**** BEGIN LOGGING AT Sun Sep 07 02:59:57 2008 Sep 07 08:59:46 juhosg * r12543 /trunk/target/linux/ar71xx/files/ (4 files in 2 dirs): [ar71xx] simplify leds-gpio device registration Sep 07 09:47:42 juhosg * r12544 /trunk/target/linux/ar71xx/files/ (4 files in 2 dirs): [ar71xx] add gpio-buttons devices Sep 07 10:01:51 xMff: Now I need to have a list of defined items .. like a set of enums Sep 07 10:02:47 xMff: I have some codecs (ulaw, alaw, gsm etc) - and the user can chose any combination Sep 07 10:36:17 Hi, I am trying to add sata support... and did a make kernel_menuconfig to do so... followed by a make target/linux/{clean,compile} but trunk/build_dir/linux-x86_generic/linux-2.6.25.16/drivers/ata doesn't have any .o files afterwards? Sep 07 10:37:13 Do I need to do a make menuconfig after a make kernel_menuconfig maybe? Sep 07 10:38:33 don't put stuff you want built as module into the kernel config Sep 07 10:38:38 put it into package/kernel/modules/*.mk Sep 07 10:38:46 with the appropriate KCONFIG lines Sep 07 10:38:52 after that run make oldconfig Sep 07 10:38:53 then make Sep 07 10:39:06 which should give you kmod-*.ipk files with your kernel modules Sep 07 10:41:08 Hi ndb, I was just reading your pdf http://nbd.name/openwrt.pdf where it shows how to add stuff... in package/kernel/modules/*.mk Sep 07 10:41:14 hi Sep 07 10:41:26 johnrw: that pdf hasn't been updated yet Sep 07 10:41:37 but the examples are rather simple Sep 07 10:41:43 just check the stuff that is already in there Sep 07 10:41:51 It is outdated already? Sep 07 10:41:52 and if you have further questions, just ask Sep 07 10:41:58 i haven't rebuilt it from trunk lately Sep 07 10:44:32 well I just built the whole repository... every package from the 3 main feeds... which took me days on an old pentium 3... ( I was mostly interested to see what builds and what doesn't) so I absolutely do not want to trigger a complete rebuild of all packages too. Sep 07 10:47:25 it doesn't rebuild all packages, but there is indeed a faster way: Sep 07 10:47:36 How abouut if I tell make kernel_menuconfig to build sata (libata it?) it into the kernel? Sep 07 10:47:41 (after make oldconfig) make target/compile package/kernel/compile Sep 07 10:47:56 you can build it into the kernel, of course Sep 07 10:48:11 but patches for definitions for the sata kernel modules would be more useful Sep 07 10:48:15 (to us) Sep 07 10:49:03 after I did a make target/linux/{clean,compile} V=99 I could not find a kernel image anywhere... Sep 07 10:49:51 would make target/linux/{clean, compile, install} do it? Sep 07 10:50:43 Yes... I was thinking of giving a try at adding the sata .mk files... Sep 07 10:54:41 but I would do that on a tree that has just the base packages symlinked... not 1292 ipks! Sep 07 10:57:06 the good news is out of all those packages in the feeds... I'd say less than ten were broken(or wouldn't compile with the toolchain as delivered) Sep 07 10:57:18 make target/linux/clean is unnecessary Sep 07 10:57:26 and because you did it, it nuked your kernel tree Sep 07 10:57:34 which means you have to run a regular make to get images again Sep 07 10:57:44 because it also nuked all kernel dependent packages along with it Sep 07 10:58:01 oh no.... Sep 07 10:58:27 but don't worry Sep 07 10:58:31 it won't rebuild all your packages Sep 07 10:58:33 just the kernel dependent ones Sep 07 10:59:23 oh okay... that's a lot better. Sep 07 10:59:46 it'll check each package for modifications, though Sep 07 10:59:54 so you will see it run through all packages Sep 07 10:59:59 but it won't actually rebuild most of those Sep 07 11:00:30 Well I am picking up another p3/1000 today to be a kind of dedicated build machine.... Sep 07 11:04:31 since I want the build to occur on the same procesor as my target... so my target machine can get used... and once that's done... ie..target machine running openwrt 24/7 ... maybe I can do more to help out around here. Sep 07 11:07:32 but i really do need to read up on the build system, imagebuilding sdk etc. Sep 07 11:07:39 I admit. Sep 07 11:19:09 I felt kinda lucky to have found your pdf though! Thanks for that. Sep 07 11:27:41 johnrw: make help Sep 07 11:28:09 Hi loswillios! Sep 07 11:28:16 hello Sep 07 11:28:20 hello johnrw Sep 07 11:28:27 hello Sep 07 11:28:31 yeah, you too Sep 07 11:29:31 well I am taking a stab at package/kernel/modules/sata.mk Sep 07 11:30:17 I'll try and start small... and add just one option... mine. :) Sep 07 11:30:55 sata_sil Sep 07 11:33:35 ya know... I bought a Siig sata card to run on this p3... and a 1tb seagate drive... and the drive was too big for the bios! Thankfully, there was an upgrade from Silicon Image. Sep 07 11:35:59 But that was tricky to do without msdos and a floppy drive. Thank the lord for floppy emulation... So I wrote a howto for Siig's tech support. Sep 07 11:36:18 heh. you could have tried to make a small boot partition Sep 07 11:36:28 small < 32G Sep 07 11:37:38 frogonwheels: specify them es enum with the multival flag set to true Sep 07 11:45:31 My Main linux box...(still sits on a hdinstall of knoppix 5.1.1) does have an old (but still working) win98 partition I can boot to, and has a floppy and a Benq (my favorite burner) so I made a floppy image on that machine... and burned it to cdr. The machine that had the card in it... has all those new media readers but no floppy. Took 5 minutes... but updateing a Bios from cdr was... Sep 07 11:45:33 ...potentially full of gotcha's... but worked just fine. Sep 07 11:51:41 Tech support at Siig had never tried it... and they swore I could not do it. He said he could remotely update the card's bios (which I thought might be a neat trick) or I could use a bootable floppy. Just those 2 options. Hah! Sep 07 11:57:47 xMff: thanks. Nearly got it ready to go Sep 07 12:02:20 xMff: I'm sure this isn't totally right.. but it is a start: http://pastebin.com/dec22848 Sep 07 12:04:33 frogonwheels: nice work! Sep 07 12:04:52 Thanks - it took some effort. Sep 07 12:05:05 frogonwheels: now you can put it in /lib/uci/schema/default Sep 07 12:05:13 and do uvl verify asterisk Sep 07 12:05:21 can you post me an example config too? Sep 07 12:05:42 sure -I'm going to have to modify my example Sep 07 12:09:02 Error in scheme "asterisk": Sep 07 12:09:02 Missing required scheme field "variable" in "asterisk.(nil).(nil).cfg4783b71" Sep 07 12:09:08 xMff: ^^ ?? Sep 07 12:09:48 hehe, I'll take a look Sep 07 12:09:57 wait - I'll update Sep 07 12:10:54 http://pastebin.com/d3c2d9065 Sep 07 12:18:36 frogonwheels: found a few errors, but I guess it's ok if I correct them :) Sep 07 12:18:45 will submit corrected version in a fewminutes Sep 07 12:18:57 ta Sep 07 12:19:14 it's a complex config :| Sep 07 12:32:10 johnrw: btw. i updated the pdf Sep 07 12:32:47 frogonwheels: http://pastebin.com/d69839d41 Sep 07 12:32:54 now it compiles fine here Sep 07 12:33:51 nbd... oh yeah?! Cool. Sep 07 12:34:15 frogonwheels: in some variables you confused type with datatype in the enums you specified option variable as option name, but it's all fixed now Sep 07 12:34:25 *and in the enums Sep 07 12:34:54 but search & replace + regxp ftw! :D Sep 07 12:36:43 cool Sep 07 12:37:43 if you like, I can maintain it from now on. I just need a good example config, then I can try to start implementing web interface pages Sep 07 12:39:18 I did notice... the Form change... for example... the FILES:=$(MODULES_DIR)/kernel/drivers/i2c/*.$(LINUX_KMOD_SUFFIX) Is now FILES:=$(LINUX_DIR)/drivers/i2c/i2c-core.$(LINUX_KMOD_SUFFIX) and i2c-dev.$(LINUX_KMOD_SUFFIX) Sep 07 12:40:27 That would be being explained in section 2.1.3 Sep 07 12:40:36 Okay now I'll reload it... Sep 07 12:43:32 old=FILES:=$(MODULES_DIR)/kernel/drivers/i2c/*.$(LINUX_KMOD_SUFFIX) Sep 07 12:43:33 frogonwheels: I guess we have to precompile the scheme for the router, because I think interpeting nearly 4000 lines of uci will take some time Sep 07 12:43:47 new FILES:=$(LINUX_DIR)/drivers/i2c/i2c-core.$(LINUX_KMOD_SUFFIX) and i2c-dev.$(LINUX_KMOD_SUFFIX) Sep 07 12:44:24 xMff: we'll see Sep 07 12:44:51 xMff: hmm. yes Sep 07 12:54:33 nbd: well, I think the C-UCI can easily cope with it, but I have to test the Lua performance. But I assed the possibility to compile scheme to bytecode in order to short the startup time Sep 07 12:54:41 *added Sep 07 13:04:23 btw. was the lua optimization patch that i did noticeable in any way? Sep 07 13:04:30 i mean in terms of responsiveness in luci Sep 07 13:05:21 well, after Cyrus removed lot's of unnescessary table copy operations, luci feels really snappier now Sep 07 13:06:15 i really have to check it out again soon. last time i saw it running on a fonera it still felt somewhat sluggish Sep 07 13:06:28 especially on pages where the template wasn't cached yet Sep 07 13:06:36 ah hmm, jes Sep 07 13:06:53 we have a bytecode compile cache now Sep 07 13:07:23 so first invocation ist still slow, but once its compiled it should be fastr Sep 07 13:07:36 but there is still lots of room for improvements Sep 07 13:07:55 especially the validation has a huge performance impact Sep 07 13:08:29 i'm thinking that maybe we should do the template processing in c at some point Sep 07 13:08:47 but probably after the release Sep 07 13:09:13 i have another idea for improvement: Sep 07 13:09:24 how about precompiling the templates during startup in background? Sep 07 13:09:49 at least frequently used ones Sep 07 13:09:55 on systems with enough ram Sep 07 13:10:44 nbd: should be possible Sep 07 13:11:12 nbd: but I think the best performance can gain can be achived when precompiling the whole stuff in the build phase Sep 07 13:13:31 there are some really fat modules, where the most processing time is spend in the startup phase Sep 07 13:14:58 I think that boa could help a lot since we're able to hold the context in memory then Sep 07 13:15:33 which would be more efficient then throw the binaries in /tmp I think Sep 07 13:18:44 yes Sep 07 13:19:04 and i should add a function that looks at the available amount of ram Sep 07 13:19:10 and does caching decisions based on that Sep 07 13:19:44 jep, perhaps you could overload the builtin require() or sth like that Sep 07 13:22:36 This is old... but may still be valid... http://linuxmafia.com/faq/Hardware/sata.html but it says: Note that enabling libata support for this chipset requires enabling CONFIG_BROKEN (under "Code maturity level options") with explanation from Jeff Garzik. How would I confirm if that is still a required depend? Sep 07 13:28:14 juhosg * r12545 /trunk/target/linux/ar71xx/image/Makefile: [ar71xx] build more image types Sep 07 13:46:49 And can I enable Config_Broken? Well now scsi_mod.ko has disappeared... http://pastebin.ca/1196419 Sep 07 14:09:58 i think CONFIG_BROKEN should not be necessary for anything that works properly Sep 07 14:10:01 so i assume the howto is old Sep 07 14:39:39 well the howto was too old... as it told me to enable something no longer there... SI support in scsi->low level support... so now I am just trying to make sure that HAS_IO_MEM and BLOCK are in fact enabled... because I never get a libata built for some reason... Sep 07 14:40:22 s/HAS_IO_MEM/HAS_IOMEM/ Sep 07 14:40:39 did you see in package/kernel/modules/*.mk how to set config options to y? Sep 07 14:41:50 not yet... because I went ahead with a toplevel make which died because scsi_mod.ko couldn't be found... Sep 07 14:55:50 well I started to read that... in your pdf... but I think something is keeping the kernel module from being built... Sep 07 14:56:59 how can I just compile the kernel without clean? make target/linux V=99 says it is up to date? And make target/linux/{compile} doesn't work? Sep 07 15:15:40 aha! make target/linux/compile V=99 works! Sep 07 15:16:17 and it is compiling the sata drivers now! Sep 07 15:17:14 :) Sep 07 15:20:05 man... that had me chasing ghosts... but all is well... now I got to meet the guy with the old boat anchor pc to make a build machine with... but I will try and make some sata mk's later today hopefully. Sep 07 15:22:39 nbd * r12546 /trunk/package/dnsmasq/files/dnsmasq.init: restore the /tmp/resolv.conf symlink when stopping dnsmasq Sep 07 17:27:14 nbd: since you are working on the dnsmasq startup, 'killall dnsmasq' doesn't really work in the stop function, sometimes kit kills the script before it kills dnsmasq Sep 07 17:27:39 i'll fix that when we have start-stop-daemon integration Sep 07 17:28:16 I have an ugly fix if your interested, :) Sep 07 17:28:33 i'd prefer a proper one Sep 07 17:28:36 ;) Sep 07 17:29:21 hehe, yeah, I just used PS and AWK to get the PID of dnsmasq, 4 lines is much less clean than 'killall dnsmasq' Sep 07 17:30:09 I'm curious about start-stop-daemon integration... Sep 07 17:30:25 what about "pidof dnsmasq" ? Sep 07 17:30:44 kill $(pidof dnsmasq) Sep 07 17:31:16 the init script is named dnsmasq too Sep 07 17:36:10 argh, right Sep 07 17:38:04 which is why i think start-stop-daemon is the right solution Sep 07 17:38:10 it's in busybox already Sep 07 17:38:26 just needs some integration work to make it easy to use from scripts using /etc/rc.common Sep 07 18:26:25 nbd: performance with the asterisk scheme seems to be okay: http://pastebin.com/d1201414e Sep 07 18:26:42 first is without, second with bytecode compiled scheme Sep 07 18:26:58 this was done on a wrt54g Sep 07 18:27:07 interesting Sep 07 18:27:17 though i still wonder where the bottleneck with precompiled bytecode is Sep 07 18:27:22 even 0.5s seems like a lot to me Sep 07 18:28:54 I guess my code needs some performance improvements as well Sep 07 18:29:20 many oop stuff and redundant data structures etc. Sep 07 18:58:36 Hello ! Sep 07 18:58:42 exovii? Sep 07 19:01:05 nbd: here ? Sep 07 19:01:13 yes Sep 07 19:01:23 Ah ! Sep 07 19:01:30 I have a question for you Sep 07 19:02:39 I own a Netgear fvx 538, and i am able to run openwrt and use the two wan interfaces Sep 07 19:04:10 but the bcm5380 switch, which is properly initialized in reverse mii mode in Redboot (it works !), don't work when linux booted Sep 07 19:04:39 I think the initialization done by Redboot has gone when linux is up. Sep 07 19:05:21 So my question is: do you think it could be possible for the "switch" package to do bcm5380 management through spi ? Sep 07 19:06:17 if you don't know the fvx 538, please see it here: http://wiki.openwrt.org/OpenWrtDocs/Hardware/Netgear/FVX538 Sep 07 19:07:50 it's a great machine, but the Netgear firmware is crappy.... Sep 07 19:08:56 And as i can speak to you, many thanks for this great project, i think you and the whole openwrt staff did a very good work. Sep 07 19:11:48 the switch package itself can't do spi Sep 07 19:11:52 but it could be extended to do it Sep 07 19:13:10 if you want to do it and you need any hints, let me know Sep 07 19:13:18 yes, because i saw that etc53xx.h has some spi constants... Sep 07 19:14:24 i want to try to do it, but i certainly won't have the dev skills to do it... Sep 07 19:16:15 Wwhen i start the ixp mii that is connected to the bcm5380, the drivers says "MII read failed". It seems that the read failed because the bcm5380 MAC is not in reverse mii mode... Sep 07 20:08:02 cyrus * r12547 /trunk/package/dnsmasq/files/dnsmasq.init: dnsmasq: Use a more intelligent way of parsing dhcp-options using the new UCI list datatype Sep 08 01:23:21 xMff: SO.. what's next Sep 08 01:23:47 btw I still get: Enum "g726" in scheme "asterisk" references non-enum option "asterisk.iaxgeneral.allow.g726" **** ENDING LOGGING AT Mon Sep 08 02:59:57 2008