Re: [PATCH v2 29/39] xen/riscv: add definition of __read_mostly

[Jan Beulich <jbeulich@xxxxxxxx>]

On 21.12.2023 16:23, Andrew Cooper wrote:
> On 12/12/2023 5:04 pm, Jan Beulich wrote:
>> On 24.11.2023 11:30, Oleksii Kurochko wrote:
>>> The definition of __read_mostly should be removed in:
>>> https://lore.kernel.org/xen-devel/f25eb5c9-7c14-6e23-8535-2c66772b333e@xxxxxxxx/
>> Andrew, can we settle on what to do with that patch? If you don't like me
>> putting __read_mostly in xen/cache.h (consistent with __ro_after_init),
>> would you please make an alternative suggestion?
> 
> xen/linkage.h?  xen/sections.h?

Well, that's the problem: There's no xen/sections.h (and I don't see why we'd
introduce one just for this), and while xen/linkage.h is about to appear for
the entry point annotations, I don't think it is a reasonable fit.

Otoh with __ro_after_init ...

> Sorry - I didn't mean to block it specifically, but I do think
> xen/cache.h is the wrong place for both to live and that it's a small
> enough change to warrant sorting out nicely once and for all.

... living in xen/cache.h already, __read_mostly is even more logical to put
there than __ro_after_init. For the latter I agree the purpose isn't really
cache related, while for the former it is. From your original reply to the
patch submission, I have to admit I don't really understand what
"micro-architectural" detail you mean: The goal of avoiding unnecessary cache
line ping-pong doesn't look to be what you mean, as imo that's a valid (and
generic) goal. And that's what I think is the reason the #define presently
lives in asm/cache.h, justifying my moving of it to xen/cache.h.

So if you can't get yourself to accept xen/cache.h as the new (even if only
temporary) location, I think you can be expected to make a better proposal.
With "better" there meaning you supplying a reason for why you think that
placement is better than xen/cache.h. For example, if I knew what you
expected xen/sections.h to further contain (in the long run), I might find
myself agreeing to that. Yet other section annotation #define-s live
elsewhere anyway, with - in particular - __init and friends imo not likely
to move out of their present header (xen/init.h).

Jan