You are not logged in.

#1 2016-11-28 20:50:48

mcloaked
Member
From: Yorkshire, UK
Registered: 2012-02-02
Posts: 1,222

[SOLVED] Early boot problems with kernel 4.9-rcX

An arch linux user that I communicate regularly with, has found that he cannot boot any of the 4.9-rcX kernels built in ABS on his machine ( without CONFIG_MODVERSIONS=y), and I have been trying to help diagnose the issue - the problem is that the boot attempt leads to a single kernel line on the screen, and then a boot fail with no diagnostics, and nothing at all in the logs that leads to any hint about what the problem might be. All 4.8 kernels boot without any issues.

I have only seen very few bug reports that seem even vaguely related using google searches - but this evening I found:

https://bugs.debian.org/cgi-bin/bugrepo … bug=845690

which is for debian and not arch but don't know if this issue might be a problem in arch linux also?

Does anyone on this forum have any experience that might enlighten this problem - does everyone building 4.9 for arch linux lead to kernels that boot without problems, or has anyone seen a very early boot failure with 4.9-rcX (mainline) kernels?

Thanks.

[Edit: the system involved has a luks encrypted disk but I don't know if that has any bearing on this problem, but I do note that there is a comment in the mainline kernel in AUR from October 16 that "After the upgrade to 4.9rc1-1, boot could not find the dev mapper for my root device (systemd-boot + luks + lvm) so I had an unbootable machine (didn't get any errors/warnings during mkinit or install).

I've downgraded back to linux-mainline 4.8-1 but word of warning and/or any advice/input is welcome"]

Last edited by mcloaked (2017-01-27 07:47:25)


Mike C

Offline

#2 2016-11-28 21:07:44

loqs
Member
Registered: 2014-03-06
Posts: 17,196

Re: [SOLVED] Early boot problems with kernel 4.9-rcX

Is git bisect not applicable?

Offline

#3 2016-11-28 21:10:07

mcloaked
Member
From: Yorkshire, UK
Registered: 2012-02-02
Posts: 1,222

Re: [SOLVED] Early boot problems with kernel 4.9-rcX

loqs wrote:

Is git bisect not applicable?

Quite possibly though the arch user involved has little time to run a lot of builds needed due to large working hours - but I was interested since it won't be long before 4.9 gets released by Linus with possibly one more rc8 or maybe not even one additional rc - and if this issue is only manifest for a particular case (such as luks encryption) then it might see a lot of instances once released unless there are other users who have reported the problem to the kernel bug list. I don't have encrypted disks which is why I tried to search for any reports.

The nub of the problem seems to be related to a bug in gcc version 7 but the link I reported above relates also to some gcc version 6.x that seems to hit the 4.9-rcX kernels but not earlier ones,  i.e.

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=78512

Last edited by mcloaked (2016-11-28 21:19:18)


Mike C

Offline

#4 2016-11-28 22:05:34

loqs
Member
Registered: 2014-03-06
Posts: 17,196

Re: [SOLVED] Early boot problems with kernel 4.9-rcX

I suspect the subset of the arch base that use the forums + lvm + luks + prerelease gcc + prerelease kernel is not going to be large.
Would test with gcc 6 vs gcc 7 to see what difference if any that makes.
Also do you think Linus would accept a bug that relies upon a prerelease version of gcc to trigger it as you said it is late in the release cycle another option would be he simply marks gcc7 as unsupported similar to how he marked CONFIG_MODVERSIONS as broken.

Offline

#5 2016-11-28 22:12:00

mcloaked
Member
From: Yorkshire, UK
Registered: 2012-02-02
Posts: 1,222

Re: [SOLVED] Early boot problems with kernel 4.9-rcX

loqs wrote:

I suspect the subset of the arch base that use the forums + lvm + luks + prerelease gcc + prerelease kernel is not going to be large.
Would test with gcc 6 vs gcc 7 to see what difference if any that makes.
Also do you think Linus would accept a bug that relies upon a prerelease version of gcc to trigger it as you said it is late in the release cycle another option would be he simply marks gcc7 as unsupported similar to how he marked CONFIG_MODVERSIONS as broken.

Indeed - though it would appear that for some people kernel 4.9 builds using gcc versions 6.X which might include the current version in arch could lead to an unbootable kernel if the reports linked to above are indicative of where the problem might be initiated.  It isn't clear that the arch current version of  gcc 6.2.1-1 is a problem, but in the Debian report at https://bugs.debian.org/cgi-bin/bugrepo … bug=845690 there is a statement "6.2.1-4 and 6.2.1-5 create nonbootable kernels on all x86-64 machines tested so far (P4 era 64-bit Xeons, 51xx era Xeons, Opteron 2xx, i5 660 at least). Kernel configurations vary ..." so whether the problem can also be triggered by gcc 6.2.1-1 isn't clear, and would need further exploration? i.e. gcc versions earlier than 7.x as per the gcc bug report may still produce buggy binaries for the 4.9 series kernels?

Last edited by mcloaked (2016-11-28 22:14:09)


Mike C

Offline

#6 2016-11-28 22:27:19

loqs
Member
Registered: 2014-03-06
Posts: 17,196

Re: [SOLVED] Early boot problems with kernel 4.9-rcX

Having some more data points may help narrow down the cause of this particular failure.
Locally building the linux package from ABS with the development tools from the official packages.
Building 4.9 rc with that tool chain.
Locally building the linux package from ABS with the prerelease gcc packages.

Offline

#7 2016-11-29 09:23:23

mcloaked
Member
From: Yorkshire, UK
Registered: 2012-02-02
Posts: 1,222

Re: [SOLVED] Early boot problems with kernel 4.9-rcX

loqs wrote:

Having some more data points may help narrow down the cause of this particular failure.
Locally building the linux package from ABS with the development tools from the official packages.
Building 4.9 rc with that tool chain.
Locally building the linux package from ABS with the prerelease gcc packages.

The arch user who is impacted by this will try those ideas when he gets back to his machine after working away this week. I will post any progress when the tests have been run.


Mike C

Offline

#8 2016-12-01 09:47:46

Dokter Bibber
Member
Registered: 2016-09-19
Posts: 65

Re: [SOLVED] Early boot problems with kernel 4.9-rcX

If the Arch Linux user is using LVM, s/he is better off staying far away from the 4.9.0-RC kernels : https://bugs.archlinux.org/task/51818
Check just above the middle in the 4th screenshot (under EDIT3:) in this post to see what happens in that bug report : https://forum.manjaro.org/t/stable-upda … x/12868/24
Upon rebooting, your system is dead. See 5th screenshot in that post at the green forum.

Last edited by Dokter Bibber (2016-12-01 09:52:03)

Offline

#9 2016-12-01 10:15:21

mcloaked
Member
From: Yorkshire, UK
Registered: 2012-02-02
Posts: 1,222

Re: [SOLVED] Early boot problems with kernel 4.9-rcX

Dokter Bibber: Thank you for posting, But I believe that LVM is not being utilised in his system so hopefully that is not the issue - but it needs tests to see if it is a gcc issue producing code for the kernel that fails to boot.

Last edited by mcloaked (2016-12-01 10:21:58)


Mike C

Offline

#10 2016-12-01 12:36:29

Dokter Bibber
Member
Registered: 2016-09-19
Posts: 65

Re: [SOLVED] Early boot problems with kernel 4.9-rcX

mcloaked wrote:

....., boot could not find the dev mapper for my root device (systemd-boot + luks + lvm) .....

You do mention lvm though.
That's why I assumed that it was involved. wink

Offline

#11 2016-12-01 15:27:42

mcloaked
Member
From: Yorkshire, UK
Registered: 2012-02-02
Posts: 1,222

Re: [SOLVED] Early boot problems with kernel 4.9-rcX

Dokter Bibber wrote:
mcloaked wrote:

....., boot could not find the dev mapper for my root device (systemd-boot + luks + lvm) .....

You do mention lvm though.
That's why I assumed that it was involved. wink

Yes I did quote from a comment in the AUR mainline kernel package from a different user so apologies if that led to confusion.


Mike C

Offline

#12 2017-01-03 16:22:40

mcloaked
Member
From: Yorkshire, UK
Registered: 2012-02-02
Posts: 1,222

Re: [SOLVED] Early boot problems with kernel 4.9-rcX


Mike C

Offline

#13 2017-01-27 07:47:05

mcloaked
Member
From: Yorkshire, UK
Registered: 2012-02-02
Posts: 1,222

Re: [SOLVED] Early boot problems with kernel 4.9-rcX

Resolved in kernel 4.9.6-1-ARCH


Mike C

Offline

Board footer

Powered by FluxBB