Use __ehdr_start to compute load address

So far, uclibc relied on the informal behaviour of GNU ld to put the
link address of the _DYNAMIC symbol in the first GOT entry. This does
not work with LLVM lld which does not follow this convention.
Consequently, glibc has abandoned its usage in favour of using
__ehdr_start as magic symbol to infer the load address [1].

Note that this implies that the link time address of __ehdr_start is
*always* 0! So far, this seems to be the case on all platforms.

[1] https://sourceware.org/bugzilla/show_bug.cgi?id=28203

Signed-off-by: Marcus Haehnel <marcus.haehnel@kernkonzept.com>

ldso/ldso/aarch64/dl-sysdep.h

@@ -54,28 +54,21 @@ unsigned long _dl_linux_resolver(struct elf_resolve * tpnt, int reloc_entry);
      || (type) == R_AARCH64_TLSDESC) * ELF_RTYPE_CLASS_PLT)		\
    | (((type) == R_AARCH64_COPY) * ELF_RTYPE_CLASS_COPY))
 
-/* Return the link-time address of _DYNAMIC.  Conveniently, this is the
-   first element of the GOT. */
-extern const ElfW(Addr) _GLOBAL_OFFSET_TABLE_[] attribute_hidden;
-static __always_inline ElfW(Addr) __attribute__ ((unused))
-elf_machine_dynamic (void)
-{
-  return _GLOBAL_OFFSET_TABLE_[0];
-}
-
 /* Return the run-time load address of the shared object.  */
 
 static __always_inline ElfW(Addr) __attribute__ ((unused))
 elf_machine_load_address (void)
 {
-  /* To figure out the load address we use the definition that for any symbol:
-     dynamic_addr(symbol) = static_addr(symbol) + load_addr
-
-    _DYNAMIC sysmbol is used here as its link-time address stored in
-    the special unrelocated first GOT entry.  */
+  extern const ElfW(Ehdr) __ehdr_start attribute_hidden;
+  return (ElfW(Addr)) &__ehdr_start;
+}
 
-    extern ElfW(Dyn) _DYNAMIC[] attribute_hidden;
-    return (ElfW(Addr)) &_DYNAMIC - elf_machine_dynamic ();
+/* Return the link-time address of _DYNAMIC. */
+static __always_inline ElfW(Addr) __attribute__ ((unused))
+elf_machine_dynamic (void)
+{
+  extern ElfW(Dyn) _DYNAMIC[] attribute_hidden;
+  return (ElfW(Addr)) _DYNAMIC - elf_machine_load_address ();
 }
 
 static __always_inline void

ldso/ldso/arm/dl-sysdep.h

@@ -96,43 +96,6 @@ unsigned long _dl_linux_resolver(struct elf_resolve * tpnt, int reloc_entry);
    | (((type) == R_ARM_COPY) * ELF_RTYPE_CLASS_COPY))
 #endif /* __FDPIC__ */
 
-/* Return the link-time address of _DYNAMIC.  Conveniently, this is the
-   first element of the GOT.  We used to use the PIC register to do this
-   without a constant pool reference, but GCC 4.2 will use a pseudo-register
-   for the PIC base, so it may not be in r10.  */
-static __always_inline Elf32_Addr __attribute__ ((unused))
-elf_machine_dynamic (void)
-{
-  Elf32_Addr dynamic;
-#if !defined __thumb__
-  __asm__ ("ldr %0, 2f\n"
-       "1: ldr %0, [pc, %0]\n"
-       "b 3f\n"
-       "2: .word _GLOBAL_OFFSET_TABLE_ - (1b+8)\n"
-       "3:" : "=r" (dynamic));
-#else
-  int tmp;
-  __asm__ (".align 2\n"
-       "bx     pc\n"
-       "nop\n"
-       ".arm\n"
-       "ldr %0, 2f\n"
-       "1: ldr %0, [pc, %0]\n"
-       "b 3f\n"
-       "2: .word _GLOBAL_OFFSET_TABLE_ - (1b+8)\n"
-       "3:"
-       ".align  2\n"
-        "orr     %1, pc, #1\n"
-        "bx      %1\n"
-        ".force_thumb\n"
-       : "=r" (dynamic), "=&r" (tmp));
-#endif
-
-  return dynamic;
-}
-
-extern char __dl_start[] __asm__("_dl_start");
-
 #ifdef __FDPIC__
 /* We must force strings used early in the bootstrap into the data
    segment.  */
@@ -148,28 +111,16 @@ extern char __dl_start[] __asm__("_dl_start");
 static __always_inline Elf32_Addr __attribute__ ((unused))
 elf_machine_load_address (void)
 {
-#if defined(__FDPIC__)
-	return 0;
-#else
-	Elf32_Addr got_addr = (Elf32_Addr) &__dl_start;
-	Elf32_Addr pcrel_addr;
-#if defined __OPTIMIZE__ && !defined __thumb__
-	__asm__ ("adr %0, _dl_start" : "=r" (pcrel_addr));
-#else
-	/* A simple adr does not work in Thumb mode because the offset is
-	   negative, and for debug builds may be too large.  */
-	int tmp;
-	__asm__ ("adr %1, 1f\n\t"
-		 "ldr %0, [%1]\n\t"
-		 "add %0, %0, %1\n\t"
-		 "b 2f\n\t"
-		 ".align 2\n\t"
-		 "1: .word _dl_start - 1b\n\t"
-		 "2:"
-		 : "=r" (pcrel_addr), "=r" (tmp));
-#endif
-	return pcrel_addr - got_addr;
-#endif
+  extern const Elf32_Ehdr __ehdr_start attribute_hidden;
+  return (Elf32_Addr) &__ehdr_start;
+}
+
+/* Return the link-time address of _DYNAMIC. */
+static __always_inline Elf32_Addr __attribute__ ((unused))
+elf_machine_dynamic (void)
+{
+  extern Elf32_Dyn _DYNAMIC[] attribute_hidden;
+  return (Elf32_Addr) _DYNAMIC - elf_machine_load_address ();
 }
 
 static __always_inline void

ldso/ldso/i386/dl-sysdep.h

@@ -35,28 +35,21 @@ extern unsigned long _dl_linux_resolver(struct elf_resolve * tpnt, int reloc_ent
      || (type) == R_386_TLS_TPOFF) * ELF_RTYPE_CLASS_PLT)				  \
    | (((type) == R_386_COPY) * ELF_RTYPE_CLASS_COPY))
 
-/* Return the link-time address of _DYNAMIC.  Conveniently, this is the
-   first element of the GOT, a special entry that is never relocated.  */
-extern const Elf32_Addr _GLOBAL_OFFSET_TABLE_[] attribute_hidden;
-static __always_inline Elf32_Addr __attribute__ ((unused, const))
-elf_machine_dynamic (void)
-{
-	/* This produces a GOTOFF reloc that resolves to zero at link time, so in
-	   fact just loads from the GOT register directly.  By doing it without
-	   an asm we can let the compiler choose any register.  */
-	return _GLOBAL_OFFSET_TABLE_[0];
-}
-
 
-extern Elf32_Dyn bygotoff[] __asm__ ("_DYNAMIC") attribute_hidden;
 /* Return the run-time load address of the shared object.  */
 static __always_inline Elf32_Addr attribute_unused
 elf_machine_load_address (void)
 {
-	/* Compute the difference between the runtime address of _DYNAMIC as seen
-	   by a GOTOFF reference, and the link-time address found in the special
-	   unrelocated first GOT entry.  */
-	return (Elf32_Addr) &bygotoff - elf_machine_dynamic ();
+	extern const Elf32_Ehdr __ehdr_start attribute_hidden;
+	return (Elf32_Addr) &__ehdr_start;
+}
+
+/* Return the link-time address of _DYNAMIC. */
+static __always_inline Elf32_Addr __attribute__ ((unused, const))
+elf_machine_dynamic (void)
+{
+	extern Elf32_Dyn _DYNAMIC[] attribute_hidden;
+	return (Elf32_Addr) _DYNAMIC - elf_machine_load_address ();
 }
 
 static __always_inline void

ldso/ldso/riscv32/dl-sysdep.h

@@ -59,22 +59,20 @@ unsigned long _dl_linux_resolver(struct elf_resolve * tpnt, int reloc_entry);
    | (ELF_RTYPE_CLASS_COPY * ((type) == R_RISCV_COPY)))
 
 
-/* Return the link-time address of _DYNAMIC.  */
-static inline ElfW(Addr)
-elf_machine_dynamic (void)
-{
-  extern ElfW(Addr) _GLOBAL_OFFSET_TABLE_ __attribute__ ((visibility ("hidden")));
-  return _GLOBAL_OFFSET_TABLE_;
-}
-
-
 /* Return the run-time load address of the shared object.  */
 static __always_inline ElfW(Addr) __attribute__ ((unused))
 elf_machine_load_address (void)
 {
-  ElfW(Addr) load_addr;
-  __asm__ ("lla %0, _DYNAMIC" : "=r" (load_addr));
-  return load_addr - elf_machine_dynamic ();
+  extern const ElfW(Ehdr) __ehdr_start attribute_hidden;
+  return (ElfW(Addr)) &__ehdr_start;
+}
+
+/* Return the link-time address of _DYNAMIC.  */
+static inline ElfW(Addr)
+elf_machine_dynamic (void)
+{
+  extern ElfW(Dyn) _DYNAMIC[] attribute_hidden;
+  return (ElfW(Addr)) _DYNAMIC - elf_machine_load_address ();
 }
 
 static __always_inline void

ldso/ldso/x86_64/dl-sysdep.h

@@ -52,48 +52,21 @@ extern unsigned long _dl_linux_resolver(struct elf_resolve * tpnt, int reloc_ent
     * ELF_RTYPE_CLASS_PLT)						      \
    | (((type) == R_X86_64_COPY) * ELF_RTYPE_CLASS_COPY))
 
-/* Return the link-time address of _DYNAMIC.  Conveniently, this is the
-   first element of the GOT.  This must be inlined in a function which
-   uses global data.  */
-static __always_inline Elf64_Addr __attribute__ ((unused))
-elf_machine_dynamic (void)
-{
-  Elf64_Addr addr;
-
-  /* This works because we have our GOT address available in the small PIC
-     model.  */
-  addr = (Elf64_Addr) &_DYNAMIC;
-
-  return addr;
-}
-
 
 /* Return the run-time load address of the shared object.  */
 static __always_inline Elf64_Addr __attribute__ ((unused))
 elf_machine_load_address (void)
 {
-  register Elf64_Addr addr, tmp;
-
-  /* The easy way is just the same as on x86:
-       leaq _dl_start, %0
-       leaq _dl_start(%%rip), %1
-       subq %0, %1
-     but this does not work with binutils since we then have
-     a R_X86_64_32S relocation in a shared lib.
-
-     Instead we store the address of _dl_start in the data section
-     and compare it with the current value that we can get via
-     an RIP relative addressing mode.  */
-
-  __asm__ ("movq 1f(%%rip), %1\n"
-       "0:\tleaq _dl_start(%%rip), %0\n\t"
-       "subq %1, %0\n\t"
-       ".section\t.data\n"
-       "1:\t.quad _dl_start\n\t"
-       ".previous\n\t"
-       : "=r" (addr), "=r" (tmp) : : "cc");
+  extern const Elf64_Ehdr __ehdr_start attribute_hidden;
+  return (Elf64_Addr) &__ehdr_start;
+}
 
-  return addr;
+/* Return the link-time address of _DYNAMIC. */
+static __always_inline Elf64_Addr __attribute__ ((unused))
+elf_machine_dynamic (void)
+{
+  extern Elf64_Dyn _DYNAMIC[] attribute_hidden;
+  return (Elf64_Addr) _DYNAMIC - elf_machine_load_address ();
 }
 
 static __always_inline void