diff options
Diffstat (limited to 'src/core')
-rw-r--r-- | src/core/arm/arm_interface.cpp | 17 | ||||
-rw-r--r-- | src/core/arm/arm_interface.h | 1 | ||||
-rw-r--r-- | src/core/arm/dynarmic/arm_dynarmic_32.cpp | 42 | ||||
-rw-r--r-- | src/core/arm/dynarmic/arm_dynarmic_64.cpp | 28 | ||||
-rw-r--r-- | src/core/core_timing.cpp | 146 | ||||
-rw-r--r-- | src/core/core_timing.h | 27 | ||||
-rw-r--r-- | src/core/cpu_manager.cpp | 4 | ||||
-rw-r--r-- | src/core/hle/kernel/k_scheduler.h | 5 |
8 files changed, 185 insertions, 85 deletions
diff --git a/src/core/arm/arm_interface.cpp b/src/core/arm/arm_interface.cpp index 8e095cdcd..0efc3732f 100644 --- a/src/core/arm/arm_interface.cpp +++ b/src/core/arm/arm_interface.cpp @@ -119,16 +119,23 @@ void ARM_Interface::Run() { } system.ExitDynarmicProfile(); - // Notify the debugger and go to sleep if a breakpoint was hit. - if (Has(hr, breakpoint)) { + // Notify the debugger and go to sleep if a breakpoint was hit, + // or if the thread is unable to continue for any reason. + if (Has(hr, breakpoint) || Has(hr, no_execute)) { RewindBreakpointInstruction(); - system.GetDebugger().NotifyThreadStopped(current_thread); - current_thread->RequestSuspend(SuspendType::Debug); + if (system.DebuggerEnabled()) { + system.GetDebugger().NotifyThreadStopped(current_thread); + } + current_thread->RequestSuspend(Kernel::SuspendType::Debug); break; } + + // Notify the debugger and go to sleep if a watchpoint was hit. if (Has(hr, watchpoint)) { RewindBreakpointInstruction(); - system.GetDebugger().NotifyThreadWatchpoint(current_thread, *HaltedWatchpoint()); + if (system.DebuggerEnabled()) { + system.GetDebugger().NotifyThreadWatchpoint(current_thread, *HaltedWatchpoint()); + } current_thread->RequestSuspend(SuspendType::Debug); break; } diff --git a/src/core/arm/arm_interface.h b/src/core/arm/arm_interface.h index 4e431e27a..8a066ed91 100644 --- a/src/core/arm/arm_interface.h +++ b/src/core/arm/arm_interface.h @@ -204,6 +204,7 @@ public: static constexpr Dynarmic::HaltReason svc_call = Dynarmic::HaltReason::UserDefined3; static constexpr Dynarmic::HaltReason breakpoint = Dynarmic::HaltReason::UserDefined4; static constexpr Dynarmic::HaltReason watchpoint = Dynarmic::HaltReason::UserDefined5; + static constexpr Dynarmic::HaltReason no_execute = Dynarmic::HaltReason::UserDefined6; protected: /// System context that this ARM interface is running under. diff --git a/src/core/arm/dynarmic/arm_dynarmic_32.cpp b/src/core/arm/dynarmic/arm_dynarmic_32.cpp index 8c90c8be0..10cf72a45 100644 --- a/src/core/arm/dynarmic/arm_dynarmic_32.cpp +++ b/src/core/arm/dynarmic/arm_dynarmic_32.cpp @@ -48,6 +48,12 @@ public: CheckMemoryAccess(vaddr, 8, Kernel::DebugWatchpointType::Read); return memory.Read64(vaddr); } + std::optional<u32> MemoryReadCode(u32 vaddr) override { + if (!memory.IsValidVirtualAddressRange(vaddr, sizeof(u32))) { + return std::nullopt; + } + return MemoryRead32(vaddr); + } void MemoryWrite8(u32 vaddr, u8 value) override { if (CheckMemoryAccess(vaddr, 1, Kernel::DebugWatchpointType::Write)) { @@ -89,21 +95,28 @@ public: void InterpreterFallback(u32 pc, std::size_t num_instructions) override { parent.LogBacktrace(); - UNIMPLEMENTED_MSG("This should never happen, pc = {:08X}, code = {:08X}", pc, - MemoryReadCode(pc)); + LOG_ERROR(Core_ARM, + "Unimplemented instruction @ 0x{:X} for {} instructions (instr = {:08X})", pc, + num_instructions, MemoryRead32(pc)); } void ExceptionRaised(u32 pc, Dynarmic::A32::Exception exception) override { - if (debugger_enabled) { - parent.SaveContext(parent.breakpoint_context); - parent.jit.load()->HaltExecution(ARM_Interface::breakpoint); + switch (exception) { + case Dynarmic::A32::Exception::NoExecuteFault: + LOG_CRITICAL(Core_ARM, "Cannot execute instruction at unmapped address {:#08x}", pc); + ReturnException(pc, ARM_Interface::no_execute); return; - } + default: + if (debugger_enabled) { + ReturnException(pc, ARM_Interface::breakpoint); + return; + } - parent.LogBacktrace(); - LOG_CRITICAL(Core_ARM, - "ExceptionRaised(exception = {}, pc = {:08X}, code = {:08X}, thumb = {})", - exception, pc, MemoryReadCode(pc), parent.IsInThumbMode()); + parent.LogBacktrace(); + LOG_CRITICAL(Core_ARM, + "ExceptionRaised(exception = {}, pc = {:08X}, code = {:08X}, thumb = {})", + exception, pc, MemoryRead32(pc), parent.IsInThumbMode()); + } } void CallSVC(u32 swi) override { @@ -141,15 +154,20 @@ public: const auto match{parent.MatchingWatchpoint(addr, size, type)}; if (match) { - parent.SaveContext(parent.breakpoint_context); - parent.jit.load()->HaltExecution(ARM_Interface::watchpoint); parent.halted_watchpoint = match; + ReturnException(parent.jit.load()->Regs()[15], ARM_Interface::watchpoint); return false; } return true; } + void ReturnException(u32 pc, Dynarmic::HaltReason hr) { + parent.SaveContext(parent.breakpoint_context); + parent.breakpoint_context.cpu_registers[15] = pc; + parent.jit.load()->HaltExecution(hr); + } + ARM_Dynarmic_32& parent; Core::Memory::Memory& memory; std::size_t num_interpreted_instructions{}; diff --git a/src/core/arm/dynarmic/arm_dynarmic_64.cpp b/src/core/arm/dynarmic/arm_dynarmic_64.cpp index 4370ca294..92266aa9e 100644 --- a/src/core/arm/dynarmic/arm_dynarmic_64.cpp +++ b/src/core/arm/dynarmic/arm_dynarmic_64.cpp @@ -52,6 +52,12 @@ public: CheckMemoryAccess(vaddr, 16, Kernel::DebugWatchpointType::Read); return {memory.Read64(vaddr), memory.Read64(vaddr + 8)}; } + std::optional<u32> MemoryReadCode(u64 vaddr) override { + if (!memory.IsValidVirtualAddressRange(vaddr, sizeof(u32))) { + return std::nullopt; + } + return MemoryRead32(vaddr); + } void MemoryWrite8(u64 vaddr, u8 value) override { if (CheckMemoryAccess(vaddr, 1, Kernel::DebugWatchpointType::Write)) { @@ -105,7 +111,7 @@ public: parent.LogBacktrace(); LOG_ERROR(Core_ARM, "Unimplemented instruction @ 0x{:X} for {} instructions (instr = {:08X})", pc, - num_instructions, MemoryReadCode(pc)); + num_instructions, MemoryRead32(pc)); } void InstructionCacheOperationRaised(Dynarmic::A64::InstructionCacheOperation op, @@ -138,16 +144,19 @@ public: case Dynarmic::A64::Exception::SendEventLocal: case Dynarmic::A64::Exception::Yield: return; + case Dynarmic::A64::Exception::NoExecuteFault: + LOG_CRITICAL(Core_ARM, "Cannot execute instruction at unmapped address {:#016x}", pc); + ReturnException(pc, ARM_Interface::no_execute); + return; default: if (debugger_enabled) { - parent.SaveContext(parent.breakpoint_context); - parent.jit.load()->HaltExecution(ARM_Interface::breakpoint); + ReturnException(pc, ARM_Interface::breakpoint); return; } parent.LogBacktrace(); - ASSERT_MSG(false, "ExceptionRaised(exception = {}, pc = {:08X}, code = {:08X})", - static_cast<std::size_t>(exception), pc, MemoryReadCode(pc)); + LOG_CRITICAL(Core_ARM, "ExceptionRaised(exception = {}, pc = {:08X}, code = {:08X})", + static_cast<std::size_t>(exception), pc, MemoryRead32(pc)); } } @@ -188,15 +197,20 @@ public: const auto match{parent.MatchingWatchpoint(addr, size, type)}; if (match) { - parent.SaveContext(parent.breakpoint_context); - parent.jit.load()->HaltExecution(ARM_Interface::watchpoint); parent.halted_watchpoint = match; + ReturnException(parent.jit.load()->GetPC(), ARM_Interface::watchpoint); return false; } return true; } + void ReturnException(u64 pc, Dynarmic::HaltReason hr) { + parent.SaveContext(parent.breakpoint_context); + parent.breakpoint_context.pc = pc; + parent.jit.load()->HaltExecution(hr); + } + ARM_Dynarmic_64& parent; Core::Memory::Memory& memory; u64 tpidrro_el0 = 0; diff --git a/src/core/core_timing.cpp b/src/core/core_timing.cpp index 29e7dba9b..140578069 100644 --- a/src/core/core_timing.cpp +++ b/src/core/core_timing.cpp @@ -6,7 +6,9 @@ #include <string> #include <tuple> +#include "common/logging/log.h" #include "common/microprofile.h" +#include "common/thread.h" #include "core/core_timing.h" #include "core/core_timing_util.h" #include "core/hardware_properties.h" @@ -41,11 +43,11 @@ CoreTiming::CoreTiming() CoreTiming::~CoreTiming() = default; -void CoreTiming::ThreadEntry(CoreTiming& instance) { - constexpr char name[] = "yuzu:HostTiming"; - MicroProfileOnThreadCreate(name); - Common::SetCurrentThreadName(name); - Common::SetCurrentThreadPriority(Common::ThreadPriority::VeryHigh); +void CoreTiming::ThreadEntry(CoreTiming& instance, size_t id) { + const std::string name = "yuzu:HostTiming_" + std::to_string(id); + MicroProfileOnThreadCreate(name.c_str()); + Common::SetCurrentThreadName(name.c_str()); + Common::SetCurrentThreadPriority(Common::ThreadPriority::Critical); instance.on_thread_init(); instance.ThreadLoop(); MicroProfileOnThreadExit(); @@ -59,68 +61,97 @@ void CoreTiming::Initialize(std::function<void()>&& on_thread_init_) { const auto empty_timed_callback = [](std::uintptr_t, std::chrono::nanoseconds) {}; ev_lost = CreateEvent("_lost_event", empty_timed_callback); if (is_multicore) { - timer_thread = std::make_unique<std::thread>(ThreadEntry, std::ref(*this)); + const auto hardware_concurrency = std::thread::hardware_concurrency(); + size_t id = 0; + worker_threads.emplace_back(ThreadEntry, std::ref(*this), id++); + if (hardware_concurrency > 8) { + worker_threads.emplace_back(ThreadEntry, std::ref(*this), id++); + } } } void CoreTiming::Shutdown() { - paused = true; + is_paused = true; shutting_down = true; - pause_event.Set(); - event.Set(); - if (timer_thread) { - timer_thread->join(); + std::atomic_thread_fence(std::memory_order_release); + + event_cv.notify_all(); + wait_pause_cv.notify_all(); + for (auto& thread : worker_threads) { + thread.join(); } + worker_threads.clear(); ClearPendingEvents(); - timer_thread.reset(); has_started = false; } -void CoreTiming::Pause(bool is_paused) { - paused = is_paused; - pause_event.Set(); +void CoreTiming::Pause(bool is_paused_) { + std::unique_lock main_lock(event_mutex); + if (is_paused_ == paused_state.load(std::memory_order_relaxed)) { + return; + } + if (is_multicore) { + is_paused = is_paused_; + event_cv.notify_all(); + if (!is_paused_) { + wait_pause_cv.notify_all(); + } + } + paused_state.store(is_paused_, std::memory_order_relaxed); } -void CoreTiming::SyncPause(bool is_paused) { - if (is_paused == paused && paused_set == paused) { +void CoreTiming::SyncPause(bool is_paused_) { + std::unique_lock main_lock(event_mutex); + if (is_paused_ == paused_state.load(std::memory_order_relaxed)) { return; } - Pause(is_paused); - if (timer_thread) { - if (!is_paused) { - pause_event.Set(); + + if (is_multicore) { + is_paused = is_paused_; + event_cv.notify_all(); + if (!is_paused_) { + wait_pause_cv.notify_all(); + } + } + paused_state.store(is_paused_, std::memory_order_relaxed); + if (is_multicore) { + if (is_paused_) { + wait_signal_cv.wait(main_lock, [this] { return pause_count == worker_threads.size(); }); + } else { + wait_signal_cv.wait(main_lock, [this] { return pause_count == 0; }); } - event.Set(); - while (paused_set != is_paused) - ; } } bool CoreTiming::IsRunning() const { - return !paused_set; + return !paused_state.load(std::memory_order_acquire); } bool CoreTiming::HasPendingEvents() const { - return !(wait_set && event_queue.empty()); + std::unique_lock main_lock(event_mutex); + return !event_queue.empty() || pending_events.load(std::memory_order_relaxed) != 0; } void CoreTiming::ScheduleEvent(std::chrono::nanoseconds ns_into_future, const std::shared_ptr<EventType>& event_type, std::uintptr_t user_data) { - { - std::scoped_lock scope{basic_lock}; - const u64 timeout = static_cast<u64>((GetGlobalTimeNs() + ns_into_future).count()); - event_queue.emplace_back(Event{timeout, event_fifo_id++, user_data, event_type}); + std::unique_lock main_lock(event_mutex); + const u64 timeout = static_cast<u64>((GetGlobalTimeNs() + ns_into_future).count()); + + event_queue.emplace_back(Event{timeout, event_fifo_id++, user_data, event_type}); + pending_events.fetch_add(1, std::memory_order_relaxed); - std::push_heap(event_queue.begin(), event_queue.end(), std::greater<>()); + std::push_heap(event_queue.begin(), event_queue.end(), std::greater<>()); + + if (is_multicore) { + event_cv.notify_one(); } - event.Set(); } void CoreTiming::UnscheduleEvent(const std::shared_ptr<EventType>& event_type, std::uintptr_t user_data) { - std::scoped_lock scope{basic_lock}; + std::unique_lock main_lock(event_mutex); const auto itr = std::remove_if(event_queue.begin(), event_queue.end(), [&](const Event& e) { return e.type.lock().get() == event_type.get() && e.user_data == user_data; }); @@ -129,6 +160,7 @@ void CoreTiming::UnscheduleEvent(const std::shared_ptr<EventType>& event_type, if (itr != event_queue.end()) { event_queue.erase(itr, event_queue.end()); std::make_heap(event_queue.begin(), event_queue.end(), std::greater<>()); + pending_events.fetch_sub(1, std::memory_order_relaxed); } } @@ -168,11 +200,12 @@ u64 CoreTiming::GetClockTicks() const { } void CoreTiming::ClearPendingEvents() { + std::unique_lock main_lock(event_mutex); event_queue.clear(); } void CoreTiming::RemoveEvent(const std::shared_ptr<EventType>& event_type) { - std::scoped_lock lock{basic_lock}; + std::unique_lock main_lock(event_mutex); const auto itr = std::remove_if(event_queue.begin(), event_queue.end(), [&](const Event& e) { return e.type.lock().get() == event_type.get(); @@ -186,21 +219,28 @@ void CoreTiming::RemoveEvent(const std::shared_ptr<EventType>& event_type) { } std::optional<s64> CoreTiming::Advance() { - std::scoped_lock lock{advance_lock, basic_lock}; global_timer = GetGlobalTimeNs().count(); + std::unique_lock main_lock(event_mutex); while (!event_queue.empty() && event_queue.front().time <= global_timer) { Event evt = std::move(event_queue.front()); std::pop_heap(event_queue.begin(), event_queue.end(), std::greater<>()); event_queue.pop_back(); - basic_lock.unlock(); if (const auto event_type{evt.type.lock()}) { - event_type->callback( - evt.user_data, std::chrono::nanoseconds{static_cast<s64>(global_timer - evt.time)}); + sequence_mutex.lock(); + event_mutex.unlock(); + + event_type->guard.lock(); + sequence_mutex.unlock(); + const s64 delay = static_cast<s64>(GetGlobalTimeNs().count() - evt.time); + event_type->callback(evt.user_data, std::chrono::nanoseconds{delay}); + event_type->guard.unlock(); + + event_mutex.lock(); + pending_events.fetch_sub(1, std::memory_order_relaxed); } - basic_lock.lock(); global_timer = GetGlobalTimeNs().count(); } @@ -213,26 +253,34 @@ std::optional<s64> CoreTiming::Advance() { } void CoreTiming::ThreadLoop() { + const auto predicate = [this] { return !event_queue.empty() || is_paused; }; has_started = true; while (!shutting_down) { - while (!paused) { - paused_set = false; + while (!is_paused && !shutting_down) { const auto next_time = Advance(); if (next_time) { if (*next_time > 0) { std::chrono::nanoseconds next_time_ns = std::chrono::nanoseconds(*next_time); - event.WaitFor(next_time_ns); + std::unique_lock main_lock(event_mutex); + event_cv.wait_for(main_lock, next_time_ns, predicate); } } else { - wait_set = true; - event.Wait(); + std::unique_lock main_lock(event_mutex); + event_cv.wait(main_lock, predicate); } - wait_set = false; } - paused_set = true; - clock->Pause(true); - pause_event.Wait(); - clock->Pause(false); + std::unique_lock main_lock(event_mutex); + pause_count++; + if (pause_count == worker_threads.size()) { + clock->Pause(true); + wait_signal_cv.notify_all(); + } + wait_pause_cv.wait(main_lock, [this] { return !is_paused || shutting_down; }); + pause_count--; + if (pause_count == 0) { + clock->Pause(false); + wait_signal_cv.notify_all(); + } } } diff --git a/src/core/core_timing.h b/src/core/core_timing.h index d27773009..a86553e08 100644 --- a/src/core/core_timing.h +++ b/src/core/core_timing.h @@ -5,6 +5,7 @@ #include <atomic> #include <chrono> +#include <condition_variable> #include <functional> #include <memory> #include <mutex> @@ -14,7 +15,6 @@ #include <vector> #include "common/common_types.h" -#include "common/thread.h" #include "common/wall_clock.h" namespace Core::Timing { @@ -32,6 +32,7 @@ struct EventType { TimedCallback callback; /// A pointer to the name of the event. const std::string name; + mutable std::mutex guard; }; /** @@ -131,7 +132,7 @@ private: /// Clear all pending events. This should ONLY be done on exit. void ClearPendingEvents(); - static void ThreadEntry(CoreTiming& instance); + static void ThreadEntry(CoreTiming& instance, size_t id); void ThreadLoop(); std::unique_ptr<Common::WallClock> clock; @@ -144,21 +145,25 @@ private: // accomodated by the standard adaptor class. std::vector<Event> event_queue; u64 event_fifo_id = 0; + std::atomic<size_t> pending_events{}; std::shared_ptr<EventType> ev_lost; - Common::Event event{}; - Common::Event pause_event{}; - std::mutex basic_lock; - std::mutex advance_lock; - std::unique_ptr<std::thread> timer_thread; - std::atomic<bool> paused{}; - std::atomic<bool> paused_set{}; - std::atomic<bool> wait_set{}; - std::atomic<bool> shutting_down{}; std::atomic<bool> has_started{}; std::function<void()> on_thread_init{}; + std::vector<std::thread> worker_threads; + + std::condition_variable event_cv; + std::condition_variable wait_pause_cv; + std::condition_variable wait_signal_cv; + mutable std::mutex event_mutex; + mutable std::mutex sequence_mutex; + + std::atomic<bool> paused_state{}; + bool is_paused{}; + bool shutting_down{}; bool is_multicore{}; + size_t pause_count{}; /// Cycle timing u64 ticks{}; diff --git a/src/core/cpu_manager.cpp b/src/core/cpu_manager.cpp index fd6928105..9fc78f033 100644 --- a/src/core/cpu_manager.cpp +++ b/src/core/cpu_manager.cpp @@ -194,7 +194,9 @@ void CpuManager::PreemptSingleCore(bool from_running_enviroment) { { auto& scheduler = system.Kernel().Scheduler(current_core); scheduler.Reload(scheduler.GetSchedulerCurrentThread()); - idle_count = 0; + if (!scheduler.IsIdle()) { + idle_count = 0; + } } } diff --git a/src/core/hle/kernel/k_scheduler.h b/src/core/hle/kernel/k_scheduler.h index 3f90656ee..cc3da33f5 100644 --- a/src/core/hle/kernel/k_scheduler.h +++ b/src/core/hle/kernel/k_scheduler.h @@ -55,6 +55,11 @@ public: return idle_thread; } + /// Returns true if the scheduler is idle + [[nodiscard]] bool IsIdle() const { + return GetSchedulerCurrentThread() == idle_thread; + } + /// Gets the timestamp for the last context switch in ticks. [[nodiscard]] u64 GetLastContextSwitchTicks() const; |