Re: Explicit pointer provenance APIs for CHERI

Mark Millard <[email protected]> Wed, 24 Jun 2026 11:39:35 -0700
Newsgroups gmane.os.freebsd.architechture
Message-ID <[email protected]>
On 6/24/26 09:31, Alexander Ziaee wrote:
> On 2026-06-24 10:37 -04:00 EDT, "Konstantin Belousov" <[email protected]> wrote:
>> On Wed, Jun 24, 2026 at 10:54:51AM +0000, Poul-Henning Kamp wrote:
>>> --------
>>> Brooks Davis writes:
>>>
>>>> On Mon, Jun 22, 2026 at 10:11:36AM -0700, Adrian Chadd wrote:
>>>
>>>>> Specifically, I'd like a manpage which defines something top level for these
>>>>> terms and link to that manpage from manpages that talk about pointer
>>>>> provenance and anything else that leads into CHERI related stuff.
>>>>
>>>> Lacking an obviously better place I've added a section to arch(7) and
>>>> linked it from the various manpages.  https://reviews.freebsd.org/D57812
>>>
>>> Not bad, but I dont think there's any way around having a proper CHERI(7)
>>> manual page too ?
>>>
>>> It doesn't have to be long, but it has to be there.
>>
>> The concept of the provenance is not specific to cheri, it is needed and
>> used by the compilers on regular architectures.
>>
>> I think that instead of adding stuff to arch(7), we need a man page
>> like memory_model(7).  There is already some items that would be more
>> natural in memory_model.7 than in atomic.9, for instance.  Provenance
>> is also the good subject for discussion there.
> 
> Can we massage it into memory(7)?

My guess is that the "for a non-developer audience" status for memory(7)
may be a big distinction with the proposed memory_model(7), along with
memory(7) being very FreeBSD OS specific. Memory model is common
terminology in various language standards. Most of what I see in the
memory(7) draft would not fit with such material as used in those
standards. CHERI is not OS specific of itself: It is a hardware
architecture extension. Linux and others have been integrating CHERI
support as well, not just one variation on the FreeBSD OS.

> I think that'd be the most obvious place.
> The draft is very focused on just memory statistics, but I think it wouldn't hurt
> it to tweak that a little bit.
> 
> https://reviews.freebsd.org/D18768
> 
> Best,
> Alex
> 


-- 
===
Mark Millard
marklmi at yahoo.com