blob: 5f97eab602cd831dec54baaf7bafd970920d0c60 [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>
Steven Rostedt016a8d52013-05-28 17:32:53 -040030#include <linux/irq_work.h>
Paul E. McKenney9f77da92009-08-22 13:56:45 -070031
32/*
Paul E. McKenney8932a632012-04-19 12:20:14 -070033 * Define shape of hierarchy based on NR_CPUS, CONFIG_RCU_FANOUT, and
34 * CONFIG_RCU_FANOUT_LEAF.
Paul E. McKenney9f77da92009-08-22 13:56:45 -070035 * In theory, it should be possible to add more levels straightforwardly.
Paul E. McKenney0209f642010-12-14 16:07:52 -080036 * In practice, this did work well going from three levels to four.
37 * Of course, your mileage may vary.
Paul E. McKenney9f77da92009-08-22 13:56:45 -070038 */
Paul E. McKenneycf244dc2009-12-02 12:10:14 -080039#define MAX_RCU_LVLS 4
Paul E. McKenney8932a632012-04-19 12:20:14 -070040#define RCU_FANOUT_1 (CONFIG_RCU_FANOUT_LEAF)
Paul E. McKenney0209f642010-12-14 16:07:52 -080041#define RCU_FANOUT_2 (RCU_FANOUT_1 * CONFIG_RCU_FANOUT)
42#define RCU_FANOUT_3 (RCU_FANOUT_2 * CONFIG_RCU_FANOUT)
43#define RCU_FANOUT_4 (RCU_FANOUT_3 * CONFIG_RCU_FANOUT)
Paul E. McKenney9f77da92009-08-22 13:56:45 -070044
Paul E. McKenney0209f642010-12-14 16:07:52 -080045#if NR_CPUS <= RCU_FANOUT_1
Paul E. McKenneyf885b7f2012-04-23 15:52:53 -070046# define RCU_NUM_LVLS 1
Paul E. McKenney9f77da92009-08-22 13:56:45 -070047# define NUM_RCU_LVL_0 1
48# define NUM_RCU_LVL_1 (NR_CPUS)
49# define NUM_RCU_LVL_2 0
50# define NUM_RCU_LVL_3 0
Paul E. McKenneycf244dc2009-12-02 12:10:14 -080051# define NUM_RCU_LVL_4 0
Paul E. McKenney0209f642010-12-14 16:07:52 -080052#elif NR_CPUS <= RCU_FANOUT_2
Paul E. McKenneyf885b7f2012-04-23 15:52:53 -070053# define RCU_NUM_LVLS 2
Paul E. McKenney9f77da92009-08-22 13:56:45 -070054# define NUM_RCU_LVL_0 1
Paul E. McKenney0209f642010-12-14 16:07:52 -080055# define NUM_RCU_LVL_1 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
Paul E. McKenney9f77da92009-08-22 13:56:45 -070056# define NUM_RCU_LVL_2 (NR_CPUS)
57# define NUM_RCU_LVL_3 0
Paul E. McKenneycf244dc2009-12-02 12:10:14 -080058# define NUM_RCU_LVL_4 0
Paul E. McKenney0209f642010-12-14 16:07:52 -080059#elif NR_CPUS <= RCU_FANOUT_3
Paul E. McKenneyf885b7f2012-04-23 15:52:53 -070060# define RCU_NUM_LVLS 3
Paul E. McKenney9f77da92009-08-22 13:56:45 -070061# define NUM_RCU_LVL_0 1
Paul E. McKenney0209f642010-12-14 16:07:52 -080062# define NUM_RCU_LVL_1 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_2)
63# define NUM_RCU_LVL_2 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
64# define NUM_RCU_LVL_3 (NR_CPUS)
Paul E. McKenneycf244dc2009-12-02 12:10:14 -080065# define NUM_RCU_LVL_4 0
Paul E. McKenney0209f642010-12-14 16:07:52 -080066#elif NR_CPUS <= RCU_FANOUT_4
Paul E. McKenneyf885b7f2012-04-23 15:52:53 -070067# define RCU_NUM_LVLS 4
Paul E. McKenneycf244dc2009-12-02 12:10:14 -080068# define NUM_RCU_LVL_0 1
Paul E. McKenney0209f642010-12-14 16:07:52 -080069# define NUM_RCU_LVL_1 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_3)
70# define NUM_RCU_LVL_2 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_2)
71# define NUM_RCU_LVL_3 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1)
72# define NUM_RCU_LVL_4 (NR_CPUS)
Paul E. McKenney9f77da92009-08-22 13:56:45 -070073#else
74# error "CONFIG_RCU_FANOUT insufficient for NR_CPUS"
Paul E. McKenney0209f642010-12-14 16:07:52 -080075#endif /* #if (NR_CPUS) <= RCU_FANOUT_1 */
Paul E. McKenney9f77da92009-08-22 13:56:45 -070076
Paul E. McKenneycf244dc2009-12-02 12:10:14 -080077#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 -070078#define NUM_RCU_NODES (RCU_SUM - NR_CPUS)
79
Paul E. McKenneyf885b7f2012-04-23 15:52:53 -070080extern int rcu_num_lvls;
81extern int rcu_num_nodes;
82
Paul E. McKenney9f77da92009-08-22 13:56:45 -070083/*
84 * Dynticks per-CPU state.
85 */
86struct rcu_dynticks {
Paul E. McKenney9b2e4f12011-09-30 12:10:22 -070087 long long dynticks_nesting; /* Track irq/process nesting level. */
88 /* Process level is worth LLONG_MAX/2. */
89 int dynticks_nmi_nesting; /* Track NMI nesting level. */
90 atomic_t dynticks; /* Even value for idle, else odd. */
Paul E. McKenney23332102013-06-21 12:34:33 -070091#ifdef CONFIG_NO_HZ_FULL_SYSIDLE
92 long long dynticks_idle_nesting;
93 /* irq/process nesting level from idle. */
94 atomic_t dynticks_idle; /* Even value for idle, else odd. */
95 /* "Idle" excludes userspace execution. */
96 unsigned long dynticks_idle_jiffies;
97 /* End of last non-NMI non-idle period. */
98#endif /* #ifdef CONFIG_NO_HZ_FULL_SYSIDLE */
Paul E. McKenney5955f7e2012-05-09 12:07:05 -070099#ifdef CONFIG_RCU_FAST_NO_HZ
Paul E. McKenneyc0f4dfd2012-12-28 11:30:36 -0800100 bool all_lazy; /* Are all CPU's CBs lazy? */
Paul E. McKenney5955f7e2012-05-09 12:07:05 -0700101 unsigned long nonlazy_posted;
102 /* # times non-lazy CBs posted to CPU. */
103 unsigned long nonlazy_posted_snap;
104 /* idle-period nonlazy_posted snapshot. */
Paul E. McKenneyc0f4dfd2012-12-28 11:30:36 -0800105 unsigned long last_accelerate;
106 /* Last jiffy CBs were accelerated. */
Paul E. McKenney9d2ad242012-06-24 10:15:02 -0700107 int tick_nohz_enabled_snap; /* Previously seen value from sysfs. */
Paul E. McKenney5955f7e2012-05-09 12:07:05 -0700108#endif /* #ifdef CONFIG_RCU_FAST_NO_HZ */
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700109};
110
Paul E. McKenneyd71df902011-03-29 17:48:28 -0700111/* RCU's kthread states for tracing. */
112#define RCU_KTHREAD_STOPPED 0
113#define RCU_KTHREAD_RUNNING 1
114#define RCU_KTHREAD_WAITING 2
Paul E. McKenney15ba0ba2011-04-06 16:01:16 -0700115#define RCU_KTHREAD_OFFCPU 3
116#define RCU_KTHREAD_YIELDING 4
117#define RCU_KTHREAD_MAX 4
Paul E. McKenneyd71df902011-03-29 17:48:28 -0700118
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700119/*
120 * Definition for node within the RCU grace-period-detection hierarchy.
121 */
122struct rcu_node {
Paul E. McKenney1304afb2010-02-22 17:05:02 -0800123 raw_spinlock_t lock; /* Root rcu_node's lock protects some */
Paul E. McKenney1eba8f82009-09-23 09:50:42 -0700124 /* rcu_state fields as well as following. */
Paul E. McKenney20133cf2010-02-22 17:05:01 -0800125 unsigned long gpnum; /* Current grace period for this node. */
Paul E. McKenney86848962009-08-27 15:00:12 -0700126 /* This will either be equal to or one */
127 /* behind the root rcu_node's gpnum. */
Paul E. McKenney20133cf2010-02-22 17:05:01 -0800128 unsigned long completed; /* Last GP completed for this node. */
Paul E. McKenneyd09b62d2009-11-02 13:52:28 -0800129 /* This will either be equal to or one */
130 /* behind the root rcu_node's gpnum. */
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700131 unsigned long qsmask; /* CPUs or groups that need to switch in */
132 /* order for current grace period to proceed.*/
Paul E. McKenney1eba8f82009-09-23 09:50:42 -0700133 /* In leaf rcu_node, each bit corresponds to */
134 /* an rcu_data structure, otherwise, each */
135 /* bit corresponds to a child rcu_node */
136 /* structure. */
Paul E. McKenney12f5f522010-11-29 21:56:39 -0800137 unsigned long expmask; /* Groups that have ->blkd_tasks */
Paul E. McKenneyd9a3da02009-12-02 12:10:15 -0800138 /* elements that need to drain to allow the */
139 /* current expedited grace period to */
140 /* complete (only for TREE_PREEMPT_RCU). */
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700141 unsigned long qsmaskinit;
Paul E. McKenneyd9a3da02009-12-02 12:10:15 -0800142 /* Per-GP initial value for qsmask & expmask. */
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700143 unsigned long grpmask; /* Mask to apply to parent qsmask. */
Paul E. McKenney1eba8f82009-09-23 09:50:42 -0700144 /* Only one bit will be set in this mask. */
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700145 int grplo; /* lowest-numbered CPU or group here. */
146 int grphi; /* highest-numbered CPU or group here. */
147 u8 grpnum; /* CPU/group number for next level up. */
148 u8 level; /* root is at level 0. */
149 struct rcu_node *parent;
Paul E. McKenney12f5f522010-11-29 21:56:39 -0800150 struct list_head blkd_tasks;
151 /* Tasks blocked in RCU read-side critical */
152 /* section. Tasks are placed at the head */
153 /* of this list and age towards the tail. */
154 struct list_head *gp_tasks;
155 /* Pointer to the first task blocking the */
156 /* current grace period, or NULL if there */
157 /* is no such task. */
158 struct list_head *exp_tasks;
159 /* Pointer to the first task blocking the */
160 /* current expedited grace period, or NULL */
161 /* if there is no such task. If there */
162 /* is no current expedited grace period, */
163 /* then there can cannot be any such task. */
Paul E. McKenney27f4d282011-02-07 12:47:15 -0800164#ifdef CONFIG_RCU_BOOST
165 struct list_head *boost_tasks;
166 /* Pointer to first task that needs to be */
167 /* priority boosted, or NULL if no priority */
168 /* boosting is needed for this rcu_node */
169 /* structure. If there are no tasks */
170 /* queued on this rcu_node structure that */
171 /* are blocking the current grace period, */
172 /* there can be no such task. */
173 unsigned long boost_time;
174 /* When to start boosting (jiffies). */
175 struct task_struct *boost_kthread_task;
176 /* kthread that takes care of priority */
177 /* boosting for this rcu_node structure. */
Paul E. McKenneyd71df902011-03-29 17:48:28 -0700178 unsigned int boost_kthread_status;
179 /* State of boost_kthread_task for tracing. */
Paul E. McKenney0ea1f2e2011-02-22 13:42:43 -0800180 unsigned long n_tasks_boosted;
181 /* Total number of tasks boosted. */
182 unsigned long n_exp_boosts;
183 /* Number of tasks boosted for expedited GP. */
184 unsigned long n_normal_boosts;
185 /* Number of tasks boosted for normal GP. */
186 unsigned long n_balk_blkd_tasks;
187 /* Refused to boost: no blocked tasks. */
188 unsigned long n_balk_exp_gp_tasks;
189 /* Refused to boost: nothing blocking GP. */
190 unsigned long n_balk_boost_tasks;
191 /* Refused to boost: already boosting. */
192 unsigned long n_balk_notblocked;
193 /* Refused to boost: RCU RS CS still running. */
194 unsigned long n_balk_notyet;
195 /* Refused to boost: not yet time. */
196 unsigned long n_balk_nos;
197 /* Refused to boost: not sure why, though. */
198 /* This can happen due to race conditions. */
Paul E. McKenney27f4d282011-02-07 12:47:15 -0800199#endif /* #ifdef CONFIG_RCU_BOOST */
Paul E. McKenneydae6e642013-02-10 20:48:58 -0800200#ifdef CONFIG_RCU_NOCB_CPU
201 wait_queue_head_t nocb_gp_wq[2];
202 /* Place for rcu_nocb_kthread() to wait GP. */
Paul E. McKenneydae6e642013-02-10 20:48:58 -0800203#endif /* #ifdef CONFIG_RCU_NOCB_CPU */
Paul E. McKenney8b425aa82012-12-30 13:06:35 -0800204 int need_future_gp[2];
205 /* Counts of upcoming no-CB GP requests. */
Paul E. McKenney394f2762012-06-26 17:00:35 -0700206 raw_spinlock_t fqslock ____cacheline_internodealigned_in_smp;
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700207} ____cacheline_internodealigned_in_smp;
208
Paul E. McKenneya0b6c9a2009-09-28 07:46:33 -0700209/*
210 * Do a full breadth-first scan of the rcu_node structures for the
211 * specified rcu_state structure.
212 */
213#define rcu_for_each_node_breadth_first(rsp, rnp) \
214 for ((rnp) = &(rsp)->node[0]; \
Paul E. McKenneyf885b7f2012-04-23 15:52:53 -0700215 (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++)
Paul E. McKenneya0b6c9a2009-09-28 07:46:33 -0700216
Paul E. McKenneyd9a3da02009-12-02 12:10:15 -0800217/*
218 * Do a breadth-first scan of the non-leaf rcu_node structures for the
219 * specified rcu_state structure. Note that if there is a singleton
220 * rcu_node tree with but one rcu_node structure, this loop is a no-op.
221 */
222#define rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) \
223 for ((rnp) = &(rsp)->node[0]; \
Paul E. McKenneyf885b7f2012-04-23 15:52:53 -0700224 (rnp) < (rsp)->level[rcu_num_lvls - 1]; (rnp)++)
Paul E. McKenneyd9a3da02009-12-02 12:10:15 -0800225
226/*
227 * Scan the leaves of the rcu_node hierarchy for the specified rcu_state
228 * structure. Note that if there is a singleton rcu_node tree with but
229 * one rcu_node structure, this loop -will- visit the rcu_node structure.
230 * It is still a leaf node, even if it is also the root node.
231 */
Paul E. McKenneya0b6c9a2009-09-28 07:46:33 -0700232#define rcu_for_each_leaf_node(rsp, rnp) \
Paul E. McKenneyf885b7f2012-04-23 15:52:53 -0700233 for ((rnp) = (rsp)->level[rcu_num_lvls - 1]; \
234 (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++)
Paul E. McKenneya0b6c9a2009-09-28 07:46:33 -0700235
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700236/* Index values for nxttail array in struct rcu_data. */
237#define RCU_DONE_TAIL 0 /* Also RCU_WAIT head. */
238#define RCU_WAIT_TAIL 1 /* Also RCU_NEXT_READY head. */
239#define RCU_NEXT_READY_TAIL 2 /* Also RCU_NEXT head. */
240#define RCU_NEXT_TAIL 3
241#define RCU_NEXT_SIZE 4
242
243/* Per-CPU data for read-copy update. */
244struct rcu_data {
245 /* 1) quiescent-state and grace-period handling : */
Paul E. McKenney20133cf2010-02-22 17:05:01 -0800246 unsigned long completed; /* Track rsp->completed gp number */
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700247 /* in order to detect GP end. */
Paul E. McKenney20133cf2010-02-22 17:05:01 -0800248 unsigned long gpnum; /* Highest gp number that this CPU */
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700249 /* is aware of having started. */
Paul E. McKenneye4cc1f22011-06-27 00:17:43 -0700250 bool passed_quiesce; /* User-mode/idle loop etc. */
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700251 bool qs_pending; /* Core waits for quiesc state. */
252 bool beenonline; /* CPU online at least once. */
Paul E. McKenney6cc68792011-03-02 13:15:15 -0800253 bool preemptible; /* Preemptible RCU? */
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700254 struct rcu_node *mynode; /* This CPU's leaf of hierarchy */
255 unsigned long grpmask; /* Mask to apply to leaf qsmask. */
Paul E. McKenneya858af22012-01-16 13:29:10 -0800256#ifdef CONFIG_RCU_CPU_STALL_INFO
257 unsigned long ticks_this_gp; /* The number of scheduling-clock */
258 /* ticks this CPU has handled */
259 /* during and after the last grace */
260 /* period it is aware of. */
261#endif /* #ifdef CONFIG_RCU_CPU_STALL_INFO */
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700262
263 /* 2) batch handling */
264 /*
265 * If nxtlist is not NULL, it is partitioned as follows.
266 * Any of the partitions might be empty, in which case the
267 * pointer to that partition will be equal to the pointer for
268 * the following partition. When the list is empty, all of
Paul E. McKenney1eba8f82009-09-23 09:50:42 -0700269 * the nxttail elements point to the ->nxtlist pointer itself,
270 * which in that case is NULL.
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700271 *
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700272 * [nxtlist, *nxttail[RCU_DONE_TAIL]):
273 * Entries that batch # <= ->completed
274 * The grace period for these entries has completed, and
275 * the other grace-period-completed entries may be moved
276 * here temporarily in rcu_process_callbacks().
Paul E. McKenney1eba8f82009-09-23 09:50:42 -0700277 * [*nxttail[RCU_DONE_TAIL], *nxttail[RCU_WAIT_TAIL]):
278 * Entries that batch # <= ->completed - 1: waiting for current GP
279 * [*nxttail[RCU_WAIT_TAIL], *nxttail[RCU_NEXT_READY_TAIL]):
280 * Entries known to have arrived before current GP ended
281 * [*nxttail[RCU_NEXT_READY_TAIL], *nxttail[RCU_NEXT_TAIL]):
282 * Entries that might have arrived after current GP ended
283 * Note that the value of *nxttail[RCU_NEXT_TAIL] will
284 * always be NULL, as this is the end of the list.
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700285 */
286 struct rcu_head *nxtlist;
287 struct rcu_head **nxttail[RCU_NEXT_SIZE];
Paul E. McKenneydc35c892012-12-03 13:52:00 -0800288 unsigned long nxtcompleted[RCU_NEXT_SIZE];
289 /* grace periods for sublists. */
Paul E. McKenney486e2592012-01-06 14:11:30 -0800290 long qlen_lazy; /* # of lazy queued callbacks */
291 long qlen; /* # of queued callbacks, incl lazy */
Paul E. McKenney37c72e52009-10-14 10:15:55 -0700292 long qlen_last_fqs_check;
293 /* qlen at last check for QS forcing */
Paul E. McKenney269dcc12010-09-07 14:23:09 -0700294 unsigned long n_cbs_invoked; /* count of RCU cbs invoked. */
Paul E. McKenneyc635a4e12012-10-29 07:29:20 -0700295 unsigned long n_nocbs_invoked; /* count of no-CBs RCU cbs invoked. */
Lai Jiangshan29494be2010-10-20 14:13:06 +0800296 unsigned long n_cbs_orphaned; /* RCU cbs orphaned by dying CPU */
297 unsigned long n_cbs_adopted; /* RCU cbs adopted from dying CPU */
Paul E. McKenney37c72e52009-10-14 10:15:55 -0700298 unsigned long n_force_qs_snap;
299 /* did other CPU force QS recently? */
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700300 long blimit; /* Upper limit on a processed batch */
301
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700302 /* 3) dynticks interface. */
303 struct rcu_dynticks *dynticks; /* Shared per-CPU dynticks state. */
304 int dynticks_snap; /* Per-GP tracking for dynticks. */
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700305
306 /* 4) reasons this CPU needed to be kicked by force_quiescent_state */
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700307 unsigned long dynticks_fqs; /* Kicked due to dynticks idle. */
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700308 unsigned long offline_fqs; /* Kicked due to being offline. */
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700309
310 /* 5) __rcu_pending() statistics. */
Paul E. McKenney20133cf2010-02-22 17:05:01 -0800311 unsigned long n_rcu_pending; /* rcu_pending() calls since boot. */
312 unsigned long n_rp_qs_pending;
Paul E. McKenneyd21670a2010-04-14 17:39:26 -0700313 unsigned long n_rp_report_qs;
Paul E. McKenney20133cf2010-02-22 17:05:01 -0800314 unsigned long n_rp_cb_ready;
315 unsigned long n_rp_cpu_needs_gp;
316 unsigned long n_rp_gp_completed;
317 unsigned long n_rp_gp_started;
Paul E. McKenney20133cf2010-02-22 17:05:01 -0800318 unsigned long n_rp_need_nothing;
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700319
Paul E. McKenneyb626c1b2012-06-11 17:39:43 -0700320 /* 6) _rcu_barrier() and OOM callbacks. */
Paul E. McKenney06668ef2012-05-28 23:57:46 -0700321 struct rcu_head barrier_head;
Paul E. McKenneyb626c1b2012-06-11 17:39:43 -0700322#ifdef CONFIG_RCU_FAST_NO_HZ
323 struct rcu_head oom_head;
324#endif /* #ifdef CONFIG_RCU_FAST_NO_HZ */
Paul E. McKenney06668ef2012-05-28 23:57:46 -0700325
Paul E. McKenney3fbfbf72012-08-19 21:35:53 -0700326 /* 7) Callback offloading. */
327#ifdef CONFIG_RCU_NOCB_CPU
328 struct rcu_head *nocb_head; /* CBs waiting for kthread. */
329 struct rcu_head **nocb_tail;
330 atomic_long_t nocb_q_count; /* # CBs waiting for kthread */
331 atomic_long_t nocb_q_count_lazy; /* (approximate). */
332 int nocb_p_count; /* # CBs being invoked by kthread */
333 int nocb_p_count_lazy; /* (approximate). */
334 wait_queue_head_t nocb_wq; /* For nocb kthreads to sleep on. */
335 struct task_struct *nocb_kthread;
336#endif /* #ifdef CONFIG_RCU_NOCB_CPU */
337
Paul E. McKenney62310692013-03-06 13:37:09 -0800338 /* 8) RCU CPU stall data. */
339#ifdef CONFIG_RCU_CPU_STALL_INFO
340 unsigned int softirq_snap; /* Snapshot of softirq activity. */
341#endif /* #ifdef CONFIG_RCU_CPU_STALL_INFO */
342
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700343 int cpu;
Paul E. McKenneyd4c08f22011-06-25 06:36:56 -0700344 struct rcu_state *rsp;
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700345};
346
Paul E. McKenneyaf446b72011-09-10 21:54:08 -0700347/* Values for fqs_state field in struct rcu_state. */
Paul E. McKenney83f5b012009-10-28 08:14:49 -0700348#define RCU_GP_IDLE 0 /* No grace period in progress. */
349#define RCU_GP_INIT 1 /* Grace period being initialized. */
350#define RCU_SAVE_DYNTICK 2 /* Need to scan dyntick state. */
Paul E. McKenneyee47eb92010-01-04 15:09:07 -0800351#define RCU_FORCE_QS 3 /* Need to force quiescent state. */
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700352#define RCU_SIGNAL_INIT RCU_SAVE_DYNTICK
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700353
Paul E. McKenney026ad282013-04-03 22:14:11 -0700354#define RCU_JIFFIES_TILL_FORCE_QS (1 + (HZ > 250) + (HZ > 500))
355 /* For jiffies_till_first_fqs and */
356 /* and jiffies_till_next_fqs. */
Paul E. McKenney007b0922010-03-05 15:03:26 -0800357
Paul E. McKenney026ad282013-04-03 22:14:11 -0700358#define RCU_JIFFIES_FQS_DIV 256 /* Very large systems need more */
359 /* delay between bouts of */
360 /* quiescent-state forcing. */
361
362#define RCU_STALL_RAT_DELAY 2 /* Allow other CPUs time to take */
363 /* at least one scheduling clock */
364 /* irq before ratting on them. */
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700365
Peter Zijlstra08bca602011-05-20 16:06:29 -0700366#define rcu_wait(cond) \
367do { \
368 for (;;) { \
369 set_current_state(TASK_INTERRUPTIBLE); \
370 if (cond) \
371 break; \
372 schedule(); \
373 } \
374 __set_current_state(TASK_RUNNING); \
375} while (0)
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700376
377/*
378 * RCU global state, including node hierarchy. This hierarchy is
379 * represented in "heap" form in a dense array. The root (first level)
380 * of the hierarchy is in ->node[0] (referenced by ->level[0]), the second
381 * level in ->node[1] through ->node[m] (->node[1] referenced by ->level[1]),
382 * and the third level in ->node[m+1] and following (->node[m+1] referenced
383 * by ->level[2]). The number of levels is determined by the number of
384 * CPUs and by CONFIG_RCU_FANOUT. Small systems will have a "hierarchy"
385 * consisting of a single rcu_node.
386 */
387struct rcu_state {
388 struct rcu_node node[NUM_RCU_NODES]; /* Hierarchy. */
Paul E. McKenneyf885b7f2012-04-23 15:52:53 -0700389 struct rcu_node *level[RCU_NUM_LVLS]; /* Hierarchy levels. */
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700390 u32 levelcnt[MAX_RCU_LVLS + 1]; /* # nodes in each level. */
Paul E. McKenneyf885b7f2012-04-23 15:52:53 -0700391 u8 levelspread[RCU_NUM_LVLS]; /* kids/node in each level. */
Lai Jiangshan394f99a2010-06-28 16:25:04 +0800392 struct rcu_data __percpu *rda; /* pointer of percu rcu_data. */
Paul E. McKenney037b64e2012-05-28 23:26:01 -0700393 void (*call)(struct rcu_head *head, /* call_rcu() flavor. */
394 void (*func)(struct rcu_head *head));
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700395
396 /* The following fields are guarded by the root rcu_node's lock. */
397
Paul E. McKenneyaf446b72011-09-10 21:54:08 -0700398 u8 fqs_state ____cacheline_internodealigned_in_smp;
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700399 /* Force QS state. */
Paul E. McKenneya46e0892011-06-15 15:47:09 -0700400 u8 boost; /* Subject to priority boost. */
Paul E. McKenney20133cf2010-02-22 17:05:01 -0800401 unsigned long gpnum; /* Current gp number. */
402 unsigned long completed; /* # of last completed gp. */
Paul E. McKenneyb3dbec72012-06-18 18:36:08 -0700403 struct task_struct *gp_kthread; /* Task for grace periods. */
404 wait_queue_head_t gp_wq; /* Where GP task waits. */
405 int gp_flags; /* Commands for GP task. */
Paul E. McKenney1eba8f82009-09-23 09:50:42 -0700406
Paul E. McKenneyd9a3da02009-12-02 12:10:15 -0800407 /* End of fields guarded by root rcu_node's lock. */
Paul E. McKenney1eba8f82009-09-23 09:50:42 -0700408
Paul E. McKenney7b2e6012012-10-08 10:54:03 -0700409 raw_spinlock_t orphan_lock ____cacheline_internodealigned_in_smp;
410 /* Protect following fields. */
Paul E. McKenneyb1420f12012-03-01 13:18:08 -0800411 struct rcu_head *orphan_nxtlist; /* Orphaned callbacks that */
412 /* need a grace period. */
413 struct rcu_head **orphan_nxttail; /* Tail of above. */
414 struct rcu_head *orphan_donelist; /* Orphaned callbacks that */
415 /* are ready to invoke. */
416 struct rcu_head **orphan_donetail; /* Tail of above. */
417 long qlen_lazy; /* Number of lazy callbacks. */
418 long qlen; /* Total number of callbacks. */
Paul E. McKenney7b2e6012012-10-08 10:54:03 -0700419 /* End of fields guarded by orphan_lock. */
Paul E. McKenneya4fbe352012-10-07 08:36:12 -0700420
421 struct mutex onoff_mutex; /* Coordinate hotplug & GPs. */
422
Paul E. McKenney7be7f0b2012-05-29 05:18:53 -0700423 struct mutex barrier_mutex; /* Guards barrier fields. */
Paul E. McKenney24ebbca2012-05-29 00:34:56 -0700424 atomic_t barrier_cpu_count; /* # CPUs waiting on. */
Paul E. McKenney7db74df2012-05-29 03:03:37 -0700425 struct completion barrier_completion; /* Wake at barrier end. */
Paul E. McKenneycf3a9c42012-05-29 14:56:46 -0700426 unsigned long n_barrier_done; /* ++ at start and end of */
427 /* _rcu_barrier(). */
Paul E. McKenneya4fbe352012-10-07 08:36:12 -0700428 /* End of fields guarded by barrier_mutex. */
429
Paul E. McKenney40694d62012-10-11 15:24:03 -0700430 atomic_long_t expedited_start; /* Starting ticket. */
431 atomic_long_t expedited_done; /* Done ticket. */
Paul E. McKenneya30489c2012-10-11 16:18:09 -0700432 atomic_long_t expedited_wrap; /* # near-wrap incidents. */
433 atomic_long_t expedited_tryfail; /* # acquisition failures. */
434 atomic_long_t expedited_workdone1; /* # done by others #1. */
435 atomic_long_t expedited_workdone2; /* # done by others #2. */
436 atomic_long_t expedited_normal; /* # fallbacks to normal. */
437 atomic_long_t expedited_stoppedcpus; /* # successful stop_cpus. */
438 atomic_long_t expedited_done_tries; /* # tries to update _done. */
439 atomic_long_t expedited_done_lost; /* # times beaten to _done. */
440 atomic_long_t expedited_done_exit; /* # times exited _done loop. */
Paul E. McKenney40694d62012-10-11 15:24:03 -0700441
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700442 unsigned long jiffies_force_qs; /* Time at which to invoke */
443 /* force_quiescent_state(). */
444 unsigned long n_force_qs; /* Number of calls to */
445 /* force_quiescent_state(). */
446 unsigned long n_force_qs_lh; /* ~Number of calls leaving */
447 /* due to lock unavailable. */
448 unsigned long n_force_qs_ngp; /* Number of calls leaving */
449 /* due to no GP active. */
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700450 unsigned long gp_start; /* Time at which GP started, */
451 /* but in jiffies. */
452 unsigned long jiffies_stall; /* Time at which to check */
453 /* for CPU stalls. */
Paul E. McKenney15ba0ba2011-04-06 16:01:16 -0700454 unsigned long gp_max; /* Maximum GP duration in */
455 /* jiffies. */
Steven Rostedt (Red Hat)e66c33d2013-07-12 16:50:28 -0400456 const char *name; /* Name of structure. */
Paul E. McKenneya4889852012-12-03 08:16:28 -0800457 char abbr; /* Abbreviated name. */
Paul E. McKenney6ce75a22012-06-12 11:01:13 -0700458 struct list_head flavors; /* List of RCU flavors. */
Steven Rostedt016a8d52013-05-28 17:32:53 -0400459 struct irq_work wakeup_work; /* Postponed wakeups */
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700460};
461
Paul E. McKenney4cdfc172012-06-22 17:06:26 -0700462/* Values for rcu_state structure's gp_flags field. */
463#define RCU_GP_FLAG_INIT 0x1 /* Need grace-period initialization. */
464#define RCU_GP_FLAG_FQS 0x2 /* Need grace-period quiescent-state forcing. */
465
Paul E. McKenney6ce75a22012-06-12 11:01:13 -0700466extern struct list_head rcu_struct_flavors;
Paul E. McKenney3fbfbf72012-08-19 21:35:53 -0700467
468/* Sequence through rcu_state structures for each RCU flavor. */
Paul E. McKenney6ce75a22012-06-12 11:01:13 -0700469#define for_each_rcu_flavor(rsp) \
470 list_for_each_entry((rsp), &rcu_struct_flavors, flavors)
471
Paul E. McKenneyd9a3da02009-12-02 12:10:15 -0800472/* Return values for rcu_preempt_offline_tasks(). */
473
474#define RCU_OFL_TASKS_NORM_GP 0x1 /* Tasks blocking normal */
475 /* GP were moved to root. */
476#define RCU_OFL_TASKS_EXP_GP 0x2 /* Tasks blocking expedited */
477 /* GP were moved to root. */
478
Ingo Molnar6258c4f2009-03-25 16:42:24 +0100479/*
480 * RCU implementation internal declarations:
481 */
Paul E. McKenneyd6714c22009-08-22 13:56:46 -0700482extern struct rcu_state rcu_sched_state;
483DECLARE_PER_CPU(struct rcu_data, rcu_sched_data);
Ingo Molnar6258c4f2009-03-25 16:42:24 +0100484
485extern struct rcu_state rcu_bh_state;
486DECLARE_PER_CPU(struct rcu_data, rcu_bh_data);
487
Paul E. McKenneyf41d9112009-08-22 13:56:52 -0700488#ifdef CONFIG_TREE_PREEMPT_RCU
489extern struct rcu_state rcu_preempt_state;
490DECLARE_PER_CPU(struct rcu_data, rcu_preempt_data);
491#endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */
492
Paul E. McKenneyeab09932011-06-21 01:59:33 -0700493#ifdef CONFIG_RCU_BOOST
494DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_status);
495DECLARE_PER_CPU(int, rcu_cpu_kthread_cpu);
496DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_loops);
497DECLARE_PER_CPU(char, rcu_cpu_has_work);
498#endif /* #ifdef CONFIG_RCU_BOOST */
499
Paul E. McKenney017c4262010-01-14 16:10:58 -0800500#ifndef RCU_TREE_NONCORE
Paul E. McKenney9f77da92009-08-22 13:56:45 -0700501
Paul E. McKenney9b2619a2009-09-23 09:50:43 -0700502/* Forward declarations for rcutree_plugin.h */
Paul E. McKenneydbe01352009-11-10 13:37:19 -0800503static void rcu_bootup_announce(void);
Paul E. McKenney9b2619a2009-09-23 09:50:43 -0700504long rcu_batches_completed(void);
Paul E. McKenneycba6d0d2012-07-02 07:08:42 -0700505static void rcu_preempt_note_context_switch(int cpu);
Paul E. McKenney27f4d282011-02-07 12:47:15 -0800506static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp);
Paul E. McKenneyb668c9c2009-11-22 08:53:48 -0800507#ifdef CONFIG_HOTPLUG_CPU
Paul E. McKenneyd3f6bad2009-12-02 12:10:13 -0800508static void rcu_report_unblock_qs_rnp(struct rcu_node *rnp,
509 unsigned long flags);
Paul E. McKenneyb668c9c2009-11-22 08:53:48 -0800510#endif /* #ifdef CONFIG_HOTPLUG_CPU */
Paul E. McKenney1ed509a2010-02-22 17:05:05 -0800511static void rcu_print_detail_task_stall(struct rcu_state *rsp);
Paul E. McKenney9bc8b552011-08-13 13:31:47 -0700512static int rcu_print_task_stall(struct rcu_node *rnp);
Paul E. McKenney9b2619a2009-09-23 09:50:43 -0700513static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp);
514#ifdef CONFIG_HOTPLUG_CPU
Paul E. McKenney237c80c2009-10-15 09:26:14 -0700515static int rcu_preempt_offline_tasks(struct rcu_state *rsp,
516 struct rcu_node *rnp,
517 struct rcu_data *rdp);
Paul E. McKenney9b2619a2009-09-23 09:50:43 -0700518#endif /* #ifdef CONFIG_HOTPLUG_CPU */
519static void rcu_preempt_check_callbacks(int cpu);
Paul E. McKenney9b2619a2009-09-23 09:50:43 -0700520void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu));
Paul E. McKenneyd9a3da02009-12-02 12:10:15 -0800521#if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_TREE_PREEMPT_RCU)
Thomas Gleixnerb40d2932011-10-22 07:12:34 -0700522static void rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp,
523 bool wake);
Paul E. McKenneyd9a3da02009-12-02 12:10:15 -0800524#endif /* #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_TREE_PREEMPT_RCU) */
Paul E. McKenney9b2619a2009-09-23 09:50:43 -0700525static void __init __rcu_init_preempt(void);
Paul E. McKenney1217ed12011-05-04 21:43:49 -0700526static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags);
Paul E. McKenneya46e0892011-06-15 15:47:09 -0700527static void rcu_preempt_boost_start_gp(struct rcu_node *rnp);
528static void invoke_rcu_callbacks_kthread(void);
Paul E. McKenneydff16722011-11-29 15:57:13 -0800529static bool rcu_is_callbacks_kthread(void);
Paul E. McKenneya46e0892011-06-15 15:47:09 -0700530#ifdef CONFIG_RCU_BOOST
531static void rcu_preempt_do_callbacks(void);
Paul Gortmaker49fb4c62013-06-19 14:52:21 -0400532static int rcu_spawn_one_boost_kthread(struct rcu_state *rsp,
Thomas Gleixner5d01bbd2012-07-16 10:42:35 +0000533 struct rcu_node *rnp);
Paul E. McKenneya46e0892011-06-15 15:47:09 -0700534#endif /* #ifdef CONFIG_RCU_BOOST */
Paul Gortmaker49fb4c62013-06-19 14:52:21 -0400535static void rcu_prepare_kthreads(int cpu);
Paul E. McKenney7cb92492011-11-28 12:28:34 -0800536static void rcu_cleanup_after_idle(int cpu);
Paul E. McKenneyaea1b352011-11-02 06:54:54 -0700537static void rcu_prepare_for_idle(int cpu);
Paul E. McKenneyc57afe82012-02-28 11:02:21 -0800538static void rcu_idle_count_callbacks_posted(void);
Paul E. McKenneya858af22012-01-16 13:29:10 -0800539static void print_cpu_stall_info_begin(void);
540static void print_cpu_stall_info(struct rcu_state *rsp, int cpu);
541static void print_cpu_stall_info_end(void);
542static void zero_cpu_stall_ticks(struct rcu_data *rdp);
543static void increment_cpu_stall_ticks(void);
Paul E. McKenneydae6e642013-02-10 20:48:58 -0800544static int rcu_nocb_needs_gp(struct rcu_state *rsp);
545static void rcu_nocb_gp_set(struct rcu_node *rnp, int nrq);
Paul E. McKenney0446be42012-12-30 15:21:01 -0800546static void rcu_nocb_gp_cleanup(struct rcu_state *rsp, struct rcu_node *rnp);
Paul E. McKenneydae6e642013-02-10 20:48:58 -0800547static void rcu_init_one_nocb(struct rcu_node *rnp);
Paul E. McKenney3fbfbf72012-08-19 21:35:53 -0700548static bool __call_rcu_nocb(struct rcu_data *rdp, struct rcu_head *rhp,
549 bool lazy);
550static bool rcu_nocb_adopt_orphan_cbs(struct rcu_state *rsp,
551 struct rcu_data *rdp);
Paul E. McKenney3fbfbf72012-08-19 21:35:53 -0700552static void rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp);
553static void rcu_spawn_nocb_kthreads(struct rcu_state *rsp);
Paul E. McKenney65d798f2013-04-12 16:19:10 -0700554static void rcu_kick_nohz_cpu(int cpu);
Paul E. McKenney34ed62462013-01-07 13:37:42 -0800555static bool init_nocb_callback_list(struct rcu_data *rdp);
Paul E. McKenneyeb348b82013-06-21 13:00:57 -0700556static void rcu_sysidle_enter(struct rcu_dynticks *rdtp, int irq);
557static void rcu_sysidle_exit(struct rcu_dynticks *rdtp, int irq);
Paul E. McKenney0edd1b12013-06-21 16:37:22 -0700558static void rcu_sysidle_check_cpu(struct rcu_data *rdp, bool *isidle,
559 unsigned long *maxj);
560static bool is_sysidle_rcu_state(struct rcu_state *rsp);
561static void rcu_sysidle_report_gp(struct rcu_state *rsp, int isidle,
562 unsigned long maxj);
Paul E. McKenneyeb757672013-06-21 17:10:40 -0700563static void rcu_bind_gp_kthread(void);
Paul E. McKenney23332102013-06-21 12:34:33 -0700564static void rcu_sysidle_init_percpu_data(struct rcu_dynticks *rdtp);
Paul E. McKenney9b2619a2009-09-23 09:50:43 -0700565
Paul E. McKenney017c4262010-01-14 16:10:58 -0800566#endif /* #ifndef RCU_TREE_NONCORE */
Paul E. McKenney3fbfbf72012-08-19 21:35:53 -0700567
568#ifdef CONFIG_RCU_TRACE
569#ifdef CONFIG_RCU_NOCB_CPU
570/* Sum up queue lengths for tracing. */
571static inline void rcu_nocb_q_lengths(struct rcu_data *rdp, long *ql, long *qll)
572{
573 *ql = atomic_long_read(&rdp->nocb_q_count) + rdp->nocb_p_count;
574 *qll = atomic_long_read(&rdp->nocb_q_count_lazy) + rdp->nocb_p_count_lazy;
575}
576#else /* #ifdef CONFIG_RCU_NOCB_CPU */
577static inline void rcu_nocb_q_lengths(struct rcu_data *rdp, long *ql, long *qll)
578{
579 *ql = 0;
580 *qll = 0;
581}
582#endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */
583#endif /* #ifdef CONFIG_RCU_TRACE */