erc-c/src/mos6502/mos6502.c

507 lines
15 KiB
C

/*
* mos6502.c
*
* These functions are kind of the "top-level", if you will, for the MOS
* 6502 processor. You can create the processor struct, operate on the
* stack, etc.
*/
#include <stdbool.h>
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include "log.h"
#include "mos6502/mos6502.h"
#include "mos6502/dis.h"
#include "vm_debug.h"
// All of our address modes, instructions, etc. are defined here.
#include "mos6502/enums.h"
/*
* This is a table which defines what instruction each opcode is mapped
* to. All possible (256) values are defined here. You will note many
* cases where we use NOP where opcodes are not _technically_ defined;
* this may or may not be the best behavior. It's quite possible we should
* instead crash the program when we stumble upon such malformed opcodes
*/
static int instructions[] = {
// 00 01 02 03 04 05 06 07 08 09 0A 0B 0C 0D 0E 0F
BRK, ORA, NP2, NOP, TSB, ORA, ASL, NOP, PHP, ORA, ASL, NOP, TSB, ORA, ASL, NOP, // 0x
BPL, ORA, ORA, NOP, TRB, ORA, ASL, NOP, CLC, ORA, INC, NOP, TRB, ORA, ASL, NOP, // 1x
JSR, AND, NP2, NOP, BIT, AND, ROL, NOP, PLP, AND, ROL, NOP, BIT, AND, ROL, NOP, // 2x
BMI, AND, AND, NOP, BIT, AND, ROL, NOP, SEC, AND, DEC, NOP, BIT, AND, ROL, NOP, // 3x
RTI, EOR, NP2, NOP, NP2, EOR, LSR, NOP, PHA, EOR, LSR, NOP, JMP, EOR, LSR, NOP, // 4x
BVC, EOR, EOR, NOP, NP2, EOR, LSR, NOP, CLI, EOR, PHY, NOP, NP3, EOR, LSR, NOP, // 5x
RTS, ADC, NP2, NOP, STZ, ADC, ROR, NOP, PLA, ADC, ROR, NOP, JMP, ADC, ROR, NOP, // 6x
BVS, ADC, ADC, NOP, STZ, ADC, ROR, NOP, SEI, ADC, PLY, NOP, JMP, ADC, ROR, NOP, // 7x
BRA, STA, NP2, NOP, STY, STA, STX, NOP, DEY, BIM, TXA, NOP, STY, STA, STX, NOP, // 8x
BCC, STA, STA, NOP, STY, STA, STX, NOP, TYA, STA, TXS, NOP, STZ, STA, STZ, NOP, // 9x
LDY, LDA, LDX, NOP, LDY, LDA, LDX, NOP, TAY, LDA, TAX, NOP, LDY, LDA, LDX, NOP, // Ax
BCS, LDA, LDA, NOP, LDY, LDA, LDX, NOP, CLV, LDA, TSX, NOP, LDY, LDA, LDX, NOP, // Bx
CPY, CMP, NP2, NOP, CPY, CMP, DEC, NOP, INY, CMP, DEX, NOP, CPY, CMP, DEC, NOP, // Cx
BNE, CMP, CMP, NOP, NP2, CMP, DEC, NOP, CLD, CMP, PHX, NOP, NP3, CMP, DEC, NOP, // Dx
CPX, SBC, NP2, NOP, CPX, SBC, INC, NOP, INX, SBC, NOP, NOP, CPX, SBC, INC, NOP, // Ex
BEQ, SBC, SBC, NOP, NP2, SBC, INC, NOP, SED, SBC, PLX, NOP, NP3, SBC, INC, NOP, // Fx
};
/*
* A small convenience for defining instruction handlers below.
*/
#define INST_HANDLER(x) \
mos6502_handle_##x
/*
* Here's another table, this time mapping instruction codes to
* instruction handler functions. They are listed in the order defined
* in the instruction enum (in mos6502.enums.h).
*/
static mos6502_instruction_handler instruction_handlers[] = {
INST_HANDLER(adc),
INST_HANDLER(and),
INST_HANDLER(asl),
INST_HANDLER(bad),
INST_HANDLER(bcc),
INST_HANDLER(bcs),
INST_HANDLER(beq),
INST_HANDLER(bit),
INST_HANDLER(bim),
INST_HANDLER(bmi),
INST_HANDLER(bne),
INST_HANDLER(bpl),
INST_HANDLER(bra),
INST_HANDLER(brk),
INST_HANDLER(bvc),
INST_HANDLER(bvs),
INST_HANDLER(clc),
INST_HANDLER(cld),
INST_HANDLER(cli),
INST_HANDLER(clv),
INST_HANDLER(cmp),
INST_HANDLER(cpx),
INST_HANDLER(cpy),
INST_HANDLER(dec),
INST_HANDLER(dex),
INST_HANDLER(dey),
INST_HANDLER(eor),
INST_HANDLER(inc),
INST_HANDLER(inx),
INST_HANDLER(iny),
INST_HANDLER(jmp),
INST_HANDLER(jsr),
INST_HANDLER(lda),
INST_HANDLER(ldx),
INST_HANDLER(ldy),
INST_HANDLER(lsr),
INST_HANDLER(nop),
INST_HANDLER(np2),
INST_HANDLER(np3),
INST_HANDLER(ora),
INST_HANDLER(pha),
INST_HANDLER(php),
INST_HANDLER(phx),
INST_HANDLER(ply),
INST_HANDLER(pla),
INST_HANDLER(plp),
INST_HANDLER(plx),
INST_HANDLER(ply),
INST_HANDLER(rol),
INST_HANDLER(ror),
INST_HANDLER(rti),
INST_HANDLER(rts),
INST_HANDLER(sbc),
INST_HANDLER(sec),
INST_HANDLER(sed),
INST_HANDLER(sei),
INST_HANDLER(sta),
INST_HANDLER(stx),
INST_HANDLER(sty),
INST_HANDLER(stz),
INST_HANDLER(tax),
INST_HANDLER(tay),
INST_HANDLER(trb),
INST_HANDLER(tsb),
INST_HANDLER(tsx),
INST_HANDLER(txa),
INST_HANDLER(txs),
INST_HANDLER(tya),
};
/*
* Here we have a table that maps opcodes to the number of cycles each
* should cost. In cases where no opcode is defined, we set the number
* of cycles to zero.
*/
static int cycles[] = {
// 00 01 02 03 04 05 06 07 08 09 0A 0B 0C 0D 0E 0F
7, 6, 2, 1, 5, 3, 5, 1, 3, 2, 2, 1, 6, 4, 6, 1, // 0x
2, 5, 5, 1, 5, 4, 6, 1, 2, 4, 2, 1, 6, 4, 6, 1, // 1x
6, 6, 2, 1, 3, 3, 5, 1, 4, 2, 2, 1, 4, 4, 6, 1, // 2x
2, 5, 5, 1, 4, 4, 6, 1, 2, 4, 2, 1, 4, 4, 6, 1, // 3x
6, 6, 2, 1, 3, 3, 5, 1, 3, 2, 2, 1, 3, 4, 6, 1, // 4x
2, 5, 5, 1, 4, 4, 6, 1, 2, 4, 3, 1, 8, 4, 6, 1, // 5x
6, 6, 2, 1, 3, 3, 5, 1, 4, 2, 2, 1, 5, 4, 6, 1, // 6x
2, 5, 5, 1, 4, 4, 6, 1, 2, 4, 4, 1, 6, 4, 6, 1, // 7x
3, 6, 2, 1, 3, 3, 3, 1, 2, 2, 2, 1, 4, 4, 4, 1, // 8x
2, 6, 5, 1, 4, 4, 4, 1, 2, 5, 2, 1, 4, 5, 5, 1, // 9x
2, 6, 2, 1, 3, 3, 3, 1, 2, 2, 2, 1, 4, 4, 4, 1, // Ax
2, 5, 5, 1, 4, 4, 4, 1, 2, 4, 2, 1, 4, 4, 4, 1, // Bx
2, 6, 2, 1, 3, 3, 5, 1, 2, 2, 2, 1, 4, 4, 3, 1, // Cx
2, 5, 5, 1, 4, 4, 6, 1, 2, 4, 3, 1, 4, 4, 7, 1, // Dx
2, 6, 2, 1, 3, 3, 5, 1, 2, 2, 2, 1, 4, 4, 6, 1, // Ex
2, 5, 5, 1, 4, 4, 6, 1, 2, 4, 4, 1, 4, 4, 7, 1, // Fx
};
/*
* Build a new mos6502 struct object, and also build the memory contents
* used therein. All registers should be zeroed out.
*/
mos6502 *
mos6502_create(vm_segment *rmem, vm_segment *wmem)
{
mos6502 *cpu;
cpu = malloc(sizeof(mos6502));
if (cpu == NULL) {
log_crit("Not enough memory to allocate mos6502");
exit(1);
}
mos6502_set_memory(cpu, rmem, wmem);
cpu->eff_addr = 0;
cpu->PC = 0;
cpu->A = 0;
cpu->X = 0;
cpu->Y = 0;
cpu->P = MOS_STATUS_DEFAULT;
cpu->S = 0xff;
return cpu;
}
/*
* Free the memory consumed by the mos6502 struct.
*/
void
mos6502_free(mos6502 *cpu)
{
// Note we do not free rmem or wmem; we consider this to be the
// responsibility of the caller that passed us those values.
free(cpu);
}
/*
* Push a _16-bit_ number to the stack. Generally speaking, only
* addresses are pushed to the stack, such that would be contained in
* the PC register (which is 16-bit).
*
* The stack is contained within a single page of memory, so you would
* be right in observing that the stack can contain at most 128, not
* 256, addresses.
*/
void
mos6502_push_stack(mos6502 *cpu, vm_8bit addr)
{
mos6502_set(cpu, 0x100 + cpu->S, addr);
// And finally we need to increment S by 2 (since we've used two
// bytes in the stack).
cpu->S--;
}
/*
* Pop an address from the stack and return that.
*/
vm_8bit
mos6502_pop_stack(mos6502 *cpu)
{
// The first thing we want to do here is to decrement S by 2, since
// the value we want to return is two positions back.
cpu->S++;
// We need to use a bitwise-or operation to combine the hi and lo
// bytes we retrieve from the stack into the actual position we
// would use for the PC register.
return mos6502_get(cpu, 0x0100 + cpu->S);
}
/*
* Here we set the status register to a given status value, regardless
* of its past contents.
*/
void
mos6502_set_status(mos6502 *cpu, vm_8bit status)
{
cpu->P = status;
}
/*
* Return the instruction that is mapped to a given opcode.
*/
int
mos6502_instruction(vm_8bit opcode)
{
return instructions[opcode];
}
/*
* Return the number of cycles an opcode may consume. The cpu is a
* required parameter, because the number of opcodes is conditional upon
* the effective address of the instruction we're executing.
*/
int
mos6502_cycles(mos6502 *cpu, vm_8bit opcode)
{
// In some contexts, we may need to return an additional cycle.
int modif = 0;
int addr_mode;
int lo_addr;
addr_mode = mos6502_addr_mode(opcode);
// Mainly we care about the lo byte of the last effective address
lo_addr = cpu->eff_addr & 0xFF;
// Ok, here's the deal: if you are using an address mode that uses
// any of the index registers, you need to return an additional
// cycle if the lo byte of the address plus that index would cross a
// memory page boundary
switch (addr_mode) {
case ABX:
if (lo_addr + cpu->X > 255) {
modif = 1;
}
break;
case ABY:
case INY:
if (lo_addr + cpu->Y > 255) {
modif = 1;
}
break;
default:
break;
}
return cycles[opcode] + modif;
}
/*
* Here we intend to return the proper resolver function for any given
* instruction.
*/
mos6502_instruction_handler
mos6502_get_instruction_handler(vm_8bit opcode)
{
return instruction_handlers[mos6502_instruction(opcode)];
}
/*
* This code does the execution step that the 6502 processor would take,
* from soup to nuts.
*/
void
mos6502_execute(mos6502 *cpu)
{
vm_8bit opcode, operand = 0;
int /*cycles,*/ bytes;
mos6502_address_resolver resolver;
mos6502_instruction_handler handler;
// We shouldn't normally get here, if there was a breakpoint, but we
// will in testing.
if (vm_debug_broke(cpu->PC)) {
return;
}
opcode = mos6502_get(cpu, cpu->PC);
// The disassembler knows how many bytes each operand requires
// (maybe this code doesn't belong in the disassembler); let's use
// that to figure out the total number of bytes to skip. We add 1
// because we need to account for the opcode as well.
bytes = 1 + mos6502_dis_expected_bytes(mos6502_addr_mode(opcode));
// First, we need to know how to resolve our effective address and
// how to execute anything.
resolver = mos6502_get_address_resolver(mos6502_addr_mode(opcode));
handler = mos6502_get_instruction_handler(opcode);
// The operand is the effective operand, the value that the
// instruction handler cares about (if it cares about any such
// value). For example, the operand could be the literal value that
// you pass into an instruction via immediate mode. As a
// side-effect, resolver will set the eff_addr field in cpu to the
// effective address where the operand can be found in memory, or
// zero if that does not apply (such as in immediate mode).
//
// Note also that resolver may be NULL, as there may not be any
// operand for this instruction! If so, we let the default for
// operand stand, which is zero.
if (resolver) {
operand = resolver(cpu);
}
// Here's where the magic happens. Whatever the instruction does, it
// happens in the handler function.
handler(cpu, operand);
// This will be the number of cycles we should spend on the
// instruction. Of course, we can execute instructions pretty
// quickly in a modern architecture, but a lot of code was written
// with the idea that certain instructions -- in certain address
// modes -- were more expensive than others, and you want those
// programs to feel faster or slower in relation to that.
//cycles = mos6502_cycles(cpu, opcode);
// If we need to jump, then the handler has to take care of updating
// PC. If not, then we need to do it.
if (!mos6502_would_jump(mos6502_instruction(opcode))) {
cpu->PC += bytes;
}
// FIXME: uh this probably isn't right, but I wanted to do
// something.
//usleep(cycles);
// We need to record the opcode and the effective address for
// anything which might need to reference it.
cpu->last_opcode = opcode;
cpu->last_addr = cpu->eff_addr;
cpu->last_operand = operand;
// Ok -- we're done! This wasn't so hard, was it?
return;
}
/*
* Given pointers for an opcode, operand, and effective address, set the
* dereferenced values of those pointers to what the CPU knows to have
* been the last of each.
*/
void
mos6502_last_executed(mos6502 *cpu, vm_8bit *opcode,
vm_8bit *operand, vm_16bit *addr)
{
if (opcode) {
*opcode = cpu->last_opcode;
}
if (operand) {
*operand = cpu->last_operand;
}
if (addr) {
*addr = cpu->last_addr;
}
}
/*
* Return true if the given instruction would require that we jump
* to somewhere else in the program.
*/
inline bool
mos6502_would_jump(int inst_code)
{
return
inst_code == BCC ||
inst_code == BCS ||
inst_code == BEQ ||
inst_code == BMI ||
inst_code == BNE ||
inst_code == BPL ||
inst_code == BRK ||
inst_code == BVC ||
inst_code == BVS ||
inst_code == JMP ||
inst_code == JSR ||
inst_code == NP2 || // this is KIND of a hack, but it works!
inst_code == NP3 || // these jump ahead by 2 or 3 bytes (respectively)
inst_code == RTS ||
inst_code == RTI;
}
/*
* This is a _kind_ of factory method, except we're obviously not
* instantiating an object. Given an address mode, we return the
* resolver function which will give you the right value (for a given
* cpu) that an instruction will use.
*/
mos6502_address_resolver
mos6502_get_address_resolver(int addr_mode)
{
switch (addr_mode) {
case ACC: return mos6502_resolve_acc;
case ABS: return mos6502_resolve_abs;
case ABX: return mos6502_resolve_abx;
case ABY: return mos6502_resolve_aby;
case IMM: return mos6502_resolve_imm;
case IND: return mos6502_resolve_ind;
case IDX: return mos6502_resolve_idx;
case IDY: return mos6502_resolve_idy;
case REL: return mos6502_resolve_rel;
case ZPG: return mos6502_resolve_zpg;
case ZPX: return mos6502_resolve_zpx;
case ZPY: return mos6502_resolve_zpy;
case IMP: // FALLTHRU
default: break;
}
return NULL;
}
/*
* Get the byte at a given address from whatever the read memory segment
* is.
*/
inline vm_8bit
mos6502_get(mos6502 *cpu, size_t addr)
{
return vm_segment_get(cpu->rmem, addr);
}
/*
* Return the 16-bit value from a given address using the read memory
* segment.
*/
inline vm_16bit
mos6502_get16(mos6502 *cpu, size_t addr)
{
return vm_segment_get16(cpu->rmem, addr);
}
/*
* Set the byte at a given address to the given value, using whatever
* the write segment is.
*/
inline void
mos6502_set(mos6502 *cpu, size_t addr, vm_8bit value)
{
vm_segment_set(cpu->wmem, addr, value);
}
/*
* Set the two bytes at a given address to the given value, using the
* write segment.
*/
inline void
mos6502_set16(mos6502 *cpu, size_t addr, vm_16bit value)
{
vm_segment_set16(cpu->wmem, addr, value);
}
/*
* Given a read and write memory segment, update the CPU to use those
* implicitly when handling the mos6502_set/get functions.
*/
void
mos6502_set_memory(mos6502 *cpu, vm_segment *rmem, vm_segment *wmem)
{
cpu->rmem = rmem;
cpu->wmem = wmem;
}