video_core: Fix gpu thread deadlock

src/video_core/dma_pusher.cpp

@@ -34,7 +34,6 @@ void DmaPusher::DispatchCalls() {
     }
     gpu.FlushCommands();
     gpu.SyncGuestHost();
-    gpu.OnCommandListEnd();
 }
 
 bool DmaPusher::Step() {

src/video_core/gpu.cpp

@@ -107,14 +107,6 @@ struct GPU::Impl {
         rasterizer->SyncGuestHost();
     }
 
-    /// Signal the ending of command list.
-    void OnCommandListEnd() {
-        if (is_async) {
-            // This command only applies to asynchronous GPU mode
-            gpu_thread.OnCommandListEnd();
-        }
-    }
-
     /// Request a host GPU memory flush from the CPU.
     [[nodiscard]] u64 RequestFlush(VAddr addr, std::size_t size) {
         std::unique_lock lck{flush_request_mutex};
@@ -765,10 +757,6 @@ void GPU::SyncGuestHost() {
     impl->SyncGuestHost();
 }
 
-void GPU::OnCommandListEnd() {
-    impl->OnCommandListEnd();
-}
-
 u64 GPU::RequestFlush(VAddr addr, std::size_t size) {
     return impl->RequestFlush(addr, size);
 }

src/video_core/gpu_thread.cpp

@@ -77,6 +77,9 @@ void ThreadManager::StartThread(VideoCore::RendererBase& renderer,
 
 void ThreadManager::SubmitList(Tegra::CommandList&& entries) {
     PushCommand(SubmitListCommand(std::move(entries)));
+    if (is_async) {
+        PushCommand(OnCommandListEndCommand());
+    }
 }
 
 void ThreadManager::SwapBuffers(const Tegra::FramebufferConfig* framebuffer) {
@@ -107,10 +110,6 @@ void ThreadManager::FlushAndInvalidateRegion(VAddr addr, u64 size) {
     rasterizer->OnCPUWrite(addr, size);
 }
 
-void ThreadManager::OnCommandListEnd() {
-    PushCommand(OnCommandListEndCommand());
-}
-
 u64 ThreadManager::PushCommand(CommandData&& command_data, bool block) {
     if (!is_async) {
         // In synchronous GPU mode, block the caller until the command has executed

src/video_core/gpu_thread.h

@@ -129,8 +129,6 @@ public:
     /// Notify rasterizer that any caches of the specified region should be flushed and invalidated
     void FlushAndInvalidateRegion(VAddr addr, u64 size);
 
-    void OnCommandListEnd();
-
 private:
     /// Pushes a command to be executed by the GPU thread
     u64 PushCommand(CommandData&& command_data, bool block = false);