Mint 19: Ryzen 2200G grey empty desktop after installing amdgpu-pro-18.40-676022-ubuntu-18.04.tar.xz

[MatNieuw]

Hi,

System: Mint 19, all updates.
Because I want to use OpenCL, I installed amdgpu_pro 18.40 . When I look in dmesg and journalctl all seems well. I get to the login screen very quick. If I login, I get a light grey screen and a movable cursor arrow, but nothing else. If I go to a real terminal e.g. alt-ctrl-f2, and look in the X.org log, alls seems well too.
So what causes the xserver to start properly (I think) but no desktop to appear? Before installing amdpu_pro all was fine.

Mat

20181112 Changed this title from "no desktop" to "grey empty desktop".

[deepakdeshp]

The driver is compatible only with
https://support.amd.com/en-us/kb-articl ... Notes.aspx
If your CPU isn't listed, you will have to uninstall the driver

[MatNieuw]

Your link points to a 2 year old statement when the Ryzen wasn't even released, and refers to driver 16.40 . The latest one is 18.40 .
If it didn't support the chip, why does all look OK looking in dmesg, journalctl, and the Xorg log?

Is there a log somewhere where I can see what happens after login? Is the login screen already handled by a Xserver? Or does the Xserver start after login?
I think in "recoverymode" at least it starts afterwards, and login is in vesa fb mode.

Mat

[Hoser Rob]

What makes you think xorg has loaded successfully if you get a black screen??? You need to look at that list of supported cards on that link, never mind the version or whenever the article came out.

[MatNieuw]

If you read my original port, you'll see I did not have a black screen, it switched to grey and displayed a movable mouse cursor. I believe that movement is handled by xorg. In addition, no errors were shown in the logs.

You argument of referring to a link which details the processors supported for a two year old software product is strange. Next you are going to advice me to look at support documents for Mint 17? Or Redhat 5.1?

I think the problem is not at the xorg level, but a "higher" level. I use the XFCE dekstop. If I compare file .xsession-errors when booting with vesa or amdgpu-pro, two differences are noticable.
With vesa:
(xfwm4:1403): xfwm4-WARNING **: 19:49:39.669: GLX extension missing, GLX support disabled.
(xfwm4:1403): xfwm4-WARNING **: 19:49:39.669: No vsync support in compositor

With amdgpu-pro:
(xfsettingsd:1341): xfsettingsd-WARNING **: 16:19:19.993: Failed to get the _NET_NUMBER_OF_DESKTOPS property.

So, is there a clue here? Is the _NET_NUMBER_OF_DESKTOPS property critical? Does XFCE work equally well with and without GLX?

Mat

[MatNieuw]

Some more details about this problem.

AMD link for download 18.40: https://www.amd.com/en/support/kb/relea ... -lin-18-40, choose the
Radeon™ Software for Linux® version 18.40 for Ubuntu 18.04.1

Installation instructions: https://amdgpu-install.readthedocs.io/en/latest/

Supported Vega GPU are listed as RX Vega Series Graphics and Vega Frontier Edition. Ryzen 3 2200G has Vega8, not
sure where this falls under.

Local temporary mods before the install:
As per the "How do I install AMDGPU-Pro on Mint 19?"thread, viewtopic.php?t=272074 .
/etc/os-release modified:

Code: Select all

line 3 modified from ID=linuxmint to ID=ubuntu
line 6 modified from VERSION_ID="19" to VERSION_ID="18.04"

Then install amdgpu-pro as per instructions

Revert /etc/os-release change.


On reboot, login screen appears. Not logging in but directly doing ctl-alt-f1 to get a non-x terminal and then logging in, there is no .xsession-error file
No graphics related errors in /var/log/Xorg.0.log, or dmesg, or journalctl output.

The output from ps -ef | grep -i x

Code: Select all

[ext4-rsv-conver]
[gfx]
/usr/lib/xorg/Xorg -core :0 -seat seat0 -auth /var/run/lightdm/root/:0 -nolisten tcp vt7 -novtswitch

It looks to me that xorg is running.
Going back to the login screen with alt-f7, and logging in, I get a grey screen with mouse cursor.
Going back to the terminal with ctl-alt-f1, now the output from ps -ef | grep -i x
(description from wrapper clients omitted)

Code: Select all

[ext4-rsv-conver]
[gfx]
/usr/lib/xorg/Xorg -core :0 -seat seat0 -auth /var/run/lightdm/root/:0 -nolisten tcp vt7 -novtswitch
/bin/sh /etc/xdg/xfce4/xinitrc -- /etc/X11/xinit/xserverrc
/usr/bin/ssh-agent /usr/bin/im-launch startxfce4
xfce4-session
/usr/lib/x86_64-linux-gnu/xfce4/xfconf/xfconfd
xfwm4
xfce4-panel
xfdesktop
xfsettingsd
xfce4-power-manager
xfce4-volumed
blueberry-obex-agent
/usr/lib/x86_64-linux-gnu/xfce4/notifyd/xfce4-notifyd
/usr/lib/x86_64-linux-gnu/tumbler-1/tumblerd
/usr/lib/x86_64-linux-gnu/xfce4/panel/wrapper-2.0 /usr/lib/x86_64-linux-gnu/xfce4/panel/plugins/libwhiskermenu.so 1 12582947 whiskermenu Whisker Menu
/usr/lib/bluetooth/obexd
/usr/lib/xdg-desktop-portal/xdg-desktop-portal
/usr/lib/xdg-desktop-portal/xdg-document-portal
/usr/lib/xdg-desktop-portal/xdg-permission-store
/usr/lib/x86_64-linux-gnu/xfce4/panel/wrapper-1.0 /usr/lib/x86_64-linux-gnu/xfce4/panel/plugins/libsystray.so 8 12582948 systray Notification Area 
/usr/lib/x86_64-linux-gnu/xfce4/panel/wrapper-2.0 /usr/lib/x86_64-linux-gnu/xfce4/panel/plugins/libnotification-plugin.so 9 12582949 notification-plugin
/usr/lib/xdg-desktop-portal/xdg-desktop-portal-gtk
/usr/lib/x86_64-linux-gnu/xfce4/panel/wrapper-2.0 /usr/lib/x86_64-linux-gnu/xfce4/panel/plugins/libxfce4powermanager.so 10 12582950 power-manager-plugin  
/usr/lib/x86_64-linux-gnu/xfce4/panel/wrapper-2.0 /usr/lib/x86_64-linux-gnu/xfce4/panel/plugins/libpulseaudio-plugin.so 11 12582951 pulseaudio PulseAudio Plugin

After getting the grey screen, the is a .xsession-error file .

So, it looks to me that xorg as well as Xfce is running, but I just have an empty grey screen.

Any ideas where to look?

[roblm]

Try testing this kernel parameter by editing the kernel command line. The effect will only last for one session. At the GRUB boot menu select the correct line but don’t press Enter. Press the “E” key to edit. If no menu shows, then hold down the shift key while starting. Use the Down arrow key to go down to the line that starts with:
linux /boot/vmlinuz.............ro quiet splash $vt_handoff

Go to the end of the line. Add a space and add the parameter: amdgpu.dc=0

Press F10 to start your system.

If that doesn’t help, then repeat the steps and try the parameter iommu=soft

If that doesn’t help, then upload the Xorg.0.log file to pastebin. This can be done in the virtual console tty1 by pressing Ctrl+Alt+F1. Use this command to install pastebinit: apt install pastebinit

Then upload the file: pastebinit /var/log/Xorg.0.log
Post the URL link.

You may need a BIOS or kernel update for better support. Do a Google search using these words to find other relevant Mint forum posts: site:linuxmint.com AMD Ryzen 2200G

Such as this:
viewtopic.php?f=59&t=272682

[MatNieuw]

Thanks for replying.

Bios is very latest for the Asrock B450 Pro4 motherboard, 1.50 from 2018-11-13 . URL https://www.asrock.com/MB/AMD/B450%20Pr ... .asp#BIOS . I saw no difference after upgrading from the previously used version 1.30. Kernel is 4.15.0-39 .

- Adding amdgpu.dc=0 resulted in a frozen system, black screen, kbd dead.
- Adding iommu=soft did not make a difference, same grey screen.

I also tested (after logging in so I got the grey screen) from a alt-ctl-f1 terminal:
xev -display :0.0 shows keyboard and mouse events happening on the grey screen, obviously you seen them only when returning to alt-ctl-f1.
xeyes -display :0.0 shows nothing on the grey screen. If booted in recovery mode (the vesa driver is used) and doing the same test, I get the eyes on the graphical screen.

So it seems to me xorg in itself is working properly, it just cannot get its output to the screen.

Xorg.0.log uploaded to pastebin, URL https://pastebin.com/zdhUUTVA

Thanks, Mat

[roblm]

There is an existing xorg.conf file. You can upload it to pastebin from tty1 using this command:
pastebinit /etc/X11/xorg.conf

Also use the command inxi -Gx to get the bus-ID, which will look similar to this: bus-ID: 02:00.0

[MatNieuw]

Thanks.
/etc/X11/xorg.conf at https://pastebin.com/yGzEnknx

Output of inxi -Gx at https://pastebin.com/6DVZ4DVw . Bus ID is 38:00.0
Note that this inxi was done from the alt-ctl-F1 terminal while VT7 was sitting on the login screen. The driver is shown as ati,amdgpu .

Mat

[roblm]

Use this command to remove the incorrectly configured xorg.conf file and reboot:
sudo rm /etc/X11/xorg.conf; reboot

[MatNieuw]

Hi,
removing the existing xorg.conf and rebooting did not change anything. A new xorg.conf did not appear.

I see in xorg.0.log a line:
[ 7.222] (--) AMDGPU(0): Chipset: "Unknown AMD Radeon GPU" (ChipID = 0x15dd)
I'm not sure this has any relevance.

Since testing the amdgpu-pro, I had some rare Linux crashes (kernel panic). I don't experience those on Mint 17.3 , 18,3 and 19 without amdgpu-pro installed (different partitions, same system).

Mat

[roblm]

Upload the Xorg.0.log file to pastebin.

[MatNieuw]

Uploaded as Xorg.0.log_without_xorg.conf to https://pastebin.com/k4rpzLmp

Mat

[roblm]

Use this command in tty1 to create a new xorg.conf file: sudo touch /etc/X11/xorg.conf

Open the file: sudo nano -t /etc/X11/xorg.conf

Add these lines:

Code: Select all

Section "Device"
   Identifier  "Device0"
   Driver      "amdgpu"
   BusId       "PCI:38:0:0"
EndSection

Press Ctrl+X to save and exit. Type reboot. If there is no change, then upload the Xorg.0.log file.

[MatNieuw]

Hi,
xorg.conf as proposed froze the system early on, no Xorg.0.log file was produced, and the journal was very short. I check this by booting from another Linux on the same system, e.g. with journalctl's --file option.

Looking at man xorg.conf, it states that a Device section is only active if referenced by a Screen section. So I took the original xorg.conf, and modified the Device section, so it became:

Code: Select all

Section "Device"
   Identifier  "Device0"
   Driver      "amdgpu"
   BusId       "PCI:38:0:0"
EndSection

Section "Monitor"
	Identifier	"Configured Monitor"
EndSection

Section "Screen"
	Identifier	"Default Screen"
	Monitor		"Configured Monitor"
	Device		"Device0"
EndSection

Now it booted a little further, but no active screen (very dark grey). This time there was an Xorg.0.log file of 0 bytes, but the Xorg.0.log.old file had the same timestamp and looked useful, see https://pastebin.com/2VnwAJsF .

It looks like that with the original xorg.conf file it got further, looking at the (EE) entries.
The content of the gpu-manager.log file in /var/log might be of interest:

Code: Select all

log_file: /var/log/gpu-manager.log
last_boot_file: /var/lib/ubuntu-drivers-common/last_gfx_boot
new_boot_file: /var/lib/ubuntu-drivers-common/last_gfx_boot
can't access /run/u-d-c-nvidia-was-loaded file
can't access /opt/amdgpu-pro/bin/amdgpu-pro-px
Looking for nvidia modules in /lib/modules/4.15.0-39-generic/updates/dkms
Looking for amdgpu modules in /lib/modules/4.15.0-39-generic/updates/dkms
Is nvidia loaded? no
Was nvidia unloaded? no
Is nvidia blacklisted? no
Is intel loaded? no
Is radeon loaded? no
Is radeon blacklisted? yes
Is amdgpu loaded? yes
Is amdgpu blacklisted? no
Is amdgpu versioned? yes
Is amdgpu pro stack? no
Is nouveau loaded? no
Is nouveau blacklisted? no
Is nvidia kernel module available? no
Is amdgpu kernel module available? no
Vendor/Device Id: 1002:15dd
BusID "PCI:56@0:0:0"
Is boot vga? yes
Found "/dev/dri/card0", driven by "amdgpu"
output 0:
	card0-DP-2
Number of connected outputs for /dev/dri/card0: 1
Skipping "/dev/dri/card0", driven by "amdgpu"
Skipping "/dev/dri/card0", driven by "amdgpu"
Skipping "/dev/dri/card0", driven by "amdgpu"
Does it require offloading? no
last cards number = 1
Has amd? yes
Has intel? no
Has nvidia? no
How many cards? 1
Has the system changed? No
Single card detected
Nothing to do

Interesting is also that both the Xorg.0.log and gpu-manager.log refer to the device ID in what looked to be the decimal version of x38 . But it correctly identifies the DeviceID, so no problem.

Mat

[roblm]

Looking at man xorg.conf, it states that a Device section is only active if referenced by a Screen section. So I took the original xorg.conf, and modified the Device section, so it became:

I have used only a Device section without problems many times. If a Screen section is not listed, then Xorg will automatically create virtual Screen and Monitor sections to configure the display but won’t physically add them to the file.

That’s what these lines in the second Xorg.0.log file you uploaded show:

Code: Select all

[     6.238] (==) No Layout section.  Using the first Screen section.
[     6.238] (==) No screen section available. Using defaults.
[     6.238] (**) |-->Screen "Default Screen Section" (0)
[     6.238] (**) |   |-->Monitor "<default monitor>"
[     6.239] (==) No monitor specified for screen "Default Screen Section".
    Using a default monitor configuration.

I noticed that the BusID from the gpu-manager.log is "PCI:56:0:0", the same as listed in the Xorg.0.log.
When I copied the output you uploaded to pastebin from using the inxi -Gx command and saved it in a text editor to a file named inxi and then opened it with the command cat inxi, which removes the extra characters, then this is the output:

Code: Select all

Graphics:  Card: Advanced Micro Devices [AMD/ATI] Vega [Radeon Vega 8 Mobile] bus-ID: 38:00.0
           Display Server: X.org 1.19.6 drivers: ati,amdgpu (unloaded: modesetting,fbdev,radeon,vesa)
           tty size: 200x75 Advanced Data: N/A out of X

The BusId is "PCI:38:0:0". They should be the same.

Try changing the line in the xorg.conf file to: BusId "PCI:56:0:0"

[MatNieuw]

Hi,

1. It appears that the xorg.conf file is not needed, if it is missing it find the proper drivers etc.
2. The PCI ID is most definitely 38:0:0. The number 56 in xorg.0.log and gpu-manager.log is the decimal version. They should do it in hex to IMHO. Xorg.0.log properly reports the vendor/device IDs.

In general the variations I tried (different partitions, same hardware) are:
Mint 19, the GUI didn't work straight out of the box for me, had to use recovery mode. The GUI worked once the kernel was upgraded to 4.15.0-39 , then it uses the ati,amdgpu driver. But it seems slow (local HD MP4 movies not smooth), va_info shows an error. And keyboard is slow, while cpu load 0.

If I upgrade the driver to amdgpu-pro 18.40 it doesn't work, just a grey screen and cursor. Recovery mode works. And my Lenovo mouse keeps disconnecting/connecting. From an alt-ctl-1 terminal, vainfo shows multiple entry points, clinfo also seems OK.

Mint 18.3, GUI works after upgrading to 4.15.0-39 (ati driver), before that either split screen or use recovery mode (vesa driver).

Mint 17.3, GUI works straight after install using the vesa driver, kernel 4.4.0-139.

AV-Linux, GUI works straight after install using the vesa driver, kernel 4.16.12-rt5-avl1 .

I am contemplating using a later kernel for the system with amdgpu-pro.

Mat

[roblm]

1. It appears that the xorg.conf file is not needed, if it is missing it find the proper drivers etc.

It’s not needed most of the time. The only reason for using it was to rule out a system driver misconfiguration, which happened in the first xorg.conf file, where the vesa driver was listed.

2. The PCI ID is most definitely 38:0:0. The number 56 in xorg.0.log and gpu-manager.log is the decimal version. They should do it in hex to IMHO. Xorg.0.log properly reports the vendor/device IDs.

Use this command to verify what the BusId is: lspci | egrep -i "3D|VGA"
It should be the same as the inxi output.

Mint 19, the GUI didn't work straight out of the box for me, had to use recovery mode. The GUI worked once the kernel was upgraded to 4.15.0-39 , then it uses the ati,amdgpu driver. But it seems slow (local HD MP4 movies not smooth), va_info shows an error.

For VA-API and VDPAU video support for the radeon and amdgpu drivers, install these packages:
mesa-va-driver and mesa-vdpau-driver

I am contemplating using a later kernel for the system with amdgpu-pro.

Here is another option if nothing else works. Install the amdgpu-pro driver with the --headless option, which installs the OpenCL libraries but not the AMD-Pro OpenGL libraries. The Mesa OpenGL libraries are used:
./amdgpu-pro-install --opencl=pal,legacy --headless

This is to rule out a conflict with AMD’s OpenGL libraries.
Also, the AMD-Pro Vulcan component is not intalled, which may be desirable for gaming.

[MatNieuw]

Hi,

Indeed lspci shows also 38:0:0
38:00.0 VGA compatible controller: Advanced Micro Devices, Inc. [AMD/ATI] Vega [Radeon Vega 8 Mobile] (rev c8)

Thanks for the tips on the mesa-va-drivers and mesa-vdpau-drivers. On Mint 19 without amdgpu-pro they indeed make vainfo give entry points.
Because on my Mint 19 with amdgpu-pro drivers vainfo already shows entry points, I have not installed it there. Would it conflict it I did?

The --headless install option did not change anything, all logs look fine too. The --opencl=pal,legacy I already did before. Just curious: is there any way of uninstalling amdgpu-pro? Do the opencl pal and legacy options bite each other?

Vulkan is not needed by me.

It really seems a failure to communicate between Xorg and amdgpu-pro . Can a kernel have influence on this?

Thanks, Mat