Date:   Fri, 20 Oct 2023 11:29:24 +0200
From:   Jonas Oberhauser <jonas.oberhauser@...weicloud.com>
To:     paulmck@...nel.org
Cc:     linux-kernel@...r.kernel.org, linux-arch@...r.kernel.org,
        linux-doc@...r.kernel.org, Alan Stern <stern@...land.harvard.edu>,
        Andrea Parri <parri.andrea@...il.com>,
        Will Deacon <will@...nel.org>,
        Peter Zijlstra <peterz@...radead.org>,
        Boqun Feng <boqun.feng@...il.com>,
        Nicholas Piggin <npiggin@...il.com>,
        David Howells <dhowells@...hat.com>,
        Jade Alglave <j.alglave@....ac.uk>,
        Luc Maranget <luc.maranget@...ia.fr>,
        Akira Yokosawa <akiyks@...il.com>,
        Daniel Lustig <dlustig@...dia.com>,
        Joel Fernandes <joel@...lfernandes.org>,
        Jonathan Corbet <corbet@....net>
Subject: Re: [PATCH memory-model] docs: memory-barriers: Add note on compiler
 transformation and address deps


Am 10/19/2023 um 6:39 PM schrieb Paul E. McKenney:
> On Wed, Oct 18, 2023 at 12:11:58PM +0200, Jonas Oberhauser wrote:
>> Hi Paul,
>> [...]
> The compiler is forbidden from inventing pointer comparisons.

TIL :) Btw, do you remember a discussion where this is clarified? A 
quick search didn't turn up anything.


>> Best wishes,
>>
>> jonas
>>
>> Am 10/6/2023 um 6:39 PM schrieb Jonas Oberhauser:
>>> Hi Paul,
>>>
>>> The "more up-to-date information" makes it sound like (some of) the
>>> information in this section is out-of-date/no longer valid.
> The old smp_read_barrier_depends() that these section cover really
> does no longer exist.


You mean that they *intend to* cover? smp_read_barrier_depends never 
appears in the text, so anyone reading this section without prior 
knowledge has no way of realizing that this is what the sections are 
talking about.

On the other hand the implicit address dependency barriers that do exist 
are mentioned in the text. And that part is still true.


>
>>> But after reading the sections, it seems the information is valid, but
>>> discusses mostly the history of address dependency barriers.
>>>
>>> Given that the sepcond part  specifically already starts with a
>>> disclaimer that this information is purely relevant to people interested
>>> in history or working on alpha, I think it would make more sense to
>>> modify things slightly differently.
>>>
>>> Firstly I'd remove the "historical" part in the first section, and add
>>> two short paragraphs explaining that
>>>
>>> - every marked access implies a address dependency barrier
> This is covered in rcu_dereference.rst.

Let me quote a much wiser man than myself here: "

The problem is that people insist on diving into the middle of documents,
so sometimes repetition is a necessary form of self defense.  ;-)

"

The main reason I would like to add this here at the very top is that

- this section serves to frigthen children about the dangers of address 
dependencies,

- never mentions a way to add them - I need to happen to read another 
section of the manual to find that out

- and says this information is historical without specifying which parts 
are still relevant

(and the parts that are still there are all still relevant, while the 
parts that only the authors know was intended to be there and is 
out-of-date is already gone).

So I would add a disclaimer specifying that (since 4.15) *all* marked 
accesses imply read dependency barriers which resolve most of the issues 
mentioned in the remainder of the article.
However, some issues remain because the dependencies that are preserved 
by such barriers are just *semantic* dependencies, and readers should 
check rcu_dereference.rst for examples of what that implies.


> [...]
> most situations would be better served by an _acquire() suffix than by
> a relaxed version of [...] an atomic [...]


I completely agree. I even considered removing address dependencies 
altogether from the company-internal memory models.
But people sometimes get a little bit angry and start asking many questions.
The valuable time of the model maintainer should be considered when 
designing memory models.


>
>>> - address dependencies considered by the model are *semantic*
>>> dependencies, meaning that a *syntactic* dependency is not sufficient to
>>> imply ordering; see the rcu file for some examples where compilers can
>>> elide syntactic dependencies
> There is a bunch of text in rcu_dereference.rst to this effect.  Or
> is there some aspect that is missing from that document?


That's what I meant by "see the rcu file" --- include a link to 
rcu_dereference.rst in that paragraph.
So that people know to check out rcu_dereference.rst for more 
explanations to this effect.


> The longer-term direction, perhaps a few years from now, is for the
> first section to simply reference rcu_dereference.rst and for the second
> section to be removed completely.


Sounds good to me, but that doesn't mean we need to compromise the 
readability in the interim :)


>
> [...]
> The problem is that people insist on diving into the middle of documents,
> so sometimes repetition is a necessary form of self defense.  ;-)
>
> But I very much appreciate your review and feedback, and I also apologize
> for my slowness.
>
> 							Thanx, Paul
>

Thanks for the response, I started thinking my mails aren't getting 
through again.

Have fun,

jonas

Powered by blists - more mailing lists