Regular compiler *crashes* on build machine running as VBox guest

[MReimer]

I don't know if a bug report for my issue already exists.

If you need further information, please assist me with getting the required logfiles. I *don't know* how to get a VBox.log file!

I'm running a build machine in context of Virtualbox. With this build machine, I regularly get compiler crahes. If I restart the compile process, then sometimes anything works as expected and sometimes I get another crash at another position and often even with other crash reasons.

I've attached two screenshots and my /proc/cpuinfo

I tried with "Nested Paging" enabled and disabled, but this didn't change anything.

[MReimer]

I didn't expect that Virtualbox stores the log file without any need to configure something.

I've attached a VBox.log file. For me it seems to not have anything special about the crash which can be seen in latest_error.png.

[Frank Mehnert]

Is the guest a clean installation or was the VM disk copied from somewhere? Also, is there any change if you decrease the number of guest CPUs to 1? And of course, are you 100% sure that the guest is set up correctly?

[MReimer]

Clean installation made with an older version of VirtualBox. I decreased the CPU-Number to 1 but no change...

See attached VBox.log and screenshot

[Frank Mehnert]

You say that you created this guest with an older version of VirtualBox. Did you perform the same tasks on the older version (btw, which one?) and it did not crash? Or did you just install the guest on the older version and then switched to 4.2.6? Is there any chance that your virtual hard disk is broken somehow, that is, can you check the integrity of the tools installed in your guest? For instance, in Debian there is a 'debsums -c' command to verify the checksum of all installed packages.

Are these crashes really random or does the compiler always crash as the same time when compiling a certain file?

[MReimer]

As Archlinux is a rolling release system, I can't tell you exactly in which version I initially created the virtual machine.

I've been able to build my package set in the "older version". But I didn't use this build machine for some time, so there may have been several updates in the meantime.

I don't think that Archlinux has some kind of "integrity check".

About the crashes: That's why I added all those screenshots. They show some of the crashes. I always started from the same base (means: I killed anything that already has been built and started from scratch). The crashes happen at different stages while building. The interesting thing is that the virtual system itself always keeps stable. I'm able to restart the build process and a clean shutdown still works.

As I need a working build machine to continue my development work, I decided to give KVM a try.

I used "VBoxManage clonehd" to transform the virtual machine, that caused problems on VirtualBox, to the "raw" image format and then used "qemu-img convert" to convert to qcow2.

I use the following command to launch the virtual machine:

kvm -m 1G \
    -smp 2 \
    -usbdevice tablet \
    -hda "$HARDDRIVE"

And the result of this try: Exacly the same (clone!) machine, that caused problems on Virtual Box, runs well with KVM! I compiled my full package set twice (each run takes about 15 minutes). Then I did a second try with "MAKEFLAGS=-j3" to speed up the build process and stress the KVM emulated machine even more. Workload on both physical CPU's at 100% but no problems at all!

Then I did a few more tries on Virtual Box. I removed the shared folder and uninstalled the guest utilities on the VM. No changes. Still get those crashes...

[Frank Mehnert]

Thanks for the additional tests. Is there any chance to make your VM available to us? If so, you could put the VM on our anonymous FTP server at ftp.oracle.com:/appsdev/incoming and mail me at frank _dot_ mehnert _at_ oracle _dot_ com. I fear, as long cannot reproduce your problem we cannot debug it.

[Frank Mehnert]

Another thing: I saw that you run the guest with 256MB on VBox and with 1G on KVM. Did you try to run the guest with 1G on VBox as well? I want to rule out any guest issue (for example OOM killer).

[Frank Mehnert]

MReimer, any updates here?

[Frank Mehnert]

No response, closing. Many changes in VT-x / AMD-V with VBox 4.3, so perhaps even fixed for you.