rsx: Avoid semaphore acquire deadlock

rpcs3/Emu/RSX/D3D12/D3D12GSRender.cpp

@@ -301,10 +301,15 @@ void D3D12GSRender::on_exit()
 	return GSRender::on_exit();
 }
 
-void D3D12GSRender::do_local_task(bool)
+void D3D12GSRender::do_local_task(rsx::FIFO_state state)
 {
-	//TODO
+	if (state != rsx::FIFO_state::lock_wait)
-	m_frame->clear_wm_events();
+	{
+		//TODO
+		m_frame->clear_wm_events();
+	}
+
+	rsx::thread::do_local_task(state);
 }
 
 bool D3D12GSRender::do_method(u32 cmd, u32 arg)

rpcs3/Emu/RSX/D3D12/D3D12GSRender.h

@@ -173,7 +173,7 @@ private:
 protected:
 	virtual void on_init_thread() override;
 	virtual void on_exit() override;
-	virtual void do_local_task(bool idle) override;
+	virtual void do_local_task(rsx::FIFO_state state) override;
 	virtual bool do_method(u32 cmd, u32 arg) override;
 	virtual void end() override;
 	virtual void flip(int buffer) override;

rpcs3/Emu/RSX/GL/GLGSRender.cpp

@@ -1582,10 +1582,8 @@ void GLGSRender::on_invalidate_memory_range(u32 address_base, u32 size)
 	}
 }
 
-void GLGSRender::do_local_task(bool idle)
+void GLGSRender::do_local_task(rsx::FIFO_state state)
 {
-	m_frame->clear_wm_events();
-
 	if (!work_queue.empty())
 	{
 		std::lock_guard<shared_mutex> lock(queue_guard);
@@ -1605,13 +1603,23 @@ void GLGSRender::do_local_task(bool idle)
 			q.cv.notify_one();
 		}
 	}
-	else if (!in_begin_end)
+	else if (!in_begin_end && state != rsx::FIFO_state::lock_wait)
 	{
 		//This will re-engage locks and break the texture cache if another thread is waiting in access violation handler!
 		//Only call when there are no waiters
 		m_gl_texture_cache.do_update();
 	}
 
+	rsx::thread::do_local_task(state);
+
+	if (state == rsx::FIFO_state::lock_wait)
+	{
+		// Critical check finished
+		return;
+	}
+
+	m_frame->clear_wm_events();
+
 	if (m_overlay_manager)
 	{
 		if (!in_begin_end && native_ui_flip_request.load())
@@ -1620,8 +1628,6 @@ void GLGSRender::do_local_task(bool idle)
 			flip((s32)current_display_buffer);
 		}
 	}
-
-	rsx::thread::do_local_task(idle);
 }
 
 work_item& GLGSRender::post_flush_request(u32 address, gl::texture_cache::thrashed_set& flush_data)

rpcs3/Emu/RSX/GL/GLGSRender.h

@@ -381,7 +381,7 @@ protected:
 	bool do_method(u32 id, u32 arg) override;
 	void flip(int buffer) override;
 
-	void do_local_task(bool idle) override;
+	void do_local_task(rsx::FIFO_state state) override;
 
 	bool on_access_violation(u32 address, bool is_writing) override;
 	void on_invalidate_memory_range(u32 address_base, u32 size) override;

rpcs3/Emu/RSX/RSXThread.cpp

@@ -513,7 +513,7 @@ namespace rsx
 			}
 
 			//Execute backend-local tasks first
-			do_local_task(performance_counters.state != FIFO_state::running);
+			do_local_task(performance_counters.state);
 
 			//Update sub-units
 			zcull_ctrl->update(this);
@@ -1290,9 +1290,9 @@ namespace rsx
 		}
 	}
 
-	void thread::do_local_task(bool /*idle*/)
+	void thread::do_local_task(FIFO_state state)
 	{
-		if (!in_begin_end)
+		if (!in_begin_end && state != FIFO_state::lock_wait)
 		{
 			if (!m_invalidated_memory_ranges.empty())
 			{

rpcs3/Emu/RSX/RSXThread.h

@@ -87,6 +87,7 @@ namespace rsx
 		empty = 1,    // PUT == GET
 		spinning = 2, // Puller continuously jumps to self addr (synchronization technique)
 		nop = 3,      // Puller is processing a NOP command
+		lock_wait = 4 // Puller is processing a lock acquire
 	};
 
 	u32 get_vertex_type_size_on_host(vertex_base_type type, u32 size);
@@ -417,9 +418,8 @@ namespace rsx
 
 		/**
 		 * Execute a backend local task queue
-		 * Idle argument checks that the FIFO queue is in an idle state
 		 */
-		virtual void do_local_task(bool idle);
+		virtual void do_local_task(FIFO_state state);
 
 	public:
 		virtual std::string get_name() const override;
@@ -545,6 +545,11 @@ namespace rsx
 		 */
 		void on_notify_memory_unmapped(u32 address_base, u32 size);
 
+		/**
+		 * Notify to check internal state during semaphore wait
+		 */
+		void on_semaphore_acquire_wait() { do_local_task(FIFO_state::lock_wait); }
+
 		/**
 		 * Copy rtt values to buffer.
 		 * TODO: It's more efficient to combine multiple call of this function into one.

rpcs3/Emu/RSX/VK/VKGSRender.cpp

@@ -857,9 +857,6 @@ bool VKGSRender::on_access_violation(u32 address, bool is_writing)
 
 			if (target_cb)
 				target_cb->wait();
-
-			if (is_rsxthr)
-				m_last_flushable_cb = -1;
 		}
 
 		if (has_queue_ref)
@@ -1858,7 +1855,6 @@ void VKGSRender::copy_render_targets_to_dma_location()
 
 	vk::leave_uninterruptible();
 
-	m_last_flushable_cb = m_current_cb_index;
 	flush_command_queue();
 
 	m_flush_draw_buffers = false;
@@ -1884,7 +1880,6 @@ void VKGSRender::flush_command_queue(bool hard_sync)
 				cb.poke();
 		}
 
-		m_last_flushable_cb = -1;
 		m_flush_requests.clear_pending_flag();
 	}
 	else
@@ -1904,9 +1899,6 @@ void VKGSRender::flush_command_queue(bool hard_sync)
 		}
 
 		m_current_command_buffer->reset();
-
-		if (m_last_flushable_cb == m_current_cb_index)
-			m_last_flushable_cb = -1;
 	}
 
 	open_command_buffer();
@@ -2089,7 +2081,7 @@ void VKGSRender::process_swap_request(frame_context_t *ctx, bool free_resources)
 	ctx->swap_command_buffer = nullptr;
 }
 
-void VKGSRender::do_local_task(bool idle)
+void VKGSRender::do_local_task(rsx::FIFO_state state)
 {
 	if (m_flush_requests.pending())
 	{
@@ -2102,22 +2094,19 @@ void VKGSRender::do_local_task(bool idle)
 		m_flush_requests.clear_pending_flag();
 		m_flush_requests.consumer_wait();
 	}
-	else if (!in_begin_end)
+	else if (!in_begin_end && state != rsx::FIFO_state::lock_wait)
 	{
 		//This will re-engage locks and break the texture cache if another thread is waiting in access violation handler!
 		//Only call when there are no waiters
 		m_texture_cache.do_update();
 	}
 
-	if (m_last_flushable_cb > -1)
+	rsx::thread::do_local_task(state);
+
+	if (state == rsx::FIFO_state::lock_wait)
 	{
-		auto cb = &m_primary_cb_list[m_last_flushable_cb];
+		// Critical check finished
-
+		return;
-		if (cb->pending)
-			cb->poke();
-
-		if (!cb->pending)
-			m_last_flushable_cb = -1;
 	}
 
 #ifdef _WIN32
@@ -2208,8 +2197,6 @@ void VKGSRender::do_local_task(bool idle)
 			flip((s32)current_display_buffer);
 		}
 	}
-
-	rsx::thread::do_local_task(idle);
 }
 
 bool VKGSRender::do_method(u32 cmd, u32 arg)

rpcs3/Emu/RSX/VK/VKGSRender.h

@@ -351,7 +351,6 @@ private:
 
 	u8 m_draw_buffers_count = 0;
 	bool m_flush_draw_buffers = false;
-	std::atomic<int> m_last_flushable_cb = {-1 };
 	
 	shared_mutex m_flush_queue_mutex;
 	flush_request_task m_flush_requests;
@@ -420,7 +419,7 @@ protected:
 	bool do_method(u32 id, u32 arg) override;
 	void flip(int buffer) override;
 
-	void do_local_task(bool idle) override;
+	void do_local_task(rsx::FIFO_state state) override;
 	bool scaled_image_from_memory(rsx::blit_src_info& src, rsx::blit_dst_info& dst, bool interpolate) override;
 	void notify_tile_unbound(u32 tile) override;
 

rpcs3/Emu/RSX/rsx_methods.cpp

@@ -78,35 +78,31 @@ namespace rsx
 				if (Emu.IsStopped())
 					return;
 
-				const auto tdr = (u64)g_cfg.video.driver_recovery_timeout;
+				if (const auto tdr = (u64)g_cfg.video.driver_recovery_timeout)
-				if (tdr == 0)
 				{
-					//No timeout
+					if (Emu.IsPaused())
-					std::this_thread::yield();
-					continue;
-				}
-
-				if (Emu.IsPaused())
-				{
-					while (Emu.IsPaused())
 					{
-						std::this_thread::yield();
+						while (Emu.IsPaused())
-					}
+						{
+							std::this_thread::sleep_for(1ms);
+						}
 
-					//Reset
+						// Reset
-					start = get_system_time();
+						start = get_system_time();
-				}
+					}
-				else
+					else
-				{
-					if ((get_system_time() - start) > tdr)
 					{
-						//If longer than driver timeout force exit
+						if ((get_system_time() - start) > tdr)
-						LOG_ERROR(RSX, "nv406e::semaphore_acquire has timed out. semaphore_address=0x%X", addr);
+						{
-						break;
+							// If longer than driver timeout force exit
+							LOG_ERROR(RSX, "nv406e::semaphore_acquire has timed out. semaphore_address=0x%X", addr);
+							break;
+						}
 					}
-
-					std::this_thread::yield();
 				}
+
+				rsx->on_semaphore_acquire_wait();
+				std::this_thread::yield();
 			}
 
 			rsx->performance_counters.idle_time += (get_system_time() - start);