"cpuset -n prefer:?" --what values for "?" are supposed to be allowed? (only 1 is, despite two numa domains)

[Mark Millard]

On 2019-Sep-11, at 08:15, Mark Johnston <markj at freebsd.org> wrote:

> On Wed, Sep 11, 2019 at 07:57:26AM -0700, Mark Millard wrote:
>> 
>> 
>> On 2019-Sep-11, at 07:31, Mark Johnston <markj at freebsd.org> wrote:
>> 
>>> On Tue, Sep 10, 2019 at 10:58:05PM -0700, Mark Millard wrote:
>>>> In a context with:
>>>> 
>>>> # cpuset -g
>>>> pid -1 mask: 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27
>>>> pid -1 domain policy: first-touch mask: 0, 1
>>>> 
>>>> I get:
>>>> 
>>>> # cpuset -l0 -n prefer:0 COMMAND
>>>> cpuset: setdomain: Invalid argument
>>>> 
>>>> # cpuset -l0 -n prefer:2 COMMAND
>>>> cpuset: setdomain: Invalid argument
>>>> 
>>>> But one prefer:? value does allow the COMMAND
>>>> to run:
>>>> 
>>>> # cpuset -l0 -n prefer:1 COMMAND
>>>> 
>>>> This seem odd to me. Am I missing something?
>>>> 
>>>> For reference: I'm using a ThreadRipper 1950X
>>>> with a head -r351227 based context for this
>>>> activity. The above happens to have been run
>>>> in a Windows 10 Pro HyperV session, instead
>>>> of in a native-boot of the same media. (A
>>>> native-boot would have had 32 CPUs.)
>>> 
>>> Can you please show the output of "sysctl vm.phys_segs" from this
>>> setup?
>> 
>> Sure:
> 
> I was wondering if you had only one domain populated, but it seems not
> to be the case.  Could you try updating to r351672 or later and see if
> the behaviour persists?

It may be a bit before I do that.

FYI: I had set MAXMEMDOM to match the number of
actual domains for the context:

/usr/src/sys/amd64/conf/GENERIC-DBG:options     MAXMEMDOM=2
/usr/src/sys/amd64/conf/GENERIC-NODBG:options   MAXMEMDOM=2

(These kernel configuration files include GENERIC.)

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