Re: [PATCH v3] io: add io_pgtable abstraction

[Boris Brezillon]

On Mon, 17 Nov 2025 16:34:06 +0000
Alice Ryhl <aliceryhl@xxxxxxxxxx> wrote:

> On Wed, Nov 12, 2025 at 09:57:09AM -0300, Daniel Almeida wrote:
> > Hi Alice,
> >   
> > > On 12 Nov 2025, at 07:15, Alice Ryhl <aliceryhl@xxxxxxxxxx> wrote:
> > > +    /// Map a physically contiguous range of pages of the same size.
> > > +    ///
> > > +    /// # Safety
> > > +    ///
> > > +    /// * This page table must not contain any mapping that overlaps with the mapping created by
> > > +    ///   this call.  
> > 
> > I don't think there is a restriction that forbids you from mapping a region that has already been mapped.
> > 
> > If A->B->C are adjacent regions, and you try to map A -> C when B->C is already mapped, I think this
> > call will simply return length(A->B).  
> 
> I think that such overlapping ranges are not allowed. It triggers a
> warn:
> 
>         } else if (pte) {                                                           
>                 /* We require an unmap first */                                 
>                 WARN_ON(!(cfg->quirks & IO_PGTABLE_QUIRK_NO_WARN));             
>                 return -EEXIST;

I confirm the io_pgtable framework (or at least some of its
implementations) doesn't allow you to do that (remap without an
explicit unmap coming first).