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 30 /* Update-side state. */ 31 spinlock_t __private lock ____cacheline_internodealigned_in_smp; 32 struct rcu_segcblist srcu_cblist; /* List of callbacks.*/ 33 unsigned long srcu_gp_seq_needed; /* Furthest future GP needed. */ 34 unsigned long srcu_gp_seq_needed_exp; /* Furthest future exp GP. */ 35 bool srcu_cblist_invoking; /* Invoking these CBs? */ 36 struct timer_list delay_work; /* Delay for CB invoking */ 37 struct work_struct work; /* Context for CB invoking. */ 38 struct rcu_head srcu_barrier_head; /* For srcu_barrier() use. */ 39 struct srcu_node *mynode; /* Leaf srcu_node. */ 40 unsigned long grpmask; /* Mask for leaf srcu_node */ 41 /* ->srcu_data_have_cbs[]. */ 42 int cpu; 43 struct srcu_struct *ssp; 44 }; 45 46 /* Values for ->srcu_reader_flavor. */ 47 #define SRCU_READ_FLAVOR_NORMAL 0x1 // srcu_read_lock(). 48 #define SRCU_READ_FLAVOR_NMI 0x2 // srcu_read_lock_nmisafe(). 49 #define SRCU_READ_FLAVOR_LITE 0x4 // srcu_read_lock_lite(). 50 51 /* 52 * Node in SRCU combining tree, similar in function to rcu_data. 53 */ 54 struct srcu_node { 55 spinlock_t __private lock; 56 unsigned long srcu_have_cbs[4]; /* GP seq for children having CBs, but only */ 57 /* if greater than ->srcu_gp_seq. */ 58 unsigned long srcu_data_have_cbs[4]; /* Which srcu_data structs have CBs for given GP? */ 59 unsigned long srcu_gp_seq_needed_exp; /* Furthest future exp GP. */ 60 struct srcu_node *srcu_parent; /* Next up in tree. */ 61 int grplo; /* Least CPU for node. */ 62 int grphi; /* Biggest CPU for node. */ 63 }; 64 65 /* 66 * Per-SRCU-domain structure, update-side data linked from srcu_struct. 67 */ 68 struct srcu_usage { 69 struct srcu_node *node; /* Combining tree. */ 70 struct srcu_node *level[RCU_NUM_LVLS + 1]; 71 /* First node at each level. */ 72 int srcu_size_state; /* Small-to-big transition state. */ 73 struct mutex srcu_cb_mutex; /* Serialize CB preparation. */ 74 spinlock_t __private lock; /* Protect counters and size state. */ 75 struct mutex srcu_gp_mutex; /* Serialize GP work. */ 76 unsigned long srcu_gp_seq; /* Grace-period seq #. */ 77 unsigned long srcu_gp_seq_needed; /* Latest gp_seq needed. */ 78 unsigned long srcu_gp_seq_needed_exp; /* Furthest future exp GP. */ 79 unsigned long srcu_gp_start; /* Last GP start timestamp (jiffies) */ 80 unsigned long srcu_last_gp_end; /* Last GP end timestamp (ns) */ 81 unsigned long srcu_size_jiffies; /* Current contention-measurement interval. */ 82 unsigned long srcu_n_lock_retries; /* Contention events in current interval. */ 83 unsigned long srcu_n_exp_nodelay; /* # expedited no-delays in current GP phase. */ 84 bool sda_is_static; /* May ->sda be passed to free_percpu()? */ 85 unsigned long srcu_barrier_seq; /* srcu_barrier seq #. */ 86 struct mutex srcu_barrier_mutex; /* Serialize barrier ops. */ 87 struct completion srcu_barrier_completion; 88 /* Awaken barrier rq at end. */ 89 atomic_t srcu_barrier_cpu_cnt; /* # CPUs not yet posting a */ 90 /* callback for the barrier */ 91 /* operation. */ 92 unsigned long reschedule_jiffies; 93 unsigned long reschedule_count; 94 struct delayed_work work; 95 struct srcu_struct *srcu_ssp; 96 }; 97 98 /* 99 * Per-SRCU-domain structure, similar in function to rcu_state. 100 */ 101 struct srcu_struct { 102 unsigned int srcu_idx; /* Current rdr array element. */ 103 struct srcu_data __percpu *sda; /* Per-CPU srcu_data array. */ 104 struct lockdep_map dep_map; 105 struct srcu_usage *srcu_sup; /* Update-side data. */ 106 }; 107 108 // Values for size state variable (->srcu_size_state). Once the state 109 // has been set to SRCU_SIZE_ALLOC, the grace-period code advances through 110 // this state machine one step per grace period until the SRCU_SIZE_BIG state 111 // is reached. Otherwise, the state machine remains in the SRCU_SIZE_SMALL 112 // state indefinitely. 113 #define SRCU_SIZE_SMALL 0 // No srcu_node combining tree, ->node == NULL 114 #define SRCU_SIZE_ALLOC 1 // An srcu_node tree is being allocated, initialized, 115 // and then referenced by ->node. It will not be used. 116 #define SRCU_SIZE_WAIT_BARRIER 2 // The srcu_node tree starts being used by everything 117 // except call_srcu(), especially by srcu_barrier(). 118 // By the end of this state, all CPUs and threads 119 // are aware of this tree's existence. 120 #define SRCU_SIZE_WAIT_CALL 3 // The srcu_node tree starts being used by call_srcu(). 121 // By the end of this state, all of the call_srcu() 122 // invocations that were running on a non-boot CPU 123 // and using the boot CPU's callback queue will have 124 // completed. 125 #define SRCU_SIZE_WAIT_CBS1 4 // Don't trust the ->srcu_have_cbs[] grace-period 126 #define SRCU_SIZE_WAIT_CBS2 5 // sequence elements or the ->srcu_data_have_cbs[] 127 #define SRCU_SIZE_WAIT_CBS3 6 // CPU-bitmask elements until all four elements of 128 #define SRCU_SIZE_WAIT_CBS4 7 // each array have been initialized. 129 #define SRCU_SIZE_BIG 8 // The srcu_node combining tree is fully initialized 130 // and all aspects of it are being put to use. 131 132 /* Values for state variable (bottom bits of ->srcu_gp_seq). */ 133 #define SRCU_STATE_IDLE 0 134 #define SRCU_STATE_SCAN1 1 135 #define SRCU_STATE_SCAN2 2 136 137 /* 138 * Values for initializing gp sequence fields. Higher values allow wrap arounds to 139 * occur earlier. 140 * The second value with state is useful in the case of static initialization of 141 * srcu_usage where srcu_gp_seq_needed is expected to have some state value in its 142 * lower bits (or else it will appear to be already initialized within 143 * the call check_init_srcu_struct()). 144 */ 145 #define SRCU_GP_SEQ_INITIAL_VAL ((0UL - 100UL) << RCU_SEQ_CTR_SHIFT) 146 #define SRCU_GP_SEQ_INITIAL_VAL_WITH_STATE (SRCU_GP_SEQ_INITIAL_VAL - 1) 147 148 #define __SRCU_USAGE_INIT(name) \ 149 { \ 150 .lock = __SPIN_LOCK_UNLOCKED(name.lock), \ 151 .srcu_gp_seq = SRCU_GP_SEQ_INITIAL_VAL, \ 152 .srcu_gp_seq_needed = SRCU_GP_SEQ_INITIAL_VAL_WITH_STATE, \ 153 .srcu_gp_seq_needed_exp = SRCU_GP_SEQ_INITIAL_VAL, \ 154 .work = __DELAYED_WORK_INITIALIZER(name.work, NULL, 0), \ 155 } 156 157 #define __SRCU_STRUCT_INIT_COMMON(name, usage_name) \ 158 .srcu_sup = &usage_name, \ 159 __SRCU_DEP_MAP_INIT(name) 160 161 #define __SRCU_STRUCT_INIT_MODULE(name, usage_name) \ 162 { \ 163 __SRCU_STRUCT_INIT_COMMON(name, usage_name) \ 164 } 165 166 #define __SRCU_STRUCT_INIT(name, usage_name, pcpu_name) \ 167 { \ 168 .sda = &pcpu_name, \ 169 __SRCU_STRUCT_INIT_COMMON(name, usage_name) \ 170 } 171 172 /* 173 * Define and initialize a srcu struct at build time. 174 * Do -not- call init_srcu_struct() nor cleanup_srcu_struct() on it. 175 * 176 * Note that although DEFINE_STATIC_SRCU() hides the name from other 177 * files, the per-CPU variable rules nevertheless require that the 178 * chosen name be globally unique. These rules also prohibit use of 179 * DEFINE_STATIC_SRCU() within a function. If these rules are too 180 * restrictive, declare the srcu_struct manually. For example, in 181 * each file: 182 * 183 * static struct srcu_struct my_srcu; 184 * 185 * Then, before the first use of each my_srcu, manually initialize it: 186 * 187 * init_srcu_struct(&my_srcu); 188 * 189 * See include/linux/percpu-defs.h for the rules on per-CPU variables. 190 */ 191 #ifdef MODULE 192 # define __DEFINE_SRCU(name, is_static) \ 193 static struct srcu_usage name##_srcu_usage = __SRCU_USAGE_INIT(name##_srcu_usage); \ 194 is_static struct srcu_struct name = __SRCU_STRUCT_INIT_MODULE(name, name##_srcu_usage); \ 195 extern struct srcu_struct * const __srcu_struct_##name; \ 196 struct srcu_struct * const __srcu_struct_##name \ 197 __section("___srcu_struct_ptrs") = &name 198 #else 199 # define __DEFINE_SRCU(name, is_static) \ 200 static DEFINE_PER_CPU(struct srcu_data, name##_srcu_data); \ 201 static struct srcu_usage name##_srcu_usage = __SRCU_USAGE_INIT(name##_srcu_usage); \ 202 is_static struct srcu_struct name = \ 203 __SRCU_STRUCT_INIT(name, name##_srcu_usage, name##_srcu_data) 204 #endif 205 #define DEFINE_SRCU(name) __DEFINE_SRCU(name, /* not static */) 206 #define DEFINE_STATIC_SRCU(name) __DEFINE_SRCU(name, static) 207 208 void synchronize_srcu_expedited(struct srcu_struct *ssp); 209 void srcu_barrier(struct srcu_struct *ssp); 210 void srcu_torture_stats_print(struct srcu_struct *ssp, char *tt, char *tf); 211 212 /* 213 * Counts the new reader in the appropriate per-CPU element of the 214 * srcu_struct. Returns an index that must be passed to the matching 215 * srcu_read_unlock_lite(). 216 * 217 * Note that this_cpu_inc() is an RCU read-side critical section either 218 * because it disables interrupts, because it is a single instruction, 219 * or because it is a read-modify-write atomic operation, depending on 220 * the whims of the architecture. 221 */ 222 static inline int __srcu_read_lock_lite(struct srcu_struct *ssp) 223 { 224 int idx; 225 226 RCU_LOCKDEP_WARN(!rcu_is_watching(), "RCU must be watching srcu_read_lock_lite()."); 227 idx = READ_ONCE(ssp->srcu_idx) & 0x1; 228 this_cpu_inc(ssp->sda->srcu_lock_count[idx].counter); /* Y */ 229 barrier(); /* Avoid leaking the critical section. */ 230 return idx; 231 } 232 233 /* 234 * Removes the count for the old reader from the appropriate 235 * per-CPU element of the srcu_struct. Note that this may well be a 236 * different CPU than that which was incremented by the corresponding 237 * srcu_read_lock_lite(), but it must be within the same task. 238 * 239 * Note that this_cpu_inc() is an RCU read-side critical section either 240 * because it disables interrupts, because it is a single instruction, 241 * or because it is a read-modify-write atomic operation, depending on 242 * the whims of the architecture. 243 */ 244 static inline void __srcu_read_unlock_lite(struct srcu_struct *ssp, int idx) 245 { 246 barrier(); /* Avoid leaking the critical section. */ 247 this_cpu_inc(ssp->sda->srcu_unlock_count[idx].counter); /* Z */ 248 RCU_LOCKDEP_WARN(!rcu_is_watching(), "RCU must be watching srcu_read_unlock_lite()."); 249 } 250 251 void __srcu_check_read_flavor(struct srcu_struct *ssp, int read_flavor); 252 253 // Record _lite() usage even for CONFIG_PROVE_RCU=n kernels. 254 static inline void srcu_check_read_flavor_lite(struct srcu_struct *ssp) 255 { 256 struct srcu_data *sdp = raw_cpu_ptr(ssp->sda); 257 258 if (likely(READ_ONCE(sdp->srcu_reader_flavor) & SRCU_READ_FLAVOR_LITE)) 259 return; 260 261 // Note that the cmpxchg() in srcu_check_read_flavor() is fully ordered. 262 __srcu_check_read_flavor(ssp, SRCU_READ_FLAVOR_LITE); 263 } 264 265 // Record non-_lite() usage only for CONFIG_PROVE_RCU=y kernels. 266 static inline void srcu_check_read_flavor(struct srcu_struct *ssp, int read_flavor) 267 { 268 if (IS_ENABLED(CONFIG_PROVE_RCU)) 269 __srcu_check_read_flavor(ssp, read_flavor); 270 } 271 272 #endif 273