Re: [PATCH v2 2/2] nohz: make nohz_full imply isolcpus

[Chris Metcalf <cmetcalf@ezchip.com>]

On 04/08/2015 01:27 PM, Peter Zijlstra wrote:
> On Wed, Apr 08, 2015 at 11:21:56AM -0400, Chris Metcalf wrote:
>> Apparently the body of the commit message isn't as clear as it might be :-)
>>
>> It does say the same thing, though, basically that if nohz_full DOESN'T
>> imply isolcpus, that's a bad thing.  I'm happy to reword the text to avoid
>> the double negative and say:
>>
>>    nohz_full is only useful with isolcpus also set, since otherwise the
>>    scheduler has to run periodically to try to determine whether to steal
>>    work from other cores.
> But you're doing the reverse! You're setting nohz_full for isolcpus, not
> limiting the nohz_full mask to isolcpus.

Ah, I see.  Yes, that's right.  The idea is that if you are saying
"nohz_full=1-15" on the command line, you would like that to
imply "isolcpus=1-15" as well, without having to actually say so
explicitly.  If we instead limit nohz_full based on isolcpus, it's not
clear that it's actually worth making any change in this area.

I still maintain that the text has always correctly (if perhaps
confusingly) said what it is that the code was doing; where the
text says "x implies y", that means "x being set forces y to be set".
But I'm respinning it anyway for Frederic so I will avoid using the
word "imply" altogether to make this clearer.

The larger question is whether you agree with the proposed semantics.

-- 
Chris Metcalf, EZChip Semiconductor
http://www.ezchip.com