ANDROID音频系统散记之一：A2dpAudioInterface

写在之前

本来有打算写写Android音频系统的，但是仔细研究了如下链接的三篇文章，果断中断了我的想法。毫不夸张来说，这是我看过的最好的阐述Android音频系统的文章了，简练精辟，将音频系统各个方面的重要的脉络都描述出来了。有这三篇文章，理解Android音频系统何止加快了10倍。

Android Audio System 之一：AudioTrack如何与AudioFlinger交换音频数据

Android Audio System 之二：AudioFlinger

Android Audio System 之三： AudioPolicyService 和 AudioPolicyManager

A2dpAudioInterface

Android音频系统有两大服务：一是AudioFlinger，二是AudioPolicyService。AudioFlinger负责向下访问AudioHardwareInterface，实现音频PCM数据的混音/输入/输出，实现音量调节；AudioPolicyService负责音频输入输出设备的连接状态，音频策略调度即音频设备（如本地CODEC、Bluetooth A2DP、Headset）的切换策略（注意它只是负责策略，真正的切换操作是在AudioFlinger中的openOutput，毕竟AudioFlinger负责操作底层音频硬件）。AudioPolicyService在以后的章节详细分析，这里主要探讨A2DP-Audio是如何注册到AudioFlinger中，并简要提及音频PCM数据流向。

好的平台软件应有这样的一个抽象层：向下提供一套固定的接口，不同的硬件设备根据这些接口实现各自的方法，然后注册到这个抽象层中去。这样对于上层应用而言并没有任何区别，因为上层只需调用抽象层接口就行了，不管底层硬件的差异性。AudioFlinger就是这样的一个抽象层，无论底层是ALSA设备还是BluetoothHeadset，上层都只会看到AudioFlinger的接口。至于何时切换到ALSA设备何时切换到BluetoothHeadset，这就属于音频策略调度范畴了即AudioPolicyService。

AudioFlinger::AudioFlinger()
    : BnAudioFlinger(),
        mAudioHardware(0), mMasterVolume(1.0f), mMasterMute(false), mNextUniqueId(1)
{
    mHardwareStatus = AUDIO_HW_IDLE;

    mAudioHardware = AudioHardwareInterface::create();
    ......

再看AudioHardwareInterface::create()：

AudioHardwareInterface* AudioHardwareInterface::create()
{
    /*
     * FIXME: This code needs to instantiate the correct audio device
     * interface. For now - we use compile-time switches.
     */
    AudioHardwareInterface* hw = 0;
    char value[PROPERTY_VALUE_MAX];

#ifdef GENERIC_AUDIO
    hw = new AudioHardwareGeneric();
#else
    // if running in emulation - use the emulator driver
    if (property_get("ro.kernel.qemu", value, 0)) {
        LOGD("Running in emulation - using generic audio driver");
        hw = new AudioHardwareGeneric();
    }
    else {
        LOGV("Creating Vendor Specific AudioHardware");
        hw = createAudioHardware();
    }
#endif
    if (hw->initCheck() != NO_ERROR) {
        LOGW("Using stubbed audio hardware. No sound will be produced.");
        delete hw;
        hw = new AudioHardwareStub();
    }
    
#ifdef WITH_A2DP
    hw = new A2dpAudioInterface(hw);
#endif

#ifdef ENABLE_AUDIO_DUMP
    // This code adds a record of buffers in a file to write calls made by AudioFlinger.
    // It replaces the current AudioHardwareInterface object by an intermediate one which
    // will record buffers in a file (after sending them to hardware) for testing purpose.
    // This feature is enabled by defining symbol ENABLE_AUDIO_DUMP.
    // The output file is set with setParameters("test_cmd_file_name=<name>"). Pause are not recorded in the file.
    LOGV("opening PCM dump interface");
    hw = new AudioDumpInterface(hw);    // replace interface
#endif
    return hw;
}

这个函数我在 ANDROID2.3音频系统HAL有简要的分析，现在我们接着往下看看A2DP的注册：

hw = new A2dpAudioInterface(hw);

注意红色部分hw，为什么A2dpAudioInterface还需要createAudioHardware()打开的AudioHardwareInterface（我们假设这是ALSA设备接口）呢？如我们所知，BluetoothA2DP与ALSA设备并不走同一套接口，因此Android的设计者就把ALSA设备接口扔到A2DP接口里面管理了。这又是如何管理呢？简单来说，就是根据上层传下来的参数devices，判断devices是否是DEVICE_OUT_BLUETOOTH_A2DP，如果是则走A2DP接口，如果不是则走ALSA设备接口。例如需要打开一个音频输出流时：

AudioStreamOut* A2dpAudioInterface::openOutputStream(
        uint32_t devices, int *format, uint32_t *channels, uint32_t *sampleRate, status_t *status)
{
    if (!AudioSystem::isA2dpDevice((AudioSystem::audio_devices)devices)) {
        LOGV("A2dpAudioInterface::openOutputStream() open HW device: %x", devices);
        return mHardwareInterface->openOutputStream(devices, format, channels, sampleRate, status);
    }

    status_t err = 0;

    // only one output stream allowed
    if (mOutput) {
        if (status)
            *status = -1;
        return NULL;
    }

    // create new output stream
    A2dpAudioStreamOut* out = new A2dpAudioStreamOut();
    if ((err = out->set(devices, format, channels, sampleRate)) == NO_ERROR) {
        mOutput = out;
        mOutput->setBluetoothEnabled(mBluetoothEnabled);
        mOutput->setSuspended(mSuspended);
    } else {
        delete out;
    }

    if (status)
        *status = err;
    return mOutput;
}

当上层传下来的devices不属于A2DP设备时，则return mHardwareInterface->openOutputStream(devices, format, channels, sampleRate, status);其中mHardwareInterface保存的是ALSA的hw。否则A2dpAudioStreamOut* out = new A2dpAudioStreamOut();为A2DP打开一个音频输出流。

liba2dp

到了A2dpAudioInterface这层，就是访问BlueZ的音频操作接口了，主要是external\bluetooth\bluez\audio\liba2dp.c。liba2dp.c代码或许很复杂，我也没有深入了解过，但是接口却非常简单易用。看liba2dp.h，仅仅只有几个接口：

int a2dp_init(int rate, int channels, a2dpData* dataPtr);
void a2dp_set_sink(a2dpData data, const char* address);
int a2dp_write(a2dpData data, const void* buffer, int count);
int a2dp_stop(a2dpData data);
void a2dp_cleanup(a2dpData data);

a2dp_init：根据传入来的采样率rate，声道数channels初始化一个a2dpData；

a2dp_set_sink：绑定一个蓝牙地址address到a2dpData上；

a2dp_write：往a2dp写入音频PCM数据；

a2dp_stop：停止a2dp播放。

例如，每当有音频PCM数据需要送入Bluetooth时：

ssize_t A2dpAudioInterface::A2dpAudioStreamOut::write(const void* buffer, size_t bytes)
{
    Mutex::Autolock lock(mLock);

    size_t remaining = bytes;
    status_t status = -1;

    if (!mBluetoothEnabled || mClosing || mSuspended) {
        LOGV("A2dpAudioStreamOut::write(), but bluetooth disabled \
               mBluetoothEnabled %d, mClosing %d, mSuspended %d",
                mBluetoothEnabled, mClosing, mSuspended);
        goto Error;
    }

    status = init();
    if (status < 0)
        goto Error;

    while (remaining > 0) {
        status = a2dp_write(mData, buffer, remaining);
        if (status <= 0) {
            LOGE("a2dp_write failed err: %d\n", status);
            goto Error;
        }
        remaining -= status;
        buffer = ((char *)buffer) + status;
    }

    mStandby = false;

    return bytes;

Error:
    // Simulate audio output timing in case of error
    usleep(((bytes * 1000 )/ frameSize() / sampleRate()) * 1000);

    return status;
}

核心语句：status = a2dp_write(mData, buffer, remaining); 只需要传入音频数据的首地址和大小就行了。

该函数在AudioFlinger::MixerThread::threadLoop()调用，下面简要介绍音频数据从上层到底层硬件设备的传输流向过程。