Messages in this thread | | | Date | Tue, 17 Jan 2023 07:10:50 -0600 | Subject | Re: [PATCH 4/4] iommu/amd: Force SNP-enabled VFIO domain to 4K page size | From | Eric van Tassell <> |
| |
On 1/10/23 08:31, Suravee Suthikulpanit wrote: > SNP only supports 2M and 4K page sizes. Other page sizes requires > page smashing to supported sizes. For SNP-enabled guests > with pass-through devices (via VFIO), it also requires RMP and IOMMU > page sizes to match. > > To simplify the support, for SNP-enabled guest, SNP will smash guest pages > to 4K page size only, and IOMMU driver will setup the IOMMU v1 page table > for the VFIO domain of the guest to match the 4K page size. > > Co-developed-by: Vasant Hegde <vasant.hegde@amd.com> > Signed-off-by: Vasant Hegde <vasant.hegde@amd.com> > Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com> > --- > drivers/iommu/amd/amd_iommu_types.h | 2 ++ > drivers/iommu/amd/iommu.c | 28 ++++++++++++++++++++++++++++ > 2 files changed, 30 insertions(+) > > diff --git a/drivers/iommu/amd/amd_iommu_types.h b/drivers/iommu/amd/amd_iommu_types.h > index ad124959a26a..5249ac18ce6e 100644 > --- a/drivers/iommu/amd/amd_iommu_types.h > +++ b/drivers/iommu/amd/amd_iommu_types.h > @@ -279,6 +279,7 @@ > #define AMD_IOMMU_PGSIZES ((~0xFFFUL) & ~(2ULL << 38)) > /* 4K, 2MB, 1G page sizes are supported */ > #define AMD_IOMMU_PGSIZES_V2 (PAGE_SIZE | (1ULL << 21) | (1ULL << 30)) > +#define AMD_IOMMU_PGSIZES_4K (PAGE_SIZE) > > /* Bit value definition for dte irq remapping fields*/ > #define DTE_IRQ_PHYS_ADDR_MASK (((1ULL << 45)-1) << 6) > @@ -440,6 +441,7 @@ > #define PD_IOMMUV2_MASK (1UL << 3) /* domain has gcr3 table */ > #define PD_GIOV_MASK (1UL << 4) /* domain enable GIOV support */ > #define PD_VFIO_MASK (1UL << 5) /* domain enable VFIO support */ > +#define PD_SNP_MASK (1UL << 6) /* domain enable SNP support */ > > extern bool amd_iommu_dump; > #define DUMP_printk(format, arg...) \ > diff --git a/drivers/iommu/amd/iommu.c b/drivers/iommu/amd/iommu.c > index a03723930f70..9a1b010a7d00 100644 > --- a/drivers/iommu/amd/iommu.c > +++ b/drivers/iommu/amd/iommu.c > @@ -2422,6 +2422,33 @@ static bool amd_iommu_enforce_cache_coherency(struct iommu_domain *domain) > return true;97921e769dda1 > } > > +static void amd_iommu_set_kvm(struct iommu_domain *domain, struct kvm *kvm) > +{ > + struct protection_domain *pdom = to_pdomain(domain); > + > + if (!amd_iommu_snp_en || !amd_iommu_svm_ops || > + !pdom || !(pdom->flags & PD_VFIO_MASK)) > + return; > + > + /* > + * The parameter kvm can be NULL when calling from kvm_vfio_group_del() > + * and kvm_vfio_destroy(). > + */ > + if (!kvm || > + !amd_iommu_svm_ops->is_snp_guest || > + !amd_iommu_svm_ops->is_snp_guest(kvm)) > + return; > + > + /* > + * VFIO Domain for SNP guest requires IOMMU and RMP page-size to match, > + * which can only support 4K and 2M. Currently, only support 4K > + * IOMMU page-size. > + */ > + pdom->flags |= PD_SNP_MASK; > + pdom->domain.pgsize_bitmap = AMD_IOMMU_PGSIZES_4K; > + pr_debug("%s: Force domain %u page size to 4K.\n", __func__, pdom->id); > +} > +
In my opinion the name of this function is to generic and doesn't describe what it does.
I'd prefer something like amd_iommu_set_4k_pgsz()
> const struct iommu_ops amd_iommu_ops = { > .capable = amd_iommu_capable, > .domain_alloc = amd_iommu_domain_alloc, > @@ -2444,6 +2471,7 @@ const struct iommu_ops amd_iommu_ops = { > .iotlb_sync = amd_iommu_iotlb_sync, > .free = amd_iommu_domain_free, > .enforce_cache_coherency = amd_iommu_enforce_cache_coherency, > + .set_kvm = amd_iommu_set_kvm, > } > }; >
| |