Re: NetBSD 10.1 vs 9.4 x86 stability and gcc ICE

[Dave B <spam%netbsd.dberg.net@localhost>]

On Thu, Sep 10, 2026 at 07:53:26AM +0200, Riccardo Mottola wrote:
...
> However, it was an overheating issue. I did blow off some dust in the fan
> and put the ThinkPad on a laptop stand and it did compile successfully
> through 8 hours.
> 
> No real "code" bugs.
> 
> I would have expected in case of overheating to read something in dmesg, but
> I didn't. Possibly it was just marginal and a memory bank or CPU cache
> started acting.
> 
> RIccardo
...

A few years ago, I started having problems too with large-ish
builds (e.g., pkgsrc packages, build.sh) on one of my long
time trustworthy NetBSD machines--a 2009-ish amd64 system
with 4GB RAM.  I don't remember which NetBSD version it had
back then, but my guess would be 9.x.

It turned out mine was a thermal issue as well, although it
took me awhile to figure that out.

At the time, I solved this by customizing various entries in

  /etc/powerd/scripts/sensor_temperature

to dial down heat generation, such as "offline"-ing extra
cores, or stopping "estd" & explicitly sysctl-ing CPU clock
rate to the minimum--depending on the severity of the temp
event--as temp increased; and progressively ratcheting back
toward normal operation as temp decreased.

This worked like a charm--props to the "powerd"
implementation; and also made it interesting to watch the
system's automated changes to CPU throttling in real time as
powerd adapted to thermal issues.

Then one morning several months ago, the machine had
completely powered itself off, and it refused to turn back on
at all.

I eventually tracked this down to the CPU cooling fan, which
had fully failed and needed to be replaced.  For all I know,
the fan's performance had probably been degrading for awhile,
and could've caused the earlier thermal issues with builds.

Anyway, I was grateful to powerd; it did a solid job of
keeping my system stable all that time until the fan failure.

Best, -D