Re: port-xen/52016: Mismatch between mem_clusters and pmap_enter bounds check

[Manuel Bouyer <bouyer%antioche.eu.org@localhost>]

On Wed, Mar 01, 2017 at 12:05:00PM +0000, coypu%SDF.ORG@localhost wrote:
> The following reply was made to PR port-xen/52016; it has been noted by GNATS.
> 
> From: coypu%SDF.ORG@localhost
> To: Manuel Bouyer <bouyer%antioche.eu.org@localhost>
> Cc: gnats-bugs%NetBSD.org@localhost
> Subject: Re: port-xen/52016: Mismatch between mem_clusters and pmap_enter
>  bounds check
> Date: Wed, 1 Mar 2017 12:03:10 +0000
> 
>  From a really far from sufficiently thorough look at linux, it looks
>  like it might be equivalent to putting stuff that is directly accessed
>  within [pmap_pa_start,pmap_pa_end) and having magic to map it to
>  "identity".

But then you may have holes in between pmap_pa_start and pmap_pa_end.
I think we should keep pmap_pa* for pseudo-physical addresses only,
and add something else for machine addresses - if we have a way to
know mappable machine addresses.
One problem is that you may want to map machine addresses which falls into
the pseudo-physical address range via /dev/mem. In this case you don't have
a way to know which kind of address this is.

Using /dev/mem for both kind of addresses is bogus.

>  
>  And we can decide for domU that some stuff points to a dummy page and
>  so HYPERVISOR_mmu_update doesn't fail for it.

a domU doesn't need to map machine addresses via /dev/mem (and the hypervisor
won't allow it).

-- 
Manuel Bouyer <bouyer%antioche.eu.org@localhost>
     NetBSD: 26 ans d'experience feront toujours la difference
--