23 files changed, 336 insertions, 85 deletions

diff --git a/src/core/core.cpp b/src/core/core.cpp
index ac0fb7872..06fba4ce5 100644
--- a/src/core/core.cpp
+++ b/src/core/core.cpp
@@ -293,6 +293,7 @@ struct System::Impl {
         ASSERT(Kernel::KProcess::Initialize(main_process, system, "main",
                                             Kernel::KProcess::ProcessType::Userland, resource_limit)
                    .IsSuccess());
+        Kernel::KProcess::Register(system.Kernel(), main_process);
         kernel.MakeApplicationProcess(main_process);
         const auto [load_result, load_parameters] = app_loader->Load(*main_process, system);
         if (load_result != Loader::ResultStatus::Success) {
diff --git a/src/core/hle/kernel/k_auto_object.h b/src/core/hle/kernel/k_auto_object.h
index 9b71fe371..f384b1568 100644
--- a/src/core/hle/kernel/k_auto_object.h
+++ b/src/core/hle/kernel/k_auto_object.h
@@ -182,8 +182,8 @@ public:
     explicit KAutoObjectWithList(KernelCore& kernel) : KAutoObject(kernel) {}
 
     static int Compare(const KAutoObjectWithList& lhs, const KAutoObjectWithList& rhs) {
-        const u64 lid = lhs.GetId();
-        const u64 rid = rhs.GetId();
+        const uintptr_t lid = reinterpret_cast<uintptr_t>(std::addressof(lhs));
+        const uintptr_t rid = reinterpret_cast<uintptr_t>(std::addressof(rhs));
 
         if (lid < rid) {
             return -1;
diff --git a/src/core/hle/kernel/kernel.cpp b/src/core/hle/kernel/kernel.cpp
index 4f3366c9d..f33600ca5 100644
--- a/src/core/hle/kernel/kernel.cpp
+++ b/src/core/hle/kernel/kernel.cpp
@@ -95,7 +95,7 @@ struct KernelCore::Impl {
                                        pt_heap_region.GetSize());
         }
 
-        InitializeHackSharedMemory();
+        InitializeHackSharedMemory(kernel);
         RegisterHostThread(nullptr);
     }
 
@@ -216,10 +216,12 @@ struct KernelCore::Impl {
             auto* main_thread{Kernel::KThread::Create(system.Kernel())};
             main_thread->SetCurrentCore(core);
             ASSERT(Kernel::KThread::InitializeMainThread(system, main_thread, core).IsSuccess());
+            KThread::Register(system.Kernel(), main_thread);
 
             auto* idle_thread{Kernel::KThread::Create(system.Kernel())};
             idle_thread->SetCurrentCore(core);
             ASSERT(Kernel::KThread::InitializeIdleThread(system, idle_thread, core).IsSuccess());
+            KThread::Register(system.Kernel(), idle_thread);
 
             schedulers[i]->Initialize(main_thread, idle_thread, core);
         }
@@ -230,6 +232,7 @@ struct KernelCore::Impl {
                                        const Core::Timing::CoreTiming& core_timing) {
         system_resource_limit = KResourceLimit::Create(system.Kernel());
         system_resource_limit->Initialize(&core_timing);
+        KResourceLimit::Register(kernel, system_resource_limit);
 
         const auto sizes{memory_layout->GetTotalAndKernelMemorySizes()};
         const auto total_size{sizes.first};
@@ -355,6 +358,7 @@ struct KernelCore::Impl {
             ASSERT(KThread::InitializeHighPriorityThread(system, shutdown_threads[core_id], {}, {},
                                                          core_id)
                        .IsSuccess());
+            KThread::Register(system.Kernel(), shutdown_threads[core_id]);
         }
     }
 
@@ -729,7 +733,7 @@ struct KernelCore::Impl {
         memory_manager->Initialize(management_region.GetAddress(), management_region.GetSize());
     }
 
-    void InitializeHackSharedMemory() {
+    void InitializeHackSharedMemory(KernelCore& kernel) {
         // Setup memory regions for emulated processes
         // TODO(bunnei): These should not be hardcoded regions initialized within the kernel
         constexpr std::size_t hid_size{0x40000};
@@ -746,14 +750,23 @@ struct KernelCore::Impl {
 
         hid_shared_mem->Initialize(system.DeviceMemory(), nullptr, Svc::MemoryPermission::None,
                                    Svc::MemoryPermission::Read, hid_size);
+        KSharedMemory::Register(kernel, hid_shared_mem);
+
         font_shared_mem->Initialize(system.DeviceMemory(), nullptr, Svc::MemoryPermission::None,
                                     Svc::MemoryPermission::Read, font_size);
+        KSharedMemory::Register(kernel, font_shared_mem);
+
         irs_shared_mem->Initialize(system.DeviceMemory(), nullptr, Svc::MemoryPermission::None,
                                    Svc::MemoryPermission::Read, irs_size);
+        KSharedMemory::Register(kernel, irs_shared_mem);
+
         time_shared_mem->Initialize(system.DeviceMemory(), nullptr, Svc::MemoryPermission::None,
                                     Svc::MemoryPermission::Read, time_size);
+        KSharedMemory::Register(kernel, time_shared_mem);
+
         hidbus_shared_mem->Initialize(system.DeviceMemory(), nullptr, Svc::MemoryPermission::None,
                                       Svc::MemoryPermission::Read, hidbus_size);
+        KSharedMemory::Register(kernel, hidbus_shared_mem);
     }
 
     std::mutex registered_objects_lock;
@@ -1072,12 +1085,15 @@ static std::jthread RunHostThreadFunc(KernelCore& kernel, KProcess* process,
     // Commit the thread reservation.
     thread_reservation.Commit();
 
+    // Register the thread.
+    KThread::Register(kernel, thread);
+
     return std::jthread(
         [&kernel, thread, thread_name{std::move(thread_name)}, func{std::move(func)}] {
             // Set the thread name.
             Common::SetCurrentThreadName(thread_name.c_str());
 
-            // Register the thread.
+            // Set the thread as current.
             kernel.RegisterHostThread(thread);
 
             // Run the callback.
@@ -1099,6 +1115,9 @@ std::jthread KernelCore::RunOnHostCoreProcess(std::string&& process_name,
     // Ensure that we don't hold onto any extra references.
     SCOPE_EXIT({ process->Close(); });
 
+    // Register the new process.
+    KProcess::Register(*this, process);
+
     // Run the host thread.
     return RunHostThreadFunc(*this, process, std::move(process_name), std::move(func));
 }
@@ -1124,6 +1143,9 @@ void KernelCore::RunOnGuestCoreProcess(std::string&& process_name, std::function
     // Ensure that we don't hold onto any extra references.
     SCOPE_EXIT({ process->Close(); });
 
+    // Register the new process.
+    KProcess::Register(*this, process);
+
     // Reserve a new thread from the process resource limit.
     KScopedResourceReservation thread_reservation(process, LimitableResource::ThreadCountMax);
     ASSERT(thread_reservation.Succeeded());
@@ -1136,6 +1158,9 @@ void KernelCore::RunOnGuestCoreProcess(std::string&& process_name, std::function
     // Commit the thread reservation.
     thread_reservation.Commit();
 
+    // Register the new thread.
+    KThread::Register(*this, thread);
+
     // Begin running the thread.
     ASSERT(R_SUCCEEDED(thread->Run()));
 }
diff --git a/src/core/hle/service/ipc_helpers.h b/src/core/hle/service/ipc_helpers.h
index e4cb4e1f2..0e222362e 100644
--- a/src/core/hle/service/ipc_helpers.h
+++ b/src/core/hle/service/ipc_helpers.h
@@ -156,6 +156,7 @@ public:
 
             auto* session = Kernel::KSession::Create(kernel);
             session->Initialize(nullptr, 0);
+            Kernel::KSession::Register(kernel, session);
 
             auto next_manager = std::make_shared<Service::SessionRequestManager>(
                 kernel, manager->GetServerManager());
diff --git a/src/core/hle/service/kernel_helpers.cpp b/src/core/hle/service/kernel_helpers.cpp
index a39ce5212..6a313a03b 100644
--- a/src/core/hle/service/kernel_helpers.cpp
+++ b/src/core/hle/service/kernel_helpers.cpp
@@ -25,6 +25,9 @@ ServiceContext::ServiceContext(Core::System& system_, std::string name_)
                                         Kernel::KProcess::ProcessType::KernelInternal,
                                         kernel.GetSystemResourceLimit())
                .IsSuccess());
+
+    // Register the process.
+    Kernel::KProcess::Register(kernel, process);
     process_created = true;
 }
 
diff --git a/src/core/hle/service/mutex.cpp b/src/core/hle/service/mutex.cpp
index 07589a0f0..b0ff71d1b 100644
--- a/src/core/hle/service/mutex.cpp
+++ b/src/core/hle/service/mutex.cpp
@@ -12,6 +12,9 @@ Mutex::Mutex(Core::System& system) : m_system(system) {
     m_event = Kernel::KEvent::Create(system.Kernel());
     m_event->Initialize(nullptr);
 
+    // Register the event.
+    Kernel::KEvent::Register(system.Kernel(), m_event);
+
     ASSERT(R_SUCCEEDED(m_event->Signal()));
 }
 
diff --git a/src/core/hle/service/server_manager.cpp b/src/core/hle/service/server_manager.cpp
index 6b4a1291e..156bc27d8 100644
--- a/src/core/hle/service/server_manager.cpp
+++ b/src/core/hle/service/server_manager.cpp
@@ -33,6 +33,9 @@ ServerManager::ServerManager(Core::System& system) : m_system{system}, m_serve_m
     // Initialize event.
     m_event = Kernel::KEvent::Create(system.Kernel());
     m_event->Initialize(nullptr);
+
+    // Register event.
+    Kernel::KEvent::Register(system.Kernel(), m_event);
 }
 
 ServerManager::~ServerManager() {
@@ -160,6 +163,9 @@ Result ServerManager::ManageDeferral(Kernel::KEvent** out_event) {
     // Initialize the event.
     m_deferral_event->Initialize(nullptr);
 
+    // Register the event.
+    Kernel::KEvent::Register(m_system.Kernel(), m_deferral_event);
+
     // Set the output.
     *out_event = m_deferral_event;
 
diff --git a/src/core/hle/service/sm/sm.cpp b/src/core/hle/service/sm/sm.cpp
index c45be5726..1608fa24c 100644
--- a/src/core/hle/service/sm/sm.cpp
+++ b/src/core/hle/service/sm/sm.cpp
@@ -64,6 +64,9 @@ Result ServiceManager::RegisterService(std::string name, u32 max_sessions,
     auto* port = Kernel::KPort::Create(kernel);
     port->Initialize(ServerSessionCountMax, false, 0);
 
+    // Register the port.
+    Kernel::KPort::Register(kernel, port);
+
     service_ports.emplace(name, port);
     registered_services.emplace(name, handler);
     if (deferral_event) {
diff --git a/src/core/hle/service/sm/sm_controller.cpp b/src/core/hle/service/sm/sm_controller.cpp
index 419c1df2b..7dce28fe0 100644
--- a/src/core/hle/service/sm/sm_controller.cpp
+++ b/src/core/hle/service/sm/sm_controller.cpp
@@ -49,6 +49,9 @@ void Controller::CloneCurrentObject(HLERequestContext& ctx) {
     // Commit the session reservation.
     session_reservation.Commit();
 
+    // Register the session.
+    Kernel::KSession::Register(system.Kernel(), session);
+
     // Register with server manager.
     session_manager->GetServerManager().RegisterSession(&session->GetServerSession(),
                                                         session_manager);
diff --git a/src/core/memory.cpp b/src/core/memory.cpp
index 432310632..a9667463f 100644
--- a/src/core/memory.cpp
+++ b/src/core/memory.cpp
@@ -462,7 +462,7 @@ struct Memory::Impl {
         }
 
         if (Settings::IsFastmemEnabled()) {
-            const bool is_read_enable = Settings::IsGPULevelHigh() || !cached;
+            const bool is_read_enable = !Settings::IsGPULevelExtreme() || !cached;
             system.DeviceMemory().buffer.Protect(vaddr, size, is_read_enable, !cached);
         }
 
diff --git a/src/video_core/fence_manager.h b/src/video_core/fence_manager.h
index c390ac91b..3b2f6aab6 100644
--- a/src/video_core/fence_manager.h
+++ b/src/video_core/fence_manager.h
@@ -4,13 +4,20 @@
 #pragma once
 
 #include <algorithm>
+#include <condition_variable>
 #include <cstring>
 #include <deque>
 #include <functional>
 #include <memory>
+#include <mutex>
+#include <thread>
 #include <queue>
 
 #include "common/common_types.h"
+#include "common/microprofile.h"
+#include "common/scope_exit.h"
+#include "common/settings.h"
+#include "common/thread.h"
 #include "video_core/delayed_destruction_ring.h"
 #include "video_core/gpu.h"
 #include "video_core/host1x/host1x.h"
@@ -23,15 +30,26 @@ class FenceBase {
 public:
     explicit FenceBase(bool is_stubbed_) : is_stubbed{is_stubbed_} {}
 
+    bool IsStubbed() const {
+        return is_stubbed;
+    }
+
 protected:
     bool is_stubbed;
 };
 
-template <typename TFence, typename TTextureCache, typename TTBufferCache, typename TQueryCache>
+template <typename Traits>
 class FenceManager {
+    using TFence = typename Traits::FenceType;
+    using TTextureCache = typename Traits::TextureCacheType;
+    using TBufferCache = typename Traits::BufferCacheType;
+    using TQueryCache = typename Traits::QueryCacheType;
+    static constexpr bool can_async_check = Traits::HAS_ASYNC_CHECK;
+
 public:
     /// Notify the fence manager about a new frame
     void TickFrame() {
+        std::unique_lock lock(ring_guard);
         delayed_destruction_ring.Tick();
     }
 
@@ -46,17 +64,33 @@ public:
     }
 
     void SignalFence(std::function<void()>&& func) {
-        TryReleasePendingFences();
+        rasterizer.InvalidateGPUCache();
+        bool delay_fence = Settings::IsGPULevelHigh();
+        if constexpr (!can_async_check) {
+            TryReleasePendingFences<false>();
+        }
         const bool should_flush = ShouldFlush();
         CommitAsyncFlushes();
-        uncommitted_operations.emplace_back(std::move(func));
-        CommitOperations();
         TFence new_fence = CreateFence(!should_flush);
-        fences.push(new_fence);
+        if constexpr (can_async_check) {
+            guard.lock();
+        }
+        if (delay_fence) {
+            uncommitted_operations.emplace_back(std::move(func));
+        }
+        pending_operations.emplace_back(std::move(uncommitted_operations));
         QueueFence(new_fence);
+        if (!delay_fence) {
+            func();
+        }
+        fences.push(std::move(new_fence));
         if (should_flush) {
             rasterizer.FlushCommands();
         }
+        if constexpr (can_async_check) {
+            guard.unlock();
+            cv.notify_all();
+        }
     }
 
     void SignalSyncPoint(u32 value) {
@@ -66,29 +100,30 @@ public:
     }
 
     void WaitPendingFences() {
-        while (!fences.empty()) {
-            TFence& current_fence = fences.front();
-            if (ShouldWait()) {
-                WaitFence(current_fence);
-            }
-            PopAsyncFlushes();
-            auto operations = std::move(pending_operations.front());
-            pending_operations.pop_front();
-            for (auto& operation : operations) {
-                operation();
-            }
-            PopFence();
+        if constexpr (!can_async_check) {
+            TryReleasePendingFences<true>();
         }
     }
 
 protected:
     explicit FenceManager(VideoCore::RasterizerInterface& rasterizer_, Tegra::GPU& gpu_,
-                          TTextureCache& texture_cache_, TTBufferCache& buffer_cache_,
+                          TTextureCache& texture_cache_, TBufferCache& buffer_cache_,
                           TQueryCache& query_cache_)
         : rasterizer{rasterizer_}, gpu{gpu_}, syncpoint_manager{gpu.Host1x().GetSyncpointManager()},
-          texture_cache{texture_cache_}, buffer_cache{buffer_cache_}, query_cache{query_cache_} {}
+          texture_cache{texture_cache_}, buffer_cache{buffer_cache_}, query_cache{query_cache_} {
+        if constexpr (can_async_check) {
+            fence_thread =
+                std::jthread([this](std::stop_token token) { ReleaseThreadFunc(token); });
+        }
+    }
 
-    virtual ~FenceManager() = default;
+    virtual ~FenceManager() {
+        if constexpr (can_async_check) {
+            fence_thread.request_stop();
+            cv.notify_all();
+            fence_thread.join();
+        }
+    }
 
     /// Creates a Fence Interface, does not create a backend fence if 'is_stubbed' is
     /// true
@@ -104,15 +139,20 @@ protected:
     Tegra::GPU& gpu;
     Tegra::Host1x::SyncpointManager& syncpoint_manager;
     TTextureCache& texture_cache;
-    TTBufferCache& buffer_cache;
+    TBufferCache& buffer_cache;
     TQueryCache& query_cache;
 
 private:
+    template <bool force_wait>
     void TryReleasePendingFences() {
         while (!fences.empty()) {
             TFence& current_fence = fences.front();
             if (ShouldWait() && !IsFenceSignaled(current_fence)) {
-                return;
+                if constexpr (force_wait) {
+                    WaitFence(current_fence);
+                } else {
+                    return;
+                }
             }
             PopAsyncFlushes();
             auto operations = std::move(pending_operations.front());
@@ -120,7 +160,49 @@ private:
             for (auto& operation : operations) {
                 operation();
             }
-            PopFence();
+            {
+                std::unique_lock lock(ring_guard);
+                delayed_destruction_ring.Push(std::move(current_fence));
+            }
+            fences.pop();
+        }
+    }
+
+    void ReleaseThreadFunc(std::stop_token stop_token) {
+        std::string name = "GPUFencingThread";
+        MicroProfileOnThreadCreate(name.c_str());
+
+        // Cleanup
+        SCOPE_EXIT({ MicroProfileOnThreadExit(); });
+
+        Common::SetCurrentThreadName(name.c_str());
+        Common::SetCurrentThreadPriority(Common::ThreadPriority::High);
+
+        TFence current_fence;
+        std::deque<std::function<void()>> current_operations;
+        while (!stop_token.stop_requested()) {
+            {
+                std::unique_lock lock(guard);
+                cv.wait(lock, [&] { return stop_token.stop_requested() || !fences.empty(); });
+                if (stop_token.stop_requested()) [[unlikely]] {
+                    return;
+                }
+                current_fence = std::move(fences.front());
+                current_operations = std::move(pending_operations.front());
+                fences.pop();
+                pending_operations.pop_front();
+            }
+            if (!current_fence->IsStubbed()) {
+                WaitFence(current_fence);
+            }
+            PopAsyncFlushes();
+            for (auto& operation : current_operations) {
+                operation();
+            }
+            {
+                std::unique_lock lock(ring_guard);
+                delayed_destruction_ring.Push(std::move(current_fence));
+            }
         }
     }
 
@@ -154,19 +236,16 @@ private:
         query_cache.CommitAsyncFlushes();
     }
 
-    void PopFence() {
-        delayed_destruction_ring.Push(std::move(fences.front()));
-        fences.pop();
-    }
-
-    void CommitOperations() {
-        pending_operations.emplace_back(std::move(uncommitted_operations));
-    }
-
     std::queue<TFence> fences;
     std::deque<std::function<void()>> uncommitted_operations;
     std::deque<std::deque<std::function<void()>>> pending_operations;
 
+    std::mutex guard;
+    std::mutex ring_guard;
+    std::condition_variable cv;
+
+    std::jthread fence_thread;
+
     DelayedDestructionRing<TFence, 6> delayed_destruction_ring;
 };
 
diff --git a/src/video_core/memory_manager.cpp b/src/video_core/memory_manager.cpp
index 01fb5b546..e06ce5d14 100644
--- a/src/video_core/memory_manager.cpp
+++ b/src/video_core/memory_manager.cpp
@@ -170,6 +170,7 @@ void MemoryManager::BindRasterizer(VideoCore::RasterizerInterface* rasterizer_)
 
 GPUVAddr MemoryManager::Map(GPUVAddr gpu_addr, VAddr cpu_addr, std::size_t size, PTEKind kind,
                             bool is_big_pages) {
+    std::unique_lock<std::mutex> lock(guard);
     if (is_big_pages) [[likely]] {
         return BigPageTableOp<EntryType::Mapped>(gpu_addr, cpu_addr, size, kind);
     }
@@ -177,6 +178,7 @@ GPUVAddr MemoryManager::Map(GPUVAddr gpu_addr, VAddr cpu_addr, std::size_t size,
 }
 
 GPUVAddr MemoryManager::MapSparse(GPUVAddr gpu_addr, std::size_t size, bool is_big_pages) {
+    std::unique_lock<std::mutex> lock(guard);
     if (is_big_pages) [[likely]] {
         return BigPageTableOp<EntryType::Reserved>(gpu_addr, 0, size, PTEKind::INVALID);
     }
@@ -187,6 +189,7 @@ void MemoryManager::Unmap(GPUVAddr gpu_addr, std::size_t size) {
     if (size == 0) {
         return;
     }
+    std::unique_lock<std::mutex> lock(guard);
     GetSubmappedRangeImpl<false>(gpu_addr, size, page_stash);
 
     for (const auto& [map_addr, map_size] : page_stash) {
@@ -553,6 +556,7 @@ size_t MemoryManager::MaxContinuousRange(GPUVAddr gpu_addr, size_t size) const {
 }
 
 size_t MemoryManager::GetMemoryLayoutSize(GPUVAddr gpu_addr, size_t max_size) const {
+    std::unique_lock<std::mutex> lock(guard);
     return kind_map.GetContinuousSizeFrom(gpu_addr);
 }
 
@@ -745,10 +749,10 @@ void MemoryManager::FlushCaching() {
         return;
     }
     accumulator->Callback([this](GPUVAddr addr, size_t size) {
-        GetSubmappedRangeImpl<false>(addr, size, page_stash);
+        GetSubmappedRangeImpl<false>(addr, size, page_stash2);
     });
-    rasterizer->InnerInvalidation(page_stash);
-    page_stash.clear();
+    rasterizer->InnerInvalidation(page_stash2);
+    page_stash2.clear();
     accumulator->Clear();
 }
 
diff --git a/src/video_core/memory_manager.h b/src/video_core/memory_manager.h
index fbbe856c4..794535122 100644
--- a/src/video_core/memory_manager.h
+++ b/src/video_core/memory_manager.h
@@ -5,6 +5,7 @@
 
 #include <atomic>
 #include <map>
+#include <mutex>
 #include <optional>
 #include <vector>
 
@@ -215,6 +216,9 @@ private:
 
     std::vector<u64> big_page_continuous;
     std::vector<std::pair<VAddr, std::size_t>> page_stash{};
+    std::vector<std::pair<VAddr, std::size_t>> page_stash2{};
+
+    mutable std::mutex guard;
 
     static constexpr size_t continuous_bits = 64;
 
diff --git a/src/video_core/query_cache.h b/src/video_core/query_cache.h
index 8906ba6d8..941de95c1 100644
--- a/src/video_core/query_cache.h
+++ b/src/video_core/query_cache.h
@@ -6,6 +6,7 @@
 #include <algorithm>
 #include <array>
 #include <cstring>
+#include <functional>
 #include <iterator>
 #include <list>
 #include <memory>
@@ -17,13 +18,19 @@
 
 #include "common/assert.h"
 #include "common/settings.h"
+#include "core/memory.h"
 #include "video_core/control/channel_state_cache.h"
 #include "video_core/engines/maxwell_3d.h"
 #include "video_core/memory_manager.h"
 #include "video_core/rasterizer_interface.h"
+#include "video_core/texture_cache/slot_vector.h"
 
 namespace VideoCommon {
 
+using AsyncJobId = SlotId;
+
+static constexpr AsyncJobId NULL_ASYNC_JOB_ID{0};
+
 template <class QueryCache, class HostCounter>
 class CounterStreamBase {
 public:
@@ -93,9 +100,13 @@ private:
 template <class QueryCache, class CachedQuery, class CounterStream, class HostCounter>
 class QueryCacheBase : public VideoCommon::ChannelSetupCaches<VideoCommon::ChannelInfo> {
 public:
-    explicit QueryCacheBase(VideoCore::RasterizerInterface& rasterizer_)
-        : rasterizer{rasterizer_}, streams{{CounterStream{static_cast<QueryCache&>(*this),
-                                                          VideoCore::QueryType::SamplesPassed}}} {}
+    explicit QueryCacheBase(VideoCore::RasterizerInterface& rasterizer_,
+                            Core::Memory::Memory& cpu_memory_)
+        : rasterizer{rasterizer_},
+          cpu_memory{cpu_memory_}, streams{{CounterStream{static_cast<QueryCache&>(*this),
+                                                          VideoCore::QueryType::SamplesPassed}}} {
+        (void)slot_async_jobs.insert(); // Null value
+    }
 
     void InvalidateRegion(VAddr addr, std::size_t size) {
         std::unique_lock lock{mutex};
@@ -126,10 +137,15 @@ public:
             query = Register(type, *cpu_addr, host_ptr, timestamp.has_value());
         }
 
-        query->BindCounter(Stream(type).Current(), timestamp);
-        if (Settings::values.use_asynchronous_gpu_emulation.GetValue()) {
-            AsyncFlushQuery(*cpu_addr);
+        auto result = query->BindCounter(Stream(type).Current(), timestamp);
+        if (result) {
+            auto async_job_id = query->GetAsyncJob();
+            auto& async_job = slot_async_jobs[async_job_id];
+            async_job.collected = true;
+            async_job.value = *result;
+            query->SetAsyncJob(NULL_ASYNC_JOB_ID);
         }
+        AsyncFlushQuery(query, timestamp, lock);
     }
 
     /// Updates counters from GPU state. Expected to be called once per draw, clear or dispatch.
@@ -173,15 +189,18 @@ public:
     }
 
     void CommitAsyncFlushes() {
+        std::unique_lock lock{mutex};
         committed_flushes.push_back(uncommitted_flushes);
         uncommitted_flushes.reset();
     }
 
     bool HasUncommittedFlushes() const {
+        std::unique_lock lock{mutex};
         return uncommitted_flushes != nullptr;
     }
 
     bool ShouldWaitAsyncFlushes() const {
+        std::unique_lock lock{mutex};
         if (committed_flushes.empty()) {
             return false;
         }
@@ -189,6 +208,7 @@ public:
     }
 
     void PopAsyncFlushes() {
+        std::unique_lock lock{mutex};
         if (committed_flushes.empty()) {
             return;
         }
@@ -197,15 +217,25 @@ public:
             committed_flushes.pop_front();
             return;
         }
-        for (VAddr query_address : *flush_list) {
-            FlushAndRemoveRegion(query_address, 4);
+        for (AsyncJobId async_job_id : *flush_list) {
+            AsyncJob& async_job = slot_async_jobs[async_job_id];
+            if (!async_job.collected) {
+                FlushAndRemoveRegion(async_job.query_location, 2, true);
+            }
         }
         committed_flushes.pop_front();
     }
 
 private:
+    struct AsyncJob {
+        bool collected = false;
+        u64 value = 0;
+        VAddr query_location = 0;
+        std::optional<u64> timestamp{};
+    };
+
     /// Flushes a memory range to guest memory and removes it from the cache.
-    void FlushAndRemoveRegion(VAddr addr, std::size_t size) {
+    void FlushAndRemoveRegion(VAddr addr, std::size_t size, bool async = false) {
         const u64 addr_begin = addr;
         const u64 addr_end = addr_begin + size;
         const auto in_range = [addr_begin, addr_end](const CachedQuery& query) {
@@ -226,7 +256,16 @@ private:
                     continue;
                 }
                 rasterizer.UpdatePagesCachedCount(query.GetCpuAddr(), query.SizeInBytes(), -1);
-                query.Flush();
+                AsyncJobId async_job_id = query.GetAsyncJob();
+                auto flush_result = query.Flush(async);
+                if (async_job_id == NULL_ASYNC_JOB_ID) {
+                    ASSERT_MSG(false, "This should not be reachable at all");
+                    continue;
+                }
+                AsyncJob& async_job = slot_async_jobs[async_job_id];
+                async_job.collected = true;
+                async_job.value = flush_result;
+                query.SetAsyncJob(NULL_ASYNC_JOB_ID);
             }
             std::erase_if(contents, in_range);
         }
@@ -253,26 +292,60 @@ private:
         return found != std::end(contents) ? &*found : nullptr;
     }
 
-    void AsyncFlushQuery(VAddr addr) {
-        if (!uncommitted_flushes) {
-            uncommitted_flushes = std::make_shared<std::vector<VAddr>>();
+    void AsyncFlushQuery(CachedQuery* query, std::optional<u64> timestamp,
+                         std::unique_lock<std::recursive_mutex>& lock) {
+        const AsyncJobId new_async_job_id = slot_async_jobs.insert();
+        {
+            AsyncJob& async_job = slot_async_jobs[new_async_job_id];
+            query->SetAsyncJob(new_async_job_id);
+            async_job.query_location = query->GetCpuAddr();
+            async_job.collected = false;
+
+            if (!uncommitted_flushes) {
+                uncommitted_flushes = std::make_shared<std::vector<AsyncJobId>>();
+            }
+            uncommitted_flushes->push_back(new_async_job_id);
         }
-        uncommitted_flushes->push_back(addr);
+        lock.unlock();
+        std::function<void()> operation([this, new_async_job_id, timestamp] {
+            std::unique_lock local_lock{mutex};
+            AsyncJob& async_job = slot_async_jobs[new_async_job_id];
+            u64 value = async_job.value;
+            VAddr address = async_job.query_location;
+            slot_async_jobs.erase(new_async_job_id);
+            local_lock.unlock();
+            if (timestamp) {
+                u64 timestamp_value = *timestamp;
+                cpu_memory.WriteBlockUnsafe(address + sizeof(u64), &timestamp_value, sizeof(u64));
+                cpu_memory.WriteBlockUnsafe(address, &value, sizeof(u64));
+                rasterizer.InvalidateRegion(address, sizeof(u64) * 2,
+                                            VideoCommon::CacheType::NoQueryCache);
+            } else {
+                u32 small_value = static_cast<u32>(value);
+                cpu_memory.WriteBlockUnsafe(address, &small_value, sizeof(u32));
+                rasterizer.InvalidateRegion(address, sizeof(u32),
+                                            VideoCommon::CacheType::NoQueryCache);
+            }
+        });
+        rasterizer.SyncOperation(std::move(operation));
     }
 
     static constexpr std::uintptr_t YUZU_PAGESIZE = 4096;
     static constexpr unsigned YUZU_PAGEBITS = 12;
 
+    SlotVector<AsyncJob> slot_async_jobs;
+
     VideoCore::RasterizerInterface& rasterizer;
+    Core::Memory::Memory& cpu_memory;
 
-    std::recursive_mutex mutex;
+    mutable std::recursive_mutex mutex;
 
     std::unordered_map<u64, std::vector<CachedQuery>> cached_queries;
 
     std::array<CounterStream, VideoCore::NumQueryTypes> streams;
 
-    std::shared_ptr<std::vector<VAddr>> uncommitted_flushes{};
-    std::list<std::shared_ptr<std::vector<VAddr>>> committed_flushes;
+    std::shared_ptr<std::vector<AsyncJobId>> uncommitted_flushes{};
+    std::list<std::shared_ptr<std::vector<AsyncJobId>>> committed_flushes;
 };
 
 template <class QueryCache, class HostCounter>
@@ -291,12 +364,12 @@ public:
     virtual ~HostCounterBase() = default;
 
     /// Returns the current value of the query.
-    u64 Query() {
+    u64 Query(bool async = false) {
         if (result) {
             return *result;
         }
 
-        u64 value = BlockingQuery() + base_result;
+        u64 value = BlockingQuery(async) + base_result;
         if (dependency) {
             value += dependency->Query();
             dependency = nullptr;
@@ -317,7 +390,7 @@ public:
 
 protected:
     /// Returns the value of query from the backend API blocking as needed.
-    virtual u64 BlockingQuery() const = 0;
+    virtual u64 BlockingQuery(bool async = false) const = 0;
 
 private:
     std::shared_ptr<HostCounter> dependency; ///< Counter to add to this value.
@@ -340,26 +413,33 @@ public:
     CachedQueryBase& operator=(const CachedQueryBase&) = delete;
 
     /// Flushes the query to guest memory.
-    virtual void Flush() {
+    virtual u64 Flush(bool async = false) {
         // When counter is nullptr it means that it's just been reset. We are supposed to write a
         // zero in these cases.
-        const u64 value = counter ? counter->Query() : 0;
+        const u64 value = counter ? counter->Query(async) : 0;
+        if (async) {
+            return value;
+        }
         std::memcpy(host_ptr, &value, sizeof(u64));
 
         if (timestamp) {
             std::memcpy(host_ptr + TIMESTAMP_OFFSET, &*timestamp, sizeof(u64));
         }
+        return value;
     }
 
     /// Binds a counter to this query.
-    void BindCounter(std::shared_ptr<HostCounter> counter_, std::optional<u64> timestamp_) {
+    std::optional<u64> BindCounter(std::shared_ptr<HostCounter> counter_,
+                                   std::optional<u64> timestamp_) {
+        std::optional<u64> result{};
         if (counter) {
             // If there's an old counter set it means the query is being rewritten by the game.
             // To avoid losing the data forever, flush here.
-            Flush();
+            result = std::make_optional(Flush());
         }
         counter = std::move(counter_);
         timestamp = timestamp_;
+        return result;
     }
 
     VAddr GetCpuAddr() const noexcept {
@@ -374,6 +454,14 @@ public:
         return with_timestamp ? LARGE_QUERY_SIZE : SMALL_QUERY_SIZE;
     }
 
+    void SetAsyncJob(AsyncJobId assigned_async_job_) {
+        assigned_async_job = assigned_async_job_;
+    }
+
+    AsyncJobId GetAsyncJob() const {
+        return assigned_async_job;
+    }
+
 protected:
     /// Returns true when querying the counter may potentially block.
     bool WaitPending() const noexcept {
@@ -389,6 +477,7 @@ private:
     u8* host_ptr;                         ///< Writable host pointer.
     std::shared_ptr<HostCounter> counter; ///< Host counter to query, owns the dependency tree.
     std::optional<u64> timestamp;         ///< Timestamp to flush to guest memory.
+    AsyncJobId assigned_async_job;
 };
 
 } // namespace VideoCommon
diff --git a/src/video_core/renderer_opengl/gl_fence_manager.h b/src/video_core/renderer_opengl/gl_fence_manager.h
index f1446e732..e21b19dcc 100644
--- a/src/video_core/renderer_opengl/gl_fence_manager.h
+++ b/src/video_core/renderer_opengl/gl_fence_manager.h
@@ -30,7 +30,17 @@ private:
 };
 
 using Fence = std::shared_ptr<GLInnerFence>;
-using GenericFenceManager = VideoCommon::FenceManager<Fence, TextureCache, BufferCache, QueryCache>;
+
+struct FenceManagerParams {
+    using FenceType = Fence;
+    using BufferCacheType = BufferCache;
+    using TextureCacheType = TextureCache;
+    using QueryCacheType = QueryCache;
+
+    static constexpr bool HAS_ASYNC_CHECK = false;
+};
+
+using GenericFenceManager = VideoCommon::FenceManager<FenceManagerParams>;
 
 class FenceManagerOpenGL final : public GenericFenceManager {
 public:
diff --git a/src/video_core/renderer_opengl/gl_query_cache.cpp b/src/video_core/renderer_opengl/gl_query_cache.cpp
index 5070db441..99d7347f5 100644
--- a/src/video_core/renderer_opengl/gl_query_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_query_cache.cpp
@@ -26,8 +26,8 @@ constexpr GLenum GetTarget(VideoCore::QueryType type) {
 
 } // Anonymous namespace
 
-QueryCache::QueryCache(RasterizerOpenGL& rasterizer_)
-    : QueryCacheBase(rasterizer_), gl_rasterizer{rasterizer_} {}
+QueryCache::QueryCache(RasterizerOpenGL& rasterizer_, Core::Memory::Memory& cpu_memory_)
+    : QueryCacheBase(rasterizer_, cpu_memory_), gl_rasterizer{rasterizer_} {}
 
 QueryCache::~QueryCache() = default;
 
@@ -74,7 +74,7 @@ void HostCounter::EndQuery() {
     glEndQuery(GetTarget(type));
 }
 
-u64 HostCounter::BlockingQuery() const {
+u64 HostCounter::BlockingQuery([[maybe_unused]] bool async) const {
     GLint64 value;
     glGetQueryObjecti64v(query.handle, GL_QUERY_RESULT, &value);
     return static_cast<u64>(value);
@@ -96,7 +96,7 @@ CachedQuery& CachedQuery::operator=(CachedQuery&& rhs) noexcept {
     return *this;
 }
 
-void CachedQuery::Flush() {
+u64 CachedQuery::Flush([[maybe_unused]] bool async) {
     // Waiting for a query while another query of the same target is enabled locks Nvidia's driver.
     // To avoid this disable and re-enable keeping the dependency stream.
     // But we only have to do this if we have pending waits to be done.
@@ -106,11 +106,13 @@ void CachedQuery::Flush() {
         stream.Update(false);
     }
 
-    VideoCommon::CachedQueryBase<HostCounter>::Flush();
+    auto result = VideoCommon::CachedQueryBase<HostCounter>::Flush();
 
     if (slice_counter) {
         stream.Update(true);
     }
+
+    return result;
 }
 
 } // namespace OpenGL
diff --git a/src/video_core/renderer_opengl/gl_query_cache.h b/src/video_core/renderer_opengl/gl_query_cache.h
index 14ce59990..872513f22 100644
--- a/src/video_core/renderer_opengl/gl_query_cache.h
+++ b/src/video_core/renderer_opengl/gl_query_cache.h
@@ -28,7 +28,7 @@ using CounterStream = VideoCommon::CounterStreamBase<QueryCache, HostCounter>;
 class QueryCache final
     : public VideoCommon::QueryCacheBase<QueryCache, CachedQuery, CounterStream, HostCounter> {
 public:
-    explicit QueryCache(RasterizerOpenGL& rasterizer_);
+    explicit QueryCache(RasterizerOpenGL& rasterizer_, Core::Memory::Memory& cpu_memory_);
     ~QueryCache();
 
     OGLQuery AllocateQuery(VideoCore::QueryType type);
@@ -51,7 +51,7 @@ public:
     void EndQuery();
 
 private:
-    u64 BlockingQuery() const override;
+    u64 BlockingQuery(bool async = false) const override;
 
     QueryCache& cache;
     const VideoCore::QueryType type;
@@ -70,7 +70,7 @@ public:
     CachedQuery(const CachedQuery&) = delete;
     CachedQuery& operator=(const CachedQuery&) = delete;
 
-    void Flush() override;
+    u64 Flush(bool async = false) override;
 
 private:
     QueryCache* cache;
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp
index 4993d4709..0089b4b27 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer.cpp
+++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp
@@ -63,7 +63,7 @@ RasterizerOpenGL::RasterizerOpenGL(Core::Frontend::EmuWindow& emu_window_, Tegra
       buffer_cache(*this, cpu_memory_, buffer_cache_runtime),
       shader_cache(*this, emu_window_, device, texture_cache, buffer_cache, program_manager,
                    state_tracker, gpu.ShaderNotify()),
-      query_cache(*this), accelerate_dma(buffer_cache, texture_cache),
+      query_cache(*this, cpu_memory_), accelerate_dma(buffer_cache, texture_cache),
       fence_manager(*this, gpu, texture_cache, buffer_cache, query_cache),
       blit_image(program_manager_) {}
 
diff --git a/src/video_core/renderer_vulkan/vk_fence_manager.cpp b/src/video_core/renderer_vulkan/vk_fence_manager.cpp
index 0214b103a..fad9e3832 100644
--- a/src/video_core/renderer_vulkan/vk_fence_manager.cpp
+++ b/src/video_core/renderer_vulkan/vk_fence_manager.cpp
@@ -5,6 +5,7 @@
 
 #include "video_core/renderer_vulkan/vk_buffer_cache.h"
 #include "video_core/renderer_vulkan/vk_fence_manager.h"
+#include "video_core/renderer_vulkan/vk_query_cache.h"
 #include "video_core/renderer_vulkan/vk_scheduler.h"
 #include "video_core/renderer_vulkan/vk_texture_cache.h"
 #include "video_core/vulkan_common/vulkan_device.h"
diff --git a/src/video_core/renderer_vulkan/vk_fence_manager.h b/src/video_core/renderer_vulkan/vk_fence_manager.h
index 7fe2afcd9..145359d4e 100644
--- a/src/video_core/renderer_vulkan/vk_fence_manager.h
+++ b/src/video_core/renderer_vulkan/vk_fence_manager.h
@@ -40,7 +40,16 @@ private:
 };
 using Fence = std::shared_ptr<InnerFence>;
 
-using GenericFenceManager = VideoCommon::FenceManager<Fence, TextureCache, BufferCache, QueryCache>;
+struct FenceManagerParams {
+    using FenceType = Fence;
+    using BufferCacheType = BufferCache;
+    using TextureCacheType = TextureCache;
+    using QueryCacheType = QueryCache;
+
+    static constexpr bool HAS_ASYNC_CHECK = true;
+};
+
+using GenericFenceManager = VideoCommon::FenceManager<FenceManagerParams>;
 
 class FenceManager final : public GenericFenceManager {
 public:
diff --git a/src/video_core/renderer_vulkan/vk_query_cache.cpp b/src/video_core/renderer_vulkan/vk_query_cache.cpp
index 929c8ece6..d67490449 100644
--- a/src/video_core/renderer_vulkan/vk_query_cache.cpp
+++ b/src/video_core/renderer_vulkan/vk_query_cache.cpp
@@ -66,9 +66,10 @@ void QueryPool::Reserve(std::pair<VkQueryPool, u32> query) {
     }
 }
 
-QueryCache::QueryCache(VideoCore::RasterizerInterface& rasterizer_, const Device& device_,
+QueryCache::QueryCache(VideoCore::RasterizerInterface& rasterizer_,
+                       Core::Memory::Memory& cpu_memory_, const Device& device_,
                        Scheduler& scheduler_)
-    : QueryCacheBase{rasterizer_}, device{device_}, scheduler{scheduler_},
+    : QueryCacheBase{rasterizer_, cpu_memory_}, device{device_}, scheduler{scheduler_},
       query_pools{
           QueryPool{device_, scheduler_, QueryType::SamplesPassed},
       } {}
@@ -98,8 +99,10 @@ HostCounter::HostCounter(QueryCache& cache_, std::shared_ptr<HostCounter> depend
       query{cache_.AllocateQuery(type_)}, tick{cache_.GetScheduler().CurrentTick()} {
     const vk::Device* logical = &cache.GetDevice().GetLogical();
     cache.GetScheduler().Record([logical, query = query](vk::CommandBuffer cmdbuf) {
+        const bool use_precise = Settings::IsGPULevelHigh();
         logical->ResetQueryPool(query.first, query.second, 1);
-        cmdbuf.BeginQuery(query.first, query.second, VK_QUERY_CONTROL_PRECISE_BIT);
+        cmdbuf.BeginQuery(query.first, query.second,
+                          use_precise ? VK_QUERY_CONTROL_PRECISE_BIT : 0);
     });
 }
 
@@ -112,8 +115,10 @@ void HostCounter::EndQuery() {
         [query = query](vk::CommandBuffer cmdbuf) { cmdbuf.EndQuery(query.first, query.second); });
 }
 
-u64 HostCounter::BlockingQuery() const {
-    cache.GetScheduler().Wait(tick);
+u64 HostCounter::BlockingQuery(bool async) const {
+    if (!async) {
+        cache.GetScheduler().Wait(tick);
+    }
     u64 data;
     const VkResult query_result = cache.GetDevice().GetLogical().GetQueryResults(
         query.first, query.second, 1, sizeof(data), &data, sizeof(data),
diff --git a/src/video_core/renderer_vulkan/vk_query_cache.h b/src/video_core/renderer_vulkan/vk_query_cache.h
index 26762ee09..c1b9552eb 100644
--- a/src/video_core/renderer_vulkan/vk_query_cache.h
+++ b/src/video_core/renderer_vulkan/vk_query_cache.h
@@ -52,7 +52,8 @@ private:
 class QueryCache final
     : public VideoCommon::QueryCacheBase<QueryCache, CachedQuery, CounterStream, HostCounter> {
 public:
-    explicit QueryCache(VideoCore::RasterizerInterface& rasterizer_, const Device& device_,
+    explicit QueryCache(VideoCore::RasterizerInterface& rasterizer_,
+                        Core::Memory::Memory& cpu_memory_, const Device& device_,
                         Scheduler& scheduler_);
     ~QueryCache();
 
@@ -83,7 +84,7 @@ public:
     void EndQuery();
 
 private:
-    u64 BlockingQuery() const override;
+    u64 BlockingQuery(bool async = false) const override;
 
     QueryCache& cache;
     const VideoCore::QueryType type;
diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp
index 2559a3aa7..d1489fc95 100644
--- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp
+++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp
@@ -172,7 +172,8 @@ RasterizerVulkan::RasterizerVulkan(Core::Frontend::EmuWindow& emu_window_, Tegra
       buffer_cache(*this, cpu_memory_, buffer_cache_runtime),
       pipeline_cache(*this, device, scheduler, descriptor_pool, update_descriptor_queue,
                      render_pass_cache, buffer_cache, texture_cache, gpu.ShaderNotify()),
-      query_cache{*this, device, scheduler}, accelerate_dma(buffer_cache, texture_cache, scheduler),
+      query_cache{*this, cpu_memory_, device, scheduler},
+      accelerate_dma(buffer_cache, texture_cache, scheduler),
       fence_manager(*this, gpu, texture_cache, buffer_cache, query_cache, device, scheduler),
       wfi_event(device.GetLogical().CreateEvent()) {
     scheduler.SetQueryCache(query_cache);
@@ -675,7 +676,8 @@ bool RasterizerVulkan::AccelerateConditionalRendering() {
     const GPUVAddr condition_address{maxwell3d->regs.render_enable.Address()};
     Maxwell::ReportSemaphore::Compare cmp;
     if (gpu_memory->IsMemoryDirty(condition_address, sizeof(cmp),
-                                  VideoCommon::CacheType::BufferCache)) {
+                                  VideoCommon::CacheType::BufferCache |
+                                      VideoCommon::CacheType::QueryCache)) {
         return true;
     }
     return false;