1 // Copyright (c) 2008, Google Inc. 2 // All rights reserved. 3 // 4 // Redistribution and use in source and binary forms, with or without 5 // modification, are permitted provided that the following conditions are 6 // met: 7 // 8 // * Redistributions of source code must retain the above copyright 9 // notice, this list of conditions and the following disclaimer. 10 // * Redistributions in binary form must reproduce the above 11 // copyright notice, this list of conditions and the following disclaimer 12 // in the documentation and/or other materials provided with the 13 // distribution. 14 // * Neither the name of Google Inc. nor the names of its 15 // contributors may be used to endorse or promote products derived from 16 // this software without specific prior written permission. 17 // 18 // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 19 // "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 20 // LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 21 // A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 22 // OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 23 // SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 24 // LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 25 // DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 26 // THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 27 // (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 28 // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 29 30 // --- 31 // Author: Sanjay Ghemawat <opensource (at) google.com> 32 33 #include "config.h" 34 #include "common.h" 35 #include "system-alloc.h" 36 37 namespace tcmalloc { 38 39 // Note: the following only works for "n"s that fit in 32-bits, but 40 // that is fine since we only use it for small sizes. 41 static inline int LgFloor(size_t n) { 42 int log = 0; 43 for (int i = 4; i >= 0; --i) { 44 int shift = (1 << i); 45 size_t x = n >> shift; 46 if (x != 0) { 47 n = x; 48 log += shift; 49 } 50 } 51 ASSERT(n == 1); 52 return log; 53 } 54 55 int AlignmentForSize(size_t size) { 56 int alignment = kAlignment; 57 if (size > kMaxSize) { 58 // Cap alignment at kPageSize for large sizes. 59 alignment = kPageSize; 60 } else if (size >= 128) { 61 // Space wasted due to alignment is at most 1/8, i.e., 12.5%. 62 alignment = (1 << LgFloor(size)) / 8; 63 } else if (size >= 16) { 64 // We need an alignment of at least 16 bytes to satisfy 65 // requirements for some SSE types. 66 alignment = 16; 67 } 68 // Maximum alignment allowed is page size alignment. 69 if (alignment > kPageSize) { 70 alignment = kPageSize; 71 } 72 CHECK_CONDITION(size < 16 || alignment >= 16); 73 CHECK_CONDITION((alignment & (alignment - 1)) == 0); 74 return alignment; 75 } 76 77 int SizeMap::NumMoveSize(size_t size) { 78 if (size == 0) return 0; 79 // Use approx 64k transfers between thread and central caches. 80 int num = static_cast<int>(64.0 * 1024.0 / size); 81 if (num < 2) num = 2; 82 83 // Avoid bringing too many objects into small object free lists. 84 // If this value is too large: 85 // - We waste memory with extra objects sitting in the thread caches. 86 // - The central freelist holds its lock for too long while 87 // building a linked list of objects, slowing down the allocations 88 // of other threads. 89 // If this value is too small: 90 // - We go to the central freelist too often and we have to acquire 91 // its lock each time. 92 // This value strikes a balance between the constraints above. 93 if (num > 32) num = 32; 94 95 return num; 96 } 97 98 // Initialize the mapping arrays 99 void SizeMap::Init() { 100 // Do some sanity checking on add_amount[]/shift_amount[]/class_array[] 101 if (ClassIndex(0) < 0) { 102 Log(kCrash, __FILE__, __LINE__, 103 "Invalid class index for size 0", ClassIndex(0)); 104 } 105 if (ClassIndex(kMaxSize) >= sizeof(class_array_)) { 106 Log(kCrash, __FILE__, __LINE__, 107 "Invalid class index for kMaxSize", ClassIndex(kMaxSize)); 108 } 109 110 // Compute the size classes we want to use 111 int sc = 1; // Next size class to assign 112 int alignment = kAlignment; 113 CHECK_CONDITION(kAlignment <= 16); 114 for (size_t size = kAlignment; size <= kMaxSize; size += alignment) { 115 alignment = AlignmentForSize(size); 116 CHECK_CONDITION((size % alignment) == 0); 117 118 int blocks_to_move = NumMoveSize(size) / 4; 119 size_t psize = 0; 120 do { 121 psize += kPageSize; 122 // Allocate enough pages so leftover is less than 1/8 of total. 123 // This bounds wasted space to at most 12.5%. 124 while ((psize % size) > (psize >> 3)) { 125 psize += kPageSize; 126 } 127 // Continue to add pages until there are at least as many objects in 128 // the span as are needed when moving objects from the central 129 // freelists and spans to the thread caches. 130 } while ((psize / size) < (blocks_to_move)); 131 const size_t my_pages = psize >> kPageShift; 132 133 if (sc > 1 && my_pages == class_to_pages_[sc-1]) { 134 // See if we can merge this into the previous class without 135 // increasing the fragmentation of the previous class. 136 const size_t my_objects = (my_pages << kPageShift) / size; 137 const size_t prev_objects = (class_to_pages_[sc-1] << kPageShift) 138 / class_to_size_[sc-1]; 139 if (my_objects == prev_objects) { 140 // Adjust last class to include this size 141 class_to_size_[sc-1] = size; 142 continue; 143 } 144 } 145 146 // Add new class 147 class_to_pages_[sc] = my_pages; 148 class_to_size_[sc] = size; 149 sc++; 150 } 151 if (sc != kNumClasses) { 152 Log(kCrash, __FILE__, __LINE__, 153 "wrong number of size classes: (found vs. expected )", sc, kNumClasses); 154 } 155 156 // Initialize the mapping arrays 157 int next_size = 0; 158 for (int c = 1; c < kNumClasses; c++) { 159 const int max_size_in_class = class_to_size_[c]; 160 for (int s = next_size; s <= max_size_in_class; s += kAlignment) { 161 class_array_[ClassIndex(s)] = c; 162 } 163 next_size = max_size_in_class + kAlignment; 164 } 165 166 // Double-check sizes just to be safe 167 for (size_t size = 0; size <= kMaxSize; size++) { 168 const int sc = SizeClass(size); 169 if (sc <= 0 || sc >= kNumClasses) { 170 Log(kCrash, __FILE__, __LINE__, 171 "Bad size class (class, size)", sc, size); 172 } 173 if (sc > 1 && size <= class_to_size_[sc-1]) { 174 Log(kCrash, __FILE__, __LINE__, 175 "Allocating unnecessarily large class (class, size)", sc, size); 176 } 177 const size_t s = class_to_size_[sc]; 178 if (size > s || s == 0) { 179 Log(kCrash, __FILE__, __LINE__, 180 "Bad (class, size, requested)", sc, s, size); 181 } 182 } 183 184 // Initialize the num_objects_to_move array. 185 for (size_t cl = 1; cl < kNumClasses; ++cl) { 186 num_objects_to_move_[cl] = NumMoveSize(ByteSizeForClass(cl)); 187 } 188 } 189 190 // Metadata allocator -- keeps stats about how many bytes allocated. 191 static uint64_t metadata_system_bytes_ = 0; 192 void* MetaDataAlloc(size_t bytes) { 193 void* result = TCMalloc_SystemAlloc(bytes, NULL); 194 if (result != NULL) { 195 metadata_system_bytes_ += bytes; 196 } 197 return result; 198 } 199 200 uint64_t metadata_system_bytes() { return metadata_system_bytes_; } 201 202 } // namespace tcmalloc 203