NetBSD

History

mycroft cf4aca11b5 In the `I can't believe it's been broken all this time' department: _rtld_bind_start must save and restore the condition codes. Varargs functions (like, say, printf()) depend on the state of cr1 to determine whether they need to store floating point registers in the save area. Without this, the first call to any particular varargs function will fail if floating point values were passed.		2001-09-26 04:06:43 +00:00
..
arch	In the `I can't believe it's been broken all this time' department:	2001-09-26 04:06:43 +00:00
Makefile	Enable arm if OBJECT_FMT is ELF. Initialize first few GOT entries for ARM.	2001-07-15 03:09:22 +00:00
README	reflect reality	1999-08-20 21:12:47 +00:00
debug.c	…
debug.h	…
headers.c	Sync with <sys/exec_elf.h> rev. 1.45: ElfNN_RelA -> ElfNN_Rela.	2001-04-25 12:24:50 +00:00
load.c	s/warn/xwarnx/ cuts down size by 50% by not pulling in stdio.	2001-05-27 23:26:47 +00:00
malloc.c	MAP_COPY -> MAP_PRIVATE	1999-06-17 21:09:47 +00:00
map_object.c	Return an error when there are too many segments rather than abort()ing.	2000-11-26 05:22:15 +00:00
paths.c	Don't include <uvm/uvm_param.h>. It does not appear to be needed by anything.	2001-05-24 17:41:42 +00:00
reloc.c	Reorganize PowerPC cold a little. The return value from _rtld_bind_pltgot()	2001-09-10 06:09:41 +00:00
rtld.c	Unloading objects, can cause other objects whose reference count becomes	2001-05-10 20:19:25 +00:00
rtld.h	Sync with <sys/exec_elf.h> rev. 1.45: ElfNN_RelA -> ElfNN_Rela.	2001-04-25 12:24:50 +00:00
rtldenv.h	…
search.c	Adjust a comment after the Great Renaming.	2000-07-27 10:44:39 +00:00
symbol.c	format warning fixes from he, after testing on 64-bit elf (alpha)	2000-10-11 20:46:07 +00:00
sysident.h	Don't use multi-line string literals.	2001-06-19 12:07:21 +00:00
xmalloc.c	…
xprintf.c	Pointers are NULL. Characters are '\0'.	2001-08-14 20:13:56 +00:00

README

$NetBSD: README,v 1.6 1999/08/20 21:12:47 christos Exp $

BUGS/PROBLEMS:

* rtld.c around line 184:

	#iffdef __alpha__
	/* XXX XXX XXX */
	_rtld_objself.pltgot = NULL;
	#endif

  It's a workaround for a linker bug, and will go away when the linker is
  fixed.

* Formerly, _DYNAMIC was _GLOBAL_OFFSET_TABLE_[0], but that isn't true with
  newer snapshots of gas/binutils (at least on the Alpha), which support
  multiple GOTs.  That's what the OLD_GOT #ifdefs are about.
  
* The method used to relocate ld.so is shoddy and fragile.  Currently it's
  something like:

  alpha, powerpc:
	(1) assembly code relocates RELATIVE relocations in the GOT
	    (on some ports, all relative relocations; this will probably
	    have to be #ifdef'd in reloc.c), assuming that everything
	    in the GOT is a RELATIVE relocation.

	(2) C code relocates the rest of the loader, excluding those
	    RELATIVE relocations already done (via an ... iffy heuristic).

  i386, m68k, sparc:
	(1) C code relocates RELATIVE relocations in _rtld_init() carefully
	    so that it does not touch any globals.

	(2) C code relocates the rest of the loader, excluding those
	    RELATIVE relocations already done (via an ... iffy heuristic).

  mips:
	(1) loads ld.elf_so at 0, so no relocating is necessary.

  Something better (like that the glibc ELF ld.so does) should be done
  eventually, where a single set of C code is compiled twice, once for
  normal use, and once to be used to relocate the shared linker.  No
  relocations should be processed in assembly language.

* On the Alpha, this code Really needs to be checked to be sure that
  it doesn't make calls that could use FP, while doing lazy binding.

TO DO:

* Support for other platforms

* Support for coexistance of 32-bit and 64-bit ELF on platforms that can
  do that.

* Possible support for ldd (standalone operation) and program loading
  a la the glibc shared loader.

* Support for relocating ld.elf_so on mips