Repository navigation
[Bug]: Windows 10 guest bugchecks and triple-faults on VirtualBox 7.2.22, Linux kernel 7.0; was stable on 7.1.6 #973
Description
Version
7.2.22
Host OS Type
Linux
Host OS name + version
Ubuntu 26.04.1 LTS, kernel 7.0.0-42
Host Architecture
x86
Guest OS Type
Windows
Guest Architecture
x86
Guest OS name + version
Windows 10 19045
Component
Unspecified
What happened?
A Windows 10 guest that ran stably under VirtualBox 7.1.6 on kernel 6.8.0-147 now bugchecks under VirtualBox 7.2.22 r175245 on kernel 7.0.0-42. I expected the guest to keep running, as it did before the host upgrade on 22 October 2026.
The guest raises the bugcheck. About 13 seconds later Windows resets itself through the Hyper-V reset MSR, which looks like the VM rebooting. VirtualBox stays up. Twice the guest then triple-faulted (VINF_EM_TRIPLE_FAULT); the VM was left Stuck until I powered it off. There is no host assertion and no VirtualBox process crash.
Host: Ubuntu 26.04, kernel 7.0.0-42-generic, ThinkPad E14 Gen 3, AMD-V. Package LINUX_64BITS_UBUNTU_26_04.
Guest: Windows 10 10.0.19045.4894, 2 CPUs, 3 GB RAM, VMSVGA with 3D disabled, Hyper-V paravirtualization. Guest Additions 7.2.22 r175245.
Previously working: Oracle VirtualBox 7.1.6 r167084 on kernel 6.8.0-147.
Four sessions, all on 7.2.22:
| Start | Fault |
|---|---|
| 9 Nov, restored saved state | 6 h 20 m: DRIVER_OVERRAN_STACK_BUFFER (0xF7). After Windows rebooted, 9 min later: IRQL_NOT_LESS_OR_EQUAL (0xA), IRQL 0xFF |
| 10 Nov 04:59, restored saved state | 10 min: 0xA, IRQL 0xFF. Faulting instruction is at a fixed offset 0x43418 from an address ending in bea00. 3 h 53 m: triple fault, both CPUs executing the same ret (+0x3b70f from an address ending in bea00) while delivering a double fault; cr2 is the double-fault IDT slot |
| 10 Nov 10:36, cold boot | 1 h 23 m: UNEXPECTED_KERNEL_MODE_TRAP (0x7F), trap 8. The two stack addresses are 0x5F00 bytes apart |
| 11 Nov 05:52, restored the saved state of that cold boot | 0.25 s: PAGE_FAULT_IN_NONPAGED_AREA (0x50), a write at IRQL 3. After Windows rebooted, 2 min 46 s: 0xA at the same bea00 / +0x43418 site. 4 min 21 s: the same triple fault, both CPUs on the same ret |
VBox.log
VBox.1.log
VBox.2.log
VBox.3.log
This may be related to #487 (varying Windows guest bugchecks on a Linux host with kernel 6.18+, absent on 6.12). One 0xF7 still occurs on 7.2.22, which already contains the fix for #590.
VBoxVBVAEnable failed is printed on every boot because 3D is off, including sessions that ran for hours. AHCI#0: Port 0 reset about 11 minutes after a completed boot is a separate stalled command to the VDI; the other resets happen while Windows is already dying.
How can we reproduce this?
I do not have a minimal test case. The guest is an existing Windows 10 installation, and the bugcheck time varies from a fraction of a second to several hours.
- Host was Ubuntu with kernel 6.8.0-147 and VirtualBox 7.1.6. This Windows 10 guest ran there without these bugchecks.
- On 22 October 2026 the host moved to Ubuntu 26.04, kernel 7.0.0-42. VirtualBox 7.1.6 was removed and 7.2.22 r175245 installed. Guest Additions inside Windows were updated to 7.2.22.
- Cold boot (
VBox.1.log): the guest runs, then at 1 h 23 m bugchecks with0x7Ftrap 8 and Windows restarts. The session was later saved. - Restoring that saved state (
VBox.log): 0.25 s after the guest reaches Running it bugchecks with0x50. Windows restarts, then bugchecks with0xA, restarts again, and triple-faults. The VM stays Stuck until powered off. - An earlier restored session (
VBox.3.log) ran 6 h 20 m, bugchecked with0xF7, rebooted, and bugchecked with0xAnine minutes later.
Opening the VM settings dialog does not cause it. During the 11 Nov run the settings window was opened while Windows was already on the 0x50 blue screen, and the reset 13 seconds later is Windows automatic restart.
Did you upload all of your necessary log files, screenshots, etc.?
- Yes, I've uploaded all pertinent files to this issue.
Activity