当前位置：首页 > news >正文

Android audio(6)-audiopolicyservice介绍

news 来源：原创 2025/9/19 9:50:00

AudioPolicyService 是策略的制定者，比如某种 Stream 类型不同设备的音量（index/DB）是多少、某种 Stream 类型的音频数据流对应什么设备等等。而 AudioFlinger 则是策略的执行者，例如具体如何与音频设备通信，维护现有系统中可用的音频设备，以及多个音频流的混音处理，音频数据流的算法处理（重采样，音效），音量调节，音频数据搬运等等都由它完成。

AudioPolicyService 根据用户配置来指导 AudioFlinger 加载设备接口，起到路由功能。在 Android Audio 系统中主要完成以下几个任务：
1）管理输入输出设备，包括设备的连接/断开状态，设备的选择和切换等
2）管理系统的音频策略，比如通话时播放音乐、或者播放音乐时来电话的一系列处理
3）管理系统的音量/静音
4）上层的一些音频参数也可以通过AudioPolicyService设置到底层
5）管理音效和流的配置

一、初始化

与AudioPolicyService 生命周期关联的有三个函数，构造函数 AudioPolicyService、onFirstRef 和初始化有关，析构函数 ~AudioPolicyService 和销毁有关。通过对android代码的学习，我们可以看到与成员变量相关的初始化都是在构造函数里面做，而与业务相关的初始化都是在onFirstRef 里面完成。这是关注点分离思想的一种体现。

源码位置：/frameworks/av/services/audiopolicy/service/AudioPolicyService.cpp

1、构造函数

AudioPolicyService::AudioPolicyService(): BnAudioPolicyService(), // 定义在IAudioPolicyService.h中， 作为Binder调用的Bn端mAudioPolicyManager(NULL),  // APMmAudioPolicyClient(NULL), mPhoneState(AUDIO_MODE_INVALID), // 通话状态 mCaptureStateNotifier(false) {
}

2、onFirstRef

void AudioPolicyService::onFirstRef()
{{Mutex::Autolock _l(mLock);//一个系统的服务通常都需要线程去承载，不断处理命令。// 启动音频命令线程，audio命令相关，如音量控制、音频参数设置mAudioCommandThread = new AudioCommandThread(String8("ApmAudio"), this);// 启动输出命令线程，Output管理mOutputCommandThread = new AudioCommandThread(String8("ApmOutput"), this);mAudioPolicyClient = new AudioPolicyClient(this);mAudioPolicyManager = createAudioPolicyManager(mAudioPolicyClient);}sp<AudioPolicyEffects> audioPolicyEffects = new AudioPolicyEffects();sp<UidPolicy> uidPolicy = new UidPolicy(this);sp<SensorPrivacyPolicy> sensorPrivacyPolicy = new SensorPrivacyPolicy(this);{Mutex::Autolock _l(mLock);mAudioPolicyEffects = audioPolicyEffects;mUidPolicy = uidPolicy;mSensorPrivacyPolicy = sensorPrivacyPolicy;}uidPolicy->registerSelf();sensorPrivacyPolicy->registerSelf();
}

可以看到这里创建了两个 AudioCommandThread 和 AudioPolicyManager，而在创建 AudioPolicyManager 时创建了AudioPolicyClient作为参数传入其中。

3、析构函数

AudioPolicyService::~AudioPolicyService()
{mAudioCommandThread->exit();mOutputCommandThread->exit();destroyAudioPolicyManager(mAudioPolicyManager);delete mAudioPolicyClient;mNotificationClients.clear();mAudioPolicyEffects.clear();mUidPolicy->unregisterSelf();mSensorPrivacyPolicy->unregisterSelf();mUidPolicy.clear();mSensorPrivacyPolicy.clear();
}

这里主要是做销毁，情基本是对初始化创建对象的回收。

二、AudioPolicyManager

AudioPolicyManager是 AudioPolicyService 服务进程下的功能模块，主要负责解析各种 Audio 配置 xml 文件，例如 audio_policy_configuration.xml 中音频的设备、流以及路由关系等。从audiopolicyservice和audiopolicymanager的设计我们可以学到，服务承载实体（进程，service等）可以和服务的业务逻辑代码分离！！！！！服务实体是为了实现异步以及序列化操作

1、AudioPolicyManager创建
上面的代码可以看到 AudioPolicyManager 是在 AudioPolicyService 的 onFirstRef 中调用 createAudioPolicyManager() 方法创建的。传入参数为 new AudioPolicyClient(this)。这样audiopolicymanager就可以调用audiopolicyservice的接口。

createAudioPolicyManager

源码位置：/frameworks/av/services/audiopolicy/manager/AudioPolicyFactory.cpp

extern "C" AudioPolicyInterface* createAudioPolicyManager(AudioPolicyClientInterface *clientInterface)
{AudioPolicyManager *apm = new AudioPolicyManager(clientInterface);status_t status = apm->initialize();if (status != NO_ERROR) {delete apm;apm = nullptr;}return apm;
}

AudioPolicyClient
AudioPolicyClient 类定义在 AudioPolicyService.h 中。源码位置：/frameworks/av/services/audiopolicy/service/AudioPolicyService.h

class AudioPolicyClient : public AudioPolicyClientInterface
{
public:
explicit AudioPolicyClient(AudioPolicyService *service) : mAudioPolicyService(service) {}
…
}

它的实现在 frameworks/av/services/audiopolicy/service/AudioPolicyClientImpl.cpp 中。

AudioPolicyManager

源码位置：/frameworks/av/services/audiopolicy/managerdefault/AudioPolicyManager.cpp

AudioPolicyManager::AudioPolicyManager(AudioPolicyClientInterface *clientInterface, bool /*forTesting*/):mUidCached(AID_AUDIOSERVER), // no need to call getuid(), there's only one of us running.mpClientInterface(clientInterface),mLimitRingtoneVolume(false), mLastVoiceVolume(-1.0f),mA2dpSuspended(false),mConfig(mHwModulesAll, mOutputDevicesAll, mInputDevicesAll, mDefaultOutputDevice),mAudioPortGeneration(1),mBeaconMuteRefCount(0),mBeaconPlayingRefCount(0),mBeaconMuted(false),mTtsOutputAvailable(false),mMasterMono(false),mMusicEffectOutput(AUDIO_IO_HANDLE_NONE)
{
}AudioPolicyManager::AudioPolicyManager(AudioPolicyClientInterface *clientInterface): AudioPolicyManager(clientInterface, false /*forTesting*/)
{loadConfig();
}void AudioPolicyManager::loadConfig() {// 处理音频配置xmlif (deserializeAudioPolicyXmlConfig(getConfig()) != NO_ERROR) {ALOGE("无法加载音频策略配置文件，设置默认值");getConfig().setDefault();}
}

对于解析音频配置相关的 xml 文件我们会在后面单独分析。

三、Engine创建

Engine 是主要负责音频路由/音量相关的业务。Engine 的创建是在 AudioPolicyManager 的 initialize() 中。

status_t AudioPolicyManager::initialize() {{auto engLib = EngineLibrary::load("libaudiopolicyengine" + getConfig().getEngineLibraryNameSuffix() + ".so");……mEngine = engLib->createEngine();……mEngine->setObserver(this);status_t status = mEngine->initCheck();……}……return status;
}

EngineLibrary.cpp：createEngine
源码位置：/frameworks/av/services/audiopolicy/managerdefault/EngineLibrary.cpp

bool EngineLibrary::init(std::string libraryPath)
{……mCreateEngineInstance = (EngineInterface* (*)())dlsym(mLibraryHandle, "createEngineInstance");……return true;}EngineInstance EngineLibrary::createEngine()
{if (mCreateEngineInstance == nullptr || mDestroyEngineInstance == nullptr) {return EngineInstance();}return EngineInstance(mCreateEngineInstance(),[lib = shared_from_this(), destroy = mDestroyEngineInstance] (EngineInterface* e) {destroy(e);});
}

EngineInstance .cpp：createEngineInstance
源码位置：/frameworks/av/services/audiopolicy/engineconfigurable/src/EngineInstance.cpp

EngineInstance *EngineInstance::getInstance()
{static EngineInstance instance;return &instance;
}Engine *EngineInstance::getEngine() const
{static Engine engine;return &engine;
}template <>
EngineInterface *EngineInstance::queryInterface() const
{return getEngine()->queryInterface<EngineInterface>();
}extern "C" EngineInterface* createEngineInstance()
{return audio_policy::EngineInstance::getInstance()->queryInterface<EngineInterface>();
}

在这里最后创建了 Engine。
Engine.cpp
源码位置：/frameworks/av/services/audiopolicy/enginedefault/src/Engine.cpp

Engine::Engine()
{auto result = EngineBase::loadAudioPolicyEngineConfig();auto legacyStrategy = getLegacyStrategy();for (const auto &strategy : legacyStrategy) {mLegacyStrategyMap[getProductStrategyByName(strategy.name)] = strategy.id;}
}

这里调用了一个比较重要的函数 loadAudioPolicyEngineConfig()。

EngineBase.cpp： loadAudioPolicyEngineConfig()

engineConfig::ParsingResult EngineBase::loadAudioPolicyEngineConfig()
{auto loadVolumeConfig = [](auto &volumeGroups, auto &volumeConfig) {// 确保名称唯一LOG_ALWAYS_FATAL_IF(std::any_of(std::begin(volumeGroups), std::end(volumeGroups), [&volumeConfig](const auto &volumeGroup) {return volumeConfig.name == volumeGroup.second->getName(); }), "group name %s defined twice, review the configuration", volumeConfig.name.c_str());// 表示当前VolumeGroup还没有被加载过，开始创建加载sp<VolumeGroup> volumeGroup = new VolumeGroup(volumeConfig.name, volumeConfig.indexMin, volumeConfig.indexMax);volumeGroups[volumeGroup->getId()] = volumeGroup;for (auto &configCurve : volumeConfig.volumeCurves) {device_category deviceCat = DEVICE_CATEGORY_SPEAKER;if (!DeviceCategoryConverter::fromString(configCurve.deviceCategory, deviceCat)) {continue;}sp<VolumeCurve> curve = new VolumeCurve(deviceCat);for (auto &point : configCurve.curvePoints) {curve->add({point.index, point.attenuationInMb});}volumeGroup->add(curve);}return volumeGroup;};auto addSupportedAttributesToGroup = [](auto &group, auto &volumeGroup, auto &strategy) {for (const auto &attr : group.attributesVect) {strategy->addAttributes({group.stream, volumeGroup->getId(), attr});volumeGroup->addSupportedAttributes(attr);}};auto checkStreamForGroups = [](auto streamType, const auto &volumeGroups) {const auto &iter = std::find_if(std::begin(volumeGroups), std::end(volumeGroups), [&streamType](const auto &volumeGroup) {const auto& streams = volumeGroup.second->getStreamTypes();return std::find(std::begin(streams), std::end(streams), streamType) != std::end(streams);});return iter != end(volumeGroups);};// 这里开始进行解析，最终会解析出来策略、标准、标准类型以及音量组四个内容auto result = engineConfig::parse();if (result.parsedConfig == nullptr) {// 如果上面没有解析没有找到配置，使用默认的配置engineConfig::Config config = gDefaultEngineConfig;android::status_t ret = engineConfig::parseLegacyVolumes(config.volumeGroups);result = {std::make_unique<engineConfig::Config>(config), static_cast<size_t>(ret == NO_ERROR ? 0 : 1)};}……engineConfig::VolumeGroup defaultVolumeConfig;engineConfig::VolumeGroup defaultSystemVolumeConfig;// 循环解析所有的音量组for (auto &volumeConfig : result.parsedConfig->volumeGroups) {// 保存未在配置中定义的流的默认音量配置 将music和patch作为未定义流类型的默认配置if (volumeConfig.name.compare("AUDIO_STREAM_MUSIC") == 0) {defaultVolumeConfig = volumeConfig;}if (volumeConfig.name.compare("AUDIO_STREAM_PATCH") == 0) {defaultSystemVolumeConfig = volumeConfig;}// 这里调用上面第一个lamabda表达式// 这里定义的mVolumeGroups是一个map容器，其中second是VolumeGroup指针,定义在VolumeGroup.h中// 这里调用这个lambda表达式是为了讲volumeConfig中包含的volumeGroups解析到mVolumeGroups中loadVolumeConfig(mVolumeGroups, volumeConfig);}// 循环遍历所有的音频策略for (auto& strategyConfig : result.parsedConfig->productStrategies) {sp<ProductStrategy> strategy = new ProductStrategy(strategyConfig.name);// 查找该策略是否有相应的音量组for (const auto &group : strategyConfig.attributesGroups) {const auto &iter = std::find_if(begin(mVolumeGroups), end(mVolumeGroups), [&group](const auto &volumeGroup) {return group.volumeGroup == volumeGroup.second->getName(); });sp<VolumeGroup> volumeGroup = nullptr;// 如果没有为此策略提供音量组，则使用音乐音量组配置创建一个新的音量组（视为默认设置） if (iter == end(mVolumeGroups)) {engineConfig::VolumeGroup volumeConfig;if (group.stream >= AUDIO_STREAM_PUBLIC_CNT) {volumeConfig = defaultSystemVolumeConfig;} else {volumeConfig = defaultVolumeConfig;}volumeConfig.name = group.volumeGroup;volumeGroup = loadVolumeConfig(mVolumeGroups, volumeConfig);} else {volumeGroup = iter->second;}if (group.stream != AUDIO_STREAM_DEFAULT) {// 可以将旧流一次分组 LOG_ALWAYS_FATAL_IF(checkStreamForGroups(group.stream, mVolumeGroups), "stream %s already assigned to a volume group, " "review the configuration", toString(group.stream).c_str());volumeGroup->addSupportedStream(group.stream);}// 为策略添加相应的属性addSupportedAttributesToGroup(group, volumeGroup, strategy);}product_strategy_t strategyId = strategy->getId();mProductStrategies[strategyId] = strategy;}// 将新创建的strategy保存到mProductStrategies中并分配一个单独的IDmProductStrategies.initialize();return result;
}

总结

通过前面几篇文章的学习，我们也了解了音频服务初始化的大致流程，下面总结一下：

audioserver实例化audioflinger
audioserver实例化audiopolicyservice
audiopolicyservice创建audiopolicymanager
audiopolicymanager解析xml配置文件
audiopolicymanager创建engine
audiopolicymanager调用audioflinger接口打开audio interface
audiopolicymanager调用audioflinger接口打开音频通道（output，注意区分stream和output的概念）

经过上面的流程系统音频服务已经启动处于ready状态，如果有应用需要播放则会通过服务选择合适的硬件播出。