Re: [Gumstix-users] uclibc's buildroot

[Alexandre P. N. <al...@om...>]

Craig Hughes escreveu:

> I definitely want to move to the latest buildroot and merge all the 
> gumstix in.  Right now I'm balancing 2 trees that I'm focusing on -- 
> bugfixes etc on the 2.6.11 trunk branch, and trying to finalize 
> 2.6.13 to make it the new trunk.  Once I've finalized 2.6.13 I'm 
> going to focus on re-syncing with the latest mainline buildroot, 
> including sending a bunch of patches upstream once I'm in sync.
>
> The biggest roadblock right now is the 2.6.13 stuff, and within that, 
> specifically multi-NIC support under the smc91x driver.  I've taken 2 
> or 3 cuts at getting that working now, and run into brick walls with 
> each attempt.  My next step is to post a question or two about it on 
> the linux-arm mailing list, but I just haven't had the time to do so 
> properly yet.  I think rather than forking a branch now to sync up 
> with the latest buildroot, the shortest path to getting to the new 
> buildroot setup would be to help get the dual-NIC stuff working, if 
> you want to help with that :)  If you are interested in helping get 
> that working, I don't think you have a netDUO, so I can hook up a 
> netDUO setup here with remote access and poke some holes in our 
> firewall to let you in to play on it :)


I can try to take a look at that, indeed I'll need you to set up a
remote access for me. I'll wait for you on this... I'm sending my pgp
key, so you can write for me securely if you prefer to pass along login
information.

>
> As for native compilation, Gordon took I think 6-8 of the latest 
> batch of gumstix and etherstix when they came in from the CM and hung 
> them off his 100Mbit switch, with the intention of having them run 
> distcc with some NFS filesystem mounted to them all.  That cluster 
> right now just has the standard root_fs flashed on it, but a root_fs 
> and NFS tree setup which would facilitate native clustered 
> compilation would be another great thing to work on -- our intention 
> is to open up access to the native-compile-cluster to anyone that 
> wants to use it, with some kind of throttling TBD so that we don't 
> have too many people trying to use it at once.  You're welcome to be 
> the first person to use it though if you want to setup a root_fs/NFS 
> combo, I can flash the 'stix and set up the NFS share on our LAN.

Right now, I've found a way to match my needs regarding compilation
issues, but I like this idea very much. I'll keep thinking about it and
let's see what come out.

>
> I've also done some very preliminary looking into having the "make" 
> run on the gumstix cluster, with distcc making use of some bigger 
> hardware with cross-compilers on it so that the packages can believe 
> they're compiling on the gumstix, but through distcc magic will 
> actually be using CPU resources elsewhere for the actually 
> compilation.  It haven't been able to fully to the magic-foo required 
> to get this working yet though.  Even having a single distccd cross-
> compiler with a single gumstic distcc client would be a good starting 
> point, but that hasn't worked right for me yet.


That's a very cool thing to play with! And if a cross compiler doesn't
work, I can think of a really, really dirty way of doing something which
would work somewhat similar.  Later, when you have the time, tell me
more about your steps and results on this, so that we can try something
without starting from scratch.