Merge pull request #2164 from SAP/pr-jdk-26+33

Merge to tag jdk-26+33

Author: RealCLanger

doc/testing.html

@@ -72,6 +72,7 @@ <h1 class="title">Testing the JDK</h1>
 <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 @@ <h3 id="pkcs11-tests">PKCS11 Tests</h3>
     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

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/g1CollectedHeap.hpp

@@ -1267,7 +1267,6 @@ class G1CollectedHeap : public CollectedHeap {
 
   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

@@ -38,6 +38,7 @@
 #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"
@@ -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/parallel/psParallelCompact.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2005, 2025, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2005, 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
@@ -44,6 +44,7 @@
 #include "gc/parallel/psStringDedup.hpp"
 #include "gc/parallel/psYoungGen.hpp"
 #include "gc/shared/classUnloadingContext.hpp"
+#include "gc/shared/collectedHeap.inline.hpp"
 #include "gc/shared/fullGCForwarding.inline.hpp"
 #include "gc/shared/gcCause.hpp"
 #include "gc/shared/gcHeapSummary.hpp"
@@ -932,6 +933,17 @@ void PSParallelCompact::summary_phase(bool should_do_max_compaction)
   }
 }
 
+void PSParallelCompact::report_object_count_after_gc() {
+  GCTraceTime(Debug, gc, phases) tm("Report Object Count", &_gc_timer);
+  // The heap is compacted, all objects are iterable. However there may be
+  // filler objects in the heap which we should ignore.
+  class SkipFillerObjectClosure : public BoolObjectClosure {
+  public:
+    bool do_object_b(oop obj) override { return !CollectedHeap::is_filler_object(obj); }
+  } cl;
+  _gc_tracer.report_object_count_after_gc(&cl, &ParallelScavengeHeap::heap()->workers());
+}
+
 bool PSParallelCompact::invoke(bool clear_all_soft_refs, bool should_do_max_compaction) {
   assert(SafepointSynchronize::is_at_safepoint(), "should be at safepoint");
   assert(Thread::current() == (Thread*)VMThread::vm_thread(),
@@ -1027,6 +1039,8 @@ bool PSParallelCompact::invoke(bool clear_all_soft_refs, bool should_do_max_comp
 
     heap->print_heap_change(pre_gc_values);
 
+    report_object_count_after_gc();
+
     // Track memory usage and detect low memory
     MemoryService::track_memory_usage();
     heap->update_counters();
@@ -1274,10 +1288,6 @@ void PSParallelCompact::marking_phase(ParallelOldTracer *gc_tracer) {
     }
   }
 
-  {
-    GCTraceTime(Debug, gc, phases) tm("Report Object Count", &_gc_timer);
-    _gc_tracer.report_object_count_after_gc(is_alive_closure(), &ParallelScavengeHeap::heap()->workers());
-  }
 #if TASKQUEUE_STATS
   ParCompactionManager::print_and_reset_taskqueue_stats();
 #endif
@@ -1835,8 +1845,7 @@ void PSParallelCompact::verify_filler_in_dense_prefix() {
       oop obj = cast_to_oop(cur_addr);
       oopDesc::verify(obj);
       if (!mark_bitmap()->is_marked(cur_addr)) {
-        Klass* k = cast_to_oop(cur_addr)->klass();
-        assert(k == Universe::fillerArrayKlass() || k == vmClasses::FillerObject_klass(), "inv");
+        assert(CollectedHeap::is_filler_object(cast_to_oop(cur_addr)), "inv");
       }
       cur_addr += obj->size();
     }

src/hotspot/share/gc/parallel/psParallelCompact.hpp

@@ -749,6 +749,7 @@ class PSParallelCompact : AllStatic {
   // Move objects to new locations.
   static void compact();
 
+  static void report_object_count_after_gc();
   // Add available regions to the stack and draining tasks to the task queue.
   static void prepare_region_draining_tasks(uint parallel_gc_threads);
 

src/hotspot/share/gc/shared/collectedHeap.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
@@ -309,6 +309,8 @@ class CollectedHeap : public CHeapObj<mtGC> {
     fill_with_object(start, pointer_delta(end, start), zap);
   }
 
+  inline static bool is_filler_object(oop obj);
+
   virtual void fill_with_dummy_object(HeapWord* start, HeapWord* end, bool zap);
   static size_t min_dummy_object_size() {
     return oopDesc::header_size();

src/hotspot/share/gc/shared/collectedHeap.inline.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2001, 2019, 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 +27,9 @@
 
 #include "gc/shared/collectedHeap.hpp"
 
+#include "classfile/vmClasses.hpp"
 #include "gc/shared/memAllocator.hpp"
+#include "memory/universe.hpp"
 #include "oops/oop.inline.hpp"
 #include "utilities/align.hpp"
 
@@ -50,4 +52,9 @@ inline void CollectedHeap::add_vmthread_cpu_time(jlong time) {
   _vmthread_cpu_time += time;
 }
 
+inline bool CollectedHeap::is_filler_object(oop obj) {
+  Klass* k = obj->klass_without_asserts();
+  return k == Universe::fillerArrayKlass() || k == vmClasses::FillerObject_klass();
+}
+
 #endif // SHARE_GC_SHARED_COLLECTEDHEAP_INLINE_HPP

src/hotspot/share/opto/subnode.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
@@ -745,71 +745,40 @@ const Type* CmpINode::Value(PhaseGVN* phase) const {
 
 // Simplify a CmpU (compare 2 integers) node, based on local information.
 // If both inputs are constants, compare them.
-const Type *CmpUNode::sub( const Type *t1, const Type *t2 ) const {
-  assert(!t1->isa_ptr(), "obsolete usage of CmpU");
+const Type* CmpUNode::sub(const Type* t1, const Type* t2) const {
+  const TypeInt* r0 = t1->is_int();
+  const TypeInt* r1 = t2->is_int();
 
-  // comparing two unsigned ints
-  const TypeInt *r0 = t1->is_int();   // Handy access
-  const TypeInt *r1 = t2->is_int();
-
-  // Current installed version
-  // Compare ranges for non-overlap
-  juint lo0 = r0->_lo;
-  juint hi0 = r0->_hi;
-  juint lo1 = r1->_lo;
-  juint hi1 = r1->_hi;
-
-  // If either one has both negative and positive values,
-  // it therefore contains both 0 and -1, and since [0..-1] is the
-  // full unsigned range, the type must act as an unsigned bottom.
-  bool bot0 = ((jint)(lo0 ^ hi0) < 0);
-  bool bot1 = ((jint)(lo1 ^ hi1) < 0);
-
-  if (bot0 || bot1) {
-    // All unsigned values are LE -1 and GE 0.
-    if (lo0 == 0 && hi0 == 0) {
-      return TypeInt::CC_LE;            //   0 <= bot
-    } else if ((jint)lo0 == -1 && (jint)hi0 == -1) {
-      return TypeInt::CC_GE;            // -1 >= bot
-    } else if (lo1 == 0 && hi1 == 0) {
-      return TypeInt::CC_GE;            // bot >= 0
-    } else if ((jint)lo1 == -1 && (jint)hi1 == -1) {
-      return TypeInt::CC_LE;            // bot <= -1
-    }
-  } else {
-    // We can use ranges of the form [lo..hi] if signs are the same.
-    assert(lo0 <= hi0 && lo1 <= hi1, "unsigned ranges are valid");
-    // results are reversed, '-' > '+' for unsigned compare
-    if (hi0 < lo1) {
-      return TypeInt::CC_LT;            // smaller
-    } else if (lo0 > hi1) {
-      return TypeInt::CC_GT;            // greater
-    } else if (hi0 == lo1 && lo0 == hi1) {
-      return TypeInt::CC_EQ;            // Equal results
-    } else if (lo0 >= hi1) {
-      return TypeInt::CC_GE;
-    } else if (hi0 <= lo1) {
-      // Check for special case in Hashtable::get.  (See below.)
-      if ((jint)lo0 >= 0 && (jint)lo1 >= 0 && is_index_range_check())
-        return TypeInt::CC_LT;
-      return TypeInt::CC_LE;
-    }
-  }
   // Check for special case in Hashtable::get - the hash index is
   // mod'ed to the table size so the following range check is useless.
   // Check for: (X Mod Y) CmpU Y, where the mod result and Y both have
   // to be positive.
   // (This is a gross hack, since the sub method never
   // looks at the structure of the node in any other case.)
-  if ((jint)lo0 >= 0 && (jint)lo1 >= 0 && is_index_range_check())
+  if (r0->_lo >= 0 && r1->_lo >= 0 && is_index_range_check()) {
     return TypeInt::CC_LT;
+  }
+
+  if (r0->_uhi < r1->_ulo) {
+    return TypeInt::CC_LT;
+  } else if (r0->_ulo > r1->_uhi) {
+    return TypeInt::CC_GT;
+  } else if (r0->is_con() && r1->is_con()) {
+    // Since r0->_ulo == r0->_uhi == r0->get_con(), we only reach here if the constants are equal
+    assert(r0->get_con() == r1->get_con(), "must reach a previous branch otherwise");
+    return TypeInt::CC_EQ;
+  } else if (r0->_uhi == r1->_ulo) {
+    return TypeInt::CC_LE;
+  } else if (r0->_ulo == r1->_uhi) {
+    return TypeInt::CC_GE;
+  }
 
   const Type* joined = r0->join(r1);
   if (joined == Type::TOP) {
     return TypeInt::CC_NE;
   }
 
-  return TypeInt::CC;                   // else use worst case results
+  return TypeInt::CC;
 }
 
 const Type* CmpUNode::Value(PhaseGVN* phase) const {
@@ -963,59 +932,29 @@ const Type *CmpLNode::sub( const Type *t1, const Type *t2 ) const {
 // Simplify a CmpUL (compare 2 unsigned longs) node, based on local information.
 // If both inputs are constants, compare them.
 const Type* CmpULNode::sub(const Type* t1, const Type* t2) const {
-  assert(!t1->isa_ptr(), "obsolete usage of CmpUL");
-
-  // comparing two unsigned longs
-  const TypeLong* r0 = t1->is_long();   // Handy access
+  const TypeLong* r0 = t1->is_long();
   const TypeLong* r1 = t2->is_long();
 
-  // Current installed version
-  // Compare ranges for non-overlap
-  julong lo0 = r0->_lo;
-  julong hi0 = r0->_hi;
-  julong lo1 = r1->_lo;
-  julong hi1 = r1->_hi;
-
-  // If either one has both negative and positive values,
-  // it therefore contains both 0 and -1, and since [0..-1] is the
-  // full unsigned range, the type must act as an unsigned bottom.
-  bool bot0 = ((jlong)(lo0 ^ hi0) < 0);
-  bool bot1 = ((jlong)(lo1 ^ hi1) < 0);
-
-  if (bot0 || bot1) {
-    // All unsigned values are LE -1 and GE 0.
-    if (lo0 == 0 && hi0 == 0) {
-      return TypeInt::CC_LE;            //   0 <= bot
-    } else if ((jlong)lo0 == -1 && (jlong)hi0 == -1) {
-      return TypeInt::CC_GE;            // -1 >= bot
-    } else if (lo1 == 0 && hi1 == 0) {
-      return TypeInt::CC_GE;            // bot >= 0
-    } else if ((jlong)lo1 == -1 && (jlong)hi1 == -1) {
-      return TypeInt::CC_LE;            // bot <= -1
-    }
-  } else {
-    // We can use ranges of the form [lo..hi] if signs are the same.
-    assert(lo0 <= hi0 && lo1 <= hi1, "unsigned ranges are valid");
-    // results are reversed, '-' > '+' for unsigned compare
-    if (hi0 < lo1) {
-      return TypeInt::CC_LT;            // smaller
-    } else if (lo0 > hi1) {
-      return TypeInt::CC_GT;            // greater
-    } else if (hi0 == lo1 && lo0 == hi1) {
-      return TypeInt::CC_EQ;            // Equal results
-    } else if (lo0 >= hi1) {
-      return TypeInt::CC_GE;
-    } else if (hi0 <= lo1) {
-      return TypeInt::CC_LE;
-    }
+  if (r0->_uhi < r1->_ulo) {
+    return TypeInt::CC_LT;
+  } else if (r0->_ulo > r1->_uhi) {
+    return TypeInt::CC_GT;
+  } else if (r0->is_con() && r1->is_con()) {
+    // Since r0->_ulo == r0->_uhi == r0->get_con(), we only reach here if the constants are equal
+    assert(r0->get_con() == r1->get_con(), "must reach a previous branch otherwise");
+    return TypeInt::CC_EQ;
+  } else if (r0->_uhi == r1->_ulo) {
+    return TypeInt::CC_LE;
+  } else if (r0->_ulo == r1->_uhi) {
+    return TypeInt::CC_GE;
   }
 
   const Type* joined = r0->join(r1);
   if (joined == Type::TOP) {
     return TypeInt::CC_NE;
   }
 
-  return TypeInt::CC;                   // else use worst case results
+  return TypeInt::CC;
 }
 
 //=============================================================================