Re: [PATCH v2 6/8] iommu/arm-smmu-v3: Refactor write_ctx_desc

[Michael Shavit]

On Tue, Aug 1, 2023 at 10:18 PM Jason Gunthorpe <jgg@xxxxxxxxxx> wrote:
> You know, you should try to keep the function instead of duplicating
> these
>
> arm_smmu_write_ctx_desc_devices()
>
> And put the four lines in there?

Urhhh yes, I thought I had a reason for this but probably just a lapse
of judgement. Done.

> -     ret = arm_smmu_write_ctx_desc(smmu_domain, mm->pasid, cd);
> +     spin_lock_irqsave(&smmu_domain->devices_lock, flags);
> +     list_for_each_entry(master, &smmu_domain->devices, domain_head) {
> +             ret = arm_smmu_write_ctx_desc(master, mm->pasid, cd);
> +     }
> +     spin_unlock_irqrestore(&smmu_domain->devices_lock, flags);

Just noticed that this is problematic; we may not notice a failure if
it occurs on an earlier iteration of the loop. Will fix.

>
> > @@ -987,19 +985,14 @@ static void arm_smmu_sync_cd(struct arm_smmu_domain *smmu_domain,
> >               },
> >       };
> >
> > -     if (!smmu_domain->cd_table.installed)
> > +     if (!master->domain->cd_table.installed)
> >               return;
>
> BTW, do you have locking for this? I didn't check carefully

This is one of the reasons I wanted to take this as a parameter to the
function. This relies on callers guaranteeing that the cd table not be
attached/detached while this call is in progress. This works now
because:
1. No domain may be attached/detached while SVA is enabled, which is
most of the calls that lead to arm_smmu_sync_cd
2. The other call to arm_smmu_write_ctx_desc in arm-smmu-v3.c is more
obviously serialized with operations that detach/attach the cd table.

Maybe this should at least be a comment on arm_smmu_write_ctx_desc, if
not a lock?


Speaking of.... I should probably flip this bit to false in patch 5
when the cd table is detached.