Commit | Line | Data |
---|---|---|
9f77da9f PM |
1 | /* |
2 | * Read-Copy Update mechanism for mutual exclusion (tree-based version) | |
3 | * Internal non-public definitions. | |
4 | * | |
5 | * This program is free software; you can redistribute it and/or modify | |
6 | * it under the terms of the GNU General Public License as published by | |
7 | * the Free Software Foundation; either version 2 of the License, or | |
8 | * (at your option) any later version. | |
9 | * | |
10 | * This program is distributed in the hope that it will be useful, | |
11 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
13 | * GNU General Public License for more details. | |
14 | * | |
15 | * You should have received a copy of the GNU General Public License | |
87de1cfd PM |
16 | * along with this program; if not, you can access it online at |
17 | * http://www.gnu.org/licenses/gpl-2.0.html. | |
9f77da9f PM |
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> | |
016a8d5b | 30 | #include <linux/irq_work.h> |
9f77da9f PM |
31 | |
32 | /* | |
8932a63d PM |
33 | * Define shape of hierarchy based on NR_CPUS, CONFIG_RCU_FANOUT, and |
34 | * CONFIG_RCU_FANOUT_LEAF. | |
9f77da9f | 35 | * In theory, it should be possible to add more levels straightforwardly. |
0209f649 PM |
36 | * In practice, this did work well going from three levels to four. |
37 | * Of course, your mileage may vary. | |
9f77da9f | 38 | */ |
cf244dc0 | 39 | #define MAX_RCU_LVLS 4 |
8932a63d | 40 | #define RCU_FANOUT_1 (CONFIG_RCU_FANOUT_LEAF) |
0209f649 PM |
41 | #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) | |
9f77da9f | 44 | |
0209f649 | 45 | #if NR_CPUS <= RCU_FANOUT_1 |
f885b7f2 | 46 | # define RCU_NUM_LVLS 1 |
9f77da9f PM |
47 | # 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 | |
cf244dc0 | 51 | # define NUM_RCU_LVL_4 0 |
0209f649 | 52 | #elif NR_CPUS <= RCU_FANOUT_2 |
f885b7f2 | 53 | # define RCU_NUM_LVLS 2 |
9f77da9f | 54 | # define NUM_RCU_LVL_0 1 |
0209f649 | 55 | # define NUM_RCU_LVL_1 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1) |
9f77da9f PM |
56 | # define NUM_RCU_LVL_2 (NR_CPUS) |
57 | # define NUM_RCU_LVL_3 0 | |
cf244dc0 | 58 | # define NUM_RCU_LVL_4 0 |
0209f649 | 59 | #elif NR_CPUS <= RCU_FANOUT_3 |
f885b7f2 | 60 | # define RCU_NUM_LVLS 3 |
9f77da9f | 61 | # define NUM_RCU_LVL_0 1 |
0209f649 PM |
62 | # 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) | |
cf244dc0 | 65 | # define NUM_RCU_LVL_4 0 |
0209f649 | 66 | #elif NR_CPUS <= RCU_FANOUT_4 |
f885b7f2 | 67 | # define RCU_NUM_LVLS 4 |
cf244dc0 | 68 | # define NUM_RCU_LVL_0 1 |
0209f649 PM |
69 | # 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) | |
9f77da9f PM |
73 | #else |
74 | # error "CONFIG_RCU_FANOUT insufficient for NR_CPUS" | |
0209f649 | 75 | #endif /* #if (NR_CPUS) <= RCU_FANOUT_1 */ |
9f77da9f | 76 | |
cf244dc0 | 77 | #define RCU_SUM (NUM_RCU_LVL_0 + NUM_RCU_LVL_1 + NUM_RCU_LVL_2 + NUM_RCU_LVL_3 + NUM_RCU_LVL_4) |
9f77da9f PM |
78 | #define NUM_RCU_NODES (RCU_SUM - NR_CPUS) |
79 | ||
f885b7f2 PM |
80 | extern int rcu_num_lvls; |
81 | extern int rcu_num_nodes; | |
82 | ||
9f77da9f PM |
83 | /* |
84 | * Dynticks per-CPU state. | |
85 | */ | |
86 | struct rcu_dynticks { | |
9b2e4f18 PM |
87 | 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. */ | |
2333210b PM |
91 | #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 */ | |
5955f7ee | 99 | #ifdef CONFIG_RCU_FAST_NO_HZ |
c0f4dfd4 | 100 | bool all_lazy; /* Are all CPU's CBs lazy? */ |
5955f7ee PM |
101 | unsigned long nonlazy_posted; |
102 | /* # times non-lazy CBs posted to CPU. */ | |
103 | unsigned long nonlazy_posted_snap; | |
104 | /* idle-period nonlazy_posted snapshot. */ | |
c0f4dfd4 PM |
105 | unsigned long last_accelerate; |
106 | /* Last jiffy CBs were accelerated. */ | |
c229828c PM |
107 | unsigned long last_advance_all; |
108 | /* Last jiffy CBs were all advanced. */ | |
9d2ad243 | 109 | int tick_nohz_enabled_snap; /* Previously seen value from sysfs. */ |
5955f7ee | 110 | #endif /* #ifdef CONFIG_RCU_FAST_NO_HZ */ |
9f77da9f PM |
111 | }; |
112 | ||
d71df90e PM |
113 | /* RCU's kthread states for tracing. */ |
114 | #define RCU_KTHREAD_STOPPED 0 | |
115 | #define RCU_KTHREAD_RUNNING 1 | |
116 | #define RCU_KTHREAD_WAITING 2 | |
15ba0ba8 PM |
117 | #define RCU_KTHREAD_OFFCPU 3 |
118 | #define RCU_KTHREAD_YIELDING 4 | |
119 | #define RCU_KTHREAD_MAX 4 | |
d71df90e | 120 | |
9f77da9f PM |
121 | /* |
122 | * Definition for node within the RCU grace-period-detection hierarchy. | |
123 | */ | |
124 | struct rcu_node { | |
1304afb2 | 125 | raw_spinlock_t lock; /* Root rcu_node's lock protects some */ |
1eba8f84 | 126 | /* rcu_state fields as well as following. */ |
20133cfc | 127 | unsigned long gpnum; /* Current grace period for this node. */ |
86848966 PM |
128 | /* This will either be equal to or one */ |
129 | /* behind the root rcu_node's gpnum. */ | |
20133cfc | 130 | unsigned long completed; /* Last GP completed for this node. */ |
d09b62df PM |
131 | /* This will either be equal to or one */ |
132 | /* behind the root rcu_node's gpnum. */ | |
9f77da9f PM |
133 | unsigned long qsmask; /* CPUs or groups that need to switch in */ |
134 | /* order for current grace period to proceed.*/ | |
1eba8f84 PM |
135 | /* In leaf rcu_node, each bit corresponds to */ |
136 | /* an rcu_data structure, otherwise, each */ | |
137 | /* bit corresponds to a child rcu_node */ | |
138 | /* structure. */ | |
12f5f524 | 139 | unsigned long expmask; /* Groups that have ->blkd_tasks */ |
d9a3da06 PM |
140 | /* elements that need to drain to allow the */ |
141 | /* current expedited grace period to */ | |
142 | /* complete (only for TREE_PREEMPT_RCU). */ | |
9f77da9f | 143 | unsigned long qsmaskinit; |
d9a3da06 | 144 | /* Per-GP initial value for qsmask & expmask. */ |
9f77da9f | 145 | unsigned long grpmask; /* Mask to apply to parent qsmask. */ |
1eba8f84 | 146 | /* Only one bit will be set in this mask. */ |
9f77da9f PM |
147 | int grplo; /* lowest-numbered CPU or group here. */ |
148 | int grphi; /* highest-numbered CPU or group here. */ | |
149 | u8 grpnum; /* CPU/group number for next level up. */ | |
150 | u8 level; /* root is at level 0. */ | |
151 | struct rcu_node *parent; | |
12f5f524 PM |
152 | struct list_head blkd_tasks; |
153 | /* Tasks blocked in RCU read-side critical */ | |
154 | /* section. Tasks are placed at the head */ | |
155 | /* of this list and age towards the tail. */ | |
156 | struct list_head *gp_tasks; | |
157 | /* Pointer to the first task blocking the */ | |
158 | /* current grace period, or NULL if there */ | |
159 | /* is no such task. */ | |
160 | struct list_head *exp_tasks; | |
161 | /* Pointer to the first task blocking the */ | |
162 | /* current expedited grace period, or NULL */ | |
163 | /* if there is no such task. If there */ | |
164 | /* is no current expedited grace period, */ | |
165 | /* then there can cannot be any such task. */ | |
27f4d280 PM |
166 | #ifdef CONFIG_RCU_BOOST |
167 | struct list_head *boost_tasks; | |
168 | /* Pointer to first task that needs to be */ | |
169 | /* priority boosted, or NULL if no priority */ | |
170 | /* boosting is needed for this rcu_node */ | |
171 | /* structure. If there are no tasks */ | |
172 | /* queued on this rcu_node structure that */ | |
173 | /* are blocking the current grace period, */ | |
174 | /* there can be no such task. */ | |
dfeb9765 PM |
175 | struct completion boost_completion; |
176 | /* Used to ensure that the rt_mutex used */ | |
177 | /* to carry out the boosting is fully */ | |
178 | /* released with no future boostee accesses */ | |
179 | /* before that rt_mutex is re-initialized. */ | |
abaa93d9 PM |
180 | struct rt_mutex boost_mtx; |
181 | /* Used only for the priority-boosting */ | |
182 | /* side effect, not as a lock. */ | |
27f4d280 PM |
183 | unsigned long boost_time; |
184 | /* When to start boosting (jiffies). */ | |
185 | struct task_struct *boost_kthread_task; | |
186 | /* kthread that takes care of priority */ | |
187 | /* boosting for this rcu_node structure. */ | |
d71df90e PM |
188 | unsigned int boost_kthread_status; |
189 | /* State of boost_kthread_task for tracing. */ | |
0ea1f2eb PM |
190 | unsigned long n_tasks_boosted; |
191 | /* Total number of tasks boosted. */ | |
192 | unsigned long n_exp_boosts; | |
193 | /* Number of tasks boosted for expedited GP. */ | |
194 | unsigned long n_normal_boosts; | |
195 | /* Number of tasks boosted for normal GP. */ | |
196 | unsigned long n_balk_blkd_tasks; | |
197 | /* Refused to boost: no blocked tasks. */ | |
198 | unsigned long n_balk_exp_gp_tasks; | |
199 | /* Refused to boost: nothing blocking GP. */ | |
200 | unsigned long n_balk_boost_tasks; | |
201 | /* Refused to boost: already boosting. */ | |
202 | unsigned long n_balk_notblocked; | |
203 | /* Refused to boost: RCU RS CS still running. */ | |
204 | unsigned long n_balk_notyet; | |
205 | /* Refused to boost: not yet time. */ | |
206 | unsigned long n_balk_nos; | |
207 | /* Refused to boost: not sure why, though. */ | |
208 | /* This can happen due to race conditions. */ | |
27f4d280 | 209 | #endif /* #ifdef CONFIG_RCU_BOOST */ |
dae6e64d PM |
210 | #ifdef CONFIG_RCU_NOCB_CPU |
211 | wait_queue_head_t nocb_gp_wq[2]; | |
212 | /* Place for rcu_nocb_kthread() to wait GP. */ | |
dae6e64d | 213 | #endif /* #ifdef CONFIG_RCU_NOCB_CPU */ |
8b425aa8 PM |
214 | int need_future_gp[2]; |
215 | /* Counts of upcoming no-CB GP requests. */ | |
394f2769 | 216 | raw_spinlock_t fqslock ____cacheline_internodealigned_in_smp; |
9f77da9f PM |
217 | } ____cacheline_internodealigned_in_smp; |
218 | ||
a0b6c9a7 PM |
219 | /* |
220 | * Do a full breadth-first scan of the rcu_node structures for the | |
221 | * specified rcu_state structure. | |
222 | */ | |
223 | #define rcu_for_each_node_breadth_first(rsp, rnp) \ | |
224 | for ((rnp) = &(rsp)->node[0]; \ | |
f885b7f2 | 225 | (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++) |
a0b6c9a7 | 226 | |
d9a3da06 PM |
227 | /* |
228 | * Do a breadth-first scan of the non-leaf rcu_node structures for the | |
229 | * specified rcu_state structure. Note that if there is a singleton | |
230 | * rcu_node tree with but one rcu_node structure, this loop is a no-op. | |
231 | */ | |
232 | #define rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) \ | |
233 | for ((rnp) = &(rsp)->node[0]; \ | |
f885b7f2 | 234 | (rnp) < (rsp)->level[rcu_num_lvls - 1]; (rnp)++) |
d9a3da06 PM |
235 | |
236 | /* | |
237 | * Scan the leaves of the rcu_node hierarchy for the specified rcu_state | |
238 | * structure. Note that if there is a singleton rcu_node tree with but | |
239 | * one rcu_node structure, this loop -will- visit the rcu_node structure. | |
240 | * It is still a leaf node, even if it is also the root node. | |
241 | */ | |
a0b6c9a7 | 242 | #define rcu_for_each_leaf_node(rsp, rnp) \ |
f885b7f2 PM |
243 | for ((rnp) = (rsp)->level[rcu_num_lvls - 1]; \ |
244 | (rnp) < &(rsp)->node[rcu_num_nodes]; (rnp)++) | |
a0b6c9a7 | 245 | |
9f77da9f PM |
246 | /* Index values for nxttail array in struct rcu_data. */ |
247 | #define RCU_DONE_TAIL 0 /* Also RCU_WAIT head. */ | |
248 | #define RCU_WAIT_TAIL 1 /* Also RCU_NEXT_READY head. */ | |
249 | #define RCU_NEXT_READY_TAIL 2 /* Also RCU_NEXT head. */ | |
250 | #define RCU_NEXT_TAIL 3 | |
251 | #define RCU_NEXT_SIZE 4 | |
252 | ||
253 | /* Per-CPU data for read-copy update. */ | |
254 | struct rcu_data { | |
255 | /* 1) quiescent-state and grace-period handling : */ | |
20133cfc | 256 | unsigned long completed; /* Track rsp->completed gp number */ |
9f77da9f | 257 | /* in order to detect GP end. */ |
20133cfc | 258 | unsigned long gpnum; /* Highest gp number that this CPU */ |
9f77da9f | 259 | /* is aware of having started. */ |
e4cc1f22 | 260 | bool passed_quiesce; /* User-mode/idle loop etc. */ |
9f77da9f PM |
261 | bool qs_pending; /* Core waits for quiesc state. */ |
262 | bool beenonline; /* CPU online at least once. */ | |
263 | struct rcu_node *mynode; /* This CPU's leaf of hierarchy */ | |
264 | unsigned long grpmask; /* Mask to apply to leaf qsmask. */ | |
a858af28 PM |
265 | #ifdef CONFIG_RCU_CPU_STALL_INFO |
266 | unsigned long ticks_this_gp; /* The number of scheduling-clock */ | |
267 | /* ticks this CPU has handled */ | |
268 | /* during and after the last grace */ | |
269 | /* period it is aware of. */ | |
270 | #endif /* #ifdef CONFIG_RCU_CPU_STALL_INFO */ | |
9f77da9f PM |
271 | |
272 | /* 2) batch handling */ | |
273 | /* | |
274 | * If nxtlist is not NULL, it is partitioned as follows. | |
275 | * Any of the partitions might be empty, in which case the | |
276 | * pointer to that partition will be equal to the pointer for | |
277 | * the following partition. When the list is empty, all of | |
1eba8f84 PM |
278 | * the nxttail elements point to the ->nxtlist pointer itself, |
279 | * which in that case is NULL. | |
9f77da9f | 280 | * |
9f77da9f PM |
281 | * [nxtlist, *nxttail[RCU_DONE_TAIL]): |
282 | * Entries that batch # <= ->completed | |
283 | * The grace period for these entries has completed, and | |
284 | * the other grace-period-completed entries may be moved | |
285 | * here temporarily in rcu_process_callbacks(). | |
1eba8f84 PM |
286 | * [*nxttail[RCU_DONE_TAIL], *nxttail[RCU_WAIT_TAIL]): |
287 | * Entries that batch # <= ->completed - 1: waiting for current GP | |
288 | * [*nxttail[RCU_WAIT_TAIL], *nxttail[RCU_NEXT_READY_TAIL]): | |
289 | * Entries known to have arrived before current GP ended | |
290 | * [*nxttail[RCU_NEXT_READY_TAIL], *nxttail[RCU_NEXT_TAIL]): | |
291 | * Entries that might have arrived after current GP ended | |
292 | * Note that the value of *nxttail[RCU_NEXT_TAIL] will | |
293 | * always be NULL, as this is the end of the list. | |
9f77da9f PM |
294 | */ |
295 | struct rcu_head *nxtlist; | |
296 | struct rcu_head **nxttail[RCU_NEXT_SIZE]; | |
dc35c893 PM |
297 | unsigned long nxtcompleted[RCU_NEXT_SIZE]; |
298 | /* grace periods for sublists. */ | |
486e2593 PM |
299 | long qlen_lazy; /* # of lazy queued callbacks */ |
300 | long qlen; /* # of queued callbacks, incl lazy */ | |
37c72e56 PM |
301 | long qlen_last_fqs_check; |
302 | /* qlen at last check for QS forcing */ | |
269dcc1c | 303 | unsigned long n_cbs_invoked; /* count of RCU cbs invoked. */ |
c635a4e1 | 304 | unsigned long n_nocbs_invoked; /* count of no-CBs RCU cbs invoked. */ |
29494be7 LJ |
305 | unsigned long n_cbs_orphaned; /* RCU cbs orphaned by dying CPU */ |
306 | unsigned long n_cbs_adopted; /* RCU cbs adopted from dying CPU */ | |
37c72e56 PM |
307 | unsigned long n_force_qs_snap; |
308 | /* did other CPU force QS recently? */ | |
9f77da9f PM |
309 | long blimit; /* Upper limit on a processed batch */ |
310 | ||
9f77da9f PM |
311 | /* 3) dynticks interface. */ |
312 | struct rcu_dynticks *dynticks; /* Shared per-CPU dynticks state. */ | |
313 | int dynticks_snap; /* Per-GP tracking for dynticks. */ | |
9f77da9f PM |
314 | |
315 | /* 4) reasons this CPU needed to be kicked by force_quiescent_state */ | |
9f77da9f | 316 | unsigned long dynticks_fqs; /* Kicked due to dynticks idle. */ |
9f77da9f | 317 | unsigned long offline_fqs; /* Kicked due to being offline. */ |
4a81e832 PM |
318 | unsigned long cond_resched_completed; |
319 | /* Grace period that needs help */ | |
320 | /* from cond_resched(). */ | |
9f77da9f PM |
321 | |
322 | /* 5) __rcu_pending() statistics. */ | |
20133cfc PM |
323 | unsigned long n_rcu_pending; /* rcu_pending() calls since boot. */ |
324 | unsigned long n_rp_qs_pending; | |
d21670ac | 325 | unsigned long n_rp_report_qs; |
20133cfc PM |
326 | unsigned long n_rp_cb_ready; |
327 | unsigned long n_rp_cpu_needs_gp; | |
328 | unsigned long n_rp_gp_completed; | |
329 | unsigned long n_rp_gp_started; | |
96d3fd0d | 330 | unsigned long n_rp_nocb_defer_wakeup; |
20133cfc | 331 | unsigned long n_rp_need_nothing; |
9f77da9f | 332 | |
b626c1b6 | 333 | /* 6) _rcu_barrier() and OOM callbacks. */ |
06668efa | 334 | struct rcu_head barrier_head; |
b626c1b6 PM |
335 | #ifdef CONFIG_RCU_FAST_NO_HZ |
336 | struct rcu_head oom_head; | |
337 | #endif /* #ifdef CONFIG_RCU_FAST_NO_HZ */ | |
06668efa | 338 | |
3fbfbf7a PM |
339 | /* 7) Callback offloading. */ |
340 | #ifdef CONFIG_RCU_NOCB_CPU | |
341 | struct rcu_head *nocb_head; /* CBs waiting for kthread. */ | |
342 | struct rcu_head **nocb_tail; | |
343 | atomic_long_t nocb_q_count; /* # CBs waiting for kthread */ | |
344 | atomic_long_t nocb_q_count_lazy; /* (approximate). */ | |
fbce7497 PM |
345 | struct rcu_head *nocb_follower_head; /* CBs ready to invoke. */ |
346 | struct rcu_head **nocb_follower_tail; | |
347 | atomic_long_t nocb_follower_count; /* # CBs ready to invoke. */ | |
348 | atomic_long_t nocb_follower_count_lazy; /* (approximate). */ | |
3fbfbf7a PM |
349 | int nocb_p_count; /* # CBs being invoked by kthread */ |
350 | int nocb_p_count_lazy; /* (approximate). */ | |
351 | wait_queue_head_t nocb_wq; /* For nocb kthreads to sleep on. */ | |
352 | struct task_struct *nocb_kthread; | |
9fdd3bc9 | 353 | int nocb_defer_wakeup; /* Defer wakeup of nocb_kthread. */ |
fbce7497 PM |
354 | |
355 | /* The following fields are used by the leader, hence own cacheline. */ | |
356 | struct rcu_head *nocb_gp_head ____cacheline_internodealigned_in_smp; | |
357 | /* CBs waiting for GP. */ | |
358 | struct rcu_head **nocb_gp_tail; | |
359 | long nocb_gp_count; | |
360 | long nocb_gp_count_lazy; | |
11ed7f93 | 361 | bool nocb_leader_sleep; /* Is the nocb leader thread asleep? */ |
fbce7497 PM |
362 | struct rcu_data *nocb_next_follower; |
363 | /* Next follower in wakeup chain. */ | |
364 | ||
365 | /* The following fields are used by the follower, hence new cachline. */ | |
366 | struct rcu_data *nocb_leader ____cacheline_internodealigned_in_smp; | |
367 | /* Leader CPU takes GP-end wakeups. */ | |
3fbfbf7a PM |
368 | #endif /* #ifdef CONFIG_RCU_NOCB_CPU */ |
369 | ||
6231069b PM |
370 | /* 8) RCU CPU stall data. */ |
371 | #ifdef CONFIG_RCU_CPU_STALL_INFO | |
372 | unsigned int softirq_snap; /* Snapshot of softirq activity. */ | |
373 | #endif /* #ifdef CONFIG_RCU_CPU_STALL_INFO */ | |
374 | ||
9f77da9f | 375 | int cpu; |
d4c08f2a | 376 | struct rcu_state *rsp; |
9f77da9f PM |
377 | }; |
378 | ||
af446b70 | 379 | /* Values for fqs_state field in struct rcu_state. */ |
83f5b01f PM |
380 | #define RCU_GP_IDLE 0 /* No grace period in progress. */ |
381 | #define RCU_GP_INIT 1 /* Grace period being initialized. */ | |
382 | #define RCU_SAVE_DYNTICK 2 /* Need to scan dyntick state. */ | |
ee47eb9f | 383 | #define RCU_FORCE_QS 3 /* Need to force quiescent state. */ |
9f77da9f | 384 | #define RCU_SIGNAL_INIT RCU_SAVE_DYNTICK |
9f77da9f | 385 | |
9fdd3bc9 PM |
386 | /* Values for nocb_defer_wakeup field in struct rcu_data. */ |
387 | #define RCU_NOGP_WAKE_NOT 0 | |
388 | #define RCU_NOGP_WAKE 1 | |
389 | #define RCU_NOGP_WAKE_FORCE 2 | |
390 | ||
026ad283 PM |
391 | #define RCU_JIFFIES_TILL_FORCE_QS (1 + (HZ > 250) + (HZ > 500)) |
392 | /* For jiffies_till_first_fqs and */ | |
393 | /* and jiffies_till_next_fqs. */ | |
007b0924 | 394 | |
026ad283 PM |
395 | #define RCU_JIFFIES_FQS_DIV 256 /* Very large systems need more */ |
396 | /* delay between bouts of */ | |
397 | /* quiescent-state forcing. */ | |
398 | ||
399 | #define RCU_STALL_RAT_DELAY 2 /* Allow other CPUs time to take */ | |
400 | /* at least one scheduling clock */ | |
401 | /* irq before ratting on them. */ | |
9f77da9f | 402 | |
08bca60a PZ |
403 | #define rcu_wait(cond) \ |
404 | do { \ | |
405 | for (;;) { \ | |
406 | set_current_state(TASK_INTERRUPTIBLE); \ | |
407 | if (cond) \ | |
408 | break; \ | |
409 | schedule(); \ | |
410 | } \ | |
411 | __set_current_state(TASK_RUNNING); \ | |
412 | } while (0) | |
9f77da9f PM |
413 | |
414 | /* | |
415 | * RCU global state, including node hierarchy. This hierarchy is | |
416 | * represented in "heap" form in a dense array. The root (first level) | |
417 | * of the hierarchy is in ->node[0] (referenced by ->level[0]), the second | |
418 | * level in ->node[1] through ->node[m] (->node[1] referenced by ->level[1]), | |
419 | * and the third level in ->node[m+1] and following (->node[m+1] referenced | |
420 | * by ->level[2]). The number of levels is determined by the number of | |
421 | * CPUs and by CONFIG_RCU_FANOUT. Small systems will have a "hierarchy" | |
422 | * consisting of a single rcu_node. | |
423 | */ | |
424 | struct rcu_state { | |
425 | struct rcu_node node[NUM_RCU_NODES]; /* Hierarchy. */ | |
f885b7f2 | 426 | struct rcu_node *level[RCU_NUM_LVLS]; /* Hierarchy levels. */ |
9f77da9f | 427 | u32 levelcnt[MAX_RCU_LVLS + 1]; /* # nodes in each level. */ |
f885b7f2 | 428 | u8 levelspread[RCU_NUM_LVLS]; /* kids/node in each level. */ |
4a81e832 | 429 | u8 flavor_mask; /* bit in flavor mask. */ |
394f99a9 | 430 | struct rcu_data __percpu *rda; /* pointer of percu rcu_data. */ |
037b64ed PM |
431 | void (*call)(struct rcu_head *head, /* call_rcu() flavor. */ |
432 | void (*func)(struct rcu_head *head)); | |
9f77da9f PM |
433 | |
434 | /* The following fields are guarded by the root rcu_node's lock. */ | |
435 | ||
af446b70 | 436 | u8 fqs_state ____cacheline_internodealigned_in_smp; |
9f77da9f | 437 | /* Force QS state. */ |
a46e0899 | 438 | u8 boost; /* Subject to priority boost. */ |
20133cfc PM |
439 | unsigned long gpnum; /* Current gp number. */ |
440 | unsigned long completed; /* # of last completed gp. */ | |
b3dbec76 PM |
441 | struct task_struct *gp_kthread; /* Task for grace periods. */ |
442 | wait_queue_head_t gp_wq; /* Where GP task waits. */ | |
afea227f PM |
443 | short gp_flags; /* Commands for GP task. */ |
444 | short gp_state; /* GP kthread sleep state. */ | |
1eba8f84 | 445 | |
d9a3da06 | 446 | /* End of fields guarded by root rcu_node's lock. */ |
1eba8f84 | 447 | |
7b2e6011 PM |
448 | raw_spinlock_t orphan_lock ____cacheline_internodealigned_in_smp; |
449 | /* Protect following fields. */ | |
b1420f1c PM |
450 | struct rcu_head *orphan_nxtlist; /* Orphaned callbacks that */ |
451 | /* need a grace period. */ | |
452 | struct rcu_head **orphan_nxttail; /* Tail of above. */ | |
453 | struct rcu_head *orphan_donelist; /* Orphaned callbacks that */ | |
454 | /* are ready to invoke. */ | |
455 | struct rcu_head **orphan_donetail; /* Tail of above. */ | |
456 | long qlen_lazy; /* Number of lazy callbacks. */ | |
457 | long qlen; /* Total number of callbacks. */ | |
7b2e6011 | 458 | /* End of fields guarded by orphan_lock. */ |
a4fbe35a PM |
459 | |
460 | struct mutex onoff_mutex; /* Coordinate hotplug & GPs. */ | |
461 | ||
7be7f0be | 462 | struct mutex barrier_mutex; /* Guards barrier fields. */ |
24ebbca8 | 463 | atomic_t barrier_cpu_count; /* # CPUs waiting on. */ |
7db74df8 | 464 | struct completion barrier_completion; /* Wake at barrier end. */ |
cf3a9c48 PM |
465 | unsigned long n_barrier_done; /* ++ at start and end of */ |
466 | /* _rcu_barrier(). */ | |
a4fbe35a PM |
467 | /* End of fields guarded by barrier_mutex. */ |
468 | ||
40694d66 PM |
469 | atomic_long_t expedited_start; /* Starting ticket. */ |
470 | atomic_long_t expedited_done; /* Done ticket. */ | |
a30489c5 PM |
471 | atomic_long_t expedited_wrap; /* # near-wrap incidents. */ |
472 | atomic_long_t expedited_tryfail; /* # acquisition failures. */ | |
473 | atomic_long_t expedited_workdone1; /* # done by others #1. */ | |
474 | atomic_long_t expedited_workdone2; /* # done by others #2. */ | |
475 | atomic_long_t expedited_normal; /* # fallbacks to normal. */ | |
476 | atomic_long_t expedited_stoppedcpus; /* # successful stop_cpus. */ | |
477 | atomic_long_t expedited_done_tries; /* # tries to update _done. */ | |
478 | atomic_long_t expedited_done_lost; /* # times beaten to _done. */ | |
479 | atomic_long_t expedited_done_exit; /* # times exited _done loop. */ | |
40694d66 | 480 | |
9f77da9f PM |
481 | unsigned long jiffies_force_qs; /* Time at which to invoke */ |
482 | /* force_quiescent_state(). */ | |
483 | unsigned long n_force_qs; /* Number of calls to */ | |
484 | /* force_quiescent_state(). */ | |
485 | unsigned long n_force_qs_lh; /* ~Number of calls leaving */ | |
486 | /* due to lock unavailable. */ | |
487 | unsigned long n_force_qs_ngp; /* Number of calls leaving */ | |
488 | /* due to no GP active. */ | |
9f77da9f PM |
489 | unsigned long gp_start; /* Time at which GP started, */ |
490 | /* but in jiffies. */ | |
491 | unsigned long jiffies_stall; /* Time at which to check */ | |
492 | /* for CPU stalls. */ | |
6193c76a PM |
493 | unsigned long jiffies_resched; /* Time at which to resched */ |
494 | /* a reluctant CPU. */ | |
15ba0ba8 PM |
495 | unsigned long gp_max; /* Maximum GP duration in */ |
496 | /* jiffies. */ | |
e66c33d5 | 497 | const char *name; /* Name of structure. */ |
a4889858 | 498 | char abbr; /* Abbreviated name. */ |
6ce75a23 | 499 | struct list_head flavors; /* List of RCU flavors. */ |
9f77da9f PM |
500 | }; |
501 | ||
4cdfc175 PM |
502 | /* Values for rcu_state structure's gp_flags field. */ |
503 | #define RCU_GP_FLAG_INIT 0x1 /* Need grace-period initialization. */ | |
504 | #define RCU_GP_FLAG_FQS 0x2 /* Need grace-period quiescent-state forcing. */ | |
505 | ||
afea227f PM |
506 | /* Values for rcu_state structure's gp_flags field. */ |
507 | #define RCU_GP_WAIT_INIT 0 /* Initial state. */ | |
508 | #define RCU_GP_WAIT_GPS 1 /* Wait for grace-period start. */ | |
509 | #define RCU_GP_WAIT_FQS 2 /* Wait for force-quiescent-state time. */ | |
510 | ||
6ce75a23 | 511 | extern struct list_head rcu_struct_flavors; |
3fbfbf7a PM |
512 | |
513 | /* Sequence through rcu_state structures for each RCU flavor. */ | |
6ce75a23 PM |
514 | #define for_each_rcu_flavor(rsp) \ |
515 | list_for_each_entry((rsp), &rcu_struct_flavors, flavors) | |
516 | ||
d9a3da06 PM |
517 | /* Return values for rcu_preempt_offline_tasks(). */ |
518 | ||
519 | #define RCU_OFL_TASKS_NORM_GP 0x1 /* Tasks blocking normal */ | |
520 | /* GP were moved to root. */ | |
521 | #define RCU_OFL_TASKS_EXP_GP 0x2 /* Tasks blocking expedited */ | |
522 | /* GP were moved to root. */ | |
523 | ||
6258c4fb IM |
524 | /* |
525 | * RCU implementation internal declarations: | |
526 | */ | |
d6714c22 PM |
527 | extern struct rcu_state rcu_sched_state; |
528 | DECLARE_PER_CPU(struct rcu_data, rcu_sched_data); | |
6258c4fb IM |
529 | |
530 | extern struct rcu_state rcu_bh_state; | |
531 | DECLARE_PER_CPU(struct rcu_data, rcu_bh_data); | |
532 | ||
f41d911f PM |
533 | #ifdef CONFIG_TREE_PREEMPT_RCU |
534 | extern struct rcu_state rcu_preempt_state; | |
535 | DECLARE_PER_CPU(struct rcu_data, rcu_preempt_data); | |
536 | #endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */ | |
537 | ||
eab0993c PM |
538 | #ifdef CONFIG_RCU_BOOST |
539 | DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_status); | |
540 | DECLARE_PER_CPU(int, rcu_cpu_kthread_cpu); | |
541 | DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_loops); | |
542 | DECLARE_PER_CPU(char, rcu_cpu_has_work); | |
543 | #endif /* #ifdef CONFIG_RCU_BOOST */ | |
544 | ||
017c4261 | 545 | #ifndef RCU_TREE_NONCORE |
9f77da9f | 546 | |
9b2619af | 547 | /* Forward declarations for rcutree_plugin.h */ |
dbe01350 | 548 | static void rcu_bootup_announce(void); |
9b2619af | 549 | long rcu_batches_completed(void); |
cba6d0d6 | 550 | static void rcu_preempt_note_context_switch(int cpu); |
27f4d280 | 551 | static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp); |
b668c9cf | 552 | #ifdef CONFIG_HOTPLUG_CPU |
d3f6bad3 PM |
553 | static void rcu_report_unblock_qs_rnp(struct rcu_node *rnp, |
554 | unsigned long flags); | |
b668c9cf | 555 | #endif /* #ifdef CONFIG_HOTPLUG_CPU */ |
1ed509a2 | 556 | static void rcu_print_detail_task_stall(struct rcu_state *rsp); |
9bc8b558 | 557 | static int rcu_print_task_stall(struct rcu_node *rnp); |
9b2619af PM |
558 | static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp); |
559 | #ifdef CONFIG_HOTPLUG_CPU | |
237c80c5 PM |
560 | static int rcu_preempt_offline_tasks(struct rcu_state *rsp, |
561 | struct rcu_node *rnp, | |
562 | struct rcu_data *rdp); | |
9b2619af PM |
563 | #endif /* #ifdef CONFIG_HOTPLUG_CPU */ |
564 | static void rcu_preempt_check_callbacks(int cpu); | |
9b2619af | 565 | void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu)); |
d9a3da06 | 566 | #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_TREE_PREEMPT_RCU) |
b40d293e TG |
567 | static void rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp, |
568 | bool wake); | |
d9a3da06 | 569 | #endif /* #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_TREE_PREEMPT_RCU) */ |
9b2619af | 570 | static void __init __rcu_init_preempt(void); |
1217ed1b | 571 | static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags); |
a46e0899 PM |
572 | static void rcu_preempt_boost_start_gp(struct rcu_node *rnp); |
573 | static void invoke_rcu_callbacks_kthread(void); | |
dff1672d | 574 | static bool rcu_is_callbacks_kthread(void); |
a46e0899 PM |
575 | #ifdef CONFIG_RCU_BOOST |
576 | static void rcu_preempt_do_callbacks(void); | |
49fb4c62 | 577 | static int rcu_spawn_one_boost_kthread(struct rcu_state *rsp, |
5d01bbd1 | 578 | struct rcu_node *rnp); |
a46e0899 | 579 | #endif /* #ifdef CONFIG_RCU_BOOST */ |
9386c0b7 | 580 | static void __init rcu_spawn_boost_kthreads(void); |
49fb4c62 | 581 | static void rcu_prepare_kthreads(int cpu); |
7cb92499 | 582 | static void rcu_cleanup_after_idle(int cpu); |
aea1b35e | 583 | static void rcu_prepare_for_idle(int cpu); |
c57afe80 | 584 | static void rcu_idle_count_callbacks_posted(void); |
a858af28 PM |
585 | static void print_cpu_stall_info_begin(void); |
586 | static void print_cpu_stall_info(struct rcu_state *rsp, int cpu); | |
587 | static void print_cpu_stall_info_end(void); | |
588 | static void zero_cpu_stall_ticks(struct rcu_data *rdp); | |
589 | static void increment_cpu_stall_ticks(void); | |
d7e29933 | 590 | static bool rcu_nocb_cpu_needs_barrier(struct rcu_state *rsp, int cpu); |
dae6e64d | 591 | static void rcu_nocb_gp_set(struct rcu_node *rnp, int nrq); |
0446be48 | 592 | static void rcu_nocb_gp_cleanup(struct rcu_state *rsp, struct rcu_node *rnp); |
dae6e64d | 593 | static void rcu_init_one_nocb(struct rcu_node *rnp); |
3fbfbf7a | 594 | static bool __call_rcu_nocb(struct rcu_data *rdp, struct rcu_head *rhp, |
96d3fd0d | 595 | bool lazy, unsigned long flags); |
3fbfbf7a | 596 | static bool rcu_nocb_adopt_orphan_cbs(struct rcu_state *rsp, |
96d3fd0d PM |
597 | struct rcu_data *rdp, |
598 | unsigned long flags); | |
9fdd3bc9 | 599 | static int rcu_nocb_need_deferred_wakeup(struct rcu_data *rdp); |
96d3fd0d | 600 | static void do_nocb_deferred_wakeup(struct rcu_data *rdp); |
3fbfbf7a | 601 | static void rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp); |
35ce7f29 PM |
602 | static void rcu_spawn_all_nocb_kthreads(int cpu); |
603 | static void __init rcu_spawn_nocb_kthreads(void); | |
604 | #ifdef CONFIG_RCU_NOCB_CPU | |
605 | static void __init rcu_organize_nocb_kthreads(struct rcu_state *rsp); | |
606 | #endif /* #ifdef CONFIG_RCU_NOCB_CPU */ | |
4a81e832 | 607 | static void __maybe_unused rcu_kick_nohz_cpu(int cpu); |
34ed6246 | 608 | static bool init_nocb_callback_list(struct rcu_data *rdp); |
eb348b89 PM |
609 | static void rcu_sysidle_enter(struct rcu_dynticks *rdtp, int irq); |
610 | static void rcu_sysidle_exit(struct rcu_dynticks *rdtp, int irq); | |
0edd1b17 PM |
611 | static void rcu_sysidle_check_cpu(struct rcu_data *rdp, bool *isidle, |
612 | unsigned long *maxj); | |
613 | static bool is_sysidle_rcu_state(struct rcu_state *rsp); | |
614 | static void rcu_sysidle_report_gp(struct rcu_state *rsp, int isidle, | |
615 | unsigned long maxj); | |
eb75767b | 616 | static void rcu_bind_gp_kthread(void); |
2333210b | 617 | static void rcu_sysidle_init_percpu_data(struct rcu_dynticks *rdtp); |
a096932f | 618 | static bool rcu_nohz_full_cpu(struct rcu_state *rsp); |
176f8f7a PM |
619 | static void rcu_dynticks_task_enter(void); |
620 | static void rcu_dynticks_task_exit(void); | |
9b2619af | 621 | |
017c4261 | 622 | #endif /* #ifndef RCU_TREE_NONCORE */ |
3fbfbf7a PM |
623 | |
624 | #ifdef CONFIG_RCU_TRACE | |
625 | #ifdef CONFIG_RCU_NOCB_CPU | |
626 | /* Sum up queue lengths for tracing. */ | |
627 | static inline void rcu_nocb_q_lengths(struct rcu_data *rdp, long *ql, long *qll) | |
628 | { | |
fbce7497 PM |
629 | *ql = atomic_long_read(&rdp->nocb_q_count) + |
630 | rdp->nocb_p_count + | |
631 | atomic_long_read(&rdp->nocb_follower_count) + | |
632 | rdp->nocb_p_count + rdp->nocb_gp_count; | |
633 | *qll = atomic_long_read(&rdp->nocb_q_count_lazy) + | |
634 | rdp->nocb_p_count_lazy + | |
635 | atomic_long_read(&rdp->nocb_follower_count_lazy) + | |
636 | rdp->nocb_p_count_lazy + rdp->nocb_gp_count_lazy; | |
3fbfbf7a PM |
637 | } |
638 | #else /* #ifdef CONFIG_RCU_NOCB_CPU */ | |
639 | static inline void rcu_nocb_q_lengths(struct rcu_data *rdp, long *ql, long *qll) | |
640 | { | |
641 | *ql = 0; | |
642 | *qll = 0; | |
643 | } | |
644 | #endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */ | |
645 | #endif /* #ifdef CONFIG_RCU_TRACE */ |