diff options
Diffstat (limited to 'kernel/rcutree.c')
-rw-r--r-- | kernel/rcutree.c | 330 |
1 files changed, 216 insertions, 114 deletions
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 | ||
53 | static struct lock_class_key rcu_lock_key; | ||
54 | struct lockdep_map rcu_lock_map = | ||
55 | STATIC_LOCKDEP_MAP_INIT("rcu_read_lock", &rcu_lock_key); | ||
56 | EXPORT_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); | |||
81 | struct rcu_state rcu_bh_state = RCU_STATE_INITIALIZER(rcu_bh_state); | 77 | struct rcu_state rcu_bh_state = RCU_STATE_INITIALIZER(rcu_bh_state); |
82 | DEFINE_PER_CPU(struct rcu_data, rcu_bh_data); | 78 | DEFINE_PER_CPU(struct rcu_data, rcu_bh_data); |
83 | 79 | ||
84 | extern long rcu_batches_completed_sched(void); | ||
85 | static struct rcu_node *rcu_get_root(struct rcu_state *rsp); | ||
86 | static void cpu_quiet_msk(unsigned long mask, struct rcu_state *rsp, | ||
87 | struct rcu_node *rnp, unsigned long flags); | ||
88 | static void cpu_quiet_msk_finish(struct rcu_state *rsp, unsigned long flags); | ||
89 | #ifdef CONFIG_HOTPLUG_CPU | ||
90 | static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp); | ||
91 | #endif /* #ifdef CONFIG_HOTPLUG_CPU */ | ||
92 | static void __rcu_process_callbacks(struct rcu_state *rsp, | ||
93 | struct rcu_data *rdp); | ||
94 | static void __call_rcu(struct rcu_head *head, | ||
95 | void (*func)(struct rcu_head *rcu), | ||
96 | struct rcu_state *rsp); | ||
97 | static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp); | ||
98 | static 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 | */ | ||
86 | static 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. */ | |||
137 | static int qhimark = 10000; /* If this many pending, ignore blimit. */ | 125 | static int qhimark = 10000; /* If this many pending, ignore blimit. */ |
138 | static int qlowmark = 100; /* Once only this many pending, use blimit. */ | 126 | static int qlowmark = 100; /* Once only this many pending, use blimit. */ |
139 | 127 | ||
128 | module_param(blimit, int, 0); | ||
129 | module_param(qhimark, int, 0); | ||
130 | module_param(qlowmark, int, 0); | ||
131 | |||
140 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed); | 132 | static void force_quiescent_state(struct rcu_state *rsp, int relaxed); |
141 | static int rcu_pending(int cpu); | 133 | static int rcu_pending(int cpu); |
142 | 134 | ||
@@ -173,9 +165,7 @@ cpu_has_callbacks_ready_to_invoke(struct rcu_data *rdp) | |||
173 | static int | 165 | static int |
174 | cpu_needs_another_gp(struct rcu_state *rsp, struct rcu_data *rdp) | 166 | cpu_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 | */ |
374 | static int dyntick_save_progress_counter(struct rcu_data *rdp) | 364 | static 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 | */ |
705 | static void cpu_quiet_msk_finish(struct rcu_state *rsp, unsigned long flags) | 704 | static 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 | */ | ||
850 | static 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 | */ | ||
871 | static 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 | */ |
848 | static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp) | 895 | static 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 | ||
943 | static void rcu_send_cbs_to_orphanage(struct rcu_state *rsp) | ||
944 | { | ||
945 | } | ||
946 | |||
947 | static void rcu_adopt_orphan_cbs(struct rcu_state *rsp) | ||
948 | { | ||
949 | } | ||
950 | |||
921 | static void rcu_offline_cpu(int cpu) | 951 | static 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 | ||
1400 | static DEFINE_PER_CPU(struct rcu_head, rcu_barrier_head) = {NULL}; | ||
1401 | static atomic_t rcu_barrier_cpu_count; | ||
1402 | static DEFINE_MUTEX(rcu_barrier_mutex); | ||
1403 | static struct completion rcu_barrier_completion; | ||
1404 | |||
1405 | static 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 | */ | ||
1414 | static 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 | */ | ||
1430 | static 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 | */ | ||
1461 | void rcu_barrier_bh(void) | ||
1462 | { | ||
1463 | _rcu_barrier(&rcu_bh_state, call_rcu_bh); | ||
1464 | } | ||
1465 | EXPORT_SYMBOL_GPL(rcu_barrier_bh); | ||
1466 | |||
1467 | /** | ||
1468 | * rcu_barrier_sched - Wait for in-flight call_rcu_sched() callbacks. | ||
1469 | */ | ||
1470 | void rcu_barrier_sched(void) | ||
1471 | { | ||
1472 | _rcu_barrier(&rcu_sched_state, call_rcu_sched); | ||
1473 | } | ||
1474 | EXPORT_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) \ |
1560 | do { \ | 1683 | do { \ |
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 | |||
1575 | void __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 | |||
1586 | void __init __rcu_init_preempt(void) | ||
1587 | { | ||
1588 | } | ||
1589 | |||
1590 | #endif /* #else #ifdef CONFIG_TREE_PREEMPT_RCU */ | ||
1591 | |||
1592 | void __init __rcu_init(void) | 1700 | void __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 | ||
1608 | module_param(blimit, int, 0); | 1712 | #include "rcutree_plugin.h" |
1609 | module_param(qhimark, int, 0); | ||
1610 | module_param(qlowmark, int, 0); | ||