On Thu, 13 Aug 2026 15:06:29 +0530
<[email protected]> wrote:
> From: Manish Honap <[email protected]>
>
> Add an opt-out so users can keep vfio-pci's CXL extensions out of the
> path for individual devices or for an entire vfio-pci instance. The
> runtime gates are:
>
> - Module parameter vfio_pci.disable_cxl (bool, 0444). Setting
> disable_cxl=1 at modprobe time makes vfio_pci_probe() set
> vdev->disable_cxl on every device it binds.
>
> - Variant drivers (nvgrace, mlx5, and others) may set vdev->disable_cxl
> in their own probe for per-device control without the module
> parameter. The bit lives on struct vfio_pci_core_device so it is
> reachable from any variant.
>
> vfio_pci_core_init_dev() consults vdev->disable_cxl before it probes for
> a CXL device, so a device that opts out is driven as plain vfio-pci:
> vfio-cxl is not loaded, init_device() never runs, and the device gets no
> VFIO_DEVICE_FLAGS_CXL, no HDM or component-register regions, and no DVSEC
> virtualization.
>
> The module parameter is built only when CONFIG_VFIO_CXL is enabled; the
> disable_cxl bit itself is unconditional so a variant driver can set it
> regardless. This mirrors the long-standing disable_denylist opt-out.
>
> Signed-off-by: Manish Honap <[email protected]>
> ---
> drivers/vfio/pci/vfio_pci.c | 9 +++++++++
> drivers/vfio/pci/vfio_pci_core.c | 8 +++++---
> include/linux/vfio_pci_core.h | 1 +
> 3 files changed, 15 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/vfio/pci/vfio_pci.c b/drivers/vfio/pci/vfio_pci.c
> index 830369ff878d..0ad041fffe48 100644
> --- a/drivers/vfio/pci/vfio_pci.c
> +++ b/drivers/vfio/pci/vfio_pci.c
> @@ -60,6 +60,12 @@ static bool disable_denylist;
> module_param(disable_denylist, bool, 0444);
> MODULE_PARM_DESC(disable_denylist, "Disable use of device denylist.
> Disabling the denylist allows binding to devices with known errata that may
> lead to exploitable stability or security issues when accessed by untrusted
> users.");
>
> +#if IS_ENABLED(CONFIG_VFIO_CXL)
> +static bool disable_cxl;
> +module_param(disable_cxl, bool, 0444);
> +MODULE_PARM_DESC(disable_cxl, "Disable CXL Type-2 extensions for all devices
> bound to vfio-pci. A variant driver may instead set vdev->disable_cxl in its
> own .init callback.");
The latter sentence isn't module parameter description material, it's
aimed at users. Why does this need to be read-only since it's
following the latch-at-init flow? Thanks,
Alex
> +#endif
> +
> static bool vfio_pci_dev_in_denylist(struct pci_dev *pdev)
> {
> switch (pdev->vendor) {
> @@ -142,6 +148,9 @@ static int vfio_pci_init_dev(struct vfio_device
> *core_vdev)
> #ifdef CONFIG_VFIO_PCI_VGA
> vdev->disable_vga = disable_vga;
> #endif
> +#if IS_ENABLED(CONFIG_VFIO_CXL)
> + vdev->disable_cxl = disable_cxl;
> +#endif
>
> return vfio_pci_core_init_dev(core_vdev);
> }
> diff --git a/drivers/vfio/pci/vfio_pci_core.c
> b/drivers/vfio/pci/vfio_pci_core.c
> index 0fed8e00bc1d..4b51a0f1e847 100644
> --- a/drivers/vfio/pci/vfio_pci_core.c
> +++ b/drivers/vfio/pci/vfio_pci_core.c
> @@ -2423,10 +2423,12 @@ int vfio_pci_core_init_dev(struct vfio_device
> *core_vdev)
> xa_init(&vdev->ctx);
>
> /*
> - * Load vfio-cxl on demand for a CXL device. If it is absent, drive the
> - * device as plain vfio-pci rather than failing the bind.
> + * Load vfio-cxl on demand for a CXL device unless the user opted out.
> + * If it is opted out or absent, drive the device as plain vfio-pci
> + * rather than failing the bind.
> */
> - if (pcie_is_cxl(vdev->pdev) && vfio_pci_is_cxl_type2(vdev->pdev)) {
> + if (!vdev->disable_cxl && pcie_is_cxl(vdev->pdev) &&
> + vfio_pci_is_cxl_type2(vdev->pdev)) {
> const struct vfio_cxl_ops *ops;
>
> request_module("vfio-cxl");
> diff --git a/include/linux/vfio_pci_core.h b/include/linux/vfio_pci_core.h
> index 18e206a35d8c..8e3723a55c17 100644
> --- a/include/linux/vfio_pci_core.h
> +++ b/include/linux/vfio_pci_core.h
> @@ -158,6 +158,7 @@ struct vfio_pci_core_device {
> bool disable_idle_d3:1;
> bool nointxmask:1;
> bool disable_vga:1;
> + bool disable_cxl:1;
> /* Flags modified at runtime - dedicated storage unit */
> bool needs_reset;
> bool pm_intx_masked;
