const std = @import("std"); const Arm7tdmi = @import("cpu.zig").Arm7tdmi; const Bios = @import("bus/Bios.zig"); const Ewram = @import("bus/Ewram.zig"); const GamePak = @import("bus/GamePak.zig"); const Io = @import("bus/io.zig").Io; const Iwram = @import("bus/Iwram.zig"); const Ppu = @import("ppu.zig").Ppu; const Apu = @import("apu.zig").Apu; const DmaTuple = @import("bus/dma.zig").DmaTuple; const TimerTuple = @import("bus/timer.zig").TimerTuple; const Scheduler = @import("scheduler.zig").Scheduler; const FilePaths = @import("../util.zig").FilePaths; const io = @import("bus/io.zig"); const Allocator = std.mem.Allocator; const log = std.log.scoped(.Bus); const createDmaTuple = @import("bus/dma.zig").create; const createTimerTuple = @import("bus/timer.zig").create; const rotr = @import("../util.zig").rotr; const timings: [2][0x10]u8 = [_][0x10]u8{ // BIOS, Unused, EWRAM, IWRAM, I/0, PALRAM, VRAM, OAM, ROM0, ROM0, ROM1, ROM1, ROM2, ROM2, SRAM, Unused [_]u8{ 1, 1, 3, 1, 1, 1, 1, 1, 5, 5, 5, 5, 5, 5, 5, 5 }, // 8-bit & 16-bit [_]u8{ 1, 1, 6, 1, 1, 2, 2, 1, 8, 8, 8, 8, 8, 8, 8, 8 }, // 32-bit }; pub const fetch_timings: [2][0x10]u8 = [_][0x10]u8{ // BIOS, Unused, EWRAM, IWRAM, I/0, PALRAM, VRAM, OAM, ROM0, ROM0, ROM1, ROM1, ROM2, ROM2, SRAM, Unused [_]u8{ 1, 1, 3, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 5, 5 }, // 8-bit & 16-bit [_]u8{ 1, 1, 6, 1, 1, 2, 2, 1, 4, 4, 4, 4, 4, 4, 8, 8 }, // 32-bit }; // Fastmem Related const page_size = 1 * 0x400; // 1KiB const address_space_size = 0x1000_0000; const table_len = address_space_size / page_size; const Self = @This(); pak: GamePak, bios: Bios, ppu: Ppu, apu: Apu, dma: DmaTuple, tim: TimerTuple, iwram: Iwram, ewram: Ewram, io: Io, cpu: *Arm7tdmi, sched: *Scheduler, read_table: *const [table_len]?*const anyopaque, write_table: *const [table_len]?*anyopaque, allocator: Allocator, pub fn init(self: *Self, allocator: Allocator, sched: *Scheduler, cpu: *Arm7tdmi, paths: FilePaths) !void { const read_table = try allocator.create([table_len]?*const anyopaque); const write_table = try allocator.create([table_len]?*anyopaque); self.* = .{ .pak = try GamePak.init(allocator, cpu, paths.rom, paths.save), .bios = try Bios.init(allocator, paths.bios), .ppu = try Ppu.init(allocator, sched), .apu = Apu.init(sched), .iwram = try Iwram.init(allocator), .ewram = try Ewram.init(allocator), .dma = createDmaTuple(), .tim = createTimerTuple(sched), .io = Io.init(), .cpu = cpu, .sched = sched, .read_table = read_table, .write_table = write_table, .allocator = allocator, }; // read_table, write_table, and *Self are not restricted to the lifetime // of this init function so we can initialize our tables here fillReadTable(self, read_table); fillWriteTable(self, write_table); } pub fn deinit(self: *Self) void { self.iwram.deinit(); self.ewram.deinit(); self.pak.deinit(); self.bios.deinit(); self.ppu.deinit(); self.allocator.destroy(self.read_table); self.allocator.destroy(self.write_table); self.* = undefined; } fn fillReadTable(bus: *Self, table: *[table_len]?*const anyopaque) void { const vramMirror = @import("ppu.zig").Vram.mirror; for (table) |*ptr, i| { const addr = page_size * i; ptr.* = switch (addr) { // General Internal Memory 0x0000_0000...0x0000_3FFF => null, // BIOS has it's own checks 0x0200_0000...0x02FF_FFFF => &bus.ewram.buf[addr & 0x3FFFF], 0x0300_0000...0x03FF_FFFF => &bus.iwram.buf[addr & 0x7FFF], 0x0400_0000...0x0400_03FF => null, // I/O // Internal Display Memory 0x0500_0000...0x05FF_FFFF => &bus.ppu.palette.buf[addr & 0x3FF], 0x0600_0000...0x06FF_FFFF => &bus.ppu.vram.buf[vramMirror(addr)], 0x0700_0000...0x07FF_FFFF => &bus.ppu.oam.buf[addr & 0x3FF], // External Memory (Game Pak) 0x0800_0000...0x0DFF_FFFF => fillTableExternalMemory(bus, addr), 0x0E00_0000...0x0FFF_FFFF => null, // SRAM else => null, }; } } fn fillWriteTable(bus: *Self, table: *[table_len]?*const anyopaque) void { for (table) |*ptr, i| { const addr = page_size * i; ptr.* = switch (addr) { // General Internal Memory 0x0000_0000...0x0000_3FFF => null, // BIOS has it's own checks 0x0200_0000...0x02FF_FFFF => &bus.ewram.buf[addr & 0x3FFFF], 0x0300_0000...0x03FF_FFFF => &bus.iwram.buf[addr & 0x7FFF], 0x0400_0000...0x0400_03FF => null, // I/O // Internal Display Memory // FIXME: Different table for different integer width writes? 0x0500_0000...0x05FF_FFFF => null, // unique behaviour on 8-bit reads 0x0600_0000...0x06FF_FFFF => null, // Has some behaviour depending on DISPCNT values 0x0700_0000...0x07FF_FFFF => &bus.ppu.oam.buf[addr & 0x3FF], // 8-bit values ignored // External Memory (Game Pak) 0x0800_0000...0x0DFF_FFFF => null, // ROM 0x0E00_0000...0x0FFF_FFFF => null, // SRAM else => null, }; } } fn fillTableExternalMemory(bus: *Self, addr: usize) ?*anyopaque { // see `GamePak.zig` for more information about what conditions need to be true // so that a simple pointer dereference isn't possible const start_addr = addr; const end_addr = addr + page_size; const gpio_data = start_addr <= 0x0800_00C4 and 0x0800_00C4 < end_addr; const gpio_direction = start_addr <= 0x0800_00C6 and 0x0800_00C6 < end_addr; const gpio_control = start_addr <= 0x0800_00C8 and 0x0800_00C8 < end_addr; if (bus.pak.gpio.device.kind != .None and (gpio_data or gpio_direction or gpio_control)) { // We found a GPIO device, and this page a GPIO register. We want to handle this in slowmem return null; } if (bus.pak.backup.kind == .Eeprom) { if (bus.pak.buf.len > 0x100_000) { // We are using a "large" EEPROM which means that if the below check is true // this page has an address that's reserved for the EEPROM and therefore must // be handled in slowmem if (addr & 0x1FF_FFFF > 0x1FF_FEFF) return null; } else { // We are using a "small" EEPROM which means that if the below check is true // (that is, we're in the 0xD address page) then we must handle at least one // address in this page in slowmem if (@truncate(u4, addr >> 24) == 0xD) return null; } } // Finally, the GamePak has some unique behaviour for reads past the end of the ROM, // so those will be handled by slowmem as well const masked_addr = addr & 0x1FF_FFFF; if (masked_addr >= bus.pak.buf.len) return null; return &bus.pak.buf[masked_addr]; } // TODO: Take advantage of fastmem here too? pub fn dbgRead(self: *const Self, comptime T: type, unaligned_address: u32) T { const page = @truncate(u8, unaligned_address >> 24); const address = forceAlign(T, unaligned_address); return switch (page) { // General Internal Memory 0x00 => blk: { if (address < Bios.size) break :blk self.bios.dbgRead(T, self.cpu.r[15], address); break :blk self.openBus(T, address); }, 0x02 => self.ewram.read(T, address), 0x03 => self.iwram.read(T, address), 0x04 => self.readIo(T, address), // Internal Display Memory 0x05 => self.ppu.palette.read(T, address), 0x06 => self.ppu.vram.read(T, address), 0x07 => self.ppu.oam.read(T, address), // External Memory (Game Pak) 0x08...0x0D => self.pak.dbgRead(T, address), 0x0E...0x0F => blk: { const value = self.pak.backup.read(unaligned_address); const multiplier = switch (T) { u32 => 0x01010101, u16 => 0x0101, u8 => 1, else => @compileError("Backup: Unsupported read width"), }; break :blk @as(T, value) * multiplier; }, else => self.openBus(T, address), }; } fn readIo(self: *const Self, comptime T: type, address: u32) T { return io.read(self, T, address) orelse self.openBus(T, address); } fn openBus(self: *const Self, comptime T: type, address: u32) T { const r15 = self.cpu.r[15]; const word = blk: { // If Arm, get the most recently fetched instruction (PC + 8) // // FIXME: This is most likely a faulty assumption. // I think what *actually* happens is that the Bus has a latch for the most // recently fetched piece of data, which is then returned during Open Bus (also DMA open bus?) // I can "get away" with this because it's very statistically likely that the most recently latched value is // the most recently fetched instruction by the pipeline if (!self.cpu.cpsr.t.read()) break :blk self.cpu.pipe.stage[1].?; const page = @truncate(u8, r15 >> 24); // PC + 2 = stage[0] // PC + 4 = stage[1] // PC + 6 = Need a Debug Read for this? switch (page) { // EWRAM, PALRAM, VRAM, and Game ROM (16-bit) 0x02, 0x05, 0x06, 0x08...0x0D => { const halfword: u32 = @truncate(u16, self.cpu.pipe.stage[1].?); break :blk halfword << 16 | halfword; }, // BIOS or OAM (32-bit) 0x00, 0x07 => { // Aligned: (PC + 6) | (PC + 4) // Unaligned: (PC + 4) | (PC + 2) const aligned = address & 3 == 0b00; // TODO: What to do on PC + 6? const high: u32 = if (aligned) self.dbgRead(u16, r15 + 4) else @truncate(u16, self.cpu.pipe.stage[1].?); const low: u32 = @truncate(u16, self.cpu.pipe.stage[@boolToInt(aligned)].?); break :blk high << 16 | low; }, // IWRAM (16-bit but special) 0x03 => { // Aligned: (PC + 2) | (PC + 4) // Unaligned: (PC + 4) | (PC + 2) const aligned = address & 3 == 0b00; const high: u32 = @truncate(u16, self.cpu.pipe.stage[1 - @boolToInt(aligned)].?); const low: u32 = @truncate(u16, self.cpu.pipe.stage[@boolToInt(aligned)].?); break :blk high << 16 | low; }, else => { log.err("THUMB open bus read from 0x{X:0>2} page @0x{X:0>8}", .{ page, address }); @panic("invariant most-likely broken"); }, } }; return @truncate(T, word); } pub fn read(self: *Self, comptime T: type, unaligned_address: u32) T { const bits = @typeInfo(std.math.IntFittingRange(0, page_size - 1)).Int.bits; const page = unaligned_address >> bits; const offset = unaligned_address & (page_size - 1); // whether or not we do this in slowmem or fastmem, we should advance the scheduler self.sched.tick += timings[@boolToInt(T == u32)][@truncate(u4, page)]; // We're doing some serious out-of-bounds open-bus reads if (page > table_len) return self.slowRead(T, unaligned_address); if (self.read_table[page]) |some_ptr| { // We have a pointer to a page, cast the pointer to it's underlying type const Ptr = [*]const T; const alignment = @alignOf(std.meta.Child(Ptr)); const ptr = @ptrCast(Ptr, @alignCast(alignment, some_ptr)); // Note: We don't check array length, since we force align the // lower bits of the address as the GBA would return ptr[forceAlign(T, offset) / @sizeOf(T)]; } return self.slowRead(T, unaligned_address); } fn slowRead(self: *Self, comptime T: type, unaligned_address: u32) T { @setCold(true); const page = @truncate(u8, unaligned_address >> 24); const address = forceAlign(T, unaligned_address); return switch (page) { // General Internal Memory 0x00 => blk: { if (address < Bios.size) break :blk self.bios.read(T, self.cpu.r[15], address); break :blk self.openBus(T, address); }, 0x02 => unreachable, // completely handled by fastmeme 0x03 => unreachable, // completely handled by fastmeme 0x04 => self.readIo(T, address), // Internal Display Memory 0x05 => unreachable, // completely handled by fastmeme 0x06 => unreachable, // completely handled by fastmeme 0x07 => unreachable, // completely handled by fastmeme // External Memory (Game Pak) 0x08...0x0D => self.pak.read(T, address), 0x0E...0x0F => blk: { const value = self.pak.backup.read(unaligned_address); const multiplier = switch (T) { u32 => 0x01010101, u16 => 0x0101, u8 => 1, else => @compileError("Backup: Unsupported read width"), }; break :blk @as(T, value) * multiplier; }, else => self.openBus(T, address), }; } pub fn write(self: *Self, comptime T: type, unaligned_address: u32, value: T) void { const bits = @typeInfo(std.math.IntFittingRange(0, page_size - 1)).Int.bits; const page = unaligned_address >> bits; const offset = unaligned_address & (page_size - 1); // whether or not we do this in slowmem or fastmem, we should advance the scheduler self.sched.tick += timings[@boolToInt(T == u32)][@truncate(u4, page)]; // We're doing some serious out-of-bounds open-bus writes, they do nothing though if (page > table_len) return; if (self.write_table[page]) |some_ptr| { // We have a pointer to a page, cast the pointer to it's underlying type // 8-bit OAM reads do nothing if (T == u8 and @truncate(u8, unaligned_address >> 24) == 0x07) return; const Ptr = [*]T; const alignment = @alignOf(std.meta.Child(Ptr)); const ptr = @ptrCast(Ptr, @alignCast(alignment, some_ptr)); // Note: We don't check array length, since we force align the // lower bits of the address as the GBA would ptr[forceAlign(T, offset) / @sizeOf(T)] = value; } else { self.slowWrite(T, unaligned_address, value); } } pub fn slowWrite(self: *Self, comptime T: type, unaligned_address: u32, value: T) void { // @setCold(true); const page = @truncate(u8, unaligned_address >> 24); const address = forceAlign(T, unaligned_address); switch (page) { // General Internal Memory 0x00 => self.bios.write(T, address, value), 0x02 => unreachable, // completely handled by fastmem 0x03 => unreachable, // completely handled by fastmem 0x04 => io.write(self, T, address, value), // Internal Display Memory 0x05 => self.ppu.palette.write(T, address, value), 0x06 => self.ppu.vram.write(T, self.ppu.dispcnt, address, value), 0x07 => unreachable, // completely handled by fastmem (TODO: Is it faster if I dont?) // External Memory (Game Pak) 0x08...0x0D => self.pak.write(T, self.dma[3].word_count, address, value), 0x0E...0x0F => self.pak.backup.write(unaligned_address, @truncate(u8, rotr(T, value, 8 * rotateBy(T, unaligned_address)))), else => {}, } } inline fn rotateBy(comptime T: type, address: u32) u32 { return switch (T) { u32 => address & 3, u16 => address & 1, u8 => 0, else => @compileError("Backup: Unsupported write width"), }; } inline fn forceAlign(comptime T: type, address: u32) u32 { return switch (T) { u32 => address & ~@as(u32, 3), u16 => address & ~@as(u32, 1), u8 => address, else => @compileError("Bus: Invalid read/write type"), }; }