Opened 3 years ago
Closed 3 years ago
#20456 closed defect (duplicate)
Guru Mediation Crash when 5-level page tables are enabled
Reported by: | Larry Finger | Owned by: | |
---|---|---|---|
Component: | other | Version: | VirtualBox 6.1.22 |
Keywords: | Cc: | ||
Guest type: | other | Host type: | other |
Description
When 5-level page tables are enabled in the host, starting most VMs result in a guest crash with the Guru Mediation page being shown. This happens with 6.1.20, 6.1.22, and test build 6.1.x revision 145620. The machine log for Windows 10 is attached.
Attachments (3)
Change History (11)
by , 3 years ago
Attachment: | Windows 10 Clone-2021-07-12-15-12-18.log added |
---|
comment:1 by , 3 years ago
When this crash happens, the host log has
[ 234.088947]
!!Assertion Failed!! Expression: pCritSect->s.Core.NativeThreadOwner == hNativeSelf Location : /home/vbox/tinderbox/6.1-lnx64-rel/src/VBox/VMM/VMMAll/PDMAllCritSect.cpp(579) int PDMCritSectLeave(PDMCRITSECT*)
[ 234.088963] ffffae7c82e9e000 <R3_STRING>: ffffffffffffffff != 00007f479c2b2640; cLockers=-1 cNestings=1
comment:2 by , 3 years ago
This is weird. Why would enabling 5 level page tables have any real effect on your system? Your CPU was launched in 2013, and the first CPUs which have 5 level page tables became available this year. I don't believe you have a time machine in your basement :)
comment:3 by , 3 years ago
In addition to klaus' question above:
Can you easily reproduce the issue after rebooting the Linux host? If so, can you prevent it by booting the Linux host kernel 5.13 with the no5lvl parameter?
comment:4 by , 3 years ago
Are you sure it's the 5-level page tables and not CONFIG_RANDOMIZE_KSTACK_OFFSET_DEFAULT
being enabled? If the latter is enabled, try adding randomize_kstack_offset=0
to the linux kernel command line and see if it helps.
comment:5 by , 3 years ago
Have you noticed the weird display size of the vmsvga adapter, similar to Larry's:
00:00:00.833768 Display size: 4294967295x4294967295 4294967295bpp
On another VM, I found this in addition:
00:00:00.945302 !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! 00:00:00.945322 emR3Debug: rc=VERR_INVALID_STATE 00:00:01.379554 GUI: UIMachineViewNormal::resendSizeHint: Restoring guest size-hint for screen 0 to 800x600 00:00:01.380324 ERROR [COM]: aRC=NS_ERROR_INVALID_ARG (0x80070057) aIID={4680b2de-8690-11e9-b83d-5719e53cf1de} aComponent={DisplayWrap} aText={Argument aWidth is invalid (must be aWidth != 0 && aWidth <= 32767)}, preserve=false aResultDetail=0 00:02:06.590415 GUI: User request to power VM off on Guru Meditation. 00:02:06.590435 GUI: Passing request to power VM off from machine-logic to UI session. 00:02:06.590436 GUI: Powering VM down on UI session power off request...
Unfortunately, it's not telling us the value of aWidth, but 0xffffffff qualifies easily as being outside this range. I'll try bird's suggestion in a minute.
comment:6 by , 3 years ago
Please try the VirtualBox test builds 6.1.23r145697 (or newer) that are supposed to fix VirtualBox issues on hosts with Linux kernel 5.13, and report back. TIA.
by , 3 years ago
Attachment: | changes-linux-5.13-randomize-kstack-offset.diff added |
---|
comment:7 by , 3 years ago
The diff contains the fix. Useful for people who build VirtualBox packages themselves, because this isn't a tweak to the vboxdrv kernel module source code.
VM Log