Re: [PATCH v2 03/10] irqchip/riscv-intc: Introduce Andes IRQ chip

[Yu-Chien Peter Lin <peterlin@andestech.com>]

Hi Thomas,

On Fri, Oct 27, 2023 at 09:08:56AM +0200, Thomas Gleixner wrote:
> On Thu, Oct 19 2023 at 21:57, Yu Chien Peter Lin wrote:
> > This commit adds support for the Andes IRQ chip, which provides
> 
> "This commit" is not any different from "This patch" and equaly
> pointless. See Documentation/process/submitting-patches.rst

Sure, will rewrite the commit message.
Thanks for the pointer.

> Also please write out interrupt instead of IRQ. Changelogs are text and
> not subject to twitter limitations.

OK!

> > IRQ mask/unmask functions to access the custom CSR (SLIE)
> 
> What is a CSR? These acronyms are really annoying for people who are not
> familiar with the chip specific details.

OK!

> > +static void andes_intc_irq_mask(struct irq_data *d)
> > +{
> > +	unsigned int mask = BIT(d->hwirq % BITS_PER_LONG);
> > +
> > +	if (d->hwirq < ANDES_SLI_CAUSE_BASE)
> 
> This lacks a comment which explains why these hardware interrupts are
> special.

Sure, will add a comment here.

> > +		csr_clear(CSR_IE, mask);
> > +	else
> > +		csr_clear(ANDES_CSR_SLIE, mask);
> > +}
> > +
> > +static void andes_intc_irq_unmask(struct irq_data *d)
> > +{
> > +	unsigned int mask = BIT(d->hwirq % BITS_PER_LONG);
> > +
> > +	if (d->hwirq < ANDES_SLI_CAUSE_BASE)
> > +		csr_set(CSR_IE, mask);
> > +	else
> > +		csr_set(ANDES_CSR_SLIE, mask);
> > +}
> > +
> >  static void riscv_intc_irq_eoi(struct irq_data *d)
> >  {
> >  	/*
> > @@ -68,12 +89,35 @@ static struct irq_chip riscv_intc_chip = {
> >  	.irq_eoi = riscv_intc_irq_eoi,
> >  };
> >  
> > +static struct irq_chip andes_intc_chip = {
> > +	.name = "RISC-V INTC",
> > +	.irq_mask = andes_intc_irq_mask,
> > +	.irq_unmask = andes_intc_irq_unmask,
> > +	.irq_eoi = riscv_intc_irq_eoi,
> 
> https://www.kernel.org/doc/html/latest/process/maintainer-tip.html#struct-declarations-and-initializers

OK, will fix.

> > +};
> > +
> >  static int riscv_intc_domain_map(struct irq_domain *d, unsigned int irq,
> >  				 irq_hw_number_t hwirq)
> >  {
> > +	struct fwnode_handle *fn = riscv_get_intc_hwnode();
> > +	struct irq_chip *chip;
> > +	const char *cp;
> > +	int rc;
> > +
> >  	irq_set_percpu_devid(irq);
> 
> Why is this not moved after the failure condition too?
> 
> > -	irq_domain_set_info(d, irq, hwirq, &riscv_intc_chip, d->host_data,
> > -			    handle_percpu_devid_irq, NULL, NULL);
> > +
> > +	rc = fwnode_property_read_string(fn, "compatible", &cp);
> > +	if (rc)
> > +		return rc;
> > +
> > +	if (strcmp(cp, "riscv,cpu-intc") == 0)
> > +		chip = &riscv_intc_chip;
> > +	else if (strcmp(cp, "andestech,cpu-intc") == 0)
> > +		chip = &andes_intc_chip;
> 
> How is this supposed to work with ACPI?
> 
> The obvious solution for this is to have two different init functions
> 
> riscv_intc_init()
> {
>         riscv_intc_init_common(..., &risc_intc_chip);
> }
> 
> riscv_andes_init()
> {
>         riscv_intc_init_common(..., &andes_intc_chip);
> }
> 
> riscv_intc_init_common(...., *chip)
> {
>         // Set the interrupt chip pointer as domain host data
>         irqdomain_create_linear(...., chip);
> }
> 
> and then you can use that in the map function:
> 
>     chip = domain->host_data;
> 
> See?

Got it! Will reimplement according to this method.

> > diff --git a/include/linux/irqchip/irq-riscv-intc.h b/include/linux/irqchip/irq-riscv-intc.h
> > new file mode 100644
> > index 000000000000..87c105b5b545
> > --- /dev/null
> > +++ b/include/linux/irqchip/irq-riscv-intc.h
> > @@ -0,0 +1,12 @@
> > +/* SPDX-License-Identifier: GPL-2.0-only */
> > +/*
> > + * Copyright (C) 2023 Andes Technology Corporation
> > + */
> > +#ifndef __INCLUDE_LINUX_IRQCHIP_IRQ_RISCV_INTC_H
> > +#define __INCLUDE_LINUX_IRQCHIP_IRQ_RISCV_INTC_H
> > +
> > +#define ANDES_SLI_CAUSE_BASE	256
> > +#define ANDES_CSR_SLIE		0x9c4
> > +#define ANDES_CSR_SLIP		0x9c5
> > +
> > +#endif /* __INCLUDE_LINUX_IRQCHIP_IRQ_RISCV_INTC_H */
> 
> What's the purpose of this header file? The defines are only used in the
> interrupt chip driver code, so they can just be in the C file. No?
> 
> Thanks,

These definitions are shared with drivers/perf/riscv_pmu_sbi.c,
however, I should not add vendor specific things here, they
will be move to include/linux/soc/andes/irq.h.

Thanks for the review!

Best regards,
Peter Lin

>         tglx
>