PowerMac G5 fans going crazy

[Mark Millard]

On 2021-Feb-16, at 18:49, Julio Merino <julio at meroh.net> wrote:

> Hi,
> 
> I have this liquid-cooled dual G5 PowerMac sitting in my cold garage. The
> machine runs fine for a while and can even do a buildworld in silence...
> but sometimes, pretty much at random, decides to spin up its fans to the
> maximum. The machine is pretty much idle when this happens so usage doesn't
> explain this.
> 
> I haven't correlated this fully, but when the machine does this, it seems
> to also become unresponsive. Moving the mouse on the console shows the
> pointer moving but it eventually freezes. Except... the machine responds to
> pings, and even to SSH. *However*, it is sluggish. SSH takes forever to
> respond, but eventually does. Similarly, the console eventually recovers
> too and allows input... until things become slow again. And while it's
> responding, running "top" doesn't show anything strange, not in user CPU
> consumption nor in sys.
> 
> Interestingly, running a "sysctl -a" makes the fans quiet down, maybe for a
> long time, maybe for a couple of minutes only. See the diff below of
> "sysctl -a" from while the fans are full-speed to right after, and notice
> that the temperatures look just fine.
> 
> Any ideas about what might be happening here?
> 
> Thanks!
> 
> 
> @@ -1,7 +1,9 @@
> -dev.smusat.0.cpu_a1_core_current: 10
> +device smu
> +kern.consmute: 0
> +dev.smusat.0.cpu_a1_core_current: 11
> dev.smusat.0.cpu_a0_core_current: 10
> -dev.smusat.0.cpu_a1_diode_temp: 27.7C
> -dev.smusat.0.cpu_a0_diode_temp: 26.5C
> +dev.smusat.0.cpu_a1_diode_temp: 28.1C
> +dev.smusat.0.cpu_a0_diode_temp: 26.7C
> dev.smusat.0.cpu_a1_vcore: 1
> dev.smusat.0.cpu_a0_vcore: 1
> dev.smusat.0.%parent: iicbus2
> @@ -12,25 +14,25 @@
> dev.smusat.%parent:
> dev.smu.0.server_mode: 0
> dev.smu.0.sensors.slots_power: 16816
> -dev.smu.0.fans.expansion_slots_intake.rpm: 4487
> +dev.smu.0.fans.expansion_slots_intake.rpm: 1572
> dev.smu.0.fans.expansion_slots_intake.maxrpm: 4500
> dev.smu.0.fans.expansion_slots_intake.minrpm: 1560
> -dev.smu.0.fans.cpu_b_exhaust.rpm: 3168
> +dev.smu.0.fans.cpu_b_exhaust.rpm: 551
> dev.smu.0.fans.cpu_b_exhaust.maxrpm: 3200
> dev.smu.0.fans.cpu_b_exhaust.minrpm: 500
> -dev.smu.0.fans.cpu_a_exhaust.rpm: 3031
> +dev.smu.0.fans.cpu_a_exhaust.rpm: 548
> dev.smu.0.fans.cpu_a_exhaust.maxrpm: 3200
> dev.smu.0.fans.cpu_a_exhaust.minrpm: 500
> -dev.smu.0.fans.cpu_b_intake.rpm: 3045
> +dev.smu.0.fans.cpu_b_intake.rpm: 534
> dev.smu.0.fans.cpu_b_intake.maxrpm: 3200
> dev.smu.0.fans.cpu_b_intake.minrpm: 485
> -dev.smu.0.fans.cpu_a_intake.rpm: 3173
> +dev.smu.0.fans.cpu_a_intake.rpm: 538
> dev.smu.0.fans.cpu_a_intake.maxrpm: 3200
> dev.smu.0.fans.cpu_a_intake.minrpm: 485
> -dev.smu.0.fans.backside.rpm: 2851
> +dev.smu.0.fans.backside.rpm: 1174
> dev.smu.0.fans.backside.maxrpm: 3000
> dev.smu.0.fans.backside.minrpm: 1100
> -dev.smu.0.fans.drive_bay_a_intake.rpm: 3792
> +dev.smu.0.fans.drive_bay_a_intake.rpm: 1011
> dev.smu.0.fans.drive_bay_a_intake.maxrpm: 4000
> dev.smu.0.fans.drive_bay_a_intake.minrpm: 1000
> dev.smu.0.%parent: ofwbus0

See: https://reviews.freebsd.org/D23376

Basically the times used to control sleeping
is mismatched by too much across cpus/cores
and, with context switching, processes/threads
end up stuck sleeping for long periods
compared to the intent. [pmac_thermal] is
not the only process/thread that can end up
with a problem becoming obvious. There can
be less obvious problems as well.

This has been an issue for years for
multi-socket PowerMacs and possibly
single-socket/2-core ones as well. (I've
never had access to such.)

(This is not the only significant PowerMac
issue around but I'll keep it simple here.)

===
Mark Millard
marklmi at yahoo.com
( dsl-only.net went
away in early 2018-Mar)