Build regression in next-20150331

Build regression in next-20150331

[Tyler Baker]

Hi Thomas, Rafael,

I was notified this morning by the kernelci.org system that a new
build error has been detected in next-20150331[0][1][2]. It seems that
"clockevents: Remove CONFIG_GENERIC_CLOCKEVENTS_BUILD"
c9439b1d6eb4ada5c2faf3970ac0d2bc4bd20e14 is the culprit.

Initially, I reported these failures to John Stultz and his response is below.

*snip*

I suspect we either need to enable GENERIC_CLOCKEVENTS on those three
hardware types, or if that's not possible, rework the definitions.

Or something like (copy-paste whitespace corruption below.. only for
reference, don't apply):

diff --git a/kernel/time/tick-internal.h b/kernel/time/tick-internal.h
index 2a1563a..6da40c0 100644
--- a/kernel/time/tick-internal.h
+++ b/kernel/time/tick-internal.h
@@ -107,12 +107,13 @@ static inline void tick_resume_broadcast(void) { }
 static inline bool tick_resume_check_broadcast(void) { return false; }
 static inline void tick_broadcast_init(void) { }
 static inline int tick_broadcast_update_freq(struct
clock_event_device *dev, u32 freq) { return -ENODEV; }
-
+#ifdef CONFIG_GENERIC_CLOCKEVENTS
 /* Set the periodic handler in non broadcast mode */
 static inline void tick_set_periodic_handler(struct
clock_event_device *dev, int broadcast)
 {
        dev->event_handler = tick_handle_periodic;
 }
+#endif
 #endif /* !BROADCAST */

*snip*

Any chance either of you can reproduce this issue on your end?

Cheers,

Tyler

[0] http://storage.kernelci.org/next/next-20150331/arm-ep93xx_defconfig/build.log
[1] http://storage.kernelci.org/next/next-20150331/arm-ebsa110_defconfig/build.log
[2] http://storage.kernelci.org/next/next-20150331/arm-rpc_defconfig/build.log

Re: Build regression in next-20150331

[Rafael J. Wysocki]

On Tuesday, March 31, 2015 11:39:37 AM Tyler Baker wrote:
> Hi Thomas, Rafael,
> 
> I was notified this morning by the kernelci.org system that a new
> build error has been detected in next-20150331[0][1][2]. It seems that
> "clockevents: Remove CONFIG_GENERIC_CLOCKEVENTS_BUILD"
> c9439b1d6eb4ada5c2faf3970ac0d2bc4bd20e14 is the culprit.
> 
> Initially, I reported these failures to John Stultz and his response is below.
> 
> *snip*
> 
> I suspect we either need to enable GENERIC_CLOCKEVENTS on those three
> hardware types, or if that's not possible, rework the definitions.
> 
> Or something like (copy-paste whitespace corruption below.. only for
> reference, don't apply):
> 
> diff --git a/kernel/time/tick-internal.h b/kernel/time/tick-internal.h
> index 2a1563a..6da40c0 100644
> --- a/kernel/time/tick-internal.h
> +++ b/kernel/time/tick-internal.h
> @@ -107,12 +107,13 @@ static inline void tick_resume_broadcast(void) { }
>  static inline bool tick_resume_check_broadcast(void) { return false; }
>  static inline void tick_broadcast_init(void) { }
>  static inline int tick_broadcast_update_freq(struct
> clock_event_device *dev, u32 freq) { return -ENODEV; }
> -
> +#ifdef CONFIG_GENERIC_CLOCKEVENTS
>  /* Set the periodic handler in non broadcast mode */
>  static inline void tick_set_periodic_handler(struct
> clock_event_device *dev, int broadcast)
>  {
>         dev->event_handler = tick_handle_periodic;
>  }
> +#endif
>  #endif /* !BROADCAST */
> 
> *snip*
> 
> Any chance either of you can reproduce this issue on your end?

Can you please tell me if the appended patch helps?

Rafael


---
 kernel/time/tick-internal.h |    4 ++++
 1 file changed, 4 insertions(+)

Index: linux-pm/kernel/time/tick-internal.h
===================================================================
--- linux-pm.orig/kernel/time/tick-internal.h
+++ linux-pm/kernel/time/tick-internal.h
@@ -107,11 +107,15 @@ static inline bool tick_resume_check_bro
 static inline void tick_broadcast_init(void) { }
 static inline int tick_broadcast_update_freq(struct clock_event_device *dev, u32 freq) { return -ENODEV; }
 
+#ifdef CONFIG_GENERIC_CLOCKEVENTS
 /* Set the periodic handler in non broadcast mode */
 static inline void tick_set_periodic_handler(struct clock_event_device *dev, int broadcast)
 {
 	dev->event_handler = tick_handle_periodic;
 }
+#else
+static void tick_handle_periodic(struct clock_event_device *dev) { }
+#endif
 #endif /* !BROADCAST */
 
 /* Functions related to oneshot broadcasting */

Re: Build regression in next-20150331

[Rafael J. Wysocki]

On Wednesday, April 01, 2015 01:03:52 AM Rafael J. Wysocki wrote:
> On Tuesday, March 31, 2015 11:39:37 AM Tyler Baker wrote:
> > Hi Thomas, Rafael,
> > 
> > I was notified this morning by the kernelci.org system that a new
> > build error has been detected in next-20150331[0][1][2]. It seems that
> > "clockevents: Remove CONFIG_GENERIC_CLOCKEVENTS_BUILD"
> > c9439b1d6eb4ada5c2faf3970ac0d2bc4bd20e14 is the culprit.
> > 
> > Initially, I reported these failures to John Stultz and his response is below.
> > 
> > *snip*
> > 
> > I suspect we either need to enable GENERIC_CLOCKEVENTS on those three
> > hardware types, or if that's not possible, rework the definitions.
> > 
> > Or something like (copy-paste whitespace corruption below.. only for
> > reference, don't apply):
> > 
> > diff --git a/kernel/time/tick-internal.h b/kernel/time/tick-internal.h
> > index 2a1563a..6da40c0 100644
> > --- a/kernel/time/tick-internal.h
> > +++ b/kernel/time/tick-internal.h
> > @@ -107,12 +107,13 @@ static inline void tick_resume_broadcast(void) { }
> >  static inline bool tick_resume_check_broadcast(void) { return false; }
> >  static inline void tick_broadcast_init(void) { }
> >  static inline int tick_broadcast_update_freq(struct
> > clock_event_device *dev, u32 freq) { return -ENODEV; }
> > -
> > +#ifdef CONFIG_GENERIC_CLOCKEVENTS
> >  /* Set the periodic handler in non broadcast mode */
> >  static inline void tick_set_periodic_handler(struct
> > clock_event_device *dev, int broadcast)
> >  {
> >         dev->event_handler = tick_handle_periodic;
> >  }
> > +#endif
> >  #endif /* !BROADCAST */
> > 
> > *snip*
> > 
> > Any chance either of you can reproduce this issue on your end?
> 
> Can you please tell me if the appended patch helps?

Scratch that, wrong patch.

Please try the one below instead.

Rafael


---
 kernel/time/tick-internal.h |    2 ++
 1 file changed, 2 insertions(+)

Index: linux-pm/kernel/time/tick-internal.h
===================================================================
--- linux-pm.orig/kernel/time/tick-internal.h
+++ linux-pm/kernel/time/tick-internal.h
@@ -110,7 +110,9 @@ static inline int tick_broadcast_update_
 /* Set the periodic handler in non broadcast mode */
 static inline void tick_set_periodic_handler(struct clock_event_device *dev, int broadcast)
 {
+#ifdef CONFIG_GENERIC_CLOCKEVENTS
 	dev->event_handler = tick_handle_periodic;
+#endif
 }
 #endif /* !BROADCAST */
 

Re: Build regression in next-20150331

[Tyler Baker]

Hi Rafael,

On 31 March 2015 at 16:27, Rafael J. Wysocki <rjw@rjwysocki.net> wrote:
> On Wednesday, April 01, 2015 01:03:52 AM Rafael J. Wysocki wrote:
>> On Tuesday, March 31, 2015 11:39:37 AM Tyler Baker wrote:
>> > Hi Thomas, Rafael,
>> >
>> > I was notified this morning by the kernelci.org system that a new
>> > build error has been detected in next-20150331[0][1][2]. It seems that
>> > "clockevents: Remove CONFIG_GENERIC_CLOCKEVENTS_BUILD"
>> > c9439b1d6eb4ada5c2faf3970ac0d2bc4bd20e14 is the culprit.
>> >
>> > Initially, I reported these failures to John Stultz and his response is below.
>> >
>> > *snip*
>> >
>> > I suspect we either need to enable GENERIC_CLOCKEVENTS on those three
>> > hardware types, or if that's not possible, rework the definitions.
>> >
>> > Or something like (copy-paste whitespace corruption below.. only for
>> > reference, don't apply):
>> >
>> > diff --git a/kernel/time/tick-internal.h b/kernel/time/tick-internal.h
>> > index 2a1563a..6da40c0 100644
>> > --- a/kernel/time/tick-internal.h
>> > +++ b/kernel/time/tick-internal.h
>> > @@ -107,12 +107,13 @@ static inline void tick_resume_broadcast(void) { }
>> >  static inline bool tick_resume_check_broadcast(void) { return false; }
>> >  static inline void tick_broadcast_init(void) { }
>> >  static inline int tick_broadcast_update_freq(struct
>> > clock_event_device *dev, u32 freq) { return -ENODEV; }
>> > -
>> > +#ifdef CONFIG_GENERIC_CLOCKEVENTS
>> >  /* Set the periodic handler in non broadcast mode */
>> >  static inline void tick_set_periodic_handler(struct
>> > clock_event_device *dev, int broadcast)
>> >  {
>> >         dev->event_handler = tick_handle_periodic;
>> >  }
>> > +#endif
>> >  #endif /* !BROADCAST */
>> >
>> > *snip*
>> >
>> > Any chance either of you can reproduce this issue on your end?
>>
>> Can you please tell me if the appended patch helps?
>
> Scratch that, wrong patch.
>
> Please try the one below instead.
>
> Rafael
>
>
> ---
>  kernel/time/tick-internal.h |    2 ++
>  1 file changed, 2 insertions(+)
>
> Index: linux-pm/kernel/time/tick-internal.h
> ===================================================================
> --- linux-pm.orig/kernel/time/tick-internal.h
> +++ linux-pm/kernel/time/tick-internal.h
> @@ -110,7 +110,9 @@ static inline int tick_broadcast_update_
>  /* Set the periodic handler in non broadcast mode */
>  static inline void tick_set_periodic_handler(struct clock_event_device *dev, int broadcast)
>  {
> +#ifdef CONFIG_GENERIC_CLOCKEVENTS
>         dev->event_handler = tick_handle_periodic;
> +#endif
>  }
>  #endif /* !BROADCAST */

Thanks for the quick response. I've applied this patch on top of
next-20150331 and I can confirm this fixes the build error reported in
this thread. The test results can be found here[0].

>
>

Cheers,

Tyler

[0] http://kernelci.org/build/tbaker/kernel/v4.0-rc6-7993-gb448f49+gcc-linaro-4.9-2015.02/

Re: Build regression in next-20150331

[Rafael J. Wysocki]

On Tuesday, March 31, 2015 08:37:32 PM Tyler Baker wrote:
> Hi Rafael,
> 
> On 31 March 2015 at 16:27, Rafael J. Wysocki <rjw@rjwysocki.net> wrote:
> > On Wednesday, April 01, 2015 01:03:52 AM Rafael J. Wysocki wrote:
> >> On Tuesday, March 31, 2015 11:39:37 AM Tyler Baker wrote:
> >> > Hi Thomas, Rafael,
> >> >
> >> > I was notified this morning by the kernelci.org system that a new
> >> > build error has been detected in next-20150331[0][1][2]. It seems that
> >> > "clockevents: Remove CONFIG_GENERIC_CLOCKEVENTS_BUILD"
> >> > c9439b1d6eb4ada5c2faf3970ac0d2bc4bd20e14 is the culprit.

[cut]

> >
> > Please try the one below instead.
> >
> > Rafael
> >
> >
> > ---
> >  kernel/time/tick-internal.h |    2 ++
> >  1 file changed, 2 insertions(+)
> >
> > Index: linux-pm/kernel/time/tick-internal.h
> > ===================================================================
> > --- linux-pm.orig/kernel/time/tick-internal.h
> > +++ linux-pm/kernel/time/tick-internal.h
> > @@ -110,7 +110,9 @@ static inline int tick_broadcast_update_
> >  /* Set the periodic handler in non broadcast mode */
> >  static inline void tick_set_periodic_handler(struct clock_event_device *dev, int broadcast)
> >  {
> > +#ifdef CONFIG_GENERIC_CLOCKEVENTS
> >         dev->event_handler = tick_handle_periodic;
> > +#endif
> >  }
> >  #endif /* !BROADCAST */
> 
> Thanks for the quick response. I've applied this patch on top of
> next-20150331 and I can confirm this fixes the build error reported in
> this thread. The test results can be found here[0].

OK, thanks for verifying.

Peter, do you think the above is acceptable or do I need to do anything more
sophisticated to fix this?  [The alternative would be probably to prepare an
empty definition of tick_handle_periodic() for CONFIG_GENERIC_CLOCKEVENTS and
move the definition of struct clock_event_device from under that Kconfig
option.]

Rafael

Re: Build regression in next-20150331

[Peter Zijlstra]

On Wed, Apr 01, 2015 at 11:46:10PM +0200, Rafael J. Wysocki wrote:

> > > ---
> > >  kernel/time/tick-internal.h |    2 ++
> > >  1 file changed, 2 insertions(+)
> > >
> > > Index: linux-pm/kernel/time/tick-internal.h
> > > ===================================================================
> > > --- linux-pm.orig/kernel/time/tick-internal.h
> > > +++ linux-pm/kernel/time/tick-internal.h
> > > @@ -110,7 +110,9 @@ static inline int tick_broadcast_update_
> > >  /* Set the periodic handler in non broadcast mode */
> > >  static inline void tick_set_periodic_handler(struct clock_event_device *dev, int broadcast)
> > >  {
> > > +#ifdef CONFIG_GENERIC_CLOCKEVENTS
> > >         dev->event_handler = tick_handle_periodic;
> > > +#endif
> > >  }
> > >  #endif /* !BROADCAST */

> Peter, do you think the above is acceptable or do I need to do anything more
> sophisticated to fix this?  [The alternative would be probably to prepare an
> empty definition of tick_handle_periodic() for CONFIG_GENERIC_CLOCKEVENTS and
> move the definition of struct clock_event_device from under that Kconfig
> option.]


Does not something like the below make more sense? The entire broadcast
thing doesn't make sense if we don't have generic_clockevents.

Should we wrap more in generic_clockevents there?

---
 kernel/time/tick-internal.h |    2 ++
 1 file changed, 2 insertions(+)

diff --git a/kernel/time/tick-internal.h b/kernel/time/tick-internal.h
index 2a1563a..5569e65 100644
--- a/kernel/time/tick-internal.h
+++ b/kernel/time/tick-internal.h
@@ -81,6 +81,7 @@ static inline int tick_check_oneshot_change(int allow_nohz) { return 0; }
 #endif /* !TICK_ONESHOT */
 
 /* Broadcasting support */
+#ifdef CONFIG_GENERIC_CLOCKEVENTS
 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
 extern int tick_device_uses_broadcast(struct clock_event_device *dev, int cpu);
 extern void tick_install_broadcast_device(struct clock_event_device *dev);
@@ -114,6 +115,7 @@ static inline void tick_set_periodic_handler(struct clock_event_device *dev, int
 	dev->event_handler = tick_handle_periodic;
 }
 #endif /* !BROADCAST */
+#endif /* GENERIC */
 
 /* Functions related to oneshot broadcasting */
 #if defined(CONFIG_GENERIC_CLOCKEVENTS_BROADCAST) && defined(CONFIG_TICK_ONESHOT)

Re: Build regression in next-20150331

[Rafael J. Wysocki]

On Thursday, April 02, 2015 10:29:17 AM Peter Zijlstra wrote:
> On Wed, Apr 01, 2015 at 11:46:10PM +0200, Rafael J. Wysocki wrote:
> 
> > > > ---
> > > >  kernel/time/tick-internal.h |    2 ++
> > > >  1 file changed, 2 insertions(+)
> > > >
> > > > Index: linux-pm/kernel/time/tick-internal.h
> > > > ===================================================================
> > > > --- linux-pm.orig/kernel/time/tick-internal.h
> > > > +++ linux-pm/kernel/time/tick-internal.h
> > > > @@ -110,7 +110,9 @@ static inline int tick_broadcast_update_
> > > >  /* Set the periodic handler in non broadcast mode */
> > > >  static inline void tick_set_periodic_handler(struct clock_event_device *dev, int broadcast)
> > > >  {
> > > > +#ifdef CONFIG_GENERIC_CLOCKEVENTS
> > > >         dev->event_handler = tick_handle_periodic;
> > > > +#endif
> > > >  }
> > > >  #endif /* !BROADCAST */
> 
> > Peter, do you think the above is acceptable or do I need to do anything more
> > sophisticated to fix this?  [The alternative would be probably to prepare an
> > empty definition of tick_handle_periodic() for CONFIG_GENERIC_CLOCKEVENTS and
> > move the definition of struct clock_event_device from under that Kconfig
> > option.]
> 
> 
> Does not something like the below make more sense? The entire broadcast
> thing doesn't make sense if we don't have generic_clockevents.
> 
> Should we wrap more in generic_clockevents there?

Well, it does make sense, but does it work?

Tyler, can you please test the Peter's patch too?

> ---
>  kernel/time/tick-internal.h |    2 ++
>  1 file changed, 2 insertions(+)
> 
> diff --git a/kernel/time/tick-internal.h b/kernel/time/tick-internal.h
> index 2a1563a..5569e65 100644
> --- a/kernel/time/tick-internal.h
> +++ b/kernel/time/tick-internal.h
> @@ -81,6 +81,7 @@ static inline int tick_check_oneshot_change(int allow_nohz) { return 0; }
>  #endif /* !TICK_ONESHOT */
>  
>  /* Broadcasting support */
> +#ifdef CONFIG_GENERIC_CLOCKEVENTS
>  #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
>  extern int tick_device_uses_broadcast(struct clock_event_device *dev, int cpu);
>  extern void tick_install_broadcast_device(struct clock_event_device *dev);
> @@ -114,6 +115,7 @@ static inline void tick_set_periodic_handler(struct clock_event_device *dev, int
>  	dev->event_handler = tick_handle_periodic;
>  }
>  #endif /* !BROADCAST */
> +#endif /* GENERIC */
>  
>  /* Functions related to oneshot broadcasting */
>  #if defined(CONFIG_GENERIC_CLOCKEVENTS_BROADCAST) && defined(CONFIG_TICK_ONESHOT)

-- 
I speak only for myself.
Rafael J. Wysocki, Intel Open Source Technology Center.

Re: Build regression in next-20150331

[Tyler Baker]

On 2 April 2015 at 01:29, Peter Zijlstra <peterz@infradead.org> wrote:
> On Wed, Apr 01, 2015 at 11:46:10PM +0200, Rafael J. Wysocki wrote:
>
>> > > ---
>> > >  kernel/time/tick-internal.h |    2 ++
>> > >  1 file changed, 2 insertions(+)
>> > >
>> > > Index: linux-pm/kernel/time/tick-internal.h
>> > > ===================================================================
>> > > --- linux-pm.orig/kernel/time/tick-internal.h
>> > > +++ linux-pm/kernel/time/tick-internal.h
>> > > @@ -110,7 +110,9 @@ static inline int tick_broadcast_update_
>> > >  /* Set the periodic handler in non broadcast mode */
>> > >  static inline void tick_set_periodic_handler(struct clock_event_device *dev, int broadcast)
>> > >  {
>> > > +#ifdef CONFIG_GENERIC_CLOCKEVENTS
>> > >         dev->event_handler = tick_handle_periodic;
>> > > +#endif
>> > >  }
>> > >  #endif /* !BROADCAST */
>
>> Peter, do you think the above is acceptable or do I need to do anything more
>> sophisticated to fix this?  [The alternative would be probably to prepare an
>> empty definition of tick_handle_periodic() for CONFIG_GENERIC_CLOCKEVENTS and
>> move the definition of struct clock_event_device from under that Kconfig
>> option.]
>
>
> Does not something like the below make more sense? The entire broadcast
> thing doesn't make sense if we don't have generic_clockevents.
>
> Should we wrap more in generic_clockevents there?
>
> ---
>  kernel/time/tick-internal.h |    2 ++
>  1 file changed, 2 insertions(+)
>
> diff --git a/kernel/time/tick-internal.h b/kernel/time/tick-internal.h
> index 2a1563a..5569e65 100644
> --- a/kernel/time/tick-internal.h
> +++ b/kernel/time/tick-internal.h
> @@ -81,6 +81,7 @@ static inline int tick_check_oneshot_change(int allow_nohz) { return 0; }
>  #endif /* !TICK_ONESHOT */
>
>  /* Broadcasting support */
> +#ifdef CONFIG_GENERIC_CLOCKEVENTS
>  #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
>  extern int tick_device_uses_broadcast(struct clock_event_device *dev, int cpu);
>  extern void tick_install_broadcast_device(struct clock_event_device *dev);
> @@ -114,6 +115,7 @@ static inline void tick_set_periodic_handler(struct clock_event_device *dev, int
>         dev->event_handler = tick_handle_periodic;
>  }
>  #endif /* !BROADCAST */
> +#endif /* GENERIC */
>
>  /* Functions related to oneshot broadcasting */
>  #if defined(CONFIG_GENERIC_CLOCKEVENTS_BROADCAST) && defined(CONFIG_TICK_ONESHOT)

Tested-by: Tyler Baker <tyler.baker@linaro.org>

Looks good to me.

Applied on top of next-20150402[0]:

---
 kernel/time/tick-internal.h | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/kernel/time/tick-internal.h b/kernel/time/tick-internal.h
index 2a1563a..e332bb4 100644
--- a/kernel/time/tick-internal.h
+++ b/kernel/time/tick-internal.h
@@ -81,6 +81,7 @@ static inline int tick_check_oneshot_change(int
allow_nohz) { return 0; }
 #endif /* !TICK_ONESHOT */

 /* Broadcasting support */
+#ifdef CONFIG_GENERIC_CLOCKEVENTS
 #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
 extern int tick_device_uses_broadcast(struct clock_event_device *dev, int cpu);
 extern void tick_install_broadcast_device(struct clock_event_device *dev);
@@ -114,7 +115,7 @@ static inline void
tick_set_periodic_handler(struct clock_event_device *dev, int
  dev->event_handler = tick_handle_periodic;
 }
 #endif /* !BROADCAST */
+#endif /* GENERIC */
 /* Functions related to oneshot broadcasting */
 #if defined(CONFIG_GENERIC_CLOCKEVENTS_BROADCAST) &&
defined(CONFIG_TICK_ONESHOT)
 extern void tick_broadcast_setup_oneshot(struct clock_event_device *bc);
---

Tyler

[0] http://kernelci.org/build/tbaker/kernel/v4.0-rc6-8742-g76c8cce3+gcc-linaro-4.9-2015.02/

Re: Build regression in next-20150331

[Rafael J. Wysocki]

On Thursday, April 02, 2015 07:59:53 PM Tyler Baker wrote:
> On 2 April 2015 at 01:29, Peter Zijlstra <peterz@infradead.org> wrote:
> > On Wed, Apr 01, 2015 at 11:46:10PM +0200, Rafael J. Wysocki wrote:
> >
> >> > > ---
> >> > >  kernel/time/tick-internal.h |    2 ++
> >> > >  1 file changed, 2 insertions(+)
> >> > >
> >> > > Index: linux-pm/kernel/time/tick-internal.h
> >> > > ===================================================================
> >> > > --- linux-pm.orig/kernel/time/tick-internal.h
> >> > > +++ linux-pm/kernel/time/tick-internal.h
> >> > > @@ -110,7 +110,9 @@ static inline int tick_broadcast_update_
> >> > >  /* Set the periodic handler in non broadcast mode */
> >> > >  static inline void tick_set_periodic_handler(struct clock_event_device *dev, int broadcast)
> >> > >  {
> >> > > +#ifdef CONFIG_GENERIC_CLOCKEVENTS
> >> > >         dev->event_handler = tick_handle_periodic;
> >> > > +#endif
> >> > >  }
> >> > >  #endif /* !BROADCAST */
> >
> >> Peter, do you think the above is acceptable or do I need to do anything more
> >> sophisticated to fix this?  [The alternative would be probably to prepare an
> >> empty definition of tick_handle_periodic() for CONFIG_GENERIC_CLOCKEVENTS and
> >> move the definition of struct clock_event_device from under that Kconfig
> >> option.]
> >
> >
> > Does not something like the below make more sense? The entire broadcast
> > thing doesn't make sense if we don't have generic_clockevents.
> >
> > Should we wrap more in generic_clockevents there?
> >
> > ---
> >  kernel/time/tick-internal.h |    2 ++
> >  1 file changed, 2 insertions(+)
> >
> > diff --git a/kernel/time/tick-internal.h b/kernel/time/tick-internal.h
> > index 2a1563a..5569e65 100644
> > --- a/kernel/time/tick-internal.h
> > +++ b/kernel/time/tick-internal.h
> > @@ -81,6 +81,7 @@ static inline int tick_check_oneshot_change(int allow_nohz) { return 0; }
> >  #endif /* !TICK_ONESHOT */
> >
> >  /* Broadcasting support */
> > +#ifdef CONFIG_GENERIC_CLOCKEVENTS
> >  #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
> >  extern int tick_device_uses_broadcast(struct clock_event_device *dev, int cpu);
> >  extern void tick_install_broadcast_device(struct clock_event_device *dev);
> > @@ -114,6 +115,7 @@ static inline void tick_set_periodic_handler(struct clock_event_device *dev, int
> >         dev->event_handler = tick_handle_periodic;
> >  }
> >  #endif /* !BROADCAST */
> > +#endif /* GENERIC */
> >
> >  /* Functions related to oneshot broadcasting */
> >  #if defined(CONFIG_GENERIC_CLOCKEVENTS_BROADCAST) && defined(CONFIG_TICK_ONESHOT)
> 
> Tested-by: Tyler Baker <tyler.baker@linaro.org>
> 
> Looks good to me.
> 
> Applied on top of next-20150402[0]:

Cool, so FWIW

Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>

Peter, I'm assiming that you'll make sure this goes into tip/timers/core.

> ---
>  kernel/time/tick-internal.h | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/kernel/time/tick-internal.h b/kernel/time/tick-internal.h
> index 2a1563a..e332bb4 100644
> --- a/kernel/time/tick-internal.h
> +++ b/kernel/time/tick-internal.h
> @@ -81,6 +81,7 @@ static inline int tick_check_oneshot_change(int
> allow_nohz) { return 0; }
>  #endif /* !TICK_ONESHOT */
> 
>  /* Broadcasting support */
> +#ifdef CONFIG_GENERIC_CLOCKEVENTS
>  #ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
>  extern int tick_device_uses_broadcast(struct clock_event_device *dev, int cpu);
>  extern void tick_install_broadcast_device(struct clock_event_device *dev);
> @@ -114,7 +115,7 @@ static inline void
> tick_set_periodic_handler(struct clock_event_device *dev, int
>   dev->event_handler = tick_handle_periodic;
>  }
>  #endif /* !BROADCAST */
> +#endif /* GENERIC */
>  /* Functions related to oneshot broadcasting */
>  #if defined(CONFIG_GENERIC_CLOCKEVENTS_BROADCAST) &&
> defined(CONFIG_TICK_ONESHOT)
>  extern void tick_broadcast_setup_oneshot(struct clock_event_device *bc);
> ---
> 
> Tyler
> 
> [0] http://kernelci.org/build/tbaker/kernel/v4.0-rc6-8742-g76c8cce3+gcc-linaro-4.9-2015.02/

-- 
I speak only for myself.
Rafael J. Wysocki, Intel Open Source Technology Center.

Re: Build regression in next-20150331

[Peter Zijlstra]

On Fri, Apr 03, 2015 at 01:07:19PM +0200, Rafael J. Wysocki wrote:
> Peter, I'm assiming that you'll make sure this goes into tip/timers/core.

I'll double check, but I think Ingo did a similar thing when he applied
your patches.

Re: Build regression in next-20150331

[Tyler Baker]

On 3 April 2015 at 07:24, Peter Zijlstra <peterz@infradead.org> wrote:
> On Fri, Apr 03, 2015 at 01:07:19PM +0200, Rafael J. Wysocki wrote:
>> Peter, I'm assiming that you'll make sure this goes into tip/timers/core.
>
> I'll double check, but I think Ingo did a similar thing when he applied
> your patches.

Fix has been included next-20150407, this build regression is resolved[0].

Thanks,

Tyler

[0] http://kernelci.org/build/next/kernel/next-20150407/