Double fault in cpu_entry_area while resuming from suspend

[Mateusz Jończyk]

Hello,

On kernel 5.15.0-60-generic from Ubuntu, I had an occurrence of a double fault while resuming from suspend. I'm attaching logs from pstore.

I run mostly upstream kernels. The same (or a similar) problem has happened to me a few times on rc releases of Linux 6.1, the first time in November 2022 on 6.1.0-rc5. It stopped happening since I have moved to rc releases of Linux 6.2. So I think that a similar bug was introduced in Linux 6.1 and fixed in Linux 6.2 and probably carried over with stable patches to Ubuntu's 5.15.0-60-generic.

I was not able to reproduce this bug reliably on Linux 6.1 back then, and I did not report the problem to LKML (I'm sorry).

In all occurrences (on Linux 5.15 and 6.1), the fault has happened at address 0xfffffe000000b000, which according to https://www.kernel.org/doc/Documentation/x86/x86_64/mm.txt lies in the cpu_entry_area mapping, I think that it is in a guard page there. Every time, this has happened while resuming from suspend. Every time, there were some RCU functions present in the pstore stack traces.

I have some patches in mainline Linux and I'm happy to assist in any troubleshooting.

ProblemType: Bug
DistroRelease: Ubuntu 20.04
Package: linux-image-5.15.0-60-generic 5.15.0-60.66~20.04.1
ProcVersionSignature: Ubuntu 5.15.0-60.66~20.04.1-generic 5.15.78
Uname: Linux 5.15.0-60-generic x86_64
ApportVersion: 2.20.11-0ubuntu27.25
Architecture: amd64
CasperMD5CheckResult: skip
CurrentDesktop: GNOME-Classic:GNOME
Date: Wed Feb 22 21:06:51 2023
InstallationDate: Installed on 2020-09-12 (892 days ago)
InstallationMedia: Ubuntu 20.04 LTS "Focal Fossa" - Release amd64 (20200423)
ProcEnviron:
 TERM=xterm-256color
 PATH=(custom, no user)
 XDG_RUNTIME_DIR=<set>
 LANG=pl_PL.UTF-8
 SHELL=/bin/bash
SourcePackage: linux-signed-hwe-5.15
UpgradeStatus: No upgrade log present (probably fresh install)

[Mateusz Jończyk (mat-jonczyk)]

Both pstore logs are from the same crash.
They were obtained by concatenating the parts from systemd-pstore:

cat $(ls /var/lib/systemd/pstore/167684323/dmesg-efi*001 -1 | tac) > pstore1_on_Linux_5.15.0-60.txt
cat $(ls /var/lib/systemd/pstore/167684323/dmesg-efi*002 -1 | tac) > pstore2_on_Linux_5.15.0-60.txt

[Mateusz Jończyk (mat-jonczyk)]

dmesg from before crash, obtained from "journalctl -k"

[Mateusz Jończyk (mat-jonczyk)]

Pstore from Linux 6.1.0 with this Bluetooth patch: https://lore<email address hidden>/ , which is probably unrelated.

"unif9" is my kernel "local version" (CONFIG_LOCALVERSION) suffix that denotes the Kconfig I was using.

[Mateusz Jończyk (mat-jonczyk)]

Kernel 6.1.0-rc7 patched with a different patch: https://<email address hidden>/T/