[solved] Broke my kernel trying to switch to systemd

[Aquous]

Hi folks,

Up until a few minutes ago, I ran a 3.10.9 kernel that was working just fine. I recently decided I wanted to switch from OpenRC to systemd, and in doing so I was asked to change some kernel configuration options. IIRC what I changed in my config file was:

• disable CONFIG_GENTOO_LINUX_INIT_SCRIPT
  
• enable CONFIG_GENTOO_LINUX_INIT_SYSTEMD
  
• enable CONFIG_FHANDLE
  
• enable CONFIG_FANOTIFY
  
• disable CONFIG_{EXT4,REISERFS}_FS_SECURITY because I wasn't using security labels and couldn't for the life of me remember why I enabled this option

And now my kernel doesn't boot any more: http://img856.imageshack.us/img856/411/e1g6.jpg

This is my new .config from the broken kernel: http://pastebin.com/i9qmindD
This is my working .config from 3.10.7. My previously working .config for 3.10.9 was generated from this config by using make oldconfig: http://pastebin.com/J0h2ypnv
(note: changing DRM and RADEON from modules to built-in is a red herring, I did this to be able to capture the stack trace from a full-resolution screen)

Can anyone help me fix my kernel?

[Aquous]

"Fixed" by disabling IPV6 and BFQ. Since the error message mentions BFQ, it seems likely that this was the culprit, however since I had been on BFQ all along and the problem didn't begin until I set CONFIG_GENTOO_LINUX_INIT_SYSTEMD=y which changed CONFIG_IPV6 from 'n' to 'y', I'd be tempted to say this was a really, really weird conflict between BFQ and IPV6. Any thoughts?

[cauchy49]

I have the exact same problem. No idea what to do except use cfq instead of bfq

[Aquous]

I've done some more testing around and it's not related to CONFIG_IPV6. It's probably CONFIG_FANOTIFY or CONFIG_FHANDLE. Unfortunately, I can't test this hypothesis any more because I'm now using systemd which requires both these options in order to function

[ulenrich]

https://bugs.gentoo.org/show_bug.cgi?id=481820#c6
That explains status of Bfq.

Even if you try to disable it
the Bfq patch invasively changes some return types
EINVAL to 0
Better to use the pure upstream vanilla-sources?

[TomWij]

[TomWij]

[Aquous]

Thank you both, re-emerging gentoo-sources fixed the issue for me

[ulenrich]