RE: [PATCH v2 1/4] iommu/vt-d: Check before setting PGSNP bit in pasid table entry

["Tian, Kevin" <kevin.tian@intel.com>]

> From: Lu Baolu <baolu.lu@linux.intel.com>
> Sent: Friday, April 22, 2022 9:04 PM
> 
> On 2022/4/22 10:47, Tian, Kevin wrote:
> >> From: Lu Baolu
> >> Sent: Thursday, April 21, 2022 7:36 PM
> >>
> >> The latest VT-d specification states that the PGSNP field in the pasid
> >> table entry should be treated as Reserved(0) for implementations not
> >> supporting Snoop Control (SC=0 in the Extended Capability Register).
> >> This adds a check before setting the field.
> >>
> >> Signed-off-by: Lu Baolu<baolu.lu@linux.intel.com>
> >> ---
> >>   drivers/iommu/intel/pasid.c | 13 ++++++++++---
> >>   1 file changed, 10 insertions(+), 3 deletions(-)
> >>
> >> diff --git a/drivers/iommu/intel/pasid.c b/drivers/iommu/intel/pasid.c
> >> index f8d215d85695..5cb2daa2b8cb 100644
> >> --- a/drivers/iommu/intel/pasid.c
> >> +++ b/drivers/iommu/intel/pasid.c
> >> @@ -625,8 +625,14 @@ int intel_pasid_setup_first_level(struct
> intel_iommu
> >> *iommu,
> >>   		}
> >>   	}
> >>
> >> -	if (flags & PASID_FLAG_PAGE_SNOOP)
> >> -		pasid_set_pgsnp(pte);
> >> +	if (flags & PASID_FLAG_PAGE_SNOOP) {
> >> +		if (ecap_sc_support(iommu->ecap)) {
> >> +			pasid_set_pgsnp(pte);
> >> +		} else {
> >> +			pasid_clear_entry(pte);
> >> +			return -EINVAL;
> >> +		}
> >> +	}
> >>
> >>   	pasid_set_domain_id(pte, did);
> >>   	pasid_set_address_width(pte, iommu->agaw);
> >> @@ -710,7 +716,8 @@ int intel_pasid_setup_second_level(struct
> >> intel_iommu *iommu,
> >>   	pasid_set_fault_enable(pte);
> >>   	pasid_set_page_snoop(pte, !!ecap_smpwc(iommu->ecap));
> >>
> >> -	if (domain->domain.type == IOMMU_DOMAIN_UNMANAGED)
> >> +	if (ecap_sc_support(iommu->ecap) &&
> >> +	    domain->domain.type == IOMMU_DOMAIN_UNMANAGED)
> >>   		pasid_set_pgsnp(pte);
> >>
> > This should be rebased on top of Jason's enforce coherency series
> > instead of blindly setting it. No matter whether it's legacy mode
> > where we set SNP in PTE or scalable mode where we set PGSNP
> > in PASID entry for entire page table, the trigger point should be
> > same i.e. when someone calls enforce_cache_coherency().
> 
> With Jason's enforce coherency series merged, we even don't need to set
> PGSNP bit of a pasid entry for second level translation. 2nd level
> always supports SNP in PTEs, so set PGSNP in pasid table entry is
> unnecessary.
> 

Yes, this sounds correct for 2nd-level.

but setting PGSNP of 1st level translation is also relevant to that
change when talking about enforcing coherency in the guest. In
this case PASID_FLAG_PAGE_SNOOP should be set also after
enforce_cache_coherency() is called.

Currently it's always set for unmanaged domain in
domain_setup_first_level():

	if (domain->domain.type == IOMMU_DOMAIN_UNMANAGED)
		flags |= PASID_FLAG_PAGE_SNOOP;

Suppose we need a separate interface to update PGSNP after pasid
entry is set up.

Thanks
Kevin