[PATCH] aarch64: Make elf_machine_{load_address, dynamic} robust [BZ #28203]

[Fangrui Song]

The AArch64 ABI is largely platform agnostic and does not specify
_GLOBAL_OFFSET_TABLE_[0] ([1]). glibc ld.so turns out to be probably the
only user of _GLOBAL_OFFSET_TABLE_[0] and GNU ld defines the value
to the link-time address _DYNAMIC. [2]

In 2012, __ehdr_start was implemented in GNU ld and gold.
Using adrp+addr to access __ehdr_start/_DYNAMIC gives us a robust way to
get the load address and the link-time address of _DYNAMIC.

With https://sourceware.org/pipermail/libc-alpha/2021-August/129864.html,
this patch, and disabling traditional TLSGD tests (neither Clang nor
LLD's aarch64 port supports), LLD linked glibc has the same number of
`make check` failures.

With this, I think
https://github.com/ARM-software/abi-aa/pull/57
doesn't need to specify _GLOBAL_OFFSET_TABLE_[0].
musl, FreeBSD, and NetBSD don't use _GLOBAL_OFFSET_TABLE_[0].

A bonus with the new asm approach: we drop reliance on compiler generating
PC-relative relocations to get the runtime address.

[1]: From a psABI maintainer, https://bugs.llvm.org/show_bug.cgi?id=49672#c2
[2]: LLD's aarch64 port does not set _GLOBAL_OFFSET_TABLE_[0] to the
link-time address _DYNAMIC.
LLD is widely used on aarch64 Android and ChromeOS devices.  Software
just works without the need for _GLOBAL_OFFSET_TABLE_[0].
---
 sysdeps/aarch64/dl-machine.h | 28 +++++++++++++---------------
 1 file changed, 13 insertions(+), 15 deletions(-)

diff --git a/sysdeps/aarch64/dl-machine.h b/sysdeps/aarch64/dl-machine.h
index d29d827ab3..bf251d2972 100644
--- a/sysdeps/aarch64/dl-machine.h
+++ b/sysdeps/aarch64/dl-machine.h
@@ -37,28 +37,26 @@ elf_machine_matches_host (const ElfW(Ehdr) *ehdr)
   return ehdr->e_machine == EM_AARCH64;
 }
 
-/* Return the link-time address of _DYNAMIC.  Conveniently, this is the
-   first element of the GOT. */
-static inline ElfW(Addr) __attribute__ ((unused))
-elf_machine_dynamic (void)
-{
-  extern const ElfW(Addr) _GLOBAL_OFFSET_TABLE_[] attribute_hidden;
-  return _GLOBAL_OFFSET_TABLE_[0];
-}
-
 /* Return the run-time load address of the shared object.  */
 
 static 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
+  ElfW(Addr) addr;
+  asm ("adrp %0, __ehdr_start\n\t"
+       "add %0, %0, :lo12:__ehdr_start" : "=r"(addr));
+  return addr;
+}
 
-    _DYNAMIC sysmbol is used here as its link-time address stored in
-    the special unrelocated first GOT entry.  */
+/* Return the link-time address of _DYNAMIC.  */
 
-    extern ElfW(Dyn) _DYNAMIC[] attribute_hidden;
-    return (ElfW(Addr)) &_DYNAMIC - elf_machine_dynamic ();
+static inline ElfW(Addr) __attribute__ ((unused))
+elf_machine_dynamic (void)
+{
+  ElfW(Addr) addr;
+  asm ("adrp %0, _DYNAMIC\n\t"
+       "add %0, %0, :lo12:_DYNAMIC" : "=r"(addr));
+  return addr - elf_machine_load_address ();
 }
 
 /* Set up the loaded object described by L so its unrelocated PLT
-- 
2.32.0.605.g8dce9f2422-goog