1 /* 2 * Read-Copy Update mechanism for mutual exclusion 3 * 4 * This program is free software; you can redistribute it and/or modify 5 * it under the terms of the GNU General Public License as published by 6 * the Free Software Foundation; either version 2 of the License, or 7 * (at your option) any later version. 8 * 9 * This program is distributed in the hope that it will be useful, 10 * but WITHOUT ANY WARRANTY; without even the implied warranty of 11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 12 * GNU General Public License for more details. 13 * 14 * You should have received a copy of the GNU General Public License 15 * along with this program; if not, you can access it online at 16 * http://www.gnu.org/licenses/gpl-2.0.html. 17 * 18 * Copyright IBM Corporation, 2001 19 * 20 * Authors: Dipankar Sarma <dipankar@in.ibm.com> 21 * Manfred Spraul <manfred@colorfullife.com> 22 * 23 * Based on the original work by Paul McKenney <paulmck@us.ibm.com> 24 * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen. 25 * Papers: 26 * http://www.rdrop.com/users/paulmck/paper/rclockpdcsproof.pdf 27 * http://lse.sourceforge.net/locking/rclock_OLS.2001.05.01c.sc.pdf (OLS2001) 28 * 29 * For detailed explanation of Read-Copy Update mechanism see - 30 * http://lse.sourceforge.net/locking/rcupdate.html 31 * 32 */ 33 #include <linux/types.h> 34 #include <linux/kernel.h> 35 #include <linux/init.h> 36 #include <linux/spinlock.h> 37 #include <linux/smp.h> 38 #include <linux/interrupt.h> 39 #include <linux/sched.h> 40 #include <linux/atomic.h> 41 #include <linux/bitops.h> 42 #include <linux/percpu.h> 43 #include <linux/notifier.h> 44 #include <linux/cpu.h> 45 #include <linux/mutex.h> 46 #include <linux/export.h> 47 #include <linux/hardirq.h> 48 #include <linux/delay.h> 49 #include <linux/module.h> 50 51 #define CREATE_TRACE_POINTS 52 53 #include "rcu.h" 54 55 MODULE_ALIAS("rcupdate"); 56 #ifdef MODULE_PARAM_PREFIX 57 #undef MODULE_PARAM_PREFIX 58 #endif 59 #define MODULE_PARAM_PREFIX "rcupdate." 60 61 module_param(rcu_expedited, int, 0); 62 63 #ifdef CONFIG_PREEMPT_RCU 64 65 /* 66 * Preemptible RCU implementation for rcu_read_lock(). 67 * Just increment ->rcu_read_lock_nesting, shared state will be updated 68 * if we block. 69 */ 70 void __rcu_read_lock(void) 71 { 72 current->rcu_read_lock_nesting++; 73 barrier(); /* critical section after entry code. */ 74 } 75 EXPORT_SYMBOL_GPL(__rcu_read_lock); 76 77 /* 78 * Preemptible RCU implementation for rcu_read_unlock(). 79 * Decrement ->rcu_read_lock_nesting. If the result is zero (outermost 80 * rcu_read_unlock()) and ->rcu_read_unlock_special is non-zero, then 81 * invoke rcu_read_unlock_special() to clean up after a context switch 82 * in an RCU read-side critical section and other special cases. 83 */ 84 void __rcu_read_unlock(void) 85 { 86 struct task_struct *t = current; 87 88 if (t->rcu_read_lock_nesting != 1) { 89 --t->rcu_read_lock_nesting; 90 } else { 91 barrier(); /* critical section before exit code. */ 92 t->rcu_read_lock_nesting = INT_MIN; 93 #ifdef CONFIG_PROVE_RCU_DELAY 94 udelay(10); /* Make preemption more probable. */ 95 #endif /* #ifdef CONFIG_PROVE_RCU_DELAY */ 96 barrier(); /* assign before ->rcu_read_unlock_special load */ 97 if (unlikely(ACCESS_ONCE(t->rcu_read_unlock_special))) 98 rcu_read_unlock_special(t); 99 barrier(); /* ->rcu_read_unlock_special load before assign */ 100 t->rcu_read_lock_nesting = 0; 101 } 102 #ifdef CONFIG_PROVE_LOCKING 103 { 104 int rrln = ACCESS_ONCE(t->rcu_read_lock_nesting); 105 106 WARN_ON_ONCE(rrln < 0 && rrln > INT_MIN / 2); 107 } 108 #endif /* #ifdef CONFIG_PROVE_LOCKING */ 109 } 110 EXPORT_SYMBOL_GPL(__rcu_read_unlock); 111 112 #endif /* #ifdef CONFIG_PREEMPT_RCU */ 113 114 #ifdef CONFIG_DEBUG_LOCK_ALLOC 115 static struct lock_class_key rcu_lock_key; 116 struct lockdep_map rcu_lock_map = 117 STATIC_LOCKDEP_MAP_INIT("rcu_read_lock", &rcu_lock_key); 118 EXPORT_SYMBOL_GPL(rcu_lock_map); 119 120 static struct lock_class_key rcu_bh_lock_key; 121 struct lockdep_map rcu_bh_lock_map = 122 STATIC_LOCKDEP_MAP_INIT("rcu_read_lock_bh", &rcu_bh_lock_key); 123 EXPORT_SYMBOL_GPL(rcu_bh_lock_map); 124 125 static struct lock_class_key rcu_sched_lock_key; 126 struct lockdep_map rcu_sched_lock_map = 127 STATIC_LOCKDEP_MAP_INIT("rcu_read_lock_sched", &rcu_sched_lock_key); 128 EXPORT_SYMBOL_GPL(rcu_sched_lock_map); 129 130 static struct lock_class_key rcu_callback_key; 131 struct lockdep_map rcu_callback_map = 132 STATIC_LOCKDEP_MAP_INIT("rcu_callback", &rcu_callback_key); 133 EXPORT_SYMBOL_GPL(rcu_callback_map); 134 135 int notrace debug_lockdep_rcu_enabled(void) 136 { 137 return rcu_scheduler_active && debug_locks && 138 current->lockdep_recursion == 0; 139 } 140 EXPORT_SYMBOL_GPL(debug_lockdep_rcu_enabled); 141 142 /** 143 * rcu_read_lock_bh_held() - might we be in RCU-bh read-side critical section? 144 * 145 * Check for bottom half being disabled, which covers both the 146 * CONFIG_PROVE_RCU and not cases. Note that if someone uses 147 * rcu_read_lock_bh(), but then later enables BH, lockdep (if enabled) 148 * will show the situation. This is useful for debug checks in functions 149 * that require that they be called within an RCU read-side critical 150 * section. 151 * 152 * Check debug_lockdep_rcu_enabled() to prevent false positives during boot. 153 * 154 * Note that rcu_read_lock() is disallowed if the CPU is either idle or 155 * offline from an RCU perspective, so check for those as well. 156 */ 157 int rcu_read_lock_bh_held(void) 158 { 159 if (!debug_lockdep_rcu_enabled()) 160 return 1; 161 if (!rcu_is_watching()) 162 return 0; 163 if (!rcu_lockdep_current_cpu_online()) 164 return 0; 165 return in_softirq() || irqs_disabled(); 166 } 167 EXPORT_SYMBOL_GPL(rcu_read_lock_bh_held); 168 169 #endif /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */ 170 171 struct rcu_synchronize { 172 struct rcu_head head; 173 struct completion completion; 174 }; 175 176 /* 177 * Awaken the corresponding synchronize_rcu() instance now that a 178 * grace period has elapsed. 179 */ 180 static void wakeme_after_rcu(struct rcu_head *head) 181 { 182 struct rcu_synchronize *rcu; 183 184 rcu = container_of(head, struct rcu_synchronize, head); 185 complete(&rcu->completion); 186 } 187 188 void wait_rcu_gp(call_rcu_func_t crf) 189 { 190 struct rcu_synchronize rcu; 191 192 init_rcu_head_on_stack(&rcu.head); 193 init_completion(&rcu.completion); 194 /* Will wake me after RCU finished. */ 195 crf(&rcu.head, wakeme_after_rcu); 196 /* Wait for it. */ 197 wait_for_completion(&rcu.completion); 198 destroy_rcu_head_on_stack(&rcu.head); 199 } 200 EXPORT_SYMBOL_GPL(wait_rcu_gp); 201 202 #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD 203 void init_rcu_head(struct rcu_head *head) 204 { 205 debug_object_init(head, &rcuhead_debug_descr); 206 } 207 208 void destroy_rcu_head(struct rcu_head *head) 209 { 210 debug_object_free(head, &rcuhead_debug_descr); 211 } 212 213 /* 214 * fixup_activate is called when: 215 * - an active object is activated 216 * - an unknown object is activated (might be a statically initialized object) 217 * Activation is performed internally by call_rcu(). 218 */ 219 static int rcuhead_fixup_activate(void *addr, enum debug_obj_state state) 220 { 221 struct rcu_head *head = addr; 222 223 switch (state) { 224 225 case ODEBUG_STATE_NOTAVAILABLE: 226 /* 227 * This is not really a fixup. We just make sure that it is 228 * tracked in the object tracker. 229 */ 230 debug_object_init(head, &rcuhead_debug_descr); 231 debug_object_activate(head, &rcuhead_debug_descr); 232 return 0; 233 default: 234 return 1; 235 } 236 } 237 238 /** 239 * init_rcu_head_on_stack() - initialize on-stack rcu_head for debugobjects 240 * @head: pointer to rcu_head structure to be initialized 241 * 242 * This function informs debugobjects of a new rcu_head structure that 243 * has been allocated as an auto variable on the stack. This function 244 * is not required for rcu_head structures that are statically defined or 245 * that are dynamically allocated on the heap. This function has no 246 * effect for !CONFIG_DEBUG_OBJECTS_RCU_HEAD kernel builds. 247 */ 248 void init_rcu_head_on_stack(struct rcu_head *head) 249 { 250 debug_object_init_on_stack(head, &rcuhead_debug_descr); 251 } 252 EXPORT_SYMBOL_GPL(init_rcu_head_on_stack); 253 254 /** 255 * destroy_rcu_head_on_stack() - destroy on-stack rcu_head for debugobjects 256 * @head: pointer to rcu_head structure to be initialized 257 * 258 * This function informs debugobjects that an on-stack rcu_head structure 259 * is about to go out of scope. As with init_rcu_head_on_stack(), this 260 * function is not required for rcu_head structures that are statically 261 * defined or that are dynamically allocated on the heap. Also as with 262 * init_rcu_head_on_stack(), this function has no effect for 263 * !CONFIG_DEBUG_OBJECTS_RCU_HEAD kernel builds. 264 */ 265 void destroy_rcu_head_on_stack(struct rcu_head *head) 266 { 267 debug_object_free(head, &rcuhead_debug_descr); 268 } 269 EXPORT_SYMBOL_GPL(destroy_rcu_head_on_stack); 270 271 struct debug_obj_descr rcuhead_debug_descr = { 272 .name = "rcu_head", 273 .fixup_activate = rcuhead_fixup_activate, 274 }; 275 EXPORT_SYMBOL_GPL(rcuhead_debug_descr); 276 #endif /* #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD */ 277 278 #if defined(CONFIG_TREE_RCU) || defined(CONFIG_TREE_PREEMPT_RCU) || defined(CONFIG_RCU_TRACE) 279 void do_trace_rcu_torture_read(const char *rcutorturename, struct rcu_head *rhp, 280 unsigned long secs, 281 unsigned long c_old, unsigned long c) 282 { 283 trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c); 284 } 285 EXPORT_SYMBOL_GPL(do_trace_rcu_torture_read); 286 #else 287 #define do_trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c) \ 288 do { } while (0) 289 #endif 290 291 #ifdef CONFIG_RCU_STALL_COMMON 292 293 #ifdef CONFIG_PROVE_RCU 294 #define RCU_STALL_DELAY_DELTA (5 * HZ) 295 #else 296 #define RCU_STALL_DELAY_DELTA 0 297 #endif 298 299 int rcu_cpu_stall_suppress __read_mostly; /* 1 = suppress stall warnings. */ 300 static int rcu_cpu_stall_timeout __read_mostly = CONFIG_RCU_CPU_STALL_TIMEOUT; 301 302 module_param(rcu_cpu_stall_suppress, int, 0644); 303 module_param(rcu_cpu_stall_timeout, int, 0644); 304 305 int rcu_jiffies_till_stall_check(void) 306 { 307 int till_stall_check = ACCESS_ONCE(rcu_cpu_stall_timeout); 308 309 /* 310 * Limit check must be consistent with the Kconfig limits 311 * for CONFIG_RCU_CPU_STALL_TIMEOUT. 312 */ 313 if (till_stall_check < 3) { 314 ACCESS_ONCE(rcu_cpu_stall_timeout) = 3; 315 till_stall_check = 3; 316 } else if (till_stall_check > 300) { 317 ACCESS_ONCE(rcu_cpu_stall_timeout) = 300; 318 till_stall_check = 300; 319 } 320 return till_stall_check * HZ + RCU_STALL_DELAY_DELTA; 321 } 322 323 void rcu_sysrq_start(void) 324 { 325 if (!rcu_cpu_stall_suppress) 326 rcu_cpu_stall_suppress = 2; 327 } 328 329 void rcu_sysrq_end(void) 330 { 331 if (rcu_cpu_stall_suppress == 2) 332 rcu_cpu_stall_suppress = 0; 333 } 334 335 static int rcu_panic(struct notifier_block *this, unsigned long ev, void *ptr) 336 { 337 rcu_cpu_stall_suppress = 1; 338 return NOTIFY_DONE; 339 } 340 341 static struct notifier_block rcu_panic_block = { 342 .notifier_call = rcu_panic, 343 }; 344 345 static int __init check_cpu_stall_init(void) 346 { 347 atomic_notifier_chain_register(&panic_notifier_list, &rcu_panic_block); 348 return 0; 349 } 350 early_initcall(check_cpu_stall_init); 351 352 #endif /* #ifdef CONFIG_RCU_STALL_COMMON */ 353 354 /* 355 * Hooks for cond_resched() and friends to avoid RCU CPU stall warnings. 356 */ 357 358 DEFINE_PER_CPU(int, rcu_cond_resched_count); 359 360 /* 361 * Report a set of RCU quiescent states, for use by cond_resched() 362 * and friends. Out of line due to being called infrequently. 363 */ 364 void rcu_resched(void) 365 { 366 preempt_disable(); 367 __this_cpu_write(rcu_cond_resched_count, 0); 368 rcu_note_context_switch(smp_processor_id()); 369 preempt_enable(); 370 } 371