Android(安卓)5.0 Camera系统源码分析(1)：CameraService启动流程
1. 前言
本文将分析android系统源码，从frameworks层到hal层，暂不涉及app层和kernel层。由于某些函数比较复杂，在贴出代码时会适当对其进行简化。本文属于自己对源码的总结，仅仅是贯穿代码流程，不会深入分析各个细节。
分析android系统源码，需要对android系统的某些知识点有所了解
涉及的知识点有：
（1）Android系统的智能指针 - 参考老罗的Android系统的智能指针（轻量级指针、强指针和弱指针）的实现原理分析
（2）Android进程间通信Binder - 参考老罗的Android进程间通信（IPC）机制Binder简要介绍和学习计划
（3）Android硬件抽象层(HAL) - 参考老罗的Android硬件抽象层（HAL）概要介绍和学习计划
2. frameworks层
Android的各个子模块的启动都是从它们的Service的启动开始的，所以我们将从CameraService的启动开始分析。CameraService的启动就在MediaServer的main函数中，代码路径在：frameworks/av/media/mediaserver/main_mediaserver.cpp
int main(int argc __unused, char** argv){ ...... CameraService::instantiate(); ......}CameraService类定义如下：
class CameraService : public BinderService, public BnCameraService, public IBinder::DeathRecipient, public camera_module_callbacks_t{ static char const* getServiceName() { return "media.camera"; } ......} mediaserver的main函数中调用了CameraService的instantiate函数来创建实例，该函数的实现在其父类BinderService中实现
templateclass BinderService{ static status_t publish(bool allowIsolated = false) { sp sm(defaultServiceManager()); return sm->addService( String16(SERVICE::getServiceName()), new SERVICE(), allowIsolated); } static void instantiate() { publish(); }} 1. instantiate函数只是简单的调用了publish函数
2. publish函数先构造CameraService，再通过addService函数将它注册到ServiceManager当中，而getServiceName函数获取到的值为“media camera”。这一切都是为了binder通信做准备
3. 这里使用了c++模版，从上面的CameraService类定义中可以看出，这里的SERVICE等于CameraService，也就是说publish函数中的new SERVICE等于new CameraService
4. 同时还使用了智能指针，也就是说除了调用CameraService的构造函数外，还会调用onFirstRef函数
CameraService::CameraService() :mSoundRef(0), mModule(0){ ALOGI("CameraService started (pid=%d)", getpid()); gCameraService = this; for (size_t i = 0; i < MAX_CAMERAS; ++i) { mStatusList[i] = ICameraServiceListener::STATUS_PRESENT; } this->camera_device_status_change = android::camera_device_status_change;}void CameraService::onFirstRef(){ LOG1("CameraService::onFirstRef"); BnCameraService::onFirstRef(); if (hw_get_module(CAMERA_HARDWARE_MODULE_ID, (const hw_module_t **)&mModule) < 0) { ALOGE("Could not load camera HAL module"); mNumberOfCameras = 0; } else { ALOGI("Loaded \"%s\" camera module", mModule->common.name); mNumberOfCameras = mModule->get_number_of_cameras(); if (mNumberOfCameras > MAX_CAMERAS) { ALOGE("Number of cameras(%d) > MAX_CAMERAS(%d).", mNumberOfCameras, MAX_CAMERAS); mNumberOfCameras = MAX_CAMERAS; } for (int i = 0; i < mNumberOfCameras; i++) { LOG1("setCameraFree(%d)", i); setCameraFree(i); } if (mModule->common.module_api_version >= CAMERA_MODULE_API_VERSION_2_1) { mModule->set_callbacks(this); } VendorTagDescriptor::clearGlobalVendorTagDescriptor(); if (mModule->common.module_api_version >= CAMERA_MODULE_API_VERSION_2_2) { setUpVendorTags(); } CameraDeviceFactory::registerService(this); } }第20行. 通过hw_get_module函数加载了一个hw_module_t模块，这个模块是与hal层对接的接口，ID为CAMERA_HARDWARE_MODULE_ID，并将它保存在mModule成员变量中。
第27行. 通过mModule->get_number_of_cameras函数进入到hal层，获取到了camera的个数。这个函数很重要，对于frameworks层来说只是拿到了camera的个数，但对于hal层和drivers层来说Camera的上电和初始化流程都是从这里开始的
3. hal层-基于MTK平台
先来看看mtk camera module的定义，代码路径在：vendor/mediatek/proprietary/hardware/mtkcam/module_hal/module/module.h
staticcamera_moduleget_camera_module(){ camera_module module = { common:{ tag : HARDWARE_MODULE_TAG, #if (PLATFORM_SDK_VERSION >= 21) module_api_version : CAMERA_MODULE_API_VERSION_2_3, #else module_api_version : CAMERA_DEVICE_API_VERSION_1_0, #endif hal_api_version : HARDWARE_HAL_API_VERSION, id : CAMERA_HARDWARE_MODULE_ID, name : "MediaTek Camera Module", author : "MediaTek", methods : get_module_methods(), dso : NULL, reserved : {0}, }, get_number_of_cameras : get_number_of_cameras, get_camera_info : get_camera_info, set_callbacks : set_callbacks, get_vendor_tag_ops : get_vendor_tag_ops, #if (PLATFORM_SDK_VERSION >= 21) open_legacy : open_legacy, #endif reserved : {0}, }; return module;};2. 当frameworks层调用mModule->get_number_of_cameras函数时，实际就是调用上面结构体的get_number_of_cameras函数
CamDeviceManagerImp gCamDeviceManager;ICamDeviceManager*getCamDeviceManager(){ return &gCamDeviceManager;}staticintget_number_of_cameras(void){ return NSCam::getCamDeviceManager()->getNumberOfDevices();}2. CamDeviceManagerImp继承了CamDeviceManagerBase，这里的getNumberOfDevices方法将由父类CamDeviceManagerBase实现
int32_tCamDeviceManagerBase::getNumberOfDevices(){ mi4DeviceNum = enumDeviceLocked(); return mi4DeviceNum;}这里只是调用了enumDeviceLocked函数，并将它的返回值（代表了camera的个数）返回到frameworks层。接着看enumDeviceLocked的实现
int32_tCamDeviceManagerImp::enumDeviceLocked(){ IHalSensorList*const pHalSensorList = IHalSensorList::get(); size_t const sensorNum = pHalSensorList->searchSensors(); for (size_t i = 0; i < sensorNum; i++) { int32_t const deviceId = i; sp pInfo = new EnumInfo; mEnumMap.add(deviceId, pInfo); IMetadataProvider> pMetadataProvider = IMetadataProvider::create(deviceId); pInfo->pMetadata = pMetadataProvider->getStaticCharacteristics(); pInfo->iFacing = (pMetadataProvider->getDeviceFacing() == MTK_LENS_FACING_FRONT) ? CAMERA_FACING_FRONT : CAMERA_FACING_BACK ; pInfo->iWantedOrientation = pMetadataProvider->getDeviceWantedOrientation(); pInfo->iSetupOrientation = pMetadataProvider->getDeviceSetupOrientation(); i4DeviceNum++; } return i4DeviceNum;} 第5-6行. 这里需要重点关注pHalSensorList->searchSensors函数，它的返回值就是camera的个数
第8-24行. 循环构造并初始化一个EnumInfo对象，并把它保存在mEnumMap中MUINTHalSensorList::enumerateSensor_Locked(){ int ret_count = 0; SensorDrv *const pSensorDrv = SensorDrv::get(); int const iSensorsList = pSensorDrv->impSearchSensor(NULL); if((iSensorsList & SENSOR_DEV_MAIN) == SENSOR_DEV_MAIN) { halSensorDev = SENSOR_DEV_MAIN; pSensorInfo = pSensorDrv->getMainSensorInfo(); addAndInitSensorEnumInfo_Locked(halSensorDev, ret_count, mapToSensorType(pSensorInfo->GetType()), pSensorInfo->getDrvMacroName()); ret_count++; } if((iSensorsList & SENSOR_DEV_SUB) == SENSOR_DEV_SUB) { halSensorDev = SENSOR_DEV_SUB; pSensorInfo = pSensorDrv->getSubSensorInfo(); addAndInitSensorEnumInfo_Locked(halSensorDev, ret_count, mapToSensorType(pSensorInfo->GetType()), pSensorInfo->getDrvMacroName()); ret_count++; } mEnumSensorCount = ret_count; return ret_count;}MUINTHalSensorList::searchSensors(){ return enumerateSensor_Locked();} 第33行. searchSensors函数只是调用了enumerateSensor_Locked函数，这里并没有贴出enumerateSensor_Locked函数的所有代码，删减了一些我们暂时不关注的东西
第7行. 重点函数pSensorDrv->impSearchSensor，它的返回值决定了enumerateSensor_Locked的返回值，也就是camera的个数
MINT32ImgSensorDrv::impSearchSensor(pfExIdChk pExIdChkCbf){ MUINT32 SensorEnum = (MUINT32) DUAL_CAMERA_MAIN_SENSOR; MUINT32 i,id[KDIMGSENSOR_MAX_INVOKE_DRIVERS] = {0,0}; MINT32 sensorDevs = SENSOR_NONE; GetSensorInitFuncList(&m_pstSensorInitFunc); m_fdSensor = ::open("/dev/kd_camera_hw", O_RDWR); for (SensorEnum = DUAL_CAMERA_MAIN_SENSOR; SensorEnum <= DUAL_CAMERA_SUB_SENSOR; SensorEnum <<= 1) { for (i = 0; i < MAX_NUM_OF_SUPPORT_SENSOR; i++) { //end of driver list if (m_pstSensorInitFunc[i].getCameraDefault == NULL) { LOG_MSG("m_pstSensorInitFunc[i].getCameraDefault is NULL: %d \n", i); break; } id[KDIMGSENSOR_INVOKE_DRIVER_0] = (SensorEnum << KDIMGSENSOR_DUAL_SHIFT) | i; err = ioctl(m_fdSensor, KDIMGSENSORIOC_X_SET_DRIVER,&id[KDIMGSENSOR_INVOKE_DRIVER_0] ); err = ioctl(m_fdSensor, KDIMGSENSORIOC_T_CHECK_IS_ALIVE); if (err < 0 || err2 < 0) { LOG_MSG("sensor ID mismatch\n"); continue; } if (SensorEnum == DUAL_CAMERA_MAIN_SENSOR) { m_mainSensorDrv.index[m_mainSensorDrv.number] = i; m_mainSensorDrv.type[m_mainSensorDrv.number] = sensorType; m_mainSensorDrv.position = socketPos; m_mainSensorDrv.sensorID = m_pstSensorInitFunc[m_mainSensorDrv.index[m_mainSensorDrv.number]].SensorId; m_mainSensorDrv.number++; } else if (SensorEnum == DUAL_CAMERA_SUB_SENSOR) { m_subSensorDrv.index[m_subSensorDrv.number] = i; m_subSensorDrv.type[m_subSensorDrv.number] = sensorType; m_subSensorDrv.position = socketPos; m_subSensorDrv.sensorID = m_pstSensorInitFunc[m_subSensorDrv.index[m_subSensorDrv.number]].SensorId; m_subSensorDrv.number++; } } } if (BAD_SENSOR_INDEX != m_mainSensorDrv.index[0]) { m_mainSensorId = m_mainSensorDrv.sensorID; m_mainSensorIdx = m_mainSensorDrv.index[0]; sensorDevs |= SENSOR_MAIN; } if (BAD_SENSOR_INDEX != m_subSensorDrv.index[0]) { m_subSensorId = m_subSensorDrv.sensorID; m_subSensorIdx = m_subSensorDrv.index[0]; sensorDevs |= SENSOR_SUB; } return sensorDevs;}这个函数比较长，所以只贴出关键代码
第8行, 调用GetSensorInitFuncList函数来获取hal层的sersors列表，并把它保存在m_pstSensorInitFunc变量中
第9行, 通过系统调用open函数打开camera的设备节点，后面会通过这个节点来进入到kernel层
第11-12行, 通过两个for循环来遍历sensorlist中所有可能存在的camera
第20行, 通过ioctl下达setDriver指令，并下传正在遍历的sensorlist中的ID。Driver层根据这个ID，挂载Driver层sensorlist中对应的操作接口
第21行, 通过ioctl下达check ID指令，Driver层为对应sensor上电，通过I2C读取预存在寄存器中的sensor id。然后比较读取结果，如果不匹配return error后继续遍历
第29-41行, 将sensor相关的信息保存在m_mainSensorDrv和m_subSensorDrv中
第45-56行, 给sensroDevs变量赋值，并将它返回给上一级
这里暂不分析kernel层的代码，先来看看GetSensorInitFuncList函数，代码在sensorlist.cpp中
MSDK_SENSOR_INIT_FUNCTION_STRUCT SensorList[] ={#if defined(IMX175_MIPI_RAW) RAW_INFO(IMX175_SENSOR_ID, SENSOR_DRVNAME_IMX175_MIPI_RAW,NULL),#endif#if defined(IMX179_MIPI_RAW) RAW_INFO(IMX179_SENSOR_ID, SENSOR_DRVNAME_IMX179_MIPI_RAW,NULL),#endif#if defined(IMX219_MIPI_RAW) RAW_INFO(IMX219_SENSOR_ID, SENSOR_DRVNAME_IMX219_MIPI_RAW, NULL),#endif#if defined(IMX214_MIPI_RAW) RAW_INFO(IMX214_SENSOR_ID, SENSOR_DRVNAME_IMX214_MIPI_RAW,NULL),#endif#if defined(GC2235_RAW) RAW_INFO(GC2235_SENSOR_ID, SENSOR_DRVNAME_GC2235_RAW, NULL),#endif#if defined(GC2035_YUV) YUV_INFO(GC2035_SENSOR_ID, SENSOR_DRVNAME_GC2035_YUV, NULL),#endif ......}UINT32 GetSensorInitFuncList(MSDK_SENSOR_INIT_FUNCTION_STRUCT **ppSensorList){ *ppSensorList = &SensorList[0]; return MHAL_NO_ERROR;}4. 总结
至此，除kernel层外，简述了CameraService的启动流程，大概过程如下图所示更多相关文章
- 没有一行代码，「2020 新冠肺炎记忆」这个项目却登上了 GitHub 中
- Android(安卓)Phone进程启动过程详解
- Android(安卓)限制textview Edittext 只能输入电话号码和特定字
- Android(安卓)SurfaceView入门学习
- android应用创建子进程的方法探究
- android xml解析生成探讨
- 一个android文本比对APP的实现（三）－设计模式在文件选择模块中的运用
- Android(安卓)运行时权限申请之电话权限（兼容6.0以上）
- Android逆向之旅---动态方式破解apk进阶篇(IDA调试so源码)[转]