diff options
-rw-r--r-- | contrib/binutils/gas/config/tc-i386.c | 3570 |
1 files changed, 2392 insertions, 1178 deletions
diff --git a/contrib/binutils/gas/config/tc-i386.c b/contrib/binutils/gas/config/tc-i386.c index 7953327..db2e18b 100644 --- a/contrib/binutils/gas/config/tc-i386.c +++ b/contrib/binutils/gas/config/tc-i386.c @@ -1,6 +1,6 @@ /* 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 (C) 1989, 91, 92, 93, 94, 95, 96, 97, 98, 99, 2000, 2001 + Free Software Foundation, Inc. This file is part of GAS, the GNU Assembler. @@ -19,19 +19,21 @@ Software Foundation, 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. */ + /* $FreeBSD$ */ -/* - Intel 80386 machine specific gas. - Written by Eliot Dresselhaus (eliot@mgm.mit.edu). - Bugs & suggestions are completely welcome. This is free software. - Please help us make it better. - */ + +/* Intel 80386 machine specific gas. + Written by Eliot Dresselhaus (eliot@mgm.mit.edu). + 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 <ctype.h> #include "as.h" #include "subsegs.h" +#include "dwarf2dbg.h" #include "opcode/i386.h" #ifndef REGISTER_WARNINGS @@ -54,23 +56,32 @@ #define false 0 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 int fits_in_signed_byte PARAMS ((offsetT)); +static int fits_in_unsigned_byte PARAMS ((offsetT)); +static int fits_in_unsigned_word PARAMS ((offsetT)); +static int fits_in_signed_word PARAMS ((offsetT)); +static int fits_in_unsigned_long PARAMS ((offsetT)); +static int fits_in_signed_long PARAMS ((offsetT)); +static int smallest_imm_type PARAMS ((offsetT)); +static offsetT offset_in_range PARAMS ((offsetT, int)); static int add_prefix PARAMS ((unsigned int)); -static void set_16bit_code_flag PARAMS ((int)); -static void set_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)); +static void set_cpu_arch PARAMS ((int)); #ifdef BFD_ASSEMBLER static bfd_reloc_code_real_type reloc - PARAMS ((int, int, bfd_reloc_code_real_type)); + PARAMS ((int, int, int, bfd_reloc_code_real_type)); #endif +#ifndef DEFAULT_ARCH +#define DEFAULT_ARCH "i386" +#endif +static char *default_arch = DEFAULT_ARCH; + /* 'md_assemble ()' gathers together information and puts it into a - i386_insn. */ + i386_insn. */ union i386_op { @@ -81,19 +92,19 @@ union i386_op 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 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 @@ -104,6 +115,10 @@ struct _i386_insn operand. */ union i386_op op[MAX_OPERANDS]; + /* Flags for operands. */ + unsigned int flags[MAX_OPERANDS]; +#define Operand_PCrel 1 + /* Relocation type for operand */ #ifdef BFD_ASSEMBLER enum bfd_reloc_code_real disp_reloc[MAX_OPERANDS]; @@ -118,8 +133,8 @@ struct _i386_insn 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. */ @@ -130,6 +145,7 @@ struct _i386_insn addressing modes of this insn are encoded. */ modrm_byte rm; + rex_byte rex; sib_byte sib; }; @@ -144,8 +160,8 @@ const char extra_symbol_chars[] = "*%-("; #endif /* This array holds the chars that always start a comment. If the - pre-processor is disabled, these aren't very useful */ -#if defined (TE_I386AIX) || ((defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)) && !defined (TE_LINUX) && !defined(TE_FreeBSD)) + pre-processor is disabled, these aren't very useful. */ +#if defined (TE_I386AIX) || ((defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)) && ! defined (TE_LINUX) && !defined(TE_FreeBSD)) /* Putting '/' here makes it impossible to use the divide operator. However, we need it for compatibility with SVR4 systems. */ const char comment_chars[] = "#/"; @@ -157,36 +173,37 @@ const char comment_chars[] = "#"; /* This array holds the chars that only start a comment at the beginning of a line. If the line seems to have the form '# 123 filename' - .line and .file directives will appear in the pre-processed output */ -/* Note that input_file.c hand checks for '#' at the beginning of the + .line and .file directives will appear in the pre-processed output. + Note that input_file.c hand checks for '#' at the beginning of the first line of the input file. This is because the compiler outputs - #NO_APP at the beginning of its output. */ -/* Also note that comments started like this one will always work if + #NO_APP at the beginning of its output. + Also note that comments started like this one will always work if '/' isn't otherwise defined. */ -#if defined (TE_I386AIX) || ((defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)) && !defined (TE_LINUX) && !defined(TE_FreeBSD)) +#if defined (TE_I386AIX) || ((defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)) && ! defined (TE_LINUX) && !defined(TE_FreeBSD)) const char line_comment_chars[] = ""; #else const char line_comment_chars[] = "/"; #endif -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]) @@ -194,43 +211,71 @@ static char digit_chars[256]; #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; -static int flag_do_long_jump; /* FIXME what does this do? */ +/* We support four different modes. FLAG_CODE variable is used to distinguish + these. */ -static int flag_16bit_code; /* 1 if we're writing 16-bit code, 0 if 32-bit */ +enum flag_code { + CODE_32BIT, + CODE_16BIT, + CODE_64BIT }; + +static enum flag_code flag_code; +static int use_rela_relocations = 0; + +/* The names used to print error messages. */ +static const char *flag_code_names[] = + { + "32", + "16", + "64" + }; -static int intel_syntax = 0; /* 1 for intel syntax, 0 if att syntax */ +/* 1 for intel syntax, + 0 if att syntax. */ +static int intel_syntax = 0; -static int allow_naked_reg = 0; /* 1 if register prefix % not required */ +/* 1 if register prefix % not required. */ +static int allow_naked_reg = 0; -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. */ +/* Used in 16 bit gcc mode to add an l suffix to call, ret, enter, + leave, push, and pop instructions so that gcc has the same stack + frame as in 32 bit mode. */ +static char stackop_size = '\0'; + +/* Non-zero to quieten some warnings. */ +static int quiet_warnings = 0; + +/* CPU name. */ +static const char *cpu_arch_name = NULL; + +/* CPU feature flags. */ +static unsigned int cpu_arch_flags = CpuUnknownFlags|CpuNo64; /* Interface to relax_segment. There are 2 relax states for 386 jump insns: one for conditional & @@ -238,10 +283,10 @@ static char stackop_size = '\0'; /* Used in 16 bit gcc mode to add an l 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 */ +/* Types. */ +#define COND_JUMP 1 +#define UNCOND_JUMP 2 +/* Sizes. */ #define CODE16 1 #define SMALL 0 #define SMALL16 (SMALL|CODE16) @@ -257,7 +302,7 @@ static char stackop_size = '\0'; /* Used in 16 bit gcc mode to add an l #endif #define ENCODE_RELAX_STATE(type,size) \ - ((relax_substateT)((type<<2) | (size))) + ((relax_substateT) ((type<<2) | (size))) #define SIZE_FROM_RELAX_STATE(s) \ ( (((s) & 0x3) == BIG ? 4 : (((s) & 0x3) == BIG16 ? 2 : 1)) ) @@ -275,8 +320,7 @@ const relax_typeS md_relax_table[] = 1) most positive reach of this state, 2) most negative reach of this state, 3) how many bytes this mode will add to the size of the current frag - 4) which index into the table to try if we can't fit into this one. - */ + 4) which index into the table to try if we can't fit into this one. */ {1, 1, 0, 0}, {1, 1, 0, 0}, {1, 1, 0, 0}, @@ -302,15 +346,31 @@ const relax_typeS md_relax_table[] = }; +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 } +}; 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[] = @@ -375,18 +435,24 @@ i386_align_code (fragP, count) 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; } } @@ -401,65 +467,125 @@ static const reg_entry *parse_register PARAMS ((char *reg_string, static void s_bss PARAMS ((int)); #endif -symbolS *GOT_symbol; /* Pre-defined "_GLOBAL_OFFSET_TABLE_" */ +symbolS *GOT_symbol; /* Pre-defined "_GLOBAL_OFFSET_TABLE_". */ static INLINE unsigned int mode_from_disp_size (t) unsigned int t; { - return (t & Disp8) ? 1 : (t & (Disp16|Disp32)) ? 2 : 0; + return (t & Disp8) ? 1 : (t & (Disp16 | Disp32 | Disp32S)) ? 2 : 0; } static INLINE int 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) + && !(cpu_arch_flags & (CpuUnknown))) + { + /* This code is disabled on the 486 because all the Imm1 forms + in the opcode table are slower on the i486. They're the + 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 @@ -471,40 +597,43 @@ add_prefix (prefix) 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 >= 0x40 && prefix < 0x50 && flag_code == CODE_64BIT) + q = REX_PREFIX; + else + switch (prefix) + { + default: + abort (); + + case CS_PREFIX_OPCODE: + case DS_PREFIX_OPCODE: + case ES_PREFIX_OPCODE: + case FS_PREFIX_OPCODE: + case GS_PREFIX_OPCODE: + case SS_PREFIX_OPCODE: + q = SEG_PREFIX; + break; + + case REPNE_PREFIX_OPCODE: + case REPE_PREFIX_OPCODE: + ret = 2; + /* fall thru */ + case LOCK_PREFIX_OPCODE: + q = LOCKREP_PREFIX; + break; + + case 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]) { @@ -518,19 +647,31 @@ add_prefix (prefix) } 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 @@ -546,9 +687,9 @@ set_intel_syntax (syntax_flag) 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.")); @@ -564,59 +705,110 @@ set_intel_syntax (syntax_flag) 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); } +static void +set_cpu_arch (dummy) + int dummy ATTRIBUTE_UNUSED; +{ + SKIP_WHITESPACE (); + + if (! is_end_of_line[(unsigned char) *input_line_pointer]) + { + char *string = input_line_pointer; + int e = get_symbol_end (); + int i; + + for (i = 0; cpu_arch[i].name; i++) + { + if (strcmp (string, cpu_arch[i].name) == 0) + { + cpu_arch_name = cpu_arch[i].name; + cpu_arch_flags = cpu_arch[i].flags | (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")); + + demand_empty_rest_of_line (); +} + const pseudo_typeS md_pseudo_table[] = { -#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 + {"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, 1}, - {"code16", set_16bit_code_flag, 1}, - {"code32", set_16bit_code_flag, 0}, + {"code16gcc", set_16bit_gcc_code_flag, CODE_16BIT}, + {"code16", set_code_flag, CODE_16BIT}, + {"code32", set_code_flag, CODE_32BIT}, + {"code64", set_code_flag, CODE_64BIT}, {"intel_syntax", set_intel_syntax, 1}, {"att_syntax", set_intel_syntax, 0}, + {"file", dwarf2_directive_file, 0}, + {"loc", dwarf2_directive_loc, 0}, {0, 0, 0} }; -/* for interface with expression () */ +/* For interface with expression (). */ extern char *input_line_pointer; -/* hash table for instruction mnemonic lookup */ +/* Hash table for instruction mnemonic lookup. */ static struct hash_control *op_hash; -/* hash table for register lookup */ + +/* Hash table for register lookup. */ static struct hash_control *reg_hash; - +#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 + 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; - optab = i386_optab; /* setup for loop */ + /* Setup for loop. */ + optab = i386_optab; core_optab = (templates *) xmalloc (sizeof (templates)); core_optab->start = optab; @@ -627,14 +819,13 @@ md_begin () || 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); @@ -647,7 +838,7 @@ md_begin () } } - /* initialize reg_hash hash table */ + /* Initialize reg_hash hash table. */ reg_hash = hash_new (); { register const reg_entry *regtab; @@ -658,11 +849,13 @@ md_begin () { 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; @@ -727,10 +920,9 @@ i386_print_statistics (file) hash_print_statistics (file, "i386 register", reg_hash); } - #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)); @@ -742,15 +934,20 @@ pi (line, x) 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.mode64, x->rex.extX, x->rex.extY, x->rex.extZ); for (i = 0; i < x->operands; i++) { fprintf (stdout, " #%d: ", i + 1); @@ -770,10 +967,9 @@ static void 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) @@ -831,15 +1027,20 @@ 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" }, @@ -863,16 +1064,9 @@ pt (t) { register 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); } @@ -888,25 +1082,27 @@ tc_i386_force_relocation (fixp) return 1; return 0; #else - /* For COFF */ + /* For COFF. */ return fixp->fx_r_type == 7; #endif } #ifdef BFD_ASSEMBLER -static bfd_reloc_code_real_type reloc - PARAMS ((int, int, bfd_reloc_code_real_type)); static bfd_reloc_code_real_type -reloc (size, pcrel, other) +reloc (size, pcrel, sign, other) int size; int pcrel; + int sign; bfd_reloc_code_real_type other; { - if (other != NO_RELOC) 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; @@ -917,64 +1113,76 @@ reloc (size, pcrel, other) } 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 (OBJ_MAYBE_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_EXTERNAL (fixP->fx_addsy) || S_IS_WEAK (fixP->fx_addsy)) return 0; #endif - /* adjust_reloc_syms doesn't know about the GOT */ + /* adjust_reloc_syms doesn't know about the GOT. */ if (fixP->fx_r_type == BFD_RELOC_386_GOTOFF || fixP->fx_r_type == BFD_RELOC_386_PLT32 || fixP->fx_r_type == BFD_RELOC_386_GOT32 - || fixP->fx_r_type == BFD_RELOC_RVA + || 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_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 ((char *mnemonic)); static int intel_float_operand (mnemonic) char *mnemonic; { - if (mnemonic[0] == 'f' && mnemonic[1] =='i') + if (mnemonic[0] == 'f' && mnemonic[1] == 'i') return 2; if (mnemonic[0] == 'f') @@ -991,7 +1199,7 @@ void md_assemble (line) char *line; { - /* Points to template once we've found it. */ + /* Points to template once we've found it. */ const template *t; /* Count the size of the instruction generated. */ @@ -1001,23 +1209,23 @@ md_assemble (line) 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; 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. */ + start of a (possibly prefixed) mnemonic. */ { char *l = line; char *token_start = l; char *mnem_p; - /* Non-zero if we found a prefix only acceptable with string insns. */ + /* Non-zero if we found a prefix only acceptable with string insns. */ const char *expecting_string_instruction = NULL; while (1) @@ -1028,7 +1236,7 @@ md_assemble (line) mnem_p++; if (mnem_p >= mnemonic + sizeof (mnemonic)) { - as_bad (_("no such 386 instruction: `%s'"), token_start); + as_bad (_("no such instruction: `%s'"), token_start); return; } l++; @@ -1062,7 +1270,7 @@ md_assemble (line) Similarly, in 32-bit mode, do not allow addr32 or data32. */ if ((current_templates->start->opcode_modifier & (Size16 | Size32)) && (((current_templates->start->opcode_modifier & Size32) != 0) - ^ flag_16bit_code)) + ^ (flag_code == CODE_16BIT))) { as_bad (_("redundant %s prefix"), current_templates->start->name); @@ -1074,8 +1282,7 @@ md_assemble (line) case 0: return; case 2: - expecting_string_instruction = - current_templates->start->name; + expecting_string_instruction = current_templates->start->name; break; } /* Skip past PREFIX_SEPARATOR and reset token_start. */ @@ -1092,31 +1299,57 @@ md_assemble (line) { case WORD_MNEM_SUFFIX: case BYTE_MNEM_SUFFIX: - case SHORT_MNEM_SUFFIX: - case LONG_MNEM_SUFFIX: + case QWORD_MNEM_SUFFIX: i.suffix = mnem_p[-1]; mnem_p[-1] = '\0'; current_templates = hash_find (op_hash, mnemonic); break; + case SHORT_MNEM_SUFFIX: + case LONG_MNEM_SUFFIX: + if (!intel_syntax) + { + i.suffix = mnem_p[-1]; + mnem_p[-1] = '\0'; + current_templates = hash_find (op_hash, mnemonic); + } + break; - /* Intel Syntax */ - case DWORD_MNEM_SUFFIX: + /* Intel Syntax. */ + case 'd': if (intel_syntax) { - i.suffix = mnem_p[-1]; + if (intel_float_operand (mnemonic)) + i.suffix = SHORT_MNEM_SUFFIX; + else + i.suffix = LONG_MNEM_SUFFIX; mnem_p[-1] = '\0'; current_templates = hash_find (op_hash, mnemonic); - break; } + break; } if (!current_templates) { - as_bad (_("no such 386 instruction: `%s'"), token_start); + as_bad (_("no such instruction: `%s'"), token_start); return; } } - /* check for rep/repne without a string instruction */ + /* Check if instruction is supported on specified architecture. */ + if (cpu_arch_flags != 0) + { + 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)) { @@ -1125,20 +1358,18 @@ md_assemble (line) return; } - /* There may be operands to parse. */ + /* There may be operands to parse. */ if (*l != END_OF_INSN) { - /* parse operands */ - - /* 1 if operand is pending after ','. */ + /* 1 if operand is pending after ','. */ unsigned int expecting_operand = 0; - /* Non-zero if operand parens not balanced. */ + /* Non-zero if operand parens not balanced. */ unsigned int paren_not_balanced; do { - /* skip optional white space before operand */ + /* Skip optional white space before operand. */ if (is_space_char (*l)) ++l; if (!is_operand_char (*l) && *l != END_OF_INSN) @@ -1191,7 +1422,7 @@ md_assemble (line) l++; } if (l != token_start) - { /* yes, we've read in another operand */ + { /* Yes, we've read in another operand. */ unsigned int operand_ok; this_operand = i.operands++; if (i.operands > MAX_OPERANDS) @@ -1200,15 +1431,17 @@ md_assemble (line) MAX_OPERANDS); return; } - /* now parse operand adding info to 'i' as we go along */ + /* Now parse operand adding info to 'i' as we go along. */ END_STRING_AND_SAVE (l); if (intel_syntax) - operand_ok = i386_intel_operand (token_start, intel_float_operand (mnemonic)); + 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 */ + RESTORE_END_STRING (l); if (!operand_ok) return; } @@ -1227,17 +1460,18 @@ md_assemble (line) } } - /* now *l must be either ',' or END_OF_INSN */ + /* Now *l must be either ',' or END_OF_INSN. */ if (*l == ',') { if (*++l == END_OF_INSN) - { /* just skip it, if it's \n complain */ + { + /* Just skip it, if it's \n complain. */ goto expecting_operand_after_comma; } expecting_operand = 1; } } - while (*l != END_OF_INSN); /* until we get end of insn */ + while (*l != END_OF_INSN); } } @@ -1246,7 +1480,7 @@ md_assemble (line) 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. */ + with the template operand types. */ #define MATCH(overlap, given, template) \ ((overlap & ~JumpAbsolute) \ @@ -1278,6 +1512,11 @@ md_assemble (line) { union i386_op temp_op; unsigned int temp_type; +#ifdef BFD_ASSEMBLER + enum bfd_reloc_code_real temp_reloc; +#else + int temp_reloc; +#endif int xchg1 = 0; int xchg2 = 0; @@ -1297,6 +1536,9 @@ md_assemble (line) temp_op = i.op[xchg2]; i.op[xchg2] = i.op[xchg1]; i.op[xchg1] = temp_op; + temp_reloc = i.disp_reloc[xchg2]; + i.disp_reloc[xchg2] = i.disp_reloc[xchg1]; + i.disp_reloc[xchg1] = temp_reloc; if (i.mem_operands == 2) { @@ -1322,45 +1564,131 @@ md_assemble (line) 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; ) + for (op = i.operands; --op >= 0;) if (i.types[op] & Reg) { if (i.types[op] & Reg8) guess_suffix = BYTE_MNEM_SUFFIX; else if (i.types[op] & Reg16) guess_suffix = WORD_MNEM_SUFFIX; + else if (i.types[op] & Reg32) + guess_suffix = LONG_MNEM_SUFFIX; + else if (i.types[op] & Reg64) + guess_suffix = QWORD_MNEM_SUFFIX; break; } } - else if (flag_16bit_code ^ (i.prefix[DATA_PREFIX] != 0)) + else if ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)) guess_suffix = WORD_MNEM_SUFFIX; - for (op = i.operands; --op >= 0; ) - if ((i.types[op] & Imm) - && i.op[op].imms->X_op == O_constant) + for (op = i.operands; --op >= 0;) + if (i.types[op] & Imm) { - /* If a suffix is given, this operand may be shortened. */ - switch (guess_suffix) + switch (i.op[op].imms->X_op) { - case WORD_MNEM_SUFFIX: - i.types[op] |= Imm16; - break; - case BYTE_MNEM_SUFFIX: - i.types[op] |= Imm16 | Imm8 | Imm8S; - break; + case O_constant: + /* If a suffix is given, this operand may be shortened. */ + switch (guess_suffix) + { + case LONG_MNEM_SUFFIX: + i.types[op] |= Imm32 | Imm64; + break; + case WORD_MNEM_SUFFIX: + i.types[op] |= Imm16 | Imm32S | Imm32 | Imm64; + break; + case BYTE_MNEM_SUFFIX: + i.types[op] |= Imm16 | Imm8 | Imm8S | Imm32S | Imm32 | Imm64; + break; + } + + /* If this operand is at most 16 bits, convert it + to a signed 16 bit number before trying to see + whether it will fit in an even smaller size. + This allows a 16-bit operand such as $0xffe0 to + be recognised as within Imm8S range. */ + if ((i.types[op] & Imm16) + && (i.op[op].imms->X_add_number & ~(offsetT) 0xffff) == 0) + { + i.op[op].imms->X_add_number = + (((i.op[op].imms->X_add_number & 0xffff) ^ 0x8000) - 0x8000); + } + if ((i.types[op] & Imm32) + && (i.op[op].imms->X_add_number & ~(((offsetT) 2 << 31) - 1)) == 0) + { + i.op[op].imms->X_add_number = + (i.op[op].imms->X_add_number ^ ((offsetT) 1 << 31)) - ((addressT) 1 << 31); + } + i.types[op] |= smallest_imm_type (i.op[op].imms->X_add_number); + /* We must avoid matching of Imm32 templates when 64bit only immediate is available. */ + if (guess_suffix == QWORD_MNEM_SUFFIX) + i.types[op] &= ~Imm32; + break; + case O_absent: + case O_register: + abort (); + /* Symbols and expressions. */ + default: + /* Convert symbolic operand to proper sizes for matching. */ + switch (guess_suffix) + { + case QWORD_MNEM_SUFFIX: + i.types[op] = Imm64 | Imm32S; + break; + case LONG_MNEM_SUFFIX: + i.types[op] = Imm32 | Imm64; + break; + case WORD_MNEM_SUFFIX: + i.types[op] = Imm16 | Imm32 | Imm64; + break; + break; + case BYTE_MNEM_SUFFIX: + i.types[op] = Imm8 | Imm8S | Imm16 | Imm32S | Imm32; + break; + break; + } + break; } + } + } + + if (i.disp_operands) + { + /* Try to use the smallest displacement type too. */ + int op; + + for (op = i.operands; --op >= 0;) + if ((i.types[op] & Disp) + && i.op[op].imms->X_op == O_constant) + { + offsetT disp = i.op[op].disps->X_add_number; + + if (i.types[op] & Disp16) + { + /* We know this operand is at most 16 bits, so + convert to a signed 16 bit number before trying + to see whether it will fit in an even smaller + size. */ - /* 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) + disp = (((disp & 0xffff) ^ 0x8000) - 0x8000); + } + else if (i.types[op] & Disp32) { - i.op[op].imms->X_add_number = - (((i.op[op].imms->X_add_number & 0xffff) ^ 0x8000) - 0x8000); + /* 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); } - i.types[op] |= smallest_imm_type ((long) i.op[op].imms->X_add_number); + if (flag_code == CODE_64BIT) + { + if (fits_in_signed_long (disp)) + i.types[op] |= Disp32S; + if (fits_in_unsigned_long (disp)) + i.types[op] |= Disp32; + } + if ((i.types[op] & (Disp32 | Disp32S | Disp16)) + && fits_in_signed_byte (disp)) + i.types[op] |= Disp8; } } @@ -1376,28 +1704,36 @@ md_assemble (line) ? No_sSuf : (i.suffix == LONG_MNEM_SUFFIX ? No_lSuf - : (i.suffix == DWORD_MNEM_SUFFIX - ? No_dSuf + : (i.suffix == QWORD_MNEM_SUFFIX + ? No_qSuf : (i.suffix == LONG_DOUBLE_MNEM_SUFFIX ? No_xSuf : 0)))))); for (t = current_templates->start; t < current_templates->end; t++) { - /* Must have right number of operands. */ + /* Must have right number of operands. */ if (i.operands != t->operands) continue; /* Check the suffix, except for some instructions in intel mode. */ if ((t->opcode_modifier & suffix_check) && !(intel_syntax + && (t->opcode_modifier & IgnoreSize)) + && !(intel_syntax && t->base_opcode == 0xd9 - && (t->extension_opcode == 5 /* 0xd9,5 "fldcw" */ - || t->extension_opcode == 7))) /* 0xd9,7 "f{n}stcw" */ + && (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) - break; /* 0 operands always matches */ + { + if (t->cpu_flags & ~cpu_arch_flags) + continue; + /* We've found a match; break out of loop. */ + break; + } overlap0 = i.types[0] & t->operand_types[0]; switch (t->operands) @@ -1416,12 +1752,11 @@ md_assemble (line) overlap1, i.types[1], t->operand_types[1])) { - - /* check if other direction is valid ... */ + /* Check if other direction is valid ... */ if ((t->opcode_modifier & (D|FloatD)) == 0) continue; - /* try reversing direction of operands */ + /* Try reversing direction of operands. */ overlap0 = i.types[0] & t->operand_types[1]; overlap1 = i.types[1] & t->operand_types[0]; if (!MATCH (overlap0, i.types[0], t->operand_types[1]) @@ -1431,16 +1766,15 @@ md_assemble (line) overlap1, i.types[1], t->operand_types[0])) { - /* does not match either direction */ + /* 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) + /* 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 @@ -1455,29 +1789,41 @@ md_assemble (line) continue; } - /* found either forward/reverse 2 or 3 operand match here: - slip through to break */ + /* Found either forward/reverse 2 or 3 operand match here: + slip through to break. */ + } + if (t->cpu_flags & ~cpu_arch_flags) + { + found_reverse_match = 0; + continue; } - break; /* we've found a match; break out of loop */ - } /* for (t = ... */ + /* We've found a match; break out of loop. */ + break; + } if (t == current_templates->end) - { /* we found no match */ + { + /* We found no match. */ as_bad (_("suffix or operands invalid for `%s'"), current_templates->start->name); return; } - if (!intel_syntax - && (i.types[0] & JumpAbsolute) != (t->operand_types[0] & JumpAbsolute)) + if (!quiet_warnings) { - as_warn (_("indirect %s without `*'"), t->name); - } + if (!intel_syntax + && ((i.types[0] & JumpAbsolute) + != (t->operand_types[0] & JumpAbsolute))) + { + as_warn (_("indirect %s without `*'"), t->name); + } - if ((t->opcode_modifier & (IsPrefix|IgnoreSize)) == (IsPrefix|IgnoreSize)) - { - /* Warn them that a data or address size prefix doesn't affect - assembly of the next line of code. */ - as_warn (_("stand-alone `%s' prefix"), t->name); + if ((t->opcode_modifier & (IsPrefix|IgnoreSize)) + == (IsPrefix|IgnoreSize)) + { + /* Warn them that a data or address size prefix doesn't + affect assembly of the next line of code. */ + as_warn (_("stand-alone `%s' prefix"), t->name); + } } /* Copy the template we found. */ @@ -1495,16 +1841,16 @@ md_assemble (line) } /* Undo SYSV386_COMPAT brokenness when in Intel mode. See i386.h */ - if (SYSV386_COMPAT - && intel_syntax - && (i.tm.base_opcode & 0xfffffde0) == 0xdce0) - i.tm.base_opcode ^= FloatR; + if (SYSV386_COMPAT + && intel_syntax + && (i.tm.base_opcode & 0xfffffde0) == 0xdce0) + i.tm.base_opcode ^= FloatR; if (i.tm.opcode_modifier & FWait) if (! add_prefix (FWAIT_OPCODE)) return; - /* Check string instruction segment overrides */ + /* Check string instruction segment overrides. */ if ((i.tm.opcode_modifier & IsString) != 0 && i.mem_operands != 0) { int mem_op = (i.types[0] & AnyMem) ? 0 : 1; @@ -1535,31 +1881,47 @@ md_assemble (line) } } + if (i.reg_operands && flag_code < CODE_64BIT) + { + int op; + for (op = i.operands; --op >= 0;) + if ((i.types[op] & Reg) + && (i.op[op].regs->reg_flags & (RegRex64|RegRex))) + { + as_bad (_("Extended register `%%%s' available only in 64bit mode."), + i.op[op].regs->reg_name); + return; + } + } + /* If matched instruction specifies an explicit instruction mnemonic suffix, use it. */ - if (i.tm.opcode_modifier & (Size16 | Size32)) + if (i.tm.opcode_modifier & (Size16 | 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. */ + 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; ) + 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; } @@ -1567,18 +1929,19 @@ md_assemble (line) else if (i.suffix == BYTE_MNEM_SUFFIX) { int op; - for (op = i.operands; --op >= 0; ) + for (op = i.operands; --op >= 0;) { /* If this is an eight bit register, it's OK. If it's the 16 or 32 bit version of an eight bit register, - we will just use the low portion, and that's OK too. */ + 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. */ + /* 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; @@ -1592,16 +1955,27 @@ md_assemble (line) #endif ) { + /* Prohibit these changes in the 64bit mode, since + the lowering is more complicated. */ + if (flag_code == CODE_64BIT + && (i.tm.operand_types[op] & InOutPortReg) == 0) + as_bad (_("Incorrect register `%%%s' used with`%c' suffix"), + i.op[op].regs->reg_name, + i.suffix); #if REGISTER_WARNINGS - if ((i.tm.operand_types[op] & InOutPortReg) == 0) + if (!quiet_warnings + && (i.tm.operand_types[op] & InOutPortReg) == 0) as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"), - (i.op[op].regs - (i.types[op] & Reg16 ? 8 : 16))->reg_name, + (i.op[op].regs + + (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 */ + /* Any other register is bad. */ if (i.types[op] & (Reg | RegMMX | RegXMM | SReg2 | SReg3 | Control | Debug | Test @@ -1618,6 +1992,51 @@ md_assemble (line) else if (i.suffix == LONG_MNEM_SUFFIX) { int op; + + for (op = i.operands; --op >= 0;) + /* Reject eight bit registers, except where the template + requires them. (eg. movzb) */ + if ((i.types[op] & Reg8) != 0 + && (i.tm.operand_types[op] & (Reg16 | Reg32 | Acc)) != 0) + { + as_bad (_("`%%%s' not allowed with `%s%c'"), + i.op[op].regs->reg_name, + i.tm.name, + i.suffix); + return; + } + /* 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) + { + /* Prohibit these changes in the 64bit mode, since + the lowering is more complicated. */ + if (flag_code == CODE_64BIT) + as_bad (_("Incorrect register `%%%s' used with`%c' suffix"), + i.op[op].regs->reg_name, + i.suffix); +#if REGISTER_WARNINGS + else + as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"), + (i.op[op].regs + REGNAM_EAX - REGNAM_AX)->reg_name, + i.op[op].regs->reg_name, + i.suffix); +#endif + } + /* Warn if the r prefix on a general reg is missing. */ + else if ((i.types[op] & Reg64) != 0 + && (i.tm.operand_types[op] & (Reg32|Acc)) != 0) + { + as_bad (_("Incorrect register `%%%s' used with`%c' suffix"), + i.op[op].regs->reg_name, + i.suffix); + } + } + else if (i.suffix == QWORD_MNEM_SUFFIX) + { + int op; + for (op = i.operands; --op >= 0; ) /* Reject eight bit registers, except where the template requires them. (eg. movzb) */ @@ -1630,22 +2049,22 @@ md_assemble (line) i.suffix); return; } -#if REGISTER_WARNINGS /* Warn if the e prefix on a general reg is missing. */ - else if ((i.types[op] & Reg16) != 0 + else if (((i.types[op] & Reg16) != 0 + || (i.types[op] & Reg32) != 0) && (i.tm.operand_types[op] & (Reg32|Acc)) != 0) { - as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"), - (i.op[op].regs + 8)->reg_name, - i.op[op].regs->reg_name, - i.suffix); + /* Prohibit these changes in the 64bit mode, since + the lowering is more complicated. */ + as_bad (_("Incorrect register `%%%s' used with`%c' suffix"), + i.op[op].regs->reg_name, + i.suffix); } -#endif } else if (i.suffix == WORD_MNEM_SUFFIX) { int op; - for (op = i.operands; --op >= 0; ) + for (op = i.operands; --op >= 0;) /* Reject eight bit registers, except where the template requires them. (eg. movzb) */ if ((i.types[op] & Reg8) != 0 @@ -1657,65 +2076,87 @@ md_assemble (line) i.suffix); return; } -#if REGISTER_WARNINGS /* Warn if the e prefix on a general reg is present. */ - else if ((i.types[op] & Reg32) != 0 + else if ((!quiet_warnings || flag_code == CODE_64BIT) + && (i.types[op] & Reg32) != 0 && (i.tm.operand_types[op] & (Reg16|Acc)) != 0) { - as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"), - (i.op[op].regs - 8)->reg_name, - i.op[op].regs->reg_name, - i.suffix); - } + /* Prohibit these changes in the 64bit mode, since + the lowering is more complicated. */ + if (flag_code == CODE_64BIT) + as_bad (_("Incorrect register `%%%s' used with`%c' suffix"), + i.op[op].regs->reg_name, + i.suffix); + else +#if REGISTER_WARNINGS + as_warn (_("using `%%%s' instead of `%%%s' due to `%c' suffix"), + (i.op[op].regs + REGNAM_AX - REGNAM_EAX)->reg_name, + i.op[op].regs->reg_name, + i.suffix); #endif + } } + else if (intel_syntax && (i.tm.opcode_modifier & IgnoreSize)) + /* Do nothing if the instruction is going to ignore the prefix. */ + ; else - abort(); + 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)) + if ((overlap0 & (Imm8 | Imm8S | Imm16 | Imm32 | Imm32S)) && overlap0 != Imm8 && overlap0 != Imm8S - && overlap0 != Imm16 && overlap0 != Imm32) + && overlap0 != Imm16 && overlap0 != Imm32S + && overlap0 != Imm32 && overlap0 != Imm64) { if (i.suffix) { overlap0 &= (i.suffix == BYTE_MNEM_SUFFIX ? (Imm8 | Imm8S) : - (i.suffix == WORD_MNEM_SUFFIX ? Imm16 : Imm32)); + (i.suffix == WORD_MNEM_SUFFIX ? Imm16 : + (i.suffix == QWORD_MNEM_SUFFIX ? Imm64 | Imm32S : Imm32))); } - else if (overlap0 == (Imm16 | Imm32)) + else if (overlap0 == (Imm16 | Imm32S | Imm32) + || overlap0 == (Imm16 | Imm32) + || overlap0 == (Imm16 | Imm32S)) { overlap0 = - (flag_16bit_code ^ (i.prefix[DATA_PREFIX] != 0)) ? Imm16 : Imm32; + ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)) ? Imm16 : Imm32S; } - else + if (overlap0 != Imm8 && overlap0 != Imm8S + && overlap0 != Imm16 && overlap0 != Imm32S + && overlap0 != Imm32 && overlap0 != Imm64) { as_bad (_("no instruction mnemonic suffix given; can't determine immediate size")); return; } } - if ((overlap1 & (Imm8 | Imm8S | Imm16 | Imm32)) + if ((overlap1 & (Imm8 | Imm8S | Imm16 | Imm32S | Imm32)) && overlap1 != Imm8 && overlap1 != Imm8S - && overlap1 != Imm16 && overlap1 != Imm32) + && overlap1 != Imm16 && overlap1 != Imm32S + && overlap1 != Imm32 && overlap1 != Imm64) { if (i.suffix) { overlap1 &= (i.suffix == BYTE_MNEM_SUFFIX ? (Imm8 | Imm8S) : - (i.suffix == WORD_MNEM_SUFFIX ? Imm16 : Imm32)); + (i.suffix == WORD_MNEM_SUFFIX ? Imm16 : + (i.suffix == QWORD_MNEM_SUFFIX ? Imm64 | Imm32S : Imm32))); } - else if (overlap1 == (Imm16 | Imm32)) + else if (overlap1 == (Imm16 | Imm32 | Imm32S) + || overlap1 == (Imm16 | Imm32) + || overlap1 == (Imm16 | Imm32S)) { overlap1 = - (flag_16bit_code ^ (i.prefix[DATA_PREFIX] != 0)) ? Imm16 : Imm32; + ((flag_code == CODE_16BIT) ^ (i.prefix[DATA_PREFIX] != 0)) ? Imm16 : Imm32S; } - else + if (overlap1 != Imm8 && overlap1 != Imm8S + && overlap1 != Imm16 && overlap1 != Imm32S + && overlap1 != Imm32 && overlap1 != Imm64) { - as_bad (_("no instruction mnemonic suffix given; can't determine immediate size")); + as_bad (_("no instruction mnemonic suffix given; can't determine immediate size %x %c"),overlap1, i.suffix); return; } } @@ -1725,7 +2166,7 @@ md_assemble (line) if (overlap0 & ImplicitRegister) i.reg_operands--; if (overlap0 & Imm1) - i.imm_operands = 0; /* kludge for shift insns */ + i.imm_operands = 0; /* kludge for shift insns. */ i.types[1] = overlap1; if (overlap1 & ImplicitRegister) @@ -1744,7 +2185,7 @@ md_assemble (line) return; } - /* For movzx and movsx, need to check the register type */ + /* For movzx and movsx, need to check the register type. */ if (intel_syntax && (i.tm.base_opcode == 0xfb6 || i.tm.base_opcode == 0xfbe)) if (i.suffix && i.suffix == BYTE_MNEM_SUFFIX) @@ -1769,8 +2210,8 @@ md_assemble (line) /* Now select between word & dword operations via the operand size prefix, except for instructions that will ignore this prefix anyway. */ - if (((intel_syntax && (i.suffix == DWORD_MNEM_SUFFIX)) - || i.suffix == LONG_MNEM_SUFFIX) == flag_16bit_code + if (i.suffix != QWORD_MNEM_SUFFIX + && (i.suffix == LONG_MNEM_SUFFIX) == (flag_code == CODE_16BIT) && !(i.tm.opcode_modifier & IgnoreSize)) { unsigned int prefix = DATA_PREFIX_OPCODE; @@ -1780,9 +2221,21 @@ md_assemble (line) if (! add_prefix (prefix)) return; } + + /* Set mode64 for an operand. */ + if (i.suffix == QWORD_MNEM_SUFFIX + && !(i.tm.opcode_modifier & NoRex64)) + { + i.rex.mode64 = 1; + if (flag_code < CODE_64BIT) + { + as_bad (_("64bit operations available only in 64bit modes.")); + return; + } + } + /* Size floating point instruction. */ - if (i.suffix == LONG_MNEM_SUFFIX - || (intel_syntax && i.suffix == DWORD_MNEM_SUFFIX)) + if (i.suffix == LONG_MNEM_SUFFIX) { if (i.tm.opcode_modifier & FloatMF) i.tm.base_opcode ^= 4; @@ -1798,7 +2251,7 @@ md_assemble (line) expressionS *exp; - assert(i.imm_operands == 0 && i.operands <= 2 && 2 < MAX_OPERANDS); + assert (i.imm_operands == 0 && i.operands <= 2 && 2 < MAX_OPERANDS); exp = &im_expressions[i.imm_operands++]; i.op[i.operands].imms = exp; @@ -1808,7 +2261,7 @@ md_assemble (line) i.tm.extension_opcode = None; } - /* For insns with operands there are more diddles to do to the opcode. */ + /* For insns with operands there are more diddles to do to the opcode. */ if (i.operands) { /* Default segment register this instruction will use @@ -1822,33 +2275,35 @@ md_assemble (line) if (i.tm.opcode_modifier & regKludge) { 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]; + /* 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 (i.tm.opcode_modifier & ShortForm) { - /* The register or float register operand is in operand 0 or 1. */ + /* The register or float register operand is in operand 0 or 1. */ unsigned int op = (i.types[0] & (Reg | FloatReg)) ? 0 : 1; - /* Register goes in low 3 bits of opcode. */ + /* Register goes in low 3 bits of opcode. */ i.tm.base_opcode |= i.op[op].regs->reg_num; - if ((i.tm.opcode_modifier & Ugh) != 0) + if (i.op[op].regs->reg_flags & RegRex) + i.rex.extZ = 1; + if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0) { /* Warn about some common errors, but press on regardless. The first case can be generated by gcc (<= 2.8.1). */ if (i.operands == 2) { - /* reversed arguments on faddp, fsubp, etc. */ + /* 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 */ + /* Extraneous `l' suffix on fp insn. */ as_warn (_("translating to `%s %%%s'"), i.tm.name, i.op[0].regs->reg_name); } @@ -1859,10 +2314,10 @@ md_assemble (line) /* 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. */ + info we've collected. */ /* i.reg_operands MUST be the number of real register operands; - implicit registers do not count. */ + implicit registers do not count. */ if (i.reg_operands == 2) { unsigned int source, dest; @@ -1885,15 +2340,23 @@ md_assemble (line) { i.rm.reg = i.op[dest].regs->reg_num; i.rm.regmem = i.op[source].regs->reg_num; + if (i.op[dest].regs->reg_flags & RegRex) + i.rex.extX = 1; + if (i.op[source].regs->reg_flags & RegRex) + i.rex.extZ = 1; } else { i.rm.reg = i.op[source].regs->reg_num; i.rm.regmem = i.op[dest].regs->reg_num; + if (i.op[dest].regs->reg_flags & RegRex) + i.rex.extZ = 1; + if (i.op[source].regs->reg_flags & RegRex) + i.rex.extX = 1; } } else - { /* if it's not 2 reg operands... */ + { /* If it's not 2 reg operands... */ if (i.mem_operands) { unsigned int fake_zero_displacement = 0; @@ -1910,65 +2373,104 @@ md_assemble (line) fake_zero_displacement = 1; if (! i.index_reg) { - /* Operand is just <disp> */ - if (flag_16bit_code ^ (i.prefix[ADDR_PREFIX] != 0)) + /* Operand is just <disp> */ + if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0)) { i.rm.regmem = NO_BASE_REGISTER_16; i.types[op] &= ~Disp; i.types[op] |= Disp16; } - else + else if (flag_code != CODE_64BIT) { i.rm.regmem = NO_BASE_REGISTER; i.types[op] &= ~Disp; i.types[op] |= Disp32; } + else + { + /* 64bit mode overwrites the 32bit + absolute addressing by RIP relative + addressing and absolute addressing + is encoded by one of the redundant + SIB forms. */ + + i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING; + i.sib.base = NO_BASE_REGISTER; + i.sib.index = NO_INDEX_REGISTER; + i.types[op] &= ~Disp; + i.types[op] |= Disp32S; + } } - else /* ! i.base_reg && i.index_reg */ + 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 */ + if (flag_code != CODE_64BIT) + i.types[op] |= Disp32; /* Must be 32 bit */ + else + i.types[op] |= Disp32S; + if (i.index_reg->reg_flags & RegRex) + i.rex.extY = 1; } } + /* RIP addressing for 64bit mode. */ + else if (i.base_reg->reg_type == BaseIndex) + { + i.rm.regmem = NO_BASE_REGISTER; + i.types[op] &= ~Disp; + i.types[op] |= Disp32S; + i.flags[op] = Operand_PCrel; + } else if (i.base_reg->reg_type & Reg16) { switch (i.base_reg->reg_num) { - case 3: /* (%bx) */ + case 3: /* (%bx) */ if (! i.index_reg) i.rm.regmem = 7; - else /* (%bx,%si) -> 0, or (%bx,%di) -> 1 */ + else /* (%bx,%si) -> 0, or (%bx,%di) -> 1 */ i.rm.regmem = i.index_reg->reg_num - 6; break; - case 5: /* (%bp) */ + 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) */ + /* fake (%bp) into 0(%bp) */ i.types[op] |= Disp8; fake_zero_displacement = 1; } } - else /* (%bp,%si) -> 2, or (%bp,%di) -> 3 */ + 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 */ + 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 */ + else /* i.base_reg and 32/64 bit mode */ { + if (flag_code == CODE_64BIT + && (i.types[op] & Disp)) + { + if (i.types[op] & Disp8) + i.types[op] = Disp8 | Disp32S; + else + i.types[op] = Disp32S; + } i.rm.regmem = i.base_reg->reg_num; + if (i.base_reg->reg_flags & RegRex) + i.rex.extZ = 1; i.sib.base = i.base_reg->reg_num; - if (i.base_reg->reg_num == EBP_REG_NUM) + /* x86-64 ignores REX prefix bit here to avoid + decoder complications. */ + if ((i.base_reg->reg_num & 7) == EBP_REG_NUM) { default_seg = &ss; if (i.disp_operands == 0) @@ -2002,6 +2504,8 @@ md_assemble (line) { i.sib.index = i.index_reg->reg_num; i.rm.regmem = ESCAPE_TO_TWO_BYTE_ADDRESSING; + if (i.index_reg->reg_flags & RegRex) + i.rex.extY = 1; } i.rm.mode = mode_from_disp_size (i.types[op]); } @@ -2009,7 +2513,7 @@ md_assemble (line) if (fake_zero_displacement) { /* Fakes a zero displacement assuming that i.types[op] - holds the correct displacement size. */ + holds the correct displacement size. */ expressionS *exp; assert (i.op[op].disps == 0); @@ -2026,7 +2530,7 @@ md_assemble (line) 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. */ + into the i.rm.reg field. */ if (i.reg_operands) { unsigned int op = @@ -2042,11 +2546,19 @@ md_assemble (line) ? 1 : 2)); /* If there is an extension opcode to put here, the - register number must be put into the regmem field. */ + register number must be put into the regmem field. */ if (i.tm.extension_opcode != None) - i.rm.regmem = i.op[op].regs->reg_num; + { + i.rm.regmem = i.op[op].regs->reg_num; + if (i.op[op].regs->reg_flags & RegRex) + i.rex.extZ = 1; + } else - i.rm.reg = i.op[op].regs->reg_num; + { + i.rm.reg = i.op[op].regs->reg_num; + if (i.op[op].regs->reg_flags & RegRex) + i.rex.extX = 1; + } /* Now, if no memory operand has set i.rm.mode = 0, 1, 2 we must set it to 3 to indicate this is a register @@ -2055,19 +2567,22 @@ md_assemble (line) i.rm.mode = 3; } - /* Fill in i.rm.reg field with extension opcode (if any). */ + /* Fill in i.rm.reg field with extension opcode (if any). */ if (i.tm.extension_opcode != None) i.rm.reg = i.tm.extension_opcode; } } else if (i.tm.opcode_modifier & (Seg2ShortForm | Seg3ShortForm)) { - if (i.tm.base_opcode == POP_SEG_SHORT && i.op[0].regs->reg_num == 1) + if (i.tm.base_opcode == POP_SEG_SHORT + && i.op[0].regs->reg_num == 1) { as_bad (_("you can't `pop %%cs'")); return; } i.tm.base_opcode |= (i.op[0].regs->reg_num << 3); + if (i.op[0].regs->reg_flags & RegRex) + i.rex.extZ = 1; } else if ((i.tm.base_opcode & ~(D|W)) == MOV_AX_DISP32) { @@ -2092,14 +2607,14 @@ md_assemble (line) return; } } - else if ((i.tm.opcode_modifier & Ugh) != 0) + else if (!quiet_warnings && (i.tm.opcode_modifier & Ugh) != 0) { /* UnixWare fsub no args is alias for fsubp, fadd -> faddp, etc. */ as_warn (_("translating to `%sp'"), i.tm.name); } } - /* Handle conversion of 'int $3' --> special int3 insn. */ + /* 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; @@ -2116,11 +2631,52 @@ md_assemble (line) i.op[0].disps->X_op = O_symbol; } - /* We are ready to output the insn. */ + if (i.tm.opcode_modifier & Rex64) + i.rex.mode64 = 1; + + /* For 8bit registers we would need an empty rex prefix. + Also in the case instruction is already having prefix, + we need to convert old registers to new ones. */ + + if (((i.types[0] & Reg8) && (i.op[0].regs->reg_flags & RegRex64)) + || ((i.types[1] & Reg8) && (i.op[1].regs->reg_flags & RegRex64)) + || ((i.rex.mode64 || i.rex.extX || i.rex.extY || i.rex.extZ || i.rex.empty) + && ((i.types[0] & Reg8) || (i.types[1] & Reg8)))) + { + int x; + i.rex.empty = 1; + for (x = 0; x < 2; x++) + { + /* Look for 8bit operand that does use old registers. */ + if (i.types[x] & Reg8 + && !(i.op[x].regs->reg_flags & RegRex64)) + { + /* In case it is "hi" register, give up. */ + if (i.op[x].regs->reg_num > 3) + as_bad (_("Can't encode registers '%%%s' in the instruction requiring REX prefix.\n"), + i.op[x].regs->reg_name); + + /* Otherwise it is equivalent to the extended register. + Since the encoding don't change this is merely cosmetical + cleanup for debug output. */ + + i.op[x].regs = i.op[x].regs + 8; + } + } + } + + if (i.rex.mode64 || i.rex.extX || i.rex.extY || i.rex.extZ || i.rex.empty) + add_prefix (0x40 + | (i.rex.mode64 ? 8 : 0) + | (i.rex.extX ? 4 : 0) + | (i.rex.extY ? 2 : 0) + | (i.rex.extZ ? 1 : 0)); + + /* We are ready to output the insn. */ { register char *p; - /* Output jumps. */ + /* Output jumps. */ if (i.tm.opcode_modifier & Jump) { int size; @@ -2128,7 +2684,7 @@ md_assemble (line) int prefix; code16 = 0; - if (flag_16bit_code) + if (flag_code == CODE_16BIT) code16 = CODE16; prefix = 0; @@ -2138,6 +2694,11 @@ md_assemble (line) i.prefixes -= 1; code16 ^= CODE16; } + if (i.prefix[REX_PREFIX]) + { + prefix++; + i.prefixes--; + } size = 4; if (code16) @@ -2155,8 +2716,10 @@ md_assemble (line) insn_size += prefix + 1; /* Prefix and 1 opcode byte go in fr_fix. */ p = frag_more (prefix + 1); - if (prefix) + if (i.prefix[DATA_PREFIX]) *p++ = DATA_PREFIX_OPCODE; + if (i.prefix[REX_PREFIX]) + *p++ = i.prefix[REX_PREFIX]; *p = i.tm.base_opcode; /* 1 possible extra opcode + displacement go in var part. Pass reloc in fr_var. */ @@ -2190,7 +2753,7 @@ md_assemble (line) int code16; code16 = 0; - if (flag_16bit_code) + if (flag_code == CODE_16BIT) code16 = CODE16; if (i.prefix[DATA_PREFIX]) @@ -2206,6 +2769,13 @@ md_assemble (line) size = 2; } + if (i.prefix[REX_PREFIX]) + { + FRAG_APPEND_1_CHAR (i.prefix[REX_PREFIX]); + insn_size++; + i.prefixes -= 1; + } + if (i.prefixes != 0 && !intel_syntax) as_warn (_("skipping prefixes on this instruction")); @@ -2216,7 +2786,7 @@ md_assemble (line) } else { - /* opcode can be at most two bytes */ + /* Opcode can be at most two bytes. */ insn_size += 2 + size; p = frag_more (2 + size); *p++ = (i.tm.base_opcode >> 8) & 0xff; @@ -2224,7 +2794,7 @@ md_assemble (line) *p++ = i.tm.base_opcode & 0xff; fix_new_exp (frag_now, p - frag_now->fr_literal, size, - i.op[0].disps, 1, reloc (size, 1, i.disp_reloc[0])); + i.op[0].disps, 1, reloc (size, 1, 1, i.disp_reloc[0])); } else if (i.tm.opcode_modifier & JumpInterSegment) { @@ -2233,7 +2803,7 @@ md_assemble (line) int code16; code16 = 0; - if (flag_16bit_code) + if (flag_code == CODE_16BIT) code16 = CODE16; prefix = 0; @@ -2243,6 +2813,11 @@ md_assemble (line) i.prefixes -= 1; code16 ^= CODE16; } + if (i.prefix[REX_PREFIX]) + { + prefix++; + i.prefixes -= 1; + } size = 4; if (code16) @@ -2251,14 +2826,20 @@ md_assemble (line) if (i.prefixes != 0 && !intel_syntax) as_warn (_("skipping prefixes on this instruction")); - insn_size += prefix + 1 + 2 + size; /* 1 opcode; 2 segment; offset */ + /* 1 opcode; 2 segment; offset */ + insn_size += prefix + 1 + 2 + size; p = frag_more (prefix + 1 + 2 + size); - if (prefix) + + if (i.prefix[DATA_PREFIX]) *p++ = DATA_PREFIX_OPCODE; + + if (i.prefix[REX_PREFIX]) + *p++ = i.prefix[REX_PREFIX]; + *p++ = i.tm.base_opcode; if (i.op[1].imms->X_op == O_constant) { - long n = (long) i.op[1].imms->X_add_number; + offsetT n = i.op[1].imms->X_add_number; if (size == 2 && !fits_in_unsigned_word (n) @@ -2267,11 +2848,11 @@ md_assemble (line) as_bad (_("16-bit jump out of range")); return; } - md_number_to_chars (p, (valueT) n, size); + md_number_to_chars (p, n, size); } else fix_new_exp (frag_now, p - frag_now->fr_literal, size, - i.op[1].imms, 0, reloc (size, 0, i.disp_reloc[0])); + i.op[1].imms, 0, reloc (size, 0, 0, i.disp_reloc[0])); if (i.op[0].imms->X_op != O_constant) as_bad (_("can't handle non absolute segment in `%s'"), i.tm.name); @@ -2279,10 +2860,15 @@ md_assemble (line) } else { - /* Output normal instructions here. */ + /* Output normal instructions here. */ unsigned char *q; - /* The prefix bytes. */ + /* All opcodes on i386 have eighter 1 or 2 bytes. We may use third + byte for the SSE instructions to specify 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++) @@ -2295,37 +2881,20 @@ md_assemble (line) } } - /* Now the opcode; be careful about word order here! */ + /* 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)) + else { insn_size += 2; p = frag_more (2); - /* put out high byte first: can't use md_number_to_chars! */ + /* Put out high byte first: can't use md_number_to_chars! */ *p++ = (i.tm.base_opcode >> 8) & 0xff; *p = i.tm.base_opcode & 0xff; } - else - { /* opcode is either 3 or 4 bytes */ - if (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) @@ -2365,47 +2934,78 @@ md_assemble (line) { if (i.op[n].disps->X_op == O_constant) { - int size = 4; - long val = (long) i.op[n].disps->X_add_number; + int size; + offsetT val; - if (i.types[n] & (Disp8 | Disp16)) + size = 4; + if (i.types[n] & (Disp8 | Disp16 | Disp64)) { - 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); + size = 1; + if (i.types[n] & Disp64) + size = 8; } + val = offset_in_range (i.op[n].disps->X_add_number, + size); insn_size += size; p = frag_more (size); - md_number_to_chars (p, (valueT) val, size); + md_number_to_chars (p, val, size); } else { int size = 4; + int sign = 0; + int pcrel = (i.flags[n] & Operand_PCrel) != 0; - if (i.types[n] & Disp16) - size = 2; + /* The PC relative address is computed relative + to the instruction boundary, so in case immediate + fields follows, we need to adjust the value. */ + if (pcrel && i.imm_operands) + { + int imm_size = 4; + register unsigned int n1; + + for (n1 = 0; n1 < i.operands; n1++) + if (i.types[n1] & Imm) + { + if (i.types[n1] & (Imm8 | Imm8S | Imm16 | Imm64)) + { + imm_size = 2; + if (i.types[n1] & (Imm8 | Imm8S)) + imm_size = 1; + if (i.types[n1] & Imm64) + imm_size = 8; + } + break; + } + /* We should find the immediate. */ + if (n1 == i.operands) + abort (); + i.op[n].disps->X_add_number -= imm_size; + } + + if (i.types[n] & Disp32S) + sign = 1; + + if (i.types[n] & (Disp16 | Disp64)) + { + size = 2; + if (i.types[n] & Disp64) + size = 8; + } insn_size += size; p = frag_more (size); fix_new_exp (frag_now, p - frag_now->fr_literal, size, - i.op[n].disps, 0, - reloc (size, 0, i.disp_reloc[n])); + i.op[n].disps, pcrel, + reloc (size, pcrel, sign, i.disp_reloc[n])); } } } - } /* end displacement output */ + } - /* output immediate */ + /* Output immediate. */ if (i.imm_operands) { register unsigned int n; @@ -2416,48 +3016,53 @@ md_assemble (line) { if (i.op[n].imms->X_op == O_constant) { - int size = 4; - long val = (long) i.op[n].imms->X_add_number; + int size; + offsetT val; - if (i.types[n] & (Imm8 | Imm8S | Imm16)) + size = 4; + if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64)) { - 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); + size = 1; + else if (i.types[n] & Imm64) + size = 8; } + val = offset_in_range (i.op[n].imms->X_add_number, + size); insn_size += size; p = frag_more (size); - md_number_to_chars (p, (valueT) val, size); + md_number_to_chars (p, val, size); } else - { /* not absolute_section */ - /* Need a 32-bit fixup (don't support 8bit + { + /* Not absolute_section. + Need a 32-bit fixup (don't support 8bit non-absolute imms). Try to support other - sizes ... */ + sizes ... */ #ifdef BFD_ASSEMBLER enum bfd_reloc_code_real reloc_type; #else int reloc_type; #endif int size = 4; + int sign = 0; - if (i.types[n] & Imm16) - size = 2; - else if (i.types[n] & (Imm8 | Imm8S)) - size = 1; + if ((i.types[n] & (Imm32S)) + && i.suffix == QWORD_MNEM_SUFFIX) + sign = 1; + if (i.types[n] & (Imm8 | Imm8S | Imm16 | Imm64)) + { + size = 2; + if (i.types[n] & (Imm8 | Imm8S)) + size = 1; + if (i.types[n] & Imm64) + size = 8; + } insn_size += size; p = frag_more (size); - reloc_type = reloc (size, 0, i.disp_reloc[0]); + reloc_type = reloc (size, 0, sign, i.disp_reloc[0]); #ifdef BFD_ASSEMBLER if (reloc_type == BFD_RELOC_32 && GOT_symbol @@ -2468,6 +3073,9 @@ md_assemble (line) (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; } @@ -2477,15 +3085,17 @@ md_assemble (line) } } } - } /* end immediate output */ + } } + dwarf2_emit_insn (insn_size); + #ifdef DEBUG386 if (flag_debug) { pi (line, &i); } -#endif /* DEBUG386 */ +#endif /* DEBUG386 */ } } @@ -2497,7 +3107,7 @@ i386_immediate (imm_start) { char *save_input_line_pointer; segT exp_seg = 0; - expressionS * exp; + expressionS *exp; if (i.imm_operands == MAX_IMMEDIATE_OPERANDS) { @@ -2516,12 +3126,10 @@ i386_immediate (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... - */ + /* 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, '@'); @@ -2531,8 +3139,8 @@ i386_immediate (imm_start) int len = 0; int first; - /* GOT relocations are not supported in 16 bit mode */ - if (flag_16bit_code) + /* GOT relocations are not supported in 16 bit mode. */ + if (flag_code == CODE_16BIT) as_bad (_("GOT relocations not supported in 16 bit mode")); if (GOT_symbol == NULL) @@ -2540,17 +3148,33 @@ i386_immediate (imm_start) if (strncmp (cp + 1, "PLT", 3) == 0) { - i.disp_reloc[this_operand] = BFD_RELOC_386_PLT32; + if (flag_code == CODE_64BIT) + i.disp_reloc[this_operand] = BFD_RELOC_X86_64_PLT32; + else + i.disp_reloc[this_operand] = BFD_RELOC_386_PLT32; len = 3; } else if (strncmp (cp + 1, "GOTOFF", 6) == 0) { + if (flag_code == CODE_64BIT) + as_bad ("GOTOFF relocations are unsupported in 64bit mode."); i.disp_reloc[this_operand] = BFD_RELOC_386_GOTOFF; len = 6; } + else if (strncmp (cp + 1, "GOTPCREL", 8) == 0) + { + if (flag_code == CODE_64BIT) + i.disp_reloc[this_operand] = BFD_RELOC_X86_64_GOTPCREL; + else + as_bad ("GOTPCREL relocations are supported only in 64bit mode."); + len = 8; + } else if (strncmp (cp + 1, "GOT", 3) == 0) { - i.disp_reloc[this_operand] = BFD_RELOC_386_GOT32; + if (flag_code == CODE_64BIT) + i.disp_reloc[this_operand] = BFD_RELOC_X86_64_GOT32; + else + i.disp_reloc[this_operand] = BFD_RELOC_386_GOT32; len = 3; } else @@ -2559,7 +3183,7 @@ i386_immediate (imm_start) /* 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)); + tmpbuf = (char *) alloca (strlen (input_line_pointer)); memcpy (tmpbuf, input_line_pointer, first); tmpbuf[first] = ' '; strcpy (tmpbuf + first + 1, cp + 1 + len); @@ -2578,7 +3202,7 @@ i386_immediate (imm_start) if (exp->X_op == O_absent || exp->X_op == O_big) { - /* missing or bad expr becomes absolute 0 */ + /* 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; @@ -2586,17 +3210,21 @@ i386_immediate (imm_start) exp->X_add_symbol = (symbolS *) 0; exp->X_op_symbol = (symbolS *) 0; } - - if (exp->X_op == O_constant) + else if (exp->X_op == O_constant) { - i.types[this_operand] |= Imm32; /* Size it properly later. */ + /* 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 ( + else if (1 #ifdef BFD_ASSEMBLER - OUTPUT_FLAVOR == bfd_target_aout_flavour && + && OUTPUT_FLAVOR == bfd_target_aout_flavour #endif - exp_seg != text_section + && exp_seg != text_section && exp_seg != data_section && exp_seg != bss_section && exp_seg != undefined_section @@ -2617,10 +3245,8 @@ i386_immediate (imm_start) { /* This is an address. The size of the address will be determined later, depending on destination register, - suffix, or the default for the section. We exclude - Imm8S here so that `push $foo' and other instructions - with an Imm8S form will use Imm16 or Imm32. */ - i.types[this_operand] |= (Imm8 | Imm16 | Imm32); + suffix, or the default for the section. */ + i.types[this_operand] |= Imm8 | Imm16 | Imm32 | Imm32S | Imm64; } return 1; @@ -2679,8 +3305,10 @@ i386_displacement (disp_start, disp_end) char *save_input_line_pointer; int bigdisp = Disp32; - if (flag_16bit_code ^ (i.prefix[ADDR_PREFIX] != 0)) + if ((flag_code == CODE_16BIT) ^ (i.prefix[ADDR_PREFIX] != 0)) bigdisp = Disp16; + if (flag_code == CODE_64BIT) + bigdisp = Disp64; i.types[this_operand] |= bigdisp; exp = &disp_expressions[i.disp_operands]; @@ -2729,20 +3357,17 @@ i386_displacement (disp_start, disp_end) rorl $16,%edx #NO_APP - So here we provide the missing zero. - */ + So here we provide the missing zero. */ *displacement_string_end = '0'; } #endif #ifndef LEX_AT { - /* - * We can have operands of the form - * <symbol>@GOTOFF+<nnn> - * Take the easy way out here and copy everything - * into a temporary buffer... - */ + /* 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, '@'); @@ -2752,8 +3377,8 @@ i386_displacement (disp_start, disp_end) int len = 0; int first; - /* GOT relocations are not supported in 16 bit mode */ - if (flag_16bit_code) + /* GOT relocations are not supported in 16 bit mode. */ + if (flag_code == CODE_16BIT) as_bad (_("GOT relocations not supported in 16 bit mode")); if (GOT_symbol == NULL) @@ -2761,17 +3386,32 @@ i386_displacement (disp_start, disp_end) if (strncmp (cp + 1, "PLT", 3) == 0) { - i.disp_reloc[this_operand] = BFD_RELOC_386_PLT32; + if (flag_code == CODE_64BIT) + i.disp_reloc[this_operand] = BFD_RELOC_X86_64_PLT32; + else + i.disp_reloc[this_operand] = BFD_RELOC_386_PLT32; len = 3; } else if (strncmp (cp + 1, "GOTOFF", 6) == 0) { + if (flag_code == CODE_64BIT) + as_bad ("GOTOFF relocation is not supported in 64bit mode."); i.disp_reloc[this_operand] = BFD_RELOC_386_GOTOFF; len = 6; } + else if (strncmp (cp + 1, "GOTPCREL", 8) == 0) + { + if (flag_code != CODE_64BIT) + as_bad ("GOTPCREL relocation is supported only in 64bit mode."); + i.disp_reloc[this_operand] = BFD_RELOC_X86_64_GOTPCREL; + len = 8; + } else if (strncmp (cp + 1, "GOT", 3) == 0) { - i.disp_reloc[this_operand] = BFD_RELOC_386_GOT32; + if (flag_code == CODE_64BIT) + i.disp_reloc[this_operand] = BFD_RELOC_X86_64_GOT32; + else + i.disp_reloc[this_operand] = BFD_RELOC_386_GOT32; len = 3; } else @@ -2780,7 +3420,7 @@ i386_displacement (disp_start, disp_end) /* 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)); + tmpbuf = (char *) alloca (strlen (input_line_pointer)); memcpy (tmpbuf, input_line_pointer, first); tmpbuf[first] = ' '; strcpy (tmpbuf + first + 1, cp + 1 + len); @@ -2794,16 +3434,20 @@ i386_displacement (disp_start, disp_end) #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) + to be relative to the beginning of the section. */ + if (i.disp_reloc[this_operand] == BFD_RELOC_386_GOTOFF + || i.disp_reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL) { - if (S_IS_LOCAL(exp->X_add_symbol) + 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; + if (i.disp_reloc[this_operand] == BFD_RELOC_X86_64_GOTPCREL) + i.disp_reloc[this_operand] = BFD_RELOC_32_PCREL; + else + i.disp_reloc[this_operand] = BFD_RELOC_32; } #endif @@ -2819,7 +3463,7 @@ i386_displacement (disp_start, disp_end) if (exp->X_op == O_absent || exp->X_op == O_big) { - /* missing or bad expr becomes absolute 0 */ + /* 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; @@ -2828,28 +3472,15 @@ i386_displacement (disp_start, disp_end) exp->X_op_symbol = (symbolS *) 0; } - if (exp->X_op == O_constant) - { - if (i.types[this_operand] & Disp16) - { - /* We know this operand is at most 16 bits, so convert to a - signed 16 bit number before trying to see whether it will - fit in an even smaller size. */ - exp->X_add_number = - (((exp->X_add_number & 0xffff) ^ 0x8000) - 0x8000); - } - if (fits_in_signed_byte (exp->X_add_number)) - i.types[this_operand] |= Disp8; - } #if (defined (OBJ_AOUT) || defined (OBJ_MAYBE_AOUT)) - else if ( + if (exp->X_op != O_constant #ifdef BFD_ASSEMBLER - OUTPUT_FLAVOR == bfd_target_aout_flavour && + && OUTPUT_FLAVOR == bfd_target_aout_flavour #endif - exp_seg != text_section - && exp_seg != data_section - && exp_seg != bss_section - && exp_seg != undefined_section) + && 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); @@ -2859,239 +3490,72 @@ i386_displacement (disp_start, disp_end) return 0; } #endif + else if (flag_code == CODE_64BIT) + i.types[this_operand] |= Disp32S | Disp32; return 1; } -static int i386_operand_modifier PARAMS ((char **, int)); - -static int -i386_operand_modifier (op_string, got_a_float) - char **op_string; - int got_a_float; -{ - if (!strncasecmp (*op_string, "BYTE PTR", 8)) - { - i.suffix = BYTE_MNEM_SUFFIX; - *op_string += 8; - return BYTE_PTR; - - } - else if (!strncasecmp (*op_string, "WORD PTR", 8)) - { - if (got_a_float == 2) /* "fi..." */ - i.suffix = SHORT_MNEM_SUFFIX; - else - i.suffix = WORD_MNEM_SUFFIX; - *op_string += 8; - return WORD_PTR; - } - - else if (!strncasecmp (*op_string, "DWORD PTR", 9)) - { - if (got_a_float == 1) /* "f..." */ - i.suffix = SHORT_MNEM_SUFFIX; - else - i.suffix = LONG_MNEM_SUFFIX; - *op_string += 9; - return DWORD_PTR; - } - - else if (!strncasecmp (*op_string, "QWORD PTR", 9)) - { - i.suffix = DWORD_MNEM_SUFFIX; - *op_string += 9; - return QWORD_PTR; - } - - else if (!strncasecmp (*op_string, "XWORD PTR", 9)) - { - i.suffix = LONG_DOUBLE_MNEM_SUFFIX; - *op_string += 9; - return XWORD_PTR; - } - - else if (!strncasecmp (*op_string, "SHORT", 5)) - { - *op_string += 5; - return SHORT; - } - - else if (!strncasecmp (*op_string, "OFFSET FLAT:", 12)) - { - *op_string += 12; - return OFFSET_FLAT; - } - - else if (!strncasecmp (*op_string, "FLAT", 4)) - { - *op_string += 4; - return FLAT; - } - - else return NONE_FOUND; -} - -static char * build_displacement_string PARAMS ((int, char *)); - -static char * -build_displacement_string (initial_disp, op_string) - int initial_disp; - char *op_string; -{ - char *temp_string = (char *) malloc (strlen (op_string) + 1); - char *end_of_operand_string; - char *tc; - char *temp_disp; - - temp_string[0] = '\0'; - tc = end_of_operand_string = strchr (op_string, '['); - if (initial_disp && !end_of_operand_string) - { - strcpy (temp_string, op_string); - return temp_string; - } - - /* Build the whole displacement string */ - if (initial_disp) - { - strncpy (temp_string, op_string, end_of_operand_string - op_string); - temp_string[end_of_operand_string - op_string] = '\0'; - temp_disp = tc; - } - else - temp_disp = op_string; - - while (*temp_disp != '\0') - { - char *end_op; - int add_minus = (*temp_disp == '-'); - - if (*temp_disp == '+' || *temp_disp == '-' || *temp_disp == '[') - temp_disp++; - - if (is_space_char (*temp_disp)) - temp_disp++; - - /* Don't consider registers */ - if ( !((*temp_disp == REGISTER_PREFIX || allow_naked_reg) - && parse_register (temp_disp, &end_op)) ) - { - char *string_start = temp_disp; - - while (*temp_disp != ']' - && *temp_disp != '+' - && *temp_disp != '-' - && *temp_disp != '*') - ++temp_disp; - - if (add_minus) - strcat (temp_string, "-"); - else - strcat (temp_string, "+"); - - strncat (temp_string, string_start, temp_disp - string_start); - if (*temp_disp == '+' || *temp_disp == '-') - --temp_disp; - } - - while (*temp_disp != '\0' - && *temp_disp != '+' - && *temp_disp != '-') - ++temp_disp; - } - - return temp_string; -} - -static int i386_parse_seg PARAMS ((char *)); - -static int -i386_parse_seg (op_string) - char *op_string; -{ - if (is_space_char (*op_string)) - ++op_string; - - /* Should be one of es, cs, ss, ds fs or gs */ - switch (*op_string++) - { - case 'e': - i.seg[i.mem_operands] = &es; - break; - case 'c': - i.seg[i.mem_operands] = &cs; - break; - case 's': - i.seg[i.mem_operands] = &ss; - break; - case 'd': - i.seg[i.mem_operands] = &ds; - break; - case 'f': - i.seg[i.mem_operands] = &fs; - break; - case 'g': - i.seg[i.mem_operands] = &gs; - break; - default: - as_bad (_("bad segment name `%s'"), op_string); - return 0; - } - - if (*op_string++ != 's') - { - as_bad (_("bad segment name `%s'"), op_string); - return 0; - } - - if (is_space_char (*op_string)) - ++op_string; - - if (*op_string != ':') - { - as_bad (_("bad segment name `%s'"), op_string); - return 0; - } - - return 1; - -} - -static int i386_index_check PARAMS((const char *)); +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. -*/ + 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 - 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))))) + ok = 1; + if (flag_code == CODE_64BIT) + { + /* 64bit checks. */ + if ((i.base_reg + && ((i.base_reg->reg_type & Reg64) == 0) + && (i.base_reg->reg_type != BaseIndex + || i.index_reg)) + || (i.index_reg + && ((i.index_reg->reg_type & (Reg64|BaseIndex)) + != (Reg64|BaseIndex)))) + ok = 0; + } + else + { + if ((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 (i.prefix[ADDR_PREFIX] == 0 && stackop_size != '\0') + if (flag_code != CODE_64BIT + && i.prefix[ADDR_PREFIX] == 0 && stackop_size != '\0') { i.prefix[ADDR_PREFIX] = ADDR_PREFIX_OPCODE; i.prefixes += 1; @@ -3099,8 +3563,7 @@ i386_index_check (operand_string) 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. - */ + 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; @@ -3113,260 +3576,14 @@ i386_index_check (operand_string) #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"); + flag_code_names[flag_code]); return 0; } return 1; } -static int i386_intel_memory_operand PARAMS ((char *)); - -static int -i386_intel_memory_operand (operand_string) - char *operand_string; -{ - char *op_string = operand_string; - char *end_of_operand_string; - - if ((i.mem_operands == 1 - && (current_templates->start->opcode_modifier & IsString) == 0) - || i.mem_operands == 2) - { - as_bad (_("too many memory references for `%s'"), - current_templates->start->name); - return 0; - } - - /* First check for a segment override. */ - if (*op_string != '[') - { - char *end_seg; - - end_seg = strchr (op_string, ':'); - if (end_seg) - { - if (!i386_parse_seg (op_string)) - return 0; - op_string = end_seg + 1; - } - } - - /* Look for displacement preceding open bracket */ - if (*op_string != '[') - { - char *temp_string; - - if (i.disp_operands) - return 0; - - temp_string = build_displacement_string (true, op_string); - - if (!i386_displacement (temp_string, temp_string + strlen (temp_string))) - { - free (temp_string); - return 0; - } - free (temp_string); - - end_of_operand_string = strchr (op_string, '['); - if (!end_of_operand_string) - end_of_operand_string = op_string + strlen (op_string); - - if (is_space_char (*end_of_operand_string)) - --end_of_operand_string; - - op_string = end_of_operand_string; - } - - if (*op_string == '[') - { - ++op_string; - - /* Pick off each component and figure out where it belongs */ - - end_of_operand_string = op_string; - - while (*op_string != ']') - { - const reg_entry *temp_reg; - char *end_op; - char *temp_string; - - while (*end_of_operand_string != '+' - && *end_of_operand_string != '-' - && *end_of_operand_string != '*' - && *end_of_operand_string != ']') - end_of_operand_string++; - - temp_string = op_string; - if (*temp_string == '+') - { - ++temp_string; - if (is_space_char (*temp_string)) - ++temp_string; - } - - if ((*temp_string == REGISTER_PREFIX || allow_naked_reg) - && (temp_reg = parse_register (temp_string, &end_op)) != NULL) - { - if (i.base_reg == NULL) - i.base_reg = temp_reg; - else - i.index_reg = temp_reg; - - i.types[this_operand] |= BaseIndex; - } - else if (*temp_string == REGISTER_PREFIX) - { - as_bad (_("bad register name `%s'"), temp_string); - return 0; - } - else if (is_digit_char (*op_string) - || *op_string == '+' || *op_string == '-') - { - char *temp_str; - - if (i.disp_operands != 0) - return 0; - - temp_string = build_displacement_string (false, op_string); - - temp_str = temp_string; - if (*temp_str == '+') - ++temp_str; - - if (!i386_displacement (temp_str, temp_str + strlen (temp_str))) - { - free (temp_string); - return 0; - } - free (temp_string); - - ++op_string; - end_of_operand_string = op_string; - while (*end_of_operand_string != ']' - && *end_of_operand_string != '+' - && *end_of_operand_string != '-' - && *end_of_operand_string != '*') - ++end_of_operand_string; - } - else if (*op_string == '*') - { - ++op_string; - - if (i.base_reg && !i.index_reg) - { - i.index_reg = i.base_reg; - i.base_reg = 0; - } - - if (!i386_scale (op_string)) - return 0; - } - op_string = end_of_operand_string; - ++end_of_operand_string; - } - } - - if (i386_index_check (operand_string) == 0) - return 0; - - i.mem_operands++; - return 1; -} - -static int -i386_intel_operand (operand_string, got_a_float) - char *operand_string; - int got_a_float; -{ - const reg_entry * r; - char *end_op; - char *op_string = operand_string; - - int operand_modifier = i386_operand_modifier (&op_string, got_a_float); - if (is_space_char (*op_string)) - ++op_string; - - switch (operand_modifier) - { - case BYTE_PTR: - case WORD_PTR: - case DWORD_PTR: - case QWORD_PTR: - case XWORD_PTR: - if (!i386_intel_memory_operand (op_string)) - return 0; - break; - - case FLAT: - case OFFSET_FLAT: - if (!i386_immediate (op_string)) - return 0; - break; - - case SHORT: - case NONE_FOUND: - /* Should be register or immediate */ - if (is_digit_char (*op_string) - && strchr (op_string, '[') == 0) - { - if (!i386_immediate (op_string)) - return 0; - } - else if ((*op_string == REGISTER_PREFIX || allow_naked_reg) - && (r = parse_register (op_string, &end_op)) != NULL) - { - /* Check for a segment override by searching for ':' after a - segment register. */ - op_string = end_op; - if (is_space_char (*op_string)) - ++op_string; - if (*op_string == ':' && (r->reg_type & (SReg2 | SReg3))) - { - switch (r->reg_num) - { - case 0: - i.seg[i.mem_operands] = &es; - break; - case 1: - i.seg[i.mem_operands] = &cs; - break; - case 2: - i.seg[i.mem_operands] = &ss; - break; - case 3: - i.seg[i.mem_operands] = &ds; - break; - case 4: - i.seg[i.mem_operands] = &fs; - break; - case 5: - i.seg[i.mem_operands] = &gs; - break; - } - - } - i.types[this_operand] |= r->reg_type & ~BaseIndex; - i.op[this_operand].regs = r; - i.reg_operands++; - } - else if (*op_string == REGISTER_PREFIX) - { - as_bad (_("bad register name `%s'"), op_string); - return 0; - } - else if (!i386_intel_memory_operand (op_string)) - return 0; - - break; - } /* end switch */ - - return 1; -} - /* Parse OPERAND_STRING into the i386_insn structure I. Returns non-zero - on error. */ + on error. */ static int i386_operand (operand_string) @@ -3380,7 +3597,7 @@ i386_operand (operand_string) ++op_string; /* We check for an absolute prefix (differentiating, - for example, 'jmp pc_relative_label' from 'jmp *absolute_label'. */ + for example, 'jmp pc_relative_label' from 'jmp *absolute_label'. */ if (*op_string == ABSOLUTE_PREFIX) { ++op_string; @@ -3389,7 +3606,7 @@ i386_operand (operand_string) i.types[this_operand] |= JumpAbsolute; } - /* Check if operand is a register. */ + /* Check if operand is a register. */ if ((*op_string == REGISTER_PREFIX || allow_naked_reg) && (r = parse_register (op_string, &end_op)) != NULL) { @@ -3435,7 +3652,7 @@ i386_operand (operand_string) as_bad (_("bad memory operand `%s'"), op_string); return 0; } - /* Handle case of %es:*foo. */ + /* Handle case of %es:*foo. */ if (*op_string == ABSOLUTE_PREFIX) { ++op_string; @@ -3460,7 +3677,7 @@ i386_operand (operand_string) return 0; } else if (*op_string == IMMEDIATE_PREFIX) - { /* ... or an immediate */ + { ++op_string; if (i.types[this_operand] & JumpAbsolute) { @@ -3472,12 +3689,12 @@ i386_operand (operand_string) } else if (is_digit_char (*op_string) || is_identifier_char (*op_string) - || *op_string == '(' ) + || *op_string == '(') { - /* This is a memory reference of some sort. */ + /* This is a memory reference of some sort. */ char *base_string; - /* Start and end of displacement string expression (if found). */ + /* Start and end of displacement string expression (if found). */ char *displacement_string_start; char *displacement_string_end; @@ -3501,7 +3718,7 @@ i386_operand (operand_string) if (is_space_char (*base_string)) --base_string; - /* If we only have a displacement, set-up for it to be parsed later. */ + /* 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; @@ -3510,7 +3727,7 @@ i386_operand (operand_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. */ + equal, so this loop will not be infinite. */ do { base_string--; @@ -3562,7 +3779,7 @@ i386_operand (operand_string) if (is_space_char (*base_string)) ++base_string; } - else if (*base_string != ')' ) + else if (*base_string != ')') { as_bad (_("expecting `,' or `)' after index register in `%s'"), operand_string); @@ -3575,7 +3792,7 @@ i386_operand (operand_string) return 0; } - /* Check for scale factor. */ + /* Check for scale factor. */ if (isdigit ((unsigned char) *base_string)) { if (!i386_scale (base_string)) @@ -3639,14 +3856,15 @@ i386_operand (operand_string) i.mem_operands++; } else - { /* it's not a memory operand; argh! */ + { + /* It's not a memory operand; argh! */ as_bad (_("invalid char %s beginning operand %d `%s'"), output_invalid (*op_string), this_operand + 1, op_string); return 0; } - return 1; /* normal return */ + return 1; /* Normal return. */ } /* md_estimate_size_before_relax() @@ -3672,7 +3890,7 @@ md_estimate_size_before_relax (fragP, segment) 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) || defined (TE_PE) +#if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) || S_IS_EXTERNAL (fragP->fr_symbol) || S_IS_WEAK (fragP->fr_symbol) #endif @@ -3701,8 +3919,9 @@ md_estimate_size_before_relax (fragP, segment) switch (opcode[0]) { - case JUMP_PC_RELATIVE: /* make jmp (0xeb) a dword displacement jump */ - opcode[0] = 0xe9; /* dword disp jmp */ + case JUMP_PC_RELATIVE: + /* Make jmp (0xeb) a dword displacement jump. */ + opcode[0] = 0xe9; fragP->fr_fix += size; fix_new (fragP, old_fr_fix, size, fragP->fr_symbol, @@ -3715,7 +3934,8 @@ md_estimate_size_before_relax (fragP, segment) 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 */ + /* 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, @@ -3725,20 +3945,19 @@ md_estimate_size_before_relax (fragP, segment) frag_wane (fragP); return fragP->fr_fix - old_fr_fix; } - return 1; /* Guess a short jump. */ + /* Guess a short jump. */ + return 1; } -/* - * 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". - */ +/* 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) @@ -3755,23 +3974,29 @@ md_convert_frag (abfd, sec, fragP) { register unsigned char *opcode; unsigned char *where_to_put_displacement = NULL; - unsigned int target_address; - unsigned int opcode_address; + offsetT target_address; + offsetT opcode_address; unsigned int extension = 0; - int displacement_from_opcode_start; + offsetT displacement_from_opcode_start; opcode = (unsigned char *) fragP->fr_opcode; - /* Address we want to reach in file space. */ + /* Address we want to reach in file space. */ target_address = S_GET_VALUE (fragP->fr_symbol) + fragP->fr_offset; -#ifdef BFD_ASSEMBLER /* not needed otherwise? */ - target_address += symbol_get_frag (fragP->fr_symbol)->fr_address; +#ifdef BFD_ASSEMBLER + /* Not needed otherwise? */ + { + /* Local symbols which have already been resolved have a NULL frags. */ + fragS *sym_frag = symbol_get_frag (fragP->fr_symbol); + if (sym_frag) + target_address += sym_frag->fr_address; + } #endif - /* Address opcode resides at in file space. */ + /* Address opcode resides at in file space. */ opcode_address = fragP->fr_address + fragP->fr_fix; - /* Displacement from opcode start to fill into instruction. */ + /* Displacement from opcode start to fill into instruction. */ displacement_from_opcode_start = target_address - opcode_address; switch (fragP->fr_subtype) @@ -3780,33 +4005,33 @@ md_convert_frag (abfd, sec, fragP) 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 */ + /* Don't have to change opcode. */ + extension = 1; /* 1 opcode + 1 displacement */ where_to_put_displacement = &opcode[1]; break; case ENCODE_RELAX_STATE (COND_JUMP, BIG): - extension = 5; /* 2 opcode + 4 displacement */ + extension = 5; /* 2 opcode + 4 displacement */ opcode[1] = opcode[0] + 0x10; opcode[0] = TWO_BYTE_OPCODE_ESCAPE; where_to_put_displacement = &opcode[2]; break; case ENCODE_RELAX_STATE (UNCOND_JUMP, BIG): - extension = 4; /* 1 opcode + 4 displacement */ + extension = 4; /* 1 opcode + 4 displacement */ opcode[0] = 0xe9; where_to_put_displacement = &opcode[1]; break; case ENCODE_RELAX_STATE (COND_JUMP, BIG16): - extension = 3; /* 2 opcode + 2 displacement */ + 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 (UNCOND_JUMP, BIG16): - extension = 2; /* 1 opcode + 2 displacement */ + extension = 2; /* 1 opcode + 2 displacement */ opcode[0] = 0xe9; where_to_put_displacement = &opcode[1]; break; @@ -3815,17 +4040,21 @@ md_convert_frag (abfd, sec, fragP) BAD_CASE (fragP->fr_subtype); break; } - /* now put displacement after opcode */ + /* Now put displacement after opcode. */ md_number_to_chars ((char *) where_to_put_displacement, (valueT) (displacement_from_opcode_start - extension), SIZE_FROM_RELAX_STATE (fragP->fr_subtype)); fragP->fr_fix += extension; } +/* Size of byte displacement jmp. */ +int md_short_jump_size = 2; -int md_short_jump_size = 2; /* size of byte displacement jmp */ -int md_long_jump_size = 5; /* size of dword displacement jmp */ -const int md_reloc_size = 8; /* Size of relocation record */ +/* Size of 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) @@ -3834,10 +4063,11 @@ md_create_short_jump (ptr, from_addr, to_addr, frag, to_symbol) fragS *frag ATTRIBUTE_UNUSED; symbolS *to_symbol ATTRIBUTE_UNUSED; { - long offset; + offsetT offset; offset = to_addr - (from_addr + 2); - md_number_to_chars (ptr, (valueT) 0xeb, 1); /* opcode for byte-disp jump */ + /* Opcode for byte-disp jump. */ + md_number_to_chars (ptr, (valueT) 0xeb, 1); md_number_to_chars (ptr + 1, (valueT) offset, 1); } @@ -3845,25 +4075,14 @@ void md_create_long_jump (ptr, from_addr, to_addr, frag, to_symbol) char *ptr; addressT from_addr, to_addr; - fragS *frag; - symbolS *to_symbol; + fragS *frag ATTRIBUTE_UNUSED; + symbolS *to_symbol ATTRIBUTE_UNUSED; { - long offset; + offsetT offset; - if (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); - } + offset = to_addr - (from_addr + 5); + md_number_to_chars (ptr, (valueT) 0xe9, 1); + md_number_to_chars (ptr + 1, (valueT) offset, 4); } /* Apply a fixup (fixS) to segment data, once it has been determined @@ -3875,9 +4094,14 @@ md_create_long_jump (ptr, from_addr, to_addr, frag, to_symbol) 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. */ + /* The fix we're to put in. */ + fixS *fixP; + + /* Pointer to the value of the bits. */ + valueT *valp; + + /* Segment fix is from. */ + segT seg ATTRIBUTE_UNUSED; { register char *p = fixP->fx_where + fixP->fx_frag->fr_literal; valueT value = *valp; @@ -3909,7 +4133,7 @@ md_apply_fix3 (fixP, valp, seg) 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) + && fixP->fx_addsy && !use_rela_relocations) { #ifndef OBJ_AOUT if (OUTPUT_FLAVOR == bfd_target_elf_flavour @@ -3945,49 +4169,25 @@ md_apply_fix3 (fixP, valp, seg) 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 } -#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))) - { - /* 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); - } -#endif /* Fix a few things - the dynamic linker expects certain values here, - and we must not dissappoint it. */ + 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 + 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 @@ -4022,55 +4222,55 @@ md_apply_fix3 (fixP, valp, seg) * explicitly mentioned, and I wonder whether it would simplify matters * to do it this way. Who knows. In earlier versions of the PIC patches, * the pcrel_adjust field was used to store the correction, but since the - * expression is not pcrel, I felt it would be confusing to do it this way. - */ - value -= 1; - break; - case BFD_RELOC_386_GOT32: - value = 0; /* Fully resolved at runtime. No addend. */ - break; - case BFD_RELOC_386_GOTOFF: - break; - - case BFD_RELOC_VTABLE_INHERIT: - case BFD_RELOC_VTABLE_ENTRY: - fixP->fx_done = 0; - return 1; + * 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_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 1; + + default: + break; + } +#endif /* defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) */ + *valp = value; +#endif /* defined (BFD_ASSEMBLER) && !defined (TE_Mach) */ - default: - break; +#ifndef BFD_ASSEMBLER + md_number_to_chars (p, value, fixP->fx_size); +#else + /* Are we finished with this relocation now? */ + if (fixP->fx_addsy == 0 && fixP->fx_pcrel == 0) + fixP->fx_done = 1; + else if (use_rela_relocations) + { + fixP->fx_no_overflow = 1; + value = 0; } -#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); +#endif 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. */ -{ - long retval; - for (retval = 0, con += nbytes - 1; nbytes--; con--) - { - retval <<= BITS_PER_CHAR; - retval |= *con; - } - return retval; -} -#endif /* 0 */ - #define MAX_LITTLENUMS 6 -/* Turn the string pointed to by litP into a floating point constant of type - type, and emit the appropriate bytes. The number of LITTLENUMS emitted - is stored in *sizeP . An error message is returned, or NULL on OK. */ +/* Turn the string pointed to by litP into a floating point constant + of type TYPE, and emit the appropriate bytes. The number of + LITTLENUMS emitted is stored in *SIZEP. An error message is + returned, or NULL on OK. */ + char * md_atof (type, litP, sizeP) int type; @@ -4120,8 +4320,6 @@ md_atof (type, litP, sizeP) char output_invalid_buf[8]; -static char * output_invalid PARAMS ((int)); - static char * output_invalid (c) int c; @@ -4133,7 +4331,6 @@ output_invalid (c) return output_invalid_buf; } - /* REG_STRING starts *before* REGISTER_PREFIX. */ static const reg_entry * @@ -4161,12 +4358,18 @@ parse_register (reg_string, end_op) 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 (r == i386_regtab /* %st is first entry of table */) { if (is_space_char (*s)) ++s; @@ -4187,7 +4390,7 @@ parse_register (reg_string, end_op) return r; } } - /* We have "%st(" then garbage */ + /* We have "%st(" then garbage. */ return (const reg_entry *) NULL; } } @@ -4196,11 +4399,18 @@ parse_register (reg_string, end_op) } #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) -CONST char *md_shortopts = "kmVQ:sq"; +const char *md_shortopts = "kVQ:sq"; #else -CONST char *md_shortopts = "m"; +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); @@ -4212,13 +4422,14 @@ md_parse_option (c, arg) { switch (c) { - case 'm': - flag_do_long_jump = 1; + case 'q': + quiet_warnings = 1; break; #if defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF) - /* -k: Ignore for FreeBSD compatibility. */ - case 'k': + /* -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. */ @@ -4226,9 +4437,8 @@ md_parse_option (c, arg) print_version_id (); break; - /* -Qy, -Qn: SVR4 arguments controlling whether a .comment section - should be emitted or not. FIXME: Not implemented. */ - case 'Q': + /* -k: Ignore for FreeBSD compatibility. */ + case 'k': break; case 's': @@ -4236,12 +4446,28 @@ md_parse_option (c, arg) .stab instead of .stab.excl. We always use .stab anyhow. */ break; - case 'q': - /* -q: On i386 Solaris, this tells the native assembler does - fewer checks. */ + 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; } @@ -4252,41 +4478,51 @@ 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")); + -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_ELF) && defined (OBJ_MAYBE_COFF)) \ - || (defined (OBJ_MAYBE_ELF) && defined (OBJ_MAYBE_AOUT)) \ - || (defined (OBJ_MAYBE_COFF) && defined (OBJ_MAYBE_AOUT))) +#if ((defined (OBJ_MAYBE_COFF) && defined (OBJ_MAYBE_AOUT)) \ + || defined (OBJ_ELF) || defined (OBJ_MAYBE_ELF)) /* Pick the target format to use. */ -const char * +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; + return AOUT_TARGET_FORMAT; #endif #ifdef OBJ_MAYBE_COFF case bfd_target_coff_flavour: return "coff-i386"; #endif -#ifdef OBJ_MAYBE_ELF +#if defined (OBJ_MAYBE_ELF) || defined (OBJ_ELF) case bfd_target_elf_flavour: - return "elf32-i386"; + { + if (flag_code == CODE_64BIT) + use_rela_relocations = 1; + return flag_code == CODE_64BIT ? "elf64-x86-64" : "elf32-i386"; + } #endif default: abort (); @@ -4294,8 +4530,8 @@ i386_target_format () } } -#endif /* OBJ_MAYBE_ more than one */ -#endif /* BFD_ASSEMBLER */ +#endif /* OBJ_MAYBE_ more than one */ +#endif /* BFD_ASSEMBLER */ symbolS * md_undefined_symbol (name) @@ -4319,6 +4555,7 @@ md_undefined_symbol (name) } /* Round up a section size to the appropriate boundary. */ + valueT md_section_align (segment, size) segT segment ATTRIBUTE_UNUSED; @@ -4370,7 +4607,6 @@ s_bss (ignore) #endif - #ifdef BFD_ASSEMBLER void @@ -4379,7 +4615,19 @@ i386_validate_fix (fixp) { if (fixp->fx_subsy && fixp->fx_subsy == GOT_symbol) { - fixp->fx_r_type = BFD_RELOC_386_GOTOFF; + /* 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; } } @@ -4394,10 +4642,14 @@ tc_gen_reloc (section, fixp) switch (fixp->fx_r_type) { + case BFD_RELOC_X86_64_PLT32: + case BFD_RELOC_X86_64_GOT32: + case BFD_RELOC_X86_64_GOTPCREL: case BFD_RELOC_386_PLT32: case BFD_RELOC_386_GOT32: case BFD_RELOC_386_GOTOFF: case BFD_RELOC_386_GOTPC: + case BFD_RELOC_X86_64_32S: case BFD_RELOC_RVA: case BFD_RELOC_VTABLE_ENTRY: case BFD_RELOC_VTABLE_INHERIT: @@ -4429,6 +4681,7 @@ tc_gen_reloc (section, fixp) case 1: code = BFD_RELOC_8; break; case 2: code = BFD_RELOC_16; break; case 4: code = BFD_RELOC_32; break; + case 8: code = BFD_RELOC_64; break; } } break; @@ -4437,22 +4690,37 @@ tc_gen_reloc (section, fixp) if (code == BFD_RELOC_32 && GOT_symbol && fixp->fx_addsy == GOT_symbol) - code = BFD_RELOC_386_GOTPC; + { + /* We don't support GOTPC on 64bit targets. */ + if (flag_code == CODE_64BIT) + abort (); + code = BFD_RELOC_386_GOTPC; + } rel = (arelent *) xmalloc (sizeof (arelent)); rel->sym_ptr_ptr = (asymbol **) xmalloc (sizeof (asymbol *)); *rel->sym_ptr_ptr = symbol_get_bfdsym (fixp->fx_addsy); rel->address = fixp->fx_frag->fr_address + fixp->fx_where; - /* HACK: Since i386 ELF uses Rel instead of Rela, encode the - vtable entry to be used in the relocation's section offset. */ - if (fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY) - rel->address = fixp->fx_offset; + if (!use_rela_relocations) + { + /* HACK: Since i386 ELF uses Rel instead of Rela, encode the + vtable entry to be used in the relocation's section offset. */ + if (fixp->fx_r_type == BFD_RELOC_VTABLE_ENTRY) + rel->address = fixp->fx_offset; - if (fixp->fx_pcrel) - rel->addend = fixp->fx_addnumber; + if (fixp->fx_pcrel) + rel->addend = fixp->fx_addnumber; + else + rel->addend = 0; + } + /* Use the rela in 64bit mode. */ else - rel->addend = 0; + { + rel->addend = fixp->fx_offset; + if (fixp->fx_pcrel) + rel->addend -= fixp->fx_size; + } rel->howto = bfd_reloc_type_lookup (stdoutput, code); if (rel->howto == NULL) @@ -4468,7 +4736,7 @@ tc_gen_reloc (section, fixp) return rel; } -#else /* ! BFD_ASSEMBLER */ +#else /* ! BFD_ASSEMBLER */ #if (defined(OBJ_AOUT) | defined(OBJ_BOUT)) void @@ -4477,12 +4745,10 @@ tc_aout_fix_to_chars (where, fixP, segment_address_in_file) fixS *fixP; relax_addressT segment_address_in_file; { - /* - * In: length of relocation (or of address) in chars: 1, 2 or 4. - * Out: GNU LD relocation length code: 0, 1, or 2. - */ + /* In: length of relocation (or of address) in chars: 1, 2 or 4. + Out: GNU LD relocation length code: 0, 1, or 2. */ - static const unsigned char nbytes_r_length[] = {42, 0, 1, 42, 2}; + static const unsigned char nbytes_r_length[] = { 42, 0, 1, 42, 2 }; long r_symbolnum; know (fixP->fx_addsy != NULL); @@ -4504,7 +4770,7 @@ tc_aout_fix_to_chars (where, fixP, segment_address_in_file) | (((fixP->fx_pcrel << 0) & 0x01) & 0x0f)); } -#endif /* OBJ_AOUT or OBJ_BOUT */ +#endif /* OBJ_AOUT or OBJ_BOUT. */ #if defined (I386COFF) @@ -4534,8 +4800,956 @@ tc_coff_sizemachdep (frag) return 0; } -#endif /* I386COFF */ +#endif /* I386COFF */ -#endif /* ! BFD_ASSEMBLER */ +#endif /* ! BFD_ASSEMBLER */ -/* end of tc-i386.c */ +/* Parse operands using Intel syntax. This implements a recursive descent + parser based on the BNF grammar published in Appendix B of the MASM 6.1 + Programmer's Guide. + + FIXME: We do not recognize the full operand grammar defined in the MASM + documentation. In particular, all the structure/union and + high-level macro operands are missing. + + Uppercase words are terminals, lower case words are non-terminals. + Objects surrounded by double brackets '[[' ']]' are optional. Vertical + bars '|' denote choices. Most grammar productions are implemented in + functions called 'intel_<production>'. + + Initial production is 'expr'. + + addOp + | - + + alpha [a-zA-Z] + + byteRegister AL | AH | BL | BH | CL | CH | DL | DH + + constant digits [[ radixOverride ]] + + dataType BYTE | WORD | DWORD | QWORD | XWORD + + digits decdigit + | digits decdigit + | digits hexdigit + + decdigit [0-9] + + e05 e05 addOp e06 + | e06 + + e06 e06 mulOp e09 + | e09 + + e09 OFFSET e10 + | e09 PTR e10 + | e09 : e10 + | e10 + + e10 e10 [ expr ] + | e11 + + e11 ( expr ) + | [ expr ] + | constant + | dataType + | id + | $ + | register + + => expr SHORT e05 + | e05 + + gpRegister AX | EAX | BX | EBX | CX | ECX | DX | EDX + | BP | EBP | SP | ESP | DI | EDI | SI | ESI + + hexdigit a | b | c | d | e | f + | A | B | C | D | E | F + + id alpha + | id alpha + | id decdigit + + mulOp * | / | MOD + + quote " | ' + + register specialRegister + | gpRegister + | byteRegister + + segmentRegister CS | DS | ES | FS | GS | SS + + specialRegister CR0 | CR2 | CR3 + | DR0 | DR1 | DR2 | DR3 | DR6 | DR7 + | TR3 | TR4 | TR5 | TR6 | TR7 + + We simplify the grammar in obvious places (e.g., register parsing is + done by calling parse_register) and eliminate immediate left recursion + to implement a recursive-descent parser. + + expr SHORT e05 + | e05 + + e05 e06 e05' + + e05' addOp e06 e05' + | Empty + + e06 e09 e06' + + e06' mulOp e09 e06' + | Empty + + e09 OFFSET e10 e09' + | e10 e09' + + e09' PTR e10 e09' + | : e10 e09' + | Empty + + e10 e11 e10' + + e10' [ expr ] e10' + | Empty + + e11 ( expr ) + | [ expr ] + | BYTE + | WORD + | DWORD + | QWORD + | XWORD + | . + | $ + | register + | id + | constant */ + +/* Parsing structure for the intel syntax parser. Used to implement the + semantic actions for the operand grammar. */ +struct intel_parser_s + { + char *op_string; /* The string being parsed. */ + int got_a_float; /* Whether the operand is a float. */ + int op_modifier; /* Operand modifier. */ + int is_mem; /* 1 if operand is memory reference. */ + const reg_entry *reg; /* Last register reference found. */ + char *disp; /* Displacement string being built. */ + }; + +static struct intel_parser_s intel_parser; + +/* Token structure for parsing intel syntax. */ +struct intel_token + { + int code; /* Token code. */ + const reg_entry *reg; /* Register entry for register tokens. */ + char *str; /* String representation. */ + }; + +static struct intel_token cur_token, prev_token; + +/* Token codes for the intel parser. Since T_SHORT is already used + by COFF, undefine it first to prevent a warning. */ +#define T_NIL -1 +#define T_CONST 1 +#define T_REG 2 +#define T_BYTE 3 +#define T_WORD 4 +#define T_DWORD 5 +#define T_QWORD 6 +#define T_XWORD 7 +#undef T_SHORT +#define T_SHORT 8 +#define T_OFFSET 9 +#define T_PTR 10 +#define T_ID 11 + +/* Prototypes for intel parser functions. */ +static int intel_match_token PARAMS ((int code)); +static void intel_get_token PARAMS ((void)); +static void intel_putback_token PARAMS ((void)); +static int intel_expr PARAMS ((void)); +static int intel_e05 PARAMS ((void)); +static int intel_e05_1 PARAMS ((void)); +static int intel_e06 PARAMS ((void)); +static int intel_e06_1 PARAMS ((void)); +static int intel_e09 PARAMS ((void)); +static int intel_e09_1 PARAMS ((void)); +static int intel_e10 PARAMS ((void)); +static int intel_e10_1 PARAMS ((void)); +static int intel_e11 PARAMS ((void)); + +static int +i386_intel_operand (operand_string, got_a_float) + char *operand_string; + int got_a_float; +{ + int ret; + char *p; + + /* Initialize token holders. */ + cur_token.code = prev_token.code = T_NIL; + cur_token.reg = prev_token.reg = NULL; + cur_token.str = prev_token.str = NULL; + + /* Initialize parser structure. */ + p = intel_parser.op_string = (char *) malloc (strlen (operand_string) + 1); + if (p == NULL) + abort (); + strcpy (intel_parser.op_string, operand_string); + intel_parser.got_a_float = got_a_float; + intel_parser.op_modifier = -1; + intel_parser.is_mem = 0; + intel_parser.reg = NULL; + intel_parser.disp = (char *) malloc (strlen (operand_string) + 1); + if (intel_parser.disp == NULL) + abort (); + intel_parser.disp[0] = '\0'; + + /* Read the first token and start the parser. */ + intel_get_token (); + ret = intel_expr (); + + if (ret) + { + /* If we found a memory reference, hand it over to i386_displacement + to fill in the rest of the operand fields. */ + if (intel_parser.is_mem) + { + if ((i.mem_operands == 1 + && (current_templates->start->opcode_modifier & IsString) == 0) + || i.mem_operands == 2) + { + as_bad (_("too many memory references for '%s'"), + current_templates->start->name); + ret = 0; + } + else + { + char *s = intel_parser.disp; + i.mem_operands++; + + /* Add the displacement expression. */ + if (*s != '\0') + ret = i386_displacement (s, s + strlen (s)) + && i386_index_check (s); + } + } + + /* Constant and OFFSET expressions are handled by i386_immediate. */ + else if (intel_parser.op_modifier == OFFSET_FLAT + || intel_parser.reg == NULL) + ret = i386_immediate (intel_parser.disp); + } + + free (p); + free (intel_parser.disp); + + return ret; +} + +/* expr SHORT e05 + | e05 */ +static int +intel_expr () +{ + /* expr SHORT e05 */ + if (cur_token.code == T_SHORT) + { + intel_parser.op_modifier = SHORT; + intel_match_token (T_SHORT); + + return (intel_e05 ()); + } + + /* expr e05 */ + else + return intel_e05 (); +} + +/* e05 e06 e05' + + e05' addOp e06 e05' + | Empty */ +static int +intel_e05 () +{ + return (intel_e06 () && intel_e05_1 ()); +} + +static int +intel_e05_1 () +{ + /* e05' addOp e06 e05' */ + if (cur_token.code == '+' || cur_token.code == '-') + { + strcat (intel_parser.disp, cur_token.str); + intel_match_token (cur_token.code); + + return (intel_e06 () && intel_e05_1 ()); + } + + /* e05' Empty */ + else + return 1; +} + +/* e06 e09 e06' + + e06' mulOp e09 e06' + | Empty */ +static int +intel_e06 () +{ + return (intel_e09 () && intel_e06_1 ()); +} + +static int +intel_e06_1 () +{ + /* e06' mulOp e09 e06' */ + if (cur_token.code == '*' || cur_token.code == '/') + { + strcat (intel_parser.disp, cur_token.str); + intel_match_token (cur_token.code); + + return (intel_e09 () && intel_e06_1 ()); + } + + /* e06' Empty */ + else + return 1; +} + +/* e09 OFFSET e10 e09' + | e10 e09' + + e09' PTR e10 e09' + | : e10 e09' + | Empty */ +static int +intel_e09 () +{ + /* e09 OFFSET e10 e09' */ + if (cur_token.code == T_OFFSET) + { + intel_parser.is_mem = 0; + intel_parser.op_modifier = OFFSET_FLAT; + intel_match_token (T_OFFSET); + + return (intel_e10 () && intel_e09_1 ()); + } + + /* e09 e10 e09' */ + else + return (intel_e10 () && intel_e09_1 ()); +} + +static int +intel_e09_1 () +{ + /* e09' PTR e10 e09' */ + if (cur_token.code == T_PTR) + { + if (prev_token.code == T_BYTE) + i.suffix = BYTE_MNEM_SUFFIX; + + else if (prev_token.code == T_WORD) + { + if (intel_parser.got_a_float == 2) /* "fi..." */ + i.suffix = SHORT_MNEM_SUFFIX; + else + i.suffix = WORD_MNEM_SUFFIX; + } + + else if (prev_token.code == T_DWORD) + { + if (intel_parser.got_a_float == 1) /* "f..." */ + i.suffix = SHORT_MNEM_SUFFIX; + else + i.suffix = LONG_MNEM_SUFFIX; + } + + else if (prev_token.code == T_QWORD) + { + if (intel_parser.got_a_float == 1) /* "f..." */ + i.suffix = LONG_MNEM_SUFFIX; + else + i.suffix = QWORD_MNEM_SUFFIX; + } + + else if (prev_token.code == T_XWORD) + i.suffix = LONG_DOUBLE_MNEM_SUFFIX; + + else + { + as_bad (_("Unknown operand modifier `%s'\n"), prev_token.str); + return 0; + } + + intel_match_token (T_PTR); + + return (intel_e10 () && intel_e09_1 ()); + } + + /* e09 : e10 e09' */ + else if (cur_token.code == ':') + { + /* Mark as a memory operand only if it's not already known to be an + offset expression. */ + if (intel_parser.op_modifier != OFFSET_FLAT) + intel_parser.is_mem = 1; + + return (intel_match_token (':') && intel_e10 () && intel_e09_1 ()); + } + + /* e09' Empty */ + else + return 1; +} + +/* e10 e11 e10' + + e10' [ expr ] e10' + | Empty */ +static int +intel_e10 () +{ + return (intel_e11 () && intel_e10_1 ()); +} + +static int +intel_e10_1 () +{ + /* e10' [ expr ] e10' */ + if (cur_token.code == '[') + { + intel_match_token ('['); + + /* Mark as a memory operand only if it's not already known to be an + offset expression. If it's an offset expression, we need to keep + the brace in. */ + if (intel_parser.op_modifier != OFFSET_FLAT) + intel_parser.is_mem = 1; + else + strcat (intel_parser.disp, "["); + + /* Add a '+' to the displacement string if necessary. */ + if (*intel_parser.disp != '\0' + && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+') + strcat (intel_parser.disp, "+"); + + if (intel_expr () && intel_match_token (']')) + { + /* Preserve brackets when the operand is an offset expression. */ + if (intel_parser.op_modifier == OFFSET_FLAT) + strcat (intel_parser.disp, "]"); + + return intel_e10_1 (); + } + else + return 0; + } + + /* e10' Empty */ + else + return 1; +} + +/* e11 ( expr ) + | [ expr ] + | BYTE + | WORD + | DWORD + | QWORD + | XWORD + | $ + | . + | register + | id + | constant */ +static int +intel_e11 () +{ + /* e11 ( expr ) */ + if (cur_token.code == '(') + { + intel_match_token ('('); + strcat (intel_parser.disp, "("); + + if (intel_expr () && intel_match_token (')')) + { + strcat (intel_parser.disp, ")"); + return 1; + } + else + return 0; + } + + /* e11 [ expr ] */ + else if (cur_token.code == '[') + { + intel_match_token ('['); + + /* Mark as a memory operand only if it's not already known to be an + offset expression. If it's an offset expression, we need to keep + the brace in. */ + if (intel_parser.op_modifier != OFFSET_FLAT) + intel_parser.is_mem = 1; + else + strcat (intel_parser.disp, "["); + + /* Operands for jump/call inside brackets denote absolute addresses. */ + if (current_templates->start->opcode_modifier & Jump + || current_templates->start->opcode_modifier & JumpDword + || current_templates->start->opcode_modifier & JumpByte + || current_templates->start->opcode_modifier & JumpInterSegment) + i.types[this_operand] |= JumpAbsolute; + + /* Add a '+' to the displacement string if necessary. */ + if (*intel_parser.disp != '\0' + && *(intel_parser.disp + strlen (intel_parser.disp) - 1) != '+') + strcat (intel_parser.disp, "+"); + + if (intel_expr () && intel_match_token (']')) + { + /* Preserve brackets when the operand is an offset expression. */ + if (intel_parser.op_modifier == OFFSET_FLAT) + strcat (intel_parser.disp, "]"); + + return 1; + } + else + return 0; + } + + /* e11 BYTE + | WORD + | DWORD + | QWORD + | XWORD */ + else if (cur_token.code == T_BYTE + || cur_token.code == T_WORD + || cur_token.code == T_DWORD + || cur_token.code == T_QWORD + || cur_token.code == T_XWORD) + { + intel_match_token (cur_token.code); + + return 1; + } + + /* e11 $ + | . */ + else if (cur_token.code == '$' || cur_token.code == '.') + { + strcat (intel_parser.disp, cur_token.str); + intel_match_token (cur_token.code); + + /* Mark as a memory operand only if it's not already known to be an + offset expression. */ + if (intel_parser.op_modifier != OFFSET_FLAT) + intel_parser.is_mem = 1; + + return 1; + } + + /* e11 register */ + else if (cur_token.code == T_REG) + { + const reg_entry *reg = intel_parser.reg = cur_token.reg; + + intel_match_token (T_REG); + + /* Check for segment change. */ + if (cur_token.code == ':') + { + if (reg->reg_type & (SReg2 | SReg3)) + { + switch (reg->reg_num) + { + case 0: + i.seg[i.mem_operands] = &es; + break; + case 1: + i.seg[i.mem_operands] = &cs; + break; + case 2: + i.seg[i.mem_operands] = &ss; + break; + case 3: + i.seg[i.mem_operands] = &ds; + break; + case 4: + i.seg[i.mem_operands] = &fs; + break; + case 5: + i.seg[i.mem_operands] = &gs; + break; + } + } + else + { + as_bad (_("`%s' is not a valid segment register"), reg->reg_name); + return 0; + } + } + + /* Not a segment register. Check for register scaling. */ + else if (cur_token.code == '*') + { + if (!intel_parser.is_mem) + { + as_bad (_("Register scaling only allowed in memory operands.")); + return 0; + } + + /* What follows must be a valid scale. */ + if (intel_match_token ('*') + && strchr ("01248", *cur_token.str)) + { + i.index_reg = reg; + i.types[this_operand] |= BaseIndex; + + /* Set the scale after setting the register (otherwise, + i386_scale will complain) */ + i386_scale (cur_token.str); + intel_match_token (T_CONST); + } + else + { + as_bad (_("expecting scale factor of 1, 2, 4, or 8: got `%s'"), + cur_token.str); + return 0; + } + } + + /* No scaling. If this is a memory operand, the register is either a + base register (first occurrence) or an index register (second + occurrence). */ + else if (intel_parser.is_mem && !(reg->reg_type & (SReg2 | SReg3))) + { + if (i.base_reg && i.index_reg) + { + as_bad (_("Too many register references in memory operand.\n")); + return 0; + } + + if (i.base_reg == NULL) + i.base_reg = reg; + else + i.index_reg = reg; + + i.types[this_operand] |= BaseIndex; + } + + /* Offset modifier. Add the register to the displacement string to be + parsed as an immediate expression after we're done. */ + else if (intel_parser.op_modifier == OFFSET_FLAT) + strcat (intel_parser.disp, reg->reg_name); + + /* It's neither base nor index nor offset. */ + else + { + i.types[this_operand] |= reg->reg_type & ~BaseIndex; + i.op[this_operand].regs = reg; + i.reg_operands++; + } + + /* Since registers are not part of the displacement string (except + when we're parsing offset operands), we may need to remove any + preceding '+' from the displacement string. */ + if (*intel_parser.disp != '\0' + && intel_parser.op_modifier != OFFSET_FLAT) + { + char *s = intel_parser.disp; + s += strlen (s) - 1; + if (*s == '+') + *s = '\0'; + } + + return 1; + } + + /* e11 id */ + else if (cur_token.code == T_ID) + { + /* Add the identifier to the displacement string. */ + strcat (intel_parser.disp, cur_token.str); + intel_match_token (T_ID); + + /* The identifier represents a memory reference only if it's not + preceded by an offset modifier. */ + if (intel_parser.op_modifier != OFFSET_FLAT) + intel_parser.is_mem = 1; + + return 1; + } + + /* e11 constant */ + else if (cur_token.code == T_CONST + || cur_token.code == '-' + || cur_token.code == '+') + { + char *save_str; + + /* Allow constants that start with `+' or `-'. */ + if (cur_token.code == '-' || cur_token.code == '+') + { + strcat (intel_parser.disp, cur_token.str); + intel_match_token (cur_token.code); + if (cur_token.code != T_CONST) + { + as_bad (_("Syntax error. Expecting a constant. Got `%s'.\n"), + cur_token.str); + return 0; + } + } + + save_str = (char *) malloc (strlen (cur_token.str) + 1); + if (save_str == NULL) + abort (); + strcpy (save_str, cur_token.str); + + /* Get the next token to check for register scaling. */ + intel_match_token (cur_token.code); + + /* Check if this constant is a scaling factor for an index register. */ + if (cur_token.code == '*') + { + if (intel_match_token ('*') && cur_token.code == T_REG) + { + if (!intel_parser.is_mem) + { + as_bad (_("Register scaling only allowed in memory operands.")); + return 0; + } + + /* The constant is followed by `* reg', so it must be + a valid scale. */ + if (strchr ("01248", *save_str)) + { + i.index_reg = cur_token.reg; + i.types[this_operand] |= BaseIndex; + + /* Set the scale after setting the register (otherwise, + i386_scale will complain) */ + i386_scale (save_str); + intel_match_token (T_REG); + + /* Since registers are not part of the displacement + string, we may need to remove any preceding '+' from + the displacement string. */ + if (*intel_parser.disp != '\0') + { + char *s = intel_parser.disp; + s += strlen (s) - 1; + if (*s == '+') + *s = '\0'; + } + + free (save_str); + + return 1; + } + else + return 0; + } + + /* The constant was not used for register scaling. Since we have + already consumed the token following `*' we now need to put it + back in the stream. */ + else + intel_putback_token (); + } + + /* Add the constant to the displacement string. */ + strcat (intel_parser.disp, save_str); + free (save_str); + + return 1; + } + + as_bad (_("Unrecognized token '%s'"), cur_token.str); + return 0; +} + +/* Match the given token against cur_token. If they match, read the next + token from the operand string. */ +static int +intel_match_token (code) + int code; +{ + if (cur_token.code == code) + { + intel_get_token (); + return 1; + } + else + { + as_bad (_("Unexpected token `%s'\n"), cur_token.str); + return 0; + } +} + +/* Read a new token from intel_parser.op_string and store it in cur_token. */ +static void +intel_get_token () +{ + char *end_op; + const reg_entry *reg; + struct intel_token new_token; + + new_token.code = T_NIL; + new_token.reg = NULL; + new_token.str = NULL; + + /* Free the memory allocated to the previous token and move + cur_token to prev_token. */ + if (prev_token.str) + free (prev_token.str); + + prev_token = cur_token; + + /* Skip whitespace. */ + while (is_space_char (*intel_parser.op_string)) + intel_parser.op_string++; + + /* Return an empty token if we find nothing else on the line. */ + if (*intel_parser.op_string == '\0') + { + cur_token = new_token; + return; + } + + /* The new token cannot be larger than the remainder of the operand + string. */ + new_token.str = (char *) malloc (strlen (intel_parser.op_string) + 1); + if (new_token.str == NULL) + abort (); + new_token.str[0] = '\0'; + + if (strchr ("0123456789", *intel_parser.op_string)) + { + char *p = new_token.str; + char *q = intel_parser.op_string; + new_token.code = T_CONST; + + /* Allow any kind of identifier char to encompass floating point and + hexadecimal numbers. */ + while (is_identifier_char (*q)) + *p++ = *q++; + *p = '\0'; + + /* Recognize special symbol names [0-9][bf]. */ + if (strlen (intel_parser.op_string) == 2 + && (intel_parser.op_string[1] == 'b' + || intel_parser.op_string[1] == 'f')) + new_token.code = T_ID; + } + + else if (strchr ("+-/*:[]()", *intel_parser.op_string)) + { + new_token.code = *intel_parser.op_string; + new_token.str[0] = *intel_parser.op_string; + new_token.str[1] = '\0'; + } + + else if ((*intel_parser.op_string == REGISTER_PREFIX || allow_naked_reg) + && ((reg = parse_register (intel_parser.op_string, &end_op)) != NULL)) + { + new_token.code = T_REG; + new_token.reg = reg; + + if (*intel_parser.op_string == REGISTER_PREFIX) + { + new_token.str[0] = REGISTER_PREFIX; + new_token.str[1] = '\0'; + } + + strcat (new_token.str, reg->reg_name); + } + + else if (is_identifier_char (*intel_parser.op_string)) + { + char *p = new_token.str; + char *q = intel_parser.op_string; + + /* A '.' or '$' followed by an identifier char is an identifier. + Otherwise, it's operator '.' followed by an expression. */ + if ((*q == '.' || *q == '$') && !is_identifier_char (*(q + 1))) + { + new_token.code = *q; + new_token.str[0] = *q; + new_token.str[1] = '\0'; + } + else + { + while (is_identifier_char (*q) || *q == '@') + *p++ = *q++; + *p = '\0'; + + if (strcasecmp (new_token.str, "BYTE") == 0) + new_token.code = T_BYTE; + + else if (strcasecmp (new_token.str, "WORD") == 0) + new_token.code = T_WORD; + + else if (strcasecmp (new_token.str, "DWORD") == 0) + new_token.code = T_DWORD; + + else if (strcasecmp (new_token.str, "QWORD") == 0) + new_token.code = T_QWORD; + + else if (strcasecmp (new_token.str, "XWORD") == 0) + new_token.code = T_XWORD; + + else if (strcasecmp (new_token.str, "PTR") == 0) + new_token.code = T_PTR; + + else if (strcasecmp (new_token.str, "SHORT") == 0) + new_token.code = T_SHORT; + + else if (strcasecmp (new_token.str, "OFFSET") == 0) + { + new_token.code = T_OFFSET; + + /* ??? This is not mentioned in the MASM grammar but gcc + makes use of it with -mintel-syntax. OFFSET may be + followed by FLAT: */ + if (strncasecmp (q, " FLAT:", 6) == 0) + strcat (new_token.str, " FLAT:"); + } + + /* ??? This is not mentioned in the MASM grammar. */ + else if (strcasecmp (new_token.str, "FLAT") == 0) + new_token.code = T_OFFSET; + + else + new_token.code = T_ID; + } + } + + else + as_bad (_("Unrecognized token `%s'\n"), intel_parser.op_string); + + intel_parser.op_string += strlen (new_token.str); + cur_token = new_token; +} + +/* Put cur_token back into the token stream and make cur_token point to + prev_token. */ +static void +intel_putback_token () +{ + intel_parser.op_string -= strlen (cur_token.str); + free (cur_token.str); + cur_token = prev_token; + + /* Forget prev_token. */ + prev_token.code = T_NIL; + prev_token.reg = NULL; + prev_token.str = NULL; +} |