[4.3.0 regression] VirtualBox keeps crashing the host OS (constant kernel panics)

[Artem S. Tashkinov]

Host OS: Linux 3.12 vanilla, i686, PAE, 16GB RAM, Intel Core i5 2500
Guest OS: Windows XP SP3, i686, English, vanilla, no patches, no software, just VBox additions.

GuestOS parameters:

RAM: 640MB
IO Apic: ON
Processor: 1 or 2 (crashes anyway)
Execution Cap: Off (100%)
Acceleration: Full (VT-x + Nested Pages)
3D acceleration: Off (I'm not using 3D apps anyway)
2D video acceleration: on
VidMem: 30M
Monitor: 1
PAE/NX: disabled
Chipset: PIIX3
Pointing Device: PS/2 Mouse
EFI: Off
UTC HW Clock: Off

Guest has only 2D acceleration turned on and it has no audio enabled. I'm not doing any fancy stuff - crashes occur when the guest is fully idling. It more often crashes when I'm playing audio on my host (seems like playing audio on the host triggers the bug faster).

4.3.3 build 90796 crashes too. 4.2.x has never crashed here.

I cannot show you kernel panics, because my /var/log/messages is empty on reboot.

[Frank Mehnert]

Any change if you disable 2D video acceleration?

[amaxware]

I am having "extreme" problems booting Windows 7 Pro guest, using vbox 4.3.2. Vbox crashes the boot process with Guru remediation.

I have removed 2d acceleration and it just booted the guest.... maybe there is something to that.

[dmitrya]

I have the similar problem — spontaneous host kernel freezes. Host is x86_64 linux, guest is i686 linux. VirtualBox 4.2.x has never crashed, but VirtualBox 4.3.0 and 4.3.2 freeze host. The problem may occur and 10 minutes after start of guest and 2 hours after its start. Guest may be fully idle or not. 3.8.x and 3.10.x host kernels are affected. I've tried to collect logs using netconsole (see attachments for example). Every time logs are different.

[dmitrya]

host kernel logs after freeze

[dmitrya]

VM Settings

[Ramshankar Venkataraman]

Please give output of your linux kernel parameters.

cat /boot/config-`uname -r` | grep -i preempt

or

zcat /proc/config.gz | grep -i preempt

[Eugene]

Here are the relevant options from that kernel config on that dmitrya's system.

CONFIG_TREE_PREEMPT_RCU=y
CONFIG_PREEMPT_RCU=y
CONFIG_PREEMPT_NOTIFIERS=y
# CONFIG_PREEMPT_NONE is not set
# CONFIG_PREEMPT_VOLUNTARY is not set
CONFIG_PREEMPT=y
CONFIG_PREEMPT_COUNT=y
# CONFIG_DEBUG_PREEMPT is not set
# CONFIG_PREEMPT_TRACER is not set

Do you need the full .config?

[Eugene]

Besides, Memtest has shown no errors so far on that system.

As many of the failures we observed there were when a kernel tried to execute an instruction at a bad address (RIP=0x180, for example), I suspect some kernel-mode component to corrupt stack, i.e. saved return addresses.

However, we could not get a valid stack trace so far.

[Ramshankar Venkataraman]

Please try with 4.3.4. There are fixes that may address this problem.

[amaxware]

Same problem, I did a Windows reboot from the client, some set, crashes on start up. I have to keep retryng until I get the Windows screen, winows failed last boot option. Pick boot windows normally, then it works.

[Ramshankar Venkataraman]

Replying to amaxware:

I am having "extreme" problems booting Windows 7 Pro guest, using vbox 4.3.2. Vbox crashes the boot process with Guru remediation.

I have removed 2d acceleration and it just booted the guest.... maybe there is something to that.

This ticket is NOT about guru meditations and crashes of the guest. Please open a different ticket for it.

[Eugene]

Replying to ramshankar:

Please try with 4.3.4. There are fixes that may address this problem.

I built and tried 4.3.4 - the problem has not shown up so far. Still, it sometimes took a while to reproduce it in 4.3.2, so one cannot be sure it is fixed.

I do see system hangs with 4.3.4 when I use some particular debugging tools to check the VirtualBox host kernel modules but this is likely to be a separate issue and may be not in VirtualBox but rather in these tools.

I have also rebuilt the host modules from 4.3.4 with stack protector enabled and used these for some time. The stack protector did not detect stack corruptions.

When I have time, I will build the kernel with DEBUG_OBJECTS_* and other debugging facilities enabled and will try to use VirtualBox 4.3.4 there.

[Eugene]

Tried 4.3.6 on kernel 3.10 built with lockdep, DEBUG_OBJECTS_* and some other debugging facilities enabled. No suspicious activity in the host kernel so far.

[Artem S. Tashkinov]

This was fixed a long time ago (at least with 4.3.4).

Please, mark fixed and close this bug report.

[Frank Mehnert]

Thanks for the feedback.