Re: [PATCH v5 6/9] iommu/arm-smmu-v3: Move CD table to arm_smmu_master

[Jason Gunthorpe <jgg@nvidia.com>]

On Thu, Aug 10, 2023 at 03:34:49PM +0100, Will Deacon wrote:
> On Thu, Aug 10, 2023 at 05:45:03PM +0800, Michael Shavit wrote:
> > On Wed, Aug 9, 2023 at 9:50 PM Will Deacon <will@kernel.org> wrote:
> > >
> > > On Wed, Aug 09, 2023 at 01:12:02AM +0800, Michael Shavit wrote:
> > > > @@ -2203,7 +2186,7 @@ static int arm_smmu_domain_finalise(struct iommu_domain *domain,
> > > >               ias = min_t(unsigned long, ias, VA_BITS);
> > > >               oas = smmu->ias;
> > > >               fmt = ARM_64_LPAE_S1;
> > > > -             finalise_stage_fn = arm_smmu_domain_finalise_s1;
> > > > +             finalise_stage_fn = arm_smmu_domain_finalise_cd;
> > >
> > > Why is this a better name? Now we have inconsistency with
> > > arm_smmu_domain_finalise_s2().
> > 
> > There was a time where s1cfg represented the entire STE and carried
> > the entire cd table. We've gotten rid of s1cfg, and now only store
> > arm_smmu_ctx_desc in the arm_smmu_domain for stage 1 domains.
> > arm_smmu_domain_finalise_cd is IMO more clear, especially given the
> > historical baggage around `s1`.
> 
> Ok, but it's the inconsistency I object to. I don't think it's clear at
> all to have arm_smmu_domain_finalise_cd() and arm_smmu_domain_finalise_s2().
> 
> The easiest thing is to leave it as-is.

Well the names have become wonky.

arm_smmu_domain_finalise_cd() is filling in the struct
arm_smmu_ctx_desc which is mostly the precomputed value for the CD
table entry, which is mostly redundant copies of the values of the
underlying pgtbl_cfg :\

But I agree keeping it named s1 is more consistent with the naming I
think we should use - domains are called S1 or S2 domains depending on
their IOPTE format. 

But arm_smmu_domain_assign_asid/vmid is a generally clearer name for
both :\

Jason