[PATCH] sched, cpuidle: Do not access cpuidle_devices when !CONFIG_CPU_IDLE

[PATCH] sched, cpuidle: Do not access cpuidle_devices when !CONFIG_CPU_IDLE

[Catalin Marinas]

The cpuidle_devices per-CPU variable is only defined when CPU_IDLE is
enabled. Commit c8cc7d4de7a4 ("sched/idle: Reorganize the idle loop")
removed the #ifdef CONFIG_CPU_IDLE around cpuidle_idle_call() with the
compiler optimising away __this_cpu_read(cpuidle_devices). However, with
CONFIG_UBSAN && !CONFIG_CPU_IDLE, this optimisation no longer happens
and the kernel fails to link since cpuidle_devices is not defined.

This patch introduces an accessor function for the current CPU cpuidle
device (returning NULL when !CONFIG_CPU_IDLE) and uses it in
cpuidle_idle_call().

Cc: Rafael J. Wysocki <rjw@rjwysocki.net>
Cc: Daniel Lezcano <daniel.lezcano@linaro.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
---

This fixes a build failure on arm64 with UBSAN && !CPU_IDLE and gcc
5.3.1 (probably other compiler versions would behave in the same way).

I tried to keep close to the existing coding style in cpuidle.h.

 include/linux/cpuidle.h | 3 +++
 kernel/sched/idle.c     | 2 +-
 2 files changed, 4 insertions(+), 1 deletion(-)

diff --git a/include/linux/cpuidle.h b/include/linux/cpuidle.h
index 786ad32631a6..07b83d32f66c 100644
--- a/include/linux/cpuidle.h
+++ b/include/linux/cpuidle.h
@@ -152,6 +152,8 @@ extern void cpuidle_disable_device(struct cpuidle_device *dev);
 extern int cpuidle_play_dead(void);
 
 extern struct cpuidle_driver *cpuidle_get_cpu_driver(struct cpuidle_device *dev);
+static inline struct cpuidle_device *cpuidle_get_device(void)
+{return __this_cpu_read(cpuidle_devices); }
 #else
 static inline void disable_cpuidle(void) { }
 static inline bool cpuidle_not_available(struct cpuidle_driver *drv,
@@ -187,6 +189,7 @@ static inline void cpuidle_disable_device(struct cpuidle_device *dev) { }
 static inline int cpuidle_play_dead(void) {return -ENODEV; }
 static inline struct cpuidle_driver *cpuidle_get_cpu_driver(
 	struct cpuidle_device *dev) {return NULL; }
+static inline struct cpuidle_device *cpuidle_get_device(void) {return NULL; }
 #endif
 
 #if defined(CONFIG_CPU_IDLE) && defined(CONFIG_SUSPEND)
diff --git a/kernel/sched/idle.c b/kernel/sched/idle.c
index bd12c6c714ec..c5aeedf4e93a 100644
--- a/kernel/sched/idle.c
+++ b/kernel/sched/idle.c
@@ -127,7 +127,7 @@ static int call_cpuidle(struct cpuidle_driver *drv, struct cpuidle_device *dev,
  */
 static void cpuidle_idle_call(void)
 {
-	struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices);
+	struct cpuidle_device *dev = cpuidle_get_device();
 	struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
 	int next_state, entered_state;
 

Re: [PATCH] sched, cpuidle: Do not access cpuidle_devices when !CONFIG_CPU_IDLE

[Rafael J. Wysocki]

On Wednesday, June 01, 2016 06:52:16 PM Catalin Marinas wrote:
> The cpuidle_devices per-CPU variable is only defined when CPU_IDLE is
> enabled. Commit c8cc7d4de7a4 ("sched/idle: Reorganize the idle loop")
> removed the #ifdef CONFIG_CPU_IDLE around cpuidle_idle_call() with the
> compiler optimising away __this_cpu_read(cpuidle_devices). However, with
> CONFIG_UBSAN && !CONFIG_CPU_IDLE, this optimisation no longer happens
> and the kernel fails to link since cpuidle_devices is not defined.
> 
> This patch introduces an accessor function for the current CPU cpuidle
> device (returning NULL when !CONFIG_CPU_IDLE) and uses it in
> cpuidle_idle_call().
> 
> Cc: Rafael J. Wysocki <rjw@rjwysocki.net>
> Cc: Daniel Lezcano <daniel.lezcano@linaro.org>
> Cc: Ingo Molnar <mingo@redhat.com>
> Cc: Peter Zijlstra <peterz@infradead.org>
> Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
> Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>

OK

Is this needed in -stable?

> ---
> 
> This fixes a build failure on arm64 with UBSAN && !CPU_IDLE and gcc
> 5.3.1 (probably other compiler versions would behave in the same way).
> 
> I tried to keep close to the existing coding style in cpuidle.h.
> 
>  include/linux/cpuidle.h | 3 +++
>  kernel/sched/idle.c     | 2 +-
>  2 files changed, 4 insertions(+), 1 deletion(-)
> 
> diff --git a/include/linux/cpuidle.h b/include/linux/cpuidle.h
> index 786ad32631a6..07b83d32f66c 100644
> --- a/include/linux/cpuidle.h
> +++ b/include/linux/cpuidle.h
> @@ -152,6 +152,8 @@ extern void cpuidle_disable_device(struct cpuidle_device *dev);
>  extern int cpuidle_play_dead(void);
>  
>  extern struct cpuidle_driver *cpuidle_get_cpu_driver(struct cpuidle_device *dev);
> +static inline struct cpuidle_device *cpuidle_get_device(void)
> +{return __this_cpu_read(cpuidle_devices); }
>  #else
>  static inline void disable_cpuidle(void) { }
>  static inline bool cpuidle_not_available(struct cpuidle_driver *drv,
> @@ -187,6 +189,7 @@ static inline void cpuidle_disable_device(struct cpuidle_device *dev) { }
>  static inline int cpuidle_play_dead(void) {return -ENODEV; }
>  static inline struct cpuidle_driver *cpuidle_get_cpu_driver(
>  	struct cpuidle_device *dev) {return NULL; }
> +static inline struct cpuidle_device *cpuidle_get_device(void) {return NULL; }
>  #endif
>  
>  #if defined(CONFIG_CPU_IDLE) && defined(CONFIG_SUSPEND)
> diff --git a/kernel/sched/idle.c b/kernel/sched/idle.c
> index bd12c6c714ec..c5aeedf4e93a 100644
> --- a/kernel/sched/idle.c
> +++ b/kernel/sched/idle.c
> @@ -127,7 +127,7 @@ static int call_cpuidle(struct cpuidle_driver *drv, struct cpuidle_device *dev,
>   */
>  static void cpuidle_idle_call(void)
>  {
> -	struct cpuidle_device *dev = __this_cpu_read(cpuidle_devices);
> +	struct cpuidle_device *dev = cpuidle_get_device();
>  	struct cpuidle_driver *drv = cpuidle_get_cpu_driver(dev);
>  	int next_state, entered_state;
>  

Re: [PATCH] sched, cpuidle: Do not access cpuidle_devices when !CONFIG_CPU_IDLE

[Catalin Marinas]

On Wed, Jun 01, 2016 at 10:20:03PM +0200, Rafael J. Wysocki wrote:
> On Wednesday, June 01, 2016 06:52:16 PM Catalin Marinas wrote:
> > The cpuidle_devices per-CPU variable is only defined when CPU_IDLE is
> > enabled. Commit c8cc7d4de7a4 ("sched/idle: Reorganize the idle loop")
> > removed the #ifdef CONFIG_CPU_IDLE around cpuidle_idle_call() with the
> > compiler optimising away __this_cpu_read(cpuidle_devices). However, with
> > CONFIG_UBSAN && !CONFIG_CPU_IDLE, this optimisation no longer happens
> > and the kernel fails to link since cpuidle_devices is not defined.
> > 
> > This patch introduces an accessor function for the current CPU cpuidle
> > device (returning NULL when !CONFIG_CPU_IDLE) and uses it in
> > cpuidle_idle_call().
> > 
> > Cc: Rafael J. Wysocki <rjw@rjwysocki.net>
> > Cc: Daniel Lezcano <daniel.lezcano@linaro.org>
> > Cc: Ingo Molnar <mingo@redhat.com>
> > Cc: Peter Zijlstra <peterz@infradead.org>
> > Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
> > Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
> 
> OK
> 
> Is this needed in -stable?

Not sure how common this config combination is (I was just testing
different options and came across it). Even though it's not a fix for
UBSAN, we could add it up to 4.5 since that's when UBSAN was merged:

Cc: <stable@vger.kernel.org> # 4.5+

-- 
Catalin