**** BEGIN LOGGING AT Mon May 13 02:59:56 2019 May 13 08:14:17 After a make distclean, mvebu is now building proper for me May 13 08:14:37 With GCC 8 May 13 12:43:31 nbd: ping - do/have you tried building 4.19 on macos with out of tree modules e.g. wireguard, cake, apu2_leds ? May 13 13:04:45 * ldir widens that out May 13 13:05:23 has *anyone* tried building out of tree modules with 4.19 ? May 13 13:05:59 ldir: I can build ath10k-ct successfully on 4.19. May 13 13:07:01 and I'm assuming macos isn't involved May 13 13:17:42 These are the sort of errors I'm getting https://pastebin.com/XwcHQ2pi May 13 13:33:22 ath79: Trying to chase down why with `CONFIG_PHY_AR7100_USB` not set it still gets built and installed as a module. May 13 13:34:09 Shows as [=n] in `make kernel_menuconfig` -- anyone here looked into this before? May 13 13:36:20 ldir: Nope, I'm on Linux. (Ubuntu 19.10, specifically) May 13 13:41:34 jeffsf: kernel_*config doesn't use the .config values and instead assumes all kmods as =n for the linux config (so your current config has no influence on it) May 13 13:44:41 Hmmm, `# CONFIG_PHY_AR7100_USB is not set` is present in ath79/config-4.1x, as well as in ath79/nand/config-default May 13 13:45:57 It's not a big thing, as I'm hoping/trusting that the DTS results in the 7200 driver being bound to the device -- just one of those prickly annoyances May 13 13:46:28 (since I don't know of any NAND-equipped board on the ar71xx-series SoCs) May 13 13:46:49 I think I have to give up trying to build openwrt on macos May 13 13:47:54 was trying to avoid a VM type environment. May 13 13:48:38 ldir: I beat on it for a while on both macOS and FreeBSD, then decided to just use VirtualBox and Debian. Side effect is that I know "my" build environment is very close to the "official" builders May 13 13:49:50 I've been on macos for 18months or so. Sadly samba4 is off limits because of macos and now it looks like building the kernel itself is going the same way. So to be blunt I'm a bit pissed off. May 13 13:54:27 ldir: are you using this new testing_kernel option? May 13 13:55:14 yes - that's the quick way of doing it. But I have exactly the same problem if I edit the target kernel version directly. May 13 13:55:29 ok May 13 13:55:32 so it's a '4.19' related issue. May 13 13:59:05 it turns out I can make it break on 4.14 if I install 'libelf' which then make 4.14 try to build objtool, which barfs for other reasons. May 13 14:00:05 e.g. https://pastebin.com/2M66hR3C May 13 14:02:25 ldir: if you want a decent machine for building, apply for an account https://cfarm.tetaneutral.net/ May 13 14:02:34 though working on a remote server is sometimes annoying May 13 14:02:55 but fast build is appreciable :) May 13 14:04:59 * Hellsenberg has to try building openwrt with distcc May 13 14:11:43 distcc and cross-compilation? good luck :) May 13 14:13:13 right, forgot about that little detail May 13 14:38:09 Is underscore _ allowed in package names? May 13 14:48:57 tmn505: well there's package/utils/spidev_test May 13 14:50:20 ynezz: the directory is called that way but package has PKG_NAME:=spidev-test May 13 14:52:38 ok, then libs/dmx_usb_module May 13 14:53:00 so it's probably ok May 13 14:58:54 ynezz: I got weird issue when building all kernel modules https://pastebin.com/NRzQc7vw. But the kernel modules built fine, dependencies are correct, but only when I replaced the underscore with dash in package name it completes the build. May 13 15:13:22 tmn505: hm, mozno jakisz lokalny blad v /tmp ? :) http://downloads.openwrt.org/snapshots/targets/mvebu/cortexa9/packages/kmod-bluetooth_6lowpan_4.14.115-1_arm_cortex-a9_vfpv3.ipk May 13 15:14:41 s,/tmp,openwrt/tmp, May 13 15:15:45 I remember hitting something similar as well, maybe it's related to some weird issue with v4/v6 kernel module handling? May 13 15:17:32 that's what makes it weird, might be that my host system makes it weird. I usually remove tmp when weird issues occure. I'll try with older revision, remember it was ok 2 weeks ago. May 13 15:29:02 Is the tree-wide plan to (eventually) move to mfgr,board-name for all boards? May 13 17:36:24 * ldir was wondering when xback would be along :-) May 13 21:07:05 Any clever ways to deal with a single file that changed on `master` without rebasing an entire PR? May 13 21:07:44 Anything better than heckout the file from `master` in one "pre-commit" then squash in with the rest? May 13 21:12:30 Ugh, NM -- board names changed, so I've got more work to do May 13 23:20:53 /nick jeffsf May 13 23:55:54 I need to build a version of libc.so for OpenWrt with symbols May 13 23:56:16 Is there an option in menuconfig to do this? May 13 23:56:29 disable stripping and rebuild May 13 23:56:32 but it will be huge... May 13 23:56:37 (i tried this recently) May 13 23:56:55 rebuild the toolchain? May 13 23:57:07 like, huge to the point of the firmware images won't build because theyre too large for the flash May 13 23:57:25 shouldnt be necessary to do the whole toolchain May 13 23:57:41 I only find libc as part of the poolchain May 13 23:57:44 toolchain May 13 23:58:19 idk what the target for libc is called May 13 23:59:06 you're still dealing with that python+libsodium thing? May 13 23:59:19 did you try rebuilding with just normal strip instead of sstrip? May 13 23:59:32 Yes. May 13 23:59:54 Which package? May 13 23:59:59 everything May 14 00:00:13 I got a version of libsodium with symbols now. May 14 00:00:21 I am getting fut=rther May 14 00:01:12 But the PyNacl package is still complaining about missing symbols in libc May 14 00:01:25 So including the symbols definirtely works May 14 00:02:09 I only need specific libraries with symbols May 14 00:02:24 I need libsodium (done) and libc. May 14 00:02:53 did you try rebuilding with just normal strip instead of sstrip? May 14 00:03:31 The way I got this to work with libsodium was to build the libsodium package wioth debig info. May 14 00:05:06 did you try rebuilding with just normal strip instead of sstrip? May 14 00:05:16 I don't understand this. May 14 00:05:28 my thought was that it could be an issue with CONFIG_USE_SSTRIP=y, which is the default. i would try with CONFIG_USE_STRIP=y May 14 00:06:47 I was able to get rid of stripping oflibsoium by using RSTRIP:= in the Makefile May 14 00:07:01 The version of libsodium in pynacl May 14 00:07:06 also make sure CONFIG_USE_MKLIBS=y isn't set May 14 00:10:41 I am a little confused. Can you point me to an example of the usage of CONFIG_USE_STRIP and CONFIG_USE_MKLIBS?? May 14 00:11:48 in 'make menuconfig'. Global build settings ---> Binary stripping method May 14 00:11:53 I can't find it in any Makefiles May 14 00:12:02 Ah..... May 14 00:12:06 Understand May 14 00:12:16 Global build settings ---> Strip unnecessary functions from libraries May 14 00:12:33 'Strip unnecessary functions from libraries' isn't set by default May 14 00:12:44 if you manually set it, it could cause issues May 14 00:12:45 Let me check May 14 00:15:11 None of the strip options is set May 14 00:16:04 ok May 14 00:16:18 what is the actual error message you're seeing? can you pastebin? May 14 00:17:10 The error is in the initialization for pynacl. It's trying to do a relocate on libsodium. This requires symbols. May 14 00:17:39 Libsodium uses functions in libc. May 14 00:18:14 Let me check the actual python statement thats giving an error. May 14 00:19:23 Here's the python statement: May 14 00:20:39 from nacl._sodium import ffi, lib May 14 00:20:53 Pretty basic stuff in python. May 14 00:21:01 no, what is the *error* message? May 14 00:21:33 what exactly is it saying that leads you to believe you need to recompile libc with debug symbols? May 14 00:22:19 ImportError: Error relocating /usr/lib/python2.7/site-packages/nacl/_soidium.so May 14 00:22:37 explicit_bzero: symbol not found. May 14 00:22:47 bzero is a function in libc May 14 00:23:15 I used to get tge same error for symbols in libsodium until I included a version that had symbols May 14 00:23:34 I have to go out for a little. I'll be back in 10 minutes. May 14 00:25:07 have you tried instructing pynacl to use the openwrt-provided version of libsodium rather than the pynacl-vendored version of libsodium? May 14 00:40:40 There's no symbols in that version of libsodium May 14 00:40:58 i'm curious what you see if you run nm on your libc May 14 00:41:04 No symbols May 14 00:41:07 e.g. on my ath79 it's: staging_dir/toolchain-mips_24kc_gcc-8.2.0_musl/bin/mips-openwrt-linux-nm staging_dir/target-mips_24kc_musl/root-ath79/lib/libc.so May 14 00:41:40 I have been looking in build_dir May 14 00:41:50 actually that one is not stripped on mine May 14 00:41:53 uhm May 14 00:41:58 I can looking the staging_dir May 14 00:42:52 FYI. The problem is more than just stripping. May 14 00:43:18 Even if libsodium is built without stripping, all the symbols come up as undefined. May 14 00:43:48 I had to build libsodium using the debug option in makuconfig May 14 00:43:57 menuconfig May 14 00:46:42 i wonder if you run 'file' on your libc.so, if you see "no section header" May 14 00:47:12 that is a sign of sstrip e.g. ./build_dir/target-mips_24kc_musl/toolchain/ipkg-mips_24kc/libc/lib/libc.so: ELF 32-bit MSB shared object, MIPS, MIPS32 rel2 version 1 (SYSV), dynamically linked, no section header May 14 00:47:42 I just checked the libc in staging_dir/ toolchain. Looks like it has symbols May 14 00:47:55 staging_dir is unstripped, yeah May 14 00:48:02 OK May 14 00:48:30 sstrip will break readelf/nm, quite possibly it breaks python too May 14 00:49:12 your issues really might go away if you just rebuilt everything from scratch with CONFIG_USE_STRIP=y, rather than the default CONFIG_USE_SSTRIP=y May 14 00:49:19 Let me try that version of libc. May 14 00:49:20 but maybe someone who knows more than me can speak to this May 14 00:49:55 by rebuilding with *no* stripping you sidestepped the fact that things were stripped with 'sstrip' previously May 14 00:50:16 If it works, I'm going to need to understand how to build libsodium (at least the version bundled in pynacl) and a version of libc which has symbols. May 14 00:50:20 ¯\_(ツ)_/¯ May 14 00:50:27 ????? May 14 00:50:53 shrug May 14 00:51:02 your issues really might go away if you just rebuilt everything from scratch with CONFIG_USE_STRIP=y, rather than the default CONFIG_USE_SSTRIP=y May 14 00:51:06 I can try that. May 14 00:51:24 But that's not the default method for building packages, right? May 14 00:51:28 nope May 14 00:51:30 sstrip is May 14 00:51:35 Right. May 14 00:51:42 it wasn't always iirc May 14 00:51:43 In any case, let me see if this works. May 14 00:52:01 Thanks greatly for your help. May 14 00:55:04 brucethompson: so if the *unstripped* version works, there is something you could try, before rebuilding everything May 14 00:55:22 alongside the compiler there should be a binutils strip (not the sstrip...) May 14 00:55:31 OMG. I just made a mistake and moved libc.so to libc.sav May 14 00:55:37 er May 14 00:55:44 Now I'm dead May 14 00:55:48 not quite May 14 00:55:55 OK May 14 00:56:16 How to recover?? May 14 00:56:18 you can possibly reboot into failsafe, mount the jffs2 overlay, and undo that change May 14 00:56:27 OK May 14 00:56:55 i was going to suggest using binutils strip to strip the libc.so in staging_dir, and install that one May 14 00:57:08 it would let you know if normal strip works without rebuilding everything May 14 00:57:38 Nope. I'm dead May 14 00:57:49 :( May 14 00:58:28 I will try to fix this first. May 14 00:58:32 Gotta be a way May 14 01:20:16 Yeah!!! Fixed. May 14 01:21:03 I'm still getting the same error with the new lic with symbols. May 14 01:21:06 libc May 14 01:22:31 I might be wrong abut libc. I'm going to check libsodium some more. It may be a symbol I am missing there. **** ENDING LOGGING AT Tue May 14 02:59:57 2019