Migrating a Business to Linux — 20
One of our workstations in our migration continues to be problematic. Sometimes the system goes a day without freezing the desktop and other days the system will freeze three times. The system has a discrete NVidia GT218 GeForce 8400 GS video card.
The Ubuntu Software & Updates informs users that the recommend proprietary NVidia drivers is the 340 series.
The desktop freezes are not a one-off. At home I have a test system with an Asus M3N78-EM motherboard. The motherboard has an on-board GeForce 8300 chip. I installed the NVidia 340 drivers and the desktop froze. Then worked fine for days.
Is the root cause the 340 drivers or something related to Ubuntu? Both? I don’t know.
We had a spare AMD/ATI Radeon HD 3600. The card worked fine with a live Ubuntu MATE 16.04.5 ISO and with a live Slackware ISO, but refused to boot into graphical mode from a hard drive made from our image partitions. The system booted fine into console. The
radeon kernel module was loaded. Nothing but a black screen of death rather than the lightdm login manager.
The image partitions use the 4.4.0 kernel. I updated to the 4.15 HWE kernel. Still no change. Sometimes alternate consoles were available, sometimes not. Once in a great while the lightdm login manager appeared but there would be no mouse or keyboard. Oddly the kernel magic
SysRq keys worked.
We tried the card in another system. Same result.
We tried the card in a Windows 10 system. Worked fine.
I scoured the web looking for clues. Nothing.
On a whim I compared the installed packages list of the existing and newly made image disks, I found several missing packages on the existing disk:
I installed the missing packages. The system booted to graphical mode just fine with the AMD/ATI card. All according to Hoyle.
I wasn’t yet out of the woods. I restored the NVidia card and installed the proprietary NVidia 340 package and dependencies. The system booted as expected.
I installed the AMD/ATI card and again witnessed the same black screen of death.
I rebooted to console and removed the proprietary NVidia package and dependencies.
The system booted as expected with the AMD/ATI card.
I temporarily removed the HWE kernel packages. The system booted as expected with the AMD/ATI card. Temporarily, because the HWE kernel packages are needed.
I played round-robin with this sequence. Each time the system booted with the correct
radeon module when the AMD/ATI card was installed, but refused to complete booting into graphical mode unless the NVidia packages were purged.
Possibly I could boot with special kernel boot parameters to quash the NVidia drivers but I did not test that.
Perhaps purging the NVidia packages seems intuitive to some people, but consider that the AMD/ATI card runs just fine without purging the
nouveau module. The Windows system has proprietary NVidia drivers installed yet booted fine with the AMD/ATI card. Further, most people would not think about purging packages in this manner after replacing a video card.
Why does swapping video cards cause a black screen of death? The kernel automatically detects the card and loads the correct driver module. The installed NVidia packages should make no difference.
What happened originally?
I suspect when we installed the proprietary NVidia drivers we somehow removed the missing xorg packages. As we were having problems with the system desktop freezing, we ran through several rounds of installing and removing various packages. This is one aspect I never liked about dependency checking. Dependency checking is a time-saver when installing packages but can be a real pain when removing packages.
Shooting myself in the foot does not explain why the NVidia packages must be purged.
Yet this was not the end of the story. Some days after installing the AMD/ATI card the desktop again froze. Then the card failed.
There comes a time when a person has to cut losses and move on. The workstation is now configured to boot into Windows 7.