| 1 | // SPDX-License-Identifier: GPL-2.0+ | 
|---|
| 2 | /* | 
|---|
| 3 | * Read-Copy Update mechanism for mutual exclusion | 
|---|
| 4 | * | 
|---|
| 5 | * Copyright IBM Corporation, 2001 | 
|---|
| 6 | * | 
|---|
| 7 | * Authors: Dipankar Sarma <dipankar@in.ibm.com> | 
|---|
| 8 | *	    Manfred Spraul <manfred@colorfullife.com> | 
|---|
| 9 | * | 
|---|
| 10 | * Based on the original work by Paul McKenney <paulmck@linux.ibm.com> | 
|---|
| 11 | * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen. | 
|---|
| 12 | * Papers: | 
|---|
| 13 | * http://www.rdrop.com/users/paulmck/paper/rclockpdcsproof.pdf | 
|---|
| 14 | * http://lse.sourceforge.net/locking/rclock_OLS.2001.05.01c.sc.pdf (OLS2001) | 
|---|
| 15 | * | 
|---|
| 16 | * For detailed explanation of Read-Copy Update mechanism see - | 
|---|
| 17 | *		http://lse.sourceforge.net/locking/rcupdate.html | 
|---|
| 18 | * | 
|---|
| 19 | */ | 
|---|
| 20 | #include <linux/types.h> | 
|---|
| 21 | #include <linux/kernel.h> | 
|---|
| 22 | #include <linux/init.h> | 
|---|
| 23 | #include <linux/spinlock.h> | 
|---|
| 24 | #include <linux/smp.h> | 
|---|
| 25 | #include <linux/interrupt.h> | 
|---|
| 26 | #include <linux/sched/signal.h> | 
|---|
| 27 | #include <linux/sched/debug.h> | 
|---|
| 28 | #include <linux/torture.h> | 
|---|
| 29 | #include <linux/atomic.h> | 
|---|
| 30 | #include <linux/bitops.h> | 
|---|
| 31 | #include <linux/percpu.h> | 
|---|
| 32 | #include <linux/notifier.h> | 
|---|
| 33 | #include <linux/cpu.h> | 
|---|
| 34 | #include <linux/mutex.h> | 
|---|
| 35 | #include <linux/export.h> | 
|---|
| 36 | #include <linux/hardirq.h> | 
|---|
| 37 | #include <linux/delay.h> | 
|---|
| 38 | #include <linux/moduleparam.h> | 
|---|
| 39 | #include <linux/kthread.h> | 
|---|
| 40 | #include <linux/tick.h> | 
|---|
| 41 | #include <linux/rcupdate_wait.h> | 
|---|
| 42 | #include <linux/sched/isolation.h> | 
|---|
| 43 | #include <linux/kprobes.h> | 
|---|
| 44 | #include <linux/slab.h> | 
|---|
| 45 | #include <linux/irq_work.h> | 
|---|
| 46 | #include <linux/rcupdate_trace.h> | 
|---|
| 47 |  | 
|---|
| 48 | #define CREATE_TRACE_POINTS | 
|---|
| 49 |  | 
|---|
| 50 | #include "rcu.h" | 
|---|
| 51 |  | 
|---|
| 52 | #ifdef MODULE_PARAM_PREFIX | 
|---|
| 53 | #undef MODULE_PARAM_PREFIX | 
|---|
| 54 | #endif | 
|---|
| 55 | #define MODULE_PARAM_PREFIX "rcupdate." | 
|---|
| 56 |  | 
|---|
| 57 | #ifndef CONFIG_TINY_RCU | 
|---|
| 58 | module_param(rcu_expedited, int, 0444); | 
|---|
| 59 | module_param(rcu_normal, int, 0444); | 
|---|
| 60 | static int rcu_normal_after_boot = IS_ENABLED(CONFIG_PREEMPT_RT); | 
|---|
| 61 | #if !defined(CONFIG_PREEMPT_RT) || defined(CONFIG_NO_HZ_FULL) | 
|---|
| 62 | module_param(rcu_normal_after_boot, int, 0444); | 
|---|
| 63 | #endif | 
|---|
| 64 | #endif /* #ifndef CONFIG_TINY_RCU */ | 
|---|
| 65 |  | 
|---|
| 66 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | 
|---|
| 67 | /** | 
|---|
| 68 | * rcu_read_lock_held_common() - might we be in RCU-sched read-side critical section? | 
|---|
| 69 | * @ret:	Best guess answer if lockdep cannot be relied on | 
|---|
| 70 | * | 
|---|
| 71 | * Returns true if lockdep must be ignored, in which case ``*ret`` contains | 
|---|
| 72 | * the best guess described below.  Otherwise returns false, in which | 
|---|
| 73 | * case ``*ret`` tells the caller nothing and the caller should instead | 
|---|
| 74 | * consult lockdep. | 
|---|
| 75 | * | 
|---|
| 76 | * If CONFIG_DEBUG_LOCK_ALLOC is selected, set ``*ret`` to nonzero iff in an | 
|---|
| 77 | * RCU-sched read-side critical section.  In absence of | 
|---|
| 78 | * CONFIG_DEBUG_LOCK_ALLOC, this assumes we are in an RCU-sched read-side | 
|---|
| 79 | * critical section unless it can prove otherwise.  Note that disabling | 
|---|
| 80 | * of preemption (including disabling irqs) counts as an RCU-sched | 
|---|
| 81 | * read-side critical section.  This is useful for debug checks in functions | 
|---|
| 82 | * that required that they be called within an RCU-sched read-side | 
|---|
| 83 | * critical section. | 
|---|
| 84 | * | 
|---|
| 85 | * Check debug_lockdep_rcu_enabled() to prevent false positives during boot | 
|---|
| 86 | * and while lockdep is disabled. | 
|---|
| 87 | * | 
|---|
| 88 | * Note that if the CPU is in the idle loop from an RCU point of view (ie: | 
|---|
| 89 | * that we are in the section between ct_idle_enter() and ct_idle_exit()) | 
|---|
| 90 | * then rcu_read_lock_held() sets ``*ret`` to false even if the CPU did an | 
|---|
| 91 | * rcu_read_lock().  The reason for this is that RCU ignores CPUs that are | 
|---|
| 92 | * in such a section, considering these as in extended quiescent state, | 
|---|
| 93 | * so such a CPU is effectively never in an RCU read-side critical section | 
|---|
| 94 | * regardless of what RCU primitives it invokes.  This state of affairs is | 
|---|
| 95 | * required --- we need to keep an RCU-free window in idle where the CPU may | 
|---|
| 96 | * possibly enter into low power mode. This way we can notice an extended | 
|---|
| 97 | * quiescent state to other CPUs that started a grace period. Otherwise | 
|---|
| 98 | * we would delay any grace period as long as we run in the idle task. | 
|---|
| 99 | * | 
|---|
| 100 | * Similarly, we avoid claiming an RCU read lock held if the current | 
|---|
| 101 | * CPU is offline. | 
|---|
| 102 | */ | 
|---|
| 103 | static bool rcu_read_lock_held_common(bool *ret) | 
|---|
| 104 | { | 
|---|
| 105 | if (!debug_lockdep_rcu_enabled()) { | 
|---|
| 106 | *ret = true; | 
|---|
| 107 | return true; | 
|---|
| 108 | } | 
|---|
| 109 | if (!rcu_is_watching()) { | 
|---|
| 110 | *ret = false; | 
|---|
| 111 | return true; | 
|---|
| 112 | } | 
|---|
| 113 | if (!rcu_lockdep_current_cpu_online()) { | 
|---|
| 114 | *ret = false; | 
|---|
| 115 | return true; | 
|---|
| 116 | } | 
|---|
| 117 | return false; | 
|---|
| 118 | } | 
|---|
| 119 |  | 
|---|
| 120 | int rcu_read_lock_sched_held(void) | 
|---|
| 121 | { | 
|---|
| 122 | bool ret; | 
|---|
| 123 |  | 
|---|
| 124 | if (rcu_read_lock_held_common(&ret)) | 
|---|
| 125 | return ret; | 
|---|
| 126 | return lock_is_held(&rcu_sched_lock_map) || !preemptible(); | 
|---|
| 127 | } | 
|---|
| 128 | EXPORT_SYMBOL(rcu_read_lock_sched_held); | 
|---|
| 129 | #endif | 
|---|
| 130 |  | 
|---|
| 131 | #ifndef CONFIG_TINY_RCU | 
|---|
| 132 |  | 
|---|
| 133 | /* | 
|---|
| 134 | * Should expedited grace-period primitives always fall back to their | 
|---|
| 135 | * non-expedited counterparts?  Intended for use within RCU.  Note | 
|---|
| 136 | * that if the user specifies both rcu_expedited and rcu_normal, then | 
|---|
| 137 | * rcu_normal wins.  (Except during the time period during boot from | 
|---|
| 138 | * when the first task is spawned until the rcu_set_runtime_mode() | 
|---|
| 139 | * core_initcall() is invoked, at which point everything is expedited.) | 
|---|
| 140 | */ | 
|---|
| 141 | bool rcu_gp_is_normal(void) | 
|---|
| 142 | { | 
|---|
| 143 | return READ_ONCE(rcu_normal) && | 
|---|
| 144 | rcu_scheduler_active != RCU_SCHEDULER_INIT; | 
|---|
| 145 | } | 
|---|
| 146 | EXPORT_SYMBOL_GPL(rcu_gp_is_normal); | 
|---|
| 147 |  | 
|---|
| 148 | static atomic_t rcu_async_hurry_nesting = ATOMIC_INIT(1); | 
|---|
| 149 | /* | 
|---|
| 150 | * Should call_rcu() callbacks be processed with urgency or are | 
|---|
| 151 | * they OK being executed with arbitrary delays? | 
|---|
| 152 | */ | 
|---|
| 153 | bool rcu_async_should_hurry(void) | 
|---|
| 154 | { | 
|---|
| 155 | return !IS_ENABLED(CONFIG_RCU_LAZY) || | 
|---|
| 156 | atomic_read(v: &rcu_async_hurry_nesting); | 
|---|
| 157 | } | 
|---|
| 158 | EXPORT_SYMBOL_GPL(rcu_async_should_hurry); | 
|---|
| 159 |  | 
|---|
| 160 | /** | 
|---|
| 161 | * rcu_async_hurry - Make future async RCU callbacks not lazy. | 
|---|
| 162 | * | 
|---|
| 163 | * After a call to this function, future calls to call_rcu() | 
|---|
| 164 | * will be processed in a timely fashion. | 
|---|
| 165 | */ | 
|---|
| 166 | void rcu_async_hurry(void) | 
|---|
| 167 | { | 
|---|
| 168 | if (IS_ENABLED(CONFIG_RCU_LAZY)) | 
|---|
| 169 | atomic_inc(v: &rcu_async_hurry_nesting); | 
|---|
| 170 | } | 
|---|
| 171 | EXPORT_SYMBOL_GPL(rcu_async_hurry); | 
|---|
| 172 |  | 
|---|
| 173 | /** | 
|---|
| 174 | * rcu_async_relax - Make future async RCU callbacks lazy. | 
|---|
| 175 | * | 
|---|
| 176 | * After a call to this function, future calls to call_rcu() | 
|---|
| 177 | * will be processed in a lazy fashion. | 
|---|
| 178 | */ | 
|---|
| 179 | void rcu_async_relax(void) | 
|---|
| 180 | { | 
|---|
| 181 | if (IS_ENABLED(CONFIG_RCU_LAZY)) | 
|---|
| 182 | atomic_dec(v: &rcu_async_hurry_nesting); | 
|---|
| 183 | } | 
|---|
| 184 | EXPORT_SYMBOL_GPL(rcu_async_relax); | 
|---|
| 185 |  | 
|---|
| 186 | static atomic_t rcu_expedited_nesting = ATOMIC_INIT(1); | 
|---|
| 187 | /* | 
|---|
| 188 | * Should normal grace-period primitives be expedited?  Intended for | 
|---|
| 189 | * use within RCU.  Note that this function takes the rcu_expedited | 
|---|
| 190 | * sysfs/boot variable and rcu_scheduler_active into account as well | 
|---|
| 191 | * as the rcu_expedite_gp() nesting.  So looping on rcu_unexpedite_gp() | 
|---|
| 192 | * until rcu_gp_is_expedited() returns false is a -really- bad idea. | 
|---|
| 193 | */ | 
|---|
| 194 | bool rcu_gp_is_expedited(void) | 
|---|
| 195 | { | 
|---|
| 196 | return rcu_expedited || atomic_read(v: &rcu_expedited_nesting); | 
|---|
| 197 | } | 
|---|
| 198 | EXPORT_SYMBOL_GPL(rcu_gp_is_expedited); | 
|---|
| 199 |  | 
|---|
| 200 | /** | 
|---|
| 201 | * rcu_expedite_gp - Expedite future RCU grace periods | 
|---|
| 202 | * | 
|---|
| 203 | * After a call to this function, future calls to synchronize_rcu() and | 
|---|
| 204 | * friends act as the corresponding synchronize_rcu_expedited() function | 
|---|
| 205 | * had instead been called. | 
|---|
| 206 | */ | 
|---|
| 207 | void rcu_expedite_gp(void) | 
|---|
| 208 | { | 
|---|
| 209 | atomic_inc(v: &rcu_expedited_nesting); | 
|---|
| 210 | } | 
|---|
| 211 | EXPORT_SYMBOL_GPL(rcu_expedite_gp); | 
|---|
| 212 |  | 
|---|
| 213 | /** | 
|---|
| 214 | * rcu_unexpedite_gp - Cancel prior rcu_expedite_gp() invocation | 
|---|
| 215 | * | 
|---|
| 216 | * Undo a prior call to rcu_expedite_gp().  If all prior calls to | 
|---|
| 217 | * rcu_expedite_gp() are undone by a subsequent call to rcu_unexpedite_gp(), | 
|---|
| 218 | * and if the rcu_expedited sysfs/boot parameter is not set, then all | 
|---|
| 219 | * subsequent calls to synchronize_rcu() and friends will return to | 
|---|
| 220 | * their normal non-expedited behavior. | 
|---|
| 221 | */ | 
|---|
| 222 | void rcu_unexpedite_gp(void) | 
|---|
| 223 | { | 
|---|
| 224 | atomic_dec(v: &rcu_expedited_nesting); | 
|---|
| 225 | } | 
|---|
| 226 | EXPORT_SYMBOL_GPL(rcu_unexpedite_gp); | 
|---|
| 227 |  | 
|---|
| 228 | static bool rcu_boot_ended __read_mostly; | 
|---|
| 229 |  | 
|---|
| 230 | /* | 
|---|
| 231 | * Inform RCU of the end of the in-kernel boot sequence. | 
|---|
| 232 | */ | 
|---|
| 233 | void rcu_end_inkernel_boot(void) | 
|---|
| 234 | { | 
|---|
| 235 | rcu_unexpedite_gp(); | 
|---|
| 236 | rcu_async_relax(); | 
|---|
| 237 | if (rcu_normal_after_boot) | 
|---|
| 238 | WRITE_ONCE(rcu_normal, 1); | 
|---|
| 239 | rcu_boot_ended = true; | 
|---|
| 240 | } | 
|---|
| 241 |  | 
|---|
| 242 | /* | 
|---|
| 243 | * Let rcutorture know when it is OK to turn it up to eleven. | 
|---|
| 244 | */ | 
|---|
| 245 | bool rcu_inkernel_boot_has_ended(void) | 
|---|
| 246 | { | 
|---|
| 247 | return rcu_boot_ended; | 
|---|
| 248 | } | 
|---|
| 249 | EXPORT_SYMBOL_GPL(rcu_inkernel_boot_has_ended); | 
|---|
| 250 |  | 
|---|
| 251 | #endif /* #ifndef CONFIG_TINY_RCU */ | 
|---|
| 252 |  | 
|---|
| 253 | /* | 
|---|
| 254 | * Test each non-SRCU synchronous grace-period wait API.  This is | 
|---|
| 255 | * useful just after a change in mode for these primitives, and | 
|---|
| 256 | * during early boot. | 
|---|
| 257 | */ | 
|---|
| 258 | void rcu_test_sync_prims(void) | 
|---|
| 259 | { | 
|---|
| 260 | if (!IS_ENABLED(CONFIG_PROVE_RCU)) | 
|---|
| 261 | return; | 
|---|
| 262 | pr_info( "Running RCU synchronous self tests\n"); | 
|---|
| 263 | synchronize_rcu(); | 
|---|
| 264 | synchronize_rcu_expedited(); | 
|---|
| 265 | } | 
|---|
| 266 |  | 
|---|
| 267 | #if !defined(CONFIG_TINY_RCU) | 
|---|
| 268 |  | 
|---|
| 269 | /* | 
|---|
| 270 | * Switch to run-time mode once RCU has fully initialized. | 
|---|
| 271 | */ | 
|---|
| 272 | static int __init rcu_set_runtime_mode(void) | 
|---|
| 273 | { | 
|---|
| 274 | rcu_test_sync_prims(); | 
|---|
| 275 | rcu_scheduler_active = RCU_SCHEDULER_RUNNING; | 
|---|
| 276 | kfree_rcu_scheduler_running(); | 
|---|
| 277 | rcu_test_sync_prims(); | 
|---|
| 278 | return 0; | 
|---|
| 279 | } | 
|---|
| 280 | core_initcall(rcu_set_runtime_mode); | 
|---|
| 281 |  | 
|---|
| 282 | #endif /* #if !defined(CONFIG_TINY_RCU) */ | 
|---|
| 283 |  | 
|---|
| 284 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | 
|---|
| 285 | static struct lock_class_key rcu_lock_key; | 
|---|
| 286 | struct lockdep_map rcu_lock_map = { | 
|---|
| 287 | .name = "rcu_read_lock", | 
|---|
| 288 | .key = &rcu_lock_key, | 
|---|
| 289 | .wait_type_outer = LD_WAIT_FREE, | 
|---|
| 290 | .wait_type_inner = LD_WAIT_CONFIG, /* PREEMPT_RT implies PREEMPT_RCU */ | 
|---|
| 291 | }; | 
|---|
| 292 | EXPORT_SYMBOL_GPL(rcu_lock_map); | 
|---|
| 293 |  | 
|---|
| 294 | static struct lock_class_key rcu_bh_lock_key; | 
|---|
| 295 | struct lockdep_map rcu_bh_lock_map = { | 
|---|
| 296 | .name = "rcu_read_lock_bh", | 
|---|
| 297 | .key = &rcu_bh_lock_key, | 
|---|
| 298 | .wait_type_outer = LD_WAIT_FREE, | 
|---|
| 299 | .wait_type_inner = LD_WAIT_CONFIG, /* PREEMPT_RT makes BH preemptible. */ | 
|---|
| 300 | }; | 
|---|
| 301 | EXPORT_SYMBOL_GPL(rcu_bh_lock_map); | 
|---|
| 302 |  | 
|---|
| 303 | static struct lock_class_key rcu_sched_lock_key; | 
|---|
| 304 | struct lockdep_map rcu_sched_lock_map = { | 
|---|
| 305 | .name = "rcu_read_lock_sched", | 
|---|
| 306 | .key = &rcu_sched_lock_key, | 
|---|
| 307 | .wait_type_outer = LD_WAIT_FREE, | 
|---|
| 308 | .wait_type_inner = LD_WAIT_SPIN, | 
|---|
| 309 | }; | 
|---|
| 310 | EXPORT_SYMBOL_GPL(rcu_sched_lock_map); | 
|---|
| 311 |  | 
|---|
| 312 | // Tell lockdep when RCU callbacks are being invoked. | 
|---|
| 313 | static struct lock_class_key rcu_callback_key; | 
|---|
| 314 | struct lockdep_map rcu_callback_map = | 
|---|
| 315 | STATIC_LOCKDEP_MAP_INIT( "rcu_callback", &rcu_callback_key); | 
|---|
| 316 | EXPORT_SYMBOL_GPL(rcu_callback_map); | 
|---|
| 317 |  | 
|---|
| 318 | noinstr int notrace debug_lockdep_rcu_enabled(void) | 
|---|
| 319 | { | 
|---|
| 320 | return rcu_scheduler_active != RCU_SCHEDULER_INACTIVE && READ_ONCE(debug_locks) && | 
|---|
| 321 | current->lockdep_recursion == 0; | 
|---|
| 322 | } | 
|---|
| 323 | EXPORT_SYMBOL_GPL(debug_lockdep_rcu_enabled); | 
|---|
| 324 |  | 
|---|
| 325 | /** | 
|---|
| 326 | * rcu_read_lock_held() - might we be in RCU read-side critical section? | 
|---|
| 327 | * | 
|---|
| 328 | * If CONFIG_DEBUG_LOCK_ALLOC is selected, returns nonzero iff in an RCU | 
|---|
| 329 | * read-side critical section.  In absence of CONFIG_DEBUG_LOCK_ALLOC, | 
|---|
| 330 | * this assumes we are in an RCU read-side critical section unless it can | 
|---|
| 331 | * prove otherwise.  This is useful for debug checks in functions that | 
|---|
| 332 | * require that they be called within an RCU read-side critical section. | 
|---|
| 333 | * | 
|---|
| 334 | * Checks debug_lockdep_rcu_enabled() to prevent false positives during boot | 
|---|
| 335 | * and while lockdep is disabled. | 
|---|
| 336 | * | 
|---|
| 337 | * Note that rcu_read_lock() and the matching rcu_read_unlock() must | 
|---|
| 338 | * occur in the same context, for example, it is illegal to invoke | 
|---|
| 339 | * rcu_read_unlock() in process context if the matching rcu_read_lock() | 
|---|
| 340 | * was invoked from within an irq handler. | 
|---|
| 341 | * | 
|---|
| 342 | * Note that rcu_read_lock() is disallowed if the CPU is either idle or | 
|---|
| 343 | * offline from an RCU perspective, so check for those as well. | 
|---|
| 344 | */ | 
|---|
| 345 | int rcu_read_lock_held(void) | 
|---|
| 346 | { | 
|---|
| 347 | bool ret; | 
|---|
| 348 |  | 
|---|
| 349 | if (rcu_read_lock_held_common(&ret)) | 
|---|
| 350 | return ret; | 
|---|
| 351 | return lock_is_held(&rcu_lock_map); | 
|---|
| 352 | } | 
|---|
| 353 | EXPORT_SYMBOL_GPL(rcu_read_lock_held); | 
|---|
| 354 |  | 
|---|
| 355 | /** | 
|---|
| 356 | * rcu_read_lock_bh_held() - might we be in RCU-bh read-side critical section? | 
|---|
| 357 | * | 
|---|
| 358 | * Check for bottom half being disabled, which covers both the | 
|---|
| 359 | * CONFIG_PROVE_RCU and not cases.  Note that if someone uses | 
|---|
| 360 | * rcu_read_lock_bh(), but then later enables BH, lockdep (if enabled) | 
|---|
| 361 | * will show the situation.  This is useful for debug checks in functions | 
|---|
| 362 | * that require that they be called within an RCU read-side critical | 
|---|
| 363 | * section. | 
|---|
| 364 | * | 
|---|
| 365 | * Check debug_lockdep_rcu_enabled() to prevent false positives during boot. | 
|---|
| 366 | * | 
|---|
| 367 | * Note that rcu_read_lock_bh() is disallowed if the CPU is either idle or | 
|---|
| 368 | * offline from an RCU perspective, so check for those as well. | 
|---|
| 369 | */ | 
|---|
| 370 | int rcu_read_lock_bh_held(void) | 
|---|
| 371 | { | 
|---|
| 372 | bool ret; | 
|---|
| 373 |  | 
|---|
| 374 | if (rcu_read_lock_held_common(&ret)) | 
|---|
| 375 | return ret; | 
|---|
| 376 | return in_softirq() || irqs_disabled(); | 
|---|
| 377 | } | 
|---|
| 378 | EXPORT_SYMBOL_GPL(rcu_read_lock_bh_held); | 
|---|
| 379 |  | 
|---|
| 380 | int rcu_read_lock_any_held(void) | 
|---|
| 381 | { | 
|---|
| 382 | bool ret; | 
|---|
| 383 |  | 
|---|
| 384 | if (rcu_read_lock_held_common(&ret)) | 
|---|
| 385 | return ret; | 
|---|
| 386 | if (lock_is_held(&rcu_lock_map) || | 
|---|
| 387 | lock_is_held(&rcu_bh_lock_map) || | 
|---|
| 388 | lock_is_held(&rcu_sched_lock_map)) | 
|---|
| 389 | return 1; | 
|---|
| 390 | return !preemptible(); | 
|---|
| 391 | } | 
|---|
| 392 | EXPORT_SYMBOL_GPL(rcu_read_lock_any_held); | 
|---|
| 393 |  | 
|---|
| 394 | #endif /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */ | 
|---|
| 395 |  | 
|---|
| 396 | /** | 
|---|
| 397 | * wakeme_after_rcu() - Callback function to awaken a task after grace period | 
|---|
| 398 | * @head: Pointer to rcu_head member within rcu_synchronize structure | 
|---|
| 399 | * | 
|---|
| 400 | * Awaken the corresponding task now that a grace period has elapsed. | 
|---|
| 401 | */ | 
|---|
| 402 | void wakeme_after_rcu(struct rcu_head *head) | 
|---|
| 403 | { | 
|---|
| 404 | struct rcu_synchronize *rcu; | 
|---|
| 405 |  | 
|---|
| 406 | rcu = container_of(head, struct rcu_synchronize, head); | 
|---|
| 407 | complete(&rcu->completion); | 
|---|
| 408 | } | 
|---|
| 409 | EXPORT_SYMBOL_GPL(wakeme_after_rcu); | 
|---|
| 410 |  | 
|---|
| 411 | void __wait_rcu_gp(bool checktiny, unsigned int state, int n, call_rcu_func_t *crcu_array, | 
|---|
| 412 | struct rcu_synchronize *rs_array) | 
|---|
| 413 | { | 
|---|
| 414 | int i; | 
|---|
| 415 | int j; | 
|---|
| 416 |  | 
|---|
| 417 | /* Initialize and register callbacks for each crcu_array element. */ | 
|---|
| 418 | for (i = 0; i < n; i++) { | 
|---|
| 419 | if (checktiny && | 
|---|
| 420 | (crcu_array[i] == call_rcu)) { | 
|---|
| 421 | might_sleep(); | 
|---|
| 422 | continue; | 
|---|
| 423 | } | 
|---|
| 424 | for (j = 0; j < i; j++) | 
|---|
| 425 | if (crcu_array[j] == crcu_array[i]) | 
|---|
| 426 | break; | 
|---|
| 427 | if (j == i) { | 
|---|
| 428 | init_rcu_head_on_stack(head: &rs_array[i].head); | 
|---|
| 429 | init_completion(x: &rs_array[i].completion); | 
|---|
| 430 | (crcu_array[i])(&rs_array[i].head, wakeme_after_rcu); | 
|---|
| 431 | } | 
|---|
| 432 | } | 
|---|
| 433 |  | 
|---|
| 434 | /* Wait for all callbacks to be invoked. */ | 
|---|
| 435 | for (i = 0; i < n; i++) { | 
|---|
| 436 | if (checktiny && | 
|---|
| 437 | (crcu_array[i] == call_rcu)) | 
|---|
| 438 | continue; | 
|---|
| 439 | for (j = 0; j < i; j++) | 
|---|
| 440 | if (crcu_array[j] == crcu_array[i]) | 
|---|
| 441 | break; | 
|---|
| 442 | if (j == i) { | 
|---|
| 443 | wait_for_completion_state(x: &rs_array[i].completion, state); | 
|---|
| 444 | destroy_rcu_head_on_stack(head: &rs_array[i].head); | 
|---|
| 445 | } | 
|---|
| 446 | } | 
|---|
| 447 | } | 
|---|
| 448 | EXPORT_SYMBOL_GPL(__wait_rcu_gp); | 
|---|
| 449 |  | 
|---|
| 450 | void finish_rcuwait(struct rcuwait *w) | 
|---|
| 451 | { | 
|---|
| 452 | rcu_assign_pointer(w->task, NULL); | 
|---|
| 453 | __set_current_state(TASK_RUNNING); | 
|---|
| 454 | } | 
|---|
| 455 | EXPORT_SYMBOL_GPL(finish_rcuwait); | 
|---|
| 456 |  | 
|---|
| 457 | #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD | 
|---|
| 458 | void init_rcu_head(struct rcu_head *head) | 
|---|
| 459 | { | 
|---|
| 460 | debug_object_init(head, &rcuhead_debug_descr); | 
|---|
| 461 | } | 
|---|
| 462 | EXPORT_SYMBOL_GPL(init_rcu_head); | 
|---|
| 463 |  | 
|---|
| 464 | void destroy_rcu_head(struct rcu_head *head) | 
|---|
| 465 | { | 
|---|
| 466 | debug_object_free(head, &rcuhead_debug_descr); | 
|---|
| 467 | } | 
|---|
| 468 | EXPORT_SYMBOL_GPL(destroy_rcu_head); | 
|---|
| 469 |  | 
|---|
| 470 | static bool rcuhead_is_static_object(void *addr) | 
|---|
| 471 | { | 
|---|
| 472 | return true; | 
|---|
| 473 | } | 
|---|
| 474 |  | 
|---|
| 475 | /** | 
|---|
| 476 | * init_rcu_head_on_stack() - initialize on-stack rcu_head for debugobjects | 
|---|
| 477 | * @head: pointer to rcu_head structure to be initialized | 
|---|
| 478 | * | 
|---|
| 479 | * This function informs debugobjects of a new rcu_head structure that | 
|---|
| 480 | * has been allocated as an auto variable on the stack.  This function | 
|---|
| 481 | * is not required for rcu_head structures that are statically defined or | 
|---|
| 482 | * that are dynamically allocated on the heap.  This function has no | 
|---|
| 483 | * effect for !CONFIG_DEBUG_OBJECTS_RCU_HEAD kernel builds. | 
|---|
| 484 | */ | 
|---|
| 485 | void init_rcu_head_on_stack(struct rcu_head *head) | 
|---|
| 486 | { | 
|---|
| 487 | debug_object_init_on_stack(head, &rcuhead_debug_descr); | 
|---|
| 488 | } | 
|---|
| 489 | EXPORT_SYMBOL_GPL(init_rcu_head_on_stack); | 
|---|
| 490 |  | 
|---|
| 491 | /** | 
|---|
| 492 | * destroy_rcu_head_on_stack() - destroy on-stack rcu_head for debugobjects | 
|---|
| 493 | * @head: pointer to rcu_head structure to be initialized | 
|---|
| 494 | * | 
|---|
| 495 | * This function informs debugobjects that an on-stack rcu_head structure | 
|---|
| 496 | * is about to go out of scope.  As with init_rcu_head_on_stack(), this | 
|---|
| 497 | * function is not required for rcu_head structures that are statically | 
|---|
| 498 | * defined or that are dynamically allocated on the heap.  Also as with | 
|---|
| 499 | * init_rcu_head_on_stack(), this function has no effect for | 
|---|
| 500 | * !CONFIG_DEBUG_OBJECTS_RCU_HEAD kernel builds. | 
|---|
| 501 | */ | 
|---|
| 502 | void destroy_rcu_head_on_stack(struct rcu_head *head) | 
|---|
| 503 | { | 
|---|
| 504 | debug_object_free(head, &rcuhead_debug_descr); | 
|---|
| 505 | } | 
|---|
| 506 | EXPORT_SYMBOL_GPL(destroy_rcu_head_on_stack); | 
|---|
| 507 |  | 
|---|
| 508 | const struct debug_obj_descr rcuhead_debug_descr = { | 
|---|
| 509 | .name = "rcu_head", | 
|---|
| 510 | .is_static_object = rcuhead_is_static_object, | 
|---|
| 511 | }; | 
|---|
| 512 | EXPORT_SYMBOL_GPL(rcuhead_debug_descr); | 
|---|
| 513 | #endif /* #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD */ | 
|---|
| 514 |  | 
|---|
| 515 | #if defined(CONFIG_TREE_RCU) || defined(CONFIG_RCU_TRACE) | 
|---|
| 516 | void do_trace_rcu_torture_read(const char *rcutorturename, struct rcu_head *rhp, | 
|---|
| 517 | unsigned long secs, | 
|---|
| 518 | unsigned long c_old, unsigned long c) | 
|---|
| 519 | { | 
|---|
| 520 | trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c); | 
|---|
| 521 | } | 
|---|
| 522 | EXPORT_SYMBOL_GPL(do_trace_rcu_torture_read); | 
|---|
| 523 | #else | 
|---|
| 524 | #define do_trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c) \ | 
|---|
| 525 | do { } while (0) | 
|---|
| 526 | #endif | 
|---|
| 527 |  | 
|---|
| 528 | #if IS_ENABLED(CONFIG_RCU_TORTURE_TEST) || IS_MODULE(CONFIG_RCU_TORTURE_TEST) || IS_ENABLED(CONFIG_LOCK_TORTURE_TEST) || IS_MODULE(CONFIG_LOCK_TORTURE_TEST) | 
|---|
| 529 | /* Get rcutorture access to sched_setaffinity(). */ | 
|---|
| 530 | long torture_sched_setaffinity(pid_t pid, const struct cpumask *in_mask, bool dowarn) | 
|---|
| 531 | { | 
|---|
| 532 | int ret; | 
|---|
| 533 |  | 
|---|
| 534 | ret = sched_setaffinity(pid, in_mask); | 
|---|
| 535 | WARN_ONCE(dowarn && ret, "%s: sched_setaffinity(%d) returned %d\n", __func__, pid, ret); | 
|---|
| 536 | return ret; | 
|---|
| 537 | } | 
|---|
| 538 | EXPORT_SYMBOL_GPL(torture_sched_setaffinity); | 
|---|
| 539 | #endif | 
|---|
| 540 |  | 
|---|
| 541 | int rcu_cpu_stall_notifiers __read_mostly; // !0 = provide stall notifiers (rarely useful) | 
|---|
| 542 | EXPORT_SYMBOL_GPL(rcu_cpu_stall_notifiers); | 
|---|
| 543 |  | 
|---|
| 544 | #ifdef CONFIG_RCU_STALL_COMMON | 
|---|
| 545 | int rcu_cpu_stall_ftrace_dump __read_mostly; | 
|---|
| 546 | module_param(rcu_cpu_stall_ftrace_dump, int, 0644); | 
|---|
| 547 | #ifdef CONFIG_RCU_CPU_STALL_NOTIFIER | 
|---|
| 548 | module_param(rcu_cpu_stall_notifiers, int, 0444); | 
|---|
| 549 | #endif // #ifdef CONFIG_RCU_CPU_STALL_NOTIFIER | 
|---|
| 550 | int rcu_cpu_stall_suppress __read_mostly; // !0 = suppress stall warnings. | 
|---|
| 551 | EXPORT_SYMBOL_GPL(rcu_cpu_stall_suppress); | 
|---|
| 552 | module_param(rcu_cpu_stall_suppress, int, 0644); | 
|---|
| 553 | int rcu_cpu_stall_timeout __read_mostly = CONFIG_RCU_CPU_STALL_TIMEOUT; | 
|---|
| 554 | module_param(rcu_cpu_stall_timeout, int, 0644); | 
|---|
| 555 | int rcu_exp_cpu_stall_timeout __read_mostly = CONFIG_RCU_EXP_CPU_STALL_TIMEOUT; | 
|---|
| 556 | module_param(rcu_exp_cpu_stall_timeout, int, 0644); | 
|---|
| 557 | int rcu_cpu_stall_cputime __read_mostly = IS_ENABLED(CONFIG_RCU_CPU_STALL_CPUTIME); | 
|---|
| 558 | module_param(rcu_cpu_stall_cputime, int, 0644); | 
|---|
| 559 | bool rcu_exp_stall_task_details __read_mostly; | 
|---|
| 560 | module_param(rcu_exp_stall_task_details, bool, 0644); | 
|---|
| 561 | #endif /* #ifdef CONFIG_RCU_STALL_COMMON */ | 
|---|
| 562 |  | 
|---|
| 563 | // Suppress boot-time RCU CPU stall warnings and rcutorture writer stall | 
|---|
| 564 | // warnings.  Also used by rcutorture even if stall warnings are excluded. | 
|---|
| 565 | int rcu_cpu_stall_suppress_at_boot __read_mostly; // !0 = suppress boot stalls. | 
|---|
| 566 | EXPORT_SYMBOL_GPL(rcu_cpu_stall_suppress_at_boot); | 
|---|
| 567 | module_param(rcu_cpu_stall_suppress_at_boot, int, 0444); | 
|---|
| 568 |  | 
|---|
| 569 | /** | 
|---|
| 570 | * get_completed_synchronize_rcu - Return a pre-completed polled state cookie | 
|---|
| 571 | * | 
|---|
| 572 | * Returns a value that will always be treated by functions like | 
|---|
| 573 | * poll_state_synchronize_rcu() as a cookie whose grace period has already | 
|---|
| 574 | * completed. | 
|---|
| 575 | */ | 
|---|
| 576 | unsigned long get_completed_synchronize_rcu(void) | 
|---|
| 577 | { | 
|---|
| 578 | return RCU_GET_STATE_COMPLETED; | 
|---|
| 579 | } | 
|---|
| 580 | EXPORT_SYMBOL_GPL(get_completed_synchronize_rcu); | 
|---|
| 581 |  | 
|---|
| 582 | #ifdef CONFIG_PROVE_RCU | 
|---|
| 583 |  | 
|---|
| 584 | /* | 
|---|
| 585 | * Early boot self test parameters. | 
|---|
| 586 | */ | 
|---|
| 587 | static bool rcu_self_test; | 
|---|
| 588 | module_param(rcu_self_test, bool, 0444); | 
|---|
| 589 |  | 
|---|
| 590 | static int rcu_self_test_counter; | 
|---|
| 591 |  | 
|---|
| 592 | static void test_callback(struct rcu_head *r) | 
|---|
| 593 | { | 
|---|
| 594 | rcu_self_test_counter++; | 
|---|
| 595 | pr_info( "RCU test callback executed %d\n", rcu_self_test_counter); | 
|---|
| 596 | } | 
|---|
| 597 |  | 
|---|
| 598 | DEFINE_STATIC_SRCU(early_srcu); | 
|---|
| 599 | static unsigned long early_srcu_cookie; | 
|---|
| 600 |  | 
|---|
| 601 | struct early_boot_kfree_rcu { | 
|---|
| 602 | struct rcu_head rh; | 
|---|
| 603 | }; | 
|---|
| 604 |  | 
|---|
| 605 | static void early_boot_test_call_rcu(void) | 
|---|
| 606 | { | 
|---|
| 607 | static struct rcu_head head; | 
|---|
| 608 | int idx; | 
|---|
| 609 | static struct rcu_head shead; | 
|---|
| 610 | struct early_boot_kfree_rcu *rhp; | 
|---|
| 611 |  | 
|---|
| 612 | idx = srcu_down_read(&early_srcu); | 
|---|
| 613 | srcu_up_read(&early_srcu, idx); | 
|---|
| 614 | call_rcu(&head, test_callback); | 
|---|
| 615 | early_srcu_cookie = start_poll_synchronize_srcu(&early_srcu); | 
|---|
| 616 | call_srcu(&early_srcu, &shead, test_callback); | 
|---|
| 617 | rhp = kmalloc(sizeof(*rhp), GFP_KERNEL); | 
|---|
| 618 | if (!WARN_ON_ONCE(!rhp)) | 
|---|
| 619 | kfree_rcu(rhp, rh); | 
|---|
| 620 | } | 
|---|
| 621 |  | 
|---|
| 622 | void rcu_early_boot_tests(void) | 
|---|
| 623 | { | 
|---|
| 624 | pr_info( "Running RCU self tests\n"); | 
|---|
| 625 |  | 
|---|
| 626 | if (rcu_self_test) | 
|---|
| 627 | early_boot_test_call_rcu(); | 
|---|
| 628 | rcu_test_sync_prims(); | 
|---|
| 629 | } | 
|---|
| 630 |  | 
|---|
| 631 | static int rcu_verify_early_boot_tests(void) | 
|---|
| 632 | { | 
|---|
| 633 | int ret = 0; | 
|---|
| 634 | int early_boot_test_counter = 0; | 
|---|
| 635 |  | 
|---|
| 636 | if (rcu_self_test) { | 
|---|
| 637 | early_boot_test_counter++; | 
|---|
| 638 | rcu_barrier(); | 
|---|
| 639 | early_boot_test_counter++; | 
|---|
| 640 | srcu_barrier(&early_srcu); | 
|---|
| 641 | WARN_ON_ONCE(!poll_state_synchronize_srcu(&early_srcu, early_srcu_cookie)); | 
|---|
| 642 | cleanup_srcu_struct(&early_srcu); | 
|---|
| 643 | } | 
|---|
| 644 | if (rcu_self_test_counter != early_boot_test_counter) { | 
|---|
| 645 | WARN_ON(1); | 
|---|
| 646 | ret = -1; | 
|---|
| 647 | } | 
|---|
| 648 |  | 
|---|
| 649 | return ret; | 
|---|
| 650 | } | 
|---|
| 651 | late_initcall(rcu_verify_early_boot_tests); | 
|---|
| 652 | #else | 
|---|
| 653 | void rcu_early_boot_tests(void) {} | 
|---|
| 654 | #endif /* CONFIG_PROVE_RCU */ | 
|---|
| 655 |  | 
|---|
| 656 | #include "tasks.h" | 
|---|
| 657 |  | 
|---|
| 658 | #ifndef CONFIG_TINY_RCU | 
|---|
| 659 |  | 
|---|
| 660 | /* | 
|---|
| 661 | * Print any significant non-default boot-time settings. | 
|---|
| 662 | */ | 
|---|
| 663 | void __init rcupdate_announce_bootup_oddness(void) | 
|---|
| 664 | { | 
|---|
| 665 | if (rcu_normal) | 
|---|
| 666 | pr_info( "\tNo expedited grace period (rcu_normal).\n"); | 
|---|
| 667 | else if (rcu_normal_after_boot) | 
|---|
| 668 | pr_info( "\tNo expedited grace period (rcu_normal_after_boot).\n"); | 
|---|
| 669 | else if (rcu_expedited) | 
|---|
| 670 | pr_info( "\tAll grace periods are expedited (rcu_expedited).\n"); | 
|---|
| 671 | if (rcu_cpu_stall_suppress) | 
|---|
| 672 | pr_info( "\tRCU CPU stall warnings suppressed (rcu_cpu_stall_suppress).\n"); | 
|---|
| 673 | if (rcu_cpu_stall_timeout != CONFIG_RCU_CPU_STALL_TIMEOUT) | 
|---|
| 674 | pr_info( "\tRCU CPU stall warnings timeout set to %d (rcu_cpu_stall_timeout).\n", rcu_cpu_stall_timeout); | 
|---|
| 675 | rcu_tasks_bootup_oddness(); | 
|---|
| 676 | } | 
|---|
| 677 |  | 
|---|
| 678 | #endif /* #ifndef CONFIG_TINY_RCU */ | 
|---|
| 679 |  | 
|---|