Opened 12 years ago
Closed 10 years ago
#10729 closed defect (fixed)
GM(VERR_VMX_INVALID_GUEST_STATE) with NetWare 6.5 SP8 SMP guest
Reported by: | kumba | Owned by: | |
---|---|---|---|
Component: | guest smp | Version: | VirtualBox 4.0.16 |
Keywords: | netware, smp, guru meditation | Cc: | |
Guest type: | other | Host type: | Windows |
Description
Under both VirtualBox 4.0.16 and 4.1.14, using either the ICH9 or PIIX3 chipsets, NetWare 6.5 SP8 is unable to use SMP capabilities. Issues range from generating an ABEND that cites a memory parity error, to an issue initializing Processor #1 at boot (which triggered a GuruMeditation -4014, VERR_VMX_INVALID_GUEST_STATE).
Under 4.0.16, I get the ABEND if I cycle the VM enough times. Sometimes, it just locks up loading several NLMs, but usually, you get the ABEND and the system either hangs or auto-resets itself. Under 4.1.18, no ABEND -- it simply freezes at random points. Tried anywhere from 2 to 4 CPUs, effect is the same, only faster if you have 4 CPUs active.
Drop it to one CPU, and it loads fine under 4.0.16, and only on 4.1.14 (See Ticket #10728 for the Guru on 4.1.18). Logs and screenshots attached.
Attachments (3)
Change History (9)
by , 12 years ago
Attachment: | vbox4016-guru-20120704-nw65sp8.txt added |
---|
by , 12 years ago
Attachment: | vbox4016-screenshot-20120704-nw65sp8-SMP-CPU1-fail.png added |
---|
Screenshot of the NW65SP8 console on Vbox 4.0.16 when the GuruMeditation happened in trying to init CPU1
by , 12 years ago
Attachment: | vbox4016-screenshot-20120704-nw65sp8-SMP-ABEND.png added |
---|
Screenshot of the NW65SP8 console on Vbox 4.0.16 of the ABEND while trying to use SMP.
comment:1 by , 12 years ago
Summary: | NetWare 6.5 SP8 SMP broken in Vbox 4.0.x, 4.1.x + GuruMeditation → GM(VERR_VMX_INVALID_GUEST_STATE) with NetWare 6.5 SP8 SMP guest |
---|
comment:2 by , 12 years ago
Managed to trigger several more ABENDs, but only with two processors. Any more than two, and you typically get a Guru when it is trying to start up the disk drivers.
So far, these ABENDs don't point at a memory parity error -- that's just a symptom. It appears there's some loss of synchronization between the different CPUs, and this causes one or more of the CPUs to hog the CPU timer, which the SERVER.NLM module catches and forces the ABEND. Quick guess is maybe the interprocessor interrupt handling isn't working right under NetWare?
I managed to dig up the virtual debugger toolkit off of Novell's site that lets me analyze the core dumps from within Windows. Not 100% sure what I am looking for, but perhaps something with stand out.
comment:3 by , 12 years ago
Oh, one interesting observation -- if the OS is stuck loading a module for a period of time, I found that by jiggling the mouse cursor inside the window somehow gets it moving again. I am using the USB Tablet emulation, however, and I noticed that NetWare loads OHCI and EHCI drivers extremely early in the boot phase, so it's apparently aware of USB and any mouse movement inside the window generates additional interrupts that cause one of the two CPUs to "move along".
comment:4 by , 10 years ago
This behavior persists under 4.3.6 (tested PUEL on openSUSE 12.3, 4.3.6, r91406). There are some additional lingering issues with VTx (apparently), but I'm waiting to open a real bug until I have more time to get some current logs and such. The SMP issue is consistent, however.
comment:5 by , 10 years ago
Please attach a VBox.log file when running this guest on VBox 4.3.6. If "lingering issues with VT-x" means that it wouldn't run with VT-x disabled then this is a different problem.
VBox.log from NetWare 6.5sp8 under 4.0.16, Guru Meditation w/ SMP enabled.