Skip to contentSkip to conversation

__crc_... is intended to be absolute

PATCH In review Al Viro 2 participants 2 messages Feb 1, 2007 – Feb 2, 2007

lore
Share & tools
lore

Al Viro

author
#0 lore raw

PATCH 1/1 __crc_... is intended to be absolute

i386 boot/compressed/relocs checks for absolute symbols and warns about unexpected ones. If you build with modversions, you get ~2500 warnings about __crc_<symbol>. These suckers are really absolute symbols - we do _not_ want to modify them on relocation.

They are generated by genksyms - EXPORT_... generates a weak alias, then genksyms produces an ld script with __crc_<symbol> = <checksum> and it's fed to ld to produce the final object file. Their only use is to match kernel and module at modprobe time; they _must_ be absolute.

boot/compressed/relocs has a whitelist of known absolute symbols, but it doesn't know about __crc_... stuff. As the result, we get shitloads of false positives on any ld(1) version.

---
 arch/i386/boot/compressed/relocs.c |    2 ++
 1 files changed, 2 insertions(+), 0 deletions(-)
arch/i386/boot/compressed/relocs.c +2 -0
index 468da89..881951c 100644
@@ -43,6 +43,8 @@ static int is_safe_abs_reloc(const char* sym_name)
 			/* Match found */
 			return 1;
 	}
+	if (strncmp(sym_name, "__crc_", 6) == 0)
+		return 1;
 	return 0;
 }
 
Signature · 1 line
1.5.0-rc2.GIT
#1 lore raw
Quoted 25 lines · Al Viro

i386 boot/compressed/relocs checks for absolute symbols and warns about unexpected ones. If you build with modversions, you get ~2500 warnings about __crc_<symbol>. These suckers are really absolute symbols - we do _not_ want to modify them on relocation.

They are generated by genksyms - EXPORT_... generates a weak alias, then genksyms produces an ld script with __crc_<symbol> = <checksum> and it's fed to ld to produce the final object file. Their only use is to match kernel and module at modprobe time; they _must_ be absolute.

boot/compressed/relocs has a whitelist of known absolute symbols, but it doesn't know about __crc_... stuff. As the result, we get shitloads of false positives on any ld(1) version.

---
 arch/i386/boot/compressed/relocs.c |    2 ++
 1 files changed, 2 insertions(+), 0 deletions(-)
arch/i386/boot/compressed/relocs.c +0 -0
index 468da89..881951c 100644

arch/i386/boot/compressed/relocs.c static int is_safe_abs_reloc(const char* sym_name) · line 49

Quoted 7 lines
arch/i386/boot/compressed/relocs.c +2 -0
@@ -43,6 +43,8 @@ static int is_safe_abs_reloc(const char* sym_name)
 			/* Match found */
 			return 1;
 	}
+	if (strncmp(sym_name, "__crc_", 6) == 0)
+		return 1;
 	return 0;
 }

Thanks Viro. Sorry, I missed this case.

Thanks
Vivek