**** BEGIN LOGGING AT Fri Mar 13 03:00:43 2020 Mar 13 09:51:04 ynezz: ack for switching to x86 to 5.4 Mar 13 09:52:50 I know this will break building on !linux but the breakage will get more visibility and (I hope) that will prompt a fix from someone who can work out how to fix it. Mar 13 09:54:38 sometimes things have to really break before they get fixed Mar 13 10:16:38 ldir: have you tried to symlink those missing include files? Mar 13 10:17:07 ldir: I mean, the one in the staging dir to your macOS /usr/include Mar 13 12:03:30 mamarley: The AC pro is more or less the same as an AC lite Mar 13 12:03:49 i have an opened AC-Lite with serial. Will look into this later Mar 13 12:05:07 nbd: ping Mar 13 12:06:20 nbd: was there any particular reason for not allow keep-alive for post requests in uhttpd? After allow keep-alive for POST I am seeing vast improvements in performance (~1.3s average page load time including all assets and ubus calls vs. ~4.5 without) Mar 13 12:08:25 i don't remember Mar 13 12:08:45 ok Mar 13 12:48:13 blocktrron: Awesome, thanks! Mar 13 12:48:23 hmmmz Mar 13 12:48:31 schools just got shut down for 5 weeks Mar 13 12:48:40 that is pretty drastic Mar 13 12:49:52 not here yet, but I expect them to announce it any minute Mar 13 12:50:17 the first school in the town just got closed down due to a positve case Mar 13 12:50:23 blogic: babysitting next week? :p Mar 13 12:56:01 ynezz: didn't you recently touch pkg-config ? Mar 13 12:56:28 jow: yes, guilty Mar 13 12:56:33 :) Mar 13 12:56:49 what did I break this time? Mar 13 12:56:50 GH#2832 might be related Mar 13 12:57:10 thats luci? Mar 13 12:57:25 https://github.com/openwrt/openwrt/pull/2832 Mar 13 12:57:36 no, breaking kmod build for x86/64 Mar 13 12:57:52 kbuild uses pkg-config to discover libelf Mar 13 12:58:13 ok, where is that breakage happening? Mar 13 12:58:15 we force it to use our own wrapper to discover our host build libelf and not the system wide one Mar 13 12:58:45 on x86/64, wirth external kmods Mar 13 12:58:50 I just stumbled over it myself Mar 13 12:58:57 make[5]: *** No rule to make target 'tools/objtool/objtool', needed by '/home/jow/devel/lede/staging.git/build_dir/target-x86_64_musl/linux-x86_64/siit-1.2/siit.o'. Stop. Mar 13 12:59:40 will try to revert that pkg-config bump, lets see if it works Mar 13 13:00:39 ynezz: babysitting next 5 weeks Mar 13 13:00:46 doing it myself Mar 13 13:00:54 the advantage of home office ;) Mar 13 13:01:13 oh, I've thought it's bi-weekly for you Mar 13 13:01:25 * mamarley doesn't think it counts as babysitting if they're your kids… Mar 13 13:02:38 ok, wrong wording Mar 13 13:02:45 but the result is same Mar 13 13:03:04 ynezz: saver with me for now Mar 13 13:03:38 mamarley: correct its more being a servant when its your own Mar 13 13:03:48 ynezz: after reverting the change to pkgconf, the kkmods work normally for me Mar 13 13:04:02 build, rather Mar 13 13:04:30 seems that pkgconf behaves differently in subtle ways Mar 13 13:04:53 schools will be shutdown in berlin next tuesday Mar 13 13:05:49 jow: indeed, seems so Mar 13 13:12:55 jow: anyway, it really seems to be maintained somehow, some mainstream distributions use it already (choosed randomly Fedora) Mar 13 13:13:42 ynezz: if "maintained" means "change behaviour in subtle way" then yes, its maintained Mar 13 13:13:51 * karlp laughs Mar 13 13:14:45 I mean its pkg-config, its purpose is to read a value out of a plaintext file and perform some minimal variable expansion, how much maintenance does it need? Mar 13 13:14:58 I guess I know, it probably lacked systemd integration Mar 13 13:15:26 (or, someone just said, "last commit is more than a year ago, it must be broken and useless, replace it immediately") Mar 13 13:15:45 or they send some patches/fixes and nobody cared Mar 13 13:15:56 so someone forked it and cares now Mar 13 13:16:55 "pkgconf makes optimizations to the linker flags in both the case of static and shared linking in order to avoid overlinking binaries and also simplifies the CFLAGS and LIBS output of the pkgconf tool for improved readability." Mar 13 13:17:09 that sounds exactly like the kind of thing breaking the kbuild now Mar 13 13:17:18 "optimizing ldflags" Mar 13 13:17:33 some patches / fixes aren't. Mar 13 13:22:09 jow: this is 4.19? Mar 13 13:22:29 and how is that pkgconfig called, I dont see it in build output Mar 13 13:23:17 it is invoked early by kbuild to determine libelf linker flags Mar 13 13:23:34 later objtool is built as prereq for some kmods Mar 13 13:23:44 which fails due to bad flags Mar 13 13:23:53 leading to "No rule to make target 'tools/objtool/objtool', needed by" Mar 13 13:24:10 it took me several hours the last time to trace that down, I forgot about the details Mar 13 13:24:37 trying to find a reproducer first Mar 13 13:24:53 and yes, it is 4.19 Mar 13 13:24:55 I'm on 5.4, did `ln -s $(pwd)/../packages.git/net/siit package/network/siit; make package/siit/{clean,compile} V=sc -j1 JOBS=1 2>&1 | tee build.log` Mar 13 13:25:04 and not seeing that error, different one Mar 13 13:25:43 what's the JOBS=1 for? Mar 13 13:25:54 do you have libelf installed on your host system? Mar 13 13:25:59 are some packages automatically running more jobs? Mar 13 13:27:20 jow: libelf1 Mar 13 13:33:01 anyway, it makes me wonder why we don't have same issue with gpio-button-hotplug Mar 13 13:38:20 jow: and you're building with SDK as well? Mar 13 13:41:40 git show 5f8e587240 Mar 13 13:47:16 why there is no siit here https://downloads.openwrt.org/snapshots/faillogs/x86_64/packages/ ? Mar 13 13:48:15 that should be 4.19 with SDK Mar 13 13:52:26 ynezz: I build with buildroot Mar 13 13:52:35 anyhow, pkgconf is the culprit indeed Mar 13 13:52:48 it "optimized" the library search path away Mar 13 13:55:26 nice optimization Mar 13 14:03:39 ynezz: https://pastebin.com/ufdbJ8Ez Mar 13 14:04:08 (line 9 should read "with pkgconf") Mar 13 14:05:12 compare line 15 with line 41, thats the main culprit Mar 13 14:07:20 hm, getting a bunch of ICEs with cmake bump 3.16.5 Mar 13 14:07:28 the main kernel build will skip building objtool (and forcibly disable CONFIG_STACK_VALIDATION) because it cannot link libelf Mar 13 14:08:02 any external kmod will assume CONFIG_STACK_VALIDATION is enabled and kbuild will try to invoke objtool which doesn't exist in the precompiled kernel soure tree Mar 13 14:16:31 ok, so it seems like mangix (or myself) needs to fix this in pkgconf and send the fix upstream OR we should revert it Mar 13 14:19:21 I found a simpler reproeducer Mar 13 14:20:45 ynezz: https://pastebin.com/Gu74ANde Mar 13 14:20:59 ynezz: can you run the command in your buildroot and tell me what it prints? Mar 13 14:21:50 -lelf Mar 13 14:22:00 yeah, thats wrong Mar 13 14:22:34 so it's using my host's libelf, right? Mar 13 14:22:38 that means the kernel's objtool will link your system libelf which might or might not be linkable Mar 13 14:22:50 in my case it is not linkable Mar 13 14:23:05 yep, it's pretty clear now, thanks Mar 13 14:35:45 ynezz: I think I found the issue Mar 13 14:36:05 ynezz: pkgconf has the concept of a "system library dir" which defaults to ${exec_prefix}/lib Mar 13 14:36:17 any -L flag matching this system library dir is filtered Mar 13 14:36:35 by default, exec_prefix is ./staging_dir/host Mar 13 14:36:59 which means the system library dir is ./staging_dir/host/lib Mar 13 14:37:20 exactly the directory we want to retain Mar 13 14:37:27 luckily there's a configure knob Mar 13 14:38:27 --with-system-libdir Mar 13 14:38:50 we should probably set --with-system-includedir too Mar 13 14:39:45 alternatively we could add --keep-system-cflags and --keep-system-libs to our pkg-config wrapper Mar 13 14:39:58 nice find, going to be afk for a few hours (kids) Mar 13 14:40:21 which is probably safer, since the configure flags will probably be "optimized" away by someone in the future Mar 13 15:03:34 It looks like I've just got a 5.4 build on macos with the aid of a horrible hack. Going to need to discuss the proper way to do this with an adult. Mar 13 15:07:05 Doing cp -r ./stagind_dir/toolchain-x86_64_gcc-9.2.0_musl/include/asm* ./staging_dir/host/include allows 'make target/linux/install' to complete. Mar 13 15:09:02 what exactly is failing? Mar 13 15:09:34 coming to that... have a lot of scrollback to go through :-) Mar 13 15:10:23 https://paste.ubuntu.com/p/gPcbfWZsKc/ Mar 13 15:11:31 remember that relative include thing that broke mips kernel on OS X the last time= Mar 13 15:11:34 ? Mar 13 15:11:37 Maybe this one is similar Mar 13 15:11:50 it again seems to be a bootstrap issue. Kernel build expecting installed kernel to complete Mar 13 15:12:52 I had a more useful log somewhere... sorry, still building, or hoping for a build. Mar 13 15:15:24 https://elixir.bootlin.com/linux/v5.4.25/source/arch/x86/entry/vdso/Makefile#L61 Mar 13 15:16:28 and in your build: ... -I./tools/include -I./include/uapi -I./arch/x86/include/uapi ... Mar 13 15:17:06 while make claims to be in /Users/kevin/wrt/build_dir/target-x86_64_musl/linux-x86_64/linux-5.4.24 Mar 13 15:18:03 so you don't have asm/ in any of /Users/kevin/wrt/build_dir/target-x86_64_musl/linux-x86_64/linux-5.4.24/tools/include, /Users/kevin/wrt/build_dir/target-x86_64_musl/linux-x86_64/linux-5.4.24/include/uapi or /Users/kevin/wrt/build_dir/target-x86_64_musl/linux-x86_64/linux-5.4.24/arch/x86/include/uapi ? Mar 13 15:18:51 https://paste.ubuntu.com/p/Th3WMnp9Xd/ - I think is a better log... might be exactly the same. Mar 13 15:19:01 let me go and parse those paths Mar 13 15:20:59 so /Users/kevin/wrt/build_dir/target-x86_64_musl/linux-x86_64/linux-5.4.24/tools/include - has asm but not types.h in it Mar 13 15:21:30 I could imagine that this a genuine bug Mar 13 15:21:52 just nobody noticed because most probably have some kernel headers installed already when building linux Mar 13 15:22:24 and /Users/kevin/wrt/build_dir/target-x86_64_musl/linux-x86_64/linux-5.4.24/include/uapi - has asm-generic but no asm (!) Mar 13 15:23:27 the asm-generic does contain a types.h Mar 13 15:23:58 yeah, asm/types.h is supposed to be installed system wide by the header isntall rule Mar 13 15:24:12 in our case thats not system wide but within the toolchain prefix Mar 13 15:24:31 and /Users/kevin/wrt/build_dir/target-x86_64_musl/linux-x86_64/linux-5.4.24/arch/x86/include/uapi only contains asm (!) but no types.h within it. Mar 13 15:28:10 shouldn't I be looking at staging dir? Or are we saying it's going wrong before we get to place things in staging ? Mar 13 15:32:31 can you try thins random hack? https://pastebin.com/Re905d0w Mar 13 15:32:48 I don't expect it to be working but try anyway Mar 13 15:33:12 no need to clean, you can run make target/linux/install right away Mar 13 15:33:42 * ldir needs at least 3 heads - a sound supervisor head - a linux/openwrt building head - a domestic bliss head - the paging of data between those into 1 active head is too much Mar 13 15:34:08 I do need to clear those headers that I copied though... Mar 13 15:34:15 ah yes, right Mar 13 15:38:48 You're right - it didn't work - gcc got invoke - gcc -Wp,-MD,arch/x86/entry/vdso/.vdso2c.d -Wall -Wmissing-prototypes -Wstrict-prototypes -O2 -fomit-frame-pointer -std=gnu89 -O2 -I/Users/kevin/wrt/staging_dir/host/include -Wall -Wmissing-prototypes -Wstrict-prototypes -I/Users/kevin/wrt/staging_dir/host/include -I./tools/include -I./include/uapi -I./arch/x86/include/uapi -o arch/x86/entry/vdso/vdso2c arch/x86/entry/vdso/vdso2c.c Mar 13 15:38:48 -L/Users/kevin/wrt/staging_dir/host/lib Mar 13 15:42:33 the next try would be: https://pastebin.com/e06RNQ5w Mar 13 15:43:03 but that feels wrong Mar 13 15:48:10 doesn't it also get overwritten by export HOST_EXTRACFLAGS=-I$(STAGING_DIR_HOST)/include below it? Mar 13 15:48:20 Hey all. So attempting to build an image and set the default governor to conservative but simply doesn't want to work. Relevant lines from .config (make defconfig checked) are: Mar 13 15:48:23 ## Kernel Governor#CONFIG_KERNEL_CPU_FREQ_GOV_PERFORMANCE=y# CONFIG_KERNEL_CPU_FREQ_GOV_POWERSAVE is not set# CONFIG_KERNEL_CPU_FREQ_GOV_SCHEDUTIL is not setCONFIG_KERNEL_CPU_FREQ_GOV_CONSERVATIVE=y# CONFIG_KERNEL_CPU_FREQ_GOV_USERSPACE is not setCONFIG_KERNEL_CPU_FREQ_GOV_ONDEMAND=y# CONFIG_KERNEL_CONFIG_CPU_FREQ_DEFAULT_GOV_ONDEMAND is not set# Mar 13 15:48:24 CONFIG_KERNEL_CONFIG_CPU_FREQ_DEFAULT_GOV_PERFORMANCE is not setCONFIG_KERNEL_CONFIG_CPU_FREQ_DEFAULT_GOV_CONSERVATIVE=yCONFIG_KERNEL_PREEMPT_NONE=y# CONFIG_KERNEL_PREEMPT_VOLUNTARY is not set# CONFIG_KERNEL_PREEMPT is not set Mar 13 15:48:38 ldir: oops yes, I missed that the export also (re)sets the value Mar 13 15:48:51 ldir: so best is to append the toolchain dir to the existing export ocmmand Mar 13 15:49:20 (default governor remains performance but if I change with echo 'conservative... works fine) Mar 13 15:50:58 leperet: CONFIG_KERNEL_* only works for kernel options known to openwrt (those registered in config/*) Mar 13 15:51:06 all others will be purged on make defconfig Mar 13 15:51:33 jow: that produces all sorts of typedef redefined horribleness - hmm Mar 13 15:51:40 ldir: yeah, as expected Mar 13 15:54:31 ldir: the toolchain include would be wrong anyway since vdso2c is a host util which expects a Linux host, so we need to supply it with linux host headers Mar 13 15:54:59 ldir: question is how much of asm/types.h it actually needs, maybe it can be satisfied with a stub header Mar 13 15:55:11 should say that I'm using this patch https://github.com/openwrt/openwrt/pull/1456 Mar 13 15:55:34 how would i change the default governor if I weren't using this patch? Mar 13 15:55:49 leperet: in thise case via make kernel_menuconfig Mar 13 15:55:51 (others do report it working for them and I suspect this will be merged soon) Mar 13 15:56:04 does that have limitations? Mar 13 15:56:16 it modifies the actual kernel config templates shipped by git Mar 13 15:56:27 which might or might not be a problem for your workflow Mar 13 15:56:49 I suspect it will be OK. Thank you jow :) Mar 13 15:58:36 let's provide it with an empty types.h and see what blows up :-) Mar 13 16:03:13 https://paste.ubuntu.com/p/39xsxSwSzK/ ok so that's what it wants Mar 13 16:13:56 cp toolchain-x86_64_gcc-9.2.0_musl/include/asm-generic/int-ll64.h host/include/asm/types.h Mar 13 16:22:15 jow: Is it that the thing that compiles vdso2c.c isn't cross-compile aware or is it not being invoked in a cross-compile manner? Mar 13 16:25:47 ldir: as I undertand it, vdso2c is a host utility which is mean to run on the host Mar 13 16:26:03 ldir: its just not portable, it assumes a Linux host with Linux kernel headers Mar 13 16:28:11 * ldir is being harassed by the dog... it wants its walkies Mar 13 17:17:31 * ldir is back Mar 13 17:26:46 jow: https://paste.ubuntu.com/p/7wpyKCTf8k/ makes it happy enough. Mar 13 17:31:36 So, how, and is it reasonable, to put that into stagind_dir/host/include/asm/types.h Mar 13 18:23:59 mamarley: i've just installed k5.4 on my ac lite and it works without any issue Mar 13 18:24:14 no crashes whatsoever Mar 13 18:24:54 Are you able to get a serial log from there? the only difference between these two is the additional switch chip, realized by an atheros switch Mar 13 18:25:14 blocktrron: Hmm, I wonder if I somehow accidentally installed one of my previous builds on mine, because on the one occasion that I did get it to quit looping and actually boot, the Ethernet also seemed dead. Mar 13 18:25:18 I will try it again. Mar 13 18:25:25 thanks Mar 13 18:25:49 i'm going to merge this patch anyway, as it seems to be solid improvement for 5.4 compared to the current state Mar 13 18:25:56 Yes, certainly. Mar 13 18:38:53 blocktrron: Yep, I must have just flashed the wrong image on it. I just tried again and it seems to be working fine. Mar 13 18:43:35 I installed it on my in-production UAP-AC-PRO as well and that also appears to work fine. Mar 13 18:52:18 awesome Mar 13 18:52:40 I'll incorporate the changes suggested on the ML, test with 4.19 and 5.4 ans merge then Mar 13 19:16:42 build #315 of arc770/generic is complete: Failure [failed tools] Build details are at http://buildbot.openwrt.org/master/images/builders/arc770%2Fgeneric/builds/315 blamelist: Adrian Schmutzler , Jun Su , Kevin Darbyshire-Bryant , Sungbo Eo Mar 13 20:46:58 jow: thoughts? https://git.openwrt.org/?p=openwrt/staging/ldir.git;a=commitdiff;h=4eb983b1c752bb3c23a8ba631799348c010c66c7 Mar 13 21:00:59 Hi any one used this? https://github.com/AdguardTeam/AdGuardHome Mar 13 21:52:10 ldir: that looks very well Mar 13 21:52:16 ldir: even less invasive than I though Mar 13 22:25:29 question for ath79 gurus: assuming I start working on porting mikrotik ar71xx devices to ath79, do we want to keep the "single image for all devices" approach or is this a no go? thx Mar 13 22:29:21 6aaa5ce2 and 8f93c05a59 would suggest there's no interest in this (although personally I like the single image approach), just want to confirm Mar 13 22:36:36 its a nice to have not mandatory Mar 13 22:38:00 blogic: i.e. single image? I agree, but I suspect that means writing an intermediary loader to apply the correct DTS at load time. Am I wrong? **** ENDING LOGGING AT Sat Mar 14 03:02:45 2020