On Sun, Sep 16, 2018 at 08:07:55PM +0100, Marc Zyngier wrote:
> On Sun, 16 Sep 2018 09:50:02 +0100,
> Guo Ren <ren_...@c-sky.com> wrote:
> > +static void csky_mpintc_handler(struct pt_regs *regs)
> > +{
> > + void __iomem *reg_base = this_cpu_read(intcl_reg);
> > +
> > + do {
> > + handle_domain_irq(NULL,
>
> It is definitely odd to call into handle_domain_irq without a
> domain. A new architecture (which C-SKY apparently is) shouldn't
> depend on this, and should always provide a domain.
Ok, I'll change it to handle_domain_irq(root_domain and prepare the
root_domain definition.
> > + ret = of_property_read_u32(node, "csky,num-irqs", &nr_irq);
> > + if (ret < 0) {
> > + nr_irq = INTC_IRQS;
> > + }
>
> Drop the extra braces.
Ok, and change it to:
if (ret < 0)
nr_irq = INTC_IRQS
> > +
> > + irq_set_default_host(root_domain);
>
> Please drop this. There is no reason to use this on any modern, DT
> based architecture.
Please let me keep this and in my arch/csky/kernel/smp.c:
void __init setup_smp_ipi(void)
{
...
irq_create_mapping(NULL, IPI_IRQ);
I need irq_set_default_host to pass the domain and this api is used by
a lot of drivers:
$ grep irq_set_default_host drivers/irqchip -r | wc -l
16
In future we could find a better solution.
> > diff --git a/irq-csky-apb-intc.c b/irq-csky-apb-intc.c
> > new file mode 100644
> > index 0000000..d56e6b5
> > --- /dev/null
> > +++ b/irq-csky-apb-intc.c
>
> This is a separate driver, right? Please make it a separate patch.
Ok, no problem.
> > +static inline int handle_irq_perbit(struct pt_regs *regs, u32 hwirq, u32
> > irq_base)
>
> Consider using bool as the return type, and use true/false as return
> values.
Ok, no problem.
> > + int ret;
>
> bool.
Ok, no problem.
> > + int ret;
>
> bool.
Ok, no problem.
