1 | /* |
2 | * Copyright (c) 2014, 2019, Oracle and/or its affiliates. All rights reserved. |
3 | * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. |
4 | * |
5 | * This code is free software; you can redistribute it and/or modify it |
6 | * under the terms of the GNU General Public License version 2 only, as |
7 | * published by the Free Software Foundation. |
8 | * |
9 | * This code is distributed in the hope that it will be useful, but WITHOUT |
10 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or |
11 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
12 | * version 2 for more details (a copy is included in the LICENSE file that |
13 | * accompanied this code). |
14 | * |
15 | * You should have received a copy of the GNU General Public License version |
16 | * 2 along with this work; if not, write to the Free Software Foundation, |
17 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. |
18 | * |
19 | * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA |
20 | * or visit www.oracle.com if you need additional information or have any |
21 | * questions. |
22 | * |
23 | */ |
24 | |
25 | #ifndef SHARE_SERVICES_MALLOCSITETABLE_HPP |
26 | #define SHARE_SERVICES_MALLOCSITETABLE_HPP |
27 | |
28 | #if INCLUDE_NMT |
29 | |
30 | #include "memory/allocation.hpp" |
31 | #include "runtime/atomic.hpp" |
32 | #include "services/allocationSite.hpp" |
33 | #include "services/mallocTracker.hpp" |
34 | #include "services/nmtCommon.hpp" |
35 | #include "utilities/nativeCallStack.hpp" |
36 | |
37 | // MallocSite represents a code path that eventually calls |
38 | // os::malloc() to allocate memory |
39 | class MallocSite : public AllocationSite<MemoryCounter> { |
40 | public: |
41 | MallocSite() : |
42 | AllocationSite<MemoryCounter>(NativeCallStack::empty_stack(), mtNone) {} |
43 | |
44 | MallocSite(const NativeCallStack& stack, MEMFLAGS flags) : |
45 | AllocationSite<MemoryCounter>(stack, flags) {} |
46 | |
47 | |
48 | void allocate(size_t size) { data()->allocate(size); } |
49 | void deallocate(size_t size) { data()->deallocate(size); } |
50 | |
51 | // Memory allocated from this code path |
52 | size_t size() const { return peek()->size(); } |
53 | // The number of calls were made |
54 | size_t count() const { return peek()->count(); } |
55 | }; |
56 | |
57 | // Malloc site hashtable entry |
58 | class MallocSiteHashtableEntry : public CHeapObj<mtNMT> { |
59 | private: |
60 | MallocSite _malloc_site; |
61 | MallocSiteHashtableEntry* volatile _next; |
62 | |
63 | public: |
64 | MallocSiteHashtableEntry() : _next(NULL) { } |
65 | |
66 | MallocSiteHashtableEntry(NativeCallStack stack, MEMFLAGS flags): |
67 | _malloc_site(stack, flags), _next(NULL) { |
68 | assert(flags != mtNone, "Expect a real memory type" ); |
69 | } |
70 | |
71 | inline const MallocSiteHashtableEntry* next() const { |
72 | return _next; |
73 | } |
74 | |
75 | // Insert an entry atomically. |
76 | // Return true if the entry is inserted successfully. |
77 | // The operation can be failed due to contention from other thread. |
78 | bool atomic_insert(MallocSiteHashtableEntry* entry); |
79 | |
80 | void set_callsite(const MallocSite& site) { |
81 | _malloc_site = site; |
82 | } |
83 | |
84 | inline const MallocSite* peek() const { return &_malloc_site; } |
85 | inline MallocSite* data() { return &_malloc_site; } |
86 | |
87 | inline long hash() const { return _malloc_site.hash(); } |
88 | inline bool equals(const NativeCallStack& stack) const { |
89 | return _malloc_site.equals(stack); |
90 | } |
91 | // Allocation/deallocation on this allocation site |
92 | inline void allocate(size_t size) { _malloc_site.allocate(size); } |
93 | inline void deallocate(size_t size) { _malloc_site.deallocate(size); } |
94 | // Memory counters |
95 | inline size_t size() const { return _malloc_site.size(); } |
96 | inline size_t count() const { return _malloc_site.count(); } |
97 | }; |
98 | |
99 | // The walker walks every entry on MallocSiteTable |
100 | class MallocSiteWalker : public StackObj { |
101 | public: |
102 | virtual bool do_malloc_site(const MallocSite* e) { return false; } |
103 | }; |
104 | |
105 | /* |
106 | * Native memory tracking call site table. |
107 | * The table is only needed when detail tracking is enabled. |
108 | */ |
109 | class MallocSiteTable : AllStatic { |
110 | private: |
111 | // The number of hash bucket in this hashtable. The number should |
112 | // be tuned if malloc activities changed significantly. |
113 | // The statistics data can be obtained via Jcmd |
114 | // jcmd <pid> VM.native_memory statistics. |
115 | |
116 | // Currently, (number of buckets / number of entires) ratio is |
117 | // about 1 / 6 |
118 | enum { |
119 | table_base_size = 128, // The base size is calculated from statistics to give |
120 | // table ratio around 1:6 |
121 | table_size = (table_base_size * NMT_TrackingStackDepth - 1) |
122 | }; |
123 | |
124 | |
125 | // This is a very special lock, that allows multiple shared accesses (sharedLock), but |
126 | // once exclusive access (exclusiveLock) is requested, all shared accesses are |
127 | // rejected forever. |
128 | class AccessLock : public StackObj { |
129 | enum LockState { |
130 | NoLock, |
131 | SharedLock, |
132 | ExclusiveLock |
133 | }; |
134 | |
135 | private: |
136 | // A very large negative number. The only possibility to "overflow" |
137 | // this number is when there are more than -min_jint threads in |
138 | // this process, which is not going to happen in foreseeable future. |
139 | const static int _MAGIC_ = min_jint; |
140 | |
141 | LockState _lock_state; |
142 | volatile int* _lock; |
143 | public: |
144 | AccessLock(volatile int* lock) : |
145 | _lock_state(NoLock), _lock(lock) { |
146 | } |
147 | |
148 | ~AccessLock() { |
149 | if (_lock_state == SharedLock) { |
150 | Atomic::dec(_lock); |
151 | } |
152 | } |
153 | // Acquire shared lock. |
154 | // Return true if shared access is granted. |
155 | inline bool sharedLock() { |
156 | jint res = Atomic::add(1, _lock); |
157 | if (res < 0) { |
158 | Atomic::dec(_lock); |
159 | return false; |
160 | } |
161 | _lock_state = SharedLock; |
162 | return true; |
163 | } |
164 | // Acquire exclusive lock |
165 | void exclusiveLock(); |
166 | }; |
167 | |
168 | public: |
169 | static bool initialize(); |
170 | static void shutdown(); |
171 | |
172 | NOT_PRODUCT(static int access_peak_count() { return _peak_count; }) |
173 | |
174 | // Number of hash buckets |
175 | static inline int hash_buckets() { return (int)table_size; } |
176 | |
177 | // Access and copy a call stack from this table. Shared lock should be |
178 | // acquired before access the entry. |
179 | static inline bool access_stack(NativeCallStack& stack, size_t bucket_idx, |
180 | size_t pos_idx) { |
181 | AccessLock locker(&_access_count); |
182 | if (locker.sharedLock()) { |
183 | NOT_PRODUCT(_peak_count = MAX2(_peak_count, _access_count);) |
184 | MallocSite* site = malloc_site(bucket_idx, pos_idx); |
185 | if (site != NULL) { |
186 | stack = *site->call_stack(); |
187 | return true; |
188 | } |
189 | } |
190 | return false; |
191 | } |
192 | |
193 | // Record a new allocation from specified call path. |
194 | // Return true if the allocation is recorded successfully, bucket_idx |
195 | // and pos_idx are also updated to indicate the entry where the allocation |
196 | // information was recorded. |
197 | // Return false only occurs under rare scenarios: |
198 | // 1. out of memory |
199 | // 2. overflow hash bucket |
200 | static inline bool allocation_at(const NativeCallStack& stack, size_t size, |
201 | size_t* bucket_idx, size_t* pos_idx, MEMFLAGS flags) { |
202 | AccessLock locker(&_access_count); |
203 | if (locker.sharedLock()) { |
204 | NOT_PRODUCT(_peak_count = MAX2(_peak_count, _access_count);) |
205 | MallocSite* site = lookup_or_add(stack, bucket_idx, pos_idx, flags); |
206 | if (site != NULL) site->allocate(size); |
207 | return site != NULL; |
208 | } |
209 | return false; |
210 | } |
211 | |
212 | // Record memory deallocation. bucket_idx and pos_idx indicate where the allocation |
213 | // information was recorded. |
214 | static inline bool deallocation_at(size_t size, size_t bucket_idx, size_t pos_idx) { |
215 | AccessLock locker(&_access_count); |
216 | if (locker.sharedLock()) { |
217 | NOT_PRODUCT(_peak_count = MAX2(_peak_count, _access_count);) |
218 | MallocSite* site = malloc_site(bucket_idx, pos_idx); |
219 | if (site != NULL) { |
220 | site->deallocate(size); |
221 | return true; |
222 | } |
223 | } |
224 | return false; |
225 | } |
226 | |
227 | // Walk this table. |
228 | static bool walk_malloc_site(MallocSiteWalker* walker); |
229 | |
230 | private: |
231 | static MallocSiteHashtableEntry* new_entry(const NativeCallStack& key, MEMFLAGS flags); |
232 | static void reset(); |
233 | |
234 | // Delete a bucket linked list |
235 | static void delete_linked_list(MallocSiteHashtableEntry* head); |
236 | |
237 | static MallocSite* lookup_or_add(const NativeCallStack& key, size_t* bucket_idx, size_t* pos_idx, MEMFLAGS flags); |
238 | static MallocSite* malloc_site(size_t bucket_idx, size_t pos_idx); |
239 | static bool walk(MallocSiteWalker* walker); |
240 | |
241 | static inline unsigned int hash_to_index(unsigned int hash) { |
242 | return (hash % table_size); |
243 | } |
244 | |
245 | static inline const NativeCallStack* hash_entry_allocation_stack() { |
246 | assert(_hash_entry_allocation_stack != NULL, "Must be set" ); |
247 | return _hash_entry_allocation_stack; |
248 | } |
249 | |
250 | static inline const MallocSiteHashtableEntry* hash_entry_allocation_site() { |
251 | assert(_hash_entry_allocation_site != NULL, "Must be set" ); |
252 | return _hash_entry_allocation_site; |
253 | } |
254 | |
255 | private: |
256 | // Counter for counting concurrent access |
257 | static volatile int _access_count; |
258 | |
259 | // The callsite hashtable. It has to be a static table, |
260 | // since malloc call can come from C runtime linker. |
261 | static MallocSiteHashtableEntry* _table[table_size]; |
262 | static const NativeCallStack* _hash_entry_allocation_stack; |
263 | static const MallocSiteHashtableEntry* _hash_entry_allocation_site; |
264 | |
265 | |
266 | NOT_PRODUCT(static int _peak_count;) |
267 | }; |
268 | |
269 | #endif // INCLUDE_NMT |
270 | #endif // SHARE_SERVICES_MALLOCSITETABLE_HPP |
271 | |