[12:22] <lamont> jbailey: because on some architecture families (ppc, hppa) there isn't _one_ kernel that works on all supported instances of the architecture.
[12:28] <jbailey> Okay, so discard that concept then and it can depend on an either or.
[12:28] <jbailey> We really need versioned provides.
[12:29] <jbailey> Hmm, it could depend on hppa32 | hppa64 or something.
[12:29] <jbailey> but still be made by linux-meta so that we have something we can depend on as the installed version.
[06:22] <fabbione> morning
[11:50] <chmj> fabbione: have you started backporting ipw2100, are u sticking with that solution for now ?
[11:50] <fabbione> chmj: i am working on it right now...
[11:50] <fabbione> i need to dig into some stuff first
[11:50] <chmj> ok 
[12:18] <fabbione> ok i found a matching set
[12:18] <fabbione> ipw2100-1.0.0 with ipw2200-1.0.1
[12:18] <fabbione> they share the same ieee subset
[12:19] <fabbione> after that release the ieee has been changed a lot
[12:19] <fabbione> that could explain the mess
[12:27] <chmj> seems like it 
[02:48] <jbailey> fabbione: *poke*
[03:00] <chmj> fabbione: ipw2100 failed 
[03:00] <fabbione> chmj: how?
[03:01] <chmj> strangely 
[03:01] <chmj> when I boot, dhcp fails 
[03:02] <fabbione> chmj: i already received 3 reports of it working
[03:02] <chmj> it just assigns to a 192.168.0.195 
[03:02] <fabbione> chmj: that's not a kernel bug clearly...
[03:02] <fabbione> what happens if you do dhcp manually?
[03:03] <chmj> when i run dhclient again, it works 
[03:03] <fabbione> than it's a dhcpd problem or an AP problem
[03:03] <fabbione> the driver works
[03:03] <chmj> but after I use it for a while, it looses the IP address 
[03:03] <fabbione> yeah.. dhcpd problem
[03:03] <fabbione> check the server
[03:03] <fabbione> they are not talking properly
[03:04] <chmj> all kern.log says it "no ipv6 routers found"
[03:04] <fabbione> yeah.. that's ok.. it means you are not in an ipv6 network
[03:08] <chmj> strange becouse it work fine with .10 
[03:08] <chmj> no problems whatsoever 
[06:42] <Lathiat> Hey Guys, is there any chance of inclusion of the host skas3 patch in ubuntu's kernel?
[06:43] <fabbione> no
[06:44] <fabbione> all this kind of patches are unmaintanable in the long run
[06:47] <Lathiat> right
[06:47] <Lathiat> i wonder if the evil module hack is still maintained :)
[07:33] <jbailey> Last year, there was a guy who had moved the IDE drivers into userspace at OLS.
[07:33] <jbailey> I wonder if that's gotten anywhere?
[07:34] <Lathiat> uh, why?
[07:34] <Lathiat> one thing thatd be nice to have in userspace is ppp
[07:34] <jbailey> Because once you can get something as key as your IDE driver efficiently into userspace, you've got a reasonable model for alot of things.
[07:35] <Lathiat> doesnt that mean that your goign to be context switchign all over the joint tho?
[07:35] <jbailey> Read in the word 'efficiently'
[07:35] <Lathiat> sure, but how can you do it efficiently?
[07:35] <jbailey> His tests showed a speed improvement over in-kernel IDE drivers.
[07:35] <jbailey> Dunno, he was still doing the research on why it was faster.
[07:36] <jbailey> I know how it works in the L4 world, where context switches are cheap.
[07:36] <jbailey> On Linux I'm not sure how it would work.
[07:36] <Lathiat> so speed improvement in L4 or speed improvement in linux?
[07:36] <jbailey> It was entirely in Linux.
[07:36] <Lathiat> hm
[07:36] <jbailey> The only research *i've* done on userspace drivers was in the context of Hurd and L4.
[07:37] <Lathiat> What are the benefits of having this in user space?
[07:37] <jbailey> IDE drivers?  Very little.  Drivers in general?  Fewer needs for custom kernels.
[07:37] <Lathiat> right
[07:37] <Lathiat> itd be nice to have things like ppp in userspace
[07:37] <Lathiat> so things like the mpppe kernel module dont need to exist
[07:38] <jbailey> Yup.
[07:38] <Lathiat> and then morton doesnt need to sneak them in with other stuff ;)
[07:39] <jbailey> I can see it being possible that the speed up was from some user->user buffer optimisation with userspace file system so that context switch penalty is negated.
[07:39] <jbailey> But that wouldn't do anything for streams/character devices.