[PATCH 5/5] irqchip: GIC: Switch ACPI support to stacked domains

[Marc Zyngier]

On 22/07/15 09:35, Hanjun Guo wrote:
> On 07/22/2015 02:12 AM, Marc Zyngier wrote:
>> On 21/07/15 19:05, Lorenzo Pieralisi wrote:
>>> On Tue, Jul 21, 2015 at 11:08:00AM +0100, Marc Zyngier wrote:
>>>> Now that the basic ACPI GSI code is irq domain aware, make sure
>>>> that the ACPI support in the GIC doesn't pointlessly deviate from
>>>> the DT path.
>>>>
>>>> Signed-off-by: Marc Zyngier <marc.zyngier at arm.com>
>>>> ---
>>>>   drivers/irqchip/irq-gic.c       | 17 ++++++-----------
>>>>   include/linux/irqchip/arm-gic.h |  2 +-
>>>>   2 files changed, 7 insertions(+), 12 deletions(-)
>>>>
>>>> diff --git a/drivers/irqchip/irq-gic.c b/drivers/irqchip/irq-gic.c
>>>> index b41ccf5..f5d365d 100644
>>>> --- a/drivers/irqchip/irq-gic.c
>>>> +++ b/drivers/irqchip/irq-gic.c
>>>> @@ -813,8 +813,6 @@ static int gic_irq_domain_xlate(struct irq_domain *d,
>>>>   {
>>>>   	unsigned long ret = 0;
>>>>
>>>> -	if (irq_domain_get_of_node(d) != controller)
>>>> -		return -EINVAL;
>>>>   	if (intsize < 3)
>>>>   		return -EINVAL;
>>>>
>>>> @@ -887,7 +885,7 @@ void gic_set_irqchip_flags(unsigned long flags)
>>>>
>>>>   void __init gic_init_bases(unsigned int gic_nr, int irq_start,
>>>>   			   void __iomem *dist_base, void __iomem *cpu_base,
>>>> -			   u32 percpu_offset, struct device_node *node)
>>>> +			   u32 percpu_offset, void *domain_token)
>>>>   {
>>>>   	irq_hw_number_t hwirq_base;
>>>>   	struct gic_chip_data *gic;
>>>> @@ -946,8 +944,8 @@ void __init gic_init_bases(unsigned int gic_nr, int irq_start,
>>>>   		gic_irqs = 1020;
>>>>   	gic->gic_irqs = gic_irqs;
>>>>
>>>> -	if (node) {		/* DT case */
>>>> -		gic->domain = irq_domain_add_linear(node, gic_irqs,
>>>> +	if (domain_token) {		/* DT/ACPI case */
>>>> +		gic->domain = irq_domain_add_linear(domain_token, gic_irqs,
>>>>   						    &gic_irq_domain_hierarchy_ops,
>>>>   						    gic);
>>>>   	} else {		/* Non-DT case */
>>>> @@ -973,7 +971,7 @@ void __init gic_init_bases(unsigned int gic_nr, int irq_start,
>>>>   			irq_base = irq_start;
>>>>   		}
>>>>
>>>> -		gic->domain = irq_domain_add_legacy(node, gic_irqs, irq_base,
>>>> +		gic->domain = irq_domain_add_legacy(NULL, gic_irqs, irq_base,
>>>>   					hwirq_base, &gic_irq_domain_ops, gic);
>>>>   	}
>>>>
>>>> @@ -1132,12 +1130,9 @@ gic_v2_acpi_init(struct acpi_table_header *table)
>>>>   	}
>>>>
>>>>   	/*
>>>> -	 * Initialize zero GIC instance (no multi-GIC support). Also, set GIC
>>>> -	 * as default IRQ domain to allow for GSI registration and GSI to IRQ
>>>> -	 * number translation (see acpi_register_gsi() and acpi_gsi_to_irq()).
>>>> +	 * Initialize zero GIC instance (no multi-GIC support).
>>>>   	 */
>>>> -	gic_init_bases(0, -1, dist_base, cpu_base, 0, NULL);
>>>> -	irq_set_default_host(gic_data[0].domain);
>>>> +	gic_init_bases(0, -1, dist_base, cpu_base, 0, (void *)ACPI_IRQ_MODEL_GIC);
>>>
>>> Nit: the acpi_irq_model_id enum starts from 0, I do not think we will
>>> use the IRQ domain look-up for the ACPI_IRQ_MODEL_PIC but we have
>>> to be careful anyway.
>>
>> Yeah, I noticed that one too, but couldn't imagine the PIC being
>> migrated to that model just yet. It looks like it would be pretty
>> harmless to set ACPI_IRQ_MODEL_PIC to 1, and introduce
>> ACPI_IRQ_MODEL_ILLEGAL as zero.
> 
> I think this will be a problem, because acpi_irq_model_id enum actually
> is defined by the ACPI spec, and the value is used to report to BIOS
> the current interrupt model used by OS, see section 5.8.1 _PIC Method
> in ACPI 6.0:
> 
> 0 – PIC mode
> 1 – APIC mode
> 2 – SAPIC mode
> Other values –Reserved

Ah, right.

> so we can't set ACPI_IRQ_MODEL_PIC to 1 as it may break the firmware,
> also _PIC method actually is not needed for ARM platform at all, we
> don't need to report to firmware the interrupt model used by OS on
> ARM, it only used by legacy IA platforms, actually I'm planning to
> remove acpi_irq_model_id on ARM64.

Remove? I don't get it. Either ACPI_IRQ_MODEL_GIC is a leval value, and
you can't remove it, or it is not, and I wonder how it ended up here the
first place.

> So to me acpi_irq_model_id is suitable for the token, can we use
                             is *not* ?
> another one as the token? how about the GIC ID in the MADT table?
> and this also can be used for x86 (IOAPIC IDs) too.

You can use whatever you want, just not a pointer. I'll add a token
parameter to the acpi_set_irq_model function that I mentioned in my
reply to Lorenzo.

Thanks,

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