Avoid large reads using parport_read() ?

I've been involved with the cpia_pp driver (parallel port Webcam), and found that X becomes very sluggish when capturing a stream of images. I suspect that this is due to reading the whole of each frame using parport_read(), which could be up to 200K.

In the 2.0 kernel days (pre-parport layer), the cpia_pp driver would do its own parallel port banging, and would actually only read 16 bytes (fifo size) at a time before checking if it needed to reschedule. Now it fetches the whole image using a single call to parport_read().

I did a quick check, and instead of issuing one parport_read for the whole image, I looped round issuing reads of 4096 bytes. The responsiveness of the machine was much improved.

Is there any problem with looping calls to parport_read() with a smaller buffer size, and performing cond_resched() each time through the loop? What size of buffer should I use?

 `O O'  | [email protected]
// ^ \\ | http://www.pyrites.org.uk/