=== prpplague is now known as prpplague_afk
sijiHow to do the screen alignment (LCD /touch 7") in Ubuntu07:29
sijiMy ubuntu natty on beagle board needs some left aligment07:30
sijiI have tried to modify /sys/devices/omapdss/display0/timings , but not happening07:30
sijipersia, there?07:59
persiasiji, Yeah, but a bit busy just now.08:00
persiaJust ask your question: maybe someone else has an answer.08:00
sijiok persia carry on08:00
sijiI have already put the question here :)08:01
persiaAh, I thought it might be something else.  I don't know the answer to that one anyway.08:01
lilsteviehowdy persia08:12
persiaHey lilstevie08:14
lilsteviepersia: with that source package the config needs to be edited08:15
lilsteviekernel config in its current form will gimp networking (I forgot about it)08:15
persiaDid you sort the rest of it as well?  I should have time to review it again in a couple hours.08:16
lilsteviewhat was wrong with the rest of it?08:18
persialilstevie, I can check my logs, but I remember package names, consolidation between two models, changelog.08:20
lilstevieah yeah08:21
lilstevieI remember niw08:22
lilsteviealso had problems with using a ppa for hw enablement, I found out standard ppa accounts don't have access to armel builders :p08:22
persiaYet another reason why the archive is superior to PPAs.08:24
sundar_hi all08:25
lilstevieit was more a stopgap solution until things make it into backports08:25
sundar_i am not sure if i am on the right channel to ask this question. I am on an embedded arm board running linux. I would like to open a virtual console on ctrl+alt+f1. any idea how i can achieve this?08:26
persiaBackports is two testers away once it hits the archives.08:26
lilstevieoh really?08:26
persiasundar_, If you're running some flavour of Ubuntu, it ought just work.  if not, I'm unsure if we can help you.08:26
lilsteviebtw I have the firmware issue taken care of :) injection done at install time08:27
lilsteviefrom android image08:27
persialilstevie, Yep.  Just needs a backports request, and two people to confirm the posted package works.  Then a backporter checks for sanity, and does the backport.08:27
lilstevieok well how do we get 2 testers :) cause I have about 250 users08:27
sundar_persia, thanks for quick response. I am running a custom compiled linux kernel and cutom built rootfs08:27
lilstevieand this kernel is newer than what they are currently running08:28
persialilstevie, We get the kernel in the archive, we prepare a backports bug with instructions on how to get the kernel, and we hope 1% of your users are willing to comment on the backports bug.08:29
lilsteviealso most of the ac100 stuff in flash kernel will work for this device08:29
persiaYou'll want to prepare a patch that lets a single flash-kernel script work for both devices.08:29
sundar_persia, if i were to do it on a ubuntu based board, how can i do it?08:29
ogra_already existing :)08:29
lilsteviepersia: ok awesome :) well I am sure I can get a few08:29
lilstevieogra_: what is already existing?08:30
persiasundar_, It would just work.  You can compare your X config and VC config to Ubuntu's, but I wouldn't know how.08:30
ogra_juliank has merged my ac100 stuff in the "flash-kernel-next" tree08:30
ogra_in a way that it handles all android based devices08:30
* persia goes back to try to finish stuff and be honest about the "couple hours" above08:30
ogra_well, everything that can use abootimg images08:30
lilstevieogra_: I only need 1 thing for the tf08:30
lilsteviemy SOS and LNX kernels are available, and depending on config (chosen at install time) depends where the primary is08:31
lilsteviebut on the pure linux cfg LNX boots normal and SOS boots in single user mode08:31
lilstevieogra_: is flash-kernel-next available anywhere?08:36
ogra_yes, lool has a branch on the debian git server somewhere08:36
ogra_its not used anywhere yet08:36
ogra_i.e. not packaged08:37
ogra_but it will be the new flash-kernel (note that i made up the -next)08:37
lilsteviewhen should that be out?08:37
lilsteviethis kern has some huge improvements over the one that is currently in use :p08:38
lilsteviefaster boot time, sound, improved handling of the keyboard08:38
sijipersia, it's solved :)08:39
ogra_lilstevie, not sure if it will ever hit armel in debian08:44
ogra_there are to many old arches in it that would need very very heavy testing08:45
ogra_i suspect we will see it in debian armhf before where you dont have to retain that much backwards compatibility08:45
ogra_but that a decision of debian08:46
lilsteviewhat about for us?08:46
ogra_i personally would like to switch to the next gen version next release08:46
ogra_i wanted to do it this time but got to much other stuff on my plate this round08:46
lilstevieah I see08:50
lilsteviejust would be nice to have an easier solution while trying to get my tf kernel into backports08:51
lilsteviepersia, I am working on fixing up those things now09:10
loollilstevie: it's in the official flash-kernel.git09:50
lilstevielool, ok cool, just wondering whether we will see support in natty09:52
LoqusHas anyone got any experience using a Gumstix platform, with 11.04, and 802.1q VLANs?09:55
lilsteviepersia, uploading the new source now'09:55
loollilstevie: *natty*?  no way09:57
loollilstevie: I mean, you could work on backports if you like, but this is a critical piece of infrastructure and there are major changes in there09:58
loolit's not that it's huge, but you wouldn't backport e.g. grub 2 to a grub 1-based ubuntu release that easily09:58
loolLoqus: I do have gumstix but didn't try VLANs, they don't work>09:58
loolthey don't work?09:59
LoqusWell, if I use Angstrom:09:59
Loqusthe vanilla supplied distro, the package manager doesn't heven have the module in its feeds.10:00
Loqus...under my desktop ubuntu environment I would get the module with "apt-get install vlan", then modprobe support into the kernel "modprobe 8021q", the use "vconfig add eth0 11" to add a virtual LAN adaptor on the Virtual LAN ID of 11.10:00
loolok, that should work with an Ubuntu armel install too10:01
LoqusWhen I created a 11.04 rootfs with rootstock, I had no option but to use the 2.6.36 kernel from bitbake because it (I'm told) isn't up to 2.6.38 yet10:01
lilstevielool, ok, the only reason I was wondering is cause I am working on bringing my kernel to backports for natty for the transformer10:01
loollilstevie: It might indeed be a better idea to base of tip and backport that; just dont expect it in official natty-updates10:02
loolA natty base is not in itself a problem, I just wanted to underline that the stable update policy doesn't leave room for intrusive changes like a flash-kernel change10:03
loollilstevie: I mean git10:03
looltip of git10:03
lilstevielool,  so you have no problems with getting it into backports then :)10:04
loolIt's unlikely that someone finds the time to confirm that there is no regression with any of the supported platforms in a backported package10:05
loolI suspect you would be masking support for some platforms in the backport10:05
loolI have no objection to a PPA   ;-)10:05
lilstevieI don't have an armel supporting PPA10:06
loolit's arch:all10:06
loolor is it10:06
looloh no it's not10:06
loolbut it's actually arch:all  :-)10:06
ogra_lool, do you see a chance for a flash-kernel-ng package in debian ?10:06
loolit's only present on certain architectures, but it only contains arch: all data10:06
lilstevieok so it is still a script rather than a binary10:07
lilstevienext generation?10:07
ogra_well, so we can keep the old one around10:07
ogra_-ng or -next or -new10:07
loologra_: just don't sync it if you dont want the new one?10:07
ogra_lool, the new one is in debian ?10:07
ogra_thats what i mean10:07
ogra_i know its used in armhf10:07
looloh god, I really need to go test it and upload it10:08
loolI'll JFDI10:08
ogra_well, do you see a chance to get it tested on all arches soon ?10:08
lilsteviewhen is ubuntu transitioning to hard float10:08
ogra_ask infinity ;)10:08
loolI can't ever test it on all platforms, but I can upload it to experimental and send a call for testing afterwards10:08
ogra_he is working on it this week10:08
ogra_lool, awesome10:09
lilstevieheh cool10:09
loolTBH, I took responsibility for this rewrite and I have not secured the time to actually make it happen in Debain10:09
loolso I am really late in this Debian work of mine10:09
ogra_lool, well, i want to switch in P in any case10:09
lilsteviewould be cool to see how much of a difference there is performance wise10:09
loologra_: Ideally, we'd switch in oneiric to have less intrusive changes in oneiric+110:09
ogra_and i will ignore debian if they arent on the new version10:09
loollilstevie: performance wise?10:10
loollilstevie: oh armhf10:10
loollilstevie: in Ubuntu, you'll see little difference10:10
lilstevielool,  hard float10:10
loolbut in Debian it's a huge gap10:10
lilsteviedebian will though10:10
lilsteviehf is a huge jump over armv5t10:10
lilstevieoh shoot10:11
lilsteviethat far behind10:11
loolYeah, I can't really support people still caring for ARMv410:11
lilsteviearmv6 is the minimum I would support but even then10:12
suihkulokkiARMv5 has a wide install base, and ARMv4t misses only few instructions from ARMv510:15
suihkulokkiCLZ, PLD being ones usually bumped into10:15
lilstevieI wish tegra had NEON10:16
suihkulokkiPeople have strange ideas that incredimental ARMvX versions give revolutional performance increases10:17
suihkulokkiI guess that is successful marketing from ARM =)10:17
lilsteviethere are some things that do have performance increases10:18
lilsteviet2 being one of them10:18
suihkulokkibut if you have an O(n^n) performance problem you still have it in T2 =)10:19
lilstevieaside from things like that though t2+NEON is faster than the equiv for armv610:20
suihkulokkiNEON is good, but you essentially need to some handwritten code take advantage of it10:21
lilstevieby equiv I mean the same program compiled without t2 and NEON10:21
lilsteviearm-darwin-gcc takes good advantage of it10:22
suihkulokkijust compiling generic C code wit neon is not going to be huge performance win10:22
suihkulokkichances that gcc will autovectorize is unlikely10:22
lilsteviepersia, when you are about ping10:34
=== prpplague_afk is now known as prpplague^2
persialilstevie, Sorry: that took much longer than I thought it would.13:34
lilstevieheh thats cool13:35
lilsteviesame location as last time13:35
lilsteviepersia: that is what we discussed yes?13:40
persiaI was hoping for just "transformer", but this might work: we can see what the archive-admins say if I don't find anything else.13:41
lilstevietransformer is fine13:41
persiaYeah, but not worth rebuilding if this can pass.13:41
lilstevieif needed13:41
lilstevieheh :)13:42
persiaIf there's something else to fix, then it makes sense to do both at once.13:42
persiaMight be worth a versioned recommends on flash-kernel, for the version that supports the transformer.13:42
persiaMind you, this just complicates backports, so it may not be worthwhile.13:43
persiaYou still have "TBD" as the board identifier.  Dunno if that ought be "transformer", or if you don't care.13:43
persiaEasily fixed in an update, and not important though.13:43
persiaThere's some other irregularities in the descriptions, like recommending installation of the "linux-tegra" metapackage, which doesn't exist (and won't).13:44
lilstevieok, well that was just from building13:45
lilstevieto the instructions13:45
persiaYeah, the instructions need extension: it leaves a few things messy still :(13:46
persiaBut that's all just textual: I haven't found anything functional yet.13:47
ogra_mahmoh, yo, did you recently say you plan to use elevator=noop on server or was that deadline ?13:47
mahmohogra_: I've been testing with deadline13:47
lilsteviepersia: :)13:47
ogra_i just ran into kernel bug 15426 here13:47
ubot2Launchpad bug 15426 in kdebase "kdesu (dup-of: 15001)" [Medium,Invalid] https://launchpad.net/bugs/1542613:47
ubot2Launchpad bug 15001 in kdebase "Administrator mode not working" [Critical,Fix released] https://launchpad.net/bugs/1500113:47
ubot2bugzilla.kernel.org bug 15426 in VFS "Running many copies of bonnie++ on different filesystems seems to deadlock in sync" [Normal,New]13:48
ogra_which causes bug 62487713:48
ubot2Launchpad bug 624877 in linux "INFO: task dpkg:23317 blocked for more than 120 seconds." [Medium,Confirmed] https://launchpad.net/bugs/62487713:48
ogra_seems the only scheduler i dont get that issue is cfq here13:48
mahmohogra_: that vfs layer bug looks familiar (really familiar in fact) but cfq may just reduce your likelihood of hitting it13:51
ogra_yeah, that could indeed be13:52
ogra_well, keep an eye open for it in your server testing ;)13:52
mahmohogra_: if it is the same bug it affects ext3/4 and should be fixed in two months or so13:52
ogra_i wonder if we should actually switch to cfq on the desktop images13:53
mahmohogra_: I've already hit it in another project, that's why it's getting fixed ;)13:53
ogra_seems many people hit it already according to the different open bugs :)13:53
mahmohogra_: I thought the desktop image is cfq already?  the x86 image is so.13:53
ogra_not armel13:53
mahmohyeah, it's a bad problem13:54
ogra_we use SD cards13:54
mahmohinteresting, so what's the benefit of noop vs. cfq on an sd card?13:54
ogra_with rootfs on SD noop is what you want13:54
ogra_or deadline13:54
ogra_the reading and writing is different on MMC than on HDD13:54
mahmohwrite out everything at once, avoid thrashing on the SD?13:55
ogra_leave the caching to the HW, to let it do its wear leveling13:55
ogra_we also adjust commit times etc13:55
ogra_which means you sould always properly shut down ;)13:56
mahmohogra_: so for the arm server kernel (whatever that is), SD won't really be a good root option, that's why I'm pushing for at least a different kernel command line add (elevator=deadline, preempt=0) if not a different kernel all together13:58
ogra_deadline is fine for Sd as well13:58
ogra_noop is better but deadline is ok, we should just set that cmdline on all server builds13:58
mahmohthe arm server kernel should look more like the x86 server kernel than the arm preinstalled image kernel that runs off of sd13:58
persialilstevie, I don't see anything else from source inspection.  I've started a build, but I expect to be asleep before it finishes.  I'll let you know if I find anything from the build in the morning.13:59
ogra_unless you can convince the kernel team to actually roll a -server binary indeed :)13:59
lilsteviepersia: how long it usually take?13:59
persiamahmoh, So, the "ARM Server" image is a preinstall that runs off SD :)13:59
persialilstevie, I'm guessing 2-3 hours, but maybe even 4.14:00
ogra_well, that will likely change to alternate at some point14:00
mahmohthey're the one's rolling the kernels?14:00
ogra_kernel team ? yeah14:00
mahmohpersia: I'm just looking for a server kernel that can be installed via net-install14:00
persiaogra_, Even so, for some targets it ends up still being installed to SD.  Depends on the device.14:00
ogra_linux and linux-ti-omap414:00
ogra_persia, sure14:00
lilsteviepersia: oh, one of the normal build systems14:00
persiamahmoh, Or apt-get, sure.  To support net-install probably needs some fiddling, but nothing too outrageous.14:01
ogra_we only have the preinstalled server images atm because it was easier to achieve and needs a lot less time for QA14:01
persialilstevie, Yeah, just one of my boards that I use for building stuff.  I don't have any magic :p14:01
mahmohtasksel ubuntu-server should install a server kernel or apt-get it sep. yes14:01
ogra_i dont expect it to stay like that forever :)14:01
mahmohthe sooner the better, I have a bug already but no action, :(14:01
lilsteviepersia: my machine takes half an our but that is cross compiling14:01
persiaI expect server to stay preinstalled until there is hardware that doesn't need preinstall available.14:01
ogra_which is likely the next release :)14:02
persialilstevie, Yeah, well, that's not quite the same as what the buildds will do :)14:02
lilsteviepersia: hehe14:02
persiaogra_, With luck, but I won't hold my breath.14:02
lilsteviepersia: well I will be unavailable most of the day tomorrow, uni day14:03
ogra_no, thats also bad for smoking14:03
ogra_(holding your breath i mean)14:03
persialilstevie, No worries.  We're not tight on a deadline.  Ping me when you get back.14:03
persiaogra_, See, I figure developing software for hardware that is merely theoretical is *good* for smoking: significantly increased chance of getting magic blue smoke.14:04
ogra_sure, as long as you dont hold your breath until the HW exists at least :)14:05
lilsteviepersia: well I will be here probably from 4 or 514:05
persialilstevie, In that case, I'll probably get back a bit later than you (it's only +9 here), and catch you when I do.14:05
lilstevieheh no problems :)14:06
persiaogra_, I'm not certain that holding one's breath correlates with release of magic blue smoke, although out of perversity, attempting to inhale magic blue smoke may potentially extend component lifespan.14:06
ogra_GrueMaster, can you make a note to monitor dmesg for "hung task" messages (like in bug 624877) in your next dist-upgrade test ?14:06
ubot2Launchpad bug 624877 in linux "INFO: task dpkg:23317 blocked for more than 120 seconds." [Medium,Confirmed] https://launchpad.net/bugs/62487714:06
ogra_i want to know if that also happens on panda14:06
ogra_(i see it a lot on ac100 which uses largely the same kernel config as omap4 nowadays)14:07
ogra_if thats the case we really need to change the default scheduler14:08
GrueMasterogra_: I have seen it during Natty.  I also see it when running IO tests on oneiric.14:09
ogra_thats bad :)14:11
lilsteviepersia: so what is so special about this build?14:11
persialilstevie, Nothing: it's just building for oneiric/armel on oneiric/armel, in an environment as close to that of the buildds as I know how to generate.14:12
persiaThe idea being to catch anything before it is submitted, as this makes it less likely the archive admins will reject it with prejudice.14:14
persiaKernels sponsored by me are likely to get a special double-check, as the last time I uploaded a kernel, it FTBFS, which was a bit embarassing.14:14
lilsteviewell I noticed one issue with this tree14:15
lilsteviemake mrproper breaks shit14:16
lilstevieI have no idea why14:16
lilstevieor even how14:16
persia!ohmy > lilstevie14:16
ubot2lilstevie, please see my private message14:16
persiaHrm?  What's "make mrproper" supposed to do?14:16
lilsteviemakes the entire tree "virgin"14:17
persiaAh, yeah, a pristine tree is kinda useful.14:17
lilsteviebut it breaks it14:17
ogra_whats embarrasing about FTBFS kernels ?14:18
lilstevieogra_: I guess sponsoring it14:18
persiaogra_, Archive Admins suggesting I need to do more build testing, mostly.14:19
persiaYeah, well.  Some of us understand the concept called "shame", and use it to improve our work.14:20
ogra_well, its not that i dont feel shame ... but i'm not embarrased by FTBFS of something that takes huge efforts to build at home ...14:22
infinityHow Japanese of you.14:22
ogra_---compared to just uploading it and let it fail14:22
persiaogra_, Um, except folk might be using it, and -meta skew is annoying, etc.14:23
ogra_pfft ... users14:23
persiaProbably builds faster at home than on the buildds anyway, given the current buildd HW.14:23
persiaSome folk subscribe to the philosophy that users are a priority, even :p14:23
GrueMastermahmoh: Any luck getting ipv6 tests going?15:01
mahmohGrueMaster: haven't tried ipv6, unsure if our local net supports it15:50
GrueMastermahmoh: You were going to work on getting the tahi.org testsuite converted.16:03
mahmohGrueMaster: I was going to take a look yes, still on todo ;)16:03
GrueMasterI can run it.16:03
mahmohGrueMaster: the u-boot bugs are bogging me down16:03
=== daurnima1or is now known as daurn
* ogra_ glares at the bottom of https://launchpad.net/project-rootstock/trunk/+ubuntupkg16:19
ogra_why does ramana own all these packages ?16:19
GrueMasterbug 819899 bug 81990016:21
ubot2Launchpad bug 819899 in livecd-rootfs "package pools need to correctly parse override info to create tasks" [Medium,In progress] https://launchpad.net/bugs/81989916:21
ubot2Launchpad bug 819900 in livecd-rootfs "package pool implementation needs to update apt" [Medium,In progress] https://launchpad.net/bugs/81990016:21
* GrueMaster thanks ubot2.16:23
persiaogra_, Report a bug: might be a DB offset issue.16:27
ogra_well, i asked in #launchpad but got no answer16:27
ogra_its likely a bug16:27
persiaPoke abently if you want an answer.16:29
persialilstevie, So, I didn't end up happily going to sleep whist this compiled.  Firstly, there's calls to ccache inserted in various places, which breaks.  After removing those, there's hard dependencies on cross-compilation stuff, which makes it not build.16:30
persiaI'll still try to catch you tomorrow, but if you're up late, or you check backscroll ...16:30
LoqusIs anyone using a Tobi Duo16:39
GrueMasterLoqus: ???16:42
* ogra_ guesses GrueMaster's wife uses a tobin from time to time, but whats a tobi ?16:42
* GrueMaster doesn't get used often enough.16:43
ogra_an addon board for the gumstix16:43
ogra_i dont think we even have many gumstix users around here (i would be happy to be wrong indeed)16:44
* GrueMaster has been called many things, but never a board.16:44
ogra_you would have to slightly work on your shape to be called a board i guess16:44
GrueMaster(although some ask how the "twins" are doing, either in reference to my jolly size, or moobs - unsure which).16:45
LoqusI was asking in #gumstix, but they are a bunch of akfkers ;)16:45
* GrueMaster reverts back to image testing before the level of conversation degrades any further. :P16:45
* ogra_ waits for persia 16:45
Loqus...yeah, it'sa duel NIC expansion board for the Gumstix... I've moved over from Angstrom to Ubuntu for Arm.16:46
persiaLoqus, Are the NICs recognised, or is the kernel missing the driver?16:47
ogra_i had a ohmy for less !16:47
LoqusI don't know... in the wisdom of whoever designed this board, it doesn't have any other console interface, except the NICs. I was asking in case someone knew that i had to modify the board support driver in some way,16:48
* persia is tired and doesn't get the reference, so the offender is free. Others are welcome to hint about channel guidelines if they like.16:49
LoqusI *think* that one of them is requesting a DHCP address, but I'm not sure on that one...16:49
persiaLoqus, Do you have a DHCP server?16:49
LoqusSSH on the address which is is given, but not connecting16:49
persiaIf not, try setting an address in /etc/network/interfaces (the interfaces(5) manpage explains the format).16:50
persiaDo you have openssh-server installed?16:50
Loqussetting it statically?16:50
LoqusIf I pop the board of the Tobi Duo (duel NIC) and put it on a single NIC Tbi, it boots just fine16:50
Loqus...the IP is differen because of the different MAC, but aside from that no changes to the FS or kernel16:51
ogra_is your power supply able to cope with the additional power needs from the board ?16:51
Loqus...I'll admit that asking here was a long shot - just in case someone knew that there was an extra module or somethign I had to include.16:52
Loqus^Yes, PSU is rated to about 20 amps :)16:52
hank_I am looking for the source code to build natty Narwhal (Ubuntu 11.04) for omap4. Can someone help me to locate the source?16:58
persiahank_, So, we typically distribute binary packages, and adjust them one-by-one, using something like `apt-get source ${PACKAGE}` or `bzr branch lp:ubuntu/${PACKAGE}` to get the source for a specific package.17:00
persiaMost of the sources for packages in the images is also available as CD images, from http://cdimage.ubuntu.com/source/current/source/17:01
persiaBut that's not usually the easiest way to sort things.17:01
lilsteviewho highlighted me before? client glitched out17:01
persialilstevie, I did.  I didn't get to bed, the compile broke first because it was using ccache, and then, after fixing that, because it was looking for cross-compiling stuff.17:02
persiaBack to you :)17:02
lilstevieheh :)17:02
lilstevieok so no ccache cause you don't use it :)17:03
lilsteviewhat cross_compile stuff though17:03
persiahank_, So, what's your ultimate goal?  Would a prebuilt environment work for you?17:03
hank_persia, I want to cross compile into omap4, currently, downloaded 10.10 soruce, wish to do 11.04. Does that mean I only need 11.04 kernel?17:04
ogra_what exactly do you want to cross compile and why17:04
persia1) What's your host environment?  2) Once you're done cross-compiling, what do you expect to get?17:05
persiahank_, And no, we don't want the details of your code, if you're working on a secret project: just a general idea so we can give the best advice for your needs.17:06
hank_ogra_, and persia, Host is a OMAP4 on panda board, prefer to do on x-86 platform because of performance. Finally boot into pandaboard.17:06
persiahank_, Are you running Ubuntu on the x86 host?  If so, which release?17:07
hank_persia, I am on 10.10 release on x-86 host.17:07
ogra_and what do you want to compile ? we offer binaries for all the source you can download already, do you want to change something existing or do you want to cross build some new code that isnt in ubuntu ?17:07
hank_ogra_,Cross compile with my specific configuration for the embedded.17:09
persiahank_, So, we usually say "We don't do embedded", but our definition is devices with <256M and no MMU, which may not match yours.17:09
ogra_given it is a panda ...17:10
persiaWe don't actually have any facility to rebuild the entire rootfs with different configuration defaults: instead our software stack is organised into packages, each of which would need customisation and rebuild if you need to change things at that level.17:10
persiaWe strive to have all the packages be as flexible as possible, so often you can just have a settings package that can change some or all of the defaults, but this is documented in a per-package manner, rather than the all-at-once model that one might have for e.g. LTIB17:11
hank_persia, and ogra_, Omap4 manufacturer TI has sent me to chat on this channel. Does this mean I should get back to TI with my issue?17:12
persiahank_, Not necessarily :)17:13
ogra_i dont think av500 works for TI :)17:13
persiaWe're happy to help get you a working rootfs for your pandaboard.17:13
persiaBut we use a different model of development, which makes it a bit different from traditional "embedded" environments.17:13
persiahank_, So, if you want to deploy something desktop-like, we'll recommend you start from the Ubuntu netbook image for 11.04.  Then use the armel cross-compiler distributed with Ubuntu 10.10 to rebuild any packages you want to change.  Install the rebuilt packages into the target, and you ought see the behaviour change.17:16
persiaIf you want something smaller, you might start from the headless image, which is a significantly smaller base.17:17
hank_Thanks, persia. As far as I am concern, I wish to start out from a set of source that has been developed to work with panda board.17:17
ogra_hank_, like that https://wiki.ubuntu.com/ARM/OMAP ?17:18
hank_persia, I have tried both notebook 11.04 and 10.10 pre-build images. 11.04 is more perferrable than 10.10 because I can restart from cold with 11.04.17:19
persiahank_, Well, OK.  Check the manifests from the image download page, and get the source for the packages concerned, from any of the ISOs I mentioned earlier, archive.ubuntu.com/ubuntu/pool/*/*/*/${PACKAGE}.dsc, or ports.ubuntu.com (same path as archive).17:19
persiaWe don't either create or build any packages specifically for the pandaboard, but rather attempt to ensure that all the packages in Ubuntu also work with the pandaboard.17:20
ogra_apart from kernel and bootloader indeed :)17:20
persiaSo, for the pandaboard, 11.04 has much more complete porting, and seems to work more reliably.  Last I heard, TI hadn't released some of the binary drivers for it, so if you need certain functions, you may need 10.10 (someone please correct me if I'm mistaken)17:21
ogra_the binary drivers are fine, the video codecs arent17:21
persiaOh, heh.  Yeah, bootloader/kernel are device specific (although we consider this a bug)17:21
ogra_in natty that is17:21
persiaogra_, Thanks for the clarification.17:21
GrueMasterThe only thing I know of that doesn't work ootb on natty is BlueTooth, but there is some workaround for that which requires pulling a small program from gitorious.17:22
GrueMasterAnd video codecs, of course.17:22
hank_Thanks, persia and ogra_. I will view the desc content and try to get the source. Appreciate it.17:22
ogra_come back if you have more questions :)17:23
persiahank_, Good luck.  If you change your mind about rebuilding *everything*, let us know, and we'd be happy to help you with more specific goals.17:23
GrueMasterAnyone have any ideas for testing ubuntu-core (beyond what I have documented at http://testcases.qa.ubuntu.com/Install/ARM/Core )?17:23
* ogra_ goes reading17:23
persiaGrueMaster, I've been thinking about that much of the day, and came to the conclusion it might be untestable.17:24
ogra_well, that looks like a good attempt17:24
hank_Thanks for the insights, persia.17:24
ogra_you at least test that apt works this way17:24
persiaUnless I'm confused, the only packages of interest in core that aren't tested by the process of creating all the other images are mountall, ifupdown, and upstart.17:24
GrueMasterWe should have some tests for the core apps (not necessarily for this release, but in the future).17:25
persiaAnd these can only be tested if it's booted, which this image doesn't support.17:25
ogra_well, what are the core apps ? thats all super low level17:25
ogra_what do you want to test ? ls ?17:25
GrueMasterI have no idea.  That is why I am asking.17:26
persiaGrueMaster, Also, do you think it's worth expanding the instructions at https://wiki.ubuntu.com/Core ?17:26
ogra_i think apt is a good usecase17:26
persiaapt is the *key* usecase, but I can't imagine that not working if we ended up with any of the other images.17:26
ogra_there are probably more than that, but wasnt the purpose of core to be exactly enough OS to be able to run apt to install additional stuff ?17:27
GrueMasterI was thinking of more of a unified test.  Apt is a good one, because if it breaks here, it usually indicates a missing dependency or something.17:28
ogra_GrueMaster, oh, typo ... resolf should be resolv17:29
GrueMasterKind of like with the desktop images.  If X isn't installed, the gui won't boot.17:29
ogra_and i'm not sure you still need to copy it actually17:29
ogra_i havent had to do that for ages in a fresh debootstrapped chroot ... though live-build might remove it17:30
ogra_oh, and the bind mounting will definitely break if your host is intel17:30
GrueMasterYou need it if you are behind a gateway.17:30
persiaI don't actually seem to have /etc/resolv.conf on any of the systems in which I have an open termina.17:31
ogra_you could add a test case using qemu-static17:31
GrueMasterThe bind mount won't break if you follow the instructions:  To test this image on an already running armv7 system, ...17:31
persiaAh, rather, it's not delivered.  Nevermind.17:31
GrueMasterSomeone who knows how to use qemu can add that.17:31
ogra_so people can test on intel17:31
ogra_its one cp more17:31
ogra_and indeed having the package installed17:31
persiaogra_, That pollutes the rootfs ...17:31
ogra_persia, indeed it does, but it helps getting testers17:32
ogra_since its not hw bound17:32
persiaI suppose.17:32
GrueMasterGetting testers is not much of a concern if we don't have any defined tests.17:32
ogra_and you get tests of qemu as a sideffect :)17:33
persiaUm, I suppose.17:36
persiaogra_, I'm reminded: you might be interested in Debian bug #63538517:38
ubot2Debian bug 635385 in qemu-user-static "qemu-user-static should upgrade static libraries in qemu-debootstrap created chroots" [Minor,Open] http://bugs.debian.org/63538517:38
mahmohdavidm: did you see https://bugs.launchpad.net/linux-linaro/+bug/709245/comments/2518:14
ubot2Ubuntu bug 709245 in linux-ti-omap4 "panda: USB disk IO slow" [High,Confirmed]18:14
infinitymahmoh: Ahh, nice.  That confirms the suspicion I had earlier today that it might be spinlock related.18:20
GrueMasterWould be interesting if someone could reproduce this on a non-omap4 Coretex-A9 system.18:23
infinityThey can.18:23
infinityIn fact, every report we've had of it seems to be dual core A9s, hence the earlier suspicions about SMP spinlocks.18:24
GrueMasterSo, is it the SMP code, or deeper (say Cortex-A9 core).18:24
infinityWell, that combined with the timing revelation.18:24
mahmohinfinity: so how can we cap the spinlock easily without nosmp?18:24
infinitymahmoh: We need to trace the actual bug in play here.18:24
GrueMasterIsn't there a different scheduler in the kernel that we could try?18:24
infinitymahmoh: This isn't something we want to work around, this needs fixing.  Broken SMP is, uhm.  Bad.18:24
mahmohinfinity: so besides nosmp, there's not easy way without starting tracing?18:25
infinitymahmoh: nosmp ends up making every spinlock a no-op.18:25
mahmohinfinity: yeah, I',m not suggesting w-o, suggesting narrowing the problem and validating18:25
infinitymahmoh: There are people at both Linaro and RedHat working on this.  Now that we're off the wrong "looks like a USB issue" track and onto a spinlock witch hunt, I suspect it'll go well.18:26
mahmohinfinity: fair enough18:26
infinityThe SMP scheduler is actually pretty well-audited and sane code.  So, if some naive ARM commiter broke it, it should jump out at someone.18:27
prpplagueGrueMaster: i haven't performed the tests myself, but i have been told from several sources that it has been replicated on the Snowball and tegra2 platforms18:39
GrueMasterprpplague: So I've heard.  My only question is if it is indeed kernel SMP code or possibly Cortex-A9 design flaw.  I have seen similar issues with SMP and bus timing eons ago on the P6 (circa 1996).18:41
prpplagueGrueMaster: indeed18:41
GrueMasterUnfortunately, I don't have the means to test at that level.18:41
GrueMasterAnd it wouldn't be the first time we have seen a hw bug expressed in code.18:43
prpplagueGrueMaster: i'm sending an internal TI email now18:50
prpplagueGrueMaster: i've got some time scheduled with the hardcore smp guys tomorrow afternoon to do some debugging19:09
GrueMaster Cool.19:10
martynGrueMaster: I just walked in on the conversaton -- what's going on with SMP on Cortex A9?19:19
martynI have multiple non-OMAP systems here.  (Versatile express, tegra2, Calxeda, STMicro )19:20
GrueMastermartyn: See bug 709245  Especially later comments.19:26
ubot2Launchpad bug 709245 in linux-ti-omap4 "panda: USB disk IO slow" [High,Confirmed] https://launchpad.net/bugs/70924519:26
martynAHHHhh.. we're also seeing slow IO19:27
martynin SATA19:27
martyn(and other tests)19:27
martynsometimes -very- slow19:27
robherI don't think nosmp affects the spinlock code itself. On x86, I believe spinlock code is patched at boot time for UP vs. SMP.19:54
robherBut on ARM, there is no run-time patching of spinlocks. Only virt_to_phys is patched.19:55
prpplaguemartyn: i have a test shell script20:34
prpplaguemartyn: you just need a usb hd to test with20:34
prpplaguemartyn: which platform?20:35
prpplagueGrueMaster: interesting enough, i just did some tests on two x86 machines. the transfer rates from the usb hd increase while pinging localhost20:35
GrueMasterHrm.  That's not right.20:36
prpplagueGrueMaster: http://pastebin.pandaboard.org/index.php/view/5288767120:37
prpplagueGrueMaster: mostly on the low end size20:38
prpplagueGrueMaster: smp - http://pastebin.pandaboard.org/index.php/view/324141920:38
GrueMasterDid you run one of the tests from TheSeven?20:39
prpplagueGrueMaster: no smp - http://pastebin.pandaboard.org/index.php/view/213833820:39
GrueMasterWhoa!!  Same board?  UP vs SMP?20:39
prpplagueGrueMaster: based on TheSeven 's stuff20:40
GrueMasterI'll try to reproduce that here once I get a spare minute.  I have a spare Core2Duo that is currently acting as a dust collector.20:41
TheSevenprpplague: so these pastes are x86 data?20:49
TheSevenhm, if the first one is x86 and the other two ones pandaboard it makes sense20:50
TheSevenand i'd call that increase on x86 with ping a fluke... it's way smaller than the fluctuations between the repeated measurements20:51
martynprpplague : Okay, back at terminal20:55
TheSevenprpplague: my shell script has a small bug btw, you need to kill the perl background tasks with SIGTERM, not SIGINT, or they won't go away20:55
martynhit me with the script20:55
martynI can try it on a VExpress20:55
TheSevenrun that as root and pass some storage device node as the only argument20:57
TheSevensudo ./speedtest.sh /dev/sda is what I'm running on my board20:57
martynah, I"ll have to install perl21:03
TheSevenyou can probably skip the perl tests21:03
TheSeventhe nothing vs. ping difference is the biggest one21:04
TheSeventhe various perl commands are somewhere in between21:04
GrueMasterAnd the perl tests could fairly easily be converted to bash scripts (esp the while loop).21:07
TheSeveni used perl to get rid of possible process invocations that could affect the threading behavior in non-obvious ways21:07
martynokay, the machines in the lab are in use for a test, I'll run the script as soon as I can get access21:08
davidmmahmoh, yes saw that comment21:16
GrueMasterdavidm: It may possibly extend beyond arm.  More testing is being done.  See scrollback.21:17
davidmGrueMaster, I'm reading the scroll back now.21:19
davidmGrueMaster, in any case, it's becoming clear it's not USB, the symptoms are just easiest to see there21:21
TheSeventhe same thing happens to a lesser degree for SD card access on the pandaboard as well21:25
prpplagueTheSeven: i yanked the perl stuff from the script21:25
prpplaguehehe martyn quites about the time i get back to my console21:26
=== robbiew is now known as robbiew-afk

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