Paul E. McKenney | 01c1c66 | 2008-01-25 21:08:24 +0100 | [diff] [blame] | 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, write to the Free Software |
| 16 | * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. |
| 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 | * Documentation/RCU |
| 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/rcupdate.h> |
| 39 | #include <linux/interrupt.h> |
| 40 | #include <linux/sched.h> |
| 41 | #include <asm/atomic.h> |
| 42 | #include <linux/bitops.h> |
| 43 | #include <linux/module.h> |
| 44 | #include <linux/completion.h> |
| 45 | #include <linux/moduleparam.h> |
| 46 | #include <linux/percpu.h> |
| 47 | #include <linux/notifier.h> |
Paul E. McKenney | 01c1c66 | 2008-01-25 21:08:24 +0100 | [diff] [blame] | 48 | #include <linux/cpu.h> |
| 49 | #include <linux/mutex.h> |
| 50 | |
| 51 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
| 52 | static struct lock_class_key rcu_lock_key; |
| 53 | struct lockdep_map rcu_lock_map = |
| 54 | STATIC_LOCKDEP_MAP_INIT("rcu_read_lock", &rcu_lock_key); |
| 55 | EXPORT_SYMBOL_GPL(rcu_lock_map); |
| 56 | #endif |
| 57 | |
| 58 | |
| 59 | /* Definition for rcupdate control block. */ |
| 60 | static struct rcu_ctrlblk rcu_ctrlblk = { |
| 61 | .cur = -300, |
| 62 | .completed = -300, |
| 63 | .lock = __SPIN_LOCK_UNLOCKED(&rcu_ctrlblk.lock), |
| 64 | .cpumask = CPU_MASK_NONE, |
| 65 | }; |
| 66 | static struct rcu_ctrlblk rcu_bh_ctrlblk = { |
| 67 | .cur = -300, |
| 68 | .completed = -300, |
| 69 | .lock = __SPIN_LOCK_UNLOCKED(&rcu_bh_ctrlblk.lock), |
| 70 | .cpumask = CPU_MASK_NONE, |
| 71 | }; |
| 72 | |
| 73 | DEFINE_PER_CPU(struct rcu_data, rcu_data) = { 0L }; |
| 74 | DEFINE_PER_CPU(struct rcu_data, rcu_bh_data) = { 0L }; |
| 75 | |
| 76 | static int blimit = 10; |
| 77 | static int qhimark = 10000; |
| 78 | static int qlowmark = 100; |
| 79 | |
| 80 | #ifdef CONFIG_SMP |
| 81 | static void force_quiescent_state(struct rcu_data *rdp, |
| 82 | struct rcu_ctrlblk *rcp) |
| 83 | { |
| 84 | int cpu; |
| 85 | cpumask_t cpumask; |
| 86 | set_need_resched(); |
| 87 | if (unlikely(!rcp->signaled)) { |
| 88 | rcp->signaled = 1; |
| 89 | /* |
| 90 | * Don't send IPI to itself. With irqs disabled, |
| 91 | * rdp->cpu is the current cpu. |
Gautham R Shenoy | 8558f8f | 2008-06-27 10:17:38 +0530 | [diff] [blame] | 92 | * |
| 93 | * cpu_online_map is updated by the _cpu_down() |
| 94 | * using stop_machine_run(). Since we're in irqs disabled |
| 95 | * section, stop_machine_run() is not exectuting, hence |
| 96 | * the cpu_online_map is stable. |
| 97 | * |
| 98 | * However, a cpu might have been offlined _just_ before |
| 99 | * we disabled irqs while entering here. |
| 100 | * And rcu subsystem might not yet have handled the CPU_DEAD |
| 101 | * notification, leading to the offlined cpu's bit |
| 102 | * being set in the rcp->cpumask. |
| 103 | * |
| 104 | * Hence cpumask = (rcp->cpumask & cpu_online_map) to prevent |
| 105 | * sending smp_reschedule() to an offlined CPU. |
Paul E. McKenney | 01c1c66 | 2008-01-25 21:08:24 +0100 | [diff] [blame] | 106 | */ |
Gautham R Shenoy | 8558f8f | 2008-06-27 10:17:38 +0530 | [diff] [blame] | 107 | cpus_and(cpumask, rcp->cpumask, cpu_online_map); |
Paul E. McKenney | 01c1c66 | 2008-01-25 21:08:24 +0100 | [diff] [blame] | 108 | cpu_clear(rdp->cpu, cpumask); |
Mike Travis | 363ab6f | 2008-05-12 21:21:13 +0200 | [diff] [blame] | 109 | for_each_cpu_mask_nr(cpu, cpumask) |
Paul E. McKenney | 01c1c66 | 2008-01-25 21:08:24 +0100 | [diff] [blame] | 110 | smp_send_reschedule(cpu); |
| 111 | } |
| 112 | } |
| 113 | #else |
| 114 | static inline void force_quiescent_state(struct rcu_data *rdp, |
| 115 | struct rcu_ctrlblk *rcp) |
| 116 | { |
| 117 | set_need_resched(); |
| 118 | } |
| 119 | #endif |
| 120 | |
| 121 | /** |
| 122 | * call_rcu - Queue an RCU callback for invocation after a grace period. |
| 123 | * @head: structure to be used for queueing the RCU updates. |
| 124 | * @func: actual update function to be invoked after the grace period |
| 125 | * |
| 126 | * The update function will be invoked some time after a full grace |
| 127 | * period elapses, in other words after all currently executing RCU |
| 128 | * read-side critical sections have completed. RCU read-side critical |
| 129 | * sections are delimited by rcu_read_lock() and rcu_read_unlock(), |
| 130 | * and may be nested. |
| 131 | */ |
| 132 | void call_rcu(struct rcu_head *head, |
| 133 | void (*func)(struct rcu_head *rcu)) |
| 134 | { |
| 135 | unsigned long flags; |
| 136 | struct rcu_data *rdp; |
| 137 | |
| 138 | head->func = func; |
| 139 | head->next = NULL; |
| 140 | local_irq_save(flags); |
| 141 | rdp = &__get_cpu_var(rcu_data); |
| 142 | *rdp->nxttail = head; |
| 143 | rdp->nxttail = &head->next; |
| 144 | if (unlikely(++rdp->qlen > qhimark)) { |
| 145 | rdp->blimit = INT_MAX; |
| 146 | force_quiescent_state(rdp, &rcu_ctrlblk); |
| 147 | } |
| 148 | local_irq_restore(flags); |
| 149 | } |
| 150 | EXPORT_SYMBOL_GPL(call_rcu); |
| 151 | |
| 152 | /** |
| 153 | * call_rcu_bh - Queue an RCU for invocation after a quicker grace period. |
| 154 | * @head: structure to be used for queueing the RCU updates. |
| 155 | * @func: actual update function to be invoked after the grace period |
| 156 | * |
| 157 | * The update function will be invoked some time after a full grace |
| 158 | * period elapses, in other words after all currently executing RCU |
| 159 | * read-side critical sections have completed. call_rcu_bh() assumes |
| 160 | * that the read-side critical sections end on completion of a softirq |
| 161 | * handler. This means that read-side critical sections in process |
| 162 | * context must not be interrupted by softirqs. This interface is to be |
| 163 | * used when most of the read-side critical sections are in softirq context. |
| 164 | * RCU read-side critical sections are delimited by rcu_read_lock() and |
| 165 | * rcu_read_unlock(), * if in interrupt context or rcu_read_lock_bh() |
| 166 | * and rcu_read_unlock_bh(), if in process context. These may be nested. |
| 167 | */ |
| 168 | void call_rcu_bh(struct rcu_head *head, |
| 169 | void (*func)(struct rcu_head *rcu)) |
| 170 | { |
| 171 | unsigned long flags; |
| 172 | struct rcu_data *rdp; |
| 173 | |
| 174 | head->func = func; |
| 175 | head->next = NULL; |
| 176 | local_irq_save(flags); |
| 177 | rdp = &__get_cpu_var(rcu_bh_data); |
| 178 | *rdp->nxttail = head; |
| 179 | rdp->nxttail = &head->next; |
| 180 | |
| 181 | if (unlikely(++rdp->qlen > qhimark)) { |
| 182 | rdp->blimit = INT_MAX; |
| 183 | force_quiescent_state(rdp, &rcu_bh_ctrlblk); |
| 184 | } |
| 185 | |
| 186 | local_irq_restore(flags); |
| 187 | } |
| 188 | EXPORT_SYMBOL_GPL(call_rcu_bh); |
| 189 | |
| 190 | /* |
| 191 | * Return the number of RCU batches processed thus far. Useful |
| 192 | * for debug and statistics. |
| 193 | */ |
| 194 | long rcu_batches_completed(void) |
| 195 | { |
| 196 | return rcu_ctrlblk.completed; |
| 197 | } |
| 198 | EXPORT_SYMBOL_GPL(rcu_batches_completed); |
| 199 | |
| 200 | /* |
| 201 | * Return the number of RCU batches processed thus far. Useful |
| 202 | * for debug and statistics. |
| 203 | */ |
| 204 | long rcu_batches_completed_bh(void) |
| 205 | { |
| 206 | return rcu_bh_ctrlblk.completed; |
| 207 | } |
| 208 | EXPORT_SYMBOL_GPL(rcu_batches_completed_bh); |
| 209 | |
| 210 | /* Raises the softirq for processing rcu_callbacks. */ |
| 211 | static inline void raise_rcu_softirq(void) |
| 212 | { |
| 213 | raise_softirq(RCU_SOFTIRQ); |
| 214 | /* |
| 215 | * The smp_mb() here is required to ensure that this cpu's |
| 216 | * __rcu_process_callbacks() reads the most recently updated |
| 217 | * value of rcu->cur. |
| 218 | */ |
| 219 | smp_mb(); |
| 220 | } |
| 221 | |
| 222 | /* |
| 223 | * Invoke the completed RCU callbacks. They are expected to be in |
| 224 | * a per-cpu list. |
| 225 | */ |
| 226 | static void rcu_do_batch(struct rcu_data *rdp) |
| 227 | { |
| 228 | struct rcu_head *next, *list; |
| 229 | int count = 0; |
| 230 | |
| 231 | list = rdp->donelist; |
| 232 | while (list) { |
| 233 | next = list->next; |
| 234 | prefetch(next); |
| 235 | list->func(list); |
| 236 | list = next; |
| 237 | if (++count >= rdp->blimit) |
| 238 | break; |
| 239 | } |
| 240 | rdp->donelist = list; |
| 241 | |
| 242 | local_irq_disable(); |
| 243 | rdp->qlen -= count; |
| 244 | local_irq_enable(); |
| 245 | if (rdp->blimit == INT_MAX && rdp->qlen <= qlowmark) |
| 246 | rdp->blimit = blimit; |
| 247 | |
| 248 | if (!rdp->donelist) |
| 249 | rdp->donetail = &rdp->donelist; |
| 250 | else |
| 251 | raise_rcu_softirq(); |
| 252 | } |
| 253 | |
| 254 | /* |
| 255 | * Grace period handling: |
| 256 | * The grace period handling consists out of two steps: |
| 257 | * - A new grace period is started. |
| 258 | * This is done by rcu_start_batch. The start is not broadcasted to |
| 259 | * all cpus, they must pick this up by comparing rcp->cur with |
| 260 | * rdp->quiescbatch. All cpus are recorded in the |
| 261 | * rcu_ctrlblk.cpumask bitmap. |
| 262 | * - All cpus must go through a quiescent state. |
| 263 | * Since the start of the grace period is not broadcasted, at least two |
| 264 | * calls to rcu_check_quiescent_state are required: |
| 265 | * The first call just notices that a new grace period is running. The |
| 266 | * following calls check if there was a quiescent state since the beginning |
| 267 | * of the grace period. If so, it updates rcu_ctrlblk.cpumask. If |
| 268 | * the bitmap is empty, then the grace period is completed. |
| 269 | * rcu_check_quiescent_state calls rcu_start_batch(0) to start the next grace |
| 270 | * period (if necessary). |
| 271 | */ |
| 272 | /* |
| 273 | * Register a new batch of callbacks, and start it up if there is currently no |
| 274 | * active batch and the batch to be registered has not already occurred. |
| 275 | * Caller must hold rcu_ctrlblk.lock. |
| 276 | */ |
| 277 | static void rcu_start_batch(struct rcu_ctrlblk *rcp) |
| 278 | { |
| 279 | if (rcp->next_pending && |
| 280 | rcp->completed == rcp->cur) { |
| 281 | rcp->next_pending = 0; |
| 282 | /* |
| 283 | * next_pending == 0 must be visible in |
| 284 | * __rcu_process_callbacks() before it can see new value of cur. |
| 285 | */ |
| 286 | smp_wmb(); |
| 287 | rcp->cur++; |
| 288 | |
| 289 | /* |
| 290 | * Accessing nohz_cpu_mask before incrementing rcp->cur needs a |
| 291 | * Barrier Otherwise it can cause tickless idle CPUs to be |
| 292 | * included in rcp->cpumask, which will extend graceperiods |
| 293 | * unnecessarily. |
| 294 | */ |
| 295 | smp_mb(); |
| 296 | cpus_andnot(rcp->cpumask, cpu_online_map, nohz_cpu_mask); |
| 297 | |
| 298 | rcp->signaled = 0; |
| 299 | } |
| 300 | } |
| 301 | |
| 302 | /* |
| 303 | * cpu went through a quiescent state since the beginning of the grace period. |
| 304 | * Clear it from the cpu mask and complete the grace period if it was the last |
| 305 | * cpu. Start another grace period if someone has further entries pending |
| 306 | */ |
| 307 | static void cpu_quiet(int cpu, struct rcu_ctrlblk *rcp) |
| 308 | { |
| 309 | cpu_clear(cpu, rcp->cpumask); |
| 310 | if (cpus_empty(rcp->cpumask)) { |
| 311 | /* batch completed ! */ |
| 312 | rcp->completed = rcp->cur; |
| 313 | rcu_start_batch(rcp); |
| 314 | } |
| 315 | } |
| 316 | |
| 317 | /* |
| 318 | * Check if the cpu has gone through a quiescent state (say context |
| 319 | * switch). If so and if it already hasn't done so in this RCU |
| 320 | * quiescent cycle, then indicate that it has done so. |
| 321 | */ |
| 322 | static void rcu_check_quiescent_state(struct rcu_ctrlblk *rcp, |
| 323 | struct rcu_data *rdp) |
| 324 | { |
| 325 | if (rdp->quiescbatch != rcp->cur) { |
| 326 | /* start new grace period: */ |
| 327 | rdp->qs_pending = 1; |
| 328 | rdp->passed_quiesc = 0; |
| 329 | rdp->quiescbatch = rcp->cur; |
| 330 | return; |
| 331 | } |
| 332 | |
| 333 | /* Grace period already completed for this cpu? |
| 334 | * qs_pending is checked instead of the actual bitmap to avoid |
| 335 | * cacheline trashing. |
| 336 | */ |
| 337 | if (!rdp->qs_pending) |
| 338 | return; |
| 339 | |
| 340 | /* |
| 341 | * Was there a quiescent state since the beginning of the grace |
| 342 | * period? If no, then exit and wait for the next call. |
| 343 | */ |
| 344 | if (!rdp->passed_quiesc) |
| 345 | return; |
| 346 | rdp->qs_pending = 0; |
| 347 | |
| 348 | spin_lock(&rcp->lock); |
| 349 | /* |
| 350 | * rdp->quiescbatch/rcp->cur and the cpu bitmap can come out of sync |
| 351 | * during cpu startup. Ignore the quiescent state. |
| 352 | */ |
| 353 | if (likely(rdp->quiescbatch == rcp->cur)) |
| 354 | cpu_quiet(rdp->cpu, rcp); |
| 355 | |
| 356 | spin_unlock(&rcp->lock); |
| 357 | } |
| 358 | |
| 359 | |
| 360 | #ifdef CONFIG_HOTPLUG_CPU |
| 361 | |
| 362 | /* warning! helper for rcu_offline_cpu. do not use elsewhere without reviewing |
| 363 | * locking requirements, the list it's pulling from has to belong to a cpu |
| 364 | * which is dead and hence not processing interrupts. |
| 365 | */ |
| 366 | static void rcu_move_batch(struct rcu_data *this_rdp, struct rcu_head *list, |
| 367 | struct rcu_head **tail) |
| 368 | { |
| 369 | local_irq_disable(); |
| 370 | *this_rdp->nxttail = list; |
| 371 | if (list) |
| 372 | this_rdp->nxttail = tail; |
| 373 | local_irq_enable(); |
| 374 | } |
| 375 | |
| 376 | static void __rcu_offline_cpu(struct rcu_data *this_rdp, |
| 377 | struct rcu_ctrlblk *rcp, struct rcu_data *rdp) |
| 378 | { |
| 379 | /* if the cpu going offline owns the grace period |
| 380 | * we can block indefinitely waiting for it, so flush |
| 381 | * it here |
| 382 | */ |
| 383 | spin_lock_bh(&rcp->lock); |
| 384 | if (rcp->cur != rcp->completed) |
| 385 | cpu_quiet(rdp->cpu, rcp); |
| 386 | spin_unlock_bh(&rcp->lock); |
Paul E. McKenney | e0ecfa7 | 2008-01-25 21:08:24 +0100 | [diff] [blame] | 387 | rcu_move_batch(this_rdp, rdp->donelist, rdp->donetail); |
Paul E. McKenney | 01c1c66 | 2008-01-25 21:08:24 +0100 | [diff] [blame] | 388 | rcu_move_batch(this_rdp, rdp->curlist, rdp->curtail); |
| 389 | rcu_move_batch(this_rdp, rdp->nxtlist, rdp->nxttail); |
Lai Jiangshan | 199a952 | 2008-06-26 10:06:43 +0800 | [diff] [blame] | 390 | |
| 391 | local_irq_disable(); |
| 392 | this_rdp->qlen += rdp->qlen; |
| 393 | local_irq_enable(); |
Paul E. McKenney | 01c1c66 | 2008-01-25 21:08:24 +0100 | [diff] [blame] | 394 | } |
| 395 | |
| 396 | static void rcu_offline_cpu(int cpu) |
| 397 | { |
| 398 | struct rcu_data *this_rdp = &get_cpu_var(rcu_data); |
| 399 | struct rcu_data *this_bh_rdp = &get_cpu_var(rcu_bh_data); |
| 400 | |
| 401 | __rcu_offline_cpu(this_rdp, &rcu_ctrlblk, |
| 402 | &per_cpu(rcu_data, cpu)); |
| 403 | __rcu_offline_cpu(this_bh_rdp, &rcu_bh_ctrlblk, |
| 404 | &per_cpu(rcu_bh_data, cpu)); |
| 405 | put_cpu_var(rcu_data); |
| 406 | put_cpu_var(rcu_bh_data); |
| 407 | } |
| 408 | |
| 409 | #else |
| 410 | |
| 411 | static void rcu_offline_cpu(int cpu) |
| 412 | { |
| 413 | } |
| 414 | |
| 415 | #endif |
| 416 | |
| 417 | /* |
| 418 | * This does the RCU processing work from softirq context. |
| 419 | */ |
| 420 | static void __rcu_process_callbacks(struct rcu_ctrlblk *rcp, |
| 421 | struct rcu_data *rdp) |
| 422 | { |
| 423 | if (rdp->curlist && !rcu_batch_before(rcp->completed, rdp->batch)) { |
| 424 | *rdp->donetail = rdp->curlist; |
| 425 | rdp->donetail = rdp->curtail; |
| 426 | rdp->curlist = NULL; |
| 427 | rdp->curtail = &rdp->curlist; |
| 428 | } |
| 429 | |
| 430 | if (rdp->nxtlist && !rdp->curlist) { |
| 431 | local_irq_disable(); |
| 432 | rdp->curlist = rdp->nxtlist; |
| 433 | rdp->curtail = rdp->nxttail; |
| 434 | rdp->nxtlist = NULL; |
| 435 | rdp->nxttail = &rdp->nxtlist; |
| 436 | local_irq_enable(); |
| 437 | |
| 438 | /* |
| 439 | * start the next batch of callbacks |
| 440 | */ |
| 441 | |
| 442 | /* determine batch number */ |
| 443 | rdp->batch = rcp->cur + 1; |
| 444 | /* see the comment and corresponding wmb() in |
| 445 | * the rcu_start_batch() |
| 446 | */ |
| 447 | smp_rmb(); |
| 448 | |
| 449 | if (!rcp->next_pending) { |
| 450 | /* and start it/schedule start if it's a new batch */ |
| 451 | spin_lock(&rcp->lock); |
| 452 | rcp->next_pending = 1; |
| 453 | rcu_start_batch(rcp); |
| 454 | spin_unlock(&rcp->lock); |
| 455 | } |
| 456 | } |
| 457 | |
| 458 | rcu_check_quiescent_state(rcp, rdp); |
| 459 | if (rdp->donelist) |
| 460 | rcu_do_batch(rdp); |
| 461 | } |
| 462 | |
| 463 | static void rcu_process_callbacks(struct softirq_action *unused) |
| 464 | { |
| 465 | __rcu_process_callbacks(&rcu_ctrlblk, &__get_cpu_var(rcu_data)); |
| 466 | __rcu_process_callbacks(&rcu_bh_ctrlblk, &__get_cpu_var(rcu_bh_data)); |
| 467 | } |
| 468 | |
| 469 | static int __rcu_pending(struct rcu_ctrlblk *rcp, struct rcu_data *rdp) |
| 470 | { |
| 471 | /* This cpu has pending rcu entries and the grace period |
| 472 | * for them has completed. |
| 473 | */ |
| 474 | if (rdp->curlist && !rcu_batch_before(rcp->completed, rdp->batch)) |
| 475 | return 1; |
| 476 | |
| 477 | /* This cpu has no pending entries, but there are new entries */ |
| 478 | if (!rdp->curlist && rdp->nxtlist) |
| 479 | return 1; |
| 480 | |
| 481 | /* This cpu has finished callbacks to invoke */ |
| 482 | if (rdp->donelist) |
| 483 | return 1; |
| 484 | |
| 485 | /* The rcu core waits for a quiescent state from the cpu */ |
| 486 | if (rdp->quiescbatch != rcp->cur || rdp->qs_pending) |
| 487 | return 1; |
| 488 | |
| 489 | /* nothing to do */ |
| 490 | return 0; |
| 491 | } |
| 492 | |
| 493 | /* |
| 494 | * Check to see if there is any immediate RCU-related work to be done |
| 495 | * by the current CPU, returning 1 if so. This function is part of the |
| 496 | * RCU implementation; it is -not- an exported member of the RCU API. |
| 497 | */ |
| 498 | int rcu_pending(int cpu) |
| 499 | { |
| 500 | return __rcu_pending(&rcu_ctrlblk, &per_cpu(rcu_data, cpu)) || |
| 501 | __rcu_pending(&rcu_bh_ctrlblk, &per_cpu(rcu_bh_data, cpu)); |
| 502 | } |
| 503 | |
| 504 | /* |
| 505 | * Check to see if any future RCU-related work will need to be done |
| 506 | * by the current CPU, even if none need be done immediately, returning |
| 507 | * 1 if so. This function is part of the RCU implementation; it is -not- |
| 508 | * an exported member of the RCU API. |
| 509 | */ |
| 510 | int rcu_needs_cpu(int cpu) |
| 511 | { |
| 512 | struct rcu_data *rdp = &per_cpu(rcu_data, cpu); |
| 513 | struct rcu_data *rdp_bh = &per_cpu(rcu_bh_data, cpu); |
| 514 | |
| 515 | return (!!rdp->curlist || !!rdp_bh->curlist || rcu_pending(cpu)); |
| 516 | } |
| 517 | |
| 518 | void rcu_check_callbacks(int cpu, int user) |
| 519 | { |
| 520 | if (user || |
| 521 | (idle_cpu(cpu) && !in_softirq() && |
| 522 | hardirq_count() <= (1 << HARDIRQ_SHIFT))) { |
Paul E. McKenney | 8db559b | 2008-05-12 21:21:05 +0200 | [diff] [blame] | 523 | |
| 524 | /* |
| 525 | * Get here if this CPU took its interrupt from user |
| 526 | * mode or from the idle loop, and if this is not a |
| 527 | * nested interrupt. In this case, the CPU is in |
| 528 | * a quiescent state, so count it. |
| 529 | * |
| 530 | * Also do a memory barrier. This is needed to handle |
| 531 | * the case where writes from a preempt-disable section |
| 532 | * of code get reordered into schedule() by this CPU's |
| 533 | * write buffer. The memory barrier makes sure that |
| 534 | * the rcu_qsctr_inc() and rcu_bh_qsctr_inc() are see |
| 535 | * by other CPUs to happen after any such write. |
| 536 | */ |
| 537 | |
| 538 | smp_mb(); /* See above block comment. */ |
Paul E. McKenney | 01c1c66 | 2008-01-25 21:08:24 +0100 | [diff] [blame] | 539 | rcu_qsctr_inc(cpu); |
| 540 | rcu_bh_qsctr_inc(cpu); |
Paul E. McKenney | 8db559b | 2008-05-12 21:21:05 +0200 | [diff] [blame] | 541 | |
| 542 | } else if (!in_softirq()) { |
| 543 | |
| 544 | /* |
| 545 | * Get here if this CPU did not take its interrupt from |
| 546 | * softirq, in other words, if it is not interrupting |
| 547 | * a rcu_bh read-side critical section. This is an _bh |
| 548 | * critical section, so count it. The memory barrier |
| 549 | * is needed for the same reason as is the above one. |
| 550 | */ |
| 551 | |
| 552 | smp_mb(); /* See above block comment. */ |
Paul E. McKenney | 01c1c66 | 2008-01-25 21:08:24 +0100 | [diff] [blame] | 553 | rcu_bh_qsctr_inc(cpu); |
Paul E. McKenney | 8db559b | 2008-05-12 21:21:05 +0200 | [diff] [blame] | 554 | } |
Paul E. McKenney | 01c1c66 | 2008-01-25 21:08:24 +0100 | [diff] [blame] | 555 | raise_rcu_softirq(); |
| 556 | } |
| 557 | |
| 558 | static void rcu_init_percpu_data(int cpu, struct rcu_ctrlblk *rcp, |
| 559 | struct rcu_data *rdp) |
| 560 | { |
| 561 | memset(rdp, 0, sizeof(*rdp)); |
| 562 | rdp->curtail = &rdp->curlist; |
| 563 | rdp->nxttail = &rdp->nxtlist; |
| 564 | rdp->donetail = &rdp->donelist; |
| 565 | rdp->quiescbatch = rcp->completed; |
| 566 | rdp->qs_pending = 0; |
| 567 | rdp->cpu = cpu; |
| 568 | rdp->blimit = blimit; |
| 569 | } |
| 570 | |
| 571 | static void __cpuinit rcu_online_cpu(int cpu) |
| 572 | { |
| 573 | struct rcu_data *rdp = &per_cpu(rcu_data, cpu); |
| 574 | struct rcu_data *bh_rdp = &per_cpu(rcu_bh_data, cpu); |
| 575 | |
| 576 | rcu_init_percpu_data(cpu, &rcu_ctrlblk, rdp); |
| 577 | rcu_init_percpu_data(cpu, &rcu_bh_ctrlblk, bh_rdp); |
Carlos R. Mafra | 962cf36 | 2008-05-15 11:15:37 -0300 | [diff] [blame] | 578 | open_softirq(RCU_SOFTIRQ, rcu_process_callbacks); |
Paul E. McKenney | 01c1c66 | 2008-01-25 21:08:24 +0100 | [diff] [blame] | 579 | } |
| 580 | |
| 581 | static int __cpuinit rcu_cpu_notify(struct notifier_block *self, |
| 582 | unsigned long action, void *hcpu) |
| 583 | { |
| 584 | long cpu = (long)hcpu; |
| 585 | |
| 586 | switch (action) { |
| 587 | case CPU_UP_PREPARE: |
| 588 | case CPU_UP_PREPARE_FROZEN: |
| 589 | rcu_online_cpu(cpu); |
| 590 | break; |
| 591 | case CPU_DEAD: |
| 592 | case CPU_DEAD_FROZEN: |
| 593 | rcu_offline_cpu(cpu); |
| 594 | break; |
| 595 | default: |
| 596 | break; |
| 597 | } |
| 598 | return NOTIFY_OK; |
| 599 | } |
| 600 | |
| 601 | static struct notifier_block __cpuinitdata rcu_nb = { |
| 602 | .notifier_call = rcu_cpu_notify, |
| 603 | }; |
| 604 | |
| 605 | /* |
| 606 | * Initializes rcu mechanism. Assumed to be called early. |
| 607 | * That is before local timer(SMP) or jiffie timer (uniproc) is setup. |
| 608 | * Note that rcu_qsctr and friends are implicitly |
| 609 | * initialized due to the choice of ``0'' for RCU_CTR_INVALID. |
| 610 | */ |
| 611 | void __init __rcu_init(void) |
| 612 | { |
| 613 | rcu_cpu_notify(&rcu_nb, CPU_UP_PREPARE, |
| 614 | (void *)(long)smp_processor_id()); |
| 615 | /* Register notifier for non-boot CPUs */ |
| 616 | register_cpu_notifier(&rcu_nb); |
| 617 | } |
| 618 | |
| 619 | module_param(blimit, int, 0); |
| 620 | module_param(qhimark, int, 0); |
| 621 | module_param(qlowmark, int, 0); |