[PATCH] kbuild: Makefile.lib: fix offsets generation with debug info

[PATCH] kbuild: Makefile.lib: fix offsets generation with debug info

[Dmitry Kandybka]

When debug information is enabled (e.g. -ggdb3), the assembler
output file (e.g. kernel/sched/rq-offsets.s) contains additional
directives such as .ascii with offset info and debug metadata.

For example:

    .ascii "->RQ_nr_pinned $3048 offsetof(struct rq, nr_pinned)"  #
    ...
    .ascii "->func); __data = (it_func_ptr)->data; ((void(*)(void *, pro"

The current sed-offsets macro uses [^ ]* to match constant names,
which is too permissive and can match unexpected characters when
debug directives are present. Additionally, the fallback command
's:->::; p;' strips '->' from any unmatched line and prints it,
which causes debug metadata to leak into the generated header
file (include/generated/rq-offsets.h), breaking the build.

Replace [^ ]* with [[:alnum:]_]* to match only valid C identifiers,
and remove the fallback 's:->::' command so that only successfully
transformed lines are emitted into the generated header.

Fixes: 7dd47b95b0f5 ("kbuild: consolidate redundant sed script ASM offset generation")
Signed-off-by: Dmitry Kandybka <d.kandybka@gmail.com>
---
 scripts/Makefile.lib | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

diff --git a/scripts/Makefile.lib b/scripts/Makefile.lib
index 0a4fdd8bd975..32251eb6ca33 100644
--- a/scripts/Makefile.lib
+++ b/scripts/Makefile.lib
@@ -493,9 +493,8 @@ quiet_cmd_zstd22_with_size = ZSTD22  $@
 # the .ascii directive whereas GCC keeps the <space> as-is.
 define sed-offsets
 	's:^[[:space:]]*\.ascii[[:space:]]*"\(.*\)".*:\1:; \
-	/^->/{s:->#\(.*\):/* \1 */:; \
-	s:^->\([^ ]*\) [\$$#]*\([^ ]*\) \(.*\):#define \1 \2 /* \3 */:; \
-	s:->::; p;}'
+	/^->/{s:->#\(.*\):/* \1 */:p; \
+	s:^->\([[:alnum:]_]*\) [\$$#]*\([^ ]*\) \(.*\):#define \1 \2 /* \3 */:p;}'
 endef
 
 # Use filechk to avoid rebuilds when a header changes, but the resulting file
-- 
2.54.0

Re: [PATCH] kbuild: Makefile.lib: fix offsets generation with debug info

[Nathan Chancellor]

Hi Dmitry,

> When debug information is enabled (e.g. -ggdb3), the assembler
> output file (e.g. kernel/sched/rq-offsets.s) contains additional
> directives such as .ascii with offset info and debug metadata.
> 
> For example:
> 
>     .ascii "->RQ_nr_pinned $3048 offsetof(struct rq, nr_pinned)"  #
>     ...
>     .ascii "->func); __data = (it_func_ptr)->data; ((void(*)(void *, pro"
> 
> The current sed-offsets macro uses [^ ]* to match constant names,
> which is too permissive and can match unexpected characters when
> debug directives are present. Additionally, the fallback command
> 's:->::; p;' strips '->' from any unmatched line and prints it,
> which causes debug metadata to leak into the generated header
> file (include/generated/rq-offsets.h), breaking the build.
> 
> Replace [^ ]* with [[:alnum:]_]* to match only valid C identifiers,
> and remove the fallback 's:->::' command so that only successfully
> transformed lines are emitted into the generated header.
> 
> Fixes: 7dd47b95b0f5 ("kbuild: consolidate redundant sed script ASM offset generation")
> Signed-off-by: Dmitry Kandybka <d.kandybka@gmail.com>
> ---
>  scripts/Makefile.lib | 5 ++---
>  1 file changed, 2 insertions(+), 3 deletions(-)
>
> diff --git a/scripts/Makefile.lib b/scripts/Makefile.lib
> index 0a4fdd8bd975..32251eb6ca33 100644
> --- a/scripts/Makefile.lib
> +++ b/scripts/Makefile.lib
> @@ -493,9 +493,8 @@ quiet_cmd_zstd22_with_size = ZSTD22  $@
>  # the .ascii directive whereas GCC keeps the <space> as-is.
>  define sed-offsets
>  	's:^[[:space:]]*\.ascii[[:space:]]*"\(.*\)".*:\1:; \
> -	/^->/{s:->#\(.*\):/* \1 */:; \
> -	s:^->\([^ ]*\) [\$$#]*\([^ ]*\) \(.*\):#define \1 \2 /* \3 */:; \
> -	s:->::; p;}'
> +	/^->/{s:->#\(.*\):/* \1 */:p; \
> +	s:^->\([[:alnum:]_]*\) [\$$#]*\([^ ]*\) \(.*\):#define \1 \2 /* \3 */:p;}'

Sashiko notes [1] that this results in BLANK() being dropped, which does
not result in any functional change as far as I can tell but it does
kind of make these generated headers a little harder to read in some
cases. Obviously the default fallback is what causes the problem here,
so maybe we can change BLANK() to include something that we can look for
like the name of the macro? This diff on top of your patchset appears to
keep the generated files the same while resolving your issue.

[1]: https://sashiko.dev/#/patchset/20260928105330.25427-1-d.kandybka@gmail.com

diff --git a/include/linux/kbuild.h b/include/linux/kbuild.h
index e7be517aaaf6..945006392cde 100644
--- a/include/linux/kbuild.h
+++ b/include/linux/kbuild.h
@@ -5,7 +5,7 @@
 #define DEFINE(sym, val) \
 	asm volatile("\n.ascii \"->" #sym " %0 " #val "\"" : : "i" (val))
 
-#define BLANK() asm volatile("\n.ascii \"->\"" : : )
+#define BLANK() asm volatile("\n.ascii \"->BLANK()\"" : : )
 
 #define OFFSET(sym, str, mem) \
 	DEFINE(sym, offsetof(struct str, mem))
diff --git a/scripts/Makefile.lib b/scripts/Makefile.lib
index 32251eb6ca33..96c2ec50f06e 100644
--- a/scripts/Makefile.lib
+++ b/scripts/Makefile.lib
@@ -494,7 +494,8 @@ quiet_cmd_zstd22_with_size = ZSTD22  $@
 define sed-offsets
 	's:^[[:space:]]*\.ascii[[:space:]]*"\(.*\)".*:\1:; \
 	/^->/{s:->#\(.*\):/* \1 */:p; \
-	s:^->\([[:alnum:]_]*\) [\$$#]*\([^ ]*\) \(.*\):#define \1 \2 /* \3 */:p;}'
+	s:^->\([[:alnum:]_]*\) [\$$#]*\([^ ]*\) \(.*\):#define \1 \2 /* \3 */:p; \
+	s:->BLANK()::p;}'
 endef
 
 # Use filechk to avoid rebuilds when a header changes, but the resulting file

-- 
Cheers,
Nathan