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 <[email protected]>
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
17 struct srcu_node;
18 struct srcu_struct;
19
20 /*
21 * Per-CPU structure feeding into leaf srcu_node, similar in function
22 * to rcu_node.
23 */
24 struct srcu_data {
25 /* Read-side state. */
26 atomic_long_t srcu_lock_count[2]; /* Locks per CPU. */
27 atomic_long_t srcu_unlock_count[2]; /* Unlocks per CPU. */
28 int srcu_reader_flavor; /* Reader flavor for srcu_struct structure? */
29 /* Values: SRCU_READ_FLAVOR_.* */
30
31 /* Update-side state. */
32 spinlock_t __private lock ____cacheline_internodealigned_in_smp;
33 struct rcu_segcblist srcu_cblist; /* List of callbacks.*/
34 unsigned long srcu_gp_seq_needed; /* Furthest future GP needed. */
35 unsigned long srcu_gp_seq_needed_exp; /* Furthest future exp GP. */
36 bool srcu_cblist_invoking; /* Invoking these CBs? */
37 struct timer_list delay_work; /* Delay for CB invoking */
38 struct work_struct work; /* Context for CB invoking. */
39 struct rcu_head srcu_barrier_head; /* For srcu_barrier() use. */
40 struct srcu_node *mynode; /* Leaf srcu_node. */
41 unsigned long grpmask; /* Mask for leaf srcu_node */
42 /* ->srcu_data_have_cbs[]. */
43 int cpu;
44 struct srcu_struct *ssp;
45 };
46
47 /*
48 * Node in SRCU combining tree, similar in function to rcu_data.
49 */
50 struct srcu_node {
51 spinlock_t __private lock;
52 unsigned long srcu_have_cbs[4]; /* GP seq for children having CBs, but only */
53 /* if greater than ->srcu_gp_seq. */
54 unsigned long srcu_data_have_cbs[4]; /* Which srcu_data structs have CBs for given GP? */
55 unsigned long srcu_gp_seq_needed_exp; /* Furthest future exp GP. */
56 struct srcu_node *srcu_parent; /* Next up in tree. */
57 int grplo; /* Least CPU for node. */
58 int grphi; /* Biggest CPU for node. */
59 };
60
61 /*
62 * Per-SRCU-domain structure, update-side data linked from srcu_struct.
63 */
64 struct srcu_usage {
65 struct srcu_node *node; /* Combining tree. */
66 struct srcu_node *level[RCU_NUM_LVLS + 1];
67 /* First node at each level. */
68 int srcu_size_state; /* Small-to-big transition state. */
69 struct mutex srcu_cb_mutex; /* Serialize CB preparation. */
70 spinlock_t __private lock; /* Protect counters and size state. */
71 struct mutex srcu_gp_mutex; /* Serialize GP work. */
72 unsigned long srcu_gp_seq; /* Grace-period seq #. */
73 unsigned long srcu_gp_seq_needed; /* Latest gp_seq needed. */
74 unsigned long srcu_gp_seq_needed_exp; /* Furthest future exp GP. */
75 unsigned long srcu_gp_start; /* Last GP start timestamp (jiffies) */
76 unsigned long srcu_last_gp_end; /* Last GP end timestamp (ns) */
77 unsigned long srcu_size_jiffies; /* Current contention-measurement interval. */
78 unsigned long srcu_n_lock_retries; /* Contention events in current interval. */
79 unsigned long srcu_n_exp_nodelay; /* # expedited no-delays in current GP phase. */
80 bool sda_is_static; /* May ->sda be passed to free_percpu()? */
81 unsigned long srcu_barrier_seq; /* srcu_barrier seq #. */
82 struct mutex srcu_barrier_mutex; /* Serialize barrier ops. */
83 struct completion srcu_barrier_completion;
84 /* Awaken barrier rq at end. */
85 atomic_t srcu_barrier_cpu_cnt; /* # CPUs not yet posting a */
86 /* callback for the barrier */
87 /* operation. */
88 unsigned long reschedule_jiffies;
89 unsigned long reschedule_count;
90 struct delayed_work work;
91 struct srcu_struct *srcu_ssp;
92 };
93
94 /*
95 * Per-SRCU-domain structure, similar in function to rcu_state.
96 */
97 struct srcu_struct {
98 unsigned int srcu_idx; /* Current rdr array element. */
99 struct srcu_data __percpu *sda; /* Per-CPU srcu_data array. */
100 struct lockdep_map dep_map;
101 struct srcu_usage *srcu_sup; /* Update-side data. */
102 };
103
104 // Values for size state variable (->srcu_size_state). Once the state
105 // has been set to SRCU_SIZE_ALLOC, the grace-period code advances through
106 // this state machine one step per grace period until the SRCU_SIZE_BIG state
107 // is reached. Otherwise, the state machine remains in the SRCU_SIZE_SMALL
108 // state indefinitely.
109 #define SRCU_SIZE_SMALL 0 // No srcu_node combining tree, ->node == NULL
110 #define SRCU_SIZE_ALLOC 1 // An srcu_node tree is being allocated, initialized,
111 // and then referenced by ->node. It will not be used.
112 #define SRCU_SIZE_WAIT_BARRIER 2 // The srcu_node tree starts being used by everything
113 // except call_srcu(), especially by srcu_barrier().
114 // By the end of this state, all CPUs and threads
115 // are aware of this tree's existence.
116 #define SRCU_SIZE_WAIT_CALL 3 // The srcu_node tree starts being used by call_srcu().
117 // By the end of this state, all of the call_srcu()
118 // invocations that were running on a non-boot CPU
119 // and using the boot CPU's callback queue will have
120 // completed.
121 #define SRCU_SIZE_WAIT_CBS1 4 // Don't trust the ->srcu_have_cbs[] grace-period
122 #define SRCU_SIZE_WAIT_CBS2 5 // sequence elements or the ->srcu_data_have_cbs[]
123 #define SRCU_SIZE_WAIT_CBS3 6 // CPU-bitmask elements until all four elements of
124 #define SRCU_SIZE_WAIT_CBS4 7 // each array have been initialized.
125 #define SRCU_SIZE_BIG 8 // The srcu_node combining tree is fully initialized
126 // and all aspects of it are being put to use.
127
128 /* Values for state variable (bottom bits of ->srcu_gp_seq). */
129 #define SRCU_STATE_IDLE 0
130 #define SRCU_STATE_SCAN1 1
131 #define SRCU_STATE_SCAN2 2
132
133 /*
134 * Values for initializing gp sequence fields. Higher values allow wrap arounds to
135 * occur earlier.
136 * The second value with state is useful in the case of static initialization of
137 * srcu_usage where srcu_gp_seq_needed is expected to have some state value in its
138 * lower bits (or else it will appear to be already initialized within
139 * the call check_init_srcu_struct()).
140 */
141 #define SRCU_GP_SEQ_INITIAL_VAL ((0UL - 100UL) << RCU_SEQ_CTR_SHIFT)
142 #define SRCU_GP_SEQ_INITIAL_VAL_WITH_STATE (SRCU_GP_SEQ_INITIAL_VAL - 1)
143
144 #define __SRCU_USAGE_INIT(name) \
145 { \
146 .lock = __SPIN_LOCK_UNLOCKED(name.lock), \
147 .srcu_gp_seq = SRCU_GP_SEQ_INITIAL_VAL, \
148 .srcu_gp_seq_needed = SRCU_GP_SEQ_INITIAL_VAL_WITH_STATE, \
149 .srcu_gp_seq_needed_exp = SRCU_GP_SEQ_INITIAL_VAL, \
150 .work = __DELAYED_WORK_INITIALIZER(name.work, NULL, 0), \
151 }
152
153 #define __SRCU_STRUCT_INIT_COMMON(name, usage_name) \
154 .srcu_sup = &usage_name, \
155 __SRCU_DEP_MAP_INIT(name)
156
157 #define __SRCU_STRUCT_INIT_MODULE(name, usage_name) \
158 { \
159 __SRCU_STRUCT_INIT_COMMON(name, usage_name) \
160 }
161
162 #define __SRCU_STRUCT_INIT(name, usage_name, pcpu_name) \
163 { \
164 .sda = &pcpu_name, \
165 __SRCU_STRUCT_INIT_COMMON(name, usage_name) \
166 }
167
168 /*
169 * Define and initialize a srcu struct at build time.
170 * Do -not- call init_srcu_struct() nor cleanup_srcu_struct() on it.
171 *
172 * Note that although DEFINE_STATIC_SRCU() hides the name from other
173 * files, the per-CPU variable rules nevertheless require that the
174 * chosen name be globally unique. These rules also prohibit use of
175 * DEFINE_STATIC_SRCU() within a function. If these rules are too
176 * restrictive, declare the srcu_struct manually. For example, in
177 * each file:
178 *
179 * static struct srcu_struct my_srcu;
180 *
181 * Then, before the first use of each my_srcu, manually initialize it:
182 *
183 * init_srcu_struct(&my_srcu);
184 *
185 * See include/linux/percpu-defs.h for the rules on per-CPU variables.
186 */
187 #ifdef MODULE
188 # define __DEFINE_SRCU(name, is_static) \
189 static struct srcu_usage name##_srcu_usage = __SRCU_USAGE_INIT(name##_srcu_usage); \
190 is_static struct srcu_struct name = __SRCU_STRUCT_INIT_MODULE(name, name##_srcu_usage); \
191 extern struct srcu_struct * const __srcu_struct_##name; \
192 struct srcu_struct * const __srcu_struct_##name \
193 __section("___srcu_struct_ptrs") = &name
194 #else
195 # define __DEFINE_SRCU(name, is_static) \
196 static DEFINE_PER_CPU(struct srcu_data, name##_srcu_data); \
197 static struct srcu_usage name##_srcu_usage = __SRCU_USAGE_INIT(name##_srcu_usage); \
198 is_static struct srcu_struct name = \
199 __SRCU_STRUCT_INIT(name, name##_srcu_usage, name##_srcu_data)
200 #endif
201 #define DEFINE_SRCU(name) __DEFINE_SRCU(name, /* not static */)
202 #define DEFINE_STATIC_SRCU(name) __DEFINE_SRCU(name, static)
203
204 void synchronize_srcu_expedited(struct srcu_struct *ssp);
205 void srcu_barrier(struct srcu_struct *ssp);
206 void srcu_torture_stats_print(struct srcu_struct *ssp, char *tt, char *tf);
207
208 /*
209 * Counts the new reader in the appropriate per-CPU element of the
210 * srcu_struct. Returns an index that must be passed to the matching
211 * srcu_read_unlock_lite().
212 *
213 * Note that this_cpu_inc() is an RCU read-side critical section either
214 * because it disables interrupts, because it is a single instruction,
215 * or because it is a read-modify-write atomic operation, depending on
216 * the whims of the architecture.
217 */
__srcu_read_lock_lite(struct srcu_struct * ssp)218 static inline int __srcu_read_lock_lite(struct srcu_struct *ssp)
219 {
220 int idx;
221
222 RCU_LOCKDEP_WARN(!rcu_is_watching(), "RCU must be watching srcu_read_lock_lite().");
223 idx = READ_ONCE(ssp->srcu_idx) & 0x1;
224 this_cpu_inc(ssp->sda->srcu_lock_count[idx].counter); /* Y */
225 barrier(); /* Avoid leaking the critical section. */
226 return idx;
227 }
228
229 /*
230 * Removes the count for the old reader from the appropriate
231 * per-CPU element of the srcu_struct. Note that this may well be a
232 * different CPU than that which was incremented by the corresponding
233 * srcu_read_lock_lite(), but it must be within the same task.
234 *
235 * Note that this_cpu_inc() is an RCU read-side critical section either
236 * because it disables interrupts, because it is a single instruction,
237 * or because it is a read-modify-write atomic operation, depending on
238 * the whims of the architecture.
239 */
__srcu_read_unlock_lite(struct srcu_struct * ssp,int idx)240 static inline void __srcu_read_unlock_lite(struct srcu_struct *ssp, int idx)
241 {
242 barrier(); /* Avoid leaking the critical section. */
243 this_cpu_inc(ssp->sda->srcu_unlock_count[idx].counter); /* Z */
244 RCU_LOCKDEP_WARN(!rcu_is_watching(), "RCU must be watching srcu_read_unlock_lite().");
245 }
246
247 void __srcu_check_read_flavor(struct srcu_struct *ssp, int read_flavor);
248
249 // Record _lite() usage even for CONFIG_PROVE_RCU=n kernels.
srcu_check_read_flavor_lite(struct srcu_struct * ssp)250 static inline void srcu_check_read_flavor_lite(struct srcu_struct *ssp)
251 {
252 struct srcu_data *sdp = raw_cpu_ptr(ssp->sda);
253
254 if (likely(READ_ONCE(sdp->srcu_reader_flavor) & SRCU_READ_FLAVOR_LITE))
255 return;
256
257 // Note that the cmpxchg() in __srcu_check_read_flavor() is fully ordered.
258 __srcu_check_read_flavor(ssp, SRCU_READ_FLAVOR_LITE);
259 }
260
261 // Record non-_lite() usage only for CONFIG_PROVE_RCU=y kernels.
srcu_check_read_flavor(struct srcu_struct * ssp,int read_flavor)262 static inline void srcu_check_read_flavor(struct srcu_struct *ssp, int read_flavor)
263 {
264 if (IS_ENABLED(CONFIG_PROVE_RCU))
265 __srcu_check_read_flavor(ssp, read_flavor);
266 }
267
268 #endif
269