Greetings,

On Mon, 28 Jul 2025 21:33:38 +0200,
Tim Leslie <tleslie@protonmail.com> wrote:
>
> Scheduler modifications are tricky territory, so wider testing appreciated
> with more varied workloads.
>

I have no opinion on this diff, but I can run a test: build
ungoogled-chromium with unpack archive on machine with 64 cores.

I've used following snapshot:

~ $ cat /var/db/installed.BUILDINFO                                                                                                        
Build date: 1753384565 - Thu Jul 24 19:16:05 UTC 2025
~ $ 

with kernel which is build from sources similar to this git commit:
https://github.com/openbsd/src/commit/fff03f7679b757d5a2031c6ea272f253ab1ec264

Ok, results. Without your patch:

$ time make build MAKE_JOBS=64
...
  398m31.34s real  4711m38.21s user 18081m09.67s system

With your patch:

$ time make build MAKE_JOBS=64
...
  394m55.14s real  4718m42.05s user 17159m05.12s system


--
wbr, Kirill