rtld: do not rely on a populated GOT on amd64

On rela architectures GNU BFD ld and gold store the relocation addend
in GOT entries (in addition to the relocation's r_addend field).
rtld previously relied on this to access its own _DYNAMIC symbol in
order to apply its own relocations.

However, recording addends in the GOT is not specified by the ABI,
and some versions of LLVM's LLD linker leave the GOT uninitialized on
rela architectures.

BFD ld does not populate the GOT on sparc64, and sparc64 rtld has a
machine-dependent rtld_dynamic_addr() function that returns the
_DYNAMIC address. Use the same approach on amd64, obtaining the %rip-
relative _DYNAMIC address following a suggestion from Rafael Espíndola.

Architectures other than amd64 should be addressed in future work.

PR:		214972
Reviewed by:	kib
MFC after:	2 weeks
Sponsored by:	The FreeBSD Foundation
Differential Revision:	https://reviews.freebsd.org/D9180

libexec/rtld-elf/amd64/rtld_machdep.h

@@ -35,8 +35,8 @@
 struct Struct_Obj_Entry;
 
 /* Return the address of the .dynamic section in the dynamic linker. */
-#define rtld_dynamic(obj) \
-    ((const Elf_Dyn *)((obj)->relocbase + (Elf_Addr)&_DYNAMIC))
+Elf_Dyn *rtld_dynamic_addr(void);
+#define	rtld_dynamic(obj)	rtld_dynamic_addr()
 
 /* Fixup the jump slot at "where" to transfer control to "target". */
 static inline Elf_Addr

libexec/rtld-elf/amd64/rtld_start.S

@@ -156,4 +156,16 @@ _rtld_bind_start:
 	.cfi_endproc
 	.size	_rtld_bind_start, . - _rtld_bind_start
 
+	.align	4
+	.globl	rtld_dynamic_addr
+	.type	rtld_dynamic_addr,@function
+rtld_dynamic_addr:
+	.cfi_startproc
+	.weak _DYNAMIC
+	.hidden _DYNAMIC
+	lea	_DYNAMIC(%rip),%rax
+	ret
+	.cfi_endproc
+	.size	rtld_dynamic_addr, . - rtld_dynamic_addr
+
 	.section .note.GNU-stack,"",%progbits