IRC Logs for #crux Saturday, 2012-09-15

Romsterfrinnst, over 200 load average yet?00:01
RomsterhorrorStruck, with this newer 3.5.x kernel can i use default txquelength again or is it a benefit to keep using the reduced values as i do now?00:01
Romsteralso what other tune-tables are there to play with?00:02
*** SiFuh has quit IRC00:02
Romstererr tune-ables00:02
*** SiFuh has joined #crux00:10
*** SiFuh has quit IRC00:17
*** himynameisphil has quit IRC00:18
*** himynameisphil has joined #crux00:18
horrorStruckRomster: if you use bql you can forget about txqlength and play with /sys/class/net/wlan0/queues/tx-0/byte_queue_limits/* but that's for ethernet only, wireless driver dont support bql yet00:45
horrorStruckRomster: you can give fq_codel a try too, supposed to do wonders00:46
horrorStruckhmm well, it most likely will be /sys/class/net/eth0/queues/tx-0/byte_queue_limits/* instead00:47
horrorStruckwell you get the point00:47
Romsterhow do i know when i'm using bql?00:47
Romsteris it default or i must choose it?00:47
horrorStruckcan't remember, i think it's on by default. are the knobs there?00:48
Romstereth0  eth1  eth2  lo  ppp0  sit000:49
horrorStruckbql stands for byte queue limits so if you have this then it's ok00:49
Romsteryeah i know the abbreviation00:49
Romsterok i got /sys/class/net/eth0/queues/tx-0/byte_queue_limits/ for eth0 which is my ppp0 interface.00:50
Romsterwonder what i can do with rx too.00:51
Romsteri really ought to search these values to know what exactly they do.00:51
Romsterright now my txquelen is 1000:51
horrorStruckthe problem (we discussed this is the past IIRC) is darkbuffers00:53
RomsterhorrorStruck, thanks for the info i'll look into this.00:53
Romsterhmm indeed i did patch my kernel on the older version to lower that value as it was hard coded for that nic00:53
Romsterhaving 3 different models of nics means i can touch one and not affect the other two.00:53
Romsterbkl can't do darkbuffers stuff can it?00:54
Romsteroops bql i mean00:54
horrorStrucki'm not sure Romster but i think it helps. sorry for this dumb answer00:54
Romsteri'm still on the hunt for tcp window manipulation instead of doing IMQ for ingress shaping.00:54
Romsteroh i was goign to ask wonder about the rp-ppppoe update when/if it'll be merged into ppp, or is the ppp module for atm encapsulation pretty much dead to updates.00:55
Romsterasking you in case you've already looked into it.00:56
Romsteri'm qite happy using ppp instead of rp-ppppoe as that avoids that ugly hack for serial console.00:56
horrorStruckRomster: has some nice comments about BQL and dark buffers00:56
Romsterhmm haven't got that one bookmarked, corrected ta.00:57
horrorStrucki havent looked at ppp since inital setup of this crux router/fw box01:00
Romsteri bumped my IMQ stuff like 1-2 weeks ago that all works but i haven't looked at ppp01:02
Romsterall i do remember is gentoo has a fuckton of patches for ppp01:02
Romsterso i'm thinking upstream maintainer is not there doing there job.01:03
*** mike_k_ has joined #crux01:03
horrorStrucki dunno about gentoo patches but it seems ppp is still somehow active;a=history;f=drivers/net/ppp;hb=HEAD01:08
Romsterppp-2.4.5.tar.gz17-Nov-2009 01:5601:14
Romsterwonder why they haven't released a tarball in ages then01:14
horrorStruckaaah sorry, userspace ppp01:14
horrorStruckmy link is about the kernel bits01:15
Romsteryeah the userspace not kenrelspace01:15
horrorStrucki'll try that;a=summary01:16
Romsterah they have imported the fixes from rp-pppoe so why have they never made a new tarball...01:19
Romsteri'll make a snapshot i think, later when i'm not pre-occupied fixing other stuff.01:19
horrorStruckbuilding git master right now01:20
*** horrorSt1uck has joined #crux01:21
horrorSt1uckstill alive :P01:21
Romsterperformance improvements?01:21
horrorSt1uckcan't tell just right now01:22
Romstermight be an idea to ask jue to roll a newer ppp port01:22
horrorSt1uckdo we have a policy WRT using development snapshots or it's a case by case thing?01:23
*** horrorStruck has quit IRC01:24
*** horrorSt1uck is now known as horrorStruck01:25
Romsterprobably case by case *shrugs*01:30
Romsterif not i'll just run it myself.01:30
horrorStruckwhat do you normally use to benchmark networking? i used netperf but it's based on client/server design so i dont think my isp will install it to make me happy :P01:36
*** SiFuh has joined #crux01:37
Romsteri do unscientific measurements with what ever i fancy. large upload/download to saturate link look at mtr/ping01:38
Romstermtr is nice as it will average out latency01:38
Romsteri really do need a better way of measuring.01:38
Romsterand i've been lazy to setup some way to check latency of every tc qdisc i have too.01:38
horrorStruckok thanks Romster. i'll install mtr and give it a try01:41
*** SiFuh has quit IRC01:42
Romsteri should relaly look into that idea of some latency measurement in real time and adjust tc bandwidth in real time.01:42
horrorStrucklook at netperf, it's been added in the repo recently01:42
Romsterah k01:43
Romsterhmp hal is broken01:48
horrorStruckanyone using xbmc git mster branch, they added java dependency and i can't get a way to disable this01:50
Romsterkinky :P j/k01:52
Romsterno i haven't used xbmc01:52
cruxbot[opt.git/2.7]: taglib: 1.7.2 -> 1.801:54
*** vaddi has joined #crux01:55
horrorStruckthey also added versioned dep to taglib 1.8 :P thanks Romster01:56
cruxbot[contrib.git/2.7]: gst-plugins-good: remove hal dependency01:56
Romsterseems i have a few very important ports now, haven't really thought about what i have maintained.01:59
cruxbot[opt.git/2.7]: libxslt: 1.1.26 -> 1.1.2702:03
Romsterbtw i did check all my opt ports like a day ago so this was very recent.02:04
Romstergod damn it services have gone down again02:05
*** sh4rm4 has quit IRC02:17
cruxbot[opt.git/2.7]: wine: 1.5.12 -> 1.5.1302:59
*** horrorStruck has quit IRC03:09
*** horrorStruck has joined #crux03:09
*** mike_k has joined #crux03:24
*** mike_k_ has quit IRC03:25
*** kinozawa2 has quit IRC03:45
*** osX64 has quit IRC03:45
frinnstRomster: dunno, not at work05:01
*** SiFuh has joined #crux05:35
*** SiFuh has quit IRC06:22
cruxbot[xorg.git/2.7]: xorg-libpixman: fix circular dependency with cairo, thanks Alan06:57
cruxbot[xorg.git/2.7]: [notify] xorg-xcb-util: 0.3.9 does not rely on gperf and m4 following removal of deprecated atoms06:57
cruxbot[xorg.git/2.7]: [notify] xorg-xbacklight: 1.1.2 -> 1.2.0, removed dep xorg-libxrandr, added dep xorg-xcb-util06:57
cruxbot[xorg.git/2.7]: xorg-xf86-video-modesetting: initial import07:29
poulecacaI'm wondering if they are some guys here that have already managed to build a crosscompilation toolcahin for arm07:38
*** horrorStruck has quit IRC07:42
pitillopoulecaca: yes, we are working on it. Both abis supported atm (softfp and hardfp) without optimizations
pitilloit can be customized to add any kind of fpu optimizations from toolchain to pkgutils07:46
*** horrorStruck has joined #crux07:48
pitillotoolchain gitrepo is softfp and toolchain-hardfp adds HF support. toolchain-noeabi is the older one ready for old devices (<armv5te if I'm right) and another one based on uclibc but not tested/used, as we try to follow CRUX at that level too07:50
poulecacapitillo: I asked that because I encounter problem copiling gcc (for the final stage)07:54
pitillopoulecaca: are you doing it from scratch?07:54
pitillowhich versions are you using to build it?07:55
pitillodo you have a template/script or something to take a look?07:55
poulecacahere are the version of all I use binutils-2.22, gcc-4.7.1, glibc-2.16.0, gmp-5.0.5, mpc-1.0, mpfr-3.1.107:57
poulecacaI have no script unfortunately07:57
pitillono cloog/graphite optimizations then07:57
pitillocan you pastebin the error?07:57
poulecacagcc complains about "error: cannot compute suffix of object files: cannot compile"07:57
poulecacaI can pastbin the config.log07:57
poulecacabut I see nothing really usefull07:58
pitillowe haven't tested yet gcc 4.7.1, as we are waiting 4.7.2 (there are some bugs in 4.7.1 atm for arm which some distros have patched)07:58
pitillothis next week we'll put hands on the new toolchain based on 4.7.2 (if the news about the release are right) and same CRUX versions for other components (and if core developers go ahead with cloog optimizations, we'll check them too)07:59
pitilloare you using a jail for your toolchain development?07:59
pitillopoulecaca: I recommend you to stay near gcc 4.5.x atm, better than 4.7.1, as you will hit the bugs if you go ahead compiling more ports. If you have some time, check out our repository (toolchain.git if you don't want hardfp ABI), Toolchain is working quite stable in some devices (efikamx and rpi are the last ones where we are making lots of tests and are running quite stable)08:02
poulecacaI actually compil everything into a directory in which I plained to chroot to compile everything08:02
pitillothen in that way, shouldn't be there any problems if you chroot later and don't mix your host with it (cross is quite a pain because it will find things which souldn't be detected)08:04
*** sh4rm4 has joined #crux08:05
pitillowe started building releases from cross, and finished buiding a first stage release born from cross and deployed to a device, where we rebuild a second stage to provide a native and clean release (without optimizations... or that could be really hard to maintain as day by day, there are more devices and newer cpus)08:05
poulecacaseems to be fastidious08:06
poulecacaI have to go but I will come back and try with a gcc 4.5.x08:06
pitilloif you want to see how fastidious is, check any of the *-cross repositories and open any of the Pkgfiles to see how we remove pkgconfig support and work directly with *_LIBS/CFLAGS....08:07
poulecacaanyway thx pitillo08:07
pitillothis was very good to understand lot of things and learn... but it's time eater and non-practical08:07
pitillonp poulecaca08:08
poulecacahaha I feel like I've just involving myself into something I won't never finished :)08:09
pitilloif you start playing with toolchain... and cross... take your time08:10
pitilloall our work didn't born in one day...08:10
*** SiFuh has joined #crux08:16
Romstertoolchains are very finicky.08:32
*** himynameisphil has quit IRC08:32
cruxbot[opt.git/2.7]: squid: update to 3.2.108:35
pitilloRomster: they are a world...08:38
frinnstfun to mess with :)08:57
*** SiFuh has quit IRC09:08
*** SiFuh has joined #crux09:09
pitillotrue, but sometimes frustrating and irritating09:12
*** SiFuh has quit IRC09:15
*** jue has quit IRC09:19
*** jue has joined #crux09:24
*** ChanServ sets mode: +o jue09:24
*** SiFuh has joined #crux09:42
*** joacim has quit IRC10:13
*** SiFuh has quit IRC10:37
Amnesiais anyone using crux on arm over here?10:55
pitilloyes Amnesia11:04
Amnesiapitillo: I've ordered my rpi, and am making myself an image11:05
Amnesiabut for some reason the supplied kernel doesn't like the hardware emulated by qemu11:05
pitilloI've deployed 2.7.1 softfp image and it's working right11:05
Amnesiaby any chance have an idea how to kickstart it:)?11:05
pitillothose are different machines... versatile vs rpi... quite differents11:05
Amnesiahm ok, as in different hardware ?11:06
pitilloI think it is... checking that link11:07
Amnesiathat suggests that the rpi would be almost the same as arm117611:08
pitilloqemu emulates the cpu... but not the entire hardware11:08
Amnesia /arm113611:08
pitilloThat's normal, since QEMU cannot emulate ALL of the hardware exactly.11:09
Amnesiamakes sense11:09
pitilloqemu has its own amulated hardware11:09
Amnesiabut the essential hardware should be able to ne emulated right?11:09
pitilloit's enought to build native packages inside and then, deploy them to the toy11:09
Amnesiausing it does boot :)11:10
pitilloyou'll launch rpi kernel... ports won't depend on hardware, just in the cpu11:10
AmnesiaI know11:10
Amnesiait isn't a userland issue that it isn't booting11:10
Amnesiaunfortunately I;m just getting a black screen though with the crux.arm supplied kernel.img11:10
Amnesiaso no debugging output:/11:11
pitilloummmm I can't understand11:11
pitillokernel.img is used to boot the rpi, not qemu11:11
Amnesiahm ok11:12
Amnesiamakes sense11:12
pitillowhat are you doing or what do you want to do?11:12
Amnesiaboot crux.arm in qemu:)11:12
pitillodo you want to test crux-arm in qemu? test it and build optimized packages for the rpi?11:13
Amnesiaand I sorta achieved this using the kernel stated above:)11:13
pitilloummmm you can't do that, as kernel.img is built for the rpi adding needed boot headers to boot it11:13
Amnesiaah I see;)11:13
Amnesiahow does everything working exactly with the bootloader?11:13
pitillothat's qemu's "generic" cpu, which we use to test releases before deploying them into real devices11:14
Amnesialol thx11:14
pitillodo you mean how rpi is working?11:14
pitillorpi is an special case, as it hasn't a bootloader (like other devices which have uboot for example)11:15
Amnesiaguess it's the code on the broadcom soc ?11:15
Amnesia*that looks for the stuff in the first boot fat32 boot partition11:15
pitillothe bootloader in rpi, if I'm right, is deployed in the boot partition, as you need to download some blobs to boot it and some needed headers in the kernel image11:15
pitilloin rpi's case, it hasn't nand/nor/... to store the bootloader... and bootloaders isn't in soc (or I think so), in this case, there is something in the gpu, but I don't understand really well this last fact11:18
pitillothis is interesting about rpi
Amnesianeat thx11:19
pitillogpu it's part of the soc, then in this case... the bootloader is in the soc xD11:19
Amnesiathought so11:19
Amnesiabeen also using network block devices11:20
Amnesiait' pretty neat11:20
pitilloummm I think for the next release, I should put hands on qemu and check newer supported cpu... with current versatile it's slower than in a real device11:21
pitilloI use nfs and it does its job fine. I played with nbd but I got problems while building big ports and I finally put in a side and went ahead with nfs again (as people told that is more stable nbd than nfs, and gives less problems... but not in my case)11:22
Amnesiahm well nbd is slow though11:22
Amnesiabut it's ideal for the mounting of qcow2 images11:22
pitillointeresting, I didn't know about qcow211:24
pitillothe only thing which made me feel interesting about nbd was the fact of making swap over the net... (with old devices which had 32/64MB ram could be interesting)11:25
*** SiFuh has joined #crux11:29
Amnesiapitillo: it's also neat that you can there by mount an indivual partition from a block device11:37
pitillosure it is, but in my case, it's enought to get space where I can develop/build without touching SD cards or usb drives. Faster and less destructive11:39
AmnesiaSD cards suck11:42
*** lasso has joined #crux11:42
pitilloand they suck more if you stress them...11:43
Amnesiahm the versatile branch isn;t booting^^11:47
*** SiFuh has quit IRC11:48
Amnesiaapparantly the ext2 superblock is missing o011:48
pitilloare you booting using versatile kernel?11:48
pitillohow do you create the .img? following the same commands on the page?11:49
pitilloreally strange11:50
AmnesiaI;m also unable to press ctrl+D in order to get into maintenancem ode11:50
Amnesiamaintenance mode *11:50
pitilloall those steps are reviewed and tested in CRUX hosts11:50
Amnesiaand a fsck on the partition from my host didn't result in anything11:50
pitilloummmm have you setup all the needed files in the rootfs?11:50
Amnesiaaccording to
Amnesiaall it needs is one big partition11:51
pitilloyes, only that... rootfs partition, but it needs to be configured before booting it (as you do in CRUX when you chroot to the mount point)11:52
pitillocan you pastebin your current fstab?11:52
pitillotry to put pass in / from 1 to 011:53
pitilloare modules extracted to the right location?11:55
Amnesiahm that work11:56
Amnesiawhen pass has been set to 011:56
AmnesiaI'll fsck from the guest itself, sec..11:56
Amnesiahm input isn't being picked up11:56
pitillohave you modified inittab/securetty?11:58
Amnesiashouldn't be necessary according to the wiki12:00
pitilloI think qemu talks to guests by ttyAMA0 (at least older versions)12:01
*** SiFuh has joined #crux12:01
Amnesiahm yeah12:03
Amnesiagot it working after changing securetty and inittab12:04
Amnesiadon't think it'd work using sdl out of the box12:05
pitilloummm I don't understand that12:06
Amnesiaqemu's default output is sdl12:08
Amnesiawhich attaches to tty1 or w/e12:08
pitillothat scapes my knowledge. In CRUX qemu is depending in sdl... not sure if it can run without that support (may be the maintainer knows more options)12:09
Amnesiameh cba12:13
Amnesiait;s working now using ttAMA012:13
AmnesiattyAMA0 *12:13
pitillotake some cups of patience if you start working in qemu :)12:16
pitilloand if you put hands on newer supported cpus in qemu, feel free to send any kind of info/patch/whatever12:17
Amnesiawill do12:18
*** horrorStruck has quit IRC12:39
*** horrorStruck has joined #crux12:47
horrorStruckBACK ON THE MAP12:47
pitilloAmnesia: yes, I think there is another project overthere to use uboot from the sd as a kernel.img too, but if I remember right, they haven't full support (no hdmi) which means this is atm for developers who access by serial12:48
Amnesiastill got to wait 11 week till I arive my rpi's :(12:48
Amnesiacan't wait^^12:49
horrorStruckglibc+ricing=utmost lose12:49
pitillobtw, rpi's bootloader is far, as we can use the boot partition on the sd and let the rootfs in USB disk12:49
pitillo11 weeks??12:49
pitilloummmm I was waiting 3 weeks... without counting since feb when they started deploying 10k units...12:50
pitilloI got it on july and I didn't put hands on it since last week12:51
Amnesiamight be due to the fact I ordered it at RS12:51
Amnesiaholy moly12:51
Amnesiathat's ages12:51
AmnesiaI ordered 3^^12:51
pitilloI got it from RS too. I subscribed and then, I got an email telling about "my time" arrived... then payed for it and one month later, got it12:51
Amnesiahm I already paid for it12:52
pitilloummm were you able to order more than one?12:52
pitilloor were you using different identities?12:52
pitilloidentities=email if I remember right12:52
Amnesiaordered 3 with one email12:53
pitillointeresting, I'd have liked to order 2, one for sepen too, but I wasn't able to do that :(12:54
Amnesiaelement has a shorting waiting period iirc12:55
Amnesia3 weeks ish12:55
pitillomay be there is high demand for the rpi yet12:57
*** tilman has quit IRC12:59
Amnesiathat for sure13:01
*** tilman has joined #crux13:01
*** ChanServ sets mode: +o tilman13:01
AmnesiaI want to replace ps3 with a rpi:)13:01
pitillops3? do you want your rpi for gaming?13:02
Amnesiano hd streaming13:04
Amnesiasince that's the only thing I'm using my ps3 for13:04
pitilloah, that's another thing :) on thursday's night I tried to play a 720p clip on the rpi (using e17) and it wasn't able to play it (using nfs on a 100Mb link)13:06
Amnesiadue to performance issues?13:07
pitilloI need more reading, because people use xbmc on the rpi and they are able to play, then, something is missing atm (not sure if at userspace or at boot lvl, as there are some config files to change gpu reserved memory)13:07
pitilloyes, really slow... (TV res 1360x768 and without playing the clip at full screen)13:08
Amnesiawere you by any chance also using any usb device?13:09
pitillobtw, I'm testing softfp release. May be with HF bump it's another history. This needs some time to research and test13:09
pitilloyes, I'm using an external IDE case over USB, where I have 2.7 and a jail to 2.7.1 up to date13:09
Amnesiawell usb/ethernet are on the same buss13:10
pitilloand... input devices (kbd/mouse)13:10
Amnesiaso that might be a cause?13:10
pitilloI don't think so13:10
pitilloI think it's related to the rootfs or ports directly. HF should bump performance (at least linaro/ubuntu people made some performance tests and it grows 30% vs softfp)13:11
Amnesiadid you do any benchmarks?13:12
pitilloand I noticed that bump in the imx51 where I started researching about HF toolchain/rootfs (and it was smoother)13:12
pitillonot atm. I put hands on the kernel, then I put on rootfs. Not much things done atm... I'm just feeding it, but there are a lot of things to read ahead13:13
Amnesiadoes nfs have a lot of overhead?13:13
pitilloI don't know really, here it's working very well at build time (not sure how it press the toy)13:14
rmullAmnesia: Overhead where, on the network?13:15
rmullIIRC it's less talkative than samba/cifs13:15
Amnesiahm ok13:15
Amnesia does look rather smooth13:15
pitilloyes Amnesia, there are lot of videos showing 720p performance and it plays them right13:16
pitillothat's the point about how they do that :) all things optimized for the toy13:17
Amnesiaand the crux.arm rootfs/kernel isn't optimized that well?13:18
niklasswegood evening13:23
pitilloAmnesia: we are working without optimizations, rootfs for softfp isn't optimized for any device (you can do later setting up pkgmk.conf) Currently, hardfp only supports imx51, which is optimized. For next release, we'll do the same as softfp, providing a generic HF release and letting users set it up to their device13:24
Amnesiahm ok13:25
pitillotoo many devices and too many optimizations... and maintaing that only by 2 men, is hard13:25
Amnesiapartially understood^^13:25
Amnesiacan believe that13:25
pitilloand without much feedback, is harder13:25
AmnesiaI'll try and help when I'll have my rpi's13:25
Amnesiaalso planning to run netbsd on it13:25
pitillobut all things are in the development section. You can build whatever you want, from toolchain adapting it for your device or do it later in the rootfs (this is one of the things I need to understand, as I don't know if a generic toolchain will affect in performance later when you try to optimize via CFLAGS)13:26
pitilloboth sounds great, all any kind of help will be welcome, at any level13:27
pitilloabout bsd, I haven't experienced with arm really13:27
AmnesiaI'm still plannning to switch from debian to crux on my laptop13:27
Amnesiaso this'll be a nice beta experience with crux13:27
Amnesiapitillo: should work though13:27
pitilloI moved from debian all machines to CRUX, from my woody router to my sid desktop...13:27
Amnesiasome patches specifically for the rpi have already been commited13:28
pitilloyes, it should as arm arch is supported by bsd13:28
Amnesiadebian's not too bad13:28
pitilloI had experience with openbsd on my router too, and I moved the old duron700 and I used a jail in the efikamx to run all what I had under obsd13:28
Amnesiaalthough I;ve got the feeling they're following fedora/ubuntu a little bit with bloat/useless dependencies13:28
pitilloI'm not telling it is good or bad :)13:28
pitilloI used it and I was very happy with it, but not really confortable. I found CRUX and put it in my desktop, then, moved to all places13:29
*** mike_k has quit IRC14:03
*** vaddi has quit IRC14:08
*** Rotwang has quit IRC14:20
*** lasso has quit IRC14:45
*** spider44 has quit IRC15:13
*** spider44 has joined #crux15:15
niklasswegood evening17:13
*** SiFuh has quit IRC17:26
*** SiFuh has joined #crux17:46
*** SiFuh has quit IRC18:12
*** SiFuh has joined #crux18:41
*** SiFuh has quit IRC19:11
*** tilman has quit IRC19:32
*** joacim has joined #crux19:33
*** tilman has joined #crux19:33
*** ChanServ sets mode: +o tilman19:33
*** nogagplz has quit IRC21:45
*** mavrick61 has quit IRC21:47
*** mavrick61 has joined #crux21:48

Generated by 2.11.0 by Marius Gedminas - find it at!