#qi-hardware IRC log for Saturday, 2012-06-02

wpwrakheh, now the price hike is out :) let's see what happens02:17
wolfspra1lnothing will happen02:40
wpwraki guess i should try to organize some angry protests, just to prove you wrong :)02:50
rohprice-hike? ah.05:35
rohwell.. i was actually thinking about buying a efika mx from tuxbrain. they seem to be difficult to get over here05:35
rohwolfspra1l: are there no working watches anymore or why does the public mean to give 10mio to these guys?!05:42
wolfspra1ldon't understand05:44
wolfspra1lit's something people like, obviously05:44
wolfspra1lI certainly don't ask "why are 10 mio spent on this or that" - if you just look around the world a bit you could ask yourself that question for billions and billions spent every day on total bs05:45
wolfspra1lI do like how they pulled it off though, will watch whether they can establish strong/new sales channels.05:45
rohwolfspra1l: well.. it doesnt happen that often that one asks for 100k and gets 10mio05:47
rohjust too sad that i do not carry watches anymore ;)05:47
qi-botThe build has FAILED: http://fidelio.qi-hardware.com/~xiangfu/building/Nanonote/Ben/openwrt-xburst.full_system-20120601-0638 08:07
qi-bot[commit] Werner Almesberger: b2/: rename a bit too general "dump" to "dump_param" (master) http://qi-hw.com/p/eda-tools/b723c8808:57
qi-bot[commit] Werner Almesberger: b2/: new option -dCHARS to dump specific db; replaces use of -v (master) http://qi-hw.com/p/eda-tools/3ebac6808:57
qi-bot[commit] Werner Almesberger: b2/boom.c (main): rearrange to reduce indentation depth (master) http://qi-hw.com/p/eda-tools/be1277e08:57
qi-bot[commit] Werner Almesberger: b2/db.c: new function parts_dump to dump the whole parts database (master) http://qi-hw.com/p/eda-tools/224127608:57
qi-bot[commit] Werner Almesberger: b2/: move parts dumping from lang.y to boom.c and make optional (-dc) (master) http://qi-hw.com/p/eda-tools/f09e4b208:57
qi-bot[commit] Werner Almesberger: b2/test/Common: support multiple files of the same kind (!-c1, !-c2, etc.) (master) http://qi-hw.com/p/eda-tools/69701f408:57
qi-bot[commit] Werner Almesberger: b2/lang.y (part): show part ID in error message (master) http://qi-hw.com/p/eda-tools/ea5732308:57
qi-bot[commit] Werner Almesberger: b2/test/char: part characteristics database test (master) http://qi-hw.com/p/eda-tools/dd727b108:57
virichttp://vincentpetry.net/blog/?p=151 worth for openwrt/nanonote?11:44
Aylashould run out of the box on the dingoo11:54
viricah nice :)11:57
viricI even did not care to see if it is packaged already, sorry11:57
viricso in the dingoo you run even python games. nice11:58
DocScrutinizer51wolfspraul: which nfs version?12:09
wolfsprauldon't know12:42
DocScrutinizer51well, I'm just having fun with copying a complete image of $HOMe (~27GB) from my old laptop to a new one, via scp15:45
DocScrutinizer51recovers nicely (when mc is managing that copy via /#sh://) but suuuucks for speed15:46
DocScrutinizer51buzzword fishfs it seems15:46
viricthe network link is your bottleneck?15:49
DocScrutinizer51friggin fs operations via ssh are15:50
DocScrutinizer51~20 files/s15:50
viricwhy don't you use rsync for example?15:51
DocScrutinizer51facing another ~12h ETA, plus penalty for each file on source that's not readable by user15:51
DocScrutinizer51viric: because I'm a lazy asshole and just realized mc can connect to the ssh of source PC15:52
viricnot a bad reason15:52
viricat least mc provides ETA.15:52
viricI don't know any other tool that des15:52
DocScrutinizer51I nevertheless evaluated option a tiny bit before, and concluded rsync via ssh would do exactly the same, but without nice progress indicator and requesters etc15:53
viricI'm more of:        ssh pc "tar c /home" | tar x15:53
viricwell, rsync via ssh would not have a slowdown of fs operations through ssh15:53
viricrsync gives work to the remote rsync.15:54
DocScrutinizer51you *might* be right actually15:54
DocScrutinizer51initially I thought "whatever time it takes, I'm not in a hurry"15:54
DocScrutinizer51now I realize I have to observe this process for 12h, to eventually hit return to make a warning requester go away, whenever there's a file owned by root and not user, and not readable by user15:55
viricha. :)15:57
viricif you know the size of the data... use 'pv'15:57
viricssh pc "tar cp /home" | pv -S 23g | tar xp15:58
viricand you'll get an ETA. :)15:58
viricssh pc "tar cp /home" | pv -a -S 23g | tar xp15:58
viricthe averaging may work better for ETA.15:58
viricstore the 'tar' stderr, and you'll have a list of the errors.15:59
viricssh pc "tar cp /home" 2> tarc-errors.txt | pv -a -S 23g | tar xp15:59
DocScrutinizer51I used sftp from console now, logging in to source PC as root --- rocks!17:23
DocScrutinizer51from scroll spreed in xterm I estimate I'm up from ~20fileops/s to >500 now17:25
DocScrutinizer51well, this might have been the part where dest files already been existing17:27
DocScrutinizer51suddenly it dropped significantly17:27
DocScrutinizer51still, I'm confident it won't take longer than 2 hours now17:36
DocScrutinizer51mhm, 2 minutes for qi ML17:56
DocScrutinizer51/home/jr/.kde/share/apps/kmail/mail/.lists.directory/qi/cur/1292839187.22703.sSFlI                                                  100% 5131     5.0KB/s   00:00    17:56
DocScrutinizer51those 130,000 mails suck for sftp, even at an estimated rate of 100/s18:00
viricit depends on your filesystem too18:02
viricif it seeks once per letter... on a 7200rpm disk it will do little more than 100/s18:03
viricwith sftp you still have each file requested remotely, one by one.18:03
DocScrutinizer51/home/jr/.kde/share/apps/kmail/mail/.lists.directory/.openmoko.directory/.public.directory/cmty/cur/1326897176.26507.33lRp          100% 7094     6.9KB/s   00:00    18:07
viricgood luck.18:08
DocScrutinizer51I hope it'll be thru with mail eventually18:08
viricI don't think sftp will respect symlinks, dates, ...18:09
DocScrutinizer51-pr will18:11
DocScrutinizer51bottleneck CPU still, not HDD: Avg: 25.1% sy:  5.1% ni:  0.0% hi:  0.0% si:  0.6% wa: 35.4% 18:12
DocScrutinizer51which makes a total of 100% on one of both cores, plus a bit on other one18:13
DocScrutinizer51wa is it18:13
DocScrutinizer51device reads SDA (8.0): ~80/s max18:14
DocScrutinizer51usually way ,ower18:15
DocScrutinizer51well, no, that's actually average. max is 50018:15
DocScrutinizer51those 80 though could actually be the msg files read one by one18:17
DocScrutinizer51watching make runs is more fun ;-)18:21
DocScrutinizer51*yaaaawn* /home/jr/.kde/share/apps/kmail/mail/.lists.directory/.openmoko.directory/.public.directory/cmty/cur/1221116234.3915.HLS8k:2,S       100% 3816     3.7KB/s   00:00    18:24
Action: DocScrutinizer51 sobs for all the inodes18:25
wpwrakif you delete all the files, your inodes will enjoy freedom again :)18:26
DocScrutinizer51jr@halley:~> find /home/jr/.kde/share/apps/kmail/mail/ |wc -l18:27
DocScrutinizer51took an amazing ~5s to complete, despite this sftp monster running concurently18:28
DocScrutinizer51YAY OM-kernel ML18:29
DocScrutinizer51through with community18:29
DocScrutinizer51anyway mail takes >30% of the files to copy18:30
DocScrutinizer51number of, not volume18:30
DocScrutinizer51\o/ internal18:32
qi-bot[commit] Werner Almesberger: fisl2011/Makefile (DL): Atben_atusb_prod_03.jpg was missing (master) http://qi-hw.com/p/wernermisc/8418a1918:43
DocScrutinizer51and stalled >:-(18:43
DocScrutinizer51fsck that d-link dir615, which obviously has problems with occasionally dropping ARP table or whatever18:48
DocScrutinizer51or firewall/NAT hangups18:49
DocScrutinizer51fuuuuuck, sftp won't resume, I.E. will start from very first file again, ignoring all the stuff already copied. Worse: rsync does that too :-o19:43
DocScrutinizer51seems rsync hold a local file where it keeps a list of files already synced, and it doesn't care at all about actually existing real files on destination19:44
DocScrutinizer51or maybe not19:47
DocScrutinizer51already at "xfer#25322"19:47
DocScrutinizer5144k done 61k left19:55
DocScrutinizer51left 1900 of 17100020:23
DocScrutinizer51and friggin rsync is way slower than sftp: while sftp maxed out a ~11MB/s on large files, rsync gets to 3.xMB/s when lucky20:24
DocScrutinizer51and the WTF: it's not been dir615 but the sis190 kernel module of source machine. Had to modprobe-cycle it to make the NIC work again20:25
DocScrutinizer51"it"=the lockup20:25
DocScrutinizer51sent 8096479 bytes  received 19742271984 bytes  2132984.34 bytes/sec22:21
DocScrutinizer51total size is 28602138243  speedup is 1.4522:21
DocScrutinizer51rsync error: some files/attrs were not transferred (see previous errors) (code 23) at main.c(1530) [generator=3.0.9]22:21
wpwrakmake a tar and scp it ? or would that be too easy ? :)22:23
wpwrakextra points for rsyncing the tar, since this will let you restart22:23
DocScrutinizer51make a tar of 28GB, on a disk that has 300MB free?22:24
wpwrakdelete all the porn first :)22:24
DocScrutinizer51well, seems it's thru now. but now I have a dir /home/jr/halley with lots of stuff in it. And a dir /home/jr/halley/home/jr, with all the stuff in it again :-S22:28
wpwrakthat's called a "backup" :)22:29
DocScrutinizer51could you tell from the cmdline >> rsync -vaRzxP --fake-super  root@ ~/halley/jr/  << which is the real stuff now?22:29
wpwrakmy bet would be on the subdirectory22:30
wpwrakbut you can just move it out. then do a  du -sh  to compare size22:31
wpwrakor diff -r if you're bold and they should be the same :)22:31
DocScrutinizer51already thought abizt du -sh, but thanks for the moving suggestion22:31
DocScrutinizer51jr@HaleBopp:~/halley/jr> du -sh /home/jr/halley2/home/jr/; du -sh /home/jr/halley22:36
DocScrutinizer5128G     /home/jr/halley2/home/jr/22:36
DocScrutinizer5113G     /home/jr/halley22:36
DocScrutinizer51wpwrak: thanks22:36
DocScrutinizer51it seems like the destination argument on that rsync cmd took no effect at all22:37
DocScrutinizer51anyway 28G is what I expected22:38
wpwrakyou could du a find with md5sum on source and origin to be 100% sure22:39
wpwraks/du /do 22:39
wpwrakcd /home/jr/halley; find . -type f -exec md5sum '{}' \; >~/output22:41
wpwrakthen do a  sort out1 out2 | uniq -u22:41
DocScrutinizer51what for?22:41
rjeffrieswolfspraul expresed interest in making hos own PCBs. this article is not bad. http://hackaday.com/2012/06/02/pcb-manufacturing-tutorial/22:41
wpwrakthat will allow you to see if a) all the files made it and b) if they were copied correctly22:42
rjeffriess /hos/his/22:42
DocScrutinizer51this will take longer than doing another rsync22:43
wpwrakno. worst case, it takes just as long :)22:46
DocScrutinizer51friggin idiot I am, giving -R parameter without thinking22:47
wpwrakbetter than --remove-source-files :)22:50
wpwrakrjeffries: seems reasonably up to date. the etchant tank is a bit overkill, though, unless you plan to do monster boards22:52
rjeffrieswprak did you see [not talking about this artcle] where one person made a spray chamber so he could etch boards quickly with no muss or fuss? amusing.22:53
wpwrakhe seems a bit confused about what 3D printers do, too22:53
DocScrutinizer51HAH, now the benefits of rsync hit22:55
DocScrutinizer51+mwDGO1eWKyOiCJnpYZVFtUcItXrsync: readlink_stat("/home/jr/.gvfs") failed: Permission denied (13)22:55
DocScrutinizer51rsync error: some files/attrs were not transferred (see previous errors) (code 23) at main.c(1530) [generator=3.0.9]22:55
DocScrutinizer51second run was "lightning fast"22:56
DocScrutinizer51nfc what's this >>key_read: uudecode AAAAB3NzaC1kc3MAAAEBAP8Bo+QhQ...<< foo at beginning though22:57
DocScrutinizer51friggin 7 lines of uuencoded key, WHY oh WHY?22:57
wpwrakrjeffries: the chemistry link is interesting, though: http://www.stephenhobley.com/blog/2011/03/02/still-messing-with-forces-i-dont-understand-the-formula/22:58
wpwrakreplacing the muriatic acid (HCl) with household chemicals22:58
wpwraki just hope those people never get their hands on higher grade peroxide. the graveyard of overly adventurous pioneers is already quite full :)22:59
DocScrutinizer51he could rub some marshmallows on it and let ants eat away the copper22:59
DocScrutinizer51~curse gnome and all g*23:03
infobotMay you be reincarnated as a Windows XP administrator, gnome and all g* !23:03
wpwrakthat could work. probably a bit slow, though23:03
wpwrakfaster than just waiting for proton decay, though :)23:04
DocScrutinizer51wpwrak: would you be interested in getting involved into a bit of kernel development discussion?23:09
wpwrakthe kernel is vast. what area of it ?23:11
DocScrutinizer51there's this guy FreemanGordon over there at #maemo-ssu, and he investigated for almost a felt year now how to make thumb interworking working on N900 which has an OMAP with a silicon erratum that fscks up the branch prediction when you (mem)swap a thumb branch for a ARM branch, or sth like that23:12
DocScrutinizer51actually there are (at least) two silicon errata, and common urban legend is the fixes for those two are mutually exclusive23:13
wpwraknice ;-)23:13
wpwraksounds like a topic for the linux-arm list ?23:13
DocScrutinizer51freemanG now claims he made it work, but none of the "experts" is willing to listen23:14
DocScrutinizer51which is kinda expected, since that OMAP3430 is rather old silicon anyway23:15
wpwrakdid he try that list ? i'm no longer on it, so it can't check quickly23:15
DocScrutinizer51and not much to earn on fixing kernel stuff to make thumb interworking fixed for that particular chip - at least unless you're a fan of N90023:15
DocScrutinizer51list not yet afaik23:16
wpwrakthumb is for running closed binaries ?23:16
DocScrutinizer51well, thumb is a smaller (16bit) alternative opcode set23:18
DocScrutinizer51or what's been the question?23:18
DocScrutinizer51his interest is in reducing RAM footprint of certain libraries like Qt etc23:19
DocScrutinizer51since N900 is in a permanent swap hell23:19
wpwrakah, i see23:20
DocScrutinizer51thumb instruction set is supposed to be ~30% smaller than ARM, but possibly slower in execution23:20
wpwraki think (conceptually) what thumb is. just never saw anyone actually use it23:20
DocScrutinizer51you usually don't even realize I guess23:21
wpwrakyou mean s/instruction set/code size/ ?23:21
DocScrutinizer51err yep23:21
DocScrutinizer51e.g. STE modem is all thumb afaik23:22
wpwrakevil blobs :)23:22
DocScrutinizer51do you think you could give some guidance what to post to linux-arm ML23:24
DocScrutinizer51*my* problem with FreemanG is that our ideas of patch validation differ vastly23:26
wpwraki'm only an occasional visitor there myself. haven't posted there since openmoko. so i wouldn't have any specific recommendations. just do the usual - describe problem, solution, and see what happens23:26
wpwrakpatch validation ?23:26
DocScrutinizer51somehow I think such a patch for a silicon erratum has to be validated somehow23:27
DocScrutinizer51to be effective23:27
DocScrutinizer51and actually target the problem comprehensively23:27
wpwrakwell it needs testing, like anything else. if the erratum is version-dependent, it may need a switch, too23:27
DocScrutinizer51esp since here we got 2 bugs in chip, with common notion (seemingly based on rumour) being that the already esisting patches are mutually exclusive23:28
DocScrutinizer51and my approach is based on building a testbed that 100% reproducably and traceably invokes the problem23:30
DocScrutinizer51while FMG's approach is "it's not possible to write such testbed. but ubuntu is all thumb and it runs on my N900 since 3 weeks now. That's proof enough"23:30
DocScrutinizer51maybe slightly different rationale and wording, but the general idea AIUI is like that23:32
DocScrutinizer51I'm affraid with this approach of his, he won't see benevolent reviews on linux-arm ML23:34
wpwrakhe does seem to have a certain, erm, assertive style, yes23:35
DocScrutinizer51that's where your comments are welcome23:35
wpwrakjust tell him the key guys on linux-arm have had daily exercise as top of the foodchain bullies for the last 20 years. he's not going to out-stubborn them ;-)23:37
wpwrakwell, some of them have. others are gentle :)23:37
DocScrutinizer51(version dependent) one of the fixes - published by TI itself - is by setting some bit in auxiliary control register of CPU, to flush branch prediction on every context switch (whatever that means, now that I come to think of it)23:37
DocScrutinizer51obviously there is a performance penalty in flushing branch prediction frequently23:38
wpwrakif it's just per context switch, that should be negligible23:39
DocScrutinizer51usually yes23:39
DocScrutinizer51then there's another bug that kicks in on that (or not), related to commands that flush branch prediction which may result in some nonsensical CPU action (like branching to a destination offset by an unrelated obscure register)23:40
DocScrutinizer51and patch for that is to either not use the flush-branch-prediction-queue instructions, or to initialize that random bogus unrelated register to 0, so it won't affect stuff23:41
wpwrakmaybe whitequark could write a thumb decompiler for you ;-) then you could recompile the whole mess with something that's actually been tested :)23:41
wpwrakso zero-initialize and flush, to kill both birds ?23:42
DocScrutinizer51all disclaimer: IIRC & AIUI23:42
DocScrutinizer51yeah, and that's where fun starts: so far nobody was able to do that (except Nokia), as you can access that auxiliary register only in secure mode / context. The secure context gets irreversibly locked during bootloader execution23:44
qi-bot[commit] Werner Almesberger: b2/test/curr: currency exchange test (master) http://qi-hw.com/p/eda-tools/0ba0abb23:44
qi-bot[commit] Werner Almesberger: b2/lang.y (provider): don't allow provider to be redefined (master) http://qi-hw.com/p/eda-tools/994ed0023:44
qi-bot[commit] Werner Almesberger: b2/test/prov: provider database test (master) http://qi-hw.com/p/eda-tools/4184c0123:44
qi-bot[commit] Werner Almesberger: b2/test/Common (run_boom): place inventory after parts and currency (master) http://qi-hw.com/p/eda-tools/e62bb0623:44
qi-bot[commit] Werner Almesberger: b2/test/inv: inventory database test (master) http://qi-hw.com/p/eda-tools/cc5d9a923:44
qi-bot[commit] Werner Almesberger: b2/: move substitutions dump from parser to boom.c (master) http://qi-hw.com/p/eda-tools/3027dac23:44
qi-bot[commit] Werner Almesberger: b2/: insert a virtual empty hierarchy if test input starts with other file (master) http://qi-hw.com/p/eda-tools/d1593b623:44
DocScrutinizer51FMG now claims he found a "BIOS call" to write to that register from normal linux context23:44
wpwrakOMG :)23:45
wpwrakhow can the BIOS write to an "irreversibly locked" register ?23:45
DocScrutinizer51it's basically like a usual age old sw irq23:45
DocScrutinizer51"BIOS" = ROM bootloader code here23:46
wpwrakso it depends on the location of the executing code ? because it doesn't seem to be locked. else it wouldn't matter what you do once linux runs23:47
DocScrutinizer51AIUI there's a way to trigger an exception that switches state of CPU to secure context and same time jumps to the "BIOS" code - just like good old INTxx23:47
DocScrutinizer51it's just the RING0, RING-N concept23:48
wpwrakah, i see. well, if that register doesn't get changed by anything else, then he can put that magic code in the platform-specific initialization23:48
wpwrakand still flush branch prediction on each context switch23:49
DocScrutinizer51yep, probably that's what he does, or plans to do, or whatever23:49
DocScrutinizer51I not *completely* wrapped my head around all that yet23:49
--- Sun Jun 3 201200:00

Generated by irclog2html.py 2.9.2 by Marius Gedminas - find it at mg.pov.lt!