ArkCompiler: Detecting Runtime Issues

Multi Thread Check

In the JavaScript runtime environment, the safety of multi-threading is an important consideration. Since JavaScript is a single-threaded language, any JavaScript object can only be operated on the thread where it is created. Violating this rule can lead to multi-threading safety issues. Here are some detailed explanations on how to determine and handle these issues. Enabling multi-thread check will cause significant performance loss. Enable it as needed.

[h2]Technical Overview

Single-threaded execution
JavaScript is a single-threaded language, which means that it can only execute code on one thread at a time. Any JavaScript object can only be operated on the thread where it is created.
Node-API
The Node-API is directly involved in the operation of JavaScript objects. The vast majority (about 95%) of Node-APIs can only be called on the JavaScript thread that created these objects.

Multi thread check mechanism
The multi thread check mechanism detects whether the current thread is consistent with the JavaScript thread ID in the JavaScript virtual machine environment (vm/env) being used. If they are not consistent, it indicates that the virtual machine environment is being used across threads, and there is a multithreading safety issue.

[h2]Common Multithreading Safety Issues

Non-JavaScript thread using Node-API

A non-JavaScript thread attempting to call the Node-API may lead to undefined behavior or crashes.

// Index.ets
Text(this.message)
  .fontSize($r('app.float.page_text_font_size'))
  .fontWeight(FontWeight.Bold)
  .onClick(() => {
    testNapi.multiCheck();
  })

// napi_init.cpp
static napi_value MultiCheck(napi_env env, napi_callback_info info)
{
    std::thread([](napi_env env) {
        napi_value obj = nullptr;
        napi_create_object(env, &obj);
    }, env).join();

    return nullptr;
}

Node-API using another thread's env

A thread attempting to use a JavaScript environment (env) created by another thread can also lead to multithreading safety issues.

// Index.ets
Text(this.message)
  .fontSize($r('app.float.page_text_font_size'))
  .fontWeight(FontWeight.Bold)
  .onClick(() => {
    testNapi.saveEnv();
    const task = new taskpool.Task(createObject);
    taskpool.execute(task);
  })

// napi_init.cpp
napi_env env_ = nullptr;
static napi_value SaveEnv(napi_env env, napi_callback_info info)
{
    env_ = env;
    return nullptr;
}
static napi_value CreateObject(napi_env env, napi_callback_info info)
{
    napi_value obj = nullptr;
    napi_create_object(env_, &obj);
    return nullptr;
}

Identifying Multithreading Safety Issues

If you encounter the following fatal error message during runtime, it means a multithreading safety issue has occurred:

Fatal: ecma_vm cannot run in multi-thread! thread:3096 currentThread:3550

The current thread number is 3550, and the JavaScript environment used is created by thread 3096. This indicates that the virtual machine environment (vm/env) is being used across threads, leading to a multithreading safety issue.

If multi-thread check is not enabled, an error may not be reported in runtime even if there is an incorrect multi-thread implementation.

[h2]Constraints

Multi thread check can only be initiated through the command line.

[h2]Enabling Multi Thread Check

You can enable multi thread check in either of the following ways:

Method 1
Choose Run > Edit Configurations > Diagnostics and select Multi Thread Check.

Method 2

Run the following command:

hdc shell aa start -a {abilityName} -b {bundleName} -R

[h2]Using Multi Thread Check

Run or debug the app.
When a multithreading safety issue occurs, the crash log message will be displayed. You can click the link provided in the message to navigate to the code that causes the issue.

[h2]ArkCompiler Error Codes

The fatal message "Fatal: ecma_vm cannot run in multi-thread! thread:20296 currentThread:19953" indicates a multithreading safety issue. It means that the current thread (ID: 19953) uses a JavaScript environment created by thread 20296, resulting in cross-thread VM access.

目录