[go: up one dir, main page]
1/* SPDX-License-Identifier: GPL-2.0+ */
2/*
3 * Sleepable Read-Copy Update mechanism for mutual exclusion,
4 * tree variant.
5 *
6 * Copyright (C) IBM Corporation, 2017
7 *
8 * Author: Paul McKenney <paulmck@linux.ibm.com>
9 */
10
11#ifndef _LINUX_SRCU_TREE_H
12#define _LINUX_SRCU_TREE_H
13
14#include <linux/rcu_node_tree.h>
15#include <linux/completion.h>
16
17struct srcu_node;
18struct srcu_struct;
19
20/* One element of the srcu_data srcu_ctrs array. */
21struct srcu_ctr {
22 atomic_long_t srcu_locks; /* Locks per CPU. */
23 atomic_long_t srcu_unlocks; /* Unlocks per CPU. */
24};
25
26/*
27 * Per-CPU structure feeding into leaf srcu_node, similar in function
28 * to rcu_node.
29 */
30struct srcu_data {
31 /* Read-side state. */
32 struct srcu_ctr srcu_ctrs[2]; /* Locks and unlocks per CPU. */
33 int srcu_reader_flavor; /* Reader flavor for srcu_struct structure? */
34 /* Values: SRCU_READ_FLAVOR_.* */
35
36 /* Update-side state. */
37 spinlock_t __private lock ____cacheline_internodealigned_in_smp;
38 struct rcu_segcblist srcu_cblist; /* List of callbacks.*/
39 unsigned long srcu_gp_seq_needed; /* Furthest future GP needed. */
40 unsigned long srcu_gp_seq_needed_exp; /* Furthest future exp GP. */
41 bool srcu_cblist_invoking; /* Invoking these CBs? */
42 struct timer_list delay_work; /* Delay for CB invoking */
43 struct work_struct work; /* Context for CB invoking. */
44 struct rcu_head srcu_barrier_head; /* For srcu_barrier() use. */
45 struct rcu_head srcu_ec_head; /* For srcu_expedite_current() use. */
46 int srcu_ec_state; /* State for srcu_expedite_current(). */
47 struct srcu_node *mynode; /* Leaf srcu_node. */
48 unsigned long grpmask; /* Mask for leaf srcu_node */
49 /* ->srcu_data_have_cbs[]. */
50 int cpu;
51 struct srcu_struct *ssp;
52};
53
54/*
55 * Node in SRCU combining tree, similar in function to rcu_data.
56 */
57struct srcu_node {
58 spinlock_t __private lock;
59 unsigned long srcu_have_cbs[4]; /* GP seq for children having CBs, but only */
60 /* if greater than ->srcu_gp_seq. */
61 unsigned long srcu_data_have_cbs[4]; /* Which srcu_data structs have CBs for given GP? */
62 unsigned long srcu_gp_seq_needed_exp; /* Furthest future exp GP. */
63 struct srcu_node *srcu_parent; /* Next up in tree. */
64 int grplo; /* Least CPU for node. */
65 int grphi; /* Biggest CPU for node. */
66};
67
68/*
69 * Per-SRCU-domain structure, update-side data linked from srcu_struct.
70 */
71struct srcu_usage {
72 struct srcu_node *node; /* Combining tree. */
73 struct srcu_node *level[RCU_NUM_LVLS + 1];
74 /* First node at each level. */
75 int srcu_size_state; /* Small-to-big transition state. */
76 struct mutex srcu_cb_mutex; /* Serialize CB preparation. */
77 spinlock_t __private lock; /* Protect counters and size state. */
78 struct mutex srcu_gp_mutex; /* Serialize GP work. */
79 unsigned long srcu_gp_seq; /* Grace-period seq #. */
80 unsigned long srcu_gp_seq_needed; /* Latest gp_seq needed. */
81 unsigned long srcu_gp_seq_needed_exp; /* Furthest future exp GP. */
82 unsigned long srcu_gp_start; /* Last GP start timestamp (jiffies) */
83 unsigned long srcu_last_gp_end; /* Last GP end timestamp (ns) */
84 unsigned long srcu_size_jiffies; /* Current contention-measurement interval. */
85 unsigned long srcu_n_lock_retries; /* Contention events in current interval. */
86 unsigned long srcu_n_exp_nodelay; /* # expedited no-delays in current GP phase. */
87 bool sda_is_static; /* May ->sda be passed to free_percpu()? */
88 unsigned long srcu_barrier_seq; /* srcu_barrier seq #. */
89 struct mutex srcu_barrier_mutex; /* Serialize barrier ops. */
90 struct completion srcu_barrier_completion;
91 /* Awaken barrier rq at end. */
92 atomic_t srcu_barrier_cpu_cnt; /* # CPUs not yet posting a */
93 /* callback for the barrier */
94 /* operation. */
95 unsigned long reschedule_jiffies;
96 unsigned long reschedule_count;
97 struct delayed_work work;
98 struct srcu_struct *srcu_ssp;
99};
100
101/*
102 * Per-SRCU-domain structure, similar in function to rcu_state.
103 */
104struct srcu_struct {
105 struct srcu_ctr __percpu *srcu_ctrp;
106 struct srcu_data __percpu *sda; /* Per-CPU srcu_data array. */
107 u8 srcu_reader_flavor;
108 struct lockdep_map dep_map;
109 struct srcu_usage *srcu_sup; /* Update-side data. */
110};
111
112// Values for size state variable (->srcu_size_state). Once the state
113// has been set to SRCU_SIZE_ALLOC, the grace-period code advances through
114// this state machine one step per grace period until the SRCU_SIZE_BIG state
115// is reached. Otherwise, the state machine remains in the SRCU_SIZE_SMALL
116// state indefinitely.
117#define SRCU_SIZE_SMALL 0 // No srcu_node combining tree, ->node == NULL
118#define SRCU_SIZE_ALLOC 1 // An srcu_node tree is being allocated, initialized,
119 // and then referenced by ->node. It will not be used.
120#define SRCU_SIZE_WAIT_BARRIER 2 // The srcu_node tree starts being used by everything
121 // except call_srcu(), especially by srcu_barrier().
122 // By the end of this state, all CPUs and threads
123 // are aware of this tree's existence.
124#define SRCU_SIZE_WAIT_CALL 3 // The srcu_node tree starts being used by call_srcu().
125 // By the end of this state, all of the call_srcu()
126 // invocations that were running on a non-boot CPU
127 // and using the boot CPU's callback queue will have
128 // completed.
129#define SRCU_SIZE_WAIT_CBS1 4 // Don't trust the ->srcu_have_cbs[] grace-period
130#define SRCU_SIZE_WAIT_CBS2 5 // sequence elements or the ->srcu_data_have_cbs[]
131#define SRCU_SIZE_WAIT_CBS3 6 // CPU-bitmask elements until all four elements of
132#define SRCU_SIZE_WAIT_CBS4 7 // each array have been initialized.
133#define SRCU_SIZE_BIG 8 // The srcu_node combining tree is fully initialized
134 // and all aspects of it are being put to use.
135
136/* Values for state variable (bottom bits of ->srcu_gp_seq). */
137#define SRCU_STATE_IDLE 0
138#define SRCU_STATE_SCAN1 1
139#define SRCU_STATE_SCAN2 2
140
141/* Values for srcu_expedite_current() state (->srcu_ec_state). */
142#define SRCU_EC_IDLE 0
143#define SRCU_EC_PENDING 1
144#define SRCU_EC_REPOST 2
145
146/*
147 * Values for initializing gp sequence fields. Higher values allow wrap arounds to
148 * occur earlier.
149 * The second value with state is useful in the case of static initialization of
150 * srcu_usage where srcu_gp_seq_needed is expected to have some state value in its
151 * lower bits (or else it will appear to be already initialized within
152 * the call check_init_srcu_struct()).
153 */
154#define SRCU_GP_SEQ_INITIAL_VAL ((0UL - 100UL) << RCU_SEQ_CTR_SHIFT)
155#define SRCU_GP_SEQ_INITIAL_VAL_WITH_STATE (SRCU_GP_SEQ_INITIAL_VAL - 1)
156
157#define __SRCU_USAGE_INIT(name) \
158{ \
159 .lock = __SPIN_LOCK_UNLOCKED(name.lock), \
160 .srcu_gp_seq = SRCU_GP_SEQ_INITIAL_VAL, \
161 .srcu_gp_seq_needed = SRCU_GP_SEQ_INITIAL_VAL_WITH_STATE, \
162 .srcu_gp_seq_needed_exp = SRCU_GP_SEQ_INITIAL_VAL, \
163 .work = __DELAYED_WORK_INITIALIZER(name.work, NULL, 0), \
164}
165
166#define __SRCU_STRUCT_INIT_COMMON(name, usage_name, fast) \
167 .srcu_sup = &usage_name, \
168 .srcu_reader_flavor = fast, \
169 __SRCU_DEP_MAP_INIT(name)
170
171#define __SRCU_STRUCT_INIT_MODULE(name, usage_name, fast) \
172{ \
173 __SRCU_STRUCT_INIT_COMMON(name, usage_name, fast) \
174}
175
176#define __SRCU_STRUCT_INIT(name, usage_name, pcpu_name, fast) \
177{ \
178 .sda = &pcpu_name, \
179 .srcu_ctrp = &pcpu_name.srcu_ctrs[0], \
180 __SRCU_STRUCT_INIT_COMMON(name, usage_name, fast) \
181}
182
183/*
184 * Define and initialize a srcu struct at build time.
185 * Do -not- call init_srcu_struct() nor cleanup_srcu_struct() on it.
186 *
187 * Note that although DEFINE_STATIC_SRCU() hides the name from other
188 * files, the per-CPU variable rules nevertheless require that the
189 * chosen name be globally unique. These rules also prohibit use of
190 * DEFINE_STATIC_SRCU() within a function. If these rules are too
191 * restrictive, declare the srcu_struct manually. For example, in
192 * each file:
193 *
194 * static struct srcu_struct my_srcu;
195 *
196 * Then, before the first use of each my_srcu, manually initialize it:
197 *
198 * init_srcu_struct(&my_srcu);
199 *
200 * See include/linux/percpu-defs.h for the rules on per-CPU variables.
201 *
202 * DEFINE_SRCU_FAST() and DEFINE_STATIC_SRCU_FAST create an srcu_struct
203 * and associated structures whose readers must be of the SRCU-fast variety.
204 * DEFINE_SRCU_FAST_UPDOWN() and DEFINE_STATIC_SRCU_FAST_UPDOWN() create
205 * an srcu_struct and associated structures whose readers must be of the
206 * SRCU-fast-updown variety. The key point (aside from error checking) with
207 * both varieties is that the grace periods must use synchronize_rcu()
208 * instead of smp_mb(), and given that the first (for example)
209 * srcu_read_lock_fast() might race with the first synchronize_srcu(),
210 * this different must be specified at initialization time.
211 */
212#ifdef MODULE
213# define __DEFINE_SRCU(name, fast, is_static) \
214 static struct srcu_usage name##_srcu_usage = __SRCU_USAGE_INIT(name##_srcu_usage); \
215 is_static struct srcu_struct name = __SRCU_STRUCT_INIT_MODULE(name, name##_srcu_usage, \
216 fast); \
217 extern struct srcu_struct * const __srcu_struct_##name; \
218 struct srcu_struct * const __srcu_struct_##name \
219 __section("___srcu_struct_ptrs") = &name
220#else
221# define __DEFINE_SRCU(name, fast, is_static) \
222 static DEFINE_PER_CPU(struct srcu_data, name##_srcu_data); \
223 static struct srcu_usage name##_srcu_usage = __SRCU_USAGE_INIT(name##_srcu_usage); \
224 is_static struct srcu_struct name = \
225 __SRCU_STRUCT_INIT(name, name##_srcu_usage, name##_srcu_data, fast)
226#endif
227#define DEFINE_SRCU(name) __DEFINE_SRCU(name, 0, /* not static */)
228#define DEFINE_STATIC_SRCU(name) __DEFINE_SRCU(name, 0, static)
229#define DEFINE_SRCU_FAST(name) __DEFINE_SRCU(name, SRCU_READ_FLAVOR_FAST, /* not static */)
230#define DEFINE_STATIC_SRCU_FAST(name) __DEFINE_SRCU(name, SRCU_READ_FLAVOR_FAST, static)
231#define DEFINE_SRCU_FAST_UPDOWN(name) __DEFINE_SRCU(name, SRCU_READ_FLAVOR_FAST_UPDOWN, \
232 /* not static */)
233#define DEFINE_STATIC_SRCU_FAST_UPDOWN(name) \
234 __DEFINE_SRCU(name, SRCU_READ_FLAVOR_FAST_UPDOWN, static)
235
236int __srcu_read_lock(struct srcu_struct *ssp) __acquires(ssp);
237void synchronize_srcu_expedited(struct srcu_struct *ssp);
238void srcu_barrier(struct srcu_struct *ssp);
239void srcu_expedite_current(struct srcu_struct *ssp);
240void srcu_torture_stats_print(struct srcu_struct *ssp, char *tt, char *tf);
241
242// Converts a per-CPU pointer to an ->srcu_ctrs[] array element to that
243// element's index.
244static inline bool __srcu_ptr_to_ctr(struct srcu_struct *ssp, struct srcu_ctr __percpu *scpp)
245{
246 return scpp - &ssp->sda->srcu_ctrs[0];
247}
248
249// Converts an integer to a per-CPU pointer to the corresponding
250// ->srcu_ctrs[] array element.
251static inline struct srcu_ctr __percpu *__srcu_ctr_to_ptr(struct srcu_struct *ssp, int idx)
252{
253 return &ssp->sda->srcu_ctrs[idx];
254}
255
256/*
257 * Counts the new reader in the appropriate per-CPU element of the
258 * srcu_struct. Returns a pointer that must be passed to the matching
259 * srcu_read_unlock_fast().
260 *
261 * Note that both this_cpu_inc() and atomic_long_inc() are RCU read-side
262 * critical sections either because they disables interrupts, because
263 * they are a single instruction, or because they are read-modify-write
264 * atomic operations, depending on the whims of the architecture.
265 * This matters because the SRCU-fast grace-period mechanism uses either
266 * synchronize_rcu() or synchronize_rcu_expedited(), that is, RCU,
267 * *not* SRCU, in order to eliminate the need for the read-side smp_mb()
268 * invocations that are used by srcu_read_lock() and srcu_read_unlock().
269 * The __srcu_read_unlock_fast() function also relies on this same RCU
270 * (again, *not* SRCU) trick to eliminate the need for smp_mb().
271 *
272 * The key point behind this RCU trick is that if any part of a given
273 * RCU reader precedes the beginning of a given RCU grace period, then
274 * the entirety of that RCU reader and everything preceding it happens
275 * before the end of that same RCU grace period. Similarly, if any part
276 * of a given RCU reader follows the end of a given RCU grace period,
277 * then the entirety of that RCU reader and everything following it
278 * happens after the beginning of that same RCU grace period. Therefore,
279 * the operations labeled Y in __srcu_read_lock_fast() and those labeled Z
280 * in __srcu_read_unlock_fast() are ordered against the corresponding SRCU
281 * read-side critical section from the viewpoint of the SRCU grace period.
282 * This is all the ordering that is required, hence no calls to smp_mb().
283 *
284 * This means that __srcu_read_lock_fast() is not all that fast
285 * on architectures that support NMIs but do not supply NMI-safe
286 * implementations of this_cpu_inc().
287 */
288static inline struct srcu_ctr __percpu notrace *__srcu_read_lock_fast(struct srcu_struct *ssp)
289{
290 struct srcu_ctr __percpu *scp = READ_ONCE(ssp->srcu_ctrp);
291
292 if (!IS_ENABLED(CONFIG_NEED_SRCU_NMI_SAFE))
293 this_cpu_inc(scp->srcu_locks.counter); // Y, and implicit RCU reader.
294 else
295 atomic_long_inc(raw_cpu_ptr(&scp->srcu_locks)); // Y, and implicit RCU reader.
296 barrier(); /* Avoid leaking the critical section. */
297 return scp;
298}
299
300/*
301 * Removes the count for the old reader from the appropriate
302 * per-CPU element of the srcu_struct. Note that this may well be a
303 * different CPU than that which was incremented by the corresponding
304 * srcu_read_lock_fast(), but it must be within the same task.
305 *
306 * Please see the __srcu_read_lock_fast() function's header comment for
307 * information on implicit RCU readers and NMI safety.
308 */
309static inline void notrace
310__srcu_read_unlock_fast(struct srcu_struct *ssp, struct srcu_ctr __percpu *scp)
311{
312 barrier(); /* Avoid leaking the critical section. */
313 if (!IS_ENABLED(CONFIG_NEED_SRCU_NMI_SAFE))
314 this_cpu_inc(scp->srcu_unlocks.counter); // Z, and implicit RCU reader.
315 else
316 atomic_long_inc(raw_cpu_ptr(&scp->srcu_unlocks)); // Z, and implicit RCU reader.
317}
318
319/*
320 * Counts the new reader in the appropriate per-CPU element of the
321 * srcu_struct. Returns a pointer that must be passed to the matching
322 * srcu_read_unlock_fast_updown(). This type of reader is compatible
323 * with srcu_down_read_fast() and srcu_up_read_fast().
324 *
325 * See the __srcu_read_lock_fast() comment for more details.
326 */
327static inline
328struct srcu_ctr __percpu notrace *__srcu_read_lock_fast_updown(struct srcu_struct *ssp)
329{
330 struct srcu_ctr __percpu *scp = READ_ONCE(ssp->srcu_ctrp);
331
332 if (!IS_ENABLED(CONFIG_NEED_SRCU_NMI_SAFE))
333 this_cpu_inc(scp->srcu_locks.counter); // Y, and implicit RCU reader.
334 else
335 atomic_long_inc(raw_cpu_ptr(&scp->srcu_locks)); // Y, and implicit RCU reader.
336 barrier(); /* Avoid leaking the critical section. */
337 return scp;
338}
339
340/*
341 * Removes the count for the old reader from the appropriate
342 * per-CPU element of the srcu_struct. Note that this may well be a
343 * different CPU than that which was incremented by the corresponding
344 * srcu_read_lock_fast(), but it must be within the same task.
345 *
346 * Please see the __srcu_read_lock_fast() function's header comment for
347 * information on implicit RCU readers and NMI safety.
348 */
349static inline void notrace
350__srcu_read_unlock_fast_updown(struct srcu_struct *ssp, struct srcu_ctr __percpu *scp)
351{
352 barrier(); /* Avoid leaking the critical section. */
353 if (!IS_ENABLED(CONFIG_NEED_SRCU_NMI_SAFE))
354 this_cpu_inc(scp->srcu_unlocks.counter); // Z, and implicit RCU reader.
355 else
356 atomic_long_inc(raw_cpu_ptr(&scp->srcu_unlocks)); // Z, and implicit RCU reader.
357}
358
359void __srcu_check_read_flavor(struct srcu_struct *ssp, int read_flavor);
360
361// Record SRCU-reader usage type only for CONFIG_PROVE_RCU=y kernels.
362static inline void srcu_check_read_flavor(struct srcu_struct *ssp, int read_flavor)
363{
364 if (IS_ENABLED(CONFIG_PROVE_RCU))
365 __srcu_check_read_flavor(ssp, read_flavor);
366}
367
368#endif
369

source code of linux/include/linux/srcutree.h