Windows 10 guest on 5.8 kernel STATUS_DATATYPE_MISALIGNMENT exception

[Macy]

As referenced also in threads t=99789 and t=99028
Windows 10 guest on Linux 5.8 kernel experiences random exceptions of type "STATUS_DATATYPE_MISALIGNMENT"

This occurs to me on Ubuntu Linux 20.04.1 with hwe kernel 5.8.0-36-generic and Virtualbox 6.1.16-140961~Ubuntu~eoan installed through the official virtualbox focal apt repository. (This is the only Virtualbox version compiling on 5.8 kernel)

This problem occurs even with a virgin installation of Windows 10 (guest), so it is not attributable to 3rd party drivers. Furthermore, it does not occur with the non-hwe kernel 5.4 shipped with Ubuntu 20.04.1.

Any advice or request for debugging is appreciated.

My error log:

Code: Select all

00:21:26.789581 GIM: HyperV: Guest indicates a fatal condition! P0=0x1e P1=0xffffffff80000002 P2=0xfffff8065acd7d93 P3=0xffffe0077a4663bb P4=0x7010008004002001
00:21:26.789648 GIMHv: BugCheck 1e {ffffffff80000002, fffff8065acd7d93, ffffe0077a4663bb, 7010008004002001}
00:21:26.789649 KMODE_EXCEPTION_NOT_HANDLED
00:21:26.789649 P1: ffffffff80000002 - exception code - STATUS_DATATYPE_MISALIGNMENT
00:21:26.789649 P2: fffff8065acd7d93 - EIP/RIP
00:21:26.789649 P3: ffffe0077a4663bb - Xcpt param #0
00:21:26.789649 P4: 7010008004002001 - Xcpt param #1
00:21:30.726903 VMMDev: vmmDevHeartbeatFlatlinedTimer: Guest seems to be unresponsive. Last heartbeat received 4 seconds ago

[fth0]

In the VM configuration, just for a test, disable audio and network completely. If the problem persists, post a zipped VBox.log file (after closing any error dialog).

[Macy]

Thanks for your reply fth0. Here attached are the log files for your review:
- with disabled USB, Audio, Networking
- with disabled Audio, Networking
- with enabled USB, Audio, Networking

The Windows 10 Guest has latest Guest Additions installed matching the version of the Virtualbox software.
A similar installation of Windows 10 Guest does not suffer from such problem in Qemu on the same host with linux kernel 5.8.

Should you need further tests, please don't hesitate to let me know. Many thanks!

[fth0]

Thank you for performing the tests. They all show similar behavior and unfortunately do not point to the cause, but that's also a gain in knowledge.

Should you need further tests, please don't hesitate to let me know. Many thanks!

You could experiment with the display settings (Graphics Controller, 3D Acceleration) and with/without the Guest Additions.

I think another way forward would be to find someone who is both capable of Windows kernel and driver debugging and of reproducing the problem. They could (1) let the guest crash into a debugger and (2) analyze the crash, to possibly (3) discover something about the real cause of the crash. I cannot do (2) and (3) myself (yet), so I cannot tell anybody else (like you) what to do.

[Macy]

Thanks for replying fth0.

I performed the following tests, all resulted in the same exception STATUS_DATATYPE_MISALIGNMENT:
a) with/without 3D accelleration
b) with/without Guest Additions
c) with VBoxVGA/VBoxSVGA/VMSVGA graphic controllers

Such resets occur simply leaving the virtual machine idle with me away from keyboard.
During a 4 hour test, they occurred 12 times at: 56' ; 75' ; 76' ; 124' ; 131' ; 132' ; 139' ; 167' ; 193' ; 219' ; 235' ; 237'
In some cases, Windows 10 booted into system recovery mode, and (without any keyboard/mouse input) later experienced a reset even so.
Attached is the log.

[fth0]

I performed the following tests, all resulted in the same exception STATUS_DATATYPE_MISALIGNMENT

Thanks for your additional tests. With your help, we have eliminated the usual suspects (I think). You fell in your own trap, BTW: Search for GIMHv: BugCheck to also find the other types of exceptions you seem to have missed.

[fth0]

After reading several forum threads about similar problems, IMHO the following conditions apply:

1. On the host, a Linux kernel 5.8 or later is running, but not Linux kernel 5.4.

2. VirtualBox versions 6.1.14, 6.1.16 and 6.1.18 are being used. Running a VM with the same virtual disk in KVM or VMware does not exhibit the crash.

3. In the guest, Windows 10 or Windows Server 2016|2019 are running, but not Windows 7 or Linux. It suffices to start a VM from a Windows 10 ISO and to wait, so no VirtualBox component is running in the guest.

4. The crash occurs independent of the chosen Graphics Controller and the chosen Paravirtualization Interface.

---

I'd like you to try one or all of the following experiments with a VM with an installed Windows 10:

A. Start the VM with Windows in Safe Mode and wait up to 1 hour. Does the crash occur?

B. In the VM configuration, disable System > Acceleration > Nested Paging. Start the VM and wait up to 1 hour (note that without nested paging, a VM can be up to 10 times slower). Does the crash occur?

C. In the VM configuration, set System > Processor > Processor(s) to 1 CPU, and disable System > Motherboard > I/O APIC. Start the VM from a Windows 10 ISO and wait up to 1 hour. Does the crash occur?

[Macy]

Hi fth0, yes I confirm all the conditions. Following your request I additionally tested:

A: VM with Windows in Safe Mode. No reset within 60 minutes.
B: VM configuration with disabled Nested Paging. Reset occurs within 10 minutes.
C: VM configuration with 1 CPU. Booting from Windows 10 ISO. Reset occurs within 10 minutes.
D: VM configuration with Legacy paravirtualization. Reset occurs within 30 minutes.
E: VM configuration with None paravirtualization and unticked nested paging. Booting from Windows 10 ISO. Reset occurs within 10 minutes.

I could not test a Windows 10 ISO (new VM with empty hard disk) with disabled I/O APIC as this immediately leads to Windows error 0xc0000225.
The above tests occurred with version 6.1.18 r142142 installed from Oracle apt repositories running on Linux 5.8.0-43-generic #49~20.04.1-Ubuntu

[fth0]

Thanks for testing and reporting. Please see Ubuntu 20.04 host, Win 10 guest constant reboot (STATUS_DATATYPE_MISALIGNMENT) for the latest details and the ongoing research.