Re: [PATCH v5 14/17] irqchip/riscv-imsic: Add ACPI support

[Sunil V L <sunilvl@ventanamicro.com>]

On Fri, May 24, 2024 at 12:00:21AM +0200, Thomas Gleixner wrote:
> On Wed, May 01 2024 at 17:47, Sunil V L wrote:
> 
> > RISC-V IMSIC interrupt controller provides IPI and MSI support.
> > Currently, DT based drivers setup the IPI feature early during boot but
> > defer setting up the MSI functionality. However, in ACPI systems, ACPI,
> > both IPI and MSI features need to be initialized early itself.
> 
> Why?
> 
Sorry, commit message got truncated by mistake. Basically, in ACPI PCI
scan happens very early and there is no concept of msi-parent/dependency
on MSI controller like in DT. It just assumes MSI is setup already. Due
to this, we need to setup MSI controller early as well.

> > +
> > +#ifdef CONFIG_ACPI
> > +
> > +static struct fwnode_handle *imsic_acpi_fwnode;
> > +
> > +struct fwnode_handle *imsic_acpi_get_fwnode(struct device *dev)
> 
> Why is this function global? It's only used in the very same file and
> under the same CONFIG_ACPI #ifdef, no?
> 
For platform devices using MSIs, we need a way to determine the MSI
domain. This function is exported so that platform device like
APLIC/IOMMU can find the MSI irqdomain.

For PCI, pci_msi_register_fwnode_provider() is registered by the MSI
driver for this purpose.

Let me know if this can be made better.

> > +{
> > +	return imsic_acpi_fwnode;
> > +}
> > +
> > +static int __init imsic_early_acpi_init(union acpi_subtable_headers *header,
> > +					const unsigned long end)
> > +{
> > +	struct acpi_madt_imsic *imsic = (struct acpi_madt_imsic *)header;
> > +	int rc;
> > +
> > +	imsic_acpi_fwnode = irq_domain_alloc_named_fwnode("imsic");
> > +	if (!imsic_acpi_fwnode) {
> > +		pr_err("unable to allocate IMSIC FW node\n");
> > +		return -ENOMEM;
> > +	}
> > +
> > +	/* Setup IMSIC state */
> > +	rc = imsic_setup_state(imsic_acpi_fwnode, (void *)imsic);
> 
> Pointless (void *) cast.
> 
Okay.

> > +	if (rc) {
> > +		pr_err("%pfwP: failed to setup state (error %d)\n", imsic_acpi_fwnode, rc);
> > +		return rc;
> > +	}
> > +
> > +	/* Do early setup of IMSIC state and IPIs */
> > +	rc = imsic_early_probe(imsic_acpi_fwnode);
> > +	if (rc)
> > +		return rc;
> > +
> > +	rc = imsic_platform_acpi_probe(imsic_acpi_fwnode);
> > +
> > +#ifdef CONFIG_PCI
> > +	if (!rc)
> > +		pci_msi_register_fwnode_provider(&imsic_acpi_get_fwnode);
> > +#endif
> > +
> > +	return rc;
> 
> Any error return in this function leaks the firmware node and probably
> some more stuff.
>
Yeah, fwnode needs free up and need to update the code a bit. Thanks!
 
> > +}
> > +
> > +IRQCHIP_ACPI_DECLARE(riscv_imsic, ACPI_MADT_TYPE_IMSIC, NULL,
> > +		     1, imsic_early_acpi_init);
> > +#endif
> 
> ...
> 
> > -	/* Find number of interrupt identities */
> > -	rc = of_property_read_u32(to_of_node(fwnode), "riscv,num-ids",
> > -				  &global->nr_ids);
> > -	if (rc) {
> > -		pr_err("%pfwP: number of interrupt identities not found\n", fwnode);
> > -		return rc;
> > +		/* Find number of guest interrupt identities */
> > +		rc = of_property_read_u32(to_of_node(fwnode), "riscv,num-guest-ids",
> > +					  &global->nr_guest_ids);
> > +		if (rc)
> > +			global->nr_guest_ids = global->nr_ids;
> > +	} else {
> > +		global->guest_index_bits = imsic->guest_index_bits;
> > +		global->hart_index_bits = imsic->hart_index_bits;
> > +		global->group_index_bits = imsic->group_index_bits;
> > +		global->group_index_shift = imsic->group_index_shift;
> > +		global->nr_ids = imsic->num_ids;
> > +		global->nr_guest_ids = imsic->num_guest_ids;
> >  	}
> 
> Seriously?
> 
> Why can't you just split out the existing DT code into a separate
> function in an initial patch which avoulds all of this unreviewable
> churn of making the DT stuff indented ?
> 
Sure, makes sense. let me create separate patch first as you suggested.

> > +#ifdef CONFIG_ACPI
> > +int imsic_platform_acpi_probe(struct fwnode_handle *fwnode);
> > +struct fwnode_handle *imsic_acpi_get_fwnode(struct device *dev);
> > +#else
> > +static inline struct fwnode_handle *imsic_acpi_get_fwnode(struct device *dev)
> > +{
> > +	return NULL;
> > +}
> > +#endif
> 
> Oh well.
> 
I guess this is related to your prior comment about the need to make
this public function. Let me know if I am missing something.

Thanks!
Sunil