RE: [RFC PATCH 5/6] iommu/amd: Introduce helper functions to setup GCR3TRPMode

[Tian, Kevin]

> From: Jason Gunthorpe <jgg@xxxxxxxxxx>
> Sent: Wednesday, December 13, 2023 9:53 PM
> 
> On Tue, Dec 12, 2023 at 10:01:38AM -0600, Suravee Suthikulpanit wrote:
> > +/*
> > + * For GCR3TRPMode, user-space provides GPA for the GCR3 Root Pointer
> Table.
> > + */
> > +int amd_iommu_set_gcr3tbl_trp(struct amd_iommu *iommu, struct
> pci_dev *pdev,
> > +			      u64 gcr3_tbl, u16 glx, u16 guest_paging_mode)
> > +{
> > +	struct iommu_dev_data *dev_data = dev_iommu_priv_get(&pdev-
> >dev);
> > +	struct dev_table_entry *dev_table = get_dev_table(iommu);
> > +	struct gcr3_tbl_info *gcr3_info = &dev_data->gcr3_info;
> > +	int devid = pci_dev_id(pdev);
> > +	u64 data0 = dev_table[devid].data[0];
> > +	u64 data1 = dev_table[devid].data[1];
> > +	u64 data2 = dev_table[devid].data[2];
> > +	u64 tmp;
> 
> Like I said in my other email, this whole function is conceptually
> wrong - you can't read the DTE to learn the parent domain's
> contribution to the nesting DTE and you can't write to the DTE during
> allocation of a domain!
> 

Agree. DTE is updated only at attach/detach. domain allocation should
involve things only about the domain itself.