[PATCH v3 03/27] ring: use compare-and-swap wrapper
From: Stephen Hemminger <stephen@networkplumber.org>
Date: 2026-05-23 19:56:31
Subsystem:
library code, the rest · Maintainers:
Andrew Morton, Linus Torvalds
The rte_atomic32_cmpset is deprecated. Initial attempts at changing this with direct conversion to rte_atomic_compare_exchange_weak_explicit() regressed MP/MC contended performance on x86 by 10-30%, because the C11 builtin's failure-writeback semantic forces GCC to emit extra instructions on the CAS critical path. Add an internal __rte_ring_compare_and_swap() wrapper that calls __sync_bool_compare_and_swap() directly, which keeps the original instruction sequence. Add equivalent function for MSVC. Signed-off-by: Stephen Hemminger <stephen@networkplumber.org> --- lib/ring/rte_ring_generic_pvt.h | 32 ++++++++++++++++++++++++++++---- 1 file changed, 28 insertions(+), 4 deletions(-)
diff --git a/lib/ring/rte_ring_generic_pvt.h b/lib/ring/rte_ring_generic_pvt.h
index affd2d5ba7..0fb972de9e 100644
--- a/lib/ring/rte_ring_generic_pvt.h
+++ b/lib/ring/rte_ring_generic_pvt.h@@ -18,6 +18,30 @@ * For more information please refer to <rte_ring.h>. */ +/** + * @internal optimized version of compare exchange + * + * The C11 builtin's failure-writeback semantic generates worse code on x86. + * Unlike rte_atomic_compare_exchange_*_explicit(), this wrapper does NOT + * write the actual value back to a pointer on failure. Callers in a retry + * loop must reload the expected value explicitly on the next iteration. + * + * Full memory barrier, equivalent to rte_memory_order_seq_cst on both + * success and failure. + */ +static __rte_always_inline bool +__rte_ring_compare_and_swap(volatile uint32_t *dst, + uint32_t expected, uint32_t desired) +{ +#if defined(RTE_TOOLCHAIN_MSVC) + return _InterlockedCompareExchange((volatile long *)dst, + (long)desired, (long)expected) + == (long)expected; +#else + return __sync_bool_compare_and_swap(dst, expected, desired); +#endif +} + /** * @internal This function updates tail values. */
@@ -108,10 +132,10 @@ __rte_ring_headtail_move_head(struct rte_ring_headtail *d, if (is_st) { d->head = *new_head; success = 1; - } else - success = rte_atomic32_cmpset( - (uint32_t *)(uintptr_t)&d->head, - *old_head, *new_head); + } else { + success = __rte_ring_compare_and_swap( + &d->head, *old_head, *new_head); + } } while (unlikely(success == 0)); return n; }
--
2.53.0