ANDROID音频系统散记之四：4.0音频系统HAL初探

谁借莪１个温暖的怀抱￠ 2022-05-31 09:29 410阅读 0赞

昨天（2011-11-15）发布了Android4.0的源码，今天download下来，开始挺进4.0时代。简单看了一下，发现音频系统方面与2.3的有较多地方不同，下面逐一描述。

## 一、代码模块位置 ##

###  ###

### 1、AudioFlinger ###

**\[plain\]** [ view plain][view plain] [copy][view plain]

1.  frameworks/base/services/audioflinger/
2.  \+-- Android.mk
3.  \+-- AudioBufferProvider.h
4.  \+-- AudioFlinger.cpp
5.  \+-- AudioFlinger.h
6.  \+-- AudioMixer.cpp
7.  \+-- AudioMixer.h
8.  \+-- AudioPolicyService.cpp
9.  \+-- AudioPolicyService.h
10. \+-- AudioResampler.cpp
11. \+-- AudioResamplerCubic.cpp
12. \+-- AudioResamplerCubic.h
13. \+-- AudioResampler.h
14. \+-- AudioResamplerSinc.cpp
15. \+-- AudioResamplerSinc.h

AudioFlinger相关代码，好像这部分与2.3相差不大，至少接口是兼容的。值得注意的是：2.3位于这里的还有AudioHardwareGeneric、AudioHardwareInterface、A2dpAudioInterface等一系列接口代码，现在都移除了。实际上，这些接口变更为legacy（有另外更好的实现方式，但也兼容之前的方法），取而代之的是要实现hardware/libhardware/include/hardware/audio.h提供的接口，这是一个较大的变化。

两种Audio Hardware HAL接口定义：

1/ legacy：hardware/libhardware\_legacy/include/hardware\_legacy/AudioHardwareInterface.h

2/ 非legacy：hardware/libhardware/include/hardware/audio.h

### 2、audio\_hw ###

**\[plain\]** [ view plain][view plain] [copy][view plain]

1.  hardware/libhardware\_legacy/audio/
2.  \+-- A2dpAudioInterface.cpp
3.  \+-- A2dpAudioInterface.h
4.  \+-- Android.mk
5.  \+-- AudioDumpInterface.cpp
6.  \+-- AudioDumpInterface.h
7.  \+-- AudioHardwareGeneric.cpp
8.  \+-- AudioHardwareGeneric.h
9.  \+-- AudioHardwareInterface.cpp
10. \+-- AudioHardwareStub.cpp
11. \+-- AudioHardwareStub.h
12. \+-- audio\_hw\_hal.cpp
13. \+-- AudioPolicyCompatClient.cpp
14. \+-- AudioPolicyCompatClient.h
15. \+-- audio\_policy\_hal.cpp
16. \+-- AudioPolicyManagerBase.cpp
17. \+-- AudioPolicyManagerDefault.cpp
18. \+-- AudioPolicyManagerDefault.h

上面提及的AudioHardwareGeneric、AudioHardwareInterface、A2dpAudioInterface等都放到libhardware\_legacy里。

事实上legacy也要封装成非legacy中的audio.h，确切的说需要一个联系legacy interface和not legacy interface的中间层，这里的audio\_hw\_hal.cpp就充当这样的一个角色了。因此，我们其实也可以把2.3之前的alsa\_sound这一套东西也搬过来。

**\[plain\]** [ view plain][view plain] [copy][view plain]

1.  hardware/libhardware/modules/audio/
2.  \+-- Android.mk
3.  \+-- audio\_hw.c
4.  \+-- audio\_policy.c

这是一个stub（类似于2.3中的AudioHardwareStub），大多数函数只是简单的返回一个值，并没有实际操作，只是保证Android能得到一个audio hardware hal实例，从而启动运行，当然声音没有输出到外设的。在底层音频驱动或audio hardware hal还没有实现好的情况下，可以使用这个stub device，先让Android跑起来。

**\[plain\]** [ view plain][view plain] [copy][view plain]

1.  device/samsung/tuna/audio/
2.  \+-- Android.mk
3.  \+-- audio\_hw.c
4.  \+-- ril\_interface.c
5.  \+-- ril\_interface.h

这是Samsung Tuna的音频设备抽象层，很有参考价值，计划以后就在它的基础上进行移植。它调用tinyalsa的接口，可见这个方案的底层音频驱动是alsa。

### 3、tinyalsa ###

**\[plain\]** [ view plain][view plain] [copy][view plain]

1.  external/tinyalsa/
2.  \+-- Android.mk
3.  \+-- include
4.  | +-- tinyalsa
5.  | +-- asoundlib.h
6.  \+-- mixer.c \#\#类alsa-lib的control，作用音频部件开关、音量调节等
7.  \+-- pcm.c \#\#类alsa-lib的pcm，作用音频pcm数据回放录制
8.  \+-- README
9.  \+-- tinycap.c \#\#类alsa\_arecord
10. \+-- tinymix.c \#\#类alsa\_amixer
11. \+-- tinyplay.c \#\#类alsa\_aplay

在2.3时代，Android还隐晦把它放在android2.3.1-gingerbread/device/samsung/crespo/libaudio，现在终于把alsa-lib一脚踢开，小三变正室了，正名tinyalsa。

这其实是历史的必然了，alsa-lib太过复杂繁琐了，我看得也很不爽；更重要的商业上面的考虑，必须移除被GNU GPL授权证所约束的部份，alsa-lib并不是个例。  
  
注意：上面的hardware/libhardware\_legacy/audio/、hardware/libhardware/modules/audio/、device/samsung/tuna/audio/是同层的。之一是legacy audio，用于兼容2.2时代的alsa\_sound；之二是stub audio接口；之三是Samsung Tuna的音频抽象层实现。调用层次：AudioFlinger -> audio\_hw -> tinyalsa。

## 二、Audio Hardware HAL加载 ##

### 1、AudioFlinger ###

**\[cpp\]** [ view plain][view plain] [copy][view plain]

1.  //加载audio hardware hal
2.  **static** int load\_audio\_interface(**const** char \*if\_name, **const** hw\_module\_t \*\*mod,
3.  audio\_hw\_device\_t \*\*dev)
4.  \{
5.  int rc;
6.  
7.  //根据classid和if\_name找到指定的动态库并加载，这里加载的是音频动态库，如libaudio.primary.tuna.so
8.  rc = hw\_get\_module\_by\_class(AUDIO\_HARDWARE\_MODULE\_ID, if\_name, mod);
9.  **if** (rc)
10. **goto** out;
11. 
12. //加载好的动态库模块必有个open方法，调用open方法打开音频设备模块
13. rc = audio\_hw\_device\_open(\*mod, dev);
14. LOGE\_IF(rc, "couldn't open audio hw device in %s.%s (%s)",
15. AUDIO\_HARDWARE\_MODULE\_ID, if\_name, strerror(-rc));
16. **if** (rc)
17. **goto** out;
18. 
19. **return** 0;
20. 
21. out:
22. \*mod = NULL;
23. \*dev = NULL;
24. **return** rc;
25. \}
26. 
27. //音频设备接口，hw\_get\_module\_by\_class需要根据这些字符串找到相关的音频模块库
28. **static** **const** char \*audio\_interfaces\[\] = \{
29. "primary", //主音频设备，一般为本机codec
30. "a2dp", //a2dp设备，蓝牙高保真音频
31. "usb", //usb-audio设备，这个东东我2.3就考虑要实现了，现在终于支持了
32. \};
33. \#define ARRAY\_SIZE(x) (sizeof((x))/sizeof(((x)\[0\])))
34. 
35. // ----------------------------------------------------------------------------
36. 
37. AudioFlinger::AudioFlinger()
38. : BnAudioFlinger(),
39. mPrimaryHardwareDev(0), mMasterVolume(1.0f), mMasterMute(**false**), mNextUniqueId(1),
40. mBtNrecIsOff(**false**)
41. \{
42. \}
43. 
44. **void** AudioFlinger::onFirstRef()
45. \{
46. int rc = 0;
47. 
48. Mutex::Autolock \_l(mLock);
49. 
50. /\* TODO: move all this work into an Init() function \*/
51. mHardwareStatus = AUDIO\_HW\_IDLE;
52. 
53. //打开audio\_interfaces数组定义的所有音频设备
54. **for** (size\_t i = 0; i < ARRAY\_SIZE(audio\_interfaces); i++) \{
55. **const** hw\_module\_t \*mod;
56. audio\_hw\_device\_t \*dev;
57. 
58. rc = load\_audio\_interface(audio\_interfaces\[i\], &mod, &dev);
59. **if** (rc)
60. **continue**;
61. 
62. LOGI("Loaded %s audio interface from %s (%s)", audio\_interfaces\[i\],
63. mod->name, mod->id);
64. mAudioHwDevs.push(dev); //mAudioHwDevs是一个Vector，存储已打开的audio hw devices
65. 
66. **if** (!mPrimaryHardwareDev) \{
67. mPrimaryHardwareDev = dev;
68. LOGI("Using '%s' (%s.%s) as the primary audio interface",
69. mod->name, mod->id, audio\_interfaces\[i\]);
70. \}
71. \}
72. 
73. mHardwareStatus = AUDIO\_HW\_INIT;
74. 
75. **if** (!mPrimaryHardwareDev || mAudioHwDevs.size() == 0) \{
76. LOGE("Primary audio interface not found");
77. **return**;
78. \}
79. 
80. //对audio hw devices进行一些初始化，如mode、master volume的设置
81. **for** (size\_t i = 0; i < mAudioHwDevs.size(); i++) \{
82. audio\_hw\_device\_t \*dev = mAudioHwDevs\[i\];
83. 
84. mHardwareStatus = AUDIO\_HW\_INIT;
85. rc = dev->init\_check(dev);
86. **if** (rc == 0) \{
87. AutoMutex lock(mHardwareLock);
88. 
89. mMode = AUDIO\_MODE\_NORMAL;
90. mHardwareStatus = AUDIO\_HW\_SET\_MODE;
91. dev->set\_mode(dev, mMode);
92. mHardwareStatus = AUDIO\_HW\_SET\_MASTER\_VOLUME;
93. dev->set\_master\_volume(dev, 1.0f);
94. mHardwareStatus = AUDIO\_HW\_IDLE;
95. \}
96. \}
97. \}

以上对AudioFlinger进行的分析，主要是通过hw\_get\_module\_by\_class()找到模块接口名字if\_name相匹配的模块库，加载，然后audio\_hw\_device\_open()调用模块的open方法，完成音频设备模块的初始化。

留意AudioFlinger的构造函数只有简单的私有变量的初始化操作了，把音频设备初始化放到onFirstRef()，Android终于改进了这一点，好的设计根本不应该把可能会失败的操作放到构造函数中。onFirstRef是RefBase类的一个虚函数，在构造sp的时候就会被调用。因此，在构造sp<AudioFlinger>的时候就会触发onFirstRef方法，从而完成音频设备模块初始化。

### 2、hw\_get\_module\_by\_class ###

我们接下来看看hw\_get\_module\_by\_class，实现在hardware/libhardware/ hardware.c中，它作用加载指定名字的模块库（.so文件），这个应该是用于加载所有硬件设备相关的库文件，并不只是音频设备。

**\[cpp\]** [ view plain][view plain] [copy][view plain]

1.  int hw\_get\_module\_by\_class(**const** char \*class\_id, **const** char \*inst,
2.  **const** **struct** hw\_module\_t \*\*module)
3.  \{
4.  int status;
5.  int i;
6.  **const** **struct** hw\_module\_t \*hmi = NULL;
7.  char prop\[PATH\_MAX\];
8.  char path\[PATH\_MAX\];
9.  char name\[PATH\_MAX\];
10. 
11. **if** (inst)
12. snprintf(name, PATH\_MAX, "%s.%s", class\_id, inst);
13. **else**
14. strlcpy(name, class\_id, PATH\_MAX);
15. 
16. //这里我们以音频库为例，AudioFlinger调用到这个函数时，
17. //class\_id=AUDIO\_HARDWARE\_MODULE\_ID="audio"，inst="primary"(或"a2dp"或"usb")
18. //那么此时name="audio.primary"
19. 
20. /\*
21. \* Here we rely on the fact that calling dlopen multiple times on
22. \* the same .so will simply increment a refcount (and not load
23. \* a new copy of the library).
24. \* We also assume that dlopen() is thread-safe.
25. \*/
26. 
27. /\* Loop through the configuration variants looking for a module \*/
28. **for** (i=0 ; i<HAL\_VARIANT\_KEYS\_COUNT+1 ; i++) \{
29. **if** (i < HAL\_VARIANT\_KEYS\_COUNT) \{
30. //通过property\_get找到厂家标记如"ro.product.board=tuna"，这时prop="tuna"
31. **if** (property\_get(variant\_keys\[i\], prop, NULL) == 0) \{
32. **continue**;
33. \}
34. snprintf(path, **sizeof**(path), "%s/%s.%s.so",
35. HAL\_LIBRARY\_PATH2, name, prop); //\#define HAL\_LIBRARY\_PATH2 "/vendor/lib/hw"
36. **if** (access(path, R\_OK) == 0) **break**;
37. 
38. snprintf(path, **sizeof**(path), "%s/%s.%s.so",
39. HAL\_LIBRARY\_PATH1, name, prop); //\#define HAL\_LIBRARY\_PATH1 "/system/lib/hw"
40. **if** (access(path, R\_OK) == 0) **break**;
41. \} **else** \{
42. snprintf(path, **sizeof**(path), "%s/%s.default.so", //如没有指定的库文件，则加载default.so，即stub-device
43. HAL\_LIBRARY\_PATH1, name);
44. **if** (access(path, R\_OK) == 0) **break**;
45. \}
46. \}
47. //到这里，完成一个模块库的完整路径名称，如path="/system/lib/hw/audio.primary.tuna.so"
48. //如何生成audio.primary.tuna.so？请看相关的Android.mk文件，其中有定义LOCAL\_MODULE := audio.primary.tuna
49. 
50. status = -ENOENT;
51. **if** (i < HAL\_VARIANT\_KEYS\_COUNT+1) \{
52. /\* load the module, if this fails, we're doomed, and we should not try
53. \* to load a different variant. \*/
54. status = load(class\_id, path, module); //加载模块库
55. \}
56. 
57. **return** status;
58. \}

load()函数不详细分析了，它通过dlopen加载库文件，然后dlsym找到hal\_module\_info的首地址。我们先看看hal\_module\_info的定义：

**\[cpp\]** [ view plain][view plain] [copy][view plain]

1.  /\*\*
2.  \* Every hardware module must have a data structure named HAL\_MODULE\_INFO\_SYM
3.  \* and the fields of this data structure must begin with hw\_module\_t
4.  \* followed by module specific information.
5.  \*/
6.  **typedef** **struct** hw\_module\_t \{
7.  /\*\* tag must be initialized to HARDWARE\_MODULE\_TAG \*/
8.  uint32\_t tag;
9.  
10. /\*\* major version number for the module \*/
11. uint16\_t version\_major;
12. 
13. /\*\* minor version number of the module \*/
14. uint16\_t version\_minor;
15. 
16. /\*\* Identifier of module \*/
17. **const** char \*id;
18. 
19. /\*\* Name of this module \*/
20. **const** char \*name;
21. 
22. /\*\* Author/owner/implementor of the module \*/
23. **const** char \*author;
24. 
25. /\*\* Modules methods \*/
26. **struct** hw\_module\_methods\_t\* methods;
27. 
28. /\*\* module's dso \*/
29. **void**\* dso;
30. 
31. /\*\* padding to 128 bytes, reserved for future use \*/
32. uint32\_t reserved\[32-7\];
33. 
34. \} hw\_module\_t;
35. 
36. **typedef** **struct** hw\_module\_methods\_t \{
37. /\*\* Open a specific device \*/
38. int (\*open)(**const** **struct** hw\_module\_t\* module, **const** char\* id,
39. **struct** hw\_device\_t\*\* device);
40. 
41. \} hw\_module\_methods\_t;

这个结构体很重要，注释很详细。dlsym拿到这个结构体的首地址后，就可以调用Modules methods进行设备模块的初始化了。设备模块中，都应该按照这个格式初始化好这个结构体，否则dlsym找不到它，也就无法调用Modules methods进行初始化了。

例如，在audio\_hw.c中，它是这样定义的：

**\[cpp\]** [ view plain][view plain] [copy][view plain]

1.  **static** **struct** hw\_module\_methods\_t hal\_module\_methods = \{
2.  .open = adev\_open,
3.  \};
4.  
5.  **struct** audio\_module HAL\_MODULE\_INFO\_SYM = \{
6.  .common = \{
7.  .tag = HARDWARE\_MODULE\_TAG,
8.  .version\_major = 1,
9.  .version\_minor = 0,
10. .id = AUDIO\_HARDWARE\_MODULE\_ID,
11. .name = "Tuna audio HW HAL",
12. .author = "The Android Open Source Project",
13. .methods = &hal\_module\_methods,
14. \},
15. \};

### 3、audio\_hw ###

好了，经过一番周折，又dlopen又dlsym的，终于进入我们的audio\_hw。这部分没什么好说的，按照hardware/libhardware/include/hardware/audio.h定义的接口实现就行了。这些接口全扔到一个结构体里面的，这样做的好处是：不必用大量的dlsym来获取各个接口函数的地址，只需找到这个结构体即可，从易用性和可扩充性来说，都是首选方式。

接口定义如下：

**\[cpp\]** [ view plain][view plain] [copy][view plain]

1.  **struct** audio\_hw\_device \{
2.  **struct** hw\_device\_t common;
3.  
4.  /\*\*
5.  \* used by audio flinger to enumerate what devices are supported by
6.  \* each audio\_hw\_device implementation.
7.  \*
8.  \* Return value is a bitmask of 1 or more values of audio\_devices\_t
9.  \*/
10. uint32\_t (\*get\_supported\_devices)(**const** **struct** audio\_hw\_device \*dev);
11. 
12. /\*\*
13. \* check to see if the audio hardware interface has been initialized.
14. \* returns 0 on success, -ENODEV on failure.
15. \*/
16. int (\*init\_check)(**const** **struct** audio\_hw\_device \*dev);
17. 
18. /\*\* set the audio volume of a voice call. Range is between 0.0 and 1.0 \*/
19. int (\*set\_voice\_volume)(**struct** audio\_hw\_device \*dev, float volume);
20. 
21. /\*\*
22. \* set the audio volume for all audio activities other than voice call.
23. \* Range between 0.0 and 1.0. If any value other than 0 is returned,
24. \* the software mixer will emulate this capability.
25. \*/
26. int (\*set\_master\_volume)(**struct** audio\_hw\_device \*dev, float volume);
27. 
28. /\*\*
29. \* setMode is called when the audio mode changes. AUDIO\_MODE\_NORMAL mode
30. \* is for standard audio playback, AUDIO\_MODE\_RINGTONE when a ringtone is
31. \* playing, and AUDIO\_MODE\_IN\_CALL when a call is in progress.
32. \*/
33. int (\*set\_mode)(**struct** audio\_hw\_device \*dev, int mode);
34. 
35. /\* mic mute \*/
36. int (\*set\_mic\_mute)(**struct** audio\_hw\_device \*dev, bool state);
37. int (\*get\_mic\_mute)(**const** **struct** audio\_hw\_device \*dev, bool \*state);
38. 
39. /\* set/get global audio parameters \*/
40. int (\*set\_parameters)(**struct** audio\_hw\_device \*dev, **const** char \*kv\_pairs);
41. 
42. /\*
43. \* Returns a pointer to a heap allocated string. The caller is responsible
44. \* for freeing the memory for it.
45. \*/
46. char \* (\*get\_parameters)(**const** **struct** audio\_hw\_device \*dev,
47. **const** char \*keys);
48. 
49. /\* Returns audio input buffer size according to parameters passed or
50. \* 0 if one of the parameters is not supported
51. \*/
52. size\_t (\*get\_input\_buffer\_size)(**const** **struct** audio\_hw\_device \*dev,
53. uint32\_t sample\_rate, int format,
54. int channel\_count);
55. 
56. /\*\* This method creates and opens the audio hardware output stream \*/
57. int (\*open\_output\_stream)(**struct** audio\_hw\_device \*dev, uint32\_t devices,
58. int \*format, uint32\_t \*channels,
59. uint32\_t \*sample\_rate,
60. **struct** audio\_stream\_out \*\*out);
61. 
62. **void** (\*close\_output\_stream)(**struct** audio\_hw\_device \*dev,
63. **struct** audio\_stream\_out\* out);
64. 
65. /\*\* This method creates and opens the audio hardware input stream \*/
66. int (\*open\_input\_stream)(**struct** audio\_hw\_device \*dev, uint32\_t devices,
67. int \*format, uint32\_t \*channels,
68. uint32\_t \*sample\_rate,
69. audio\_in\_acoustics\_t acoustics,
70. **struct** audio\_stream\_in \*\*stream\_in);
71. 
72. **void** (\*close\_input\_stream)(**struct** audio\_hw\_device \*dev,
73. **struct** audio\_stream\_in \*in);
74. 
75. /\*\* This method dumps the state of the audio hardware \*/
76. int (\*dump)(**const** **struct** audio\_hw\_device \*dev, int fd);
77. \};
78. **typedef** **struct** audio\_hw\_device audio\_hw\_device\_t;

注：这是比较标准的C接口设计方法了，但是个人感觉还是用C++比较好，直观易读。2.3之前都是用C++实现这些接口设计的，到了4.0，不知道为何采纳用C？不会理由是做底层的不懂C++吧？！

## 三、Audio Hardware HAL的legacy实现 ##

之前提到两种Audio Hardware HAL接口定义：  
1/ legacy：hardware/libhardware\_legacy/include/hardware\_legacy/AudioHardwareInterface.h  
2/ 非legacy：hardware/libhardware/include/hardware/audio.h  
前者是2.3及之前的音频设备接口定义，后者是4.0的接口定义。

为了兼容以前的设计，4.0实现一个中间层：hardware/libhardware\_legacy/audio/audio\_hw\_hal.cpp，结构与其他的audio\_hw.c大同小异，差别在于open方法：

**\[cpp\]** [ view plain][view plain] [copy][view plain]

1.  **static** int legacy\_adev\_open(**const** hw\_module\_t\* module, **const** char\* name,
2.  hw\_device\_t\*\* device)
3.  \{
4.  ......
5.  
6.  ladev->hwif = createAudioHardware();
7.  **if** (!ladev->hwif) \{
8.  ret = -EIO;
9.  **goto** err\_create\_audio\_hw;
10. \}
11. 
12. ......
13. \}

看到那个熟悉的createAudioHardware()没有？这是以前我提到的Vendor Specific Audio接口，然后新的接口再调用ladev->hwif的函数就是了。

因此老一套的alsa-lib、alsa-utils和alsa\_sound也可以照搬过来，这里的文件被编译成静态库的，因此你需要修改alsa\_sound里面的Android.mk文件，链接这个静态库。还有alsa\_sound的命名空间原来是“android”，现在需要改成“android\_audio\_legacy”。

## 四、a2dp Audio HAL的实现 ##

4.0的a2dp audio hal放到bluez里实现了，我找了好一会才找到：  
external/Bluetooth/bluez/audio/android\_audio\_hw.c  
大致与上面提到的audio\_hw.c类似，因为都是基于audio.h定义的接口来实现的。  
如果需要编译这个库，须在BoardConfig.mk里定义：  
BOARD\_HAVE\_BLUETOOTH := true  
  
开始还提到现在支持3种audio设备了，分别是primary、a2dp和usb。目前剩下usb audio hal我没有找到，不知是否需要自己去实现？其实alsa-driver都支持大部分的usb-audio设备了，因此上层也可调用tinyalsa的接口，就像samsung tuna的audio\_hw.c那样。

## 五、音质改进？？？ ##

可使用audio echo cancel和更好的resampler（SRC）？？？

[view plain]: http://blog.csdn.net/sepnic/article/details/6978484