// Copyright 2012 Michael Kang, 2014 Citra Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
// We can provide simple Thumb simulation by decoding the Thumb instruction into its corresponding
// ARM instruction, and using the existing ARM simulator.
#include "core/arm/skyeye_common/skyeye_defs.h"
#ifndef MODET // Required for the Thumb instruction support
#if 1
#error "MODET needs to be defined for the Thumb world to work"
#else
#define MODET (1)
#endif
#endif
#include "core/arm/skyeye_common/armos.h"
#include "core/arm/dyncom/arm_dyncom_thumb.h"
// Decode a 16bit Thumb instruction. The instruction is in the low 16-bits of the tinstr field,
// with the following Thumb instruction held in the high 16-bits. Passing in two Thumb instructions
// allows easier simulation of the special dual BL instruction.
tdstate thumb_translate (addr_t addr, uint32_t instr, uint32_t* ainstr, uint32_t* inst_size) {
tdstate valid = t_uninitialized;
ARMword tinstr;
tinstr = instr;
// The endian should be judge here
if((addr & 0x3) != 0)
tinstr = instr >> 16;
else
tinstr &= 0xFFFF;
*ainstr = 0xDEADC0DE; // Debugging to catch non updates
switch ((tinstr & 0xF800) >> 11) {
case 0: // LSL
case 1: // LSR
case 2: // ASR
*ainstr = 0xE1B00000 // base opcode
| ((tinstr & 0x1800) >> (11 - 5)) // shift type
|((tinstr & 0x07C0) << (7 - 6)) // imm5
|((tinstr & 0x0038) >> 3) // Rs
|((tinstr & 0x0007) << 12); // Rd
break;
case 3: // ADD/SUB
{
static const ARMword subset[4] = {
0xE0900000, // ADDS Rd,Rs,Rn
0xE0500000, // SUBS Rd,Rs,Rn
0xE2900000, // ADDS Rd,Rs,#imm3
0xE2500000 // SUBS Rd,Rs,#imm3
};
// It is quicker indexing into a table, than performing switch or conditionals:
*ainstr = subset[(tinstr & 0x0600) >> 9] // base opcode
|((tinstr & 0x01C0) >> 6) // Rn or imm3
|((tinstr & 0x0038) << (16 - 3)) // Rs
|((tinstr & 0x0007) << (12 - 0)); // Rd
}
break;
case 4: // MOV
case 5: // CMP
case 6: // ADD
case 7: // SUB
{
static const ARMword subset[4] = {
0xE3B00000, // MOVS Rd,#imm8
0xE3500000, // CMP Rd,#imm8
0xE2900000, // ADDS Rd,Rd,#imm8
0xE2500000, // SUBS Rd,Rd,#imm8
};
*ainstr = subset[(tinstr & 0x1800) >> 11] // base opcode
|((tinstr & 0x00FF) >> 0) // imm8
|((tinstr & 0x0700) << (16 - 8)) // Rn
|((tinstr & 0x0700) << (12 - 8)); // Rd
}
break;
case 8: // Arithmetic and high register transfers
// TODO: Since the subsets for both Format 4 and Format 5 instructions are made up of
// different ARM encodings, we could save the following conditional, and just have one
// large subset
if ((tinstr & (1 << 10)) == 0) {
enum otype {
t_norm,
t_shift,
t_neg,
t_mul
};
static const struct {
ARMword opcode;
otype type;
} subset[16] = {
{ 0xE0100000, t_norm }, // ANDS Rd,Rd,Rs
{ 0xE0300000, t_norm }, // EORS Rd,Rd,Rs
{ 0xE1B00010, t_shift }, // MOVS Rd,Rd,LSL Rs
{ 0xE1B00030, t_shift }, // MOVS Rd,Rd,LSR Rs
{ 0xE1B00050, t_shift }, // MOVS Rd,Rd,ASR Rs
{ 0xE0B00000, t_norm }, // ADCS Rd,Rd,Rs
{ 0xE0D00000, t_norm }, // SBCS Rd,Rd,Rs
{ 0xE1B00070, t_shift }, // MOVS Rd,Rd,ROR Rs
{ 0xE1100000, t_norm }, // TST Rd,Rs
{ 0xE2700000, t_neg }, // RSBS Rd,Rs,#0
{ 0xE1500000, t_norm }, // CMP Rd,Rs
{ 0xE1700000, t_norm }, // CMN Rd,Rs
{ 0xE1900000, t_norm }, // ORRS Rd,Rd,Rs
{ 0xE0100090, t_mul }, // MULS Rd,Rd,Rs
{ 0xE1D00000, t_norm }, // BICS Rd,Rd,Rs
{ 0xE1F00000, t_norm } // MVNS Rd,Rs
};
*ainstr = subset[(tinstr & 0x03C0) >> 6].opcode; // base
switch (subset[(tinstr & 0x03C0) >> 6].type) {
case t_norm:
*ainstr |= ((tinstr & 0x0007) << 16) // Rn
|((tinstr & 0x0007) << 12) // Rd
|((tinstr & 0x0038) >> 3); // Rs
break;
case t_shift:
*ainstr |= ((tinstr & 0x0007) << 12) // Rd
|((tinstr & 0x0007) >> 0) // Rm
|((tinstr & 0x0038) << (8 - 3)); // Rs
break;
case t_neg:
*ainstr |= ((tinstr & 0x0007) << 12) // Rd
|((tinstr & 0x0038) << (16 - 3)); // Rn
break;
case t_mul:
*ainstr |= ((tinstr & 0x0007) << 16) // Rd
|((tinstr & 0x0007) << 8) // Rs
|((tinstr & 0x0038) >> 3); // Rm
break;
}
} else {
ARMword Rd = ((tinstr & 0x0007) >> 0);
ARMword Rs = ((tinstr & 0x0038) >> 3);
if (tinstr & (1 << 7))
Rd += 8;
if (tinstr & (1 << 6))
Rs += 8;
switch ((tinstr & 0x03C0) >> 6) {
case 0x1: // ADD Rd,Rd,Hs
case 0x2: // ADD Hd,Hd,Rs
case 0x3: // ADD Hd,Hd,Hs
*ainstr = 0xE0800000 // base
| (Rd << 16) // Rn
|(Rd << 12) // Rd
|(Rs << 0); // Rm
break;
case 0x5: // CMP Rd,Hs
case 0x6: // CMP Hd,Rs
case 0x7: // CMP Hd,Hs
*ainstr = 0xE1500000 // base
| (Rd << 16) // Rn
|(Rd << 12) // Rd
|(Rs << 0); // Rm
break;
case 0x9: // MOV Rd,Hs
case 0xA: // MOV Hd,Rs
case 0xB: // MOV Hd,Hs
*ainstr = 0xE1A00000 // base
| (Rd << 16) // Rn
|(Rd << 12) // Rd
|(Rs << 0); // Rm
break;
case 0xC: // BX Rs
case 0xD: // BX Hs
*ainstr = 0xE12FFF10 // base
| ((tinstr & 0x0078) >> 3); // Rd
break;
case 0x0: // UNDEFINED
case 0x4: // UNDEFINED
case 0x8: // UNDEFINED
valid = t_undefined;
break;
case 0xE: // BLX
case 0xF: // BLX
*ainstr = 0xE1200030 // base
| (Rs << 0); // Rm
break;
}
}
break;
case 9: // LDR Rd,[PC,#imm8]
*ainstr = 0xE59F0000 // base
| ((tinstr & 0x0700) << (12 - 8)) // Rd
|((tinstr & 0x00FF) << (2 - 0)); // off8
break;
case 10:
case 11:
// TODO: Format 7 and Format 8 perform the same ARM encoding, so the following could be
// merged into a single subset, saving on the following boolean:
if ((tinstr & (1 << 9)) == 0) {
static const ARMword subset[4] = {
0xE7800000, // STR Rd,[Rb,Ro]
0xE7C00000, // STRB Rd,[Rb,Ro]
0xE7900000, // LDR Rd,[Rb,Ro]
0xE7D00000 // LDRB Rd,[Rb,Ro]
};
*ainstr = subset[(tinstr & 0x0C00) >> 10] // base
|((tinstr & 0x0007) << (12 - 0)) // Rd
|((tinstr & 0x0038) << (16 - 3)) // Rb
|((tinstr & 0x01C0) >> 6); // Ro
} else {
static const ARMword subset[4] = {
0xE18000B0, // STRH Rd,[Rb,Ro]
0xE19000D0, // LDRSB Rd,[Rb,Ro]
0xE19000B0, // LDRH Rd,[Rb,Ro]
0xE19000F0 // LDRSH Rd,[Rb,Ro]
};
*ainstr = subset[(tinstr & 0x0C00) >> 10] // base
|((tinstr & 0x0007) << (12 - 0)) // Rd
|((tinstr & 0x0038) << (16 - 3)) // Rb
|((tinstr & 0x01C0) >> 6); // Ro
}
break;
case 12: // STR Rd,[Rb,#imm5]
case 13: // LDR Rd,[Rb,#imm5]
case 14: // STRB Rd,[Rb,#imm5]
case 15: // LDRB Rd,[Rb,#imm5]
{
static const ARMword subset[4] = {
0xE5800000, // STR Rd,[Rb,#imm5]
0xE5900000, // LDR Rd,[Rb,#imm5]
0xE5C00000, // STRB Rd,[Rb,#imm5]
0xE5D00000 // LDRB Rd,[Rb,#imm5]
};
// The offset range defends on whether we are transferring a byte or word value:
*ainstr = subset[(tinstr & 0x1800) >> 11] // base
|((tinstr & 0x0007) << (12 - 0)) // Rd
|((tinstr & 0x0038) << (16 - 3)) // Rb
|((tinstr & 0x07C0) >> (6 - ((tinstr & (1 << 12)) ? 0 : 2))); // off5
}
break;
case 16: // STRH Rd,[Rb,#imm5]
case 17: // LDRH Rd,[Rb,#imm5]
*ainstr = ((tinstr & (1 << 11)) // base
? 0xE1D000B0 // LDRH
: 0xE1C000B0) // STRH
|((tinstr & 0x0007) << (12 - 0)) // Rd
|((tinstr & 0x0038) << (16 - 3)) // Rb
|((tinstr & 0x01C0) >> (6 - 1)) // off5, low nibble
|((tinstr & 0x0600) >> (9 - 8)); // off5, high nibble
break;
case 18: // STR Rd,[SP,#imm8]
case 19: // LDR Rd,[SP,#imm8]
*ainstr = ((tinstr & (1 << 11)) // base
? 0xE59D0000 // LDR
: 0xE58D0000) // STR
|((tinstr & 0x0700) << (12 - 8)) // Rd
|((tinstr & 0x00FF) << 2); // off8
break;
case 20: // ADD Rd,PC,#imm8
case 21: // ADD Rd,SP,#imm8
if ((tinstr & (1 << 11)) == 0) {
// NOTE: The PC value used here should by word aligned. We encode shift-left-by-2 in the
// rotate immediate field, so no shift of off8 is needed.
*ainstr = 0xE28F0F00 // base
| ((tinstr & 0x0700) << (12 - 8)) // Rd
|(tinstr & 0x00FF); // off8
} else {
// We encode shift-left-by-2 in the rotate immediate field, so no shift of off8 is needed.
*ainstr = 0xE28D0F00 // base
| ((tinstr & 0x0700) << (12 - 8)) // Rd
|(tinstr & 0x00FF); // off8
}
break;
case 22:
case 23:
if ((tinstr & 0x0F00) == 0x0000) {
// NOTE: The instruction contains a shift left of 2 equivalent (implemented as ROR #30):
*ainstr = ((tinstr & (1 << 7)) // base
? 0xE24DDF00 // SUB
: 0xE28DDF00) // ADD
|(tinstr & 0x007F); // off7
} else if ((tinstr & 0x0F00) == 0x0e00)
*ainstr = 0xEF000000 | SWI_Breakpoint;
else {
static const ARMword subset[4] = {
0xE92D0000, // STMDB sp!,{rlist}
0xE92D4000, // STMDB sp!,{rlist,lr}
0xE8BD0000, // LDMIA sp!,{rlist}
0xE8BD8000 // LDMIA sp!,{rlist,pc}
};
*ainstr = subset[((tinstr & (1 << 11)) >> 10) | ((tinstr & (1 << 8)) >> 8)] // base
|(tinstr & 0x00FF); // mask8
}
break;
case 24: // STMIA
case 25: // LDMIA
*ainstr = ((tinstr & (1 << 11)) // base
? 0xE8B00000 // LDMIA
: 0xE8A00000) // STMIA
|((tinstr & 0x0700) << (16 - 8)) // Rb
|(tinstr & 0x00FF); // mask8
break;
case 26: // Bcc
case 27: // Bcc/SWI
if ((tinstr & 0x0F00) == 0x0F00) {
// Format 17 : SWI
*ainstr = 0xEF000000;
// Breakpoint must be handled specially.
if ((tinstr & 0x00FF) == 0x18)
*ainstr |= ((tinstr & 0x00FF) << 16);
// New breakpoint value. See gdb/arm-tdep.c
else if ((tinstr & 0x00FF) == 0xFE)
*ainstr |= SWI_Breakpoint;
else
*ainstr |= (tinstr & 0x00FF);
} else if ((tinstr & 0x0F00) != 0x0E00)
valid = t_branch;
else // UNDEFINED : cc=1110(AL) uses different format
valid = t_undefined;
break;
case 28: // B
valid = t_branch;
break;
case 29:
if(tinstr & 0x1)
valid = t_undefined;
else
valid = t_branch;
break;
case 30: // BL instruction 1
// There is no single ARM instruction equivalent for this Thumb instruction. To keep the
// simulation simple (from the user perspective) we check if the following instruction is
// the second half of this BL, and if it is we simulate it immediately
valid = t_branch;
break;
case 31: // BL instruction 2
// There is no single ARM instruction equivalent for this instruction. Also, it should only
// ever be matched with the fmt19 "BL instruction 1" instruction. However, we do allow the
// simulation of it on its own, with undefined results if r14 is not suitably initialised.
valid = t_branch;
break;
}
*inst_size = 2;
return valid;
}