Date: Sat, 17 Mar 2001 12:40:20 -0500 (EST)
From: "Mark S. Mathews" <mark@linux-wlan.com>
To: linux-wlan-user@absoval.com
Subject: Re: [lwlan-user]: Re: linux-wlan-ng-0.1.8-pre5 works perfectly(?)with Orinoco Silver
In-Reply-To: <Pine.LNX.4.10.10103171709060.11265-100000@localhost.localdomain>
Message-ID: <Pine.LNX.4.21.0103171231530.24501-100000@chef.cc.absoval.com>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
Sender: owner-linux-wlan-user@nightsky.com
Precedence: bulk
Reply-To: linux-wlan-user@absoval.com

On Sat, 17 Mar 2001, Karlis Peisenieks wrote:

> On Fri, 16 Mar 2001, Mark S. Mathews wrote:
> 
> > I think this is a issue w/ 2.2.17 and 2.2.18.  I haven't spent much time
> > with them, but I think there's a symbol issue. (DUH!)  I haven't spent
> 
> Everything works fine on 2.2.18. Just netlink (CONFIG_NETLINK) has to be
> enabled.
> 
> Still you have to make sure that you compile wlan with kernel source that
> is configured as your running kernel (due to modversions).
> 
> What is output of commands:
> 
> grep netlink_kernel_create /proc/ksyms
> and
> objdump -r prism2sta_cs.o | grep netlink_kernel_create
> ?

Good tip, I overlooked the running kernel vs. source tree thing.  Once
again, I've missed the obvious while looking for a more complicated
cause.  Forest and trees......   ;-)

> > One thing I'm considering is not using netlink at all.  I'm thinking about
> > using a Unix domain socket instead of nelink for usermode event
> > notification.  We'll see.
> 
> Why? I suspect you will have a lot more trouble implementing kernel space
> unix socket "server" than making a FAQ with answer to given problem :).
> Especially taking into account that netlink was created exactly for such
> purposes (kernel/user space app communication). Netlink gives clean and
> easy way to iplement kernel-userapp interface for both - kernel side
> (which is the most important thing I guess) and user app.
> 
> I think that it would even be better (read - easier) to make "virtual"
> character device driver that would do this communication rather than using
> unix sockets.

The primary reason I was considering a move away from netlink is because
of portability.  We've been getting pinged for Xbsd support lately (and I
have virtually no experience w/ bsd).  I don't think there's a netlink
equivalent in the bsd's.   The reason I was considering a
unix domain socket rather than a char device is that I felt it might be a
tad more portable.  If I were to do a char device, I'd probably implement
it as a misc device with the minor number allocated at runtime
(usermode apps would retrieve the minor number via an ioctl query)....and
I'm not sure if _that's_ supported on other platforms.

I agree that netlink is a perfect solution (and very nicely done, thanks
AC).  Additionally, I love its simplicity.  It isn't particularly well
documented, but the source is pretty straightforward.

-Mark

-- 

Mark S. Mathews

AbsoluteValue Systems      Web:    http://www.linux-wlan.com
P.O. Box 410670            e-mail: mark@linux-wlan.com
Melbourne, FL 32941-0670   Phone:  321.259.0737
USA                        Fax:    321.259.0286

-----------------------------------------------
The Linux WLAN User's Mailing List
For more information about this list see:
http://www.absoval.com/linux-wlan/lists.html