[PATCH RFC v3 2/4] iommufd: Add IOMMUFD_CMD_DEVICE_SET_DATA and IOMMUFD_CMD_DEVICE_UNSET_DATA

[Baolu Lu]

On 4/23/23 3:40 PM, Nicolin Chen wrote:
> Add a new pair of ioctls to allow user space to set and unset its iommu-
> specific device data for a passthrough device that's behind the iommu.
> 
> On platforms with SMMUv3, this new uAPIs will be used to forward a user
> space virtual Stream ID of a passthrough device to link to its physical
> Stream ID and log into a lookup table, in order for the host kernel to
> later run sanity on ATC invalidation requests from the user space, with
> ATC_INV commands that have SID fields (virtual Stream IDs).
> 
> Signed-off-by: Nicolin Chen <nicolinc at nvidia.com>
> ---
>   drivers/iommu/iommufd/device.c          | 81 +++++++++++++++++++++++++
>   drivers/iommu/iommufd/iommufd_private.h |  3 +
>   drivers/iommu/iommufd/main.c            |  4 ++
>   include/uapi/linux/iommufd.h            | 32 ++++++++++
>   4 files changed, 120 insertions(+)
> 
> diff --git a/drivers/iommu/iommufd/device.c b/drivers/iommu/iommufd/device.c
> index c649a3403797..9480cd36a8bd 100644
> --- a/drivers/iommu/iommufd/device.c
> +++ b/drivers/iommu/iommufd/device.c
> @@ -136,6 +136,8 @@ void iommufd_device_destroy(struct iommufd_object *obj)
>   	struct iommufd_device *idev =
>   		container_of(obj, struct iommufd_device, obj);
>   
> +	if (WARN_ON(idev->has_user_data))
> +		dev_iommu_ops(idev->dev)->unset_dev_user_data(idev->dev);

Do you really need this WARN_ON()? The user space application can easily
trigger this kernel WARN() by setting the user data and forgetting to
unset it.

Best regards,
baolu