[RFC PATCH 0/2] irqchip: irq-ti-sci-inta: Add IRQ affinity support

[RFC PATCH 0/2] irqchip: irq-ti-sci-inta: Add IRQ affinity support

[Vignesh Raghavendra]

Interrupt Aggregator (INTA) IP INTA on TI's K3 SoCs convert DMA global
events (MSI like) to wired interrupts (VINT). 64 events can be mapped to
single VINT. Currently driver maps multiple events to single wired
interrupt line.  This makes setting IRQ affinity impossible as
migrating wired interrupt to different core will end up migrating all
events to that core.  And since DMA events related to networking IPs and
other high IRQ load IPs are behind this INTA logic, it creates load on a
single CPU, thus limiting overall performance of these peripherals

This series rewrites events to VINT mapping logic to have only one event
mapped to VINT as much as possible. It falls back to aggregating events
when run of unique VINTS.  In most systems there are enough VINTs to not
need aggregation. Also, if IRQ affinity and hint is set by client
driver, such VINTs are reserved and no more events are mapped to those
VINTs.

This allows to implement IRQ affinity callback as simple propagation to
affinity setting to parent irqchip (GIC).

Vignesh Raghavendra (2):
  irqchip: irq-ti-sci-inta: Don't aggregate MSI events until necessary
  irqchip: irq-ti-sci-inta: Introduce IRQ affinity support

 drivers/irqchip/irq-ti-sci-inta.c | 84 +++++++++++++++++++++++--------
 1 file changed, 64 insertions(+), 20 deletions(-)

-- 
2.39.0


_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

[RFC PATCH 1/2] irqchip: irq-ti-sci-inta: Don't aggregate MSI events until necessary

[Vignesh Raghavendra]

INTA on K3 SoCs convert DMA global events (MSI) to wired interrupts
(VINT). Currently driver maps multiple events to single wired interrupt
lines.  This makes setting IRQ affinity impossible as migrating wired
interrupt to different core will end up migrating all events to that
core.

Allocate a dedicated VINT for each DMA event request until there are no
more VINTs. This creates a 1:1 DMA event to VINT mapping and thus
provides unique IRQ line. This will help allocate dedicated IRQs for
high performance DMA channels which can thus be mapped to particular
CPUs using IRQ affinity.

Signed-off-by: Vignesh Raghavendra <vigneshr@ti.com>
---
 drivers/irqchip/irq-ti-sci-inta.c | 45 +++++++++++++++++--------------
 1 file changed, 25 insertions(+), 20 deletions(-)

diff --git a/drivers/irqchip/irq-ti-sci-inta.c b/drivers/irqchip/irq-ti-sci-inta.c
index a6ecc53d055c..f1419d24568e 100644
--- a/drivers/irqchip/irq-ti-sci-inta.c
+++ b/drivers/irqchip/irq-ti-sci-inta.c
@@ -202,7 +202,8 @@ static int ti_sci_inta_xlate_irq(struct ti_sci_inta_irq_domain *inta,
  *
  * Return 0 if all went well else corresponding error value.
  */
-static struct ti_sci_inta_vint_desc *ti_sci_inta_alloc_parent_irq(struct irq_domain *domain)
+static struct ti_sci_inta_vint_desc *ti_sci_inta_alloc_parent_irq(struct irq_domain *domain,
+								  u16 vint_id)
 {
 	struct ti_sci_inta_irq_domain *inta = domain->host_data;
 	struct ti_sci_inta_vint_desc *vint_desc;
@@ -210,11 +211,6 @@ static struct ti_sci_inta_vint_desc *ti_sci_inta_alloc_parent_irq(struct irq_dom
 	struct device_node *parent_node;
 	unsigned int parent_virq;
 	int p_hwirq, ret;
-	u16 vint_id;
-
-	vint_id = ti_sci_get_free_resource(inta->vint);
-	if (vint_id == TI_SCI_RESOURCE_NULL)
-		return ERR_PTR(-EINVAL);
 
 	p_hwirq = ti_sci_inta_xlate_irq(inta, vint_id);
 	if (p_hwirq < 0) {
@@ -328,29 +324,38 @@ static struct ti_sci_inta_event_desc *ti_sci_inta_alloc_irq(struct irq_domain *d
 	struct ti_sci_inta_vint_desc *vint_desc = NULL;
 	struct ti_sci_inta_event_desc *event_desc;
 	u16 free_bit;
+	u16 vint_id;
 
 	mutex_lock(&inta->vint_mutex);
-	list_for_each_entry(vint_desc, &inta->vint_list, list) {
+	/*
+	 * Allocate new VINT each time until we runout, then start
+	 * aggregating
+	 */
+	vint_id = ti_sci_get_free_resource(inta->vint);
+	if (vint_id == TI_SCI_RESOURCE_NULL) {
+		list_for_each_entry(vint_desc, &inta->vint_list, list) {
+			free_bit = find_first_zero_bit(vint_desc->event_map,
+						       MAX_EVENTS_PER_VINT);
+			if (free_bit != MAX_EVENTS_PER_VINT)
+				set_bit(free_bit, vint_desc->event_map);
+		}
+	} else  {
+		vint_desc = ti_sci_inta_alloc_parent_irq(domain, vint_id);
+		if (IS_ERR(vint_desc)) {
+			event_desc = ERR_CAST(vint_desc);
+			goto unlock;
+		}
+
 		free_bit = find_first_zero_bit(vint_desc->event_map,
 					       MAX_EVENTS_PER_VINT);
-		if (free_bit != MAX_EVENTS_PER_VINT) {
-			set_bit(free_bit, vint_desc->event_map);
-			goto alloc_event;
-		}
+		set_bit(free_bit, vint_desc->event_map);
 	}
 
-	/* No free bits available. Allocate a new vint */
-	vint_desc = ti_sci_inta_alloc_parent_irq(domain);
-	if (IS_ERR(vint_desc)) {
-		event_desc = ERR_CAST(vint_desc);
+	if (free_bit == MAX_EVENTS_PER_VINT) {
+		event_desc = ERR_PTR(-EINVAL);
 		goto unlock;
 	}
 
-	free_bit = find_first_zero_bit(vint_desc->event_map,
-				       MAX_EVENTS_PER_VINT);
-	set_bit(free_bit, vint_desc->event_map);
-
-alloc_event:
 	event_desc = ti_sci_inta_alloc_event(vint_desc, free_bit, hwirq);
 	if (IS_ERR(event_desc))
 		clear_bit(free_bit, vint_desc->event_map);
-- 
2.39.0


_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

[RFC PATCH 2/2] irqchip: irq-ti-sci-inta: Introduce IRQ affinity support

[Vignesh Raghavendra]

Add support for setting IRQ affinity for VINTs which have only one event
mapped to them. This just involves changing the parent IRQs affinity
(GIC/INTR). Flag VINTs which have affinity configured so as to not
aggregate/map more events to such VINTs.

Signed-off-by: Vignesh Raghavendra <vigneshr@ti.com>
---
 drivers/irqchip/irq-ti-sci-inta.c | 39 +++++++++++++++++++++++++++++++
 1 file changed, 39 insertions(+)

diff --git a/drivers/irqchip/irq-ti-sci-inta.c b/drivers/irqchip/irq-ti-sci-inta.c
index f1419d24568e..237cb4707cb8 100644
--- a/drivers/irqchip/irq-ti-sci-inta.c
+++ b/drivers/irqchip/irq-ti-sci-inta.c
@@ -64,6 +64,7 @@ struct ti_sci_inta_event_desc {
  * @events:		Array of event descriptors assigned to this vint.
  * @parent_virq:	Linux IRQ number that gets attached to parent
  * @vint_id:		TISCI vint ID
+ * @affinity_managed	flag to indicate VINT affinity is managed
  */
 struct ti_sci_inta_vint_desc {
 	struct irq_domain *domain;
@@ -72,6 +73,7 @@ struct ti_sci_inta_vint_desc {
 	struct ti_sci_inta_event_desc events[MAX_EVENTS_PER_VINT];
 	unsigned int parent_virq;
 	u16 vint_id;
+	bool affinity_managed;
 };
 
 /**
@@ -334,6 +336,8 @@ static struct ti_sci_inta_event_desc *ti_sci_inta_alloc_irq(struct irq_domain *d
 	vint_id = ti_sci_get_free_resource(inta->vint);
 	if (vint_id == TI_SCI_RESOURCE_NULL) {
 		list_for_each_entry(vint_desc, &inta->vint_list, list) {
+			if (vint_desc->affinity_managed)
+				continue;
 			free_bit = find_first_zero_bit(vint_desc->event_map,
 						       MAX_EVENTS_PER_VINT);
 			if (free_bit != MAX_EVENTS_PER_VINT)
@@ -434,6 +438,7 @@ static int ti_sci_inta_request_resources(struct irq_data *data)
 		return PTR_ERR(event_desc);
 
 	data->chip_data = event_desc;
+	irq_data_update_effective_affinity(data, cpu_online_mask);
 
 	return 0;
 }
@@ -504,11 +509,45 @@ static void ti_sci_inta_ack_irq(struct irq_data *data)
 		ti_sci_inta_manage_event(data, VINT_STATUS_OFFSET);
 }
 
+#ifdef CONFIG_SMP
+static int ti_sci_inta_set_affinity(struct irq_data *d,
+				    const struct cpumask *mask_val, bool force)
+{
+	struct ti_sci_inta_event_desc *event_desc;
+	struct ti_sci_inta_vint_desc *vint_desc;
+	struct irq_data *parent_irq_data;
+
+	if (cpumask_equal(irq_data_get_effective_affinity_mask(d), mask_val))
+		return 0;
+
+	event_desc = irq_data_get_irq_chip_data(d);
+	if (event_desc) {
+		vint_desc = to_vint_desc(event_desc, event_desc->vint_bit);
+
+		/*
+		 * Cannot set affinity if there is more than one event
+		 * mapped to same VINT
+		 */
+		if (bitmap_weight(vint_desc->event_map, MAX_EVENTS_PER_VINT) > 1)
+			return -EINVAL;
+
+		vint_desc->affinity_managed = true;
+
+		irq_data_update_effective_affinity(d, mask_val);
+		parent_irq_data = irq_get_irq_data(vint_desc->parent_virq);
+		if (parent_irq_data->chip->irq_set_affinity)
+			return parent_irq_data->chip->irq_set_affinity(parent_irq_data, mask_val, force);
+	}
+
+	return -EINVAL;
+}
+#else
 static int ti_sci_inta_set_affinity(struct irq_data *d,
 				    const struct cpumask *mask_val, bool force)
 {
 	return -EINVAL;
 }
+#endif
 
 /**
  * ti_sci_inta_set_type() - Update the trigger type of the irq.
-- 
2.39.0


_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

Re: [RFC PATCH 2/2] irqchip: irq-ti-sci-inta: Introduce IRQ affinity support

[Marc Zyngier]

On Sun, 22 Jan 2023 08:16:07 +0000,
Vignesh Raghavendra <vigneshr@ti.com> wrote:
> 
> Add support for setting IRQ affinity for VINTs which have only one event
> mapped to them. This just involves changing the parent IRQs affinity
> (GIC/INTR). Flag VINTs which have affinity configured so as to not
> aggregate/map more events to such VINTs.



> 
> Signed-off-by: Vignesh Raghavendra <vigneshr@ti.com>
> ---
>  drivers/irqchip/irq-ti-sci-inta.c | 39 +++++++++++++++++++++++++++++++
>  1 file changed, 39 insertions(+)
> 
> diff --git a/drivers/irqchip/irq-ti-sci-inta.c b/drivers/irqchip/irq-ti-sci-inta.c
> index f1419d24568e..237cb4707cb8 100644
> --- a/drivers/irqchip/irq-ti-sci-inta.c
> +++ b/drivers/irqchip/irq-ti-sci-inta.c
> @@ -64,6 +64,7 @@ struct ti_sci_inta_event_desc {
>   * @events:		Array of event descriptors assigned to this vint.
>   * @parent_virq:	Linux IRQ number that gets attached to parent
>   * @vint_id:		TISCI vint ID
> + * @affinity_managed	flag to indicate VINT affinity is managed
>   */
>  struct ti_sci_inta_vint_desc {
>  	struct irq_domain *domain;
> @@ -72,6 +73,7 @@ struct ti_sci_inta_vint_desc {
>  	struct ti_sci_inta_event_desc events[MAX_EVENTS_PER_VINT];
>  	unsigned int parent_virq;
>  	u16 vint_id;
> +	bool affinity_managed;
>  };
>  
>  /**
> @@ -334,6 +336,8 @@ static struct ti_sci_inta_event_desc *ti_sci_inta_alloc_irq(struct irq_domain *d
>  	vint_id = ti_sci_get_free_resource(inta->vint);
>  	if (vint_id == TI_SCI_RESOURCE_NULL) {
>  		list_for_each_entry(vint_desc, &inta->vint_list, list) {
> +			if (vint_desc->affinity_managed)
> +				continue;
>  			free_bit = find_first_zero_bit(vint_desc->event_map,
>  						       MAX_EVENTS_PER_VINT);
>  			if (free_bit != MAX_EVENTS_PER_VINT)
> @@ -434,6 +438,7 @@ static int ti_sci_inta_request_resources(struct irq_data *data)
>  		return PTR_ERR(event_desc);
>  
>  	data->chip_data = event_desc;
> +	irq_data_update_effective_affinity(data, cpu_online_mask);
>  
>  	return 0;
>  }
> @@ -504,11 +509,45 @@ static void ti_sci_inta_ack_irq(struct irq_data *data)
>  		ti_sci_inta_manage_event(data, VINT_STATUS_OFFSET);
>  }
>  
> +#ifdef CONFIG_SMP
> +static int ti_sci_inta_set_affinity(struct irq_data *d,
> +				    const struct cpumask *mask_val, bool force)
> +{
> +	struct ti_sci_inta_event_desc *event_desc;
> +	struct ti_sci_inta_vint_desc *vint_desc;
> +	struct irq_data *parent_irq_data;
> +
> +	if (cpumask_equal(irq_data_get_effective_affinity_mask(d), mask_val))
> +		return 0;
> +
> +	event_desc = irq_data_get_irq_chip_data(d);
> +	if (event_desc) {
> +		vint_desc = to_vint_desc(event_desc, event_desc->vint_bit);
> +
> +		/*
> +		 * Cannot set affinity if there is more than one event
> +		 * mapped to same VINT
> +		 */
> +		if (bitmap_weight(vint_desc->event_map, MAX_EVENTS_PER_VINT) > 1)
> +			return -EINVAL;
> +
> +		vint_desc->affinity_managed = true;
> +
> +		irq_data_update_effective_affinity(d, mask_val);
> +		parent_irq_data = irq_get_irq_data(vint_desc->parent_virq);
> +		if (parent_irq_data->chip->irq_set_affinity)
> +			return parent_irq_data->chip->irq_set_affinity(parent_irq_data, mask_val, force);

This looks completely wrong.

You still have a chained irqchip on all paths, and have to do some
horrible probing to work out:

- which parent interrupt this is

- how many interrupts are connected to it

And then the fun begins:

- You have one interrupt that is standalone, so its affinity can be
  moved

- An unrelated driver gets probed, and one of its interrupts gets
  lumped together with the one above

- Now it cannot be moved anymore, and userspace complains

The rule is very simple: chained irqchip, no affinity management.
Either you reserve a poll of direct interrupts that have affinity
management and no muxing, or you keep the current approach.

But I'm strongly opposed to this sort of approach.

Thanks,

	M.

-- 
Without deviation from the norm, progress is not possible.

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

Re: [RFC PATCH 2/2] irqchip: irq-ti-sci-inta: Introduce IRQ affinity support

[Raghavendra, Vignesh]

Hi Marc,d

On 1/26/2023 7:48 PM, Marc Zyngier wrote:
> On Sun, 22 Jan 2023 08:16:07 +0000,
> Vignesh Raghavendra <vigneshr@ti.com> wrote:
>>
>> Add support for setting IRQ affinity for VINTs which have only one event
>> mapped to them. This just involves changing the parent IRQs affinity
>> (GIC/INTR). Flag VINTs which have affinity configured so as to not
>> aggregate/map more events to such VINTs.
> 
> 
> 
>>
>> Signed-off-by: Vignesh Raghavendra <vigneshr@ti.com>
>> ---
>>  drivers/irqchip/irq-ti-sci-inta.c | 39 +++++++++++++++++++++++++++++++
>>  1 file changed, 39 insertions(+)
>>
>> diff --git a/drivers/irqchip/irq-ti-sci-inta.c b/drivers/irqchip/irq-ti-sci-inta.c
>> index f1419d24568e..237cb4707cb8 100644
>> --- a/drivers/irqchip/irq-ti-sci-inta.c
>> +++ b/drivers/irqchip/irq-ti-sci-inta.c
>> @@ -64,6 +64,7 @@ struct ti_sci_inta_event_desc {
>>   * @events:		Array of event descriptors assigned to this vint.
>>   * @parent_virq:	Linux IRQ number that gets attached to parent
>>   * @vint_id:		TISCI vint ID
>> + * @affinity_managed	flag to indicate VINT affinity is managed
>>   */
>>  struct ti_sci_inta_vint_desc {
>>  	struct irq_domain *domain;
>> @@ -72,6 +73,7 @@ struct ti_sci_inta_vint_desc {
>>  	struct ti_sci_inta_event_desc events[MAX_EVENTS_PER_VINT];
>>  	unsigned int parent_virq;
>>  	u16 vint_id;
>> +	bool affinity_managed;
>>  };
>>  
>>  /**
>> @@ -334,6 +336,8 @@ static struct ti_sci_inta_event_desc *ti_sci_inta_alloc_irq(struct irq_domain *d
>>  	vint_id = ti_sci_get_free_resource(inta->vint);
>>  	if (vint_id == TI_SCI_RESOURCE_NULL) {
>>  		list_for_each_entry(vint_desc, &inta->vint_list, list) {
>> +			if (vint_desc->affinity_managed)
>> +				continue;
>>  			free_bit = find_first_zero_bit(vint_desc->event_map,
>>  						       MAX_EVENTS_PER_VINT);
>>  			if (free_bit != MAX_EVENTS_PER_VINT)
>> @@ -434,6 +438,7 @@ static int ti_sci_inta_request_resources(struct irq_data *data)
>>  		return PTR_ERR(event_desc);
>>  
>>  	data->chip_data = event_desc;
>> +	irq_data_update_effective_affinity(data, cpu_online_mask);
>>  
>>  	return 0;
>>  }
>> @@ -504,11 +509,45 @@ static void ti_sci_inta_ack_irq(struct irq_data *data)
>>  		ti_sci_inta_manage_event(data, VINT_STATUS_OFFSET);
>>  }
>>  
>> +#ifdef CONFIG_SMP
>> +static int ti_sci_inta_set_affinity(struct irq_data *d,
>> +				    const struct cpumask *mask_val, bool force)
>> +{
>> +	struct ti_sci_inta_event_desc *event_desc;
>> +	struct ti_sci_inta_vint_desc *vint_desc;
>> +	struct irq_data *parent_irq_data;
>> +
>> +	if (cpumask_equal(irq_data_get_effective_affinity_mask(d), mask_val))
>> +		return 0;
>> +
>> +	event_desc = irq_data_get_irq_chip_data(d);
>> +	if (event_desc) {
>> +		vint_desc = to_vint_desc(event_desc, event_desc->vint_bit);
>> +
>> +		/*
>> +		 * Cannot set affinity if there is more than one event
>> +		 * mapped to same VINT
>> +		 */
>> +		if (bitmap_weight(vint_desc->event_map, MAX_EVENTS_PER_VINT) > 1)
>> +			return -EINVAL;
>> +
>> +		vint_desc->affinity_managed = true;
>> +
>> +		irq_data_update_effective_affinity(d, mask_val);
>> +		parent_irq_data = irq_get_irq_data(vint_desc->parent_virq);
>> +		if (parent_irq_data->chip->irq_set_affinity)
>> +			return parent_irq_data->chip->irq_set_affinity(parent_irq_data, mask_val, force);
> 
> This looks completely wrong.
> 
> You still have a chained irqchip on all paths, and have to do some
> horrible probing to work out:
> 
> - which parent interrupt this is
> 
> - how many interrupts are connected to it
> 
> And then the fun begins:
> 
> - You have one interrupt that is standalone, so its affinity can be
>   moved
> 
> - An unrelated driver gets probed, and one of its interrupts gets
>   lumped together with the one above
> 
> - Now it cannot be moved anymore, and userspace complains
> 
> The rule is very simple: chained irqchip, no affinity management.
> Either you reserve a poll of direct interrupts that have affinity

This is what I am trying to accomplish, that is, reserve a pool of
direct interrupts that can be used by certain drivers that require IRQ
steering for performance. But I don't see a way to indicate from client
drivers to allocate from this reserved pool (there is no hint in
request_irq() call that ends up in .irq_request_resources() that I can use)

I can try and virtually split INTA into two irqchips perhaps, with one
part modeled as chained irqchip and other as stacked for the reserved
pool (and would have to spawn of two child msi-domains I presume).
But, there is only one DT node for this irqchip and thus clients cannot
request IRQ for reserved pool.

Wondering if you have any pointers here?

Thanks for your patience.

Regards
Vignesh

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

Re: [RFC PATCH 2/2] irqchip: irq-ti-sci-inta: Introduce IRQ affinity support

[Marc Zyngier]

On Fri, 27 Jan 2023 17:53:55 +0000,
"Raghavendra, Vignesh" <vigneshr@ti.com> wrote:
>

[...]

> >> @@ -504,11 +509,45 @@ static void ti_sci_inta_ack_irq(struct irq_data *data)
> >>  		ti_sci_inta_manage_event(data, VINT_STATUS_OFFSET);
> >>  }
> >>  
> >> +#ifdef CONFIG_SMP
> >> +static int ti_sci_inta_set_affinity(struct irq_data *d,
> >> +				    const struct cpumask *mask_val, bool force)
> >> +{
> >> +	struct ti_sci_inta_event_desc *event_desc;
> >> +	struct ti_sci_inta_vint_desc *vint_desc;
> >> +	struct irq_data *parent_irq_data;
> >> +
> >> +	if (cpumask_equal(irq_data_get_effective_affinity_mask(d), mask_val))
> >> +		return 0;
> >> +
> >> +	event_desc = irq_data_get_irq_chip_data(d);
> >> +	if (event_desc) {
> >> +		vint_desc = to_vint_desc(event_desc, event_desc->vint_bit);
> >> +
> >> +		/*
> >> +		 * Cannot set affinity if there is more than one event
> >> +		 * mapped to same VINT
> >> +		 */
> >> +		if (bitmap_weight(vint_desc->event_map, MAX_EVENTS_PER_VINT) > 1)
> >> +			return -EINVAL;
> >> +
> >> +		vint_desc->affinity_managed = true;
> >> +
> >> +		irq_data_update_effective_affinity(d, mask_val);
> >> +		parent_irq_data = irq_get_irq_data(vint_desc->parent_virq);
> >> +		if (parent_irq_data->chip->irq_set_affinity)
> >> +			return parent_irq_data->chip->irq_set_affinity(parent_irq_data, mask_val, force);
> > 
> > This looks completely wrong.
> > 
> > You still have a chained irqchip on all paths, and have to do some
> > horrible probing to work out:
> > 
> > - which parent interrupt this is
> > 
> > - how many interrupts are connected to it
> > 
> > And then the fun begins:
> > 
> > - You have one interrupt that is standalone, so its affinity can be
> >   moved
> > 
> > - An unrelated driver gets probed, and one of its interrupts gets
> >   lumped together with the one above
> > 
> > - Now it cannot be moved anymore, and userspace complains
> > 
> > The rule is very simple: chained irqchip, no affinity management.
> > Either you reserve a poll of direct interrupts that have affinity
> 
> This is what I am trying to accomplish, that is, reserve a pool of
> direct interrupts that can be used by certain drivers that require IRQ
> steering for performance. But I don't see a way to indicate from client
> drivers to allocate from this reserved pool (there is no hint in
> request_irq() call that ends up in .irq_request_resources() that I can use)
> 
> I can try and virtually split INTA into two irqchips perhaps, with one
> part modeled as chained irqchip and other as stacked for the reserved
> pool (and would have to spawn of two child msi-domains I presume).
> But, there is only one DT node for this irqchip and thus clients cannot
> request IRQ for reserved pool.

I don't see why DT should be aware of this. You only need to decide at
allocation time which one is where, and plug it at the right level.

And you probably only need *one* chained interrupt that muxes
everything that cannot be allocated direct path.

> Wondering if you have any pointers here?

I don't. But this shouldn't be too hard to bolt onto the existing
framework.

	M.

-- 
Without deviation from the norm, progress is not possible.

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

Re: [RFC PATCH 2/2] irqchip: irq-ti-sci-inta: Introduce IRQ affinity support

[Raghavendra, Vignesh]

On 2/20/2023 2:17 PM, Marc Zyngier wrote:
> On Fri, 27 Jan 2023 17:53:55 +0000,
> "Raghavendra, Vignesh" <vigneshr@ti.com> wrote:
>>
> 
> [...]
> 
>>>> @@ -504,11 +509,45 @@ static void ti_sci_inta_ack_irq(struct irq_data *data)
>>>>  		ti_sci_inta_manage_event(data, VINT_STATUS_OFFSET);
>>>>  }
>>>>  
>>>> +#ifdef CONFIG_SMP
>>>> +static int ti_sci_inta_set_affinity(struct irq_data *d,
>>>> +				    const struct cpumask *mask_val, bool force)
>>>> +{
>>>> +	struct ti_sci_inta_event_desc *event_desc;
>>>> +	struct ti_sci_inta_vint_desc *vint_desc;
>>>> +	struct irq_data *parent_irq_data;
>>>> +
>>>> +	if (cpumask_equal(irq_data_get_effective_affinity_mask(d), mask_val))
>>>> +		return 0;
>>>> +
>>>> +	event_desc = irq_data_get_irq_chip_data(d);
>>>> +	if (event_desc) {
>>>> +		vint_desc = to_vint_desc(event_desc, event_desc->vint_bit);
>>>> +
>>>> +		/*
>>>> +		 * Cannot set affinity if there is more than one event
>>>> +		 * mapped to same VINT
>>>> +		 */
>>>> +		if (bitmap_weight(vint_desc->event_map, MAX_EVENTS_PER_VINT) > 1)
>>>> +			return -EINVAL;
>>>> +
>>>> +		vint_desc->affinity_managed = true;
>>>> +
>>>> +		irq_data_update_effective_affinity(d, mask_val);
>>>> +		parent_irq_data = irq_get_irq_data(vint_desc->parent_virq);
>>>> +		if (parent_irq_data->chip->irq_set_affinity)
>>>> +			return parent_irq_data->chip->irq_set_affinity(parent_irq_data, mask_val, force);
>>>
>>> This looks completely wrong.
>>>
>>> You still have a chained irqchip on all paths, and have to do some
>>> horrible probing to work out:
>>>
>>> - which parent interrupt this is
>>>
>>> - how many interrupts are connected to it
>>>
>>> And then the fun begins:
>>>
>>> - You have one interrupt that is standalone, so its affinity can be
>>>   moved
>>>
>>> - An unrelated driver gets probed, and one of its interrupts gets
>>>   lumped together with the one above
>>>
>>> - Now it cannot be moved anymore, and userspace complains
>>>
>>> The rule is very simple: chained irqchip, no affinity management.
>>> Either you reserve a poll of direct interrupts that have affinity
>>
>> This is what I am trying to accomplish, that is, reserve a pool of
>> direct interrupts that can be used by certain drivers that require IRQ
>> steering for performance. But I don't see a way to indicate from client
>> drivers to allocate from this reserved pool (there is no hint in
>> request_irq() call that ends up in .irq_request_resources() that I can use)
>>
>> I can try and virtually split INTA into two irqchips perhaps, with one
>> part modeled as chained irqchip and other as stacked for the reserved
>> pool (and would have to spawn of two child msi-domains I presume).
>> But, there is only one DT node for this irqchip and thus clients cannot
>> request IRQ for reserved pool.
> 
> I don't see why DT should be aware of this. You only need to decide at
> allocation time which one is where, and plug it at the right level.
> 

Ah, okay. I can certainly update .irq_set_affinity to point to INTA's
parent during ti_sci_inta_irq_domain_alloc() for reserved pool of direct
interrupts.

> And you probably only need *one* chained interrupt that muxes
> everything that cannot be allocated direct path.

IIRC, suggestion is that driver is aware (eg via static list) of
incoming downstream events that need to be directly mapped and thus does
so during allocation, right?

But this is a bit tricky on TI K3 SoCs. INTA is meant to aggregate DMA
channel events. On certain SoCs, any DMA channel can talk to any
peripheral and there is no static binding. We, ideally, need to do load
balancing for certain networking (and such high IRQ load) peripherals only.

One way I see is create a special DMA channel pool at DMA driver as well
that correspond to reserved VINT pool here. Else I am not aware of a way
for DMA driver to inform INTA driver to allocate IRQ line from reserved
pool.

> 
>> Wondering if you have any pointers here?
> 
> I don't. But this shouldn't be too hard to bolt onto the existing
> framework.


Really appreciate feedback! Thanks

Regards
Vignesh

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel