**** BEGIN LOGGING AT Wed Jan 27 02:59:57 2021 Jan 27 09:34:15 Morning! Seems internet and ZNC were down here since 1:30 ;) Jan 27 09:34:28 Did I miss something? Jan 27 09:35:18 Morning! Jan 27 09:35:30 Herrie: nothing happened :) Jan 27 09:35:38 Tofe: OK :) Jan 27 09:35:58 For this Volla kernel... More or less figured out what it's doing.... Jan 27 09:36:54 Seems it kicks off these bits which will generate a new .config based on auto.conf: https://github.com/JamiKettunen/android_kernel_volla_mt6763/blob/android-9.0/Makefile#L563 Jan 27 09:37:25 Seems for building the kernel they use this script: https://github.com/JamiKettunen/android_kernel_volla_mt6763/blob/android-9.0/build.sh Jan 27 09:37:58 Seems they there generate the .config based on the provided defconfig by default Jan 27 09:38:10 So we'd need to emulate that in our build somehow... Jan 27 09:39:57 So I guess I could do something in the do_configure step (i.e. prevent it from running this) or add a step to run https://github.com/JamiKettunen/android_kernel_volla_mt6763/blob/android-9.0/build.sh#L38 Jan 27 09:43:38 Ah seems something like this might work :) https://community.nxp.com/t5/i-MX-Processors/Kernel-config-file-used-by-Yocto/m-p/399859/highlight/true#M58909 Jan 27 09:49:19 ah, yes, in the end it's similar to what we do for other target, only we copy it directly in the "cache" .config Jan 27 09:50:00 but .config is usually generated by do_configure, I'm a bit surprised Jan 27 09:54:00 Tofe: Well when I copy the defconfig like we usually do, it overwrites it, it seems Jan 27 09:56:09 So I end up having flags in the defconfig that shouldn't be there Jan 27 10:05:51 I end up having a .config in my kernel build directory that has different contents (way bigger) v.s. the provided defconfig Jan 27 10:06:18 And it has some flags that cause issues such as CONFIG_USB_LIBCOMPOSITE=m which come from auto.conf Jan 27 10:14:21 I wonder where it takes these values Jan 27 10:18:17 from auto.conf ;) Jan 27 10:18:28 By running auto.conf.cmd it seems Jan 27 10:19:01 I.e. my provided defconfig = 12kb, I end up with a .config that's 100kb larger Jan 27 10:21:01 Herrie: but the Makefile is a non-standard one ? Jan 27 10:21:12 Seems standard Jan 27 10:21:22 Seems just that the values in auto.conf differ from our regular kernels Jan 27 10:21:29 I.e. tissot has CONFIG_USB_LIBCOMPOSITE=y instead of CONFIG_USB_LIBCOMPOSITE=m Jan 27 10:21:36 ok, I never remarked that this possibility was there Jan 27 10:21:42 Also for Tissot we end up with a way larger .config Jan 27 10:23:31 Just where this auto.conf comes from I wonder, let me dig a bit Jan 27 10:49:58 Hmmz all a bit of a mystery to me... Jan 27 10:54:19 I mean I have .config in my build directory, no clue where it comes from... Same for autoconf.h etc :S Jan 27 10:57:46 autoconf ? for a kernel build ? how bizarre Jan 27 11:03:23 https://paste.ubuntu.com/p/f9fQVC8NPB/ Jan 27 11:48:34 It's confusing Jan 27 11:51:34 Yeah.... Jan 27 11:52:54 JaMa: You have any idea how all this works? ^ Jan 27 12:14:42 the part which makes defconfig smaller is calling savedefconfig which produces minimized version (which doesn't list all the values) Jan 27 12:15:12 so with this small defconfig, if you copy it as .config and run e.g. make oldconfig it should produce the same big .config file you had before Jan 27 12:15:32 and for kernel upgrades it will just follow whatever default values are set for new options Jan 27 12:16:05 so it makes it easier to compare minimalized defconfigs without all the churn from new options Jan 27 12:16:50 btw: I'm finishing upgrade to Qt 5.15.2, it seems to work fine on emulator, but I haven't rebased qtwebengine yet (doing it now), do you want this backported to gatesgarth? Jan 27 12:29:16 JaMa: I'm not really sure even how this autoconf.h is generated by the build and where it's values come from... It's all a big mystery :S Jan 27 12:37:39 https://serverfault.com/questions/568395/what-is-creating-the-generated-autoconf-h Jan 27 13:35:50 JaMa: Where should I run make oldconfig? In do_configure_prepend or... ? Jan 27 15:58:02 Herrie: kernel recipe should do it automatically already in do_configure Jan 27 15:58:04 meta/classes/kernel.bbclass:KERNEL_CONFIG_COMMAND ?= "oe_runmake_call -C ${S} CC="${KERNEL_CC}" LD="${KERNEL_LD}" O=${B} olddefconfig || oe_runmake -C ${S} O=${B} CC="${KERNEL_CC}" LD="${KERNEL_LD}" oldnoconfig" Jan 27 15:58:30 JaMa: This is kernel for Vollaphone Jan 27 15:58:40 Some Mediatek fancy 4.x one Jan 27 15:58:45 They use a custom build script it seems Jan 27 15:58:52 but it's going to inherit kernel.bbclass, right? Jan 27 15:59:03 JaMa: Yes seems so Jan 27 15:59:05 https://github.com/JamiKettunen/android_kernel_volla_mt6763/blob/halium-9.0/build.sh Jan 27 15:59:20 Basically they seem to run https://github.com/JamiKettunen/android_kernel_volla_mt6763/blob/halium-9.0/build.sh#L36 Jan 27 15:59:26 And then regular make Jan 27 16:00:56 looks like all of this build.sh should be already supported by regular kernel.bbclass/kernel_android.bbclass Jan 27 16:01:32 JaMa: Yes I would say so... Just when I run our kernel like normal I end up with some m flags in defconfig that cause kernel build to fail Jan 27 16:01:47 When using their build script these are set to y instead of m Jan 27 16:02:03 And these values come from auto.conf it seems Jan 27 16:02:23 most likely some dependency of these flags is m in our config causing them to end as m as well Jan 27 16:02:25 Specifically CONFIG_USB_LIBCOMPOSITE=m is problemnatic Jan 27 16:02:49 check the Kconfig file what this flag depends on and if the dependencies are set as y in our defconfig Jan 27 16:07:17 https://github.com/shr-distribution/meta-smartphone/commit/3b6578ea0ebe4acaa1bb8be4ed28c2af7900ba85 I believe there was better example with more proper fix, but already in 2012 and haven't noticed it in git log --oneline Jan 27 16:08:07 JaMa: Ah that might be a good workaround ;) Jan 27 16:08:18 I think their build script is interfering a bit with our build Jan 27 16:08:31 Since they use a .out directory and we have a build directory when we're not building in-tree Jan 27 16:08:48 yes, I would be surprised if it worked in our setup Jan 27 16:09:07 https://github.com/JamiKettunen/android_kernel_volla_mt6763/blob/halium-9.0/build.sh#L10 Jan 27 16:09:11 kernel.bbclass does a lot of things and expects them to be done this way (e.g. for packaging) Jan 27 16:13:36 is Vollaphone your next target? I haven't noticed it before and from the homepage not sure how it compares with pinephone, librem5 after watching indiegogo, I still trust the later 2 more on delivering something usable Jan 27 16:15:57 JaMa: not ours, but looks like someone would like to try a port Jan 27 16:16:08 Tofe/JaMa; Yes correct Jan 27 16:16:17 I don't have a Vollaphone Jan 27 16:16:21 Just someone would like to try Jan 27 16:16:29 They have UBPorts running with Halium 9 Jan 27 16:16:38 So LuneOS should be easy (except for the kernel for now) Jan 27 16:16:39 ah I see Jan 27 16:16:52 I'll guess I'll do the hackidy way you pointed to update the values Jan 27 16:17:38 Qt 5.15.2 seems to work fine on emulator even with qtwebengine updated, I will probably merge it to meta-qt5/gatesgarth as well, then it will be just 2 commits for meta-webos-ports Jan 27 16:19:06 JaMa: thanks, yes should be a minor update Jan 27 17:01:14 JaMa: Seems it's somehow not working Jan 27 17:01:53 Recipe: https://paste.ubuntu.com/p/fzKKcxT6Xz/ Jan 27 17:02:34 https://paste.ubuntu.com/p/y7WmjkpB6S/ Jan 27 17:11:05 strange, please paste log.do_configure as well Jan 27 17:14:47 JaMa: https://paste.ubuntu.com/p/JCb6vnG7Xn/ Jan 27 17:57:24 interesting it doesn't show CONFIG_USB_LIBCOMPOSITE at all in log.do_configure, maybe some bug in Kconfig stuff in this kernel, but I'm out of ideas without actually trying it locally Jan 27 17:59:19 JaMa: Ah OK Jan 27 17:59:35 It goes via de USB Gadget I think Jan 27 18:03:19 Herrie: your defconfig is taken from ./arch/arm64/configs/k63v2_64_bsp_defconfig ? Jan 27 18:03:44 JaMa: Well I have a local one that I copy Jan 27 18:03:45 it seems to be missing CONFIG_CONFIGFS_FS Jan 27 18:04:55 This one: https://paste.ubuntu.com/p/7kSQ9g6TwX/ Jan 27 18:05:41 try to add CONFIG_CONFIGFS_FS into it Jan 27 18:06:10 otherwise it ends as M Jan 27 18:06:12 Userspace-driven configuration filesystem (CONFIGFS_FS) [M/y/?] (NEW) Jan 27 18:06:42 OK Jan 27 18:09:12 Building now, should know soon I guess Jan 27 18:11:48 I still have it as m in defconfig, but getting another error now it seems: /home/herrie/LuneOS/volla/webos-ports/tmp-glibc/work-shared/yggdrasil/kernel-source/kernel/sched/tune.c:76:13: error: 'met_dvfs_info3' defined but not used [-Werror=unused-variable] Jan 27 18:21:19 that's probably just caused by newer gcc, either add -Wno-error=unused-variable or better fix that (backport should be available) Jan 27 18:21:58 JaMa: Yeah I'm looking into it... define is triggered by a defconfig flag that comes from somewhere Jan 27 18:22:20 However I'll try with the halium-9.0 kernel branhc instead of android-9.0 since there's some changes to this specific bit it seems Jan 27 18:38:54 That branch has other issues, I guess I'll try to cherry pick a fiix Jan 27 21:45:51 JaMa: This builds now: https://paste.ubuntu.com/p/D99pM776MV/ Jan 27 21:46:24 Seems the trick was to add "oe_runmake oldnoconfig" after the kernel_conf_variable_fixups **** BEGIN LOGGING AT Wed Jan 27 21:48:20 2021 Jan 27 22:05:41 Herrie: I'm still curious as to why, for this kernel in particular, our good old "copy to defconfig" doesn't work Jan 27 22:06:17 anyway, off to bed now :) Jan 27 22:11:11 Tofe: Yeah same here... It's MTK so who knows Jan 27 22:11:19 And quite modified by the looks of it as well Jan 27 22:11:23 So who knows Jan 27 22:11:27 Here in bed too :O Jan 27 22:11:30 :P Jan 27 22:11:32 gn8 **** BEGIN LOGGING AT Thu Jan 28 02:20:38 2021 **** ENDING LOGGING AT Thu Jan 28 03:00:03 2021