**** BEGIN LOGGING AT Thu Feb 07 02:59:58 2013 Feb 07 08:52:29 ccxCZ: you can bindmount the pulseaudio socket or something Feb 07 08:52:33 and use that Feb 07 08:59:57 yeah, but I'd prefer not having to deal with pa in myc chroot Feb 07 09:00:10 mpd would be quite handy here Feb 07 10:53:58 are there any wine(windows for linux) experts here? Feb 07 10:55:26 mad_dev: I'd say you'd find those in #wine Feb 07 10:56:16 teotwaki: I know...its just that they are idle Feb 07 10:56:30 teotwaki: its #winehq Feb 07 10:57:18 I don't think people in IRC should expect immediate answers. Feb 07 10:57:31 I tried to buy a Renault in a Renault shop, but there were no salesmen available, so I came into this Ford shop. Why won't you sell me a Renault?! Feb 07 10:58:16 teotwaki: point taken Feb 07 11:02:42 mad_dev: if you're thinking about running wine on maemo: it has no x86 cpu to even execute usual windows .exe Feb 07 11:03:13 DocScrutinizer05: wine+qemu :D Feb 07 11:03:40 yeah, I heard they ran WinXP and MacOS10 this way, on N900 Feb 07 11:03:54 or was it winNT? Feb 07 11:04:31 time on macOS til first sign of life on screen was about 2h iirc Feb 07 11:04:52 which macos? Feb 07 11:04:57 DocScrutinizer05: no..i am trying to install some drivers Feb 07 11:05:29 where, and for what? Feb 07 11:06:56 well, someone could make winece, to run wince binaries on arm, while wincing. Feb 07 11:07:12 (surgeon general does not recommend) Feb 07 11:07:14 http://www.tuug.fi/~toni/serendipity/index.php?/archives/9-Windows-NT-4.0-running-on-N900!.html http://www.youtube.com/watch?v=UNo6pn-dnSQ Feb 07 11:07:36 that's bochs, though Feb 07 11:07:38 that's cheating Feb 07 11:07:44 ;-P Feb 07 11:08:48 bochs runs like shit *everywhere* Feb 07 11:09:48 http://www.itpro.co.uk/609997/video-two-operating-systems-on-one-phone Feb 07 11:17:14 http://talk.maemo.org/archive/index.php/t-42908.html Feb 07 11:22:09 jaska: yo dawg! heard you like emulators so we placed a a VM into a VM so you can emulate emulations emulating emulations Feb 07 11:24:09 http://www.engadget.com/2010/01/31/mac-os-x-10-3-installed-on-nokia-n900-via-pearpc-barely-usable/ Feb 07 11:26:31 nice Feb 07 11:27:39 >>As you can see from the heavily edited video, it took almost 2 hours to reach the "About my Mac..." window.<< Feb 07 11:32:04 the scrolling at 1:00 into that video is incredible ;-P Feb 07 11:33:39 fadeout of menu at 2:20 :-D Feb 07 11:33:41 text console on old sun workstations was about similar :p Feb 07 11:37:40 DocScrutinizer05: i'm not sure if pearpc is the best choice there Feb 07 11:38:04 what's faster to emulate, x86 or ppc? Feb 07 11:38:26 I'm not sure if the idea at large been the best choice, except for mere shits'n'giggles Feb 07 11:40:35 oh yeah, running a PPC mac binary in that PPC emulator in OSX running in a x86 emulator running in maemo Feb 07 11:41:19 DocScrutinizer05: wine on windows Feb 07 11:42:16 now you just should run maemo of above in qemu in scratchbox on your linux PC Feb 07 11:42:36 DocScrutinizer05: clearly, you should run that in qemu in scratchbox on your scratchbox VM Feb 07 11:42:50 yeah Feb 07 11:42:54 sorry, forgot Feb 07 11:44:25 winCE java program in java VM on winCE run in bochs on.... *cough* Feb 07 11:44:53 do arm cpus have facilities for virtualization? Feb 07 11:45:06 basically yes, afaik Feb 07 11:45:47 they even have two physicaly separated worls: secure and user Feb 07 11:45:53 worlds* Feb 07 11:46:08 wince java program running in java vm in wince emulated by *the n900's dsp*? Feb 07 11:46:24 /wrist Feb 07 11:46:54 cmd unknown: wrist Feb 07 12:09:50 heh, the tracker ETA is funny. 44951 minutes 20 seconds Feb 07 12:20:49 merlin1991: Any news on the packages list with wrong MD5s in extras? Feb 07 12:27:43 eta 1 month :D Feb 07 12:39:28 jumped to 9 mins and now to 4 mins Feb 07 12:39:40 probably it's my huge maildir Feb 07 12:46:14 why, oh why can't people bother to write properly on a forum Feb 07 12:52:11 -on a forum Feb 07 12:55:07 -to write properly Feb 07 12:56:26 hi someone use rockbox? Feb 07 13:51:40 hello Feb 07 13:52:58 hi Feb 07 13:54:47 hello I try irc for the first time... Feb 07 13:55:47 str: congrats Feb 07 13:57:29 :) Feb 07 13:57:44 cool I use it from emacs...even more compelling Feb 07 13:58:03 sorry I maybe need a guide who to bother for what... Feb 07 14:05:12 hah emacs Feb 07 16:25:33 If I were to reflash my phone now and want a nice, optimal clean setup, what would be the recommendation to use? I've been reading up on cssu-thumb a bit and that seems like a good idea. Any other good ideas? Feb 07 16:26:25 cssu-thumb is pretty fucking awesome, yes Feb 07 16:31:07 first good idea is to flash a clean PR1.3 and maybe even VANILLA(eMMC) Feb 07 16:31:34 the rest is installation either way Feb 07 16:32:06 IOW there's no alternative to PR1.3 fiasco image for flashing Feb 07 16:33:40 after that, actually CSSU is recommended. Your call whether this will be CSSU-Stable, CSSU-Testing, or CSSU-thumb on top of testing Feb 07 16:35:09 kerio: so, reflash, install kp & cssu-thumb, done? is it even possible at the moment with the constant state of flux of everything maemo.org? Feb 07 16:35:41 repository.maemo.org is back to working Feb 07 16:35:46 we still have some problems with the repos Feb 07 16:35:48 thumb is for those brave at heart, since you might face ... complications that you'd not see with any other setup, like problems when maybe flashing a normal kernel later Feb 07 16:36:21 and iirc there's a problem with cssu-testing right now, because it eventually depends on a couple of packages that are in extras-* Feb 07 16:36:53 if it's not a problem, you could wait until merlin1991 fixes them Feb 07 16:37:04 or use Stable Feb 07 16:37:26 you could use stable, but then you can't install cssu-thumb on top of it Feb 07 16:37:33 meh Feb 07 16:37:37 nonsense Feb 07 16:37:51 you can switch to testing any time, no? Feb 07 16:37:54 well, yes Feb 07 16:38:03 but not right now Feb 07 16:38:08 which is the problem :) Feb 07 16:38:12 pfff Feb 07 16:38:36 i'm not even sure of which packages are problematic - one is libopenjpeg2, but that's in the plain extras too, so it shouldn't be an issue Feb 07 16:38:40 you could use stable, but then you can't install cssu-thumb on top of it <-- WRONG Feb 07 16:38:49 fine, fine :( Feb 07 16:39:58 ErwinJunge: anyway, right now the recommended procedure is to flash pr1.3 (both combined and vanilla, to be sure), then install some flavour of cssu Feb 07 16:40:13 cssu-stable is "officially" recommended to any n900 user Feb 07 16:40:25 cssu-s -> cssu-t -> cssu-thumb is always a valid path. If it's working or if cssu-t is fsckdup by a enabler that depends on stuff it shouldn't depend on is another question Feb 07 16:45:36 and I can't help but suspect that those rogue dependencies in cssu-t enabler are introduced by uninstall function that should get nuked anyway Feb 07 16:47:01 nah Feb 07 16:47:18 at least, not libopenjpeg2, that's pulled in by tracker Feb 07 16:47:19 :) Feb 07 16:47:20 if that's actually true, then this uninstall option is a nice example what MUST NOT be done in distro maintenance Feb 07 16:47:23 ~2110 Feb 07 16:47:27 ~2119 Feb 07 16:47:27 The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in RFC 2119. Feb 07 16:49:10 any other interesting stuff aside from cssu-thumb? Feb 07 16:49:14 hell, how can cssu tracker depend on a lib from extras-devel? And even if that's true how the heck tracker made it into cssu then? Feb 07 16:49:42 the usual useful little apps Feb 07 16:49:48 and tweaks Feb 07 16:50:08 not much changed regarding validity of suggestions in Feb 07 16:50:11 ~jrtools Feb 07 16:50:12 extra, extra, read all about it, jrtools is http://wiki.maemo.org/User:Joerg_rw/tools Feb 07 16:51:03 oh yeah, extras (testing, devel). Currently I have all enabled, because a lot of stuff never leaves devel. Is that still a valid assessment? Feb 07 16:51:24 ErwinJunge: the best you can do for your N900 is to NOT install a few certain apps Feb 07 16:51:50 DocScrutinizer05: got a list of those? :) Feb 07 16:52:01 ErwinJunge: testing and devel in parallel is absolute nonsense Feb 07 16:52:15 ~speedpatch Feb 07 16:52:15 extra, extra, read all about it, speedpatch is http://talk.maemo.org/showthread.php?p=1012405#post1012405 >>first i don't realy understand what does this patch do (that is why it is called miracle patch)<< [/quote original-author-of-speedpatch] Feb 07 16:52:30 ~batterypatch Feb 07 16:52:30 extra, extra, read all about it, speedpatch is http://talk.maemo.org/showthread.php?p=1012405#post1012405 >>first i don't realy understand what does this patch do (that is why it is called miracle patch)<< [/quote original-author-of-speedpatch] Feb 07 16:52:43 and auto-disconnect Feb 07 16:54:39 I've never trusted either of those patches, so never installed them :) What is auto-disconnect supposed to do? Feb 07 16:54:43 *patch are a random(!) patchwork of arbitrary other workarounds or tweaks that obviously sounded interesting to that guy, but never really been understood by him Feb 07 16:55:28 so the better 50% is useless cruft, while the nasty 50% are not doing good to your beloved N900 Feb 07 16:56:53 auto-disconnect is not really doing what you essentially expect it to do, and can't get uninstalled Feb 07 16:59:05 wow, i just make an update after a long time, 125 upgradable packages Feb 07 16:59:12 Hi, I've installed remind on my n900 and run "play-sound Beep.acc.wav" on each triggered reminder, however it looks like it only works once (only once for the first reminder and does not reoccur and the rest of reminders are ignored). 1) Why? 2) Do I have to put something like this "remind -a ~/.reminders" to /etc/profile or any other place in order for remind to work after reboot? Feb 07 16:59:55 ok, so just extras-devel should be enough? Feb 07 17:00:19 zeev: no clue about reminder, I recommend alarmed Feb 07 17:02:00 DocScrutinizer05: what's that? I doubt it is as powerful as remind... for instance can it generate alarms based on sunset time? what about question 2) - it is pretty general Feb 07 17:02:07 ErwinJunge: basically yes, though it's still not really recommended to keep it for global updates. The SOP is: enable extras-devel, install (or update) that one package you're interested in, then *disable* extras-devel again Feb 07 17:03:03 DocScrutinizer05: is alarmed maemo specific or linux general? Feb 07 17:03:30 zeev: no, it doesn't care about sunset, just about time of day and date. and 2) see my previous comment Feb 07 17:03:59 alarmed is a maemo specific app Feb 07 17:06:08 ok, so I'll stay with remind. 3) can I trigger alarmed from command line? 4) How can I run a CL command in maemo automatically on each boot? Feb 07 17:06:38 Ok DocScrutinizer05 thanks for the help, I'll probably redo the phone somewhere in the next couple of days Feb 07 17:08:52 zeev: 3) Trigger? alarmed has a rich cli interface that allows configuration and listing of alarms. 4) depends Feb 07 17:10:03 if you're asking re remind - it obviously isn't properly adapted to maemo, otherwise you wouldn't need to bother about that by yourself. Which suggests it has other flaws as well Feb 07 17:10:33 DocScrutinizer05: 3) great, so I might be able to use it with remind - where can I read about it? 4) so I want to run this command on each boot: "remind -a ~/.reminders" where to place it? Feb 07 17:10:55 no idea Feb 07 17:11:06 obviously remind is broken Feb 07 17:11:18 so I won't look into it Feb 07 17:11:25 where can i get kp-v52? google doesn't seem to be too aware of it Feb 07 17:11:35 why broken? Feb 07 17:11:41 WizardNumberNext: tmp Feb 07 17:11:44 er, tmo Feb 07 17:11:55 because it seems to be designed for systems that never get booted Feb 07 17:12:24 DocScrutinizer05: its a regular unix command Feb 07 17:12:47 thus, designed for systems that never get booted Feb 07 17:12:53 see - regular unix commands relying on cron etc don't work on maemo Feb 07 17:13:21 even worse if they don't rely on cron but do their own nonsense Feb 07 17:13:32 DocScrutinizer05: so it doesn't hurt to restart it manually on each boot, right? Feb 07 17:13:32 with sleep or timers/alarms Feb 07 17:14:06 it probably DOES hurt, since that tool isn't aware of basic design rules for embedded, particularly maemo Feb 07 17:14:44 for all I know it could do a busy wait until next event due Feb 07 17:15:17 on maemo apps are supposed to use alarmd daemon. Alarmed does use alarmd Feb 07 17:15:39 remind quite evidently doesn't Feb 07 17:16:18 how can I check whether it has busy waits? Feb 07 17:16:44 standard maemo alarm is too limited for me... Feb 07 17:16:50 so, in pretty usual unix manner, it might wake up every second to check if an alarm event is pending Feb 07 17:17:04 and drain battery? Feb 07 17:17:11 yes Feb 07 17:17:27 well, every second won't drain much battery Feb 07 17:17:36 haha Feb 07 17:18:52 let's say, I don't care about memory - sooner or later n900 will have to be attached to power line permanently... Feb 07 17:18:57 btw, alarmed can invoke arbitrary cli cmds Feb 07 17:19:00 about battery Feb 07 17:19:44 it could even invoke this remind command Feb 07 17:20:01 will backup-menu be sufficient for full device backup (and restore) in case of something software-wise broken Feb 07 17:20:02 I need to make alarms depending on sunset Feb 07 17:20:49 WizardNumberNext: almost Feb 07 17:21:10 so close, but it might fail? Feb 07 17:21:32 it will work perfectly if you restore a backup made when the kernel was the same as the current one Feb 07 17:21:56 that is answer, which I wanted! Feb 07 17:22:02 or, rather, if the backup you're restoring has the modules for the current kernel Feb 07 17:22:18 DocScrutinizer05: so, again, If I don't care about battery - how do I run "remind ~/.reminder" on each boot? Feb 07 17:22:32 zeev: does it daemonize by itself? Feb 07 17:23:01 so, basically if something fails, I just check which dir is linked to sth-current in /lib/modules and change it actual kernel installed? Feb 07 17:23:20 -z[n] Feb 07 17:23:20 Runs Remind in the daemon mode. If n is supplied, it specifies how often (in minutes) Remind should wake up to check if the reminder script has been changed. N defaults to 5, and can range from 5 to 60. Feb 07 17:23:30 WizardNumberNext: nah, you don't even need that Feb 07 17:23:57 WizardNumberNext: just make sure that the kernel that's currently flashed has its modules in /lib/modules Feb 07 17:24:02 alternatively, flash the kernel that has the modules Feb 07 17:24:10 (you need a pc for that, though) Feb 07 17:24:31 kerio: actually I'm not going to change the script... so I'm not sure whether it is needed. I also didn't see remind in my ps -x output - which is strange.... Feb 07 17:25:20 kerio: lets say I would reflash it after backup and get to stock kernel, restore and then it seam obvious to me I have to change link in /lib/modules. or I dont? Feb 07 17:25:35 zeev: alarmed --title "invoke remind each morning at 3:00" --command "remind -a ~/.reminders" --cron '0 3 * * *' Feb 07 17:25:41 the link is meaningless Feb 07 17:25:58 all it matters is that there's the /lib/modules/2.6.28-omap1 directory Feb 07 17:26:08 ok, then it is as I was thinking, but checking is always better then running into stupid problems Feb 07 17:26:21 just... don't change kernel Feb 07 17:27:08 kerio: I understand. I only check what can cause it to fail and it case how to fix it. I think that is quite valid approach Feb 07 17:27:29 WizardNumberNext: i usually grab the .fiasco image for the kernel i made the backup with Feb 07 17:27:30 and store both Feb 07 17:27:53 how to grab .fiasco then? Feb 07 17:28:03 other then stock obviously Feb 07 17:28:37 download the .deb and unpack it Feb 07 17:28:41 I have original kernel and kp-v51, so in case I can use it to boot device and fix whatever Feb 07 17:28:42 DocScrutinizer05: if run remind while it runs already all the events will be triggered twice, if I'm not mistaken. That's why I ask how to do it on boot, so there will be only one instance of remind... Feb 07 17:28:43 it'll be in the boot/ directory Feb 07 17:29:02 so I do have everything need, as I have both .deb and unpacked .deb Feb 07 17:29:22 oh, obvious warning: if you're running cssu-thumb, kernels without the errata fix are worthless Feb 07 17:29:27 tell zeev about x y Feb 07 17:29:30 WizardNumberNext: i suppose so Feb 07 17:29:31 tell zeev about xy Feb 07 17:29:39 ~tell zeev about xy Feb 07 17:29:49 ~botsnack Feb 07 17:29:49 kerio: thanks Feb 07 17:30:10 kerio: I know - I read it somewhere and even have some idea why I need errata, but do not remember exactly Feb 07 17:30:21 What happens at sunset? Feb 07 17:30:28 :) Feb 07 17:30:44 except that the sun sets.. Feb 07 17:30:54 it is an xyz Feb 07 17:31:11 zeev: I told you remind is obviously broken resp not properly ported to maemo. So I won't do the meintainer's work, your question should get directed to maintainer of remind, it's a mandatory part of such a tool Feb 07 17:31:47 DocScrutinizer05: hm, do you know if anyone made or proposed an alarmd "compatibility layer" for crontabs? Feb 07 17:32:01 So you a table with the sunset times? Feb 07 17:32:37 DocScrutinizer05: it was not really ported, just compiled Feb 07 17:33:00 kerio: alarmed is as compatible as it gets. I suggested alarmed to also scan crontabs, but shapeshifter rejected that Feb 07 17:33:40 so there is no way to run it on boot, correct? Feb 07 17:33:58 kerio: a "compatibility layer" crontab (the command) -> alarmed is a shell 3liner Feb 07 17:34:29 n900-dk: want to have ability not to miss sunsets or sunrises :) Feb 07 17:34:32 zeev: there's plenty of ways Feb 07 17:34:50 zeev: there for sure are several ways, but only one correct way, and I'm not inclined to analyze remind to find out which is the right one Feb 07 17:34:55 kerio: so give me at least one of them Feb 07 17:35:34 zeev: odds are it won't work Feb 07 17:35:45 zeev: So you download a table of sunset time? And use a script to look up this time of the current day and generate an alarm? Feb 07 17:36:42 n900-dk: its more than just that - remind has very sophisticated syntax for defining alarms which I miss in standard maemo alarm Feb 07 17:37:02 but how will you define the alarm time? Feb 07 17:37:12 zeev: a file in /etc/event.d/ maybe Feb 07 17:37:32 you dont want to set it manually every day, right? Feb 07 17:37:42 yes Feb 07 17:38:11 kerio: how should it look? Feb 07 17:38:24 RTFM Feb 07 17:38:39 zeev: indeed, read the fine manual Feb 07 17:38:48 ok :) Feb 07 17:39:06 ~xy Feb 07 17:39:07 i heard xy is The XY problem: You want to do X, but don't know how. You think you can solve it using Y, but don't know how to do that, either. You ask about Y, which is a strange thing to want to do. Just ask about X. Feb 07 17:39:17 yeah, we still haven't heard X Feb 07 17:39:33 should I actually see remind on ps -x? Feb 07 17:40:17 zeev: how the F**k are *we* supposed to know? Feb 07 17:40:56 can it function without be a deamon? Feb 07 17:41:03 ran into a nice problem today, trying to set rtc with hwclock Feb 07 17:41:25 hwclock cies that the date on the rtc is invalid and thus refuses to set the time Feb 07 17:41:26 theoretically... Feb 07 17:41:29 ? Feb 07 17:41:37 merlin1991: wut Feb 07 17:41:42 yes, it could use cron, or alarmd Feb 07 17:41:43 what's the actual error message? Feb 07 17:41:58 or rename iself to schweinebraten Feb 07 17:42:14 it doesn't speak german Feb 07 17:42:20 merlin1991: there's a flag to override that Feb 07 17:42:24 iirc Feb 07 17:42:33 DocScrutinizer05: nah it's foobar code in the version I had Feb 07 17:42:46 newest util-linux does not have the problem Feb 07 17:42:54 creating 4GiB file on eMMC (MyDocs) using dd takes ages (I do not want sparse-file). Any ideas why? OK, I know using data=journal will make writes two times longer, but this is extremely slow Feb 07 17:43:18 basically if you never initalized the rtc with any value that specific version of hwclock refused to work :D Feb 07 17:44:23 WizardNumberNext: because 4GiB at 2MB/s or something is a lot Feb 07 17:44:51 damn I would expect it to be a lot faster, then 2MB/s! Feb 07 17:45:12 then it is 1MiB/s, because same dat is written twice Feb 07 17:46:10 zeev: to finally make you happy: queen beecon is a tool frequently suggested for a low-effort way to run stuff on startup of X11/HD Feb 07 17:46:45 afaik it has a "run once" option Feb 07 17:47:14 theoreticaly it should finish almost hour ago! And I have only 1470MiB written Feb 07 17:47:20 in 8000 seconds! Feb 07 17:47:37 WizardNumberNext: did you set bs ? Feb 07 17:47:49 bs=1048576 Feb 07 17:47:49 because you know writing stuff to flash byte by byte is as silly as it gets Feb 07 17:48:14 to make it easy to write exactly 4GiB Feb 07 17:49:04 hm no idea iif there is a such a thing as a "too big" blocksize Feb 07 17:49:12 but too small defenitely exists Feb 07 17:49:41 merlin1991: with the n900's ram, a 4GiB blocksize would be too big Feb 07 17:49:58 bs=1048576 count=1 seek=$1, where it is in loop, which adds to $1, added sync on end of loop Feb 07 17:50:21 it might skip instead of seek - I am always confused and have to check it before using Feb 07 17:50:33 WUT? Feb 07 17:51:21 also, what's your source? Feb 07 17:51:35 what the hell Feb 07 17:51:46 should I write down my one-lie script to make it as clear as it gets? Feb 07 17:51:55 dd if=/dev/zero of=file bs=1m count=4096 Feb 07 17:52:33 just wondering, does it really make a difference increasing the swap on the n900? Feb 07 17:52:50 kerio: thanks :-) Feb 07 17:53:24 and don't use if=/dev/random Feb 07 17:53:48 hahaha what the hell, you're using random? Feb 07 17:53:56 that's... cryptographically secure Feb 07 17:54:06 honestly, i don't know where's the bottleneck in that Feb 07 17:54:27 either the writing, the generation of the cryptosecure data, or the constant write/close/open/seek Feb 07 17:54:33 MiB=0; while [ $MiB -lt 4096 ]; do dd if=/dev/zero of=file bs=1048576 count=1 skip=$MiB; sync; echo "written $number MiB"; MiB=$(($MiB+1)); done Feb 07 17:54:42 ....WHAT Feb 07 17:54:45 what the fuck Feb 07 17:54:48 what the fucking fuck Feb 07 17:54:56 DocScrutinizer05: can i swear? Feb 07 17:55:00 kerio in entropy! Feb 07 17:55:24 s/$number/#MiB/ Feb 07 17:55:34 WizardNumberNext: but... why? :s Feb 07 17:55:37 ~ping Feb 07 17:55:37 ~pong Feb 07 17:55:44 what why? Feb 07 17:55:50 plenty of things, actually Feb 07 17:55:57 but mostly the sync Feb 07 17:56:22 I want to test ext4 with different setting, so I need some other FS to do it Feb 07 17:56:54 you need what? to do what? Feb 07 17:57:14 sync - easy answer - to not create bottle-neck of cache being overused - I have seen it already many times, so sync helps to keep sustained speed Feb 07 17:57:45 WizardNumberNext: pretty please stop whining about slow writes to eMMC! Feb 07 17:57:49 and anyway for some reason I have issue with constant writting to eMMC - N900 reboots after some time Feb 07 17:57:56 WizardNumberNext: why? Feb 07 17:58:01 forgot to metnion sleep 1s there Feb 07 17:58:01 ah yes, the fscking watchdog Feb 07 17:58:08 WizardNumberNext: dd has option to do direct io Feb 07 17:58:29 conv=direct Feb 07 17:58:38 and you get no buffering Feb 07 17:58:59 also google for swappolube Feb 07 17:59:04 you think it would help - I can try (it does help with hdparm, so it could help with dd) Feb 07 17:59:16 I have swappolube Feb 07 17:59:25 but I do not use default settings Feb 07 17:59:34 I have more agresive settings Feb 07 18:00:15 * DocScrutinizer05 idly wonders if that now been ~yx or even something more screwed Feb 07 18:00:28 hahaha Feb 07 18:01:14 worst thing I dever did on a n900 was tar -xf on 2Gbyte tar.gz Feb 07 18:01:21 crashed somewhere halfway through :D Feb 07 18:01:37 and took the fat fs with it :D Feb 07 18:02:40 swappiness=20 dirty_ratio=80 dirty_bgd_ratio=20 dirty_expire=750 dirty_writeback=500 page_cluster=0 vfs_cache_pressure=100 laptop_mode=1 Feb 07 18:03:05 merlin1991 - same problem here - cannot seem to pass 1GiB mark with tar Feb 07 18:03:33 maybe I should use -xjf, but that is senseless as all data is already compressed Feb 07 18:06:54 got to the 1,1GB mark, it's slowing down a bi Feb 07 18:06:55 t Feb 07 18:07:00 it's down to 3.1MB/s Feb 07 18:07:15 now 2.9 :( Feb 07 18:20:59 WizardNumberNext: Feb 07 18:21:01 /etc/init.d/rcS:echo 4 >/sys/block/mmcblk0/queue/nr_requests || echo "ohnoes, hope it boots1" Feb 07 18:21:02 /etc/init.d/rcS:echo 4 >/sys/block/mmcblk1/queue/nr_requests || echo "ohnoes, hope it boots2" Feb 07 18:21:13 DocScrutinizer05: 4? :s Feb 07 18:21:28 ShadowJK advises making that much, much bigger Feb 07 18:21:53 well, depends on what you wanna do Feb 07 18:22:49 anyway, that's a dial to screw, when you wanna tackle that lockup/reboot issue during large writes to eMMC or uSD Feb 07 18:23:49 for bs=1m a 4 entries on wait seems fine for buffers Feb 07 18:24:12 and for bs=128K? Feb 07 18:24:21 just changed my script Feb 07 18:24:45 if on the other hand you have a bs=1, you probably want nr_requests = 500000 Feb 07 18:24:51 anyway I was using 64 there already, instead of default 128 Feb 07 18:25:32 64M is still a tad too much, for dd bs=1m Feb 07 18:26:58 assume I copy some big file (hundreds of MiB) from NFS to eMMC patition with ext4 (4KiB block-size I think - don't remeber). What you would propose then? Feb 07 18:27:00 aiui each request is one buffer holding one IO event's data Feb 07 18:27:43 and too many (large) buffers cause swap hell Feb 07 18:27:59 which in turn causes reboot Feb 07 18:28:05 my take on it Feb 07 18:28:06 block-size=4096 inode-size=8192 Feb 07 18:28:12 no warranties implied Feb 07 18:28:46 DocScrutinizer05: so I was thinking very right to lower it even more, then it is at the moment Feb 07 18:29:26 generally for speed use high numbers for nr_requests, to avoid reboots use very low numbers, particularly when using huge buffersizes Feb 07 18:31:54 so maybe adjusting dirty_expire and dirty_writeback would good idea as well Feb 07 18:32:23 if I should make up a rule of thumb, I'd say: nr_requests * buffersize < 20MB, and you should be fine Feb 07 18:32:39 but I do not really want to lower it much as I do not want cpu to be awaken too often Feb 07 18:33:02 err? Feb 07 18:33:10 I should be a lot belowe it right now Feb 07 18:33:26 128k*16=2M Feb 07 18:33:46 sounds good Feb 07 18:34:03 ask ShadowJK if you want to speak an expert Feb 07 18:34:04 well, i don't know about you guys, but 4294967295 bytes (4,3 GB) copied, 1982,34 s, 2,2 MB/s Feb 07 18:34:13 to me too - eats less mem and swappiness wuld be more effective, then it was Feb 07 18:34:26 and sadly, it errored out right at the end, because i'm on vfat and 4GiB is just too big Feb 07 18:34:28 :D Feb 07 18:34:40 1024+0 records in, 1023+0 records out Feb 07 18:34:49 LOL Feb 07 18:34:58 hahaha, then you should use 4095MiB instead of 4096MiB Feb 07 18:35:05 yeah, yeah, the file is almost completely written Feb 07 18:35:40 actually everything >2GB is problematic on vfat Feb 07 18:35:44 in fact, wtf did dd report? it's only missing one byte Feb 07 18:35:48 kerio it is afair exact size of biggest possible file on FAT32 = 4GiB-1MiB Feb 07 18:36:03 WizardNumberNext: clearly not Feb 07 18:36:10 4294967295 2013-02-07 19:33 test Feb 07 18:36:29 as I said - afair Feb 07 18:36:37 then it is 4GiB-1B Feb 07 18:36:53 a lot of system calls seem to fail on >2GB Feb 07 18:36:58 not too big difference , taking in account size of file Feb 07 18:37:06 hopefully unlink() worked Feb 07 18:37:10 shouldn't of Linux Feb 07 18:37:33 windows won't bite for sure Feb 07 18:37:38 using int instead of uint, or whatever Feb 07 18:37:46 in windows it is 2GiB-1B Feb 07 18:38:04 might be the reason and it seem to be valid Feb 07 18:38:51 recorder for example always stopped recording on 2GB .wav size Feb 07 18:39:28 probably some lib call using int<0 = error Feb 07 18:39:52 int32 to be precise Feb 07 18:39:52 could be Feb 07 18:40:14 some prevention, instead of recovery Feb 07 18:41:11 anyway, whatever I use ext4, not FAT32, so I can fill eMMC and would be still very damn far from limit Feb 07 18:41:30 well, it's pretty clear that my settings are better than your settings >:D Feb 07 18:41:43 you even cannot get such big HDD yet to able to create maximum size file on ext4 Feb 07 18:42:00 kerio: your settings? Feb 07 18:42:22 so writing one more byte to a file of size 0b0111 1111 1111 1111 1111 1111 1111 1111 will result in filesize 0b1000 0000 0000 0000 0000 0000 0000 0000 Feb 07 18:42:29 and remeber - nothing FAT related - I hate FAT only slightly less then NTFS Feb 07 18:42:39 8192 queue/nr_requests, 0 queue/iosched/slice_idle Feb 07 18:42:47 function call returns int < 0 Feb 07 18:42:49 and everything else stock Feb 07 18:43:31 resulted in 4294967295 bytes (4,3 GB) copied, 1982,34 s, 2,2 MB/s with a bs=4M when writing on vfat Feb 07 18:43:33 slice_idle might be the answer to many of my questions Feb 07 18:44:17 vfat isn't too slow, so that won't be huge difference with ext4, then Feb 07 18:46:18 last time I killed my system been when I tried to mv my mp3 collection from eMMC to uSD Feb 07 18:46:58 concurrent access is probably still as slow as a dead snail Feb 07 18:47:21 kerio: where is it in proc? Feb 07 18:47:30 I kinda do not use sysctl Feb 07 18:49:01 it's not in proc, it's in sys Feb 07 18:49:16 /sys/block/mmcblk*/queue/ Feb 07 18:49:30