On Mon, 29 Oct 2007 20:04:57 +0200, Vitaliy Ivanov <vitalivanov@gmail.com> wrote:

quoted text
> Finally had a time on my weekend to perform tests and fix Pete's patch a little. Now it seems to be correct.

Great!

quoted text
> 1. One device per user space application. Old driver allowed many users application to work with same device which can lead to IO mess.

OK. This trick was popular in UNIX. Personally I think it's in a bad
taste, because good applications still need to verify if only one
instance is running, and threfore can use application level locking.
But if you are gunning for the maintenership I'm not going to argue
your style. The busy lock-out certainly works better than "/dev/cua" :-)

However, this looks wrong:

quoted text
> +			 dev->interrupt_in_endpoint->bInterval);
> +	dev->read_urb_finished = 0;
> +	retval = usb_submit_urb(dev->interrupt_in_urb, GFP_KERNEL);
> +	/* we ignore failure */
> +	/* end of fixup for first read */
> +
> +	/* initialize out direction */
> +	dev->out_urb_finished = 1;

The finished flag is only set when URB is not in use anymore. Did you
observe an anomaly with my code? Any hangs? If so, I assure you this
is not the fix. As it's written, even if we ignore the failure (e.g.
do not pass it to userland), we sill have to maintain the correct
flag state.

-- Pete
-
unsubscribe notice
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/