Skip to contentSkip to conversation

genksyms: Teach parser about 128-bit built-in types

PATCH In review Will Deacon 3 participants 4 messages Jun 18, 2019 – Jun 21, 2019

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 genksyms: Teach parser about 128-bit built-in types

__uint128_t crops up in a few files that export symbols to modules, so teach genksyms about it and the other GCC built-in 128-bit integer types so that we don't end up skipping the CRC generation for some symbols due to the parser failing to spot them:

  | WARNING: EXPORT symbol "kernel_neon_begin" [vmlinux] version
  |          generation failed, symbol will not be versioned.
  | ld: arch/arm64/kernel/fpsimd.o: relocation R_AARCH64_ABS32 against
  |     `__crc_kernel_neon_begin' can not be used when making a shared
  |     object
  | ld: arch/arm64/kernel/fpsimd.o:(.data+0x0): dangerous relocation:
  |     unsupported relocation
Cc Richard Henderson <[email protected]>
Cc Masahiro Yamada <[email protected]>
Cc Ard Biesheuvel <[email protected]>

---

Without this patch, we're seeing arm64 build breakage in linux-next under some configurations.

 scripts/genksyms/keywords.c | 4 ++++
 scripts/genksyms/parse.y    | 2 ++
 2 files changed, 6 insertions(+)
scripts/genksyms/keywords.c +4 -0
index 9f40bcd17d07..f6956aa41366 100644
@@ -24,6 +24,10 @@ static struct resword {
 	{ "__volatile__", VOLATILE_KEYW },
 	{ "__builtin_va_list", VA_LIST_KEYW },
 
+	{ "__int128", BUILTIN_INT_KEYW },
+	{ "__int128_t", BUILTIN_INT_KEYW },
+	{ "__uint128_t", BUILTIN_INT_KEYW },
+
 	// According to rth, c99 defines "_Bool", __restrict", __restrict__", "restrict".  KAO
 	{ "_Bool", BOOL_KEYW },
 	{ "_restrict", RESTRICT_KEYW },
scripts/genksyms/parse.y +2 -0
index 00a6d7e54971..1ebcf52cd0f9 100644
@@ -76,6 +76,7 @@ static void record_compound(struct string_list **keyw,
 %token ATTRIBUTE_KEYW
 %token AUTO_KEYW
 %token BOOL_KEYW
+%token BUILTIN_INT_KEYW
 %token CHAR_KEYW
 %token CONST_KEYW
 %token DOUBLE_KEYW
@@ -263,6 +264,7 @@ simple_type_specifier:
 	| VOID_KEYW
 	| BOOL_KEYW
 	| VA_LIST_KEYW
+	| BUILTIN_INT_KEYW
 	| TYPE			{ (*$1)->tag = SYM_TYPEDEF; $$ = $1; }
 	;
 
Signature · 1 line
2.11.0
#1 lore raw

scripts/genksyms/keywords.c static struct resword { · line 29

Quoted 4 lines · Will Deacon
scripts/genksyms/keywords.c +3 -0
+       { "__int128", BUILTIN_INT_KEYW },
+       { "__int128_t", BUILTIN_INT_KEYW },
+       { "__uint128_t", BUILTIN_INT_KEYW },

I wonder if it's safe to treat them as the same type, since __int128_t and __uint128_t differ in signedness.

If someone exports a symbol with one and changes it to the other, they would appear to be the same type.

Arnd

#2 lore raw

Hi Arnd,

Quoted 11 lines · Arnd Bergmann

scripts/genksyms/keywords.c static struct resword { · line 29

Quoted 4 lines · Will Deacon
scripts/genksyms/keywords.c +3 -0
+       { "__int128", BUILTIN_INT_KEYW },
+       { "__int128_t", BUILTIN_INT_KEYW },
+       { "__uint128_t", BUILTIN_INT_KEYW },

I wonder if it's safe to treat them as the same type, since __int128_t and __uint128_t differ in signedness.

If someone exports a symbol with one and changes it to the other, they would appear to be the same type.

My understanding is that the actual CRC generation for normal symbols is based purely on the string-representation of the function signature, and so the underlying type information isn't relevant to that. I can also confirm that the CRC for an exported symbol that returns a __uint128_t is not the same if you change it to return a __int128_t instead.

Will

#3 lore raw
Quoted 13 lines · Will Deacon

Hi Arnd,

Quoted 9 lines · Arnd Bergmann

scripts/genksyms/keywords.c static struct resword { · line 29

Quoted 4 lines · Will Deacon
scripts/genksyms/keywords.c +3 -0
+       { "__int128", BUILTIN_INT_KEYW },
+       { "__int128_t", BUILTIN_INT_KEYW },
+       { "__uint128_t", BUILTIN_INT_KEYW },

I wonder if it's safe to treat them as the same type, since __int128_t and __uint128_t differ in signedness.

Quoted 8 lines
Quoted 2 lines

If someone exports a symbol with one and changes it to the other, they would appear to be the same type.

My understanding is that the actual CRC generation for normal symbols is based purely on the string-representation of the function signature, and so the underlying type information isn't relevant to that. I can also confirm that the CRC for an exported symbol that returns a __uint128_t is not the same if you change it to return a __int128_t instead.

Right.
Applied to linux-kbuild. Thanks!

Signature · 2 lines
Best Regards
Masahiro Yamada