Paul E. McKenney | 64db4cf | 2008-12-18 21:55:32 +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, 2008 |
| 19 | * |
| 20 | * Authors: Dipankar Sarma <dipankar@in.ibm.com> |
| 21 | * Manfred Spraul <manfred@colorfullife.com> |
| 22 | * Paul E. McKenney <paulmck@linux.vnet.ibm.com> Hierarchical version |
| 23 | * |
| 24 | * Based on the original work by Paul McKenney <paulmck@us.ibm.com> |
| 25 | * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen. |
| 26 | * |
| 27 | * For detailed explanation of Read-Copy Update mechanism see - |
| 28 | * Documentation/RCU |
| 29 | */ |
| 30 | #include <linux/types.h> |
| 31 | #include <linux/kernel.h> |
| 32 | #include <linux/init.h> |
| 33 | #include <linux/spinlock.h> |
| 34 | #include <linux/smp.h> |
| 35 | #include <linux/rcupdate.h> |
| 36 | #include <linux/interrupt.h> |
| 37 | #include <linux/sched.h> |
| 38 | #include <asm/atomic.h> |
| 39 | #include <linux/bitops.h> |
| 40 | #include <linux/module.h> |
| 41 | #include <linux/completion.h> |
| 42 | #include <linux/moduleparam.h> |
| 43 | #include <linux/percpu.h> |
| 44 | #include <linux/notifier.h> |
| 45 | #include <linux/cpu.h> |
| 46 | #include <linux/mutex.h> |
| 47 | #include <linux/time.h> |
| 48 | |
| 49 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
| 50 | static struct lock_class_key rcu_lock_key; |
| 51 | struct lockdep_map rcu_lock_map = |
| 52 | STATIC_LOCKDEP_MAP_INIT("rcu_read_lock", &rcu_lock_key); |
| 53 | EXPORT_SYMBOL_GPL(rcu_lock_map); |
| 54 | #endif |
| 55 | |
| 56 | /* Data structures. */ |
| 57 | |
| 58 | #define RCU_STATE_INITIALIZER(name) { \ |
| 59 | .level = { &name.node[0] }, \ |
| 60 | .levelcnt = { \ |
| 61 | NUM_RCU_LVL_0, /* root of hierarchy. */ \ |
| 62 | NUM_RCU_LVL_1, \ |
| 63 | NUM_RCU_LVL_2, \ |
| 64 | NUM_RCU_LVL_3, /* == MAX_RCU_LVLS */ \ |
| 65 | }, \ |
| 66 | .signaled = RCU_SIGNAL_INIT, \ |
| 67 | .gpnum = -300, \ |
| 68 | .completed = -300, \ |
| 69 | .onofflock = __SPIN_LOCK_UNLOCKED(&name.onofflock), \ |
| 70 | .fqslock = __SPIN_LOCK_UNLOCKED(&name.fqslock), \ |
| 71 | .n_force_qs = 0, \ |
| 72 | .n_force_qs_ngp = 0, \ |
| 73 | } |
| 74 | |
Ingo Molnar | 6258c4f | 2009-03-25 16:42:24 +0100 | [diff] [blame] | 75 | struct rcu_state rcu_state = RCU_STATE_INITIALIZER(rcu_state); |
| 76 | DEFINE_PER_CPU(struct rcu_data, rcu_data); |
Paul E. McKenney | 64db4cf | 2008-12-18 21:55:32 +0100 | [diff] [blame] | 77 | |
Ingo Molnar | 6258c4f | 2009-03-25 16:42:24 +0100 | [diff] [blame] | 78 | struct rcu_state rcu_bh_state = RCU_STATE_INITIALIZER(rcu_bh_state); |
| 79 | DEFINE_PER_CPU(struct rcu_data, rcu_bh_data); |
Ingo Molnar | b1f77b0 | 2009-03-13 03:20:49 +0100 | [diff] [blame] | 80 | |
| 81 | /* |
| 82 | * Increment the quiescent state counter. |
| 83 | * The counter is a bit degenerated: We do not need to know |
| 84 | * how many quiescent states passed, just if there was at least |
| 85 | * one since the start of the grace period. Thus just a flag. |
| 86 | */ |
| 87 | void rcu_qsctr_inc(int cpu) |
| 88 | { |
| 89 | struct rcu_data *rdp = &per_cpu(rcu_data, cpu); |
| 90 | rdp->passed_quiesc = 1; |
| 91 | rdp->passed_quiesc_completed = rdp->completed; |
| 92 | } |
| 93 | |
| 94 | void rcu_bh_qsctr_inc(int cpu) |
| 95 | { |
| 96 | struct rcu_data *rdp = &per_cpu(rcu_bh_data, cpu); |
| 97 | rdp->passed_quiesc = 1; |
| 98 | rdp->passed_quiesc_completed = rdp->completed; |
| 99 | } |
Paul E. McKenney | 64db4cf | 2008-12-18 21:55:32 +0100 | [diff] [blame] | 100 | |
| 101 | #ifdef CONFIG_NO_HZ |
Paul E. McKenney | 90a4d2c | 2009-01-04 11:41:11 -0800 | [diff] [blame] | 102 | DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks) = { |
| 103 | .dynticks_nesting = 1, |
| 104 | .dynticks = 1, |
| 105 | }; |
Paul E. McKenney | 64db4cf | 2008-12-18 21:55:32 +0100 | [diff] [blame] | 106 | #endif /* #ifdef CONFIG_NO_HZ */ |
| 107 | |
| 108 | static int blimit = 10; /* Maximum callbacks per softirq. */ |
| 109 | static int qhimark = 10000; /* If this many pending, ignore blimit. */ |
| 110 | static int qlowmark = 100; /* Once only this many pending, use blimit. */ |
| 111 | |
| 112 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed); |
| 113 | |
| 114 | /* |
| 115 | * Return the number of RCU batches processed thus far for debug & stats. |
| 116 | */ |
| 117 | long rcu_batches_completed(void) |
| 118 | { |
| 119 | return rcu_state.completed; |
| 120 | } |
| 121 | EXPORT_SYMBOL_GPL(rcu_batches_completed); |
| 122 | |
| 123 | /* |
| 124 | * Return the number of RCU BH batches processed thus far for debug & stats. |
| 125 | */ |
| 126 | long rcu_batches_completed_bh(void) |
| 127 | { |
| 128 | return rcu_bh_state.completed; |
| 129 | } |
| 130 | EXPORT_SYMBOL_GPL(rcu_batches_completed_bh); |
| 131 | |
| 132 | /* |
| 133 | * Does the CPU have callbacks ready to be invoked? |
| 134 | */ |
| 135 | static int |
| 136 | cpu_has_callbacks_ready_to_invoke(struct rcu_data *rdp) |
| 137 | { |
| 138 | return &rdp->nxtlist != rdp->nxttail[RCU_DONE_TAIL]; |
| 139 | } |
| 140 | |
| 141 | /* |
| 142 | * Does the current CPU require a yet-as-unscheduled grace period? |
| 143 | */ |
| 144 | static int |
| 145 | cpu_needs_another_gp(struct rcu_state *rsp, struct rcu_data *rdp) |
| 146 | { |
| 147 | /* ACCESS_ONCE() because we are accessing outside of lock. */ |
| 148 | return *rdp->nxttail[RCU_DONE_TAIL] && |
| 149 | ACCESS_ONCE(rsp->completed) == ACCESS_ONCE(rsp->gpnum); |
| 150 | } |
| 151 | |
| 152 | /* |
| 153 | * Return the root node of the specified rcu_state structure. |
| 154 | */ |
| 155 | static struct rcu_node *rcu_get_root(struct rcu_state *rsp) |
| 156 | { |
| 157 | return &rsp->node[0]; |
| 158 | } |
| 159 | |
| 160 | #ifdef CONFIG_SMP |
| 161 | |
| 162 | /* |
| 163 | * If the specified CPU is offline, tell the caller that it is in |
| 164 | * a quiescent state. Otherwise, whack it with a reschedule IPI. |
| 165 | * Grace periods can end up waiting on an offline CPU when that |
| 166 | * CPU is in the process of coming online -- it will be added to the |
| 167 | * rcu_node bitmasks before it actually makes it online. The same thing |
| 168 | * can happen while a CPU is in the process of coming online. Because this |
| 169 | * race is quite rare, we check for it after detecting that the grace |
| 170 | * period has been delayed rather than checking each and every CPU |
| 171 | * each and every time we start a new grace period. |
| 172 | */ |
| 173 | static int rcu_implicit_offline_qs(struct rcu_data *rdp) |
| 174 | { |
| 175 | /* |
| 176 | * If the CPU is offline, it is in a quiescent state. We can |
| 177 | * trust its state not to change because interrupts are disabled. |
| 178 | */ |
| 179 | if (cpu_is_offline(rdp->cpu)) { |
| 180 | rdp->offline_fqs++; |
| 181 | return 1; |
| 182 | } |
| 183 | |
| 184 | /* The CPU is online, so send it a reschedule IPI. */ |
| 185 | if (rdp->cpu != smp_processor_id()) |
| 186 | smp_send_reschedule(rdp->cpu); |
| 187 | else |
| 188 | set_need_resched(); |
| 189 | rdp->resched_ipi++; |
| 190 | return 0; |
| 191 | } |
| 192 | |
| 193 | #endif /* #ifdef CONFIG_SMP */ |
| 194 | |
| 195 | #ifdef CONFIG_NO_HZ |
| 196 | static DEFINE_RATELIMIT_STATE(rcu_rs, 10 * HZ, 5); |
| 197 | |
| 198 | /** |
| 199 | * rcu_enter_nohz - inform RCU that current CPU is entering nohz |
| 200 | * |
| 201 | * Enter nohz mode, in other words, -leave- the mode in which RCU |
| 202 | * read-side critical sections can occur. (Though RCU read-side |
| 203 | * critical sections can occur in irq handlers in nohz mode, a possibility |
| 204 | * handled by rcu_irq_enter() and rcu_irq_exit()). |
| 205 | */ |
| 206 | void rcu_enter_nohz(void) |
| 207 | { |
| 208 | unsigned long flags; |
| 209 | struct rcu_dynticks *rdtp; |
| 210 | |
| 211 | smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */ |
| 212 | local_irq_save(flags); |
| 213 | rdtp = &__get_cpu_var(rcu_dynticks); |
| 214 | rdtp->dynticks++; |
| 215 | rdtp->dynticks_nesting--; |
| 216 | WARN_ON_RATELIMIT(rdtp->dynticks & 0x1, &rcu_rs); |
| 217 | local_irq_restore(flags); |
| 218 | } |
| 219 | |
| 220 | /* |
| 221 | * rcu_exit_nohz - inform RCU that current CPU is leaving nohz |
| 222 | * |
| 223 | * Exit nohz mode, in other words, -enter- the mode in which RCU |
| 224 | * read-side critical sections normally occur. |
| 225 | */ |
| 226 | void rcu_exit_nohz(void) |
| 227 | { |
| 228 | unsigned long flags; |
| 229 | struct rcu_dynticks *rdtp; |
| 230 | |
| 231 | local_irq_save(flags); |
| 232 | rdtp = &__get_cpu_var(rcu_dynticks); |
| 233 | rdtp->dynticks++; |
| 234 | rdtp->dynticks_nesting++; |
| 235 | WARN_ON_RATELIMIT(!(rdtp->dynticks & 0x1), &rcu_rs); |
| 236 | local_irq_restore(flags); |
| 237 | smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */ |
| 238 | } |
| 239 | |
| 240 | /** |
| 241 | * rcu_nmi_enter - inform RCU of entry to NMI context |
| 242 | * |
| 243 | * If the CPU was idle with dynamic ticks active, and there is no |
| 244 | * irq handler running, this updates rdtp->dynticks_nmi to let the |
| 245 | * RCU grace-period handling know that the CPU is active. |
| 246 | */ |
| 247 | void rcu_nmi_enter(void) |
| 248 | { |
| 249 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); |
| 250 | |
| 251 | if (rdtp->dynticks & 0x1) |
| 252 | return; |
| 253 | rdtp->dynticks_nmi++; |
| 254 | WARN_ON_RATELIMIT(!(rdtp->dynticks_nmi & 0x1), &rcu_rs); |
| 255 | smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */ |
| 256 | } |
| 257 | |
| 258 | /** |
| 259 | * rcu_nmi_exit - inform RCU of exit from NMI context |
| 260 | * |
| 261 | * If the CPU was idle with dynamic ticks active, and there is no |
| 262 | * irq handler running, this updates rdtp->dynticks_nmi to let the |
| 263 | * RCU grace-period handling know that the CPU is no longer active. |
| 264 | */ |
| 265 | void rcu_nmi_exit(void) |
| 266 | { |
| 267 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); |
| 268 | |
| 269 | if (rdtp->dynticks & 0x1) |
| 270 | return; |
| 271 | smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */ |
| 272 | rdtp->dynticks_nmi++; |
| 273 | WARN_ON_RATELIMIT(rdtp->dynticks_nmi & 0x1, &rcu_rs); |
| 274 | } |
| 275 | |
| 276 | /** |
| 277 | * rcu_irq_enter - inform RCU of entry to hard irq context |
| 278 | * |
| 279 | * If the CPU was idle with dynamic ticks active, this updates the |
| 280 | * rdtp->dynticks to let the RCU handling know that the CPU is active. |
| 281 | */ |
| 282 | void rcu_irq_enter(void) |
| 283 | { |
| 284 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); |
| 285 | |
| 286 | if (rdtp->dynticks_nesting++) |
| 287 | return; |
| 288 | rdtp->dynticks++; |
| 289 | WARN_ON_RATELIMIT(!(rdtp->dynticks & 0x1), &rcu_rs); |
| 290 | smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */ |
| 291 | } |
| 292 | |
| 293 | /** |
| 294 | * rcu_irq_exit - inform RCU of exit from hard irq context |
| 295 | * |
| 296 | * If the CPU was idle with dynamic ticks active, update the rdp->dynticks |
| 297 | * to put let the RCU handling be aware that the CPU is going back to idle |
| 298 | * with no ticks. |
| 299 | */ |
| 300 | void rcu_irq_exit(void) |
| 301 | { |
| 302 | struct rcu_dynticks *rdtp = &__get_cpu_var(rcu_dynticks); |
| 303 | |
| 304 | if (--rdtp->dynticks_nesting) |
| 305 | return; |
| 306 | smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */ |
| 307 | rdtp->dynticks++; |
| 308 | WARN_ON_RATELIMIT(rdtp->dynticks & 0x1, &rcu_rs); |
| 309 | |
| 310 | /* If the interrupt queued a callback, get out of dyntick mode. */ |
| 311 | if (__get_cpu_var(rcu_data).nxtlist || |
| 312 | __get_cpu_var(rcu_bh_data).nxtlist) |
| 313 | set_need_resched(); |
| 314 | } |
| 315 | |
| 316 | /* |
| 317 | * Record the specified "completed" value, which is later used to validate |
| 318 | * dynticks counter manipulations. Specify "rsp->completed - 1" to |
| 319 | * unconditionally invalidate any future dynticks manipulations (which is |
| 320 | * useful at the beginning of a grace period). |
| 321 | */ |
| 322 | static void dyntick_record_completed(struct rcu_state *rsp, long comp) |
| 323 | { |
| 324 | rsp->dynticks_completed = comp; |
| 325 | } |
| 326 | |
| 327 | #ifdef CONFIG_SMP |
| 328 | |
| 329 | /* |
| 330 | * Recall the previously recorded value of the completion for dynticks. |
| 331 | */ |
| 332 | static long dyntick_recall_completed(struct rcu_state *rsp) |
| 333 | { |
| 334 | return rsp->dynticks_completed; |
| 335 | } |
| 336 | |
| 337 | /* |
| 338 | * Snapshot the specified CPU's dynticks counter so that we can later |
| 339 | * credit them with an implicit quiescent state. Return 1 if this CPU |
| 340 | * is already in a quiescent state courtesy of dynticks idle mode. |
| 341 | */ |
| 342 | static int dyntick_save_progress_counter(struct rcu_data *rdp) |
| 343 | { |
| 344 | int ret; |
| 345 | int snap; |
| 346 | int snap_nmi; |
| 347 | |
| 348 | snap = rdp->dynticks->dynticks; |
| 349 | snap_nmi = rdp->dynticks->dynticks_nmi; |
| 350 | smp_mb(); /* Order sampling of snap with end of grace period. */ |
| 351 | rdp->dynticks_snap = snap; |
| 352 | rdp->dynticks_nmi_snap = snap_nmi; |
| 353 | ret = ((snap & 0x1) == 0) && ((snap_nmi & 0x1) == 0); |
| 354 | if (ret) |
| 355 | rdp->dynticks_fqs++; |
| 356 | return ret; |
| 357 | } |
| 358 | |
| 359 | /* |
| 360 | * Return true if the specified CPU has passed through a quiescent |
| 361 | * state by virtue of being in or having passed through an dynticks |
| 362 | * idle state since the last call to dyntick_save_progress_counter() |
| 363 | * for this same CPU. |
| 364 | */ |
| 365 | static int rcu_implicit_dynticks_qs(struct rcu_data *rdp) |
| 366 | { |
| 367 | long curr; |
| 368 | long curr_nmi; |
| 369 | long snap; |
| 370 | long snap_nmi; |
| 371 | |
| 372 | curr = rdp->dynticks->dynticks; |
| 373 | snap = rdp->dynticks_snap; |
| 374 | curr_nmi = rdp->dynticks->dynticks_nmi; |
| 375 | snap_nmi = rdp->dynticks_nmi_snap; |
| 376 | smp_mb(); /* force ordering with cpu entering/leaving dynticks. */ |
| 377 | |
| 378 | /* |
| 379 | * If the CPU passed through or entered a dynticks idle phase with |
| 380 | * no active irq/NMI handlers, then we can safely pretend that the CPU |
| 381 | * already acknowledged the request to pass through a quiescent |
| 382 | * state. Either way, that CPU cannot possibly be in an RCU |
| 383 | * read-side critical section that started before the beginning |
| 384 | * of the current RCU grace period. |
| 385 | */ |
| 386 | if ((curr != snap || (curr & 0x1) == 0) && |
| 387 | (curr_nmi != snap_nmi || (curr_nmi & 0x1) == 0)) { |
| 388 | rdp->dynticks_fqs++; |
| 389 | return 1; |
| 390 | } |
| 391 | |
| 392 | /* Go check for the CPU being offline. */ |
| 393 | return rcu_implicit_offline_qs(rdp); |
| 394 | } |
| 395 | |
| 396 | #endif /* #ifdef CONFIG_SMP */ |
| 397 | |
| 398 | #else /* #ifdef CONFIG_NO_HZ */ |
| 399 | |
| 400 | static void dyntick_record_completed(struct rcu_state *rsp, long comp) |
| 401 | { |
| 402 | } |
| 403 | |
| 404 | #ifdef CONFIG_SMP |
| 405 | |
| 406 | /* |
| 407 | * If there are no dynticks, then the only way that a CPU can passively |
| 408 | * be in a quiescent state is to be offline. Unlike dynticks idle, which |
| 409 | * is a point in time during the prior (already finished) grace period, |
| 410 | * an offline CPU is always in a quiescent state, and thus can be |
| 411 | * unconditionally applied. So just return the current value of completed. |
| 412 | */ |
| 413 | static long dyntick_recall_completed(struct rcu_state *rsp) |
| 414 | { |
| 415 | return rsp->completed; |
| 416 | } |
| 417 | |
| 418 | static int dyntick_save_progress_counter(struct rcu_data *rdp) |
| 419 | { |
| 420 | return 0; |
| 421 | } |
| 422 | |
| 423 | static int rcu_implicit_dynticks_qs(struct rcu_data *rdp) |
| 424 | { |
| 425 | return rcu_implicit_offline_qs(rdp); |
| 426 | } |
| 427 | |
| 428 | #endif /* #ifdef CONFIG_SMP */ |
| 429 | |
| 430 | #endif /* #else #ifdef CONFIG_NO_HZ */ |
| 431 | |
| 432 | #ifdef CONFIG_RCU_CPU_STALL_DETECTOR |
| 433 | |
| 434 | static void record_gp_stall_check_time(struct rcu_state *rsp) |
| 435 | { |
| 436 | rsp->gp_start = jiffies; |
| 437 | rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_CHECK; |
| 438 | } |
| 439 | |
| 440 | static void print_other_cpu_stall(struct rcu_state *rsp) |
| 441 | { |
| 442 | int cpu; |
| 443 | long delta; |
| 444 | unsigned long flags; |
| 445 | struct rcu_node *rnp = rcu_get_root(rsp); |
| 446 | struct rcu_node *rnp_cur = rsp->level[NUM_RCU_LVLS - 1]; |
| 447 | struct rcu_node *rnp_end = &rsp->node[NUM_RCU_NODES]; |
| 448 | |
| 449 | /* Only let one CPU complain about others per time interval. */ |
| 450 | |
| 451 | spin_lock_irqsave(&rnp->lock, flags); |
| 452 | delta = jiffies - rsp->jiffies_stall; |
| 453 | if (delta < RCU_STALL_RAT_DELAY || rsp->gpnum == rsp->completed) { |
| 454 | spin_unlock_irqrestore(&rnp->lock, flags); |
| 455 | return; |
| 456 | } |
| 457 | rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_RECHECK; |
| 458 | spin_unlock_irqrestore(&rnp->lock, flags); |
| 459 | |
| 460 | /* OK, time to rat on our buddy... */ |
| 461 | |
| 462 | printk(KERN_ERR "INFO: RCU detected CPU stalls:"); |
| 463 | for (; rnp_cur < rnp_end; rnp_cur++) { |
| 464 | if (rnp_cur->qsmask == 0) |
| 465 | continue; |
| 466 | for (cpu = 0; cpu <= rnp_cur->grphi - rnp_cur->grplo; cpu++) |
| 467 | if (rnp_cur->qsmask & (1UL << cpu)) |
| 468 | printk(" %d", rnp_cur->grplo + cpu); |
| 469 | } |
| 470 | printk(" (detected by %d, t=%ld jiffies)\n", |
| 471 | smp_processor_id(), (long)(jiffies - rsp->gp_start)); |
| 472 | force_quiescent_state(rsp, 0); /* Kick them all. */ |
| 473 | } |
| 474 | |
| 475 | static void print_cpu_stall(struct rcu_state *rsp) |
| 476 | { |
| 477 | unsigned long flags; |
| 478 | struct rcu_node *rnp = rcu_get_root(rsp); |
| 479 | |
| 480 | printk(KERN_ERR "INFO: RCU detected CPU %d stall (t=%lu jiffies)\n", |
| 481 | smp_processor_id(), jiffies - rsp->gp_start); |
| 482 | dump_stack(); |
| 483 | spin_lock_irqsave(&rnp->lock, flags); |
| 484 | if ((long)(jiffies - rsp->jiffies_stall) >= 0) |
| 485 | rsp->jiffies_stall = |
| 486 | jiffies + RCU_SECONDS_TILL_STALL_RECHECK; |
| 487 | spin_unlock_irqrestore(&rnp->lock, flags); |
| 488 | set_need_resched(); /* kick ourselves to get things going. */ |
| 489 | } |
| 490 | |
| 491 | static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp) |
| 492 | { |
| 493 | long delta; |
| 494 | struct rcu_node *rnp; |
| 495 | |
| 496 | delta = jiffies - rsp->jiffies_stall; |
| 497 | rnp = rdp->mynode; |
| 498 | if ((rnp->qsmask & rdp->grpmask) && delta >= 0) { |
| 499 | |
| 500 | /* We haven't checked in, so go dump stack. */ |
| 501 | print_cpu_stall(rsp); |
| 502 | |
| 503 | } else if (rsp->gpnum != rsp->completed && |
| 504 | delta >= RCU_STALL_RAT_DELAY) { |
| 505 | |
| 506 | /* They had two time units to dump stack, so complain. */ |
| 507 | print_other_cpu_stall(rsp); |
| 508 | } |
| 509 | } |
| 510 | |
| 511 | #else /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */ |
| 512 | |
| 513 | static void record_gp_stall_check_time(struct rcu_state *rsp) |
| 514 | { |
| 515 | } |
| 516 | |
| 517 | static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp) |
| 518 | { |
| 519 | } |
| 520 | |
| 521 | #endif /* #else #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */ |
| 522 | |
| 523 | /* |
| 524 | * Update CPU-local rcu_data state to record the newly noticed grace period. |
| 525 | * This is used both when we started the grace period and when we notice |
| 526 | * that someone else started the grace period. |
| 527 | */ |
| 528 | static void note_new_gpnum(struct rcu_state *rsp, struct rcu_data *rdp) |
| 529 | { |
| 530 | rdp->qs_pending = 1; |
| 531 | rdp->passed_quiesc = 0; |
| 532 | rdp->gpnum = rsp->gpnum; |
Paul E. McKenney | 64db4cf | 2008-12-18 21:55:32 +0100 | [diff] [blame] | 533 | } |
| 534 | |
| 535 | /* |
| 536 | * Did someone else start a new RCU grace period start since we last |
| 537 | * checked? Update local state appropriately if so. Must be called |
| 538 | * on the CPU corresponding to rdp. |
| 539 | */ |
| 540 | static int |
| 541 | check_for_new_grace_period(struct rcu_state *rsp, struct rcu_data *rdp) |
| 542 | { |
| 543 | unsigned long flags; |
| 544 | int ret = 0; |
| 545 | |
| 546 | local_irq_save(flags); |
| 547 | if (rdp->gpnum != rsp->gpnum) { |
| 548 | note_new_gpnum(rsp, rdp); |
| 549 | ret = 1; |
| 550 | } |
| 551 | local_irq_restore(flags); |
| 552 | return ret; |
| 553 | } |
| 554 | |
| 555 | /* |
| 556 | * Start a new RCU grace period if warranted, re-initializing the hierarchy |
| 557 | * in preparation for detecting the next grace period. The caller must hold |
| 558 | * the root node's ->lock, which is released before return. Hard irqs must |
| 559 | * be disabled. |
| 560 | */ |
| 561 | static void |
| 562 | rcu_start_gp(struct rcu_state *rsp, unsigned long flags) |
| 563 | __releases(rcu_get_root(rsp)->lock) |
| 564 | { |
| 565 | struct rcu_data *rdp = rsp->rda[smp_processor_id()]; |
| 566 | struct rcu_node *rnp = rcu_get_root(rsp); |
| 567 | struct rcu_node *rnp_cur; |
| 568 | struct rcu_node *rnp_end; |
| 569 | |
| 570 | if (!cpu_needs_another_gp(rsp, rdp)) { |
| 571 | spin_unlock_irqrestore(&rnp->lock, flags); |
| 572 | return; |
| 573 | } |
| 574 | |
| 575 | /* Advance to a new grace period and initialize state. */ |
| 576 | rsp->gpnum++; |
| 577 | rsp->signaled = RCU_GP_INIT; /* Hold off force_quiescent_state. */ |
| 578 | rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS; |
Paul E. McKenney | 64db4cf | 2008-12-18 21:55:32 +0100 | [diff] [blame] | 579 | record_gp_stall_check_time(rsp); |
| 580 | dyntick_record_completed(rsp, rsp->completed - 1); |
| 581 | note_new_gpnum(rsp, rdp); |
| 582 | |
| 583 | /* |
| 584 | * Because we are first, we know that all our callbacks will |
| 585 | * be covered by this upcoming grace period, even the ones |
| 586 | * that were registered arbitrarily recently. |
| 587 | */ |
| 588 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; |
| 589 | rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; |
| 590 | |
| 591 | /* Special-case the common single-level case. */ |
| 592 | if (NUM_RCU_NODES == 1) { |
| 593 | rnp->qsmask = rnp->qsmaskinit; |
Paul E. McKenney | c12172c | 2009-01-04 20:30:06 -0800 | [diff] [blame] | 594 | rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state OK. */ |
Paul E. McKenney | 64db4cf | 2008-12-18 21:55:32 +0100 | [diff] [blame] | 595 | spin_unlock_irqrestore(&rnp->lock, flags); |
| 596 | return; |
| 597 | } |
| 598 | |
| 599 | spin_unlock(&rnp->lock); /* leave irqs disabled. */ |
| 600 | |
| 601 | |
| 602 | /* Exclude any concurrent CPU-hotplug operations. */ |
| 603 | spin_lock(&rsp->onofflock); /* irqs already disabled. */ |
| 604 | |
| 605 | /* |
| 606 | * Set the quiescent-state-needed bits in all the non-leaf RCU |
| 607 | * nodes for all currently online CPUs. This operation relies |
| 608 | * on the layout of the hierarchy within the rsp->node[] array. |
| 609 | * Note that other CPUs will access only the leaves of the |
| 610 | * hierarchy, which still indicate that no grace period is in |
| 611 | * progress. In addition, we have excluded CPU-hotplug operations. |
| 612 | * |
| 613 | * We therefore do not need to hold any locks. Any required |
| 614 | * memory barriers will be supplied by the locks guarding the |
| 615 | * leaf rcu_nodes in the hierarchy. |
| 616 | */ |
| 617 | |
| 618 | rnp_end = rsp->level[NUM_RCU_LVLS - 1]; |
| 619 | for (rnp_cur = &rsp->node[0]; rnp_cur < rnp_end; rnp_cur++) |
| 620 | rnp_cur->qsmask = rnp_cur->qsmaskinit; |
| 621 | |
| 622 | /* |
| 623 | * Now set up the leaf nodes. Here we must be careful. First, |
| 624 | * we need to hold the lock in order to exclude other CPUs, which |
| 625 | * might be contending for the leaf nodes' locks. Second, as |
| 626 | * soon as we initialize a given leaf node, its CPUs might run |
| 627 | * up the rest of the hierarchy. We must therefore acquire locks |
| 628 | * for each node that we touch during this stage. (But we still |
| 629 | * are excluding CPU-hotplug operations.) |
| 630 | * |
| 631 | * Note that the grace period cannot complete until we finish |
| 632 | * the initialization process, as there will be at least one |
| 633 | * qsmask bit set in the root node until that time, namely the |
| 634 | * one corresponding to this CPU. |
| 635 | */ |
| 636 | rnp_end = &rsp->node[NUM_RCU_NODES]; |
| 637 | rnp_cur = rsp->level[NUM_RCU_LVLS - 1]; |
| 638 | for (; rnp_cur < rnp_end; rnp_cur++) { |
| 639 | spin_lock(&rnp_cur->lock); /* irqs already disabled. */ |
| 640 | rnp_cur->qsmask = rnp_cur->qsmaskinit; |
| 641 | spin_unlock(&rnp_cur->lock); /* irqs already disabled. */ |
| 642 | } |
| 643 | |
| 644 | rsp->signaled = RCU_SIGNAL_INIT; /* force_quiescent_state now OK. */ |
| 645 | spin_unlock_irqrestore(&rsp->onofflock, flags); |
| 646 | } |
| 647 | |
| 648 | /* |
| 649 | * Advance this CPU's callbacks, but only if the current grace period |
| 650 | * has ended. This may be called only from the CPU to whom the rdp |
| 651 | * belongs. |
| 652 | */ |
| 653 | static void |
| 654 | rcu_process_gp_end(struct rcu_state *rsp, struct rcu_data *rdp) |
| 655 | { |
| 656 | long completed_snap; |
| 657 | unsigned long flags; |
| 658 | |
| 659 | local_irq_save(flags); |
| 660 | completed_snap = ACCESS_ONCE(rsp->completed); /* outside of lock. */ |
| 661 | |
| 662 | /* Did another grace period end? */ |
| 663 | if (rdp->completed != completed_snap) { |
| 664 | |
| 665 | /* Advance callbacks. No harm if list empty. */ |
| 666 | rdp->nxttail[RCU_DONE_TAIL] = rdp->nxttail[RCU_WAIT_TAIL]; |
| 667 | rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_READY_TAIL]; |
| 668 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; |
| 669 | |
| 670 | /* Remember that we saw this grace-period completion. */ |
| 671 | rdp->completed = completed_snap; |
| 672 | } |
| 673 | local_irq_restore(flags); |
| 674 | } |
| 675 | |
| 676 | /* |
| 677 | * Similar to cpu_quiet(), for which it is a helper function. Allows |
| 678 | * a group of CPUs to be quieted at one go, though all the CPUs in the |
| 679 | * group must be represented by the same leaf rcu_node structure. |
| 680 | * That structure's lock must be held upon entry, and it is released |
| 681 | * before return. |
| 682 | */ |
| 683 | static void |
| 684 | cpu_quiet_msk(unsigned long mask, struct rcu_state *rsp, struct rcu_node *rnp, |
| 685 | unsigned long flags) |
| 686 | __releases(rnp->lock) |
| 687 | { |
| 688 | /* Walk up the rcu_node hierarchy. */ |
| 689 | for (;;) { |
| 690 | if (!(rnp->qsmask & mask)) { |
| 691 | |
| 692 | /* Our bit has already been cleared, so done. */ |
| 693 | spin_unlock_irqrestore(&rnp->lock, flags); |
| 694 | return; |
| 695 | } |
| 696 | rnp->qsmask &= ~mask; |
| 697 | if (rnp->qsmask != 0) { |
| 698 | |
| 699 | /* Other bits still set at this level, so done. */ |
| 700 | spin_unlock_irqrestore(&rnp->lock, flags); |
| 701 | return; |
| 702 | } |
| 703 | mask = rnp->grpmask; |
| 704 | if (rnp->parent == NULL) { |
| 705 | |
| 706 | /* No more levels. Exit loop holding root lock. */ |
| 707 | |
| 708 | break; |
| 709 | } |
| 710 | spin_unlock_irqrestore(&rnp->lock, flags); |
| 711 | rnp = rnp->parent; |
| 712 | spin_lock_irqsave(&rnp->lock, flags); |
| 713 | } |
| 714 | |
| 715 | /* |
| 716 | * Get here if we are the last CPU to pass through a quiescent |
| 717 | * state for this grace period. Clean up and let rcu_start_gp() |
| 718 | * start up the next grace period if one is needed. Note that |
| 719 | * we still hold rnp->lock, as required by rcu_start_gp(), which |
| 720 | * will release it. |
| 721 | */ |
| 722 | rsp->completed = rsp->gpnum; |
| 723 | rcu_process_gp_end(rsp, rsp->rda[smp_processor_id()]); |
| 724 | rcu_start_gp(rsp, flags); /* releases rnp->lock. */ |
| 725 | } |
| 726 | |
| 727 | /* |
| 728 | * Record a quiescent state for the specified CPU, which must either be |
| 729 | * the current CPU or an offline CPU. The lastcomp argument is used to |
| 730 | * make sure we are still in the grace period of interest. We don't want |
| 731 | * to end the current grace period based on quiescent states detected in |
| 732 | * an earlier grace period! |
| 733 | */ |
| 734 | static void |
| 735 | cpu_quiet(int cpu, struct rcu_state *rsp, struct rcu_data *rdp, long lastcomp) |
| 736 | { |
| 737 | unsigned long flags; |
| 738 | unsigned long mask; |
| 739 | struct rcu_node *rnp; |
| 740 | |
| 741 | rnp = rdp->mynode; |
| 742 | spin_lock_irqsave(&rnp->lock, flags); |
| 743 | if (lastcomp != ACCESS_ONCE(rsp->completed)) { |
| 744 | |
| 745 | /* |
| 746 | * Someone beat us to it for this grace period, so leave. |
| 747 | * The race with GP start is resolved by the fact that we |
| 748 | * hold the leaf rcu_node lock, so that the per-CPU bits |
| 749 | * cannot yet be initialized -- so we would simply find our |
| 750 | * CPU's bit already cleared in cpu_quiet_msk() if this race |
| 751 | * occurred. |
| 752 | */ |
| 753 | rdp->passed_quiesc = 0; /* try again later! */ |
| 754 | spin_unlock_irqrestore(&rnp->lock, flags); |
| 755 | return; |
| 756 | } |
| 757 | mask = rdp->grpmask; |
| 758 | if ((rnp->qsmask & mask) == 0) { |
| 759 | spin_unlock_irqrestore(&rnp->lock, flags); |
| 760 | } else { |
| 761 | rdp->qs_pending = 0; |
| 762 | |
| 763 | /* |
| 764 | * This GP can't end until cpu checks in, so all of our |
| 765 | * callbacks can be processed during the next GP. |
| 766 | */ |
| 767 | rdp = rsp->rda[smp_processor_id()]; |
| 768 | rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; |
| 769 | |
| 770 | cpu_quiet_msk(mask, rsp, rnp, flags); /* releases rnp->lock */ |
| 771 | } |
| 772 | } |
| 773 | |
| 774 | /* |
| 775 | * Check to see if there is a new grace period of which this CPU |
| 776 | * is not yet aware, and if so, set up local rcu_data state for it. |
| 777 | * Otherwise, see if this CPU has just passed through its first |
| 778 | * quiescent state for this grace period, and record that fact if so. |
| 779 | */ |
| 780 | static void |
| 781 | rcu_check_quiescent_state(struct rcu_state *rsp, struct rcu_data *rdp) |
| 782 | { |
| 783 | /* If there is now a new grace period, record and return. */ |
| 784 | if (check_for_new_grace_period(rsp, rdp)) |
| 785 | return; |
| 786 | |
| 787 | /* |
| 788 | * Does this CPU still need to do its part for current grace period? |
| 789 | * If no, return and let the other CPUs do their part as well. |
| 790 | */ |
| 791 | if (!rdp->qs_pending) |
| 792 | return; |
| 793 | |
| 794 | /* |
| 795 | * Was there a quiescent state since the beginning of the grace |
| 796 | * period? If no, then exit and wait for the next call. |
| 797 | */ |
| 798 | if (!rdp->passed_quiesc) |
| 799 | return; |
| 800 | |
| 801 | /* Tell RCU we are done (but cpu_quiet() will be the judge of that). */ |
| 802 | cpu_quiet(rdp->cpu, rsp, rdp, rdp->passed_quiesc_completed); |
| 803 | } |
| 804 | |
| 805 | #ifdef CONFIG_HOTPLUG_CPU |
| 806 | |
| 807 | /* |
| 808 | * Remove the outgoing CPU from the bitmasks in the rcu_node hierarchy |
| 809 | * and move all callbacks from the outgoing CPU to the current one. |
| 810 | */ |
| 811 | static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp) |
| 812 | { |
| 813 | int i; |
| 814 | unsigned long flags; |
| 815 | long lastcomp; |
| 816 | unsigned long mask; |
| 817 | struct rcu_data *rdp = rsp->rda[cpu]; |
| 818 | struct rcu_data *rdp_me; |
| 819 | struct rcu_node *rnp; |
| 820 | |
| 821 | /* Exclude any attempts to start a new grace period. */ |
| 822 | spin_lock_irqsave(&rsp->onofflock, flags); |
| 823 | |
| 824 | /* Remove the outgoing CPU from the masks in the rcu_node hierarchy. */ |
| 825 | rnp = rdp->mynode; |
| 826 | mask = rdp->grpmask; /* rnp->grplo is constant. */ |
| 827 | do { |
| 828 | spin_lock(&rnp->lock); /* irqs already disabled. */ |
| 829 | rnp->qsmaskinit &= ~mask; |
| 830 | if (rnp->qsmaskinit != 0) { |
| 831 | spin_unlock(&rnp->lock); /* irqs already disabled. */ |
| 832 | break; |
| 833 | } |
| 834 | mask = rnp->grpmask; |
| 835 | spin_unlock(&rnp->lock); /* irqs already disabled. */ |
| 836 | rnp = rnp->parent; |
| 837 | } while (rnp != NULL); |
| 838 | lastcomp = rsp->completed; |
| 839 | |
| 840 | spin_unlock(&rsp->onofflock); /* irqs remain disabled. */ |
| 841 | |
| 842 | /* Being offline is a quiescent state, so go record it. */ |
| 843 | cpu_quiet(cpu, rsp, rdp, lastcomp); |
| 844 | |
| 845 | /* |
| 846 | * Move callbacks from the outgoing CPU to the running CPU. |
| 847 | * Note that the outgoing CPU is now quiscent, so it is now |
| 848 | * (uncharacteristically) safe to access it rcu_data structure. |
| 849 | * Note also that we must carefully retain the order of the |
| 850 | * outgoing CPU's callbacks in order for rcu_barrier() to work |
| 851 | * correctly. Finally, note that we start all the callbacks |
| 852 | * afresh, even those that have passed through a grace period |
| 853 | * and are therefore ready to invoke. The theory is that hotplug |
| 854 | * events are rare, and that if they are frequent enough to |
| 855 | * indefinitely delay callbacks, you have far worse things to |
| 856 | * be worrying about. |
| 857 | */ |
| 858 | rdp_me = rsp->rda[smp_processor_id()]; |
| 859 | if (rdp->nxtlist != NULL) { |
| 860 | *rdp_me->nxttail[RCU_NEXT_TAIL] = rdp->nxtlist; |
| 861 | rdp_me->nxttail[RCU_NEXT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; |
| 862 | rdp->nxtlist = NULL; |
| 863 | for (i = 0; i < RCU_NEXT_SIZE; i++) |
| 864 | rdp->nxttail[i] = &rdp->nxtlist; |
| 865 | rdp_me->qlen += rdp->qlen; |
| 866 | rdp->qlen = 0; |
| 867 | } |
| 868 | local_irq_restore(flags); |
| 869 | } |
| 870 | |
| 871 | /* |
| 872 | * Remove the specified CPU from the RCU hierarchy and move any pending |
| 873 | * callbacks that it might have to the current CPU. This code assumes |
| 874 | * that at least one CPU in the system will remain running at all times. |
| 875 | * Any attempt to offline -all- CPUs is likely to strand RCU callbacks. |
| 876 | */ |
| 877 | static void rcu_offline_cpu(int cpu) |
| 878 | { |
| 879 | __rcu_offline_cpu(cpu, &rcu_state); |
| 880 | __rcu_offline_cpu(cpu, &rcu_bh_state); |
| 881 | } |
| 882 | |
| 883 | #else /* #ifdef CONFIG_HOTPLUG_CPU */ |
| 884 | |
| 885 | static void rcu_offline_cpu(int cpu) |
| 886 | { |
| 887 | } |
| 888 | |
| 889 | #endif /* #else #ifdef CONFIG_HOTPLUG_CPU */ |
| 890 | |
| 891 | /* |
| 892 | * Invoke any RCU callbacks that have made it to the end of their grace |
| 893 | * period. Thottle as specified by rdp->blimit. |
| 894 | */ |
| 895 | static void rcu_do_batch(struct rcu_data *rdp) |
| 896 | { |
| 897 | unsigned long flags; |
| 898 | struct rcu_head *next, *list, **tail; |
| 899 | int count; |
| 900 | |
| 901 | /* If no callbacks are ready, just return.*/ |
| 902 | if (!cpu_has_callbacks_ready_to_invoke(rdp)) |
| 903 | return; |
| 904 | |
| 905 | /* |
| 906 | * Extract the list of ready callbacks, disabling to prevent |
| 907 | * races with call_rcu() from interrupt handlers. |
| 908 | */ |
| 909 | local_irq_save(flags); |
| 910 | list = rdp->nxtlist; |
| 911 | rdp->nxtlist = *rdp->nxttail[RCU_DONE_TAIL]; |
| 912 | *rdp->nxttail[RCU_DONE_TAIL] = NULL; |
| 913 | tail = rdp->nxttail[RCU_DONE_TAIL]; |
| 914 | for (count = RCU_NEXT_SIZE - 1; count >= 0; count--) |
| 915 | if (rdp->nxttail[count] == rdp->nxttail[RCU_DONE_TAIL]) |
| 916 | rdp->nxttail[count] = &rdp->nxtlist; |
| 917 | local_irq_restore(flags); |
| 918 | |
| 919 | /* Invoke callbacks. */ |
| 920 | count = 0; |
| 921 | while (list) { |
| 922 | next = list->next; |
| 923 | prefetch(next); |
| 924 | list->func(list); |
| 925 | list = next; |
| 926 | if (++count >= rdp->blimit) |
| 927 | break; |
| 928 | } |
| 929 | |
| 930 | local_irq_save(flags); |
| 931 | |
| 932 | /* Update count, and requeue any remaining callbacks. */ |
| 933 | rdp->qlen -= count; |
| 934 | if (list != NULL) { |
| 935 | *tail = rdp->nxtlist; |
| 936 | rdp->nxtlist = list; |
| 937 | for (count = 0; count < RCU_NEXT_SIZE; count++) |
| 938 | if (&rdp->nxtlist == rdp->nxttail[count]) |
| 939 | rdp->nxttail[count] = tail; |
| 940 | else |
| 941 | break; |
| 942 | } |
| 943 | |
| 944 | /* Reinstate batch limit if we have worked down the excess. */ |
| 945 | if (rdp->blimit == LONG_MAX && rdp->qlen <= qlowmark) |
| 946 | rdp->blimit = blimit; |
| 947 | |
| 948 | local_irq_restore(flags); |
| 949 | |
| 950 | /* Re-raise the RCU softirq if there are callbacks remaining. */ |
| 951 | if (cpu_has_callbacks_ready_to_invoke(rdp)) |
| 952 | raise_softirq(RCU_SOFTIRQ); |
| 953 | } |
| 954 | |
| 955 | /* |
| 956 | * Check to see if this CPU is in a non-context-switch quiescent state |
| 957 | * (user mode or idle loop for rcu, non-softirq execution for rcu_bh). |
| 958 | * Also schedule the RCU softirq handler. |
| 959 | * |
| 960 | * This function must be called with hardirqs disabled. It is normally |
| 961 | * invoked from the scheduling-clock interrupt. If rcu_pending returns |
| 962 | * false, there is no point in invoking rcu_check_callbacks(). |
| 963 | */ |
| 964 | void rcu_check_callbacks(int cpu, int user) |
| 965 | { |
| 966 | if (user || |
Paul E. McKenney | a682604 | 2009-02-25 18:03:42 -0800 | [diff] [blame] | 967 | (idle_cpu(cpu) && rcu_scheduler_active && |
| 968 | !in_softirq() && hardirq_count() <= (1 << HARDIRQ_SHIFT))) { |
Paul E. McKenney | 64db4cf | 2008-12-18 21:55:32 +0100 | [diff] [blame] | 969 | |
| 970 | /* |
| 971 | * Get here if this CPU took its interrupt from user |
| 972 | * mode or from the idle loop, and if this is not a |
| 973 | * nested interrupt. In this case, the CPU is in |
| 974 | * a quiescent state, so count it. |
| 975 | * |
| 976 | * No memory barrier is required here because both |
| 977 | * rcu_qsctr_inc() and rcu_bh_qsctr_inc() reference |
| 978 | * only CPU-local variables that other CPUs neither |
| 979 | * access nor modify, at least not while the corresponding |
| 980 | * CPU is online. |
| 981 | */ |
| 982 | |
| 983 | rcu_qsctr_inc(cpu); |
| 984 | rcu_bh_qsctr_inc(cpu); |
| 985 | |
| 986 | } else if (!in_softirq()) { |
| 987 | |
| 988 | /* |
| 989 | * Get here if this CPU did not take its interrupt from |
| 990 | * softirq, in other words, if it is not interrupting |
| 991 | * a rcu_bh read-side critical section. This is an _bh |
| 992 | * critical section, so count it. |
| 993 | */ |
| 994 | |
| 995 | rcu_bh_qsctr_inc(cpu); |
| 996 | } |
| 997 | raise_softirq(RCU_SOFTIRQ); |
| 998 | } |
| 999 | |
| 1000 | #ifdef CONFIG_SMP |
| 1001 | |
| 1002 | /* |
| 1003 | * Scan the leaf rcu_node structures, processing dyntick state for any that |
| 1004 | * have not yet encountered a quiescent state, using the function specified. |
| 1005 | * Returns 1 if the current grace period ends while scanning (possibly |
| 1006 | * because we made it end). |
| 1007 | */ |
| 1008 | static int rcu_process_dyntick(struct rcu_state *rsp, long lastcomp, |
| 1009 | int (*f)(struct rcu_data *)) |
| 1010 | { |
| 1011 | unsigned long bit; |
| 1012 | int cpu; |
| 1013 | unsigned long flags; |
| 1014 | unsigned long mask; |
| 1015 | struct rcu_node *rnp_cur = rsp->level[NUM_RCU_LVLS - 1]; |
| 1016 | struct rcu_node *rnp_end = &rsp->node[NUM_RCU_NODES]; |
| 1017 | |
| 1018 | for (; rnp_cur < rnp_end; rnp_cur++) { |
| 1019 | mask = 0; |
| 1020 | spin_lock_irqsave(&rnp_cur->lock, flags); |
| 1021 | if (rsp->completed != lastcomp) { |
| 1022 | spin_unlock_irqrestore(&rnp_cur->lock, flags); |
| 1023 | return 1; |
| 1024 | } |
| 1025 | if (rnp_cur->qsmask == 0) { |
| 1026 | spin_unlock_irqrestore(&rnp_cur->lock, flags); |
| 1027 | continue; |
| 1028 | } |
| 1029 | cpu = rnp_cur->grplo; |
| 1030 | bit = 1; |
| 1031 | for (; cpu <= rnp_cur->grphi; cpu++, bit <<= 1) { |
| 1032 | if ((rnp_cur->qsmask & bit) != 0 && f(rsp->rda[cpu])) |
| 1033 | mask |= bit; |
| 1034 | } |
| 1035 | if (mask != 0 && rsp->completed == lastcomp) { |
| 1036 | |
| 1037 | /* cpu_quiet_msk() releases rnp_cur->lock. */ |
| 1038 | cpu_quiet_msk(mask, rsp, rnp_cur, flags); |
| 1039 | continue; |
| 1040 | } |
| 1041 | spin_unlock_irqrestore(&rnp_cur->lock, flags); |
| 1042 | } |
| 1043 | return 0; |
| 1044 | } |
| 1045 | |
| 1046 | /* |
| 1047 | * Force quiescent states on reluctant CPUs, and also detect which |
| 1048 | * CPUs are in dyntick-idle mode. |
| 1049 | */ |
| 1050 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed) |
| 1051 | { |
| 1052 | unsigned long flags; |
| 1053 | long lastcomp; |
Paul E. McKenney | 64db4cf | 2008-12-18 21:55:32 +0100 | [diff] [blame] | 1054 | struct rcu_node *rnp = rcu_get_root(rsp); |
| 1055 | u8 signaled; |
| 1056 | |
| 1057 | if (ACCESS_ONCE(rsp->completed) == ACCESS_ONCE(rsp->gpnum)) |
| 1058 | return; /* No grace period in progress, nothing to force. */ |
| 1059 | if (!spin_trylock_irqsave(&rsp->fqslock, flags)) { |
| 1060 | rsp->n_force_qs_lh++; /* Inexact, can lose counts. Tough! */ |
| 1061 | return; /* Someone else is already on the job. */ |
| 1062 | } |
| 1063 | if (relaxed && |
Paul E. McKenney | ef631b0 | 2009-04-13 21:31:16 -0700 | [diff] [blame] | 1064 | (long)(rsp->jiffies_force_qs - jiffies) >= 0) |
Paul E. McKenney | 64db4cf | 2008-12-18 21:55:32 +0100 | [diff] [blame] | 1065 | goto unlock_ret; /* no emergency and done recently. */ |
| 1066 | rsp->n_force_qs++; |
| 1067 | spin_lock(&rnp->lock); |
| 1068 | lastcomp = rsp->completed; |
| 1069 | signaled = rsp->signaled; |
| 1070 | rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS; |
Paul E. McKenney | 64db4cf | 2008-12-18 21:55:32 +0100 | [diff] [blame] | 1071 | if (lastcomp == rsp->gpnum) { |
| 1072 | rsp->n_force_qs_ngp++; |
| 1073 | spin_unlock(&rnp->lock); |
| 1074 | goto unlock_ret; /* no GP in progress, time updated. */ |
| 1075 | } |
| 1076 | spin_unlock(&rnp->lock); |
| 1077 | switch (signaled) { |
| 1078 | case RCU_GP_INIT: |
| 1079 | |
| 1080 | break; /* grace period still initializing, ignore. */ |
| 1081 | |
| 1082 | case RCU_SAVE_DYNTICK: |
| 1083 | |
| 1084 | if (RCU_SIGNAL_INIT != RCU_SAVE_DYNTICK) |
| 1085 | break; /* So gcc recognizes the dead code. */ |
| 1086 | |
| 1087 | /* Record dyntick-idle state. */ |
| 1088 | if (rcu_process_dyntick(rsp, lastcomp, |
| 1089 | dyntick_save_progress_counter)) |
| 1090 | goto unlock_ret; |
| 1091 | |
| 1092 | /* Update state, record completion counter. */ |
| 1093 | spin_lock(&rnp->lock); |
| 1094 | if (lastcomp == rsp->completed) { |
| 1095 | rsp->signaled = RCU_FORCE_QS; |
| 1096 | dyntick_record_completed(rsp, lastcomp); |
| 1097 | } |
| 1098 | spin_unlock(&rnp->lock); |
| 1099 | break; |
| 1100 | |
| 1101 | case RCU_FORCE_QS: |
| 1102 | |
| 1103 | /* Check dyntick-idle state, send IPI to laggarts. */ |
| 1104 | if (rcu_process_dyntick(rsp, dyntick_recall_completed(rsp), |
| 1105 | rcu_implicit_dynticks_qs)) |
| 1106 | goto unlock_ret; |
| 1107 | |
| 1108 | /* Leave state in case more forcing is required. */ |
| 1109 | |
| 1110 | break; |
| 1111 | } |
| 1112 | unlock_ret: |
| 1113 | spin_unlock_irqrestore(&rsp->fqslock, flags); |
| 1114 | } |
| 1115 | |
| 1116 | #else /* #ifdef CONFIG_SMP */ |
| 1117 | |
| 1118 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed) |
| 1119 | { |
| 1120 | set_need_resched(); |
| 1121 | } |
| 1122 | |
| 1123 | #endif /* #else #ifdef CONFIG_SMP */ |
| 1124 | |
| 1125 | /* |
| 1126 | * This does the RCU processing work from softirq context for the |
| 1127 | * specified rcu_state and rcu_data structures. This may be called |
| 1128 | * only from the CPU to whom the rdp belongs. |
| 1129 | */ |
| 1130 | static void |
| 1131 | __rcu_process_callbacks(struct rcu_state *rsp, struct rcu_data *rdp) |
| 1132 | { |
| 1133 | unsigned long flags; |
| 1134 | |
| 1135 | /* |
| 1136 | * If an RCU GP has gone long enough, go check for dyntick |
| 1137 | * idle CPUs and, if needed, send resched IPIs. |
| 1138 | */ |
Paul E. McKenney | ef631b0 | 2009-04-13 21:31:16 -0700 | [diff] [blame] | 1139 | if ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0) |
Paul E. McKenney | 64db4cf | 2008-12-18 21:55:32 +0100 | [diff] [blame] | 1140 | force_quiescent_state(rsp, 1); |
| 1141 | |
| 1142 | /* |
| 1143 | * Advance callbacks in response to end of earlier grace |
| 1144 | * period that some other CPU ended. |
| 1145 | */ |
| 1146 | rcu_process_gp_end(rsp, rdp); |
| 1147 | |
| 1148 | /* Update RCU state based on any recent quiescent states. */ |
| 1149 | rcu_check_quiescent_state(rsp, rdp); |
| 1150 | |
| 1151 | /* Does this CPU require a not-yet-started grace period? */ |
| 1152 | if (cpu_needs_another_gp(rsp, rdp)) { |
| 1153 | spin_lock_irqsave(&rcu_get_root(rsp)->lock, flags); |
| 1154 | rcu_start_gp(rsp, flags); /* releases above lock */ |
| 1155 | } |
| 1156 | |
| 1157 | /* If there are callbacks ready, invoke them. */ |
| 1158 | rcu_do_batch(rdp); |
| 1159 | } |
| 1160 | |
| 1161 | /* |
| 1162 | * Do softirq processing for the current CPU. |
| 1163 | */ |
| 1164 | static void rcu_process_callbacks(struct softirq_action *unused) |
| 1165 | { |
| 1166 | /* |
| 1167 | * Memory references from any prior RCU read-side critical sections |
| 1168 | * executed by the interrupted code must be seen before any RCU |
| 1169 | * grace-period manipulations below. |
| 1170 | */ |
| 1171 | smp_mb(); /* See above block comment. */ |
| 1172 | |
| 1173 | __rcu_process_callbacks(&rcu_state, &__get_cpu_var(rcu_data)); |
| 1174 | __rcu_process_callbacks(&rcu_bh_state, &__get_cpu_var(rcu_bh_data)); |
| 1175 | |
| 1176 | /* |
| 1177 | * Memory references from any later RCU read-side critical sections |
| 1178 | * executed by the interrupted code must be seen after any RCU |
| 1179 | * grace-period manipulations above. |
| 1180 | */ |
| 1181 | smp_mb(); /* See above block comment. */ |
| 1182 | } |
| 1183 | |
| 1184 | static void |
| 1185 | __call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu), |
| 1186 | struct rcu_state *rsp) |
| 1187 | { |
| 1188 | unsigned long flags; |
| 1189 | struct rcu_data *rdp; |
| 1190 | |
| 1191 | head->func = func; |
| 1192 | head->next = NULL; |
| 1193 | |
| 1194 | smp_mb(); /* Ensure RCU update seen before callback registry. */ |
| 1195 | |
| 1196 | /* |
| 1197 | * Opportunistically note grace-period endings and beginnings. |
| 1198 | * Note that we might see a beginning right after we see an |
| 1199 | * end, but never vice versa, since this CPU has to pass through |
| 1200 | * a quiescent state betweentimes. |
| 1201 | */ |
| 1202 | local_irq_save(flags); |
| 1203 | rdp = rsp->rda[smp_processor_id()]; |
| 1204 | rcu_process_gp_end(rsp, rdp); |
| 1205 | check_for_new_grace_period(rsp, rdp); |
| 1206 | |
| 1207 | /* Add the callback to our list. */ |
| 1208 | *rdp->nxttail[RCU_NEXT_TAIL] = head; |
| 1209 | rdp->nxttail[RCU_NEXT_TAIL] = &head->next; |
| 1210 | |
| 1211 | /* Start a new grace period if one not already started. */ |
| 1212 | if (ACCESS_ONCE(rsp->completed) == ACCESS_ONCE(rsp->gpnum)) { |
| 1213 | unsigned long nestflag; |
| 1214 | struct rcu_node *rnp_root = rcu_get_root(rsp); |
| 1215 | |
| 1216 | spin_lock_irqsave(&rnp_root->lock, nestflag); |
| 1217 | rcu_start_gp(rsp, nestflag); /* releases rnp_root->lock. */ |
| 1218 | } |
| 1219 | |
| 1220 | /* Force the grace period if too many callbacks or too long waiting. */ |
| 1221 | if (unlikely(++rdp->qlen > qhimark)) { |
| 1222 | rdp->blimit = LONG_MAX; |
| 1223 | force_quiescent_state(rsp, 0); |
Paul E. McKenney | ef631b0 | 2009-04-13 21:31:16 -0700 | [diff] [blame] | 1224 | } else if ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0) |
Paul E. McKenney | 64db4cf | 2008-12-18 21:55:32 +0100 | [diff] [blame] | 1225 | force_quiescent_state(rsp, 1); |
| 1226 | local_irq_restore(flags); |
| 1227 | } |
| 1228 | |
| 1229 | /* |
| 1230 | * Queue an RCU callback for invocation after a grace period. |
| 1231 | */ |
| 1232 | void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) |
| 1233 | { |
| 1234 | __call_rcu(head, func, &rcu_state); |
| 1235 | } |
| 1236 | EXPORT_SYMBOL_GPL(call_rcu); |
| 1237 | |
| 1238 | /* |
| 1239 | * Queue an RCU for invocation after a quicker grace period. |
| 1240 | */ |
| 1241 | void call_rcu_bh(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) |
| 1242 | { |
| 1243 | __call_rcu(head, func, &rcu_bh_state); |
| 1244 | } |
| 1245 | EXPORT_SYMBOL_GPL(call_rcu_bh); |
| 1246 | |
| 1247 | /* |
| 1248 | * Check to see if there is any immediate RCU-related work to be done |
| 1249 | * by the current CPU, for the specified type of RCU, returning 1 if so. |
| 1250 | * The checks are in order of increasing expense: checks that can be |
| 1251 | * carried out against CPU-local state are performed first. However, |
| 1252 | * we must check for CPU stalls first, else we might not get a chance. |
| 1253 | */ |
| 1254 | static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp) |
| 1255 | { |
| 1256 | rdp->n_rcu_pending++; |
| 1257 | |
| 1258 | /* Check for CPU stalls, if enabled. */ |
| 1259 | check_cpu_stall(rsp, rdp); |
| 1260 | |
| 1261 | /* Is the RCU core waiting for a quiescent state from this CPU? */ |
Paul E. McKenney | 7ba5c84 | 2009-04-13 21:31:17 -0700 | [diff] [blame] | 1262 | if (rdp->qs_pending) { |
| 1263 | rdp->n_rp_qs_pending++; |
Paul E. McKenney | 64db4cf | 2008-12-18 21:55:32 +0100 | [diff] [blame] | 1264 | return 1; |
Paul E. McKenney | 7ba5c84 | 2009-04-13 21:31:17 -0700 | [diff] [blame] | 1265 | } |
Paul E. McKenney | 64db4cf | 2008-12-18 21:55:32 +0100 | [diff] [blame] | 1266 | |
| 1267 | /* Does this CPU have callbacks ready to invoke? */ |
Paul E. McKenney | 7ba5c84 | 2009-04-13 21:31:17 -0700 | [diff] [blame] | 1268 | if (cpu_has_callbacks_ready_to_invoke(rdp)) { |
| 1269 | rdp->n_rp_cb_ready++; |
Paul E. McKenney | 64db4cf | 2008-12-18 21:55:32 +0100 | [diff] [blame] | 1270 | return 1; |
Paul E. McKenney | 7ba5c84 | 2009-04-13 21:31:17 -0700 | [diff] [blame] | 1271 | } |
Paul E. McKenney | 64db4cf | 2008-12-18 21:55:32 +0100 | [diff] [blame] | 1272 | |
| 1273 | /* Has RCU gone idle with this CPU needing another grace period? */ |
Paul E. McKenney | 7ba5c84 | 2009-04-13 21:31:17 -0700 | [diff] [blame] | 1274 | if (cpu_needs_another_gp(rsp, rdp)) { |
| 1275 | rdp->n_rp_cpu_needs_gp++; |
Paul E. McKenney | 64db4cf | 2008-12-18 21:55:32 +0100 | [diff] [blame] | 1276 | return 1; |
Paul E. McKenney | 7ba5c84 | 2009-04-13 21:31:17 -0700 | [diff] [blame] | 1277 | } |
Paul E. McKenney | 64db4cf | 2008-12-18 21:55:32 +0100 | [diff] [blame] | 1278 | |
| 1279 | /* Has another RCU grace period completed? */ |
Paul E. McKenney | 7ba5c84 | 2009-04-13 21:31:17 -0700 | [diff] [blame] | 1280 | if (ACCESS_ONCE(rsp->completed) != rdp->completed) { /* outside lock */ |
| 1281 | rdp->n_rp_gp_completed++; |
Paul E. McKenney | 64db4cf | 2008-12-18 21:55:32 +0100 | [diff] [blame] | 1282 | return 1; |
Paul E. McKenney | 7ba5c84 | 2009-04-13 21:31:17 -0700 | [diff] [blame] | 1283 | } |
Paul E. McKenney | 64db4cf | 2008-12-18 21:55:32 +0100 | [diff] [blame] | 1284 | |
| 1285 | /* Has a new RCU grace period started? */ |
Paul E. McKenney | 7ba5c84 | 2009-04-13 21:31:17 -0700 | [diff] [blame] | 1286 | if (ACCESS_ONCE(rsp->gpnum) != rdp->gpnum) { /* outside lock */ |
| 1287 | rdp->n_rp_gp_started++; |
Paul E. McKenney | 64db4cf | 2008-12-18 21:55:32 +0100 | [diff] [blame] | 1288 | return 1; |
Paul E. McKenney | 7ba5c84 | 2009-04-13 21:31:17 -0700 | [diff] [blame] | 1289 | } |
Paul E. McKenney | 64db4cf | 2008-12-18 21:55:32 +0100 | [diff] [blame] | 1290 | |
| 1291 | /* Has an RCU GP gone long enough to send resched IPIs &c? */ |
| 1292 | if (ACCESS_ONCE(rsp->completed) != ACCESS_ONCE(rsp->gpnum) && |
Paul E. McKenney | 7ba5c84 | 2009-04-13 21:31:17 -0700 | [diff] [blame] | 1293 | ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0)) { |
| 1294 | rdp->n_rp_need_fqs++; |
Paul E. McKenney | 64db4cf | 2008-12-18 21:55:32 +0100 | [diff] [blame] | 1295 | return 1; |
Paul E. McKenney | 7ba5c84 | 2009-04-13 21:31:17 -0700 | [diff] [blame] | 1296 | } |
Paul E. McKenney | 64db4cf | 2008-12-18 21:55:32 +0100 | [diff] [blame] | 1297 | |
| 1298 | /* nothing to do */ |
Paul E. McKenney | 7ba5c84 | 2009-04-13 21:31:17 -0700 | [diff] [blame] | 1299 | rdp->n_rp_need_nothing++; |
Paul E. McKenney | 64db4cf | 2008-12-18 21:55:32 +0100 | [diff] [blame] | 1300 | return 0; |
| 1301 | } |
| 1302 | |
| 1303 | /* |
| 1304 | * Check to see if there is any immediate RCU-related work to be done |
| 1305 | * by the current CPU, returning 1 if so. This function is part of the |
| 1306 | * RCU implementation; it is -not- an exported member of the RCU API. |
| 1307 | */ |
| 1308 | int rcu_pending(int cpu) |
| 1309 | { |
| 1310 | return __rcu_pending(&rcu_state, &per_cpu(rcu_data, cpu)) || |
| 1311 | __rcu_pending(&rcu_bh_state, &per_cpu(rcu_bh_data, cpu)); |
| 1312 | } |
| 1313 | |
| 1314 | /* |
| 1315 | * Check to see if any future RCU-related work will need to be done |
| 1316 | * by the current CPU, even if none need be done immediately, returning |
| 1317 | * 1 if so. This function is part of the RCU implementation; it is -not- |
| 1318 | * an exported member of the RCU API. |
| 1319 | */ |
| 1320 | int rcu_needs_cpu(int cpu) |
| 1321 | { |
| 1322 | /* RCU callbacks either ready or pending? */ |
| 1323 | return per_cpu(rcu_data, cpu).nxtlist || |
| 1324 | per_cpu(rcu_bh_data, cpu).nxtlist; |
| 1325 | } |
| 1326 | |
| 1327 | /* |
| 1328 | * Initialize a CPU's per-CPU RCU data. We take this "scorched earth" |
| 1329 | * approach so that we don't have to worry about how long the CPU has |
| 1330 | * been gone, or whether it ever was online previously. We do trust the |
| 1331 | * ->mynode field, as it is constant for a given struct rcu_data and |
| 1332 | * initialized during early boot. |
| 1333 | * |
| 1334 | * Note that only one online or offline event can be happening at a given |
| 1335 | * time. Note also that we can accept some slop in the rsp->completed |
| 1336 | * access due to the fact that this CPU cannot possibly have any RCU |
| 1337 | * callbacks in flight yet. |
| 1338 | */ |
Lai Jiangshan | e4fa4c9 | 2009-01-14 14:58:15 +0800 | [diff] [blame] | 1339 | static void __cpuinit |
Paul E. McKenney | 64db4cf | 2008-12-18 21:55:32 +0100 | [diff] [blame] | 1340 | rcu_init_percpu_data(int cpu, struct rcu_state *rsp) |
| 1341 | { |
| 1342 | unsigned long flags; |
| 1343 | int i; |
| 1344 | long lastcomp; |
| 1345 | unsigned long mask; |
| 1346 | struct rcu_data *rdp = rsp->rda[cpu]; |
| 1347 | struct rcu_node *rnp = rcu_get_root(rsp); |
| 1348 | |
| 1349 | /* Set up local state, ensuring consistent view of global state. */ |
| 1350 | spin_lock_irqsave(&rnp->lock, flags); |
| 1351 | lastcomp = rsp->completed; |
| 1352 | rdp->completed = lastcomp; |
| 1353 | rdp->gpnum = lastcomp; |
| 1354 | rdp->passed_quiesc = 0; /* We could be racing with new GP, */ |
| 1355 | rdp->qs_pending = 1; /* so set up to respond to current GP. */ |
| 1356 | rdp->beenonline = 1; /* We have now been online. */ |
| 1357 | rdp->passed_quiesc_completed = lastcomp - 1; |
| 1358 | rdp->grpmask = 1UL << (cpu - rdp->mynode->grplo); |
| 1359 | rdp->nxtlist = NULL; |
| 1360 | for (i = 0; i < RCU_NEXT_SIZE; i++) |
| 1361 | rdp->nxttail[i] = &rdp->nxtlist; |
| 1362 | rdp->qlen = 0; |
| 1363 | rdp->blimit = blimit; |
| 1364 | #ifdef CONFIG_NO_HZ |
| 1365 | rdp->dynticks = &per_cpu(rcu_dynticks, cpu); |
| 1366 | #endif /* #ifdef CONFIG_NO_HZ */ |
| 1367 | rdp->cpu = cpu; |
| 1368 | spin_unlock(&rnp->lock); /* irqs remain disabled. */ |
| 1369 | |
| 1370 | /* |
| 1371 | * A new grace period might start here. If so, we won't be part |
| 1372 | * of it, but that is OK, as we are currently in a quiescent state. |
| 1373 | */ |
| 1374 | |
| 1375 | /* Exclude any attempts to start a new GP on large systems. */ |
| 1376 | spin_lock(&rsp->onofflock); /* irqs already disabled. */ |
| 1377 | |
| 1378 | /* Add CPU to rcu_node bitmasks. */ |
| 1379 | rnp = rdp->mynode; |
| 1380 | mask = rdp->grpmask; |
| 1381 | do { |
| 1382 | /* Exclude any attempts to start a new GP on small systems. */ |
| 1383 | spin_lock(&rnp->lock); /* irqs already disabled. */ |
| 1384 | rnp->qsmaskinit |= mask; |
| 1385 | mask = rnp->grpmask; |
| 1386 | spin_unlock(&rnp->lock); /* irqs already disabled. */ |
| 1387 | rnp = rnp->parent; |
| 1388 | } while (rnp != NULL && !(rnp->qsmaskinit & mask)); |
| 1389 | |
| 1390 | spin_unlock(&rsp->onofflock); /* irqs remain disabled. */ |
| 1391 | |
| 1392 | /* |
| 1393 | * A new grace period might start here. If so, we will be part of |
| 1394 | * it, and its gpnum will be greater than ours, so we will |
| 1395 | * participate. It is also possible for the gpnum to have been |
| 1396 | * incremented before this function was called, and the bitmasks |
| 1397 | * to not be filled out until now, in which case we will also |
| 1398 | * participate due to our gpnum being behind. |
| 1399 | */ |
| 1400 | |
| 1401 | /* Since it is coming online, the CPU is in a quiescent state. */ |
| 1402 | cpu_quiet(cpu, rsp, rdp, lastcomp); |
| 1403 | local_irq_restore(flags); |
| 1404 | } |
| 1405 | |
| 1406 | static void __cpuinit rcu_online_cpu(int cpu) |
| 1407 | { |
Paul E. McKenney | 64db4cf | 2008-12-18 21:55:32 +0100 | [diff] [blame] | 1408 | rcu_init_percpu_data(cpu, &rcu_state); |
| 1409 | rcu_init_percpu_data(cpu, &rcu_bh_state); |
| 1410 | open_softirq(RCU_SOFTIRQ, rcu_process_callbacks); |
| 1411 | } |
| 1412 | |
| 1413 | /* |
| 1414 | * Handle CPU online/offline notifcation events. |
| 1415 | */ |
| 1416 | static int __cpuinit rcu_cpu_notify(struct notifier_block *self, |
| 1417 | unsigned long action, void *hcpu) |
| 1418 | { |
| 1419 | long cpu = (long)hcpu; |
| 1420 | |
| 1421 | switch (action) { |
| 1422 | case CPU_UP_PREPARE: |
| 1423 | case CPU_UP_PREPARE_FROZEN: |
| 1424 | rcu_online_cpu(cpu); |
| 1425 | break; |
| 1426 | case CPU_DEAD: |
| 1427 | case CPU_DEAD_FROZEN: |
| 1428 | case CPU_UP_CANCELED: |
| 1429 | case CPU_UP_CANCELED_FROZEN: |
| 1430 | rcu_offline_cpu(cpu); |
| 1431 | break; |
| 1432 | default: |
| 1433 | break; |
| 1434 | } |
| 1435 | return NOTIFY_OK; |
| 1436 | } |
| 1437 | |
| 1438 | /* |
| 1439 | * Compute the per-level fanout, either using the exact fanout specified |
| 1440 | * or balancing the tree, depending on CONFIG_RCU_FANOUT_EXACT. |
| 1441 | */ |
| 1442 | #ifdef CONFIG_RCU_FANOUT_EXACT |
| 1443 | static void __init rcu_init_levelspread(struct rcu_state *rsp) |
| 1444 | { |
| 1445 | int i; |
| 1446 | |
| 1447 | for (i = NUM_RCU_LVLS - 1; i >= 0; i--) |
| 1448 | rsp->levelspread[i] = CONFIG_RCU_FANOUT; |
| 1449 | } |
| 1450 | #else /* #ifdef CONFIG_RCU_FANOUT_EXACT */ |
| 1451 | static void __init rcu_init_levelspread(struct rcu_state *rsp) |
| 1452 | { |
| 1453 | int ccur; |
| 1454 | int cprv; |
| 1455 | int i; |
| 1456 | |
| 1457 | cprv = NR_CPUS; |
| 1458 | for (i = NUM_RCU_LVLS - 1; i >= 0; i--) { |
| 1459 | ccur = rsp->levelcnt[i]; |
| 1460 | rsp->levelspread[i] = (cprv + ccur - 1) / ccur; |
| 1461 | cprv = ccur; |
| 1462 | } |
| 1463 | } |
| 1464 | #endif /* #else #ifdef CONFIG_RCU_FANOUT_EXACT */ |
| 1465 | |
| 1466 | /* |
| 1467 | * Helper function for rcu_init() that initializes one rcu_state structure. |
| 1468 | */ |
| 1469 | static void __init rcu_init_one(struct rcu_state *rsp) |
| 1470 | { |
| 1471 | int cpustride = 1; |
| 1472 | int i; |
| 1473 | int j; |
| 1474 | struct rcu_node *rnp; |
| 1475 | |
| 1476 | /* Initialize the level-tracking arrays. */ |
| 1477 | |
| 1478 | for (i = 1; i < NUM_RCU_LVLS; i++) |
| 1479 | rsp->level[i] = rsp->level[i - 1] + rsp->levelcnt[i - 1]; |
| 1480 | rcu_init_levelspread(rsp); |
| 1481 | |
| 1482 | /* Initialize the elements themselves, starting from the leaves. */ |
| 1483 | |
| 1484 | for (i = NUM_RCU_LVLS - 1; i >= 0; i--) { |
| 1485 | cpustride *= rsp->levelspread[i]; |
| 1486 | rnp = rsp->level[i]; |
| 1487 | for (j = 0; j < rsp->levelcnt[i]; j++, rnp++) { |
| 1488 | spin_lock_init(&rnp->lock); |
| 1489 | rnp->qsmask = 0; |
| 1490 | rnp->qsmaskinit = 0; |
| 1491 | rnp->grplo = j * cpustride; |
| 1492 | rnp->grphi = (j + 1) * cpustride - 1; |
| 1493 | if (rnp->grphi >= NR_CPUS) |
| 1494 | rnp->grphi = NR_CPUS - 1; |
| 1495 | if (i == 0) { |
| 1496 | rnp->grpnum = 0; |
| 1497 | rnp->grpmask = 0; |
| 1498 | rnp->parent = NULL; |
| 1499 | } else { |
| 1500 | rnp->grpnum = j % rsp->levelspread[i - 1]; |
| 1501 | rnp->grpmask = 1UL << rnp->grpnum; |
| 1502 | rnp->parent = rsp->level[i - 1] + |
| 1503 | j / rsp->levelspread[i - 1]; |
| 1504 | } |
| 1505 | rnp->level = i; |
| 1506 | } |
| 1507 | } |
| 1508 | } |
| 1509 | |
| 1510 | /* |
| 1511 | * Helper macro for __rcu_init(). To be used nowhere else! |
| 1512 | * Assigns leaf node pointers into each CPU's rcu_data structure. |
| 1513 | */ |
| 1514 | #define RCU_DATA_PTR_INIT(rsp, rcu_data) \ |
| 1515 | do { \ |
| 1516 | rnp = (rsp)->level[NUM_RCU_LVLS - 1]; \ |
| 1517 | j = 0; \ |
| 1518 | for_each_possible_cpu(i) { \ |
| 1519 | if (i > rnp[j].grphi) \ |
| 1520 | j++; \ |
| 1521 | per_cpu(rcu_data, i).mynode = &rnp[j]; \ |
| 1522 | (rsp)->rda[i] = &per_cpu(rcu_data, i); \ |
| 1523 | } \ |
| 1524 | } while (0) |
| 1525 | |
| 1526 | static struct notifier_block __cpuinitdata rcu_nb = { |
| 1527 | .notifier_call = rcu_cpu_notify, |
| 1528 | }; |
| 1529 | |
| 1530 | void __init __rcu_init(void) |
| 1531 | { |
| 1532 | int i; /* All used by RCU_DATA_PTR_INIT(). */ |
| 1533 | int j; |
| 1534 | struct rcu_node *rnp; |
| 1535 | |
| 1536 | printk(KERN_WARNING "Experimental hierarchical RCU implementation.\n"); |
| 1537 | #ifdef CONFIG_RCU_CPU_STALL_DETECTOR |
| 1538 | printk(KERN_INFO "RCU-based detection of stalled CPUs is enabled.\n"); |
| 1539 | #endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */ |
| 1540 | rcu_init_one(&rcu_state); |
| 1541 | RCU_DATA_PTR_INIT(&rcu_state, rcu_data); |
| 1542 | rcu_init_one(&rcu_bh_state); |
| 1543 | RCU_DATA_PTR_INIT(&rcu_bh_state, rcu_bh_data); |
| 1544 | |
| 1545 | for_each_online_cpu(i) |
| 1546 | rcu_cpu_notify(&rcu_nb, CPU_UP_PREPARE, (void *)(long)i); |
| 1547 | /* Register notifier for non-boot CPUs */ |
| 1548 | register_cpu_notifier(&rcu_nb); |
| 1549 | printk(KERN_WARNING "Experimental hierarchical RCU init done.\n"); |
| 1550 | } |
| 1551 | |
| 1552 | module_param(blimit, int, 0); |
| 1553 | module_param(qhimark, int, 0); |
| 1554 | module_param(qlowmark, int, 0); |