Re: kbuild: Section mismatch warnings

[Greg KH <gregkh@suse.de>]

On Fri, Feb 17, 2006 at 04:48:37PM -0800, David Brownell wrote:
> On Friday 17 February 2006 4:09 pm, Greg KH wrote:
> > On Fri, Feb 17, 2006 at 11:47:02PM +0100, Sam Ravnborg wrote:
> 
> > > WARNING: drivers/usb/gadget/g_ether.o - Section mismatch: reference to .init.text from .data between 'eth_driver' (at offset 0x10) and 'stringtab'
> > > WARNING: drivers/usb/gadget/g_file_storage.o - ... etc ...
> >
> > David, these all look like they are due to the calls in the
> > drivers/usb/gadget/epautoconf.c file from functions within the gadget
> > drivers.  It looks like it's all safe, but can you verify that the bind
> > callback is finished before module_init() exits?
> 
> If they just usb_gadget_register_driver(), that's how it's defined to
> work yes.  (Though a spec lawyer might want more explicit language ...)
> 
> 
> > And if so, we should mark the bind functions __init also, to prevent
> > this from being flagged in the future.
> 
> And the unbind functions __exit/__exit_p()?  Smaller runtime footprints
> are good.  I don't like leaving the driver->init() method invalid, which
> is I think why I didn't do that before, but saving space is the right
> thing to do.

Ok, care to create a patch for these?

> > > WARNING: drivers/usb/host/isp116x-hcd.o - Section mismatch: reference to .init.text from .data between '' (at offset 0x0) and 'isp116x_hc_driver'
> > 
> > This looks like the isp116x_remove function just needs to get the looney
> > __init_or_module marking of of it.  Again, David, do you agree?
> 
> Right; that marking is for infrastructure, not drivers.

Ok, I'll take care of this, thanks for verifying it.

greg k-h