Re: [RFC GIT PULL] nohz: Kconfig layout improvements

[Ingo Molnar <mingo@kernel.org>]

* Frederic Weisbecker <fweisbec@gmail.com> wrote:

> 2013/4/8 Ingo Molnar <mingo@kernel.org>:
> > I pulled it into tip:timers/nohz and will push it out if it passes testing because
> > I like the improvements - but there's still a few things missing I think.
> >
> > Firstly, I performed this "how are users exposed to this new feature" test:
> >
> >    git checkout v3.9-rc6
> >    make defconfig
> >    git checkout tip/master
> >    make oldconfig
> >
> > the x86 (64-bit) defconfig has NO_HZ enabled. When I did the 'make oldconfig', I
> > was given:
> >
> >   *
> >   * Timers subsystem
> >   *
> >   Timer tick handling
> >   > 1. Periodic timer ticks (constant rate, no dynticks) (PERIODIC_HZ) (NEW)
> >     2. Idle dynticks system (tickless idle) (NO_HZ_IDLE) (NEW)
> >   choice[1-2]:
> >
> > [ Firstly, while at it: that should be 'Timer subsystem' or 'Timers'. ]
> 
> Isn't "Timers" too general? I really don't mind changing that though.

I'd suggest consulting Thomas.

> > More importantly, the new Kconfig behavior is still not quite right for two
> > reasons:
> >
> > 1)
> >
> > the default is not set to NO_HZ_IDLE. The oldconfig .config had
> > CONFIG_NO_HZ set - this should be grandfathered over into the new config's
> > default choice. That can probably be done by still keeping CONFIG_NO_HZ as
> > a migration helper entry.
> 
> Ah I did keep it for backward compatibility. We default to
> CONFIG_NO_HZ_IDLE if CONFIG_NO_HZ is set. This is just not working
> because CONFIG_NO_HZ isn't visible. It's an arbirtrary Kconfig
> limitation. I'll just make it visible by adding it a title text and
> this will work.

Okay, cool!

> > 2)
> >
> > there's still no extended idle tick option offered - due to it not meeting
> > the CONFIG_VIRT_CPU_ACCOUNTING_GEN dependency.
> >
> > Even if I read the Kconfig rules and figure out the dependency, I have to
> > perform _two_ steps to get extended ticks:
> >
> > I had to manually find CONFIG_VIRT_CPU_ACCOUNTING_GEN in the .config and
> > delete it, so that I'm given the choice on the next 'make oldconfig'.
> >
> > Then NO_HZ_EXTENDED was set to disabled in the .config silently, because
> > NO_HZ_IDLE was already set. So I had to delete that and re-configure it
> > again.
> 
> Agreed. I mentioned that in the pull request. It's again due to an
> arbitrary Kconfig limitation. The following:
> 
>     config X
>         select Y
> 
> doesn't work if Y is part of a Kconfig "choice".
> I have a patch that fixes in Kconfig, will submit it to Michal and fix
> the nohz passive dependency once we get that sorted.

Wow, you are fixing kconfig.

99% of the people work it around in some fashion.

Kudos!

Thanks,

	Ingo