Re: [PATCH] xen/arm: fix iomem_ranges cfg in map_range_to_domain()

[Julien Grall <julien@xxxxxxx>]

Hi Andrew,

On 14/02/2025 14:25, Andrew Cooper wrote:
> On 14/02/2025 2:10 pm, Grygorii Strashko wrote:
> > > > For example, requested range:
> > > >     00e6140000 - 00e6141004 should set e6140:e6141 nr=2, but will
> > > > configure
> > > > e6140 e6142 nr=3 instead.
> > > I am not sure what 'nr' is referring to here.
> > Sorry, will change to "num_pages"?
> I agree Xen needs to be better (and by that, I mean consistent and
> clear), but there are subtle bugs with most approaches like this.
>
> Any exclusive bound, as well as counts like this need $n+1 bits of
> arithmetic when you want to describe the boundaries of the space.
>
> There is also a boundary condition for counts.  What map_foo(x, 0) mean?
I would consider "0" has invalid.

> Real hardware uses "last" for describing boundaries (x86 specifically
> calls this "limit" in the ISA, but it's a trick used by other
> architectures too).  Unlike "end", it's clearly an inclusive bound.
>
> Personally, I'd like to see Xen switch to "start, last" pairs.  It's
> unambiguous and has fewest boundary cases.
"start, last" would also work for me.

Cheers,

--
Julien Grall