Fix multi-threading issues (#2013)

- Implement atomic.fence to ensure a proper memory synchronization order
- Destroy exec_env_singleton first in wasm/aot deinstantiation
- Change terminate other threads to wait for other threads in
  wasm_exec_env_destroy
- Fix detach thread in thread_manager_start_routine
- Fix duplicated lock cluster->lock in wasm_cluster_cancel_thread
- Add lib-pthread and lib-wasi-threads compilation to Windows CI

core/shared/platform/include/platform_api_extension.h

@@ -92,6 +92,41 @@ int os_thread_detach(korp_tid);
 void
 os_thread_exit(void *retval);
 
+/* Try to define os_atomic_thread_fence if it isn't defined in
+   platform's platform_internal.h */
+#ifndef os_atomic_thread_fence
+
+#if !defined(__GNUC_PREREQ) && (defined(__GNUC__) || defined(__GNUG__)) \
+    && !defined(__clang__) && defined(__GNUC_MINOR__)
+#define __GNUC_PREREQ(maj, min) \
+    ((__GNUC__ << 16) + __GNUC_MINOR__ >= ((maj) << 16) + (min))
+#endif
+
+/* Clang's __GNUC_PREREQ macro has a different meaning than GCC one,
+   so we have to handle this case specially */
+#if defined(__clang__)
+/* Clang provides stdatomic.h since 3.6.0
+   See https://releases.llvm.org/3.6.0/tools/clang/docs/ReleaseNotes.html */
+#if __clang_major__ > 3 || (__clang_major__ == 3 && __clang_minor__ >= 6)
+#define BH_HAS_STD_ATOMIC
+#endif
+#elif defined(__GNUC_PREREQ)
+/* Even though older versions of GCC support C11, atomics were
+   not implemented until 4.9. See
+   https://gcc.gnu.org/bugzilla/show_bug.cgi?id=58016 */
+#if __GNUC_PREREQ(4, 9)
+#define BH_HAS_STD_ATOMIC
+#endif /* end of __GNUC_PREREQ(4, 9) */
+#endif /* end of defined(__GNUC_PREREQ) */
+
+#if defined(BH_HAS_STD_ATOMIC) && !defined(__cplusplus)
+#include <stdatomic.h>
+#define os_memory_order_release memory_order_release
+#define os_atomic_thread_fence atomic_thread_fence
+#endif
+
+#endif /* end of os_atomic_thread_fence */
+
 /**
  * Initialize current thread environment if current thread
  * is created by developer but not runtime

core/shared/platform/linux-sgx/platform_internal.h

@@ -63,6 +63,9 @@ os_set_print_function(os_print_function_t pf);
 char *
 strcpy(char *dest, const char *src);
 
+#define os_memory_order_release __ATOMIC_RELEASE
+#define os_atomic_thread_fence __atomic_thread_fence
+
 #ifdef __cplusplus
 }
 #endif

core/shared/platform/windows/platform_internal.h

@@ -116,6 +116,20 @@ os_thread_signal_inited();
 #endif /* end of BUILD_TARGET_X86_64/AMD_64 */
 #endif /* end of WASM_DISABLE_HW_BOUND_CHECK */
 
+typedef enum os_memory_order {
+    os_memory_order_relaxed,
+    os_memory_order_consume,
+    os_memory_order_acquire,
+    os_memory_order_release,
+    os_memory_order_acq_rel,
+    os_memory_order_seq_cst,
+} os_memory_order;
+
+void
+bh_atomic_thread_fence(int mem_order);
+
+#define os_atomic_thread_fence bh_atomic_thread_fence
+
 #ifdef __cplusplus
 }
 #endif

core/shared/platform/windows/shared_platform.cmake

@@ -10,7 +10,8 @@ add_definitions(-DHAVE_STRUCT_TIMESPEC)
 include_directories(${PLATFORM_SHARED_DIR})
 include_directories(${PLATFORM_SHARED_DIR}/../include)
 
-file (GLOB_RECURSE source_all ${PLATFORM_SHARED_DIR}/*.c)
+file (GLOB_RECURSE source_all ${PLATFORM_SHARED_DIR}/*.c
+			      ${PLATFORM_SHARED_DIR}/*.cpp)
 
 set (PLATFORM_SHARED_SOURCE ${source_all})
 

core/shared/platform/windows/win_atomic.cpp

@@ -0,0 +1,22 @@
+/*
+ * Copyright (C) 2023 Intel Corporation.  All rights reserved.
+ * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+ */
+
+#include "platform_api_vmcore.h"
+#include "platform_api_extension.h"
+
+#if WASM_ENABLE_SHARED_MEMORY != 0
+
+#include <atomic>
+
+void
+bh_atomic_thread_fence(int mem_order)
+{
+    std::memory_order order =
+        (std::memory_order)(std::memory_order::memory_order_relaxed + mem_order
+                            - os_memory_order_relaxed);
+    std::atomic_thread_fence(order);
+}
+
+#endif