Messages in this thread | | | Date | Wed, 19 Jun 2013 11:32:31 -0700 | From | Greg Kroah-Hartman <> | Subject | Re: [PATCH] lp: implement proper detach function for parport_driver lp |
| |
On Wed, Jun 19, 2013 at 07:04:51PM +0200, Hannes Weisbach wrote: > >> Signed-off-by: Hannes Weisbach <hannes_weisbach@gmx.net> > >> --- > >> Granted, for normal parport drivers this is usually not an issue, > >> because the device does not go away. However, I am currently writing a > >> Linux device driver for a USB to parallel port converter [0] and > >> therefore need proper detaching. Additionally, the wrong ref count > >> keeps me from simply rmmod my driver and insmod a new version while > >> developing and testing. > > > > Really? We already have a usb to parallel port driver in the kernel > > tree that seems to work just fine. > > > It's been there since the 2.3 kernel > > days, so either it has the same problem, or your driver is doing > > something odd. > > Do you mean the USB Printer class driver for pseudo parallel port > adapters? They don't use the char/lp.c printer driver. (Or I didn't > see it). I'm writing a driver for USB2LPT [0], which gives you a real > /dev/parportN-device in user space, with which you can do all the > bit-twiddling like with a real parallel port. > > Or do you mean USS720 (misc/uss720.c) and MOS7715 (serial/mos7720.c) > drivers.
Yes, I mean these.
> They are doing what I am doing: translating whatever the > user does on a /dev/parportN-node and sending device-specific commands > over USB. When they do parport_announce_port(), lp.c should also be > initialized and they should have the same problem.
Why hasn't anyone reported that problem then? Surely someone must use these, they've been in the kernel tree for over a decade now.
> On second thought, my patch might not be optimal. lp.c stores > instance-structs in an array of size 8. So after 8 re-plugs, lp.c > will not instantiate any more printer devices. I think I better go > all the way and replace that array with a list, to have a proper > solution.
Proper solutions are always good :)
thanks,
greg k-h
| |