be9a236222
macro USER_CFLAGS to Makefile. Add stratus config frags to configure.in.
1637 lines
48 KiB
C
1637 lines
48 KiB
C
/* Machine-dependent code which would otherwise be in inflow.c and core.c,
|
||
for GDB, the GNU debugger.
|
||
Copyright (C) 1992 Free Software Foundation, Inc.
|
||
This code is for the i860 cpu.
|
||
Contributed by Peggy Fieland (pfieland@stratus.com)
|
||
|
||
GDB is distributed in the hope that it will be useful, but WITHOUT ANY
|
||
WARRANTY. No author or distributor accepts responsibility to anyone
|
||
for the consequences of using it or for whether it serves any
|
||
particular purpose or works at all, unless he says so in writing.
|
||
Refer to the GDB General Public License for full details.
|
||
|
||
Everyone is granted permission to copy, modify and redistribute GDB,
|
||
but only under the conditions described in the GDB General Public
|
||
License. A copy of this license is supposed to have been given to you
|
||
along with GDB so you can know your rights and responsibilities. It
|
||
should be in a file named COPYING. Among other things, the copyright
|
||
notice and this notice must be preserved on all copies.
|
||
|
||
In other words, go ahead and share GDB, but don't try to stop
|
||
anyone else from sharing it farther. Help stamp out software hoarding!
|
||
*/
|
||
|
||
#include "defs.h"
|
||
#include "tm-i860.h"
|
||
#include "frame.h"
|
||
#include "inferior.h"
|
||
#include "obstack.h"
|
||
#include "symtab.h"
|
||
#include "value.h"
|
||
|
||
#include "tm-i860.h"
|
||
#include "i860-opcode.h"
|
||
|
||
#include <stdio.h>
|
||
#include "break.h"
|
||
|
||
#ifdef notdef
|
||
|
||
#include <sys/types.h>
|
||
#include <sys/param.h>
|
||
#include <sys/dir.h>
|
||
|
||
#endif
|
||
|
||
#include <signal.h>
|
||
#include <sys/ioctl.h>
|
||
#include <fcntl.h>
|
||
|
||
/* #include <sys/reg.h> */
|
||
#include "i860_reg.h"
|
||
|
||
#include <a.out.h>
|
||
#include <sys/file.h>
|
||
#include <sys/stat.h>
|
||
#include <core.h>
|
||
|
||
#include <sys/user.h>
|
||
|
||
#include <elf.h>
|
||
#include <sys/elftypes.h>
|
||
#include <sys/elf_860.h>
|
||
#include <libelf.h>
|
||
|
||
|
||
int btdebug = 0; /* change value to 1 to enable debugging code */
|
||
|
||
#define BTDEBUG if (btdebug) btdebug_message
|
||
|
||
#include <stdarg.h>
|
||
|
||
int read_memory();
|
||
int write_memory();
|
||
|
||
btdebug_message(char *format, ...)
|
||
{
|
||
va_list arglist;
|
||
va_start( arglist, format );
|
||
|
||
if( btdebug )
|
||
vfprintf (stderr, format, arglist );
|
||
va_end ( arglist );
|
||
}
|
||
|
||
extern int errno;
|
||
extern int attach_flag;
|
||
|
||
|
||
/* This is used when GDB is exiting. It gives less chance of error.*/
|
||
|
||
|
||
/* Simulate single-step ptrace call for sun4. Code written by Gary
|
||
Beihl (beihl@mcc.com). */
|
||
/* Modified for i860 by Jim Hanko (hanko@orc.olivetti.com) */
|
||
|
||
|
||
static struct breakpoint brk;
|
||
typedef char binsn_quantum[sizeof break_insn];
|
||
static binsn_quantum break_mem[2];
|
||
|
||
/* Non-zero if we just simulated a single-step ptrace call. This is
|
||
needed because we cannot remove the breakpoints in the inferior
|
||
process until after the `wait' in `wait_for_inferior'. Used for
|
||
i860. */
|
||
|
||
int one_stepped;
|
||
void
|
||
single_step (signal)
|
||
int signal;
|
||
{
|
||
CORE_ADDR pc;
|
||
branch_type place_brk();
|
||
|
||
pc = read_register (PC_REGNUM);
|
||
|
||
if (!one_stepped)
|
||
{
|
||
brk.address = pc;
|
||
place_brk (pc, SINGLE_STEP_MODE, &brk);
|
||
brk.shadow_contents[0] = brk.shadow_contents[1] = 0;
|
||
brk.shadow_contents[2] = brk.shadow_contents[3] = 0;
|
||
|
||
if (brk.mode == DIM)
|
||
{
|
||
if ( brk.address1 )
|
||
{
|
||
printf(" DIM1 ->");
|
||
printf(" %x : ", brk.act_addr[3]);
|
||
print_insn( brk.act_addr[3], stdout);
|
||
printf("\t -|- ");
|
||
printf(" %x : ", brk.act_addr[2]);
|
||
print_insn( brk.act_addr[2], stdout);
|
||
printf("\n");
|
||
fflush(stdout);
|
||
|
||
adj_read_memory (brk.act_addr[2], &brk.shadow_contents[2], 4);
|
||
adj_write_memory (brk.act_addr[2], break_insn, 4);
|
||
adj_read_memory (brk.act_addr[3], &brk.shadow_contents[3], 4);
|
||
/* adj_write_memory (brk.act_addr[3], float_insn, 4); */
|
||
|
||
}
|
||
if ( brk.address1)
|
||
printf(" DIM2 ->");
|
||
else
|
||
printf(" DIM1 ->");
|
||
|
||
printf(" %x : ", brk.act_addr[1]);
|
||
print_insn( brk.act_addr[1], stdout);
|
||
printf("\t -|- ");
|
||
printf(" %x : ", brk.act_addr[0]);
|
||
print_insn( brk.act_addr[0], stdout);
|
||
printf("\n");
|
||
fflush(stdout);
|
||
|
||
adj_read_memory (brk.act_addr[0], &brk.shadow_contents[0], 4);
|
||
adj_write_memory (brk.act_addr[0], break_insn, 4);
|
||
adj_read_memory (brk.act_addr[1], &brk.shadow_contents[1], 4);
|
||
/* adj_write_memory (brk.act_addr[1], float_insn, 4); */
|
||
|
||
}
|
||
else {
|
||
if (brk.address1)
|
||
{
|
||
if (btdebug)
|
||
{
|
||
printf(" SIM1 ->");
|
||
printf(" %x : ", brk.act_addr[2]);
|
||
print_insn( brk.act_addr[2], stdout);
|
||
printf("\n");
|
||
fflush(stdout);
|
||
}
|
||
adj_read_memory (brk.act_addr[2], &brk.shadow_contents[2], 4);
|
||
adj_write_memory (brk.act_addr[2], break_insn, 4);
|
||
}
|
||
if (btdebug)
|
||
{
|
||
if ( brk.address1)
|
||
printf(" SIM2 ->");
|
||
else
|
||
printf(" SIM1 ->");
|
||
|
||
printf(" %x : ", brk.act_addr[0]);
|
||
print_insn( brk.act_addr[0], stdout);
|
||
printf("\n");
|
||
fflush(stdout);
|
||
}
|
||
adj_read_memory (brk.act_addr[0], &brk.shadow_contents[0], 4);
|
||
adj_write_memory (brk.act_addr[0], break_insn, 4);
|
||
}
|
||
|
||
/* Let it go */
|
||
one_stepped = 1;
|
||
return;
|
||
}
|
||
else
|
||
{
|
||
/* Remove breakpoints */
|
||
if (brk.mode == DIM)
|
||
{
|
||
adj_write_memory (brk.act_addr[0], &brk.shadow_contents[0], 4);
|
||
adj_write_memory (brk.act_addr[1], &brk.shadow_contents[1], 4);
|
||
} else {
|
||
adj_write_memory (brk.act_addr[0], &brk.shadow_contents[0], 4);
|
||
}
|
||
|
||
if (brk.address1)
|
||
{
|
||
if (brk.mode == DIM)
|
||
{
|
||
adj_write_memory (brk.act_addr[2], &brk.shadow_contents[2], 4);
|
||
adj_write_memory (brk.act_addr[3], &brk.shadow_contents[3], 4);
|
||
} else {
|
||
adj_write_memory (brk.act_addr[2], &brk.shadow_contents[2], 4);
|
||
}
|
||
}
|
||
one_stepped = 0;
|
||
}
|
||
}
|
||
|
||
|
||
|
||
|
||
/* return nonzero if the routine containing pc has been
|
||
* compiled with -g. We assume -g if the first instruction is
|
||
* an addu|adds -X,sp and the second is st.l fp,XX(sp)
|
||
*
|
||
* based on skip_prologue();
|
||
*/
|
||
|
||
g_routine(pc)
|
||
CORE_ADDR pc;
|
||
{
|
||
long instr;
|
||
int regno;
|
||
CORE_ADDR top_pc;
|
||
|
||
top_pc = get_pc_function_start(pc);
|
||
if (top_pc)
|
||
{
|
||
instr = adj_read_memory_integer (top_pc);
|
||
/* Recognize "addu|adds -X,sp,sp" insn. */
|
||
|
||
if ((instr & 0xEFFF0000) == 0x84420000)
|
||
{
|
||
top_pc += 4;
|
||
instr = adj_read_memory_integer (top_pc);
|
||
|
||
if( (instr & 0xFFE0F801) == 0x1C401801 ) /* st.l fp,X(sp) */
|
||
return(1);
|
||
}
|
||
}
|
||
return(0);
|
||
}
|
||
|
||
|
||
/* Written for i860 by Jim Hanko (hanko@orc.olivetti.com) */
|
||
/* This code was based on SPARC code written by Gary Beihl (beihl@mcc.com),
|
||
by Michael Tiemann (tiemann@corto.inria.fr). */
|
||
|
||
struct command_line *get_breakpoint_commands ();
|
||
|
||
CORE_ADDR
|
||
skip_prologue (pc)
|
||
CORE_ADDR pc;
|
||
{
|
||
long instr;
|
||
int regno;
|
||
|
||
instr = adj_read_memory_integer (pc);
|
||
|
||
/* Recognize "addu|adds -X,sp,sp" insn. */
|
||
if ((instr & 0xEFFF0000) == 0x84420000)
|
||
{
|
||
pc += 4;
|
||
instr = adj_read_memory_integer (pc);
|
||
}
|
||
else
|
||
return(pc); /* No frame! */
|
||
|
||
/* Recognize store of return addr and frame pointer into frame */
|
||
while (1)
|
||
{
|
||
if ((instr & 0xFFE0F801) == 0x1C400801 || /* st.l r1,X(sp) */
|
||
(instr & 0xFFE0F801) == 0x1C401801) /* st.l fp,X(sp) */
|
||
{
|
||
pc += 4;
|
||
instr = adj_read_memory_integer (pc);
|
||
}
|
||
else
|
||
break;
|
||
}
|
||
|
||
/* Recognize "addu|adds X,sp,fp" insn. */
|
||
if ((instr & 0xEFFF0000) == 0x84430000)
|
||
{
|
||
pc += 4;
|
||
instr = adj_read_memory_integer (pc);
|
||
}
|
||
|
||
/* Now recognize stores into the frame from the registers. */
|
||
|
||
while (1)
|
||
{
|
||
if ((instr & 0xFFA00003) == 0x1C200001 || /* st.l rn,X(fp|sp) */
|
||
(instr & 0xFFA00001) == 0x4C200000) /* fst.y fn,X(fp|sp) */
|
||
{
|
||
regno = (instr >> 11) & 0x1f;
|
||
if (regno == 0) /* source reg == 0? quit */
|
||
break;
|
||
pc += 4;
|
||
instr = adj_read_memory_integer (pc);
|
||
}
|
||
else
|
||
break;
|
||
}
|
||
|
||
return(pc);
|
||
}
|
||
|
||
|
||
/* Set *nextpc to branch target if we find a branch. If it is not a branch,
|
||
set it to the next instruction (addr + 4) */
|
||
|
||
|
||
branch_type
|
||
isabranch (addr, nextpc)
|
||
CORE_ADDR addr, *nextpc;
|
||
{
|
||
long instr;
|
||
branch_type val = not_branch;
|
||
long offset; /* Must be signed for sign-extend */
|
||
|
||
printf(" isabranch\n");
|
||
*nextpc = addr;
|
||
instr = adj_read_memory_integer (addr);
|
||
|
||
if ((instr & 0xE0000000) == 0x60000000 && /* CTRL format */
|
||
(instr & 0xF8000000) != 0x60000000) /* not pfld.y */
|
||
{
|
||
if ((instr & 0xF8000000) == 0x68000000) /* br or call */
|
||
val = uncond_d;
|
||
else if ((instr & 0xF4000000) == 0x74000000) /* bc.t or bnc.t */
|
||
val = cond_d;
|
||
else if ((instr & 0xF4000000) == 0x70000000) /* bc or bnc */
|
||
val = cond;
|
||
|
||
offset = (instr & 0x03ffffff);
|
||
if (offset & 0x02000000) /* sign extend? */
|
||
offset |= 0xFC000000;
|
||
*nextpc = addr + 4 + (offset << 2);
|
||
}
|
||
else if ((instr & 0xFC00003F) == 0x4C000002 || /* calli */
|
||
(instr & 0xFC000000) == 0x40000000) /* bri */
|
||
{
|
||
val = uncond_d;
|
||
offset = ((instr & 0x0000F800) >> 11);
|
||
*nextpc = (read_register(offset) & 0xFFFFFFFC);
|
||
}
|
||
else if ((instr & 0xF0000000) == 0x50000000) /* bte or btne */
|
||
{
|
||
val = cond;
|
||
|
||
offset = SIGN_EXT16(((instr & 0x001F0000) >> 5) | (instr & 0x000007FF));
|
||
*nextpc = addr + 4 + (offset << 2);
|
||
}
|
||
else if ((instr & 0xFC000000) == 0xB4000000) /* bla */
|
||
{
|
||
val = cond_d;
|
||
|
||
offset = SIGN_EXT16(((instr & 0x001F0000) >> 5) | (instr & 0x000007FF));
|
||
*nextpc = addr + 4 + (offset << 2);
|
||
}
|
||
|
||
printf(" Final addr - %x\n", *nextpc);
|
||
/*printf("isabranch ret: %d\n",val); */
|
||
return val;
|
||
}
|
||
|
||
/* set in call_function() [valops.c] to the address of the "call dummy" code
|
||
so dummy frames can be easily recognized; also used in wait_for_inferior()
|
||
[infrun.c]. When not used, it points into the ABI's 'reserved area' */
|
||
|
||
CORE_ADDR call_dummy_set = 0; /* true if dummy call being done */
|
||
CORE_ADDR call_dummy_start; /* address of call dummy code */
|
||
|
||
frame_find_saved_regs(frame_info, frame_saved_regs)
|
||
struct frame_info *frame_info;
|
||
struct frame_saved_regs *frame_saved_regs;
|
||
{
|
||
register CORE_ADDR pc;
|
||
long instr, spdelta = 0, offset;
|
||
int i, size, reg;
|
||
int r1_off = -1, fp_off = -1;
|
||
int framesize;
|
||
|
||
bzero (frame_saved_regs, sizeof(*frame_saved_regs));
|
||
|
||
if (call_dummy_set && frame_info->pc >= call_dummy_start &&
|
||
frame_info->pc <= call_dummy_start + CALL_DUMMY_LENGTH)
|
||
{
|
||
/* DUMMY frame - all registers stored in order at fp; old sp is
|
||
at fp + NUM_REGS*4 */
|
||
|
||
for (i = 1; i < NUM_REGS; i++) /* skip reg 0 */
|
||
if (i != SP_REGNUM && i != FP0_REGNUM && i != FP0_REGNUM + 1)
|
||
frame_saved_regs->regs[i] = frame_info->frame + i*4;
|
||
|
||
frame_saved_regs->regs[SP_REGNUM] = frame_info->frame + NUM_REGS*4;
|
||
|
||
call_dummy_set = 0;
|
||
return;
|
||
}
|
||
|
||
pc = get_pc_function_start (frame_info->pc);
|
||
if (pc)
|
||
{
|
||
instr = adj_read_memory_integer (pc);
|
||
/* Recognize "addu|adds -X,sp,sp" insn. */
|
||
if ((instr & 0xEFFF0000) == 0x84420000)
|
||
{
|
||
framesize = -SIGN_EXT16(instr & 0x0000FFFF);
|
||
pc += 4;
|
||
instr = adj_read_memory_integer (pc);
|
||
}
|
||
}
|
||
else
|
||
goto punt; /* No frame! */
|
||
|
||
/* Recognize store of return addr and frame pointer into frame */
|
||
while (1)
|
||
{
|
||
if ((instr & 0xFFE0F801) == 0x1C400801) /* st.l r1,X(sp) */
|
||
{
|
||
r1_off = SIGN_EXT16(((instr&0x001F0000) >> 5) | (instr&0x000007FE));
|
||
pc += 4;
|
||
instr = adj_read_memory_integer (pc);
|
||
}
|
||
else if ((instr & 0xFFE0F801) == 0x1C401801) /* st.l fp,X(sp) */
|
||
{
|
||
fp_off = SIGN_EXT16(((instr&0x001F0000) >> 5) | (instr&0x000007FE));
|
||
pc += 4;
|
||
instr = adj_read_memory_integer (pc);
|
||
}
|
||
else
|
||
break;
|
||
}
|
||
|
||
/* Recognize "addu|adds X,sp,fp" insn. */
|
||
if ((instr & 0xEFFF0000) == 0x84430000)
|
||
{
|
||
spdelta = SIGN_EXT16(instr & 0x0000FFFF);
|
||
pc += 4;
|
||
instr = adj_read_memory_integer (pc);
|
||
}
|
||
|
||
/* Now recognize stores into the frame from the registers. */
|
||
|
||
while (1)
|
||
{
|
||
if ((instr & 0xFFC00003) == 0x1C400001) /* st.l rn,X(fp|sp) */
|
||
{
|
||
offset = SIGN_EXT16(((instr&0x001F0000) >> 5) | (instr&0x000007FE));
|
||
reg = (instr >> 11) & 0x1F;
|
||
if (reg == 0)
|
||
break;
|
||
if ((instr & 0x00200000) == 0) /* was this using sp? */
|
||
if (spdelta) /* and we know sp-fp delta */
|
||
offset -= spdelta; /* if so, adjust the offset */
|
||
else
|
||
break; /* if not, give up */
|
||
|
||
|
||
/* Handle the case where the return address is stored after the fp
|
||
is adjusted */
|
||
|
||
if (reg == 1)
|
||
frame_saved_regs->regs[PC_REGNUM] = frame_info->frame + offset;
|
||
else
|
||
frame_saved_regs->regs[reg] = frame_info->frame + offset;
|
||
|
||
pc += 4;
|
||
instr = adj_read_memory_integer (pc);
|
||
}
|
||
else if ((instr & 0xFFC00001) == 0x2C400000) /* fst.y fn,X(fp|sp) */
|
||
{
|
||
/*
|
||
* The number of words in a floating store based on 3 LSB of instr
|
||
*/
|
||
static int fst_sizes[] = {2, 0, 1, 0, 4, 0, 1, 0};
|
||
|
||
size = fst_sizes[instr & 7];
|
||
reg = ((instr >> 16) & 0x1F) + FP0_REGNUM;
|
||
if (reg == 0)
|
||
break;
|
||
|
||
if (size > 1) /* align the offset */
|
||
offset = SIGN_EXT16(instr & 0x0000FFF8); /* drop 3 bits */
|
||
else
|
||
offset = SIGN_EXT16(instr & 0x0000FFFC); /* drop 2 bits */
|
||
|
||
if ((instr & 0x00200000) == 0) /* was this using sp? */
|
||
if (spdelta) /* and we know sp-fp delta */
|
||
offset -= spdelta; /* if so, adjust the offset */
|
||
else
|
||
break; /* if not, give up */
|
||
|
||
for (i = 0; i < size; i++)
|
||
{
|
||
frame_saved_regs->regs[reg] = frame_info->frame + offset;
|
||
|
||
offset += 4;
|
||
reg++;
|
||
}
|
||
|
||
pc += 4;
|
||
instr = adj_read_memory_integer (pc);
|
||
}
|
||
else
|
||
break;
|
||
}
|
||
|
||
punt: ;
|
||
if (framesize != 0 && spdelta != 0)
|
||
frame_saved_regs->regs[SP_REGNUM] = frame_info->frame+(framesize-spdelta);
|
||
else
|
||
frame_saved_regs->regs[SP_REGNUM] = frame_info->frame + 8;
|
||
|
||
if (spdelta && fp_off != -1)
|
||
frame_saved_regs->regs[FP_REGNUM] = frame_info->frame - spdelta + fp_off;
|
||
else
|
||
frame_saved_regs->regs[FP_REGNUM] = frame_info->frame;
|
||
|
||
if (spdelta && r1_off != -1)
|
||
frame_saved_regs->regs[PC_REGNUM] = frame_info->frame - spdelta + r1_off;
|
||
else
|
||
frame_saved_regs->regs[PC_REGNUM] = frame_info->frame + 4;
|
||
}
|
||
|
||
/* return the stack offset where the fp register is stored */
|
||
find_fp_offset(pc)
|
||
{
|
||
int fp_off,i;
|
||
long instr;
|
||
|
||
/* look for the instruction and examine the offset */
|
||
|
||
for(i=4; i<16; i+=4){
|
||
instr = adj_read_memory_integer(pc+i);
|
||
if( (instr & 0xFFE0F801) == 0x1C401801) { /* st.l fp,X(sp) */
|
||
|
||
fp_off = SIGN_EXT16(((instr&0x001F0000) >> 5) |
|
||
(instr&0x000007FE));
|
||
return(fp_off);
|
||
}
|
||
}
|
||
return(0);
|
||
}
|
||
|
||
/* return the stack offset where r1 (return linkage ) register is stored */
|
||
|
||
find_r1_offset(pc)
|
||
{
|
||
int r1_off,i;
|
||
long instr;
|
||
|
||
/* look for the instruction and examine the offset */
|
||
|
||
for(i=4; i<16; i+=4){
|
||
|
||
instr = adj_read_memory_integer(pc+i);
|
||
if ((instr & 0xFFE0F801) == 0x1C400801) { /* st.l r1,X(sp) */
|
||
|
||
r1_off = SIGN_EXT16(((instr&0x001F0000) >> 5) |
|
||
(instr&0x000007FE));
|
||
return(r1_off);
|
||
}
|
||
}
|
||
return(-1);
|
||
}
|
||
|
||
|
||
/* dose routine starting at pc build a stack frame of any kind?? */
|
||
has_a_frame(pc)
|
||
{
|
||
if( skip_prologue(pc) != pc )return(1);
|
||
else return(0);
|
||
}
|
||
|
||
/* get the frame pointer of the caller.
|
||
* note that only routines that have been compiled with
|
||
* -g have full (XX)fp style stack frames
|
||
* if we are not returning to a non -g caller then we
|
||
* return the sp at entry to us as it is the caller's
|
||
* frame reference.
|
||
*/
|
||
|
||
frame_chain(thisframe)
|
||
FRAME thisframe;
|
||
{
|
||
unsigned long fp, sp, pc;
|
||
unsigned long func_start;
|
||
long instr;
|
||
int offset;
|
||
unsigned long thisfp = thisframe->frame;
|
||
|
||
/* get the frame pointer actually sp for a non -g
|
||
* for the routine that called us routine
|
||
*/
|
||
|
||
BTDEBUG("FRAME_CHAIN(%x)\n",thisframe);
|
||
|
||
if ( !read_memory_integer (thisframe->frame,4))
|
||
{
|
||
return (0);
|
||
}
|
||
|
||
if( ! g_routine(thisframe->pc) ){
|
||
BTDEBUG( "non g at %x\n",thisframe->pc);
|
||
caller_pc(thisframe->pc,thisframe->sp,&pc,&fp);
|
||
BTDEBUG("caller_pc returned %x %x \n",pc,fp);
|
||
return(fp);
|
||
|
||
}/* else a -g routine */
|
||
|
||
|
||
fp = read_memory_integer (thisfp, 4);
|
||
|
||
if (fp < thisfp || fp > STACK_END_ADDR)
|
||
{
|
||
/* handle the Metaware-type pseudo-frame */
|
||
|
||
func_start = get_pc_function_start(thisframe->pc);
|
||
|
||
if (func_start)
|
||
{
|
||
|
||
instr = adj_read_memory_integer (func_start);
|
||
/* Recognize "addu|adds -X,sp,sp" insn. */
|
||
if ((instr & 0xEFFF0000) == 0x84420000)
|
||
offset = SIGN_EXT16(instr & 0x0000FFFF);
|
||
|
||
}
|
||
|
||
fp = 0;
|
||
if (offset < 0)
|
||
fp = thisfp - offset;
|
||
}
|
||
BTDEBUG("frame_chain returned %d\n",fp);
|
||
return(fp);
|
||
}
|
||
|
||
/* get the pc and frame pointer (or sp )
|
||
* for the routine that called us
|
||
* when we (this_pc) is not within a -g routine
|
||
* if caller is non g we return sp for fp
|
||
*/
|
||
|
||
/* note this is written for Metaware version R2.1d compiler */
|
||
|
||
caller_pc(this_pc,this_sp,to_pc,to_fp)
|
||
int * to_pc, *to_fp;
|
||
unsigned long this_pc,this_sp;
|
||
{
|
||
unsigned long func_start;
|
||
int sp_offset,offset;
|
||
unsigned long sp,pc,fp,instr;
|
||
|
||
BTDEBUG("caller_pc %x sp = %x\n",this_pc,this_sp);
|
||
|
||
func_start = get_pc_function_start(this_pc);
|
||
|
||
BTDEBUG("caller_pc func_start %x\n", func_start);
|
||
|
||
if (func_start)
|
||
{
|
||
if( has_a_frame(func_start) ){
|
||
|
||
BTDEBUG("has_a_frame\n");
|
||
|
||
/* if our caller has a preamble and
|
||
* declares space for a stack frame
|
||
* then we must work to find our return address
|
||
*/
|
||
instr = adj_read_memory_integer (func_start);
|
||
/* Recognize "addu|adds -X,sp,sp" insn. */
|
||
|
||
if ((instr & 0xEFFF0000) == 0x84420000)
|
||
sp_offset=SIGN_EXT16(instr&0x0000FFFF);
|
||
}
|
||
else { printf("error frame_chain\n");return(0); }
|
||
|
||
BTDEBUG("sp_offset = %d %x\n",sp_offset,sp_offset);
|
||
|
||
offset = find_r1_offset(func_start);
|
||
|
||
if( offset < 0 ){
|
||
printf("cant find return address for routine at %x\n",
|
||
func_start);
|
||
return(0);
|
||
}
|
||
pc = read_memory_integer(this_sp+offset,4);
|
||
sp= this_sp - sp_offset;
|
||
|
||
BTDEBUG("callers pc = %x sp = %x\n",pc,sp);
|
||
|
||
/* our caller a -g routine ?
|
||
* if he is we have to find his real fp
|
||
* else provide the sp as his fp
|
||
*/
|
||
|
||
if( g_routine(pc) ){
|
||
|
||
BTDEBUG("caller_a_g\n");
|
||
|
||
if( ! (offset = find_fp_offset(func_start)) ) {
|
||
printf("error fp_offset\n");
|
||
return(0);
|
||
}
|
||
BTDEBUG("offset = %x %d\n",offset,offset);
|
||
|
||
fp = read_memory_integer(this_sp+offset,4);
|
||
*to_pc = CLEAN_PC(pc);
|
||
*to_fp = fp;
|
||
return(1);
|
||
}else
|
||
*to_pc = CLEAN_PC(pc);
|
||
*to_fp = sp;
|
||
return(1);
|
||
} else {
|
||
/* pc = read_register(RP_REGNUM); */
|
||
pc = 0;
|
||
BTDEBUG("read_register pc %x\n",pc);
|
||
if( g_routine(pc) ){
|
||
|
||
*to_pc = CLEAN_PC(pc);
|
||
*to_fp = read_register(FP_REGNUM);
|
||
return(1);
|
||
}else {
|
||
*to_pc = CLEAN_PC(pc);
|
||
*to_fp = this_sp;
|
||
return(1);
|
||
}
|
||
}
|
||
}
|
||
|
||
int outside_startup_file();
|
||
|
||
/* get the PC of the caller */
|
||
frame_saved_pc(frame_struct)
|
||
FRAME frame_struct;
|
||
{
|
||
unsigned long frame;
|
||
unsigned long pc;
|
||
unsigned long pc1;
|
||
unsigned long sp;
|
||
|
||
CORE_ADDR fp;
|
||
|
||
CORE_ADDR rp;
|
||
|
||
frame = frame_struct->frame;
|
||
pc = frame_struct->pc;
|
||
BTDEBUG("frame_saved_pc input: frame %x, pc %x, sp %x ",
|
||
frame, pc, sp);
|
||
|
||
/* First see if this is the current frame. If it is, return the value in r1,
|
||
as it may not have been stored */
|
||
|
||
fp = read_register(FP_REGNUM);
|
||
|
||
/* check to see if we are in an entry sequence, where the return pointer has not yet been stored */
|
||
if (fp == frame && no_stored_rp(pc))
|
||
{
|
||
pc = read_register(RP_REGNUM);
|
||
frame_struct->rp = pc;
|
||
}
|
||
else if( ! g_routine(pc) )
|
||
{
|
||
caller_pc(pc,sp,&pc,&frame);
|
||
}
|
||
else
|
||
{
|
||
|
||
pc = read_memory_integer (frame + 4, 4);
|
||
|
||
if (!outside_startup_file(pc))
|
||
{
|
||
|
||
BTDEBUG("pc %x outside startup file \n",pc);
|
||
|
||
pc1 = read_memory_integer (frame, 4);
|
||
|
||
if (outside_startup_file(pc1))
|
||
pc = pc1;
|
||
else
|
||
pc = 0;
|
||
}
|
||
}
|
||
BTDEBUG(" returning pc %x\n", CLEAN_PC(pc));
|
||
return(CLEAN_PC(pc));
|
||
|
||
}
|
||
|
||
/* Pass arguments to a function in the inferior process - ABI compliant
|
||
*/
|
||
|
||
pass_function_arguments(args, nargs, struct_return)
|
||
value *args;
|
||
int nargs;
|
||
int struct_return;
|
||
{
|
||
int ireg = (struct_return) ? 17 : 16;
|
||
int freg = FP0_REGNUM + 8;
|
||
int i;
|
||
struct type *type;
|
||
value arg;
|
||
long tmp;
|
||
value value_arg_coerce();
|
||
|
||
|
||
for (i = 0; i < nargs; i++)
|
||
{
|
||
arg = value_arg_coerce(args[i]);
|
||
type = VALUE_TYPE(arg);
|
||
if (type == builtin_type_double)
|
||
{
|
||
write_register_bytes(REGISTER_BYTE(freg), VALUE_CONTENTS(arg), 8);
|
||
freg += 2;
|
||
}
|
||
else
|
||
{
|
||
bcopy(VALUE_CONTENTS(arg), &tmp, 4);
|
||
write_register(ireg, tmp);
|
||
ireg++;
|
||
}
|
||
}
|
||
if (ireg >= 28 || freg >= FP0_REGNUM + 16)
|
||
error("Too many arguments to function");
|
||
}
|
||
|
||
|
||
#define SPACES " "
|
||
#define P_SPACES " "
|
||
#define BYTE 0xff
|
||
|
||
int screen_lines=24;
|
||
|
||
char *spec_reg[] = {
|
||
"fsr", "db", "dirbase", "fir", "psr", "epsr",
|
||
};
|
||
|
||
char *doro_reg[] = {
|
||
"scp", "cbsp", "pt_cs", "intmsk", "intack",
|
||
};
|
||
#define NREGS 32
|
||
|
||
|
||
get_reg(regno)
|
||
{
|
||
char raw_buffer[32];
|
||
int addr;
|
||
int virtual_buffer;
|
||
|
||
read_relative_register_raw_bytes (regno, raw_buffer);
|
||
REGISTER_CONVERT_TO_VIRTUAL (addr, raw_buffer, &virtual_buffer);
|
||
return(virtual_buffer);
|
||
}
|
||
|
||
int jhdebug = 0;
|
||
/*
|
||
** Figure out address to place next breakpoint. Avoid tricky spots,
|
||
** ie. delayed instruction slots etc.
|
||
** Need to upgrade this later to allow delayed instruction breakpoints
|
||
** with fix-up work done AFTER breakpoint.
|
||
** Note that this routine DOES deal with dual instruction mode
|
||
*/
|
||
#define BIM 0x8008
|
||
|
||
branch_type
|
||
place_brk (addr, mode, brk)
|
||
CORE_ADDR addr;
|
||
int mode;
|
||
struct breakpoint *brk;
|
||
{
|
||
long nextadr, prevadr, instr;
|
||
int val = not_branch;
|
||
long offset; /* Must be signed for sign-extend */
|
||
extern char registers[];
|
||
prevadr = nextadr = 0;
|
||
|
||
brk->address1 = 0;
|
||
|
||
if (mode == SINGLE_STEP_MODE)
|
||
{
|
||
if (INDIM || ENDIM)
|
||
{
|
||
nextadr = brk->address = (addr + 8);
|
||
instr = adj_read_memory_integer ((addr + 4));
|
||
brk->mode = DIM;
|
||
}
|
||
else
|
||
{
|
||
nextadr = brk->address = (addr + 4);
|
||
instr = adj_read_memory_integer (addr);
|
||
if (STDIM)
|
||
brk->mode = DIM;
|
||
else
|
||
brk->mode = SIM;
|
||
}
|
||
|
||
|
||
/*
|
||
** For br/call one more sequential instruction gets executed and then we
|
||
** continue at the current addr + offset. We are definitely going to
|
||
** the dest. We are NOT allowed to place a breakpoint in the "delay"
|
||
** slot - (the next sequential instruction) so we only place 1 breakpoint
|
||
** at the destination.
|
||
** For the bc/bnc the next instruction executed is EITHER the next sequential
|
||
** or the destination of the branch, we therefore place 2 breakpoints one
|
||
** at each location.
|
||
** For the bc.t/bnc.t either 1 more sequential instruction is performed
|
||
** followed by a branch (like br/call) OR we skip the sequential
|
||
** instruction and keep going. We therefore place a breakpoint at the
|
||
** destination of the branch AND the second sequential instruction after
|
||
** the branch. Again a breakpoint is NOT allowed in the "delay slot"
|
||
*/
|
||
if ((instr & 0xE0000000) == 0x60000000 && /* CTRL format */
|
||
(instr & 0xF8000000) != 0x60000000) /* not pfld.y */
|
||
{
|
||
if ((instr & 0xF8000000) == 0x68000000) /* br or call */
|
||
val = uncond_d;
|
||
else if ((instr & 0xF4000000) == 0x74000000) /* bc.t/bnc.t */
|
||
val = cond_d;
|
||
else if ((instr & 0xF4000000) == 0x70000000) /* bc or bnc */
|
||
val = cond;
|
||
offset = (instr & 0x03ffffff);
|
||
if (offset & 0x02000000) /*?sign extend*/
|
||
offset |= 0xFC000000;
|
||
if (val == uncond_d) /* br/call*/
|
||
prevadr = 0;
|
||
else if (val == cond_d) /* bc.t/bnc.t */
|
||
{
|
||
if ((INDIM) && !(ENDIM))
|
||
prevadr = nextadr + 8;
|
||
else
|
||
prevadr = nextadr + 4;
|
||
} else { /* bc /bnc */
|
||
if ((INDIM) && !(ENDIM))
|
||
prevadr = nextadr;
|
||
else
|
||
prevadr = nextadr;
|
||
}
|
||
nextadr += (offset << 2);
|
||
}
|
||
/*
|
||
** We treat the bri/calli the same way as the br/call case.
|
||
*/
|
||
else if ((instr & 0xFC00003F) == 0x4C000002 || /* calli */
|
||
(instr & 0xFC000000) == 0x40000000) /* bri */
|
||
{
|
||
val = uncond_d;
|
||
offset = ((instr & 0x0000F800) >> 11);
|
||
nextadr = (read_register(offset + R0) & 0xFFFFFFFC);
|
||
prevadr = 0;
|
||
}
|
||
/*
|
||
** We treat the bte/btne the same way as the bc/bnc case.
|
||
*/
|
||
else if ((instr & 0xF0000000) == 0x50000000) /* bte/btne */
|
||
{
|
||
val = cond;
|
||
offset = SIGN_EXT16(((instr & 0x001F0000) >> 5) |
|
||
(instr & 0x000007FF));
|
||
if ((INDIM) && !(ENDIM))
|
||
prevadr = nextadr;
|
||
else
|
||
prevadr = nextadr;
|
||
|
||
nextadr += (offset << 2);
|
||
}
|
||
/*
|
||
** We treat the bte/btne the same way as the bc/bnc case.
|
||
** With the caveat that the 2 breakpoints may turn out to be at the same
|
||
** address in which case we ignore one of them.
|
||
*/
|
||
else if ((instr & 0xFC000000) == 0xB4000000) /* bla */
|
||
{
|
||
val = cond_d;
|
||
offset = SIGN_EXT16(((instr & 0x001F0000) >> 5) |
|
||
(instr & 0x000007FF));
|
||
if ((INDIM) && !(ENDIM))
|
||
{
|
||
prevadr = nextadr + 8;
|
||
} else {
|
||
prevadr = nextadr + 4;
|
||
}
|
||
nextadr += (offset << 2);
|
||
if (prevadr == nextadr) prevadr = 0;
|
||
}
|
||
} else {
|
||
int adjust = 0;
|
||
|
||
nextadr = addr;
|
||
|
||
if (ISDIM(FOPADR(addr)))
|
||
{
|
||
if (ISDIM(FOPADR(nextadr-8)))
|
||
{
|
||
instr = adj_read_memory_integer(CORADR(addr-8));
|
||
brk->mode = DIM;
|
||
} else {
|
||
instr = adj_read_memory_integer(addr-4);
|
||
brk->mode = RIM;
|
||
}
|
||
} else {
|
||
if (ISDIM(addr-4))
|
||
{
|
||
instr = adj_read_memory_integer(addr-4);
|
||
brk->mode = BIM;
|
||
} else {
|
||
instr = adj_read_memory_integer (addr-4);
|
||
brk->mode = SIM;
|
||
}
|
||
}
|
||
|
||
/* examine the PREVIOUS instruction to determine if we are in a branch delay
|
||
slot. If we are, dont set a break here -- set it on the previous instruction.
|
||
This code also accounts for dual instruction mode */
|
||
if ((instr & 0xE0000000) == 0x60000000 &&
|
||
(instr & 0xF8000000) != 0x60000000) /* not pfld.y */
|
||
{
|
||
adjust++;
|
||
/* br /call */
|
||
/* bc /bnc */
|
||
/* bc.t /bnc.t*/
|
||
if ((instr & 0xF8000000) == 0x68000000) /* br or call */
|
||
printf(" Breakpoint adjusted to avoid br/call delay slot and multiple breakpoints\n");
|
||
|
||
if ((instr & 0xF4000000) == 0x74000000) /* bc.t or bnc.t */
|
||
printf(" Breakpoint adjusted to avoid bc.t/bnc.t delay slot and multiple breakpoints\n");
|
||
/* it IS really OK to set a break on the instruction AFTER the conditional branch
|
||
-- it DOESN't have a delay slot */
|
||
if ((instr & 0xF4000000) == 0x70000000) /* bc / bnc */
|
||
/* printf(" Breakpoint adjusted to avoid bc/bnc delay slot and multiple breakpoints\n"); */
|
||
adjust = 0;
|
||
} else if
|
||
((instr & 0xFC00003F) == 0x4C000002 || /* bri/ calli */
|
||
(instr & 0xFC000000) == 0x40000000)
|
||
{
|
||
adjust++;
|
||
printf(" Breakpoint adjusted to avoid calli/bri delay slot and multiple breakpoints\n");
|
||
} else if
|
||
((instr & 0xF0000000) == 0x50000000) /* bte - btne */
|
||
{
|
||
/* it's OK to set a break here -- we are NOT in aa branch delay slot */
|
||
/*
|
||
adjust++;
|
||
printf(" Breakpoint adjusted to avoid bte/btne multiple breakpoints\n");
|
||
*/
|
||
adjust = 0;
|
||
} else if
|
||
((instr & 0xFC000000) == 0xB4000000)
|
||
{
|
||
adjust++;
|
||
printf(" Breakpoint adjusted to avoid bla delay slot and multiple breakpoints\n");
|
||
}
|
||
if (adjust)
|
||
{
|
||
if (brk->mode == DIM)
|
||
{
|
||
nextadr -= 8;
|
||
nextadr = CORADR(nextadr);
|
||
}
|
||
else
|
||
nextadr -=4;
|
||
}
|
||
|
||
}
|
||
|
||
if (brk->mode == RIM)
|
||
brk->mode = DIM;
|
||
if (brk->mode == BIM)
|
||
brk->mode = SIM;
|
||
|
||
if (nextadr)
|
||
{
|
||
if (brk->mode == DIM)
|
||
{
|
||
brk->act_addr[0] = CORADR(nextadr);
|
||
brk->act_addr[1] = FOPADR(nextadr);
|
||
} else {
|
||
brk->act_addr[0] = nextadr;
|
||
brk->act_addr[1] = 0;
|
||
}
|
||
}
|
||
|
||
if (prevadr)
|
||
{
|
||
brk->address1 = prevadr;
|
||
if (brk->mode == DIM)
|
||
{
|
||
brk->act_addr[2] = CORADR(prevadr);
|
||
brk->act_addr[3] = FOPADR(prevadr);
|
||
} else {
|
||
brk->act_addr[2] = prevadr;
|
||
brk->act_addr[3] = 0;
|
||
}
|
||
} else {
|
||
brk->act_addr[2] = brk->act_addr[3] = 0;
|
||
}
|
||
return val;
|
||
}
|
||
|
||
/*
|
||
** Figure out whether we are in a delayed slot and if so then take necessary
|
||
** action to resume properly - remember trap pre-empts instruction
|
||
*/
|
||
int
|
||
wasabranch (addr, nextpc, ss)
|
||
CORE_ADDR addr, *nextpc;
|
||
int ss;
|
||
{
|
||
long nextadr, instr;
|
||
int val = not_branch;
|
||
long offset; /* Must be signed for sign-extend */
|
||
|
||
if (ss)
|
||
{
|
||
if (INDIM)
|
||
{
|
||
nextadr = CORADR((int)(addr + 8));
|
||
instr = adj_read_memory_integer (CORADR(addr));
|
||
}
|
||
else
|
||
{
|
||
nextadr = addr + 4;
|
||
instr = adj_read_memory_integer (addr);
|
||
}
|
||
} else {
|
||
if (ISDIM(addr))
|
||
{
|
||
nextadr = CORADR(addr);
|
||
instr = adj_read_memory_integer (nextadr);
|
||
}
|
||
else
|
||
{
|
||
nextadr = addr;
|
||
instr = adj_read_memory_integer (addr);
|
||
}
|
||
}
|
||
|
||
|
||
if ((instr & 0xE0000000) == 0x60000000 && /* CTRL format */
|
||
(instr & 0xF8000000) != 0x60000000) /* not pfld.y */
|
||
{
|
||
if ((instr & 0xF8000000) == 0x68000000) /* br or call */
|
||
val = uncond_d;
|
||
else if ((instr & 0xF4000000) == 0x74000000) /* bc.t or bnc.t */
|
||
val = cond_d;
|
||
else if ((instr & 0xF4000000) == 0x70000000) /* bc or bnc */
|
||
val = cond;
|
||
|
||
offset = (instr & 0x03ffffff);
|
||
if (offset & 0x02000000) /* sign extend? */
|
||
offset |= 0xFC000000;
|
||
nextadr += (offset << 2);
|
||
}
|
||
else if ((instr & 0xFC00003F) == 0x4C000002 || /* calli */
|
||
(instr & 0xFC000000) == 0x40000000) /* bri */
|
||
{
|
||
if (ss)
|
||
{
|
||
val = uncond_d;
|
||
offset = ((instr & 0x0000F800) >> 11);
|
||
nextadr = (read_register(offset) & 0xFFFFFFFC);
|
||
} else {
|
||
val = uncond_d;
|
||
}
|
||
}
|
||
else if ((instr & 0xF0000000) == 0x50000000) /* bte or btne */
|
||
{
|
||
val = cond;
|
||
|
||
offset = SIGN_EXT16(((instr & 0x001F0000) >> 5) | (instr & 0x000007FF));
|
||
nextadr += (offset << 2);
|
||
}
|
||
else if ((instr & 0xFC000000) == 0xB4000000) /* bla */
|
||
{
|
||
val = cond_d;
|
||
|
||
offset = SIGN_EXT16(((instr & 0x001F0000) >> 5) | (instr & 0x000007FF));
|
||
nextadr += (offset << 2);
|
||
}
|
||
|
||
*nextpc = nextadr;
|
||
return val;
|
||
}
|
||
|
||
extern char registers[];
|
||
|
||
i860_do_registers_info(regnum,fpregs)
|
||
int regnum;
|
||
int fpregs;
|
||
{
|
||
register int i;
|
||
unsigned int val;
|
||
unsigned int j,k;
|
||
|
||
if (regnum == -1)
|
||
printf_filtered (
|
||
"Register Contents (relative to selected stack frame)\n\n");
|
||
|
||
if (regnum != -1) /* print one register */
|
||
{
|
||
if ((regnum >=F0 ) && (regnum <= F31))
|
||
bcopy (®isters[ADJ_FREG(regnum)<<2], &val, sizeof (long));
|
||
else
|
||
bcopy (®isters[regnum<<2], &val, sizeof (long));
|
||
printf("%-4s 0x%08x\t", reg_names[regnum], val);
|
||
printf("\n\t"); fflush(stdout);
|
||
}
|
||
else /* print all registers */
|
||
{
|
||
|
||
printf("\n Control/Status Registers :- \n\t");
|
||
for (j=0; j<=DB; j++)
|
||
{
|
||
bcopy (®isters[j<<2], &val, sizeof (long));
|
||
printf("%-4s 0x%08x\t", reg_names[j], val);
|
||
}
|
||
printf("\n\t"); fflush(stdout);
|
||
|
||
/* EPSR */
|
||
bcopy (®isters[EPSR<<2], &val, sizeof (long));
|
||
printf("%-4s 0x%08x\t", reg_names[EPSR], val);
|
||
|
||
/* FSR */
|
||
bcopy (®isters[FSR<<2], &val, sizeof (long));
|
||
printf("%-4s 0x%08x\t", reg_names[FSR], val);
|
||
|
||
/* CCR */
|
||
bcopy (®isters[CCR<<2], &val, sizeof (long));
|
||
printf("%-4s 0x%08x\t", reg_names[CCR], val);
|
||
/* BEAR*/
|
||
bcopy (®isters[BEAR<<2], &val, sizeof (long));
|
||
printf("%-4s 0x%08x\t", reg_names[BEAR], val);
|
||
|
||
|
||
#ifdef JIM_ADD_PRIV
|
||
for (j=P0; j<=P3; j++)
|
||
{
|
||
bcopy (®isters[j<<2], &val, sizeof (long));
|
||
printf("%-4s 0x%08x\t", reg_names[j], val);
|
||
}
|
||
#endif
|
||
|
||
printf("\n Integer Registers :- \n\t");
|
||
for (j=R0; j<=R31; j++)
|
||
{
|
||
if (j != IREGS && (j % 4 == 0))
|
||
{
|
||
printf("\n\t"); fflush(stdout);
|
||
}
|
||
bcopy (®isters[j<<2], &val, sizeof (long));
|
||
printf("%-4s 0x%08x\t", reg_names[j], val);
|
||
}
|
||
|
||
printf("\n Floating Registers :- \n\t");
|
||
for (j=F0; j<=F31; j++)
|
||
{
|
||
if (j != FREGS && (j % 4 == 0))
|
||
{
|
||
printf("\n\t"); fflush(stdout);
|
||
}
|
||
bcopy (®isters[ADJ_FREG(j)<<2], &val, sizeof (long));
|
||
printf("%-4s 0x%08x\t", reg_names[j], val);
|
||
}
|
||
|
||
printf("\n Special Registers :- \n\t");
|
||
for (j=SPC_KI; j<=SPC_MERGE; j+=2)
|
||
{
|
||
unsigned int valh;
|
||
if (j == SPC_T)
|
||
{
|
||
printf("\n\t"); fflush(stdout);
|
||
}
|
||
bcopy (®isters[j<<2], &val, sizeof (long));
|
||
bcopy (®isters[(j+1)<<2], &valh, sizeof (long));
|
||
printf("%-6s 0x%08x %08x\t", reg_names[j], val,valh);
|
||
}
|
||
|
||
printf("\n Graphics Pipeline :- \n");
|
||
{
|
||
unsigned int valh, val2,val3;
|
||
j = PSV_I1;
|
||
bcopy (®isters[j<<2], &val, sizeof (long));
|
||
bcopy (®isters[(j+1)<<2], &valh, sizeof (long));
|
||
printf("\t\t\t%-8s 0x%08x %08x \n", reg_names[j], val,valh);
|
||
}
|
||
|
||
printf(" Memory Load Pipeline :- \n");
|
||
for (j=PSV_L1; j<=PSV_L3; j+=4)
|
||
{
|
||
unsigned int valh, val2,val3;
|
||
bcopy (®isters[j<<2], &val, sizeof (long));
|
||
bcopy (®isters[(j+1)<<2], &valh, sizeof (long));
|
||
bcopy (®isters[(j+2)<<2], &val2, sizeof (long));
|
||
bcopy (®isters[(j+3)<<2], &val3, sizeof (long));
|
||
printf("\t\t%-8s 0x%08x %08x %08x %08x\n", reg_names[j],
|
||
val,valh,val2,val3);
|
||
}
|
||
|
||
printf("\n Adder Pipeline :-\t\tMultiplier Pipeline :-\t\tFSR results :-\n");
|
||
for (i=PSV_FSR1,j=PSV_A1,k=PSV_M1; j<=PSV_A3; i++,j+=2,k+=2)
|
||
{
|
||
unsigned int valh,val2,val3,val4;
|
||
bcopy (®isters[i<<2], &val4, sizeof (long));
|
||
bcopy (®isters[j<<2], &val, sizeof (long));
|
||
bcopy (®isters[(j+1)<<2], &valh, sizeof (long));
|
||
bcopy (®isters[k<<2], &val2, sizeof (long));
|
||
bcopy (®isters[(k+1)<<2], &val3, sizeof (long));
|
||
printf(" %-4s 0x%08x %08x\t", reg_names[j], val,valh);
|
||
printf("%-4s 0x%08x %08x\t", reg_names[k], val2,val3);
|
||
printf("%-4s 0x%08x\n", reg_names[i], val4);
|
||
}
|
||
|
||
}
|
||
|
||
|
||
}
|
||
|
||
int has_stored_r1(CORE_ADDR prologue_start, CORE_ADDR prologue_end)
|
||
{
|
||
long instr;
|
||
CORE_ADDR addr;
|
||
|
||
BTDEBUG("has_stored_r1, prologue_start %x, prologue_end %x\n",
|
||
prologue_start, prologue_end);
|
||
|
||
for (addr = prologue_start; addr <= prologue_end; addr += 4)
|
||
{
|
||
|
||
instr = adj_read_memory_integer (addr);
|
||
if ((instr & 0xFFE0F801) == 0x1C400801 /* st.l r1,X(sp) */
|
||
|| (instr & 0xFFE0F801) == 0x1C600801) /* st.l r1,X(fp) */
|
||
return (1);
|
||
}
|
||
return 0;
|
||
}
|
||
|
||
/* This function returns 1 if there is no stored r1, 0 otherwise.
|
||
The function returns 1 if the pc is in a function prologue,
|
||
or the function prologue didn't save the return pointer in
|
||
the stack frame, 0 otherwise */
|
||
|
||
int no_stored_rp(CORE_ADDR pc)
|
||
{
|
||
CORE_ADDR func_start, prologue_end;
|
||
|
||
func_start = get_pc_function_start(pc);
|
||
if (func_start)
|
||
{
|
||
prologue_end = func_start;
|
||
SKIP_PROLOGUE(prologue_end);
|
||
if ( (pc >= func_start) && (pc <= prologue_end))
|
||
{
|
||
BTDEBUG("no_stored_rp: pc %x is in prologue \n",pc);
|
||
return 1;
|
||
}
|
||
/* otherwise, see if the entry sequence stored the return pointer.
|
||
If it didn't, return 1 */
|
||
/* Some procedures , at least, store the return pointer AFTER the prologue sequence! */
|
||
if (!has_stored_r1(func_start, pc))
|
||
{
|
||
BTDEBUG("no_stored_rp, for pc %x, prologue didn't store r1\n",pc);
|
||
return 1;
|
||
}
|
||
}
|
||
BTDEBUG("no_stored_rp for pc %x return pointer was stored \n", pc);
|
||
|
||
return 0;
|
||
}
|
||
|
||
/* The following set of routines was adapted from existing code previously
|
||
in an i860-specific version of breakpoint.c by Peggy Fieland
|
||
(Margaret_Fieland@vos.stratus.com) */
|
||
unsigned int dbrkval, dbrkmod;
|
||
void i860_dbrk_breakpoint()
|
||
{
|
||
BTDEBUG("i860_dbrk_breakpoint was called , dbrkval %x\n", dbrkval);
|
||
|
||
if (dbrkval)
|
||
{
|
||
*(int *)®isters[DB<<2] = dbrkval;
|
||
}
|
||
else
|
||
{
|
||
*(int *)®isters[DB<<2] = 0;
|
||
}
|
||
|
||
*(int *)®isters[PSR<<2] &= ~3;
|
||
*(int *)®isters[PSR<<2] |= dbrkmod;
|
||
|
||
store_inferior_registers(DB);
|
||
store_inferior_registers(PSR);
|
||
|
||
}
|
||
|
||
void
|
||
d_ro_break_command(arg)
|
||
char *arg;
|
||
{
|
||
dbrkval = strtoul(arg, NULL, 0);
|
||
dbrkmod = 0x01;
|
||
printf(" ro_dbreak - %x %x\n", dbrkval, dbrkmod);
|
||
}
|
||
|
||
void
|
||
d_wo_break_command(arg)
|
||
char *arg;
|
||
{
|
||
dbrkval = strtoul(arg, NULL, 0);
|
||
dbrkmod = 0x02;
|
||
printf(" wo_dbreak - %x %x\n", dbrkval, dbrkmod);
|
||
}
|
||
|
||
void
|
||
d_rw_break_command(arg)
|
||
char *arg;
|
||
{
|
||
dbrkval = strtoul(arg, NULL, 0);
|
||
dbrkmod = 0x03;
|
||
printf(" rw_dbreak - %x %x\n", dbrkval, dbrkmod);
|
||
}
|
||
|
||
void
|
||
clear_dbreak()
|
||
{
|
||
dbrkval = 0;
|
||
dbrkmod = 0;
|
||
}
|
||
|
||
void
|
||
i860_init_breakpoints()
|
||
{
|
||
dbrkval = dbrkmod = 0;
|
||
add_com ("dbro", class_breakpoint, d_ro_break_command,
|
||
"Set a data breakpoint READ ONLY, 32-bit data element.");
|
||
add_com ("dbwo", class_breakpoint, d_wo_break_command,
|
||
"Set a data breakpoint WRITE ONLY, 32-bit data element.");
|
||
add_com ("dbrw", class_breakpoint, d_rw_break_command,
|
||
"Set a data breakpoint READ/WRITE, 32-bit data element.");
|
||
add_com ("dclear", class_breakpoint, clear_dbreak,
|
||
"clear the current data breakpoint.");
|
||
add_com_alias ("dc", "dclear", class_breakpoint, 1);
|
||
|
||
}
|
||
|
||
int i860_insert_breakpoint(b)
|
||
struct breakpoint *b;
|
||
{
|
||
int val;
|
||
|
||
place_brk( b->address, BREAK_MODE, b );
|
||
if (b->mode == DIM)
|
||
{
|
||
|
||
adj_read_memory (b->act_addr[0], &b->shadow_contents[0], 4);
|
||
val = adj_write_memory (b->act_addr[0], break_insn, 4);
|
||
if (val) return val;
|
||
adj_read_memory (b->act_addr[1], &b->shadow_contents[1], 4);
|
||
/* val = adj_write_memory (b->act_addr[1], float_insn, 4); */
|
||
if (val) return val;
|
||
}
|
||
else
|
||
{
|
||
adj_read_memory (b->act_addr[0], &b->shadow_contents[0], 4);
|
||
val = adj_write_memory (b->act_addr[0], break_insn, 4);
|
||
}
|
||
if (b->address1)
|
||
{
|
||
if (b->mode == DIM)
|
||
{
|
||
|
||
adj_read_memory (b->act_addr[2], &b->shadow_contents[2], 4);
|
||
val = adj_write_memory (b->act_addr[2], break_insn, 4);
|
||
if (val) return val;
|
||
adj_read_memory (b->act_addr[3], &b->shadow_contents[3], 4);
|
||
/* val = adj_write_memory (b->act_addr[3], float_insn, 4); */
|
||
if (val) return val;
|
||
}
|
||
else
|
||
{
|
||
adj_read_memory (b->act_addr[2], &b->shadow_contents[0], 4);
|
||
val = adj_write_memory (b->act_addr[2], break_insn, 4);
|
||
}
|
||
}
|
||
if (val)
|
||
return val;
|
||
BTDEBUG("Inserted breakpoint at 0x%x, shadow 0x%x, 0x%x.\n",
|
||
b->address, b->shadow_contents[0], b->shadow_contents[1]);
|
||
b->inserted = 1;
|
||
return 0;
|
||
}
|
||
|
||
int i860_remove_breakpoint(b)
|
||
struct breakpoint *b;
|
||
{
|
||
int val;
|
||
|
||
if (b->inserted)
|
||
{
|
||
if (b->mode == DIM)
|
||
{
|
||
val =adj_write_memory (b->act_addr[0], &(b->shadow_contents[0]), 4);
|
||
val =adj_write_memory (b->act_addr[1], &(b->shadow_contents[1]), 4);
|
||
if (b->address1)
|
||
{
|
||
val =adj_write_memory (b->act_addr[2], &(b->shadow_contents[2]), 4);
|
||
val =adj_write_memory (b->act_addr[3], &(b->shadow_contents[3]), 4);
|
||
}
|
||
}
|
||
else
|
||
{
|
||
val =adj_write_memory (b->act_addr[0], b->shadow_contents, 4);
|
||
if (b->address1)
|
||
{
|
||
val =adj_write_memory (b->act_addr[2], b->shadow_contents, 4);
|
||
}
|
||
}
|
||
if (val)
|
||
return val;
|
||
b->inserted = 0;
|
||
BTDEBUG( "Removed breakpoint at 0x%x, shadow 0x%x, 0x%x.\n",
|
||
b->address, b->shadow_contents[0], b->shadow_contents[1]);
|
||
}
|
||
|
||
return 0;
|
||
|
||
|
||
}
|
||
|
||
|
||
#ifdef USE_PROC_FS /* Target dependent support for /proc */
|
||
|
||
#include <sys/procfs.h>
|
||
|
||
/* The following routines were added by Peggy Fieland (Margaret_Fieland@vos.stratus.com)
|
||
They were adapted from the m-68k versions of the routines .*/
|
||
|
||
/* Given a pointer to a floating point register set in /proc format
|
||
(fpregset_t *), unpack the register contents and supply them as gdb's
|
||
idea of the current floating point register values. */
|
||
|
||
void
|
||
supply_fpregset (fpregsetp)
|
||
fpregset_t *fpregsetp;
|
||
{
|
||
register int regno;
|
||
|
||
BTDEBUG("supply_fregset called \n");
|
||
|
||
for (regno = F0 ; regno <= F31 ; regno++)
|
||
{
|
||
supply_register (regno, (char *) &(fpregsetp -> fpu.r_freg[regno-F0]));
|
||
}
|
||
}
|
||
|
||
/* Given a pointer to a floating point register set in /proc format
|
||
(fpregset_t *), update the register specified by REGNO from gdb's idea
|
||
of the current floating point register set. If REGNO is -1, update
|
||
them all. */
|
||
|
||
void
|
||
fill_fpregset (fpregsetp, regno)
|
||
fpregset_t *fpregsetp;
|
||
int regno;
|
||
{
|
||
int regi;
|
||
char *to;
|
||
char *from;
|
||
extern char registers[];
|
||
BTDEBUG("fill_fregset regno %d\n",regno);
|
||
|
||
for (regi = F0 ; regi <= F31 ; regi++)
|
||
{
|
||
if ((regno == -1) || (regno == regi))
|
||
{
|
||
from = (char *) ®isters[REGISTER_BYTE (regi)];
|
||
to = (char *) &(fpregsetp -> fpu.r_freg[regi-F0]);
|
||
bcopy (from, to, REGISTER_RAW_SIZE (regno));
|
||
}
|
||
}
|
||
}
|
||
|
||
|
||
/* Given a pointer to a general register set in /proc format (gregset_t *),
|
||
unpack the register contents and supply them as gdb's idea of the current
|
||
register values. */
|
||
|
||
void
|
||
supply_gregset (gregsetp)
|
||
gregset_t *gregsetp;
|
||
{
|
||
register int regno;
|
||
register greg_t *regp = (greg_t *) gregsetp;
|
||
|
||
BTDEBUG("supply_gregset called \n");
|
||
|
||
for (regno = 0 ; regno <= R31 ; regno++)
|
||
{
|
||
supply_register (regno, (char *) (regp + regno));
|
||
}
|
||
}
|
||
|
||
void
|
||
fill_gregset (gregsetp, regno)
|
||
gregset_t *gregsetp;
|
||
int regno;
|
||
{
|
||
int regi;
|
||
extern char registers[];
|
||
register greg_t *regp = (greg_t *) gregsetp;
|
||
BTDEBUG("fill_gregset regno %d \n",regno);
|
||
|
||
for (regi = 0 ; regi <= R31 ; regi++)
|
||
{
|
||
if ((regno == -1) || (regno == regi))
|
||
{
|
||
*(regp + regi) = *(int *) ®isters[REGISTER_BYTE (regi)];
|
||
}
|
||
|
||
}
|
||
}
|
||
#endif
|
||
|