Re: [GIT PULL v2] sched: Make sleep inside atomic detection work on !PREEMPT

[Ingo Molnar <mingo@elte.hu>]

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

> On Fri, Jul 01, 2011 at 07:29:34PM +0200, Ingo Molnar wrote:
> > 
> > * Frederic Weisbecker <fweisbec@gmail.com> wrote:
> > 
> > > On Fri, Jul 01, 2011 at 05:05:17PM +0200, Ingo Molnar wrote:
> > > > 
> > > > -tip testing also found that m32r fails to build:
> > > > 
> > > > /home/mingo/tip/kernel/sched.c: In function 'preempt_schedule':
> > > > /home/mingo/tip/kernel/sched.c:4364: error: implicit declaration of function 'add_preempt_count_notrace'
> > > > /home/mingo/tip/kernel/sched.c:4366: error: implicit declaration of function 'sub_preempt_count_notrace'
> > > > 
> > > > due to:
> > > > 
> > > > bdd4e85dc36cdbcfc1608a5b2a17c80a9db8986a is the first bad commit
> > > > commit bdd4e85dc36cdbcfc1608a5b2a17c80a9db8986a
> > > > Author: Frederic Weisbecker <fweisbec@gmail.com>
> > > > Date:   Wed Jun 8 01:13:27 2011 +0200
> > > > 
> > > >     sched: Isolate preempt counting in its own config option
> > > > 
> > > > Thanks,
> > > > 
> > > > 	Ingo
> > > 
> > > I have no clue how that did happen.
> > > 
> > > Can you send me your config?
> > 
> > Just the m32r defconfig: m32700ut.smp_defconfig.
> > 
> > My guess would be that one of these:
> > 
> >  include/linux/bit_spinlock.h |    2 +-
> >  include/linux/hardirq.h      |    4 ++--
> >  include/linux/pagemap.h      |    4 ++--
> >  include/linux/preempt.h      |   26 +++++++++++++++++---------
> >  include/linux/rcupdate.h     |   12 ++++++------
> >  include/linux/sched.h        |    2 +-
> > 
> > changed/exposed some header dependency quirk on m32r, failing the 
> > build. Or m32r has some strange .config combo.
> > 
> > Thanks,
> > 
> > 	Ingo
> 
> That's weird.
> 
> We have this:
> 
> 	config PREEMPT
> 		bool
> 		select PREEMPT_COUNT
> 
> 	config PREEMPT_COUNT
> 		bool
> 
> Then I run "make ARCH=m32r defconfig" that has CONFIG_PREEMPT=y
> but it doesn't select PREEMPT_COUNT and happily runs to break few
> times after.
> 
> Even running make ARCH=m32r oldconfig on the defconfig doesn't fix up
> the issue.
> 
> I thought "select" was an unconditional action. What am I missing?
> 
> I'm adding in Cc more people who know better kconfig/kbuild than me.
> Every time I finally consider I know well the Kconfig language, I quickly
> get caught up by the fate...

Architectures have to include kernel/Kconfig.preempt explicitly, 
like:

  arch/x86/Kconfig:source "kernel/Kconfig.preempt"

m32r does not appear to do this - so it does not know about 
PREEMPT_COUNT.

Thanks,

	Ingo