Hello again list, I wrote: > We're seeing problems using cameras based on Geo GC6500 in Linux > kernels 4.19+ in raw mode (as opposed to H.264 mode, to avoid IP > issues). > Putting frame size reporting/calculations aside for now Now, on to frame size calculations. There's a small gadget in the code seemingly existing since the dawn of time: https://github.com/torvalds/linux/blob/95f5cbff90b9e4324839a5c28ee3153a3c9921a5/drivers/media/usb/uvc/uvc_video.c#L119 if (!(format->flags & UVC_FMT_FLAG_COMPRESSED) || (ctrl->dwMaxVideoFrameSize == 0 && stream->dev->uvc_version < 0x0110)) ctrl->dwMaxVideoFrameSize = frame->dwMaxVideoFrameBufferSize; It seems to be overwriting a closer-to-the-truth value of 3110412 in dwMaxVideoFrameSize with frame->dwMaxVideoFrameBufferSize which happens to be 4147200 and is used from then on, including the uvc_video_validate_buffer() check. FWIW uvc_version is 0x0110. The value is still not quite 3110400, but even if it were so (e.g. to please uvc_video_validate_buffer()), it would have been overwritten anyway. The condition block logic seems worth double-checking (sure it's || not &&?). My original question on whether a value below the maximum is valid remains (hope the answer is obvious). Best regards, -- DoubleF
