[PATCH 05/10] vfio: Use IOMMU_CAP_ENFORCE_CACHE_COHERENCY for vfio_file_enforced_coherent()

[Tian, Kevin]

> From: Jason Gunthorpe <jgg at nvidia.com>
> Sent: Wednesday, October 26, 2022 2:17 AM
> 
> iommufd doesn't establish the iommu_domains until after the device FD is
> opened, even if the container has been set. This design is part of moving
> away from the group centric iommu APIs.
> 
> This is fine, except that the normal sequence of establishing the kvm
> wbindv won't work:

wbindv -> wbinvd

> 
>    group = open("/dev/vfio/XX")
>    ioctl(group, VFIO_GROUP_SET_CONTAINER)
>    ioctl(kvm, KVM_DEV_VFIO_GROUP_ADD)
>    ioctl(group, VFIO_GROUP_GET_DEVICE_FD)
> 
> As the domains don't start existing until GET_DEVICE_FD. Further,
> GET_DEVICE_FD requires that KVM_DEV_VFIO_GROUP_ADD already be
> done as that
> is what sets the group->kvm and thus device->kvm for the driver to use
> during open.
> 
> Now that we have device centric cap ops and the new
> IOMMU_CAP_ENFORCE_CACHE_COHERENCY we know what the
> iommu_domain will be
> capable of without having to create it. Use this to compute

it's worth noting that the prerequisite is that vfio always enforces
cache coherency on a domain according to the iommu capability
of the devices attached to that domain. There is no mix of attaching
a device supporting the cap to a domain which doesn't enforce
coherency. With that we know what the domain will be w/o having
to create it.

> vfio_file_enforced_coherent() and resolve the ordering problems.
> 
> Signed-off-by: Jason Gunthorpe <jgg at nvidia.com>
> ---
>  drivers/vfio/container.c |  5 +++--
>  drivers/vfio/vfio.h      |  2 --
>  drivers/vfio/vfio_main.c | 27 ++++++++++++++-------------
>  3 files changed, 17 insertions(+), 17 deletions(-)
> 
> diff --git a/drivers/vfio/container.c b/drivers/vfio/container.c
> index 499777930b08fa..d97747dfb05d02 100644
> --- a/drivers/vfio/container.c
> +++ b/drivers/vfio/container.c
> @@ -188,8 +188,9 @@ void vfio_device_container_unregister(struct
> vfio_device *device)
>  			device->group->container->iommu_data, device);
>  }
> 
> -long vfio_container_ioctl_check_extension(struct vfio_container *container,
> -					  unsigned long arg)
> +static long
> +vfio_container_ioctl_check_extension(struct vfio_container *container,
> +				     unsigned long arg)
>  {
>  	struct vfio_iommu_driver *driver;
>  	long ret = 0;
> diff --git a/drivers/vfio/vfio.h b/drivers/vfio/vfio.h
> index 54e5a8e0834ccb..247590334e14b0 100644
> --- a/drivers/vfio/vfio.h
> +++ b/drivers/vfio/vfio.h
> @@ -119,8 +119,6 @@ int vfio_container_attach_group(struct
> vfio_container *container,
>  void vfio_group_detach_container(struct vfio_group *group);
>  void vfio_device_container_register(struct vfio_device *device);
>  void vfio_device_container_unregister(struct vfio_device *device);
> -long vfio_container_ioctl_check_extension(struct vfio_container *container,
> -					  unsigned long arg);
>  int __init vfio_container_init(void);
>  void vfio_container_cleanup(void);
> 
> diff --git a/drivers/vfio/vfio_main.c b/drivers/vfio/vfio_main.c
> index 1e414b2c48a511..a8d1fbfcc3ddad 100644
> --- a/drivers/vfio/vfio_main.c
> +++ b/drivers/vfio/vfio_main.c
> @@ -1625,24 +1625,25 @@ EXPORT_SYMBOL_GPL(vfio_file_is_group);
>  bool vfio_file_enforced_coherent(struct file *file)
>  {
>  	struct vfio_group *group = file->private_data;
> -	bool ret;
> +	struct vfio_device *device;
> +	bool ret = true;
> 
>  	if (!vfio_file_is_group(file))
>  		return true;
> 
> -	mutex_lock(&group->group_lock);
> -	if (group->container) {
> -		ret = vfio_container_ioctl_check_extension(group->container,
> -
> VFIO_DMA_CC_IOMMU);
> -	} else {
> -		/*
> -		 * Since the coherency state is determined only once a
> container
> -		 * is attached the user must do so before they can prove they
> -		 * have permission.
> -		 */
> -		ret = true;
> +	/*
> +	 * If the device does not have
> IOMMU_CAP_ENFORCE_CACHE_COHERENCY then
> +	 * any domain later attached to it will also not support it.
> +	 */

also add the other part i.e. if the device does have the cap then any domain
later attached to it will have the cap enabled. Only with both clarified
we can safely use the device cap here.

> +	mutex_lock(&group->device_lock);
> +	list_for_each_entry(device, &group->device_list, group_next) {
> +		if (!device_iommu_capable(device->dev,
> +
> IOMMU_CAP_ENFORCE_CACHE_COHERENCY)) {
> +			ret = false;
> +			break;
> +		}
>  	}
> -	mutex_unlock(&group->group_lock);
> +	mutex_unlock(&group->device_lock);
>  	return ret;
>  }
>  EXPORT_SYMBOL_GPL(vfio_file_enforced_coherent);
> --
> 2.38.0
>