Re: [PATCH] ARC: atomics: Implement arch_atomic64_cmpxchg using _relaxed

[Vineet Gupta <vgupta@kernel.org>]

On 4/8/25 10:22, Jason Gunthorpe wrote:
> The core atomic code has a number of macros where it elaborates
> architecture primitives into more functions. ARC uses
> arch_atomic64_cmpxchg() as it's architecture primitive which disable alot
> of the additional functions.
>
> Instead provide arch_cmpxchg64_relaxed() as the primitive and rely on the
> core macros to create arch_cmpxchg64().
>
> The macros will also provide other functions, for instance,
> try_cmpxchg64_release(), giving a more complete implementation.
>
> Suggested-by: Mark Rutland <mark.rutland@arm.com>
> Link: https://lore.kernel.org/r/Z0747n5bSep4_1VX@J2N7QTR9R3
> Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>

Acked-by: Vineet Gupta <vgupta@kernel.org>

> ---
>  arch/arc/include/asm/atomic64-arcv2.h | 15 +++++----------
>  1 file changed, 5 insertions(+), 10 deletions(-)
>
> diff --git a/arch/arc/include/asm/atomic64-arcv2.h b/arch/arc/include/asm/atomic64-arcv2.h
> index 9b5791b8547133..73080a664369b4 100644
> --- a/arch/arc/include/asm/atomic64-arcv2.h
> +++ b/arch/arc/include/asm/atomic64-arcv2.h
> @@ -137,12 +137,9 @@ ATOMIC64_OPS(xor, xor, xor)
>  #undef ATOMIC64_OP_RETURN
>  #undef ATOMIC64_OP
>  
> -static inline s64
> -arch_atomic64_cmpxchg(atomic64_t *ptr, s64 expected, s64 new)
> +static inline u64 __arch_cmpxchg64_relaxed(volatile void *ptr, u64 old, u64 new)
>  {
> -	s64 prev;
> -
> -	smp_mb();
> +	u64 prev;
>  
>  	__asm__ __volatile__(
>  	"1:	llockd  %0, [%1]	\n"
> @@ -152,14 +149,12 @@ arch_atomic64_cmpxchg(atomic64_t *ptr, s64 expected, s64 new)
>  	"	bnz     1b		\n"
>  	"2:				\n"
>  	: "=&r"(prev)
> -	: "r"(ptr), "ir"(expected), "r"(new)
> -	: "cc");	/* memory clobber comes from smp_mb() */
> -
> -	smp_mb();
> +	: "r"(ptr), "ir"(old), "r"(new)
> +	: "memory", "cc");
>  
>  	return prev;
>  }
> -#define arch_atomic64_cmpxchg arch_atomic64_cmpxchg
> +#define arch_cmpxchg64_relaxed __arch_cmpxchg64_relaxed
>  
>  static inline s64 arch_atomic64_xchg(atomic64_t *ptr, s64 new)
>  {
>
> base-commit: ea8f6ee2111cd78b32d0363ea630ba9b08ada22d

Thx for this.

FWIW I missed the cmpxchg API when the rest of atomics were relaxed in commit
301014cf6d728 ("ARC: atomic_cmpxchg/atomic_xchg: implement relaxed variants")

Added to ARC for-curr (for next)

Thx,
-Vineet