1 /* nasm.h main header file for the Netwide Assembler: inter-module interface
3 * The Netwide Assembler is copyright (C) 1996 Simon Tatham and
4 * Julian Hall. All rights reserved. The software is
5 * redistributable under the licence given in the file "Licence"
6 * distributed in the NASM archive.
8 * initial version: 27/iii/95 by Simon Tatham
16 #include "version.h" /* generated NASM version macros */
19 #include "insnsi.h" /* For enum opcode */
26 #define FALSE 0 /* comes in handy */
32 #define NO_SEG -1L /* null segment value */
33 #define SEG_ABS 0x40000000L /* mask for far-absolute segments */
36 #define FILENAME_MAX 256
44 #define POSTFIX_MAX 10
47 #define IDLEN_MAX 4096
50 * Name pollution problems: <time.h> on Digital UNIX pulls in some
51 * strange hardware header file which sees fit to define R_SP. We
52 * undefine it here so as not to break the enum below.
59 * We must declare the existence of this structure type up here,
60 * since we have to reference it before we define it...
65 * -----------------------
66 * Other function typedefs
67 * -----------------------
71 * A label-lookup function should look like this.
73 typedef int (*lfunc) (char *label, int32_t *segment, int32_t *offset);
76 * And a label-definition function like this. The boolean parameter
77 * `is_norm' states whether the label is a `normal' label (which
78 * should affect the local-label system), or something odder like
79 * an EQU or a segment-base symbol, which shouldn't.
81 typedef void (*ldfunc) (char *label, int32_t segment, int32_t offset,
82 char *special, int is_norm, int isextrn,
83 struct ofmt * ofmt, efunc error);
86 * List-file generators should look like this:
90 * Called to initialize the listing file generator. Before this
91 * is called, the other routines will silently do nothing when
92 * called. The `char *' parameter is the file name to write the
95 void (*init) (char *, efunc);
98 * Called to clear stuff up and close the listing file.
100 void (*cleanup) (void);
103 * Called to output binary data. Parameters are: the offset;
104 * the data; the data type. Data types are similar to the
105 * output-format interface, only OUT_ADDRESS will _always_ be
106 * displayed as if it's relocatable, so ensure that any non-
107 * relocatable address has been converted to OUT_RAWDATA by
108 * then. Note that OUT_RAWDATA+0 is a valid data type, and is a
109 * dummy call used to give the listing generator an offset to
110 * work with when doing things like uplevel(LIST_TIMES) or
111 * uplevel(LIST_INCBIN).
113 void (*output) (int32_t, const void *, uint32_t);
116 * Called to send a text line to the listing generator. The
117 * `int' parameter is LIST_READ or LIST_MACRO depending on
118 * whether the line came directly from an input file or is the
119 * result of a multi-line macro expansion.
121 void (*line) (int, char *);
124 * Called to change one of the various levelled mechanisms in
125 * the listing generator. LIST_INCLUDE and LIST_MACRO can be
126 * used to increase the nesting level of include files and
127 * macro expansions; LIST_TIMES and LIST_INCBIN switch on the
128 * two binary-output-suppression mechanisms for large-scale
129 * pseudo-instructions.
131 * LIST_MACRO_NOLIST is synonymous with LIST_MACRO except that
132 * it indicates the beginning of the expansion of a `nolist'
133 * macro, so anything under that level won't be expanded unless
134 * it includes another file.
136 void (*uplevel) (int);
139 * Reverse the effects of uplevel.
141 void (*downlevel) (int);
145 * The expression evaluator must be passed a scanner function; a
146 * standard scanner is provided as part of nasmlib.c. The
147 * preprocessor will use a different one. Scanners, and the
148 * token-value structures they return, look like this.
150 * The return value from the scanner is always a copy of the
151 * `t_type' field in the structure.
155 int64_t t_integer, t_inttwo;
158 typedef int (*scanner) (void *private_data, struct tokenval * tv);
161 * Token types returned by the scanner, in addition to ordinary
162 * ASCII character values, and zero for end-of-string.
164 enum { /* token types, other than chars */
165 TOKEN_INVALID = -1, /* a placeholder value */
166 TOKEN_EOS = 0, /* end of string */
167 TOKEN_EQ = '=', TOKEN_GT = '>', TOKEN_LT = '<', /* aliases */
168 TOKEN_ID = 256, TOKEN_NUM, TOKEN_REG, TOKEN_INSN, /* major token types */
169 TOKEN_ERRNUM, /* numeric constant with error in */
170 TOKEN_HERE, TOKEN_BASE, /* $ and $$ */
171 TOKEN_SPECIAL, /* BYTE, WORD, DWORD, QWORD, FAR, NEAR, etc */
172 TOKEN_PREFIX, /* A32, O16, LOCK, REPNZ, TIMES, etc */
173 TOKEN_SHL, TOKEN_SHR, /* << and >> */
174 TOKEN_SDIV, TOKEN_SMOD, /* // and %% */
175 TOKEN_GE, TOKEN_LE, TOKEN_NE, /* >=, <= and <> (!= is same as <>) */
176 TOKEN_DBL_AND, TOKEN_DBL_OR, TOKEN_DBL_XOR, /* &&, || and ^^ */
177 TOKEN_SEG, TOKEN_WRT, /* SEG and WRT */
178 TOKEN_FLOAT /* floating-point constant */
188 * Expression-evaluator datatype. Expressions, within the
189 * evaluator, are stored as an array of these beasts, terminated by
190 * a record with type==0. Mostly, it's a vector type: each type
191 * denotes some kind of a component, and the value denotes the
192 * multiple of that component present in the expression. The
193 * exception is the WRT type, whose `value' field denotes the
194 * segment to which the expression is relative. These segments will
195 * be segment-base types, i.e. either odd segment values or SEG_ABS
196 * types. So it is still valid to assume that anything with a
197 * `value' field of zero is insignificant.
200 int32_t type; /* a register, or EXPR_xxx */
201 int64_t value; /* must be >= 32 bits */
205 * Library routines to manipulate expression data types.
207 int is_reloc(expr *);
208 int is_simple(expr *);
209 int is_really_simple(expr *);
210 int is_unknown(expr *);
211 int is_just_unknown(expr *);
212 int64_t reloc_value(expr *);
213 int32_t reloc_seg(expr *);
214 int32_t reloc_wrt(expr *);
217 * The evaluator can also return hints about which of two registers
218 * used in an expression should be the base register. See also the
219 * `operand' structure.
227 * The actual expression evaluator function looks like this. When
228 * called, it expects the first token of its expression to already
229 * be in `*tv'; if it is not, set tv->t_type to TOKEN_INVALID and
230 * it will start by calling the scanner.
232 * If a forward reference happens during evaluation, the evaluator
233 * must set `*fwref' to TRUE if `fwref' is non-NULL.
235 * `critical' is non-zero if the expression may not contain forward
236 * references. The evaluator will report its own error if this
237 * occurs; if `critical' is 1, the error will be "symbol not
238 * defined before use", whereas if `critical' is 2, the error will
239 * be "symbol undefined".
241 * If `critical' has bit 8 set (in addition to its main value: 0x101
242 * and 0x102 correspond to 1 and 2) then an extended expression
243 * syntax is recognised, in which relational operators such as =, <
244 * and >= are accepted, as well as low-precedence logical operators
247 * If `hints' is non-NULL, it gets filled in with some hints as to
248 * the base register in complex effective addresses.
250 #define CRITICAL 0x100
251 typedef expr *(*evalfunc) (scanner sc, void *scprivate,
252 struct tokenval * tv, int *fwref, int critical,
253 efunc error, struct eval_hints * hints);
256 * Special values for expr->type. These come after EXPR_REG_END
257 * as defined in regs.h.
260 #define EXPR_UNKNOWN (EXPR_REG_END+1) /* forward references */
261 #define EXPR_SIMPLE (EXPR_REG_END+2)
262 #define EXPR_WRT (EXPR_REG_END+3)
263 #define EXPR_SEGBASE (EXPR_REG_END+4)
266 * Preprocessors ought to look like this:
268 typedef struct preproc_ops {
270 * Called at the start of a pass; given a file name, the number
271 * of the pass, an error reporting function, an evaluator
272 * function, and a listing generator to talk to.
274 void (*reset) (char *, int, efunc, evalfunc, ListGen *);
277 * Called to fetch a line of preprocessed source. The line
278 * returned has been malloc'ed, and so should be freed after
281 char *(*getline) (void);
284 * Called at the end of a pass.
286 void (*cleanup) (int);
289 extern Preproc nasmpp;
292 * ----------------------------------------------------------------
293 * Some lexical properties of the NASM source language, included
294 * here because they are shared between the parser and preprocessor
295 * ----------------------------------------------------------------
299 * isidstart matches any character that may start an identifier, and isidchar
300 * matches any character that may appear at places other than the start of an
301 * identifier. E.g. a period may only appear at the start of an identifier
302 * (for local labels), whereas a number may appear anywhere *but* at the
306 #define isidstart(c) ( isalpha(c) || (c)=='_' || (c)=='.' || (c)=='?' \
308 #define isidchar(c) ( isidstart(c) || isdigit(c) || (c)=='$' || (c)=='#' \
311 /* Ditto for numeric constants. */
313 #define isnumstart(c) ( isdigit(c) || (c)=='$' )
314 #define isnumchar(c) ( isalnum(c) )
316 /* This returns the numeric value of a given 'digit'. */
318 #define numvalue(c) ((c)>='a' ? (c)-'a'+10 : (c)>='A' ? (c)-'A'+10 : (c)-'0')
321 * Data-type flags that get passed to listing-file routines.
324 LIST_READ, LIST_MACRO, LIST_MACRO_NOLIST, LIST_INCLUDE,
325 LIST_INCBIN, LIST_TIMES
329 * -----------------------------------------------------------
330 * Format of the `insn' structure returned from `parser.c' and
331 * passed into `assemble.c'
332 * -----------------------------------------------------------
336 * Here we define the operand types. These are implemented as bit
337 * masks, since some are subsets of others; e.g. AX in a MOV
338 * instruction is a special operand type, whereas AX in other
339 * contexts is just another 16-bit register. (Also, consider CL in
340 * shift instructions, DX in OUT, etc.)
342 * The basic concept here is that
343 * (class & ~operand) == 0
345 * if and only if "operand" belongs to class type "class".
347 * The bits are assigned as follows:
349 * Bits 0-7, 29: sizes
358 * 29: 128 bits (OWORD)
360 * Bits 8-11 modifiers
366 * Bits 12-15: type of operand
369 * 14: MEMORY (always has REGMEM attribute as well)
370 * 15: REGMEM (valid EA operand)
372 * Bits 16-19: subclasses
379 * 16: REG_ACCUM (AL, AX, EAX, RAX)
380 * 17: REG_COUNT (CL, CX, ECX, RCX)
381 * 18: REG_DATA (DL, DX, EDX, RDX)
382 * 19: REG_HIGH (AH, CH, DH, BH)
386 * 17: REG_DESS (DS, ES, SS)
397 * 16: MEM_OFFS (this is a simple offset)
398 * 17: IP_REL (IP-relative offset)
402 * 17: BYTENESS (-128..127)
404 * Bits 20-26: register classes
405 * 20: REG_CDT (CRx, DRx, TRx)
406 * 21: RM_GPR (REG_GPR) (integer register)
408 * 23: IP_REG (RIP or EIP) [unused]
410 * 25: RM_MMX (MMXREG)
411 * 26: RM_XMM (XMMREG)
413 * Bits 27-29 & 31 are currently unallocated.
416 * Special flag only used in instruction patterns; means this operand
417 * has to be identical to another operand. Currently only supported
421 typedef uint32_t opflags_t;
423 /* Size, and other attributes, of the operand */
424 #define BITS8 0x00000001L
425 #define BITS16 0x00000002L
426 #define BITS32 0x00000004L
427 #define BITS64 0x00000008L /* x64 and FPU only */
428 #define BITS80 0x00000010L /* FPU only */
429 #define BITS128 0x20000000L
430 #define FAR 0x00000020L /* grotty: this means 16:16 or */
431 /* 16:32, like in CALL/JMP */
432 #define NEAR 0x00000040L
433 #define SHORT 0x00000080L /* and this means what it says :) */
435 #define SIZE_MASK 0x200000FFL /* all the size attributes */
438 #define MODIFIER_MASK 0x00000f00L
439 #define TO 0x00000100L /* reverse effect in FADD, FSUB &c */
440 #define COLON 0x00000200L /* operand is followed by a colon */
441 #define STRICT 0x00000400L /* do not optimize this operand */
443 /* Type of operand: memory reference, register, etc. */
444 #define OPTYPE_MASK 0x0000f000L
445 #define REGISTER 0x00001000L /* register number in 'basereg' */
446 #define IMMEDIATE 0x00002000L
447 #define MEMORY 0x0000c000L
448 #define REGMEM 0x00008000L /* for r/m, ie EA, operands */
450 /* Register classes */
451 #define REG_EA 0x00009000L /* 'normal' reg, qualifies as EA */
452 #define RM_GPR 0x00208000L /* integer operand */
453 #define REG_GPR 0x00209000L /* integer register */
454 #define REG8 0x00209001L /* 8-bit GPR */
455 #define REG16 0x00209002L /* 16-bit GPR */
456 #define REG32 0x00209004L /* 32-bit GPR */
457 #define REG64 0x00209008L /* 64-bit GPR */
458 #define IP_REG 0x00801000L /* RIP or EIP register */
459 #define RIPREG 0x00801008L /* RIP */
460 #define EIPREG 0x00801004L /* EIP */
461 #define FPUREG 0x01001000L /* floating point stack registers */
462 #define FPU0 0x01011000L /* FPU stack register zero */
463 #define RM_MMX 0x02008000L /* MMX operand */
464 #define MMXREG 0x02009000L /* MMX register */
465 #define RM_XMM 0x04008000L /* XMM (SSE) operand */
466 #define XMMREG 0x04009000L /* XMM (SSE) register */
467 #define XMM0 0x04019000L /* XMM register zero */
468 #define REG_CDT 0x00101004L /* CRn, DRn and TRn */
469 #define REG_CREG 0x00111004L /* CRn */
470 #define REG_DREG 0x00121004L /* DRn */
471 #define REG_TREG 0x00141004L /* TRn */
472 #define REG_SREG 0x00401002L /* any segment register */
473 #define REG_CS 0x00411002L /* CS */
474 #define REG_DESS 0x00421002L /* DS, ES, SS */
475 #define REG_FSGS 0x00441002L /* FS, GS */
476 #define REG_SEG67 0x00481002L /* Unimplemented segment registers */
478 #define REG_RIP 0x00801008L /* RIP relative addressing */
479 #define REG_EIP 0x00801004L /* EIP relative addressing */
482 #define REG_SMASK 0x000f0000L /* a mask for the following */
483 #define REG_ACCUM 0x00219000L /* accumulator: AL, AX, EAX, RAX */
484 #define REG_AL 0x00219001L
485 #define REG_AX 0x00219002L
486 #define REG_EAX 0x00219004L
487 #define REG_RAX 0x00219008L
488 #define REG_COUNT 0x00229000L /* counter: CL, CX, ECX, RCX */
489 #define REG_CL 0x00229001L
490 #define REG_CX 0x00229002L
491 #define REG_ECX 0x00229004L
492 #define REG_RCX 0x00229008L
493 #define REG_DL 0x00249001L /* data: DL, DX, EDX, RDX */
494 #define REG_DX 0x00249002L
495 #define REG_EDX 0x00249004L
496 #define REG_RDX 0x00249008L
497 #define REG_HIGH 0x00289001L /* high regs: AH, CH, DH, BH */
499 /* special types of EAs */
500 #define MEM_OFFS 0x0001c000L /* simple [address] offset - absolute! */
501 #define IP_REL 0x0002c000L /* IP-relative offset */
503 /* memory which matches any type of r/m operand */
504 #define MEMORY_ANY (MEMORY|RM_GPR|RM_MMX|RM_XMM)
506 /* special type of immediate operand */
507 #define UNITY 0x00012000L /* for shift/rotate instructions */
508 #define SBYTE 0x00022000L /* for op r16/32,immediate instrs. */
511 #define SAME_AS 0x40000000L
513 /* Register names automatically generated from regs.dat */
516 enum ccode { /* condition code names */
517 C_A, C_AE, C_B, C_BE, C_C, C_E, C_G, C_GE, C_L, C_LE, C_NA, C_NAE,
518 C_NB, C_NBE, C_NC, C_NE, C_NG, C_NGE, C_NL, C_NLE, C_NO, C_NP,
519 C_NS, C_NZ, C_O, C_P, C_PE, C_PO, C_S, C_Z,
526 #define REX_OC 0x0200 /* DREX suffix has the OC0 bit set */
527 #define REX_D 0x0100 /* Instruction uses DREX instead of REX */
528 #define REX_H 0x80 /* High register present, REX forbidden */
529 #define REX_P 0x40 /* REX prefix present/required */
530 #define REX_L 0x20 /* Use LOCK prefix instead of REX.R */
531 #define REX_W 0x08 /* 64-bit operand size */
532 #define REX_R 0x04 /* ModRM reg extension */
533 #define REX_X 0x02 /* SIB index extension */
534 #define REX_B 0x01 /* ModRM r/m extension */
535 #define REX_REAL 0x4f /* Actual REX prefix bits */
538 * Note that because segment registers may be used as instruction
539 * prefixes, we must ensure the enumerations for prefixes and
540 * register names do not overlap.
542 enum prefixes { /* instruction prefixes */
543 PREFIX_ENUM_START = REG_ENUM_LIMIT,
544 P_A16 = PREFIX_ENUM_START, P_A32, P_LOCK, P_O16, P_O32,
545 P_REP, P_REPE, P_REPNE, P_REPNZ, P_REPZ, P_TIMES
548 enum { /* extended operand types */
549 EOT_NOTHING, EOT_DB_STRING, EOT_DB_NUMBER
552 enum { /* special EA flags */
553 EAF_BYTEOFFS = 1, /* force offset part to byte size */
554 EAF_WORDOFFS = 2, /* force offset part to [d]word size */
555 EAF_TIMESTWO = 4, /* really do EAX*2 not EAX+EAX */
556 EAF_REL = 8, /* IP-relative addressing */
557 EAF_ABS = 16, /* non-IP-relative addressing */
558 EAF_FSGS = 32 /* fs/gs segment override present */
561 enum eval_hint { /* values for `hinttype' */
562 EAH_NOHINT = 0, /* no hint at all - our discretion */
563 EAH_MAKEBASE = 1, /* try to make given reg the base */
564 EAH_NOTBASE = 2 /* try _not_ to make reg the base */
567 typedef struct { /* operand to an instruction */
568 int32_t type; /* type of operand */
569 int addr_size; /* 0 means default; 16; 32; 64 */
570 enum reg_enum basereg, indexreg; /* address registers */
571 int scale; /* index scale */
573 enum eval_hint hinttype; /* hint as to real base register */
574 int32_t segment; /* immediate segment, if needed */
575 int64_t offset; /* any immediate number */
576 int32_t wrt; /* segment base it's relative to */
577 int eaflags; /* special EA flags */
578 int opflags; /* see OPFLAG_* defines below */
581 #define OPFLAG_FORWARD 1 /* operand is a forward reference */
582 #define OPFLAG_EXTERN 2 /* operand is an external reference */
584 typedef struct extop { /* extended operand */
585 struct extop *next; /* linked list */
586 int32_t type; /* defined above */
587 char *stringval; /* if it's a string, then here it is */
588 int stringlen; /* ... and here's how long it is */
589 int32_t segment; /* if it's a number/address, then... */
590 int64_t offset; /* ... it's given here ... */
591 int32_t wrt; /* ... and here */
595 #define MAX_OPERANDS 4
597 typedef struct { /* an instruction itself */
598 char *label; /* the label defined, or NULL */
599 enum prefixes prefixes[MAXPREFIX]; /* instruction prefixes, if any */
600 int nprefix; /* number of entries in above */
601 enum opcode opcode; /* the opcode - not just the string */
602 enum ccode condition; /* the condition code, if Jcc/SETcc */
603 int operands; /* how many operands? 0-3
604 * (more if db et al) */
605 operand oprs[MAX_OPERANDS]; /* the operands, defined as above */
606 extop *eops; /* extended operands */
607 int eops_float; /* true if DD and floating */
608 int32_t times; /* repeat count (TIMES prefix) */
609 int forw_ref; /* is there a forward reference? */
610 int rex; /* Special REX Prefix */
611 int drexdst; /* Destination register for DREX suffix */
614 enum geninfo { GI_SWITCH };
616 * ------------------------------------------------------------
617 * The data structure defining an output format driver, and the
618 * interfaces to the functions therein.
619 * ------------------------------------------------------------
624 * This is a short (one-liner) description of the type of
625 * output generated by the driver.
627 const char *fullname;
630 * This is a single keyword used to select the driver.
632 const char *shortname;
636 * this is reserved for out module specific help.
637 * It is set to NULL in all the out modules and is not implemented
638 * in the main program
640 const char *helpstring;
643 * this is a pointer to the first element of the debug information
645 struct dfmt **debug_formats;
648 * and a pointer to the element that is being used
649 * note: this is set to the default at compile time and changed if the
650 * -F option is selected. If developing a set of new debug formats for
651 * an output format, be sure to set this to whatever default you want
654 struct dfmt *current_dfmt;
657 * This, if non-NULL, is a NULL-terminated list of `char *'s
658 * pointing to extra standard macros supplied by the object
659 * format (e.g. a sensible initial default value of __SECT__,
660 * and user-level equivalents for any format-specific
666 * This procedure is called at the start of an output session.
667 * It tells the output format what file it will be writing to,
668 * what routine to report errors through, and how to interface
669 * to the label manager and expression evaluator if necessary.
670 * It also gives it a chance to do other initialisation.
672 void (*init) (FILE * fp, efunc error, ldfunc ldef, evalfunc eval);
675 * This procedure is called to pass generic information to the
676 * object file. The first parameter gives the information type
677 * (currently only command line switches)
678 * and the second parameter gives the value. This function returns
679 * 1 if recognized, 0 if unrecognized
681 int (*setinfo) (enum geninfo type, char **string);
684 * This procedure is called by assemble() to write actual
685 * generated code or data to the object file. Typically it
686 * doesn't have to actually _write_ it, just store it for
689 * The `type' argument specifies the type of output data, and
690 * usually the size as well: its contents are described below.
692 void (*output) (int32_t segto, const void *data, uint32_t type,
693 int32_t segment, int32_t wrt);
696 * This procedure is called once for every symbol defined in
697 * the module being assembled. It gives the name and value of
698 * the symbol, in NASM's terms, and indicates whether it has
699 * been declared to be global. Note that the parameter "name",
700 * when passed, will point to a piece of static storage
701 * allocated inside the label manager - it's safe to keep using
702 * that pointer, because the label manager doesn't clean up
703 * until after the output driver has.
705 * Values of `is_global' are: 0 means the symbol is local; 1
706 * means the symbol is global; 2 means the symbol is common (in
707 * which case `offset' holds the _size_ of the variable).
708 * Anything else is available for the output driver to use
711 * This routine explicitly _is_ allowed to call the label
712 * manager to define further symbols, if it wants to, even
713 * though it's been called _from_ the label manager. That much
714 * re-entrancy is guaranteed in the label manager. However, the
715 * label manager will in turn call this routine, so it should
716 * be prepared to be re-entrant itself.
718 * The `special' parameter contains special information passed
719 * through from the command that defined the label: it may have
720 * been an EXTERN, a COMMON or a GLOBAL. The distinction should
721 * be obvious to the output format from the other parameters.
723 void (*symdef) (char *name, int32_t segment, int32_t offset, int is_global,
727 * This procedure is called when the source code requests a
728 * segment change. It should return the corresponding segment
729 * _number_ for the name, or NO_SEG if the name is not a valid
732 * It may also be called with NULL, in which case it is to
733 * return the _default_ section number for starting assembly in.
735 * It is allowed to modify the string it is given a pointer to.
737 * It is also allowed to specify a default instruction size for
738 * the segment, by setting `*bits' to 16 or 32. Or, if it
739 * doesn't wish to define a default, it can leave `bits' alone.
741 int32_t (*section) (char *name, int pass, int *bits);
744 * This procedure is called to modify the segment base values
745 * returned from the SEG operator. It is given a segment base
746 * value (i.e. a segment value with the low bit set), and is
747 * required to produce in return a segment value which may be
748 * different. It can map segment bases to absolute numbers by
749 * means of returning SEG_ABS types.
751 * It should return NO_SEG if the segment base cannot be
752 * determined; the evaluator (which calls this routine) is
753 * responsible for throwing an error condition if that occurs
754 * in pass two or in a critical expression.
756 int32_t (*segbase) (int32_t segment);
759 * This procedure is called to allow the output driver to
760 * process its own specific directives. When called, it has the
761 * directive word in `directive' and the parameter string in
762 * `value'. It is called in both assembly passes, and `pass'
763 * will be either 1 or 2.
765 * This procedure should return zero if it does not _recognise_
766 * the directive, so that the main program can report an error.
767 * If it recognises the directive but then has its own errors,
768 * it should report them itself and then return non-zero. It
769 * should also return non-zero if it correctly processes the
772 int (*directive) (char *directive, char *value, int pass);
775 * This procedure is called before anything else - even before
776 * the "init" routine - and is passed the name of the input
777 * file from which this output file is being generated. It
778 * should return its preferred name for the output file in
779 * `outname', if outname[0] is not '\0', and do nothing to
780 * `outname' otherwise. Since it is called before the driver is
781 * properly initialized, it has to be passed its error handler
784 * This procedure may also take its own copy of the input file
785 * name for use in writing the output file: it is _guaranteed_
786 * that it will be called before the "init" routine.
788 * The parameter `outname' points to an area of storage
789 * guaranteed to be at least FILENAME_MAX in size.
791 void (*filename) (char *inname, char *outname, efunc error);
794 * This procedure is called after assembly finishes, to allow
795 * the output driver to clean itself up and free its memory.
796 * Typically, it will also be the point at which the object
797 * file actually gets _written_.
799 * One thing the cleanup routine should always do is to close
800 * the output file pointer.
802 void (*cleanup) (int debuginfo);
806 * values for the `type' parameter to an output function. Each one
807 * must have the actual number of _bytes_ added to it.
809 * Exceptions are OUT_RELxADR, which denote an x-byte relocation
810 * which will be a relative jump. For this we need to know the
811 * distance in bytes from the start of the relocated record until
812 * the end of the containing instruction. _This_ is what is stored
813 * in the size part of the parameter, in this case.
815 * Also OUT_RESERVE denotes reservation of N bytes of BSS space,
816 * and the contents of the "data" parameter is irrelevant.
818 * The "data" parameter for the output function points to a "int32_t",
819 * containing the address in question, unless the type is
820 * OUT_RAWDATA, in which case it points to an "uint8_t"
823 #define OUT_RAWDATA 0x00000000UL
824 #define OUT_ADDRESS 0x10000000UL
825 #define OUT_REL2ADR 0x20000000UL
826 #define OUT_REL4ADR 0x30000000UL
827 #define OUT_RESERVE 0x40000000UL
828 #define OUT_TYPMASK 0xF0000000UL
829 #define OUT_SIZMASK 0x0FFFFFFFUL
832 * ------------------------------------------------------------
833 * The data structure defining a debug format driver, and the
834 * interfaces to the functions therein.
835 * ------------------------------------------------------------
841 * This is a short (one-liner) description of the type of
842 * output generated by the driver.
844 const char *fullname;
847 * This is a single keyword used to select the driver.
849 const char *shortname;
852 * init - called initially to set up local pointer to object format,
853 * void pointer to implementation defined data, file pointer (which
854 * probably won't be used, but who knows?), and error function.
856 void (*init) (struct ofmt * of, void *id, FILE * fp, efunc error);
859 * linenum - called any time there is output with a change of
860 * line number or file.
862 void (*linenum) (const char *filename, int32_t linenumber, int32_t segto);
865 * debug_deflabel - called whenever a label is defined. Parameters
866 * are the same as to 'symdef()' in the output format. This function
867 * would be called before the output format version.
870 void (*debug_deflabel) (char *name, int32_t segment, int32_t offset,
871 int is_global, char *special);
873 * debug_directive - called whenever a DEBUG directive other than 'LINE'
874 * is encountered. 'directive' contains the first parameter to the
875 * DEBUG directive, and params contains the rest. For example,
876 * 'DEBUG VAR _somevar:int' would translate to a call to this
877 * function with 'directive' equal to "VAR" and 'params' equal to
880 void (*debug_directive) (const char *directive, const char *params);
883 * typevalue - called whenever the assembler wishes to register a type
884 * for the last defined label. This routine MUST detect if a type was
885 * already registered and not re-register it.
887 void (*debug_typevalue) (int32_t type);
890 * debug_output - called whenever output is required
891 * 'type' is the type of info required, and this is format-specific
893 void (*debug_output) (int type, void *param);
896 * cleanup - called after processing of file is complete
898 void (*cleanup) (void);
902 * The type definition macros
905 * low 3 bits: reserved
907 * next 24 bits: number of elements for arrays (0 for labels)
910 #define TY_UNKNOWN 0x00
911 #define TY_LABEL 0x08
914 #define TY_DWORD 0x20
915 #define TY_FLOAT 0x28
916 #define TY_QWORD 0x30
917 #define TY_TBYTE 0x38
918 #define TY_COMMON 0xE0
920 #define TY_EXTERN 0xF0
923 #define TYM_TYPE(x) ((x) & 0xF8)
924 #define TYM_ELEMENTS(x) (((x) & 0xFFFFFF00) >> 8)
926 #define TYS_ELEMENTS(x) ((x) << 8)
934 enum special_tokens {
935 S_ABS, S_BYTE, S_DWORD, S_FAR, S_LONG, S_NEAR, S_NOSPLIT,
936 S_OWORD, S_QWORD, S_REL, S_SHORT, S_STRICT, S_TO, S_TWORD, S_WORD
946 * This is a useful #define which I keep meaning to use more often:
947 * the number of elements of a statically defined array.
950 #define elements(x) ( sizeof(x) / sizeof(*(x)) )
959 * This declaration passes the "pass" number to all other modules
960 * "pass0" assumes the values: 0, 0, ..., 0, 1, 2
961 * where 0 = optimizing pass
968 extern int tasm_compatible_mode;
969 extern int optimizing;
970 extern int globalbits; /* 16, 32 or 64-bit mode */
971 extern int globalrel; /* default to relative addressing? */
972 extern int maxbits; /* max bits supported by output */