diff options
Diffstat (limited to 'src/core/arm')
| -rw-r--r-- | src/core/arm/arm_interface.cpp | 71 | ||||
| -rw-r--r-- | src/core/arm/arm_interface.h | 48 | ||||
| -rw-r--r-- | src/core/arm/cpu_interrupt_handler.cpp | 25 | ||||
| -rw-r--r-- | src/core/arm/cpu_interrupt_handler.h | 40 | ||||
| -rw-r--r-- | src/core/arm/dynarmic/arm_dynarmic_32.cpp | 139 | ||||
| -rw-r--r-- | src/core/arm/dynarmic/arm_dynarmic_32.h | 14 | ||||
| -rw-r--r-- | src/core/arm/dynarmic/arm_dynarmic_64.cpp | 170 | ||||
| -rw-r--r-- | src/core/arm/dynarmic/arm_dynarmic_64.h | 30 | ||||
| -rw-r--r-- | src/core/arm/dynarmic/arm_dynarmic_cp15.cpp | 87 | ||||
| -rw-r--r-- | src/core/arm/dynarmic/arm_dynarmic_cp15.h | 126 | ||||
| -rw-r--r-- | src/core/arm/dynarmic/arm_exclusive_monitor.cpp | 76 | ||||
| -rw-r--r-- | src/core/arm/dynarmic/arm_exclusive_monitor.h | 48 | ||||
| -rw-r--r-- | src/core/arm/exclusive_monitor.cpp | 2 | ||||
| -rw-r--r-- | src/core/arm/exclusive_monitor.h | 8 | ||||
| -rw-r--r-- | src/core/arm/unicorn/arm_unicorn.cpp | 274 | ||||
| -rw-r--r-- | src/core/arm/unicorn/arm_unicorn.h | 55 |
16 files changed, 602 insertions, 611 deletions
diff --git a/src/core/arm/arm_interface.cpp b/src/core/arm/arm_interface.cpp index 7e846ddd5..0951e1976 100644 --- a/src/core/arm/arm_interface.cpp +++ b/src/core/arm/arm_interface.cpp @@ -60,7 +60,7 @@ static_assert(sizeof(ELFSymbol) == 0x18, "ELFSymbol has incorrect size."); using Symbols = std::vector<std::pair<ELFSymbol, std::string>>; -Symbols GetSymbols(VAddr text_offset, Memory::Memory& memory) { +Symbols GetSymbols(VAddr text_offset, Core::Memory::Memory& memory) { const auto mod_offset = text_offset + memory.Read32(text_offset + 4); if (mod_offset < text_offset || (mod_offset & 0b11) != 0 || @@ -123,7 +123,7 @@ Symbols GetSymbols(VAddr text_offset, Memory::Memory& memory) { std::optional<std::string> GetSymbolName(const Symbols& symbols, VAddr func_address) { const auto iter = std::find_if(symbols.begin(), symbols.end(), [func_address](const auto& pair) { - const auto& [symbol, name] = pair; + const auto& symbol = pair.first; const auto end_address = symbol.value + symbol.size; return func_address >= symbol.value && func_address < end_address; }); @@ -139,6 +139,71 @@ std::optional<std::string> GetSymbolName(const Symbols& symbols, VAddr func_addr constexpr u64 SEGMENT_BASE = 0x7100000000ull; +std::vector<ARM_Interface::BacktraceEntry> ARM_Interface::GetBacktraceFromContext( + System& system, const ThreadContext64& ctx) { + std::vector<BacktraceEntry> out; + auto& memory = system.Memory(); + + auto fp = ctx.cpu_registers[29]; + auto lr = ctx.cpu_registers[30]; + while (true) { + out.push_back({ + .module = "", + .address = 0, + .original_address = lr, + .offset = 0, + .name = {}, + }); + + if (fp == 0) { + break; + } + + lr = memory.Read64(fp + 8) - 4; + fp = memory.Read64(fp); + } + + std::map<VAddr, std::string> modules; + auto& loader{system.GetAppLoader()}; + if (loader.ReadNSOModules(modules) != Loader::ResultStatus::Success) { + return {}; + } + + std::map<std::string, Symbols> symbols; + for (const auto& module : modules) { + symbols.insert_or_assign(module.second, GetSymbols(module.first, memory)); + } + + for (auto& entry : out) { + VAddr base = 0; + for (auto iter = modules.rbegin(); iter != modules.rend(); ++iter) { + const auto& module{*iter}; + if (entry.original_address >= module.first) { + entry.module = module.second; + base = module.first; + break; + } + } + + entry.offset = entry.original_address - base; + entry.address = SEGMENT_BASE + entry.offset; + + if (entry.module.empty()) + entry.module = "unknown"; + + const auto symbol_set = symbols.find(entry.module); + if (symbol_set != symbols.end()) { + const auto symbol = GetSymbolName(symbol_set->second, entry.offset); + if (symbol.has_value()) { + // TODO(DarkLordZach): Add demangling of symbol names. + entry.name = *symbol; + } + } + } + + return out; +} + std::vector<ARM_Interface::BacktraceEntry> ARM_Interface::GetBacktrace() const { std::vector<BacktraceEntry> out; auto& memory = system.Memory(); @@ -146,7 +211,7 @@ std::vector<ARM_Interface::BacktraceEntry> ARM_Interface::GetBacktrace() const { auto fp = GetReg(29); auto lr = GetReg(30); while (true) { - out.push_back({"", 0, lr, 0}); + out.push_back({"", 0, lr, 0, ""}); if (!fp) { break; } diff --git a/src/core/arm/arm_interface.h b/src/core/arm/arm_interface.h index 57eae839e..1f24051e4 100644 --- a/src/core/arm/arm_interface.h +++ b/src/core/arm/arm_interface.h @@ -7,6 +7,7 @@ #include <array> #include <vector> #include "common/common_types.h" +#include "core/hardware_properties.h" namespace Common { struct PageTable; @@ -18,36 +19,40 @@ enum class VMAPermission : u8; namespace Core { class System; +class CPUInterruptHandler; + +using CPUInterrupts = std::array<CPUInterruptHandler, Core::Hardware::NUM_CPU_CORES>; /// Generic ARMv8 CPU interface class ARM_Interface : NonCopyable { public: - explicit ARM_Interface(System& system_) : system{system_} {} + explicit ARM_Interface(System& system_, CPUInterrupts& interrupt_handlers, bool uses_wall_clock) + : system{system_}, interrupt_handlers{interrupt_handlers}, uses_wall_clock{ + uses_wall_clock} {} virtual ~ARM_Interface() = default; struct ThreadContext32 { - std::array<u32, 16> cpu_registers; - u32 cpsr; - std::array<u8, 4> padding; - std::array<u64, 32> fprs; - u32 fpscr; - u32 fpexc; - u32 tpidr; + std::array<u32, 16> cpu_registers{}; + std::array<u32, 64> extension_registers{}; + u32 cpsr{}; + u32 fpscr{}; + u32 fpexc{}; + u32 tpidr{}; }; // Internally within the kernel, it expects the AArch32 version of the // thread context to be 344 bytes in size. - static_assert(sizeof(ThreadContext32) == 0x158); + static_assert(sizeof(ThreadContext32) == 0x150); struct ThreadContext64 { - std::array<u64, 31> cpu_registers; - u64 sp; - u64 pc; - u32 pstate; - std::array<u8, 4> padding; - std::array<u128, 32> vector_registers; - u32 fpcr; - u32 fpsr; - u64 tpidr; + std::array<u64, 31> cpu_registers{}; + u64 sp{}; + u64 pc{}; + u32 pstate{}; + std::array<u8, 4> padding{}; + std::array<u128, 32> vector_registers{}; + u32 fpcr{}; + u32 fpsr{}; + u64 tpidr{}; }; // Internally within the kernel, it expects the AArch64 version of the // thread context to be 800 bytes in size. @@ -143,6 +148,8 @@ public: */ virtual void SetTPIDR_EL0(u64 value) = 0; + virtual void ChangeProcessorID(std::size_t new_core_id) = 0; + virtual void SaveContext(ThreadContext32& ctx) = 0; virtual void SaveContext(ThreadContext64& ctx) = 0; virtual void LoadContext(const ThreadContext32& ctx) = 0; @@ -162,6 +169,9 @@ public: std::string name; }; + static std::vector<BacktraceEntry> GetBacktraceFromContext(System& system, + const ThreadContext64& ctx); + std::vector<BacktraceEntry> GetBacktrace() const; /// fp (= r29) points to the last frame record. @@ -175,6 +185,8 @@ public: protected: /// System context that this ARM interface is running under. System& system; + CPUInterrupts& interrupt_handlers; + bool uses_wall_clock; }; } // namespace Core diff --git a/src/core/arm/cpu_interrupt_handler.cpp b/src/core/arm/cpu_interrupt_handler.cpp new file mode 100644 index 000000000..9c8898700 --- /dev/null +++ b/src/core/arm/cpu_interrupt_handler.cpp @@ -0,0 +1,25 @@ +// Copyright 2020 yuzu emulator team +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#include "common/thread.h" +#include "core/arm/cpu_interrupt_handler.h" + +namespace Core { + +CPUInterruptHandler::CPUInterruptHandler() : interrupt_event{std::make_unique<Common::Event>()} {} + +CPUInterruptHandler::~CPUInterruptHandler() = default; + +void CPUInterruptHandler::SetInterrupt(bool is_interrupted_) { + if (is_interrupted_) { + interrupt_event->Set(); + } + is_interrupted = is_interrupted_; +} + +void CPUInterruptHandler::AwaitInterrupt() { + interrupt_event->Wait(); +} + +} // namespace Core diff --git a/src/core/arm/cpu_interrupt_handler.h b/src/core/arm/cpu_interrupt_handler.h new file mode 100644 index 000000000..c20c280f1 --- /dev/null +++ b/src/core/arm/cpu_interrupt_handler.h @@ -0,0 +1,40 @@ +// Copyright 2020 yuzu emulator team +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#pragma once + +#include <atomic> +#include <memory> + +namespace Common { +class Event; +} + +namespace Core { + +class CPUInterruptHandler { +public: + CPUInterruptHandler(); + ~CPUInterruptHandler(); + + CPUInterruptHandler(const CPUInterruptHandler&) = delete; + CPUInterruptHandler& operator=(const CPUInterruptHandler&) = delete; + + CPUInterruptHandler(CPUInterruptHandler&&) = delete; + CPUInterruptHandler& operator=(CPUInterruptHandler&&) = delete; + + bool IsInterrupted() const { + return is_interrupted; + } + + void SetInterrupt(bool is_interrupted); + + void AwaitInterrupt(); + +private: + std::unique_ptr<Common::Event> interrupt_event; + std::atomic_bool is_interrupted{false}; +}; + +} // namespace Core diff --git a/src/core/arm/dynarmic/arm_dynarmic_32.cpp b/src/core/arm/dynarmic/arm_dynarmic_32.cpp index 187a972ac..6dc03f3b1 100644 --- a/src/core/arm/dynarmic/arm_dynarmic_32.cpp +++ b/src/core/arm/dynarmic/arm_dynarmic_32.cpp @@ -7,15 +7,18 @@ #include <dynarmic/A32/a32.h> #include <dynarmic/A32/config.h> #include <dynarmic/A32/context.h> -#include "common/microprofile.h" +#include "common/assert.h" +#include "common/logging/log.h" +#include "common/page_table.h" +#include "core/arm/cpu_interrupt_handler.h" #include "core/arm/dynarmic/arm_dynarmic_32.h" -#include "core/arm/dynarmic/arm_dynarmic_64.h" #include "core/arm/dynarmic/arm_dynarmic_cp15.h" +#include "core/arm/dynarmic/arm_exclusive_monitor.h" #include "core/core.h" -#include "core/core_manager.h" #include "core/core_timing.h" #include "core/hle/kernel/svc.h" #include "core/memory.h" +#include "core/settings.h" namespace Core { @@ -49,8 +52,22 @@ public: parent.system.Memory().Write64(vaddr, value); } + bool MemoryWriteExclusive8(u32 vaddr, u8 value, u8 expected) override { + return parent.system.Memory().WriteExclusive8(vaddr, value, expected); + } + bool MemoryWriteExclusive16(u32 vaddr, u16 value, u16 expected) override { + return parent.system.Memory().WriteExclusive16(vaddr, value, expected); + } + bool MemoryWriteExclusive32(u32 vaddr, u32 value, u32 expected) override { + return parent.system.Memory().WriteExclusive32(vaddr, value, expected); + } + bool MemoryWriteExclusive64(u32 vaddr, u64 value, u64 expected) override { + return parent.system.Memory().WriteExclusive64(vaddr, value, expected); + } + void InterpreterFallback(u32 pc, std::size_t num_instructions) override { - UNIMPLEMENTED(); + UNIMPLEMENTED_MSG("This should never happen, pc = {:08X}, code = {:08X}", pc, + MemoryReadCode(pc)); } void ExceptionRaised(u32 pc, Dynarmic::A32::Exception exception) override { @@ -61,36 +78,46 @@ public: case Dynarmic::A32::Exception::Breakpoint: break; } - LOG_CRITICAL(HW_GPU, "ExceptionRaised(exception = {}, pc = {:08X}, code = {:08X})", + LOG_CRITICAL(Core_ARM, "ExceptionRaised(exception = {}, pc = {:08X}, code = {:08X})", static_cast<std::size_t>(exception), pc, MemoryReadCode(pc)); UNIMPLEMENTED(); } void CallSVC(u32 swi) override { - Kernel::CallSVC(parent.system, swi); + Kernel::Svc::Call(parent.system, swi); } void AddTicks(u64 ticks) override { + if (parent.uses_wall_clock) { + return; + } // Divide the number of ticks by the amount of CPU cores. TODO(Subv): This yields only a // rough approximation of the amount of executed ticks in the system, it may be thrown off // if not all cores are doing a similar amount of work. Instead of doing this, we should // device a way so that timing is consistent across all cores without increasing the ticks 4 // times. - u64 amortized_ticks = (ticks - num_interpreted_instructions) / Core::NUM_CPU_CORES; + u64 amortized_ticks = + (ticks - num_interpreted_instructions) / Core::Hardware::NUM_CPU_CORES; // Always execute at least one tick. amortized_ticks = std::max<u64>(amortized_ticks, 1); parent.system.CoreTiming().AddTicks(amortized_ticks); num_interpreted_instructions = 0; } + u64 GetTicksRemaining() override { - return std::max(parent.system.CoreTiming().GetDowncount(), {}); + if (parent.uses_wall_clock) { + if (!parent.interrupt_handlers[parent.core_index].IsInterrupted()) { + return minimum_run_cycles; + } + return 0U; + } + return std::max<s64>(parent.system.CoreTiming().GetDowncount(), 0); } ARM_Dynarmic_32& parent; std::size_t num_interpreted_instructions{}; - u64 tpidrro_el0{}; - u64 tpidr_el0{}; + static constexpr u64 minimum_run_cycles = 1000U; }; std::shared_ptr<Dynarmic::A32::Jit> ARM_Dynarmic_32::MakeJit(Common::PageTable& page_table, @@ -99,26 +126,79 @@ std::shared_ptr<Dynarmic::A32::Jit> ARM_Dynarmic_32::MakeJit(Common::PageTable& config.callbacks = cb.get(); // TODO(bunnei): Implement page table for 32-bit // config.page_table = &page_table.pointers; - config.coprocessors[15] = std::make_shared<DynarmicCP15>((u32*)&CP15_regs[0]); + config.coprocessors[15] = cp15; config.define_unpredictable_behaviour = true; + static constexpr std::size_t PAGE_BITS = 12; + static constexpr std::size_t NUM_PAGE_TABLE_ENTRIES = 1 << (32 - PAGE_BITS); + config.page_table = reinterpret_cast<std::array<std::uint8_t*, NUM_PAGE_TABLE_ENTRIES>*>( + page_table.pointers.data()); + config.absolute_offset_page_table = true; + config.detect_misaligned_access_via_page_table = 16 | 32 | 64 | 128; + config.only_detect_misalignment_via_page_table_on_page_boundary = true; + + // Multi-process state + config.processor_id = core_index; + config.global_monitor = &exclusive_monitor.monitor; + + // Timing + config.wall_clock_cntpct = uses_wall_clock; + + // Safe optimizations + if (Settings::values.cpu_accuracy == Settings::CPUAccuracy::DebugMode) { + if (!Settings::values.cpuopt_page_tables) { + config.page_table = nullptr; + } + if (!Settings::values.cpuopt_block_linking) { + config.optimizations &= ~Dynarmic::OptimizationFlag::BlockLinking; + } + if (!Settings::values.cpuopt_return_stack_buffer) { + config.optimizations &= ~Dynarmic::OptimizationFlag::ReturnStackBuffer; + } + if (!Settings::values.cpuopt_fast_dispatcher) { + config.optimizations &= ~Dynarmic::OptimizationFlag::FastDispatch; + } + if (!Settings::values.cpuopt_context_elimination) { + config.optimizations &= ~Dynarmic::OptimizationFlag::GetSetElimination; + } + if (!Settings::values.cpuopt_const_prop) { + config.optimizations &= ~Dynarmic::OptimizationFlag::ConstProp; + } + if (!Settings::values.cpuopt_misc_ir) { + config.optimizations &= ~Dynarmic::OptimizationFlag::MiscIROpt; + } + if (!Settings::values.cpuopt_reduce_misalign_checks) { + config.only_detect_misalignment_via_page_table_on_page_boundary = false; + } + } + + // Unsafe optimizations + if (Settings::values.cpu_accuracy == Settings::CPUAccuracy::Unsafe) { + config.unsafe_optimizations = true; + if (Settings::values.cpuopt_unsafe_unfuse_fma) { + config.optimizations |= Dynarmic::OptimizationFlag::Unsafe_UnfuseFMA; + } + if (Settings::values.cpuopt_unsafe_reduce_fp_error) { + config.optimizations |= Dynarmic::OptimizationFlag::Unsafe_ReducedErrorFP; + } + } + return std::make_unique<Dynarmic::A32::Jit>(config); } -MICROPROFILE_DEFINE(ARM_Jit_Dynarmic_32, "ARM JIT", "Dynarmic", MP_RGB(255, 64, 64)); - void ARM_Dynarmic_32::Run() { - MICROPROFILE_SCOPE(ARM_Jit_Dynarmic_32); jit->Run(); } void ARM_Dynarmic_32::Step() { - cb->InterpreterFallback(jit->Regs()[15], 1); + jit->Step(); } -ARM_Dynarmic_32::ARM_Dynarmic_32(System& system, ExclusiveMonitor& exclusive_monitor, +ARM_Dynarmic_32::ARM_Dynarmic_32(System& system, CPUInterrupts& interrupt_handlers, + bool uses_wall_clock, ExclusiveMonitor& exclusive_monitor, std::size_t core_index) - : ARM_Interface{system}, - cb(std::make_unique<DynarmicCallbacks32>(*this)), core_index{core_index}, + : ARM_Interface{system, interrupt_handlers, uses_wall_clock}, + cb(std::make_unique<DynarmicCallbacks32>(*this)), + cp15(std::make_shared<DynarmicCP15>(*this)), core_index{core_index}, exclusive_monitor{dynamic_cast<DynarmicExclusiveMonitor&>(exclusive_monitor)} {} ARM_Dynarmic_32::~ARM_Dynarmic_32() = default; @@ -154,32 +234,40 @@ void ARM_Dynarmic_32::SetPSTATE(u32 cpsr) { } u64 ARM_Dynarmic_32::GetTlsAddress() const { - return CP15_regs[static_cast<std::size_t>(CP15Register::CP15_THREAD_URO)]; + return cp15->uro; } void ARM_Dynarmic_32::SetTlsAddress(VAddr address) { - CP15_regs[static_cast<std::size_t>(CP15Register::CP15_THREAD_URO)] = static_cast<u32>(address); + cp15->uro = static_cast<u32>(address); } u64 ARM_Dynarmic_32::GetTPIDR_EL0() const { - return cb->tpidr_el0; + return cp15->uprw; } void ARM_Dynarmic_32::SetTPIDR_EL0(u64 value) { - cb->tpidr_el0 = value; + cp15->uprw = static_cast<u32>(value); +} + +void ARM_Dynarmic_32::ChangeProcessorID(std::size_t new_core_id) { + jit->ChangeProcessorID(new_core_id); } void ARM_Dynarmic_32::SaveContext(ThreadContext32& ctx) { Dynarmic::A32::Context context; jit->SaveContext(context); ctx.cpu_registers = context.Regs(); + ctx.extension_registers = context.ExtRegs(); ctx.cpsr = context.Cpsr(); + ctx.fpscr = context.Fpscr(); } void ARM_Dynarmic_32::LoadContext(const ThreadContext32& ctx) { Dynarmic::A32::Context context; context.Regs() = ctx.cpu_registers; + context.ExtRegs() = ctx.extension_registers; context.SetCpsr(ctx.cpsr); + context.SetFpscr(ctx.fpscr); jit->LoadContext(context); } @@ -188,10 +276,15 @@ void ARM_Dynarmic_32::PrepareReschedule() { } void ARM_Dynarmic_32::ClearInstructionCache() { + if (!jit) { + return; + } jit->ClearCache(); } -void ARM_Dynarmic_32::ClearExclusiveState() {} +void ARM_Dynarmic_32::ClearExclusiveState() { + jit->ClearExclusiveState(); +} void ARM_Dynarmic_32::PageTableChanged(Common::PageTable& page_table, std::size_t new_address_space_size_in_bits) { diff --git a/src/core/arm/dynarmic/arm_dynarmic_32.h b/src/core/arm/dynarmic/arm_dynarmic_32.h index 143e46e4d..2bab31b92 100644 --- a/src/core/arm/dynarmic/arm_dynarmic_32.h +++ b/src/core/arm/dynarmic/arm_dynarmic_32.h @@ -9,25 +9,28 @@ #include <dynarmic/A32/a32.h> #include <dynarmic/A64/a64.h> -#include <dynarmic/A64/exclusive_monitor.h> +#include <dynarmic/exclusive_monitor.h> #include "common/common_types.h" #include "common/hash.h" #include "core/arm/arm_interface.h" #include "core/arm/exclusive_monitor.h" -namespace Memory { +namespace Core::Memory { class Memory; } namespace Core { +class CPUInterruptHandler; class DynarmicCallbacks32; +class DynarmicCP15; class DynarmicExclusiveMonitor; class System; class ARM_Dynarmic_32 final : public ARM_Interface { public: - ARM_Dynarmic_32(System& system, ExclusiveMonitor& exclusive_monitor, std::size_t core_index); + ARM_Dynarmic_32(System& system, CPUInterrupts& interrupt_handlers, bool uses_wall_clock, + ExclusiveMonitor& exclusive_monitor, std::size_t core_index); ~ARM_Dynarmic_32() override; void SetPC(u64 pc) override; @@ -44,6 +47,7 @@ public: void SetTlsAddress(VAddr address) override; void SetTPIDR_EL0(u64 value) override; u64 GetTPIDR_EL0() const override; + void ChangeProcessorID(std::size_t new_core_id) override; void SaveContext(ThreadContext32& ctx) override; void SaveContext(ThreadContext64& ctx) override {} @@ -66,12 +70,14 @@ private: std::unordered_map<JitCacheKey, std::shared_ptr<Dynarmic::A32::Jit>, Common::PairHash>; friend class DynarmicCallbacks32; + friend class DynarmicCP15; + std::unique_ptr<DynarmicCallbacks32> cb; JitCacheType jit_cache; std::shared_ptr<Dynarmic::A32::Jit> jit; + std::shared_ptr<DynarmicCP15> cp15; std::size_t core_index; DynarmicExclusiveMonitor& exclusive_monitor; - std::array<u32, 84> CP15_regs{}; }; } // namespace Core diff --git a/src/core/arm/dynarmic/arm_dynarmic_64.cpp b/src/core/arm/dynarmic/arm_dynarmic_64.cpp index a53a58ba0..9f170a224 100644 --- a/src/core/arm/dynarmic/arm_dynarmic_64.cpp +++ b/src/core/arm/dynarmic/arm_dynarmic_64.cpp @@ -6,20 +6,21 @@ #include <memory> #include <dynarmic/A64/a64.h> #include <dynarmic/A64/config.h> +#include "common/assert.h" #include "common/logging/log.h" -#include "common/microprofile.h" +#include "common/page_table.h" +#include "core/arm/cpu_interrupt_handler.h" #include "core/arm/dynarmic/arm_dynarmic_64.h" +#include "core/arm/dynarmic/arm_exclusive_monitor.h" #include "core/core.h" -#include "core/core_manager.h" #include "core/core_timing.h" -#include "core/core_timing_util.h" #include "core/gdbstub/gdbstub.h" #include "core/hardware_properties.h" #include "core/hle/kernel/process.h" #include "core/hle/kernel/scheduler.h" #include "core/hle/kernel/svc.h" -#include "core/hle/kernel/vm_manager.h" #include "core/memory.h" +#include "core/settings.h" namespace Core { @@ -64,17 +65,26 @@ public: memory.Write64(vaddr + 8, value[1]); } + bool MemoryWriteExclusive8(u64 vaddr, std::uint8_t value, std::uint8_t expected) override { + return parent.system.Memory().WriteExclusive8(vaddr, value, expected); + } + bool MemoryWriteExclusive16(u64 vaddr, std::uint16_t value, std::uint16_t expected) override { + return parent.system.Memory().WriteExclusive16(vaddr, value, expected); + } + bool MemoryWriteExclusive32(u64 vaddr, std::uint32_t value, std::uint32_t expected) override { + return parent.system.Memory().WriteExclusive32(vaddr, value, expected); + } + bool MemoryWriteExclusive64(u64 vaddr, std::uint64_t value, std::uint64_t expected) override { + return parent.system.Memory().WriteExclusive64(vaddr, value, expected); + } + bool MemoryWriteExclusive128(u64 vaddr, Vector value, Vector expected) override { + return parent.system.Memory().WriteExclusive128(vaddr, value, expected); + } + void InterpreterFallback(u64 pc, std::size_t num_instructions) override { - LOG_INFO(Core_ARM, "Unicorn fallback @ 0x{:X} for {} instructions (instr = {:08X})", pc, - num_instructions, MemoryReadCode(pc)); - - ARM_Interface::ThreadContext64 ctx; - parent.SaveContext(ctx); - parent.inner_unicorn.LoadContext(ctx); - parent.inner_unicorn.ExecuteInstructions(num_instructions); - parent.inner_unicorn.SaveContext(ctx); - parent.LoadContext(ctx); - num_interpreted_instructions += num_instructions; + LOG_ERROR(Core_ARM, + "Unimplemented instruction @ 0x{:X} for {} instructions (instr = {:08X})", pc, + num_instructions, MemoryReadCode(pc)); } void ExceptionRaised(u64 pc, Dynarmic::A64::Exception exception) override { @@ -97,39 +107,50 @@ public: } [[fallthrough]]; default: - ASSERT_MSG(false, "ExceptionRaised(exception = {}, pc = {:X})", - static_cast<std::size_t>(exception), pc); + ASSERT_MSG(false, "ExceptionRaised(exception = {}, pc = {:08X}, code = {:08X})", + static_cast<std::size_t>(exception), pc, MemoryReadCode(pc)); } } void CallSVC(u32 swi) override { - Kernel::CallSVC(parent.system, swi); + Kernel::Svc::Call(parent.system, swi); } void AddTicks(u64 ticks) override { + if (parent.uses_wall_clock) { + return; + } + // Divide the number of ticks by the amount of CPU cores. TODO(Subv): This yields only a // rough approximation of the amount of executed ticks in the system, it may be thrown off // if not all cores are doing a similar amount of work. Instead of doing this, we should // device a way so that timing is consistent across all cores without increasing the ticks 4 // times. - u64 amortized_ticks = (ticks - num_interpreted_instructions) / Core::NUM_CPU_CORES; + u64 amortized_ticks = ticks / Core::Hardware::NUM_CPU_CORES; // Always execute at least one tick. amortized_ticks = std::max<u64>(amortized_ticks, 1); parent.system.CoreTiming().AddTicks(amortized_ticks); - num_interpreted_instructions = 0; } + u64 GetTicksRemaining() override { - return std::max(parent.system.CoreTiming().GetDowncount(), s64{0}); + if (parent.uses_wall_clock) { + if (!parent.interrupt_handlers[parent.core_index].IsInterrupted()) { + return minimum_run_cycles; + } + return 0U; + } + return std::max<s64>(parent.system.CoreTiming().GetDowncount(), 0); } + u64 GetCNTPCT() override { - return Timing::CpuCyclesToClockCycles(parent.system.CoreTiming().GetTicks()); + return parent.system.CoreTiming().GetClockTicks(); } ARM_Dynarmic_64& parent; - std::size_t num_interpreted_instructions = 0; u64 tpidrro_el0 = 0; u64 tpidr_el0 = 0; + static constexpr u64 minimum_run_cycles = 1000U; }; std::shared_ptr<Dynarmic::A64::Jit> ARM_Dynarmic_64::MakeJit(Common::PageTable& page_table, @@ -144,6 +165,8 @@ std::shared_ptr<Dynarmic::A64::Jit> ARM_Dynarmic_64::MakeJit(Common::PageTable& config.page_table_address_space_bits = address_space_bits; config.silently_mirror_page_table = false; config.absolute_offset_page_table = true; + config.detect_misaligned_access_via_page_table = 16 | 32 | 64 | 128; + config.only_detect_misalignment_via_page_table_on_page_boundary = true; // Multi-process state config.processor_id = core_index; @@ -159,14 +182,52 @@ std::shared_ptr<Dynarmic::A64::Jit> ARM_Dynarmic_64::MakeJit(Common::PageTable& // Unpredictable instructions config.define_unpredictable_behaviour = true; + // Timing + config.wall_clock_cntpct = uses_wall_clock; + + // Safe optimizations + if (Settings::values.cpu_accuracy == Settings::CPUAccuracy::DebugMode) { + if (!Settings::values.cpuopt_page_tables) { + config.page_table = nullptr; + } + if (!Settings::values.cpuopt_block_linking) { + config.optimizations &= ~Dynarmic::OptimizationFlag::BlockLinking; + } + if (!Settings::values.cpuopt_return_stack_buffer) { + config.optimizations &= ~Dynarmic::OptimizationFlag::ReturnStackBuffer; + } + if (!Settings::values.cpuopt_fast_dispatcher) { + config.optimizations &= ~Dynarmic::OptimizationFlag::FastDispatch; + } + if (!Settings::values.cpuopt_context_elimination) { + config.optimizations &= ~Dynarmic::OptimizationFlag::GetSetElimination; + } + if (!Settings::values.cpuopt_const_prop) { + config.optimizations &= ~Dynarmic::OptimizationFlag::ConstProp; + } + if (!Settings::values.cpuopt_misc_ir) { + config.optimizations &= ~Dynarmic::OptimizationFlag::MiscIROpt; + } + if (!Settings::values.cpuopt_reduce_misalign_checks) { + config.only_detect_misalignment_via_page_table_on_page_boundary = false; + } + } + + // Unsafe optimizations + if (Settings::values.cpu_accuracy == Settings::CPUAccuracy::Unsafe) { + config.unsafe_optimizations = true; + if (Settings::values.cpuopt_unsafe_unfuse_fma) { + config.optimizations |= Dynarmic::OptimizationFlag::Unsafe_UnfuseFMA; + } + if (Settings::values.cpuopt_unsafe_reduce_fp_error) { + config.optimizations |= Dynarmic::OptimizationFlag::Unsafe_ReducedErrorFP; + } + } + return std::make_shared<Dynarmic::A64::Jit>(config); } -MICROPROFILE_DEFINE(ARM_Jit_Dynarmic_64, "ARM JIT", "Dynarmic", MP_RGB(255, 64, 64)); - void ARM_Dynarmic_64::Run() { - MICROPROFILE_SCOPE(ARM_Jit_Dynarmic_64); - jit->Run(); } @@ -174,12 +235,12 @@ void ARM_Dynarmic_64::Step() { cb->InterpreterFallback(jit->GetPC(), 1); } -ARM_Dynarmic_64::ARM_Dynarmic_64(System& system, ExclusiveMonitor& exclusive_monitor, +ARM_Dynarmic_64::ARM_Dynarmic_64(System& system, CPUInterrupts& interrupt_handlers, + bool uses_wall_clock, ExclusiveMonitor& exclusive_monitor, std::size_t core_index) - : ARM_Interface{system}, - cb(std::make_unique<DynarmicCallbacks64>(*this)), inner_unicorn{system}, - core_index{core_index}, exclusive_monitor{ - dynamic_cast<DynarmicExclusiveMonitor&>(exclusive_monitor)} {} + : ARM_Interface{system, interrupt_handlers, uses_wall_clock}, + cb(std::make_unique<DynarmicCallbacks64>(*this)), core_index{core_index}, + exclusive_monitor{dynamic_cast<DynarmicExclusiveMonitor&>(exclusive_monitor)} {} ARM_Dynarmic_64::~ARM_Dynarmic_64() = default; @@ -231,6 +292,10 @@ void ARM_Dynarmic_64::SetTPIDR_EL0(u64 value) { cb->tpidr_el0 = value; } +void ARM_Dynarmic_64::ChangeProcessorID(std::size_t new_core_id) { + jit->ChangeProcessorID(new_core_id); +} + void ARM_Dynarmic_64::SaveContext(ThreadContext64& ctx) { ctx.cpu_registers = jit->GetRegisters(); ctx.sp = jit->GetSP(); @@ -258,6 +323,9 @@ void ARM_Dynarmic_64::PrepareReschedule() { } void ARM_Dynarmic_64::ClearInstructionCache() { + if (!jit) { + return; + } jit->ClearCache(); } @@ -277,44 +345,4 @@ void ARM_Dynarmic_64::PageTableChanged(Common::PageTable& page_table, jit_cache.emplace(key, jit); } -DynarmicExclusiveMonitor::DynarmicExclusiveMonitor(Memory::Memory& memory, std::size_t core_count) - : monitor(core_count), memory{memory} {} - -DynarmicExclusiveMonitor::~DynarmicExclusiveMonitor() = default; - -void DynarmicExclusiveMonitor::SetExclusive(std::size_t core_index, VAddr addr) { - // Size doesn't actually matter. - monitor.Mark(core_index, addr, 16); -} - -void DynarmicExclusiveMonitor::ClearExclusive() { - monitor.Clear(); -} - -bool DynarmicExclusiveMonitor::ExclusiveWrite8(std::size_t core_index, VAddr vaddr, u8 value) { - return monitor.DoExclusiveOperation(core_index, vaddr, 1, [&] { memory.Write8(vaddr, value); }); -} - -bool DynarmicExclusiveMonitor::ExclusiveWrite16(std::size_t core_index, VAddr vaddr, u16 value) { - return monitor.DoExclusiveOperation(core_index, vaddr, 2, - [&] { memory.Write16(vaddr, value); }); -} - -bool DynarmicExclusiveMonitor::ExclusiveWrite32(std::size_t core_index, VAddr vaddr, u32 value) { - return monitor.DoExclusiveOperation(core_index, vaddr, 4, - [&] { memory.Write32(vaddr, value); }); -} - -bool DynarmicExclusiveMonitor::ExclusiveWrite64(std::size_t core_index, VAddr vaddr, u64 value) { - return monitor.DoExclusiveOperation(core_index, vaddr, 8, - [&] { memory.Write64(vaddr, value); }); -} - -bool DynarmicExclusiveMonitor::ExclusiveWrite128(std::size_t core_index, VAddr vaddr, u128 value) { - return monitor.DoExclusiveOperation(core_index, vaddr, 16, [&] { - memory.Write64(vaddr + 0, value[0]); - memory.Write64(vaddr + 8, value[1]); - }); -} - } // namespace Core diff --git a/src/core/arm/dynarmic/arm_dynarmic_64.h b/src/core/arm/dynarmic/arm_dynarmic_64.h index e71240a96..28e11a17d 100644 --- a/src/core/arm/dynarmic/arm_dynarmic_64.h +++ b/src/core/arm/dynarmic/arm_dynarmic_64.h @@ -8,26 +8,26 @@ #include <unordered_map> #include <dynarmic/A64/a64.h> -#include <dynarmic/A64/exclusive_monitor.h> #include "common/common_types.h" #include "common/hash.h" #include "core/arm/arm_interface.h" #include "core/arm/exclusive_monitor.h" -#include "core/arm/unicorn/arm_unicorn.h" -namespace Memory { +namespace Core::Memory { class Memory; } namespace Core { class DynarmicCallbacks64; +class CPUInterruptHandler; class DynarmicExclusiveMonitor; class System; class ARM_Dynarmic_64 final : public ARM_Interface { public: - ARM_Dynarmic_64(System& system, ExclusiveMonitor& exclusive_monitor, std::size_t core_index); + ARM_Dynarmic_64(System& system, CPUInterrupts& interrupt_handlers, bool uses_wall_clock, + ExclusiveMonitor& exclusive_monitor, std::size_t core_index); ~ARM_Dynarmic_64() override; void SetPC(u64 pc) override; @@ -44,6 +44,7 @@ public: void SetTlsAddress(VAddr address) override; void SetTPIDR_EL0(u64 value) override; u64 GetTPIDR_EL0() const override; + void ChangeProcessorID(std::size_t new_core_id) override; void SaveContext(ThreadContext32& ctx) override {} void SaveContext(ThreadContext64& ctx) override; @@ -69,30 +70,9 @@ private: std::unique_ptr<DynarmicCallbacks64> cb; JitCacheType jit_cache; std::shared_ptr<Dynarmic::A64::Jit> jit; - ARM_Unicorn inner_unicorn; std::size_t core_index; DynarmicExclusiveMonitor& exclusive_monitor; }; -class DynarmicExclusiveMonitor final : public ExclusiveMonitor { -public: - explicit DynarmicExclusiveMonitor(Memory::Memory& memory, std::size_t core_count); - ~DynarmicExclusiveMonitor() override; - - void SetExclusive(std::size_t core_index, VAddr addr) override; - void ClearExclusive() override; - - bool ExclusiveWrite8(std::size_t core_index, VAddr vaddr, u8 value) override; - bool ExclusiveWrite16(std::size_t core_index, VAddr vaddr, u16 value) override; - bool ExclusiveWrite32(std::size_t core_index, VAddr vaddr, u32 value) override; - bool ExclusiveWrite64(std::size_t core_index, VAddr vaddr, u64 value) override; - bool ExclusiveWrite128(std::size_t core_index, VAddr vaddr, u128 value) override; - -private: - friend class ARM_Dynarmic_64; - Dynarmic::A64::ExclusiveMonitor monitor; - Memory::Memory& memory; -}; - } // namespace Core diff --git a/src/core/arm/dynarmic/arm_dynarmic_cp15.cpp b/src/core/arm/dynarmic/arm_dynarmic_cp15.cpp index 3fdcdebde..caefc09f4 100644 --- a/src/core/arm/dynarmic/arm_dynarmic_cp15.cpp +++ b/src/core/arm/dynarmic/arm_dynarmic_cp15.cpp @@ -2,79 +2,132 @@ // Licensed under GPLv2 or any later version // Refer to the license.txt file included. +#include <fmt/format.h> +#include "common/logging/log.h" +#include "core/arm/dynarmic/arm_dynarmic_32.h" #include "core/arm/dynarmic/arm_dynarmic_cp15.h" +#include "core/core.h" +#include "core/core_timing.h" +#include "core/core_timing_util.h" using Callback = Dynarmic::A32::Coprocessor::Callback; using CallbackOrAccessOneWord = Dynarmic::A32::Coprocessor::CallbackOrAccessOneWord; using CallbackOrAccessTwoWords = Dynarmic::A32::Coprocessor::CallbackOrAccessTwoWords; +template <> +struct fmt::formatter<Dynarmic::A32::CoprocReg> { + constexpr auto parse(format_parse_context& ctx) { + return ctx.begin(); + } + template <typename FormatContext> + auto format(const Dynarmic::A32::CoprocReg& reg, FormatContext& ctx) { + return format_to(ctx.out(), "cp{}", static_cast<size_t>(reg)); + } +}; + +namespace Core { + +static u32 dummy_value; + std::optional<Callback> DynarmicCP15::CompileInternalOperation(bool two, unsigned opc1, CoprocReg CRd, CoprocReg CRn, CoprocReg CRm, unsigned opc2) { - return {}; + LOG_CRITICAL(Core_ARM, "CP15: cdp{} p15, {}, {}, {}, {}, {}", two ? "2" : "", opc1, CRd, CRn, + CRm, opc2); + return std::nullopt; } CallbackOrAccessOneWord DynarmicCP15::CompileSendOneWord(bool two, unsigned opc1, CoprocReg CRn, CoprocReg CRm, unsigned opc2) { - // TODO(merry): Privileged CP15 registers - if (!two && CRn == CoprocReg::C7 && opc1 == 0 && CRm == CoprocReg::C5 && opc2 == 4) { + // CP15_FLUSH_PREFETCH_BUFFER // This is a dummy write, we ignore the value written here. - return &CP15[static_cast<std::size_t>(CP15Register::CP15_FLUSH_PREFETCH_BUFFER)]; + return &dummy_value; } if (!two && CRn == CoprocReg::C7 && opc1 == 0 && CRm == CoprocReg::C10) { switch (opc2) { case 4: + // CP15_DATA_SYNC_BARRIER // This is a dummy write, we ignore the value written here. - return &CP15[static_cast<std::size_t>(CP15Register::CP15_DATA_SYNC_BARRIER)]; + return &dummy_value; case 5: + // CP15_DATA_MEMORY_BARRIER // This is a dummy write, we ignore the value written here. - return &CP15[static_cast<std::size_t>(CP15Register::CP15_DATA_MEMORY_BARRIER)]; - default: - return {}; + return &dummy_value; } } if (!two && CRn == CoprocReg::C13 && opc1 == 0 && CRm == CoprocReg::C0 && opc2 == 2) { - return &CP15[static_cast<std::size_t>(CP15Register::CP15_THREAD_UPRW)]; + // CP15_THREAD_UPRW + return &uprw; } + LOG_CRITICAL(Core_ARM, "CP15: mcr{} p15, {}, <Rt>, {}, {}, {}", two ? "2" : "", opc1, CRn, CRm, + opc2); return {}; } CallbackOrAccessTwoWords DynarmicCP15::CompileSendTwoWords(bool two, unsigned opc, CoprocReg CRm) { + LOG_CRITICAL(Core_ARM, "CP15: mcrr{} p15, {}, <Rt>, <Rt2>, {}", two ? "2" : "", opc, CRm); return {}; } CallbackOrAccessOneWord DynarmicCP15::CompileGetOneWord(bool two, unsigned opc1, CoprocReg CRn, CoprocReg CRm, unsigned opc2) { - // TODO(merry): Privileged CP15 registers - if (!two && CRn == CoprocReg::C13 && opc1 == 0 && CRm == CoprocReg::C0) { switch (opc2) { case 2: - return &CP15[static_cast<std::size_t>(CP15Register::CP15_THREAD_UPRW)]; + // CP15_THREAD_UPRW + return &uprw; case 3: - return &CP15[static_cast<std::size_t>(CP15Register::CP15_THREAD_URO)]; - default: - return {}; + // CP15_THREAD_URO + return &uro; } } + LOG_CRITICAL(Core_ARM, "CP15: mrc{} p15, {}, <Rt>, {}, {}, {}", two ? "2" : "", opc1, CRn, CRm, + opc2); return {}; } CallbackOrAccessTwoWords DynarmicCP15::CompileGetTwoWords(bool two, unsigned opc, CoprocReg CRm) { + if (!two && opc == 0 && CRm == CoprocReg::C14) { + // CNTPCT + const auto callback = static_cast<u64 (*)(Dynarmic::A32::Jit*, void*, u32, u32)>( + [](Dynarmic::A32::Jit*, void* arg, u32, u32) -> u64 { + ARM_Dynarmic_32& parent = *(ARM_Dynarmic_32*)arg; + return parent.system.CoreTiming().GetClockTicks(); + }); + return Dynarmic::A32::Coprocessor::Callback{callback, (void*)&parent}; + } + + LOG_CRITICAL(Core_ARM, "CP15: mrrc{} p15, {}, <Rt>, <Rt2>, {}", two ? "2" : "", opc, CRm); return {}; } std::optional<Callback> DynarmicCP15::CompileLoadWords(bool two, bool long_transfer, CoprocReg CRd, std::optional<u8> option) { - return {}; + if (option) { + LOG_CRITICAL(Core_ARM, "CP15: mrrc{}{} p15, {}, [...], {}", two ? "2" : "", + long_transfer ? "l" : "", CRd, *option); + } else { + LOG_CRITICAL(Core_ARM, "CP15: mrrc{}{} p15, {}, [...]", two ? "2" : "", + long_transfer ? "l" : "", CRd); + } + return std::nullopt; } std::optional<Callback> DynarmicCP15::CompileStoreWords(bool two, bool long_transfer, CoprocReg CRd, std::optional<u8> option) { - return {}; + if (option) { + LOG_CRITICAL(Core_ARM, "CP15: mrrc{}{} p15, {}, [...], {}", two ? "2" : "", + long_transfer ? "l" : "", CRd, *option); + } else { + LOG_CRITICAL(Core_ARM, "CP15: mrrc{}{} p15, {}, [...]", two ? "2" : "", + long_transfer ? "l" : "", CRd); + } + return std::nullopt; } + +} // namespace Core diff --git a/src/core/arm/dynarmic/arm_dynarmic_cp15.h b/src/core/arm/dynarmic/arm_dynarmic_cp15.h index 07bcde5f9..dc6f4af3a 100644 --- a/src/core/arm/dynarmic/arm_dynarmic_cp15.h +++ b/src/core/arm/dynarmic/arm_dynarmic_cp15.h @@ -10,128 +10,15 @@ #include <dynarmic/A32/coprocessor.h> #include "common/common_types.h" -enum class CP15Register { - // c0 - Information registers - CP15_MAIN_ID, - CP15_CACHE_TYPE, - CP15_TCM_STATUS, - CP15_TLB_TYPE, - CP15_CPU_ID, - CP15_PROCESSOR_FEATURE_0, - CP15_PROCESSOR_FEATURE_1, - CP15_DEBUG_FEATURE_0, - CP15_AUXILIARY_FEATURE_0, - CP15_MEMORY_MODEL_FEATURE_0, - CP15_MEMORY_MODEL_FEATURE_1, - CP15_MEMORY_MODEL_FEATURE_2, - CP15_MEMORY_MODEL_FEATURE_3, - CP15_ISA_FEATURE_0, - CP15_ISA_FEATURE_1, - CP15_ISA_FEATURE_2, - CP15_ISA_FEATURE_3, - CP15_ISA_FEATURE_4, +namespace Core { - // c1 - Control registers - CP15_CONTROL, - CP15_AUXILIARY_CONTROL, - CP15_COPROCESSOR_ACCESS_CONTROL, - - // c2 - Translation table registers - CP15_TRANSLATION_BASE_TABLE_0, - CP15_TRANSLATION_BASE_TABLE_1, - CP15_TRANSLATION_BASE_CONTROL, - CP15_DOMAIN_ACCESS_CONTROL, - CP15_RESERVED, - - // c5 - Fault status registers - CP15_FAULT_STATUS, - CP15_INSTR_FAULT_STATUS, - CP15_COMBINED_DATA_FSR = CP15_FAULT_STATUS, - CP15_INST_FSR, - - // c6 - Fault Address registers - CP15_FAULT_ADDRESS, - CP15_COMBINED_DATA_FAR = CP15_FAULT_ADDRESS, - CP15_WFAR, - CP15_IFAR, - - // c7 - Cache operation registers - CP15_WAIT_FOR_INTERRUPT, - CP15_PHYS_ADDRESS, - CP15_INVALIDATE_INSTR_CACHE, - CP15_INVALIDATE_INSTR_CACHE_USING_MVA, - CP15_INVALIDATE_INSTR_CACHE_USING_INDEX, - CP15_FLUSH_PREFETCH_BUFFER, - CP15_FLUSH_BRANCH_TARGET_CACHE, - CP15_FLUSH_BRANCH_TARGET_CACHE_ENTRY, - CP15_INVALIDATE_DATA_CACHE, - CP15_INVALIDATE_DATA_CACHE_LINE_USING_MVA, - CP15_INVALIDATE_DATA_CACHE_LINE_USING_INDEX, - CP15_INVALIDATE_DATA_AND_INSTR_CACHE, - CP15_CLEAN_DATA_CACHE, - CP15_CLEAN_DATA_CACHE_LINE_USING_MVA, - CP15_CLEAN_DATA_CACHE_LINE_USING_INDEX, - CP15_DATA_SYNC_BARRIER, - CP15_DATA_MEMORY_BARRIER, - CP15_CLEAN_AND_INVALIDATE_DATA_CACHE, - CP15_CLEAN_AND_INVALIDATE_DATA_CACHE_LINE_USING_MVA, - CP15_CLEAN_AND_INVALIDATE_DATA_CACHE_LINE_USING_INDEX, - - // c8 - TLB operations - CP15_INVALIDATE_ITLB, - CP15_INVALIDATE_ITLB_SINGLE_ENTRY, - CP15_INVALIDATE_ITLB_ENTRY_ON_ASID_MATCH, - CP15_INVALIDATE_ITLB_ENTRY_ON_MVA, - CP15_INVALIDATE_DTLB, - CP15_INVALIDATE_DTLB_SINGLE_ENTRY, - CP15_INVALIDATE_DTLB_ENTRY_ON_ASID_MATCH, - CP15_INVALIDATE_DTLB_ENTRY_ON_MVA, - CP15_INVALIDATE_UTLB, - CP15_INVALIDATE_UTLB_SINGLE_ENTRY, - CP15_INVALIDATE_UTLB_ENTRY_ON_ASID_MATCH, - CP15_INVALIDATE_UTLB_ENTRY_ON_MVA, - - // c9 - Data cache lockdown register - CP15_DATA_CACHE_LOCKDOWN, - - // c10 - TLB/Memory map registers - CP15_TLB_LOCKDOWN, - CP15_PRIMARY_REGION_REMAP, - CP15_NORMAL_REGION_REMAP, - - // c13 - Thread related registers - CP15_PID, - CP15_CONTEXT_ID, - CP15_THREAD_UPRW, // Thread ID register - User/Privileged Read/Write - CP15_THREAD_URO, // Thread ID register - User Read Only (Privileged R/W) - CP15_THREAD_PRW, // Thread ID register - Privileged R/W only. - - // c15 - Performance and TLB lockdown registers - CP15_PERFORMANCE_MONITOR_CONTROL, - CP15_CYCLE_COUNTER, - CP15_COUNT_0, - CP15_COUNT_1, - CP15_READ_MAIN_TLB_LOCKDOWN_ENTRY, - CP15_WRITE_MAIN_TLB_LOCKDOWN_ENTRY, - CP15_MAIN_TLB_LOCKDOWN_VIRT_ADDRESS, - CP15_MAIN_TLB_LOCKDOWN_PHYS_ADDRESS, - CP15_MAIN_TLB_LOCKDOWN_ATTRIBUTE, - CP15_TLB_DEBUG_CONTROL, - - // Skyeye defined - CP15_TLB_FAULT_ADDR, - CP15_TLB_FAULT_STATUS, - - // Not an actual register. - // All registers should be defined above this. - CP15_REGISTER_COUNT, -}; +class ARM_Dynarmic_32; class DynarmicCP15 final : public Dynarmic::A32::Coprocessor { public: using CoprocReg = Dynarmic::A32::CoprocReg; - explicit DynarmicCP15(u32* cp15) : CP15(cp15){}; + explicit DynarmicCP15(ARM_Dynarmic_32& parent) : parent(parent) {} std::optional<Callback> CompileInternalOperation(bool two, unsigned opc1, CoprocReg CRd, CoprocReg CRn, CoprocReg CRm, @@ -147,6 +34,9 @@ public: std::optional<Callback> CompileStoreWords(bool two, bool long_transfer, CoprocReg CRd, std::optional<u8> option) override; -private: - u32* CP15{}; + ARM_Dynarmic_32& parent; + u32 uprw = 0; + u32 uro = 0; }; + +} // namespace Core diff --git a/src/core/arm/dynarmic/arm_exclusive_monitor.cpp b/src/core/arm/dynarmic/arm_exclusive_monitor.cpp new file mode 100644 index 000000000..4e209f6a5 --- /dev/null +++ b/src/core/arm/dynarmic/arm_exclusive_monitor.cpp @@ -0,0 +1,76 @@ +// Copyright 2018 yuzu emulator team +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#include <cinttypes> +#include <memory> +#include "core/arm/dynarmic/arm_exclusive_monitor.h" +#include "core/memory.h" + +namespace Core { + +DynarmicExclusiveMonitor::DynarmicExclusiveMonitor(Memory::Memory& memory, std::size_t core_count) + : monitor(core_count), memory{memory} {} + +DynarmicExclusiveMonitor::~DynarmicExclusiveMonitor() = default; + +u8 DynarmicExclusiveMonitor::ExclusiveRead8(std::size_t core_index, VAddr addr) { + return monitor.ReadAndMark<u8>(core_index, addr, [&]() -> u8 { return memory.Read8(addr); }); +} + +u16 DynarmicExclusiveMonitor::ExclusiveRead16(std::size_t core_index, VAddr addr) { + return monitor.ReadAndMark<u16>(core_index, addr, [&]() -> u16 { return memory.Read16(addr); }); +} + +u32 DynarmicExclusiveMonitor::ExclusiveRead32(std::size_t core_index, VAddr addr) { + return monitor.ReadAndMark<u32>(core_index, addr, [&]() -> u32 { return memory.Read32(addr); }); +} + +u64 DynarmicExclusiveMonitor::ExclusiveRead64(std::size_t core_index, VAddr addr) { + return monitor.ReadAndMark<u64>(core_index, addr, [&]() -> u64 { return memory.Read64(addr); }); +} + +u128 DynarmicExclusiveMonitor::ExclusiveRead128(std::size_t core_index, VAddr addr) { + return monitor.ReadAndMark<u128>(core_index, addr, [&]() -> u128 { + u128 result; + result[0] = memory.Read64(addr); + result[1] = memory.Read64(addr + 8); + return result; + }); +} + +void DynarmicExclusiveMonitor::ClearExclusive() { + monitor.Clear(); +} + +bool DynarmicExclusiveMonitor::ExclusiveWrite8(std::size_t core_index, VAddr vaddr, u8 value) { + return monitor.DoExclusiveOperation<u8>(core_index, vaddr, [&](u8 expected) -> bool { + return memory.WriteExclusive8(vaddr, value, expected); + }); +} + +bool DynarmicExclusiveMonitor::ExclusiveWrite16(std::size_t core_index, VAddr vaddr, u16 value) { + return monitor.DoExclusiveOperation<u16>(core_index, vaddr, [&](u16 expected) -> bool { + return memory.WriteExclusive16(vaddr, value, expected); + }); +} + +bool DynarmicExclusiveMonitor::ExclusiveWrite32(std::size_t core_index, VAddr vaddr, u32 value) { + return monitor.DoExclusiveOperation<u32>(core_index, vaddr, [&](u32 expected) -> bool { + return memory.WriteExclusive32(vaddr, value, expected); + }); +} + +bool DynarmicExclusiveMonitor::ExclusiveWrite64(std::size_t core_index, VAddr vaddr, u64 value) { + return monitor.DoExclusiveOperation<u64>(core_index, vaddr, [&](u64 expected) -> bool { + return memory.WriteExclusive64(vaddr, value, expected); + }); +} + +bool DynarmicExclusiveMonitor::ExclusiveWrite128(std::size_t core_index, VAddr vaddr, u128 value) { + return monitor.DoExclusiveOperation<u128>(core_index, vaddr, [&](u128 expected) -> bool { + return memory.WriteExclusive128(vaddr, value, expected); + }); +} + +} // namespace Core diff --git a/src/core/arm/dynarmic/arm_exclusive_monitor.h b/src/core/arm/dynarmic/arm_exclusive_monitor.h new file mode 100644 index 000000000..964f4a55d --- /dev/null +++ b/src/core/arm/dynarmic/arm_exclusive_monitor.h @@ -0,0 +1,48 @@ +// Copyright 2020 yuzu emulator team +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#pragma once + +#include <memory> +#include <unordered_map> + +#include <dynarmic/exclusive_monitor.h> + +#include "common/common_types.h" +#include "core/arm/dynarmic/arm_dynarmic_32.h" +#include "core/arm/dynarmic/arm_dynarmic_64.h" +#include "core/arm/exclusive_monitor.h" + +namespace Core::Memory { +class Memory; +} + +namespace Core { + +class DynarmicExclusiveMonitor final : public ExclusiveMonitor { +public: + explicit DynarmicExclusiveMonitor(Memory::Memory& memory, std::size_t core_count); + ~DynarmicExclusiveMonitor() override; + + u8 ExclusiveRead8(std::size_t core_index, VAddr addr) override; + u16 ExclusiveRead16(std::size_t core_index, VAddr addr) override; + u32 ExclusiveRead32(std::size_t core_index, VAddr addr) override; + u64 ExclusiveRead64(std::size_t core_index, VAddr addr) override; + u128 ExclusiveRead128(std::size_t core_index, VAddr addr) override; + void ClearExclusive() override; + + bool ExclusiveWrite8(std::size_t core_index, VAddr vaddr, u8 value) override; + bool ExclusiveWrite16(std::size_t core_index, VAddr vaddr, u16 value) override; + bool ExclusiveWrite32(std::size_t core_index, VAddr vaddr, u32 value) override; + bool ExclusiveWrite64(std::size_t core_index, VAddr vaddr, u64 value) override; + bool ExclusiveWrite128(std::size_t core_index, VAddr vaddr, u128 value) override; + +private: + friend class ARM_Dynarmic_32; + friend class ARM_Dynarmic_64; + Dynarmic::ExclusiveMonitor monitor; + Core::Memory::Memory& memory; +}; + +} // namespace Core diff --git a/src/core/arm/exclusive_monitor.cpp b/src/core/arm/exclusive_monitor.cpp index b32401e0b..d8cba369d 100644 --- a/src/core/arm/exclusive_monitor.cpp +++ b/src/core/arm/exclusive_monitor.cpp @@ -3,7 +3,7 @@ // Refer to the license.txt file included. #ifdef ARCHITECTURE_x86_64 -#include "core/arm/dynarmic/arm_dynarmic_64.h" +#include "core/arm/dynarmic/arm_exclusive_monitor.h" #endif #include "core/arm/exclusive_monitor.h" #include "core/memory.h" diff --git a/src/core/arm/exclusive_monitor.h b/src/core/arm/exclusive_monitor.h index 4ef418b90..62f6e6023 100644 --- a/src/core/arm/exclusive_monitor.h +++ b/src/core/arm/exclusive_monitor.h @@ -8,7 +8,7 @@ #include "common/common_types.h" -namespace Memory { +namespace Core::Memory { class Memory; } @@ -18,7 +18,11 @@ class ExclusiveMonitor { public: virtual ~ExclusiveMonitor(); - virtual void SetExclusive(std::size_t core_index, VAddr addr) = 0; + virtual u8 ExclusiveRead8(std::size_t core_index, VAddr addr) = 0; + virtual u16 ExclusiveRead16(std::size_t core_index, VAddr addr) = 0; + virtual u32 ExclusiveRead32(std::size_t core_index, VAddr addr) = 0; + virtual u64 ExclusiveRead64(std::size_t core_index, VAddr addr) = 0; + virtual u128 ExclusiveRead128(std::size_t core_index, VAddr addr) = 0; virtual void ClearExclusive() = 0; virtual bool ExclusiveWrite8(std::size_t core_index, VAddr vaddr, u8 value) = 0; diff --git a/src/core/arm/unicorn/arm_unicorn.cpp b/src/core/arm/unicorn/arm_unicorn.cpp deleted file mode 100644 index 8a9800a96..000000000 --- a/src/core/arm/unicorn/arm_unicorn.cpp +++ /dev/null @@ -1,274 +0,0 @@ -// Copyright 2018 yuzu emulator team -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#include <algorithm> -#include <unicorn/arm64.h> -#include "common/assert.h" -#include "common/microprofile.h" -#include "core/arm/unicorn/arm_unicorn.h" -#include "core/core.h" -#include "core/core_timing.h" -#include "core/hle/kernel/scheduler.h" -#include "core/hle/kernel/svc.h" - -namespace Core { - -// Load Unicorn DLL once on Windows using RAII -#ifdef _MSC_VER -#include <unicorn_dynload.h> -struct LoadDll { -private: - LoadDll() { - ASSERT(uc_dyn_load(NULL, 0)); - } - ~LoadDll() { - ASSERT(uc_dyn_free()); - } - static LoadDll g_load_dll; -}; -LoadDll LoadDll::g_load_dll; -#endif - -#define CHECKED(expr) \ - do { \ - if (auto _cerr = (expr)) { \ - ASSERT_MSG(false, "Call " #expr " failed with error: {} ({})\n", _cerr, \ - uc_strerror(_cerr)); \ - } \ - } while (0) - -static void CodeHook(uc_engine* uc, uint64_t address, uint32_t size, void* user_data) { - GDBStub::BreakpointAddress bkpt = - GDBStub::GetNextBreakpointFromAddress(address, GDBStub::BreakpointType::Execute); - if (GDBStub::IsMemoryBreak() || - (bkpt.type != GDBStub::BreakpointType::None && address == bkpt.address)) { - auto core = static_cast<ARM_Unicorn*>(user_data); - core->RecordBreak(bkpt); - uc_emu_stop(uc); - } -} - -static bool UnmappedMemoryHook(uc_engine* uc, uc_mem_type type, u64 addr, int size, u64 value, - void* user_data) { - auto* const system = static_cast<System*>(user_data); - - ARM_Interface::ThreadContext64 ctx{}; - system->CurrentArmInterface().SaveContext(ctx); - ASSERT_MSG(false, "Attempted to read from unmapped memory: 0x{:X}, pc=0x{:X}, lr=0x{:X}", addr, - ctx.pc, ctx.cpu_registers[30]); - - return false; -} - -ARM_Unicorn::ARM_Unicorn(System& system) : ARM_Interface{system} { - CHECKED(uc_open(UC_ARCH_ARM64, UC_MODE_ARM, &uc)); - - auto fpv = 3 << 20; - CHECKED(uc_reg_write(uc, UC_ARM64_REG_CPACR_EL1, &fpv)); - - uc_hook hook{}; - CHECKED(uc_hook_add(uc, &hook, UC_HOOK_INTR, (void*)InterruptHook, this, 0, UINT64_MAX)); - CHECKED(uc_hook_add(uc, &hook, UC_HOOK_MEM_INVALID, (void*)UnmappedMemoryHook, &system, 0, - UINT64_MAX)); - if (GDBStub::IsServerEnabled()) { - CHECKED(uc_hook_add(uc, &hook, UC_HOOK_CODE, (void*)CodeHook, this, 0, UINT64_MAX)); - last_bkpt_hit = false; - } -} - -ARM_Unicorn::~ARM_Unicorn() { - CHECKED(uc_close(uc)); -} - -void ARM_Unicorn::SetPC(u64 pc) { - CHECKED(uc_reg_write(uc, UC_ARM64_REG_PC, &pc)); -} - -u64 ARM_Unicorn::GetPC() const { - u64 val{}; - CHECKED(uc_reg_read(uc, UC_ARM64_REG_PC, &val)); - return val; -} - -u64 ARM_Unicorn::GetReg(int regn) const { - u64 val{}; - auto treg = UC_ARM64_REG_SP; - if (regn <= 28) { - treg = (uc_arm64_reg)(UC_ARM64_REG_X0 + regn); - } else if (regn < 31) { - treg = (uc_arm64_reg)(UC_ARM64_REG_X29 + regn - 29); - } - CHECKED(uc_reg_read(uc, treg, &val)); - return val; -} - -void ARM_Unicorn::SetReg(int regn, u64 val) { - auto treg = UC_ARM64_REG_SP; - if (regn <= 28) { - treg = (uc_arm64_reg)(UC_ARM64_REG_X0 + regn); - } else if (regn < 31) { - treg = (uc_arm64_reg)(UC_ARM64_REG_X29 + regn - 29); - } - CHECKED(uc_reg_write(uc, treg, &val)); -} - -u128 ARM_Unicorn::GetVectorReg(int /*index*/) const { - UNIMPLEMENTED(); - static constexpr u128 res{}; - return res; -} - -void ARM_Unicorn::SetVectorReg(int /*index*/, u128 /*value*/) { - UNIMPLEMENTED(); -} - -u32 ARM_Unicorn::GetPSTATE() const { - u64 nzcv{}; - CHECKED(uc_reg_read(uc, UC_ARM64_REG_NZCV, &nzcv)); - return static_cast<u32>(nzcv); -} - -void ARM_Unicorn::SetPSTATE(u32 pstate) { - u64 nzcv = pstate; - CHECKED(uc_reg_write(uc, UC_ARM64_REG_NZCV, &nzcv)); -} - -VAddr ARM_Unicorn::GetTlsAddress() const { - u64 base{}; - CHECKED(uc_reg_read(uc, UC_ARM64_REG_TPIDRRO_EL0, &base)); - return base; -} - -void ARM_Unicorn::SetTlsAddress(VAddr base) { - CHECKED(uc_reg_write(uc, UC_ARM64_REG_TPIDRRO_EL0, &base)); -} - -u64 ARM_Unicorn::GetTPIDR_EL0() const { - u64 value{}; - CHECKED(uc_reg_read(uc, UC_ARM64_REG_TPIDR_EL0, &value)); - return value; -} - -void ARM_Unicorn::SetTPIDR_EL0(u64 value) { - CHECKED(uc_reg_write(uc, UC_ARM64_REG_TPIDR_EL0, &value)); -} - -void ARM_Unicorn::Run() { - if (GDBStub::IsServerEnabled()) { - ExecuteInstructions(std::max(4000000U, 0U)); - } else { - ExecuteInstructions( - std::max(std::size_t(system.CoreTiming().GetDowncount()), std::size_t{0})); - } -} - -void ARM_Unicorn::Step() { - ExecuteInstructions(1); -} - -MICROPROFILE_DEFINE(ARM_Jit_Unicorn, "ARM JIT", "Unicorn", MP_RGB(255, 64, 64)); - -void ARM_Unicorn::ExecuteInstructions(std::size_t num_instructions) { - MICROPROFILE_SCOPE(ARM_Jit_Unicorn); - CHECKED(uc_emu_start(uc, GetPC(), 1ULL << 63, 0, num_instructions)); - system.CoreTiming().AddTicks(num_instructions); - if (GDBStub::IsServerEnabled()) { - if (last_bkpt_hit && last_bkpt.type == GDBStub::BreakpointType::Execute) { - uc_reg_write(uc, UC_ARM64_REG_PC, &last_bkpt.address); - } - - Kernel::Thread* const thread = system.CurrentScheduler().GetCurrentThread(); - SaveContext(thread->GetContext64()); - if (last_bkpt_hit || GDBStub::IsMemoryBreak() || GDBStub::GetCpuStepFlag()) { - last_bkpt_hit = false; - GDBStub::Break(); - GDBStub::SendTrap(thread, 5); - } - } -} - -void ARM_Unicorn::SaveContext(ThreadContext64& ctx) { - int uregs[32]; - void* tregs[32]; - - CHECKED(uc_reg_read(uc, UC_ARM64_REG_SP, &ctx.sp)); - CHECKED(uc_reg_read(uc, UC_ARM64_REG_PC, &ctx.pc)); - CHECKED(uc_reg_read(uc, UC_ARM64_REG_NZCV, &ctx.pstate)); - - for (auto i = 0; i < 29; ++i) { - uregs[i] = UC_ARM64_REG_X0 + i; - tregs[i] = &ctx.cpu_registers[i]; - } - uregs[29] = UC_ARM64_REG_X29; - tregs[29] = (void*)&ctx.cpu_registers[29]; - uregs[30] = UC_ARM64_REG_X30; - tregs[30] = (void*)&ctx.cpu_registers[30]; - - CHECKED(uc_reg_read_batch(uc, uregs, tregs, 31)); - - for (int i = 0; i < 32; ++i) { - uregs[i] = UC_ARM64_REG_Q0 + i; - tregs[i] = &ctx.vector_registers[i]; - } - - CHECKED(uc_reg_read_batch(uc, uregs, tregs, 32)); -} - -void ARM_Unicorn::LoadContext(const ThreadContext64& ctx) { - int uregs[32]; - void* tregs[32]; - - CHECKED(uc_reg_write(uc, UC_ARM64_REG_SP, &ctx.sp)); - CHECKED(uc_reg_write(uc, UC_ARM64_REG_PC, &ctx.pc)); - CHECKED(uc_reg_write(uc, UC_ARM64_REG_NZCV, &ctx.pstate)); - - for (int i = 0; i < 29; ++i) { - uregs[i] = UC_ARM64_REG_X0 + i; - tregs[i] = (void*)&ctx.cpu_registers[i]; - } - uregs[29] = UC_ARM64_REG_X29; - tregs[29] = (void*)&ctx.cpu_registers[29]; - uregs[30] = UC_ARM64_REG_X30; - tregs[30] = (void*)&ctx.cpu_registers[30]; - - CHECKED(uc_reg_write_batch(uc, uregs, tregs, 31)); - - for (auto i = 0; i < 32; ++i) { - uregs[i] = UC_ARM64_REG_Q0 + i; - tregs[i] = (void*)&ctx.vector_registers[i]; - } - - CHECKED(uc_reg_write_batch(uc, uregs, tregs, 32)); -} - -void ARM_Unicorn::PrepareReschedule() { - CHECKED(uc_emu_stop(uc)); -} - -void ARM_Unicorn::ClearExclusiveState() {} - -void ARM_Unicorn::ClearInstructionCache() {} - -void ARM_Unicorn::RecordBreak(GDBStub::BreakpointAddress bkpt) { - last_bkpt = bkpt; - last_bkpt_hit = true; -} - -void ARM_Unicorn::InterruptHook(uc_engine* uc, u32 int_no, void* user_data) { - u32 esr{}; - CHECKED(uc_reg_read(uc, UC_ARM64_REG_ESR, &esr)); - - const auto ec = esr >> 26; - const auto iss = esr & 0xFFFFFF; - - auto* const arm_instance = static_cast<ARM_Unicorn*>(user_data); - - switch (ec) { - case 0x15: // SVC - Kernel::CallSVC(arm_instance->system, iss); - break; - } -} - -} // namespace Core diff --git a/src/core/arm/unicorn/arm_unicorn.h b/src/core/arm/unicorn/arm_unicorn.h deleted file mode 100644 index f30d13cb6..000000000 --- a/src/core/arm/unicorn/arm_unicorn.h +++ /dev/null @@ -1,55 +0,0 @@ -// Copyright 2018 yuzu emulator team -// Licensed under GPLv2 or any later version -// Refer to the license.txt file included. - -#pragma once - -#include <unicorn/unicorn.h> -#include "common/common_types.h" -#include "core/arm/arm_interface.h" -#include "core/gdbstub/gdbstub.h" - -namespace Core { - -class System; - -class ARM_Unicorn final : public ARM_Interface { -public: - explicit ARM_Unicorn(System& system); - ~ARM_Unicorn() override; - - void SetPC(u64 pc) override; - u64 GetPC() const override; - u64 GetReg(int index) const override; - void SetReg(int index, u64 value) override; - u128 GetVectorReg(int index) const override; - void SetVectorReg(int index, u128 value) override; - u32 GetPSTATE() const override; - void SetPSTATE(u32 pstate) override; - VAddr GetTlsAddress() const override; - void SetTlsAddress(VAddr address) override; - void SetTPIDR_EL0(u64 value) override; - u64 GetTPIDR_EL0() const override; - void PrepareReschedule() override; - void ClearExclusiveState() override; - void ExecuteInstructions(std::size_t num_instructions); - void Run() override; - void Step() override; - void ClearInstructionCache() override; - void PageTableChanged(Common::PageTable&, std::size_t) override {} - void RecordBreak(GDBStub::BreakpointAddress bkpt); - - void SaveContext(ThreadContext32& ctx) override {} - void SaveContext(ThreadContext64& ctx) override; - void LoadContext(const ThreadContext32& ctx) override {} - void LoadContext(const ThreadContext64& ctx) override; - -private: - static void InterruptHook(uc_engine* uc, u32 int_no, void* user_data); - - uc_engine* uc{}; - GDBStub::BreakpointAddress last_bkpt{}; - bool last_bkpt_hit = false; -}; - -} // namespace Core |
