aboutsummaryrefslogtreecommitdiffstats
path: root/include/linux/rcutree.h
diff options
context:
space:
mode:
Diffstat (limited to 'include/linux/rcutree.h')
-rw-r--r--include/linux/rcutree.h262
1 files changed, 30 insertions, 232 deletions
diff --git a/include/linux/rcutree.h b/include/linux/rcutree.h
index 5a5153806c42..a89307717825 100644
--- a/include/linux/rcutree.h
+++ b/include/linux/rcutree.h
@@ -30,264 +30,57 @@
30#ifndef __LINUX_RCUTREE_H 30#ifndef __LINUX_RCUTREE_H
31#define __LINUX_RCUTREE_H 31#define __LINUX_RCUTREE_H
32 32
33#include <linux/cache.h> 33extern void rcu_sched_qs(int cpu);
34#include <linux/spinlock.h> 34extern void rcu_bh_qs(int cpu);
35#include <linux/threads.h>
36#include <linux/cpumask.h>
37#include <linux/seqlock.h>
38 35
39/* 36extern int rcu_needs_cpu(int cpu);
40 * Define shape of hierarchy based on NR_CPUS and CONFIG_RCU_FANOUT.
41 * In theory, it should be possible to add more levels straightforwardly.
42 * In practice, this has not been tested, so there is probably some
43 * bug somewhere.
44 */
45#define MAX_RCU_LVLS 3
46#define RCU_FANOUT (CONFIG_RCU_FANOUT)
47#define RCU_FANOUT_SQ (RCU_FANOUT * RCU_FANOUT)
48#define RCU_FANOUT_CUBE (RCU_FANOUT_SQ * RCU_FANOUT)
49
50#if NR_CPUS <= RCU_FANOUT
51# define NUM_RCU_LVLS 1
52# define NUM_RCU_LVL_0 1
53# define NUM_RCU_LVL_1 (NR_CPUS)
54# define NUM_RCU_LVL_2 0
55# define NUM_RCU_LVL_3 0
56#elif NR_CPUS <= RCU_FANOUT_SQ
57# define NUM_RCU_LVLS 2
58# define NUM_RCU_LVL_0 1
59# define NUM_RCU_LVL_1 (((NR_CPUS) + RCU_FANOUT - 1) / RCU_FANOUT)
60# define NUM_RCU_LVL_2 (NR_CPUS)
61# define NUM_RCU_LVL_3 0
62#elif NR_CPUS <= RCU_FANOUT_CUBE
63# define NUM_RCU_LVLS 3
64# define NUM_RCU_LVL_0 1
65# define NUM_RCU_LVL_1 (((NR_CPUS) + RCU_FANOUT_SQ - 1) / RCU_FANOUT_SQ)
66# define NUM_RCU_LVL_2 (((NR_CPUS) + (RCU_FANOUT) - 1) / (RCU_FANOUT))
67# define NUM_RCU_LVL_3 NR_CPUS
68#else
69# error "CONFIG_RCU_FANOUT insufficient for NR_CPUS"
70#endif /* #if (NR_CPUS) <= RCU_FANOUT */
71
72#define RCU_SUM (NUM_RCU_LVL_0 + NUM_RCU_LVL_1 + NUM_RCU_LVL_2 + NUM_RCU_LVL_3)
73#define NUM_RCU_NODES (RCU_SUM - NR_CPUS)
74
75/*
76 * Dynticks per-CPU state.
77 */
78struct rcu_dynticks {
79 int dynticks_nesting; /* Track nesting level, sort of. */
80 int dynticks; /* Even value for dynticks-idle, else odd. */
81 int dynticks_nmi; /* Even value for either dynticks-idle or */
82 /* not in nmi handler, else odd. So this */
83 /* remains even for nmi from irq handler. */
84};
85
86/*
87 * Definition for node within the RCU grace-period-detection hierarchy.
88 */
89struct rcu_node {
90 spinlock_t lock;
91 unsigned long qsmask; /* CPUs or groups that need to switch in */
92 /* order for current grace period to proceed.*/
93 unsigned long qsmaskinit;
94 /* Per-GP initialization for qsmask. */
95 unsigned long grpmask; /* Mask to apply to parent qsmask. */
96 int grplo; /* lowest-numbered CPU or group here. */
97 int grphi; /* highest-numbered CPU or group here. */
98 u8 grpnum; /* CPU/group number for next level up. */
99 u8 level; /* root is at level 0. */
100 struct rcu_node *parent;
101} ____cacheline_internodealigned_in_smp;
102
103/* Index values for nxttail array in struct rcu_data. */
104#define RCU_DONE_TAIL 0 /* Also RCU_WAIT head. */
105#define RCU_WAIT_TAIL 1 /* Also RCU_NEXT_READY head. */
106#define RCU_NEXT_READY_TAIL 2 /* Also RCU_NEXT head. */
107#define RCU_NEXT_TAIL 3
108#define RCU_NEXT_SIZE 4
109
110/* Per-CPU data for read-copy update. */
111struct rcu_data {
112 /* 1) quiescent-state and grace-period handling : */
113 long completed; /* Track rsp->completed gp number */
114 /* in order to detect GP end. */
115 long gpnum; /* Highest gp number that this CPU */
116 /* is aware of having started. */
117 long passed_quiesc_completed;
118 /* Value of completed at time of qs. */
119 bool passed_quiesc; /* User-mode/idle loop etc. */
120 bool qs_pending; /* Core waits for quiesc state. */
121 bool beenonline; /* CPU online at least once. */
122 struct rcu_node *mynode; /* This CPU's leaf of hierarchy */
123 unsigned long grpmask; /* Mask to apply to leaf qsmask. */
124
125 /* 2) batch handling */
126 /*
127 * If nxtlist is not NULL, it is partitioned as follows.
128 * Any of the partitions might be empty, in which case the
129 * pointer to that partition will be equal to the pointer for
130 * the following partition. When the list is empty, all of
131 * the nxttail elements point to nxtlist, which is NULL.
132 *
133 * [*nxttail[RCU_NEXT_READY_TAIL], NULL = *nxttail[RCU_NEXT_TAIL]):
134 * Entries that might have arrived after current GP ended
135 * [*nxttail[RCU_WAIT_TAIL], *nxttail[RCU_NEXT_READY_TAIL]):
136 * Entries known to have arrived before current GP ended
137 * [*nxttail[RCU_DONE_TAIL], *nxttail[RCU_WAIT_TAIL]):
138 * Entries that batch # <= ->completed - 1: waiting for current GP
139 * [nxtlist, *nxttail[RCU_DONE_TAIL]):
140 * Entries that batch # <= ->completed
141 * The grace period for these entries has completed, and
142 * the other grace-period-completed entries may be moved
143 * here temporarily in rcu_process_callbacks().
144 */
145 struct rcu_head *nxtlist;
146 struct rcu_head **nxttail[RCU_NEXT_SIZE];
147 long qlen; /* # of queued callbacks */
148 long blimit; /* Upper limit on a processed batch */
149
150#ifdef CONFIG_NO_HZ
151 /* 3) dynticks interface. */
152 struct rcu_dynticks *dynticks; /* Shared per-CPU dynticks state. */
153 int dynticks_snap; /* Per-GP tracking for dynticks. */
154 int dynticks_nmi_snap; /* Per-GP tracking for dynticks_nmi. */
155#endif /* #ifdef CONFIG_NO_HZ */
156
157 /* 4) reasons this CPU needed to be kicked by force_quiescent_state */
158#ifdef CONFIG_NO_HZ
159 unsigned long dynticks_fqs; /* Kicked due to dynticks idle. */
160#endif /* #ifdef CONFIG_NO_HZ */
161 unsigned long offline_fqs; /* Kicked due to being offline. */
162 unsigned long resched_ipi; /* Sent a resched IPI. */
163
164 /* 5) __rcu_pending() statistics. */
165 long n_rcu_pending; /* rcu_pending() calls since boot. */
166 long n_rp_qs_pending;
167 long n_rp_cb_ready;
168 long n_rp_cpu_needs_gp;
169 long n_rp_gp_completed;
170 long n_rp_gp_started;
171 long n_rp_need_fqs;
172 long n_rp_need_nothing;
173
174 int cpu;
175};
176
177/* Values for signaled field in struct rcu_state. */
178#define RCU_GP_INIT 0 /* Grace period being initialized. */
179#define RCU_SAVE_DYNTICK 1 /* Need to scan dyntick state. */
180#define RCU_FORCE_QS 2 /* Need to force quiescent state. */
181#ifdef CONFIG_NO_HZ
182#define RCU_SIGNAL_INIT RCU_SAVE_DYNTICK
183#else /* #ifdef CONFIG_NO_HZ */
184#define RCU_SIGNAL_INIT RCU_FORCE_QS
185#endif /* #else #ifdef CONFIG_NO_HZ */
186
187#define RCU_JIFFIES_TILL_FORCE_QS 3 /* for rsp->jiffies_force_qs */
188#ifdef CONFIG_RCU_CPU_STALL_DETECTOR
189#define RCU_SECONDS_TILL_STALL_CHECK (10 * HZ) /* for rsp->jiffies_stall */
190#define RCU_SECONDS_TILL_STALL_RECHECK (30 * HZ) /* for rsp->jiffies_stall */
191#define RCU_STALL_RAT_DELAY 2 /* Allow other CPUs time */
192 /* to take at least one */
193 /* scheduling clock irq */
194 /* before ratting on them. */
195
196#endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
197
198/*
199 * RCU global state, including node hierarchy. This hierarchy is
200 * represented in "heap" form in a dense array. The root (first level)
201 * of the hierarchy is in ->node[0] (referenced by ->level[0]), the second
202 * level in ->node[1] through ->node[m] (->node[1] referenced by ->level[1]),
203 * and the third level in ->node[m+1] and following (->node[m+1] referenced
204 * by ->level[2]). The number of levels is determined by the number of
205 * CPUs and by CONFIG_RCU_FANOUT. Small systems will have a "hierarchy"
206 * consisting of a single rcu_node.
207 */
208struct rcu_state {
209 struct rcu_node node[NUM_RCU_NODES]; /* Hierarchy. */
210 struct rcu_node *level[NUM_RCU_LVLS]; /* Hierarchy levels. */
211 u32 levelcnt[MAX_RCU_LVLS + 1]; /* # nodes in each level. */
212 u8 levelspread[NUM_RCU_LVLS]; /* kids/node in each level. */
213 struct rcu_data *rda[NR_CPUS]; /* array of rdp pointers. */
214
215 /* The following fields are guarded by the root rcu_node's lock. */
216
217 u8 signaled ____cacheline_internodealigned_in_smp;
218 /* Force QS state. */
219 long gpnum; /* Current gp number. */
220 long completed; /* # of last completed gp. */
221 spinlock_t onofflock; /* exclude on/offline and */
222 /* starting new GP. */
223 spinlock_t fqslock; /* Only one task forcing */
224 /* quiescent states. */
225 unsigned long jiffies_force_qs; /* Time at which to invoke */
226 /* force_quiescent_state(). */
227 unsigned long n_force_qs; /* Number of calls to */
228 /* force_quiescent_state(). */
229 unsigned long n_force_qs_lh; /* ~Number of calls leaving */
230 /* due to lock unavailable. */
231 unsigned long n_force_qs_ngp; /* Number of calls leaving */
232 /* due to no GP active. */
233#ifdef CONFIG_RCU_CPU_STALL_DETECTOR
234 unsigned long gp_start; /* Time at which GP started, */
235 /* but in jiffies. */
236 unsigned long jiffies_stall; /* Time at which to check */
237 /* for CPU stalls. */
238#endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
239#ifdef CONFIG_NO_HZ
240 long dynticks_completed; /* Value of completed @ snap. */
241#endif /* #ifdef CONFIG_NO_HZ */
242};
243 37
244extern void rcu_qsctr_inc(int cpu); 38#ifdef CONFIG_TREE_PREEMPT_RCU
245extern void rcu_bh_qsctr_inc(int cpu);
246 39
247extern int rcu_pending(int cpu); 40extern void __rcu_read_lock(void);
248extern int rcu_needs_cpu(int cpu); 41extern void __rcu_read_unlock(void);
42extern void exit_rcu(void);
249 43
250#ifdef CONFIG_DEBUG_LOCK_ALLOC 44#else /* #ifdef CONFIG_TREE_PREEMPT_RCU */
251extern struct lockdep_map rcu_lock_map;
252# define rcu_read_acquire() \
253 lock_acquire(&rcu_lock_map, 0, 0, 2, 1, NULL, _THIS_IP_)
254# define rcu_read_release() lock_release(&rcu_lock_map, 1, _THIS_IP_)
255#else
256# define rcu_read_acquire() do { } while (0)
257# define rcu_read_release() do { } while (0)
258#endif
259 45
260static inline void __rcu_read_lock(void) 46static inline void __rcu_read_lock(void)
261{ 47{
262 preempt_disable(); 48 preempt_disable();
263 __acquire(RCU);
264 rcu_read_acquire();
265} 49}
50
266static inline void __rcu_read_unlock(void) 51static inline void __rcu_read_unlock(void)
267{ 52{
268 rcu_read_release();
269 __release(RCU);
270 preempt_enable(); 53 preempt_enable();
271} 54}
55
56static inline void exit_rcu(void)
57{
58}
59
60#endif /* #else #ifdef CONFIG_TREE_PREEMPT_RCU */
61
272static inline void __rcu_read_lock_bh(void) 62static inline void __rcu_read_lock_bh(void)
273{ 63{
274 local_bh_disable(); 64 local_bh_disable();
275 __acquire(RCU_BH);
276 rcu_read_acquire();
277} 65}
278static inline void __rcu_read_unlock_bh(void) 66static inline void __rcu_read_unlock_bh(void)
279{ 67{
280 rcu_read_release();
281 __release(RCU_BH);
282 local_bh_enable(); 68 local_bh_enable();
283} 69}
284 70
285#define __synchronize_sched() synchronize_rcu() 71#define __synchronize_sched() synchronize_rcu()
286 72
287#define call_rcu_sched(head, func) call_rcu(head, func) 73extern void call_rcu_sched(struct rcu_head *head,
74 void (*func)(struct rcu_head *rcu));
288 75
289static inline void rcu_init_sched(void) 76static inline void synchronize_rcu_expedited(void)
290{ 77{
78 synchronize_sched_expedited();
79}
80
81static inline void synchronize_rcu_bh_expedited(void)
82{
83 synchronize_sched_expedited();
291} 84}
292 85
293extern void __rcu_init(void); 86extern void __rcu_init(void);
@@ -296,6 +89,11 @@ extern void rcu_restart_cpu(int cpu);
296 89
297extern long rcu_batches_completed(void); 90extern long rcu_batches_completed(void);
298extern long rcu_batches_completed_bh(void); 91extern long rcu_batches_completed_bh(void);
92extern long rcu_batches_completed_sched(void);
93
94static inline void rcu_init_sched(void)
95{
96}
299 97
300#ifdef CONFIG_NO_HZ 98#ifdef CONFIG_NO_HZ
301void rcu_enter_nohz(void); 99void rcu_enter_nohz(void);
generated by cgit v1.2.2 (git 2.25.0) at 2024-11-15 13:35:23 -0500