1 /* SPDX-License-Identifier: GPL-2.0+ */ 1 /* SPDX-License-Identifier: GPL-2.0+ */
2 /* 2 /*
3 * Read-Copy Update mechanism for mutual exclu 3 * Read-Copy Update mechanism for mutual exclusion (tree-based version)
4 * Internal non-public definitions. 4 * Internal non-public definitions.
5 * 5 *
6 * Copyright IBM Corporation, 2008 6 * Copyright IBM Corporation, 2008
7 * 7 *
8 * Author: Ingo Molnar <mingo@elte.hu> 8 * Author: Ingo Molnar <mingo@elte.hu>
9 * Paul E. McKenney <paulmck@linux.ibm 9 * Paul E. McKenney <paulmck@linux.ibm.com>
10 */ 10 */
11 11
12 #include <linux/cache.h> 12 #include <linux/cache.h>
13 #include <linux/kthread.h> <<
14 #include <linux/spinlock.h> 13 #include <linux/spinlock.h>
15 #include <linux/rtmutex.h> 14 #include <linux/rtmutex.h>
16 #include <linux/threads.h> 15 #include <linux/threads.h>
17 #include <linux/cpumask.h> 16 #include <linux/cpumask.h>
18 #include <linux/seqlock.h> 17 #include <linux/seqlock.h>
19 #include <linux/swait.h> 18 #include <linux/swait.h>
>> 19 #include <linux/stop_machine.h>
20 #include <linux/rcu_node_tree.h> 20 #include <linux/rcu_node_tree.h>
21 21
22 #include "rcu_segcblist.h" 22 #include "rcu_segcblist.h"
23 23
24 /* Communicate arguments to a kthread worker h !! 24 /* Communicate arguments to a workqueue handler. */
25 struct rcu_exp_work { 25 struct rcu_exp_work {
26 unsigned long rew_s; 26 unsigned long rew_s;
27 struct kthread_work rew_work; !! 27 struct work_struct rew_work;
28 }; 28 };
29 29
30 /* RCU's kthread states for tracing. */ 30 /* RCU's kthread states for tracing. */
31 #define RCU_KTHREAD_STOPPED 0 31 #define RCU_KTHREAD_STOPPED 0
32 #define RCU_KTHREAD_RUNNING 1 32 #define RCU_KTHREAD_RUNNING 1
33 #define RCU_KTHREAD_WAITING 2 33 #define RCU_KTHREAD_WAITING 2
34 #define RCU_KTHREAD_OFFCPU 3 34 #define RCU_KTHREAD_OFFCPU 3
35 #define RCU_KTHREAD_YIELDING 4 35 #define RCU_KTHREAD_YIELDING 4
36 #define RCU_KTHREAD_MAX 4 36 #define RCU_KTHREAD_MAX 4
37 37
38 /* 38 /*
39 * Definition for node within the RCU grace-pe 39 * Definition for node within the RCU grace-period-detection hierarchy.
40 */ 40 */
41 struct rcu_node { 41 struct rcu_node {
42 raw_spinlock_t __private lock; /* Roo 42 raw_spinlock_t __private lock; /* Root rcu_node's lock protects */
43 /* so 43 /* some rcu_state fields as well as */
44 /* fo 44 /* following. */
45 unsigned long gp_seq; /* Track rsp-> !! 45 unsigned long gp_seq; /* Track rsp->rcu_gp_seq. */
46 unsigned long gp_seq_needed; /* Track 46 unsigned long gp_seq_needed; /* Track furthest future GP request. */
47 unsigned long completedqs; /* All QSes 47 unsigned long completedqs; /* All QSes done for this node. */
48 unsigned long qsmask; /* CPUs or gro 48 unsigned long qsmask; /* CPUs or groups that need to switch in */
49 /* order for 49 /* order for current grace period to proceed.*/
50 /* In leaf rc 50 /* In leaf rcu_node, each bit corresponds to */
51 /* an rcu_dat 51 /* an rcu_data structure, otherwise, each */
52 /* bit corres 52 /* bit corresponds to a child rcu_node */
53 /* structure. 53 /* structure. */
54 unsigned long rcu_gp_init_mask; /* Mas 54 unsigned long rcu_gp_init_mask; /* Mask of offline CPUs at GP init. */
55 unsigned long qsmaskinit; 55 unsigned long qsmaskinit;
56 /* Per-GP init 56 /* Per-GP initial value for qsmask. */
57 /* Initialize 57 /* Initialized from ->qsmaskinitnext at the */
58 /* beginning 58 /* beginning of each grace period. */
59 unsigned long qsmaskinitnext; 59 unsigned long qsmaskinitnext;
>> 60 /* Online CPUs for next grace period. */
60 unsigned long expmask; /* CPUs or gro 61 unsigned long expmask; /* CPUs or groups that need to check in */
61 /* to allow t 62 /* to allow the current expedited GP */
62 /* to complet 63 /* to complete. */
63 unsigned long expmaskinit; 64 unsigned long expmaskinit;
64 /* Per-GP init 65 /* Per-GP initial values for expmask. */
65 /* Initialize 66 /* Initialized from ->expmaskinitnext at the */
66 /* beginning 67 /* beginning of each expedited GP. */
67 unsigned long expmaskinitnext; 68 unsigned long expmaskinitnext;
68 /* Online CPUs 69 /* Online CPUs for next expedited GP. */
69 /* Any CPU th 70 /* Any CPU that has ever been online will */
70 /* have its b 71 /* have its bit set. */
71 struct kthread_worker *exp_kworker; <<
72 /* Workers per <<
73 /* initializat <<
74 unsigned long cbovldmask; <<
75 /* CPUs experi <<
76 unsigned long ffmask; /* Fully funct 72 unsigned long ffmask; /* Fully functional CPUs. */
77 unsigned long grpmask; /* Mask to app 73 unsigned long grpmask; /* Mask to apply to parent qsmask. */
78 /* Only one b 74 /* Only one bit will be set in this mask. */
79 int grplo; /* lowest-numb !! 75 int grplo; /* lowest-numbered CPU or group here. */
80 int grphi; /* highest-num !! 76 int grphi; /* highest-numbered CPU or group here. */
81 u8 grpnum; /* group numbe !! 77 u8 grpnum; /* CPU/group number for next level up. */
82 u8 level; /* root is at 78 u8 level; /* root is at level 0. */
83 bool wait_blkd_tasks;/* Necessary t 79 bool wait_blkd_tasks;/* Necessary to wait for blocked tasks to */
84 /* exit RCU r 80 /* exit RCU read-side critical sections */
85 /* before pro 81 /* before propagating offline up the */
86 /* rcu_node t 82 /* rcu_node tree? */
87 struct rcu_node *parent; 83 struct rcu_node *parent;
88 struct list_head blkd_tasks; 84 struct list_head blkd_tasks;
89 /* Tasks block 85 /* Tasks blocked in RCU read-side critical */
90 /* section. 86 /* section. Tasks are placed at the head */
91 /* of this li 87 /* of this list and age towards the tail. */
92 struct list_head *gp_tasks; 88 struct list_head *gp_tasks;
93 /* Pointer to 89 /* Pointer to the first task blocking the */
94 /* current gr 90 /* current grace period, or NULL if there */
95 /* is no such 91 /* is no such task. */
96 struct list_head *exp_tasks; 92 struct list_head *exp_tasks;
97 /* Pointer to 93 /* Pointer to the first task blocking the */
98 /* current ex 94 /* current expedited grace period, or NULL */
99 /* if there i 95 /* if there is no such task. If there */
100 /* is no curr 96 /* is no current expedited grace period, */
101 /* then there 97 /* then there can cannot be any such task. */
102 struct list_head *boost_tasks; 98 struct list_head *boost_tasks;
103 /* Pointer to 99 /* Pointer to first task that needs to be */
104 /* priority b 100 /* priority boosted, or NULL if no priority */
105 /* boosting i 101 /* boosting is needed for this rcu_node */
106 /* structure. 102 /* structure. If there are no tasks */
107 /* queued on 103 /* queued on this rcu_node structure that */
108 /* are blocki 104 /* are blocking the current grace period, */
109 /* there can 105 /* there can be no such task. */
110 struct rt_mutex boost_mtx; 106 struct rt_mutex boost_mtx;
111 /* Used only f 107 /* Used only for the priority-boosting */
112 /* side effec 108 /* side effect, not as a lock. */
113 unsigned long boost_time; 109 unsigned long boost_time;
114 /* When to sta 110 /* When to start boosting (jiffies). */
115 struct mutex kthread_mutex; <<
116 /* Exclusion f <<
117 /* manipulati <<
118 struct task_struct *boost_kthread_task 111 struct task_struct *boost_kthread_task;
119 /* kthread tha 112 /* kthread that takes care of priority */
120 /* boosting f 113 /* boosting for this rcu_node structure. */
121 unsigned int boost_kthread_status; 114 unsigned int boost_kthread_status;
122 /* State of bo 115 /* State of boost_kthread_task for tracing. */
123 unsigned long n_boosts; /* Number of b <<
124 #ifdef CONFIG_RCU_NOCB_CPU 116 #ifdef CONFIG_RCU_NOCB_CPU
125 struct swait_queue_head nocb_gp_wq[2]; 117 struct swait_queue_head nocb_gp_wq[2];
126 /* Place for r 118 /* Place for rcu_nocb_kthread() to wait GP. */
127 #endif /* #ifdef CONFIG_RCU_NOCB_CPU */ 119 #endif /* #ifdef CONFIG_RCU_NOCB_CPU */
128 raw_spinlock_t fqslock ____cacheline_i 120 raw_spinlock_t fqslock ____cacheline_internodealigned_in_smp;
129 121
130 spinlock_t exp_lock ____cacheline_inte 122 spinlock_t exp_lock ____cacheline_internodealigned_in_smp;
131 unsigned long exp_seq_rq; 123 unsigned long exp_seq_rq;
132 wait_queue_head_t exp_wq[4]; 124 wait_queue_head_t exp_wq[4];
133 struct rcu_exp_work rew; 125 struct rcu_exp_work rew;
134 bool exp_need_flush; /* Need to flu 126 bool exp_need_flush; /* Need to flush workitem? */
135 raw_spinlock_t exp_poll_lock; <<
136 /* Lock and da <<
137 unsigned long exp_seq_poll_rq; <<
138 struct work_struct exp_poll_wq; <<
139 } ____cacheline_internodealigned_in_smp; 127 } ____cacheline_internodealigned_in_smp;
140 128
141 /* 129 /*
142 * Bitmasks in an rcu_node cover the interval 130 * Bitmasks in an rcu_node cover the interval [grplo, grphi] of CPU IDs, and
143 * are indexed relative to this interval rathe 131 * are indexed relative to this interval rather than the global CPU ID space.
144 * This generates the bit for a CPU in node-lo 132 * This generates the bit for a CPU in node-local masks.
145 */ 133 */
146 #define leaf_node_cpu_bit(rnp, cpu) (BIT((cpu) 134 #define leaf_node_cpu_bit(rnp, cpu) (BIT((cpu) - (rnp)->grplo))
147 135
148 /* 136 /*
149 * Union to allow "aggregate OR" operation on 137 * Union to allow "aggregate OR" operation on the need for a quiescent
150 * state by the normal and expedited grace per 138 * state by the normal and expedited grace periods.
151 */ 139 */
152 union rcu_noqs { 140 union rcu_noqs {
153 struct { 141 struct {
154 u8 norm; 142 u8 norm;
155 u8 exp; 143 u8 exp;
156 } b; /* Bits. */ 144 } b; /* Bits. */
157 u16 s; /* Set of bits, aggregate OR he 145 u16 s; /* Set of bits, aggregate OR here. */
158 }; 146 };
159 147
160 /* <<
161 * Record the snapshot of the core stats at ha <<
162 * The member gp_seq is used to ensure that al <<
163 * during the sampling period. The snapshot is <<
164 * equal to rdp->gp_seq. <<
165 */ <<
166 struct rcu_snap_record { <<
167 unsigned long gp_seq; /* Tra <<
168 u64 cputime_irq; /* Acc <<
169 u64 cputime_softirq;/* Acc <<
170 u64 cputime_system; /* Acc <<
171 unsigned long nr_hardirqs; /* Acc <<
172 unsigned int nr_softirqs; /* Acc <<
173 unsigned long long nr_csw; /* Acc <<
174 unsigned long jiffies; /* Tra <<
175 }; <<
176 <<
177 /* Per-CPU data for read-copy update. */ 148 /* Per-CPU data for read-copy update. */
178 struct rcu_data { 149 struct rcu_data {
179 /* 1) quiescent-state and grace-period 150 /* 1) quiescent-state and grace-period handling : */
180 unsigned long gp_seq; /* Tra !! 151 unsigned long gp_seq; /* Track rsp->rcu_gp_seq counter. */
181 unsigned long gp_seq_needed; /* Tra 152 unsigned long gp_seq_needed; /* Track furthest future GP request. */
182 union rcu_noqs cpu_no_qs; /* No 153 union rcu_noqs cpu_no_qs; /* No QSes yet for this CPU. */
183 bool core_needs_qs; /* Cor !! 154 bool core_needs_qs; /* Core waits for quiesc state. */
184 bool beenonline; /* CPU 155 bool beenonline; /* CPU online at least once. */
185 bool gpwrap; /* Pos 156 bool gpwrap; /* Possible ->gp_seq wrap. */
186 bool cpu_started; /* RCU !! 157 bool deferred_qs; /* This CPU awaiting a deferred QS? */
187 struct rcu_node *mynode; /* Thi 158 struct rcu_node *mynode; /* This CPU's leaf of hierarchy */
188 unsigned long grpmask; /* Mas 159 unsigned long grpmask; /* Mask to apply to leaf qsmask. */
189 unsigned long ticks_this_gp; /* The 160 unsigned long ticks_this_gp; /* The number of scheduling-clock */
190 /* ti 161 /* ticks this CPU has handled */
191 /* du 162 /* during and after the last grace */
192 /* per 163 /* period it is aware of. */
193 struct irq_work defer_qs_iw; /* Obt <<
194 bool defer_qs_iw_pending; /* Sch <<
195 struct work_struct strict_work; /* Sch <<
196 164
197 /* 2) batch handling */ 165 /* 2) batch handling */
198 struct rcu_segcblist cblist; /* Seg 166 struct rcu_segcblist cblist; /* Segmented callback list, with */
199 /* dif 167 /* different callbacks waiting for */
200 /* dif 168 /* different grace periods. */
201 long qlen_last_fqs_check; 169 long qlen_last_fqs_check;
202 /* qle 170 /* qlen at last check for QS forcing */
203 unsigned long n_cbs_invoked; /* # c <<
204 unsigned long n_force_qs_snap; 171 unsigned long n_force_qs_snap;
205 /* did 172 /* did other CPU force QS recently? */
206 long blimit; /* Upp 173 long blimit; /* Upper limit on a processed batch */
207 174
208 /* 3) dynticks interface. */ 175 /* 3) dynticks interface. */
209 int watching_snap; /* Per !! 176 int dynticks_snap; /* Per-GP tracking for dynticks. */
>> 177 long dynticks_nesting; /* Track process nesting level. */
>> 178 long dynticks_nmi_nesting; /* Track irq/NMI nesting level. */
>> 179 atomic_t dynticks; /* Even value for idle, else odd. */
210 bool rcu_need_heavy_qs; /* GP 180 bool rcu_need_heavy_qs; /* GP old, so heavy quiescent state! */
211 bool rcu_urgent_qs; /* GP 181 bool rcu_urgent_qs; /* GP old need light quiescent state. */
212 bool rcu_forced_tick; /* For !! 182 #ifdef CONFIG_RCU_FAST_NO_HZ
213 bool rcu_forced_tick_exp; /* . !! 183 bool all_lazy; /* All CPU's CBs lazy at idle start? */
>> 184 unsigned long last_accelerate; /* Last jiffy CBs were accelerated. */
>> 185 unsigned long last_advance_all; /* Last jiffy CBs were all advanced. */
>> 186 int tick_nohz_enabled_snap; /* Previously seen value from sysfs. */
>> 187 #endif /* #ifdef CONFIG_RCU_FAST_NO_HZ */
214 188
215 /* 4) rcu_barrier(), OOM callbacks, an 189 /* 4) rcu_barrier(), OOM callbacks, and expediting. */
216 unsigned long barrier_seq_snap; /* Sna <<
217 struct rcu_head barrier_head; 190 struct rcu_head barrier_head;
218 int exp_watching_snap; /* Dou !! 191 int exp_dynticks_snap; /* Double-check need for IPI. */
219 192
220 /* 5) Callback offloading. */ 193 /* 5) Callback offloading. */
221 #ifdef CONFIG_RCU_NOCB_CPU 194 #ifdef CONFIG_RCU_NOCB_CPU
222 struct swait_queue_head nocb_cb_wq; /* !! 195 struct rcu_head *nocb_head; /* CBs waiting for kthread. */
223 struct swait_queue_head nocb_state_wq; !! 196 struct rcu_head **nocb_tail;
224 struct task_struct *nocb_gp_kthread; !! 197 atomic_long_t nocb_q_count; /* # CBs waiting for nocb */
>> 198 atomic_long_t nocb_q_count_lazy; /* invocation (all stages). */
>> 199 struct rcu_head *nocb_follower_head; /* CBs ready to invoke. */
>> 200 struct rcu_head **nocb_follower_tail;
>> 201 struct swait_queue_head nocb_wq; /* For nocb kthreads to sleep on. */
>> 202 struct task_struct *nocb_kthread;
225 raw_spinlock_t nocb_lock; /* Gua 203 raw_spinlock_t nocb_lock; /* Guard following pair of fields. */
226 int nocb_defer_wakeup; /* Def 204 int nocb_defer_wakeup; /* Defer wakeup of nocb_kthread. */
227 struct timer_list nocb_timer; /* Enf 205 struct timer_list nocb_timer; /* Enforce finite deferral. */
228 unsigned long nocb_gp_adv_time; /* Las !! 206
229 struct mutex nocb_gp_kthread_mutex; /* !! 207 /* The following fields are used by the leader, hence own cacheline. */
230 /* !! 208 struct rcu_head *nocb_gp_head ____cacheline_internodealigned_in_smp;
231 !! 209 /* CBs waiting for GP. */
232 /* The following fields are used by ca !! 210 struct rcu_head **nocb_gp_tail;
233 raw_spinlock_t nocb_bypass_lock ____ca !! 211 bool nocb_leader_sleep; /* Is the nocb leader thread asleep? */
234 struct rcu_cblist nocb_bypass; /* Loc !! 212 struct rcu_data *nocb_next_follower;
235 unsigned long nocb_bypass_first; /* Ti !! 213 /* Next follower in wakeup chain. */
236 unsigned long nocb_nobypass_last; /* L !! 214
237 int nocb_nobypass_count; /* # - !! 215 /* The following fields are used by the follower, hence new cachline. */
238 !! 216 struct rcu_data *nocb_leader ____cacheline_internodealigned_in_smp;
239 /* The following fields are used by GP !! 217 /* Leader CPU takes GP-end wakeups. */
240 raw_spinlock_t nocb_gp_lock ____cachel <<
241 u8 nocb_gp_sleep; /* Is <<
242 u8 nocb_gp_bypass; /* Fou <<
243 u8 nocb_gp_gp; /* GP <<
244 unsigned long nocb_gp_seq; /* If <<
245 unsigned long nocb_gp_loops; /* # p <<
246 struct swait_queue_head nocb_gp_wq; /* <<
247 bool nocb_cb_sleep; /* Is <<
248 struct task_struct *nocb_cb_kthread; <<
249 struct list_head nocb_head_rdp; /* <<
250 * Hea <<
251 * if <<
252 */ <<
253 struct list_head nocb_entry_rdp; /* rc <<
254 struct rcu_data *nocb_toggling_rdp; /* <<
255 <<
256 /* The following fields are used by CB <<
257 struct rcu_data *nocb_gp_rdp ____cache <<
258 /* GP <<
259 #endif /* #ifdef CONFIG_RCU_NOCB_CPU */ 218 #endif /* #ifdef CONFIG_RCU_NOCB_CPU */
260 219
261 /* 6) RCU priority boosting. */ 220 /* 6) RCU priority boosting. */
262 struct task_struct *rcu_cpu_kthread_ta 221 struct task_struct *rcu_cpu_kthread_task;
263 /* rcu 222 /* rcuc per-CPU kthread or NULL. */
264 unsigned int rcu_cpu_kthread_status; 223 unsigned int rcu_cpu_kthread_status;
265 char rcu_cpu_has_work; 224 char rcu_cpu_has_work;
266 unsigned long rcuc_activity; <<
267 225
268 /* 7) Diagnostic data, including RCU C 226 /* 7) Diagnostic data, including RCU CPU stall warnings. */
269 unsigned int softirq_snap; /* Sna 227 unsigned int softirq_snap; /* Snapshot of softirq activity. */
270 /* ->rcu_iw* fields protected by leaf 228 /* ->rcu_iw* fields protected by leaf rcu_node ->lock. */
271 struct irq_work rcu_iw; /* Che 229 struct irq_work rcu_iw; /* Check for non-irq activity. */
272 bool rcu_iw_pending; /* Is 230 bool rcu_iw_pending; /* Is ->rcu_iw pending? */
273 unsigned long rcu_iw_gp_seq; /* ->g 231 unsigned long rcu_iw_gp_seq; /* ->gp_seq associated with ->rcu_iw. */
274 unsigned long rcu_ofl_gp_seq; /* ->g 232 unsigned long rcu_ofl_gp_seq; /* ->gp_seq at last offline. */
275 short rcu_ofl_gp_state; /* ->g !! 233 short rcu_ofl_gp_flags; /* ->gp_flags at last offline. */
276 unsigned long rcu_onl_gp_seq; /* ->g 234 unsigned long rcu_onl_gp_seq; /* ->gp_seq at last online. */
277 short rcu_onl_gp_state; /* ->g !! 235 short rcu_onl_gp_flags; /* ->gp_flags at last online. */
278 unsigned long last_fqs_resched; /* Tim 236 unsigned long last_fqs_resched; /* Time of last rcu_resched(). */
279 unsigned long last_sched_clock; /* Jif <<
280 struct rcu_snap_record snap_record; /* <<
281 /* <<
282 237
283 long lazy_len; /* Len <<
284 int cpu; 238 int cpu;
285 }; 239 };
286 240
287 /* Values for nocb_defer_wakeup field in struc 241 /* Values for nocb_defer_wakeup field in struct rcu_data. */
288 #define RCU_NOCB_WAKE_NOT 0 242 #define RCU_NOCB_WAKE_NOT 0
289 #define RCU_NOCB_WAKE_BYPASS 1 !! 243 #define RCU_NOCB_WAKE 1
290 #define RCU_NOCB_WAKE_LAZY 2 !! 244 #define RCU_NOCB_WAKE_FORCE 2
291 #define RCU_NOCB_WAKE 3 <<
292 #define RCU_NOCB_WAKE_FORCE 4 <<
293 245
294 #define RCU_JIFFIES_TILL_FORCE_QS (1 + (HZ > 2 246 #define RCU_JIFFIES_TILL_FORCE_QS (1 + (HZ > 250) + (HZ > 500))
295 /* For 247 /* For jiffies_till_first_fqs and */
296 /* an 248 /* and jiffies_till_next_fqs. */
297 249
298 #define RCU_JIFFIES_FQS_DIV 256 /* Ver 250 #define RCU_JIFFIES_FQS_DIV 256 /* Very large systems need more */
299 /* de 251 /* delay between bouts of */
300 /* qu 252 /* quiescent-state forcing. */
301 253
302 #define RCU_STALL_RAT_DELAY 2 /* All 254 #define RCU_STALL_RAT_DELAY 2 /* Allow other CPUs time to take */
303 /* at 255 /* at least one scheduling clock */
304 /* ir 256 /* irq before ratting on them. */
305 257
306 #define rcu_wait(cond) 258 #define rcu_wait(cond) \
307 do { 259 do { \
308 for (;;) { 260 for (;;) { \
309 set_current_state(TASK_INTERRU 261 set_current_state(TASK_INTERRUPTIBLE); \
310 if (cond) 262 if (cond) \
311 break; 263 break; \
312 schedule(); 264 schedule(); \
313 } 265 } \
314 __set_current_state(TASK_RUNNING); 266 __set_current_state(TASK_RUNNING); \
315 } while (0) 267 } while (0)
316 268
317 /* 269 /*
318 * A max threshold for synchronize_rcu() users <<
319 * awaken directly by the rcu_gp_kthread(). Le <<
320 * deferred to the main worker. <<
321 */ <<
322 #define SR_MAX_USERS_WAKE_FROM_GP 5 <<
323 #define SR_NORMAL_GP_WAIT_HEAD_MAX 5 <<
324 <<
325 struct sr_wait_node { <<
326 atomic_t inuse; <<
327 struct llist_node node; <<
328 }; <<
329 <<
330 /* <<
331 * RCU global state, including node hierarchy. 270 * RCU global state, including node hierarchy. This hierarchy is
332 * represented in "heap" form in a dense array 271 * represented in "heap" form in a dense array. The root (first level)
333 * of the hierarchy is in ->node[0] (reference 272 * of the hierarchy is in ->node[0] (referenced by ->level[0]), the second
334 * level in ->node[1] through ->node[m] (->nod 273 * level in ->node[1] through ->node[m] (->node[1] referenced by ->level[1]),
335 * and the third level in ->node[m+1] and foll 274 * and the third level in ->node[m+1] and following (->node[m+1] referenced
336 * by ->level[2]). The number of levels is de 275 * by ->level[2]). The number of levels is determined by the number of
337 * CPUs and by CONFIG_RCU_FANOUT. Small syste 276 * CPUs and by CONFIG_RCU_FANOUT. Small systems will have a "hierarchy"
338 * consisting of a single rcu_node. 277 * consisting of a single rcu_node.
339 */ 278 */
340 struct rcu_state { 279 struct rcu_state {
341 struct rcu_node node[NUM_RCU_NODES]; 280 struct rcu_node node[NUM_RCU_NODES]; /* Hierarchy. */
342 struct rcu_node *level[RCU_NUM_LVLS + 281 struct rcu_node *level[RCU_NUM_LVLS + 1];
343 282 /* Hierarchy levels (+1 to */
344 283 /* shut bogus gcc warning) */
345 int ncpus; 284 int ncpus; /* # CPUs seen so far. */
346 int n_online_cpus; <<
347 285
348 /* The following fields are guarded by 286 /* The following fields are guarded by the root rcu_node's lock. */
349 287
350 unsigned long gp_seq ____cacheline_int !! 288 u8 boost ____cacheline_internodealigned_in_smp;
351 !! 289 /* Subject to priority boost. */
352 unsigned long gp_max; !! 290 unsigned long gp_seq; /* Grace-period sequence #. */
353 <<
354 struct task_struct *gp_kthread; 291 struct task_struct *gp_kthread; /* Task for grace periods. */
355 struct swait_queue_head gp_wq; 292 struct swait_queue_head gp_wq; /* Where GP task waits. */
356 short gp_flags; 293 short gp_flags; /* Commands for GP task. */
357 short gp_state; 294 short gp_state; /* GP kthread sleep state. */
358 unsigned long gp_wake_time; 295 unsigned long gp_wake_time; /* Last GP kthread wake. */
359 unsigned long gp_wake_seq; 296 unsigned long gp_wake_seq; /* ->gp_seq at ^^^. */
360 unsigned long gp_seq_polled; <<
361 unsigned long gp_seq_polled_snap; <<
362 unsigned long gp_seq_polled_exp_snap; <<
363 297
364 /* End of fields guarded by root rcu_n 298 /* End of fields guarded by root rcu_node's lock. */
365 299
366 struct mutex barrier_mutex; 300 struct mutex barrier_mutex; /* Guards barrier fields. */
367 atomic_t barrier_cpu_count; 301 atomic_t barrier_cpu_count; /* # CPUs waiting on. */
368 struct completion barrier_completion; 302 struct completion barrier_completion; /* Wake at barrier end. */
369 unsigned long barrier_sequence; 303 unsigned long barrier_sequence; /* ++ at start and end of */
370 304 /* rcu_barrier(). */
371 /* End of fields guarded by barrier_mu 305 /* End of fields guarded by barrier_mutex. */
372 306
373 raw_spinlock_t barrier_lock; <<
374 <<
375 struct mutex exp_mutex; 307 struct mutex exp_mutex; /* Serialize expedited GP. */
376 struct mutex exp_wake_mutex; 308 struct mutex exp_wake_mutex; /* Serialize wakeup. */
377 unsigned long expedited_sequence; 309 unsigned long expedited_sequence; /* Take a ticket. */
378 atomic_t expedited_need_qs; 310 atomic_t expedited_need_qs; /* # CPUs left to check in. */
379 struct swait_queue_head expedited_wq; 311 struct swait_queue_head expedited_wq; /* Wait for check-ins. */
380 int ncpus_snap; 312 int ncpus_snap; /* # CPUs seen last time. */
381 u8 cbovld; <<
382 u8 cbovldnext; <<
383 313
384 unsigned long jiffies_force_qs; 314 unsigned long jiffies_force_qs; /* Time at which to invoke */
385 315 /* force_quiescent_state(). */
386 unsigned long jiffies_kick_kthreads; 316 unsigned long jiffies_kick_kthreads; /* Time at which to kick */
387 317 /* kthreads, if configured. */
388 unsigned long n_force_qs; 318 unsigned long n_force_qs; /* Number of calls to */
389 319 /* force_quiescent_state(). */
390 unsigned long gp_start; 320 unsigned long gp_start; /* Time at which GP started, */
391 321 /* but in jiffies. */
392 unsigned long gp_end; 322 unsigned long gp_end; /* Time last GP ended, again */
393 323 /* in jiffies. */
394 unsigned long gp_activity; 324 unsigned long gp_activity; /* Time of last GP kthread */
395 325 /* activity in jiffies. */
396 unsigned long gp_req_activity; 326 unsigned long gp_req_activity; /* Time of last GP request */
397 327 /* in jiffies. */
398 unsigned long jiffies_stall; 328 unsigned long jiffies_stall; /* Time at which to check */
399 329 /* for CPU stalls. */
400 int nr_fqs_jiffies_stall; <<
401 <<
402 <<
403 <<
404 unsigned long jiffies_resched; 330 unsigned long jiffies_resched; /* Time at which to resched */
405 331 /* a reluctant CPU. */
406 unsigned long n_force_qs_gpstart; 332 unsigned long n_force_qs_gpstart; /* Snapshot of n_force_qs at */
407 333 /* GP start. */
>> 334 unsigned long gp_max; /* Maximum GP duration in */
>> 335 /* jiffies. */
408 const char *name; 336 const char *name; /* Name of structure. */
409 char abbr; 337 char abbr; /* Abbreviated name. */
410 338
411 arch_spinlock_t ofl_lock ____cacheline !! 339 raw_spinlock_t ofl_lock ____cacheline_internodealigned_in_smp;
412 340 /* Synchronize offline with */
413 341 /* GP pre-initialization. */
414 <<
415 /* synchronize_rcu() part. */ <<
416 struct llist_head srs_next; /* req <<
417 struct llist_node *srs_wait_tail; /* w <<
418 struct llist_node *srs_done_tail; /* r <<
419 struct sr_wait_node srs_wait_nodes[SR_ <<
420 struct work_struct srs_cleanup_work; <<
421 atomic_t srs_cleanups_pending; /* srs <<
422 <<
423 #ifdef CONFIG_RCU_NOCB_CPU <<
424 struct mutex nocb_mutex; <<
425 int nocb_is_setup; <<
426 #endif <<
427 }; 342 };
428 343
429 /* Values for rcu_state structure's gp_flags f 344 /* Values for rcu_state structure's gp_flags field. */
430 #define RCU_GP_FLAG_INIT 0x1 /* Need grace- 345 #define RCU_GP_FLAG_INIT 0x1 /* Need grace-period initialization. */
431 #define RCU_GP_FLAG_FQS 0x2 /* Need grace- 346 #define RCU_GP_FLAG_FQS 0x2 /* Need grace-period quiescent-state forcing. */
432 #define RCU_GP_FLAG_OVLD 0x4 /* Experiencin <<
433 347
434 /* Values for rcu_state structure's gp_state f 348 /* Values for rcu_state structure's gp_state field. */
435 #define RCU_GP_IDLE 0 /* Initial sta 349 #define RCU_GP_IDLE 0 /* Initial state and no GP in progress. */
436 #define RCU_GP_WAIT_GPS 1 /* Wait for gr 350 #define RCU_GP_WAIT_GPS 1 /* Wait for grace-period start. */
437 #define RCU_GP_DONE_GPS 2 /* Wait done f 351 #define RCU_GP_DONE_GPS 2 /* Wait done for grace-period start. */
438 #define RCU_GP_ONOFF 3 /* Grace-perio 352 #define RCU_GP_ONOFF 3 /* Grace-period initialization hotplug. */
439 #define RCU_GP_INIT 4 /* Grace-perio 353 #define RCU_GP_INIT 4 /* Grace-period initialization. */
440 #define RCU_GP_WAIT_FQS 5 /* Wait for fo 354 #define RCU_GP_WAIT_FQS 5 /* Wait for force-quiescent-state time. */
441 #define RCU_GP_DOING_FQS 6 /* Wait done f 355 #define RCU_GP_DOING_FQS 6 /* Wait done for force-quiescent-state time. */
442 #define RCU_GP_CLEANUP 7 /* Grace-perio 356 #define RCU_GP_CLEANUP 7 /* Grace-period cleanup started. */
443 #define RCU_GP_CLEANED 8 /* Grace-perio 357 #define RCU_GP_CLEANED 8 /* Grace-period cleanup complete. */
444 358
>> 359 static const char * const gp_state_names[] = {
>> 360 "RCU_GP_IDLE",
>> 361 "RCU_GP_WAIT_GPS",
>> 362 "RCU_GP_DONE_GPS",
>> 363 "RCU_GP_ONOFF",
>> 364 "RCU_GP_INIT",
>> 365 "RCU_GP_WAIT_FQS",
>> 366 "RCU_GP_DOING_FQS",
>> 367 "RCU_GP_CLEANUP",
>> 368 "RCU_GP_CLEANED",
>> 369 };
>> 370
445 /* 371 /*
446 * In order to export the rcu_state name to th 372 * In order to export the rcu_state name to the tracing tools, it
447 * needs to be added in the __tracepoint_strin 373 * needs to be added in the __tracepoint_string section.
448 * This requires defining a separate variable 374 * This requires defining a separate variable tp_<sname>_varname
449 * that points to the string being used, and t 375 * that points to the string being used, and this will allow
450 * the tracing userspace tools to be able to d 376 * the tracing userspace tools to be able to decipher the string
451 * address to the matching string. 377 * address to the matching string.
452 */ 378 */
453 #ifdef CONFIG_PREEMPT_RCU 379 #ifdef CONFIG_PREEMPT_RCU
454 #define RCU_ABBR 'p' 380 #define RCU_ABBR 'p'
455 #define RCU_NAME_RAW "rcu_preempt" 381 #define RCU_NAME_RAW "rcu_preempt"
456 #else /* #ifdef CONFIG_PREEMPT_RCU */ 382 #else /* #ifdef CONFIG_PREEMPT_RCU */
457 #define RCU_ABBR 's' 383 #define RCU_ABBR 's'
458 #define RCU_NAME_RAW "rcu_sched" 384 #define RCU_NAME_RAW "rcu_sched"
459 #endif /* #else #ifdef CONFIG_PREEMPT_RCU */ 385 #endif /* #else #ifdef CONFIG_PREEMPT_RCU */
460 #ifndef CONFIG_TRACING 386 #ifndef CONFIG_TRACING
461 #define RCU_NAME RCU_NAME_RAW 387 #define RCU_NAME RCU_NAME_RAW
462 #else /* #ifdef CONFIG_TRACING */ 388 #else /* #ifdef CONFIG_TRACING */
463 static char rcu_name[] = RCU_NAME_RAW; 389 static char rcu_name[] = RCU_NAME_RAW;
464 static const char *tp_rcu_varname __used __tra 390 static const char *tp_rcu_varname __used __tracepoint_string = rcu_name;
465 #define RCU_NAME rcu_name 391 #define RCU_NAME rcu_name
466 #endif /* #else #ifdef CONFIG_TRACING */ 392 #endif /* #else #ifdef CONFIG_TRACING */
467 393
>> 394 int rcu_dynticks_snap(struct rcu_data *rdp);
>> 395
468 /* Forward declarations for tree_plugin.h */ 396 /* Forward declarations for tree_plugin.h */
469 static void rcu_bootup_announce(void); 397 static void rcu_bootup_announce(void);
470 static void rcu_qs(void); 398 static void rcu_qs(void);
471 static int rcu_preempt_blocked_readers_cgp(str 399 static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp);
472 #ifdef CONFIG_HOTPLUG_CPU 400 #ifdef CONFIG_HOTPLUG_CPU
473 static bool rcu_preempt_has_tasks(struct rcu_n 401 static bool rcu_preempt_has_tasks(struct rcu_node *rnp);
474 #endif /* #ifdef CONFIG_HOTPLUG_CPU */ 402 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
475 static int rcu_print_task_exp_stall(struct rcu 403 static int rcu_print_task_exp_stall(struct rcu_node *rnp);
476 static void rcu_preempt_check_blocked_tasks(st 404 static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp);
477 static void rcu_flavor_sched_clock_irq(int use 405 static void rcu_flavor_sched_clock_irq(int user);
>> 406 void call_rcu(struct rcu_head *head, rcu_callback_t func);
478 static void dump_blkd_tasks(struct rcu_node *r 407 static void dump_blkd_tasks(struct rcu_node *rnp, int ncheck);
479 static void rcu_initiate_boost(struct rcu_node 408 static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags);
480 static void rcu_preempt_boost_start_gp(struct 409 static void rcu_preempt_boost_start_gp(struct rcu_node *rnp);
481 static bool rcu_is_callbacks_kthread(struct rc !! 410 static void invoke_rcu_callbacks_kthread(void);
482 static void rcu_cpu_kthread_setup(unsigned int !! 411 static bool rcu_is_callbacks_kthread(void);
483 static void rcu_spawn_one_boost_kthread(struct !! 412 static void __init rcu_spawn_boost_kthreads(void);
>> 413 static void rcu_prepare_kthreads(int cpu);
>> 414 static void rcu_cleanup_after_idle(void);
>> 415 static void rcu_prepare_for_idle(void);
484 static bool rcu_preempt_has_tasks(struct rcu_n 416 static bool rcu_preempt_has_tasks(struct rcu_node *rnp);
485 static bool rcu_preempt_need_deferred_qs(struc 417 static bool rcu_preempt_need_deferred_qs(struct task_struct *t);
>> 418 static void rcu_preempt_deferred_qs(struct task_struct *t);
486 static void zero_cpu_stall_ticks(struct rcu_da 419 static void zero_cpu_stall_ticks(struct rcu_data *rdp);
>> 420 static bool rcu_nocb_cpu_needs_barrier(int cpu);
487 static struct swait_queue_head *rcu_nocb_gp_ge 421 static struct swait_queue_head *rcu_nocb_gp_get(struct rcu_node *rnp);
488 static void rcu_nocb_gp_cleanup(struct swait_q 422 static void rcu_nocb_gp_cleanup(struct swait_queue_head *sq);
489 static void rcu_init_one_nocb(struct rcu_node 423 static void rcu_init_one_nocb(struct rcu_node *rnp);
490 static bool wake_nocb_gp(struct rcu_data *rdp, !! 424 static bool __call_rcu_nocb(struct rcu_data *rdp, struct rcu_head *rhp,
491 static bool rcu_nocb_flush_bypass(struct rcu_d !! 425 bool lazy, unsigned long flags);
492 unsigned lon !! 426 static bool rcu_nocb_adopt_orphan_cbs(struct rcu_data *my_rdp,
493 static void call_rcu_nocb(struct rcu_data *rdp !! 427 struct rcu_data *rdp,
494 rcu_callback_t func, !! 428 unsigned long flags);
495 static void __maybe_unused __call_rcu_nocb_wak !! 429 static int rcu_nocb_need_deferred_wakeup(struct rcu_data *rdp);
496 !! 430 static void do_nocb_deferred_wakeup(struct rcu_data *rdp);
497 static int rcu_nocb_need_deferred_wakeup(struc <<
498 static bool do_nocb_deferred_wakeup(struct rcu <<
499 static void rcu_boot_init_nocb_percpu_data(str 431 static void rcu_boot_init_nocb_percpu_data(struct rcu_data *rdp);
500 static void rcu_spawn_cpu_nocb_kthread(int cpu 432 static void rcu_spawn_cpu_nocb_kthread(int cpu);
501 static void show_rcu_nocb_state(struct rcu_dat !! 433 static void __init rcu_spawn_nocb_kthreads(void);
502 static void rcu_nocb_lock(struct rcu_data *rdp <<
503 static void rcu_nocb_unlock(struct rcu_data *r <<
504 static void rcu_nocb_unlock_irqrestore(struct <<
505 unsigne <<
506 static void rcu_lockdep_assert_cblist_protecte <<
507 #ifdef CONFIG_RCU_NOCB_CPU 434 #ifdef CONFIG_RCU_NOCB_CPU
508 static void __init rcu_organize_nocb_kthreads( 435 static void __init rcu_organize_nocb_kthreads(void);
509 !! 436 #endif /* #ifdef CONFIG_RCU_NOCB_CPU */
510 /* !! 437 static bool init_nocb_callback_list(struct rcu_data *rdp);
511 * Disable IRQs before checking offloaded stat !! 438 static unsigned long rcu_get_n_cbs_nocb_cpu(struct rcu_data *rdp);
512 * locking is safe against concurrent de-offlo <<
513 */ <<
514 #define rcu_nocb_lock_irqsave(rdp, flags) <<
515 do { <<
516 local_irq_save(flags); <<
517 if (rcu_segcblist_is_offloaded(&(rdp)- <<
518 raw_spin_lock(&(rdp)->nocb_loc <<
519 } while (0) <<
520 #else /* #ifdef CONFIG_RCU_NOCB_CPU */ <<
521 #define rcu_nocb_lock_irqsave(rdp, flags) loca <<
522 #endif /* #else #ifdef CONFIG_RCU_NOCB_CPU */ <<
523 <<
524 static void rcu_bind_gp_kthread(void); 439 static void rcu_bind_gp_kthread(void);
525 static bool rcu_nohz_full_cpu(void); 440 static bool rcu_nohz_full_cpu(void);
>> 441 static void rcu_dynticks_task_enter(void);
>> 442 static void rcu_dynticks_task_exit(void);
526 443
527 /* Forward declarations for tree_stall.h */ 444 /* Forward declarations for tree_stall.h */
528 static void record_gp_stall_check_time(void); 445 static void record_gp_stall_check_time(void);
529 static void rcu_iw_handler(struct irq_work *iw 446 static void rcu_iw_handler(struct irq_work *iwp);
530 static void check_cpu_stall(struct rcu_data *r 447 static void check_cpu_stall(struct rcu_data *rdp);
531 static void rcu_check_gp_start_stall(struct rc 448 static void rcu_check_gp_start_stall(struct rcu_node *rnp, struct rcu_data *rdp,
532 const uns 449 const unsigned long gpssdelay);
533 <<
534 /* Forward declarations for tree_exp.h. */ <<
535 static void sync_rcu_do_polled_gp(struct work_ <<
536 450
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.