early-access version 2668

README.md

@@ -1,7 +1,7 @@
 yuzu emulator early access
 =============
 
-This is the source code for early-access 2667.
+This is the source code for early-access 2668.
 
 ## Legal Notice
 

src/common/fiber.cpp

@@ -2,10 +2,9 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
-#include <mutex>
-
 #include "common/assert.h"
 #include "common/fiber.h"
+#include "common/spin_lock.h"
 #include "common/virtual_buffer.h"
 
 #include <boost/context/detail/fcontext.hpp>
@@ -20,7 +19,7 @@ struct Fiber::FiberImpl {
     VirtualBuffer<u8> stack;
     VirtualBuffer<u8> rewind_stack;
 
-    std::mutex guard;
+    SpinLock guard{};
     std::function<void(void*)> entry_point;
     std::function<void(void*)> rewind_point;
     void* rewind_parameter{};

src/core/core_timing.h

@@ -8,13 +8,13 @@
 #include <chrono>
 #include <functional>
 #include <memory>
-#include <mutex>
 #include <optional>
 #include <string>
 #include <thread>
 #include <vector>
 
 #include "common/common_types.h"
+#include "common/spin_lock.h"
 #include "common/thread.h"
 #include "common/wall_clock.h"
 
@@ -149,8 +149,8 @@ private:
     std::shared_ptr<EventType> ev_lost;
     Common::Event event{};
     Common::Event pause_event{};
-    std::mutex basic_lock;
-    std::mutex advance_lock;
+    Common::SpinLock basic_lock{};
+    Common::SpinLock advance_lock{};
     std::unique_ptr<std::thread> timer_thread;
     std::atomic<bool> paused{};
     std::atomic<bool> paused_set{};

src/core/hle/kernel/global_scheduler_context.h

@@ -8,6 +8,7 @@
 #include <vector>
 
 #include "common/common_types.h"
+#include "common/spin_lock.h"
 #include "core/hardware_properties.h"
 #include "core/hle/kernel/k_priority_queue.h"
 #include "core/hle/kernel/k_scheduler_lock.h"
@@ -79,7 +80,7 @@ private:
 
     /// Lists all thread ids that aren't deleted/etc.
     std::vector<KThread*> thread_list;
-    std::mutex global_list_guard;
+    Common::SpinLock global_list_guard{};
 };
 
 } // namespace Kernel

src/core/hle/kernel/k_scheduler.cpp

@@ -405,6 +405,9 @@ void KScheduler::EnableScheduling(KernelCore& kernel, u64 cores_needing_scheduli
     } else {
         RescheduleCores(kernel, cores_needing_scheduling);
     }
+
+    // Special case to ensure dummy threads that are waiting block.
+    current_thread->IfDummyThreadTryWait();
 }
 
 u64 KScheduler::UpdateHighestPriorityThreads(KernelCore& kernel) {
@@ -702,7 +705,7 @@ void KScheduler::Unload(KThread* thread) {
         prev_thread = nullptr;
     }
 
-    thread->context_guard.unlock();
+    thread->context_guard.Unlock();
 }
 
 void KScheduler::Reload(KThread* thread) {
@@ -791,13 +794,13 @@ void KScheduler::SwitchToCurrent() {
         do {
             auto next_thread = current_thread.load();
             if (next_thread != nullptr) {
-                const auto locked = next_thread->context_guard.try_lock();
+                const auto locked = next_thread->context_guard.TryLock();
                 if (state.needs_scheduling.load()) {
-                    next_thread->context_guard.unlock();
+                    next_thread->context_guard.Unlock();
                     break;
                 }
                 if (next_thread->GetActiveCore() != core_id) {
-                    next_thread->context_guard.unlock();
+                    next_thread->context_guard.Unlock();
                     break;
                 }
                 if (!locked) {

src/core/hle/kernel/k_spin_lock.cpp

@@ -4,18 +4,51 @@
 
 #include "core/hle/kernel/k_spin_lock.h"
 
+#if _MSC_VER
+#include <intrin.h>
+#if _M_AMD64
+#define __x86_64__ 1
+#endif
+#if _M_ARM64
+#define __aarch64__ 1
+#endif
+#else
+#if __x86_64__
+#include <xmmintrin.h>
+#endif
+#endif
+
+namespace {
+
+void ThreadPause() {
+#if __x86_64__
+    _mm_pause();
+#elif __aarch64__ && _MSC_VER
+    __yield();
+#elif __aarch64__
+    asm("yield");
+#endif
+}
+
+} // namespace
+
 namespace Kernel {
 
 void KSpinLock::Lock() {
-    lck.lock();
+    while (lck.test_and_set(std::memory_order_acquire)) {
+        ThreadPause();
+    }
 }
 
 void KSpinLock::Unlock() {
-    lck.unlock();
+    lck.clear(std::memory_order_release);
 }
 
 bool KSpinLock::TryLock() {
-    return lck.try_lock();
+    if (lck.test_and_set(std::memory_order_acquire)) {
+        return false;
+    }
+    return true;
 }
 
 } // namespace Kernel

src/core/hle/kernel/k_spin_lock.h

@@ -4,7 +4,7 @@
 
 #pragma once
 
-#include <mutex>
+#include <atomic>
 
 #include "core/hle/kernel/k_scoped_lock.h"
 
@@ -25,7 +25,7 @@ public:
     [[nodiscard]] bool TryLock();
 
 private:
-    std::mutex lck;
+    std::atomic_flag lck = ATOMIC_FLAG_INIT;
 };
 
 // TODO(bunnei): Alias for now, in case we want to implement these accurately in the future.

src/core/hle/kernel/k_thread.cpp

@@ -1070,12 +1070,46 @@ ResultCode KThread::Sleep(s64 timeout) {
     return ResultSuccess;
 }
 
+void KThread::IfDummyThreadTryWait() {
+    if (!IsDummyThread()) {
+        return;
+    }
+
+    if (GetState() != ThreadState::Waiting) {
+        return;
+    }
+
+    // Block until we can grab the lock.
+    KScopedSpinLock lk{dummy_wait_lock};
+}
+
+void KThread::IfDummyThreadBeginWait() {
+    if (!IsDummyThread()) {
+        return;
+    }
+
+    // Ensure the thread will block when IfDummyThreadTryWait is called.
+    dummy_wait_lock.Lock();
+}
+
+void KThread::IfDummyThreadEndWait() {
+    if (!IsDummyThread()) {
+        return;
+    }
+
+    // Ensure the thread will no longer block.
+    dummy_wait_lock.Unlock();
+}
+
 void KThread::BeginWait(KThreadQueue* queue) {
     // Set our state as waiting.
     SetState(ThreadState::Waiting);
 
     // Set our wait queue.
     wait_queue = queue;
+
+    // Special case for dummy threads to ensure they block.
+    IfDummyThreadBeginWait();
 }
 
 void KThread::NotifyAvailable(KSynchronizationObject* signaled_object, ResultCode wait_result_) {
@@ -1101,6 +1135,9 @@ void KThread::EndWait(ResultCode wait_result_) {
         }
 
         wait_queue->EndWait(this, wait_result_);
+
+        // Special case for dummy threads to wakeup if necessary.
+        IfDummyThreadEndWait();
     }
 }
 

src/core/hle/kernel/k_thread.h

@@ -15,7 +15,6 @@
 
 #include "common/common_types.h"
 #include "common/intrusive_red_black_tree.h"
-#include "common/spin_lock.h"
 #include "core/arm/arm_interface.h"
 #include "core/hle/kernel/k_affinity_mask.h"
 #include "core/hle/kernel/k_light_lock.h"
@@ -638,6 +637,14 @@ public:
         return condvar_key;
     }
 
+    // Dummy threads (used for HLE host threads) cannot wait based on the guest scheduler, and
+    // therefore will not block on guest kernel synchronization primitives. These methods handle
+    // blocking as needed.
+
+    void IfDummyThreadTryWait();
+    void IfDummyThreadBeginWait();
+    void IfDummyThreadEndWait();
+
 private:
     static constexpr size_t PriorityInheritanceCountMax = 10;
     union SyncObjectBuffer {
@@ -755,7 +762,7 @@ private:
     s8 priority_inheritance_count{};
     bool resource_limit_release_hint{};
     StackParameters stack_parameters{};
-    Common::SpinLock context_guard{};
+    KSpinLock context_guard{};
     KSpinLock dummy_wait_lock{};
 
     // For emulation

src/core/hle/kernel/physical_core.cpp

@@ -2,6 +2,7 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
+#include "common/spin_lock.h"
 #include "core/arm/cpu_interrupt_handler.h"
 #include "core/arm/dynarmic/arm_dynarmic_32.h"
 #include "core/arm/dynarmic/arm_dynarmic_64.h"
@@ -15,7 +16,7 @@ namespace Kernel {
 PhysicalCore::PhysicalCore(std::size_t core_index_, Core::System& system_, KScheduler& scheduler_,
                            Core::CPUInterrupts& interrupts_)
     : core_index{core_index_}, system{system_}, scheduler{scheduler_},
-      interrupts{interrupts_}, guard{std::make_unique<std::mutex>()} {
+      interrupts{interrupts_}, guard{std::make_unique<Common::SpinLock>()} {
 #ifdef ARCHITECTURE_x86_64
     // TODO(bunnei): Initialization relies on a core being available. We may later replace this with
     // a 32-bit instance of Dynarmic. This should be abstracted out to a CPU manager.

src/core/hle/kernel/physical_core.h

@@ -6,10 +6,13 @@
 
 #include <cstddef>
 #include <memory>
-#include <mutex>
 
 #include "core/arm/arm_interface.h"
 
+namespace Common {
+class SpinLock;
+}
+
 namespace Kernel {
 class KScheduler;
 } // namespace Kernel
@@ -88,7 +91,7 @@ private:
     Core::System& system;
     Kernel::KScheduler& scheduler;
     Core::CPUInterrupts& interrupts;
-    std::unique_ptr<std::mutex> guard;
+    std::unique_ptr<Common::SpinLock> guard;
     std::unique_ptr<Core::ARM_Interface> arm_interface;
 };
 

src/core/hle/service/service.h

@@ -9,6 +9,7 @@
 #include <string>
 #include <boost/container/flat_map.hpp>
 #include "common/common_types.h"
+#include "common/spin_lock.h"
 #include "core/hle/kernel/hle_ipc.h"
 
 ////////////////////////////////////////////////////////////////////////////////////////////////////
@@ -89,7 +90,7 @@ protected:
     using HandlerFnP = void (Self::*)(Kernel::HLERequestContext&);
 
     /// Used to gain exclusive access to the service members, e.g. from CoreTiming thread.
-    [[nodiscard]] std::scoped_lock<std::mutex> LockService() {
+    [[nodiscard]] std::scoped_lock<Common::SpinLock> LockService() {
         return std::scoped_lock{lock_service};
     }
 
@@ -134,7 +135,7 @@ private:
     boost::container::flat_map<u32, FunctionInfoBase> handlers_tipc;
 
     /// Used to gain exclusive access to the service members, e.g. from CoreTiming thread.
-    std::mutex lock_service;
+    Common::SpinLock lock_service;
 };
 
 /**