Skip to contentSkip to conversation

OMAP2/3: GPIO: remove recursion in IRQ wakeup path

PATCH In review Kevin Hilman 3 participants 7 messages Jan 28, 2009 – Feb 20, 2009

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 OMAP2/3: GPIO: remove recursion in IRQ wakeup path

Now that the generic IRQ and GPIO frameworks are used for enabling and disabling GPIO IRQ wakeup sources, there is no longer a need to call [enable|disable]_irq_wake() in the low-level code. Doing so results in recursive calls to [enable|disable]_irq_wake().

This was discovered in the suspend/resume path on OMAP3/Beagle using the gpio-keys driver which disables/re-enables GPIO IRQ wakeups in the suspend/resume path.

---
 arch/arm/plat-omap/gpio.c |   14 ++++----------
 1 files changed, 4 insertions(+), 10 deletions(-)
arch/arm/plat-omap/gpio.c +4 -10
index f856a90..798a8cd 100644
@@ -837,13 +837,10 @@ static int _set_gpio_wakeup(struct gpio_bank *bank, int gpio, int enable)
 	case METHOD_MPUIO:
 	case METHOD_GPIO_1610:
 		spin_lock_irqsave(&bank->lock, flags);
-		if (enable) {
+		if (enable)
 			bank->suspend_wakeup |= (1 << gpio);
-			enable_irq_wake(bank->irq);
-		} else {
-			disable_irq_wake(bank->irq);
+		else
 			bank->suspend_wakeup &= ~(1 << gpio);
-		}
 		spin_unlock_irqrestore(&bank->lock, flags);
 		return 0;
 #endif
@@ -856,13 +853,10 @@ static int _set_gpio_wakeup(struct gpio_bank *bank, int gpio, int enable)
 			return -EINVAL;
 		}
 		spin_lock_irqsave(&bank->lock, flags);
-		if (enable) {
+		if (enable)
 			bank->suspend_wakeup |= (1 << gpio);
-			enable_irq_wake(bank->irq);
-		} else {
-			disable_irq_wake(bank->irq);
+		else
 			bank->suspend_wakeup &= ~(1 << gpio);
-		}
 		spin_unlock_irqrestore(&bank->lock, flags);
 		return 0;
 #endif
Signature · 1 line
1.6.1
#1 lore raw
Quoted 10 lines · Kevin Hilman

Now that the generic IRQ and GPIO frameworks are used for enabling and disabling GPIO IRQ wakeup sources, there is no longer a need to call [enable|disable]_irq_wake() in the low-level code. Doing so results in recursive calls to [enable|disable]_irq_wake().

This was discovered in the suspend/resume path on OMAP3/Beagle using the gpio-keys driver which disables/re-enables GPIO IRQ wakeups in the suspend/resume path.

Tony,

Not sure if it's too late, but this could go into omap-fixes too.

Kevin

Quoted 42 lines
---
 arch/arm/plat-omap/gpio.c |   14 ++++----------
 1 files changed, 4 insertions(+), 10 deletions(-)
arch/arm/plat-omap/gpio.c +4 -10
index f856a90..798a8cd 100644
@@ -837,13 +837,10 @@ static int _set_gpio_wakeup(struct gpio_bank *bank, int gpio, int enable)
 	case METHOD_MPUIO:
 	case METHOD_GPIO_1610:
 		spin_lock_irqsave(&bank->lock, flags);
-		if (enable) {
+		if (enable)
 			bank->suspend_wakeup |= (1 << gpio);
-			enable_irq_wake(bank->irq);
-		} else {
-			disable_irq_wake(bank->irq);
+		else
 			bank->suspend_wakeup &= ~(1 << gpio);
-		}
 		spin_unlock_irqrestore(&bank->lock, flags);
 		return 0;
 #endif
@@ -856,13 +853,10 @@ static int _set_gpio_wakeup(struct gpio_bank *bank, int gpio, int enable)
 			return -EINVAL;
 		}
 		spin_lock_irqsave(&bank->lock, flags);
-		if (enable) {
+		if (enable)
 			bank->suspend_wakeup |= (1 << gpio);
-			enable_irq_wake(bank->irq);
-		} else {
-			disable_irq_wake(bank->irq);
+		else
 			bank->suspend_wakeup &= ~(1 << gpio);
-		}
 		spin_unlock_irqrestore(&bank->lock, flags);
 		return 0;
 #endif
Signature · 1 line
1.6.1
#2 lore raw
Quoted 11 lines · Kevin Hilman
Quoted 9 lines · Kevin Hilman

Now that the generic IRQ and GPIO frameworks are used for enabling and disabling GPIO IRQ wakeup sources, there is no longer a need to call [enable|disable]_irq_wake() in the low-level code. Doing so results in recursive calls to [enable|disable]_irq_wake().

This was discovered in the suspend/resume path on OMAP3/Beagle using the gpio-keys driver which disables/re-enables GPIO IRQ wakeups in the suspend/resume path.

Quoted 5 lines
Quoted 1 line

Tony,

Not sure if it's too late, but this could go into omap-fixes too.

I'll add it after the previous omap-fixes are merged to the mainline.

Tony

Quoted 50 lines

Kevin

Quoted 42 lines
---
 arch/arm/plat-omap/gpio.c |   14 ++++----------
 1 files changed, 4 insertions(+), 10 deletions(-)
arch/arm/plat-omap/gpio.c +4 -10
index f856a90..798a8cd 100644
@@ -837,13 +837,10 @@ static int _set_gpio_wakeup(struct gpio_bank *bank, int gpio, int enable)
 	case METHOD_MPUIO:
 	case METHOD_GPIO_1610:
 		spin_lock_irqsave(&bank->lock, flags);
-		if (enable) {
+		if (enable)
 			bank->suspend_wakeup |= (1 << gpio);
-			enable_irq_wake(bank->irq);
-		} else {
-			disable_irq_wake(bank->irq);
+		else
 			bank->suspend_wakeup &= ~(1 << gpio);
-		}
 		spin_unlock_irqrestore(&bank->lock, flags);
 		return 0;
 #endif
@@ -856,13 +853,10 @@ static int _set_gpio_wakeup(struct gpio_bank *bank, int gpio, int enable)
 			return -EINVAL;
 		}
 		spin_lock_irqsave(&bank->lock, flags);
-		if (enable) {
+		if (enable)
 			bank->suspend_wakeup |= (1 << gpio);
-			enable_irq_wake(bank->irq);
-		} else {
-			disable_irq_wake(bank->irq);
+		else
 			bank->suspend_wakeup &= ~(1 << gpio);
-		}
 		spin_unlock_irqrestore(&bank->lock, flags);
 		return 0;
 #endif
Signature · 1 line
1.6.1

--
To unsubscribe from this list: send the line "unsubscribe linux-omap" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

#3 lore raw
In reply to Kevin Hilman
Quoted 4 lines · Kevin Hilman

Now that the generic IRQ and GPIO frameworks are used for enabling and disabling GPIO IRQ wakeup sources, there is no longer a need to call [enable|disable]_irq_wake() in the low-level code. Doing so results in recursive calls to [enable|disable]_irq_wake().

Could you clarify what actually made that requirement go away?

The recursion was introduced -- using the generic IRQ framework! -- as a simple way to ensure the parent IRQ was properly wake-enabled. Is the issue that something changed, so that something else wake-enables the parent?

- Dave

Quoted 49 lines

This was discovered in the suspend/resume path on OMAP3/Beagle using the gpio-keys driver which disables/re-enables GPIO IRQ wakeups in the suspend/resume path.

---
 arch/arm/plat-omap/gpio.c |   14 ++++----------
 1 files changed, 4 insertions(+), 10 deletions(-)
arch/arm/plat-omap/gpio.c +4 -10
index f856a90..798a8cd 100644
@@ -837,13 +837,10 @@ static int _set_gpio_wakeup(struct gpio_bank *bank, int gpio, int enable)
 	case METHOD_MPUIO:
 	case METHOD_GPIO_1610:
 		spin_lock_irqsave(&bank->lock, flags);
-		if (enable) {
+		if (enable)
 			bank->suspend_wakeup |= (1 << gpio);
-			enable_irq_wake(bank->irq);
-		} else {
-			disable_irq_wake(bank->irq);
+		else
 			bank->suspend_wakeup &= ~(1 << gpio);
-		}
 		spin_unlock_irqrestore(&bank->lock, flags);
 		return 0;
 #endif
@@ -856,13 +853,10 @@ static int _set_gpio_wakeup(struct gpio_bank *bank, int gpio, int enable)
 			return -EINVAL;
 		}
 		spin_lock_irqsave(&bank->lock, flags);
-		if (enable) {
+		if (enable)
 			bank->suspend_wakeup |= (1 << gpio);
-			enable_irq_wake(bank->irq);
-		} else {
-			disable_irq_wake(bank->irq);
+		else
 			bank->suspend_wakeup &= ~(1 << gpio);
-		}
 		spin_unlock_irqrestore(&bank->lock, flags);
 		return 0;
 #endif
Signature · 1 line
1.6.1
#4 lore raw
Quoted 6 lines · David Brownell
Quoted 4 lines · Kevin Hilman

Now that the generic IRQ and GPIO frameworks are used for enabling and disabling GPIO IRQ wakeup sources, there is no longer a need to call [enable|disable]_irq_wake() in the low-level code. Doing so results in recursive calls to [enable|disable]_irq_wake().

Quoted 7 lines

Could you clarify what actually made that requirement go away?

The recursion was introduced -- using the generic IRQ framework! -- as a simple way to ensure the parent IRQ was properly wake-enabled. Is the issue that something changed, so that something else wake-enables the parent?

My description was not very descriptive... sorry...

Quoted 1 line

From what I can understand here, I don't see the point in

wake-enabling the parent IRQ since there is no wakeup glue for the bank IRQs, thus these calls are actually failing and causing 'unbalanced IRQ disable' noise in the generic IRQ layer.

Here is what is happening. Consider the gpio-keys driver. Upon suspend, it enables the IRQ wake for its GPIO. The OMAP GPIO code then calls enable_wake_irq() for the parent irq (the GPIO bank IRQ). This call is actually failing because the bank IRQ has no 'set_wake' method. Because of this failure, the generic IRQ code doesn't actually do anything, and sets the 'disable_depth' to zero for the bank IRQ.

Then, upon resume, the resume path disables IRQ wakeups for the GPIO. The OMAP GPIO code then tries to call disable_irq_wake() for the bank IRQ and you get noisy 'unbalanced IRQ disable' warnings from the generic IRQ layere because of the attempt to disable the IRQ wake of an IRQ that was never enabled.

So the options that I see are:

1) fix the OMAP GPIO code to check the return code of the parent enable, or
2) drop the parent enable/disable

I prefer (1) since those calls will always fail AFAICT.

Does that make any sense?

If you're OK with (1), I will re-issue the patch with a better discription.

Kevin

#5 lore raw
Quoted 39 lines · Kevin Hilman
Quoted 6 lines · David Brownell
Quoted 4 lines · Kevin Hilman

Now that the generic IRQ and GPIO frameworks are used for enabling and disabling GPIO IRQ wakeup sources, there is no longer a need to call [enable|disable]_irq_wake() in the low-level code. Doing so results in recursive calls to [enable|disable]_irq_wake().

Quoted 7 lines

Could you clarify what actually made that requirement go away?

The recursion was introduced -- using the generic IRQ framework! -- as a simple way to ensure the parent IRQ was properly wake-enabled. Is the issue that something changed, so that something else wake-enables the parent?

My description was not very descriptive... sorry...

From what I can understand here, I don't see the point in wake-enabling the parent IRQ since there is no wakeup glue for the bank IRQs, thus these calls are actually failing and causing 'unbalanced IRQ disable' noise in the generic IRQ layer.

Here is what is happening. Consider the gpio-keys driver. Upon suspend, it enables the IRQ wake for its GPIO. The OMAP GPIO code then calls enable_wake_irq() for the parent irq (the GPIO bank IRQ). This call is actually failing because the bank IRQ has no 'set_wake' method. Because of this failure, the generic IRQ code doesn't actually do anything, and sets the 'disable_depth' to zero for the bank IRQ.

Then, upon resume, the resume path disables IRQ wakeups for the GPIO. The OMAP GPIO code then tries to call disable_irq_wake() for the bank IRQ and you get noisy 'unbalanced IRQ disable' warnings from the generic IRQ layere because of the attempt to disable the IRQ wake of an IRQ that was never enabled.

So the options that I see are:

Quoted 8 lines

1) fix the OMAP GPIO code to check the return code of the parent enable, or
2) drop the parent enable/disable

I prefer (1) since those calls will always fail AFAICT.

Does that make any sense?

If you're OK with (1), I will re-issue the patch with a better discription.

Ignoring this for now, please let me know if you want me to queue this for omap-upstream with the updates mentioned above.

Regards,

Tony

#6 lore raw
Quoted 41 lines · Tony Lindgren
Quoted 40 lines · Kevin Hilman
Quoted 5 lines · David Brownell
Quoted 4 lines · Kevin Hilman

Now that the generic IRQ and GPIO frameworks are used for enabling and disabling GPIO IRQ wakeup sources, there is no longer a need to call [enable|disable]_irq_wake() in the low-level code. Doing so results in recursive calls to [enable|disable]_irq_wake().

Quoted 7 lines

Could you clarify what actually made that requirement go away?

The recursion was introduced -- using the generic IRQ framework! -- as a simple way to ensure the parent IRQ was properly wake-enabled. Is the issue that something changed, so that something else wake-enables the parent?

My description was not very descriptive... sorry...

From what I can understand here, I don't see the point in wake-enabling the parent IRQ since there is no wakeup glue for the bank IRQs, thus these calls are actually failing and causing 'unbalanced IRQ disable' noise in the generic IRQ layer.

Here is what is happening. Consider the gpio-keys driver. Upon suspend, it enables the IRQ wake for its GPIO. The OMAP GPIO code then calls enable_wake_irq() for the parent irq (the GPIO bank IRQ). This call is actually failing because the bank IRQ has no 'set_wake' method. Because of this failure, the generic IRQ code doesn't actually do anything, and sets the 'disable_depth' to zero for the bank IRQ.

Then, upon resume, the resume path disables IRQ wakeups for the GPIO. The OMAP GPIO code then tries to call disable_irq_wake() for the bank IRQ and you get noisy 'unbalanced IRQ disable' warnings from the generic IRQ layere because of the attempt to disable the IRQ wake of an IRQ that was never enabled.

So the options that I see are:

1) fix the OMAP GPIO code to check the return code of the parent enable, or
2) drop the parent enable/disable

Quoted 8 lines
Quoted 5 lines

I prefer (1) since those calls will always fail AFAICT.

Does that make any sense?

If you're OK with (1), I will re-issue the patch with a better discription.

Ignoring this for now, please let me know if you want me to queue this for omap-upstream with the updates mentioned above.

OK. I'll re-send with updated description if above is OK with Dave.

Kevin