Virtualbox 5 on Linux Host causes Host to crash.

[z5d69]

Hi, before I provide the vbox.log file, etc. I wanted to see if anyone was having problems with their host crashing when Virtualbox is running? I hadn't had any problems with my server prior to upgrading from VirtualBox-4.3-4.3.20_96996_openSUSE123-1.x86_64
to VBox 5. At that point, I can't run for more than 3 days without the server crashing (ONE Windows 7 64-bit guest running). When I went back to VirtualBox-4.3-4.3.20_96996_openSUSE123-1.x86_64
, the host runs without crashing. AS WELL, I notice that, in 5, the load average appears to be 1.0 with NOTHING running (barring some new processes that are apparently part of 5) while in 4.3, the load is maybe 0.2 or LESS unless the guest is working something? The crash did NOT appear to log anything nor did VBox but, again, first I'm just trying to figure out if other people are running into this.

Thank you in advance!

[Ramshankar]

The load average will be fixed in the upcoming 5.0 maintenance release. It is basically linux accounting uninterruptible sleep as "load" which technically is not the kind of load which keeps a CPU busy. So it is a somewhat unexpected and quirky way in which Linux accounts for what is "load". You can verify this because the "iprt-VBoxTscThr" thread in your "ps -efl output" will be in state "D" (uninterruptible sleep).

As for host crashes, I have no idea without getting more information from you like which host kernel, VBox.log when the VM is running, dmesg output of kernel panic etc.

[z5d69]

Okay, that resolves the load average part.

At this point, I'm letting the system run at the lower VirtualBox version to make sure everything looks okay, before going back up VB 5.0.

uname -a
Linux dlquad 3.16.7-21-default #1 SMP Tue Apr 14 07:11:37 UTC 2015 (93c1539) x86_64 x86_64 x86_64 GNU/Linux

I'm using opensuse 13.2, 64-bit on an AMD quad processor. For the rest, I'll get back to you once I've upgraded again.

Can you please provide me with the information you need as well as how to get it? VBox.log, got it. I can attach that once I've upgraded again (and the system has crashed). dmesg... just redirected to a file? Is there a particular command you want me to run or just 'dmesg >> output.log'? Is that just the next time it crashes and I have rebooted? What else? It's just that once I've provided this, I'll need to go down to the lower version again until there's a 'fix' (or more information or testing)...

[nmartinho]

I confirm this problem.
Causes an unexpected shutdown daily.

Host: Linux butler 3.2.0-4-amd64 #1 SMP Debian 3.2.68-1+deb7u2 x86_64 GNU/Linux
VirtualBox: 5.0.0r101573

[mpack]

I don't think we need any more confirmations of the problem, given that the devs have already acknowledged it and announced a forthcoming fix (see Ramshankar's post above).

[z5d69]

The forthcoming fix was for the running process though and NOT the host crash? I think that we were getting confirmation that 5.0 causes host crashes. The load is what was stated by a developer as a known condition.

[z5d69]

I'm trying to be thorough here so I am trying to do the following testing, in order:

1. Run the host for a couple days to see if there was a crash (there wasn't).
2. Run stress using up all CPU resources and almost all of the RAM (no crash however this has only been running for about 8 hours... no CPU overheat problems or other issues though).
3. Install the VBox drivers while 2 is still in place (./vboxdrv start) and let sit for a day.
4. ONLY IF 1 - 3 executed without problem, start a virtual machine and see what happens. I DID set paravirtualization for Hyper-V because it's a windows guest running on an AMD 64 host. I'm unsure if this can cause a crash. I DO run this client under a non-root userid. Is it possible for a non-root instance of a virtualbox guest to cause a system crash?

None of the logs (opensuse 13.2 host) indicate WHY or WHAT is causing the server crash, BTW... is there something I need to do to turn logging UP for root cause analysis or does this potentially indicate a hardware issue, if there is no logging of a kernel panic, etc.?

[mpack]

The forthcoming fix was for the running process though and NOT the host crash?

Right, I didn't notice the last sentence of Ramshankar. In that case the details he asked for should be provided with any confirmation.

[z5d69]

UPDATE:

I'm trying to be thorough here so I am trying to do the following testing, in order:

1. Run the host for a couple days to see if there was a crash (there wasn't).
2. Run stress using up all CPU resources and almost all of the RAM (no crash after 2 days at 100%).
3. Install the VBox drivers while 2 is still in place (./vboxdrv start) and let sit for a day. No problems
4. Started a virtual machine up (windows 7). I updated paravirtualization to "None" and started it in regular mode vs. headless, etc. I'm unsure if this can cause a crash. I DO run this client under a non-root userid. Is it possible for a non-root instance of a virtualbox guest to cause a system crash? Can paravirtualization of a guest cause a crash?

None of the logs (opensuse 13.2 host) indicate WHY or WHAT is causing the server crash, BTW... is there something I need to do to turn logging UP for root cause analysis or does this potentially indicate a hardware issue, if there is no logging of a kernel panic, etc.?

[z5d69]

Not trying to muddy the water here however can paravirtualization settings in virtualbox guests cause a host crash? I changed the setting to None and, so far, no problems. I'm using a Phenom II AMD processor (quad core) and I'm wondering if the motherboard bios back then didn't handle the configurations correctly?

[Perryg]

Not trying to muddy the water here however can paravirtualization settings in virtualbox guests cause a host crash? I changed the setting to None and, so far, no problems. I'm using a Phenom II AMD processor (quad core) and I'm wondering if the motherboard bios back then didn't handle the configurations correctly?

Would it be possible for you to test with the legacy setting instead of none?

[z5d69]

I'll shut down and switch to that now.

[z5d69]

If you could tell me how each option works for paravirtualization? I'm unclear the difference between none and legacy? I didn't see a lot on this in the manual.

[z5d69]

And yes, rebooted in legacy mode now!

[Perryg]

This whole paravirtualization section is new and as far as I can determine is still being tested and worked on with version 5. I believe there will be an improvement when the next maintenance release comes out thanks in no small part by the people that report issues here. Legacy is supposed to be the old way to control the time sync of the guest and the best general setting for version 5 is the default setting which end users should not change. Meaning that users should not try to change it unless they know what they are doing and even then it may not work as expected. I see repairs and added changes done daily in trunk which is where I work. I suspect more information will be coming when there is time and the code has matured enough that changes will not be effecting what you see now.