* itbl-ops.c: New file. Add support for dynamically read

instruction registers, opcodes and formats.  Build internal table
	for new instructions and provide callbacks for assembler and
	disassembler.
	* itbl-lex.l, itbl-parse.y: Lex and yacc parsers for instruction
	spec table.
	* itbl-ops.h: New file.  Header file for itbl support.
	* config/itbl-mips.h: New file.  Mips specific definitions for
	itbl support.
This commit is contained in:
Dawn Perchik 1997-02-11 02:07:57 +00:00
parent 242ce3d857
commit 8e5c905e99
8 changed files with 1456 additions and 0 deletions

View File

@ -73,6 +73,10 @@ link.cmd
listing.c
listing.h
literal.c
itbl-lex.l
itbl-ops.c
itbl-ops.h
itbl-parse.y
mac-as.r
macro.c
macro.h

View File

@ -1,3 +1,15 @@
Mon Feb 10 18:09:00 1997 Dawn Perchik (dawn@cygnus.com)
* itbl-ops.c: New file. Add support for dynamically read
instruction registers, opcodes and formats. Build internal table
for new instructions and provide callbacks for assembler and
disassembler.
* itbl-lex.l, itbl-parse.y: Lex and yacc parsers for instruction
spec table.
* itbl-ops.h: New file. Header file for itbl support.
* config/itbl-mips.h: New file. Mips specific definitions for
itbl support.
Fri Feb 7 09:52:34 1997 Jeffrey A Law (law@cygnus.com)
* config/tc-mn10200.c (md_assemble): If a constant operand won't

View File

@ -65,6 +65,7 @@ go32.cfg
e-mipsecoff.c
e-mipself.c
i386coff.mt
itbl-mips.h
m68k-parse.h
m68k-parse.y
m68kcoff.mt

27
gas/config/itbl-mips.h Normal file
View File

@ -0,0 +1,27 @@
/* Defines for Mips itbl cop support */
#include "opcode/mips.h"
/* Values for processors will be from 0 to NUMBER_OF_PROCESSORS-1 */
#define NUMBER_OF_PROCESSORS 4
#define MAX_BITPOS 31
/* Mips specifics */
#define MIPS_OPCODE_COP0 (0x21) /* COPz+CO, bits 31-25: 0100zz1 */
#define MIPS_ENCODE_COP_NUM(z) ((MIPS_OPCODE_COP0|z<<1)<<25)
#define MIPS_IS_COP_INSN(insn) ((MIPS_OPCODE_COP0&(insn>>25)) \
== MIPS_OPCODE_COP0)
#define MIPS_DECODE_COP_NUM(insn) ((~MIPS_OPCODE_COP0&(insn>>25))>>1)
#define MIPS_DECODE_COP_COFUN(insn) ((~MIPS_ENCODE_COP_NUM(3))&(insn))
/* definitions required by generic code */
#define ITBL_IS_INSN(insn) MIPS_IS_COP_INSN(insn)
#define ITBL_DECODE_PNUM(insn) MIPS_DECODE_COP_NUM(insn)
#define ITBL_ENCODE_PNUM(pnum) MIPS_ENCODE_COP_NUM(pnum)
#define ITBL_OPCODE_STRUCT mips_opcode
#define ITBL_OPCODES mips_opcodes
#define ITBL_NUM_OPCODES NUMOPCODES
#define ITBL_NUM_MACROS M_NUM_MACROS

86
gas/itbl-lex.l Normal file
View File

@ -0,0 +1,86 @@
%{
#include <stdio.h>
#include <string.h>
#include <stdlib.h>
#include <ctype.h>
#include "itbl-parse.h"
#ifdef DEBUG
#define DBG(x) printf x
#define MDBG(x) printf x
#else
#define DBG(x)
#define MDBG(x)
#endif
int insntbl_line = 1;
%}
ALNUM [A-Za-z0-9_]
DIGIT [0-9]
ALPHA [A-Za-z_]
HEX [0-9A-Fa-f]
%%
"creg"|"CREG" {
return CREG;
}
"dreg"|"DREG" {
return DREG;
}
"greg"|"GREG" {
return GREG;
}
"immed"|"IMMED" {
return IMMED;
}
"addr"|"ADDR" {
return ADDR;
}
"insn"|"INSN" {
return INSN;
}
"p"{DIGIT} {
yytext[yyleng]=0;
yylval.processor = strtoul(yytext+1,0,0);
return PNUM;
}
{DIGIT}+ {
yytext[yyleng]=0;
yylval.num = strtoul(yytext,0,0);
return NUM;
}
"0x"{HEX}+ {
yytext[yyleng]=0;
yylval.num = strtoul(yytext,0,0);
return NUM;
}
{ALPHA}{ALNUM}* {
yytext[yyleng]=0;
yylval.str = strdup(yytext);
return ID;
}
";"|"#" {
int c;
while ((c = input()) != EOF) {
if (c == '\n')
{
unput(c);
break;
}
}
}
"\n" {
insntbl_line++;
MDBG(("in lex, NL=%d (x%x)\n",NL,NL));
return NL;
}
" "|"\t" { }
. {
MDBG(("char=%x,%d\n",yytext[0],yytext[0]));
return yytext[0];
}
%%
int yywrap() { return 1; }

838
gas/itbl-ops.c Normal file
View File

@ -0,0 +1,838 @@
/*======================================================================*/
/*
* Herein lies the support for dynamic specification of processor
* instructions and registers. Mnemonics, values, and formats for each
* instruction and register are specified in an ascii file consisting of
* table entries. The grammar for the table is defined in the document
* "Processor instruction table specification".
*
* Instructions use the gnu assembler syntax, with the addition of
* allowing mnemonics for register.
* Eg. "func $2,reg3,0x100,symbol ; comment"
* func - opcode name
* $n - register n
* reg3 - mnemonic for processor's register defined in table
* 0xddd..d - immediate value
* symbol - address of label or external symbol
*
* First, itbl_parse reads in the table of register and instruction
* names and formats, and builds a list of entries for each
* processor/type combination. lex and yacc are used to parse
* the entries in the table and call functions defined here to
* add each entry to our list.
*
* Then, when assembling or disassembling, these functions are called to
* 1) get information on a processor's registers and
* 2) assemble/disassemble an instruction.
* To assemble(disassemble) an instruction, the function
* itbl_assemble(itbl_disassemble) is called to search the list of
* instruction entries, and if a match is found, uses the format
* described in the instruction entry structure to complete the action.
*
* Eg. Suppose we have a Mips coprocessor "cop3" with data register "d2"
* and we want to define function "pig" which takes two operands.
*
* Given the table entries:
* "p3 insn pig 0x1:24-21 dreg:20-16 immed:15-0"
* "p3 dreg d2 0x2"
* and that the instruction encoding for coprocessor pz has encoding:
* #define MIPS_ENCODE_COP_NUM(z) ((0x21|(z<<1))<<25)
* #define ITBL_ENCODE_PNUM(pnum) MIPS_ENCODE_COP_NUM(pnum)
*
* a structure to describe the instruction might look something like:
* struct itbl_entry = {
* e_processor processor = e_p3
* e_type type = e_insn
* char *name = "pig"
* uint value = 0x1
* uint flags = 0
* struct itbl_range range = 24-21
* struct itbl_field *field = {
* e_type type = e_dreg
* struct itbl_range range = 20-16
* struct itbl_field *next = {
* e_type type = e_immed
* struct itbl_range range = 15-0
* struct itbl_field *next = 0
* };
* };
* struct itbl_entry *next = 0
* };
*
* And the assembler instructions:
* "pig d2,0x100"
* "pig $2,0x100"
*
* would both assemble to the hex value:
* "0x4e220100"
*
*/
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include "itbl-ops.h"
#include "itbl-parse.h"
#define DEBUG
#ifdef DEBUG
#include <assert.h>
#define ASSERT(x) assert(x)
#define DBG(x) printf x
#else
#define ASSERT(x)
#define DBG(x)
#endif
#ifndef min
#define min(a,b) (a<b?a:b)
#endif
/*======================================================================*/
/* structures for keeping itbl format entries */
struct itbl_range {
int sbit; /* mask starting bit position */
int ebit; /* mask ending bit position */
};
struct itbl_field {
e_type type; /* dreg/creg/greg/immed/symb */
struct itbl_range range; /* field's bitfield range within instruction */
unsigned long flags; /* field flags */
struct itbl_field *next; /* next field in list */
};
/* These structures define the instructions and registers for a processor.
* If the type is an instruction, the structure defines the format of an
* instruction where the fields are the list of operands.
* The flags field below uses the same values as those defined in the
* gnu assembler and are machine specific. */
struct itbl_entry {
e_processor processor; /* processor number */
e_type type; /* dreg/creg/greg/insn */
char *name; /* mnemionic name for insn/register */
unsigned long value; /* opcode/instruction mask/register number */
unsigned long flags; /* effects of the instruction */
struct itbl_range range;/* bit range within instruction for value */
struct itbl_field *fields; /* list of operand definitions (if any) */
struct itbl_entry *next; /* next entry */
};
/* local data and structures */
static int itbl_num_opcodes = 0;
/* Array of entries for each processor and entry type */
static struct itbl_entry *entries[e_nprocs][e_ntypes] =
{
{ 0, 0, 0, 0, 0, 0 },
{ 0, 0, 0, 0, 0, 0 },
{ 0, 0, 0, 0, 0, 0 },
{ 0, 0, 0, 0, 0, 0 }
};
/* local prototypes */
static unsigned long build_opcode(struct itbl_entry *e);
static e_type get_type(int yytype);
static e_processor get_processor(int yyproc);
static struct itbl_entry **get_entries(e_processor processor, e_type type);
static struct itbl_entry *find_entry_byname(e_processor processor, e_type type,
char *name);
static struct itbl_entry *find_entry_byval(e_processor processor, e_type type,
unsigned long val, struct itbl_range *r);
static struct itbl_entry *alloc_entry(e_processor processor, e_type type, char *name,
unsigned long value);
static unsigned long apply_range(unsigned long value, struct itbl_range r);
static unsigned long extract_range(unsigned long value, struct itbl_range r);
static struct itbl_field *alloc_field(e_type type, int sbit, int ebit, unsigned long flags);
/*======================================================================*/
/* Interfaces to the parser */
/* Open the table and use lex and yacc to parse the entries.
* Return 1 for failure; 0 for success. */
int itbl_parse(char* insntbl)
{
extern FILE *yyin;
extern int yyparse(void);
yyin = fopen(insntbl, "r");
if (yyin == 0)
{
printf("Can't open processor instruction specification file \"%s\"\n",
insntbl);
return 1;
}
else
{
while (yyparse());
}
fclose(yyin);
return 0;
}
/* Add a register entry */
struct itbl_entry *itbl_add_reg(int yyprocessor, int yytype, char *regname,
int regnum)
{
#if 0 /* ndef STAND_ALONE */
#include "as.h"
#include "symbols.h"
/* Since register names don't have a prefix, we put them in the symbol table so
they can't be used as symbols. This also simplifies argument parsing as
we can let gas parse registers for us. The recorded register number is
regnum. */
/* Use symbol_create here instead of symbol_new so we don't try to
output registers into the object file's symbol table. */
symbol_table_insert (symbol_create (regname, reg_section,
regnum, &zero_address_frag));
#endif
return alloc_entry(get_processor(yyprocessor),get_type(yytype),regname,
(unsigned long)regnum);
}
/* Add an instruction entry */
struct itbl_entry *itbl_add_insn(int yyprocessor, char *name, unsigned long value,
int sbit, int ebit, unsigned long flags)
{
struct itbl_entry *e;
e = alloc_entry(get_processor(yyprocessor),e_insn,name,value);
if (e)
{
e->range.sbit=sbit;
e->range.ebit=ebit;
e->flags=flags;
itbl_num_opcodes++;
}
return e;
}
/* Add an operand to an instruction entry */
struct itbl_field *itbl_add_operand(struct itbl_entry *e, int yytype, int sbit,
int ebit, unsigned long flags)
{
struct itbl_field *f, **last_f;
if (!e)
return 0;
/* Add to end of fields' list. */
f = alloc_field(get_type(yytype),sbit,ebit,flags);
if (f)
{
last_f = &e->fields;
while (*last_f)
last_f = &(*last_f)->next;
*last_f = f;
f->next = 0;
}
return f;
}
/*======================================================================*/
/* Interfaces for assembler and disassembler */
#ifndef STAND_ALONE
#include "as.h"
#include "symbols.h"
static void append_insns_as_macros(void);
/* initialize for gas */
void itbl_init(void)
{
struct itbl_entry *e, **es;
e_processor procn;
e_type type;
/* Since register names don't have a prefix, put them in the symbol table so
they can't be used as symbols. This simplifies argument parsing as
we can let gas parse registers for us. */
/* Use symbol_create instead of symbol_new so we don't try to
output registers into the object file's symbol table. */
for (type=e_regtype0; type<e_nregtypes; type++)
for (procn=e_p0; procn<e_nprocs; procn++)
{
es = get_entries(procn, type);
for (e=*es; e; e=e->next)
{
symbol_table_insert (symbol_create (e->name, reg_section,
e->value, &zero_address_frag));
}
}
append_insns_as_macros();
}
/* Append insns to opcodes table and increase number of opcodes */
/* Structure of opcodes table: */
/* struct itbl_opcode
/* {
/* const char *name;
/* const char *args; /* string describing the arguments . */
/* unsigned long match; /* opcode, or ISA level if pinfo=INSN_MACRO */
/* unsigned long mask; /* opcode mask, or macro id if pinfo=INSN_MACRO */
/* unsigned long pinfo; /* insn flags, or INSN_MACRO */
/* };
/* examples:
* {"li", "t,i", 0x34000000, 0xffe00000, WR_t },
* {"li", "t,I", 0, (int) M_LI, INSN_MACRO },
*/
static char *form_args(struct itbl_entry *e);
static void append_insns_as_macros(void)
{
struct ITBL_OPCODE_STRUCT *new_opcodes, *o;
struct itbl_entry *e, **es;
int n, id, size, new_size, new_num_opcodes;
ASSERT(itbl_num_opcodes > 0);
if (!itbl_num_opcodes) /* no new instructions to add! */
{
return;
}
DBG(("previous num_opcodes=%d\n",ITBL_NUM_OPCODES));
new_num_opcodes = ITBL_NUM_OPCODES + itbl_num_opcodes;
ASSERT(new_num_opcodes >= itbl_num_opcodes);
size = sizeof(struct ITBL_OPCODE_STRUCT) * ITBL_NUM_OPCODES;
ASSERT(size >= 0);
DBG(("I get=%d\n", size / sizeof(ITBL_OPCODES[0])));
new_size = sizeof(struct ITBL_OPCODE_STRUCT) * new_num_opcodes;
ASSERT(new_size > size);
/* FIXME since ITBL_OPCODES culd be a static table,
we can't realloc or delete the old memory. */
new_opcodes = (struct ITBL_OPCODE_STRUCT*)malloc(new_size);
if (!new_opcodes)
{
printf("Unable to allocate memory for new instructions\n");
return;
}
if (size) /* copy prexisting opcodes table */
memcpy(new_opcodes, ITBL_OPCODES, size);
/* FIXME! some NUMOPCODES are calculated expressions.
These need to be changed before itbls can be supported. */
id = ITBL_NUM_MACROS; /* begin the next macro id after the last */
o = &new_opcodes[ITBL_NUM_OPCODES]; /* append macro to opcodes list */
for (n=e_p0; n<e_nprocs; n++)
{
es = get_entries(n,e_insn);
for (e=*es; e; e=e->next)
{
/* name, args, mask, match, pinfo
* {"li", "t,i", 0x34000000, 0xffe00000, WR_t },
* {"li", "t,I", 0, (int) M_LI, INSN_MACRO },
* Construct args from itbl_fields.
*/
o->name = e->name;
o->args = strdup(form_args(e));
o->mask = apply_range(e->value,e->range);
/* FIXME how to catch durring assembly? */
/* mask to identify this insn */
o->match = apply_range(e->value,e->range);
o->pinfo = 0;
#ifdef USE_MACROS
o->mask = id++; /* FIXME how to catch durring assembly? */
o->match = 0; /* for macros, the insn_isa number */
o->pinfo = INSN_MACRO;
#endif
/* Don't add instructions which caused an error */
if (o->args)
o++;
else
new_num_opcodes--;
}
}
ITBL_OPCODES = new_opcodes;
ITBL_NUM_OPCODES = new_num_opcodes;
/* FIXME
At this point, we can free the entries, as they should have
been added to the assembler's tables.
Don't free name though, since name is being used by the new
opcodes table.
Eventually, we should also free the new opcodes table itself on exit.
*/
}
static char *form_args(struct itbl_entry *e)
{
static char s[31];
char c=0, *p=s;
struct itbl_field *f;
ASSERT(e);
for (f=e->fields; f; f=f->next)
{
switch (f->type)
{
case e_dreg: c='d'; break;
case e_creg: c='t'; break;
case e_greg: c='s'; break;
case e_immed: c='i'; break;
case e_addr: c='a'; break;
default: c=0; /* ignore; unknown field type */
}
if (c)
{
if (p!=s)
*p++=',';
*p++=c;
}
}
*p=0;
return s;
}
#endif /* !STAND_ALONE */
/* Get processor's register name from val */
unsigned long itbl_get_reg_val(char *name)
{
e_type t;
e_processor p;
int r=0;
for (p=e_p0; p<e_nprocs; p++)
for (t=e_regtype0; t<e_nregtypes; t++)
{
if (r = itbl_get_val(p, t, name), r)
return r;
}
return 0;
}
char *itbl_get_name(e_processor processor, e_type type, unsigned long val)
{
struct itbl_entry *r;
/* type depends on instruction passed */
r = find_entry_byval(processor,type,val,0);
if (r)
return r->name;
else
return 0; /* error; invalid operand */
}
/* Get processor's register value from name */
unsigned long itbl_get_val(e_processor processor, e_type type, char *name)
{
struct itbl_entry *r;
/* type depends on instruction passed */
r = find_entry_byname(processor,type,name);
if (r)
return r->value;
else
return 0; /* error; invalid operand */
}
/* Assemble instruction "name" with operands "s".
* name - name of instruction
* s - operands
* returns - long word for assembled instruction */
unsigned long itbl_assemble(char *name, char *s)
{
unsigned long opcode;
struct itbl_entry *e;
struct itbl_field *f;
char *n;
int processor;
if (!name || !*name)
return 0; /* error! must have a opcode name/expr */
/* find entry in list of instructions for all processors */
for (processor=0; processor<e_nprocs; processor++)
{
e = find_entry_byname(processor, e_insn, name);
if (e) break;
}
if (!e)
return 0; /* opcode not in table; invalid instrustion */
opcode = build_opcode(e);
/* parse opcode's args (if any) */
for (f=e->fields; f; f=f->next) /* for each arg, ... */
{
struct itbl_entry *r;
unsigned long value;
if (!s || !*s)
return 0; /* error - not enough operands */
n = itbl_get_field(&s);
/* n should be in form $n or 0xhhh (are symbol names valid?? */
switch (f->type)
{
case e_dreg:
case e_creg:
case e_greg:
/* Accept either a string name
* or '$' followed by the register number */
if (*n == '$')
{
n++;
value = strtol(n,0,10);
/* FIXME! could have "0l"... then what?? */
if (value == 0 && *n!='0')
return 0; /* error; invalid operand */
}
else
{
r = find_entry_byname(e->processor,f->type,n);
if (r)
value = r->value;
else
return 0; /* error; invalid operand */
}
break;
case e_addr:
/* use assembler's symbol table to find symbol */
/* FIXME!! Do we need this?
if so, what about relocs??
my_getExpression (&imm_expr, s);
return 0; /-* error; invalid operand *-/
break;
*/
/* If not a symbol, fall thru to IMMED */
case e_immed:
if (*n=='0' && *(n+1)=='x') /* hex begins 0x... */
{
n+=2;
value = strtol(n,0,16);
/* FIXME! could have "0xl"... then what?? */
}
else
{
value = strtol(n,0,10);
/* FIXME! could have "0l"... then what?? */
if (value == 0 && *n!='0')
return 0; /* error; invalid operand */
}
break;
default:
return 0; /* error; invalid field spec */
}
opcode |= apply_range(value,f->range);
}
if (s && *s)
return 0; /* error - too many operands */
return opcode; /* done! */
}
/* Disassemble instruction "insn".
* insn - instruction
* s - buffer to hold disassembled instruction
* returns - 1 if succeeded; 0 if failed
*/
int itbl_disassemble(char *s, unsigned long insn)
{
e_processor processor;
struct itbl_entry *e;
struct itbl_field *f;
if (!ITBL_IS_INSN(insn))
return 0; /* error*/
processor = get_processor(ITBL_DECODE_PNUM(insn));
/* find entry in list */
e = find_entry_byval(processor, e_insn, insn, 0);
if (!e)
return 0; /* opcode not in table; invalid instrustion */
strcpy(s, e->name);
/* parse insn's args (if any) */
for (f=e->fields; f; f=f->next) /* for each arg, ... */
{
struct itbl_entry *r;
unsigned long value;
if (f==e->fields) /* first operand is preceeded by tab */
strcat(s,"\t");
else /* ','s separate following operands */
strcat(s,",");
value = extract_range(insn, f->range);
/* n should be in form $n or 0xhhh (are symbol names valid?? */
switch (f->type)
{
case e_dreg:
case e_creg:
case e_greg:
/* Accept either a string name
* or '$' followed by the register number */
r = find_entry_byval(e->processor,f->type,value,&f->range);
if (r)
strcat(s,r->name);
else
sprintf(s,"%s$%d",s,value);
break;
case e_addr:
/* use assembler's symbol table to find symbol */
/* FIXME!! Do we need this?
* if so, what about relocs??
*/
/* If not a symbol, fall thru to IMMED */
case e_immed:
sprintf(s,"%s0x%x",s,value);
break;
default:
return 0; /* error; invalid field spec */
}
}
return 1; /* done! */
}
/*======================================================================*/
/*
* Local functions for manipulating private structures containing
* the names and format for the new instructions and registers
* for each processor.
*/
/* Calculate instruction's opcode and function values from entry */
static unsigned long build_opcode(struct itbl_entry *e)
{
unsigned long opcode;
opcode = apply_range(e->value,e->range);
opcode |= ITBL_ENCODE_PNUM(e->processor);
return opcode;
}
/* Calculate absolute value given the relative value and bit position range
* within the instruction.
* The range is inclusive where 0 is least significant bit.
* A range of { 24, 20 } will have a mask of
* bit 3 2 1
* pos: 1098 7654 3210 9876 5432 1098 7654 3210
* bin: 0000 0001 1111 0000 0000 0000 0000 0000
* hex: 0 1 f 0 0 0 0 0
* mask: 0x01f00000.
*/
static unsigned long apply_range(unsigned long rval, struct itbl_range r)
{
unsigned long mask;
unsigned long aval;
int len = MAX_BITPOS - r.sbit;
ASSERT(r.sbit >= r.ebit);
ASSERT(MAX_BITPOS >= r.sbit);
ASSERT(r.ebit >= 0);
/* create mask by truncating 1s by shifting */
mask = 0xffffffff << len;
mask = mask >> len;
mask = mask >> r.ebit;
mask = mask << r.ebit;
aval = (rval << r.ebit) & mask;
return aval;
}
/* Calculate relative value given the absolute value and bit position range
* within the instruction. */
static unsigned long extract_range(unsigned long aval, struct itbl_range r)
{
unsigned long mask;
unsigned long rval;
int len = MAX_BITPOS - r.sbit;
/* create mask by truncating 1s by shifting */
mask = 0xffffffff << len;
mask = mask >> len;
mask = mask >> r.ebit;
mask = mask << r.ebit;
rval = (aval & mask) >> r.ebit;
return rval;
}
/* Extract processor's assembly instruction field name from s;
* forms are "n args" "n,args" or "n" */
/* Return next argument from string pointer "s" and advance s.
* delimiters are " ,\0" */
char *itbl_get_field(char **S)
{
static char n[128];
char *p, *ps, *s;
int len;
s = *S;
if (!s || !*s)
return 0;
p = s+strlen(s);
if (ps=strchr(s,','),ps) p = ps;
if (ps=strchr(s,' '),ps) p = min(p,ps);
if (ps=strchr(s,'\0'),ps) p = min(p,ps);
if (p==0)
return 0; /* error! */
len = p-s;
ASSERT(128>len+1);
strncpy(n,s,len);
n[len]=0;
if (s[len]=='\0') s=0; /* no more args */
else s+=len+1; /* advance to next arg */
*S = s;
return n;
}
/* Search entries for a given processor and type
* to find one matching the name "n".
* Return a pointer to the entry */
static struct itbl_entry *find_entry_byname(e_processor processor,
e_type type, char *n)
{
struct itbl_entry *e, **es;
es = get_entries(processor, type);
for (e=*es; e; e=e->next) /* for each entry, ... */
{
if (!strcmp(e->name,n))
return e;
}
return 0;
}
/* Search entries for a given processor and type
* to find one matching the value "val" for the range "r".
* Return a pointer to the entry.
* This function is used for disassembling fields of an instruction.
*/
static struct itbl_entry *find_entry_byval(e_processor processor, e_type type,
unsigned long val, struct itbl_range *r)
{
struct itbl_entry *e, **es;
unsigned long eval;
es = get_entries(processor, type);
for (e=*es; e; e=e->next) /* for each entry, ... */
{
if (processor != e->processor)
continue;
/* For insns, we might not know the range of the opcode,
* so a range of 0 will allow this routine to match against
* the range of the entry to be compared with.
* This could cause ambiguities.
* For operands, we get an extracted value and a range.
*/
/* if range is 0, mask val against the range of the compared entry. */
if (r==0) /* if no range passed, must be whole 32-bits
* so create 32-bit value from entry's range */
{
eval = apply_range(e->value,e->range);
val &= apply_range(0xffffffff,e->range);
}
else if (r->sbit == e->range.sbit && r->ebit == e->range.ebit
|| e->range.sbit == 0 && e->range.ebit == 0)
{
eval = apply_range(e->value,*r);
val = apply_range(val, *r);
}
else
continue;
if (val==eval)
return e;
}
return 0;
}
/* Return a pointer to the list of entries for a given processor and type. */
static struct itbl_entry **get_entries(e_processor processor, e_type type)
{
return &entries[processor][type];
}
/* Return an integral value for the processor passed from yyparse. */
static e_processor get_processor(int yyproc)
{
/* translate from yacc's processor to enum */
if (yyproc >= e_p0 && yyproc < e_nprocs)
return (e_processor) yyproc;
return e_invproc; /* error; invalid processor */
}
/* Return an integral value for the entry type passed from yyparse. */
static e_type get_type(int yytype)
{
switch(yytype)
{
/* translate from yacc's type to enum */
case INSN: return e_insn;
case DREG: return e_dreg;
case CREG: return e_creg;
case GREG: return e_greg;
case ADDR: return e_addr;
case IMMED: return e_immed;
default:
return e_invtype; /* error; invalid type */
}
}
/* Allocate and initialize an entry */
static struct itbl_entry *alloc_entry(e_processor processor, e_type type,
char *name, unsigned long value)
{
struct itbl_entry *e, **es;
if (!name) return 0;
e = (struct itbl_entry*) malloc(sizeof(struct itbl_entry));
if (e)
{
memset(e,0,sizeof(struct itbl_entry));
e->name = (char *) malloc(sizeof(strlen(name))+1);
if (e->name) strcpy(e->name,name);
e->processor = processor;
e->type = type;
e->value = value;
es = get_entries(e->processor,e->type);
e->next = *es;
*es = e;
}
return e;
}
/* Allocate and initialize an entry's field */
static struct itbl_field *alloc_field(e_type type, int sbit, int ebit,
unsigned long flags)
{
struct itbl_field *f;
f = (struct itbl_field*) malloc(sizeof(struct itbl_field));
if (f)
{
memset(f,0,sizeof(struct itbl_field));
f->type = type;
f->range.sbit = sbit;
f->range.ebit = ebit;
f->flags = flags;
}
return f;
}

75
gas/itbl-ops.h Normal file
View File

@ -0,0 +1,75 @@
/* External functions, constants and defines for itbl support */
#include "itbl-cpu.h"
/* Defaults for definitions required by generic code */
#ifndef ITBL_NUMBER_OF_PROCESSORS
#define ITBL_NUMBER_OF_PROCESSORS 1
#endif
#ifndef ITBL_MAX_BITPOS
#define ITBL_MAX_BITPOS 31
#endif
#ifndef ITBL_TYPE
#define ITBL_TYPE unsigned long
#endif
#ifndef ITBL_IS_INSN
#define ITBL_IS_INSN(insn) 1
#endif
#ifndef ITBL_DECODE_PNUM
#define ITBL_DECODE_PNUM(insn) 0
#endif
#ifndef ITBL_ENCODE_PNUM
#define ITBL_ENCODE_PNUM(pnum) 0
#endif
typedef ITBL_TYPE t_insn;
/* types of entries */
typedef enum
{
e_insn,
e_dreg,
e_regtype0 = e_dreg,
e_creg,
e_greg,
e_addr,
e_nregtypes = e_greg+1,
e_immed,
e_ntypes,
e_invtype /* invalid type */
} e_type;
typedef enum
{
e_p0,
e_nprocs=NUMBER_OF_PROCESSORS,
e_invproc /* invalid processor */
} e_processor;
/* These routines are visible to the main part of the assembler */
int itbl_parse(char* insntbl);
void itbl_init(void);
char *itbl_get_field(char **s);
unsigned long itbl_assemble(char *name, char *operands);
int itbl_disassemble(char *str, unsigned long insn);
int itbl_parse(char *tbl); /* parses insn tbl */
unsigned long itbl_get_reg_val(char *name);
unsigned long itbl_get_val(e_processor processor, e_type type, char *name);
char *itbl_get_name(e_processor processor, e_type type, unsigned long val);
/* These routines are called by the table parser used to build the
* dynamic list of new processor instructions and registers. */
struct itbl_entry *itbl_add_reg(int yyproc, int yytype, char *regname, int regnum);
struct itbl_entry *itbl_add_insn(int yyproc, char *name, unsigned long value,
int sbit, int ebit, unsigned long flags);
struct itbl_field *itbl_add_operand(struct itbl_entry *e, int yytype,
int sbit, int ebit, unsigned long flags);

413
gas/itbl-parse.y Normal file
View File

@ -0,0 +1,413 @@
%{
/*
Yacc grammar for instruction table entries.
=======================================================================
Original Instruction table specification document:
MIPS Coprocessor Table Specification
====================================
This document describes the format of the MIPS coprocessor table. The
table specifies a list of valid functions, data registers and control
registers that can be used in coprocessor instructions. This list,
together with the coprocessor instruction classes listed below,
specifies the complete list of coprocessor instructions that will
be recognized and assembled by the GNU assembler. In effect,
this makes the GNU assembler table-driven, where the table is
specified by the programmer.
The table is an ordinary text file that the GNU assembler reads when
it starts. Using the information in the table, the assembler
generates an internal list of valid coprocessor registers and
functions. The assembler uses this internal list in addition to the
standard MIPS registers and instructions which are built-in to the
assembler during code generation.
To specify the coprocessor table when invoking the GNU assembler, use
the command line option "--itbl file", where file is the
complete name of the table, including path and extension.
Examples:
gas -t cop.tbl test.s -o test.o
gas -t /usr/local/lib/cop.tbl test.s -o test.o
gas --itbl d:\gnu\data\cop.tbl test.s -o test.o
Only one table may be supplied during a single invocation of
the assembler.
Instruction classes
===================
Below is a list of the valid coprocessor instruction classes for
any given coprocessor "z". These instructions are already recognized
by the assembler, and are listed here only for reference.
Class format instructions
-------------------------------------------------
Class1: op base rt offset
LWCz rt,offset(base)
SWCz rt,offset(base)
Class2: COPz sub rt rd 0
MTCz rt,rd
MFCz rt,rd
CTCz rt,rd
CFCz rt,rd
Class3: COPz CO cofun
COPz cofun
Class4: COPz BC br offset
BCzT offset
BCzF offset
Class5: COPz sub rt rd 0
DMFCz rt,rd
DMTCz rt,rd
Class6: op base rt offset
LDCz rt,offset(base)
SDCz rt,offset(base)
Class7: COPz BC br offset
BCzTL offset
BCzFL offset
The coprocessor table defines coprocessor-specific registers that can
be used with all of the above classes of instructions, where
appropriate. It also defines additional coprocessor-specific
functions for Class3 (COPz cofun) instructions, Thus, the table allows
the programmer to use convenient mnemonics and operands for these
functions, instead of the COPz mmenmonic and cofun operand.
The names of the MIPS general registers and their aliases are defined
by the assembler and will be recognized as valid register names by the
assembler when used (where allowed) in coprocessor instructions.
However, the names and values of all coprocessor data and control
register mnemonics must be specified in the coprocessor table.
Table Grammar
=============
Here is the grammar for the coprocessor table:
table -> entry*
entry -> [z entrydef] [comment] '\n'
entrydef -> type name val
entrydef -> 'insn' name val funcdef ; type of entry (instruction)
z -> 'p'['0'..'3'] ; processor number
type -> ['dreg' | 'creg' | 'greg' ] ; type of entry (register)
; 'dreg', 'creg' or 'greg' specifies a data, control, or general
; register mnemonic, respectively
name -> [ltr|dec]* ; mnemonic of register/function
val -> [dec|hex] ; register/function number (integer constant)
funcdef -> frange flags fields
; bitfield range for opcode
; list of fields' formats
fields -> field*
field -> [','] ftype frange flags
flags -> ['*' flagexpr]
flagexpr -> '[' flagexpr ']'
flagexpr -> val '|' flagexpr
ftype -> [ type | 'immed' | 'addr' ]
; 'immed' specifies an immediate value; see grammar for "val" above
; 'addr' specifies a C identifier; name of symbol to be resolved at
; link time
frange -> ':' val '-' val ; starting to ending bit positions, where
; where 0 is least significant bit
frange -> (null) ; default range of 31-0 will be assumed
comment -> [';'|'#'] [char]*
char -> any printable character
ltr -> ['a'..'z'|'A'..'Z']
dec -> ['0'..'9']* ; value in decimal
hex -> '0x'['0'..'9' | 'a'..'f' | 'A'..'F']* ; value in hexidecimal
Examples
========
Example 1:
The table:
p1 dreg d1 1 ; data register "d1" for COP1 has value 1
p1 creg c3 3 ; ctrl register "c3" for COP1 has value 3
p3 func fill 0x1f:24-20 ; function "fill" for COP3 has value 31 and
; no fields
will allow the assembler to accept the following coprocessor instructions:
LWC1 d1,0x100($2)
fill
Here, the general purpose register "$2", and instruction "LWC1", are standard
mnemonics built-in to the MIPS assembler.
Example 2:
The table:
p3 dreg d3 3 ; data register "d3" for COP3 has value 3
p3 creg c2 22 ; control register "c2" for COP3 has value 22
p3 func fee 0x1f:24-20 dreg:17-13 creg:12-8 immed:7-0
; function "fee" for COP3 has value 31, and 3 fields
; consisting of a data register, a control register,
; and an immediate value.
will allow the assembler to accept the following coprocessor instruction:
fee d3,c2,0x1
and will emit the object code:
31-26 25 24-20 19-18 17-13 12-8 7-0
COPz CO fun dreg creg immed
010011 1 11111 00 00011 10110 00000001
0x4ff07601
Example 3:
The table:
p3 dreg d3 3 ; data register "d3" for COP3 has value 3
p3 creg c2 22 ; control register "c2" for COP3 has value 22
p3 func fuu 0x01f00001 dreg:17-13 creg:12-8
will allow the assembler to accept the following coprocessor
instruction:
fuu d3,c2
and will emit the object code:
31-26 25 24-20 19-18 17-13 12-8 7-0
COPz CO fun dreg creg
010011 1 11111 00 00011 10110 00000001
0x4ff07601
In this way, the programmer can force arbitrary bits of an instruction
to have predefined values.
=======================================================================
Additional notes:
Encoding of ranges:
To handle more than one bit position range within an instruction,
use 0s to mask out the ranges which don't apply.
May decide to modify the syntax to allow commas separate multiple
ranges within an instruction (range','range).
Changes in grammar: The number of parms argument to the function entry
was deleted from the original format such that we now count the fields.
----
FIXME! should really change lexical analyzer
to recognize 'dreg' etc. in context sensative way.
Currently function names or mnemonics may be incorrectly parsed as keywords
FIXME! hex is ambiguous with any digit
*/
#include <stdio.h>
#include "itbl-ops.h"
/* #define DEBUG */
#ifdef DEBUG
#ifndef DBG_LVL
#define DBG_LVL 1
#endif
#else
#define DBG_LVL 0
#endif
#if DBG_LVL >= 1
#define DBG(x) printf x
#else
#define DBG(x)
#endif
#if DBG_LVL >= 2
#define DBGL2(x) printf x
#else
#define DBGL2(x)
#endif
static int sbit, ebit;
static struct itbl_entry *insn=0;
extern int insntbl_line;
int yyparse(void);
int yylex(void);
%}
%union {
char *str;
int num;
int processor;
unsigned long val;
}
%token DREG CREG GREG IMMED ADDR INSN NUM ID NL PNUM
%type <val> value flags flagexpr
%type <num> number NUM ftype regtype pnum PNUM
%type <str> ID name
%start insntbl
%%
insntbl: entrys
;
entrys: entry entrys
|
;
entry: pnum regtype name value NL
{
DBG(("line %d: entry pnum=%d type=%d name=%s value=x%x\n",
insntbl_line, $1, $2, $3, $4));
itbl_add_reg($1, $2, $3, $4);
}
| pnum INSN name value range flags
{
DBG(("line %d: entry pnum=%d type=INSN name=%s value=x%x",
insntbl_line, $1, $3, $4));
DBG((" sbit=%d ebit=%d flags=0x%x\n", sbit, ebit, $6));
insn=itbl_add_insn($1, $3, $4, sbit, ebit, $6);
}
fieldspecs NL
| NL
| error NL
;
fieldspecs: ',' fieldspec fieldspecs
| fieldspec fieldspecs
|
;
ftype: regtype
{
DBGL2(("ftype\n"));
$$ = $1;
}
| ADDR
{
DBGL2(("addr\n"));
$$ = ADDR;
}
| IMMED
{
DBGL2(("immed\n"));
$$ = IMMED;
}
;
fieldspec: ftype range flags
{
DBG(("line %d: field type=%d sbit=%d ebit=%d, flags=0x%x\n",
insntbl_line, $1, sbit, ebit, $3));
itbl_add_operand(insn, $1, sbit, ebit, $3);
}
;
flagexpr: NUM '|' flagexpr
{
$$ = $1 | $3;
}
| '[' flagexpr ']'
{
$$ = $2;
}
| NUM
{
$$ = $1;
}
;
flags: '*' flagexpr
{
DBGL2(("flags=%d\n", $2));
$$ = $2;
}
|
{
$$ = 0;
}
;
range: ':' NUM '-' NUM
{
DBGL2(("range %d %d\n", $2, $4));
sbit = $2;
ebit = $4;
}
|
{
sbit = 31;
ebit = 0;
}
;
pnum: PNUM
{
DBGL2(("pnum=%d\n",$1));
$$ = $1;
}
;
regtype: DREG
{
DBGL2(("dreg\n"));
$$ = DREG;
}
| CREG
{
DBGL2(("creg\n"));
$$ = CREG;
}
| GREG
{
DBGL2(("greg\n"));
$$ = GREG;
}
;
name: ID
{
DBGL2(("name=%s\n",$1));
$$ = $1;
}
;
number: NUM
{
DBGL2(("num=%d\n",$1));
$$ = $1;
}
;
value: NUM
{
DBGL2(("val=x%x\n",$1));
$$ = $1;
}
;
%%
void yyerror(char *msg)
{
printf("line %d: %s\n", insntbl_line, msg);
}