1/* 2 * Read-Copy Update mechanism for mutual exclusion (tree-based version) 3 * Internal non-public definitions. 4 * 5 * This program is free software; you can redistribute it and/or modify 6 * it under the terms of the GNU General Public License as published by 7 * the Free Software Foundation; either version 2 of the License, or 8 * (at your option) any later version. 9 * 10 * This program is distributed in the hope that it will be useful, 11 * but WITHOUT ANY WARRANTY; without even the implied warranty of 12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 13 * GNU General Public License for more details. 14 * 15 * You should have received a copy of the GNU General Public License 16 * along with this program; if not, write to the Free Software 17 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. 18 * 19 * Copyright IBM Corporation, 2008 20 * 21 * Author: Ingo Molnar <mingo@elte.hu> 22 * Paul E. McKenney <paulmck@linux.vnet.ibm.com> 23 */ 24 25#include <linux/cache.h> 26#include <linux/spinlock.h> 27#include <linux/threads.h> 28#include <linux/cpumask.h> 29#include <linux/seqlock.h> 30 31/* 32 * Define shape of hierarchy based on NR_CPUS and CONFIG_RCU_FANOUT. 33 * In theory, it should be possible to add more levels straightforwardly. 34 * In practice, this did work well going from three levels to four. 35 * Of course, your mileage may vary. 36 */ 37#define MAX_RCU_LVLS 4 38#if CONFIG_RCU_FANOUT > 16 39#define RCU_FANOUT_LEAF 16 40#else /* #if CONFIG_RCU_FANOUT > 16 */ 41#define RCU_FANOUT_LEAF (CONFIG_RCU_FANOUT) 42#endif /* #else #if CONFIG_RCU_FANOUT > 16 */ 43#define RCU_FANOUT_1 (RCU_FANOUT_LEAF) 44#define RCU_FANOUT_2 (RCU_FANOUT_1 * CONFIG_RCU_FANOUT) 45#define RCU_FANOUT_3 (RCU_FANOUT_2 * CONFIG_RCU_FANOUT) 46#define RCU_FANOUT_4 (RCU_FANOUT_3 * CONFIG_RCU_FANOUT) 47 48#if NR_CPUS <= RCU_FANOUT_1 49# define NUM_RCU_LVLS 1 50# define NUM_RCU_LVL_0 1 51# define NUM_RCU_LVL_1 (NR_CPUS) 52# define NUM_RCU_LVL_2 0 53# define NUM_RCU_LVL_3 0 54# define NUM_RCU_LVL_4 0 55#elif NR_CPUS <= RCU_FANOUT_2 56# define NUM_RCU_LVLS 2 57# define NUM_RCU_LVL_0 1 58# define NUM_RCU_LVL_1 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1) 59# define NUM_RCU_LVL_2 (NR_CPUS) 60# define NUM_RCU_LVL_3 0 61# define NUM_RCU_LVL_4 0 62#elif NR_CPUS <= RCU_FANOUT_3 63# define NUM_RCU_LVLS 3 64# define NUM_RCU_LVL_0 1 65# define NUM_RCU_LVL_1 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_2) 66# define NUM_RCU_LVL_2 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1) 67# define NUM_RCU_LVL_3 (NR_CPUS) 68# define NUM_RCU_LVL_4 0 69#elif NR_CPUS <= RCU_FANOUT_4 70# define NUM_RCU_LVLS 4 71# define NUM_RCU_LVL_0 1 72# define NUM_RCU_LVL_1 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_3) 73# define NUM_RCU_LVL_2 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_2) 74# define NUM_RCU_LVL_3 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_1) 75# define NUM_RCU_LVL_4 (NR_CPUS) 76#else 77# error "CONFIG_RCU_FANOUT insufficient for NR_CPUS" 78#endif /* #if (NR_CPUS) <= RCU_FANOUT_1 */ 79 80#define RCU_SUM (NUM_RCU_LVL_0 + NUM_RCU_LVL_1 + NUM_RCU_LVL_2 + NUM_RCU_LVL_3 + NUM_RCU_LVL_4) 81#define NUM_RCU_NODES (RCU_SUM - NR_CPUS) 82 83/* 84 * Dynticks per-CPU state. 85 */ 86struct rcu_dynticks { 87 int dynticks_nesting; /* Track nesting level, sort of. */ 88 int dynticks; /* Even value for dynticks-idle, else odd. */ 89 int dynticks_nmi; /* Even value for either dynticks-idle or */ 90 /* not in nmi handler, else odd. So this */ 91 /* remains even for nmi from irq handler. */ 92}; 93 94/* 95 * Definition for node within the RCU grace-period-detection hierarchy. 96 */ 97struct rcu_node { 98 raw_spinlock_t lock; /* Root rcu_node's lock protects some */ 99 /* rcu_state fields as well as following. */ 100 unsigned long gpnum; /* Current grace period for this node. */ 101 /* This will either be equal to or one */ 102 /* behind the root rcu_node's gpnum. */ 103 unsigned long completed; /* Last GP completed for this node. */ 104 /* This will either be equal to or one */ 105 /* behind the root rcu_node's gpnum. */ 106 unsigned long qsmask; /* CPUs or groups that need to switch in */ 107 /* order for current grace period to proceed.*/ 108 /* In leaf rcu_node, each bit corresponds to */ 109 /* an rcu_data structure, otherwise, each */ 110 /* bit corresponds to a child rcu_node */ 111 /* structure. */ 112 unsigned long expmask; /* Groups that have ->blocked_tasks[] */ 113 /* elements that need to drain to allow the */ 114 /* current expedited grace period to */ 115 /* complete (only for TREE_PREEMPT_RCU). */ 116 unsigned long qsmaskinit; 117 /* Per-GP initial value for qsmask & expmask. */ 118 unsigned long grpmask; /* Mask to apply to parent qsmask. */ 119 /* Only one bit will be set in this mask. */ 120 int grplo; /* lowest-numbered CPU or group here. */ 121 int grphi; /* highest-numbered CPU or group here. */ 122 u8 grpnum; /* CPU/group number for next level up. */ 123 u8 level; /* root is at level 0. */ 124 struct rcu_node *parent; 125 struct list_head blocked_tasks[4]; 126 /* Tasks blocked in RCU read-side critsect. */ 127 /* Grace period number (->gpnum) x blocked */ 128 /* by tasks on the (x & 0x1) element of the */ 129 /* blocked_tasks[] array. */ 130} ____cacheline_internodealigned_in_smp; 131 132/* 133 * Do a full breadth-first scan of the rcu_node structures for the 134 * specified rcu_state structure. 135 */ 136#define rcu_for_each_node_breadth_first(rsp, rnp) \ 137 for ((rnp) = &(rsp)->node[0]; \ 138 (rnp) < &(rsp)->node[NUM_RCU_NODES]; (rnp)++) 139 140/* 141 * Do a breadth-first scan of the non-leaf rcu_node structures for the 142 * specified rcu_state structure. Note that if there is a singleton 143 * rcu_node tree with but one rcu_node structure, this loop is a no-op. 144 */ 145#define rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) \ 146 for ((rnp) = &(rsp)->node[0]; \ 147 (rnp) < (rsp)->level[NUM_RCU_LVLS - 1]; (rnp)++) 148 149/* 150 * Scan the leaves of the rcu_node hierarchy for the specified rcu_state 151 * structure. Note that if there is a singleton rcu_node tree with but 152 * one rcu_node structure, this loop -will- visit the rcu_node structure. 153 * It is still a leaf node, even if it is also the root node. 154 */ 155#define rcu_for_each_leaf_node(rsp, rnp) \ 156 for ((rnp) = (rsp)->level[NUM_RCU_LVLS - 1]; \ 157 (rnp) < &(rsp)->node[NUM_RCU_NODES]; (rnp)++) 158 159/* Index values for nxttail array in struct rcu_data. */ 160#define RCU_DONE_TAIL 0 /* Also RCU_WAIT head. */ 161#define RCU_WAIT_TAIL 1 /* Also RCU_NEXT_READY head. */ 162#define RCU_NEXT_READY_TAIL 2 /* Also RCU_NEXT head. */ 163#define RCU_NEXT_TAIL 3 164#define RCU_NEXT_SIZE 4 165 166/* Per-CPU data for read-copy update. */ 167struct rcu_data { 168 /* 1) quiescent-state and grace-period handling : */ 169 unsigned long completed; /* Track rsp->completed gp number */ 170 /* in order to detect GP end. */ 171 unsigned long gpnum; /* Highest gp number that this CPU */ 172 /* is aware of having started. */ 173 unsigned long passed_quiesc_completed; 174 /* Value of completed at time of qs. */ 175 bool passed_quiesc; /* User-mode/idle loop etc. */ 176 bool qs_pending; /* Core waits for quiesc state. */ 177 bool beenonline; /* CPU online at least once. */ 178 bool preemptable; /* Preemptable RCU? */ 179 struct rcu_node *mynode; /* This CPU's leaf of hierarchy */ 180 unsigned long grpmask; /* Mask to apply to leaf qsmask. */ 181 182 /* 2) batch handling */ 183 /* 184 * If nxtlist is not NULL, it is partitioned as follows. 185 * Any of the partitions might be empty, in which case the 186 * pointer to that partition will be equal to the pointer for 187 * the following partition. When the list is empty, all of 188 * the nxttail elements point to the ->nxtlist pointer itself, 189 * which in that case is NULL. 190 * 191 * [nxtlist, *nxttail[RCU_DONE_TAIL]): 192 * Entries that batch # <= ->completed 193 * The grace period for these entries has completed, and 194 * the other grace-period-completed entries may be moved 195 * here temporarily in rcu_process_callbacks(). 196 * [*nxttail[RCU_DONE_TAIL], *nxttail[RCU_WAIT_TAIL]): 197 * Entries that batch # <= ->completed - 1: waiting for current GP 198 * [*nxttail[RCU_WAIT_TAIL], *nxttail[RCU_NEXT_READY_TAIL]): 199 * Entries known to have arrived before current GP ended 200 * [*nxttail[RCU_NEXT_READY_TAIL], *nxttail[RCU_NEXT_TAIL]): 201 * Entries that might have arrived after current GP ended 202 * Note that the value of *nxttail[RCU_NEXT_TAIL] will 203 * always be NULL, as this is the end of the list. 204 */ 205 struct rcu_head *nxtlist; 206 struct rcu_head **nxttail[RCU_NEXT_SIZE]; 207 long qlen; /* # of queued callbacks */ 208 long qlen_last_fqs_check; 209 /* qlen at last check for QS forcing */ 210 unsigned long n_cbs_invoked; /* count of RCU cbs invoked. */ 211 unsigned long n_cbs_orphaned; /* RCU cbs orphaned by dying CPU */ 212 unsigned long n_cbs_adopted; /* RCU cbs adopted from dying CPU */ 213 unsigned long n_force_qs_snap; 214 /* did other CPU force QS recently? */ 215 long blimit; /* Upper limit on a processed batch */ 216 217#ifdef CONFIG_NO_HZ 218 /* 3) dynticks interface. */ 219 struct rcu_dynticks *dynticks; /* Shared per-CPU dynticks state. */ 220 int dynticks_snap; /* Per-GP tracking for dynticks. */ 221 int dynticks_nmi_snap; /* Per-GP tracking for dynticks_nmi. */ 222#endif /* #ifdef CONFIG_NO_HZ */ 223 224 /* 4) reasons this CPU needed to be kicked by force_quiescent_state */ 225#ifdef CONFIG_NO_HZ 226 unsigned long dynticks_fqs; /* Kicked due to dynticks idle. */ 227#endif /* #ifdef CONFIG_NO_HZ */ 228 unsigned long offline_fqs; /* Kicked due to being offline. */ 229 unsigned long resched_ipi; /* Sent a resched IPI. */ 230 231 /* 5) __rcu_pending() statistics. */ 232 unsigned long n_rcu_pending; /* rcu_pending() calls since boot. */ 233 unsigned long n_rp_qs_pending; 234 unsigned long n_rp_report_qs; 235 unsigned long n_rp_cb_ready; 236 unsigned long n_rp_cpu_needs_gp; 237 unsigned long n_rp_gp_completed; 238 unsigned long n_rp_gp_started; 239 unsigned long n_rp_need_fqs; 240 unsigned long n_rp_need_nothing; 241 242 int cpu; 243}; 244 245/* Values for signaled field in struct rcu_state. */ 246#define RCU_GP_IDLE 0 /* No grace period in progress. */ 247#define RCU_GP_INIT 1 /* Grace period being initialized. */ 248#define RCU_SAVE_DYNTICK 2 /* Need to scan dyntick state. */ 249#define RCU_FORCE_QS 3 /* Need to force quiescent state. */ 250#ifdef CONFIG_NO_HZ 251#define RCU_SIGNAL_INIT RCU_SAVE_DYNTICK 252#else /* #ifdef CONFIG_NO_HZ */ 253#define RCU_SIGNAL_INIT RCU_FORCE_QS 254#endif /* #else #ifdef CONFIG_NO_HZ */ 255 256#define RCU_JIFFIES_TILL_FORCE_QS 3 /* for rsp->jiffies_force_qs */ 257#ifdef CONFIG_RCU_CPU_STALL_DETECTOR 258 259#ifdef CONFIG_PROVE_RCU 260#define RCU_STALL_DELAY_DELTA (5 * HZ) 261#else 262#define RCU_STALL_DELAY_DELTA 0 263#endif 264 265#define RCU_SECONDS_TILL_STALL_CHECK (CONFIG_RCU_CPU_STALL_TIMEOUT * HZ + \ 266 RCU_STALL_DELAY_DELTA) 267 /* for rsp->jiffies_stall */ 268#define RCU_SECONDS_TILL_STALL_RECHECK (3 * RCU_SECONDS_TILL_STALL_CHECK + 30) 269 /* for rsp->jiffies_stall */ 270#define RCU_STALL_RAT_DELAY 2 /* Allow other CPUs time */ 271 /* to take at least one */ 272 /* scheduling clock irq */ 273 /* before ratting on them. */ 274 275#ifdef CONFIG_RCU_CPU_STALL_DETECTOR_RUNNABLE 276#define RCU_CPU_STALL_SUPPRESS_INIT 0 277#else 278#define RCU_CPU_STALL_SUPPRESS_INIT 1 279#endif 280 281#endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */ 282 283/* 284 * RCU global state, including node hierarchy. This hierarchy is 285 * represented in "heap" form in a dense array. The root (first level) 286 * of the hierarchy is in ->node[0] (referenced by ->level[0]), the second 287 * level in ->node[1] through ->node[m] (->node[1] referenced by ->level[1]), 288 * and the third level in ->node[m+1] and following (->node[m+1] referenced 289 * by ->level[2]). The number of levels is determined by the number of 290 * CPUs and by CONFIG_RCU_FANOUT. Small systems will have a "hierarchy" 291 * consisting of a single rcu_node. 292 */ 293struct rcu_state { 294 struct rcu_node node[NUM_RCU_NODES]; /* Hierarchy. */ 295 struct rcu_node *level[NUM_RCU_LVLS]; /* Hierarchy levels. */ 296 u32 levelcnt[MAX_RCU_LVLS + 1]; /* # nodes in each level. */ 297 u8 levelspread[NUM_RCU_LVLS]; /* kids/node in each level. */ 298 struct rcu_data __percpu *rda; /* pointer of percu rcu_data. */ 299 300 /* The following fields are guarded by the root rcu_node's lock. */ 301 302 u8 signaled ____cacheline_internodealigned_in_smp; 303 /* Force QS state. */ 304 u8 fqs_active; /* force_quiescent_state() */ 305 /* is running. */ 306 u8 fqs_need_gp; /* A CPU was prevented from */ 307 /* starting a new grace */ 308 /* period because */ 309 /* force_quiescent_state() */ 310 /* was running. */ 311 unsigned long gpnum; /* Current gp number. */ 312 unsigned long completed; /* # of last completed gp. */ 313 314 /* End of fields guarded by root rcu_node's lock. */ 315 316 raw_spinlock_t onofflock; /* exclude on/offline and */ 317 /* starting new GP. */ 318 raw_spinlock_t fqslock; /* Only one task forcing */ 319 /* quiescent states. */ 320 unsigned long jiffies_force_qs; /* Time at which to invoke */ 321 /* force_quiescent_state(). */ 322 unsigned long n_force_qs; /* Number of calls to */ 323 /* force_quiescent_state(). */ 324 unsigned long n_force_qs_lh; /* ~Number of calls leaving */ 325 /* due to lock unavailable. */ 326 unsigned long n_force_qs_ngp; /* Number of calls leaving */ 327 /* due to no GP active. */ 328#ifdef CONFIG_RCU_CPU_STALL_DETECTOR 329 unsigned long gp_start; /* Time at which GP started, */ 330 /* but in jiffies. */ 331 unsigned long jiffies_stall; /* Time at which to check */ 332 /* for CPU stalls. */ 333#endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */ 334 char *name; /* Name of structure. */ 335}; 336 337/* Return values for rcu_preempt_offline_tasks(). */ 338 339#define RCU_OFL_TASKS_NORM_GP 0x1 /* Tasks blocking normal */ 340 /* GP were moved to root. */ 341#define RCU_OFL_TASKS_EXP_GP 0x2 /* Tasks blocking expedited */ 342 /* GP were moved to root. */ 343 344/* 345 * RCU implementation internal declarations: 346 */ 347extern struct rcu_state rcu_sched_state; 348DECLARE_PER_CPU(struct rcu_data, rcu_sched_data); 349 350extern struct rcu_state rcu_bh_state; 351DECLARE_PER_CPU(struct rcu_data, rcu_bh_data); 352 353#ifdef CONFIG_TREE_PREEMPT_RCU 354extern struct rcu_state rcu_preempt_state; 355DECLARE_PER_CPU(struct rcu_data, rcu_preempt_data); 356#endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */ 357 358#ifndef RCU_TREE_NONCORE 359 360/* Forward declarations for rcutree_plugin.h */ 361static void rcu_bootup_announce(void); 362long rcu_batches_completed(void); 363static void rcu_preempt_note_context_switch(int cpu); 364static int rcu_preempted_readers(struct rcu_node *rnp); 365#ifdef CONFIG_HOTPLUG_CPU 366static void rcu_report_unblock_qs_rnp(struct rcu_node *rnp, 367 unsigned long flags); 368#endif /* #ifdef CONFIG_HOTPLUG_CPU */ 369#ifdef CONFIG_RCU_CPU_STALL_DETECTOR 370static void rcu_print_detail_task_stall(struct rcu_state *rsp); 371static void rcu_print_task_stall(struct rcu_node *rnp); 372static void rcu_preempt_stall_reset(void); 373#endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */ 374static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp); 375#ifdef CONFIG_HOTPLUG_CPU 376static int rcu_preempt_offline_tasks(struct rcu_state *rsp, 377 struct rcu_node *rnp, 378 struct rcu_data *rdp); 379static void rcu_preempt_offline_cpu(int cpu); 380#endif /* #ifdef CONFIG_HOTPLUG_CPU */ 381static void rcu_preempt_check_callbacks(int cpu); 382static void rcu_preempt_process_callbacks(void); 383void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu)); 384#if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_TREE_PREEMPT_RCU) 385static void rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp); 386#endif /* #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_TREE_PREEMPT_RCU) */ 387static int rcu_preempt_pending(int cpu); 388static int rcu_preempt_needs_cpu(int cpu); 389static void __cpuinit rcu_preempt_init_percpu_data(int cpu); 390static void rcu_preempt_send_cbs_to_online(void); 391static void __init __rcu_init_preempt(void); 392static void rcu_needs_cpu_flush(void); 393 394#endif /* #ifndef RCU_TREE_NONCORE */ 395