/srv/irclogs.ubuntu.com/2010/02/23/#ubuntu-arm.txt

james_lMy apologies, but is anyone familiar with zubuntu here?04:26
persiajames_l: No apologies needed.  Omegamoon was hanging out in here for a while some time back, but we never managed to get the Zubuntu kernel into Ubuntu.04:30
persiaAnd only Ubuntu 9.04 can even run on that hardware: more recent versions require newer processors.04:31
persia(unless you're talking about https://launchpad.net/zubuntu , but I wouldn't expect this to be the right forum for that)04:31
james_lThat's a pity (I'm trying to get a working setup of either zubuntu or angstrom, and so far I've found little out about zubuntu, so I figured I'd ask)04:31
persiaWell, you *shoud* be able to install Ubuntu on your Zaurus, as long as you restrict yourself to 9.04.04:32
james_lWhy is 9.04 the latest?04:33
persiaBe warned that most of the default applications don't work very well at 640x480, so you may need to construct a special rootfs (the rootstock tool can help with this).04:33
persiaBecause the processor in those devices is old, and newer versions of Ubuntu require newer instruction sets that aren't supported.04:33
persiaSame issue with the SheevaPlug, SmartQ series, etc.04:33
persiaactually, I think the SmartQ series or Nokia n810 can run 9.10, but won't be able to run 10.04.04:34
* persia isn't quite sure04:34
persiaNote that *all* of these devices need custom kernels.04:34
james_lYeah, I've played with that, unfortunately for every kernel I've tried, it's dying after Kernel Panic: aiiee attempted to kill init.04:35
james_lEven ones that work (minus a few things) on angstrom, so I'm kind of at a loss on them.04:35
persiaIf you want something that supports the newer instruction sets, the Netwalker and Efika MX seem to be widely available.  LOts of people use the BeagleBoard as well, but it's low on RAM04:35
persia(and these also all each require custom kernels).04:35
persiaHrm.  I suspect there's something special in the kernel configs.  Take a look at the kernel config for one of the Ubuntu-supplied kernels for 9.04 (-versatile might be a good candidate or the nslu2 kernel), and see if there are any obvious differences.04:37
james_lCustom kernels aren't much of a problem, I just can't seem to find the kernel command line options that it wants, does ubuntu use the standard init, or something else?04:37
persiaUbuntu uses upstart04:37
james_lWhere is that stored?04:38
persiaHow do you mean?  Where can you download it, or where is it stored in the image?04:39
james_lWhere it's stored in the image04:39
james_lie, is it started as /sbin/init04:40
persiaIf you have an initrd or initramfs, it's going to be in there.  It provides /sbin/init, /sbin/initctl, /sbin/reboot, etc.04:40
persiaDefinitely /sbin/init04:40
james_lSo it's not actually stored on the rootfs image?04:41
persiaNo, it's in *both* places.  If you're using initrd or initramfs, you need to have a copy there to deal with early boot, and you'll also have a copy in the rootfs to deal with keeping the system running.04:42
james_lOk, sorry the way you mentioned it made it sound as if it was only in the initial ramdisk.04:43
persiaSorry.04:45
james_lDo you know if it requires a minimum kernel version?04:46
persiaI don't offhand, but it's usually best to try to use the kernel that was released with a given release of Ubuntu.  For 9.04 that was 2.6.2804:47
james_lThanks, I'll keep messing with it.04:48
persiaGood luck.  If you get it working, please share: having a clean procedure to install Ubuntu on the Zauri would be appreciated by many.04:49
james_lDo you know where Omegamoon went? I've tried to contact him about a few things, but alas have had no response.04:52
persiaI don't.  I haven't seen traffic from him in about a year in this channel.04:58
james_lAnyone else work on zubuntu?05:05
james_lpersia: Any idea why when it boots, it would be killing init?05:37
james_lHuzzah, got it to not kill init.06:54
james_lOn the other hand the touchscreen doesn't work06:55
james_lOoops06:55
persiaThat's just a driver thing :)  Nice work.  What was the magic change?06:57
james_lNot much, I'm using the 1.0 from http://www.oesf.org/forum/index.php?showtopic=26510&st=0 on a tosa07:05
james_ltosa = SL-600007:05
james_lI know the kernel works for TS, and I think I should have fixed that.07:08
james_lor not07:10
persiaOh, that should work.  The 2.0 stuff seems suspicious to me because I *know* that large chunks of the archive were recompiled for an instruction set not available on those devices.07:12
james_lWhy was that, and does ubuntu have a build system capable of rebuilding them as armv5 (or even armv4)07:16
persiaIf you want ARMv4, I'll strongly recommend Debian armel.  The team there has been maintaining support for this ISA, and does a great job.07:17
persiaIts potentially possible to attempt to rebuild everything that has changed since 9.04 with a toolchain that supports ARMv5, but that's 1) going to take months, and 2) likely to require a significant number of build machines and a separate archive.07:18
persiaI know it *can* be done, because some people recompiled 7.04, 7.10, and 8.04 for armel even before it was properly supported.07:18
james_lI've got two SL-5500 which are strongarm (armv4 (l?)), and a SL-6000 which is armv5 (something or other)07:19
james_lmonths?07:19
persiaBut I expect that such an effort would take 4-8 months.07:19
persiaYeah, there are thousands of packages involved.07:19
james_lWhy is ubuntu's build system so slow compared to others?07:19
persiaI'm not sure to what you're comparing it, but I suspect the answer is that it's all native builds.07:19
persiaI think our builds are faster than Debians.07:19
persiaI'm not sure about other places.07:20
persiaBut to complete the answer to your question, ARMv7 was the initial target for Ubuntu armel, but it was impossible to get hardware at the time the project started.  As hardware has become available, each release has gotten closer to the target.  I'm hoping that we'll have enough hardware to be able to have a sane, stable, supported platform for buildds with our October release, but the path to get there has been confusing.07:22
james_lWell, I've done cross compiling with a number of things, gentoo, angstrom, and even some debian and while each is probably far smaller than ALL the packages, it hasn't been that bad. Maybe a week for any of those.07:22
persiaYeah.  Recompiling for a target rootfs is usually at most a tenth the size of recompiling the archive in terms of package count.07:23
persiaBut there's a couple other issues involved, as follows:07:24
persia1) Not all packages in Ubuntu are set up for correct cross-compilation07:24
persia2) Build-time test-suites run under cross-compilation don't tend to generate the expected results07:24
persia3) The combination of native-built and cross-compiled binaries is poorly tested, and it's unknown whether there may be ABI changes.07:25
persiaSince working around 1) involves 20,000 patches, we chose native compilation.07:25
persiaAnd because of 3) we discourage cross-compilation because we don't know how it would affect things.07:26
persia2) is a bit more of a happy accident, but would have likely been dealt with if we had to patch for 1)07:26
ynezzd'oh http://pastebin.ca/180650007:41
james_lTo be honest, that's a bit disappointing, considering that debian had the best CC years ago07:42
persiaWell, it depends on the package.  Anything that cross-compiles well in Debian likely cross-compiles well in Ubuntu.07:44
persiaBut the vast majority of packages have never been tested.07:44
persia*so*, if you're just looking to do something small, you could rebootstrap with cross-compilation.07:44
persiaBut if you want the entirety of the archive, it's not necessarily safe.07:44
loolynezz: I saw that double free once, but it's usually not fatal08:27
loolynezz: Try with MALLOC_CHECK_=008:27
ynezzok, building again08:34
ogralool, usually not fatal ? it usually kills apt during package install :)08:51
ogralool, i think we should target that one for release and get doko on the case if possible08:52
ograynezz, hey, the screenshow in your new bug looks different from the old one :)08:57
ynezzit's same URL, isn't it? :)08:58
ograynezz, --serial ttyS2 .... vs console=ttyAMA0,115200 ;) you didnt build your image with the right serial device08:58
ograrootstock needs to use --serial ttyAMA0 in this case08:59
ericmogra, what did imx51 do to enable the serial console at boot by the way?08:59
ericmogra, by modifying flash-kernel?09:00
ograericm, adding console=ttymxc0,115200 to the cmdline09:00
ericmogra, I mean not in boot loader09:00
ograno, we dont touch flash-kernel, the default cmdline is set durign image build09:00
ericmogra, but the default will only be used when no cmdline is passed09:00
ograright09:00
ograas always09:00
ericmand u-boot will read the boot.scr scripts on disk to setup the command line, which is configured by flash-kernel09:01
ynezzogra: ok, rebuilding with correct --serial09:01
ericmso for dove - I guess to modify flash-kernel will be a better approach09:01
ograericm, no, we dont want to have a console option by default in any of the images09:01
ograwe only added that for the dev release and will drop it in one of the betas09:02
ograthe cmdline on arm should not differ from the default in any other images09:02
ericmogra, I'm going to close bug 45273709:02
ograthe boot.scr for dove is created by the image build scripts on the cdimage server09:02
ubot4ericm: Bug 452737 on http://launchpad.net/bugs/452737 is private09:02
ograericm, hmm, while its a bit late to enable that for lucid it might be helpful for L+1 until beta09:04
ericmogra, ok - got you09:04
ograericm, but essentially it is fixed btw09:04
loolynezz: note that you need to set MALLOC_CHECK_ in the installer script which gets copied by rootstock into the VM09:05
ograsee my last comment09:05
ericmogra, right - if it is only wanted a way to enable that instead of every time09:05
ogralool, i'll set that by default in my next upload(risking that we lose duplicate copunt for the bug)09:05
ograericm, yes, the liveimage specifically got that serialtty= option in casper to create the console09:06
ograericm, but it will only come up after boot for the login prompt, what i suspect Li wants on that bug is kernel bootmessages :)09:06
ericmogra, exactly what he was talking about09:07
ograright, so he needs to set console= on the images boot.scr ... its not that hard to do from an x86 system on the mounted USB key/SD card09:07
ogra(manually before booting)09:08
ericmogra, see - so I'll close that one09:11
ograyeah, but add proper comments :)09:11
=== hrw|gone is now known as hrw
hrwmorning09:13
ynezzlool: something like that? http://paste.ubuntu.com/382126/09:15
ograynezz, yes09:16
ograi'll add the same to rootstock as well09:17
ynezzogra: I've rebuild that karmic image with this http://paste.ubuntu.com/382129/09:20
ynezzogra: and it's same, that serial device had no effect on it09:20
ogratry without the --kernel-image parameter09:22
ograthats actually only for the beagel board09:22
ograbetter apt-get install linux-versatile after first boot manually :)09:22
hrwbtw - rootstack under Debian/sid + DIST=lucid hangs after extracting last package09:23
ograthough you should get a login prompt09:23
ograhrw on first stage ?09:23
ogranote that rootstock isnt really written for debian and i dont even test it in that context09:24
ynezzKernel image must be specified09:24
ograynezz, no09:24
ogranot in rootstock09:24
ogra--kernel-image is an addition for the beagleboard and it extracts a kernel directly, circumventing the package system09:25
ogra(because beagle has no kernel in the archive yet)09:25
ynezzah09:25
hrwogra: http://hrw.pastebin.ca/180655909:26
ynezzhm, had the same issue09:27
ograhrw, yeah, problem with the difference of our qemu packages i think09:27
ynezzit's missing qemu-arm-static09:27
ogradebian doesnt have qemu-arm-static in any stable release ... and in testing its named differently09:27
hrwok09:27
ogra(if it even entered testing already, not sure, might still be unstable)09:28
ograynezz, thats why we have packages for rootstock ;) they install the needed deps09:29
ynezzthere was no qemu-arm-static in 9.0409:30
ograright09:30
ograthats why the 9.04 package uses a VM for that stage09:30
ynezzah, ok09:31
=== ogra changed the topic of #ubuntu-arm to: Ubuntu ARM Discussion & Development | https://wiki.ubuntu.com/ARM | Want to Submit a Bug? https://bugs.launchpad.net/ubuntu/+filebug | Debian ARMel TODO: http://wiki.debian.org/ArmEabiTodo | Build a rootfs from scratch: https://wiki.ubuntu.com/ARM/RootfsFromScratch | got package build issues in lucid ? see https://wiki.ubuntu.com/ARM/Thumb2 | no, we do not cross compile
* ogra thinks it became a FAQ so makes sense to have it in the topic :)09:33
hrwogra: which HW is used for native builds?09:34
ograimx5109:34
hrwcpu speed? memsize?09:34
ogra800MHz, 512MB09:35
hrwwhere are times when 64MB 400MHz was good machine ;D09:36
ograheh that was when we were young :) ... back then09:37
hrwI still have slower devices on desk09:37
hrwogra: rootstock should have option to just generate images. as all is done in temporary dir it is harder to find where is image which was generated/used/failing09:38
* ogra too09:38
ograhrw, thats done with the --keepimage or --notarball options09:39
ograapart from that both, the gtk GUI and the script tell you wheer the image was stored at the end of the build (the GUI also has an option to select the target dir)09:40
hrwdpkg-divert: cannot open diversions: No such file or directory09:41
hrweggonlea: Second stage build in Virtual Machine failed !09:41
hrweggonlea: Please see the log to see what went wrong.09:41
hrwian_brasil: Cleaning up...09:41
hrw~curse irssi tabcompletion09:41
hrwian_brasil, eggonlea: sorry09:41
ogralool, so i'm considering to change the image fs from ext2, what would you prefer, ext3 or 4 ?09:41
ograhrw, heh09:41
hrwext2 is fine for images09:41
ograhrw, no, it causes fsck to kick in if the clock is off09:41
ograext3 should be common on all kernels nowadays09:42
hrwso go for ext3 and disable fsck by tune2fs09:42
ograi wont screw around ion the rootfs :)09:42
ograso no tune2fs09:43
ynezzogra, lool: little change in error message :) http://paste.ubuntu.com/382146/09:43
ograrootstock is supposed to build as identical installs as possible to a normal ubuntu09:43
ograwhich is why the --user and --passwd options are not used by default anymore in the latest commits for example09:44
ynezzogra: I've rebuilt that karmic image with http://paste.ubuntu.com/382148/09:46
ynezzogra: but it's same09:46
ograthats weird, it should work09:47
ogrado you see dmesg output on boot ? and do you see ttyAMA0 in there ?09:48
ynezzyes I see dmesg09:48
ynezzthat screenshot was just the last part09:48
ograyou should be able to scroll back and there should be some message about ttyAMA009:49
ynezzand I see ttyAMA0 there09:49
ogravery weird09:49
ynezz1,2,3 too09:49
ograthen it should work09:49
ynezzthe problem is with devtmpfs I think09:49
ograno09:49
ogradevtmpfs is in all ubuntu kernels for lucid09:50
ynezzit get's mounted and is missing /dev/pts /dev/shm etc.09:50
ograthat should be handled by udev09:50
* ogra goes for some breakfast09:52
* hrw goes though rootstock and removes all removals09:53
hrw~hail 25Mbps download09:53
hrwhmm. rootstack generated rootfs thinks that only dpkg is installed ;(10:11
loologra_cmpc: Please use ext4 for images of karmic+10:25
looland if you care, ext3 for << karmic10:25
persiaWhy, particularly?10:26
loolynezz: that's "interesting"10:27
loolynezz: Do you reproduce with "ldconfig"?10:27
loolpersia: are you asking me?10:27
persialool: Yes, and specifically about your filesystem selection suggestions.10:28
loolynezz: Yes, that might be the issue, but mountall should cope with that I think10:28
persiaI don't disagree: rather I'm curious why one ought select those.10:28
loolpersia: because ext4 was the default fs starting with karmic installs10:28
loolAnd I think all Ubuntu subprojects should be consistent here10:29
loolJust like we patched vmbuilder to do this as well for instance10:29
persiaThis makes lots of sense.  Thanks for the explanation.10:30
ynezzlool: how do you mean that 'ldconfig' ? mount that image and run it manualy using qemu-arm-static?10:30
loolynezz: boot with init=/bin/sh10:30
ynezzlool: sorry, you're talking about that apt-get segfault?10:31
loolynezz: Yes10:35
loolynezz: I htink it's ldconfig10:35
ynezzlool: since it crashed I need to create qemu image manualy and copy the unfinished fs from /tmp/ right?10:39
loolynezz: Sorry, I'm not sure about that part10:39
loolProbably there's a leftowver .img somewhere, unless it cleans it up on crash10:39
ynezzit cleans it10:40
ynezzok I'll rebuild without that cleanup task and try it10:41
* lool &10:41
ynezzah it's in /tmp also, good10:42
ynezzlool: if I run /sbin/ldconfig nothing happens10:51
ynezzlool: running it with --verbose it do something10:53
loolynezz: try apt-get install vim ... as in the log instead?11:08
loolynezz: It would be great to have a smaller testcase than running rootstock11:08
loolI saw that once myself, but couldn't trigger it again afterwards11:09
* ogra doesnt see it at all under lucid 11:09
ograand i'm doing about ten testbuilds with rootstock per day currently11:09
ograthough most of the time only minimal with no seeds11:10
ynezzogra: would you mind trying this 'sudo project-rootstock/rootstock --fqdn ubuntu --login ubuntu --password ubuntu --imagesize 2G --seed wget,vim,network-manager --dist lucid --serial ttyAMA0'11:11
ynezz?11:11
ynezzit does happen to me when there's network-manager in the seed...11:11
ograi'll try it in the next gui test later today11:12
ynezzok, thanks11:12
ynezzwithout seed it's ok11:12
ogra##################### REMINDER: mobile team meeting in #ubuntu-meeting in 5 min ######################12:56
loologra: Did you look at where rootstock would pull lucid kernels from?15:20
ograi didnt change the code you know15:20
loologra: You might want to check 43179015:21
loollp #43179015:21
ubot4Launchpad bug 431790 in debian-installer (Ubuntu) (and 1 other project) "debian-installer images aren't signed in the archive (affects: 1)" [Undecided,New] https://launchpad.net/bugs/43179015:21
ograbut its not active yet either15:21
ograoh, thanks15:22
ogralooks good, what key would be used, the normal archive key ?15:23
loolI guess15:25
ogralool, so it seems i can reproduce the issue with --seed wget,vim,network-manager here16:48
ograi get the same segfault16:48
loologra: Ok great, can you investigate?16:53
loolStarting by a bug report etc.16:53
* lool is about to leave for the day in 20 minutes or so16:53
* ogra too16:55
ograbut i'll put it on my list16:55
=== dmart is now known as Guest81189
=== Guest81189 is now known as dmart_
kikolrg, ping when you're around!18:43
ogralool, the segfault does only happen with the non archive kernel ... i cant reproduce it with the archive kernel here, i guess i'll switch on the rootstock code to pull the deb with the next commit :)19:34
asacok uploaded libv4l with workaround for arm gcc issue19:50
asacoff19:50
ograyay19:50
* ogra is off too19:50
=== dl9pf_ is now known as dl9pf
loologra: Could you comment on whether you intend to use d-i kernel in https://bugs.launchpad.net/bugs/431790 ?21:14
ubot4Launchpad bug 431790 in debian-installer (Ubuntu) (and 1 other project) "debian-installer images aren't signed in the archive (affects: 1)" [Undecided,New]21:15
lool~[6~[6~[6~[6~[6~[6~[6~[6~[6~[6~[6~[6~[6~[6~[6~[6~[6~[6~[6~[6~[6~[6~[6~[6~[6~[6~[6~[6~[6~[6~[6~[6~[6~[6~[6~[6~[6~/c21:18

Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!