]>
Commit | Line | Data |
---|---|---|
7c673cae | 1 | // Copyright (c) 2011-present, Facebook, Inc. All rights reserved. |
11fdf7f2 TL |
2 | // This source code is licensed under both the GPLv2 (found in the |
3 | // COPYING file in the root directory) and Apache 2.0 License | |
4 | // (found in the LICENSE.Apache file in the root directory). | |
7c673cae FG |
5 | // |
6 | // Copyright (c) 2011 The LevelDB Authors. All rights reserved. | |
7 | // Use of this source code is governed by a BSD-style license that can be | |
8 | // found in the LICENSE file. See the AUTHORS file for names of contributors. | |
9 | ||
20effc67 TL |
10 | #if !defined(OS_WIN) |
11 | ||
7c673cae FG |
12 | #include "port/port_posix.h" |
13 | ||
14 | #include <assert.h> | |
15 | #if defined(__i386__) || defined(__x86_64__) | |
16 | #include <cpuid.h> | |
17 | #endif | |
18 | #include <errno.h> | |
11fdf7f2 | 19 | #include <sched.h> |
7c673cae FG |
20 | #include <signal.h> |
21 | #include <stdio.h> | |
22 | #include <string.h> | |
7c673cae | 23 | #include <sys/resource.h> |
f67539c2 | 24 | #include <sys/time.h> |
7c673cae FG |
25 | #include <unistd.h> |
26 | #include <cstdlib> | |
7c673cae | 27 | |
f67539c2 | 28 | namespace ROCKSDB_NAMESPACE { |
494da23a TL |
29 | |
30 | // We want to give users opportunity to default all the mutexes to adaptive if | |
31 | // not specified otherwise. This enables a quick way to conduct various | |
32 | // performance related experiements. | |
33 | // | |
34 | // NB! Support for adaptive mutexes is turned on by definining | |
35 | // ROCKSDB_PTHREAD_ADAPTIVE_MUTEX during the compilation. If you use RocksDB | |
36 | // build environment then this happens automatically; otherwise it's up to the | |
37 | // consumer to define the identifier. | |
38 | #ifdef ROCKSDB_DEFAULT_TO_ADAPTIVE_MUTEX | |
39 | extern const bool kDefaultToAdaptiveMutex = true; | |
40 | #else | |
41 | extern const bool kDefaultToAdaptiveMutex = false; | |
42 | #endif | |
43 | ||
7c673cae FG |
44 | namespace port { |
45 | ||
46 | static int PthreadCall(const char* label, int result) { | |
47 | if (result != 0 && result != ETIMEDOUT) { | |
48 | fprintf(stderr, "pthread %s: %s\n", label, strerror(result)); | |
49 | abort(); | |
50 | } | |
51 | return result; | |
52 | } | |
53 | ||
54 | Mutex::Mutex(bool adaptive) { | |
11fdf7f2 | 55 | (void) adaptive; |
7c673cae FG |
56 | #ifdef ROCKSDB_PTHREAD_ADAPTIVE_MUTEX |
57 | if (!adaptive) { | |
58 | PthreadCall("init mutex", pthread_mutex_init(&mu_, nullptr)); | |
59 | } else { | |
60 | pthread_mutexattr_t mutex_attr; | |
61 | PthreadCall("init mutex attr", pthread_mutexattr_init(&mutex_attr)); | |
62 | PthreadCall("set mutex attr", | |
63 | pthread_mutexattr_settype(&mutex_attr, | |
64 | PTHREAD_MUTEX_ADAPTIVE_NP)); | |
65 | PthreadCall("init mutex", pthread_mutex_init(&mu_, &mutex_attr)); | |
66 | PthreadCall("destroy mutex attr", | |
67 | pthread_mutexattr_destroy(&mutex_attr)); | |
68 | } | |
69 | #else | |
70 | PthreadCall("init mutex", pthread_mutex_init(&mu_, nullptr)); | |
71 | #endif // ROCKSDB_PTHREAD_ADAPTIVE_MUTEX | |
72 | } | |
73 | ||
74 | Mutex::~Mutex() { PthreadCall("destroy mutex", pthread_mutex_destroy(&mu_)); } | |
75 | ||
76 | void Mutex::Lock() { | |
77 | PthreadCall("lock", pthread_mutex_lock(&mu_)); | |
78 | #ifndef NDEBUG | |
79 | locked_ = true; | |
80 | #endif | |
81 | } | |
82 | ||
83 | void Mutex::Unlock() { | |
84 | #ifndef NDEBUG | |
85 | locked_ = false; | |
86 | #endif | |
87 | PthreadCall("unlock", pthread_mutex_unlock(&mu_)); | |
88 | } | |
89 | ||
90 | void Mutex::AssertHeld() { | |
91 | #ifndef NDEBUG | |
92 | assert(locked_); | |
93 | #endif | |
94 | } | |
95 | ||
96 | CondVar::CondVar(Mutex* mu) | |
97 | : mu_(mu) { | |
98 | PthreadCall("init cv", pthread_cond_init(&cv_, nullptr)); | |
99 | } | |
100 | ||
101 | CondVar::~CondVar() { PthreadCall("destroy cv", pthread_cond_destroy(&cv_)); } | |
102 | ||
103 | void CondVar::Wait() { | |
104 | #ifndef NDEBUG | |
105 | mu_->locked_ = false; | |
106 | #endif | |
107 | PthreadCall("wait", pthread_cond_wait(&cv_, &mu_->mu_)); | |
108 | #ifndef NDEBUG | |
109 | mu_->locked_ = true; | |
110 | #endif | |
111 | } | |
112 | ||
113 | bool CondVar::TimedWait(uint64_t abs_time_us) { | |
114 | struct timespec ts; | |
115 | ts.tv_sec = static_cast<time_t>(abs_time_us / 1000000); | |
116 | ts.tv_nsec = static_cast<suseconds_t>((abs_time_us % 1000000) * 1000); | |
117 | ||
118 | #ifndef NDEBUG | |
119 | mu_->locked_ = false; | |
120 | #endif | |
121 | int err = pthread_cond_timedwait(&cv_, &mu_->mu_, &ts); | |
122 | #ifndef NDEBUG | |
123 | mu_->locked_ = true; | |
124 | #endif | |
125 | if (err == ETIMEDOUT) { | |
126 | return true; | |
127 | } | |
128 | if (err != 0) { | |
129 | PthreadCall("timedwait", err); | |
130 | } | |
131 | return false; | |
132 | } | |
133 | ||
134 | void CondVar::Signal() { | |
135 | PthreadCall("signal", pthread_cond_signal(&cv_)); | |
136 | } | |
137 | ||
138 | void CondVar::SignalAll() { | |
139 | PthreadCall("broadcast", pthread_cond_broadcast(&cv_)); | |
140 | } | |
141 | ||
142 | RWMutex::RWMutex() { | |
143 | PthreadCall("init mutex", pthread_rwlock_init(&mu_, nullptr)); | |
144 | } | |
145 | ||
146 | RWMutex::~RWMutex() { PthreadCall("destroy mutex", pthread_rwlock_destroy(&mu_)); } | |
147 | ||
148 | void RWMutex::ReadLock() { PthreadCall("read lock", pthread_rwlock_rdlock(&mu_)); } | |
149 | ||
150 | void RWMutex::WriteLock() { PthreadCall("write lock", pthread_rwlock_wrlock(&mu_)); } | |
151 | ||
152 | void RWMutex::ReadUnlock() { PthreadCall("read unlock", pthread_rwlock_unlock(&mu_)); } | |
153 | ||
154 | void RWMutex::WriteUnlock() { PthreadCall("write unlock", pthread_rwlock_unlock(&mu_)); } | |
155 | ||
156 | int PhysicalCoreID() { | |
11fdf7f2 TL |
157 | #if defined(ROCKSDB_SCHED_GETCPU_PRESENT) && defined(__x86_64__) && \ |
158 | (__GNUC__ > 2 || (__GNUC__ == 2 && __GNUC_MINOR__ >= 22)) | |
159 | // sched_getcpu uses VDSO getcpu() syscall since 2.22. I believe Linux offers VDSO | |
160 | // support only on x86_64. This is the fastest/preferred method if available. | |
161 | int cpuno = sched_getcpu(); | |
162 | if (cpuno < 0) { | |
163 | return -1; | |
164 | } | |
165 | return cpuno; | |
166 | #elif defined(__x86_64__) || defined(__i386__) | |
167 | // clang/gcc both provide cpuid.h, which defines __get_cpuid(), for x86_64 and i386. | |
7c673cae | 168 | unsigned eax, ebx = 0, ecx, edx; |
11fdf7f2 TL |
169 | if (!__get_cpuid(1, &eax, &ebx, &ecx, &edx)) { |
170 | return -1; | |
171 | } | |
7c673cae FG |
172 | return ebx >> 24; |
173 | #else | |
11fdf7f2 | 174 | // give up, the caller can generate a random number or something. |
7c673cae FG |
175 | return -1; |
176 | #endif | |
177 | } | |
178 | ||
179 | void InitOnce(OnceType* once, void (*initializer)()) { | |
180 | PthreadCall("once", pthread_once(once, initializer)); | |
181 | } | |
182 | ||
183 | void Crash(const std::string& srcfile, int srcline) { | |
184 | fprintf(stdout, "Crashing at %s:%d\n", srcfile.c_str(), srcline); | |
185 | fflush(stdout); | |
186 | kill(getpid(), SIGTERM); | |
187 | } | |
188 | ||
189 | int GetMaxOpenFiles() { | |
190 | #if defined(RLIMIT_NOFILE) | |
191 | struct rlimit no_files_limit; | |
192 | if (getrlimit(RLIMIT_NOFILE, &no_files_limit) != 0) { | |
193 | return -1; | |
194 | } | |
195 | // protect against overflow | |
f67539c2 TL |
196 | if (static_cast<uintmax_t>(no_files_limit.rlim_cur) >= |
197 | static_cast<uintmax_t>(std::numeric_limits<int>::max())) { | |
7c673cae FG |
198 | return std::numeric_limits<int>::max(); |
199 | } | |
200 | return static_cast<int>(no_files_limit.rlim_cur); | |
201 | #endif | |
202 | return -1; | |
203 | } | |
204 | ||
11fdf7f2 TL |
205 | void *cacheline_aligned_alloc(size_t size) { |
206 | #if __GNUC__ < 5 && defined(__SANITIZE_ADDRESS__) | |
207 | return malloc(size); | |
208 | #elif ( _POSIX_C_SOURCE >= 200112L || _XOPEN_SOURCE >= 600 || defined(__APPLE__)) | |
209 | void *m; | |
210 | errno = posix_memalign(&m, CACHE_LINE_SIZE, size); | |
211 | return errno ? nullptr : m; | |
212 | #else | |
213 | return malloc(size); | |
214 | #endif | |
215 | } | |
216 | ||
217 | void cacheline_aligned_free(void *memblock) { | |
218 | free(memblock); | |
219 | } | |
220 | ||
f67539c2 TL |
221 | static size_t GetPageSize() { |
222 | #if defined(OS_LINUX) || defined(_SC_PAGESIZE) | |
223 | long v = sysconf(_SC_PAGESIZE); | |
224 | if (v >= 1024) { | |
225 | return static_cast<size_t>(v); | |
226 | } | |
227 | #endif | |
228 | // Default assume 4KB | |
229 | return 4U * 1024U; | |
230 | } | |
231 | ||
232 | const size_t kPageSize = GetPageSize(); | |
11fdf7f2 | 233 | |
20effc67 TL |
234 | void SetCpuPriority(ThreadId id, CpuPriority priority) { |
235 | #ifdef OS_LINUX | |
236 | sched_param param; | |
237 | param.sched_priority = 0; | |
238 | switch (priority) { | |
239 | case CpuPriority::kHigh: | |
240 | sched_setscheduler(id, SCHED_OTHER, ¶m); | |
241 | setpriority(PRIO_PROCESS, id, -20); | |
242 | break; | |
243 | case CpuPriority::kNormal: | |
244 | sched_setscheduler(id, SCHED_OTHER, ¶m); | |
245 | setpriority(PRIO_PROCESS, id, 0); | |
246 | break; | |
247 | case CpuPriority::kLow: | |
248 | sched_setscheduler(id, SCHED_OTHER, ¶m); | |
249 | setpriority(PRIO_PROCESS, id, 19); | |
250 | break; | |
251 | case CpuPriority::kIdle: | |
252 | sched_setscheduler(id, SCHED_IDLE, ¶m); | |
253 | break; | |
254 | default: | |
255 | assert(false); | |
256 | } | |
257 | #else | |
258 | (void)id; | |
259 | (void)priority; | |
260 | #endif | |
261 | } | |
262 | ||
7c673cae | 263 | } // namespace port |
f67539c2 | 264 | } // namespace ROCKSDB_NAMESPACE |
20effc67 TL |
265 | |
266 | #endif |