fix: avoid joining thread in deferred deleter

Join thread in DllMain (which we are not aware of)
can result in hang occurring in DeferredDeleter, if the library is
freed before FreeLibraryAndExitThread call from within the worker thread,
the thread gets stuck, thus the main thread is stuck on worker->join().

Related-To: NEO-14121

Signed-off-by: Oskar Hubert Weber <oskar.hubert.weber@intel.com>

shared/source/memory_manager/deferred_deleter.cpp

@@ -27,8 +27,11 @@ void DeferredDeleter::stop() {
         doWorkInBackground = false;
         lock.unlock();
         condition.notify_one();
-        // Wait for the working job to exit
-        worker->join();
+        worker->detach();
+        // Wait for the working job to exit main loop
+        while (!exitedMainLoop) {
+            std::this_thread::yield();
+        }
         // Delete working thread
         worker.reset();
     }
@@ -76,6 +79,7 @@ void DeferredDeleter::ensureThread() {
     if (worker != nullptr) {
         return;
     }
+    exitedMainLoop = false;
     worker = Thread::createFunc(run, reinterpret_cast<void *>(this));
 }
 
@@ -103,6 +107,7 @@ void *DeferredDeleter::run(void *arg) {
         lock.lock();
         // Check whether working thread should be stopped
     } while (!self->shouldStop());
+    self->exitedMainLoop = true;
     lock.unlock();
     return nullptr;
 }

shared/source/memory_manager/deferred_deleter.h

@@ -40,6 +40,7 @@ class DeferredDeleter : NEO::NonCopyableAndNonMovableClass {
 
     static void *run(void *);
 
+    std::atomic<bool> exitedMainLoop = false;
     std::atomic<bool> doWorkInBackground = false;
     std::atomic<int> elementsToRelease = 0;
     std::atomic<int> hostptrsToRelease = 0;

shared/source/os_interface/linux/os_thread_linux.cpp

@@ -24,6 +24,10 @@ void ThreadLinux::join() {
     pthread_join(threadId, nullptr);
 }
 
+void ThreadLinux::detach() {
+    pthread_detach(threadId);
+}
+
 void ThreadLinux::yield() {
     sched_yield();
 }

shared/source/os_interface/linux/os_thread_linux.h

@@ -16,6 +16,7 @@ class ThreadLinux : public Thread {
   public:
     ThreadLinux(pthread_t threadId);
     void join() override;
+    void detach() override;
     void yield() override;
     ~ThreadLinux() override = default;
 

shared/source/os_interface/os_thread.h

@@ -15,6 +15,7 @@ class Thread {
   public:
     static decltype(&Thread::create) createFunc;
     virtual void join() = 0;
+    virtual void detach() = 0;
     virtual ~Thread() = default;
     virtual void yield() = 0;
 };

shared/source/os_interface/windows/os_thread_win.cpp

@@ -21,7 +21,12 @@ void ThreadWin::join() {
     thread->join();
 }
 
+void ThreadWin::detach() {
+    thread->detach();
+}
+
 void ThreadWin::yield() {
     std::this_thread::yield();
 }
+
 } // namespace NEO

shared/source/os_interface/windows/os_thread_win.h

@@ -16,6 +16,7 @@ class ThreadWin : public Thread {
   public:
     ThreadWin(std::thread *thread);
     void join() override;
+    void detach() override;
     void yield() override;
     ~ThreadWin() override = default;
 

shared/test/unit_test/memory_manager/deferrable_allocation_deletion_tests.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2018-2024 Intel Corporation
+ * Copyright (C) 2018-2025 Intel Corporation
  *
  * SPDX-License-Identifier: MIT
  *
@@ -10,6 +10,7 @@
 #include "shared/source/memory_manager/deferrable_allocation_deletion.h"
 #include "shared/source/memory_manager/deferred_deleter.h"
 #include "shared/source/os_interface/os_context.h"
+#include "shared/test/common/helpers/memory_management.h"
 #include "shared/test/common/libult/ult_command_stream_receiver.h"
 #include "shared/test/common/mocks/mock_allocation_properties.h"
 #include "shared/test/common/mocks/mock_device.h"
@@ -36,6 +37,7 @@ struct DeferredDeleterPublic : DeferredDeleter {
 
 struct DeferrableAllocationDeletionTest : ::testing::Test {
     void SetUp() override {
+        MemoryManagement::fastLeaksDetectionMode = MemoryManagement::LeakDetectionMode::TURN_OFF_LEAK_DETECTION;
         executionEnvironment = new MockExecutionEnvironment(defaultHwInfo.get(), false, 1u);
         executionEnvironment->incRefInternal();
         memoryManager = new MockMemoryManager(*executionEnvironment);