diff libsanitizer/sanitizer_common/sanitizer_allocator_primary64.h @ 111:04ced10e8804

gcc 7
author kono
date Fri, 27 Oct 2017 22:46:09 +0900
parents
children 1830386684a0
line wrap: on
line diff
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/libsanitizer/sanitizer_common/sanitizer_allocator_primary64.h	Fri Oct 27 22:46:09 2017 +0900
@@ -0,0 +1,821 @@
+//===-- sanitizer_allocator_primary64.h -------------------------*- C++ -*-===//
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// Part of the Sanitizer Allocator.
+//
+//===----------------------------------------------------------------------===//
+#ifndef SANITIZER_ALLOCATOR_H
+#error This file must be included inside sanitizer_allocator.h
+#endif
+
+template<class SizeClassAllocator> struct SizeClassAllocator64LocalCache;
+
+// SizeClassAllocator64 -- allocator for 64-bit address space.
+// The template parameter Params is a class containing the actual parameters.
+//
+// Space: a portion of address space of kSpaceSize bytes starting at SpaceBeg.
+// If kSpaceBeg is ~0 then SpaceBeg is chosen dynamically my mmap.
+// Otherwise SpaceBeg=kSpaceBeg (fixed address).
+// kSpaceSize is a power of two.
+// At the beginning the entire space is mprotect-ed, then small parts of it
+// are mapped on demand.
+//
+// Region: a part of Space dedicated to a single size class.
+// There are kNumClasses Regions of equal size.
+//
+// UserChunk: a piece of memory returned to user.
+// MetaChunk: kMetadataSize bytes of metadata associated with a UserChunk.
+
+// FreeArray is an array free-d chunks (stored as 4-byte offsets)
+//
+// A Region looks like this:
+// UserChunk1 ... UserChunkN <gap> MetaChunkN ... MetaChunk1 FreeArray
+
+struct SizeClassAllocator64FlagMasks {  //  Bit masks.
+  enum {
+    kRandomShuffleChunks = 1,
+  };
+};
+
+template <class Params>
+class SizeClassAllocator64 {
+ public:
+  static const uptr kSpaceBeg = Params::kSpaceBeg;
+  static const uptr kSpaceSize = Params::kSpaceSize;
+  static const uptr kMetadataSize = Params::kMetadataSize;
+  typedef typename Params::SizeClassMap SizeClassMap;
+  typedef typename Params::MapUnmapCallback MapUnmapCallback;
+
+  static const bool kRandomShuffleChunks =
+      Params::kFlags & SizeClassAllocator64FlagMasks::kRandomShuffleChunks;
+
+  typedef SizeClassAllocator64<Params> ThisT;
+  typedef SizeClassAllocator64LocalCache<ThisT> AllocatorCache;
+
+  // When we know the size class (the region base) we can represent a pointer
+  // as a 4-byte integer (offset from the region start shifted right by 4).
+  typedef u32 CompactPtrT;
+  static const uptr kCompactPtrScale = 4;
+  CompactPtrT PointerToCompactPtr(uptr base, uptr ptr) const {
+    return static_cast<CompactPtrT>((ptr - base) >> kCompactPtrScale);
+  }
+  uptr CompactPtrToPointer(uptr base, CompactPtrT ptr32) const {
+    return base + (static_cast<uptr>(ptr32) << kCompactPtrScale);
+  }
+
+  void Init(s32 release_to_os_interval_ms) {
+    uptr TotalSpaceSize = kSpaceSize + AdditionalSize();
+    if (kUsingConstantSpaceBeg) {
+      CHECK_EQ(kSpaceBeg, reinterpret_cast<uptr>(
+                              MmapFixedNoAccess(kSpaceBeg, TotalSpaceSize)));
+    } else {
+      NonConstSpaceBeg =
+          reinterpret_cast<uptr>(MmapNoAccess(TotalSpaceSize));
+      CHECK_NE(NonConstSpaceBeg, ~(uptr)0);
+    }
+    SetReleaseToOSIntervalMs(release_to_os_interval_ms);
+    MapWithCallbackOrDie(SpaceEnd(), AdditionalSize());
+  }
+
+  s32 ReleaseToOSIntervalMs() const {
+    return atomic_load(&release_to_os_interval_ms_, memory_order_relaxed);
+  }
+
+  void SetReleaseToOSIntervalMs(s32 release_to_os_interval_ms) {
+    atomic_store(&release_to_os_interval_ms_, release_to_os_interval_ms,
+                 memory_order_relaxed);
+  }
+
+  static bool CanAllocate(uptr size, uptr alignment) {
+    return size <= SizeClassMap::kMaxSize &&
+      alignment <= SizeClassMap::kMaxSize;
+  }
+
+  NOINLINE void ReturnToAllocator(AllocatorStats *stat, uptr class_id,
+                                  const CompactPtrT *chunks, uptr n_chunks) {
+    RegionInfo *region = GetRegionInfo(class_id);
+    uptr region_beg = GetRegionBeginBySizeClass(class_id);
+    CompactPtrT *free_array = GetFreeArray(region_beg);
+
+    BlockingMutexLock l(&region->mutex);
+    uptr old_num_chunks = region->num_freed_chunks;
+    uptr new_num_freed_chunks = old_num_chunks + n_chunks;
+    // Failure to allocate free array space while releasing memory is non
+    // recoverable.
+    if (UNLIKELY(!EnsureFreeArraySpace(region, region_beg,
+                                       new_num_freed_chunks)))
+      DieOnFailure::OnOOM();
+    for (uptr i = 0; i < n_chunks; i++)
+      free_array[old_num_chunks + i] = chunks[i];
+    region->num_freed_chunks = new_num_freed_chunks;
+    region->stats.n_freed += n_chunks;
+
+    MaybeReleaseToOS(class_id);
+  }
+
+  NOINLINE bool GetFromAllocator(AllocatorStats *stat, uptr class_id,
+                                 CompactPtrT *chunks, uptr n_chunks) {
+    RegionInfo *region = GetRegionInfo(class_id);
+    uptr region_beg = GetRegionBeginBySizeClass(class_id);
+    CompactPtrT *free_array = GetFreeArray(region_beg);
+
+    BlockingMutexLock l(&region->mutex);
+    if (UNLIKELY(region->num_freed_chunks < n_chunks)) {
+      if (UNLIKELY(!PopulateFreeArray(stat, class_id, region,
+                                      n_chunks - region->num_freed_chunks)))
+        return false;
+      CHECK_GE(region->num_freed_chunks, n_chunks);
+    }
+    region->num_freed_chunks -= n_chunks;
+    uptr base_idx = region->num_freed_chunks;
+    for (uptr i = 0; i < n_chunks; i++)
+      chunks[i] = free_array[base_idx + i];
+    region->stats.n_allocated += n_chunks;
+    return true;
+  }
+
+  bool PointerIsMine(const void *p) {
+    uptr P = reinterpret_cast<uptr>(p);
+    if (kUsingConstantSpaceBeg && (kSpaceBeg % kSpaceSize) == 0)
+      return P / kSpaceSize == kSpaceBeg / kSpaceSize;
+    return P >= SpaceBeg() && P < SpaceEnd();
+  }
+
+  uptr GetRegionBegin(const void *p) {
+    if (kUsingConstantSpaceBeg)
+      return reinterpret_cast<uptr>(p) & ~(kRegionSize - 1);
+    uptr space_beg = SpaceBeg();
+    return ((reinterpret_cast<uptr>(p)  - space_beg) & ~(kRegionSize - 1)) +
+        space_beg;
+  }
+
+  uptr GetRegionBeginBySizeClass(uptr class_id) const {
+    return SpaceBeg() + kRegionSize * class_id;
+  }
+
+  uptr GetSizeClass(const void *p) {
+    if (kUsingConstantSpaceBeg && (kSpaceBeg % kSpaceSize) == 0)
+      return ((reinterpret_cast<uptr>(p)) / kRegionSize) % kNumClassesRounded;
+    return ((reinterpret_cast<uptr>(p) - SpaceBeg()) / kRegionSize) %
+           kNumClassesRounded;
+  }
+
+  void *GetBlockBegin(const void *p) {
+    uptr class_id = GetSizeClass(p);
+    uptr size = ClassIdToSize(class_id);
+    if (!size) return nullptr;
+    uptr chunk_idx = GetChunkIdx((uptr)p, size);
+    uptr reg_beg = GetRegionBegin(p);
+    uptr beg = chunk_idx * size;
+    uptr next_beg = beg + size;
+    if (class_id >= kNumClasses) return nullptr;
+    RegionInfo *region = GetRegionInfo(class_id);
+    if (region->mapped_user >= next_beg)
+      return reinterpret_cast<void*>(reg_beg + beg);
+    return nullptr;
+  }
+
+  uptr GetActuallyAllocatedSize(void *p) {
+    CHECK(PointerIsMine(p));
+    return ClassIdToSize(GetSizeClass(p));
+  }
+
+  uptr ClassID(uptr size) { return SizeClassMap::ClassID(size); }
+
+  void *GetMetaData(const void *p) {
+    uptr class_id = GetSizeClass(p);
+    uptr size = ClassIdToSize(class_id);
+    uptr chunk_idx = GetChunkIdx(reinterpret_cast<uptr>(p), size);
+    uptr region_beg = GetRegionBeginBySizeClass(class_id);
+    return reinterpret_cast<void *>(GetMetadataEnd(region_beg) -
+                                    (1 + chunk_idx) * kMetadataSize);
+  }
+
+  uptr TotalMemoryUsed() {
+    uptr res = 0;
+    for (uptr i = 0; i < kNumClasses; i++)
+      res += GetRegionInfo(i)->allocated_user;
+    return res;
+  }
+
+  // Test-only.
+  void TestOnlyUnmap() {
+    UnmapWithCallbackOrDie(SpaceBeg(), kSpaceSize + AdditionalSize());
+  }
+
+  static void FillMemoryProfile(uptr start, uptr rss, bool file, uptr *stats,
+                           uptr stats_size) {
+    for (uptr class_id = 0; class_id < stats_size; class_id++)
+      if (stats[class_id] == start)
+        stats[class_id] = rss;
+  }
+
+  void PrintStats(uptr class_id, uptr rss) {
+    RegionInfo *region = GetRegionInfo(class_id);
+    if (region->mapped_user == 0) return;
+    uptr in_use = region->stats.n_allocated - region->stats.n_freed;
+    uptr avail_chunks = region->allocated_user / ClassIdToSize(class_id);
+    Printf(
+        "%s %02zd (%6zd): mapped: %6zdK allocs: %7zd frees: %7zd inuse: %6zd "
+        "num_freed_chunks %7zd avail: %6zd rss: %6zdK releases: %6zd "
+        "last released: %6zdK region: 0x%zx\n",
+        region->exhausted ? "F" : " ", class_id, ClassIdToSize(class_id),
+        region->mapped_user >> 10, region->stats.n_allocated,
+        region->stats.n_freed, in_use, region->num_freed_chunks, avail_chunks,
+        rss >> 10, region->rtoi.num_releases,
+        region->rtoi.last_released_bytes >> 10,
+        SpaceBeg() + kRegionSize * class_id);
+  }
+
+  void PrintStats() {
+    uptr total_mapped = 0;
+    uptr n_allocated = 0;
+    uptr n_freed = 0;
+    for (uptr class_id = 1; class_id < kNumClasses; class_id++) {
+      RegionInfo *region = GetRegionInfo(class_id);
+      total_mapped += region->mapped_user;
+      n_allocated += region->stats.n_allocated;
+      n_freed += region->stats.n_freed;
+    }
+    Printf("Stats: SizeClassAllocator64: %zdM mapped in %zd allocations; "
+           "remains %zd\n",
+           total_mapped >> 20, n_allocated, n_allocated - n_freed);
+    uptr rss_stats[kNumClasses];
+    for (uptr class_id = 0; class_id < kNumClasses; class_id++)
+      rss_stats[class_id] = SpaceBeg() + kRegionSize * class_id;
+    GetMemoryProfile(FillMemoryProfile, rss_stats, kNumClasses);
+    for (uptr class_id = 1; class_id < kNumClasses; class_id++)
+      PrintStats(class_id, rss_stats[class_id]);
+  }
+
+  // ForceLock() and ForceUnlock() are needed to implement Darwin malloc zone
+  // introspection API.
+  void ForceLock() {
+    for (uptr i = 0; i < kNumClasses; i++) {
+      GetRegionInfo(i)->mutex.Lock();
+    }
+  }
+
+  void ForceUnlock() {
+    for (int i = (int)kNumClasses - 1; i >= 0; i--) {
+      GetRegionInfo(i)->mutex.Unlock();
+    }
+  }
+
+  // Iterate over all existing chunks.
+  // The allocator must be locked when calling this function.
+  void ForEachChunk(ForEachChunkCallback callback, void *arg) {
+    for (uptr class_id = 1; class_id < kNumClasses; class_id++) {
+      RegionInfo *region = GetRegionInfo(class_id);
+      uptr chunk_size = ClassIdToSize(class_id);
+      uptr region_beg = SpaceBeg() + class_id * kRegionSize;
+      for (uptr chunk = region_beg;
+           chunk < region_beg + region->allocated_user;
+           chunk += chunk_size) {
+        // Too slow: CHECK_EQ((void *)chunk, GetBlockBegin((void *)chunk));
+        callback(chunk, arg);
+      }
+    }
+  }
+
+  static uptr ClassIdToSize(uptr class_id) {
+    return SizeClassMap::Size(class_id);
+  }
+
+  static uptr AdditionalSize() {
+    return RoundUpTo(sizeof(RegionInfo) * kNumClassesRounded,
+                     GetPageSizeCached());
+  }
+
+  typedef SizeClassMap SizeClassMapT;
+  static const uptr kNumClasses = SizeClassMap::kNumClasses;
+  static const uptr kNumClassesRounded = SizeClassMap::kNumClassesRounded;
+
+  // A packed array of counters. Each counter occupies 2^n bits, enough to store
+  // counter's max_value. Ctor will try to allocate the required buffer via
+  // mapper->MapPackedCounterArrayBuffer and the caller is expected to check
+  // whether the initialization was successful by checking IsAllocated() result.
+  // For the performance sake, none of the accessors check the validity of the
+  // arguments, it is assumed that index is always in [0, n) range and the value
+  // is not incremented past max_value.
+  template<class MemoryMapperT>
+  class PackedCounterArray {
+   public:
+    PackedCounterArray(u64 num_counters, u64 max_value, MemoryMapperT *mapper)
+        : n(num_counters), memory_mapper(mapper) {
+      CHECK_GT(num_counters, 0);
+      CHECK_GT(max_value, 0);
+      constexpr u64 kMaxCounterBits = sizeof(*buffer) * 8ULL;
+      // Rounding counter storage size up to the power of two allows for using
+      // bit shifts calculating particular counter's index and offset.
+      uptr counter_size_bits =
+          RoundUpToPowerOfTwo(MostSignificantSetBitIndex(max_value) + 1);
+      CHECK_LE(counter_size_bits, kMaxCounterBits);
+      counter_size_bits_log = Log2(counter_size_bits);
+      counter_mask = ~0ULL >> (kMaxCounterBits - counter_size_bits);
+
+      uptr packing_ratio = kMaxCounterBits >> counter_size_bits_log;
+      CHECK_GT(packing_ratio, 0);
+      packing_ratio_log = Log2(packing_ratio);
+      bit_offset_mask = packing_ratio - 1;
+
+      buffer_size =
+          (RoundUpTo(n, 1ULL << packing_ratio_log) >> packing_ratio_log) *
+          sizeof(*buffer);
+      buffer = reinterpret_cast<u64*>(
+          memory_mapper->MapPackedCounterArrayBuffer(buffer_size));
+    }
+    ~PackedCounterArray() {
+      if (buffer) {
+        memory_mapper->UnmapPackedCounterArrayBuffer(
+            reinterpret_cast<uptr>(buffer), buffer_size);
+      }
+    }
+
+    bool IsAllocated() const {
+      return !!buffer;
+    }
+
+    u64 GetCount() const {
+      return n;
+    }
+
+    uptr Get(uptr i) const {
+      DCHECK_LT(i, n);
+      uptr index = i >> packing_ratio_log;
+      uptr bit_offset = (i & bit_offset_mask) << counter_size_bits_log;
+      return (buffer[index] >> bit_offset) & counter_mask;
+    }
+
+    void Inc(uptr i) const {
+      DCHECK_LT(Get(i), counter_mask);
+      uptr index = i >> packing_ratio_log;
+      uptr bit_offset = (i & bit_offset_mask) << counter_size_bits_log;
+      buffer[index] += 1ULL << bit_offset;
+    }
+
+    void IncRange(uptr from, uptr to) const {
+      DCHECK_LE(from, to);
+      for (uptr i = from; i <= to; i++)
+        Inc(i);
+    }
+
+   private:
+    const u64 n;
+    u64 counter_size_bits_log;
+    u64 counter_mask;
+    u64 packing_ratio_log;
+    u64 bit_offset_mask;
+
+    MemoryMapperT* const memory_mapper;
+    u64 buffer_size;
+    u64* buffer;
+  };
+
+  template<class MemoryMapperT>
+  class FreePagesRangeTracker {
+   public:
+    explicit FreePagesRangeTracker(MemoryMapperT* mapper)
+        : memory_mapper(mapper),
+          page_size_scaled_log(Log2(GetPageSizeCached() >> kCompactPtrScale)),
+          in_the_range(false), current_page(0), current_range_start_page(0) {}
+
+    void NextPage(bool freed) {
+      if (freed) {
+        if (!in_the_range) {
+          current_range_start_page = current_page;
+          in_the_range = true;
+        }
+      } else {
+        CloseOpenedRange();
+      }
+      current_page++;
+    }
+
+    void Done() {
+      CloseOpenedRange();
+    }
+
+   private:
+    void CloseOpenedRange() {
+      if (in_the_range) {
+        memory_mapper->ReleasePageRangeToOS(
+            current_range_start_page << page_size_scaled_log,
+            current_page << page_size_scaled_log);
+        in_the_range = false;
+      }
+    }
+
+    MemoryMapperT* const memory_mapper;
+    const uptr page_size_scaled_log;
+    bool in_the_range;
+    uptr current_page;
+    uptr current_range_start_page;
+  };
+
+  // Iterates over the free_array to identify memory pages containing freed
+  // chunks only and returns these pages back to OS.
+  // allocated_pages_count is the total number of pages allocated for the
+  // current bucket.
+  template<class MemoryMapperT>
+  static void ReleaseFreeMemoryToOS(CompactPtrT *free_array,
+                                    uptr free_array_count, uptr chunk_size,
+                                    uptr allocated_pages_count,
+                                    MemoryMapperT *memory_mapper) {
+    const uptr page_size = GetPageSizeCached();
+
+    // Figure out the number of chunks per page and whether we can take a fast
+    // path (the number of chunks per page is the same for all pages).
+    uptr full_pages_chunk_count_max;
+    bool same_chunk_count_per_page;
+    if (chunk_size <= page_size && page_size % chunk_size == 0) {
+      // Same number of chunks per page, no cross overs.
+      full_pages_chunk_count_max = page_size / chunk_size;
+      same_chunk_count_per_page = true;
+    } else if (chunk_size <= page_size && page_size % chunk_size != 0 &&
+        chunk_size % (page_size % chunk_size) == 0) {
+      // Some chunks are crossing page boundaries, which means that the page
+      // contains one or two partial chunks, but all pages contain the same
+      // number of chunks.
+      full_pages_chunk_count_max = page_size / chunk_size + 1;
+      same_chunk_count_per_page = true;
+    } else if (chunk_size <= page_size) {
+      // Some chunks are crossing page boundaries, which means that the page
+      // contains one or two partial chunks.
+      full_pages_chunk_count_max = page_size / chunk_size + 2;
+      same_chunk_count_per_page = false;
+    } else if (chunk_size > page_size && chunk_size % page_size == 0) {
+      // One chunk covers multiple pages, no cross overs.
+      full_pages_chunk_count_max = 1;
+      same_chunk_count_per_page = true;
+    } else if (chunk_size > page_size) {
+      // One chunk covers multiple pages, Some chunks are crossing page
+      // boundaries. Some pages contain one chunk, some contain two.
+      full_pages_chunk_count_max = 2;
+      same_chunk_count_per_page = false;
+    } else {
+      UNREACHABLE("All chunk_size/page_size ratios must be handled.");
+    }
+
+    PackedCounterArray<MemoryMapperT> counters(allocated_pages_count,
+                                               full_pages_chunk_count_max,
+                                               memory_mapper);
+    if (!counters.IsAllocated())
+      return;
+
+    const uptr chunk_size_scaled = chunk_size >> kCompactPtrScale;
+    const uptr page_size_scaled = page_size >> kCompactPtrScale;
+    const uptr page_size_scaled_log = Log2(page_size_scaled);
+
+    // Iterate over free chunks and count how many free chunks affect each
+    // allocated page.
+    if (chunk_size <= page_size && page_size % chunk_size == 0) {
+      // Each chunk affects one page only.
+      for (uptr i = 0; i < free_array_count; i++)
+        counters.Inc(free_array[i] >> page_size_scaled_log);
+    } else {
+      // In all other cases chunks might affect more than one page.
+      for (uptr i = 0; i < free_array_count; i++) {
+        counters.IncRange(
+            free_array[i] >> page_size_scaled_log,
+            (free_array[i] + chunk_size_scaled - 1) >> page_size_scaled_log);
+      }
+    }
+
+    // Iterate over pages detecting ranges of pages with chunk counters equal
+    // to the expected number of chunks for the particular page.
+    FreePagesRangeTracker<MemoryMapperT> range_tracker(memory_mapper);
+    if (same_chunk_count_per_page) {
+      // Fast path, every page has the same number of chunks affecting it.
+      for (uptr i = 0; i < counters.GetCount(); i++)
+        range_tracker.NextPage(counters.Get(i) == full_pages_chunk_count_max);
+    } else {
+      // Show path, go through the pages keeping count how many chunks affect
+      // each page.
+      const uptr pn =
+          chunk_size < page_size ? page_size_scaled / chunk_size_scaled : 1;
+      const uptr pnc = pn * chunk_size_scaled;
+      // The idea is to increment the current page pointer by the first chunk
+      // size, middle portion size (the portion of the page covered by chunks
+      // except the first and the last one) and then the last chunk size, adding
+      // up the number of chunks on the current page and checking on every step
+      // whether the page boundary was crossed.
+      uptr prev_page_boundary = 0;
+      uptr current_boundary = 0;
+      for (uptr i = 0; i < counters.GetCount(); i++) {
+        uptr page_boundary = prev_page_boundary + page_size_scaled;
+        uptr chunks_per_page = pn;
+        if (current_boundary < page_boundary) {
+          if (current_boundary > prev_page_boundary)
+            chunks_per_page++;
+          current_boundary += pnc;
+          if (current_boundary < page_boundary) {
+            chunks_per_page++;
+            current_boundary += chunk_size_scaled;
+          }
+        }
+        prev_page_boundary = page_boundary;
+
+        range_tracker.NextPage(counters.Get(i) == chunks_per_page);
+      }
+    }
+    range_tracker.Done();
+  }
+
+ private:
+  friend class MemoryMapper;
+
+  static const uptr kRegionSize = kSpaceSize / kNumClassesRounded;
+  // FreeArray is the array of free-d chunks (stored as 4-byte offsets).
+  // In the worst case it may reguire kRegionSize/SizeClassMap::kMinSize
+  // elements, but in reality this will not happen. For simplicity we
+  // dedicate 1/8 of the region's virtual space to FreeArray.
+  static const uptr kFreeArraySize = kRegionSize / 8;
+
+  static const bool kUsingConstantSpaceBeg = kSpaceBeg != ~(uptr)0;
+  uptr NonConstSpaceBeg;
+  uptr SpaceBeg() const {
+    return kUsingConstantSpaceBeg ? kSpaceBeg : NonConstSpaceBeg;
+  }
+  uptr SpaceEnd() const { return  SpaceBeg() + kSpaceSize; }
+  // kRegionSize must be >= 2^32.
+  COMPILER_CHECK((kRegionSize) >= (1ULL << (SANITIZER_WORDSIZE / 2)));
+  // kRegionSize must be <= 2^36, see CompactPtrT.
+  COMPILER_CHECK((kRegionSize) <= (1ULL << (SANITIZER_WORDSIZE / 2 + 4)));
+  // Call mmap for user memory with at least this size.
+  static const uptr kUserMapSize = 1 << 16;
+  // Call mmap for metadata memory with at least this size.
+  static const uptr kMetaMapSize = 1 << 16;
+  // Call mmap for free array memory with at least this size.
+  static const uptr kFreeArrayMapSize = 1 << 16;
+
+  atomic_sint32_t release_to_os_interval_ms_;
+
+  struct Stats {
+    uptr n_allocated;
+    uptr n_freed;
+  };
+
+  struct ReleaseToOsInfo {
+    uptr n_freed_at_last_release;
+    uptr num_releases;
+    u64 last_release_at_ns;
+    u64 last_released_bytes;
+  };
+
+  struct RegionInfo {
+    BlockingMutex mutex;
+    uptr num_freed_chunks;  // Number of elements in the freearray.
+    uptr mapped_free_array;  // Bytes mapped for freearray.
+    uptr allocated_user;  // Bytes allocated for user memory.
+    uptr allocated_meta;  // Bytes allocated for metadata.
+    uptr mapped_user;  // Bytes mapped for user memory.
+    uptr mapped_meta;  // Bytes mapped for metadata.
+    u32 rand_state;  // Seed for random shuffle, used if kRandomShuffleChunks.
+    bool exhausted;  // Whether region is out of space for new chunks.
+    Stats stats;
+    ReleaseToOsInfo rtoi;
+  };
+  COMPILER_CHECK(sizeof(RegionInfo) >= kCacheLineSize);
+
+  u32 Rand(u32 *state) {  // ANSI C linear congruential PRNG.
+    return (*state = *state * 1103515245 + 12345) >> 16;
+  }
+
+  u32 RandN(u32 *state, u32 n) { return Rand(state) % n; }  // [0, n)
+
+  void RandomShuffle(u32 *a, u32 n, u32 *rand_state) {
+    if (n <= 1) return;
+    for (u32 i = n - 1; i > 0; i--)
+      Swap(a[i], a[RandN(rand_state, i + 1)]);
+  }
+
+  RegionInfo *GetRegionInfo(uptr class_id) const {
+    CHECK_LT(class_id, kNumClasses);
+    RegionInfo *regions =
+        reinterpret_cast<RegionInfo *>(SpaceBeg() + kSpaceSize);
+    return &regions[class_id];
+  }
+
+  uptr GetMetadataEnd(uptr region_beg) const {
+    return region_beg + kRegionSize - kFreeArraySize;
+  }
+
+  uptr GetChunkIdx(uptr chunk, uptr size) const {
+    if (!kUsingConstantSpaceBeg)
+      chunk -= SpaceBeg();
+
+    uptr offset = chunk % kRegionSize;
+    // Here we divide by a non-constant. This is costly.
+    // size always fits into 32-bits. If the offset fits too, use 32-bit div.
+    if (offset >> (SANITIZER_WORDSIZE / 2))
+      return offset / size;
+    return (u32)offset / (u32)size;
+  }
+
+  CompactPtrT *GetFreeArray(uptr region_beg) const {
+    return reinterpret_cast<CompactPtrT *>(GetMetadataEnd(region_beg));
+  }
+
+  bool MapWithCallback(uptr beg, uptr size) {
+    uptr mapped = reinterpret_cast<uptr>(MmapFixedOrDieOnFatalError(beg, size));
+    if (UNLIKELY(!mapped))
+      return false;
+    CHECK_EQ(beg, mapped);
+    MapUnmapCallback().OnMap(beg, size);
+    return true;
+  }
+
+  void MapWithCallbackOrDie(uptr beg, uptr size) {
+    CHECK_EQ(beg, reinterpret_cast<uptr>(MmapFixedOrDie(beg, size)));
+    MapUnmapCallback().OnMap(beg, size);
+  }
+
+  void UnmapWithCallbackOrDie(uptr beg, uptr size) {
+    MapUnmapCallback().OnUnmap(beg, size);
+    UnmapOrDie(reinterpret_cast<void *>(beg), size);
+  }
+
+  bool EnsureFreeArraySpace(RegionInfo *region, uptr region_beg,
+                            uptr num_freed_chunks) {
+    uptr needed_space = num_freed_chunks * sizeof(CompactPtrT);
+    if (region->mapped_free_array < needed_space) {
+      uptr new_mapped_free_array = RoundUpTo(needed_space, kFreeArrayMapSize);
+      CHECK_LE(new_mapped_free_array, kFreeArraySize);
+      uptr current_map_end = reinterpret_cast<uptr>(GetFreeArray(region_beg)) +
+                             region->mapped_free_array;
+      uptr new_map_size = new_mapped_free_array - region->mapped_free_array;
+      if (UNLIKELY(!MapWithCallback(current_map_end, new_map_size)))
+        return false;
+      region->mapped_free_array = new_mapped_free_array;
+    }
+    return true;
+  }
+
+  NOINLINE bool PopulateFreeArray(AllocatorStats *stat, uptr class_id,
+                                  RegionInfo *region, uptr requested_count) {
+    // region->mutex is held.
+    const uptr size = ClassIdToSize(class_id);
+    const uptr new_space_beg = region->allocated_user;
+    const uptr new_space_end = new_space_beg + requested_count * size;
+    const uptr region_beg = GetRegionBeginBySizeClass(class_id);
+
+    // Map more space for chunks, if necessary.
+    if (new_space_end > region->mapped_user) {
+      if (!kUsingConstantSpaceBeg && region->mapped_user == 0)
+        region->rand_state = static_cast<u32>(region_beg >> 12);  // From ASLR.
+      // Do the mmap for the user memory.
+      uptr map_size = kUserMapSize;
+      while (new_space_end > region->mapped_user + map_size)
+        map_size += kUserMapSize;
+      CHECK_GE(region->mapped_user + map_size, new_space_end);
+      if (UNLIKELY(!MapWithCallback(region_beg + region->mapped_user,
+                                    map_size)))
+        return false;
+      stat->Add(AllocatorStatMapped, map_size);
+      region->mapped_user += map_size;
+    }
+    const uptr new_chunks_count = (region->mapped_user - new_space_beg) / size;
+
+    // Calculate the required space for metadata.
+    const uptr requested_allocated_meta =
+        region->allocated_meta + new_chunks_count * kMetadataSize;
+    uptr requested_mapped_meta = region->mapped_meta;
+    while (requested_allocated_meta > requested_mapped_meta)
+      requested_mapped_meta += kMetaMapSize;
+    // Check whether this size class is exhausted.
+    if (region->mapped_user + requested_mapped_meta >
+        kRegionSize - kFreeArraySize) {
+      if (!region->exhausted) {
+        region->exhausted = true;
+        Printf("%s: Out of memory. ", SanitizerToolName);
+        Printf("The process has exhausted %zuMB for size class %zu.\n",
+               kRegionSize >> 20, size);
+      }
+      return false;
+    }
+    // Map more space for metadata, if necessary.
+    if (requested_mapped_meta > region->mapped_meta) {
+      if (UNLIKELY(!MapWithCallback(
+              GetMetadataEnd(region_beg) - requested_mapped_meta,
+              requested_mapped_meta - region->mapped_meta)))
+        return false;
+      region->mapped_meta = requested_mapped_meta;
+    }
+
+    // If necessary, allocate more space for the free array and populate it with
+    // newly allocated chunks.
+    const uptr total_freed_chunks = region->num_freed_chunks + new_chunks_count;
+    if (UNLIKELY(!EnsureFreeArraySpace(region, region_beg, total_freed_chunks)))
+      return false;
+    CompactPtrT *free_array = GetFreeArray(region_beg);
+    for (uptr i = 0, chunk = new_space_beg; i < new_chunks_count;
+         i++, chunk += size)
+      free_array[total_freed_chunks - 1 - i] = PointerToCompactPtr(0, chunk);
+    if (kRandomShuffleChunks)
+      RandomShuffle(&free_array[region->num_freed_chunks], new_chunks_count,
+                    &region->rand_state);
+
+    // All necessary memory is mapped and now it is safe to advance all
+    // 'allocated_*' counters.
+    region->num_freed_chunks += new_chunks_count;
+    region->allocated_user += new_chunks_count * size;
+    CHECK_LE(region->allocated_user, region->mapped_user);
+    region->allocated_meta = requested_allocated_meta;
+    CHECK_LE(region->allocated_meta, region->mapped_meta);
+    region->exhausted = false;
+
+    // TODO(alekseyshl): Consider bumping last_release_at_ns here to prevent
+    // MaybeReleaseToOS from releasing just allocated pages or protect these
+    // not yet used chunks some other way.
+
+    return true;
+  }
+
+  class MemoryMapper {
+   public:
+    MemoryMapper(const ThisT& base_allocator, uptr class_id)
+        : allocator(base_allocator),
+          region_base(base_allocator.GetRegionBeginBySizeClass(class_id)),
+          released_ranges_count(0),
+          released_bytes(0) {
+    }
+
+    uptr GetReleasedRangesCount() const {
+      return released_ranges_count;
+    }
+
+    uptr GetReleasedBytes() const {
+      return released_bytes;
+    }
+
+    uptr MapPackedCounterArrayBuffer(uptr buffer_size) {
+      // TODO(alekseyshl): The idea to explore is to check if we have enough
+      // space between num_freed_chunks*sizeof(CompactPtrT) and
+      // mapped_free_array to fit buffer_size bytes and use that space instead
+      // of mapping a temporary one.
+      return reinterpret_cast<uptr>(
+          MmapOrDieOnFatalError(buffer_size, "ReleaseToOSPageCounters"));
+    }
+
+    void UnmapPackedCounterArrayBuffer(uptr buffer, uptr buffer_size) {
+      UnmapOrDie(reinterpret_cast<void *>(buffer), buffer_size);
+    }
+
+    // Releases [from, to) range of pages back to OS.
+    void ReleasePageRangeToOS(CompactPtrT from, CompactPtrT to) {
+      const uptr from_page = allocator.CompactPtrToPointer(region_base, from);
+      const uptr to_page = allocator.CompactPtrToPointer(region_base, to);
+      ReleaseMemoryPagesToOS(from_page, to_page);
+      released_ranges_count++;
+      released_bytes += to_page - from_page;
+    }
+
+   private:
+    const ThisT& allocator;
+    const uptr region_base;
+    uptr released_ranges_count;
+    uptr released_bytes;
+  };
+
+  // Attempts to release RAM occupied by freed chunks back to OS. The region is
+  // expected to be locked.
+  void MaybeReleaseToOS(uptr class_id) {
+    RegionInfo *region = GetRegionInfo(class_id);
+    const uptr chunk_size = ClassIdToSize(class_id);
+    const uptr page_size = GetPageSizeCached();
+
+    uptr n = region->num_freed_chunks;
+    if (n * chunk_size < page_size)
+      return;  // No chance to release anything.
+    if ((region->stats.n_freed -
+         region->rtoi.n_freed_at_last_release) * chunk_size < page_size) {
+      return;  // Nothing new to release.
+    }
+
+    s32 interval_ms = ReleaseToOSIntervalMs();
+    if (interval_ms < 0)
+      return;
+
+    if (region->rtoi.last_release_at_ns + interval_ms * 1000000ULL > NanoTime())
+      return;  // Memory was returned recently.
+
+    MemoryMapper memory_mapper(*this, class_id);
+
+    ReleaseFreeMemoryToOS<MemoryMapper>(
+        GetFreeArray(GetRegionBeginBySizeClass(class_id)), n, chunk_size,
+        RoundUpTo(region->allocated_user, page_size) / page_size,
+        &memory_mapper);
+
+    if (memory_mapper.GetReleasedRangesCount() > 0) {
+      region->rtoi.n_freed_at_last_release = region->stats.n_freed;
+      region->rtoi.num_releases += memory_mapper.GetReleasedRangesCount();
+      region->rtoi.last_released_bytes = memory_mapper.GetReleasedBytes();
+    }
+    region->rtoi.last_release_at_ns = NanoTime();
+  }
+};