| Git Clone URL: | https://aur.archlinux.org/linux-git.git (read-only, click to copy) |
|---|---|
| Package Base: | linux-git |
| Description: | Headers and scripts for building modules for the Linus Torvalds' Mainline Linux kernel |
| Upstream URL: | https://www.kernel.org |
| Keywords: | git kernel linux |
| Licenses: | GPL-2.0-only |
| Provides: | LINUX-HEADERS |
| Submitter: | jonathanio |
| Maintainer: | osimarr |
| Last Packager: | osimarr |
| Votes: | 13 |
| Popularity: | 0.003633 |
| First Submitted: | 2019-12-22 10:10 (UTC) |
| Last Updated: | 2026-04-20 08:57 (UTC) |
For reference, these are the recent commits I've built successfully.
#225 24-Aug-2020 04:46 linux-git 5.9rc2.r0.gd012a7190fc1-1
#224 17-Aug-2020 04:46 linux-git 5.9rc1.r0.g9123e3a74ec7-1
#223 10-Aug-2020 04:46 linux-git 5.8.r11991.gfc80c51fd4b2-1
#222 03-Aug-2020 04:46 linux-git 5.8.r0.gbcf876870b95-1
#221 27-Jul-2020 04:46 linux-git 5.8rc7.r0.g92ed30191993-1
I have a Jenkins job on my build server that builds the kernel every Monday morning in the early hours. This includes a full wipe of the build workspace beforehand so it's a fresh build. There are no current issues with the build for me right now.
I don't use dracut, so like @gardotd426 suggests, it seems that darcut is probably the issue as it's the only difference (that we're aware of) between the three of us.
Have you tried building it without dracut and from a clean workspace?
@gardotd426 I knew the kernel is not broken for the reasons outlined by you. My suspect was that something is changed in kernel build process which required changes in PKGBUILD, so this would affect only Arch.
Regarding dracut. I can assure, that dracut is fully fine in my setup, which is quite complicated (full disk encryption). I even have created custom scripts to facilitate my minor needs. I would recommend to any advanced Arch user to use dracut because it is at least as flexible as mkinitcpio.
Regarding the problem. It turns out, that depmod and modinfo (which were called by dracut) were complaining on garbage '/usr/lib/modules' previous kernel version installation which no longer existed (trashed by power failure). After clearing this not tracked by pacman gargabe I was able to compile package. Thanks for quick response!
@mxfm um, the "git repo" is literally the actual main Linux kernel repo. There's nothing wrong with it, if there were, the whole world would know. It would also mean that no other 5.9-rc1 kernels could be built using that source, which isn't true (I'm running one right now).
It's very likely a Dracut issue. Dracut on Arch is nowhere near ready for prime time on Arch Linux. An Arch dev reported not being able to use their keyboard and mouse when testing Dracut, etc.
But it's definitely not the git repo. If it's not the PKGBUILD, then it's your setup or dracut.
Approximately since 5.9.1.rc1 I cannot build correct package of linux-git. I use config file from mainline 'linux' package from core repo (I only modified two kernel options related to default localhost name and version config). After building the package, I cannot build initramfs image with dracut - it fails to include kernel modules with message like 'depmod: ERROR: failed to load symbols from /var/tmp/dracut.nBQFd0/initramfs/lib/modules/5.9.0-rc1-1-git/kernel/drivers/pinctrl/cirrus/pi nctrl-lochnagar.ko.xz: Exec format error'
Can someone confirm that he can build package from this script? Perhaps something has been broken in git repo?
I just did a complete rebuild for testing purposes with -j12 enabled. The build time came down from ~2 hours to 20 minutes while the laptop is still responsive. Wow! Thank you for this suggestion. This is more than I ever expected. I'll stick with this setting definitely ;)
@gardotd426 - Well in my case the build files stay near the source code, and are present even after reboots and /tmp is cleared.
And if you use "makepkg -e" it does perform an incremental build (assuming config was already done on the previous build).
@Harey - You should definitely go with a higher count, atleast "make -j8" if not "make -j12"
Wow, thank you for the fast and elaborate answers. I think I'll go with MAKEFLAGS -j4 (should be safe on a Ryzen 4800U) and maybe have a look at ccache.
Well I mean even if they're not using an AUR helper, /etc/makepkg.conf will have /tmp/makepkg set as the BUILDDIR so unless you set another BUILDDIR at compile time it'll still go to /tmp. But yeah, otherwise. It's definitely not impossible to make sure the files stick around, they just usually don't by default (but even if they do stick around, I'm not sure about "only rebuilding what's changed," I know that's a thing when manually compiling from source, but I don't know how well it works with PKGBUILDs, so I defer to you or someone else that knows on that part. I just use ccache).
Pinned Comments
osimarr commented on 2023-05-07 22:27 (UTC) (edited on 2023-05-12 04:05 (UTC) by osimarr)
Hi. Since the package became an orphan, I adopted it to continue what I actively maintain on https://aur.archlinux.org/pkgbase/linux-mainline-git, now on this AUR package.
The build script and config file will be up-to-date with Arch's official Linux package unless an update is required to be functional.
But I intentionally removed the documentation package because it consumes too much building time.
This package supports certain customization if compiled by the user. Check for the README file inside the package repo, or the .INSTALL message after package installation to learn how.