Re: [PATCH 3/3] irqchip/gic-v3-its: Include "msi-map-mask" for calculating nvecs

[Marc Zyngier]

+ Robin, who has dealt with a lot of the stuff here.

On Mon, 20 Sep 2021 07:41:33 +0100,
Kishon Vijay Abraham I <kishon@xxxxxx> wrote:
> 
> Using "msi-map-mask" in device tree lets multiple PCIe requestor ID to
> use the same GIC ITS device ID. So while creating the Interrupt
> Translation Table (ITT) for a specific GIC ITS device ID, the total number
> of interrupts required by all the PCIe requestor ID that maps to the
> same GIC ITS device ID should be calculated
> 
> Add support for gic-v3-its to include "msi-map-mask" property in device
> tree for calculating the total number of MSI interrupts in
> its_pci_msi_prepare().
> 
> Signed-off-by: Kishon Vijay Abraham I <kishon@xxxxxx>
> ---
>  drivers/irqchip/irq-gic-v3-its-pci-msi.c | 21 ++++++++++++++++++++-
>  1 file changed, 20 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/irqchip/irq-gic-v3-its-pci-msi.c b/drivers/irqchip/irq-gic-v3-its-pci-msi.c
> index ad2810c017ed..c79bca1a5787 100644
> --- a/drivers/irqchip/irq-gic-v3-its-pci-msi.c
> +++ b/drivers/irqchip/irq-gic-v3-its-pci-msi.c
> @@ -54,9 +54,13 @@ static int its_get_pci_alias(struct pci_dev *pdev, u16 alias, void *data)
>  static int its_pci_msi_prepare(struct irq_domain *domain, struct device *dev,
>  			       int nvec, msi_alloc_info_t *info)
>  {
> +	int alias_count = 0, map_count = 0, minnvec = 1, ret;
>  	struct pci_dev *pdev, *alias_dev;
>  	struct msi_domain_info *msi_info;
> -	int alias_count = 0, minnvec = 1;
> +	struct device *parent_dev;
> +	struct pci_bus *root_bus;
> +	struct device_node *np;
> +	u32 map_mask, rid;
>  
>  	if (!dev_is_pci(dev))
>  		return -EINVAL;
> @@ -78,6 +82,21 @@ static int its_pci_msi_prepare(struct irq_domain *domain, struct device *dev,
>  		info->flags |= MSI_ALLOC_FLAGS_PROXY_DEVICE;
>  	}
>  
> +	for (parent_dev = dev; parent_dev; parent_dev = parent_dev->parent) {

Move the declaration of np, ret and map_count here.

> +		np = parent_dev->of_node;
> +		if (!np)
> +			continue;
> +
> +		ret = of_property_read_u32(np, "msi-map-mask", &map_mask);
> +		if (!ret && map_mask != 0xffff) {
> +			rid = pci_dev_id(pdev) & map_mask;
> +			root_bus = find_pci_root_bus(pdev->bus);
> +			__pci_walk_bus(root_bus, its_pci_msi_vec_count, &map_count, rid, map_mask);
> +			break;
> +		}
> +	}
> +	alias_count = max(map_count, alias_count);

Move this after the call to __pci_walk_bus().

> +
>  	/* ITS specific DeviceID, as the core ITS ignores dev. */
>  	info->scratchpad[0].ul = pci_msi_domain_get_msi_rid(domain, pdev);

If you don't set the PROXY_DEVICE flag, bad thing will happen when the
canonical device is removed.

Ultimately, this should be moved into a separate helper, along the
line of pci_for_each_dma_alias(). It probably also needs to have an
ACPI counterpart.

Thanks,

	M.

-- 
Without deviation from the norm, progress is not possible.