[PATCH bpf-next] locking/atomic: arc, hexagon, openrisc: Define arch_cmpxchg_local()

[PATCH bpf-next] locking/atomic: arc, hexagon, openrisc: Define arch_cmpxchg_local()

[Sasha Levin]

Commit ccb88db8dac8 ("bpf: Don't hand out a linked element as extra
element") made alloc_htab_elem() take the per-cpu spare element with
try_cmpxchg_local(), the first caller of cmpxchg_local() outside of
architecture code. It doesn't build on arc, hexagon and openrisc:

  include/linux/atomic/atomic-arch-fallback.h:384:27: error: implicit declaration of function 'arch_cmpxchg_local'; did you mean 'raw_cmpxchg_local'? [-Werror=implicit-function-declaration]

raw_cmpxchg_local() maps to arch_cmpxchg_local() unconditionally, but
these three architectures don't define it, and unlike nios2 and
microblaze they don't get it from asm-generic/cmpxchg.h either.

cmpxchg_local() only has to be atomic with respect to the local CPU, so
an architecture's own arch_cmpxchg() can serve for the sizes it
supports. arc uses its relaxed LLOCK/SCOND variant when ARC_HAS_LLSC is
set and its atomic_ops_lock() variant otherwise, and openrisc uses its
LL/SC arch_cmpxchg(), the way arm64, riscv and s390 alias theirs. The
hexagon arch_cmpxchg() only handles 32-bit words, so hexagon uses the
generic version that disables interrupts, like sh, sparc32 and xtensa
do. Linux has no NMIs on hexagon, so that is sufficient there.

Fixes: ccb88db8dac8 ("bpf: Don't hand out a linked element as extra element")
Reported-by: kernelci.org bot <bot@kernelci.org>
Closes: https://d.kernelci.org/i/maestro:78b65519ee4561df1fbf1e72f1dec9da7e5b153e
Reported-by: kernel test robot <lkp@intel.com>
Closes: https://lore.kernel.org/oe-kbuild-all/202610181750.BoqAddaF-lkp@intel.com/
Closes: https://lore.kernel.org/oe-kbuild-all/202610182048.aEaiBP8W-lkp@intel.com/
Signed-off-by: Sasha Levin <sashal@kernel.org>
---
The only user is in bpf-next, so it is probably easiest to take this
through bpf-next with acks from the arc, hexagon and openrisc
maintainers. A similar change was posted in 2023 for objpool and
stalled once objpool stopped using try_cmpxchg_local():
https://lore.kernel.org/all/20231121142347.241356-1-wuqiang.matt@bytedance.com/

Build tested on arc with and without LLSC (including both KernelCI
configs), hexagon (clang) and openrisc.

 arch/arc/include/asm/cmpxchg.h      | 4 ++++
 arch/hexagon/include/asm/cmpxchg.h  | 6 ++++++
 arch/openrisc/include/asm/cmpxchg.h | 2 ++
 3 files changed, 12 insertions(+)

diff --git a/arch/arc/include/asm/cmpxchg.h b/arch/arc/include/asm/cmpxchg.h
index 66b3743c94008..93f7153eb267a 100644
--- a/arch/arc/include/asm/cmpxchg.h
+++ b/arch/arc/include/asm/cmpxchg.h
@@ -59,6 +59,8 @@
 	_prev_;								\
 })
 
+#define arch_cmpxchg_local	arch_cmpxchg_relaxed
+
 #else
 
 #define arch_cmpxchg(ptr, old, new)				        \
@@ -80,6 +82,8 @@
 	_prev_;								\
 })
 
+#define arch_cmpxchg_local	arch_cmpxchg
+
 #endif
 
 /*
diff --git a/arch/hexagon/include/asm/cmpxchg.h b/arch/hexagon/include/asm/cmpxchg.h
index 9c58fb81f7fd6..e6114e713db9f 100644
--- a/arch/hexagon/include/asm/cmpxchg.h
+++ b/arch/hexagon/include/asm/cmpxchg.h
@@ -8,6 +8,8 @@
 #ifndef _ASM_CMPXCHG_H
 #define _ASM_CMPXCHG_H
 
+#include <asm-generic/cmpxchg-local.h>
+
 /*
  * __arch_xchg - atomically exchange a register and a memory location
  * @x: value to swap
@@ -72,4 +74,8 @@ __arch_xchg(unsigned long x, volatile void *ptr, int size)
 	__oldval;						\
 })
 
+#define arch_cmpxchg_local(ptr, o, n)					\
+	((__typeof__(*(ptr)))__generic_cmpxchg_local((ptr),		\
+			(unsigned long)(o), (unsigned long)(n), sizeof(*(ptr))))
+
 #endif /* _ASM_CMPXCHG_H */
diff --git a/arch/openrisc/include/asm/cmpxchg.h b/arch/openrisc/include/asm/cmpxchg.h
index 8ee151c072e41..fcf1bc8db8fe9 100644
--- a/arch/openrisc/include/asm/cmpxchg.h
+++ b/arch/openrisc/include/asm/cmpxchg.h
@@ -140,6 +140,8 @@ static inline unsigned long __cmpxchg(volatile void *ptr, unsigned long old,
 					       sizeof(*(ptr)));		\
 	})
 
+#define arch_cmpxchg_local	arch_cmpxchg
+
 /*
  * This function doesn't exist, so you'll get a linker error if
  * something tries to do an invalidly-sized xchg().

base-commit: a6303535c2ad9d49598d52556183911f02287791
-- 
2.53.0

Re: [PATCH bpf-next] locking/atomic: arc, hexagon, openrisc: Define arch_cmpxchg_local()

[Alexei Starovoitov]

On Sun Oct 18, 2026 at 1:29 PM UTC, Sasha Levin wrote:
> Commit ccb88db8dac8 ("bpf: Don't hand out a linked element as extra
> element") made alloc_htab_elem() take the per-cpu spare element with
> try_cmpxchg_local(), the first caller of cmpxchg_local() outside of
> architecture code. It doesn't build on arc, hexagon and openrisc:
>
>   include/linux/atomic/atomic-arch-fallback.h:384:27: error: implicit declaration of function 'arch_cmpxchg_local'; did you mean 'raw_cmpxchg_local'? [-Werror=implicit-function-declaration]
>
> raw_cmpxchg_local() maps to arch_cmpxchg_local() unconditionally, but
> these three architectures don't define it, and unlike nios2 and
> microblaze they don't get it from asm-generic/cmpxchg.h either.
>
> cmpxchg_local() only has to be atomic with respect to the local CPU, so
> an architecture's own arch_cmpxchg() can serve for the sizes it
> supports. arc uses its relaxed LLOCK/SCOND variant when ARC_HAS_LLSC is
> set and its atomic_ops_lock() variant otherwise, and openrisc uses its
> LL/SC arch_cmpxchg(), the way arm64, riscv and s390 alias theirs. The
> hexagon arch_cmpxchg() only handles 32-bit words, so hexagon uses the
> generic version that disables interrupts, like sh, sparc32 and xtensa
> do. Linux has no NMIs on hexagon, so that is sufficient there.
>
> Fixes: ccb88db8dac8 ("bpf: Don't hand out a linked element as extra element")
> Reported-by: kernelci.org bot <bot@kernelci.org>
> Closes: https://d.kernelci.org/i/maestro:78b65519ee4561df1fbf1e72f1dec9da7e5b153e
> Reported-by: kernel test robot <lkp@intel.com>
> Closes: https://lore.kernel.org/oe-kbuild-all/202610181750.BoqAddaF-lkp@intel.com/
> Closes: https://lore.kernel.org/oe-kbuild-all/202610182048.aEaiBP8W-lkp@intel.com/
> Signed-off-by: Sasha Levin <sashal@kernel.org>
> ---
> The only user is in bpf-next, so it is probably easiest to take this
> through bpf-next with acks from the arc, hexagon and openrisc
> maintainers. A similar change was posted in 2023 for objpool and
> stalled once objpool stopped using try_cmpxchg_local():
> https://lore.kernel.org/all/20231121142347.241356-1-wuqiang.matt@bytedance.com/
>
> Build tested on arc with and without LLSC (including both KernelCI
> configs), hexagon (clang) and openrisc.

Thanks for the quick fix. Applied.

Re: [PATCH bpf-next] locking/atomic: arc, hexagon, openrisc: Define arch_cmpxchg_local()

[patchwork-bot+netdevbpf]

Hello:

This patch was applied to bpf/bpf-next.git (master)
by Alexei Starovoitov <ast@kernel.org>:

On Sun, 18 Oct 2026 09:29:07 -0400 you wrote:
> Commit ccb88db8dac8 ("bpf: Don't hand out a linked element as extra
> element") made alloc_htab_elem() take the per-cpu spare element with
> try_cmpxchg_local(), the first caller of cmpxchg_local() outside of
> architecture code. It doesn't build on arc, hexagon and openrisc:
> 
>   include/linux/atomic/atomic-arch-fallback.h:384:27: error: implicit declaration of function 'arch_cmpxchg_local'; did you mean 'raw_cmpxchg_local'? [-Werror=implicit-function-declaration]
> 
> [...]

Here is the summary with links:
  - [bpf-next] locking/atomic: arc, hexagon, openrisc: Define arch_cmpxchg_local()
    https://git.kernel.org/bpf/bpf-next/c/8cebaa3be829

You are awesome, thank you!
-- 
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html