Guest additions newer than 5.0.16 cause X to fail in Ubuntu 15 and 16 -> duplicate of #15860

[raynebc]

As reported on the forum a couple times, most recently here: ​https://forums.virtualbox.org/viewtopic.php?p=364318#p364318

In that thread, Michael indicated Oracle can no longer can reproduce the wide spread problems where the newer versions of the guest additions (newer than 5.0.16) caused the video interface to not launch in some versions of Linux, so he asked me to open this ticket.

(Edit (Michael): I am not aware of these "wide spread problems" at all, let alone able to reproduce them.)

My host OS is Windows 7 Pro x64 and the affected guest OSs are Ubuntu 15.x and 16.04 LTS. Originally I just grabbed the generic Ubuntu 15.10 VM image from osboxes.org and installed some various packages mostly just related to building a cross-platform application that I program for. The Ubuntu 16 VM was created by making a clone of the Ubuntu 15 VM within the Virtualbox manager and allowing the OS to upgrade to Ubuntu 16.

I will attach two sets of logs. One of them is a normal boot and shutdown while the version 5.0.16 guest additions are installed. The other set reflects the botched boot up when the 5.0.22 guest additions are installed and me manually opening a virtual terminal to issue the shutdown command.

[raynebc]

Successful boot and shutdown with version 5.0.16 additions

[raynebc]

Malfunctioning boot while 5.0.22 additions are installed and shutdown via virtual terminal.

[Michael Thayer]

Could you try to reproduce this using a fresh installation from Ubuntu installation media and without performing an upgrade after installation? If that fails, please try to find the most minimal reproduction scenario you can, as what you wrote above is still rather involved.

[cowbutt]

A CentOS 7 guest on Windows 8.1 behaves similarly with both 5.0.16 and 5.0.22. The Guest Additions have been broken in various ways since 5.0.18 for me.

[Frank Mehnert]

Replying to cowbutt:

A CentOS 7 guest on Windows 8.1 behaves similarly with both 5.0.16 and 5.0.22. The Guest Additions have been broken in various ways since 5.0.18 for me.

Sorry, but this information does not help at all. For me the Guest Additions of 5.0.18 / 5.0.20 and 5.0.22 work well with CentOS 7. So please enumerate and describe exactly what doesn't work for you.

[Michael Thayer]

Please, for issues like this provide us with the simplest possible reproduction scenario using a freshly installed guest system.

[raynebc]

Here are more simplified steps to reproduce the problem:

Download the x64 Ubuntu Wily VDI offered here: ​http://www.osboxes.org/ubuntu/

Register a new VM with that existing hard drive image, allow it automatically detect the type as Linux and the version as 64 bit Ubuntu. Set the base memory to 2048MB. Max out the video memory to 128 MB and enable 3D acceleration. Boot the new guest, insert the 5.0.22 guest additions ISO via the Devices VirtualBox menu and allow the guest to run and install them. Reboot the guest and the x11 problem is immediately observed. It's not a clean install, but it's a starting point and should at least be relatively easy to allow developers to reproduce the problem.

[raynebc]

Another finding: Disabling the 3D acceleration on the above example VM allows the GUI to work.

[Michael Thayer]

Tested it with current development code on a Linux host, but with official 5.0.22 Additions. So far I can't see any problems. I wonder if it is related to your Windows host graphics drivers. Did you actually try to reproduce this with a freshly installed machine?

[raynebc]

As requested, I tested with a fresh guest OS (Ubuntu Desktop 16.04 LTS x64).

VM setup: New VM, 2GB RAM, new VHD (VDI format, dynamic, 8GB), 128MB VRAM, 3D acceleration enabled. All other VM settings left to default configuration.

Guest OS setup: Did not enable the option to automatically install any updates during installation of the OS. Default partitioning options and language options for my region (US-English).

On the first boot after installation completed, it hung with a message saying that said to either upgrade the BIOS or to use some sort of boot option. After rebooting the VM, everything came up normally and I logged in. I installed the guest 5.0.22 additions and rebooted and once again x11 failed to start up. I found that it would briefly display a message (that I assume is the same as the one it did that first reboot after the installation of the guest OS) and then display a blank screen. I managed to screen capture the message, which says:

/dev/sda1: recovering journal
/dev/sda1: clean, 187102/393216 files, 1000629/1572608 blocks
[    6.512902] piix4_smbus 0000:00:07.0: SMBus base address uninitialized - upgrade BIOS or use force_addr=0xaddr
[    6.930460] intel_rapl: no valid rapl domains found in package 0

I'm not sure that output is very meaningful, because I see that it is displayed even before the GUI successfully starts up during boot (ie. whenever 5.0.22 guest additions are not installed).

Forcefully rebooting the VM led to the same behavior so I got to a virtual terminal and ran the guest additions uninstall script and rebooted. After this, the guest OS booted normally, GUI and all. So I installed the guest additions again by inserting the ISO via the VirtualBox devices menu and opting to install it when the prompt came up automatically. After a reboot, the GUI failed to load as usual. I shut the VM off and disabled 3D acceleration, and after this change, the VM could boot with GUI. Re-enabling 3D acceleration again prevents the GUI from working as I expected.

The old guest additions (5.0.16) work just fine if I install them in this new VM. For this reason I tend to think that it's not a problem with my system or Radeon's drivers, but that guest additions lost compatibility with some system configurations after 5.0.16.

[Michael Thayer]

Could you upload /var/log/Xorg.0.log from the guest immediately after X.Org has failed to start? The guest's dmesg output might also be useful.

[raynebc]

Xorg.0.log and dmesg output after X11 fails to start

[raynebc]

I have attached the requested logging.

[Michael Thayer]

It would be great if you could also look for any lightdm, Unity or Compiz logs on the system, as I see nothing unusual in those logs, except that the X server terminates immediately for no visible reason. Perhaps trying to start lightdm from the command line (sudo systemctl start lightdm) might give more information. By the way, is the console in normal text mode (25 lines) or graphical mode (definitely more than 25 lines)? I think we will need more back and forwards to work out what is happening here, so any investigation of your own would be very helpful too.

[raynebc]

I have no significant expertise with Linux, as I have mostly just been a Windows user most of my life. I don't know how to troubleshoot this type of problem and as such I am unfortunately only capable of providing information that I am specifically guided to collect.

When this problem occurs, virtual terminal 7 (the one used by x11) appears to be in text mode as there is a blinking cursor at the top left most character position of the screen. Based on the size of the cursor, it appears the display is configured to be 25 characters tall. The screen is otherwise entirely black, no keyboard input is reflected on screen if I type anything and no mouse pointer is displayed. I have to change to another virtual terminal to perform any command (such as copying the xorg log and dmesg output to another file I can retrieve later).

When I change to another virtual terminal to issue your suggested command to start lightdm, that terminal's console output blanks out with a blinking cursor at the top left of the screen exactly as terminal 7 behaves. It remains this way until I change to another virtual terminal and back, at which point it is back to normal and shows that I had issued the "sudo systemctl start lightdm" command. Changing to terminal 7 shows that x11 is still not functional.

[raynebc]

lightdm logs as of when x11 has failed to start

[raynebc]

I was eventually able to find and access the lightdm logs, which have been attached.

[Michael Thayer]

Still no success at reproducing this. Is there any way that you could make a virtual machine affected by this problem (the .vbox and the .vdi files) available for me to download?

[Michael Thayer]

Took a look at your lightdm logs again. In x-0-greeter.log I see repeated: "Failed to launch bus: Failed to connect to session bus". That does not look like a VirtualBox-related problem.

What prompted me to look again is that I am also having similar problems with an Ubuntu 16.04 virtual machine: if the host system is under load (e.g. compiling) the guest boots fine. If not, it gives me a message that "The system is running in low-graphics mode", but when I look on VT-7 the X server has started without any problems. Symptoms similar to this:

​https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=814760

What I am trying to get at: this may be a lightdm problem, perhaps a timing one - lightdm does seem to have timing problems on Ubuntu - which happened to be triggered by the different Additions version. You might look at your system log (execute "journalctl" to see it in Ubuntu 16.04) to see if there is any more information there. And you might want to try asking lightdm people if this might be a bug there. If they try to send you back to us, you could try getting them to talk to us directly.

[raynebc]

Regarding how wide spread this problem is considered to be, it seemed that with guest additions 5.0.18, there were X11 problems for enough people (some of whom posted in the discussion thread at ​https://forums.virtualbox.org/viewtopic.php?f=1&t=77283) that Oracle mentioned on the downloads page the possible necessity of downgrading the additions as a workaround. A subsequent release had reportedly fixed some related x11 problems, but it didn't resolve the problem I'd been experiencing.

I have compressed (7zip format) the VM's .vdi and .vbox files as requested: ​https://drive.google.com/file/d/0B1VNFqdtmy_iNXkxMEFzWnVWMjg/view?usp=sharing The password for the account is "V!rtualB0x"

I re-installed the 5.0.22 additions today to gather your requested "journalctl" output, but seemingly as a fluke, the GUI launched normally with the OS. While it was working, I used "sudo VBoxService --version" to verify the version and it did claim 5.0.22 was installed. To be sure, I removed the guest additions, rebooted, re-installed the 5.0.22 additions and rebooted. Sure enough, the GUI started up again. The only other change I could think of involved me changing the password from a weak one to a strong one before uploading a copy of the VM for your testing. However, even after re-instating my old password and re-installing the guest additions, they now seem to be working normally in this from scratch Ubuntu install VM. 3D acceleration is still enabled for it, even increasing the processor count from 1 to 2 didn't cause problems. I have no idea why it decided to work. I hadn't touched it since I gathered information for this ticket about a week ago.

I launched my other Ubuntu VMs and they all work with 5.0.22 additions now. My host OS hasn't rebooted during the past couple weeks or had any configuration changes I can think of. If I can't reproduce the problem anymore, I don't know how I'll be able to help troubleshoot it. This is a head scratcher to be certain.

[Michael Thayer]

Works without issues here too. I will close this ticket and suggest that you talk to the Ubuntu people before re-opening it.

Regarding the warning about the 5.0.18 Additions, I put that there before releasing them as I had done major changes to the non-3D X11 components (mainly switching from a user-space driver to a kernel driver for several Linux guest distributions) and minor changes to the way our 3D parts get loaded. I can only assume that my testing was reasonably thorough, as there were indeed a few problems, but it was generally much smoother than I expected.

[raynebc]

Problem occurs again today with multiple (probably all) of my Ubuntu VMs. I am remote desktop-ing into my host computer to run VirtualBox, this quite likely has something to do with it. I find that various programs (including Firefox) are known to have a history of hardware acceleration problems when an RDP session is in use. I'll let you guys decide if you want to pursue fixing this limitation.

[Michael Thayer]

Now that I have investigated ticket #15574 I can see from the log files that this issue is a duplicate (and it is our bug, not Ubuntu's). Closing it as a duplicate and adding raynebc and cowbutt to the CC list for that ticket.

[Michael Thayer]

raynebc: ticket #15574 was fixed, but this issue is still reproducible by you with the 5.1.2 Guest Additions. Could you please attach new logs?

[Michael Thayer]

I am also not aware of anyone other than you who can reproduce this issue (I would be happy if you can find some). If you have access to any other system I would be interested to know if you can reproduce it there too.

[Michael Thayer]

Ah, I also missed comment:21 I see now. My last comment still holds.

[Michael Thayer]

Do you also have direct access to the host though so that you can see if this also occurs without RDP?

[raynebc]

I don't know any other people that use VirtualBox to run Linux. I have encountered this bug when accessing the host directly in-person, but it is not happening in that particular scenario right this moment. Could you confirm which logs you want me to collect the next time I can reproduce this problem?

[Michael Thayer]

Virtual machine log, system log (journalctl) from the guest, Xorg.0.log from the guest, dmesg from the guest, lightdm logs. I am sure that some of those are redundant, but I have a hard time remembering which logs to ask for from which Linux distributions. And finding other people who have the issue was meant more by searching for people reporting similar things on the forums, or on the internet, and contacting the people to compare.

[Michael Thayer]

If you try reducing the video RAM size, does that make a difference?

[raynebc]

Today when I RDP'd into my computer and launched VirtualBox Manager and launched my VM, it exhibited the same problem. As per your most recent suggestion, I decreased the VM's memory (from 2GB to 512MB) and started it up again, but unfortunately this didn't seem to have any effect on the problem. Decreasing the amount further to just 256MB didn't change the behavior.

I have gathered all requested logs and attached them.

[Michael Thayer]

Misunderstanding: I meant the video memory, not the standard memory.

[Michael Thayer]

You might also try the latest Guest Additions test build<1>.

<1> https://www.virtualbox.org/wiki/Testbuilds

[raynebc]

Logs gathered when running the 5.1.3-109119 guest additions

[raynebc]

Decreasing the VM's video memory from 256MB (has to be manually set in the .vbox file because the GUI doesn't allow defining it higher than 128MB) to 64MB didn't seem to have an effect on the problem. Neither did setting it to 32MB.

I installed the latest beta build of the guest additions (5.1.3-109119), verified that the "sudo VBoxService --version" command confirmed that version number and then re-enabled 3D acceleration with 64MB of VRAM. On the first boot attempt after this, the OS tried multiple times to enable video mode, but then stuck on a black screen instead of either the text screen or the splash screen. After playing with the ALT+F# keys enough times I eventually got it to a virtual terminal and then back to terminal 7 to see that the splash screen was actually stuck. After cleanly shutting down the VM and starting it again, it got stuck on the splash screen. I don't know if this difference means anything, sometimes this bug results in the OS being stuck in text mode where it shows logging, and sometimes the amount of logging displayed differs. Sometimes it also doesn't allow me to change to a virtual terminal and I have to initiate a shutdown from the VM's Machine menu. In any case, I re-gathered and attached the logs with the newer guest additions in case they reveal any more information.

[zerkms]

I have the same (?) problem here.

Details: reproducible on multiple machines (Windows 10 x64 pro and home) Guest OS: Ubuntu 16.04.1, just downloaded from the official site VirtualBox: 5.1.2

Steps to reproduce: created a virtual machine with:

• 2 CPU
• 4096Mb of RAM
• With PAE/NX enabled
• With 3d acceleration
• With 128Mb of video RAM

Install the ubuntu, with just the default settings (but with "download updates during installation" checked).

After the first boot - install the v5.1.2 Guest Addons.

Logout.

Now the ubuntu does not boot in graphical mode anymore and there are errors relevant to X.org and lightdm.

Here are the all possible logs for the problem I could gathered: ​https://www.dropbox.com/s/kr6ptg1vorjchck/virtualbox-ubuntu-1604.tgz?dl=0

[raynebc]

I couldn't find my testing has much in common with your VM settings. I'm able to reproduce the problem when the affected VM has:

• Either 2 or 1 CPU
• Various amounts of RAM
• PAE/NX disabled
• Various amounts of VRAM

The only thing that makes a fully consistent difference for me is whether 3D acceleration is disabled.

[Michael Thayer]

For people's interest, test build<1> revision 109126 and later of the Additions contains a fix for a problem which could cause random (timing-related) failures, particularly in early boot.

<1> https://www.virtualbox.org/wiki/Testbuilds

[zerkms]

Nothing has changed for me with Guest Additions development revision 109126

After the boot lightdm still cannot start. At first glance in the logs - everything is still the same.

[Michael Thayer]

zerkms's log files

[Michael Thayer]

Test build 109138 has another small fix. Probably not it, but you never know.

[raynebc]

I couldn't find the version of guest additions you mentioned was available, but build 109137 was there so I just assumed it was a typo. I removed the installed version and installed guest additions 5.1.3-109137 in my VM, but the problem still occurs as previously described. I ran "sudo VBoxService --version" and it reports the installed version as 5.1.3r109138, so I guess it's the VB download page that has the typo.

It's OK for me to use a version of guest additions newer than the version of VirtualBox itself, right? I still have version 5.1.2 r108956 of VirtualBox.

[raynebc]

Logs captured while using guest additions r109137 (service reports as being 109138)

[Michael Thayer]

zerkms: I see that Ubuntu 16.04.1 comes with 5.0.18 Additions pre-installed. Did you remove those before installing the 5.1.2 ones? I have to take a look at this myself to see how they will play together.

[zerkms]

I see that Ubuntu 16.04.1 comes with 5.0.18 Additions pre-installed

Oh, that explains a lot of things.

It's not that I really mind what exact version is installed there, but subjectively the system as-is does not have proper 3d acceleration (even keeping in mind what you just said about pre-installed addons).

Anyway, that's some more food for further thoughts, thanks.

[Michael Thayer]

You are right, they just install the non-accelerated parts. I still cannot reproduce this. I will not continue investigating this until I have a usable reproduction scenario.

[raynebc]

If you can offer beta builds of the revisions that came between the known working version (guest additions that came with Virtualbox version 5.0.16-105871) and the known problematic version (guest additions that came with Virtualbox version 5.0.18-106667), I'm willing to test and find which of them was the first to break compatibility.

This will hopefully make the developers' job easier by verifying which set of changes caused it to stop working on some computer configurations. In the meantime I'll try to reproduce on some other computers.

[Michael Thayer]

I doubt that will help, as we enabled a couple of features in single change sets which had been in the code, inactive, for a long time. Those two features are the kernel graphics driver (you can disable that again for testing purposes by deleting the kernel module vboxvideo.ko in the kernel /lib/modules tree) and loading the 3D library in a standard way. Previously we used a very unorthodox method to load it, which did not work in many cases. Now it works in many cases in which it previously failed, unfortunately also exposing bugs that were hidden before. However, these are unlikely to get fixed in the foreseeable future, see the wiki page.

[Michael Thayer]

If the problem is the kernel graphics driver, we can investigate in a lot more detail by changing (or creating) the file /var/lib/VBoxGuestAdditions/config on the guest system and changing the line "BUILD_TYPE=..." to read "BUILD_TYPE='debug'", or adding that line if it is not present. Then run /sbin/rcvboxadd setup to re-create the kernel drivers.

By the way, I noticed in the log file that your machine was set up for TFTP boot. Was that just for the initial install? Is there anything else that you might have forgotten to mention that might cause differences in behaviour?

[raynebc]

Lots after rebuilding guest additions kernel drivers for debug logging

[raynebc]

I don't know enough about Linux to be able to perform your previous suggestion regarding the vboxvideo.ko file.

I didn't manually set up any such TFTP feature, that must be how it was out of the box (either by osboxes or by default Ubuntu install settings). Since this graphics problem happens with stock, all-defaults installations of Ubuntu 15 and 16, I don't immediately believe such a configuration setting plays a substantial role.

I tried to reproduce the problem with the from-scratch Ubuntu 15 VM on two other computers, but the guest acceleration's 3D acceleration seemed to work as expected. On one of those computers, the first time I booted the VM after installing guest additions, a popup message was displayed after login that said "The VirtualBox Kernel service is not running. Exiting." When I ran the "sudo VBoxService --version" command, it reported "VBoxService: error: VbglR3Init failed with rc-=VERR_FILE_NOT_FOUND". I rebooted the VM once more and then it worked without error, and the version command reported 5.1.2r108956.

I upgraded my from-scratch Ubuntu 16 VM to the latest test build of the guest additions (5.1.3-109137), edited the guest additions config file to enable debug logging as suggested and used the command you specified to rebuild the the guest additions' kernel drivers. I replicated the guest additions bug and have attached the usual log fileset.

[Michael Thayer]

It would mean removing the file /lib/modules/<version>/misc/vboxvideo.ko. You should be able to find the details of how to do that on the net. Re-running the /sbin/rcvboxadd setup command will restore it (change the 'debug' option back to 'release' first). I can't see anything immediately interesting in your log files. So I fear I am coming to the end of my ideas until someone can find a new link.

[raynebc]

I deleted /lib/modules/4.4.0-31-generic/misc/vboxvideo.ko (the only such file I found within /lib/modules) and rebooted, but it didn't change anything. Likewise, "sudo init 5" didn't achieve a different result. Would you please describe how I would "load the 3D library in a standard way" after removing that .ko file?

[Michael Thayer]

If what you did did not change anything then that shows that the problem is not with that driver, which was one of the two main changes between 5.0.16 and 5.0.18. There is no easier way to load the 3D driver in our old, non-standard way than just reverting to the 5.0.16 Additions. However this does not work with many recent Linux distributions (which was why I changed it), so there is no question of reverting the change.

Finding out what is going wrong would need detailed analysis of what is happening inside the Ubuntu guest, which is somewhat outside of my area of expertise, and since it seems to be very rare I can't really justify spending more time on it unless something new comes to light (there are many issues to be resolved which I can reproduce and which affect more people I'm afraid). Perhaps someone at Ubuntu can help you with this. You might try asking on their IRC channel.

[Michael Thayer]

I am guessing that this is the same issue as ticket #15860, namely that when you connect to a Windows host via RDP, the host reports a very low level of OpenGL support (see "GL_RENDERER: GDI Generic" in the machine log file), which is passed through to the guest and causes confusion, because the guest tries to use a 3D desktop but does not find the capabilities it needs. This was in fact your assessment in comment:21. I will add you to the CC list for that ticket and close this one as a duplicate. We can still re-open it if that turns out to be wrong.

[ez121sl]

I experienced the same problem. One of my Ubuntu VMs is version 16.04 LTS and it was upgraded from previous versions. The graphical interface did not launch with guest additions newer than 5.0.16. Another Ubuntu 16.04 LTS VM of mine was installed directly from the Ubuntu 16.04 CD, it was not upgraded. On that VM, the graphical interface worked fine from the beginning. My host is Windows 10 x64. I have been watching this bug for any solutions.

As of a few minutes ago, my first VM is no longer experiencing the problem. I did two things, either one of them, or the combination, is probably responsible for the fix.

First, I found that the Virtual Box Graphics driver was disabled in Ubuntu proprietary driver settings. I enabled the virtualbox-guest-x11 proprietary driver. Those settings can be accessed either via the "Additional Drivers" applet or on one of the tabs under the "Software & Updates" settings. Once I rebooted, I went to "Display" settings and saw that the display was now labelled as "VBX". Before this change, the display appeared as "unknown".

Second, I installed guest addition version "5.1.18". When I rebooted, the graphical interface came up just fine.

[WBaron]

The last comment is 10 months ago. First, the VirtualBox 5.2 kernel modules would not load on Windows hosts and now X11 does not work. VBoxGuestAdditions_5.2.3-119611.iso do work for me on Linux guests, but full screen adds some invisible vertical screen estate above and below the visible range. VirtualBox-5.2 has been a complete failure until now. What is the status on that and when can we expect a usable VirtualBox-5.2 version?