Kernel 5.9.x insane loads

[gieltje]

System: Asrock j5005 + cheap ssd + 8GB ram, server only (pi-hole, nextcloud, dsmr-reader, wireguard, nothing special).

When upgrading to kernel 5.9.[0|1] everything goes awry, there are load spikes that cause the system to completly lock up to the point where the only option is a forced power cycle.
Even when stopping every running program that's not absolutely required (gives a nice low memory usage though ), running something like a minecraft java server is sure to kill it withing a minute.
Reverting back to a previous kernel and everything is nice and stable again.

Anyone else experiencing issues like these?

[f1tz]

did you check any logs/dmesg for error posts?

The upgrade to 5.9.1 worked out of the box here.

One of our friends here has some performance issues, too.

[gieltje]

dmesg doesn't report anything, haveged is running, module support isn't even enabled and am using the ondemand governor, there is more then enough ram free (60% orso). The average load on any pre 5.9 kernel is about 0.8, with 5.9 the average load is about 50 with a spike every once in a while that just kills of the system.

glib and glibc are recompiled with every kernel upgrade, so I am a bit baffled by this.

[Hu]

When the load is high, what processes are in device-wait state? On what are they waiting?

[Anon-E-moose]

also when the load gets high what does

cat /proc/sys/kernel/random/entropy_avail

return
_________________
UM780, 6.12 zen kernel, gcc 13, openrc, wayland

[gieltje]

Sorry for the delay, had a busy week.

Entropy is quite stable (3000~4000), but the wait goes up (~95% of the load).
What would be the next debugging step?

[gieltje]

Kernel 5.9.10 seems to have fixed it completely

[toralf]

[gieltje]