From: Arnd Bergmann <arnd@arndb.de>

Building kernels with CONFIG_LD_DEAD_CODE_DATA_ELIMINATION using gcc-11
causes compiler warnings:

kernel/cgroup/rstat.c:65:1: error: 'retain' attribute ignored [-Werror=attributes]

It appears that this version reports __has_attribute(__retain__) as true
but doesn't actually support the feature.

Fixes: 0a5d3258d7c9 ("compiler_types.h: Define __retain for __attribute__((__retain__))")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
---
 include/linux/compiler_types.h | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/include/linux/compiler_types.h b/include/linux/compiler_types.h
index 4684277bdd3b..c295b376a305 100644
--- a/include/linux/compiler_types.h
+++ b/include/linux/compiler_types.h
@@ -153,12 +153,15 @@ static inline void __chk_io_ptr(const volatile void __iomem *ptr) { }
  * being optimized out, but operates at the compiler/IR-level and may still
  * allow unintended removal of objects during linking.
  *
- * Optional: only supported since gcc >= 11, clang >= 13
+ * Optional: only supported since gcc >= 12, clang >= 13
+ *
+ * On gcc-11 the feature is reported to be present but causes build warnings.
  *
  *   gcc: https://gcc.gnu.org/onlinedocs/gcc/Common-Function-Attributes.html#index-retain-function-attribute
  * clang: https://clang.llvm.org/docs/AttributeReference.html#retain
  */
 #if __has_attribute(__retain__) && \
+	(__GNUC__ != 11) && \
 	(defined(CONFIG_LD_DEAD_CODE_DATA_ELIMINATION) || \
 	 defined(CONFIG_LTO_CLANG))
 # define __retain			__attribute__((__retain__))
-- 
2.39.5

On Fri, Jun 20, 2025 at 1:14 PM Arnd Bergmann <arnd@kernel.org> wrote:
>
> Building kernels with CONFIG_LD_DEAD_CODE_DATA_ELIMINATION using gcc-11
> causes compiler warnings:
>
> kernel/cgroup/rstat.c:65:1: error: 'retain' attribute ignored [-Werror=attributes]
>
> It appears that this version reports __has_attribute(__retain__) as true
> but doesn't actually support the feature.

Hmm... I can't reproduce the error in Compiler Explorer:

    https://godbolt.org/z/roMrfjxGM

And locally with a GCC 11.4 I can see the attribute keeping a function
in the final binary when marked as `retain` against `retain` against
`-ffunction-sections -Wl,--gc-sections`.

Is something else going on?

Cheers,
Miguel

On Fri, Jun 20, 2025 at 3:00 PM Miguel Ojeda
<miguel.ojeda.sandonis@gmail.com> wrote:
>
> Is something else going on?

Ah, I see -- it does error on arm and arm64, but not x86:

    https://godbolt.org/z/G4Pbd9Gar

So it is architecture dependent.

Cheers,
Miguel