/srv/irclogs.ubuntu.com/2011/03/09/#ubuntu-kernel.txt

RAOFHm.  I wish aufs wouldn't go all “Apport would like to pop up and annoy you about this kernel oops” for each build I do in schroot.04:03
=== _LibertyZero is now known as LibertyZero
YokoZarHi, I've been campaigning for an increase to the hard ulimit (leaving soft ulimit in place) for some time now.  It's a simple change, but I'm worried it'll be lost by the wayside this release: https://bugs.launchpad.net/ubuntu/+source/linux/+bug/66309006:09
ubot2Launchpad bug 663090 in linux "Please raise file descriptor hard limit to 4096 (but keep soft limit at 1024)" [Wishlist,Confirmed]06:09
=== smb` is now known as smb
fairuzHi, morning09:01
fairuzif I do taskset -c 0,1 myprog, Its up to the kernel to decide which cpu myprog will run?09:02
jk-fairuz: if it's a single-process, yeah09:03
fairuzjk-: ok.. taskset -c 0,1 myprog is the same as taskset 0x3 myprog right?09:04
jk-fairuz: yep09:05
fairuzjk-: thanks09:05
fairuzjk-: I don't know it's the right channel to ask, but I'm trying to flush my L2 cache controller ( I have the register's addr and all )09:09
jk-fairuz: flush a single cache line?09:09
fairuzjk-: but when I try to write to the flush register of the controller, the kernel freezes09:09
fairuzjk-: no, flush by Way09:09
fairuzjk-: I mean I want to flush all caches09:09
fairuzjk-: Does the kernel have any protection whatsoever thats prevents me from flushing the L2 cache?09:10
jk-heh:09:10
jk-BUGS09:10
jk-       The  current  implementation  ignores  the  addr  and nbytes arguments.09:10
jk-       Therefore, the whole cache is always flushed.09:10
jk-(man 2 cacheflush)09:11
jk-fairuz: I don't *think* so09:11
jk-I'm not too familiar with x86 cache control09:11
fairuzjk-: I think cacheflush is only flushing L1 caches09:12
fairuz(I'm using ARM by the way), maybe  thats the problem09:12
jk-ah09:12
jk-so you're using mrc ... c7 ?09:13
fairuzFor flushing L1 cache, yes09:14
fairuzI use CP15 instructions to flush the cache09:14
fairuzbut since my L2 cache is external, I think I cant use MRC...... to cache L209:14
fairuz*to flush L209:14
fairuzflushing L1 cache works well in my case09:15
fairuzjust L2 that brings problems09:15
fairuzthe registers are memory mapped and I dont know how it dont want to work09:15
jk-man, the operands to mrc always get me confused09:17
jk-fairuz: which platform?09:19
fairuzjk-: OMAP409:19
fairuzcortex-A9 MP09:20
fairuzjk-: you work on ARM too?09:22
jk-fairuz: yeah, a bit09:22
jk-what are you looking to do?09:23
fairuzjk-: Actually I have a program that needs to flush both caches to make it works properly09:23
fairuzin my platform the L2 cache is external09:24
fairuzbut all the registers are memory mapped09:24
fairuzand I thought it should be easy :D09:24
jk-fairuz: take a look at v7_flush_kern_cache_all, in arch/arm/mm/cache-v7.S09:24
fairuzI can read some of the registers09:24
fairuzbut when it comes to the cache operation register, it freezes the kernel09:25
jk-I'm not sure how your L2 behaves though.09:25
fairuzjk-: Yes, I already took a look at it, I think it just flushes L1 caches09:25
fairuzjk-:  It's a PL310 L2 cache controller09:25
jk-might it snoop the L1 invalidates?09:26
fairuzjk-: You mean? Actually I follow a sequence that are recommended in PL310 TRM09:26
fairuzwhich is Clean L1, DSB, Clean & Invalidate L2, Cache Sync, Clean & invalidate L1, DSB09:27
jk-I'm just guessing :)09:27
fairuzjk-: It makes me frustrated..It should be easy =(09:29
fairuzand I dont even know how to debug09:30
fairuzdoes the kernel log even if it freezes?09:30
fairuzi mean just before it freezes09:30
jk-fairuz: I think it's going to wedge on the register write, so no logging.09:36
fairuzjk-: =( bad news to me... I already tried to disable the interrupts before writing to the register but still no luck09:37
fairuzjk-: I wonder what makes the kernel freezes09:37
fairuzjk-: I'm thinking that the cache controller trying to clean the cache but the kernel is trying to write on it? So it's a never ending work09:38
fairuzjk-: which explains the freeze?09:38
fairuzjk-: is that possible?09:38
jk-no idea sorry, I don't have any details about that cache09:39
jk-you might be able to get some help on #armlinux though09:39
jk-or #linaro, or #linaro-kernel09:39
* jk- heading out09:39
sconklin7873ca4e4401f0ecd8868bf1543113467e6bae6110:19
Kanohi, is there a linux-meta 38-6 update somewhere11:47
=== diwic is now known as diwic_afk
=== ogra is now known as Guest90542
=== Guest90542 is now known as ogra_
=== diwic_afk is now known as diwic
fairuzwhat static mapping means? can i still ioremap an address that is already mapped?14:26
=== ogra_ is now known as ogra
sconklinakgraner: you here?14:46
loolOy15:05
loollinux-image-virtual seems to be missing some .kos (nls_cp*.ko notably) in maverick/i386, but they seem enabled in the config?15:06
loolsmoser: ^15:12
loolOk; at least the modules are missing consistently since lucid on both amd64 and i38615:14
loolbut they were present pre lucid, with 2.6.32 kernels15:16
apwlool, can't be very important if they have been missing for 18 months without anyone noticing :)15:17
loolOk, debian.master/control.d/virtual.inclusion-list is the reason15:17
loolapw: Yes, that's what's reassuring me15:17
loolIt seems like a bogus entry in debian.master/control.d/virtual.inclusion-list15:17
apwlikely excluded indeed15:18
loolI would suggest that an exclusion-list should be maintained in parallel and that any mismatch shall fail the build  :-)15:19
apwheh, sounds lovely15:19
apwlool, actually the module lists in the abi stuff would catch a regression15:19
loolapw: How would you catch a new module which you fail to install?15:20
loole.g. upstream adds cpxyz15:20
apwwell we'd only want to add those if they are useful15:20
apwand generally we are including by class15:20
loolapw: I guess that's fair enough, my personal preference is for the tool to be hard on me and force me to document it both ways  :-)15:21
loolapw: What I find odd is that these things get built but not installed15:21
loolapw: Why not disable the configs instead?15:22
loolThis would make the delta apparent in the config handling mechanism15:22
loolfuse is missing too15:22
apwthat bloats the config delta, too.  this was a compromise15:23
loolI wonder whether the config factorizer could keep most common configs at the top of the tree, and have overrides for the most uncommon ones instead of splitting configs as soon as one differs15:24
looloh well15:24
apwi did try that, and it can be done, but it does make it much more confusing15:26
loolOk15:28
loolI filed LP #732046 to let smoser and others chime in15:28
ubot2Launchpad bug 732046 in linux "Missing filesystem modules in -virtual package" [Undecided,New] https://launchpad.net/bugs/73204615:28
loolI propose fixing this in SRUs as well, but feel free to dsiagree15:28
loolsmb: Hey, to workaround LP #732046, I'd like to switch to another kernel; would you know whether I can run -server or -generic under EC2 vms?  do you have hints on how to do that?15:54
ubot2Launchpad bug 732046 in linux "Missing filesystem modules in -virtual package" [Undecided,New] https://launchpad.net/bugs/73204615:54
loolDaviey: Ah I think I'm a bit naive to do it from the vm, you are supposed to pass a kernel when starting the instance, so I'm supposed to provide an alternate aki15:55
smblool, That was Maverick?15:55
loolI guess I should pass a pvgrub one15:55
loolsmb: yes15:55
* lool starts a kernel build to get the modules in the background15:56
smblool, So pvgrub is there already. I'd try to install a server kernel. 15:56
loolsmb: I did, then I did a grub-reboot on it and it didn't come up15:57
loolbut maybe I messed it up, I might not have waited long enough15:57
smblool, was that an m1.large?15:57
loolNo, c1.medium15:57
looli38615:57
smbI think that got more than one cpu too. There is a bog that make reboot from within the instance not working15:58
smbI think it is still waiting to get into proposed15:58
smbYou would need to reboot either from the web interface or the ec2 tools15:58
smbAnd even that seemed to take a long time15:59
DavieyEC2 makes me cry.15:59
loolsmb: right reboot didn't work indeed, but I used ec2-stop-instance -f15:59
* lool tries again, waiting longer for the machine to come up16:00
loolServer.InternalError: Request could not be executed due to an internal service error16:00
looloh wow16:00
smblool, I was using ec2-reboot-intance16:00
smbinstance even16:00
smbor instances16:00
smbCommand completion makes one forget details16:00
smblool, When looking for that bug it felt like taking 5minutes or so16:01
bdmurrayJFo: bug 707353 seems rather important16:08
ubot2Launchpad bug 707353 in linux "[STAGING] Broadcom 4313 firmware fails to load because it is not present" [High,Confirmed] https://launchpad.net/bugs/70735316:08
* smb wonders whether that is that firmware that needs to be obtained with fw-cutter...16:09
smbtgardner, ^ Do you happen to know that from memory?16:09
JFotagged it kerne;-key so it shows on our list16:13
JFothanks bdmurray 16:13
bdmurrayJFo: thank you16:18
=== diwic is now known as diwic_afk
JFo<-lunch17:03
loolsmb: Couldn't get this pv thing to work; I installed linux-image-server which installed the -generic-pae flavor, and it's the first entry in grub.cfg, I grub-reboot 0 and sudo halt, then stop-instance, start-instance, and it still comes up as -virtual17:11
smblool, pv-grub is based on grub1, its /boot/grub/menu.lst that matters17:12
loolapw: Sorry, was wrong earlier, lucid is unaffected17:18
loolsmb: aha, grub isn't installed anymore and grub2 is instead, that's why17:20
smblool, There should be some ec2-grub-bla-legacy too. I know its not the least confusign solution to have both files hanging around17:21
loolgrub-legacy-ec217:22
smbThat should be the one17:22
smbIt generates menu.lst in parralel to grub.cfg17:22
=== cmagina is now known as cmagina-lunch
lool/usr/sbin/update-grub-legacy-ec2 doesn't pick up the generic-pae flabor17:23
loolflavor17:23
smbBut on EC2 grub.cfg is just there to confuse people17:23
loolIgnoring non-Xen Kernel on Xen domU host: vmlinuz-2.6.35-27-generic-pae17:23
loolit seems to be looking for xen in the kernel name17:24
smbHm, I guess virtual17:24
loolI just added the entry manually17:25
smbI thought that there was some normal kernel installed too. But maybe again just to confuse17:25
smbjjohansen, Do you remember whether there was any other kernel flavour that may be bootable in ec2? Until the missing modules is solved...17:27
jjohansensmb: only -ec2 and -virtual17:28
lool2.6.35-27-generic-pae17:29
loolwee17:29
looljjohansen: I just used -server which is /boot/vmlinuz-2.6.35-27-generic-pae17:29
jjohansensmb: -generic and -server are close17:29
loolit booted fine, just had to force the grub entry as the script wouldn't want to install it17:29
smbjjohansen, sounds like it is possible 17:30
smb;)17:30
jjohansensmb: yeah possible17:30
jjohansenon maverick17:30
loolyup17:30
loolmodinfo nls_cp437 is there, cool17:30
* jjohansen was thinking there was a config missing but, that isn't right either17:30
smbpractical proof. qed. pub.17:30
loolsmb: the next thing I would have done is rebuild linux-virtual myself with the same sources17:31
jjohansenyeah I did end up booting some -server and -generic kernels during early dev17:31
loolI would hope the modules would be accepted in this case17:31
loolI do miss console output with the -generic-pae kernel17:31
smbWell that could be the missing config17:32
smbThat xen console is not enabled17:32
* smb -> away17:34
=== cmagina-lunch is now known as cmagina-afk
=== cmagina-afk is now known as cmagina
smoserlool, did you get what you were looking for ?19:11
smoseri can hopefully help19:11
smoserregarding reboot on ec2, that works fine.19:11
smoseri reboot in tests that we run for every release, and loads of times when i'm testing cloud-init.19:12
smoserit is as reliable as rebooting hardware19:12
smoserwell, i'll be around a bit today, and i'm back in tomorrow, if someone has questions. 19:14
kristian-aalborghi all... I have some kind of project going on: http://ubuntuforums.org/showthread.php?p=10541660#post10541660 - building a custom kernel, then moving it to another box19:15
loolsmoser: I found a workarund19:22
loolsmoser: I've sub-ed you to the kernel bugs19:22
lool*bug19:22
loolsmoser: thanks!19:23
smoserthe -virtual kernels should be almost identical to -server... in natty we had some issues, so there are some differences.19:23
smoserbut other than that, they're largely just subsets.19:23
smoserbut it is now a full flavour rather than sub-flavor19:23
sorensmoser: orly? Since natty?20:07
=== Guest21293 is now known as calc
smosermaverick i think it is its own.20:09
jjohansenyep, maverick is where it became its own flavor20:15
* soren is failing to keep up, apparantly. :(20:20
tumbleweedhow do I easily build an ubuntu kernel from the kernel git repo? I just want to test a single patch, so build-mkppa looks like it'll give me a source package I can build, but it seems to expect a debian/changelog to exist20:56
tumbleweeds/kernel git repo/kernel-team git repo/20:56
jjohansentumbleweed: https://wiki.ubuntu.com/Kernel/BuildYourOwnKernel21:49
tumbleweedjjohansen: oh, duh, I missed the clean21:50
=== jjohansen is now known as jj-afk

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