[PATCH] bpf: avoid warning for unused register_bpf_struct_ops()

[PATCH] bpf: avoid warning for unused register_bpf_struct_ops()

[Arnd Bergmann]

From: Arnd Bergmann <arnd@arndb.de>

The macro originally introduced in commit f6be98d19985 ("bpf, net:
switch to dynamic registration") causes a warning in the new smc code
because of the way it evaluates the arguments:

In file included from include/linux/bpf_verifier.h:7,
                 from net/smc/smc_hs_bpf.c:13:
net/smc/smc_hs_bpf.c: In function 'bpf_smc_hs_ctrl_init':
include/linux/bpf.h:2076:50: error: statement with no effect [-Werror=unused-value]
 2076 | #define register_bpf_struct_ops(st_ops, type) ({ (void *)(st_ops); 0; })
      |                                                  ^~~~~~~~~~~~~~~~
net/smc/smc_hs_bpf.c:139:16: note: in expansion of macro 'register_bpf_struct_ops'
  139 |         return register_bpf_struct_ops(&bpf_smc_hs_ctrl_ops, smc_hs_ctrl);
      |                ^~~~~~~~~~~~~~~~~~~~~~~

Work around this using an inline function that takes the argument,
the same way as the normal implementation. Since the second argument to
register_bpf_struct_ops() is a type rather than an object, this still
has to be a macro, but it can call a new inline helper internally like
the normal one does.

Fixes: 15f295f55656 ("net/smc: bpf: Introduce generic hook for handshake flow")
Cc: Kui-Feng Lee <thinker.li@gmail.com>
Cc: Martin KaFai Lau <martin.lau@kernel.org>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
---
 include/linux/bpf.h | 2 +-
 include/linux/btf.h | 5 +++++
 2 files changed, 6 insertions(+), 1 deletion(-)

diff --git a/include/linux/bpf.h b/include/linux/bpf.h
index 6498be4c44f8..eca6966d4f87 100644
--- a/include/linux/bpf.h
+++ b/include/linux/bpf.h
@@ -2065,7 +2065,7 @@ int bpf_struct_ops_desc_init(struct bpf_struct_ops_desc *st_ops_desc,
 void bpf_map_struct_ops_info_fill(struct bpf_map_info *info, struct bpf_map *map);
 void bpf_struct_ops_desc_release(struct bpf_struct_ops_desc *st_ops_desc);
 #else
-#define register_bpf_struct_ops(st_ops, type) ({ (void *)(st_ops); 0; })
+#define register_bpf_struct_ops(st_ops, type) __register_bpf_struct_ops(st_ops)
 static inline bool bpf_try_module_get(const void *data, struct module *owner)
 {
 	return try_module_get(owner);
diff --git a/include/linux/btf.h b/include/linux/btf.h
index f06976ffb63f..799a75209536 100644
--- a/include/linux/btf.h
+++ b/include/linux/btf.h
@@ -535,6 +535,11 @@ int __register_bpf_struct_ops(struct bpf_struct_ops *st_ops);
 const struct bpf_struct_ops_desc *bpf_struct_ops_find_value(struct btf *btf, u32 value_id);
 const struct bpf_struct_ops_desc *bpf_struct_ops_find(struct btf *btf, u32 type_id);
 #else
+struct bpf_struct_ops;
+static inline int __register_bpf_struct_ops(struct bpf_struct_ops *st_ops)
+{
+	return 0;
+}
 static inline const struct bpf_struct_ops_desc *bpf_struct_ops_find(struct btf *btf, u32 type_id)
 {
 	return NULL;
-- 
2.39.5

Re: [PATCH] bpf: avoid warning for unused register_bpf_struct_ops()

[Martin KaFai Lau]

On 12/4/25 1:42 AM, Arnd Bergmann wrote:
> From: Arnd Bergmann <arnd@arndb.de>
> 
> The macro originally introduced in commit f6be98d19985 ("bpf, net:
> switch to dynamic registration") causes a warning in the new smc code
> because of the way it evaluates the arguments:
> 
> In file included from include/linux/bpf_verifier.h:7,
>                   from net/smc/smc_hs_bpf.c:13:
> net/smc/smc_hs_bpf.c: In function 'bpf_smc_hs_ctrl_init':
> include/linux/bpf.h:2076:50: error: statement with no effect [-Werror=unused-value]
>   2076 | #define register_bpf_struct_ops(st_ops, type) ({ (void *)(st_ops); 0; })
>        |                                                  ^~~~~~~~~~~~~~~~
> net/smc/smc_hs_bpf.c:139:16: note: in expansion of macro 'register_bpf_struct_ops'
>    139 |         return register_bpf_struct_ops(&bpf_smc_hs_ctrl_ops, smc_hs_ctrl);
>        |                ^~~~~~~~~~~~~~~~~~~~~~~
> 
> Work around this using an inline function that takes the argument,
> the same way as the normal implementation. Since the second argument to
> register_bpf_struct_ops() is a type rather than an object, this still
> has to be a macro, but it can call a new inline helper internally like
> the normal one does.

Thanks for the patch. This has been fixed in 
"https://lore.kernel.org/bpf/988c61e5fea280872d81b3640f1f34d0619cfbbf.1764843951.git.geert@linux-m68k.org/" 
to completely remove its usage from smc. The smc usage without 
CONFIG_BPF_JIT was an overlook. This empty register_bpf_struct_ops 
should be removed from the bpf-next tree as a cleanup.