8337987: Relocate jfr and throw_exception stubs from StubGenerator to SharedRuntime

Reviewed-by: fyang, kvn, yzheng

Author: adinn

src/hotspot/cpu/aarch64/macroAssembler_aarch64.cpp

@@ -744,7 +744,7 @@ void MacroAssembler::reserved_stack_check() {
     // We have already removed our own frame.
     // throw_delayed_StackOverflowError will think that it's been
     // called by our caller.
-    lea(rscratch1, RuntimeAddress(StubRoutines::throw_delayed_StackOverflowError_entry()));
+    lea(rscratch1, RuntimeAddress(SharedRuntime::throw_delayed_StackOverflowError_entry()));
     br(rscratch1);
     should_not_reach_here();
 

src/hotspot/cpu/aarch64/methodHandles_aarch64.cpp

@@ -120,7 +120,7 @@ void MethodHandles::jump_from_method_handle(MacroAssembler* _masm, Register meth
   __ ldr(rscratch1,Address(method, entry_offset));
   __ br(rscratch1);
   __ bind(L_no_such_method);
-  __ far_jump(RuntimeAddress(StubRoutines::throw_AbstractMethodError_entry()));
+  __ far_jump(RuntimeAddress(SharedRuntime::throw_AbstractMethodError_entry()));
 }
 
 void MethodHandles::jump_to_lambda_form(MacroAssembler* _masm,
@@ -451,7 +451,7 @@ void MethodHandles::generate_method_handle_dispatch(MacroAssembler* _masm,
     jump_from_method_handle(_masm, rmethod, temp1, for_compiler_entry);
     if (iid == vmIntrinsics::_linkToInterface) {
       __ bind(L_incompatible_class_change_error);
-      __ far_jump(RuntimeAddress(StubRoutines::throw_IncompatibleClassChangeError_entry()));
+      __ far_jump(RuntimeAddress(SharedRuntime::throw_IncompatibleClassChangeError_entry()));
     }
   }
 }

src/hotspot/cpu/aarch64/sharedRuntime_aarch64.cpp

@@ -66,6 +66,12 @@
 
 #define __ masm->
 
+#ifdef PRODUCT
+#define BLOCK_COMMENT(str) /* nothing */
+#else
+#define BLOCK_COMMENT(str) __ block_comment(str)
+#endif
+
 const int StackAlignmentInSlots = StackAlignmentInBytes / VMRegImpl::stack_slot_size;
 
 // FIXME -- this is used by C1
@@ -2764,3 +2770,193 @@ RuntimeStub* SharedRuntime::generate_resolve_blob(address destination, const cha
   // frame_size_words or bytes??
   return RuntimeStub::new_runtime_stub(name, &buffer, frame_complete, frame_size_in_words, oop_maps, true);
 }
+
+// Continuation point for throwing of implicit exceptions that are
+// not handled in the current activation. Fabricates an exception
+// oop and initiates normal exception dispatching in this
+// frame. Since we need to preserve callee-saved values (currently
+// only for C2, but done for C1 as well) we need a callee-saved oop
+// map and therefore have to make these stubs into RuntimeStubs
+// rather than BufferBlobs.  If the compiler needs all registers to
+// be preserved between the fault point and the exception handler
+// then it must assume responsibility for that in
+// AbstractCompiler::continuation_for_implicit_null_exception or
+// continuation_for_implicit_division_by_zero_exception. All other
+// implicit exceptions (e.g., NullPointerException or
+// AbstractMethodError on entry) are either at call sites or
+// otherwise assume that stack unwinding will be initiated, so
+// caller saved registers were assumed volatile in the compiler.
+
+RuntimeStub* SharedRuntime::generate_throw_exception(const char* name, address runtime_entry) {
+  // Information about frame layout at time of blocking runtime call.
+  // Note that we only have to preserve callee-saved registers since
+  // the compilers are responsible for supplying a continuation point
+  // if they expect all registers to be preserved.
+  // n.b. aarch64 asserts that frame::arg_reg_save_area_bytes == 0
+  enum layout {
+    rfp_off = 0,
+    rfp_off2,
+    return_off,
+    return_off2,
+    framesize // inclusive of return address
+  };
+
+  int insts_size = 512;
+  int locs_size  = 64;
+
+  ResourceMark rm;
+  const char* timer_msg = "SharedRuntime generate_throw_exception";
+  TraceTime timer(timer_msg, TRACETIME_LOG(Info, startuptime));
+
+  CodeBuffer code(name, insts_size, locs_size);
+  OopMapSet* oop_maps  = new OopMapSet();
+  MacroAssembler* masm = new MacroAssembler(&code);
+
+  address start = __ pc();
+
+  // This is an inlined and slightly modified version of call_VM
+  // which has the ability to fetch the return PC out of
+  // thread-local storage and also sets up last_Java_sp slightly
+  // differently than the real call_VM
+
+  __ enter(); // Save FP and LR before call
+
+  assert(is_even(framesize/2), "sp not 16-byte aligned");
+
+  // lr and fp are already in place
+  __ sub(sp, rfp, ((uint64_t)framesize-4) << LogBytesPerInt); // prolog
+
+  int frame_complete = __ pc() - start;
+
+  // Set up last_Java_sp and last_Java_fp
+  address the_pc = __ pc();
+  __ set_last_Java_frame(sp, rfp, the_pc, rscratch1);
+
+  __ mov(c_rarg0, rthread);
+  BLOCK_COMMENT("call runtime_entry");
+  __ mov(rscratch1, runtime_entry);
+  __ blr(rscratch1);
+
+  // Generate oop map
+  OopMap* map = new OopMap(framesize, 0);
+
+  oop_maps->add_gc_map(the_pc - start, map);
+
+  __ reset_last_Java_frame(true);
+
+  // Reinitialize the ptrue predicate register, in case the external runtime
+  // call clobbers ptrue reg, as we may return to SVE compiled code.
+  __ reinitialize_ptrue();
+
+  __ leave();
+
+  // check for pending exceptions
+#ifdef ASSERT
+  Label L;
+  __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
+  __ cbnz(rscratch1, L);
+  __ should_not_reach_here();
+  __ bind(L);
+#endif // ASSERT
+  __ far_jump(RuntimeAddress(StubRoutines::forward_exception_entry()));
+
+  // codeBlob framesize is in words (not VMRegImpl::slot_size)
+  RuntimeStub* stub =
+    RuntimeStub::new_runtime_stub(name,
+                                  &code,
+                                  frame_complete,
+                                  (framesize >> (LogBytesPerWord - LogBytesPerInt)),
+                                  oop_maps, false);
+  return stub;
+}
+
+#if INCLUDE_JFR
+
+static void jfr_prologue(address the_pc, MacroAssembler* masm, Register thread) {
+  __ set_last_Java_frame(sp, rfp, the_pc, rscratch1);
+  __ mov(c_rarg0, thread);
+}
+
+// The handle is dereferenced through a load barrier.
+static void jfr_epilogue(MacroAssembler* masm) {
+  __ reset_last_Java_frame(true);
+}
+
+// For c2: c_rarg0 is junk, call to runtime to write a checkpoint.
+// It returns a jobject handle to the event writer.
+// The handle is dereferenced and the return value is the event writer oop.
+RuntimeStub* SharedRuntime::generate_jfr_write_checkpoint() {
+  enum layout {
+    rbp_off,
+    rbpH_off,
+    return_off,
+    return_off2,
+    framesize // inclusive of return address
+  };
+
+  int insts_size = 1024;
+  int locs_size = 64;
+  CodeBuffer code("jfr_write_checkpoint", insts_size, locs_size);
+  OopMapSet* oop_maps = new OopMapSet();
+  MacroAssembler* masm = new MacroAssembler(&code);
+
+  address start = __ pc();
+  __ enter();
+  int frame_complete = __ pc() - start;
+  address the_pc = __ pc();
+  jfr_prologue(the_pc, masm, rthread);
+  __ call_VM_leaf(CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::write_checkpoint), 1);
+  jfr_epilogue(masm);
+  __ resolve_global_jobject(r0, rscratch1, rscratch2);
+  __ leave();
+  __ ret(lr);
+
+  OopMap* map = new OopMap(framesize, 1); // rfp
+  oop_maps->add_gc_map(the_pc - start, map);
+
+  RuntimeStub* stub = // codeBlob framesize is in words (not VMRegImpl::slot_size)
+    RuntimeStub::new_runtime_stub("jfr_write_checkpoint", &code, frame_complete,
+                                  (framesize >> (LogBytesPerWord - LogBytesPerInt)),
+                                  oop_maps, false);
+  return stub;
+}
+
+// For c2: call to return a leased buffer.
+RuntimeStub* SharedRuntime::generate_jfr_return_lease() {
+  enum layout {
+    rbp_off,
+    rbpH_off,
+    return_off,
+    return_off2,
+    framesize // inclusive of return address
+  };
+
+  int insts_size = 1024;
+  int locs_size = 64;
+
+  CodeBuffer code("jfr_return_lease", insts_size, locs_size);
+  OopMapSet* oop_maps = new OopMapSet();
+  MacroAssembler* masm = new MacroAssembler(&code);
+
+  address start = __ pc();
+  __ enter();
+  int frame_complete = __ pc() - start;
+  address the_pc = __ pc();
+  jfr_prologue(the_pc, masm, rthread);
+  __ call_VM_leaf(CAST_FROM_FN_PTR(address, JfrIntrinsicSupport::return_lease), 1);
+  jfr_epilogue(masm);
+
+  __ leave();
+  __ ret(lr);
+
+  OopMap* map = new OopMap(framesize, 1); // rfp
+  oop_maps->add_gc_map(the_pc - start, map);
+
+  RuntimeStub* stub = // codeBlob framesize is in words (not VMRegImpl::slot_size)
+    RuntimeStub::new_runtime_stub("jfr_return_lease", &code, frame_complete,
+                                  (framesize >> (LogBytesPerWord - LogBytesPerInt)),
+                                  oop_maps, false);
+  return stub;
+}
+
+#endif // INCLUDE_JFR