xref: /linux-6.15/include/linux/rcutree.h (revision e73173db)
1 /*
2  * Read-Copy Update mechanism for mutual exclusion (tree-based version)
3  *
4  * This program is free software; you can redistribute it and/or modify
5  * it under the terms of the GNU General Public License as published by
6  * the Free Software Foundation; either version 2 of the License, or
7  * (at your option) any later version.
8  *
9  * This program is distributed in the hope that it will be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12  * GNU General Public License for more details.
13  *
14  * You should have received a copy of the GNU General Public License
15  * along with this program; if not, write to the Free Software
16  * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
17  *
18  * Copyright IBM Corporation, 2008
19  *
20  * Author: Dipankar Sarma <[email protected]>
21  *	   Paul E. McKenney <[email protected]> Hierarchical algorithm
22  *
23  * Based on the original work by Paul McKenney <[email protected]>
24  * and inputs from Rusty Russell, Andrea Arcangeli and Andi Kleen.
25  *
26  * For detailed explanation of Read-Copy Update mechanism see -
27  * 	Documentation/RCU
28  */
29 
30 #ifndef __LINUX_RCUTREE_H
31 #define __LINUX_RCUTREE_H
32 
33 #include <linux/cache.h>
34 #include <linux/spinlock.h>
35 #include <linux/threads.h>
36 #include <linux/cpumask.h>
37 #include <linux/seqlock.h>
38 
39 /*
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  */
78 struct 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  */
89 struct 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. */
111 struct 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) For future __rcu_pending statistics. */
165 	long n_rcu_pending;		/* rcu_pending() calls since boot. */
166 
167 	int cpu;
168 };
169 
170 /* Values for signaled field in struct rcu_state. */
171 #define RCU_GP_INIT		0	/* Grace period being initialized. */
172 #define RCU_SAVE_DYNTICK	1	/* Need to scan dyntick state. */
173 #define RCU_FORCE_QS		2	/* Need to force quiescent state. */
174 #ifdef CONFIG_NO_HZ
175 #define RCU_SIGNAL_INIT		RCU_SAVE_DYNTICK
176 #else /* #ifdef CONFIG_NO_HZ */
177 #define RCU_SIGNAL_INIT		RCU_FORCE_QS
178 #endif /* #else #ifdef CONFIG_NO_HZ */
179 
180 #define RCU_JIFFIES_TILL_FORCE_QS	 3	/* for rsp->jiffies_force_qs */
181 #ifdef CONFIG_RCU_CPU_STALL_DETECTOR
182 #define RCU_SECONDS_TILL_STALL_CHECK   (10 * HZ)  /* for rsp->jiffies_stall */
183 #define RCU_SECONDS_TILL_STALL_RECHECK (30 * HZ)  /* for rsp->jiffies_stall */
184 #define RCU_STALL_RAT_DELAY		2	  /* Allow other CPUs time */
185 						  /*  to take at least one */
186 						  /*  scheduling clock irq */
187 						  /*  before ratting on them. */
188 
189 #endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
190 
191 /*
192  * RCU global state, including node hierarchy.  This hierarchy is
193  * represented in "heap" form in a dense array.  The root (first level)
194  * of the hierarchy is in ->node[0] (referenced by ->level[0]), the second
195  * level in ->node[1] through ->node[m] (->node[1] referenced by ->level[1]),
196  * and the third level in ->node[m+1] and following (->node[m+1] referenced
197  * by ->level[2]).  The number of levels is determined by the number of
198  * CPUs and by CONFIG_RCU_FANOUT.  Small systems will have a "hierarchy"
199  * consisting of a single rcu_node.
200  */
201 struct rcu_state {
202 	struct rcu_node node[NUM_RCU_NODES];	/* Hierarchy. */
203 	struct rcu_node *level[NUM_RCU_LVLS];	/* Hierarchy levels. */
204 	u32 levelcnt[MAX_RCU_LVLS + 1];		/* # nodes in each level. */
205 	u8 levelspread[NUM_RCU_LVLS];		/* kids/node in each level. */
206 	struct rcu_data *rda[NR_CPUS];		/* array of rdp pointers. */
207 
208 	/* The following fields are guarded by the root rcu_node's lock. */
209 
210 	u8	signaled ____cacheline_internodealigned_in_smp;
211 						/* Force QS state. */
212 	long	gpnum;				/* Current gp number. */
213 	long	completed;			/* # of last completed gp. */
214 	spinlock_t onofflock;			/* exclude on/offline and */
215 						/*  starting new GP. */
216 	spinlock_t fqslock;			/* Only one task forcing */
217 						/*  quiescent states. */
218 	unsigned long jiffies_force_qs;		/* Time at which to invoke */
219 						/*  force_quiescent_state(). */
220 	unsigned long n_force_qs;		/* Number of calls to */
221 						/*  force_quiescent_state(). */
222 	unsigned long n_force_qs_lh;		/* ~Number of calls leaving */
223 						/*  due to lock unavailable. */
224 	unsigned long n_force_qs_ngp;		/* Number of calls leaving */
225 						/*  due to no GP active. */
226 #ifdef CONFIG_RCU_CPU_STALL_DETECTOR
227 	unsigned long gp_start;			/* Time at which GP started, */
228 						/*  but in jiffies. */
229 	unsigned long jiffies_stall;		/* Time at which to check */
230 						/*  for CPU stalls. */
231 #endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
232 #ifdef CONFIG_NO_HZ
233 	long dynticks_completed;		/* Value of completed @ snap. */
234 #endif /* #ifdef CONFIG_NO_HZ */
235 };
236 
237 extern void rcu_qsctr_inc(int cpu);
238 extern void rcu_bh_qsctr_inc(int cpu);
239 
240 extern int rcu_pending(int cpu);
241 extern int rcu_needs_cpu(int cpu);
242 
243 #ifdef CONFIG_DEBUG_LOCK_ALLOC
244 extern struct lockdep_map rcu_lock_map;
245 # define rcu_read_acquire()	\
246 			lock_acquire(&rcu_lock_map, 0, 0, 2, 1, NULL, _THIS_IP_)
247 # define rcu_read_release()	lock_release(&rcu_lock_map, 1, _THIS_IP_)
248 #else
249 # define rcu_read_acquire()	do { } while (0)
250 # define rcu_read_release()	do { } while (0)
251 #endif
252 
253 static inline void __rcu_read_lock(void)
254 {
255 	preempt_disable();
256 	__acquire(RCU);
257 	rcu_read_acquire();
258 }
259 static inline void __rcu_read_unlock(void)
260 {
261 	rcu_read_release();
262 	__release(RCU);
263 	preempt_enable();
264 }
265 static inline void __rcu_read_lock_bh(void)
266 {
267 	local_bh_disable();
268 	__acquire(RCU_BH);
269 	rcu_read_acquire();
270 }
271 static inline void __rcu_read_unlock_bh(void)
272 {
273 	rcu_read_release();
274 	__release(RCU_BH);
275 	local_bh_enable();
276 }
277 
278 #define __synchronize_sched() synchronize_rcu()
279 
280 #define call_rcu_sched(head, func) call_rcu(head, func)
281 
282 static inline void rcu_init_sched(void)
283 {
284 }
285 
286 extern void __rcu_init(void);
287 extern void rcu_check_callbacks(int cpu, int user);
288 extern void rcu_restart_cpu(int cpu);
289 
290 extern long rcu_batches_completed(void);
291 extern long rcu_batches_completed_bh(void);
292 
293 #ifdef CONFIG_NO_HZ
294 void rcu_enter_nohz(void);
295 void rcu_exit_nohz(void);
296 #else /* CONFIG_NO_HZ */
297 static inline void rcu_enter_nohz(void)
298 {
299 }
300 static inline void rcu_exit_nohz(void)
301 {
302 }
303 #endif /* CONFIG_NO_HZ */
304 
305 /* A context switch is a grace period for rcutree. */
306 static inline int rcu_blocking_is_gp(void)
307 {
308 	return num_online_cpus() == 1;
309 }
310 
311 #endif /* __LINUX_RCUTREE_H */
312