Message-ID: <4AB65CFE.7090101@canonical.com>
Date: Sun, 20 Sep 2009 10:49:02 -0600
From: Tim Gardner <tim.gardner@canonical.com>
Reply-To: tim.gardner@canonical.com
MIME-Version: 1.0
To: "Luis R. Rodriguez" <lrodriguez@atheros.com>
CC: linux-wireless@vger.kernel.org, Greg KH <greg@kroah.com>,
	Johannes Berg <johannes@sipsolutions.net>
Subject: Re: [RFC] compat-2.6: mangle symbols for driver-select
References: <20090919205545.GA18080@bombadil.infradead.org> 	<4AB5A5A7.3020107@canonical.com> <43e72e890909200743h74934657nc056c9e6dcb14e90@mail.gmail.com>
In-Reply-To: <43e72e890909200743h74934657nc056c9e6dcb14e90@mail.gmail.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Sender: linux-wireless-owner@vger.kernel.org

Luis R. Rodriguez wrote:
> On Sat, Sep 19, 2009 at 8:46 PM, Tim Gardner <tim.gardner@canonical.com> wrote:
>> Luis R. Rodriguez wrote:
>>> Today at the summit we spoke about mangling symbols for driver-slect.
>>> Here's a quick nasty take on this but without doing this for driver-select
>>> specifically just for testing. It seems to compile, but someone more
>>> motivated
>>> may want to test and make this apply somehow only for driver-select or
>>> perhaps
>>> when a -D define is used.
>>>
>>> Reason for this is to help distributions / OEMs / ODMs who want to replace
>>> just *one* driver with compat-wireless.
>>>
>> I think it would be better to generate the list of mangled symbols
>> dynamically.
> 
> Agreed, that is the part I left out, as a TODO to someone interested.
> 

I'm working on a patch (it works with 2.6.31), but I need to test with 
older kernels.

>> In older Ubuntu releases (before depmod behavior was
>> corrected), we have to run a 'munge' script to preface all of the exported
>> symbols so that a compat-wireless driver references the compat-wireless
>> protocol stack symbols. See the attached munge script for compat-wireless on
>> 2.6.24.
> 
> Please correct me if I'm wrong but I believe the script seems to do
> what I did just that it actually edited the files with the changes, I
> prefer the way I did this as it requires less work to maintain and
> understand IMHO.
> 

I'm fine with that, in fact its a bit faster since my munge script is a 
bit slow. It'll also make dealing with kernel version dependent symbol 
munge exceptions a bit simpler.

One thing that is worth mentioning is that the module names need to be 
changed for older user space environments, otherwise depmod mucks things 
up in interesting ways.

>> We can either do something like this for compat-wireless, or we
>> could use a subset of this logic to generate the list of symbols contained
>> within the '#ifdef CONFIG_COMPAT_WIRELESS_MANGLE' clause.
> 
> So I was under the impression you would use this only if you are using
> ./scripts/driver-select to select one driver out of the whole tree,
> but it seems you actually use this for all the drivers on
> compat-wireless for the Ubuntu linux-backports-modules package. I take
> it you put lbm stuff then into some /lib/modules/$(uname)/compat/ and
> use a sort of /etc/depmod.d/01-compat.conf to prefer compat over
> updates/ or kernel/ ?

For LBM I've taken a scorched earth approach, i.e., _all_ drivers in 
compat-wireless get built. The vast bulk of users that I deal with are 
only interested in one driver, but I don't know a priori _which_ driver. 
The use case where they would want to use a mainline driver at the same 
time as a different compat-wireless driver is fairly rare (which my 
approach makes impossible).

rtg
-- 
Tim Gardner tim.gardner@canonical.com