2 Copyright 1997, 1999, 2000, 2001, 2002, 2003, 2005, 2006, 2007
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 3, 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:
94 #include <itbl-parse.h>
100 #define ASSERT(x) assert(x)
101 #define DBG(x) printf x
108 #define min(a,b) (a<b?a:b)
111 int itbl_have_entries
= 0;
113 /*======================================================================*/
114 /* structures for keeping itbl format entries */
117 int sbit
; /* mask starting bit position */
118 int ebit
; /* mask ending bit position */
122 e_type type
; /* dreg/creg/greg/immed/symb */
123 struct itbl_range range
; /* field's bitfield range within instruction */
124 unsigned long flags
; /* field flags */
125 struct itbl_field
*next
; /* next field in list */
128 /* These structures define the instructions and registers for a processor.
129 * If the type is an instruction, the structure defines the format of an
130 * instruction where the fields are the list of operands.
131 * The flags field below uses the same values as those defined in the
132 * gnu assembler and are machine specific. */
134 e_processor processor
; /* processor number */
135 e_type type
; /* dreg/creg/greg/insn */
136 char *name
; /* mnemionic name for insn/register */
137 unsigned long value
; /* opcode/instruction mask/register number */
138 unsigned long flags
; /* effects of the instruction */
139 struct itbl_range range
; /* bit range within instruction for value */
140 struct itbl_field
*fields
; /* list of operand definitions (if any) */
141 struct itbl_entry
*next
; /* next entry */
144 /* local data and structures */
146 static int itbl_num_opcodes
= 0;
147 /* Array of entries for each processor and entry type */
148 static struct itbl_entry
*entries
[e_nprocs
][e_ntypes
];
150 /* local prototypes */
151 static unsigned long build_opcode (struct itbl_entry
*e
);
152 static e_type
get_type (int yytype
);
153 static e_processor
get_processor (int yyproc
);
154 static struct itbl_entry
**get_entries (e_processor processor
,
156 static struct itbl_entry
*find_entry_byname (e_processor processor
,
157 e_type type
, char *name
);
158 static struct itbl_entry
*find_entry_byval (e_processor processor
,
159 e_type type
, unsigned long val
, struct itbl_range
*r
);
160 static struct itbl_entry
*alloc_entry (e_processor processor
,
161 e_type type
, char *name
, unsigned long value
);
162 static unsigned long apply_range (unsigned long value
, struct itbl_range r
);
163 static unsigned long extract_range (unsigned long value
, struct itbl_range r
);
164 static struct itbl_field
*alloc_field (e_type type
, int sbit
,
165 int ebit
, unsigned long flags
);
167 /*======================================================================*/
168 /* Interfaces to the parser */
170 /* Open the table and use lex and yacc to parse the entries.
171 * Return 1 for failure; 0 for success. */
174 itbl_parse (char *insntbl
)
177 extern int yyparse (void);
179 yyin
= fopen (insntbl
, FOPEN_RT
);
182 printf ("Can't open processor instruction specification file \"%s\"\n",
191 itbl_have_entries
= 1;
195 /* Add a register entry */
198 itbl_add_reg (int yyprocessor
, int yytype
, char *regname
,
201 return alloc_entry (get_processor (yyprocessor
), get_type (yytype
), regname
,
202 (unsigned long) regnum
);
205 /* Add an instruction entry */
208 itbl_add_insn (int yyprocessor
, char *name
, unsigned long value
,
209 int sbit
, int ebit
, unsigned long flags
)
211 struct itbl_entry
*e
;
212 e
= alloc_entry (get_processor (yyprocessor
), e_insn
, name
, value
);
215 e
->range
.sbit
= sbit
;
216 e
->range
.ebit
= ebit
;
223 /* Add an operand to an instruction entry */
226 itbl_add_operand (struct itbl_entry
*e
, int yytype
, int sbit
,
227 int ebit
, unsigned long flags
)
229 struct itbl_field
*f
, **last_f
;
232 /* Add to end of fields' list. */
233 f
= alloc_field (get_type (yytype
), sbit
, ebit
, flags
);
238 last_f
= &(*last_f
)->next
;
245 /*======================================================================*/
246 /* Interfaces for assembler and disassembler */
249 static void append_insns_as_macros (void);
251 /* Initialize for gas. */
256 struct itbl_entry
*e
, **es
;
260 if (!itbl_have_entries
)
263 /* Since register names don't have a prefix, put them in the symbol table so
264 they can't be used as symbols. This simplifies argument parsing as
265 we can let gas parse registers for us. */
266 /* Use symbol_create instead of symbol_new so we don't try to
267 output registers into the object file's symbol table. */
269 for (type
= e_regtype0
; type
< e_nregtypes
; type
++)
270 for (procn
= e_p0
; procn
< e_nprocs
; procn
++)
272 es
= get_entries (procn
, type
);
273 for (e
= *es
; e
; e
= e
->next
)
275 symbol_table_insert (symbol_create (e
->name
, reg_section
,
276 e
->value
, &zero_address_frag
));
279 append_insns_as_macros ();
282 /* Append insns to opcodes table and increase number of opcodes
283 * Structure of opcodes table:
287 * const char *args; - string describing the arguments.
288 * unsigned long match; - opcode, or ISA level if pinfo=INSN_MACRO
289 * unsigned long mask; - opcode mask, or macro id if pinfo=INSN_MACRO
290 * unsigned long pinfo; - insn flags, or INSN_MACRO
293 * {"li", "t,i", 0x34000000, 0xffe00000, WR_t },
294 * {"li", "t,I", 0, (int) M_LI, INSN_MACRO },
297 static char *form_args (struct itbl_entry
*e
);
299 append_insns_as_macros (void)
301 struct ITBL_OPCODE_STRUCT
*new_opcodes
, *o
;
302 struct itbl_entry
*e
, **es
;
303 int n
, id
, size
, new_size
, new_num_opcodes
;
305 if (!itbl_have_entries
)
308 if (!itbl_num_opcodes
) /* no new instructions to add! */
312 DBG (("previous num_opcodes=%d\n", ITBL_NUM_OPCODES
));
314 new_num_opcodes
= ITBL_NUM_OPCODES
+ itbl_num_opcodes
;
315 ASSERT (new_num_opcodes
>= itbl_num_opcodes
);
317 size
= sizeof (struct ITBL_OPCODE_STRUCT
) * ITBL_NUM_OPCODES
;
319 DBG (("I get=%d\n", size
/ sizeof (ITBL_OPCODES
[0])));
321 new_size
= sizeof (struct ITBL_OPCODE_STRUCT
) * new_num_opcodes
;
322 ASSERT (new_size
> size
);
324 /* FIXME since ITBL_OPCODES culd be a static table,
325 we can't realloc or delete the old memory. */
326 new_opcodes
= (struct ITBL_OPCODE_STRUCT
*) malloc (new_size
);
329 printf (_("Unable to allocate memory for new instructions\n"));
332 if (size
) /* copy preexisting opcodes table */
333 memcpy (new_opcodes
, ITBL_OPCODES
, size
);
335 /* FIXME! some NUMOPCODES are calculated expressions.
336 These need to be changed before itbls can be supported. */
338 id
= ITBL_NUM_MACROS
; /* begin the next macro id after the last */
339 o
= &new_opcodes
[ITBL_NUM_OPCODES
]; /* append macro to opcodes list */
340 for (n
= e_p0
; n
< e_nprocs
; n
++)
342 es
= get_entries (n
, e_insn
);
343 for (e
= *es
; e
; e
= e
->next
)
345 /* name, args, mask, match, pinfo
346 * {"li", "t,i", 0x34000000, 0xffe00000, WR_t },
347 * {"li", "t,I", 0, (int) M_LI, INSN_MACRO },
348 * Construct args from itbl_fields.
351 o
->args
= strdup (form_args (e
));
352 o
->mask
= apply_range (e
->value
, e
->range
);
353 /* FIXME how to catch during assembly? */
354 /* mask to identify this insn */
355 o
->match
= apply_range (e
->value
, e
->range
);
359 o
->mask
= id
++; /* FIXME how to catch during assembly? */
360 o
->match
= 0; /* for macros, the insn_isa number */
361 o
->pinfo
= INSN_MACRO
;
364 /* Don't add instructions which caused an error */
371 ITBL_OPCODES
= new_opcodes
;
372 ITBL_NUM_OPCODES
= new_num_opcodes
;
375 At this point, we can free the entries, as they should have
376 been added to the assembler's tables.
377 Don't free name though, since name is being used by the new
380 Eventually, we should also free the new opcodes table itself
386 form_args (struct itbl_entry
*e
)
390 struct itbl_field
*f
;
393 for (f
= e
->fields
; f
; f
= f
->next
)
413 c
= 0; /* ignore; unknown field type */
425 #endif /* !STAND_ALONE */
427 /* Get processor's register name from val */
430 itbl_get_reg_val (char *name
, unsigned long *pval
)
435 for (p
= e_p0
; p
< e_nprocs
; p
++)
437 for (t
= e_regtype0
; t
< e_nregtypes
; t
++)
439 if (itbl_get_val (p
, t
, name
, pval
))
447 itbl_get_name (e_processor processor
, e_type type
, unsigned long val
)
449 struct itbl_entry
*r
;
450 /* type depends on instruction passed */
451 r
= find_entry_byval (processor
, type
, val
, 0);
455 return 0; /* error; invalid operand */
458 /* Get processor's register value from name */
461 itbl_get_val (e_processor processor
, e_type type
, char *name
,
464 struct itbl_entry
*r
;
465 /* type depends on instruction passed */
466 r
= find_entry_byname (processor
, type
, name
);
473 /* Assemble instruction "name" with operands "s".
474 * name - name of instruction
476 * returns - long word for assembled instruction */
479 itbl_assemble (char *name
, char *s
)
481 unsigned long opcode
;
482 struct itbl_entry
*e
= NULL
;
483 struct itbl_field
*f
;
488 return 0; /* error! must have an opcode name/expr */
490 /* find entry in list of instructions for all processors */
491 for (processor
= 0; processor
< e_nprocs
; processor
++)
493 e
= find_entry_byname (processor
, e_insn
, name
);
498 return 0; /* opcode not in table; invalid instruction */
499 opcode
= build_opcode (e
);
501 /* parse opcode's args (if any) */
502 for (f
= e
->fields
; f
; f
= f
->next
) /* for each arg, ... */
504 struct itbl_entry
*r
;
507 return 0; /* error - not enough operands */
508 n
= itbl_get_field (&s
);
509 /* n should be in form $n or 0xhhh (are symbol names valid?? */
515 /* Accept either a string name
516 * or '$' followed by the register number */
520 value
= strtol (n
, 0, 10);
521 /* FIXME! could have "0l"... then what?? */
522 if (value
== 0 && *n
!= '0')
523 return 0; /* error; invalid operand */
527 r
= find_entry_byname (e
->processor
, f
->type
, n
);
531 return 0; /* error; invalid operand */
535 /* use assembler's symbol table to find symbol */
536 /* FIXME!! Do we need this?
537 if so, what about relocs??
538 my_getExpression (&imm_expr, s);
539 return 0; /-* error; invalid operand *-/
542 /* If not a symbol, fall thru to IMMED */
544 if (*n
== '0' && *(n
+ 1) == 'x') /* hex begins 0x... */
547 value
= strtol (n
, 0, 16);
548 /* FIXME! could have "0xl"... then what?? */
552 value
= strtol (n
, 0, 10);
553 /* FIXME! could have "0l"... then what?? */
554 if (value
== 0 && *n
!= '0')
555 return 0; /* error; invalid operand */
559 return 0; /* error; invalid field spec */
561 opcode
|= apply_range (value
, f
->range
);
564 return 0; /* error - too many operands */
565 return opcode
; /* done! */
568 /* Disassemble instruction "insn".
570 * s - buffer to hold disassembled instruction
571 * returns - 1 if succeeded; 0 if failed
575 itbl_disassemble (char *s
, unsigned long insn
)
577 e_processor processor
;
578 struct itbl_entry
*e
;
579 struct itbl_field
*f
;
581 if (!ITBL_IS_INSN (insn
))
582 return 0; /* error */
583 processor
= get_processor (ITBL_DECODE_PNUM (insn
));
585 /* find entry in list */
586 e
= find_entry_byval (processor
, e_insn
, insn
, 0);
588 return 0; /* opcode not in table; invalid instruction */
591 /* Parse insn's args (if any). */
592 for (f
= e
->fields
; f
; f
= f
->next
) /* for each arg, ... */
594 struct itbl_entry
*r
;
597 if (f
== e
->fields
) /* First operand is preceded by tab. */
599 else /* ','s separate following operands. */
601 value
= extract_range (insn
, f
->range
);
602 /* n should be in form $n or 0xhhh (are symbol names valid?? */
608 /* Accept either a string name
609 or '$' followed by the register number. */
610 r
= find_entry_byval (e
->processor
, f
->type
, value
, &f
->range
);
614 sprintf (s
, "%s$%lu", s
, value
);
617 /* Use assembler's symbol table to find symbol. */
618 /* FIXME!! Do we need this? If so, what about relocs?? */
619 /* If not a symbol, fall through to IMMED. */
621 sprintf (s
, "%s0x%lx", s
, value
);
624 return 0; /* error; invalid field spec */
627 return 1; /* Done! */
630 /*======================================================================*/
632 * Local functions for manipulating private structures containing
633 * the names and format for the new instructions and registers
634 * for each processor.
637 /* Calculate instruction's opcode and function values from entry */
640 build_opcode (struct itbl_entry
*e
)
642 unsigned long opcode
;
644 opcode
= apply_range (e
->value
, e
->range
);
645 opcode
|= ITBL_ENCODE_PNUM (e
->processor
);
649 /* Calculate absolute value given the relative value and bit position range
650 * within the instruction.
651 * The range is inclusive where 0 is least significant bit.
652 * A range of { 24, 20 } will have a mask of
654 * pos: 1098 7654 3210 9876 5432 1098 7654 3210
655 * bin: 0000 0001 1111 0000 0000 0000 0000 0000
656 * hex: 0 1 f 0 0 0 0 0
661 apply_range (unsigned long rval
, struct itbl_range r
)
665 int len
= MAX_BITPOS
- r
.sbit
;
667 ASSERT (r
.sbit
>= r
.ebit
);
668 ASSERT (MAX_BITPOS
>= r
.sbit
);
669 ASSERT (r
.ebit
>= 0);
671 /* create mask by truncating 1s by shifting */
672 mask
= 0xffffffff << len
;
674 mask
= mask
>> r
.ebit
;
675 mask
= mask
<< r
.ebit
;
677 aval
= (rval
<< r
.ebit
) & mask
;
681 /* Calculate relative value given the absolute value and bit position range
682 * within the instruction. */
685 extract_range (unsigned long aval
, struct itbl_range r
)
689 int len
= MAX_BITPOS
- r
.sbit
;
691 /* create mask by truncating 1s by shifting */
692 mask
= 0xffffffff << len
;
694 mask
= mask
>> r
.ebit
;
695 mask
= mask
<< r
.ebit
;
697 rval
= (aval
& mask
) >> r
.ebit
;
701 /* Extract processor's assembly instruction field name from s;
702 * forms are "n args" "n,args" or "n" */
703 /* Return next argument from string pointer "s" and advance s.
704 * delimiters are " ,()" */
707 itbl_get_field (char **S
)
716 /* FIXME: This is a weird set of delimiters. */
717 len
= strcspn (s
, " \t,()");
718 ASSERT (128 > len
+ 1);
722 s
= 0; /* no more args */
724 s
+= len
+ 1; /* advance to next arg */
730 /* Search entries for a given processor and type
731 * to find one matching the name "n".
732 * Return a pointer to the entry */
734 static struct itbl_entry
*
735 find_entry_byname (e_processor processor
,
736 e_type type
, char *n
)
738 struct itbl_entry
*e
, **es
;
740 es
= get_entries (processor
, type
);
741 for (e
= *es
; e
; e
= e
->next
) /* for each entry, ... */
743 if (!strcmp (e
->name
, n
))
749 /* Search entries for a given processor and type
750 * to find one matching the value "val" for the range "r".
751 * Return a pointer to the entry.
752 * This function is used for disassembling fields of an instruction.
755 static struct itbl_entry
*
756 find_entry_byval (e_processor processor
, e_type type
,
757 unsigned long val
, struct itbl_range
*r
)
759 struct itbl_entry
*e
, **es
;
762 es
= get_entries (processor
, type
);
763 for (e
= *es
; e
; e
= e
->next
) /* for each entry, ... */
765 if (processor
!= e
->processor
)
767 /* For insns, we might not know the range of the opcode,
768 * so a range of 0 will allow this routine to match against
769 * the range of the entry to be compared with.
770 * This could cause ambiguities.
771 * For operands, we get an extracted value and a range.
773 /* if range is 0, mask val against the range of the compared entry. */
774 if (r
== 0) /* if no range passed, must be whole 32-bits
775 * so create 32-bit value from entry's range */
777 eval
= apply_range (e
->value
, e
->range
);
778 val
&= apply_range (0xffffffff, e
->range
);
780 else if ((r
->sbit
== e
->range
.sbit
&& r
->ebit
== e
->range
.ebit
)
781 || (e
->range
.sbit
== 0 && e
->range
.ebit
== 0))
783 eval
= apply_range (e
->value
, *r
);
784 val
= apply_range (val
, *r
);
794 /* Return a pointer to the list of entries for a given processor and type. */
796 static struct itbl_entry
**
797 get_entries (e_processor processor
, e_type type
)
799 return &entries
[processor
][type
];
802 /* Return an integral value for the processor passed from yyparse. */
805 get_processor (int yyproc
)
807 /* translate from yacc's processor to enum */
808 if (yyproc
>= e_p0
&& yyproc
< e_nprocs
)
809 return (e_processor
) yyproc
;
810 return e_invproc
; /* error; invalid processor */
813 /* Return an integral value for the entry type passed from yyparse. */
816 get_type (int yytype
)
820 /* translate from yacc's type to enum */
834 return e_invtype
; /* error; invalid type */
838 /* Allocate and initialize an entry */
840 static struct itbl_entry
*
841 alloc_entry (e_processor processor
, e_type type
,
842 char *name
, unsigned long value
)
844 struct itbl_entry
*e
, **es
;
847 e
= (struct itbl_entry
*) malloc (sizeof (struct itbl_entry
));
850 memset (e
, 0, sizeof (struct itbl_entry
));
851 e
->name
= (char *) malloc (sizeof (strlen (name
)) + 1);
853 strcpy (e
->name
, name
);
854 e
->processor
= processor
;
857 es
= get_entries (e
->processor
, e
->type
);
864 /* Allocate and initialize an entry's field */
866 static struct itbl_field
*
867 alloc_field (e_type type
, int sbit
, int ebit
,
870 struct itbl_field
*f
;
871 f
= (struct itbl_field
*) malloc (sizeof (struct itbl_field
));
874 memset (f
, 0, sizeof (struct itbl_field
));
876 f
->range
.sbit
= sbit
;
877 f
->range
.ebit
= ebit
;