Em 18-05-2011 16:46, Sakari Ailus escreveu: > Hans Verkuil wrote: >> Note that many video receivers cannot stall. You can't tell them to wait until >> the last buffer finished processing. This is different from some/most? sensors. > > Not even image sensors. They just output the frame data; if the receiver > runs out of buffers the data is just lost. And if any part of the frame > is lost, there's no use for other parts of it either. But that's > something the receiver must handle, i.e. discard the data and increment > frame number (field_count in v4l2_buffer). > > The interfaces used by image sensors, be they parallel or serial, do not > provide means to inform the sensor that the receiver has run out of > buffer space. These interfaces are just unidirectional. Well, it depends on how the hardware works, really. On most (all?) designs, the IP block responsible to receive data from a sensor (or to transmit data, on an output device) is capable of generating an IRQ to notify the OS that a framebuffer was filled. So, the V4L driver can mark that buffer as finished and remove it from the list of the queued buffers. Although the current API's don't allow to create a new buffer if the list is empty, it may actually make sense to allow kernel to dynamically create a new buffer, warranting that the sensor (or receiver) will never run out of buffers under normal usage. Of course, the maximum number of buffers should be specified, to avoid having an unacceptable delay. On such case, the frame will end by being discarded. It makes sense to provide a way to report userspace if this happens. Mauro. -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
