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

[Hanjun Guo]

On 07/22/2015 04:35 PM, 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
>
> 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.
>
> So to me acpi_irq_model_id is suitable for the token, can we use
> 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.

or just introduce a similar enum as acpi_irq_model for this purpose,
as acpi_irq_model is for _PIC method.

Thanks
Hanjun