Re: [RFC PATCH kvmtool 01/10] vfio: Associate vm instance with vfio fd

[Mostafa Saleh <smostafa@google.com>]

Hi Aneesh,

On Sun, May 25, 2025 at 01:19:07PM +0530, Aneesh Kumar K.V (Arm) wrote:
> This is needed for followup patches
> 
> Signed-off-by: Aneesh Kumar K.V (Arm) <aneesh.kumar@kernel.org>
> ---
>  vfio/core.c | 24 +++++++++++++++++++++++-
>  1 file changed, 23 insertions(+), 1 deletion(-)
> 
> diff --git a/vfio/core.c b/vfio/core.c
> index 3ff2c0b075df..c6b305c30cf7 100644
> --- a/vfio/core.c
> +++ b/vfio/core.c
> @@ -9,6 +9,7 @@
>  #define IOMMU_GROUP_DIR		"/sys/kernel/iommu_groups"
>  
>  static int vfio_container;
> +static int kvm_vfio_device;
>  static LIST_HEAD(vfio_groups);
>  static struct vfio_device *vfio_devices;
>  
> @@ -437,8 +438,19 @@ static int vfio_configure_groups(struct kvm *kvm)
>  		ret = vfio_configure_reserved_regions(kvm, group);
>  		if (ret)
>  			return ret;
> -	}
>  
> +		struct kvm_device_attr attr = {
> +			.group = KVM_DEV_VFIO_FILE,
> +			.attr = KVM_DEV_VFIO_FILE_ADD,
> +			.addr = (__u64)&group->fd,
> +		};
> +
> +		if (ioctl(kvm_vfio_device, KVM_SET_DEVICE_ATTR, &attr)) {
> +			pr_err("Failed KVM_SET_DEVICE_ATTR for KVM_DEV_VFIO_FILE");
> +			return -ENODEV;
> +		}
> +
I think it’s better if we get and print the error code of the ioctl in case it
fails instead of masking it as -ENODEV.

> +	}
>  	return 0;
>  }
>  
> @@ -656,6 +668,16 @@ static int vfio__init(struct kvm *kvm)
>  	if (!vfio_devices)
>  		return -ENOMEM;
>  
> +	struct kvm_create_device device = {
> +		.type = KVM_DEV_TYPE_VFIO,
> +	};
> +
> +	if (ioctl(kvm->vm_fd, KVM_CREATE_DEVICE, &device)) {
> +		pr_err("Failed KVM_CREATE_DEVICE ioctl");
> +		return -ENODEV;
> +	}
> +	kvm_vfio_device = device.fd;

Do we need to close this file in vfio_group_exit()?

Thanks,
Mostafa

> +
>  	ret = vfio_container_init(kvm);
>  	if (ret)
>  		return ret;
> -- 
> 2.43.0
>