/srv/irclogs.ubuntu.com/2008/05/27/#ubuntu-kernel.txt

justdaveit was "current release" earlier this week, so I assume Hardy00:00
pwnguinyea00:00
justdavethe thing did have Linux on it as shipped, but all binaries, and no dev toolkit on it00:01
justdaveand a 2.2.18 kernel00:01
justdaveso I'm assuming swiping the drivers off the original software probably won't be much use00:01
justdavethe lack of a dev toolchain is the main reason I reloaded it00:01
justdavethe software that shipped on it was all java and I hate java anyway (not to mention it was all compiled class files anyway, so not easily hackable)00:02
pwnguinim not sure there's a big ubuntu embedded community00:02
pwnguinyou might seek out embeddian00:02
justdaveFigured it'd be more fun to do touchscreen stuff in python ;)00:02
pwnguinor start an ubuntu embedded community if you like something in particular to ubuntu00:02
pwnguinerr emdebian00:03
pwnguinif you're in germany, emdebian will be at fosdem in brussels00:03
pwnguinlike tomorrow00:04
pwnguiner no00:04
justdaveheh.  nope, USA.00:04
pwnguinthats right, its linuxtag thats in germany soon00:05
guijemontisn't there that moblin project ?00:11
guijemontbased on ubuntu afaik00:11
guijemontmoblin.org00:11
pwnguinthere is00:11
pwnguinbut00:11
pwnguinits basically a group for hire working on specific hardware and requirements00:12
justdavethat looks like it's mostly for modern embedded hardware :)00:12
justdavethis thing is like 7 years old00:12
guijemontyeah, I think they're mainly targetting intel Atom00:12
guijemontit's an intel sponsored project if I understand correctly00:13
justdavelooks like the Geode stuff was manufactured by several folks...  AMD claims all the drivers got upstreamed into the mainline kernel00:13
justdaveI found 4 kernel modules in Hardy that appear to deal with GPIO ports, but none of them seem to work on this thing00:14
pwnguinif you dont like emdebian, there's also the gumstix community00:14
pwnguinthey also deal with gpio ports00:15
justdaveemdebian might work, still reading. :)00:15
=== Ben1 is now known as BenC
justdavehaha, is this timing or what04:46
justdavejust updated the package list and had a geode video driver show up04:47
justdavemaybe I can do something better than VESA on this thing after all04:47
justdavealthough vesa isn't exactly working poorly on it, so I'm not too concerned :)04:48
=== asac_ is now known as asac
tseliotcan anyone of the kernel-team approve my subscription to the mailing list, please?11:30
SwishI was successful in compiling the kernel with make menuconfig followed by make-kpkg etc, but what's the right way to modify some .config options in the debian/config/i386/config file -then- build it the "ubuntu" way with the syntax:  AUTOBUILD=1 NOEXTRAS=1 fakeroot debian/rules custom-binary-FLAVOUR  ?11:59
Swishi should say, what's "a" right way to do that, considering the config files are split up and later concatenated, and if I just overwrite a config file in debian/config/ARCH, that doesn't work at all and when I initiate a compile, i get asked a ton of questions12:01
sorentseliot: You don't need approval to subscribe to the mailing list.12:10
sorentseliot: I certainly didn't, so I can't imagine why you would.12:10
tseliotsoren: is this the right link to the mailing list? https://lists.ubuntu.com/mailman/listinfo/kernel-team12:16
sorenLooks like it.12:17
tseliotsoren: I did it yesterday but I didn't receive the email to confirm my subscription12:17
tseliotI have just tried it now and it works12:17
tselioto_O12:18
emgentmorning12:22
tseliotemgent: moring12:24
amitkSwish: you can use the "debian/rules prepare-<flavour>' to do that12:35
Swishamitk, oooh, /me googles12:35
amitkSwish: e.g. debian/rules prepare-generic12:36
Swishokay, so that would create a .config file in debian/build/build-generic/ ... and then... I would edit that .config file in make menuconfig, then run some other script to re-split that new .config file into the multiple files in the debian/config/whatever dirs?12:38
emgentheya tseliot 12:38
amitkSwish: yes. We did talk about simplifying that a bit by directly allowing an interactive 'make menuconfig' and then split up the result..12:43
amitksplitconfig.pl in the scripts directory will help you split up the config now12:43
Swishah ha.  so splitconfig uses as input the debian/build/* dirs, and outputs back into the debian/config/* dirs?12:44
* Swish edits it12:44
amitkSwish: we'll accept patches to do this automagically ;)12:45
Swishhee hee ;)12:45
Swishthis will have to wait until my kernel compile is done.12:46
Swishthanks for the info!  :)12:47
alex_joniBenC: short question. is 2.6.24-18 going to be pushed soon?14:17
BenChow do you mean pushed?14:21
alex_joniI mean released14:21
alex_joniI saw that git has that ABI number14:22
BenCit's still being prepared by the security team I guess14:23
rtgBenC: I think the security upload will happen today. or as soon as kees comes online.14:37
BenCrtg: sounds good14:44
BenCrtg: FYI, current intrepid locks up on my 1420 laptop (using 64-bit kernel) whenever I close the lid14:44
BenCgetting ready to rebase to -rc4 to see if it fixes the problem14:44
BenCit's a strange lockup though.. the entire system locks up, except the screen is still on, and the mouse still moves...everything else is dead (no screen updates other than mouse)14:45
rtgBenC: I've been so busy with Hardy SRU's that I haven't even loaded Intrepid in several weeks.14:46
BenCrtg: Do you have any systems that duplicate the suspend/resume problem for testing ehci-hcd removal?14:46
rtgBenC: At least 2. Both of my XPS laptops14:46
rtgI've also fixed some Dell 640's using the same method, though I don't know if they were running -1714:47
BenCDo you have a way to test with a USB2.0 device attached to see if it causes some issues?14:48
rtgBenC: I think the case that we need to test is when your root device is USB. I have only one external USB drive, and I'm loath to trash it. Maybe I have a flash stick thats big enough.14:50
rtgI wonder if the classmate is USB?14:50
BenCit is14:51
BenCWhat does removing ehci-hcd do when you have a usb2.0 device connected?14:51
rtgBenC: thats what I'm trying to find out. It for sure dumps bluetooth devices correctly.14:52
BenCdoes it revert it to usb1.0 transparently, fail to be removed, or does it remove the devices completely?14:52
rtgBenC: I think it removes them, Looking at dmesg implies that for bluetooth.14:52
BenCAnd if the device is in use, I suspect it will fail to unload then (e.g. usb root fs)14:53
rtgBenC: right, that was the case I was asking mjg59 about.14:54
BenCrtg: shouldn't need to make it rootfs...should be able to test with it mounted, and open files on it14:54
mjg59No, there's no locking to prevent you from removing it14:55
BenCmjg59: then does it play nice and drop devices to usb1.0?14:55
mjg59Not as far as I know14:55
mjg59You'd need to trigger the 1.0 controller to power down the port and power it up again to get the handshaking redone14:56
BenCso in-use devices just get removed...that's pretty shitty...and makes it dangerous to add to pm-utils for removal14:56
mjg59There's no way you can drop from 2.0 to 1.0 without a simulated reconnection14:56
mjg59The protocol just doesn't allow it, AFAIK14:57
rtgBenC: correct. I think its the reason some file systems have gotten corrupted.14:57
BenCCan reconnections be done for in-use devices without loosing state?14:57
BenCin-kernel state I mean14:57
BenC*losing14:57
mjg59You could possibly abuse the USB persist stuff14:58
mjg59But I suspect it'd be unhappy about it being on a different controller14:58
* BenC is totally into abusing things14:58
amitkBenC: you want to checkout the USB persist patch that I applied to get classmate suspend working? :)14:59
rtgamitk: how does persist work?14:59
amitkrtg: by working around the issue of the USB controller powering down on suspend. So the data structures are maintained on suspend. On resume, the kernel checks the persistence flag and prevents 'reset' of the device15:03
rtgamitk: does that work for the general case, e.g., for more then just the classmate?15:03
amitkor actually does just a port reset..15:04
amitkrtg: works across all devices, it is now default in 2.6.26, no kconfig option AFAIK15:04
rtgamitk: hmm.15:05
amitkrtg: ohh.. and each device has a persist file in sysfs where you tell whether it is persistent or not15:09
BenCrtg: I finally re-installed my laptop using 64-bit15:10
BenCnetscape+flash seems to be working perfectly fine15:11
BenCs/netscape/firefox/15:11
* BenC jumped back about 10 years for a moment there15:11
amitkBenC: wait till npviewer.bin starts crashing on you...15:13
BenCamitk: youtube+myspace+random-junk for 4 days and no crash yet15:15
rtgBenC: I've been running 64 bit for a couple months on both XPS's with no issues.15:15
rtgamitk: I think the noviwer stuff must have gotten fixed. I've not had any problems with it sine about alpha 415:16
rtg*npviwer15:16
BenCforced push of rebase to -rc4 for intrepid15:17
amitkBenC: did you do a sync first?15:17
BenChopefully no one committed anything in the last 20 minutes :)15:17
amitkwe just cleaned up the tree..15:17
amitkBenC: did you pull after friday?15:18
BenCI pulled this morning15:19
* amitk sighs with relief...15:20
rtgBenC: yeah, amitk cleaned up all of my Intrepid merge history. made it a lot nicer to look at with gitk. it might have also caught a couple of merge problems.15:23
BenCubuntu-intrepid-ports is also being pushed15:23
BenCHopefully I can get an upload of that today some time15:23
BenCand be done with it15:24
rtgBenC: I gotta get working on LRM and finish packaging the new broadcom driver.15:24
BenCWho's handling the dkms nvidia/fglrx packaging?15:26
rtgBenC: mario  and alberto milone.15:27
rtgthe envy guy15:27
amitkrtg: did I tell you alberto volunteered to handle madwifi too..15:28
rtgamitk: handle it how?15:28
amitkrtg: DKMS'ify it15:28
rtgamitk: maybe we wanna get some experience with DKMS before we jump into the deep end.15:29
rtgwe haven't exactly turned it loose on a million desktop users yet.15:30
amitkagreed...15:30
BenCWhen does he think he will have something in the archive?15:33
rtgBenC: I think they were shooting for sometime this week.15:33
rtgBenC: at least that was Mario's goal. I may distract him with Hardy issues, though.15:33
sorenIf a vendor/device ID combo is listed twice in modules.pcimap, which gets chosen? First match? Last match? Undefined?15:35
rtgsoren: its sorted in modules.dep.15:36
BenCsoren: undefined though15:36
BenCthere's no guarantee which will get loaded15:36
sorenSo the list of modules that support the device is generated, and the one of those to appear first in modules.dep is chosen?15:37
sorenOr something to the same effect?15:37
rtgunless the driver is blacklisted, its the first one in modules.dep that gets loaded.15:37
BenCthat's generally what happens, but don't depend on it15:37
rtgBenC: well, LBM _does_ depend on it.15:37
BenCrtg: No, it doesn't15:37
rtgthen I have an imperfect understanding.15:38
BenCrtg: modprobe will prefer things in updates/ before all others15:39
BenCmodules.*map defines the module name, not the location it is loaded from15:39
rtgBenC: depmod  will prefer things in updates/ before all others15:39
BenCrtg: But that affects modules.dep, but modules.pcimap15:42
BenC*not15:42
BenCmodules.pcimap is just a listing, modules.dep is where it gets the load location/dependencies from15:43
rtgBenC: modules.*map merely defines the module(s) name for a matching ID, right?15:43
BenCright15:43
BenCwhich means you cannot rely on the ordering in that file15:44
BenCmodules.dep is a different story15:44
rtgBenC: right, order in modules.*map is unsorted.15:44
rtgbut, moduels.dep is sorted according to depmod search rules.15:44
BenCright15:44
BenCbut soren was asking about the ordering in modules.pcimap, which is unreliable :)15:45
rtgso, modprobe uses a straightforward algorithm to choose which module to load.15:45
rtgoh, maybe I misread sorens question.15:45
rtgsoren: back to my original statement :) Unless blacklisted, the first module in modules.dep that is referenced in a modules.*map is the one to get loaded.15:47
Xsss4hellIs the 2.6.25.4 Kernel out jet?16:04
Xsss4hellfor Ubuntu16:04
Xsss4hellI need it, because of ATI gfx cards and Catalys 8.516:05
rtgXsss4hell: Are you talking abut Intrepid?16:05
Xsss4hellno16:06
Xsss4hellI have Ubuntu Hardy..16:06
Xsss4hellJust wanted to update the kernel..16:06
rtgXsss4hell: Hardy will continue to be based on a 2.6.24 kernel.16:07
Xsss4hellWLAN doesn't work either with the 2.6.17 Kernel.. Fritz!Wlan16:07
Xsss4hell2.6.24.17 I mean16:07
rtgXsss4hell: is there a bug report on it somewhere that I can read?16:08
Xsss4hellhttps://bugs.launchpad.net/ubuntu/+source/linux-meta/+bug/22535316:11
rtgXsss4hell: ok, I'll look in a bit. on a conf call for awhile.16:11
Xsss4hellBut I doesn work with 2.6.24.16-17 as far as I've tested it. WPA works on 2.6.24.16 but only with ndiswrapper and windows drivers.. Think about it, this type of wlan stick is the most common in Germany16:12
Xsss4hellWPA2 doesn't work on any of these kernels16:12
Xsss4hellrtg thank you ;)16:28
BenCrtg: -rc4 seems to have fixed the lockup I had when closing the laptop lid17:06
rtgBenC: you must have seen something on LKML that lead you in that direction?17:07
BenCrtg: my 1420 works out-of-the-box with stock kernel (no lum/lbm/lrm)17:07
rtgBenC: it ought to. I turned on all of ALSA. 17:07
rtgin the kernel, that is.17:08
rtgBenC: have you tried the microphone?17:08
BenCI did have to copy firmware17:09
BenCfor iwl17:10
BenCrtg: Let me check...17:10
BenCHmm, no built-in mic I guess...I'll need to find one to hook up17:11
BenCit seems to be working, just recording dead air tho17:11
rtgBenC: you should have iwl firmware in LUM. there is a LUM package on the kernel PPA. I'll start an upload with the -rc4 rebase.17:12
johanbrThe 1420 I have does have a built-in mic.17:12
johanbrMaybe it's different depending on your options at purchase time.17:13
rtgjohanbr: I have one as well, I just have to unearth it. Some are digital mics, some are analog.17:14
johanbrWhat do you mean by "unearth" ?17:14
rtgjohanbr: I have a bunch of laptops from Dell. the 1420 is somewhere in the pile.17:15
johanbrAhh, okay. I thought you were talking about the mic. :)17:15
rtgjohanbr: the 1420 built-in mic was difficult to get working. I really hope there is no regression there.17:16
BenCjohanbr: if you get the webcam option, it has built-in mic17:20
BenCI wonder if my other 1420 has the webcam17:21
BenCAh, I have an E1420 that has built-in webcam+mic17:23
BenCI can swap my drive to it and test17:23
rtgBenC: what a PIA. wouldn't it be faster to just reinstall?17:23
BenCtwo screws vs. reinstall?17:24
* BenC goes for the 2 screws17:24
rtgBenC: oh, much simpler then some I've dealt with.17:24
BenCit's too bad the 1420 doesn't have a wwan slot for my verizon card though :/17:24
rtgBenC: is that the laptop you're using since you wrecked your XPS1330?17:25
BenCyep...I was using the 1535, but that was too bulky17:30
BenClow battery life too17:30
Xsss4hellhi17:39
Xsss4hellTo the kernel developers! Why don't you create a dynamic kernel tick rate, that automatically adjusts from 100Hz to >1000Hz if needed?17:39
mjg59Xsss4hell: The kernel already has a dynamic tick rate, so really what you're asking for is the default timeslice setting17:41
Xsss4hellyes, that what audio production system needs.. I think that's what you said17:41
Xsss4hell:)17:42
Xsss4hellso?17:42
mjg59Xsss4hell: So, the question you want to ask is "Can the default CONFIG_HZ be changed to 1000"17:42
mjg59:)17:42
Xsss4hellno, not really17:42
Xsss4hellmy qustion then is17:42
Xsss4hellCan the default CONFIG_HZ be dynamically changed to 100017:43
mjg59No17:43
Xsss4hellwhy not?17:43
BenCCONFIG_HZ is not a dynamic value17:43
Xsss4hellI know ^17:43
BenCso we can't dynamically change it17:43
Xsss4hellThat's why I'm asking you, if you make it dynamic in the next kernel version17:43
BenChow do you propose we do that when it isn't a dynamic value?17:44
Xsss4hellmodify the kernel's source17:44
BenCDo you have a patch to do this?17:44
Xsss4hellI wished I had one =)17:45
BenCYou do realize it's not dynamic for a reason?17:45
BenCbecause it's hardcoded in a lot of fast path functions17:45
Xsss4hellI don't understand the reason why it is hardcoded.17:46
BenCdynticks is about as good as you are going to get17:46
BenCand we already have that17:46
BenCif you need something better, I would suggest the -rt kernel17:46
BenCwhich is there specifically for hardcore audio work17:46
BenCyou could also try recompiling the kernel with CONFIG_HZ=1000 and let us know how that works for you17:47
Xsss4hellexample audio production midi etc.. needs a very high tick. but these producers don't need the high tickrate 24/7 so why not set it dynamically down, to boost non-time cricital applications17:47
BenCit can't be done17:48
BenCyou are saying it like it's a something simple we can just turn on in the kernel, when it would require tons of time to implement, even more time to regression test, and most likely impact performance to a point beyond acceptable limits17:49
Xsss4hellyou mean, to much work. or technically not possible. because I don't understand the reasone why it doesn't work these way.17:49
Xsss4hellI did compile the 2.6.25.4 kernel with  CONFIG_HZ=1000  by the way. it's great but performance is not good a non time-critical apps and 3d somehow.17:50
BenCIt would take way too long to explain why we can't just do this :)17:50
Xsss4hellhehe I see17:50
Xsss4hellbut can you say if it is technically physically not possible, or it is just too much work?17:50
BenCCONFIG_HZ is hardcoded in the kernel, which means replacing it with a variable reference, which means overhead in time critical functions17:51
BenCit's possible, just not worth the effort because of the performance loss17:51
BenCIOW, you would lose performance by simply replacing CONFIG_HZ with a variable17:51
Xsss4hellthen maybe running time-critically marked apps with CONFIG_HZ=1000 and all other with the CONFIG_HZ rate. Similar to QoS priority tagging17:52
BenCThat makes no sense17:52
BenCHZ isn't a per process setting17:53
Xsss4hellIt is just an idea. I know it isn't dynamic.17:54
BenCthat's not even the point17:54
BenCHZ isn't a for applications...it's for the kernel17:54
stgraberyou can't change the rate for a single app, it's for the whole system17:54
stgraberso if you implement something like that, it'd mean you change HZ to 1000 as soon as one of those "tagged" app is started17:55
BenCHZ is (basically) the number of times the kernel hits a timer interrupt, in which it can do things like scheduling processes17:55
BenCthe more times you do a timer interrupt, the more changes per second the kernel can do context switching between processes, but it also means more time the kernel spends doing those things, instead of letting your processes run17:57
BenCs/changes/chances/17:57
Xsss4helltrue17:57
Xsss4hellsuboptimal17:57
Xsss4hellit isn't a generic solution, but a special: like generic user? take generic kernel | hardcore audio user? take rt kernel17:58
Xsss4helllike said it was a thought. I hoped that there is no need for a different CONFIG_HZ rate and things could work with one variable CONFIG_HZ rate..sadly it costs to much effort to create such a thing18:00
BenCThere's lots of ways to adjust the scheduler though...perhaps there's some way to tune things to your liking18:00
Xsss4hellty for your patience and this informative chat session18:13
Xsss4hellgood bye18:13
fokartg, Hi!  Just a follow-up to the seemingly corrupted dmesg output seen in RTL8102E bug reports:19:11
fokaBenC, https://bugs.edge.launchpad.net/ubuntu/+source/klibc/+bug/23528219:12
fokartg, It is not memory corruption, but rather a buggy dmesg  :-p19:12
Tophatwheee - hey mates ive read over the wiki and all the fun articles i can find.  but im still not quite sure how to get developing on the ubuntu kernel.....you know being a windows developer and all.19:19
rtgfoka: right, it was the result of memory corruption in video ram because the network driver crashed.19:19
lagaTophat: well, what do you want to develop?19:20
fokartg, Apparently not... In my case, it actually looked alright on the scrollback buffer, but not the "dmesg > dmesg.log" output.19:20
fokartg, https://bugs.edge.launchpad.net/ubuntu/+source/klibc/+bug/23528219:20
rtgfoka: anything after a crash is suspect.19:20
fokartg, You see, we began seeing the same problem on machines that weren't crashed by a kernel oops...19:20
fokartg, Actually, you may do an experiment to see if it is the case.19:21
fokartg, Try /usr/lib/klibc/bin/dmesg19:21
Tophatlaga - im not to sure, theres tons of things i would like to do.  but im not sure if you use git or cvs.  how you do your pushing (who decides whats good and whats not) and all that fun stuff19:21
rtgfoka: hm, I didn't realize it occured even when the NIC driver was sane.19:21
fokartg, And compare its output to /bin/dmesg (from util-linux)19:22
fokartg, Let's just say we ran into yet another motherboard which refuses to boot due to SATA timeout, and then we saw the same weird dmesg output... (No kernel oops)19:22
fokartg, I have yet to find time to report the bug about that motherboard though (a new ECS motherboard with nVidia MCP78 chipset)19:23
rtgfoka: I'll assign the bug to myself so I don't forget it, but I've some other stuff thats gotta get done first.19:23
lagaTophat: you can clone the kernel source code for ubuntu using git19:23
lagaTophat: if you got patches, you can send them to the ubuntu kernel mailing list.19:23
fokartg, No problem, it is no hurry, just so that you know /bin/dmesg currently included in initrd.gz is buggy.19:24
lagafor big features, it's probably a good idea to use the vanilla kernel from kernel.org. but i'm not a kernel developer, so someone in here might know better :)19:24
fokartg, I ended up copying a normal /bin/dmesg to a USB stick to get a correct dmesg output.19:24
Tophatthanks laga ill do that and play around for a while19:27
=== graner is now known as pgraner

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