blob: e51373c0b74840edf4380b47ca44840a2ad0dfc9 [file] [log] [blame]
Paul E. McKenney9f77da92009-08-22 13:56:45 -07001/*
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/*
Paul E. McKenney8932a632012-04-19 12:20:14 -070032 * Define shape of hierarchy based on NR_CPUS, CONFIG_RCU_FANOUT, and
33 * CONFIG_RCU_FANOUT_LEAF.
Paul E. McKenney9f77da92009-08-22 13:56:45 -070034 * In theory, it should be possible to add more levels straightforwardly.
Paul E. McKenney0209f642010-12-14 16:07:52 -080035 * In practice, this did work well going from three levels to four.
36 * Of course, your mileage may vary.
Paul E. McKenney9f77da92009-08-22 13:56:45 -070037 */
Paul E. McKenneycf244dc2009-12-02 12:10:14 -080038#define MAX_RCU_LVLS 4
Paul E. McKenney8932a632012-04-19 12:20:14 -070039#define RCU_FANOUT_1 (CONFIG_RCU_FANOUT_LEAF)
Paul E. McKenney0209f642010-12-14 16:07:52 -080040#define RCU_FANOUT_2 (RCU_FANOUT_1 * CONFIG_RCU_FANOUT)
41#define RCU_FANOUT_3 (RCU_FANOUT_2 * CONFIG_RCU_FANOUT)
42#define RCU_FANOUT_4 (RCU_FANOUT_3 * CONFIG_RCU_FANOUT)
Paul E. McKenney9f77da92009-08-22 13:56:45 -070043
Paul E. McKenney0209f642010-12-14 16:07:52 -080044#if NR_CPUS <= RCU_FANOUT_1
Paul E. McKenneyf885b7f2012-04-23 15:52:53 -070045# define RCU_NUM_LVLS 1
Paul E. McKenney9f77da92009-08-22 13:56:45 -070046# define NUM_RCU_LVL_0 1
47# define NUM_RCU_LVL_1 (NR_CPUS)
48# define NUM_RCU_LVL_2 0
49# define NUM_RCU_LVL_3 0
Paul E. McKenneycf244dc2009-12-02 12:10:14 -080050# define NUM_RCU_LVL_4 0
Paul E. McKenney0209f642010-12-14 16:07:52 -080051#elif NR_CPUS <= RCU_FANOUT_2
Paul E. McKenneyf885b7f2012-04-23 15:52:53 -070052# define RCU_NUM_LVLS 2
Paul E. McKenney9f77da92009-08-22 13:56:45 -070053# define NUM_RCU_LVL_0 1
Paul E. McKenney0209f642010-12-14 16:07:52 -080054# define NUM_RCU_LVL_1 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
Paul E. McKenney9f77da92009-08-22 13:56:45 -070055# define NUM_RCU_LVL_2 (NR_CPUS)
56# define NUM_RCU_LVL_3 0
Paul E. McKenneycf244dc2009-12-02 12:10:14 -080057# define NUM_RCU_LVL_4 0
Paul E. McKenney0209f642010-12-14 16:07:52 -080058#elif NR_CPUS <= RCU_FANOUT_3
Paul E. McKenneyf885b7f2012-04-23 15:52:53 -070059# define RCU_NUM_LVLS 3
Paul E. McKenney9f77da92009-08-22 13:56:45 -070060# define NUM_RCU_LVL_0 1
Paul E. McKenney0209f642010-12-14 16:07:52 -080061# define NUM_RCU_LVL_1 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_2)
62# define NUM_RCU_LVL_2 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
63# define NUM_RCU_LVL_3 (NR_CPUS)
Paul E. McKenneycf244dc2009-12-02 12:10:14 -080064# define NUM_RCU_LVL_4 0
Paul E. McKenney0209f642010-12-14 16:07:52 -080065#elif NR_CPUS <= RCU_FANOUT_4
Paul E. McKenneyf885b7f2012-04-23 15:52:53 -070066# define RCU_NUM_LVLS 4
Paul E. McKenneycf244dc2009-12-02 12:10:14 -080067# define NUM_RCU_LVL_0 1
Paul E. McKenney0209f642010-12-14 16:07:52 -080068# define NUM_RCU_LVL_1 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_3)
69# define NUM_RCU_LVL_2 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_2)
70# define NUM_RCU_LVL_3 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
71# define NUM_RCU_LVL_4 (NR_CPUS)
Paul E. McKenney9f77da92009-08-22 13:56:45 -070072#else
73# error "CONFIG_RCU_FANOUT insufficient for NR_CPUS"
Paul E. McKenney0209f642010-12-14 16:07:52 -080074#endif /* #if (NR_CPUS) <= RCU_FANOUT_1 */
Paul E. McKenney9f77da92009-08-22 13:56:45 -070075
Paul E. McKenneycf244dc2009-12-02 12:10:14 -080076#define RCU_SUM (NUM_RCU_LVL_0 + NUM_RCU_LVL_1 + NUM_RCU_LVL_2 + NUM_RCU_LVL_3 + NUM_RCU_LVL_4)
Paul E. McKenney9f77da92009-08-22 13:56:45 -070077#define NUM_RCU_NODES (RCU_SUM - NR_CPUS)
78
Paul E. McKenneyf885b7f2012-04-23 15:52:53 -070079extern int rcu_num_lvls;
80extern int rcu_num_nodes;
81
Paul E. McKenney9f77da92009-08-22 13:56:45 -070082/*
83 * Dynticks per-CPU state.
84 */
85struct rcu_dynticks {
Paul E. McKenney9b2e4f12011-09-30 12:10:22 -070086 long long dynticks_nesting; /* Track irq/process nesting level. */
87 /* Process level is worth LLONG_MAX/2. */
88 int dynticks_nmi_nesting; /* Track NMI nesting level. */
89 atomic_t dynticks; /* Even value for idle, else odd. */
Paul E. McKenney5955f7e2012-05-09 12:07:05 -070090#ifdef CONFIG_RCU_FAST_NO_HZ
91 int dyntick_drain; /* Prepare-for-idle state variable. */
92 unsigned long dyntick_holdoff;
93 /* No retries for the jiffy of failure. */
94 struct timer_list idle_gp_timer;
95 /* Wake up CPU sleeping with callbacks. */
96 unsigned long idle_gp_timer_expires;
97 /* When to wake up CPU (for repost). */
98 bool idle_first_pass; /* First pass of attempt to go idle? */
99 unsigned long nonlazy_posted;
100 /* # times non-lazy CBs posted to CPU. */
101 unsigned long nonlazy_posted_snap;
102 /* idle-period nonlazy_posted snapshot. */
Paul E. McKenney9d2ad242012-06-24 10:15:02 -0700103 int tick_nohz_enabled_snap; /* Previously seen value from sysfs. */
Paul E. McKenney5955f7e2012-05-09 12:07:05 -0700104#endif /* #ifdef CONFIG_RCU_FAST_NO_HZ */
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700105};
106
Paul E. McKenneyd71df902011-03-29 17:48:28 -0700107/* RCU's kthread states for tracing. */
108#define RCU_KTHREAD_STOPPED 0
109#define RCU_KTHREAD_RUNNING 1
110#define RCU_KTHREAD_WAITING 2
Paul E. McKenney15ba0ba2011-04-06 16:01:16 -0700111#define RCU_KTHREAD_OFFCPU 3
112#define RCU_KTHREAD_YIELDING 4
113#define RCU_KTHREAD_MAX 4
Paul E. McKenneyd71df902011-03-29 17:48:28 -0700114
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700115/*
116 * Definition for node within the RCU grace-period-detection hierarchy.
117 */
118struct rcu_node {
Paul E. McKenney1304afb2010-02-22 17:05:02 -0800119 raw_spinlock_t lock; /* Root rcu_node's lock protects some */
Paul E. McKenney1eba8f82009-09-23 09:50:42 -0700120 /* rcu_state fields as well as following. */
Paul E. McKenney20133cf2010-02-22 17:05:01 -0800121 unsigned long gpnum; /* Current grace period for this node. */
Paul E. McKenney86848962009-08-27 15:00:12 -0700122 /* This will either be equal to or one */
123 /* behind the root rcu_node's gpnum. */
Paul E. McKenney20133cf2010-02-22 17:05:01 -0800124 unsigned long completed; /* Last GP completed for this node. */
Paul E. McKenneyd09b62d2009-11-02 13:52:28 -0800125 /* This will either be equal to or one */
126 /* behind the root rcu_node's gpnum. */
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700127 unsigned long qsmask; /* CPUs or groups that need to switch in */
128 /* order for current grace period to proceed.*/
Paul E. McKenney1eba8f82009-09-23 09:50:42 -0700129 /* In leaf rcu_node, each bit corresponds to */
130 /* an rcu_data structure, otherwise, each */
131 /* bit corresponds to a child rcu_node */
132 /* structure. */
Paul E. McKenney12f5f522010-11-29 21:56:39 -0800133 unsigned long expmask; /* Groups that have ->blkd_tasks */
Paul E. McKenneyd9a3da02009-12-02 12:10:15 -0800134 /* elements that need to drain to allow the */
135 /* current expedited grace period to */
136 /* complete (only for TREE_PREEMPT_RCU). */
Paul E. McKenney8826f3b2011-05-11 05:41:41 -0700137 atomic_t wakemask; /* CPUs whose kthread needs to be awakened. */
138 /* Since this has meaning only for leaf */
139 /* rcu_node structures, 32 bits suffices. */
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700140 unsigned long qsmaskinit;
Paul E. McKenneyd9a3da02009-12-02 12:10:15 -0800141 /* Per-GP initial value for qsmask & expmask. */
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700142 unsigned long grpmask; /* Mask to apply to parent qsmask. */
Paul E. McKenney1eba8f82009-09-23 09:50:42 -0700143 /* Only one bit will be set in this mask. */
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700144 int grplo; /* lowest-numbered CPU or group here. */
145 int grphi; /* highest-numbered CPU or group here. */
146 u8 grpnum; /* CPU/group number for next level up. */
147 u8 level; /* root is at level 0. */
148 struct rcu_node *parent;
Paul E. McKenney12f5f522010-11-29 21:56:39 -0800149 struct list_head blkd_tasks;
150 /* Tasks blocked in RCU read-side critical */
151 /* section. Tasks are placed at the head */
152 /* of this list and age towards the tail. */
153 struct list_head *gp_tasks;
154 /* Pointer to the first task blocking the */
155 /* current grace period, or NULL if there */
156 /* is no such task. */
157 struct list_head *exp_tasks;
158 /* Pointer to the first task blocking the */
159 /* current expedited grace period, or NULL */
160 /* if there is no such task. If there */
161 /* is no current expedited grace period, */
162 /* then there can cannot be any such task. */
Paul E. McKenney27f4d282011-02-07 12:47:15 -0800163#ifdef CONFIG_RCU_BOOST
164 struct list_head *boost_tasks;
165 /* Pointer to first task that needs to be */
166 /* priority boosted, or NULL if no priority */
167 /* boosting is needed for this rcu_node */
168 /* structure. If there are no tasks */
169 /* queued on this rcu_node structure that */
170 /* are blocking the current grace period, */
171 /* there can be no such task. */
172 unsigned long boost_time;
173 /* When to start boosting (jiffies). */
174 struct task_struct *boost_kthread_task;
175 /* kthread that takes care of priority */
176 /* boosting for this rcu_node structure. */
Paul E. McKenneyd71df902011-03-29 17:48:28 -0700177 unsigned int boost_kthread_status;
178 /* State of boost_kthread_task for tracing. */
Paul E. McKenney0ea1f2e2011-02-22 13:42:43 -0800179 unsigned long n_tasks_boosted;
180 /* Total number of tasks boosted. */
181 unsigned long n_exp_boosts;
182 /* Number of tasks boosted for expedited GP. */
183 unsigned long n_normal_boosts;
184 /* Number of tasks boosted for normal GP. */
185 unsigned long n_balk_blkd_tasks;
186 /* Refused to boost: no blocked tasks. */
187 unsigned long n_balk_exp_gp_tasks;
188 /* Refused to boost: nothing blocking GP. */
189 unsigned long n_balk_boost_tasks;
190 /* Refused to boost: already boosting. */
191 unsigned long n_balk_notblocked;
192 /* Refused to boost: RCU RS CS still running. */
193 unsigned long n_balk_notyet;
194 /* Refused to boost: not yet time. */
195 unsigned long n_balk_nos;
196 /* Refused to boost: not sure why, though. */
197 /* This can happen due to race conditions. */
Paul E. McKenney27f4d282011-02-07 12:47:15 -0800198#endif /* #ifdef CONFIG_RCU_BOOST */
Paul E. McKenneydae6e642013-02-10 20:48:58 -0800199#ifdef CONFIG_RCU_NOCB_CPU
200 wait_queue_head_t nocb_gp_wq[2];
201 /* Place for rcu_nocb_kthread() to wait GP. */
202 int n_nocb_gp_requests[2];
203 /* Counts of upcoming no-CB GP requests. */
204#endif /* #ifdef CONFIG_RCU_NOCB_CPU */
Paul E. McKenney394f2762012-06-26 17:00:35 -0700205 raw_spinlock_t fqslock ____cacheline_internodealigned_in_smp;
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700206} ____cacheline_internodealigned_in_smp;
207
Paul E. McKenneya0b6c9a2009-09-28 07:46:33 -0700208/*
209 * Do a full breadth-first scan of the rcu_node structures for the
210 * specified rcu_state structure.
211 */
212#define rcu_for_each_node_breadth_first(rsp, rnp) \
213 for ((rnp) = &(rsp)->node[0]; \
Paul E. McKenneyf885b7f2012-04-23 15:52:53 -0700214 (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++)
Paul E. McKenneya0b6c9a2009-09-28 07:46:33 -0700215
Paul E. McKenneyd9a3da02009-12-02 12:10:15 -0800216/*
217 * Do a breadth-first scan of the non-leaf rcu_node structures for the
218 * specified rcu_state structure. Note that if there is a singleton
219 * rcu_node tree with but one rcu_node structure, this loop is a no-op.
220 */
221#define rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) \
222 for ((rnp) = &(rsp)->node[0]; \
Paul E. McKenneyf885b7f2012-04-23 15:52:53 -0700223 (rnp) < (rsp)->level[rcu_num_lvls - 1]; (rnp)++)
Paul E. McKenneyd9a3da02009-12-02 12:10:15 -0800224
225/*
226 * Scan the leaves of the rcu_node hierarchy for the specified rcu_state
227 * structure. Note that if there is a singleton rcu_node tree with but
228 * one rcu_node structure, this loop -will- visit the rcu_node structure.
229 * It is still a leaf node, even if it is also the root node.
230 */
Paul E. McKenneya0b6c9a2009-09-28 07:46:33 -0700231#define rcu_for_each_leaf_node(rsp, rnp) \
Paul E. McKenneyf885b7f2012-04-23 15:52:53 -0700232 for ((rnp) = (rsp)->level[rcu_num_lvls - 1]; \
233 (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++)
Paul E. McKenneya0b6c9a2009-09-28 07:46:33 -0700234
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700235/* Index values for nxttail array in struct rcu_data. */
236#define RCU_DONE_TAIL 0 /* Also RCU_WAIT head. */
237#define RCU_WAIT_TAIL 1 /* Also RCU_NEXT_READY head. */
238#define RCU_NEXT_READY_TAIL 2 /* Also RCU_NEXT head. */
239#define RCU_NEXT_TAIL 3
240#define RCU_NEXT_SIZE 4
241
242/* Per-CPU data for read-copy update. */
243struct rcu_data {
244 /* 1) quiescent-state and grace-period handling : */
Paul E. McKenney20133cf2010-02-22 17:05:01 -0800245 unsigned long completed; /* Track rsp->completed gp number */
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700246 /* in order to detect GP end. */
Paul E. McKenney20133cf2010-02-22 17:05:01 -0800247 unsigned long gpnum; /* Highest gp number that this CPU */
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700248 /* is aware of having started. */
Paul E. McKenneye4cc1f22011-06-27 00:17:43 -0700249 bool passed_quiesce; /* User-mode/idle loop etc. */
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700250 bool qs_pending; /* Core waits for quiesc state. */
251 bool beenonline; /* CPU online at least once. */
Paul E. McKenney6cc68792011-03-02 13:15:15 -0800252 bool preemptible; /* Preemptible RCU? */
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700253 struct rcu_node *mynode; /* This CPU's leaf of hierarchy */
254 unsigned long grpmask; /* Mask to apply to leaf qsmask. */
Paul E. McKenneya858af22012-01-16 13:29:10 -0800255#ifdef CONFIG_RCU_CPU_STALL_INFO
256 unsigned long ticks_this_gp; /* The number of scheduling-clock */
257 /* ticks this CPU has handled */
258 /* during and after the last grace */
259 /* period it is aware of. */
260#endif /* #ifdef CONFIG_RCU_CPU_STALL_INFO */
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700261
262 /* 2) batch handling */
263 /*
264 * If nxtlist is not NULL, it is partitioned as follows.
265 * Any of the partitions might be empty, in which case the
266 * pointer to that partition will be equal to the pointer for
267 * the following partition. When the list is empty, all of
Paul E. McKenney1eba8f82009-09-23 09:50:42 -0700268 * the nxttail elements point to the ->nxtlist pointer itself,
269 * which in that case is NULL.
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700270 *
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700271 * [nxtlist, *nxttail[RCU_DONE_TAIL]):
272 * Entries that batch # <= ->completed
273 * The grace period for these entries has completed, and
274 * the other grace-period-completed entries may be moved
275 * here temporarily in rcu_process_callbacks().
Paul E. McKenney1eba8f82009-09-23 09:50:42 -0700276 * [*nxttail[RCU_DONE_TAIL], *nxttail[RCU_WAIT_TAIL]):
277 * Entries that batch # <= ->completed - 1: waiting for current GP
278 * [*nxttail[RCU_WAIT_TAIL], *nxttail[RCU_NEXT_READY_TAIL]):
279 * Entries known to have arrived before current GP ended
280 * [*nxttail[RCU_NEXT_READY_TAIL], *nxttail[RCU_NEXT_TAIL]):
281 * Entries that might have arrived after current GP ended
282 * Note that the value of *nxttail[RCU_NEXT_TAIL] will
283 * always be NULL, as this is the end of the list.
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700284 */
285 struct rcu_head *nxtlist;
286 struct rcu_head **nxttail[RCU_NEXT_SIZE];
Paul E. McKenneydc35c892012-12-03 13:52:00 -0800287 unsigned long nxtcompleted[RCU_NEXT_SIZE];
288 /* grace periods for sublists. */
Paul E. McKenney486e2592012-01-06 14:11:30 -0800289 long qlen_lazy; /* # of lazy queued callbacks */
290 long qlen; /* # of queued callbacks, incl lazy */
Paul E. McKenney37c72e52009-10-14 10:15:55 -0700291 long qlen_last_fqs_check;
292 /* qlen at last check for QS forcing */
Paul E. McKenney269dcc12010-09-07 14:23:09 -0700293 unsigned long n_cbs_invoked; /* count of RCU cbs invoked. */
Paul E. McKenneyc635a4e12012-10-29 07:29:20 -0700294 unsigned long n_nocbs_invoked; /* count of no-CBs RCU cbs invoked. */
Lai Jiangshan29494be2010-10-20 14:13:06 +0800295 unsigned long n_cbs_orphaned; /* RCU cbs orphaned by dying CPU */
296 unsigned long n_cbs_adopted; /* RCU cbs adopted from dying CPU */
Paul E. McKenney37c72e52009-10-14 10:15:55 -0700297 unsigned long n_force_qs_snap;
298 /* did other CPU force QS recently? */
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700299 long blimit; /* Upper limit on a processed batch */
300
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700301 /* 3) dynticks interface. */
302 struct rcu_dynticks *dynticks; /* Shared per-CPU dynticks state. */
303 int dynticks_snap; /* Per-GP tracking for dynticks. */
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700304
305 /* 4) reasons this CPU needed to be kicked by force_quiescent_state */
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700306 unsigned long dynticks_fqs; /* Kicked due to dynticks idle. */
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700307 unsigned long offline_fqs; /* Kicked due to being offline. */
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700308
309 /* 5) __rcu_pending() statistics. */
Paul E. McKenney20133cf2010-02-22 17:05:01 -0800310 unsigned long n_rcu_pending; /* rcu_pending() calls since boot. */
311 unsigned long n_rp_qs_pending;
Paul E. McKenneyd21670a2010-04-14 17:39:26 -0700312 unsigned long n_rp_report_qs;
Paul E. McKenney20133cf2010-02-22 17:05:01 -0800313 unsigned long n_rp_cb_ready;
314 unsigned long n_rp_cpu_needs_gp;
315 unsigned long n_rp_gp_completed;
316 unsigned long n_rp_gp_started;
Paul E. McKenney20133cf2010-02-22 17:05:01 -0800317 unsigned long n_rp_need_nothing;
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700318
Paul E. McKenneyb626c1b2012-06-11 17:39:43 -0700319 /* 6) _rcu_barrier() and OOM callbacks. */
Paul E. McKenney06668ef2012-05-28 23:57:46 -0700320 struct rcu_head barrier_head;
Paul E. McKenneyb626c1b2012-06-11 17:39:43 -0700321#ifdef CONFIG_RCU_FAST_NO_HZ
322 struct rcu_head oom_head;
323#endif /* #ifdef CONFIG_RCU_FAST_NO_HZ */
Paul E. McKenney06668ef2012-05-28 23:57:46 -0700324
Paul E. McKenney3fbfbf72012-08-19 21:35:53 -0700325 /* 7) Callback offloading. */
326#ifdef CONFIG_RCU_NOCB_CPU
327 struct rcu_head *nocb_head; /* CBs waiting for kthread. */
328 struct rcu_head **nocb_tail;
329 atomic_long_t nocb_q_count; /* # CBs waiting for kthread */
330 atomic_long_t nocb_q_count_lazy; /* (approximate). */
331 int nocb_p_count; /* # CBs being invoked by kthread */
332 int nocb_p_count_lazy; /* (approximate). */
333 wait_queue_head_t nocb_wq; /* For nocb kthreads to sleep on. */
334 struct task_struct *nocb_kthread;
335#endif /* #ifdef CONFIG_RCU_NOCB_CPU */
336
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700337 int cpu;
Paul E. McKenneyd4c08f22011-06-25 06:36:56 -0700338 struct rcu_state *rsp;
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700339};
340
Paul E. McKenneyaf446b72011-09-10 21:54:08 -0700341/* Values for fqs_state field in struct rcu_state. */
Paul E. McKenney83f5b012009-10-28 08:14:49 -0700342#define RCU_GP_IDLE 0 /* No grace period in progress. */
343#define RCU_GP_INIT 1 /* Grace period being initialized. */
344#define RCU_SAVE_DYNTICK 2 /* Need to scan dyntick state. */
Paul E. McKenneyee47eb92010-01-04 15:09:07 -0800345#define RCU_FORCE_QS 3 /* Need to force quiescent state. */
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700346#define RCU_SIGNAL_INIT RCU_SAVE_DYNTICK
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700347
348#define RCU_JIFFIES_TILL_FORCE_QS 3 /* for rsp->jiffies_force_qs */
Paul E. McKenney007b0922010-03-05 15:03:26 -0800349
Paul E. McKenney007b0922010-03-05 15:03:26 -0800350#define RCU_STALL_RAT_DELAY 2 /* Allow other CPUs time */
351 /* to take at least one */
352 /* scheduling clock irq */
353 /* before ratting on them. */
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700354
Peter Zijlstra08bca602011-05-20 16:06:29 -0700355#define rcu_wait(cond) \
356do { \
357 for (;;) { \
358 set_current_state(TASK_INTERRUPTIBLE); \
359 if (cond) \
360 break; \
361 schedule(); \
362 } \
363 __set_current_state(TASK_RUNNING); \
364} while (0)
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700365
366/*
367 * RCU global state, including node hierarchy. This hierarchy is
368 * represented in "heap" form in a dense array. The root (first level)
369 * of the hierarchy is in ->node[0] (referenced by ->level[0]), the second
370 * level in ->node[1] through ->node[m] (->node[1] referenced by ->level[1]),
371 * and the third level in ->node[m+1] and following (->node[m+1] referenced
372 * by ->level[2]). The number of levels is determined by the number of
373 * CPUs and by CONFIG_RCU_FANOUT. Small systems will have a "hierarchy"
374 * consisting of a single rcu_node.
375 */
376struct rcu_state {
377 struct rcu_node node[NUM_RCU_NODES]; /* Hierarchy. */
Paul E. McKenneyf885b7f2012-04-23 15:52:53 -0700378 struct rcu_node *level[RCU_NUM_LVLS]; /* Hierarchy levels. */
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700379 u32 levelcnt[MAX_RCU_LVLS + 1]; /* # nodes in each level. */
Paul E. McKenneyf885b7f2012-04-23 15:52:53 -0700380 u8 levelspread[RCU_NUM_LVLS]; /* kids/node in each level. */
Lai Jiangshan394f99a2010-06-28 16:25:04 +0800381 struct rcu_data __percpu *rda; /* pointer of percu rcu_data. */
Paul E. McKenney037b64e2012-05-28 23:26:01 -0700382 void (*call)(struct rcu_head *head, /* call_rcu() flavor. */
383 void (*func)(struct rcu_head *head));
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700384
385 /* The following fields are guarded by the root rcu_node's lock. */
386
Paul E. McKenneyaf446b72011-09-10 21:54:08 -0700387 u8 fqs_state ____cacheline_internodealigned_in_smp;
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700388 /* Force QS state. */
Paul E. McKenneya46e0892011-06-15 15:47:09 -0700389 u8 boost; /* Subject to priority boost. */
Paul E. McKenney20133cf2010-02-22 17:05:01 -0800390 unsigned long gpnum; /* Current gp number. */
391 unsigned long completed; /* # of last completed gp. */
Paul E. McKenneyb3dbec72012-06-18 18:36:08 -0700392 struct task_struct *gp_kthread; /* Task for grace periods. */
393 wait_queue_head_t gp_wq; /* Where GP task waits. */
394 int gp_flags; /* Commands for GP task. */
Paul E. McKenney1eba8f82009-09-23 09:50:42 -0700395
Paul E. McKenneyd9a3da02009-12-02 12:10:15 -0800396 /* End of fields guarded by root rcu_node's lock. */
Paul E. McKenney1eba8f82009-09-23 09:50:42 -0700397
Paul E. McKenney7b2e6012012-10-08 10:54:03 -0700398 raw_spinlock_t orphan_lock ____cacheline_internodealigned_in_smp;
399 /* Protect following fields. */
Paul E. McKenneyb1420f12012-03-01 13:18:08 -0800400 struct rcu_head *orphan_nxtlist; /* Orphaned callbacks that */
401 /* need a grace period. */
402 struct rcu_head **orphan_nxttail; /* Tail of above. */
403 struct rcu_head *orphan_donelist; /* Orphaned callbacks that */
404 /* are ready to invoke. */
405 struct rcu_head **orphan_donetail; /* Tail of above. */
406 long qlen_lazy; /* Number of lazy callbacks. */
407 long qlen; /* Total number of callbacks. */
Paul E. McKenney7b2e6012012-10-08 10:54:03 -0700408 /* End of fields guarded by orphan_lock. */
Paul E. McKenneya4fbe352012-10-07 08:36:12 -0700409
410 struct mutex onoff_mutex; /* Coordinate hotplug & GPs. */
411
Paul E. McKenney7be7f0b2012-05-29 05:18:53 -0700412 struct mutex barrier_mutex; /* Guards barrier fields. */
Paul E. McKenney24ebbca2012-05-29 00:34:56 -0700413 atomic_t barrier_cpu_count; /* # CPUs waiting on. */
Paul E. McKenney7db74df2012-05-29 03:03:37 -0700414 struct completion barrier_completion; /* Wake at barrier end. */
Paul E. McKenneycf3a9c42012-05-29 14:56:46 -0700415 unsigned long n_barrier_done; /* ++ at start and end of */
416 /* _rcu_barrier(). */
Paul E. McKenneya4fbe352012-10-07 08:36:12 -0700417 /* End of fields guarded by barrier_mutex. */
418
Paul E. McKenney40694d62012-10-11 15:24:03 -0700419 atomic_long_t expedited_start; /* Starting ticket. */
420 atomic_long_t expedited_done; /* Done ticket. */
Paul E. McKenneya30489c2012-10-11 16:18:09 -0700421 atomic_long_t expedited_wrap; /* # near-wrap incidents. */
422 atomic_long_t expedited_tryfail; /* # acquisition failures. */
423 atomic_long_t expedited_workdone1; /* # done by others #1. */
424 atomic_long_t expedited_workdone2; /* # done by others #2. */
425 atomic_long_t expedited_normal; /* # fallbacks to normal. */
426 atomic_long_t expedited_stoppedcpus; /* # successful stop_cpus. */
427 atomic_long_t expedited_done_tries; /* # tries to update _done. */
428 atomic_long_t expedited_done_lost; /* # times beaten to _done. */
429 atomic_long_t expedited_done_exit; /* # times exited _done loop. */
Paul E. McKenney40694d62012-10-11 15:24:03 -0700430
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700431 unsigned long jiffies_force_qs; /* Time at which to invoke */
432 /* force_quiescent_state(). */
433 unsigned long n_force_qs; /* Number of calls to */
434 /* force_quiescent_state(). */
435 unsigned long n_force_qs_lh; /* ~Number of calls leaving */
436 /* due to lock unavailable. */
437 unsigned long n_force_qs_ngp; /* Number of calls leaving */
438 /* due to no GP active. */
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700439 unsigned long gp_start; /* Time at which GP started, */
440 /* but in jiffies. */
441 unsigned long jiffies_stall; /* Time at which to check */
442 /* for CPU stalls. */
Paul E. McKenney15ba0ba2011-04-06 16:01:16 -0700443 unsigned long gp_max; /* Maximum GP duration in */
444 /* jiffies. */
Paul E. McKenney4300aa62010-04-13 16:18:22 -0700445 char *name; /* Name of structure. */
Paul E. McKenney6ce75a22012-06-12 11:01:13 -0700446 struct list_head flavors; /* List of RCU flavors. */
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700447};
448
Paul E. McKenney4cdfc172012-06-22 17:06:26 -0700449/* Values for rcu_state structure's gp_flags field. */
450#define RCU_GP_FLAG_INIT 0x1 /* Need grace-period initialization. */
451#define RCU_GP_FLAG_FQS 0x2 /* Need grace-period quiescent-state forcing. */
452
Paul E. McKenney6ce75a22012-06-12 11:01:13 -0700453extern struct list_head rcu_struct_flavors;
Paul E. McKenney3fbfbf72012-08-19 21:35:53 -0700454
455/* Sequence through rcu_state structures for each RCU flavor. */
Paul E. McKenney6ce75a22012-06-12 11:01:13 -0700456#define for_each_rcu_flavor(rsp) \
457 list_for_each_entry((rsp), &rcu_struct_flavors, flavors)
458
Paul E. McKenneyd9a3da02009-12-02 12:10:15 -0800459/* Return values for rcu_preempt_offline_tasks(). */
460
461#define RCU_OFL_TASKS_NORM_GP 0x1 /* Tasks blocking normal */
462 /* GP were moved to root. */
463#define RCU_OFL_TASKS_EXP_GP 0x2 /* Tasks blocking expedited */
464 /* GP were moved to root. */
465
Ingo Molnar6258c4f2009-03-25 16:42:24 +0100466/*
467 * RCU implementation internal declarations:
468 */
Paul E. McKenneyd6714c22009-08-22 13:56:46 -0700469extern struct rcu_state rcu_sched_state;
470DECLARE_PER_CPU(struct rcu_data, rcu_sched_data);
Ingo Molnar6258c4f2009-03-25 16:42:24 +0100471
472extern struct rcu_state rcu_bh_state;
473DECLARE_PER_CPU(struct rcu_data, rcu_bh_data);
474
Paul E. McKenneyf41d9112009-08-22 13:56:52 -0700475#ifdef CONFIG_TREE_PREEMPT_RCU
476extern struct rcu_state rcu_preempt_state;
477DECLARE_PER_CPU(struct rcu_data, rcu_preempt_data);
478#endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */
479
Paul E. McKenneyeab09932011-06-21 01:59:33 -0700480#ifdef CONFIG_RCU_BOOST
481DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_status);
482DECLARE_PER_CPU(int, rcu_cpu_kthread_cpu);
483DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_loops);
484DECLARE_PER_CPU(char, rcu_cpu_has_work);
485#endif /* #ifdef CONFIG_RCU_BOOST */
486
Paul E. McKenney017c4262010-01-14 16:10:58 -0800487#ifndef RCU_TREE_NONCORE
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700488
Paul E. McKenney9b2619a2009-09-23 09:50:43 -0700489/* Forward declarations for rcutree_plugin.h */
Paul E. McKenneydbe01352009-11-10 13:37:19 -0800490static void rcu_bootup_announce(void);
Paul E. McKenney9b2619a2009-09-23 09:50:43 -0700491long rcu_batches_completed(void);
Paul E. McKenneycba6d0d2012-07-02 07:08:42 -0700492static void rcu_preempt_note_context_switch(int cpu);
Paul E. McKenney27f4d282011-02-07 12:47:15 -0800493static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp);
Paul E. McKenneyb668c9c2009-11-22 08:53:48 -0800494#ifdef CONFIG_HOTPLUG_CPU
Paul E. McKenneyd3f6bad2009-12-02 12:10:13 -0800495static void rcu_report_unblock_qs_rnp(struct rcu_node *rnp,
496 unsigned long flags);
Paul E. McKenneyb668c9c2009-11-22 08:53:48 -0800497#endif /* #ifdef CONFIG_HOTPLUG_CPU */
Paul E. McKenney1ed509a2010-02-22 17:05:05 -0800498static void rcu_print_detail_task_stall(struct rcu_state *rsp);
Paul E. McKenney9bc8b552011-08-13 13:31:47 -0700499static int rcu_print_task_stall(struct rcu_node *rnp);
Paul E. McKenney9b2619a2009-09-23 09:50:43 -0700500static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp);
501#ifdef CONFIG_HOTPLUG_CPU
Paul E. McKenney237c80c2009-10-15 09:26:14 -0700502static int rcu_preempt_offline_tasks(struct rcu_state *rsp,
503 struct rcu_node *rnp,
504 struct rcu_data *rdp);
Paul E. McKenney9b2619a2009-09-23 09:50:43 -0700505#endif /* #ifdef CONFIG_HOTPLUG_CPU */
506static void rcu_preempt_check_callbacks(int cpu);
Paul E. McKenney9b2619a2009-09-23 09:50:43 -0700507void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu));
Paul E. McKenneyd9a3da02009-12-02 12:10:15 -0800508#if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_TREE_PREEMPT_RCU)
Thomas Gleixnerb40d2932011-10-22 07:12:34 -0700509static void rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp,
510 bool wake);
Paul E. McKenneyd9a3da02009-12-02 12:10:15 -0800511#endif /* #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_TREE_PREEMPT_RCU) */
Paul E. McKenney9b2619a2009-09-23 09:50:43 -0700512static void __init __rcu_init_preempt(void);
Paul E. McKenney1217ed12011-05-04 21:43:49 -0700513static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags);
Paul E. McKenneya46e0892011-06-15 15:47:09 -0700514static void rcu_preempt_boost_start_gp(struct rcu_node *rnp);
515static void invoke_rcu_callbacks_kthread(void);
Paul E. McKenneydff16722011-11-29 15:57:13 -0800516static bool rcu_is_callbacks_kthread(void);
Paul E. McKenneya46e0892011-06-15 15:47:09 -0700517#ifdef CONFIG_RCU_BOOST
518static void rcu_preempt_do_callbacks(void);
Paul E. McKenney27f4d282011-02-07 12:47:15 -0800519static int __cpuinit rcu_spawn_one_boost_kthread(struct rcu_state *rsp,
Thomas Gleixner5d01bbd2012-07-16 10:42:35 +0000520 struct rcu_node *rnp);
Paul E. McKenneya46e0892011-06-15 15:47:09 -0700521#endif /* #ifdef CONFIG_RCU_BOOST */
Paul E. McKenneyf8b7fc62011-06-16 08:26:32 -0700522static void __cpuinit rcu_prepare_kthreads(int cpu);
Paul E. McKenney7cb92492011-11-28 12:28:34 -0800523static void rcu_prepare_for_idle_init(int cpu);
524static void rcu_cleanup_after_idle(int cpu);
Paul E. McKenneyaea1b352011-11-02 06:54:54 -0700525static void rcu_prepare_for_idle(int cpu);
Paul E. McKenneyc57afe82012-02-28 11:02:21 -0800526static void rcu_idle_count_callbacks_posted(void);
Paul E. McKenneya858af22012-01-16 13:29:10 -0800527static void print_cpu_stall_info_begin(void);
528static void print_cpu_stall_info(struct rcu_state *rsp, int cpu);
529static void print_cpu_stall_info_end(void);
530static void zero_cpu_stall_ticks(struct rcu_data *rdp);
531static void increment_cpu_stall_ticks(void);
Paul E. McKenneydae6e642013-02-10 20:48:58 -0800532static int rcu_nocb_needs_gp(struct rcu_state *rsp);
533static void rcu_nocb_gp_set(struct rcu_node *rnp, int nrq);
534static int rcu_nocb_gp_cleanup(struct rcu_state *rsp, struct rcu_node *rnp);
535static void rcu_init_one_nocb(struct rcu_node *rnp);
Paul E. McKenney3fbfbf72012-08-19 21:35:53 -0700536static bool is_nocb_cpu(int cpu);
537static bool __call_rcu_nocb(struct rcu_data *rdp, struct rcu_head *rhp,
538 bool lazy);
539static bool rcu_nocb_adopt_orphan_cbs(struct rcu_state *rsp,
540 struct rcu_data *rdp);
Paul E. McKenney3fbfbf72012-08-19 21:35:53 -0700541static void rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp);
542static void rcu_spawn_nocb_kthreads(struct rcu_state *rsp);
Paul E. McKenney34ed62462013-01-07 13:37:42 -0800543static bool init_nocb_callback_list(struct rcu_data *rdp);
Paul E. McKenney9b2619a2009-09-23 09:50:43 -0700544
Paul E. McKenney017c4262010-01-14 16:10:58 -0800545#endif /* #ifndef RCU_TREE_NONCORE */
Paul E. McKenney3fbfbf72012-08-19 21:35:53 -0700546
547#ifdef CONFIG_RCU_TRACE
548#ifdef CONFIG_RCU_NOCB_CPU
549/* Sum up queue lengths for tracing. */
550static inline void rcu_nocb_q_lengths(struct rcu_data *rdp, long *ql, long *qll)
551{
552 *ql = atomic_long_read(&rdp->nocb_q_count) + rdp->nocb_p_count;
553 *qll = atomic_long_read(&rdp->nocb_q_count_lazy) + rdp->nocb_p_count_lazy;
554}
555#else /* #ifdef CONFIG_RCU_NOCB_CPU */
556static inline void rcu_nocb_q_lengths(struct rcu_data *rdp, long *ql, long *qll)
557{
558 *ql = 0;
559 *qll = 0;
560}
561#endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */
562#endif /* #ifdef CONFIG_RCU_TRACE */