From 6eb7d5bd3ee7f0b79f3fd7adbe931dccbacf7e1b Mon Sep 17 00:00:00 2001 From: Gilles Siberlin Date: Sat, 1 Jun 2024 05:31:50 +0200 Subject: [PATCH] Implement hook insertion (#73) * Implement function hook insertion * Fix recompiled code indentation * Add _matherr to renamed_funcs * Replace after_vram by before_vram * Emit dummy value if relocatable_sections_ordered is empty --- RSPRecomp/src/rsp_recomp.cpp | 26 +- include/recomp_port.h | 8 + src/analysis.cpp | 476 +++++++++++++++++------------------ src/config.cpp | 151 +++++++---- src/main.cpp | 62 ++++- src/recompilation.cpp | 47 ++-- 6 files changed, 440 insertions(+), 330 deletions(-) diff --git a/RSPRecomp/src/rsp_recomp.cpp b/RSPRecomp/src/rsp_recomp.cpp index dd8bf74..38b914a 100644 --- a/RSPRecomp/src/rsp_recomp.cpp +++ b/RSPRecomp/src/rsp_recomp.cpp @@ -564,25 +564,25 @@ struct RSPRecompilerConfig { }; std::filesystem::path concat_if_not_empty(const std::filesystem::path& parent, const std::filesystem::path& child) { - if (!child.empty()) { - return parent / child; - } - return child; + if (!child.empty()) { + return parent / child; + } + return child; } template std::vector toml_to_vec(const toml::array* array) { - std::vector ret; + std::vector ret; - // Reserve room for all the funcs in the map. - ret.reserve(array->size()); + // Reserve room for all the funcs in the map. + ret.reserve(array->size()); array->for_each([&ret](auto&& el) { if constexpr (toml::is_integer) { ret.push_back(*el); } }); - return ret; + return ret; } template @@ -601,9 +601,9 @@ std::unordered_set toml_to_set(const toml::array* array) { bool read_config(const std::filesystem::path& config_path, RSPRecompilerConfig& out) { RSPRecompilerConfig ret{}; - try { + try { const toml::table config_data = toml::parse_file(config_path.u8string()); - std::filesystem::path basedir = std::filesystem::path{ config_path }.parent_path(); + std::filesystem::path basedir = std::filesystem::path{ config_path }.parent_path(); std::optional text_offset = config_data["text_offset"].value(); if (text_offset.has_value()) { @@ -653,20 +653,20 @@ bool read_config(const std::filesystem::path& config_path, RSPRecompilerConfig& throw toml::parse_error("Missing output_function_name in config file", config_data.source()); } - // Extra indirect branch targets (optional) + // Extra indirect branch targets (optional) const toml::node_view branch_targets_data = config_data["extra_indirect_branch_targets"]; if (branch_targets_data.is_array()) { const toml::array* branch_targets_array = branch_targets_data.as_array(); ret.extra_indirect_branch_targets = toml_to_vec(branch_targets_array); } - // Unsupported_instructions (optional) + // Unsupported_instructions (optional) const toml::node_view unsupported_instructions_data = config_data["unsupported_instructions"]; if (unsupported_instructions_data.is_array()) { const toml::array* unsupported_instructions_array = unsupported_instructions_data.as_array(); ret.unsupported_instructions = toml_to_set(unsupported_instructions_array); } - } + } catch (const toml::parse_error& err) { std::cerr << "Syntax error parsing toml: " << *err.source().path << " (" << err.source().begin << "):\n" << err.description() << std::endl; return false; diff --git a/include/recomp_port.h b/include/recomp_port.h index a74a8b5..3828f01 100644 --- a/include/recomp_port.h +++ b/include/recomp_port.h @@ -40,6 +40,12 @@ namespace RecompPort { uint32_t value; }; + struct FunctionHook { + std::string func_name; + int32_t before_vram; + std::string text; + }; + struct FunctionSize { std::string func_name; uint32_t size_bytes; @@ -71,6 +77,7 @@ namespace RecompPort { std::vector ignored_funcs; DeclaredFunctionMap declared_funcs; std::vector instruction_patches; + std::vector function_hooks; std::vector manual_func_sizes; std::vector manual_functions; std::string bss_section_suffix; @@ -110,6 +117,7 @@ namespace RecompPort { bool ignored; bool reimplemented; bool stubbed; + std::unordered_map function_hooks; Function(uint32_t vram, uint32_t rom, std::vector words, std::string name, ELFIO::Elf_Half section_index, bool ignored = false, bool reimplemented = false, bool stubbed = false) : vram(vram), rom(rom), words(std::move(words)), name(std::move(name)), section_index(section_index), ignored(ignored), reimplemented(reimplemented), stubbed(stubbed) {} diff --git a/src/analysis.cpp b/src/analysis.cpp index f98b737..5a689a0 100644 --- a/src/analysis.cpp +++ b/src/analysis.cpp @@ -10,271 +10,271 @@ extern "C" const char* RabbitizerRegister_getNameGpr(uint8_t regValue); // If 64-bit addressing is ever implemented, these will need to be changed to 64-bit values struct RegState { - // For tracking a register that will be used to load from RAM - uint32_t prev_lui; - uint32_t prev_addiu_vram; - uint32_t prev_addu_vram; - uint8_t prev_addend_reg; - bool valid_lui; - bool valid_addiu; - bool valid_addend; - // For tracking a register that has been loaded from RAM - uint32_t loaded_lw_vram; - uint32_t loaded_addu_vram; - uint32_t loaded_address; - uint8_t loaded_addend_reg; - bool valid_loaded; + // For tracking a register that will be used to load from RAM + uint32_t prev_lui; + uint32_t prev_addiu_vram; + uint32_t prev_addu_vram; + uint8_t prev_addend_reg; + bool valid_lui; + bool valid_addiu; + bool valid_addend; + // For tracking a register that has been loaded from RAM + uint32_t loaded_lw_vram; + uint32_t loaded_addu_vram; + uint32_t loaded_address; + uint8_t loaded_addend_reg; + bool valid_loaded; - RegState() = default; + RegState() = default; - void invalidate() { - prev_lui = 0; - prev_addiu_vram = 0; - prev_addu_vram = 0; - prev_addend_reg = 0; + void invalidate() { + prev_lui = 0; + prev_addiu_vram = 0; + prev_addu_vram = 0; + prev_addend_reg = 0; - valid_lui = false; - valid_addiu = false; - valid_addend = false; + valid_lui = false; + valid_addiu = false; + valid_addend = false; - loaded_lw_vram = 0; - loaded_addu_vram = 0; - loaded_address = 0; - loaded_addend_reg = 0; + loaded_lw_vram = 0; + loaded_addu_vram = 0; + loaded_address = 0; + loaded_addend_reg = 0; - valid_loaded = false; - } + valid_loaded = false; + } }; using InstrId = rabbitizer::InstrId::UniqueId; using RegId = rabbitizer::Registers::Cpu::GprO32; bool analyze_instruction(const rabbitizer::InstructionCpu& instr, const RecompPort::Function& func, RecompPort::FunctionStats& stats, - RegState reg_states[32], std::vector& stack_states) { - // Temporary register state for tracking the register being operated on - RegState temp{}; + RegState reg_states[32], std::vector& stack_states) { + // Temporary register state for tracking the register being operated on + RegState temp{}; - int rd = (int)instr.GetO32_rd(); - int rs = (int)instr.GetO32_rs(); - int base = rs; - int rt = (int)instr.GetO32_rt(); - int sa = (int)instr.Get_sa(); + int rd = (int)instr.GetO32_rd(); + int rs = (int)instr.GetO32_rs(); + int base = rs; + int rt = (int)instr.GetO32_rt(); + int sa = (int)instr.Get_sa(); - uint16_t imm = instr.Get_immediate(); + uint16_t imm = instr.Get_immediate(); - auto check_move = [&]() { - if (rs == 0) { - // rs is zero so copy rt to rd - reg_states[rd] = reg_states[rt]; - } else if (rt == 0) { - // rt is zero so copy rs to rd - reg_states[rd] = reg_states[rs]; - } else { - // Not a move, invalidate rd - reg_states[rd].invalidate(); - } - }; + auto check_move = [&]() { + if (rs == 0) { + // rs is zero so copy rt to rd + reg_states[rd] = reg_states[rt]; + } else if (rt == 0) { + // rt is zero so copy rs to rd + reg_states[rd] = reg_states[rs]; + } else { + // Not a move, invalidate rd + reg_states[rd].invalidate(); + } + }; - switch (instr.getUniqueId()) { - case InstrId::cpu_lui: - // rt has been completely overwritten, so invalidate it - reg_states[rt].invalidate(); - reg_states[rt].prev_lui = (int16_t)imm << 16; - reg_states[rt].valid_lui = true; - break; - case InstrId::cpu_addiu: - // The target reg is a copy of the source reg plus an immediate, so copy the source reg's state - reg_states[rt] = reg_states[rs]; - // Set the addiu state if and only if there hasn't been an addiu already - if (!reg_states[rt].valid_addiu) { - reg_states[rt].prev_addiu_vram = (int16_t)imm; - reg_states[rt].valid_addiu = true; - } else { - // Otherwise, there have been 2 or more consecutive addius so invalidate the whole register - reg_states[rt].invalidate(); - } - break; - case InstrId::cpu_addu: - // rd has been completely overwritten, so invalidate it - temp.invalidate(); - // Exactly one of the two addend register states should have a valid lui at this time - if (reg_states[rs].valid_lui != reg_states[rt].valid_lui) { - // Track which of the two registers has the valid lui state and which is the addend - int valid_lui_reg = reg_states[rs].valid_lui ? rs : rt; - int addend_reg = reg_states[rs].valid_lui ? rt : rs; + switch (instr.getUniqueId()) { + case InstrId::cpu_lui: + // rt has been completely overwritten, so invalidate it + reg_states[rt].invalidate(); + reg_states[rt].prev_lui = (int16_t)imm << 16; + reg_states[rt].valid_lui = true; + break; + case InstrId::cpu_addiu: + // The target reg is a copy of the source reg plus an immediate, so copy the source reg's state + reg_states[rt] = reg_states[rs]; + // Set the addiu state if and only if there hasn't been an addiu already + if (!reg_states[rt].valid_addiu) { + reg_states[rt].prev_addiu_vram = (int16_t)imm; + reg_states[rt].valid_addiu = true; + } else { + // Otherwise, there have been 2 or more consecutive addius so invalidate the whole register + reg_states[rt].invalidate(); + } + break; + case InstrId::cpu_addu: + // rd has been completely overwritten, so invalidate it + temp.invalidate(); + // Exactly one of the two addend register states should have a valid lui at this time + if (reg_states[rs].valid_lui != reg_states[rt].valid_lui) { + // Track which of the two registers has the valid lui state and which is the addend + int valid_lui_reg = reg_states[rs].valid_lui ? rs : rt; + int addend_reg = reg_states[rs].valid_lui ? rt : rs; - // Copy the lui reg's state into the destination reg, then set the destination reg's addend to the other operand - temp = reg_states[valid_lui_reg]; - temp.valid_addend = true; - temp.prev_addend_reg = addend_reg; - temp.prev_addu_vram = instr.getVram(); - } else { - // Check if this is a move - check_move(); - } - reg_states[rd] = temp; - break; - case InstrId::cpu_daddu: - case InstrId::cpu_or: - check_move(); - break; - case InstrId::cpu_sw: - // If this is a store to the stack, copy the state of rt into the stack at the given offset - if (base == (int)RegId::GPR_O32_sp) { - if ((imm & 0b11) != 0) { - fmt::print(stderr, "Invalid alignment on offset for sw to stack: {}\n", (int16_t)imm); - return false; - } - if (((int16_t)imm) < 0) { - fmt::print(stderr, "Negative offset for sw to stack: {}\n", (int16_t)imm); - return false; - } - size_t stack_offset = imm / 4; - if (stack_offset >= stack_states.size()) { - stack_states.resize(stack_offset + 1); - } - stack_states[stack_offset] = reg_states[rt]; - } - break; - case InstrId::cpu_lw: - // rt has been completely overwritten, so invalidate it - temp.invalidate(); - // If this is a load from the stack, copy the state of the stack at the given offset to rt - if (base == (int)RegId::GPR_O32_sp) { - if ((imm & 0b11) != 0) { - fmt::print(stderr, "Invalid alignment on offset for lw from stack: {}\n", (int16_t)imm); - return false; - } - if (((int16_t)imm) < 0) { - fmt::print(stderr, "Negative offset for lw from stack: {}\n", (int16_t)imm); - return false; - } - size_t stack_offset = imm / 4; - if (stack_offset >= stack_states.size()) { - stack_states.resize(stack_offset + 1); - } - temp = stack_states[stack_offset]; - } - // If the base register has a valid lui state and a valid addend before this, then this may be a load from a jump table - else if (reg_states[base].valid_lui && reg_states[base].valid_addend) { - // Exactly one of the lw and the base reg should have a valid lo16 value - bool nonzero_immediate = imm != 0; - if (nonzero_immediate != reg_states[base].valid_addiu) { - uint32_t lo16; - if (nonzero_immediate) { - lo16 = (int16_t)imm; - } else { - lo16 = reg_states[base].prev_addiu_vram; - } + // Copy the lui reg's state into the destination reg, then set the destination reg's addend to the other operand + temp = reg_states[valid_lui_reg]; + temp.valid_addend = true; + temp.prev_addend_reg = addend_reg; + temp.prev_addu_vram = instr.getVram(); + } else { + // Check if this is a move + check_move(); + } + reg_states[rd] = temp; + break; + case InstrId::cpu_daddu: + case InstrId::cpu_or: + check_move(); + break; + case InstrId::cpu_sw: + // If this is a store to the stack, copy the state of rt into the stack at the given offset + if (base == (int)RegId::GPR_O32_sp) { + if ((imm & 0b11) != 0) { + fmt::print(stderr, "Invalid alignment on offset for sw to stack: {}\n", (int16_t)imm); + return false; + } + if (((int16_t)imm) < 0) { + fmt::print(stderr, "Negative offset for sw to stack: {}\n", (int16_t)imm); + return false; + } + size_t stack_offset = imm / 4; + if (stack_offset >= stack_states.size()) { + stack_states.resize(stack_offset + 1); + } + stack_states[stack_offset] = reg_states[rt]; + } + break; + case InstrId::cpu_lw: + // rt has been completely overwritten, so invalidate it + temp.invalidate(); + // If this is a load from the stack, copy the state of the stack at the given offset to rt + if (base == (int)RegId::GPR_O32_sp) { + if ((imm & 0b11) != 0) { + fmt::print(stderr, "Invalid alignment on offset for lw from stack: {}\n", (int16_t)imm); + return false; + } + if (((int16_t)imm) < 0) { + fmt::print(stderr, "Negative offset for lw from stack: {}\n", (int16_t)imm); + return false; + } + size_t stack_offset = imm / 4; + if (stack_offset >= stack_states.size()) { + stack_states.resize(stack_offset + 1); + } + temp = stack_states[stack_offset]; + } + // If the base register has a valid lui state and a valid addend before this, then this may be a load from a jump table + else if (reg_states[base].valid_lui && reg_states[base].valid_addend) { + // Exactly one of the lw and the base reg should have a valid lo16 value + bool nonzero_immediate = imm != 0; + if (nonzero_immediate != reg_states[base].valid_addiu) { + uint32_t lo16; + if (nonzero_immediate) { + lo16 = (int16_t)imm; + } else { + lo16 = reg_states[base].prev_addiu_vram; + } - uint32_t address = reg_states[base].prev_lui + lo16; - temp.valid_loaded = true; - temp.loaded_lw_vram = instr.getVram(); - temp.loaded_address = address; - temp.loaded_addend_reg = reg_states[base].prev_addend_reg; - temp.loaded_addu_vram = reg_states[base].prev_addu_vram; - } - } - reg_states[rt] = temp; - break; - case InstrId::cpu_jr: - // Ignore jr $ra - if (rs == (int)rabbitizer::Registers::Cpu::GprO32::GPR_O32_ra) { - break; - } - // Check if the source reg has a valid loaded state and if so record that as a jump table - if (reg_states[rs].valid_loaded) { - stats.jump_tables.emplace_back( - reg_states[rs].loaded_address, - reg_states[rs].loaded_addend_reg, - 0, - reg_states[rs].loaded_lw_vram, - reg_states[rs].loaded_addu_vram, - instr.getVram(), - std::vector{} - ); - } else if (reg_states[rs].valid_lui && reg_states[rs].valid_addiu && !reg_states[rs].valid_addend && !reg_states[rs].valid_loaded) { - uint32_t address = reg_states[rs].prev_addiu_vram + reg_states[rs].prev_lui; - stats.absolute_jumps.emplace_back( - address, - instr.getVram() - ); - } - // Allow tail calls (TODO account for trailing nops due to bad function splits) - else if (instr.getVram() != func.vram + (func.words.size() - 2) * sizeof(func.words[0])) { - // Inconclusive analysis - fmt::print(stderr, "Failed to to find jump table for `jr {}` at 0x{:08X} in {}\n", RabbitizerRegister_getNameGpr(rs), instr.getVram(), func.name); - return false; - } - break; - default: - if (instr.modifiesRd()) { - reg_states[rd].invalidate(); - } - if (instr.modifiesRt()) { - reg_states[rt].invalidate(); - } - break; - } - return true; + uint32_t address = reg_states[base].prev_lui + lo16; + temp.valid_loaded = true; + temp.loaded_lw_vram = instr.getVram(); + temp.loaded_address = address; + temp.loaded_addend_reg = reg_states[base].prev_addend_reg; + temp.loaded_addu_vram = reg_states[base].prev_addu_vram; + } + } + reg_states[rt] = temp; + break; + case InstrId::cpu_jr: + // Ignore jr $ra + if (rs == (int)rabbitizer::Registers::Cpu::GprO32::GPR_O32_ra) { + break; + } + // Check if the source reg has a valid loaded state and if so record that as a jump table + if (reg_states[rs].valid_loaded) { + stats.jump_tables.emplace_back( + reg_states[rs].loaded_address, + reg_states[rs].loaded_addend_reg, + 0, + reg_states[rs].loaded_lw_vram, + reg_states[rs].loaded_addu_vram, + instr.getVram(), + std::vector{} + ); + } else if (reg_states[rs].valid_lui && reg_states[rs].valid_addiu && !reg_states[rs].valid_addend && !reg_states[rs].valid_loaded) { + uint32_t address = reg_states[rs].prev_addiu_vram + reg_states[rs].prev_lui; + stats.absolute_jumps.emplace_back( + address, + instr.getVram() + ); + } + // Allow tail calls (TODO account for trailing nops due to bad function splits) + else if (instr.getVram() != func.vram + (func.words.size() - 2) * sizeof(func.words[0])) { + // Inconclusive analysis + fmt::print(stderr, "Failed to to find jump table for `jr {}` at 0x{:08X} in {}\n", RabbitizerRegister_getNameGpr(rs), instr.getVram(), func.name); + return false; + } + break; + default: + if (instr.modifiesRd()) { + reg_states[rd].invalidate(); + } + if (instr.modifiesRt()) { + reg_states[rt].invalidate(); + } + break; + } + return true; } bool RecompPort::analyze_function(const RecompPort::Context& context, const RecompPort::Function& func, - const std::vector& instructions, RecompPort::FunctionStats& stats) { - // Create a state to track each register (r0 won't be used) - RegState reg_states[32] {}; - std::vector stack_states{}; + const std::vector& instructions, RecompPort::FunctionStats& stats) { + // Create a state to track each register (r0 won't be used) + RegState reg_states[32] {}; + std::vector stack_states{}; - // Look for jump tables - // A linear search through the func won't be accurate due to not taking control flow into account, but it'll work for finding jtables - for (const auto& instr : instructions) { - if (!analyze_instruction(instr, func, stats, reg_states, stack_states)) { - return false; - } - } + // Look for jump tables + // A linear search through the func won't be accurate due to not taking control flow into account, but it'll work for finding jtables + for (const auto& instr : instructions) { + if (!analyze_instruction(instr, func, stats, reg_states, stack_states)) { + return false; + } + } - // Sort jump tables by their address - std::sort(stats.jump_tables.begin(), stats.jump_tables.end(), - [](const JumpTable& a, const JumpTable& b) - { - return a.vram < b.vram; - }); + // Sort jump tables by their address + std::sort(stats.jump_tables.begin(), stats.jump_tables.end(), + [](const JumpTable& a, const JumpTable& b) + { + return a.vram < b.vram; + }); - // Determine jump table sizes - for (size_t i = 0; i < stats.jump_tables.size(); i++) { - JumpTable& cur_jtbl = stats.jump_tables[i]; - uint32_t end_address = (uint32_t)-1; - uint32_t entry_count = 0; - uint32_t vram = cur_jtbl.vram; + // Determine jump table sizes + for (size_t i = 0; i < stats.jump_tables.size(); i++) { + JumpTable& cur_jtbl = stats.jump_tables[i]; + uint32_t end_address = (uint32_t)-1; + uint32_t entry_count = 0; + uint32_t vram = cur_jtbl.vram; - if (i < stats.jump_tables.size() - 1) { - end_address = stats.jump_tables[i + 1].vram; - } + if (i < stats.jump_tables.size() - 1) { + end_address = stats.jump_tables[i + 1].vram; + } - // TODO this assumes that the jump table is in the same section as the function itself - cur_jtbl.rom = cur_jtbl.vram + func.rom - func.vram; + // TODO this assumes that the jump table is in the same section as the function itself + cur_jtbl.rom = cur_jtbl.vram + func.rom - func.vram; - while (vram < end_address) { - // Retrieve the current entry of the jump table - // TODO same as above - uint32_t rom_addr = vram + func.rom - func.vram; - uint32_t jtbl_word = byteswap(*reinterpret_cast(&context.rom[rom_addr])); - // Check if the entry is a valid address in the current function - if (jtbl_word < func.vram || jtbl_word > func.vram + func.words.size() * sizeof(func.words[0])) { - // If it's not then this is the end of the jump table - break; - } - cur_jtbl.entries.push_back(jtbl_word); - vram += 4; - } + while (vram < end_address) { + // Retrieve the current entry of the jump table + // TODO same as above + uint32_t rom_addr = vram + func.rom - func.vram; + uint32_t jtbl_word = byteswap(*reinterpret_cast(&context.rom[rom_addr])); + // Check if the entry is a valid address in the current function + if (jtbl_word < func.vram || jtbl_word > func.vram + func.words.size() * sizeof(func.words[0])) { + // If it's not then this is the end of the jump table + break; + } + cur_jtbl.entries.push_back(jtbl_word); + vram += 4; + } - if (cur_jtbl.entries.size() == 0) { - fmt::print("Failed to determine size of jump table at 0x{:08X} for instruction at 0x{:08X}\n", cur_jtbl.vram, cur_jtbl.jr_vram); - return false; - } + if (cur_jtbl.entries.size() == 0) { + fmt::print("Failed to determine size of jump table at 0x{:08X} for instruction at 0x{:08X}\n", cur_jtbl.vram, cur_jtbl.jr_vram); + return false; + } - //fmt::print("Jtbl at 0x{:08X} (rom 0x{:08X}) with {} entries used by instr at 0x{:08X}\n", cur_jtbl.vram, cur_jtbl.rom, cur_jtbl.entries.size(), cur_jtbl.jr_vram); - } + //fmt::print("Jtbl at 0x{:08X} (rom 0x{:08X}) with {} entries used by instr at 0x{:08X}\n", cur_jtbl.vram, cur_jtbl.rom, cur_jtbl.entries.size(), cur_jtbl.jr_vram); + } - return true; + return true; } diff --git a/src/config.cpp b/src/config.cpp index b22f9bf..e106ae6 100644 --- a/src/config.cpp +++ b/src/config.cpp @@ -5,10 +5,10 @@ #include "recomp_port.h" std::vector get_manual_funcs(const toml::array* manual_funcs_array) { - std::vector ret; + std::vector ret; - // Reserve room for all the funcs in the map. - ret.reserve(manual_funcs_array->size()); + // Reserve room for all the funcs in the map. + ret.reserve(manual_funcs_array->size()); manual_funcs_array->for_each([&ret](auto&& el) { if constexpr (toml::is_table) { std::optional func_name = el["name"].template value(); @@ -27,13 +27,13 @@ std::vector get_manual_funcs(const toml::array* manu } }); - return ret; + return ret; } std::vector get_stubbed_funcs(const toml::table* patches_data) { - std::vector stubbed_funcs{}; + std::vector stubbed_funcs{}; - // Check if the stubs array exists. + // Check if the stubs array exists. const toml::node_view stubs_data = (*patches_data)["stubs"]; if (stubs_data.is_array()) { @@ -53,13 +53,13 @@ std::vector get_stubbed_funcs(const toml::table* patches_data) { }); } - return stubbed_funcs; + return stubbed_funcs; } std::vector get_ignored_funcs(const toml::table* patches_data) { - std::vector ignored_funcs{}; + std::vector ignored_funcs{}; - // Check if the ignored funcs array exists. + // Check if the ignored funcs array exists. const toml::node_view ignored_funcs_data = (*patches_data)["ignored"]; if (ignored_funcs_data.is_array()) { @@ -76,16 +76,16 @@ std::vector get_ignored_funcs(const toml::table* patches_data) { }); } - return ignored_funcs; + return ignored_funcs; } std::unordered_map arg_type_map{ - {"u32", RecompPort::FunctionArgType::u32}, - {"s32", RecompPort::FunctionArgType::s32}, + {"u32", RecompPort::FunctionArgType::u32}, + {"s32", RecompPort::FunctionArgType::s32}, }; std::vector parse_args(const toml::array* args_in) { - std::vector ret(args_in->size()); + std::vector ret(args_in->size()); args_in->for_each([&ret](auto&& el) { if constexpr (toml::is_string) { @@ -104,13 +104,13 @@ std::vector parse_args(const toml::array* args_in) } }); - return ret; + return ret; } RecompPort::DeclaredFunctionMap get_declared_funcs(const toml::table* patches_data) { - RecompPort::DeclaredFunctionMap declared_funcs{}; + RecompPort::DeclaredFunctionMap declared_funcs{}; - // Check if the func array exists. + // Check if the func array exists. const toml::node_view funcs_data = (*patches_data)["func"]; if (funcs_data.is_array()) { @@ -138,13 +138,13 @@ RecompPort::DeclaredFunctionMap get_declared_funcs(const toml::table* patches_da }); } - return declared_funcs; + return declared_funcs; } std::vector get_func_sizes(const toml::table* patches_data) { - std::vector func_sizes{}; + std::vector func_sizes{}; - // Check if the func size array exists. + // Check if the func size array exists. const toml::node_view funcs_data = (*patches_data)["function_sizes"]; if (funcs_data.is_array()) { const toml::array* sizes_array = funcs_data.as_array(); @@ -177,13 +177,13 @@ std::vector get_func_sizes(const toml::table* patches_ }); } - return func_sizes; + return func_sizes; } std::vector get_instruction_patches(const toml::table* patches_data) { - std::vector ret; + std::vector ret; - // Check if the instruction patch array exists. + // Check if the instruction patch array exists. const toml::node_view insn_patch_data = (*patches_data)["instruction"]; if (insn_patch_data.is_array()) { @@ -221,20 +221,64 @@ std::vector get_instruction_patches(const toml::ta }); } - return ret; + return ret; +} + +std::vector get_function_hooks(const toml::table* patches_data) { + std::vector ret; + + // Check if the function hook array exists. + const toml::node_view func_hook_data = (*patches_data)["hook"]; + + if (func_hook_data.is_array()) { + const toml::array* func_hook_array = func_hook_data.as_array(); + ret.reserve(func_hook_array->size()); + + // Copy all the hooks into the output vector. + func_hook_array->for_each([&ret](auto&& el) { + if constexpr (toml::is_table) { + const toml::table& cur_hook = *el.as_table(); + + // Get the vram and make sure it's 4-byte aligned. + std::optional before_vram = cur_hook["before_vram"].value(); + std::optional func_name = cur_hook["func"].value(); + std::optional text = cur_hook["text"].value(); + + if (!func_name.has_value() || !text.has_value()) { + throw toml::parse_error("Function hook is missing required value(s)", el.source()); + } + + if (before_vram.has_value() && before_vram.value() & 0b11) { + // Not properly aligned, so throw an error (and make it look like a normal toml one). + throw toml::parse_error("before_vram is not word-aligned", el.source()); + } + + ret.push_back(RecompPort::FunctionHook{ + .func_name = func_name.value(), + .before_vram = before_vram.has_value() ? (int32_t)before_vram.value() : 0, + .text = text.value(), + }); + } + else { + throw toml::parse_error("Invalid function hook entry", el.source()); + } + }); + } + + return ret; } std::filesystem::path concat_if_not_empty(const std::filesystem::path& parent, const std::filesystem::path& child) { - if (!child.empty()) { - return parent / child; - } - return child; + if (!child.empty()) { + return parent / child; + } + return child; } RecompPort::Config::Config(const char* path) { - // Start this config out as bad so that it has to finish parsing without errors to be good. - entrypoint = 0; - bad = true; + // Start this config out as bad so that it has to finish parsing without errors to be good. + entrypoint = 0; + bad = true; toml::table config_data{}; try { @@ -348,6 +392,9 @@ RecompPort::Config::Config(const char* path) { // Manual function sizes (optional) manual_func_sizes = get_func_sizes(table); + + // Fonction hooks (optional) + function_hooks = get_function_hooks(table); } } catch (const toml::parse_error& err) { @@ -355,34 +402,34 @@ RecompPort::Config::Config(const char* path) { return; } - // No errors occured, so mark this config file as good. - bad = false; + // No errors occured, so mark this config file as good. + bad = false; } const std::unordered_map reloc_type_name_map { - { "R_MIPS_NONE", RecompPort::RelocType::R_MIPS_NONE }, - { "R_MIPS_16", RecompPort::RelocType::R_MIPS_16 }, - { "R_MIPS_32", RecompPort::RelocType::R_MIPS_32 }, - { "R_MIPS_REL32", RecompPort::RelocType::R_MIPS_REL32 }, - { "R_MIPS_26", RecompPort::RelocType::R_MIPS_26 }, - { "R_MIPS_HI16", RecompPort::RelocType::R_MIPS_HI16 }, - { "R_MIPS_LO16", RecompPort::RelocType::R_MIPS_LO16 }, - { "R_MIPS_GPREL16", RecompPort::RelocType::R_MIPS_GPREL16 }, + { "R_MIPS_NONE", RecompPort::RelocType::R_MIPS_NONE }, + { "R_MIPS_16", RecompPort::RelocType::R_MIPS_16 }, + { "R_MIPS_32", RecompPort::RelocType::R_MIPS_32 }, + { "R_MIPS_REL32", RecompPort::RelocType::R_MIPS_REL32 }, + { "R_MIPS_26", RecompPort::RelocType::R_MIPS_26 }, + { "R_MIPS_HI16", RecompPort::RelocType::R_MIPS_HI16 }, + { "R_MIPS_LO16", RecompPort::RelocType::R_MIPS_LO16 }, + { "R_MIPS_GPREL16", RecompPort::RelocType::R_MIPS_GPREL16 }, }; RecompPort::RelocType reloc_type_from_name(const std::string& reloc_type_name) { - auto find_it = reloc_type_name_map.find(reloc_type_name); - if (find_it != reloc_type_name_map.end()) { - return find_it->second; - } - return RecompPort::RelocType::R_MIPS_NONE; + auto find_it = reloc_type_name_map.find(reloc_type_name); + if (find_it != reloc_type_name_map.end()) { + return find_it->second; + } + return RecompPort::RelocType::R_MIPS_NONE; } bool RecompPort::Context::from_symbol_file(const std::filesystem::path& symbol_file_path, std::vector&& rom, RecompPort::Context& out) { - RecompPort::Context ret{}; + RecompPort::Context ret{}; - try { - const toml::table config_data = toml::parse_file(symbol_file_path.u8string()); + try { + const toml::table config_data = toml::parse_file(symbol_file_path.u8string()); const toml::node_view config_sections_value = config_data["section"]; if (!config_sections_value.is_array()) { @@ -518,13 +565,13 @@ bool RecompPort::Context::from_symbol_file(const std::filesystem::path& symbol_f throw toml::parse_error("Invalid section entry", el.source()); } }); - } + } catch (const toml::parse_error& err) { std::cerr << "Syntax error parsing toml: " << *err.source().path << " (" << err.source().begin << "):\n" << err.description() << std::endl; return false; } - ret.rom = std::move(rom); - out = std::move(ret); - return true; + ret.rom = std::move(rom); + out = std::move(ret); + return true; } diff --git a/src/main.cpp b/src/main.cpp index 00f6d9e..8faae53 100644 --- a/src/main.cpp +++ b/src/main.cpp @@ -628,6 +628,7 @@ std::unordered_set renamed_funcs{ "div64_64", "div64_32", "__moddi3", + "_matherr", }; bool read_symbols(RecompPort::Context& context, const ELFIO::elfio& elf_file, ELFIO::section* symtab_section, uint32_t entrypoint, bool has_entrypoint, bool use_absolute_symbols) { @@ -1499,6 +1500,41 @@ int main(int argc, char** argv) { func.words[instruction_index] = byteswap(patch.value); } + // Apply any function hooks. + for (const RecompPort::FunctionHook& patch : config.function_hooks) { + // Check if the specified function exists. + auto func_find = context.functions_by_name.find(patch.func_name); + if (func_find == context.functions_by_name.end()) { + // Function doesn't exist, present an error to the user instead of silently failing to stub it out. + // This helps prevent typos in the config file or functions renamed between versions from causing issues. + exit_failure(fmt::format("Function {} has a function hook but does not exist!", patch.func_name)); + } + + RecompPort::Function& func = context.functions[func_find->second]; + int32_t func_vram = func.vram; + + // Check that the function actually contains this vram address. + if (patch.before_vram < func_vram || patch.before_vram >= func_vram + func.words.size() * sizeof(func.words[0])) { + exit_failure(fmt::format("Function {} has a function hook for vram 0x{:08X} but doesn't contain that vram address!", patch.func_name, (uint32_t)patch.before_vram)); + } + + // No after_vram means this will be placed at the start of the function + size_t instruction_index = -1; + + // Calculate the instruction index. + if (patch.before_vram != 0) { + instruction_index = (static_cast(patch.before_vram) - func_vram) / sizeof(uint32_t); + } + + // Check if a function hook already exits for that instruction index. + auto hook_find = func.function_hooks.find(instruction_index); + if (hook_find != func.function_hooks.end()) { + exit_failure(fmt::format("Function {} already has a function hook for vram 0x{:08X}!", patch.func_name, (uint32_t)patch.before_vram)); + } + + func.function_hooks[instruction_index] = patch.text; + } + std::ofstream single_output_file; if (config.single_file_output) { @@ -1700,18 +1736,22 @@ int main(int argc, char** argv) { fmt::print(overlay_file, "static int overlay_sections_by_index[] = {{\n"); - for (const std::string& section : relocatable_sections_ordered) { - // Check if this is an empty overlay - if (section == "*") { - fmt::print(overlay_file, " -1,\n"); - } - else { - auto find_it = relocatable_section_indices.find(section); - if (find_it == relocatable_section_indices.end()) { - fmt::print(stderr, "Failed to find written section index of relocatable section: {}\n", section); - std::exit(EXIT_FAILURE); + if (relocatable_sections_ordered.empty()) { + fmt::print(overlay_file, " -1,\n"); + } else { + for (const std::string& section : relocatable_sections_ordered) { + // Check if this is an empty overlay + if (section == "*") { + fmt::print(overlay_file, " -1,\n"); + } + else { + auto find_it = relocatable_section_indices.find(section); + if (find_it == relocatable_section_indices.end()) { + fmt::print(stderr, "Failed to find written section index of relocatable section: {}\n", section); + std::exit(EXIT_FAILURE); + } + fmt::print(overlay_file, " {},\n", relocatable_section_indices[section]); } - fmt::print(overlay_file, " {},\n", relocatable_section_indices[section]); } } fmt::print(overlay_file, "}};\n"); diff --git a/src/recompilation.cpp b/src/recompilation.cpp index 4b29725..e34ed3c 100644 --- a/src/recompilation.cpp +++ b/src/recompilation.cpp @@ -24,23 +24,33 @@ bool process_instruction(const RecompPort::Context& context, const RecompPort::C const auto& instr = instructions[instr_index]; needs_link_branch = false; is_branch_likely = false; + uint32_t instr_vram = instr.getVram(); + + auto print_indent = [&]() { + fmt::print(output_file, " "); + }; + + auto hook_find = func.function_hooks.find(instr_index); + if (hook_find != func.function_hooks.end()) { + fmt::print(output_file, " {}\n", hook_find->second); + if (indent) { + print_indent(); + } + } // Output a comment with the original instruction if (instr.isBranch() || instr.getUniqueId() == InstrId::cpu_j) { - fmt::print(output_file, " // {}\n", instr.disassemble(0, fmt::format("L_{:08X}", (uint32_t)instr.getBranchVramGeneric()))); + fmt::print(output_file, " // 0x{:08X}: {}\n", instr_vram, instr.disassemble(0, fmt::format("L_{:08X}", (uint32_t)instr.getBranchVramGeneric()))); } else if (instr.getUniqueId() == InstrId::cpu_jal) { - fmt::print(output_file, " // {}\n", instr.disassemble(0, fmt::format("0x{:08X}", (uint32_t)instr.getBranchVramGeneric()))); + fmt::print(output_file, " // 0x{:08X}: {}\n", instr_vram, instr.disassemble(0, fmt::format("0x{:08X}", (uint32_t)instr.getBranchVramGeneric()))); } else { - fmt::print(output_file, " // {}\n", instr.disassemble(0)); + fmt::print(output_file, " // 0x{:08X}: {}\n", instr_vram, instr.disassemble(0)); } - uint32_t instr_vram = instr.getVram(); - if (skipped_insns.contains(instr_vram)) { return true; } - bool at_reloc = false; bool reloc_handled = false; RecompPort::RelocType reloc_type = RecompPort::RelocType::R_MIPS_NONE; @@ -71,10 +81,6 @@ bool process_instruction(const RecompPort::Context& context, const RecompPort::C } } - auto print_indent = [&]() { - fmt::print(output_file, " "); - }; - auto print_line = [&](fmt::format_string fmt_str, Ts ...args) { print_indent(); fmt::vprint(output_file, fmt_str, fmt::make_format_args(args...)); @@ -106,7 +112,7 @@ bool process_instruction(const RecompPort::Context& context, const RecompPort::C } }; - auto print_func_call = [&](uint32_t target_func_vram, bool link_branch = true) { + auto print_func_call = [&](uint32_t target_func_vram, bool link_branch = true, bool indent = false) { const auto matching_funcs_find = context.functions_by_vram.find(target_func_vram); std::string jal_target_name; uint32_t section_vram_start = section.ram_addr; @@ -173,7 +179,11 @@ bool process_instruction(const RecompPort::Context& context, const RecompPort::C } } needs_link_branch = link_branch; - print_unconditional_branch("{}(rdram, ctx)", jal_target_name); + if (indent) { + print_unconditional_branch(" {}(rdram, ctx)", jal_target_name); + } else { + print_unconditional_branch("{}(rdram, ctx)", jal_target_name); + } return true; }; @@ -183,9 +193,9 @@ bool process_instruction(const RecompPort::Context& context, const RecompPort::C if (context.functions_by_vram.find(branch_target) != context.functions_by_vram.end()) { fmt::print(output_file, "{{\n "); fmt::print("Tail call in {} to 0x{:08X}\n", func.name, branch_target); - print_func_call(branch_target, false); - print_line("return"); - fmt::print(output_file, ";\n }}\n"); + print_func_call(branch_target, false, true); + print_line(" return"); + fmt::print(output_file, " }}\n"); return; } @@ -1103,7 +1113,7 @@ bool RecompPort::recompile_function(const RecompPort::Context& context, const Re // these variables shouldn't need to be preserved across function boundaries, so make them local for more efficient output " uint64_t hi = 0, lo = 0, result = 0;\n" " unsigned int rounding_mode = DEFAULT_ROUNDING_MODE;\n" - " int c1cs = 0; \n", // cop1 conditional signal + " int c1cs = 0;\n", // cop1 conditional signal func.name); // Skip analysis and recompilation of this function is stubbed. @@ -1112,6 +1122,11 @@ bool RecompPort::recompile_function(const RecompPort::Context& context, const Re std::set branch_labels; instructions.reserve(func.words.size()); + auto hook_find = func.function_hooks.find(-1); + if (hook_find != func.function_hooks.end()) { + fmt::print(output_file, " {}\n", hook_find->second); + } + // First pass, disassemble each instruction and collect branch labels uint32_t vram = func.vram; for (uint32_t word : func.words) {