8331253: 16 bits is not enough for nmethod::_skipped_instructions_size field

Reviewed-by: dlong, thartmann

src/hotspot/cpu/aarch64/gc/z/zBarrierSetAssembler_aarch64.cpp

@@ -1141,7 +1141,6 @@ public:
 #define __ masm->
 
 void ZBarrierSetAssembler::generate_c2_load_barrier_stub(MacroAssembler* masm, ZLoadBarrierStubC2* stub) const {
-  Assembler::InlineSkippedInstructionsCounter skipped_counter(masm);
   BLOCK_COMMENT("ZLoadBarrierStubC2");
 
   // Stub entry
@@ -1160,7 +1159,6 @@ void ZBarrierSetAssembler::generate_c2_load_barrier_stub(MacroAssembler* masm, Z
 }
 
 void ZBarrierSetAssembler::generate_c2_store_barrier_stub(MacroAssembler* masm, ZStoreBarrierStubC2* stub) const {
-  Assembler::InlineSkippedInstructionsCounter skipped_counter(masm);
   BLOCK_COMMENT("ZStoreBarrierStubC2");
 
   // Stub entry

src/hotspot/cpu/ppc/gc/z/zBarrierSetAssembler_ppc.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2021, 2023, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2021, 2024, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2021, 2023 SAP SE. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
@@ -887,7 +887,6 @@ class ZSetupArguments {
 #define __ masm->
 
 void ZBarrierSetAssembler::generate_c2_load_barrier_stub(MacroAssembler* masm, ZLoadBarrierStubC2* stub) const {
-  Assembler::InlineSkippedInstructionsCounter skipped_counter(masm);
   __ block_comment("generate_c2_load_barrier_stub (zgc) {");
 
   __ bind(*stub->entry());
@@ -911,7 +910,6 @@ void ZBarrierSetAssembler::generate_c2_load_barrier_stub(MacroAssembler* masm, Z
 }
 
 void ZBarrierSetAssembler::generate_c2_store_barrier_stub(MacroAssembler* masm, ZStoreBarrierStubC2* stub) const {
-  Assembler::InlineSkippedInstructionsCounter skipped_counter(masm);
   __ block_comment("ZStoreBarrierStubC2");
 
   // Stub entry

src/hotspot/cpu/x86/gc/z/zBarrierSetAssembler_x86.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018, 2023, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2018, 2024, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -1213,7 +1213,6 @@ public:
 #define __ masm->
 
 void ZBarrierSetAssembler::generate_c2_load_barrier_stub(MacroAssembler* masm, ZLoadBarrierStubC2* stub) const {
-  Assembler::InlineSkippedInstructionsCounter skipped_counter(masm);
   BLOCK_COMMENT("ZLoadBarrierStubC2");
 
   // Stub entry
@@ -1233,7 +1232,6 @@ void ZBarrierSetAssembler::generate_c2_load_barrier_stub(MacroAssembler* masm, Z
 }
 
 void ZBarrierSetAssembler::generate_c2_store_barrier_stub(MacroAssembler* masm, ZStoreBarrierStubC2* stub) const {
-  Assembler::InlineSkippedInstructionsCounter skipped_counter(masm);
   BLOCK_COMMENT("ZStoreBarrierStubC2");
 
   // Stub entry

src/hotspot/share/asm/codeBuffer.cpp

@@ -1012,6 +1012,8 @@ void CodeBuffer::log_section_sizes(const char* name) {
 }
 
 bool CodeBuffer::finalize_stubs() {
+  // Record size of code before we generate stubs in instructions section
+  _main_code_size = _insts.size();
   if (_finalize_stubs && !pd_finalize_stubs()) {
     // stub allocation failure
     return false;

src/hotspot/share/asm/codeBuffer.hpp

@@ -427,6 +427,9 @@ class CodeBuffer: public StackObj DEBUG_ONLY(COMMA private Scrubber) {
   address      _total_start;    // first address of combined memory buffer
   csize_t      _total_size;     // size in bytes of combined memory buffer
 
+  // Size of code without stubs generated at the end of instructions section
+  csize_t      _main_code_size;
+
   OopRecorder* _oop_recorder;
 
   OopRecorder  _default_oop_recorder;  // override with initialize_oop_recorder
@@ -457,6 +460,7 @@ class CodeBuffer: public StackObj DEBUG_ONLY(COMMA private Scrubber) {
     _oop_recorder    = nullptr;
     _overflow_arena  = nullptr;
     _last_insn       = nullptr;
+    _main_code_size  = 0;
     _finalize_stubs  = false;
     _shared_stub_to_interp_requests = nullptr;
     _shared_trampoline_requests = nullptr;
@@ -630,6 +634,9 @@ class CodeBuffer: public StackObj DEBUG_ONLY(COMMA private Scrubber) {
   // number of bytes remaining in the insts section
   csize_t insts_remaining() const        { return _insts.remaining(); }
 
+  // size of code without stubs in instruction section
+  csize_t main_code_size() const         { return _main_code_size; }
+
   // is a given address in the insts section?  (2nd version is end-inclusive)
   bool insts_contains(address pc) const  { return _insts.contains(pc); }
   bool insts_contains2(address pc) const { return _insts.contains2(pc); }

src/hotspot/share/ci/ciMethod.cpp

@@ -1126,13 +1126,14 @@ int ciMethod::code_size_for_inlining() {
 // not highly relevant to an inlined method.  So we use the more
 // specific accessor nmethod::insts_size.
 // Also some instructions inside the code are excluded from inline
-// heuristic (e.g. post call nop instructions; see InlineSkippedInstructionsCounter)
+// heuristic (e.g. post call nop instructions and GC barriers;
+// see InlineSkippedInstructionsCounter).
 int ciMethod::inline_instructions_size() {
   if (_inline_instructions_size == -1) {
     GUARDED_VM_ENTRY(
       nmethod* code = get_Method()->code();
       if (code != nullptr && (code->comp_level() == CompLevel_full_optimization)) {
-        int isize = code->insts_end() - code->verified_entry_point() - code->skipped_instructions_size();
+        int isize = code->inline_insts_size();
         _inline_instructions_size = isize > 0 ? isize : 0;
       } else {
         _inline_instructions_size = 0;

src/hotspot/share/code/nmethod.cpp

@@ -131,6 +131,7 @@ struct java_nmethod_stats_struct {
   uint relocation_size;
   uint consts_size;
   uint insts_size;
+  uint inline_insts_size;
   uint stub_size;
   uint oops_size;
   uint metadata_size;
@@ -151,6 +152,7 @@ struct java_nmethod_stats_struct {
     relocation_size     += nm->relocation_size();
     consts_size         += nm->consts_size();
     insts_size          += nm->insts_size();
+    inline_insts_size   += nm->inline_insts_size();
     stub_size           += nm->stub_size();
     oops_size           += nm->oops_size();
     metadata_size       += nm->metadata_size();
@@ -185,6 +187,9 @@ struct java_nmethod_stats_struct {
     if (insts_size != 0) {
       tty->print_cr("   main code     = %u (%f%%)", insts_size, (insts_size * 100.0f)/total_nm_size);
     }
+    if (inline_insts_size != 0) {
+      tty->print_cr("     inline code = %u (%f%%)", inline_insts_size, (inline_insts_size * 100.0f)/total_nm_size);
+    }
     if (stub_size != 0) {
       tty->print_cr("   stub code     = %u (%f%%)", stub_size, (stub_size * 100.0f)/total_nm_size);
     }
@@ -1253,7 +1258,15 @@ void nmethod::init_defaults(CodeBuffer *code_buffer, CodeOffsets* offsets) {
 
   CHECKED_CAST(_entry_offset,              uint16_t, (offsets->value(CodeOffsets::Entry)));
   CHECKED_CAST(_verified_entry_offset,     uint16_t, (offsets->value(CodeOffsets::Verified_Entry)));
-  CHECKED_CAST(_skipped_instructions_size, uint16_t, (code_buffer->total_skipped_instructions_size()));
+
+  int size = code_buffer->main_code_size();
+  assert(size >= 0, "should be initialized");
+  // Use instructions section size if it is 0 (e.g. native wrapper)
+  if (size == 0) size = code_size(); // requires _stub_offset to be set
+  assert(size <= code_size(), "incorrect size: %d > %d", size, code_size());
+  _inline_insts_size = size - _verified_entry_offset
+                     - code_buffer->total_skipped_instructions_size();
+  assert(_inline_insts_size >= 0, "sanity");
 }
 
 // Post initialization

src/hotspot/share/code/nmethod.hpp

@@ -218,6 +218,8 @@ class nmethod : public CodeBlob {
 
   // _consts_offset == _content_offset because SECT_CONSTS is first in code buffer
 
+  int _inline_insts_size;
+
   int _stub_offset;
 
   // Offsets for different stubs section parts
@@ -232,7 +234,6 @@ class nmethod : public CodeBlob {
   int16_t  _unwind_handler_offset;
   // Number of arguments passed on the stack
   uint16_t _num_stack_arg_slots;
-  uint16_t _skipped_instructions_size;
 
   // Offsets in mutable data section
   // _oops_offset == _data_offset,  offset where embedded oop table begins (inside data)
@@ -589,7 +590,7 @@ public:
   int     oops_count() const { assert(oops_size() % oopSize == 0, "");  return (oops_size() / oopSize) + 1; }
   int metadata_count() const { assert(metadata_size() % wordSize == 0, ""); return (metadata_size() / wordSize) + 1; }
 
-  int skipped_instructions_size () const { return _skipped_instructions_size; }
+  int inline_insts_size() const { return _inline_insts_size; }
   int total_size() const;
 
   // Containment

src/hotspot/share/opto/output.cpp

@@ -1324,8 +1324,9 @@ CodeBuffer* PhaseOutput::init_buffer() {
 
   int pad_req   = NativeCall::instruction_size;
 
+  // GC barrier stubs are generated in code section
   BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
-  stub_req += bs->estimate_stub_size();
+  code_req += bs->estimate_stub_size();
 
   // nmethod and CodeBuffer count stubs & constants as part of method's code.
   // class HandlerImpl is platform-specific and defined in the *.ad files.
@@ -1334,9 +1335,9 @@ CodeBuffer* PhaseOutput::init_buffer() {
   stub_req += MAX_stubs_size;   // ensure per-stub margin
   code_req += MAX_inst_size;    // ensure per-instruction margin
 
-  if (StressCodeBuffers)
+  if (StressCodeBuffers) {
     code_req = const_req = stub_req = exception_handler_req = deopt_handler_req = 0x10;  // force expansion
-
+  }
   int total_req =
           const_req +
           code_req +
@@ -1345,9 +1346,10 @@ CodeBuffer* PhaseOutput::init_buffer() {
           exception_handler_req +
           deopt_handler_req;               // deopt handler
 
-  if (C->has_method_handle_invokes())
+  if (C->has_method_handle_invokes()) {
-    total_req += deopt_handler_req;  // deopt MH handler
+    total_req += deopt_handler_req;        // deopt MH handler
-
+    stub_req  += deopt_handler_req;
+  }
   CodeBuffer* cb = code_buffer();
   cb->initialize(total_req, _buf_sizes._reloc);