[PATCH] iommu/riscv: prefer WSI on IGS=BOTH when wired IRQs are described

[Sunil V L]

On Tue, May 19, 2026 at 6:28 PM <fangyu.yu at linux.alibaba.com> wrote:
>
> From: Fangyu Yu <fangyu.yu at linux.alibaba.com>
>
> The RISC-V IOMMU spec defines IGS=BOTH as supporting both MSI and
> WSI, with software selecting the path. The DT path already behaves
> as expected by selecting WSI when wired IRQ resources are described.
> The ACPI path, however, currently falls back to MSI even when
> firmware describes wired IRQ resources.
>
> Use firmware-described wired IRQ resources as the trigger to select
> WSI for IGS=BOTH:
>   - DT: "interrupts" present, no "msi-parent"
>   - ACPI: DSDT _CRS Interrupt() descriptors
>     (mainline does not yet parse the RIMT Interrupt Wire Array)
>
> When triggered, rewrite igs to IGS_WSI and reuse the existing WSI
> handling. Keep the existing behaviour otherwise.
>
> Fixes: d5f88acdd6ff ("iommu/riscv: Add support for platform msi")
> Signed-off-by: Fangyu Yu <fangyu.yu at linux.alibaba.com>
> ---
>  drivers/iommu/riscv/iommu-platform.c | 15 +++++++++++++++
>  1 file changed, 15 insertions(+)
>
> diff --git a/drivers/iommu/riscv/iommu-platform.c b/drivers/iommu/riscv/iommu-platform.c
> index 399ba8fe1b3e..bd7712231140 100644
> --- a/drivers/iommu/riscv/iommu-platform.c
> +++ b/drivers/iommu/riscv/iommu-platform.c
> @@ -71,6 +71,21 @@ static int riscv_iommu_platform_probe(struct platform_device *pdev)
>         iommu->irqs_count = RISCV_IOMMU_INTR_COUNT;
>
>         igs = FIELD_GET(RISCV_IOMMU_CAPABILITIES_IGS, iommu->caps);
> +
> +       /*
> +        * IGS=BOTH means the IOMMU supports either MSI or WSI;
> +        * the spec leaves the choice to software. Use the firmware-described
> +        * wired interrupt resources as the trigger:
> +        *   - DT  : "interrupts" property present, no "msi-parent"  -> WSI
> +        *   - ACPI: DSDT _CRS Interrupt() present  -> WSI
> +        * Otherwise default to the MSI path.
> +        */
> +       if (igs == RISCV_IOMMU_CAPABILITIES_IGS_BOTH &&
> +           platform_irq_count(pdev) > 0) {
> +               dev_info(dev, "firmware describes wired IRQs; preferring WSI on IGS=BOTH\n");
> +               igs = RISCV_IOMMU_CAPABILITIES_IGS_WSI;
> +       }
> +
Won't it change the DT behavior as it doesn't check msi-parent
anymore?  IOW, should this be made specific to ACPI
by checking whether the device node is acpi node?

Thanks,
Sunil