[kernel,v2,09/11] vfio iommu: Add helpers to (un)register blocking notifiers per group

c086de81 "vfio iommu: Add blocking notifier to notify DMA_UNMAP" added
notifiers to a VFIO group. However even though the code underneath
uses groups, the API takes device struct pointers.

This adds helpers which do the same thing but take IOMMU groups instead.

This adds vfio_iommu_group_set_kvm() which is a wrapper on top of
vfio_group_set_kvm() but also takes an iommu_group.

Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
---
 include/linux/vfio.h |  6 +++++
 drivers/vfio/vfio.c  | 76 ++++++++++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 82 insertions(+)

On Sun, Dec 18, 2016 at 12:28:58PM +1100, Alexey Kardashevskiy wrote:
> c086de81 "vfio iommu: Add blocking notifier to notify DMA_UNMAP" added
> notifiers to a VFIO group. However even though the code underneath
> uses groups, the API takes device struct pointers.
> 
> This adds helpers which do the same thing but take IOMMU groups instead.
> 
> This adds vfio_iommu_group_set_kvm() which is a wrapper on top of
> vfio_group_set_kvm() but also takes an iommu_group.
> 
> Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>

Adding a second interface in parallel seems dubious.

Should the existing interface just be replaced with this one?

Or can the existing interface be re-implemented in terms of this one?

> ---
>  include/linux/vfio.h |  6 +++++
>  drivers/vfio/vfio.c  | 76 ++++++++++++++++++++++++++++++++++++++++++++++++++++
>  2 files changed, 82 insertions(+)
> 
> diff --git a/include/linux/vfio.h b/include/linux/vfio.h
> index edf9b2cad277..8a3488ba4732 100644
> --- a/include/linux/vfio.h
> +++ b/include/linux/vfio.h
> @@ -127,9 +127,15 @@ extern int vfio_register_notifier(struct device *dev,
>  extern int vfio_unregister_notifier(struct device *dev,
>  				    enum vfio_notify_type type,
>  				    struct notifier_block *nb);
> +extern int vfio_iommu_group_register_notifier(struct iommu_group *grp,
> +		enum vfio_notify_type type, unsigned long *events,
> +		struct notifier_block *nb);
> +extern int vfio_iommu_group_unregister_notifier(struct iommu_group *grp,
> +		enum vfio_notify_type type, struct notifier_block *nb);
>  
>  struct kvm;
>  extern void vfio_group_set_kvm(struct vfio_group *group, struct kvm *kvm);
> +extern void vfio_iommu_group_set_kvm(struct iommu_group *grp, struct kvm *kvm);
>  
>  /*
>   * Sub-module helpers
> diff --git a/drivers/vfio/vfio.c b/drivers/vfio/vfio.c
> index 9901c4671e2f..6b9a98508939 100644
> --- a/drivers/vfio/vfio.c
> +++ b/drivers/vfio/vfio.c
> @@ -2077,6 +2077,23 @@ void vfio_group_set_kvm(struct vfio_group *group, struct kvm *kvm)
>  }
>  EXPORT_SYMBOL_GPL(vfio_group_set_kvm);
>  
> +void vfio_iommu_group_set_kvm(struct iommu_group *grp, struct kvm *kvm)
> +{
> +	struct vfio_group *group;
> +
> +	if (!grp)
> +		return;
> +
> +	group = vfio_group_get_from_iommu(grp);
> +	if (!group)
> +		return;
> +
> +	vfio_group_set_kvm(group, kvm);
> +
> +	vfio_group_put(group);
> +}
> +EXPORT_SYMBOL_GPL(vfio_iommu_group_set_kvm);
> +
>  static int vfio_register_group_notifier(struct vfio_group *group,
>  					unsigned long *events,
>  					struct notifier_block *nb)
> @@ -2197,6 +2214,65 @@ int vfio_unregister_notifier(struct device *dev, enum vfio_notify_type type,
>  }
>  EXPORT_SYMBOL(vfio_unregister_notifier);
>  
> +int vfio_iommu_group_register_notifier(struct iommu_group *iommugroup,
> +		enum vfio_notify_type type,
> +		unsigned long *events, struct notifier_block *nb)
> +{
> +	struct vfio_group *group;
> +	int ret;
> +
> +	if (!iommugroup || !nb || !events || (*events == 0))
> +		return -EINVAL;
> +
> +	group = vfio_group_get_from_iommu(iommugroup);
> +	if (!group)
> +		return -ENODEV;
> +
> +	switch (type) {
> +	case VFIO_IOMMU_NOTIFY:
> +		ret = vfio_register_iommu_notifier(group, events, nb);
> +		break;
> +	case VFIO_GROUP_NOTIFY:
> +		ret = vfio_register_group_notifier(group, events, nb);
> +		break;
> +	default:
> +		ret = -EINVAL;
> +	}
> +
> +	vfio_group_put(group);
> +	return ret;
> +}
> +EXPORT_SYMBOL(vfio_iommu_group_register_notifier);
> +
> +int vfio_iommu_group_unregister_notifier(struct iommu_group *grp,
> +		enum vfio_notify_type type, struct notifier_block *nb)
> +{
> +	struct vfio_group *group;
> +	int ret;
> +
> +	if (!grp || !nb)
> +		return -EINVAL;
> +
> +	group = vfio_group_get_from_iommu(grp);
> +	if (!group)
> +		return -ENODEV;
> +
> +	switch (type) {
> +	case VFIO_IOMMU_NOTIFY:
> +		ret = vfio_unregister_iommu_notifier(group, nb);
> +		break;
> +	case VFIO_GROUP_NOTIFY:
> +		ret = vfio_unregister_group_notifier(group, nb);
> +		break;
> +	default:
> +		ret = -EINVAL;
> +	}
> +
> +	vfio_group_put(group);
> +	return ret;
> +}
> +EXPORT_SYMBOL(vfio_iommu_group_unregister_notifier);
> +
>  /**
>   * Module/class support
>   */

On 21/12/16 17:04, David Gibson wrote:
> On Sun, Dec 18, 2016 at 12:28:58PM +1100, Alexey Kardashevskiy wrote:
>> c086de81 "vfio iommu: Add blocking notifier to notify DMA_UNMAP" added
>> notifiers to a VFIO group. However even though the code underneath
>> uses groups, the API takes device struct pointers.
>>
>> This adds helpers which do the same thing but take IOMMU groups instead.
>>
>> This adds vfio_iommu_group_set_kvm() which is a wrapper on top of
>> vfio_group_set_kvm() but also takes an iommu_group.
>>
>> Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
> 
> Adding a second interface in parallel seems dubious.
> 
> Should the existing interface just be replaced with this one?
> 
> Or can the existing interface be re-implemented in terms of this one?

imho this should have been done in the first place but since Alex and I
came to a conclusion that this does not simplify anything in my patchset
(rather the opposite), I am not going to push it further now.

09/11, 10/11, 11/11 from this patchset are superseded by:
[PATCH kernel v3] KVM: PPC: Add in-kernel acceleration for VFIO