Re: [PATCH v2 09/12] iommu/vt-d: Use cache helpers in arch_invalidate_secondary_tlbs

[Baolu Lu]

On 2024/4/10 23:55, Jason Gunthorpe wrote:
> On Wed, Apr 10, 2024 at 10:08:41AM +0800, Lu Baolu wrote:
> >   /* Pages have been freed at this point */
> >   static void intel_arch_invalidate_secondary_tlbs(struct mmu_notifier *mn,
> >   					struct mm_struct *mm,
> >   					unsigned long start, unsigned long end)
> >   {
> >   	struct intel_svm *svm = container_of(mn, struct intel_svm, notifier);
> > +	struct dmar_domain *domain = svm->domain;
> >   
> >   	if (start == 0 && end == -1UL) {
>
> ULONG_MAX ideally.

Done.

> > -		intel_flush_svm_all(svm);
> > +		cache_tag_flush_all(domain);
> >   		return;
> >   	}
> >   
> > -	intel_flush_svm_range(svm, start,
> > -			      (end - start + PAGE_SIZE - 1) >> VTD_PAGE_SHIFT, 0);
> > +	cache_tag_flush_range(domain, start, end, 0);
>
> Be mindful of the note from the ARM driver:
>
>          /*
>           * The mm_types defines vm_end as the first byte after the end address,
>           * different from IOMMU subsystem using the last address of an address
>           * range. So do a simple translation here by calculating size correctly.
>           */
>          size = end - start;

I didn't find any documentation about the @end in this callback, but in
mm subsystem, it does like this,

flush_tlb_mm_range(mm, va, va + nr_pages * PAGE_SIZE, PAGE_SHIFT, false);

So, yes, the @end in arch_invalidate_secondary_tlbs callback is
different from the iommu gather.

I was not aware of this. Thanks for pointing this out.

> Given that the cache_tag_flush_range's are all tied directly to the
> iommu gather API, this is probably missing a -1 though perhaps it does
> not cause a functional problem here.

I will change it like below,

diff --git a/drivers/iommu/intel/svm.c b/drivers/iommu/intel/svm.c
index 858a64fbdaab..15dcd1b30df1 100644
--- a/drivers/iommu/intel/svm.c
+++ b/drivers/iommu/intel/svm.c
@@ -146,7 +146,12 @@ static void 
intel_arch_invalidate_secondary_tlbs(struct mmu_notifier *mn,
                return;
        }

-       cache_tag_flush_range(domain, start, end, 0);
+       /*
+        * The mm_types defines vm_end as the first byte after the end 
address,
+        * different from IOMMU subsystem using the last address of an 
address
+        * range.
+        */
+       cache_tag_flush_range(domain, start, end - 1, 0);
 }

 static void intel_mm_release(struct mmu_notifier *mn, struct mm_struct 
*mm)

Best regards,
baolu