Re: [PATCH 31/34] move virtrng_remove to .devexit.text

["Uwe Kleine-König" <u.kleine-koenig@pengutronix.de>]

Hello,

On Thu, Oct 01, 2009 at 11:45:18AM +0200, Michael S. Tsirkin wrote:
> On Thu, Oct 01, 2009 at 11:31:06AM +0200, Uwe Kleine-König wrote:
> > Hello,
> > 
> > On Thu, Oct 01, 2009 at 11:12:18AM +0200, Michael S. Tsirkin wrote:
> > > On Thu, Oct 01, 2009 at 10:44:48AM +0200, Christian Borntraeger wrote:
> > > > Am Donnerstag 01 Oktober 2009 10:28:35 schrieb Uwe Kleine-König:
> > > > > The function virtrng_remove is used only wrapped by __devexit_p so define
> > > > > it using __devexit.
> > > > > 
> > > > > Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
> > > > > Acked-by: Sam Ravnborg <sam@ravnborg.org>
> > > > > Cc: Rusty Russell <rusty@rustcorp.com.au>
> > > > > Cc: Michael S. Tsirkin <mst@redhat.com>
> > > > > Cc: Christian Borntraeger <borntraeger@de.ibm.com>
> > > > 
> > > > Acked-by: Christian Borntraeger <borntraeger@de.ibm.com>
> > > 
> > > FWIW
> > > Acked-by: Michael S. Tsirkin <mst@redhat.com>
> > ok
> > 
> > > > It seems that there are similar changes possible in other virtio drivers (e.g. 
> > > > virtio_net). 
> > http://thread.gmane.org/gmane.linux.kernel/896297/focus=896309
> > 
> > > Yes, and more importantly drivers/virtio/virtio.c as well.
> > Hm, I don't see it:
> > 
> > 	$ git grep -E '__(dev)?exit_p' linus/master:drivers/virtio/virtio.c
> > 
> > 	$
> > 
> > Well, you could add something, but adding __devexit is a noop for most
> > kernels.  It matters only if you don't have CONFIG_HOTPLUG.
> 
> And MODULE.
Well, discarding does not really depend on CONFIG_MODULE.
.devexit sections are only discarded from vmlinux (and not modules) and
only if CONFIG_HOTPLUG=n.  So my statement could be extended to:

Adding __devexit is a noop for most kernels. It matters only if you
don't have CONFIG_HOTPLUG and then only for code that is not compiled as
a module.

> > In this series I only addressed drivers that use __{,dev}exit and
> > __{,dev}exit_p inconsistenly.  I.e. my script greps for __{,dev}exit_p
> > and checks the prototype of the wrapped function.  I have another
> > script that does a similar check for platform_devices in general.  This
> > one also notices if you have a __devexit function that isn't wrapped by
> > __devexit_p.
> 
> Can we teach sparse about this?
I don't know much about sparse, better ask on linux-sparse.
 
> > So if you want to see drivers/virtio/virtio.c improving, send patches
> > yourself :-)
> 
> Here's my reasoning:
> include/linux/virtio.h defines virtio_driver, and remove pointer
> there is only used on hot-unplug or module removal.
> This is the only reason I see that we can make device removal as devexit.
> So we can make all of them devexit then?
Exactly the same applies to platform_drivers.  The remove callback is
only called if the driver is unregistered or the device is unbound.

But note it's not an error in general to use a .text function as remove
callback.  E.g. take drivers/gpio/twl4030-gpio.c.  gpio_twl4030_remove
is used in gpio_twl4030_probe which is defined using __devinit.  So
using __devexit for gpio_twl4030_remove is wrong.  (So there is a bug,
as gpio_twl4030_remove uses __devexit.)  I didn't try, but as far as I
understand this will result in a compile error if the driver is built-in
with HOTPLUG=n.

Best regards
Uwe

-- 
Pengutronix e.K.                              | Uwe Kleine-König            |
Industrial Linux Solutions                    | http://www.pengutronix.de/  |