Hi Gianfranco,

Great! Let's work this out together.

I have strong doubts that:

commit ef40d28f17bd384d7e0b630c7d83f108a526351b
Author: Kees Cook <k...@kernel.org>
Date:   Wed Jun 19 14:47:15 2024 -0700
Subject: randomize_kstack: Remove non-functional per-arch entropy filtering
Link: 
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=ef40d28f17bd384d7e0b630c7d83f108a526351b

will fix this issue. If you read the commit log and the code, it clearly
removes all per-arch entropy values, and just sticks with a universal
1KiB of space for all architectures.

The root cause of the issue is that "randomize_kstack: Improve entropy
diffusion" changed the amd64 kernel stack consumption for randomisation,
from 0.25KiB to a full 1KiB of space. The per thread kernel stacks are
only 16KiB in size, so we went from VirtualBox having 15.75KiB of stack
space down to 15KiB. VirtualBox must have been really pushing the limit
and needing that extra 0.75KiB of space, since without it, we panic.

They probably made some architectural changes in 7.0.x that reduce the
total kernel thread stack consumption, and now fall under the 15KiB
limit that "randomize_kstack: Improve entropy diffusion" imposes.

Anyway, I still made you a test kernel. It is based on
5.15.0-117-generic + "randomize_kstack: Remove non-functional per-arch
entropy filtering".

This test kernel is for Gianfranco Costamagna ONLY! Other users, please,
try my other test kernel in the above comment instead.

Gianfranco, the kernel will be ready in 3 hours from this message. They
are still building.

Please note this package is NOT SUPPORTED by Canonical, and is for TESTING
PURPOSES ONLY. ONLY Install in a dedicated test environment.

Instructions to Install (On a focal or jammy system):
1) sudo add-apt-repository ppa:mruffell/lp2073267-test-2
2) sudo apt update
3) sudo apt install linux-image-unsigned-5.15.0-117-generic 
linux-modules-5.15.0-117-generic linux-modules-extra-5.15.0-117-generic 
linux-headers-5.15.0-117-generic
4) sudo reboot
5) uname -rv
Look for +TEST2073267v20240731b2 

Anyway, I think we either need to figure out how to get the virtualbox
kernel module stack consumption down, or we revert "randomize_kstack:
Improve entropy diffusion" for focal, focal HWE, jammy, jammy HWE (but
not noble).

 virtualbox | 6.1.6-dfsg-1                        | focal/multiverse           
| source, amd64
 virtualbox | 6.1.32-dfsg-1build1                 | jammy/multiverse           
| source, amd64
 virtualbox | 6.1.50-dfsg-1~ubuntu1.20.04.1       | focal-security/multiverse  
| source, amd64
 virtualbox | 6.1.50-dfsg-1~ubuntu1.20.04.1       | focal-updates/multiverse   
| source, amd64
 virtualbox | 6.1.50-dfsg-1~ubuntu1.22.04.1       | jammy-updates/multiverse   
| source, amd64
 virtualbox | 6.1.50-dfsg-1~ubuntu1.22.04.2       | jammy-proposed/multiverse  
| source, amd64
 virtualbox | 7.0.16-dfsg-2                       | noble/multiverse           
| source, amd64
 virtualbox | 7.0.16-dfsg-2ubuntu1                | noble-updates/multiverse   
| source, amd64
 virtualbox | 7.0.20-dfsg-1                       | oracular/multiverse        
| source, amd64
 
Thanks,
Matthew

-- 
You received this bug notification because you are a member of Kernel
Packages, which is subscribed to linux-signed-hwe-5.15 in Ubuntu.
https://bugs.launchpad.net/bugs/2073267

Title:
  Virtualbox Guru meditation on VM start caused by kernel commit in
  v6.9-rc4

Status in linux package in Ubuntu:
  Triaged
Status in linux-signed-hwe-5.15 package in Ubuntu:
  Confirmed
Status in virtualbox package in Ubuntu:
  Confirmed

Bug description:
  It worked yesterday, but today I get a Guru Meditation trying to start
  some of my virtual machines. This shows up in VBox.log as "VCPU0: Guru
  Meditation -2708 (VERR_VMM_SET_JMP_ABORTED_RESUME)". I suspect this
  may have started due to a Linux kernel upgrade I installed this
  morning.

  A fresh VM with no disk shows the issue. Sometimes turning off the I/O
  APIC makes the issue go away, sometimes not. Turning off nested paging
  sometimes lets VirtualBox make a little bit of progress w.r.t. booting
  VMs, but that usually still crashes before the VM finishes starting.

  This may be related to this bug reported on the VirtualBox forums:
  
https://forums.virtualbox.org/viewtopic.php?t=111889&sid=5cd33c0872a03b689e7e9f84d850f538

  https://forums.virtualbox.org/viewtopic.php?t=111918

  Ubuntu is 22.04.4 LTS, kernel is 5.15.0-116-generic, VirtualBox is
  6.1.50-dfsg-1~ubuntu1.22.04.1.

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/linux/+bug/2073267/+subscriptions


-- 
Mailing list: https://launchpad.net/~kernel-packages
Post to     : kernel-packages@lists.launchpad.net
Unsubscribe : https://launchpad.net/~kernel-packages
More help   : https://help.launchpad.net/ListHelp

Reply via email to