service: nvflinger: Improve synchronization for BufferQueue.

- Use proper mechanisms for blocking on DequeueBuffer.
- Ensure service thread terminates on emulation Shutdown.

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

@@ -25,7 +25,12 @@ 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,
@@ -41,10 +46,20 @@ 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 (free_buffers.empty()) {
+    if (!is_connect) {
+        // 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();
 
@@ -97,7 +112,11 @@ 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();
 }
@@ -127,15 +146,28 @@ 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

@@ -4,7 +4,9 @@
 
 #pragma once
 
+#include <condition_variable>
 #include <list>
+#include <mutex>
 #include <optional>
 #include <vector>
 
@@ -99,6 +101,7 @@ 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);
 
@@ -106,18 +109,28 @@ public:
         return id;
     }
 
+    bool IsConnected() const {
+        return is_connect;
+    }
+
     std::shared_ptr<Kernel::WritableEvent> GetWritableBufferWaitEvent() const;
 
     std::shared_ptr<Kernel::ReadableEvent> GetBufferWaitEvent() const;
 
 private:
-    u32 id;
-    u64 layer_id;
+    BufferQueue(const BufferQueue&) = delete;
+
+    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

@@ -88,6 +88,10 @@ 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();
@@ -132,8 +136,9 @@ 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(system.Kernel(), buffer_queue_id, layer_id);
-    display->CreateLayer(layer_id, buffer_queues.back());
+    buffer_queues.emplace_back(
+        std::make_unique<BufferQueue>(system.Kernel(), buffer_queue_id, layer_id));
+    display->CreateLayer(layer_id, *buffer_queues.back());
     return layer_id;
 }
 
@@ -170,13 +175,13 @@ std::shared_ptr<Kernel::ReadableEvent> NVFlinger::FindVsyncEvent(u64 display_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()) {
         return nullptr;
     }
 
-    return &*itr;
+    return itr->get();
 }
 
 VI::Display* NVFlinger::FindDisplay(u64 display_id) {

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

@@ -107,7 +107,7 @@ private:
     std::shared_ptr<Nvidia::Module> nvdrv;
 
     std::vector<VI::Display> displays;
-    std::vector<BufferQueue> buffer_queues;
+    std::vector<std::unique_ptr<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/vi/vi.cpp

@@ -544,6 +544,12 @@ private:
                                  Settings::values.resolution_factor.GetValue()),
                 static_cast<u32>(static_cast<u32>(DisplayResolution::UndockedHeight) *
                                  Settings::values.resolution_factor.GetValue())};
+
+            {
+                auto& buffer_queue = *nv_flinger.FindBufferQueue(id);
+                buffer_queue.Connect();
+            }
+
             ctx.WriteBuffer(response.Serialize());
             break;
         }
@@ -565,18 +571,15 @@ private:
             const u32 width{request.data.width};
             const u32 height{request.data.height};
 
-            std::optional<std::pair<u32, Service::Nvidia::MultiFence*>> result;
-
-            while (!result) {
             auto& buffer_queue = *nv_flinger.FindBufferQueue(id);
-                result = buffer_queue.DequeueBuffer(width, height);
-
-                if (result) {
+            do {
+                if (auto result = buffer_queue.DequeueBuffer(width, height); result) {
                     // Buffer is available
                     IGBPDequeueBufferResponseParcel response{result->first, *result->second};
                     ctx.WriteBuffer(response.Serialize());
+                    break;
                 }
-            }
+            } while (buffer_queue.IsConnected());
 
             break;
         }