Upgrade to F36, trouble with jitter entropy so rngd init fails

Ask Fedora
, , , ,
1

I know I have an old CPU (AMD Phenom 9650 Quad-Core Processor) but things were fine on F34. Just requesting suggestions on how to fix this.
Thanks,
Ralph

10:27 AM rngd.service: Consumed 17.883s CPU time. systemd
10:27 AM rngd.service: Failed with result ‘exit-code’. systemd
10:27 AM rngd.service: Main process exited, code=exited, status=1/FAILURE systemd
10:27 AM Maybe RNG device modules are not loaded rngd
10:27 AM Can’t open any entropy source rngd
10:27 AM [rtlsdr]: Initialization Failed rngd
10:27 AM [jitter]: Initialization Failed rngd
10:27 AM [jitter]: Unable to obtain AES key, disabling JITTER source rngd
10:27 AM [jitter]: Initializing AES buffer rngd
10:27 AM [jitter]: JITTER timeout set to 5 sec rngd

2

Hm. I’m seeing something similar, perhaps-not-coincidentally also on an older AMD CPU. (But mine is MUUUCH older: a ca. 2008 AMD Athlon™ 64 X2 Dual Core Processor 4800+.)

I wonder if the failures are a consequence of this mid-2020 change? Which based on the PR title sounds like it may have fixed a bug that was causing false successful inits. (IOW, things may not have been working under F34 either. A bug may have just been hiding the failures.) I haven’t interpreted that PR enough to be sure I really grok it, though.

3

Curious. It still happens in Fedora 41 (with a similarly old Athlon II X2 280).
Fortunately haveged works well as an alternative,

4

Given the age of the processors involved, I don’t imagine this will ever get fixed. It’ll keep happening to anyone with those CPUs, until there are no more of those CPUs.

(My own Athlon X2 is gone, collateral damage from the motherboard it resided on dying — I replaced it with a Dell motherboard containing a Core2 Duo E7500, which was basically a lateral move except it doesn’t tick off rngd.)

Good tip on haveged, though, which is in the Fedora repos for anyone who needs.

5

Heh! I take that back. I forgot that I’d disabled rngd.service on that machine (why not, if it was always failing), and never turned it back on when i swapped the motherboards.

I tried that just now, and it turns out rngd fails the same way on my Core2 Duo E7500. So this isn’t even just limited to old AMD processsors. (It does seem limited to old processors, though — rngd.service starts up just fine on my i5-2400 machine.)

1 Like
6

I just learned a new thing (which definitively closes the issue):

remove rng-tools because in newer linux kernel versions /dev/random is not blocking anymore (5.6+) and the standard kernel provided entropy pool is strong enough (5.10+) so that tools like rng-tools or haveged are not required anymore and just unnecessarily consume CPU time.

4 Likes
7

This topic is about F36, a lot has changed since. Even if an issue feels comparable, you need to get new data and cannot rely on old still being comparable. Blurring such topics by mixing data and analysis of different eras is likely to hinder problem solving. Therefore, please open a new topic. You might reference this old one in the new one if you think it is comparable, but that way data of both attempts remains comprehensible and separable without blurs.