fix for hang on windows

Author: isanghao

src/inference/dev_api/openvino/runtime/iplugin.hpp

@@ -236,6 +236,11 @@ class OPENVINO_RUNTIME_API IPlugin : public std::enable_shared_from_this<IPlugin
 
     virtual ~IPlugin() = default;
 
+    /**
+     * @brief Explicit resource cleanup before plugin unload
+     */
+    virtual void cleanup() {};
+
 protected:
     IPlugin();
 

src/inference/src/cpp/core.cpp

@@ -271,7 +271,7 @@ void Core::unload_plugin(const std::string& device_name) {
     OV_CORE_CALL_STATEMENT({
         ov::DeviceIDParser parser(device_name);
         std::string devName = parser.get_device_name();
-
+        _impl->get_plugin(devName).cleanup();
         _impl->unload_plugin(devName);
     });
 }

src/inference/src/dev/plugin.cpp

@@ -20,10 +20,6 @@
         OPENVINO_THROW("Unexpected exception");                                        \
     }
 
-ov::Plugin::~Plugin() {
-    m_ptr = {};
-}
-
 ov::Plugin::Plugin(const std::shared_ptr<ov::IPlugin>& ptr, const std::shared_ptr<void>& so) : m_ptr{ptr}, m_so{so} {
     OV_PLUGIN_CALL_STATEMENT();
 }

src/inference/src/dev/plugin.hpp

@@ -28,8 +28,6 @@ class Plugin {
 public:
     Plugin() = default;
 
-    ~Plugin();
-
     Plugin(const std::shared_ptr<ov::IPlugin>& ptr, const std::shared_ptr<void>& so);
 
     void set_name(const std::string& deviceName);
@@ -81,6 +79,8 @@ class Plugin {
         return get_property(property.name(), arguments).template as<T>();
     }
     bool supports_model_caching(const AnyMap& arguments = {}) const;
+
+    void cleanup() { m_ptr->cleanup(); }
 };
 
 }  // namespace ov

src/plugins/intel_gpu/include/intel_gpu/plugin/plugin.hpp

@@ -20,6 +20,8 @@ class Plugin : public ov::IPlugin {
     std::map<std::string, ExecutionConfig> m_configs_map;
     ov::AnyMap m_compiled_model_runtime_properties;
 
+    mutable std::map<std::string, std::shared_ptr<RemoteContextImpl>> m_default_contexts;
+    mutable std::once_flag m_default_contexts_once;
     std::map<std::string, std::shared_ptr<RemoteContextImpl>> get_default_contexts() const;
 
     std::shared_ptr<ov::Model> clone_and_transform_model(const std::shared_ptr<const ov::Model>& network,
@@ -67,6 +69,7 @@ class Plugin : public ov::IPlugin {
                                     const ov::AnyMap& properties) const override;
     ov::SoPtr<ov::IRemoteContext> create_context(const ov::AnyMap& remote_properties) const override;
     ov::SoPtr<ov::IRemoteContext> get_default_context(const ov::AnyMap& remote_properties) const override;
+    void cleanup() override;
 };
 
 }  // namespace ov::intel_gpu

src/plugins/intel_gpu/src/plugin/plugin.cpp

@@ -200,17 +200,31 @@ std::shared_ptr<ov::Model> Plugin::clone_and_transform_model(const std::shared_p
     return cloned_model;
 }
 
-std::map<std::string, RemoteContextImpl::Ptr> Plugin::get_default_contexts() const {
-    static std::map<std::string, std::shared_ptr<RemoteContextImpl>> m_default_contexts;
-    static std::once_flag m_default_contexts_once;
+// weak map to hold singleton default contexts
+// Weak singleton map is used to share contexts between multiple plugin(or Core) instances.
+// It is needed to ensure that context is released before plugin is unloaded.
+// As the actual ownership is in plugin class, the ownership is released when plugin is destructed.
+std::map<std::string, std::weak_ptr<RemoteContextImpl>> weak_singleton_default_contexts;
+std::mutex singleton_default_contexts_mutex;
 
+std::map<std::string, RemoteContextImpl::Ptr> Plugin::get_default_contexts() const {
     std::call_once(m_default_contexts_once, [this]() {
-        // Create default context
+        std::lock_guard<std::mutex> lock(singleton_default_contexts_mutex);
         for (auto& device : m_device_map) {
             const auto device_name = get_device_name() + "." + device.first;
+
+            // If already initialized, use existing one
+            if (weak_singleton_default_contexts.find(device.first) != weak_singleton_default_contexts.end()) {
+                if (auto ctx = weak_singleton_default_contexts[device.first].lock()) {
+                    m_default_contexts[device.first] = ctx;
+                    continue;
+                }
+            }
+            // If context is not created yet or expired, create new one
             const auto initialize = false;
             auto ctx = std::make_shared<RemoteContextImpl>(device_name, std::vector<cldnn::device::ptr>{device.second}, initialize);
-            m_default_contexts.insert({device.first, ctx});
+            weak_singleton_default_contexts[device.first] = ctx;
+            m_default_contexts[device.first] = ctx;
         }
     });
     return m_default_contexts;
@@ -972,6 +986,10 @@ uint32_t Plugin::get_optimal_batch_size(const ov::AnyMap& options) const {
 
     return batch;
 }
+void Plugin::cleanup() {
+    GPU_DEBUG_INFO << "[GPU] Plugin shutdown" << std::endl;
+    m_default_contexts.clear();
+}
 
 }  // namespace ov::intel_gpu