irclog2html for blob on 2002.01.08

02:45:05SammyBZFlag: you here ?
02:45:20BZFlag was last seen on #blob 5 hours, 58 minutes and 25 seconds ago, saying: is the SF project mentioned in that doc? the diff lines included the .nl site... [Mon Jan  7 20:46:55 2002]
02:45:20Sammyibot: seen BZFlag ?
04:43:23Sammylunch
05:54:00Sammyis back ...
09:50:00erikmmorning
10:08:34Sammymorning erikm :)
12:35:49seletzhi all
12:46:12erikmhello seletz
13:56:01rsaUsneed some help porting blob
14:21:38sammy_wmsloading the pre2 ...
14:31:05erikmthinks rsaUs needed a little bit more patience
15:49:15seletzakira|work: hi
15:49:49seletztrying to understand freeware ATA flash driver from SANDISK
15:51:52prpplaguemorning seletz
15:52:04seletzhi
15:52:08prpplagueseletz: hows life as a "linux consultant"
15:52:18seletz:D
15:52:26seletzfine.
15:53:35seletzprpplague: just told a student which does his diploma here that I'll not accept his code he wrote over Xmas for my  driver.
15:54:08akira|workhahahaha
15:54:12akira|workhi btw :)
15:54:21akira|worksomehow that sounds oddly familiar to me :>
15:55:48seletzwell, i told him _before_ that he has to use some code we wrote before. He decided not to do so. I decided not to accept it.
15:58:47akira|work<g>
15:59:04akira|workluckily htat des _not_ sound familiar :)
16:01:12erikmseletz: lol :)
16:01:16prpplagueseletz: hey now, you'd probably laugh at my code....
16:01:23seletznah.
16:01:48prpplagueseletz: i don't have anyone looking over my shoulder very often, so i get lazy sometimes
16:02:21seletzHe is to become an degree in E-eng, but has _no_ C exp. Wrote #include <foo.c> and such. But partout wants to write a linux device driver. sigh.
16:02:31prpplagueman this aquapad hardware is nice, but the rootfs sucks
16:02:35prpplagueseletz: ohh
16:03:17seletzprpplague: now, i told him: lets first write a rinbuffer module, we'll need it anyway. Then test it un userspace.
16:03:38seletzprpplague: _then_ get "linux device drivers" and read it thrice :)
16:03:51akira|worklet me guess
16:04:00seletzprpplague: _then_ write a module with init() exit() open() and close()
16:04:05erikmseletz: give him a couple of good C books to read. K&R ANSI edition,  for example
16:04:21seletzprpplague: then read() which returns a scalar value
16:04:38seletzerikm: to late. he's to finish due 02-02
16:04:43erikmseletz: aiee
16:04:52seletzs/to/too/
16:05:46seletzwell, at last he wrote a driver which _seems_ to not lock up when multiple reads/writes open/closes happe.
16:06:03seletz(with help, tough. LOT of help)
16:06:36akira|workoh and when he's done u can start smacking me around a bit eh seletz :)
16:06:51seletzbut then: instead of finally using the rinbuffer module heself wrote, he hacked another (buggy) version of a ringbuffer directly into the driver. ARGH!
16:07:15akira|worki fear my coding style has gone too far into the "koch/keller/gampp" direction :>
16:07:38seletzakira|work: we'll fix that, that'll be np :)
16:07:45akira|workthouhgt so :)
16:08:25seletzcooling down slowly
16:08:37erikmtwaps around with Documentation/CodingStyle
16:08:50seletzerikm: in blob?
16:09:17erikmseletz: no, in the linux source
16:10:25seletzerikm: hmmm, is there really one? /me read this week's KernelCousin digest ....
16:11:04erikmfile /usr/src/linux-2.4.17/Documentation/CodingStyle
16:11:05erikm/usr/src/linux-2.4.17/Documentation/CodingStyle: ASCII English text
16:11:16seletzerikm: lol
16:11:26seletzerikm: yeah, i know this one.
16:11:42erikm"First off, I'd suggest printing out a copy of the GNU coding standards,
16:11:43erikmand NOT read it.  Burn them, it's a great symbolic gesture.
16:11:43erikm"
16:11:59seletzthats one of the best parts, really.
16:12:01prpplaguebelieves creating good/beautiful code is an art not a science
16:12:10seletzyup
16:12:32erikmprpplague: it's something you only learn by experience
16:12:59erikmseletz: did that student of yours have *any* programming experience?
16:13:31prpplaguei also believe that alot of it comes from being mentored by someone, books and howtos are good, but feedback from another is great
16:13:44erikmseletz: over here the students learn C++, which is a horror in itself if you get them for their masters thesis
16:13:49erikmprpplague: *nod*
16:14:08seletzerikm: well, nowadays they don't even make a difference between "C" and "C++". They use "C++" compilers and do procedural programming using cin and cout.
16:14:28prpplaguelives in east texas where the FHA gets more education money than the computer department
16:14:36erikmseletz: they happily feed a C++ istream to an ioctl() and wonder why the serial port speed doesn't change at all
16:14:51erikmprpplague: FHA?
16:15:01seletzerikm: yup. But the don't know of serial speeds either, at most.
16:15:29prpplagueerikm: aggreculture and homemaker stuff
16:15:39seletz_ugh_
16:16:01seletzprpplague: did you mean it or was that a joke?
16:16:59prpplaguenope it is true, my high school was in the middle of a cow pasture, 1/2 the day was devoted to agg technology!
16:17:53seletzprpplague: Only in texas i guess? :)
16:17:59prpplagueyep
16:18:08erikmseletz: I got a good student a couple of months ago. one that actually understood how a 4-way SMP machine worked and how you could take advantage of that
16:18:22prpplaguewhere i work is out on a 3800 ranch
16:18:54seletzerikm: man, i wish i had'nt chosen to go to an FH ....
16:19:23seletzerikm: "SMP" i heard of _never_ in any course.
16:20:02erikmseletz: SMP == Symmetric Multi Processing
16:20:16erikmseletz: dual PIII is a 2-way SMP machine
16:20:32seletzerikm: lol, no, i know what it means, but i didnt learn it at the FH.
16:21:08erikmseletz: the "symmetric" part is from the fact that both CPUs have the same view on the machine
16:21:51seletzerikm: ...
16:22:22erikmseletz: want me to continue on NUMA and ccNUMA? ;)
16:22:32seletzhmmm.
16:22:42seletzNUMA i know of, but "cc"NUMA?
16:23:03erikmNUMA is Non Uniform Memory Architecture. the cc part is Cache Coherent
16:23:08akira|workhmm that explains why i don't know wither what SMP is :)
16:23:15akira|worksame FH ;)
16:23:26akira|works/wither/either
16:23:46seletzakira|work: you forgot the last slash :)
16:23:50akira|work<gotta adapt to the local language>
16:23:52akira|workdang
16:23:59akira|work:)
16:24:25erikmseletz: an sgi Origin 2000/3000 is a ccNUMA machine, while a beowulf cluster with mosix is more like a NUMA machine
16:24:50seletzerikm: ok. /me thinks the "cc" part is the hardest ...
16:25:18seletzerikm: how does it work? do they have _one_ 2nd level cache?
16:25:22erikmseletz: yes. ccNUMA is much easier for the programmer, but the disadvantage is that there is *lots* of cache update/invalidate traffic on the bus
16:25:46seletzerikm: or do the cpus talk to each other?
16:25:48erikmseletz: no, each CPU has its own L2 cache
16:26:24seletzseletz: man, that gets _heavy_ traffic
16:26:45erikmseletz: the origin series is organised with compute boards with 2 CPUs with each their own L2 cache and local memory. sort of an SMP machine on a single board
16:27:32seletzerikm: well, SGI/IRIX then does a goot job on balancing data/processes to minimize cache failures?
16:27:39erikmseletz: two of those boards can be connected together to make a 4-way ccNUMA machine. if you want more CPUs, you have to add routers to create a hypercube interconnect
16:28:17seletzerikm: hmm, the interconnects have to be _fast_
16:28:22erikmseletz: and yes, the traffic is extreme
16:28:38erikmlooks for the Origin2000 hardware shortform
16:28:50erikmhmm. not here.
16:29:20erikmIIRC, the speed to/from a nodeboard is 3.2GB/s (that's byte, not bit)
16:30:01erikmthat is: for the origin2000. I don't know the specs for the 3000 series
16:30:22seletzerikm: i met the dev leader of IRIX/germany at the "linux tag" in frankfurt this spring. He told me that IRIS is not longer developed on, they'll use LINUX on their high-end machines. true or buzzwords?
16:31:06seletzs/IRIS/IRIX/
16:31:51seletzhmm, that's not too fast, tough.
16:32:12erikmseletz: true. they will support IRIX on their MIPS line (origin 2000/3000 and onyx 2/3), but they'l switch to linux for their IA64 machines
16:32:37erikmseletz: it's faster than their memory subsystem
16:32:43seletzhmm.
16:32:53erikmseletz: it's also a *lot* faster than any PC does
16:34:03seletzerikm: hmm. i think about an modern, say 800Mhz CPU. Then i think about a interconnect at full CPU speed, then 32bit busses.
16:34:22seletzor even 64 bit.
16:34:32seletzerikm: is it a serial link?
16:34:47erikmseletz: ehm, we're talking over here about off-board interfaces, not about the 10cm from CPU to chipset
16:35:13seletzerikm: yes, a lot of interference. It's actually HF.
16:35:26seletzerikm: ok. i give in. its fast.... :)
16:35:39seletzerikm: and the connectors i forgot also...
16:35:43erikmseletz: FYI: current CPUs just start to do 2GB/s from CPU to chipset :)
16:36:18seletzerikm: ok, i'm not a hw guru at last ....
16:36:22seletz:)
16:36:41erikmseletz: I'm not sure if sgi uses serial interfaces. I think they'll do, using some kind of LVDS interface
16:37:11erikmseletz: their old PowerChallenge/Onyx used a 256 bits wide bus
16:37:40seletzerikm: btw, my admin is ATM sruggeling with lotus notes to deliver emails. :)
16:37:55erikmseletz: /me did electrical engineering :)
16:37:57seletzerikm: they switched to notes for _email_
16:38:03erikmseletz: aargh! bloated notes!
16:38:27seletzerikm: the client takes 19MB RAM in no-op mode ...
16:38:38erikmseletz: friend of mine works at IBM and he *has* to use notes.
16:38:48seletzerikm: doesn't run anyway ....
16:38:56seletzerikm: oh my god.
16:39:12seletzerikm: guess he's to use the "calendar" a lut ...
16:39:18seletzs/lut/lot/
16:39:49erikmseletz: it's actually quite expensive for IBM. notes databases tend to become large, and it's about 17US$/MB/year to maintain them, so...
16:41:22seletzerikm: btw, i'm ATM reading through a freeware ATA driver from SANDISK. Its written for DOS (...), but neatly documented. Perhaps I'll get it running within blob.
16:41:56seletzerikm: they provide this code as an ex for their CF cards.
16:42:00erikmseletz: btw, how are the pcmcia/ide patches going?
16:42:20erikmseletz: but CF memory cards *are* IDE....?
16:42:32seletzyup
16:43:33seletzthe pcmcia code _seems_ to work, but i'm not sure. At last the cf card gets tetected, and the CIS tuples get read correctly. But the IDE part from brad does'nt work for me.
16:43:45seletzs/tet/det/
16:43:51erikmseletz: so I'd think that as soon as brad's IDE stuff works, it should work for CF as well
16:43:59seletzyup
16:44:15seletzThat's why i read through the sandisk code.
16:44:26erikmhmm. /me has a "normal" IDE disk connected to his LART
16:44:33seletzI think it's something with not resetting the CF card
16:45:05seletzOk, when we have an IDE layer, you can access the HD within blob
16:45:14erikmseletz: the sa11x0 PCMCIA driver might give you some ideas
16:45:19seletzThat's what im working on right now.
16:45:22erikm(in the linux kernel, I mean)
16:45:28seletzyup.
16:45:34seletzI read it.
16:45:47akira|workgonna eat some now... bbl
16:46:03seletzAs i said, i think the PCMCIA part is fine. Just the IDE part is missing/buggy.
16:46:03erikmis gonna eat with lots of co workers in about 30 minutes
16:46:17seletzakira|work: Maahlzeit!
16:46:18erikmseletz: did you try it on an assabet?
16:46:44erikmseletz: just to rule out potential hardware bugs
16:46:52seletzerikm: no, i don't have an assabet any more.  I used them up :)
16:46:57erikm:(
16:47:21seletzerikm: erm, i write this whole bunch of code to detect hw bugs.
16:47:27erikmlol :)
16:47:58seletzerikm: the linux pcmcia driver i ported to system3 hangs itself by locking irqs
16:48:26seletzerikm: but the hw guru says all's ok
16:49:03erikmhmm
16:49:28seletzerikm: so, i try to use the pcmcia/ide stuf in polling mode to test at last that the data/address busses are ok, and the sa1111 is ok etc
16:50:49seletzerikm: sadly enough the PT digital board doesnt deliver all lines for PCMCIA, so i have to fake them in the low-level sa1111 pcmcia stuff.
16:51:14prpplagueseletz: you little faker you
16:51:33seletzerikm: (voltage selection stuff mainly)
16:51:40seletzprpplague: lol
16:51:53erikmseletz: LART also doesn't, so I just skipped that
16:52:05erikmseletz: actually, we use only three lines for PCMCIA
16:52:23seletzerikm: pt system 3 only delivers 3V3, no real pcmcia, no 12V
16:52:33seletzerikm: no 5V also
16:52:35erikmseletz: lart also doesn't
16:52:46erikmseletz: lart can only do 5V
16:53:33erikmseletz: not that we ever *tested* the hardware. I wrote a driver for Aleph1, they're programming the CPLD on the PCMCIA board
16:53:47erikmseletz: want me to send you the driver?
16:54:02seletzerikm: would be nice ...
16:54:44seletzerikm: know what the BVD lines are for?
16:55:39erikmBattery Voltage Detect. original PCMCIA standard allowed you to put in battery cards, IIRC
16:56:03seletzok, i don't care of this either.
16:56:45erikmgenerates a diff
16:57:55seletzerikm: my pcmcia_get_socket_state() function fakes all lines whic i don't have: bvd*, vs_*
16:58:51seletzerikm: btw, i also applied the latest patch rmk sent on linux-kernel for sa1111 pcmcia stuff. I think it was back-ported from 2.5 ....
16:59:06seletzerikm: and changed my driver of course.
16:59:59seletzerikm: but no change: i always get "IRQ XXX locks up the system. disabled."
17:00:09seletzerikm: (forever and system hangs)
17:00:57erikmis glad that he uses riel's VM again in 2.4.17
17:01:07seletzerikm: "cardctl status" works quite nice, tough.
17:01:41seletzupdating kernel later ...
17:02:18erikmfor ARM I'm still on 2.4.13-ac5-rmk2
17:03:04seletzis using a 2.2.16 as a dev machine
17:03:35erikmbut for PC I use 2.4.17. with rik van riel's VM, because I believe his VM is conceptually better than andrea's
17:03:51erikmseletz: ehm you should *really* upgrade to 2.2.20...
17:04:09seletzerikm: i know :)
17:04:36seletzerikm: i'll get a new machine the next two weeks or so
17:04:55seletzswitches over to his own office in januar
17:07:14seletzerikm: just a strange idea: is it possible (for diag) to use irqs?
17:07:17erikmpatch is on its way
17:07:23seletzthanks
17:07:26erikmin theory yes
17:08:24erikmstarts cleaning up
17:09:02erikmcu tomorrow
20:15:06Russ|werkhmmm
20:15:16Russ|werk16 week lead on 0.068ohm resitors...
20:15:21Russ|werkthat's not happy
20:17:1530.7692307692308
20:17:15Russ|werkibot: 100/(1.3 * 2.5)
20:17:280.0307692307692308
20:17:28Russ|werkibot: 100/(1.3 * 2.5)/1000
21:17:360.333333333333333
21:17:36Russ|werkibot: (.1 * .1) / .03
21:18:220.39999999996
21:18:22Russ|werkibot: .3333333333 * 1.20
21:18:510.4
21:18:51BZFlagibot 1.20 / 3
21:19:36Russ|werkhmmm...you think thats odd maybe?
21:19:44Russ|werker, wait, nm
21:20:03Russ|werkI'm lazy
21:22:38BZFlagheh

Generated by irclog2html.pl by Jeff Waugh - find it at freshmeat.net! Modified by Tim Riker to work with infobot logs, split per channel and by date, etc.