AMD Ryzen on Linux

[brmather]

I am able to install Fedora 25 without any problems. Which leads me to believe that it is an Ubuntu problem, not the kernel.

The motherboard displays error code 56 - "invalid CPU type or speed" - so perhaps it is something in the way Ubuntu recognises the CPU.

I've noticed that acpi=off works but acpi=ht doesn't, which indicates the error probably lies in the acpi table.

[fiedzia]

I've got some progress on kernel 4.11rc1 when I'll add "quiet loglevel=3" to boot params. This allows me to boot into recovery mode, and kernel finally notices 8 cores. Dmesg shows gigabytes of "handle_irq <some hex numbers>" messages. Normal boot still fails. I'll try fedora.

[fiedzia]

Still not working, reported here: https://bugs.launchpad.net/ubuntu/+s...x/+bug/1671360

[sebastian.riedel]

Same mainboard and issues as fiedzia here (Gigabyte GA-AB350 Gaming 3, Ryzen 1700), newest F5 bios also doesn't work for me. Gigabyte support responded with:

We have ask our colleagues in the laboratory and get the following answer.

Please suggest user to consult with Linux vendor for installation issue, we don’t guarantee third party O.S. performance due to it’s an open source.

As long as Windows O.S. works fine, it’s not our issue.

Kind regards
GIGABYTE-Team Germany

I pointed them at fiedzia's launchpad report, maybe they can make use of the logs (although it doesn't seem they care much).

[puleglot]

... due to it’s an open source ...

Good reason!

[fiedzia]

amd_gpio_irq_handler

Try disable CONFIG_GPIOLIB and all other pin controllers. You can try my custom kernel where is many features disabled. how does it boot?. Copy .config file from deb package for editing. .http://s000.tinyupload.com/?file_id=...40821255277866

Will do, thanks for suggestion. Do you know what will I lose by disabling that?

[fiedzia]

I've had to hunt several places related to amd and gpio (there were at least three different settings for that in various places), but its better now.
The problem with messages spamming kernel is gone. System boots normally. Nvidia drivers do not work with 4.11, and nouveau doesn't work either,
but I'm booting into text mode without other issues. Audio works, though even switching terminals causes it to stop for a second. I may have more luck with going back to 4.10 and disabling same things, that should at least give me nvidia drivers.

[fiedzia]

Everything seems to work fine on 4.10 as well as 4.11rc1. Problems with sound turned out to have more to do with my mplayer config than with hardware, so that works fine too, though I probably need to read more about how sound works in recent Linux to be sure.

Brief instruction for anyone in the same situation:

1. Find instruction how to compile kernel for your distribution
2. When configuring kernel (I used make menuconfig) disable following things:

Everything under device drivers/gpio support, especially:
memory mapped files/amd promontory support (GPIO_AMDPT)
and pci gpio expanders / amd 8111 gpio driver (GPIO_AMD8111)
but I've turned of everything to make sure nothing is using gpio.

device drivers / pin controllers / amd gpio pin control (PINCTRL_AMD)

Perhaps only one of them could be really necessary, but I didn't tested that.

System boots normally, no problems with journald and no weird messages from dmesg.

[protomucca]

Hello, got the same problem using Kernel version 4.8 and 4.10, the system seems to work fine with Linux Kernel 4.4.0-64-generic (I'm on Kubuntu 16.10).

[mossman44]

I got the same "unexpected irq trap at vector 07" error when trying to boot a new install of openSUSE tumbleweed (kernel was 4.10.3). Using a Ryzen 1700x and Gigabyte Aorus AX370 Gaming 5 with latest BIOS (F5d). Tried installing Fedora 25 from live CD instead and the system booted correctly with both the 4.8 kernel on the live CD and the most recent 4.9.14 kernel once I completed the install and updated.

I wonder if the Fedora kernel is compiled with PINCTRL_AMD disabled or has some patch that prevents this issue from being triggered?