Return-Path: <cyrus@holtmann.org>
Message-ID: <1417791601.32209.13.camel@jrissane-mobl.ger.corp.intel.com>
Subject: Re: 6lowpan next header compression framwork: module handling or
 not?
From: Jukka Rissanen <jukka.rissanen@linux.intel.com>
To: Alexander Aring <alex.aring@gmail.com>
Cc: linux-bluetooth@vger.kernel.org, marcel@holtmann.org,
	linux-wpan@vger.kernel.org, martin.townsend@xsilon.com
Date: Fri, 05 Dec 2014 17:00:01 +0200
In-Reply-To: <20141205143156.GC478@omega>
References: <20141205124602.GB478@omega>
	 <1417786926.32209.9.camel@jrissane-mobl.ger.corp.intel.com>
	 <20141205143156.GC478@omega>
Content-Type: text/plain; charset="UTF-8"
Mime-Version: 1.0
List-ID: <linux-bluetooth.vger.kernel.org>

Hi Alex,

On pe, 2014-12-05 at 15:31 +0100, Alexander Aring wrote:
> Hi Jukka,
> 
> On Fri, Dec 05, 2014 at 03:42:06PM +0200, Jukka Rissanen wrote:
> ...
> > I am fine if the "default" being a module, as long as it is
> > automatically loaded. From the end user point of view, he/she should not
> > configure or do any magic in order to use 6lowpan.
> > 
> 
> Seems that I am not a module linux guru. I don't know how I can
> implement such feature. The nhc compression module depends on 6lowpan.ko
> module but 6lowpan.ko can't depend on the nhc compression modules. If
> there would be some dependency then 6lowpan.ko would load the
> compression modules at module_init. Means some symbols are missing and
> "something" try to load the right modules to get these symbols.
> 
> I tried to create such behaviour but failed. I don't know if this is
> really possible or I don't know how I can implement this behaviour.
> 
> What I tried was to implement a "dummy" function, which seems to be also
> a hack in my opnion. This "dummy" function was implemented in the nhc module.
> 
> void lowpan_nhc_rfc6282_udp_load(void) { }
> EXPORT_SYMBOL(lowpan_nhc_rfc6282_udp_load);
> 
> Then lowpan_nhc_rfc6282_udp_load was called at module_init of 6lowpan.ko
> but I got some linking issues.
> 

I am not an expert on kernel modules but perhaps one option could be the
request_module() function that can load the desired module.

http://lxr.free-electrons.com/source/kernel/kmod.c#L118


> 
> Maybe I need some time to getting the idea how I can implement such
> behaviour without ugly hacks. Currently I don't know how I can implement
> something like that. Maybe the solution is easy.
> 
> - Alex


Cheers,
Jukka