binutils-gdb/sim/arm/armsupp.c
Mike Frysinger 6df01ab8ab sim: switch config.h usage to defs.h
The defs.h header will take care of including the various config.h
headers.  For now, it's just config.h, but we'll add more when we
integrate gnulib in.

This header should be used instead of config.h, and should be the
first include in every .c file.  We won't rely on the old behavior
where we expected files to include the port's sim-main.h which then
includes the common sim-basics.h which then includes config.h.  We
have a ton of code that includes things before sim-main.h, and it
sometimes needs to be that way.  Creating a dedicated header avoids
the ordering mess and implicit inclusion that shows up otherwise.
2021-05-16 22:38:41 -04:00

1705 lines
37 KiB
C

/* armsupp.c -- ARMulator support code: ARM6 Instruction Emulator.
Copyright (C) 1994 Advanced RISC Machines Ltd.
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
the Free Software Foundation; either version 3 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program; if not, see <http://www.gnu.org/licenses/>. */
/* This must come before any other includes. */
#include "defs.h"
#include "armdefs.h"
#include "armemu.h"
#include "ansidecl.h"
#include "libiberty.h"
#include <math.h>
/* Definitions for the support routines. */
static ARMword ModeToBank (ARMword);
static void EnvokeList (ARMul_State *, unsigned long, unsigned long);
struct EventNode
{ /* An event list node. */
unsigned (*func) (ARMul_State *); /* The function to call. */
struct EventNode *next;
};
/* This routine returns the value of a register from a mode. */
ARMword
ARMul_GetReg (ARMul_State * state, unsigned mode, unsigned reg)
{
mode &= MODEBITS;
if (mode != state->Mode)
return (state->RegBank[ModeToBank ((ARMword) mode)][reg]);
else
return (state->Reg[reg]);
}
/* This routine sets the value of a register for a mode. */
void
ARMul_SetReg (ARMul_State * state, unsigned mode, unsigned reg, ARMword value)
{
mode &= MODEBITS;
if (mode != state->Mode)
state->RegBank[ModeToBank ((ARMword) mode)][reg] = value;
else
state->Reg[reg] = value;
}
/* This routine returns the value of the PC, mode independently. */
ARMword
ARMul_GetPC (ARMul_State * state)
{
if (state->Mode > SVC26MODE)
return state->Reg[15];
else
return R15PC;
}
/* This routine returns the value of the PC, mode independently. */
ARMword
ARMul_GetNextPC (ARMul_State * state)
{
if (state->Mode > SVC26MODE)
return state->Reg[15] + isize;
else
return (state->Reg[15] + isize) & R15PCBITS;
}
/* This routine sets the value of the PC. */
void
ARMul_SetPC (ARMul_State * state, ARMword value)
{
if (ARMul_MODE32BIT)
state->Reg[15] = value & PCBITS;
else
state->Reg[15] = R15CCINTMODE | (value & R15PCBITS);
FLUSHPIPE;
}
/* This routine returns the value of register 15, mode independently. */
ARMword
ARMul_GetR15 (ARMul_State * state)
{
if (state->Mode > SVC26MODE)
return (state->Reg[15]);
else
return (R15PC | ECC | ER15INT | EMODE);
}
/* This routine sets the value of Register 15. */
void
ARMul_SetR15 (ARMul_State * state, ARMword value)
{
if (ARMul_MODE32BIT)
state->Reg[15] = value & PCBITS;
else
{
state->Reg[15] = value;
ARMul_R15Altered (state);
}
FLUSHPIPE;
}
/* This routine returns the value of the CPSR. */
ARMword
ARMul_GetCPSR (ARMul_State * state)
{
return (CPSR | state->Cpsr);
}
/* This routine sets the value of the CPSR. */
void
ARMul_SetCPSR (ARMul_State * state, ARMword value)
{
state->Cpsr = value;
ARMul_CPSRAltered (state);
}
/* This routine does all the nasty bits involved in a write to the CPSR,
including updating the register bank, given a MSR instruction. */
void
ARMul_FixCPSR (ARMul_State * state, ARMword instr, ARMword rhs)
{
state->Cpsr = ARMul_GetCPSR (state);
if (state->Mode != USER26MODE
&& state->Mode != USER32MODE)
{
/* In user mode, only write flags. */
if (BIT (16))
SETPSR_C (state->Cpsr, rhs);
if (BIT (17))
SETPSR_X (state->Cpsr, rhs);
if (BIT (18))
SETPSR_S (state->Cpsr, rhs);
}
if (BIT (19))
SETPSR_F (state->Cpsr, rhs);
ARMul_CPSRAltered (state);
}
/* Get an SPSR from the specified mode. */
ARMword
ARMul_GetSPSR (ARMul_State * state, ARMword mode)
{
ARMword bank = ModeToBank (mode & MODEBITS);
if (! BANK_CAN_ACCESS_SPSR (bank))
return ARMul_GetCPSR (state);
return state->Spsr[bank];
}
/* This routine does a write to an SPSR. */
void
ARMul_SetSPSR (ARMul_State * state, ARMword mode, ARMword value)
{
ARMword bank = ModeToBank (mode & MODEBITS);
if (BANK_CAN_ACCESS_SPSR (bank))
state->Spsr[bank] = value;
}
/* This routine does a write to the current SPSR, given an MSR instruction. */
void
ARMul_FixSPSR (ARMul_State * state, ARMword instr, ARMword rhs)
{
if (BANK_CAN_ACCESS_SPSR (state->Bank))
{
if (BIT (16))
SETPSR_C (state->Spsr[state->Bank], rhs);
if (BIT (17))
SETPSR_X (state->Spsr[state->Bank], rhs);
if (BIT (18))
SETPSR_S (state->Spsr[state->Bank], rhs);
if (BIT (19))
SETPSR_F (state->Spsr[state->Bank], rhs);
}
}
/* This routine updates the state of the emulator after the Cpsr has been
changed. Both the processor flags and register bank are updated. */
void
ARMul_CPSRAltered (ARMul_State * state)
{
ARMword oldmode;
if (state->prog32Sig == LOW)
state->Cpsr &= (CCBITS | INTBITS | R15MODEBITS);
oldmode = state->Mode;
if (state->Mode != (state->Cpsr & MODEBITS))
{
state->Mode =
ARMul_SwitchMode (state, state->Mode, state->Cpsr & MODEBITS);
state->NtransSig = (state->Mode & 3) ? HIGH : LOW;
}
state->Cpsr &= ~MODEBITS;
ASSIGNINT (state->Cpsr & INTBITS);
state->Cpsr &= ~INTBITS;
ASSIGNN ((state->Cpsr & NBIT) != 0);
state->Cpsr &= ~NBIT;
ASSIGNZ ((state->Cpsr & ZBIT) != 0);
state->Cpsr &= ~ZBIT;
ASSIGNC ((state->Cpsr & CBIT) != 0);
state->Cpsr &= ~CBIT;
ASSIGNV ((state->Cpsr & VBIT) != 0);
state->Cpsr &= ~VBIT;
ASSIGNS ((state->Cpsr & SBIT) != 0);
state->Cpsr &= ~SBIT;
#ifdef MODET
ASSIGNT ((state->Cpsr & TBIT) != 0);
state->Cpsr &= ~TBIT;
#endif
if (oldmode > SVC26MODE)
{
if (state->Mode <= SVC26MODE)
{
state->Emulate = CHANGEMODE;
state->Reg[15] = ECC | ER15INT | EMODE | R15PC;
}
}
else
{
if (state->Mode > SVC26MODE)
{
state->Emulate = CHANGEMODE;
state->Reg[15] = R15PC;
}
else
state->Reg[15] = ECC | ER15INT | EMODE | R15PC;
}
}
/* This routine updates the state of the emulator after register 15 has
been changed. Both the processor flags and register bank are updated.
This routine should only be called from a 26 bit mode. */
void
ARMul_R15Altered (ARMul_State * state)
{
if (state->Mode != R15MODE)
{
state->Mode = ARMul_SwitchMode (state, state->Mode, R15MODE);
state->NtransSig = (state->Mode & 3) ? HIGH : LOW;
}
if (state->Mode > SVC26MODE)
state->Emulate = CHANGEMODE;
ASSIGNR15INT (R15INT);
ASSIGNN ((state->Reg[15] & NBIT) != 0);
ASSIGNZ ((state->Reg[15] & ZBIT) != 0);
ASSIGNC ((state->Reg[15] & CBIT) != 0);
ASSIGNV ((state->Reg[15] & VBIT) != 0);
}
/* This routine controls the saving and restoring of registers across mode
changes. The regbank matrix is largely unused, only rows 13 and 14 are
used across all modes, 8 to 14 are used for FIQ, all others use the USER
column. It's easier this way. old and new parameter are modes numbers.
Notice the side effect of changing the Bank variable. */
ARMword
ARMul_SwitchMode (ARMul_State * state, ARMword oldmode, ARMword newmode)
{
unsigned i;
ARMword oldbank;
ARMword newbank;
oldbank = ModeToBank (oldmode);
newbank = state->Bank = ModeToBank (newmode);
/* Do we really need to do it? */
if (oldbank != newbank)
{
/* Save away the old registers. */
switch (oldbank)
{
case USERBANK:
case IRQBANK:
case SVCBANK:
case ABORTBANK:
case UNDEFBANK:
if (newbank == FIQBANK)
for (i = 8; i < 13; i++)
state->RegBank[USERBANK][i] = state->Reg[i];
state->RegBank[oldbank][13] = state->Reg[13];
state->RegBank[oldbank][14] = state->Reg[14];
break;
case FIQBANK:
for (i = 8; i < 15; i++)
state->RegBank[FIQBANK][i] = state->Reg[i];
break;
case DUMMYBANK:
for (i = 8; i < 15; i++)
state->RegBank[DUMMYBANK][i] = 0;
break;
default:
abort ();
}
/* Restore the new registers. */
switch (newbank)
{
case USERBANK:
case IRQBANK:
case SVCBANK:
case ABORTBANK:
case UNDEFBANK:
if (oldbank == FIQBANK)
for (i = 8; i < 13; i++)
state->Reg[i] = state->RegBank[USERBANK][i];
state->Reg[13] = state->RegBank[newbank][13];
state->Reg[14] = state->RegBank[newbank][14];
break;
case FIQBANK:
for (i = 8; i < 15; i++)
state->Reg[i] = state->RegBank[FIQBANK][i];
break;
case DUMMYBANK:
for (i = 8; i < 15; i++)
state->Reg[i] = 0;
break;
default:
abort ();
}
}
return newmode;
}
/* Given a processor mode, this routine returns the
register bank that will be accessed in that mode. */
static ARMword
ModeToBank (ARMword mode)
{
static ARMword bankofmode[] =
{
USERBANK, FIQBANK, IRQBANK, SVCBANK,
DUMMYBANK, DUMMYBANK, DUMMYBANK, DUMMYBANK,
DUMMYBANK, DUMMYBANK, DUMMYBANK, DUMMYBANK,
DUMMYBANK, DUMMYBANK, DUMMYBANK, DUMMYBANK,
USERBANK, FIQBANK, IRQBANK, SVCBANK,
DUMMYBANK, DUMMYBANK, DUMMYBANK, ABORTBANK,
DUMMYBANK, DUMMYBANK, DUMMYBANK, UNDEFBANK,
DUMMYBANK, DUMMYBANK, DUMMYBANK, SYSTEMBANK
};
if (mode >= ARRAY_SIZE (bankofmode))
return DUMMYBANK;
return bankofmode[mode];
}
/* Returns the register number of the nth register in a reg list. */
unsigned
ARMul_NthReg (ARMword instr, unsigned number)
{
unsigned bit, upto;
for (bit = 0, upto = 0; upto <= number; bit ++)
if (BIT (bit))
upto ++;
return (bit - 1);
}
/* Assigns the N and Z flags depending on the value of result. */
void
ARMul_NegZero (ARMul_State * state, ARMword result)
{
if (NEG (result))
{
SETN;
CLEARZ;
}
else if (result == 0)
{
CLEARN;
SETZ;
}
else
{
CLEARN;
CLEARZ;
}
}
/* Compute whether an addition of A and B, giving RESULT, overflowed. */
int
AddOverflow (ARMword a, ARMword b, ARMword result)
{
return ((NEG (a) && NEG (b) && POS (result))
|| (POS (a) && POS (b) && NEG (result)));
}
/* Compute whether a subtraction of A and B, giving RESULT, overflowed. */
int
SubOverflow (ARMword a, ARMword b, ARMword result)
{
return ((NEG (a) && POS (b) && POS (result))
|| (POS (a) && NEG (b) && NEG (result)));
}
/* Assigns the C flag after an addition of a and b to give result. */
void
ARMul_AddCarry (ARMul_State * state, ARMword a, ARMword b, ARMword result)
{
ASSIGNC ((NEG (a) && NEG (b)) ||
(NEG (a) && POS (result)) || (NEG (b) && POS (result)));
}
/* Assigns the V flag after an addition of a and b to give result. */
void
ARMul_AddOverflow (ARMul_State * state, ARMword a, ARMword b, ARMword result)
{
ASSIGNV (AddOverflow (a, b, result));
}
/* Assigns the C flag after an subtraction of a and b to give result. */
void
ARMul_SubCarry (ARMul_State * state, ARMword a, ARMword b, ARMword result)
{
ASSIGNC ((NEG (a) && POS (b)) ||
(NEG (a) && POS (result)) || (POS (b) && POS (result)));
}
/* Assigns the V flag after an subtraction of a and b to give result. */
void
ARMul_SubOverflow (ARMul_State * state, ARMword a, ARMword b, ARMword result)
{
ASSIGNV (SubOverflow (a, b, result));
}
static void
handle_VFP_xfer (ARMul_State * state, ARMword instr)
{
if (TOPBITS (28) == NV)
{
fprintf (stderr, "SIM: UNDEFINED VFP instruction\n");
return;
}
if (BITS (25, 27) != 0x6)
{
fprintf (stderr, "SIM: ISE: VFP handler called incorrectly\n");
return;
}
switch (BITS (20, 24))
{
case 0x04:
case 0x05:
{
/* VMOV double precision to/from two ARM registers. */
int vm = BITS (0, 3);
int rt1 = BITS (12, 15);
int rt2 = BITS (16, 19);
/* FIXME: UNPREDICTABLE if rt1 == 15 or rt2 == 15. */
if (BIT (20))
{
/* Transfer to ARM. */
/* FIXME: UPPREDICTABLE if rt1 == rt2. */
state->Reg[rt1] = VFP_dword (vm) & 0xffffffff;
state->Reg[rt2] = VFP_dword (vm) >> 32;
}
else
{
VFP_dword (vm) = state->Reg[rt2];
VFP_dword (vm) <<= 32;
VFP_dword (vm) |= (state->Reg[rt1] & 0xffffffff);
}
return;
}
case 0x08:
case 0x0A:
case 0x0C:
case 0x0E:
{
/* VSTM with PUW=011 or PUW=010. */
int n = BITS (16, 19);
int imm8 = BITS (0, 7);
ARMword address = state->Reg[n];
if (BIT (21))
state->Reg[n] = address + (imm8 << 2);
if (BIT (8))
{
int src = (BIT (22) << 4) | BITS (12, 15);
imm8 >>= 1;
while (imm8--)
{
if (state->bigendSig)
{
ARMul_StoreWordN (state, address, VFP_dword (src) >> 32);
ARMul_StoreWordN (state, address + 4, VFP_dword (src));
}
else
{
ARMul_StoreWordN (state, address, VFP_dword (src));
ARMul_StoreWordN (state, address + 4, VFP_dword (src) >> 32);
}
address += 8;
src += 1;
}
}
else
{
int src = (BITS (12, 15) << 1) | BIT (22);
while (imm8--)
{
ARMul_StoreWordN (state, address, VFP_uword (src));
address += 4;
src += 1;
}
}
}
return;
case 0x10:
case 0x14:
case 0x18:
case 0x1C:
{
/* VSTR */
ARMword imm32 = BITS (0, 7) << 2;
int base = state->Reg[LHSReg];
ARMword address;
int dest;
if (LHSReg == 15)
base = (base + 3) & ~3;
address = base + (BIT (23) ? imm32 : - imm32);
if (CPNum == 10)
{
dest = (DESTReg << 1) + BIT (22);
ARMul_StoreWordN (state, address, VFP_uword (dest));
}
else
{
dest = (BIT (22) << 4) + DESTReg;
if (state->bigendSig)
{
ARMul_StoreWordN (state, address, VFP_dword (dest) >> 32);
ARMul_StoreWordN (state, address + 4, VFP_dword (dest));
}
else
{
ARMul_StoreWordN (state, address, VFP_dword (dest));
ARMul_StoreWordN (state, address + 4, VFP_dword (dest) >> 32);
}
}
}
return;
case 0x12:
case 0x16:
if (BITS (16, 19) == 13)
{
/* VPUSH */
ARMword address = state->Reg[13] - (BITS (0, 7) << 2);
state->Reg[13] = address;
if (BIT (8))
{
int dreg = (BIT (22) << 4) | BITS (12, 15);
int num = BITS (0, 7) >> 1;
while (num--)
{
if (state->bigendSig)
{
ARMul_StoreWordN (state, address, VFP_dword (dreg) >> 32);
ARMul_StoreWordN (state, address + 4, VFP_dword (dreg));
}
else
{
ARMul_StoreWordN (state, address, VFP_dword (dreg));
ARMul_StoreWordN (state, address + 4, VFP_dword (dreg) >> 32);
}
address += 8;
dreg += 1;
}
}
else
{
int sreg = (BITS (12, 15) << 1) | BIT (22);
int num = BITS (0, 7);
while (num--)
{
ARMul_StoreWordN (state, address, VFP_uword (sreg));
address += 4;
sreg += 1;
}
}
}
else if (BITS (9, 11) != 0x5)
break;
else
{
/* VSTM PUW=101 */
int n = BITS (16, 19);
int imm8 = BITS (0, 7);
ARMword address = state->Reg[n] - (imm8 << 2);
state->Reg[n] = address;
if (BIT (8))
{
int src = (BIT (22) << 4) | BITS (12, 15);
imm8 >>= 1;
while (imm8--)
{
if (state->bigendSig)
{
ARMul_StoreWordN (state, address, VFP_dword (src) >> 32);
ARMul_StoreWordN (state, address + 4, VFP_dword (src));
}
else
{
ARMul_StoreWordN (state, address, VFP_dword (src));
ARMul_StoreWordN (state, address + 4, VFP_dword (src) >> 32);
}
address += 8;
src += 1;
}
}
else
{
int src = (BITS (12, 15) << 1) | BIT (22);
while (imm8--)
{
ARMul_StoreWordN (state, address, VFP_uword (src));
address += 4;
src += 1;
}
}
}
return;
case 0x13:
case 0x17:
/* VLDM PUW=101 */
case 0x09:
case 0x0D:
/* VLDM PUW=010 */
{
int n = BITS (16, 19);
int imm8 = BITS (0, 7);
ARMword address = state->Reg[n];
if (BIT (23) == 0)
address -= imm8 << 2;
if (BIT (21))
state->Reg[n] = BIT (23) ? address + (imm8 << 2) : address;
if (BIT (8))
{
int dest = (BIT (22) << 4) | BITS (12, 15);
imm8 >>= 1;
while (imm8--)
{
if (state->bigendSig)
{
VFP_dword (dest) = ARMul_LoadWordN (state, address);
VFP_dword (dest) <<= 32;
VFP_dword (dest) |= ARMul_LoadWordN (state, address + 4);
}
else
{
VFP_dword (dest) = ARMul_LoadWordN (state, address + 4);
VFP_dword (dest) <<= 32;
VFP_dword (dest) |= ARMul_LoadWordN (state, address);
}
if (trace)
fprintf (stderr, " VFP: VLDM: D%d = %g\n", dest, VFP_dval (dest));
address += 8;
dest += 1;
}
}
else
{
int dest = (BITS (12, 15) << 1) | BIT (22);
while (imm8--)
{
VFP_uword (dest) = ARMul_LoadWordN (state, address);
address += 4;
dest += 1;
}
}
}
return;
case 0x0B:
case 0x0F:
if (BITS (16, 19) == 13)
{
/* VPOP */
ARMword address = state->Reg[13];
state->Reg[13] = address + (BITS (0, 7) << 2);
if (BIT (8))
{
int dest = (BIT (22) << 4) | BITS (12, 15);
int num = BITS (0, 7) >> 1;
while (num--)
{
if (state->bigendSig)
{
VFP_dword (dest) = ARMul_LoadWordN (state, address);
VFP_dword (dest) <<= 32;
VFP_dword (dest) |= ARMul_LoadWordN (state, address + 4);
}
else
{
VFP_dword (dest) = ARMul_LoadWordN (state, address + 4);
VFP_dword (dest) <<= 32;
VFP_dword (dest) |= ARMul_LoadWordN (state, address);
}
if (trace)
fprintf (stderr, " VFP: VPOP: D%d = %g\n", dest, VFP_dval (dest));
address += 8;
dest += 1;
}
}
else
{
int sreg = (BITS (12, 15) << 1) | BIT (22);
int num = BITS (0, 7);
while (num--)
{
VFP_uword (sreg) = ARMul_LoadWordN (state, address);
address += 4;
sreg += 1;
}
}
}
else if (BITS (9, 11) != 0x5)
break;
else
{
/* VLDM PUW=011 */
int n = BITS (16, 19);
int imm8 = BITS (0, 7);
ARMword address = state->Reg[n];
state->Reg[n] += imm8 << 2;
if (BIT (8))
{
int dest = (BIT (22) << 4) | BITS (12, 15);
imm8 >>= 1;
while (imm8--)
{
if (state->bigendSig)
{
VFP_dword (dest) = ARMul_LoadWordN (state, address);
VFP_dword (dest) <<= 32;
VFP_dword (dest) |= ARMul_LoadWordN (state, address + 4);
}
else
{
VFP_dword (dest) = ARMul_LoadWordN (state, address + 4);
VFP_dword (dest) <<= 32;
VFP_dword (dest) |= ARMul_LoadWordN (state, address);
}
if (trace)
fprintf (stderr, " VFP: VLDM: D%d = %g\n", dest, VFP_dval (dest));
address += 8;
dest += 1;
}
}
else
{
int dest = (BITS (12, 15) << 1) | BIT (22);
while (imm8--)
{
VFP_uword (dest) = ARMul_LoadWordN (state, address);
address += 4;
dest += 1;
}
}
}
return;
case 0x11:
case 0x15:
case 0x19:
case 0x1D:
{
/* VLDR */
ARMword imm32 = BITS (0, 7) << 2;
int base = state->Reg[LHSReg];
ARMword address;
int dest;
if (LHSReg == 15)
base = (base + 3) & ~3;
address = base + (BIT (23) ? imm32 : - imm32);
if (CPNum == 10)
{
dest = (DESTReg << 1) + BIT (22);
VFP_uword (dest) = ARMul_LoadWordN (state, address);
}
else
{
dest = (BIT (22) << 4) + DESTReg;
if (state->bigendSig)
{
VFP_dword (dest) = ARMul_LoadWordN (state, address);
VFP_dword (dest) <<= 32;
VFP_dword (dest) |= ARMul_LoadWordN (state, address + 4);
}
else
{
VFP_dword (dest) = ARMul_LoadWordN (state, address + 4);
VFP_dword (dest) <<= 32;
VFP_dword (dest) |= ARMul_LoadWordN (state, address);
}
if (trace)
fprintf (stderr, " VFP: VLDR: D%d = %g\n", dest, VFP_dval (dest));
}
}
return;
}
fprintf (stderr, "SIM: VFP: Unimplemented: %0x\n", BITS (20, 24));
}
/* This function does the work of generating the addresses used in an
LDC instruction. The code here is always post-indexed, it's up to the
caller to get the input address correct and to handle base register
modification. It also handles the Busy-Waiting. */
void
ARMul_LDC (ARMul_State * state, ARMword instr, ARMword address)
{
unsigned cpab;
ARMword data;
if (CPNum == 10 || CPNum == 11)
{
handle_VFP_xfer (state, instr);
return;
}
UNDEF_LSCPCBaseWb;
if (! CP_ACCESS_ALLOWED (state, CPNum))
{
ARMul_UndefInstr (state, instr);
return;
}
if (ADDREXCEPT (address))
INTERNALABORT (address);
cpab = (state->LDC[CPNum]) (state, ARMul_FIRST, instr, 0);
while (cpab == ARMul_BUSY)
{
ARMul_Icycles (state, 1, 0);
if (IntPending (state))
{
cpab = (state->LDC[CPNum]) (state, ARMul_INTERRUPT, instr, 0);
return;
}
else
cpab = (state->LDC[CPNum]) (state, ARMul_BUSY, instr, 0);
}
if (cpab == ARMul_CANT)
{
CPTAKEABORT;
return;
}
cpab = (state->LDC[CPNum]) (state, ARMul_TRANSFER, instr, 0);
data = ARMul_LoadWordN (state, address);
BUSUSEDINCPCN;
if (BIT (21))
LSBase = state->Base;
cpab = (state->LDC[CPNum]) (state, ARMul_DATA, instr, data);
while (cpab == ARMul_INC)
{
address += 4;
data = ARMul_LoadWordN (state, address);
cpab = (state->LDC[CPNum]) (state, ARMul_DATA, instr, data);
}
if (state->abortSig || state->Aborted)
TAKEABORT;
}
/* This function does the work of generating the addresses used in an
STC instruction. The code here is always post-indexed, it's up to the
caller to get the input address correct and to handle base register
modification. It also handles the Busy-Waiting. */
void
ARMul_STC (ARMul_State * state, ARMword instr, ARMword address)
{
unsigned cpab;
ARMword data;
if (CPNum == 10 || CPNum == 11)
{
handle_VFP_xfer (state, instr);
return;
}
UNDEF_LSCPCBaseWb;
if (! CP_ACCESS_ALLOWED (state, CPNum))
{
ARMul_UndefInstr (state, instr);
return;
}
if (ADDREXCEPT (address) || VECTORACCESS (address))
INTERNALABORT (address);
cpab = (state->STC[CPNum]) (state, ARMul_FIRST, instr, &data);
while (cpab == ARMul_BUSY)
{
ARMul_Icycles (state, 1, 0);
if (IntPending (state))
{
cpab = (state->STC[CPNum]) (state, ARMul_INTERRUPT, instr, 0);
return;
}
else
cpab = (state->STC[CPNum]) (state, ARMul_BUSY, instr, &data);
}
if (cpab == ARMul_CANT)
{
CPTAKEABORT;
return;
}
#ifndef MODE32
if (ADDREXCEPT (address) || VECTORACCESS (address))
INTERNALABORT (address);
#endif
BUSUSEDINCPCN;
if (BIT (21))
LSBase = state->Base;
cpab = (state->STC[CPNum]) (state, ARMul_DATA, instr, &data);
ARMul_StoreWordN (state, address, data);
while (cpab == ARMul_INC)
{
address += 4;
cpab = (state->STC[CPNum]) (state, ARMul_DATA, instr, &data);
ARMul_StoreWordN (state, address, data);
}
if (state->abortSig || state->Aborted)
TAKEABORT;
}
/* This function does the Busy-Waiting for an MCR instruction. */
void
ARMul_MCR (ARMul_State * state, ARMword instr, ARMword source)
{
unsigned cpab;
if (! CP_ACCESS_ALLOWED (state, CPNum))
{
ARMul_UndefInstr (state, instr);
return;
}
cpab = (state->MCR[CPNum]) (state, ARMul_FIRST, instr, source);
while (cpab == ARMul_BUSY)
{
ARMul_Icycles (state, 1, 0);
if (IntPending (state))
{
cpab = (state->MCR[CPNum]) (state, ARMul_INTERRUPT, instr, 0);
return;
}
else
cpab = (state->MCR[CPNum]) (state, ARMul_BUSY, instr, source);
}
if (cpab == ARMul_CANT)
ARMul_Abort (state, ARMul_UndefinedInstrV);
else
{
BUSUSEDINCPCN;
ARMul_Ccycles (state, 1, 0);
}
}
/* This function does the Busy-Waiting for an MRC instruction. */
ARMword
ARMul_MRC (ARMul_State * state, ARMword instr)
{
unsigned cpab;
ARMword result = 0;
if (! CP_ACCESS_ALLOWED (state, CPNum))
{
ARMul_UndefInstr (state, instr);
return result;
}
cpab = (state->MRC[CPNum]) (state, ARMul_FIRST, instr, &result);
while (cpab == ARMul_BUSY)
{
ARMul_Icycles (state, 1, 0);
if (IntPending (state))
{
cpab = (state->MRC[CPNum]) (state, ARMul_INTERRUPT, instr, 0);
return (0);
}
else
cpab = (state->MRC[CPNum]) (state, ARMul_BUSY, instr, &result);
}
if (cpab == ARMul_CANT)
{
ARMul_Abort (state, ARMul_UndefinedInstrV);
/* Parent will destroy the flags otherwise. */
result = ECC;
}
else
{
BUSUSEDINCPCN;
ARMul_Ccycles (state, 1, 0);
ARMul_Icycles (state, 1, 0);
}
return result;
}
static void
handle_VFP_op (ARMul_State * state, ARMword instr)
{
int dest;
int srcN;
int srcM;
if (BITS (9, 11) != 0x5 || BIT (4) != 0)
{
fprintf (stderr, "SIM: VFP: Unimplemented: Float op: %08x\n", BITS (0,31));
return;
}
if (BIT (8))
{
dest = BITS(12,15) + (BIT (22) << 4);
srcN = LHSReg + (BIT (7) << 4);
srcM = BITS (0,3) + (BIT (5) << 4);
}
else
{
dest = (BITS(12,15) << 1) + BIT (22);
srcN = (LHSReg << 1) + BIT (7);
srcM = (BITS (0,3) << 1) + BIT (5);
}
switch (BITS (20, 27))
{
case 0xE0:
case 0xE4:
/* VMLA VMLS */
if (BIT (8))
{
ARMdval val = VFP_dval (srcN) * VFP_dval (srcM);
if (BIT (6))
{
if (trace)
fprintf (stderr, " VFP: VMLS: %g = %g - %g * %g\n",
VFP_dval (dest) - val,
VFP_dval (dest), VFP_dval (srcN), VFP_dval (srcM));
VFP_dval (dest) -= val;
}
else
{
if (trace)
fprintf (stderr, " VFP: VMLA: %g = %g + %g * %g\n",
VFP_dval (dest) + val,
VFP_dval (dest), VFP_dval (srcN), VFP_dval (srcM));
VFP_dval (dest) += val;
}
}
else
{
ARMfval val = VFP_fval (srcN) * VFP_fval (srcM);
if (BIT (6))
{
if (trace)
fprintf (stderr, " VFP: VMLS: %g = %g - %g * %g\n",
VFP_fval (dest) - val,
VFP_fval (dest), VFP_fval (srcN), VFP_fval (srcM));
VFP_fval (dest) -= val;
}
else
{
if (trace)
fprintf (stderr, " VFP: VMLA: %g = %g + %g * %g\n",
VFP_fval (dest) + val,
VFP_fval (dest), VFP_fval (srcN), VFP_fval (srcM));
VFP_fval (dest) += val;
}
}
return;
case 0xE1:
case 0xE5:
if (BIT (8))
{
ARMdval product = VFP_dval (srcN) * VFP_dval (srcM);
if (BIT (6))
{
/* VNMLA */
if (trace)
fprintf (stderr, " VFP: VNMLA: %g = -(%g + (%g * %g))\n",
-(VFP_dval (dest) + product),
VFP_dval (dest), VFP_dval (srcN), VFP_dval (srcM));
VFP_dval (dest) = -(product + VFP_dval (dest));
}
else
{
/* VNMLS */
if (trace)
fprintf (stderr, " VFP: VNMLS: %g = -(%g + (%g * %g))\n",
-(VFP_dval (dest) + product),
VFP_dval (dest), VFP_dval (srcN), VFP_dval (srcM));
VFP_dval (dest) = product - VFP_dval (dest);
}
}
else
{
ARMfval product = VFP_fval (srcN) * VFP_fval (srcM);
if (BIT (6))
/* VNMLA */
VFP_fval (dest) = -(product + VFP_fval (dest));
else
/* VNMLS */
VFP_fval (dest) = product - VFP_fval (dest);
}
return;
case 0xE2:
case 0xE6:
if (BIT (8))
{
ARMdval product = VFP_dval (srcN) * VFP_dval (srcM);
if (BIT (6))
{
if (trace)
fprintf (stderr, " VFP: VMUL: %g = %g * %g\n",
- product, VFP_dval (srcN), VFP_dval (srcM));
/* VNMUL */
VFP_dval (dest) = - product;
}
else
{
if (trace)
fprintf (stderr, " VFP: VMUL: %g = %g * %g\n",
product, VFP_dval (srcN), VFP_dval (srcM));
/* VMUL */
VFP_dval (dest) = product;
}
}
else
{
ARMfval product = VFP_fval (srcN) * VFP_fval (srcM);
if (BIT (6))
{
if (trace)
fprintf (stderr, " VFP: VNMUL: %g = %g * %g\n",
- product, VFP_fval (srcN), VFP_fval (srcM));
VFP_fval (dest) = - product;
}
else
{
if (trace)
fprintf (stderr, " VFP: VMUL: %g = %g * %g\n",
product, VFP_fval (srcN), VFP_fval (srcM));
VFP_fval (dest) = product;
}
}
return;
case 0xE3:
case 0xE7:
if (BIT (6) == 0)
{
/* VADD */
if (BIT(8))
{
if (trace)
fprintf (stderr, " VFP: VADD %g = %g + %g\n",
VFP_dval (srcN) + VFP_dval (srcM),
VFP_dval (srcN),
VFP_dval (srcM));
VFP_dval (dest) = VFP_dval (srcN) + VFP_dval (srcM);
}
else
VFP_fval (dest) = VFP_fval (srcN) + VFP_fval (srcM);
}
else
{
/* VSUB */
if (BIT(8))
{
if (trace)
fprintf (stderr, " VFP: VSUB %g = %g - %g\n",
VFP_dval (srcN) - VFP_dval (srcM),
VFP_dval (srcN),
VFP_dval (srcM));
VFP_dval (dest) = VFP_dval (srcN) - VFP_dval (srcM);
}
else
VFP_fval (dest) = VFP_fval (srcN) - VFP_fval (srcM);
}
return;
case 0xE8:
case 0xEC:
if (BIT (6) == 1)
break;
/* VDIV */
if (BIT (8))
{
ARMdval res = VFP_dval (srcN) / VFP_dval (srcM);
if (trace)
fprintf (stderr, " VFP: VDIV (64bit): %g = %g / %g\n",
res, VFP_dval (srcN), VFP_dval (srcM));
VFP_dval (dest) = res;
}
else
{
if (trace)
fprintf (stderr, " VFP: VDIV: %g = %g / %g\n",
VFP_fval (srcN) / VFP_fval (srcM),
VFP_fval (srcN), VFP_fval (srcM));
VFP_fval (dest) = VFP_fval (srcN) / VFP_fval (srcM);
}
return;
case 0xEB:
case 0xEF:
if (BIT (6) != 1)
break;
switch (BITS (16, 19))
{
case 0x0:
if (BIT (7) == 0)
{
if (BIT (8))
{
/* VMOV.F64 <Dd>, <Dm>. */
VFP_dval (dest) = VFP_dval (srcM);
if (trace)
fprintf (stderr, " VFP: VMOV d%d, d%d: %g\n", dest, srcM, VFP_dval (srcM));
}
else
{
/* VMOV.F32 <Sd>, <Sm>. */
VFP_fval (dest) = VFP_fval (srcM);
if (trace)
fprintf (stderr, " VFP: VMOV s%d, s%d: %g\n", dest, srcM, VFP_fval (srcM));
}
}
else
{
/* VABS */
if (BIT (8))
{
ARMdval src = VFP_dval (srcM);
VFP_dval (dest) = fabs (src);
if (trace)
fprintf (stderr, " VFP: VABS (%g) = %g\n", src, VFP_dval (dest));
}
else
{
ARMfval src = VFP_fval (srcM);
VFP_fval (dest) = fabsf (src);
if (trace)
fprintf (stderr, " VFP: VABS (%g) = %g\n", src, VFP_fval (dest));
}
}
return;
case 0x1:
if (BIT (7) == 0)
{
/* VNEG */
if (BIT (8))
VFP_dval (dest) = - VFP_dval (srcM);
else
VFP_fval (dest) = - VFP_fval (srcM);
}
else
{
/* VSQRT */
if (BIT (8))
{
if (trace)
fprintf (stderr, " VFP: %g = root(%g)\n",
sqrt (VFP_dval (srcM)), VFP_dval (srcM));
VFP_dval (dest) = sqrt (VFP_dval (srcM));
}
else
{
if (trace)
fprintf (stderr, " VFP: %g = root(%g)\n",
sqrtf (VFP_fval (srcM)), VFP_fval (srcM));
VFP_fval (dest) = sqrtf (VFP_fval (srcM));
}
}
return;
case 0x4:
case 0x5:
/* VCMP, VCMPE */
if (BIT(8))
{
ARMdval res = VFP_dval (dest);
if (BIT (16) == 0)
{
ARMdval src = VFP_dval (srcM);
if (isinf (res) && isinf (src))
{
if (res > 0.0 && src > 0.0)
res = 0.0;
else if (res < 0.0 && src < 0.0)
res = 0.0;
/* else leave res alone. */
}
else
res -= src;
}
/* FIXME: Add handling of signalling NaNs and the E bit. */
state->FPSCR &= 0x0FFFFFFF;
if (res < 0.0)
state->FPSCR |= NBIT;
else
state->FPSCR |= CBIT;
if (res == 0.0)
state->FPSCR |= ZBIT;
if (isnan (res))
state->FPSCR |= VBIT;
if (trace)
fprintf (stderr, " VFP: VCMP (64bit) %g vs %g res %g, flags: %c%c%c%c\n",
VFP_dval (dest), BIT (16) ? 0.0 : VFP_dval (srcM), res,
state->FPSCR & NBIT ? 'N' : '-',
state->FPSCR & ZBIT ? 'Z' : '-',
state->FPSCR & CBIT ? 'C' : '-',
state->FPSCR & VBIT ? 'V' : '-');
}
else
{
ARMfval res = VFP_fval (dest);
if (BIT (16) == 0)
{
ARMfval src = VFP_fval (srcM);
if (isinf (res) && isinf (src))
{
if (res > 0.0 && src > 0.0)
res = 0.0;
else if (res < 0.0 && src < 0.0)
res = 0.0;
/* else leave res alone. */
}
else
res -= src;
}
/* FIXME: Add handling of signalling NaNs and the E bit. */
state->FPSCR &= 0x0FFFFFFF;
if (res < 0.0)
state->FPSCR |= NBIT;
else
state->FPSCR |= CBIT;
if (res == 0.0)
state->FPSCR |= ZBIT;
if (isnan (res))
state->FPSCR |= VBIT;
if (trace)
fprintf (stderr, " VFP: VCMP (32bit) %g vs %g res %g, flags: %c%c%c%c\n",
VFP_fval (dest), BIT (16) ? 0.0 : VFP_fval (srcM), res,
state->FPSCR & NBIT ? 'N' : '-',
state->FPSCR & ZBIT ? 'Z' : '-',
state->FPSCR & CBIT ? 'C' : '-',
state->FPSCR & VBIT ? 'V' : '-');
}
return;
case 0x7:
if (BIT (8))
{
dest = (DESTReg << 1) + BIT (22);
VFP_fval (dest) = VFP_dval (srcM);
}
else
{
dest = DESTReg + (BIT (22) << 4);
VFP_dval (dest) = VFP_fval (srcM);
}
return;
case 0x8:
case 0xC:
case 0xD:
/* VCVT integer <-> FP */
if (BIT (18))
{
/* To integer. */
if (BIT (8))
{
dest = (BITS(12,15) << 1) + BIT (22);
if (BIT (16))
VFP_sword (dest) = VFP_dval (srcM);
else
VFP_uword (dest) = VFP_dval (srcM);
}
else
{
if (BIT (16))
VFP_sword (dest) = VFP_fval (srcM);
else
VFP_uword (dest) = VFP_fval (srcM);
}
}
else
{
/* From integer. */
if (BIT (8))
{
srcM = (BITS (0,3) << 1) + BIT (5);
if (BIT (7))
VFP_dval (dest) = VFP_sword (srcM);
else
VFP_dval (dest) = VFP_uword (srcM);
}
else
{
if (BIT (7))
VFP_fval (dest) = VFP_sword (srcM);
else
VFP_fval (dest) = VFP_uword (srcM);
}
}
return;
}
fprintf (stderr, "SIM: VFP: Unimplemented: Float op3: %03x\n", BITS (16,27));
return;
}
fprintf (stderr, "SIM: VFP: Unimplemented: Float op2: %02x\n", BITS (20, 27));
return;
}
/* This function does the Busy-Waiting for an CDP instruction. */
void
ARMul_CDP (ARMul_State * state, ARMword instr)
{
unsigned cpab;
if (CPNum == 10 || CPNum == 11)
{
handle_VFP_op (state, instr);
return;
}
if (! CP_ACCESS_ALLOWED (state, CPNum))
{
ARMul_UndefInstr (state, instr);
return;
}
cpab = (state->CDP[CPNum]) (state, ARMul_FIRST, instr);
while (cpab == ARMul_BUSY)
{
ARMul_Icycles (state, 1, 0);
if (IntPending (state))
{
cpab = (state->CDP[CPNum]) (state, ARMul_INTERRUPT, instr);
return;
}
else
cpab = (state->CDP[CPNum]) (state, ARMul_BUSY, instr);
}
if (cpab == ARMul_CANT)
ARMul_Abort (state, ARMul_UndefinedInstrV);
else
BUSUSEDN;
}
/* This function handles Undefined instructions, as CP isntruction. */
void
ARMul_UndefInstr (ARMul_State * state, ARMword instr ATTRIBUTE_UNUSED)
{
ARMul_Abort (state, ARMul_UndefinedInstrV);
}
/* Return TRUE if an interrupt is pending, FALSE otherwise. */
unsigned
IntPending (ARMul_State * state)
{
if (state->Exception)
{
/* Any exceptions. */
if (state->NresetSig == LOW)
{
ARMul_Abort (state, ARMul_ResetV);
return TRUE;
}
else if (!state->NfiqSig && !FFLAG)
{
ARMul_Abort (state, ARMul_FIQV);
return TRUE;
}
else if (!state->NirqSig && !IFLAG)
{
ARMul_Abort (state, ARMul_IRQV);
return TRUE;
}
}
return FALSE;
}
/* Align a word access to a non word boundary. */
ARMword
ARMul_Align (ARMul_State *state ATTRIBUTE_UNUSED, ARMword address, ARMword data)
{
/* This code assumes the address is really unaligned,
as a shift by 32 is undefined in C. */
address = (address & 3) << 3; /* Get the word address. */
return ((data >> address) | (data << (32 - address))); /* rot right */
}
/* This routine is used to call another routine after a certain number of
cycles have been executed. The first parameter is the number of cycles
delay before the function is called, the second argument is a pointer
to the function. A delay of zero doesn't work, just call the function. */
void
ARMul_ScheduleEvent (ARMul_State * state, unsigned long delay,
unsigned (*what) (ARMul_State *))
{
unsigned long when;
struct EventNode *event;
if (state->EventSet++ == 0)
state->Now = ARMul_Time (state);
when = (state->Now + delay) % EVENTLISTSIZE;
event = (struct EventNode *) malloc (sizeof (struct EventNode));
event->func = what;
event->next = *(state->EventPtr + when);
*(state->EventPtr + when) = event;
}
/* This routine is called at the beginning of
every cycle, to envoke scheduled events. */
void
ARMul_EnvokeEvent (ARMul_State * state)
{
static unsigned long then;
then = state->Now;
state->Now = ARMul_Time (state) % EVENTLISTSIZE;
if (then < state->Now)
/* Schedule events. */
EnvokeList (state, then, state->Now);
else if (then > state->Now)
{
/* Need to wrap around the list. */
EnvokeList (state, then, EVENTLISTSIZE - 1L);
EnvokeList (state, 0L, state->Now);
}
}
/* Envokes all the entries in a range. */
static void
EnvokeList (ARMul_State * state, unsigned long from, unsigned long to)
{
for (; from <= to; from++)
{
struct EventNode *anevent;
anevent = *(state->EventPtr + from);
while (anevent)
{
(anevent->func) (state);
state->EventSet--;
anevent = anevent->next;
}
*(state->EventPtr + from) = NULL;
}
}
/* This routine is returns the number of clock ticks since the last reset. */
unsigned long
ARMul_Time (ARMul_State * state)
{
return (state->NumScycles + state->NumNcycles +
state->NumIcycles + state->NumCcycles + state->NumFcycles);
}