/*
    * Copyright 2022 The Netty Project
    *
    * The Netty Project licenses this file to you under the Apache License,
    * version 2.0 (the "License"); you may not use this file except in compliance
    * with the License. You may obtain a copy of the License at:
    *
    *   https://www.apache.org/licenses/LICENSE-2.0
    *
   * Unless required by applicable law or agreed to in writing, software
   * distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
   * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
   * License for the specific language governing permissions and limitations
   * under the License.
   */
  package io.netty.buffer;
  
  import io.netty.util.ByteProcessor;
  import io.netty.util.CharsetUtil;
  import io.netty.util.IllegalReferenceCountException;
  import io.netty.util.NettyRuntime;
  import io.netty.util.Recycler;
  import io.netty.util.Recycler.EnhancedHandle;
  import io.netty.util.concurrent.ConcurrentSkipListIntObjMultimap;
  import io.netty.util.concurrent.ConcurrentSkipListIntObjMultimap.IntEntry;
  import io.netty.util.concurrent.FastThreadLocal;
  import io.netty.util.concurrent.FastThreadLocalThread;
  import io.netty.util.concurrent.MpscIntQueue;
  import io.netty.util.internal.MathUtil;
  import io.netty.util.internal.ObjectUtil;
  import io.netty.util.internal.PlatformDependent;
  import io.netty.util.internal.RefCnt;
  import io.netty.util.internal.SystemPropertyUtil;
  import io.netty.util.internal.ThreadExecutorMap;
  import io.netty.util.internal.UnstableApi;
  
  import java.io.IOException;
  import java.io.InputStream;
  import java.io.OutputStream;
  import java.nio.ByteBuffer;
  import java.nio.ByteOrder;
  import java.nio.channels.ClosedChannelException;
  import java.nio.channels.FileChannel;
  import java.nio.channels.GatheringByteChannel;
  import java.nio.channels.ScatteringByteChannel;
  import java.nio.charset.Charset;
  import java.util.ArrayList;
  import java.util.Arrays;
  import java.util.Iterator;
  import java.util.Queue;
  import java.util.concurrent.atomic.AtomicInteger;
  import java.util.concurrent.atomic.AtomicIntegerFieldUpdater;
  import java.util.concurrent.atomic.AtomicLong;
  import java.util.concurrent.atomic.AtomicReferenceFieldUpdater;
  import java.util.concurrent.atomic.LongAdder;
  import java.util.concurrent.locks.StampedLock;
  import java.util.function.IntConsumer;
  
  /**
   * An auto-tuning pooling allocator, that follows an anti-generational hypothesis.
   * <p>
   * The allocator is organized into a list of Magazines, and each magazine has a chunk-buffer that they allocate buffers
   * from.
   * <p>
   * The magazines hold the mutexes that ensure the thread-safety of the allocator, and each thread picks a magazine
   * based on the id of the thread. This spreads the contention of multi-threaded access across the magazines.
   * If contention is detected above a certain threshold, the number of magazines are increased in response to the
   * contention.
   * <p>
   * The magazines maintain histograms of the sizes of the allocations they do. The histograms are used to compute the
   * preferred chunk size. The preferred chunk size is one that is big enough to service 10 allocations of the
   * 99-percentile size. This way, the chunk size is adapted to the allocation patterns.
   * <p>
   * Computing the preferred chunk size is a somewhat expensive operation. Therefore, the frequency with which this is
   * done, is also adapted to the allocation pattern. If a newly computed preferred chunk is the same as the previous
   * preferred chunk size, then the frequency is reduced. Otherwise, the frequency is increased.
   * <p>
   * This allows the allocator to quickly respond to changes in the application workload,
   * without suffering undue overhead from maintaining its statistics.
   * <p>
   * Since magazines are "relatively thread-local", the allocator has a chunk cache that allows excess chunks from any
   * magazine to be shared with other magazines.
   */
  @UnstableApi
  final class AdaptivePoolingAllocator {
      private static final int LOW_MEM_THRESHOLD = 512 * 1024 * 1024;
      private static final boolean IS_LOW_MEM = SystemPropertyUtil.getBoolean(
              "io.netty.allocator.lowMemory",
              Runtime.getRuntime().maxMemory() <= LOW_MEM_THRESHOLD);
  
      /**
       * Whether the IS_LOW_MEM setting should disable thread-local magazines.
       * This can have fairly high performance overhead.
       */
      private static final boolean DISABLE_THREAD_LOCAL_MAGAZINES_ON_LOW_MEM = SystemPropertyUtil.getBoolean(
              "io.netty.allocator.disableThreadLocalMagazinesOnLowMemory", true);
  
      /**
       * The 128 KiB minimum chunk size is chosen to encourage the system allocator to delegate to mmap for chunk
      * allocations. For instance, glibc will do this.
      * This pushes any fragmentation from chunk size deviations off physical memory, onto virtual memory,
      * which is a much, much larger space. Chunks are also allocated in whole multiples of the minimum
      * chunk size, which itself is a whole multiple of popular page sizes like 4 KiB, 16 KiB, and 64 KiB.
      */
     static final int MIN_CHUNK_SIZE = 128 * 1024;
     private static final int EXPANSION_ATTEMPTS = 3;
     private static final int INITIAL_MAGAZINES = 1;
     private static final int RETIRE_CAPACITY = 256;
     private static final int MAX_STRIPES = IS_LOW_MEM ? 1 : NettyRuntime.availableProcessors() * 2;
     private static final int BUFS_PER_CHUNK = 8; // For large buffers, aim to have about this many buffers per chunk.
 
     /**
      * The maximum size of a pooled chunk, in bytes. Allocations bigger than this will never be pooled.
      * <p>
      * This number is 8 MiB, and is derived from the limitations of internal histograms.
      */
     private static final int MAX_CHUNK_SIZE = IS_LOW_MEM ?
             2 * 1024 * 1024 : // 2 MiB for systems with small heaps.
             8 * 1024 * 1024; // 8 MiB.
     private static final int MAX_POOLED_BUF_SIZE = MAX_CHUNK_SIZE / BUFS_PER_CHUNK;
 
     /**
      * The capacity if the chunk reuse queues, that allow chunks to be shared across magazines in a group.
      * The default size is twice {@link NettyRuntime#availableProcessors()},
      * same as the maximum number of magazines per magazine group.
      */
     static final int CHUNK_REUSE_QUEUE = Math.max(2, SystemPropertyUtil.getInt(
             "io.netty.allocator.chunkReuseQueueCapacity", NettyRuntime.availableProcessors() * 2));
 
     static final long CHUNK_PURGE_POLLS_THREAD_LOCAL = Math.max(1, SystemPropertyUtil.getLong(
             "io.netty.allocator.chunkPurgePollsThreadLocal", 16L));
 
     static final long CHUNK_PURGE_POLLS_SHARED = Math.max(1, SystemPropertyUtil.getLong(
             "io.netty.allocator.chunkPurgePollsShared", 128L));
 
     static final int CHUNK_PURGE_THRESHOLD = Math.max(1, SystemPropertyUtil.getInt(
             "io.netty.allocator.chunkPurgeThreshold", 3));
 
     /**
      * Per-size-class upper bound (in bytes) on the thread-local chunk cache.
      * When a size class cache holds this many bytes worth of chunks,
      * further offers are rejected and the chunk is marked for immediate deallocation.
      * Chunks already in the cache are only evicted by the purge mechanism (they must be full and idle for
      * {@link #CHUNK_PURGE_THRESHOLD} consecutive purge cycles).
      */
     static final int THREAD_LOCAL_CACHE_MAX_BYTES = Math.max(1, SystemPropertyUtil.getInt(
             "io.netty.allocator.threadLocalChunkCacheMaxBytes", 8 * 1024 * 1024));
 
     /**
      * Per-size-class lower bound (in bytes) on the thread-local chunk cache.
      * The purge mechanism will not evict chunks below this retention floor, even if they are full and idle.
      * Clamped to {@link #THREAD_LOCAL_CACHE_MAX_BYTES} if the configured value exceeds it.
      * When equal to {@link #THREAD_LOCAL_CACHE_MAX_BYTES}, purge eviction is effectively disabled.
      */
     static final int THREAD_LOCAL_CACHE_MIN_BYTES = Math.min(THREAD_LOCAL_CACHE_MAX_BYTES,
             Math.max(1, SystemPropertyUtil.getInt(
                     "io.netty.allocator.threadLocalChunkCacheMinBytes",
                     THREAD_LOCAL_CACHE_MAX_BYTES / 2)));
 
     /**
      * The capacity if the magazine local buffer queue. This queue just pools the outer ByteBuf instance and not
      * the actual memory and so helps to reduce GC pressure.
      */
     private static final int MAGAZINE_BUFFER_QUEUE_CAPACITY = SystemPropertyUtil.getInt(
             "io.netty.allocator.magazineBufferQueueCapacity", 1024);
 
     /**
      * The size classes are chosen based on the following observation:
      * <p>
      * Most allocations, particularly ones above 256 bytes, aim to be a power-of-2. However, many use cases, such
      * as framing protocols, are themselves operating or moving power-of-2 sized payloads, to which they add a
      * small amount of overhead, such as headers or checksums.
      * This means we seem to get a lot of mileage out of having both power-of-2 sizes, and power-of-2-plus-a-bit.
      * <p>
      * On the conflicting requirements of both having as few chunks as possible, and having as little wasted
      * memory within each chunk as possible, this seems to strike a surprisingly good balance for the use cases
      * tested so far.
      */
     private static final int[] SIZE_CLASSES = {
             32,
             64,
             128,
             256,
             512,
             640, // 512 + 128
             1024,
             1152, // 1024 + 128
             2048,
             2304, // 2048 + 256
             4096,
             4352, // 4096 + 256
             8192,
             8704, // 8192 + 512
             16384,
             16896, // 16384 + 512
     };
 
     private static final int SIZE_CLASSES_COUNT = SIZE_CLASSES.length;
     private static final byte[] SIZE_INDEXES = new byte[SIZE_CLASSES[SIZE_CLASSES_COUNT - 1] / 32 + 1];
 
     static {
         if (MAGAZINE_BUFFER_QUEUE_CAPACITY < 2) {
             throw new IllegalArgumentException("MAGAZINE_BUFFER_QUEUE_CAPACITY: " + MAGAZINE_BUFFER_QUEUE_CAPACITY
                     + " (expected: >= " + 2 + ')');
         }
         int lastIndex = 0;
         for (int i = 0; i < SIZE_CLASSES_COUNT; i++) {
             int sizeClass = SIZE_CLASSES[i];
             //noinspection ConstantValue
             assert (sizeClass & 31) == 0 : "Size class must be a multiple of 32";
             int sizeIndex = sizeIndexOf(sizeClass);
             Arrays.fill(SIZE_INDEXES, lastIndex + 1, sizeIndex + 1, (byte) i);
             lastIndex = sizeIndex;
         }
     }
 
     private final ChunkAllocator chunkAllocator;
     private final ChunkRegistry chunkRegistry;
     private final MagazineGroup[] sizeClassedMagazineGroups;
     private final MagazineGroup largeBufferMagazineGroup;
     private final FastThreadLocal<MagazineGroup[]> threadLocalGroup;
 
     AdaptivePoolingAllocator(ChunkAllocator chunkAllocator, boolean useCacheForNonEventLoopThreads) {
         this.chunkAllocator = ObjectUtil.checkNotNull(chunkAllocator, "chunkAllocator");
         chunkRegistry = new ChunkRegistry();
         sizeClassedMagazineGroups = createMagazineGroupSizeClasses(this, false);
         largeBufferMagazineGroup = new MagazineGroup(
                 this, chunkAllocator, new BuddyChunkManagementStrategy(), false);
 
         boolean disableThreadLocalGroups = IS_LOW_MEM && DISABLE_THREAD_LOCAL_MAGAZINES_ON_LOW_MEM;
         threadLocalGroup = disableThreadLocalGroups ? null : new FastThreadLocal<MagazineGroup[]>() {
             @Override
             protected MagazineGroup[] initialValue() {
                 if (useCacheForNonEventLoopThreads || ThreadExecutorMap.currentExecutor() != null) {
                     return createMagazineGroupSizeClasses(AdaptivePoolingAllocator.this, true);
                 }
                 return null;
             }
 
             @Override
             protected void onRemoval(final MagazineGroup[] groups) throws Exception {
                 if (groups != null) {
                     for (MagazineGroup group : groups) {
                         group.free();
                     }
                 }
             }
         };
     }
 
     private static MagazineGroup[] createMagazineGroupSizeClasses(
             AdaptivePoolingAllocator allocator, boolean isThreadLocal) {
         MagazineGroup[] groups = new MagazineGroup[SIZE_CLASSES.length];
         for (int i = 0; i < SIZE_CLASSES.length; i++) {
             int segmentSize = SIZE_CLASSES[i];
             groups[i] = new MagazineGroup(allocator, allocator.chunkAllocator,
                     new SizeClassChunkManagementStrategy(segmentSize), isThreadLocal);
         }
         return groups;
     }
 
     ByteBuf allocate(int size, int maxCapacity) {
         return allocate(size, maxCapacity, Thread.currentThread(), null);
     }
 
     private AdaptiveByteBuf allocate(int size, int maxCapacity, Thread currentThread, AdaptiveByteBuf buf) {
         AdaptiveByteBuf allocated = null;
         if (size <= MAX_POOLED_BUF_SIZE) {
             final int index = sizeClassIndexOf(size);
             MagazineGroup[] magazineGroups;
             if (!FastThreadLocalThread.currentThreadWillCleanupFastThreadLocals() ||
                     IS_LOW_MEM ||
                     (magazineGroups = threadLocalGroup.get()) == null) {
                 magazineGroups = sizeClassedMagazineGroups;
             }
             if (index < magazineGroups.length) {
                 allocated = magazineGroups[index].allocate(size, maxCapacity, currentThread, buf);
             } else if (!IS_LOW_MEM) {
                 allocated = largeBufferMagazineGroup.allocate(size, maxCapacity, currentThread, buf);
             }
         }
         if (allocated == null) {
             allocated = allocateFallback(size, maxCapacity, currentThread, buf);
         }
         return allocated;
     }
 
     private static int sizeIndexOf(final int size) {
         // this is aligning the size to the next multiple of 32 and dividing by 32 to get the size index.
         return size + 31 >> 5;
     }
 
     static int sizeClassIndexOf(int size) {
         int sizeIndex = sizeIndexOf(size);
         if (sizeIndex < SIZE_INDEXES.length) {
             return SIZE_INDEXES[sizeIndex];
         }
         return SIZE_CLASSES_COUNT;
     }
 
     static int[] getSizeClasses() {
         return SIZE_CLASSES.clone();
     }
 
     private AdaptiveByteBuf allocateFallback(int size, int maxCapacity, Thread currentThread, AdaptiveByteBuf buf) {
         // If we don't already have a buffer, obtain one from the most conveniently available magazine.
         Magazine magazine;
         if (buf != null) {
             Chunk chunk = buf.chunk;
             if (chunk == null || chunk == Magazine.MAGAZINE_FREED || (magazine = chunk.currentMagazine()) == null) {
                 magazine = getFallbackMagazine(currentThread);
             }
         } else {
             magazine = getFallbackMagazine(currentThread);
             buf = magazine.newBuffer();
         }
         // Create a one-off chunk for this allocation.
         AbstractByteBuf innerChunk = chunkAllocator.allocate(size, maxCapacity);
         Chunk chunk = new Chunk(innerChunk, magazine, false);
         chunkRegistry.add(chunk);
         try {
             boolean success = chunk.readInitInto(buf, size, size, maxCapacity);
             assert success : "Failed to initialize ByteBuf with dedicated chunk";
         } finally {
             // As the chunk is an one-off we need to always call release explicitly as readInitInto(...)
             // will take care of retain once when successful. Once The AdaptiveByteBuf is released it will
             // completely release the Chunk and so the contained innerChunk.
             chunk.release();
         }
         return buf;
     }
 
     private Magazine getFallbackMagazine(Thread currentThread) {
         Magazine[] mags = largeBufferMagazineGroup.magazines;
         return mags[(int) currentThread.getId() & mags.length - 1];
     }
 
     /**
      * Allocate into the given buffer. Used by {@link AdaptiveByteBuf#capacity(int)}.
      */
     void reallocate(int size, int maxCapacity, AdaptiveByteBuf into) {
         AdaptiveByteBuf result = allocate(size, maxCapacity, Thread.currentThread(), into);
         assert result == into : "Re-allocation created separate buffer instance";
     }
 
     long usedMemory() {
         return chunkRegistry.totalCapacity();
     }
 
     // Ensure that we release all previous pooled resources when this object is finalized. This is needed as otherwise
     // we might end up with leaks. While these leaks are usually harmless in reality it would still at least be
     // very confusing for users.
     @SuppressWarnings({"FinalizeDeclaration", "deprecation"})
     @Override
     protected void finalize() throws Throwable {
         try {
             free();
         } finally {
             super.finalize();
         }
     }
 
     private void free() {
         largeBufferMagazineGroup.free();
     }
 
     private static final class MagazineGroup {
         private final AdaptivePoolingAllocator allocator;
         private final ChunkAllocator chunkAllocator;
         private final ChunkManagementStrategy chunkManagementStrategy;
         private final ChunkCache chunkCache;
         private final StampedLock magazineExpandLock;
         private final Magazine threadLocalMagazine;
         private Thread ownerThread;
         private volatile Magazine[] magazines;
         private volatile boolean freed;
 
         MagazineGroup(AdaptivePoolingAllocator allocator,
                       ChunkAllocator chunkAllocator,
                       ChunkManagementStrategy chunkManagementStrategy,
                       boolean isThreadLocal) {
             this.allocator = allocator;
             this.chunkAllocator = chunkAllocator;
             this.chunkManagementStrategy = chunkManagementStrategy;
             chunkCache = chunkManagementStrategy.createChunkCache(isThreadLocal);
             if (isThreadLocal) {
                 ownerThread = Thread.currentThread();
                 magazineExpandLock = null;
                 threadLocalMagazine = new Magazine(this, false, chunkManagementStrategy.createController(this));
             } else {
                 ownerThread = null;
                 magazineExpandLock = new StampedLock();
                 threadLocalMagazine = null;
                 Magazine[] mags = new Magazine[INITIAL_MAGAZINES];
                 for (int i = 0; i < mags.length; i++) {
                     mags[i] = new Magazine(this, true, chunkManagementStrategy.createController(this));
                 }
                 magazines = mags;
             }
         }
 
         public AdaptiveByteBuf allocate(int size, int maxCapacity, Thread currentThread, AdaptiveByteBuf buf) {
             boolean reallocate = buf != null;
 
             // Path for thread-local allocation.
             Magazine tlMag = threadLocalMagazine;
             if (tlMag != null) {
                 if (buf == null) {
                     buf = tlMag.newBuffer();
                 }
                 boolean allocated = tlMag.tryAllocate(size, maxCapacity, buf, reallocate);
                 assert allocated : "Allocation of threadLocalMagazine must always succeed";
                 return buf;
             }
 
             // Path for concurrent allocation.
             long threadId = currentThread.getId();
             Magazine[] mags;
             int expansions = 0;
             do {
                 mags = magazines;
                 int mask = mags.length - 1;
                 int index = (int) (threadId & mask);
                 for (int i = 0, m = mags.length << 1; i < m; i++) {
                     Magazine mag = mags[index + i & mask];
                     if (buf == null) {
                         buf = mag.newBuffer();
                     }
                     if (mag.tryAllocate(size, maxCapacity, buf, reallocate)) {
                         // Was able to allocate.
                         return buf;
                     }
                 }
                 expansions++;
             } while (expansions <= EXPANSION_ATTEMPTS && tryExpandMagazines(mags.length));
 
             // The magazines failed us; contention too high and we don't want to spend more effort expanding the array.
             if (!reallocate && buf != null) {
                 buf.release(); // Release the previously claimed buffer before we return.
             }
             return null;
         }
 
         private boolean tryExpandMagazines(int currentLength) {
             if (currentLength >= MAX_STRIPES) {
                 return true;
             }
             final Magazine[] mags;
             long writeLock = magazineExpandLock.tryWriteLock();
             if (writeLock != 0) {
                 try {
                     mags = magazines;
                     if (mags.length >= MAX_STRIPES || mags.length > currentLength || freed) {
                         return true;
                     }
                     Magazine[] expanded = new Magazine[mags.length * 2];
                     for (int i = 0, l = expanded.length; i < l; i++) {
                         expanded[i] = new Magazine(this, true, chunkManagementStrategy.createController(this));
                     }
                     magazines = expanded;
                 } finally {
                     magazineExpandLock.unlockWrite(writeLock);
                 }
                 for (Magazine magazine : mags) {
                     magazine.free();
                 }
             }
             return true;
         }
 
         Chunk pollChunk(int size) {
             return chunkCache.pollChunk(size);
         }
 
         boolean offerChunk(Chunk chunk) {
             if (freed) {
                 return false;
             }
 
             if (chunk.hasUnprocessedFreelistEntries()) {
                 chunk.processFreelistEntries();
             }
             boolean isAdded = chunkCache.offerChunk(chunk);
 
             if (freed && isAdded) {
                 // Help to free the reuse queue.
                 freeChunkReuseQueue(ownerThread);
             }
             return isAdded;
         }
 
         private void free() {
             freed = true;
             Thread ownerThread = this.ownerThread;
             if (threadLocalMagazine != null) {
                 this.ownerThread = null;
                 threadLocalMagazine.free();
             } else {
                 long stamp = magazineExpandLock.writeLock();
                 try {
                     Magazine[] mags = magazines;
                     for (Magazine magazine : mags) {
                         magazine.free();
                     }
                 } finally {
                     magazineExpandLock.unlockWrite(stamp);
                 }
             }
             freeChunkReuseQueue(ownerThread);
         }
 
         private void freeChunkReuseQueue(Thread ownerThread) {
             if (ownerThread != null && chunkCache instanceof ThreadLocalSizeClassedChunkCache) {
                 ThreadLocalSizeClassedChunkCache tlCache = (ThreadLocalSizeClassedChunkCache) chunkCache;
                 int mask = tlCache.chunks.length - 1;
                 for (int i = 0; i < tlCache.count; i++) {
                     SizeClassedChunk chunk = tlCache.chunks[(tlCache.head + i) & mask];
                     assert ownerThread == chunk.ownerThread;
                     chunk.ownerThread = null;
                 }
             }
             chunkCache.free();
         }
     }
 
     interface ChunkCache {
         Chunk pollChunk(int size);
 
         boolean offerChunk(Chunk chunk);
 
         void free();
 
         boolean isEmpty();
     }
 
     // Cached chunks are detached from magazines: no readInitInto can happen, so segment count
     // can only grow (external releaseSegment returns) and never shrink. Once a chunk reaches
     // full capacity (hasFullCapacity), it stays idle while in the cache.
     //
     // Epoch-based aging invariants (both caches):
     //
     // 1. CLASSIFICATION: purge scans all chunks. Idle (hasFullCapacity) → epoch++.
     //    Non-idle → epoch = 0. Only idle chunks can accumulate epoch.
     //
     // 2. EVICTION: idle chunks with epoch > CHUNK_PURGE_THRESHOLD are evicted (markToDeallocate).
     //    Eviction is immediate — all segments are in, no outstanding references.
     //    Non-idle chunks are never evicted (deallocation would be deferred, not immediate).
     //    A retention floor prevents over-eviction: CHUNK_REUSE_QUEUE for the shared cache,
     //    purgeRetentionFloor (from THREAD_LOCAL_CACHE_MIN_BYTES) for the thread-local cache.
     //
     // 3. SCAN RESET: scanForCapacity resets purgeEpoch = 0 on the chunk it picks. The scan
     //    knows the chunk is being used. The chunk gets allocated from, becomes non-idle, and
     //    the next purge resets its epoch anyway (non-idle → 0). The scan reset covers the case
     //    where all segments return before the next purge (short-lived buffers).
     //
     // 4. CONVERGENCE: idle chunks that are never picked by scan age undisturbed across
     //    purge cycles. After CHUNK_PURGE_THRESHOLD + 1 consecutive cycles of being idle and
     //    unpolled, they are evicted. Chunks picked by scan get epoch reset — aging interrupted.
     //    Thread-local: partition orders [epoch=0 | 0<epoch<T | epoch>=T | noCap]. Scan takes
     //    from head (epoch=0 first). Chunks with epoch>=threshold are placed at the back of
     //    the hasCap zone so scan doesn't reach them — they age to threshold+1 and get evicted.
     //    Shared: approximate, converges over multiple cycles (FIFO queue ordering,
     //    LRU preference in scan, retained counter in purge).
     abstract static class SizeClassedChunkCache implements ChunkCache {
         static SizeClassedChunkCache create(boolean isThreadLocal, int chunkSize) {
             return isThreadLocal ? new ThreadLocalSizeClassedChunkCache(chunkSize) :
                     new SharedSizeClassedChunkCache();
         }
 
         @Override
         public abstract SizeClassedChunk pollChunk(int size);
 
         // Visible for testing: triggers a purge scan bypassing the budget counter.
         abstract SizeClassedChunk forcePurge();
     }
 
     /**
      * Ring buffer cache for thread-local chunk reuse (SPSC — only the owner thread accesses it).
      *
      * <p>Logical layout after purge:
      * <pre>
      *   head                          tail
      *   v                             v
      *   [..., notEmpty, notEmpty, ..., empty, empty, ..., null, ...]
      *        |--- notEmptyCount ---|--- emptyCount --|
      *        |------------ count ------------------|
      * </pre>
      *
      * <p>Physical layout when the ring wraps:
      * <pre>
      *   0         tail          head          length
      *   v         v             v             v
      *   [...tail] [  unused  ]  [head................]
      *             ^             |--- content wraps ---|
      *             wrap point
      * </pre>
      *
      * <p><b>scanForCapacity</b> — O(1) fast path takes from head while {@code notEmptyCount > 0}:
      * <pre>
      *   before: notEmptyCount=2, count=5
      *   [NE, NE, E, E, E, _, _, _]
      *    ^head            ^tail
      *
      *   after: returns NE, notEmptyCount=1, count=4
      *   [_,  NE, E, E, E, _, _, _]
      *        ^head        ^tail
      * </pre>
      * Fallback when {@code notEmptyCount == 0}: linear scan of the empty zone for chunks
      * that gained capacity from external segment returns.
      *
      * <p><b>offerChunk</b> — write at tail, grow (double + linearize) if full:
      * <pre>
      *   before: count=4
      *   [_,  NE, E, E, E, _, _, _]
      *        ^head        ^tail
      *
      *   after: count=5
      *   [_,  NE, E, E, E, X, _, _]
      *        ^head           ^tail
      * </pre>
      *
      * <p><b>runPurgeScan</b> (every {@link #CHUNK_PURGE_POLLS_THREAD_LOCAL} polls) —
      * two passes. Pass 1: age idle chunks (full → epoch++, non-full → epoch=0), evict
      * past threshold, compact survivors (nulls stale slots inline). Pass 2: partition
      * hasCap to front / noCap to back, then three-way Dutch-flag within hasCap into
      * [epoch=0 | 0&lt;epoch&lt;threshold | epoch&gt;=threshold]. Chunks with epoch&gt;=threshold
      * are placed at the back of hasCap so scan doesn't reach them — they age to
      * threshold+1 and get evicted. Never selects — selection is always
      * {@code scanForCapacity}.
      *
      * <p>Case 1 — no eviction, an empty chunk gained capacity externally (common):
      * <pre>
      *   before (E* gained capacity since last purge):
      *   [NE, NE, E*, E, _, _, _, _]
      *    ^head            ^tail
      *    notEmptyCount=2
      *
      *   pass 1: age idle chunks. None past threshold. No compaction needed.
      *   pass 2 (partition): E* now has capacity → placed in notEmpty zone.
      *
      *   after:
      *   [NE, NE, E*, E, _, _, _, _]
      *    ^head            ^tail
      *    notEmptyCount=3
      * </pre>
      *
      * <p>Case 2 — eviction (uncommon, burst wind-down):
      * <pre>
      *   before (ring wraps, IDLE* = idle past threshold):
      *   [E, NE, _,  IDLE*, NE, E, E, NE]
      *          ^tail ^head
      *
      *   pass 1: IDLE* evicted (markToDeallocate), survivors compacted, stale slots nulled.
      *   [_, _, _,  NE, E, E, NE, E]
      *     ^tail    ^head
      *              |--- kept=6 ---|
      *
      *   pass 2 (partition): [epoch=0 hasCap | 0&lt;epoch&lt;T hasCap | epoch&gt;=T hasCap | noCap].
      *   [_, _, _,  NE, NE, E, E, E]
      *     ^tail    ^head
      *              notEmptyCount=2, count=6
      * </pre>
      * Idle chunks ({@code remainingCapacity == capacity}) age via purgeEpoch and are evicted
      * past threshold, but at least {@code purgeRetentionFloor} chunks are always retained.
      */
     static final class ThreadLocalSizeClassedChunkCache extends SizeClassedChunkCache {
         SizeClassedChunk[] chunks; // package-private for testing
         int head;
         int tail;
         int count;
         int notEmptyCount;
         private long purgeBudget;
         final int maxCachedChunks; // package-private for testing
         final int purgeRetentionFloor; // package-private for testing
 
         ThreadLocalSizeClassedChunkCache(int chunkSize) {
             chunks = new SizeClassedChunk[8];
             purgeBudget = CHUNK_PURGE_POLLS_THREAD_LOCAL;
             maxCachedChunks = Math.max(1, THREAD_LOCAL_CACHE_MAX_BYTES / chunkSize);
             purgeRetentionFloor = Math.min(maxCachedChunks,
                     Math.max(1, THREAD_LOCAL_CACHE_MIN_BYTES / chunkSize));
         }
 
         @Override
         SizeClassedChunk forcePurge() {
             purgeBudget = 1;
             return pollChunk(0);
         }
 
         @Override
         public SizeClassedChunk pollChunk(int size) {
             if (--purgeBudget == 0) {
                 runPurgeScan();
             }
             return scanForCapacity();
         }
 
         private SizeClassedChunk scanForCapacity() {
             if (notEmptyCount > 0) {
                 SizeClassedChunk chunk = chunks[head];
                 assert chunk.hasRemainingCapacity();
                 chunk.purgeEpoch = 0;
                 chunks[head] = null;
                 head = (head + 1) & (chunks.length - 1);
                 count--;
                 notEmptyCount--;
                 return chunk;
             }
             return scanForCapacityFallback();
         }
 
         private SizeClassedChunk scanForCapacityFallback() {
             int mask = chunks.length - 1;
             int emptyCount = count - notEmptyCount;
             int pos = (head + notEmptyCount) & mask;
             for (int i = 0; i < emptyCount; i++) {
                 SizeClassedChunk chunk = chunks[pos];
                 if (chunk.hasRemainingCapacity()) {
                     chunk.purgeEpoch = 0;
                     int lastIdx = (tail - 1) & mask;
                     chunks[pos] = chunks[lastIdx];
                     chunks[lastIdx] = null;
                     tail = lastIdx;
                     count--;
                     return chunk;
                 }
                 pos = (pos + 1) & mask;
             }
             return null;
         }
 
         private void runPurgeScan() {
             int mask = chunks.length - 1;
             int kept = 0;
             int survivors = count;
             for (int i = 0; i < count; i++) {
                 int readIdx = (head + i) & mask;
                 SizeClassedChunk chunk = chunks[readIdx];
                 if (chunk.purgeEpoch > 0) {
                     assert chunk.hasFullCapacity();
                     chunk.purgeEpoch++;
                     if (chunk.purgeEpoch > CHUNK_PURGE_THRESHOLD && survivors > purgeRetentionFloor) {
                         chunk.markToDeallocate();
                         chunks[readIdx] = null;
                         survivors--;
                         continue;
                     }
                 } else if (chunk.hasFullCapacity()) {
                     chunk.purgeEpoch = 1;
                 }
                 int writeIdx = (head + kept) & mask;
                 if (writeIdx != readIdx) {
                     chunks[writeIdx] = chunk;
                     chunks[readIdx] = null;
                 }
                 kept++;
             }
             tail = (head + kept) & mask;
             count = kept;
             partition(kept);
             purgeBudget = CHUNK_PURGE_POLLS_THREAD_LOCAL;
         }
 
         private void partition(int size) {
             int mask = chunks.length - 1;
             // Pass 1: hasCapacity to front, noCapacity to back.
             int lo = 0;
             int hi = size - 1;
             while (lo <= hi) {
                 int loIdx = (head + lo) & mask;
                 if (chunks[loIdx].hasRemainingCapacity()) {
                     lo++;
                 } else {
                     int hiIdx = (head + hi) & mask;
                     SizeClassedChunk tmp = chunks[loIdx];
                     chunks[loIdx] = chunks[hiIdx];
                     chunks[hiIdx] = tmp;
                     hi--;
                 }
             }
             notEmptyCount = lo;
             // Pass 2: three-way Dutch-flag within notEmpty:
             //   [epoch=0 | 0<epoch<threshold | epoch>=threshold]
             //
             // Epoch=0 (recently used) at head — scan picks these first.
             // Epoch>=threshold (about to be evicted) at back — scan doesn't reach them,
             // so they age one more cycle to threshold+1 and get evicted.
             //
             // This ordering guarantees convergence regardless of count/polls ratio.
             // Without it (e.g., a simple epoch=0/epoch>0 split with mid++), when
             // count/polls == threshold the groups rotate perfectly and max epoch never
             // exceeds threshold — eviction stalls at threshold * polls chunks.
             int elo = 0;
             int emid = 0;
             int ehi = lo - 1;
             while (emid <= ehi) {
                 int emidIdx = (head + emid) & mask;
                 SizeClassedChunk c = chunks[emidIdx];
                 if (c.purgeEpoch == 0) {
                     if (elo != emid) {
                         int eloIdx = (head + elo) & mask;
                         chunks[emidIdx] = chunks[eloIdx];
                         chunks[eloIdx] = c;
                     }
                     elo++;
                     emid++;
                 } else if (c.purgeEpoch < CHUNK_PURGE_THRESHOLD) {
                     emid++;
                 } else {
                     int ehiIdx = (head + ehi) & mask;
                     chunks[emidIdx] = chunks[ehiIdx];
                     chunks[ehiIdx] = c;
                     ehi--;
                 }
             }
         }
 
         @Override
         public boolean offerChunk(Chunk chunk) {
             if (count >= maxCachedChunks) {
                 return false;
             }
             if (count == chunks.length) {
                 SizeClassedChunk[] newChunks = new SizeClassedChunk[chunks.length * 2];
                 for (int i = 0; i < count; i++) {
                     newChunks[i] = chunks[(head + i) & (chunks.length - 1)];
                 }
                 chunks = newChunks;
                 head = 0;
                 tail = count;
             }
             chunks[tail] = (SizeClassedChunk) chunk;
             tail = (tail + 1) & (chunks.length - 1);
             count++;
             return true;
         }
 
         @Override
         public String toString() {
             int mask = chunks.length - 1;
             StringBuilder sb = new StringBuilder();
             sb.append("ThreadLocalCache[head=").append(head)
                     .append(", tail=").append(tail)
                     .append(", count=").append(count)
                     .append(", notEmpty=").append(notEmptyCount)
                     .append(", length=").append(chunks.length)
                     .append("]\n  ");
             for (int i = 0; i < count; i++) {
                 if (i > 0) {
                     sb.append(", ");
                 }
                 if (i == notEmptyCount) {
                     sb.append("| ");
                 }
                 SizeClassedChunk c = chunks[(head + i) & mask];
                 String region = i < notEmptyCount ? "notEmpty" : "empty";
                 String actual = c == null ? "null" :
                         c.hasRemainingCapacity() ? "hasCap" : "noCap";
                 sb.append('[').append(region).append(':').append(actual)
                         .append(",ep=").append(c == null ? -1 : c.purgeEpoch).append(']');
             }
             return sb.toString();
         }
 
         @Override
         public void free() {
             int mask = chunks.length - 1;
             for (int i = 0; i < count; i++) {
                 int idx = (head + i) & mask;
                 chunks[idx].markToDeallocate();
                 chunks[idx] = null;
             }
             head = 0;
             tail = 0;
             count = 0;
             notEmptyCount = 0;
         }
 
         @Override
         public boolean isEmpty() {
             return count == 0;
         }
     }
 
     /**
      * MPMC queue cache for shared (cross-thread) chunk reuse.
      *
      * <p><b>scanForCapacity</b> — LRU preference with fallback:
      * <pre>
      *   fast path: head chunk has purgeEpoch == 0 and capacity → return O(1)
      *
      *   slow path: scan for epoch=0 chunk, hold first idle (epoch &gt; 0) as fallback
      *     queue: [E&gt;0, E&gt;0, E=0, E&gt;0, ...]
      *             skip   skip  ↑ return (put fallback back)
      *
      *   no epoch=0 found → use fallback, reset its epoch to 0
      * </pre>
      *
      * <p>The LRU preference creates a natural separation: recently-used chunks (epoch=0,
      * returned via {@link #offerChunk} after magazine use) cycle at the front. Idle chunks
      * (epoch &gt; 0, aged by purge) are scanned past but never returned — they age undisturbed.
      * When no recently-used chunks exist, idle ones are reused (fallback) rather than
      * allocating new chunks.
      *
      * <p>All re-offered chunks are stamped with {@code lastScanGeneration} for cycle detection.
      * The {@code >=} check terminates the scan when encountering any chunk already processed
      * by this or a later scan, preventing livelock under concurrent access.
      *
      * <p><b>runPurgeScan</b> (every {@link #CHUNK_PURGE_POLLS_SHARED} polls):
      * drains the queue, ages full chunks (epoch++), resets non-full (epoch=0).
      * Non-candidate capacity chunks are re-offered inline. Eviction candidates (full,
      * epoch past threshold) and no-capacity chunks are deferred to a buffer. After the drain,
      * the buffer is walked with the known total: candidates are evicted while above
      * {@link #CHUNK_REUSE_QUEUE}, remainder re-offered. No selection — that is
      * {@code scanForCapacity}'s job (called after purge via {@code pollChunk}).
      */
     static final class SharedSizeClassedChunkCache extends SizeClassedChunkCache {
         // Must exceed CHUNK_REUSE_QUEUE (the retention floor) to leave room for burst absorption.
         // TODO replace with an unbounded concurrent collection once available.
         private static final int SHARED_CACHE_CAPACITY = Math.max(16, CHUNK_REUSE_QUEUE * 2);
         private final Queue<SizeClassedChunk> queue;
         private final AtomicLong purgeBudget;
         private final ArrayList<SizeClassedChunk> deferredBuffer = new ArrayList<>();
         private long purgeGeneration;
         private final AtomicLong scanGeneration = new AtomicLong();
 
         SharedSizeClassedChunkCache() {
             queue = PlatformDependent.newFixedMpmcQueue(SHARED_CACHE_CAPACITY);
             purgeBudget = new AtomicLong(CHUNK_PURGE_POLLS_SHARED);
         }
 
         @Override
         SizeClassedChunk forcePurge() {
             purgeBudget.set(1);
             return pollChunk(0);
         }
 
         @Override
         public SizeClassedChunk pollChunk(int size) {
             long budget = purgeBudget.decrementAndGet();
             if (budget == 0) {
                 runPurgeScan();
             }
             return scanForCapacity();
         }
 
         private SizeClassedChunk scanForCapacity() {
             SizeClassedChunk first = queue.poll();
             if (first == null) {
                 return null;
             }
             if (first.purgeEpoch == 0 && first.hasRemainingCapacity()) {
                 return first;
             }
             long generation = scanGeneration.incrementAndGet();
             first.lastScanGeneration = generation;
             if (first.hasRemainingCapacity()) {
                 return scanForCapacitySlow(generation, first);
             }
             offerOrDeallocate(first);
             return scanForCapacitySlow(generation, null);
         }
 
         private SizeClassedChunk scanForCapacitySlow(long generation, SizeClassedChunk fallback) {
             SizeClassedChunk chunk;
             while ((chunk = queue.poll()) != null) {
                 if (chunk.lastScanGeneration >= generation) {
                     offerOrDeallocate(chunk);
                     break;
                 }
                 if (chunk.hasRemainingCapacity()) {
                     if (chunk.purgeEpoch == 0) {
                         if (fallback != null) {
                             offerOrDeallocate(fallback);
                         }
                         return chunk;
                     }
                     if (fallback == null) {
                         fallback = chunk;
                         continue;
                     }
                 }
                 chunk.lastScanGeneration = generation;
                 offerOrDeallocate(chunk);
             }
             if (fallback != null) {
                 fallback.purgeEpoch = 0;
                 return fallback;
             }
             return null;
         }
 
         private boolean offerOrDeallocate(SizeClassedChunk chunk) {
             if (!queue.offer(chunk)) {
                 chunk.markToDeallocate();
                 return false;
             }
             return true;
         }
 
         private boolean offerOrDeallocate(SizeClassedChunk chunk, long generation) {
             chunk.lastPurgeGeneration = generation;
             return offerOrDeallocate(chunk);
         }
 
         private void runPurgeScan() {
             long generation = ++purgeGeneration;
             int retained = 0;
             ArrayList<SizeClassedChunk> deferred = deferredBuffer;
             SizeClassedChunk chunk;
             while ((chunk = queue.poll()) != null) {
                 if (chunk.lastPurgeGeneration == generation) {
                     offerOrDeallocate(chunk, generation);
                     break;
                 }
                 retained++;
                 if (chunk.hasFullCapacity()) {
                     chunk.purgeEpoch++;
                     if (chunk.purgeEpoch > CHUNK_PURGE_THRESHOLD) {
                         deferred.add(chunk);
                         continue;
                     }
                 } else {
                     chunk.purgeEpoch = 0;
                 }
                 int remaining = chunk.remainingCapacity();
                 if (remaining > 0) {
                     if (!offerOrDeallocate(chunk, generation)) {
                         retained--;
                     }
                 } else {
                     deferred.add(chunk);
                 }
             }
             for (int i = 0, size = deferred.size(); i < size; i++) {
                 chunk = deferred.get(i);
                 if (chunk.purgeEpoch > CHUNK_PURGE_THRESHOLD && retained > CHUNK_REUSE_QUEUE) {
                     chunk.markToDeallocate();
                     retained--;
                 } else {
                     if (!offerOrDeallocate(chunk, generation)) {
                         retained--;
                     }
                 }
             }
             deferred.clear();
             purgeBudget.lazySet(CHUNK_PURGE_POLLS_SHARED);
         }
 
         @Override
         public boolean offerChunk(Chunk chunk) {
             return queue.offer((SizeClassedChunk) chunk);
         }
 
         @Override
         public void free() {
             SizeClassedChunk chunk;
             while ((chunk = queue.poll()) != null) {
                 chunk.markToDeallocate();
             }
         }
 
         @Override
         public boolean isEmpty() {
             return queue.isEmpty();
         }
     }
 
     private static final class ConcurrentSkipListChunkCache implements ChunkCache {
         private final ConcurrentSkipListIntObjMultimap<Chunk> chunks;
 
         private ConcurrentSkipListChunkCache() {
             chunks = new ConcurrentSkipListIntObjMultimap<>(-1);
         }
 
         @Override
         public Chunk pollChunk(int size) {
             if (chunks.isEmpty()) {
                 return null;
             }
             IntEntry<Chunk> entry = chunks.pollCeilingEntry(size);
             if (entry != null) {
                 Chunk chunk = entry.getValue();
                 if (chunk.hasUnprocessedFreelistEntries()) {
                     chunk.processFreelistEntries();
                 }
                 return chunk;
             }
 
             Chunk bestChunk = null;
             int bestRemainingCapacity = 0;
             Iterator<IntEntry<Chunk>> itr = chunks.iterator();
             while (itr.hasNext()) {
                 entry = itr.next();
                 final Chunk chunk;
                 if (entry != null && (chunk = entry.getValue()).hasUnprocessedFreelistEntries()) {
                     if (!chunks.remove(entry.getKey(), entry.getValue())) {
                         continue;
                     }
                     chunk.processFreelistEntries();
                     int remainingCapacity = chunk.remainingCapacity();
                     if (remainingCapacity >= size &&
                             (bestChunk == null || remainingCapacity > bestRemainingCapacity)) {
                         if (bestChunk != null) {
                             chunks.put(bestRemainingCapacity, bestChunk);
                         }
                         bestChunk = chunk;
                         bestRemainingCapacity = remainingCapacity;
                     } else {
                         chunks.put(remainingCapacity, chunk);
                     }
                 }
             }
 
             return bestChunk;
         }
 
         @Override
         public boolean offerChunk(Chunk chunk) {
             chunks.put(chunk.remainingCapacity(), chunk);
 
             int size = chunks.size();
             while (size > CHUNK_REUSE_QUEUE) {
                 int key = -1;
                 Chunk toDeallocate = null;
                 for (IntEntry<Chunk> entry : chunks) {
                     Chunk candidate = entry.getValue();
                     if (candidate != null && RefCnt.refCnt(candidate.refCnt) == 1) {
                         toDeallocate = candidate;
                         key = entry.getKey();
                         break;
                     }
                 }
                 if (toDeallocate == null) {
                     break;
                 }
                 if (chunks.remove(key, toDeallocate)) {
                     toDeallocate.markToDeallocate();
                 }
                 size = chunks.size();
             }
             return true;
         }
 
         @Override
         public void free() {
             for (IntEntry<Chunk> entry : chunks) {
                 Chunk chunk = entry.getValue();
                 if (chunk != null && chunks.remove(entry.getKey(), chunk)) {
                     chunk.markToDeallocate();
                 }
             }
         }
 
         @Override
         public boolean isEmpty() {
             return chunks.isEmpty();
         }
     }
 
     private interface ChunkManagementStrategy {
         ChunkController createController(MagazineGroup group);
 
         ChunkCache createChunkCache(boolean isThreadLocal);
     }
 
     private interface ChunkController {
         /**
          * Compute the "fast max capacity" value for the buffer.
          */
         int computeBufferCapacity(int requestedSize, int maxCapacity, boolean isReallocation);
 
         /**
          * Allocate a new {@link Chunk} for the given {@link Magazine}.
          */
         Chunk newChunkAllocation(int promptingSize, Magazine magazine);
     }
 
     private static final class SizeClassChunkManagementStrategy implements ChunkManagementStrategy {
         // To amortize activation/deactivation of chunks, we should have a minimum number of segments per chunk.
         // We choose 32 because it seems neither too small nor too big.
         // For segments of 16 KiB, the chunks will be half a megabyte.
         private static final int MIN_SEGMENTS_PER_CHUNK = 32;
         private final int segmentSize;
         private final int chunkSize;
 
         private SizeClassChunkManagementStrategy(int segmentSize) {
             this.segmentSize = ObjectUtil.checkPositive(segmentSize, "segmentSize");
             chunkSize = Math.max(MIN_CHUNK_SIZE, segmentSize * MIN_SEGMENTS_PER_CHUNK);
         }
 
         @Override
         public ChunkController createController(MagazineGroup group) {
             return new SizeClassChunkController(group, segmentSize, chunkSize);
         }
 
         @Override
         public ChunkCache createChunkCache(boolean isThreadLocal) {
             return SizeClassedChunkCache.create(isThreadLocal, chunkSize);
         }
     }
 
     private static final class SizeClassChunkController implements ChunkController {
 
         private final ChunkAllocator chunkAllocator;
         private final int segmentSize;
         private final int chunkSize;
         private final ChunkRegistry chunkRegistry;
 
         private SizeClassChunkController(MagazineGroup group, int segmentSize, int chunkSize) {
             chunkAllocator = group.chunkAllocator;
             this.segmentSize = segmentSize;
             this.chunkSize = chunkSize;
             chunkRegistry = group.allocator.chunkRegistry;
         }
 
         private MpscIntQueue createEmptyFreeList() {
             return MpscIntQueue.create(chunkSize / segmentSize, SizeClassedChunk.FREE_LIST_EMPTY);
         }
 
         private MpscIntQueue createFreeList() {
             final int segmentsCount = chunkSize / segmentSize;
             final MpscIntQueue freeList = MpscIntQueue.create(segmentsCount, SizeClassedChunk.FREE_LIST_EMPTY);
             int segmentOffset = 0;
             for (int i = 0; i < segmentsCount; i++) {
                 freeList.offer(segmentOffset);
                 segmentOffset += segmentSize;
             }
             return freeList;
         }
 
         private IntStack createLocalFreeList() {
             final int segmentsCount = chunkSize / segmentSize;
             int segmentOffset = chunkSize;
             int[] offsets = new int[segmentsCount];
             for (int i = 0; i < segmentsCount; i++) {
                 segmentOffset -= segmentSize;
                 offsets[i] = segmentOffset;
             }
             return new IntStack(offsets);
         }
 
         @Override
         public int computeBufferCapacity(
                 int requestedSize, int maxCapacity, boolean isReallocation) {
             return Math.min(segmentSize, maxCapacity);
         }
 
         @Override
         public Chunk newChunkAllocation(int promptingSize, Magazine magazine) {
             AbstractByteBuf chunkBuffer = chunkAllocator.allocate(chunkSize, chunkSize);
             assert chunkBuffer.capacity() == chunkSize;
             SizeClassedChunk chunk = new SizeClassedChunk(chunkBuffer, magazine, this);
             chunkRegistry.add(chunk);
             return chunk;
         }
     }
 
     private static final class BuddyChunkManagementStrategy implements ChunkManagementStrategy {
         private final AtomicInteger maxChunkSize = new AtomicInteger();
 
         @Override
         public ChunkController createController(MagazineGroup group) {
             return new BuddyChunkController(group, maxChunkSize);
         }
 
         @Override
         public ChunkCache createChunkCache(boolean isThreadLocal) {
             return new ConcurrentSkipListChunkCache();
         }
     }
 
     private static final class BuddyChunkController implements ChunkController {
         private final ChunkAllocator chunkAllocator;
         private final ChunkRegistry chunkRegistry;
         private final AtomicInteger maxChunkSize;
 
         BuddyChunkController(MagazineGroup group, AtomicInteger maxChunkSize) {
             chunkAllocator = group.chunkAllocator;
             chunkRegistry = group.allocator.chunkRegistry;
             this.maxChunkSize = maxChunkSize;
         }
 
         @Override
         public int computeBufferCapacity(int requestedSize, int maxCapacity, boolean isReallocation) {
             return MathUtil.safeFindNextPositivePowerOfTwo(requestedSize);
         }
 
         @Override
         public Chunk newChunkAllocation(int promptingSize, Magazine magazine) {
             int maxChunkSize = this.maxChunkSize.get();
             int proposedChunkSize = MathUtil.safeFindNextPositivePowerOfTwo(BUFS_PER_CHUNK * promptingSize);
             int chunkSize = Math.min(MAX_CHUNK_SIZE, Math.max(maxChunkSize, proposedChunkSize));
             if (chunkSize > maxChunkSize) {
                 // Update our stored max chunk size. It's fine that this is racy.
                 this.maxChunkSize.set(chunkSize);
             }
             BuddyChunk chunk = new BuddyChunk(chunkAllocator.allocate(chunkSize, chunkSize), magazine);
             chunkRegistry.add(chunk);
             return chunk;
         }
     }
 
     private static final class Magazine {
         private static final AtomicReferenceFieldUpdater<Magazine, Chunk> NEXT_IN_LINE;
 
         static {
             NEXT_IN_LINE = AtomicReferenceFieldUpdater.newUpdater(Magazine.class, Chunk.class, "nextInLine");
         }
 
         private static final Chunk MAGAZINE_FREED = new Chunk();
 
         private static final class AdaptiveRecycler extends Recycler<AdaptiveByteBuf> {
 
             private AdaptiveRecycler(boolean unguarded) {
                 // uses fast thread local
                 super(unguarded);
             }
 
             private AdaptiveRecycler(int maxCapacity, boolean unguarded) {
                 // doesn't use fast thread local, shared
                 super(maxCapacity, unguarded);
             }
 
             @Override
             protected AdaptiveByteBuf newObject(final Handle<AdaptiveByteBuf> handle) {
                 return new AdaptiveByteBuf((EnhancedHandle<AdaptiveByteBuf>) handle);
             }
 
             public static AdaptiveRecycler threadLocal() {
                 return new AdaptiveRecycler(true);
             }
 
             public static AdaptiveRecycler sharedWith(int maxCapacity) {
                 return new AdaptiveRecycler(maxCapacity, true);
             }
         }
 
         private static final AdaptiveRecycler EVENT_LOOP_LOCAL_BUFFER_POOL = AdaptiveRecycler.threadLocal();
 
         private Chunk current;
         @SuppressWarnings("unused") // updated via NEXT_IN_LINE
         private volatile Chunk nextInLine;
         private final MagazineGroup group;
         private final ChunkController chunkController;
         private final StampedLock allocationLock;
         private final AdaptiveRecycler recycler;
 
         Magazine(MagazineGroup group, boolean shareable, ChunkController chunkController) {
             this.group = group;
             this.chunkController = chunkController;
 
             if (shareable) {
                 // We only need the StampedLock if this Magazine will be shared across threads.
                 allocationLock = new StampedLock();
                 recycler = AdaptiveRecycler.sharedWith(MAGAZINE_BUFFER_QUEUE_CAPACITY);
             } else {
                 allocationLock = null;
                 recycler = null;
             }
         }
 
         public boolean tryAllocate(int size, int maxCapacity, AdaptiveByteBuf buf, boolean reallocate) {
             if (allocationLock == null) {
                 // This magazine is not shared across threads, just allocate directly.
                 return allocate(size, maxCapacity, buf, reallocate);
             }
 
             // Try to retrieve the lock and if successful allocate.
             long writeLock = allocationLock.tryWriteLock();
             if (writeLock != 0) {
                 try {
                     return allocate(size, maxCapacity, buf, reallocate);
                 } finally {
                     allocationLock.unlockWrite(writeLock);
                 }
             }
             return allocateWithoutLock(size, maxCapacity, buf);
         }
 
         private boolean allocateWithoutLock(int size, int maxCapacity, AdaptiveByteBuf buf) {
             Chunk curr = NEXT_IN_LINE.getAndSet(this, null);
             if (curr == MAGAZINE_FREED) {
                 // Allocation raced with a stripe-resize that freed this magazine.
                 restoreMagazineFreed();
                 return false;
             }
             if (curr == null) {
                 curr = group.pollChunk(size);
                 if (curr == null) {
                     return false;
                 }
                 curr.attachToMagazine(this);
             }
             boolean allocated = false;
             int remainingCapacity = curr.remainingCapacity();
             int startingCapacity = chunkController.computeBufferCapacity(
                     size, maxCapacity, true /* never update stats as we don't hold the magazine lock */);
             if (remainingCapacity >= size &&
                     curr.readInitInto(buf, size, Math.min(remainingCapacity, startingCapacity), maxCapacity)) {
                 allocated = true;
                 remainingCapacity = curr.remainingCapacity();
             }
             try {
                 if (remainingCapacity >= RETIRE_CAPACITY) {
                     transferToNextInLineOrRelease(curr);
                     curr = null;
                 }
             } finally {
                 if (curr != null) {
                     curr.releaseFromMagazine();
                 }
             }
             return allocated;
         }
 
         private boolean allocate(int size, int maxCapacity, AdaptiveByteBuf buf, boolean reallocate) {
             int startingCapacity = chunkController.computeBufferCapacity(size, maxCapacity, reallocate);
             Chunk curr = current;
             if (curr != null) {
                 boolean success = curr.readInitInto(buf, size, startingCapacity, maxCapacity);
                 int remainingCapacity = curr.remainingCapacity();
                 if (!success && remainingCapacity > 0) {
                     current = null;
                     transferToNextInLineOrRelease(curr);
                 } else if (remainingCapacity == 0) {
                     current = null;
                     curr.releaseFromMagazine();
                 }
                 if (success) {
                     return true;
                 }
             }
 
             assert current == null;
             // The fast-path for allocations did not work.
             //
             // Try to fetch the next "Magazine local" Chunk first, if this fails because we don't have a
             // next-in-line chunk available, we will poll our centralQueue.
             // If this fails as well we will just allocate a new Chunk.
             //
             // In any case we will store the Chunk as the current so it will be used again for the next allocation and
             // thus be "reserved" by this Magazine for exclusive usage.
             curr = NEXT_IN_LINE.getAndSet(this, null);
             if (curr != null) {
                 if (curr == MAGAZINE_FREED) {
                     // Allocation raced with a stripe-resize that freed this magazine.
                     restoreMagazineFreed();
                     return false;
                 }
 
                 int remainingCapacity = curr.remainingCapacity();
                 if (remainingCapacity > startingCapacity &&
                         curr.readInitInto(buf, size, startingCapacity, maxCapacity)) {
                     // We have a Chunk that has some space left.
                     current = curr;
                     return true;
                 }
 
                 try {
                     if (remainingCapacity >= size) {
                         // At this point we know that this will be the last time curr will be used, so directly set it
                         // to null and release it once we are done.
                         return curr.readInitInto(buf, size, remainingCapacity, maxCapacity);
                     }
                 } finally {
                     // Release in a finally block so even if readInitInto(...) would throw we would still correctly
                     // release the current chunk before null it out.
                     curr.releaseFromMagazine();
                 }
             }
 
             // Now try to poll from the central queue first
             curr = group.pollChunk(size);
             if (curr == null) {
                 curr = chunkController.newChunkAllocation(size, this);
             } else {
                 curr.attachToMagazine(this);
 
                 int remainingCapacity = curr.remainingCapacity();
                 if (remainingCapacity == 0 || remainingCapacity < size) {
                     // Check if we either retain the chunk in the nextInLine cache or releasing it.
                     if (remainingCapacity < RETIRE_CAPACITY) {
                         curr.releaseFromMagazine();
                     } else {
                         // See if it makes sense to transfer the Chunk to the nextInLine cache for later usage.
                         // This method will release curr if this is not the case
                         transferToNextInLineOrRelease(curr);
                     }
                     curr = chunkController.newChunkAllocation(size, this);
                 }
             }
 
             current = curr;
             boolean success;
             try {
                 int remainingCapacity = curr.remainingCapacity();
                 assert remainingCapacity >= size;
                 if (remainingCapacity > startingCapacity) {
                     success = curr.readInitInto(buf, size, startingCapacity, maxCapacity);
                     curr = null;
                 } else {
                     success = curr.readInitInto(buf, size, remainingCapacity, maxCapacity);
                 }
             } finally {
                 if (curr != null) {
                     // Release in a finally block so even if readInitInto(...) would throw we would still correctly
                     // release the current chunk before null it out.
                     curr.releaseFromMagazine();
                     current = null;
                 }
             }
             return success;
         }
 
         private void restoreMagazineFreed() {
             Chunk next = NEXT_IN_LINE.getAndSet(this, MAGAZINE_FREED);
             if (next != null && next != MAGAZINE_FREED) {
                 // A chunk snuck in through a race. Release it after restoring MAGAZINE_FREED state.
                 next.releaseFromMagazine();
             }
         }
 
         private void transferToNextInLineOrRelease(Chunk chunk) {
             if (NEXT_IN_LINE.compareAndSet(this, null, chunk)) {
                 return;
             }
 
             Chunk nextChunk = NEXT_IN_LINE.get(this);
             if (nextChunk != null && nextChunk != MAGAZINE_FREED
                     && chunk.remainingCapacity() > nextChunk.remainingCapacity()) {
                 if (NEXT_IN_LINE.compareAndSet(this, nextChunk, chunk)) {
                     nextChunk.releaseFromMagazine();
                     return;
                 }
             }
             // Next-in-line is occupied. We don't try to add it to the central queue yet as it might still be used
             // by some buffers and so is attached to a Magazine.
             // Once a Chunk is completely released by Chunk.release() it will try to move itself to the queue
             // as last resort.
             chunk.releaseFromMagazine();
         }
 
         void free() {
             // Release the current Chunk and the next that was stored for later usage.
             restoreMagazineFreed();
             long stamp = allocationLock != null ? allocationLock.writeLock() : 0;
             try {
                 if (current != null) {
                     current.releaseFromMagazine();
                     current = null;
                 }
             } finally {
                 if (allocationLock != null) {
                     allocationLock.unlockWrite(stamp);
                 }
             }
         }
 
         public AdaptiveByteBuf newBuffer() {
             AdaptiveRecycler recycler = this.recycler;
             AdaptiveByteBuf buf = recycler == null ? EVENT_LOOP_LOCAL_BUFFER_POOL.get() : recycler.get();
             buf.resetRefCnt();
             buf.discardMarks();
             return buf;
         }
 
         boolean offerToQueue(Chunk chunk) {
             return group.offerChunk(chunk);
         }
     }
 
     private static final class ChunkRegistry {
         private final LongAdder totalCapacity = new LongAdder();
 
         public long totalCapacity() {
             return totalCapacity.sum();
         }
 
         public void add(Chunk chunk) {
             totalCapacity.add(chunk.capacity());
         }
 
         public void remove(Chunk chunk) {
             totalCapacity.add(-chunk.capacity());
         }
     }
 
     static class Chunk implements ChunkInfo {
         protected final AbstractByteBuf delegate;
         protected Magazine magazine;
         private final AdaptivePoolingAllocator allocator;
         // Always populate the refCnt field, so HotSpot doesn't emit `null` checks.
         // This is safe to do even on native-image.
         private final RefCnt refCnt = new RefCnt();
         private final int capacity;
         private final boolean pooled;
         protected int allocatedBytes;
 
         Chunk() {
             // Constructor only used by the MAGAZINE_FREED sentinel.
             delegate = null;
             magazine = null;
             allocator = null;
             capacity = 0;
             pooled = false;
         }
 
         Chunk(AbstractByteBuf delegate, Magazine magazine, boolean pooled) {
             this.delegate = delegate;
             this.pooled = pooled;
             capacity = delegate.capacity();
             attachToMagazine(magazine);
 
             // We need the top-level allocator so ByteBuf.capacity(int) can call reallocate()
             allocator = magazine.group.allocator;
 
             if (PlatformDependent.isJfrEnabled() && AllocateChunkEvent.isEventEnabled()) {
                 AllocateChunkEvent event = new AllocateChunkEvent();
                 if (event.shouldCommit()) {
                     event.fill(this, AdaptiveByteBufAllocator.class);
                     event.pooled = pooled;
                     event.threadLocal = magazine.allocationLock == null;
                     event.commit();
                 }
             }
         }
 
         Magazine currentMagazine() {
             return magazine;
         }
 
         void detachFromMagazine() {
             if (magazine != null) {
                 magazine = null;
             }
         }
 
         void attachToMagazine(Magazine magazine) {
             assert this.magazine == null;
             this.magazine = magazine;
         }
 
         /**
          * Called when a magazine is done using this chunk, probably because it was emptied.
          */
         void releaseFromMagazine() {
             // Chunks can be reused before they become empty.
             // We can therefor put them in the shared queue as soon as the magazine is done with this chunk.
             Magazine mag = magazine;
             detachFromMagazine();
             if (!mag.offerToQueue(this)) {
                 markToDeallocate();
             }
         }
 
         /**
          * Called when a ByteBuf is done using its allocation in this chunk.
          */
         void releaseSegment(int ignoredSegmentId, int size) {
             release();
         }
 
         void markToDeallocate() {
             release();
         }
 
         private void retain() {
             RefCnt.retain(refCnt);
         }
 
         protected boolean release() {
             boolean deallocate = RefCnt.release(refCnt);
             if (deallocate) {
                 deallocate();
             }
             return deallocate;
         }
 
         protected void deallocate() {
             onRelease();
             allocator.chunkRegistry.remove(this);
             delegate.release();
         }
 
         private void onRelease() {
             if (PlatformDependent.isJfrEnabled() && FreeChunkEvent.isEventEnabled()) {
                 FreeChunkEvent event = new FreeChunkEvent();
                 if (event.shouldCommit()) {
                     event.fill(this, AdaptiveByteBufAllocator.class);
                     event.pooled = pooled;
                     event.commit();
                 }
             }
         }
 
         public boolean readInitInto(AdaptiveByteBuf buf, int size, int startingCapacity, int maxCapacity) {
             int startIndex = allocatedBytes;
             allocatedBytes = startIndex + startingCapacity;
             Chunk chunk = this;
             chunk.retain();
             try {
                 buf.init(delegate, chunk, 0, 0, startIndex, size, startingCapacity, maxCapacity);
                 chunk = null;
             } finally {
                 if (chunk != null) {
                     // If chunk is not null we know that buf.init(...) failed and so we need to manually release
                     // the chunk again as we retained it before calling buf.init(...). Beside this we also need to
                     // restore the old allocatedBytes value.
                     allocatedBytes = startIndex;
                     chunk.release();
                 }
             }
             return true;
         }
 
         public int remainingCapacity() {
             return capacity - allocatedBytes;
         }
 
         public boolean hasUnprocessedFreelistEntries() {
             return false;
         }
 
         public void processFreelistEntries() {
         }
 
         @Override
         public int capacity() {
             return capacity;
         }
 
         @Override
         public boolean isDirect() {
             return delegate.isDirect();
         }
 
         @Override
         public long memoryAddress() {
             return delegate._memoryAddress();
         }
     }
 
     private static final class IntStack {
 
         private final int[] stack;
         private int top;
 
         IntStack(int[] initialValues) {
             stack = initialValues;
             top = initialValues.length - 1;
         }
 
         public boolean isEmpty() {
             return top == -1;
         }
 
         public int pop() {
             final int last = stack[top];
             top--;
             return last;
         }
 
         public void push(int value) {
             stack[top + 1] = value;
             top++;
         }
 
         public int size() {
             return top + 1;
         }
     }
 
     /**
      * Removes per-allocation retain()/release() atomic ops from the hot path by replacing ref counting
      * with a segment-count state machine. Atomics are only needed on the cold deallocation path
      * ({@link #markToDeallocate()}), which is rare for long-lived chunks that cycle segments many times.
      * The tradeoff is a {@link MpscIntQueue#size()} call (volatile reads, no RMW) per remaining segment
      * return after mark — acceptable since it avoids atomic RMWs entirely.
      * <p>
      * State transitions:
      * <ul>
      *   <li>{@link #AVAILABLE} (-1): chunk is in use, no deallocation tracking needed</li>
      *   <li>0..N: local free list size at the time {@link #markToDeallocate()} was called;
      *       used to track when all segments have been returned</li>
      *   <li>{@link #DEALLOCATED} (Integer.MIN_VALUE): all segments returned, chunk deallocated</li>
      * </ul>
      * <p>
      * Ordering: external {@link #releaseSegment} pushes to the MPSC queue (which has an implicit
      * StoreLoad barrier via its {@code offer()}), then reads {@code state} — this guarantees
      * visibility of any preceding {@link #markToDeallocate()} write.
      */
     static class SizeClassedChunk extends Chunk {
         private static final int FREE_LIST_EMPTY = -1;
         private static final int AVAILABLE = -1;
         // Integer.MIN_VALUE so that `DEALLOCATED + externalFreeList.size()` can never equal `segments`,
         // making late-arriving releaseSegment calls on external threads arithmetically harmless.
         private static final int DEALLOCATED = Integer.MIN_VALUE;
         private static final AtomicIntegerFieldUpdater<SizeClassedChunk> STATE =
                 AtomicIntegerFieldUpdater.newUpdater(SizeClassedChunk.class, "state");
         private volatile int state;
         private final int segments;
         private final int segmentSize;
         private final MpscIntQueue externalFreeList;
         private final IntStack localFreeList;
         private Thread ownerThread;
         int purgeEpoch;
         long lastPurgeGeneration;
         long lastScanGeneration;
 
         SizeClassedChunk(AbstractByteBuf delegate, Magazine magazine,
                          SizeClassChunkController controller) {
             super(delegate, magazine, true);
             segmentSize = controller.segmentSize;
             segments = controller.chunkSize / segmentSize;
             STATE.lazySet(this, AVAILABLE);
             ownerThread = magazine.group.ownerThread;
             if (ownerThread == null) {
                 externalFreeList = controller.createFreeList();
                 localFreeList = null;
             } else {
                 externalFreeList = controller.createEmptyFreeList();
                 localFreeList = controller.createLocalFreeList();
             }
         }
 
         @Override
         public boolean readInitInto(AdaptiveByteBuf buf, int size, int startingCapacity, int maxCapacity) {
             assert state == AVAILABLE;
             final int startIndex = nextAvailableSegmentOffset();
             if (startIndex == FREE_LIST_EMPTY) {
                 return false;
             }
             allocatedBytes += segmentSize;
             try {
                 buf.init(delegate, this, 0, 0, startIndex, size, startingCapacity, maxCapacity);
             } catch (Throwable t) {
                 allocatedBytes -= segmentSize;
                 releaseSegmentOffsetIntoFreeList(startIndex);
                 throw t;
             }
             return true;
         }
 
         private int nextAvailableSegmentOffset() {
             final int startIndex;
             IntStack localFreeList = this.localFreeList;
             if (localFreeList != null) {
                 assert Thread.currentThread() == ownerThread;
                 if (localFreeList.isEmpty()) {
                     startIndex = externalFreeList.poll();
                 } else {
                     startIndex = localFreeList.pop();
                 }
             } else {
                 startIndex = externalFreeList.poll();
             }
             return startIndex;
         }
 
         // this can be used by the ConcurrentQueueChunkCache to find the first buffer to use:
         // it doesn't update the remaining capacity and it's not consider a single segmentSize
         // case as not suitable to be reused
         public boolean hasRemainingCapacity() {
             int remaining = super.remainingCapacity();
             if (remaining > 0) {
                 return true;
             }
             if (localFreeList != null) {
                 return !localFreeList.isEmpty();
             }
             return !externalFreeList.isEmpty();
         }
 
         boolean hasFullCapacity() {
             int free = externalFreeList.size();
             IntStack local = localFreeList;
             if (local != null) {
                 free += local.size();
             }
             return free == segments;
         }
 
         @Override
         public int remainingCapacity() {
             int remaining = super.remainingCapacity();
             return remaining > segmentSize ? remaining : updateRemainingCapacity(remaining);
         }
 
         private int updateRemainingCapacity(int snapshotted) {
             int freeSegments = externalFreeList.size();
             IntStack localFreeList = this.localFreeList;
             if (localFreeList != null) {
                 freeSegments += localFreeList.size();
             }
             int updated = freeSegments * segmentSize;
             if (updated != snapshotted) {
                 allocatedBytes = capacity() - updated;
             }
             return updated;
         }
 
         private void releaseSegmentOffsetIntoFreeList(int startIndex) {
             IntStack localFreeList = this.localFreeList;
             if (localFreeList != null && Thread.currentThread() == ownerThread) {
                 localFreeList.push(startIndex);
             } else {
                 boolean segmentReturned = externalFreeList.offer(startIndex);
                 assert segmentReturned : "Unable to return segment " + startIndex + " to free list";
             }
         }
 
         @Override
         void releaseSegment(int startIndex, int size) {
             IntStack localFreeList = this.localFreeList;
             if (localFreeList != null && Thread.currentThread() == ownerThread) {
                 localFreeList.push(startIndex);
                 int state = this.state;
                 if (state != AVAILABLE) {
                     updateStateOnLocalReleaseSegment(state, localFreeList);
                 }
             } else {
                 boolean segmentReturned = externalFreeList.offer(startIndex);
                 assert segmentReturned;
                 // implicit StoreLoad barrier from MPSC offer()
                 int state = this.state;
                 if (state != AVAILABLE) {
                     deallocateIfNeeded(state);
                 }
             }
         }
 
         private void updateStateOnLocalReleaseSegment(int previousLocalSize, IntStack localFreeList) {
             int newLocalSize = localFreeList.size();
             boolean alwaysTrue = STATE.compareAndSet(this, previousLocalSize, newLocalSize);
             assert alwaysTrue : "this shouldn't happen unless double release in the local free list";
             deallocateIfNeeded(newLocalSize);
         }
 
         private void deallocateIfNeeded(int localSize) {
             // Check if all segments have been returned.
             int totalFreeSegments = localSize + externalFreeList.size();
             if (totalFreeSegments == segments && STATE.compareAndSet(this, localSize, DEALLOCATED)) {
                 deallocate();
             }
         }
 
         @Override
         void markToDeallocate() {
             IntStack localFreeList = this.localFreeList;
             int localSize = localFreeList != null ? localFreeList.size() : 0;
             STATE.set(this, localSize);
             deallocateIfNeeded(localSize);
         }
     }
 
     private static final class BuddyChunk extends Chunk implements IntConsumer {
         private static final int MIN_BUDDY_SIZE = 32768;
         private static final byte IS_CLAIMED = (byte) (1 << 7);
         private static final byte HAS_CLAIMED_CHILDREN = 1 << 6;
         private static final byte SHIFT_MASK = ~(IS_CLAIMED | HAS_CLAIMED_CHILDREN);
         private static final int PACK_OFFSET_MASK = 0xFFFF;
         private static final int PACK_SIZE_SHIFT = Integer.SIZE - Integer.numberOfLeadingZeros(PACK_OFFSET_MASK);
 
         private final MpscIntQueue freeList;
         // The bits of each buddy: [1: is claimed][1: has claimed children][30: MIN_BUDDY_SIZE shift to get size]
         private final byte[] buddies;
         private final int freeListCapacity;
 
         BuddyChunk(AbstractByteBuf delegate, Magazine magazine) {
             super(delegate, magazine, true);
             freeListCapacity = delegate.capacity() / MIN_BUDDY_SIZE;
             int maxShift = Integer.numberOfTrailingZeros(freeListCapacity);
             assert maxShift <= 30; // The top 2 bits are used for marking.
             freeList = MpscIntQueue.create(freeListCapacity, -1); // At most half of tree (all leaf nodes) can be freed.
             buddies = new byte[freeListCapacity << 1];
 
             // Generate the buddies entries.
             int index = 1;
             int runLength = 1;
             int currentRun = 0;
             while (maxShift > 0) {
                 buddies[index++] = (byte) maxShift;
                 if (++currentRun == runLength) {
                     currentRun = 0;
                     runLength <<= 1;
                     maxShift--;
                 }
             }
         }
 
         @Override
         public boolean readInitInto(AdaptiveByteBuf buf, int size, int startingCapacity, int maxCapacity) {
             if (!freeList.isEmpty()) {
                 freeList.drain(freeListCapacity, this);
             }
             int startIndex = chooseFirstFreeBuddy(1, startingCapacity, 0);
             if (startIndex == -1) {
                 return false;
             }
             Chunk chunk = this;
             chunk.retain();
             try {
                 buf.init(delegate, this, 0, 0, startIndex, size, startingCapacity, maxCapacity);
                 allocatedBytes += startingCapacity;
                 chunk = null;
             } finally {
                 if (chunk != null) {
                     unreserveMatchingBuddy(1, startingCapacity, startIndex, 0);
                     // If chunk is not null we know that buf.init(...) failed and so we need to manually release
                     // the chunk again as we retained it before calling buf.init(...).
                     chunk.release();
                 }
             }
             return true;
         }
 
         @Override
         public void accept(int packed) {
             // Called by allocating thread when draining freeList.
             int size = unpackSize(packed);
             int offset = unpackOffset(packed);
             unreserveMatchingBuddy(1, size, offset, 0);
             allocatedBytes -= size;
         }
 
         private static int unpackSize(int packed) {
             return MIN_BUDDY_SIZE << (packed >> PACK_SIZE_SHIFT);
         }
 
         private static int unpackOffset(int packed) {
             return (packed & PACK_OFFSET_MASK) * MIN_BUDDY_SIZE;
         }
 
         @Override
         void releaseSegment(int startingIndex, int size) {
             int packedOffset = startingIndex / MIN_BUDDY_SIZE;
             int packedSize = Integer.numberOfTrailingZeros(size / MIN_BUDDY_SIZE) << PACK_SIZE_SHIFT;
             int packed = packedOffset | packedSize;
             freeList.offer(packed);
             release();
         }
 
         @Override
         public int remainingCapacity() {
             int capacityInFreeList = 0;
             if (!freeList.isEmpty()) {
                 capacityInFreeList = freeList.weakPeekReduce(freeListCapacity, 0,
                         (sum, entry) -> sum + unpackSize(entry));
             }
             return super.remainingCapacity() + capacityInFreeList;
         }
 
         @Override
         public boolean hasUnprocessedFreelistEntries() {
             return !freeList.isEmpty();
         }
 
         @Override
         public void processFreelistEntries() {
             freeList.drain(freeListCapacity, this);
         }
 
         /**
          * Claim a suitable buddy and return its start offset into the delegate chunk, or return -1 if nothing claimed.
          */
         private int chooseFirstFreeBuddy(int index, int size, int currOffset) {
             byte[] buddies = this.buddies;
             while (index < buddies.length) {
                 byte buddy = buddies[index];
                 int currValue = MIN_BUDDY_SIZE << (buddy & SHIFT_MASK);
                 if (currValue < size || (buddy & IS_CLAIMED) == IS_CLAIMED) {
                     return -1;
                 }
                 if (currValue == size && (buddy & HAS_CLAIMED_CHILDREN) == 0) {
                     buddies[index] |= IS_CLAIMED;
                     return currOffset;
                 }
                 int found = chooseFirstFreeBuddy(index << 1, size, currOffset);
                 if (found != -1) {
                     buddies[index] |= HAS_CLAIMED_CHILDREN;
                     return found;
                 }
                 index = (index << 1) + 1;
                 currOffset += currValue >> 1; // Bump offset to skip first half of this layer.
             }
             return -1;
         }
 
         /**
          * Un-reserve the matching buddy and return whether there are any other child or sibling reservations.
          */
         private boolean unreserveMatchingBuddy(int index, int size, int offset, int currOffset) {
             byte[] buddies = this.buddies;
             if (buddies.length <= index) {
                 return false;
             }
             byte buddy = buddies[index];
             int currSize = MIN_BUDDY_SIZE << (buddy & SHIFT_MASK);
 
             if (currSize == size) {
                 // We're at the right size level.
                 if (currOffset == offset) {
                     buddies[index] &= SHIFT_MASK;
                     return false;
                 }
                 throw new IllegalStateException("The intended segment was not found at index " +
                         index + ", for size " + size + " and offset " + offset);
             }
 
             // We're at a parent size level. Use the target offset to guide our drill-down path.
             boolean claims;
             int siblingIndex;
             if (offset < currOffset + (currSize >> 1)) {
                 // Must be down the left path.
                 claims = unreserveMatchingBuddy(index << 1, size, offset, currOffset);
                 siblingIndex = (index << 1) + 1;
             } else {
                 // Must be down the rigth path.
                 claims = unreserveMatchingBuddy((index << 1) + 1, size, offset, currOffset + (currSize >> 1));
                 siblingIndex = index << 1;
             }
             if (!claims) {
                 // No other claims down the path we took. Check if the sibling has claims.
                 byte sibling = buddies[siblingIndex];
                 if ((sibling & SHIFT_MASK) == sibling) {
                     // No claims in the sibling. We can clear this level as well.
                     buddies[index] &= SHIFT_MASK;
                     return false;
                 }
             }
             return true;
         }
 
         @Override
         public String toString() {
             int capacity = delegate.capacity();
             int remaining = capacity - allocatedBytes;
             return "BuddyChunk[capacity: " + capacity +
                     ", remaining: " + remaining +
                     ", free list: " + freeList.size() + ']';
         }
     }
 
     static final class AdaptiveByteBuf extends AbstractReferenceCountedByteBuf {
 
         private final EnhancedHandle<AdaptiveByteBuf> handle;
 
         // this both act as adjustment and the start index for a free list segment allocation
         private int startIndex;
         private AbstractByteBuf rootParent;
         Chunk chunk;
         private int length;
         private int maxFastCapacity;
         private ByteBuffer tmpNioBuf;
         private boolean hasArray;
         private boolean hasMemoryAddress;
 
         AdaptiveByteBuf(EnhancedHandle<AdaptiveByteBuf> recyclerHandle) {
             super(0);
             handle = ObjectUtil.checkNotNull(recyclerHandle, "recyclerHandle");
         }
 
         void init(AbstractByteBuf unwrapped, Chunk wrapped, int readerIndex, int writerIndex,
                   int startIndex, int size, int capacity, int maxCapacity) {
             this.startIndex = startIndex;
             chunk = wrapped;
             length = size;
             maxFastCapacity = capacity;
             maxCapacity(maxCapacity);
             setIndex0(readerIndex, writerIndex);
             hasArray = unwrapped.hasArray();
             hasMemoryAddress = unwrapped.hasMemoryAddress();
             rootParent = unwrapped;
             tmpNioBuf = null;
 
             if (PlatformDependent.isJfrEnabled() && AllocateBufferEvent.isEventEnabled()) {
                 AllocateBufferEvent event = new AllocateBufferEvent();
                 if (event.shouldCommit()) {
                     event.fill(this, AdaptiveByteBufAllocator.class);
                     event.chunkPooled = wrapped.pooled;
                     Magazine m = wrapped.magazine;
                     event.chunkThreadLocal = m != null && m.allocationLock == null;
                     event.commit();
                 }
             }
         }
 
         private AbstractByteBuf rootParent() {
             final AbstractByteBuf rootParent = this.rootParent;
             if (rootParent != null) {
                 return rootParent;
             }
             throw new IllegalReferenceCountException();
         }
 
         @Override
         public int capacity() {
             return length;
         }
 
         @Override
         public int maxFastWritableBytes() {
             return Math.min(maxFastCapacity, maxCapacity()) - writerIndex;
         }
 
         @Override
         public ByteBuf capacity(int newCapacity) {
             checkNewCapacity(newCapacity);
             if (length <= newCapacity && newCapacity <= maxFastCapacity) {
                 length = newCapacity;
                 return this;
             }
             if (newCapacity < capacity()) {
                 length = newCapacity;
                 trimIndicesToCapacity(newCapacity);
                 return this;
             }
 
             if (PlatformDependent.isJfrEnabled() && ReallocateBufferEvent.isEventEnabled()) {
                 ReallocateBufferEvent event = new ReallocateBufferEvent();
                 if (event.shouldCommit()) {
                     event.fill(this, AdaptiveByteBufAllocator.class);
                     event.newCapacity = newCapacity;
                     event.commit();
                 }
             }
 
             // Reallocation required.
             Chunk chunk = this.chunk;
             AdaptivePoolingAllocator allocator = chunk.allocator;
             int readerIndex = this.readerIndex;
             int writerIndex = this.writerIndex;
             int baseOldRootIndex = startIndex;
             int oldLength = length;
             int oldCapacity = maxFastCapacity;
             AbstractByteBuf oldRoot = rootParent();
             allocator.reallocate(newCapacity, maxCapacity(), this);
             oldRoot.getBytes(baseOldRootIndex, this, 0, oldLength);
             chunk.releaseSegment(baseOldRootIndex, oldCapacity);
             assert oldCapacity < maxFastCapacity && newCapacity <= maxFastCapacity :
                     "Capacity increase failed";
             this.readerIndex = readerIndex;
             this.writerIndex = writerIndex;
             return this;
         }
 
         @Override
         public ByteBufAllocator alloc() {
             return rootParent().alloc();
         }
 
         @SuppressWarnings("deprecation")
         @Override
         public ByteOrder order() {
             return rootParent().order();
         }
 
         @Override
         public ByteBuf unwrap() {
             return null;
         }
 
         @Override
         public boolean isDirect() {
             return rootParent().isDirect();
         }
 
         @Override
         public int arrayOffset() {
             return idx(rootParent().arrayOffset());
         }
 
         @Override
         public boolean hasMemoryAddress() {
             return hasMemoryAddress;
         }
 
         @Override
         public long memoryAddress() {
             ensureAccessible();
             return _memoryAddress();
         }
 
         @Override
         long _memoryAddress() {
             AbstractByteBuf root = rootParent;
             return root != null ? root._memoryAddress() + startIndex : 0L;
         }
 
         @Override
         boolean _isDirect() {
             AbstractByteBuf root = rootParent;
             return root != null && root.isDirect();
         }
 
         @Override
         public ByteBuffer nioBuffer(int index, int length) {
             checkIndex(index, length);
             return rootParent().nioBuffer(idx(index), length);
         }
 
         @Override
         public ByteBuffer internalNioBuffer(int index, int length) {
             checkIndex(index, length);
             return (ByteBuffer) internalNioBuffer().position(index).limit(index + length);
         }
 
         private ByteBuffer internalNioBuffer() {
             if (tmpNioBuf == null) {
                 tmpNioBuf = rootParent().nioBuffer(startIndex, maxFastCapacity);
             }
             return (ByteBuffer) tmpNioBuf.clear();
         }
 
         @Override
         public ByteBuffer[] nioBuffers(int index, int length) {
             checkIndex(index, length);
             return rootParent().nioBuffers(idx(index), length);
         }
 
         @Override
         public boolean hasArray() {
             return hasArray;
         }
 
         @Override
         public byte[] array() {
             ensureAccessible();
             return rootParent().array();
         }
 
         @Override
         public ByteBuf copy(int index, int length) {
             checkIndex(index, length);
             return rootParent().copy(idx(index), length);
         }
 
         @Override
         public int nioBufferCount() {
             return rootParent().nioBufferCount();
         }
 
         @Override
         protected byte _getByte(int index) {
             return rootParent()._getByte(idx(index));
         }
 
         @Override
         protected short _getShort(int index) {
             return rootParent()._getShort(idx(index));
         }
 
         @Override
         protected short _getShortLE(int index) {
             return rootParent()._getShortLE(idx(index));
         }
 
         @Override
         protected int _getUnsignedMedium(int index) {
             return rootParent()._getUnsignedMedium(idx(index));
         }
 
         @Override
         protected int _getUnsignedMediumLE(int index) {
             return rootParent()._getUnsignedMediumLE(idx(index));
         }
 
         @Override
         protected int _getInt(int index) {
             return rootParent()._getInt(idx(index));
         }
 
         @Override
         protected int _getIntLE(int index) {
             return rootParent()._getIntLE(idx(index));
         }
 
         @Override
         protected long _getLong(int index) {
             return rootParent()._getLong(idx(index));
         }
 
         @Override
         protected long _getLongLE(int index) {
             return rootParent()._getLongLE(idx(index));
         }
 
         @Override
         public ByteBuf getBytes(int index, ByteBuf dst, int dstIndex, int length) {
             checkIndex(index, length);
             rootParent().getBytes(idx(index), dst, dstIndex, length);
             return this;
         }
 
         @Override
         public ByteBuf getBytes(int index, byte[] dst, int dstIndex, int length) {
             checkIndex(index, length);
             rootParent().getBytes(idx(index), dst, dstIndex, length);
             return this;
         }
 
         @Override
         public ByteBuf getBytes(int index, ByteBuffer dst) {
             checkIndex(index, dst.remaining());
             rootParent().getBytes(idx(index), dst);
             return this;
         }
 
         @Override
         protected void _setByte(int index, int value) {
             rootParent()._setByte(idx(index), value);
         }
 
         @Override
         protected void _setShort(int index, int value) {
             rootParent()._setShort(idx(index), value);
         }
 
         @Override
         protected void _setShortLE(int index, int value) {
             rootParent()._setShortLE(idx(index), value);
         }
 
         @Override
         protected void _setMedium(int index, int value) {
             rootParent()._setMedium(idx(index), value);
         }
 
         @Override
         protected void _setMediumLE(int index, int value) {
             rootParent()._setMediumLE(idx(index), value);
         }
 
         @Override
         protected void _setInt(int index, int value) {
             rootParent()._setInt(idx(index), value);
         }
 
         @Override
         protected void _setIntLE(int index, int value) {
             rootParent()._setIntLE(idx(index), value);
         }
 
         @Override
         protected void _setLong(int index, long value) {
             rootParent()._setLong(idx(index), value);
         }
 
         @Override
         protected void _setLongLE(int index, long value) {
             rootParent()._setLongLE(idx(index), value);
         }
 
         @Override
         public ByteBuf setBytes(int index, byte[] src, int srcIndex, int length) {
             checkIndex(index, length);
             if (tmpNioBuf == null && PlatformDependent.javaVersion() >= 13) {
                 ByteBuffer dstBuffer = rootParent()._internalNioBuffer();
                 PlatformDependent.absolutePut(dstBuffer, idx(index), src, srcIndex, length);
             } else {
                 ByteBuffer tmp = (ByteBuffer) internalNioBuffer().clear().position(index);
                 tmp.put(src, srcIndex, length);
             }
             return this;
         }
 
         @Override
         public ByteBuf setBytes(int index, ByteBuf src, int srcIndex, int length) {
             checkIndex(index, length);
             if (src instanceof AdaptiveByteBuf && PlatformDependent.javaVersion() >= 16) {
                 AdaptiveByteBuf srcBuf = (AdaptiveByteBuf) src;
                 srcBuf.checkIndex(srcIndex, length);
                 ByteBuffer dstBuffer = rootParent()._internalNioBuffer();
                 ByteBuffer srcBuffer = srcBuf.rootParent()._internalNioBuffer();
                 PlatformDependent.absolutePut(dstBuffer, idx(index), srcBuffer, srcBuf.idx(srcIndex), length);
             } else {
                 ByteBuffer tmp = internalNioBuffer();
                 tmp.position(index);
                 tmp.put(src.nioBuffer(srcIndex, length));
             }
             return this;
         }
 
         @Override
         public ByteBuf setBytes(int index, ByteBuffer src) {
             int length = src.remaining();
             checkIndex(index, length);
             ByteBuffer tmp = internalNioBuffer();
             if (PlatformDependent.javaVersion() >= 16) {
                 int offset = src.position();
                 PlatformDependent.absolutePut(tmp, index, src, offset, length);
                 src.position(offset + length);
             } else {
                 tmp.position(index);
                 tmp.put(src);
             }
             return this;
         }
 
         @Override
         public ByteBuf getBytes(int index, OutputStream out, int length)
                 throws IOException {
             checkIndex(index, length);
             if (length != 0) {
                 ByteBuffer tmp = internalNioBuffer();
                 ByteBufUtil.readBytes(alloc(), tmp.hasArray() ? tmp : tmp.duplicate(), index, length, out);
             }
             return this;
         }
 
         @Override
         public int getBytes(int index, GatheringByteChannel out, int length)
                 throws IOException {
             ByteBuffer buf = internalNioBuffer().duplicate();
             buf.clear().position(index).limit(index + length);
             return out.write(buf);
         }
 
         @Override
         public int getBytes(int index, FileChannel out, long position, int length)
                 throws IOException {
             ByteBuffer buf = internalNioBuffer().duplicate();
             buf.clear().position(index).limit(index + length);
             return out.write(buf, position);
         }
 
         @Override
         public int setBytes(int index, InputStream in, int length)
                 throws IOException {
             checkIndex(index, length);
             final AbstractByteBuf rootParent = rootParent();
             if (rootParent.hasArray()) {
                 return rootParent.setBytes(idx(index), in, length);
             }
             byte[] tmp = ByteBufUtil.threadLocalTempArray(length);
             int readBytes = in.read(tmp, 0, length);
             if (readBytes <= 0) {
                 return readBytes;
             }
             setBytes(index, tmp, 0, readBytes);
             return readBytes;
         }
 
         @Override
         public int setBytes(int index, ScatteringByteChannel in, int length)
                 throws IOException {
             try {
                 return in.read(internalNioBuffer(index, length));
             } catch (ClosedChannelException ignored) {
                 return -1;
             }
         }
 
         @Override
         public int setBytes(int index, FileChannel in, long position, int length)
                 throws IOException {
             try {
                 return in.read(internalNioBuffer(index, length), position);
             } catch (ClosedChannelException ignored) {
                 return -1;
             }
         }
 
         @Override
         public int setCharSequence(int index, CharSequence sequence, Charset charset) {
             return setCharSequence0(index, sequence, charset, false);
         }
 
         private int setCharSequence0(int index, CharSequence sequence, Charset charset, boolean expand) {
             if (charset.equals(CharsetUtil.UTF_8)) {
                 int length = ByteBufUtil.utf8MaxBytes(sequence);
                 if (expand) {
                     ensureWritable0(length);
                     checkIndex0(index, length);
                 } else {
                     checkIndex(index, length);
                 }
                 return ByteBufUtil.writeUtf8(this, index, length, sequence, sequence.length());
             }
             if (charset.equals(CharsetUtil.US_ASCII) || charset.equals(CharsetUtil.ISO_8859_1)) {
                 int length = sequence.length();
                 if (expand) {
                     ensureWritable0(length);
                     checkIndex0(index, length);
                 } else {
                     checkIndex(index, length);
                 }
                 return ByteBufUtil.writeAscii(this, index, sequence, length);
             }
             byte[] bytes = sequence.toString().getBytes(charset);
             if (expand) {
                 ensureWritable0(bytes.length);
                 // setBytes(...) will take care of checking the indices.
             }
             setBytes(index, bytes);
             return bytes.length;
         }
 
         @Override
         public int writeCharSequence(CharSequence sequence, Charset charset) {
             int written = setCharSequence0(writerIndex, sequence, charset, true);
             writerIndex += written;
             return written;
         }
 
         @Override
         public int forEachByte(int index, int length, ByteProcessor processor) {
             checkIndex(index, length);
             int ret = rootParent().forEachByte(idx(index), length, processor);
             return forEachResult(ret);
         }
 
         @Override
         public int forEachByteDesc(int index, int length, ByteProcessor processor) {
             checkIndex(index, length);
             int ret = rootParent().forEachByteDesc(idx(index), length, processor);
             return forEachResult(ret);
         }
 
         @Override
         public ByteBuf setZero(int index, int length) {
             checkIndex(index, length);
             rootParent().setZero(idx(index), length);
             return this;
         }
 
         @Override
         public ByteBuf writeZero(int length) {
             ensureWritable(length);
             rootParent().setZero(idx(writerIndex), length);
             writerIndex += length;
             return this;
         }
 
         private int forEachResult(int ret) {
             if (ret < startIndex) {
                 return -1;
             }
             return ret - startIndex;
         }
 
         @Override
         public boolean isContiguous() {
             return rootParent().isContiguous();
         }
 
         private int idx(int index) {
             return index + startIndex;
         }
 
         @Override
         protected void deallocate() {
             if (PlatformDependent.isJfrEnabled() && FreeBufferEvent.isEventEnabled()) {
                 FreeBufferEvent event = new FreeBufferEvent();
                 if (event.shouldCommit()) {
                     event.fill(this, AdaptiveByteBufAllocator.class);
                     event.commit();
                 }
             }
 
             if (chunk != null) {
                 chunk.releaseSegment(startIndex, maxFastCapacity);
             }
             tmpNioBuf = null;
             chunk = null;
             rootParent = null;
             handle.unguardedRecycle(this);
         }
     }
 
     /**
      * The strategy for how {@link AdaptivePoolingAllocator} should allocate chunk buffers.
      */
     interface ChunkAllocator {
         /**
          * Allocate a buffer for a chunk. This can be any kind of {@link AbstractByteBuf} implementation.
          *
          * @param initialCapacity The initial capacity of the returned {@link AbstractByteBuf}.
          * @param maxCapacity     The maximum capacity of the returned {@link AbstractByteBuf}.
          * @return The buffer that represents the chunk memory.
          */
         AbstractByteBuf allocate(int initialCapacity, int maxCapacity);
     }
 }