| 1 | /* SPDX-License-Identifier: GPL-2.0+ */ | 
|---|
| 2 | /* | 
|---|
| 3 | * Read-Copy Update definitions shared among RCU implementations. | 
|---|
| 4 | * | 
|---|
| 5 | * Copyright IBM Corporation, 2011 | 
|---|
| 6 | * | 
|---|
| 7 | * Author: Paul E. McKenney <paulmck@linux.ibm.com> | 
|---|
| 8 | */ | 
|---|
| 9 |  | 
|---|
| 10 | #ifndef __LINUX_RCU_H | 
|---|
| 11 | #define __LINUX_RCU_H | 
|---|
| 12 |  | 
|---|
| 13 | #include <linux/slab.h> | 
|---|
| 14 | #include <trace/events/rcu.h> | 
|---|
| 15 |  | 
|---|
| 16 | /* | 
|---|
| 17 | * Grace-period counter management. | 
|---|
| 18 | * | 
|---|
| 19 | * The two least significant bits contain the control flags. | 
|---|
| 20 | * The most significant bits contain the grace-period sequence counter. | 
|---|
| 21 | * | 
|---|
| 22 | * When both control flags are zero, no grace period is in progress. | 
|---|
| 23 | * When either bit is non-zero, a grace period has started and is in | 
|---|
| 24 | * progress. When the grace period completes, the control flags are reset | 
|---|
| 25 | * to 0 and the grace-period sequence counter is incremented. | 
|---|
| 26 | * | 
|---|
| 27 | * However some specific RCU usages make use of custom values. | 
|---|
| 28 | * | 
|---|
| 29 | * SRCU special control values: | 
|---|
| 30 | * | 
|---|
| 31 | *	SRCU_SNP_INIT_SEQ	:	Invalid/init value set when SRCU node | 
|---|
| 32 | *					is initialized. | 
|---|
| 33 | * | 
|---|
| 34 | *	SRCU_STATE_IDLE		:	No SRCU gp is in progress | 
|---|
| 35 | * | 
|---|
| 36 | *	SRCU_STATE_SCAN1	:	State set by rcu_seq_start(). Indicates | 
|---|
| 37 | *					we are scanning the readers on the slot | 
|---|
| 38 | *					defined as inactive (there might well | 
|---|
| 39 | *					be pending readers that will use that | 
|---|
| 40 | *					index, but their number is bounded). | 
|---|
| 41 | * | 
|---|
| 42 | *	SRCU_STATE_SCAN2	:	State set manually via rcu_seq_set_state() | 
|---|
| 43 | *					Indicates we are flipping the readers | 
|---|
| 44 | *					index and then scanning the readers on the | 
|---|
| 45 | *					slot newly designated as inactive (again, | 
|---|
| 46 | *					the number of pending readers that will use | 
|---|
| 47 | *					this inactive index is bounded). | 
|---|
| 48 | * | 
|---|
| 49 | * RCU polled GP special control value: | 
|---|
| 50 | * | 
|---|
| 51 | *	RCU_GET_STATE_COMPLETED :	State value indicating an already-completed | 
|---|
| 52 | *					polled GP has completed.  This value covers | 
|---|
| 53 | *					both the state and the counter of the | 
|---|
| 54 | *					grace-period sequence number. | 
|---|
| 55 | */ | 
|---|
| 56 |  | 
|---|
| 57 | /* Low-order bit definition for polled grace-period APIs. */ | 
|---|
| 58 | #define RCU_GET_STATE_COMPLETED	0x1 | 
|---|
| 59 |  | 
|---|
| 60 | /* A complete grace period count */ | 
|---|
| 61 | #define RCU_SEQ_GP (RCU_SEQ_STATE_MASK + 1) | 
|---|
| 62 |  | 
|---|
| 63 | extern int sysctl_sched_rt_runtime; | 
|---|
| 64 |  | 
|---|
| 65 | /* | 
|---|
| 66 | * Return the counter portion of a sequence number previously returned | 
|---|
| 67 | * by rcu_seq_snap() or rcu_seq_current(). | 
|---|
| 68 | */ | 
|---|
| 69 | static inline unsigned long rcu_seq_ctr(unsigned long s) | 
|---|
| 70 | { | 
|---|
| 71 | return s >> RCU_SEQ_CTR_SHIFT; | 
|---|
| 72 | } | 
|---|
| 73 |  | 
|---|
| 74 | /* | 
|---|
| 75 | * Return the state portion of a sequence number previously returned | 
|---|
| 76 | * by rcu_seq_snap() or rcu_seq_current(). | 
|---|
| 77 | */ | 
|---|
| 78 | static inline int rcu_seq_state(unsigned long s) | 
|---|
| 79 | { | 
|---|
| 80 | return s & RCU_SEQ_STATE_MASK; | 
|---|
| 81 | } | 
|---|
| 82 |  | 
|---|
| 83 | /* | 
|---|
| 84 | * Set the state portion of the pointed-to sequence number. | 
|---|
| 85 | * The caller is responsible for preventing conflicting updates. | 
|---|
| 86 | */ | 
|---|
| 87 | static inline void rcu_seq_set_state(unsigned long *sp, int newstate) | 
|---|
| 88 | { | 
|---|
| 89 | WARN_ON_ONCE(newstate & ~RCU_SEQ_STATE_MASK); | 
|---|
| 90 | WRITE_ONCE(*sp, (*sp & ~RCU_SEQ_STATE_MASK) + newstate); | 
|---|
| 91 | } | 
|---|
| 92 |  | 
|---|
| 93 | /* Adjust sequence number for start of update-side operation. */ | 
|---|
| 94 | static inline void rcu_seq_start(unsigned long *sp) | 
|---|
| 95 | { | 
|---|
| 96 | WRITE_ONCE(*sp, *sp + 1); | 
|---|
| 97 | smp_mb(); /* Ensure update-side operation after counter increment. */ | 
|---|
| 98 | WARN_ON_ONCE(rcu_seq_state(*sp) != 1); | 
|---|
| 99 | } | 
|---|
| 100 |  | 
|---|
| 101 | /* Compute the end-of-grace-period value for the specified sequence number. */ | 
|---|
| 102 | static inline unsigned long rcu_seq_endval(unsigned long *sp) | 
|---|
| 103 | { | 
|---|
| 104 | return (*sp | RCU_SEQ_STATE_MASK) + 1; | 
|---|
| 105 | } | 
|---|
| 106 |  | 
|---|
| 107 | /* Adjust sequence number for end of update-side operation. */ | 
|---|
| 108 | static inline void rcu_seq_end(unsigned long *sp) | 
|---|
| 109 | { | 
|---|
| 110 | smp_mb(); /* Ensure update-side operation before counter increment. */ | 
|---|
| 111 | WARN_ON_ONCE(!rcu_seq_state(*sp)); | 
|---|
| 112 | WRITE_ONCE(*sp, rcu_seq_endval(sp)); | 
|---|
| 113 | } | 
|---|
| 114 |  | 
|---|
| 115 | /* | 
|---|
| 116 | * rcu_seq_snap - Take a snapshot of the update side's sequence number. | 
|---|
| 117 | * | 
|---|
| 118 | * This function returns the earliest value of the grace-period sequence number | 
|---|
| 119 | * that will indicate that a full grace period has elapsed since the current | 
|---|
| 120 | * time.  Once the grace-period sequence number has reached this value, it will | 
|---|
| 121 | * be safe to invoke all callbacks that have been registered prior to the | 
|---|
| 122 | * current time. This value is the current grace-period number plus two to the | 
|---|
| 123 | * power of the number of low-order bits reserved for state, then rounded up to | 
|---|
| 124 | * the next value in which the state bits are all zero. | 
|---|
| 125 | */ | 
|---|
| 126 | static inline unsigned long rcu_seq_snap(unsigned long *sp) | 
|---|
| 127 | { | 
|---|
| 128 | unsigned long s; | 
|---|
| 129 |  | 
|---|
| 130 | s = (READ_ONCE(*sp) + 2 * RCU_SEQ_STATE_MASK + 1) & ~RCU_SEQ_STATE_MASK; | 
|---|
| 131 | smp_mb(); /* Above access must not bleed into critical section. */ | 
|---|
| 132 | return s; | 
|---|
| 133 | } | 
|---|
| 134 |  | 
|---|
| 135 | /* Return the current value the update side's sequence number, no ordering. */ | 
|---|
| 136 | static inline unsigned long rcu_seq_current(unsigned long *sp) | 
|---|
| 137 | { | 
|---|
| 138 | return READ_ONCE(*sp); | 
|---|
| 139 | } | 
|---|
| 140 |  | 
|---|
| 141 | /* | 
|---|
| 142 | * Given a snapshot from rcu_seq_snap(), determine whether or not the | 
|---|
| 143 | * corresponding update-side operation has started. | 
|---|
| 144 | */ | 
|---|
| 145 | static inline bool rcu_seq_started(unsigned long *sp, unsigned long s) | 
|---|
| 146 | { | 
|---|
| 147 | return ULONG_CMP_LT((s - 1) & ~RCU_SEQ_STATE_MASK, READ_ONCE(*sp)); | 
|---|
| 148 | } | 
|---|
| 149 |  | 
|---|
| 150 | /* | 
|---|
| 151 | * Given a snapshot from rcu_seq_snap(), determine whether or not a | 
|---|
| 152 | * full update-side operation has occurred. | 
|---|
| 153 | */ | 
|---|
| 154 | static inline bool rcu_seq_done(unsigned long *sp, unsigned long s) | 
|---|
| 155 | { | 
|---|
| 156 | return ULONG_CMP_GE(READ_ONCE(*sp), s); | 
|---|
| 157 | } | 
|---|
| 158 |  | 
|---|
| 159 | /* | 
|---|
| 160 | * Given a snapshot from rcu_seq_snap(), determine whether or not a | 
|---|
| 161 | * full update-side operation has occurred, but do not allow the | 
|---|
| 162 | * (ULONG_MAX / 2) safety-factor/guard-band. | 
|---|
| 163 | * | 
|---|
| 164 | * The token returned by get_state_synchronize_rcu_full() is based on | 
|---|
| 165 | * rcu_state.gp_seq but it is tested in poll_state_synchronize_rcu_full() | 
|---|
| 166 | * against the root rnp->gp_seq. Since rcu_seq_start() is first called | 
|---|
| 167 | * on rcu_state.gp_seq and only later reflected on the root rnp->gp_seq, | 
|---|
| 168 | * it is possible that rcu_seq_snap(rcu_state.gp_seq) returns 2 full grace | 
|---|
| 169 | * periods ahead of the root rnp->gp_seq. To prevent false-positives with the | 
|---|
| 170 | * full polling API that a wrap around instantly completed the GP, when nothing | 
|---|
| 171 | * like that happened, adjust for the 2 GPs in the ULONG_CMP_LT(). | 
|---|
| 172 | */ | 
|---|
| 173 | static inline bool rcu_seq_done_exact(unsigned long *sp, unsigned long s) | 
|---|
| 174 | { | 
|---|
| 175 | unsigned long cur_s = READ_ONCE(*sp); | 
|---|
| 176 |  | 
|---|
| 177 | return ULONG_CMP_GE(cur_s, s) || ULONG_CMP_LT(cur_s, s - (2 * RCU_SEQ_GP)); | 
|---|
| 178 | } | 
|---|
| 179 |  | 
|---|
| 180 | /* | 
|---|
| 181 | * Has a grace period completed since the time the old gp_seq was collected? | 
|---|
| 182 | */ | 
|---|
| 183 | static inline bool rcu_seq_completed_gp(unsigned long old, unsigned long new) | 
|---|
| 184 | { | 
|---|
| 185 | return ULONG_CMP_LT(old, new & ~RCU_SEQ_STATE_MASK); | 
|---|
| 186 | } | 
|---|
| 187 |  | 
|---|
| 188 | /* | 
|---|
| 189 | * Has a grace period started since the time the old gp_seq was collected? | 
|---|
| 190 | */ | 
|---|
| 191 | static inline bool rcu_seq_new_gp(unsigned long old, unsigned long new) | 
|---|
| 192 | { | 
|---|
| 193 | return ULONG_CMP_LT((old + RCU_SEQ_STATE_MASK) & ~RCU_SEQ_STATE_MASK, | 
|---|
| 194 | new); | 
|---|
| 195 | } | 
|---|
| 196 |  | 
|---|
| 197 | /* | 
|---|
| 198 | * Roughly how many full grace periods have elapsed between the collection | 
|---|
| 199 | * of the two specified grace periods? | 
|---|
| 200 | */ | 
|---|
| 201 | static inline unsigned long rcu_seq_diff(unsigned long new, unsigned long old) | 
|---|
| 202 | { | 
|---|
| 203 | unsigned long rnd_diff; | 
|---|
| 204 |  | 
|---|
| 205 | if (old == new) | 
|---|
| 206 | return 0; | 
|---|
| 207 | /* | 
|---|
| 208 | * Compute the number of grace periods (still shifted up), plus | 
|---|
| 209 | * one if either of new and old is not an exact grace period. | 
|---|
| 210 | */ | 
|---|
| 211 | rnd_diff = (new & ~RCU_SEQ_STATE_MASK) - | 
|---|
| 212 | ((old + RCU_SEQ_STATE_MASK) & ~RCU_SEQ_STATE_MASK) + | 
|---|
| 213 | ((new & RCU_SEQ_STATE_MASK) || (old & RCU_SEQ_STATE_MASK)); | 
|---|
| 214 | if (ULONG_CMP_GE(RCU_SEQ_STATE_MASK, rnd_diff)) | 
|---|
| 215 | return 1; /* Definitely no grace period has elapsed. */ | 
|---|
| 216 | return ((rnd_diff - RCU_SEQ_STATE_MASK - 1) >> RCU_SEQ_CTR_SHIFT) + 2; | 
|---|
| 217 | } | 
|---|
| 218 |  | 
|---|
| 219 | /* | 
|---|
| 220 | * debug_rcu_head_queue()/debug_rcu_head_unqueue() are used internally | 
|---|
| 221 | * by call_rcu() and rcu callback execution, and are therefore not part | 
|---|
| 222 | * of the RCU API. These are in rcupdate.h because they are used by all | 
|---|
| 223 | * RCU implementations. | 
|---|
| 224 | */ | 
|---|
| 225 |  | 
|---|
| 226 | #ifdef CONFIG_DEBUG_OBJECTS_RCU_HEAD | 
|---|
| 227 | # define STATE_RCU_HEAD_READY	0 | 
|---|
| 228 | # define STATE_RCU_HEAD_QUEUED	1 | 
|---|
| 229 |  | 
|---|
| 230 | extern const struct debug_obj_descr rcuhead_debug_descr; | 
|---|
| 231 |  | 
|---|
| 232 | static inline int debug_rcu_head_queue(struct rcu_head *head) | 
|---|
| 233 | { | 
|---|
| 234 | int r1; | 
|---|
| 235 |  | 
|---|
| 236 | r1 = debug_object_activate(head, &rcuhead_debug_descr); | 
|---|
| 237 | debug_object_active_state(head, &rcuhead_debug_descr, | 
|---|
| 238 | STATE_RCU_HEAD_READY, | 
|---|
| 239 | STATE_RCU_HEAD_QUEUED); | 
|---|
| 240 | return r1; | 
|---|
| 241 | } | 
|---|
| 242 |  | 
|---|
| 243 | static inline void debug_rcu_head_unqueue(struct rcu_head *head) | 
|---|
| 244 | { | 
|---|
| 245 | debug_object_active_state(head, &rcuhead_debug_descr, | 
|---|
| 246 | STATE_RCU_HEAD_QUEUED, | 
|---|
| 247 | STATE_RCU_HEAD_READY); | 
|---|
| 248 | debug_object_deactivate(head, &rcuhead_debug_descr); | 
|---|
| 249 | } | 
|---|
| 250 | #else	/* !CONFIG_DEBUG_OBJECTS_RCU_HEAD */ | 
|---|
| 251 | static inline int debug_rcu_head_queue(struct rcu_head *head) | 
|---|
| 252 | { | 
|---|
| 253 | return 0; | 
|---|
| 254 | } | 
|---|
| 255 |  | 
|---|
| 256 | static inline void debug_rcu_head_unqueue(struct rcu_head *head) | 
|---|
| 257 | { | 
|---|
| 258 | } | 
|---|
| 259 | #endif	/* #else !CONFIG_DEBUG_OBJECTS_RCU_HEAD */ | 
|---|
| 260 |  | 
|---|
| 261 | static inline void debug_rcu_head_callback(struct rcu_head *rhp) | 
|---|
| 262 | { | 
|---|
| 263 | if (unlikely(!rhp->func)) | 
|---|
| 264 | kmem_dump_obj(object: rhp); | 
|---|
| 265 | } | 
|---|
| 266 |  | 
|---|
| 267 | static inline bool rcu_barrier_cb_is_done(struct rcu_head *rhp) | 
|---|
| 268 | { | 
|---|
| 269 | return rhp->next == rhp; | 
|---|
| 270 | } | 
|---|
| 271 |  | 
|---|
| 272 | extern int rcu_cpu_stall_suppress_at_boot; | 
|---|
| 273 |  | 
|---|
| 274 | static inline bool rcu_stall_is_suppressed_at_boot(void) | 
|---|
| 275 | { | 
|---|
| 276 | return rcu_cpu_stall_suppress_at_boot && !rcu_inkernel_boot_has_ended(); | 
|---|
| 277 | } | 
|---|
| 278 |  | 
|---|
| 279 | extern int rcu_cpu_stall_notifiers; | 
|---|
| 280 |  | 
|---|
| 281 | #ifdef CONFIG_RCU_STALL_COMMON | 
|---|
| 282 |  | 
|---|
| 283 | extern int rcu_cpu_stall_ftrace_dump; | 
|---|
| 284 | extern int rcu_cpu_stall_suppress; | 
|---|
| 285 | extern int rcu_cpu_stall_timeout; | 
|---|
| 286 | extern int rcu_exp_cpu_stall_timeout; | 
|---|
| 287 | extern int rcu_cpu_stall_cputime; | 
|---|
| 288 | extern bool rcu_exp_stall_task_details __read_mostly; | 
|---|
| 289 | int rcu_jiffies_till_stall_check(void); | 
|---|
| 290 | int rcu_exp_jiffies_till_stall_check(void); | 
|---|
| 291 |  | 
|---|
| 292 | static inline bool rcu_stall_is_suppressed(void) | 
|---|
| 293 | { | 
|---|
| 294 | return rcu_stall_is_suppressed_at_boot() || rcu_cpu_stall_suppress; | 
|---|
| 295 | } | 
|---|
| 296 |  | 
|---|
| 297 | #define rcu_ftrace_dump_stall_suppress() \ | 
|---|
| 298 | do { \ | 
|---|
| 299 | if (!rcu_cpu_stall_suppress) \ | 
|---|
| 300 | rcu_cpu_stall_suppress = 3; \ | 
|---|
| 301 | } while (0) | 
|---|
| 302 |  | 
|---|
| 303 | #define rcu_ftrace_dump_stall_unsuppress() \ | 
|---|
| 304 | do { \ | 
|---|
| 305 | if (rcu_cpu_stall_suppress == 3) \ | 
|---|
| 306 | rcu_cpu_stall_suppress = 0; \ | 
|---|
| 307 | } while (0) | 
|---|
| 308 |  | 
|---|
| 309 | #else /* #endif #ifdef CONFIG_RCU_STALL_COMMON */ | 
|---|
| 310 |  | 
|---|
| 311 | static inline bool rcu_stall_is_suppressed(void) | 
|---|
| 312 | { | 
|---|
| 313 | return rcu_stall_is_suppressed_at_boot(); | 
|---|
| 314 | } | 
|---|
| 315 | #define rcu_ftrace_dump_stall_suppress() | 
|---|
| 316 | #define rcu_ftrace_dump_stall_unsuppress() | 
|---|
| 317 | #endif /* #ifdef CONFIG_RCU_STALL_COMMON */ | 
|---|
| 318 |  | 
|---|
| 319 | /* | 
|---|
| 320 | * Strings used in tracepoints need to be exported via the | 
|---|
| 321 | * tracing system such that tools like perf and trace-cmd can | 
|---|
| 322 | * translate the string address pointers to actual text. | 
|---|
| 323 | */ | 
|---|
| 324 | #define TPS(x)  tracepoint_string(x) | 
|---|
| 325 |  | 
|---|
| 326 | /* | 
|---|
| 327 | * Dump the ftrace buffer, but only one time per callsite per boot. | 
|---|
| 328 | */ | 
|---|
| 329 | #define rcu_ftrace_dump(oops_dump_mode) \ | 
|---|
| 330 | do { \ | 
|---|
| 331 | static atomic_t ___rfd_beenhere = ATOMIC_INIT(0); \ | 
|---|
| 332 | \ | 
|---|
| 333 | if (!atomic_read(&___rfd_beenhere) && \ | 
|---|
| 334 | !atomic_xchg(&___rfd_beenhere, 1)) { \ | 
|---|
| 335 | tracing_off(); \ | 
|---|
| 336 | rcu_ftrace_dump_stall_suppress(); \ | 
|---|
| 337 | ftrace_dump(oops_dump_mode); \ | 
|---|
| 338 | rcu_ftrace_dump_stall_unsuppress(); \ | 
|---|
| 339 | } \ | 
|---|
| 340 | } while (0) | 
|---|
| 341 |  | 
|---|
| 342 | void rcu_early_boot_tests(void); | 
|---|
| 343 | void rcu_test_sync_prims(void); | 
|---|
| 344 |  | 
|---|
| 345 | /* | 
|---|
| 346 | * This function really isn't for public consumption, but RCU is special in | 
|---|
| 347 | * that context switches can allow the state machine to make progress. | 
|---|
| 348 | */ | 
|---|
| 349 | extern void resched_cpu(int cpu); | 
|---|
| 350 |  | 
|---|
| 351 | #if !defined(CONFIG_TINY_RCU) | 
|---|
| 352 |  | 
|---|
| 353 | #include <linux/rcu_node_tree.h> | 
|---|
| 354 |  | 
|---|
| 355 | extern int rcu_num_lvls; | 
|---|
| 356 | extern int num_rcu_lvl[]; | 
|---|
| 357 | extern int rcu_num_nodes; | 
|---|
| 358 | static bool rcu_fanout_exact; | 
|---|
| 359 | static int rcu_fanout_leaf; | 
|---|
| 360 |  | 
|---|
| 361 | /* | 
|---|
| 362 | * Compute the per-level fanout, either using the exact fanout specified | 
|---|
| 363 | * or balancing the tree, depending on the rcu_fanout_exact boot parameter. | 
|---|
| 364 | */ | 
|---|
| 365 | static inline void rcu_init_levelspread(int *levelspread, const int *levelcnt) | 
|---|
| 366 | { | 
|---|
| 367 | int i; | 
|---|
| 368 |  | 
|---|
| 369 | for (i = 0; i < RCU_NUM_LVLS; i++) | 
|---|
| 370 | levelspread[i] = INT_MIN; | 
|---|
| 371 | if (rcu_fanout_exact) { | 
|---|
| 372 | levelspread[rcu_num_lvls - 1] = rcu_fanout_leaf; | 
|---|
| 373 | for (i = rcu_num_lvls - 2; i >= 0; i--) | 
|---|
| 374 | levelspread[i] = RCU_FANOUT; | 
|---|
| 375 | } else { | 
|---|
| 376 | int ccur; | 
|---|
| 377 | int cprv; | 
|---|
| 378 |  | 
|---|
| 379 | cprv = nr_cpu_ids; | 
|---|
| 380 | for (i = rcu_num_lvls - 1; i >= 0; i--) { | 
|---|
| 381 | ccur = levelcnt[i]; | 
|---|
| 382 | levelspread[i] = (cprv + ccur - 1) / ccur; | 
|---|
| 383 | cprv = ccur; | 
|---|
| 384 | } | 
|---|
| 385 | } | 
|---|
| 386 | } | 
|---|
| 387 |  | 
|---|
| 388 | extern void rcu_init_geometry(void); | 
|---|
| 389 |  | 
|---|
| 390 | /* Returns a pointer to the first leaf rcu_node structure. */ | 
|---|
| 391 | #define rcu_first_leaf_node() (rcu_state.level[rcu_num_lvls - 1]) | 
|---|
| 392 |  | 
|---|
| 393 | /* Is this rcu_node a leaf? */ | 
|---|
| 394 | #define rcu_is_leaf_node(rnp) ((rnp)->level == rcu_num_lvls - 1) | 
|---|
| 395 |  | 
|---|
| 396 | /* Is this rcu_node the last leaf? */ | 
|---|
| 397 | #define rcu_is_last_leaf_node(rnp) ((rnp) == &rcu_state.node[rcu_num_nodes - 1]) | 
|---|
| 398 |  | 
|---|
| 399 | /* | 
|---|
| 400 | * Do a full breadth-first scan of the {s,}rcu_node structures for the | 
|---|
| 401 | * specified state structure (for SRCU) or the only rcu_state structure | 
|---|
| 402 | * (for RCU). | 
|---|
| 403 | */ | 
|---|
| 404 | #define _rcu_for_each_node_breadth_first(sp, rnp) \ | 
|---|
| 405 | for ((rnp) = &(sp)->node[0]; \ | 
|---|
| 406 | (rnp) < &(sp)->node[rcu_num_nodes]; (rnp)++) | 
|---|
| 407 | #define rcu_for_each_node_breadth_first(rnp) \ | 
|---|
| 408 | _rcu_for_each_node_breadth_first(&rcu_state, rnp) | 
|---|
| 409 | #define srcu_for_each_node_breadth_first(ssp, rnp) \ | 
|---|
| 410 | _rcu_for_each_node_breadth_first(ssp->srcu_sup, rnp) | 
|---|
| 411 |  | 
|---|
| 412 | /* | 
|---|
| 413 | * Scan the leaves of the rcu_node hierarchy for the rcu_state structure. | 
|---|
| 414 | * Note that if there is a singleton rcu_node tree with but one rcu_node | 
|---|
| 415 | * structure, this loop -will- visit the rcu_node structure.  It is still | 
|---|
| 416 | * a leaf node, even if it is also the root node. | 
|---|
| 417 | */ | 
|---|
| 418 | #define rcu_for_each_leaf_node(rnp) \ | 
|---|
| 419 | for ((rnp) = rcu_first_leaf_node(); \ | 
|---|
| 420 | (rnp) < &rcu_state.node[rcu_num_nodes]; (rnp)++) | 
|---|
| 421 |  | 
|---|
| 422 | /* | 
|---|
| 423 | * Iterate over all possible CPUs in a leaf RCU node. | 
|---|
| 424 | */ | 
|---|
| 425 | #define for_each_leaf_node_possible_cpu(rnp, cpu) \ | 
|---|
| 426 | for (WARN_ON_ONCE(!rcu_is_leaf_node(rnp)), \ | 
|---|
| 427 | (cpu) = cpumask_next((rnp)->grplo - 1, cpu_possible_mask); \ | 
|---|
| 428 | (cpu) <= rnp->grphi; \ | 
|---|
| 429 | (cpu) = cpumask_next((cpu), cpu_possible_mask)) | 
|---|
| 430 |  | 
|---|
| 431 | /* | 
|---|
| 432 | * Iterate over all CPUs in a leaf RCU node's specified mask. | 
|---|
| 433 | */ | 
|---|
| 434 | #define rcu_find_next_bit(rnp, cpu, mask) \ | 
|---|
| 435 | ((rnp)->grplo + find_next_bit(&(mask), BITS_PER_LONG, (cpu))) | 
|---|
| 436 | #define for_each_leaf_node_cpu_mask(rnp, cpu, mask) \ | 
|---|
| 437 | for (WARN_ON_ONCE(!rcu_is_leaf_node(rnp)), \ | 
|---|
| 438 | (cpu) = rcu_find_next_bit((rnp), 0, (mask)); \ | 
|---|
| 439 | (cpu) <= rnp->grphi; \ | 
|---|
| 440 | (cpu) = rcu_find_next_bit((rnp), (cpu) + 1 - (rnp->grplo), (mask))) | 
|---|
| 441 |  | 
|---|
| 442 | #endif /* !defined(CONFIG_TINY_RCU) */ | 
|---|
| 443 |  | 
|---|
| 444 | #if !defined(CONFIG_TINY_RCU) || defined(CONFIG_TASKS_RCU_GENERIC) | 
|---|
| 445 |  | 
|---|
| 446 | /* | 
|---|
| 447 | * Wrappers for the rcu_node::lock acquire and release. | 
|---|
| 448 | * | 
|---|
| 449 | * Because the rcu_nodes form a tree, the tree traversal locking will observe | 
|---|
| 450 | * different lock values, this in turn means that an UNLOCK of one level | 
|---|
| 451 | * followed by a LOCK of another level does not imply a full memory barrier; | 
|---|
| 452 | * and most importantly transitivity is lost. | 
|---|
| 453 | * | 
|---|
| 454 | * In order to restore full ordering between tree levels, augment the regular | 
|---|
| 455 | * lock acquire functions with smp_mb__after_unlock_lock(). | 
|---|
| 456 | * | 
|---|
| 457 | * As ->lock of struct rcu_node is a __private field, therefore one should use | 
|---|
| 458 | * these wrappers rather than directly call raw_spin_{lock,unlock}* on ->lock. | 
|---|
| 459 | */ | 
|---|
| 460 | #define raw_spin_lock_rcu_node(p)					\ | 
|---|
| 461 | do {									\ | 
|---|
| 462 | raw_spin_lock(&ACCESS_PRIVATE(p, lock));			\ | 
|---|
| 463 | smp_mb__after_unlock_lock();					\ | 
|---|
| 464 | } while (0) | 
|---|
| 465 |  | 
|---|
| 466 | #define raw_spin_unlock_rcu_node(p)					\ | 
|---|
| 467 | do {									\ | 
|---|
| 468 | lockdep_assert_irqs_disabled();					\ | 
|---|
| 469 | raw_spin_unlock(&ACCESS_PRIVATE(p, lock));			\ | 
|---|
| 470 | } while (0) | 
|---|
| 471 |  | 
|---|
| 472 | #define raw_spin_lock_irq_rcu_node(p)					\ | 
|---|
| 473 | do {									\ | 
|---|
| 474 | raw_spin_lock_irq(&ACCESS_PRIVATE(p, lock));			\ | 
|---|
| 475 | smp_mb__after_unlock_lock();					\ | 
|---|
| 476 | } while (0) | 
|---|
| 477 |  | 
|---|
| 478 | #define raw_spin_unlock_irq_rcu_node(p)					\ | 
|---|
| 479 | do {									\ | 
|---|
| 480 | lockdep_assert_irqs_disabled();					\ | 
|---|
| 481 | raw_spin_unlock_irq(&ACCESS_PRIVATE(p, lock));			\ | 
|---|
| 482 | } while (0) | 
|---|
| 483 |  | 
|---|
| 484 | #define raw_spin_lock_irqsave_rcu_node(p, flags)			\ | 
|---|
| 485 | do {									\ | 
|---|
| 486 | raw_spin_lock_irqsave(&ACCESS_PRIVATE(p, lock), flags);	\ | 
|---|
| 487 | smp_mb__after_unlock_lock();					\ | 
|---|
| 488 | } while (0) | 
|---|
| 489 |  | 
|---|
| 490 | #define raw_spin_unlock_irqrestore_rcu_node(p, flags)			\ | 
|---|
| 491 | do {									\ | 
|---|
| 492 | lockdep_assert_irqs_disabled();					\ | 
|---|
| 493 | raw_spin_unlock_irqrestore(&ACCESS_PRIVATE(p, lock), flags);	\ | 
|---|
| 494 | } while (0) | 
|---|
| 495 |  | 
|---|
| 496 | #define raw_spin_trylock_rcu_node(p)					\ | 
|---|
| 497 | ({									\ | 
|---|
| 498 | bool ___locked = raw_spin_trylock(&ACCESS_PRIVATE(p, lock));	\ | 
|---|
| 499 | \ | 
|---|
| 500 | if (___locked)							\ | 
|---|
| 501 | smp_mb__after_unlock_lock();				\ | 
|---|
| 502 | ___locked;							\ | 
|---|
| 503 | }) | 
|---|
| 504 |  | 
|---|
| 505 | #define raw_lockdep_assert_held_rcu_node(p)				\ | 
|---|
| 506 | lockdep_assert_held(&ACCESS_PRIVATE(p, lock)) | 
|---|
| 507 |  | 
|---|
| 508 | #endif // #if !defined(CONFIG_TINY_RCU) || defined(CONFIG_TASKS_RCU_GENERIC) | 
|---|
| 509 |  | 
|---|
| 510 | #ifdef CONFIG_TINY_RCU | 
|---|
| 511 | /* Tiny RCU doesn't expedite, as its purpose in life is instead to be tiny. */ | 
|---|
| 512 | static inline bool rcu_gp_is_normal(void) { return true; } | 
|---|
| 513 | static inline bool rcu_gp_is_expedited(void) { return false; } | 
|---|
| 514 | static inline bool rcu_async_should_hurry(void) { return false; } | 
|---|
| 515 | static inline void rcu_expedite_gp(void) { } | 
|---|
| 516 | static inline void rcu_unexpedite_gp(void) { } | 
|---|
| 517 | static inline void rcu_async_hurry(void) { } | 
|---|
| 518 | static inline void rcu_async_relax(void) { } | 
|---|
| 519 | static inline bool rcu_cpu_online(int cpu) { return true; } | 
|---|
| 520 | #else /* #ifdef CONFIG_TINY_RCU */ | 
|---|
| 521 | bool rcu_gp_is_normal(void);     /* Internal RCU use. */ | 
|---|
| 522 | bool rcu_gp_is_expedited(void);  /* Internal RCU use. */ | 
|---|
| 523 | bool rcu_async_should_hurry(void);  /* Internal RCU use. */ | 
|---|
| 524 | void rcu_expedite_gp(void); | 
|---|
| 525 | void rcu_unexpedite_gp(void); | 
|---|
| 526 | void rcu_async_hurry(void); | 
|---|
| 527 | void rcu_async_relax(void); | 
|---|
| 528 | void rcupdate_announce_bootup_oddness(void); | 
|---|
| 529 | bool rcu_cpu_online(int cpu); | 
|---|
| 530 | #ifdef CONFIG_TASKS_RCU_GENERIC | 
|---|
| 531 | void show_rcu_tasks_gp_kthreads(void); | 
|---|
| 532 | #else /* #ifdef CONFIG_TASKS_RCU_GENERIC */ | 
|---|
| 533 | static inline void show_rcu_tasks_gp_kthreads(void) {} | 
|---|
| 534 | #endif /* #else #ifdef CONFIG_TASKS_RCU_GENERIC */ | 
|---|
| 535 | #endif /* #else #ifdef CONFIG_TINY_RCU */ | 
|---|
| 536 |  | 
|---|
| 537 | #ifdef CONFIG_TASKS_RCU | 
|---|
| 538 | struct task_struct *get_rcu_tasks_gp_kthread(void); | 
|---|
| 539 | void rcu_tasks_get_gp_data(int *flags, unsigned long *gp_seq); | 
|---|
| 540 | #endif // # ifdef CONFIG_TASKS_RCU | 
|---|
| 541 |  | 
|---|
| 542 | #ifdef CONFIG_TASKS_RUDE_RCU | 
|---|
| 543 | struct task_struct *get_rcu_tasks_rude_gp_kthread(void); | 
|---|
| 544 | void rcu_tasks_rude_get_gp_data(int *flags, unsigned long *gp_seq); | 
|---|
| 545 | #endif // # ifdef CONFIG_TASKS_RUDE_RCU | 
|---|
| 546 |  | 
|---|
| 547 | #ifdef CONFIG_TASKS_TRACE_RCU | 
|---|
| 548 | void rcu_tasks_trace_get_gp_data(int *flags, unsigned long *gp_seq); | 
|---|
| 549 | #endif | 
|---|
| 550 |  | 
|---|
| 551 | #ifdef CONFIG_TASKS_RCU_GENERIC | 
|---|
| 552 | void tasks_cblist_init_generic(void); | 
|---|
| 553 | #else /* #ifdef CONFIG_TASKS_RCU_GENERIC */ | 
|---|
| 554 | static inline void tasks_cblist_init_generic(void) { } | 
|---|
| 555 | #endif /* #else #ifdef CONFIG_TASKS_RCU_GENERIC */ | 
|---|
| 556 |  | 
|---|
| 557 | #define RCU_SCHEDULER_INACTIVE	0 | 
|---|
| 558 | #define RCU_SCHEDULER_INIT	1 | 
|---|
| 559 | #define RCU_SCHEDULER_RUNNING	2 | 
|---|
| 560 |  | 
|---|
| 561 | enum rcutorture_type { | 
|---|
| 562 | RCU_FLAVOR, | 
|---|
| 563 | RCU_TASKS_FLAVOR, | 
|---|
| 564 | RCU_TASKS_RUDE_FLAVOR, | 
|---|
| 565 | RCU_TASKS_TRACING_FLAVOR, | 
|---|
| 566 | RCU_TRIVIAL_FLAVOR, | 
|---|
| 567 | SRCU_FLAVOR, | 
|---|
| 568 | INVALID_RCU_FLAVOR | 
|---|
| 569 | }; | 
|---|
| 570 |  | 
|---|
| 571 | #if defined(CONFIG_RCU_LAZY) | 
|---|
| 572 | unsigned long rcu_get_jiffies_lazy_flush(void); | 
|---|
| 573 | void rcu_set_jiffies_lazy_flush(unsigned long j); | 
|---|
| 574 | #else | 
|---|
| 575 | static inline unsigned long rcu_get_jiffies_lazy_flush(void) { return 0; } | 
|---|
| 576 | static inline void rcu_set_jiffies_lazy_flush(unsigned long j) { } | 
|---|
| 577 | #endif | 
|---|
| 578 |  | 
|---|
| 579 | #if defined(CONFIG_TREE_RCU) | 
|---|
| 580 | void rcutorture_get_gp_data(int *flags, unsigned long *gp_seq); | 
|---|
| 581 | void do_trace_rcu_torture_read(const char *rcutorturename, | 
|---|
| 582 | struct rcu_head *rhp, | 
|---|
| 583 | unsigned long secs, | 
|---|
| 584 | unsigned long c_old, | 
|---|
| 585 | unsigned long c); | 
|---|
| 586 | void rcu_gp_set_torture_wait(int duration); | 
|---|
| 587 | void rcu_set_gpwrap_lag(unsigned long lag); | 
|---|
| 588 | int rcu_get_gpwrap_count(int cpu); | 
|---|
| 589 | #else | 
|---|
| 590 | static inline void rcutorture_get_gp_data(int *flags, unsigned long *gp_seq) | 
|---|
| 591 | { | 
|---|
| 592 | *flags = 0; | 
|---|
| 593 | *gp_seq = 0; | 
|---|
| 594 | } | 
|---|
| 595 | #ifdef CONFIG_RCU_TRACE | 
|---|
| 596 | void do_trace_rcu_torture_read(const char *rcutorturename, | 
|---|
| 597 | struct rcu_head *rhp, | 
|---|
| 598 | unsigned long secs, | 
|---|
| 599 | unsigned long c_old, | 
|---|
| 600 | unsigned long c); | 
|---|
| 601 | #else | 
|---|
| 602 | #define do_trace_rcu_torture_read(rcutorturename, rhp, secs, c_old, c) \ | 
|---|
| 603 | do { } while (0) | 
|---|
| 604 | #endif | 
|---|
| 605 | static inline void rcu_gp_set_torture_wait(int duration) { } | 
|---|
| 606 | static inline void rcu_set_gpwrap_lag(unsigned long lag) { } | 
|---|
| 607 | static inline int rcu_get_gpwrap_count(int cpu) { return 0; } | 
|---|
| 608 | #endif | 
|---|
| 609 | unsigned long long rcutorture_gather_gp_seqs(void); | 
|---|
| 610 | void rcutorture_format_gp_seqs(unsigned long long seqs, char *cp, size_t len); | 
|---|
| 611 |  | 
|---|
| 612 | #ifdef CONFIG_TINY_SRCU | 
|---|
| 613 |  | 
|---|
| 614 | static inline void srcutorture_get_gp_data(struct srcu_struct *sp, int *flags, | 
|---|
| 615 | unsigned long *gp_seq) | 
|---|
| 616 | { | 
|---|
| 617 | *flags = 0; | 
|---|
| 618 | *gp_seq = sp->srcu_idx; | 
|---|
| 619 | } | 
|---|
| 620 |  | 
|---|
| 621 | #elif defined(CONFIG_TREE_SRCU) | 
|---|
| 622 |  | 
|---|
| 623 | void srcutorture_get_gp_data(struct srcu_struct *sp, int *flags, | 
|---|
| 624 | unsigned long *gp_seq); | 
|---|
| 625 |  | 
|---|
| 626 | #endif | 
|---|
| 627 |  | 
|---|
| 628 | #ifdef CONFIG_TINY_RCU | 
|---|
| 629 | static inline bool rcu_watching_zero_in_eqs(int cpu, int *vp) { return false; } | 
|---|
| 630 | static inline unsigned long rcu_get_gp_seq(void) { return 0; } | 
|---|
| 631 | static inline unsigned long rcu_exp_batches_completed(void) { return 0; } | 
|---|
| 632 | static inline void rcu_force_quiescent_state(void) { } | 
|---|
| 633 | static inline bool rcu_check_boost_fail(unsigned long gp_state, int *cpup) { return true; } | 
|---|
| 634 | static inline void show_rcu_gp_kthreads(void) { } | 
|---|
| 635 | static inline int rcu_get_gp_kthreads_prio(void) { return 0; } | 
|---|
| 636 | static inline void rcu_fwd_progress_check(unsigned long j) { } | 
|---|
| 637 | static inline void rcu_gp_slow_register(atomic_t *rgssp) { } | 
|---|
| 638 | static inline void rcu_gp_slow_unregister(atomic_t *rgssp) { } | 
|---|
| 639 | #else /* #ifdef CONFIG_TINY_RCU */ | 
|---|
| 640 | bool rcu_watching_zero_in_eqs(int cpu, int *vp); | 
|---|
| 641 | unsigned long rcu_get_gp_seq(void); | 
|---|
| 642 | unsigned long rcu_exp_batches_completed(void); | 
|---|
| 643 | bool rcu_check_boost_fail(unsigned long gp_state, int *cpup); | 
|---|
| 644 | void show_rcu_gp_kthreads(void); | 
|---|
| 645 | int rcu_get_gp_kthreads_prio(void); | 
|---|
| 646 | void rcu_fwd_progress_check(unsigned long j); | 
|---|
| 647 | void rcu_force_quiescent_state(void); | 
|---|
| 648 | extern struct workqueue_struct *rcu_gp_wq; | 
|---|
| 649 | extern struct kthread_worker *rcu_exp_gp_kworker; | 
|---|
| 650 | void rcu_gp_slow_register(atomic_t *rgssp); | 
|---|
| 651 | void rcu_gp_slow_unregister(atomic_t *rgssp); | 
|---|
| 652 | #endif /* #else #ifdef CONFIG_TINY_RCU */ | 
|---|
| 653 |  | 
|---|
| 654 | #ifdef CONFIG_TINY_SRCU | 
|---|
| 655 | static inline unsigned long srcu_batches_completed(struct srcu_struct *sp) { return 0; } | 
|---|
| 656 | #else // #ifdef CONFIG_TINY_SRCU | 
|---|
| 657 | unsigned long srcu_batches_completed(struct srcu_struct *sp); | 
|---|
| 658 | #endif // #else // #ifdef CONFIG_TINY_SRCU | 
|---|
| 659 |  | 
|---|
| 660 | #ifdef CONFIG_RCU_NOCB_CPU | 
|---|
| 661 | void rcu_bind_current_to_nocb(void); | 
|---|
| 662 | #else | 
|---|
| 663 | static inline void rcu_bind_current_to_nocb(void) { } | 
|---|
| 664 | #endif | 
|---|
| 665 |  | 
|---|
| 666 | #if !defined(CONFIG_TINY_RCU) && defined(CONFIG_TASKS_RCU) | 
|---|
| 667 | void show_rcu_tasks_classic_gp_kthread(void); | 
|---|
| 668 | #else | 
|---|
| 669 | static inline void show_rcu_tasks_classic_gp_kthread(void) {} | 
|---|
| 670 | #endif | 
|---|
| 671 | #if !defined(CONFIG_TINY_RCU) && defined(CONFIG_TASKS_RUDE_RCU) | 
|---|
| 672 | void show_rcu_tasks_rude_gp_kthread(void); | 
|---|
| 673 | #else | 
|---|
| 674 | static inline void show_rcu_tasks_rude_gp_kthread(void) {} | 
|---|
| 675 | #endif | 
|---|
| 676 | #if !defined(CONFIG_TINY_RCU) && defined(CONFIG_TASKS_TRACE_RCU) | 
|---|
| 677 | void show_rcu_tasks_trace_gp_kthread(void); | 
|---|
| 678 | #else | 
|---|
| 679 | static inline void show_rcu_tasks_trace_gp_kthread(void) {} | 
|---|
| 680 | #endif | 
|---|
| 681 |  | 
|---|
| 682 | #ifdef CONFIG_TINY_RCU | 
|---|
| 683 | static inline bool rcu_cpu_beenfullyonline(int cpu) { return true; } | 
|---|
| 684 | #else | 
|---|
| 685 | bool rcu_cpu_beenfullyonline(int cpu); | 
|---|
| 686 | #endif | 
|---|
| 687 |  | 
|---|
| 688 | #if defined(CONFIG_RCU_STALL_COMMON) && defined(CONFIG_RCU_CPU_STALL_NOTIFIER) | 
|---|
| 689 | int rcu_stall_notifier_call_chain(unsigned long val, void *v); | 
|---|
| 690 | #else // #if defined(CONFIG_RCU_STALL_COMMON) && defined(CONFIG_RCU_CPU_STALL_NOTIFIER) | 
|---|
| 691 | static inline int rcu_stall_notifier_call_chain(unsigned long val, void *v) { return NOTIFY_DONE; } | 
|---|
| 692 | #endif // #else // #if defined(CONFIG_RCU_STALL_COMMON) && defined(CONFIG_RCU_CPU_STALL_NOTIFIER) | 
|---|
| 693 |  | 
|---|
| 694 | #endif /* __LINUX_RCU_H */ | 
|---|
| 695 |  | 
|---|