
+output_invalid (c)
+ int c;
+{
+ if (ISPRINT (c))
+ sprintf (output_invalid_buf, "'%c'", c);
+ else
+ sprintf (output_invalid_buf, "(0x%x)", (unsigned) c);
+ return output_invalid_buf;
+}
+
+/* REG_STRING starts *before* REGISTER_PREFIX. */
+
+static const reg_entry *
+parse_register (reg_string, end_op)
+ char *reg_string;
+ char **end_op;
+{
+ char *s = reg_string;
+ char *p;
+ char reg_name_given[MAX_REG_NAME_SIZE + 1];
+ const reg_entry *r;
+
+ /* 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')
+ {
+ if (p >= reg_name_given + MAX_REG_NAME_SIZE)
+ 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;
+ }
+ }
+
+ if (r != NULL
+ && (r->reg_flags & (RegRex64 | RegRex)) != 0
+ && flag_code != CODE_64BIT)
+ {
+ return (const reg_entry *) NULL;
+ }
+
+ 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[] = {
+#define OPTION_32 (OPTION_MD_BASE + 0)
+ {"32", no_argument, NULL, OPTION_32},
+#if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
+#define OPTION_64 (OPTION_MD_BASE + 1)
+ {"64", no_argument, NULL, OPTION_64},
+#endif
+ {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;
+
+#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;
+
+ /* -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;
+
+ case OPTION_64:
+ {
+ const char **list, **l;
+
+ list = bfd_target_list ();
+ for (l = list; *l != NULL; l++)
+ if (strcmp (*l, "elf64-x86-64") == 0)
+ {
+ default_arch = "x86_64";
+ break;
+ }
+ if (*l == NULL)
+ as_fatal (_("No compiled in support for x86_64"));
+ free (list);
+ }
+ break;
+#endif
+
+ case OPTION_32:
+ default_arch = "i386";
+ break;
+
+ default:
+ return 0;
+ }
+ return 1;
+}
+
+void
+md_show_usage (stream)
+ FILE *stream;
+{
+#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
+}
+
+#ifdef BFD_ASSEMBLER
+#if ((defined (OBJ_MAYBE_COFF) && defined (OBJ_MAYBE_AOUT)) \
+ || defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
+
+/* Pick the target format to use. */
+
+const char *
+i386_target_format ()
+{
+ if (!strcmp (default_arch, "x86_64"))
+ set_code_flag (CODE_64BIT);
+ else if (!strcmp (default_arch, "i386"))
+ set_code_flag (CODE_32BIT);
+ else
+ as_fatal (_("Unknown architecture"));
+ switch (OUTPUT_FLAVOR)
+ {
+#ifdef OBJ_MAYBE_AOUT
+ case bfd_target_aout_flavour:
+ return AOUT_TARGET_FORMAT;
+#endif
+#ifdef OBJ_MAYBE_COFF
+ case bfd_target_coff_flavour:
+ return "coff-i386";
+#endif
+#if defined (OBJ_MAYBE_ELF) || defined (OBJ_ELF)
+ case bfd_target_elf_flavour:
+ {
+ if (flag_code == CODE_64BIT)
+ use_rela_relocations = 1;
+ return flag_code == CODE_64BIT ? "elf64-x86-64" : "elf32-i386";
+ }
+#endif
+ default:
+ abort ();
+ return NULL;
+ }
+}
+
+#endif /* OBJ_MAYBE_ more than one */
+
+#if (defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF))
+void i386_elf_emit_arch_note ()
+{
+ if (OUTPUT_FLAVOR == bfd_target_elf_flavour
+ && cpu_arch_name != NULL)
+ {
+ char *p;
+ asection *seg = now_seg;
+ subsegT subseg = now_subseg;
+ Elf_Internal_Note i_note;
+ Elf_External_Note e_note;
+ asection *note_secp;
+ int len;
+
+ /* Create the .note section. */
+ note_secp = subseg_new (".note", 0);
+ bfd_set_section_flags (stdoutput,
+ note_secp,
+ SEC_HAS_CONTENTS | SEC_READONLY);
+
+ /* Process the arch string. */
+ len = strlen (cpu_arch_name);
+
+ i_note.namesz = len + 1;
+ i_note.descsz = 0;
+ i_note.type = NT_ARCH;
+ p = frag_more (sizeof (e_note.namesz));
+ md_number_to_chars (p, (valueT) i_note.namesz, sizeof (e_note.namesz));
+ p = frag_more (sizeof (e_note.descsz));
+ md_number_to_chars (p, (valueT) i_note.descsz, sizeof (e_note.descsz));
+ p = frag_more (sizeof (e_note.type));
+ md_number_to_chars (p, (valueT) i_note.type, sizeof (e_note.type));
+ p = frag_more (len + 1);
+ strcpy (p, cpu_arch_name);
+
+ frag_align (2, 0, 0);
+
+ subseg_set (seg, subseg);
+ }
+}
+#endif
+#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 ATTRIBUTE_UNUSED;
+ valueT size;
+{
+#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;
+}
+
+/* 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;
+{
+ return fixP->fx_size + fixP->fx_where + fixP->fx_frag->fr_address;
+}
+
+#ifndef I386COFF
+
+static void
+s_bss (ignore)
+ int ignore ATTRIBUTE_UNUSED;
+{
+ int temp;
+
+ temp = get_absolute_expression ();
+ subseg_set (bss_section, (subsegT) temp);
+ 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)
+ {
+ /* GOTOFF relocation are nonsense in 64bit mode. */
+ if (fixp->fx_r_type == BFD_RELOC_32_PCREL)
+ {
+ if (flag_code != CODE_64BIT)
+ abort ();
+ fixp->fx_r_type = BFD_RELOC_X86_64_GOTPCREL;
+ }
+ else
+ {
+ if (flag_code == CODE_64BIT)
+ abort ();
+ fixp->fx_r_type = BFD_RELOC_386_GOTOFF;
+ }
+ fixp->fx_subsy = 0;
+ }
+}
+
+arelent *
+tc_gen_reloc (section, fixp)
+ asection *section ATTRIBUTE_UNUSED;
+ fixS *fixp;
+{
+ arelent *rel;
+ bfd_reloc_code_real_type code;
+
+ switch (fixp->fx_r_type)
+ {
+ case BFD_RELOC_X86_64_PLT32:
+ case BFD_RELOC_X86_64_GOT32:
+ case BFD_RELOC_X86_64_GOTPCREL:
+ case BFD_RELOC_386_PLT32:
+ case BFD_RELOC_386_GOT32:
+ case BFD_RELOC_386_GOTOFF:
+ case BFD_RELOC_386_GOTPC:
+ case BFD_RELOC_386_TLS_GD:
+ case BFD_RELOC_386_TLS_LDM:
+ case BFD_RELOC_386_TLS_LDO_32:
+ case BFD_RELOC_386_TLS_IE_32:
+ case BFD_RELOC_386_TLS_LE_32:
+ case BFD_RELOC_386_TLS_LE:
+ case BFD_RELOC_X86_64_32S:
+ case BFD_RELOC_RVA:
+ case BFD_RELOC_VTABLE_ENTRY:
+ case BFD_RELOC_VTABLE_INHERIT:
+ code = fixp->fx_r_type;
+ break;
+ default:
+ if (fixp->fx_pcrel)
+ {
+ switch (fixp->fx_size)
+ {
+ default:
+ as_bad_where (fixp->fx_file, fixp->fx_line,
+ _("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_where (fixp->fx_file, fixp->fx_line,
+ _("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;
+#ifdef BFD64
+ case 8: code = BFD_RELOC_64; break;
+#endif
+ }
}
+ break;
+ }
- /* 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);
+ if (code == BFD_RELOC_32
+ && GOT_symbol
+ && fixp->fx_addsy == GOT_symbol)
+ {
+ /* We don't support GOTPC on 64bit targets. */
+ if (flag_code == CODE_64BIT)
+ abort ();
+ code = BFD_RELOC_386_GOTPC;
+ }
- --base_string;
- if (is_space_char (*base_string))
- --base_string;
+ rel = (arelent *) xmalloc (sizeof (arelent));
+ rel->sym_ptr_ptr = (asymbol **) xmalloc (sizeof (asymbol *));
+ *rel->sym_ptr_ptr = symbol_get_bfdsym (fixp->fx_addsy);
- /* 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;
+ rel->address = fixp->fx_frag->fr_address + fixp->fx_where;
+ if (!use_rela_relocations)
+ {
+ /* HACK: Since i386 ELF uses Rel instead of Rela, encode the
+ vtable entry to be used in the relocation's section offset. */
+ if (fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY)
+ rel->address = fixp->fx_offset;
- if (*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
- {
- base_string--;
- if (*base_string == ')')
- parens_balanced++;
- if (*base_string == '(')
- parens_balanced--;
- }
- while (parens_balanced);
+ if (fixp->fx_pcrel)
+ rel->addend = fixp->fx_addnumber;
+ else
+ rel->addend = 0;
+ }
+ /* Use the rela in 64bit mode. */
+ else
+ {
+ if (!fixp->fx_pcrel)
+ rel->addend = fixp->fx_offset;
+ else
+ switch (code)
+ {
+ case BFD_RELOC_X86_64_PLT32:
+ case BFD_RELOC_X86_64_GOT32:
+ case BFD_RELOC_X86_64_GOTPCREL:
+ rel->addend = fixp->fx_offset - fixp->fx_size;
+ break;
+ default:
+ rel->addend = (section->vma
+ - fixp->fx_size
+ + fixp->fx_addnumber
+ + md_pcrel_from (fixp));
+ break;
+ }
+ }
- temp_string = base_string;
+ 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);
+ }
- /* Skip past '(' and whitespace. */
- ++base_string;
- if (is_space_char (*base_string))
- ++base_string;
+ return rel;
+}
- if (*base_string == ','
- || ((*base_string == REGISTER_PREFIX || allow_naked_reg)
- && (i.base_reg = parse_register (base_string, &end_op)) != NULL))
- {
- displacement_string_end = temp_string;
+#else /* !BFD_ASSEMBLER */
- i.types[this_operand] |= BaseIndex;
+#if (defined(OBJ_AOUT) | defined(OBJ_BOUT))
+void
+tc_aout_fix_to_chars (where, fixP, segment_address_in_file)
+ char *where;
+ 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. */
- if (i.base_reg)
- {
- base_string = end_op;
- if (is_space_char (*base_string))
- ++base_string;
- }
+ static const unsigned char nbytes_r_length[] = { 42, 0, 1, 42, 2 };
+ long r_symbolnum;
- /* There may be an index reg or scale factor here. */
- if (*base_string == ',')
- {
- ++base_string;
- if (is_space_char (*base_string))
- ++base_string;
+ know (fixP->fx_addsy != NULL);
- 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;
- }
+ md_number_to_chars (where,
+ (valueT) (fixP->fx_frag->fr_address
+ + fixP->fx_where - segment_address_in_file),
+ 4);
- /* Check for scale factor. */
- if (isdigit ((unsigned char) *base_string))
- {
- if (!i386_scale (base_string))
- return 0;
+ r_symbolnum = (S_IS_DEFINED (fixP->fx_addsy)
+ ? S_GET_TYPE (fixP->fx_addsy)
+ : fixP->fx_addsy->sy_number);
- ++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 (*base_string != ')')
- {
- as_bad (_("expecting `,' or `)' after base register in `%s'"),
- operand_string);
- return 0;
- }
- }
- else if (*base_string == REGISTER_PREFIX)
- {
- as_bad (_("bad register name `%s'"), base_string);
- return 0;
- }
- }
+ where[6] = (r_symbolnum >> 16) & 0x0ff;
+ where[5] = (r_symbolnum >> 8) & 0x0ff;
+ where[4] = r_symbolnum & 0x0ff;
+ where[7] = ((((!S_IS_DEFINED (fixP->fx_addsy)) << 3) & 0x08)
+ | ((nbytes_r_length[fixP->fx_size] << 1) & 0x06)
+ | (((fixP->fx_pcrel << 0) & 0x01) & 0x0f));
+}
- /* 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)
- {
- if (!i386_displacement (displacement_string_start,
- displacement_string_end))
- return 0;
- }
+#endif /* OBJ_AOUT or OBJ_BOUT. */
- /* 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;
- return 1;
- }
+#if defined (I386COFF)
- 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 beginning operand %d `%s'"),
- output_invalid (*op_string),
- this_operand + 1,
- op_string);
- return 0;
- }
- return 1; /* normal return */
+short
+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 :
+ R_PCRLONG) :
+ (fixP->fx_size == 1 ? R_RELBYTE :
+ fixP->fx_size == 2 ? R_RELWORD :
+ R_DIR32));
}
-\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.
- */
+
int
-md_estimate_size_before_relax (fragP, segment)
- register fragS *fragP;
- register segT segment;
+tc_coff_sizemachdep (frag)
+ fragS *frag;
{
- register unsigned char *opcode;
- register int old_fr_fix;
+ if (frag->fr_next)
+ return (frag->fr_next->fr_address - frag->fr_address);
+ else
+ return 0;
+}
+
+#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
- 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)
- {
- /* symbol is undefined in this segment */
- int code16 = fragP->fr_subtype & CODE16;
- int size = code16 ? 2 : 4;
-#ifdef BFD_ASSEMBLER
- enum bfd_reloc_code_real reloc_type;
-#else
- int reloc_type;
-#endif
+ e09 OFFSET e10
+ | e09 PTR e10
+ | e09 : e10
+ | e10
- if (GOT_symbol /* Not quite right - we should switch on presence of
- @PLT, but I cannot see how to get to that from
- here. We should have done this in md_assemble to
- really get it right all of the time, but I think it
- does not matter that much, as this will be right
- most of the time. ERY */
- && S_GET_SEGMENT(fragP->fr_symbol) == undefined_section)
- reloc_type = BFD_RELOC_386_PLT32;
- else if (code16)
- reloc_type = BFD_RELOC_16_PCREL;
- else
- reloc_type = BFD_RELOC_32_PCREL;
+ e10 e10 [ expr ]
+ | e11
- switch (opcode[0])
- {
- case JUMP_PC_RELATIVE: /* make jmp (0xeb) a dword displacement jump */
- opcode[0] = 0xe9; /* dword disp jmp */
- fragP->fr_fix += size;
- fix_new (fragP, old_fr_fix, size,
- fragP->fr_symbol,
- fragP->fr_offset, 1,
- reloc_type);
- break;
+ e11 ( expr )
+ | [ expr ]
+ | constant
+ | dataType
+ | id
+ | $
+ | register
- default:
- /* 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;
- fragP->fr_fix += 1 + size; /* we've added an opcode byte */
- fix_new (fragP, old_fr_fix + 1, size,
- fragP->fr_symbol,
- fragP->fr_offset, 1,
- reloc_type);
- break;
- }
- frag_wane (fragP);
- }
- 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".
- */
-#ifndef BFD_ASSEMBLER
-void
-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 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;
- unsigned int extension = 0;
- int displacement_from_opcode_start;
+ => expr SHORT e05
+ | e05
- opcode = (unsigned char *) fragP->fr_opcode;
+ gpRegister AX | EAX | BX | EBX | CX | ECX | DX | EDX
+ | BP | EBP | SP | ESP | DI | EDI | SI | ESI
- /* 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
+ hexdigit a | b | c | d | e | f
+ | A | B | C | D | E | F
- /* Address opcode resides at in file space. */
- opcode_address = fragP->fr_address + fragP->fr_fix;
+ id alpha
+ | id alpha
+ | id decdigit
- /* Displacement from opcode start to fill into instruction. */
- displacement_from_opcode_start = target_address - opcode_address;
+ mulOp * | / | MOD
- switch (fragP->fr_subtype)
- {
- 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;
+ quote " | '
- 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;
+ register specialRegister
+ | gpRegister
+ | byteRegister
- case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG):
- extension = 4; /* 1 opcode + 4 displacement */
- opcode[0] = 0xe9;
- where_to_put_displacement = &opcode[1];
- break;
+ segmentRegister CS | DS | ES | FS | GS | SS
- case ENCODE_RELAX_STATE (COND_JUMP, BIG16):
- extension = 3; /* 2 opcode + 2 displacement */
- opcode[1] = opcode[0] + 0x10;
- opcode[0] = TWO_BYTE_OPCODE_ESCAPE;
- where_to_put_displacement = &opcode[2];
- break;
+ specialRegister CR0 | CR2 | CR3
+ | DR0 | DR1 | DR2 | DR3 | DR6 | DR7
+ | TR3 | TR4 | TR5 | TR6 | TR7
- case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG16):
- extension = 2; /* 1 opcode + 2 displacement */
- opcode[0] = 0xe9;
- where_to_put_displacement = &opcode[1];
- break;
+ 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.
- default:
- BAD_CASE (fragP->fr_subtype);
- break;
- }
- /* now put displacement after opcode */
- md_number_to_chars ((char *) where_to_put_displacement,
- (valueT) (displacement_from_opcode_start - extension),
- SIZE_FROM_RELAX_STATE (fragP->fr_subtype));
- fragP->fr_fix += extension;
-}
-\f
+ expr SHORT e05
+ | e05
-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 */
+ e05 e06 e05'
-void
-md_create_short_jump (ptr, from_addr, to_addr, frag, to_symbol)
- char *ptr;
- addressT from_addr, to_addr;
- fragS *frag ATTRIBUTE_UNUSED;
- symbolS *to_symbol ATTRIBUTE_UNUSED;
-{
- long offset;
+ e05' addOp e06 e05'
+ | Empty
- offset = to_addr - (from_addr + 2);
- md_number_to_chars (ptr, (valueT) 0xeb, 1); /* opcode for byte-disp jump */
- md_number_to_chars (ptr + 1, (valueT) offset, 1);
-}
+ e06 e09 e06'
-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;
-{
- long offset;
+ e06' mulOp e09 e06'
+ | Empty
- if (flag_do_long_jump)
- {
- offset = to_addr - S_GET_VALUE (to_symbol);
- md_number_to_chars (ptr, (valueT) 0xe9, 1);/* opcode for long jmp */
- md_number_to_chars (ptr + 1, (valueT) offset, 4);
- fix_new (frag, (ptr + 1) - frag->fr_literal, 4,
- to_symbol, (offsetT) 0, 0, BFD_RELOC_32);
- }
- else
- {
- offset = to_addr - (from_addr + 5);
- md_number_to_chars (ptr, (valueT) 0xe9, 1);
- md_number_to_chars (ptr + 1, (valueT) offset, 4);
- }
-}
-\f
-/* 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.
+ e09 OFFSET e10 e09'
+ | e10 e09'
- On the 386, immediates, displacements, and data pointers are all in
- the same (little-endian) format, so we don't need to care about which
- we are handling. */
+ e09' PTR e10 e09'
+ | : e10 e09'
+ | Empty
-int
-md_apply_fix3 (fixP, valp, seg)
- fixS *fixP; /* The fix we're to put in. */
- valueT *valp; /* Pointer to the value of the bits. */
- segT seg ATTRIBUTE_UNUSED; /* Segment fix is from. */
+ e10 e11 e10'
+
+ e10' [ expr ] e10'
+ | Empty
+
+ e11 ( expr )
+ | [ expr ]
+ | BYTE
+ | WORD
+ | DWORD
+ | QWORD
+ | XWORD
+ | .
+ | $
+ | register
+ | id
+ | constant */
+
+/* Parsing structure for the intel syntax parser. Used to implement the
+ semantic actions for the operand grammar. */
+struct intel_parser_s
+ {
+ char *op_string; /* The string being parsed. */
+ int got_a_float; /* Whether the operand is a float. */
+ int op_modifier; /* Operand modifier. */
+ int is_mem; /* 1 if operand is memory reference. */
+ const reg_entry *reg; /* Last register reference found. */
+ char *disp; /* Displacement string being built. */
+ };
+
+static struct intel_parser_s intel_parser;
+
+/* Token structure for parsing intel syntax. */
+struct intel_token
+ {
+ int code; /* Token code. */
+ const reg_entry *reg; /* Register entry for register tokens. */
+ char *str; /* String representation. */
+ };
+
+static struct intel_token cur_token, prev_token;
+
+/* Token codes for the intel parser. Since T_SHORT is already used
+ by COFF, undefine it first to prevent a warning. */
+#define T_NIL -1
+#define T_CONST 1
+#define T_REG 2
+#define T_BYTE 3
+#define T_WORD 4
+#define T_DWORD 5
+#define T_QWORD 6
+#define T_XWORD 7
+#undef T_SHORT
+#define T_SHORT 8
+#define T_OFFSET 9
+#define T_PTR 10
+#define T_ID 11
+
+/* Prototypes for intel parser functions. */
+static int intel_match_token PARAMS ((int code));
+static void intel_get_token PARAMS ((void));
+static void intel_putback_token PARAMS ((void));
+static int intel_expr PARAMS ((void));
+static int intel_e05 PARAMS ((void));
+static int intel_e05_1 PARAMS ((void));
+static int intel_e06 PARAMS ((void));
+static int intel_e06_1 PARAMS ((void));
+static int intel_e09 PARAMS ((void));
+static int intel_e09_1 PARAMS ((void));
+static int intel_e10 PARAMS ((void));
+static int intel_e10_1 PARAMS ((void));
+static int intel_e11 PARAMS ((void));
+
+static int
+i386_intel_operand (operand_string, got_a_float)
+ char *operand_string;
+ int got_a_float;
{
- register char *p = fixP->fx_where + fixP->fx_frag->fr_literal;
- valueT value = *valp;
+ int ret;
+ char *p;
-#if defined (BFD_ASSEMBLER) && !defined (TE_Mach)
- if (fixP->fx_pcrel)
+ /* 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)
{
- switch (fixP->fx_r_type)
+ /* 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)
{
- default:
- break;
+ 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++;
- 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;
+ /* 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);
}
- /* 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)
+ 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)
{
-#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);
+ intel_parser.op_modifier = SHORT;
+ intel_match_token (T_SHORT);
- 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);
- else if (S_IS_EXTERNAL (fixP->fx_addsy)
- || S_IS_WEAK (fixP->fx_addsy))
- {
- /* We are generating an external relocation for this defined
- symbol. We add the address, because
- bfd_install_relocation will subtract it. VALUE already
- holds the symbol value, because fixup_segment added it
- in. We subtract it out, and then we subtract it out
- again because bfd_install_relocation will add it in
- again. */
- value += md_pcrel_from (fixP);
- value -= 2 * S_GET_VALUE (fixP->fx_addsy);
- }
-#endif
+ return (intel_e05 ());
}
-#ifdef TE_PE
- else if (fixP->fx_addsy != NULL
- && S_IS_DEFINED (fixP->fx_addsy)
- && (S_IS_EXTERNAL (fixP->fx_addsy)
- || S_IS_WEAK (fixP->fx_addsy)))
+
+ /* 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 == '-')
{
- /* We are generating an external relocation for this defined
- symbol. VALUE already holds the symbol value, and
- bfd_install_relocation will add it in again. We don't want
- either addition. */
- value -= 2 * S_GET_VALUE (fixP->fx_addsy);
+ strcat (intel_parser.disp, cur_token.str);
+ intel_match_token (cur_token.code);
+
+ return (intel_e06 () && intel_e05_1 ());
}
-#endif
- /* 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 = 0xfffffffc;
- 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;
+ /* e05' Empty */
+ else
+ return 1;
+}
- case BFD_RELOC_VTABLE_INHERIT:
- case BFD_RELOC_VTABLE_ENTRY:
- fixP->fx_done = 0;
- return 1;
+/* e06 e09 e06'
- default:
- break;
+ e06' mulOp e09 e06'
+ | Empty */
+static int
+intel_e06 ()
+{
+ return (intel_e09 () && intel_e06_1 ());
+}
+
+static int
+intel_e06_1 ()
+{
+ /* e06' mulOp e09 e06' */
+ if (cur_token.code == '*' || cur_token.code == '/')
+ {
+ strcat (intel_parser.disp, cur_token.str);
+ intel_match_token (cur_token.code);
+
+ return (intel_e09 () && intel_e06_1 ());
}
-#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;
+ /* e06' Empty */
+ else
+ return 1;
}
-#if 0
-/* This is never used. */
-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. */
+/* e09 OFFSET e10 e09'
+ | e10 e09'
+
+ e09' PTR e10 e09'
+ | : e10 e09'
+ | Empty */
+static int
+intel_e09 ()
{
- long retval;
- for (retval = 0, con += nbytes - 1; nbytes--; con--)
+ /* e09 OFFSET e10 e09' */
+ if (cur_token.code == T_OFFSET)
{
- retval <<= BITS_PER_CHAR;
- retval |= *con;
+ intel_parser.is_mem = 0;
+ intel_parser.op_modifier = OFFSET_FLAT;
+ intel_match_token (T_OFFSET);
+
+ return (intel_e10 () && intel_e09_1 ());
}
- return retval;
-}
-#endif /* 0 */
-\f
-#define MAX_LITTLENUMS 6
+ /* e09 e10 e09' */
+ else
+ return (intel_e10 () && intel_e09_1 ());
+}
-/* 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)
- int type;
- char *litP;
- int *sizeP;
+static int
+intel_e09_1 ()
{
- int prec;
- LITTLENUM_TYPE words[MAX_LITTLENUMS];
- LITTLENUM_TYPE *wordP;
- char *t;
-
- switch (type)
+ /* e09' PTR e10 e09' */
+ if (cur_token.code == T_PTR)
{
- case 'f':
- case 'F':
- prec = 2;
- break;
+ if (prev_token.code == T_BYTE)
+ i.suffix = BYTE_MNEM_SUFFIX;
- case 'd':
- case 'D':
- prec = 4;
- break;
+ 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;
+ }
- case 'x':
- case 'X':
- prec = 5;
- break;
+ 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;
+ }
- default:
- *sizeP = 0;
- return _("Bad call to md_atof ()");
+ else if (prev_token.code == T_QWORD)
+ {
+ if (intel_parser.got_a_float == 1) /* "f..." */
+ i.suffix = LONG_MNEM_SUFFIX;
+ else
+ i.suffix = QWORD_MNEM_SUFFIX;
+ }
+
+ else if (prev_token.code == T_XWORD)
+ i.suffix = LONG_DOUBLE_MNEM_SUFFIX;
+
+ else
+ {
+ as_bad (_("Unknown operand modifier `%s'\n"), prev_token.str);
+ return 0;
+ }
+
+ intel_match_token (T_PTR);
+
+ return (intel_e10 () && intel_e09_1 ());
}
- t = atof_ieee (input_line_pointer, type, words);
- if (t)
- input_line_pointer = t;
- *sizeP = prec * sizeof (LITTLENUM_TYPE);
- /* This loops outputs the LITTLENUMs in REVERSE order; in accord with
- the bigendian 386. */
- for (wordP = words + prec - 1; prec--;)
+ /* e09 : e10 e09' */
+ else if (cur_token.code == ':')
{
- md_number_to_chars (litP, (valueT) (*wordP--), sizeof (LITTLENUM_TYPE));
- litP += sizeof (LITTLENUM_TYPE);
+ /* Mark as a memory operand only if it's not already known to be an
+ offset expression. */
+ if (intel_parser.op_modifier != OFFSET_FLAT)
+ intel_parser.is_mem = 1;
+
+ return (intel_match_token (':') && intel_e10 () && intel_e09_1 ());
}
- return 0;
+
+ /* e09' Empty */
+ else
+ return 1;
}
-\f
-char output_invalid_buf[8];
-static char * output_invalid PARAMS ((int));
+/* e10 e11 e10'
-static char *
-output_invalid (c)
- int c;
+ e10' [ expr ] e10'
+ | Empty */
+static int
+intel_e10 ()
{
- if (isprint (c))
- sprintf (output_invalid_buf, "'%c'", c);
- else
- sprintf (output_invalid_buf, "(0x%x)", (unsigned) c);
- return output_invalid_buf;
+ return (intel_e11 () && intel_e10_1 ());
}
+static int
+intel_e10_1 ()
+{
+ /* e10' [ expr ] e10' */
+ if (cur_token.code == '[')
+ {
+ intel_match_token ('[');
-/* REG_STRING starts *before* REGISTER_PREFIX. */
+ /* Mark as a memory operand only if it's not already known to be an
+ offset expression. If it's an offset expression, we need to keep
+ the brace in. */
+ if (intel_parser.op_modifier != OFFSET_FLAT)
+ intel_parser.is_mem = 1;
+ else
+ strcat (intel_parser.disp, "[");
-static const reg_entry *
-parse_register (reg_string, end_op)
- char *reg_string;
- char **end_op;
+ /* Add a '+' to the displacement string if necessary. */
+ if (*intel_parser.disp != '\0'
+ && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
+ strcat (intel_parser.disp, "+");
+
+ if (intel_expr () && intel_match_token (']'))
+ {
+ /* Preserve brackets when the operand is an offset expression. */
+ if (intel_parser.op_modifier == OFFSET_FLAT)
+ strcat (intel_parser.disp, "]");
+
+ return intel_e10_1 ();
+ }
+ else
+ return 0;
+ }
+
+ /* e10' Empty */
+ else
+ return 1;
+}
+
+/* e11 ( expr )
+ | [ expr ]
+ | BYTE
+ | WORD
+ | DWORD
+ | QWORD
+ | XWORD
+ | $
+ | .
+ | register
+ | id
+ | constant */
+static int
+intel_e11 ()
{
- char *s = reg_string;
- char *p;
- char reg_name_given[MAX_REG_NAME_SIZE + 1];
- const reg_entry *r;
+ /* e11 ( expr ) */
+ if (cur_token.code == '(')
+ {
+ intel_match_token ('(');
+ strcat (intel_parser.disp, "(");
- /* Skip possible REGISTER_PREFIX and possible whitespace. */
- if (*s == REGISTER_PREFIX)
- ++s;
+ if (intel_expr () && intel_match_token (')'))
+ {
+ strcat (intel_parser.disp, ")");
+ return 1;
+ }
+ else
+ return 0;
+ }
- if (is_space_char (*s))
- ++s;
+ /* e11 [ expr ] */
+ else if (cur_token.code == '[')
+ {
+ intel_match_token ('[');
- p = reg_name_given;
- while ((*p++ = register_chars[(unsigned char) *s]) != '\0')
+ /* Mark as a memory operand only if it's not already known to be an
+ offset expression. If it's an offset expression, we need to keep
+ the brace in. */
+ if (intel_parser.op_modifier != OFFSET_FLAT)
+ intel_parser.is_mem = 1;
+ else
+ strcat (intel_parser.disp, "[");
+
+ /* Operands for jump/call inside brackets denote absolute addresses. */
+ if (current_templates->start->opcode_modifier & Jump
+ || current_templates->start->opcode_modifier & JumpDword
+ || current_templates->start->opcode_modifier & JumpByte
+ || current_templates->start->opcode_modifier & JumpInterSegment)
+ i.types[this_operand] |= JumpAbsolute;
+
+ /* Add a '+' to the displacement string if necessary. */
+ if (*intel_parser.disp != '\0'
+ && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+')
+ strcat (intel_parser.disp, "+");
+
+ if (intel_expr () && intel_match_token (']'))
+ {
+ /* Preserve brackets when the operand is an offset expression. */
+ if (intel_parser.op_modifier == OFFSET_FLAT)
+ strcat (intel_parser.disp, "]");
+
+ return 1;
+ }
+ else
+ return 0;
+ }
+
+ /* e11 BYTE
+ | WORD
+ | DWORD
+ | QWORD
+ | XWORD */
+ else if (cur_token.code == T_BYTE
+ || cur_token.code == T_WORD
+ || cur_token.code == T_DWORD
+ || cur_token.code == T_QWORD
+ || cur_token.code == T_XWORD)
{
- if (p >= reg_name_given + MAX_REG_NAME_SIZE)
- return (const reg_entry *) NULL;
- s++;
+ intel_match_token (cur_token.code);
+
+ return 1;
}
- *end_op = s;
+ /* e11 $
+ | . */
+ else if (cur_token.code == '$' || cur_token.code == '.')
+ {
+ strcat (intel_parser.disp, cur_token.str);
+ intel_match_token (cur_token.code);
- r = (const reg_entry *) hash_find (reg_hash, reg_name_given);
+ /* Mark as a memory operand only if it's not already known to be an
+ offset expression. */
+ if (intel_parser.op_modifier != OFFSET_FLAT)
+ intel_parser.is_mem = 1;
- /* Handle floating point regs, allowing spaces in the (i) part. */
- if (r == i386_regtab /* %st is first entry of table */)
+ return 1;
+ }
+
+ /* e11 register */
+ else if (cur_token.code == T_REG)
{
- if (is_space_char (*s))
- ++s;
- if (*s == '(')
+ const reg_entry *reg = intel_parser.reg = cur_token.reg;
+
+ intel_match_token (T_REG);
+
+ /* Check for segment change. */
+ if (cur_token.code == ':')
{
- ++s;
- if (is_space_char (*s))
- ++s;
- if (*s >= '0' && *s <= '7')
+ if (reg->reg_type & (SReg2 | SReg3))
{
- r = &i386_float_regtab[*s - '0'];
- ++s;
- if (is_space_char (*s))
- ++s;
- if (*s == ')')
+ switch (reg->reg_num)
{
- *end_op = s + 1;
- return r;
+ 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;
}
}
- /* We have "%st(" then garbage */
- return (const reg_entry *) NULL;
+ else
+ {
+ as_bad (_("`%s' is not a valid segment register"), reg->reg_name);
+ return 0;
+ }
}
- }
-
- return r;
-}
-\f
-#if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
-CONST char *md_shortopts = "kmVQ:sq";
-#else
-CONST char *md_shortopts = "m";
-#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 'm':
- flag_do_long_jump = 1;
- break;
+ /* 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;
+ }
-#if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
- /* -k: Ignore for FreeBSD compatibility. */
- case 'k':
- break;
+ /* 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;
- /* -V: SVR4 argument to print version ID. */
- case 'V':
- print_version_id ();
- break;
+ /* 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;
+ }
+ }
- /* -Qy, -Qn: SVR4 arguments controlling whether a .comment section
- should be emitted or not. FIXME: Not implemented. */
- case 'Q':
- break;
+ /* 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;
+ }
- case 's':
- /* -s: On i386 Solaris, this tells the native assembler to use
- .stab instead of .stab.excl. We always use .stab anyhow. */
- break;
+ if (i.base_reg == NULL)
+ i.base_reg = reg;
+ else
+ i.index_reg = reg;
- case 'q':
- /* -q: On i386 Solaris, this tells the native assembler does
- fewer checks. */
- break;
-#endif
+ i.types[this_operand] |= BaseIndex;
+ }
- default:
- return 0;
- }
- return 1;
-}
+ /* 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);
-void
-md_show_usage (stream)
- FILE *stream;
-{
- fprintf (stream, _("\
- -m do long jump\n"));
-#if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)
- fprintf (stream, _("\
- -V print assembler version number\n\
- -k ignored\n\
- -Qy, -Qn ignored\n\
- -q ignored\n\
- -s ignored\n"));
-#endif
-}
+ /* 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++;
+ }
-#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)))
+ /* 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';
+ }
-/* Pick the target format to use. */
+ return 1;
+ }
-const char *
-i386_target_format ()
-{
- switch (OUTPUT_FLAVOR)
+ /* e11 id */
+ else if (cur_token.code == T_ID)
{
-#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;
+ /* Add the identifier to the displacement string. */
+ strcat (intel_parser.disp, cur_token.str);
+ intel_match_token (T_ID);
+
+ /* The identifier represents a memory reference only if it's not
+ preceded by an offset modifier. */
+ if (intel_parser.op_modifier != OFFSET_FLAT)
+ intel_parser.is_mem = 1;
+
+ return 1;
}
-}
-#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)
+ /* e11 constant */
+ else if (cur_token.code == T_CONST
+ || cur_token.code == '-'
+ || cur_token.code == '+')
{
- if (!GOT_symbol)
+ char *save_str;
+
+ /* Allow constants that start with `+' or `-'. */
+ if (cur_token.code == '-' || cur_token.code == '+')
{
- 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;
-}
+ 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;
+ }
+ }
-/* Round up a section size to the appropriate boundary. */
-valueT
-md_section_align (segment, size)
- segT segment ATTRIBUTE_UNUSED;
- valueT size;
-{
-#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;
+ save_str = (char *) malloc (strlen (cur_token.str) + 1);
+ if (save_str == NULL)
+ abort ();
+ strcpy (save_str, cur_token.str);
- align = bfd_get_section_alignment (stdoutput, segment);
- size = ((size + (1 << align) - 1) & ((valueT) -1 << align));
+ /* 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;
}
-#endif
-#endif
- return size;
+ as_bad (_("Unrecognized token '%s'"), cur_token.str);
+ return 0;
}
-/* 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;
+/* 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;
{
- return fixP->fx_size + fixP->fx_where + fixP->fx_frag->fr_address;
+ if (cur_token.code == code)
+ {
+ intel_get_token ();
+ return 1;
+ }
+ else
+ {
+ as_bad (_("Unexpected token `%s'\n"), cur_token.str);
+ return 0;
+ }
}
-#ifndef I386COFF
-
+/* Read a new token from intel_parser.op_string and store it in cur_token. */
static void
-s_bss (ignore)
- int ignore ATTRIBUTE_UNUSED;
+intel_get_token ()
{
- register int temp;
+ char *end_op;
+ const reg_entry *reg;
+ struct intel_token new_token;
- temp = get_absolute_expression ();
- subseg_set (bss_section, (subsegT) temp);
- demand_empty_rest_of_line ();
-}
+ new_token.code = T_NIL;
+ new_token.reg = NULL;
+ new_token.str = NULL;
-#endif
+ /* 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;
-#ifdef BFD_ASSEMBLER
+ /* Skip whitespace. */
+ while (is_space_char (*intel_parser.op_string))
+ intel_parser.op_string++;
-void
-i386_validate_fix (fixp)
- fixS *fixp;
-{
- if (fixp->fx_subsy && fixp->fx_subsy == GOT_symbol)
+ /* Return an empty token if we find nothing else on the line. */
+ if (*intel_parser.op_string == '\0')
{
- fixp->fx_r_type = BFD_RELOC_386_GOTOFF;
- fixp->fx_subsy = 0;
+ cur_token = new_token;
+ return;
}
-}
-arelent *
-tc_gen_reloc (section, fixp)
- asection *section ATTRIBUTE_UNUSED;
- fixS *fixp;
-{
- arelent *rel;
- bfd_reloc_code_real_type code;
+ /* 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';
- switch (fixp->fx_r_type)
+ if (strchr ("0123456789", *intel_parser.op_string))
{
- 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:
- 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;
+ 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;
}
- if (code == BFD_RELOC_32
- && GOT_symbol
- && fixp->fx_addsy == GOT_symbol)
- code = BFD_RELOC_386_GOTPC;
+ 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';
+ }
- rel = (arelent *) xmalloc (sizeof (arelent));
- rel->sym_ptr_ptr = (asymbol **) xmalloc (sizeof (asymbol *));
- *rel->sym_ptr_ptr = symbol_get_bfdsym (fixp->fx_addsy);
+ 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;
- 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 (*intel_parser.op_string == REGISTER_PREFIX)
+ {
+ new_token.str[0] = REGISTER_PREFIX;
+ new_token.str[1] = '\0';
+ }
- if (fixp->fx_pcrel)
- rel->addend = fixp->fx_addnumber;
- else
- rel->addend = 0;
+ strcat (new_token.str, reg->reg_name);
+ }
- rel->howto = bfd_reloc_type_lookup (stdoutput, code);
- if (rel->howto == NULL)
+ else if (is_identifier_char (*intel_parser.op_string))
{
- 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);
- }
+ char *p = new_token.str;
+ char *q = intel_parser.op_string;
- return rel;
-}
+ /* 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';
-#else /* ! BFD_ASSEMBLER */
+ if (strcasecmp (new_token.str, "BYTE") == 0)
+ new_token.code = T_BYTE;
-#if (defined(OBJ_AOUT) | defined(OBJ_BOUT))
-void
-tc_aout_fix_to_chars (where, fixP, segment_address_in_file)
- char *where;
- 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.
- */
+ else if (strcasecmp (new_token.str, "WORD") == 0)
+ new_token.code = T_WORD;
- static const unsigned char nbytes_r_length[] = {42, 0, 1, 42, 2};
- long r_symbolnum;
+ else if (strcasecmp (new_token.str, "DWORD") == 0)
+ new_token.code = T_DWORD;
- know (fixP->fx_addsy != NULL);
+ else if (strcasecmp (new_token.str, "QWORD") == 0)
+ new_token.code = T_QWORD;
- md_number_to_chars (where,
- (valueT) (fixP->fx_frag->fr_address
- + fixP->fx_where - segment_address_in_file),
- 4);
+ else if (strcasecmp (new_token.str, "XWORD") == 0)
+ new_token.code = T_XWORD;
- r_symbolnum = (S_IS_DEFINED (fixP->fx_addsy)
- ? S_GET_TYPE (fixP->fx_addsy)
- : fixP->fx_addsy->sy_number);
+ else if (strcasecmp (new_token.str, "PTR") == 0)
+ new_token.code = T_PTR;
- where[6] = (r_symbolnum >> 16) & 0x0ff;
- where[5] = (r_symbolnum >> 8) & 0x0ff;
- where[4] = r_symbolnum & 0x0ff;
- where[7] = ((((!S_IS_DEFINED (fixP->fx_addsy)) << 3) & 0x08)
- | ((nbytes_r_length[fixP->fx_size] << 1) & 0x06)
- | (((fixP->fx_pcrel << 0) & 0x01) & 0x0f));
-}
+ 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;
-#endif /* OBJ_AOUT or OBJ_BOUT */
+ /* ??? 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:");
+ }
-#if defined (I386COFF)
+ /* ??? This is not mentioned in the MASM grammar. */
+ else if (strcasecmp (new_token.str, "FLAT") == 0)
+ new_token.code = T_OFFSET;
-short
-tc_coff_fix2rtype (fixP)
- fixS *fixP;
-{
- if (fixP->fx_r_type == R_IMAGEBASE)
- return R_IMAGEBASE;
+ else
+ new_token.code = T_ID;
+ }
+ }
- return (fixP->fx_pcrel ?
- (fixP->fx_size == 1 ? R_PCRBYTE :
- fixP->fx_size == 2 ? R_PCRWORD :
- R_PCRLONG) :
- (fixP->fx_size == 1 ? R_RELBYTE :
- fixP->fx_size == 2 ? R_RELWORD :
- R_DIR32));
+ else
+ as_bad (_("Unrecognized token `%s'\n"), intel_parser.op_string);
+
+ intel_parser.op_string += strlen (new_token.str);
+ cur_token = new_token;
}
-int
-tc_coff_sizemachdep (frag)
- fragS *frag;
+/* Put cur_token back into the token stream and make cur_token point to
+ prev_token. */
+static void
+intel_putback_token ()
{
- if (frag->fr_next)
- return (frag->fr_next->fr_address - frag->fr_address);
- else
- return 0;
+ 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;
}
-
-#endif /* I386COFF */
-
-#endif /* ! BFD_ASSEMBLER */
-\f
-/* end of tc-i386.c */
projects / deliverable / binutils-gdb.git / blobdiff