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 struct rcu_gp {
 123         /*
 124          * Global quiescent period counter with low-order bits unused.
 125          * Using a int rather than a char to eliminate false register
 126          * dependencies causing stalls on some architectures.
 127          */
 128         unsigned long ctr;
 129
 130         int32_t futex;
 131 } __attribute__((aligned(CAA_CACHE_LINE_SIZE)));
 132
 133 extern struct rcu_gp rcu_gp;
 134
 135 struct rcu_reader {
 136         /* Data used by both reader and synchronize_rcu() */
 137         unsigned long ctr;
 138         /* Data used for registry */
 139         struct cds_list_head node __attribute__((aligned(CAA_CACHE_LINE_SIZE)));
 140         int waiting;
 141         pthread_t tid;
 142 };
 143
 144 extern DECLARE_URCU_TLS(struct rcu_reader, rcu_reader);
 145
 146 /*
 147  * Wake-up waiting synchronize_rcu(). Called from many concurrent threads.
 148  */
 149 static inline void wake_up_gp(void)
 150 {
 151         if (caa_unlikely(_CMM_LOAD_SHARED(URCU_TLS(rcu_reader).waiting))) {
 152                 _CMM_STORE_SHARED(URCU_TLS(rcu_reader).waiting, 0);
 153                 cmm_smp_mb();
 154                 if (uatomic_read(&rcu_gp.futex) != -1)
 155                         return;
 156                 uatomic_set(&rcu_gp.futex, 0);
 157                 futex_noasync(&rcu_gp.futex, FUTEX_WAKE, 1,
 158                       NULL, NULL, 0);
 159         }
 160 }
 161
 162 static inline enum rcu_state rcu_reader_state(unsigned long *ctr)
 163 {
 164         unsigned long v;
 165
 166         v = CMM_LOAD_SHARED(*ctr);
 167         if (!v)
 168                 return RCU_READER_INACTIVE;
 169         if (v == rcu_gp.ctr)
 170                 return RCU_READER_ACTIVE_CURRENT;
 171         return RCU_READER_ACTIVE_OLD;
 172 }
 173
 174 /*
 175  * Enter an RCU read-side critical section.
 176  *
 177  * This function is less than 10 lines long.  The intent is that this
 178  * function meets the 10-line criterion for LGPL, allowing this function
 179  * to be invoked directly from non-LGPL code.
 180  */
 181 static inline void _rcu_read_lock(void)
 182 {
 183         rcu_assert(URCU_TLS(rcu_reader).ctr);
 184 }
 185
 186 /*
 187  * Exit an RCU read-side critical section.
 188  *
 189  * This function is less than 10 lines long.  The intent is that this
 190  * function meets the 10-line criterion for LGPL, allowing this function
 191  * to be invoked directly from non-LGPL code.
 192  */
 193 static inline void _rcu_read_unlock(void)
 194 {
 195 }
 196
 197 /*
 198  * Returns whether within a RCU read-side critical section.
 199  *
 200  * This function is less than 10 lines long.  The intent is that this
 201  * function meets the 10-line criterion for LGPL, allowing this function
 202  * to be invoked directly from non-LGPL code.
 203  */
 204 static inline int _rcu_read_ongoing(void)
 205 {
 206         return URCU_TLS(rcu_reader).ctr;
 207 }
 208
 209 /*
 210  * This is a helper function for _rcu_quiescent_state().
 211  * The first cmm_smp_mb() ensures memory accesses in the prior read-side
 212  * critical sections are not reordered with store to
 213  * URCU_TLS(rcu_reader).ctr, and ensures that mutexes held within an
 214  * offline section that would happen to end with this
 215  * rcu_quiescent_state() call are not reordered with
 216  * store to URCU_TLS(rcu_reader).ctr.
 217  */
 218 static inline void _rcu_quiescent_state_update_and_wakeup(unsigned long gp_ctr)
 219 {
 220         cmm_smp_mb();
 221         _CMM_STORE_SHARED(URCU_TLS(rcu_reader).ctr, gp_ctr);
 222         cmm_smp_mb();   /* write URCU_TLS(rcu_reader).ctr before read futex */
 223         wake_up_gp();
 224         cmm_smp_mb();
 225 }
 226
 227 /*
 228  * Inform RCU of a quiescent state.
 229  *
 230  * This function is less than 10 lines long.  The intent is that this
 231  * function meets the 10-line criterion for LGPL, allowing this function
 232  * to be invoked directly from non-LGPL code.
 233  *
 234  * We skip the memory barriers and gp store if our local ctr already
 235  * matches the global rcu_gp.ctr value: this is OK because a prior
 236  * _rcu_quiescent_state() or _rcu_thread_online() already updated it
 237  * within our thread, so we have no quiescent state to report.
 238  */
 239 static inline void _rcu_quiescent_state(void)
 240 {
 241         unsigned long gp_ctr;
 242
 243         if ((gp_ctr = CMM_LOAD_SHARED(rcu_gp.ctr)) == URCU_TLS(rcu_reader).ctr)
 244                 return;
 245         _rcu_quiescent_state_update_and_wakeup(gp_ctr);
 246 }
 247
 248 /*
 249  * Take a thread offline, prohibiting it from entering further RCU
 250  * read-side critical sections.
 251  *
 252  * This function is less than 10 lines long.  The intent is that this
 253  * function meets the 10-line criterion for LGPL, allowing this function
 254  * to be invoked directly from non-LGPL code.
 255  */
 256 static inline void _rcu_thread_offline(void)
 257 {
 258         cmm_smp_mb();
 259         CMM_STORE_SHARED(URCU_TLS(rcu_reader).ctr, 0);
 260         cmm_smp_mb();   /* write URCU_TLS(rcu_reader).ctr before read futex */
 261         wake_up_gp();
 262         cmm_barrier();  /* Ensure the compiler does not reorder us with mutex */
 263 }
 264
 265 /*
 266  * Bring a thread online, allowing it to once again enter RCU
 267  * read-side critical sections.
 268  *
 269  * This function is less than 10 lines long.  The intent is that this
 270  * function meets the 10-line criterion for LGPL, allowing this function
 271  * to be invoked directly from non-LGPL code.
 272  */
 273 static inline void _rcu_thread_online(void)
 274 {
 275         cmm_barrier();  /* Ensure the compiler does not reorder us with mutex */
 276         _CMM_STORE_SHARED(URCU_TLS(rcu_reader).ctr, CMM_LOAD_SHARED(rcu_gp.ctr));
 277         cmm_smp_mb();
 278 }
 279
 280 #ifdef __cplusplus
 281 }
 282 #endif
 283
 284 #endif /* _URCU_QSBR_STATIC_H */