Guest Additions do not work with ArchLinux -> fixed in releases 5.0.22 and higher

[GonzO]

Title says it all: A Linux guest (on an OS X Host) that had 3D accel functioning just fine on 5.16 now doesn't have accel on 5.20 (5.18 was the original breaking release, and 5.20 didn't fix the regression). GLXInfo says I should be able to have it, but for some reason I'm still on llvmpipe.

GLXInfo data:

[gonzo@ArchVM ~]$ glxinfo | grep -i render direct rendering: Yes

GLX_MESA_multithread_makecurrent, GLX_MESA_query_renderer, GLX_MESA_multithread_makecurrent, GLX_MESA_query_renderer,

Extended renderer info (GLX_MESA_query_renderer): OpenGL renderer string: Gallium 0.4 on llvmpipe (LLVM 3.7, 256 bits)

GL_ARB_conditional_render_inverted, GL_ARB_conservative_depth, GL_NV_conditional_render, GL_NV_depth_clamp, GL_NV_packed_depth_stencil, GL_ARB_conditional_render_inverted, GL_ARB_conservative_depth, GL_NV_blend_square, GL_NV_conditional_render, GL_NV_depth_clamp, GL_OES_element_index_uint, GL_OES_fbo_render_mipmap,

Xorg log:

[gonzo@ArchVM ~]$ cat /var/log/Xorg.0.log | grep "(EE)"

(WW) warning, (EE) error, (NI) not implemented, (??) unknown.

[ 5.567] (EE) modeset(0): glamor initialization failed [ 5.633] (EE) AIGLX: reverting to software rendering

*

Guest additions are updated as well.

[GonzO]

The VBox Log file for this VM

[Michael Thayer]

If this is still relevant with the current test builds, do you have the file /etc/ld.so.conf.d/00vboxvideo.conf on your guest, and the libraries in /tmp/VBoxOGL?

[johndoe0]

dmesg

[johndoe0]

This seems to be an issue with the latest VirtualBox from the official repository[1].

On host (Fedora 23):

$ rpm -q VirtualBox-5.0 
VirtualBox-5.0-5.0.20_106931_fedora22-1.x86_64
$ dmesg|grep -i vbox
[Jun 9 22:48] vboxdrv: Found 4 processor cores
[  +0.015887] vboxdrv: TSC mode is Invariant, tentative frequency 3392279880 Hz
[  +0.000003] vboxdrv: Successfully loaded version 5.0.20 (interface 0x00240000)
[  +0.205578] VBoxNetFlt: Successfully started.
[  +0.001756] VBoxNetAdp: Successfully started.
[  +0.002375] VBoxPciLinuxInit
[  +0.000006] vboxpci: IOMMU found
[  +0.397869] vboxdrv: ffffffffa0855020 VMMR0.r0
[  +0.168000] vboxdrv: ffffffffa0950020 VBoxDDR0.r0
[  +0.001047] vboxdrv: ffffffffa0006020 VBoxDD2R0.r0
[  +0.038213] vboxpci: created IOMMU domain ffff880399d5ff78

On guest (Xubuntu Xenial):

# dpkg -l virtualbox\*|grep \^ii
ii  virtualbox-guest-dkms    5.0.18-dfsg-2build1 all          x86 virtualization solution - guest addition module source for dkms
ii  virtualbox-guest-utils   5.0.18-dfsg-2build1 amd64        x86 virtualization solution - non-X11 guest utilities
ii  virtualbox-guest-x11     5.0.18-dfsg-2build1 amd64        x86 virtualization solution - X11 guest utilities
# dmesg|egrep -i '(virtualbox|vbox)'
[    0.000000] DMI: innotek GmbH VirtualBox/VirtualBox, BIOS VirtualBox 12/01/2006
[    0.000000] ACPI: RSDP 0x00000000000E0000 000024 (v02 VBOX  )
[    0.000000] ACPI: XSDT 0x000000005FFF0030 00003C (v01 VBOX   VBOXXSDT 00000001 ASL  00000061)
[    0.000000] ACPI: FACP 0x000000005FFF00F0 0000F4 (v04 VBOX   VBOXFACP 00000001 ASL  00000061)
[    0.000000] ACPI: DSDT 0x000000005FFF0470 002106 (v01 VBOX   VBOXBIOS 00000002 INTL 20100528)
[    0.000000] ACPI: APIC 0x000000005FFF0240 000054 (v02 VBOX   VBOXAPIC 00000001 ASL  00000061)
[    0.000000] ACPI: SSDT 0x000000005FFF02A0 0001CC (v01 VBOX   VBOXCPUT 00000002 INTL 20100528)
[    0.770019] ata2.00: ATAPI: VBOX CD-ROM, 1.0, max UDMA/133
[    0.771075] scsi 1:0:0:0: CD-ROM            VBOX     CD-ROM           1.0  PQ: 0 ANSI: 5
[    1.354230] usb 1-1: Manufacturer: VirtualBox
[    1.378962] input: VirtualBox USB Tablet as /devices/pci0000:00/0000:00:06.0/usb1/1-1/1-1:1.0/0003:80EE:0021.0001/input/input6
[    1.379372] hid-generic 0003:80EE:0021.0001: input,hidraw0: USB HID v1.10 Mouse [VirtualBox USB Tablet] on usb-0000:00:06.0-1/input0
[    1.540641] ata3.00: ATA-6: VBOX HARDDISK, 1.0, max UDMA/133
[    1.541164] scsi 2:0:0:0: Direct-Access     ATA      VBOX HARDDISK    1.0  PQ: 0 ANSI: 5
[    4.823954] vboxguest: misc device minor 55, IRQ 20, I/O port d020, MMIO at 00000000f0400000 (size 0x400000)
[    4.823957] vboxguest: Successfully loaded version 5.0.18_Ubuntu (interface 0x00010004)
[    4.894107] fbcon: vboxdrmfb (fb0) is primary device
[    4.944719] vboxvideo 0000:00:02.0: fb0: vboxdrmfb frame buffer device
[    4.969704] [drm] Initialized vboxvideo 1.0.0 20130823 for 0000:00:02.0 on minor 0
[    5.409758] vboxsf: Successfully loaded version 5.0.18_Ubuntu (interface 0x00010004)
[    5.417615] VBoxService 5.0.18_Ubuntu r106667 (verbosity: 0) linux.amd64 (Apr 20 2016 09:32:17) release log
[    5.418593] 00:00:00.001064 main     Executable: /usr/sbin/VBoxService

VirtualBox log from host, and dmesg and Xorg.0.log from guest will be attached soon.

[1] ​http://download.virtualbox.org/virtualbox/rpm/fedora/virtualbox.repo

[johndoe0]

I tried it with guest additions from Xubuntu Yakkety repository to match VirtualBox version:

# dpkg -l virtualbox\*|grep \^ii
ii  virtualbox-guest-dkms    5.0.20-dfsg-2 all          x86 virtualization solution - guest addition module source for dkms
ii  virtualbox-guest-utils   5.0.20-dfsg-2 amd64        x86 virtualization solution - non-X11 guest utilities
ii  virtualbox-guest-x11     5.0.20-dfsg-2 amd64        x86 virtualization solution - X11 guest utilities

The result seems to be same:

$ less /var/log/Xorg.0.log
[...SNIP...]
[     5.598] (II) Loading sub module "glamoregl"
[     5.598] (II) LoadModule: "glamoregl"
[     5.598] (II) Loading /usr/lib/xorg/modules/libglamoregl.so
[     5.609] (II) Module glamoregl: vendor="X.Org Foundation"
[     5.609]    compiled for 1.18.3, module version = 1.0.0
[     5.609]    ABI class: X.Org ANSI C Emulation, version 0.4
[     5.609] (II) glamor: OpenGL accelerated X.org driver based.
[     5.823] (II) glamor: EGL version 1.4 (DRI2):
[     5.823] EGL_MESA_drm_image required.
[     5.824] (EE) modeset(0): glamor initialization failed
[...SNIP...]
[     5.845] (II) AIGLX: Screen 0 is not DRI2 capable
[     5.845] (EE) AIGLX: reverting to software rendering
[     5.882] (II) AIGLX: enabled GLX_MESA_copy_sub_buffer
[     5.883] (II) AIGLX: Loaded and initialized swrast
[     5.883] (II) GLX: Initialized DRISWRAST GL provider for screen 0
[...SNIP...]

Please poke me if you'd like to see more information from this test as well.

[johndoe0]

dmesg

[johndoe0]

I've just tried it with VirtualBox guest additions 5.0.20_106931 from the official CD-ROM, on the same Xubuntu Xenial. The result is unfortunately same.

On guest:

# lsmod|grep ^vbox
vboxsf                 49152  0
vboxvideo              49152  3
vboxguest             286720  6 vboxsf,vboxvideo
# modinfo vbox{video,guest}
filename:       /lib/modules/4.4.0-22-generic/updates/dkms/vboxvideo.ko
version:        5.0.20 r106931
license:        GPL and additional rights
description:    Oracle VM VirtualBox Graphics Card
author:         Oracle Corporation
srcversion:     8513E22BC5DFD1C4C3631FA
alias:          pci:v000080EEd0000BEEFsv*sd*bc*sc*i*
depends:        drm,drm_kms_helper,ttm,vboxguest,syscopyarea,sysfillrect,sysimgblt
vermagic:       4.4.0-22-generic SMP mod_unload modversions 
parm:           modeset:Disable/Enable modesetting (int)
filename:       /lib/modules/4.4.0-22-generic/updates/dkms/vboxguest.ko
version:        5.0.20 r106931
license:        GPL
description:    Oracle VM VirtualBox Guest Additions for Linux Module
author:         Oracle Corporation
srcversion:     4552320B5D61597A101E58F
alias:          pci:v000080EEd0000CAFEsv00000000sd00000000bc*sc*i*
depends:        
vermagic:       4.4.0-22-generic SMP mod_unload modversions 

[Kevin Brodsky]

I encounter exactly the same problem with VirtualBox 5.0.22, on an Arch Linux host with an Apricity guest (Arch-based with Cinnamon). Even though everything seems to be configured properly (glxinfo looks fine, etc.), I see the same error in Xorg.0.log as johndoe0.

[GonzO]

5.0.22 still not working.

"If this is still relevant with the current test builds, do you have the file /etc/ld.so.conf.d/00vboxvideo.conf on your guest, and the libraries in /tmp/VBoxOGL?"

I do not have that file in /etc/ld.so.conf.d/, nor do I even have a /tmp/VBoxOGL directory on the guest. I'm not sure I ever did on any of the older builds that worked, but I know I don't have that on the guest now.

[Michael Thayer]

johndoe0: I am not even quite sure what issue you are having, but it does not seem to be the same as the others, and it is with Ubuntu's packages, so you should open a Launchpad bug report for it.

To the others: if I understand right our Installer and Additions scripts do not seem to be working quite right with ArchLinux. Would one of you have time to investigate why not? Executing "/sbin/rcvboxadd setup" is roughly similar to re-running the installer and a good place to start looking.

[Kevin Brodsky]

Replying to michael:

johndoe0: I am not even quite sure what issue you are having, but it does not seem to be the same as the others, and it is with Ubuntu's packages, so you should open a Launchpad bug report for it.

To the others: if I understand right our Installer and Additions scripts do not seem to be working quite right with ArchLinux. Would one of you have time to investigate why not? Executing "/sbin/rcvboxadd setup" is roughly similar to re-running the installer and a good place to start looking.

I've tried with another VM, essentially Fedora with Cinnamon. I see exactly the same things in Xorg.0.log as previously mentioned, however Cinnamon does not display the notification about running without hardware acceleration and I seem to be getting some transparency... Not sure what's going on.

[Michael Thayer]

Regarding the warning in Xorg.0.log: we support 3D acceleration in X11 clients (applications), but not AIGLX inside the X server itself, so the X server is expected to revert to software rendering. Applications should still be accelerated. Did that make sense?

[GonzO]

Replying to michael:

Ubuntu guests are also broken. It isn't just Arch guests. And it's worse in Ubuntu, because instead of turning all the animations off (which is what Gnome does), Unity keeps them all but running v-e-r-y slowly in software.

I also know that it's the guest additions that are broken. I upgraded from 5.0.16 to 5.0.18 with ext pack, and things were still good, but as soon as I upgraded the guest additions to 5.0.18, GL-enhanced desktops stopped working.

I can confirm that the AIGLX message occurs regardless of whether or not 3D acceleration is working.

[GonzO]

Replying to michael:

THIS JUST IN:

I upgraded VirtualBox and its Extension pack to 5.0.22, but downgraded the guest additions in both Arch and Ubuntu to 5.0.16 -- poof, 3D acceleration returned in Unity and Gnome. This experiment can be very easily replicated, but easier in Ubuntu (the 5.0.16 ISO is available and actually compiles there, whereas in Arch I reverted to the 4.4 LTS kernel and then had to dig up a 5.0.16 guest utils package from the archive).

I would request that the "ArchLinux" renaming of the ticket be reverted and this be filed as a guest additions issue.

[Michael Thayer]

I just tried this, installing official 5.0.22 Additions into a 64-bit Ubuntu 16.04 VM running in a 5.1 beta 4-ish VirtualBox on an Ubuntu 16.04 host. 3D pass-through is working. Please, if you are interested in getting this fixed, provide me with precise instructions on how to reproduce this with a fresh virtual machine, including all the steps you take to create the machine, so that I can find out what the critical difference between your set-up and mine is.

[GonzO]

Are you sure it _is_ working, though? The Unity animations play regardless, just slower and choppier if in the software renderer. A better bet would be to use the GNOME desktop, as it turns off animations/fades when it figures out it only has a software renderer to work with.

My exact steps to reproduce, however, are as follows:

1) Be running OS X El Capitan. I'm on a MacBook Pro if that helps.

2) Install VB 5.0.22 and the 5.0.22 Extension pack.

3) Create a VM and install (say) Ubuntu on it. Ubuntu: Gnome Edition would be the best way to see the differences.

4) In the Guest: Install the corresponding virtualbox-guest-utils pack, or use the one that comes with Ubuntu (which is 5.0.18). Neither will give you proper 3D, so *shrug*. Reboot and notice how long it takes the Dashboard (if in Unity) to fade in, or how Gnome has no animations or fades whatsoever (if you're using that environment).

5) In the Guest: Grab the 5.0.16 guest utilities iso, install the guest additions from that, reboot. Notice that Gnome now has desktop animations active, and Unity runs at full speed and with a very noticeable smoothness increase.

Other than that, I don't really know what else I can provide that I haven't already. If there's a particular log you need, I can fetch it. I can even stream a movie of how it looks with newer guest-utils as opposed to older, and I can input that (but I'm not sure how useful it'll be).

But I do believe that the evidence strongly points to something breaking in the virtualbox-guest-utils between 5.0.16 and 5.0.18 on any Linux guest I throw at it (not just Arch).

I have NOT tried 5.0.22 guest-utils with VBox 5.1 beta as a host. The only Mac I have is a work laptop and I use Vagrant on the daily, so I don't think I can use this environment to test that.

[Kevin Brodsky]

Replying to GonzO:

I upgraded VirtualBox and its Extension pack to 5.0.22, but downgraded the guest additions in both Arch and Ubuntu to 5.0.16 -- poof, 3D acceleration returned in Unity and Gnome. This experiment can be very easily replicated, but easier in Ubuntu (the 5.0.16 ISO is available and actually compiles there, whereas in Arch I reverted to the 4.4 LTS kernel and then had to dig up a 5.0.16 guest utils package from the archive).

I would request that the "ArchLinux" renaming of the ticket be reverted and this be filed as a guest additions issue.

I confirm this! I just tried it in my Arch guest. I don't install the guest additions from the VirtualBox iso, I use the Arch packages (​virtualbox-guest-utils and ​virtualbox-guest-modules-arch). Reverting virtualbox-guest-utils (libraries and binaries) to 5.16.x made HW acceleration work again (Cinnamon doesn't complain anymore), while 5.18 is also broken. Note that I did not revert the kernel modules (virtualbox-guest-modules-arch): I stayed with the 5.22 version.

FWIW, the virtualbox-guest-utils package contains these files:

/etc/xdg/autostart/vboxclient.desktop
/usr/bin/VBoxClient
/usr/bin/VBoxClient-all
/usr/bin/VBoxControl
/usr/bin/VBoxService
/usr/bin/mount.vboxsf
/usr/lib/VBoxOGL.so
/usr/lib/VBoxOGLarrayspu.so
/usr/lib/VBoxOGLcrutil.so
/usr/lib/VBoxOGLerrorspu.so
/usr/lib/VBoxOGLfeedbackspu.so
/usr/lib/VBoxOGLpackspu.so
/usr/lib/VBoxOGLpassthroughspu.so
/usr/lib/security/pam_vbox.so
/usr/lib/systemd/system/vboxservice.service
/usr/lib/sysusers.d/virtualbox-guest-utils.conf
/usr/lib/udev/rules.d/60-vboxguest.rules
/usr/lib/xorg/modules/dri/vboxvideo_dri.so
/usr/lib/xorg/modules/drivers/vboxvideo.so
/usr/share/licenses/virtualbox-guest-utils/LICENSE

[Michael Thayer]

Did I understand correctly - and if I did, I definitely should have asked earlier: are you using the Guest Additions provided by your distribution, not by us? If that is the case, then you should be filing a bug report with them. Of course, it may be (though I think it is less likely) that it is still our "fault", but in that case I would also rather be working directly with the distribution packagers to solve the issue.

[GonzO]

Replying to michael:

Did I understand correctly - and if I did, I definitely should have asked earlier: are you using the Guest Additions provided by your distribution, not by us?

Yes and No.

With Arch (and in Corax's case), the answer is "yes" because with Arch you have no choice. Y'all's ISO -- in particular the installer -- has no compile branch that recognizes Arch as a legitimate distribution. If we examine the PKGBUILD file at ​https://git.archlinux.org/svntogit/community.git/tree/trunk/PKGBUILD?h=packages/virtualbox, line 52 makes it clear that they build their package from the tar.bz file grabbed from your server at ​http://download.virtualbox.org/virtualbox/5.0.22/VirtualBox-5.0.22.tar.bz2 and then patch the installer to get it recognized.

However, in my case and with Ubuntu: the answer is "No". I used y'all's ISO to confirm the effect, and I did confirm it in two desktop environments.

The problem is almost definitely a guest-utils code change that occurred between the 5.0.16 and 5.0.18 guest-utils ISO, which affects OSX hosts, and which has been carried forward ever since to both subsequent guest-utils versions and the packages that are generated from it.

I stopped using the Ubuntu packages specifically to eliminate the possibility of there being a packaging problem (as Ubuntu has it's own packages as well), and the issue remains.

[Michael Thayer]

I wonder whether that is still the case with the 5.1 beta. I tried to remove as much as possible of that distribution-specific script and replace it with script which is less magic but more general and hopefully reliable.

[GonzO]

I can try to test that (albeit on a Linux or Windows host) and get back to you, if you'd like.

[tico-tico]

The official vb additions can be installed in Manjaro without problems (only common ones like broken Chrome), I guess with Arch it should be the same. Kernel 4.4, systemd 229 here.

/etc/ld.so.conf.d/00vboxvideo.conf points to /var/lib/VBoxGuestAdditions/lib and there are two symlinks libEGL.so and libGL.so. They points to /opt/VboxGuestAdditions....

Edit: I mean sudo ./VBoxLinuxAdditions.run just works.

[GonzO]

Replying to michael:

I wonder whether that is still the case with the 5.1 beta. I tried to remove as much as possible of that distribution-specific script and replace it with script which is less magic but more general and hopefully reliable.

Michael, you're a _genius_. I said "hey, they're just guest additions" and downloaded the 5.1 beta 3 iso on my guests.

It builds in Arch. It also works in Arch. It also builds on Ubuntu and Ubuntu Gnome Edition, and works there as well.

Is there a way to mark a bug "fixed in the future"? Because it seems this might be the case.

[GonzO]

OK, wow. Learned some new things:

The 5.0.22 guest utils (from the ISO) also compile and install in modern Arch. They didn't at 5.0.18 time. They also work, with the exception of hardware acceleration in Chrome blowing up the guest GUI entirely (I'm assuming this is what tico-tico is referring to). So I have to eat crow there.

I guess we should just mark this one closed, file a ticket with Arch packagers for those who want to use the packages, and just stick to the install ISO in all cases. My stuff is running and I have nothing further to complain about, except the Chrome thing (which sounds like a known issue with a simple workaround).

[Michael Thayer]

Thanks for your testing! The Chrome problem has its own ticket (#15417). Other than that, I will re-label the bug matching your suggestion. I thought we already had a bug for the problem with the Additions not working in ArchLinux, but I can't find it now, so I will re-label this one similarly to your suggestion. We can re-mark this as a duplicate if we do find it.

[Kevin Brodsky]

Replying to GonzO:

OK, wow. Learned some new things:

The 5.0.22 guest utils (from the ISO) also compile and install in modern Arch. They didn't at 5.0.18 time. They also work, with the exception of hardware acceleration in Chrome blowing up the guest GUI entirely (I'm assuming this is what tico-tico is referring to). So I have to eat crow there.

I guess we should just mark this one closed, file a ticket with Arch packagers for those who want to use the packages, and just stick to the install ISO in all cases. My stuff is running and I have nothing further to complain about, except the Chrome thing (which sounds like a known issue with a simple workaround).

Again I confirm - the guest utils from the 5.0.22 ISO work on Arch. Now there's something I still don't understand: you said that it didn't work when installing from the ISO on Ubuntu. What's the the matter with that? Do you need to use 5.1 Beta in this case?

[Frank Mehnert]

Fix is part of 5.0.22 so let's close this ticket.