(+explicit Cc: Paul)
Hi Elad,
On Tue, 21 Dec 2021 10:20:52 -0500, Elad Lahav wrote:
> Hi Paul,
>
> As promised, I have a potentially more interesting comment regarding
> Section 9.5. Throughout this section, from the very first example, the
> writer uses release semantics, but the readers are not obligated to
> use acquire semantics, at least on sensible architectures (with
> apologies to employees of HP nee Compaq nee Digital).
;-)
>
> If I understand correctly, these relaxed semantics are the result of
> an address dependency, with the data protected by the RCU critical
> section residing in a structure whose address is stored by the pointer
> dereferenced by a reader. The reader cannot consider any data outside
> of this structure as protected by the critical section. This is a
> critical point without which the examples won't work reliably.
I think this is mostly covered in Section 9.5.2.1 "Publish-Subscribe
Mechanism" and Figure 9.10 "Publication/Subscription Constraints".
>
> Am I missing something? If not, I think that this point should be
> emphasized early on, with a link to Section 15.2.3.
There is such a reference in Section 9.5.2.1 just below Quick Quiz
9.28. You mean you find this too late?
Thanks, Akira
BTW, I couldn't figure out what you meant by "the Spanish
Acquisition"...
>
> --Elad
>
