2 Copyright 1997, 1999, 2000, 2001, 2002, 2003, 2005
3 Free Software Foundation, Inc.
5 This file is part of GAS, the GNU Assembler.
7 GAS is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 2, or (at your option)
12 GAS is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with GAS; see the file COPYING. If not, write to the Free
19 Software Foundation, 51 Franklin Street - Fifth Floor, Boston, MA
22 /*======================================================================*/
24 * Herein lies the support for dynamic specification of processor
25 * instructions and registers. Mnemonics, values, and formats for each
26 * instruction and register are specified in an ascii file consisting of
27 * table entries. The grammar for the table is defined in the document
28 * "Processor instruction table specification".
30 * Instructions use the gnu assembler syntax, with the addition of
31 * allowing mnemonics for register.
32 * Eg. "func $2,reg3,0x100,symbol ; comment"
35 * reg3 - mnemonic for processor's register defined in table
36 * 0xddd..d - immediate value
37 * symbol - address of label or external symbol
39 * First, itbl_parse reads in the table of register and instruction
40 * names and formats, and builds a list of entries for each
41 * processor/type combination. lex and yacc are used to parse
42 * the entries in the table and call functions defined here to
43 * add each entry to our list.
45 * Then, when assembling or disassembling, these functions are called to
46 * 1) get information on a processor's registers and
47 * 2) assemble/disassemble an instruction.
48 * To assemble(disassemble) an instruction, the function
49 * itbl_assemble(itbl_disassemble) is called to search the list of
50 * instruction entries, and if a match is found, uses the format
51 * described in the instruction entry structure to complete the action.
53 * Eg. Suppose we have a Mips coprocessor "cop3" with data register "d2"
54 * and we want to define function "pig" which takes two operands.
56 * Given the table entries:
57 * "p3 insn pig 0x1:24-21 dreg:20-16 immed:15-0"
59 * and that the instruction encoding for coprocessor pz has encoding:
60 * #define MIPS_ENCODE_COP_NUM(z) ((0x21|(z<<1))<<25)
61 * #define ITBL_ENCODE_PNUM(pnum) MIPS_ENCODE_COP_NUM(pnum)
63 * a structure to describe the instruction might look something like:
64 * struct itbl_entry = {
65 * e_processor processor = e_p3
66 * e_type type = e_insn
70 * struct itbl_range range = 24-21
71 * struct itbl_field *field = {
72 * e_type type = e_dreg
73 * struct itbl_range range = 20-16
74 * struct itbl_field *next = {
75 * e_type type = e_immed
76 * struct itbl_range range = 15-0
77 * struct itbl_field *next = 0
80 * struct itbl_entry *next = 0
83 * And the assembler instructions:
87 * would both assemble to the hex value:
96 #include <itbl-parse.h>
102 #define ASSERT(x) assert(x)
103 #define DBG(x) printf x
110 #define min(a,b) (a<b?a:b)
113 int itbl_have_entries
= 0;
115 /*======================================================================*/
116 /* structures for keeping itbl format entries */
119 int sbit
; /* mask starting bit position */
120 int ebit
; /* mask ending bit position */
124 e_type type
; /* dreg/creg/greg/immed/symb */
125 struct itbl_range range
; /* field's bitfield range within instruction */
126 unsigned long flags
; /* field flags */
127 struct itbl_field
*next
; /* next field in list */
130 /* These structures define the instructions and registers for a processor.
131 * If the type is an instruction, the structure defines the format of an
132 * instruction where the fields are the list of operands.
133 * The flags field below uses the same values as those defined in the
134 * gnu assembler and are machine specific. */
136 e_processor processor
; /* processor number */
137 e_type type
; /* dreg/creg/greg/insn */
138 char *name
; /* mnemionic name for insn/register */
139 unsigned long value
; /* opcode/instruction mask/register number */
140 unsigned long flags
; /* effects of the instruction */
141 struct itbl_range range
; /* bit range within instruction for value */
142 struct itbl_field
*fields
; /* list of operand definitions (if any) */
143 struct itbl_entry
*next
; /* next entry */
146 /* local data and structures */
148 static int itbl_num_opcodes
= 0;
149 /* Array of entries for each processor and entry type */
150 static struct itbl_entry
*entries
[e_nprocs
][e_ntypes
] = {
157 /* local prototypes */
158 static unsigned long build_opcode (struct itbl_entry
*e
);
159 static e_type
get_type (int yytype
);
160 static e_processor
get_processor (int yyproc
);
161 static struct itbl_entry
**get_entries (e_processor processor
,
163 static struct itbl_entry
*find_entry_byname (e_processor processor
,
164 e_type type
, char *name
);
165 static struct itbl_entry
*find_entry_byval (e_processor processor
,
166 e_type type
, unsigned long val
, struct itbl_range
*r
);
167 static struct itbl_entry
*alloc_entry (e_processor processor
,
168 e_type type
, char *name
, unsigned long value
);
169 static unsigned long apply_range (unsigned long value
, struct itbl_range r
);
170 static unsigned long extract_range (unsigned long value
, struct itbl_range r
);
171 static struct itbl_field
*alloc_field (e_type type
, int sbit
,
172 int ebit
, unsigned long flags
);
174 /*======================================================================*/
175 /* Interfaces to the parser */
177 /* Open the table and use lex and yacc to parse the entries.
178 * Return 1 for failure; 0 for success. */
181 itbl_parse (char *insntbl
)
184 extern int yyparse (void);
186 yyin
= fopen (insntbl
, FOPEN_RT
);
189 printf ("Can't open processor instruction specification file \"%s\"\n",
198 itbl_have_entries
= 1;
202 /* Add a register entry */
205 itbl_add_reg (int yyprocessor
, int yytype
, char *regname
,
208 return alloc_entry (get_processor (yyprocessor
), get_type (yytype
), regname
,
209 (unsigned long) regnum
);
212 /* Add an instruction entry */
215 itbl_add_insn (int yyprocessor
, char *name
, unsigned long value
,
216 int sbit
, int ebit
, unsigned long flags
)
218 struct itbl_entry
*e
;
219 e
= alloc_entry (get_processor (yyprocessor
), e_insn
, name
, value
);
222 e
->range
.sbit
= sbit
;
223 e
->range
.ebit
= ebit
;
230 /* Add an operand to an instruction entry */
233 itbl_add_operand (struct itbl_entry
*e
, int yytype
, int sbit
,
234 int ebit
, unsigned long flags
)
236 struct itbl_field
*f
, **last_f
;
239 /* Add to end of fields' list. */
240 f
= alloc_field (get_type (yytype
), sbit
, ebit
, flags
);
245 last_f
= &(*last_f
)->next
;
252 /*======================================================================*/
253 /* Interfaces for assembler and disassembler */
258 static void append_insns_as_macros (void);
260 /* Initialize for gas. */
265 struct itbl_entry
*e
, **es
;
269 if (!itbl_have_entries
)
272 /* Since register names don't have a prefix, put them in the symbol table so
273 they can't be used as symbols. This simplifies argument parsing as
274 we can let gas parse registers for us. */
275 /* Use symbol_create instead of symbol_new so we don't try to
276 output registers into the object file's symbol table. */
278 for (type
= e_regtype0
; type
< e_nregtypes
; type
++)
279 for (procn
= e_p0
; procn
< e_nprocs
; procn
++)
281 es
= get_entries (procn
, type
);
282 for (e
= *es
; e
; e
= e
->next
)
284 symbol_table_insert (symbol_create (e
->name
, reg_section
,
285 e
->value
, &zero_address_frag
));
288 append_insns_as_macros ();
291 /* Append insns to opcodes table and increase number of opcodes
292 * Structure of opcodes table:
296 * const char *args; - string describing the arguments.
297 * unsigned long match; - opcode, or ISA level if pinfo=INSN_MACRO
298 * unsigned long mask; - opcode mask, or macro id if pinfo=INSN_MACRO
299 * unsigned long pinfo; - insn flags, or INSN_MACRO
302 * {"li", "t,i", 0x34000000, 0xffe00000, WR_t },
303 * {"li", "t,I", 0, (int) M_LI, INSN_MACRO },
306 static char *form_args (struct itbl_entry
*e
);
308 append_insns_as_macros (void)
310 struct ITBL_OPCODE_STRUCT
*new_opcodes
, *o
;
311 struct itbl_entry
*e
, **es
;
312 int n
, id
, size
, new_size
, new_num_opcodes
;
314 if (!itbl_have_entries
)
317 if (!itbl_num_opcodes
) /* no new instructions to add! */
321 DBG (("previous num_opcodes=%d\n", ITBL_NUM_OPCODES
));
323 new_num_opcodes
= ITBL_NUM_OPCODES
+ itbl_num_opcodes
;
324 ASSERT (new_num_opcodes
>= itbl_num_opcodes
);
326 size
= sizeof (struct ITBL_OPCODE_STRUCT
) * ITBL_NUM_OPCODES
;
328 DBG (("I get=%d\n", size
/ sizeof (ITBL_OPCODES
[0])));
330 new_size
= sizeof (struct ITBL_OPCODE_STRUCT
) * new_num_opcodes
;
331 ASSERT (new_size
> size
);
333 /* FIXME since ITBL_OPCODES culd be a static table,
334 we can't realloc or delete the old memory. */
335 new_opcodes
= (struct ITBL_OPCODE_STRUCT
*) malloc (new_size
);
338 printf (_("Unable to allocate memory for new instructions\n"));
341 if (size
) /* copy preexisting opcodes table */
342 memcpy (new_opcodes
, ITBL_OPCODES
, size
);
344 /* FIXME! some NUMOPCODES are calculated expressions.
345 These need to be changed before itbls can be supported. */
347 id
= ITBL_NUM_MACROS
; /* begin the next macro id after the last */
348 o
= &new_opcodes
[ITBL_NUM_OPCODES
]; /* append macro to opcodes list */
349 for (n
= e_p0
; n
< e_nprocs
; n
++)
351 es
= get_entries (n
, e_insn
);
352 for (e
= *es
; e
; e
= e
->next
)
354 /* name, args, mask, match, pinfo
355 * {"li", "t,i", 0x34000000, 0xffe00000, WR_t },
356 * {"li", "t,I", 0, (int) M_LI, INSN_MACRO },
357 * Construct args from itbl_fields.
360 o
->args
= strdup (form_args (e
));
361 o
->mask
= apply_range (e
->value
, e
->range
);
362 /* FIXME how to catch during assembly? */
363 /* mask to identify this insn */
364 o
->match
= apply_range (e
->value
, e
->range
);
368 o
->mask
= id
++; /* FIXME how to catch during assembly? */
369 o
->match
= 0; /* for macros, the insn_isa number */
370 o
->pinfo
= INSN_MACRO
;
373 /* Don't add instructions which caused an error */
380 ITBL_OPCODES
= new_opcodes
;
381 ITBL_NUM_OPCODES
= new_num_opcodes
;
384 At this point, we can free the entries, as they should have
385 been added to the assembler's tables.
386 Don't free name though, since name is being used by the new
389 Eventually, we should also free the new opcodes table itself
395 form_args (struct itbl_entry
*e
)
399 struct itbl_field
*f
;
402 for (f
= e
->fields
; f
; f
= f
->next
)
422 c
= 0; /* ignore; unknown field type */
434 #endif /* !STAND_ALONE */
436 /* Get processor's register name from val */
439 itbl_get_reg_val (char *name
, unsigned long *pval
)
444 for (p
= e_p0
; p
< e_nprocs
; p
++)
446 for (t
= e_regtype0
; t
< e_nregtypes
; t
++)
448 if (itbl_get_val (p
, t
, name
, pval
))
456 itbl_get_name (e_processor processor
, e_type type
, unsigned long val
)
458 struct itbl_entry
*r
;
459 /* type depends on instruction passed */
460 r
= find_entry_byval (processor
, type
, val
, 0);
464 return 0; /* error; invalid operand */
467 /* Get processor's register value from name */
470 itbl_get_val (e_processor processor
, e_type type
, char *name
,
473 struct itbl_entry
*r
;
474 /* type depends on instruction passed */
475 r
= find_entry_byname (processor
, type
, name
);
482 /* Assemble instruction "name" with operands "s".
483 * name - name of instruction
485 * returns - long word for assembled instruction */
488 itbl_assemble (char *name
, char *s
)
490 unsigned long opcode
;
491 struct itbl_entry
*e
= NULL
;
492 struct itbl_field
*f
;
497 return 0; /* error! must have an opcode name/expr */
499 /* find entry in list of instructions for all processors */
500 for (processor
= 0; processor
< e_nprocs
; processor
++)
502 e
= find_entry_byname (processor
, e_insn
, name
);
507 return 0; /* opcode not in table; invalid instruction */
508 opcode
= build_opcode (e
);
510 /* parse opcode's args (if any) */
511 for (f
= e
->fields
; f
; f
= f
->next
) /* for each arg, ... */
513 struct itbl_entry
*r
;
516 return 0; /* error - not enough operands */
517 n
= itbl_get_field (&s
);
518 /* n should be in form $n or 0xhhh (are symbol names valid?? */
524 /* Accept either a string name
525 * or '$' followed by the register number */
529 value
= strtol (n
, 0, 10);
530 /* FIXME! could have "0l"... then what?? */
531 if (value
== 0 && *n
!= '0')
532 return 0; /* error; invalid operand */
536 r
= find_entry_byname (e
->processor
, f
->type
, n
);
540 return 0; /* error; invalid operand */
544 /* use assembler's symbol table to find symbol */
545 /* FIXME!! Do we need this?
546 if so, what about relocs??
547 my_getExpression (&imm_expr, s);
548 return 0; /-* error; invalid operand *-/
551 /* If not a symbol, fall thru to IMMED */
553 if (*n
== '0' && *(n
+ 1) == 'x') /* hex begins 0x... */
556 value
= strtol (n
, 0, 16);
557 /* FIXME! could have "0xl"... then what?? */
561 value
= strtol (n
, 0, 10);
562 /* FIXME! could have "0l"... then what?? */
563 if (value
== 0 && *n
!= '0')
564 return 0; /* error; invalid operand */
568 return 0; /* error; invalid field spec */
570 opcode
|= apply_range (value
, f
->range
);
573 return 0; /* error - too many operands */
574 return opcode
; /* done! */
577 /* Disassemble instruction "insn".
579 * s - buffer to hold disassembled instruction
580 * returns - 1 if succeeded; 0 if failed
584 itbl_disassemble (char *s
, unsigned long insn
)
586 e_processor processor
;
587 struct itbl_entry
*e
;
588 struct itbl_field
*f
;
590 if (!ITBL_IS_INSN (insn
))
591 return 0; /* error */
592 processor
= get_processor (ITBL_DECODE_PNUM (insn
));
594 /* find entry in list */
595 e
= find_entry_byval (processor
, e_insn
, insn
, 0);
597 return 0; /* opcode not in table; invalid instruction */
600 /* Parse insn's args (if any). */
601 for (f
= e
->fields
; f
; f
= f
->next
) /* for each arg, ... */
603 struct itbl_entry
*r
;
606 if (f
== e
->fields
) /* First operand is preceded by tab. */
608 else /* ','s separate following operands. */
610 value
= extract_range (insn
, f
->range
);
611 /* n should be in form $n or 0xhhh (are symbol names valid?? */
617 /* Accept either a string name
618 or '$' followed by the register number. */
619 r
= find_entry_byval (e
->processor
, f
->type
, value
, &f
->range
);
623 sprintf (s
, "%s$%lu", s
, value
);
626 /* Use assembler's symbol table to find symbol. */
627 /* FIXME!! Do we need this? If so, what about relocs?? */
628 /* If not a symbol, fall through to IMMED. */
630 sprintf (s
, "%s0x%lx", s
, value
);
633 return 0; /* error; invalid field spec */
636 return 1; /* Done! */
639 /*======================================================================*/
641 * Local functions for manipulating private structures containing
642 * the names and format for the new instructions and registers
643 * for each processor.
646 /* Calculate instruction's opcode and function values from entry */
649 build_opcode (struct itbl_entry
*e
)
651 unsigned long opcode
;
653 opcode
= apply_range (e
->value
, e
->range
);
654 opcode
|= ITBL_ENCODE_PNUM (e
->processor
);
658 /* Calculate absolute value given the relative value and bit position range
659 * within the instruction.
660 * The range is inclusive where 0 is least significant bit.
661 * A range of { 24, 20 } will have a mask of
663 * pos: 1098 7654 3210 9876 5432 1098 7654 3210
664 * bin: 0000 0001 1111 0000 0000 0000 0000 0000
665 * hex: 0 1 f 0 0 0 0 0
670 apply_range (unsigned long rval
, struct itbl_range r
)
674 int len
= MAX_BITPOS
- r
.sbit
;
676 ASSERT (r
.sbit
>= r
.ebit
);
677 ASSERT (MAX_BITPOS
>= r
.sbit
);
678 ASSERT (r
.ebit
>= 0);
680 /* create mask by truncating 1s by shifting */
681 mask
= 0xffffffff << len
;
683 mask
= mask
>> r
.ebit
;
684 mask
= mask
<< r
.ebit
;
686 aval
= (rval
<< r
.ebit
) & mask
;
690 /* Calculate relative value given the absolute value and bit position range
691 * within the instruction. */
694 extract_range (unsigned long aval
, struct itbl_range r
)
698 int len
= MAX_BITPOS
- r
.sbit
;
700 /* create mask by truncating 1s by shifting */
701 mask
= 0xffffffff << len
;
703 mask
= mask
>> r
.ebit
;
704 mask
= mask
<< r
.ebit
;
706 rval
= (aval
& mask
) >> r
.ebit
;
710 /* Extract processor's assembly instruction field name from s;
711 * forms are "n args" "n,args" or "n" */
712 /* Return next argument from string pointer "s" and advance s.
713 * delimiters are " ,()" */
716 itbl_get_field (char **S
)
725 /* FIXME: This is a weird set of delimiters. */
726 len
= strcspn (s
, " \t,()");
727 ASSERT (128 > len
+ 1);
731 s
= 0; /* no more args */
733 s
+= len
+ 1; /* advance to next arg */
739 /* Search entries for a given processor and type
740 * to find one matching the name "n".
741 * Return a pointer to the entry */
743 static struct itbl_entry
*
744 find_entry_byname (e_processor processor
,
745 e_type type
, char *n
)
747 struct itbl_entry
*e
, **es
;
749 es
= get_entries (processor
, type
);
750 for (e
= *es
; e
; e
= e
->next
) /* for each entry, ... */
752 if (!strcmp (e
->name
, n
))
758 /* Search entries for a given processor and type
759 * to find one matching the value "val" for the range "r".
760 * Return a pointer to the entry.
761 * This function is used for disassembling fields of an instruction.
764 static struct itbl_entry
*
765 find_entry_byval (e_processor processor
, e_type type
,
766 unsigned long val
, struct itbl_range
*r
)
768 struct itbl_entry
*e
, **es
;
771 es
= get_entries (processor
, type
);
772 for (e
= *es
; e
; e
= e
->next
) /* for each entry, ... */
774 if (processor
!= e
->processor
)
776 /* For insns, we might not know the range of the opcode,
777 * so a range of 0 will allow this routine to match against
778 * the range of the entry to be compared with.
779 * This could cause ambiguities.
780 * For operands, we get an extracted value and a range.
782 /* if range is 0, mask val against the range of the compared entry. */
783 if (r
== 0) /* if no range passed, must be whole 32-bits
784 * so create 32-bit value from entry's range */
786 eval
= apply_range (e
->value
, e
->range
);
787 val
&= apply_range (0xffffffff, e
->range
);
789 else if ((r
->sbit
== e
->range
.sbit
&& r
->ebit
== e
->range
.ebit
)
790 || (e
->range
.sbit
== 0 && e
->range
.ebit
== 0))
792 eval
= apply_range (e
->value
, *r
);
793 val
= apply_range (val
, *r
);
803 /* Return a pointer to the list of entries for a given processor and type. */
805 static struct itbl_entry
**
806 get_entries (e_processor processor
, e_type type
)
808 return &entries
[processor
][type
];
811 /* Return an integral value for the processor passed from yyparse. */
814 get_processor (int yyproc
)
816 /* translate from yacc's processor to enum */
817 if (yyproc
>= e_p0
&& yyproc
< e_nprocs
)
818 return (e_processor
) yyproc
;
819 return e_invproc
; /* error; invalid processor */
822 /* Return an integral value for the entry type passed from yyparse. */
825 get_type (int yytype
)
829 /* translate from yacc's type to enum */
843 return e_invtype
; /* error; invalid type */
847 /* Allocate and initialize an entry */
849 static struct itbl_entry
*
850 alloc_entry (e_processor processor
, e_type type
,
851 char *name
, unsigned long value
)
853 struct itbl_entry
*e
, **es
;
856 e
= (struct itbl_entry
*) malloc (sizeof (struct itbl_entry
));
859 memset (e
, 0, sizeof (struct itbl_entry
));
860 e
->name
= (char *) malloc (sizeof (strlen (name
)) + 1);
862 strcpy (e
->name
, name
);
863 e
->processor
= processor
;
866 es
= get_entries (e
->processor
, e
->type
);
873 /* Allocate and initialize an entry's field */
875 static struct itbl_field
*
876 alloc_field (e_type type
, int sbit
, int ebit
,
879 struct itbl_field
*f
;
880 f
= (struct itbl_field
*) malloc (sizeof (struct itbl_field
));
883 memset (f
, 0, sizeof (struct itbl_field
));
885 f
->range
.sbit
= sbit
;
886 f
->range
.ebit
= ebit
;