Merge branch 'master' into 8375477_coreutils

Merge

doc/testing.html

@@ -72,6 +72,7 @@ id="toc-notes-for-specific-tests">Notes for Specific Tests</a>
 <li><a href="#non-us-locale" id="toc-non-us-locale">Non-US
 locale</a></li>
 <li><a href="#pkcs11-tests" id="toc-pkcs11-tests">PKCS11 Tests</a></li>
+<li><a href="#sctp-tests" id="toc-sctp-tests">SCTP Tests</a></li>
 <li><a href="#testing-ahead-of-time-optimizations"
 id="toc-testing-ahead-of-time-optimizations">Testing Ahead-of-time
 Optimizations</a></li>
@@ -621,6 +622,21 @@ element of the appropriate <code>@Artifact</code> class. (See
     JTREG=&quot;JAVA_OPTIONS=-Djdk.test.lib.artifacts.nsslib-linux_aarch64=/path/to/NSS-libs&quot;</code></pre>
 <p>For more notes about the PKCS11 tests, please refer to
 test/jdk/sun/security/pkcs11/README.</p>
+<h3 id="sctp-tests">SCTP Tests</h3>
+<p>The SCTP tests require the SCTP runtime library, which is often not
+installed by default in popular Linux distributions. Without this
+library, the SCTP tests will be skipped. If you want to enable the SCTP
+tests, you should install the SCTP library before running the tests.</p>
+<p>For distributions using the .deb packaging format and the apt tool
+(such as Debian, Ubuntu, etc.), try this:</p>
+<pre><code>sudo apt install libsctp1
+sudo modprobe sctp
+lsmod | grep sctp</code></pre>
+<p>For distributions using the .rpm packaging format and the dnf tool
+(such as Fedora, Red Hat, etc.), try this:</p>
+<pre><code>sudo dnf install -y lksctp-tools
+sudo modprobe sctp
+lsmod | grep sctp</code></pre>
 <h3 id="testing-ahead-of-time-optimizations">Testing Ahead-of-time
 Optimizations</h3>
 <p>One way to improve test coverage of ahead-of-time (AOT) optimizations

doc/testing.md

@@ -640,6 +640,32 @@ $ make test TEST="jtreg:sun/security/pkcs11/Secmod/AddTrustedCert.java" \
 For more notes about the PKCS11 tests, please refer to
 test/jdk/sun/security/pkcs11/README.
 
+
+### SCTP Tests
+
+The SCTP tests require the SCTP runtime library, which is often not installed
+by default in popular Linux distributions. Without this library, the SCTP tests
+will be skipped. If you want to enable the SCTP tests, you should install the
+SCTP library before running the tests.
+
+For distributions using the .deb packaging format and the apt tool
+(such as Debian, Ubuntu, etc.), try this:
+
+```
+sudo apt install libsctp1
+sudo modprobe sctp
+lsmod | grep sctp
+```
+
+For distributions using the .rpm packaging format and the dnf tool
+(such as Fedora, Red Hat, etc.), try this:
+
+```
+sudo dnf install -y lksctp-tools
+sudo modprobe sctp
+lsmod | grep sctp
+```
+
 ### Testing Ahead-of-time Optimizations
 
 One way to improve test coverage of ahead-of-time (AOT) optimizations in

make/autoconf/flags-cflags.m4

@@ -69,22 +69,18 @@ AC_DEFUN([FLAGS_SETUP_DEBUG_SYMBOLS],
   # Debug prefix mapping if supported by compiler
   DEBUG_PREFIX_CFLAGS=
 
-  UTIL_ARG_WITH(NAME: native-debug-symbols-level, TYPE: string,
-    DEFAULT: "",
-    RESULT: DEBUG_SYMBOLS_LEVEL,
+  UTIL_ARG_WITH(NAME: native-debug-symbols-level, TYPE: literal,
+    DEFAULT: [auto], VALID_VALUES: [auto 1 2 3],
+    CHECK_AVAILABLE: [
+      if test x$TOOLCHAIN_TYPE = xmicrosoft; then
+        AVAILABLE=false
+      fi
+    ],
     DESC: [set the native debug symbol level (GCC and Clang only)],
-    DEFAULT_DESC: [toolchain default])
-  AC_SUBST(DEBUG_SYMBOLS_LEVEL)
-
-  if test "x${TOOLCHAIN_TYPE}" = xgcc || \
-     test "x${TOOLCHAIN_TYPE}" = xclang; then
-    DEBUG_SYMBOLS_LEVEL_FLAGS="-g"
-    if test "x${DEBUG_SYMBOLS_LEVEL}" != "x"; then
-      DEBUG_SYMBOLS_LEVEL_FLAGS="-g${DEBUG_SYMBOLS_LEVEL}"
-      FLAGS_COMPILER_CHECK_ARGUMENTS(ARGUMENT: [${DEBUG_SYMBOLS_LEVEL_FLAGS}],
-          IF_FALSE: AC_MSG_ERROR("Debug info level ${DEBUG_SYMBOLS_LEVEL} is not supported"))
-    fi
-  fi
+    DEFAULT_DESC: [toolchain default],
+    IF_AUTO: [
+      RESULT=""
+    ])
 
   # Debug symbols
   if test "x$TOOLCHAIN_TYPE" = xgcc; then
@@ -111,8 +107,8 @@ AC_DEFUN([FLAGS_SETUP_DEBUG_SYMBOLS],
     fi
 
     # Debug info level should follow the debug format to be effective.
-    CFLAGS_DEBUG_SYMBOLS="-gdwarf-4 ${DEBUG_SYMBOLS_LEVEL_FLAGS}"
-    ASFLAGS_DEBUG_SYMBOLS="${DEBUG_SYMBOLS_LEVEL_FLAGS}"
+    CFLAGS_DEBUG_SYMBOLS="-gdwarf-4 -g${NATIVE_DEBUG_SYMBOLS_LEVEL}"
+    ASFLAGS_DEBUG_SYMBOLS="-g${NATIVE_DEBUG_SYMBOLS_LEVEL}"
   elif test "x$TOOLCHAIN_TYPE" = xclang; then
     if test "x$ALLOW_ABSOLUTE_PATHS_IN_OUTPUT" = "xfalse"; then
       # Check if compiler supports -fdebug-prefix-map. If so, use that to make
@@ -132,8 +128,8 @@ AC_DEFUN([FLAGS_SETUP_DEBUG_SYMBOLS],
         IF_FALSE: [GDWARF_FLAGS=""])
 
     # Debug info level should follow the debug format to be effective.
-    CFLAGS_DEBUG_SYMBOLS="${GDWARF_FLAGS} ${DEBUG_SYMBOLS_LEVEL_FLAGS}"
-    ASFLAGS_DEBUG_SYMBOLS="${DEBUG_SYMBOLS_LEVEL_FLAGS}"
+    CFLAGS_DEBUG_SYMBOLS="${GDWARF_FLAGS} -g${NATIVE_DEBUG_SYMBOLS_LEVEL}"
+    ASFLAGS_DEBUG_SYMBOLS="-g${NATIVE_DEBUG_SYMBOLS_LEVEL}"
   elif test "x$TOOLCHAIN_TYPE" = xmicrosoft; then
     CFLAGS_DEBUG_SYMBOLS="-Z7"
   fi

make/hotspot/lib/CompileGtest.gmk

@@ -1,5 +1,5 @@
 #
-# Copyright (c) 2016, 2025, Oracle and/or its affiliates. All rights reserved.
+# Copyright (c) 2016, 2026, 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
@@ -61,7 +61,8 @@ $(eval $(call SetupJdkLibrary, BUILD_GTEST_LIBGTEST, \
     INCLUDE_FILES := gtest-all.cc gmock-all.cc, \
     DISABLED_WARNINGS_gcc := format-nonliteral maybe-uninitialized undef \
         unused-result zero-as-null-pointer-constant, \
-    DISABLED_WARNINGS_clang := format-nonliteral undef unused-result, \
+    DISABLED_WARNINGS_clang := format-nonliteral undef unused-result \
+        zero-as-null-pointer-constant, \
     DISABLED_WARNINGS_microsoft := 4530, \
     DEFAULT_CFLAGS := false, \
     CFLAGS := $(JVM_CFLAGS) \

make/modules/jdk.security.auth/Lib.gmk

@@ -31,13 +31,14 @@ include LibCommon.gmk
 ## Build libjaas
 ################################################################################
 
-$(eval $(call SetupJdkLibrary, BUILD_LIBJAAS, \
-    NAME := jaas, \
-    OPTIMIZATION := LOW, \
-    EXTRA_HEADER_DIRS := java.base:libjava, \
-    LIBS_windows := advapi32.lib mpr.lib netapi32.lib user32.lib, \
-))
-
-TARGETS += $(BUILD_LIBJAAS)
+ifeq ($(call isTargetOs, windows), true)
+  $(eval $(call SetupJdkLibrary, BUILD_LIBJAAS, \
+      NAME := jaas, \
+      OPTIMIZATION := LOW, \
+      EXTRA_HEADER_DIRS := java.base:libjava, \
+      LIBS_windows := advapi32.lib mpr.lib netapi32.lib user32.lib, \
+  ))
 
+  TARGETS += $(BUILD_LIBJAAS)
+endif
 ################################################################################

src/hotspot/cpu/aarch64/macroAssembler_aarch64.cpp

@@ -5782,6 +5782,9 @@ address MacroAssembler::arrays_equals(Register a1, Register a2, Register tmp3,
     //      return false;
     bind(A_IS_NOT_NULL);
     ldrw(cnt1, Address(a1, length_offset));
+    ldrw(tmp5, Address(a2, length_offset));
+    cmp(cnt1, tmp5);
+    br(NE, DONE); // If lengths differ, return false
     // Increase loop counter by diff between base- and actual start-offset.
     addw(cnt1, cnt1, extra_length);
     lea(a1, Address(a1, start_offset));
@@ -5848,6 +5851,9 @@ address MacroAssembler::arrays_equals(Register a1, Register a2, Register tmp3,
     cbz(a1, DONE);
     ldrw(cnt1, Address(a1, length_offset));
     cbz(a2, DONE);
+    ldrw(tmp5, Address(a2, length_offset));
+    cmp(cnt1, tmp5);
+    br(NE, DONE); // If lengths differ, return false
     // Increase loop counter by diff between base- and actual start-offset.
     addw(cnt1, cnt1, extra_length);
 

src/hotspot/cpu/aarch64/sharedRuntime_aarch64.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2003, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2003, 2026, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2014, 2021, Red Hat Inc. All rights reserved.
  * Copyright (c) 2021, Azul Systems, Inc. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
@@ -722,22 +722,20 @@ void SharedRuntime::generate_i2c2i_adapters(MacroAssembler *masm,
 
   // Class initialization barrier for static methods
   entry_address[AdapterBlob::C2I_No_Clinit_Check] = nullptr;
-  if (VM_Version::supports_fast_class_init_checks()) {
-    Label L_skip_barrier;
+  assert(VM_Version::supports_fast_class_init_checks(), "sanity");
+  Label L_skip_barrier;
 
-    { // Bypass the barrier for non-static methods
-      __ ldrh(rscratch1, Address(rmethod, Method::access_flags_offset()));
-      __ andsw(zr, rscratch1, JVM_ACC_STATIC);
-      __ br(Assembler::EQ, L_skip_barrier); // non-static
-    }
+  // Bypass the barrier for non-static methods
+  __ ldrh(rscratch1, Address(rmethod, Method::access_flags_offset()));
+  __ andsw(zr, rscratch1, JVM_ACC_STATIC);
+  __ br(Assembler::EQ, L_skip_barrier); // non-static
 
-    __ load_method_holder(rscratch2, rmethod);
-    __ clinit_barrier(rscratch2, rscratch1, &L_skip_barrier);
-    __ far_jump(RuntimeAddress(SharedRuntime::get_handle_wrong_method_stub()));
+  __ load_method_holder(rscratch2, rmethod);
+  __ clinit_barrier(rscratch2, rscratch1, &L_skip_barrier);
+  __ far_jump(RuntimeAddress(SharedRuntime::get_handle_wrong_method_stub()));
 
-    __ bind(L_skip_barrier);
-    entry_address[AdapterBlob::C2I_No_Clinit_Check] = __ pc();
-  }
+  __ bind(L_skip_barrier);
+  entry_address[AdapterBlob::C2I_No_Clinit_Check] = __ pc();
 
   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
   bs->c2i_entry_barrier(masm);
@@ -1508,7 +1506,8 @@ nmethod* SharedRuntime::generate_native_wrapper(MacroAssembler* masm,
   // SVC, HVC, or SMC.  Make it a NOP.
   __ nop();
 
-  if (VM_Version::supports_fast_class_init_checks() && method->needs_clinit_barrier()) {
+  if (method->needs_clinit_barrier()) {
+    assert(VM_Version::supports_fast_class_init_checks(), "sanity");
     Label L_skip_barrier;
     __ mov_metadata(rscratch2, method->method_holder()); // InstanceKlass*
     __ clinit_barrier(rscratch2, rscratch1, &L_skip_barrier);

src/hotspot/cpu/aarch64/templateTable_aarch64.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2003, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2003, 2026, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2014, Red Hat Inc. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
@@ -2290,7 +2290,8 @@ void TemplateTable::resolve_cache_and_index_for_method(int byte_no,
   __ subs(zr, temp, (int) code);  // have we resolved this bytecode?
 
   // Class initialization barrier for static methods
-  if (VM_Version::supports_fast_class_init_checks() && bytecode() == Bytecodes::_invokestatic) {
+  if (bytecode() == Bytecodes::_invokestatic) {
+    assert(VM_Version::supports_fast_class_init_checks(), "sanity");
     __ br(Assembler::NE, L_clinit_barrier_slow);
     __ ldr(temp, Address(Rcache, in_bytes(ResolvedMethodEntry::method_offset())));
     __ load_method_holder(temp, temp);
@@ -2340,8 +2341,8 @@ void TemplateTable::resolve_cache_and_index_for_field(int byte_no,
   __ subs(zr, temp, (int) code);  // have we resolved this bytecode?
 
   // Class initialization barrier for static fields
-  if (VM_Version::supports_fast_class_init_checks() &&
-      (bytecode() == Bytecodes::_getstatic || bytecode() == Bytecodes::_putstatic)) {
+  if (bytecode() == Bytecodes::_getstatic || bytecode() == Bytecodes::_putstatic) {
+    assert(VM_Version::supports_fast_class_init_checks(), "sanity");
     const Register field_holder = temp;
 
     __ br(Assembler::NE, L_clinit_barrier_slow);

src/hotspot/cpu/arm/frame_arm.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2008, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2008, 2026, 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
@@ -356,10 +356,10 @@ frame frame::sender_for_interpreter_frame(RegisterMap* map) const {
 bool frame::is_interpreted_frame_valid(JavaThread* thread) const {
   assert(is_interpreted_frame(), "Not an interpreted frame");
   // These are reasonable sanity checks
-  if (fp() == 0 || (intptr_t(fp()) & (wordSize-1)) != 0) {
+  if (fp() == nullptr || (intptr_t(fp()) & (wordSize-1)) != 0) {
     return false;
   }
-  if (sp() == 0 || (intptr_t(sp()) & (wordSize-1)) != 0) {
+  if (sp() == nullptr || (intptr_t(sp()) & (wordSize-1)) != 0) {
     return false;
   }
   if (fp() + interpreter_frame_initial_sp_offset < sp()) {

src/hotspot/cpu/arm/nativeInst_arm_32.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2008, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2008, 2026, 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
@@ -172,7 +172,7 @@ void NativeMovConstReg::set_data(intptr_t x, address pc) {
 
     address addr = oop_addr != nullptr ? (address)oop_addr : (address)metadata_addr;
 
-    if(pc == 0) {
+    if (pc == nullptr) {
       offset = addr - instruction_address() - 8;
     } else {
       offset = addr - pc - 8;
@@ -228,7 +228,7 @@ void NativeMovConstReg::set_data(intptr_t x, address pc) {
 
 void NativeMovConstReg::set_pc_relative_offset(address addr, address pc) {
   int offset;
-  if (pc == 0) {
+  if (pc == nullptr) {
     offset = addr - instruction_address() - 8;
   } else {
     offset = addr - pc - 8;

src/hotspot/cpu/arm/nativeInst_arm_32.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2008, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2008, 2026, 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
@@ -371,7 +371,7 @@ class NativeMovConstReg: public NativeInstruction {
  public:
 
   intptr_t data() const;
-  void set_data(intptr_t x, address pc = 0);
+  void set_data(intptr_t x, address pc = nullptr);
   bool is_pc_relative() {
     return !is_movw();
   }

src/hotspot/cpu/ppc/interp_masm_ppc_64.cpp

@@ -1109,11 +1109,11 @@ void InterpreterMacroAssembler::verify_method_data_pointer() {
   lhz(R11_scratch1, in_bytes(DataLayout::bci_offset()), R28_mdx);
   ld(R12_scratch2, in_bytes(Method::const_offset()), R19_method);
   addi(R11_scratch1, R11_scratch1, in_bytes(ConstMethod::codes_offset()));
-  add(R11_scratch1, R12_scratch2, R12_scratch2);
+  add(R11_scratch1, R11_scratch1, R12_scratch2);
   cmpd(CR0, R11_scratch1, R14_bcp);
   beq(CR0, verify_continue);
 
-  call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp ), R19_method, R14_bcp, R28_mdx);
+  call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp), R19_method, R14_bcp, R28_mdx);
 
   bind(verify_continue);
 #endif

src/hotspot/cpu/ppc/sharedRuntime_ppc.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1997, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 2026, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2012, 2025 SAP SE. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
@@ -1237,26 +1237,24 @@ void SharedRuntime::generate_i2c2i_adapters(MacroAssembler *masm,
 
   // Class initialization barrier for static methods
   entry_address[AdapterBlob::C2I_No_Clinit_Check] = nullptr;
-  if (VM_Version::supports_fast_class_init_checks()) {
-    Label L_skip_barrier;
+  assert(VM_Version::supports_fast_class_init_checks(), "sanity");
+  Label L_skip_barrier;
 
-    { // Bypass the barrier for non-static methods
-      __ lhz(R0, in_bytes(Method::access_flags_offset()), R19_method);
-      __ andi_(R0, R0, JVM_ACC_STATIC);
-      __ beq(CR0, L_skip_barrier); // non-static
-    }
+  // Bypass the barrier for non-static methods
+  __ lhz(R0, in_bytes(Method::access_flags_offset()), R19_method);
+  __ andi_(R0, R0, JVM_ACC_STATIC);
+  __ beq(CR0, L_skip_barrier); // non-static
 
-    Register klass = R11_scratch1;
-    __ load_method_holder(klass, R19_method);
-    __ clinit_barrier(klass, R16_thread, &L_skip_barrier /*L_fast_path*/);
+  Register klass = R11_scratch1;
+  __ load_method_holder(klass, R19_method);
+  __ clinit_barrier(klass, R16_thread, &L_skip_barrier /*L_fast_path*/);
 
-    __ load_const_optimized(klass, SharedRuntime::get_handle_wrong_method_stub(), R0);
-    __ mtctr(klass);
-    __ bctr();
+  __ load_const_optimized(klass, SharedRuntime::get_handle_wrong_method_stub(), R0);
+  __ mtctr(klass);
+  __ bctr();
 
-    __ bind(L_skip_barrier);
-    entry_address[AdapterBlob::C2I_No_Clinit_Check] = __ pc();
-  }
+  __ bind(L_skip_barrier);
+  entry_address[AdapterBlob::C2I_No_Clinit_Check] = __ pc();
 
   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
   bs->c2i_entry_barrier(masm, /* tmp register*/ ic_klass, /* tmp register*/ receiver_klass, /* tmp register*/ code);
@@ -2210,7 +2208,8 @@ nmethod *SharedRuntime::generate_native_wrapper(MacroAssembler *masm,
   // --------------------------------------------------------------------------
   vep_start_pc = (intptr_t)__ pc();
 
-  if (VM_Version::supports_fast_class_init_checks() && method->needs_clinit_barrier()) {
+  if (method->needs_clinit_barrier()) {
+    assert(VM_Version::supports_fast_class_init_checks(), "sanity");
     Label L_skip_barrier;
     Register klass = r_temp_1;
     // Notify OOP recorder (don't need the relocation)

src/hotspot/cpu/ppc/templateTable_ppc_64.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2014, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2014, 2026, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2013, 2025 SAP SE. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
@@ -2199,7 +2199,8 @@ void TemplateTable::resolve_cache_and_index_for_method(int byte_no, Register Rca
   __ isync(); // Order load wrt. succeeding loads.
 
   // Class initialization barrier for static methods
-  if (VM_Version::supports_fast_class_init_checks() && bytecode() == Bytecodes::_invokestatic) {
+  if (bytecode() == Bytecodes::_invokestatic) {
+    assert(VM_Version::supports_fast_class_init_checks(), "sanity");
     const Register method = Rscratch;
     const Register klass  = Rscratch;
 
@@ -2244,8 +2245,8 @@ void TemplateTable::resolve_cache_and_index_for_field(int byte_no, Register Rcac
   __ isync(); // Order load wrt. succeeding loads.
 
   // Class initialization barrier for static fields
-  if (VM_Version::supports_fast_class_init_checks() &&
-      (bytecode() == Bytecodes::_getstatic || bytecode() == Bytecodes::_putstatic)) {
+  if (bytecode() == Bytecodes::_getstatic || bytecode() == Bytecodes::_putstatic) {
+    assert(VM_Version::supports_fast_class_init_checks(), "sanity");
     const Register field_holder = R4_ARG2;
 
     // InterpreterRuntime::resolve_get_put sets field_holder and finally release-stores put_code.

src/hotspot/cpu/riscv/sharedRuntime_riscv.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2003, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2003, 2026, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2014, 2020, Red Hat Inc. All rights reserved.
  * Copyright (c) 2020, 2023, Huawei Technologies Co., Ltd. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
@@ -213,7 +213,7 @@ void RegisterSaver::restore_live_registers(MacroAssembler* masm) {
 // Is vector's size (in bytes) bigger than a size saved by default?
 // riscv does not ovlerlay the floating-point registers on vector registers like aarch64.
 bool SharedRuntime::is_wide_vector(int size) {
-  return UseRVV;
+  return UseRVV && size > 0;
 }
 
 // ---------------------------------------------------------------------------
@@ -637,22 +637,20 @@ void SharedRuntime::generate_i2c2i_adapters(MacroAssembler *masm,
 
   // Class initialization barrier for static methods
   entry_address[AdapterBlob::C2I_No_Clinit_Check] = nullptr;
-  if (VM_Version::supports_fast_class_init_checks()) {
-    Label L_skip_barrier;
+  assert(VM_Version::supports_fast_class_init_checks(), "sanity");
+  Label L_skip_barrier;
 
-    { // Bypass the barrier for non-static methods
-      __ load_unsigned_short(t0, Address(xmethod, Method::access_flags_offset()));
-      __ test_bit(t1, t0, exact_log2(JVM_ACC_STATIC));
-      __ beqz(t1, L_skip_barrier); // non-static
-    }
+  // Bypass the barrier for non-static methods
+  __ load_unsigned_short(t0, Address(xmethod, Method::access_flags_offset()));
+  __ test_bit(t1, t0, exact_log2(JVM_ACC_STATIC));
+  __ beqz(t1, L_skip_barrier); // non-static
 
-    __ load_method_holder(t1, xmethod);
-    __ clinit_barrier(t1, t0, &L_skip_barrier);
-    __ far_jump(RuntimeAddress(SharedRuntime::get_handle_wrong_method_stub()));
+  __ load_method_holder(t1, xmethod);
+  __ clinit_barrier(t1, t0, &L_skip_barrier);
+  __ far_jump(RuntimeAddress(SharedRuntime::get_handle_wrong_method_stub()));
 
-    __ bind(L_skip_barrier);
-    entry_address[AdapterBlob::C2I_No_Clinit_Check] = __ pc();
-  }
+  __ bind(L_skip_barrier);
+  entry_address[AdapterBlob::C2I_No_Clinit_Check] = __ pc();
 
   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
   bs->c2i_entry_barrier(masm);
@@ -1443,7 +1441,8 @@ nmethod* SharedRuntime::generate_native_wrapper(MacroAssembler* masm,
     __ nop();  // 4 bytes
   }
 
-  if (VM_Version::supports_fast_class_init_checks() && method->needs_clinit_barrier()) {
+  if (method->needs_clinit_barrier()) {
+    assert(VM_Version::supports_fast_class_init_checks(), "sanity");
     Label L_skip_barrier;
     __ mov_metadata(t1, method->method_holder()); // InstanceKlass*
     __ clinit_barrier(t1, t0, &L_skip_barrier);

src/hotspot/cpu/riscv/templateTable_riscv.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2003, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2003, 2026, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2014, Red Hat Inc. All rights reserved.
  * Copyright (c) 2020, 2023, Huawei Technologies Co., Ltd. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
@@ -2192,7 +2192,8 @@ void TemplateTable::resolve_cache_and_index_for_method(int byte_no,
   __ mv(t0, (int) code);
 
   // Class initialization barrier for static methods
-  if (VM_Version::supports_fast_class_init_checks() && bytecode() == Bytecodes::_invokestatic) {
+  if (bytecode() == Bytecodes::_invokestatic) {
+    assert(VM_Version::supports_fast_class_init_checks(), "sanity");
     __ bne(temp, t0, L_clinit_barrier_slow);  // have we resolved this bytecode?
     __ ld(temp, Address(Rcache, in_bytes(ResolvedMethodEntry::method_offset())));
     __ load_method_holder(temp, temp);
@@ -2243,8 +2244,8 @@ void TemplateTable::resolve_cache_and_index_for_field(int byte_no,
   __ mv(t0, (int) code);  // have we resolved this bytecode?
 
   // Class initialization barrier for static fields
-  if (VM_Version::supports_fast_class_init_checks() &&
-      (bytecode() == Bytecodes::_getstatic || bytecode() == Bytecodes::_putstatic)) {
+  if (bytecode() == Bytecodes::_getstatic || bytecode() == Bytecodes::_putstatic) {
+    assert(VM_Version::supports_fast_class_init_checks(), "sanity");
     const Register field_holder = temp;
 
     __ bne(temp, t0, L_clinit_barrier_slow);

src/hotspot/cpu/s390/sharedRuntime_s390.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2016, 2026, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2016, 2024 SAP SE. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
@@ -1567,7 +1567,8 @@ nmethod *SharedRuntime::generate_native_wrapper(MacroAssembler *masm,
   //---------------------------------------------------------------------
   wrapper_VEPStart = __ offset();
 
-  if (VM_Version::supports_fast_class_init_checks() && method->needs_clinit_barrier()) {
+  if (method->needs_clinit_barrier()) {
+    assert(VM_Version::supports_fast_class_init_checks(), "sanity");
     Label L_skip_barrier;
     Register klass = Z_R1_scratch;
     // Notify OOP recorder (don't need the relocation)
@@ -2378,24 +2379,22 @@ void SharedRuntime::generate_i2c2i_adapters(MacroAssembler *masm,
 
   // Class initialization barrier for static methods
   entry_address[AdapterBlob::C2I_No_Clinit_Check] = nullptr;
-  if (VM_Version::supports_fast_class_init_checks()) {
-    Label L_skip_barrier;
+  assert(VM_Version::supports_fast_class_init_checks(), "sanity");
+  Label L_skip_barrier;
 
-    { // Bypass the barrier for non-static methods
-      __ testbit_ushort(Address(Z_method, Method::access_flags_offset()), JVM_ACC_STATIC_BIT);
-      __ z_bfalse(L_skip_barrier); // non-static
-    }
+  // Bypass the barrier for non-static methods
+  __ testbit_ushort(Address(Z_method, Method::access_flags_offset()), JVM_ACC_STATIC_BIT);
+  __ z_bfalse(L_skip_barrier); // non-static
 
-    Register klass = Z_R11;
-    __ load_method_holder(klass, Z_method);
-    __ clinit_barrier(klass, Z_thread, &L_skip_barrier /*L_fast_path*/);
+  Register klass = Z_R11;
+  __ load_method_holder(klass, Z_method);
+  __ clinit_barrier(klass, Z_thread, &L_skip_barrier /*L_fast_path*/);
 
-    __ load_const_optimized(klass, SharedRuntime::get_handle_wrong_method_stub());
-    __ z_br(klass);
+  __ load_const_optimized(klass, SharedRuntime::get_handle_wrong_method_stub());
+  __ z_br(klass);
 
-    __ bind(L_skip_barrier);
-    entry_address[AdapterBlob::C2I_No_Clinit_Check] = __ pc();
-  }
+  __ bind(L_skip_barrier);
+  entry_address[AdapterBlob::C2I_No_Clinit_Check] = __ pc();
 
   gen_c2i_adapter(masm, total_args_passed, comp_args_on_stack, sig_bt, regs, skip_fixup);
   return;

src/hotspot/cpu/s390/templateTable_s390.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2016, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2016, 2026, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2016, 2024 SAP SE. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
@@ -2377,7 +2377,8 @@ void TemplateTable::resolve_cache_and_index_for_method(int byte_no,
   __ z_cli(Address(Rcache, bc_offset), code);
 
   // Class initialization barrier for static methods
-  if (VM_Version::supports_fast_class_init_checks() && bytecode() == Bytecodes::_invokestatic) {
+  if (bytecode() == Bytecodes::_invokestatic) {
+    assert(VM_Version::supports_fast_class_init_checks(), "sanity");
     const Register method = Z_R1_scratch;
     const Register klass  = Z_R1_scratch;
     __ z_brne(L_clinit_barrier_slow);
@@ -2427,8 +2428,8 @@ void TemplateTable::resolve_cache_and_index_for_field(int byte_no,
   __ z_cli(Address(cache, code_offset), code);
 
   // Class initialization barrier for static fields
-  if (VM_Version::supports_fast_class_init_checks() &&
-      (bytecode() == Bytecodes::_getstatic || bytecode() == Bytecodes::_putstatic)) {
+  if (bytecode() == Bytecodes::_getstatic || bytecode() == Bytecodes::_putstatic) {
+    assert(VM_Version::supports_fast_class_init_checks(), "sanity");
     const Register field_holder = index;
 
     __ z_brne(L_clinit_barrier_slow);

src/hotspot/cpu/x86/sharedRuntime_x86_64.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2003, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2003, 2026, 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
@@ -1043,26 +1043,24 @@ void SharedRuntime::generate_i2c2i_adapters(MacroAssembler *masm,
 
   // Class initialization barrier for static methods
   entry_address[AdapterBlob::C2I_No_Clinit_Check] = nullptr;
-  if (VM_Version::supports_fast_class_init_checks()) {
-    Label L_skip_barrier;
-    Register method = rbx;
+  assert(VM_Version::supports_fast_class_init_checks(), "sanity");
+  Label L_skip_barrier;
+  Register method = rbx;
 
-    { // Bypass the barrier for non-static methods
-      Register flags = rscratch1;
-      __ load_unsigned_short(flags, Address(method, Method::access_flags_offset()));
-      __ testl(flags, JVM_ACC_STATIC);
-      __ jcc(Assembler::zero, L_skip_barrier); // non-static
-    }
+  // Bypass the barrier for non-static methods
+  Register flags = rscratch1;
+  __ load_unsigned_short(flags, Address(method, Method::access_flags_offset()));
+  __ testl(flags, JVM_ACC_STATIC);
+  __ jcc(Assembler::zero, L_skip_barrier); // non-static
 
-    Register klass = rscratch1;
-    __ load_method_holder(klass, method);
-    __ clinit_barrier(klass, &L_skip_barrier /*L_fast_path*/);
+  Register klass = rscratch1;
+  __ load_method_holder(klass, method);
+  __ clinit_barrier(klass, &L_skip_barrier /*L_fast_path*/);
 
-    __ jump(RuntimeAddress(SharedRuntime::get_handle_wrong_method_stub())); // slow path
+  __ jump(RuntimeAddress(SharedRuntime::get_handle_wrong_method_stub())); // slow path
 
-    __ bind(L_skip_barrier);
-    entry_address[AdapterBlob::C2I_No_Clinit_Check] = __ pc();
-  }
+  __ bind(L_skip_barrier);
+  entry_address[AdapterBlob::C2I_No_Clinit_Check] = __ pc();
 
   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
   bs->c2i_entry_barrier(masm);
@@ -1904,7 +1902,8 @@ nmethod* SharedRuntime::generate_native_wrapper(MacroAssembler* masm,
 
   int vep_offset = ((intptr_t)__ pc()) - start;
 
-  if (VM_Version::supports_fast_class_init_checks() && method->needs_clinit_barrier()) {
+  if (method->needs_clinit_barrier()) {
+    assert(VM_Version::supports_fast_class_init_checks(), "sanity");
     Label L_skip_barrier;
     Register klass = r10;
     __ mov_metadata(klass, method->method_holder()); // InstanceKlass*
@@ -3602,4 +3601,3 @@ RuntimeStub* SharedRuntime::generate_jfr_return_lease() {
 }
 
 #endif // INCLUDE_JFR
-

src/hotspot/cpu/x86/stubGenerator_x86_64_kyber.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2025, 2026, 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
@@ -64,6 +64,39 @@ static address kyberAvx512ConstsAddr(int offset) {
 
 const Register scratch = r10;
 
+ATTRIBUTE_ALIGNED(64) static const uint8_t kyberAvx512_12To16Dup[] = {
+// 0 - 63
+    0, 1, 1, 2, 3, 4, 4, 5, 6, 7, 7, 8, 9, 10, 10, 11, 12, 13, 13, 14, 15, 16,
+    16, 17, 18, 19, 19, 20, 21, 22, 22, 23, 24, 25, 25, 26, 27, 28, 28, 29, 30,
+    31, 31, 32, 33, 34, 34, 35, 36, 37, 37, 38, 39, 40, 40, 41, 42, 43, 43, 44,
+    45, 46, 46, 47
+  };
+
+static address kyberAvx512_12To16DupAddr() {
+  return (address) kyberAvx512_12To16Dup;
+}
+
+ATTRIBUTE_ALIGNED(64) static const uint16_t kyberAvx512_12To16Shift[] = {
+// 0 - 31
+    0, 4, 0, 4, 0, 4, 0, 4, 0, 4, 0, 4, 0, 4, 0, 4, 0, 4, 0, 4, 0, 4, 0, 4, 0,
+    4, 0, 4, 0, 4, 0, 4
+  };
+
+static address kyberAvx512_12To16ShiftAddr() {
+  return (address) kyberAvx512_12To16Shift;
+}
+
+ATTRIBUTE_ALIGNED(64) static const uint64_t kyberAvx512_12To16And[] = {
+// 0 - 7
+    0x0FFF0FFF0FFF0FFF, 0x0FFF0FFF0FFF0FFF, 0x0FFF0FFF0FFF0FFF,
+    0x0FFF0FFF0FFF0FFF, 0x0FFF0FFF0FFF0FFF, 0x0FFF0FFF0FFF0FFF,
+    0x0FFF0FFF0FFF0FFF, 0x0FFF0FFF0FFF0FFF
+  };
+
+static address kyberAvx512_12To16AndAddr() {
+  return (address) kyberAvx512_12To16And;
+}
+
 ATTRIBUTE_ALIGNED(64) static const uint16_t kyberAvx512NttPerms[] = {
 // 0
     0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17,
@@ -822,10 +855,65 @@ address generate_kyber12To16_avx512(StubGenerator *stubgen,
 
   const Register perms = r11;
 
-  Label Loop;
+  Label Loop, VBMILoop;
 
   __ addptr(condensed, condensedOffs);
 
+  if (VM_Version::supports_avx512_vbmi()) {
+    // mask load for the first 48 bytes of each vector
+    __ mov64(rax, 0x0000FFFFFFFFFFFF);
+    __ kmovql(k1, rax);
+
+    __ lea(perms, ExternalAddress(kyberAvx512_12To16DupAddr()));
+    __ evmovdqub(xmm20, Address(perms), Assembler::AVX_512bit);
+
+    __ lea(perms, ExternalAddress(kyberAvx512_12To16ShiftAddr()));
+    __ evmovdquw(xmm21, Address(perms), Assembler::AVX_512bit);
+
+    __ lea(perms, ExternalAddress(kyberAvx512_12To16AndAddr()));
+    __ evmovdquq(xmm22, Address(perms), Assembler::AVX_512bit);
+
+    __ align(OptoLoopAlignment);
+    __ BIND(VBMILoop);
+
+      __ evmovdqub(xmm0, k1, Address(condensed, 0), false,
+                   Assembler::AVX_512bit);
+      __ evmovdqub(xmm1, k1, Address(condensed, 48), false,
+                   Assembler::AVX_512bit);
+      __ evmovdqub(xmm2, k1, Address(condensed, 96), false,
+                   Assembler::AVX_512bit);
+      __ evmovdqub(xmm3, k1, Address(condensed, 144), false,
+                   Assembler::AVX_512bit);
+
+      __ evpermb(xmm4, k0, xmm20, xmm0, false, Assembler::AVX_512bit);
+      __ evpermb(xmm5, k0, xmm20, xmm1, false, Assembler::AVX_512bit);
+      __ evpermb(xmm6, k0, xmm20, xmm2, false, Assembler::AVX_512bit);
+      __ evpermb(xmm7, k0, xmm20, xmm3, false, Assembler::AVX_512bit);
+
+      __ evpsrlvw(xmm4, xmm4, xmm21, Assembler::AVX_512bit);
+      __ evpsrlvw(xmm5, xmm5, xmm21, Assembler::AVX_512bit);
+      __ evpsrlvw(xmm6, xmm6, xmm21, Assembler::AVX_512bit);
+      __ evpsrlvw(xmm7, xmm7, xmm21, Assembler::AVX_512bit);
+
+      __ evpandq(xmm0, xmm22, xmm4, Assembler::AVX_512bit);
+      __ evpandq(xmm1, xmm22, xmm5, Assembler::AVX_512bit);
+      __ evpandq(xmm2, xmm22, xmm6, Assembler::AVX_512bit);
+      __ evpandq(xmm3, xmm22, xmm7, Assembler::AVX_512bit);
+
+      store4regs(parsed, 0, xmm0_3, _masm);
+
+      __ addptr(condensed, 192);
+      __ addptr(parsed, 256);
+      __ subl(parsedLength, 128);
+      __ jcc(Assembler::greater, VBMILoop);
+
+    __ leave(); // required for proper stackwalking of RuntimeStub frame
+    __ mov64(rax, 0); // return 0
+    __ ret(0);
+
+    return start;
+  }
+
   __ lea(perms, ExternalAddress(kyberAvx512_12To16PermsAddr()));
 
   load4regs(xmm24_27, perms, 0, _masm);

src/hotspot/cpu/x86/templateTable_x86.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1997, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 2026, 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
@@ -2216,7 +2216,8 @@ void TemplateTable::resolve_cache_and_index_for_method(int byte_no,
   __ cmpl(temp, code);  // have we resolved this bytecode?
 
   // Class initialization barrier for static methods
-  if (VM_Version::supports_fast_class_init_checks() && bytecode() == Bytecodes::_invokestatic) {
+  if (bytecode() == Bytecodes::_invokestatic) {
+    assert(VM_Version::supports_fast_class_init_checks(), "sanity");
     const Register method = temp;
     const Register klass  = temp;
 
@@ -2264,8 +2265,8 @@ void TemplateTable::resolve_cache_and_index_for_field(int byte_no,
   __ cmpl(temp, code);  // have we resolved this bytecode?
 
   // Class initialization barrier for static fields
-  if (VM_Version::supports_fast_class_init_checks() &&
-      (bytecode() == Bytecodes::_getstatic || bytecode() == Bytecodes::_putstatic)) {
+  if (bytecode() == Bytecodes::_getstatic || bytecode() == Bytecodes::_putstatic) {
+    assert(VM_Version::supports_fast_class_init_checks(), "sanity");
     const Register field_holder = temp;
 
     __ jcc(Assembler::notEqual, L_clinit_barrier_slow);

src/hotspot/os/aix/os_aix.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1999, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1999, 2026, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2012, 2025 SAP SE. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
@@ -258,10 +258,18 @@ bool os::free_memory(physical_memory_size_type& value) {
   return Aix::available_memory(value);
 }
 
+bool os::Machine::free_memory(physical_memory_size_type& value) {
+  return Aix::available_memory(value);
+}
+
 bool os::available_memory(physical_memory_size_type& value) {
   return Aix::available_memory(value);
 }
 
+bool os::Machine::available_memory(physical_memory_size_type& value) {
+  return Aix::available_memory(value);
+}
+
 bool os::Aix::available_memory(physical_memory_size_type& value) {
   os::Aix::meminfo_t mi;
   if (os::Aix::get_meminfo(&mi)) {
@@ -273,6 +281,10 @@ bool os::Aix::available_memory(physical_memory_size_type& value) {
 }
 
 bool os::total_swap_space(physical_memory_size_type& value) {
+  return Machine::total_swap_space(value);
+}
+
+bool os::Machine::total_swap_space(physical_memory_size_type& value) {
   perfstat_memory_total_t memory_info;
   if (libperfstat::perfstat_memory_total(nullptr, &memory_info, sizeof(perfstat_memory_total_t), 1) == -1) {
     return false;
@@ -282,6 +294,10 @@ bool os::total_swap_space(physical_memory_size_type& value) {
 }
 
 bool os::free_swap_space(physical_memory_size_type& value) {
+  return Machine::free_swap_space(value);
+}
+
+bool os::Machine::free_swap_space(physical_memory_size_type& value) {
   perfstat_memory_total_t memory_info;
   if (libperfstat::perfstat_memory_total(nullptr, &memory_info, sizeof(perfstat_memory_total_t), 1) == -1) {
     return false;
@@ -294,6 +310,10 @@ physical_memory_size_type os::physical_memory() {
   return Aix::physical_memory();
 }
 
+physical_memory_size_type os::Machine::physical_memory() {
+  return Aix::physical_memory();
+}
+
 size_t os::rss() { return (size_t)0; }
 
 // Cpu architecture string
@@ -2264,6 +2284,10 @@ int os::active_processor_count() {
     return ActiveProcessorCount;
   }
 
+  return Machine::active_processor_count();
+}
+
+int os::Machine::active_processor_count() {
   int online_cpus = ::sysconf(_SC_NPROCESSORS_ONLN);
   assert(online_cpus > 0 && online_cpus <= processor_count(), "sanity check");
   return online_cpus;

src/hotspot/os/bsd/memMapPrinter_macosx.cpp

@@ -132,7 +132,7 @@ public:
   static const char* tagToStr(uint32_t user_tag) {
     switch (user_tag) {
       case 0:
-        return 0;
+        return nullptr;
       X1(MALLOC, malloc);
       X1(MALLOC_SMALL, malloc_small);
       X1(MALLOC_LARGE, malloc_large);

src/hotspot/os/bsd/os_bsd.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1999, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1999, 2026, 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
@@ -137,10 +137,18 @@ bool os::available_memory(physical_memory_size_type& value) {
   return Bsd::available_memory(value);
 }
 
+bool os::Machine::available_memory(physical_memory_size_type& value) {
+  return Bsd::available_memory(value);
+}
+
 bool os::free_memory(physical_memory_size_type& value) {
   return Bsd::available_memory(value);
 }
 
+bool os::Machine::free_memory(physical_memory_size_type& value) {
+  return Bsd::available_memory(value);
+}
+
 // Available here means free. Note that this number is of no much use. As an estimate
 // for future memory pressure it is far too conservative, since MacOS will use a lot
 // of unused memory for caches, and return it willingly in case of needs.
@@ -181,6 +189,10 @@ void os::Bsd::print_uptime_info(outputStream* st) {
 }
 
 bool os::total_swap_space(physical_memory_size_type& value) {
+  return Machine::total_swap_space(value);
+}
+
+bool os::Machine::total_swap_space(physical_memory_size_type& value) {
 #if defined(__APPLE__)
   struct xsw_usage vmusage;
   size_t size = sizeof(vmusage);
@@ -195,6 +207,10 @@ bool os::total_swap_space(physical_memory_size_type& value) {
 }
 
 bool os::free_swap_space(physical_memory_size_type& value) {
+  return Machine::free_swap_space(value);
+}
+
+bool os::Machine::free_swap_space(physical_memory_size_type& value) {
 #if defined(__APPLE__)
   struct xsw_usage vmusage;
   size_t size = sizeof(vmusage);
@@ -212,6 +228,10 @@ physical_memory_size_type os::physical_memory() {
   return Bsd::physical_memory();
 }
 
+physical_memory_size_type os::Machine::physical_memory() {
+  return Bsd::physical_memory();
+}
+
 size_t os::rss() {
   size_t rss = 0;
 #ifdef __APPLE__
@@ -608,7 +628,7 @@ static void *thread_native_entry(Thread *thread) {
   log_info(os, thread)("Thread finished (tid: %zu, pthread id: %zu).",
     os::current_thread_id(), (uintx) pthread_self());
 
-  return 0;
+  return nullptr;
 }
 
 bool os::create_thread(Thread* thread, ThreadType thr_type,
@@ -1400,7 +1420,7 @@ int os::get_loaded_modules_info(os::LoadedModulesCallbackFunc callback, void *pa
 #elif defined(__APPLE__)
   for (uint32_t i = 1; i < _dyld_image_count(); i++) {
     // Value for top_address is returned as 0 since we don't have any information about module size
-    if (callback(_dyld_get_image_name(i), (address)_dyld_get_image_header(i), (address)0, param)) {
+    if (callback(_dyld_get_image_name(i), (address)_dyld_get_image_header(i), nullptr, param)) {
       return 1;
     }
   }
@@ -2189,6 +2209,10 @@ int os::active_processor_count() {
     return ActiveProcessorCount;
   }
 
+  return Machine::active_processor_count();
+}
+
+int os::Machine::active_processor_count() {
   return _processor_count;
 }
 

src/hotspot/os/linux/cgroupSubsystem_linux.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2019, 2026, 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
@@ -631,22 +631,20 @@ void CgroupSubsystemFactory::cleanup(CgroupInfo* cg_infos) {
  * return:
  *    true if there were no errors. false otherwise.
  */
-bool CgroupSubsystem::active_processor_count(int& value) {
-  int cpu_count;
-  int result = -1;
-
+bool CgroupSubsystem::active_processor_count(double& value) {
   // We use a cache with a timeout to avoid performing expensive
   // computations in the event this function is called frequently.
   // [See 8227006].
-  CachingCgroupController<CgroupCpuController>* contrl = cpu_controller();
-  CachedMetric* cpu_limit = contrl->metrics_cache();
+  CachingCgroupController<CgroupCpuController, double>* contrl = cpu_controller();
+  CachedMetric<double>* cpu_limit = contrl->metrics_cache();
   if (!cpu_limit->should_check_metric()) {
-    value = (int)cpu_limit->value();
-    log_trace(os, container)("CgroupSubsystem::active_processor_count (cached): %d", value);
+    value = cpu_limit->value();
+    log_trace(os, container)("CgroupSubsystem::active_processor_count (cached): %.2f", value);
     return true;
   }
 
-  cpu_count = os::Linux::active_processor_count();
+  int cpu_count = os::Linux::active_processor_count();
+  double result = -1;
   if (!CgroupUtil::processor_count(contrl->controller(), cpu_count, result)) {
     return false;
   }
@@ -671,8 +669,8 @@ bool CgroupSubsystem::active_processor_count(int& value) {
  */
 bool CgroupSubsystem::memory_limit_in_bytes(physical_memory_size_type upper_bound,
                                             physical_memory_size_type& value) {
-  CachingCgroupController<CgroupMemoryController>* contrl = memory_controller();
-  CachedMetric* memory_limit = contrl->metrics_cache();
+  CachingCgroupController<CgroupMemoryController, physical_memory_size_type>* contrl = memory_controller();
+  CachedMetric<physical_memory_size_type>* memory_limit = contrl->metrics_cache();
   if (!memory_limit->should_check_metric()) {
     value = memory_limit->value();
     return true;

src/hotspot/os/linux/cgroupSubsystem_linux.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2019, 2026, 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
@@ -181,20 +181,21 @@ class CgroupController: public CHeapObj<mtInternal> {
     static bool limit_from_str(char* limit_str, physical_memory_size_type& value);
 };
 
+template <typename MetricType>
 class CachedMetric : public CHeapObj<mtInternal>{
   private:
-    volatile physical_memory_size_type _metric;
+    volatile MetricType _metric;
     volatile jlong _next_check_counter;
   public:
     CachedMetric() {
-      _metric = value_unlimited;
+      _metric = static_cast<MetricType>(value_unlimited);
       _next_check_counter = min_jlong;
     }
     bool should_check_metric() {
       return os::elapsed_counter() > _next_check_counter;
     }
-    physical_memory_size_type value() { return _metric; }
-    void set_value(physical_memory_size_type value, jlong timeout) {
+    MetricType value() { return _metric; }
+    void set_value(MetricType value, jlong timeout) {
       _metric = value;
       // Metric is unlikely to change, but we want to remain
       // responsive to configuration changes. A very short grace time
@@ -205,19 +206,19 @@ class CachedMetric : public CHeapObj<mtInternal>{
     }
 };
 
-template <class T>
+template <class T, typename MetricType>
 class CachingCgroupController : public CHeapObj<mtInternal> {
   private:
     T* _controller;
-    CachedMetric* _metrics_cache;
+    CachedMetric<MetricType>* _metrics_cache;
 
   public:
     CachingCgroupController(T* cont) {
       _controller = cont;
-      _metrics_cache = new CachedMetric();
+      _metrics_cache = new CachedMetric<MetricType>();
     }
 
-    CachedMetric* metrics_cache() { return _metrics_cache; }
+    CachedMetric<MetricType>* metrics_cache() { return _metrics_cache; }
     T* controller() { return _controller; }
 };
 
@@ -277,7 +278,7 @@ class CgroupMemoryController: public CHeapObj<mtInternal> {
 class CgroupSubsystem: public CHeapObj<mtInternal> {
   public:
     bool memory_limit_in_bytes(physical_memory_size_type upper_bound, physical_memory_size_type& value);
-    bool active_processor_count(int& value);
+    bool active_processor_count(double& value);
 
     virtual bool pids_max(uint64_t& value) = 0;
     virtual bool pids_current(uint64_t& value) = 0;
@@ -286,8 +287,8 @@ class CgroupSubsystem: public CHeapObj<mtInternal> {
     virtual char * cpu_cpuset_cpus() = 0;
     virtual char * cpu_cpuset_memory_nodes() = 0;
     virtual const char * container_type() = 0;
-    virtual CachingCgroupController<CgroupMemoryController>* memory_controller() = 0;
-    virtual CachingCgroupController<CgroupCpuController>* cpu_controller() = 0;
+    virtual CachingCgroupController<CgroupMemoryController, physical_memory_size_type>* memory_controller() = 0;
+    virtual CachingCgroupController<CgroupCpuController, double>* cpu_controller() = 0;
     virtual CgroupCpuacctController* cpuacct_controller() = 0;
 
     bool cpu_quota(int& value);

src/hotspot/os/linux/cgroupUtil_linux.cpp

@@ -1,5 +1,6 @@
 /*
  * Copyright (c) 2024, 2025, Red Hat, Inc.
+ * Copyright (c) 2026, 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
@@ -25,9 +26,8 @@
 #include "cgroupUtil_linux.hpp"
 #include "os_linux.hpp"
 
-bool CgroupUtil::processor_count(CgroupCpuController* cpu_ctrl, int upper_bound, int& value) {
+bool CgroupUtil::processor_count(CgroupCpuController* cpu_ctrl, int upper_bound, double& value) {
   assert(upper_bound > 0, "upper bound of cpus must be positive");
-  int limit_count = upper_bound;
   int quota = -1;
   int period = -1;
   if (!cpu_ctrl->cpu_quota(quota)) {
@@ -37,20 +37,15 @@ bool CgroupUtil::processor_count(CgroupCpuController* cpu_ctrl, int upper_bound,
     return false;
   }
   int quota_count = 0;
-  int result = upper_bound;
+  double result = upper_bound;
 
-  if (quota > -1 && period > 0) {
-    quota_count = ceilf((float)quota / (float)period);
-    log_trace(os, container)("CPU Quota count based on quota/period: %d", quota_count);
+  if (quota > 0 && period > 0) { // Use quotas
+    double cpu_quota = static_cast<double>(quota) / period;
+    log_trace(os, container)("CPU Quota based on quota/period: %.2f", cpu_quota);
+    result = MIN2(result, cpu_quota);
   }
 
-  // Use quotas
-  if (quota_count != 0) {
-    limit_count = quota_count;
-  }
-
-  result = MIN2(upper_bound, limit_count);
-  log_trace(os, container)("OSContainer::active_processor_count: %d", result);
+  log_trace(os, container)("OSContainer::active_processor_count: %.2f", result);
   value = result;
   return true;
 }
@@ -73,11 +68,11 @@ physical_memory_size_type CgroupUtil::get_updated_mem_limit(CgroupMemoryControll
 
 // Get an updated cpu limit. The return value is strictly less than or equal to the
 // passed in 'lowest' value.
-int CgroupUtil::get_updated_cpu_limit(CgroupCpuController* cpu,
+double CgroupUtil::get_updated_cpu_limit(CgroupCpuController* cpu,
                                      int lowest,
                                      int upper_bound) {
   assert(lowest > 0 && lowest <= upper_bound, "invariant");
-  int cpu_limit_val = -1;
+  double cpu_limit_val = -1;
   if (CgroupUtil::processor_count(cpu, upper_bound, cpu_limit_val) && cpu_limit_val != upper_bound) {
     assert(cpu_limit_val <= upper_bound, "invariant");
     if (lowest > cpu_limit_val) {
@@ -172,7 +167,7 @@ void CgroupUtil::adjust_controller(CgroupCpuController* cpu) {
   assert(cg_path[0] == '/', "cgroup path must start with '/'");
   int host_cpus = os::Linux::active_processor_count();
   int lowest_limit = host_cpus;
-  int cpus = get_updated_cpu_limit(cpu, lowest_limit, host_cpus);
+  double cpus = get_updated_cpu_limit(cpu, lowest_limit, host_cpus);
   int orig_limit = lowest_limit != host_cpus ? lowest_limit : host_cpus;
   char* limit_cg_path = nullptr;
   while ((last_slash = strrchr(cg_path, '/')) != cg_path) {

src/hotspot/os/linux/cgroupUtil_linux.hpp

@@ -1,5 +1,6 @@
 /*
  * Copyright (c) 2024, Red Hat, Inc.
+ * Copyright (c) 2026, 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
@@ -31,7 +32,7 @@
 class CgroupUtil: AllStatic {
 
   public:
-    static bool processor_count(CgroupCpuController* cpu, int upper_bound, int& value);
+    static bool processor_count(CgroupCpuController* cpu, int upper_bound, double& value);
     // Given a memory controller, adjust its path to a point in the hierarchy
     // that represents the closest memory limit.
     static void adjust_controller(CgroupMemoryController* m);
@@ -42,9 +43,7 @@ class CgroupUtil: AllStatic {
     static physical_memory_size_type get_updated_mem_limit(CgroupMemoryController* m,
                                                            physical_memory_size_type lowest,
                                                            physical_memory_size_type upper_bound);
-    static int get_updated_cpu_limit(CgroupCpuController* c,
-                                     int lowest,
-                                     int upper_bound);
+    static double get_updated_cpu_limit(CgroupCpuController* c, int lowest, int upper_bound);
 };
 
 #endif // CGROUP_UTIL_LINUX_HPP

src/hotspot/os/linux/cgroupV1Subsystem_linux.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2019, 2026, 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
@@ -328,8 +328,8 @@ CgroupV1Subsystem::CgroupV1Subsystem(CgroupV1Controller* cpuset,
     _pids(pids) {
   CgroupUtil::adjust_controller(memory);
   CgroupUtil::adjust_controller(cpu);
-  _memory = new CachingCgroupController<CgroupMemoryController>(memory);
-  _cpu = new CachingCgroupController<CgroupCpuController>(cpu);
+  _memory = new CachingCgroupController<CgroupMemoryController, physical_memory_size_type>(memory);
+  _cpu = new CachingCgroupController<CgroupCpuController, double>(cpu);
 }
 
 bool CgroupV1Subsystem::is_containerized() {

src/hotspot/os/linux/cgroupV1Subsystem_linux.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2019, 2026, 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
@@ -214,15 +214,15 @@ class CgroupV1Subsystem: public CgroupSubsystem {
     const char * container_type() override {
       return "cgroupv1";
     }
-    CachingCgroupController<CgroupMemoryController>* memory_controller() override { return _memory; }
-    CachingCgroupController<CgroupCpuController>* cpu_controller() override { return _cpu; }
+    CachingCgroupController<CgroupMemoryController, physical_memory_size_type>* memory_controller() override { return _memory; }
+    CachingCgroupController<CgroupCpuController, double>* cpu_controller() override { return _cpu; }
     CgroupCpuacctController* cpuacct_controller() override { return _cpuacct; }
 
   private:
     /* controllers */
-    CachingCgroupController<CgroupMemoryController>* _memory = nullptr;
+    CachingCgroupController<CgroupMemoryController, physical_memory_size_type>* _memory = nullptr;
     CgroupV1Controller* _cpuset = nullptr;
-    CachingCgroupController<CgroupCpuController>* _cpu = nullptr;
+    CachingCgroupController<CgroupCpuController, double>* _cpu = nullptr;
     CgroupV1CpuacctController* _cpuacct = nullptr;
     CgroupV1Controller* _pids = nullptr;
 

src/hotspot/os/linux/cgroupV2Subsystem_linux.cpp

@@ -1,6 +1,6 @@
 /*
  * Copyright (c) 2020, 2025, Red Hat Inc.
- * Copyright (c) 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2025, 2026, 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
@@ -156,8 +156,8 @@ CgroupV2Subsystem::CgroupV2Subsystem(CgroupV2MemoryController * memory,
                                      _unified(unified) {
   CgroupUtil::adjust_controller(memory);
   CgroupUtil::adjust_controller(cpu);
-  _memory = new CachingCgroupController<CgroupMemoryController>(memory);
-  _cpu = new CachingCgroupController<CgroupCpuController>(cpu);
+  _memory = new CachingCgroupController<CgroupMemoryController, physical_memory_size_type>(memory);
+  _cpu = new CachingCgroupController<CgroupCpuController, double>(cpu);
   _cpuacct = cpuacct;
 }
 

src/hotspot/os/linux/cgroupV2Subsystem_linux.hpp

@@ -1,6 +1,6 @@
 /*
  * Copyright (c) 2020, 2024, Red Hat Inc.
- * Copyright (c) 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2025, 2026, 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
@@ -152,8 +152,8 @@ class CgroupV2Subsystem: public CgroupSubsystem {
     /* One unified controller */
     CgroupV2Controller _unified;
     /* Caching wrappers for cpu/memory metrics */
-    CachingCgroupController<CgroupMemoryController>* _memory = nullptr;
-    CachingCgroupController<CgroupCpuController>* _cpu = nullptr;
+    CachingCgroupController<CgroupMemoryController, physical_memory_size_type>* _memory = nullptr;
+    CachingCgroupController<CgroupCpuController, double>* _cpu = nullptr;
 
     CgroupCpuacctController* _cpuacct = nullptr;
 
@@ -175,8 +175,8 @@ class CgroupV2Subsystem: public CgroupSubsystem {
     const char * container_type() override {
       return "cgroupv2";
     }
-    CachingCgroupController<CgroupMemoryController>* memory_controller() override { return _memory; }
-    CachingCgroupController<CgroupCpuController>* cpu_controller() override { return _cpu; }
+    CachingCgroupController<CgroupMemoryController, physical_memory_size_type>* memory_controller() override { return _memory; }
+    CachingCgroupController<CgroupCpuController, double>* cpu_controller() override { return _cpu; }
     CgroupCpuacctController* cpuacct_controller() override { return _cpuacct; };
 };
 

src/hotspot/os/linux/osContainer_linux.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2017, 2026, 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
@@ -86,8 +86,8 @@ void OSContainer::init() {
     //  2.) On a physical Linux system with a limit enforced by other means (like systemd slice)
     physical_memory_size_type mem_limit_val = value_unlimited;
     (void)memory_limit_in_bytes(mem_limit_val);  // discard error and use default
-    int host_cpus = os::Linux::active_processor_count();
-    int cpus = host_cpus;
+    double host_cpus = os::Linux::active_processor_count();
+    double cpus = host_cpus;
     (void)active_processor_count(cpus);  // discard error and use default
     any_mem_cpu_limit_present = mem_limit_val != value_unlimited || host_cpus != cpus;
     if (any_mem_cpu_limit_present) {
@@ -127,8 +127,7 @@ bool OSContainer::available_memory_in_bytes(physical_memory_size_type& value) {
   return false;
 }
 
-bool OSContainer::available_swap_in_bytes(physical_memory_size_type host_free_swap,
-                                          physical_memory_size_type& value) {
+bool OSContainer::available_swap_in_bytes(physical_memory_size_type& value) {
   physical_memory_size_type mem_limit = 0;
   physical_memory_size_type mem_swap_limit = 0;
   if (memory_limit_in_bytes(mem_limit) &&
@@ -179,8 +178,7 @@ bool OSContainer::available_swap_in_bytes(physical_memory_size_type host_free_sw
     assert(num < 25, "buffer too small");
     mem_limit_buf[num] = '\0';
     log_trace(os,container)("OSContainer::available_swap_in_bytes: container_swap_limit=%s"
-                            " container_mem_limit=%s, host_free_swap: " PHYS_MEM_TYPE_FORMAT,
-                            mem_swap_buf, mem_limit_buf, host_free_swap);
+                            " container_mem_limit=%s", mem_swap_buf, mem_limit_buf);
   }
   return false;
 }
@@ -252,7 +250,7 @@ char * OSContainer::cpu_cpuset_memory_nodes() {
   return cgroup_subsystem->cpu_cpuset_memory_nodes();
 }
 
-bool OSContainer::active_processor_count(int& value) {
+bool OSContainer::active_processor_count(double& value) {
   assert(cgroup_subsystem != nullptr, "cgroup subsystem not available");
   return cgroup_subsystem->active_processor_count(value);
 }
@@ -291,11 +289,13 @@ template<typename T> struct metric_fmt;
 template<> struct metric_fmt<unsigned long long int> { static constexpr const char* fmt = "%llu"; };
 template<> struct metric_fmt<unsigned long int> { static constexpr const char* fmt = "%lu"; };
 template<> struct metric_fmt<int> { static constexpr const char* fmt = "%d"; };
+template<> struct metric_fmt<double> { static constexpr const char* fmt = "%.2f"; };
 template<> struct metric_fmt<const char*> { static constexpr const char* fmt = "%s"; };
 
 template void OSContainer::print_container_metric<unsigned long long int>(outputStream*, const char*, unsigned long long int, const char*);
 template void OSContainer::print_container_metric<unsigned long int>(outputStream*, const char*, unsigned long int, const char*);
 template void OSContainer::print_container_metric<int>(outputStream*, const char*, int, const char*);
+template void OSContainer::print_container_metric<double>(outputStream*, const char*, double, const char*);
 template void OSContainer::print_container_metric<const char*>(outputStream*, const char*, const char*, const char*);
 
 template <typename T>
@@ -304,12 +304,13 @@ void OSContainer::print_container_metric(outputStream* st, const char* metrics,
   constexpr int longest_value = max_length - 11; // Max length - shortest "metric: " string ("cpu_quota: ")
   char value_str[longest_value + 1] = {};
   os::snprintf_checked(value_str, longest_value, metric_fmt<T>::fmt, value);
-  st->print("%s: %*s", metrics, max_length - static_cast<int>(strlen(metrics)) - 2, value_str); // -2 for the ": "
-  if (unit[0] != '\0') {
-    st->print_cr(" %s", unit);
-  } else {
-    st->print_cr("");
-  }
+
+  const int pad_width = max_length - static_cast<int>(strlen(metrics)) - 2; // -2 for the ": "
+  const char* unit_prefix = unit[0] != '\0' ? " " : "";
+
+  char line[128] = {};
+  os::snprintf_checked(line, sizeof(line), "%s: %*s%s%s", metrics, pad_width, value_str, unit_prefix, unit);
+  st->print_cr("%s", line);
 }
 
 void OSContainer::print_container_helper(outputStream* st, MetricResult& res, const char* metrics) {

src/hotspot/os/linux/osContainer_linux.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2017, 2026, 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
@@ -72,8 +72,7 @@ class OSContainer: AllStatic {
   static const char * container_type();
 
   static bool available_memory_in_bytes(physical_memory_size_type& value);
-  static bool available_swap_in_bytes(physical_memory_size_type host_free_swap,
-                                      physical_memory_size_type& value);
+  static bool available_swap_in_bytes(physical_memory_size_type& value);
   static bool memory_limit_in_bytes(physical_memory_size_type& value);
   static bool memory_and_swap_limit_in_bytes(physical_memory_size_type& value);
   static bool memory_and_swap_usage_in_bytes(physical_memory_size_type& value);
@@ -84,7 +83,7 @@ class OSContainer: AllStatic {
   static bool rss_usage_in_bytes(physical_memory_size_type& value);
   static bool cache_usage_in_bytes(physical_memory_size_type& value);
 
-  static bool active_processor_count(int& value);
+  static bool active_processor_count(double& value);
 
   static char * cpu_cpuset_cpus();
   static char * cpu_cpuset_memory_nodes();

src/hotspot/os/linux/os_linux.cpp

@@ -211,15 +211,58 @@ static bool suppress_primordial_thread_resolution = false;
 
 // utility functions
 
+bool os::is_containerized() {
+  return OSContainer::is_containerized();
+}
+
+bool os::Container::memory_limit(physical_memory_size_type& value) {
+  physical_memory_size_type result = 0;
+  if (OSContainer::memory_limit_in_bytes(result) && result != value_unlimited) {
+    value = result;
+    return true;
+  }
+  return false;
+}
+
+bool os::Container::memory_soft_limit(physical_memory_size_type& value) {
+  physical_memory_size_type result = 0;
+  if (OSContainer::memory_soft_limit_in_bytes(result) && result != 0 && result != value_unlimited) {
+    value = result;
+    return true;
+  }
+  return false;
+}
+
+bool os::Container::memory_throttle_limit(physical_memory_size_type& value) {
+  physical_memory_size_type result = 0;
+  if (OSContainer::memory_throttle_limit_in_bytes(result) && result != value_unlimited) {
+    value = result;
+    return true;
+  }
+  return false;
+}
+
+bool os::Container::used_memory(physical_memory_size_type& value) {
+  return OSContainer::memory_usage_in_bytes(value);
+}
+
 bool os::available_memory(physical_memory_size_type& value) {
-  if (OSContainer::is_containerized() && OSContainer::available_memory_in_bytes(value)) {
+  if (is_containerized() && Container::available_memory(value)) {
     log_trace(os)("available container memory: " PHYS_MEM_TYPE_FORMAT, value);
     return true;
   }
 
+  return Machine::available_memory(value);
+}
+
+bool os::Machine::available_memory(physical_memory_size_type& value) {
   return Linux::available_memory(value);
 }
 
+bool os::Container::available_memory(physical_memory_size_type& value) {
+  return OSContainer::available_memory_in_bytes(value);
+}
+
 bool os::Linux::available_memory(physical_memory_size_type& value) {
   physical_memory_size_type avail_mem = 0;
 
@@ -251,11 +294,15 @@ bool os::Linux::available_memory(physical_memory_size_type& value) {
 }
 
 bool os::free_memory(physical_memory_size_type& value) {
-  if (OSContainer::is_containerized() && OSContainer::available_memory_in_bytes(value)) {
+  if (is_containerized() && Container::available_memory(value)) {
     log_trace(os)("free container memory: " PHYS_MEM_TYPE_FORMAT, value);
     return true;
   }
 
+  return Machine::free_memory(value);
+}
+
+bool os::Machine::free_memory(physical_memory_size_type& value) {
   return Linux::free_memory(value);
 }
 
@@ -274,21 +321,30 @@ bool os::Linux::free_memory(physical_memory_size_type& value) {
 }
 
 bool os::total_swap_space(physical_memory_size_type& value) {
-  if (OSContainer::is_containerized()) {
-    physical_memory_size_type mem_swap_limit = value_unlimited;
-    physical_memory_size_type memory_limit = value_unlimited;
-    if (OSContainer::memory_and_swap_limit_in_bytes(mem_swap_limit) &&
-        OSContainer::memory_limit_in_bytes(memory_limit)) {
-      if (memory_limit != value_unlimited && mem_swap_limit != value_unlimited &&
-          mem_swap_limit >= memory_limit /* ensure swap is >= 0 */) {
-        value = mem_swap_limit - memory_limit;
-        return true;
-      }
-    }
-  } // fallback to the host swap space if the container returned unlimited
+  if (is_containerized() && Container::total_swap_space(value)) {
+    return true;
+  } // fallback to the host swap space if the container value fails
+  return Machine::total_swap_space(value);
+}
+
+bool os::Machine::total_swap_space(physical_memory_size_type& value) {
   return Linux::host_swap(value);
 }
 
+bool os::Container::total_swap_space(physical_memory_size_type& value) {
+  physical_memory_size_type mem_swap_limit = value_unlimited;
+  physical_memory_size_type memory_limit = value_unlimited;
+  if (OSContainer::memory_and_swap_limit_in_bytes(mem_swap_limit) &&
+      OSContainer::memory_limit_in_bytes(memory_limit)) {
+    if (memory_limit != value_unlimited && mem_swap_limit != value_unlimited &&
+        mem_swap_limit >= memory_limit /* ensure swap is >= 0 */) {
+      value = mem_swap_limit - memory_limit;
+      return true;
+    }
+  }
+  return false;
+}
+
 static bool host_free_swap_f(physical_memory_size_type& value) {
   struct sysinfo si;
   int ret = sysinfo(&si);
@@ -309,32 +365,45 @@ bool os::free_swap_space(physical_memory_size_type& value) {
     return false;
   }
   physical_memory_size_type host_free_swap_val = MIN2(total_swap_space, host_free_swap);
-  if (OSContainer::is_containerized()) {
-    if (OSContainer::available_swap_in_bytes(host_free_swap_val, value)) {
+  if (is_containerized()) {
+    if (Container::free_swap_space(value)) {
       return true;
     }
     // Fall through to use host value
     log_trace(os,container)("os::free_swap_space: containerized value unavailable"
                             " returning host value: " PHYS_MEM_TYPE_FORMAT, host_free_swap_val);
   }
+
   value = host_free_swap_val;
   return true;
 }
 
+bool os::Machine::free_swap_space(physical_memory_size_type& value) {
+  return host_free_swap_f(value);
+}
+
+bool os::Container::free_swap_space(physical_memory_size_type& value) {
+  return OSContainer::available_swap_in_bytes(value);
+}
+
 physical_memory_size_type os::physical_memory() {
-  if (OSContainer::is_containerized()) {
+  if (is_containerized()) {
     physical_memory_size_type mem_limit = value_unlimited;
-    if (OSContainer::memory_limit_in_bytes(mem_limit) && mem_limit != value_unlimited) {
+    if (Container::memory_limit(mem_limit) && mem_limit != value_unlimited) {
       log_trace(os)("total container memory: " PHYS_MEM_TYPE_FORMAT, mem_limit);
       return mem_limit;
     }
   }
 
-  physical_memory_size_type phys_mem = Linux::physical_memory();
+  physical_memory_size_type phys_mem = Machine::physical_memory();
   log_trace(os)("total system memory: " PHYS_MEM_TYPE_FORMAT, phys_mem);
   return phys_mem;
 }
 
+physical_memory_size_type os::Machine::physical_memory() {
+  return Linux::physical_memory();
+}
+
 // Returns the resident set size (RSS) of the process.
 // Falls back to using VmRSS from /proc/self/status if /proc/self/smaps_rollup is unavailable.
 // Note: On kernels with memory cgroups or shared memory, VmRSS may underreport RSS.
@@ -2439,20 +2508,21 @@ bool os::Linux::print_container_info(outputStream* st) {
   OSContainer::print_container_metric(st, "cpu_memory_nodes", p != nullptr ? p : "not supported");
   free(p);
 
-  int i = -1;
-  bool supported = OSContainer::active_processor_count(i);
+  double cpus = -1;
+  bool supported = OSContainer::active_processor_count(cpus);
   if (supported) {
-    assert(i > 0, "must be");
+    assert(cpus > 0, "must be");
     if (ActiveProcessorCount > 0) {
       OSContainer::print_container_metric(st, "active_processor_count", ActiveProcessorCount, "(from -XX:ActiveProcessorCount)");
     } else {
-      OSContainer::print_container_metric(st, "active_processor_count", i);
+      OSContainer::print_container_metric(st, "active_processor_count", cpus);
     }
   } else {
     OSContainer::print_container_metric(st, "active_processor_count", "not supported");
   }
 
 
+  int i = -1;
   supported = OSContainer::cpu_quota(i);
   if (supported && i > 0) {
     OSContainer::print_container_metric(st, "cpu_quota", i);
@@ -4737,15 +4807,26 @@ int os::active_processor_count() {
     return ActiveProcessorCount;
   }
 
-  int active_cpus = -1;
-  if (OSContainer::is_containerized() && OSContainer::active_processor_count(active_cpus)) {
-    log_trace(os)("active_processor_count: determined by OSContainer: %d",
-                   active_cpus);
-  } else {
-    active_cpus = os::Linux::active_processor_count();
+  if (is_containerized()) {
+    double cpu_quota;
+    if (Container::processor_count(cpu_quota)) {
+      int active_cpus = ceilf(cpu_quota); // Round fractional CPU quota up.
+      assert(active_cpus <= Machine::active_processor_count(), "must be");
+      log_trace(os)("active_processor_count: determined by OSContainer: %d",
+                     active_cpus);
+      return active_cpus;
+    }
   }
 
-  return active_cpus;
+  return Machine::active_processor_count();
+}
+
+int os::Machine::active_processor_count() {
+  return os::Linux::active_processor_count();
+}
+
+bool os::Container::processor_count(double& value) {
+  return OSContainer::active_processor_count(value);
 }
 
 static bool should_warn_invalid_processor_id() {
@@ -4882,9 +4963,14 @@ int os::open(const char *path, int oflag, int mode) {
   oflag |= O_CLOEXEC;
 
   int fd = ::open(path, oflag, mode);
-  if (fd == -1) return -1;
+  // No further checking is needed if open() returned an error or
+  // access mode is not read only.
+  if (fd == -1 || (oflag & O_ACCMODE) != O_RDONLY) {
+    return fd;
+  }
 
-  //If the open succeeded, the file might still be a directory
+  // If the open succeeded and is read only, the file might be a directory
+  // which the JVM doesn't allow to be read.
   {
     struct stat buf;
     int ret = ::fstat(fd, &buf);

src/hotspot/os/posix/perfMemory_posix.cpp

@@ -112,6 +112,10 @@ static void save_memory_to_file(char* addr, size_t size) {
     result = ::close(fd);
     if (result == OS_ERR) {
       warning("Could not close %s: %s\n", destfile, os::strerror(errno));
+    } else {
+      if (!successful_write) {
+        remove(destfile);
+      }
     }
   }
   FREE_C_HEAP_ARRAY(char, destfile);
@@ -949,6 +953,7 @@ static int create_sharedmem_file(const char* dirname, const char* filename, size
         warning("Insufficient space for shared memory file: %s/%s\n", dirname, filename);
       }
       result = OS_ERR;
+      remove(filename);
       break;
     }
   }

src/hotspot/os/windows/os_windows.cpp

@@ -839,10 +839,18 @@ bool os::available_memory(physical_memory_size_type& value) {
   return win32::available_memory(value);
 }
 
+bool os::Machine::available_memory(physical_memory_size_type& value) {
+  return win32::available_memory(value);
+}
+
 bool os::free_memory(physical_memory_size_type& value) {
   return win32::available_memory(value);
 }
 
+bool os::Machine::free_memory(physical_memory_size_type& value) {
+  return win32::available_memory(value);
+}
+
 bool os::win32::available_memory(physical_memory_size_type& value) {
   // Use GlobalMemoryStatusEx() because GlobalMemoryStatus() may return incorrect
   // value if total memory is larger than 4GB
@@ -858,7 +866,11 @@ bool os::win32::available_memory(physical_memory_size_type& value) {
   }
 }
 
-bool os::total_swap_space(physical_memory_size_type& value) {
+bool os::total_swap_space(physical_memory_size_type& value)  {
+  return Machine::total_swap_space(value);
+}
+
+bool os::Machine::total_swap_space(physical_memory_size_type& value) {
   MEMORYSTATUSEX ms;
   ms.dwLength = sizeof(ms);
   BOOL res = GlobalMemoryStatusEx(&ms);
@@ -872,6 +884,10 @@ bool os::total_swap_space(physical_memory_size_type& value) {
 }
 
 bool os::free_swap_space(physical_memory_size_type& value) {
+  return Machine::free_swap_space(value);
+}
+
+bool os::Machine::free_swap_space(physical_memory_size_type& value) {
   MEMORYSTATUSEX ms;
   ms.dwLength = sizeof(ms);
   BOOL res = GlobalMemoryStatusEx(&ms);
@@ -888,6 +904,10 @@ physical_memory_size_type os::physical_memory() {
   return win32::physical_memory();
 }
 
+physical_memory_size_type os::Machine::physical_memory() {
+  return win32::physical_memory();
+}
+
 size_t os::rss() {
   size_t rss = 0;
   PROCESS_MEMORY_COUNTERS_EX pmex;
@@ -911,6 +931,10 @@ int os::active_processor_count() {
     return ActiveProcessorCount;
   }
 
+  return Machine::active_processor_count();
+}
+
+int os::Machine::active_processor_count() {
   bool schedules_all_processor_groups = win32::is_windows_11_or_greater() || win32::is_windows_server_2022_or_greater();
   if (UseAllWindowsProcessorGroups && !schedules_all_processor_groups && !win32::processor_group_warning_displayed()) {
     win32::set_processor_group_warning_displayed(true);

src/hotspot/share/cds/archiveBuilder.cpp

@@ -571,7 +571,12 @@ ArchiveBuilder::FollowMode ArchiveBuilder::get_follow_mode(MetaspaceClosure::Ref
       }
       if (is_excluded(klass)) {
         ResourceMark rm;
-        log_debug(cds, dynamic)("Skipping class (excluded): %s", klass->external_name());
+        aot_log_trace(aot)("pointer set to null: class (excluded): %s", klass->external_name());
+        return set_to_null;
+      }
+      if (klass->is_array_klass() && CDSConfig::is_dumping_dynamic_archive()) {
+        ResourceMark rm;
+        aot_log_trace(aot)("pointer set to null: array class not supported in dynamic region: %s", klass->external_name());
         return set_to_null;
       }
     }

src/hotspot/share/ci/ciField.cpp

@@ -216,6 +216,10 @@ ciField::ciField(fieldDescriptor *fd) :
 static bool trust_final_non_static_fields(ciInstanceKlass* holder) {
   if (holder == nullptr)
     return false;
+  if (holder->trust_final_fields()) {
+    // Explicit opt-in from system classes
+    return true;
+  }
   // Even if general trusting is disabled, trust system-built closures in these packages.
   if (holder->is_in_package("java/lang/invoke") || holder->is_in_package("sun/invoke") ||
       holder->is_in_package("java/lang/reflect") || holder->is_in_package("jdk/internal/reflect") ||
@@ -230,14 +234,6 @@ static bool trust_final_non_static_fields(ciInstanceKlass* holder) {
   // Trust final fields in records
   if (holder->is_record())
     return true;
-  // Trust Atomic*FieldUpdaters: they are very important for performance, and make up one
-  // more reason not to use Unsafe, if their final fields are trusted. See more in JDK-8140483.
-  if (holder->name() == ciSymbols::java_util_concurrent_atomic_AtomicIntegerFieldUpdater_Impl() ||
-      holder->name() == ciSymbols::java_util_concurrent_atomic_AtomicLongFieldUpdater_CASUpdater() ||
-      holder->name() == ciSymbols::java_util_concurrent_atomic_AtomicLongFieldUpdater_LockedUpdater() ||
-      holder->name() == ciSymbols::java_util_concurrent_atomic_AtomicReferenceFieldUpdater_Impl()) {
-    return true;
-  }
   return TrustFinalNonStaticFields;
 }
 

src/hotspot/share/ci/ciInstanceKlass.cpp

@@ -65,6 +65,7 @@ ciInstanceKlass::ciInstanceKlass(Klass* k) :
   _has_nonstatic_concrete_methods = ik->has_nonstatic_concrete_methods();
   _is_hidden = ik->is_hidden();
   _is_record = ik->is_record();
+  _trust_final_fields = ik->trust_final_fields();
   _nonstatic_fields = nullptr; // initialized lazily by compute_nonstatic_fields:
   _has_injected_fields = -1;
   _implementor = nullptr; // we will fill these lazily

src/hotspot/share/ci/ciInstanceKlass.hpp

@@ -59,6 +59,7 @@ private:
   bool                   _has_nonstatic_concrete_methods;
   bool                   _is_hidden;
   bool                   _is_record;
+  bool                   _trust_final_fields;
   bool                   _has_trusted_loader;
 
   ciFlags                _flags;
@@ -207,6 +208,10 @@ public:
     return _is_record;
   }
 
+  bool trust_final_fields() const {
+    return _trust_final_fields;
+  }
+
   ciInstanceKlass* get_canonical_holder(int offset);
   ciField* get_field_by_offset(int field_offset, bool is_static);
   ciField* get_field_by_name(ciSymbol* name, ciSymbol* signature, bool is_static);

src/hotspot/share/classfile/classFileParser.cpp

@@ -943,6 +943,7 @@ public:
     _java_lang_Deprecated_for_removal,
     _jdk_internal_vm_annotation_AOTSafeClassInitializer,
     _method_AOTRuntimeSetup,
+    _jdk_internal_vm_annotation_TrustFinalFields,
     _annotation_LIMIT
   };
   const Location _location;
@@ -1878,6 +1879,11 @@ AnnotationCollector::annotation_index(const ClassLoaderData* loader_data,
       if (!privileged)              break;  // only allow in privileged code
       return _field_Stable;
     }
+    case VM_SYMBOL_ENUM_NAME(jdk_internal_vm_annotation_TrustFinalFields_signature): {
+      if (_location != _in_class)   break;  // only allow for classes
+      if (!privileged)              break;  // only allow in privileged code
+      return _jdk_internal_vm_annotation_TrustFinalFields;
+    }
     case VM_SYMBOL_ENUM_NAME(jdk_internal_vm_annotation_Contended_signature): {
       if (_location != _in_field && _location != _in_class) {
         break;  // only allow for fields and classes
@@ -1992,6 +1998,9 @@ void ClassFileParser::ClassAnnotationCollector::apply_to(InstanceKlass* ik) {
   if (has_annotation(_jdk_internal_vm_annotation_AOTSafeClassInitializer)) {
     ik->set_has_aot_safe_initializer();
   }
+  if (has_annotation(_jdk_internal_vm_annotation_TrustFinalFields)) {
+    ik->set_trust_final_fields(true);
+  }
 }
 
 #define MAX_ARGS_SIZE 255

src/hotspot/share/classfile/stringTable.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1997, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 2026, 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
@@ -614,6 +614,10 @@ struct StringTableDeleteCheck : StackObj {
 };
 
 void StringTable::clean_dead_entries(JavaThread* jt) {
+  // BulkDeleteTask::prepare() may take ConcurrentHashTableResize_lock (nosafepoint-2).
+  // When NativeHeapTrimmer is enabled, SuspendMark may take NativeHeapTrimmer::_lock (nosafepoint).
+  // Take SuspendMark first to keep lock order and avoid deadlock.
+  NativeHeapTrimmer::SuspendMark sm("stringtable");
   StringTableHash::BulkDeleteTask bdt(_local_table);
   if (!bdt.prepare(jt)) {
     return;
@@ -621,7 +625,6 @@ void StringTable::clean_dead_entries(JavaThread* jt) {
 
   StringTableDeleteCheck stdc;
   StringTableDoDelete stdd;
-  NativeHeapTrimmer::SuspendMark sm("stringtable");
   {
     TraceTime timer("Clean", TRACETIME_LOG(Debug, stringtable, perf));
     while(bdt.do_task(jt, stdc, stdd)) {

src/hotspot/share/classfile/symbolTable.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1997, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 2026, 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
@@ -763,6 +763,10 @@ struct SymbolTableDeleteCheck : StackObj {
 };
 
 void SymbolTable::clean_dead_entries(JavaThread* jt) {
+  // BulkDeleteTask::prepare() may take ConcurrentHashTableResize_lock (nosafepoint-2).
+  // When NativeHeapTrimmer is enabled, SuspendMark may take NativeHeapTrimmer::_lock (nosafepoint).
+  // Take SuspendMark first to keep lock order and avoid deadlock.
+  NativeHeapTrimmer::SuspendMark sm("symboltable");
   SymbolTableHash::BulkDeleteTask bdt(_local_table);
   if (!bdt.prepare(jt)) {
     return;
@@ -770,7 +774,6 @@ void SymbolTable::clean_dead_entries(JavaThread* jt) {
 
   SymbolTableDeleteCheck stdc;
   SymbolTableDoDelete stdd;
-  NativeHeapTrimmer::SuspendMark sm("symboltable");
   {
     TraceTime timer("Clean", TRACETIME_LOG(Debug, symboltable, perf));
     while (bdt.do_task(jt, stdc, stdd)) {

src/hotspot/share/classfile/vmSymbols.hpp

@@ -245,10 +245,6 @@ class SerializeClosure;
                                                                                                   \
   /* Concurrency support */                                                                       \
   template(java_util_concurrent_locks_AbstractOwnableSynchronizer,           "java/util/concurrent/locks/AbstractOwnableSynchronizer") \
-  template(java_util_concurrent_atomic_AtomicIntegerFieldUpdater_Impl,       "java/util/concurrent/atomic/AtomicIntegerFieldUpdater$AtomicIntegerFieldUpdaterImpl") \
-  template(java_util_concurrent_atomic_AtomicLongFieldUpdater_CASUpdater,    "java/util/concurrent/atomic/AtomicLongFieldUpdater$CASUpdater") \
-  template(java_util_concurrent_atomic_AtomicLongFieldUpdater_LockedUpdater, "java/util/concurrent/atomic/AtomicLongFieldUpdater$LockedUpdater") \
-  template(java_util_concurrent_atomic_AtomicReferenceFieldUpdater_Impl,     "java/util/concurrent/atomic/AtomicReferenceFieldUpdater$AtomicReferenceFieldUpdaterImpl") \
   template(jdk_internal_vm_annotation_Contended_signature,                   "Ljdk/internal/vm/annotation/Contended;")    \
   template(jdk_internal_vm_annotation_ReservedStackAccess_signature,         "Ljdk/internal/vm/annotation/ReservedStackAccess;") \
   template(jdk_internal_ValueBased_signature,                                "Ljdk/internal/ValueBased;") \
@@ -302,6 +298,7 @@ class SerializeClosure;
   template(jdk_internal_misc_Scoped_signature,               "Ljdk/internal/misc/ScopedMemoryAccess$Scoped;") \
   template(jdk_internal_vm_annotation_IntrinsicCandidate_signature, "Ljdk/internal/vm/annotation/IntrinsicCandidate;") \
   template(jdk_internal_vm_annotation_Stable_signature,      "Ljdk/internal/vm/annotation/Stable;") \
+  template(jdk_internal_vm_annotation_TrustFinalFields_signature, "Ljdk/internal/vm/annotation/TrustFinalFields;") \
                                                                                                   \
   template(jdk_internal_vm_annotation_ChangesCurrentThread_signature,  "Ljdk/internal/vm/annotation/ChangesCurrentThread;")  \
   template(jdk_internal_vm_annotation_JvmtiHideEvents_signature,       "Ljdk/internal/vm/annotation/JvmtiHideEvents;")  \

src/hotspot/share/gc/g1/g1Arguments.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2018, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2018, 2026, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2017, Red Hat, Inc. and/or its affiliates.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
@@ -209,6 +209,17 @@ void G1Arguments::initialize() {
     FLAG_SET_DEFAULT(GCTimeRatio, 24);
   }
 
+  // Do not interfere with GC-Pressure driven heap resizing unless the user
+  // explicitly sets otherwise. G1 heap sizing should be free to grow or shrink
+  // the heap based on GC pressure, rather than being forced to satisfy
+  // MinHeapFreeRatio or MaxHeapFreeRatio defaults that the user did not set.
+  if (FLAG_IS_DEFAULT(MinHeapFreeRatio)) {
+    FLAG_SET_DEFAULT(MinHeapFreeRatio, 0);
+  }
+  if (FLAG_IS_DEFAULT(MaxHeapFreeRatio)) {
+    FLAG_SET_DEFAULT(MaxHeapFreeRatio, 100);
+  }
+
   // Below, we might need to calculate the pause time interval based on
   // the pause target. When we do so we are going to give G1 maximum
   // flexibility and allow it to do pauses when it needs to. So, we'll

src/hotspot/share/gc/g1/g1BarrierSet.inline.hpp

@@ -70,7 +70,11 @@ inline void G1BarrierSet::write_ref_field_pre(T* field) {
 
 template <DecoratorSet decorators, typename T>
 inline void G1BarrierSet::write_ref_field_post(T* field) {
-  volatile CardValue* byte = _card_table->byte_for(field);
+  // Make sure that the card table reference is read only once. Otherwise the compiler
+  // might reload that value in the two accesses below, that could cause writes to
+  // the wrong card table.
+  CardTable* card_table = AtomicAccess::load(&_card_table);
+  CardValue* byte = card_table->byte_for(field);
   if (*byte == G1CardTable::clean_card_val()) {
     *byte = G1CardTable::dirty_card_val();
   }

src/hotspot/share/gc/g1/g1BatchedTask.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2021, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2021, 2026, 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
@@ -26,7 +26,6 @@
 #include "gc/g1/g1BatchedTask.hpp"
 #include "gc/g1/g1CollectedHeap.inline.hpp"
 #include "gc/g1/g1GCParPhaseTimesTracker.hpp"
-#include "runtime/atomicAccess.hpp"
 #include "utilities/growableArray.hpp"
 
 void G1AbstractSubTask::record_work_item(uint worker_id, uint index, size_t count) {
@@ -40,7 +39,7 @@ const char* G1AbstractSubTask::name() const {
 }
 
 bool G1BatchedTask::try_claim_serial_task(int& task) {
-  task = AtomicAccess::fetch_then_add(&_num_serial_tasks_done, 1);
+  task = _num_serial_tasks_done.fetch_then_add(1);
   return task < _serial_tasks.length();
 }
 
@@ -96,8 +95,8 @@ void G1BatchedTask::work(uint worker_id) {
 }
 
 G1BatchedTask::~G1BatchedTask() {
-  assert(AtomicAccess::load(&_num_serial_tasks_done) >= _serial_tasks.length(),
-         "Only %d tasks of %d claimed", AtomicAccess::load(&_num_serial_tasks_done), _serial_tasks.length());
+  assert(_num_serial_tasks_done.load_relaxed() >= _serial_tasks.length(),
+         "Only %d tasks of %d claimed", _num_serial_tasks_done.load_relaxed(), _serial_tasks.length());
 
   for (G1AbstractSubTask* task : _parallel_tasks) {
     delete task;

src/hotspot/share/gc/g1/g1BatchedTask.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2021, 2024, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2021, 2026, 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
@@ -28,6 +28,7 @@
 #include "gc/g1/g1GCPhaseTimes.hpp"
 #include "gc/shared/workerThread.hpp"
 #include "memory/allocation.hpp"
+#include "runtime/atomic.hpp"
 
 template <typename E, MemTag MT>
 class GrowableArrayCHeap;
@@ -120,7 +121,7 @@ public:
 // 5) ~T()
 //
 class G1BatchedTask : public WorkerTask {
-  volatile int _num_serial_tasks_done;
+  Atomic<int> _num_serial_tasks_done;
   G1GCPhaseTimes* _phase_times;
 
   bool try_claim_serial_task(int& task);

src/hotspot/share/gc/g1/g1CardSet.cpp

@@ -29,7 +29,6 @@
 #include "gc/shared/gcLogPrecious.hpp"
 #include "gc/shared/gcTraceTime.inline.hpp"
 #include "memory/allocation.inline.hpp"
-#include "runtime/atomicAccess.hpp"
 #include "runtime/globals_extension.hpp"
 #include "runtime/java.hpp"
 #include "utilities/bitMap.inline.hpp"
@@ -192,32 +191,32 @@ const char* G1CardSetConfiguration::mem_object_type_name_str(uint index) {
 void G1CardSetCoarsenStats::reset() {
   STATIC_ASSERT(ARRAY_SIZE(_coarsen_from) == ARRAY_SIZE(_coarsen_collision));
   for (uint i = 0; i < ARRAY_SIZE(_coarsen_from); i++) {
-    _coarsen_from[i] = 0;
-    _coarsen_collision[i] = 0;
+    _coarsen_from[i].store_relaxed(0);
+    _coarsen_collision[i].store_relaxed(0);
   }
 }
 
 void G1CardSetCoarsenStats::set(G1CardSetCoarsenStats& other) {
   STATIC_ASSERT(ARRAY_SIZE(_coarsen_from) == ARRAY_SIZE(_coarsen_collision));
   for (uint i = 0; i < ARRAY_SIZE(_coarsen_from); i++) {
-    _coarsen_from[i] = other._coarsen_from[i];
-    _coarsen_collision[i] = other._coarsen_collision[i];
+    _coarsen_from[i].store_relaxed(other._coarsen_from[i].load_relaxed());
+    _coarsen_collision[i].store_relaxed(other._coarsen_collision[i].load_relaxed());
   }
 }
 
 void G1CardSetCoarsenStats::subtract_from(G1CardSetCoarsenStats& other) {
   STATIC_ASSERT(ARRAY_SIZE(_coarsen_from) == ARRAY_SIZE(_coarsen_collision));
   for (uint i = 0; i < ARRAY_SIZE(_coarsen_from); i++) {
-    _coarsen_from[i] = other._coarsen_from[i] - _coarsen_from[i];
-    _coarsen_collision[i] = other._coarsen_collision[i] - _coarsen_collision[i];
+    _coarsen_from[i].store_relaxed(other._coarsen_from[i].load_relaxed() - _coarsen_from[i].load_relaxed());
+    _coarsen_collision[i].store_relaxed(other._coarsen_collision[i].load_relaxed() - _coarsen_collision[i].load_relaxed());
   }
 }
 
 void G1CardSetCoarsenStats::record_coarsening(uint tag, bool collision) {
   assert(tag < ARRAY_SIZE(_coarsen_from), "tag %u out of bounds", tag);
-  AtomicAccess::inc(&_coarsen_from[tag], memory_order_relaxed);
+  _coarsen_from[tag].add_then_fetch(1u, memory_order_relaxed);
   if (collision) {
-    AtomicAccess::inc(&_coarsen_collision[tag], memory_order_relaxed);
+    _coarsen_collision[tag].add_then_fetch(1u, memory_order_relaxed);
   }
 }
 
@@ -228,13 +227,13 @@ void G1CardSetCoarsenStats::print_on(outputStream* out) {
                 "Inline->AoC %zu (%zu) "
                 "AoC->BitMap %zu (%zu) "
                 "BitMap->Full %zu (%zu) ",
-                _coarsen_from[0], _coarsen_collision[0],
-                _coarsen_from[1], _coarsen_collision[1],
+                _coarsen_from[0].load_relaxed(), _coarsen_collision[0].load_relaxed(),
+                _coarsen_from[1].load_relaxed(), _coarsen_collision[1].load_relaxed(),
                 // There is no BitMap at the first level so we can't .
-                _coarsen_from[3], _coarsen_collision[3],
-                _coarsen_from[4], _coarsen_collision[4],
-                _coarsen_from[5], _coarsen_collision[5],
-                _coarsen_from[6], _coarsen_collision[6]
+                _coarsen_from[3].load_relaxed(), _coarsen_collision[3].load_relaxed(),
+                _coarsen_from[4].load_relaxed(), _coarsen_collision[4].load_relaxed(),
+                _coarsen_from[5].load_relaxed(), _coarsen_collision[5].load_relaxed(),
+                _coarsen_from[6].load_relaxed(), _coarsen_collision[6].load_relaxed()
                );
 }
 
@@ -248,7 +247,7 @@ class G1CardSetHashTable : public CHeapObj<mtGCCardSet> {
   // the per region cardsets.
   const static uint GroupBucketClaimSize = 4;
   // Did we insert at least one card in the table?
-  bool volatile _inserted_card;
+  Atomic<bool> _inserted_card;
 
   G1CardSetMemoryManager* _mm;
   CardSetHash _table;
@@ -311,10 +310,10 @@ public:
     G1CardSetHashTableValue value(region_idx, G1CardSetInlinePtr());
     bool inserted = _table.insert_get(Thread::current(), lookup, value, found, should_grow);
 
-    if (!_inserted_card && inserted) {
+    if (!_inserted_card.load_relaxed() && inserted) {
       // It does not matter to us who is setting the flag so a regular atomic store
       // is sufficient.
-      AtomicAccess::store(&_inserted_card, true);
+      _inserted_card.store_relaxed(true);
     }
 
     return found.value();
@@ -343,9 +342,9 @@ public:
   }
 
   void reset() {
-    if (AtomicAccess::load(&_inserted_card)) {
+    if (_inserted_card.load_relaxed()) {
       _table.unsafe_reset(InitialLogTableSize);
-      AtomicAccess::store(&_inserted_card, false);
+      _inserted_card.store_relaxed(false);
     }
   }
 
@@ -455,14 +454,14 @@ void G1CardSet::free_mem_object(ContainerPtr container) {
   _mm->free(container_type_to_mem_object_type(type), value);
 }
 
-G1CardSet::ContainerPtr G1CardSet::acquire_container(ContainerPtr volatile* container_addr) {
+G1CardSet::ContainerPtr G1CardSet::acquire_container(Atomic<ContainerPtr>* container_addr) {
   // Update reference counts under RCU critical section to avoid a
   // use-after-cleapup bug where we increment a reference count for
   // an object whose memory has already been cleaned up and reused.
   GlobalCounter::CriticalSection cs(Thread::current());
   while (true) {
     // Get ContainerPtr and increment refcount atomically wrt to memory reuse.
-    ContainerPtr container = AtomicAccess::load_acquire(container_addr);
+    ContainerPtr container = container_addr->load_acquire();
     uint cs_type = container_type(container);
     if (container == FullCardSet || cs_type == ContainerInlinePtr) {
       return container;
@@ -503,15 +502,15 @@ class G1ReleaseCardsets : public StackObj {
   G1CardSet* _card_set;
   using ContainerPtr = G1CardSet::ContainerPtr;
 
-  void coarsen_to_full(ContainerPtr* container_addr) {
+  void coarsen_to_full(Atomic<ContainerPtr>* container_addr) {
     while (true) {
-      ContainerPtr cur_container = AtomicAccess::load_acquire(container_addr);
+      ContainerPtr cur_container = container_addr->load_acquire();
       uint cs_type = G1CardSet::container_type(cur_container);
       if (cur_container == G1CardSet::FullCardSet) {
         return;
       }
 
-      ContainerPtr old_value = AtomicAccess::cmpxchg(container_addr, cur_container, G1CardSet::FullCardSet);
+      ContainerPtr old_value = container_addr->compare_exchange(cur_container, G1CardSet::FullCardSet);
 
       if (old_value == cur_container) {
         _card_set->release_and_maybe_free_container(cur_container);
@@ -523,7 +522,7 @@ class G1ReleaseCardsets : public StackObj {
 public:
   explicit G1ReleaseCardsets(G1CardSet* card_set) : _card_set(card_set) { }
 
-  void operator ()(ContainerPtr* container_addr) {
+  void operator ()(Atomic<ContainerPtr>* container_addr) {
     coarsen_to_full(container_addr);
   }
 };
@@ -544,10 +543,10 @@ G1AddCardResult G1CardSet::add_to_howl(ContainerPtr parent_container,
   ContainerPtr container;
 
   uint bucket = _config->howl_bucket_index(card_in_region);
-  ContainerPtr volatile* bucket_entry = howl->container_addr(bucket);
+  Atomic<ContainerPtr>* bucket_entry = howl->container_addr(bucket);
 
   while (true) {
-    if (AtomicAccess::load(&howl->_num_entries) >= _config->cards_in_howl_threshold()) {
+    if (howl->_num_entries.load_relaxed() >= _config->cards_in_howl_threshold()) {
       return Overflow;
     }
 
@@ -571,7 +570,7 @@ G1AddCardResult G1CardSet::add_to_howl(ContainerPtr parent_container,
   }
 
   if (increment_total && add_result == Added) {
-    AtomicAccess::inc(&howl->_num_entries, memory_order_relaxed);
+    howl->_num_entries.add_then_fetch(1u, memory_order_relaxed);
   }
 
   if (to_transfer != nullptr) {
@@ -588,7 +587,7 @@ G1AddCardResult G1CardSet::add_to_bitmap(ContainerPtr container, uint card_in_re
   return bitmap->add(card_offset, _config->cards_in_howl_bitmap_threshold(), _config->max_cards_in_howl_bitmap());
 }
 
-G1AddCardResult G1CardSet::add_to_inline_ptr(ContainerPtr volatile* container_addr, ContainerPtr container, uint card_in_region) {
+G1AddCardResult G1CardSet::add_to_inline_ptr(Atomic<ContainerPtr>* container_addr, ContainerPtr container, uint card_in_region) {
   G1CardSetInlinePtr value(container_addr, container);
   return value.add(card_in_region, _config->inline_ptr_bits_per_card(), _config->max_cards_in_inline_ptr());
 }
@@ -610,7 +609,7 @@ G1CardSet::ContainerPtr G1CardSet::create_coarsened_array_of_cards(uint card_in_
   return new_container;
 }
 
-bool G1CardSet::coarsen_container(ContainerPtr volatile* container_addr,
+bool G1CardSet::coarsen_container(Atomic<ContainerPtr>* container_addr,
                                   ContainerPtr cur_container,
                                   uint card_in_region,
                                   bool within_howl) {
@@ -640,7 +639,7 @@ bool G1CardSet::coarsen_container(ContainerPtr volatile* container_addr,
       ShouldNotReachHere();
   }
 
-  ContainerPtr old_value = AtomicAccess::cmpxchg(container_addr, cur_container, new_container); // Memory order?
+  ContainerPtr old_value = container_addr->compare_exchange(cur_container, new_container); // Memory order?
   if (old_value == cur_container) {
     // Success. Indicate that the cards from the current card set must be transferred
     // by this caller.
@@ -687,7 +686,7 @@ void G1CardSet::transfer_cards(G1CardSetHashTableValue* table_entry, ContainerPt
     assert(container_type(source_container) == ContainerHowl, "must be");
     // Need to correct for that the Full remembered set occupies more cards than the
     // AoCS before.
-    AtomicAccess::add(&_num_occupied, _config->max_cards_in_region() - table_entry->_num_occupied, memory_order_relaxed);
+    _num_occupied.add_then_fetch(_config->max_cards_in_region() - table_entry->_num_occupied.load_relaxed(), memory_order_relaxed);
   }
 }
 
@@ -713,18 +712,18 @@ void G1CardSet::transfer_cards_in_howl(ContainerPtr parent_container,
     diff -= 1;
 
     G1CardSetHowl* howling_array = container_ptr<G1CardSetHowl>(parent_container);
-    AtomicAccess::add(&howling_array->_num_entries, diff, memory_order_relaxed);
+    howling_array->_num_entries.add_then_fetch(diff, memory_order_relaxed);
 
     G1CardSetHashTableValue* table_entry = get_container(card_region);
     assert(table_entry != nullptr, "Table entry not found for transferred cards");
 
-    AtomicAccess::add(&table_entry->_num_occupied, diff, memory_order_relaxed);
+    table_entry->_num_occupied.add_then_fetch(diff, memory_order_relaxed);
 
-    AtomicAccess::add(&_num_occupied, diff, memory_order_relaxed);
+    _num_occupied.add_then_fetch(diff, memory_order_relaxed);
   }
 }
 
-G1AddCardResult G1CardSet::add_to_container(ContainerPtr volatile* container_addr,
+G1AddCardResult G1CardSet::add_to_container(Atomic<ContainerPtr>* container_addr,
                                             ContainerPtr container,
                                             uint card_region,
                                             uint card_in_region,
@@ -827,8 +826,8 @@ G1AddCardResult G1CardSet::add_card(uint card_region, uint card_in_region, bool
   }
 
   if (increment_total && add_result == Added) {
-    AtomicAccess::inc(&table_entry->_num_occupied, memory_order_relaxed);
-    AtomicAccess::inc(&_num_occupied, memory_order_relaxed);
+    table_entry->_num_occupied.add_then_fetch(1u, memory_order_relaxed);
+    _num_occupied.add_then_fetch(1u, memory_order_relaxed);
   }
   if (should_grow_table) {
     _table->grow();
@@ -853,7 +852,7 @@ bool G1CardSet::contains_card(uint card_region, uint card_in_region) {
     return false;
   }
 
-  ContainerPtr container = table_entry->_container;
+  ContainerPtr container = table_entry->_container.load_relaxed();
   if (container == FullCardSet) {
     // contains_card() is not a performance critical method so we do not hide that
     // case in the switch below.
@@ -889,7 +888,7 @@ void G1CardSet::print_info(outputStream* st, uintptr_t card) {
     return;
   }
 
-  ContainerPtr container = table_entry->_container;
+  ContainerPtr container = table_entry->_container.load_relaxed();
   if (container == FullCardSet) {
     st->print("FULL card set)");
     return;
@@ -940,7 +939,7 @@ void G1CardSet::iterate_cards_during_transfer(ContainerPtr const container, Card
 void G1CardSet::iterate_containers(ContainerPtrClosure* cl, bool at_safepoint) {
   auto do_value =
     [&] (G1CardSetHashTableValue* value) {
-      cl->do_containerptr(value->_region_idx, value->_num_occupied, value->_container);
+      cl->do_containerptr(value->_region_idx, value->_num_occupied.load_relaxed(), value->_container.load_relaxed());
       return true;
     };
 
@@ -1001,11 +1000,11 @@ bool G1CardSet::occupancy_less_or_equal_to(size_t limit) const {
 }
 
 bool G1CardSet::is_empty() const {
-  return _num_occupied == 0;
+  return _num_occupied.load_relaxed() == 0;
 }
 
 size_t G1CardSet::occupied() const {
-  return _num_occupied;
+  return _num_occupied.load_relaxed();
 }
 
 size_t G1CardSet::num_containers() {
@@ -1051,7 +1050,7 @@ size_t G1CardSet::static_mem_size() {
 
 void G1CardSet::clear() {
   _table->reset();
-  _num_occupied = 0;
+  _num_occupied.store_relaxed(0);
   _mm->flush();
 }
 

src/hotspot/share/gc/g1/g1CardSet.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2021, 2024, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2021, 2026, 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
@@ -27,6 +27,7 @@
 
 #include "memory/allocation.hpp"
 #include "memory/memRegion.hpp"
+#include "runtime/atomic.hpp"
 #include "utilities/concurrentHashTable.hpp"
 
 class G1CardSetAllocOptions;
@@ -154,8 +155,8 @@ public:
 
 private:
   // Indices are "from" indices.
-  size_t _coarsen_from[NumCoarsenCategories];
-  size_t _coarsen_collision[NumCoarsenCategories];
+  Atomic<size_t> _coarsen_from[NumCoarsenCategories];
+  Atomic<size_t> _coarsen_collision[NumCoarsenCategories];
 
 public:
   G1CardSetCoarsenStats() { reset(); }
@@ -271,11 +272,11 @@ private:
 
   // Total number of cards in this card set. This is a best-effort value, i.e. there may
   // be (slightly) more cards in the card set than this value in reality.
-  size_t _num_occupied;
+  Atomic<size_t> _num_occupied;
 
   ContainerPtr make_container_ptr(void* value, uintptr_t type);
 
-  ContainerPtr acquire_container(ContainerPtr volatile* container_addr);
+  ContainerPtr acquire_container(Atomic<ContainerPtr>* container_addr);
   // Returns true if the card set container should be released
   bool release_container(ContainerPtr container);
   // Release card set and free if needed.
@@ -288,7 +289,7 @@ private:
   // coarsen_container does not transfer cards from cur_container
   // to the new container. Transfer is achieved by transfer_cards.
   // Returns true if this was the thread that coarsened the container (and added the card).
-  bool coarsen_container(ContainerPtr volatile* container_addr,
+  bool coarsen_container(Atomic<ContainerPtr>* container_addr,
                          ContainerPtr cur_container,
                          uint card_in_region, bool within_howl = false);
 
@@ -300,9 +301,9 @@ private:
   void transfer_cards(G1CardSetHashTableValue* table_entry, ContainerPtr source_container, uint card_region);
   void transfer_cards_in_howl(ContainerPtr parent_container, ContainerPtr source_container, uint card_region);
 
-  G1AddCardResult add_to_container(ContainerPtr volatile* container_addr, ContainerPtr container, uint card_region, uint card, bool increment_total = true);
+  G1AddCardResult add_to_container(Atomic<ContainerPtr>* container_addr, ContainerPtr container, uint card_region, uint card, bool increment_total = true);
 
-  G1AddCardResult add_to_inline_ptr(ContainerPtr volatile* container_addr, ContainerPtr container, uint card_in_region);
+  G1AddCardResult add_to_inline_ptr(Atomic<ContainerPtr>* container_addr, ContainerPtr container, uint card_in_region);
   G1AddCardResult add_to_array(ContainerPtr container, uint card_in_region);
   G1AddCardResult add_to_bitmap(ContainerPtr container, uint card_in_region);
   G1AddCardResult add_to_howl(ContainerPtr parent_container, uint card_region, uint card_in_region, bool increment_total = true);
@@ -366,7 +367,6 @@ public:
 
   size_t num_containers();
 
-  static G1CardSetCoarsenStats coarsen_stats();
   static void print_coarsen_stats(outputStream* out);
 
   // Returns size of the actual remembered set containers in bytes.
@@ -412,8 +412,15 @@ public:
   using ContainerPtr = G1CardSet::ContainerPtr;
 
   const uint _region_idx;
-  uint volatile _num_occupied;
-  ContainerPtr volatile _container;
+  Atomic<uint> _num_occupied;
+  Atomic<ContainerPtr> _container;
+
+  // Copy constructor needed for use in ConcurrentHashTable.
+  G1CardSetHashTableValue(const G1CardSetHashTableValue& other) :
+    _region_idx(other._region_idx),
+    _num_occupied(other._num_occupied.load_relaxed()),
+    _container(other._container.load_relaxed())
+  { }
 
   G1CardSetHashTableValue(uint region_idx, ContainerPtr container) : _region_idx(region_idx), _num_occupied(0), _container(container) { }
 };

src/hotspot/share/gc/g1/g1CardSetContainers.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2023, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2023, 2026, 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
@@ -27,7 +27,7 @@
 
 #include "gc/g1/g1CardSet.hpp"
 #include "memory/allocation.hpp"
-#include "runtime/atomicAccess.hpp"
+#include "runtime/atomic.hpp"
 #include "utilities/bitMap.hpp"
 #include "utilities/globalDefinitions.hpp"
 
@@ -67,7 +67,7 @@ class G1CardSetInlinePtr : public StackObj {
 
   using ContainerPtr = G1CardSet::ContainerPtr;
 
-  ContainerPtr volatile * _value_addr;
+  Atomic<ContainerPtr>* _value_addr;
   ContainerPtr _value;
 
   static const uint SizeFieldLen = 3;
@@ -103,7 +103,7 @@ public:
   explicit G1CardSetInlinePtr(ContainerPtr value) :
     G1CardSetInlinePtr(nullptr, value) {}
 
-  G1CardSetInlinePtr(ContainerPtr volatile* value_addr, ContainerPtr value) : _value_addr(value_addr), _value(value) {
+  G1CardSetInlinePtr(Atomic<ContainerPtr>* value_addr, ContainerPtr value) : _value_addr(value_addr), _value(value) {
     assert(G1CardSet::container_type(_value) == G1CardSet::ContainerInlinePtr, "Value " PTR_FORMAT " is not a valid G1CardSetInlinePtr.", p2i(_value));
   }
 
@@ -145,13 +145,13 @@ public:
 // All but inline pointers are of this kind. For those, card entries are stored
 // directly in the ContainerPtr of the ConcurrentHashTable node.
 class G1CardSetContainer {
-  uintptr_t _ref_count;
+  Atomic<uintptr_t> _ref_count;
 protected:
   ~G1CardSetContainer() = default;
 public:
   G1CardSetContainer() : _ref_count(3) { }
 
-  uintptr_t refcount() const { return AtomicAccess::load_acquire(&_ref_count); }
+  uintptr_t refcount() const { return _ref_count.load_acquire(); }
 
   bool try_increment_refcount();
 
@@ -172,7 +172,7 @@ public:
   using ContainerPtr = G1CardSet::ContainerPtr;
 private:
   EntryCountType _size;
-  EntryCountType volatile _num_entries;
+  Atomic<EntryCountType> _num_entries;
   // VLA implementation.
   EntryDataType _data[1];
 
@@ -180,10 +180,10 @@ private:
   static const EntryCountType EntryMask = LockBitMask - 1;
 
   class G1CardSetArrayLocker : public StackObj {
-    EntryCountType volatile* _num_entries_addr;
+    Atomic<EntryCountType>* _num_entries_addr;
     EntryCountType _local_num_entries;
   public:
-    G1CardSetArrayLocker(EntryCountType volatile* value);
+    G1CardSetArrayLocker(Atomic<EntryCountType>* value);
 
     EntryCountType num_entries() const { return _local_num_entries; }
     void inc_num_entries() {
@@ -192,7 +192,7 @@ private:
     }
 
     ~G1CardSetArrayLocker() {
-      AtomicAccess::release_store(_num_entries_addr, _local_num_entries);
+      _num_entries_addr->release_store(_local_num_entries);
     }
   };
 
@@ -213,7 +213,7 @@ public:
   template <class CardVisitor>
   void iterate(CardVisitor& found);
 
-  size_t num_entries() const { return _num_entries & EntryMask; }
+  size_t num_entries() const { return _num_entries.load_relaxed() & EntryMask; }
 
   static size_t header_size_in_bytes();
 
@@ -223,7 +223,7 @@ public:
 };
 
 class G1CardSetBitMap : public G1CardSetContainer {
-  size_t _num_bits_set;
+  Atomic<size_t> _num_bits_set;
   BitMap::bm_word_t _bits[1];
 
 public:
@@ -236,7 +236,7 @@ public:
     return bm.at(card_idx);
   }
 
-  uint num_bits_set() const { return (uint)_num_bits_set; }
+  uint num_bits_set() const { return (uint)_num_bits_set.load_relaxed(); }
 
   template <class CardVisitor>
   void iterate(CardVisitor& found, size_t const size_in_bits, uint offset);
@@ -255,10 +255,10 @@ class G1CardSetHowl : public G1CardSetContainer {
 public:
   typedef uint EntryCountType;
   using ContainerPtr = G1CardSet::ContainerPtr;
-  EntryCountType volatile _num_entries;
+  Atomic<EntryCountType> _num_entries;
 private:
   // VLA implementation.
-  ContainerPtr _buckets[1];
+  Atomic<ContainerPtr> _buckets[1];
   // Do not add class member variables beyond this point.
 
   // Iterates over the given ContainerPtr with at index in this Howl card set,
@@ -268,14 +268,14 @@ private:
 
   ContainerPtr at(EntryCountType index) const;
 
-  ContainerPtr const* buckets() const;
+  Atomic<ContainerPtr> const* buckets() const;
 
 public:
   G1CardSetHowl(EntryCountType card_in_region, G1CardSetConfiguration* config);
 
-  ContainerPtr const* container_addr(EntryCountType index) const;
+  Atomic<ContainerPtr> const* container_addr(EntryCountType index) const;
 
-  ContainerPtr* container_addr(EntryCountType index);
+  Atomic<ContainerPtr>* container_addr(EntryCountType index);
 
   bool contains(uint card_idx, G1CardSetConfiguration* config);
   // Iterates over all ContainerPtrs in this Howl card set, applying a CardOrRangeVisitor

src/hotspot/share/gc/g1/g1CardSetContainers.inline.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2021, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2021, 2026, 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
@@ -67,7 +67,7 @@ inline G1AddCardResult G1CardSetInlinePtr::add(uint card_idx, uint bits_per_card
       return Overflow;
     }
     ContainerPtr new_value = merge(_value, card_idx, num_cards, bits_per_card);
-    ContainerPtr old_value = AtomicAccess::cmpxchg(_value_addr, _value, new_value, memory_order_relaxed);
+    ContainerPtr old_value = _value_addr->compare_exchange(_value, new_value, memory_order_relaxed);
     if (_value == old_value) {
       return Added;
     }
@@ -126,7 +126,7 @@ inline bool G1CardSetContainer::try_increment_refcount() {
     }
 
     uintptr_t new_value = old_value + 2;
-    uintptr_t ref_count = AtomicAccess::cmpxchg(&_ref_count, old_value, new_value);
+    uintptr_t ref_count = _ref_count.compare_exchange(old_value, new_value);
     if (ref_count == old_value) {
       return true;
     }
@@ -137,7 +137,7 @@ inline bool G1CardSetContainer::try_increment_refcount() {
 inline uintptr_t G1CardSetContainer::decrement_refcount() {
   uintptr_t old_value = refcount();
   assert((old_value & 0x1) != 0 && old_value >= 3, "precondition");
-  return AtomicAccess::sub(&_ref_count, 2u);
+  return _ref_count.sub_then_fetch(2u);
 }
 
 inline G1CardSetArray::G1CardSetArray(uint card_in_region, EntryCountType num_cards) :
@@ -149,14 +149,13 @@ inline G1CardSetArray::G1CardSetArray(uint card_in_region, EntryCountType num_ca
   *entry_addr(0) = checked_cast<EntryDataType>(card_in_region);
 }
 
-inline G1CardSetArray::G1CardSetArrayLocker::G1CardSetArrayLocker(EntryCountType volatile* num_entries_addr) :
+inline G1CardSetArray::G1CardSetArrayLocker::G1CardSetArrayLocker(Atomic<EntryCountType>* num_entries_addr) :
   _num_entries_addr(num_entries_addr) {
   SpinYield s;
-  EntryCountType num_entries = AtomicAccess::load(_num_entries_addr) & EntryMask;
+  EntryCountType num_entries = _num_entries_addr->load_relaxed() & EntryMask;
   while (true) {
-    EntryCountType old_value = AtomicAccess::cmpxchg(_num_entries_addr,
-                                                     num_entries,
-                                                     (EntryCountType)(num_entries | LockBitMask));
+    EntryCountType old_value = _num_entries_addr->compare_exchange(num_entries,
+                                                                   (EntryCountType)(num_entries | LockBitMask));
     if (old_value == num_entries) {
       // Succeeded locking the array.
       _local_num_entries = num_entries;
@@ -174,7 +173,7 @@ inline G1CardSetArray::EntryDataType const* G1CardSetArray::base_addr() const {
 }
 
 inline G1CardSetArray::EntryDataType const* G1CardSetArray::entry_addr(EntryCountType index) const {
-  assert(index < _num_entries, "precondition");
+  assert(index < _num_entries.load_relaxed(), "precondition");
   return base_addr() + index;
 }
 
@@ -189,7 +188,7 @@ inline G1CardSetArray::EntryDataType G1CardSetArray::at(EntryCountType index) co
 inline G1AddCardResult G1CardSetArray::add(uint card_idx) {
   assert(card_idx < (1u << (sizeof(EntryDataType) * BitsPerByte)),
          "Card index %u does not fit allowed card value range.", card_idx);
-  EntryCountType num_entries = AtomicAccess::load_acquire(&_num_entries) & EntryMask;
+  EntryCountType num_entries = _num_entries.load_acquire() & EntryMask;
   EntryCountType idx = 0;
   for (; idx < num_entries; idx++) {
     if (at(idx) == card_idx) {
@@ -223,7 +222,7 @@ inline G1AddCardResult G1CardSetArray::add(uint card_idx) {
 }
 
 inline bool G1CardSetArray::contains(uint card_idx) {
-  EntryCountType num_entries = AtomicAccess::load_acquire(&_num_entries) & EntryMask;
+  EntryCountType num_entries = _num_entries.load_acquire() & EntryMask;
 
   for (EntryCountType idx = 0; idx < num_entries; idx++) {
     if (at(idx) == card_idx) {
@@ -235,7 +234,7 @@ inline bool G1CardSetArray::contains(uint card_idx) {
 
 template <class CardVisitor>
 void G1CardSetArray::iterate(CardVisitor& found) {
-  EntryCountType num_entries = AtomicAccess::load_acquire(&_num_entries) & EntryMask;
+  EntryCountType num_entries = _num_entries.load_acquire() & EntryMask;
   for (EntryCountType idx = 0; idx < num_entries; idx++) {
     found(at(idx));
   }
@@ -256,11 +255,11 @@ inline G1CardSetBitMap::G1CardSetBitMap(uint card_in_region, uint size_in_bits)
 
 inline G1AddCardResult G1CardSetBitMap::add(uint card_idx, size_t threshold, size_t size_in_bits) {
   BitMapView bm(_bits, size_in_bits);
-  if (_num_bits_set >= threshold) {
+  if (_num_bits_set.load_relaxed() >= threshold) {
     return bm.at(card_idx) ? Found : Overflow;
   }
   if (bm.par_set_bit(card_idx)) {
-    AtomicAccess::inc(&_num_bits_set, memory_order_relaxed);
+    _num_bits_set.add_then_fetch(1u, memory_order_relaxed);
     return Added;
   }
   return Found;
@@ -276,22 +275,22 @@ inline size_t G1CardSetBitMap::header_size_in_bytes() {
     return offset_of(G1CardSetBitMap, _bits);
 }
 
-inline G1CardSetHowl::ContainerPtr const* G1CardSetHowl::container_addr(EntryCountType index) const {
-  assert(index < _num_entries, "precondition");
+inline Atomic<G1CardSetHowl::ContainerPtr> const* G1CardSetHowl::container_addr(EntryCountType index) const {
+  assert(index < _num_entries.load_relaxed(), "precondition");
   return buckets() + index;
 }
 
-inline G1CardSetHowl::ContainerPtr* G1CardSetHowl::container_addr(EntryCountType index) {
-  return const_cast<ContainerPtr*>(const_cast<const G1CardSetHowl*>(this)->container_addr(index));
+inline Atomic<G1CardSetHowl::ContainerPtr>* G1CardSetHowl::container_addr(EntryCountType index) {
+  return const_cast<Atomic<ContainerPtr>*>(const_cast<const G1CardSetHowl*>(this)->container_addr(index));
 }
 
 inline G1CardSetHowl::ContainerPtr G1CardSetHowl::at(EntryCountType index) const {
-  return *container_addr(index);
+  return (*container_addr(index)).load_relaxed();
 }
 
-inline G1CardSetHowl::ContainerPtr const* G1CardSetHowl::buckets() const {
+inline Atomic<G1CardSetHowl::ContainerPtr> const* G1CardSetHowl::buckets() const {
   const void* ptr = reinterpret_cast<const char*>(this) + header_size_in_bytes();
-  return reinterpret_cast<ContainerPtr const*>(ptr);
+  return reinterpret_cast<Atomic<ContainerPtr> const*>(ptr);
 }
 
 inline G1CardSetHowl::G1CardSetHowl(EntryCountType card_in_region, G1CardSetConfiguration* config) :
@@ -300,7 +299,7 @@ inline G1CardSetHowl::G1CardSetHowl(EntryCountType card_in_region, G1CardSetConf
   EntryCountType num_buckets = config->num_buckets_in_howl();
   EntryCountType bucket = config->howl_bucket_index(card_in_region);
   for (uint i = 0; i < num_buckets; ++i) {
-    *container_addr(i) = G1CardSetInlinePtr();
+    container_addr(i)->store_relaxed(G1CardSetInlinePtr());
     if (i == bucket) {
       G1CardSetInlinePtr value(container_addr(i), at(i));
       value.add(card_in_region, config->inline_ptr_bits_per_card(), config->max_cards_in_inline_ptr());
@@ -310,8 +309,8 @@ inline G1CardSetHowl::G1CardSetHowl(EntryCountType card_in_region, G1CardSetConf
 
 inline bool G1CardSetHowl::contains(uint card_idx, G1CardSetConfiguration* config) {
   EntryCountType bucket = config->howl_bucket_index(card_idx);
-  ContainerPtr* array_entry = container_addr(bucket);
-  ContainerPtr container = AtomicAccess::load_acquire(array_entry);
+  Atomic<ContainerPtr>* array_entry = container_addr(bucket);
+  ContainerPtr container = array_entry->load_acquire();
 
   switch (G1CardSet::container_type(container)) {
     case G1CardSet::ContainerArrayOfCards: {

src/hotspot/share/gc/g1/g1CardSetMemory.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2021, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2021, 2026, 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
@@ -26,7 +26,6 @@
 #include "gc/g1/g1CardSetContainers.inline.hpp"
 #include "gc/g1/g1CardSetMemory.inline.hpp"
 #include "gc/g1/g1MonotonicArena.inline.hpp"
-#include "runtime/atomicAccess.hpp"
 #include "utilities/ostream.hpp"
 
 G1CardSetAllocator::G1CardSetAllocator(const char* name,

src/hotspot/share/gc/g1/g1CardTableClaimTable.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2025, 2026, 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
@@ -44,20 +44,20 @@ G1CardTableClaimTable::~G1CardTableClaimTable() {
 
 void G1CardTableClaimTable::initialize(uint max_reserved_regions) {
   assert(_card_claims == nullptr, "Must not be initialized twice");
-  _card_claims = NEW_C_HEAP_ARRAY(uint, max_reserved_regions, mtGC);
+  _card_claims = NEW_C_HEAP_ARRAY(Atomic<uint>, max_reserved_regions, mtGC);
   _max_reserved_regions = max_reserved_regions;
   reset_all_to_unclaimed();
 }
 
 void G1CardTableClaimTable::reset_all_to_unclaimed() {
   for (uint i = 0; i < _max_reserved_regions; i++) {
-    _card_claims[i] = 0;
+    _card_claims[i].store_relaxed(0);
   }
 }
 
 void G1CardTableClaimTable::reset_all_to_claimed() {
   for (uint i = 0; i < _max_reserved_regions; i++) {
-    _card_claims[i] = (uint)G1HeapRegion::CardsPerRegion;
+    _card_claims[i].store_relaxed((uint)G1HeapRegion::CardsPerRegion);
   }
 }
 

src/hotspot/share/gc/g1/g1CardTableClaimTable.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2025, 2026, 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
@@ -27,6 +27,7 @@
 
 #include "gc/g1/g1CardTable.hpp"
 #include "memory/allocation.hpp"
+#include "runtime/atomic.hpp"
 
 class G1HeapRegionClosure;
 
@@ -45,7 +46,7 @@ class G1CardTableClaimTable : public CHeapObj<mtGC> {
 
   // Card table iteration claim values for every heap region, from 0 (completely unclaimed)
   // to (>=) G1HeapRegion::CardsPerRegion (completely claimed).
-  uint volatile* _card_claims;
+  Atomic<uint>* _card_claims;
 
   uint _cards_per_chunk;           // For conversion between card index and chunk index.
 

src/hotspot/share/gc/g1/g1CardTableClaimTable.inline.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2025, 2026, 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
@@ -29,26 +29,25 @@
 
 #include "gc/g1/g1CollectedHeap.inline.hpp"
 #include "gc/g1/g1HeapRegion.inline.hpp"
-#include "runtime/atomicAccess.hpp"
 
 bool G1CardTableClaimTable::has_unclaimed_cards(uint region) {
   assert(region < _max_reserved_regions, "Tried to access invalid region %u", region);
-  return AtomicAccess::load(&_card_claims[region]) < G1HeapRegion::CardsPerRegion;
+  return _card_claims[region].load_relaxed() < G1HeapRegion::CardsPerRegion;
 }
 
 void G1CardTableClaimTable::reset_to_unclaimed(uint region) {
   assert(region < _max_reserved_regions, "Tried to access invalid region %u", region);
-  AtomicAccess::store(&_card_claims[region], 0u);
+  _card_claims[region].store_relaxed(0u);
 }
 
 uint G1CardTableClaimTable::claim_cards(uint region, uint increment) {
   assert(region < _max_reserved_regions, "Tried to access invalid region %u", region);
-  return AtomicAccess::fetch_then_add(&_card_claims[region], increment, memory_order_relaxed);
+  return _card_claims[region].fetch_then_add(increment, memory_order_relaxed);
 }
 
 uint G1CardTableClaimTable::claim_chunk(uint region) {
   assert(region < _max_reserved_regions, "Tried to access invalid region %u", region);
-  return AtomicAccess::fetch_then_add(&_card_claims[region], cards_per_chunk(), memory_order_relaxed);
+  return _card_claims[region].fetch_then_add(cards_per_chunk(), memory_order_relaxed);
 }
 
 uint G1CardTableClaimTable::claim_all_cards(uint region) {

src/hotspot/share/gc/g1/g1CodeRootSet.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2014, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2014, 2026, 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
@@ -28,7 +28,7 @@
 #include "gc/g1/g1HeapRegion.hpp"
 #include "memory/allocation.hpp"
 #include "oops/oop.inline.hpp"
-#include "runtime/atomicAccess.hpp"
+#include "runtime/atomic.hpp"
 #include "utilities/concurrentHashTable.inline.hpp"
 #include "utilities/concurrentHashTableTasks.inline.hpp"
 
@@ -60,7 +60,7 @@ class G1CodeRootSetHashTable : public CHeapObj<mtGC> {
   HashTable _table;
   HashTableScanTask _table_scanner;
 
-  size_t volatile _num_entries;
+  Atomic<size_t> _num_entries;
 
   bool is_empty() const { return number_of_entries() == 0; }
 
@@ -120,7 +120,7 @@ public:
     bool grow_hint = false;
     bool inserted = _table.insert(Thread::current(), lookup, method, &grow_hint);
     if (inserted) {
-      AtomicAccess::inc(&_num_entries);
+      _num_entries.add_then_fetch(1u);
     }
     if (grow_hint) {
       _table.grow(Thread::current());
@@ -131,7 +131,7 @@ public:
     HashTableLookUp lookup(method);
     bool removed = _table.remove(Thread::current(), lookup);
     if (removed) {
-      AtomicAccess::dec(&_num_entries);
+      _num_entries.sub_then_fetch(1u);
     }
     return removed;
   }
@@ -182,7 +182,7 @@ public:
     guarantee(succeeded, "unable to clean table");
 
     if (num_deleted != 0) {
-      size_t current_size = AtomicAccess::sub(&_num_entries, num_deleted);
+      size_t current_size = _num_entries.sub_then_fetch(num_deleted);
       shrink_to_match(current_size);
     }
   }
@@ -226,7 +226,7 @@ public:
 
   size_t mem_size() { return sizeof(*this) + _table.get_mem_size(Thread::current()); }
 
-  size_t number_of_entries() const { return AtomicAccess::load(&_num_entries); }
+  size_t number_of_entries() const { return _num_entries.load_relaxed(); }
 };
 
 uintx G1CodeRootSetHashTable::HashTableLookUp::get_hash() const {

src/hotspot/share/gc/g1/g1CollectedHeap.cpp

@@ -103,7 +103,6 @@
 #include "oops/access.inline.hpp"
 #include "oops/compressedOops.inline.hpp"
 #include "oops/oop.inline.hpp"
-#include "runtime/atomicAccess.hpp"
 #include "runtime/cpuTimeCounters.hpp"
 #include "runtime/handles.inline.hpp"
 #include "runtime/init.hpp"

src/hotspot/share/gc/g1/g1CollectedHeap.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2001, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2001, 2026, 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
@@ -54,6 +54,7 @@
 #include "memory/allocation.hpp"
 #include "memory/iterator.hpp"
 #include "memory/memRegion.hpp"
+#include "runtime/atomic.hpp"
 #include "runtime/mutexLocker.hpp"
 #include "runtime/threadSMR.hpp"
 #include "utilities/bitMap.hpp"
@@ -124,7 +125,7 @@ class G1JavaThreadsListClaimer : public StackObj {
   ThreadsListHandle _list;
   uint _claim_step;
 
-  volatile uint _cur_claim;
+  Atomic<uint> _cur_claim;
 
   // Attempts to claim _claim_step JavaThreads, returning an array of claimed
   // JavaThread* with count elements. Returns null (and a zero count) if there
@@ -1267,7 +1268,6 @@ public:
 
   bool is_marked(oop obj) const;
 
-  inline static bool is_obj_filler(const oop obj);
   // Determine if an object is dead, given the object and also
   // the region to which the object belongs.
   inline bool is_obj_dead(const oop obj, const G1HeapRegion* hr) const;

src/hotspot/share/gc/g1/g1CollectedHeap.inline.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2001, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2001, 2026, 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
@@ -31,6 +31,7 @@
 #include "gc/g1/g1CollectorState.hpp"
 #include "gc/g1/g1ConcurrentMark.inline.hpp"
 #include "gc/g1/g1EvacFailureRegions.hpp"
+#include "gc/g1/g1EvacStats.inline.hpp"
 #include "gc/g1/g1HeapRegion.inline.hpp"
 #include "gc/g1/g1HeapRegionManager.inline.hpp"
 #include "gc/g1/g1HeapRegionRemSet.hpp"
@@ -38,10 +39,10 @@
 #include "gc/g1/g1Policy.hpp"
 #include "gc/g1/g1RegionPinCache.inline.hpp"
 #include "gc/g1/g1RemSet.hpp"
+#include "gc/shared/collectedHeap.inline.hpp"
 #include "gc/shared/markBitMap.inline.hpp"
 #include "gc/shared/taskqueue.inline.hpp"
 #include "oops/stackChunkOop.hpp"
-#include "runtime/atomicAccess.hpp"
 #include "runtime/threadSMR.inline.hpp"
 #include "utilities/bitMap.inline.hpp"
 
@@ -53,10 +54,10 @@ inline bool G1STWIsAliveClosure::do_object_b(oop p) {
 
 inline JavaThread* const* G1JavaThreadsListClaimer::claim(uint& count) {
   count = 0;
-  if (AtomicAccess::load(&_cur_claim) >= _list.length()) {
+  if (_cur_claim.load_relaxed() >= _list.length()) {
     return nullptr;
   }
-  uint claim = AtomicAccess::fetch_then_add(&_cur_claim, _claim_step);
+  uint claim = _cur_claim.fetch_then_add(_claim_step);
   if (claim >= _list.length()) {
     return nullptr;
   }
@@ -230,16 +231,11 @@ inline bool G1CollectedHeap::requires_barriers(stackChunkOop obj) const {
   return !heap_region_containing(obj)->is_young(); // is_in_young does an unnecessary null check
 }
 
-inline bool G1CollectedHeap::is_obj_filler(const oop obj) {
-  Klass* k = obj->klass_without_asserts();
-  return k == Universe::fillerArrayKlass() || k == vmClasses::FillerObject_klass();
-}
-
 inline bool G1CollectedHeap::is_obj_dead(const oop obj, const G1HeapRegion* hr) const {
   assert(!hr->is_free(), "looking up obj " PTR_FORMAT " in Free region %u", p2i(obj), hr->hrm_index());
   if (hr->is_in_parsable_area(obj)) {
     // This object is in the parsable part of the heap, live unless scrubbed.
-    return is_obj_filler(obj);
+    return is_filler_object(obj);
   } else {
     // From Remark until a region has been concurrently scrubbed, parts of the
     // region is not guaranteed to be parsable. Use the bitmap for liveness.

src/hotspot/share/gc/g1/g1CollectionSetChooser.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2001, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2001, 2026, 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
@@ -27,7 +27,7 @@
 #include "gc/g1/g1CollectionSetChooser.hpp"
 #include "gc/g1/g1HeapRegionRemSet.inline.hpp"
 #include "gc/shared/space.hpp"
-#include "runtime/atomicAccess.hpp"
+#include "runtime/atomic.hpp"
 #include "utilities/quickSort.hpp"
 
 // Determine collection set candidates (from marking): For all regions determine
@@ -50,7 +50,7 @@ class G1BuildCandidateRegionsTask : public WorkerTask {
 
     G1HeapRegion** _data;
 
-    uint volatile _cur_claim_idx;
+    Atomic<uint> _cur_claim_idx;
 
     static int compare_region_gc_efficiency(G1HeapRegion** rr1, G1HeapRegion** rr2) {
       G1HeapRegion* r1 = *rr1;
@@ -105,7 +105,7 @@ class G1BuildCandidateRegionsTask : public WorkerTask {
 
     // Claim a new chunk, returning its bounds [from, to[.
     void claim_chunk(uint& from, uint& to) {
-      uint result = AtomicAccess::add(&_cur_claim_idx, _chunk_size);
+      uint result = _cur_claim_idx.add_then_fetch(_chunk_size);
       assert(_max_size > result - 1,
              "Array too small, is %u should be %u with chunk size %u.",
              _max_size, result, _chunk_size);
@@ -121,14 +121,15 @@ class G1BuildCandidateRegionsTask : public WorkerTask {
     }
 
     void sort_by_gc_efficiency() {
-      if (_cur_claim_idx == 0) {
+      uint length = _cur_claim_idx.load_relaxed();
+      if (length == 0) {
         return;
       }
-      for (uint i = _cur_claim_idx; i < _max_size; i++) {
+      for (uint i = length; i < _max_size; i++) {
         assert(_data[i] == nullptr, "must be");
       }
-      qsort(_data, _cur_claim_idx, sizeof(_data[0]), (_sort_Fn)compare_region_gc_efficiency);
-      for (uint i = _cur_claim_idx; i < _max_size; i++) {
+      qsort(_data, length, sizeof(_data[0]), (_sort_Fn)compare_region_gc_efficiency);
+      for (uint i = length; i < _max_size; i++) {
         assert(_data[i] == nullptr, "must be");
       }
     }
@@ -202,13 +203,13 @@ class G1BuildCandidateRegionsTask : public WorkerTask {
   G1CollectedHeap* _g1h;
   G1HeapRegionClaimer _hrclaimer;
 
-  uint volatile _num_regions_added;
+  Atomic<uint> _num_regions_added;
 
   G1BuildCandidateArray _result;
 
   void update_totals(uint num_regions) {
     if (num_regions > 0) {
-      AtomicAccess::add(&_num_regions_added, num_regions);
+      _num_regions_added.add_then_fetch(num_regions);
     }
   }
 
@@ -220,7 +221,7 @@ class G1BuildCandidateRegionsTask : public WorkerTask {
   void prune(G1HeapRegion** data) {
     G1Policy* p = G1CollectedHeap::heap()->policy();
 
-    uint num_candidates = AtomicAccess::load(&_num_regions_added);
+    uint num_candidates = _num_regions_added.load_relaxed();
 
     uint min_old_cset_length = p->calc_min_old_cset_length(num_candidates);
     uint num_pruned = 0;
@@ -253,7 +254,7 @@ class G1BuildCandidateRegionsTask : public WorkerTask {
                               wasted_bytes,
                               allowed_waste);
 
-    AtomicAccess::sub(&_num_regions_added, num_pruned, memory_order_relaxed);
+    _num_regions_added.sub_then_fetch(num_pruned, memory_order_relaxed);
   }
 
 public:
@@ -274,7 +275,7 @@ public:
     _result.sort_by_gc_efficiency();
     prune(_result.array());
     candidates->set_candidates_from_marking(_result.array(),
-                                            _num_regions_added);
+                                            _num_regions_added.load_relaxed());
   }
 };
 

src/hotspot/share/gc/g1/g1ConcurrentMark.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2001, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2001, 2026, 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
@@ -51,6 +51,9 @@
 #include "gc/shared/gcTimer.hpp"
 #include "gc/shared/gcTraceTime.inline.hpp"
 #include "gc/shared/gcVMOperations.hpp"
+#include "gc/shared/partialArraySplitter.inline.hpp"
+#include "gc/shared/partialArrayState.hpp"
+#include "gc/shared/partialArrayTaskStats.hpp"
 #include "gc/shared/referencePolicy.hpp"
 #include "gc/shared/suspendibleThreadSet.hpp"
 #include "gc/shared/taskqueue.inline.hpp"
@@ -67,7 +70,6 @@
 #include "nmt/memTracker.hpp"
 #include "oops/access.inline.hpp"
 #include "oops/oop.inline.hpp"
-#include "runtime/atomicAccess.hpp"
 #include "runtime/globals_extension.hpp"
 #include "runtime/handles.inline.hpp"
 #include "runtime/java.hpp"
@@ -76,6 +78,7 @@
 #include "runtime/prefetch.inline.hpp"
 #include "runtime/threads.hpp"
 #include "utilities/align.hpp"
+#include "utilities/checkedCast.hpp"
 #include "utilities/formatBuffer.hpp"
 #include "utilities/growableArray.hpp"
 #include "utilities/powerOfTwo.hpp"
@@ -99,7 +102,7 @@ bool G1CMBitMapClosure::do_addr(HeapWord* const addr) {
   // We move that task's local finger along.
   _task->move_finger_to(addr);
 
-  _task->scan_task_entry(G1TaskQueueEntry::from_oop(cast_to_oop(addr)));
+  _task->process_entry(G1TaskQueueEntry(cast_to_oop(addr)), false /* stolen */);
   // we only partially drain the local queue and global stack
   _task->drain_local_queue(true);
   _task->drain_global_stack(true);
@@ -148,25 +151,25 @@ bool G1CMMarkStack::initialize() {
 }
 
 G1CMMarkStack::TaskQueueEntryChunk* G1CMMarkStack::ChunkAllocator::allocate_new_chunk() {
-  if (_size >= _max_capacity) {
+  if (_size.load_relaxed() >= _max_capacity) {
     return nullptr;
   }
 
-  size_t cur_idx = AtomicAccess::fetch_then_add(&_size, 1u);
+  size_t cur_idx = _size.fetch_then_add(1u);
 
   if (cur_idx >= _max_capacity) {
     return nullptr;
   }
 
   size_t bucket = get_bucket(cur_idx);
-  if (AtomicAccess::load_acquire(&_buckets[bucket]) == nullptr) {
+  if (_buckets[bucket].load_acquire() == nullptr) {
     if (!_should_grow) {
       // Prefer to restart the CM.
       return nullptr;
     }
 
     MutexLocker x(G1MarkStackChunkList_lock, Mutex::_no_safepoint_check_flag);
-    if (AtomicAccess::load_acquire(&_buckets[bucket]) == nullptr) {
+    if (_buckets[bucket].load_acquire() == nullptr) {
       size_t desired_capacity = bucket_size(bucket) * 2;
       if (!try_expand_to(desired_capacity)) {
         return nullptr;
@@ -175,7 +178,7 @@ G1CMMarkStack::TaskQueueEntryChunk* G1CMMarkStack::ChunkAllocator::allocate_new_
   }
 
   size_t bucket_idx = get_bucket_index(cur_idx);
-  TaskQueueEntryChunk* result = ::new (&_buckets[bucket][bucket_idx]) TaskQueueEntryChunk;
+  TaskQueueEntryChunk* result = ::new (&_buckets[bucket].load_relaxed()[bucket_idx]) TaskQueueEntryChunk;
   result->next = nullptr;
   return result;
 }
@@ -197,10 +200,10 @@ bool G1CMMarkStack::ChunkAllocator::initialize(size_t initial_capacity, size_t m
   _max_capacity = max_capacity;
   _num_buckets  = get_bucket(_max_capacity) + 1;
 
-  _buckets = NEW_C_HEAP_ARRAY(TaskQueueEntryChunk*, _num_buckets, mtGC);
+  _buckets = NEW_C_HEAP_ARRAY(Atomic<TaskQueueEntryChunk*>, _num_buckets, mtGC);
 
   for (size_t i = 0; i < _num_buckets; i++) {
-    _buckets[i] = nullptr;
+    _buckets[i].store_relaxed(nullptr);
   }
 
   size_t new_capacity = bucket_size(0);
@@ -240,9 +243,9 @@ G1CMMarkStack::ChunkAllocator::~ChunkAllocator() {
   }
 
   for (size_t i = 0; i < _num_buckets; i++) {
-    if (_buckets[i] != nullptr) {
-      MmapArrayAllocator<TaskQueueEntryChunk>::free(_buckets[i],  bucket_size(i));
-      _buckets[i] = nullptr;
+    if (_buckets[i].load_relaxed() != nullptr) {
+      MmapArrayAllocator<TaskQueueEntryChunk>::free(_buckets[i].load_relaxed(),  bucket_size(i));
+      _buckets[i].store_relaxed(nullptr);
     }
   }
 
@@ -259,7 +262,7 @@ bool G1CMMarkStack::ChunkAllocator::reserve(size_t new_capacity) {
   // and the new capacity (new_capacity). This step ensures that there are no gaps in the
   // array and that the capacity accurately reflects the reserved memory.
   for (; i <= highest_bucket; i++) {
-    if (AtomicAccess::load_acquire(&_buckets[i]) != nullptr) {
+    if (_buckets[i].load_acquire() != nullptr) {
       continue; // Skip over already allocated buckets.
     }
 
@@ -279,7 +282,7 @@ bool G1CMMarkStack::ChunkAllocator::reserve(size_t new_capacity) {
       return false;
     }
     _capacity += bucket_capacity;
-    AtomicAccess::release_store(&_buckets[i], bucket_base);
+    _buckets[i].release_store(bucket_base);
   }
   return true;
 }
@@ -288,9 +291,9 @@ void G1CMMarkStack::expand() {
   _chunk_allocator.try_expand();
 }
 
-void G1CMMarkStack::add_chunk_to_list(TaskQueueEntryChunk* volatile* list, TaskQueueEntryChunk* elem) {
-  elem->next = *list;
-  *list = elem;
+void G1CMMarkStack::add_chunk_to_list(Atomic<TaskQueueEntryChunk*>* list, TaskQueueEntryChunk* elem) {
+  elem->next = list->load_relaxed();
+  list->store_relaxed(elem);
 }
 
 void G1CMMarkStack::add_chunk_to_chunk_list(TaskQueueEntryChunk* elem) {
@@ -304,10 +307,10 @@ void G1CMMarkStack::add_chunk_to_free_list(TaskQueueEntryChunk* elem) {
   add_chunk_to_list(&_free_list, elem);
 }
 
-G1CMMarkStack::TaskQueueEntryChunk* G1CMMarkStack::remove_chunk_from_list(TaskQueueEntryChunk* volatile* list) {
-  TaskQueueEntryChunk* result = *list;
+G1CMMarkStack::TaskQueueEntryChunk* G1CMMarkStack::remove_chunk_from_list(Atomic<TaskQueueEntryChunk*>* list) {
+  TaskQueueEntryChunk* result = list->load_relaxed();
   if (result != nullptr) {
-    *list = (*list)->next;
+    list->store_relaxed(list->load_relaxed()->next);
   }
   return result;
 }
@@ -361,8 +364,8 @@ bool G1CMMarkStack::par_pop_chunk(G1TaskQueueEntry* ptr_arr) {
 
 void G1CMMarkStack::set_empty() {
   _chunks_in_chunk_list = 0;
-  _chunk_list = nullptr;
-  _free_list = nullptr;
+  _chunk_list.store_relaxed(nullptr);
+  _free_list.store_relaxed(nullptr);
   _chunk_allocator.reset();
 }
 
@@ -490,6 +493,7 @@ G1ConcurrentMark::G1ConcurrentMark(G1CollectedHeap* g1h,
 
   _task_queues(new G1CMTaskQueueSet(_max_num_tasks)),
   _terminator(_max_num_tasks, _task_queues),
+  _partial_array_state_manager(new PartialArrayStateManager(_max_num_tasks)),
 
   _first_overflow_barrier_sync(),
   _second_overflow_barrier_sync(),
@@ -556,6 +560,10 @@ G1ConcurrentMark::G1ConcurrentMark(G1CollectedHeap* g1h,
   reset_at_marking_complete();
 }
 
+PartialArrayStateManager* G1ConcurrentMark::partial_array_state_manager() const {
+  return _partial_array_state_manager;
+}
+
 void G1ConcurrentMark::reset() {
   _has_aborted = false;
 
@@ -650,7 +658,26 @@ void G1ConcurrentMark::set_concurrency_and_phase(uint active_tasks, bool concurr
   }
 }
 
+#if TASKQUEUE_STATS
+void G1ConcurrentMark::print_and_reset_taskqueue_stats() {
+
+  _task_queues->print_and_reset_taskqueue_stats("G1ConcurrentMark Oop Queue");
+
+  auto get_pa_stats = [&](uint i) {
+    return _tasks[i]->partial_array_task_stats();
+  };
+
+  PartialArrayTaskStats::log_set(_max_num_tasks, get_pa_stats,
+                                 "G1ConcurrentMark Partial Array Task Stats");
+
+  for (uint i = 0; i < _max_num_tasks; ++i) {
+    get_pa_stats(i)->reset();
+  }
+}
+#endif
+
 void G1ConcurrentMark::reset_at_marking_complete() {
+  TASKQUEUE_STATS_ONLY(print_and_reset_taskqueue_stats());
   // We set the global marking state to some default values when we're
   // not doing marking.
   reset_marking_for_restart();
@@ -804,11 +831,25 @@ void G1ConcurrentMark::cleanup_for_next_mark() {
 
   clear_bitmap(_concurrent_workers, true);
 
+  reset_partial_array_state_manager();
+
   // Repeat the asserts from above.
   guarantee(cm_thread()->in_progress(), "invariant");
   guarantee(!_g1h->collector_state()->mark_or_rebuild_in_progress(), "invariant");
 }
 
+void G1ConcurrentMark::reset_partial_array_state_manager() {
+  for (uint i = 0; i < _max_num_tasks; ++i) {
+    _tasks[i]->unregister_partial_array_splitter();
+  }
+
+  partial_array_state_manager()->reset();
+
+  for (uint i = 0; i < _max_num_tasks; ++i) {
+    _tasks[i]->register_partial_array_splitter();
+  }
+}
+
 void G1ConcurrentMark::clear_bitmap(WorkerThreads* workers) {
   assert_at_safepoint_on_vm_thread();
   // To avoid fragmentation the full collection requesting to clear the bitmap
@@ -1789,17 +1830,18 @@ public:
   { }
 
   void operator()(G1TaskQueueEntry task_entry) const {
-    if (task_entry.is_array_slice()) {
-      guarantee(_g1h->is_in_reserved(task_entry.slice()), "Slice " PTR_FORMAT " must be in heap.", p2i(task_entry.slice()));
+    if (task_entry.is_partial_array_state()) {
+      oop obj = task_entry.to_partial_array_state()->source();
+      guarantee(_g1h->is_in_reserved(obj), "Partial Array " PTR_FORMAT " must be in heap.", p2i(obj));
       return;
     }
-    guarantee(oopDesc::is_oop(task_entry.obj()),
+    guarantee(oopDesc::is_oop(task_entry.to_oop()),
               "Non-oop " PTR_FORMAT ", phase: %s, info: %d",
-              p2i(task_entry.obj()), _phase, _info);
-    G1HeapRegion* r = _g1h->heap_region_containing(task_entry.obj());
+              p2i(task_entry.to_oop()), _phase, _info);
+    G1HeapRegion* r = _g1h->heap_region_containing(task_entry.to_oop());
     guarantee(!(r->in_collection_set() || r->has_index_in_opt_cset()),
               "obj " PTR_FORMAT " from %s (%d) in region %u in (optional) collection set",
-              p2i(task_entry.obj()), _phase, _info, r->hrm_index());
+              p2i(task_entry.to_oop()), _phase, _info, r->hrm_index());
   }
 };
 
@@ -2055,6 +2097,17 @@ void G1CMTask::reset(G1CMBitMap* mark_bitmap) {
   _mark_stats_cache.reset();
 }
 
+void G1CMTask::register_partial_array_splitter() {
+
+  ::new (&_partial_array_splitter) PartialArraySplitter(_cm->partial_array_state_manager(),
+                                                        _cm->max_num_tasks(),
+                                                        ObjArrayMarkingStride);
+}
+
+void G1CMTask::unregister_partial_array_splitter() {
+  _partial_array_splitter.~PartialArraySplitter();
+}
+
 bool G1CMTask::should_exit_termination() {
   if (!regular_clock_call()) {
     return true;
@@ -2185,7 +2238,7 @@ bool G1CMTask::get_entries_from_global_stack() {
     if (task_entry.is_null()) {
       break;
     }
-    assert(task_entry.is_array_slice() || oopDesc::is_oop(task_entry.obj()), "Element " PTR_FORMAT " must be an array slice or oop", p2i(task_entry.obj()));
+    assert(task_entry.is_partial_array_state() || oopDesc::is_oop(task_entry.to_oop()), "Element " PTR_FORMAT " must be an array slice or oop", p2i(task_entry.to_oop()));
     bool success = _task_queue->push(task_entry);
     // We only call this when the local queue is empty or under a
     // given target limit. So, we do not expect this push to fail.
@@ -2216,7 +2269,7 @@ void G1CMTask::drain_local_queue(bool partially) {
     G1TaskQueueEntry entry;
     bool ret = _task_queue->pop_local(entry);
     while (ret) {
-      scan_task_entry(entry);
+      process_entry(entry, false /* stolen */);
       if (_task_queue->size() <= target_size || has_aborted()) {
         ret = false;
       } else {
@@ -2226,6 +2279,37 @@ void G1CMTask::drain_local_queue(bool partially) {
   }
 }
 
+size_t G1CMTask::start_partial_array_processing(oop obj) {
+  assert(should_be_sliced(obj), "Must be an array object %d and large %zu", obj->is_objArray(), obj->size());
+
+  objArrayOop obj_array = objArrayOop(obj);
+  size_t array_length = obj_array->length();
+
+  size_t initial_chunk_size = _partial_array_splitter.start(_task_queue, obj_array, nullptr, array_length);
+
+  // Mark objArray klass metadata
+  if (_cm_oop_closure->do_metadata()) {
+    _cm_oop_closure->do_klass(obj_array->klass());
+  }
+
+  process_array_chunk(obj_array, 0, initial_chunk_size);
+
+  // Include object header size
+  return objArrayOopDesc::object_size(checked_cast<int>(initial_chunk_size));
+}
+
+size_t G1CMTask::process_partial_array(const G1TaskQueueEntry& task, bool stolen) {
+  PartialArrayState* state = task.to_partial_array_state();
+  // Access state before release by claim().
+  objArrayOop obj = objArrayOop(state->source());
+
+  PartialArraySplitter::Claim claim =
+    _partial_array_splitter.claim(state, _task_queue, stolen);
+
+  process_array_chunk(obj, claim._start, claim._end);
+  return heap_word_size((claim._end - claim._start) * heapOopSize);
+}
+
 void G1CMTask::drain_global_stack(bool partially) {
   if (has_aborted()) {
     return;
@@ -2430,7 +2514,7 @@ void G1CMTask::attempt_stealing() {
   while (!has_aborted()) {
     G1TaskQueueEntry entry;
     if (_cm->try_stealing(_worker_id, entry)) {
-      scan_task_entry(entry);
+      process_entry(entry, true /* stolen */);
 
       // And since we're towards the end, let's totally drain the
       // local queue and global stack.
@@ -2759,12 +2843,12 @@ G1CMTask::G1CMTask(uint worker_id,
                    G1ConcurrentMark* cm,
                    G1CMTaskQueue* task_queue,
                    G1RegionMarkStats* mark_stats) :
-  _objArray_processor(this),
   _worker_id(worker_id),
   _g1h(G1CollectedHeap::heap()),
   _cm(cm),
   _mark_bitmap(nullptr),
   _task_queue(task_queue),
+  _partial_array_splitter(_cm->partial_array_state_manager(), _cm->max_num_tasks(), ObjArrayMarkingStride),
   _mark_stats_cache(mark_stats, G1RegionMarkStatsCache::RegionMarkStatsCacheSize),
   _calls(0),
   _time_target_ms(0.0),

src/hotspot/share/gc/g1/g1ConcurrentMark.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2001, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2001, 2026, 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
@@ -26,17 +26,20 @@
 #define SHARE_GC_G1_G1CONCURRENTMARK_HPP
 
 #include "gc/g1/g1ConcurrentMarkBitMap.hpp"
-#include "gc/g1/g1ConcurrentMarkObjArrayProcessor.hpp"
 #include "gc/g1/g1HeapRegionSet.hpp"
 #include "gc/g1/g1HeapVerifier.hpp"
 #include "gc/g1/g1RegionMarkStatsCache.hpp"
 #include "gc/shared/gcCause.hpp"
+#include "gc/shared/partialArraySplitter.hpp"
+#include "gc/shared/partialArrayState.hpp"
+#include "gc/shared/partialArrayTaskStats.hpp"
 #include "gc/shared/taskqueue.hpp"
 #include "gc/shared/taskTerminator.hpp"
 #include "gc/shared/verifyOption.hpp"
 #include "gc/shared/workerThread.hpp"
 #include "gc/shared/workerUtils.hpp"
 #include "memory/allocation.hpp"
+#include "runtime/atomic.hpp"
 #include "utilities/compilerWarnings.hpp"
 #include "utilities/numberSeq.hpp"
 
@@ -53,41 +56,7 @@ class G1RegionToSpaceMapper;
 class G1SurvivorRegions;
 class ThreadClosure;
 
-// This is a container class for either an oop or a continuation address for
-// mark stack entries. Both are pushed onto the mark stack.
-class G1TaskQueueEntry {
-private:
-  void* _holder;
-
-  static const uintptr_t ArraySliceBit = 1;
-
-  G1TaskQueueEntry(oop obj) : _holder(obj) {
-    assert(_holder != nullptr, "Not allowed to set null task queue element");
-  }
-  G1TaskQueueEntry(HeapWord* addr) : _holder((void*)((uintptr_t)addr | ArraySliceBit)) { }
-public:
-
-  G1TaskQueueEntry() : _holder(nullptr) { }
-  // Trivially copyable, for use in GenericTaskQueue.
-
-  static G1TaskQueueEntry from_slice(HeapWord* what) { return G1TaskQueueEntry(what); }
-  static G1TaskQueueEntry from_oop(oop obj) { return G1TaskQueueEntry(obj); }
-
-  oop obj() const {
-    assert(!is_array_slice(), "Trying to read array slice " PTR_FORMAT " as oop", p2i(_holder));
-    return cast_to_oop(_holder);
-  }
-
-  HeapWord* slice() const {
-    assert(is_array_slice(), "Trying to read oop " PTR_FORMAT " as array slice", p2i(_holder));
-    return (HeapWord*)((uintptr_t)_holder & ~ArraySliceBit);
-  }
-
-  bool is_oop() const { return !is_array_slice(); }
-  bool is_array_slice() const { return ((uintptr_t)_holder & ArraySliceBit) != 0; }
-  bool is_null() const { return _holder == nullptr; }
-};
-
+typedef ScannerTask G1TaskQueueEntry;
 typedef GenericTaskQueue<G1TaskQueueEntry, mtGC> G1CMTaskQueue;
 typedef GenericTaskQueueSet<G1CMTaskQueue, mtGC> G1CMTaskQueueSet;
 
@@ -172,9 +141,9 @@ private:
     size_t _capacity;
     size_t _num_buckets;
     bool _should_grow;
-    TaskQueueEntryChunk* volatile* _buckets;
+    Atomic<TaskQueueEntryChunk*>* _buckets;
     char _pad0[DEFAULT_PADDING_SIZE];
-    volatile size_t _size;
+    Atomic<size_t> _size;
     char _pad4[DEFAULT_PADDING_SIZE - sizeof(size_t)];
 
     size_t bucket_size(size_t bucket) {
@@ -212,7 +181,7 @@ private:
     bool initialize(size_t initial_capacity, size_t max_capacity);
 
     void reset() {
-      _size = 0;
+      _size.store_relaxed(0);
       _should_grow = false;
     }
 
@@ -241,17 +210,17 @@ private:
   ChunkAllocator _chunk_allocator;
 
   char _pad0[DEFAULT_PADDING_SIZE];
-  TaskQueueEntryChunk* volatile _free_list;  // Linked list of free chunks that can be allocated by users.
+  Atomic<TaskQueueEntryChunk*> _free_list;  // Linked list of free chunks that can be allocated by users.
   char _pad1[DEFAULT_PADDING_SIZE - sizeof(TaskQueueEntryChunk*)];
-  TaskQueueEntryChunk* volatile _chunk_list; // List of chunks currently containing data.
+  Atomic<TaskQueueEntryChunk*> _chunk_list; // List of chunks currently containing data.
   volatile size_t _chunks_in_chunk_list;
   char _pad2[DEFAULT_PADDING_SIZE - sizeof(TaskQueueEntryChunk*) - sizeof(size_t)];
 
   // Atomically add the given chunk to the list.
-  void add_chunk_to_list(TaskQueueEntryChunk* volatile* list, TaskQueueEntryChunk* elem);
+  void add_chunk_to_list(Atomic<TaskQueueEntryChunk*>* list, TaskQueueEntryChunk* elem);
   // Atomically remove and return a chunk from the given list. Returns null if the
   // list is empty.
-  TaskQueueEntryChunk* remove_chunk_from_list(TaskQueueEntryChunk* volatile* list);
+  TaskQueueEntryChunk* remove_chunk_from_list(Atomic<TaskQueueEntryChunk*>* list);
 
   void add_chunk_to_chunk_list(TaskQueueEntryChunk* elem);
   void add_chunk_to_free_list(TaskQueueEntryChunk* elem);
@@ -283,7 +252,7 @@ private:
 
   // Return whether the chunk list is empty. Racy due to unsynchronized access to
   // _chunk_list.
-  bool is_empty() const { return _chunk_list == nullptr; }
+  bool is_empty() const { return _chunk_list.load_relaxed() == nullptr; }
 
   size_t capacity() const  { return _chunk_allocator.capacity(); }
 
@@ -411,6 +380,8 @@ class G1ConcurrentMark : public CHeapObj<mtGC> {
   G1CMTaskQueueSet*       _task_queues; // Task queue set
   TaskTerminator          _terminator;  // For termination
 
+  PartialArrayStateManager* _partial_array_state_manager;
+
   // Two sync barriers that are used to synchronize tasks when an
   // overflow occurs. The algorithm is the following. All tasks enter
   // the first one to ensure that they have all stopped manipulating
@@ -488,6 +459,8 @@ class G1ConcurrentMark : public CHeapObj<mtGC> {
   // Prints all gathered CM-related statistics
   void print_stats();
 
+  void print_and_reset_taskqueue_stats();
+
   HeapWord*           finger()       { return _finger;   }
   bool                concurrent()   { return _concurrent; }
   uint                active_tasks() { return _num_active_tasks; }
@@ -556,14 +529,14 @@ public:
   // mark_in_bitmap call. Updates various statistics data.
   void add_to_liveness(uint worker_id, oop const obj, size_t size);
   // Did the last marking find a live object between bottom and TAMS?
-  bool contains_live_object(uint region) const { return _region_mark_stats[region]._live_words != 0; }
+  bool contains_live_object(uint region) const { return _region_mark_stats[region].live_words() != 0; }
   // Live bytes in the given region as determined by concurrent marking, i.e. the amount of
   // live bytes between bottom and TAMS.
-  size_t live_bytes(uint region) const { return _region_mark_stats[region]._live_words * HeapWordSize; }
+  size_t live_bytes(uint region) const { return _region_mark_stats[region].live_words() * HeapWordSize; }
   // Set live bytes for concurrent marking.
-  void set_live_bytes(uint region, size_t live_bytes) { _region_mark_stats[region]._live_words = live_bytes / HeapWordSize; }
+  void set_live_bytes(uint region, size_t live_bytes) { _region_mark_stats[region]._live_words.store_relaxed(live_bytes / HeapWordSize); }
   // Approximate number of incoming references found during marking.
-  size_t incoming_refs(uint region) const { return _region_mark_stats[region]._incoming_refs; }
+  size_t incoming_refs(uint region) const { return _region_mark_stats[region].incoming_refs(); }
 
   // Update the TAMS for the given region to the current top.
   inline void update_top_at_mark_start(G1HeapRegion* r);
@@ -582,6 +555,8 @@ public:
 
   uint worker_id_offset() const { return _worker_id_offset; }
 
+  uint max_num_tasks() const {return _max_num_tasks; }
+
   // Clear statistics gathered during the concurrent cycle for the given region after
   // it has been reclaimed.
   void clear_statistics(G1HeapRegion* r);
@@ -631,6 +606,8 @@ public:
   // Calculates the number of concurrent GC threads to be used in the marking phase.
   uint calc_active_marking_workers();
 
+  PartialArrayStateManager* partial_array_state_manager() const;
+
   // Resets the global marking data structures, as well as the
   // task local ones; should be called during concurrent start.
   void reset();
@@ -642,6 +619,10 @@ public:
   // to be called concurrently to the mutator. It will yield to safepoint requests.
   void cleanup_for_next_mark();
 
+  // Recycle the memory that has been requested by allocators associated with
+  // this manager.
+  void reset_partial_array_state_manager();
+
   // Clear the next marking bitmap during safepoint.
   void clear_bitmap(WorkerThreads* workers);
 
@@ -732,14 +713,13 @@ private:
     refs_reached_period           = 1024,
   };
 
-  G1CMObjArrayProcessor       _objArray_processor;
-
   uint                        _worker_id;
   G1CollectedHeap*            _g1h;
   G1ConcurrentMark*           _cm;
   G1CMBitMap*                 _mark_bitmap;
   // the task queue of this task
   G1CMTaskQueue*              _task_queue;
+  PartialArraySplitter        _partial_array_splitter;
 
   G1RegionMarkStatsCache      _mark_stats_cache;
   // Number of calls to this task
@@ -850,13 +830,24 @@ private:
   // mark bitmap scan, and so needs to be pushed onto the mark stack.
   bool is_below_finger(oop obj, HeapWord* global_finger) const;
 
-  template<bool scan> void process_grey_task_entry(G1TaskQueueEntry task_entry);
+  template<bool scan> void process_grey_task_entry(G1TaskQueueEntry task_entry, bool stolen);
+
+  static bool should_be_sliced(oop obj);
+  // Start processing the given objArrayOop by first pushing its continuations and
+  // then scanning the first chunk including the header.
+  size_t start_partial_array_processing(oop obj);
+  // Process the given continuation. Returns the number of words scanned.
+  size_t process_partial_array(const G1TaskQueueEntry& task, bool stolen);
+  // Apply the closure to the given range of elements in the objArray.
+  inline void process_array_chunk(objArrayOop obj, size_t start, size_t end);
 public:
-  // Apply the closure on the given area of the objArray. Return the number of words
-  // scanned.
-  inline size_t scan_objArray(objArrayOop obj, MemRegion mr);
   // Resets the task; should be called right at the beginning of a marking phase.
   void reset(G1CMBitMap* mark_bitmap);
+  // Register/unregister Partial Array Splitter Allocator with the PartialArrayStateManager.
+  // This allows us to discard memory arenas used for partial object array states at the end
+  // of a concurrent mark cycle.
+  void register_partial_array_splitter();
+  void unregister_partial_array_splitter();
   // Clears all the fields that correspond to a claimed region.
   void clear_region_fields();
 
@@ -912,7 +903,7 @@ public:
   inline bool deal_with_reference(T* p);
 
   // Scans an object and visits its children.
-  inline void scan_task_entry(G1TaskQueueEntry task_entry);
+  inline void process_entry(G1TaskQueueEntry task_entry, bool stolen);
 
   // Pushes an object on the local queue.
   inline void push(G1TaskQueueEntry task_entry);
@@ -957,6 +948,11 @@ public:
   Pair<size_t, size_t> flush_mark_stats_cache();
   // Prints statistics associated with this task
   void print_stats();
+#if TASKQUEUE_STATS
+  PartialArrayTaskStats* partial_array_task_stats() {
+    return _partial_array_splitter.stats();
+  }
+#endif
 };
 
 // Class that's used to to print out per-region liveness

src/hotspot/share/gc/g1/g1ConcurrentMark.inline.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2001, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2001, 2026, 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
@@ -29,7 +29,6 @@
 
 #include "gc/g1/g1CollectedHeap.inline.hpp"
 #include "gc/g1/g1ConcurrentMarkBitMap.inline.hpp"
-#include "gc/g1/g1ConcurrentMarkObjArrayProcessor.inline.hpp"
 #include "gc/g1/g1HeapRegion.hpp"
 #include "gc/g1/g1HeapRegionRemSet.inline.hpp"
 #include "gc/g1/g1OopClosures.inline.hpp"
@@ -39,6 +38,7 @@
 #include "gc/shared/suspendibleThreadSet.hpp"
 #include "gc/shared/taskqueue.inline.hpp"
 #include "utilities/bitMap.inline.hpp"
+#include "utilities/checkedCast.hpp"
 
 inline bool G1CMIsAliveClosure::do_object_b(oop obj) {
   // Check whether the passed in object is null. During discovery the referent
@@ -90,7 +90,7 @@ inline void G1CMMarkStack::iterate(Fn fn) const {
 
   size_t num_chunks = 0;
 
-  TaskQueueEntryChunk* cur = _chunk_list;
+  TaskQueueEntryChunk* cur = _chunk_list.load_relaxed();
   while (cur != nullptr) {
     guarantee(num_chunks <= _chunks_in_chunk_list, "Found %zu oop chunks which is more than there should be", num_chunks);
 
@@ -107,13 +107,15 @@ inline void G1CMMarkStack::iterate(Fn fn) const {
 #endif
 
 // It scans an object and visits its children.
-inline void G1CMTask::scan_task_entry(G1TaskQueueEntry task_entry) { process_grey_task_entry<true>(task_entry); }
+inline void G1CMTask::process_entry(G1TaskQueueEntry task_entry, bool stolen) {
+  process_grey_task_entry<true>(task_entry, stolen);
+}
 
 inline void G1CMTask::push(G1TaskQueueEntry task_entry) {
-  assert(task_entry.is_array_slice() || _g1h->is_in_reserved(task_entry.obj()), "invariant");
-  assert(task_entry.is_array_slice() || !_g1h->is_on_master_free_list(
-              _g1h->heap_region_containing(task_entry.obj())), "invariant");
-  assert(task_entry.is_array_slice() || _mark_bitmap->is_marked(cast_from_oop<HeapWord*>(task_entry.obj())), "invariant");
+  assert(task_entry.is_partial_array_state() || _g1h->is_in_reserved(task_entry.to_oop()), "invariant");
+  assert(task_entry.is_partial_array_state() || !_g1h->is_on_master_free_list(
+              _g1h->heap_region_containing(task_entry.to_oop())), "invariant");
+  assert(task_entry.is_partial_array_state() || _mark_bitmap->is_marked(cast_from_oop<HeapWord*>(task_entry.to_oop())), "invariant");
 
   if (!_task_queue->push(task_entry)) {
     // The local task queue looks full. We need to push some entries
@@ -159,29 +161,34 @@ inline bool G1CMTask::is_below_finger(oop obj, HeapWord* global_finger) const {
 }
 
 template<bool scan>
-inline void G1CMTask::process_grey_task_entry(G1TaskQueueEntry task_entry) {
-  assert(scan || (task_entry.is_oop() && task_entry.obj()->is_typeArray()), "Skipping scan of grey non-typeArray");
-  assert(task_entry.is_array_slice() || _mark_bitmap->is_marked(cast_from_oop<HeapWord*>(task_entry.obj())),
+inline void G1CMTask::process_grey_task_entry(G1TaskQueueEntry task_entry, bool stolen) {
+  assert(scan || (!task_entry.is_partial_array_state() && task_entry.to_oop()->is_typeArray()), "Skipping scan of grey non-typeArray");
+  assert(task_entry.is_partial_array_state() || _mark_bitmap->is_marked(cast_from_oop<HeapWord*>(task_entry.to_oop())),
          "Any stolen object should be a slice or marked");
 
   if (scan) {
-    if (task_entry.is_array_slice()) {
-      _words_scanned += _objArray_processor.process_slice(task_entry.slice());
+    if (task_entry.is_partial_array_state()) {
+      _words_scanned += process_partial_array(task_entry, stolen);
     } else {
-      oop obj = task_entry.obj();
-      if (G1CMObjArrayProcessor::should_be_sliced(obj)) {
-        _words_scanned += _objArray_processor.process_obj(obj);
+      oop obj = task_entry.to_oop();
+      if (should_be_sliced(obj)) {
+        _words_scanned += start_partial_array_processing(obj);
       } else {
-        _words_scanned += obj->oop_iterate_size(_cm_oop_closure);;
+        _words_scanned += obj->oop_iterate_size(_cm_oop_closure);
       }
     }
   }
   check_limits();
 }
 
-inline size_t G1CMTask::scan_objArray(objArrayOop obj, MemRegion mr) {
-  obj->oop_iterate(_cm_oop_closure, mr);
-  return mr.word_size();
+inline bool G1CMTask::should_be_sliced(oop obj) {
+  return obj->is_objArray() && ((objArrayOop)obj)->length() >= (int)ObjArrayMarkingStride;
+}
+
+inline void G1CMTask::process_array_chunk(objArrayOop obj, size_t start, size_t end) {
+  obj->oop_iterate_elements_range(_cm_oop_closure,
+                                  checked_cast<int>(start),
+                                  checked_cast<int>(end));
 }
 
 inline void G1ConcurrentMark::update_top_at_mark_start(G1HeapRegion* r) {
@@ -265,7 +272,7 @@ inline bool G1CMTask::make_reference_grey(oop obj) {
   // be pushed on the stack. So, some duplicate work, but no
   // correctness problems.
   if (is_below_finger(obj, global_finger)) {
-    G1TaskQueueEntry entry = G1TaskQueueEntry::from_oop(obj);
+    G1TaskQueueEntry entry(obj);
     if (obj->is_typeArray()) {
       // Immediately process arrays of primitive types, rather
       // than pushing on the mark stack.  This keeps us from
@@ -277,7 +284,7 @@ inline bool G1CMTask::make_reference_grey(oop obj) {
       // by only doing a bookkeeping update and avoiding the
       // actual scan of the object - a typeArray contains no
       // references, and the metadata is built-in.
-      process_grey_task_entry<false>(entry);
+      process_grey_task_entry<false>(entry, false /* stolen */);
     } else {
       push(entry);
     }

src/hotspot/share/gc/g1/g1ConcurrentMarkObjArrayProcessor.cpp

@@ -1,80 +0,0 @@
-/*
- * Copyright (c) 2016, 2025, 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
- * under the terms of the GNU General Public License version 2 only, as
- * published by the Free Software Foundation.
- *
- * This code is distributed in the hope that it will be useful, but WITHOUT
- * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
- * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
- * version 2 for more details (a copy is included in the LICENSE file that
- * accompanied this code).
- *
- * You should have received a copy of the GNU General Public License version
- * 2 along with this work; if not, write to the Free Software Foundation,
- * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
- *
- * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
- * or visit www.oracle.com if you need additional information or have any
- * questions.
- *
- */
-
-#include "gc/g1/g1CollectedHeap.inline.hpp"
-#include "gc/g1/g1ConcurrentMark.inline.hpp"
-#include "gc/g1/g1ConcurrentMarkObjArrayProcessor.inline.hpp"
-#include "gc/g1/g1HeapRegion.inline.hpp"
-#include "gc/shared/gc_globals.hpp"
-#include "memory/memRegion.hpp"
-#include "utilities/globalDefinitions.hpp"
-
-void G1CMObjArrayProcessor::push_array_slice(HeapWord* what) {
-  _task->push(G1TaskQueueEntry::from_slice(what));
-}
-
-size_t G1CMObjArrayProcessor::process_array_slice(objArrayOop obj, HeapWord* start_from, size_t remaining) {
-  size_t words_to_scan = MIN2(remaining, (size_t)ObjArrayMarkingStride);
-
-  if (remaining > ObjArrayMarkingStride) {
-    push_array_slice(start_from + ObjArrayMarkingStride);
-  }
-
-  // Then process current area.
-  MemRegion mr(start_from, words_to_scan);
-  return _task->scan_objArray(obj, mr);
-}
-
-size_t G1CMObjArrayProcessor::process_obj(oop obj) {
-  assert(should_be_sliced(obj), "Must be an array object %d and large %zu", obj->is_objArray(), obj->size());
-
-  return process_array_slice(objArrayOop(obj), cast_from_oop<HeapWord*>(obj), objArrayOop(obj)->size());
-}
-
-size_t G1CMObjArrayProcessor::process_slice(HeapWord* slice) {
-
-  // Find the start address of the objArrayOop.
-  // Shortcut the BOT access if the given address is from a humongous object. The BOT
-  // slide is fast enough for "smaller" objects in non-humongous regions, but is slower
-  // than directly using heap region table.
-  G1CollectedHeap* g1h = G1CollectedHeap::heap();
-  G1HeapRegion* r = g1h->heap_region_containing(slice);
-
-  HeapWord* const start_address = r->is_humongous() ?
-                                  r->humongous_start_region()->bottom() :
-                                  r->block_start(slice);
-
-  assert(cast_to_oop(start_address)->is_objArray(), "Address " PTR_FORMAT " does not refer to an object array ", p2i(start_address));
-  assert(start_address < slice,
-         "Object start address " PTR_FORMAT " must be smaller than decoded address " PTR_FORMAT,
-         p2i(start_address),
-         p2i(slice));
-
-  objArrayOop objArray = objArrayOop(cast_to_oop(start_address));
-
-  size_t already_scanned = pointer_delta(slice, start_address);
-  size_t remaining = objArray->size() - already_scanned;
-
-  return process_array_slice(objArray, slice, remaining);
-}

src/hotspot/share/gc/g1/g1ConcurrentMarkObjArrayProcessor.hpp

@@ -1,59 +0,0 @@
-/*
- * Copyright (c) 2016, 2019, 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
- * under the terms of the GNU General Public License version 2 only, as
- * published by the Free Software Foundation.
- *
- * This code is distributed in the hope that it will be useful, but WITHOUT
- * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
- * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
- * version 2 for more details (a copy is included in the LICENSE file that
- * accompanied this code).
- *
- * You should have received a copy of the GNU General Public License version
- * 2 along with this work; if not, write to the Free Software Foundation,
- * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
- *
- * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
- * or visit www.oracle.com if you need additional information or have any
- * questions.
- *
- */
-
-#ifndef SHARE_GC_G1_G1CONCURRENTMARKOBJARRAYPROCESSOR_HPP
-#define SHARE_GC_G1_G1CONCURRENTMARKOBJARRAYPROCESSOR_HPP
-
-#include "oops/oopsHierarchy.hpp"
-
-class G1CMTask;
-
-// Helper class to mark through large objArrays during marking in an efficient way.
-// Instead of pushing large object arrays, we push continuations onto the
-// mark stack. These continuations are identified by having their LSB set.
-// This allows incremental processing of large objects.
-class G1CMObjArrayProcessor {
-private:
-  // Reference to the task for doing the actual work.
-  G1CMTask* _task;
-
-  // Push the continuation at the given address onto the mark stack.
-  void push_array_slice(HeapWord* addr);
-
-  // Process (apply the closure) on the given continuation of the given objArray.
-  size_t process_array_slice(objArrayOop const obj, HeapWord* start_from, size_t remaining);
-public:
-  static bool should_be_sliced(oop obj);
-
-  G1CMObjArrayProcessor(G1CMTask* task) : _task(task) {
-  }
-
-  // Process the given continuation. Returns the number of words scanned.
-  size_t process_slice(HeapWord* slice);
-  // Start processing the given objArrayOop by scanning the header and pushing its
-  // continuation.
-  size_t process_obj(oop obj);
-};
-
-#endif // SHARE_GC_G1_G1CONCURRENTMARKOBJARRAYPROCESSOR_HPP

src/hotspot/share/gc/g1/g1ConcurrentMarkRemarkTasks.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2025, 2026, 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
@@ -30,7 +30,6 @@
 #include "gc/g1/g1HeapRegionPrinter.hpp"
 #include "gc/g1/g1RemSetTrackingPolicy.hpp"
 #include "logging/log.hpp"
-#include "runtime/atomicAccess.hpp"
 #include "runtime/mutexLocker.hpp"
 
 struct G1UpdateRegionLivenessAndSelectForRebuildTask::G1OnRegionClosure : public G1HeapRegionClosure {
@@ -154,7 +153,7 @@ void G1UpdateRegionLivenessAndSelectForRebuildTask::work(uint worker_id) {
   G1OnRegionClosure on_region_cl(_g1h, _cm, &local_cleanup_list);
   _g1h->heap_region_par_iterate_from_worker_offset(&on_region_cl, &_hrclaimer, worker_id);
 
-  AtomicAccess::add(&_total_selected_for_rebuild, on_region_cl._num_selected_for_rebuild);
+  _total_selected_for_rebuild.add_then_fetch(on_region_cl._num_selected_for_rebuild);
 
   // Update the old/humongous region sets
   _g1h->remove_from_old_gen_sets(on_region_cl._num_old_regions_removed,

src/hotspot/share/gc/g1/g1ConcurrentMarkRemarkTasks.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2025, 2026, 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
@@ -29,6 +29,7 @@
 #include "gc/g1/g1HeapRegionManager.hpp"
 #include "gc/g1/g1HeapRegionSet.hpp"
 #include "gc/shared/workerThread.hpp"
+#include "runtime/atomic.hpp"
 
 class G1CollectedHeap;
 class G1ConcurrentMark;
@@ -41,7 +42,7 @@ class G1UpdateRegionLivenessAndSelectForRebuildTask : public WorkerTask {
   G1ConcurrentMark* _cm;
   G1HeapRegionClaimer _hrclaimer;
 
-  uint volatile _total_selected_for_rebuild;
+  Atomic<uint> _total_selected_for_rebuild;
 
   // Reclaimed empty regions
   G1FreeRegionList _cleanup_list;
@@ -57,7 +58,9 @@ public:
 
   void work(uint worker_id) override;
 
-  uint total_selected_for_rebuild() const { return _total_selected_for_rebuild; }
+  uint total_selected_for_rebuild() const {
+    return _total_selected_for_rebuild.load_relaxed();
+  }
 
   static uint desired_num_workers(uint num_regions);
 };

src/hotspot/share/gc/g1/g1ConcurrentRefine.cpp

@@ -28,6 +28,7 @@
 #include "gc/g1/g1CollectedHeap.inline.hpp"
 #include "gc/g1/g1CollectionSet.hpp"
 #include "gc/g1/g1ConcurrentRefine.hpp"
+#include "gc/g1/g1ConcurrentRefineStats.inline.hpp"
 #include "gc/g1/g1ConcurrentRefineSweepTask.hpp"
 #include "gc/g1/g1ConcurrentRefineThread.hpp"
 #include "gc/g1/g1HeapRegion.inline.hpp"

src/hotspot/share/gc/g1/g1ConcurrentRefineStats.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2020, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2020, 2026, 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
@@ -22,7 +22,7 @@
  *
  */
 
-#include "gc/g1/g1ConcurrentRefineStats.hpp"
+#include "gc/g1/g1ConcurrentRefineStats.inline.hpp"
 #include "runtime/atomicAccess.hpp"
 #include "runtime/timer.hpp"
 
@@ -39,19 +39,27 @@ G1ConcurrentRefineStats::G1ConcurrentRefineStats() :
 {}
 
 void G1ConcurrentRefineStats::add_atomic(G1ConcurrentRefineStats* other) {
-  AtomicAccess::add(&_sweep_duration, other->_sweep_duration, memory_order_relaxed);
-  AtomicAccess::add(&_yield_during_sweep_duration, other->_yield_during_sweep_duration, memory_order_relaxed);
+  _sweep_duration.add_then_fetch(other->_sweep_duration.load_relaxed(), memory_order_relaxed);
+  _yield_during_sweep_duration.add_then_fetch(other->yield_during_sweep_duration(), memory_order_relaxed);
 
-  AtomicAccess::add(&_cards_scanned, other->_cards_scanned, memory_order_relaxed);
-  AtomicAccess::add(&_cards_clean, other->_cards_clean, memory_order_relaxed);
-  AtomicAccess::add(&_cards_not_parsable, other->_cards_not_parsable, memory_order_relaxed);
-  AtomicAccess::add(&_cards_already_refer_to_cset, other->_cards_already_refer_to_cset, memory_order_relaxed);
-  AtomicAccess::add(&_cards_refer_to_cset, other->_cards_refer_to_cset, memory_order_relaxed);
-  AtomicAccess::add(&_cards_no_cross_region, other->_cards_no_cross_region, memory_order_relaxed);
+  _cards_scanned.add_then_fetch(other->cards_scanned(), memory_order_relaxed);
+  _cards_clean.add_then_fetch(other->cards_clean(), memory_order_relaxed);
+  _cards_not_parsable.add_then_fetch(other->cards_not_parsable(), memory_order_relaxed);
+  _cards_already_refer_to_cset.add_then_fetch(other->cards_already_refer_to_cset(), memory_order_relaxed);
+  _cards_refer_to_cset.add_then_fetch(other->cards_refer_to_cset(), memory_order_relaxed);
+  _cards_no_cross_region.add_then_fetch(other->cards_no_cross_region(), memory_order_relaxed);
 
-  AtomicAccess::add(&_refine_duration, other->_refine_duration, memory_order_relaxed);
+  _refine_duration.add_then_fetch(other->refine_duration(), memory_order_relaxed);
 }
 
 void G1ConcurrentRefineStats::reset() {
-  *this = G1ConcurrentRefineStats();
+  _sweep_duration.store_relaxed(0);
+  _yield_during_sweep_duration.store_relaxed(0);
+  _cards_scanned.store_relaxed(0);
+  _cards_clean.store_relaxed(0);
+  _cards_not_parsable.store_relaxed(0);
+  _cards_already_refer_to_cset.store_relaxed(0);
+  _cards_refer_to_cset.store_relaxed(0);
+  _cards_no_cross_region.store_relaxed(0);
+  _refine_duration.store_relaxed(0);
 }

src/hotspot/share/gc/g1/g1ConcurrentRefineStats.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2020, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2020, 2026, 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
@@ -26,61 +26,61 @@
 #define SHARE_GC_G1_G1CONCURRENTREFINESTATS_HPP
 
 #include "memory/allocation.hpp"
+#include "runtime/atomic.hpp"
 #include "utilities/globalDefinitions.hpp"
-#include "utilities/ticks.hpp"
 
 // Collection of statistics for concurrent refinement processing.
 // Used for collecting per-thread statistics and for summaries over a
 // collection of threads.
 class G1ConcurrentRefineStats : public CHeapObj<mtGC> {
-  jlong _sweep_duration;              // Time spent sweeping the table finding non-clean cards
-                                      // and refining them.
-  jlong _yield_during_sweep_duration; // Time spent yielding during the sweep (not doing the sweep).
+  Atomic<jlong> _sweep_duration;              // Time spent sweeping the table finding non-clean cards
+                                              // and refining them.
+  Atomic<jlong> _yield_during_sweep_duration; // Time spent yielding during the sweep (not doing the sweep).
 
-  size_t _cards_scanned;              // Total number of cards scanned.
-  size_t _cards_clean;                // Number of cards found clean.
-  size_t _cards_not_parsable;         // Number of cards we could not parse and left unrefined.
-  size_t _cards_already_refer_to_cset;// Number of cards marked found to be already young.
-  size_t _cards_refer_to_cset;        // Number of dirty cards that were recently found to contain a to-cset reference.
-  size_t _cards_no_cross_region;      // Number of dirty cards that were dirtied, but then cleaned again by the mutator.
+  Atomic<size_t> _cards_scanned;              // Total number of cards scanned.
+  Atomic<size_t> _cards_clean;                // Number of cards found clean.
+  Atomic<size_t> _cards_not_parsable;         // Number of cards we could not parse and left unrefined.
+  Atomic<size_t> _cards_already_refer_to_cset;// Number of cards marked found to be already young.
+  Atomic<size_t> _cards_refer_to_cset;        // Number of dirty cards that were recently found to contain a to-cset reference.
+  Atomic<size_t> _cards_no_cross_region;      // Number of dirty cards that were dirtied, but then cleaned again by the mutator.
 
-  jlong _refine_duration;             // Time spent during actual refinement.
+  Atomic<jlong> _refine_duration;             // Time spent during actual refinement.
 
 public:
   G1ConcurrentRefineStats();
 
   // Time spent performing sweeping the refinement table (includes actual refinement,
   // but not yield time).
-  jlong sweep_duration() const { return _sweep_duration - _yield_during_sweep_duration; }
-  jlong yield_during_sweep_duration() const { return _yield_during_sweep_duration; }
-  jlong refine_duration() const { return _refine_duration; }
+  inline jlong sweep_duration() const;
+  inline jlong yield_during_sweep_duration() const;
+  inline jlong refine_duration() const;
 
   // Number of refined cards.
-  size_t refined_cards() const { return cards_not_clean(); }
+  inline size_t refined_cards() const;
 
-  size_t cards_scanned() const { return _cards_scanned; }
-  size_t cards_clean() const { return _cards_clean; }
-  size_t cards_not_clean() const { return _cards_scanned - _cards_clean; }
-  size_t cards_not_parsable() const { return _cards_not_parsable; }
-  size_t cards_already_refer_to_cset() const { return _cards_already_refer_to_cset; }
-  size_t cards_refer_to_cset() const { return _cards_refer_to_cset; }
-  size_t cards_no_cross_region() const { return _cards_no_cross_region; }
+  inline size_t cards_scanned() const;
+  inline size_t cards_clean() const;
+  inline size_t cards_not_clean() const;
+  inline size_t cards_not_parsable() const;
+  inline size_t cards_already_refer_to_cset() const;
+  inline size_t cards_refer_to_cset() const;
+  inline size_t cards_no_cross_region() const;
   // Number of cards that were marked dirty and in need of refinement. This includes cards recently
   // found to refer to the collection set as they originally were dirty.
-  size_t cards_pending() const { return cards_not_clean() - _cards_already_refer_to_cset; }
+  inline size_t cards_pending() const;
 
-  size_t cards_to_cset() const { return _cards_already_refer_to_cset + _cards_refer_to_cset; }
+  inline size_t cards_to_cset() const;
 
-  void inc_sweep_time(jlong t) { _sweep_duration += t; }
-  void inc_yield_during_sweep_duration(jlong t) { _yield_during_sweep_duration += t; }
-  void inc_refine_duration(jlong t) { _refine_duration += t; }
+  inline void inc_sweep_time(jlong t);
+  inline void inc_yield_during_sweep_duration(jlong t);
+  inline void inc_refine_duration(jlong t);
 
-  void inc_cards_scanned(size_t increment) { _cards_scanned += increment; }
-  void inc_cards_clean(size_t increment) { _cards_clean += increment; }
-  void inc_cards_not_parsable() { _cards_not_parsable++; }
-  void inc_cards_already_refer_to_cset() { _cards_already_refer_to_cset++; }
-  void inc_cards_refer_to_cset() { _cards_refer_to_cset++; }
-  void inc_cards_no_cross_region() { _cards_no_cross_region++; }
+  inline void inc_cards_scanned(size_t increment);
+  inline void inc_cards_clean(size_t increment);
+  inline void inc_cards_not_parsable();
+  inline void inc_cards_already_refer_to_cset();
+  inline void inc_cards_refer_to_cset();
+  inline void inc_cards_no_cross_region();
 
   void add_atomic(G1ConcurrentRefineStats* other);
 

src/hotspot/share/gc/g1/g1ConcurrentRefineStats.inline.hpp

@@ -0,0 +1,118 @@
+/*
+ * Copyright (c) 2026, 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
+ * under the terms of the GNU General Public License version 2 only, as
+ * published by the Free Software Foundation.
+ *
+ * This code is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
+ * version 2 for more details (a copy is included in the LICENSE file that
+ * accompanied this code).
+ *
+ * You should have received a copy of the GNU General Public License version
+ * 2 along with this work; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
+ * or visit www.oracle.com if you need additional information or have any
+ * questions.
+ *
+ */
+
+#ifndef SHARE_GC_G1_G1CONCURRENTREFINESTATS_INLINE_HPP
+#define SHARE_GC_G1_G1CONCURRENTREFINESTATS_INLINE_HPP
+
+#include "gc/g1/g1ConcurrentRefineStats.hpp"
+
+inline jlong G1ConcurrentRefineStats::sweep_duration() const {
+  return _sweep_duration.load_relaxed() - yield_during_sweep_duration();
+}
+
+inline jlong G1ConcurrentRefineStats::yield_during_sweep_duration() const {
+  return _yield_during_sweep_duration.load_relaxed();
+}
+
+inline jlong G1ConcurrentRefineStats::refine_duration() const {
+  return _refine_duration.load_relaxed();
+}
+
+inline size_t G1ConcurrentRefineStats::refined_cards() const {
+  return cards_not_clean();
+}
+
+inline size_t G1ConcurrentRefineStats::cards_scanned() const {
+  return _cards_scanned.load_relaxed();
+}
+
+inline size_t G1ConcurrentRefineStats::cards_clean() const {
+  return _cards_clean.load_relaxed();
+}
+
+inline size_t G1ConcurrentRefineStats::cards_not_clean() const {
+  return cards_scanned() - cards_clean();
+}
+
+inline size_t G1ConcurrentRefineStats::cards_not_parsable() const {
+  return _cards_not_parsable.load_relaxed();
+}
+
+inline size_t G1ConcurrentRefineStats::cards_already_refer_to_cset() const {
+  return _cards_already_refer_to_cset.load_relaxed();
+}
+
+inline size_t G1ConcurrentRefineStats::cards_refer_to_cset() const {
+  return _cards_refer_to_cset.load_relaxed();
+}
+
+inline size_t G1ConcurrentRefineStats::cards_no_cross_region() const {
+  return _cards_no_cross_region.load_relaxed();
+}
+
+inline size_t G1ConcurrentRefineStats::cards_pending() const {
+  return cards_not_clean() - cards_already_refer_to_cset();
+}
+
+inline size_t G1ConcurrentRefineStats::cards_to_cset() const {
+  return cards_already_refer_to_cset() + cards_refer_to_cset();
+}
+
+inline void G1ConcurrentRefineStats::inc_sweep_time(jlong t) {
+  _sweep_duration.store_relaxed(_sweep_duration.load_relaxed() + t);
+}
+
+inline void G1ConcurrentRefineStats::inc_yield_during_sweep_duration(jlong t) {
+  _yield_during_sweep_duration.store_relaxed(yield_during_sweep_duration() + t);
+}
+
+inline void G1ConcurrentRefineStats::inc_refine_duration(jlong t) {
+  _refine_duration.store_relaxed(refine_duration() + t);
+}
+
+inline void G1ConcurrentRefineStats::inc_cards_scanned(size_t increment) {
+  _cards_scanned.store_relaxed(cards_scanned() + increment);
+}
+
+inline void G1ConcurrentRefineStats::inc_cards_clean(size_t increment) {
+  _cards_clean.store_relaxed(cards_clean() + increment);
+}
+
+inline void G1ConcurrentRefineStats::inc_cards_not_parsable() {
+  _cards_not_parsable.store_relaxed(cards_not_parsable() + 1);
+}
+
+inline void G1ConcurrentRefineStats::inc_cards_already_refer_to_cset() {
+  _cards_already_refer_to_cset.store_relaxed(cards_already_refer_to_cset() + 1);
+}
+
+inline void G1ConcurrentRefineStats::inc_cards_refer_to_cset() {
+  _cards_refer_to_cset.store_relaxed(cards_refer_to_cset() + 1);
+}
+
+inline void G1ConcurrentRefineStats::inc_cards_no_cross_region() {
+  _cards_no_cross_region.store_relaxed(cards_no_cross_region() + 1);
+}
+
+#endif // SHARE_GC_G1_G1CONCURRENTREFINESTATS_INLINE_HPP

src/hotspot/share/gc/g1/g1ConcurrentRefineSweepTask.cpp

@@ -24,6 +24,7 @@
 
 #include "gc/g1/g1CardTableClaimTable.inline.hpp"
 #include "gc/g1/g1CollectedHeap.inline.hpp"
+#include "gc/g1/g1ConcurrentRefineStats.inline.hpp"
 #include "gc/g1/g1ConcurrentRefineSweepTask.hpp"
 
 class G1RefineRegionClosure : public G1HeapRegionClosure {

src/hotspot/share/gc/g1/g1ConcurrentRefineSweepTask.hpp

@@ -25,10 +25,10 @@
 #ifndef SHARE_GC_G1_G1CONCURRENTREFINESWEEPTASK_HPP
 #define SHARE_GC_G1_G1CONCURRENTREFINESWEEPTASK_HPP
 
-#include "gc/g1/g1ConcurrentRefineStats.hpp"
 #include "gc/shared/workerThread.hpp"
 
 class G1CardTableClaimTable;
+class G1ConcurrentRefineStats;
 
 class G1ConcurrentRefineSweepTask : public WorkerTask {
   G1CardTableClaimTable* _scan_state;

src/hotspot/share/gc/g1/g1ConcurrentRefineThread.cpp

@@ -26,7 +26,7 @@
 #include "gc/g1/g1CardTableClaimTable.inline.hpp"
 #include "gc/g1/g1CollectedHeap.inline.hpp"
 #include "gc/g1/g1ConcurrentRefine.hpp"
-#include "gc/g1/g1ConcurrentRefineStats.hpp"
+#include "gc/g1/g1ConcurrentRefineStats.inline.hpp"
 #include "gc/g1/g1ConcurrentRefineSweepTask.hpp"
 #include "gc/g1/g1ConcurrentRefineThread.hpp"
 #include "gc/shared/gcTraceTime.inline.hpp"

src/hotspot/share/gc/g1/g1ConcurrentRefineThread.hpp

@@ -25,7 +25,6 @@
 #ifndef SHARE_GC_G1_G1CONCURRENTREFINETHREAD_HPP
 #define SHARE_GC_G1_G1CONCURRENTREFINETHREAD_HPP
 
-#include "gc/g1/g1ConcurrentRefineStats.hpp"
 #include "gc/shared/concurrentGCThread.hpp"
 #include "runtime/mutex.hpp"
 #include "utilities/globalDefinitions.hpp"

src/hotspot/share/gc/g1/g1EvacFailureRegions.cpp

@@ -1,5 +1,6 @@
 /*
  * Copyright (c) 2021, 2022, Huawei Technologies Co., Ltd. All rights reserved.
+ * Copyright (c) 2026, 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
@@ -28,7 +29,6 @@
 #include "gc/g1/g1EvacFailureRegions.inline.hpp"
 #include "gc/g1/g1HeapRegion.hpp"
 #include "memory/allocation.hpp"
-#include "runtime/atomicAccess.hpp"
 #include "utilities/bitMap.inline.hpp"
 
 G1EvacFailureRegions::G1EvacFailureRegions() :
@@ -43,7 +43,7 @@ G1EvacFailureRegions::~G1EvacFailureRegions() {
 }
 
 void G1EvacFailureRegions::pre_collection(uint max_regions) {
-  AtomicAccess::store(&_num_regions_evac_failed, 0u);
+  _num_regions_evac_failed.store_relaxed(0u);
   _regions_evac_failed.resize(max_regions);
   _regions_pinned.resize(max_regions);
   _regions_alloc_failed.resize(max_regions);
@@ -69,6 +69,6 @@ void G1EvacFailureRegions::par_iterate(G1HeapRegionClosure* closure,
   G1CollectedHeap::heap()->par_iterate_regions_array(closure,
                                                      hrclaimer,
                                                      _evac_failed_regions,
-                                                     AtomicAccess::load(&_num_regions_evac_failed),
+                                                     num_regions_evac_failed(),
                                                      worker_id);
 }

src/hotspot/share/gc/g1/g1EvacFailureRegions.hpp

@@ -1,5 +1,6 @@
 /*
  * Copyright (c) 2021, 2022, Huawei Technologies Co., Ltd. All rights reserved.
+ * Copyright (c) 2026, 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
@@ -25,6 +26,7 @@
 #ifndef SHARE_GC_G1_G1EVACFAILUREREGIONS_HPP
 #define SHARE_GC_G1_G1EVACFAILUREREGIONS_HPP
 
+#include "runtime/atomic.hpp"
 #include "utilities/bitMap.hpp"
 
 class G1AbstractSubTask;
@@ -53,14 +55,14 @@ class G1EvacFailureRegions {
   // Evacuation failed regions (indexes) in the current collection.
   uint* _evac_failed_regions;
   // Number of regions evacuation failed in the current collection.
-  volatile uint _num_regions_evac_failed;
+  Atomic<uint> _num_regions_evac_failed;
 
 public:
   G1EvacFailureRegions();
   ~G1EvacFailureRegions();
 
   uint get_region_idx(uint idx) const {
-    assert(idx < _num_regions_evac_failed, "precondition");
+    assert(idx < _num_regions_evac_failed.load_relaxed(), "precondition");
     return _evac_failed_regions[idx];
   }
 

src/hotspot/share/gc/g1/g1EvacFailureRegions.inline.hpp

@@ -1,5 +1,6 @@
 /*
  * Copyright (c) 2021, 2022, Huawei Technologies Co., Ltd. All rights reserved.
+ * Copyright (c) 2026, 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
@@ -29,10 +30,9 @@
 
 #include "gc/g1/g1CollectedHeap.inline.hpp"
 #include "gc/g1/g1GCPhaseTimes.hpp"
-#include "runtime/atomicAccess.hpp"
 
 uint G1EvacFailureRegions::num_regions_evac_failed() const {
-  return AtomicAccess::load(&_num_regions_evac_failed);
+  return _num_regions_evac_failed.load_relaxed();
 }
 
 bool G1EvacFailureRegions::has_regions_evac_failed() const {
@@ -57,7 +57,7 @@ bool G1EvacFailureRegions::record(uint worker_id, uint region_idx, bool cause_pi
   bool success = _regions_evac_failed.par_set_bit(region_idx,
                                                   memory_order_relaxed);
   if (success) {
-    size_t offset = AtomicAccess::fetch_then_add(&_num_regions_evac_failed, 1u);
+    size_t offset = _num_regions_evac_failed.fetch_then_add(1u);
     _evac_failed_regions[offset] = region_idx;
 
     G1CollectedHeap* g1h = G1CollectedHeap::heap();

src/hotspot/share/gc/g1/g1EvacStats.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2015, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2015, 2026, 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
@@ -22,13 +22,24 @@
  *
  */
 
-#include "gc/g1/g1EvacStats.hpp"
+#include "gc/g1/g1EvacStats.inline.hpp"
 #include "gc/shared/gc_globals.hpp"
 #include "gc/shared/gcId.hpp"
 #include "logging/log.hpp"
 #include "memory/allocation.inline.hpp"
 #include "runtime/globals.hpp"
 
+void G1EvacStats::reset() {
+  PLABStats::reset();
+  _region_end_waste.store_relaxed(0);
+  _regions_filled.store_relaxed(0);
+  _num_plab_filled.store_relaxed(0);
+  _direct_allocated.store_relaxed(0);
+  _num_direct_allocated.store_relaxed(0);
+  _failure_used.store_relaxed(0);
+  _failure_waste.store_relaxed(0);
+}
+
 void G1EvacStats::log_plab_allocation() {
   log_debug(gc, plab)("%s PLAB allocation: "
                       "allocated: %zuB, "
@@ -51,13 +62,13 @@ void G1EvacStats::log_plab_allocation() {
                       "failure used: %zuB, "
                       "failure wasted: %zuB",
                       _description,
-                      _region_end_waste * HeapWordSize,
-                      _regions_filled,
-                      _num_plab_filled,
-                      _direct_allocated * HeapWordSize,
-                      _num_direct_allocated,
-                      _failure_used * HeapWordSize,
-                      _failure_waste * HeapWordSize);
+                      region_end_waste() * HeapWordSize,
+                      regions_filled(),
+                      num_plab_filled(),
+                      direct_allocated() * HeapWordSize,
+                      num_direct_allocated(),
+                      failure_used() * HeapWordSize,
+                      failure_waste() * HeapWordSize);
 }
 
 void G1EvacStats::log_sizing(size_t calculated_words, size_t net_desired_words) {
@@ -109,7 +120,7 @@ size_t G1EvacStats::compute_desired_plab_size() const {
   // threads do not allocate anything but a few rather large objects. In this
   // degenerate case the PLAB size would simply quickly tend to minimum PLAB size,
   // which is an okay reaction.
-  size_t const used_for_waste_calculation = used() > _region_end_waste ? used() - _region_end_waste : 0;
+  size_t const used_for_waste_calculation = used() > region_end_waste() ? used() - region_end_waste() : 0;
 
   size_t const total_waste_allowed = used_for_waste_calculation * TargetPLABWastePct;
   return (size_t)((double)total_waste_allowed / (100 - G1LastPLABAverageOccupancy));

src/hotspot/share/gc/g1/g1EvacStats.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2015, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2015, 2026, 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
@@ -27,6 +27,7 @@
 
 #include "gc/shared/gcUtil.hpp"
 #include "gc/shared/plab.hpp"
+#include "runtime/atomic.hpp"
 
 // Records various memory allocation statistics gathered during evacuation. All sizes
 // are in HeapWords.
@@ -36,30 +37,21 @@ class G1EvacStats : public PLABStats {
   AdaptiveWeightedAverage
          _net_plab_size_filter;  // Integrator with decay
 
-  size_t _region_end_waste; // Number of words wasted due to skipping to the next region.
-  uint   _regions_filled;   // Number of regions filled completely.
-  size_t _num_plab_filled; // Number of PLABs filled and retired.
-  size_t _direct_allocated; // Number of words allocated directly into the regions.
-  size_t _num_direct_allocated; // Number of direct allocation attempts.
+  Atomic<size_t> _region_end_waste; // Number of words wasted due to skipping to the next region.
+  Atomic<uint>   _regions_filled;   // Number of regions filled completely.
+  Atomic<size_t> _num_plab_filled; // Number of PLABs filled and retired.
+  Atomic<size_t> _direct_allocated; // Number of words allocated directly into the regions.
+  Atomic<size_t> _num_direct_allocated; // Number of direct allocation attempts.
 
   // Number of words in live objects remaining in regions that ultimately suffered an
   // evacuation failure. This is used in the regions when the regions are made old regions.
-  size_t _failure_used;
+  Atomic<size_t> _failure_used;
   // Number of words wasted in regions which failed evacuation. This is the sum of space
   // for objects successfully copied out of the regions (now dead space) plus waste at the
   // end of regions.
-  size_t _failure_waste;
+  Atomic<size_t> _failure_waste;
 
-  virtual void reset() {
-    PLABStats::reset();
-    _region_end_waste = 0;
-    _regions_filled = 0;
-    _num_plab_filled = 0;
-    _direct_allocated = 0;
-    _num_direct_allocated = 0;
-    _failure_used = 0;
-    _failure_waste = 0;
-  }
+  virtual void reset();
 
   void log_plab_allocation();
   void log_sizing(size_t calculated_words, size_t net_desired_words);
@@ -77,16 +69,16 @@ public:
   // Should be called at the end of a GC pause.
   void adjust_desired_plab_size();
 
-  uint regions_filled() const { return _regions_filled; }
-  size_t num_plab_filled() const { return _num_plab_filled; }
-  size_t region_end_waste() const { return _region_end_waste; }
-  size_t direct_allocated() const { return _direct_allocated; }
-  size_t num_direct_allocated() const { return _num_direct_allocated; }
+  uint regions_filled() const;
+  size_t num_plab_filled() const;
+  size_t region_end_waste() const;
+  size_t direct_allocated() const;
+  size_t num_direct_allocated() const;
 
   // Amount of space in heapwords used in the failing regions when an evacuation failure happens.
-  size_t failure_used() const { return _failure_used; }
+  size_t failure_used() const;
   // Amount of space in heapwords wasted (unused) in the failing regions when an evacuation failure happens.
-  size_t failure_waste() const { return _failure_waste; }
+  size_t failure_waste() const;
 
   inline void add_num_plab_filled(size_t value);
   inline void add_direct_allocated(size_t value);

src/hotspot/share/gc/g1/g1EvacStats.inline.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2015, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2015, 2026, 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
@@ -27,28 +27,54 @@
 
 #include "gc/g1/g1EvacStats.hpp"
 
-#include "runtime/atomicAccess.hpp"
+inline uint G1EvacStats::regions_filled() const {
+  return _regions_filled.load_relaxed();
+}
+
+inline size_t G1EvacStats::num_plab_filled() const {
+  return _num_plab_filled.load_relaxed();
+}
+
+inline size_t G1EvacStats::region_end_waste() const {
+  return _region_end_waste.load_relaxed();
+}
+
+inline size_t G1EvacStats::direct_allocated() const {
+  return _direct_allocated.load_relaxed();
+}
+
+inline size_t G1EvacStats::num_direct_allocated() const {
+  return _num_direct_allocated.load_relaxed();
+}
+
+inline size_t G1EvacStats::failure_used() const {
+  return _failure_used.load_relaxed();
+}
+
+inline size_t G1EvacStats::failure_waste() const {
+  return _failure_waste.load_relaxed();
+}
 
 inline void G1EvacStats::add_direct_allocated(size_t value) {
-  AtomicAccess::add(&_direct_allocated, value, memory_order_relaxed);
+  _direct_allocated.add_then_fetch(value, memory_order_relaxed);
 }
 
 inline void G1EvacStats::add_num_plab_filled(size_t value) {
-  AtomicAccess::add(&_num_plab_filled, value, memory_order_relaxed);
+  _num_plab_filled.add_then_fetch(value, memory_order_relaxed);
 }
 
 inline void G1EvacStats::add_num_direct_allocated(size_t value) {
-  AtomicAccess::add(&_num_direct_allocated, value, memory_order_relaxed);
+  _num_direct_allocated.add_then_fetch(value, memory_order_relaxed);
 }
 
 inline void G1EvacStats::add_region_end_waste(size_t value) {
-  AtomicAccess::add(&_region_end_waste, value, memory_order_relaxed);
-  AtomicAccess::inc(&_regions_filled, memory_order_relaxed);
+  _region_end_waste.add_then_fetch(value, memory_order_relaxed);
+  _regions_filled.add_then_fetch(1u, memory_order_relaxed);
 }
 
 inline void G1EvacStats::add_failure_used_and_waste(size_t used, size_t waste) {
-  AtomicAccess::add(&_failure_used, used, memory_order_relaxed);
-  AtomicAccess::add(&_failure_waste, waste, memory_order_relaxed);
+  _failure_used.add_then_fetch(used, memory_order_relaxed);
+  _failure_waste.add_then_fetch(waste, memory_order_relaxed);
 }
 
 #endif // SHARE_GC_G1_G1EVACSTATS_INLINE_HPP

src/hotspot/share/gc/g1/g1FullCollector.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2017, 2026, 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
@@ -116,8 +116,8 @@ G1FullCollector::G1FullCollector(G1CollectedHeap* heap,
     _num_workers(calc_active_workers()),
     _has_compaction_targets(false),
     _has_humongous(false),
-    _oop_queue_set(_num_workers),
-    _array_queue_set(_num_workers),
+    _marking_task_queues(_num_workers),
+    _partial_array_state_manager(nullptr),
     _preserved_marks_set(true),
     _serial_compaction_point(this, nullptr),
     _humongous_compaction_point(this, nullptr),
@@ -134,32 +134,40 @@ G1FullCollector::G1FullCollector(G1CollectedHeap* heap,
   _compaction_points = NEW_C_HEAP_ARRAY(G1FullGCCompactionPoint*, _num_workers, mtGC);
 
   _live_stats = NEW_C_HEAP_ARRAY(G1RegionMarkStats, _heap->max_num_regions(), mtGC);
-  _compaction_tops = NEW_C_HEAP_ARRAY(HeapWord*, _heap->max_num_regions(), mtGC);
+  _compaction_tops = NEW_C_HEAP_ARRAY(Atomic<HeapWord*>, _heap->max_num_regions(), mtGC);
   for (uint j = 0; j < heap->max_num_regions(); j++) {
     _live_stats[j].clear();
-    _compaction_tops[j] = nullptr;
+    ::new (&_compaction_tops[j]) Atomic<HeapWord*>{};
   }
 
+  _partial_array_state_manager = new PartialArrayStateManager(_num_workers);
+
   for (uint i = 0; i < _num_workers; i++) {
     _markers[i] = new G1FullGCMarker(this, i, _live_stats);
     _compaction_points[i] = new G1FullGCCompactionPoint(this, _preserved_marks_set.get(i));
-    _oop_queue_set.register_queue(i, marker(i)->oop_stack());
-    _array_queue_set.register_queue(i, marker(i)->objarray_stack());
+    _marking_task_queues.register_queue(i, marker(i)->task_queue());
   }
+
   _serial_compaction_point.set_preserved_stack(_preserved_marks_set.get(0));
   _humongous_compaction_point.set_preserved_stack(_preserved_marks_set.get(0));
   _region_attr_table.initialize(heap->reserved(), G1HeapRegion::GrainBytes);
 }
 
+PartialArrayStateManager* G1FullCollector::partial_array_state_manager() const {
+  return _partial_array_state_manager;
+}
+
 G1FullCollector::~G1FullCollector() {
   for (uint i = 0; i < _num_workers; i++) {
     delete _markers[i];
     delete _compaction_points[i];
   }
 
+  delete _partial_array_state_manager;
+
   FREE_C_HEAP_ARRAY(G1FullGCMarker*, _markers);
   FREE_C_HEAP_ARRAY(G1FullGCCompactionPoint*, _compaction_points);
-  FREE_C_HEAP_ARRAY(HeapWord*, _compaction_tops);
+  FREE_C_HEAP_ARRAY(Atomic<HeapWord*>, _compaction_tops);
   FREE_C_HEAP_ARRAY(G1RegionMarkStats, _live_stats);
 }
 
@@ -279,8 +287,8 @@ public:
     uint index = (_tm == RefProcThreadModel::Single) ? 0 : worker_id;
     G1FullKeepAliveClosure keep_alive(_collector.marker(index));
     BarrierEnqueueDiscoveredFieldClosure enqueue;
-    G1FollowStackClosure* complete_gc = _collector.marker(index)->stack_closure();
-    _rp_task->rp_work(worker_id, &is_alive, &keep_alive, &enqueue, complete_gc);
+    G1MarkStackClosure* complete_marking = _collector.marker(index)->stack_closure();
+    _rp_task->rp_work(worker_id, &is_alive, &keep_alive, &enqueue, complete_marking);
   }
 };
 
@@ -302,7 +310,7 @@ void G1FullCollector::phase1_mark_live_objects() {
     const ReferenceProcessorStats& stats = reference_processor()->process_discovered_references(task, _heap->workers(), pt);
     scope()->tracer()->report_gc_reference_stats(stats);
     pt.print_all_references();
-    assert(marker(0)->oop_stack()->is_empty(), "Should be no oops on the stack");
+    assert(marker(0)->task_queue()->is_empty(), "Should be no oops on the stack");
   }
 
   {
@@ -328,8 +336,7 @@ void G1FullCollector::phase1_mark_live_objects() {
     scope()->tracer()->report_object_count_after_gc(&_is_alive, _heap->workers());
   }
 #if TASKQUEUE_STATS
-  oop_queue_set()->print_and_reset_taskqueue_stats("Oop Queue");
-  array_queue_set()->print_and_reset_taskqueue_stats("ObjArrayOop Queue");
+  marking_task_queues()->print_and_reset_taskqueue_stats("Marking Task Queue");
 #endif
 }
 

src/hotspot/share/gc/g1/g1FullCollector.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2017, 2026, 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
@@ -79,8 +79,8 @@ class G1FullCollector : StackObj {
   bool                      _has_humongous;
   G1FullGCMarker**          _markers;
   G1FullGCCompactionPoint** _compaction_points;
-  OopQueueSet               _oop_queue_set;
-  ObjArrayTaskQueueSet      _array_queue_set;
+  G1MarkTasksQueueSet       _marking_task_queues;
+  PartialArrayStateManager* _partial_array_state_manager;
   PreservedMarksSet         _preserved_marks_set;
   G1FullGCCompactionPoint   _serial_compaction_point;
   G1FullGCCompactionPoint   _humongous_compaction_point;
@@ -96,7 +96,7 @@ class G1FullCollector : StackObj {
 
   G1FullGCHeapRegionAttr _region_attr_table;
 
-  HeapWord* volatile* _compaction_tops;
+  Atomic<HeapWord*>* _compaction_tops;
 
 public:
   G1FullCollector(G1CollectedHeap* heap,
@@ -113,8 +113,7 @@ public:
   uint                     workers() { return _num_workers; }
   G1FullGCMarker*          marker(uint id) { return _markers[id]; }
   G1FullGCCompactionPoint* compaction_point(uint id) { return _compaction_points[id]; }
-  OopQueueSet*             oop_queue_set() { return &_oop_queue_set; }
-  ObjArrayTaskQueueSet*    array_queue_set() { return &_array_queue_set; }
+  G1MarkTasksQueueSet*     marking_task_queues() { return &_marking_task_queues; }
   PreservedMarksSet*       preserved_mark_set() { return &_preserved_marks_set; }
   G1FullGCCompactionPoint* serial_compaction_point() { return &_serial_compaction_point; }
   G1FullGCCompactionPoint* humongous_compaction_point() { return &_humongous_compaction_point; }
@@ -122,9 +121,11 @@ public:
   ReferenceProcessor*      reference_processor();
   size_t live_words(uint region_index) const {
     assert(region_index < _heap->max_num_regions(), "sanity");
-    return _live_stats[region_index]._live_words;
+    return _live_stats[region_index].live_words();
   }
 
+  PartialArrayStateManager* partial_array_state_manager() const;
+
   void before_marking_update_attribute_table(G1HeapRegion* hr);
 
   inline bool is_compacting(oop obj) const;

src/hotspot/share/gc/g1/g1FullCollector.inline.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2020, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2020, 2026, 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
@@ -63,11 +63,11 @@ void G1FullCollector::update_from_skip_compacting_to_compacting(uint region_idx)
 }
 
 void G1FullCollector::set_compaction_top(G1HeapRegion* r, HeapWord* value) {
-  AtomicAccess::store(&_compaction_tops[r->hrm_index()], value);
+  _compaction_tops[r->hrm_index()].store_relaxed(value);
 }
 
 HeapWord* G1FullCollector::compaction_top(G1HeapRegion* r) const {
-  return AtomicAccess::load(&_compaction_tops[r->hrm_index()]);
+  return _compaction_tops[r->hrm_index()].load_relaxed();
 }
 
 void G1FullCollector::set_has_compaction_targets() {

src/hotspot/share/gc/g1/g1FullGCAdjustTask.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2017, 2026, 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
@@ -37,7 +37,6 @@
 #include "gc/shared/weakProcessor.inline.hpp"
 #include "logging/log.hpp"
 #include "memory/iterator.inline.hpp"
-#include "runtime/atomicAccess.hpp"
 
 class G1AdjustLiveClosure : public StackObj {
   G1AdjustClosure* _adjust_closure;

src/hotspot/share/gc/g1/g1FullGCMarkTask.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2017, 2026, 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
@@ -34,7 +34,7 @@
 G1FullGCMarkTask::G1FullGCMarkTask(G1FullCollector* collector) :
     G1FullGCTask("G1 Parallel Marking Task", collector),
     _root_processor(G1CollectedHeap::heap(), collector->workers()),
-    _terminator(collector->workers(), collector->array_queue_set()) {
+    _terminator(collector->workers(), collector->marking_task_queues()) {
 }
 
 void G1FullGCMarkTask::work(uint worker_id) {
@@ -54,10 +54,9 @@ void G1FullGCMarkTask::work(uint worker_id) {
   }
 
   // Mark stack is populated, now process and drain it.
-  marker->complete_marking(collector()->oop_queue_set(), collector()->array_queue_set(), &_terminator);
+  marker->complete_marking(collector()->marking_task_queues(), &_terminator);
 
   // This is the point where the entire marking should have completed.
-  assert(marker->oop_stack()->is_empty(), "Marking should have completed");
-  assert(marker->objarray_stack()->is_empty(), "Array marking should have completed");
+  assert(marker->task_queue()->is_empty(), "Marking should have completed");
   log_task("Marking task", worker_id, start);
 }

src/hotspot/share/gc/g1/g1FullGCMarker.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2017, 2026, 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
@@ -25,6 +25,8 @@
 #include "classfile/classLoaderData.hpp"
 #include "classfile/classLoaderDataGraph.hpp"
 #include "gc/g1/g1FullGCMarker.inline.hpp"
+#include "gc/shared/partialArraySplitter.inline.hpp"
+#include "gc/shared/partialArrayState.hpp"
 #include "gc/shared/referenceProcessor.hpp"
 #include "gc/shared/taskTerminator.hpp"
 #include "gc/shared/verifyOption.hpp"
@@ -36,8 +38,8 @@ G1FullGCMarker::G1FullGCMarker(G1FullCollector* collector,
     _collector(collector),
     _worker_id(worker_id),
     _bitmap(collector->mark_bitmap()),
-    _oop_stack(),
-    _objarray_stack(),
+    _task_queue(),
+    _partial_array_splitter(collector->partial_array_state_manager(), collector->workers(), ObjArrayMarkingStride),
     _mark_closure(worker_id, this, ClassLoaderData::_claim_stw_fullgc_mark, G1CollectedHeap::heap()->ref_processor_stw()),
     _stack_closure(this),
     _cld_closure(mark_closure(), ClassLoaderData::_claim_stw_fullgc_mark),
@@ -47,24 +49,36 @@ G1FullGCMarker::G1FullGCMarker(G1FullCollector* collector,
 }
 
 G1FullGCMarker::~G1FullGCMarker() {
-  assert(is_empty(), "Must be empty at this point");
+  assert(is_task_queue_empty(), "Must be empty at this point");
 }
 
-void G1FullGCMarker::complete_marking(OopQueueSet* oop_stacks,
-                                      ObjArrayTaskQueueSet* array_stacks,
+void G1FullGCMarker::process_partial_array(PartialArrayState* state, bool stolen) {
+  // Access state before release by claim().
+  objArrayOop obj_array = objArrayOop(state->source());
+  PartialArraySplitter::Claim claim =
+    _partial_array_splitter.claim(state, task_queue(), stolen);
+  process_array_chunk(obj_array, claim._start, claim._end);
+}
+
+void G1FullGCMarker::start_partial_array_processing(objArrayOop obj) {
+  mark_closure()->do_klass(obj->klass());
+  // Don't push empty arrays to avoid unnecessary work.
+  size_t array_length = obj->length();
+  if (array_length > 0) {
+    size_t initial_chunk_size = _partial_array_splitter.start(task_queue(), obj, nullptr, array_length);
+    process_array_chunk(obj, 0, initial_chunk_size);
+  }
+}
+
+void G1FullGCMarker::complete_marking(G1ScannerTasksQueueSet* task_queues,
                                       TaskTerminator* terminator) {
   do {
-    follow_marking_stacks();
-    ObjArrayTask steal_array;
-    if (array_stacks->steal(_worker_id, steal_array)) {
-      follow_array_chunk(objArrayOop(steal_array.obj()), steal_array.index());
-    } else {
-      oop steal_oop;
-      if (oop_stacks->steal(_worker_id, steal_oop)) {
-        follow_object(steal_oop);
-      }
+    process_marking_stacks();
+    ScannerTask stolen_task;
+    if (task_queues->steal(_worker_id, stolen_task)) {
+      dispatch_task(stolen_task, true);
     }
-  } while (!is_empty() || !terminator->offer_termination());
+  } while (!is_task_queue_empty() || !terminator->offer_termination());
 }
 
 void G1FullGCMarker::flush_mark_stats_cache() {

src/hotspot/share/gc/g1/g1FullGCMarker.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017, 2024, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2017, 2026, 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
@@ -28,6 +28,8 @@
 #include "gc/g1/g1FullGCOopClosures.hpp"
 #include "gc/g1/g1OopClosures.hpp"
 #include "gc/g1/g1RegionMarkStatsCache.hpp"
+#include "gc/shared/partialArraySplitter.hpp"
+#include "gc/shared/partialArrayState.hpp"
 #include "gc/shared/stringdedup/stringDedup.hpp"
 #include "gc/shared/taskqueue.hpp"
 #include "memory/iterator.hpp"
@@ -38,16 +40,15 @@
 #include "utilities/growableArray.hpp"
 #include "utilities/stack.hpp"
 
-typedef OverflowTaskQueue<oop, mtGC>                 OopQueue;
-typedef OverflowTaskQueue<ObjArrayTask, mtGC>        ObjArrayTaskQueue;
 
-typedef GenericTaskQueueSet<OopQueue, mtGC>          OopQueueSet;
-typedef GenericTaskQueueSet<ObjArrayTaskQueue, mtGC> ObjArrayTaskQueueSet;
 
 class G1CMBitMap;
 class G1FullCollector;
 class TaskTerminator;
 
+typedef OverflowTaskQueue<ScannerTask, mtGC>        G1MarkTasksQueue;
+typedef GenericTaskQueueSet<G1MarkTasksQueue, mtGC> G1MarkTasksQueueSet;
+
 class G1FullGCMarker : public CHeapObj<mtGC> {
   G1FullCollector*   _collector;
 
@@ -56,56 +57,50 @@ class G1FullGCMarker : public CHeapObj<mtGC> {
   G1CMBitMap*        _bitmap;
 
   // Mark stack
-  OopQueue           _oop_stack;
-  ObjArrayTaskQueue  _objarray_stack;
+  G1MarkTasksQueue     _task_queue;
+  PartialArraySplitter _partial_array_splitter;
 
   // Marking closures
   G1MarkAndPushClosure  _mark_closure;
-  G1FollowStackClosure  _stack_closure;
+  G1MarkStackClosure    _stack_closure;
   CLDToOopClosure       _cld_closure;
   StringDedup::Requests _string_dedup_requests;
 
 
   G1RegionMarkStatsCache _mark_stats_cache;
 
-  inline bool is_empty();
-  inline void push_objarray(oop obj, size_t index);
+  inline bool is_task_queue_empty();
   inline bool mark_object(oop obj);
 
   // Marking helpers
-  inline void follow_object(oop obj);
-  inline void follow_array(objArrayOop array);
-  inline void follow_array_chunk(objArrayOop array, int index);
+  inline void process_array_chunk(objArrayOop obj, size_t start, size_t end);
+  inline void dispatch_task(const ScannerTask& task, bool stolen);
+  // Start processing the given objArrayOop by first pushing its continuations and
+  // then scanning the first chunk.
+  void start_partial_array_processing(objArrayOop obj);
+  // Process the given continuation.
+  void process_partial_array(PartialArrayState* state, bool stolen);
 
   inline void publish_and_drain_oop_tasks();
-  // Try to publish all contents from the objArray task queue overflow stack to
-  // the shared objArray stack.
-  // Returns true and a valid task if there has not been enough space in the shared
-  // objArray stack, otherwise returns false and the task is invalid.
-  inline bool publish_or_pop_objarray_tasks(ObjArrayTask& task);
-
 public:
   G1FullGCMarker(G1FullCollector* collector,
                  uint worker_id,
                  G1RegionMarkStats* mark_stats);
   ~G1FullGCMarker();
 
-  // Stack getters
-  OopQueue*          oop_stack()       { return &_oop_stack; }
-  ObjArrayTaskQueue* objarray_stack()  { return &_objarray_stack; }
+  G1MarkTasksQueue* task_queue() { return &_task_queue; }
 
   // Marking entry points
   template <class T> inline void mark_and_push(T* p);
 
-  inline void follow_marking_stacks();
-  void complete_marking(OopQueueSet* oop_stacks,
-                        ObjArrayTaskQueueSet* array_stacks,
+  inline void process_marking_stacks();
+  void complete_marking(G1MarkTasksQueueSet* task_queues,
                         TaskTerminator* terminator);
 
   // Closure getters
   CLDToOopClosure*      cld_closure()   { return &_cld_closure; }
   G1MarkAndPushClosure* mark_closure()  { return &_mark_closure; }
-  G1FollowStackClosure* stack_closure() { return &_stack_closure; }
+  G1MarkStackClosure*   stack_closure() { return &_stack_closure; }
 
   // Flush live bytes to regions
   void flush_mark_stats_cache();

src/hotspot/share/gc/g1/g1FullGCMarker.inline.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017, 2024, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2017, 2026, 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
@@ -42,6 +42,7 @@
 #include "oops/access.inline.hpp"
 #include "oops/compressedOops.inline.hpp"
 #include "oops/oop.inline.hpp"
+#include "utilities/checkedCast.hpp"
 #include "utilities/debug.hpp"
 
 inline bool G1FullGCMarker::mark_object(oop obj) {
@@ -71,94 +72,55 @@ template <class T> inline void G1FullGCMarker::mark_and_push(T* p) {
   if (!CompressedOops::is_null(heap_oop)) {
     oop obj = CompressedOops::decode_not_null(heap_oop);
     if (mark_object(obj)) {
-      _oop_stack.push(obj);
+      _task_queue.push(ScannerTask(obj));
     }
     assert(_bitmap->is_marked(obj), "Must be marked");
   }
 }
 
-inline bool G1FullGCMarker::is_empty() {
-  return _oop_stack.is_empty() && _objarray_stack.is_empty();
+inline bool G1FullGCMarker::is_task_queue_empty() {
+  return _task_queue.is_empty();
 }
 
-inline void G1FullGCMarker::push_objarray(oop obj, size_t index) {
-  ObjArrayTask task(obj, index);
-  assert(task.is_valid(), "bad ObjArrayTask");
-  _objarray_stack.push(task);
+inline void G1FullGCMarker::process_array_chunk(objArrayOop obj, size_t start, size_t end) {
+  obj->oop_iterate_elements_range(mark_closure(),
+                                  checked_cast<int>(start),
+                                  checked_cast<int>(end));
 }
 
-inline void G1FullGCMarker::follow_array(objArrayOop array) {
-  mark_closure()->do_klass(array->klass());
-  // Don't push empty arrays to avoid unnecessary work.
-  if (array->length() > 0) {
-    push_objarray(array, 0);
-  }
-}
-
-void G1FullGCMarker::follow_array_chunk(objArrayOop array, int index) {
-  const int len = array->length();
-  const int beg_index = index;
-  assert(beg_index < len || len == 0, "index too large");
-
-  const int stride = MIN2(len - beg_index, (int) ObjArrayMarkingStride);
-  const int end_index = beg_index + stride;
-
-  // Push the continuation first to allow more efficient work stealing.
-  if (end_index < len) {
-    push_objarray(array, end_index);
-  }
-
-  array->oop_iterate_elements_range(mark_closure(), beg_index, end_index);
-}
-
-inline void G1FullGCMarker::follow_object(oop obj) {
-  assert(_bitmap->is_marked(obj), "should be marked");
-  if (obj->is_objArray()) {
-    // Handle object arrays explicitly to allow them to
-    // be split into chunks if needed.
-    follow_array((objArrayOop)obj);
+inline void G1FullGCMarker::dispatch_task(const ScannerTask& task, bool stolen) {
+  if (task.is_partial_array_state()) {
+    assert(_bitmap->is_marked(task.to_partial_array_state()->source()), "should be marked");
+    process_partial_array(task.to_partial_array_state(), stolen);
   } else {
-    obj->oop_iterate(mark_closure());
+    oop obj = task.to_oop();
+    assert(_bitmap->is_marked(obj), "should be marked");
+    if (obj->is_objArray()) {
+      // Handle object arrays explicitly to allow them to
+      // be split into chunks if needed.
+      start_partial_array_processing((objArrayOop)obj);
+    } else {
+      obj->oop_iterate(mark_closure());
+    }
   }
 }
 
 inline void G1FullGCMarker::publish_and_drain_oop_tasks() {
-  oop obj;
-  while (_oop_stack.pop_overflow(obj)) {
-    if (!_oop_stack.try_push_to_taskqueue(obj)) {
-      assert(_bitmap->is_marked(obj), "must be marked");
-      follow_object(obj);
+  ScannerTask task;
+  while (_task_queue.pop_overflow(task)) {
+    if (!_task_queue.try_push_to_taskqueue(task)) {
+      dispatch_task(task, false);
     }
   }
-  while (_oop_stack.pop_local(obj)) {
-    assert(_bitmap->is_marked(obj), "must be marked");
-    follow_object(obj);
+  while (_task_queue.pop_local(task)) {
+    dispatch_task(task, false);
   }
 }
 
-inline bool G1FullGCMarker::publish_or_pop_objarray_tasks(ObjArrayTask& task) {
-  // It is desirable to move as much as possible work from the overflow queue to
-  // the shared queue as quickly as possible.
-  while (_objarray_stack.pop_overflow(task)) {
-    if (!_objarray_stack.try_push_to_taskqueue(task)) {
-      return true;
-    }
-  }
-  return false;
-}
-
-void G1FullGCMarker::follow_marking_stacks() {
+void G1FullGCMarker::process_marking_stacks() {
   do {
-    // First, drain regular oop stack.
     publish_and_drain_oop_tasks();
-
-    // Then process ObjArrays one at a time to avoid marking stack bloat.
-    ObjArrayTask task;
-    if (publish_or_pop_objarray_tasks(task) ||
-        _objarray_stack.pop_local(task)) {
-      follow_array_chunk(objArrayOop(task.obj()), task.index());
-    }
-  } while (!is_empty());
+  } while (!is_task_queue_empty());
 }
 
 #endif // SHARE_GC_G1_G1FULLGCMARKER_INLINE_HPP

src/hotspot/share/gc/g1/g1FullGCOopClosures.cpp

@@ -35,7 +35,7 @@
 G1IsAliveClosure::G1IsAliveClosure(G1FullCollector* collector) :
   G1IsAliveClosure(collector, collector->mark_bitmap()) { }
 
-void G1FollowStackClosure::do_void() { _marker->follow_marking_stacks(); }
+void G1MarkStackClosure::do_void() { _marker->process_marking_stacks(); }
 
 void G1FullKeepAliveClosure::do_oop(oop* p) { do_oop_work(p); }
 void G1FullKeepAliveClosure::do_oop(narrowOop* p) { do_oop_work(p); }

src/hotspot/share/gc/g1/g1FullGCOopClosures.hpp

@@ -86,11 +86,11 @@ public:
   virtual ReferenceIterationMode reference_iteration_mode() { return DO_FIELDS; }
 };
 
-class G1FollowStackClosure: public VoidClosure {
+class G1MarkStackClosure: public VoidClosure {
   G1FullGCMarker* _marker;
 
 public:
-  G1FollowStackClosure(G1FullGCMarker* marker) : _marker(marker) {}
+  G1MarkStackClosure(G1FullGCMarker* marker) : _marker(marker) {}
   virtual void do_void();
 };
 

src/hotspot/share/gc/g1/g1HeapRegionRemSet.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2001, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2001, 2026, 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
@@ -31,7 +31,6 @@
 #include "memory/allocation.hpp"
 #include "memory/padded.inline.hpp"
 #include "oops/oop.inline.hpp"
-#include "runtime/atomicAccess.hpp"
 #include "runtime/globals_extension.hpp"
 #include "runtime/java.hpp"
 #include "runtime/mutexLocker.hpp"

src/hotspot/share/gc/g1/g1HeapRegionRemSet.hpp

@@ -30,7 +30,6 @@
 #include "gc/g1/g1CodeRootSet.hpp"
 #include "gc/g1/g1CollectionSetCandidates.hpp"
 #include "gc/g1/g1FromCardCache.hpp"
-#include "runtime/atomicAccess.hpp"
 #include "runtime/mutexLocker.hpp"
 #include "runtime/safepoint.hpp"
 #include "utilities/bitMap.hpp"

src/hotspot/share/gc/g1/g1HeapRegionRemSet.inline.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1997, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 2026, 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
@@ -30,7 +30,6 @@
 #include "gc/g1/g1CardSet.inline.hpp"
 #include "gc/g1/g1CollectedHeap.inline.hpp"
 #include "gc/g1/g1HeapRegion.inline.hpp"
-#include "runtime/atomicAccess.hpp"
 #include "utilities/bitMap.inline.hpp"
 
 void G1HeapRegionRemSet::set_state_untracked() {

src/hotspot/share/gc/g1/g1MonotonicArena.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2022, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2022, 2026, 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
@@ -24,7 +24,6 @@
 
 #include "gc/g1/g1MonotonicArena.inline.hpp"
 #include "memory/allocation.hpp"
-#include "runtime/atomicAccess.hpp"
 #include "runtime/vmOperations.hpp"
 #include "utilities/globalCounter.inline.hpp"
 
@@ -61,13 +60,13 @@ void G1MonotonicArena::SegmentFreeList::bulk_add(Segment& first,
                                                  size_t num,
                                                  size_t mem_size) {
   _list.prepend(first, last);
-  AtomicAccess::add(&_num_segments, num, memory_order_relaxed);
-  AtomicAccess::add(&_mem_size, mem_size, memory_order_relaxed);
+  _num_segments.add_then_fetch(num, memory_order_relaxed);
+  _mem_size.add_then_fetch(mem_size, memory_order_relaxed);
 }
 
 void G1MonotonicArena::SegmentFreeList::print_on(outputStream* out, const char* prefix) {
   out->print_cr("%s: segments %zu size %zu",
-                prefix, AtomicAccess::load(&_num_segments), AtomicAccess::load(&_mem_size));
+                prefix, _num_segments.load_relaxed(), _mem_size.load_relaxed());
 }
 
 G1MonotonicArena::Segment* G1MonotonicArena::SegmentFreeList::get_all(size_t& num_segments,
@@ -75,12 +74,12 @@ G1MonotonicArena::Segment* G1MonotonicArena::SegmentFreeList::get_all(size_t& nu
   GlobalCounter::CriticalSection cs(Thread::current());
 
   Segment* result = _list.pop_all();
-  num_segments = AtomicAccess::load(&_num_segments);
-  mem_size = AtomicAccess::load(&_mem_size);
+  num_segments = _num_segments.load_relaxed();
+  mem_size = _mem_size.load_relaxed();
 
   if (result != nullptr) {
-    AtomicAccess::sub(&_num_segments, num_segments, memory_order_relaxed);
-    AtomicAccess::sub(&_mem_size, mem_size, memory_order_relaxed);
+    _num_segments.sub_then_fetch(num_segments, memory_order_relaxed);
+    _mem_size.sub_then_fetch(mem_size, memory_order_relaxed);
   }
   return result;
 }
@@ -96,8 +95,8 @@ void G1MonotonicArena::SegmentFreeList::free_all() {
     Segment::delete_segment(cur);
   }
 
-  AtomicAccess::sub(&_num_segments, num_freed, memory_order_relaxed);
-  AtomicAccess::sub(&_mem_size, mem_size_freed, memory_order_relaxed);
+  _num_segments.sub_then_fetch(num_freed, memory_order_relaxed);
+  _mem_size.sub_then_fetch(mem_size_freed, memory_order_relaxed);
 }
 
 G1MonotonicArena::Segment* G1MonotonicArena::new_segment(Segment* const prev) {
@@ -115,7 +114,7 @@ G1MonotonicArena::Segment* G1MonotonicArena::new_segment(Segment* const prev) {
   }
 
   // Install it as current allocation segment.
-  Segment* old = AtomicAccess::cmpxchg(&_first, prev, next);
+  Segment* old = _first.compare_exchange(prev, next);
   if (old != prev) {
     // Somebody else installed the segment, use that one.
     Segment::delete_segment(next);
@@ -126,9 +125,9 @@ G1MonotonicArena::Segment* G1MonotonicArena::new_segment(Segment* const prev) {
       _last = next;
     }
     // Successfully installed the segment into the list.
-    AtomicAccess::inc(&_num_segments, memory_order_relaxed);
-    AtomicAccess::add(&_mem_size, next->mem_size(), memory_order_relaxed);
-    AtomicAccess::add(&_num_total_slots, next->num_slots(), memory_order_relaxed);
+    _num_segments.add_then_fetch(1u, memory_order_relaxed);
+    _mem_size.add_then_fetch(next->mem_size(), memory_order_relaxed);
+    _num_total_slots.add_then_fetch(next->num_slots(), memory_order_relaxed);
     return next;
   }
 }
@@ -155,7 +154,7 @@ uint G1MonotonicArena::slot_size() const {
 }
 
 void G1MonotonicArena::drop_all() {
-  Segment* cur = AtomicAccess::load_acquire(&_first);
+  Segment* cur = _first.load_acquire();
 
   if (cur != nullptr) {
     assert(_last != nullptr, "If there is at least one segment, there must be a last one.");
@@ -175,25 +174,25 @@ void G1MonotonicArena::drop_all() {
       cur = next;
     }
 #endif
-    assert(num_segments == _num_segments, "Segment count inconsistent %u %u", num_segments, _num_segments);
-    assert(mem_size == _mem_size, "Memory size inconsistent");
+    assert(num_segments == _num_segments.load_relaxed(), "Segment count inconsistent %u %u", num_segments, _num_segments.load_relaxed());
+    assert(mem_size == _mem_size.load_relaxed(), "Memory size inconsistent");
     assert(last == _last, "Inconsistent last segment");
 
-    _segment_free_list->bulk_add(*first, *_last, _num_segments, _mem_size);
+    _segment_free_list->bulk_add(*first, *_last, _num_segments.load_relaxed(), _mem_size.load_relaxed());
   }
 
-  _first = nullptr;
+  _first.store_relaxed(nullptr);
   _last = nullptr;
-  _num_segments = 0;
-  _mem_size = 0;
-  _num_total_slots = 0;
-  _num_allocated_slots = 0;
+  _num_segments.store_relaxed(0);
+  _mem_size.store_relaxed(0);
+  _num_total_slots.store_relaxed(0);
+  _num_allocated_slots.store_relaxed(0);
 }
 
 void* G1MonotonicArena::allocate() {
   assert(slot_size() > 0, "instance size not set.");
 
-  Segment* cur = AtomicAccess::load_acquire(&_first);
+  Segment* cur = _first.load_acquire();
   if (cur == nullptr) {
     cur = new_segment(cur);
   }
@@ -201,7 +200,7 @@ void* G1MonotonicArena::allocate() {
   while (true) {
     void* slot = cur->allocate_slot();
     if (slot != nullptr) {
-      AtomicAccess::inc(&_num_allocated_slots, memory_order_relaxed);
+      _num_allocated_slots.add_then_fetch(1u, memory_order_relaxed);
       guarantee(is_aligned(slot, _alloc_options->slot_alignment()),
                 "result " PTR_FORMAT " not aligned at %u", p2i(slot), _alloc_options->slot_alignment());
       return slot;
@@ -213,7 +212,7 @@ void* G1MonotonicArena::allocate() {
 }
 
 uint G1MonotonicArena::num_segments() const {
-  return AtomicAccess::load(&_num_segments);
+  return _num_segments.load_relaxed();
 }
 
 #ifdef ASSERT
@@ -238,7 +237,7 @@ uint G1MonotonicArena::calculate_length() const {
 
 template <typename SegmentClosure>
 void G1MonotonicArena::iterate_segments(SegmentClosure& closure) const {
-  Segment* cur = AtomicAccess::load_acquire(&_first);
+  Segment* cur = _first.load_acquire();
 
   assert((cur != nullptr) == (_last != nullptr),
          "If there is at least one segment, there must be a last one");

src/hotspot/share/gc/g1/g1MonotonicArena.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2021, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2021, 2026, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2021, 2022, Huawei Technologies Co., Ltd. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
@@ -28,6 +28,7 @@
 
 #include "gc/shared/freeListAllocator.hpp"
 #include "nmt/memTag.hpp"
+#include "runtime/atomic.hpp"
 #include "utilities/globalDefinitions.hpp"
 #include "utilities/lockFreeStack.hpp"
 
@@ -65,27 +66,27 @@ private:
   // AllocOptions provides parameters for Segment sizing and expansion.
   const AllocOptions* _alloc_options;
 
-  Segment* volatile _first;       // The (start of the) list of all segments.
-  Segment* _last;                 // The last segment of the list of all segments.
-  volatile uint _num_segments;    // Number of assigned segments to this allocator.
-  volatile size_t _mem_size;      // Memory used by all segments.
+  Atomic<Segment*> _first;       // The (start of the) list of all segments.
+  Segment* _last;                // The last segment of the list of all segments.
+  Atomic<uint> _num_segments;    // Number of assigned segments to this allocator.
+  Atomic<size_t> _mem_size;      // Memory used by all segments.
 
   SegmentFreeList* _segment_free_list;  // The global free segment list to preferentially
                                         // get new segments from.
 
-  volatile uint _num_total_slots; // Number of slots available in all segments (allocated + not yet used).
-  volatile uint _num_allocated_slots; // Number of total slots allocated ever (including free and pending).
+  Atomic<uint> _num_total_slots; // Number of slots available in all segments (allocated + not yet used).
+  Atomic<uint> _num_allocated_slots; // Number of total slots allocated ever (including free and pending).
 
   inline Segment* new_segment(Segment* const prev);
 
   DEBUG_ONLY(uint calculate_length() const;)
 
 public:
-  const Segment* first_segment() const { return AtomicAccess::load(&_first); }
+  const Segment* first_segment() const { return _first.load_relaxed(); }
 
-  uint num_total_slots() const { return AtomicAccess::load(&_num_total_slots); }
+  uint num_total_slots() const { return _num_total_slots.load_relaxed(); }
   uint num_allocated_slots() const {
-    uint allocated = AtomicAccess::load(&_num_allocated_slots);
+    uint allocated = _num_allocated_slots.load_relaxed();
     assert(calculate_length() == allocated, "Must be");
     return allocated;
   }
@@ -116,11 +117,11 @@ static constexpr uint SegmentPayloadMaxAlignment = 8;
 class alignas(SegmentPayloadMaxAlignment) G1MonotonicArena::Segment {
   const uint _slot_size;
   const uint _num_slots;
-  Segment* volatile _next;
+  Atomic<Segment*> _next;
   // Index into the next free slot to allocate into. Full if equal (or larger)
   // to _num_slots (can be larger because we atomically increment this value and
   // check only afterwards if the allocation has been successful).
-  uint volatile _next_allocate;
+  Atomic<uint> _next_allocate;
   const MemTag _mem_tag;
 
   static size_t header_size() { return align_up(sizeof(Segment), SegmentPayloadMaxAlignment); }
@@ -139,21 +140,21 @@ class alignas(SegmentPayloadMaxAlignment) G1MonotonicArena::Segment {
   Segment(uint slot_size, uint num_slots, Segment* next, MemTag mem_tag);
   ~Segment() = default;
 public:
-  Segment* volatile* next_addr() { return &_next; }
+  Atomic<Segment*>* next_addr() { return &_next; }
 
   void* allocate_slot();
 
   uint num_slots() const { return _num_slots; }
 
-  Segment* next() const { return _next; }
+  Segment* next() const { return _next.load_relaxed(); }
 
   void set_next(Segment* next) {
     assert(next != this, " loop condition");
-    _next = next;
+    _next.store_relaxed(next);
   }
 
   void reset(Segment* next) {
-    _next_allocate = 0;
+    _next_allocate.store_relaxed(0);
     assert(next != this, " loop condition");
     set_next(next);
     memset(payload(0), 0, payload_size());
@@ -166,7 +167,7 @@ public:
   uint length() const {
     // _next_allocate might grow larger than _num_slots in multi-thread environments
     // due to races.
-    return MIN2(_next_allocate, _num_slots);
+    return MIN2(_next_allocate.load_relaxed(), _num_slots);
   }
 
   static size_t size_in_bytes(uint slot_size, uint num_slots) {
@@ -176,7 +177,7 @@ public:
   static Segment* create_segment(uint slot_size, uint num_slots, Segment* next, MemTag mem_tag);
   static void delete_segment(Segment* segment);
 
-  bool is_full() const { return _next_allocate >= _num_slots; }
+  bool is_full() const { return _next_allocate.load_relaxed() >= _num_slots; }
 };
 
 static_assert(alignof(G1MonotonicArena::Segment) >= SegmentPayloadMaxAlignment, "assert alignment of Segment (and indirectly its payload)");
@@ -186,15 +187,15 @@ static_assert(alignof(G1MonotonicArena::Segment) >= SegmentPayloadMaxAlignment,
 // performed by multiple threads concurrently.
 // Counts and memory usage are current on a best-effort basis if accessed concurrently.
 class G1MonotonicArena::SegmentFreeList {
-  static Segment* volatile* next_ptr(Segment& segment) {
+  static Atomic<Segment*>* next_ptr(Segment& segment) {
     return segment.next_addr();
   }
   using SegmentStack = LockFreeStack<Segment, &SegmentFreeList::next_ptr>;
 
   SegmentStack _list;
 
-  volatile size_t _num_segments;
-  volatile size_t _mem_size;
+  Atomic<size_t> _num_segments;
+  Atomic<size_t> _mem_size;
 
 public:
   SegmentFreeList() : _list(), _num_segments(0), _mem_size(0) { }
@@ -210,8 +211,8 @@ public:
 
   void print_on(outputStream* out, const char* prefix = "");
 
-  size_t num_segments() const { return AtomicAccess::load(&_num_segments); }
-  size_t mem_size() const { return AtomicAccess::load(&_mem_size); }
+  size_t num_segments() const { return _num_segments.load_relaxed(); }
+  size_t mem_size() const { return _mem_size.load_relaxed(); }
 };
 
 // Configuration for G1MonotonicArena, e.g slot size, slot number of next Segment.

src/hotspot/share/gc/g1/g1MonotonicArena.inline.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2021, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2021, 2026, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2021, 2022, Huawei Technologies Co., Ltd. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
@@ -28,14 +28,13 @@
 
 #include "gc/g1/g1MonotonicArena.hpp"
 
-#include "runtime/atomicAccess.hpp"
 #include "utilities/globalCounter.inline.hpp"
 
 inline void* G1MonotonicArena::Segment::allocate_slot() {
-  if (_next_allocate >= _num_slots) {
+  if (_next_allocate.load_relaxed() >= _num_slots) {
     return nullptr;
   }
-  uint result = AtomicAccess::fetch_then_add(&_next_allocate, 1u, memory_order_relaxed);
+  uint result = _next_allocate.fetch_then_add(1u, memory_order_relaxed);
   if (result >= _num_slots) {
     return nullptr;
   }
@@ -48,8 +47,8 @@ inline G1MonotonicArena::Segment* G1MonotonicArena::SegmentFreeList::get() {
 
   Segment* result = _list.pop();
   if (result != nullptr) {
-    AtomicAccess::dec(&_num_segments, memory_order_relaxed);
-    AtomicAccess::sub(&_mem_size, result->mem_size(), memory_order_relaxed);
+    _num_segments.sub_then_fetch(1u, memory_order_relaxed);
+    _mem_size.sub_then_fetch(result->mem_size(), memory_order_relaxed);
   }
   return result;
 }

src/hotspot/share/gc/g1/g1PageBasedVirtualSpace.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2014, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2014, 2026, 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
@@ -28,7 +28,6 @@
 #include "nmt/memTracker.hpp"
 #include "oops/markWord.hpp"
 #include "oops/oop.inline.hpp"
-#include "runtime/atomicAccess.hpp"
 #include "runtime/os.hpp"
 #include "utilities/align.hpp"
 #include "utilities/bitMap.inline.hpp"