Hi Carl, On Friday 17 June 2011 23:33:25 Carl Michal wrote:
[snip] > I've been looking at lengths of packets and headers in packets where these > unexpected STI, RES, EOH and ERROR bits are set. > > There are many things that seem inconsistent, but there are some patterns > that come up repeatedly. For example getting a packet with an ERROR bit > set that has a length of 2048 and a header length (from data[0]) of 127 or > 125 or 110 or some other big number. That packet gets discarded because of > the error bit, but then when the frame ends, it is 2048 bytes shorter than > it should be. I've also seen that with a packet length of 1024 and the > frame ending 1024 bytes short. > > It looks like the header has somehow disappeared and the image data is > being interpreted as header. Good catch. > To test this, I put in checks in uvc_video_decode_isoc for any EOH, STI, RES > or ERROR - if any are set (or unset for EOH) then to return a header length > of 0, so that the entire packet gets used as image data. > > This by no means solves all my problems, but with that - I do find at least > some frames assembled this way are complete and uncorrupted. > > So somehow some headers are disappearing? That's totally against the UVC specs. We could work around it in the driver if there was a reliable way to detect that a frame doesn't start with a header, but that's not possible. I would consider the camera as completely broken. -- Regards, Laurent Pinchart _______________________________________________ Linux-uvc-devel mailing list Linux-uvc-devel@lists.berlios.de https://lists.berlios.de/mailman/listinfo/linux-uvc-devel
