nasm.h

开源的nasm编译器源码,研究编译器原理很有帮且

/* special type of immediate operand */
#define ONENESS   0x00800000L          /* so UNITY == IMMEDIATE | ONENESS */
#define UNITY     0x00802000L	       /* for shift/rotate instructions */
#define BYTENESS  0x40000000L          /* so SBYTE == IMMEDIATE | BYTENESS */
#define SBYTE 	  0x40002000L	       /* for op r16/32,immediate instrs. */
		
/* Register names automatically generated from regs.dat */
#include "regs.h"

enum {				       /* condition code names */
    C_A, C_AE, C_B, C_BE, C_C, C_E, C_G, C_GE, C_L, C_LE, C_NA, C_NAE,
    C_NB, C_NBE, C_NC, C_NE, C_NG, C_NGE, C_NL, C_NLE, C_NO, C_NP,
    C_NS, C_NZ, C_O, C_P, C_PE, C_PO, C_S, C_Z
};

/*
 * Note that because segment registers may be used as instruction
 * prefixes, we must ensure the enumerations for prefixes and
 * register names do not overlap.
 */
enum {				       /* instruction prefixes */
    PREFIX_ENUM_START = REG_ENUM_LIMIT,
    P_A16 = PREFIX_ENUM_START, P_A32, P_LOCK, P_O16, P_O32, P_REP, P_REPE,
    P_REPNE, P_REPNZ, P_REPZ, P_TIMES
};

enum {				       /* extended operand types */
    EOT_NOTHING, EOT_DB_STRING, EOT_DB_NUMBER
};

enum {				       /* special EA flags */
    EAF_BYTEOFFS = 1,		       /* force offset part to byte size */
    EAF_WORDOFFS = 2,		       /* force offset part to [d]word size */
    EAF_TIMESTWO = 4		       /* really do EAX*2 not EAX+EAX */
};

enum {				       /* values for `hinttype' */
    EAH_NOHINT = 0,		       /* no hint at all - our discretion */
    EAH_MAKEBASE = 1,		       /* try to make given reg the base */
    EAH_NOTBASE = 2		       /* try _not_ to make reg the base */
};

typedef struct {		       /* operand to an instruction */
    long type;			       /* type of operand */
    int addr_size;		       /* 0 means default; 16; 32 */
    int basereg, indexreg, scale;      /* registers and scale involved */
    int hintbase, hinttype;	       /* hint as to real base register */
    long segment;		       /* immediate segment, if needed */
    long offset;		       /* any immediate number */
    long wrt;			       /* segment base it's relative to */
    int eaflags;		       /* special EA flags */
    int opflags;		       /* see OPFLAG_* defines below */
} operand;

#define OPFLAG_FORWARD		1      /* operand is a forward reference */
#define OPFLAG_EXTERN		2      /* operand is an external reference */

typedef struct extop {		       /* extended operand */
    struct extop *next;		       /* linked list */
    long type;			       /* defined above */
    char *stringval;		       /* if it's a string, then here it is */
    int stringlen;		       /* ... and here's how long it is */
    long segment;		       /* if it's a number/address, then... */
    long offset;		       /* ... it's given here ... */
    long wrt;			       /* ... and here */
} extop;

#define MAXPREFIX 4

typedef struct {		       /* an instruction itself */
    char *label;		       /* the label defined, or NULL */
    int prefixes[MAXPREFIX];	       /* instruction prefixes, if any */
    int nprefix;		       /* number of entries in above */
    int opcode;			       /* the opcode - not just the string */
    int condition;		       /* the condition code, if Jcc/SETcc */
    int operands;		       /* how many operands? 0-3 
                                        * (more if db et al) */
    operand oprs[3];	   	       /* the operands, defined as above */
    extop *eops;		       /* extended operands */
    int eops_float;                    /* true if DD and floating */
    long times;			       /* repeat count (TIMES prefix) */
    int forw_ref;		       /* is there a forward reference? */
} insn;

enum geninfo { GI_SWITCH };
/*
 * ------------------------------------------------------------
 * The data structure defining an output format driver, and the
 * interfaces to the functions therein.
 * ------------------------------------------------------------
 */

struct ofmt {
    /*
     * This is a short (one-liner) description of the type of
     * output generated by the driver.
     */
    const char *fullname;

    /*
     * This is a single keyword used to select the driver.
     */
    const char *shortname;

    /*
     * this is reserved for out module specific help.
     * It is set to NULL in all the out modules but is not implemented
     * in the main program
     */
    const char *helpstring;

    /*
     * this is a pointer to the first element of the debug information
     */
    struct dfmt **debug_formats;

    /*
     * and a pointer to the element that is being used
     * note: this is set to the default at compile time and changed if the
     * -F option is selected.  If developing a set of new debug formats for
     * an output format, be sure to set this to whatever default you want
     *
     */
    struct dfmt *current_dfmt;

    /*
     * This, if non-NULL, is a NULL-terminated list of `char *'s
     * pointing to extra standard macros supplied by the object
     * format (e.g. a sensible initial default value of __SECT__,
     * and user-level equivalents for any format-specific
     * directives).
     */
    const char **stdmac;

    /*
     * This procedure is called at the start of an output session.
     * It tells the output format what file it will be writing to,
     * what routine to report errors through, and how to interface
     * to the label manager and expression evaluator if necessary.
     * It also gives it a chance to do other initialisation.
     */
    void (*init) (FILE *fp, efunc error, ldfunc ldef, evalfunc eval);

    /*
     * This procedure is called to pass generic information to the
     * object file.  The first parameter gives the information type
     * (currently only command line switches)
     * and the second parameter gives the value.  This function returns
     * 1 if recognized, 0 if unrecognized
     */
    int (*setinfo)(enum geninfo type, char **string);

    /*
     * This procedure is called by assemble() to write actual
     * generated code or data to the object file. Typically it
     * doesn't have to actually _write_ it, just store it for
     * later.
     *
     * The `type' argument specifies the type of output data, and
     * usually the size as well: its contents are described below.
     */
    void (*output) (long segto, const void *data, unsigned long type,
		    long segment, long wrt);

    /*
     * This procedure is called once for every symbol defined in
     * the module being assembled. It gives the name and value of
     * the symbol, in NASM's terms, and indicates whether it has
     * been declared to be global. Note that the parameter "name",
     * when passed, will point to a piece of static storage
     * allocated inside the label manager - it's safe to keep using
     * that pointer, because the label manager doesn't clean up
     * until after the output driver has.
     *
     * Values of `is_global' are: 0 means the symbol is local; 1
     * means the symbol is global; 2 means the symbol is common (in
     * which case `offset' holds the _size_ of the variable).
     * Anything else is available for the output driver to use
     * internally.
     *
     * This routine explicitly _is_ allowed to call the label
     * manager to define further symbols, if it wants to, even
     * though it's been called _from_ the label manager. That much
     * re-entrancy is guaranteed in the label manager. However, the
     * label manager will in turn call this routine, so it should
     * be prepared to be re-entrant itself.
     *
     * The `special' parameter contains special information passed
     * through from the command that defined the label: it may have
     * been an EXTERN, a COMMON or a GLOBAL. The distinction should
     * be obvious to the output format from the other parameters.
     */
    void (*symdef) (char *name, long segment, long offset, int is_global,
		    char *special);

    /*
     * This procedure is called when the source code requests a
     * segment change. It should return the corresponding segment
     * _number_ for the name, or NO_SEG if the name is not a valid
     * segment name.
     *
     * It may also be called with NULL, in which case it is to
     * return the _default_ section number for starting assembly in.
     *
     * It is allowed to modify the string it is given a pointer to.
     *
     * It is also allowed to specify a default instruction size for
     * the segment, by setting `*bits' to 16 or 32. Or, if it
     * doesn't wish to define a default, it can leave `bits' alone.
     */
    long (*section) (char *name, int pass, int *bits);

    /*
     * This procedure is called to modify the segment base values
     * returned from the SEG operator. It is given a segment base
     * value (i.e. a segment value with the low bit set), and is
     * required to produce in return a segment value which may be
     * different. It can map segment bases to absolute numbers by
     * means of returning SEG_ABS types.
     *
     * It should return NO_SEG if the segment base cannot be
     * determined; the evaluator (which calls this routine) is
     * responsible for throwing an error condition if that occurs
     * in pass two or in a critical expression.
     */
    long (*segbase) (long segment);

    /*
     * This procedure is called to allow the output driver to
     * process its own specific directives. When called, it has the
     * directive word in `directive' and the parameter string in
     * `value'. It is called in both assembly passes, and `pass'
     * will be either 1 or 2.
     *
     * This procedure should return zero if it does not _recognise_
     * the directive, so that the main program can report an error.
     * If it recognises the directive but then has its own errors,
     * it should report them itself and then return non-zero. It
     * should also return non-zero if it correctly processes the
     * directive.
     */
    int (*directive) (char *directive, char *value, int pass);

    /*
     * This procedure is called before anything else - even before
     * the "init" routine - and is passed the name of the input
     * file from which this output file is being generated. It
     * should return its preferred name for the output file in
     * `outname', if outname[0] is not '\0', and do nothing to
     * `outname' otherwise. Since it is called before the driver is
     * properly initialised, it has to be passed its error handler
     * separately.
     *
     * This procedure may also take its own copy of the input file
     * name for use in writing the output file: it is _guaranteed_
     * that it will be called before the "init" routine.
     *
     * The parameter `outname' points to an area of storage
     * guaranteed to be at least FILENAME_MAX in size.
     */
    void (*filename) (char *inname, char *outname, efunc error);

    /*
     * This procedure is called after assembly finishes, to allow
     * the output driver to clean itself up and free its memory.
     * Typically, it will also be the point at which the object
     * file actually gets _written_.
     *
     * One thing the cleanup routine should always do is to close
     * the output file pointer.
     */
    void (*cleanup) (int debuginfo);
};

/*
 * values for the `type' parameter to an output function. Each one
 * must have the actual number of _bytes_ added to it.
 *
 * Exceptions are OUT_RELxADR, which denote an x-byte relocation
 * which will be a relative jump. For this we need to know the
 * distance in bytes from the start of the relocated record until
 * the end of the containing instruction. _This_ is what is stored
 * in the size part of the parameter, in this case.
 *
 * Also OUT_RESERVE denotes reservation of N bytes of BSS space,
 * and the contents of the "data" parameter is irrelevant.
 *
 * The "data" parameter for the output function points to a "long",
 * containing the address in question, unless the type is
 * OUT_RAWDATA, in which case it points to an "unsigned char"
 * array.
 */
#define OUT_RAWDATA 0x00000000UL
#define OUT_ADDRESS 0x10000000UL
#define OUT_REL2ADR 0x20000000UL
#define OUT_REL4ADR 0x30000000UL
#define OUT_RESERVE 0x40000000UL
#define OUT_TYPMASK 0xF0000000UL
#define OUT_SIZMASK 0x0FFFFFFFUL

/*
 * ------------------------------------------------------------
 * The data structure defining a debug format driver, and the
 * interfaces to the functions therein.
 * ------------------------------------------------------------
 */

struct dfmt {
    
    /*
     * This is a short (one-liner) description of the type of
     * output generated by the driver.
     */
    const char *fullname;

    /*
     * This is a single keyword used to select the driver.
     */
    const char *shortname;


    /*
     * init - called initially to set up local pointer to object format, 
     * void pointer to implementation defined data, file pointer (which
     * probably won't be used, but who knows?), and error function.
     */
    void (*init) (struct ofmt * of, void * id, FILE * fp, efunc error);

    /*
     * linenum - called any time there is output with a change of
     * line number or file.
     */
    void (*linenum) (const char * filename, long linenumber, long segto);

    /*
     * debug_deflabel - called whenever a label is defined. Parameters
     * are the same as to 'symdef()' in the output format. This function
     * would be called before the output format version.
     */

    void (*debug_deflabel) (char * name, long segment, long offset,
                            int is_global, char * special);
    /*
     * debug_directive - called whenever a DEBUG directive other than 'LINE'
     * is encountered. 'directive' contains the first parameter to the
     * DEBUG directive, and params contains the rest. For example,
     * 'DEBUG VAR _somevar:int' would translate to a call to this
     * function with 'directive' equal to "VAR" and 'params' equal to 
     * "_somevar:int".
     */
    void (*debug_directive) (const char * directive, const char * params);

    /*
     * typevalue - called whenever the assembler wishes to register a type
     * for the last defined label.  This routine MUST detect if a type was
     * already registered and not re-register it.
     */
    void (*debug_typevalue) (long type);

    /*
     * debug_output - called whenever output is required
     * 'type' is the type of info required, and this is format-specific
     */
    void (*debug_output) (int type, void *param);

    /*
     * cleanup - called after processing of file is complete
     */
    void (*cleanup) (void);

};
/*
 * The type definition macros
 * for debugging
 *
 * low 3 bits: reserved
 * next 5 bits: type
 * next 24 bits: number of elements for arrays (0 for labels)
 */

#define TY_UNKNOWN 0x00
#define TY_LABEL   0x08
#define TY_BYTE    0x10
#define TY_WORD    0x18
#define TY_DWORD   0x20
#define TY_FLOAT   0x28
#define TY_QWORD   0x30
#define TY_TBYTE   0x38
#define TY_COMMON  0xE0
#define TY_SEG     0xE8
#define TY_EXTERN  0xF0
#define TY_EQU     0xF8

#define TYM_TYPE(x) ((x) & 0xF8)
#define TYM_ELEMENTS(x) (((x) & 0xFFFFFF00) >> 8)

#define TYS_ELEMENTS(x)  ((x) << 8)
/*
 * -----
 * Other
 * -----
 */

/*
 * This is a useful #define which I keep meaning to use more often:
 * the number of elements of a statically defined array.
 */

#define elements(x)     ( sizeof(x) / sizeof(*(x)) )

extern int tasm_compatible_mode;

/*
 * This declaration passes the "pass" number to all other modules
 * "pass0" assumes the values: 0, 0, ..., 0, 1, 2
 * where 0 = optimizing pass
 *       1 = pass 1
 *       2 = pass 2
 */

extern int pass0;	/* this is globally known */
extern int optimizing;

#endif