Re: [PATCH 7/7] iommu: Add iommu_domain::domain_ops

[Jason Gunthorpe]

On Tue, Jan 25, 2022 at 12:59:14PM +0800, Lu Baolu wrote:
> On 1/24/22 5:58 PM, Tian, Kevin wrote:
> > > From: Lu Baolu <baolu.lu@xxxxxxxxxxxxxxx>
> > > Sent: Monday, January 24, 2022 3:11 PM
> > > +/**
> > > + * struct domain_ops - per-domain ops
> > > + * @attach_dev: attach an iommu domain to a device
> > > + * @detach_dev: detach an iommu domain from a device
> > 
> > What is the criteria about whether an op should be iommu_ops or domain_ops
> > when it requires both domain and device pointers like above two (and future
> > PASID-based attach)?
> 
> Generally ops belong to iommu_ops if they are only device oriented, and
> domain_ops if they are only domain oriented. But it's really hard to
> judge when both device and domain are involved. Good question. :-)
> 
> Perhaps we should leave the attach/detach interfaces in iommu_ops since
> they are related to device capabilities. For example, some devices
> support attach_dev_pasid, while others not.

Isn't the attach process for something like SVA or the KVM version
actually rather different code?

Ideally the function pointers should help minimize case statements/etc
inside the driver..

Or stated another way, I expect drivers will have different structs
container_of'ing back to the iommu_domain (ie intel_iommu_domain_sva,
say). Any code that needs to work differently depending on the struct
should have an op in the domain so it can sanely container_of without
a mess.

Jason