Re: NetBSD identifies then shuns cpu core

[christos%astron.com@localhost (Christos Zoulas)]

In article <86lhwfqv40.fsf%shell.gmplib.org@localhost>,
Torbjorn Granlund  <tg%gmplib.org@localhost> wrote:
>I am migrating several systems from FreeBSD to NetBSD in an effort to
>consolidate my testing environment.  (NetBSD's Xen support and
>up-to-date GNU stuff are the main reasons for this move.)
>
>One system is an "Asrock ION 330", and it behaves quite strange with
>NetBSD.
>
>The CPU in this system, an Atom 330 has 2 real cores, but with SMT it
>presents 4.  I have however disabled SMT in BIOS, leaving just two CPUs.
>
>NetBSD correctly identifies the two CPUs during boot, and sysctl later
>agrees on the number of CPUs.
>
>But it is hard to make cpu0 be used.
>
>With two cpu-bound processes, both seem to be scheduled to cpu1, while
>cpu0 is 100% (or close to 100%) idle.
>
>With three cpu-bound processes, both cores become used.
>
>I use /usr/bin/top as the main tool for making these conclusions.  But
>timing tests confirm the conclusions; the n seconds of cpu time need 2n
>seconds of waal clock time when running two cpu-bound processes.
>
>BIOS bugs?  NetBSD bugs?  Solutions?

Can you see if you can force things with psrset(8)?

christos