Refine aot call func procedure and fix zephyr timer overflow issue (#617)

1. Refine the aot call function procedure

2. fix timer integer overflow issue on zephyr platform

3. move wasm_exec_env_set_thread_info into lower layer

core/config.h

@@ -267,6 +267,9 @@
    stack overflow exception if the guard boudary is reached */
 #define RESERVED_BYTES_TO_NATIVE_STACK_BOUNDARY (512)
 
+/* Guard page count for stack overflow check with hardware trap */
+#define STACK_OVERFLOW_CHECK_GUARD_PAGE_COUNT 3
+
 /* Default wasm block address cache size and conflict list size */
 #ifndef BLOCK_ADDR_CACHE_SIZE
 #define BLOCK_ADDR_CACHE_SIZE 64

core/iwasm/aot/aot_runtime.c

@@ -1110,17 +1110,8 @@ aot_lookup_function(const AOTModuleInstance *module_inst,
 
 #ifdef OS_ENABLE_HW_BOUND_CHECK
 
-#define STACK_OVERFLOW_CHECK_GUARD_PAGE_COUNT 3
-
 static os_thread_local_attribute WASMExecEnv *aot_exec_env = NULL;
 
-static inline uint8 *
-get_stack_min_addr(WASMExecEnv *exec_env, uint32 page_size)
-{
-    uintptr_t stack_bound = (uintptr_t)exec_env->native_stack_boundary;
-    return (uint8*)(stack_bound & ~(uintptr_t)(page_size -1 ));
-}
-
 static void
 aot_signal_handler(void *sig_addr)
 {
@@ -1149,7 +1140,7 @@ aot_signal_handler(void *sig_addr)
 
         /* Get stack info of current thread */
         page_size = os_getpagesize();
-        stack_min_addr = get_stack_min_addr(aot_exec_env, page_size);
+        stack_min_addr = os_thread_get_stack_boundary();
 
         if (memory_inst
             && (mapped_mem_start_addr <= (uint8*)sig_addr
@@ -1182,27 +1173,6 @@ aot_signal_destroy()
     os_signal_destroy();
 }
 
-#if defined(__GNUC__)
-__attribute__((no_sanitize_address)) static uint32
-#else
-static uint32
-#endif
-touch_pages(uint8 *stack_min_addr, uint32 page_size)
-{
-    uint8 sum = 0;
-    while (1) {
-        volatile uint8 *touch_addr =
-            (volatile uint8*)os_alloca(page_size / 2);
-        if (touch_addr < stack_min_addr + page_size) {
-            sum += *(stack_min_addr + page_size - 1);
-            break;
-        }
-        *touch_addr = 0;
-        sum += *touch_addr;
-    }
-    return sum;
-}
-
 static bool
 invoke_native_with_hw_bound_check(WASMExecEnv *exec_env, void *func_ptr,
                                  const WASMType *func_type, const char *signature,
@@ -1211,10 +1181,9 @@ invoke_native_with_hw_bound_check(WASMExecEnv *exec_env, void *func_ptr,
 {
     AOTModuleInstance *module_inst = (AOTModuleInstance*)exec_env->module_inst;
     WASMExecEnv **p_aot_exec_env = &aot_exec_env;
-    WASMJmpBuf *jmpbuf_node, *jmpbuf_node_pop;
+    WASMJmpBuf jmpbuf_node = { 0 }, *jmpbuf_node_pop;
     uint32 page_size = os_getpagesize();
     uint32 guard_page_count = STACK_OVERFLOW_CHECK_GUARD_PAGE_COUNT;
-    uint8 *stack_min_addr = get_stack_min_addr(exec_env, page_size);
     bool ret;
 
     /* Check native stack overflow firstly to ensure we have enough
@@ -1226,33 +1195,17 @@ invoke_native_with_hw_bound_check(WASMExecEnv *exec_env, void *func_ptr,
         return false;
     }
 
-    if (aot_exec_env
-        && (aot_exec_env != exec_env)) {
+    if (aot_exec_env && (aot_exec_env != exec_env)) {
         aot_set_exception(module_inst, "invalid exec env");
         return false;
     }
 
-    if (!exec_env->jmpbuf_stack_top) {
-        /* Touch each stack page to ensure that it has been mapped: the OS may
-           lazily grow the stack mapping as a guard page is hit. */
-        (void)touch_pages(stack_min_addr, page_size);
-        /* First time to call aot function, protect one page */
-        if (os_mprotect(stack_min_addr, page_size * guard_page_count,
-                        MMAP_PROT_NONE) != 0) {
-            aot_set_exception(module_inst, "set protected page failed");
-            return false;
-        }
-    }
+    os_thread_init_stack_guard_pages();
 
-    if (!(jmpbuf_node = wasm_runtime_malloc(sizeof(WASMJmpBuf)))) {
-        aot_set_exception_with_id(module_inst, EXCE_OUT_OF_MEMORY);
-        return false;
-    }
-
-    wasm_exec_env_push_jmpbuf(exec_env, jmpbuf_node);
+    wasm_exec_env_push_jmpbuf(exec_env, &jmpbuf_node);
 
     aot_exec_env = exec_env;
-    if (os_setjmp(jmpbuf_node->jmpbuf) == 0) {
+    if (os_setjmp(jmpbuf_node.jmpbuf) == 0) {
         ret = wasm_runtime_invoke_native(exec_env, func_ptr, func_type,
                                          signature, attachment,
                                          argv, argc, argv_ret);
@@ -1263,12 +1216,8 @@ invoke_native_with_hw_bound_check(WASMExecEnv *exec_env, void *func_ptr,
     }
 
     jmpbuf_node_pop = wasm_exec_env_pop_jmpbuf(exec_env);
-    bh_assert(jmpbuf_node == jmpbuf_node_pop);
-    wasm_runtime_free(jmpbuf_node);
+    bh_assert(&jmpbuf_node == jmpbuf_node_pop);
     if (!exec_env->jmpbuf_stack_top) {
-        /* Unprotect the guard page when the nested call depth is zero */
-        os_mprotect(stack_min_addr, page_size * guard_page_count,
-                    MMAP_PROT_READ | MMAP_PROT_WRITE);
         *p_aot_exec_env = NULL;
     }
     os_sigreturn();
@@ -1293,6 +1242,9 @@ aot_call_function(WASMExecEnv *exec_env,
     uint32 ext_ret_count = result_count > 1 ? result_count - 1 : 0;
     bool ret;
 
+    /* set thread handle and stack boundary */
+    wasm_exec_env_set_thread_info(exec_env);
+
     if (ext_ret_count > 0) {
         uint32 cell_num = 0, i;
         uint8 *ext_ret_types = func_type->types + func_type->param_count + 1;
@@ -1375,7 +1327,7 @@ aot_call_function(WASMExecEnv *exec_env,
             case VALUE_TYPE_V128:
                 argv_ret += 4;
                 break;
-#endif                -
+#endif
             default:
                 bh_assert(0);
                 break;
@@ -1436,8 +1388,6 @@ aot_create_exec_env_and_call_function(AOTModuleInstance *module_inst,
             return false;
         }
 
-        /* set thread handle and stack boundary */
-        wasm_exec_env_set_thread_info(exec_env);
 #if WASM_ENABLE_THREAD_MGR != 0
     }
 #endif
@@ -2265,7 +2215,7 @@ aot_call_indirect(WASMExecEnv *exec_env,
             case VALUE_TYPE_V128:
                 argv_ret += 4;
                 break;
-#endif                -
+#endif
             default:
                 bh_assert(0);
                 break;

core/iwasm/common/wasm_runtime_common.c

@@ -1122,9 +1122,6 @@ wasm_runtime_call_wasm(WASMExecEnv *exec_env,
         return false;
     }
 
-    /* set thread handle and stack boundary */
-    wasm_exec_env_set_thread_info(exec_env);
-
 #if WASM_ENABLE_REF_TYPES != 0
     wasm_runtime_prepare_call_function(exec_env, function);
 #endif

core/iwasm/interpreter/wasm_runtime.c

@@ -1647,6 +1647,10 @@ wasm_call_function(WASMExecEnv *exec_env,
                    unsigned argc, uint32 argv[])
 {
     WASMModuleInstance *module_inst = (WASMModuleInstance*)exec_env->module_inst;
+
+    /* set thread handle and stack boundary */
+    wasm_exec_env_set_thread_info(exec_env);
+
     wasm_interp_call_wasm(module_inst, exec_env, function, argc, argv);
     (void)clear_wasi_proc_exit_exception(module_inst);
     return !wasm_get_exception(module_inst) ? true : false;
@@ -1674,8 +1678,6 @@ wasm_create_exec_env_and_call_function(WASMModuleInstance *module_inst,
             return false;
         }
 
-        /* set thread handle and stack boundary */
-        wasm_exec_env_set_thread_info(exec_env);
 #if WASM_ENABLE_THREAD_MGR != 0
     }
 #endif

core/shared/platform/android/platform_init.c

@@ -20,6 +20,9 @@ bh_platform_init()
 void
 bh_platform_destroy()
 {
+#ifdef OS_ENABLE_HW_BOUND_CHECK
+    os_thread_destroy_stack_guard_pages();
+#endif
 }
 
 int os_printf(const char *fmt, ...)

core/shared/platform/android/platform_internal.h

@@ -56,6 +56,8 @@ typedef pthread_mutex_t korp_mutex;
 typedef pthread_cond_t korp_cond;
 typedef pthread_t korp_thread;
 
+#define os_thread_local_attribute __thread
+
 #if WASM_DISABLE_HW_BOUND_CHECK == 0
 #if defined(BUILD_TARGET_X86_64) \
     || defined(BUILD_TARGET_AMD_64) \
@@ -65,8 +67,6 @@ typedef pthread_t korp_thread;
 
 #define OS_ENABLE_HW_BOUND_CHECK
 
-#define os_thread_local_attribute __thread
-
 typedef jmp_buf korp_jmpbuf;
 
 #define os_setjmp setjmp
@@ -75,6 +75,10 @@ typedef jmp_buf korp_jmpbuf;
 
 #define os_getpagesize getpagesize
 
+bool os_thread_init_stack_guard_pages();
+
+void os_thread_destroy_stack_guard_pages();
+
 typedef void (*os_signal_handler)(void *sig_addr);
 
 int os_signal_init(os_signal_handler handler);

core/shared/platform/common/posix/posix_thread.c

@@ -25,6 +25,9 @@ static void *os_thread_wrapper(void *arg)
     targ->stack = (void *)((uintptr_t)(&arg) & (uintptr_t)~0xfff);
     BH_FREE(targ);
     start_func(thread_arg);
+#ifdef OS_ENABLE_HW_BOUND_CHECK
+    os_thread_destroy_stack_guard_pages();
+#endif
     return NULL;
 }
 
@@ -239,22 +242,32 @@ int os_thread_detach(korp_tid thread)
 
 void os_thread_exit(void *retval)
 {
+#ifdef OS_ENABLE_HW_BOUND_CHECK
+    os_thread_destroy_stack_guard_pages();
+#endif
     return pthread_exit(retval);
 }
 
+static os_thread_local_attribute uint8 *thread_stack_boundary = NULL;
+
 uint8 *os_thread_get_stack_boundary()
 {
-    pthread_t self = pthread_self();
+    pthread_t self;
 #ifdef __linux__
     pthread_attr_t attr;
     size_t guard_size;
 #endif
     uint8 *addr = NULL;
-    size_t stack_size;
-    int page_size = getpagesize();
-    size_t max_stack_size = (size_t)
-                            (APP_THREAD_STACK_SIZE_MAX + page_size - 1)
-                            & ~(page_size - 1);
+    size_t stack_size, max_stack_size;
+    int page_size;
+
+    if (thread_stack_boundary)
+        return thread_stack_boundary;
+
+    page_size = getpagesize();
+    self = pthread_self();
+    max_stack_size = (size_t)(APP_THREAD_STACK_SIZE_MAX + page_size - 1)
+                     & ~(page_size - 1);
 
     if (max_stack_size < APP_THREAD_STACK_SIZE_DEFAULT)
         max_stack_size = APP_THREAD_STACK_SIZE_DEFAULT;
@@ -284,6 +297,7 @@ uint8 *os_thread_get_stack_boundary()
     }
 #endif
 
+    thread_stack_boundary = addr;
     return addr;
 }
 
@@ -291,12 +305,71 @@ uint8 *os_thread_get_stack_boundary()
 
 #define SIG_ALT_STACK_SIZE (32 * 1024)
 
+/* Whether the stack pages are touched and guard pages are set */
+static os_thread_local_attribute bool stack_guard_pages_inited = false;
+
 /* The signal alternate stack base addr */
 static uint8 *sigalt_stack_base_addr;
 
 /* The signal handler passed to os_signal_init() */
 static os_signal_handler signal_handler;
 
+#if defined(__GNUC__)
+__attribute__((no_sanitize_address)) static uint32
+#else
+static uint32
+#endif
+touch_pages(uint8 *stack_min_addr, uint32 page_size)
+{
+    uint8 sum = 0;
+    while (1) {
+        volatile uint8 *touch_addr =
+            (volatile uint8*)os_alloca(page_size / 2);
+        if (touch_addr < stack_min_addr + page_size) {
+            sum += *(stack_min_addr + page_size - 1);
+            break;
+        }
+        *touch_addr = 0;
+        sum += *touch_addr;
+    }
+    return sum;
+}
+
+bool
+os_thread_init_stack_guard_pages()
+{
+    uint32 page_size = os_getpagesize();
+    uint32 guard_page_count = STACK_OVERFLOW_CHECK_GUARD_PAGE_COUNT;
+    uint8 *stack_min_addr = os_thread_get_stack_boundary();
+
+    if (!stack_guard_pages_inited) {
+        /* Touch each stack page to ensure that it has been mapped: the OS
+           may lazily grow the stack mapping as a guard page is hit. */
+        (void)touch_pages(stack_min_addr, page_size);
+        /* First time to call aot function, protect guard pages */
+        if (os_mprotect(stack_min_addr, page_size * guard_page_count,
+                        MMAP_PROT_NONE) != 0) {
+            return false;
+        }
+        stack_guard_pages_inited = true;
+    }
+    return true;
+}
+
+void
+os_thread_destroy_stack_guard_pages()
+{
+    if (stack_guard_pages_inited) {
+        uint32 page_size = os_getpagesize();
+        uint32 guard_page_count = STACK_OVERFLOW_CHECK_GUARD_PAGE_COUNT;
+        uint8 *stack_min_addr = os_thread_get_stack_boundary();
+
+        os_mprotect(stack_min_addr, page_size * guard_page_count,
+                    MMAP_PROT_READ | MMAP_PROT_WRITE);
+        stack_guard_pages_inited = false;
+    }
+}
+
 static void
 mask_signals(int how)
 {

core/shared/platform/darwin/platform_init.c

@@ -14,6 +14,9 @@ bh_platform_init()
 void
 bh_platform_destroy()
 {
+#ifdef OS_ENABLE_HW_BOUND_CHECK
+    os_thread_destroy_stack_guard_pages();
+#endif
 }
 
 int

core/shared/platform/darwin/platform_internal.h

@@ -57,6 +57,8 @@ typedef pthread_mutex_t korp_mutex;
 typedef pthread_cond_t korp_cond;
 typedef pthread_t korp_thread;
 
+#define os_thread_local_attribute __thread
+
 #if WASM_DISABLE_HW_BOUND_CHECK == 0
 #if defined(BUILD_TARGET_X86_64) \
     || defined(BUILD_TARGET_AMD_64) \
@@ -66,8 +68,6 @@ typedef pthread_t korp_thread;
 
 #define OS_ENABLE_HW_BOUND_CHECK
 
-#define os_thread_local_attribute __thread
-
 typedef jmp_buf korp_jmpbuf;
 
 #define os_setjmp setjmp
@@ -76,6 +76,10 @@ typedef jmp_buf korp_jmpbuf;
 
 #define os_getpagesize getpagesize
 
+bool os_thread_init_stack_guard_pages();
+
+void os_thread_destroy_stack_guard_pages();
+
 typedef void (*os_signal_handler)(void *sig_addr);
 
 int os_signal_init(os_signal_handler handler);

core/shared/platform/linux/platform_init.c

@@ -14,6 +14,9 @@ bh_platform_init()
 void
 bh_platform_destroy()
 {
+#ifdef OS_ENABLE_HW_BOUND_CHECK
+    os_thread_destroy_stack_guard_pages();
+#endif
 }
 
 int

core/shared/platform/linux/platform_internal.h

@@ -56,6 +56,8 @@ typedef pthread_mutex_t korp_mutex;
 typedef pthread_cond_t korp_cond;
 typedef pthread_t korp_thread;
 
+#define os_thread_local_attribute __thread
+
 #if WASM_DISABLE_HW_BOUND_CHECK == 0
 #if defined(BUILD_TARGET_X86_64) \
     || defined(BUILD_TARGET_AMD_64) \
@@ -65,8 +67,6 @@ typedef pthread_t korp_thread;
 
 #define OS_ENABLE_HW_BOUND_CHECK
 
-#define os_thread_local_attribute __thread
-
 typedef jmp_buf korp_jmpbuf;
 
 #define os_setjmp setjmp
@@ -75,6 +75,10 @@ typedef jmp_buf korp_jmpbuf;
 
 #define os_getpagesize getpagesize
 
+bool os_thread_init_stack_guard_pages();
+
+void os_thread_destroy_stack_guard_pages();
+
 typedef void (*os_signal_handler)(void *sig_addr);
 
 int os_signal_init(os_signal_handler handler);

core/shared/platform/vxworks/platform_init.c

@@ -14,6 +14,9 @@ bh_platform_init()
 void
 bh_platform_destroy()
 {
+#ifdef OS_ENABLE_HW_BOUND_CHECK
+    os_thread_destroy_stack_guard_pages();
+#endif
 }
 
 int

core/shared/platform/vxworks/platform_internal.h

@@ -55,6 +55,8 @@ typedef pthread_mutex_t korp_mutex;
 typedef pthread_cond_t korp_cond;
 typedef pthread_t korp_thread;
 
+#define os_thread_local_attribute __thread
+
 #if WASM_DISABLE_HW_BOUND_CHECK == 0
 #if defined(BUILD_TARGET_X86_64) \
     || defined(BUILD_TARGET_AMD_64) \
@@ -64,8 +66,6 @@ typedef pthread_t korp_thread;
 
 #define OS_ENABLE_HW_BOUND_CHECK
 
-#define os_thread_local_attribute __thread
-
 typedef jmp_buf korp_jmpbuf;
 
 #define os_setjmp setjmp
@@ -74,6 +74,10 @@ typedef jmp_buf korp_jmpbuf;
 
 #define os_getpagesize getpagesize
 
+bool os_thread_init_stack_guard_pages();
+
+void os_thread_destroy_stack_guard_pages();
+
 typedef void (*os_signal_handler)(void *sig_addr);
 
 int os_signal_init(os_signal_handler handler);

core/shared/platform/zephyr/zephyr_time.c

@@ -8,6 +8,6 @@
 uint64
 os_time_get_boot_microsecond()
 {
-    return k_uptime_get_32() * 1000;
+    return k_uptime_get() * 1000;
 }