Re: [REGRESSION] "bind" a device to a driver doesn't not work anymore

[Greg KH <gregkh@suse.de>]

On Mon, Oct 12, 2009 at 08:45:54AM -0700, Dmitry Torokhov wrote:
> On Mon, Oct 12, 2009 at 07:44:58AM -0700, Greg KH wrote:
> > On Mon, Oct 12, 2009 at 01:46:17PM +0200, Éric Piel wrote:
> > > Op 12-10-09 06:35, Dmitry Torokhov schreef:
> > > > On Saturday 10 October 2009 08:00:29 pm Greg KH wrote:
> > > >> On Sun, Oct 11, 2009 at 02:04:02AM +0200, ?ric Piel wrote:
> > > >>> Hello,
> > > >>>
> > > >>> After a suspend, I used to do this (to get the PS2 keyboard working
> > > >>> again), and it used to work fine:
> > > >>> echo -n "i8042" > /sys/bus/platform/drivers/i8042/unbind ; sleep 1 ; echo
> > > >>> -n "i8042" > /sys/bus/platform/drivers/i8042/bind
> > > >>>
> > > >>> Unfortunately since 2.6.32-rc1, the unbind works fine, but the bind
> > > >>> returns a "no such device" error now (and the binding is not done). After
> > > >>> that, additional unbinds also report the error. I think 2.6.31 worked
> > > >>> fine. Any idea what's going wrong?
> > > >> Odd.  Any chance you can run 'git bisect' to figure out which patch
> > > >> might have caused it?
> > > > 
> > > > I8042 has been converted to use platform_driver_probe() - we are creating 
> > > > platform device and platform driver together, there is no chance that some 
> > > > other device might be driven by i8042 driver.
> > > I haven't finished bisecting, but indeed, the problem seemed to converge
> > > toward this patch.
> > > 
> > > 
> > > > Eric, do you still need to fiddle with i8042 to get your keyboard working with 
> > > > 2.6.32-rc3? We need to make sure that resube works without users needing to 
> > > > mess with bind/unbind.
> > > Yes, the keyboard doesn't work after the first resume from ram (the
> > > later resumes do work). I haven't tried any special option to the i8042
> > > driver. This is on a HP 2510, in 64 bits.
> > > 
> > > Indeed, it would be great to fix this root problem! Is there any bug
> > > report opened to work on this?
> > > 
> > > However, currently the first problem is that unbind works, while bind
> > > doesn't. Which means anyone doing an unbind will have keyboard and mouse
> > > lost until the next reboot. That's rather bad because it has been
> > > recommended for a long time to have this kind of script at resume:
> > > http://intertrusion.com/files/hp-s2ram.patch
> > > http://en.opensuse.org/Pm-utils
> > > 
> > > So at least, unbind should fail as well as bind.
> > 
> > Ah, I bet I know why bind fails...  Yeah, the i8042_platform_device is
> > never freed up in the i8042_remove() function in
> > drivers/input/serio/i8042.c, so when you try to register it again, it
> > fails.
> > 
> 
> Huh? Platform device and driver are created in i8042_init() and removed
> in i8042_exit(); i8042_remove() is a remove() method of the platform
> driver so it never should be freeing the platform device.

Ah, but you do not have the ability to allow a bind to ever happen
again.  I guess that's fine if you don't want that to happen.  And for
platform devices, it's kind of difficult now that I think about it some
more.

> The issue is that with platform drivers using i8042_driver_probe() the
> probe frunction is discarded afterwards so the driver can't be bound to
> anything after initialization has been completed; however unbinding
> through 'unbind' attribute still works (once). You may recall a thread
> with me and David arguing about it a couple of weeks ago...

Yeah, I remember that...

At least we now know the problem.  Just rmmod and insmod the module to
unbind/bind and everyone will be happy :)

thanks,

greg k-h