Re: [PATCH v8 01/12] iommu: Add iommu_attach_device_pasid_handle()

[Jason Gunthorpe <jgg@nvidia.com>]

On Fri, Feb 28, 2025 at 02:48:22PM +0800, Yi Liu wrote:
> But you indeed reminds me. It's improper to have the
> __iommu_attach_device_pasid() in header.

If you do it would be wrapped in an inline for users.

Ie you'd have  iommu_attach_device_pasid() &
iommu_attach_device_pasid_handle() with the !NULL validation just call
out to __iommu_attach_device_pasid()

> > So, overall it doesn't feel very necessary to me...
> > 
> > The iommu_attach_group/iommu_attach_group_handle() case, on the
> > other hand, does take care of some underlying difference. Thus,
> > having two separate functions could make things clear.
> 
> actually, they can share code as well. The below patch actually has started
> it. Though dropped to not blocking the pasid series.

Yeah, I think we can get to the point where all domain attach,
handle/nohandle/pasid/nopasid just calls one function that manipulates
the xarray. You got pretty close in that series

The PASID0 special cases can be handled with flags.

Jason