Re: [PATCH v3 02/23] vfio: enable per-IRQ MSI-X masking

[John Levon <john.levon@nutanix.com>]

On Fri, Jun 06, 2025 at 05:10:34PM -0700, John Levon wrote:

> If VFIO_IRQ_INFO_MASKABLE is set for VFIO_PCI_MSIX_IRQ_INDEX, record
> this in ->can_mask_msix, and use it to individually mask MSI-X
> interrupts as needed.

I'm just going to drop this patch. Neither vfio nor libvfio-user (including
qemu-as-server) report MASKABLE for MSI-X anyway, so it doesn't seem relevant.

I'm not sure if Oracle had some other use case in mind or had previously tested
it somehow.

Furthermore, this:

> @@ -702,6 +721,13 @@ static void vfio_msix_enable(VFIOPCIDevice *vdev)
>              error_report("vfio: failed to enable vectors, %s",
>                           strerror(-ret));
>          }
> +    } else if (vdev->can_mask_msix) {
> +        /*
> +         * If we can use single irq masking, send an invalid fd on vector 0
> +         * to enable MSI-X without any vectors enabled.
> +         */
> +        vfio_device_irq_set_signaling(&vdev->vbasedev, VFIO_PCI_MSIX_IRQ_INDEX,
> +                                      0, VFIO_IRQ_SET_ACTION_TRIGGER, -1, NULL);
>      } else {
>          /*
>           * Some communication channels between VF & PF or PF & fw rely on the

Seems odd as it doesn't pass DATA_EVENTFD, unlike the vfio_enable_msix_no_vec()
below it; I have no idea why the difference or if it makes sense, but it doesn't
seem so.

regards
john