Re: [vlan_device_event] BUG: unable to handle kernel paging request at 6b6b6ccf

[Fengguang Wu]

On Thu, Nov 09, 2017 at 10:43:08AM +0800, Fengguang Wu wrote:
> > > Of course, if it's bisectable, that would be great too.
> >
> > Yes, bisect is on the way. So far it's bisecting in the 4.12 commits.
>
> The bisect was unsuccessful due to an unrelated DRM_BOCHS oops in 4.11.
> Disabling the buggy driver, I managed to reproduce the
> vlan_device_event bug in 4.11. However only to find the older kernels
> suffer from different kind of oops, which make the bisect troublesome.
>
> % grep -E 'dmesg.(BUG|EIP)' v4.*/matrix.json
> v4.10/matrix.json:  "dmesg.BUG:unable_to_handle_kernel": [
> v4.10/matrix.json:  "dmesg.EIP:kobject_get": [
> v4.10/matrix.json:  "dmesg.BUG:kernel_reboot-without-warning_in_test_stage": [
> v4.11/matrix.json:  "dmesg.BUG:unable_to_handle_kernel": [
> v4.11/matrix.json:  "dmesg.EIP:vlan_device_event": [
> v4.8/matrix.json:  "dmesg.BUG:kernel_reboot-without-warning_in_early-boot_stage,last_printk:Decompressing_Linux": [
> v4.9/matrix.json:  "dmesg.BUG:key_not_in.data": [
>
> I'll try tuning kconfig to get a good bisect base.
>
> Hopefully we'll make the coming RC releases free from such bisect
> stoppers with better regression tracking and bisect automation.

However if the situation turns out to be not that optimistic, we might
track "bisect stopper" kconfig options over kernel releases. For
bisect scripts to auto pick them up.

An example kconfig bisect-blacklist file would be:

       $ cat tools/testing/configs/bisect-4.12.config
       # CONFIG_DRM_BOCHS is not set

That would mean:

       CONFIG_DRM_BOCHS will reliably stop kernel from booting before 4.12.
       If ever bisecting into pre-4.12 kernels, better disable it.

The best fit would be the uncommon drivers and optional features.

Warnings are typically not bisect stoppers so can be ignored. Kernel
panics that happen rare enough to impact bisect may also be ignored.

Thanks,
Fengguang