Skip to contentSkip to conversation

AMD RDRAND

DISCUSSION Germano Massullo 2 participants 3 messages Jan 11, 2021 – Jan 12, 2021

lore
Share & tools
lore
#1 lore raw

Hi Germano,

Quoted 11 lines · Germano Massullo

I would like to inform Wireguard developers about the following. it seems that this [1] old AMD RDRAND problem is still present:
RDRAND on AMD Ryzen 9 5900X is flakey -
https://github.com/systemd/systemd/issues/18184

I own 2x Ryzen 3600, feel free to write me in case you would like to run some tests

[1]:
https://arstechnica.com/gadgets/2019/10/how-a-months-old-amd-microcode-bug-destroyed-my-weekend/

I fixed this already, by removing the use of RDRAND from get_random_{u32,u64,int,long} entirely:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=69efea712f5b0489e67d07565aad5c94e09a3e52

This happened almost a year ago back in February, and was backported to the stable kernels too, which means you should have no issues finding a working kernel for your brand new Ryzen.

Is this actually a problem you're experiencing or is your report mostly just speculation based on that systemd issue you linked to?

Jason

#2 lore raw

Il 12/01/21 00:20, Jason A. Donenfeld ha scritto:

Quoted 6 lines

Hi Germano,

I fixed this already, by removing the use of RDRAND from get_random_{u32,u64,int,long} entirely:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=69efea712f5b0489e67d07565aad5c94e09a3e52

Quoted 8 lines

This happened almost a year ago back in February, and was backported to the stable kernels too, which means you should have no issues finding a working kernel for your brand new Ryzen.

Is this actually a problem you're experiencing or is your report mostly just speculation based on that systemd issue you linked to?

Jason

Hi Jason, I did not know you already addressed the problem in the Linux kernel, I just made a quick search in Wireguard mailing list but I did not find any discussion related to RDRAND
I did not experience the problem (I haven't run proper tests yet), and I haven't studied yet in detail all new replies in the thread. I just wanted to let Wireguard developers know about this problem because I thought it may affect Wireguard