nasm.h

nasm早期的源代码,比较简单是学习汇编和编译原理的好例子

/* nasm.h   main header file for the Netwide Assembler: inter-module interface
 *
 * The Netwide Assembler is copyright (C) 1996 Simon Tatham and
 * Julian Hall. All rights reserved. The software is
 * redistributable under the licence given in the file "Licence"
 * distributed in the NASM archive.
 *
 * initial version: 27/iii/95 by Simon Tatham
 */

#ifndef NASM_NASM_H
#define NASM_NASM_H

#include "compiler.h"

#include <stdio.h>
#include <inttypes.h>
#include "version.h"            /* generated NASM version macros */
#include "nasmlib.h"
#include "insnsi.h"		/* For enum opcode */

#define NO_SEG -1L              /* null segment value */
#define SEG_ABS 0x40000000L     /* mask for far-absolute segments */

#ifndef FILENAME_MAX
#define FILENAME_MAX 256
#endif

#ifndef PREFIX_MAX
#define PREFIX_MAX 10
#endif

#ifndef POSTFIX_MAX
#define POSTFIX_MAX 10
#endif

#define IDLEN_MAX 4096

/*
 * Name pollution problems: <time.h> on Digital UNIX pulls in some
 * strange hardware header file which sees fit to define R_SP. We
 * undefine it here so as not to break the enum below.
 */
#ifdef R_SP
#undef R_SP
#endif

/*
 * We must declare the existence of this structure type up here,
 * since we have to reference it before we define it...
 */
struct ofmt;

/*
 * -----------------------
 * Other function typedefs
 * -----------------------
 */

/*
 * A label-lookup function should look like this.
 */
typedef bool (*lfunc) (char *label, int32_t *segment, int32_t *offset);

/*
 * And a label-definition function like this. The boolean parameter
 * `is_norm' states whether the label is a `normal' label (which
 * should affect the local-label system), or something odder like
 * an EQU or a segment-base symbol, which shouldn't.
 */
typedef void (*ldfunc) (char *label, int32_t segment, int32_t offset,
                        char *special, bool is_norm, bool isextrn,
                        struct ofmt * ofmt, efunc error);

/*
 * List-file generators should look like this:
 */
typedef struct {
    /*
     * Called to initialize the listing file generator. Before this
     * is called, the other routines will silently do nothing when
     * called. The `char *' parameter is the file name to write the
     * listing to.
     */
    void (*init) (char *, efunc);

    /*
     * Called to clear stuff up and close the listing file.
     */
    void (*cleanup) (void);

    /*
     * Called to output binary data. Parameters are: the offset;
     * the data; the data type. Data types are similar to the
     * output-format interface, only OUT_ADDRESS will _always_ be
     * displayed as if it's relocatable, so ensure that any non-
     * relocatable address has been converted to OUT_RAWDATA by
     * then. Note that OUT_RAWDATA+0 is a valid data type, and is a
     * dummy call used to give the listing generator an offset to
     * work with when doing things like uplevel(LIST_TIMES) or
     * uplevel(LIST_INCBIN).
     */
    void (*output) (int32_t, const void *, uint32_t);

    /*
     * Called to send a text line to the listing generator. The
     * `int' parameter is LIST_READ or LIST_MACRO depending on
     * whether the line came directly from an input file or is the
     * result of a multi-line macro expansion.
     */
    void (*line) (int, char *);

    /*
     * Called to change one of the various levelled mechanisms in
     * the listing generator. LIST_INCLUDE and LIST_MACRO can be
     * used to increase the nesting level of include files and
     * macro expansions; LIST_TIMES and LIST_INCBIN switch on the
     * two binary-output-suppression mechanisms for large-scale
     * pseudo-instructions.
     *
     * LIST_MACRO_NOLIST is synonymous with LIST_MACRO except that
     * it indicates the beginning of the expansion of a `nolist'
     * macro, so anything under that level won't be expanded unless
     * it includes another file.
     */
    void (*uplevel) (int);

    /*
     * Reverse the effects of uplevel.
     */
    void (*downlevel) (int);
} ListGen;

/*
 * The expression evaluator must be passed a scanner function; a
 * standard scanner is provided as part of nasmlib.c. The
 * preprocessor will use a different one. Scanners, and the
 * token-value structures they return, look like this.
 *
 * The return value from the scanner is always a copy of the
 * `t_type' field in the structure.
 */
struct tokenval {
    int t_type;
    int64_t t_integer, t_inttwo;
    char *t_charptr;
};
typedef int (*scanner) (void *private_data, struct tokenval * tv);

/*
 * Token types returned by the scanner, in addition to ordinary
 * ASCII character values, and zero for end-of-string.
 */
enum {                          /* token types, other than chars */
    TOKEN_INVALID = -1,         /* a placeholder value */
    TOKEN_EOS = 0,              /* end of string */
    TOKEN_EQ = '=', TOKEN_GT = '>', TOKEN_LT = '<',     /* aliases */
    TOKEN_ID = 256, TOKEN_NUM, TOKEN_REG, TOKEN_INSN,   /* major token types */
    TOKEN_ERRNUM,               /* numeric constant with error in */
    TOKEN_HERE, TOKEN_BASE,     /* $ and $$ */
    TOKEN_SPECIAL,              /* BYTE, WORD, DWORD, QWORD, FAR, NEAR, etc */
    TOKEN_PREFIX,               /* A32, O16, LOCK, REPNZ, TIMES, etc */
    TOKEN_SHL, TOKEN_SHR,       /* << and >> */
    TOKEN_SDIV, TOKEN_SMOD,     /* // and %% */
    TOKEN_GE, TOKEN_LE, TOKEN_NE,       /* >=, <= and <> (!= is same as <>) */
    TOKEN_DBL_AND, TOKEN_DBL_OR, TOKEN_DBL_XOR, /* &&, || and ^^ */
    TOKEN_SEG, TOKEN_WRT,       /* SEG and WRT */
    TOKEN_FLOAT,                /* floating-point constant */
    TOKEN_FLOATIZE,		/* __floatX__ */
};

enum floatize {
    FLOAT_16,
    FLOAT_32,
    FLOAT_64,
    FLOAT_80M,
    FLOAT_80E,
    FLOAT_128L,
    FLOAT_128H,
};

struct location {
    int64_t offset;
    int32_t segment;
    int known;
};

/*
 * Expression-evaluator datatype. Expressions, within the
 * evaluator, are stored as an array of these beasts, terminated by
 * a record with type==0. Mostly, it's a vector type: each type
 * denotes some kind of a component, and the value denotes the
 * multiple of that component present in the expression. The
 * exception is the WRT type, whose `value' field denotes the
 * segment to which the expression is relative. These segments will
 * be segment-base types, i.e. either odd segment values or SEG_ABS
 * types. So it is still valid to assume that anything with a
 * `value' field of zero is insignificant.
 */
typedef struct {
    int32_t type;                  /* a register, or EXPR_xxx */
    int64_t value;                 /* must be >= 32 bits */
} expr;

/*
 * Library routines to manipulate expression data types.
 */
int is_reloc(expr *);
int is_simple(expr *);
int is_really_simple(expr *);
int is_unknown(expr *);
int is_just_unknown(expr *);
int64_t reloc_value(expr *);
int32_t reloc_seg(expr *);
int32_t reloc_wrt(expr *);

/*
 * The evaluator can also return hints about which of two registers
 * used in an expression should be the base register. See also the
 * `operand' structure.
 */
struct eval_hints {
    int64_t base;
    int type;
};

/*
 * The actual expression evaluator function looks like this. When
 * called, it expects the first token of its expression to already
 * be in `*tv'; if it is not, set tv->t_type to TOKEN_INVALID and
 * it will start by calling the scanner.
 *
 * If a forward reference happens during evaluation, the evaluator
 * must set `*fwref' to true if `fwref' is non-NULL.
 *
 * `critical' is non-zero if the expression may not contain forward
 * references. The evaluator will report its own error if this
 * occurs; if `critical' is 1, the error will be "symbol not
 * defined before use", whereas if `critical' is 2, the error will
 * be "symbol undefined".
 *
 * If `critical' has bit 8 set (in addition to its main value: 0x101
 * and 0x102 correspond to 1 and 2) then an extended expression
 * syntax is recognised, in which relational operators such as =, <
 * and >= are accepted, as well as low-precedence logical operators
 * &&, ^^ and ||.
 *
 * If `hints' is non-NULL, it gets filled in with some hints as to
 * the base register in complex effective addresses.
 */
#define CRITICAL 0x100
typedef expr *(*evalfunc) (scanner sc, void *scprivate,
                           struct tokenval * tv, int *fwref, int critical,
                           efunc error, struct eval_hints * hints);

/*
 * Special values for expr->type.  These come after EXPR_REG_END
 * as defined in regs.h.
 */

#define EXPR_UNKNOWN	(EXPR_REG_END+1) /* forward references */
#define EXPR_SIMPLE	(EXPR_REG_END+2)
#define EXPR_WRT	(EXPR_REG_END+3)
#define EXPR_SEGBASE	(EXPR_REG_END+4)

/*
 * Preprocessors ought to look like this:
 */
typedef struct preproc_ops {
    /*
     * Called at the start of a pass; given a file name, the number
     * of the pass, an error reporting function, an evaluator
     * function, and a listing generator to talk to.
     */
    void (*reset) (char *, int, efunc, evalfunc, ListGen *);

    /*
     * Called to fetch a line of preprocessed source. The line
     * returned has been malloc'ed, and so should be freed after
     * use.
     */
    char *(*getline) (void);

    /*
     * Called at the end of a pass.
     */
    void (*cleanup) (int);
} Preproc;

extern Preproc nasmpp;

/*
 * ----------------------------------------------------------------
 * Some lexical properties of the NASM source language, included
 * here because they are shared between the parser and preprocessor
 * ----------------------------------------------------------------
 */

/*
 * isidstart matches any character that may start an identifier, and isidchar
 * matches any character that may appear at places other than the start of an
 * identifier. E.g. a period may only appear at the start of an identifier
 * (for local labels), whereas a number may appear anywhere *but* at the
 * start. 
 */

#define isidstart(c) ( isalpha(c) || (c)=='_' || (c)=='.' || (c)=='?' \
                                  || (c)=='@' )
#define isidchar(c)  ( isidstart(c) || isdigit(c) || (c)=='$' || (c)=='#' \
                                                  || (c)=='~' )

/* Ditto for numeric constants. */

#define isnumstart(c)  ( isdigit(c) || (c)=='$' )
#define isnumchar(c)   ( isalnum(c) )

/* This returns the numeric value of a given 'digit'. */

#define numvalue(c)  ((c)>='a' ? (c)-'a'+10 : (c)>='A' ? (c)-'A'+10 : (c)-'0')

/*
 * Data-type flags that get passed to listing-file routines.
 */
enum {
    LIST_READ, LIST_MACRO, LIST_MACRO_NOLIST, LIST_INCLUDE,
    LIST_INCBIN, LIST_TIMES