[06:13] <fabbione> morning
[06:49] <fabbione> Mithrandir: apt-cache search pistachio
[09:24] <Mithrandir> fabbione: no hits, yes
[09:32] <fabbione> Mithrandir: it's in debian.. but a very old version
[12:14] <fabbione> i just uploaded 1.2
[12:14] <fabbione> please wait to commit around that i need to do the baz dance
[12:14] <fabbione> and get some food before that :)
[01:40] <zul> hola
[01:41] <fabbione> hey zul
[01:41] <zul> hey fabbione how goes it
[01:41] <fabbione> the main baz repo seems to be fucked
[01:41] <zul> huh?
[01:41] <fabbione> we are investigating the problem
[01:41] <fabbione> there are some strange merge issues
[01:41] <zul> i didnt do it :)
[01:41] <fabbione> and i can't tag 1.2 upload
[01:41] <fabbione> i know :)
[01:41] <fabbione> it's a baz bug apparently
[01:42] <zul> bah...lets use cvs ;)
[01:42] <fabbione> i can take your word for it :)
[01:42] <zul> lol
[02:00] <fabbione> zul: i am almost done with the tagging :)
[02:00] <zul> ok im trying to track down something
[02:24] <fabbione> 1.3 is open :)
[02:25] <fabbione> have fun
[02:25] <fabbione> i am off for a bit
[02:41] <jbailey> Hmm.  depmod will have to be done at boot time to deal with module assembly. =(
[02:43] <jbailey> I know the kernel can cause modules to be loaded automatically.  Does it shell out to insmod or something, or does it rely on a hotplug event to get that right?
[02:44] <jbailey> I'm thinking things like sd_mod
[02:44] <zul> i think it relies on hotplug
[02:44] <zul> but im not 100% sure
[02:44] <jbailey> Cool.  Thanks, Chuck.
[02:46] <zul> np
[03:07] <fabbione> jbailey: depmod can run at initramfs creation time, but it's always a good idea to run it at boot
[03:07] <fabbione> and if can.. please stop using insmod and switch to modprobe
[03:07] <jbailey> This is simply for the modules that exist in the initramfs for hardware autodetection.
[03:08] <jbailey> The problem is that at build time I can't assume that I have all the modules that will eventually be present in there, so to get the modules.aliases file right, I need to run it on the bits in the initramfs
[03:08] <jbailey> Already stopped using insmod in my local tree.
[03:09] <fabbione> jbailey: ah ok.. i see
[03:32] <jbailey> Booyah!  She boots!
[03:32] <jbailey> The only modules I have to hardcode now are md, raid1, and sd_mod
[03:33] <fabbione> Total 2075 package(s)
[03:33] <fabbione> ^ sparc needs-build
[03:33] <fabbione> 24 hours ago they were 2200
[03:34] <jbailey> Hmm, I could drop md from that list, nice.
[04:07] <jbailey> Is there any way to tell which things in kernel/drivers/net are actually for hardware devices and which ones are just things for ppp/slip etc?
[04:12] <fabbione> not automatically
[04:13] <jbailey> 'k, thanks.
[04:13] <fabbione> ppp is hardware.. it gives you a device back
[04:13] <fabbione> even if it's "virtual"
[04:13] <jbailey> Yeah, I was hoping for a device tag or something.
[04:13] <jbailey> No biggy, they're pretty small, and I can always do a blacklist later.
[04:13] <jbailey> Or hey, perhaps we can support NFSRoot over PPP
[04:13] <jbailey> *puke*
[04:14] <fabbione> ahahhaa
[04:16] <jbailey> Lesse..  Reasonable root filesystems:  ext2, ext3, jfs, nfs, reiserfs, xfs.
[04:16] <jbailey> Would anyone try over afs, hfs or ufs?
[04:16] <zul> reiserfs4..hah!
[04:17] <jbailey> I don't see that in the Ubuntu kernel...
[04:18] <fabbione> ahaha
[04:23] <dilinger> jbailey: i'd like to get our root filesystems using afs, but i'm not sure if that's possible
[04:23] <dilinger> given afs has a global namespace that always has to be /afs/<cell>
[04:24] <dilinger> it would require some ugliness involving symlinking, for example, /usr into /afs/<cell>/root/usr or somethihng
[04:24] <dilinger> or bind mounting
[04:24] <dilinger> or something
[04:25] <jbailey> dilinger: Mind if I ignore that case for now?  We can always add the drivers in later.
[04:25] <jbailey> That sounds in the same class as root over NFSv4 =)
[04:26] <dilinger> does nfs4 employ some weird namespace?
[04:26] <jbailey> It uses strong auth like Kerberos as well. =)
[04:26] <dilinger> interesting
[04:27] <jbailey> dilinger: Eh, you might know the question chuck answered earlier.
[04:27] <jbailey> You mentioned before that SCSI can autoload things like sd_mod when it loads the appropriate driver.
[04:28] <lamont>   CC [M]   drivers/net/fealnx.o
[04:28] <lamont> drivers/net/fealnx.c:907:2: warning: #warning Processor architecture undefined!
[04:28] <jbailey> Does it shell out to insmod, or rely on a hotplug event?
[04:28] <lamont> fabbione: I think we can disable that driver for hpp
[04:28] <lamont> a
[04:29] <fabbione> lamont: go ahead :)
[04:29] <lamont> heh
[04:29] <fabbione> baz commit -s'Fix FTBFS on sparc64'
[04:31] <fabbione> lamont: nope.. it knows nothing about baz
[04:31] <fabbione> we can make another bot for it
[04:31] <jbailey> Wire it up to the commits bot?
[04:31] <fabbione> hmmm actually.. given that sparc32 is not supported.. i could make sparc64 the default and kill sparc
[04:32] <lamont> kernel-team@ubuntu.com--2005/kernel-debian--pre1,2--2.6.11.94
[04:32] <jbailey> It would probably be easy to write a module for CIA, and nice for baz projects to have it anyway.
[04:32] <lamont> that's the right version, yes?
[04:32] <fabbione> jbailey: no. i don't like to mix too much
[04:32] <fabbione> lamont: nope..
[04:32] <fabbione> that's been released today
[04:32] <fabbione> 1,3
[04:32] <lamont> you uploaded -1.2, ok
[04:33] <lamont> -1.2 went to main?
[04:33] <jbailey> fabbione: Why not?  There are enough people who look at the #commits channel, and it keeps stats and such.  Might just be interesting PR for Ubuntu.
[04:33] <lamont> s/went to/is headed for/
[04:34] <dilinger> jbailey: i don't think it calls out to insmod, i think it calls the internal stuff that does module loading
[04:34] <fabbione> jbailey: because i don't want another extra service to run on my own server.
[04:34] <dilinger> jbailey: map the driver into memory, call the init function, etc.  that's just a guess, though; i haven't looked at the module loading code
[04:34] <fabbione> jbailey: if somebody wants to set it up.. more than welcome :)
[04:35] <jbailey> dilinger: 'kay thanks.  I'll look and see what I can find.
[04:37] <fabbione> brb
[04:47] <jbailey> Ugh.  Now the network drivers are detected in the initramfs, which is correct, but hotplug on the boot system isn't triggered by the module loading, so the network doesn't get started. =(
[04:58] <lamont> jbailey: don't make us hurt you....
[04:58] <jbailey> What if I beg?
[06:18] <zul> hey...the live cds dont have a System.map do they?
[07:45] <zul> huh...where is the kernel-image for powerpc 2.6.12
[08:07] <fabbione> they should....
[08:08] <zul> not on the iso
[08:08] <fabbione> strange...
[08:09] <zul> chuck@homer:/media/cdrom$ find . -name "System.map" -print
[08:09] <zul> chuck@homer:/media/cdrom$
[08:09] <fabbione> we should ask Kamion if they get removed to save space
[08:10] <zul> i was trying to track down that amd64 problem to see what is causing it to crash
[08:16] <Micksa> alright, hibernate works
[08:16] <Micksa> that was easy
[08:16] <Micksa> heh, even my ssh sessions stayed running 8)
[08:36] <jbailey> mdz: Around?
[08:43] <mdz> jbailey: yep
[08:44] <jbailey> mdz: I have all the hardware autodetection working right, but it has the side effect of not letting hotplug generate a colplug event (since the module is already loaded).  The end result is that ifup is never called.
[08:45] <jbailey> mdz: I think the fix is in hotplug.  I can upload what I have for now if you'd like to test it, though (or post .deb's or whatnot)
[08:45] <jbailey> mdz: s/is in/needs to go in/
[08:45] <mdz> jbailey: go ahead and upload
[08:46] <mdz> we don't rely on a plug event to call ifup by default anyway, right?
[08:46] <mdz> we just do ifup -a at boot
[08:47] <jbailey> Hmm, it didn't on my box.
[08:47] <jbailey> I can look for that, though.
[08:53] <mdz> that's certainly how it works here
[09:00] <jbailey> mdz: If you're testing it locally you still might have to add ide-general or sd_mod, but all the actual hardware drivers are autodetected.  The kernel is supposed to autoload those, but I'm guessing that I don't have something quite right.
[09:01] <mdz> jbailey: the kernel should load sd_mod when needed, but ide-generic has never worked that way (unfortunately)
[09:02] <mdz> jbailey: wasn't it you who implemented the workaround in the hotplug package for that?
[09:03] <jbailey> I did, I haven't ported that to it yet.
[09:03] <jbailey> dilinger said that a recent rc version included the patch to ide to make it use the driver model.
[09:03] <jbailey> So it might do the autoloading on its own.
[09:03] <jbailey> In any event, my test system for this is sata, and sd_mod isn't being autoloaded, so I'm trying to see what I need to make that happen as well.
[09:03] <jbailey> The rest of the hacks are nice known things, so I can just toss them in easily enough.
[09:07] <jbailey> I didn't have auto eth0 in my interfaces file... =(
[09:11] <jbailey> mdz: btw, thanks for the fix to initramfs.
[10:11] <mdz> jbailey: no problem
[10:11] <mdz> jbailey: is busybox-cvs-initramfs waiting in queue/new or something?
[10:12] <jbailey> Filename: pool/universe/b/busybox-cvs/busybox-cvs-initramfs_20040623-1ubuntu18_powerpc.deb
[10:12] <mdz> ah, no, it's in universe
[10:12] <jbailey> I thought dependancies were automatically taken care of?
[10:13] <mdz> where "taken care of" means "show up in the anastacia output the next day", yes
[10:13] <jbailey> Ah. =)
[10:15] <mdz> jbailey: which MODULES choice should I used to get the network drivers included?
[10:15] <jbailey> It does it unconditionally at the moment.  Did it miss yours?
[10:15] <mdz> s/used/use/
[10:16] <mdz> no, I didn't look
[10:16] <mdz> I had MODULES=list, so I assumed it wouldn't add them
[10:16] <jbailey> That will be a valid assumption soon. =)
[10:16] <mdz> jbailey: does it get af_packet as well (for dhcp)?
[10:17] <jbailey> It appears not.
[10:17] <jbailey> (I'm in my initramfs atm)
[10:18] <mdz> my ethernet driver didn't get loaded
[10:18] <jbailey> Which driver?
[10:19] <mdz> oh, silly me; I installed busybox-cvs-initramfs and then forgot to go back and upgrade initramfs-tools
[10:19] <jbailey> =)
[10:19] <mdz> that's more like it; taking much longer to build the initramfs now :-)
[10:20] <jbailey> If you want to take a look inside it, I have a script locally that I use that basically does mkdir /tmp/initramfs; (cd /tmp/initramfs && zcat /boot/initramfs <cpio -i)
[10:20] <jbailey> I current save my initramfs as /boot/initramfs for testing.
[10:20] <mdz> jbailey: I get some error messages from invalid modprobe invocations, but it does work
[10:21] <jbailey> Yup, I haven't done noise reduction at all on this.
[10:21] <jbailey> Did you have to add af_packet?
[10:21] <mdz> I didn't take it out of mkinitramfs/modules
[10:22] <jbailey> 'k
[10:22] <mdz> I added it in the first place when ipconfig failed to work without it
[10:22] <mdz> so I assume it's still needed
[10:22] <jbailey> Yup, and it's definetly not getting pulled in atm.
[10:22] <mdz> it's probably a safe assumption that all nfs-root setups should have it
[10:25] <mdz> jbailey: hmm, taking nfs out of modules fails too
[10:25] <mdz> I end up with an initramfs which has the nfs module in it, but not its dependencies
[10:25] <mdz> also, the kernel-package hook doesn't seem to work
[10:25] <mdz> root@mizar:/usr/bin/X11 # dpkg-reconfigure linux-image-2.6.12-1-386
[10:25] <mdz> Cannot find /lib/modules//lib/modules/2.6.12-1-386
[10:25] <mdz> Failed to create initrd image.
[10:26] <mdz> looks like mkinitramfs syntax differs slightly from mkinitrd
[10:26] <jbailey> Hmm.  I'm copying whole directory trees at this point into the initramfs.  I thought I had most of what we needed for nfs.  I'll add more in.
[10:29] <mdz> jbailey: why does it work via /etc/mkinitramfs/modules, but not via the nfs magic?
[10:29] <mdz> do you resolve deps in one case and not the other
[10:29] <mdz> ?
[10:30] <jbailey> Right.
[10:30] <jbailey> In modules, I'm actually resolving dependancies.  In the other case, I'm using cp -a to get thigns from /lib/modules into the initramfs
[10:31] <jbailey> I can add specific things to resolve dependancies on, but I don't want to wind up with a list of drivers that we copy in - that just means we have to maintain it.
[10:32] <mdz> yep
[10:33] <mdz> jbailey: is the kernel-package thing simple to fix?  it'd simplify my life
[10:33] <jbailey> I'm rebooting into the working system to test it now.
[10:33] <jbailey> I'll also add the dependancy check for nfs and af_packet
[10:34] <mdz> cool
[10:34] <jbailey> And force those to modprobe in the nfs case.
[10:34] <jbailey> At that point will the modules file be empty on your nfs system?
[10:34] <mdz> af_packet and nfs should both autoload fine, no?
[10:35] <jbailey> There's nothing in there to autoload them right now.  If you list them in modules, they get modprobed.
[10:35] <jbailey> But in nfs mode, they should be modprobed too
[10:38] <mdz> oh, kmod doesn't work?
[10:41] <jbailey> Doesn't seem to, I haven't tracked that down.
[10:41] <jbailey> but sd_mod isn't being autoloaded either.
[10:41] <jbailey> I was asking dilinger about that earlier.
[11:32] <jbailey> mdz: Looks like 0.11 will make :33, it has your fixes.
[11:33] <mdz> jbailey: cool, thanks
[11:50] <jbailey> mdz: One of the things in the nfsroot spec currently is: Add ability to specify module loading from the kernel command line and additional module details (IRQ, IO Address, Port) to support older network cards.
[11:50] <jbailey> mdz the ltsp folks didn't think it was so necessary, since it's pretty rare to see non-pci devices (and our hardware detection won't find it atm anyway)
[11:51] <jbailey> I will hack it so that if module information is specified it will get used, but do you otherwise see any problem with dropping that?
[11:51] <jbailey> Either that, or I can also take a parameter for additional modules to load from the command line.