Qualcomm Camera HAL 2.0

Qualcomm Camera HAL 2.0

Posted on 2013/08/25 by guohai  705 views

我们知道在HAL的Vendor实现当中会动态去load一个名字为camera.$platform$.so的档案，然后去加载Android HAL当中定义的方法，这里以Camera HAL 2.0并且Qualcomm msm8960为例子看下，结合之前的一篇文章(http://guoh.org/lifelog/2013/07/glance-at-camera-hal-2-0/)。

(注：这篇文章已经草稿比较久了，但是一直没有发出来，因为手里的这版代码没有设备可以跑，另外也无法确定代码是否完全正确，至少发现了一些地方都是stub实现，文中可能存在一些错误，如发现不正确的地方欢迎指出，我也会尽量发现错误并修正！)

我们知道在camera2.h当中定义了很多方法，那么在msm8960 HAL就是在如下地方
/path/to/qcam-hal/QCamera/HAL2
这编译出来就是一个camera.$platform$.so，请看它的实现
首先是HAL2/wrapper/QualcommCamera.h|cpp

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102

/**   * The functions need to be provided by the camera HAL.   *   * If getNumberOfCameras() returns N, the valid cameraId for getCameraInfo()   * and openCameraHardware() is 0 to N-1.   */    static hw_module_methods_t camera_module_methods = {      open: camera_device_open, };    static hw_module_t camera_common  = {      tag: HARDWARE_MODULE_TAG,      module_api_version: CAMERA_MODULE_API_VERSION_2_0, // 这样Camera Service才会去初始化Camera2Client一系列      hal_api_version: HARDWARE_HAL_API_VERSION,      id: CAMERA_HARDWARE_MODULE_ID,      name: "Qcamera" ,      author: "Qcom" ,      methods: &camera_module_methods,      dso: NULL,      reserved:  { 0 }, };    camera_module_t HAL_MODULE_INFO_SYM = { // 这个HMI，每个HAL模块都必须有的      common: camera_common,      get_number_of_cameras: get_number_of_cameras,      get_camera_info: get_camera_info, };    camera2_device_ops_t camera_ops = { // 注意这些绑定的函数      set_request_queue_src_ops:           android::set_request_queue_src_ops,      notify_request_queue_not_empty:      android::notify_request_queue_not_empty,      set_frame_queue_dst_ops:             android::set_frame_queue_dst_ops,      get_in_progress_count:               android::get_in_progress_count,      flush_captures_in_progress:          android::flush_captures_in_progress,      construct_default_request:           android::construct_default_request,         allocate_stream:                     android::allocate_stream,      register_stream_buffers:             android::register_stream_buffers,      release_stream:                      android::release_stream,         allocate_reprocess_stream:           android::allocate_reprocess_stream,      allocate_reprocess_stream_from_stream: android::allocate_reprocess_stream_from_stream,      release_reprocess_stream:            android::release_reprocess_stream,         trigger_action:                      android::trigger_action,      set_notify_callback:                 android::set_notify_callback,      get_metadata_vendor_tag_ops:         android::get_metadata_vendor_tag_ops,      dump:                                android::dump, };    typedef struct { // 注意这个是Qualcomm自己定义的一个wrap结构    camera2_device_t hw_dev; // 这里是标准的    QCameraHardwareInterface *hardware;    int camera_released;    int cameraId; } camera_hardware_t;    /* HAL should return NULL if it fails to open camera hardware. */ extern "C" int  camera_device_open(    const struct hw_module_t* module, const char* id,            struct hw_device_t** hw_device) {      int rc = -1;      int mode = 0;      camera2_device_t *device = NULL;      if (module && id && hw_device) {          int cameraId = atoi(id);             if (!strcmp(module->name, camera_common.name)) {              camera_hardware_t *camHal =                  (camera_hardware_t *) malloc(sizeof (camera_hardware_t));              if (!camHal) {                  *hw_device = NULL;                  ALOGE("%s:  end in no mem", __func__);                  return rc;              }              /* we have the camera_hardware obj malloced */              memset(camHal, 0, sizeof (camera_hardware_t));              camHal->hardware = new QCameraHardwareInterface(cameraId, mode);              if (camHal->hardware && camHal->hardware->isCameraReady()) {                  camHal->cameraId = cameraId;                  device = &camHal->hw_dev; // 这里camera2_device_t                  device->common.close = close_camera_device; // 初始化camera2_device_t                  device->common.version = CAMERA_DEVICE_API_VERSION_2_0;                  device->ops = &camera_ops;                  device->priv = (void *)camHal;                  rc =  0;              } else {                  if (camHal->hardware) {                      delete camHal->hardware;                      camHal->hardware = NULL;                  }                  free(camHal);                  device = NULL;              }          }      }      /* pass actual hw_device ptr to framework. This amkes that we actally be use memberof() macro */      *hw_device = (hw_device_t*)&device->common; // 这就是kernel或者Android native framework常用的一招      return rc; }

看看allocate stream

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

int allocate_stream( const struct camera2_device *device,          uint32_t width,          uint32_t height,          int      format,          const camera2_stream_ops_t *stream_ops,          uint32_t *stream_id,          uint32_t *format_actual,          uint32_t *usage,          uint32_t *max_buffers) {      QCameraHardwareInterface *hardware = util_get_Hal_obj(device);      hardware->allocate_stream(width, height, format, stream_ops,              stream_id, format_actual, usage, max_buffers);      return rc; }

这里注意QCameraHardwareInterface在QCameraHWI.h|cpp当中

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32

int QCameraHardwareInterface::allocate_stream(      uint32_t width,      uint32_t height, int format,      const camera2_stream_ops_t *stream_ops,      uint32_t *stream_id,      uint32_t *format_actual,      uint32_t *usage,      uint32_t *max_buffers) {      int ret = OK;      QCameraStream *stream = NULL;      camera_mode_t myMode = (camera_mode_t)(CAMERA_MODE_2D|CAMERA_NONZSL_MODE);         stream = QCameraStream_preview::createInstance(                          mCameraHandle->camera_handle,                          mChannelId,                          width,                          height,                          format,                          mCameraHandle,                          myMode);         stream->setPreviewWindow(stream_ops); // 这里，也就是只要通过该方法创建的stream，都会有对应的ANativeWindow进来      *stream_id = stream->getStreamId();      *max_buffers= stream->getMaxBuffers(); // 从HAL得到的      *usage = GRALLOC_USAGE_HW_CAMERA_WRITE | CAMERA_GRALLOC_HEAP_ID          | CAMERA_GRALLOC_FALLBACK_HEAP_ID;      /* Set to an arbitrary format SUPPORTED by gralloc */      *format_actual = HAL_PIXEL_FORMAT_YCrCb_420_SP;         return ret; }

QCameraStream_preview::createInstance直接调用自己的构造方法，也就是下面
(相关class在QCameraStream.h|cpp和QCameraStream_Preview.cpp)

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33

QCameraStream_preview::QCameraStream_preview(uint32_t CameraHandle,                          uint32_t ChannelId,                          uint32_t Width,                          uint32_t Height,                          int requestedFormat,                          mm_camera_vtbl_t *mm_ops,                          camera_mode_t mode) :                   QCameraStream(CameraHandle,                          ChannelId,                          Width,                          Height,                          mm_ops,                          mode),                   mLastQueuedFrame(NULL),                   mDisplayBuf(NULL),                   mNumFDRcvd( 0 ) {      mStreamId = allocateStreamId(); // 分配stream id(根据mStreamTable)         switch (requestedFormat) { // max buffer number      case CAMERA2_HAL_PIXEL_FORMAT_OPAQUE:          mMaxBuffers = 5 ;          break ;      case HAL_PIXEL_FORMAT_BLOB:          mMaxBuffers = 1 ;          break ;      default :          ALOGE( "Unsupported requested format %d" , requestedFormat);          mMaxBuffers = 1 ;          break ;      }      /*TODO: There has to be a better way to do this*/ }

再看看
/path/to/qcam-hal/QCamera/stack/mm-camera-interface/
mm_camera_interface.h
当中

1 2 3 4 5

typedef struct {      uint32_t camera_handle;        /* camera object handle */      mm_camera_info_t *camera_info; /* reference pointer of camear info */      mm_camera_ops_t *ops;          /* API call table */ } mm_camera_vtbl_t;

mm_camera_interface.c
当中

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29

/* camera ops v-table */ static mm_camera_ops_t mm_camera_ops = {      .sync = mm_camera_intf_sync,      .is_event_supported = mm_camera_intf_is_event_supported,      .register_event_notify = mm_camera_intf_register_event_notify,      .qbuf = mm_camera_intf_qbuf,      .camera_close = mm_camera_intf_close,      .query_2nd_sensor_info = mm_camera_intf_query_2nd_sensor_info,      .is_parm_supported = mm_camera_intf_is_parm_supported,      .set_parm = mm_camera_intf_set_parm,      .get_parm = mm_camera_intf_get_parm,      .ch_acquire = mm_camera_intf_add_channel,      .ch_release = mm_camera_intf_del_channel,      .add_stream = mm_camera_intf_add_stream,      .del_stream = mm_camera_intf_del_stream,      .config_stream = mm_camera_intf_config_stream,      .init_stream_bundle = mm_camera_intf_bundle_streams,      .destroy_stream_bundle = mm_camera_intf_destroy_bundle,      .start_streams = mm_camera_intf_start_streams,      .stop_streams = mm_camera_intf_stop_streams,      .async_teardown_streams = mm_camera_intf_async_teardown_streams,      .request_super_buf = mm_camera_intf_request_super_buf,      .cancel_super_buf_request = mm_camera_intf_cancel_super_buf_request,      .start_focus = mm_camera_intf_start_focus,      .abort_focus = mm_camera_intf_abort_focus,      .prepare_snapshot = mm_camera_intf_prepare_snapshot,      .set_stream_parm = mm_camera_intf_set_stream_parm,      .get_stream_parm = mm_camera_intf_get_stream_parm };

以start stream为例子

1 2 3 4 5 6 7 8 9

mm_camera_intf_start_streams(mm_camera_interface      mm_camera_start_streams(mm_camera          mm_channel_fsm_fn(mm_camera_channel              mm_channel_fsm_fn_active(mm_camera_channel                  mm_channel_start_streams(mm_camera_channel                      mm_stream_fsm_fn(mm_camera_stream                          mm_stream_fsm_reg(mm_camera_stream                              mm_camera_cmd_thread_launch(mm_camera_data                              mm_stream_streamon(mm_camera_stream

注意：本文当中，如上这种梯度摆放，表示是调用关系，如果梯度是一样的，就表示这些方法是在上层同一个方法里面被调用的

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23

int32_t mm_stream_streamon(mm_stream_t *my_obj) {      int32_t rc;      enum v4l2_buf_type buf_type = V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE;         /* Add fd to data poll thread */      rc = mm_camera_poll_thread_add_poll_fd(&my_obj->ch_obj->poll_thread[0],                                             my_obj->my_hdl,                                             my_obj->fd,                                             mm_stream_data_notify,                                             (void*)my_obj);      if (rc < 0) {          return rc;      }      rc = ioctl(my_obj->fd, VIDIOC_STREAMON, &buf_type);      if (rc < 0) {          CDBG_ERROR("%s: ioctl VIDIOC_STREAMON failed: rc=%d\n",                     __func__, rc);          /* remove fd from data poll thread in case of failure */          mm_camera_poll_thread_del_poll_fd(&my_obj->ch_obj->poll_thread[ 0 ], my_obj->my_hdl);      }      return rc; }

看到ioctl，VIDIOC_STREAMON，可以高兴一下了，这就是V4L2规范当中用户空间和内核空间通信的方法，V4L2(Video for Linux Two)是一种经典而且成熟的视频通信协议，之前是V4L，不清楚的可以去下载它的规范，另外The Video4Linux2(http://lwn.net/Articles/203924/)也是很好的资料。
这里简单介绍下：

open(VIDEO_DEVICE_NAME, ...) // 开启视频设备，一般在程序初始化的时候调用

ioctl(...) // 主要是一些需要传输数据量很小的控制操作
这里可以用的参数很多，并且通常来说我们会按照以下方式来使用，比如
VIDIOC_QUERYCAP // 查询设备能干什么
VIDIOC_CROPCAP // 查询设备crop能力
VIDIOC_S_* // set/get方法，设置/获取参数
VIDIOC_G_*
VIDIOC_REQBUFS // 分配buffer，可以有多种方式
VIDIOC_QUERYBUF // 查询分配的buffer的信息
VIDIOC_QBUF // QUEUE BUFFER 把buffer压入DRV缓存队列(这时候buffer是空的)
VIDIOC_STREAMON // 开始视频数据传输
VIDIOC_DQBUF // DEQUEUE BUFFER 把buffer从DRV缓存队列中取出(这时候buffer是有数据的)

[0...n]
QBUF -> DQBUF // 可以一直重复这个动作

VIDIOC_STREAMOFF // 停止视频数据传输

close(VIDEO_DEVICE_FD) // 关闭设备
上面就是主要的函数和简单的调用顺序，另外还有几个函数

select() // 等待事件发生，主要用在我们把存frame的buffer推给DRV以后，等待它的反应
mmap/munmap // 主要处理我们request的buffer的，buffer分配在设备的内存空间的时候需要

并且看看mm_camera_stream这个文件里面也都是这么实现的。

看完这里，我们回过头来继续看QCam HAL，当然它实现的细节也不是我上面start stream所列的那么简单，但是其实也不算复杂，觉得重要的就是状态和用到的结构。

首先是channel状态，目前只支持1个channel，但是可以有多个streams(后面会介绍，而且目前最多支持8个streams)

1 2 3 4 5 6 7 8

/* mm_channel */ typedef enum {      MM_CHANNEL_STATE_NOTUSED = 0,   /* not used */      MM_CHANNEL_STATE_STOPPED,       /* stopped */      MM_CHANNEL_STATE_ACTIVE,        /* active, at least one stream active */      MM_CHANNEL_STATE_PAUSED,        /* paused */      MM_CHANNEL_STATE_MAX } mm_channel_state_type_t;

它可以执行的事件

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21

typedef enum {      MM_CHANNEL_EVT_ADD_STREAM,      MM_CHANNEL_EVT_DEL_STREAM,      MM_CHANNEL_EVT_START_STREAM,      MM_CHANNEL_EVT_STOP_STREAM,      MM_CHANNEL_EVT_TEARDOWN_STREAM,      MM_CHANNEL_EVT_CONFIG_STREAM,      MM_CHANNEL_EVT_PAUSE,      MM_CHANNEL_EVT_RESUME,      MM_CHANNEL_EVT_INIT_BUNDLE,      MM_CHANNEL_EVT_DESTROY_BUNDLE,      MM_CHANNEL_EVT_REQUEST_SUPER_BUF,      MM_CHANNEL_EVT_CANCEL_REQUEST_SUPER_BUF,      MM_CHANNEL_EVT_START_FOCUS,      MM_CHANNEL_EVT_ABORT_FOCUS,      MM_CHANNEL_EVT_PREPARE_SNAPSHOT,      MM_CHANNEL_EVT_SET_STREAM_PARM,      MM_CHANNEL_EVT_GET_STREAM_PARM,      MM_CHANNEL_EVT_DELETE,      MM_CHANNEL_EVT_MAX } mm_channel_evt_type_t;

1 2 3 4 5 6 7 8 9 10 11 12

/* mm_stream */ typedef enum { // 这里的状态要仔细，每执行一次方法，状态就需要变化      MM_STREAM_STATE_NOTUSED = 0,      /* not used */      MM_STREAM_STATE_INITED,           /* inited  */      MM_STREAM_STATE_ACQUIRED,         /* acquired, fd opened  */      MM_STREAM_STATE_CFG,              /* fmt & dim configured */      MM_STREAM_STATE_BUFFED,           /* buf allocated */      MM_STREAM_STATE_REG,              /* buf regged, stream off */      MM_STREAM_STATE_ACTIVE_STREAM_ON, /* active with stream on */      MM_STREAM_STATE_ACTIVE_STREAM_OFF, /* active with stream off */      MM_STREAM_STATE_MAX } mm_stream_state_type_t;

同样，stream可以执行的事件

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

typedef enum {      MM_STREAM_EVT_ACQUIRE,      MM_STREAM_EVT_RELEASE,      MM_STREAM_EVT_SET_FMT,      MM_STREAM_EVT_GET_BUF,      MM_STREAM_EVT_PUT_BUF,      MM_STREAM_EVT_REG_BUF,      MM_STREAM_EVT_UNREG_BUF,      MM_STREAM_EVT_START,      MM_STREAM_EVT_STOP,      MM_STREAM_EVT_QBUF,      MM_STREAM_EVT_SET_PARM,      MM_STREAM_EVT_GET_PARM,      MM_STREAM_EVT_MAX } mm_stream_evt_type_t;

这里每次执行函数的时候都需要检查channel/stream的状态，只有状态正确的时候才会去执行

比如你可以观察到
mm_channel的mm_channel_state_type_t state;
mm_stream的mm_stream_state_type_t state;
均表示这个结构当前的状态

另外
struct mm_camera_obj
struct mm_channel
struct mm_stream
这三个也是自上而下包含的，并且stream和channel还会持有父结构(暂且这么称呼，实际为container关系)的引用。

实际上Vendor的HAL每个都有自己实现的方法，也可能包含很多特有的东西，比如这里它会喂给ioctl一些特有的命令或者数据结构，这些我们就只有在做特定平台的时候去考虑了。这些都可能千变万化，比如OMAP4它同DRV沟通是透过rpmsg，并用OpenMAX的一套规范来实现的。

理论就这么多，接着看一个实例，比如我们在Camera Service要去start preview：

1 2 3 4 5 6 7 8 9 10 11

Camera2Client::startPreviewL      StreamingProcessor->updatePreviewStream          Camera2Device->createStream              StreamAdapter->connectToDevice                  camera2_device_t->ops->allocate_stream // 上面有分析                  native_window_api_*或者native_window_*         StreamingProcessor->startStream          Camera2Device->setStreamingRequest              Camera2Device::RequestQueue->setStreamSlot // 创建一个stream slot                  Camera2Device::RequestQueue->signalConsumerLocked

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

status_t Camera2Device::MetadataQueue::signalConsumerLocked() {      status_t res = OK;      notEmpty.signal();      if (mSignalConsumer && mDevice != NULL) {          mSignalConsumer = false ;          mMutex.unlock();          res = mDevice->ops->notify_request_queue_not_empty(mDevice); // 通知Vendor HAL的run command thread去运行，                                                                       // notify_request_queue_not_empty这个事件不是每次都会触发的，只有初始化时候                                                                       // 或者run command thread在dequeue的时候发现数据为NULL，                                                                       // 而Camera Service之变又有新的request进来的时候才会去触发                                                                       // 可以说是减轻负担吧，不用没有请求的时候，thread也一直在那里                                                                       // 不过通常碰到这样的情况都是利用锁让thread停在那里          mMutex.lock();      }      return res; }

然而在Qualcomm HAL当中

1 2 3

int notify_request_queue_not_empty( const struct camera2_device *device) // 这个方法注册到camera2_device_ops_t当中      QCameraHardwareInterface->notify_request_queue_not_empty()          pthread_create(&mCommandThread, &attr, command_thread, ( void *) this ) != 0 )

1 2 3 4 5

void *command_thread( void *obj) {      ...      pme->runCommandThread(obj); }

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69

void QCameraHardwareInterface::runCommandThread( void *data) {      /**       * This function implements the main service routine for the incoming       * frame requests, this thread routine is started everytime we get a       * notify_request_queue_not_empty trigger, this thread makes the       * assumption that once it receives a NULL on a dequest_request call       * there will be a fresh notify_request_queue_not_empty call that is       * invoked thereby launching a new instance of this thread. Therefore,       * once we get a NULL on a dequeue request we simply let this thread die       */      int res;      camera_metadata_t *request=NULL;      mPendingRequests= 0 ;         while (mRequestQueueSrc) { // mRequestQueueSrc是通过set_request_queue_src_ops设置进来的                                 // 参见Camera2Device::MetadataQueue::setConsumerDevice                                 // 在Camera2Device::initialize当中被调用          ALOGV( "%s:Dequeue request using mRequestQueueSrc:%p" ,__func__,mRequestQueueSrc);          mRequestQueueSrc->dequeue_request(mRequestQueueSrc, &request); // 取framework request          if (request==NULL) {              ALOGE("%s:No more requests available from src command \                      thread dying",__func__);              return ;          }          mPendingRequests++;             /* Set the metadata values */             /* Wait for the SOF for the new metadata values to be applied */             /* Check the streams that need to be active in the stream request */          sort_camera_metadata(request);             camera_metadata_entry_t streams;          res = find_camera_metadata_entry(request,                  ANDROID_REQUEST_OUTPUT_STREAMS,                  &streams);          if (res != NO_ERROR) {              ALOGE("%s: error reading output stream tag", __FUNCTION__);              return;          }             res = tryRestartStreams(streams); // 会去prepareStream和streamOn，后面有详细代码          if (res != NO_ERROR) {              ALOGE("error tryRestartStreams %d", res);              return;          }             /* 3rd pass: Turn on all streams requested */          for (uint32_t i = 0; i < streams.count; i++) {              int streamId = streams.data.u8[i];              QCameraStream *stream = QCameraStream::getStreamAtId(streamId);                 /* Increment the frame pending count in each stream class */                 /* Assuming we will have the stream obj in had at this point may be               * may be multiple objs in which case we loop through array of streams */              stream->onNewRequest();          }          ALOGV("%s:Freeing request using mRequestQueueSrc:%p",__func__,mRequestQueueSrc);          /* Free the request buffer */          mRequestQueueSrc->free_request(mRequestQueueSrc,request);          mPendingRequests--;          ALOGV( "%s:Completed request" ,__func__);      }         QCameraStream::streamOffAll(); }

下面这个方法解释mRequestQueueSrc来自何处

1 2 3 4 5 6 7 8 9 10

// Connect to camera2 HAL as consumer (input requests/reprocessing) status_t Camera2Device::MetadataQueue::setConsumerDevice(camera2_device_t *d) {      ATRACE_CALL();      status_t res;      res = d->ops->set_request_queue_src_ops(d,              this );      if (res != OK) return res;      mDevice = d;      return OK; }

因为

1 2 3 4 5 6 7 8

QCameraStream_preview->prepareStream      QCameraStream->initStream          mm_camera_vtbl_t->ops->add_stream(... stream_cb_routine ...) // 这是用来返回数据的callback，带mm_camera_super_buf_t*和void*两参数              mm_camera_add_stream                  mm_channel_fsm_fn(..., MM_CHANNEL_EVT_ADD_STREAM, ..., mm_evt_paylod_add_stream_t)                      mm_channel_fsm_fn_stopped                          mm_channel_add_stream(..., mm_camera_buf_notify_t, ...)                              mm_stream_fsm_inited

而
在mm_channel_add_stream当中有把mm_camera_buf_notify_t包装到mm_stream_t

1 2 3 4 5 6 7

mm_stream_t *stream_obj = NULL; /* initialize stream object */ memset(stream_obj, 0, sizeof(mm_stream_t)); /* cd through intf always palced at idx 0 of buf_cb */ stream_obj->buf_cb[0].cb = buf_cb; // callback stream_obj->buf_cb[0].user_data = user_data; stream_obj->buf_cb[0].cb_count = -1; /* infinite by default */ // 默认无限次数

并且mm_stream_fsm_inited，传进来的event参数也是MM_STREAM_EVT_ACQUIRE

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53

int32_t mm_stream_fsm_inited(mm_stream_t *my_obj,                               mm_stream_evt_type_t evt,                               void * in_val,                               void * out_val) {      int32_t rc = 0 ;      char dev_name[MM_CAMERA_DEV_NAME_LEN];         switch (evt) {      case MM_STREAM_EVT_ACQUIRE:          if ((NULL == my_obj->ch_obj) || (NULL == my_obj->ch_obj->cam_obj)) {              CDBG_ERROR( "%s: NULL channel or camera obj\n" , __func__);              rc = - 1 ;              break ;          }             snprintf(dev_name, sizeof(dev_name), "/dev/%s" ,                   mm_camera_util_get_dev_name(my_obj->ch_obj->cam_obj->my_hdl));             my_obj->fd = open(dev_name, O_RDWR | O_NONBLOCK); // 打开视频设备          if (my_obj->fd <= 0 ) {              CDBG_ERROR( "%s: open dev returned %d\n" , __func__, my_obj->fd);              rc = - 1 ;              break ;          }          rc = mm_stream_set_ext_mode(my_obj);          if ( 0 == rc) {              my_obj->state = MM_STREAM_STATE_ACQUIRED; // mm_stream_state_type_t          } else {              /* failed setting ext_mode               * close fd */              if (my_obj->fd > 0 ) {                  close(my_obj->fd);                  my_obj->fd = - 1 ;              }              break ;          }          rc = get_stream_inst_handle(my_obj);          if (rc) {              if (my_obj->fd > 0 ) {                  close(my_obj->fd);                  my_obj->fd = - 1 ;              }          }          break ;      default :          CDBG_ERROR( "%s: Invalid evt=%d, stream_state=%d" ,                     __func__,evt,my_obj->state);          rc = - 1 ;          break ;      }      return rc; }

还有

1 2 3 4 5 6 7 8 9

QCameraStream->streamOn      mm_camera_vtbl_t->ops->start_streams          mm_camera_intf_start_streams              mm_camera_start_streams                  mm_channel_fsm_fn(..., MM_CHANNEL_EVT_START_STREAM, ...)                      mm_stream_fsm_fn(..., MM_STREAM_EVT_START, ...)                          mm_camera_cmd_thread_launch // 启动CB线程                          mm_stream_streamon(mm_stream_t)                              mm_camera_poll_thread_add_poll_fd(..., mm_stream_data_notify , ...)

而

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52

static void mm_stream_data_notify( void * user_data) {      mm_stream_t *my_obj = (mm_stream_t*)user_data;      int32_t idx = - 1 , i, rc;      uint8_t has_cb = 0 ;      mm_camera_buf_info_t buf_info;         if (NULL == my_obj) {          return ;      }         if (MM_STREAM_STATE_ACTIVE_STREAM_ON != my_obj->state) {          /* this Cb will only received in active_stream_on state           * if not so, return here */          CDBG_ERROR("%s: ERROR!! Wrong state (%d) to receive data notify!",                     __func__, my_obj->state);          return;      }         memset(&buf_info, 0, sizeof(mm_camera_buf_info_t));         pthread_mutex_lock(&my_obj->buf_lock);      rc = mm_stream_read_msm_frame(my_obj, &buf_info); // 通过ioctl(..., VIDIOC_DQBUF, ...)读取frame数据      if (rc != 0) {          pthread_mutex_unlock(&my_obj->buf_lock);          return;      }      idx = buf_info.buf->buf_idx;         /* update buffer location */      my_obj->buf_status[idx].in_kernel = 0;         /* update buf ref count */      if (my_obj->is_bundled) {          /* need to add into super buf since bundled, add ref count */          my_obj->buf_status[idx].buf_refcnt++;      }         for (i=0; i < MM_CAMERA_STREAM_BUF_CB_MAX; i++) {          if(NULL != my_obj->buf_cb[i].cb) {              /* for every CB, add ref count */              my_obj->buf_status[idx].buf_refcnt++;              has_cb = 1 ;          }      }      pthread_mutex_unlock(&my_obj->buf_lock);         mm_stream_handle_rcvd_buf(my_obj, &buf_info); // mm_camera_queue_enq，往queue里面丢frame数据(                                                    // 前提是有注册callback)，并透过sem_post通知queue                                                    // 然后mm_camera_cmd_thread_launch启动的线程会                                                    // 轮循读取数据，然后执行CB }

这样就会导致在stream on的时候stream_cb_routine(实现在QCameraStream当中)就会一直执行

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

void stream_cb_routine(mm_camera_super_buf_t *bufs,                         void *userdata) {      QCameraStream *p_obj=(QCameraStream*) userdata;      switch (p_obj->mExtImgMode) { // 这个mode在prepareStream的时候就会确定      case MM_CAMERA_PREVIEW:          ALOGE( "%s : callback for MM_CAMERA_PREVIEW" , __func__);          ((QCameraStream_preview *)p_obj)->dataCallback(bufs); // CAMERA_PREVIEW和CAMERA_VIDEO是一样的？          break ;      case MM_CAMERA_VIDEO:          ALOGE( "%s : callback for MM_CAMERA_VIDEO" , __func__);          ((QCameraStream_preview *)p_obj)->dataCallback(bufs);          break ;      case MM_CAMERA_SNAPSHOT_MAIN:          ALOGE( "%s : callback for MM_CAMERA_SNAPSHOT_MAIN" , __func__);          p_obj->p_mm_ops->ops->qbuf(p_obj->mCameraHandle,                                     p_obj->mChannelId,                                     bufs->bufs[ 0 ]);          break ;      case MM_CAMERA_SNAPSHOT_THUMBNAIL:          break ;      default :          break ;      } }

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

void QCameraStream::dataCallback(mm_camera_super_buf_t *bufs) {      if (mPendingCount != 0 ) { // 这个dataCallback是一直在都在回来么？                                 // 而且从代码来看设置下去的callback次数默认是-1，-1就表示infinite。                                 // 似乎只能这样才能解释，否则没人触发的话，即使mPendingCount在onNewRequest当中加1了                                 // 这里也感知不到          ALOGD( "Got frame request" );          pthread_mutex_lock(&mFrameDeliveredMutex);          mPendingCount--;          ALOGD( "Completed frame request" );          pthread_cond_signal(&mFrameDeliveredCond);          pthread_mutex_unlock(&mFrameDeliveredMutex);          processPreviewFrame(bufs);      } else {          p_mm_ops->ops->qbuf(mCameraHandle,                  mChannelId, bufs->bufs[ 0 ]); // 如果没有需要数据的情况，直接把buffer压入DRV的队列当中，会call到V4L2的QBUF      } }

比较好奇的是在手里这版QCam HAL的code当中camera2_frame_queue_dst_ops_t没有被用到

1 2 3 4 5 6

int QCameraHardwareInterface::set_frame_queue_dst_ops(      const camera2_frame_queue_dst_ops_t *frame_dst_ops) {      mFrameQueueDst = frame_dst_ops; // 这个现在似乎没有用到嘛      return OK; }

这样Camera Service的FrameProcessor的Camera2Device->getNextFrame就永远也获取不到数据，不知道是不是我手里的这版代码的问题，而且在最新的Qualcomm Camera HAL代码也不在AOSP树当中了，而是直接以proprietary形式给的so档，这只是题外话。

所以总体来看，这里可能有几个QCameraStream，每个stream负责自己的事情。
他们之间也有相互关系，比如有可能新的stream进来会导致其他已经stream-on的stream重新启动。

在Camera HAL 2.0当中我们还有个重点就是re-process stream
简单的说就是把output stream作为input stream再次添加到BufferQueue中，让其他的consumer来处理，就类似一个chain一样。
目前在ZslProcessor当中有用到。

1 2 3 4 5 6

ZslProcessor->updateStream      Camera2Device->createStream      Camera2Device->createReprocessStreamFromStream // release的时候是先delete re-process          new ReprocessStreamAdapter          ReprocessStreamAdapter->connectToDevice              camera2_device_t->ops->allocate_reprocess_stream_from_stream

这里ReprocessStreamAdapter实际就是camera2_stream_in_ops_t，负责管理re-process的stream。

但是这版的代码Qualcomm也似乎没有去实现，所以暂时到此为止，如果后面找到相应的代码，再来看。

所以看完这么多不必觉得惊讶，站在Camera Service的立场，它持有两个MetadataQueue，mRequestQueue和mFrameQueue。
app请求的动作，比如set parameter/start preview/start recording会直接转化为request，放到mRequestQueue，然后去重启preview/recording stream。
比如capture也会转换为request，放到mRequestQueue。
如果有必要，会通过notify_request_queue_not_empty去通知QCam HAL有请求需要处理，然后QCam HAL会启动一个线程(QCameraHardwareInterface::runCommandThread)去做处理。直到所有request处理完毕退出线程。
在这个处理的过程当中会分别调用到每个stream的processPreviewFrame，有必要的话它每个都会调用自己后续的callback。
还有一个实现的细节就是，stream_cb_routine是从start stream就有开始注册在同一个channel上的，而stream_cb_routine间接调用QCameraStream::dataCallback(当然stream_cb_routine有去指定这个callback回来的原因是什么，就好调用对应的dataCallback)，这个callback是一直都在回来，所以每次new request让mPendingCount加1之后，dataCallback回来才会调用processPreviewFrame，否则就直接把buffer再次压回DRV队列当中。

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

void QCameraStream::dataCallback(mm_camera_super_buf_t *bufs) {      if (mPendingCount != 0 ) { // 这个dataCallback是一直在都在回来么？                                 // 而且从代码来看设置下去的callback次数默认是-1，-1就表示infinite。                                 // 似乎只能这样才能解释，否则没人触发的话，即使mPendingCount在onNewRequest当中加1了                                 // 这里也感知不到          ALOGD( "Got frame request" );          pthread_mutex_lock(&mFrameDeliveredMutex);          mPendingCount--;          ALOGD( "Completed frame request" );          pthread_cond_signal(&mFrameDeliveredCond);          pthread_mutex_unlock(&mFrameDeliveredMutex);          processPreviewFrame(bufs);      } else {          p_mm_ops->ops->qbuf(mCameraHandle,                  mChannelId, bufs->bufs[ 0 ]); // 如果没有需要数据的情况，直接把buffer压入DRV的队列当中，会call到V4L2的QBUF      } }

1 2 3 4 5 6 7 8 9 10 11

void QCameraStream::onNewRequest() {      ALOGI( "%s:E" ,__func__);      pthread_mutex_lock(&mFrameDeliveredMutex);      ALOGI( "Sending Frame request" );      mPendingCount++;      pthread_cond_wait(&mFrameDeliveredCond, &mFrameDeliveredMutex); // 等带一个请求处理完，再做下一个请求      ALOGV( "Got frame" );      pthread_mutex_unlock(&mFrameDeliveredMutex);      ALOGV( "%s:X" ,__func__); }

processPreviewFrame会调用到创建这个stream的时候关联进来的那个BufferQueue的enqueue_buffer方法，把数据塞到BufferQueue中，然后对应的consumer就会收到了。
比如在Android Camera HAL 2.0当中目前有
camera2/BurstCapture.h
camera2/CallbackProcessor.h
camera2/JpegProcessor.h
camera2/StreamingProcessor.h
camera2/ZslProcessor.h
实现了对应的Consumer::FrameAvailableListener，但是burst-capture现在可以不考虑，因为都还只是stub实现。

ZslProcessor.h和CaptureSequencer.h都有去实现FrameProcessor::FilteredListener的onFrameAvailable(...)
但是我们之前讲过这版QCam HAL没有实现，所以FrameProcessor是无法获取到meta data的。
所以这样来看onFrameAbailable都不会得到通知。(我相信是我手里的这版代码的问题啦)

之前我们说过QCam HAL有部分东西没有实现，所以mFrameQueue就不会有数据，但是它本来应该是DRV回来的元数据会queue到这里面。

另外
CaptureSequencer.h还有去实现onCaptureAvailable，当JpegProcessor处理完了会通知它。

好奇？多个stream(s)不是同时返回的，这样如果CPU处理快慢不同就会有时间差？还有很好奇DRV是如何处理Video snapshot的，如果buffer是顺序的，就会存在Video少一个frame，如果不是顺序的，那就是DRV一次返回多个buffer？以前真没有想过这个问题@_@

This entry was posted in Android, C++, Multimedia and tagged Camera. Bookmark the permalink.

← Lesson Learn之文件被mv给覆盖