ADDITIONAL SYSTEM INFORMATION :
4.18.0-553.34.1.
RHEL8_10.x86_64

A DESCRIPTION OF THE PROBLEM :
After upgrading to Java 21 with G1 from Java 11 we've noticed constant crashes with SIGSEGV in different places - JVM code, core and 3d party libraries
First we've identified an issue with corrupted pointers and AddressSanitizer issues (ERROR: AddressSanitizer:
use-after-poison on address 0x7cb7f70115e8 at pc 0x7fc7de8a5c7a bp 0x7cb7b7bf9160 sp 0x7cb7b7bf9158 READ of size 4 at 0x7cb7f701158 thread T339 (GC Thread#78))
------------------------------------------------------------------------------------------------------
[2025-02-13T11:33:28.864+0200][info ][gc ] GC(2461) g1RemSet.cpp assert failed: Region 38073 should be free region but is EDEN
...
[2025-02-13T11:33:29.660+0200][info ][gc ] Class pointer is corrupted: 0x0000000000000020 CompressedKlassPointers::decode_raw result

Then figured out that there's some mismatch between expected and actual region collection sizes:
-------------------------------------------------------------------------------------
[2025-02-14T16:12:04.902+0200][info ][gc ] GC(192) g1CollectionSet.cpp assert failed: Young region length 976 should match collection set length 975
[2025-02-14T16:12:04.943+0200][info ][gc ] GC(192) Stack trace for Young region length 976 should match collection set length 975:
Native frames: (J=compiled Java code, j=interpreted, Vv=VM code, C=native code)
V [libjvm.so+0x7e6639] G1CollectionSet::finalize_young_part(double, G1SurvivorRegions*)+0xe9 (g1CollectionSet.cpp:97)
V [libjvm.so+0x7e6b31] G1CollectionSet::finalize_initial_collection_set(double, G1SurvivorRegions*)+0x11 (g1CollectionSet.cpp:397)
V [libjvm.so+0x862050] G1YoungCollector::calculate_collection_set(G1EvacInfo*, double)+0x40 (g1YoungCollector.cpp:269)
V [libjvm.so+0x86222c] G1YoungCollector::pre_evacuate_collection_set(G1EvacInfo*)+0xec (g1YoungCollector.cpp:475)
V [libjvm.so+0x863603] G1YoungCollector::collect()+0x2f3 (g1YoungCollector.cpp:1042)
V [libjvm.so+0x7e05f6] G1CollectedHeap::do_collection_pause_at_safepoint_helper()+0xf6 (g1CollectedHeap.cpp:2573)
V [libjvm.so+0x7e06b8] G1CollectedHeap::do_collection_pause_at_safepoint()+0x28 (g1CollectedHeap.cpp:2495)
V [libjvm.so+0x861c21] VM_G1CollectForAllocation::doit()+0x61 (g1VMOperations.cpp:142)
V [libjvm.so+0x102094e] VM_Operation::evaluate()+0xfe (vmOperations.cpp:73)
V [libjvm.so+0x1024b88] VMThread::inner_execute(VM_Operation*)+0x2a8 (vmThread.cpp:281)
V [libjvm.so+0x102500e] VMThread::loop()+0x11e (vmThread.cpp:502)
V [libjvm.so+0x102527c] VMThread::run()+0x6c (vmThread.cpp:175)
V [libjvm.so+0xf9061f] Thread::call_run()+0x9f (thread.cpp:221)
V [libjvm.so+0xcd62aa] thread_native_entry(Thread*)+0xda (os_linux.cpp:790)
-------------------------------------------------------------------------------------

Then we've reproduced this issue with fast-debug build and identified failing assert:
Internal Error (/src/hotspot/share/gc/g1/g1AllocRegion.cpp:135), pid=3110139, tid=3239892
# assert((_alloc_region == _dummy_region)) failed: [Mutator Alloc Region] pre-condition c: 165 r: 0x00007c8203cfde90 u: 0
#
# Problematic frame:
# V [libjvm.so+0xb94112] G1AllocRegion::new_alloc_region_and_allocate(unsigned long, bool)+0x1f2

Unfortunately our application is pretty big and requires lots of memory (terabytes) so it's hard to come up with small reproducer but it's reproduces on our environments consistently every time with HotSpot, Adoptium and Coretto


REGRESSION : Last worked in version 11.0.26


CUSTOMER SUBMITTED WORKAROUND :
ZGC works fine but we would like to find workaround for G1 as well
Java 24 doesn't have this issue

FREQUENCY : always