Re: [PATCH v6 3/4] iommu: debug-pagealloc: Track IOMMU pages

[Jason Gunthorpe <jgg@ziepe.ca>]

On Mon, Jan 12, 2026 at 10:20:14AM +0000, Mostafa Saleh wrote:
> On Fri, Jan 9, 2026 at 7:51 PM Jason Gunthorpe <jgg@ziepe.ca> wrote:
> >
> > On Fri, Jan 09, 2026 at 05:18:04PM +0000, Mostafa Saleh wrote:
> > > +static struct page_ext *get_iommu_page_ext(phys_addr_t phys)
> > > +{
> > > +     struct page *page = phys_to_page(phys);
> > > +     struct page_ext *page_ext = page_ext_get(page);
> > > +
> > > +     return page_ext;
> > > +}
> > > +
> > > +static struct iommu_debug_metadata *get_iommu_data(struct page_ext *page_ext)
> > > +{
> > > +     return page_ext_data(page_ext, &page_iommu_debug_ops);
> > > +}
> > > +
> > > +static void iommu_debug_inc_page(phys_addr_t phys)
> > > +{
> > > +     struct page_ext *page_ext = get_iommu_page_ext(phys);
> > > +     struct iommu_debug_metadata *d = get_iommu_data(page_ext);
> >
> > You cannot do this - phys_to_page() can only be called if we already
> > know that phys is a struct page backed item and by the time you get
> > here that information is lost.
> >
> > Probably the only way to resolve this is to somehow pass in an iommu
> > prot flag that can tell the difference between struct page and
> > non-struct page addresses.
> >
> > But I have to NAK this approach of blindly calling phys_to_page().
>
> The callers to this, first will check "pfn_valid", which is the right
> check AFAICT (looking at similar patterns in page_owner for example).

I'm not sure pfn_valid really works in all cases, it has a number of
exclusions that can be relevant when phys_addr_t can be a MMIO
address..

So far we haven't been using it in the DMA paths at all, I'm not so
keen to see that start..

Jason