Android音频系统探究——从SoundPool到AudioHardware

对音频系统的探索起源于工作中遇到的一个bug。平时都是力求快速解决问题，不问原因。这次时间比较宽裕，正好借着解决问题的机会，把Android的音频系统了解一下。既然由bug引发，那就从bug开始说。

一. bug现象

    Android的照相机在拍照的时候会播放一个按键音。最近的一个MID项目（基于RK3188，Android 4.2）中，测试部门反馈，拍照时按键音播放异常情况如下：

    （1）进入应用程序以后，第一次拍照，没有按键音

    （2）连续拍照，有按键音

    （3）停止连拍，等待几秒钟后，再次拍照，又没有按键音

二. 问题简化

    看CameraApp代码可以知道，播放按键音使用了SoundPool类。做一个使用SoundPool播放声音的应用程序，界面上只有一个Button，点击后播放声音。这样就能确定这单纯是声音播放问题还是复合性问题。代码很简单：

protected void onCreate(Bundle savedInstanceState) {
        super.onCreate(savedInstanceState);
        setContentView(R.layout.activity_test_sound_pool);
        mSoundPool = new SoundPool(10, AudioManager.STREAM_SYSTEM, 5);
        mSoundId = mSoundPool.load(this, R.raw.camera_click, 1);   //这里R.raw.camera_click是ogg格式的音频资源

        vBtnShut = (Button) findViewById(R.id.btn_click);
        vBtnShut.setOnClickListener(new OnClickListener() {
            @Override
            public void onClick(View v) {
                mSoundPool.play(mSoundId, 1, 1, 0, 0, 1);
            }
        });
    }

protected void onCreate(Bundle savedInstanceState) {
		super.onCreate(savedInstanceState);
		setContentView(R.layout.activity_test_sound_pool);
		mSoundPool = new SoundPool(10, AudioManager.STREAM_SYSTEM, 5);
		mSoundId = mSoundPool.load(this, R.raw.camera_click, 1);   //这里R.raw.camera_click是ogg格式的音频资源
		
		vBtnShut = (Button) findViewById(R.id.btn_click);
		vBtnShut.setOnClickListener(new OnClickListener() {
			@Override
			public void onClick(View v) {
				mSoundPool.play(mSoundId, 1, 1, 0, 0, 1);
			}
		});
	}
	

    结果表明，BUG现象仍然是一样的。我们将BUG现象做一次简化：

    idle-->play failed-->idle-->play failed-->play success-->play success-->idle-->play failed-->...

    可以总结为，每间隔几秒钟后，第一次播放音频无声音输出。 

三. 初步分析

 

   理清了现象，简化了环境，我们可以开始分析问题了：

    显而易见的是，BUG非常规律，只有相隔几秒钟后的第一次播放才出现问题，与软件逻辑密切相关，可以排除硬件问题。本质上来讲，无论使用什么软件系统，声音播放的流程一般都是——用户指定要播放的声音数据，可能是文件，可能是Buffer；Audio系统对声音数据解码，可能采用软解码，也可能采用硬解码；将解码出来的数字音频信号传给功放设备，经过D/A转换后送到扬声器，声音就播放出来了。可以说,这个流程中的第一部分，是应用程序的行为；第二部分，是Android系统的职责；第三部分，是kernel中驱动的工作。应用程序的问题可以排除，现在要解决的疑问是，是解码程序出了问题，还是驱动程序出了问题？出现了什么情况，导致了idle后播放不出来？

    

四.  代码研究

1. Android Audio框架

    首先网络上找找资料，要搞清楚Android音频的框架层次结构，才容易定位问题。用图说明——

    有了大致的概念，开始以SoundPool为入口，摸清播放流程。其中在每个层次中要了解两点：数据如何传递，播放的动作如何执行。 也就是沿着SoundPool.load()和Sound.play()顺藤摸瓜。

2. SoundPool和AudioFlinger

    SoundPool.java基本是个空壳，直接使用了Native接口，代码没什么可看的。不过可以先看下这个类的介绍，就在SoundPool.java的开头，整一页的英文注释。幸运的是，很快就找到了我们需要看的资料：

/**
 * The SoundPool class manages and plays audio resources for applications.
 *
 * <p>A SoundPool is a collection of samples that can be loaded into memory
 * from a resource inside the APK or from a file in the file system. The
 * SoundPool library uses the MediaPlayer service to decode the audio
 * into a raw 16-bit PCM mono or stereo stream. This allows applications
 * to ship with compressed streams without having to suffer the CPU load
 * and latency of decompressing during playback.</p>
... ...
... ...
*/

/**
 * The SoundPool class manages and plays audio resources for applications.
 *
 * <p>A SoundPool is a collection of samples that can be loaded into memory
 * from a resource inside the APK or from a file in the file system. The
 * SoundPool library uses the MediaPlayer service to decode the audio
 * into a raw 16-bit PCM mono or stereo stream. This allows applications
 * to ship with compressed streams without having to suffer the CPU load
 * and latency of decompressing during playback.</p>
... ...
... ...
*/

     挑重要的说，SoundPool是Sample的集合，能把APK里的资源或者文件系统中的文件加载到内存中，使用MediaPlayer服务把音频解码成原始的16位PCM单声道或立体声数据流。好嘛，原来解码在这里就做了。还是看看代码实现吧，免得心里不踏实。

     不去理会Jni那些手续，直接看SoundPool.cpp。上面那个测试APK的代码，调用了SoundPool的load，play两个接口，就把声音播放出来了。load一次后，可多次播放，这两个接口之所以要分开，应该就是load做了解码。先看load的实现，为满足不同音频资源的需要，load被重载了，看其中一个就行了。

int SoundPool::load(int fd, int64_t offset, int64_t length, int priority)
{
    ALOGV("load: fd=%d, offset=%lld, length=%lld, priority=%d",
            fd, offset, length, priority);
    Mutex::Autolock lock(&mLock);
    sp<Sample> sample = new Sample(++mNextSampleID, fd, offset, length);
    mSamples.add(sample->sampleID(), sample);   //将sample对象加入管理
    doLoad(sample);   //load所在
    return sample->sampleID();
}

int SoundPool::load(int fd, int64_t offset, int64_t length, int priority)
{
    ALOGV("load: fd=%d, offset=%lld, length=%lld, priority=%d",
            fd, offset, length, priority);
    Mutex::Autolock lock(&mLock);
    sp<Sample> sample = new Sample(++mNextSampleID, fd, offset, length);
    mSamples.add(sample->sampleID(), sample);   //将sample对象加入管理
    doLoad(sample);   //load所在
    return sample->sampleID();
}

    数据处理角度来说，真正的load在doLoad中：

void SoundPool::doLoad(sp<Sample>& sample)
{
    ALOGV("doLoad: loading sample sampleID=%d", sample->sampleID());
    sample->startLoad();   //只是改变了状态
    mDecodeThread->loadSample(sample->sampleID());  //真正加载的地方
}

void SoundPool::doLoad(sp<Sample>& sample)
{
    ALOGV("doLoad: loading sample sampleID=%d", sample->sampleID());
    sample->startLoad();   //只是改变了状态
    mDecodeThread->loadSample(sample->sampleID());  //真正加载的地方
}

    看到了mDecodeThread，眼前一亮，很可能这里就是将ogg解码成PCM的地方了。所以进入loadSample看一看：

void SoundPoolThread::loadSample(int sampleID) {
    write(SoundPoolMsg(SoundPoolMsg::LOAD_SAMPLE, sampleID));
}

void SoundPoolThread::loadSample(int sampleID) {
    write(SoundPoolMsg(SoundPoolMsg::LOAD_SAMPLE, sampleID));
}

    只是消息传递而已，找到LOAD_SAMPLE消息处理的地方：

int SoundPoolThread::run() {
    ALOGV("run");
    for (;;) {
        SoundPoolMsg msg = read();
        ALOGV("Got message m=%d, mData=%d", msg.mMessageType, msg.mData);
        switch (msg.mMessageType) {
        case SoundPoolMsg::KILL:
            ALOGV("goodbye");
            return NO_ERROR;
        case SoundPoolMsg::LOAD_SAMPLE:   //在这里处理LOAD_SAMPLE
            doLoadSample(msg.mData);
            break;
        default:
            ALOGW("run: Unrecognized message %d\n",
                    msg.mMessageType);
            break;
        }
    }
}

int SoundPoolThread::run() {
    ALOGV("run");
    for (;;) {
        SoundPoolMsg msg = read();
        ALOGV("Got message m=%d, mData=%d", msg.mMessageType, msg.mData);
        switch (msg.mMessageType) {
        case SoundPoolMsg::KILL:
            ALOGV("goodbye");
            return NO_ERROR;
        case SoundPoolMsg::LOAD_SAMPLE:   //在这里处理LOAD_SAMPLE
            doLoadSample(msg.mData);
            break;
        default:
            ALOGW("run: Unrecognized message %d\n",
                    msg.mMessageType);
            break;
        }
    }
}

void SoundPoolThread::doLoadSample(int sampleID) {
    sp <Sample> sample = mSoundPool->findSample(sampleID);
    status_t status = -1;
    if (sample != 0) {
        status = sample->doLoad();
    }
    mSoundPool->notify(SoundPoolEvent(SoundPoolEvent::SAMPLE_LOADED, sampleID, status));
}

void SoundPoolThread::doLoadSample(int sampleID) {
    sp <Sample> sample = mSoundPool->findSample(sampleID);
    status_t status = -1;
    if (sample != 0) {
        status = sample->doLoad();
    }
    mSoundPool->notify(SoundPoolEvent(SoundPoolEvent::SAMPLE_LOADED, sampleID, status));
}

    看来最后是在sample->doLoad()中做的处理。进去看看，颇有惊喜：

status_t Sample::doLoad()
{
    uint32_t sampleRate;
    int numChannels;
    audio_format_t format;
    sp<IMemory> p;
    ALOGV("Start decode");
    if (mUrl) {
        p = MediaPlayer::decode(mUrl, &sampleRate, &numChannels, &format);
    } else {
        p = MediaPlayer::decode(mFd, mOffset, mLength, &sampleRate, &numChannels, &format);
        ALOGV("close(%d)", mFd);
        ::close(mFd);
        mFd = -1;
    }
    if (p == 0) {
        ALOGE("Unable to load sample: %s", mUrl);
        return -1;
    }
    ALOGV("pointer = %p, size = %u, sampleRate = %u, numChannels = %d",
            p->pointer(), p->size(), sampleRate, numChannels);

    if (sampleRate > kMaxSampleRate) {
       ALOGE("Sample rate (%u) out of range", sampleRate);
       return - 1;
    }

    if ((numChannels < 1) || (numChannels > 2)) {
        ALOGE("Sample channel count (%d) out of range", numChannels);
        return - 1;
    }

    //_dumpBuffer(p->pointer(), p->size());
    uint8_t* q = static_cast<uint8_t*>(p->pointer()) + p->size() - 10;
    //_dumpBuffer(q, 10, 10, false);

    mData = p;
    mSize = p->size();
    mSampleRate = sampleRate;
    mNumChannels = numChannels;
    mFormat = format;
    mState = READY;
    return 0;
}

status_t Sample::doLoad()
{
    uint32_t sampleRate;
    int numChannels;
    audio_format_t format;
    sp<IMemory> p;
    ALOGV("Start decode");
    if (mUrl) {
        p = MediaPlayer::decode(mUrl, &sampleRate, &numChannels, &format);
    } else {
        p = MediaPlayer::decode(mFd, mOffset, mLength, &sampleRate, &numChannels, &format);
        ALOGV("close(%d)", mFd);
        ::close(mFd);
        mFd = -1;
    }
    if (p == 0) {
        ALOGE("Unable to load sample: %s", mUrl);
        return -1;
    }
    ALOGV("pointer = %p, size = %u, sampleRate = %u, numChannels = %d",
            p->pointer(), p->size(), sampleRate, numChannels);

    if (sampleRate > kMaxSampleRate) {
       ALOGE("Sample rate (%u) out of range", sampleRate);
       return - 1;
    }

    if ((numChannels < 1) || (numChannels > 2)) {
        ALOGE("Sample channel count (%d) out of range", numChannels);
        return - 1;
    }

    //_dumpBuffer(p->pointer(), p->size());
    uint8_t* q = static_cast<uint8_t*>(p->pointer()) + p->size() - 10;
    //_dumpBuffer(q, 10, 10, false);

    mData = p;
    mSize = p->size();
    mSampleRate = sampleRate;
    mNumChannels = numChannels;
    mFormat = format;
    mState = READY;
    return 0;
}

    原来Sample请来了MediaPlayer帮其解码，并计算出了采样率和帧数。到这里数据已经准备好了。接下来我们就要看Framework能否把数据正确的传递给HAL，至于MediaPlayer是如何解码的我们先不研究。

    弄清楚SoundPool的Play做了什么，也就能找到HAL的代码了。下面看只看play中的关键代码：

int SoundPool::play(int sampleID, float leftVolume, float rightVolume,
        int priority, int loop, float rate)
{
    //...
    channel = allocateChannel_l(priority);
    //...
    channel->play(sample, channelID, leftVolume, rightVolume, priority, loop, rate);
    //...
}

int SoundPool::play(int sampleID, float leftVolume, float rightVolume,
        int priority, int loop, float rate)
{
	//...
	channel = allocateChannel_l(priority);
	//...
	channel->play(sample, channelID, leftVolume, rightVolume, priority, loop, rate);
	//...
}

    调用了SoundChannel的play.好读书而不求甚解，先把代码一路追下去，不作细究。

void SoundChannel::play(const sp<Sample>& sample, int nextChannelID, float leftVolume,
        float rightVolume, int priority, int loop, float rate)
{
    AudioTrack* newTrack;
    //....
    newTrack = new AudioTrack(streamType, sampleRate, sample->format(),
                channels, frameCount, AUDIO_OUTPUT_FLAG_FAST, callback, userData,                bufferFrames);
    //...
    mState = PLAYING;
    mAudioTrack->start();
    //...
}

void SoundChannel::play(const sp<Sample>& sample, int nextChannelID, float leftVolume,
        float rightVolume, int priority, int loop, float rate)
{
	AudioTrack* newTrack;
	//....
	newTrack = new AudioTrack(streamType, sampleRate, sample->format(),
                channels, frameCount, AUDIO_OUTPUT_FLAG_FAST, callback, userData,                bufferFrames);
	//...
	mState = PLAYING;
	mAudioTrack->start();
	//...
}

    SoundChannel::play创建了一个AudioTrack对象，在AudioTrack的构造函数中，调用了set，set又调用了createTrack_l。createTrack_I中，通过IAudioFlinger创建了一个IAudioTrack。关于AudioTrack和AudioFlinger是为何物，两者如何交换音频数据，就说来话长了。而且有很多大大分析得很详细，就不赘述了。有几篇写得很好——

• AudioTrack分析：http://www.cnblogs.com/innost/archive/2011/01/09/1931457.html
• AudioFlinger分析：http://www.cnblogs.com/innost/archive/2011/01/15/1936425.html
• AudioTrack如何与AudioFlinger交换数据: http://blog.chinaunix.net/uid-26533928-id-3052398.html

    阅读这些资料我们可以知道，Android Framework的音频子系统中，每一个音频流对应着一个AudioTrack类的一个实例，每个AudioTrack会在创建时注册到AudioFlinger中，由AudioFlinger把所有的AudioTrack进行混合（Mixer），然后输送到AudioHardware中进行播放。换言之，AudioFlinger是Audio系统的核心服务之一，起到了承上启下的衔接作用。

    我们现在已经让SoundPool牵线，抓到AudioFlinger这条大鱼。下面着重来看AudioFlinger如何向下调用AudioHardware的。

3. AudioFlinger与AudioHardware

    这里需要一点基础知识，先要了解Android的硬件抽象接口机制，才能理解AudioFlinger如何调用到AudioHardware，相关资料：

http://blog.csdn.net/myarrow/article/details/7175204

    因为对Audio系统一无所知，所以很惭愧用了反相的代码搜索，在hardware/xxx/audio目录下查找HAL_MODULE_INFO_SYM，然后反过来到framework找HAL_MODULE_INFO_SYM的id "AUDIO_HARDWARE_MODULE_ID"，过程非常笨拙，不足为道。他山之石可以攻玉，看到一篇好文，借助其中的一段分析来完成对AudioFlinger和AudioHardware关联的分析。原文地址：http://blog.csdn.net/xuesen_lin/article/details/8805108  

    当AudioPolicyService构造时创建了一个AudioPolicyDevice(mpAudioPolicyDev)并由此打开一个AudioPolicy(mpAudioPolicy)——这个Policy默认情况下的实现是legacy_audio_policy::policy(数据类型audio_policy)。同时legacy_audio_policy还包含了一个AudioPolicyInterface成员变量，它会被初始化为一个AudioPolicyManagerDefault。AudioPolicyManagerDefault的父类，即AudioPolicyManagerBase，它的构造函数中调用了mpClientInterface->loadHwModule()。

AudioPolicyManagerBase::AudioPolicyManagerBase(AudioPolicyClientInterface*clientInterface)…
{
    //......
    for (size_t i = 0; i < mHwModules.size();i++) {
       mHwModules[i]->mHandle = mpClientInterface->loadHwModule(mHwModules[i]->mName);
        if(mHwModules[i]->mHandle == 0) {
            continue;
        }
    //......
}

AudioPolicyManagerBase::AudioPolicyManagerBase(AudioPolicyClientInterface*clientInterface)…
{   
	//......
    for (size_t i = 0; i < mHwModules.size();i++) {
       mHwModules[i]->mHandle = mpClientInterface->loadHwModule(mHwModules[i]->mName);
        if(mHwModules[i]->mHandle == 0) {
            continue;
        }
	//......
}

    很明显的mpClientInterface这个变量在AudioPolicyManagerBase构造函数中做了初始化，再回溯追踪，可以发现它的根源在AudioPolicyService的构造函数中，对应的代码语句如下：

rc =mpAudioPolicyDev->create_audio_policy(mpAudioPolicyDev, &aps_ops, this, &mpAudioPolicy);

rc =mpAudioPolicyDev->create_audio_policy(mpAudioPolicyDev, &aps_ops, this, &mpAudioPolicy);      

    在这个场景下，函数create_audio_policy对应的是create_legacy_ap，并将传入的aps_ops组装到一个AudioPolicyCompatClient对象中，也就是mpClientInterface所指向的那个对象。
    换句话说，mpClientInterface->loadHwModule实际上调用的就是aps_ops->loadHwModule，即：

static audio_module_handle_t  aps_load_hw_module(void*service,const char *name)
{
    sp<IAudioFlinger> af= AudioSystem::get_audio_flinger();
    …
    return af->loadHwModule(name);
}

static audio_module_handle_t  aps_load_hw_module(void*service,const char *name)
{
    sp<IAudioFlinger> af= AudioSystem::get_audio_flinger();
    …
    return af->loadHwModule(name);
}

    AudioFlinger终于出现了，同样的情况也适用于mpClientInterface->openOutput，代码如下：

static audio_io_handle_t  aps_open_output(…)
{
    sp<IAudioFlinger> af= AudioSystem::get_audio_flinger();
    …
    return  af->openOutput((audio_module_handle_t)0,pDevices, pSamplingRate, pFormat, pChannelMask,
                         pLatencyMs, flags);
}

static audio_io_handle_t  aps_open_output(…)
{
    sp<IAudioFlinger> af= AudioSystem::get_audio_flinger();
    …
    return  af->openOutput((audio_module_handle_t)0,pDevices, pSamplingRate, pFormat, pChannelMask,
                         pLatencyMs, flags);
}

   现在前方就是AudioHardware了，终于打开了从APK到HAL的通路。

4. AudioHardware

    AudioHardware有两个内部类，AudioStreamOutALSA和AudioStreamInALSA，我们要解决的是声音播放的问题，看AudioStreamOutALSA即可。 AudioStreamOutALSA代码很清晰，很快找到了我们需要的代码，写PCM数据用的函数：

AudioHardware::AudioStreamOutALSA::AudioStreamOutALSA() :
    mHardware(0), mPcm(0), mMixer(0), mRouteCtl(0),
    mStandby(true), mDevices(0), mChannels(AUDIO_HW_OUT_CHANNELS),
    mSampleRate(AUDIO_HW_OUT_SAMPLERATE), mBufferSize(AUDIO_HW_OUT_PERIOD_BYTES),
    mDriverOp(DRV_NONE), mStandbyCnt(0)
{
#ifdef DEBUG_ALSA_OUT
    if(alsa_out_fp== NULL)
        alsa_out_fp = fopen("/data/data/out.pcm","a+");
    if(alsa_out_fp)
        ALOGI("------------>openfile success");
#endif
}
ssize_t AudioHardware::AudioStreamOutALSA::write(const void* buffer, size_t bytes)
{
    //...
#ifdef DEBUG_ALSA_OUT
    if(alsa_out_fp)
        fwrite(buffer,1,bytes,alsa_out_fp);
#endif
    //...
    if (mStandby) {
        open_l();   //重新open音频设备
       mStandby = false;
    }
    //...
    ret = pcm_write(mPcm,(void*) p, bytes);
    //...
}

AudioHardware::AudioStreamOutALSA::AudioStreamOutALSA() :
    mHardware(0), mPcm(0), mMixer(0), mRouteCtl(0),
    mStandby(true), mDevices(0), mChannels(AUDIO_HW_OUT_CHANNELS),
    mSampleRate(AUDIO_HW_OUT_SAMPLERATE), mBufferSize(AUDIO_HW_OUT_PERIOD_BYTES),
    mDriverOp(DRV_NONE), mStandbyCnt(0)
{
#ifdef DEBUG_ALSA_OUT
	if(alsa_out_fp== NULL)
		alsa_out_fp = fopen("/data/data/out.pcm","a+");
	if(alsa_out_fp)
		ALOGI("------------>openfile success");         
#endif                                                         
}
ssize_t AudioHardware::AudioStreamOutALSA::write(const void* buffer, size_t bytes)
{
	//...
#ifdef DEBUG_ALSA_OUT
	if(alsa_out_fp)
		fwrite(buffer,1,bytes,alsa_out_fp);
#endif 
	//...
	if (mStandby) {
		open_l();   //重新open音频设备
	   mStandby = false;
	}
	//...
	ret = pcm_write(mPcm,(void*) p, bytes);
	//...
}

    这里提供了一个很容易验证PCM数据是否正确的方法，打开DEBUG_ALSA_OUT开关后，可以将PCM流保存到“/data/data/out.pcm”文件中。到了验证数据是否正确的时候了，打开这个编译开关，得到out.pcm，将它pull到PC上，用coolEdit打开播放，发现正常播放正常。好了，我们现在可以知道，问题并不出在解码程序上了。那又是什么原因导致的呢，我们从write函数开始，研究播放的流程。

    首先，AudioStreamOutALSA的构造函数中将mStandby初始化为true。这个变量显然是作为记录音频设备待机状态用的。当mStandby==true时，每次调用write，都会调用open_l()重新开启一次音频设备，然后再做pcm_write。

    再看看open_l():

status_t AudioHardware::AudioStreamOutALSA::open_l()
{
    //...
    mPcm = mHardware->openPcmOut_l();
    if (mPcm == NULL) {
        return NO_INIT;
    }
    //...
}
struct pcm *AudioHardware::openPcmOut_l()
{
        //...
        mPcm = pcm_open(flags);
        //...
        if (!pcm_ready(mPcm)) {
            pcm_close(mPcm);
            //...
        }
    }
    return mPcm;
}

status_t AudioHardware::AudioStreamOutALSA::open_l()
{
 	//...
    mPcm = mHardware->openPcmOut_l();
    if (mPcm == NULL) {
        return NO_INIT;
    }
	//...
}
struct pcm *AudioHardware::openPcmOut_l()
{
		//...
        mPcm = pcm_open(flags);
		//...
        if (!pcm_ready(mPcm)) {
            pcm_close(mPcm);
            //...
        }
    }
    return mPcm;
}

open_l中调用了AudioHardware::openPcmOut_l,AudioHardware::openPcmOut_l中调用了pcm_open。再到pcm_open 中去看看：

struct pcm *pcm_open(unsigned flags)
{
    //... ...

    if (flags & PCM_IN) {
        dname = "/dev/snd/pcmC0D0c";
        channalFlags = -1;
        startCheckCount = 0;
    } else {
#ifdef SUPPORT_USB
        dname = "/dev/snd/pcmC1D0p";
#else
        dname = "/dev/snd/pcmC0D0p";
#endif
    }

    pcm->fd = open(dname, O_RDWR);
    if (pcm->fd < 0) {
        oops(pcm, errno, "cannot open device '%s'", dname);
        return pcm;
    }

    if (ioctl(pcm->fd, SNDRV_PCM_IOCTL_INFO, &info)) {
        oops(pcm, errno, "cannot get info - %s", dname);
        goto fail;
    }

    if (ioctl(pcm->fd, SNDRV_PCM_IOCTL_HW_PARAMS, &params)) {
        oops(pcm, errno, "cannot set hw params");
        goto fail;
    }
    if (ioctl(pcm->fd, SNDRV_PCM_IOCTL_SW_PARAMS, &sparams)) {
        oops(pcm, errno, "cannot set sw params");
        goto fail;
    }

fail:
    close(pcm->fd);
    pcm->fd = -1;
    return pcm;
}

struct pcm *pcm_open(unsigned flags)
{
	//... ...

    if (flags & PCM_IN) {
        dname = "/dev/snd/pcmC0D0c";
		channalFlags = -1;
		startCheckCount = 0;
    } else {
#ifdef SUPPORT_USB
        dname = "/dev/snd/pcmC1D0p";
#else
		dname = "/dev/snd/pcmC0D0p";
#endif
    }

    pcm->fd = open(dname, O_RDWR);
    if (pcm->fd < 0) {
        oops(pcm, errno, "cannot open device '%s'", dname);
        return pcm;
    }

    if (ioctl(pcm->fd, SNDRV_PCM_IOCTL_INFO, &info)) {
        oops(pcm, errno, "cannot get info - %s", dname);
        goto fail;
    }
    
    if (ioctl(pcm->fd, SNDRV_PCM_IOCTL_HW_PARAMS, &params)) {
        oops(pcm, errno, "cannot set hw params");
        goto fail;
    }
    if (ioctl(pcm->fd, SNDRV_PCM_IOCTL_SW_PARAMS, &sparams)) {
        oops(pcm, errno, "cannot set sw params");
        goto fail;
    }

fail:
    close(pcm->fd);
    pcm->fd = -1;
    return pcm;
}

    果然，这里就是操作设备节点的地方了。我们先在AudioStreamOutALSA的write中加打印信息，看看第一次播放和后续播放究竟有何不同。测试结果发现，每次播放不出声音的情况，都发生mStandby==true之后，这个时候做了一次打开音频设备的动作，但此时PCM数据是正确的。我们先来看看什么时候会导致mStandby==true。

<PRE class=cpp name="code"><PRE class=cpp name="code">status_t AudioHardware::AudioStreamOutALSA::standby()
{
        doStandby_l();
}

void AudioHardware::AudioStreamOutALSA::doStandby_l()
{

    if(!mStandby)
        mStandby = true;
    close_l();
}

void AudioHardware::AudioStreamOutALSA::close_l()
{
    if (mPcm) {
        mHardware->closePcmOut_l();
        mPcm = NULL;
    }
}</PRE><BR><BR></PRE>

<div class="dp-highlighter bg_cpp"><div class="bar"><div class="tools"><strong>[cpp]</strong> <a target=_blank class="ViewSource" title="view plain" href="http://blog.csdn.net/special_lin/article/details/12849637#">view plain</a><a target=_blank class="CopyToClipboard" title="copy" href="http://blog.csdn.net/special_lin/article/details/12849637#">copy</a><a target=_blank class="PrintSource" title="print" href="http://blog.csdn.net/special_lin/article/details/12849637#">print</a><a target=_blank class="About" title="?" href="http://blog.csdn.net/special_lin/article/details/12849637#">?</a></div></div><ol class="dp-cpp"><li class="alt"><span><span><PRE </span><span class="keyword">class</span><span>=cpp name=</span><span class="string">"code"</span><span>>status_t AudioHardware::AudioStreamOutALSA::standby()  </span></span></li><li><span>{  </span></li><li class="alt"><span>        doStandby_l();  </span></li><li><span>}  </span></li><li class="alt"><span>  </span></li><li><span></span><span class="keyword">void</span><span> AudioHardware::AudioStreamOutALSA::doStandby_l()  </span></li><li class="alt"><span>{  </span></li><li><span>  </span></li><li class="alt"><span>    </span><span class="keyword">if</span><span>(!mStandby)  </span></li><li><span>        mStandby = </span><span class="keyword">true</span><span>;  </span></li><li class="alt"><span>    close_l();  </span></li><li><span>}  </span></li><li class="alt"><span>  </span></li><li><span></span><span class="keyword">void</span><span> AudioHardware::AudioStreamOutALSA::close_l()  </span></li><li class="alt"><span>{  </span></li><li><span>    </span><span class="keyword">if</span><span> (mPcm) {  </span></li><li class="alt"><span>        mHardware->closePcmOut_l();  </span></li><li><span>        mPcm = NULL;  </span></li><li class="alt"><span>    }  </span></li><li><span>}</PRE><BR><BR>  </span></li></ol></div><pre style="DISPLAY: none" class="cpp" name="code"><div class="dp-highlighter bg_cpp"><div class="bar"><div class="tools"><strong>[cpp]</strong> <a target=_blank class="ViewSource" title="view plain" href="http://blog.csdn.net/special_lin/article/details/12849637#">view plain</a><a target=_blank class="CopyToClipboard" title="copy" href="http://blog.csdn.net/special_lin/article/details/12849637#">copy</a><a target=_blank class="PrintSource" title="print" href="http://blog.csdn.net/special_lin/article/details/12849637#">print</a><a target=_blank class="About" title="?" href="http://blog.csdn.net/special_lin/article/details/12849637#">?</a></div></div><ol class="dp-cpp"><li class="alt"><span><span>status_t AudioHardware::AudioStreamOutALSA::standby()  </span></span></li><li><span>{  </span></li><li class="alt"><span>        doStandby_l();  </span></li><li><span>}  </span></li><li class="alt"><span>  </span></li><li><span></span><span class="keyword">void</span><span> AudioHardware::AudioStreamOutALSA::doStandby_l()  </span></li><li class="alt"><span>{  </span></li><li><span>  </span></li><li class="alt"><span>    </span><span class="keyword">if</span><span>(!mStandby)  </span></li><li><span>        mStandby = </span><span class="keyword">true</span><span>;  </span></li><li class="alt"><span>    close_l();  </span></li><li><span>}  </span></li><li class="alt"><span>  </span></li><li><span></span><span class="keyword">void</span><span> AudioHardware::AudioStreamOutALSA::close_l()  </span></li><li class="alt"><span>{  </span></li><li><span>    </span><span class="keyword">if</span><span> (mPcm) {  </span></li><li class="alt"><span>        mHardware->closePcmOut_l();  </span></li><li><span>        mPcm = NULL;  </span></li><li class="alt"><span>    }  </span></li><li><span>}  </span></li></ol></div><pre style="DISPLAY: none" class="cpp" name="code">status_t AudioHardware::AudioStreamOutALSA::standby()
{
        doStandby_l();
}

void AudioHardware::AudioStreamOutALSA::doStandby_l()
{

	if(!mStandby)
		mStandby = true;
    close_l();
}

void AudioHardware::AudioStreamOutALSA::close_l()
{
    if (mPcm) {
        mHardware->closePcmOut_l();
        mPcm = NULL;
    }
}

    好了，现在我们可以确定，mStandby是在调用standby的时候被设置生true了。如果不总是重新打开音频设备，会不会变正常？做了一个实验，把standby函数体的代码都注释掉。这样修改后，果然开机只有一次声音播放不出来，那就是第一次。每隔一段时间，声音就播不出来的问题不见了。

     其实到现在，问题已经定位出来了。这个问题属于kernel问题，不再属于Framework了。但是还是想弄清楚，standby为什么隔一段时间被调用一次，是被谁调用的。经过一系列反查，找到了standby的真正调用处,AudioFlinger的播放线程中。具体怎么查的，还是要参考HAL知识去，就不重复记载了。

void AudioFlinger::PlaybackThread::threadLoop_standby()
{
    ALOGV("Audio hardware entering standby, mixer %p, suspend count %d", this, mSuspended);
    mOutput->stream->common.standby(&mOutput->stream->common);
}

bool AudioFlinger::PlaybackThread::threadLoop()
{
    // ... ...
    while (!exitPending())
<PRE class=cpp name="code"> <SPAN style="FONT-FAMILY: Arial, Helvetica, sans-serif">{</SPAN></PRE>       if (CC_UNLIKELY((!mActiveTracks.size() && systemTime() > standbyTime) ||                    isSuspended())) {            if (!mStandby) {                threadLoop_standby();                mStandby = true;            }//... ...}//... ...standbyTime = systemTime() + standbyDelay;//... ...}// ... ...}

void AudioFlinger::PlaybackThread::threadLoop_standby()
{
    ALOGV("Audio hardware entering standby, mixer %p, suspend count %d", this, mSuspended);
    mOutput->stream->common.standby(&mOutput->stream->common);
}

bool AudioFlinger::PlaybackThread::threadLoop()
{
	// ... ...
	while (!exitPending())
<div class="dp-highlighter bg_cpp"><div class="bar"><div class="tools"><strong>[cpp]</strong> <a target=_blank class="ViewSource" title="view plain" href="http://blog.csdn.net/special_lin/article/details/12849637#">view plain</a><a target=_blank class="CopyToClipboard" title="copy" href="http://blog.csdn.net/special_lin/article/details/12849637#">copy</a><a target=_blank class="PrintSource" title="print" href="http://blog.csdn.net/special_lin/article/details/12849637#">print</a><a target=_blank class="About" title="?" href="http://blog.csdn.net/special_lin/article/details/12849637#">?</a></div></div><ol class="dp-cpp"><li class="alt"><span><span><SPAN style=</span><span class="string">"FONT-FAMILY: Arial, Helvetica, sans-serif"</span><span>>{</SPAN>  </span></span></li></ol></div><pre style="DISPLAY: none" class="cpp" name="code">	<span style="font-family:Arial, Helvetica, sans-serif;"><span>{</span></span>

if (CC_UNLIKELY((!mActiveTracks.size() && systemTime() > standbyTime) || isSuspended())) { if (!mStandby) { threadLoop_standby(); mStandby = true; }//... ...}//... ...standbyTime = systemTime() + standbyDelay;//... ...}// ... ...}

    这里我们看到了，standby是由AudioFlinger控制的，一旦满足以下条件后，没有AudioTrack处于活动状态并且已经到达了standbyTime这个时间就进入Standby模式。那么standbyTime=systemTime() + standbyDelay，也就是过了standbyDelay这段时间后，音频系统将进入待机，关闭音频设备。最后找到standbyDelay的值是多少。

   AudioFlinger::PlaybackThread构造函数中，将standbyDelay初始化，standbyDelay(AudioFlinger::mStandbyTimeInNsecs), 

   AudioFlinger这个类第一次被引用时，就对成员变量mStandbyTimeInNsecs 进行了初始化

void AudioFlinger::onFirstRef()
{
    //... ...
    /* TODO: move all this work into an Init() function */
    char val_str[PROPERTY_VALUE_MAX] = { 0 };
    if (property_get("ro.audio.flinger_standbytime_ms", val_str, NULL) >= 0) {
        uint32_t int_val;
        if (1 == sscanf(val_str, "%u", &int_val)) {
            mStandbyTimeInNsecs = milliseconds(int_val);
            ALOGI("Using %u mSec as standby time.", int_val);
        } else {
            mStandbyTimeInNsecs = kDefaultStandbyTimeInNsecs;
            ALOGI("Using default %u mSec as standby time.",
                    (uint32_t)(mStandbyTimeInNsecs / 1000000));
        }
    }

    mMode = AUDIO_MODE_NORMAL;
}

void AudioFlinger::onFirstRef()
{
    //... ...
    /* TODO: move all this work into an Init() function */
    char val_str[PROPERTY_VALUE_MAX] = { 0 };
    if (property_get("ro.audio.flinger_standbytime_ms", val_str, NULL) >= 0) {
        uint32_t int_val;
        if (1 == sscanf(val_str, "%u", &int_val)) {
            mStandbyTimeInNsecs = milliseconds(int_val);
            ALOGI("Using %u mSec as standby time.", int_val);
        } else {
            mStandbyTimeInNsecs = kDefaultStandbyTimeInNsecs;
            ALOGI("Using default %u mSec as standby time.",
                    (uint32_t)(mStandbyTimeInNsecs / 1000000));
        }
    }

    mMode = AUDIO_MODE_NORMAL;
}

    如果有ro.audio.flinger_standbytime_ms这个属性，就按这个属性值设定stand by的idle time（很可能是OEM代码），如果没有，取kDefaultStandbyTimeInNsecs的值。kDefaultStandbyTimeInNsecs是个常量，3s：

static const nsecs_t kDefaultStandbyTimeInNsecs = seconds(3);

static const nsecs_t kDefaultStandbyTimeInNsecs = seconds(3);

五. 结论及收获

    通过分析研究Android系统代码，我们虽然最终没有解决问题，但是已经定位出了问题所在的层次，确定这是一个驱动的BUG。Framework工程师的任务至此完成了。问题交付给驱动工程师，经过排查发现，是PA没有打开造成的问题。

    经验可以带来技巧，如果下次遇到类似问题，我们可以直接在AudioHardware中截获PCM，通过判断解码出的PCM流是否正确，较快速的定位到问题所在——是MediaPlayer Codec、AudioSystem、还是Driver。

 

原文转自：http://blog.csdn.net/special_lin/article/details/12849637