Some ARM fixes.

openjdk · May 31, 2021 · 9e283a4 · 9e283a4
1 parent d29e238
commit 9e283a4
Show file tree

Hide file tree

Showing 14 changed files with 44 additions and 33 deletions.
diff --git a/src/hotspot/cpu/aarch64/aarch64.ad b/src/hotspot/cpu/aarch64/aarch64.ad
@@ -3648,6 +3648,8 @@ encode %{
       }
     }
 
+    __ post_call_nop();
+
     // Only non uncommon_trap calls need to reinitialize ptrue.
     if (Compile::current()->max_vector_size() >= 16 && uncommon_trap_request() == 0) {
       __ reinitialize_ptrue();
@@ -3658,6 +3660,7 @@ encode %{
     C2_MacroAssembler _masm(&cbuf);
     int method_index = resolved_method_index(cbuf);
     address call = __ ic_call((address)$meth$$method, method_index);
+    __ post_call_nop();
     if (call == NULL) {
       ciEnv::current()->record_failure("CodeCache is full");
       return;
@@ -3685,6 +3688,7 @@ encode %{
     CodeBlob *cb = CodeCache::find_blob(entry);
     if (cb) {
       address call = __ trampoline_call(Address(entry, relocInfo::runtime_call_type));
+      __ post_call_nop();
       if (call == NULL) {
         ciEnv::current()->record_failure("CodeCache is full");
         return;
@@ -3697,6 +3701,7 @@ encode %{
       __ stp(zr, rscratch2, Address(__ pre(sp, -2 * wordSize)));
       __ blr(rscratch1);
       __ bind(retaddr);
+      // __ post_call_nop(); -- can we have one here? do we need one?
       __ add(sp, sp, 2 * wordSize);
     }
     if (Compile::current()->max_vector_size() >= 16) {

diff --git a/src/hotspot/cpu/aarch64/c1_LIRAssembler_aarch64.cpp b/src/hotspot/cpu/aarch64/c1_LIRAssembler_aarch64.cpp
@@ -2060,6 +2060,7 @@ void LIR_Assembler::call(LIR_OpJavaCall* op, relocInfo::relocType rtype) {
     return;
   }
   add_call_info(code_offset(), op->info());
+  __ post_call_nop();
 }
 
 
@@ -2070,6 +2071,7 @@ void LIR_Assembler::ic_call(LIR_OpJavaCall* op) {
     return;
   }
   add_call_info(code_offset(), op->info());
+  __ post_call_nop();
 }
 
 void LIR_Assembler::emit_static_call_stub() {
@@ -2910,6 +2912,7 @@ void LIR_Assembler::rt_call(LIR_Opr result, address dest, const LIR_OprList* arg
   if (info != NULL) {
     add_call_info_here(info);
   }
+  __ post_call_nop();
 }
 
 void LIR_Assembler::volatile_move_op(LIR_Opr src, LIR_Opr dest, BasicType type, CodeEmitInfo* info) {

diff --git a/src/hotspot/cpu/aarch64/frame_aarch64.cpp b/src/hotspot/cpu/aarch64/frame_aarch64.cpp
@@ -275,10 +275,12 @@ void frame::patch_pc(Thread* thread, address pc) {
   // Either the return address is the original one or we are going to
   // patch in the same address that's already there.
   assert(_pc == *pc_addr || pc == *pc_addr, "must be");
+  DEBUG_ONLY(address old_pc = _pc;)
   *pc_addr = pc;
+  _pc = pc; // must be set before call to get_deopt_original_pc
   address original_pc = CompiledMethod::get_deopt_original_pc(this);
   if (original_pc != NULL) {
-    assert(original_pc == _pc, "expected original PC to be stored before patching");
+    assert(original_pc == old_pc, "expected original PC to be stored before patching");
     _deopt_state = is_deoptimized;
     // leave _pc as is
   } else {

diff --git a/src/hotspot/cpu/aarch64/gc/shared/barrierSetNMethod_aarch64.cpp b/src/hotspot/cpu/aarch64/gc/shared/barrierSetNMethod_aarch64.cpp
@@ -28,6 +28,7 @@
 #include "gc/shared/barrierSetNMethod.hpp"
 #include "logging/log.hpp"
 #include "memory/resourceArea.hpp"
+#include "runtime/frame.inline.hpp"
 #include "runtime/sharedRuntime.hpp"
 #include "runtime/registerMap.hpp"
 #include "runtime/thread.hpp"

diff --git a/src/hotspot/cpu/aarch64/instanceStackChunkKlass_aarch64.inline.hpp b/src/hotspot/cpu/aarch64/instanceStackChunkKlass_aarch64.inline.hpp
@@ -27,6 +27,7 @@
 
 #include "interpreter/oopMapCache.hpp"
 #include "runtime/frame.inline.hpp"
+#include "runtime/registerMap.hpp"
 
 int InstanceStackChunkKlass::metadata_words() { return frame::sender_sp_offset; }
 int InstanceStackChunkKlass::align_wiggle()   { return 1; }

diff --git a/src/hotspot/cpu/aarch64/macroAssembler_aarch64.hpp b/src/hotspot/cpu/aarch64/macroAssembler_aarch64.hpp
@@ -696,6 +696,9 @@ class MacroAssembler: public Assembler {
   // Alignment
   void align(int modulus);
 
+  // nop
+  void post_call_nop() { nop(); }
+
   // Stack frame creation/removal
   void enter()
   {

diff --git a/src/hotspot/cpu/aarch64/nativeInst_aarch64.cpp b/src/hotspot/cpu/aarch64/nativeInst_aarch64.cpp
@@ -546,22 +546,12 @@ address NativeCall::trampoline_jump(CodeBuffer &cbuf, address dest) {
 }
 
 void NativePostCallNop::make_deopt() {
-  // TODO LOOM AARCH64
   NativeDeoptInstruction::insert(addr_at(0));
-  // makes the first 2 bytes into UD, the rest is thrash. Make NOPs to help debugging.
-  set_char_at(2, 0x90);
-  set_char_at(3, 0x90);
-  set_char_at(4, 0x90);
-  set_char_at(5, 0x90);
-  set_char_at(6, 0x90);
-  set_char_at(7, 0x90);
 }
 
 void NativePostCallNop::patch(jint diff) {
   // TODO LOOM AARCH64
-  assert(diff != 0, "must be");
-  int32_t *code_pos = (int32_t *) addr_at(displacement_offset);
-  *((int32_t *)(code_pos)) = (int32_t) diff;
+  // unsupported for now
 }
 
 void NativeDeoptInstruction::verify() {
@@ -570,7 +560,12 @@ void NativeDeoptInstruction::verify() {
 // Inserts an undefined instruction at a given pc
 void NativeDeoptInstruction::insert(address code_pos) {
   // TODO LOOM AARCH64
-  *code_pos = instruction_prefix;
-  *(code_pos+1) = instruction_code;
-  ICache::invalidate_range(code_pos, instruction_size);
+  // 1 1 0 1 | 0 1 0 0 | 0 0 1 imm16 0 0 0 0 0
+  // d       | 4       | 2      | 0 | 0 | 0 |
+  // 0xd4, 0x20, 0x00, 0x00
+  *code_pos = 0xd4;
+  *(code_pos+1) = 0x20;
+  *(code_pos+2) = 0x00;
+  *(code_pos+3) = 0x00;
+  ICache::invalidate_range(code_pos, 4);
 }
diff --git a/src/hotspot/cpu/aarch64/nativeInst_aarch64.hpp b/src/hotspot/cpu/aarch64/nativeInst_aarch64.hpp
@@ -71,7 +71,7 @@ class NativeInstruction {
     return (encoding() & 0xff000000) == 0x10000000;
   }
 
-  inline bool is_nop();
+  inline bool is_nop() const;
   bool is_jump();
   bool is_general_jump();
   inline bool is_jump_or_nop();
@@ -543,7 +543,7 @@ class NativeTstRegMem: public NativeInstruction {
 public:
 };
 
-inline bool NativeInstruction::is_nop() {
+inline bool NativeInstruction::is_nop() const{
   uint32_t insn = *(uint32_t*)addr_at(0);
   return insn == 0xd503201f;
 }
@@ -668,15 +668,8 @@ inline NativeLdSt* NativeLdSt_at(address addr) {
 
 class NativePostCallNop: public NativeInstruction {
 public:
-  enum Intel_specific_constants { // TODO LOOM AARCH64
-    instruction_code = 0x0f,
-    instruction_size = 8,
-    instruction_offset = 0,
-    displacement_offset = 4
-  };
-
-  bool check() const { return int_at(0) == 0x841f0f; }
-  int displacement() const { return (jint) int_at(displacement_offset); }
+  bool check() const { return is_nop(); }
+  int displacement() const { return 0; }
   void patch(jint diff);
   void make_deopt();
 };
@@ -698,9 +691,7 @@ inline NativePostCallNop* nativePostCallNop_unsafe_at(address address) {
 class NativeDeoptInstruction: public NativeInstruction {
  public:
   enum Intel_specific_constants { // TODO LOOM AARCH64
-    instruction_prefix          = 0x0F,
-    instruction_code            = 0xFF,
-    instruction_size            =    2,
+    instruction_size            =    4,
     instruction_offset          =    0,
   };
 
@@ -710,8 +701,10 @@ class NativeDeoptInstruction: public NativeInstruction {
   void  verify();
 
   static bool is_deopt_at(address instr) {
-    return ((*instr) & 0xFF) == NativeDeoptInstruction::instruction_prefix && 
-      ((*(instr+1)) & 0xFF) == NativeDeoptInstruction::instruction_code;
+    return ((*instr) & 0xFF) == 0xd4 && 
+      ((*(instr+1)) & 0xFF) == 0x20 &&
+      ((*(instr+2)) & 0xFF) == 0x00 &&
+      ((*(instr+3)) & 0xFF) == 0x00;
   }
 
   // MT-safe patching

diff --git a/src/hotspot/cpu/arm/frame_arm.cpp b/src/hotspot/cpu/arm/frame_arm.cpp
@@ -230,10 +230,12 @@ void frame::patch_pc(Thread* thread, address pc) {
     tty->print_cr("patch_pc at address" INTPTR_FORMAT " [" INTPTR_FORMAT " -> " INTPTR_FORMAT "] ",
                   p2i(pc_addr), p2i(*pc_addr), p2i(pc));
   }
+  DEBUG_ONLY(address old_pc = _pc;)
   *pc_addr = pc;
+  _pc = pc; // must be set before call to get_deopt_original_pc
   address original_pc = CompiledMethod::get_deopt_original_pc(this);
   if (original_pc != NULL) {
-    assert(original_pc == _pc, "expected original PC to be stored before patching");
+    assert(original_pc == old_pc, "expected original PC to be stored before patching");
     _deopt_state = is_deoptimized;
     // leave _pc as is
   } else {

diff --git a/src/hotspot/cpu/x86/stubGenerator_x86_64.cpp b/src/hotspot/cpu/x86/stubGenerator_x86_64.cpp
@@ -3891,6 +3891,7 @@ class StubGenerator: public StubCodeGenerator {
     StubRoutines::_arrayof_oop_disjoint_arraycopy_uninit    = StubRoutines::_oop_disjoint_arraycopy_uninit;
     StubRoutines::_arrayof_oop_arraycopy_uninit             = StubRoutines::_oop_arraycopy_uninit;
 
+    StubRoutines::_has_word_memcpy     = true;
     StubRoutines::_word_memcpy_up      = generate_disjoint_word_copy_up  (false, "word_memcpy_up");
     StubRoutines::_word_memcpy_up_nt   = generate_disjoint_word_copy_up  (true,  "word_memcpy_up_nt");
     StubRoutines::_word_memcpy_down    = generate_disjoint_word_copy_down(0,     "word_memcpy_down");

diff --git a/src/hotspot/share/code/codeBlob.hpp b/src/hotspot/share/code/codeBlob.hpp
@@ -27,6 +27,7 @@
 
 #include "asm/codeBuffer.hpp"
 #include "compiler/compilerDefinitions.hpp"
+#include "compiler/oopMap.inline.hpp"
 #include "runtime/frame.hpp"
 #include "runtime/handles.hpp"
 #include "utilities/align.hpp"

diff --git a/src/hotspot/share/oops/instanceStackChunkKlass.cpp b/src/hotspot/share/oops/instanceStackChunkKlass.cpp
@@ -22,6 +22,7 @@
  *
  */
 
+#include "compiler/oopMap.inline.hpp"
 #include "oops/instanceStackChunkKlass.hpp"
 #include "gc/shared/gc_globals.hpp"
 #include "memory/resourceArea.hpp"
@@ -157,7 +158,7 @@ MemcpyFnT InstanceStackChunkKlass::memcpy_fn_from_stack_to_chunk = nullptr;
 MemcpyFnT InstanceStackChunkKlass::memcpy_fn_from_chunk_to_stack = nullptr;
 
 void InstanceStackChunkKlass::resolve_memcpy_functions() {
-  if (UseNewCode) {
+  if (!StubRoutines::has_word_memcpy() || UseNewCode) {
     // tty->print_cr(">> Config memcpy: default");
     memcpy_fn_from_stack_to_chunk = (MemcpyFnT)InstanceStackChunkKlass::default_memcpy;
     memcpy_fn_from_chunk_to_stack = (MemcpyFnT)InstanceStackChunkKlass::default_memcpy;

diff --git a/src/hotspot/share/runtime/stubRoutines.cpp b/src/hotspot/share/runtime/stubRoutines.cpp
@@ -99,6 +99,7 @@ address StubRoutines::_jlong_disjoint_arraycopy          = CAST_FROM_FN_PTR(addr
 address StubRoutines::_oop_disjoint_arraycopy            = CAST_FROM_FN_PTR(address, StubRoutines::oop_copy);
 address StubRoutines::_oop_disjoint_arraycopy_uninit     = CAST_FROM_FN_PTR(address, StubRoutines::oop_copy_uninit);
 
+bool    StubRoutines::_has_word_memcpy             = false;
 address StubRoutines::_word_memcpy_up              = NULL; // CAST_FROM_FN_PTR(address, StubRoutines::jint_copy);
 address StubRoutines::_word_memcpy_down            = NULL; // CAST_FROM_FN_PTR(address, StubRoutines::jint_copy);
 address StubRoutines::_word_memcpy_up_nt           = NULL; // CAST_FROM_FN_PTR(address, StubRoutines::jint_copy);

diff --git a/src/hotspot/share/runtime/stubRoutines.hpp b/src/hotspot/share/runtime/stubRoutines.hpp
@@ -177,6 +177,7 @@ class StubRoutines: AllStatic {
   static address _jlong_disjoint_arraycopy;
   static address _oop_disjoint_arraycopy, _oop_disjoint_arraycopy_uninit;
 
+  static bool _has_word_memcpy;
   static address _word_memcpy_up;
   static address _word_memcpy_down;
   static address _word_memcpy_up_nt;
@@ -359,6 +360,7 @@ class StubRoutines: AllStatic {
   static address jint_arraycopy()   { return _jint_arraycopy; }
   static address jlong_arraycopy()  { return _jlong_arraycopy; }
 
+  static bool has_word_memcpy()        { return _has_word_memcpy; }
   static address word_memcpy_up()      { return _word_memcpy_up; }
   static address word_memcpy_up_nt()   { return _word_memcpy_up_nt; }
   static address word_memcpy_down()    { return _word_memcpy_down; }