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 #ifdef PTHREAD_SETAFFINITY_IN_NP_HEADER
  39 #include <pthread_np.h>
  40 #endif
  41 #endif
  42
  43 #ifdef __HAIKU__
  44 #include <OS.h>
  45 #endif
  46
  47 #ifdef __FreeBSD__
  48 /* pthread_np.h -> sys/param.h -> machine/param.h
  49  * - defines ALIGN which clashes with our ALIGN
  50  */
  51 #undef ALIGN
  52 #define cpu_set_t cpuset_t
  53 #endif
  54
  55 static inline thrd_t u_thread_create(int (*routine)(void *), void *param)
  56 {
  57    thrd_t thread;
  58 #ifdef HAVE_PTHREAD
  59    sigset_t saved_set, new_set;
  60    int ret;
  61
  62    sigfillset(&new_set);
  63    sigdelset(&new_set, SIGSYS);
  64    pthread_sigmask(SIG_BLOCK, &new_set, &saved_set);
  65    ret = thrd_create( &thread, routine, param );
  66    pthread_sigmask(SIG_SETMASK, &saved_set, NULL);
  67 #else
  68    int ret;
  69    ret = thrd_create( &thread, routine, param );
  70 #endif
  71    if (ret)
  72       return 0;
  73
  74    return thread;
  75 }
  76
  77 static inline void u_thread_setname( const char *name )
  78 {
  79 #if defined(HAVE_PTHREAD)
  80 #if DETECT_OS_LINUX || DETECT_OS_CYGWIN || DETECT_OS_SOLARIS
  81    pthread_setname_np(pthread_self(), name);
  82 #elif DETECT_OS_FREEBSD || DETECT_OS_OPENBSD
  83    pthread_set_name_np(pthread_self(), name);
  84 #elif DETECT_OS_NETBSD
  85    pthread_setname_np(pthread_self(), "%s", (void *)name);
  86 #elif DETECT_OS_APPLE
  87    pthread_setname_np(name);
  88 #elif DETECT_OS_HAIKU
  89    rename_thread(find_thread(NULL), name);
  90 #else
  91 #warning Not sure how to call pthread_setname_np
  92 #endif
  93 #endif
  94    (void)name;
  95 }
  96
  97 /**
  98  * An AMD Zen CPU consists of multiple modules where each module has its own L3
  99  * cache. Inter-thread communication such as locks and atomics between modules
 100  * is very expensive. It's desirable to pin a group of closely cooperating
 101  * threads to one group of cores sharing L3.
 102  *
 103  * \param thread        thread
 104  * \param L3_index      index of the L3 cache
 105  * \param cores_per_L3  number of CPU cores shared by one L3
 106  */
 107 static inline void
 108 util_pin_thread_to_L3(thrd_t thread, unsigned L3_index, unsigned cores_per_L3)
 109 {
 110 #if defined(HAVE_PTHREAD_SETAFFINITY)
 111    cpu_set_t cpuset;
 112
 113    CPU_ZERO(&cpuset);
 114    for (unsigned i = 0; i < cores_per_L3; i++)
 115       CPU_SET(L3_index * cores_per_L3 + i, &cpuset);
 116    pthread_setaffinity_np(thread, sizeof(cpuset), &cpuset);
 117 #endif
 118 }
 119
 120 /**
 121  * Return the index of L3 that the thread is pinned to. If the thread is
 122  * pinned to multiple L3 caches, return -1.
 123  *
 124  * \param thread        thread
 125  * \param cores_per_L3  number of CPU cores shared by one L3
 126  */
 127 static inline int
 128 util_get_L3_for_pinned_thread(thrd_t thread, unsigned cores_per_L3)
 129 {
 130 #if defined(HAVE_PTHREAD_SETAFFINITY)
 131    cpu_set_t cpuset;
 132
 133    if (pthread_getaffinity_np(thread, sizeof(cpuset), &cpuset) == 0) {
 134       int L3_index = -1;
 135
 136       for (unsigned i = 0; i < CPU_SETSIZE; i++) {
 137          if (CPU_ISSET(i, &cpuset)) {
 138             int x = i / cores_per_L3;
 139
 140             if (L3_index != x) {
 141                if (L3_index == -1)
 142                   L3_index = x;
 143                else
 144                   return -1; /* multiple L3s are set */
 145             }
 146          }
 147       }
 148       return L3_index;
 149    }
 150 #endif
 151    return -1;
 152 }
 153
 154 /*
 155  * Thread statistics.
 156  */
 157
 158 /* Return the time of a thread's CPU time clock. */
 159 static inline int64_t
 160 u_thread_get_time_nano(thrd_t thread)
 161 {
 162 #if defined(HAVE_PTHREAD) && !defined(__APPLE__) && !defined(__HAIKU__)
 163    struct timespec ts;
 164    clockid_t cid;
 165
 166    pthread_getcpuclockid(thread, &cid);
 167    clock_gettime(cid, &ts);
 168    return (int64_t)ts.tv_sec * 1000000000 + ts.tv_nsec;
 169 #else
 170    return 0;
 171 #endif
 172 }
 173
 174 static inline bool u_thread_is_self(thrd_t thread)
 175 {
 176 #if defined(HAVE_PTHREAD)
 177    return pthread_equal(pthread_self(), thread);
 178 #endif
 179    return false;
 180 }
 181
 182 /*
 183  * util_barrier
 184  */
 185
 186 #if defined(HAVE_PTHREAD) && !defined(__APPLE__)
 187
 188 typedef pthread_barrier_t util_barrier;
 189
 190 static inline void util_barrier_init(util_barrier *barrier, unsigned count)
 191 {
 192    pthread_barrier_init(barrier, NULL, count);
 193 }
 194
 195 static inline void util_barrier_destroy(util_barrier *barrier)
 196 {
 197    pthread_barrier_destroy(barrier);
 198 }
 199
 200 static inline void util_barrier_wait(util_barrier *barrier)
 201 {
 202    pthread_barrier_wait(barrier);
 203 }
 204
 205
 206 #else /* If the OS doesn't have its own, implement barriers using a mutex and a condvar */
 207
 208 typedef struct {
 209    unsigned count;
 210    unsigned waiters;
 211    uint64_t sequence;
 212    mtx_t mutex;
 213    cnd_t condvar;
 214 } util_barrier;
 215
 216 static inline void util_barrier_init(util_barrier *barrier, unsigned count)
 217 {
 218    barrier->count = count;
 219    barrier->waiters = 0;
 220    barrier->sequence = 0;
 221    (void) mtx_init(&barrier->mutex, mtx_plain);
 222    cnd_init(&barrier->condvar);
 223 }
 224
 225 static inline void util_barrier_destroy(util_barrier *barrier)
 226 {
 227    assert(barrier->waiters == 0);
 228    mtx_destroy(&barrier->mutex);
 229    cnd_destroy(&barrier->condvar);
 230 }
 231
 232 static inline void util_barrier_wait(util_barrier *barrier)
 233 {
 234    mtx_lock(&barrier->mutex);
 235
 236    assert(barrier->waiters < barrier->count);
 237    barrier->waiters++;
 238
 239    if (barrier->waiters < barrier->count) {
 240       uint64_t sequence = barrier->sequence;
 241
 242       do {
 243          cnd_wait(&barrier->condvar, &barrier->mutex);
 244       } while (sequence == barrier->sequence);
 245    } else {
 246       barrier->waiters = 0;
 247       barrier->sequence++;
 248       cnd_broadcast(&barrier->condvar);
 249    }
 250
 251    mtx_unlock(&barrier->mutex);
 252 }
 253
 254 #endif
 255
 256 #endif /* U_THREAD_H_ */