/srv/irclogs.ubuntu.com/2010/10/25/#ubuntu-kernel.txt

lucent1;2c/away06:26
lucent^^typingfail06:26
=== _LibertyZero is now known as LibertyZero
=== zul_ is now known as zul
Ober7hi guys, having14:16
Ober7hi guys, can any1 help me w/ this error? 14:18
Ober7drivers/dsp/syslink/multicore_ipc/gate.c: In function ‘gate_enter_system’:14:19
Ober7drivers/dsp/syslink/multicore_ipc/gate.c:38: error: implicit declaration of function ‘local_irq_save’14:19
Ober7thats it14:19
lagOber7: Where did you get your tree from?14:52
Ober7lag: ubuntu git tag Ubuntu-2.6.35-903.914:53
lagOkay, give me a moment14:54
Ober7thanks14:54
lagHave you changed anything in the config?14:57
lagOber7: --^14:59
lagCan you grep your config files for "CONFIG_TRACE_IRQFLAGS_SUPPORT"15:00
Ober7lag: i used localmodconfig, when i looked at it later with menuconfig the whole sys_linux was built-in 15:00
Ober7lag: CONFIG_TRACE_IRQFLAGS_SUPPORT=y15:01
lagOber7: That's a good start15:01
Ober7lag: nice15:03
lagOber7: Bear with me, I'm going to build that Tag15:07
lagOber7: Which platform are you trying to build for?15:10
Ober7lag: 68615:23
Ober7i686*15:24
=== akgraner` is now known as akgraner
Ober7lag, would appreciate your help thanks15:49
=== jdstrand_ is now known as jdstrand
lagOber7: Why are you building Ubuntu-2.6.35-903.9?16:11
lagOber7: What made you choose Ubuntu-2.6.35-903.9?16:17
Ober7lag i just like trying out the latest16:19
Ober7starting to think maybe this build is for something specific16:19
lagOber7: Correct16:19
lagOber7: Specifically, ARM16:19
lag:)16:19
lagBut it's not the lastest16:20
Ober7ahhh16:20
lagCheckout the lastest master branch16:20
lagThat is the lastest16:20
Ober7thanks lag16:20
lag:)16:20
lagnp16:20
lagThat's why we're here16:20
Ober7anyways i removed syslink form the config and it compiled fine16:20
Ober7im on it right now16:21
Ober7will rebuild the master16:21
lagYes, syslink is an ARM thinkg16:21
lagthing*16:21
lagTI specific 16:21
Ober7sorry for the trouble16:21
Ober7cheers16:21
lagS'ok16:21
jcrigby#ubuntu-uds-bonaire117:09
Ian_Cornewhat's the bonaire for?17:10
jk-Ian_Corne: the room name17:12
Ian_Corneok :)17:12
=== smb` is now known as smb
EdganDoes anyone know of a convenient way to un-Ubuntuize for patches a Ubuntu kernel package?22:17
hyperairhuh? un-ubuntuize patches?22:18
hyperairif you want a vanilla kernel, why don't you just download it from kernel.org?22:18
Edganhyperair: Remove all the ubuntu patches, but leave the build environment22:18
hyperairah22:18
hyperairthat sounds like it could be tough =p22:19
hyperaircheck debian/patches?22:19
hyperairif there's nothing there, then just copy debian/ onto a vanilla kernel22:19
Edganhyperair: What I really want is to take the linux-rt package and take out the ubuntu patches, making what Ubuntu calls a -realtime kernel.22:19
hyperaireh?22:19
Edganhyperair: I tried something like that22:19
hyperairwait, so you want a realtime kernel minus ubuntu patches?22:19
Edganhyperair: as a package, yes22:20
hyperairhmm why would you want to do that?22:20
Edganhyperair: There is one for 2.6.33, but I need 2.6.3122:20
* hyperair repeats. why would you want to do that?22:20
Edganhyperair: Because there is an issue with kernels with Ubuntu patches22:20
hyperairand that is?22:21
Edganhyperair: The error message is like info: task jbd2 312 blocked for more than 120 seconds.22:22
hyperair=\22:22
hyperaira hung task.22:22
Edganhyperair: I don't see it on Fedora kernels, or -realtime kernels. But I do see it on all Ubuntu kernels, even the latest development 2.6.36 kernels22:22
hyperairyou should report a bug22:23
hyperairand maybe bisect the patch out22:23
Edganhyperair: There are open bugs for issues like this. Mine might be slightly different, but this issue has been around since the beginning of Lucid, at least. They have been patching it, which makes it less frequent, but it still isn't gone.22:23
Edganhyperair: All it takes for me to reproduce it on a normal -generic kernel is dd a big file, and looping a apt-get remove ; apt-get install22:24
hyperairEdgan: https://bugzilla.kernel.org/show_bug.cgi?id=15370 <-- looks like fedora has it.22:24
ubot2bugzilla.kernel.org bug 15370 in LVM2/DM "task jbd2/dm-1-8:973 blocked for more than 120 seconds." [High,Resolved: invalid]22:24
Edganhyperair: That is fc12, but I don't see it on fc13, which is now on 2.6.3422:25
hyperairso basically you're seeing those i/o latencies22:25
Edganhyperair: I have also tested newer development Fedora 2.6.3622:25
hyperairi think you're seeing those i/o hangs where your load level temporarily shoots up to the 20s or 30s22:26
hyperairand suddenly everything gets swapped out, and swapped back in22:26
hyperairand so everything grinds to a halt.22:26
Edganhyperair: It doesn't shoot that high with dd and apt-get, but yes, in some cases we shoot it way higher than that.22:26
hyperairit's not due to an ubuntu patch imo22:26
hyperairi've been compiling my own vanilla kernels for some time now22:26
Edganhyperair: Well here is my evidence22:27
hyperairand until my recent RAM upgrade, i've had this issue on every kernel i've compiled22:27
hyperairsince karmic.22:27
hyperairwhich was approximately when i started compiling my own kernels22:27
Edganhyperair: 2.6.31-rt(official lucid realtime) has it, 2.6.32-25(official lucid) has it, 2.6.35-22(official maverick) has it, 2.6.36(development maverick) has it22:28
* hyperair shrugs22:28
Edganhyperair: 2.6.33-realtime(ppa lucid realtime) doesn't have it, 2.6.36(Fedora development) doesn't have it22:28
hyperairi'm just saying that i've seen the issue in vanilla kernels.22:28
Edganhyperair: I have seen this issue in the past with Fedora, but it went away long ago22:29
hyperairhmm22:29
Edganhyperair: I think there have also been multiple bugs that trigger that error22:29
hyperairapw: you seen this issue before?22:29
hyperairEdgan: are you saying that this 2.6.33-realtime kernel doesn't get the severe i/o hangs?22:30
hyperairat all?22:30
Edganhyperair: Right, though I have other issues with it, which make me want to go back to 2.6.31, but then the 120 second error comes back22:30
Edgan-realtime kernel - is based on the vanilla kernel source tree with Ingo Molnar maintained PREEMPT_RT patch applied to it. Also know as a hard real-time kernel. 22:31
hyperairwell that's really something.22:31
EdganThat is the description for -realtime22:31
Edganhyperair: I am the sysadmin at a robotics research company. We run the realtime kernel on our robots, and they have been crashing left and right. I have been working for over a week to find a truly stable kernel. It has been very frustrating.22:32
hyperairi see.22:33
hyperairi think you might have better luck digging around the git tree for ubuntu kernels22:34
Edganhyperair: My theory is that it is differences in the ext4 code. Red Hat writes more of the code, so they stay on top of it more. in something like 2.6.32-23, Ubuntu was like oh look we are missing 100+ ext4 patches, lets add those in. In this exact case I suspect it is either out dated or one-off Ubuntu patch.22:35
Edganhyperair: What doesn't help is that Ubuntu's kernel is one big monster patch, that isn't broken out into individual patches22:36
hyperairEdgan: you need to see the git tree.22:36
hyperairi believe that the big monster patch is generated from the git tree22:36
hyperairand that the actual maintenance of the separate patches happens within that git tree22:36
Edganhyperair: lame22:37
Edganhyperair: Any pointer page?22:37
hyperairEdgan: not so lame when you consider that there are potentially hundreds of those patches there, which without git's delta-compression can cause the package size to swell.22:37
hyperairEdgan: https://wiki.ubuntu.com/Kernel/Dev/KernelGitGuide22:39
Edganhyperair: Fedora's build environment is somewhat git-ized. I have to use it build non-debugging development kernels. But normal packages have all the patches broken out. In the spec file you can just comment them out one by one for a build.22:39
Edganhyperair: Plus patches are just text, which compresses well22:39
hyperairEdgan: one or two patches compress well. not hundreds.22:39
Edganhyperair: They end up as one big monster patch in the apt-get source results anyway. For 2.6.32-25, that is 2.1mb compressed.22:40
Edganhyperair: kernel tarball, 79mb22:41
hyperair1 monster patch = 1 set of context lines.22:41
hyperair100 patches = 100 sets of context lines22:41
hyperaircouple that with merges and whatnot, and you've got yourself a fine mess of patches that you don't know to apply in what order ^_^22:42
EdganIt would also be more sane if the debian* directories where outside the build tree, but that is Debian's fault22:42
Edganhyperair: That is what a spec file sorts out for you.22:42
hyperairthat is what quilt sorts out for you22:42
hyperairbut you need to generate the series file22:43
hyperairwhich is a set of patches in linear order22:43
hyperairwhen you have merges with conflicts, they're not so easy to generate patches for22:43
hyperairi don't believe the spec file handles all of that22:43
hyperairperhaps redhat just has less distro-specific patches22:43
hyperairor take great pains to make sure their patches are linear22:44
hyperairhave linear history, i mean22:44
hyperaireither way, i maintain that rpm is a joke of a package format and package manager, so *shrug*22:44
hyperairif you really want fedora kernels, then go ahead and use fedora.22:45
Edganhyperair: haha, I find dpkg far worse.22:45
hyperairthen let's agree to disagree22:45
Edganhyperair: I would but before me they settled on Ubuntu as their development platform22:45
hyperairi lack sleep and don't feel like continuing this disagreement.22:45
Edganhyperair: http://kernel.ubuntu.com/git I am not finding any realtime branches here22:46
hyperairwell redhat junkies will be redhat junkies. i remember the short period of time i administrated a centos server22:46
hyperairthat was a real nightmare.22:46
Edganhyperair: I did that for years, they have their problems, but I am finding Ubuntu far worse as a server.22:46
Edganhyperair: But I am willing to drop that topic22:47
Edganhyperair: realtime branch?22:47
hyperairEdgan: i like debian on my servers. but yes, let's drop it22:47
hyperairas for realtime branch, i don't know, i've never poked around22:47
hyperairwhat you could do, however, is wait for apw to come around and tell you things that kernelteam knows about22:48
hyperairi'm just a random non-kernel-hacker who compiles his own kernels and hangs around the channel.22:48
Edganhyperair: ok, I will look to ask him if he pokes his head up.22:48
hyperairif all else fails, what you can do is just go get ingo molnar's PREEMPT_RT tree and compile that straight22:49
hyperairmake-kpkg generates deb packages you can install22:49
Edganah, not bad22:49
hyperairbuildkernel='rm -f .version; KBUILD_BUILD_VERSION=0 AUTOBUILD=1 CONCURRENCY_LEVEL=${CONCURRENCY_LEVEL:-2} ionice -c 3 schedtool -D -e time make-kpkg --initrd --rootcmd=fakeroot --append-to-version=-hyper${KERNELREV:-1} kernel_{headers,image}'22:49
hyperairthat's my alias i use for building kernels22:49
hyperairSarvatt: what was the new and improved way of building kernels again?22:50
* hyperair still loves his make-kpkg22:50
hyperairi think there was a make deb-pkg target as well in the mainline tree, but never experimented with that22:51
hyperairwell then, my vision is hazy and the sun's about to rise, so off i go to bed.22:51

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