OLD | NEW |
1 // Copyright (c) 2008, Google Inc. | 1 // Copyright (c) 2008, Google Inc. |
2 // All rights reserved. | 2 // All rights reserved. |
3 // | 3 // |
4 // Redistribution and use in source and binary forms, with or without | 4 // Redistribution and use in source and binary forms, with or without |
5 // modification, are permitted provided that the following conditions are | 5 // modification, are permitted provided that the following conditions are |
6 // met: | 6 // met: |
7 // | 7 // |
8 // * Redistributions of source code must retain the above copyright | 8 // * Redistributions of source code must retain the above copyright |
9 // notice, this list of conditions and the following disclaimer. | 9 // notice, this list of conditions and the following disclaimer. |
10 // * Redistributions in binary form must reproduce the above | 10 // * Redistributions in binary form must reproduce the above |
(...skipping 13 matching lines...) Expand all Loading... |
24 // LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, | 24 // LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, |
25 // DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY | 25 // DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY |
26 // THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT | 26 // THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT |
27 // (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE | 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. | 28 // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
29 | 29 |
30 // --- | 30 // --- |
31 // Author: Sanjay Ghemawat <opensource@google.com> | 31 // Author: Sanjay Ghemawat <opensource@google.com> |
32 // | 32 // |
33 // Common definitions for tcmalloc code. | 33 // Common definitions for tcmalloc code. |
| 34 |
34 #ifndef TCMALLOC_COMMON_H_ | 35 #ifndef TCMALLOC_COMMON_H_ |
35 #define TCMALLOC_COMMON_H_ | 36 #define TCMALLOC_COMMON_H_ |
36 | 37 |
37 #include "config.h" | 38 #include "config.h" |
38 #include <stddef.h> // for size_t | 39 #include <stddef.h> // for size_t |
39 #ifdef HAVE_STDINT_H | 40 #ifdef HAVE_STDINT_H |
40 #include <stdint.h> // for uintptr_t, uint64_t | 41 #include <stdint.h> // for uintptr_t, uint64_t |
41 #endif | 42 #endif |
42 #include "free_list.h" // for SIZE_CLASS macros | |
43 #include "internal_logging.h" // for ASSERT, etc | 43 #include "internal_logging.h" // for ASSERT, etc |
44 | 44 |
45 // Type that can hold a page number | 45 // Type that can hold a page number |
46 typedef uintptr_t PageID; | 46 typedef uintptr_t PageID; |
47 | 47 |
48 // Type that can hold the length of a run of pages | 48 // Type that can hold the length of a run of pages |
49 typedef uintptr_t Length; | 49 typedef uintptr_t Length; |
50 | 50 |
51 //------------------------------------------------------------------- | 51 //------------------------------------------------------------------- |
52 // Configuration | 52 // Configuration |
53 //------------------------------------------------------------------- | 53 //------------------------------------------------------------------- |
54 | 54 |
55 // Using large pages speeds up the execution at a cost of larger memory use. | 55 // Using large pages speeds up the execution at a cost of larger memory use. |
56 // Deallocation may speed up by a factor as the page map gets 8x smaller, so | 56 // Deallocation may speed up by a factor as the page map gets 8x smaller, so |
57 // lookups in the page map result in fewer L2 cache misses, which translates to | 57 // lookups in the page map result in fewer L2 cache misses, which translates to |
58 // speedup for application/platform combinations with high L2 cache pressure. | 58 // speedup for application/platform combinations with high L2 cache pressure. |
59 // As the number of size classes increases with large pages, we increase | 59 // As the number of size classes increases with large pages, we increase |
60 // the thread cache allowance to avoid passing more free ranges to and from | 60 // the thread cache allowance to avoid passing more free ranges to and from |
61 // central lists. Also, larger pages are less likely to get freed. | 61 // central lists. Also, larger pages are less likely to get freed. |
62 // These two factors cause a bounded increase in memory use. | 62 // These two factors cause a bounded increase in memory use. |
63 | 63 |
64 static const size_t kAlignment = 8; | |
65 | |
66 // Constants dependant on tcmalloc configuration and archetecture. | |
67 // We need to guarantee the smallest class size is big enough to hold the | |
68 // pointers that form the free list. | |
69 static const size_t kNumFreeListPointers = | |
70 (tcmalloc::kSupportsDoublyLinkedList ? 2 : 1); | |
71 static const size_t kLinkSize = kNumFreeListPointers * sizeof(void *); | |
72 static const size_t kMinClassSize = | |
73 (kLinkSize > kAlignment ? kLinkSize : kAlignment); | |
74 static const size_t kSkippedClasses = (kAlignment < kMinClassSize ? 1 : 0); | |
75 | |
76 #if defined(TCMALLOC_LARGE_PAGES) | 64 #if defined(TCMALLOC_LARGE_PAGES) |
77 static const size_t kPageShift = 15; | 65 static const size_t kPageShift = 15; |
78 static const size_t kNumClasses = 95 - kSkippedClasses; | 66 static const size_t kNumClasses = 78; |
| 67 #else |
| 68 static const size_t kPageShift = 13; |
| 69 static const size_t kNumClasses = 86; |
| 70 #endif |
79 static const size_t kMaxThreadCacheSize = 4 << 20; | 71 static const size_t kMaxThreadCacheSize = 4 << 20; |
80 #else | |
81 static const size_t kPageShift = 12; | |
82 static const size_t kNumClasses = 61 - kSkippedClasses; | |
83 static const size_t kMaxThreadCacheSize = 2 << 20; | |
84 #endif | |
85 | 72 |
86 static const size_t kPageSize = 1 << kPageShift; | 73 static const size_t kPageSize = 1 << kPageShift; |
87 static const size_t kMaxSize = 8u * kPageSize; | 74 static const size_t kMaxSize = 256 * 1024; |
| 75 static const size_t kAlignment = 8; |
| 76 static const size_t kLargeSizeClass = 0; |
88 // For all span-lengths < kMaxPages we keep an exact-size list. | 77 // For all span-lengths < kMaxPages we keep an exact-size list. |
89 static const size_t kMaxPages = 1 << (20 - kPageShift); | 78 static const size_t kMaxPages = 1 << (20 - kPageShift); |
90 | 79 |
91 // Default bound on the total amount of thread caches. | 80 // Default bound on the total amount of thread caches. |
92 #ifdef TCMALLOC_SMALL_BUT_SLOW | 81 #ifdef TCMALLOC_SMALL_BUT_SLOW |
93 // Make the overall thread cache no bigger than that of a single thread | 82 // Make the overall thread cache no bigger than that of a single thread |
94 // for the small memory footprint case. | 83 // for the small memory footprint case. |
95 static const size_t kDefaultOverallThreadCacheSize = kMaxThreadCacheSize; | 84 static const size_t kDefaultOverallThreadCacheSize = kMaxThreadCacheSize; |
96 #else | 85 #else |
97 static const size_t kDefaultOverallThreadCacheSize = 8u * kMaxThreadCacheSize; | 86 static const size_t kDefaultOverallThreadCacheSize = 8u * kMaxThreadCacheSize; |
(...skipping 71 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
169 // ------------------------------------------------------- | 158 // ------------------------------------------------------- |
170 // 0 (0 + 7) / 8 0 | 159 // 0 (0 + 7) / 8 0 |
171 // 1 (1 + 7) / 8 1 | 160 // 1 (1 + 7) / 8 1 |
172 // ... | 161 // ... |
173 // 1024 (1024 + 7) / 8 128 | 162 // 1024 (1024 + 7) / 8 128 |
174 // 1025 (1025 + 127 + (120<<7)) / 128 129 | 163 // 1025 (1025 + 127 + (120<<7)) / 128 129 |
175 // ... | 164 // ... |
176 // 32768 (32768 + 127 + (120<<7)) / 128 376 | 165 // 32768 (32768 + 127 + (120<<7)) / 128 376 |
177 static const int kMaxSmallSize = 1024; | 166 static const int kMaxSmallSize = 1024; |
178 static const size_t kClassArraySize = | 167 static const size_t kClassArraySize = |
179 (((1 << kPageShift) * 8u + 127 + (120 << 7)) >> 7) + 1; | 168 ((kMaxSize + 127 + (120 << 7)) >> 7) + 1; |
180 unsigned char class_array_[kClassArraySize]; | 169 unsigned char class_array_[kClassArraySize]; |
181 | 170 |
182 // Compute index of the class_array[] entry for a given size | 171 // Compute index of the class_array[] entry for a given size |
183 static inline int ClassIndex(int s) { | 172 static inline int ClassIndex(int s) { |
184 ASSERT(0 <= s); | 173 ASSERT(0 <= s); |
185 ASSERT(s <= kMaxSize); | 174 ASSERT(s <= kMaxSize); |
186 const bool big = (s > kMaxSmallSize); | 175 const bool big = (s > kMaxSmallSize); |
187 const int add_amount = big ? (127 + (120<<7)) : 7; | 176 const int add_amount = big ? (127 + (120<<7)) : 7; |
188 const int shift_amount = big ? 7 : 3; | 177 const int shift_amount = big ? 7 : 3; |
189 return (s + add_amount) >> shift_amount; | 178 return (s + add_amount) >> shift_amount; |
(...skipping 35 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
225 } | 214 } |
226 | 215 |
227 // Number of objects to move between a per-thread list and a central | 216 // Number of objects to move between a per-thread list and a central |
228 // list in one shot. We want this to be not too small so we can | 217 // list in one shot. We want this to be not too small so we can |
229 // amortize the lock overhead for accessing the central list. Making | 218 // amortize the lock overhead for accessing the central list. Making |
230 // it too big may temporarily cause unnecessary memory wastage in the | 219 // it too big may temporarily cause unnecessary memory wastage in the |
231 // per-thread free list until the scavenger cleans up the list. | 220 // per-thread free list until the scavenger cleans up the list. |
232 inline int num_objects_to_move(size_t cl) { | 221 inline int num_objects_to_move(size_t cl) { |
233 return num_objects_to_move_[cl]; | 222 return num_objects_to_move_[cl]; |
234 } | 223 } |
235 | |
236 // Dump contents of the computed size map | |
237 void Dump(TCMalloc_Printer* out); | |
238 }; | 224 }; |
239 | 225 |
240 // Allocates "bytes" worth of memory and returns it. Increments | 226 // Allocates "bytes" worth of memory and returns it. Increments |
241 // metadata_system_bytes appropriately. May return NULL if allocation | 227 // metadata_system_bytes appropriately. May return NULL if allocation |
242 // fails. Requires pageheap_lock is held. | 228 // fails. Requires pageheap_lock is held. |
243 void* MetaDataAlloc(size_t bytes); | 229 void* MetaDataAlloc(size_t bytes); |
244 | 230 |
245 // Returns the total number of bytes allocated from the system. | 231 // Returns the total number of bytes allocated from the system. |
246 // Requires pageheap_lock is held. | 232 // Requires pageheap_lock is held. |
247 uint64_t metadata_system_bytes(); | 233 uint64_t metadata_system_bytes(); |
248 | 234 |
249 // Adjust metadata_system_bytes to indicate that bytes are actually committed. | |
250 // Requires pageheap_lock is held. | |
251 void increment_metadata_system_bytes(size_t bytes); | |
252 | |
253 // size/depth are made the same size as a pointer so that some generic | 235 // size/depth are made the same size as a pointer so that some generic |
254 // code below can conveniently cast them back and forth to void*. | 236 // code below can conveniently cast them back and forth to void*. |
255 static const int kMaxStackDepth = 31; | 237 static const int kMaxStackDepth = 31; |
256 struct StackTrace { | 238 struct StackTrace { |
257 uintptr_t size; // Size of object | 239 uintptr_t size; // Size of object |
258 uintptr_t depth; // Number of PC values stored in array below | 240 uintptr_t depth; // Number of PC values stored in array below |
259 void* stack[kMaxStackDepth]; | 241 void* stack[kMaxStackDepth]; |
260 }; | 242 }; |
261 | 243 |
262 } // namespace tcmalloc | 244 } // namespace tcmalloc |
263 | 245 |
264 #endif // TCMALLOC_COMMON_H_ | 246 #endif // TCMALLOC_COMMON_H_ |
OLD | NEW |