> On Mon 28-08-17 18:27:05, Stephen Rothwell wrote:
> > Hi Michal,
> >
> > On Mon, 28 Aug 2017 09:59:31 +0200 Michal Hocko <mhocko@xxxxxxxxxx> wrote:
> > >
> > > From 31d551dbcb1b7987a4cd07767c1e2805849b7a26 Mon Sep 17 00:00:00 2001
> > > From: Michal Hocko <mhocko@xxxxxxxx>
> > > Date: Mon, 28 Aug 2017 09:41:39 +0200
> > > Subject: [PATCH]
> > > mm-hmm-struct-hmm-is-only-use-by-hmm-mirror-functionality-v2-fix
> > >
> > > Compiler is complaining for allnoconfig
> > >
> > > kernel/fork.c: In function 'mm_init':
> > > kernel/fork.c:814:2: error: implicit declaration of function
> > > 'hmm_mm_init' [-Werror=implicit-function-declaration]
> > > hmm_mm_init(mm);
> > > ^
> > > kernel/fork.c: In function '__mmdrop':
> > > kernel/fork.c:893:2: error: implicit declaration of function
> > > 'hmm_mm_destroy' [-Werror=implicit-function-declaration]
> > > hmm_mm_destroy(mm);
> > >
> > > Make sure that hmm_mm_init/hmm_mm_destroy empty stups are defined when
> > > CONFIG_HMM is disabled.
> > >
> > > Signed-off-by: Michal Hocko <mhocko@xxxxxxxx>
> > > ---
> > > include/linux/hmm.h | 7 +++----
> > > 1 file changed, 3 insertions(+), 4 deletions(-)
> > >
> > > diff --git a/include/linux/hmm.h b/include/linux/hmm.h
> > > index 9583d9a15f9c..aeb94e682dda 100644
> > > --- a/include/linux/hmm.h
> > > +++ b/include/linux/hmm.h
> > > @@ -508,11 +508,10 @@ static inline void hmm_mm_init(struct mm_struct
> > > *mm)
> > > {
> > > mm->hmm = NULL;
> > > }
> > > -#else /* IS_ENABLED(CONFIG_HMM_MIRROR) */
> > > +#endif
> > > +
> > > +#else /* IS_ENABLED(CONFIG_HMM) */
> > > static inline void hmm_mm_destroy(struct mm_struct *mm) {}
> > > static inline void hmm_mm_init(struct mm_struct *mm) {}
> > > -#endif /* IS_ENABLED(CONFIG_HMM_MIRROR) */
> > > -
> > > -
> > > #endif /* IS_ENABLED(CONFIG_HMM) */
> > > #endif /* LINUX_HMM_H */
> >
> > What happens when CONFIG_HMM is defined but CONFIG_HMM_MIRROR is not?
> > Or is that not possible (in which case why would we have
> > CONFIG_HMM_MIRROR)?
>
> This is something to Jerome to answer but hmm_mm_init/hmm_mm_destroy are
> used regardless of the specific HMM configuration so an empty stub
> should be defined unconditionally AFAIU.
Sorry for the build issue i posted a patch on friday that i tested against
all combination :
https://lkml.org/lkml/2017/8/25/802
Michal is right this function needs to be defined no matter what but they
only need to be stub if HMM_MIRROR is not enabled.
The fix i posted has the correct logic. I missplaced the endif when i was
fixing Arnd build issue when HMM_MIRROR was not enabled but other HMM feature
were.
Jérôme
