**** BEGIN LOGGING AT Thu Sep 24 09:52:27 2020 Sep 24 09:52:43 in an ideal world we would look at vmlinux and/or system.map to see which kinds of symbols are exported by rthe kernel Sep 24 09:52:47 and checksum that Sep 24 09:52:52 well I have two motivation, first reproducibility meaning kernel magic and user/hostname within the image should be the same, second the kmod feed could be used for self created imagebuilders if the kmagic wouldn't change Sep 24 09:52:57 then we would not need any "kernel magic" at all Sep 24 09:53:21 oh actually maybe not because of missing symbols within the self build imagebuilder kernel Sep 24 09:54:07 in my world view, reproducibility means providing the means to tweak and respect environment settings to be able to reach the same output with the same parameters Sep 24 09:54:16 it does not mean to enforce the same faked defaults for all people Sep 24 09:55:19 true, I wanted to "slap two flies at once", but as local kernel will miss kmods symbols anyway it's not a big gain Sep 24 09:55:41 i'll look at system.map once selinux is default grin Sep 24 09:56:15 for some arcane reason I keep forgetting it is somehow not easy/impossible to first build vmlinux and later all the modules Sep 24 09:56:30 nbd might know more, but I learned that this is an intrinsic OpenWrt buildroot limitation Sep 24 09:56:50 and I know that Gluon does some unholy hacks to reverse the order (?), neoraider might know more about that Sep 24 09:59:59 one reason to build the kernel after modules and packages is that it's needed for initramfs Sep 24 10:03:52 found it: https://pastebin.com/hSYhKLsa Sep 24 10:04:41 nbd: is there no make target / tool / solution to glue vmlinux + initramfs together and recompress the result? Sep 24 10:05:05 it would have to be arch specific, since initramfs is linked into vmlinux Sep 24 10:05:27 incrementally rebuilding the kernel image should be no problem either Sep 24 10:05:43 first build bare vmlinux, thne rebuild it later for initramfs cases Sep 24 10:06:29 i guess that might work. what's the reason for wanting to build vmlinux first? Sep 24 10:06:53 to check the exported kernel symbols and build and abi fingerpint on top of that Sep 24 10:07:10 in order to make our kmod/kernel dependencies less overly specific Sep 24 10:07:27 right now the checksumming factors in a lot of unrelated config symbols Sep 24 10:07:37 which don't affect the abi at all Sep 24 10:07:57 it would be a global kernel ABI fingerprint? or something per module? Sep 24 10:08:01 global Sep 24 10:08:20 something we can put as version in the kernel metapkg and as version constraint dependency into each kmod Sep 24 10:08:47 DEPENDS:=kernel (5.4.123-) Sep 24 10:08:54 DEPENDS:=kernel (=5.4.123-) Sep 24 10:10:26 nbd: so the actual goal is not even having vmlinux but Module.symvers, I suppose the latter implies the former though **** BEGIN LOGGING AT Thu Sep 24 10:13:58 2020 Sep 24 10:15:18 jow: I'll look at this system.map stuff but I think fixing the build system is too nerdy for me Sep 24 10:15:55 I'm out, bye Sep 24 10:17:58 bye bye Sep 24 10:37:17 nbd: in my app i use uclient and in the .header_done i do: if (cl->status_code != 200) { uclient_disconnect(cl); uclient_free(cl); } Sep 24 10:37:32 nbd: that uclient_free() seems to be causing a crash Sep 24 10:37:38 https://pastebin.com/raw/kpqY2CU1 Sep 24 10:37:46 (delayed one) Sep 24 10:38:02 is there something missing in the uclient_disconnect() ? Sep 24 10:40:01 defer the free using a uloop_timeout Sep 24 10:40:17 you can't free it from within that callback, since it does other processing Sep 24 10:40:42 i see now that uclient_http_request_disconnect() does uloop_timeout_set(&uh->disconnect_t, 1); Sep 24 10:40:46 sounds a bt hacky Sep 24 10:40:49 it won't be racy? Sep 24 10:40:57 my free vs. &uh->disconnect_t Sep 24 10:41:45 hm Sep 24 10:42:07 let me check Sep 24 10:42:27 or should we have another callback in the struct uclient_cb? for freeing resources? Sep 24 10:43:49 you don't need to do do disconnect and free Sep 24 10:43:52 free implies immediate disconnect Sep 24 10:43:59 and cancels the disconnect timer Sep 24 10:44:43 oh, interesting Sep 24 10:45:14 i guess it would be possible to allow free from within the callback too Sep 24 10:45:18 but it's more complex Sep 24 10:45:57 the calls that need protection against use-after-free would have to create an on-stack variable for aborting everything Sep 24 10:46:01 and put that in the data structure Sep 24 10:46:06 so that the free call can set it Sep 24 10:46:30 let me check Sep 24 10:46:35 uclient_free() -> uclient_http_free() -> uclient_http_free_url_state() -> uclient_http_disconnect() Sep 24 10:46:41 i see that path at least now Sep 24 10:49:22 i'll try to hack up a patch to allow free from within callbacks Sep 24 10:50:07 i'll be happy to test FWIW Sep 24 11:00:56 https://pastebin.com/Udn3T7Ez Sep 24 11:00:59 completely untested Sep 24 11:01:03 but should show the general approach Sep 24 11:28:56 nbd: as an FYI, it looks like xcode12 has b0rken building dc (within bc package) in host tools. after being awake 24 hours for other reasons I can't think straight to work it out. Sep 24 11:30:36 it's a typedef redefinition blowup but bizarrely it looks like a re-def within xcode headers and not a package related thing. I'm just 'WTH?' at the moment :-) Sep 24 11:37:46 https://pastebin.com/7MxPvMrD if anyone interested Sep 24 11:38:05 seena bunch of recent xcode12 bugreports on other projects too fwiw ldir Sep 24 11:39:21 no surprise there :) Sep 24 11:40:05 seemsmostly related to "latest clang sets -Werror=implicit-function-declaration by default now" Sep 24 11:40:11 but probably others... Sep 24 12:16:19 he he - usually what happens is I struggle and put in a 20 line invasive fix to get a local build and then a few hours later nbd comes along with a single line change that's simply genius :-) Sep 24 12:38:05 can anyone tell me what generates this file: usr/bin/ld: scripts/dtc/dtc-parser.tab.o:(.bss+0x10): multiple definition of `yylloc'; scripts/dtc/dtc-lexer.lex.o:(.bss+0x0): first defined here ? Sep 24 12:38:10 that dtc-lexer.lex.o ? Sep 24 12:41:25 ldir: can you please show me config.log from the bc host build dir? Sep 24 12:43:31 tete_: liekly created by flex Sep 24 12:44:20 nbd: https://pastebin.com/w57JTjys Sep 24 12:47:37 ldir: and now grep -r ptrdiff_t /Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX.sdk/usr/include Sep 24 12:47:59 i'm still on mojave, so i can't install the latest xcode myself yet Sep 24 12:48:55 https://pastebin.com/UmF074Gd Sep 24 12:57:30 ldir: it looks to me like the configure tests are broken in a weird way. Try adding HOST_FIXUP:=autoreconf before the host-build.mk include to tools/bc/Makefile Sep 24 12:57:40 it seems to make the generated configure tests cleaner Sep 24 12:58:20 thanks, I'll give it a try Sep 24 13:09:04 nbd: that patch (diff) broke uclient, i'm not getting cl->status_code 0 every time .header_done gets called Sep 24 13:10:47 odd Sep 24 13:13:34 let me verify that Sep 24 13:20:35 jow, is there some pattern so i can know which lex file that created? Sep 24 13:20:52 because my 2 patches should remove all those yyalloc lines from the .lex files Sep 24 13:25:45 Is there any docs for initscript uci config validation? Sep 24 13:26:14 I'd like to avoid starting a service if the configuration isn't properly prepared Sep 24 13:30:17 is really no one else compiling with gcc10 and facing these problems? Sep 24 13:36:41 $ gcc -v 2>&1 | tail -n 1 Sep 24 13:36:42 gcc version 10.2.0 (Debian 10.2.0-7) Sep 24 13:36:50 I cna compile OpenWrt Sep 24 13:37:31 nbd: what do you think about https://github.com/openwrt/openwrt/pull/3445/files ? (context: https://bugs.openwrt.org/index.php?do=details&task_id=3360) Sep 24 13:37:48 I like that change and I think it would make source change detection a lot more robust Sep 24 13:38:12 -printf usage should be fine since we mandata GNU find Sep 24 13:39:01 the extra sort subshell is not ideal. I guess we could also wrap the find command in a script that takes care of appending a timestamp and sorting the output Sep 24 13:39:07 so maybe it only affects mvebu with gcc10 Sep 24 13:41:57 jow: that change looks good to me Sep 24 13:42:02 hello, I am trying to test the Qualcomm PLC chip QCA7005. I have two embedded Linux systems with apparently working qca7000 driver, but I am unsure how am I supposed to test actual communication Sep 24 13:42:36 The chips on each board are connected by the two powerline wires Sep 24 13:42:57 Just setting IP addresses with ifconfig and trying to ping the other does not work Sep 24 13:49:13 plcstat -i eth2 -t only shows local adapter Sep 24 13:49:22 not sure what I am doing wrong Sep 24 13:49:56 I did execute plctool -i eth2 -B join before that, but it did not help Sep 24 13:50:18 these are open-plc-utils commands Sep 24 14:13:22 does anyove have armada 385 linksys device, or something with LEDs controlled by leds-pca963x driver? We need someone to test some patches for linux-leds... Sep 24 14:22:22 kab-el: You mean a Turris Omnia, perhaps? Sep 24 14:25:06 On second thought, I don't think the Omnia uses the pca963x. Sep 24 14:27:44 How do I learn what is uci_load_validate good for? Sep 24 14:30:08 rsalvaterra: it does not, we have plenty of Omnias at CZ.NIC :D Sep 24 14:30:29 rsalvaterra: according to device trees linksys routers have this controller Sep 24 14:30:56 rsalvaterra: we need someone to test 2 more patches so that Pavel can apply on leds/for-next Sep 24 14:36:20 pavlix: it calls /sbin/validate_data internally which understands the following datatypes: https://lxr.openwrt.org/source/ubox/validate/validate.c#L893 Sep 24 14:36:25 kab-el: I have WRT1200AC I don't use I can test on Sep 24 14:37:09 kab-el: can you compile image for me and tell me what to test? Sep 24 14:38:03 nbd: my bad, my packages had incomplete Makefiel and it didn't get recompiled after uclient change Sep 24 14:38:33 nbd: that 0 status_code was probably a result of reading from old struct binary format Sep 24 14:38:55 nbd: your patch seems to fix crash for me! Sep 24 14:40:22 jow: Thanks. But how is that correctly used in the initscripts. Right now I just want to load a simple config but /etc/config/{system,dropbear,...} examples look really confusing in the initscripts. Sep 24 14:40:55 uci_load_validate is supposed to load the values (as the name suggests) but only config_load seems to actually do that Sep 24 14:40:58 kab-el: Oh, you're from CZ.NIC? :) Ah, well… I'm stiil looking forward to have complete LED control on my Omnia, though. ;) Sep 24 14:41:13 *still Sep 24 14:42:24 Also config_foreach seem to be used even for singletons but than the *_config function doesn't pass variables back to the start_service. Sep 24 14:43:04 it doesn't need to? Sep 24 14:43:55 kab-el: By the way I experimented a lot with the marvell switch driver. Sep 24 14:44:39 pavlix: is this a simpler example? https://paste.jvnv.net/view/vbkTv Sep 24 14:45:17 that's older than uci_load_validate though I guess. Sep 24 14:46:18 karlp: I see. But then I'd like to use it well in the initscript and that I can't probably derive from the c code... Sep 24 14:46:34 well, that paste of mine is from an initscript. Sep 24 14:46:45 well you pass a series of colon-separated tuples to uci_validate_section() Sep 24 14:46:52 I thin the c file link was just to help understand teh types available Sep 24 14:47:00 karlp: Thanks, my browser tabs got confused Sep 24 14:47:06 these tuples are uci_option_name:type_expression:default value Sep 24 14:47:10 :default is optional Sep 24 14:47:22 uci_option_name refers to the uci option that shall be validated Sep 24 14:47:51 type_expression is one of the things understood by https://lxr.openwrt.org/source/ubox/validate/validate.c#L893 Sep 24 14:49:06 karlp: This looks very simple, thanks. What's the purpose of procd_add_validation (when we already do call validation in *_add_instance? Sep 24 14:50:07 karlp: Also you're not setting USE_PROCD, is it no longer needed? Sep 24 14:50:14 it wass't the whole init script Sep 24 14:51:03 procd_add_validation was required at some point, I don't rememberanymore. Sep 24 14:54:46 karlp: I see. Thanks. Sep 24 15:37:00 jow, how did you install that gcc 10 on debian? i have a standard install of debian 10 but it has only gcc 8.x Sep 24 15:37:14 would like to test that in a virtualbox if it behaves the same. Sep 24 15:45:57 probably install bulls-eye alpha2 and then enable the experimental repository do apt-get update , apt-get upgrade , apt-get distro-sync: https://www.debian.org/devel/debian-installer/ Sep 24 15:54:46 well, the output of jow was: gcc version 10.2.0 (Debian 10.2.0-7) Sep 24 15:56:18 Hello, I have been looking for awhile and cannot find documentation on changing rootfs size in the openwrt build. Is anyone familiar with editing the rootfs size in make menuconfig? Sep 24 15:57:50 daniel__: in menuconfig hit / then search for TARGET_ROOTFS_PARTSIZE Sep 24 16:01:41 Awesome thank you @stintel Sep 24 16:01:58 daniel__: you're welcome Sep 24 16:04:35 rsalvaterra: yes, this is why I am now working on LED subsystem to clear some things. Upstream kernel driver for LED on omnia does not support HW offloading yet Sep 24 16:05:29 rsalvaterra: the current idea is transparent offloading, so that if you set netdev trigger on the WAN LED, for example, and the netdev trigger settings are offloadable to the PHY, the system will offload it to hardware Sep 24 16:05:33 kab-el: Wonderful news! I've been following the multicolor framework stuff. :) Sep 24 16:06:49 rsalvaterra: the problem is that the PCIe LEDs are connected to outputs from Mini-PCIe ports, and the function there can vary with the Mini-PCIe card Sep 24 16:07:04 rsalvaterra: and some cards do not do anything interesting with the LED pins Sep 24 16:07:06 kab-el: By the way, are the first (Indiegogo) Omnias going to get the updated u-boot? Sep 24 16:07:44 tete_: gcc-10 is available in Debian testing, so you can add that to your stable system, make apt config with APT::Default-Release "stable" and install gcc with "apt install -t testing gcc" Sep 24 16:08:04 thanks, ill try that Sep 24 16:08:08 rsalvaterra: it is planned, but we have too much stuff on our minds now and we are a little bit afraid about automatic u-boot upgrade Sep 24 16:08:09 kab-el: Well, in that case, the mini-PCIe LEDs can always be software-controlled, right? Sep 24 16:08:30 rsalvaterra: I can write a manual for how to compile and flash new u-boot, though Sep 24 16:08:36 rsalvaterra: yes Sep 24 16:08:53 kab-el: That works for me. :) Sep 24 16:08:57 rsalvaterra: in fact i think this is most reasonable choice currently Sep 24 16:09:20 pavlix: what experiments did you do? Sep 24 16:09:59 rsalvaterra: I have gentoo on my omnia, for example, with upstream 5.4 kernel Sep 24 16:10:13 rsalvaterra: so I would also like full support in upstream kernel Sep 24 16:11:51 kab-el: I've also been having some strange issues with the pca953x on my Omnia… Sep 24 16:13:09 … I had to build the driver without support for interrupts, otherwise they'd be continuously triggering (about 20 reqs/s), and the kernel would eventually give up. Sep 24 16:13:53 stintel, so if I were to edit the default value in config/Config-images.in for TARGET_ROOTFS_PARTSIZE, and then run make, would that new firmware flash factory image (not sysupgrade) use the new value I inputed? Sep 24 16:14:36 stintel, my device target does not have a TARGET_ROOTFS_PARTSIZE option under TargetImages -> in menuconfig Sep 24 16:19:32 Wait, the WDR4300 is supported by upstream u-boot…? :O https://gitlab.denx.de/u-boot/u-boot/-/commits/master/board/tplink/wdr4300 Sep 24 16:19:43 Does this also apply to the WDR3600…? Sep 24 16:40:55 jow, nbd: I haven't read the whole discussion where I was pinged, but the common solution to pack an initramfs with the kernel in a post-processing step would be a fitImage Sep 24 16:42:15 Theoretically, we could replace our lzma-loader or relocate with a small fitImage loader Sep 24 16:42:55 Hmm, this doesn't even need to use fitImages, the bootloader passes the initramfs to the kernel, it doesn't matter how this is stored on flash Sep 24 16:42:56 rsalvaterra: the gpio driver? Sep 24 16:56:52 Does anyone know the difference between RAS_ROOTFS_SIZE & TARGET_ROOTFS_PARTSIZE ? Sep 24 18:12:34 kab-el: Yes, the GPIO driver. (Sorry for the delay.) Sep 24 18:15:55 kab-el: For some reason, the interrupt handler seems to stop working, and the kernel disables it with a big fat warning. I don't know if it's SFP related (I'm not using SFP). Sep 24 18:20:19 rsalvaterra: kernel version? Sep 24 18:20:53 kab-el: It as OpenWrt master at 5.4.47, at the time, IIRC. Sep 24 18:21:19 I haven't reenabled GPIO interrupts since then. Sep 24 18:23:36 Oh, and actually it was about 100 reqs/s not 20. Sep 24 18:28:18 Anyone here that can read a gdb bt on a SIGILL? Sep 24 18:28:29 It appears to be MUSL related Sep 24 18:28:45 kab-el: I found the original email: https://lists.openwrt.org/pipermail/openwrt-devel/2020-June/029695.html Sep 24 18:29:00 https://pastebin.com/akRSQYn8 Sep 24 18:48:53 Grommish: looks like wrong FPU instruction for the CPU used Sep 24 18:50:24 Grommish: you can try something like x/i $pc Sep 24 18:50:28 To see what it is exactly Sep 24 18:55:09 neoraider: so you have an idea how to replace linuxmagic with a hash of systemmap :)? Sep 24 19:43:05 jow: why isn't no System map created by default? I have to manually run `make modules` Sep 24 19:49:24 rsalvaterra: i think it may be because of wrong devicetree Sep 24 19:50:20 rsalvaterra: make a tarball of /sys/firmware/devicetree/base and send it to me somehow, if the issue is still actual Sep 24 19:52:31 kab-el: Done. Where do I send it to? :) Sep 24 19:58:26 (If it wasn't obvious, this is a CZ11NIC13 model.) Sep 24 20:05:30 rsalvaterra: kabel@kernel.org Sep 24 20:06:59 so i got the same error now again with debian and gcc10, so it occurs at least with target mvebu Sep 24 20:07:32 maybe x86 and others are not affected, but all mvebu targets are affected... wondering how important this is and when it will be fixed Sep 24 20:11:16 kab-el: You've got mail. :) Sep 24 20:24:44 PaulFertser: Sorry, I had to step away.. Sep 24 20:36:19 rsalvaterra: hmm, your devicetree seems reasonable Sep 24 20:36:59 rsalvaterra: its weird. Maybe you have something in the sfp port that causes shortening of a gpio? :D Sep 24 20:37:21 rsalvaterra: or if you want you can send me your kernel binary and modules Sep 24 20:38:19 rsalvaterra: or you can just disable the gpio controller from u-boot via fdt command :) Sep 24 20:40:07 PaulFertser: I will try that.. i tried rebuilding with a -msoft-float gcc option, and while it built, it didn't actually boot the device Sep 24 20:40:23 So I pulled it and will try again Sep 24 21:10:36 kab-el: The only thing I have on my SFP port is a plastic cover. :P Sep 24 21:11:38 I just disabled the interrupts… I don't think I'm using GPIO for anything. LEDs, maybe…? Sep 24 21:18:24 rsalvaterra: no, LEDs are controlled via I2C Sep 24 21:18:38 rsalvaterra: you disabled interrupts on the whole system? Sep 24 21:20:31 kab-el: No, I just disabled the interrupt controller feature of the PCA9547. Sep 24 21:21:13 rsalvaterra: you mean PCA9538 Sep 24 21:21:19 And the interrupt in question doesn't appear anymore in /proc/interrupts. Sep 24 21:21:28 rsalvaterra: PCA9547 is I2C multiplexer Sep 24 21:21:36 Right, PCA9538, sorry. Sep 24 21:22:25 (It was interrupt 76, IIRC.) Sep 24 21:37:14 Fri Sep 25 00:35:57 2020 daemon.notice hostapd: handle_probe_req: send failed Sep 24 21:37:15 Fri Sep 25 00:35:58 2020 daemon.notice hostapd: send_auth_reply: send failed Sep 24 21:37:49 logread is being flooded with those, any idea what's up with that? master r14424 Sep 24 21:39:04 and at least one of my clients is offline Sep 24 21:50:25 I did a make kernel_menuconfig and changed what I needed to. How do I get it to save it to target/*? Sep 24 21:50:33 or does it do it automatically? Sep 24 21:50:56 it does Sep 24 21:51:02 Thank you Sep 24 21:51:10 but it can mess things up Sep 24 21:51:17 I just realized I never turned on the EDAC caches :( Sep 24 21:51:25 so use git diff for sanity Sep 24 21:52:00 Oh, I maintain a stub defconfig and I can run thruy menuconfig fast enough, but I want to ensure those octeon specific settings will be saved across cleans or distcleans Sep 24 21:52:12 DonkeyHotei: git diff doesn't work when there's not config in the first place (e. g. cortexa9). ;) Sep 24 21:52:20 *there's no Sep 24 21:54:20 then do a commit before using kernel_menuconfig Sep 24 21:54:24 then it will Sep 24 21:54:26 DonkeyHotei: Do i need to explicitly define -msoft-float as a target option, or will CONFIG_SOFT_FLOAT take care of it? Sep 24 21:54:52 I know you probably don't know offhand, but I figured I'd ask :) Sep 24 21:55:23 i'd check other softfloat targets Sep 24 21:55:56 *nod* I will, but cavium does wierd things with the FPU Sep 24 21:56:01 like.. not having one Sep 24 22:12:18 Grommish: Most (if not all) ath79 routers also don't have an FPU, so that's alright. Sep 24 22:13:24 I have CONFIG_SOFT_FLOAT enabled, no kernel FPU emulation enabled at all, and don't have any issues. Sep 24 22:18:37 many targets have no fpu Sep 24 23:43:40 Fri Sep 25 00:35:57 2020 daemon.notice hostapd: handle_probe_req: send failed ---> I've seen that happen when other software is messing with the interface Sep 24 23:44:51 hmmm Sep 24 23:45:05 does down mess with the interfaces directly? Sep 24 23:45:07 dawn* **** ENDING LOGGING AT Fri Sep 25 02:59:57 2020