Re: [PATCH RFC v2 03/11] iommu/sva: Add iommu_domain type for SVA

[Jason Gunthorpe <jgg@nvidia.com>]

On Tue, Mar 29, 2022 at 01:37:52PM +0800, Lu Baolu wrote:
> @@ -95,6 +101,7 @@ struct iommu_domain {
>  	void *handler_token;
>  	struct iommu_domain_geometry geometry;
>  	struct iommu_dma_cookie *iova_cookie;
> +	struct iommu_sva_cookie *sva_cookie;

Cookie is still the wrong word to use here

> +struct iommu_sva_cookie {
> +	struct mm_struct *mm;
> +	ioasid_t pasid;
> +	refcount_t users;

Really surprised to see a refcount buried inside the iommu_domain..

This design seems inside out, the SVA struct should 'enclose' the domain, not
be a pointer inside it.

struct iommu_sva_domain {
       struct kref_t kref;
       struct mm_struct *mm;
       ioasid_t pasid;

       /* All the domains that are linked to this */
       struct xarray domain_list;
};

And then you could have a pointer to that inside the mm_struct instead
of just the naked pasid.

> +static __maybe_unused struct iommu_domain *

Why maybe unused?

> +iommu_sva_get_domain(struct device *dev, struct mm_struct *mm)
> +{
> +	struct iommu_domain *domain;
> +	ioasid_t pasid = mm->pasid;
> +
> +	if (pasid == INVALID_IOASID)
> +		return NULL;
> +
> +	domain = xa_load(&sva_domain_array, pasid);
> +	if (!domain)
> +		return iommu_sva_alloc_domain(dev, mm);
> +	iommu_sva_domain_get_user(domain);

This assumes any domain is interchangeable with any device, which is
not the iommu model. We need a domain op to check if a device is
compatiable with the domain for vfio an iommufd, this should do the
same.

It means each mm can have a list of domains associated with it and a
new domain is auto-created if the device doesn't work with any of the
existing domains.

Jason