irclog2html for blob on 2001.12.19

08:06:05SammyBZFlag: when I mknod inside the rootfs , and after reboot the device node can't be safe , why ?
08:07:03BZFlagSammy: re make /dev a ramfs (normally) so that permissions etc can be changed without excessive flash writes.
08:07:39BZFlagreal path is /boot/var/dev/ which get's copied to /var/dev/ on boot where /var is ramfs
08:08:01BZFlagsame is true of any files in /boot/var
08:15:41SammyBZFlag: sorry forget tell you one thing , that rootfs is in flash (Puppy), can you answer me again , plz ?
08:22:09BZFlagyes, the rootfs is, but if you are using rootfs /dev is a link to /var/dev and /var/dev (after boot) is ramfs
08:22:27BZFlager if you are using the buildroot to get your fs.
08:23:18BZFlagor did you say you were using devfs? if so I'm not sure how to make it persist. same method should work though, have a /boot/dev and copy it over _after_ mounting /dev
08:23:58Sammyyeap ,I use devfs
08:25:03Sammybut onething bad is in my /boot don't have sub dir /dev :(
08:27:07BZFlagmake one
08:27:59Sammyok, I try that
10:47:54Sammyhi erikm :o
10:49:01Sammyiskra ?
10:49:11Sammyhave address ?
10:49:19Sammygoogle ...
11:10:17Sammydinner time ...
11:10:34Sammyerikm: lunch ?
11:10:44erikmyeah, just got a SIGLUNCH
11:47:57Sammyerikm: strange dev ...
11:48:49SammyI use devfs and mknod at /dev , then after I reboot , that's node I made was gone ...
11:48:52Sammywhy ?
11:51:22erikmdoesn't do devfs
11:51:37erikmIMHO devfs is seriously broken mean devfs broken or my jffs2 image broken ?
11:53:33erikmI consider devfs broken by design
11:55:31Sammyerikm: do you think devfs is good for big jffs2 file to use ?
11:55:48Sammyor you have better suggest ?
11:56:44Sammybecause sometimes in my /dev after download into flash , and find out lost some device ...
11:57:23erikmthat's because /dev lives in RAM (more specifically in the page cache) if you mount devfs over it
11:58:18Sammyso..... don't mount  ?
11:58:45Sammyjust support and don't mount on /dev ?
11:58:47erikmor create the device nodes you want after devfs has been mounted
12:01:02Sammy|o| , but this is inconvenient
12:01:28Sammyif I need fb or ts or etc ...
12:02:50erikm/dev/fb should be created by devfsd
12:03:34erikmanyway, /me reboots into 2.4.17-rc2, brb
12:18:57seletzhi all ...
12:19:23Sammyhi seletz :)
12:19:42Sammyerikm: I use this word ( inconvenient ) correct ?
12:20:16erikmhello seletz
12:20:52erikmSammy: which word? I just lost the xchat history when I rebooted into 2.4.17-rc2
12:22:13Sammy[20:03] <Sammy> |o| , but this is "inconvenient" <<= correct ?
12:23:49erikmSammy: if you mean "giving trouble" , or "giving annoyance"  it's correct
12:24:20Sammyok :)
12:33:54seletztrying out xchat :)
12:43:21Sammylater guys :)
12:43:30Sammygo home ...
12:46:24seletzsammy_wms: bye
13:02:13erikmlooks at the jffs2 patch
13:34:11seletzerikm: ahhh! at last some CRC32 function!
13:37:14erikmseletz: yeah, but broken
13:37:28erikmseletz: it does table lookups
13:37:37erikmseletz: which makes it *slow*
13:37:53seletzerikm: well, this patch looks likes some more general
13:38:15erikmseletz: remember that we run with dcache disabled, so every memory access costs you about 60 CPU cycles.
13:38:26seletzerikm: what do you think, could it be integrated with the pcmcia/ide stuff, once it works and is tested?
13:38:51erikmseletz: the icache is however enabled, so computing every value makes it *much* faster
13:39:03erikmseletz: I'm integrating it right npw
13:39:03seletzerikm: _ouch_
13:39:07erikmnow, I mean
13:39:48seletzerikm: normally lookups are a good thing, tough.
13:40:53erikmseletz: quite the contrary
13:41:45erikmseletz: it all depends on cache size/speed/latency, memory size/speed/latency, CPU speed, and amount of work to compute the value
13:42:06erikmseletz: and in this case it really makes sense to compute the CRC value
13:42:40seletzerikm: yes, when one has a setup that has a big speed penalty on mem access then lookup tables are not the best thing to do.
13:43:23seletzerikm: _and_ when computing values does not take too long :D
13:44:08seletzerikm: _yuck_ that table is _BIG_
13:44:43erikmseletz: yeah, 1K of data. that's serious cache poisoning even with dcache enabled
13:46:10erikmerikm: but I've done several program speedups in my group by just doing a proper analysis of the algorithm
13:46:13seletzerikm: :D comments are a fine thing ... (jffs2.c)
13:47:16erikmseletz: lol :)
13:47:29seletzerikm: well, the best speedup is archieved normally by using a better algorithmus (and implementing it correctly).
13:48:13erikmseletz: *nod* the biggest improvement we got was from an algorithm that used 200ms for processing an image on a PIII 800 to 15ms on a PII 450
13:49:05erikmseletz: we didn't have a PIII to test it on, but we thought the slower PII was already good enough (they needed < 40ms/image)
13:49:32seletzerikm: qed, so to say. :D
13:52:28seletzerikm: well, scanning through the code i see some "conding horrors" which when i would find/see them in reviews done here would result in quite some, erm, rewriting.
13:53:01seletzerikm: but i must admit that in the linux community the rules are somewhat different.
13:55:22erikmseletz: there are some coding errors, but you also have to remember that linux uses the rules as written down in Documentation/CodingStly
13:57:07seletzerikm: well, ok, we had some arguing about that. I already gave in and I try to use the common style, which makes very much sense.
14:00:30seletzerikm: thats about style. But i was talking about code quality. By quality i mean maintainability, _not_ the functionality, and not the coding style. Stlye does matter, tough, but the linux coding style is ok, IMHO.
14:01:48erikmseletz: using a consistent coding style helps a lot in maintaining the code. there was a large discussion on lkml about that a week ago. it's nicely summarised in the latest KT
14:03:09seletzerikm: ok, I'll read it.
14:04:50seletzerikm: I'ts just sometimes i can't get rid-off my pre-linux SW development work. I keep trying to get the best out of both worlds ... :D Never mind.
14:07:21seletzerikm: lol! KT is _good_
15:21:33erikmdeciced that partition table support should go first
15:21:40erikmdecides, I mean
15:21:46prpplaguemorning erikm
15:22:00erikmhello, prpplague
15:22:29prpplaguewell the company i work for has agreed to pay andersee to fix the uclibc support for c++
15:23:06erikmprpplague: that's good news!
15:23:30prpplagueandersee thinks he can have it finish by the end of jan
15:24:12erikmprpplague: I really applaud your management for such a wise decision
15:25:26prpplagueerikm: no too big of a decision "yes sir, we can use these modification very extensively, and we can get them dirt cheap!" "ok do it"
15:26:03prpplaguemorning jeffrey
15:31:45jeffreyHi prpplague
15:32:32erikmprpplague: heh, we did the same a couple of years ago. either hire a programmer to a) first study the linux kernel and b) start writing a device driver for a PCI I/O card, or pay roger wolff to do it instead. the latter was *much* cheaper
16:08:03sammy_wmshello all ...
16:10:38prpplagueerikm: i'm hoping i can convince them to spring the $$ for pthreads as well
16:10:58erikmheh, good luck :)
16:13:23sammy_wmsis that any point  need to be careful with handle touchscreen support ?
16:13:41sammy_wmsbeside need to know what chip to use ..
16:13:57sammy_wmsUCB1300  yes
16:14:49sammy_wmsor if make kernel have choose , then it's will support quite well ?
16:16:46sammy_wmsany suggest ? erikm ? prpplague ? Russ ? seletz ?
16:17:28erikmyou have to make some minor changes to the kernel driver to get it to work on your platform
16:20:25sammy_wmserikm: like ....what ?
16:20:37erikmIRQ, GPIO, etc
16:21:17sammy_wms||o|| remember that ...
16:21:43prpplaguesammy_wms: sorry sammy i'm just now getting into the lcd/ts interfacing myself
16:23:15sammy_wmsit's ok , I also like a newbies on this level too ,
16:24:27sammy_wmsso if you got any problem that I can help you , I'll do the best I can
16:26:02sammy_wmsHope someday I can put something good use stuff on blob, kernel ... etc
16:50:56sammy_wmsrain rain rain alllllllll day ............. :(
16:51:25erikmrain with occasionally some snow over here
16:54:32sammy_wmsI'd better like snow then rain ...
16:55:04sammy_wms(the voice come from someone never touch the snow )
16:57:34sammy_wmsbecause here doesn't have any chance to get snow ....
17:03:17erikmthis is *scary*
17:03:41erikmI mean the way the kernel parses bootldr partitions
17:08:54sammy_wmserikm: as I know that iPAQ when bootldr define the partition , even the kernel MTD has define different , but still choose what the bootldr say ? right ?
17:09:36erikmsammy_wms: that's right. if there is a partition table, the kernel will prefer that one over the default one
17:11:30sammy_wmsis that good ? or jsut like blob partition match with kernel MTD partition is better ?
17:13:10erikmsammy_wms: I don't think it's good, but if there is no partition table in flash, anything can happen
17:16:45sammy_wmsbtw need some sleep , night erikm & all :)
17:17:02erikmnight, sammy
17:53:30erikmconsiders trading in strcpy() for strncpy()
17:56:47erikmthinks that having strncpy() will force you to think about *what* to copy
17:56:58erikmwhich is a Good Thing [tm]
18:27:25Russ|werkI just wish it was a true limiting function, and not just a non-null terminated string function
18:29:26erikmRuss|werk: I'm just coding up a strncpy() function
18:30:25erikmRuss|werk: so you'd like to have the destination string null-terminated if n < strlen(src) ?
18:30:58erikmdoesn't think that's a good idea
18:34:11Russ|werkit'd be nice if there was a standard str copy that did that
18:34:20Russ|werklike strlcpy (l for limit)
18:34:54erikmRuss|werk: well, for blob, I'd rather stay with the description from the manual pages. principle of least surprise
18:34:55Russ|werkthen you wouldn't need a bunch of strncpy(value, buffer, 25); \ value[24] = '\0';
18:36:00erikmRuss|werk: make it a macro :)
18:42:56erikmalso removes strcmp()
18:45:07Russ|werkthat would make my code harder to read, so I'd rather just use the kludge
18:46:03erikmRuss|werk: I'll make it an inline function
18:46:37erikmRuss|werk: it makes sense, it's as safe as strncpy().
18:46:55erikmRuss|werk: but strcmp() could potentially run forever
19:20:59Russ|werkerikm: not to nitpick...
19:21:19Russ|werkstrncmp(argv[1], "230k4", 5), should technically be strncmp(argv[1], "230k4", 6) (and etc)
19:21:32erikmyeah, you're right
19:22:50Russ|werkyou could always make a macro strequ(a, const_str) strncmp(a, const_str, sizeof(const_str));
19:23:06Russ|werkalthough technically, strncmp will never burn you on const strings anyway
19:25:03erikmtoo late, I already changed everything in main.c :)
19:25:27erikmheya, BZFlag
19:31:02Russ|werkit'll always stop at the NULL in the const string
19:32:00erikmRuss|werk: I'm currently removing strcpy() in favour of strncpy(). is there any maximum length on the kernel cmdline in the param_block?
19:32:41erikmRuss|werk: if not, I'll put it at 1024 in linux.c and put a NOTE comment just above it
19:33:43erikmhmm, wait.
19:34:00Russ|werklemme see, its been a semester
19:34:10erikmsame for me :)
19:34:44erikmwait, the maximum size is COMMAND_LINE_SIZE as defined by the kernel
19:34:57erikmin include/asm-arm/setup.h
19:41:01Russ|werkya, it would just be that
19:41:35Russ|werk(altough it could technically be longer in the param block (should be limited by the utility for making the param block)
19:42:04erikmit wouldn't make sense, the kernel can't take any longer
19:42:46erikmat the time the ATAG parser runs, kmalloc() isn't yet initialised, so the kernel can only copy fixed size strings
20:02:48erikmRuss|werk: yeah, strlcpy() is quite neat :)
21:13:19prpplagueerikm: besides the tuxscreen, what has been the least expensive arm based board you seen available?
21:14:49erikmprpplague: the ipaq h3600 seems to be quite reasonable. sammy got one for $279
21:18:43prpplagueerikm: what is the creditlart price going to be?
21:20:00erikmprpplague: I really have no idea
21:20:41erikmprpplague: I even wonder if it will ever get there. the commercial consortium is a bit paralised
21:21:24prpplagueerikm: hmm what seems to be the problem?
21:22:05prpplagueerikm: the reason i ask, this guy over at earthlcd said on the phone they were producing a board based on the arm720t core, target price $99
21:23:39erikmprpplague: I'm not sure what really is the problem, but I guess part of it is due to the recesion in the US
21:25:05Russ|werkI'm really hoping that the recession starts to end around may
21:25:25Russ|werkso all the companies that laid off their old engineers are looking for new ones
21:25:58erikmRuss|werk: you'll graduate in may?
21:26:57prpplagueBZFlag: we accepted andersee's quote on the c++ support for uclibc
21:27:07prpplagueBZFlag: going to have to wait for pthreads
21:27:39Russ|werkerikm: yup
21:27:40prpplagueRuss|werk: everyone is talking about recession, but aren't seeing it here
21:28:06prpplagueRuss|werk: we just hired 5 more ppl, and the entire month of jan is booked with installations
21:28:08Russ|werkprpplague: it looks like a ressecion, smells like one, and greenspan says its one
21:28:17prpplagueRuss|werk: lol
21:28:22Russ|werkprpplague: from looking at numbers, they say it started around last may
21:28:25erikmprpplague: it's not everywhere. there is still a lot to do in embedded stuff, but telco is a bit less
21:28:42Russ|werkprpplague: motorola annonced today they are laying off 8.4%
21:29:36prpplagueRuss|werk: ok point taken, alot of ppl and out of jobs right now, but i think that it'll bounc back soon
21:30:33Russ|werkI hope a lot of it is artifical economic slow down (fears, etc)
21:30:50prpplaguei think thats alot of it
21:31:01prpplagueis not an economic guru
21:31:34Russ|werkI blame clinton for a lot of things
21:32:19Russ|werkhe apologised and made peace with the al queda somoli leader after the black hawk down incident
21:35:54erikmRuss|werk: well, the US has made more of those mistakes. iraq's hussein, panama's noriega, etc.
21:39:03Russ|werknoriega was "fixed"
21:39:28erikmRuss|werk: at quite some costs, yes.
21:39:53erikmactually liked the way they got him out of the vatican embassy :)
21:40:02Russ|werkbut this war lord attacked and destroyed UN food convoys, when special forces were sent to "remove" him and his leadership, their mission failed because they were attacked by his forces
21:40:31Russ|werkafter the incident, clinton apologized to the leader
21:41:32erikmdigs into the stack of newspapers
21:41:45erikmthere was something about that last saturday...
21:42:08Russ|werkI think there are probably a lot of similar such incedents where clinton could have done something more about al quiada (sp), but didn't, because it would mean american soldiers dying
21:42:48Russ|werka "no foriegn policy is good foreign policy" approach
21:43:20erikmRuss|werk: well, you have to be honest: one of bush points in his campaign was to decrease the level of foreign policy even further
21:44:01erikmRuss|werk: and the main reason is because foreign policy doesn't give you any votes
21:44:09Russ|werknot really, it was to reduce a lot of the older elements of military, but invest in higher tech stuff
21:44:20Russ|werk(ie, missile defense)
21:44:47erikmRuss|werk: but as the only remaining super power in the world, the US just can't affor not to use it
21:45:05Russ|werkthe hope is, if you have some form of missile defense, you no longer need the mutually assured descruction policy, and you can get rid uf your arsenal
21:45:50erikmRuss|werk: wake up. the soviet union is no more. there is no such thing anymore as "mutually assured destruction"
21:46:51erikmRuss|werk: china is not your largest problem. terrorists and rogue states are.
21:47:32Russ|werkI think missile defense is a much more long term plan, ie, 20 years
21:48:42erikmRuss|werk: I'm not against misile defense or "star wars", as long as it's used wisely
21:48:46Russ|werkthe world is likely to be as different today 20 years from now as it was 20 years ago
21:48:54erikmno doubt about that
21:49:12Russ|werkI don't think missile defense will do much if anything against terrorism
21:50:01erikmwell, the "star wars" missile shield would certainly help against the bin ladens with nuclear missiles in this world
21:50:38Russ|werkI'd think they'd go for something else, like the palo verde plant
21:51:18Russ|werk6th largest city in the US sitting next to the largest nuculear plant in the free world
21:52:47erikmthat would indeed be an easy target
21:53:32Russ|werkmy dad's been in there (sales)
21:53:42erikmanything within any open country is actually an easy target
21:53:52Russ|werkluke AFB is right next to it though
21:54:26erikmwho is luke AFB?
21:55:09Russ|werklargest traininf facility for F16 pilots
21:55:41Russ|werkthey seem to fall out of the sky quite a bit
21:56:17erikmwe didn't have that anymore. they fell out of the sky about 10 years ago, but I haven't heard about it lately
21:57:04Russ|werkit hasn't happened it about a year, but I think there were 8 incidents in a years time
21:57:30erikmlast plane that fell out of the sky was a german tornado a month ago. luckily above see, so nobody got hurt
21:58:14Russ|werkwe had a guy drop his fuel tank when his plane was starting to go down, it landed in the bed of a pickup truck wating at an intersection
21:58:35erikmoops :(
21:59:49erikmwell, I remember an F16 crashed in a street. it was a miracle it didn't crash on top of a house, but all houses in the street were severely damaged. again, nobody got hurt because most people were at work
22:01:42erikmwhat do I see over there? is that a cessna-ish plane in the upper left corner?
22:01:55erikmbut what about the rest?
22:02:17Russ|werkits the HUD of an F16
22:02:32erikmehm oops. so it's pretty close
22:04:23Russ|werkanyway, if they have a big "LF" on the tail, its from here
22:20:02erikmOK, I coded up an initial bootldr partition table parser. it's nowhere near to functional, but should I commit the code already?
22:20:20erikm(but it compiles without errors)
22:22:24Russ|werkoh, I did a bootldr thing too
22:22:38Russ|werkI made a patch so that there is an offset in blob to the bootldr parition offset
22:22:58Russ|werkso you hardcode the location of a member of the PTAG head into blob
22:23:24Russ|werkand set thet paramater to the offset of the bootldr partition table when you write the tags (the partition table is a tag)
22:23:35Russ|werkits just a small patch to bootldr.c
22:23:47erikmI don't understand the bootldr partition offset thing
22:24:09erikmI mean: I think I know what it does, but I don't like it
22:24:27Russ|werkI do something different
22:24:30erikmis that the patch you send to linux-arm-kernel?
22:25:00Russ|werksearch it for bootldr
22:25:07Russ|werkshould be able to stream to your browser
22:25:20erikmalready used wget
22:28:22Russ|werkyou just make a ptag and boot the bootldr partition table in there
22:30:04Russ|werkyou need to put the bootldr magic at 0x10 in blob or some such address
22:31:01erikmthat's what I don't like
22:31:13prpplaguelater guys, got to go do the christmas party thing
22:31:29erikmI just want to put the partition table at the start of an erase block
22:31:39erikmlater, prpplague
22:31:43erikmhmm, too late :(
22:31:55Russ|werknaw, put it in with the tags
22:32:28erikmno, I see them separately
22:32:52Russ|werkthey should be in the same erase block
22:33:00erikmyes, that's no problem
22:33:34Russ|werkso you want to put it at the start of the block, and paramaters right after that?
22:33:47Russ|werkit make sense to me just to drop it in as a param
22:34:06erikmno, it makes it much more difficult for the kernel to find the partition table
22:35:08Russ|werkyou mean, something like putting a magic in the partition table, and scanning the start of each erase block for that?
22:35:27erikmyes, because the partition table already *has* a magic value
22:37:19Russ|werkthen it wouldn't be in a fixed position, which would probably be a good thing
22:37:40Russ|werkbut then how do you decide how far to offset the paramater block from the start of the erase block?
22:37:57erikmfor example
22:38:14erikmanything should be OK, as long as blob knows where to find it
22:38:32erikmthe kernel doesn't have to know where to find it
22:39:04erikmbesides, having the partition table not in the params stuff makes it possible to put the params on a jffs2 partition
22:39:49erikmthinks he doesn't have to answer the cs89x0 question anymore :)
22:44:07Russ|werkerikm: hows that?
22:44:30Russ|werkerikm: if it was in the jffs2 partition, it wouldn't be at the start of the erase block
22:44:33erikmRuss|werk: cause both you and jdb already answered it :)
22:45:08erikmRuss|werk: no, but if it was in a jffs2 partition, blob had no way to figure that out at all. chicken vs. egg
22:46:35Russ|werkoh, for some reason I thought you said partition, not params
22:46:40erikmanyway, I almost got to go. I'll commit my flash partition stuff, so you can comment on it
22:52:39erikmdone. cya

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