aboutsummaryrefslogtreecommitdiffstats
path: root/kernel
diff options
context:
space:
mode:
Diffstat (limited to 'kernel')
-rw-r--r--kernel/exit.c2
-rw-r--r--kernel/fork.c10
-rw-r--r--kernel/futex.c3
-rw-r--r--kernel/panic.c3
-rw-r--r--kernel/rcupdate.c140
-rw-r--r--kernel/rcutorture.c4
-rw-r--r--kernel/rcutree.c330
-rw-r--r--kernel/rcutree.h86
-rw-r--r--kernel/rcutree_plugin.h103
-rw-r--r--kernel/rcutree_trace.c4
10 files changed, 386 insertions, 299 deletions
diff --git a/kernel/exit.c b/kernel/exit.c
index 5859f598c951..e61891f80123 100644
--- a/kernel/exit.c
+++ b/kernel/exit.c
@@ -991,8 +991,6 @@ NORET_TYPE void do_exit(long code)
991 tsk->mempolicy = NULL; 991 tsk->mempolicy = NULL;
992#endif 992#endif
993#ifdef CONFIG_FUTEX 993#ifdef CONFIG_FUTEX
994 if (unlikely(!list_empty(&tsk->pi_state_list)))
995 exit_pi_state_list(tsk);
996 if (unlikely(current->pi_state_cache)) 994 if (unlikely(current->pi_state_cache))
997 kfree(current->pi_state_cache); 995 kfree(current->pi_state_cache);
998#endif 996#endif
diff --git a/kernel/fork.c b/kernel/fork.c
index 266c6af6ef1b..4c20fff8c13a 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -570,12 +570,18 @@ void mm_release(struct task_struct *tsk, struct mm_struct *mm)
570 570
571 /* Get rid of any futexes when releasing the mm */ 571 /* Get rid of any futexes when releasing the mm */
572#ifdef CONFIG_FUTEX 572#ifdef CONFIG_FUTEX
573 if (unlikely(tsk->robust_list)) 573 if (unlikely(tsk->robust_list)) {
574 exit_robust_list(tsk); 574 exit_robust_list(tsk);
575 tsk->robust_list = NULL;
576 }
575#ifdef CONFIG_COMPAT 577#ifdef CONFIG_COMPAT
576 if (unlikely(tsk->compat_robust_list)) 578 if (unlikely(tsk->compat_robust_list)) {
577 compat_exit_robust_list(tsk); 579 compat_exit_robust_list(tsk);
580 tsk->compat_robust_list = NULL;
581 }
578#endif 582#endif
583 if (unlikely(!list_empty(&tsk->pi_state_list)))
584 exit_pi_state_list(tsk);
579#endif 585#endif
580 586
581 /* Get rid of any cached register state */ 587 /* Get rid of any cached register state */
diff --git a/kernel/futex.c b/kernel/futex.c
index b911adceb2c4..4949d336d88d 100644
--- a/kernel/futex.c
+++ b/kernel/futex.c
@@ -916,8 +916,8 @@ retry:
916 hb1 = hash_futex(&key1); 916 hb1 = hash_futex(&key1);
917 hb2 = hash_futex(&key2); 917 hb2 = hash_futex(&key2);
918 918
919 double_lock_hb(hb1, hb2);
920retry_private: 919retry_private:
920 double_lock_hb(hb1, hb2);
921 op_ret = futex_atomic_op_inuser(op, uaddr2); 921 op_ret = futex_atomic_op_inuser(op, uaddr2);
922 if (unlikely(op_ret < 0)) { 922 if (unlikely(op_ret < 0)) {
923 923
@@ -2117,7 +2117,6 @@ int handle_early_requeue_pi_wakeup(struct futex_hash_bucket *hb,
2117 * Unqueue the futex_q and determine which it was. 2117 * Unqueue the futex_q and determine which it was.
2118 */ 2118 */
2119 plist_del(&q->list, &q->list.plist); 2119 plist_del(&q->list, &q->list.plist);
2120 drop_futex_key_refs(&q->key);
2121 2120
2122 if (timeout && !timeout->task) 2121 if (timeout && !timeout->task)
2123 ret = -ETIMEDOUT; 2122 ret = -ETIMEDOUT;
diff --git a/kernel/panic.c b/kernel/panic.c
index bcdef26e3332..96b45d0b4ba5 100644
--- a/kernel/panic.c
+++ b/kernel/panic.c
@@ -90,6 +90,8 @@ NORET_TYPE void panic(const char * fmt, ...)
90 90
91 atomic_notifier_call_chain(&panic_notifier_list, 0, buf); 91 atomic_notifier_call_chain(&panic_notifier_list, 0, buf);
92 92
93 bust_spinlocks(0);
94
93 if (!panic_blink) 95 if (!panic_blink)
94 panic_blink = no_blink; 96 panic_blink = no_blink;
95 97
@@ -136,7 +138,6 @@ NORET_TYPE void panic(const char * fmt, ...)
136 mdelay(1); 138 mdelay(1);
137 i++; 139 i++;
138 } 140 }
139 bust_spinlocks(0);
140} 141}
141 142
142EXPORT_SYMBOL(panic); 143EXPORT_SYMBOL(panic);
diff --git a/kernel/rcupdate.c b/kernel/rcupdate.c
index 37ac45483082..400183346ad2 100644
--- a/kernel/rcupdate.c
+++ b/kernel/rcupdate.c
@@ -46,22 +46,15 @@
46#include <linux/module.h> 46#include <linux/module.h>
47#include <linux/kernel_stat.h> 47#include <linux/kernel_stat.h>
48 48
49enum rcu_barrier { 49#ifdef CONFIG_DEBUG_LOCK_ALLOC
50 RCU_BARRIER_STD, 50static struct lock_class_key rcu_lock_key;
51 RCU_BARRIER_BH, 51struct lockdep_map rcu_lock_map =
52 RCU_BARRIER_SCHED, 52 STATIC_LOCKDEP_MAP_INIT("rcu_read_lock", &rcu_lock_key);
53}; 53EXPORT_SYMBOL_GPL(rcu_lock_map);
54#endif
54 55
55static DEFINE_PER_CPU(struct rcu_head, rcu_barrier_head) = {NULL};
56static atomic_t rcu_barrier_cpu_count;
57static DEFINE_MUTEX(rcu_barrier_mutex);
58static struct completion rcu_barrier_completion;
59int rcu_scheduler_active __read_mostly; 56int rcu_scheduler_active __read_mostly;
60 57
61static atomic_t rcu_migrate_type_count = ATOMIC_INIT(0);
62static struct rcu_head rcu_migrate_head[3];
63static DECLARE_WAIT_QUEUE_HEAD(rcu_migrate_wq);
64
65/* 58/*
66 * Awaken the corresponding synchronize_rcu() instance now that a 59 * Awaken the corresponding synchronize_rcu() instance now that a
67 * grace period has elapsed. 60 * grace period has elapsed.
@@ -164,129 +157,10 @@ void synchronize_rcu_bh(void)
164} 157}
165EXPORT_SYMBOL_GPL(synchronize_rcu_bh); 158EXPORT_SYMBOL_GPL(synchronize_rcu_bh);
166 159
167static void rcu_barrier_callback(struct rcu_head *notused)
168{
169 if (atomic_dec_and_test(&rcu_barrier_cpu_count))
170 complete(&rcu_barrier_completion);
171}
172
173/*
174 * Called with preemption disabled, and from cross-cpu IRQ context.
175 */
176static void rcu_barrier_func(void *type)
177{
178 int cpu = smp_processor_id();
179 struct rcu_head *head = &per_cpu(rcu_barrier_head, cpu);
180
181 atomic_inc(&rcu_barrier_cpu_count);
182 switch ((enum rcu_barrier)type) {
183 case RCU_BARRIER_STD:
184 call_rcu(head, rcu_barrier_callback);
185 break;
186 case RCU_BARRIER_BH:
187 call_rcu_bh(head, rcu_barrier_callback);
188 break;
189 case RCU_BARRIER_SCHED:
190 call_rcu_sched(head, rcu_barrier_callback);
191 break;
192 }
193}
194
195static inline void wait_migrated_callbacks(void)
196{
197 wait_event(rcu_migrate_wq, !atomic_read(&rcu_migrate_type_count));
198 smp_mb(); /* In case we didn't sleep. */
199}
200
201/*
202 * Orchestrate the specified type of RCU barrier, waiting for all
203 * RCU callbacks of the specified type to complete.
204 */
205static void _rcu_barrier(enum rcu_barrier type)
206{
207 BUG_ON(in_interrupt());
208 /* Take cpucontrol mutex to protect against CPU hotplug */
209 mutex_lock(&rcu_barrier_mutex);
210 init_completion(&rcu_barrier_completion);
211 /*
212 * Initialize rcu_barrier_cpu_count to 1, then invoke
213 * rcu_barrier_func() on each CPU, so that each CPU also has
214 * incremented rcu_barrier_cpu_count. Only then is it safe to
215 * decrement rcu_barrier_cpu_count -- otherwise the first CPU
216 * might complete its grace period before all of the other CPUs
217 * did their increment, causing this function to return too
218 * early.
219 */
220 atomic_set(&rcu_barrier_cpu_count, 1);
221 on_each_cpu(rcu_barrier_func, (void *)type, 1);
222 if (atomic_dec_and_test(&rcu_barrier_cpu_count))
223 complete(&rcu_barrier_completion);
224 wait_for_completion(&rcu_barrier_completion);
225 mutex_unlock(&rcu_barrier_mutex);
226 wait_migrated_callbacks();
227}
228
229/**
230 * rcu_barrier - Wait until all in-flight call_rcu() callbacks complete.
231 */
232void rcu_barrier(void)
233{
234 _rcu_barrier(RCU_BARRIER_STD);
235}
236EXPORT_SYMBOL_GPL(rcu_barrier);
237
238/**
239 * rcu_barrier_bh - Wait until all in-flight call_rcu_bh() callbacks complete.
240 */
241void rcu_barrier_bh(void)
242{
243 _rcu_barrier(RCU_BARRIER_BH);
244}
245EXPORT_SYMBOL_GPL(rcu_barrier_bh);
246
247/**
248 * rcu_barrier_sched - Wait for in-flight call_rcu_sched() callbacks.
249 */
250void rcu_barrier_sched(void)
251{
252 _rcu_barrier(RCU_BARRIER_SCHED);
253}
254EXPORT_SYMBOL_GPL(rcu_barrier_sched);
255
256static void rcu_migrate_callback(struct rcu_head *notused)
257{
258 if (atomic_dec_and_test(&rcu_migrate_type_count))
259 wake_up(&rcu_migrate_wq);
260}
261
262extern int rcu_cpu_notify(struct notifier_block *self,
263 unsigned long action, void *hcpu);
264
265static int __cpuinit rcu_barrier_cpu_hotplug(struct notifier_block *self, 160static int __cpuinit rcu_barrier_cpu_hotplug(struct notifier_block *self,
266 unsigned long action, void *hcpu) 161 unsigned long action, void *hcpu)
267{ 162{
268 rcu_cpu_notify(self, action, hcpu); 163 return rcu_cpu_notify(self, action, hcpu);
269 if (action == CPU_DYING) {
270 /*
271 * preempt_disable() in on_each_cpu() prevents stop_machine(),
272 * so when "on_each_cpu(rcu_barrier_func, (void *)type, 1);"
273 * returns, all online cpus have queued rcu_barrier_func(),
274 * and the dead cpu(if it exist) queues rcu_migrate_callback()s.
275 *
276 * These callbacks ensure _rcu_barrier() waits for all
277 * RCU callbacks of the specified type to complete.
278 */
279 atomic_set(&rcu_migrate_type_count, 3);
280 call_rcu_bh(rcu_migrate_head, rcu_migrate_callback);
281 call_rcu_sched(rcu_migrate_head + 1, rcu_migrate_callback);
282 call_rcu(rcu_migrate_head + 2, rcu_migrate_callback);
283 } else if (action == CPU_DOWN_PREPARE) {
284 /* Don't need to wait until next removal operation. */
285 /* rcu_migrate_head is protected by cpu_add_remove_lock */
286 wait_migrated_callbacks();
287 }
288
289 return NOTIFY_OK;
290} 164}
291 165
292void __init rcu_init(void) 166void __init rcu_init(void)
diff --git a/kernel/rcutorture.c b/kernel/rcutorture.c
index 233768f21f97..697c0a0229d4 100644
--- a/kernel/rcutorture.c
+++ b/kernel/rcutorture.c
@@ -606,8 +606,6 @@ static struct rcu_torture_ops sched_ops_sync = {
606 .name = "sched_sync" 606 .name = "sched_sync"
607}; 607};
608 608
609extern int rcu_expedited_torture_stats(char *page);
610
611static struct rcu_torture_ops sched_expedited_ops = { 609static struct rcu_torture_ops sched_expedited_ops = {
612 .init = rcu_sync_torture_init, 610 .init = rcu_sync_torture_init,
613 .cleanup = NULL, 611 .cleanup = NULL,
@@ -650,7 +648,7 @@ rcu_torture_writer(void *arg)
650 old_rp = rcu_torture_current; 648 old_rp = rcu_torture_current;
651 rp->rtort_mbtest = 1; 649 rp->rtort_mbtest = 1;
652 rcu_assign_pointer(rcu_torture_current, rp); 650 rcu_assign_pointer(rcu_torture_current, rp);
653 smp_wmb(); 651 smp_wmb(); /* Mods to old_rp must follow rcu_assign_pointer() */
654 if (old_rp) { 652 if (old_rp) {
655 i = old_rp->rtort_pipe_count; 653 i = old_rp->rtort_pipe_count;
656 if (i > RCU_TORTURE_PIPE_LEN) 654 if (i > RCU_TORTURE_PIPE_LEN)
diff --git a/kernel/rcutree.c b/kernel/rcutree.c
index 52b06f6e158c..705f02ac7433 100644
--- a/kernel/rcutree.c
+++ b/kernel/rcutree.c
@@ -49,13 +49,6 @@
49 49
50#include "rcutree.h" 50#include "rcutree.h"
51 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/* Data structures. */ 52/* Data structures. */
60 53
61#define RCU_STATE_INITIALIZER(name) { \ 54#define RCU_STATE_INITIALIZER(name) { \
@@ -70,6 +63,9 @@ EXPORT_SYMBOL_GPL(rcu_lock_map);
70 .gpnum = -300, \ 63 .gpnum = -300, \
71 .completed = -300, \ 64 .completed = -300, \
72 .onofflock = __SPIN_LOCK_UNLOCKED(&name.onofflock), \ 65 .onofflock = __SPIN_LOCK_UNLOCKED(&name.onofflock), \
66 .orphan_cbs_list = NULL, \
67 .orphan_cbs_tail = &name.orphan_cbs_list, \
68 .orphan_qlen = 0, \
73 .fqslock = __SPIN_LOCK_UNLOCKED(&name.fqslock), \ 69 .fqslock = __SPIN_LOCK_UNLOCKED(&name.fqslock), \
74 .n_force_qs = 0, \ 70 .n_force_qs = 0, \
75 .n_force_qs_ngp = 0, \ 71 .n_force_qs_ngp = 0, \
@@ -81,24 +77,16 @@ DEFINE_PER_CPU(struct rcu_data, rcu_sched_data);
81struct rcu_state rcu_bh_state = RCU_STATE_INITIALIZER(rcu_bh_state); 77struct rcu_state rcu_bh_state = RCU_STATE_INITIALIZER(rcu_bh_state);
82DEFINE_PER_CPU(struct rcu_data, rcu_bh_data); 78DEFINE_PER_CPU(struct rcu_data, rcu_bh_data);
83 79
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 80
101#include "rcutree_plugin.h" 81/*
82 * Return true if an RCU grace period is in progress. The ACCESS_ONCE()s
83 * permit this function to be invoked without holding the root rcu_node
84 * structure's ->lock, but of course results can be subject to change.
85 */
86static int rcu_gp_in_progress(struct rcu_state *rsp)
87{
88 return ACCESS_ONCE(rsp->completed) != ACCESS_ONCE(rsp->gpnum);
89}
102 90
103/* 91/*
104 * Note a quiescent state. Because we do not need to know 92 * Note a quiescent state. Because we do not need to know
@@ -137,6 +125,10 @@ static int blimit = 10; /* Maximum callbacks per softirq. */
137static int qhimark = 10000; /* If this many pending, ignore blimit. */ 125static int qhimark = 10000; /* If this many pending, ignore blimit. */
138static int qlowmark = 100; /* Once only this many pending, use blimit. */ 126static int qlowmark = 100; /* Once only this many pending, use blimit. */
139 127
128module_param(blimit, int, 0);
129module_param(qhimark, int, 0);
130module_param(qlowmark, int, 0);
131
140static void force_quiescent_state(struct rcu_state *rsp, int relaxed); 132static void force_quiescent_state(struct rcu_state *rsp, int relaxed);
141static int rcu_pending(int cpu); 133static int rcu_pending(int cpu);
142 134
@@ -173,9 +165,7 @@ cpu_has_callbacks_ready_to_invoke(struct rcu_data *rdp)
173static int 165static int
174cpu_needs_another_gp(struct rcu_state *rsp, struct rcu_data *rdp) 166cpu_needs_another_gp(struct rcu_state *rsp, struct rcu_data *rdp)
175{ 167{
176 /* ACCESS_ONCE() because we are accessing outside of lock. */ 168 return *rdp->nxttail[RCU_DONE_TAIL] && !rcu_gp_in_progress(rsp);
177 return *rdp->nxttail[RCU_DONE_TAIL] &&
178 ACCESS_ONCE(rsp->completed) == ACCESS_ONCE(rsp->gpnum);
179} 169}
180 170
181/* 171/*
@@ -369,7 +359,7 @@ static long dyntick_recall_completed(struct rcu_state *rsp)
369/* 359/*
370 * Snapshot the specified CPU's dynticks counter so that we can later 360 * Snapshot the specified CPU's dynticks counter so that we can later
371 * credit them with an implicit quiescent state. Return 1 if this CPU 361 * credit them with an implicit quiescent state. Return 1 if this CPU
372 * is already in a quiescent state courtesy of dynticks idle mode. 362 * is in dynticks idle mode, which is an extended quiescent state.
373 */ 363 */
374static int dyntick_save_progress_counter(struct rcu_data *rdp) 364static int dyntick_save_progress_counter(struct rcu_data *rdp)
375{ 365{
@@ -475,30 +465,34 @@ static void print_other_cpu_stall(struct rcu_state *rsp)
475 long delta; 465 long delta;
476 unsigned long flags; 466 unsigned long flags;
477 struct rcu_node *rnp = rcu_get_root(rsp); 467 struct rcu_node *rnp = rcu_get_root(rsp);
478 struct rcu_node *rnp_cur = rsp->level[NUM_RCU_LVLS - 1];
479 struct rcu_node *rnp_end = &rsp->node[NUM_RCU_NODES];
480 468
481 /* Only let one CPU complain about others per time interval. */ 469 /* Only let one CPU complain about others per time interval. */
482 470
483 spin_lock_irqsave(&rnp->lock, flags); 471 spin_lock_irqsave(&rnp->lock, flags);
484 delta = jiffies - rsp->jiffies_stall; 472 delta = jiffies - rsp->jiffies_stall;
485 if (delta < RCU_STALL_RAT_DELAY || rsp->gpnum == rsp->completed) { 473 if (delta < RCU_STALL_RAT_DELAY || !rcu_gp_in_progress(rsp)) {
486 spin_unlock_irqrestore(&rnp->lock, flags); 474 spin_unlock_irqrestore(&rnp->lock, flags);
487 return; 475 return;
488 } 476 }
489 rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_RECHECK; 477 rsp->jiffies_stall = jiffies + RCU_SECONDS_TILL_STALL_RECHECK;
478
479 /*
480 * Now rat on any tasks that got kicked up to the root rcu_node
481 * due to CPU offlining.
482 */
483 rcu_print_task_stall(rnp);
490 spin_unlock_irqrestore(&rnp->lock, flags); 484 spin_unlock_irqrestore(&rnp->lock, flags);
491 485
492 /* OK, time to rat on our buddy... */ 486 /* OK, time to rat on our buddy... */
493 487
494 printk(KERN_ERR "INFO: RCU detected CPU stalls:"); 488 printk(KERN_ERR "INFO: RCU detected CPU stalls:");
495 for (; rnp_cur < rnp_end; rnp_cur++) { 489 rcu_for_each_leaf_node(rsp, rnp) {
496 rcu_print_task_stall(rnp); 490 rcu_print_task_stall(rnp);
497 if (rnp_cur->qsmask == 0) 491 if (rnp->qsmask == 0)
498 continue; 492 continue;
499 for (cpu = 0; cpu <= rnp_cur->grphi - rnp_cur->grplo; cpu++) 493 for (cpu = 0; cpu <= rnp->grphi - rnp->grplo; cpu++)
500 if (rnp_cur->qsmask & (1UL << cpu)) 494 if (rnp->qsmask & (1UL << cpu))
501 printk(" %d", rnp_cur->grplo + cpu); 495 printk(" %d", rnp->grplo + cpu);
502 } 496 }
503 printk(" (detected by %d, t=%ld jiffies)\n", 497 printk(" (detected by %d, t=%ld jiffies)\n",
504 smp_processor_id(), (long)(jiffies - rsp->gp_start)); 498 smp_processor_id(), (long)(jiffies - rsp->gp_start));
@@ -537,8 +531,7 @@ static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp)
537 /* We haven't checked in, so go dump stack. */ 531 /* We haven't checked in, so go dump stack. */
538 print_cpu_stall(rsp); 532 print_cpu_stall(rsp);
539 533
540 } else if (rsp->gpnum != rsp->completed && 534 } else if (rcu_gp_in_progress(rsp) && delta >= RCU_STALL_RAT_DELAY) {
541 delta >= RCU_STALL_RAT_DELAY) {
542 535
543 /* They had two time units to dump stack, so complain. */ 536 /* They had two time units to dump stack, so complain. */
544 print_other_cpu_stall(rsp); 537 print_other_cpu_stall(rsp);
@@ -617,9 +610,15 @@ rcu_start_gp(struct rcu_state *rsp, unsigned long flags)
617 note_new_gpnum(rsp, rdp); 610 note_new_gpnum(rsp, rdp);
618 611
619 /* 612 /*
620 * Because we are first, we know that all our callbacks will 613 * Because this CPU just now started the new grace period, we know
621 * be covered by this upcoming grace period, even the ones 614 * that all of its callbacks will be covered by this upcoming grace
622 * that were registered arbitrarily recently. 615 * period, even the ones that were registered arbitrarily recently.
616 * Therefore, advance all outstanding callbacks to RCU_WAIT_TAIL.
617 *
618 * Other CPUs cannot be sure exactly when the grace period started.
619 * Therefore, their recently registered callbacks must pass through
620 * an additional RCU_NEXT_READY stage, so that they will be handled
621 * by the next RCU grace period.
623 */ 622 */
624 rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; 623 rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
625 rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL]; 624 rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
@@ -657,7 +656,7 @@ rcu_start_gp(struct rcu_state *rsp, unsigned long flags)
657 * one corresponding to this CPU, due to the fact that we have 656 * one corresponding to this CPU, due to the fact that we have
658 * irqs disabled. 657 * irqs disabled.
659 */ 658 */
660 for (rnp = &rsp->node[0]; rnp < &rsp->node[NUM_RCU_NODES]; rnp++) { 659 rcu_for_each_node_breadth_first(rsp, rnp) {
661 spin_lock(&rnp->lock); /* irqs already disabled. */ 660 spin_lock(&rnp->lock); /* irqs already disabled. */
662 rcu_preempt_check_blocked_tasks(rnp); 661 rcu_preempt_check_blocked_tasks(rnp);
663 rnp->qsmask = rnp->qsmaskinit; 662 rnp->qsmask = rnp->qsmaskinit;
@@ -703,9 +702,9 @@ rcu_process_gp_end(struct rcu_state *rsp, struct rcu_data *rdp)
703 * hold rnp->lock, as required by rcu_start_gp(), which will release it. 702 * hold rnp->lock, as required by rcu_start_gp(), which will release it.
704 */ 703 */
705static void cpu_quiet_msk_finish(struct rcu_state *rsp, unsigned long flags) 704static void cpu_quiet_msk_finish(struct rcu_state *rsp, unsigned long flags)
706 __releases(rnp->lock) 705 __releases(rcu_get_root(rsp)->lock)
707{ 706{
708 WARN_ON_ONCE(rsp->completed == rsp->gpnum); 707 WARN_ON_ONCE(!rcu_gp_in_progress(rsp));
709 rsp->completed = rsp->gpnum; 708 rsp->completed = rsp->gpnum;
710 rcu_process_gp_end(rsp, rsp->rda[smp_processor_id()]); 709 rcu_process_gp_end(rsp, rsp->rda[smp_processor_id()]);
711 rcu_start_gp(rsp, flags); /* releases root node's rnp->lock. */ 710 rcu_start_gp(rsp, flags); /* releases root node's rnp->lock. */
@@ -842,17 +841,63 @@ rcu_check_quiescent_state(struct rcu_state *rsp, struct rcu_data *rdp)
842#ifdef CONFIG_HOTPLUG_CPU 841#ifdef CONFIG_HOTPLUG_CPU
843 842
844/* 843/*
844 * Move a dying CPU's RCU callbacks to the ->orphan_cbs_list for the
845 * specified flavor of RCU. The callbacks will be adopted by the next
846 * _rcu_barrier() invocation or by the CPU_DEAD notifier, whichever
847 * comes first. Because this is invoked from the CPU_DYING notifier,
848 * irqs are already disabled.
849 */
850static void rcu_send_cbs_to_orphanage(struct rcu_state *rsp)
851{
852 int i;
853 struct rcu_data *rdp = rsp->rda[smp_processor_id()];
854
855 if (rdp->nxtlist == NULL)
856 return; /* irqs disabled, so comparison is stable. */
857 spin_lock(&rsp->onofflock); /* irqs already disabled. */
858 *rsp->orphan_cbs_tail = rdp->nxtlist;
859 rsp->orphan_cbs_tail = rdp->nxttail[RCU_NEXT_TAIL];
860 rdp->nxtlist = NULL;
861 for (i = 0; i < RCU_NEXT_SIZE; i++)
862 rdp->nxttail[i] = &rdp->nxtlist;
863 rsp->orphan_qlen += rdp->qlen;
864 rdp->qlen = 0;
865 spin_unlock(&rsp->onofflock); /* irqs remain disabled. */
866}
867
868/*
869 * Adopt previously orphaned RCU callbacks.
870 */
871static void rcu_adopt_orphan_cbs(struct rcu_state *rsp)
872{
873 unsigned long flags;
874 struct rcu_data *rdp;
875
876 spin_lock_irqsave(&rsp->onofflock, flags);
877 rdp = rsp->rda[smp_processor_id()];
878 if (rsp->orphan_cbs_list == NULL) {
879 spin_unlock_irqrestore(&rsp->onofflock, flags);
880 return;
881 }
882 *rdp->nxttail[RCU_NEXT_TAIL] = rsp->orphan_cbs_list;
883 rdp->nxttail[RCU_NEXT_TAIL] = rsp->orphan_cbs_tail;
884 rdp->qlen += rsp->orphan_qlen;
885 rsp->orphan_cbs_list = NULL;
886 rsp->orphan_cbs_tail = &rsp->orphan_cbs_list;
887 rsp->orphan_qlen = 0;
888 spin_unlock_irqrestore(&rsp->onofflock, flags);
889}
890
891/*
845 * Remove the outgoing CPU from the bitmasks in the rcu_node hierarchy 892 * Remove the outgoing CPU from the bitmasks in the rcu_node hierarchy
846 * and move all callbacks from the outgoing CPU to the current one. 893 * and move all callbacks from the outgoing CPU to the current one.
847 */ 894 */
848static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp) 895static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp)
849{ 896{
850 int i;
851 unsigned long flags; 897 unsigned long flags;
852 long lastcomp; 898 long lastcomp;
853 unsigned long mask; 899 unsigned long mask;
854 struct rcu_data *rdp = rsp->rda[cpu]; 900 struct rcu_data *rdp = rsp->rda[cpu];
855 struct rcu_data *rdp_me;
856 struct rcu_node *rnp; 901 struct rcu_node *rnp;
857 902
858 /* Exclude any attempts to start a new grace period. */ 903 /* Exclude any attempts to start a new grace period. */
@@ -875,32 +920,9 @@ static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp)
875 } while (rnp != NULL); 920 } while (rnp != NULL);
876 lastcomp = rsp->completed; 921 lastcomp = rsp->completed;
877 922
878 spin_unlock(&rsp->onofflock); /* irqs remain disabled. */ 923 spin_unlock_irqrestore(&rsp->onofflock, flags);
879 924
880 /* 925 rcu_adopt_orphan_cbs(rsp);
881 * Move callbacks from the outgoing CPU to the running CPU.
882 * Note that the outgoing CPU is now quiscent, so it is now
883 * (uncharacteristically) safe to access its rcu_data structure.
884 * Note also that we must carefully retain the order of the
885 * outgoing CPU's callbacks in order for rcu_barrier() to work
886 * correctly. Finally, note that we start all the callbacks
887 * afresh, even those that have passed through a grace period
888 * and are therefore ready to invoke. The theory is that hotplug
889 * events are rare, and that if they are frequent enough to
890 * indefinitely delay callbacks, you have far worse things to
891 * be worrying about.
892 */
893 rdp_me = rsp->rda[smp_processor_id()];
894 if (rdp->nxtlist != NULL) {
895 *rdp_me->nxttail[RCU_NEXT_TAIL] = rdp->nxtlist;
896 rdp_me->nxttail[RCU_NEXT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
897 rdp->nxtlist = NULL;
898 for (i = 0; i < RCU_NEXT_SIZE; i++)
899 rdp->nxttail[i] = &rdp->nxtlist;
900 rdp_me->qlen += rdp->qlen;
901 rdp->qlen = 0;
902 }
903 local_irq_restore(flags);
904} 926}
905 927
906/* 928/*
@@ -918,6 +940,14 @@ static void rcu_offline_cpu(int cpu)
918 940
919#else /* #ifdef CONFIG_HOTPLUG_CPU */ 941#else /* #ifdef CONFIG_HOTPLUG_CPU */
920 942
943static void rcu_send_cbs_to_orphanage(struct rcu_state *rsp)
944{
945}
946
947static void rcu_adopt_orphan_cbs(struct rcu_state *rsp)
948{
949}
950
921static void rcu_offline_cpu(int cpu) 951static void rcu_offline_cpu(int cpu)
922{ 952{
923} 953}
@@ -1050,33 +1080,32 @@ static int rcu_process_dyntick(struct rcu_state *rsp, long lastcomp,
1050 int cpu; 1080 int cpu;
1051 unsigned long flags; 1081 unsigned long flags;
1052 unsigned long mask; 1082 unsigned long mask;
1053 struct rcu_node *rnp_cur = rsp->level[NUM_RCU_LVLS - 1]; 1083 struct rcu_node *rnp;
1054 struct rcu_node *rnp_end = &rsp->node[NUM_RCU_NODES];
1055 1084
1056 for (; rnp_cur < rnp_end; rnp_cur++) { 1085 rcu_for_each_leaf_node(rsp, rnp) {
1057 mask = 0; 1086 mask = 0;
1058 spin_lock_irqsave(&rnp_cur->lock, flags); 1087 spin_lock_irqsave(&rnp->lock, flags);
1059 if (rsp->completed != lastcomp) { 1088 if (rsp->completed != lastcomp) {
1060 spin_unlock_irqrestore(&rnp_cur->lock, flags); 1089 spin_unlock_irqrestore(&rnp->lock, flags);
1061 return 1; 1090 return 1;
1062 } 1091 }
1063 if (rnp_cur->qsmask == 0) { 1092 if (rnp->qsmask == 0) {
1064 spin_unlock_irqrestore(&rnp_cur->lock, flags); 1093 spin_unlock_irqrestore(&rnp->lock, flags);
1065 continue; 1094 continue;
1066 } 1095 }
1067 cpu = rnp_cur->grplo; 1096 cpu = rnp->grplo;
1068 bit = 1; 1097 bit = 1;
1069 for (; cpu <= rnp_cur->grphi; cpu++, bit <<= 1) { 1098 for (; cpu <= rnp->grphi; cpu++, bit <<= 1) {
1070 if ((rnp_cur->qsmask & bit) != 0 && f(rsp->rda[cpu])) 1099 if ((rnp->qsmask & bit) != 0 && f(rsp->rda[cpu]))
1071 mask |= bit; 1100 mask |= bit;
1072 } 1101 }
1073 if (mask != 0 && rsp->completed == lastcomp) { 1102 if (mask != 0 && rsp->completed == lastcomp) {
1074 1103
1075 /* cpu_quiet_msk() releases rnp_cur->lock. */ 1104 /* cpu_quiet_msk() releases rnp->lock. */
1076 cpu_quiet_msk(mask, rsp, rnp_cur, flags); 1105 cpu_quiet_msk(mask, rsp, rnp, flags);
1077 continue; 1106 continue;
1078 } 1107 }
1079 spin_unlock_irqrestore(&rnp_cur->lock, flags); 1108 spin_unlock_irqrestore(&rnp->lock, flags);
1080 } 1109 }
1081 return 0; 1110 return 0;
1082} 1111}
@@ -1092,7 +1121,7 @@ static void force_quiescent_state(struct rcu_state *rsp, int relaxed)
1092 struct rcu_node *rnp = rcu_get_root(rsp); 1121 struct rcu_node *rnp = rcu_get_root(rsp);
1093 u8 signaled; 1122 u8 signaled;
1094 1123
1095 if (ACCESS_ONCE(rsp->completed) == ACCESS_ONCE(rsp->gpnum)) 1124 if (!rcu_gp_in_progress(rsp))
1096 return; /* No grace period in progress, nothing to force. */ 1125 return; /* No grace period in progress, nothing to force. */
1097 if (!spin_trylock_irqsave(&rsp->fqslock, flags)) { 1126 if (!spin_trylock_irqsave(&rsp->fqslock, flags)) {
1098 rsp->n_force_qs_lh++; /* Inexact, can lose counts. Tough! */ 1127 rsp->n_force_qs_lh++; /* Inexact, can lose counts. Tough! */
@@ -1251,7 +1280,7 @@ __call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu),
1251 rdp->nxttail[RCU_NEXT_TAIL] = &head->next; 1280 rdp->nxttail[RCU_NEXT_TAIL] = &head->next;
1252 1281
1253 /* Start a new grace period if one not already started. */ 1282 /* Start a new grace period if one not already started. */
1254 if (ACCESS_ONCE(rsp->completed) == ACCESS_ONCE(rsp->gpnum)) { 1283 if (!rcu_gp_in_progress(rsp)) {
1255 unsigned long nestflag; 1284 unsigned long nestflag;
1256 struct rcu_node *rnp_root = rcu_get_root(rsp); 1285 struct rcu_node *rnp_root = rcu_get_root(rsp);
1257 1286
@@ -1331,7 +1360,7 @@ static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp)
1331 } 1360 }
1332 1361
1333 /* Has an RCU GP gone long enough to send resched IPIs &c? */ 1362 /* Has an RCU GP gone long enough to send resched IPIs &c? */
1334 if (ACCESS_ONCE(rsp->completed) != ACCESS_ONCE(rsp->gpnum) && 1363 if (rcu_gp_in_progress(rsp) &&
1335 ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0)) { 1364 ((long)(ACCESS_ONCE(rsp->jiffies_force_qs) - jiffies) < 0)) {
1336 rdp->n_rp_need_fqs++; 1365 rdp->n_rp_need_fqs++;
1337 return 1; 1366 return 1;
@@ -1368,6 +1397,82 @@ int rcu_needs_cpu(int cpu)
1368 rcu_preempt_needs_cpu(cpu); 1397 rcu_preempt_needs_cpu(cpu);
1369} 1398}
1370 1399
1400static DEFINE_PER_CPU(struct rcu_head, rcu_barrier_head) = {NULL};
1401static atomic_t rcu_barrier_cpu_count;
1402static DEFINE_MUTEX(rcu_barrier_mutex);
1403static struct completion rcu_barrier_completion;
1404
1405static void rcu_barrier_callback(struct rcu_head *notused)
1406{
1407 if (atomic_dec_and_test(&rcu_barrier_cpu_count))
1408 complete(&rcu_barrier_completion);
1409}
1410
1411/*
1412 * Called with preemption disabled, and from cross-cpu IRQ context.
1413 */
1414static void rcu_barrier_func(void *type)
1415{
1416 int cpu = smp_processor_id();
1417 struct rcu_head *head = &per_cpu(rcu_barrier_head, cpu);
1418 void (*call_rcu_func)(struct rcu_head *head,
1419 void (*func)(struct rcu_head *head));
1420
1421 atomic_inc(&rcu_barrier_cpu_count);
1422 call_rcu_func = type;
1423 call_rcu_func(head, rcu_barrier_callback);
1424}
1425
1426/*
1427 * Orchestrate the specified type of RCU barrier, waiting for all
1428 * RCU callbacks of the specified type to complete.
1429 */
1430static void _rcu_barrier(struct rcu_state *rsp,
1431 void (*call_rcu_func)(struct rcu_head *head,
1432 void (*func)(struct rcu_head *head)))
1433{
1434 BUG_ON(in_interrupt());
1435 /* Take mutex to serialize concurrent rcu_barrier() requests. */
1436 mutex_lock(&rcu_barrier_mutex);
1437 init_completion(&rcu_barrier_completion);
1438 /*
1439 * Initialize rcu_barrier_cpu_count to 1, then invoke
1440 * rcu_barrier_func() on each CPU, so that each CPU also has
1441 * incremented rcu_barrier_cpu_count. Only then is it safe to
1442 * decrement rcu_barrier_cpu_count -- otherwise the first CPU
1443 * might complete its grace period before all of the other CPUs
1444 * did their increment, causing this function to return too
1445 * early.
1446 */
1447 atomic_set(&rcu_barrier_cpu_count, 1);
1448 preempt_disable(); /* stop CPU_DYING from filling orphan_cbs_list */
1449 rcu_adopt_orphan_cbs(rsp);
1450 on_each_cpu(rcu_barrier_func, (void *)call_rcu_func, 1);
1451 preempt_enable(); /* CPU_DYING can again fill orphan_cbs_list */
1452 if (atomic_dec_and_test(&rcu_barrier_cpu_count))
1453 complete(&rcu_barrier_completion);
1454 wait_for_completion(&rcu_barrier_completion);
1455 mutex_unlock(&rcu_barrier_mutex);
1456}
1457
1458/**
1459 * rcu_barrier_bh - Wait until all in-flight call_rcu_bh() callbacks complete.
1460 */
1461void rcu_barrier_bh(void)
1462{
1463 _rcu_barrier(&rcu_bh_state, call_rcu_bh);
1464}
1465EXPORT_SYMBOL_GPL(rcu_barrier_bh);
1466
1467/**
1468 * rcu_barrier_sched - Wait for in-flight call_rcu_sched() callbacks.
1469 */
1470void rcu_barrier_sched(void)
1471{
1472 _rcu_barrier(&rcu_sched_state, call_rcu_sched);
1473}
1474EXPORT_SYMBOL_GPL(rcu_barrier_sched);
1475
1371/* 1476/*
1372 * Do boot-time initialization of a CPU's per-CPU RCU data. 1477 * Do boot-time initialization of a CPU's per-CPU RCU data.
1373 */ 1478 */
@@ -1464,6 +1569,22 @@ int __cpuinit rcu_cpu_notify(struct notifier_block *self,
1464 case CPU_UP_PREPARE_FROZEN: 1569 case CPU_UP_PREPARE_FROZEN:
1465 rcu_online_cpu(cpu); 1570 rcu_online_cpu(cpu);
1466 break; 1571 break;
1572 case CPU_DYING:
1573 case CPU_DYING_FROZEN:
1574 /*
1575 * preempt_disable() in _rcu_barrier() prevents stop_machine(),
1576 * so when "on_each_cpu(rcu_barrier_func, (void *)type, 1);"
1577 * returns, all online cpus have queued rcu_barrier_func().
1578 * The dying CPU clears its cpu_online_mask bit and
1579 * moves all of its RCU callbacks to ->orphan_cbs_list
1580 * in the context of stop_machine(), so subsequent calls
1581 * to _rcu_barrier() will adopt these callbacks and only
1582 * then queue rcu_barrier_func() on all remaining CPUs.
1583 */
1584 rcu_send_cbs_to_orphanage(&rcu_bh_state);
1585 rcu_send_cbs_to_orphanage(&rcu_sched_state);
1586 rcu_preempt_send_cbs_to_orphanage();
1587 break;
1467 case CPU_DEAD: 1588 case CPU_DEAD:
1468 case CPU_DEAD_FROZEN: 1589 case CPU_DEAD_FROZEN:
1469 case CPU_UP_CANCELED: 1590 case CPU_UP_CANCELED:
@@ -1526,7 +1647,8 @@ static void __init rcu_init_one(struct rcu_state *rsp)
1526 cpustride *= rsp->levelspread[i]; 1647 cpustride *= rsp->levelspread[i];
1527 rnp = rsp->level[i]; 1648 rnp = rsp->level[i];
1528 for (j = 0; j < rsp->levelcnt[i]; j++, rnp++) { 1649 for (j = 0; j < rsp->levelcnt[i]; j++, rnp++) {
1529 spin_lock_init(&rnp->lock); 1650 if (rnp != rcu_get_root(rsp))
1651 spin_lock_init(&rnp->lock);
1530 rnp->gpnum = 0; 1652 rnp->gpnum = 0;
1531 rnp->qsmask = 0; 1653 rnp->qsmask = 0;
1532 rnp->qsmaskinit = 0; 1654 rnp->qsmaskinit = 0;
@@ -1549,6 +1671,7 @@ static void __init rcu_init_one(struct rcu_state *rsp)
1549 INIT_LIST_HEAD(&rnp->blocked_tasks[1]); 1671 INIT_LIST_HEAD(&rnp->blocked_tasks[1]);
1550 } 1672 }
1551 } 1673 }
1674 spin_lock_init(&rcu_get_root(rsp)->lock);
1552} 1675}
1553 1676
1554/* 1677/*
@@ -1558,6 +1681,10 @@ static void __init rcu_init_one(struct rcu_state *rsp)
1558 */ 1681 */
1559#define RCU_INIT_FLAVOR(rsp, rcu_data) \ 1682#define RCU_INIT_FLAVOR(rsp, rcu_data) \
1560do { \ 1683do { \
1684 int i; \
1685 int j; \
1686 struct rcu_node *rnp; \
1687 \
1561 rcu_init_one(rsp); \ 1688 rcu_init_one(rsp); \
1562 rnp = (rsp)->level[NUM_RCU_LVLS - 1]; \ 1689 rnp = (rsp)->level[NUM_RCU_LVLS - 1]; \
1563 j = 0; \ 1690 j = 0; \
@@ -1570,31 +1697,8 @@ do { \
1570 } \ 1697 } \
1571} while (0) 1698} while (0)
1572 1699
1573#ifdef CONFIG_TREE_PREEMPT_RCU
1574
1575void __init __rcu_init_preempt(void)
1576{
1577 int i; /* All used by RCU_INIT_FLAVOR(). */
1578 int j;
1579 struct rcu_node *rnp;
1580
1581 RCU_INIT_FLAVOR(&rcu_preempt_state, rcu_preempt_data);
1582}
1583
1584#else /* #ifdef CONFIG_TREE_PREEMPT_RCU */
1585
1586void __init __rcu_init_preempt(void)
1587{
1588}
1589
1590#endif /* #else #ifdef CONFIG_TREE_PREEMPT_RCU */
1591
1592void __init __rcu_init(void) 1700void __init __rcu_init(void)
1593{ 1701{
1594 int i; /* All used by RCU_INIT_FLAVOR(). */
1595 int j;
1596 struct rcu_node *rnp;
1597
1598 rcu_bootup_announce(); 1702 rcu_bootup_announce();
1599#ifdef CONFIG_RCU_CPU_STALL_DETECTOR 1703#ifdef CONFIG_RCU_CPU_STALL_DETECTOR
1600 printk(KERN_INFO "RCU-based detection of stalled CPUs is enabled.\n"); 1704 printk(KERN_INFO "RCU-based detection of stalled CPUs is enabled.\n");
@@ -1605,6 +1709,4 @@ void __init __rcu_init(void)
1605 open_softirq(RCU_SOFTIRQ, rcu_process_callbacks); 1709 open_softirq(RCU_SOFTIRQ, rcu_process_callbacks);
1606} 1710}
1607 1711
1608module_param(blimit, int, 0); 1712#include "rcutree_plugin.h"
1609module_param(qhimark, int, 0);
1610module_param(qlowmark, int, 0);
diff --git a/kernel/rcutree.h b/kernel/rcutree.h
index 8e8287a983c2..b40ac5706040 100644
--- a/kernel/rcutree.h
+++ b/kernel/rcutree.h
@@ -48,14 +48,14 @@
48#elif NR_CPUS <= RCU_FANOUT_SQ 48#elif NR_CPUS <= RCU_FANOUT_SQ
49# define NUM_RCU_LVLS 2 49# define NUM_RCU_LVLS 2
50# define NUM_RCU_LVL_0 1 50# define NUM_RCU_LVL_0 1
51# define NUM_RCU_LVL_1 (((NR_CPUS) + RCU_FANOUT - 1) / RCU_FANOUT) 51# define NUM_RCU_LVL_1 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT)
52# define NUM_RCU_LVL_2 (NR_CPUS) 52# define NUM_RCU_LVL_2 (NR_CPUS)
53# define NUM_RCU_LVL_3 0 53# define NUM_RCU_LVL_3 0
54#elif NR_CPUS <= RCU_FANOUT_CUBE 54#elif NR_CPUS <= RCU_FANOUT_CUBE
55# define NUM_RCU_LVLS 3 55# define NUM_RCU_LVLS 3
56# define NUM_RCU_LVL_0 1 56# define NUM_RCU_LVL_0 1
57# define NUM_RCU_LVL_1 (((NR_CPUS) + RCU_FANOUT_SQ - 1) / RCU_FANOUT_SQ) 57# define NUM_RCU_LVL_1 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_SQ)
58# define NUM_RCU_LVL_2 (((NR_CPUS) + (RCU_FANOUT) - 1) / (RCU_FANOUT)) 58# define NUM_RCU_LVL_2 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT)
59# define NUM_RCU_LVL_3 NR_CPUS 59# define NUM_RCU_LVL_3 NR_CPUS
60#else 60#else
61# error "CONFIG_RCU_FANOUT insufficient for NR_CPUS" 61# error "CONFIG_RCU_FANOUT insufficient for NR_CPUS"
@@ -79,15 +79,21 @@ struct rcu_dynticks {
79 * Definition for node within the RCU grace-period-detection hierarchy. 79 * Definition for node within the RCU grace-period-detection hierarchy.
80 */ 80 */
81struct rcu_node { 81struct rcu_node {
82 spinlock_t lock; 82 spinlock_t lock; /* Root rcu_node's lock protects some */
83 /* rcu_state fields as well as following. */
83 long gpnum; /* Current grace period for this node. */ 84 long gpnum; /* Current grace period for this node. */
84 /* This will either be equal to or one */ 85 /* This will either be equal to or one */
85 /* behind the root rcu_node's gpnum. */ 86 /* behind the root rcu_node's gpnum. */
86 unsigned long qsmask; /* CPUs or groups that need to switch in */ 87 unsigned long qsmask; /* CPUs or groups that need to switch in */
87 /* order for current grace period to proceed.*/ 88 /* order for current grace period to proceed.*/
89 /* In leaf rcu_node, each bit corresponds to */
90 /* an rcu_data structure, otherwise, each */
91 /* bit corresponds to a child rcu_node */
92 /* structure. */
88 unsigned long qsmaskinit; 93 unsigned long qsmaskinit;
89 /* Per-GP initialization for qsmask. */ 94 /* Per-GP initialization for qsmask. */
90 unsigned long grpmask; /* Mask to apply to parent qsmask. */ 95 unsigned long grpmask; /* Mask to apply to parent qsmask. */
96 /* Only one bit will be set in this mask. */
91 int grplo; /* lowest-numbered CPU or group here. */ 97 int grplo; /* lowest-numbered CPU or group here. */
92 int grphi; /* highest-numbered CPU or group here. */ 98 int grphi; /* highest-numbered CPU or group here. */
93 u8 grpnum; /* CPU/group number for next level up. */ 99 u8 grpnum; /* CPU/group number for next level up. */
@@ -95,8 +101,23 @@ struct rcu_node {
95 struct rcu_node *parent; 101 struct rcu_node *parent;
96 struct list_head blocked_tasks[2]; 102 struct list_head blocked_tasks[2];
97 /* Tasks blocked in RCU read-side critsect. */ 103 /* Tasks blocked in RCU read-side critsect. */
104 /* Grace period number (->gpnum) x blocked */
105 /* by tasks on the (x & 0x1) element of the */
106 /* blocked_tasks[] array. */
98} ____cacheline_internodealigned_in_smp; 107} ____cacheline_internodealigned_in_smp;
99 108
109/*
110 * Do a full breadth-first scan of the rcu_node structures for the
111 * specified rcu_state structure.
112 */
113#define rcu_for_each_node_breadth_first(rsp, rnp) \
114 for ((rnp) = &(rsp)->node[0]; \
115 (rnp) < &(rsp)->node[NUM_RCU_NODES]; (rnp)++)
116
117#define rcu_for_each_leaf_node(rsp, rnp) \
118 for ((rnp) = (rsp)->level[NUM_RCU_LVLS - 1]; \
119 (rnp) < &(rsp)->node[NUM_RCU_NODES]; (rnp)++)
120
100/* Index values for nxttail array in struct rcu_data. */ 121/* Index values for nxttail array in struct rcu_data. */
101#define RCU_DONE_TAIL 0 /* Also RCU_WAIT head. */ 122#define RCU_DONE_TAIL 0 /* Also RCU_WAIT head. */
102#define RCU_WAIT_TAIL 1 /* Also RCU_NEXT_READY head. */ 123#define RCU_WAIT_TAIL 1 /* Also RCU_NEXT_READY head. */
@@ -126,19 +147,22 @@ struct rcu_data {
126 * Any of the partitions might be empty, in which case the 147 * Any of the partitions might be empty, in which case the
127 * pointer to that partition will be equal to the pointer for 148 * pointer to that partition will be equal to the pointer for
128 * the following partition. When the list is empty, all of 149 * the following partition. When the list is empty, all of
129 * the nxttail elements point to nxtlist, which is NULL. 150 * the nxttail elements point to the ->nxtlist pointer itself,
151 * which in that case is NULL.
130 * 152 *
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]): 153 * [nxtlist, *nxttail[RCU_DONE_TAIL]):
138 * Entries that batch # <= ->completed 154 * Entries that batch # <= ->completed
139 * The grace period for these entries has completed, and 155 * The grace period for these entries has completed, and
140 * the other grace-period-completed entries may be moved 156 * the other grace-period-completed entries may be moved
141 * here temporarily in rcu_process_callbacks(). 157 * here temporarily in rcu_process_callbacks().
158 * [*nxttail[RCU_DONE_TAIL], *nxttail[RCU_WAIT_TAIL]):
159 * Entries that batch # <= ->completed - 1: waiting for current GP
160 * [*nxttail[RCU_WAIT_TAIL], *nxttail[RCU_NEXT_READY_TAIL]):
161 * Entries known to have arrived before current GP ended
162 * [*nxttail[RCU_NEXT_READY_TAIL], *nxttail[RCU_NEXT_TAIL]):
163 * Entries that might have arrived after current GP ended
164 * Note that the value of *nxttail[RCU_NEXT_TAIL] will
165 * always be NULL, as this is the end of the list.
142 */ 166 */
143 struct rcu_head *nxtlist; 167 struct rcu_head *nxtlist;
144 struct rcu_head **nxttail[RCU_NEXT_SIZE]; 168 struct rcu_head **nxttail[RCU_NEXT_SIZE];
@@ -216,8 +240,19 @@ struct rcu_state {
216 /* Force QS state. */ 240 /* Force QS state. */
217 long gpnum; /* Current gp number. */ 241 long gpnum; /* Current gp number. */
218 long completed; /* # of last completed gp. */ 242 long completed; /* # of last completed gp. */
243
244 /* End of fields guarded by root rcu_node's lock. */
245
219 spinlock_t onofflock; /* exclude on/offline and */ 246 spinlock_t onofflock; /* exclude on/offline and */
220 /* starting new GP. */ 247 /* starting new GP. Also */
248 /* protects the following */
249 /* orphan_cbs fields. */
250 struct rcu_head *orphan_cbs_list; /* list of rcu_head structs */
251 /* orphaned by all CPUs in */
252 /* a given leaf rcu_node */
253 /* going offline. */
254 struct rcu_head **orphan_cbs_tail; /* And tail pointer. */
255 long orphan_qlen; /* Number of orphaned cbs. */
221 spinlock_t fqslock; /* Only one task forcing */ 256 spinlock_t fqslock; /* Only one task forcing */
222 /* quiescent states. */ 257 /* quiescent states. */
223 unsigned long jiffies_force_qs; /* Time at which to invoke */ 258 unsigned long jiffies_force_qs; /* Time at which to invoke */
@@ -255,5 +290,30 @@ extern struct rcu_state rcu_preempt_state;
255DECLARE_PER_CPU(struct rcu_data, rcu_preempt_data); 290DECLARE_PER_CPU(struct rcu_data, rcu_preempt_data);
256#endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */ 291#endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */
257 292
258#endif /* #ifdef RCU_TREE_NONCORE */ 293#else /* #ifdef RCU_TREE_NONCORE */
294
295/* Forward declarations for rcutree_plugin.h */
296static inline void rcu_bootup_announce(void);
297long rcu_batches_completed(void);
298static void rcu_preempt_note_context_switch(int cpu);
299static int rcu_preempted_readers(struct rcu_node *rnp);
300#ifdef CONFIG_RCU_CPU_STALL_DETECTOR
301static void rcu_print_task_stall(struct rcu_node *rnp);
302#endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
303static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp);
304#ifdef CONFIG_HOTPLUG_CPU
305static void rcu_preempt_offline_tasks(struct rcu_state *rsp,
306 struct rcu_node *rnp,
307 struct rcu_data *rdp);
308static void rcu_preempt_offline_cpu(int cpu);
309#endif /* #ifdef CONFIG_HOTPLUG_CPU */
310static void rcu_preempt_check_callbacks(int cpu);
311static void rcu_preempt_process_callbacks(void);
312void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu));
313static int rcu_preempt_pending(int cpu);
314static int rcu_preempt_needs_cpu(int cpu);
315static void __cpuinit rcu_preempt_init_percpu_data(int cpu);
316static void rcu_preempt_send_cbs_to_orphanage(void);
317static void __init __rcu_init_preempt(void);
259 318
319#endif /* #else #ifdef RCU_TREE_NONCORE */
diff --git a/kernel/rcutree_plugin.h b/kernel/rcutree_plugin.h
index 1cee04f627eb..c0cb783aa16a 100644
--- a/kernel/rcutree_plugin.h
+++ b/kernel/rcutree_plugin.h
@@ -150,6 +150,16 @@ void __rcu_read_lock(void)
150} 150}
151EXPORT_SYMBOL_GPL(__rcu_read_lock); 151EXPORT_SYMBOL_GPL(__rcu_read_lock);
152 152
153/*
154 * Check for preempted RCU readers blocking the current grace period
155 * for the specified rcu_node structure. If the caller needs a reliable
156 * answer, it must hold the rcu_node's ->lock.
157 */
158static int rcu_preempted_readers(struct rcu_node *rnp)
159{
160 return !list_empty(&rnp->blocked_tasks[rnp->gpnum & 0x1]);
161}
162
153static void rcu_read_unlock_special(struct task_struct *t) 163static void rcu_read_unlock_special(struct task_struct *t)
154{ 164{
155 int empty; 165 int empty;
@@ -196,7 +206,7 @@ static void rcu_read_unlock_special(struct task_struct *t)
196 break; 206 break;
197 spin_unlock(&rnp->lock); /* irqs remain disabled. */ 207 spin_unlock(&rnp->lock); /* irqs remain disabled. */
198 } 208 }
199 empty = list_empty(&rnp->blocked_tasks[rnp->gpnum & 0x1]); 209 empty = !rcu_preempted_readers(rnp);
200 list_del_init(&t->rcu_node_entry); 210 list_del_init(&t->rcu_node_entry);
201 t->rcu_blocked_node = NULL; 211 t->rcu_blocked_node = NULL;
202 212
@@ -207,7 +217,7 @@ static void rcu_read_unlock_special(struct task_struct *t)
207 * drop rnp->lock and restore irq. 217 * drop rnp->lock and restore irq.
208 */ 218 */
209 if (!empty && rnp->qsmask == 0 && 219 if (!empty && rnp->qsmask == 0 &&
210 list_empty(&rnp->blocked_tasks[rnp->gpnum & 0x1])) { 220 !rcu_preempted_readers(rnp)) {
211 struct rcu_node *rnp_p; 221 struct rcu_node *rnp_p;
212 222
213 if (rnp->parent == NULL) { 223 if (rnp->parent == NULL) {
@@ -257,12 +267,12 @@ static void rcu_print_task_stall(struct rcu_node *rnp)
257{ 267{
258 unsigned long flags; 268 unsigned long flags;
259 struct list_head *lp; 269 struct list_head *lp;
260 int phase = rnp->gpnum & 0x1; 270 int phase;
261 struct task_struct *t; 271 struct task_struct *t;
262 272
263 if (!list_empty(&rnp->blocked_tasks[phase])) { 273 if (rcu_preempted_readers(rnp)) {
264 spin_lock_irqsave(&rnp->lock, flags); 274 spin_lock_irqsave(&rnp->lock, flags);
265 phase = rnp->gpnum & 0x1; /* re-read under lock. */ 275 phase = rnp->gpnum & 0x1;
266 lp = &rnp->blocked_tasks[phase]; 276 lp = &rnp->blocked_tasks[phase];
267 list_for_each_entry(t, lp, rcu_node_entry) 277 list_for_each_entry(t, lp, rcu_node_entry)
268 printk(" P%d", t->pid); 278 printk(" P%d", t->pid);
@@ -281,20 +291,10 @@ static void rcu_print_task_stall(struct rcu_node *rnp)
281 */ 291 */
282static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp) 292static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp)
283{ 293{
284 WARN_ON_ONCE(!list_empty(&rnp->blocked_tasks[rnp->gpnum & 0x1])); 294 WARN_ON_ONCE(rcu_preempted_readers(rnp));
285 WARN_ON_ONCE(rnp->qsmask); 295 WARN_ON_ONCE(rnp->qsmask);
286} 296}
287 297
288/*
289 * Check for preempted RCU readers for the specified rcu_node structure.
290 * If the caller needs a reliable answer, it must hold the rcu_node's
291 * >lock.
292 */
293static int rcu_preempted_readers(struct rcu_node *rnp)
294{
295 return !list_empty(&rnp->blocked_tasks[rnp->gpnum & 0x1]);
296}
297
298#ifdef CONFIG_HOTPLUG_CPU 298#ifdef CONFIG_HOTPLUG_CPU
299 299
300/* 300/*
@@ -410,6 +410,15 @@ static int rcu_preempt_needs_cpu(int cpu)
410 return !!per_cpu(rcu_preempt_data, cpu).nxtlist; 410 return !!per_cpu(rcu_preempt_data, cpu).nxtlist;
411} 411}
412 412
413/**
414 * rcu_barrier - Wait until all in-flight call_rcu() callbacks complete.
415 */
416void rcu_barrier(void)
417{
418 _rcu_barrier(&rcu_preempt_state, call_rcu);
419}
420EXPORT_SYMBOL_GPL(rcu_barrier);
421
413/* 422/*
414 * Initialize preemptable RCU's per-CPU data. 423 * Initialize preemptable RCU's per-CPU data.
415 */ 424 */
@@ -419,6 +428,22 @@ static void __cpuinit rcu_preempt_init_percpu_data(int cpu)
419} 428}
420 429
421/* 430/*
431 * Move preemptable RCU's callbacks to ->orphan_cbs_list.
432 */
433static void rcu_preempt_send_cbs_to_orphanage(void)
434{
435 rcu_send_cbs_to_orphanage(&rcu_preempt_state);
436}
437
438/*
439 * Initialize preemptable RCU's state structures.
440 */
441static void __init __rcu_init_preempt(void)
442{
443 RCU_INIT_FLAVOR(&rcu_preempt_state, rcu_preempt_data);
444}
445
446/*
422 * Check for a task exiting while in a preemptable-RCU read-side 447 * Check for a task exiting while in a preemptable-RCU read-side
423 * critical section, clean up if so. No need to issue warnings, 448 * critical section, clean up if so. No need to issue warnings,
424 * as debug_check_no_locks_held() already does this if lockdep 449 * as debug_check_no_locks_held() already does this if lockdep
@@ -461,6 +486,15 @@ static void rcu_preempt_note_context_switch(int cpu)
461{ 486{
462} 487}
463 488
489/*
490 * Because preemptable RCU does not exist, there are never any preempted
491 * RCU readers.
492 */
493static int rcu_preempted_readers(struct rcu_node *rnp)
494{
495 return 0;
496}
497
464#ifdef CONFIG_RCU_CPU_STALL_DETECTOR 498#ifdef CONFIG_RCU_CPU_STALL_DETECTOR
465 499
466/* 500/*
@@ -483,15 +517,6 @@ static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp)
483 WARN_ON_ONCE(rnp->qsmask); 517 WARN_ON_ONCE(rnp->qsmask);
484} 518}
485 519
486/*
487 * Because preemptable RCU does not exist, there are never any preempted
488 * RCU readers.
489 */
490static int rcu_preempted_readers(struct rcu_node *rnp)
491{
492 return 0;
493}
494
495#ifdef CONFIG_HOTPLUG_CPU 520#ifdef CONFIG_HOTPLUG_CPU
496 521
497/* 522/*
@@ -518,7 +543,7 @@ static void rcu_preempt_offline_cpu(int cpu)
518 * Because preemptable RCU does not exist, it never has any callbacks 543 * Because preemptable RCU does not exist, it never has any callbacks
519 * to check. 544 * to check.
520 */ 545 */
521void rcu_preempt_check_callbacks(int cpu) 546static void rcu_preempt_check_callbacks(int cpu)
522{ 547{
523} 548}
524 549
@@ -526,7 +551,7 @@ void rcu_preempt_check_callbacks(int cpu)
526 * Because preemptable RCU does not exist, it never has any callbacks 551 * Because preemptable RCU does not exist, it never has any callbacks
527 * to process. 552 * to process.
528 */ 553 */
529void rcu_preempt_process_callbacks(void) 554static void rcu_preempt_process_callbacks(void)
530{ 555{
531} 556}
532 557
@@ -556,6 +581,16 @@ static int rcu_preempt_needs_cpu(int cpu)
556} 581}
557 582
558/* 583/*
584 * Because preemptable RCU does not exist, rcu_barrier() is just
585 * another name for rcu_barrier_sched().
586 */
587void rcu_barrier(void)
588{
589 rcu_barrier_sched();
590}
591EXPORT_SYMBOL_GPL(rcu_barrier);
592
593/*
559 * Because preemptable RCU does not exist, there is no per-CPU 594 * Because preemptable RCU does not exist, there is no per-CPU
560 * data to initialize. 595 * data to initialize.
561 */ 596 */
@@ -563,4 +598,18 @@ static void __cpuinit rcu_preempt_init_percpu_data(int cpu)
563{ 598{
564} 599}
565 600
601/*
602 * Because there is no preemptable RCU, there are no callbacks to move.
603 */
604static void rcu_preempt_send_cbs_to_orphanage(void)
605{
606}
607
608/*
609 * Because preemptable RCU does not exist, it need not be initialized.
610 */
611static void __init __rcu_init_preempt(void)
612{
613}
614
566#endif /* #else #ifdef CONFIG_TREE_PREEMPT_RCU */ 615#endif /* #else #ifdef CONFIG_TREE_PREEMPT_RCU */
diff --git a/kernel/rcutree_trace.c b/kernel/rcutree_trace.c
index 179e6ad80dc0..4b31c779e62e 100644
--- a/kernel/rcutree_trace.c
+++ b/kernel/rcutree_trace.c
@@ -159,13 +159,13 @@ static void print_one_rcu_state(struct seq_file *m, struct rcu_state *rsp)
159 struct rcu_node *rnp; 159 struct rcu_node *rnp;
160 160
161 seq_printf(m, "c=%ld g=%ld s=%d jfq=%ld j=%x " 161 seq_printf(m, "c=%ld g=%ld s=%d jfq=%ld j=%x "
162 "nfqs=%lu/nfqsng=%lu(%lu) fqlh=%lu\n", 162 "nfqs=%lu/nfqsng=%lu(%lu) fqlh=%lu oqlen=%ld\n",
163 rsp->completed, rsp->gpnum, rsp->signaled, 163 rsp->completed, rsp->gpnum, rsp->signaled,
164 (long)(rsp->jiffies_force_qs - jiffies), 164 (long)(rsp->jiffies_force_qs - jiffies),
165 (int)(jiffies & 0xffff), 165 (int)(jiffies & 0xffff),
166 rsp->n_force_qs, rsp->n_force_qs_ngp, 166 rsp->n_force_qs, rsp->n_force_qs_ngp,
167 rsp->n_force_qs - rsp->n_force_qs_ngp, 167 rsp->n_force_qs - rsp->n_force_qs_ngp,
168 rsp->n_force_qs_lh); 168 rsp->n_force_qs_lh, rsp->orphan_qlen);
169 for (rnp = &rsp->node[0]; rnp - &rsp->node[0] < NUM_RCU_NODES; rnp++) { 169 for (rnp = &rsp->node[0]; rnp - &rsp->node[0] < NUM_RCU_NODES; rnp++) {
170 if (rnp->level != level) { 170 if (rnp->level != level) {
171 seq_puts(m, "\n"); 171 seq_puts(m, "\n");
generated by cgit v1.2.2 (git 2.25.0) at 2024-12-15 06:43:31 -0500