AMD Ryzen: Problems and Fixes 🔧

klorax · 29 June 2020 22:11

AMD Ryzen: Problems and Fixes

Many of the problems concerning AMD Ryzen CPUs and APUs seems to have solutions/workarounds now. Here are some of the most disabling problems with possible fixes. As a newbie I had to do quite a bit of research to solve these for my self, hopefully it can be of use for others as well!

Installing

Depending on the age of the CPU or APU an installation media running a kernel (the live session) that supports the CPU (and GPU) is needed. This has not always been the case, especially for the APUs (which needs drivers for the new GPUs for the live GUI). A solution that has usually worked is to use Manjaro Architect. Please see this tutorial.

Manjaro Architect has not worked for everyone though, and you might have to add one of these boot options;

"Pass through" should work dynamically:

iommu=pt

But if you still have issues then use software-only instead with:

iommu=soft

(Soft) Lockups

There are many different types of lockups that has plagued the now many different AMD Ryzen CPUs and APUs. Many probably have had more than one type and therefore several solutions have been needed. A common error message in the journalctl (for different types of CPU/APU and problems it seems) have looked like:

NMI watchdog: BUG: soft lockup - CPU#12 stuck for 23s! [DOM Worker:1364]

First, make sure that:

Update the kernel if possible (latest stable should be fine at the time of writing).
Update UEFI (BIOS) if possible.

Most of these soft lockups seem to be solved by a combination of the following three solutions (or workarounds):

In UEFI (BIOS) set Power Supply Idle Control (or something similar) to Typical Current Idle (or something similar) .
Add the kernel boot parameter: idle=nomwait.
Disable C6 states (both core and package) with Zenstates.py.

Kernel Boot Parameter

Adding a kernel boot parameter for GRUB in Manjaro:

Open the file /etc/default/grub in a text editor.
Add the parameter, in this case idle=nomwait (notice the m), to the line starting with GRUB_CMDLINE_LINUX_DEFAULT=, at the end, within the quotes, like so: GRUB_CMDLINE_LINUX_DEFAULT="<current parameters ...> idle=nomwait".
Save the file.
Update GRUB: sudo update-grub.
Upon next reboot the default boot option should include the parameter.

Zenstates.py

Disabling C6 states in Manjaro (modified and extended this):

Install zenstates-git from AUR (for example with Pamac, Octopi, or bmenu).
The kernel module msr is needed for Zenstates. To load this module permanently (every boot): add msr to the file /etc/modules-load.d/modules.conf.
To disable C6 states permanently (every boot), we need to create a system.d unit. To do this create a file in /etc/systemd/system called something like disable_c6.service.
Add the following contents to the file:

[Unit]
Description=Ryzen Disable C6
DefaultDependencies=no
After=sysinit.target local-fs.target
Before=basic.target

[Service]
Type=oneshot
ExecStart=/usr/bin/zenstates --c6-disable

[Install]
WantedBy=basic.target

Run sudo systemctl enable disable_c6.service to enable the system.d unit.
At the next reboot msr should be loaded, to check this run: lsmod | grep msr which should return a line starting with msr.
At the next reboot zenstates should also have run, and thus C6 states disabled. To check this run sudo zenstates -l and it should return something ending with:

...
C6 State - Package - Disabled
C6 State - Core - Disabled

whm1974 · 19 August 2018 15:06

This should be helpful when I get around to building a Ryzen desktop...

klorax · 3 September 2018 15:02

Appendix

Personally I have built two very similar Ryzen desktops and have one Ryzen Laptop (Lenovo Ideapad 720s-ARR with AMD Ryzen 2500U with Vega 8).

One of the desktop is a Ryzen 1600 and have worked "flawlessly" since day one.

The other desktop, a Ryzen 1200, with similar components: same (supposedly supported) Corsair memory (but only 8 GB instead of 16 GB), similar ASUS motherboard (only mini ATX instead of ATX), had lockups. I was fortunate enough to guess correctly that the memory made it unstable, a switch at the store fixed that.

My laptop have had problems since day one. Most of the post above is related to fixing it. I have looked around everywhere in this forum (1, 2, 3, for example), all over the Internet, and also talked with @Strit here, who has a similar laptop.

Unfortunately the UEFI option is not possible on this laptop. Disabling C6 states seemed to help somewhat. However the kernel parameter idle=nomwait which I have not found anywhere in this forum, seems to have done the trick (couple of days running fine, hopefully I have not jinxed it! )! Please try it out!

Update

2018-09-03: Still no lockups, so idle=nomwait seems to be working (for me)!