Package Details: linux-git 4.17rc1.r21.ga27fc14219f2-1

Git Clone URL: (read-only)
Package Base: linux-git
Description: The Linux kernel and modules (git version)
Upstream URL:
Keywords: git kernel linux
Licenses: GPL2
Provides: linux
Submitter: xduugu
Maintainer: Boohbah
Last Packager: Boohbah
Votes: 63
Popularity: 0.059643
First Submitted: 2011-05-30 07:43
Last Updated: 2018-04-18 07:39

Required by (191)

Sources (4)

Latest Comments

RubenKelevra commented on 2018-04-13 14:57

sarvex, there are other packages to fulfill this role, this package is for users who want to build a specific git-version.

RubenKelevra commented on 2018-04-13 14:56

ssorgatem, Line 9, 28, 96

sarvex commented on 2018-03-24 08:05

Cloning the git repository for building the kernel is a huge waste of bandwidth.

It would be far more efficient to use the mainline compressed tar from the website

ssorgatem commented on 2018-03-21 11:04


I maintain a couple packages whose based on this one... and I have a minor problem I haven't been able to solve. On GRUB, the kernels from this pakage and the ones based on it all show up the same, as "git" rather than a more informative name.

Where does that "git" come from? I haven't been able to find it to change it to something else... maybe it's obvious and I'm blind, though xD

Yardanico commented on 2018-03-07 10:07

+1 to bradmccormack100 , I have 600-700 KB/sec from, 2-3 MB/sec from github, and 9-10MB/sec from google mirror

bradmccormack100 commented on 2018-02-27 12:53


Is it worth updating the git source to git+ and use Google's mirror which is a LOT faster.

The original source clones at about 20-30KB a second a lot of the time. The updated mirror clones about 10MB/s.

This is at the same time of testing for both sources. I'm from Australia and the difference in speed is huge.

shunonymous commented on 2018-02-15 15:37

Packaging error with 4.16rc1

==> Starting package_linux-git-headers()...
cp: cannot stat 'drivers/media/dvb-core/*.h': No such file or directory
==> ERROR: A failure occurred in package_linux-git-headers().

and fixed.

HarlemSquirrel commented on 2017-12-11 20:13

I'm now having a new issue

make[4]: *** No rule to make target '/usr/lib/gcc/x86_64-pc-linux-gnu/7.2.0/include/stddef.h', needed by '/home/hs/builds/linux-git/src/linux/tools/objtool/fixdep.o'.  Stop.

EDIT: I was able to fix this by removing src folder and building again.

mweinelt commented on 2017-12-08 18:41

I would like to build this package inside a docker container, headless.

Can you introduce a way to bypass make olddefconfig and instead just use the defaults, for example by setting an environment variable like NONINTERACTIVE, UNATTENDED or HEADLESS?

HarlemSquirrel commented on 2017-12-03 19:40

Having some trouble with 4.15rc2

Finished making: linux-git 4.15rc2.r0.gae64f9bd1d36-1 (Sun Dec  3 14:37:22 EST 2017)
==> Installing linux-git package group with pacman -U...
loading packages...
resolving dependencies...
looking for conflicting packages...
error: could not open file /var/lib/pacman/local/linux-git-headers-4.15rc1.r0.g4fbd8d194f06-1/desc: No such file or directory
warning: could not fully load metadata for package linux-git-headers-4.15rc1.r0.g4fbd8d194f06-1
error: failed to prepare transaction (invalid or corrupted package)
==> WARNING: Failed to install built package(s).

ishitatsuyuki commented on 2017-10-16 07:36

Looks like the parallelism is actually reduced by some non-compiling processes. Is this known? The count of gcc running gradually decreases to 1 while submake are spawned.

darkbasic commented on 2017-09-29 15:49

> I would recommend changing the make command within the build section to include the paremeter "-j 4"

You should just put MAKEFLAGS="-j4" into /etc/makepkg.conf

agowa338 commented on 2017-09-25 14:00

I would recommend changing the make command within the build section to include the paremeter "-j 4"

@A.J This aur is not outdated. It always pulls in the latest release from git.
The Version number shown here is not the one actually built. The versionnumber of the generated package is gennerated using this command:
git describe --long | sed -E 's/^v//;s/([^-]*-g)/r\1/;s/-/./g;s/\.rc/rc/'

It seams, that this website does not support that kind of dynamic version numbers.

tylerjd commented on 2017-07-25 00:37

Even after this package was updated, it is still failing to build, with the same error as before. This can be worked around by taking out the defunct DocBook install line out of the PKGBUILD lines 175-177, you're looking to remove or comment these:

# add docbook makefile
install -D -m644 Documentation/DocBook/Makefile \

Maintainer please updating the PKGBUILD to reflect the new changes in the kernel I pointed to before.

EDIT: The maintainer fixed this and is no longer an issue

tylerjd commented on 2017-07-12 21:15

CTXz is right, Documentation/DocBook/Makefile is gone, it and the rest of the DocBook directory were removed in a 4.13 merge

CTXz commented on 2017-07-12 19:47

Getting the following error on build :

install: cannot stat 'Documentation/DocBook/Makefile': No such file or directory

Looks like DocBook is no longer present in 4.12

PhotonX commented on 2017-07-01 19:12

Thanks for the hint! Actually, I got an even more convenient proposal on the forums: However, I get silly config prompts when trying to build the kernel after the first git bisect run (see later posts in the same thread). I suppose that the package should build without user interaction so I must be doing something wrong...

devopsdeluxe commented on 2017-07-01 18:37


Yes you can:

Instead of specifying the branch, you can specify a specific commit.

PhotonX commented on 2017-07-01 15:44

I need to bisect the kernel versions to find a bug. Can I somehow use this package in combination with git bisect to do so? Thanks!

nabilh2020 commented on 2017-05-17 15:24

However linux-git 4.11 built successsfully, with no HDMI i915 sound.

nabilh2020 commented on 2017-05-17 15:23

Failed to build,
on BBEN HDMI Intel Cherrytrail Compute Stick,
tried restart building twice:
DEPMOD 4.12.0-rc1-g2b6b38b04c2e
sed: -e expression #2, char 44: unknown option to `s'
==> ERROR: A failure occurred in package_linux-git().
==> ERROR: Makepkg was unable to build linux-git.
==> Restart building linux-git ? [y/N]

thx1138 commented on 2017-04-03 15:20

==> Starting package_linux-git-headers()...
rm: cannot remove '/var/build/linux-git/pkg/linux-git-headers/usr/lib/modules/4.4.0-g765bdb4/build/Documentation/kbuild/Kconfig.*-*': No such file or directory


# Fix file conflict with -doc package
rm "${pkgdir}/usr/lib/modules/${_kernver}/build/Documentation/kbuild"/Kconfig.*-* || true

adding the " || true" on the end, so that the build does not quit with an error.

kaueraal commented on 2017-03-25 16:59

On 2016-12-20 21:33 I mentioned it doesn't build for me. It seems it's enough to create those directories. See for a patch here:

Unfortunately I've got no idea why this is necessary for me but not for others.

krash86.7 commented on 2017-03-22 13:48

This package fails for me here:
-> Cloning linux git repo...
Cloning into bare repository '/tmp/yaourt-tmp-krash86/aur-linux-git/linux'...
error: RPC failed; HTTP 504 curl 22 The requested URL returned error: 504 Gateway Time-out
fatal: The remote end hung up unexpectedly
==> ERROR: Failure while downloading linux git repo

I have traced it to this line in the PKGBUILD file:
Line 17:

If I change that from "git+https" to "git" only then the PKGBUILD works fine.

kaueraal commented on 2016-12-20 21:33

Doesn't build for me:

==> Entering fakeroot environment...
==> Starting package_linux-git()...
realpath: /home/alexander/.cache/pacaur/linux-git/pkg/linux-git/lib/modules/4.9.0-ge93b1cc/kernel: No such file or directory
make: *** [Makefile:1218: _modinst_] Error 1
==> ERROR: A failure occurred in package_linux-git().
:: failed to build linux-git,linux-git-headers package(s)

Boohbah commented on 2016-12-12 20:18

The CONFIG_MODVERSIONS bug was fixed (or worked around) in rc8 so I added it back to the config.

kozzi commented on 2016-12-05 07:38

@mhussain yes it will boot ok, at least for me. CONFIG_MODVERSIONS is unset, but I belive (I will try it), it will work even with CONFIG_MODVERSIONS enabled.

mhussain commented on 2016-11-27 19:47

hi there,
once this is compiled, will it now boot?
is the CONFIG_MODVERSIONS setting set to false in the config file?
if not how would I go about changeing this setting?
thanks for any help you may have! :)

talonz commented on 2016-11-13 01:54

does dkms not work on linux-git ? i cannot get virtualbox to work on 4.9.rc4 with the same error as blinkallthetime was getting

kozzi commented on 2016-11-06 10:59

There is rc4 but still need to disable CONFIG_MODVERSIONS option, because the patch does not make it yet

talonz commented on 2016-11-03 21:50

==> Validating source files with sha256sums...
linux ... Skipped
config ... FAILED
config.x86_64 ... Passed
linux-git.preset ... Passed
==> ERROR: One or more files did not pass the validity check!
:: failed to verify linux-git integrity

Boohbah commented on 2016-11-03 21:20

I disabled CONFIG_MODVERSIONS in the default configs and it should boot now.

Sorry for the breakage.

R00KIE commented on 2016-11-03 15:48


Thank you for the tips. I'll give this another go once a few DRM fixes (possibly interesting for me) get pulled by Linus.

kozzi commented on 2016-11-03 14:03


it is known issue. You need to disable CONFIG_MODVERSIONS option

some more info:

here is a patch ( I guess it will be part of next rc (rc4)

R00KIE commented on 2016-11-02 13:34

When trying to boot linux-git I'm getting lots of:

module_name: no symbol version for __put_user_4
module_name: Unknown symbol __get_user_4 (err -22)

I'm dropped to an emergency shell and the only way to regain control of the machine is to hard power off.

I've tried googling and searching the kernel bug tracker but I didn't find anything that seems to fit this problem so it might be something with my setup.

Any ideas of what could be wrong?

mrlamud commented on 2016-10-30 00:12

While creating packages " linux-git" & linux-git-headers" there are warnings:

==> WARNING: backup entry file not in package : etc/mkinitcpio.d/linux-git.preset

madman2003 commented on 2016-10-27 18:19

Beware that the current config has DEBUG_TEST_DRIVER_REMOVE enabled, which caused me a hang during loading of kernel initial ram image (forgot the exact name). It takes quite some time to bisect.

nauticalnexus commented on 2016-10-14 04:34


amdgpu-pro-dkms is broken. Don't use it for installs.

Anonymous comment on 2016-08-22 00:40

Installation fails. Is this a problem with the kernel or with amdgpu-pro?

==> Image generation successful
Optional dependencies for linux-git
crda: to set the correct wireless channels of your country
(2/3) installing linux-git-headers [######################] 100%
(3/3) installing linux-git-docs [######################] 100%
:: Running post-transaction hooks...
(1/1) Install DKMS modules
==> dkms install -m amdgpu-pro-16.30.3 -v 306809 -k 4.8.0-rc3-g4ec656b
grep: /boot/ No such file or directory
Error! Bad return status for module build on kernel: 4.8.0-rc3-g4ec656b (x86_64)
Consult /var/lib/dkms/amdgpu-pro-16.30.3/306809/build/make.log for more information.

here is /var/lib/dkms/amdgpu-pro-16.30.3/306809/build/make.log

parkerlreed commented on 2016-05-09 01:10

Adding to Krutonium

PKGBUILD config.x86_64 sum 0b92682bc82b37494563da40db4d79098406e21b3f47da6b26e228cd177dc3bb

Actual sum 8cec372fb3c8a53d5ea06e906baa893ae7853befd788d5b9ac25d0cf051776c2 config.x86_64

Krutonium commented on 2016-05-07 04:45

config.x86_64 is not passing it's SHA256sum.

Boohbah commented on 2016-05-05 21:30

misc: Thank you. I added objtool to the _pkg-headers function and CONFIG_STACK_VALIDATION=y in config.x86_64.

misc commented on 2016-05-01 11:51

Just a heads-up, if CONFIG_STACK_VALIDATION is set later, it'll require…
mkdir -p ${pkgdir}/usr/lib/modules/${_kernver}/build/tools/objtool
cp -a tools/objtool/objtool ${pkgdir}/usr/lib/modules/${_kernver}/build/tools/objtool/
… otherwise building of out-of-tree modules will just fail without explanation.

eworm commented on 2016-02-24 16:45

Can you please update config with new option CONFIG_DEBUG_WQ_FORCE_RR_CPU?

bwidawsk commented on 2015-12-28 21:38

This problem is present in the upstream PKGBUILD. It would be good to get it addressed there so that the other kernel packages don't have to all roll their own thing. I don't understand the reason for including the Kconfig files anyway, so I am planning to just remove that for loop (

But, even if you want to keep them, I think the correct thing is to actually copy the files just named "Kconfig"
diff --git a/PKGBUILD b/PKGBUILD
index 6d9c120..2a22226 100644
@@ -227,7 +227,7 @@ _package-headers() {
mkdir -p "${pkgdir}/usr/lib/modules/${_kernver}/build/mm"

# copy in Kconfig files
- for i in $(find . -name "Kconfig*"); do
+ for i in $(find . -name "Kconfig"); do
mkdir -p "${pkgdir}"/usr/lib/modules/${_kernver}/build/`echo ${i} | sed 's|/Kconfig.*||'`
cp ${i} "${pkgdir}/usr/lib/modules/${_kernver}/build/${i}"

My guess is the Kconfig are there for record keeping, anybody know?

Boohbah commented on 2015-12-14 02:45

jyc: Thanks for the fix. I added it to the PKGBUILD.

esrevinu commented on 2015-12-02 14:59

I had the same issue as jyc reported. Jyc's workaround is working.

KozRoss commented on 2015-11-21 08:11

This fails to build for me with the following error:

cp: Cannot stat 'arch//Makefile': No such file or directory.

What am I missing?

jyc commented on 2015-11-19 03:27

I had an error where a package conflict occurred because the following files were included in both the -doc and -header package:


This is due to line 226 copying any file with Kconfig in its name.
A workaround is just to add right after that the following line:

rm "${pkgdir}/usr/lib/modules/${_kernver}/build/Documentation/kbuild"/Kconfig.*-*

This removes the extraneous files that look like Kconfig files but are actually just documentation.

ackalker commented on 2015-11-01 13:29

Some kind requests and a question:
- Please update config{,.x86_64} files to latest in core/linux (this package: "Linux/x86 4.2.0-2 Kernel Configuration", core/linux: "Linux/x86 4.2.3-1 Kernel Configuration")
- Please add git to makedepends, to make `makepkg -s` happy.
- Why also use "replaces=" in package functions? Why not rely on just "conflicts="/"provides="?

Buddlespit commented on 2015-08-09 21:45

For those of us having nvidia build issues:

diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index 5243d4b..a0339f9 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -2619,7 +2619,7 @@ void flush_workqueue(struct workqueue_struct *wq)

* drain_workqueue - drain a workqueue

and edit PKGBUILD:

# Nvidia Patch
# the main kernel config files
'config' 'config.x86_64'
# standard config files for mkinitcpio ramdisk


pkgver() {
cd "${_srcname}"

git describe --long | sed -E sed -E 's/^v//;s/([^-]*-g)/r\1/;s/-/./g;s/\.rc/rc/'

prepare() {
cd "${_srcname}"

patch -p1 -i "${srcdir}/nvidia.patch"

if [ "${CARCH}" = "x86_64" ]; then
cat "${srcdir}/config.x86_64" > ./.config
cat "${srcdir}/config" > ./.config

yurikoles commented on 2015-06-17 15:39

Please use google mirror!

ZeroBit commented on 2015-05-23 06:42

Is it possible to have linux-git 4.1rc4 ?

Boohbah commented on 2015-04-14 22:17

hobarrera: The mirror works for me and others, but if you'd like then you can clone the repo from github and change the URL after using 'git remote set-url origin'

mrlouhibi: I don't get the same error. If you're using packer as AUR helper, see this

Tazmain: The full repo is 1.2 GB. The pacman developers have already said they won't implement shallow git clones in makepkg.

Tazmain commented on 2015-04-13 13:35

Seems like this is not doing a shallow copy of the git repo and a full clone making it around 2.6 gb download ?

mrlouhibi commented on 2015-03-28 06:58

I tried twice and I have this error:

==> Finished making: linux-git 4.0rc5.r104.ga39bdfb-1 (Sat Mar 28 02:54:18 EDT 2015)
[sudo] password for user:
loading packages...
error: 'linux-git-headers-4.0rc5.r104.ga39bdfb-1-x86_64.pkg.tar.xz': duplicate target
error: 'linux-git-docs-4.0rc5.r104.ga39bdfb-1-x86_64.pkg.tar.xz': duplicate target

WhyNotHugo commented on 2015-03-22 11:50

The mirror seems extremely slow (I can't get over 40KiB/s) (is it throttled). Can we switch to the github mirror?

Boohbah commented on 2015-03-03 20:52

I added https:// to the URL with this release.

If you get an error like ==> ERROR: linux-git/linux is not a clone of

then you can do 'cd linux; git remote set-url origin' to fix the URL of the bare clone repo.

silvio commented on 2014-12-11 19:18


please integrate this

thx, Silvio

Nefelim4ag commented on 2014-10-08 13:53

Can you move this code after make ???config
Its safety?
I use my own config file and load him from xconfig, but its everride version
# set localversion to git commit
sed -i "s|CONFIG_LOCALVERSION=.*|CONFIG_LOCALVERSION=\"-${pkgver##*.}\"|g" ./.config

# set extraversion to release candidate version, or blank
if [[ ${_rcver} == rc* ]]; then
sed -ri "s|^(EXTRAVERSION =).*|\1 -${_rcver}|" Makefile

Nefelim4ag commented on 2014-10-06 01:04

Please can you change default behavior of make config, to make olddefconfig?
This just a suggestion
Like in linux and linux-mainline packages. Because some people have own configured kernel and as i think echo yes | make config useless (because as i now, it a equivalent of make make allyesconfig)

Nefelim4ag commented on 2014-10-06 00:45

Please can you change default behavior of make config, to make olddefconfig?
This just a suggestion
Like in linux and linux-mainline packages. Because some people have own configured kernel and as i think echo yes | make config useless (because as i now, it a equivalent of make defconfig)

alpha1 commented on 2014-09-22 14:29


cp: cannot stat 'include/config/dvb/*.h'

alpha1 commented on 2014-08-20 18:51

It's just the location of the file that has changed, I think. xfs_sb.h is now at fs/xfs/libxfs/xfs_sb.h.

Also, I kinda came across a way to use the PKGBUILD to build tags, seems good to me, doesn't interfere (as far as I can see) with pkgver (it does indeed change the pkgver of the PKGBUILD to the right tag/commit), but the way I did may go against AUR guidelines or something. Anyways, I don't think it's something that everybody should use, but still, I thought I'd try and give something back.

So, this might be hacky, but for my current objective seems to do the trick:

pkgver() {
cd "${_srcname}"
git checkout -b build tags/v3.16-rc5
git describe --long | sed -E 's/^v//;s/([^-]*-g)/r\1/;s/-/./g'

Of course, change "v3.16-rc5" to what you want to build.

WonderWoofy commented on 2014-08-17 05:56

I get this when trying to build:

==> Starting package_linux-git-headers()...
cp: cannot stat ‘fs/xfs/xfs_sb.h’: No such file or directory

I just commented out the whole section in the PKGBUILD since I don't use aufs. But I am sure there when aufs3 is updated for 3.17, there will be a proper solution.

WonderWoofy commented on 2014-06-18 01:12

I don't think that these packages should provide 'linux', 'linux-headers', and 'linux-docs'. That is, the provides=(linux{,-{headers,docs}}="$pkgver").

The only packages that depend on 'linux' are 3rd party modules, which would definitely not work with this package anyway. But it also has the side effect of making it so that if one is trying to build a 3rd party module against the regular Arch kernel, 'makepkg -s' won't automatically install the needed 'linux'headers' package.

Boohbah commented on 2014-05-06 00:12

Also removed all patches from the PKGBUILD.

Boohbah commented on 2014-05-06 00:10

Fixed. I had changed $startdir to $srcdir, and that didn't work.

andreyv commented on 2014-05-05 12:18

I also get these .INSTALL errors. What's more funny, however, is that makepkg just ignores these errors and carries on...

WonderWoofy commented on 2014-05-05 04:34

I didn't really look into it in detail, but something about how you set the kernel version in the install script is wonky. I get this:

-> Adding install file...
cp: cannot stat ‘/tmp/cower/linux-git/linux-git.install.pkg’: No such file or directory
chmod: cannot access ‘.INSTALL’: No such file or directory
-> Generating .MTREE file...
bsdtar: .INSTALL: Cannot stat: No such file or directory
-> Compressing package...
bsdtar: .INSTALL: Cannot stat: No such file or directory

Boohbah commented on 2014-04-15 20:15

Removed 0002-module-allow-multiple-calls-to-MODULE_DEVICE_TABLE-p.patch and 0003-module-remove-MODULE_GENERIC_TABLE.patch

WonderWoofy commented on 2014-04-14 00:45

the 0002 and 0003 patches are no longer needed it seems. At least they don't apply anymore.

ArthurBorsboom commented on 2014-04-08 21:23

This time I have this compilation problem.

==> Starting prepare()...
patching file kernel/printk/printk.c
Hunk #1 succeeded at 59 with fuzz 2 (offset 1 line).
patching file include/linux/module.h
Hunk #1 FAILED at 142.
1 out of 1 hunk FAILED -- saving rejects to file include/linux/module.h.rej
patching file scripts/mod/file2alias.c
Reversed (or previously applied) patch detected! Assume -R? [n]
Apply anyway? [n]
Skipping patch.
5 out of 5 hunks ignored -- saving rejects to file scripts/mod/file2alias.c.rej
==> ERROR: A failure occurred in prepare().

Would you mind to fix this? Thank you in advance.

Boohbah commented on 2014-04-03 21:41

Looks like the bluetooth patch made it into the 3.15 merge window.
I've removed it from the PKGBUILD.

ArthurBorsboom commented on 2014-04-03 18:37

The patches introduce some questions, which I don't understand.
Is this a package error, or should I answer yes everywhere?

==> Extracting sources...
-> Creating working copy of linux git repo...
Cloning into 'linux'...
Checking out files: 100% (46364/46364), done.
==> Starting pkgver()...
==> Updated version: linux-git 3.14.r7333.g59ecc26-1
==> Starting prepare()...
patching file kernel/printk/printk.c
Hunk #1 succeeded at 59 with fuzz 2 (offset 1 line).
patching file Documentation/devices.txt
Reversed (or previously applied) patch detected! Assume -R? [n]
Apply anyway? [n]
Skipping patch.
1 out of 1 hunk ignored -- saving rejects to file Documentation/devices.txt.rej
patching file drivers/bluetooth/hci_vhci.c
Reversed (or previously applied) patch detected! Assume -R? [n]
Apply anyway? [n]
Skipping patch.
2 out of 2 hunks ignored -- saving rejects to file drivers/bluetooth/hci_vhci.c.rej
patching file include/linux/miscdevice.h
Reversed (or previously applied) patch detected! Assume -R? [n]
Apply anyway? [n]
Skipping patch.
1 out of 1 hunk ignored -- saving rejects to file include/linux/miscdevice.h.rej
==> ERROR: A failure occurred in prepare().
==> ERROR: Makepkg was unable to build linux-git.
==> Restart building linux-git ? [y/N]
==> ----------------------------------

misc commented on 2014-02-01 12:09

package_linux-git-headers currently fails since sn9c102 was moved from drivers/media/usb to drivers/staging/media in the recent v4l_for_linus pull.

Boohbah commented on 2014-01-07 09:36

Thanks Det, updated with your suggestion.

Det commented on 2014-01-05 03:30

Instead of:

true && pkgname=("${pkgbase}" "${pkgbase}-headers" "${pkgbase}-docs")

You could also say:

pkgname+=("${pkgbase}-headers" "${pkgbase}-docs")

xduugu commented on 2014-01-01 17:44

Disowned linux-git and linux-firmware-git.

kolkstel commented on 2013-12-21 00:45

@xduugu I would like to maintain linux-git if you don't want to anymore.

xduugu commented on 2013-11-25 15:24

As you have noticed the last months, I do not have the time to properly maintain this package anymore. For this reason, I would like to hand it - and ideally also linux-firmware-git - over to someone else.

There is one issue with the current PKGBUILD that I am aware of and that I would like have fixed before the handover, but there was no time to find a workaround. So, unfortunately, building a specific git revision by setting the _gitrev variable does not work in the pkgver function and therefore, the package gets a wrong version number. Apart from that, the PKGBUILD should be fine.

xduugu commented on 2013-11-25 15:11

Updated to latest config files.

One of the kernel config scripts needs it.

@ flokli
According to the bug report, there are security issues with this config option. So I left it disabled by default.

gun26 commented on 2013-11-15 16:44

This probably needs updating to bring the config files into line with the current ARCH kernel. One specific thing I've noticed is that sysfs-efivars is now deprecated and efivarfs should be used instead.

baronmog commented on 2013-11-05 04:52

Why does this depend on extra/bc?

Ajaay1 commented on 2013-11-04 11:05

I'm using a modified version of this PKGBUILD to download a custom kernel from git. When patches are released for this build is there a way to patch and compile the impacted module only? Or do I have to recompile the whole kernel again through makepkg?

Anonymous comment on 2013-09-27 21:31

Nevermind, I think I've figured it out. Run `makepkg -o` to grab the sources, cd into $srcdir/linux-git and start bisecting, then build with `makepkg -e` to use the bisected checkout.

Anonymous comment on 2013-09-27 08:03

Does anyone have any hints on how to use this package with git-bisect to generate and test kernel packages?

flokli commented on 2013-09-21 11:18

Can you please enable CONFIG_USER_NS as planned for 3.12 in ?

The needed commit d6970d4b726cea6d7a9bc4120814f95c09571fc3 [enable building user namespace with xfs] was merged by Linus in 300893b08f3bc7057a7a5f84074090ba66c8b5ca [Merge tag 'xfs-for-linus-v3.12-rc1' of git://]

BlackLotus89 commented on 2013-07-17 12:07

Could you please add CONFIG_SND_HDA_I915=y to the default .config ?

gun26 commented on 2013-04-19 11:31

I noticed that the version I just built, linux-git-3.9.rc7.74.ga86d526-1-x86_64.pkg.tar.xz, reports itself via uname -r as:
Shouldn't the rc7 be there? This looks more what we would see during the merge window after 3.9 is released, before the merge window closes and we get a 3.10-rc1. I don't know if this is something we can control from the PKGBUILD, or maybe the kernel config.

xduugu commented on 2013-04-16 10:08

Yep, since the latest pacman release, makepkg no longer sets pkgver to the current date for vcs packages by default and I do not see an advantage in the old versioning format.

The initial clone and the updating of the repository is now handled by makepkg, but maybe the name of the folder can be changed by prefixing the uri with "linux.git::".

gun26 commented on 2013-04-15 20:56

Works well here. I notice _kernel_pkgver is gone as an option but it still puts git revision in the package number - I guess it's always on now?

The git pull targets a linux subfolder instead of linux.git as before. Was this intended? It started to clone a fresh tree before I stopped it and did cp -Ra from linux.git to linux before restarting.

xduugu commented on 2013-04-15 12:50

It should work a lot better now with pacman 4.1.

alesko commented on 2013-04-14 20:13

gun26: I have the exact same issues as you. You're not alone :) I'll try to build it now and hope that the issues are gone for me too.

gun26 commented on 2013-04-14 02:06

And now it seems to be working as intended but I can't explain why it wasn't before. Ignore my previous comment.

gun26 commented on 2013-04-09 00:26

Now I'm getting this:
==> Saving /home/gun/linux-git/linux.git-build to /home/gun/linux-git/src/linux.git-build...
mv: cannot move ‘/home/gun/linux-git/linux.git-build’ to ‘/home/gun/linux-git/src/linux.git-build’: Directory not empty
==> ERROR: A failure occurred in build().

xduugu commented on 2013-04-04 17:21

It should build now, but it still needs some work to adjust it for pacman 4.1.

gun26 commented on 2013-04-03 00:23

Since the pacman 4.1 update I can't build this package. It fails like this:
==> Creating build branch...
fatal: could not create work tree dir '/linux.git-build'.: Permission denied
It doesn't seem to matter if I uncomment the BUILDDIR variable in /etc/makepkg.conf. Other packages build ok.

xduugu commented on 2013-03-28 19:09

Thanks, I added it as make dependency.

Hazzl commented on 2013-03-28 06:41

I can confirm that bc is required to build current git kernels (for some frequency calculations)

xduugu commented on 2013-03-20 23:47

Thanks, but are you sure that it is required by the kernel and not by some aur wrapper? I will try to confirm this, but currently I do not have a chroot at hand.

hi117 commented on 2013-03-19 19:04

When building linux-git, I got the error /bin/sh bc: Command not found.
It appears linux-git requires bc and seeing how it is not in base or base-devel, it should be added to the makedepends.

Boohbah commented on 2013-02-05 10:31

Use git to apply patches and commit the changes to avoid dirtying the git tree:
Note: this won't work if you set _gitrev to build a particular {tag,commit,branch,}

Boohbah commented on 2013-02-05 10:05

Use git to apply patches and commit the changes to avoid dirtying the git tree:

Boohbah commented on 2013-02-05 09:58

Use git to apply patches and commit the changes to avoid dirtying the git tree:

Boohbah commented on 2013-02-05 09:29

Use git to apply patches and commit the changes to avoid dirtying the git tree:

gun26 commented on 2012-09-08 01:48

Good news, I got this to build again by changing line 434 of the PKGBUILD to depmod -b "$pkgdir/usr" -F "$_kernver" (adding the /usr). Hard to believe it's that simple, but seems to work. I won't kid anyone I really know what I'm doing though. :)

gun26 commented on 2012-09-07 17:41

Looks like the new kmod 10 in testing breaks this package. The Arch kernel PKGBUILD is fixed by moving the module tree from /lib to /usr/lib after depmod instead of before, but I'm not sure how to apply the same sort of fix to this PKGBUILD.

Roken commented on 2012-08-05 06:26

Oh - ignore me. Found a stray space in the patch. Fixed and working now.

Roken commented on 2012-08-03 19:33

Thank you. Switching to makepkg fixed that (I still have things to learn re ABS, obviously).

Anyway, new issue. Trying to add the following patch:

[code]diff --git a/fs/fuse/dir.c b/fs/fuse/dir.c
index 8964cf3..7c6c808 100644
--- a/fs/fuse/dir.c
+++ b/fs/fuse/dir.c
@@ -405,7 +405,7 @@ static int fuse_create_open(struct inode *dir, struct dentry *entry,
memset(&inarg, 0, sizeof(inarg));
memset(&outentry, 0, sizeof(outentry));
inarg.flags = flags;
- inarg.mode = mode;
+ inarg.mode = mode | S_IFREG; /* Userspace expects S_IFREG */
inarg.umask = current_umask();
req->in.h.opcode = FUSE_CREATE;
req->in.h.nodeid = get_node_id(dir);[/code]

fails if I add it the patches directory. As I work around I use xconfig and manually change the file before saving off the config. Any idea why it fails from the PKGBUILD though (I don't have the error now, sorry).

The patch is a testing patch to fix the ntfs-3g/fuse issue that I've mentioned around the forums. I imagine it will be committed to git soon enough.

xduugu commented on 2012-07-24 07:46

I think I prefer to support only one upstream source. A linux-stable-git package is imo a better option, because you do not get error reports from two different upstream sources in the comments of only one package.

The source is stored in the folder where you extracted the source tarball. When you extract it to /tmp, it is downloaded there. Configure your aur wrapper properly or use the plain makepkg program.

Roken commented on 2012-07-23 23:06

Any chance you could use a directory other than /tmp for the build, to prevent having to re-download the entire git tree. I keep forgetting to change it and wham, gets deleted when I reboot.

eworm commented on 2012-07-22 18:16

Just an idea... You could add a configuration option to use Greg's stable tree instead of Linus' mainline.

eworm commented on 2012-07-22 10:33

Looks good this time. Thanks a lot!

eworm commented on 2012-07-22 09:52

Damn, my fault. I have an sed script that changes some settings. I put some depmod things in there for testing purposes without reverting them. Rebuilding now...

xduugu commented on 2012-07-22 09:38

It looks good here. Your error is exactly the one that occurs when the change in the script is not made. Please check if it actually was modified.

eworm commented on 2012-07-22 09:36

It does not build now:

ERROR: could not open directory /tmp/.repo-make-21493/pkg/linux-git/usr/usr/lib/modules/ No such file or directory
FATAL: could not search modules: No such file or directory
make: *** [_modinst_post] Error 1

xduugu commented on 2012-07-22 09:07

You are right, the lsmod fix is not required anymore. The depmod issue should be fixed, but I had no chance to build the package yet to verify.

eworm commented on 2012-07-22 08:52

I think the lsmod thing actually does not do anything, does it? For me it changes nothing, the pattern simple does not match.

The only change to the source tree before building is in scripts/ Would be great to get a fix for that. Thanks a lot!

xduugu commented on 2012-07-22 08:42

That's not just because I alter (this change only exists since the /lib move three weeks ago), but may be because of the lsmod path that needs to be corrected. I will see how to fix it.

eworm commented on 2012-07-21 22:28

Your package always produces kernels whose version ends with '-dirty'. This is because you alter scripts/ Any chance we change that to get 'clean' version numbers?

xduugu commented on 2012-07-04 20:11

Updated to install the modules to /usr/lib. I guess it currently depends on the packages (kmod in particular) from testing.

gun26 commented on 2012-07-04 04:58

I guess this package needs to be revised for the migration out of /lib (i.e. /lib/modules is now going to be /usr/lib/modules, etc.) Here's the relevant post in arch-dev-public:

alesko commented on 2012-04-16 14:38

Yes I use cairo from testing but since it flickers even in framebuffer console (without X server), I'm pretty sure it's not cairo related.
The only component which remains to blame is makepkg. I will try it tonight again as I really want to use this package :)

xduugu commented on 2012-04-15 20:28

Yeah, that's strange. And you cannot reproduce it with other kernel packages? Maybe it's related to the makepkg environment. Or you could try nconfig instead of menuconfig. Do you happen to use testing so that it may be cairo-related?

alesko commented on 2012-04-14 23:40

The strange thing is I only have this flickering with this package. I tried it in terminator, gnome-terminal and the "real" console (without X) - doesn't work. It's like something in the background is working.
When cloning the kernel by myself and call "make menuconfig" manually, there is no flickering.
I will try it with xconfig as you suggested.

xduugu commented on 2012-04-14 20:18

Works fine here. I guess you experience the same flickering in any other ncurses-based program? If so, the problem lies somewhere else and should be fixed there. As a workaround, you could try to use one of the other graphical config menus.

alesko commented on 2012-04-14 11:05

I have strange flickering when menuconfig is presented. I can't do anything in the menu, only kill it.

Anonymous comment on 2011-11-26 18:28

I also suggest changing "/sbin/depmod $kernver" to "depmod $kernver"

Anonymous comment on 2011-11-26 18:21

Hi. In the linux-git.install you need to change "/sbin/mkinitcpio -p $pkgname" to "mkinitcpio -p $pkgname". In mkinitcpio 0.8.0-1, which is currently in [testing], the binary path is changed to /usr/bin [1]. So the kernel image fails to generate for users of [testing] and eventually [core] when mkinitcpio 0.8.0-1 gets moved.


xduugu commented on 2011-11-05 18:57

Seems I did not test this change carefully enough. The intention was to add the localmodversion to the directory name and not another part of the kernel version. The latest package should be fine now.

gun26 commented on 2011-11-05 00:51

it's possible to make the extramodules direcstory more useful so that out-of-tree modules don't have to be rebuild and reinstalled each time linux-git changes, if we name it something generic like extramodules-3.1-git instead of the full git version number. I'm trying this variation as an experiment by changing line 367 accordingly in this PKGBUILD. I can now install a newer version of linux-git without having to rebuild the nvidia module. Not a huge saving but it works.

xduugu commented on 2011-07-12 21:55

Thanks for the reminder. I've noticed that too, but had no time to spend on this at that moment.
Basically, depmod has to be run before mkinitcpio. Not sure why it worked before, but it will be fixed in the new packages I'll upload shortly.

gun26 commented on 2011-07-12 17:23

I've got a mkinitcpio weirdness when installing this kernel. Here's what I see: The resulting linux-git.img indeed has no modules and won't boot. BUT if I scroll up and copy the mkinitcpio line from the terminal and invoke mkinitcpio manually, namely: sudo mkinitcpio -k 3.0.0-rc7-dirty -c /etc/mkinitcpio.conf -g /boot/linux-git.img, the resulting image is fine. So this is what I've been doing each time I install a new git kernel. This happens on both my desktop (Intel quad core) and laptop (Thinkpad with Core 2 duo). The same thing happens with the kernel26-git package, btw, although I've been successfully (other than this mkinitcpio thing) using this package for several weeks.

skogler commented on 2011-06-18 10:21

is there a difference to kernel26-git at the moment?