On Wed, Apr 12, 2023 at 1:33 PM Peter Zijlstra [off-list ref] wrote:

On Wed, Apr 05, 2023 at 04:17:08PM +0200, Uros Bizjak wrote:

quoted

diff --git a/arch/powerpc/include/asm/local.h b/arch/powerpc/include/asm/local.h
index bc4bd19b7fc2..45492fb5bf22 100644
--- a/arch/powerpc/include/asm/local.h
+++ b/arch/powerpc/include/asm/local.h

@@ -90,6 +90,17 @@ static __inline__ long local_cmpxchg(local_t *l, long o, long n)
      return t;
 }

+static __inline__ bool local_try_cmpxchg(local_t *l, long *po, long n)
+{
+     long o = *po, r;
+
+     r = local_cmpxchg(l, o, n);
+     if (unlikely(r != o))
+             *po = r;
+
+     return likely(r == o);
+}
+

Why is the ppc one different from the rest? Why can't it use the
try_cmpxchg_local() fallback and needs to have it open-coded?

Please note that ppc directly defines local_cmpxchg that bypasses
cmpxchg_local/arch_cmpxchg_local machinery. The patch takes the same
approach for local_try_cmpxchg, because fallbacks are using
arch_cmpxchg_local definitions.

PPC should be converted to use arch_cmpxchg_local (to also enable
instrumentation), but this is not the scope of the proposed patchset.

Uros.