From 7ee76003adbc24bdc651710bcab9860efe3b8c22 Mon Sep 17 00:00:00 2001 From: Fernando Sahmkow Date: Mon, 24 Feb 2020 22:04:12 -0400 Subject: General: Recover Prometheus project from harddrive failure This commit: Implements CPU Interrupts, Replaces Cycle Timing for Host Timing, Reworks the Kernel's Scheduler, Introduce Idle State and Suspended State, Recreates the bootmanager, Initializes Multicore system. --- src/core/memory.cpp | 11 ++++------- 1 file changed, 4 insertions(+), 7 deletions(-) (limited to 'src/core/memory.cpp') diff --git a/src/core/memory.cpp b/src/core/memory.cpp index 9d87045a05..66634596d1 100644 --- a/src/core/memory.cpp +++ b/src/core/memory.cpp @@ -29,15 +29,12 @@ namespace Core::Memory { struct Memory::Impl { explicit Impl(Core::System& system_) : system{system_} {} - void SetCurrentPageTable(Kernel::Process& process) { + void SetCurrentPageTable(Kernel::Process& process, u32 core_id) { current_page_table = &process.PageTable().PageTableImpl(); const std::size_t address_space_width = process.PageTable().GetAddressSpaceWidth(); - system.ArmInterface(0).PageTableChanged(*current_page_table, address_space_width); - system.ArmInterface(1).PageTableChanged(*current_page_table, address_space_width); - system.ArmInterface(2).PageTableChanged(*current_page_table, address_space_width); - system.ArmInterface(3).PageTableChanged(*current_page_table, address_space_width); + system.ArmInterface(core_id).PageTableChanged(*current_page_table, address_space_width); } void MapMemoryRegion(Common::PageTable& page_table, VAddr base, u64 size, PAddr target) { @@ -689,8 +686,8 @@ struct Memory::Impl { Memory::Memory(Core::System& system) : impl{std::make_unique(system)} {} Memory::~Memory() = default; -void Memory::SetCurrentPageTable(Kernel::Process& process) { - impl->SetCurrentPageTable(process); +void Memory::SetCurrentPageTable(Kernel::Process& process, u32 core_id) { + impl->SetCurrentPageTable(process, core_id); } void Memory::MapMemoryRegion(Common::PageTable& page_table, VAddr base, u64 size, PAddr target) { -- cgit v1.2.3 From 5eac520ae3fe076ddcf55138ad45c636688c3322 Mon Sep 17 00:00:00 2001 From: Fernando Sahmkow Date: Sat, 7 Mar 2020 18:59:42 -0400 Subject: ARM/Memory: Correct Exclusive Monitor and Implement Exclusive Memory Writes. --- src/common/CMakeLists.txt | 2 + src/common/atomic_ops.cpp | 70 ++++++++++++++++++++++ src/common/atomic_ops.h | 17 ++++++ src/core/arm/dynarmic/arm_dynarmic_64.cpp | 66 +++++++++++++++++---- src/core/arm/dynarmic/arm_dynarmic_64.h | 6 +- src/core/arm/exclusive_monitor.h | 6 +- src/core/hle/kernel/address_arbiter.cpp | 6 +- src/core/hle/kernel/mutex.cpp | 5 +- src/core/hle/kernel/svc.cpp | 2 +- src/core/hle/kernel/thread.cpp | 6 +- src/core/memory.cpp | 98 +++++++++++++++++++++++++++++++ src/core/memory.h | 65 ++++++++++++++++++++ 12 files changed, 325 insertions(+), 24 deletions(-) create mode 100644 src/common/atomic_ops.cpp create mode 100644 src/common/atomic_ops.h (limited to 'src/core/memory.cpp') diff --git a/src/common/CMakeLists.txt b/src/common/CMakeLists.txt index 3cc17d0e90..d120c8d3d9 100644 --- a/src/common/CMakeLists.txt +++ b/src/common/CMakeLists.txt @@ -98,6 +98,8 @@ add_library(common STATIC algorithm.h alignment.h assert.h + atomic_ops.cpp + atomic_ops.h detached_tasks.cpp detached_tasks.h bit_field.h diff --git a/src/common/atomic_ops.cpp b/src/common/atomic_ops.cpp new file mode 100644 index 0000000000..65cdfb4fd5 --- /dev/null +++ b/src/common/atomic_ops.cpp @@ -0,0 +1,70 @@ +// Copyright 2020 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#include + +#include "common/atomic_ops.h" + +#if _MSC_VER +#include +#endif + +namespace Common { + +#if _MSC_VER + +bool AtomicCompareAndSwap(u8 volatile* pointer, u8 value, u8 expected) { + u8 result = _InterlockedCompareExchange8((char*)pointer, value, expected); + return result == expected; +} + +bool AtomicCompareAndSwap(u16 volatile* pointer, u16 value, u16 expected) { + u16 result = _InterlockedCompareExchange16((short*)pointer, value, expected); + return result == expected; +} + +bool AtomicCompareAndSwap(u32 volatile* pointer, u32 value, u32 expected) { + u32 result = _InterlockedCompareExchange((long*)pointer, value, expected); + return result == expected; +} + +bool AtomicCompareAndSwap(u64 volatile* pointer, u64 value, u64 expected) { + u64 result = _InterlockedCompareExchange64((__int64*)pointer, value, expected); + return result == expected; +} + +bool AtomicCompareAndSwap(u64 volatile* pointer, u128 value, u128 expected) { + return _InterlockedCompareExchange128((__int64*)pointer, value[1], value[0], (__int64*)expected.data()) != 0; +} + + +#else + +bool AtomicCompareAndSwap(u8 volatile* pointer, u8 value, u8 expected) { + return __sync_bool_compare_and_swap (pointer, value, expected); +} + +bool AtomicCompareAndSwap(u16 volatile* pointer, u16 value, u16 expected) { + return __sync_bool_compare_and_swap (pointer, value, expected); +} + +bool AtomicCompareAndSwap(u32 volatile* pointer, u32 value, u32 expected) { + return __sync_bool_compare_and_swap (pointer, value, expected); +} + +bool AtomicCompareAndSwap(u64 volatile* pointer, u64 value, u64 expected) { + return __sync_bool_compare_and_swap (pointer, value, expected); +} + +bool AtomicCompareAndSwap(u64 volatile* pointer, u128 value, u128 expected) { + unsigned __int128 value_a; + unsigned __int128 expected_a; + std::memcpy(&value_a, value.data(), sizeof(u128)); + std::memcpy(&expected_a, expected.data(), sizeof(u128)); + return __sync_bool_compare_and_swap ((unsigned __int128*)pointer, value_a, expected_a); +} + +#endif + +} // namespace Common diff --git a/src/common/atomic_ops.h b/src/common/atomic_ops.h new file mode 100644 index 0000000000..22cb3a402b --- /dev/null +++ b/src/common/atomic_ops.h @@ -0,0 +1,17 @@ +// Copyright 2020 yuzu Emulator Project +// Licensed under GPLv2 or any later version +// Refer to the license.txt file included. + +#pragma once + +#include "common/common_types.h" + +namespace Common { + +bool AtomicCompareAndSwap(u8 volatile * pointer, u8 value, u8 expected); +bool AtomicCompareAndSwap(u16 volatile* pointer, u16 value, u16 expected); +bool AtomicCompareAndSwap(u32 volatile* pointer, u32 value, u32 expected); +bool AtomicCompareAndSwap(u64 volatile* pointer, u64 value, u64 expected); +bool AtomicCompareAndSwap(u64 volatile* pointer, u128 value, u128 expected); + +} // namespace Common diff --git a/src/core/arm/dynarmic/arm_dynarmic_64.cpp b/src/core/arm/dynarmic/arm_dynarmic_64.cpp index 5e316ffd4d..a22c22bf00 100644 --- a/src/core/arm/dynarmic/arm_dynarmic_64.cpp +++ b/src/core/arm/dynarmic/arm_dynarmic_64.cpp @@ -66,6 +66,22 @@ 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)); @@ -284,9 +300,29 @@ DynarmicExclusiveMonitor::DynarmicExclusiveMonitor(Memory::Memory& memory, std:: 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::SetExclusive8(std::size_t core_index, VAddr addr) { + monitor.Mark(core_index, addr, 1, [&]() -> u8 { return memory.Read8(addr); }); +} + +void DynarmicExclusiveMonitor::SetExclusive16(std::size_t core_index, VAddr addr) { + monitor.Mark(core_index, addr, 2, [&]() -> u16 { return memory.Read16(addr); }); +} + +void DynarmicExclusiveMonitor::SetExclusive32(std::size_t core_index, VAddr addr) { + monitor.Mark(core_index, addr, 4, [&]() -> u32 { return memory.Read32(addr); }); +} + +void DynarmicExclusiveMonitor::SetExclusive64(std::size_t core_index, VAddr addr) { + monitor.Mark(core_index, addr, 8, [&]() -> u64 { return memory.Read64(addr); }); +} + +void DynarmicExclusiveMonitor::SetExclusive128(std::size_t core_index, VAddr addr) { + monitor.Mark(core_index, addr, 16, [&]() -> u128 { + u128 result; + result[0] = memory.Read64(addr); + result[1] = memory.Read64(addr + 8); + return result; + }); } void DynarmicExclusiveMonitor::ClearExclusive() { @@ -294,28 +330,32 @@ void DynarmicExclusiveMonitor::ClearExclusive() { } bool DynarmicExclusiveMonitor::ExclusiveWrite8(std::size_t core_index, VAddr vaddr, u8 value) { - return monitor.DoExclusiveOperation(core_index, vaddr, 1, [&] { memory.Write8(vaddr, value); }); + return monitor.DoExclusiveOperation(core_index, vaddr, 1, [&](u8 expected) -> bool { + return memory.WriteExclusive8(vaddr, value, expected); + }); } bool DynarmicExclusiveMonitor::ExclusiveWrite16(std::size_t core_index, VAddr vaddr, u16 value) { - return monitor.DoExclusiveOperation(core_index, vaddr, 2, - [&] { memory.Write16(vaddr, value); }); + return monitor.DoExclusiveOperation(core_index, vaddr, 2, [&](u16 expected) -> bool { + return memory.WriteExclusive16(vaddr, value, expected); + }); } bool DynarmicExclusiveMonitor::ExclusiveWrite32(std::size_t core_index, VAddr vaddr, u32 value) { - return monitor.DoExclusiveOperation(core_index, vaddr, 4, - [&] { memory.Write32(vaddr, value); }); + return monitor.DoExclusiveOperation(core_index, vaddr, 4, [&](u32 expected) -> bool { + return memory.WriteExclusive32(vaddr, value, expected); + }); } bool DynarmicExclusiveMonitor::ExclusiveWrite64(std::size_t core_index, VAddr vaddr, u64 value) { - return monitor.DoExclusiveOperation(core_index, vaddr, 8, - [&] { memory.Write64(vaddr, value); }); + return monitor.DoExclusiveOperation(core_index, vaddr, 8, [&](u64 expected) -> bool { + return memory.WriteExclusive64(vaddr, value, expected); + }); } 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]); + return monitor.DoExclusiveOperation(core_index, vaddr, 16, [&](u128 expected) -> bool { + return memory.WriteExclusive128(vaddr, value, expected); }); } diff --git a/src/core/arm/dynarmic/arm_dynarmic_64.h b/src/core/arm/dynarmic/arm_dynarmic_64.h index 9e94b58c2d..3ead59f166 100644 --- a/src/core/arm/dynarmic/arm_dynarmic_64.h +++ b/src/core/arm/dynarmic/arm_dynarmic_64.h @@ -82,7 +82,11 @@ public: explicit DynarmicExclusiveMonitor(Memory::Memory& memory, std::size_t core_count); ~DynarmicExclusiveMonitor() override; - void SetExclusive(std::size_t core_index, VAddr addr) override; + void SetExclusive8(std::size_t core_index, VAddr addr) override; + void SetExclusive16(std::size_t core_index, VAddr addr) override; + void SetExclusive32(std::size_t core_index, VAddr addr) override; + void SetExclusive64(std::size_t core_index, VAddr addr) override; + void SetExclusive128(std::size_t core_index, VAddr addr) override; void ClearExclusive() override; bool ExclusiveWrite8(std::size_t core_index, VAddr vaddr, u8 value) override; diff --git a/src/core/arm/exclusive_monitor.h b/src/core/arm/exclusive_monitor.h index ccd73b80ff..2ee312eeea 100644 --- a/src/core/arm/exclusive_monitor.h +++ b/src/core/arm/exclusive_monitor.h @@ -18,7 +18,11 @@ class ExclusiveMonitor { public: virtual ~ExclusiveMonitor(); - virtual void SetExclusive(std::size_t core_index, VAddr addr) = 0; + virtual void SetExclusive8(std::size_t core_index, VAddr addr) = 0; + virtual void SetExclusive16(std::size_t core_index, VAddr addr) = 0; + virtual void SetExclusive32(std::size_t core_index, VAddr addr) = 0; + virtual void SetExclusive64(std::size_t core_index, VAddr addr) = 0; + virtual void SetExclusive128(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/hle/kernel/address_arbiter.cpp b/src/core/hle/kernel/address_arbiter.cpp index ebabde9213..07acabc1d5 100644 --- a/src/core/hle/kernel/address_arbiter.cpp +++ b/src/core/hle/kernel/address_arbiter.cpp @@ -90,7 +90,7 @@ ResultCode AddressArbiter::IncrementAndSignalToAddressIfEqual(VAddr address, s32 auto& monitor = system.Monitor(); u32 current_value; do { - monitor.SetExclusive(current_core, address); + monitor.SetExclusive32(current_core, address); current_value = memory.Read32(address); if (current_value != value) { @@ -120,7 +120,7 @@ ResultCode AddressArbiter::ModifyByWaitingCountAndSignalToAddressIfEqual(VAddr a auto& monitor = system.Monitor(); s32 updated_value; do { - monitor.SetExclusive(current_core, address); + monitor.SetExclusive32(current_core, address); updated_value = memory.Read32(address); if (updated_value != value) { @@ -191,7 +191,7 @@ ResultCode AddressArbiter::WaitForAddressIfLessThan(VAddr address, s32 value, s6 const std::size_t current_core = system.CurrentCoreIndex(); auto& monitor = system.Monitor(); do { - monitor.SetExclusive(current_core, address); + monitor.SetExclusive32(current_core, address); current_value = static_cast(memory.Read32(address)); if (should_decrement) { decrement_value = current_value - 1; diff --git a/src/core/hle/kernel/mutex.cpp b/src/core/hle/kernel/mutex.cpp index ebe3f6050e..16c95782a3 100644 --- a/src/core/hle/kernel/mutex.cpp +++ b/src/core/hle/kernel/mutex.cpp @@ -10,6 +10,7 @@ #include "common/logging/log.h" #include "core/core.h" #include "core/arm/exclusive_monitor.h" +#include "core/core.h" #include "core/hle/kernel/errors.h" #include "core/hle/kernel/handle_table.h" #include "core/hle/kernel/kernel.h" @@ -138,7 +139,7 @@ std::pair> Mutex::Unlock(std::shared_ptr> Mutex::Unlock(std::shared_ptrResumeFromWait(); do { - monitor.SetExclusive(current_core, address); + monitor.SetExclusive32(current_core, address); } while (!monitor.ExclusiveWrite32(current_core, address, mutex_value)); return {RESULT_SUCCESS, new_owner}; } diff --git a/src/core/hle/kernel/svc.cpp b/src/core/hle/kernel/svc.cpp index da2f90a1d1..371beed0d7 100644 --- a/src/core/hle/kernel/svc.cpp +++ b/src/core/hle/kernel/svc.cpp @@ -1641,7 +1641,7 @@ static void SignalProcessWideKey(Core::System& system, VAddr condition_variable_ u32 update_val = 0; const VAddr mutex_address = thread->GetMutexWaitAddress(); do { - monitor.SetExclusive(current_core, mutex_address); + monitor.SetExclusive32(current_core, mutex_address); // If the mutex is not yet acquired, acquire it. mutex_val = memory.Read32(mutex_address); diff --git a/src/core/hle/kernel/thread.cpp b/src/core/hle/kernel/thread.cpp index b99e3b7a55..51cc5dcca6 100644 --- a/src/core/hle/kernel/thread.cpp +++ b/src/core/hle/kernel/thread.cpp @@ -236,7 +236,7 @@ ResultVal> Thread::Create(Core::System& system, ThreadTy ResetThreadContext64(thread->context_64, stack_top, entry_point, arg); } thread->host_context = - std::make_shared(std::move(thread_start_func), thread_start_parameter); + std::make_shared(std::move(thread_start_func), thread_start_parameter); return MakeResult>(std::move(thread)); } @@ -412,12 +412,12 @@ ResultCode Thread::SetActivity(ThreadActivity value) { } if (value == ThreadActivity::Paused) { - if (pausing_state & static_cast(ThreadSchedFlags::ThreadPauseFlag) != 0) { + if ((pausing_state & static_cast(ThreadSchedFlags::ThreadPauseFlag)) != 0) { return ERR_INVALID_STATE; } AddSchedulingFlag(ThreadSchedFlags::ThreadPauseFlag); } else { - if (pausing_state & static_cast(ThreadSchedFlags::ThreadPauseFlag) == 0) { + if ((pausing_state & static_cast(ThreadSchedFlags::ThreadPauseFlag)) == 0) { return ERR_INVALID_STATE; } RemoveSchedulingFlag(ThreadSchedFlags::ThreadPauseFlag); diff --git a/src/core/memory.cpp b/src/core/memory.cpp index 66634596d1..4cb5d05e59 100644 --- a/src/core/memory.cpp +++ b/src/core/memory.cpp @@ -8,6 +8,7 @@ #include #include "common/assert.h" +#include "common/atomic_ops.h" #include "common/common_types.h" #include "common/logging/log.h" #include "common/page_table.h" @@ -176,6 +177,22 @@ struct Memory::Impl { } } + bool WriteExclusive8(const VAddr addr, const u8 data, const u8 expected) { + return WriteExclusive(addr, data, expected); + } + + bool WriteExclusive16(const VAddr addr, const u16 data, const u16 expected) { + return WriteExclusive(addr, data, expected); + } + + bool WriteExclusive32(const VAddr addr, const u32 data, const u32 expected) { + return WriteExclusive(addr, data, expected); + } + + bool WriteExclusive64(const VAddr addr, const u64 data, const u64 expected) { + return WriteExclusive(addr, data, expected); + } + std::string ReadCString(VAddr vaddr, std::size_t max_length) { std::string string; string.reserve(max_length); @@ -679,6 +696,67 @@ struct Memory::Impl { } } + template + bool WriteExclusive(const VAddr vaddr, const T data, const T expected) { + u8* page_pointer = current_page_table->pointers[vaddr >> PAGE_BITS]; + if (page_pointer != nullptr) { + // NOTE: Avoid adding any extra logic to this fast-path block + T volatile* pointer = reinterpret_cast(&page_pointer[vaddr]); + return Common::AtomicCompareAndSwap(pointer, data, expected); + } + + const Common::PageType type = current_page_table->attributes[vaddr >> PAGE_BITS]; + switch (type) { + case Common::PageType::Unmapped: + LOG_ERROR(HW_Memory, "Unmapped Write{} 0x{:08X} @ 0x{:016X}", sizeof(data) * 8, + static_cast(data), vaddr); + return true; + case Common::PageType::Memory: + ASSERT_MSG(false, "Mapped memory page without a pointer @ {:016X}", vaddr); + break; + case Common::PageType::RasterizerCachedMemory: { + u8* host_ptr{GetPointerFromVMA(vaddr)}; + system.GPU().InvalidateRegion(ToCacheAddr(host_ptr), sizeof(T)); + T volatile* pointer = reinterpret_cast(&host_ptr); + return Common::AtomicCompareAndSwap(pointer, data, expected); + break; + } + default: + UNREACHABLE(); + } + return true; + } + + bool WriteExclusive128(const VAddr vaddr, const u128 data, const u128 expected) { + u8* const page_pointer = current_page_table->pointers[vaddr >> PAGE_BITS]; + if (page_pointer != nullptr) { + // NOTE: Avoid adding any extra logic to this fast-path block + u64 volatile* pointer = reinterpret_cast(&page_pointer[vaddr]); + return Common::AtomicCompareAndSwap(pointer, data, expected); + } + + const Common::PageType type = current_page_table->attributes[vaddr >> PAGE_BITS]; + switch (type) { + case Common::PageType::Unmapped: + LOG_ERROR(HW_Memory, "Unmapped Write{} 0x{:08X} @ 0x{:016X}{:016X}", sizeof(data) * 8, + static_cast(data[1]), static_cast(data[0]), vaddr); + return true; + case Common::PageType::Memory: + ASSERT_MSG(false, "Mapped memory page without a pointer @ {:016X}", vaddr); + break; + case Common::PageType::RasterizerCachedMemory: { + u8* host_ptr{GetPointerFromVMA(vaddr)}; + system.GPU().InvalidateRegion(ToCacheAddr(host_ptr), sizeof(u128)); + u64 volatile* pointer = reinterpret_cast(&host_ptr); + return Common::AtomicCompareAndSwap(pointer, data, expected); + break; + } + default: + UNREACHABLE(); + } + return true; + } + Common::PageTable* current_page_table = nullptr; Core::System& system; }; @@ -761,6 +839,26 @@ void Memory::Write64(VAddr addr, u64 data) { impl->Write64(addr, data); } +bool Memory::WriteExclusive8(VAddr addr, u8 data, u8 expected) { + return impl->WriteExclusive8(addr, data, expected); +} + +bool Memory::WriteExclusive16(VAddr addr, u16 data, u16 expected) { + return impl->WriteExclusive16(addr, data, expected); +} + +bool Memory::WriteExclusive32(VAddr addr, u32 data, u32 expected) { + return impl->WriteExclusive32(addr, data, expected); +} + +bool Memory::WriteExclusive64(VAddr addr, u64 data, u64 expected) { + return impl->WriteExclusive64(addr, data, expected); +} + +bool Memory::WriteExclusive128(VAddr addr, u128 data, u128 expected) { + return impl->WriteExclusive128(addr, data, expected); +} + std::string Memory::ReadCString(VAddr vaddr, std::size_t max_length) { return impl->ReadCString(vaddr, max_length); } diff --git a/src/core/memory.h b/src/core/memory.h index 93f0c1d6c4..4a1cc63f4d 100644 --- a/src/core/memory.h +++ b/src/core/memory.h @@ -244,6 +244,71 @@ public: */ void Write64(VAddr addr, u64 data); + /** + * Writes a 8-bit unsigned integer to the given virtual address in + * the current process' address space if and only if the address contains + * the expected value. This operation is atomic. + * + * @param addr The virtual address to write the 8-bit unsigned integer to. + * @param data The 8-bit unsigned integer to write to the given virtual address. + * @param expected The 8-bit unsigned integer to check against the given virtual address. + * + * @post The memory range [addr, sizeof(data)) contains the given data value. + */ + bool WriteExclusive8(VAddr addr, u8 data, u8 expected); + + /** + * Writes a 16-bit unsigned integer to the given virtual address in + * the current process' address space if and only if the address contains + * the expected value. This operation is atomic. + * + * @param addr The virtual address to write the 16-bit unsigned integer to. + * @param data The 16-bit unsigned integer to write to the given virtual address. + * @param expected The 16-bit unsigned integer to check against the given virtual address. + * + * @post The memory range [addr, sizeof(data)) contains the given data value. + */ + bool WriteExclusive16(VAddr addr, u16 data, u16 expected); + + /** + * Writes a 32-bit unsigned integer to the given virtual address in + * the current process' address space if and only if the address contains + * the expected value. This operation is atomic. + * + * @param addr The virtual address to write the 32-bit unsigned integer to. + * @param data The 32-bit unsigned integer to write to the given virtual address. + * @param expected The 32-bit unsigned integer to check against the given virtual address. + * + * @post The memory range [addr, sizeof(data)) contains the given data value. + */ + bool WriteExclusive32(VAddr addr, u32 data, u32 expected); + + /** + * Writes a 64-bit unsigned integer to the given virtual address in + * the current process' address space if and only if the address contains + * the expected value. This operation is atomic. + * + * @param addr The virtual address to write the 64-bit unsigned integer to. + * @param data The 64-bit unsigned integer to write to the given virtual address. + * @param expected The 64-bit unsigned integer to check against the given virtual address. + * + * @post The memory range [addr, sizeof(data)) contains the given data value. + */ + bool WriteExclusive64(VAddr addr, u64 data, u64 expected); + + /** + * Writes a 128-bit unsigned integer to the given virtual address in + * the current process' address space if and only if the address contains + * the expected value. This operation is atomic. + * + * @param addr The virtual address to write the 128-bit unsigned integer to. + * @param data The 128-bit unsigned integer to write to the given virtual address. + * @param expected The 128-bit unsigned integer to check against the given virtual address. + * + * @post The memory range [addr, sizeof(data)) contains the given data value. + */ + bool WriteExclusive128(VAddr addr, u128 data, u128 expected); + /** * Reads a null-terminated string from the given virtual address. * This function will continually read characters until either: -- cgit v1.2.3 From cc2f7dcd08ae9455524098d2c575eed484d8f08c Mon Sep 17 00:00:00 2001 From: Fernando Sahmkow Date: Sun, 8 Mar 2020 22:39:41 -0400 Subject: General: Initial Setup for Single Core. --- src/core/core.cpp | 3 + src/core/cpu_manager.cpp | 186 +++++++++++++++++++++++---- src/core/cpu_manager.h | 30 ++++- src/core/hle/kernel/kernel.cpp | 19 +++ src/core/hle/kernel/kernel.h | 3 + src/core/memory.cpp | 8 +- src/yuzu/configuration/configure_general.cpp | 6 + src/yuzu/configuration/configure_general.ui | 7 + 8 files changed, 228 insertions(+), 34 deletions(-) (limited to 'src/core/memory.cpp') diff --git a/src/core/core.cpp b/src/core/core.cpp index 5d4ecdce5a..fd1bdcaf0a 100644 --- a/src/core/core.cpp +++ b/src/core/core.cpp @@ -149,6 +149,9 @@ struct System::Impl { device_memory = std::make_unique(system); + kernel.SetMulticore(Settings::values.use_multi_core); + cpu_manager.SetMulticore(Settings::values.use_multi_core); + core_timing.Initialize([&system]() { system.RegisterHostThread(); }); kernel.Initialize(); cpu_manager.Initialize(); diff --git a/src/core/cpu_manager.cpp b/src/core/cpu_manager.cpp index 9a261968a7..e72f898083 100644 --- a/src/core/cpu_manager.cpp +++ b/src/core/cpu_manager.cpp @@ -26,9 +26,13 @@ void CpuManager::ThreadStart(CpuManager& cpu_manager, std::size_t core) { void CpuManager::Initialize() { running_mode = true; - for (std::size_t core = 0; core < Core::Hardware::NUM_CPU_CORES; core++) { - core_data[core].host_thread = - std::make_unique(ThreadStart, std::ref(*this), core); + if (is_multicore) { + for (std::size_t core = 0; core < Core::Hardware::NUM_CPU_CORES; core++) { + core_data[core].host_thread = + std::make_unique(ThreadStart, std::ref(*this), core); + } + } else { + core_data[0].host_thread = std::make_unique(ThreadStart, std::ref(*this), 0); } } @@ -41,52 +45,72 @@ void CpuManager::Shutdown() { } } +std::function CpuManager::GetGuestThreadStartFunc() { + return std::function(GuestThreadFunction); +} + +std::function CpuManager::GetIdleThreadStartFunc() { + return std::function(IdleThreadFunction); +} + +std::function CpuManager::GetSuspendThreadStartFunc() { + return std::function(SuspendThreadFunction); +} + void CpuManager::GuestThreadFunction(void* cpu_manager_) { CpuManager* cpu_manager = static_cast(cpu_manager_); - cpu_manager->RunGuestThread(); + if (cpu_manager->is_multicore) { + cpu_manager->MultiCoreRunGuestThread(); + } else { + cpu_manager->SingleCoreRunGuestThread(); + } } void CpuManager::GuestRewindFunction(void* cpu_manager_) { CpuManager* cpu_manager = static_cast(cpu_manager_); - cpu_manager->RunGuestLoop(); + if (cpu_manager->is_multicore) { + cpu_manager->MultiCoreRunGuestLoop(); + } else { + cpu_manager->SingleCoreRunGuestLoop(); + } } void CpuManager::IdleThreadFunction(void* cpu_manager_) { CpuManager* cpu_manager = static_cast(cpu_manager_); - cpu_manager->RunIdleThread(); + if (cpu_manager->is_multicore) { + cpu_manager->MultiCoreRunIdleThread(); + } else { + cpu_manager->SingleCoreRunIdleThread(); + } } void CpuManager::SuspendThreadFunction(void* cpu_manager_) { CpuManager* cpu_manager = static_cast(cpu_manager_); - cpu_manager->RunSuspendThread(); -} - -std::function CpuManager::GetGuestThreadStartFunc() { - return std::function(GuestThreadFunction); -} - -std::function CpuManager::GetIdleThreadStartFunc() { - return std::function(IdleThreadFunction); -} - -std::function CpuManager::GetSuspendThreadStartFunc() { - return std::function(SuspendThreadFunction); + if (cpu_manager->is_multicore) { + cpu_manager->MultiCoreRunSuspendThread(); + } else { + cpu_manager->SingleCoreRunSuspendThread(); + } } void* CpuManager::GetStartFuncParamater() { return static_cast(this); } -void CpuManager::RunGuestThread() { +/////////////////////////////////////////////////////////////////////////////// +/// MultiCore /// +/////////////////////////////////////////////////////////////////////////////// + +void CpuManager::MultiCoreRunGuestThread() { auto& kernel = system.Kernel(); { auto& sched = kernel.CurrentScheduler(); sched.OnThreadStart(); } - RunGuestLoop(); + MultiCoreRunGuestLoop(); } -void CpuManager::RunGuestLoop() { +void CpuManager::MultiCoreRunGuestLoop() { auto& kernel = system.Kernel(); auto* thread = kernel.CurrentScheduler().GetCurrentThread(); auto host_context = thread->GetHostContext(); @@ -103,7 +127,7 @@ void CpuManager::RunGuestLoop() { } } -void CpuManager::RunIdleThread() { +void CpuManager::MultiCoreRunIdleThread() { auto& kernel = system.Kernel(); while (true) { auto& physical_core = kernel.CurrentPhysicalCore(); @@ -113,7 +137,7 @@ void CpuManager::RunIdleThread() { } } -void CpuManager::RunSuspendThread() { +void CpuManager::MultiCoreRunSuspendThread() { auto& kernel = system.Kernel(); { auto& sched = kernel.CurrentScheduler(); @@ -130,7 +154,7 @@ void CpuManager::RunSuspendThread() { } } -void CpuManager::Pause(bool paused) { +void CpuManager::MultiCorePause(bool paused) { if (!paused) { bool all_not_barrier = false; while (!all_not_barrier) { @@ -171,10 +195,120 @@ void CpuManager::Pause(bool paused) { paused_state = paused; } +/////////////////////////////////////////////////////////////////////////////// +/// SingleCore /// +/////////////////////////////////////////////////////////////////////////////// + +void CpuManager::SingleCoreRunGuestThread() { + auto& kernel = system.Kernel(); + { + auto& sched = kernel.CurrentScheduler(); + sched.OnThreadStart(); + } + SingleCoreRunGuestLoop(); +} + +void CpuManager::SingleCoreRunGuestLoop() { + auto& kernel = system.Kernel(); + auto* thread = kernel.CurrentScheduler().GetCurrentThread(); + auto host_context = thread->GetHostContext(); + host_context->SetRewindPoint(std::function(GuestRewindFunction), this); + host_context.reset(); + while (true) { + auto& physical_core = kernel.CurrentPhysicalCore(); + while (!physical_core.IsInterrupted()) { + physical_core.Run(); + preemption_count++; + if (preemption_count % max_cycle_runs == 0) { + break; + } + } + physical_core.ClearExclusive(); + PreemptSingleCore(); + auto& scheduler = physical_core.Scheduler(); + scheduler.TryDoContextSwitch(); + } +} + +void CpuManager::SingleCoreRunIdleThread() { + auto& kernel = system.Kernel(); + while (true) { + auto& physical_core = kernel.CurrentPhysicalCore(); + PreemptSingleCore(); + auto& scheduler = physical_core.Scheduler(); + scheduler.TryDoContextSwitch(); + } +} + +void CpuManager::SingleCoreRunSuspendThread() { + auto& kernel = system.Kernel(); + { + auto& sched = kernel.CurrentScheduler(); + sched.OnThreadStart(); + } + while (true) { + auto core = kernel.GetCurrentHostThreadID(); + auto& scheduler = kernel.CurrentScheduler(); + Kernel::Thread* current_thread = scheduler.GetCurrentThread(); + Common::Fiber::YieldTo(current_thread->GetHostContext(), core_data[0].host_context); + ASSERT(scheduler.ContextSwitchPending()); + ASSERT(core == kernel.GetCurrentHostThreadID()); + scheduler.TryDoContextSwitch(); + } +} + +void CpuManager::PreemptSingleCore() { + preemption_count = 0; + std::size_t old_core = current_core; + current_core = (current_core + 1) % Core::Hardware::NUM_CPU_CORES; + auto& scheduler = system.Kernel().Scheduler(old_core); + Kernel::Thread* current_thread = system.Kernel().Scheduler(old_core).GetCurrentThread(); + Kernel::Thread* next_thread = system.Kernel().Scheduler(current_core).GetCurrentThread(); + Common::Fiber::YieldTo(current_thread->GetHostContext(), next_thread->GetHostContext()); +} + +void CpuManager::SingleCorePause(bool paused) { + if (!paused) { + bool all_not_barrier = false; + while (!all_not_barrier) { + all_not_barrier = !core_data[0].is_running.load() && core_data[0].initialized.load(); + } + core_data[0].enter_barrier->Set(); + if (paused_state.load()) { + bool all_barrier = false; + while (!all_barrier) { + all_barrier = core_data[0].is_paused.load() && core_data[0].initialized.load(); + } + core_data[0].exit_barrier->Set(); + } + } else { + /// Wait until all cores are paused. + bool all_barrier = false; + while (!all_barrier) { + all_barrier = core_data[0].is_paused.load() && core_data[0].initialized.load(); + } + /// Don't release the barrier + } + paused_state = paused; +} + +void CpuManager::Pause(bool paused) { + if (is_multicore) { + MultiCorePause(paused); + } else { + SingleCorePause(paused); + } +} + void CpuManager::RunThread(std::size_t core) { /// Initialization system.RegisterCoreThread(core); - std::string name = "yuzu:CoreHostThread_" + std::to_string(core); + std::string name; + if (is_multicore) { + name = "yuzu:CoreCPUThread_" + std::to_string(core); + } else { + name = "yuzu:CPUThread"; + } MicroProfileOnThreadCreate(name.c_str()); Common::SetCurrentThreadName(name.c_str()); auto& data = core_data[core]; diff --git a/src/core/cpu_manager.h b/src/core/cpu_manager.h index e83ab20f9c..1e81481ec2 100644 --- a/src/core/cpu_manager.h +++ b/src/core/cpu_manager.h @@ -30,6 +30,10 @@ public: CpuManager& operator=(const CpuManager&) = delete; CpuManager& operator=(CpuManager&&) = delete; + /// Sets if emulation is multicore or single core, must be set before Initialize + void SetMulticore(bool is_multicore) { + this->is_multicore = is_multicore; + } void Initialize(); void Shutdown(); @@ -40,21 +44,34 @@ public: std::function GetSuspendThreadStartFunc(); void* GetStartFuncParamater(); + std::size_t CurrentCore() const { + return current_core; + } + private: static void GuestThreadFunction(void* cpu_manager); static void GuestRewindFunction(void* cpu_manager); static void IdleThreadFunction(void* cpu_manager); static void SuspendThreadFunction(void* cpu_manager); - void RunGuestThread(); - void RunGuestLoop(); - void RunIdleThread(); - void RunSuspendThread(); + void MultiCoreRunGuestThread(); + void MultiCoreRunGuestLoop(); + void MultiCoreRunIdleThread(); + void MultiCoreRunSuspendThread(); + void MultiCorePause(bool paused); + + void SingleCoreRunGuestThread(); + void SingleCoreRunGuestLoop(); + void SingleCoreRunIdleThread(); + void SingleCoreRunSuspendThread(); + void SingleCorePause(bool paused); static void ThreadStart(CpuManager& cpu_manager, std::size_t core); void RunThread(std::size_t core); + void PreemptSingleCore(); + struct CoreData { std::shared_ptr host_context; std::unique_ptr enter_barrier; @@ -70,6 +87,11 @@ private: std::array core_data{}; + bool is_multicore{}; + std::size_t current_core{}; + std::size_t preemption_count{}; + static constexpr std::size_t max_cycle_runs = 5; + System& system; }; diff --git a/src/core/hle/kernel/kernel.cpp b/src/core/hle/kernel/kernel.cpp index 721ab1e708..4a091ea381 100644 --- a/src/core/hle/kernel/kernel.cpp +++ b/src/core/hle/kernel/kernel.cpp @@ -113,6 +113,10 @@ struct KernelCore::Impl { explicit Impl(Core::System& system, KernelCore& kernel) : global_scheduler{kernel}, synchronization{system}, time_manager{system}, system{system} {} + void SetMulticore(bool is_multicore) { + this->is_multicore = is_multicore; + } + void Initialize(KernelCore& kernel) { Shutdown(); @@ -237,6 +241,9 @@ struct KernelCore::Impl { void RegisterCoreThread(std::size_t core_id) { std::unique_lock lock{register_thread_mutex}; + if (!is_multicore) { + single_core_thread_id = std::this_thread::get_id(); + } const std::thread::id this_id = std::this_thread::get_id(); const auto it = host_thread_ids.find(this_id); ASSERT(core_id < Core::Hardware::NUM_CPU_CORES); @@ -258,6 +265,11 @@ struct KernelCore::Impl { u32 GetCurrentHostThreadID() const { const std::thread::id this_id = std::this_thread::get_id(); + if (!is_multicore) { + if (single_core_thread_id == this_id) { + return static_cast(system.GetCpuManager().CurrentCore()); + } + } const auto it = host_thread_ids.find(this_id); if (it == host_thread_ids.end()) { return Core::INVALID_HOST_THREAD_ID; @@ -378,6 +390,9 @@ struct KernelCore::Impl { std::array, Core::Hardware::NUM_CPU_CORES> suspend_threads{}; + bool is_multicore{}; + std::thread::id single_core_thread_id{}; + // System context Core::System& system; }; @@ -387,6 +402,10 @@ KernelCore::~KernelCore() { Shutdown(); } +void KernelCore::SetMulticore(bool is_multicore) { + impl->SetMulticore(is_multicore); +} + void KernelCore::Initialize() { impl->Initialize(*this); } diff --git a/src/core/hle/kernel/kernel.h b/src/core/hle/kernel/kernel.h index 5d32a83294..162bbd2f8d 100644 --- a/src/core/hle/kernel/kernel.h +++ b/src/core/hle/kernel/kernel.h @@ -65,6 +65,9 @@ public: KernelCore(KernelCore&&) = delete; KernelCore& operator=(KernelCore&&) = delete; + /// Sets if emulation is multicore or single core, must be set before Initialize + void SetMulticore(bool is_multicore); + /// Resets the kernel to a clean slate for use. void Initialize(); diff --git a/src/core/memory.cpp b/src/core/memory.cpp index 4cb5d05e59..7def007689 100644 --- a/src/core/memory.cpp +++ b/src/core/memory.cpp @@ -715,8 +715,8 @@ struct Memory::Impl { ASSERT_MSG(false, "Mapped memory page without a pointer @ {:016X}", vaddr); break; case Common::PageType::RasterizerCachedMemory: { - u8* host_ptr{GetPointerFromVMA(vaddr)}; - system.GPU().InvalidateRegion(ToCacheAddr(host_ptr), sizeof(T)); + u8* host_ptr{GetPointerFromRasterizerCachedMemory(vaddr)}; + system.GPU().InvalidateRegion(vaddr, sizeof(T)); T volatile* pointer = reinterpret_cast(&host_ptr); return Common::AtomicCompareAndSwap(pointer, data, expected); break; @@ -745,8 +745,8 @@ struct Memory::Impl { ASSERT_MSG(false, "Mapped memory page without a pointer @ {:016X}", vaddr); break; case Common::PageType::RasterizerCachedMemory: { - u8* host_ptr{GetPointerFromVMA(vaddr)}; - system.GPU().InvalidateRegion(ToCacheAddr(host_ptr), sizeof(u128)); + u8* host_ptr{GetPointerFromRasterizerCachedMemory(vaddr)}; + system.GPU().InvalidateRegion(vaddr, sizeof(u128)); u64 volatile* pointer = reinterpret_cast(&host_ptr); return Common::AtomicCompareAndSwap(pointer, data, expected); break; diff --git a/src/yuzu/configuration/configure_general.cpp b/src/yuzu/configuration/configure_general.cpp index cb95423e09..74b2ad537b 100644 --- a/src/yuzu/configuration/configure_general.cpp +++ b/src/yuzu/configuration/configure_general.cpp @@ -23,6 +23,11 @@ ConfigureGeneral::ConfigureGeneral(QWidget* parent) ConfigureGeneral::~ConfigureGeneral() = default; void ConfigureGeneral::SetConfiguration() { + const bool runtime_lock = !Core::System::GetInstance().IsPoweredOn(); + + ui->use_multi_core->setEnabled(runtime_lock); + ui->use_multi_core->setChecked(Settings::values.use_multi_core); + ui->toggle_check_exit->setChecked(UISettings::values.confirm_before_closing); ui->toggle_user_on_boot->setChecked(UISettings::values.select_user_on_boot); ui->toggle_background_pause->setChecked(UISettings::values.pause_when_in_background); @@ -41,6 +46,7 @@ void ConfigureGeneral::ApplyConfiguration() { Settings::values.use_frame_limit = ui->toggle_frame_limit->isChecked(); Settings::values.frame_limit = ui->frame_limit->value(); + Settings::values.use_multi_core = ui->use_multi_core->isChecked(); } void ConfigureGeneral::changeEvent(QEvent* event) { diff --git a/src/yuzu/configuration/configure_general.ui b/src/yuzu/configuration/configure_general.ui index fc3b7e65a5..f872bddecf 100644 --- a/src/yuzu/configuration/configure_general.ui +++ b/src/yuzu/configuration/configure_general.ui @@ -51,6 +51,13 @@ + + + + Emulate CPU in Multiple Cores + + + -- cgit v1.2.3