On Wed, 24 Aug 2022 21:44:22 +0200 Johannes Berg wrote:

On Tue, 2022-08-23 at 21:50 -0700, Jakub Kicinski wrote:

quoted

+/* Report that a root attribute is missing */
+#define GENL_REQ_ATTR_CHECK(info, attr) ({				\
+	struct genl_info *__info = (info);				\
+	u32 __attr = (attr);						\
+	int __retval;							\
+									\
+	__retval = !__info->attrs[__attr];				\
+	if (__retval)							\
+		NL_SET_ERR_ATTR_MISS(__info->extack,			\
+				     __info->userhdr ? : __info->genlhdr, \
+				     __attr);				\
+	__retval;							\
+})  

Not sure this needs to be a macro btw, could be an inline returning a
bool? You're not really expanding anything here, nor doing something
with strings (unlike GENL_SET_ERR_MSG for example.)

Initially I typed up both flavors with and without the message 
but I dropped the _MSG() one since I didn't find a strong enough 
reason to use it.

If we do get the _MSG() version at some point (perhaps to preserve 
an existing message during conversion?) having different case
would seem off.

I have no opinion which way is better, LMK if you prefer lower case
(ignoring existing non-MSG helpers being upper case) and I'll sed thru
the patches, no problem at all.