**** BEGIN LOGGING AT Mon Aug 22 02:59:56 2011 Aug 22 09:48:16 ogra_: Do you have a minute? Aug 22 09:48:38 ogra_: I was hoping you could help me with qemu-system-arm? :) Aug 22 11:46:02 Daviey, whats up ? Aug 22 11:48:36 ogra_: I'm really sucking in trying to get either the server cloudimg's or the core images running in qemu-system-arm Aug 22 11:48:56 They don't seem to output to tty.. and i couldn't get data over serial. Aug 22 11:49:03 So i'm not convinced i'm doing it right. Aug 22 11:49:23 I've been trying versatile(a|p)b Aug 22 11:49:34 on what host ? Aug 22 11:49:39 amd64 Aug 22 11:49:41 versatile is rather dead Aug 22 11:49:48 no, which release i mean :) Aug 22 11:49:53 (on the host) Aug 22 11:49:54 Oh, oneiric Aug 22 11:50:02 ah, better use vexpress Aug 22 11:50:14 vexpress is the favoured machine? Aug 22 11:50:28 right, omap should also work i think Aug 22 11:51:50 I am having issues installing libscrollkeeper0 from karmic/lucid repos. all i get is 404... what can i do? Aug 22 11:51:51 ogra_: Could i ask you to try, http://cloud-images.ubuntu.com/oneiric/20110821-armel/oneiric-server-cloudimg-armel.tar.gz (tarball contains a disk img and kernel)) Aug 22 11:52:33 phew, sure, let me install the qemu stuff, i havent used that since a year or so Aug 22 11:52:47 hmm, though i have no oneiric x86 machine Aug 22 11:53:18 ogra_: I can give you ssh access to a virtual machine if that helps? Aug 22 11:53:34 but no graphical interface if you want that Aug 22 11:53:46 nah, let me roll an oeniric chroot quickly Aug 22 11:53:55 ogra_: expect beer. Aug 22 12:04:29 Daviey, oh, btw, http://cdimage.ubuntu.com/ubuntu-server/daily-preinstalled/20110822/ ac100 images ... (they dont work yet, but are close) Aug 22 12:05:05 ogra_: Oh awesome! Aug 22 12:06:03 ogra_: although, for bare metal arm - i care more about the images working through D-I. :/ Aug 22 12:06:12 indeed Aug 22 12:06:31 we dont use them yet untl we have real arm server HW ... Aug 22 12:06:32 .. unless someone writes a pxe enabled dd script :) Aug 22 12:06:41 thats trivial Aug 22 12:35:55 Daviey, so how is your image expected to work ? for the beagle emulation you need a full disk image with boot partitions Aug 22 12:35:59 *partition Aug 22 12:36:20 (containing x-loader and u-boot etc) Aug 22 12:42:12 ogra_: We don't have much idea :) Aug 22 12:42:38 well, you should produce something similar to what we build on the imagebuilders Aug 22 12:43:01 a two partition image with the boot bits in the first (vfat) partition Aug 22 12:43:26 and then go from here https://wiki.linaro.org/Resources/HowTo/Qemu-beagleboard Aug 22 12:43:51 ogra_: I'm loathed to try and replicate what your team is doing. Aug 22 12:44:18 well, probably vexpress can boot more similar to what versatile was Aug 22 12:44:25 that would make it easier Aug 22 12:44:39 ogra_: Does it have isa-serial per chance? :) Aug 22 12:44:52 no idea Aug 22 12:44:57 i never used it :) Aug 22 12:45:08 yeah, libvirt seems to think everyone wants it :) Aug 22 12:45:38 https://wiki.linaro.org/PeterMaydell/QemuVersatileExpress Aug 22 12:45:57 that looks more backwards compatible to versatile than qemu Aug 22 12:46:04 and it seems to be able to emulate 1G Aug 22 12:46:40 heh, ..."If you want you can add "-smp 4" to make it boot as a 4-core SMP model." Aug 22 12:47:15 heh Aug 22 12:47:51 ogra_: Who would be the best person for utlemming to work with to fix our build process? Aug 22 12:48:12 utlemming is our cloud image master. :) Aug 22 12:48:15 well, lets find a proper concept first :) Aug 22 12:49:17 hummmm.... Aug 22 12:49:27 ogra_: Well use case is two fold: 1) Emulated ARM hardware using qemu-system-arm.. 2) Some manipulation to get it running on bare metal (probably pandaboard) via LXC. Aug 22 12:49:33 we dont seem to build vexpress netinst images anymore Aug 22 12:50:11 so there is no vmlinuz :( Aug 22 12:50:25 argh Aug 22 12:51:51 i wonder if linaro has something like that Aug 22 13:12:47 Daviey, so using vexpress it cant determine the filesystem of the img it seems Aug 22 13:13:00 argh Aug 22 13:13:10 what fs is it ? Aug 22 13:13:42 $ file oneiric-server-cloudimg-armel.img Aug 22 13:13:42 oneiric-server-cloudimg-armel.img: Linux rev 1.0 ext4 filesystem data, UUID=87915972-5866-47e0-af76-306c3a1143b4, volume name "cloudimg-rootfs" (extents) (large files) (huge files) Aug 22 13:13:54 oh Aug 22 13:14:01 it doesnt try ext4 at all Aug 22 13:14:13 ah, ok - what should it be? Aug 22 13:14:29 it checks ext2/3 above the mount error Aug 22 13:14:57 [ 1.231251] No filesystem could mount root, tried: ext3 ext2 cramfs vfat btrfs Aug 22 13:15:19 Odd that it supports btrfs but not ext4 :/ Aug 22 13:15:26 Daviey, so here is what i did ... Aug 22 13:15:52 pulling the latest vexpress hwpack from http://snapshots.linaro.org/oneiric/vexpress-oneiric/20110822/0/images/hwpack/ Aug 22 13:16:28 (there is sadly no "current" link and the hwpacks have version numbers in their filename) Aug 22 13:16:59 unpack it, dpkg -x pkgs/linux-image-3.0.0-1001-linaro-vexpress_3.0.0-1001.1~ppa~natty_armel.deb . Aug 22 13:17:22 qemu-system-arm -kernel boot/vmlinuz-3.0.0-1001-linaro-vexpress -M vexpress-a9 -cpu cortex-a9 -m 1024 -append 'root=/dev/mmcblk0 rw mem=1024M raid=noautodetect console=ttyAMA0,38400n8 rootwait vmalloc=256MB devtmpfs.mount=0' -sd oneiric-server-cloudimg-armel.img -nographic Aug 22 13:20:45 ... Aug 22 13:20:46 heh Aug 22 13:20:48 root@osiris:/root/tmp# grep EXT4 boot/config-3.0.0-1001-linaro-vexpress Aug 22 13:20:49 # CONFIG_EXT4_FS is not set Aug 22 13:21:00 intresting, not even a module Aug 22 13:21:10 ogra_: that is great! What do we need to do, to get the hwpack into Oneiric? Aug 22 13:21:28 the vexpress package might be in the archive somewhere actually Aug 22 13:21:49 Ah ok, does vexpress max out at 512MB of RAM like versatile? Aug 22 13:22:57 there you go http://ports.ubuntu.com/pool/main/l/linux-linaro-vexpress/ Aug 22 13:23:07 it should apparently support 1G Aug 22 13:24:01 groovy. Aug 22 13:26:04 Daviey, if you download the kernel-image udeb (you can also extract that with dpkg -x), that only contains vmlinuz and System.map Aug 22 13:26:37 ah, worth knowing Aug 22 13:27:22 ogra_: So, is there a reason it shouldn't support ext4? Ie, should i raise a bug - or is there a logical reason? Aug 22 13:27:57 are your instrances supposed to survive reboots ? or is it some throw away thing ? Aug 22 13:28:12 i dotn see a reason to not support ext4 Aug 22 13:29:03 if its throw-away i would actually go without journal, that speeds up I/O (and if you dont need reboot-recovery the journal is rather moot anyway) Aug 22 13:34:37 ogra_: should be persistent across reboots Aug 22 13:34:53 ah, k, then better keep a journal ;) Aug 22 13:35:18 although perf is probably more importiant than reliability at this stage IMO :) Aug 22 13:35:53 well, talk to linaor if you need config chaneges in the kernel Aug 22 13:35:58 *linaro Aug 22 13:37:00 linaro being on a different release cycle to us kind of concerns me to rely upon their kernel TBH. Aug 22 13:37:30 the packages in the ubuntu archive fall under the ubuntu release schedule Aug 22 13:37:49 (not more QA than other universe packages indeed) Aug 22 13:39:04 but indeed you can use omap instead, but with less ram and more complex image creation Aug 22 13:39:36 (since the vm behaves exactly like a beagle you also need a matching image) Aug 22 13:41:02 Yeah, i want to be able to allocate as much memory as possible. Aug 22 13:41:07 Ieally up to 16GB :) Aug 22 13:41:11 lol Aug 22 13:41:15 Ideally* Aug 22 13:41:28 so you should go into qemu hacking and invent a VM that can do that Aug 22 13:41:45 i think the vexpress is the biggest we have up to now Aug 22 13:42:17 also dont expect great performance of qemu in general Aug 22 14:43:10 I'm trying to do "sudo -s" in console, but all i get is "sudo: must be setuid root" - what can I do? Aug 22 14:44:43 The system I use (Ubuntu inside maemo on N900) arrived with that glitch. I did nothing to provoke that, I can't reinstall to fix it. Aug 22 14:56:00 help Aug 22 15:55:41 utlemming: Hello! Aug 22 15:55:47 howdy Daviey Aug 22 15:56:18 utlemming: meet ogra_ and NCommander.. they are your new best friends, which we will have to buy lots of beer for at UDS> Aug 22 15:56:50 hello ogra_ and NCommander Aug 22 15:58:54 utlemming: So ogra_ managed to get your images working by doing, http://pb.daviey.com/nINF/ Aug 22 15:59:19 ext4 isn't supported by the kernel at this stage, which is why you've changed it to ext3. Aug 22 15:59:50 I just pulled ext4 out of the recipe Aug 22 15:59:58 and I was working on the in-image kernel Aug 22 16:00:25 looking at your pastebin, do we not have a working Ubuntu-provided kernel? Aug 22 16:00:31 You should be able to use, http://ports.ubuntu.com/pool/main/l/linux-linaro-vexpress/ Aug 22 16:01:40 utlemming: I got a proof of concept with openstack starting the instance, but i lucked out with the options.. If you are able to ack that process works.. we should be able to land that soon. Aug 22 16:02:43 yeah, the change shouldn't be too difficult. I should have confirmation shortly. Aug 22 16:08:04 do we not have a 3.0.0.x kernel for arm images in the ports pool? Aug 22 16:10:04 utlemming: apparently, https://launchpad.net/ubuntu/+source/linux/3.0.0-9.12/+build/2733971 Aug 22 16:10:51 ah...okay, so I was using the right kernel :) Aug 22 16:11:22 pass. someone from the arm team is best placed to answer that. Aug 22 16:12:33 utlemming: We do for omap/omap4, not sure why there isn't one for vexpress. Aug 22 16:13:44 I was building the cloud images with the omap kernel simply because it was the only one current for the 3.0.0.x kernel tree. Aug 22 16:14:45 Might be a question for ppisati. Aug 22 16:17:12 ppisati: can you chime in on the building of vexpress arm kernels? Aug 22 16:27:36 Daviey: why am I buying utlemming beer? Aug 22 16:30:24 NCommander: Awesome! You can buy me one aswell. Aug 22 16:31:00 NCommander: I think you didn't parse it correctly, beer is being provided for you, not by you. Aug 22 16:31:00 Daviey: er,I just asked why you are buying me beer Aug 22 16:31:06 oh, awesome Aug 22 16:31:07 yay Aug 22 16:31:19 * NCommander got wired crashed and might be semi-sleep deprieved Aug 22 16:31:29 Can I get one too? Aug 22 16:32:15 charlie-tca: no. Aug 22 16:32:38 oh, well. Worth asking, anyway :) Aug 22 16:33:12 charlie-tca: heh, sure you can - utlemming is buying. Aug 22 16:34:39 * GrueMaster sighs. Everyone always ignores the QA guy when beer is involved. Aug 22 16:36:22 +1 GrueMaster Aug 22 16:38:37 Hi, ogra_ Aug 22 16:38:53 this is regarding USB OTG Aug 22 16:39:37 last tuesday u had mentioned that u will update the status on USB OTG Aug 22 17:33:59 How can I generate a new .Xauthority file? Aug 22 18:45:14 has anyone had success with the vexpress kernel and networking? Aug 22 18:45:34 I'm seeing "qemu: hardware error: lan9118: Unimplemented MAC register write: 9 = 0x8100" Aug 22 19:59:17 utlemming: lool experienced that a while ago Aug 22 19:59:56 Daviey: thanks Aug 22 20:00:54 utlemming: we need to cherry pick, http://lists.gnu.org/archive/html/qemu-devel/2011-03/msg02900.html Aug 22 20:01:30 Or with a better comment, http://git.linaro.org/gitweb?p=qemu/qemu-linaro.git;a=commitdiff;h=a0313c00fc Aug 22 20:04:30 * utlemming looks to test patch Aug 22 20:13:52 utlemming: If that resolves it, lets get a debdiff / branch together to upload. Aug 22 20:14:33 Daviey: sure thing Aug 22 20:25:13 Daviey: patch confirmed Aug 22 20:31:08 utlemming: rocking, show me the money^D debdiff Aug 22 20:36:17 Daviey: http://uec-images.ubuntu.com/oneiric/20110822-armel.2 (new images) Aug 22 20:36:51 utlemming: ROCKING.. can you throw me the qemu-system-arm command line you tested it with? Aug 22 20:40:28 Daviey: give me a minute...the image isn't pinging so I think my network setup needs some work Aug 22 20:41:51 bah, who needs network access. Aug 22 20:46:21 utlemming: vexpress networking only half works sadly; I get some packet drops when doing a netinst :-/ Aug 22 20:46:22 utlemming: but I didn't get your failure Aug 22 20:46:22 https://bugs.launchpad.net/qemu-linaro/+bug/799757 Aug 22 20:46:23 Ubuntu bug 799757 in qemu-linaro "Network unstable with vexpress model" [Undecided,New] Aug 22 20:47:30 lool: well then, whats your qemu invocation? Aug 22 20:47:34 I'm using: qemu-system-arm -kernel boot/vmlinuz-3.0.0-1001-linaro-vexpress -M vexpress-a9 -cpu cortex-a9 -m 1024 -append 'root=/dev/mmcblk0 rw mem=1024M raid=noautodetect console=ttyAMA0 ip=10.1.6.2::10.1.6.1:255.255.255.0 rootwait vmalloc=256MB devtmpfs.mount=0' -sd oneiric-server-cloudimg-armel.img -net nic,vlan=0 -net tap,vlan=0 --nographic Aug 22 20:50:45 utlemming: I'm not passing any -net, not sure what raid=noautodetect does and the one thing which is bad here is your -sd which should be -drive file=sd.img,if=sd,cache=writeback Aug 22 20:51:08 you don't actually need -cpu Aug 22 20:51:21 I believe you want console=ttyAMA0,115200 Aug 22 20:51:53 I don't know why you pass vmalloc=256MB devtmpfs.mount=0 either Aug 22 20:52:35 both those were feed to me by http://pb.daviey.com/nINF/ Aug 22 20:53:25 lool: well, that worked better Aug 22 20:55:14 Daviey: use qemu-system-arm -kernel boot/vmlinuz-3.0.0-1001-linaro-vexpress -M vexpress-a9 -cpu cortex-a9 -m 1024 -append 'root=/dev/mmcblk0 rw mem=1024M console=ttyAMA0,115200 rootwait ' -drive file=oneiric-server-cloudimg-armel.img,if=sd,cache=writeback --nographic Aug 22 20:55:34 utlemming: is this on oneiric? Aug 22 20:55:58 I'm using natty. Firing up oneiric now Aug 22 20:56:23 utlemming: --nographic should be -nographic Aug 22 20:57:05 yup, it should. But qemu didn't spawn an SDL window Aug 22 20:58:21 utlemming: Argh! That explains why it's not fixed for you :) Aug 22 20:58:36 .. and why i started ranting :( Aug 22 20:58:50 My pleasure to increase your stress level a bit Aug 22 21:02:23 utlemming: I live for stress, yeah baby! Aug 22 21:02:45 Asside from being _dog_ slow, I can confirm the image boots Aug 22 21:03:29 utlemming: drop -nographic if you want the SDL window Aug 22 21:03:45 lool: is SDL faster? Aug 22 21:03:48 no Aug 22 21:03:53 but it gives more features Aug 22 21:04:01 you can switch between qemu console, serial console and fb Aug 22 21:04:18 I usually disable graphics and use -serial stdio to get the output of ttyAMA0 on my terminal Aug 22 21:05:20 lool: do you have any suggestion for making it run any faster? Aug 22 21:06:52 utlemming: only use one CPU, not SMP (1 x cortex-a9 is fine); disable safe writes (-drive above); put the -sd in tmpfs; run on a fast intel box :-) Aug 22 21:07:06 utlemming: it will be slow though Aug 22 21:07:44 utlemming: some useful things can be run under qemu-arm-static instead; you will be using your intel kernel and hosts' files Aug 22 21:07:56 depends what you're doing, and some things also fail utterly in that mode Aug 22 21:08:34 the images are being installed under qemu-system-arm -- it takes about an hour to generate the images, but works well for debootstrap Aug 22 23:24:04 Anyone about? I'm seeing a 20 % drop in performance after updating my PandaBoard kernel to 3.0. Looking for some kernel hackers... Aug 22 23:24:56 michaelh1: Describe your test method. Aug 22 23:25:28 GrueMaster: CoreMark, which is a CPU bound benchmark. First noticed on my own NEON benchmark which shows the same results. Aug 22 23:26:21 GrueMaster: NFS root. Build CoreMark. Boot into a random 2.6.35. Run CoreMark. Boot into linux-linaro 3.0. Run same binary. Compare. Aug 22 23:26:41 GrueMaster: The bogomips in /proc/cpuinfo is also at ~80 % Aug 22 23:27:22 LP: #831683 Aug 22 23:27:45 michaelh1: I'd have to check, but I believe there was an issue with the cpu speed above 900mhz in the kernel (may have been beagleXM). Aug 22 23:28:05 lp 831683 Aug 22 23:28:06 Launchpad bug 831683 in linux-linaro "Performance regression between 2.6.35 and 3.0" [Undecided,New] https://launchpad.net/bugs/831683 Aug 22 23:28:17 thanks, ubot2. Aug 22 23:29:11 Ah, that's the incantation! Aug 22 23:29:45 GrueMaster: the board is stable with a .35 kernel (it's used in a build farm and works quite hard) Aug 22 23:31:27 It is highly possible that this may be related to bug 709245 Aug 22 23:31:29 Launchpad bug 709245 in linux-ti-omap4 "ARM SMP scheduler performance bug" [High,Confirmed] https://launchpad.net/bugs/709245 Aug 22 23:32:20 Try running with nosmp on the kernel cmdline and see if that helps performance. Or, try ping -f to the machine while it is busy. Aug 22 23:35:17 GrueMaster: will do Aug 22 23:36:29 GrueMaster: note that the machine is idle and this is a single threaded benchmark Aug 22 23:37:25 Well, the benchmarks we ran earlier were mainly hdparm and discovered a 10x boost with nosmp. Aug 22 23:41:51 GrueMaster: nope, nosmp has no effect. Noted on the bug. Aug 22 23:42:28 Ok. Worth a shot though. Aug 22 23:43:35 GrueMaster: is there a debug flag I can set to show the clocks/frequencies that have been set? Aug 22 23:44:27 It should be in the dmesg output or /var/log/syslog. Aug 22 23:48:29 GrueMaster: no, nothing. We'll see what people think of the bug... **** ENDING LOGGING AT Tue Aug 23 02:59:56 2011