#qi-hardware IRC log for Wednesday, 2013-07-17

pcercueihi wej 04:42
pcercueiis the path where GMU saves / loads its config modificable?04:43
whitequarkHowever, for those of you not familiar with the Pi, it comes with very limited hardware resources (including 512 MB of memory) which can be easily consumed by a [while] loop [in Bash] like the one I'm using.11:28
whitequarkby the way, an acquaintance of mine recently suggested another reason broadcom could be selling pis at a tiny margin11:51
whitequarkapparently if you have a few thousands of unliquidated chips (or early revisions, or slightly broken ones), it is somehow problematic to just dump them in a landfill11:52
lekerneldidn't they sell millions?12:35
whitequark"a few thousands" was a figure of speech12:39
whitequarkmillions. well, not sure12:39
ysionneau"can't sell those to real customers, so let's make a false open source design to at least squizz some internet buzz out of it" ?12:39
viricbut pis work, don't they?12:40
ysionneaumaybe they don't work at 85°C :)12:40
ysionneauor -4012:40
ysionneauor some part of the chip does not work12:41
ysionneaumaybe it's parts on the sides of the waffers12:41
viricbtw, anyone knows where to find a reasonably priced infrared camera+lamp, for night seeing? usb would be fine.12:41
viricysionneau: I've one and it always worked fine, for the limited work I gave.12:42
whitequarkysionneau: yeah12:42
whitequarkthe reason he said that: you know the ODROID thingy?12:42
whitequarkhardkernel.com etc12:42
virica friend reported many fails for odroid12:42
whitequarkhe worked quite a bit with the devices and it turns out that they indeed ship failed/early batches12:42
viricfor his odroid.12:42
ysionneauviric: indeed, but you cannot notice that some IP core not routed to the Rpi PCB (so feature not present on the Rpi) is not working12:42
ysionneaubecause you simply cannot test it12:43
ysionneauand you don't care12:43
viricysionneau: then it's fine for me. It's the same for two-core, three-core computers12:43
whitequarklol, someone actually sells three-core cpu?12:43
ysionneauyep, but they cannot sell the chip to a real customer12:43
viricoriginal 4-core processors where some core doesn't work12:43
whitequarkviric: yeah I understand how it can be12:43
ysionneauthey can only sell it soldered to some board with a very specific use case12:43
whitequarkbut never seen12:43
whitequarkysionneau: well, celerons *are* dumbed down pentiums after all12:44
viricThere are even bioses that allow *unblocking* those cores12:44
viricmotherboards with that feature. It has a name I can't recall.12:44
viricsome kind of "give the cores back, I don't mind if they don't work"12:44
ysionneauwhitequark: well yes, dividing the production of the same chips into several different parts is common for chip makers12:44
ysionneaulike AEC-Q100 chips, and non-automotive ones12:45
whitequarkyeah, that's generally what is done12:45
ysionneauAEC-Q100 are just those on the middle of the waffer, or just for some lucky reason pass the temperature tests12:45
viricso, for the pi, fine.12:47
whitequarkfor programmers who have the true C spirit: "I don't care if it works if it is fast!" :p12:47
viricyes, on that line :) but with hardware.12:47
viricPerformance boost!12:48
wpwrakprolog/lisp: i don't care of it works, as long as the experience is good :)14:28
whitequarkmissing the point ;)14:30
whitequarkI only recently talked with a self-described C programmer14:31
whitequarkhe had a program which only worked correctly when compiled with -O2 (implying he has really bad UB somewhere)14:31
whitequarkso what he did, could you guess?14:31
whitequarkadded "compile with -O2" to README14:31
whitequarkand (in my eyes) this attitude is too prevalent and too harmful. I don't care about minor issues like manual MM vs GC :p14:32
wpwrakUB = ?14:45
whitequarkundefined behavior14:46
whitequarkoverall, that would mean that his program works "correctly" (even if it does) purely by accident, and changing so much as a minor version of GCC would suffice to break it14:47
whitequarkbut I'm pretty sure you already know it14:47
wpwrakseems that he should try, in this order, -Wall, and then valgrind :)14:49
wpwrakno everything can be caught this way, but a lot14:50
whitequarkwpwrak: oh, sure, or -fsanitize=undefined14:52
whitequarkbut the problem is that he doesn't want14:52
whitequark"yeah, gotta build it with -O2, whatever, I don't care as long as it works"14:52
whitequarkand he defended that even after an explanation of what UB actually is14:52
Action: whitequark sighs14:52
wpwrakdoes he work on embedded controllers for battery management in the boeing dreamliner ? :)14:54
whitequarkno, on cloud storage stuff (I think), and plans to work on OS kernels if I'm not confused14:55
larscwell as long as it is the Windows kernel I'm fine with that14:56
whitequarkwindows kernel is actually not that bad14:56
whitequarksomething something working async I/O :p14:56
larsconly speeding up the inevitable14:57
whitequarkbtw wpwrak 17:01
whitequarkhave you seen Cyclone?17:02
whitequarkthat actually seems to have some slim chances you'd even try writing code in it :)17:02
wpwrakdo you think i obsess about pointers ? :)17:14
wpwrakalmost anything that i occasionally do wrong with pointers is easily found by valgrind.17:16
viricwpwrak: you should not use heap memory. :)17:23
wpwrakyou prefer arrays ?17:24
larscone big global state struct17:32
viricfor example :)17:35
viricstack is overrated too.17:35
viricI should back up luks headers...17:37
viriclarsc: is there any way I can see what my 'kworkers' do?17:38
virickwoerker/0:1H for example, does it indicate anything?17:38
larscdepends on the definition of 'do'17:38
viriccat /proc/PID/stack?17:39
larscno what do you mean by do17:39
larscdo you want to know what code they execute17:39
larscor when they execute17:39
viricI see them on 'top'17:39
viricdoes each PID have a specific task? or they pick tasks from queues?17:40
larsce.g. kworker:mmcqueue17:40
larscwill do mmc card related stuff17:40
larscbut there are also generic workers which just queue up 'random' things17:40
viric# cat /proc/4567/stack17:40
viric[<c002acb4>] msleep+0x24/0x3017:40
viric[<c040e0e0>] mmc_power_up+0xd0/0xd417:40
viric[<c040ef8c>] mmc_rescan+0x29c/0x3a417:40
viric^ this one appears on top often17:40
viricnamed "kworker/u:1"17:40
larscthat's a generic one though17:41
viric- I don't have any mmc card put in the slot.17:41
larscin the kernel there are so called workqueues17:41
larsca driver can create a custom workqueue17:41
larscbut there are also a couple of generic workqueues17:41
larscwhich are usually used17:41
larscunless your queue needs like a special priority or something17:42
larscthe generic queues are basicall first come first served17:42
viricwhat is that language of "/u:1" ?17:42
larscI don't know, but should be easy to find that out17:42
larsc1 is probably cpu 117:42
larscand u mabye unbound17:43
virickworker/u:0H, kworker/0:0H, kworker/0:1H, kworker/u:1, ...17:43
viricand what are "flush-8:0" for?17:43
larsc    if (pool->cpu >= 0)17:44
larsc        snprintf(id_buf, sizeof(id_buf), "%d:%d%s", pool->cpu, id,17:44
larsc             pool->attrs->nice < 0  ? "H" : "");17:44
larsc    else17:44
larsc        snprintf(id_buf, sizeof(id_buf), "u%d:%d", pool->id, id);17:44
larscfor flushing the fs cache17:44
viricH = high priority, ok. cpu:id.17:44
viricand u = without cpu assigned.    pool->id vs id, I don't know.17:45
viricthis sheevaplug crypto device isn't performing as I'd expect17:45
larscprobably not caused by the worker tasks though17:48
viricoh, Kconfig reads: "Currently the driver supports AES in ECB and CBC mode without DMA."17:49
Luke-Jrcan anyone recommend a JTAG cable with best free software support?19:52
viriccheck openocd.19:53
viricmaybe they tell about jtag adapters19:53
larscI'm not sure if I can recommend a FTDI, but usually it works19:54
viricthe sheevaplug has a ftdi jtag thing in it, and it's the only one I've used.19:58
Luke-JrI actually wrote a FT232R-based JTAG bitbanging interface.. maybe I can use that somehow19:59
Luke-Jrhmm, one of these is open hardware20:07
wpwrakthe Amontec JTAGkey works quite well. ftdi-based.20:55
wpwraksimilarly, the famous openmoko debug board is quite nice. also brings out serial and has a built-in usb hub. there, you even get the schematics.20:55
larscif you have enough spare flex cables20:56
wpwrakdrawback of the openmoko board: you need to add the headers for the cable yourself.20:56
wpwraklarsc: and a gta01 or gta02 :-)20:56
larscthose I have, a flexcable not anymore20:57
wpwrakand then there's of course the famous ubb-jtag :-) http://downloads.qi-hardware.com/people/werner/ubb/jtag/20:57
wpwrakall you need for that is one UBB, a few minutes of soldering, and a ben nanonote :)20:58
wpwrak(well, and resistors, cables, whatever connector plugs into the other side. the usual "small parts")20:58
wpwraklarsc: yeah, they're basically consumables :)20:59
wpwrakthe connector on the debug board as well. the gta0x side is fairly robust, though.21:00
wpwraklarsc: of course, real men (tm) use this: http://www.almesberger.net/misc/idbg/21:00
Luke-Jrhmm, the nanonote thing leaves most pins unconnected?21:01
Luke-JrI'm talking to an AVR32, but I'd like to have something I can use for any purpose21:02
Luke-Jrif I can justify buying a Nanonote, that'd be neat ;)21:02
Luke-JrI do have some GTA0x here somewhere21:02
wpwrakthe nanonote critter connects everything a xilinx fpga needs. not all devices are happy with so little, though. but then you still have two spare pins (in the xilinx layout). i could have added a serial console on them :)21:04
Luke-Jrhmm, https://sharism.cc/ is down? :/21:05
wpwrakthe ubb-jtag wasn't really meant to be a "real" solution. more a proof of concept. but of course, if you need only low speed, it works just fine. and you could always make an optimized driver that would then be MUCH faster than the /sys/class/gpio-based one that proof of concept uses.21:06
wpwraki think sharism.cc if dead for good. but try pulster.eu21:07
Luke-Jrany idea how many volts the openmoko debug board uses for JTAG? or can it autodetect?21:10
Luke-Jr&assuming I can still buy one of these somewhere :/21:11
wpwrakthe openmoko board is 3.3 V, no autodetect21:58
Luke-Jralso seems I'd have to pay $50 + import customs to get it "gratis" from some german store22:02
Luke-Jrwish I could just build one of the open hardware ones, but apparently chips will burn out if I try to solder them normally? :?22:03
wpwrakoh, interesting twist :) maybe mail him and ask if he sells them "normally" as well. and yes, customs always want their share :)22:30
wpwrakand no, if you solder chips normally, they don't burn out :)22:30
nutthey anyone, I just replaced a touchpad assembly for an hp pavilion dv6 laptop and had it up and running after replacing it for a couple hours. Then I rebooted to install updates and now it can't boot23:03
nuttI tried the recovery disc and it can't find an os23:03
nuttany ideas?23:03
Luke-Jrwpwrak: well, the $50 is shipping to USA and bank wire fee - so non-gratis would just be additional on top of it :/23:39
Luke-Jrwpwrak: is the solder thing something that has changed over the years? my father apparently had to stop using chips entirely when they stopped doing sockets, because of this; maybe at some point chips began to be more heat-tolerant?23:40
rohmore? nah23:42
rohthe only reason for solderchanges recent years was ROHS... basically a systematic ban on some metals like lead where possible23:42
rohbut thats industry-wide.. all consumer stuff. there are extemtions for some telco and network equipment also.23:43
rohthe moko debugboard is a quite basic ftdi2232 application. you can also solder one yourself. just keep all leads short and dont forget the proper caps ;)23:44
rohamontec also has similar boards (also in simplicity)23:44
rohhttp://www.amontec.com/jtagkey-tiny.shtml is quite similar if not identical pinout (for the jtag part, no serial output there)23:45
Luke-Jrroh: well, I mean changes to the chips so they don't burn out as easily23:47
Luke-Jrthe pinout of the moko debug board is actually wrong for me - I just prefer to get open hardware even if I can't build it myself >_<23:47
rohLuke-Jr: well.. wrong pinout in what way? for jtag?23:49
rohits standard 20pin arm afaik23:49
Luke-Jrroh: yes, my board has the 10pin design23:49
rohhm. and you also need some support in openocd afaik23:49
rohif not there yet23:50
Luke-JrI think OpenOCD already supports AVR3223:50
rohor whatever you want to use as gdbserver<->jtag connector23:50
Luke-Jrif not, I'll ignore it and write my own firmware upload thing23:50
Luke-Jror port it23:50
Action: Luke-Jr can handle software, just not hardware :23:51
rohhmmm..  i guess soldering an adapter for 20pin arm to avr32 jtag would be easier23:51
Luke-Jrthere seem to be 3-5 options for open hardware, just nobody selling them in the USA premade :/23:51
roheheh... 23:52
rohhttp://amtek.wordpress.com/2011/07/20/make-your-own-avr-jtag-debugger/ ?23:53
rohseems doable even for beginners. simple one-sided board. if somebody with a clue helps a bit it can be even done with kitchen-table etching23:53
Luke-Jrroh: not sure the interface is open (closed firmware) :/23:55
Luke-Jrso I may not have any way to use it23:55
Luke-Jr"To test if the board is working youll have to give 5V to the board by connecting a 10pin cable to one of the headers(JTAG or AVR ISP). Make sure you use a regulated 5V source or you might damage your USB port."23:56
Luke-Jrhuh? why not just use the USB power? :/23:56
rohor just buy a jtag-ice on ebay.. there is open sw on the pc afaik. the firmware is flashed onto the device23:56
Luke-Jrhmm, ebay23:57
rohi just heard from my 'off' that i should just use the jtag-ice in that box... hrrr.. i like this hackerspace23:58
rohjust no pic programmer around.23:58
rohi just soldered a GPIB interface and need to flash the fw23:59
--- Thu Jul 18 201300:00

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