nasm.h

一个汇编语言编译器源码

/* 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

#define NASM_MAJOR_VER 0
#define NASM_MINOR_VER 98
#define NASM_VER "0.98"

#ifndef NULL
#define NULL 0
#endif

#ifndef FALSE
#define FALSE 0			       /* comes in handy */
#endif
#ifndef TRUE
#define TRUE 1
#endif

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

#ifndef FILENAME_MAX
#define FILENAME_MAX 256
#endif

/*
 * 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;

/*
 * -------------------------
 * Error reporting functions
 * -------------------------
 */

/*
 * An error reporting function should look like this.
 */
typedef void (*efunc) (int severity, char *fmt, ...);

/*
 * These are the error severity codes which get passed as the first
 * argument to an efunc.
 */

#define ERR_WARNING 0		       /* warn only: no further action */
#define ERR_NONFATAL 1		       /* terminate assembly after phase */
#define ERR_FATAL 2		       /* instantly fatal: exit with error */
#define ERR_PANIC 3		       /* internal error: panic instantly
					* and dump core for reference */
#define ERR_MASK 0x0F		       /* mask off the above codes */
#define ERR_NOFILE 0x10		       /* don't give source file name/line */
#define ERR_USAGE 0x20		       /* print a usage message */
#define ERR_PASS1 0x80		       /* only print this error on pass one */

/*
 * These codes define specific types of suppressible warning.
 */
#define ERR_WARN_MNP  0x0100	       /* macro-num-parameters warning */
#define ERR_WARN_OL   0x0200	       /* orphan label (no colon, and
					* alone on line) */
#define ERR_WARN_NOV  0x0300	       /* numeric overflow */
#define ERR_WARN_MASK 0xFF00	       /* the mask for this feature */
#define ERR_WARN_SHR  8		       /* how far to shift right */
#define ERR_WARN_MAX  3		       /* the highest numbered one */

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

/*
 * A label-lookup function should look like this.
 */
typedef int (*lfunc) (char *label, long *segment, long *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, long segment, long offset, char *special,
			int is_norm, int isextrn, struct ofmt *ofmt,
			efunc error);

/*
 * List-file generators should look like this:
 */
typedef struct {
    /*
     * Called to initialise 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) (long, void *, unsigned long);

    /*
     * 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;
    long 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, 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 */
};

typedef struct {
    long segment;
    long offset;
    int  known;
} loc_t;

/*
 * 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 {
    long type;			       /* a register, or EXPR_xxx */
    long value;			       /* must be >= 32 bits */
} 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 {
    int 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 4 set (in addition to its main value: 0x11
 * and 0x12 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.
 */
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. ASSUMPTION MADE HERE: the number
 * of distinct register names (i.e. possible "type" fields for an
 * expr structure) does not exceed 124 (EXPR_REG_START through
 * EXPR_REG_END).
 */
#define EXPR_REG_START 1
#define EXPR_REG_END 124
#define EXPR_UNKNOWN 125L	       /* for forward references */
#define EXPR_SIMPLE 126L
#define EXPR_WRT 127L
#define EXPR_SEGBASE 128L

/*
 * Preprocessors ought to look like this:
 */