6 files changed, 39 insertions, 14 deletions

diff --git a/src/core/core.cpp b/src/core/core.cpp
index aa96f709b..3f9a7f44b 100644
--- a/src/core/core.cpp
+++ b/src/core/core.cpp
@@ -317,6 +317,8 @@ struct System::Impl {
         is_powered_on = false;
         exit_lock = false;
 
+        gpu_core->NotifyShutdown();
+
         services.reset();
         service_manager.reset();
         cheat_engine.reset();
diff --git a/src/core/hle/service/nvflinger/buffer_queue.cpp b/src/core/hle/service/nvflinger/buffer_queue.cpp
index b4c3a6099..5fead6d1b 100644
--- a/src/core/hle/service/nvflinger/buffer_queue.cpp
+++ b/src/core/hle/service/nvflinger/buffer_queue.cpp
@@ -186,6 +186,10 @@ u32 BufferQueue::Query(QueryType type) {
     case QueryType::NativeWindowWidth:
     case QueryType::NativeWindowHeight:
         break;
+    case QueryType::NativeWindowMinUndequeuedBuffers:
+        return 0;
+    case QueryType::NativeWindowConsumerUsageBits:
+        return 0;
     }
     UNIMPLEMENTED_MSG("Unimplemented query type={}", type);
     return 0;
diff --git a/src/core/hle/service/nvflinger/buffer_queue.h b/src/core/hle/service/nvflinger/buffer_queue.h
index 7b7baeaea..f2a579133 100644
--- a/src/core/hle/service/nvflinger/buffer_queue.h
+++ b/src/core/hle/service/nvflinger/buffer_queue.h
@@ -57,6 +57,11 @@ public:
         NativeWindowWidth = 0,
         NativeWindowHeight = 1,
         NativeWindowFormat = 2,
+        /// The minimum number of buffers that must remain un-dequeued after a buffer has been
+        /// queued
+        NativeWindowMinUndequeuedBuffers = 3,
+        /// The consumer gralloc usage bits currently set by the consumer
+        NativeWindowConsumerUsageBits = 10,
     };
 
     explicit BufferQueue(Kernel::KernelCore& kernel, u32 id_, u64 layer_id_,
diff --git a/src/core/hle/service/nvflinger/nvflinger.cpp b/src/core/hle/service/nvflinger/nvflinger.cpp
index 396cc5afa..01e69de30 100644
--- a/src/core/hle/service/nvflinger/nvflinger.cpp
+++ b/src/core/hle/service/nvflinger/nvflinger.cpp
@@ -100,9 +100,6 @@ std::optional<u64> NVFlinger::OpenDisplay(std::string_view name) {
 
     LOG_DEBUG(Service, "Opening \"{}\" display", name);
 
-    // TODO(Subv): Currently we only support the Default display.
-    ASSERT(name == "Default");
-
     const auto itr =
         std::find_if(displays.begin(), displays.end(),
                      [&](const VI::Display& display) { return display.GetName() == name; });
@@ -266,11 +263,10 @@ void NVFlinger::Compose() {
 
         auto& gpu = system.GPU();
         const auto& multi_fence = buffer->get().multi_fence;
-        const auto stop_token = vsync_thread.get_stop_token();
         guard->unlock();
         for (u32 fence_id = 0; fence_id < multi_fence.num_fences; fence_id++) {
             const auto& fence = multi_fence.fences[fence_id];
-            gpu.WaitFence(fence.id, fence.value, stop_token);
+            gpu.WaitFence(fence.id, fence.value);
         }
         guard->lock();
 
diff --git a/src/video_core/gpu.cpp b/src/video_core/gpu.cpp
index d98874150..705765c99 100644
--- a/src/video_core/gpu.cpp
+++ b/src/video_core/gpu.cpp
@@ -17,7 +17,6 @@
 #include "core/frontend/emu_window.h"
 #include "core/hardware_interrupt_manager.h"
 #include "core/hle/service/nvdrv/nvdata.h"
-#include "core/hle/service/nvflinger/buffer_queue.h"
 #include "core/perf_stats.h"
 #include "video_core/cdma_pusher.h"
 #include "video_core/dma_pusher.h"
@@ -206,7 +205,7 @@ struct GPU::Impl {
     }
 
     /// Allows the CPU/NvFlinger to wait on the GPU before presenting a frame.
-    void WaitFence(u32 syncpoint_id, u32 value, std::stop_token stop_token = {}) {
+    void WaitFence(u32 syncpoint_id, u32 value) {
         // Synced GPU, is always in sync
         if (!is_async) {
             return;
@@ -218,8 +217,13 @@ struct GPU::Impl {
         }
         MICROPROFILE_SCOPE(GPU_wait);
         std::unique_lock lock{sync_mutex};
-        sync_cv.wait(lock, stop_token,
-                     [=, this] { return syncpoints.at(syncpoint_id).load() >= value; });
+        sync_cv.wait(lock, [=, this] {
+            if (shutting_down.load(std::memory_order_relaxed)) {
+                // We're shutting down, ensure no threads continue to wait for the next syncpoint
+                return true;
+            }
+            return syncpoints.at(syncpoint_id).load() >= value;
+        });
     }
 
     void IncrementSyncPoint(u32 syncpoint_id) {
@@ -307,6 +311,12 @@ struct GPU::Impl {
         cpu_context->MakeCurrent();
     }
 
+    void NotifyShutdown() {
+        std::unique_lock lk{sync_mutex};
+        shutting_down.store(true, std::memory_order::relaxed);
+        sync_cv.notify_all();
+    }
+
     /// Obtain the CPU Context
     void ObtainContext() {
         cpu_context->MakeCurrent();
@@ -665,6 +675,8 @@ struct GPU::Impl {
     std::unique_ptr<Engines::KeplerMemory> kepler_memory;
     /// Shader build notifier
     std::unique_ptr<VideoCore::ShaderNotify> shader_notify;
+    /// When true, we are about to shut down emulation session, so terminate outstanding tasks
+    std::atomic_bool shutting_down{};
 
     std::array<std::atomic<u32>, Service::Nvidia::MaxSyncPoints> syncpoints{};
 
@@ -673,7 +685,7 @@ struct GPU::Impl {
     std::mutex sync_mutex;
     std::mutex device_mutex;
 
-    std::condition_variable_any sync_cv;
+    std::condition_variable sync_cv;
 
     struct FlushRequest {
         explicit FlushRequest(u64 fence_, VAddr addr_, std::size_t size_)
@@ -812,8 +824,8 @@ const VideoCore::ShaderNotify& GPU::ShaderNotify() const {
     return impl->ShaderNotify();
 }
 
-void GPU::WaitFence(u32 syncpoint_id, u32 value, std::stop_token stop_token) {
-    impl->WaitFence(syncpoint_id, value, stop_token);
+void GPU::WaitFence(u32 syncpoint_id, u32 value) {
+    impl->WaitFence(syncpoint_id, value);
 }
 
 void GPU::IncrementSyncPoint(u32 syncpoint_id) {
@@ -852,6 +864,10 @@ void GPU::Start() {
     impl->Start();
 }
 
+void GPU::NotifyShutdown() {
+    impl->NotifyShutdown();
+}
+
 void GPU::ObtainContext() {
     impl->ObtainContext();
 }
diff --git a/src/video_core/gpu.h b/src/video_core/gpu.h
index cc65a7870..3188b83ed 100644
--- a/src/video_core/gpu.h
+++ b/src/video_core/gpu.h
@@ -5,7 +5,6 @@
 #pragma once
 
 #include <memory>
-#include <stop_token>
 
 #include "common/bit_field.h"
 #include "common/common_types.h"
@@ -210,7 +209,7 @@ public:
     [[nodiscard]] const VideoCore::ShaderNotify& ShaderNotify() const;
 
     /// Allows the CPU/NvFlinger to wait on the GPU before presenting a frame.
-    void WaitFence(u32 syncpoint_id, u32 value, std::stop_token stop_token = {});
+    void WaitFence(u32 syncpoint_id, u32 value);
 
     void IncrementSyncPoint(u32 syncpoint_id);
 
@@ -233,6 +232,9 @@ public:
     /// core timing events.
     void Start();
 
+    /// Performs any additional necessary steps to shutdown GPU emulation.
+    void NotifyShutdown();
+
     /// Obtain the CPU Context
     void ObtainContext();