Re: [PATCH v4 10/27] vfio/cxl: Reject unsupported decoder topologies at bind

[Alex Williamson <alex@shazbot.org>]

On Thu, 13 Aug 2026 15:06:14 +0530
<mhonap@nvidia.com> wrote:

> From: Manish Honap <mhonap@nvidia.com>
> 
> The guest drives a single virtual decoder whose memory maps 1:1, so a
> device with more than one HDM decoder, or an interleaved one, cannot be
> represented. Turn that assumption into an explicit refusal at bind rather
> than a silent misprogramming later.
> 
> Signed-off-by: Manish Honap <mhonap@nvidia.com>
> ---
>  drivers/vfio/pci/cxl/vfio_cxl_core.c | 20 ++++++++++++++++++++
>  1 file changed, 20 insertions(+)
> 
> diff --git a/drivers/vfio/pci/cxl/vfio_cxl_core.c b/drivers/vfio/pci/cxl/vfio_cxl_core.c
> index 125e11354a46..7edc53b25576 100644
> --- a/drivers/vfio/pci/cxl/vfio_cxl_core.c
> +++ b/drivers/vfio/pci/cxl/vfio_cxl_core.c
> @@ -36,10 +36,30 @@ static int vfio_cxl_init_device(struct vfio_pci_core_device *vdev)
>  	if (!pdev->hdm)
>  		return -EPROBE_DEFER;
>  
> +	/* The guest drives one virtual decoder; multiple are unsupported. */
> +	if (pdev->hdm->decoder_count != 1)
> +		return -EOPNOTSUPP;
> +
>  	hdm_size = range_len(&pdev->hdm->settings[0].hpa_range);
>  	if (!hdm_size)
>  		return -ENXIO;
>  
> +	/* Interleaved decoders are unsupported. */
> +	if (pdev->hdm->settings[0].interleave_ways != 1)
> +		return -EOPNOTSUPP;
> +
> +	/*
> +	 * The guest drives resets through the CXL Device DVSEC and polls the
> +	 * shadow for completion. If the host cannot service a function-scoped
> +	 * CXL reset (no reset DVSEC, a multifunction device, or no HDM reset
> +	 * support), that guest request could never complete, so refuse the
> +	 * device rather than advertise a reset the guest would poll on forever.
> +	 */
> +	if (!cxl_reset_capable(pdev)) {
> +		pci_err(pdev, "vfio-cxl: Unsupported device: host cannot service a CXL reset request\n");
> +		return -EOPNOTSUPP;
> +	}

Why does this particular error deserve such a high priority log while
we silently fail the other two cases added here?

These don't seem particularly noteworthy to split into a separate patch
versus rolling them into the previous.  I'd probably also group the
easily testable EOPNOTSUPP errors before we actually start evaluating
things like the hdm_size.  Thanks,

Alex

> +
>  	dvsec = pci_find_dvsec_capability(pdev, PCI_VENDOR_ID_CXL,
>  					  PCI_DVSEC_CXL_DEVICE);
>  	serial = pci_get_dsn(pdev);