arm/arm64: GICv2 driver does not have irq_disable implemented

arm/arm64: GICv2 driver does not have irq_disable implemented

[Thomas Gleixner]

On Fri, 9 Oct 2015, Duc Dang wrote:
> Hi Thomas, Marc, Jason and All,
> 
> In APM ARM64 X-Gene Enet controller driver, we use disable_irq_nosync to
> disable interrupt before calling __napi_schedule to schedule packet handler
> to process the Tx/Rx packets.

Which is wrong to begin with. Disable the interrupt at the device
level not at the interrupt line level.
 
Thanks,

	tglx

arm/arm64: GICv2 driver does not have irq_disable implemented

[Duc Dang]

On Fri, Oct 9, 2015 at 10:52 AM, Thomas Gleixner <tglx@linutronix.de> wrote:
> On Fri, 9 Oct 2015, Duc Dang wrote:
>> Hi Thomas, Marc, Jason and All,
>>
>> In APM ARM64 X-Gene Enet controller driver, we use disable_irq_nosync to
>> disable interrupt before calling __napi_schedule to schedule packet handler
>> to process the Tx/Rx packets.
>
> Which is wrong to begin with. Disable the interrupt at the device
> level not at the interrupt line level.
>
Hi Thomas,

We could not disable the interrupt at Enet controller level due to the
controller limitation. As you said, using  disable_irq_nosync is wrong
but it looks like that the only option that we have.

Do you have any suggestion about different approach that we could try?

> Thanks,
>
>         tglx

Regards,
Duc Dang.

arm/arm64: GICv2 driver does not have irq_disable implemented

[Thomas Gleixner]

On Fri, 9 Oct 2015, Duc Dang wrote:
> On Fri, Oct 9, 2015 at 10:52 AM, Thomas Gleixner <tglx@linutronix.de> wrote:
> > On Fri, 9 Oct 2015, Duc Dang wrote:
> >> In APM ARM64 X-Gene Enet controller driver, we use disable_irq_nosync to
> >> disable interrupt before calling __napi_schedule to schedule packet handler
> >> to process the Tx/Rx packets.
> >
> > Which is wrong to begin with. Disable the interrupt at the device
> > level not at the interrupt line level.
> >
> We could not disable the interrupt at Enet controller level due to the
> controller limitation. As you said, using  disable_irq_nosync is wrong
> but it looks like that the only option that we have.

Oh well.
 
> Do you have any suggestion about different approach that we could try?

Try the patch below and add 

    irq_set_status_flags(irq, IRQ_DISABLE_UNLAZY);

to your driver before requesting the interrupt. Unset it when freeing
the interrupt.

Thanks,

	tglx

8<--------------

Subject: genirq: Add flag to force mask in disable_irq[_nosync]()
From: Thomas Gleixner <tglx@linutronix.de>
Date: Fri, 09 Oct 2015 23:28:58 +0200

If an irq chip does not implement the irq_disable callback, then we
use a lazy approach for disabling the interrupt. That means that the
interrupt is marked disabled, but the interrupt line is not
immediately masked in the interrupt chip. It only becomes masked if
the interrupt is raised while it's marked disabled. We use this to avoid
possibly expensive mask/unmask operations for common case operations.

Unfortunately there are devices which do not allow the interrupt to be
disabled easily at the device level. They are forced to use
disable_irq_nosync(). This can result in taking each interrupt twice.

Instead of enforcing the non lazy mode on all interrupts of a irq
chip, provide a settings flag, which can be set by the driver for that
particular interrupt line.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
 include/linux/irq.h   |    4 +++-
 kernel/irq/chip.c     |    9 +++++++++
 kernel/irq/settings.h |    7 +++++++
 3 files changed, 19 insertions(+), 1 deletion(-)

Index: tip/include/linux/irq.h
===================================================================
--- tip.orig/include/linux/irq.h
+++ tip/include/linux/irq.h
@@ -72,6 +72,7 @@ enum irqchip_irq_state;
  * IRQ_IS_POLLED		- Always polled by another interrupt. Exclude
  *				  it from the spurious interrupt detection
  *				  mechanism and from core side polling.
+ * IRQ_DISABLE_UNLAZY		- Disable lazy irq disable
  */
 enum {
 	IRQ_TYPE_NONE		= 0x00000000,
@@ -97,13 +98,14 @@ enum {
 	IRQ_NOTHREAD		= (1 << 16),
 	IRQ_PER_CPU_DEVID	= (1 << 17),
 	IRQ_IS_POLLED		= (1 << 18),
+	IRQ_DISABLE_UNLAZY	= (1 << 19),
 };
 
 #define IRQF_MODIFY_MASK	\
 	(IRQ_TYPE_SENSE_MASK | IRQ_NOPROBE | IRQ_NOREQUEST | \
 	 IRQ_NOAUTOEN | IRQ_MOVE_PCNTXT | IRQ_LEVEL | IRQ_NO_BALANCING | \
 	 IRQ_PER_CPU | IRQ_NESTED_THREAD | IRQ_NOTHREAD | IRQ_PER_CPU_DEVID | \
-	 IRQ_IS_POLLED)
+	 IRQ_IS_POLLED | IRQ_DISABLE_UNLAZY)
 
 #define IRQ_NO_BALANCING_MASK	(IRQ_PER_CPU | IRQ_NO_BALANCING)
 
Index: tip/kernel/irq/chip.c
===================================================================
--- tip.orig/kernel/irq/chip.c
+++ tip/kernel/irq/chip.c
@@ -241,6 +241,13 @@ void irq_enable(struct irq_desc *desc)
  * disabled. If an interrupt happens, then the interrupt flow
  * handler masks the line at the hardware level and marks it
  * pending.
+ *
+ * If the interrupt chip does not implement the irq_disable callback,
+ * a driver can disable the lazy approach for a particular irq line by
+ * calling 'irq_set_status_flags(irq, IRQ_DISABLE_UNLAZY)'. This can be
+ * used for devices which cannot disable the interrupt@the device
+ * level under certain circumstances and have to use
+ * disable_irq[_nosync] instead.
  */
 void irq_disable(struct irq_desc *desc)
 {
@@ -248,6 +255,8 @@ void irq_disable(struct irq_desc *desc)
 	if (desc->irq_data.chip->irq_disable) {
 		desc->irq_data.chip->irq_disable(&desc->irq_data);
 		irq_state_set_masked(desc);
+	} else if (irq_settings_disable_unlazy(desc)) {
+		mask_irq(desc);
 	}
 }
 
Index: tip/kernel/irq/settings.h
===================================================================
--- tip.orig/kernel/irq/settings.h
+++ tip/kernel/irq/settings.h
@@ -15,6 +15,7 @@ enum {
 	_IRQ_NESTED_THREAD	= IRQ_NESTED_THREAD,
 	_IRQ_PER_CPU_DEVID	= IRQ_PER_CPU_DEVID,
 	_IRQ_IS_POLLED		= IRQ_IS_POLLED,
+	_IRQ_DISABLE_UNLAZY	= IRQ_DISABLE_UNLAZY,
 	_IRQF_MODIFY_MASK	= IRQF_MODIFY_MASK,
 };
 
@@ -28,6 +29,7 @@ enum {
 #define IRQ_NESTED_THREAD	GOT_YOU_MORON
 #define IRQ_PER_CPU_DEVID	GOT_YOU_MORON
 #define IRQ_IS_POLLED		GOT_YOU_MORON
+#define IRQ_DISABLE_UNLAZY	GOT_YOU_MORON
 #undef IRQF_MODIFY_MASK
 #define IRQF_MODIFY_MASK	GOT_YOU_MORON
 
@@ -154,3 +156,8 @@ static inline bool irq_settings_is_polle
 {
 	return desc->status_use_accessors & _IRQ_IS_POLLED;
 }
+
+static inline bool irq_settings_disable_unlazy(struct irq_desc *desc)
+{
+	return desc->status_use_accessors & _IRQ_DISABLE_UNLAZY;
+}

arm/arm64: GICv2 driver does not have irq_disable implemented

[Duc Dang]

On Fri, Oct 9, 2015 at 2:52 PM, Thomas Gleixner <tglx@linutronix.de> wrote:
> On Fri, 9 Oct 2015, Duc Dang wrote:
>> On Fri, Oct 9, 2015 at 10:52 AM, Thomas Gleixner <tglx@linutronix.de> wrote:
>> > On Fri, 9 Oct 2015, Duc Dang wrote:
>> >> In APM ARM64 X-Gene Enet controller driver, we use disable_irq_nosync to
>> >> disable interrupt before calling __napi_schedule to schedule packet handler
>> >> to process the Tx/Rx packets.
>> >
>> > Which is wrong to begin with. Disable the interrupt at the device
>> > level not at the interrupt line level.
>> >
>> We could not disable the interrupt at Enet controller level due to the
>> controller limitation. As you said, using  disable_irq_nosync is wrong
>> but it looks like that the only option that we have.
>
> Oh well.
>
>> Do you have any suggestion about different approach that we could try?
>
> Try the patch below and add
>
>     irq_set_status_flags(irq, IRQ_DISABLE_UNLAZY);
>
> to your driver before requesting the interrupt. Unset it when freeing
> the interrupt.

Thanks, Thomas!

We will try and let you know soon.

>
> Thanks,
>
>         tglx
>
> 8<--------------
>
> Subject: genirq: Add flag to force mask in disable_irq[_nosync]()
> From: Thomas Gleixner <tglx@linutronix.de>
> Date: Fri, 09 Oct 2015 23:28:58 +0200
>
> If an irq chip does not implement the irq_disable callback, then we
> use a lazy approach for disabling the interrupt. That means that the
> interrupt is marked disabled, but the interrupt line is not
> immediately masked in the interrupt chip. It only becomes masked if
> the interrupt is raised while it's marked disabled. We use this to avoid
> possibly expensive mask/unmask operations for common case operations.
>
> Unfortunately there are devices which do not allow the interrupt to be
> disabled easily at the device level. They are forced to use
> disable_irq_nosync(). This can result in taking each interrupt twice.
>
> Instead of enforcing the non lazy mode on all interrupts of a irq
> chip, provide a settings flag, which can be set by the driver for that
> particular interrupt line.
>
> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
> ---
>  include/linux/irq.h   |    4 +++-
>  kernel/irq/chip.c     |    9 +++++++++
>  kernel/irq/settings.h |    7 +++++++
>  3 files changed, 19 insertions(+), 1 deletion(-)
>
> Index: tip/include/linux/irq.h
> ===================================================================
> --- tip.orig/include/linux/irq.h
> +++ tip/include/linux/irq.h
> @@ -72,6 +72,7 @@ enum irqchip_irq_state;
>   * IRQ_IS_POLLED               - Always polled by another interrupt. Exclude
>   *                               it from the spurious interrupt detection
>   *                               mechanism and from core side polling.
> + * IRQ_DISABLE_UNLAZY          - Disable lazy irq disable
>   */
>  enum {
>         IRQ_TYPE_NONE           = 0x00000000,
> @@ -97,13 +98,14 @@ enum {
>         IRQ_NOTHREAD            = (1 << 16),
>         IRQ_PER_CPU_DEVID       = (1 << 17),
>         IRQ_IS_POLLED           = (1 << 18),
> +       IRQ_DISABLE_UNLAZY      = (1 << 19),
>  };
>
>  #define IRQF_MODIFY_MASK       \
>         (IRQ_TYPE_SENSE_MASK | IRQ_NOPROBE | IRQ_NOREQUEST | \
>          IRQ_NOAUTOEN | IRQ_MOVE_PCNTXT | IRQ_LEVEL | IRQ_NO_BALANCING | \
>          IRQ_PER_CPU | IRQ_NESTED_THREAD | IRQ_NOTHREAD | IRQ_PER_CPU_DEVID | \
> -        IRQ_IS_POLLED)
> +        IRQ_IS_POLLED | IRQ_DISABLE_UNLAZY)
>
>  #define IRQ_NO_BALANCING_MASK  (IRQ_PER_CPU | IRQ_NO_BALANCING)
>
> Index: tip/kernel/irq/chip.c
> ===================================================================
> --- tip.orig/kernel/irq/chip.c
> +++ tip/kernel/irq/chip.c
> @@ -241,6 +241,13 @@ void irq_enable(struct irq_desc *desc)
>   * disabled. If an interrupt happens, then the interrupt flow
>   * handler masks the line at the hardware level and marks it
>   * pending.
> + *
> + * If the interrupt chip does not implement the irq_disable callback,
> + * a driver can disable the lazy approach for a particular irq line by
> + * calling 'irq_set_status_flags(irq, IRQ_DISABLE_UNLAZY)'. This can be
> + * used for devices which cannot disable the interrupt at the device
> + * level under certain circumstances and have to use
> + * disable_irq[_nosync] instead.
>   */
>  void irq_disable(struct irq_desc *desc)
>  {
> @@ -248,6 +255,8 @@ void irq_disable(struct irq_desc *desc)
>         if (desc->irq_data.chip->irq_disable) {
>                 desc->irq_data.chip->irq_disable(&desc->irq_data);
>                 irq_state_set_masked(desc);
> +       } else if (irq_settings_disable_unlazy(desc)) {
> +               mask_irq(desc);
>         }
>  }
>
> Index: tip/kernel/irq/settings.h
> ===================================================================
> --- tip.orig/kernel/irq/settings.h
> +++ tip/kernel/irq/settings.h
> @@ -15,6 +15,7 @@ enum {
>         _IRQ_NESTED_THREAD      = IRQ_NESTED_THREAD,
>         _IRQ_PER_CPU_DEVID      = IRQ_PER_CPU_DEVID,
>         _IRQ_IS_POLLED          = IRQ_IS_POLLED,
> +       _IRQ_DISABLE_UNLAZY     = IRQ_DISABLE_UNLAZY,
>         _IRQF_MODIFY_MASK       = IRQF_MODIFY_MASK,
>  };
>
> @@ -28,6 +29,7 @@ enum {
>  #define IRQ_NESTED_THREAD      GOT_YOU_MORON
>  #define IRQ_PER_CPU_DEVID      GOT_YOU_MORON
>  #define IRQ_IS_POLLED          GOT_YOU_MORON
> +#define IRQ_DISABLE_UNLAZY     GOT_YOU_MORON
>  #undef IRQF_MODIFY_MASK
>  #define IRQF_MODIFY_MASK       GOT_YOU_MORON
>
> @@ -154,3 +156,8 @@ static inline bool irq_settings_is_polle
>  {
>         return desc->status_use_accessors & _IRQ_IS_POLLED;
>  }
> +
> +static inline bool irq_settings_disable_unlazy(struct irq_desc *desc)
> +{
> +       return desc->status_use_accessors & _IRQ_DISABLE_UNLAZY;
> +}

Regards,
Duc Dang.

arm/arm64: GICv2 driver does not have irq_disable implemented

[Duc Dang]

On Fri, Oct 9, 2015 at 3:21 PM, Duc Dang <dhdang@apm.com> wrote:
> On Fri, Oct 9, 2015 at 2:52 PM, Thomas Gleixner <tglx@linutronix.de> wrote:
>> On Fri, 9 Oct 2015, Duc Dang wrote:
>>> On Fri, Oct 9, 2015 at 10:52 AM, Thomas Gleixner <tglx@linutronix.de> wrote:
>>> > On Fri, 9 Oct 2015, Duc Dang wrote:
>>> >> In APM ARM64 X-Gene Enet controller driver, we use disable_irq_nosync to
>>> >> disable interrupt before calling __napi_schedule to schedule packet handler
>>> >> to process the Tx/Rx packets.
>>> >
>>> > Which is wrong to begin with. Disable the interrupt at the device
>>> > level not at the interrupt line level.
>>> >
>>> We could not disable the interrupt at Enet controller level due to the
>>> controller limitation. As you said, using  disable_irq_nosync is wrong
>>> but it looks like that the only option that we have.
>>
>> Oh well.
>>
>>> Do you have any suggestion about different approach that we could try?
>>
>> Try the patch below and add
>>
>>     irq_set_status_flags(irq, IRQ_DISABLE_UNLAZY);
>>
>> to your driver before requesting the interrupt. Unset it when freeing
>> the interrupt.
>
> Thanks, Thomas!
>
> We will try and let you know soon.

Hi  Thomas,

We use irq_set_status_flags(irq, IRQ_DISABLE_UNLAZY) in our X-Gene
Ethernet driver along with your patch and interrupt count works as
expected.

Are you going to commit your patch soon?

We will post the patch for our X-Gene Ethernet driver after your patch
is available.

Thanks,
>
>>
>> Thanks,
>>
>>         tglx
>>
>> 8<--------------
>>
>> Subject: genirq: Add flag to force mask in disable_irq[_nosync]()
>> From: Thomas Gleixner <tglx@linutronix.de>
>> Date: Fri, 09 Oct 2015 23:28:58 +0200
>>
>> If an irq chip does not implement the irq_disable callback, then we
>> use a lazy approach for disabling the interrupt. That means that the
>> interrupt is marked disabled, but the interrupt line is not
>> immediately masked in the interrupt chip. It only becomes masked if
>> the interrupt is raised while it's marked disabled. We use this to avoid
>> possibly expensive mask/unmask operations for common case operations.
>>
>> Unfortunately there are devices which do not allow the interrupt to be
>> disabled easily at the device level. They are forced to use
>> disable_irq_nosync(). This can result in taking each interrupt twice.
>>
>> Instead of enforcing the non lazy mode on all interrupts of a irq
>> chip, provide a settings flag, which can be set by the driver for that
>> particular interrupt line.
>>
>> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
>> ---
>>  include/linux/irq.h   |    4 +++-
>>  kernel/irq/chip.c     |    9 +++++++++
>>  kernel/irq/settings.h |    7 +++++++
>>  3 files changed, 19 insertions(+), 1 deletion(-)
>>
>> Index: tip/include/linux/irq.h
>> ===================================================================
>> --- tip.orig/include/linux/irq.h
>> +++ tip/include/linux/irq.h
>> @@ -72,6 +72,7 @@ enum irqchip_irq_state;
>>   * IRQ_IS_POLLED               - Always polled by another interrupt. Exclude
>>   *                               it from the spurious interrupt detection
>>   *                               mechanism and from core side polling.
>> + * IRQ_DISABLE_UNLAZY          - Disable lazy irq disable
>>   */
>>  enum {
>>         IRQ_TYPE_NONE           = 0x00000000,
>> @@ -97,13 +98,14 @@ enum {
>>         IRQ_NOTHREAD            = (1 << 16),
>>         IRQ_PER_CPU_DEVID       = (1 << 17),
>>         IRQ_IS_POLLED           = (1 << 18),
>> +       IRQ_DISABLE_UNLAZY      = (1 << 19),
>>  };
>>
>>  #define IRQF_MODIFY_MASK       \
>>         (IRQ_TYPE_SENSE_MASK | IRQ_NOPROBE | IRQ_NOREQUEST | \
>>          IRQ_NOAUTOEN | IRQ_MOVE_PCNTXT | IRQ_LEVEL | IRQ_NO_BALANCING | \
>>          IRQ_PER_CPU | IRQ_NESTED_THREAD | IRQ_NOTHREAD | IRQ_PER_CPU_DEVID | \
>> -        IRQ_IS_POLLED)
>> +        IRQ_IS_POLLED | IRQ_DISABLE_UNLAZY)
>>
>>  #define IRQ_NO_BALANCING_MASK  (IRQ_PER_CPU | IRQ_NO_BALANCING)
>>
>> Index: tip/kernel/irq/chip.c
>> ===================================================================
>> --- tip.orig/kernel/irq/chip.c
>> +++ tip/kernel/irq/chip.c
>> @@ -241,6 +241,13 @@ void irq_enable(struct irq_desc *desc)
>>   * disabled. If an interrupt happens, then the interrupt flow
>>   * handler masks the line at the hardware level and marks it
>>   * pending.
>> + *
>> + * If the interrupt chip does not implement the irq_disable callback,
>> + * a driver can disable the lazy approach for a particular irq line by
>> + * calling 'irq_set_status_flags(irq, IRQ_DISABLE_UNLAZY)'. This can be
>> + * used for devices which cannot disable the interrupt at the device
>> + * level under certain circumstances and have to use
>> + * disable_irq[_nosync] instead.
>>   */
>>  void irq_disable(struct irq_desc *desc)
>>  {
>> @@ -248,6 +255,8 @@ void irq_disable(struct irq_desc *desc)
>>         if (desc->irq_data.chip->irq_disable) {
>>                 desc->irq_data.chip->irq_disable(&desc->irq_data);
>>                 irq_state_set_masked(desc);
>> +       } else if (irq_settings_disable_unlazy(desc)) {
>> +               mask_irq(desc);
>>         }
>>  }
>>
>> Index: tip/kernel/irq/settings.h
>> ===================================================================
>> --- tip.orig/kernel/irq/settings.h
>> +++ tip/kernel/irq/settings.h
>> @@ -15,6 +15,7 @@ enum {
>>         _IRQ_NESTED_THREAD      = IRQ_NESTED_THREAD,
>>         _IRQ_PER_CPU_DEVID      = IRQ_PER_CPU_DEVID,
>>         _IRQ_IS_POLLED          = IRQ_IS_POLLED,
>> +       _IRQ_DISABLE_UNLAZY     = IRQ_DISABLE_UNLAZY,
>>         _IRQF_MODIFY_MASK       = IRQF_MODIFY_MASK,
>>  };
>>
>> @@ -28,6 +29,7 @@ enum {
>>  #define IRQ_NESTED_THREAD      GOT_YOU_MORON
>>  #define IRQ_PER_CPU_DEVID      GOT_YOU_MORON
>>  #define IRQ_IS_POLLED          GOT_YOU_MORON
>> +#define IRQ_DISABLE_UNLAZY     GOT_YOU_MORON
>>  #undef IRQF_MODIFY_MASK
>>  #define IRQF_MODIFY_MASK       GOT_YOU_MORON
>>
>> @@ -154,3 +156,8 @@ static inline bool irq_settings_is_polle
>>  {
>>         return desc->status_use_accessors & _IRQ_IS_POLLED;
>>  }
>> +
>> +static inline bool irq_settings_disable_unlazy(struct irq_desc *desc)
>> +{
>> +       return desc->status_use_accessors & _IRQ_DISABLE_UNLAZY;
>> +}
>
> Regards,
> Duc Dang.

Duc Dang.