Guru Meditation -4014 (VERR_VMX_INVALID_GUEST_STATE) with 4.3.8

[Yoda]

Host: Fedora 20/64 Guest: eCS Vbox 4.3.8

Never seen a crash like this with any previous Vbox version

[Ramshankar Venkataraman]

For my referernce HM error 0x58 is VMX_IGS_RFLAGS_IF_INVALID.

Thanks for the report, will see if I have the appropriate guest here to try replicate the problem.

[Ramshankar Venkataraman]

Please tell me the exact guest OS you are using. Which version of eComStation?

[Yoda]

IIRC it is based of eCS 2.0
Kernel is IBM 14.104a_SMP
ACPI driver: 3.22.3
Video: Panorama 1.06

[Ramshankar Venkataraman]

I installed eCS 2.0 with the SMP kernel and ACPI. The guest still sees only 1 VCPU.

So I'm not sure what else needs to be done for it to use both the CPUs. Could you give instructions if any on how to enable the other CPU after installing the SMP kernel and ACPI support?

Otherwise, I'm afraid I'll need a copy of your VM to see if I can reproduce the problem here.

[Yoda]

If SMP kernel and ACPI driver is installed, it should detect the number of CPU's at boot by itself.

Check config.sys for the line:

PSD=ACPI.PSD

There should not be any other parameters on that line.

[Yoda]

BTW, I'm not getting any email notofication about updates in this ticket. Aren't we supposed to get that ?

[Ramshankar Venkataraman]

eCS with SMP kernel and ACPI installed, still doesn't bring up 2nd VCPU.

[Yoda]

Are you sure the SMP kernel is installed ? check with:

bldlevel os2krnl

from a cmd window

or use the ACPI wizzard in 'Local System' -> 'Install/remove'

[Yoda]

To figure out, why this didn't work for you, I just made a new install of eCS 2.0 It indeed doesn't work for more than 1 CPU with the ACPI driver that comes with eCS 2.0; but that is also known to be very defective. Thats why I specified the additional drivers, I had updated.

I just added new ACPI and Panorama to this test setup, and it works for 2 cpu's (My I7 has 4+4 cores) but trying 3 or more often crashes at boot - so don't do that (unless you want to fix it ;-) ) )

[Ramshankar Venkataraman]

Silly question, never used eCS before, but where/how can I update the ACPI driver to the required version?

[Yoda]

Give me your email, and I'll send it

[Ramshankar Venkataraman]

Could you please send it as an attachment as a private message through the ​VirtualBox forums? My username there is "ramshankar".

[Yoda]

I couldn't figure out how to attach it, so I sent an URL instead.

[Ramshankar Venkataraman]

Managed to install the new ACPI drivers, and now the guest recognizes 2 VCPUs. "Processors Initialized: 2" while boot up. CPU Monitor also shows 2 CPUs now.

I then tried with 3 CPUs and it seemed to work. I didn't notice any crashes on bootup. I'll keep trying a bit more.

[Ramshankar Venkataraman]

SMP working with 4 VCPUs

[Yoda]

Interesting... First, the SMP Monitor app has been defective for a long time - it doesn't show correct CPU utilization. The eCenter at the bottom shows it correctly, and vill show you a graph for each detected cpu too.

Thinking back, I think I did all my SMP tests with 4.1 and 4.2 , and haven't tried more than 2 VCPUs since.

So I just tried it on my eCS server. Just to be rough, I gave it 8 VCPUs. Previously, as I said, more than 2 VCPUs would crash very early - before boot-logo. With 8, it passed that point, and loaded all drivers from config.sys - but froze there. OK, back to 4 VCPU (matching hw). Boot was clearly much slower( than 2VCPUs) - with a long pause after last driver was loaded, but desktop came up. However, it reset the guest just after. It booted up again, just as slow as before - but this time it managed to load all the apps the server runs - and then it completely froze.

Back to 2 VCPUs - boot is fast, and everything is stable again.

Conclusion must be, that the old bug (booting causes an IPE in early boot with more than 2 VCPUs) are gone, but something else is clearly wrong.

To get back to the original topic for this ticket ( :-) ) ... the Guru error has only happened that one time, after the server had run for a few days.

[Yoda]

Crash haven't happened for 4.3.10 or 4.3.12 - I guess we can close this ticket

[Ramshankar Venkataraman]

Closing, reopen if relevant.

[Yoda]

I guess I said that too early - it just happened again. Log uploaded

[Yoda]

...and it happened again - vbox2.zip (logs+png) attached.

[Ramshankar Venkataraman]

Thanks, will look into it. Need to think how this can happen, although I have an idea, will need to double-check if that can really happen.

[Ramshankar Venkataraman]

Would you be willing to try a test build to help narrow down the issue?

[Yoda]

Sure.

BTW, I noticed from the png's, that the crash usually happens just after 05.00 . At 0500 the newsserver starts expire process - that puts quite a load on CPU and I/O - maybe this has something to do with it.

[Yoda]

On my laptop eCS just crashed with:

Guru Meditation 1155 (VINF_EM_TRIPLE_FAULT)

Vbox3 log attached

[Ramshankar Venkataraman]

Here is an instrumented build which could help narrow down the problem, feel free to give it a try but remember it's a bleeding-edge build:

https://www.virtualbox.org/download/testcase/VirtualBox-4.3.53-94551-Linux_amd64.run (link expires in ~14 days)

[Yoda]

Installed. Lets see if I can make it crash :-)

Do you have a Win7 build too ? My laptop runs Win7 - and that was crash nr 3.

[Yoda]

vbox4.zip attached. Guru with new build

[Yoda]

Guru Meditation 1155 (VINF_EM_TRIPLE_FAULT) with 4.3.14

Vbox5.zip attached.

[Yoda]

OK, these Tripple faults keep happening here. Not exactly same env as org post - this is on my laptop running Win7 host, and eCS 2.1 guest

I've had 10 so far :-/

This lap has been running this env for at least a year without any GURU's - seems to just start with 4.3.14.

[Ramshankar Venkataraman]

I've made some fixes to the event injection logic which affects Linux, Solaris hosts (which will be part of the next maintenance version of VirtualBox). I'll try upload a test build for you soon. As for the Windows triple-fault that may or may not be related, can't say yet.

[Yoda]

I just managed to reproduce the tripple-fault on Linux, using same VM. If you think this is unrelated to the original problem here, I can open another ticket for this. Vbox6.zip attached.

[Yoda]

After having updated to v4.3.20 and giving it a few heavy stress tests, it seems that the bug is finally fixed. Lets close this ticket.

[Yoda]

Well, it seems everytime I think it is fixed, it crashes again. Vbox7.zip attached version 4.3.20