Bug 198529

Summary: Reboot on kernel load due to 92a0f81d
Product: Platform Specific/Hardware Reporter: Alkis Georgopoulos (alkisg)
Component: x86-64Assignee: platform_x86_64 (platform_x86_64)
Status: NEW ---    
Severity: high CC: chenhan.hsiao.tw
Priority: P1    
Hardware: i386   
OS: Linux   
Kernel Version: 4.15rc5+ Subsystem:
Regression: Yes Bisected commit-id:

Description Alkis Georgopoulos 2018-01-19 20:19:06 UTC
Hi, it's the first time I did a kernel bisection. If I did it correctly, commit 92a0f81d89571e3e8759366e050ee05cc545ef99 (x86/cpu_entry_area: Move it out of the fixmap, on Wed Dec 20), is causing the following issue on certain hardware:
> Loading Linux 4.15 ...
> Loading initial ramdisk ...
> <immediate reboot>

 * I can reproduce this issue on an i5-4440 with any 32bit kernel after that commit, and up to yesterday's linux-stable.
 * I cannot reproduce it on 64bit kernels.
 * I cannot reproduce it on an i3-4150 on the same motherboard Gigabyte B85M-HD3, so maybe it's related to the number of cores?
 * Also if I put acpi=off in the kernel command line, it works around the issue.

About 10 schools have already reported the issue, so I believe it's rather common...

I'm able to test kernels or patches. Thank you!
Comment 1 Alkis Georgopoulos 2018-01-23 06:49:47 UTC
The problem still happens with 4.15rc9, i.e. 32 bit kernels immediately reboot on at least i5-4440 and i5-4460.
Comment 2 Alkis Georgopoulos 2018-01-25 05:58:03 UTC
I tested a few more CPUs, 32 bit kernels no longer work on:
i5-3470, i5-4440, i5-4460, i5-6200U

...and yet they work properly on:
i3-4170, i3-7100, i5-4590, Q6600, Q8300

The "acpi=off" parameter sometimes helped, sometimes not.
Comment 3 Alkis Georgopoulos 2018-02-01 10:47:41 UTC
The issue has also been reported in launchpad:

Wgrant posted a fix there which works for me.
He also sent it upstream in the kernel:

Thanks a lot William!