Having spent the past several days researching this, I realize I'm not going to get anywhere without some kind of debugging assistance. The error message is generic, possibly owing to there being several possible causes, none of which show up in any VirtualBox or system logs. The rest of the error message (either on the command line or in a pop-up window, depending on whether "VirtualBox" or "VBoxManage" is used to start the VM) is equally unhelpful (to me, anyway):
Result Code: NS_ERROR_FAILURE (0x80004005)
Component: MachineWrap
Interface: IMachine {<UUID>}
Environment is CentOS 7.5.1804 with kernel version 3.10.0-957.12.2, 64-bit, on a Dell Inspiron i7 laptop (not sitting in front of machine right now -- model to follow if requested/needed). Selinux is running in "enforced / targeted" mode, although that doesn't seem to be a factor as far as I can tell: no errors showing up in the selinux logs. Due to difficulties with multiple-monitor setup/configuration, it was necessary to download and install the latest compatible version of the "amdgpu" drivers (the non-Pro variant), the presence of which *may* be causing the problem.
The official VirtualBox 6.0.8 RPM for "el7" seems to install without any issues, and the four kernel modules build and load without error. Using "vagrant" to build and deploy a headless 64-bit Linux VM works well: the headless VM case seems to be the only one that will work.
I tried downloading the VirtualBox OSE source and rolling my own, making sure that updated libraries provided by the "amdgpu" package were linked against during the build. A non-hardened build works great when run directly out of the build's "out/(...)/bin" directory (after building, installing, and loading the modules, of course). At that point, I figured building with the updated "amdgpu" libraries fixed the problem, so did a hardened "release" build with RPM packaging (cd src/VBox/Installer/linux ; rpm/rules binary).
The hardened build worked identically to the official VirtualBox download
. Edited the "rules" file mentioned above to create an RPM installer for a non-hardened build. Same broken behavior (!) when attempting to start a non-headless VM.So at this point in the troubleshooting effort, I'm attempting to identify the differences between running a non-hardened build in its unpackaged state, and running the RPM-packaged non-hardened (or hardened) build. Frankly, I think "hardened" vs. "non-hardened" as suggested by someone else experiencing this problem (mentioned in another forum thread) is not the issue. So, what does the packaging process do that "breaks things" when the package is installed?
One obvious difference is the lack of "udev" entries for device creation when the "vbox*" modules get loaded: the VirtualBox USB entries under "/dev" don't get created without those "udev" rules. Similarly, the owner, group, and permissions of the device entries don't get set properly for the unpackaged non-hardened case, so the quick-and-dirty workaround is to simply do "chmod 666 /dev/vbox*" as hinted at near the bottom of the wiki's "Linux build instructions" page. Another difference is stripping of selected binary objects in the packaged version.
No idea what's causing the child process (started VM) to exit immediately. Is there any way to "instrument" that spawned process to determine why it's exiting?
Thanks in advance for any help provided!
