**** BEGIN LOGGING AT Tue Feb 16 02:59:59 2016 Feb 16 07:47:09 blogic r48724 trunk/package/kernel/linux/modules/crypto.mk * kernel: crypto-echainiv was missing deps Feb 16 11:15:33 hrm, I was turning off things in menuconfig to make a smaller image, and I've got a nice small image now, but I no longer have a serial console? I've got all the stripping and removing opkg and things, it's fairly harsh, but I wass't expecting to lose the serial console? Feb 16 15:12:36 karlp: have a diff of .config.old and .config between tests? Feb 16 15:22:59 I could go back I guess, was wondering if there was some known gotcha. Feb 16 15:23:14 I only need this tiny image to bootstrap around a borked bootloader. Feb 16 15:23:28 the web console works, so it was just on my desk that the serial console missing was odd. Feb 16 15:26:02 what device are you missing the console on karl ? Feb 16 15:31:17 carambola2, only after turning on kernel and library stripping and removing lots of debug. Feb 16 15:31:28 it still has printk, I see it boot up, Feb 16 15:31:42 just doesn't accept the console anymore. Feb 16 15:31:57 was just curious what had killed it, it certainly wasn't something I explicitly disabled Feb 16 15:45:25 is there a way to know why some package was compiled that is NOT in enabled in .config ? Feb 16 15:52:46 bleh. I can edit /sbin/reload_config and see that it detects a change in my config, but my process isn't restarted. isn't it just service_triggers() procd_add_reload_trigger configname ? Feb 16 15:52:51 that's all it says here https://wiki.openwrt.org/inbox/procd-init-scripts#procd_triggers_on_config_filenetwork_interface_changes Feb 16 16:09:24 karlp: your process is only restarted if the caluclated command line changed Feb 16 16:09:42 right, or if you add procd_set_param file $CONF or similar right? Feb 16 16:09:59 yes (having a deja-vu ... didn't we discuss this a while back already?) Feb 16 16:10:08 yes. Feb 16 16:10:12 I wrote half that wiki page. Feb 16 16:10:18 yeah, I remember Feb 16 16:10:22 just updating forward onto CC and tryign to use these features now. Feb 16 16:10:27 updating some old packages to use procd Feb 16 16:11:12 so if your process is doing uci natively and not relying on command line arg passing then you should treat the uci file as plain config so that procd is md5summing it Feb 16 16:11:33 no, init script is generating a config file from uci config. Feb 16 16:11:47 then running "mosquitto -c /tmp/mosq.generated.conf" Feb 16 16:11:48 then register that regenerated file Feb 16 16:11:52 yeah, I have. Feb 16 16:12:01 using procd_set_param file ? Feb 16 16:12:05 procd_set_param file /tmp/blah Feb 16 16:12:06 yeah Feb 16 16:12:15 can you pastebin the ubus call service list output? Feb 16 16:13:30 http://fpaste.org/323569/14556392/ Feb 16 16:13:38 I'm guessing hte lack of any mention of the file there is going to be the problem. Feb 16 16:14:35 procd_debug seems to say the right thing though: http://fpaste.org/323570/63927114/ Feb 16 16:15:00 karlp: try touching the file Feb 16 16:15:02 sorry, PROCD_DEBUG=1 in the init script Feb 16 16:15:43 nothing. no log, no process restarting or anything Feb 16 16:16:28 karlp: http://git.openwrt.org/?p=project/procd.git;a=blob;f=service/instance.c#l590 Feb 16 16:16:51 so the file is not registered if it does not exist yet when the instance is sent to procd Feb 16 16:17:12 should already exist? Feb 16 16:17:40 yes Feb 16 16:17:55 http://fpaste.org/323571/39472145/ Feb 16 16:18:08 I mean, it should have already existed from my init script. Feb 16 16:18:41 yeah, right - brainfart Feb 16 16:19:28 also, teh prodc debug seems to say it got the file ok, if that "file": [ array ] output is meaningful when you run the init script manuallyt Feb 16 16:19:42 file and netdev are simply not reported in instance dumps Feb 16 16:19:48 *file and netdev triggers Feb 16 16:22:56 I assume your usagei s correct and procd is bugged Feb 16 16:24:12 anything I can do to confirm? this is on CC. Feb 16 16:25:06 not without patching procd, I won't have time to do it today but I'll add some commits tomorrow to let it dump its netdev and file triggers, this makes debugging easier Feb 16 16:25:12 blogic r48725 trunk/target/linux/mvebu/ config-4.4 config-4.1 * mvebu: add missing symbols. Feb 16 16:25:17 blogic r48726 trunk/target/linux/at91/modules.mk * at91: udc driver was missing deps Feb 16 16:25:26 blogic r48727 trunk/package/kernel/linux/modules/dsa.mk * kernel: DSA is missing a OF dependend dep for mdio Feb 16 16:25:35 blogic r48728 trunk/package/kernel/linux/modules/usb.mk * kernel: add missing symbol to usb-audio Feb 16 16:25:42 blogic r48729 trunk/package/kernel/linux/modules/video.mk * kernel: add mode backlight symbols Feb 16 16:25:42 ok, no problem then. Feb 16 16:25:50 blogic r48730 trunk/target/linux/generic/config-4.4 * kernel: add missing symbols to config-4.4 Feb 16 16:25:52 and you could try stracing procd to see if it at least tries to open the trigger file you sent to it Feb 16 16:26:06 I'm just trying to set this up so I can _actually_ start testing this sd card mounting/unmounting issues :) Feb 16 16:26:41 what's the trick to run procd int eh froground under strace? Feb 16 16:39:46 karlp: strace -p $(pidof procd) Feb 16 16:40:06 oh, huh, never tried strace on running processes before. TIL... Feb 16 16:53:08 ahhh, hehe, more fun. turns out some work I did for a startup wizard in luci only works if the caching is disabled. Feb 16 16:57:53 jow_laptop: doesn't seems to: http://fpaste.org/323611/64180614/ Feb 16 17:00:06 though, I think this is related... my /etc/init.d/mosquitto is actually a symlink, and it's getting hte anme wrong.... here's the strace of procd when I manually start the init script. http://fpaste.org/323613/55642001/ Feb 16 17:00:16 it does seem to look at the file when it starts at least. Feb 16 21:30:23 nbd r48731 trunk/package/devel/gdb/patches/100-musl_fix.patch * gdb: fix powerpc build issues with musl Feb 16 21:42:02 nbd r48732 trunk/package/kernel/brcm2708-gpu-fw/Makefile Feb 16 21:42:02 brcm2708-gpu-fw: move pkg build dir to kernel build dir to fix rebuild bugs after cleaning the kernel tree Feb 16 21:58:07 jow_laptop: the calloc_a call in jsonpath/ast.c is the only user of libubox when you separate the library and app portions, but I can't see any easy way to just replace it, it's not something I've seen before I'm afraid :) Feb 16 23:42:56 Is someone already working on putting https://sourceware.org/ml/libc-alpha/2016-02/msg00416.html into OpenWRT or should I do it? Feb 16 23:54:13 mamarley: not sure many people are using glibc, so... probably not really any sort of priority? Feb 17 00:47:43 karlp: The octeon architecture uses it. Feb 17 00:49:08 only, no? and not for long **** ENDING LOGGING AT Wed Feb 17 02:59:59 2016