[PATCH RFC cmpxchg 3/8] ARC: Emulate one-byte and two-byte cmpxchg

[Arnd Bergmann]

On Thu, Apr 4, 2024, at 16:44, Paul E. McKenney wrote:
> On Thu, Apr 04, 2024 at 01:57:32PM +0200, Arnd Bergmann wrote:
>> On Tue, Apr 2, 2024, at 19:06, Paul E. McKenney wrote:
>> > Or is your question specific to the 16-bit primitives?  (Full disclosure:
>> > I have no objection to removing Alpha ev4/45/5, having several times
>> > suggested removing Alpha entirely.  And having the scars to prove it.)
>> 
>> For the native sub-word access, alpha ev5 cannot do either of
>> them, while armv3 could do byte access but not 16-bit words.
>> 
>> It sounds like the old alphas are already broken then, which
>> is a good reason to finally drop support. It would appear that
>> the arm riscpc is not affected by this though.
>
> Good point, given that single-byte load/store and emulated single-byte
> cmpxchg() has been in common code for some time.
>
> So armv3 is OK with one-byte emulated cmpxchg(), but not with two-byte,
> which is consistent with the current state of this series in the -rcu
> tree.  (My plan is to wait until Monday to re-send the series in order
> to allow the test robots to find yet more bugs.)
>
> Or is the plan to also drop support for armv3 in the near term?

Russell still has his RiscPC, which is probably the only one
using armv3 kernels (it's actually v4 but relies on -march=armv3
to work around hardware quirks). Since armv3 support was dropped
in gcc-9, it's a matter of time before we drop this as well, but
it's not now.

>> We'd still need all the combinations of cmpxchg() and xchg()
>> with the four sizes and ordering variants, but at least the
>> latter should easily collapse on most architectures. At the
>> moment, most architectures only provide the full-ordering
>> version.
>
> That does make a lot of sense to me.  Though C-language inline functions
> have some trouble with type-generic parameters.
>
> However, my patch is down at architecture-specific level, so should not
> affect the cmpxchg() macro, right?  Or am I missing some aspect of your
> proposed refactoring?

Today, arch_cmpxchg() and its variants are defined in each
architecture to handle some subset of integer sizes.

The way I'd like to do this in the future would be to remove that
macro in favor of arch_{cmp,}xchg{8,16,32,64}{,_relaxed,_release,
_acquire,_local} inline functions that each architecture needs
to provide either directly or through a generic helper, with
all the macros wrapping them moved to common code.

I've been wanting to change this for years now and it never
quite makes it to the top of my todo pile, so I guess it can wait
a little longer.

     Arnd