early-access version 1260

README.md

@@ -1,7 +1,7 @@
 yuzu emulator early access
 =============
 
-This is the source code for early-access 1259.
+This is the source code for early-access 1260.
 
 ## Legal Notice
 

src/audio_core/audio_renderer.cpp

@@ -11,6 +11,7 @@
 #include "audio_core/info_updater.h"
 #include "audio_core/voice_context.h"
 #include "common/logging/log.h"
+#include "core/hle/kernel/writable_event.h"
 #include "core/memory.h"
 #include "core/settings.h"
 
@@ -70,9 +71,10 @@ namespace {
 namespace AudioCore {
 AudioRenderer::AudioRenderer(Core::Timing::CoreTiming& core_timing, Core::Memory::Memory& memory_,
                              AudioCommon::AudioRendererParameter params,
-                             Stream::ReleaseCallback&& release_callback,
+                             std::shared_ptr<Kernel::WritableEvent> buffer_event_,
                              std::size_t instance_number)
-    : worker_params{params}, memory_pool_info(params.effect_count + params.voice_count * 4),
+    : worker_params{params}, buffer_event{buffer_event_},
+      memory_pool_info(params.effect_count + params.voice_count * 4),
       voice_context(params.voice_count), effect_context(params.effect_count), mix_context(),
       sink_context(params.sink_count), splitter_context(),
       voices(params.voice_count), memory{memory_},
@@ -83,9 +85,10 @@ AudioRenderer::AudioRenderer(Core::Timing::CoreTiming& core_timing, Core::Memory
                                 params.num_splitter_send_channels);
     mix_context.Initialize(behavior_info, params.submix_count + 1, params.effect_count);
     audio_out = std::make_unique<AudioCore::AudioOut>();
-    stream = audio_out->OpenStream(
+    stream =
-        core_timing, params.sample_rate, AudioCommon::STREAM_NUM_CHANNELS,
+        audio_out->OpenStream(core_timing, params.sample_rate, AudioCommon::STREAM_NUM_CHANNELS,
-        fmt::format("AudioRenderer-Instance{}", instance_number), std::move(release_callback));
+                              fmt::format("AudioRenderer-Instance{}", instance_number),
+                              [=]() { buffer_event_->Signal(); });
     audio_out->StartStream(stream);
 
     QueueMixedBuffer(0);

src/audio_core/audio_renderer.h

@@ -27,6 +27,10 @@ namespace Core::Timing {
 class CoreTiming;
 }
 
+namespace Kernel {
+class WritableEvent;
+}
+
 namespace Core::Memory {
 class Memory;
 }
@@ -40,7 +44,8 @@ class AudioRenderer {
 public:
     AudioRenderer(Core::Timing::CoreTiming& core_timing, Core::Memory::Memory& memory_,
                   AudioCommon::AudioRendererParameter params,
-                  Stream::ReleaseCallback&& release_callback, std::size_t instance_number);
+                  std::shared_ptr<Kernel::WritableEvent> buffer_event_,
+                  std::size_t instance_number);
     ~AudioRenderer();
 
     [[nodiscard]] ResultCode UpdateAudioRenderer(const std::vector<u8>& input_params,
@@ -56,6 +61,7 @@ private:
     BehaviorInfo behavior_info{};
 
     AudioCommon::AudioRendererParameter worker_params;
+    std::shared_ptr<Kernel::WritableEvent> buffer_event;
     std::vector<ServerMemoryPoolInfo> memory_pool_info;
     VoiceContext voice_context;
     EffectContext effect_context;

src/audio_core/stream.cpp

@@ -130,11 +130,7 @@ bool Stream::ContainsBuffer([[maybe_unused]] Buffer::Tag tag) const {
 std::vector<Buffer::Tag> Stream::GetTagsAndReleaseBuffers(std::size_t max_count) {
     std::vector<Buffer::Tag> tags;
     for (std::size_t count = 0; count < max_count && !released_buffers.empty(); ++count) {
-        if (released_buffers.front()) {
         tags.push_back(released_buffers.front()->GetTag());
-        } else {
-            ASSERT_MSG(false, "Invalid tag in released_buffers!");
-        }
         released_buffers.pop();
     }
     return tags;
@@ -144,11 +140,7 @@ std::vector<Buffer::Tag> Stream::GetTagsAndReleaseBuffers() {
     std::vector<Buffer::Tag> tags;
     tags.reserve(released_buffers.size());
     while (!released_buffers.empty()) {
-        if (released_buffers.front()) {
         tags.push_back(released_buffers.front()->GetTag());
-        } else {
-            ASSERT_MSG(false, "Invalid tag in released_buffers!");
-        }
         released_buffers.pop();
     }
     return tags;

src/core/CMakeLists.txt

@@ -201,8 +201,6 @@ add_library(core STATIC
     hle/kernel/server_port.h
     hle/kernel/server_session.cpp
     hle/kernel/server_session.h
-    hle/kernel/service_thread.cpp
-    hle/kernel/service_thread.h
     hle/kernel/session.cpp
     hle/kernel/session.h
     hle/kernel/shared_memory.cpp
@@ -501,6 +499,7 @@ add_library(core STATIC
     hle/service/sm/controller.h
     hle/service/sm/sm.cpp
     hle/service/sm/sm.h
+    hle/service/sockets/blocking_worker.h
     hle/service/sockets/bsd.cpp
     hle/service/sockets/bsd.h
     hle/service/sockets/ethc.cpp

src/core/core.cpp

@@ -159,7 +159,7 @@ struct System::Impl {
         device_memory = std::make_unique<Core::DeviceMemory>();
 
         is_multicore = Settings::values.use_multi_core.GetValue();
-        is_async_gpu = Settings::values.use_asynchronous_gpu_emulation.GetValue();
+        is_async_gpu = is_multicore || Settings::values.use_asynchronous_gpu_emulation.GetValue();
 
         kernel.SetMulticore(is_multicore);
         cpu_manager.SetMulticore(is_multicore);
@@ -307,6 +307,7 @@ struct System::Impl {
         service_manager.reset();
         cheat_engine.reset();
         telemetry_session.reset();
+        device_memory.reset();
 
         // Close all CPU/threading state
         cpu_manager.Shutdown();

src/core/hle/kernel/hle_ipc.cpp

@@ -46,6 +46,43 @@ void SessionRequestHandler::ClientDisconnected(
     boost::range::remove_erase(connected_sessions, server_session);
 }
 
+std::shared_ptr<WritableEvent> HLERequestContext::SleepClientThread(
+    const std::string& reason, u64 timeout, WakeupCallback&& callback,
+    std::shared_ptr<WritableEvent> writable_event) {
+    // Put the client thread to sleep until the wait event is signaled or the timeout expires.
+
+    if (!writable_event) {
+        // Create event if not provided
+        const auto pair = WritableEvent::CreateEventPair(kernel, "HLE Pause Event: " + reason);
+        writable_event = pair.writable;
+    }
+
+    Handle event_handle = InvalidHandle;
+    {
+        KScopedSchedulerLockAndSleep lock(kernel, event_handle, thread.get(), timeout);
+        thread->SetHLECallback(
+            [context = *this, callback](std::shared_ptr<Thread> thread) mutable -> bool {
+                ThreadWakeupReason reason = thread->GetSignalingResult() == RESULT_TIMEOUT
+                                                ? ThreadWakeupReason::Timeout
+                                                : ThreadWakeupReason::Signal;
+                callback(thread, context, reason);
+                context.WriteToOutgoingCommandBuffer(*thread);
+                return true;
+            });
+        const auto readable_event{writable_event->GetReadableEvent()};
+        writable_event->Clear();
+        thread->SetHLESyncObject(readable_event.get());
+        thread->SetStatus(ThreadStatus::WaitHLEEvent);
+        thread->SetSynchronizationResults(nullptr, RESULT_TIMEOUT);
+        readable_event->AddWaitingThread(thread);
+    }
+    thread->SetHLETimeEvent(event_handle);
+
+    is_thread_waiting = true;
+
+    return writable_event;
+}
+
 HLERequestContext::HLERequestContext(KernelCore& kernel, Core::Memory::Memory& memory,
                                      std::shared_ptr<ServerSession> server_session,
                                      std::shared_ptr<Thread> thread)

src/core/hle/kernel/hle_ipc.h

@@ -129,6 +129,23 @@ public:
     using WakeupCallback = std::function<void(
         std::shared_ptr<Thread> thread, HLERequestContext& context, ThreadWakeupReason reason)>;
 
+    /**
+     * Puts the specified guest thread to sleep until the returned event is signaled or until the
+     * specified timeout expires.
+     * @param reason Reason for pausing the thread, to be used for debugging purposes.
+     * @param timeout Timeout in nanoseconds after which the thread will be awoken and the callback
+     * invoked with a Timeout reason.
+     * @param callback Callback to be invoked when the thread is resumed. This callback must write
+     * the entire command response once again, regardless of the state of it before this function
+     * was called.
+     * @param writable_event Event to use to wake up the thread. If unspecified, an event will be
+     * created.
+     * @returns Event that when signaled will resume the thread and call the callback function.
+     */
+    std::shared_ptr<WritableEvent> SleepClientThread(
+        const std::string& reason, u64 timeout, WakeupCallback&& callback,
+        std::shared_ptr<WritableEvent> writable_event = nullptr);
+
     /// Populates this context with data from the requesting process/thread.
     ResultCode PopulateFromIncomingCommandBuffer(const HandleTable& handle_table,
                                                  u32_le* src_cmdbuf);

src/core/hle/kernel/kernel.cpp

@@ -8,7 +8,7 @@
 #include <functional>
 #include <memory>
 #include <thread>
-#include <unordered_set>
+#include <unordered_map>
 #include <utility>
 
 #include "common/assert.h"
@@ -35,7 +35,6 @@
 #include "core/hle/kernel/physical_core.h"
 #include "core/hle/kernel/process.h"
 #include "core/hle/kernel/resource_limit.h"
-#include "core/hle/kernel/service_thread.h"
 #include "core/hle/kernel/shared_memory.h"
 #include "core/hle/kernel/synchronization.h"
 #include "core/hle/kernel/thread.h"
@@ -58,8 +57,6 @@ struct KernelCore::Impl {
     }
 
     void Initialize(KernelCore& kernel) {
-        process_list.clear();
-
         RegisterHostThread();
 
         global_scheduler_context = std::make_unique<Kernel::GlobalSchedulerContext>(kernel);
@@ -92,6 +89,8 @@ struct KernelCore::Impl {
 
         cores.clear();
 
+        process_list.clear();
+
         current_process = nullptr;
 
         system_resource_limit = nullptr;
@@ -108,9 +107,6 @@ struct KernelCore::Impl {
         std::fill(register_host_thread_keys.begin(), register_host_thread_keys.end(),
                   std::thread::id{});
         std::fill(register_host_thread_values.begin(), register_host_thread_values.end(), 0);
-
-        // Ensures all service threads gracefully shutdown
-        service_threads.clear();
     }
 
     void InitializePhysicalCores() {
@@ -333,7 +329,7 @@ struct KernelCore::Impl {
     std::atomic<u32> registered_thread_ids{Core::Hardware::NUM_CPU_CORES};
 
     // Number of host threads is a relatively high number to avoid overflowing
-    static constexpr size_t NUM_REGISTRABLE_HOST_THREADS = 1024;
+    static constexpr size_t NUM_REGISTRABLE_HOST_THREADS = 64;
     std::atomic<size_t> num_host_threads{0};
     std::array<std::atomic<std::thread::id>, NUM_REGISTRABLE_HOST_THREADS>
         register_host_thread_keys{};
@@ -349,9 +345,6 @@ struct KernelCore::Impl {
     std::shared_ptr<Kernel::SharedMemory> irs_shared_mem;
     std::shared_ptr<Kernel::SharedMemory> time_shared_mem;
 
-    // Threads used for services
-    std::unordered_set<std::shared_ptr<Kernel::ServiceThread>> service_threads;
-
     std::array<std::shared_ptr<Thread>, Core::Hardware::NUM_CPU_CORES> suspend_threads{};
     std::array<Core::CPUInterruptHandler, Core::Hardware::NUM_CPU_CORES> interrupts{};
     std::array<std::unique_ptr<Kernel::KScheduler>, Core::Hardware::NUM_CPU_CORES> schedulers{};
@@ -646,16 +639,4 @@ void KernelCore::ExitSVCProfile() {
     MicroProfileLeave(MICROPROFILE_TOKEN(Kernel_SVC), impl->svc_ticks[core]);
 }
 
-std::weak_ptr<Kernel::ServiceThread> KernelCore::CreateServiceThread(const std::string& name) {
-    auto service_thread = std::make_shared<Kernel::ServiceThread>(*this, 1, name);
-    impl->service_threads.emplace(service_thread);
-    return service_thread;
-}
-
-void KernelCore::ReleaseServiceThread(std::weak_ptr<Kernel::ServiceThread> service_thread) {
-    if (auto strong_ptr = service_thread.lock()) {
-        impl->service_threads.erase(strong_ptr);
-    }
-}
-
 } // namespace Kernel

src/core/hle/kernel/kernel.h

@@ -42,7 +42,6 @@ class Process;
 class ResourceLimit;
 class KScheduler;
 class SharedMemory;
-class ServiceThread;
 class Synchronization;
 class Thread;
 class TimeManager;
@@ -228,22 +227,6 @@ public:
 
     void ExitSVCProfile();
 
-    /**
-     * Creates an HLE service thread, which are used to execute service routines asynchronously.
-     * While these are allocated per ServerSession, these need to be owned and managed outside of
-     * ServerSession to avoid a circular dependency.
-     * @param name String name for the ServerSession creating this thread, used for debug purposes.
-     * @returns The a weak pointer newly created service thread.
-     */
-    std::weak_ptr<Kernel::ServiceThread> CreateServiceThread(const std::string& name);
-
-    /**
-     * Releases a HLE service thread, instructing KernelCore to free it. This should be called when
-     * the ServerSession associated with the thread is destroyed.
-     * @param service_thread Service thread to release.
-     */
-    void ReleaseServiceThread(std::weak_ptr<Kernel::ServiceThread> service_thread);
-
 private:
     friend class Object;
     friend class Process;

src/core/hle/kernel/server_session.cpp

@@ -25,19 +25,19 @@
 namespace Kernel {
 
 ServerSession::ServerSession(KernelCore& kernel) : SynchronizationObject{kernel} {}
-
+ServerSession::~ServerSession() = default;
-ServerSession::~ServerSession() {
-    kernel.ReleaseServiceThread(service_thread);
-}
 
 ResultVal<std::shared_ptr<ServerSession>> ServerSession::Create(KernelCore& kernel,
                                                                 std::shared_ptr<Session> parent,
                                                                 std::string name) {
     std::shared_ptr<ServerSession> session{std::make_shared<ServerSession>(kernel)};
 
+    session->request_event =
+        Core::Timing::CreateEvent(name, [session](std::uintptr_t, std::chrono::nanoseconds) {
+            session->CompleteSyncRequest();
+        });
     session->name = std::move(name);
     session->parent = std::move(parent);
-    session->service_thread = kernel.CreateServiceThread(session->name);
 
     return MakeResult(std::move(session));
 }
@@ -142,16 +142,16 @@ ResultCode ServerSession::QueueSyncRequest(std::shared_ptr<Thread> thread,
         std::make_shared<HLERequestContext>(kernel, memory, SharedFrom(this), std::move(thread));
 
     context->PopulateFromIncomingCommandBuffer(kernel.CurrentProcess()->GetHandleTable(), cmd_buf);
-
+    request_queue.Push(std::move(context));
-    if (auto strong_ptr = service_thread.lock()) {
-        strong_ptr->QueueSyncRequest(*this, std::move(context));
-        return RESULT_SUCCESS;
-    }
 
     return RESULT_SUCCESS;
 }
 
-ResultCode ServerSession::CompleteSyncRequest(HLERequestContext& context) {
+ResultCode ServerSession::CompleteSyncRequest() {
+    ASSERT(!request_queue.Empty());
+
+    auto& context = *request_queue.Front();
+
     ResultCode result = RESULT_SUCCESS;
     // If the session has been converted to a domain, handle the domain request
     if (IsDomain() && context.HasDomainMessageHeader()) {
@@ -177,13 +177,18 @@ ResultCode ServerSession::CompleteSyncRequest(HLERequestContext& context) {
         }
     }
 
+    request_queue.Pop();
+
     return result;
 }
 
 ResultCode ServerSession::HandleSyncRequest(std::shared_ptr<Thread> thread,
                                             Core::Memory::Memory& memory,
                                             Core::Timing::CoreTiming& core_timing) {
-    return QueueSyncRequest(std::move(thread), memory);
+    const ResultCode result = QueueSyncRequest(std::move(thread), memory);
+    const auto delay = std::chrono::nanoseconds{kernel.IsMulticore() ? 0 : 20000};
+    core_timing.ScheduleEvent(delay, request_event, {});
+    return result;
 }
 
 } // namespace Kernel

src/core/hle/kernel/server_session.h

@@ -10,7 +10,6 @@
 #include <vector>
 
 #include "common/threadsafe_queue.h"
-#include "core/hle/kernel/service_thread.h"
 #include "core/hle/kernel/synchronization_object.h"
 #include "core/hle/result.h"
 
@@ -44,8 +43,6 @@ class Thread;
  * TLS buffer and control is transferred back to it.
  */
 class ServerSession final : public SynchronizationObject {
-    friend class ServiceThread;
-
 public:
     explicit ServerSession(KernelCore& kernel);
     ~ServerSession() override;
@@ -135,7 +132,7 @@ private:
     ResultCode QueueSyncRequest(std::shared_ptr<Thread> thread, Core::Memory::Memory& memory);
 
     /// Completes a sync request from the emulated application.
-    ResultCode CompleteSyncRequest(HLERequestContext& context);
+    ResultCode CompleteSyncRequest();
 
     /// Handles a SyncRequest to a domain, forwarding the request to the proper object or closing an
     /// object handle.
@@ -166,8 +163,11 @@ private:
     /// The name of this session (optional)
     std::string name;
 
-    /// Thread to dispatch service requests
+    /// Core timing event used to schedule the service request at some point in the future
-    std::weak_ptr<ServiceThread> service_thread;
+    std::shared_ptr<Core::Timing::EventType> request_event;
+
+    /// Queue of scheduled service requests
+    Common::MPSCQueue<std::shared_ptr<Kernel::HLERequestContext>> request_queue;
 };
 
 } // namespace Kernel

src/core/hle/service/audio/audout_u.cpp

@@ -70,10 +70,8 @@ public:
             Kernel::WritableEvent::CreateEventPair(system.Kernel(), "IAudioOutBufferReleased");
 
         stream = audio_core.OpenStream(system.CoreTiming(), audio_params.sample_rate,
-                                       audio_params.channel_count, std::move(unique_name), [this] {
+                                       audio_params.channel_count, std::move(unique_name),
-                                           const auto guard = LockService();
+                                       [this] { buffer_event.writable->Signal(); });
-                                           buffer_event.writable->Signal();
-                                       });
     }
 
 private:

src/core/hle/service/audio/audren_u.cpp

@@ -49,16 +49,16 @@ public:
 
         system_event =
             Kernel::WritableEvent::CreateEventPair(system.Kernel(), "IAudioRenderer:SystemEvent");
-        renderer = std::make_unique<AudioCore::AudioRenderer>(
+        renderer = std::make_unique<AudioCore::AudioRenderer>(system.CoreTiming(), system.Memory(),
-            system.CoreTiming(), system.Memory(), audren_params,
+                                                              audren_params, system_event.writable,
-            [this]() {
-                const auto guard = LockService();
-                system_event.writable->Signal();
-            },
                                                               instance_number);
     }
 
 private:
+    void UpdateAudioCallback() {
+        system_event.writable->Signal();
+    }
+
     void GetSampleRate(Kernel::HLERequestContext& ctx) {
         LOG_DEBUG(Service_Audio, "called");
 

src/core/hle/service/hid/hid.cpp

@@ -78,13 +78,11 @@ IAppletResource::IAppletResource(Core::System& system_)
     pad_update_event = Core::Timing::CreateEvent(
         "HID::UpdatePadCallback",
         [this](std::uintptr_t user_data, std::chrono::nanoseconds ns_late) {
-            const auto guard = LockService();
             UpdateControllers(user_data, ns_late);
         });
     motion_update_event = Core::Timing::CreateEvent(
         "HID::MotionPadCallback",
         [this](std::uintptr_t user_data, std::chrono::nanoseconds ns_late) {
-            const auto guard = LockService();
             UpdateMotion(user_data, ns_late);
         });
 

src/core/hle/service/nvdrv/devices/nvdevice.h

@@ -31,8 +31,8 @@ public:
      * @param output A buffer where the output data will be written to.
      * @returns The result code of the ioctl.
      */
-    virtual NvResult Ioctl1(Ioctl command, const std::vector<u8>& input,
+    virtual NvResult Ioctl1(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
-                            std::vector<u8>& output) = 0;
+                            IoctlCtrl& ctrl) = 0;
 
     /**
      * Handles an ioctl2 request.
@@ -43,7 +43,8 @@ public:
      * @returns The result code of the ioctl.
      */
     virtual NvResult Ioctl2(Ioctl command, const std::vector<u8>& input,
-                            const std::vector<u8>& inline_input, std::vector<u8>& output) = 0;
+                            const std::vector<u8>& inline_input, std::vector<u8>& output,
+                            IoctlCtrl& ctrl) = 0;
 
     /**
      * Handles an ioctl3 request.
@@ -54,7 +55,7 @@ public:
      * @returns The result code of the ioctl.
      */
     virtual NvResult Ioctl3(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
-                            std::vector<u8>& inline_output) = 0;
+                            std::vector<u8>& inline_output, IoctlCtrl& ctrl) = 0;
 
 protected:
     Core::System& system;

src/core/hle/service/nvdrv/devices/nvdisp_disp0.cpp

@@ -18,20 +18,21 @@ nvdisp_disp0::nvdisp_disp0(Core::System& system, std::shared_ptr<nvmap> nvmap_de
     : nvdevice(system), nvmap_dev(std::move(nvmap_dev)) {}
 nvdisp_disp0 ::~nvdisp_disp0() = default;
 
-NvResult nvdisp_disp0::Ioctl1(Ioctl command, const std::vector<u8>& input,
+NvResult nvdisp_disp0::Ioctl1(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
-                              std::vector<u8>& output) {
+                              IoctlCtrl& ctrl) {
     UNIMPLEMENTED_MSG("Unimplemented ioctl={:08X}", command.raw);
     return NvResult::NotImplemented;
 }
 
 NvResult nvdisp_disp0::Ioctl2(Ioctl command, const std::vector<u8>& input,
-                              const std::vector<u8>& inline_input, std::vector<u8>& output) {
+                              const std::vector<u8>& inline_input, std::vector<u8>& output,
+                              IoctlCtrl& ctrl) {
     UNIMPLEMENTED_MSG("Unimplemented ioctl={:08X}", command.raw);
     return NvResult::NotImplemented;
 }
 
 NvResult nvdisp_disp0::Ioctl3(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
-                              std::vector<u8>& inline_output) {
+                              std::vector<u8>& inline_output, IoctlCtrl& ctrl) {
     UNIMPLEMENTED_MSG("Unimplemented ioctl={:08X}", command.raw);
     return NvResult::NotImplemented;
 }

src/core/hle/service/nvdrv/devices/nvdisp_disp0.h

@@ -20,11 +20,13 @@ public:
     explicit nvdisp_disp0(Core::System& system, std::shared_ptr<nvmap> nvmap_dev);
     ~nvdisp_disp0() override;
 
-    NvResult Ioctl1(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) override;
+    NvResult Ioctl1(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+                    IoctlCtrl& ctrl) override;
     NvResult Ioctl2(Ioctl command, const std::vector<u8>& input,
-                    const std::vector<u8>& inline_input, std::vector<u8>& output) override;
+                    const std::vector<u8>& inline_input, std::vector<u8>& output,
+                    IoctlCtrl& ctrl) override;
     NvResult Ioctl3(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
-                    std::vector<u8>& inline_output) override;
+                    std::vector<u8>& inline_output, IoctlCtrl& ctrl) override;
 
     /// Performs a screen flip, drawing the buffer pointed to by the handle.
     void flip(u32 buffer_handle, u32 offset, u32 format, u32 width, u32 height, u32 stride,

src/core/hle/service/nvdrv/devices/nvhost_as_gpu.cpp

@@ -21,8 +21,8 @@ nvhost_as_gpu::nvhost_as_gpu(Core::System& system, std::shared_ptr<nvmap> nvmap_
     : nvdevice(system), nvmap_dev(std::move(nvmap_dev)) {}
 nvhost_as_gpu::~nvhost_as_gpu() = default;
 
-NvResult nvhost_as_gpu::Ioctl1(Ioctl command, const std::vector<u8>& input,
+NvResult nvhost_as_gpu::Ioctl1(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
-                               std::vector<u8>& output) {
+                               IoctlCtrl& ctrl) {
     switch (command.group) {
     case 'A':
         switch (command.cmd) {
@@ -55,13 +55,14 @@ NvResult nvhost_as_gpu::Ioctl1(Ioctl command, const std::vector<u8>& input,
 }
 
 NvResult nvhost_as_gpu::Ioctl2(Ioctl command, const std::vector<u8>& input,
-                               const std::vector<u8>& inline_input, std::vector<u8>& output) {
+                               const std::vector<u8>& inline_input, std::vector<u8>& output,
+                               IoctlCtrl& ctrl) {
     UNIMPLEMENTED_MSG("Unimplemented ioctl={:08X}", command.raw);
     return NvResult::NotImplemented;
 }
 
 NvResult nvhost_as_gpu::Ioctl3(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
-                               std::vector<u8>& inline_output) {
+                               std::vector<u8>& inline_output, IoctlCtrl& ctrl) {
     switch (command.group) {
     case 'A':
         switch (command.cmd) {

src/core/hle/service/nvdrv/devices/nvhost_as_gpu.h

@@ -30,11 +30,13 @@ public:
     explicit nvhost_as_gpu(Core::System& system, std::shared_ptr<nvmap> nvmap_dev);
     ~nvhost_as_gpu() override;
 
-    NvResult Ioctl1(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) override;
+    NvResult Ioctl1(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+                    IoctlCtrl& ctrl) override;
     NvResult Ioctl2(Ioctl command, const std::vector<u8>& input,
-                    const std::vector<u8>& inline_input, std::vector<u8>& output) override;
+                    const std::vector<u8>& inline_input, std::vector<u8>& output,
+                    IoctlCtrl& ctrl) override;
     NvResult Ioctl3(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
-                    std::vector<u8>& inline_output) override;
+                    std::vector<u8>& inline_output, IoctlCtrl& ctrl) override;
 
 private:
     class BufferMap final {

src/core/hle/service/nvdrv/devices/nvhost_ctrl.cpp

@@ -20,7 +20,8 @@ nvhost_ctrl::nvhost_ctrl(Core::System& system, EventInterface& events_interface,
     : nvdevice(system), events_interface{events_interface}, syncpoint_manager{syncpoint_manager} {}
 nvhost_ctrl::~nvhost_ctrl() = default;
 
-NvResult nvhost_ctrl::Ioctl1(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) {
+NvResult nvhost_ctrl::Ioctl1(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+                             IoctlCtrl& ctrl) {
     switch (command.group) {
     case 0x0:
         switch (command.cmd) {
@@ -29,9 +30,9 @@ NvResult nvhost_ctrl::Ioctl1(Ioctl command, const std::vector<u8>& input, std::v
         case 0x1c:
             return IocCtrlClearEventWait(input, output);
         case 0x1d:
-            return IocCtrlEventWait(input, output, false);
+            return IocCtrlEventWait(input, output, false, ctrl);
         case 0x1e:
-            return IocCtrlEventWait(input, output, true);
+            return IocCtrlEventWait(input, output, true, ctrl);
         case 0x1f:
             return IocCtrlEventRegister(input, output);
         case 0x20:
@@ -47,13 +48,14 @@ NvResult nvhost_ctrl::Ioctl1(Ioctl command, const std::vector<u8>& input, std::v
 }
 
 NvResult nvhost_ctrl::Ioctl2(Ioctl command, const std::vector<u8>& input,
-                             const std::vector<u8>& inline_input, std::vector<u8>& output) {
+                             const std::vector<u8>& inline_input, std::vector<u8>& output,
+                             IoctlCtrl& ctrl) {
     UNIMPLEMENTED_MSG("Unimplemented ioctl={:08X}", command.raw);
     return NvResult::NotImplemented;
 }
 
 NvResult nvhost_ctrl::Ioctl3(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
-                             std::vector<u8>& inline_outpu) {
+                             std::vector<u8>& inline_output, IoctlCtrl& ctrl) {
     UNIMPLEMENTED_MSG("Unimplemented ioctl={:08X}", command.raw);
     return NvResult::NotImplemented;
 }
@@ -67,7 +69,7 @@ NvResult nvhost_ctrl::NvOsGetConfigU32(const std::vector<u8>& input, std::vector
 }
 
 NvResult nvhost_ctrl::IocCtrlEventWait(const std::vector<u8>& input, std::vector<u8>& output,
-                                       bool is_async) {
+                                       bool is_async, IoctlCtrl& ctrl) {
     IocCtrlEventWaitParams params{};
     std::memcpy(&params, input.data(), sizeof(params));
     LOG_DEBUG(Service_NVDRV, "syncpt_id={}, threshold={}, timeout={}, is_async={}",
@@ -139,6 +141,12 @@ NvResult nvhost_ctrl::IocCtrlEventWait(const std::vector<u8>& input, std::vector
         params.value |= event_id;
         event.event.writable->Clear();
         gpu.RegisterSyncptInterrupt(params.syncpt_id, target_value);
+        if (!is_async && ctrl.fresh_call) {
+            ctrl.must_delay = true;
+            ctrl.timeout = params.timeout;
+            ctrl.event_id = event_id;
+            return NvResult::Timeout;
+        }
         std::memcpy(output.data(), &params, sizeof(params));
         return NvResult::Timeout;
     }

src/core/hle/service/nvdrv/devices/nvhost_ctrl.h

@@ -18,11 +18,13 @@ public:
                          SyncpointManager& syncpoint_manager);
     ~nvhost_ctrl() override;
 
-    NvResult Ioctl1(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) override;
+    NvResult Ioctl1(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+                    IoctlCtrl& ctrl) override;
     NvResult Ioctl2(Ioctl command, const std::vector<u8>& input,
-                    const std::vector<u8>& inline_input, std::vector<u8>& output) override;
+                    const std::vector<u8>& inline_input, std::vector<u8>& output,
+                    IoctlCtrl& ctrl) override;
     NvResult Ioctl3(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
-                    std::vector<u8>& inline_output) override;
+                    std::vector<u8>& inline_output, IoctlCtrl& ctrl) override;
 
 private:
     struct IocSyncptReadParams {
@@ -121,7 +123,8 @@ private:
     static_assert(sizeof(IocCtrlEventKill) == 8, "IocCtrlEventKill is incorrect size");
 
     NvResult NvOsGetConfigU32(const std::vector<u8>& input, std::vector<u8>& output);
-    NvResult IocCtrlEventWait(const std::vector<u8>& input, std::vector<u8>& output, bool is_async);
+    NvResult IocCtrlEventWait(const std::vector<u8>& input, std::vector<u8>& output, bool is_async,
+                              IoctlCtrl& ctrl);
     NvResult IocCtrlEventRegister(const std::vector<u8>& input, std::vector<u8>& output);
     NvResult IocCtrlEventUnregister(const std::vector<u8>& input, std::vector<u8>& output);
     NvResult IocCtrlClearEventWait(const std::vector<u8>& input, std::vector<u8>& output);

src/core/hle/service/nvdrv/devices/nvhost_ctrl_gpu.cpp

@@ -16,7 +16,7 @@ nvhost_ctrl_gpu::nvhost_ctrl_gpu(Core::System& system) : nvdevice(system) {}
 nvhost_ctrl_gpu::~nvhost_ctrl_gpu() = default;
 
 NvResult nvhost_ctrl_gpu::Ioctl1(Ioctl command, const std::vector<u8>& input,
-                                 std::vector<u8>& output) {
+                                 std::vector<u8>& output, IoctlCtrl& ctrl) {
     switch (command.group) {
     case 'G':
         switch (command.cmd) {
@@ -48,13 +48,15 @@ NvResult nvhost_ctrl_gpu::Ioctl1(Ioctl command, const std::vector<u8>& input,
 }
 
 NvResult nvhost_ctrl_gpu::Ioctl2(Ioctl command, const std::vector<u8>& input,
-                                 const std::vector<u8>& inline_input, std::vector<u8>& output) {
+                                 const std::vector<u8>& inline_input, std::vector<u8>& output,
+                                 IoctlCtrl& ctrl) {
     UNIMPLEMENTED_MSG("Unimplemented ioctl={:08X}", command.raw);
     return NvResult::NotImplemented;
 }
 
 NvResult nvhost_ctrl_gpu::Ioctl3(Ioctl command, const std::vector<u8>& input,
-                                 std::vector<u8>& output, std::vector<u8>& inline_output) {
+                                 std::vector<u8>& output, std::vector<u8>& inline_output,
+                                 IoctlCtrl& ctrl) {
     switch (command.group) {
     case 'G':
         switch (command.cmd) {

src/core/hle/service/nvdrv/devices/nvhost_ctrl_gpu.h

@@ -16,11 +16,13 @@ public:
     explicit nvhost_ctrl_gpu(Core::System& system);
     ~nvhost_ctrl_gpu() override;
 
-    NvResult Ioctl1(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) override;
+    NvResult Ioctl1(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+                    IoctlCtrl& ctrl) override;
     NvResult Ioctl2(Ioctl command, const std::vector<u8>& input,
-                    const std::vector<u8>& inline_input, std::vector<u8>& output) override;
+                    const std::vector<u8>& inline_input, std::vector<u8>& output,
+                    IoctlCtrl& ctrl) override;
     NvResult Ioctl3(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
-                    std::vector<u8>& inline_output) override;
+                    std::vector<u8>& inline_output, IoctlCtrl& ctrl) override;
 
 private:
     struct IoctlGpuCharacteristics {

src/core/hle/service/nvdrv/devices/nvhost_gpu.cpp

@@ -23,7 +23,8 @@ nvhost_gpu::nvhost_gpu(Core::System& system, std::shared_ptr<nvmap> nvmap_dev,
 
 nvhost_gpu::~nvhost_gpu() = default;
 
-NvResult nvhost_gpu::Ioctl1(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) {
+NvResult nvhost_gpu::Ioctl1(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+                            IoctlCtrl& ctrl) {
     switch (command.group) {
     case 0x0:
         switch (command.cmd) {
@@ -75,7 +76,8 @@ NvResult nvhost_gpu::Ioctl1(Ioctl command, const std::vector<u8>& input, std::ve
 };
 
 NvResult nvhost_gpu::Ioctl2(Ioctl command, const std::vector<u8>& input,
-                            const std::vector<u8>& inline_input, std::vector<u8>& output) {
+                            const std::vector<u8>& inline_input, std::vector<u8>& output,
+                            IoctlCtrl& ctrl) {
     switch (command.group) {
     case 'H':
         switch (command.cmd) {
@@ -89,7 +91,7 @@ NvResult nvhost_gpu::Ioctl2(Ioctl command, const std::vector<u8>& input,
 }
 
 NvResult nvhost_gpu::Ioctl3(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
-                            std::vector<u8>& inline_output) {
+                            std::vector<u8>& inline_output, IoctlCtrl& ctrl) {
     UNIMPLEMENTED_MSG("Unimplemented ioctl={:08X}", command.raw);
     return NvResult::NotImplemented;
 }

src/core/hle/service/nvdrv/devices/nvhost_gpu.h

@@ -26,11 +26,13 @@ public:
                         SyncpointManager& syncpoint_manager);
     ~nvhost_gpu() override;
 
-    NvResult Ioctl1(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) override;
+    NvResult Ioctl1(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+                    IoctlCtrl& ctrl) override;
     NvResult Ioctl2(Ioctl command, const std::vector<u8>& input,
-                    const std::vector<u8>& inline_input, std::vector<u8>& output) override;
+                    const std::vector<u8>& inline_input, std::vector<u8>& output,
+                    IoctlCtrl& ctrl) override;
     NvResult Ioctl3(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
-                    std::vector<u8>& inline_output) override;
+                    std::vector<u8>& inline_output, IoctlCtrl& ctrl) override;
 
 private:
     enum class CtxObjects : u32_le {

src/core/hle/service/nvdrv/devices/nvhost_nvdec.cpp

@@ -15,8 +15,8 @@ nvhost_nvdec::nvhost_nvdec(Core::System& system, std::shared_ptr<nvmap> nvmap_de
     : nvhost_nvdec_common(system, std::move(nvmap_dev)) {}
 nvhost_nvdec::~nvhost_nvdec() = default;
 
-NvResult nvhost_nvdec::Ioctl1(Ioctl command, const std::vector<u8>& input,
+NvResult nvhost_nvdec::Ioctl1(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
-                              std::vector<u8>& output) {
+                              IoctlCtrl& ctrl) {
     switch (command.group) {
     case 0x0:
         switch (command.cmd) {
@@ -59,13 +59,14 @@ NvResult nvhost_nvdec::Ioctl1(Ioctl command, const std::vector<u8>& input,
 }
 
 NvResult nvhost_nvdec::Ioctl2(Ioctl command, const std::vector<u8>& input,
-                              const std::vector<u8>& inline_input, std::vector<u8>& output) {
+                              const std::vector<u8>& inline_input, std::vector<u8>& output,
+                              IoctlCtrl& ctrl) {
     UNIMPLEMENTED_MSG("Unimplemented ioctl={:08X}", command.raw);
     return NvResult::NotImplemented;
 }
 
 NvResult nvhost_nvdec::Ioctl3(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
-                              std::vector<u8>& inline_output) {
+                              std::vector<u8>& inline_output, IoctlCtrl& ctrl) {
     UNIMPLEMENTED_MSG("Unimplemented ioctl={:08X}", command.raw);
     return NvResult::NotImplemented;
 }

src/core/hle/service/nvdrv/devices/nvhost_nvdec.h

@@ -14,11 +14,13 @@ public:
     explicit nvhost_nvdec(Core::System& system, std::shared_ptr<nvmap> nvmap_dev);
     ~nvhost_nvdec() override;
 
-    NvResult Ioctl1(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) override;
+    NvResult Ioctl1(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+                    IoctlCtrl& ctrl) override;
     NvResult Ioctl2(Ioctl command, const std::vector<u8>& input,
-                    const std::vector<u8>& inline_input, std::vector<u8>& output) override;
+                    const std::vector<u8>& inline_input, std::vector<u8>& output,
+                    IoctlCtrl& ctrl) override;
     NvResult Ioctl3(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
-                    std::vector<u8>& inline_output) override;
+                    std::vector<u8>& inline_output, IoctlCtrl& ctrl) override;
 };
 
 } // namespace Service::Nvidia::Devices

src/core/hle/service/nvdrv/devices/nvhost_nvjpg.cpp

@@ -13,8 +13,8 @@ namespace Service::Nvidia::Devices {
 nvhost_nvjpg::nvhost_nvjpg(Core::System& system) : nvdevice(system) {}
 nvhost_nvjpg::~nvhost_nvjpg() = default;
 
-NvResult nvhost_nvjpg::Ioctl1(Ioctl command, const std::vector<u8>& input,
+NvResult nvhost_nvjpg::Ioctl1(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
-                              std::vector<u8>& output) {
+                              IoctlCtrl& ctrl) {
     switch (command.group) {
     case 'H':
         switch (command.cmd) {
@@ -33,13 +33,14 @@ NvResult nvhost_nvjpg::Ioctl1(Ioctl command, const std::vector<u8>& input,
 }
 
 NvResult nvhost_nvjpg::Ioctl2(Ioctl command, const std::vector<u8>& input,
-                              const std::vector<u8>& inline_input, std::vector<u8>& output) {
+                              const std::vector<u8>& inline_input, std::vector<u8>& output,
+                              IoctlCtrl& ctrl) {
     UNIMPLEMENTED_MSG("Unimplemented ioctl={:08X}", command.raw);
     return NvResult::NotImplemented;
 }
 
 NvResult nvhost_nvjpg::Ioctl3(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
-                              std::vector<u8>& inline_output) {
+                              std::vector<u8>& inline_output, IoctlCtrl& ctrl) {
     UNIMPLEMENTED_MSG("Unimplemented ioctl={:08X}", command.raw);
     return NvResult::NotImplemented;
 }

src/core/hle/service/nvdrv/devices/nvhost_nvjpg.h

@@ -16,11 +16,13 @@ public:
     explicit nvhost_nvjpg(Core::System& system);
     ~nvhost_nvjpg() override;
 
-    NvResult Ioctl1(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) override;
+    NvResult Ioctl1(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+                    IoctlCtrl& ctrl) override;
     NvResult Ioctl2(Ioctl command, const std::vector<u8>& input,
-                    const std::vector<u8>& inline_input, std::vector<u8>& output) override;
+                    const std::vector<u8>& inline_input, std::vector<u8>& output,
+                    IoctlCtrl& ctrl) override;
     NvResult Ioctl3(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
-                    std::vector<u8>& inline_output) override;
+                    std::vector<u8>& inline_output, IoctlCtrl& ctrl) override;
 
 private:
     struct IoctlSetNvmapFD {

src/core/hle/service/nvdrv/devices/nvhost_vic.cpp

@@ -16,7 +16,8 @@ nvhost_vic::nvhost_vic(Core::System& system, std::shared_ptr<nvmap> nvmap_dev)
 }
 nvhost_vic::~nvhost_vic() = default;
 
-NvResult nvhost_vic::Ioctl1(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) {
+NvResult nvhost_vic::Ioctl1(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+                            IoctlCtrl& ctrl) {
     switch (command.group) {
     case 0x0:
         switch (command.cmd) {
@@ -51,13 +52,14 @@ NvResult nvhost_vic::Ioctl1(Ioctl command, const std::vector<u8>& input, std::ve
 }
 
 NvResult nvhost_vic::Ioctl2(Ioctl command, const std::vector<u8>& input,
-                            const std::vector<u8>& inline_input, std::vector<u8>& output) {
+                            const std::vector<u8>& inline_input, std::vector<u8>& output,
+                            IoctlCtrl& ctrl) {
     UNIMPLEMENTED_MSG("Unimplemented ioctl={:08X}", command.raw);
     return NvResult::NotImplemented;
 }
 
 NvResult nvhost_vic::Ioctl3(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
-                            std::vector<u8>& inline_output) {
+                            std::vector<u8>& inline_output, IoctlCtrl& ctrl) {
     UNIMPLEMENTED_MSG("Unimplemented ioctl={:08X}", command.raw);
     return NvResult::NotImplemented;
 }

src/core/hle/service/nvdrv/devices/nvhost_vic.h

@@ -14,10 +14,12 @@ public:
     explicit nvhost_vic(Core::System& system, std::shared_ptr<nvmap> nvmap_dev);
     ~nvhost_vic();
 
-    NvResult Ioctl1(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) override;
+    NvResult Ioctl1(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+                    IoctlCtrl& ctrl) override;
     NvResult Ioctl2(Ioctl command, const std::vector<u8>& input,
-                    const std::vector<u8>& inline_input, std::vector<u8>& output) override;
+                    const std::vector<u8>& inline_input, std::vector<u8>& output,
+                    IoctlCtrl& ctrl) override;
     NvResult Ioctl3(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
-                    std::vector<u8>& inline_output) override;
+                    std::vector<u8>& inline_output, IoctlCtrl& ctrl) override;
 };
 } // namespace Service::Nvidia::Devices

src/core/hle/service/nvdrv/devices/nvmap.cpp

@@ -19,7 +19,8 @@ nvmap::nvmap(Core::System& system) : nvdevice(system) {
 
 nvmap::~nvmap() = default;
 
-NvResult nvmap::Ioctl1(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) {
+NvResult nvmap::Ioctl1(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+                       IoctlCtrl& ctrl) {
     switch (command.group) {
     case 0x1:
         switch (command.cmd) {
@@ -48,13 +49,14 @@ NvResult nvmap::Ioctl1(Ioctl command, const std::vector<u8>& input, std::vector<
 }
 
 NvResult nvmap::Ioctl2(Ioctl command, const std::vector<u8>& input,
-                       const std::vector<u8>& inline_input, std::vector<u8>& output) {
+                       const std::vector<u8>& inline_input, std::vector<u8>& output,
+                       IoctlCtrl& ctrl) {
     UNIMPLEMENTED_MSG("Unimplemented ioctl={:08X}", command.raw);
     return NvResult::NotImplemented;
 }
 
 NvResult nvmap::Ioctl3(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
-                       std::vector<u8>& inline_output) {
+                       std::vector<u8>& inline_output, IoctlCtrl& ctrl) {
     UNIMPLEMENTED_MSG("Unimplemented ioctl={:08X}", command.raw);
     return NvResult::NotImplemented;
 }

src/core/hle/service/nvdrv/devices/nvmap.h

@@ -19,11 +19,13 @@ public:
     explicit nvmap(Core::System& system);
     ~nvmap() override;
 
-    NvResult Ioctl1(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) override;
+    NvResult Ioctl1(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
+                    IoctlCtrl& ctrl) override;
     NvResult Ioctl2(Ioctl command, const std::vector<u8>& input,
-                    const std::vector<u8>& inline_input, std::vector<u8>& output) override;
+                    const std::vector<u8>& inline_input, std::vector<u8>& output,
+                    IoctlCtrl& ctrl) override;
     NvResult Ioctl3(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output,
-                    std::vector<u8>& inline_output) override;
+                    std::vector<u8>& inline_output, IoctlCtrl& ctrl) override;
 
     /// Returns the allocated address of an nvmap object given its handle.
     VAddr GetObjectAddress(u32 handle) const;

src/core/hle/service/nvdrv/interface.cpp

@@ -61,10 +61,33 @@ void NVDRV::Ioctl1(Kernel::HLERequestContext& ctx) {
     std::vector<u8> output_buffer(ctx.GetWriteBufferSize(0));
     const auto input_buffer = ctx.ReadBuffer(0);
 
-    const auto nv_result = nvdrv->Ioctl1(fd, command, input_buffer, output_buffer);
+    IoctlCtrl ctrl{};
+
+    const auto nv_result = nvdrv->Ioctl1(fd, command, input_buffer, output_buffer, ctrl);
+    if (ctrl.must_delay) {
+        ctrl.fresh_call = false;
+        ctx.SleepClientThread(
+            "NVServices::DelayedResponse", ctrl.timeout,
+            [=, this](std::shared_ptr<Kernel::Thread> thread, Kernel::HLERequestContext& ctx_,
+                      Kernel::ThreadWakeupReason reason) {
+                IoctlCtrl ctrl2{ctrl};
+                std::vector<u8> tmp_output = output_buffer;
+                const auto nv_result2 = nvdrv->Ioctl1(fd, command, input_buffer, tmp_output, ctrl2);
+
+                if (command.is_out != 0) {
+                    ctx.WriteBuffer(tmp_output);
+                }
+
+                IPC::ResponseBuilder rb{ctx_, 3};
+                rb.Push(RESULT_SUCCESS);
+                rb.PushEnum(nv_result2);
+            },
+            nvdrv->GetEventWriteable(ctrl.event_id));
+    } else {
         if (command.is_out != 0) {
             ctx.WriteBuffer(output_buffer);
         }
+    }
 
     IPC::ResponseBuilder rb{ctx, 3};
     rb.Push(RESULT_SUCCESS);
@@ -87,8 +110,36 @@ void NVDRV::Ioctl2(Kernel::HLERequestContext& ctx) {
     const auto input_inlined_buffer = ctx.ReadBuffer(1);
     std::vector<u8> output_buffer(ctx.GetWriteBufferSize(0));
 
+    IoctlCtrl ctrl{};
+
     const auto nv_result =
-        nvdrv->Ioctl2(fd, command, input_buffer, input_inlined_buffer, output_buffer);
+        nvdrv->Ioctl2(fd, command, input_buffer, input_inlined_buffer, output_buffer, ctrl);
+    if (ctrl.must_delay) {
+        ctrl.fresh_call = false;
+        ctx.SleepClientThread(
+            "NVServices::DelayedResponse", ctrl.timeout,
+            [=, this](std::shared_ptr<Kernel::Thread> thread, Kernel::HLERequestContext& ctx_,
+                      Kernel::ThreadWakeupReason reason) {
+                IoctlCtrl ctrl2{ctrl};
+                std::vector<u8> tmp_output = output_buffer;
+                const auto nv_result2 = nvdrv->Ioctl2(fd, command, input_buffer,
+                                                      input_inlined_buffer, tmp_output, ctrl2);
+
+                if (command.is_out != 0) {
+                    ctx.WriteBuffer(tmp_output);
+                }
+
+                IPC::ResponseBuilder rb{ctx_, 3};
+                rb.Push(RESULT_SUCCESS);
+                rb.PushEnum(nv_result2);
+            },
+            nvdrv->GetEventWriteable(ctrl.event_id));
+    } else {
+        if (command.is_out != 0) {
+            ctx.WriteBuffer(output_buffer);
+        }
+    }
+
     if (command.is_out != 0) {
         ctx.WriteBuffer(output_buffer);
     }
@@ -114,12 +165,37 @@ void NVDRV::Ioctl3(Kernel::HLERequestContext& ctx) {
     std::vector<u8> output_buffer(ctx.GetWriteBufferSize(0));
     std::vector<u8> output_buffer_inline(ctx.GetWriteBufferSize(1));
 
+    IoctlCtrl ctrl{};
     const auto nv_result =
-        nvdrv->Ioctl3(fd, command, input_buffer, output_buffer, output_buffer_inline);
+        nvdrv->Ioctl3(fd, command, input_buffer, output_buffer, output_buffer_inline, ctrl);
+    if (ctrl.must_delay) {
+        ctrl.fresh_call = false;
+        ctx.SleepClientThread(
+            "NVServices::DelayedResponse", ctrl.timeout,
+            [=, this](std::shared_ptr<Kernel::Thread> thread, Kernel::HLERequestContext& ctx_,
+                      Kernel::ThreadWakeupReason reason) {
+                IoctlCtrl ctrl2{ctrl};
+                std::vector<u8> tmp_output = output_buffer;
+                std::vector<u8> tmp_output2 = output_buffer;
+                const auto nv_result2 =
+                    nvdrv->Ioctl3(fd, command, input_buffer, tmp_output, tmp_output2, ctrl2);
+
+                if (command.is_out != 0) {
+                    ctx.WriteBuffer(tmp_output, 0);
+                    ctx.WriteBuffer(tmp_output2, 1);
+                }
+
+                IPC::ResponseBuilder rb{ctx_, 3};
+                rb.Push(RESULT_SUCCESS);
+                rb.PushEnum(nv_result2);
+            },
+            nvdrv->GetEventWriteable(ctrl.event_id));
+    } else {
         if (command.is_out != 0) {
             ctx.WriteBuffer(output_buffer, 0);
             ctx.WriteBuffer(output_buffer_inline, 1);
         }
+    }
 
     IPC::ResponseBuilder rb{ctx, 3};
     rb.Push(RESULT_SUCCESS);

src/core/hle/service/nvdrv/nvdata.h

@@ -97,4 +97,15 @@ union Ioctl {
     BitField<31, 1, u32> is_out;
 };
 
+struct IoctlCtrl {
+    // First call done to the servioce for services that call itself again after a call.
+    bool fresh_call{true};
+    // Tells the Ioctl Wrapper that it must delay the IPC response and send the thread to sleep
+    bool must_delay{};
+    // Timeout for the delay
+    s64 timeout{};
+    // NV Event Id
+    s32 event_id{-1};
+};
+
 } // namespace Service::Nvidia

src/core/hle/service/nvdrv/nvdrv.cpp

@@ -91,7 +91,7 @@ DeviceFD Module::Open(const std::string& device_name) {
 }
 
 NvResult Module::Ioctl1(DeviceFD fd, Ioctl command, const std::vector<u8>& input,
-                        std::vector<u8>& output) {
+                        std::vector<u8>& output, IoctlCtrl& ctrl) {
     if (fd < 0) {
         LOG_ERROR(Service_NVDRV, "Invalid DeviceFD={}!", fd);
         return NvResult::InvalidState;
@@ -104,11 +104,12 @@ NvResult Module::Ioctl1(DeviceFD fd, Ioctl command, const std::vector<u8>& input
         return NvResult::NotImplemented;
     }
 
-    return itr->second->Ioctl1(command, input, output);
+    return itr->second->Ioctl1(command, input, output, ctrl);
 }
 
 NvResult Module::Ioctl2(DeviceFD fd, Ioctl command, const std::vector<u8>& input,
-                        const std::vector<u8>& inline_input, std::vector<u8>& output) {
+                        const std::vector<u8>& inline_input, std::vector<u8>& output,
+                        IoctlCtrl& ctrl) {
     if (fd < 0) {
         LOG_ERROR(Service_NVDRV, "Invalid DeviceFD={}!", fd);
         return NvResult::InvalidState;
@@ -121,11 +122,11 @@ NvResult Module::Ioctl2(DeviceFD fd, Ioctl command, const std::vector<u8>& input
         return NvResult::NotImplemented;
     }
 
-    return itr->second->Ioctl2(command, input, inline_input, output);
+    return itr->second->Ioctl2(command, input, inline_input, output, ctrl);
 }
 
 NvResult Module::Ioctl3(DeviceFD fd, Ioctl command, const std::vector<u8>& input,
-                        std::vector<u8>& output, std::vector<u8>& inline_output) {
+                        std::vector<u8>& output, std::vector<u8>& inline_output, IoctlCtrl& ctrl) {
     if (fd < 0) {
         LOG_ERROR(Service_NVDRV, "Invalid DeviceFD={}!", fd);
         return NvResult::InvalidState;
@@ -138,7 +139,7 @@ NvResult Module::Ioctl3(DeviceFD fd, Ioctl command, const std::vector<u8>& input
         return NvResult::NotImplemented;
     }
 
-    return itr->second->Ioctl3(command, input, output, inline_output);
+    return itr->second->Ioctl3(command, input, output, inline_output, ctrl);
 }
 
 NvResult Module::Close(DeviceFD fd) {

src/core/hle/service/nvdrv/nvdrv.h

@@ -119,13 +119,13 @@ public:
 
     /// Sends an ioctl command to the specified file descriptor.
     NvResult Ioctl1(DeviceFD fd, Ioctl command, const std::vector<u8>& input,
-                    std::vector<u8>& output);
+                    std::vector<u8>& output, IoctlCtrl& ctrl);
 
     NvResult Ioctl2(DeviceFD fd, Ioctl command, const std::vector<u8>& input,
-                    const std::vector<u8>& inline_input, std::vector<u8>& output);
+                    const std::vector<u8>& inline_input, std::vector<u8>& output, IoctlCtrl& ctrl);
 
     NvResult Ioctl3(DeviceFD fd, Ioctl command, const std::vector<u8>& input,
-                    std::vector<u8>& output, std::vector<u8>& inline_output);
+                    std::vector<u8>& output, std::vector<u8>& inline_output, IoctlCtrl& ctrl);
 
     /// Closes a device file descriptor and returns operation success.
     NvResult Close(DeviceFD fd);

src/core/hle/service/nvflinger/buffer_queue.cpp

@@ -25,12 +25,7 @@ void BufferQueue::SetPreallocatedBuffer(u32 slot, const IGBPBuffer& igbp_buffer)
     ASSERT(slot < buffer_slots);
     LOG_WARNING(Service, "Adding graphics buffer {}", slot);
 
-    {
-        std::unique_lock lock{queue_mutex};
     free_buffers.push_back(slot);
-    }
-    condition.notify_one();
-
     buffers[slot] = {
         .slot = slot,
         .status = Buffer::Status::Free,
@@ -46,20 +41,10 @@ void BufferQueue::SetPreallocatedBuffer(u32 slot, const IGBPBuffer& igbp_buffer)
 
 std::optional<std::pair<u32, Service::Nvidia::MultiFence*>> BufferQueue::DequeueBuffer(u32 width,
                                                                                        u32 height) {
-    // Wait for first request before trying to dequeue
-    {
-        std::unique_lock lock{queue_mutex};
-        condition.wait(lock, [this] { return !free_buffers.empty() || !is_connect; });
-    }
 
-    if (!is_connect) {
+    if (free_buffers.empty()) {
-        // Buffer was disconnected while the thread was blocked, this is most likely due to
-        // emulation being stopped
         return std::nullopt;
     }
-
-    std::unique_lock lock{queue_mutex};
-
     auto f_itr = free_buffers.begin();
     auto slot = buffers.size();
 
@@ -112,11 +97,7 @@ void BufferQueue::CancelBuffer(u32 slot, const Service::Nvidia::MultiFence& mult
     buffers[slot].multi_fence = multi_fence;
     buffers[slot].swap_interval = 0;
 
-    {
-        std::unique_lock lock{queue_mutex};
     free_buffers.push_back(slot);
-    }
-    condition.notify_one();
 
     buffer_wait_event.writable->Signal();
 }
@@ -146,28 +127,15 @@ void BufferQueue::ReleaseBuffer(u32 slot) {
     ASSERT(buffers[slot].slot == slot);
 
     buffers[slot].status = Buffer::Status::Free;
-    {
-        std::unique_lock lock{queue_mutex};
     free_buffers.push_back(slot);
-    }
-    condition.notify_one();
 
     buffer_wait_event.writable->Signal();
 }
 
-void BufferQueue::Connect() {
-    queue_sequence.clear();
-    id = 1;
-    layer_id = 1;
-    is_connect = true;
-}
-
 void BufferQueue::Disconnect() {
     buffers.fill({});
     queue_sequence.clear();
     buffer_wait_event.writable->Signal();
-    is_connect = false;
-    condition.notify_one();
 }
 
 u32 BufferQueue::Query(QueryType type) {

src/core/hle/service/nvflinger/buffer_queue.h

@@ -4,9 +4,7 @@
 
 #pragma once
 
-#include <condition_variable>
 #include <list>
-#include <mutex>
 #include <optional>
 #include <vector>
 
@@ -101,7 +99,6 @@ public:
     void CancelBuffer(u32 slot, const Service::Nvidia::MultiFence& multi_fence);
     std::optional<std::reference_wrapper<const Buffer>> AcquireBuffer();
     void ReleaseBuffer(u32 slot);
-    void Connect();
     void Disconnect();
     u32 Query(QueryType type);
 
@@ -109,28 +106,18 @@ public:
         return id;
     }
 
-    bool IsConnected() const {
-        return is_connect;
-    }
-
     std::shared_ptr<Kernel::WritableEvent> GetWritableBufferWaitEvent() const;
 
     std::shared_ptr<Kernel::ReadableEvent> GetBufferWaitEvent() const;
 
 private:
-    BufferQueue(const BufferQueue&) = delete;
+    u32 id;
-
+    u64 layer_id;
-    u32 id{};
-    u64 layer_id{};
-    std::atomic_bool is_connect{};
 
     std::list<u32> free_buffers;
     std::array<Buffer, buffer_slots> buffers;
     std::list<u32> queue_sequence;
     Kernel::EventPair buffer_wait_event;
-
-    std::mutex queue_mutex;
-    std::condition_variable condition;
 };
 
 } // namespace Service::NVFlinger

src/core/hle/service/nvflinger/nvflinger.cpp

@@ -88,10 +88,6 @@ NVFlinger::NVFlinger(Core::System& system) : system(system) {
 }
 
 NVFlinger::~NVFlinger() {
-    for (auto& buffer_queue : buffer_queues) {
-        buffer_queue->Disconnect();
-    }
-
     if (system.IsMulticore()) {
         is_running = false;
         wait_event->Set();
@@ -108,8 +104,6 @@ void NVFlinger::SetNVDrvInstance(std::shared_ptr<Nvidia::Module> instance) {
 }
 
 std::optional<u64> NVFlinger::OpenDisplay(std::string_view name) {
-    const auto guard = Lock();
-
     LOG_DEBUG(Service, "Opening \"{}\" display", name);
 
     // TODO(Subv): Currently we only support the Default display.
@@ -127,7 +121,6 @@ std::optional<u64> NVFlinger::OpenDisplay(std::string_view name) {
 }
 
 std::optional<u64> NVFlinger::CreateLayer(u64 display_id) {
-    const auto guard = Lock();
     auto* const display = FindDisplay(display_id);
 
     if (display == nullptr) {
@@ -136,22 +129,18 @@ std::optional<u64> NVFlinger::CreateLayer(u64 display_id) {
 
     const u64 layer_id = next_layer_id++;
     const u32 buffer_queue_id = next_buffer_queue_id++;
-    buffer_queues.emplace_back(
+    buffer_queues.emplace_back(system.Kernel(), buffer_queue_id, layer_id);
-        std::make_unique<BufferQueue>(system.Kernel(), buffer_queue_id, layer_id));
+    display->CreateLayer(layer_id, buffer_queues.back());
-    display->CreateLayer(layer_id, *buffer_queues.back());
     return layer_id;
 }
 
 void NVFlinger::CloseLayer(u64 layer_id) {
-    const auto guard = Lock();
-
     for (auto& display : displays) {
         display.CloseLayer(layer_id);
     }
 }
 
 std::optional<u32> NVFlinger::FindBufferQueueId(u64 display_id, u64 layer_id) const {
-    const auto guard = Lock();
     const auto* const layer = FindLayer(display_id, layer_id);
 
     if (layer == nullptr) {
@@ -162,7 +151,6 @@ std::optional<u32> NVFlinger::FindBufferQueueId(u64 display_id, u64 layer_id) co
 }
 
 std::shared_ptr<Kernel::ReadableEvent> NVFlinger::FindVsyncEvent(u64 display_id) const {
-    const auto guard = Lock();
     auto* const display = FindDisplay(display_id);
 
     if (display == nullptr) {
@@ -172,16 +160,20 @@ std::shared_ptr<Kernel::ReadableEvent> NVFlinger::FindVsyncEvent(u64 display_id)
     return display->GetVSyncEvent();
 }
 
-BufferQueue* NVFlinger::FindBufferQueue(u32 id) {
+BufferQueue& NVFlinger::FindBufferQueue(u32 id) {
-    const auto guard = Lock();
     const auto itr = std::find_if(buffer_queues.begin(), buffer_queues.end(),
-                                  [id](const auto& queue) { return queue->GetId() == id; });
+                                  [id](const auto& queue) { return queue.GetId() == id; });
 
-    if (itr == buffer_queues.end()) {
+    ASSERT(itr != buffer_queues.end());
-        return nullptr;
+    return *itr;
 }
 
-    return itr->get();
+const BufferQueue& NVFlinger::FindBufferQueue(u32 id) const {
+    const auto itr = std::find_if(buffer_queues.begin(), buffer_queues.end(),
+                                  [id](const auto& queue) { return queue.GetId() == id; });
+
+    ASSERT(itr != buffer_queues.end());
+    return *itr;
 }
 
 VI::Display* NVFlinger::FindDisplay(u64 display_id) {

src/core/hle/service/nvflinger/nvflinger.h

@@ -75,7 +75,10 @@ public:
     [[nodiscard]] std::shared_ptr<Kernel::ReadableEvent> FindVsyncEvent(u64 display_id) const;
 
     /// Obtains a buffer queue identified by the ID.
-    [[nodiscard]] BufferQueue* FindBufferQueue(u32 id);
+    [[nodiscard]] BufferQueue& FindBufferQueue(u32 id);
+
+    /// Obtains a buffer queue identified by the ID.
+    [[nodiscard]] const BufferQueue& FindBufferQueue(u32 id) const;
 
     /// Performs a composition request to the emulated nvidia GPU and triggers the vsync events when
     /// finished.
@@ -83,11 +86,11 @@ public:
 
     [[nodiscard]] s64 GetNextTicks() const;
 
-private:
     [[nodiscard]] std::unique_lock<std::mutex> Lock() const {
         return std::unique_lock{*guard};
     }
 
+private:
     /// Finds the display identified by the specified ID.
     [[nodiscard]] VI::Display* FindDisplay(u64 display_id);
 
@@ -107,7 +110,7 @@ private:
     std::shared_ptr<Nvidia::Module> nvdrv;
 
     std::vector<VI::Display> displays;
-    std::vector<std::unique_ptr<BufferQueue>> buffer_queues;
+    std::vector<BufferQueue> buffer_queues;
 
     /// Id to use for the next layer that is created, this counter is shared among all displays.
     u64 next_layer_id = 1;

src/core/hle/service/service.cpp

@@ -95,14 +95,9 @@ ServiceFrameworkBase::ServiceFrameworkBase(Core::System& system_, const char* se
     : system{system_}, service_name{service_name_}, max_sessions{max_sessions_},
       handler_invoker{handler_invoker_} {}
 
-ServiceFrameworkBase::~ServiceFrameworkBase() {
+ServiceFrameworkBase::~ServiceFrameworkBase() = default;
-    // Wait for other threads to release access before destroying
-    const auto guard = LockService();
-}
 
 void ServiceFrameworkBase::InstallAsService(SM::ServiceManager& service_manager) {
-    const auto guard = LockService();
-
     ASSERT(!port_installed);
 
     auto port = service_manager.RegisterService(service_name, max_sessions).Unwrap();
@@ -111,8 +106,6 @@ void ServiceFrameworkBase::InstallAsService(SM::ServiceManager& service_manager)
 }
 
 void ServiceFrameworkBase::InstallAsNamedPort(Kernel::KernelCore& kernel) {
-    const auto guard = LockService();
-
     ASSERT(!port_installed);
 
     auto [server_port, client_port] =
@@ -122,6 +115,17 @@ void ServiceFrameworkBase::InstallAsNamedPort(Kernel::KernelCore& kernel) {
     port_installed = true;
 }
 
+std::shared_ptr<Kernel::ClientPort> ServiceFrameworkBase::CreatePort(Kernel::KernelCore& kernel) {
+    ASSERT(!port_installed);
+
+    auto [server_port, client_port] =
+        Kernel::ServerPort::CreatePortPair(kernel, max_sessions, service_name);
+    auto port = MakeResult(std::move(server_port)).Unwrap();
+    port->SetHleHandler(shared_from_this());
+    port_installed = true;
+    return client_port;
+}
+
 void ServiceFrameworkBase::RegisterHandlersBase(const FunctionInfoBase* functions, std::size_t n) {
     handlers.reserve(handlers.size() + n);
     for (std::size_t i = 0; i < n; ++i) {
@@ -160,8 +164,6 @@ void ServiceFrameworkBase::InvokeRequest(Kernel::HLERequestContext& ctx) {
 }
 
 ResultCode ServiceFrameworkBase::HandleSyncRequest(Kernel::HLERequestContext& context) {
-    const auto guard = LockService();
-
     switch (context.GetCommandType()) {
     case IPC::CommandType::Close: {
         IPC::ResponseBuilder rb{context, 2};
@@ -182,11 +184,7 @@ ResultCode ServiceFrameworkBase::HandleSyncRequest(Kernel::HLERequestContext& co
         UNIMPLEMENTED_MSG("command_type={}", context.GetCommandType());
     }
 
-    // If emulation was shutdown, we are closing service threads, do not write the response back to
-    // memory that may be shutting down as well.
-    if (system.IsPoweredOn()) {
     context.WriteToOutgoingCommandBuffer(context.GetThread());
-    }
 
     return RESULT_SUCCESS;
 }

src/core/hle/service/service.h

@@ -5,11 +5,9 @@
 #pragma once
 
 #include <cstddef>
-#include <mutex>
 #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"
 #include "core/hle/kernel/object.h"
 
@@ -70,9 +68,11 @@ public:
     void InstallAsService(SM::ServiceManager& service_manager);
     /// Creates a port pair and registers it on the kernel's global port registry.
     void InstallAsNamedPort(Kernel::KernelCore& kernel);
-    /// Invokes a service request routine.
+    /// Creates and returns an unregistered port for the service.
+    std::shared_ptr<Kernel::ClientPort> CreatePort(Kernel::KernelCore& kernel);
+
     void InvokeRequest(Kernel::HLERequestContext& ctx);
-    /// Handles a synchronization request for the service.
+
     ResultCode HandleSyncRequest(Kernel::HLERequestContext& context) override;
 
 protected:
@@ -80,11 +80,6 @@ protected:
     template <typename Self>
     using HandlerFnP = void (Self::*)(Kernel::HLERequestContext&);
 
-    /// Used to gain exclusive access to the service members, e.g. from CoreTiming thread.
-    [[nodiscard]] std::scoped_lock<Common::SpinLock> LockService() {
-        return std::scoped_lock{lock_service};
-    }
-
     /// System context that the service operates under.
     Core::System& system;
 
@@ -120,9 +115,6 @@ private:
     /// Function used to safely up-cast pointers to the derived class before invoking a handler.
     InvokerFn* handler_invoker;
     boost::container::flat_map<u32, FunctionInfoBase> handlers;
-
-    /// Used to gain exclusive access to the service members, e.g. from CoreTiming thread.
-    Common::SpinLock lock_service;
 };
 
 /**

src/core/hle/service/sockets/blocking_worker.h

@@ -0,0 +1,161 @@
+// 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>
+#include <string>
+#include <string_view>
+#include <thread>
+#include <variant>
+#include <vector>
+
+#include <fmt/format.h>
+
+#include "common/assert.h"
+#include "common/microprofile.h"
+#include "common/thread.h"
+#include "core/core.h"
+#include "core/hle/kernel/hle_ipc.h"
+#include "core/hle/kernel/kernel.h"
+#include "core/hle/kernel/thread.h"
+#include "core/hle/kernel/writable_event.h"
+
+namespace Service::Sockets {
+
+/**
+ * Worker abstraction to execute blocking calls on host without blocking the guest thread
+ *
+ * @tparam Service  Service where the work is executed
+ * @tparam Types Types of work to execute
+ */
+template <class Service, class... Types>
+class BlockingWorker {
+    using This = BlockingWorker<Service, Types...>;
+    using WorkVariant = std::variant<std::monostate, Types...>;
+
+public:
+    /// Create a new worker
+    static std::unique_ptr<This> Create(Core::System& system, Service* service,
+                                        std::string_view name) {
+        return std::unique_ptr<This>(new This(system, service, name));
+    }
+
+    ~BlockingWorker() {
+        while (!is_available.load(std::memory_order_relaxed)) {
+            // Busy wait until work is finished
+            std::this_thread::yield();
+        }
+        // Monostate means to exit the thread
+        work = std::monostate{};
+        work_event.Set();
+        thread.join();
+    }
+
+    /**
+     * Try to capture the worker to send work after a success
+     * @returns True when the worker has been successfully captured
+     */
+    bool TryCapture() {
+        bool expected = true;
+        return is_available.compare_exchange_weak(expected, false, std::memory_order_relaxed,
+                                                  std::memory_order_relaxed);
+    }
+
+    /**
+     * Send work to this worker abstraction
+     * @see TryCapture must be called before attempting to call this function
+     */
+    template <class Work>
+    void SendWork(Work new_work) {
+        ASSERT_MSG(!is_available, "Trying to send work on a worker that's not captured");
+        work = std::move(new_work);
+        work_event.Set();
+    }
+
+    /// Generate a callback for @see SleepClientThread
+    template <class Work>
+    auto Callback() {
+        return [this](std::shared_ptr<Kernel::Thread>, Kernel::HLERequestContext& ctx,
+                      Kernel::ThreadWakeupReason reason) {
+            ASSERT(reason == Kernel::ThreadWakeupReason::Signal);
+            std::get<Work>(work).Response(ctx);
+            is_available.store(true);
+        };
+    }
+
+    /// Get kernel event that will be signalled by the worker when the host operation finishes
+    std::shared_ptr<Kernel::WritableEvent> KernelEvent() const {
+        return kernel_event;
+    }
+
+private:
+    explicit BlockingWorker(Core::System& system, Service* service, std::string_view name) {
+        auto pair = Kernel::WritableEvent::CreateEventPair(system.Kernel(), std::string(name));
+        kernel_event = std::move(pair.writable);
+        thread = std::thread([this, &system, service, name] { Run(system, service, name); });
+    }
+
+    void Run(Core::System& system, Service* service, std::string_view name) {
+        system.RegisterHostThread();
+
+        const std::string thread_name = fmt::format("yuzu:{}", name);
+        MicroProfileOnThreadCreate(thread_name.c_str());
+        Common::SetCurrentThreadName(thread_name.c_str());
+
+        bool keep_running = true;
+        while (keep_running) {
+            work_event.Wait();
+
+            const auto visit_fn = [service, &keep_running]<typename T>(T&& w) {
+                if constexpr (std::is_same_v<std::decay_t<T>, std::monostate>) {
+                    keep_running = false;
+                } else {
+                    w.Execute(service);
+                }
+            };
+            std::visit(visit_fn, work);
+
+            kernel_event->Signal();
+        }
+    }
+
+    std::thread thread;
+    WorkVariant work;
+    Common::Event work_event;
+    std::shared_ptr<Kernel::WritableEvent> kernel_event;
+    std::atomic_bool is_available{true};
+};
+
+template <class Service, class... Types>
+class BlockingWorkerPool {
+    using Worker = BlockingWorker<Service, Types...>;
+
+public:
+    explicit BlockingWorkerPool(Core::System& system_, Service* service_)
+        : system{system_}, service{service_} {}
+
+    /// Returns a captured worker thread, creating new ones if necessary
+    Worker* CaptureWorker() {
+        for (auto& worker : workers) {
+            if (worker->TryCapture()) {
+                return worker.get();
+            }
+        }
+        auto new_worker = Worker::Create(system, service, fmt::format("BSD:{}", workers.size()));
+        [[maybe_unused]] const bool success = new_worker->TryCapture();
+        ASSERT(success);
+
+        return workers.emplace_back(std::move(new_worker)).get();
+    }
+
+private:
+    Core::System& system;
+    Service* const service;
+
+    std::vector<std::unique_ptr<Worker>> workers;
+};
+
+} // namespace Service::Sockets

src/core/hle/service/sockets/bsd.cpp

@@ -178,7 +178,8 @@ void BSD::Poll(Kernel::HLERequestContext& ctx) {
 
     LOG_DEBUG(Service, "called. nfds={} timeout={}", nfds, timeout);
 
-    ExecuteWork(ctx, PollWork{
+    ExecuteWork(ctx, "BSD:Poll", timeout != 0,
+                PollWork{
                     .nfds = nfds,
                     .timeout = timeout,
                     .read_buffer = ctx.ReadBuffer(),
@@ -192,7 +193,8 @@ void BSD::Accept(Kernel::HLERequestContext& ctx) {
 
     LOG_DEBUG(Service, "called. fd={}", fd);
 
-    ExecuteWork(ctx, AcceptWork{
+    ExecuteWork(ctx, "BSD:Accept", IsBlockingSocket(fd),
+                AcceptWork{
                     .fd = fd,
                     .write_buffer = std::vector<u8>(ctx.GetWriteBufferSize()),
                 });
@@ -213,7 +215,8 @@ void BSD::Connect(Kernel::HLERequestContext& ctx) {
 
     LOG_DEBUG(Service, "called. fd={} addrlen={}", fd, ctx.GetReadBufferSize());
 
-    ExecuteWork(ctx, ConnectWork{
+    ExecuteWork(ctx, "BSD:Connect", IsBlockingSocket(fd),
+                ConnectWork{
                     .fd = fd,
                     .addr = ctx.ReadBuffer(),
                 });
@@ -324,7 +327,8 @@ void BSD::Recv(Kernel::HLERequestContext& ctx) {
 
     LOG_DEBUG(Service, "called. fd={} flags=0x{:x} len={}", fd, flags, ctx.GetWriteBufferSize());
 
-    ExecuteWork(ctx, RecvWork{
+    ExecuteWork(ctx, "BSD:Recv", IsBlockingSocket(fd),
+                RecvWork{
                     .fd = fd,
                     .flags = flags,
                     .message = std::vector<u8>(ctx.GetWriteBufferSize()),
@@ -340,7 +344,8 @@ void BSD::RecvFrom(Kernel::HLERequestContext& ctx) {
     LOG_DEBUG(Service, "called. fd={} flags=0x{:x} len={} addrlen={}", fd, flags,
               ctx.GetWriteBufferSize(0), ctx.GetWriteBufferSize(1));
 
-    ExecuteWork(ctx, RecvFromWork{
+    ExecuteWork(ctx, "BSD:RecvFrom", IsBlockingSocket(fd),
+                RecvFromWork{
                     .fd = fd,
                     .flags = flags,
                     .message = std::vector<u8>(ctx.GetWriteBufferSize(0)),
@@ -356,7 +361,8 @@ void BSD::Send(Kernel::HLERequestContext& ctx) {
 
     LOG_DEBUG(Service, "called. fd={} flags=0x{:x} len={}", fd, flags, ctx.GetReadBufferSize());
 
-    ExecuteWork(ctx, SendWork{
+    ExecuteWork(ctx, "BSD:Send", IsBlockingSocket(fd),
+                SendWork{
                     .fd = fd,
                     .flags = flags,
                     .message = ctx.ReadBuffer(),
@@ -371,7 +377,8 @@ void BSD::SendTo(Kernel::HLERequestContext& ctx) {
     LOG_DEBUG(Service, "called. fd={} flags=0x{} len={} addrlen={}", fd, flags,
               ctx.GetReadBufferSize(0), ctx.GetReadBufferSize(1));
 
-    ExecuteWork(ctx, SendToWork{
+    ExecuteWork(ctx, "BSD:SendTo", IsBlockingSocket(fd),
+                SendToWork{
                     .fd = fd,
                     .flags = flags,
                     .message = ctx.ReadBuffer(0),
@@ -385,7 +392,8 @@ void BSD::Write(Kernel::HLERequestContext& ctx) {
 
     LOG_DEBUG(Service, "called. fd={} len={}", fd, ctx.GetReadBufferSize());
 
-    ExecuteWork(ctx, SendWork{
+    ExecuteWork(ctx, "BSD:Write", IsBlockingSocket(fd),
+                SendWork{
                     .fd = fd,
                     .flags = 0,
                     .message = ctx.ReadBuffer(),
@@ -402,9 +410,24 @@ void BSD::Close(Kernel::HLERequestContext& ctx) {
 }
 
 template <typename Work>
-void BSD::ExecuteWork(Kernel::HLERequestContext& ctx, Work work) {
+void BSD::ExecuteWork(Kernel::HLERequestContext& ctx, std::string_view sleep_reason,
+                      bool is_blocking, Work work) {
+    if (!is_blocking) {
         work.Execute(this);
         work.Response(ctx);
+        return;
+    }
+
+    // Signal a dummy response to make IPC validation happy
+    // This will be overwritten by the SleepClientThread callback
+    work.Response(ctx);
+
+    auto worker = worker_pool.CaptureWorker();
+
+    ctx.SleepClientThread(std::string(sleep_reason), std::numeric_limits<u64>::max(),
+                          worker->Callback<Work>(), worker->KernelEvent());
+
+    worker->SendWork(std::move(work));
 }
 
 std::pair<s32, Errno> BSD::SocketImpl(Domain domain, Type type, Protocol protocol) {
@@ -784,6 +807,18 @@ bool BSD::IsFileDescriptorValid(s32 fd) const noexcept {
     return true;
 }
 
+bool BSD::IsBlockingSocket(s32 fd) const noexcept {
+    // Inform invalid sockets as non-blocking
+    // This way we avoid using a worker thread as it will fail without blocking host
+    if (fd > static_cast<s32>(MAX_FD) || fd < 0) {
+        return false;
+    }
+    if (!file_descriptors[fd]) {
+        return false;
+    }
+    return (file_descriptors[fd]->flags & FLAG_O_NONBLOCK) != 0;
+}
+
 void BSD::BuildErrnoResponse(Kernel::HLERequestContext& ctx, Errno bsd_errno) const noexcept {
     IPC::ResponseBuilder rb{ctx, 4};
 
@@ -792,7 +827,8 @@ void BSD::BuildErrnoResponse(Kernel::HLERequestContext& ctx, Errno bsd_errno) co
     rb.PushEnum(bsd_errno);
 }
 
-BSD::BSD(Core::System& system_, const char* name) : ServiceFramework{system_, name} {
+BSD::BSD(Core::System& system_, const char* name)
+    : ServiceFramework{system_, name}, worker_pool{system_, this} {
     // clang-format off
     static const FunctionInfo functions[] = {
         {0, &BSD::RegisterClient, "RegisterClient"},

src/core/hle/service/sockets/bsd.h

@@ -11,6 +11,7 @@
 #include "common/common_types.h"
 #include "core/hle/kernel/hle_ipc.h"
 #include "core/hle/service/service.h"
+#include "core/hle/service/sockets/blocking_worker.h"
 #include "core/hle/service/sockets/sockets.h"
 
 namespace Core {
@@ -137,7 +138,8 @@ private:
     void Close(Kernel::HLERequestContext& ctx);
 
     template <typename Work>
-    void ExecuteWork(Kernel::HLERequestContext& ctx, Work work);
+    void ExecuteWork(Kernel::HLERequestContext& ctx, std::string_view sleep_reason,
+                     bool is_blocking, Work work);
 
     std::pair<s32, Errno> SocketImpl(Domain domain, Type type, Protocol protocol);
     std::pair<s32, Errno> PollImpl(std::vector<u8>& write_buffer, std::vector<u8> read_buffer,
@@ -161,10 +163,15 @@ private:
 
     s32 FindFreeFileDescriptorHandle() noexcept;
     bool IsFileDescriptorValid(s32 fd) const noexcept;
+    bool IsBlockingSocket(s32 fd) const noexcept;
 
     void BuildErrnoResponse(Kernel::HLERequestContext& ctx, Errno bsd_errno) const noexcept;
 
     std::array<std::optional<FileDescriptor>, MAX_FD> file_descriptors;
+
+    BlockingWorkerPool<BSD, PollWork, AcceptWork, ConnectWork, RecvWork, RecvFromWork, SendWork,
+                       SendToWork>
+        worker_pool;
 };
 
 class BSDCFG final : public ServiceFramework<BSDCFG> {

src/core/hle/service/vi/vi.cpp

@@ -536,7 +536,8 @@ private:
         LOG_DEBUG(Service_VI, "called. id=0x{:08X} transaction={:X}, flags=0x{:08X}", id,
                   transaction, flags);
 
-        auto& buffer_queue = *nv_flinger.FindBufferQueue(id);
+        const auto guard = nv_flinger.Lock();
+        auto& buffer_queue = nv_flinger.FindBufferQueue(id);
 
         switch (transaction) {
         case TransactionId::Connect: {
@@ -546,9 +547,6 @@ private:
                                  Settings::values.resolution_factor.GetValue()),
                 static_cast<u32>(static_cast<u32>(DisplayResolution::UndockedHeight) *
                                  Settings::values.resolution_factor.GetValue())};
-
-            buffer_queue.Connect();
-
             ctx.WriteBuffer(response.Serialize());
             break;
         }
@@ -565,25 +563,40 @@ private:
             IGBPDequeueBufferRequestParcel request{ctx.ReadBuffer()};
             const u32 width{request.data.width};
             const u32 height{request.data.height};
+            auto result = buffer_queue.DequeueBuffer(width, height);
 
-            do {
+            if (result) {
-                if (auto result = buffer_queue.DequeueBuffer(width, height); result) {
                 // Buffer is available
                 IGBPDequeueBufferResponseParcel response{result->first, *result->second};
                 ctx.WriteBuffer(response.Serialize());
-                    break;
+            } else {
-                }
+                // Wait the current thread until a buffer becomes available
-            } while (buffer_queue.IsConnected());
+                ctx.SleepClientThread(
+                    "IHOSBinderDriver::DequeueBuffer", UINT64_MAX,
+                    [=, this](std::shared_ptr<Kernel::Thread> thread,
+                              Kernel::HLERequestContext& ctx, Kernel::ThreadWakeupReason reason) {
+                        // Repeat TransactParcel DequeueBuffer when a buffer is available
+                        const auto guard = nv_flinger.Lock();
+                        auto& buffer_queue = nv_flinger.FindBufferQueue(id);
+                        auto result = buffer_queue.DequeueBuffer(width, height);
+                        ASSERT_MSG(result != std::nullopt, "Could not dequeue buffer.");
 
+                        IGBPDequeueBufferResponseParcel response{result->first, *result->second};
+                        ctx.WriteBuffer(response.Serialize());
+                        IPC::ResponseBuilder rb{ctx, 2};
+                        rb.Push(RESULT_SUCCESS);
+                    },
+                    buffer_queue.GetWritableBufferWaitEvent());
+            }
             break;
         }
         case TransactionId::RequestBuffer: {
             IGBPRequestBufferRequestParcel request{ctx.ReadBuffer()};
 
             auto& buffer = buffer_queue.RequestBuffer(request.slot);
+
             IGBPRequestBufferResponseParcel response{buffer};
             ctx.WriteBuffer(response.Serialize());
-
             break;
         }
         case TransactionId::QueueBuffer: {
@@ -669,7 +682,7 @@ private:
 
         LOG_WARNING(Service_VI, "(STUBBED) called id={}, unknown={:08X}", id, unknown);
 
-        const auto& buffer_queue = *nv_flinger.FindBufferQueue(id);
+        const auto& buffer_queue = nv_flinger.FindBufferQueue(id);
 
         // TODO(Subv): Find out what this actually is.
         IPC::ResponseBuilder rb{ctx, 2, 1};

src/core/settings.cpp

@@ -148,4 +148,9 @@ void RestoreGlobalState(bool is_powered_on) {
     values.motion_enabled.SetGlobal(true);
 }
 
+void Sanitize() {
+    values.use_asynchronous_gpu_emulation.SetValue(
+        values.use_asynchronous_gpu_emulation.GetValue() || values.use_multi_core.GetValue());
+}
+
 } // namespace Settings

src/core/settings.h

@@ -257,4 +257,7 @@ void LogSettings();
 // Restore the global state of all applicable settings in the Values struct
 void RestoreGlobalState(bool is_powered_on);
 
+// Fixes settings that are known to cause issues with the emulator
+void Sanitize();
+
 } // namespace Settings

src/input_common/udp/client.cpp

@@ -225,6 +225,11 @@ void Client::OnPortInfo([[maybe_unused]] Response::PortInfo data) {
 }
 
 void Client::OnPadData(Response::PadData data, std::size_t client) {
+    // Accept packets only for the correct pad
+    if (static_cast<u8>(clients[client].pad_index) != data.info.id) {
+        return;
+    }
+
     LOG_TRACE(Input, "PadData packet received");
     if (data.packet_counter == clients[client].packet_sequence) {
         LOG_WARNING(

src/video_core/CMakeLists.txt

@@ -48,7 +48,6 @@ add_library(video_core STATIC
     engines/shader_bytecode.h
     engines/shader_header.h
     engines/shader_type.h
-    framebuffer_config.h
     macro/macro.cpp
     macro/macro.h
     macro/macro_hle.cpp
@@ -60,6 +59,10 @@ add_library(video_core STATIC
     fence_manager.h
     gpu.cpp
     gpu.h
+    gpu_asynch.cpp
+    gpu_asynch.h
+    gpu_synch.cpp
+    gpu_synch.h
     gpu_thread.cpp
     gpu_thread.h
     guest_driver.cpp

src/video_core/gpu.cpp

@@ -10,7 +10,6 @@
 #include "core/core_timing.h"
 #include "core/core_timing_util.h"
 #include "core/frontend/emu_window.h"
-#include "core/hardware_interrupt_manager.h"
 #include "core/memory.h"
 #include "core/settings.h"
 #include "video_core/engines/fermi_2d.h"
@@ -37,8 +36,7 @@ GPU::GPU(Core::System& system_, bool is_async_, bool use_nvdec_)
       kepler_compute{std::make_unique<Engines::KeplerCompute>(system, *memory_manager)},
       maxwell_dma{std::make_unique<Engines::MaxwellDMA>(system, *memory_manager)},
       kepler_memory{std::make_unique<Engines::KeplerMemory>(system, *memory_manager)},
-      shader_notify{std::make_unique<VideoCore::ShaderNotify>()}, is_async{is_async_},
+      shader_notify{std::make_unique<VideoCore::ShaderNotify>()}, is_async{is_async_} {}
-      gpu_thread{system_, is_async_} {}
 
 GPU::~GPU() = default;
 
@@ -200,6 +198,10 @@ void GPU::SyncGuestHost() {
     renderer->Rasterizer().SyncGuestHost();
 }
 
+void GPU::OnCommandListEnd() {
+    renderer->Rasterizer().ReleaseFences();
+}
+
 enum class GpuSemaphoreOperation {
     AcquireEqual = 0x1,
     WriteLong = 0x2,
@@ -459,75 +461,4 @@ void GPU::ProcessSemaphoreAcquire() {
     }
 }
 
-void GPU::Start() {
-    gpu_thread.StartThread(*renderer, renderer->Context(), *dma_pusher, *cdma_pusher);
-    cpu_context = renderer->GetRenderWindow().CreateSharedContext();
-    cpu_context->MakeCurrent();
-}
-
-void GPU::ObtainContext() {
-    cpu_context->MakeCurrent();
-}
-
-void GPU::ReleaseContext() {
-    cpu_context->DoneCurrent();
-}
-
-void GPU::PushGPUEntries(Tegra::CommandList&& entries) {
-    gpu_thread.SubmitList(std::move(entries));
-}
-
-void GPU::PushCommandBuffer(Tegra::ChCommandHeaderList& entries) {
-    if (!use_nvdec) {
-        return;
-    }
-    // This condition fires when a video stream ends, clear all intermediary data
-    if (entries[0].raw == 0xDEADB33F) {
-        cdma_pusher.reset();
-        return;
-    }
-    if (!cdma_pusher) {
-        cdma_pusher = std::make_unique<Tegra::CDmaPusher>(*this);
-    }
-
-    // SubmitCommandBuffer would make the nvdec operations async, this is not currently working
-    // TODO(ameerj): RE proper async nvdec operation
-    // gpu_thread.SubmitCommandBuffer(std::move(entries));
-
-    cdma_pusher->Push(std::move(entries));
-    cdma_pusher->DispatchCalls();
-}
-
-void GPU::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) {
-    gpu_thread.SwapBuffers(framebuffer);
-}
-
-void GPU::FlushRegion(VAddr addr, u64 size) {
-    gpu_thread.FlushRegion(addr, size);
-}
-
-void GPU::InvalidateRegion(VAddr addr, u64 size) {
-    gpu_thread.InvalidateRegion(addr, size);
-}
-
-void GPU::FlushAndInvalidateRegion(VAddr addr, u64 size) {
-    gpu_thread.FlushAndInvalidateRegion(addr, size);
-}
-
-void GPU::TriggerCpuInterrupt(const u32 syncpoint_id, const u32 value) const {
-    auto& interrupt_manager = system.InterruptManager();
-    interrupt_manager.GPUInterruptSyncpt(syncpoint_id, value);
-}
-
-void GPU::WaitIdle() const {
-    gpu_thread.WaitIdle();
-}
-
-void GPU::OnCommandListEnd() {
-    if (is_async) {
-        // This command only applies to asynchronous GPU mode
-        gpu_thread.OnCommandListEnd();
-    }
-}
-
 } // namespace Tegra

src/video_core/gpu.h

@@ -15,8 +15,6 @@
 #include "core/hle/service/nvflinger/buffer_queue.h"
 #include "video_core/cdma_pusher.h"
 #include "video_core/dma_pusher.h"
-#include "video_core/framebuffer_config.h"
-#include "video_core/gpu_thread.h"
 
 using CacheAddr = std::uintptr_t;
 [[nodiscard]] inline CacheAddr ToCacheAddr(const void* host_ptr) {
@@ -103,6 +101,28 @@ enum class DepthFormat : u32 {
 struct CommandListHeader;
 class DebugContext;
 
+/**
+ * Struct describing framebuffer configuration
+ */
+struct FramebufferConfig {
+    enum class PixelFormat : u32 {
+        A8B8G8R8_UNORM = 1,
+        RGB565_UNORM = 4,
+        B8G8R8A8_UNORM = 5,
+    };
+
+    VAddr address;
+    u32 offset;
+    u32 width;
+    u32 height;
+    u32 stride;
+    PixelFormat pixel_format;
+
+    using TransformFlags = Service::NVFlinger::BufferQueue::BufferTransformFlags;
+    TransformFlags transform_flags;
+    Common::Rectangle<int> crop_rect;
+};
+
 namespace Engines {
 class Fermi2D;
 class Maxwell3D;
@@ -121,7 +141,7 @@ enum class EngineID {
 
 class MemoryManager;
 
-class GPU final {
+class GPU {
 public:
     struct MethodCall {
         u32 method{};
@@ -139,7 +159,7 @@ public:
     };
 
     explicit GPU(Core::System& system_, bool is_async_, bool use_nvdec_);
-    ~GPU();
+    virtual ~GPU();
 
     /// Binds a renderer to the GPU.
     void BindRenderer(std::unique_ptr<VideoCore::RendererBase> renderer);
@@ -156,7 +176,7 @@ public:
     /// Synchronizes CPU writes with Host GPU memory.
     void SyncGuestHost();
     /// Signal the ending of command list.
-    void OnCommandListEnd();
+    virtual void OnCommandListEnd();
 
     /// Request a host GPU memory flush from the CPU.
     [[nodiscard]] u64 RequestFlush(VAddr addr, std::size_t size);
@@ -220,7 +240,7 @@ public:
     }
 
     // Waits for the GPU to finish working
-    void WaitIdle() const;
+    virtual void WaitIdle() const = 0;
 
     /// Allows the CPU/NvFlinger to wait on the GPU before presenting a frame.
     void WaitFence(u32 syncpoint_id, u32 value);
@@ -310,34 +330,34 @@ public:
     /// Performs any additional setup necessary in order to begin GPU emulation.
     /// This can be used to launch any necessary threads and register any necessary
     /// core timing events.
-    void Start();
+    virtual void Start() = 0;
 
     /// Obtain the CPU Context
-    void ObtainContext();
+    virtual void ObtainContext() = 0;
 
     /// Release the CPU Context
-    void ReleaseContext();
+    virtual void ReleaseContext() = 0;
 
     /// Push GPU command entries to be processed
-    void PushGPUEntries(Tegra::CommandList&& entries);
+    virtual void PushGPUEntries(Tegra::CommandList&& entries) = 0;
 
     /// Push GPU command buffer entries to be processed
-    void PushCommandBuffer(Tegra::ChCommandHeaderList& entries);
+    virtual void PushCommandBuffer(Tegra::ChCommandHeaderList& entries) = 0;
 
     /// Swap buffers (render frame)
-    void SwapBuffers(const Tegra::FramebufferConfig* framebuffer);
+    virtual void SwapBuffers(const Tegra::FramebufferConfig* framebuffer) = 0;
 
     /// Notify rasterizer that any caches of the specified region should be flushed to Switch memory
-    void FlushRegion(VAddr addr, u64 size);
+    virtual void FlushRegion(VAddr addr, u64 size) = 0;
 
     /// Notify rasterizer that any caches of the specified region should be invalidated
-    void InvalidateRegion(VAddr addr, u64 size);
+    virtual void InvalidateRegion(VAddr addr, u64 size) = 0;
 
     /// Notify rasterizer that any caches of the specified region should be flushed and invalidated
-    void FlushAndInvalidateRegion(VAddr addr, u64 size);
+    virtual void FlushAndInvalidateRegion(VAddr addr, u64 size) = 0;
 
 protected:
-    void TriggerCpuInterrupt(u32 syncpoint_id, u32 value) const;
+    virtual void TriggerCpuInterrupt(u32 syncpoint_id, u32 value) const = 0;
 
 private:
     void ProcessBindMethod(const MethodCall& method_call);
@@ -407,9 +427,6 @@ private:
     std::mutex flush_request_mutex;
 
     const bool is_async;
-
-    VideoCommon::GPUThread::ThreadManager gpu_thread;
-    std::unique_ptr<Core::Frontend::GraphicsContext> cpu_context;
 };
 
 #define ASSERT_REG_POSITION(field_name, position)                                                  \

src/video_core/gpu_asynch.cpp

@@ -0,0 +1,86 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "core/core.h"
+#include "core/hardware_interrupt_manager.h"
+#include "video_core/gpu_asynch.h"
+#include "video_core/gpu_thread.h"
+#include "video_core/renderer_base.h"
+
+namespace VideoCommon {
+
+GPUAsynch::GPUAsynch(Core::System& system_, bool use_nvdec_)
+    : GPU{system_, true, use_nvdec_}, gpu_thread{system_} {}
+
+GPUAsynch::~GPUAsynch() = default;
+
+void GPUAsynch::Start() {
+    gpu_thread.StartThread(*renderer, renderer->Context(), *dma_pusher, *cdma_pusher);
+    cpu_context = renderer->GetRenderWindow().CreateSharedContext();
+    cpu_context->MakeCurrent();
+}
+
+void GPUAsynch::ObtainContext() {
+    cpu_context->MakeCurrent();
+}
+
+void GPUAsynch::ReleaseContext() {
+    cpu_context->DoneCurrent();
+}
+
+void GPUAsynch::PushGPUEntries(Tegra::CommandList&& entries) {
+    gpu_thread.SubmitList(std::move(entries));
+}
+
+void GPUAsynch::PushCommandBuffer(Tegra::ChCommandHeaderList& entries) {
+    if (!use_nvdec) {
+        return;
+    }
+    // This condition fires when a video stream ends, clear all intermediary data
+    if (entries[0].raw == 0xDEADB33F) {
+        cdma_pusher.reset();
+        return;
+    }
+    if (!cdma_pusher) {
+        cdma_pusher = std::make_unique<Tegra::CDmaPusher>(*this);
+    }
+
+    // SubmitCommandBuffer would make the nvdec operations async, this is not currently working
+    // TODO(ameerj): RE proper async nvdec operation
+    // gpu_thread.SubmitCommandBuffer(std::move(entries));
+
+    cdma_pusher->Push(std::move(entries));
+    cdma_pusher->DispatchCalls();
+}
+
+void GPUAsynch::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) {
+    gpu_thread.SwapBuffers(framebuffer);
+}
+
+void GPUAsynch::FlushRegion(VAddr addr, u64 size) {
+    gpu_thread.FlushRegion(addr, size);
+}
+
+void GPUAsynch::InvalidateRegion(VAddr addr, u64 size) {
+    gpu_thread.InvalidateRegion(addr, size);
+}
+
+void GPUAsynch::FlushAndInvalidateRegion(VAddr addr, u64 size) {
+    gpu_thread.FlushAndInvalidateRegion(addr, size);
+}
+
+void GPUAsynch::TriggerCpuInterrupt(const u32 syncpoint_id, const u32 value) const {
+    auto& interrupt_manager = system.InterruptManager();
+    interrupt_manager.GPUInterruptSyncpt(syncpoint_id, value);
+}
+
+void GPUAsynch::WaitIdle() const {
+    gpu_thread.WaitIdle();
+}
+
+void GPUAsynch::OnCommandListEnd() {
+    gpu_thread.OnCommandListEnd();
+}
+
+} // namespace VideoCommon

src/video_core/gpu_asynch.h

@@ -0,0 +1,47 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include "video_core/gpu.h"
+#include "video_core/gpu_thread.h"
+
+namespace Core::Frontend {
+class GraphicsContext;
+}
+
+namespace VideoCore {
+class RendererBase;
+} // namespace VideoCore
+
+namespace VideoCommon {
+
+/// Implementation of GPU interface that runs the GPU asynchronously
+class GPUAsynch final : public Tegra::GPU {
+public:
+    explicit GPUAsynch(Core::System& system_, bool use_nvdec_);
+    ~GPUAsynch() override;
+
+    void Start() override;
+    void ObtainContext() override;
+    void ReleaseContext() override;
+    void PushGPUEntries(Tegra::CommandList&& entries) override;
+    void PushCommandBuffer(Tegra::ChCommandHeaderList& entries) override;
+    void SwapBuffers(const Tegra::FramebufferConfig* framebuffer) override;
+    void FlushRegion(VAddr addr, u64 size) override;
+    void InvalidateRegion(VAddr addr, u64 size) override;
+    void FlushAndInvalidateRegion(VAddr addr, u64 size) override;
+    void WaitIdle() const override;
+
+    void OnCommandListEnd() override;
+
+protected:
+    void TriggerCpuInterrupt(u32 syncpoint_id, u32 value) const override;
+
+private:
+    GPUThread::ThreadManager gpu_thread;
+    std::unique_ptr<Core::Frontend::GraphicsContext> cpu_context;
+};
+
+} // namespace VideoCommon

src/video_core/gpu_synch.cpp

@@ -0,0 +1,61 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#include "video_core/gpu_synch.h"
+#include "video_core/renderer_base.h"
+
+namespace VideoCommon {
+
+GPUSynch::GPUSynch(Core::System& system_, bool use_nvdec_) : GPU{system_, false, use_nvdec_} {}
+
+GPUSynch::~GPUSynch() = default;
+
+void GPUSynch::Start() {}
+
+void GPUSynch::ObtainContext() {
+    renderer->Context().MakeCurrent();
+}
+
+void GPUSynch::ReleaseContext() {
+    renderer->Context().DoneCurrent();
+}
+
+void GPUSynch::PushGPUEntries(Tegra::CommandList&& entries) {
+    dma_pusher->Push(std::move(entries));
+    dma_pusher->DispatchCalls();
+}
+
+void GPUSynch::PushCommandBuffer(Tegra::ChCommandHeaderList& entries) {
+    if (!use_nvdec) {
+        return;
+    }
+    // This condition fires when a video stream ends, clears all intermediary data
+    if (entries[0].raw == 0xDEADB33F) {
+        cdma_pusher.reset();
+        return;
+    }
+    if (!cdma_pusher) {
+        cdma_pusher = std::make_unique<Tegra::CDmaPusher>(*this);
+    }
+    cdma_pusher->Push(std::move(entries));
+    cdma_pusher->DispatchCalls();
+}
+
+void GPUSynch::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) {
+    renderer->SwapBuffers(framebuffer);
+}
+
+void GPUSynch::FlushRegion(VAddr addr, u64 size) {
+    renderer->Rasterizer().FlushRegion(addr, size);
+}
+
+void GPUSynch::InvalidateRegion(VAddr addr, u64 size) {
+    renderer->Rasterizer().InvalidateRegion(addr, size);
+}
+
+void GPUSynch::FlushAndInvalidateRegion(VAddr addr, u64 size) {
+    renderer->Rasterizer().FlushAndInvalidateRegion(addr, size);
+}
+
+} // namespace VideoCommon

src/video_core/gpu_synch.h

@@ -0,0 +1,41 @@
+// Copyright 2019 yuzu Emulator Project
+// Licensed under GPLv2 or any later version
+// Refer to the license.txt file included.
+
+#pragma once
+
+#include "video_core/gpu.h"
+
+namespace Core::Frontend {
+class GraphicsContext;
+}
+
+namespace VideoCore {
+class RendererBase;
+} // namespace VideoCore
+
+namespace VideoCommon {
+
+/// Implementation of GPU interface that runs the GPU synchronously
+class GPUSynch final : public Tegra::GPU {
+public:
+    explicit GPUSynch(Core::System& system_, bool use_nvdec_);
+    ~GPUSynch() override;
+
+    void Start() override;
+    void ObtainContext() override;
+    void ReleaseContext() override;
+    void PushGPUEntries(Tegra::CommandList&& entries) override;
+    void PushCommandBuffer(Tegra::ChCommandHeaderList& entries) override;
+    void SwapBuffers(const Tegra::FramebufferConfig* framebuffer) override;
+    void FlushRegion(VAddr addr, u64 size) override;
+    void InvalidateRegion(VAddr addr, u64 size) override;
+    void FlushAndInvalidateRegion(VAddr addr, u64 size) override;
+    void WaitIdle() const override {}
+
+protected:
+    void TriggerCpuInterrupt([[maybe_unused]] u32 syncpoint_id,
+                             [[maybe_unused]] u32 value) const override {}
+};
+
+} // namespace VideoCommon

src/video_core/gpu_thread.cpp

@@ -4,7 +4,6 @@
 
 #include "common/assert.h"
 #include "common/microprofile.h"
-#include "common/scope_exit.h"
 #include "common/thread.h"
 #include "core/core.h"
 #include "core/frontend/emu_window.h"
@@ -22,8 +21,6 @@ static void RunThread(Core::System& system, VideoCore::RendererBase& renderer,
                       SynchState& state, Tegra::CDmaPusher& cdma_pusher) {
     std::string name = "yuzu:GPU";
     MicroProfileOnThreadCreate(name.c_str());
-    SCOPE_EXIT({ MicroProfileOnThreadExit(); });
-
     Common::SetCurrentThreadName(name.c_str());
     Common::SetCurrentThreadPriority(Common::ThreadPriority::High);
     system.RegisterHostThread();
@@ -68,8 +65,7 @@ static void RunThread(Core::System& system, VideoCore::RendererBase& renderer,
     }
 }
 
-ThreadManager::ThreadManager(Core::System& system_, bool is_async_)
+ThreadManager::ThreadManager(Core::System& system_) : system{system_} {}
-    : system{system_}, is_async{is_async_} {}
 
 ThreadManager::~ThreadManager() {
     if (!thread.joinable()) {
@@ -101,30 +97,19 @@ void ThreadManager::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) {
 }
 
 void ThreadManager::FlushRegion(VAddr addr, u64 size) {
-    if (!is_async) {
+    if (!Settings::IsGPULevelHigh()) {
-        // Always flush with synchronous GPU mode
         PushCommand(FlushRegionCommand(addr, size));
         return;
     }
-
+    if (!Settings::IsGPULevelExtreme()) {
-    // Asynchronous GPU mode
+        return;
-    switch (Settings::values.gpu_accuracy.GetValue()) {
+    }
-    case Settings::GPUAccuracy::Normal:
+    if (system.Renderer().Rasterizer().MustFlushRegion(addr, size)) {
-        PushCommand(FlushRegionCommand(addr, size));
-        break;
-    case Settings::GPUAccuracy::High:
-        // TODO(bunnei): Is this right? Preserving existing behavior for now
-        break;
-    case Settings::GPUAccuracy::Extreme: {
         auto& gpu = system.GPU();
         u64 fence = gpu.RequestFlush(addr, size);
         PushCommand(GPUTickCommand());
         while (fence > gpu.CurrentFlushRequestFence()) {
         }
-        break;
-    }
-    default:
-        UNIMPLEMENTED_MSG("Unsupported gpu_accuracy {}", Settings::values.gpu_accuracy.GetValue());
     }
 }
 
@@ -138,8 +123,7 @@ void ThreadManager::FlushAndInvalidateRegion(VAddr addr, u64 size) {
 }
 
 void ThreadManager::WaitIdle() const {
-    while (state.last_fence > state.signaled_fence.load(std::memory_order_relaxed) &&
+    while (state.last_fence > state.signaled_fence.load(std::memory_order_relaxed)) {
-           system.IsPoweredOn()) {
     }
 }
 
@@ -150,12 +134,6 @@ void ThreadManager::OnCommandListEnd() {
 u64 ThreadManager::PushCommand(CommandData&& command_data) {
     const u64 fence{++state.last_fence};
     state.queue.Push(CommandDataContainer(std::move(command_data), fence));
-
-    if (!is_async) {
-        // In synchronous GPU mode, block the caller until the command has executed
-        WaitIdle();
-    }
-
     return fence;
 }
 

src/video_core/gpu_thread.h

@@ -10,9 +10,8 @@
 #include <optional>
 #include <thread>
 #include <variant>
-
 #include "common/threadsafe_queue.h"
-#include "video_core/framebuffer_config.h"
+#include "video_core/gpu.h"
 
 namespace Tegra {
 struct FramebufferConfig;
@@ -26,10 +25,6 @@ class GraphicsContext;
 class System;
 } // namespace Core
 
-namespace VideoCore {
-class RendererBase;
-} // namespace VideoCore
-
 namespace VideoCommon::GPUThread {
 
 /// Command to signal to the GPU thread that processing has ended
@@ -117,7 +112,7 @@ struct SynchState final {
 /// Class used to manage the GPU thread
 class ThreadManager final {
 public:
-    explicit ThreadManager(Core::System& system_, bool is_async_);
+    explicit ThreadManager(Core::System& system_);
     ~ThreadManager();
 
     /// Creates and starts the GPU thread.
@@ -155,7 +150,6 @@ private:
     Core::System& system;
     std::thread thread;
     std::thread::id thread_id;
-    const bool is_async;
 };
 
 } // namespace VideoCommon::GPUThread

src/video_core/video_core.cpp

@@ -7,6 +7,8 @@
 #include "common/logging/log.h"
 #include "core/core.h"
 #include "core/settings.h"
+#include "video_core/gpu_asynch.h"
+#include "video_core/gpu_synch.h"
 #include "video_core/renderer_base.h"
 #include "video_core/renderer_opengl/renderer_opengl.h"
 #include "video_core/renderer_vulkan/renderer_vulkan.h"
@@ -37,9 +39,13 @@ std::unique_ptr<VideoCore::RendererBase> CreateRenderer(
 namespace VideoCore {
 
 std::unique_ptr<Tegra::GPU> CreateGPU(Core::Frontend::EmuWindow& emu_window, Core::System& system) {
+    std::unique_ptr<Tegra::GPU> gpu;
     const bool use_nvdec = Settings::values.use_nvdec_emulation.GetValue();
-    std::unique_ptr<Tegra::GPU> gpu = std::make_unique<Tegra::GPU>(
+    if (Settings::values.use_asynchronous_gpu_emulation.GetValue()) {
-        system, Settings::values.use_asynchronous_gpu_emulation.GetValue(), use_nvdec);
+        gpu = std::make_unique<VideoCommon::GPUAsynch>(system, use_nvdec);
+    } else {
+        gpu = std::make_unique<VideoCommon::GPUSynch>(system, use_nvdec);
+    }
 
     auto context = emu_window.CreateSharedContext();
     const auto scope = context->Acquire();

src/yuzu/configuration/config.cpp

@@ -1589,12 +1589,14 @@ void Config::WriteSettingGlobal(const QString& name, const QVariant& value, bool
 
 void Config::Reload() {
     ReadValues();
+    Settings::Sanitize();
     // To apply default value changes
     SaveValues();
     Settings::Apply(Core::System::GetInstance());
 }
 
 void Config::Save() {
+    Settings::Sanitize();
     SaveValues();
 }
 

src/yuzu/main.cpp

@@ -580,8 +580,9 @@ void GMainWindow::InitializeWidgets() {
         if (emulation_running) {
             return;
         }
-        Settings::values.use_asynchronous_gpu_emulation.SetValue(
+        const bool is_async = !Settings::values.use_asynchronous_gpu_emulation.GetValue() ||
-            !Settings::values.use_asynchronous_gpu_emulation.GetValue());
+                              Settings::values.use_multi_core.GetValue();
+        Settings::values.use_asynchronous_gpu_emulation.SetValue(is_async);
         async_status_button->setChecked(Settings::values.use_asynchronous_gpu_emulation.GetValue());
         Settings::Apply(Core::System::GetInstance());
     });
@@ -598,13 +599,16 @@ void GMainWindow::InitializeWidgets() {
             return;
         }
         Settings::values.use_multi_core.SetValue(!Settings::values.use_multi_core.GetValue());
+        const bool is_async = Settings::values.use_asynchronous_gpu_emulation.GetValue() ||
+                              Settings::values.use_multi_core.GetValue();
+        Settings::values.use_asynchronous_gpu_emulation.SetValue(is_async);
+        async_status_button->setChecked(Settings::values.use_asynchronous_gpu_emulation.GetValue());
         multicore_status_button->setChecked(Settings::values.use_multi_core.GetValue());
         Settings::Apply(Core::System::GetInstance());
     });
     multicore_status_button->setText(tr("MULTICORE"));
     multicore_status_button->setCheckable(true);
     multicore_status_button->setChecked(Settings::values.use_multi_core.GetValue());
-
     statusBar()->insertPermanentWidget(0, multicore_status_button);
     statusBar()->insertPermanentWidget(0, async_status_button);
 
@@ -2529,6 +2533,9 @@ void GMainWindow::UpdateStatusBar() {
 void GMainWindow::UpdateStatusButtons() {
     dock_status_button->setChecked(Settings::values.use_docked_mode.GetValue());
     multicore_status_button->setChecked(Settings::values.use_multi_core.GetValue());
+    Settings::values.use_asynchronous_gpu_emulation.SetValue(
+        Settings::values.use_asynchronous_gpu_emulation.GetValue() ||
+        Settings::values.use_multi_core.GetValue());
     async_status_button->setChecked(Settings::values.use_asynchronous_gpu_emulation.GetValue());
     renderer_status_button->setChecked(Settings::values.renderer_backend.GetValue() ==
                                        Settings::RendererBackend::Vulkan);