buggy usage of pm_idle by omap1 code

buggy usage of pm_idle by omap1 code

[Nicolas Pitre]

I just noticed in mach-omap1/cpu.c:omap_pm_prepare() that the function 
pointer pm_idle is explicitly set to NULL for some vague reason.  
However this pointer is unconditionally dereferenced in 
kernel/process.c:cpu_idle().  This is just a oops waiting to happen.  
And if it doesn't happen in practice, then the code in omap_pm_prepare() 
must be useless in addition to being wrong.


Nicolas

buggy usage of pm_idle by omap1 code

[Tony Lindgren]

* Nicolas Pitre <nico@fluxnic.net> [111216 18:53]:
> 
> I just noticed in mach-omap1/cpu.c:omap_pm_prepare() that the function 
> pointer pm_idle is explicitly set to NULL for some vague reason.  
> However this pointer is unconditionally dereferenced in 
> kernel/process.c:cpu_idle().  This is just a oops waiting to happen.  
> And if it doesn't happen in practice, then the code in omap_pm_prepare() 
> must be useless in addition to being wrong.

Good catch. Here's a patch to deal with that issue, want to take this
into your idle series? I'll try out your idle series today at some
point.

Regards,

Tony


From: Tony Lindgren <tony@atomide.com>
Date: Mon, 19 Dec 2011 09:11:11 -0800
Subject: [PATCH] ARM: OMAP1: Fix pm_idle during suspend

Commit 9ccdac3662dbf3c75e8f8851a214bdf7d365a4bd ([ARM] idle:
clean up pm_idle calling, obey hlt_counter) removed a check
for NULL pm_idle.

On omap1 the system hits an equivalent of suspend during idle.
If we are suspending, we need to make sure so we don't call
omap1_pm_idle during the suspend process.

Fix this by setting the pm_idle function to a an empty
function while suspending.

Reported-by: Nicolas Pitre <nico@fluxnic.net>
Signed-off-by: Tony Lindgren <tony@atomide.com>

diff --git a/arch/arm/mach-omap1/pm.c b/arch/arm/mach-omap1/pm.c
index 89ea20c..b551a62 100644
--- a/arch/arm/mach-omap1/pm.c
+++ b/arch/arm/mach-omap1/pm.c
@@ -584,6 +584,9 @@ static void omap_pm_init_proc(void)
 #endif /* DEBUG && CONFIG_PROC_FS */
 
 static void (*saved_idle)(void) = NULL;
+static void omap1_dummy_idle(void)
+{
+}
 
 /*
  *	omap_pm_prepare - Do preliminary suspend work.
@@ -593,7 +596,7 @@ static int omap_pm_prepare(void)
 {
 	/* We cannot sleep in idle until we have resumed */
 	saved_idle = pm_idle;
-	pm_idle = NULL;
+	pm_idle = omap1_dummy_idle;
 
 	return 0;
 }

buggy usage of pm_idle by omap1 code

[Nicolas Pitre]

On Mon, 19 Dec 2011, Tony Lindgren wrote:

> * Nicolas Pitre <nico@fluxnic.net> [111216 18:53]:
> > 
> > I just noticed in mach-omap1/cpu.c:omap_pm_prepare() that the function 
> > pointer pm_idle is explicitly set to NULL for some vague reason.  
> > However this pointer is unconditionally dereferenced in 
> > kernel/process.c:cpu_idle().  This is just a oops waiting to happen.  
> > And if it doesn't happen in practice, then the code in omap_pm_prepare() 
> > must be useless in addition to being wrong.
> 
> Good catch. Here's a patch to deal with that issue, want to take this
> into your idle series?

Sure.

> I'll try out your idle series today at some
> point.

Thanks.


Nicolas

buggy usage of pm_idle by omap1 code

[Russell King - ARM Linux]

On Mon, Dec 19, 2011 at 09:35:19AM -0800, Tony Lindgren wrote:
> Good catch. Here's a patch to deal with that issue, want to take this
> into your idle series? I'll try out your idle series today at some
> point.

Two points:

1. It's good practice to call cpu_idle_wait() after modifying pm_idle,
even on UP - it allows static checking of this.

> From: Tony Lindgren <tony@atomide.com>
> Date: Mon, 19 Dec 2011 09:11:11 -0800
> Subject: [PATCH] ARM: OMAP1: Fix pm_idle during suspend
> 
> Commit 9ccdac3662dbf3c75e8f8851a214bdf7d365a4bd ([ARM] idle:
> clean up pm_idle calling, obey hlt_counter) removed a check
> for NULL pm_idle.
> 
> On omap1 the system hits an equivalent of suspend during idle.
> If we are suspending, we need to make sure so we don't call
> omap1_pm_idle during the suspend process.
> 
> Fix this by setting the pm_idle function to a an empty
> function while suspending.

2. what's wrong with calling disable_hlt() to prevent the pm_idle function
being called?  Why solve it this complicated way of changing the pm_idle
pointer when there's some infrastructure which can fix it for you already?

> 
> Reported-by: Nicolas Pitre <nico@fluxnic.net>
> Signed-off-by: Tony Lindgren <tony@atomide.com>
> 
> diff --git a/arch/arm/mach-omap1/pm.c b/arch/arm/mach-omap1/pm.c
> index 89ea20c..b551a62 100644
> --- a/arch/arm/mach-omap1/pm.c
> +++ b/arch/arm/mach-omap1/pm.c
> @@ -584,6 +584,9 @@ static void omap_pm_init_proc(void)
>  #endif /* DEBUG && CONFIG_PROC_FS */
>  
>  static void (*saved_idle)(void) = NULL;
> +static void omap1_dummy_idle(void)
> +{
> +}
>  
>  /*
>   *	omap_pm_prepare - Do preliminary suspend work.
> @@ -593,7 +596,7 @@ static int omap_pm_prepare(void)
>  {
>  	/* We cannot sleep in idle until we have resumed */
>  	saved_idle = pm_idle;
> -	pm_idle = NULL;
> +	pm_idle = omap1_dummy_idle;
>  
>  	return 0;
>  }
> 
> _______________________________________________
> linux-arm-kernel mailing list
> linux-arm-kernel at lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

buggy usage of pm_idle by omap1 code

[Nicolas Pitre]

On Mon, 19 Dec 2011, Russell King - ARM Linux wrote:

> On Mon, Dec 19, 2011 at 09:35:19AM -0800, Tony Lindgren wrote:
> > Commit 9ccdac3662dbf3c75e8f8851a214bdf7d365a4bd ([ARM] idle:
> > clean up pm_idle calling, obey hlt_counter) removed a check
> > for NULL pm_idle.
> > 
> > On omap1 the system hits an equivalent of suspend during idle.
> > If we are suspending, we need to make sure so we don't call
> > omap1_pm_idle during the suspend process.
> > 
> > Fix this by setting the pm_idle function to a an empty
> > function while suspending.
> 
> 2. what's wrong with calling disable_hlt() to prevent the pm_idle function
> being called?  Why solve it this complicated way of changing the pm_idle
> pointer when there's some infrastructure which can fix it for you already?

Absolutely.  I must have been stairing at the whole thing for too long, 
as I obviously did exactly that already elsewhere.

So I propose this alternative:

    ARM: OMAP1: Fix pm_idle during suspend
    
    Commit 9ccdac3662dbf3c75e8f8851a214bdf7d365a4bd ([ARM] idle:
    clean up pm_idle calling, obey hlt_counter) removed a check
    for NULL pm_idle.
    
    Replace the NULL assignment in the OMAP1 code with disable_hlt()
    to be in sync with the core code and provide the intended behavior.
    
    Signed-off-by: Nicolas Pitre <nicolas.pitre@linaro.org>

diff --git a/arch/arm/mach-omap1/pm.c b/arch/arm/mach-omap1/pm.c
index 89ea20ca0c..6c6a2dc554 100644
--- a/arch/arm/mach-omap1/pm.c
+++ b/arch/arm/mach-omap1/pm.c
@@ -583,8 +583,6 @@ static void omap_pm_init_proc(void)
 
 #endif /* DEBUG && CONFIG_PROC_FS */
 
-static void (*saved_idle)(void) = NULL;
-
 /*
  *	omap_pm_prepare - Do preliminary suspend work.
  *
@@ -592,8 +590,7 @@ static void (*saved_idle)(void) = NULL;
 static int omap_pm_prepare(void)
 {
 	/* We cannot sleep in idle until we have resumed */
-	saved_idle = pm_idle;
-	pm_idle = NULL;
+	disable_hlt();
 
 	return 0;
 }
@@ -630,7 +627,7 @@ static int omap_pm_enter(suspend_state_t state)
 
 static void omap_pm_finish(void)
 {
-	pm_idle = saved_idle;
+	enable_hlt();
 }
 
 

buggy usage of pm_idle by omap1 code

[Tony Lindgren]

* Russell King - ARM Linux <linux@arm.linux.org.uk> [111219 13:02]:
> On Mon, Dec 19, 2011 at 09:35:19AM -0800, Tony Lindgren wrote:
> > Good catch. Here's a patch to deal with that issue, want to take this
> > into your idle series? I'll try out your idle series today at some
> > point.
> 
> Two points:
> 
> 1. It's good practice to call cpu_idle_wait() after modifying pm_idle,
> even on UP - it allows static checking of this.

Hmm, looks like we're not using that anywhere right now. Should
we have a function or macro set_arm_idle() that does that
automatically?
 
> > From: Tony Lindgren <tony@atomide.com>
> > Date: Mon, 19 Dec 2011 09:11:11 -0800
> > Subject: [PATCH] ARM: OMAP1: Fix pm_idle during suspend
> > 
> > Commit 9ccdac3662dbf3c75e8f8851a214bdf7d365a4bd ([ARM] idle:
> > clean up pm_idle calling, obey hlt_counter) removed a check
> > for NULL pm_idle.
> > 
> > On omap1 the system hits an equivalent of suspend during idle.
> > If we are suspending, we need to make sure so we don't call
> > omap1_pm_idle during the suspend process.
> > 
> > Fix this by setting the pm_idle function to a an empty
> > function while suspending.
> 
> 2. what's wrong with calling disable_hlt() to prevent the pm_idle function
> being called?  Why solve it this complicated way of changing the pm_idle
> pointer when there's some infrastructure which can fix it for you already?

Thanks that allows removal of few lines of code. Updated
patch below.

Regards,

Tony


From: Tony Lindgren <tony@atomide.com>
Date: Mon, 19 Dec 2011 09:11:11 -0800
Subject: [PATCH] Subject: [PATCH] ARM: OMAP1: Fix pm_idle during suspend

Commit 9ccdac3662dbf3c75e8f8851a214bdf7d365a4bd ([ARM] idle:
clean up pm_idle calling, obey hlt_counter) removed a check
for NULL pm_idle.

On omap1 the system hits an equivalent of suspend during idle.
If we are suspending, we need to make sure so we don't call
omap1_pm_idle during the suspend process.

Fix this by using disable_hlt and enable_hlt for the suspend
as pointed out by Russell King <linux@arm.linux.org.uk>. This
allows us to remove the saved_idle.

Reported-by: Nicolas Pitre <nico@fluxnic.net>
Signed-off-by: Tony Lindgren <tony@atomide.com>

--- a/arch/arm/mach-omap1/pm.c
+++ b/arch/arm/mach-omap1/pm.c
@@ -583,8 +583,6 @@ static void omap_pm_init_proc(void)
 
 #endif /* DEBUG && CONFIG_PROC_FS */
 
-static void (*saved_idle)(void) = NULL;
-
 /*
  *	omap_pm_prepare - Do preliminary suspend work.
  *
@@ -592,8 +590,7 @@ static void (*saved_idle)(void) = NULL;
 static int omap_pm_prepare(void)
 {
 	/* We cannot sleep in idle until we have resumed */
-	saved_idle = pm_idle;
-	pm_idle = NULL;
+	disable_hlt();
 
 	return 0;
 }
@@ -630,7 +627,7 @@ static int omap_pm_enter(suspend_state_t state)
 
 static void omap_pm_finish(void)
 {
-	pm_idle = saved_idle;
+	enable_hlt();
 }
 
 

buggy usage of pm_idle by omap1 code

[Tony Lindgren]

* Nicolas Pitre <nico@fluxnic.net> [111219 14:32]:
> On Mon, 19 Dec 2011, Russell King - ARM Linux wrote:
> 
> > On Mon, Dec 19, 2011 at 09:35:19AM -0800, Tony Lindgren wrote:
> > > Commit 9ccdac3662dbf3c75e8f8851a214bdf7d365a4bd ([ARM] idle:
> > > clean up pm_idle calling, obey hlt_counter) removed a check
> > > for NULL pm_idle.
> > > 
> > > On omap1 the system hits an equivalent of suspend during idle.
> > > If we are suspending, we need to make sure so we don't call
> > > omap1_pm_idle during the suspend process.
> > > 
> > > Fix this by setting the pm_idle function to a an empty
> > > function while suspending.
> > 
> > 2. what's wrong with calling disable_hlt() to prevent the pm_idle function
> > being called?  Why solve it this complicated way of changing the pm_idle
> > pointer when there's some infrastructure which can fix it for you already?
> 
> Absolutely.  I must have been stairing at the whole thing for too long, 
> as I obviously did exactly that already elsewhere.
> 
> So I propose this alternative:
> 
>     ARM: OMAP1: Fix pm_idle during suspend
>     
>     Commit 9ccdac3662dbf3c75e8f8851a214bdf7d365a4bd ([ARM] idle:
>     clean up pm_idle calling, obey hlt_counter) removed a check
>     for NULL pm_idle.
>     
>     Replace the NULL assignment in the OMAP1 code with disable_hlt()
>     to be in sync with the core code and provide the intended behavior.
>     
>     Signed-off-by: Nicolas Pitre <nicolas.pitre@linaro.org>

Yeh this is better. Looks like you beat me to it by few minutes:

Acked-by: Tony Lindgren <tony@atomide.com>
 
> diff --git a/arch/arm/mach-omap1/pm.c b/arch/arm/mach-omap1/pm.c
> index 89ea20ca0c..6c6a2dc554 100644
> --- a/arch/arm/mach-omap1/pm.c
> +++ b/arch/arm/mach-omap1/pm.c
> @@ -583,8 +583,6 @@ static void omap_pm_init_proc(void)
>  
>  #endif /* DEBUG && CONFIG_PROC_FS */
>  
> -static void (*saved_idle)(void) = NULL;
> -
>  /*
>   *	omap_pm_prepare - Do preliminary suspend work.
>   *
> @@ -592,8 +590,7 @@ static void (*saved_idle)(void) = NULL;
>  static int omap_pm_prepare(void)
>  {
>  	/* We cannot sleep in idle until we have resumed */
> -	saved_idle = pm_idle;
> -	pm_idle = NULL;
> +	disable_hlt();
>  
>  	return 0;
>  }
> @@ -630,7 +627,7 @@ static int omap_pm_enter(suspend_state_t state)
>  
>  static void omap_pm_finish(void)
>  {
> -	pm_idle = saved_idle;
> +	enable_hlt();
>  }
>  
>