urcu/static/urcu-qsbr.h

   1 #ifndef _URCU_QSBR_STATIC_H
   2 #define _URCU_QSBR_STATIC_H
   3
   4 /*
   5  * urcu-qsbr-static.h
   6  *
   7  * Userspace RCU QSBR header.
   8  *
   9  * TO BE INCLUDED ONLY IN CODE THAT IS TO BE RECOMPILED ON EACH LIBURCU
  10  * RELEASE. See urcu.h for linking dynamically with the userspace rcu library.
  11  *
  12  * Copyright (c) 2009 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
  13  * Copyright (c) 2009 Paul E. McKenney, IBM Corporation.
  14  *
  15  * This library is free software; you can redistribute it and/or
  16  * modify it under the terms of the GNU Lesser General Public
  17  * License as published by the Free Software Foundation; either
  18  * version 2.1 of the License, or (at your option) any later version.
  19  *
  20  * This library is distributed in the hope that it will be useful,
  21  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  22  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  23  * Lesser General Public License for more details.
  24  *
  25  * You should have received a copy of the GNU Lesser General Public
  26  * License along with this library; if not, write to the Free Software
  27  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  28  *
  29  * IBM's contributions to this file may be relicensed under LGPLv2 or later.
  30  */
  31
  32 #include <stdlib.h>
  33 #include <pthread.h>
  34 #include <assert.h>
  35 #include <limits.h>
  36 #include <unistd.h>
  37 #include <stdint.h>
  38
  39 #include <urcu/compiler.h>
  40 #include <urcu/arch.h>
  41 #include <urcu/system.h>
  42 #include <urcu/uatomic.h>
  43 #include <urcu/list.h>
  44 #include <urcu/futex.h>
  45 #include <urcu/tls-compat.h>
  46
  47 #ifdef __cplusplus
  48 extern "C" {
  49 #endif
  50
  51 /*
  52  * This code section can only be included in LGPL 2.1 compatible source code.
  53  * See below for the function call wrappers which can be used in code meant to
  54  * be only linked with the Userspace RCU library. This comes with a small
  55  * performance degradation on the read-side due to the added function calls.
  56  * This is required to permit relinking with newer versions of the library.
  57  */
  58
  59 #ifdef DEBUG_RCU
  60 #define rcu_assert(args...)     assert(args)
  61 #else
  62 #define rcu_assert(args...)
  63 #endif
  64
  65 enum rcu_state {
  66         RCU_READER_ACTIVE_CURRENT,
  67         RCU_READER_ACTIVE_OLD,
  68         RCU_READER_INACTIVE,
  69 };
  70
  71 #ifdef DEBUG_YIELD
  72 #include <sched.h>
  73 #include <time.h>
  74 #include <pthread.h>
  75 #include <unistd.h>
  76
  77 #define RCU_YIELD_READ  (1 << 0)
  78 #define RCU_YIELD_WRITE (1 << 1)
  79
  80 /* maximum sleep delay, in us */
  81 #define MAX_SLEEP 50
  82
  83 extern unsigned int rcu_yield_active;
  84 extern DECLARE_URCU_TLS(unsigned int, rcu_rand_yield);
  85
  86 static inline void rcu_debug_yield_read(void)
  87 {
  88         if (rcu_yield_active & RCU_YIELD_READ)
  89                 if (rand_r(&URCU_TLS(rcu_rand_yield)) & 0x1)
  90                         usleep(rand_r(&URCU_TLS(rcu_rand_yield)) % MAX_SLEEP);
  91 }
  92
  93 static inline void rcu_debug_yield_write(void)
  94 {
  95         if (rcu_yield_active & RCU_YIELD_WRITE)
  96                 if (rand_r(&URCU_TLS(rcu_rand_yield)) & 0x1)
  97                         usleep(rand_r(&URCU_TLS(rcu_rand_yield)) % MAX_SLEEP);
  98 }
  99
 100 static inline void rcu_debug_yield_init(void)
 101 {
 102         URCU_TLS(rcu_rand_yield) = time(NULL) ^ (unsigned long) pthread_self();
 103 }
 104 #else
 105 static inline void rcu_debug_yield_read(void)
 106 {
 107 }
 108
 109 static inline void rcu_debug_yield_write(void)
 110 {
 111 }
 112
 113 static inline void rcu_debug_yield_init(void)
 114 {
 115
 116 }
 117 #endif
 118
 119 #define RCU_GP_ONLINE           (1UL << 0)
 120 #define RCU_GP_CTR              (1UL << 1)
 121
 122 /*
 123  * Global quiescent period counter with low-order bits unused.
 124  * Using a int rather than a char to eliminate false register dependencies
 125  * causing stalls on some architectures.
 126  */
 127 extern unsigned long rcu_gp_ctr;
 128
 129 struct rcu_reader {
 130         /* Data used by both reader and synchronize_rcu() */
 131         unsigned long ctr;
 132         /* Data used for registry */
 133         struct cds_list_head node __attribute__((aligned(CAA_CACHE_LINE_SIZE)));
 134         int waiting;
 135         pthread_t tid;
 136 };
 137
 138 extern DECLARE_URCU_TLS(struct rcu_reader, rcu_reader);
 139
 140 extern int32_t rcu_gp_futex;
 141
 142 /*
 143  * Wake-up waiting synchronize_rcu(). Called from many concurrent threads.
 144  */
 145 static inline void wake_up_gp(void)
 146 {
 147         if (caa_unlikely(_CMM_LOAD_SHARED(URCU_TLS(rcu_reader).waiting))) {
 148                 _CMM_STORE_SHARED(URCU_TLS(rcu_reader).waiting, 0);
 149                 cmm_smp_mb();
 150                 if (uatomic_read(&rcu_gp_futex) != -1)
 151                         return;
 152                 uatomic_set(&rcu_gp_futex, 0);
 153                 futex_noasync(&rcu_gp_futex, FUTEX_WAKE, 1,
 154                       NULL, NULL, 0);
 155         }
 156 }
 157
 158 static inline enum rcu_state rcu_reader_state(unsigned long *ctr)
 159 {
 160         unsigned long v;
 161
 162         v = CMM_LOAD_SHARED(*ctr);
 163         if (!v)
 164                 return RCU_READER_INACTIVE;
 165         if (v == rcu_gp_ctr)
 166                 return RCU_READER_ACTIVE_CURRENT;
 167         return RCU_READER_ACTIVE_OLD;
 168 }
 169
 170 /*
 171  * Enter an RCU read-side critical section.
 172  *
 173  * This function is less than 10 lines long.  The intent is that this
 174  * function meets the 10-line criterion for LGPL, allowing this function
 175  * to be invoked directly from non-LGPL code.
 176  */
 177 static inline void _rcu_read_lock(void)
 178 {
 179         rcu_assert(URCU_TLS(rcu_reader).ctr);
 180 }
 181
 182 /*
 183  * Exit an RCU read-side critical section.
 184  *
 185  * This function is less than 10 lines long.  The intent is that this
 186  * function meets the 10-line criterion for LGPL, allowing this function
 187  * to be invoked directly from non-LGPL code.
 188  */
 189 static inline void _rcu_read_unlock(void)
 190 {
 191 }
 192
 193 /*
 194  * This is a helper function for _rcu_quiescent_state().
 195  * The first cmm_smp_mb() ensures memory accesses in the prior read-side
 196  * critical sections are not reordered with store to
 197  * URCU_TLS(rcu_reader).ctr, and ensures that mutexes held within an
 198  * offline section that would happen to end with this
 199  * rcu_quiescent_state() call are not reordered with
 200  * store to URCU_TLS(rcu_reader).ctr.
 201  */
 202 static inline void _rcu_quiescent_state_update_and_wakeup(unsigned long gp_ctr)
 203 {
 204         cmm_smp_mb();
 205         _CMM_STORE_SHARED(URCU_TLS(rcu_reader).ctr, gp_ctr);
 206         cmm_smp_mb();   /* write URCU_TLS(rcu_reader).ctr before read futex */
 207         wake_up_gp();
 208         cmm_smp_mb();
 209 }
 210
 211 /*
 212  * Inform RCU of a quiescent state.
 213  *
 214  * This function is less than 10 lines long.  The intent is that this
 215  * function meets the 10-line criterion for LGPL, allowing this function
 216  * to be invoked directly from non-LGPL code.
 217  *
 218  * We skip the memory barriers and gp store if our local ctr already
 219  * matches the global rcu_gp_ctr value: this is OK because a prior
 220  * _rcu_quiescent_state() or _rcu_thread_online() already updated it
 221  * within our thread, so we have no quiescent state to report.
 222  */
 223 static inline void _rcu_quiescent_state(void)
 224 {
 225         unsigned long gp_ctr;
 226
 227         if ((gp_ctr = CMM_LOAD_SHARED(rcu_gp_ctr)) == URCU_TLS(rcu_reader).ctr)
 228                 return;
 229         _rcu_quiescent_state_update_and_wakeup(gp_ctr);
 230 }
 231
 232 /*
 233  * Take a thread offline, prohibiting it from entering further RCU
 234  * read-side critical sections.
 235  *
 236  * This function is less than 10 lines long.  The intent is that this
 237  * function meets the 10-line criterion for LGPL, allowing this function
 238  * to be invoked directly from non-LGPL code.
 239  */
 240 static inline void _rcu_thread_offline(void)
 241 {
 242         cmm_smp_mb();
 243         CMM_STORE_SHARED(URCU_TLS(rcu_reader).ctr, 0);
 244         cmm_smp_mb();   /* write URCU_TLS(rcu_reader).ctr before read futex */
 245         wake_up_gp();
 246         cmm_barrier();  /* Ensure the compiler does not reorder us with mutex */
 247 }
 248
 249 /*
 250  * Bring a thread online, allowing it to once again enter RCU
 251  * read-side critical sections.
 252  *
 253  * This function is less than 10 lines long.  The intent is that this
 254  * function meets the 10-line criterion for LGPL, allowing this function
 255  * to be invoked directly from non-LGPL code.
 256  */
 257 static inline void _rcu_thread_online(void)
 258 {
 259         cmm_barrier();  /* Ensure the compiler does not reorder us with mutex */
 260         _CMM_STORE_SHARED(URCU_TLS(rcu_reader).ctr, CMM_LOAD_SHARED(rcu_gp_ctr));
 261         cmm_smp_mb();
 262 }
 263
 264 #ifdef __cplusplus
 265 }
 266 #endif
 267
 268 #endif /* _URCU_QSBR_STATIC_H */