bhushan verma wrote: > in \usr\src\cmd\sgs\rtld\common/elf.c > the function elf_rtld_load() there is a macro #if defined(i386) under which > elf_reloc_relacount() is called > can anybody please help me to find out,why this is for intel not for other > sparc and AMD64
It looks like an optimization that was applied to i386, could have been applied to AMD64 too. On both architectures, the procedure linkage table resides in shared text but uses addresses in the private global offset table. These .got entries need relative relocations applied to them before their use. In i386/_setup.c we skip .plt relocations, and only apply them if we determine we'll later jump through an ld.so.1 lazy binding, in which case, elf_reloc_relacount() performs the .plt relative relocations. In amd64/_setup.c we include .plt's in the initial relocation count. Inconsistent? I believe so. -- Rod
