本文均属自己阅读源码的点滴总结,转账请注明出处谢谢。
欢迎和大家交流。qq:1037701636 email:[email protected]
Android源码版本Version:4.2.2; 硬件平台 全志A31
step1:之前在讲到CameraService处的setPreviewWindow中传入一个窗口给HAL
status_t setPreviewWindow(const sp<ANativeWindow>& buf) { ALOGV("%s(%s) buf %p", __FUNCTION__, mName.string(), buf.get()); if (mDevice->ops->set_preview_window) { mPreviewWindow = buf; mHalPreviewWindow.user = this; ALOGV("%s &mHalPreviewWindow %p mHalPreviewWindow.user %p", __FUNCTION__, &mHalPreviewWindow, mHalPreviewWindow.user); return mDevice->ops->set_preview_window(mDevice, buf.get() ? &mHalPreviewWindow.nw : 0);//调用底层硬件hal接口 } return INVALID_OPERATION; }
传入给HAL的参数buf为一个Surface,还有一个变量是关于预览窗口的数据流操作nw
struct camera_preview_window { struct preview_stream_ops nw; void *user; };
该变量的初始如下:这些函数接口看上去很熟悉,的确在SurfaceFlinger中,客户端的Surface也就是通过这些接口来向SurfaceFlinger申请图形缓存并处理图形缓存显示的,只是之前的操作都交个了OpenGL ES的eglswapbuf()来对这个本地窗口进行如下的dequeueBuffer和enqueuebuffer的操作而已。而在Camera的预览中,这些操作将手动完成。
void initHalPreviewWindow() { mHalPreviewWindow.nw.cancel_buffer = __cancel_buffer; mHalPreviewWindow.nw.lock_buffer = __lock_buffer; mHalPreviewWindow.nw.dequeue_buffer = __dequeue_buffer; mHalPreviewWindow.nw.enqueue_buffer = __enqueue_buffer; mHalPreviewWindow.nw.set_buffer_count = __set_buffer_count; mHalPreviewWindow.nw.set_buffers_geometry = __set_buffers_geometry; mHalPreviewWindow.nw.set_crop = __set_crop; mHalPreviewWindow.nw.set_timestamp = __set_timestamp; mHalPreviewWindow.nw.set_usage = __set_usage; mHalPreviewWindow.nw.set_swap_interval = __set_swap_interval; mHalPreviewWindow.nw.get_min_undequeued_buffer_count = __get_min_undequeued_buffer_count; }
step2.继续前面的preview的处理操作,在CameraService处的CameraClinet已经调用了CameraHardwareInterface的startPreview函数,实际就是操作HAL处的Camera设备如下
status_t startPreview() { ALOGV("%s(%s)", __FUNCTION__, mName.string()); if (mDevice->ops->start_preview) return mDevice->ops->start_preview(mDevice); return INVALID_OPERATION; }
step3.进入HAL来看Preview的处理
status_t CameraHardware::doStartPreview(){...........res = camera_dev->startDevice(mCaptureWidth, mCaptureHeight, org_fmt, video_hint);//启动设备......}
调用V4L2设备来启动视频流的采集,startDevice()函数更好的解释了预览的启动也就是视频采集的启动。
status_t V4L2CameraDevice::startDevice(int width, int height, uint32_t pix_fmt, bool video_hint){ LOGD("%s, wxh: %dx%d, fmt: %d", __FUNCTION__, width, height, pix_fmt); Mutex::Autolock locker(&mObjectLock); if (!isConnected()) { LOGE("%s: camera device is not connected.", __FUNCTION__); return EINVAL; } if (isStarted()) { LOGE("%s: camera device is already started.", __FUNCTION__); return EINVAL; } // VE encoder need this format mVideoFormat = pix_fmt; mCurrentV4l2buf = NULL; mVideoHint = video_hint; mCanBeDisconnected = false; // set capture mode and fps // CHECK_NO_ERROR(v4l2setCaptureParams()); // do not check this error v4l2setCaptureParams(); // set v4l2 device parameters, it maybe change the value of mFrameWidth and mFrameHeight. CHECK_NO_ERROR(v4l2SetVideoParams(width, height, pix_fmt)); // v4l2 request buffers int buf_cnt = (mTakePictureState == TAKE_PICTURE_NORMAL) ? 1 : NB_BUFFER; CHECK_NO_ERROR(v4l2ReqBufs(&buf_cnt));//buf申请 mBufferCnt = buf_cnt; // v4l2 query buffers CHECK_NO_ERROR(v4l2QueryBuf());//buffer的query,完成mmap等操作 // stream on the v4l2 device CHECK_NO_ERROR(v4l2StartStreaming());//启动视频流采集 mCameraDeviceState = STATE_STARTED; mContinuousPictureAfter = 1000000 / 10; mFaceDectectAfter = 1000000 / 15; mPreviewAfter = 1000000 / 24; return NO_ERROR;}
这个是完全参考了V4L2的视频采集处理流程:
1.v4l2setCaptureParams()设置采集的相关参数;
2.v4l2QueryBuf():获取内核图像缓存的信息,并将所有的内核图像缓存映射到当前的进程中来。方便用户空间的处理
3.v4l2StartStreaming():开启V4L2的视频采集流程。
step4: 图像采集线程bool V4L2CameraDevice::captureThread();
该函数的内容比较复杂,但核心是 ret = getPreviewFrame(&buf),获取当前一帧图像:
int V4L2CameraDevice::getPreviewFrame(v4l2_buffer *buf){ int ret = UNKNOWN_ERROR; buf->type = V4L2_BUF_TYPE_VIDEO_CAPTURE; buf->memory = V4L2_MEMORY_MMAP; ret = ioctl(mCameraFd, VIDIOC_DQBUF, buf); //获取一帧数据 if (ret < 0) { LOGW("GetPreviewFrame: VIDIOC_DQBUF Failed, %s", strerror(errno)); return __LINE__; // can not return false } return OK;}
调用了典型的VIDIOC_DQBUF命令,出列一帧图形缓存,提取到用户空间供显示。
当前的平台通过定义一个V4L2BUF_t结构体来表示当前采集到的一帧图像,分别记录到Y和C所在的物理地址和用户空间的虚拟地址。虚拟地址是对内核采集缓存的映射
typedef struct V4L2BUF_t{ unsigned int addrPhyY; // physical Y address of this frame unsigned int addrPhyC; // physical Y address of this frame unsigned int addrVirY; // virtual Y address of this frame unsigned int addrVirC; // virtual Y address of this frame unsigned int width; unsigned int height; int index; // DQUE id number long long timeStamp; // time stamp of this frame RECT_t crop_rect; int format; void* overlay_info; // thumb unsigned char isThumbAvailable; unsigned char thumbUsedForPreview; unsigned char thumbUsedForPhoto; unsigned char thumbUsedForVideo; unsigned int thumbAddrPhyY; // physical Y address of thumb buffer unsigned int thumbAddrVirY; // virtual Y address of thumb buffer unsigned int thumbWidth; unsigned int thumbHeight; RECT_t thumb_crop_rect; int thumbFormat; int refCnt; // used for releasing this frame unsigned int bytesused; // used by compressed source}V4L2BUF_t;
来看看该结构体的初始化代码:
V4L2BUF_t v4l2_buf; if (mVideoFormat != V4L2_PIX_FMT_YUYV && mCaptureFormat == V4L2_PIX_FMT_YUYV) { v4l2_buf.addrPhyY = mVideoBuffer.buf_phy_addr[buf.index]; v4l2_buf.addrVirY = mVideoBuffer.buf_vir_addr[buf.index]; } else { v4l2_buf.addrPhyY = buf.m.offset & 0x0fffffff;//内核物理地址 v4l2_buf.addrVirY = (unsigned int)mMapMem.mem[buf.index];//虚拟地址 } v4l2_buf.index = buf.index;//内部采集缓存的索引 v4l2_buf.timeStamp = mCurFrameTimestamp; v4l2_buf.width = mFrameWidth; v4l2_buf.height = mFrameHeight; v4l2_buf.crop_rect.left = mRectCrop.left; v4l2_buf.crop_rect.top = mRectCrop.top; v4l2_buf.crop_rect.width = mRectCrop.right - mRectCrop.left + 1; v4l2_buf.crop_rect.height = mRectCrop.bottom - mRectCrop.top + 1; v4l2_buf.format = mVideoFormat;
addrPhy和addrViry分别记录到Y和C所在的物理地址和用户空间的虚拟地址。而这个地址都是通过当前Buf的index直接设置的,为什么?因为内核的图像缓存区的mmap操作将每一个缓存,以其Index分别逐一的映射到了用户空间,并记录缓存的物理和虚拟地址,而这主要是方便后续图像的显示而已。
step5:bool V4L2CameraDevice::previewThread()//预览线程
获得了一帧数据必须通知预览线程进行图像的显示,采集线程和显示线程之间通过pthread_cond_wait(&mPreviewCond, &mPreviewMutex);进程间锁进行等待。
bool V4L2CameraDevice::previewThread()//预览线程{ V4L2BUF_t * pbuf = (V4L2BUF_t *)OSAL_Dequeue(&mQueueBufferPreview);//获取预览帧buffer信息 if (pbuf == NULL) { // LOGV("picture queue no buffer, sleep..."); pthread_mutex_lock(&mPreviewMutex); pthread_cond_wait(&mPreviewCond, &mPreviewMutex);//等待 pthread_mutex_unlock(&mPreviewMutex); return true; } Mutex::Autolock locker(&mObjectLock); if (mMapMem.mem[pbuf->index] == NULL || pbuf->addrPhyY == 0) { LOGV("preview buffer have been released..."); return true; } // callback mCallbackNotifier->onNextFrameAvailable((void*)pbuf, mUseHwEncoder);//回调采集到帧数据 // preview if (isPreviewTime())//预览 { mPreviewWindow->onNextFrameAvailable((void*)pbuf);//帧可以显示 } // LOGD("preview id : %d", pbuf->index); releasePreviewFrame(pbuf->index); return true;}
预览线程主要做了两件事,一是完成图像缓存数据的回调供最最上层的使用;另一件当然是送显。
step6:预览线程如何显示?
bool PreviewWindow::onNextFrameAvailable(const void* frame)//使用本地窗口surface 进行初始化{ int res; Mutex::Autolock locker(&mObjectLock); V4L2BUF_t * pv4l2_buf = (V4L2BUF_t *)frame;//一帧图像所在的地址信息...... res = mPreviewWindow->set_buffers_geometry(mPreviewWindow, mPreviewFrameWidth, mPreviewFrameHeight, format);//设置本地窗口的buffer的几何熟悉...... res = mPreviewWindow->dequeue_buffer(mPreviewWindow, &buffer, &stride);//申请SF进行bufferqueue的图形缓存操作。返回当前进程地址到buffer.................. res = grbuffer_mapper.lock(*buffer, GRALLOC_USAGE_SW_WRITE_OFTEN, rect, &img);//把映射回来的buffer信息中的地址放到img中,用来填充............. mPreviewWindow->enqueue_buffer(mPreviewWindow, buffer);//交由surfaceFlinger去做显示............}
上述代码实时了本地窗口图像向SurfaceFlinger的投递,为何这么说,看下面的分析:
1.PreviewWindow类里的mPreviewWindow成员变量是什么?
这个是从应用端的setPreviewDisplay()设置过来的,传入到HAL的地方在CameraHardwareInterface的initialize函数里:
return mDevice->ops->set_preview_window(mDevice, buf.get() ? &mHalPreviewWindow.nw : 0);//调用底层硬件hal接口 }
nw的操作在step1里面已经有说明了,初始化相关的一些操作。
2.以dequeue_buffer为例:
static int __dequeue_buffer(struct preview_stream_ops* w, buffer_handle_t** buffer, int *stride) { int rc; ANativeWindow *a = anw(w); ANativeWindowBuffer* anb; rc = native_window_dequeue_buffer_and_wait(a, &anb); if (!rc) { *buffer = &anb->handle; *stride = anb->stride; } return rc; }
调用到本地的窗口,通过w获得ANativeWindow对象,来看看该宏的实现:
static ANativeWindow *__to_anw(void *user) { CameraHardwareInterface *__this = reinterpret_cast<CameraHardwareInterface *>(user); return __this->mPreviewWindow.get(); }#define anw(n) __to_anw(((struct camera_preview_window *)n)->user)
首先获取user对象为CameraHardwareInterface对象,通过它获得之前初始化的Surface对象即成员变量mPreviewWindow(属于本地窗口ANativeWindow类)。
3.本地窗口的操作
static inline int native_window_dequeue_buffer_and_wait(ANativeWindow *anw, struct ANativeWindowBuffer** anb) { return anw->dequeueBuffer_DEPRECATED(anw, anb);}
上述的过程其实是调用应用层创建的Surface对象,该对象已经完全打包传递给了CameraService,来进行绘图和渲染的处理。如下所示:BpCamera
// pass the buffered Surface to the camera service status_t setPreviewDisplay(const sp<Surface>& surface) { ALOGV("setPreviewDisplay"); Parcel data, reply; data.writeInterfaceToken(ICamera::getInterfaceDescriptor()); Surface::writeToParcel(surface, &data);//数据打包 remote()->transact(SET_PREVIEW_DISPLAY, data, &reply); return reply.readInt32(); }
BnCamera处,内部实现了新建一个CameraService处的Surface,但是都是用客户端处的参数来初始化的。即两者再不同进程中,但所包含的信息完全一样。
case SET_PREVIEW_DISPLAY: { ALOGV("SET_PREVIEW_DISPLAY"); CHECK_INTERFACE(ICamera, data, reply); sp<Surface> surface = Surface::readFromParcel(data); reply->writeInt32(setPreviewDisplay(surface));//设置sueface return NO_ERROR; } break;
Surface::Surface(const Parcel& parcel, const sp<IBinder>& ref) : SurfaceTextureClient(){ mSurface = interface_cast<ISurface>(ref); sp<IBinder> st_binder(parcel.readStrongBinder()); sp<ISurfaceTexture> st; if (st_binder != NULL) { st = interface_cast<ISurfaceTexture>(st_binder); } else if (mSurface != NULL) { st = mSurface->getSurfaceTexture(); } mIdentity = parcel.readInt32(); init(st);}
这里的Surface建立是通过mSurface来完成和SurfaceFlinger的通信的,因为之前Camera客户端处的Surface是和SurfaceFLinger进行Binder通信,现在要将原先的Bpxxx相关的写入到CameraService进一步和SurfaceFlinger做后续的Binder通信处理,如queueBuffer()处理中和SurfaceFlinger的Bufferqueue的通信等。
4.故anw->dequeueBuffer的函数就和之前的从Android Bootanimation理解SurfaceFlinger的客户端建立完全对应起来,而且完全一样,只是Bootanimation进程创建的Surface交给OpenGL Es来进行底层的比如dequeue(缓存申请,填充当前的buffer)和enqueue(入列渲染)的绘图操作而已,见Android4.2.2 SurfaceFlinger之图形缓存区申请与分配dequeueBuffer一文。具体的绘图就不在这里说明了。通过该方法已经和SurfaceFlinger建立起连接,最终交由其进行显示。