src/util/u_thread.h

   1 /**************************************************************************
   2  *
   3  * Copyright 1999-2006 Brian Paul
   4  * Copyright 2008 VMware, Inc.
   5  * All Rights Reserved.
   6  *
   7  * Permission is hereby granted, free of charge, to any person obtaining a
   8  * copy of this software and associated documentation files (the "Software"),
   9  * to deal in the Software without restriction, including without limitation
  10  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
  11  * and/or sell copies of the Software, and to permit persons to whom the
  12  * Software is furnished to do so, subject to the following conditions:
  13  *
  14  * The above copyright notice and this permission notice shall be included
  15  * in all copies or substantial portions of the Software.
  16  *
  17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  18  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  19  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  20  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
  21  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
  22  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
  23  * OTHER DEALINGS IN THE SOFTWARE.
  24  *
  25  **************************************************************************/
  26
  27 #ifndef U_THREAD_H_
  28 #define U_THREAD_H_
  29
  30 #include <stdint.h>
  31 #include <stdbool.h>
  32
  33 #include "c11/threads.h"
  34 #include "detect_os.h"
  35
  36 #ifdef HAVE_PTHREAD
  37 #include <signal.h>
  38 #endif
  39
  40 static inline thrd_t u_thread_create(int (*routine)(void *), void *param)
  41 {
  42    thrd_t thread;
  43 #ifdef HAVE_PTHREAD
  44    sigset_t saved_set, new_set;
  45    int ret;
  46
  47    sigfillset(&new_set);
  48    sigdelset(&new_set, SIGSYS);
  49    pthread_sigmask(SIG_BLOCK, &new_set, &saved_set);
  50    ret = thrd_create( &thread, routine, param );
  51    pthread_sigmask(SIG_SETMASK, &saved_set, NULL);
  52 #else
  53    int ret;
  54    ret = thrd_create( &thread, routine, param );
  55 #endif
  56    if (ret)
  57       return 0;
  58
  59    return thread;
  60 }
  61
  62 static inline void u_thread_setname( const char *name )
  63 {
  64 #if defined(HAVE_PTHREAD)
  65 #if DETECT_OS_LINUX
  66    pthread_setname_np(pthread_self(), name);
  67 #elif DETECT_OS_FREEBSD || DETECT_OS_OPENBSD
  68    pthread_set_name_np(pthread_self(), name);
  69 #elif DETECT_OS_NETBSD
  70    pthread_setname_np(pthread_self(), "%s", name);
  71 #elif DETECT_OS_APPLE
  72    pthread_setname_np(name);
  73 #else
  74 #error Not sure how to call pthread_setname_np
  75 #endif
  76 #endif
  77    (void)name;
  78 }
  79
  80 /**
  81  * An AMD Zen CPU consists of multiple modules where each module has its own L3
  82  * cache. Inter-thread communication such as locks and atomics between modules
  83  * is very expensive. It's desirable to pin a group of closely cooperating
  84  * threads to one group of cores sharing L3.
  85  *
  86  * \param thread        thread
  87  * \param L3_index      index of the L3 cache
  88  * \param cores_per_L3  number of CPU cores shared by one L3
  89  */
  90 static inline void
  91 util_pin_thread_to_L3(thrd_t thread, unsigned L3_index, unsigned cores_per_L3)
  92 {
  93 #if defined(HAVE_PTHREAD_SETAFFINITY)
  94    cpu_set_t cpuset;
  95
  96    CPU_ZERO(&cpuset);
  97    for (unsigned i = 0; i < cores_per_L3; i++)
  98       CPU_SET(L3_index * cores_per_L3 + i, &cpuset);
  99    pthread_setaffinity_np(thread, sizeof(cpuset), &cpuset);
 100 #endif
 101 }
 102
 103 /**
 104  * Return the index of L3 that the thread is pinned to. If the thread is
 105  * pinned to multiple L3 caches, return -1.
 106  *
 107  * \param thread        thread
 108  * \param cores_per_L3  number of CPU cores shared by one L3
 109  */
 110 static inline int
 111 util_get_L3_for_pinned_thread(thrd_t thread, unsigned cores_per_L3)
 112 {
 113 #if defined(HAVE_PTHREAD_SETAFFINITY)
 114    cpu_set_t cpuset;
 115
 116    if (pthread_getaffinity_np(thread, sizeof(cpuset), &cpuset) == 0) {
 117       int L3_index = -1;
 118
 119       for (unsigned i = 0; i < CPU_SETSIZE; i++) {
 120          if (CPU_ISSET(i, &cpuset)) {
 121             int x = i / cores_per_L3;
 122
 123             if (L3_index != x) {
 124                if (L3_index == -1)
 125                   L3_index = x;
 126                else
 127                   return -1; /* multiple L3s are set */
 128             }
 129          }
 130       }
 131       return L3_index;
 132    }
 133 #endif
 134    return -1;
 135 }
 136
 137 /*
 138  * Thread statistics.
 139  */
 140
 141 /* Return the time of a thread's CPU time clock. */
 142 static inline int64_t
 143 u_thread_get_time_nano(thrd_t thread)
 144 {
 145 #if defined(__linux__) && defined(HAVE_PTHREAD)
 146    struct timespec ts;
 147    clockid_t cid;
 148
 149    pthread_getcpuclockid(thread, &cid);
 150    clock_gettime(cid, &ts);
 151    return (int64_t)ts.tv_sec * 1000000000 + ts.tv_nsec;
 152 #else
 153    return 0;
 154 #endif
 155 }
 156
 157 static inline bool u_thread_is_self(thrd_t thread)
 158 {
 159 #if defined(HAVE_PTHREAD)
 160    return pthread_equal(pthread_self(), thread);
 161 #endif
 162    return false;
 163 }
 164
 165 /*
 166  * util_barrier
 167  */
 168
 169 #if defined(HAVE_PTHREAD) && !defined(__APPLE__)
 170
 171 typedef pthread_barrier_t util_barrier;
 172
 173 static inline void util_barrier_init(util_barrier *barrier, unsigned count)
 174 {
 175    pthread_barrier_init(barrier, NULL, count);
 176 }
 177
 178 static inline void util_barrier_destroy(util_barrier *barrier)
 179 {
 180    pthread_barrier_destroy(barrier);
 181 }
 182
 183 static inline void util_barrier_wait(util_barrier *barrier)
 184 {
 185    pthread_barrier_wait(barrier);
 186 }
 187
 188
 189 #else /* If the OS doesn't have its own, implement barriers using a mutex and a condvar */
 190
 191 typedef struct {
 192    unsigned count;
 193    unsigned waiters;
 194    uint64_t sequence;
 195    mtx_t mutex;
 196    cnd_t condvar;
 197 } util_barrier;
 198
 199 static inline void util_barrier_init(util_barrier *barrier, unsigned count)
 200 {
 201    barrier->count = count;
 202    barrier->waiters = 0;
 203    barrier->sequence = 0;
 204    (void) mtx_init(&barrier->mutex, mtx_plain);
 205    cnd_init(&barrier->condvar);
 206 }
 207
 208 static inline void util_barrier_destroy(util_barrier *barrier)
 209 {
 210    assert(barrier->waiters == 0);
 211    mtx_destroy(&barrier->mutex);
 212    cnd_destroy(&barrier->condvar);
 213 }
 214
 215 static inline void util_barrier_wait(util_barrier *barrier)
 216 {
 217    mtx_lock(&barrier->mutex);
 218
 219    assert(barrier->waiters < barrier->count);
 220    barrier->waiters++;
 221
 222    if (barrier->waiters < barrier->count) {
 223       uint64_t sequence = barrier->sequence;
 224
 225       do {
 226          cnd_wait(&barrier->condvar, &barrier->mutex);
 227       } while (sequence == barrier->sequence);
 228    } else {
 229       barrier->waiters = 0;
 230       barrier->sequence++;
 231       cnd_broadcast(&barrier->condvar);
 232    }
 233
 234    mtx_unlock(&barrier->mutex);
 235 }
 236
 237 #endif
 238
 239 #endif /* U_THREAD_H_ */