Re: [PATCH V10 1/8] irqchip: add C-SKY SMP interrupt controller

[Marc Zyngier]

On 09/10/18 03:24, Guo Ren wrote:
> Hi Marc,
>
> On Mon, Oct 08, 2018 at 05:35:19PM +0100, Marc Zyngier wrote:
> > Hi Guo,
> >
> > On 04/10/18 18:22, Guo Ren wrote:
> > >   - Irq-csky-mpintc is C-SKY smp system interrupt controller and it
> > >     could support 16 soft irqs, 16 private irqs, and 992 max common
> > >     irqs.
> > >
> > > Changelog:
> > >   - pass checkpatch.pl
> > >   - Move IPI_IRQ into the driver
> > >   - Remove irq_set_default_host() and use set_ipi_irq_mapping()
>
>   [...]
>
> > > +#ifdef CONFIG_SMP
> > > +static void csky_mpintc_send_ipi(const unsigned long *mask)
> > > +{
> >
> > Why isn't this a cpumask? It should be this driver's job to convert the
> > cpumask to an interrupt-controller specific representation, and not the SMP
> > code's.
> Ok, use cpumask.
>
>   [...]
>
> > > +#ifdef CONFIG_SMP
> > > +	set_send_ipi(&csky_mpintc_send_ipi);
> > > +
> > > +	set_ipi_irq_mapping(&csky_mpintc_ipi_irq_mapping);
> >
> > Since you seem to be inventing a new set_send_ipi callback, why don't you
> > define it as:
> >
> > void set_send_ipi(void (*func)(const struct cpumask *),
> > 		  unsigned int ipi_irq);
> > after having created the mapping for the IPI interrupt? It would avoid this
> > rather pointless mapping callback.
>
> I'll define it to:
>
> void set_send_ipi(void (*func)(const struct cpumask *));
>
> IPI_IRQ only use software-irq-15 in mpintc driver, so arch needn't care
> about irq-num.

Except that your SMP code does a request_irq on the result of 
create_mapping. Who is going to do that if you don't pass the Linux irq 
around?

> 	/*
> 	 * INTCL_SIGR[3:0] INTID
> 	 * INTCL_SIGR[8:15] CPUMASK
> 	 */
> 	writel_relaxed((*mask) << 8 | IPI_IRQ, reg_base + INTCL_SIGR);
>
> We direct put IPI_IRQ to the hw-reg for the target core.

I can see that, but that doesn't solve the issue outlined above.

Thanks,

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