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