aboutsummaryrefslogtreecommitdiffstats
path: root/kernel
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@linux-foundation.org>2009-09-11 16:20:18 -0400
committerLinus Torvalds <torvalds@linux-foundation.org>2009-09-11 16:20:18 -0400
commiteee2775d9924b22643bd89b2e568cc5eed7e8a04 (patch)
tree095ad7851895c5d39596f3ff7ee1e078235a2501 /kernel
parent53e16fbd30005905168d9b75555fdc7e0a2eac58 (diff)
parent7db905e636f08ea5bc9825c1f73d77802e8ccad5 (diff)
Merge branch 'core-rcu-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip
* 'core-rcu-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip: (28 commits) rcu: Move end of special early-boot RCU operation earlier rcu: Changes from reviews: avoid casts, fix/add warnings, improve comments rcu: Create rcutree plugins to handle hotplug CPU for multi-level trees rcu: Remove lockdep annotations from RCU's _notrace() API members rcu: Add #ifdef to suppress __rcu_offline_cpu() warning in !HOTPLUG_CPU builds rcu: Add CPU-offline processing for single-node configurations rcu: Add "notrace" to RCU function headers used by ftrace rcu: Remove CONFIG_PREEMPT_RCU rcu: Merge preemptable-RCU functionality into hierarchical RCU rcu: Simplify rcu_pending()/rcu_check_callbacks() API rcu: Use debugfs_remove_recursive() simplify code. rcu: Merge per-RCU-flavor initialization into pre-existing macro rcu: Fix online/offline indication for rcudata.csv trace file rcu: Consolidate sparse and lockdep declarations in include/linux/rcupdate.h rcu: Renamings to increase RCU clarity rcu: Move private definitions from include/linux/rcutree.h to kernel/rcutree.h rcu: Expunge lingering references to CONFIG_CLASSIC_RCU, optimize on !SMP rcu: Delay rcu_barrier() wait until beginning of next CPU-hotunplug operation. rcu: Fix typo in rcu_irq_exit() comment header rcu: Make rcupreempt_trace.c look at offline CPUs ...
Diffstat (limited to 'kernel')
-rw-r--r--kernel/Makefile4
-rw-r--r--kernel/exit.c1
-rw-r--r--kernel/fork.c5
-rw-r--r--kernel/rcuclassic.c807
-rw-r--r--kernel/rcupdate.c44
-rw-r--r--kernel/rcupreempt.c1539
-rw-r--r--kernel/rcupreempt_trace.c334
-rw-r--r--kernel/rcutorture.c202
-rw-r--r--kernel/rcutree.c273
-rw-r--r--kernel/rcutree.h253
-rw-r--r--kernel/rcutree_plugin.h532
-rw-r--r--kernel/rcutree_trace.c88
-rw-r--r--kernel/sched.c131
-rw-r--r--kernel/softirq.c4
-rw-r--r--kernel/timer.c3
15 files changed, 1293 insertions, 2927 deletions
diff --git a/kernel/Makefile b/kernel/Makefile
index 2093a691f1c2..b833bd5cc127 100644
--- a/kernel/Makefile
+++ b/kernel/Makefile
@@ -80,11 +80,9 @@ obj-$(CONFIG_DETECT_HUNG_TASK) += hung_task.o
80obj-$(CONFIG_GENERIC_HARDIRQS) += irq/ 80obj-$(CONFIG_GENERIC_HARDIRQS) += irq/
81obj-$(CONFIG_SECCOMP) += seccomp.o 81obj-$(CONFIG_SECCOMP) += seccomp.o
82obj-$(CONFIG_RCU_TORTURE_TEST) += rcutorture.o 82obj-$(CONFIG_RCU_TORTURE_TEST) += rcutorture.o
83obj-$(CONFIG_CLASSIC_RCU) += rcuclassic.o
84obj-$(CONFIG_TREE_RCU) += rcutree.o 83obj-$(CONFIG_TREE_RCU) += rcutree.o
85obj-$(CONFIG_PREEMPT_RCU) += rcupreempt.o 84obj-$(CONFIG_TREE_PREEMPT_RCU) += rcutree.o
86obj-$(CONFIG_TREE_RCU_TRACE) += rcutree_trace.o 85obj-$(CONFIG_TREE_RCU_TRACE) += rcutree_trace.o
87obj-$(CONFIG_PREEMPT_RCU_TRACE) += rcupreempt_trace.o
88obj-$(CONFIG_RELAY) += relay.o 86obj-$(CONFIG_RELAY) += relay.o
89obj-$(CONFIG_SYSCTL) += utsname_sysctl.o 87obj-$(CONFIG_SYSCTL) += utsname_sysctl.o
90obj-$(CONFIG_TASK_DELAY_ACCT) += delayacct.o 88obj-$(CONFIG_TASK_DELAY_ACCT) += delayacct.o
diff --git a/kernel/exit.c b/kernel/exit.c
index c98ff7a8025f..ae5d8660ddff 100644
--- a/kernel/exit.c
+++ b/kernel/exit.c
@@ -1014,6 +1014,7 @@ NORET_TYPE void do_exit(long code)
1014 validate_creds_for_do_exit(tsk); 1014 validate_creds_for_do_exit(tsk);
1015 1015
1016 preempt_disable(); 1016 preempt_disable();
1017 exit_rcu();
1017 /* causes final put_task_struct in finish_task_switch(). */ 1018 /* causes final put_task_struct in finish_task_switch(). */
1018 tsk->state = TASK_DEAD; 1019 tsk->state = TASK_DEAD;
1019 schedule(); 1020 schedule();
diff --git a/kernel/fork.c b/kernel/fork.c
index aab8579c6093..bfee931ee3fb 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -1007,10 +1007,7 @@ static struct task_struct *copy_process(unsigned long clone_flags,
1007 copy_flags(clone_flags, p); 1007 copy_flags(clone_flags, p);
1008 INIT_LIST_HEAD(&p->children); 1008 INIT_LIST_HEAD(&p->children);
1009 INIT_LIST_HEAD(&p->sibling); 1009 INIT_LIST_HEAD(&p->sibling);
1010#ifdef CONFIG_PREEMPT_RCU 1010 rcu_copy_process(p);
1011 p->rcu_read_lock_nesting = 0;
1012 p->rcu_flipctr_idx = 0;
1013#endif /* #ifdef CONFIG_PREEMPT_RCU */
1014 p->vfork_done = NULL; 1011 p->vfork_done = NULL;
1015 spin_lock_init(&p->alloc_lock); 1012 spin_lock_init(&p->alloc_lock);
1016 1013
diff --git a/kernel/rcuclassic.c b/kernel/rcuclassic.c
deleted file mode 100644
index 0f2b0b311304..000000000000
--- a/kernel/rcuclassic.c
+++ /dev/null
@@ -1,807 +0,0 @@
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>
48#include <linux/cpu.h>
49#include <linux/mutex.h>
50#include <linux/time.h>
51
52#ifdef CONFIG_DEBUG_LOCK_ALLOC
53static struct lock_class_key rcu_lock_key;
54struct lockdep_map rcu_lock_map =
55 STATIC_LOCKDEP_MAP_INIT("rcu_read_lock", &rcu_lock_key);
56EXPORT_SYMBOL_GPL(rcu_lock_map);
57#endif
58
59
60/* Definition for rcupdate control block. */
61static struct rcu_ctrlblk rcu_ctrlblk = {
62 .cur = -300,
63 .completed = -300,
64 .pending = -300,
65 .lock = __SPIN_LOCK_UNLOCKED(&rcu_ctrlblk.lock),
66 .cpumask = CPU_BITS_NONE,
67};
68
69static struct rcu_ctrlblk rcu_bh_ctrlblk = {
70 .cur = -300,
71 .completed = -300,
72 .pending = -300,
73 .lock = __SPIN_LOCK_UNLOCKED(&rcu_bh_ctrlblk.lock),
74 .cpumask = CPU_BITS_NONE,
75};
76
77static DEFINE_PER_CPU(struct rcu_data, rcu_data);
78static DEFINE_PER_CPU(struct rcu_data, rcu_bh_data);
79
80/*
81 * Increment the quiescent state counter.
82 * The counter is a bit degenerated: We do not need to know
83 * how many quiescent states passed, just if there was at least
84 * one since the start of the grace period. Thus just a flag.
85 */
86void rcu_qsctr_inc(int cpu)
87{
88 struct rcu_data *rdp = &per_cpu(rcu_data, cpu);
89 rdp->passed_quiesc = 1;
90}
91
92void rcu_bh_qsctr_inc(int cpu)
93{
94 struct rcu_data *rdp = &per_cpu(rcu_bh_data, cpu);
95 rdp->passed_quiesc = 1;
96}
97
98static int blimit = 10;
99static int qhimark = 10000;
100static int qlowmark = 100;
101
102#ifdef CONFIG_SMP
103static void force_quiescent_state(struct rcu_data *rdp,
104 struct rcu_ctrlblk *rcp)
105{
106 int cpu;
107 unsigned long flags;
108
109 set_need_resched();
110 spin_lock_irqsave(&rcp->lock, flags);
111 if (unlikely(!rcp->signaled)) {
112 rcp->signaled = 1;
113 /*
114 * Don't send IPI to itself. With irqs disabled,
115 * rdp->cpu is the current cpu.
116 *
117 * cpu_online_mask is updated by the _cpu_down()
118 * using __stop_machine(). Since we're in irqs disabled
119 * section, __stop_machine() is not exectuting, hence
120 * the cpu_online_mask is stable.
121 *
122 * However, a cpu might have been offlined _just_ before
123 * we disabled irqs while entering here.
124 * And rcu subsystem might not yet have handled the CPU_DEAD
125 * notification, leading to the offlined cpu's bit
126 * being set in the rcp->cpumask.
127 *
128 * Hence cpumask = (rcp->cpumask & cpu_online_mask) to prevent
129 * sending smp_reschedule() to an offlined CPU.
130 */
131 for_each_cpu_and(cpu,
132 to_cpumask(rcp->cpumask), cpu_online_mask) {
133 if (cpu != rdp->cpu)
134 smp_send_reschedule(cpu);
135 }
136 }
137 spin_unlock_irqrestore(&rcp->lock, flags);
138}
139#else
140static inline void force_quiescent_state(struct rcu_data *rdp,
141 struct rcu_ctrlblk *rcp)
142{
143 set_need_resched();
144}
145#endif
146
147static void __call_rcu(struct rcu_head *head, struct rcu_ctrlblk *rcp,
148 struct rcu_data *rdp)
149{
150 long batch;
151
152 head->next = NULL;
153 smp_mb(); /* Read of rcu->cur must happen after any change by caller. */
154
155 /*
156 * Determine the batch number of this callback.
157 *
158 * Using ACCESS_ONCE to avoid the following error when gcc eliminates
159 * local variable "batch" and emits codes like this:
160 * 1) rdp->batch = rcp->cur + 1 # gets old value
161 * ......
162 * 2)rcu_batch_after(rcp->cur + 1, rdp->batch) # gets new value
163 * then [*nxttail[0], *nxttail[1]) may contain callbacks
164 * that batch# = rdp->batch, see the comment of struct rcu_data.
165 */
166 batch = ACCESS_ONCE(rcp->cur) + 1;
167
168 if (rdp->nxtlist && rcu_batch_after(batch, rdp->batch)) {
169 /* process callbacks */
170 rdp->nxttail[0] = rdp->nxttail[1];
171 rdp->nxttail[1] = rdp->nxttail[2];
172 if (rcu_batch_after(batch - 1, rdp->batch))
173 rdp->nxttail[0] = rdp->nxttail[2];
174 }
175
176 rdp->batch = batch;
177 *rdp->nxttail[2] = head;
178 rdp->nxttail[2] = &head->next;
179
180 if (unlikely(++rdp->qlen > qhimark)) {
181 rdp->blimit = INT_MAX;
182 force_quiescent_state(rdp, &rcu_ctrlblk);
183 }
184}
185
186#ifdef CONFIG_RCU_CPU_STALL_DETECTOR
187
188static void record_gp_stall_check_time(struct rcu_ctrlblk *rcp)
189{
190 rcp->gp_start = jiffies;
191 rcp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_CHECK;
192}
193
194static void print_other_cpu_stall(struct rcu_ctrlblk *rcp)
195{
196 int cpu;
197 long delta;
198 unsigned long flags;
199
200 /* Only let one CPU complain about others per time interval. */
201
202 spin_lock_irqsave(&rcp->lock, flags);
203 delta = jiffies - rcp->jiffies_stall;
204 if (delta < 2 || rcp->cur != rcp->completed) {
205 spin_unlock_irqrestore(&rcp->lock, flags);
206 return;
207 }
208 rcp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_RECHECK;
209 spin_unlock_irqrestore(&rcp->lock, flags);
210
211 /* OK, time to rat on our buddy... */
212
213 printk(KERN_ERR "INFO: RCU detected CPU stalls:");
214 for_each_possible_cpu(cpu) {
215 if (cpumask_test_cpu(cpu, to_cpumask(rcp->cpumask)))
216 printk(" %d", cpu);
217 }
218 printk(" (detected by %d, t=%ld jiffies)\n",
219 smp_processor_id(), (long)(jiffies - rcp->gp_start));
220}
221
222static void print_cpu_stall(struct rcu_ctrlblk *rcp)
223{
224 unsigned long flags;
225
226 printk(KERN_ERR "INFO: RCU detected CPU %d stall (t=%lu/%lu jiffies)\n",
227 smp_processor_id(), jiffies,
228 jiffies - rcp->gp_start);
229 dump_stack();
230 spin_lock_irqsave(&rcp->lock, flags);
231 if ((long)(jiffies - rcp->jiffies_stall) >= 0)
232 rcp->jiffies_stall =
233 jiffies + RCU_SECONDS_TILL_STALL_RECHECK;
234 spin_unlock_irqrestore(&rcp->lock, flags);
235 set_need_resched(); /* kick ourselves to get things going. */
236}
237
238static void check_cpu_stall(struct rcu_ctrlblk *rcp)
239{
240 long delta;
241
242 delta = jiffies - rcp->jiffies_stall;
243 if (cpumask_test_cpu(smp_processor_id(), to_cpumask(rcp->cpumask)) &&
244 delta >= 0) {
245
246 /* We haven't checked in, so go dump stack. */
247 print_cpu_stall(rcp);
248
249 } else if (rcp->cur != rcp->completed && delta >= 2) {
250
251 /* They had two seconds to dump stack, so complain. */
252 print_other_cpu_stall(rcp);
253 }
254}
255
256#else /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
257
258static void record_gp_stall_check_time(struct rcu_ctrlblk *rcp)
259{
260}
261
262static inline void check_cpu_stall(struct rcu_ctrlblk *rcp)
263{
264}
265
266#endif /* #else #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
267
268/**
269 * call_rcu - Queue an RCU callback for invocation after a grace period.
270 * @head: structure to be used for queueing the RCU updates.
271 * @func: actual update function to be invoked after the grace period
272 *
273 * The update function will be invoked some time after a full grace
274 * period elapses, in other words after all currently executing RCU
275 * read-side critical sections have completed. RCU read-side critical
276 * sections are delimited by rcu_read_lock() and rcu_read_unlock(),
277 * and may be nested.
278 */
279void call_rcu(struct rcu_head *head,
280 void (*func)(struct rcu_head *rcu))
281{
282 unsigned long flags;
283
284 head->func = func;
285 local_irq_save(flags);
286 __call_rcu(head, &rcu_ctrlblk, &__get_cpu_var(rcu_data));
287 local_irq_restore(flags);
288}
289EXPORT_SYMBOL_GPL(call_rcu);
290
291/**
292 * call_rcu_bh - Queue an RCU for invocation after a quicker grace period.
293 * @head: structure to be used for queueing the RCU updates.
294 * @func: actual update function to be invoked after the grace period
295 *
296 * The update function will be invoked some time after a full grace
297 * period elapses, in other words after all currently executing RCU
298 * read-side critical sections have completed. call_rcu_bh() assumes
299 * that the read-side critical sections end on completion of a softirq
300 * handler. This means that read-side critical sections in process
301 * context must not be interrupted by softirqs. This interface is to be
302 * used when most of the read-side critical sections are in softirq context.
303 * RCU read-side critical sections are delimited by rcu_read_lock() and
304 * rcu_read_unlock(), * if in interrupt context or rcu_read_lock_bh()
305 * and rcu_read_unlock_bh(), if in process context. These may be nested.
306 */
307void call_rcu_bh(struct rcu_head *head,
308 void (*func)(struct rcu_head *rcu))
309{
310 unsigned long flags;
311
312 head->func = func;
313 local_irq_save(flags);
314 __call_rcu(head, &rcu_bh_ctrlblk, &__get_cpu_var(rcu_bh_data));
315 local_irq_restore(flags);
316}
317EXPORT_SYMBOL_GPL(call_rcu_bh);
318
319/*
320 * Return the number of RCU batches processed thus far. Useful
321 * for debug and statistics.
322 */
323long rcu_batches_completed(void)
324{
325 return rcu_ctrlblk.completed;
326}
327EXPORT_SYMBOL_GPL(rcu_batches_completed);
328
329/*
330 * Return the number of RCU batches processed thus far. Useful
331 * for debug and statistics.
332 */
333long rcu_batches_completed_bh(void)
334{
335 return rcu_bh_ctrlblk.completed;
336}
337EXPORT_SYMBOL_GPL(rcu_batches_completed_bh);
338
339/* Raises the softirq for processing rcu_callbacks. */
340static inline void raise_rcu_softirq(void)
341{
342 raise_softirq(RCU_SOFTIRQ);
343}
344
345/*
346 * Invoke the completed RCU callbacks. They are expected to be in
347 * a per-cpu list.
348 */
349static void rcu_do_batch(struct rcu_data *rdp)
350{
351 unsigned long flags;
352 struct rcu_head *next, *list;
353 int count = 0;
354
355 list = rdp->donelist;
356 while (list) {
357 next = list->next;
358 prefetch(next);
359 list->func(list);
360 list = next;
361 if (++count >= rdp->blimit)
362 break;
363 }
364 rdp->donelist = list;
365
366 local_irq_save(flags);
367 rdp->qlen -= count;
368 local_irq_restore(flags);
369 if (rdp->blimit == INT_MAX && rdp->qlen <= qlowmark)
370 rdp->blimit = blimit;
371
372 if (!rdp->donelist)
373 rdp->donetail = &rdp->donelist;
374 else
375 raise_rcu_softirq();
376}
377
378/*
379 * Grace period handling:
380 * The grace period handling consists out of two steps:
381 * - A new grace period is started.
382 * This is done by rcu_start_batch. The start is not broadcasted to
383 * all cpus, they must pick this up by comparing rcp->cur with
384 * rdp->quiescbatch. All cpus are recorded in the
385 * rcu_ctrlblk.cpumask bitmap.
386 * - All cpus must go through a quiescent state.
387 * Since the start of the grace period is not broadcasted, at least two
388 * calls to rcu_check_quiescent_state are required:
389 * The first call just notices that a new grace period is running. The
390 * following calls check if there was a quiescent state since the beginning
391 * of the grace period. If so, it updates rcu_ctrlblk.cpumask. If
392 * the bitmap is empty, then the grace period is completed.
393 * rcu_check_quiescent_state calls rcu_start_batch(0) to start the next grace
394 * period (if necessary).
395 */
396
397/*
398 * Register a new batch of callbacks, and start it up if there is currently no
399 * active batch and the batch to be registered has not already occurred.
400 * Caller must hold rcu_ctrlblk.lock.
401 */
402static void rcu_start_batch(struct rcu_ctrlblk *rcp)
403{
404 if (rcp->cur != rcp->pending &&
405 rcp->completed == rcp->cur) {
406 rcp->cur++;
407 record_gp_stall_check_time(rcp);
408
409 /*
410 * Accessing nohz_cpu_mask before incrementing rcp->cur needs a
411 * Barrier Otherwise it can cause tickless idle CPUs to be
412 * included in rcp->cpumask, which will extend graceperiods
413 * unnecessarily.
414 */
415 smp_mb();
416 cpumask_andnot(to_cpumask(rcp->cpumask),
417 cpu_online_mask, nohz_cpu_mask);
418
419 rcp->signaled = 0;
420 }
421}
422
423/*
424 * cpu went through a quiescent state since the beginning of the grace period.
425 * Clear it from the cpu mask and complete the grace period if it was the last
426 * cpu. Start another grace period if someone has further entries pending
427 */
428static void cpu_quiet(int cpu, struct rcu_ctrlblk *rcp)
429{
430 cpumask_clear_cpu(cpu, to_cpumask(rcp->cpumask));
431 if (cpumask_empty(to_cpumask(rcp->cpumask))) {
432 /* batch completed ! */
433 rcp->completed = rcp->cur;
434 rcu_start_batch(rcp);
435 }
436}
437
438/*
439 * Check if the cpu has gone through a quiescent state (say context
440 * switch). If so and if it already hasn't done so in this RCU
441 * quiescent cycle, then indicate that it has done so.
442 */
443static void rcu_check_quiescent_state(struct rcu_ctrlblk *rcp,
444 struct rcu_data *rdp)
445{
446 unsigned long flags;
447
448 if (rdp->quiescbatch != rcp->cur) {
449 /* start new grace period: */
450 rdp->qs_pending = 1;
451 rdp->passed_quiesc = 0;
452 rdp->quiescbatch = rcp->cur;
453 return;
454 }
455
456 /* Grace period already completed for this cpu?
457 * qs_pending is checked instead of the actual bitmap to avoid
458 * cacheline trashing.
459 */
460 if (!rdp->qs_pending)
461 return;
462
463 /*
464 * Was there a quiescent state since the beginning of the grace
465 * period? If no, then exit and wait for the next call.
466 */
467 if (!rdp->passed_quiesc)
468 return;
469 rdp->qs_pending = 0;
470
471 spin_lock_irqsave(&rcp->lock, flags);
472 /*
473 * rdp->quiescbatch/rcp->cur and the cpu bitmap can come out of sync
474 * during cpu startup. Ignore the quiescent state.
475 */
476 if (likely(rdp->quiescbatch == rcp->cur))
477 cpu_quiet(rdp->cpu, rcp);
478
479 spin_unlock_irqrestore(&rcp->lock, flags);
480}
481
482
483#ifdef CONFIG_HOTPLUG_CPU
484
485/* warning! helper for rcu_offline_cpu. do not use elsewhere without reviewing
486 * locking requirements, the list it's pulling from has to belong to a cpu
487 * which is dead and hence not processing interrupts.
488 */
489static void rcu_move_batch(struct rcu_data *this_rdp, struct rcu_head *list,
490 struct rcu_head **tail, long batch)
491{
492 unsigned long flags;
493
494 if (list) {
495 local_irq_save(flags);
496 this_rdp->batch = batch;
497 *this_rdp->nxttail[2] = list;
498 this_rdp->nxttail[2] = tail;
499 local_irq_restore(flags);
500 }
501}
502
503static void __rcu_offline_cpu(struct rcu_data *this_rdp,
504 struct rcu_ctrlblk *rcp, struct rcu_data *rdp)
505{
506 unsigned long flags;
507
508 /*
509 * if the cpu going offline owns the grace period
510 * we can block indefinitely waiting for it, so flush
511 * it here
512 */
513 spin_lock_irqsave(&rcp->lock, flags);
514 if (rcp->cur != rcp->completed)
515 cpu_quiet(rdp->cpu, rcp);
516 rcu_move_batch(this_rdp, rdp->donelist, rdp->donetail, rcp->cur + 1);
517 rcu_move_batch(this_rdp, rdp->nxtlist, rdp->nxttail[2], rcp->cur + 1);
518 spin_unlock(&rcp->lock);
519
520 this_rdp->qlen += rdp->qlen;
521 local_irq_restore(flags);
522}
523
524static void rcu_offline_cpu(int cpu)
525{
526 struct rcu_data *this_rdp = &get_cpu_var(rcu_data);
527 struct rcu_data *this_bh_rdp = &get_cpu_var(rcu_bh_data);
528
529 __rcu_offline_cpu(this_rdp, &rcu_ctrlblk,
530 &per_cpu(rcu_data, cpu));
531 __rcu_offline_cpu(this_bh_rdp, &rcu_bh_ctrlblk,
532 &per_cpu(rcu_bh_data, cpu));
533 put_cpu_var(rcu_data);
534 put_cpu_var(rcu_bh_data);
535}
536
537#else
538
539static void rcu_offline_cpu(int cpu)
540{
541}
542
543#endif
544
545/*
546 * This does the RCU processing work from softirq context.
547 */
548static void __rcu_process_callbacks(struct rcu_ctrlblk *rcp,
549 struct rcu_data *rdp)
550{
551 unsigned long flags;
552 long completed_snap;
553
554 if (rdp->nxtlist) {
555 local_irq_save(flags);
556 completed_snap = ACCESS_ONCE(rcp->completed);
557
558 /*
559 * move the other grace-period-completed entries to
560 * [rdp->nxtlist, *rdp->nxttail[0]) temporarily
561 */
562 if (!rcu_batch_before(completed_snap, rdp->batch))
563 rdp->nxttail[0] = rdp->nxttail[1] = rdp->nxttail[2];
564 else if (!rcu_batch_before(completed_snap, rdp->batch - 1))
565 rdp->nxttail[0] = rdp->nxttail[1];
566
567 /*
568 * the grace period for entries in
569 * [rdp->nxtlist, *rdp->nxttail[0]) has completed and
570 * move these entries to donelist
571 */
572 if (rdp->nxttail[0] != &rdp->nxtlist) {
573 *rdp->donetail = rdp->nxtlist;
574 rdp->donetail = rdp->nxttail[0];
575 rdp->nxtlist = *rdp->nxttail[0];
576 *rdp->donetail = NULL;
577
578 if (rdp->nxttail[1] == rdp->nxttail[0])
579 rdp->nxttail[1] = &rdp->nxtlist;
580 if (rdp->nxttail[2] == rdp->nxttail[0])
581 rdp->nxttail[2] = &rdp->nxtlist;
582 rdp->nxttail[0] = &rdp->nxtlist;
583 }
584
585 local_irq_restore(flags);
586
587 if (rcu_batch_after(rdp->batch, rcp->pending)) {
588 unsigned long flags2;
589
590 /* and start it/schedule start if it's a new batch */
591 spin_lock_irqsave(&rcp->lock, flags2);
592 if (rcu_batch_after(rdp->batch, rcp->pending)) {
593 rcp->pending = rdp->batch;
594 rcu_start_batch(rcp);
595 }
596 spin_unlock_irqrestore(&rcp->lock, flags2);
597 }
598 }
599
600 rcu_check_quiescent_state(rcp, rdp);
601 if (rdp->donelist)
602 rcu_do_batch(rdp);
603}
604
605static void rcu_process_callbacks(struct softirq_action *unused)
606{
607 /*
608 * Memory references from any prior RCU read-side critical sections
609 * executed by the interrupted code must be see before any RCU
610 * grace-period manupulations below.
611 */
612
613 smp_mb(); /* See above block comment. */
614
615 __rcu_process_callbacks(&rcu_ctrlblk, &__get_cpu_var(rcu_data));
616 __rcu_process_callbacks(&rcu_bh_ctrlblk, &__get_cpu_var(rcu_bh_data));
617
618 /*
619 * Memory references from any later RCU read-side critical sections
620 * executed by the interrupted code must be see after any RCU
621 * grace-period manupulations above.
622 */
623
624 smp_mb(); /* See above block comment. */
625}
626
627static int __rcu_pending(struct rcu_ctrlblk *rcp, struct rcu_data *rdp)
628{
629 /* Check for CPU stalls, if enabled. */
630 check_cpu_stall(rcp);
631
632 if (rdp->nxtlist) {
633 long completed_snap = ACCESS_ONCE(rcp->completed);
634
635 /*
636 * This cpu has pending rcu entries and the grace period
637 * for them has completed.
638 */
639 if (!rcu_batch_before(completed_snap, rdp->batch))
640 return 1;
641 if (!rcu_batch_before(completed_snap, rdp->batch - 1) &&
642 rdp->nxttail[0] != rdp->nxttail[1])
643 return 1;
644 if (rdp->nxttail[0] != &rdp->nxtlist)
645 return 1;
646
647 /*
648 * This cpu has pending rcu entries and the new batch
649 * for then hasn't been started nor scheduled start
650 */
651 if (rcu_batch_after(rdp->batch, rcp->pending))
652 return 1;
653 }
654
655 /* This cpu has finished callbacks to invoke */
656 if (rdp->donelist)
657 return 1;
658
659 /* The rcu core waits for a quiescent state from the cpu */
660 if (rdp->quiescbatch != rcp->cur || rdp->qs_pending)
661 return 1;
662
663 /* nothing to do */
664 return 0;
665}
666
667/*
668 * Check to see if there is any immediate RCU-related work to be done
669 * by the current CPU, returning 1 if so. This function is part of the
670 * RCU implementation; it is -not- an exported member of the RCU API.
671 */
672int rcu_pending(int cpu)
673{
674 return __rcu_pending(&rcu_ctrlblk, &per_cpu(rcu_data, cpu)) ||
675 __rcu_pending(&rcu_bh_ctrlblk, &per_cpu(rcu_bh_data, cpu));
676}
677
678/*
679 * Check to see if any future RCU-related work will need to be done
680 * by the current CPU, even if none need be done immediately, returning
681 * 1 if so. This function is part of the RCU implementation; it is -not-
682 * an exported member of the RCU API.
683 */
684int rcu_needs_cpu(int cpu)
685{
686 struct rcu_data *rdp = &per_cpu(rcu_data, cpu);
687 struct rcu_data *rdp_bh = &per_cpu(rcu_bh_data, cpu);
688
689 return !!rdp->nxtlist || !!rdp_bh->nxtlist || rcu_pending(cpu);
690}
691
692/*
693 * Top-level function driving RCU grace-period detection, normally
694 * invoked from the scheduler-clock interrupt. This function simply
695 * increments counters that are read only from softirq by this same
696 * CPU, so there are no memory barriers required.
697 */
698void rcu_check_callbacks(int cpu, int user)
699{
700 if (user ||
701 (idle_cpu(cpu) && rcu_scheduler_active &&
702 !in_softirq() && hardirq_count() <= (1 << HARDIRQ_SHIFT))) {
703
704 /*
705 * Get here if this CPU took its interrupt from user
706 * mode or from the idle loop, and if this is not a
707 * nested interrupt. In this case, the CPU is in
708 * a quiescent state, so count it.
709 *
710 * Also do a memory barrier. This is needed to handle
711 * the case where writes from a preempt-disable section
712 * of code get reordered into schedule() by this CPU's
713 * write buffer. The memory barrier makes sure that
714 * the rcu_qsctr_inc() and rcu_bh_qsctr_inc() are see
715 * by other CPUs to happen after any such write.
716 */
717
718 smp_mb(); /* See above block comment. */
719 rcu_qsctr_inc(cpu);
720 rcu_bh_qsctr_inc(cpu);
721
722 } else if (!in_softirq()) {
723
724 /*
725 * Get here if this CPU did not take its interrupt from
726 * softirq, in other words, if it is not interrupting
727 * a rcu_bh read-side critical section. This is an _bh
728 * critical section, so count it. The memory barrier
729 * is needed for the same reason as is the above one.
730 */
731
732 smp_mb(); /* See above block comment. */
733 rcu_bh_qsctr_inc(cpu);
734 }
735 raise_rcu_softirq();
736}
737
738static void __cpuinit rcu_init_percpu_data(int cpu, struct rcu_ctrlblk *rcp,
739 struct rcu_data *rdp)
740{
741 unsigned long flags;
742
743 spin_lock_irqsave(&rcp->lock, flags);
744 memset(rdp, 0, sizeof(*rdp));
745 rdp->nxttail[0] = rdp->nxttail[1] = rdp->nxttail[2] = &rdp->nxtlist;
746 rdp->donetail = &rdp->donelist;
747 rdp->quiescbatch = rcp->completed;
748 rdp->qs_pending = 0;
749 rdp->cpu = cpu;
750 rdp->blimit = blimit;
751 spin_unlock_irqrestore(&rcp->lock, flags);
752}
753
754static void __cpuinit rcu_online_cpu(int cpu)
755{
756 struct rcu_data *rdp = &per_cpu(rcu_data, cpu);
757 struct rcu_data *bh_rdp = &per_cpu(rcu_bh_data, cpu);
758
759 rcu_init_percpu_data(cpu, &rcu_ctrlblk, rdp);
760 rcu_init_percpu_data(cpu, &rcu_bh_ctrlblk, bh_rdp);
761 open_softirq(RCU_SOFTIRQ, rcu_process_callbacks);
762}
763
764static int __cpuinit rcu_cpu_notify(struct notifier_block *self,
765 unsigned long action, void *hcpu)
766{
767 long cpu = (long)hcpu;
768
769 switch (action) {
770 case CPU_UP_PREPARE:
771 case CPU_UP_PREPARE_FROZEN:
772 rcu_online_cpu(cpu);
773 break;
774 case CPU_DEAD:
775 case CPU_DEAD_FROZEN:
776 rcu_offline_cpu(cpu);
777 break;
778 default:
779 break;
780 }
781 return NOTIFY_OK;
782}
783
784static struct notifier_block __cpuinitdata rcu_nb = {
785 .notifier_call = rcu_cpu_notify,
786};
787
788/*
789 * Initializes rcu mechanism. Assumed to be called early.
790 * That is before local timer(SMP) or jiffie timer (uniproc) is setup.
791 * Note that rcu_qsctr and friends are implicitly
792 * initialized due to the choice of ``0'' for RCU_CTR_INVALID.
793 */
794void __init __rcu_init(void)
795{
796#ifdef CONFIG_RCU_CPU_STALL_DETECTOR
797 printk(KERN_INFO "RCU-based detection of stalled CPUs is enabled.\n");
798#endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
799 rcu_cpu_notify(&rcu_nb, CPU_UP_PREPARE,
800 (void *)(long)smp_processor_id());
801 /* Register notifier for non-boot CPUs */
802 register_cpu_notifier(&rcu_nb);
803}
804
805module_param(blimit, int, 0);
806module_param(qhimark, int, 0);
807module_param(qlowmark, int, 0);
diff --git a/kernel/rcupdate.c b/kernel/rcupdate.c
index a967c9feb90a..bd5d5c8e5140 100644
--- a/kernel/rcupdate.c
+++ b/kernel/rcupdate.c
@@ -98,6 +98,30 @@ void synchronize_rcu(void)
98} 98}
99EXPORT_SYMBOL_GPL(synchronize_rcu); 99EXPORT_SYMBOL_GPL(synchronize_rcu);
100 100
101/**
102 * synchronize_rcu_bh - wait until an rcu_bh grace period has elapsed.
103 *
104 * Control will return to the caller some time after a full rcu_bh grace
105 * period has elapsed, in other words after all currently executing rcu_bh
106 * read-side critical sections have completed. RCU read-side critical
107 * sections are delimited by rcu_read_lock_bh() and rcu_read_unlock_bh(),
108 * and may be nested.
109 */
110void synchronize_rcu_bh(void)
111{
112 struct rcu_synchronize rcu;
113
114 if (rcu_blocking_is_gp())
115 return;
116
117 init_completion(&rcu.completion);
118 /* Will wake me after RCU finished. */
119 call_rcu_bh(&rcu.head, wakeme_after_rcu);
120 /* Wait for it. */
121 wait_for_completion(&rcu.completion);
122}
123EXPORT_SYMBOL_GPL(synchronize_rcu_bh);
124
101static void rcu_barrier_callback(struct rcu_head *notused) 125static void rcu_barrier_callback(struct rcu_head *notused)
102{ 126{
103 if (atomic_dec_and_test(&rcu_barrier_cpu_count)) 127 if (atomic_dec_and_test(&rcu_barrier_cpu_count))
@@ -129,6 +153,7 @@ static void rcu_barrier_func(void *type)
129static inline void wait_migrated_callbacks(void) 153static inline void wait_migrated_callbacks(void)
130{ 154{
131 wait_event(rcu_migrate_wq, !atomic_read(&rcu_migrate_type_count)); 155 wait_event(rcu_migrate_wq, !atomic_read(&rcu_migrate_type_count));
156 smp_mb(); /* In case we didn't sleep. */
132} 157}
133 158
134/* 159/*
@@ -192,9 +217,13 @@ static void rcu_migrate_callback(struct rcu_head *notused)
192 wake_up(&rcu_migrate_wq); 217 wake_up(&rcu_migrate_wq);
193} 218}
194 219
220extern int rcu_cpu_notify(struct notifier_block *self,
221 unsigned long action, void *hcpu);
222
195static int __cpuinit rcu_barrier_cpu_hotplug(struct notifier_block *self, 223static int __cpuinit rcu_barrier_cpu_hotplug(struct notifier_block *self,
196 unsigned long action, void *hcpu) 224 unsigned long action, void *hcpu)
197{ 225{
226 rcu_cpu_notify(self, action, hcpu);
198 if (action == CPU_DYING) { 227 if (action == CPU_DYING) {
199 /* 228 /*
200 * preempt_disable() in on_each_cpu() prevents stop_machine(), 229 * preempt_disable() in on_each_cpu() prevents stop_machine(),
@@ -209,7 +238,8 @@ static int __cpuinit rcu_barrier_cpu_hotplug(struct notifier_block *self,
209 call_rcu_bh(rcu_migrate_head, rcu_migrate_callback); 238 call_rcu_bh(rcu_migrate_head, rcu_migrate_callback);
210 call_rcu_sched(rcu_migrate_head + 1, rcu_migrate_callback); 239 call_rcu_sched(rcu_migrate_head + 1, rcu_migrate_callback);
211 call_rcu(rcu_migrate_head + 2, rcu_migrate_callback); 240 call_rcu(rcu_migrate_head + 2, rcu_migrate_callback);
212 } else if (action == CPU_POST_DEAD) { 241 } else if (action == CPU_DOWN_PREPARE) {
242 /* Don't need to wait until next removal operation. */
213 /* rcu_migrate_head is protected by cpu_add_remove_lock */ 243 /* rcu_migrate_head is protected by cpu_add_remove_lock */
214 wait_migrated_callbacks(); 244 wait_migrated_callbacks();
215 } 245 }
@@ -219,8 +249,18 @@ static int __cpuinit rcu_barrier_cpu_hotplug(struct notifier_block *self,
219 249
220void __init rcu_init(void) 250void __init rcu_init(void)
221{ 251{
252 int i;
253
222 __rcu_init(); 254 __rcu_init();
223 hotcpu_notifier(rcu_barrier_cpu_hotplug, 0); 255 cpu_notifier(rcu_barrier_cpu_hotplug, 0);
256
257 /*
258 * We don't need protection against CPU-hotplug here because
259 * this is called early in boot, before either interrupts
260 * or the scheduler are operational.
261 */
262 for_each_online_cpu(i)
263 rcu_barrier_cpu_hotplug(NULL, CPU_UP_PREPARE, (void *)(long)i);
224} 264}
225 265
226void rcu_scheduler_starting(void) 266void rcu_scheduler_starting(void)
diff --git a/kernel/rcupreempt.c b/kernel/rcupreempt.c
deleted file mode 100644
index beb0e659adcc..000000000000
--- a/kernel/rcupreempt.c
+++ /dev/null
@@ -1,1539 +0,0 @@
1/*
2 * Read-Copy Update mechanism for mutual exclusion, realtime implementation
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, 2006
19 *
20 * Authors: Paul E. McKenney <paulmck@us.ibm.com>
21 * With thanks to Esben Nielsen, Bill Huey, and Ingo Molnar
22 * for pushing me away from locks and towards counters, and
23 * to Suparna Bhattacharya for pushing me completely away
24 * from atomic instructions on the read side.
25 *
26 * - Added handling of Dynamic Ticks
27 * Copyright 2007 - Paul E. Mckenney <paulmck@us.ibm.com>
28 * - Steven Rostedt <srostedt@redhat.com>
29 *
30 * Papers: http://www.rdrop.com/users/paulmck/RCU
31 *
32 * Design Document: http://lwn.net/Articles/253651/
33 *
34 * For detailed explanation of Read-Copy Update mechanism see -
35 * Documentation/RCU/ *.txt
36 *
37 */
38#include <linux/types.h>
39#include <linux/kernel.h>
40#include <linux/init.h>
41#include <linux/spinlock.h>
42#include <linux/smp.h>
43#include <linux/rcupdate.h>
44#include <linux/interrupt.h>
45#include <linux/sched.h>
46#include <asm/atomic.h>
47#include <linux/bitops.h>
48#include <linux/module.h>
49#include <linux/kthread.h>
50#include <linux/completion.h>
51#include <linux/moduleparam.h>
52#include <linux/percpu.h>
53#include <linux/notifier.h>
54#include <linux/cpu.h>
55#include <linux/random.h>
56#include <linux/delay.h>
57#include <linux/cpumask.h>
58#include <linux/rcupreempt_trace.h>
59#include <asm/byteorder.h>
60
61/*
62 * PREEMPT_RCU data structures.
63 */
64
65/*
66 * GP_STAGES specifies the number of times the state machine has
67 * to go through the all the rcu_try_flip_states (see below)
68 * in a single Grace Period.
69 *
70 * GP in GP_STAGES stands for Grace Period ;)
71 */
72#define GP_STAGES 2
73struct rcu_data {
74 spinlock_t lock; /* Protect rcu_data fields. */
75 long completed; /* Number of last completed batch. */
76 int waitlistcount;
77 struct rcu_head *nextlist;
78 struct rcu_head **nexttail;
79 struct rcu_head *waitlist[GP_STAGES];
80 struct rcu_head **waittail[GP_STAGES];
81 struct rcu_head *donelist; /* from waitlist & waitschedlist */
82 struct rcu_head **donetail;
83 long rcu_flipctr[2];
84 struct rcu_head *nextschedlist;
85 struct rcu_head **nextschedtail;
86 struct rcu_head *waitschedlist;
87 struct rcu_head **waitschedtail;
88 int rcu_sched_sleeping;
89#ifdef CONFIG_RCU_TRACE
90 struct rcupreempt_trace trace;
91#endif /* #ifdef CONFIG_RCU_TRACE */
92};
93
94/*
95 * States for rcu_try_flip() and friends.
96 */
97
98enum rcu_try_flip_states {
99
100 /*
101 * Stay here if nothing is happening. Flip the counter if somthing
102 * starts happening. Denoted by "I"
103 */
104 rcu_try_flip_idle_state,
105
106 /*
107 * Wait here for all CPUs to notice that the counter has flipped. This
108 * prevents the old set of counters from ever being incremented once
109 * we leave this state, which in turn is necessary because we cannot
110 * test any individual counter for zero -- we can only check the sum.
111 * Denoted by "A".
112 */
113 rcu_try_flip_waitack_state,
114
115 /*
116 * Wait here for the sum of the old per-CPU counters to reach zero.
117 * Denoted by "Z".
118 */
119 rcu_try_flip_waitzero_state,
120
121 /*
122 * Wait here for each of the other CPUs to execute a memory barrier.
123 * This is necessary to ensure that these other CPUs really have
124 * completed executing their RCU read-side critical sections, despite
125 * their CPUs wildly reordering memory. Denoted by "M".
126 */
127 rcu_try_flip_waitmb_state,
128};
129
130/*
131 * States for rcu_ctrlblk.rcu_sched_sleep.
132 */
133
134enum rcu_sched_sleep_states {
135 rcu_sched_not_sleeping, /* Not sleeping, callbacks need GP. */
136 rcu_sched_sleep_prep, /* Thinking of sleeping, rechecking. */
137 rcu_sched_sleeping, /* Sleeping, awaken if GP needed. */
138};
139
140struct rcu_ctrlblk {
141 spinlock_t fliplock; /* Protect state-machine transitions. */
142 long completed; /* Number of last completed batch. */
143 enum rcu_try_flip_states rcu_try_flip_state; /* The current state of
144 the rcu state machine */
145 spinlock_t schedlock; /* Protect rcu_sched sleep state. */
146 enum rcu_sched_sleep_states sched_sleep; /* rcu_sched state. */
147 wait_queue_head_t sched_wq; /* Place for rcu_sched to sleep. */
148};
149
150struct rcu_dyntick_sched {
151 int dynticks;
152 int dynticks_snap;
153 int sched_qs;
154 int sched_qs_snap;
155 int sched_dynticks_snap;
156};
157
158static DEFINE_PER_CPU_SHARED_ALIGNED(struct rcu_dyntick_sched, rcu_dyntick_sched) = {
159 .dynticks = 1,
160};
161
162void rcu_qsctr_inc(int cpu)
163{
164 struct rcu_dyntick_sched *rdssp = &per_cpu(rcu_dyntick_sched, cpu);
165
166 rdssp->sched_qs++;
167}
168
169#ifdef CONFIG_NO_HZ
170
171void rcu_enter_nohz(void)
172{
173 static DEFINE_RATELIMIT_STATE(rs, 10 * HZ, 1);
174
175 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
176 __get_cpu_var(rcu_dyntick_sched).dynticks++;
177 WARN_ON_RATELIMIT(__get_cpu_var(rcu_dyntick_sched).dynticks & 0x1, &rs);
178}
179
180void rcu_exit_nohz(void)
181{
182 static DEFINE_RATELIMIT_STATE(rs, 10 * HZ, 1);
183
184 __get_cpu_var(rcu_dyntick_sched).dynticks++;
185 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
186 WARN_ON_RATELIMIT(!(__get_cpu_var(rcu_dyntick_sched).dynticks & 0x1),
187 &rs);
188}
189
190#endif /* CONFIG_NO_HZ */
191
192
193static DEFINE_PER_CPU(struct rcu_data, rcu_data);
194
195static struct rcu_ctrlblk rcu_ctrlblk = {
196 .fliplock = __SPIN_LOCK_UNLOCKED(rcu_ctrlblk.fliplock),
197 .completed = 0,
198 .rcu_try_flip_state = rcu_try_flip_idle_state,
199 .schedlock = __SPIN_LOCK_UNLOCKED(rcu_ctrlblk.schedlock),
200 .sched_sleep = rcu_sched_not_sleeping,
201 .sched_wq = __WAIT_QUEUE_HEAD_INITIALIZER(rcu_ctrlblk.sched_wq),
202};
203
204static struct task_struct *rcu_sched_grace_period_task;
205
206#ifdef CONFIG_RCU_TRACE
207static char *rcu_try_flip_state_names[] =
208 { "idle", "waitack", "waitzero", "waitmb" };
209#endif /* #ifdef CONFIG_RCU_TRACE */
210
211static DECLARE_BITMAP(rcu_cpu_online_map, NR_CPUS) __read_mostly
212 = CPU_BITS_NONE;
213
214/*
215 * Enum and per-CPU flag to determine when each CPU has seen
216 * the most recent counter flip.
217 */
218
219enum rcu_flip_flag_values {
220 rcu_flip_seen, /* Steady/initial state, last flip seen. */
221 /* Only GP detector can update. */
222 rcu_flipped /* Flip just completed, need confirmation. */
223 /* Only corresponding CPU can update. */
224};
225static DEFINE_PER_CPU_SHARED_ALIGNED(enum rcu_flip_flag_values, rcu_flip_flag)
226 = rcu_flip_seen;
227
228/*
229 * Enum and per-CPU flag to determine when each CPU has executed the
230 * needed memory barrier to fence in memory references from its last RCU
231 * read-side critical section in the just-completed grace period.
232 */
233
234enum rcu_mb_flag_values {
235 rcu_mb_done, /* Steady/initial state, no mb()s required. */
236 /* Only GP detector can update. */
237 rcu_mb_needed /* Flip just completed, need an mb(). */
238 /* Only corresponding CPU can update. */
239};
240static DEFINE_PER_CPU_SHARED_ALIGNED(enum rcu_mb_flag_values, rcu_mb_flag)
241 = rcu_mb_done;
242
243/*
244 * RCU_DATA_ME: find the current CPU's rcu_data structure.
245 * RCU_DATA_CPU: find the specified CPU's rcu_data structure.
246 */
247#define RCU_DATA_ME() (&__get_cpu_var(rcu_data))
248#define RCU_DATA_CPU(cpu) (&per_cpu(rcu_data, cpu))
249
250/*
251 * Helper macro for tracing when the appropriate rcu_data is not
252 * cached in a local variable, but where the CPU number is so cached.
253 */
254#define RCU_TRACE_CPU(f, cpu) RCU_TRACE(f, &(RCU_DATA_CPU(cpu)->trace));
255
256/*
257 * Helper macro for tracing when the appropriate rcu_data is not
258 * cached in a local variable.
259 */
260#define RCU_TRACE_ME(f) RCU_TRACE(f, &(RCU_DATA_ME()->trace));
261
262/*
263 * Helper macro for tracing when the appropriate rcu_data is pointed
264 * to by a local variable.
265 */
266#define RCU_TRACE_RDP(f, rdp) RCU_TRACE(f, &((rdp)->trace));
267
268#define RCU_SCHED_BATCH_TIME (HZ / 50)
269
270/*
271 * Return the number of RCU batches processed thus far. Useful
272 * for debug and statistics.
273 */
274long rcu_batches_completed(void)
275{
276 return rcu_ctrlblk.completed;
277}
278EXPORT_SYMBOL_GPL(rcu_batches_completed);
279
280void __rcu_read_lock(void)
281{
282 int idx;
283 struct task_struct *t = current;
284 int nesting;
285
286 nesting = ACCESS_ONCE(t->rcu_read_lock_nesting);
287 if (nesting != 0) {
288
289 /* An earlier rcu_read_lock() covers us, just count it. */
290
291 t->rcu_read_lock_nesting = nesting + 1;
292
293 } else {
294 unsigned long flags;
295
296 /*
297 * We disable interrupts for the following reasons:
298 * - If we get scheduling clock interrupt here, and we
299 * end up acking the counter flip, it's like a promise
300 * that we will never increment the old counter again.
301 * Thus we will break that promise if that
302 * scheduling clock interrupt happens between the time
303 * we pick the .completed field and the time that we
304 * increment our counter.
305 *
306 * - We don't want to be preempted out here.
307 *
308 * NMIs can still occur, of course, and might themselves
309 * contain rcu_read_lock().
310 */
311
312 local_irq_save(flags);
313
314 /*
315 * Outermost nesting of rcu_read_lock(), so increment
316 * the current counter for the current CPU. Use volatile
317 * casts to prevent the compiler from reordering.
318 */
319
320 idx = ACCESS_ONCE(rcu_ctrlblk.completed) & 0x1;
321 ACCESS_ONCE(RCU_DATA_ME()->rcu_flipctr[idx])++;
322
323 /*
324 * Now that the per-CPU counter has been incremented, we
325 * are protected from races with rcu_read_lock() invoked
326 * from NMI handlers on this CPU. We can therefore safely
327 * increment the nesting counter, relieving further NMIs
328 * of the need to increment the per-CPU counter.
329 */
330
331 ACCESS_ONCE(t->rcu_read_lock_nesting) = nesting + 1;
332
333 /*
334 * Now that we have preventing any NMIs from storing
335 * to the ->rcu_flipctr_idx, we can safely use it to
336 * remember which counter to decrement in the matching
337 * rcu_read_unlock().
338 */
339
340 ACCESS_ONCE(t->rcu_flipctr_idx) = idx;
341 local_irq_restore(flags);
342 }
343}
344EXPORT_SYMBOL_GPL(__rcu_read_lock);
345
346void __rcu_read_unlock(void)
347{
348 int idx;
349 struct task_struct *t = current;
350 int nesting;
351
352 nesting = ACCESS_ONCE(t->rcu_read_lock_nesting);
353 if (nesting > 1) {
354
355 /*
356 * We are still protected by the enclosing rcu_read_lock(),
357 * so simply decrement the counter.
358 */
359
360 t->rcu_read_lock_nesting = nesting - 1;
361
362 } else {
363 unsigned long flags;
364
365 /*
366 * Disable local interrupts to prevent the grace-period
367 * detection state machine from seeing us half-done.
368 * NMIs can still occur, of course, and might themselves
369 * contain rcu_read_lock() and rcu_read_unlock().
370 */
371
372 local_irq_save(flags);
373
374 /*
375 * Outermost nesting of rcu_read_unlock(), so we must
376 * decrement the current counter for the current CPU.
377 * This must be done carefully, because NMIs can
378 * occur at any point in this code, and any rcu_read_lock()
379 * and rcu_read_unlock() pairs in the NMI handlers
380 * must interact non-destructively with this code.
381 * Lots of volatile casts, and -very- careful ordering.
382 *
383 * Changes to this code, including this one, must be
384 * inspected, validated, and tested extremely carefully!!!
385 */
386
387 /*
388 * First, pick up the index.
389 */
390
391 idx = ACCESS_ONCE(t->rcu_flipctr_idx);
392
393 /*
394 * Now that we have fetched the counter index, it is
395 * safe to decrement the per-task RCU nesting counter.
396 * After this, any interrupts or NMIs will increment and
397 * decrement the per-CPU counters.
398 */
399 ACCESS_ONCE(t->rcu_read_lock_nesting) = nesting - 1;
400
401 /*
402 * It is now safe to decrement this task's nesting count.
403 * NMIs that occur after this statement will route their
404 * rcu_read_lock() calls through this "else" clause, and
405 * will thus start incrementing the per-CPU counter on
406 * their own. They will also clobber ->rcu_flipctr_idx,
407 * but that is OK, since we have already fetched it.
408 */
409
410 ACCESS_ONCE(RCU_DATA_ME()->rcu_flipctr[idx])--;
411 local_irq_restore(flags);
412 }
413}
414EXPORT_SYMBOL_GPL(__rcu_read_unlock);
415
416/*
417 * If a global counter flip has occurred since the last time that we
418 * advanced callbacks, advance them. Hardware interrupts must be
419 * disabled when calling this function.
420 */
421static void __rcu_advance_callbacks(struct rcu_data *rdp)
422{
423 int cpu;
424 int i;
425 int wlc = 0;
426
427 if (rdp->completed != rcu_ctrlblk.completed) {
428 if (rdp->waitlist[GP_STAGES - 1] != NULL) {
429 *rdp->donetail = rdp->waitlist[GP_STAGES - 1];
430 rdp->donetail = rdp->waittail[GP_STAGES - 1];
431 RCU_TRACE_RDP(rcupreempt_trace_move2done, rdp);
432 }
433 for (i = GP_STAGES - 2; i >= 0; i--) {
434 if (rdp->waitlist[i] != NULL) {
435 rdp->waitlist[i + 1] = rdp->waitlist[i];
436 rdp->waittail[i + 1] = rdp->waittail[i];
437 wlc++;
438 } else {
439 rdp->waitlist[i + 1] = NULL;
440 rdp->waittail[i + 1] =
441 &rdp->waitlist[i + 1];
442 }
443 }
444 if (rdp->nextlist != NULL) {
445 rdp->waitlist[0] = rdp->nextlist;
446 rdp->waittail[0] = rdp->nexttail;
447 wlc++;
448 rdp->nextlist = NULL;
449 rdp->nexttail = &rdp->nextlist;
450 RCU_TRACE_RDP(rcupreempt_trace_move2wait, rdp);
451 } else {
452 rdp->waitlist[0] = NULL;
453 rdp->waittail[0] = &rdp->waitlist[0];
454 }
455 rdp->waitlistcount = wlc;
456 rdp->completed = rcu_ctrlblk.completed;
457 }
458
459 /*
460 * Check to see if this CPU needs to report that it has seen
461 * the most recent counter flip, thereby declaring that all
462 * subsequent rcu_read_lock() invocations will respect this flip.
463 */
464
465 cpu = raw_smp_processor_id();
466 if (per_cpu(rcu_flip_flag, cpu) == rcu_flipped) {
467 smp_mb(); /* Subsequent counter accesses must see new value */
468 per_cpu(rcu_flip_flag, cpu) = rcu_flip_seen;
469 smp_mb(); /* Subsequent RCU read-side critical sections */
470 /* seen -after- acknowledgement. */
471 }
472}
473
474#ifdef CONFIG_NO_HZ
475static DEFINE_PER_CPU(int, rcu_update_flag);
476
477/**
478 * rcu_irq_enter - Called from Hard irq handlers and NMI/SMI.
479 *
480 * If the CPU was idle with dynamic ticks active, this updates the
481 * rcu_dyntick_sched.dynticks to let the RCU handling know that the
482 * CPU is active.
483 */
484void rcu_irq_enter(void)
485{
486 int cpu = smp_processor_id();
487 struct rcu_dyntick_sched *rdssp = &per_cpu(rcu_dyntick_sched, cpu);
488
489 if (per_cpu(rcu_update_flag, cpu))
490 per_cpu(rcu_update_flag, cpu)++;
491
492 /*
493 * Only update if we are coming from a stopped ticks mode
494 * (rcu_dyntick_sched.dynticks is even).
495 */
496 if (!in_interrupt() &&
497 (rdssp->dynticks & 0x1) == 0) {
498 /*
499 * The following might seem like we could have a race
500 * with NMI/SMIs. But this really isn't a problem.
501 * Here we do a read/modify/write, and the race happens
502 * when an NMI/SMI comes in after the read and before
503 * the write. But NMI/SMIs will increment this counter
504 * twice before returning, so the zero bit will not
505 * be corrupted by the NMI/SMI which is the most important
506 * part.
507 *
508 * The only thing is that we would bring back the counter
509 * to a postion that it was in during the NMI/SMI.
510 * But the zero bit would be set, so the rest of the
511 * counter would again be ignored.
512 *
513 * On return from the IRQ, the counter may have the zero
514 * bit be 0 and the counter the same as the return from
515 * the NMI/SMI. If the state machine was so unlucky to
516 * see that, it still doesn't matter, since all
517 * RCU read-side critical sections on this CPU would
518 * have already completed.
519 */
520 rdssp->dynticks++;
521 /*
522 * The following memory barrier ensures that any
523 * rcu_read_lock() primitives in the irq handler
524 * are seen by other CPUs to follow the above
525 * increment to rcu_dyntick_sched.dynticks. This is
526 * required in order for other CPUs to correctly
527 * determine when it is safe to advance the RCU
528 * grace-period state machine.
529 */
530 smp_mb(); /* see above block comment. */
531 /*
532 * Since we can't determine the dynamic tick mode from
533 * the rcu_dyntick_sched.dynticks after this routine,
534 * we use a second flag to acknowledge that we came
535 * from an idle state with ticks stopped.
536 */
537 per_cpu(rcu_update_flag, cpu)++;
538 /*
539 * If we take an NMI/SMI now, they will also increment
540 * the rcu_update_flag, and will not update the
541 * rcu_dyntick_sched.dynticks on exit. That is for
542 * this IRQ to do.
543 */
544 }
545}
546
547/**
548 * rcu_irq_exit - Called from exiting Hard irq context.
549 *
550 * If the CPU was idle with dynamic ticks active, update the
551 * rcu_dyntick_sched.dynticks to put let the RCU handling be
552 * aware that the CPU is going back to idle with no ticks.
553 */
554void rcu_irq_exit(void)
555{
556 int cpu = smp_processor_id();
557 struct rcu_dyntick_sched *rdssp = &per_cpu(rcu_dyntick_sched, cpu);
558
559 /*
560 * rcu_update_flag is set if we interrupted the CPU
561 * when it was idle with ticks stopped.
562 * Once this occurs, we keep track of interrupt nesting
563 * because a NMI/SMI could also come in, and we still
564 * only want the IRQ that started the increment of the
565 * rcu_dyntick_sched.dynticks to be the one that modifies
566 * it on exit.
567 */
568 if (per_cpu(rcu_update_flag, cpu)) {
569 if (--per_cpu(rcu_update_flag, cpu))
570 return;
571
572 /* This must match the interrupt nesting */
573 WARN_ON(in_interrupt());
574
575 /*
576 * If an NMI/SMI happens now we are still
577 * protected by the rcu_dyntick_sched.dynticks being odd.
578 */
579
580 /*
581 * The following memory barrier ensures that any
582 * rcu_read_unlock() primitives in the irq handler
583 * are seen by other CPUs to preceed the following
584 * increment to rcu_dyntick_sched.dynticks. This
585 * is required in order for other CPUs to determine
586 * when it is safe to advance the RCU grace-period
587 * state machine.
588 */
589 smp_mb(); /* see above block comment. */
590 rdssp->dynticks++;
591 WARN_ON(rdssp->dynticks & 0x1);
592 }
593}
594
595void rcu_nmi_enter(void)
596{
597 rcu_irq_enter();
598}
599
600void rcu_nmi_exit(void)
601{
602 rcu_irq_exit();
603}
604
605static void dyntick_save_progress_counter(int cpu)
606{
607 struct rcu_dyntick_sched *rdssp = &per_cpu(rcu_dyntick_sched, cpu);
608
609 rdssp->dynticks_snap = rdssp->dynticks;
610}
611
612static inline int
613rcu_try_flip_waitack_needed(int cpu)
614{
615 long curr;
616 long snap;
617 struct rcu_dyntick_sched *rdssp = &per_cpu(rcu_dyntick_sched, cpu);
618
619 curr = rdssp->dynticks;
620 snap = rdssp->dynticks_snap;
621 smp_mb(); /* force ordering with cpu entering/leaving dynticks. */
622
623 /*
624 * If the CPU remained in dynticks mode for the entire time
625 * and didn't take any interrupts, NMIs, SMIs, or whatever,
626 * then it cannot be in the middle of an rcu_read_lock(), so
627 * the next rcu_read_lock() it executes must use the new value
628 * of the counter. So we can safely pretend that this CPU
629 * already acknowledged the counter.
630 */
631
632 if ((curr == snap) && ((curr & 0x1) == 0))
633 return 0;
634
635 /*
636 * If the CPU passed through or entered a dynticks idle phase with
637 * no active irq handlers, then, as above, we can safely pretend
638 * that this CPU already acknowledged the counter.
639 */
640
641 if ((curr - snap) > 2 || (curr & 0x1) == 0)
642 return 0;
643
644 /* We need this CPU to explicitly acknowledge the counter flip. */
645
646 return 1;
647}
648
649static inline int
650rcu_try_flip_waitmb_needed(int cpu)
651{
652 long curr;
653 long snap;
654 struct rcu_dyntick_sched *rdssp = &per_cpu(rcu_dyntick_sched, cpu);
655
656 curr = rdssp->dynticks;
657 snap = rdssp->dynticks_snap;
658 smp_mb(); /* force ordering with cpu entering/leaving dynticks. */
659
660 /*
661 * If the CPU remained in dynticks mode for the entire time
662 * and didn't take any interrupts, NMIs, SMIs, or whatever,
663 * then it cannot have executed an RCU read-side critical section
664 * during that time, so there is no need for it to execute a
665 * memory barrier.
666 */
667
668 if ((curr == snap) && ((curr & 0x1) == 0))
669 return 0;
670
671 /*
672 * If the CPU either entered or exited an outermost interrupt,
673 * SMI, NMI, or whatever handler, then we know that it executed
674 * a memory barrier when doing so. So we don't need another one.
675 */
676 if (curr != snap)
677 return 0;
678
679 /* We need the CPU to execute a memory barrier. */
680
681 return 1;
682}
683
684static void dyntick_save_progress_counter_sched(int cpu)
685{
686 struct rcu_dyntick_sched *rdssp = &per_cpu(rcu_dyntick_sched, cpu);
687
688 rdssp->sched_dynticks_snap = rdssp->dynticks;
689}
690
691static int rcu_qsctr_inc_needed_dyntick(int cpu)
692{
693 long curr;
694 long snap;
695 struct rcu_dyntick_sched *rdssp = &per_cpu(rcu_dyntick_sched, cpu);
696
697 curr = rdssp->dynticks;
698 snap = rdssp->sched_dynticks_snap;
699 smp_mb(); /* force ordering with cpu entering/leaving dynticks. */
700
701 /*
702 * If the CPU remained in dynticks mode for the entire time
703 * and didn't take any interrupts, NMIs, SMIs, or whatever,
704 * then it cannot be in the middle of an rcu_read_lock(), so
705 * the next rcu_read_lock() it executes must use the new value
706 * of the counter. Therefore, this CPU has been in a quiescent
707 * state the entire time, and we don't need to wait for it.
708 */
709
710 if ((curr == snap) && ((curr & 0x1) == 0))
711 return 0;
712
713 /*
714 * If the CPU passed through or entered a dynticks idle phase with
715 * no active irq handlers, then, as above, this CPU has already
716 * passed through a quiescent state.
717 */
718
719 if ((curr - snap) > 2 || (snap & 0x1) == 0)
720 return 0;
721
722 /* We need this CPU to go through a quiescent state. */
723
724 return 1;
725}
726
727#else /* !CONFIG_NO_HZ */
728
729# define dyntick_save_progress_counter(cpu) do { } while (0)
730# define rcu_try_flip_waitack_needed(cpu) (1)
731# define rcu_try_flip_waitmb_needed(cpu) (1)
732
733# define dyntick_save_progress_counter_sched(cpu) do { } while (0)
734# define rcu_qsctr_inc_needed_dyntick(cpu) (1)
735
736#endif /* CONFIG_NO_HZ */
737
738static void save_qsctr_sched(int cpu)
739{
740 struct rcu_dyntick_sched *rdssp = &per_cpu(rcu_dyntick_sched, cpu);
741
742 rdssp->sched_qs_snap = rdssp->sched_qs;
743}
744
745static inline int rcu_qsctr_inc_needed(int cpu)
746{
747 struct rcu_dyntick_sched *rdssp = &per_cpu(rcu_dyntick_sched, cpu);
748
749 /*
750 * If there has been a quiescent state, no more need to wait
751 * on this CPU.
752 */
753
754 if (rdssp->sched_qs != rdssp->sched_qs_snap) {
755 smp_mb(); /* force ordering with cpu entering schedule(). */
756 return 0;
757 }
758
759 /* We need this CPU to go through a quiescent state. */
760
761 return 1;
762}
763
764/*
765 * Get here when RCU is idle. Decide whether we need to
766 * move out of idle state, and return non-zero if so.
767 * "Straightforward" approach for the moment, might later
768 * use callback-list lengths, grace-period duration, or
769 * some such to determine when to exit idle state.
770 * Might also need a pre-idle test that does not acquire
771 * the lock, but let's get the simple case working first...
772 */
773
774static int
775rcu_try_flip_idle(void)
776{
777 int cpu;
778
779 RCU_TRACE_ME(rcupreempt_trace_try_flip_i1);
780 if (!rcu_pending(smp_processor_id())) {
781 RCU_TRACE_ME(rcupreempt_trace_try_flip_ie1);
782 return 0;
783 }
784
785 /*
786 * Do the flip.
787 */
788
789 RCU_TRACE_ME(rcupreempt_trace_try_flip_g1);
790 rcu_ctrlblk.completed++; /* stands in for rcu_try_flip_g2 */
791
792 /*
793 * Need a memory barrier so that other CPUs see the new
794 * counter value before they see the subsequent change of all
795 * the rcu_flip_flag instances to rcu_flipped.
796 */
797
798 smp_mb(); /* see above block comment. */
799
800 /* Now ask each CPU for acknowledgement of the flip. */
801
802 for_each_cpu(cpu, to_cpumask(rcu_cpu_online_map)) {
803 per_cpu(rcu_flip_flag, cpu) = rcu_flipped;
804 dyntick_save_progress_counter(cpu);
805 }
806
807 return 1;
808}
809
810/*
811 * Wait for CPUs to acknowledge the flip.
812 */
813
814static int
815rcu_try_flip_waitack(void)
816{
817 int cpu;
818
819 RCU_TRACE_ME(rcupreempt_trace_try_flip_a1);
820 for_each_cpu(cpu, to_cpumask(rcu_cpu_online_map))
821 if (rcu_try_flip_waitack_needed(cpu) &&
822 per_cpu(rcu_flip_flag, cpu) != rcu_flip_seen) {
823 RCU_TRACE_ME(rcupreempt_trace_try_flip_ae1);
824 return 0;
825 }
826
827 /*
828 * Make sure our checks above don't bleed into subsequent
829 * waiting for the sum of the counters to reach zero.
830 */
831
832 smp_mb(); /* see above block comment. */
833 RCU_TRACE_ME(rcupreempt_trace_try_flip_a2);
834 return 1;
835}
836
837/*
838 * Wait for collective ``last'' counter to reach zero,
839 * then tell all CPUs to do an end-of-grace-period memory barrier.
840 */
841
842static int
843rcu_try_flip_waitzero(void)
844{
845 int cpu;
846 int lastidx = !(rcu_ctrlblk.completed & 0x1);
847 int sum = 0;
848
849 /* Check to see if the sum of the "last" counters is zero. */
850
851 RCU_TRACE_ME(rcupreempt_trace_try_flip_z1);
852 for_each_cpu(cpu, to_cpumask(rcu_cpu_online_map))
853 sum += RCU_DATA_CPU(cpu)->rcu_flipctr[lastidx];
854 if (sum != 0) {
855 RCU_TRACE_ME(rcupreempt_trace_try_flip_ze1);
856 return 0;
857 }
858
859 /*
860 * This ensures that the other CPUs see the call for
861 * memory barriers -after- the sum to zero has been
862 * detected here
863 */
864 smp_mb(); /* ^^^^^^^^^^^^ */
865
866 /* Call for a memory barrier from each CPU. */
867 for_each_cpu(cpu, to_cpumask(rcu_cpu_online_map)) {
868 per_cpu(rcu_mb_flag, cpu) = rcu_mb_needed;
869 dyntick_save_progress_counter(cpu);
870 }
871
872 RCU_TRACE_ME(rcupreempt_trace_try_flip_z2);
873 return 1;
874}
875
876/*
877 * Wait for all CPUs to do their end-of-grace-period memory barrier.
878 * Return 0 once all CPUs have done so.
879 */
880
881static int
882rcu_try_flip_waitmb(void)
883{
884 int cpu;
885
886 RCU_TRACE_ME(rcupreempt_trace_try_flip_m1);
887 for_each_cpu(cpu, to_cpumask(rcu_cpu_online_map))
888 if (rcu_try_flip_waitmb_needed(cpu) &&
889 per_cpu(rcu_mb_flag, cpu) != rcu_mb_done) {
890 RCU_TRACE_ME(rcupreempt_trace_try_flip_me1);
891 return 0;
892 }
893
894 smp_mb(); /* Ensure that the above checks precede any following flip. */
895 RCU_TRACE_ME(rcupreempt_trace_try_flip_m2);
896 return 1;
897}
898
899/*
900 * Attempt a single flip of the counters. Remember, a single flip does
901 * -not- constitute a grace period. Instead, the interval between
902 * at least GP_STAGES consecutive flips is a grace period.
903 *
904 * If anyone is nuts enough to run this CONFIG_PREEMPT_RCU implementation
905 * on a large SMP, they might want to use a hierarchical organization of
906 * the per-CPU-counter pairs.
907 */
908static void rcu_try_flip(void)
909{
910 unsigned long flags;
911
912 RCU_TRACE_ME(rcupreempt_trace_try_flip_1);
913 if (unlikely(!spin_trylock_irqsave(&rcu_ctrlblk.fliplock, flags))) {
914 RCU_TRACE_ME(rcupreempt_trace_try_flip_e1);
915 return;
916 }
917
918 /*
919 * Take the next transition(s) through the RCU grace-period
920 * flip-counter state machine.
921 */
922
923 switch (rcu_ctrlblk.rcu_try_flip_state) {
924 case rcu_try_flip_idle_state:
925 if (rcu_try_flip_idle())
926 rcu_ctrlblk.rcu_try_flip_state =
927 rcu_try_flip_waitack_state;
928 break;
929 case rcu_try_flip_waitack_state:
930 if (rcu_try_flip_waitack())
931 rcu_ctrlblk.rcu_try_flip_state =
932 rcu_try_flip_waitzero_state;
933 break;
934 case rcu_try_flip_waitzero_state:
935 if (rcu_try_flip_waitzero())
936 rcu_ctrlblk.rcu_try_flip_state =
937 rcu_try_flip_waitmb_state;
938 break;
939 case rcu_try_flip_waitmb_state:
940 if (rcu_try_flip_waitmb())
941 rcu_ctrlblk.rcu_try_flip_state =
942 rcu_try_flip_idle_state;
943 }
944 spin_unlock_irqrestore(&rcu_ctrlblk.fliplock, flags);
945}
946
947/*
948 * Check to see if this CPU needs to do a memory barrier in order to
949 * ensure that any prior RCU read-side critical sections have committed
950 * their counter manipulations and critical-section memory references
951 * before declaring the grace period to be completed.
952 */
953static void rcu_check_mb(int cpu)
954{
955 if (per_cpu(rcu_mb_flag, cpu) == rcu_mb_needed) {
956 smp_mb(); /* Ensure RCU read-side accesses are visible. */
957 per_cpu(rcu_mb_flag, cpu) = rcu_mb_done;
958 }
959}
960
961void rcu_check_callbacks(int cpu, int user)
962{
963 unsigned long flags;
964 struct rcu_data *rdp = RCU_DATA_CPU(cpu);
965
966 /*
967 * If this CPU took its interrupt from user mode or from the
968 * idle loop, and this is not a nested interrupt, then
969 * this CPU has to have exited all prior preept-disable
970 * sections of code. So increment the counter to note this.
971 *
972 * The memory barrier is needed to handle the case where
973 * writes from a preempt-disable section of code get reordered
974 * into schedule() by this CPU's write buffer. So the memory
975 * barrier makes sure that the rcu_qsctr_inc() is seen by other
976 * CPUs to happen after any such write.
977 */
978
979 if (user ||
980 (idle_cpu(cpu) && !in_softirq() &&
981 hardirq_count() <= (1 << HARDIRQ_SHIFT))) {
982 smp_mb(); /* Guard against aggressive schedule(). */
983 rcu_qsctr_inc(cpu);
984 }
985
986 rcu_check_mb(cpu);
987 if (rcu_ctrlblk.completed == rdp->completed)
988 rcu_try_flip();
989 spin_lock_irqsave(&rdp->lock, flags);
990 RCU_TRACE_RDP(rcupreempt_trace_check_callbacks, rdp);
991 __rcu_advance_callbacks(rdp);
992 if (rdp->donelist == NULL) {
993 spin_unlock_irqrestore(&rdp->lock, flags);
994 } else {
995 spin_unlock_irqrestore(&rdp->lock, flags);
996 raise_softirq(RCU_SOFTIRQ);
997 }
998}
999
1000/*
1001 * Needed by dynticks, to make sure all RCU processing has finished
1002 * when we go idle:
1003 */
1004void rcu_advance_callbacks(int cpu, int user)
1005{
1006 unsigned long flags;
1007 struct rcu_data *rdp = RCU_DATA_CPU(cpu);
1008
1009 if (rcu_ctrlblk.completed == rdp->completed) {
1010 rcu_try_flip();
1011 if (rcu_ctrlblk.completed == rdp->completed)
1012 return;
1013 }
1014 spin_lock_irqsave(&rdp->lock, flags);
1015 RCU_TRACE_RDP(rcupreempt_trace_check_callbacks, rdp);
1016 __rcu_advance_callbacks(rdp);
1017 spin_unlock_irqrestore(&rdp->lock, flags);
1018}
1019
1020#ifdef CONFIG_HOTPLUG_CPU
1021#define rcu_offline_cpu_enqueue(srclist, srctail, dstlist, dsttail) do { \
1022 *dsttail = srclist; \
1023 if (srclist != NULL) { \
1024 dsttail = srctail; \
1025 srclist = NULL; \
1026 srctail = &srclist;\
1027 } \
1028 } while (0)
1029
1030void rcu_offline_cpu(int cpu)
1031{
1032 int i;
1033 struct rcu_head *list = NULL;
1034 unsigned long flags;
1035 struct rcu_data *rdp = RCU_DATA_CPU(cpu);
1036 struct rcu_head *schedlist = NULL;
1037 struct rcu_head **schedtail = &schedlist;
1038 struct rcu_head **tail = &list;
1039
1040 /*
1041 * Remove all callbacks from the newly dead CPU, retaining order.
1042 * Otherwise rcu_barrier() will fail
1043 */
1044
1045 spin_lock_irqsave(&rdp->lock, flags);
1046 rcu_offline_cpu_enqueue(rdp->donelist, rdp->donetail, list, tail);
1047 for (i = GP_STAGES - 1; i >= 0; i--)
1048 rcu_offline_cpu_enqueue(rdp->waitlist[i], rdp->waittail[i],
1049 list, tail);
1050 rcu_offline_cpu_enqueue(rdp->nextlist, rdp->nexttail, list, tail);
1051 rcu_offline_cpu_enqueue(rdp->waitschedlist, rdp->waitschedtail,
1052 schedlist, schedtail);
1053 rcu_offline_cpu_enqueue(rdp->nextschedlist, rdp->nextschedtail,
1054 schedlist, schedtail);
1055 rdp->rcu_sched_sleeping = 0;
1056 spin_unlock_irqrestore(&rdp->lock, flags);
1057 rdp->waitlistcount = 0;
1058
1059 /* Disengage the newly dead CPU from the grace-period computation. */
1060
1061 spin_lock_irqsave(&rcu_ctrlblk.fliplock, flags);
1062 rcu_check_mb(cpu);
1063 if (per_cpu(rcu_flip_flag, cpu) == rcu_flipped) {
1064 smp_mb(); /* Subsequent counter accesses must see new value */
1065 per_cpu(rcu_flip_flag, cpu) = rcu_flip_seen;
1066 smp_mb(); /* Subsequent RCU read-side critical sections */
1067 /* seen -after- acknowledgement. */
1068 }
1069
1070 RCU_DATA_ME()->rcu_flipctr[0] += RCU_DATA_CPU(cpu)->rcu_flipctr[0];
1071 RCU_DATA_ME()->rcu_flipctr[1] += RCU_DATA_CPU(cpu)->rcu_flipctr[1];
1072
1073 RCU_DATA_CPU(cpu)->rcu_flipctr[0] = 0;
1074 RCU_DATA_CPU(cpu)->rcu_flipctr[1] = 0;
1075
1076 cpumask_clear_cpu(cpu, to_cpumask(rcu_cpu_online_map));
1077
1078 spin_unlock_irqrestore(&rcu_ctrlblk.fliplock, flags);
1079
1080 /*
1081 * Place the removed callbacks on the current CPU's queue.
1082 * Make them all start a new grace period: simple approach,
1083 * in theory could starve a given set of callbacks, but
1084 * you would need to be doing some serious CPU hotplugging
1085 * to make this happen. If this becomes a problem, adding
1086 * a synchronize_rcu() to the hotplug path would be a simple
1087 * fix.
1088 */
1089
1090 local_irq_save(flags); /* disable preempt till we know what lock. */
1091 rdp = RCU_DATA_ME();
1092 spin_lock(&rdp->lock);
1093 *rdp->nexttail = list;
1094 if (list)
1095 rdp->nexttail = tail;
1096 *rdp->nextschedtail = schedlist;
1097 if (schedlist)
1098 rdp->nextschedtail = schedtail;
1099 spin_unlock_irqrestore(&rdp->lock, flags);
1100}
1101
1102#else /* #ifdef CONFIG_HOTPLUG_CPU */
1103
1104void rcu_offline_cpu(int cpu)
1105{
1106}
1107
1108#endif /* #else #ifdef CONFIG_HOTPLUG_CPU */
1109
1110void __cpuinit rcu_online_cpu(int cpu)
1111{
1112 unsigned long flags;
1113 struct rcu_data *rdp;
1114
1115 spin_lock_irqsave(&rcu_ctrlblk.fliplock, flags);
1116 cpumask_set_cpu(cpu, to_cpumask(rcu_cpu_online_map));
1117 spin_unlock_irqrestore(&rcu_ctrlblk.fliplock, flags);
1118
1119 /*
1120 * The rcu_sched grace-period processing might have bypassed
1121 * this CPU, given that it was not in the rcu_cpu_online_map
1122 * when the grace-period scan started. This means that the
1123 * grace-period task might sleep. So make sure that if this
1124 * should happen, the first callback posted to this CPU will
1125 * wake up the grace-period task if need be.
1126 */
1127
1128 rdp = RCU_DATA_CPU(cpu);
1129 spin_lock_irqsave(&rdp->lock, flags);
1130 rdp->rcu_sched_sleeping = 1;
1131 spin_unlock_irqrestore(&rdp->lock, flags);
1132}
1133
1134static void rcu_process_callbacks(struct softirq_action *unused)
1135{
1136 unsigned long flags;
1137 struct rcu_head *next, *list;
1138 struct rcu_data *rdp;
1139
1140 local_irq_save(flags);
1141 rdp = RCU_DATA_ME();
1142 spin_lock(&rdp->lock);
1143 list = rdp->donelist;
1144 if (list == NULL) {
1145 spin_unlock_irqrestore(&rdp->lock, flags);
1146 return;
1147 }
1148 rdp->donelist = NULL;
1149 rdp->donetail = &rdp->donelist;
1150 RCU_TRACE_RDP(rcupreempt_trace_done_remove, rdp);
1151 spin_unlock_irqrestore(&rdp->lock, flags);
1152 while (list) {
1153 next = list->next;
1154 list->func(list);
1155 list = next;
1156 RCU_TRACE_ME(rcupreempt_trace_invoke);
1157 }
1158}
1159
1160void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
1161{
1162 unsigned long flags;
1163 struct rcu_data *rdp;
1164
1165 head->func = func;
1166 head->next = NULL;
1167 local_irq_save(flags);
1168 rdp = RCU_DATA_ME();
1169 spin_lock(&rdp->lock);
1170 __rcu_advance_callbacks(rdp);
1171 *rdp->nexttail = head;
1172 rdp->nexttail = &head->next;
1173 RCU_TRACE_RDP(rcupreempt_trace_next_add, rdp);
1174 spin_unlock_irqrestore(&rdp->lock, flags);
1175}
1176EXPORT_SYMBOL_GPL(call_rcu);
1177
1178void call_rcu_sched(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
1179{
1180 unsigned long flags;
1181 struct rcu_data *rdp;
1182 int wake_gp = 0;
1183
1184 head->func = func;
1185 head->next = NULL;
1186 local_irq_save(flags);
1187 rdp = RCU_DATA_ME();
1188 spin_lock(&rdp->lock);
1189 *rdp->nextschedtail = head;
1190 rdp->nextschedtail = &head->next;
1191 if (rdp->rcu_sched_sleeping) {
1192
1193 /* Grace-period processing might be sleeping... */
1194
1195 rdp->rcu_sched_sleeping = 0;
1196 wake_gp = 1;
1197 }
1198 spin_unlock_irqrestore(&rdp->lock, flags);
1199 if (wake_gp) {
1200
1201 /* Wake up grace-period processing, unless someone beat us. */
1202
1203 spin_lock_irqsave(&rcu_ctrlblk.schedlock, flags);
1204 if (rcu_ctrlblk.sched_sleep != rcu_sched_sleeping)
1205 wake_gp = 0;
1206 rcu_ctrlblk.sched_sleep = rcu_sched_not_sleeping;
1207 spin_unlock_irqrestore(&rcu_ctrlblk.schedlock, flags);
1208 if (wake_gp)
1209 wake_up_interruptible(&rcu_ctrlblk.sched_wq);
1210 }
1211}
1212EXPORT_SYMBOL_GPL(call_rcu_sched);
1213
1214/*
1215 * Wait until all currently running preempt_disable() code segments
1216 * (including hardware-irq-disable segments) complete. Note that
1217 * in -rt this does -not- necessarily result in all currently executing
1218 * interrupt -handlers- having completed.
1219 */
1220void __synchronize_sched(void)
1221{
1222 struct rcu_synchronize rcu;
1223
1224 if (num_online_cpus() == 1)
1225 return; /* blocking is gp if only one CPU! */
1226
1227 init_completion(&rcu.completion);
1228 /* Will wake me after RCU finished. */
1229 call_rcu_sched(&rcu.head, wakeme_after_rcu);
1230 /* Wait for it. */
1231 wait_for_completion(&rcu.completion);
1232}
1233EXPORT_SYMBOL_GPL(__synchronize_sched);
1234
1235/*
1236 * kthread function that manages call_rcu_sched grace periods.
1237 */
1238static int rcu_sched_grace_period(void *arg)
1239{
1240 int couldsleep; /* might sleep after current pass. */
1241 int couldsleepnext = 0; /* might sleep after next pass. */
1242 int cpu;
1243 unsigned long flags;
1244 struct rcu_data *rdp;
1245 int ret;
1246
1247 /*
1248 * Each pass through the following loop handles one
1249 * rcu_sched grace period cycle.
1250 */
1251 do {
1252 /* Save each CPU's current state. */
1253
1254 for_each_online_cpu(cpu) {
1255 dyntick_save_progress_counter_sched(cpu);
1256 save_qsctr_sched(cpu);
1257 }
1258
1259 /*
1260 * Sleep for about an RCU grace-period's worth to
1261 * allow better batching and to consume less CPU.
1262 */
1263 schedule_timeout_interruptible(RCU_SCHED_BATCH_TIME);
1264
1265 /*
1266 * If there was nothing to do last time, prepare to
1267 * sleep at the end of the current grace period cycle.
1268 */
1269 couldsleep = couldsleepnext;
1270 couldsleepnext = 1;
1271 if (couldsleep) {
1272 spin_lock_irqsave(&rcu_ctrlblk.schedlock, flags);
1273 rcu_ctrlblk.sched_sleep = rcu_sched_sleep_prep;
1274 spin_unlock_irqrestore(&rcu_ctrlblk.schedlock, flags);
1275 }
1276
1277 /*
1278 * Wait on each CPU in turn to have either visited
1279 * a quiescent state or been in dynticks-idle mode.
1280 */
1281 for_each_online_cpu(cpu) {
1282 while (rcu_qsctr_inc_needed(cpu) &&
1283 rcu_qsctr_inc_needed_dyntick(cpu)) {
1284 /* resched_cpu(cpu); @@@ */
1285 schedule_timeout_interruptible(1);
1286 }
1287 }
1288
1289 /* Advance callbacks for each CPU. */
1290
1291 for_each_online_cpu(cpu) {
1292
1293 rdp = RCU_DATA_CPU(cpu);
1294 spin_lock_irqsave(&rdp->lock, flags);
1295
1296 /*
1297 * We are running on this CPU irq-disabled, so no
1298 * CPU can go offline until we re-enable irqs.
1299 * The current CPU might have already gone
1300 * offline (between the for_each_offline_cpu and
1301 * the spin_lock_irqsave), but in that case all its
1302 * callback lists will be empty, so no harm done.
1303 *
1304 * Advance the callbacks! We share normal RCU's
1305 * donelist, since callbacks are invoked the
1306 * same way in either case.
1307 */
1308 if (rdp->waitschedlist != NULL) {
1309 *rdp->donetail = rdp->waitschedlist;
1310 rdp->donetail = rdp->waitschedtail;
1311
1312 /*
1313 * Next rcu_check_callbacks() will
1314 * do the required raise_softirq().
1315 */
1316 }
1317 if (rdp->nextschedlist != NULL) {
1318 rdp->waitschedlist = rdp->nextschedlist;
1319 rdp->waitschedtail = rdp->nextschedtail;
1320 couldsleep = 0;
1321 couldsleepnext = 0;
1322 } else {
1323 rdp->waitschedlist = NULL;
1324 rdp->waitschedtail = &rdp->waitschedlist;
1325 }
1326 rdp->nextschedlist = NULL;
1327 rdp->nextschedtail = &rdp->nextschedlist;
1328
1329 /* Mark sleep intention. */
1330
1331 rdp->rcu_sched_sleeping = couldsleep;
1332
1333 spin_unlock_irqrestore(&rdp->lock, flags);
1334 }
1335
1336 /* If we saw callbacks on the last scan, go deal with them. */
1337
1338 if (!couldsleep)
1339 continue;
1340
1341 /* Attempt to block... */
1342
1343 spin_lock_irqsave(&rcu_ctrlblk.schedlock, flags);
1344 if (rcu_ctrlblk.sched_sleep != rcu_sched_sleep_prep) {
1345
1346 /*
1347 * Someone posted a callback after we scanned.
1348 * Go take care of it.
1349 */
1350 spin_unlock_irqrestore(&rcu_ctrlblk.schedlock, flags);
1351 couldsleepnext = 0;
1352 continue;
1353 }
1354
1355 /* Block until the next person posts a callback. */
1356
1357 rcu_ctrlblk.sched_sleep = rcu_sched_sleeping;
1358 spin_unlock_irqrestore(&rcu_ctrlblk.schedlock, flags);
1359 ret = 0; /* unused */
1360 __wait_event_interruptible(rcu_ctrlblk.sched_wq,
1361 rcu_ctrlblk.sched_sleep != rcu_sched_sleeping,
1362 ret);
1363
1364 couldsleepnext = 0;
1365
1366 } while (!kthread_should_stop());
1367
1368 return (0);
1369}
1370
1371/*
1372 * Check to see if any future RCU-related work will need to be done
1373 * by the current CPU, even if none need be done immediately, returning
1374 * 1 if so. Assumes that notifiers would take care of handling any
1375 * outstanding requests from the RCU core.
1376 *
1377 * This function is part of the RCU implementation; it is -not-
1378 * an exported member of the RCU API.
1379 */
1380int rcu_needs_cpu(int cpu)
1381{
1382 struct rcu_data *rdp = RCU_DATA_CPU(cpu);
1383
1384 return (rdp->donelist != NULL ||
1385 !!rdp->waitlistcount ||
1386 rdp->nextlist != NULL ||
1387 rdp->nextschedlist != NULL ||
1388 rdp->waitschedlist != NULL);
1389}
1390
1391int rcu_pending(int cpu)
1392{
1393 struct rcu_data *rdp = RCU_DATA_CPU(cpu);
1394
1395 /* The CPU has at least one callback queued somewhere. */
1396
1397 if (rdp->donelist != NULL ||
1398 !!rdp->waitlistcount ||
1399 rdp->nextlist != NULL ||
1400 rdp->nextschedlist != NULL ||
1401 rdp->waitschedlist != NULL)
1402 return 1;
1403
1404 /* The RCU core needs an acknowledgement from this CPU. */
1405
1406 if ((per_cpu(rcu_flip_flag, cpu) == rcu_flipped) ||
1407 (per_cpu(rcu_mb_flag, cpu) == rcu_mb_needed))
1408 return 1;
1409
1410 /* This CPU has fallen behind the global grace-period number. */
1411
1412 if (rdp->completed != rcu_ctrlblk.completed)
1413 return 1;
1414
1415 /* Nothing needed from this CPU. */
1416
1417 return 0;
1418}
1419
1420static int __cpuinit rcu_cpu_notify(struct notifier_block *self,
1421 unsigned long action, void *hcpu)
1422{
1423 long cpu = (long)hcpu;
1424
1425 switch (action) {
1426 case CPU_UP_PREPARE:
1427 case CPU_UP_PREPARE_FROZEN:
1428 rcu_online_cpu(cpu);
1429 break;
1430 case CPU_UP_CANCELED:
1431 case CPU_UP_CANCELED_FROZEN:
1432 case CPU_DEAD:
1433 case CPU_DEAD_FROZEN:
1434 rcu_offline_cpu(cpu);
1435 break;
1436 default:
1437 break;
1438 }
1439 return NOTIFY_OK;
1440}
1441
1442static struct notifier_block __cpuinitdata rcu_nb = {
1443 .notifier_call = rcu_cpu_notify,
1444};
1445
1446void __init __rcu_init(void)
1447{
1448 int cpu;
1449 int i;
1450 struct rcu_data *rdp;
1451
1452 printk(KERN_NOTICE "Preemptible RCU implementation.\n");
1453 for_each_possible_cpu(cpu) {
1454 rdp = RCU_DATA_CPU(cpu);
1455 spin_lock_init(&rdp->lock);
1456 rdp->completed = 0;
1457 rdp->waitlistcount = 0;
1458 rdp->nextlist = NULL;
1459 rdp->nexttail = &rdp->nextlist;
1460 for (i = 0; i < GP_STAGES; i++) {
1461 rdp->waitlist[i] = NULL;
1462 rdp->waittail[i] = &rdp->waitlist[i];
1463 }
1464 rdp->donelist = NULL;
1465 rdp->donetail = &rdp->donelist;
1466 rdp->rcu_flipctr[0] = 0;
1467 rdp->rcu_flipctr[1] = 0;
1468 rdp->nextschedlist = NULL;
1469 rdp->nextschedtail = &rdp->nextschedlist;
1470 rdp->waitschedlist = NULL;
1471 rdp->waitschedtail = &rdp->waitschedlist;
1472 rdp->rcu_sched_sleeping = 0;
1473 }
1474 register_cpu_notifier(&rcu_nb);
1475
1476 /*
1477 * We don't need protection against CPU-Hotplug here
1478 * since
1479 * a) If a CPU comes online while we are iterating over the
1480 * cpu_online_mask below, we would only end up making a
1481 * duplicate call to rcu_online_cpu() which sets the corresponding
1482 * CPU's mask in the rcu_cpu_online_map.
1483 *
1484 * b) A CPU cannot go offline at this point in time since the user
1485 * does not have access to the sysfs interface, nor do we
1486 * suspend the system.
1487 */
1488 for_each_online_cpu(cpu)
1489 rcu_cpu_notify(&rcu_nb, CPU_UP_PREPARE, (void *)(long) cpu);
1490
1491 open_softirq(RCU_SOFTIRQ, rcu_process_callbacks);
1492}
1493
1494/*
1495 * Late-boot-time RCU initialization that must wait until after scheduler
1496 * has been initialized.
1497 */
1498void __init rcu_init_sched(void)
1499{
1500 rcu_sched_grace_period_task = kthread_run(rcu_sched_grace_period,
1501 NULL,
1502 "rcu_sched_grace_period");
1503 WARN_ON(IS_ERR(rcu_sched_grace_period_task));
1504}
1505
1506#ifdef CONFIG_RCU_TRACE
1507long *rcupreempt_flipctr(int cpu)
1508{
1509 return &RCU_DATA_CPU(cpu)->rcu_flipctr[0];
1510}
1511EXPORT_SYMBOL_GPL(rcupreempt_flipctr);
1512
1513int rcupreempt_flip_flag(int cpu)
1514{
1515 return per_cpu(rcu_flip_flag, cpu);
1516}
1517EXPORT_SYMBOL_GPL(rcupreempt_flip_flag);
1518
1519int rcupreempt_mb_flag(int cpu)
1520{
1521 return per_cpu(rcu_mb_flag, cpu);
1522}
1523EXPORT_SYMBOL_GPL(rcupreempt_mb_flag);
1524
1525char *rcupreempt_try_flip_state_name(void)
1526{
1527 return rcu_try_flip_state_names[rcu_ctrlblk.rcu_try_flip_state];
1528}
1529EXPORT_SYMBOL_GPL(rcupreempt_try_flip_state_name);
1530
1531struct rcupreempt_trace *rcupreempt_trace_cpu(int cpu)
1532{
1533 struct rcu_data *rdp = RCU_DATA_CPU(cpu);
1534
1535 return &rdp->trace;
1536}
1537EXPORT_SYMBOL_GPL(rcupreempt_trace_cpu);
1538
1539#endif /* #ifdef RCU_TRACE */
diff --git a/kernel/rcupreempt_trace.c b/kernel/rcupreempt_trace.c
deleted file mode 100644
index 7c2665cac172..000000000000
--- a/kernel/rcupreempt_trace.c
+++ /dev/null
@@ -1,334 +0,0 @@
1/*
2 * Read-Copy Update tracing for realtime implementation
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, 2006
19 *
20 * Papers: http://www.rdrop.com/users/paulmck/RCU
21 *
22 * For detailed explanation of Read-Copy Update mechanism see -
23 * Documentation/RCU/ *.txt
24 *
25 */
26#include <linux/types.h>
27#include <linux/kernel.h>
28#include <linux/init.h>
29#include <linux/spinlock.h>
30#include <linux/smp.h>
31#include <linux/rcupdate.h>
32#include <linux/interrupt.h>
33#include <linux/sched.h>
34#include <asm/atomic.h>
35#include <linux/bitops.h>
36#include <linux/module.h>
37#include <linux/completion.h>
38#include <linux/moduleparam.h>
39#include <linux/percpu.h>
40#include <linux/notifier.h>
41#include <linux/cpu.h>
42#include <linux/mutex.h>
43#include <linux/rcupreempt_trace.h>
44#include <linux/debugfs.h>
45
46static struct mutex rcupreempt_trace_mutex;
47static char *rcupreempt_trace_buf;
48#define RCUPREEMPT_TRACE_BUF_SIZE 4096
49
50void rcupreempt_trace_move2done(struct rcupreempt_trace *trace)
51{
52 trace->done_length += trace->wait_length;
53 trace->done_add += trace->wait_length;
54 trace->wait_length = 0;
55}
56void rcupreempt_trace_move2wait(struct rcupreempt_trace *trace)
57{
58 trace->wait_length += trace->next_length;
59 trace->wait_add += trace->next_length;
60 trace->next_length = 0;
61}
62void rcupreempt_trace_try_flip_1(struct rcupreempt_trace *trace)
63{
64 atomic_inc(&trace->rcu_try_flip_1);
65}
66void rcupreempt_trace_try_flip_e1(struct rcupreempt_trace *trace)
67{
68 atomic_inc(&trace->rcu_try_flip_e1);
69}
70void rcupreempt_trace_try_flip_i1(struct rcupreempt_trace *trace)
71{
72 trace->rcu_try_flip_i1++;
73}
74void rcupreempt_trace_try_flip_ie1(struct rcupreempt_trace *trace)
75{
76 trace->rcu_try_flip_ie1++;
77}
78void rcupreempt_trace_try_flip_g1(struct rcupreempt_trace *trace)
79{
80 trace->rcu_try_flip_g1++;
81}
82void rcupreempt_trace_try_flip_a1(struct rcupreempt_trace *trace)
83{
84 trace->rcu_try_flip_a1++;
85}
86void rcupreempt_trace_try_flip_ae1(struct rcupreempt_trace *trace)
87{
88 trace->rcu_try_flip_ae1++;
89}
90void rcupreempt_trace_try_flip_a2(struct rcupreempt_trace *trace)
91{
92 trace->rcu_try_flip_a2++;
93}
94void rcupreempt_trace_try_flip_z1(struct rcupreempt_trace *trace)
95{
96 trace->rcu_try_flip_z1++;
97}
98void rcupreempt_trace_try_flip_ze1(struct rcupreempt_trace *trace)
99{
100 trace->rcu_try_flip_ze1++;
101}
102void rcupreempt_trace_try_flip_z2(struct rcupreempt_trace *trace)
103{
104 trace->rcu_try_flip_z2++;
105}
106void rcupreempt_trace_try_flip_m1(struct rcupreempt_trace *trace)
107{
108 trace->rcu_try_flip_m1++;
109}
110void rcupreempt_trace_try_flip_me1(struct rcupreempt_trace *trace)
111{
112 trace->rcu_try_flip_me1++;
113}
114void rcupreempt_trace_try_flip_m2(struct rcupreempt_trace *trace)
115{
116 trace->rcu_try_flip_m2++;
117}
118void rcupreempt_trace_check_callbacks(struct rcupreempt_trace *trace)
119{
120 trace->rcu_check_callbacks++;
121}
122void rcupreempt_trace_done_remove(struct rcupreempt_trace *trace)
123{
124 trace->done_remove += trace->done_length;
125 trace->done_length = 0;
126}
127void rcupreempt_trace_invoke(struct rcupreempt_trace *trace)
128{
129 atomic_inc(&trace->done_invoked);
130}
131void rcupreempt_trace_next_add(struct rcupreempt_trace *trace)
132{
133 trace->next_add++;
134 trace->next_length++;
135}
136
137static void rcupreempt_trace_sum(struct rcupreempt_trace *sp)
138{
139 struct rcupreempt_trace *cp;
140 int cpu;
141
142 memset(sp, 0, sizeof(*sp));
143 for_each_possible_cpu(cpu) {
144 cp = rcupreempt_trace_cpu(cpu);
145 sp->next_length += cp->next_length;
146 sp->next_add += cp->next_add;
147 sp->wait_length += cp->wait_length;
148 sp->wait_add += cp->wait_add;
149 sp->done_length += cp->done_length;
150 sp->done_add += cp->done_add;
151 sp->done_remove += cp->done_remove;
152 atomic_add(atomic_read(&cp->done_invoked), &sp->done_invoked);
153 sp->rcu_check_callbacks += cp->rcu_check_callbacks;
154 atomic_add(atomic_read(&cp->rcu_try_flip_1),
155 &sp->rcu_try_flip_1);
156 atomic_add(atomic_read(&cp->rcu_try_flip_e1),
157 &sp->rcu_try_flip_e1);
158 sp->rcu_try_flip_i1 += cp->rcu_try_flip_i1;
159 sp->rcu_try_flip_ie1 += cp->rcu_try_flip_ie1;
160 sp->rcu_try_flip_g1 += cp->rcu_try_flip_g1;
161 sp->rcu_try_flip_a1 += cp->rcu_try_flip_a1;
162 sp->rcu_try_flip_ae1 += cp->rcu_try_flip_ae1;
163 sp->rcu_try_flip_a2 += cp->rcu_try_flip_a2;
164 sp->rcu_try_flip_z1 += cp->rcu_try_flip_z1;
165 sp->rcu_try_flip_ze1 += cp->rcu_try_flip_ze1;
166 sp->rcu_try_flip_z2 += cp->rcu_try_flip_z2;
167 sp->rcu_try_flip_m1 += cp->rcu_try_flip_m1;
168 sp->rcu_try_flip_me1 += cp->rcu_try_flip_me1;
169 sp->rcu_try_flip_m2 += cp->rcu_try_flip_m2;
170 }
171}
172
173static ssize_t rcustats_read(struct file *filp, char __user *buffer,
174 size_t count, loff_t *ppos)
175{
176 struct rcupreempt_trace trace;
177 ssize_t bcount;
178 int cnt = 0;
179
180 rcupreempt_trace_sum(&trace);
181 mutex_lock(&rcupreempt_trace_mutex);
182 snprintf(&rcupreempt_trace_buf[cnt], RCUPREEMPT_TRACE_BUF_SIZE - cnt,
183 "ggp=%ld rcc=%ld\n",
184 rcu_batches_completed(),
185 trace.rcu_check_callbacks);
186 snprintf(&rcupreempt_trace_buf[cnt], RCUPREEMPT_TRACE_BUF_SIZE - cnt,
187 "na=%ld nl=%ld wa=%ld wl=%ld da=%ld dl=%ld dr=%ld di=%d\n"
188 "1=%d e1=%d i1=%ld ie1=%ld g1=%ld a1=%ld ae1=%ld a2=%ld\n"
189 "z1=%ld ze1=%ld z2=%ld m1=%ld me1=%ld m2=%ld\n",
190
191 trace.next_add, trace.next_length,
192 trace.wait_add, trace.wait_length,
193 trace.done_add, trace.done_length,
194 trace.done_remove, atomic_read(&trace.done_invoked),
195 atomic_read(&trace.rcu_try_flip_1),
196 atomic_read(&trace.rcu_try_flip_e1),
197 trace.rcu_try_flip_i1, trace.rcu_try_flip_ie1,
198 trace.rcu_try_flip_g1,
199 trace.rcu_try_flip_a1, trace.rcu_try_flip_ae1,
200 trace.rcu_try_flip_a2,
201 trace.rcu_try_flip_z1, trace.rcu_try_flip_ze1,
202 trace.rcu_try_flip_z2,
203 trace.rcu_try_flip_m1, trace.rcu_try_flip_me1,
204 trace.rcu_try_flip_m2);
205 bcount = simple_read_from_buffer(buffer, count, ppos,
206 rcupreempt_trace_buf, strlen(rcupreempt_trace_buf));
207 mutex_unlock(&rcupreempt_trace_mutex);
208 return bcount;
209}
210
211static ssize_t rcugp_read(struct file *filp, char __user *buffer,
212 size_t count, loff_t *ppos)
213{
214 long oldgp = rcu_batches_completed();
215 ssize_t bcount;
216
217 mutex_lock(&rcupreempt_trace_mutex);
218 synchronize_rcu();
219 snprintf(rcupreempt_trace_buf, RCUPREEMPT_TRACE_BUF_SIZE,
220 "oldggp=%ld newggp=%ld\n", oldgp, rcu_batches_completed());
221 bcount = simple_read_from_buffer(buffer, count, ppos,
222 rcupreempt_trace_buf, strlen(rcupreempt_trace_buf));
223 mutex_unlock(&rcupreempt_trace_mutex);
224 return bcount;
225}
226
227static ssize_t rcuctrs_read(struct file *filp, char __user *buffer,
228 size_t count, loff_t *ppos)
229{
230 int cnt = 0;
231 int cpu;
232 int f = rcu_batches_completed() & 0x1;
233 ssize_t bcount;
234
235 mutex_lock(&rcupreempt_trace_mutex);
236
237 cnt += snprintf(&rcupreempt_trace_buf[cnt], RCUPREEMPT_TRACE_BUF_SIZE,
238 "CPU last cur F M\n");
239 for_each_online_cpu(cpu) {
240 long *flipctr = rcupreempt_flipctr(cpu);
241 cnt += snprintf(&rcupreempt_trace_buf[cnt],
242 RCUPREEMPT_TRACE_BUF_SIZE - cnt,
243 "%3d %4ld %3ld %d %d\n",
244 cpu,
245 flipctr[!f],
246 flipctr[f],
247 rcupreempt_flip_flag(cpu),
248 rcupreempt_mb_flag(cpu));
249 }
250 cnt += snprintf(&rcupreempt_trace_buf[cnt],
251 RCUPREEMPT_TRACE_BUF_SIZE - cnt,
252 "ggp = %ld, state = %s\n",
253 rcu_batches_completed(),
254 rcupreempt_try_flip_state_name());
255 cnt += snprintf(&rcupreempt_trace_buf[cnt],
256 RCUPREEMPT_TRACE_BUF_SIZE - cnt,
257 "\n");
258 bcount = simple_read_from_buffer(buffer, count, ppos,
259 rcupreempt_trace_buf, strlen(rcupreempt_trace_buf));
260 mutex_unlock(&rcupreempt_trace_mutex);
261 return bcount;
262}
263
264static struct file_operations rcustats_fops = {
265 .owner = THIS_MODULE,
266 .read = rcustats_read,
267};
268
269static struct file_operations rcugp_fops = {
270 .owner = THIS_MODULE,
271 .read = rcugp_read,
272};
273
274static struct file_operations rcuctrs_fops = {
275 .owner = THIS_MODULE,
276 .read = rcuctrs_read,
277};
278
279static struct dentry *rcudir, *statdir, *ctrsdir, *gpdir;
280static int rcupreempt_debugfs_init(void)
281{
282 rcudir = debugfs_create_dir("rcu", NULL);
283 if (!rcudir)
284 goto out;
285 statdir = debugfs_create_file("rcustats", 0444, rcudir,
286 NULL, &rcustats_fops);
287 if (!statdir)
288 goto free_out;
289
290 gpdir = debugfs_create_file("rcugp", 0444, rcudir, NULL, &rcugp_fops);
291 if (!gpdir)
292 goto free_out;
293
294 ctrsdir = debugfs_create_file("rcuctrs", 0444, rcudir,
295 NULL, &rcuctrs_fops);
296 if (!ctrsdir)
297 goto free_out;
298 return 0;
299free_out:
300 if (statdir)
301 debugfs_remove(statdir);
302 if (gpdir)
303 debugfs_remove(gpdir);
304 debugfs_remove(rcudir);
305out:
306 return 1;
307}
308
309static int __init rcupreempt_trace_init(void)
310{
311 int ret;
312
313 mutex_init(&rcupreempt_trace_mutex);
314 rcupreempt_trace_buf = kmalloc(RCUPREEMPT_TRACE_BUF_SIZE, GFP_KERNEL);
315 if (!rcupreempt_trace_buf)
316 return 1;
317 ret = rcupreempt_debugfs_init();
318 if (ret)
319 kfree(rcupreempt_trace_buf);
320 return ret;
321}
322
323static void __exit rcupreempt_trace_cleanup(void)
324{
325 debugfs_remove(statdir);
326 debugfs_remove(gpdir);
327 debugfs_remove(ctrsdir);
328 debugfs_remove(rcudir);
329 kfree(rcupreempt_trace_buf);
330}
331
332
333module_init(rcupreempt_trace_init);
334module_exit(rcupreempt_trace_cleanup);
diff --git a/kernel/rcutorture.c b/kernel/rcutorture.c
index 9b4a975a4b4a..b33db539a8ad 100644
--- a/kernel/rcutorture.c
+++ b/kernel/rcutorture.c
@@ -257,14 +257,14 @@ struct rcu_torture_ops {
257 void (*init)(void); 257 void (*init)(void);
258 void (*cleanup)(void); 258 void (*cleanup)(void);
259 int (*readlock)(void); 259 int (*readlock)(void);
260 void (*readdelay)(struct rcu_random_state *rrsp); 260 void (*read_delay)(struct rcu_random_state *rrsp);
261 void (*readunlock)(int idx); 261 void (*readunlock)(int idx);
262 int (*completed)(void); 262 int (*completed)(void);
263 void (*deferredfree)(struct rcu_torture *p); 263 void (*deferred_free)(struct rcu_torture *p);
264 void (*sync)(void); 264 void (*sync)(void);
265 void (*cb_barrier)(void); 265 void (*cb_barrier)(void);
266 int (*stats)(char *page); 266 int (*stats)(char *page);
267 int irqcapable; 267 int irq_capable;
268 char *name; 268 char *name;
269}; 269};
270static struct rcu_torture_ops *cur_ops = NULL; 270static struct rcu_torture_ops *cur_ops = NULL;
@@ -320,7 +320,7 @@ rcu_torture_cb(struct rcu_head *p)
320 rp->rtort_mbtest = 0; 320 rp->rtort_mbtest = 0;
321 rcu_torture_free(rp); 321 rcu_torture_free(rp);
322 } else 322 } else
323 cur_ops->deferredfree(rp); 323 cur_ops->deferred_free(rp);
324} 324}
325 325
326static void rcu_torture_deferred_free(struct rcu_torture *p) 326static void rcu_torture_deferred_free(struct rcu_torture *p)
@@ -329,18 +329,18 @@ static void rcu_torture_deferred_free(struct rcu_torture *p)
329} 329}
330 330
331static struct rcu_torture_ops rcu_ops = { 331static struct rcu_torture_ops rcu_ops = {
332 .init = NULL, 332 .init = NULL,
333 .cleanup = NULL, 333 .cleanup = NULL,
334 .readlock = rcu_torture_read_lock, 334 .readlock = rcu_torture_read_lock,
335 .readdelay = rcu_read_delay, 335 .read_delay = rcu_read_delay,
336 .readunlock = rcu_torture_read_unlock, 336 .readunlock = rcu_torture_read_unlock,
337 .completed = rcu_torture_completed, 337 .completed = rcu_torture_completed,
338 .deferredfree = rcu_torture_deferred_free, 338 .deferred_free = rcu_torture_deferred_free,
339 .sync = synchronize_rcu, 339 .sync = synchronize_rcu,
340 .cb_barrier = rcu_barrier, 340 .cb_barrier = rcu_barrier,
341 .stats = NULL, 341 .stats = NULL,
342 .irqcapable = 1, 342 .irq_capable = 1,
343 .name = "rcu" 343 .name = "rcu"
344}; 344};
345 345
346static void rcu_sync_torture_deferred_free(struct rcu_torture *p) 346static void rcu_sync_torture_deferred_free(struct rcu_torture *p)
@@ -370,18 +370,18 @@ static void rcu_sync_torture_init(void)
370} 370}
371 371
372static struct rcu_torture_ops rcu_sync_ops = { 372static struct rcu_torture_ops rcu_sync_ops = {
373 .init = rcu_sync_torture_init, 373 .init = rcu_sync_torture_init,
374 .cleanup = NULL, 374 .cleanup = NULL,
375 .readlock = rcu_torture_read_lock, 375 .readlock = rcu_torture_read_lock,
376 .readdelay = rcu_read_delay, 376 .read_delay = rcu_read_delay,
377 .readunlock = rcu_torture_read_unlock, 377 .readunlock = rcu_torture_read_unlock,
378 .completed = rcu_torture_completed, 378 .completed = rcu_torture_completed,
379 .deferredfree = rcu_sync_torture_deferred_free, 379 .deferred_free = rcu_sync_torture_deferred_free,
380 .sync = synchronize_rcu, 380 .sync = synchronize_rcu,
381 .cb_barrier = NULL, 381 .cb_barrier = NULL,
382 .stats = NULL, 382 .stats = NULL,
383 .irqcapable = 1, 383 .irq_capable = 1,
384 .name = "rcu_sync" 384 .name = "rcu_sync"
385}; 385};
386 386
387/* 387/*
@@ -432,33 +432,33 @@ static void rcu_bh_torture_synchronize(void)
432} 432}
433 433
434static struct rcu_torture_ops rcu_bh_ops = { 434static struct rcu_torture_ops rcu_bh_ops = {
435 .init = NULL, 435 .init = NULL,
436 .cleanup = NULL, 436 .cleanup = NULL,
437 .readlock = rcu_bh_torture_read_lock, 437 .readlock = rcu_bh_torture_read_lock,
438 .readdelay = rcu_read_delay, /* just reuse rcu's version. */ 438 .read_delay = rcu_read_delay, /* just reuse rcu's version. */
439 .readunlock = rcu_bh_torture_read_unlock, 439 .readunlock = rcu_bh_torture_read_unlock,
440 .completed = rcu_bh_torture_completed, 440 .completed = rcu_bh_torture_completed,
441 .deferredfree = rcu_bh_torture_deferred_free, 441 .deferred_free = rcu_bh_torture_deferred_free,
442 .sync = rcu_bh_torture_synchronize, 442 .sync = rcu_bh_torture_synchronize,
443 .cb_barrier = rcu_barrier_bh, 443 .cb_barrier = rcu_barrier_bh,
444 .stats = NULL, 444 .stats = NULL,
445 .irqcapable = 1, 445 .irq_capable = 1,
446 .name = "rcu_bh" 446 .name = "rcu_bh"
447}; 447};
448 448
449static struct rcu_torture_ops rcu_bh_sync_ops = { 449static struct rcu_torture_ops rcu_bh_sync_ops = {
450 .init = rcu_sync_torture_init, 450 .init = rcu_sync_torture_init,
451 .cleanup = NULL, 451 .cleanup = NULL,
452 .readlock = rcu_bh_torture_read_lock, 452 .readlock = rcu_bh_torture_read_lock,
453 .readdelay = rcu_read_delay, /* just reuse rcu's version. */ 453 .read_delay = rcu_read_delay, /* just reuse rcu's version. */
454 .readunlock = rcu_bh_torture_read_unlock, 454 .readunlock = rcu_bh_torture_read_unlock,
455 .completed = rcu_bh_torture_completed, 455 .completed = rcu_bh_torture_completed,
456 .deferredfree = rcu_sync_torture_deferred_free, 456 .deferred_free = rcu_sync_torture_deferred_free,
457 .sync = rcu_bh_torture_synchronize, 457 .sync = rcu_bh_torture_synchronize,
458 .cb_barrier = NULL, 458 .cb_barrier = NULL,
459 .stats = NULL, 459 .stats = NULL,
460 .irqcapable = 1, 460 .irq_capable = 1,
461 .name = "rcu_bh_sync" 461 .name = "rcu_bh_sync"
462}; 462};
463 463
464/* 464/*
@@ -530,17 +530,17 @@ static int srcu_torture_stats(char *page)
530} 530}
531 531
532static struct rcu_torture_ops srcu_ops = { 532static struct rcu_torture_ops srcu_ops = {
533 .init = srcu_torture_init, 533 .init = srcu_torture_init,
534 .cleanup = srcu_torture_cleanup, 534 .cleanup = srcu_torture_cleanup,
535 .readlock = srcu_torture_read_lock, 535 .readlock = srcu_torture_read_lock,
536 .readdelay = srcu_read_delay, 536 .read_delay = srcu_read_delay,
537 .readunlock = srcu_torture_read_unlock, 537 .readunlock = srcu_torture_read_unlock,
538 .completed = srcu_torture_completed, 538 .completed = srcu_torture_completed,
539 .deferredfree = rcu_sync_torture_deferred_free, 539 .deferred_free = rcu_sync_torture_deferred_free,
540 .sync = srcu_torture_synchronize, 540 .sync = srcu_torture_synchronize,
541 .cb_barrier = NULL, 541 .cb_barrier = NULL,
542 .stats = srcu_torture_stats, 542 .stats = srcu_torture_stats,
543 .name = "srcu" 543 .name = "srcu"
544}; 544};
545 545
546/* 546/*
@@ -574,32 +574,49 @@ static void sched_torture_synchronize(void)
574} 574}
575 575
576static struct rcu_torture_ops sched_ops = { 576static struct rcu_torture_ops sched_ops = {
577 .init = rcu_sync_torture_init, 577 .init = rcu_sync_torture_init,
578 .cleanup = NULL, 578 .cleanup = NULL,
579 .readlock = sched_torture_read_lock, 579 .readlock = sched_torture_read_lock,
580 .readdelay = rcu_read_delay, /* just reuse rcu's version. */ 580 .read_delay = rcu_read_delay, /* just reuse rcu's version. */
581 .readunlock = sched_torture_read_unlock, 581 .readunlock = sched_torture_read_unlock,
582 .completed = sched_torture_completed, 582 .completed = sched_torture_completed,
583 .deferredfree = rcu_sched_torture_deferred_free, 583 .deferred_free = rcu_sched_torture_deferred_free,
584 .sync = sched_torture_synchronize, 584 .sync = sched_torture_synchronize,
585 .cb_barrier = rcu_barrier_sched, 585 .cb_barrier = rcu_barrier_sched,
586 .stats = NULL, 586 .stats = NULL,
587 .irqcapable = 1, 587 .irq_capable = 1,
588 .name = "sched" 588 .name = "sched"
589}; 589};
590 590
591static struct rcu_torture_ops sched_ops_sync = { 591static struct rcu_torture_ops sched_ops_sync = {
592 .init = rcu_sync_torture_init, 592 .init = rcu_sync_torture_init,
593 .cleanup = NULL, 593 .cleanup = NULL,
594 .readlock = sched_torture_read_lock, 594 .readlock = sched_torture_read_lock,
595 .readdelay = rcu_read_delay, /* just reuse rcu's version. */ 595 .read_delay = rcu_read_delay, /* just reuse rcu's version. */
596 .readunlock = sched_torture_read_unlock, 596 .readunlock = sched_torture_read_unlock,
597 .completed = sched_torture_completed, 597 .completed = sched_torture_completed,
598 .deferredfree = rcu_sync_torture_deferred_free, 598 .deferred_free = rcu_sync_torture_deferred_free,
599 .sync = sched_torture_synchronize, 599 .sync = sched_torture_synchronize,
600 .cb_barrier = NULL, 600 .cb_barrier = NULL,
601 .stats = NULL, 601 .stats = NULL,
602 .name = "sched_sync" 602 .name = "sched_sync"
603};
604
605extern int rcu_expedited_torture_stats(char *page);
606
607static struct rcu_torture_ops sched_expedited_ops = {
608 .init = rcu_sync_torture_init,
609 .cleanup = NULL,
610 .readlock = sched_torture_read_lock,
611 .read_delay = rcu_read_delay, /* just reuse rcu's version. */
612 .readunlock = sched_torture_read_unlock,
613 .completed = sched_torture_completed,
614 .deferred_free = rcu_sync_torture_deferred_free,
615 .sync = synchronize_sched_expedited,
616 .cb_barrier = NULL,
617 .stats = rcu_expedited_torture_stats,
618 .irq_capable = 1,
619 .name = "sched_expedited"
603}; 620};
604 621
605/* 622/*
@@ -635,7 +652,7 @@ rcu_torture_writer(void *arg)
635 i = RCU_TORTURE_PIPE_LEN; 652 i = RCU_TORTURE_PIPE_LEN;
636 atomic_inc(&rcu_torture_wcount[i]); 653 atomic_inc(&rcu_torture_wcount[i]);
637 old_rp->rtort_pipe_count++; 654 old_rp->rtort_pipe_count++;
638 cur_ops->deferredfree(old_rp); 655 cur_ops->deferred_free(old_rp);
639 } 656 }
640 rcu_torture_current_version++; 657 rcu_torture_current_version++;
641 oldbatch = cur_ops->completed(); 658 oldbatch = cur_ops->completed();
@@ -700,7 +717,7 @@ static void rcu_torture_timer(unsigned long unused)
700 if (p->rtort_mbtest == 0) 717 if (p->rtort_mbtest == 0)
701 atomic_inc(&n_rcu_torture_mberror); 718 atomic_inc(&n_rcu_torture_mberror);
702 spin_lock(&rand_lock); 719 spin_lock(&rand_lock);
703 cur_ops->readdelay(&rand); 720 cur_ops->read_delay(&rand);
704 n_rcu_torture_timers++; 721 n_rcu_torture_timers++;
705 spin_unlock(&rand_lock); 722 spin_unlock(&rand_lock);
706 preempt_disable(); 723 preempt_disable();
@@ -738,11 +755,11 @@ rcu_torture_reader(void *arg)
738 755
739 VERBOSE_PRINTK_STRING("rcu_torture_reader task started"); 756 VERBOSE_PRINTK_STRING("rcu_torture_reader task started");
740 set_user_nice(current, 19); 757 set_user_nice(current, 19);
741 if (irqreader && cur_ops->irqcapable) 758 if (irqreader && cur_ops->irq_capable)
742 setup_timer_on_stack(&t, rcu_torture_timer, 0); 759 setup_timer_on_stack(&t, rcu_torture_timer, 0);
743 760
744 do { 761 do {
745 if (irqreader && cur_ops->irqcapable) { 762 if (irqreader && cur_ops->irq_capable) {
746 if (!timer_pending(&t)) 763 if (!timer_pending(&t))
747 mod_timer(&t, 1); 764 mod_timer(&t, 1);
748 } 765 }
@@ -757,7 +774,7 @@ rcu_torture_reader(void *arg)
757 } 774 }
758 if (p->rtort_mbtest == 0) 775 if (p->rtort_mbtest == 0)
759 atomic_inc(&n_rcu_torture_mberror); 776 atomic_inc(&n_rcu_torture_mberror);
760 cur_ops->readdelay(&rand); 777 cur_ops->read_delay(&rand);
761 preempt_disable(); 778 preempt_disable();
762 pipe_count = p->rtort_pipe_count; 779 pipe_count = p->rtort_pipe_count;
763 if (pipe_count > RCU_TORTURE_PIPE_LEN) { 780 if (pipe_count > RCU_TORTURE_PIPE_LEN) {
@@ -778,7 +795,7 @@ rcu_torture_reader(void *arg)
778 } while (!kthread_should_stop() && fullstop == FULLSTOP_DONTSTOP); 795 } while (!kthread_should_stop() && fullstop == FULLSTOP_DONTSTOP);
779 VERBOSE_PRINTK_STRING("rcu_torture_reader task stopping"); 796 VERBOSE_PRINTK_STRING("rcu_torture_reader task stopping");
780 rcutorture_shutdown_absorb("rcu_torture_reader"); 797 rcutorture_shutdown_absorb("rcu_torture_reader");
781 if (irqreader && cur_ops->irqcapable) 798 if (irqreader && cur_ops->irq_capable)
782 del_timer_sync(&t); 799 del_timer_sync(&t);
783 while (!kthread_should_stop()) 800 while (!kthread_should_stop())
784 schedule_timeout_uninterruptible(1); 801 schedule_timeout_uninterruptible(1);
@@ -1078,6 +1095,7 @@ rcu_torture_init(void)
1078 int firsterr = 0; 1095 int firsterr = 0;
1079 static struct rcu_torture_ops *torture_ops[] = 1096 static struct rcu_torture_ops *torture_ops[] =
1080 { &rcu_ops, &rcu_sync_ops, &rcu_bh_ops, &rcu_bh_sync_ops, 1097 { &rcu_ops, &rcu_sync_ops, &rcu_bh_ops, &rcu_bh_sync_ops,
1098 &sched_expedited_ops,
1081 &srcu_ops, &sched_ops, &sched_ops_sync, }; 1099 &srcu_ops, &sched_ops, &sched_ops_sync, };
1082 1100
1083 mutex_lock(&fullstop_mutex); 1101 mutex_lock(&fullstop_mutex);
diff --git a/kernel/rcutree.c b/kernel/rcutree.c
index 9c5fa9fc57ec..6b11b07cfe7f 100644
--- a/kernel/rcutree.c
+++ b/kernel/rcutree.c
@@ -47,6 +47,8 @@
47#include <linux/mutex.h> 47#include <linux/mutex.h>
48#include <linux/time.h> 48#include <linux/time.h>
49 49
50#include "rcutree.h"
51
50#ifdef CONFIG_DEBUG_LOCK_ALLOC 52#ifdef CONFIG_DEBUG_LOCK_ALLOC
51static struct lock_class_key rcu_lock_key; 53static struct lock_class_key rcu_lock_key;
52struct lockdep_map rcu_lock_map = 54struct lockdep_map rcu_lock_map =
@@ -73,30 +75,59 @@ EXPORT_SYMBOL_GPL(rcu_lock_map);
73 .n_force_qs_ngp = 0, \ 75 .n_force_qs_ngp = 0, \
74} 76}
75 77
76struct rcu_state rcu_state = RCU_STATE_INITIALIZER(rcu_state); 78struct rcu_state rcu_sched_state = RCU_STATE_INITIALIZER(rcu_sched_state);
77DEFINE_PER_CPU(struct rcu_data, rcu_data); 79DEFINE_PER_CPU(struct rcu_data, rcu_sched_data);
78 80
79struct rcu_state rcu_bh_state = RCU_STATE_INITIALIZER(rcu_bh_state); 81struct rcu_state rcu_bh_state = RCU_STATE_INITIALIZER(rcu_bh_state);
80DEFINE_PER_CPU(struct rcu_data, rcu_bh_data); 82DEFINE_PER_CPU(struct rcu_data, rcu_bh_data);
81 83
84extern long rcu_batches_completed_sched(void);
85static struct rcu_node *rcu_get_root(struct rcu_state *rsp);
86static void cpu_quiet_msk(unsigned long mask, struct rcu_state *rsp,
87 struct rcu_node *rnp, unsigned long flags);
88static void cpu_quiet_msk_finish(struct rcu_state *rsp, unsigned long flags);
89#ifdef CONFIG_HOTPLUG_CPU
90static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp);
91#endif /* #ifdef CONFIG_HOTPLUG_CPU */
92static void __rcu_process_callbacks(struct rcu_state *rsp,
93 struct rcu_data *rdp);
94static void __call_rcu(struct rcu_head *head,
95 void (*func)(struct rcu_head *rcu),
96 struct rcu_state *rsp);
97static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp);
98static void __cpuinit rcu_init_percpu_data(int cpu, struct rcu_state *rsp,
99 int preemptable);
100
101#include "rcutree_plugin.h"
102
82/* 103/*
83 * Increment the quiescent state counter. 104 * Note a quiescent state. Because we do not need to know
84 * The counter is a bit degenerated: We do not need to know
85 * how many quiescent states passed, just if there was at least 105 * how many quiescent states passed, just if there was at least
86 * one since the start of the grace period. Thus just a flag. 106 * one since the start of the grace period, this just sets a flag.
87 */ 107 */
88void rcu_qsctr_inc(int cpu) 108void rcu_sched_qs(int cpu)
89{ 109{
90 struct rcu_data *rdp = &per_cpu(rcu_data, cpu); 110 unsigned long flags;
111 struct rcu_data *rdp;
112
113 local_irq_save(flags);
114 rdp = &per_cpu(rcu_sched_data, cpu);
91 rdp->passed_quiesc = 1; 115 rdp->passed_quiesc = 1;
92 rdp->passed_quiesc_completed = rdp->completed; 116 rdp->passed_quiesc_completed = rdp->completed;
117 rcu_preempt_qs(cpu);
118 local_irq_restore(flags);
93} 119}
94 120
95void rcu_bh_qsctr_inc(int cpu) 121void rcu_bh_qs(int cpu)
96{ 122{
97 struct rcu_data *rdp = &per_cpu(rcu_bh_data, cpu); 123 unsigned long flags;
124 struct rcu_data *rdp;
125
126 local_irq_save(flags);
127 rdp = &per_cpu(rcu_bh_data, cpu);
98 rdp->passed_quiesc = 1; 128 rdp->passed_quiesc = 1;
99 rdp->passed_quiesc_completed = rdp->completed; 129 rdp->passed_quiesc_completed = rdp->completed;
130 local_irq_restore(flags);
100} 131}
101 132
102#ifdef CONFIG_NO_HZ 133#ifdef CONFIG_NO_HZ
@@ -111,15 +142,16 @@ static int qhimark = 10000; /* If this many pending, ignore blimit. */
111static int qlowmark = 100; /* Once only this many pending, use blimit. */ 142static int qlowmark = 100; /* Once only this many pending, use blimit. */
112 143
113static void force_quiescent_state(struct rcu_state *rsp, int relaxed); 144static void force_quiescent_state(struct rcu_state *rsp, int relaxed);
145static int rcu_pending(int cpu);
114 146
115/* 147/*
116 * Return the number of RCU batches processed thus far for debug & stats. 148 * Return the number of RCU-sched batches processed thus far for debug & stats.
117 */ 149 */
118long rcu_batches_completed(void) 150long rcu_batches_completed_sched(void)
119{ 151{
120 return rcu_state.completed; 152 return rcu_sched_state.completed;
121} 153}
122EXPORT_SYMBOL_GPL(rcu_batches_completed); 154EXPORT_SYMBOL_GPL(rcu_batches_completed_sched);
123 155
124/* 156/*
125 * Return the number of RCU BH batches processed thus far for debug & stats. 157 * Return the number of RCU BH batches processed thus far for debug & stats.
@@ -182,6 +214,10 @@ static int rcu_implicit_offline_qs(struct rcu_data *rdp)
182 return 1; 214 return 1;
183 } 215 }
184 216
217 /* If preemptable RCU, no point in sending reschedule IPI. */
218 if (rdp->preemptable)
219 return 0;
220
185 /* The CPU is online, so send it a reschedule IPI. */ 221 /* The CPU is online, so send it a reschedule IPI. */
186 if (rdp->cpu != smp_processor_id()) 222 if (rdp->cpu != smp_processor_id())
187 smp_send_reschedule(rdp->cpu); 223 smp_send_reschedule(rdp->cpu);
@@ -194,7 +230,6 @@ static int rcu_implicit_offline_qs(struct rcu_data *rdp)
194#endif /* #ifdef CONFIG_SMP */ 230#endif /* #ifdef CONFIG_SMP */
195 231
196#ifdef CONFIG_NO_HZ 232#ifdef CONFIG_NO_HZ
197static DEFINE_RATELIMIT_STATE(rcu_rs, 10 * HZ, 5);
198 233
199/** 234/**
200 * rcu_enter_nohz - inform RCU that current CPU is entering nohz 235 * rcu_enter_nohz - inform RCU that current CPU is entering nohz
@@ -214,7 +249,7 @@ void rcu_enter_nohz(void)
214 rdtp = &__get_cpu_var(rcu_dynticks); 249 rdtp = &__get_cpu_var(rcu_dynticks);
215 rdtp->dynticks++; 250 rdtp->dynticks++;
216 rdtp->dynticks_nesting--; 251 rdtp->dynticks_nesting--;
217 WARN_ON_RATELIMIT(rdtp->dynticks & 0x1, &rcu_rs); 252 WARN_ON_ONCE(rdtp->dynticks & 0x1);
218 local_irq_restore(flags); 253 local_irq_restore(flags);
219} 254}
220 255
@@ -233,7 +268,7 @@ void rcu_exit_nohz(void)
233 rdtp = &__get_cpu_var(rcu_dynticks); 268 rdtp = &__get_cpu_var(rcu_dynticks);
234 rdtp->dynticks++; 269 rdtp->dynticks++;
235 rdtp->dynticks_nesting++; 270 rdtp->dynticks_nesting++;
236 WARN_ON_RATELIMIT(!(rdtp->dynticks & 0x1), &rcu_rs); 271 WARN_ON_ONCE(!(rdtp->dynticks & 0x1));
237 local_irq_restore(flags); 272 local_irq_restore(flags);
238 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */ 273 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
239} 274}
@@ -252,7 +287,7 @@ void rcu_nmi_enter(void)
252 if (rdtp->dynticks & 0x1) 287 if (rdtp->dynticks & 0x1)
253 return; 288 return;
254 rdtp->dynticks_nmi++; 289 rdtp->dynticks_nmi++;
255 WARN_ON_RATELIMIT(!(rdtp->dynticks_nmi & 0x1), &rcu_rs); 290 WARN_ON_ONCE(!(rdtp->dynticks_nmi & 0x1));
256 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */ 291 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
257} 292}
258 293
@@ -271,7 +306,7 @@ void rcu_nmi_exit(void)
271 return; 306 return;
272 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */ 307 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
273 rdtp->dynticks_nmi++; 308 rdtp->dynticks_nmi++;
274 WARN_ON_RATELIMIT(rdtp->dynticks_nmi & 0x1, &rcu_rs); 309 WARN_ON_ONCE(rdtp->dynticks_nmi & 0x1);
275} 310}
276 311
277/** 312/**
@@ -287,7 +322,7 @@ void rcu_irq_enter(void)
287 if (rdtp->dynticks_nesting++) 322 if (rdtp->dynticks_nesting++)
288 return; 323 return;
289 rdtp->dynticks++; 324 rdtp->dynticks++;
290 WARN_ON_RATELIMIT(!(rdtp->dynticks & 0x1), &rcu_rs); 325 WARN_ON_ONCE(!(rdtp->dynticks & 0x1));
291 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */ 326 smp_mb(); /* CPUs seeing ++ must see later RCU read-side crit sects */
292} 327}
293 328
@@ -306,10 +341,10 @@ void rcu_irq_exit(void)
306 return; 341 return;
307 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */ 342 smp_mb(); /* CPUs seeing ++ must see prior RCU read-side crit sects */
308 rdtp->dynticks++; 343 rdtp->dynticks++;
309 WARN_ON_RATELIMIT(rdtp->dynticks & 0x1, &rcu_rs); 344 WARN_ON_ONCE(rdtp->dynticks & 0x1);
310 345
311 /* If the interrupt queued a callback, get out of dyntick mode. */ 346 /* If the interrupt queued a callback, get out of dyntick mode. */
312 if (__get_cpu_var(rcu_data).nxtlist || 347 if (__get_cpu_var(rcu_sched_data).nxtlist ||
313 __get_cpu_var(rcu_bh_data).nxtlist) 348 __get_cpu_var(rcu_bh_data).nxtlist)
314 set_need_resched(); 349 set_need_resched();
315} 350}
@@ -462,6 +497,7 @@ static void print_other_cpu_stall(struct rcu_state *rsp)
462 497
463 printk(KERN_ERR "INFO: RCU detected CPU stalls:"); 498 printk(KERN_ERR "INFO: RCU detected CPU stalls:");
464 for (; rnp_cur < rnp_end; rnp_cur++) { 499 for (; rnp_cur < rnp_end; rnp_cur++) {
500 rcu_print_task_stall(rnp);
465 if (rnp_cur->qsmask == 0) 501 if (rnp_cur->qsmask == 0)
466 continue; 502 continue;
467 for (cpu = 0; cpu <= rnp_cur->grphi - rnp_cur->grplo; cpu++) 503 for (cpu = 0; cpu <= rnp_cur->grphi - rnp_cur->grplo; cpu++)
@@ -679,6 +715,19 @@ rcu_process_gp_end(struct rcu_state *rsp, struct rcu_data *rdp)
679} 715}
680 716
681/* 717/*
718 * Clean up after the prior grace period and let rcu_start_gp() start up
719 * the next grace period if one is needed. Note that the caller must
720 * hold rnp->lock, as required by rcu_start_gp(), which will release it.
721 */
722static void cpu_quiet_msk_finish(struct rcu_state *rsp, unsigned long flags)
723 __releases(rnp->lock)
724{
725 rsp->completed = rsp->gpnum;
726 rcu_process_gp_end(rsp, rsp->rda[smp_processor_id()]);
727 rcu_start_gp(rsp, flags); /* releases root node's rnp->lock. */
728}
729
730/*
682 * Similar to cpu_quiet(), for which it is a helper function. Allows 731 * Similar to cpu_quiet(), for which it is a helper function. Allows
683 * a group of CPUs to be quieted at one go, though all the CPUs in the 732 * a group of CPUs to be quieted at one go, though all the CPUs in the
684 * group must be represented by the same leaf rcu_node structure. 733 * group must be represented by the same leaf rcu_node structure.
@@ -699,7 +748,7 @@ cpu_quiet_msk(unsigned long mask, struct rcu_state *rsp, struct rcu_node *rnp,
699 return; 748 return;
700 } 749 }
701 rnp->qsmask &= ~mask; 750 rnp->qsmask &= ~mask;
702 if (rnp->qsmask != 0) { 751 if (rnp->qsmask != 0 || rcu_preempted_readers(rnp)) {
703 752
704 /* Other bits still set at this level, so done. */ 753 /* Other bits still set at this level, so done. */
705 spin_unlock_irqrestore(&rnp->lock, flags); 754 spin_unlock_irqrestore(&rnp->lock, flags);
@@ -719,14 +768,10 @@ cpu_quiet_msk(unsigned long mask, struct rcu_state *rsp, struct rcu_node *rnp,
719 768
720 /* 769 /*
721 * Get here if we are the last CPU to pass through a quiescent 770 * Get here if we are the last CPU to pass through a quiescent
722 * state for this grace period. Clean up and let rcu_start_gp() 771 * state for this grace period. Invoke cpu_quiet_msk_finish()
723 * start up the next grace period if one is needed. Note that 772 * to clean up and start the next grace period if one is needed.
724 * we still hold rnp->lock, as required by rcu_start_gp(), which
725 * will release it.
726 */ 773 */
727 rsp->completed = rsp->gpnum; 774 cpu_quiet_msk_finish(rsp, flags); /* releases rnp->lock. */
728 rcu_process_gp_end(rsp, rsp->rda[smp_processor_id()]);
729 rcu_start_gp(rsp, flags); /* releases rnp->lock. */
730} 775}
731 776
732/* 777/*
@@ -833,11 +878,12 @@ static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp)
833 spin_lock(&rnp->lock); /* irqs already disabled. */ 878 spin_lock(&rnp->lock); /* irqs already disabled. */
834 rnp->qsmaskinit &= ~mask; 879 rnp->qsmaskinit &= ~mask;
835 if (rnp->qsmaskinit != 0) { 880 if (rnp->qsmaskinit != 0) {
836 spin_unlock(&rnp->lock); /* irqs already disabled. */ 881 spin_unlock(&rnp->lock); /* irqs remain disabled. */
837 break; 882 break;
838 } 883 }
884 rcu_preempt_offline_tasks(rsp, rnp);
839 mask = rnp->grpmask; 885 mask = rnp->grpmask;
840 spin_unlock(&rnp->lock); /* irqs already disabled. */ 886 spin_unlock(&rnp->lock); /* irqs remain disabled. */
841 rnp = rnp->parent; 887 rnp = rnp->parent;
842 } while (rnp != NULL); 888 } while (rnp != NULL);
843 lastcomp = rsp->completed; 889 lastcomp = rsp->completed;
@@ -850,7 +896,7 @@ static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp)
850 /* 896 /*
851 * Move callbacks from the outgoing CPU to the running CPU. 897 * Move callbacks from the outgoing CPU to the running CPU.
852 * Note that the outgoing CPU is now quiscent, so it is now 898 * Note that the outgoing CPU is now quiscent, so it is now
853 * (uncharacteristically) safe to access it rcu_data structure. 899 * (uncharacteristically) safe to access its rcu_data structure.
854 * Note also that we must carefully retain the order of the 900 * Note also that we must carefully retain the order of the
855 * outgoing CPU's callbacks in order for rcu_barrier() to work 901 * outgoing CPU's callbacks in order for rcu_barrier() to work
856 * correctly. Finally, note that we start all the callbacks 902 * correctly. Finally, note that we start all the callbacks
@@ -881,8 +927,9 @@ static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp)
881 */ 927 */
882static void rcu_offline_cpu(int cpu) 928static void rcu_offline_cpu(int cpu)
883{ 929{
884 __rcu_offline_cpu(cpu, &rcu_state); 930 __rcu_offline_cpu(cpu, &rcu_sched_state);
885 __rcu_offline_cpu(cpu, &rcu_bh_state); 931 __rcu_offline_cpu(cpu, &rcu_bh_state);
932 rcu_preempt_offline_cpu(cpu);
886} 933}
887 934
888#else /* #ifdef CONFIG_HOTPLUG_CPU */ 935#else /* #ifdef CONFIG_HOTPLUG_CPU */
@@ -968,6 +1015,8 @@ static void rcu_do_batch(struct rcu_data *rdp)
968 */ 1015 */
969void rcu_check_callbacks(int cpu, int user) 1016void rcu_check_callbacks(int cpu, int user)
970{ 1017{
1018 if (!rcu_pending(cpu))
1019 return; /* if nothing for RCU to do. */
971 if (user || 1020 if (user ||
972 (idle_cpu(cpu) && rcu_scheduler_active && 1021 (idle_cpu(cpu) && rcu_scheduler_active &&
973 !in_softirq() && hardirq_count() <= (1 << HARDIRQ_SHIFT))) { 1022 !in_softirq() && hardirq_count() <= (1 << HARDIRQ_SHIFT))) {
@@ -976,17 +1025,16 @@ void rcu_check_callbacks(int cpu, int user)
976 * Get here if this CPU took its interrupt from user 1025 * Get here if this CPU took its interrupt from user
977 * mode or from the idle loop, and if this is not a 1026 * mode or from the idle loop, and if this is not a
978 * nested interrupt. In this case, the CPU is in 1027 * nested interrupt. In this case, the CPU is in
979 * a quiescent state, so count it. 1028 * a quiescent state, so note it.
980 * 1029 *
981 * No memory barrier is required here because both 1030 * No memory barrier is required here because both
982 * rcu_qsctr_inc() and rcu_bh_qsctr_inc() reference 1031 * rcu_sched_qs() and rcu_bh_qs() reference only CPU-local
983 * only CPU-local variables that other CPUs neither 1032 * variables that other CPUs neither access nor modify,
984 * access nor modify, at least not while the corresponding 1033 * at least not while the corresponding CPU is online.
985 * CPU is online.
986 */ 1034 */
987 1035
988 rcu_qsctr_inc(cpu); 1036 rcu_sched_qs(cpu);
989 rcu_bh_qsctr_inc(cpu); 1037 rcu_bh_qs(cpu);
990 1038
991 } else if (!in_softirq()) { 1039 } else if (!in_softirq()) {
992 1040
@@ -994,11 +1042,12 @@ void rcu_check_callbacks(int cpu, int user)
994 * Get here if this CPU did not take its interrupt from 1042 * Get here if this CPU did not take its interrupt from
995 * softirq, in other words, if it is not interrupting 1043 * softirq, in other words, if it is not interrupting
996 * a rcu_bh read-side critical section. This is an _bh 1044 * a rcu_bh read-side critical section. This is an _bh
997 * critical section, so count it. 1045 * critical section, so note it.
998 */ 1046 */
999 1047
1000 rcu_bh_qsctr_inc(cpu); 1048 rcu_bh_qs(cpu);
1001 } 1049 }
1050 rcu_preempt_check_callbacks(cpu);
1002 raise_softirq(RCU_SOFTIRQ); 1051 raise_softirq(RCU_SOFTIRQ);
1003} 1052}
1004 1053
@@ -1137,6 +1186,8 @@ __rcu_process_callbacks(struct rcu_state *rsp, struct rcu_data *rdp)
1137{ 1186{
1138 unsigned long flags; 1187 unsigned long flags;
1139 1188
1189 WARN_ON_ONCE(rdp->beenonline == 0);
1190
1140 /* 1191 /*
1141 * If an RCU GP has gone long enough, go check for dyntick 1192 * If an RCU GP has gone long enough, go check for dyntick
1142 * idle CPUs and, if needed, send resched IPIs. 1193 * idle CPUs and, if needed, send resched IPIs.
@@ -1175,8 +1226,10 @@ static void rcu_process_callbacks(struct softirq_action *unused)
1175 */ 1226 */
1176 smp_mb(); /* See above block comment. */ 1227 smp_mb(); /* See above block comment. */
1177 1228
1178 __rcu_process_callbacks(&rcu_state, &__get_cpu_var(rcu_data)); 1229 __rcu_process_callbacks(&rcu_sched_state,
1230 &__get_cpu_var(rcu_sched_data));
1179 __rcu_process_callbacks(&rcu_bh_state, &__get_cpu_var(rcu_bh_data)); 1231 __rcu_process_callbacks(&rcu_bh_state, &__get_cpu_var(rcu_bh_data));
1232 rcu_preempt_process_callbacks();
1180 1233
1181 /* 1234 /*
1182 * Memory references from any later RCU read-side critical sections 1235 * Memory references from any later RCU read-side critical sections
@@ -1232,13 +1285,13 @@ __call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu),
1232} 1285}
1233 1286
1234/* 1287/*
1235 * Queue an RCU callback for invocation after a grace period. 1288 * Queue an RCU-sched callback for invocation after a grace period.
1236 */ 1289 */
1237void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) 1290void call_rcu_sched(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
1238{ 1291{
1239 __call_rcu(head, func, &rcu_state); 1292 __call_rcu(head, func, &rcu_sched_state);
1240} 1293}
1241EXPORT_SYMBOL_GPL(call_rcu); 1294EXPORT_SYMBOL_GPL(call_rcu_sched);
1242 1295
1243/* 1296/*
1244 * Queue an RCU for invocation after a quicker grace period. 1297 * Queue an RCU for invocation after a quicker grace period.
@@ -1310,10 +1363,11 @@ static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp)
1310 * by the current CPU, returning 1 if so. This function is part of the 1363 * by the current CPU, returning 1 if so. This function is part of the
1311 * RCU implementation; it is -not- an exported member of the RCU API. 1364 * RCU implementation; it is -not- an exported member of the RCU API.
1312 */ 1365 */
1313int rcu_pending(int cpu) 1366static int rcu_pending(int cpu)
1314{ 1367{
1315 return __rcu_pending(&rcu_state, &per_cpu(rcu_data, cpu)) || 1368 return __rcu_pending(&rcu_sched_state, &per_cpu(rcu_sched_data, cpu)) ||
1316 __rcu_pending(&rcu_bh_state, &per_cpu(rcu_bh_data, cpu)); 1369 __rcu_pending(&rcu_bh_state, &per_cpu(rcu_bh_data, cpu)) ||
1370 rcu_preempt_pending(cpu);
1317} 1371}
1318 1372
1319/* 1373/*
@@ -1325,27 +1379,46 @@ int rcu_pending(int cpu)
1325int rcu_needs_cpu(int cpu) 1379int rcu_needs_cpu(int cpu)
1326{ 1380{
1327 /* RCU callbacks either ready or pending? */ 1381 /* RCU callbacks either ready or pending? */
1328 return per_cpu(rcu_data, cpu).nxtlist || 1382 return per_cpu(rcu_sched_data, cpu).nxtlist ||
1329 per_cpu(rcu_bh_data, cpu).nxtlist; 1383 per_cpu(rcu_bh_data, cpu).nxtlist ||
1384 rcu_preempt_needs_cpu(cpu);
1330} 1385}
1331 1386
1332/* 1387/*
1333 * Initialize a CPU's per-CPU RCU data. We take this "scorched earth" 1388 * Do boot-time initialization of a CPU's per-CPU RCU data.
1334 * approach so that we don't have to worry about how long the CPU has
1335 * been gone, or whether it ever was online previously. We do trust the
1336 * ->mynode field, as it is constant for a given struct rcu_data and
1337 * initialized during early boot.
1338 *
1339 * Note that only one online or offline event can be happening at a given
1340 * time. Note also that we can accept some slop in the rsp->completed
1341 * access due to the fact that this CPU cannot possibly have any RCU
1342 * callbacks in flight yet.
1343 */ 1389 */
1344static void __cpuinit 1390static void __init
1345rcu_init_percpu_data(int cpu, struct rcu_state *rsp) 1391rcu_boot_init_percpu_data(int cpu, struct rcu_state *rsp)
1346{ 1392{
1347 unsigned long flags; 1393 unsigned long flags;
1348 int i; 1394 int i;
1395 struct rcu_data *rdp = rsp->rda[cpu];
1396 struct rcu_node *rnp = rcu_get_root(rsp);
1397
1398 /* Set up local state, ensuring consistent view of global state. */
1399 spin_lock_irqsave(&rnp->lock, flags);
1400 rdp->grpmask = 1UL << (cpu - rdp->mynode->grplo);
1401 rdp->nxtlist = NULL;
1402 for (i = 0; i < RCU_NEXT_SIZE; i++)
1403 rdp->nxttail[i] = &rdp->nxtlist;
1404 rdp->qlen = 0;
1405#ifdef CONFIG_NO_HZ
1406 rdp->dynticks = &per_cpu(rcu_dynticks, cpu);
1407#endif /* #ifdef CONFIG_NO_HZ */
1408 rdp->cpu = cpu;
1409 spin_unlock_irqrestore(&rnp->lock, flags);
1410}
1411
1412/*
1413 * Initialize a CPU's per-CPU RCU data. Note that only one online or
1414 * offline event can be happening at a given time. Note also that we
1415 * can accept some slop in the rsp->completed access due to the fact
1416 * that this CPU cannot possibly have any RCU callbacks in flight yet.
1417 */
1418static void __cpuinit
1419rcu_init_percpu_data(int cpu, struct rcu_state *rsp, int preemptable)
1420{
1421 unsigned long flags;
1349 long lastcomp; 1422 long lastcomp;
1350 unsigned long mask; 1423 unsigned long mask;
1351 struct rcu_data *rdp = rsp->rda[cpu]; 1424 struct rcu_data *rdp = rsp->rda[cpu];
@@ -1359,17 +1432,9 @@ rcu_init_percpu_data(int cpu, struct rcu_state *rsp)
1359 rdp->passed_quiesc = 0; /* We could be racing with new GP, */ 1432 rdp->passed_quiesc = 0; /* We could be racing with new GP, */
1360 rdp->qs_pending = 1; /* so set up to respond to current GP. */ 1433 rdp->qs_pending = 1; /* so set up to respond to current GP. */
1361 rdp->beenonline = 1; /* We have now been online. */ 1434 rdp->beenonline = 1; /* We have now been online. */
1435 rdp->preemptable = preemptable;
1362 rdp->passed_quiesc_completed = lastcomp - 1; 1436 rdp->passed_quiesc_completed = lastcomp - 1;
1363 rdp->grpmask = 1UL << (cpu - rdp->mynode->grplo);
1364 rdp->nxtlist = NULL;
1365 for (i = 0; i < RCU_NEXT_SIZE; i++)
1366 rdp->nxttail[i] = &rdp->nxtlist;
1367 rdp->qlen = 0;
1368 rdp->blimit = blimit; 1437 rdp->blimit = blimit;
1369#ifdef CONFIG_NO_HZ
1370 rdp->dynticks = &per_cpu(rcu_dynticks, cpu);
1371#endif /* #ifdef CONFIG_NO_HZ */
1372 rdp->cpu = cpu;
1373 spin_unlock(&rnp->lock); /* irqs remain disabled. */ 1438 spin_unlock(&rnp->lock); /* irqs remain disabled. */
1374 1439
1375 /* 1440 /*
@@ -1410,16 +1475,16 @@ rcu_init_percpu_data(int cpu, struct rcu_state *rsp)
1410 1475
1411static void __cpuinit rcu_online_cpu(int cpu) 1476static void __cpuinit rcu_online_cpu(int cpu)
1412{ 1477{
1413 rcu_init_percpu_data(cpu, &rcu_state); 1478 rcu_init_percpu_data(cpu, &rcu_sched_state, 0);
1414 rcu_init_percpu_data(cpu, &rcu_bh_state); 1479 rcu_init_percpu_data(cpu, &rcu_bh_state, 0);
1415 open_softirq(RCU_SOFTIRQ, rcu_process_callbacks); 1480 rcu_preempt_init_percpu_data(cpu);
1416} 1481}
1417 1482
1418/* 1483/*
1419 * Handle CPU online/offline notifcation events. 1484 * Handle CPU online/offline notification events.
1420 */ 1485 */
1421static int __cpuinit rcu_cpu_notify(struct notifier_block *self, 1486int __cpuinit rcu_cpu_notify(struct notifier_block *self,
1422 unsigned long action, void *hcpu) 1487 unsigned long action, void *hcpu)
1423{ 1488{
1424 long cpu = (long)hcpu; 1489 long cpu = (long)hcpu;
1425 1490
@@ -1491,6 +1556,7 @@ static void __init rcu_init_one(struct rcu_state *rsp)
1491 rnp = rsp->level[i]; 1556 rnp = rsp->level[i];
1492 for (j = 0; j < rsp->levelcnt[i]; j++, rnp++) { 1557 for (j = 0; j < rsp->levelcnt[i]; j++, rnp++) {
1493 spin_lock_init(&rnp->lock); 1558 spin_lock_init(&rnp->lock);
1559 rnp->gpnum = 0;
1494 rnp->qsmask = 0; 1560 rnp->qsmask = 0;
1495 rnp->qsmaskinit = 0; 1561 rnp->qsmaskinit = 0;
1496 rnp->grplo = j * cpustride; 1562 rnp->grplo = j * cpustride;
@@ -1508,16 +1574,20 @@ static void __init rcu_init_one(struct rcu_state *rsp)
1508 j / rsp->levelspread[i - 1]; 1574 j / rsp->levelspread[i - 1];
1509 } 1575 }
1510 rnp->level = i; 1576 rnp->level = i;
1577 INIT_LIST_HEAD(&rnp->blocked_tasks[0]);
1578 INIT_LIST_HEAD(&rnp->blocked_tasks[1]);
1511 } 1579 }
1512 } 1580 }
1513} 1581}
1514 1582
1515/* 1583/*
1516 * Helper macro for __rcu_init(). To be used nowhere else! 1584 * Helper macro for __rcu_init() and __rcu_init_preempt(). To be used
1517 * Assigns leaf node pointers into each CPU's rcu_data structure. 1585 * nowhere else! Assigns leaf node pointers into each CPU's rcu_data
1586 * structure.
1518 */ 1587 */
1519#define RCU_DATA_PTR_INIT(rsp, rcu_data) \ 1588#define RCU_INIT_FLAVOR(rsp, rcu_data) \
1520do { \ 1589do { \
1590 rcu_init_one(rsp); \
1521 rnp = (rsp)->level[NUM_RCU_LVLS - 1]; \ 1591 rnp = (rsp)->level[NUM_RCU_LVLS - 1]; \
1522 j = 0; \ 1592 j = 0; \
1523 for_each_possible_cpu(i) { \ 1593 for_each_possible_cpu(i) { \
@@ -1525,32 +1595,43 @@ do { \
1525 j++; \ 1595 j++; \
1526 per_cpu(rcu_data, i).mynode = &rnp[j]; \ 1596 per_cpu(rcu_data, i).mynode = &rnp[j]; \
1527 (rsp)->rda[i] = &per_cpu(rcu_data, i); \ 1597 (rsp)->rda[i] = &per_cpu(rcu_data, i); \
1598 rcu_boot_init_percpu_data(i, rsp); \
1528 } \ 1599 } \
1529} while (0) 1600} while (0)
1530 1601
1531static struct notifier_block __cpuinitdata rcu_nb = { 1602#ifdef CONFIG_TREE_PREEMPT_RCU
1532 .notifier_call = rcu_cpu_notify, 1603
1533}; 1604void __init __rcu_init_preempt(void)
1605{
1606 int i; /* All used by RCU_INIT_FLAVOR(). */
1607 int j;
1608 struct rcu_node *rnp;
1609
1610 RCU_INIT_FLAVOR(&rcu_preempt_state, rcu_preempt_data);
1611}
1612
1613#else /* #ifdef CONFIG_TREE_PREEMPT_RCU */
1614
1615void __init __rcu_init_preempt(void)
1616{
1617}
1618
1619#endif /* #else #ifdef CONFIG_TREE_PREEMPT_RCU */
1534 1620
1535void __init __rcu_init(void) 1621void __init __rcu_init(void)
1536{ 1622{
1537 int i; /* All used by RCU_DATA_PTR_INIT(). */ 1623 int i; /* All used by RCU_INIT_FLAVOR(). */
1538 int j; 1624 int j;
1539 struct rcu_node *rnp; 1625 struct rcu_node *rnp;
1540 1626
1541 printk(KERN_INFO "Hierarchical RCU implementation.\n"); 1627 rcu_bootup_announce();
1542#ifdef CONFIG_RCU_CPU_STALL_DETECTOR 1628#ifdef CONFIG_RCU_CPU_STALL_DETECTOR
1543 printk(KERN_INFO "RCU-based detection of stalled CPUs is enabled.\n"); 1629 printk(KERN_INFO "RCU-based detection of stalled CPUs is enabled.\n");
1544#endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */ 1630#endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
1545 rcu_init_one(&rcu_state); 1631 RCU_INIT_FLAVOR(&rcu_sched_state, rcu_sched_data);
1546 RCU_DATA_PTR_INIT(&rcu_state, rcu_data); 1632 RCU_INIT_FLAVOR(&rcu_bh_state, rcu_bh_data);
1547 rcu_init_one(&rcu_bh_state); 1633 __rcu_init_preempt();
1548 RCU_DATA_PTR_INIT(&rcu_bh_state, rcu_bh_data); 1634 open_softirq(RCU_SOFTIRQ, rcu_process_callbacks);
1549
1550 for_each_online_cpu(i)
1551 rcu_cpu_notify(&rcu_nb, CPU_UP_PREPARE, (void *)(long)i);
1552 /* Register notifier for non-boot CPUs */
1553 register_cpu_notifier(&rcu_nb);
1554} 1635}
1555 1636
1556module_param(blimit, int, 0); 1637module_param(blimit, int, 0);
diff --git a/kernel/rcutree.h b/kernel/rcutree.h
index 5e872bbf07f5..bf8a6f9f134d 100644
--- a/kernel/rcutree.h
+++ b/kernel/rcutree.h
@@ -1,10 +1,259 @@
1/*
2 * Read-Copy Update mechanism for mutual exclusion (tree-based version)
3 * Internal non-public definitions.
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
18 *
19 * Copyright IBM Corporation, 2008
20 *
21 * Author: Ingo Molnar <mingo@elte.hu>
22 * Paul E. McKenney <paulmck@linux.vnet.ibm.com>
23 */
24
25#include <linux/cache.h>
26#include <linux/spinlock.h>
27#include <linux/threads.h>
28#include <linux/cpumask.h>
29#include <linux/seqlock.h>
30
31/*
32 * Define shape of hierarchy based on NR_CPUS and CONFIG_RCU_FANOUT.
33 * In theory, it should be possible to add more levels straightforwardly.
34 * In practice, this has not been tested, so there is probably some
35 * bug somewhere.
36 */
37#define MAX_RCU_LVLS 3
38#define RCU_FANOUT (CONFIG_RCU_FANOUT)
39#define RCU_FANOUT_SQ (RCU_FANOUT * RCU_FANOUT)
40#define RCU_FANOUT_CUBE (RCU_FANOUT_SQ * RCU_FANOUT)
41
42#if NR_CPUS <= RCU_FANOUT
43# define NUM_RCU_LVLS 1
44# define NUM_RCU_LVL_0 1
45# define NUM_RCU_LVL_1 (NR_CPUS)
46# define NUM_RCU_LVL_2 0
47# define NUM_RCU_LVL_3 0
48#elif NR_CPUS <= RCU_FANOUT_SQ
49# define NUM_RCU_LVLS 2
50# define NUM_RCU_LVL_0 1
51# define NUM_RCU_LVL_1 (((NR_CPUS) + RCU_FANOUT - 1) / RCU_FANOUT)
52# define NUM_RCU_LVL_2 (NR_CPUS)
53# define NUM_RCU_LVL_3 0
54#elif NR_CPUS <= RCU_FANOUT_CUBE
55# define NUM_RCU_LVLS 3
56# define NUM_RCU_LVL_0 1
57# define NUM_RCU_LVL_1 (((NR_CPUS) + RCU_FANOUT_SQ - 1) / RCU_FANOUT_SQ)
58# define NUM_RCU_LVL_2 (((NR_CPUS) + (RCU_FANOUT) - 1) / (RCU_FANOUT))
59# define NUM_RCU_LVL_3 NR_CPUS
60#else
61# error "CONFIG_RCU_FANOUT insufficient for NR_CPUS"
62#endif /* #if (NR_CPUS) <= RCU_FANOUT */
63
64#define RCU_SUM (NUM_RCU_LVL_0 + NUM_RCU_LVL_1 + NUM_RCU_LVL_2 + NUM_RCU_LVL_3)
65#define NUM_RCU_NODES (RCU_SUM - NR_CPUS)
66
67/*
68 * Dynticks per-CPU state.
69 */
70struct rcu_dynticks {
71 int dynticks_nesting; /* Track nesting level, sort of. */
72 int dynticks; /* Even value for dynticks-idle, else odd. */
73 int dynticks_nmi; /* Even value for either dynticks-idle or */
74 /* not in nmi handler, else odd. So this */
75 /* remains even for nmi from irq handler. */
76};
77
78/*
79 * Definition for node within the RCU grace-period-detection hierarchy.
80 */
81struct rcu_node {
82 spinlock_t lock;
83 long gpnum; /* Current grace period for this node. */
84 /* This will either be equal to or one */
85 /* behind the root rcu_node's gpnum. */
86 unsigned long qsmask; /* CPUs or groups that need to switch in */
87 /* order for current grace period to proceed.*/
88 unsigned long qsmaskinit;
89 /* Per-GP initialization for qsmask. */
90 unsigned long grpmask; /* Mask to apply to parent qsmask. */
91 int grplo; /* lowest-numbered CPU or group here. */
92 int grphi; /* highest-numbered CPU or group here. */
93 u8 grpnum; /* CPU/group number for next level up. */
94 u8 level; /* root is at level 0. */
95 struct rcu_node *parent;
96 struct list_head blocked_tasks[2];
97 /* Tasks blocked in RCU read-side critsect. */
98} ____cacheline_internodealigned_in_smp;
99
100/* Index values for nxttail array in struct rcu_data. */
101#define RCU_DONE_TAIL 0 /* Also RCU_WAIT head. */
102#define RCU_WAIT_TAIL 1 /* Also RCU_NEXT_READY head. */
103#define RCU_NEXT_READY_TAIL 2 /* Also RCU_NEXT head. */
104#define RCU_NEXT_TAIL 3
105#define RCU_NEXT_SIZE 4
106
107/* Per-CPU data for read-copy update. */
108struct rcu_data {
109 /* 1) quiescent-state and grace-period handling : */
110 long completed; /* Track rsp->completed gp number */
111 /* in order to detect GP end. */
112 long gpnum; /* Highest gp number that this CPU */
113 /* is aware of having started. */
114 long passed_quiesc_completed;
115 /* Value of completed at time of qs. */
116 bool passed_quiesc; /* User-mode/idle loop etc. */
117 bool qs_pending; /* Core waits for quiesc state. */
118 bool beenonline; /* CPU online at least once. */
119 bool preemptable; /* Preemptable RCU? */
120 struct rcu_node *mynode; /* This CPU's leaf of hierarchy */
121 unsigned long grpmask; /* Mask to apply to leaf qsmask. */
122
123 /* 2) batch handling */
124 /*
125 * If nxtlist is not NULL, it is partitioned as follows.
126 * Any of the partitions might be empty, in which case the
127 * pointer to that partition will be equal to the pointer for
128 * the following partition. When the list is empty, all of
129 * the nxttail elements point to nxtlist, which is NULL.
130 *
131 * [*nxttail[RCU_NEXT_READY_TAIL], NULL = *nxttail[RCU_NEXT_TAIL]):
132 * Entries that might have arrived after current GP ended
133 * [*nxttail[RCU_WAIT_TAIL], *nxttail[RCU_NEXT_READY_TAIL]):
134 * Entries known to have arrived before current GP ended
135 * [*nxttail[RCU_DONE_TAIL], *nxttail[RCU_WAIT_TAIL]):
136 * Entries that batch # <= ->completed - 1: waiting for current GP
137 * [nxtlist, *nxttail[RCU_DONE_TAIL]):
138 * Entries that batch # <= ->completed
139 * The grace period for these entries has completed, and
140 * the other grace-period-completed entries may be moved
141 * here temporarily in rcu_process_callbacks().
142 */
143 struct rcu_head *nxtlist;
144 struct rcu_head **nxttail[RCU_NEXT_SIZE];
145 long qlen; /* # of queued callbacks */
146 long blimit; /* Upper limit on a processed batch */
147
148#ifdef CONFIG_NO_HZ
149 /* 3) dynticks interface. */
150 struct rcu_dynticks *dynticks; /* Shared per-CPU dynticks state. */
151 int dynticks_snap; /* Per-GP tracking for dynticks. */
152 int dynticks_nmi_snap; /* Per-GP tracking for dynticks_nmi. */
153#endif /* #ifdef CONFIG_NO_HZ */
154
155 /* 4) reasons this CPU needed to be kicked by force_quiescent_state */
156#ifdef CONFIG_NO_HZ
157 unsigned long dynticks_fqs; /* Kicked due to dynticks idle. */
158#endif /* #ifdef CONFIG_NO_HZ */
159 unsigned long offline_fqs; /* Kicked due to being offline. */
160 unsigned long resched_ipi; /* Sent a resched IPI. */
161
162 /* 5) __rcu_pending() statistics. */
163 long n_rcu_pending; /* rcu_pending() calls since boot. */
164 long n_rp_qs_pending;
165 long n_rp_cb_ready;
166 long n_rp_cpu_needs_gp;
167 long n_rp_gp_completed;
168 long n_rp_gp_started;
169 long n_rp_need_fqs;
170 long n_rp_need_nothing;
171
172 int cpu;
173};
174
175/* Values for signaled field in struct rcu_state. */
176#define RCU_GP_INIT 0 /* Grace period being initialized. */
177#define RCU_SAVE_DYNTICK 1 /* Need to scan dyntick state. */
178#define RCU_FORCE_QS 2 /* Need to force quiescent state. */
179#ifdef CONFIG_NO_HZ
180#define RCU_SIGNAL_INIT RCU_SAVE_DYNTICK
181#else /* #ifdef CONFIG_NO_HZ */
182#define RCU_SIGNAL_INIT RCU_FORCE_QS
183#endif /* #else #ifdef CONFIG_NO_HZ */
184
185#define RCU_JIFFIES_TILL_FORCE_QS 3 /* for rsp->jiffies_force_qs */
186#ifdef CONFIG_RCU_CPU_STALL_DETECTOR
187#define RCU_SECONDS_TILL_STALL_CHECK (10 * HZ) /* for rsp->jiffies_stall */
188#define RCU_SECONDS_TILL_STALL_RECHECK (30 * HZ) /* for rsp->jiffies_stall */
189#define RCU_STALL_RAT_DELAY 2 /* Allow other CPUs time */
190 /* to take at least one */
191 /* scheduling clock irq */
192 /* before ratting on them. */
193
194#endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
195
196/*
197 * RCU global state, including node hierarchy. This hierarchy is
198 * represented in "heap" form in a dense array. The root (first level)
199 * of the hierarchy is in ->node[0] (referenced by ->level[0]), the second
200 * level in ->node[1] through ->node[m] (->node[1] referenced by ->level[1]),
201 * and the third level in ->node[m+1] and following (->node[m+1] referenced
202 * by ->level[2]). The number of levels is determined by the number of
203 * CPUs and by CONFIG_RCU_FANOUT. Small systems will have a "hierarchy"
204 * consisting of a single rcu_node.
205 */
206struct rcu_state {
207 struct rcu_node node[NUM_RCU_NODES]; /* Hierarchy. */
208 struct rcu_node *level[NUM_RCU_LVLS]; /* Hierarchy levels. */
209 u32 levelcnt[MAX_RCU_LVLS + 1]; /* # nodes in each level. */
210 u8 levelspread[NUM_RCU_LVLS]; /* kids/node in each level. */
211 struct rcu_data *rda[NR_CPUS]; /* array of rdp pointers. */
212
213 /* The following fields are guarded by the root rcu_node's lock. */
214
215 u8 signaled ____cacheline_internodealigned_in_smp;
216 /* Force QS state. */
217 long gpnum; /* Current gp number. */
218 long completed; /* # of last completed gp. */
219 spinlock_t onofflock; /* exclude on/offline and */
220 /* starting new GP. */
221 spinlock_t fqslock; /* Only one task forcing */
222 /* quiescent states. */
223 unsigned long jiffies_force_qs; /* Time at which to invoke */
224 /* force_quiescent_state(). */
225 unsigned long n_force_qs; /* Number of calls to */
226 /* force_quiescent_state(). */
227 unsigned long n_force_qs_lh; /* ~Number of calls leaving */
228 /* due to lock unavailable. */
229 unsigned long n_force_qs_ngp; /* Number of calls leaving */
230 /* due to no GP active. */
231#ifdef CONFIG_RCU_CPU_STALL_DETECTOR
232 unsigned long gp_start; /* Time at which GP started, */
233 /* but in jiffies. */
234 unsigned long jiffies_stall; /* Time at which to check */
235 /* for CPU stalls. */
236#endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
237#ifdef CONFIG_NO_HZ
238 long dynticks_completed; /* Value of completed @ snap. */
239#endif /* #ifdef CONFIG_NO_HZ */
240};
241
242#ifdef RCU_TREE_NONCORE
1 243
2/* 244/*
3 * RCU implementation internal declarations: 245 * RCU implementation internal declarations:
4 */ 246 */
5extern struct rcu_state rcu_state; 247extern struct rcu_state rcu_sched_state;
6DECLARE_PER_CPU(struct rcu_data, rcu_data); 248DECLARE_PER_CPU(struct rcu_data, rcu_sched_data);
7 249
8extern struct rcu_state rcu_bh_state; 250extern struct rcu_state rcu_bh_state;
9DECLARE_PER_CPU(struct rcu_data, rcu_bh_data); 251DECLARE_PER_CPU(struct rcu_data, rcu_bh_data);
10 252
253#ifdef CONFIG_TREE_PREEMPT_RCU
254extern struct rcu_state rcu_preempt_state;
255DECLARE_PER_CPU(struct rcu_data, rcu_preempt_data);
256#endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */
257
258#endif /* #ifdef RCU_TREE_NONCORE */
259
diff --git a/kernel/rcutree_plugin.h b/kernel/rcutree_plugin.h
new file mode 100644
index 000000000000..47789369ea59
--- /dev/null
+++ b/kernel/rcutree_plugin.h
@@ -0,0 +1,532 @@
1/*
2 * Read-Copy Update mechanism for mutual exclusion (tree-based version)
3 * Internal non-public definitions that provide either classic
4 * or preemptable semantics.
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
10 *
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
15 *
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
19 *
20 * Copyright Red Hat, 2009
21 * Copyright IBM Corporation, 2009
22 *
23 * Author: Ingo Molnar <mingo@elte.hu>
24 * Paul E. McKenney <paulmck@linux.vnet.ibm.com>
25 */
26
27
28#ifdef CONFIG_TREE_PREEMPT_RCU
29
30struct rcu_state rcu_preempt_state = RCU_STATE_INITIALIZER(rcu_preempt_state);
31DEFINE_PER_CPU(struct rcu_data, rcu_preempt_data);
32
33/*
34 * Tell them what RCU they are running.
35 */
36static inline void rcu_bootup_announce(void)
37{
38 printk(KERN_INFO
39 "Experimental preemptable hierarchical RCU implementation.\n");
40}
41
42/*
43 * Return the number of RCU-preempt batches processed thus far
44 * for debug and statistics.
45 */
46long rcu_batches_completed_preempt(void)
47{
48 return rcu_preempt_state.completed;
49}
50EXPORT_SYMBOL_GPL(rcu_batches_completed_preempt);
51
52/*
53 * Return the number of RCU batches processed thus far for debug & stats.
54 */
55long rcu_batches_completed(void)
56{
57 return rcu_batches_completed_preempt();
58}
59EXPORT_SYMBOL_GPL(rcu_batches_completed);
60
61/*
62 * Record a preemptable-RCU quiescent state for the specified CPU. Note
63 * that this just means that the task currently running on the CPU is
64 * not in a quiescent state. There might be any number of tasks blocked
65 * while in an RCU read-side critical section.
66 */
67static void rcu_preempt_qs_record(int cpu)
68{
69 struct rcu_data *rdp = &per_cpu(rcu_preempt_data, cpu);
70 rdp->passed_quiesc = 1;
71 rdp->passed_quiesc_completed = rdp->completed;
72}
73
74/*
75 * We have entered the scheduler or are between softirqs in ksoftirqd.
76 * If we are in an RCU read-side critical section, we need to reflect
77 * that in the state of the rcu_node structure corresponding to this CPU.
78 * Caller must disable hardirqs.
79 */
80static void rcu_preempt_qs(int cpu)
81{
82 struct task_struct *t = current;
83 int phase;
84 struct rcu_data *rdp;
85 struct rcu_node *rnp;
86
87 if (t->rcu_read_lock_nesting &&
88 (t->rcu_read_unlock_special & RCU_READ_UNLOCK_BLOCKED) == 0) {
89
90 /* Possibly blocking in an RCU read-side critical section. */
91 rdp = rcu_preempt_state.rda[cpu];
92 rnp = rdp->mynode;
93 spin_lock(&rnp->lock);
94 t->rcu_read_unlock_special |= RCU_READ_UNLOCK_BLOCKED;
95 t->rcu_blocked_node = rnp;
96
97 /*
98 * If this CPU has already checked in, then this task
99 * will hold up the next grace period rather than the
100 * current grace period. Queue the task accordingly.
101 * If the task is queued for the current grace period
102 * (i.e., this CPU has not yet passed through a quiescent
103 * state for the current grace period), then as long
104 * as that task remains queued, the current grace period
105 * cannot end.
106 */
107 phase = !(rnp->qsmask & rdp->grpmask) ^ (rnp->gpnum & 0x1);
108 list_add(&t->rcu_node_entry, &rnp->blocked_tasks[phase]);
109 smp_mb(); /* Ensure later ctxt swtch seen after above. */
110 spin_unlock(&rnp->lock);
111 }
112
113 /*
114 * Either we were not in an RCU read-side critical section to
115 * begin with, or we have now recorded that critical section
116 * globally. Either way, we can now note a quiescent state
117 * for this CPU. Again, if we were in an RCU read-side critical
118 * section, and if that critical section was blocking the current
119 * grace period, then the fact that the task has been enqueued
120 * means that we continue to block the current grace period.
121 */
122 rcu_preempt_qs_record(cpu);
123 t->rcu_read_unlock_special &= ~(RCU_READ_UNLOCK_NEED_QS |
124 RCU_READ_UNLOCK_GOT_QS);
125}
126
127/*
128 * Tree-preemptable RCU implementation for rcu_read_lock().
129 * Just increment ->rcu_read_lock_nesting, shared state will be updated
130 * if we block.
131 */
132void __rcu_read_lock(void)
133{
134 ACCESS_ONCE(current->rcu_read_lock_nesting)++;
135 barrier(); /* needed if we ever invoke rcu_read_lock in rcutree.c */
136}
137EXPORT_SYMBOL_GPL(__rcu_read_lock);
138
139static void rcu_read_unlock_special(struct task_struct *t)
140{
141 int empty;
142 unsigned long flags;
143 unsigned long mask;
144 struct rcu_node *rnp;
145 int special;
146
147 /* NMI handlers cannot block and cannot safely manipulate state. */
148 if (in_nmi())
149 return;
150
151 local_irq_save(flags);
152
153 /*
154 * If RCU core is waiting for this CPU to exit critical section,
155 * let it know that we have done so.
156 */
157 special = t->rcu_read_unlock_special;
158 if (special & RCU_READ_UNLOCK_NEED_QS) {
159 t->rcu_read_unlock_special &= ~RCU_READ_UNLOCK_NEED_QS;
160 t->rcu_read_unlock_special |= RCU_READ_UNLOCK_GOT_QS;
161 }
162
163 /* Hardware IRQ handlers cannot block. */
164 if (in_irq()) {
165 local_irq_restore(flags);
166 return;
167 }
168
169 /* Clean up if blocked during RCU read-side critical section. */
170 if (special & RCU_READ_UNLOCK_BLOCKED) {
171 t->rcu_read_unlock_special &= ~RCU_READ_UNLOCK_BLOCKED;
172
173 /*
174 * Remove this task from the list it blocked on. The
175 * task can migrate while we acquire the lock, but at
176 * most one time. So at most two passes through loop.
177 */
178 for (;;) {
179 rnp = t->rcu_blocked_node;
180 spin_lock(&rnp->lock);
181 if (rnp == t->rcu_blocked_node)
182 break;
183 spin_unlock(&rnp->lock);
184 }
185 empty = list_empty(&rnp->blocked_tasks[rnp->gpnum & 0x1]);
186 list_del_init(&t->rcu_node_entry);
187 t->rcu_blocked_node = NULL;
188
189 /*
190 * If this was the last task on the current list, and if
191 * we aren't waiting on any CPUs, report the quiescent state.
192 * Note that both cpu_quiet_msk_finish() and cpu_quiet_msk()
193 * drop rnp->lock and restore irq.
194 */
195 if (!empty && rnp->qsmask == 0 &&
196 list_empty(&rnp->blocked_tasks[rnp->gpnum & 0x1])) {
197 t->rcu_read_unlock_special &=
198 ~(RCU_READ_UNLOCK_NEED_QS |
199 RCU_READ_UNLOCK_GOT_QS);
200 if (rnp->parent == NULL) {
201 /* Only one rcu_node in the tree. */
202 cpu_quiet_msk_finish(&rcu_preempt_state, flags);
203 return;
204 }
205 /* Report up the rest of the hierarchy. */
206 mask = rnp->grpmask;
207 spin_unlock_irqrestore(&rnp->lock, flags);
208 rnp = rnp->parent;
209 spin_lock_irqsave(&rnp->lock, flags);
210 cpu_quiet_msk(mask, &rcu_preempt_state, rnp, flags);
211 return;
212 }
213 spin_unlock(&rnp->lock);
214 }
215 local_irq_restore(flags);
216}
217
218/*
219 * Tree-preemptable RCU implementation for rcu_read_unlock().
220 * Decrement ->rcu_read_lock_nesting. If the result is zero (outermost
221 * rcu_read_unlock()) and ->rcu_read_unlock_special is non-zero, then
222 * invoke rcu_read_unlock_special() to clean up after a context switch
223 * in an RCU read-side critical section and other special cases.
224 */
225void __rcu_read_unlock(void)
226{
227 struct task_struct *t = current;
228
229 barrier(); /* needed if we ever invoke rcu_read_unlock in rcutree.c */
230 if (--ACCESS_ONCE(t->rcu_read_lock_nesting) == 0 &&
231 unlikely(ACCESS_ONCE(t->rcu_read_unlock_special)))
232 rcu_read_unlock_special(t);
233}
234EXPORT_SYMBOL_GPL(__rcu_read_unlock);
235
236#ifdef CONFIG_RCU_CPU_STALL_DETECTOR
237
238/*
239 * Scan the current list of tasks blocked within RCU read-side critical
240 * sections, printing out the tid of each.
241 */
242static void rcu_print_task_stall(struct rcu_node *rnp)
243{
244 unsigned long flags;
245 struct list_head *lp;
246 int phase = rnp->gpnum & 0x1;
247 struct task_struct *t;
248
249 if (!list_empty(&rnp->blocked_tasks[phase])) {
250 spin_lock_irqsave(&rnp->lock, flags);
251 phase = rnp->gpnum & 0x1; /* re-read under lock. */
252 lp = &rnp->blocked_tasks[phase];
253 list_for_each_entry(t, lp, rcu_node_entry)
254 printk(" P%d", t->pid);
255 spin_unlock_irqrestore(&rnp->lock, flags);
256 }
257}
258
259#endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
260
261/*
262 * Check for preempted RCU readers for the specified rcu_node structure.
263 * If the caller needs a reliable answer, it must hold the rcu_node's
264 * >lock.
265 */
266static int rcu_preempted_readers(struct rcu_node *rnp)
267{
268 return !list_empty(&rnp->blocked_tasks[rnp->gpnum & 0x1]);
269}
270
271#ifdef CONFIG_HOTPLUG_CPU
272
273/*
274 * Handle tasklist migration for case in which all CPUs covered by the
275 * specified rcu_node have gone offline. Move them up to the root
276 * rcu_node. The reason for not just moving them to the immediate
277 * parent is to remove the need for rcu_read_unlock_special() to
278 * make more than two attempts to acquire the target rcu_node's lock.
279 *
280 * The caller must hold rnp->lock with irqs disabled.
281 */
282static void rcu_preempt_offline_tasks(struct rcu_state *rsp,
283 struct rcu_node *rnp)
284{
285 int i;
286 struct list_head *lp;
287 struct list_head *lp_root;
288 struct rcu_node *rnp_root = rcu_get_root(rsp);
289 struct task_struct *tp;
290
291 if (rnp == rnp_root) {
292 WARN_ONCE(1, "Last CPU thought to be offlined?");
293 return; /* Shouldn't happen: at least one CPU online. */
294 }
295
296 /*
297 * Move tasks up to root rcu_node. Rely on the fact that the
298 * root rcu_node can be at most one ahead of the rest of the
299 * rcu_nodes in terms of gp_num value. This fact allows us to
300 * move the blocked_tasks[] array directly, element by element.
301 */
302 for (i = 0; i < 2; i++) {
303 lp = &rnp->blocked_tasks[i];
304 lp_root = &rnp_root->blocked_tasks[i];
305 while (!list_empty(lp)) {
306 tp = list_entry(lp->next, typeof(*tp), rcu_node_entry);
307 spin_lock(&rnp_root->lock); /* irqs already disabled */
308 list_del(&tp->rcu_node_entry);
309 tp->rcu_blocked_node = rnp_root;
310 list_add(&tp->rcu_node_entry, lp_root);
311 spin_unlock(&rnp_root->lock); /* irqs remain disabled */
312 }
313 }
314}
315
316/*
317 * Do CPU-offline processing for preemptable RCU.
318 */
319static void rcu_preempt_offline_cpu(int cpu)
320{
321 __rcu_offline_cpu(cpu, &rcu_preempt_state);
322}
323
324#endif /* #ifdef CONFIG_HOTPLUG_CPU */
325
326/*
327 * Check for a quiescent state from the current CPU. When a task blocks,
328 * the task is recorded in the corresponding CPU's rcu_node structure,
329 * which is checked elsewhere.
330 *
331 * Caller must disable hard irqs.
332 */
333static void rcu_preempt_check_callbacks(int cpu)
334{
335 struct task_struct *t = current;
336
337 if (t->rcu_read_lock_nesting == 0) {
338 t->rcu_read_unlock_special &=
339 ~(RCU_READ_UNLOCK_NEED_QS | RCU_READ_UNLOCK_GOT_QS);
340 rcu_preempt_qs_record(cpu);
341 return;
342 }
343 if (per_cpu(rcu_preempt_data, cpu).qs_pending) {
344 if (t->rcu_read_unlock_special & RCU_READ_UNLOCK_GOT_QS) {
345 rcu_preempt_qs_record(cpu);
346 t->rcu_read_unlock_special &= ~RCU_READ_UNLOCK_GOT_QS;
347 } else if (!(t->rcu_read_unlock_special &
348 RCU_READ_UNLOCK_NEED_QS)) {
349 t->rcu_read_unlock_special |= RCU_READ_UNLOCK_NEED_QS;
350 }
351 }
352}
353
354/*
355 * Process callbacks for preemptable RCU.
356 */
357static void rcu_preempt_process_callbacks(void)
358{
359 __rcu_process_callbacks(&rcu_preempt_state,
360 &__get_cpu_var(rcu_preempt_data));
361}
362
363/*
364 * Queue a preemptable-RCU callback for invocation after a grace period.
365 */
366void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
367{
368 __call_rcu(head, func, &rcu_preempt_state);
369}
370EXPORT_SYMBOL_GPL(call_rcu);
371
372/*
373 * Check to see if there is any immediate preemptable-RCU-related work
374 * to be done.
375 */
376static int rcu_preempt_pending(int cpu)
377{
378 return __rcu_pending(&rcu_preempt_state,
379 &per_cpu(rcu_preempt_data, cpu));
380}
381
382/*
383 * Does preemptable RCU need the CPU to stay out of dynticks mode?
384 */
385static int rcu_preempt_needs_cpu(int cpu)
386{
387 return !!per_cpu(rcu_preempt_data, cpu).nxtlist;
388}
389
390/*
391 * Initialize preemptable RCU's per-CPU data.
392 */
393static void __cpuinit rcu_preempt_init_percpu_data(int cpu)
394{
395 rcu_init_percpu_data(cpu, &rcu_preempt_state, 1);
396}
397
398/*
399 * Check for a task exiting while in a preemptable-RCU read-side
400 * critical section, clean up if so. No need to issue warnings,
401 * as debug_check_no_locks_held() already does this if lockdep
402 * is enabled.
403 */
404void exit_rcu(void)
405{
406 struct task_struct *t = current;
407
408 if (t->rcu_read_lock_nesting == 0)
409 return;
410 t->rcu_read_lock_nesting = 1;
411 rcu_read_unlock();
412}
413
414#else /* #ifdef CONFIG_TREE_PREEMPT_RCU */
415
416/*
417 * Tell them what RCU they are running.
418 */
419static inline void rcu_bootup_announce(void)
420{
421 printk(KERN_INFO "Hierarchical RCU implementation.\n");
422}
423
424/*
425 * Return the number of RCU batches processed thus far for debug & stats.
426 */
427long rcu_batches_completed(void)
428{
429 return rcu_batches_completed_sched();
430}
431EXPORT_SYMBOL_GPL(rcu_batches_completed);
432
433/*
434 * Because preemptable RCU does not exist, we never have to check for
435 * CPUs being in quiescent states.
436 */
437static void rcu_preempt_qs(int cpu)
438{
439}
440
441#ifdef CONFIG_RCU_CPU_STALL_DETECTOR
442
443/*
444 * Because preemptable RCU does not exist, we never have to check for
445 * tasks blocked within RCU read-side critical sections.
446 */
447static void rcu_print_task_stall(struct rcu_node *rnp)
448{
449}
450
451#endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
452
453/*
454 * Because preemptable RCU does not exist, there are never any preempted
455 * RCU readers.
456 */
457static int rcu_preempted_readers(struct rcu_node *rnp)
458{
459 return 0;
460}
461
462#ifdef CONFIG_HOTPLUG_CPU
463
464/*
465 * Because preemptable RCU does not exist, it never needs to migrate
466 * tasks that were blocked within RCU read-side critical sections.
467 */
468static void rcu_preempt_offline_tasks(struct rcu_state *rsp,
469 struct rcu_node *rnp)
470{
471}
472
473/*
474 * Because preemptable RCU does not exist, it never needs CPU-offline
475 * processing.
476 */
477static void rcu_preempt_offline_cpu(int cpu)
478{
479}
480
481#endif /* #ifdef CONFIG_HOTPLUG_CPU */
482
483/*
484 * Because preemptable RCU does not exist, it never has any callbacks
485 * to check.
486 */
487void rcu_preempt_check_callbacks(int cpu)
488{
489}
490
491/*
492 * Because preemptable RCU does not exist, it never has any callbacks
493 * to process.
494 */
495void rcu_preempt_process_callbacks(void)
496{
497}
498
499/*
500 * In classic RCU, call_rcu() is just call_rcu_sched().
501 */
502void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
503{
504 call_rcu_sched(head, func);
505}
506EXPORT_SYMBOL_GPL(call_rcu);
507
508/*
509 * Because preemptable RCU does not exist, it never has any work to do.
510 */
511static int rcu_preempt_pending(int cpu)
512{
513 return 0;
514}
515
516/*
517 * Because preemptable RCU does not exist, it never needs any CPU.
518 */
519static int rcu_preempt_needs_cpu(int cpu)
520{
521 return 0;
522}
523
524/*
525 * Because preemptable RCU does not exist, there is no per-CPU
526 * data to initialize.
527 */
528static void __cpuinit rcu_preempt_init_percpu_data(int cpu)
529{
530}
531
532#endif /* #else #ifdef CONFIG_TREE_PREEMPT_RCU */
diff --git a/kernel/rcutree_trace.c b/kernel/rcutree_trace.c
index fe1dcdbf1ca3..0ea1bff69727 100644
--- a/kernel/rcutree_trace.c
+++ b/kernel/rcutree_trace.c
@@ -43,6 +43,7 @@
43#include <linux/debugfs.h> 43#include <linux/debugfs.h>
44#include <linux/seq_file.h> 44#include <linux/seq_file.h>
45 45
46#define RCU_TREE_NONCORE
46#include "rcutree.h" 47#include "rcutree.h"
47 48
48static void print_one_rcu_data(struct seq_file *m, struct rcu_data *rdp) 49static void print_one_rcu_data(struct seq_file *m, struct rcu_data *rdp)
@@ -76,8 +77,12 @@ static void print_one_rcu_data(struct seq_file *m, struct rcu_data *rdp)
76 77
77static int show_rcudata(struct seq_file *m, void *unused) 78static int show_rcudata(struct seq_file *m, void *unused)
78{ 79{
79 seq_puts(m, "rcu:\n"); 80#ifdef CONFIG_TREE_PREEMPT_RCU
80 PRINT_RCU_DATA(rcu_data, print_one_rcu_data, m); 81 seq_puts(m, "rcu_preempt:\n");
82 PRINT_RCU_DATA(rcu_preempt_data, print_one_rcu_data, m);
83#endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */
84 seq_puts(m, "rcu_sched:\n");
85 PRINT_RCU_DATA(rcu_sched_data, print_one_rcu_data, m);
81 seq_puts(m, "rcu_bh:\n"); 86 seq_puts(m, "rcu_bh:\n");
82 PRINT_RCU_DATA(rcu_bh_data, print_one_rcu_data, m); 87 PRINT_RCU_DATA(rcu_bh_data, print_one_rcu_data, m);
83 return 0; 88 return 0;
@@ -102,7 +107,7 @@ static void print_one_rcu_data_csv(struct seq_file *m, struct rcu_data *rdp)
102 return; 107 return;
103 seq_printf(m, "%d,%s,%ld,%ld,%d,%ld,%d", 108 seq_printf(m, "%d,%s,%ld,%ld,%d,%ld,%d",
104 rdp->cpu, 109 rdp->cpu,
105 cpu_is_offline(rdp->cpu) ? "\"Y\"" : "\"N\"", 110 cpu_is_offline(rdp->cpu) ? "\"N\"" : "\"Y\"",
106 rdp->completed, rdp->gpnum, 111 rdp->completed, rdp->gpnum,
107 rdp->passed_quiesc, rdp->passed_quiesc_completed, 112 rdp->passed_quiesc, rdp->passed_quiesc_completed,
108 rdp->qs_pending); 113 rdp->qs_pending);
@@ -124,8 +129,12 @@ static int show_rcudata_csv(struct seq_file *m, void *unused)
124 seq_puts(m, "\"dt\",\"dt nesting\",\"dn\",\"df\","); 129 seq_puts(m, "\"dt\",\"dt nesting\",\"dn\",\"df\",");
125#endif /* #ifdef CONFIG_NO_HZ */ 130#endif /* #ifdef CONFIG_NO_HZ */
126 seq_puts(m, "\"of\",\"ri\",\"ql\",\"b\"\n"); 131 seq_puts(m, "\"of\",\"ri\",\"ql\",\"b\"\n");
127 seq_puts(m, "\"rcu:\"\n"); 132#ifdef CONFIG_TREE_PREEMPT_RCU
128 PRINT_RCU_DATA(rcu_data, print_one_rcu_data_csv, m); 133 seq_puts(m, "\"rcu_preempt:\"\n");
134 PRINT_RCU_DATA(rcu_preempt_data, print_one_rcu_data_csv, m);
135#endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */
136 seq_puts(m, "\"rcu_sched:\"\n");
137 PRINT_RCU_DATA(rcu_sched_data, print_one_rcu_data_csv, m);
129 seq_puts(m, "\"rcu_bh:\"\n"); 138 seq_puts(m, "\"rcu_bh:\"\n");
130 PRINT_RCU_DATA(rcu_bh_data, print_one_rcu_data_csv, m); 139 PRINT_RCU_DATA(rcu_bh_data, print_one_rcu_data_csv, m);
131 return 0; 140 return 0;
@@ -171,8 +180,12 @@ static void print_one_rcu_state(struct seq_file *m, struct rcu_state *rsp)
171 180
172static int show_rcuhier(struct seq_file *m, void *unused) 181static int show_rcuhier(struct seq_file *m, void *unused)
173{ 182{
174 seq_puts(m, "rcu:\n"); 183#ifdef CONFIG_TREE_PREEMPT_RCU
175 print_one_rcu_state(m, &rcu_state); 184 seq_puts(m, "rcu_preempt:\n");
185 print_one_rcu_state(m, &rcu_preempt_state);
186#endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */
187 seq_puts(m, "rcu_sched:\n");
188 print_one_rcu_state(m, &rcu_sched_state);
176 seq_puts(m, "rcu_bh:\n"); 189 seq_puts(m, "rcu_bh:\n");
177 print_one_rcu_state(m, &rcu_bh_state); 190 print_one_rcu_state(m, &rcu_bh_state);
178 return 0; 191 return 0;
@@ -193,8 +206,12 @@ static struct file_operations rcuhier_fops = {
193 206
194static int show_rcugp(struct seq_file *m, void *unused) 207static int show_rcugp(struct seq_file *m, void *unused)
195{ 208{
196 seq_printf(m, "rcu: completed=%ld gpnum=%ld\n", 209#ifdef CONFIG_TREE_PREEMPT_RCU
197 rcu_state.completed, rcu_state.gpnum); 210 seq_printf(m, "rcu_preempt: completed=%ld gpnum=%ld\n",
211 rcu_preempt_state.completed, rcu_preempt_state.gpnum);
212#endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */
213 seq_printf(m, "rcu_sched: completed=%ld gpnum=%ld\n",
214 rcu_sched_state.completed, rcu_sched_state.gpnum);
198 seq_printf(m, "rcu_bh: completed=%ld gpnum=%ld\n", 215 seq_printf(m, "rcu_bh: completed=%ld gpnum=%ld\n",
199 rcu_bh_state.completed, rcu_bh_state.gpnum); 216 rcu_bh_state.completed, rcu_bh_state.gpnum);
200 return 0; 217 return 0;
@@ -243,8 +260,12 @@ static void print_rcu_pendings(struct seq_file *m, struct rcu_state *rsp)
243 260
244static int show_rcu_pending(struct seq_file *m, void *unused) 261static int show_rcu_pending(struct seq_file *m, void *unused)
245{ 262{
246 seq_puts(m, "rcu:\n"); 263#ifdef CONFIG_TREE_PREEMPT_RCU
247 print_rcu_pendings(m, &rcu_state); 264 seq_puts(m, "rcu_preempt:\n");
265 print_rcu_pendings(m, &rcu_preempt_state);
266#endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */
267 seq_puts(m, "rcu_sched:\n");
268 print_rcu_pendings(m, &rcu_sched_state);
248 seq_puts(m, "rcu_bh:\n"); 269 seq_puts(m, "rcu_bh:\n");
249 print_rcu_pendings(m, &rcu_bh_state); 270 print_rcu_pendings(m, &rcu_bh_state);
250 return 0; 271 return 0;
@@ -264,62 +285,47 @@ static struct file_operations rcu_pending_fops = {
264}; 285};
265 286
266static struct dentry *rcudir; 287static struct dentry *rcudir;
267static struct dentry *datadir;
268static struct dentry *datadir_csv;
269static struct dentry *gpdir;
270static struct dentry *hierdir;
271static struct dentry *rcu_pendingdir;
272 288
273static int __init rcuclassic_trace_init(void) 289static int __init rcuclassic_trace_init(void)
274{ 290{
291 struct dentry *retval;
292
275 rcudir = debugfs_create_dir("rcu", NULL); 293 rcudir = debugfs_create_dir("rcu", NULL);
276 if (!rcudir) 294 if (!rcudir)
277 goto out; 295 goto free_out;
278 296
279 datadir = debugfs_create_file("rcudata", 0444, rcudir, 297 retval = debugfs_create_file("rcudata", 0444, rcudir,
280 NULL, &rcudata_fops); 298 NULL, &rcudata_fops);
281 if (!datadir) 299 if (!retval)
282 goto free_out; 300 goto free_out;
283 301
284 datadir_csv = debugfs_create_file("rcudata.csv", 0444, rcudir, 302 retval = debugfs_create_file("rcudata.csv", 0444, rcudir,
285 NULL, &rcudata_csv_fops); 303 NULL, &rcudata_csv_fops);
286 if (!datadir_csv) 304 if (!retval)
287 goto free_out; 305 goto free_out;
288 306
289 gpdir = debugfs_create_file("rcugp", 0444, rcudir, NULL, &rcugp_fops); 307 retval = debugfs_create_file("rcugp", 0444, rcudir, NULL, &rcugp_fops);
290 if (!gpdir) 308 if (!retval)
291 goto free_out; 309 goto free_out;
292 310
293 hierdir = debugfs_create_file("rcuhier", 0444, rcudir, 311 retval = debugfs_create_file("rcuhier", 0444, rcudir,
294 NULL, &rcuhier_fops); 312 NULL, &rcuhier_fops);
295 if (!hierdir) 313 if (!retval)
296 goto free_out; 314 goto free_out;
297 315
298 rcu_pendingdir = debugfs_create_file("rcu_pending", 0444, rcudir, 316 retval = debugfs_create_file("rcu_pending", 0444, rcudir,
299 NULL, &rcu_pending_fops); 317 NULL, &rcu_pending_fops);
300 if (!rcu_pendingdir) 318 if (!retval)
301 goto free_out; 319 goto free_out;
302 return 0; 320 return 0;
303free_out: 321free_out:
304 if (datadir) 322 debugfs_remove_recursive(rcudir);
305 debugfs_remove(datadir);
306 if (datadir_csv)
307 debugfs_remove(datadir_csv);
308 if (gpdir)
309 debugfs_remove(gpdir);
310 debugfs_remove(rcudir);
311out:
312 return 1; 323 return 1;
313} 324}
314 325
315static void __exit rcuclassic_trace_cleanup(void) 326static void __exit rcuclassic_trace_cleanup(void)
316{ 327{
317 debugfs_remove(datadir); 328 debugfs_remove_recursive(rcudir);
318 debugfs_remove(datadir_csv);
319 debugfs_remove(gpdir);
320 debugfs_remove(hierdir);
321 debugfs_remove(rcu_pendingdir);
322 debugfs_remove(rcudir);
323} 329}
324 330
325 331
diff --git a/kernel/sched.c b/kernel/sched.c
index 2c75f7daa439..4066241ae9f4 100644
--- a/kernel/sched.c
+++ b/kernel/sched.c
@@ -5325,7 +5325,7 @@ need_resched:
5325 preempt_disable(); 5325 preempt_disable();
5326 cpu = smp_processor_id(); 5326 cpu = smp_processor_id();
5327 rq = cpu_rq(cpu); 5327 rq = cpu_rq(cpu);
5328 rcu_qsctr_inc(cpu); 5328 rcu_sched_qs(cpu);
5329 prev = rq->curr; 5329 prev = rq->curr;
5330 switch_count = &prev->nivcsw; 5330 switch_count = &prev->nivcsw;
5331 5331
@@ -7053,6 +7053,11 @@ fail:
7053 return ret; 7053 return ret;
7054} 7054}
7055 7055
7056#define RCU_MIGRATION_IDLE 0
7057#define RCU_MIGRATION_NEED_QS 1
7058#define RCU_MIGRATION_GOT_QS 2
7059#define RCU_MIGRATION_MUST_SYNC 3
7060
7056/* 7061/*
7057 * migration_thread - this is a highprio system thread that performs 7062 * migration_thread - this is a highprio system thread that performs
7058 * thread migration by bumping thread off CPU then 'pushing' onto 7063 * thread migration by bumping thread off CPU then 'pushing' onto
@@ -7060,6 +7065,7 @@ fail:
7060 */ 7065 */
7061static int migration_thread(void *data) 7066static int migration_thread(void *data)
7062{ 7067{
7068 int badcpu;
7063 int cpu = (long)data; 7069 int cpu = (long)data;
7064 struct rq *rq; 7070 struct rq *rq;
7065 7071
@@ -7094,8 +7100,17 @@ static int migration_thread(void *data)
7094 req = list_entry(head->next, struct migration_req, list); 7100 req = list_entry(head->next, struct migration_req, list);
7095 list_del_init(head->next); 7101 list_del_init(head->next);
7096 7102
7097 spin_unlock(&rq->lock); 7103 if (req->task != NULL) {
7098 __migrate_task(req->task, cpu, req->dest_cpu); 7104 spin_unlock(&rq->lock);
7105 __migrate_task(req->task, cpu, req->dest_cpu);
7106 } else if (likely(cpu == (badcpu = smp_processor_id()))) {
7107 req->dest_cpu = RCU_MIGRATION_GOT_QS;
7108 spin_unlock(&rq->lock);
7109 } else {
7110 req->dest_cpu = RCU_MIGRATION_MUST_SYNC;
7111 spin_unlock(&rq->lock);
7112 WARN_ONCE(1, "migration_thread() on CPU %d, expected %d\n", badcpu, cpu);
7113 }
7099 local_irq_enable(); 7114 local_irq_enable();
7100 7115
7101 complete(&req->done); 7116 complete(&req->done);
@@ -10583,3 +10598,113 @@ struct cgroup_subsys cpuacct_subsys = {
10583 .subsys_id = cpuacct_subsys_id, 10598 .subsys_id = cpuacct_subsys_id,
10584}; 10599};
10585#endif /* CONFIG_CGROUP_CPUACCT */ 10600#endif /* CONFIG_CGROUP_CPUACCT */
10601
10602#ifndef CONFIG_SMP
10603
10604int rcu_expedited_torture_stats(char *page)
10605{
10606 return 0;
10607}
10608EXPORT_SYMBOL_GPL(rcu_expedited_torture_stats);
10609
10610void synchronize_sched_expedited(void)
10611{
10612}
10613EXPORT_SYMBOL_GPL(synchronize_sched_expedited);
10614
10615#else /* #ifndef CONFIG_SMP */
10616
10617static DEFINE_PER_CPU(struct migration_req, rcu_migration_req);
10618static DEFINE_MUTEX(rcu_sched_expedited_mutex);
10619
10620#define RCU_EXPEDITED_STATE_POST -2
10621#define RCU_EXPEDITED_STATE_IDLE -1
10622
10623static int rcu_expedited_state = RCU_EXPEDITED_STATE_IDLE;
10624
10625int rcu_expedited_torture_stats(char *page)
10626{
10627 int cnt = 0;
10628 int cpu;
10629
10630 cnt += sprintf(&page[cnt], "state: %d /", rcu_expedited_state);
10631 for_each_online_cpu(cpu) {
10632 cnt += sprintf(&page[cnt], " %d:%d",
10633 cpu, per_cpu(rcu_migration_req, cpu).dest_cpu);
10634 }
10635 cnt += sprintf(&page[cnt], "\n");
10636 return cnt;
10637}
10638EXPORT_SYMBOL_GPL(rcu_expedited_torture_stats);
10639
10640static long synchronize_sched_expedited_count;
10641
10642/*
10643 * Wait for an rcu-sched grace period to elapse, but use "big hammer"
10644 * approach to force grace period to end quickly. This consumes
10645 * significant time on all CPUs, and is thus not recommended for
10646 * any sort of common-case code.
10647 *
10648 * Note that it is illegal to call this function while holding any
10649 * lock that is acquired by a CPU-hotplug notifier. Failing to
10650 * observe this restriction will result in deadlock.
10651 */
10652void synchronize_sched_expedited(void)
10653{
10654 int cpu;
10655 unsigned long flags;
10656 bool need_full_sync = 0;
10657 struct rq *rq;
10658 struct migration_req *req;
10659 long snap;
10660 int trycount = 0;
10661
10662 smp_mb(); /* ensure prior mod happens before capturing snap. */
10663 snap = ACCESS_ONCE(synchronize_sched_expedited_count) + 1;
10664 get_online_cpus();
10665 while (!mutex_trylock(&rcu_sched_expedited_mutex)) {
10666 put_online_cpus();
10667 if (trycount++ < 10)
10668 udelay(trycount * num_online_cpus());
10669 else {
10670 synchronize_sched();
10671 return;
10672 }
10673 if (ACCESS_ONCE(synchronize_sched_expedited_count) - snap > 0) {
10674 smp_mb(); /* ensure test happens before caller kfree */
10675 return;
10676 }
10677 get_online_cpus();
10678 }
10679 rcu_expedited_state = RCU_EXPEDITED_STATE_POST;
10680 for_each_online_cpu(cpu) {
10681 rq = cpu_rq(cpu);
10682 req = &per_cpu(rcu_migration_req, cpu);
10683 init_completion(&req->done);
10684 req->task = NULL;
10685 req->dest_cpu = RCU_MIGRATION_NEED_QS;
10686 spin_lock_irqsave(&rq->lock, flags);
10687 list_add(&req->list, &rq->migration_queue);
10688 spin_unlock_irqrestore(&rq->lock, flags);
10689 wake_up_process(rq->migration_thread);
10690 }
10691 for_each_online_cpu(cpu) {
10692 rcu_expedited_state = cpu;
10693 req = &per_cpu(rcu_migration_req, cpu);
10694 rq = cpu_rq(cpu);
10695 wait_for_completion(&req->done);
10696 spin_lock_irqsave(&rq->lock, flags);
10697 if (unlikely(req->dest_cpu == RCU_MIGRATION_MUST_SYNC))
10698 need_full_sync = 1;
10699 req->dest_cpu = RCU_MIGRATION_IDLE;
10700 spin_unlock_irqrestore(&rq->lock, flags);
10701 }
10702 rcu_expedited_state = RCU_EXPEDITED_STATE_IDLE;
10703 mutex_unlock(&rcu_sched_expedited_mutex);
10704 put_online_cpus();
10705 if (need_full_sync)
10706 synchronize_sched();
10707}
10708EXPORT_SYMBOL_GPL(synchronize_sched_expedited);
10709
10710#endif /* #else #ifndef CONFIG_SMP */
diff --git a/kernel/softirq.c b/kernel/softirq.c
index eb5e131a0485..7db25067cd2d 100644
--- a/kernel/softirq.c
+++ b/kernel/softirq.c
@@ -227,7 +227,7 @@ restart:
227 preempt_count() = prev_count; 227 preempt_count() = prev_count;
228 } 228 }
229 229
230 rcu_bh_qsctr_inc(cpu); 230 rcu_bh_qs(cpu);
231 } 231 }
232 h++; 232 h++;
233 pending >>= 1; 233 pending >>= 1;
@@ -721,7 +721,7 @@ static int ksoftirqd(void * __bind_cpu)
721 preempt_enable_no_resched(); 721 preempt_enable_no_resched();
722 cond_resched(); 722 cond_resched();
723 preempt_disable(); 723 preempt_disable();
724 rcu_qsctr_inc((long)__bind_cpu); 724 rcu_sched_qs((long)__bind_cpu);
725 } 725 }
726 preempt_enable(); 726 preempt_enable();
727 set_current_state(TASK_INTERRUPTIBLE); 727 set_current_state(TASK_INTERRUPTIBLE);
diff --git a/kernel/timer.c b/kernel/timer.c
index a7f07d5a6241..a3d25f415019 100644
--- a/kernel/timer.c
+++ b/kernel/timer.c
@@ -1156,8 +1156,7 @@ void update_process_times(int user_tick)
1156 /* Note: this timer irq context must be accounted for as well. */ 1156 /* Note: this timer irq context must be accounted for as well. */
1157 account_process_tick(p, user_tick); 1157 account_process_tick(p, user_tick);
1158 run_local_timers(); 1158 run_local_timers();
1159 if (rcu_pending(cpu)) 1159 rcu_check_callbacks(cpu, user_tick);
1160 rcu_check_callbacks(cpu, user_tick);
1161 printk_tick(); 1160 printk_tick();
1162 scheduler_tick(); 1161 scheduler_tick();
1163 run_posix_cpu_timers(p); 1162 run_posix_cpu_timers(p);
generated by cgit v1.2.2 (git 2.25.0) at 2024-11-20 10:18:03 -0500