--- a/src/cpu/aarch64/vm/aarch64.ad	Wed Dec 09 13:08:38 2015 +0000
+++ b/src/cpu/aarch64/vm/aarch64.ad	Tue Sep 15 16:14:32 2015 +0000
@@ -2845,18 +2845,18 @@
 
 int MachCallRuntimeNode::ret_addr_offset() {
   // for generated stubs the call will be
-  //   far_call(addr)
-  // for real runtime callouts it iwll be
-  //   mov(rscratch1, RuntimeAddress(addr)
+  //   bl(addr)
+  // for real runtime callouts it will be six instructions
+  // see aarch64_enc_java_to_runtime
+  //   adr(rscratch2, retaddr)
+  //   lea(rscratch1, RuntimeAddress(addr)
+  //   stp(zr, rscratch2, Address(__ pre(sp, -2 * wordSize)))
   //   blrt rscratch1
   CodeBlob *cb = CodeCache::find_blob(_entry_point);
   if (cb) {
     return MacroAssembler::far_branch_size();
   } else {
-    // A 48-bit address.  See movptr().
-    // then a blrt
-    // return 16;
-    return 4 * NativeInstruction::instruction_size;
+    return 6 * NativeInstruction::instruction_size;
   }
 }
 
@@ -4765,8 +4765,14 @@
       int fpcnt;
       int rtype;
       getCallInfo(tf(), gpcnt, fpcnt, rtype);
+      Label retaddr;
+      __ adr(rscratch2, retaddr);
       __ lea(rscratch1, RuntimeAddress(entry));
+      // Leave a breadcrumb for JavaThread::pd_last_frame().
+      __ stp(zr, rscratch2, Address(__ pre(sp, -2 * wordSize)));
       __ blrt(rscratch1, gpcnt, fpcnt, rtype);
+      __ bind(retaddr);
+      __ add(sp, sp, 2 * wordSize);
     }
   %}
 
@@ -7813,51 +7819,6 @@
   ins_pipe(istore_mem);
 %}
 
-// Save last Java PC to thread anchor
-
-// the ideal code cannot directly encode a reference to the current
-// PC, never mind the PC of the return address which follows the
-// runtime call whcih wil be generated at a later point. so this
-// operation is encoded in the ideal code as a write of either -1 or
-// -2 via the thread register with an offset which locates the last
-// Java pc slot in the thread anchor. the encoding computes the
-// correct Java pc to write by offsetting form the current pc to allwo
-// for the intervening instructions which will precede the runtime
-// call. Note that the constant is supplied as -2 when the stub caller
-// return address is passed to the C function and -1 when no return
-// address is required. we set the primary opcode to 1 or 0
-// accordingly as this affects the offset calculation.
-
-instruct storeLastJavaPC_no_retaddr(thread_anchor_pc mem, immP_M1 dummy_m1)
-%{
-  match(Set mem (StoreP mem dummy_m1));
-
-  ins_cost(INSN_COST);
-  format %{ "str  ., $mem\t# save pc to thread (no ret addr)" %}
-
-  // use opcode to indicate that we have no return address argument
-  opcode(0x0);
-
-  ins_encode(aarch64_enc_save_pc());
-
-  ins_pipe(istore_mem);
-%}
-
-instruct storeLastJavaPC_with_retaddr(thread_anchor_pc mem, immP_M2 dummy_m2)
-%{
-  match(Set mem (StoreP mem dummy_m2));
-
-  ins_cost(INSN_COST);
-  format %{ "str  ., $mem\t# save pc to thread (w ret addr)" %}
-
-  // use opcode to indicate that we have a return address argument
-  opcode(0x1);
-
-  ins_encode(aarch64_enc_save_pc());
-
-  ins_pipe(istore_mem);
-%}
-
 // Store Compressed Pointer
 instruct storeN(iRegN src, memory mem)
 %{

--- a/src/share/vm/opto/generateOptoStub.cpp	Wed Dec 09 13:08:38 2015 +0000
+++ b/src/share/vm/opto/generateOptoStub.cpp	Tue Sep 15 16:14:32 2015 +0000
@@ -96,54 +96,6 @@
                                    in_bytes(JavaFrameAnchor::flags_offset()));
 #endif /* defined(SPARC) */
 
-#if defined(AARCH64)
-  // the x86 code does not need to write return pc which follows the
-  // VM callout to the thread anchor pc slot. the frame walking code
-  // can safely compute that a called the stub's return pc by
-  // evaluating sp[-1] where sp is the top of the stub frame saved in
-  // the anchor sp field. AArch64 cannot make this assumption because
-  // the C compiler does not have to push the link register and, even
-  // if it does, can locate it somewhere other than at the bottom of
-  // the frame.
-
-  // this poses a problem when we come to generate the opto runtime
-  // stubs. we cannot actually refer to generated code addresses in
-  // ideal code and, worse, we need to get the memory write to write a
-  // constant value derived form the code address of the instruction
-  // following the runtime callout i.e. the constant we need is only
-  // determined when we plant a separate instruction later in the
-  // instruction stream.
-  // 
-  // rather than introduce a mechanism into the opto compiler to allow
-  // us to provide some sort of label and label resolve mechanism and
-  // then backpatch the required address we can actually do this with
-  // a minor hack. we generate a store to the anchor pc field but
-  // supply a small negative constant address (-1 or -2) as the value
-  // for the memory store. in the back end we detect this type of
-  // store and compute the desired address, substituting it in place
-  // of the supplied constant.
-  // 
-  // the back end employs a special lowering rule to do this. it
-  // matches a write via the thread register with an offset which
-  // equals the thread anchor's pc slot. of course, those sort of
-  // writes only happen in these stubs! the encodign for this rule
-  // substitutes the supplied value with the current code buffer
-  // address plus an offset to the instruction following the VM
-  // callout. Of course, the offset varies according to what arguments
-  // are passed to the callout. For any given stub the arguments
-  // passed to the VM include the stub arguments, the thread and,
-  // optionally, the stub's caller's return pc. We supply -1 in the
-  // store if the call will not include the caller ret pc or -2 if it
-  // does. The back end can use this to work out exactly what the
-  // required offset is.
-
-  const TypeRawPtr *t = TypeRawPtr::make((address)(return_pc ? -2L : -1L));
-
-  Node *last_pc = new (C) ConPNode(t);
-  _gvn.set_type(last_pc, t);
-  store_to_memory(NULL, adr_last_Java_pc, last_pc, T_ADDRESS, NoAlias, MemNode::unordered);
-
-#endif /* defined(AARCH64) */
 
   // Drop in the last_Java_sp.  last_Java_fp is not touched.
   // Always do this after the other "last_Java_frame" fields are set since