[BZ 18034][AArch64] Lazy TLSDESC relocation data race fix

Lazy TLSDESC initialization needs to be synchronized with concurrent TLS
accesses.  The TLS descriptor contains a function pointer (entry) and an
argument that is accessed from the entry function.  With lazy initialization
the first call to the entry function updates the entry and the argument to
their final value.  A final entry function must make sure that it accesses an
initialized argument, this needs synchronization on systems with weak memory
ordering otherwise the writes of the first call can be observed out of order.

There are at least two issues with the current code:

tlsdesc.c (i386, x86_64, arm, aarch64) uses volatile memory accesses on the
write side (in the initial entry function) instead of C11 atomics.

And on systems with weak memory ordering (arm, aarch64) the read side
synchronization is missing from the final entry functions (dl-tlsdesc.S).

This patch only deals with aarch64.

* Write side:

Volatile accesses were replaced with C11 relaxed atomics, and a release
store was used for the initialization of entry so the read side can
synchronize with it.

* Read side:

TLS access generated by the compiler and an entry function code is roughly

  ldr x1, [x0]    // load the entry
  blr x1          // call it

entryfunc:
  ldr x0, [x0,#8] // load the arg
  ret

Various alternatives were considered to force the ordering in the entry
function between the two loads:

(1) barrier

entryfunc:
  dmb ishld
  ldr x0, [x0,#8]

(2) address dependency (if the address of the second load depends on the
result of the first one the ordering is guaranteed):

entryfunc:
  ldr x1,[x0]
  and x1,x1,#8
  orr x1,x1,#8
  ldr x0,[x0,x1]

(3) load-acquire (ARMv8 instruction that is ordered before subsequent
loads and stores)

entryfunc:
  ldar xzr,[x0]
  ldr x0,[x0,#8]

Option (1) is the simplest but slowest (note: this runs at every TLS
access), options (2) and (3) do one extra load from [x0] (same address
loads are ordered so it happens-after the load on the call site),
option (2) clobbers x1 which is problematic because existing gcc does
not expect that, so approach (3) was chosen.

A new _dl_tlsdesc_return_lazy entry function was introduced for lazily
relocated static TLS, so non-lazy static TLS can avoid the synchronization
cost.

	[BZ #18034]
	* sysdeps/aarch64/dl-tlsdesc.h (_dl_tlsdesc_return_lazy): Declare.
	* sysdeps/aarch64/dl-tlsdesc.S (_dl_tlsdesc_return_lazy): Define.
	(_dl_tlsdesc_undefweak): Guarantee TLSDESC entry and argument load-load
	ordering using ldar.
	(_dl_tlsdesc_dynamic): Likewise.
	(_dl_tlsdesc_return_lazy): Likewise.
	* sysdeps/aarch64/tlsdesc.c (_dl_tlsdesc_resolve_rela_fixup): Use
	relaxed atomics instead of volatile and synchronize with release store.
	(_dl_tlsdesc_resolve_hold_fixup): Use relaxed atomics instead of
	volatile.
	* elf/tlsdeschtab.h (_dl_tlsdesc_resolve_early_return_p): Likewise.

1 files changed, 24 insertions, 12 deletions

diff --git a/sysdeps/aarch64/tlsdesc.c b/sysdeps/aarch64/tlsdesc.c
index 4821f8c08a..9f3ff9b662 100644
--- a/sysdeps/aarch64/tlsdesc.c
+++ b/sysdeps/aarch64/tlsdesc.c
@@ -25,6 +25,7 @@
 #include <dl-tlsdesc.h>
 #include <dl-unmap-segments.h>
 #include <tlsdeschtab.h>
+#include <atomic.h>
 
 /* The following functions take an entry_check_offset argument.  It's
    computed by the caller as an offset between its entry point and the
@@ -39,11 +40,15 @@
 
 void
 attribute_hidden
-_dl_tlsdesc_resolve_rela_fixup (struct tlsdesc volatile *td,
-				struct link_map *l)
+_dl_tlsdesc_resolve_rela_fixup (struct tlsdesc *td, struct link_map *l)
 {
-  const ElfW(Rela) *reloc = td->arg;
+  const ElfW(Rela) *reloc = atomic_load_relaxed (&td->arg);
 
+  /* After GL(dl_load_lock) is grabbed only one caller can see td->entry in
+     initial state in _dl_tlsdesc_resolve_early_return_p, other concurrent
+     callers will return and retry calling td->entry.  The updated td->entry
+     synchronizes with the single writer so all read accesses here can use
+     relaxed order.  */
   if (_dl_tlsdesc_resolve_early_return_p
       (td, (void*)(D_PTR (l, l_info[ADDRIDX (DT_TLSDESC_PLT)]) + l->l_addr)))
     return;
@@ -86,8 +91,10 @@ _dl_tlsdesc_resolve_rela_fixup (struct tlsdesc volatile *td,
 
   if (!sym)
     {
-      td->arg = (void*) reloc->r_addend;
-      td->entry = _dl_tlsdesc_undefweak;
+      atomic_store_relaxed (&td->arg, (void *) reloc->r_addend);
+      /* This release store synchronizes with the ldar acquire load
+	 instruction in _dl_tlsdesc_undefweak.  */
+      atomic_store_release (&td->entry, _dl_tlsdesc_undefweak);
     }
   else
     {
@@ -96,16 +103,22 @@ _dl_tlsdesc_resolve_rela_fixup (struct tlsdesc volatile *td,
 #  else
       if (!TRY_STATIC_TLS (l, result))
 	{
-	  td->arg = _dl_make_tlsdesc_dynamic (result, sym->st_value
+	  void *p = _dl_make_tlsdesc_dynamic (result, sym->st_value
 					      + reloc->r_addend);
-	  td->entry = _dl_tlsdesc_dynamic;
+	  atomic_store_relaxed (&td->arg, p);
+	  /* This release store synchronizes with the ldar acquire load
+	     instruction in _dl_tlsdesc_dynamic.  */
+	  atomic_store_release (&td->entry, _dl_tlsdesc_dynamic);
 	}
       else
 #  endif
 	{
-	  td->arg = (void*) (sym->st_value + result->l_tls_offset
+	  void *p = (void*) (sym->st_value + result->l_tls_offset
 			     + reloc->r_addend);
-	  td->entry = _dl_tlsdesc_return;
+	  atomic_store_relaxed (&td->arg, p);
+	  /* This release store synchronizes with the ldar acquire load
+	     instruction in _dl_tlsdesc_return_lazy.  */
+	  atomic_store_release (&td->entry, _dl_tlsdesc_return_lazy);
 	}
     }
 
@@ -120,11 +133,10 @@ _dl_tlsdesc_resolve_rela_fixup (struct tlsdesc volatile *td,
 
 void
 attribute_hidden
-_dl_tlsdesc_resolve_hold_fixup (struct tlsdesc volatile *td,
-				void *caller)
+_dl_tlsdesc_resolve_hold_fixup (struct tlsdesc *td, void *caller)
 {
   /* Maybe we're lucky and can return early.  */
-  if (caller != td->entry)
+  if (caller != atomic_load_relaxed (&td->entry))
     return;
 
   /* Locking here will stop execution until the running resolver runs