start___kallsyms missing from i386 vmlinux.lds ?

arch/i386/vmlinux.lds Mon Jan 1 07:55:50 2001

Quoted 5 lines · Paul Gortmaker

arch/i386/vmlinux.lds Mon Jan 1 07:55:50 2001 renamed from arch/i386/vmlinux.lds~ Fri Jul 7 03:47:07 2000 +3 -0

+  __start___kallsyms = .;     /* All kernel symbols */
+  __kallsyms : { *(__kallsyms) }
+  __stop___kallsyms = .;

kernel/module.c defines
extern const char __start___kallsyms[] __attribute__ ((weak)); extern const char __stop___kallsyms[] __attribute__ ((weak));

The symbols are weak and do not need to be defined. If gcc is not honouring __attribute__ ((weak)) then you have a broken or obsolete version of gcc. You need at least gcc 2.91.66 for kernel 2.4.

Quoted 4 lines · Keith Owens

kernel/module.c defines
extern const char __start___kallsyms[] __attribute__ ((weak)); extern const char __stop___kallsyms[] __attribute__ ((weak));

Quoted 3 lines

The symbols are weak and do not need to be defined. If gcc is not honouring __attribute__ ((weak)) then you have a broken or obsolete version of gcc. You need at least gcc 2.91.66 for kernel 2.4.

Yep, saw the weak part - just noted while scanning test11 diff that they were defined like the above patch for arch/sparc* and wondered if the inconsistency was intentional.

Some sparc users have a slightly older version of gcc, built shortly before 'weak' support was added, which required those symbols to be defined. Dave Miller thought the compiler problem was widespread enough to justify changing the source to suit the compiler instead of forcing sparc users to upgrade. I suspect that super-h has the same problem of old compilers, I noticed that somebody added the symbols to sh/vmlinux.lds.