[PATCH RFC] irqchip: gic-v3: Support forced cpu irq affinity setting

[PATCH RFC] irqchip: gic-v3: Support forced cpu irq affinity setting

[Alim Akhtar]

Skip check for online cpu mask when setting irq affinity of per cpu timer
during early startup of not yet online cpu.

Signed-off-by: Shaik Ameer Basha <shaik.ameer@samsung.com>
Signed-off-by: Aswani Reddy <aswani.reddy@samsung.com>
Signed-off-by: Alim Akhtar <alim.akhtar@samsung.com>
---
 drivers/irqchip/irq-gic-v3.c | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)

Currently when non boot CPUs comes up and if setting of IRQ affinity gets called
before cpu online status gets updated, gic_mpidr_to_affinity() returns
undesirable cpu number. Using __force__ flag ignore cpu_online_mask and setup
the intended cpu affinity for the irq.

The same approch is used in gicv2 
(commit: ffde1de64012c: irqchip: Gic: Support forced affinity setting)

diff --git a/drivers/irqchip/irq-gic-v3.c b/drivers/irqchip/irq-gic-v3.c
index dbffb7a..c238c64 100644
--- a/drivers/irqchip/irq-gic-v3.c
+++ b/drivers/irqchip/irq-gic-v3.c
@@ -640,11 +640,16 @@ static void gic_smp_init(void)
 static int gic_set_affinity(struct irq_data *d, const struct cpumask *mask_val,
 			    bool force)
 {
-	unsigned int cpu = cpumask_any_and(mask_val, cpu_online_mask);
+	unsigned int cpu;
 	void __iomem *reg;
 	int enabled;
 	u64 val;
 
+	if (!force)
+		cpu = cpumask_any_and(mask_val, cpu_online_mask);
+	else
+		cpu = cpumask_first(mask_val);
+
 	if (cpu >= nr_cpu_ids)
 		return -EINVAL;
 
-- 
2.7.4

Re: [PATCH RFC] irqchip: gic-v3: Support forced cpu irq affinity setting

[Marc Zyngier]

On 08/08/17 10:04, Alim Akhtar wrote:
> Skip check for online cpu mask when setting irq affinity of per cpu timer
> during early startup of not yet online cpu.
> 
> Signed-off-by: Shaik Ameer Basha <shaik.ameer@samsung.com>
> Signed-off-by: Aswani Reddy <aswani.reddy@samsung.com>
> Signed-off-by: Alim Akhtar <alim.akhtar@samsung.com>
> ---
>  drivers/irqchip/irq-gic-v3.c | 7 ++++++-
>  1 file changed, 6 insertions(+), 1 deletion(-)
> 
> Currently when non boot CPUs comes up and if setting of IRQ affinity gets called
> before cpu online status gets updated, gic_mpidr_to_affinity() returns
> undesirable cpu number. Using __force__ flag ignore cpu_online_mask and setup
> the intended cpu affinity for the irq.
> 
> The same approch is used in gicv2 
> (commit: ffde1de64012c: irqchip: Gic: Support forced affinity setting)
> 
> diff --git a/drivers/irqchip/irq-gic-v3.c b/drivers/irqchip/irq-gic-v3.c
> index dbffb7a..c238c64 100644
> --- a/drivers/irqchip/irq-gic-v3.c
> +++ b/drivers/irqchip/irq-gic-v3.c
> @@ -640,11 +640,16 @@ static void gic_smp_init(void)
>  static int gic_set_affinity(struct irq_data *d, const struct cpumask *mask_val,
>  			    bool force)
>  {
> -	unsigned int cpu = cpumask_any_and(mask_val, cpu_online_mask);
> +	unsigned int cpu;
>  	void __iomem *reg;
>  	int enabled;
>  	u64 val;
>  
> +	if (!force)
> +		cpu = cpumask_any_and(mask_val, cpu_online_mask);
> +	else
> +		cpu = cpumask_first(mask_val);
> +
>  	if (cpu >= nr_cpu_ids)
>  		return -EINVAL;
>  
> 

See

https://git.kernel.org/pub/scm/linux/kernel/git/maz/arm-platforms.git/commit/?h=irq/irqchip-4.13&id=65a30f8b300107266f316d550f060ccc186201a3

	M.
-- 
Jazz is not dead. It just smells funny...

Re: [PATCH RFC] irqchip: gic-v3: Support forced cpu irq affinity setting

[Alim Akhtar]

On 08/08/2017 02:52 PM, Marc Zyngier wrote:
> On 08/08/17 10:04, Alim Akhtar wrote:
>> Skip check for online cpu mask when setting irq affinity of per cpu timer
>> during early startup of not yet online cpu.
>>
>> Signed-off-by: Shaik Ameer Basha <shaik.ameer@samsung.com>
>> Signed-off-by: Aswani Reddy <aswani.reddy@samsung.com>
>> Signed-off-by: Alim Akhtar <alim.akhtar@samsung.com>
>> ---
>>   drivers/irqchip/irq-gic-v3.c | 7 ++++++-
>>   1 file changed, 6 insertions(+), 1 deletion(-)
>>
>> Currently when non boot CPUs comes up and if setting of IRQ affinity gets called
>> before cpu online status gets updated, gic_mpidr_to_affinity() returns
>> undesirable cpu number. Using __force__ flag ignore cpu_online_mask and setup
>> the intended cpu affinity for the irq.
>>
>> The same approch is used in gicv2
>> (commit: ffde1de64012c: irqchip: Gic: Support forced affinity setting)
>>
>> diff --git a/drivers/irqchip/irq-gic-v3.c b/drivers/irqchip/irq-gic-v3.c
>> index dbffb7a..c238c64 100644
>> --- a/drivers/irqchip/irq-gic-v3.c
>> +++ b/drivers/irqchip/irq-gic-v3.c
>> @@ -640,11 +640,16 @@ static void gic_smp_init(void)
>>   static int gic_set_affinity(struct irq_data *d, const struct cpumask *mask_val,
>>   			    bool force)
>>   {
>> -	unsigned int cpu = cpumask_any_and(mask_val, cpu_online_mask);
>> +	unsigned int cpu;
>>   	void __iomem *reg;
>>   	int enabled;
>>   	u64 val;
>>   
>> +	if (!force)
>> +		cpu = cpumask_any_and(mask_val, cpu_online_mask);
>> +	else
>> +		cpu = cpumask_first(mask_val);
>> +
>>   	if (cpu >= nr_cpu_ids)
>>   		return -EINVAL;
>>   
>>
> 
> See
> 
> https://git.kernel.org/pub/scm/linux/kernel/git/maz/arm-platforms.git/commit/?h=irq/irqchip-4.13&id=65a30f8b300107266f316d550f060ccc186201a3
> 

Thanks Marc for above link. Is this already queued in some tree? I can't 
find this is linux-next.

> 	M.
> 

Re: [PATCH RFC] irqchip: gic-v3: Support forced cpu irq affinity setting

[Marc Zyngier]

On 08/08/17 10:27, Alim Akhtar wrote:
> 
> 
> On 08/08/2017 02:52 PM, Marc Zyngier wrote:
>> On 08/08/17 10:04, Alim Akhtar wrote:
>>> Skip check for online cpu mask when setting irq affinity of per cpu timer
>>> during early startup of not yet online cpu.
>>>
>>> Signed-off-by: Shaik Ameer Basha <shaik.ameer@samsung.com>
>>> Signed-off-by: Aswani Reddy <aswani.reddy@samsung.com>
>>> Signed-off-by: Alim Akhtar <alim.akhtar@samsung.com>
>>> ---
>>>   drivers/irqchip/irq-gic-v3.c | 7 ++++++-
>>>   1 file changed, 6 insertions(+), 1 deletion(-)
>>>
>>> Currently when non boot CPUs comes up and if setting of IRQ affinity gets called
>>> before cpu online status gets updated, gic_mpidr_to_affinity() returns
>>> undesirable cpu number. Using __force__ flag ignore cpu_online_mask and setup
>>> the intended cpu affinity for the irq.
>>>
>>> The same approch is used in gicv2
>>> (commit: ffde1de64012c: irqchip: Gic: Support forced affinity setting)
>>>
>>> diff --git a/drivers/irqchip/irq-gic-v3.c b/drivers/irqchip/irq-gic-v3.c
>>> index dbffb7a..c238c64 100644
>>> --- a/drivers/irqchip/irq-gic-v3.c
>>> +++ b/drivers/irqchip/irq-gic-v3.c
>>> @@ -640,11 +640,16 @@ static void gic_smp_init(void)
>>>   static int gic_set_affinity(struct irq_data *d, const struct cpumask *mask_val,
>>>   			    bool force)
>>>   {
>>> -	unsigned int cpu = cpumask_any_and(mask_val, cpu_online_mask);
>>> +	unsigned int cpu;
>>>   	void __iomem *reg;
>>>   	int enabled;
>>>   	u64 val;
>>>   
>>> +	if (!force)
>>> +		cpu = cpumask_any_and(mask_val, cpu_online_mask);
>>> +	else
>>> +		cpu = cpumask_first(mask_val);
>>> +
>>>   	if (cpu >= nr_cpu_ids)
>>>   		return -EINVAL;
>>>   
>>>
>>
>> See
>>
>> https://git.kernel.org/pub/scm/linux/kernel/git/maz/arm-platforms.git/commit/?h=irq/irqchip-4.13&id=65a30f8b300107266f316d550f060ccc186201a3
>>
> 
> Thanks Marc for above link. Is this already queued in some tree? I can't 
> find this is linux-next.

It is queued as a fix for 4.13, bug tglx is offline for the time being,
so this has not hit -next yet.

Thanks,

	M.
-- 
Jazz is not dead. It just smells funny...