merge binutils and gdb sparc disassemblers

This commit is contained in:
Jim Kingdon 1993-04-01 16:56:06 +00:00
parent d1c89c2817
commit f7ed13c7d5
11 changed files with 74 additions and 479 deletions

View File

@ -69,24 +69,3 @@ strip.1
version.c
Do-last:
v9dirty="sparc-pinsn.c"
if ( echo $* | grep keep\-v9 > /dev/null ) ; then
echo Keeping v9 in ${v9dirty}
else
for i in ${v9dirty} ; do
echo Sanitizing v9 in $i
rm -f new
sed '/^#ifndef[ ]NO_V9/,/^#endif/d' < $i > new
if grep -s -i v9 new ; then
echo ***** SANITIZING V9 IN $i FAILED *****
fi
if [ -n "${safe}" ] ; then
mv $i .Recover
else
rm $i
fi
mv new $i
done
fi

View File

@ -1,5 +1,9 @@
Wed Mar 31 10:25:29 1993 Jim Kingdon (kingdon@lioth.cygnus.com)
* objdump.c (disassemble_data): print_insn_sparc is now a
`disassemble' not a `print'.
Makefile.in: Remove sparc-pinsn.c (now in libopcodes.a).
* objdump.c (disassemble_data): Use new read_memory_func stuff.
Thu Mar 25 10:38:11 1993 Ken Raeburn (raeburn@cambridge.cygnus.com)

View File

@ -112,7 +112,7 @@ INCLUDES = -I. -I$(srcdir) -I$(BASEDIR)/include -I$(BASEDIR)/bfd
# When adding .o files, to make VPATH work in Sun Make, you have to
# also add a foo.o: foo.c line at the bottom of the file.
DISASMS = i960-pinsn.o sparc-pinsn.o am29k-pinsn.o
DISASMS = i960-pinsn.o am29k-pinsn.o
#
## Random definitions
@ -156,7 +156,7 @@ testsuite:
check: all
/bin/sh $(srcdir)/sanity.sh .
test-install:
installcheck:
/bin/sh $(srcdir)/sanity.sh $(bindir)
info: binutils.info
@ -411,7 +411,6 @@ not-ranlib.o:not-ranlib.c
not-strip.o:not-strip.c
objdump.o: objdump.c
size.o: size.c
sparc-pinsn.o: sparc-pinsn.c
strip.o:strip.c
version.o: $(srcdir)/version.c
$(CC) $(CFLAGS) $(INCLUDES) $(HDEFINES) $(TDEFINES) -DVERSION='"$(VERSION)"' -c $(srcdir)/version.c

View File

@ -311,7 +311,6 @@ disassemble_data (abfd)
disassembler_ftype disassemble = 0; /* New style */
unsigned int print_insn_a29k ();
unsigned int print_insn_i960 ();
unsigned int print_insn_sparc ();
unsigned int print_insn_h8300 ();
enum bfd_architecture a;
struct disassemble_info disasm_info;
@ -378,7 +377,7 @@ disassemble_data (abfd)
switch (a)
{
case bfd_arch_sparc:
print = print_insn_sparc;
disassemble = print_insn_sparc;
break;
case bfd_arch_z8k:
if (bfd_get_mach(abfd) == bfd_mach_z8001)

View File

@ -1,3 +1,7 @@
Wed Mar 31 12:52:12 1993 Jim Kingdon (kingdon@lioth.cygnus.com)
* sparc-pinsn.c: Much abridged, just calls version in libopcodes.a.
Wed Mar 31 21:23:41 1993 K. Richard Pixley (rich@rtl.cygnus.com)
* core.c (dis_asm_read_memory): drop fourth arg which conflicts

View File

@ -1,7 +1,7 @@
/* Print SPARC instructions for GDB, the GNU Debugger.
Copyright 1989, 1991, 1992 Free Software Foundation, Inc.
/* Print sparc instructions for GDB, the GNU debugger.
Copyright 1986, 1987, 1989, 1991, 1992, 1993 Free Software Foundation, Inc.
This file is part of GDB, the GNU debugger.
This file is part of GDB.
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
@ -18,460 +18,19 @@ along with this program; if not, write to the Free Software
Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */
#include "defs.h"
#include "symtab.h"
#include "opcode/sparc.h"
#include "gdbcore.h"
#include <string.h>
#include "target.h"
#include "dis-asm.h"
#define freg_names (&reg_names[4 * 8])
/* Print the instruction at address MEMADDR in debugged memory,
on STREAM. Returns length of the instruction, in bytes. */
union sparc_insn
{
unsigned long int code;
struct
{
unsigned int anop:2;
#define op ldst.anop
unsigned int anrd:5;
#define rd ldst.anrd
unsigned int op3:6;
unsigned int anrs1:5;
#define rs1 ldst.anrs1
unsigned int i:1;
unsigned int anasi:8;
#define asi ldst.anasi
unsigned int anrs2:5;
#define rs2 ldst.anrs2
#define shcnt rs2
} ldst;
struct
{
unsigned int anop:2, anrd:5, op3:6, anrs1:5, i:1;
unsigned int IMM13:13;
#define imm13 IMM13.IMM13
} IMM13;
struct
{
unsigned int anop:2;
unsigned int a:1;
unsigned int cond:4;
unsigned int op2:3;
unsigned int DISP22:22;
#define disp22 branch.DISP22
} branch;
#define imm22 disp22
struct
{
unsigned int anop:2;
unsigned int adisp30:30;
#define disp30 call.adisp30
} call;
};
/* Nonzero if INSN is the opcode for a delayed branch. */
static int
is_delayed_branch (insn)
union sparc_insn insn;
{
unsigned int i;
for (i = 0; i < NUMOPCODES; ++i)
{
const struct sparc_opcode *opcode = &sparc_opcodes[i];
if ((opcode->match & insn.code) == opcode->match
&& (opcode->lose & insn.code) == 0)
return (opcode->flags & F_DELAYED);
}
return 0;
}
static int opcodes_sorted = 0;
extern void qsort ();
/* Print one instruction from MEMADDR on STREAM.
We suffix the instruction with a comment that gives the absolute
address involved, as well as its symbolic form, if the instruction
is preceded by a findable `sethi' and it either adds an immediate
displacement to that register, or it is an `add' or `or' instruction
on that register. */
int
print_insn (memaddr, stream)
CORE_ADDR memaddr;
FILE *stream;
{
union sparc_insn insn;
disassemble_info info;
register unsigned int i;
GDB_INIT_DISASSEMBLE_INFO(info, stream);
if (!opcodes_sorted)
{
static int compare_opcodes ();
qsort ((char *) sparc_opcodes, NUMOPCODES,
sizeof (sparc_opcodes[0]), compare_opcodes);
opcodes_sorted = 1;
}
read_memory (memaddr, (char *) &insn, sizeof (insn));
for (i = 0; i < NUMOPCODES; ++i)
{
const struct sparc_opcode *opcode = &sparc_opcodes[i];
if ((opcode->match & insn.code) == opcode->match
&& (opcode->lose & insn.code) == 0)
{
/* Nonzero means that we have found an instruction which has
the effect of adding or or'ing the imm13 field to rs1. */
int imm_added_to_rs1 = 0;
/* Nonzero means that we have found a plus sign in the args
field of the opcode table. */
int found_plus = 0;
/* Do we have an `add' or `or' instruction where rs1 is the same
as rsd, and which has the i bit set? */
if ((opcode->match == 0x80102000 || opcode->match == 0x80002000)
/* (or) (add) */
&& insn.rs1 == insn.rd)
imm_added_to_rs1 = 1;
if (insn.rs1 != insn.rd
&& strchr (opcode->args, 'r') != 0)
/* Can't do simple format if source and dest are different. */
continue;
fputs_filtered (opcode->name, stream);
{
register const char *s;
if (opcode->args[0] != ',')
fputs_filtered (" ", stream);
for (s = opcode->args; *s != '\0'; ++s)
{
if (*s == ',')
{
fputs_filtered (",", stream);
++s;
if (*s == 'a')
{
fputs_filtered ("a", stream);
++s;
}
fputs_filtered (" ", stream);
}
switch (*s)
{
case '+':
found_plus = 1;
/* note fall-through */
default:
fprintf_filtered (stream, "%c", *s);
break;
case '#':
fputs_filtered ("0", stream);
break;
#define reg(n) fprintf_filtered (stream, "%%%s", reg_names[n])
case '1':
case 'r':
reg (insn.rs1);
break;
case '2':
reg (insn.rs2);
break;
case 'd':
reg (insn.rd);
break;
#undef reg
#define freg(n) fprintf_filtered (stream, "%%%s", freg_names[n])
case 'e':
case 'v': /* double/even */
case 'V': /* quad/multiple of 4 */
freg (insn.rs1);
break;
case 'f':
case 'B': /* double/even */
case 'R': /* quad/multiple of 4 */
freg (insn.rs2);
break;
case 'g':
case 'H': /* double/even */
case 'J': /* quad/multiple of 4 */
freg (insn.rd);
break;
#undef freg
#define creg(n) fprintf_filtered (stream, "%%c%u", (unsigned int) (n))
case 'b':
creg (insn.rs1);
break;
case 'c':
creg (insn.rs2);
break;
case 'D':
creg (insn.rd);
break;
#undef creg
case 'h':
fprintf_filtered (stream, "%%hi(%#x)",
(int) insn.imm22 << 10);
break;
case 'i':
{
/* We cannot trust the compiler to sign-extend
when extracting the bitfield, hence the shifts. */
int imm = ((int) insn.imm13 << 19) >> 19;
/* Check to see whether we have a 1+i, and take
note of that fact.
FIXME: No longer true/relavant ???
Note: because of the way we sort the table,
we will be matching 1+i rather than i+1,
so it is OK to assume that i is after +,
not before it. */
if (found_plus)
imm_added_to_rs1 = 1;
if (imm <= 9)
fprintf_filtered (stream, "%d", imm);
else
fprintf_filtered (stream, "%#x", imm);
}
break;
case 'L':
print_address ((CORE_ADDR) memaddr + insn.disp30 * 4,
stream);
break;
case 'l':
if ((insn.code >> 22) == 0)
/* Special case for `unimp'. Don't try to turn
it's operand into a function offset. */
fprintf_filtered (stream, "%#x",
(int) (((int) insn.disp22 << 10) >> 10));
else
/* We cannot trust the compiler to sign-extend
when extracting the bitfield, hence the shifts. */
print_address ((CORE_ADDR)
(memaddr
+ (((int) insn.disp22 << 10) >> 10) * 4),
stream);
break;
case 'A':
fprintf_filtered (stream, "(%d)", (int) insn.asi);
break;
case 'C':
fputs_filtered ("%csr", stream);
break;
case 'F':
fputs_filtered ("%fsr", stream);
break;
case 'p':
fputs_filtered ("%psr", stream);
break;
case 'q':
fputs_filtered ("%fq", stream);
break;
case 'Q':
fputs_filtered ("%cq", stream);
break;
case 't':
fputs_filtered ("%tbr", stream);
break;
case 'w':
fputs_filtered ("%wim", stream);
break;
case 'y':
fputs_filtered ("%y", stream);
break;
}
}
}
/* If we are adding or or'ing something to rs1, then
check to see whether the previous instruction was
a sethi to the same register as in the sethi.
If so, attempt to print the result of the add or
or (in this context add and or do the same thing)
and its symbolic value. */
if (imm_added_to_rs1)
{
union sparc_insn prev_insn;
int errcode;
errcode = target_read_memory (memaddr - 4,
(char *)&prev_insn, sizeof (prev_insn));
if (errcode == 0)
{
/* If it is a delayed branch, we need to look at the
instruction before the delayed branch. This handles
sequences such as
sethi %o1, %hi(_foo), %o1
call _printf
or %o1, %lo(_foo), %o1
*/
if (is_delayed_branch (prev_insn))
errcode = target_read_memory
(memaddr - 8, (char *)&prev_insn, sizeof (prev_insn));
}
/* If there was a problem reading memory, then assume
the previous instruction was not sethi. */
if (errcode == 0)
{
/* Is it sethi to the same register? */
if ((prev_insn.code & 0xc1c00000) == 0x01000000
&& prev_insn.rd == insn.rs1)
{
fprintf_filtered (stream, "\t! ");
/* We cannot trust the compiler to sign-extend
when extracting the bitfield, hence the shifts. */
print_address (((int) prev_insn.imm22 << 10)
| (insn.imm13 << 19) >> 19, stream);
}
}
}
return sizeof (insn);
}
}
printf_filtered ("%#8x", insn.code);
return sizeof (insn);
}
/* Compare opcodes A and B. */
static int
compare_opcodes (a, b)
char *a, *b;
{
struct sparc_opcode *op0 = (struct sparc_opcode *) a;
struct sparc_opcode *op1 = (struct sparc_opcode *) b;
unsigned long int match0 = op0->match, match1 = op1->match;
unsigned long int lose0 = op0->lose, lose1 = op1->lose;
register unsigned int i;
/* If a bit is set in both match and lose, there is something
wrong with the opcode table. */
if (match0 & lose0)
{
fprintf (stderr, "Internal error: bad sparc-opcode.h: \"%s\", %#.8lx, %#.8lx\n",
op0->name, match0, lose0);
op0->lose &= ~op0->match;
lose0 = op0->lose;
}
if (match1 & lose1)
{
fprintf (stderr, "Internal error: bad sparc-opcode.h: \"%s\", %#.8lx, %#.8lx\n",
op1->name, match1, lose1);
op1->lose &= ~op1->match;
lose1 = op1->lose;
}
/* Because the bits that are variable in one opcode are constant in
another, it is important to order the opcodes in the right order. */
for (i = 0; i < 32; ++i)
{
unsigned long int x = 1 << i;
int x0 = (match0 & x) != 0;
int x1 = (match1 & x) != 0;
if (x0 != x1)
return x1 - x0;
}
for (i = 0; i < 32; ++i)
{
unsigned long int x = 1 << i;
int x0 = (lose0 & x) != 0;
int x1 = (lose1 & x) != 0;
if (x0 != x1)
return x1 - x0;
}
/* They are functionally equal. So as long as the opcode table is
valid, we can put whichever one first we want, on aesthetic grounds. */
/* Our first aesthetic ground is that aliases defer to real insns. */
{
int alias_diff = (op0->flags & F_ALIAS) - (op1->flags & F_ALIAS);
if (alias_diff != 0)
/* Put the one that isn't an alias first. */
return alias_diff;
}
/* Except for aliases, two "identical" instructions had
better have the same opcode. This is a sanity check on the table. */
i = STRCMP (op0->name, op1->name);
if (i)
if (op0->flags & F_ALIAS) /* If they're both aliases, be arbitrary. */
return i;
else
fprintf (stderr,
"Internal error: bad sparc-opcode.h: \"%s\" == \"%s\"\n",
op0->name, op1->name);
/* Fewer arguments are preferred. */
{
int length_diff = strlen (op0->args) - strlen (op1->args);
if (length_diff != 0)
/* Put the one with fewer arguments first. */
return length_diff;
}
/* Put 1+i before i+1. */
{
char *p0 = (char *) strchr(op0->args, '+');
char *p1 = (char *) strchr(op1->args, '+');
if (p0 && p1)
{
/* There is a plus in both operands. Note that a plus
sign cannot be the first character in args,
so the following [-1]'s are valid. */
if (p0[-1] == 'i' && p1[1] == 'i')
/* op0 is i+1 and op1 is 1+i, so op1 goes first. */
return 1;
if (p0[1] == 'i' && p1[-1] == 'i')
/* op0 is 1+i and op1 is i+1, so op0 goes first. */
return -1;
}
}
/* They are, as far as we can tell, identical.
Since qsort may have rearranged the table partially, there is
no way to tell which one was first in the opcode table as
written, so just say there are equal. */
return 0;
return print_insn_sparc (memaddr, &info);
}

View File

@ -1,3 +1,7 @@
Wed Mar 31 17:40:16 1993 Jim Kingdon (kingdon@lioth.cygnus.com)
* dis-asm.h: Add print_insn_sparc.
Wed Mar 31 17:51:42 1993 Ian Lance Taylor (ian@cygnus.com)
* bfd.h: Updated for BFD_RELOC_MIPS_GPREL and bfd_[gs]et_gp_size
@ -11,7 +15,7 @@ Wed Mar 31 16:35:12 1993 Stu Grossman (grossman@cygnus.com)
Tue Mar 30 19:09:23 1993 Jim Kingdon (kingdon@lioth.cygnus.com)
* dis-asm.h (disassembler_info): Add read_memory_func,
memory_error_func, buffer, buffer_length, and buffer_vma.
memory_error_func, buffer, and length.
({GDB_,}INIT_DISASSEMBLE_INFO): Set them.
print_insn_*: Remove second argument.

View File

@ -80,3 +80,4 @@ extern int print_insn_m68k PARAMS ((bfd_vma,disassemble_info*));
extern int print_insn_z8001 PARAMS ((bfd_vma,disassemble_info*));
extern int print_insn_z8002 PARAMS ((bfd_vma,disassemble_info*));
extern int print_insn_h8500 PARAMS ((bfd_vma,disassemble_info*));
extern int print_insn_sparc PARAMS ((bfd_vma,disassemble_info*));

View File

@ -52,4 +52,25 @@ else
mv new sparc-opc.c
fi
v9dirty="sparc-pinsn.c"
if ( echo $* | grep keep\-v9 > /dev/null ) ; then
echo Keeping v9 in ${v9dirty}
else
for i in ${v9dirty} ; do
echo Sanitizing v9 in $i
rm -f new
sed '/^#ifndef[ ]NO_V9/,/^#endif/d' < $i > new
if grep -s -i v9 new ; then
echo ***** SANITIZING V9 IN $i FAILED *****
fi
if [ -n "${safe}" ] ; then
mv $i .Recover
else
rm $i
fi
mv new $i
done
fi
# End of file.

View File

@ -1,3 +1,25 @@
Wed Mar 31 10:07:04 1993 Jim Kingdon (kingdon@lioth.cygnus.com)
* Makefile.in: Add sparc-dis.c.
sparc-dis.c: New file, merges binutils and gdb versions as follows:
From GDB:
Add `add' instruction to the set that get checked
for a preceding `sethi' in order to print an absolute address.
* (print_insn): Disassembly prefers real instructions.
(is_delayed_branch): Speed up.
* sparc-opcode.h: Add ALIAS bit to aliases. Fix up opcode tables.
Still missing some float ops, and needs testing.
* sparc-pinsn.c (print_insn): Eliminate 'set' test, subsumed by
F_ALIAS. Use printf, not fprintf, when not passing a file
pointer...
(compare_opcodes): Check that identical instructions have
identical opcodes, complain otherwise.
From binutils:
* New 'm' arg.
* Include reg_names.
From neither:
Use dis-asm.h/read_memory_func interface.
Wed Mar 31 20:49:06 1993 K. Richard Pixley (rich@rtl.cygnus.com)
* h8500-dis.c, i386-dis.c, m68k-dis.c, z8k-dis.c (fetch_data):

View File

@ -61,8 +61,10 @@ DEP = mkdep
TARGETLIB = libopcodes.a
DIS_LIBS = i386-dis.o z8k-dis.o m68k-dis.o mips-dis.o h8500-dis.o dis-buf.o
# To circumvent a Sun make VPATH bug, each file listed here
# should also have a foo.o: foo.c line further along in this file.
DIS_LIBS = i386-dis.o z8k-dis.o m68k-dis.o mips-dis.o h8500-dis.o dis-buf.o \
sparc-dis.o
OFILES = $(DIS_LIBS) sparc-opc.o m68881-ext.o
#### host and target dependent Makefile fragments come in here.
@ -114,6 +116,7 @@ $(TARGETLIB): $(OFILES)
sparc-opc.o: sparc-opc.c
m68881-ext.o: m68881-ext.c
h8500-dis.o: h8500-dis.c
sparc-dis.o: sparc-dis.c
tags etags: TAGS