rcutree.h 12 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324
  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. #include <linux/cache.h>
  25. #include <linux/spinlock.h>
  26. #include <linux/threads.h>
  27. #include <linux/cpumask.h>
  28. #include <linux/seqlock.h>
  29. /*
  30. * Define shape of hierarchy based on NR_CPUS and CONFIG_RCU_FANOUT.
  31. * In theory, it should be possible to add more levels straightforwardly.
  32. * In practice, this has not been tested, so there is probably some
  33. * bug somewhere.
  34. */
  35. #define MAX_RCU_LVLS 3
  36. #define RCU_FANOUT (CONFIG_RCU_FANOUT)
  37. #define RCU_FANOUT_SQ (RCU_FANOUT * RCU_FANOUT)
  38. #define RCU_FANOUT_CUBE (RCU_FANOUT_SQ * RCU_FANOUT)
  39. #if NR_CPUS <= RCU_FANOUT
  40. # define NUM_RCU_LVLS 1
  41. # define NUM_RCU_LVL_0 1
  42. # define NUM_RCU_LVL_1 (NR_CPUS)
  43. # define NUM_RCU_LVL_2 0
  44. # define NUM_RCU_LVL_3 0
  45. #elif NR_CPUS <= RCU_FANOUT_SQ
  46. # define NUM_RCU_LVLS 2
  47. # define NUM_RCU_LVL_0 1
  48. # define NUM_RCU_LVL_1 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT)
  49. # define NUM_RCU_LVL_2 (NR_CPUS)
  50. # define NUM_RCU_LVL_3 0
  51. #elif NR_CPUS <= RCU_FANOUT_CUBE
  52. # define NUM_RCU_LVLS 3
  53. # define NUM_RCU_LVL_0 1
  54. # define NUM_RCU_LVL_1 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_SQ)
  55. # define NUM_RCU_LVL_2 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT)
  56. # define NUM_RCU_LVL_3 NR_CPUS
  57. #else
  58. # error "CONFIG_RCU_FANOUT insufficient for NR_CPUS"
  59. #endif /* #if (NR_CPUS) <= RCU_FANOUT */
  60. #define RCU_SUM (NUM_RCU_LVL_0 + NUM_RCU_LVL_1 + NUM_RCU_LVL_2 + NUM_RCU_LVL_3)
  61. #define NUM_RCU_NODES (RCU_SUM - NR_CPUS)
  62. /*
  63. * Dynticks per-CPU state.
  64. */
  65. struct rcu_dynticks {
  66. int dynticks_nesting; /* Track nesting level, sort of. */
  67. int dynticks; /* Even value for dynticks-idle, else odd. */
  68. int dynticks_nmi; /* Even value for either dynticks-idle or */
  69. /* not in nmi handler, else odd. So this */
  70. /* remains even for nmi from irq handler. */
  71. };
  72. /*
  73. * Definition for node within the RCU grace-period-detection hierarchy.
  74. */
  75. struct rcu_node {
  76. spinlock_t lock; /* Root rcu_node's lock protects some */
  77. /* rcu_state fields as well as following. */
  78. long gpnum; /* Current grace period for this node. */
  79. /* This will either be equal to or one */
  80. /* behind the root rcu_node's gpnum. */
  81. unsigned long qsmask; /* CPUs or groups that need to switch in */
  82. /* order for current grace period to proceed.*/
  83. /* In leaf rcu_node, each bit corresponds to */
  84. /* an rcu_data structure, otherwise, each */
  85. /* bit corresponds to a child rcu_node */
  86. /* structure. */
  87. unsigned long qsmaskinit;
  88. /* Per-GP initialization for qsmask. */
  89. unsigned long grpmask; /* Mask to apply to parent qsmask. */
  90. /* Only one bit will be set in this mask. */
  91. int grplo; /* lowest-numbered CPU or group here. */
  92. int grphi; /* highest-numbered CPU or group here. */
  93. u8 grpnum; /* CPU/group number for next level up. */
  94. u8 level; /* root is at level 0. */
  95. struct rcu_node *parent;
  96. struct list_head blocked_tasks[2];
  97. /* Tasks blocked in RCU read-side critsect. */
  98. /* Grace period number (->gpnum) x blocked */
  99. /* by tasks on the (x & 0x1) element of the */
  100. /* blocked_tasks[] array. */
  101. } ____cacheline_internodealigned_in_smp;
  102. /*
  103. * Do a full breadth-first scan of the rcu_node structures for the
  104. * specified rcu_state structure.
  105. */
  106. #define rcu_for_each_node_breadth_first(rsp, rnp) \
  107. for ((rnp) = &(rsp)->node[0]; \
  108. (rnp) < &(rsp)->node[NUM_RCU_NODES]; (rnp)++)
  109. #define rcu_for_each_leaf_node(rsp, rnp) \
  110. for ((rnp) = (rsp)->level[NUM_RCU_LVLS - 1]; \
  111. (rnp) < &(rsp)->node[NUM_RCU_NODES]; (rnp)++)
  112. /* Index values for nxttail array in struct rcu_data. */
  113. #define RCU_DONE_TAIL 0 /* Also RCU_WAIT head. */
  114. #define RCU_WAIT_TAIL 1 /* Also RCU_NEXT_READY head. */
  115. #define RCU_NEXT_READY_TAIL 2 /* Also RCU_NEXT head. */
  116. #define RCU_NEXT_TAIL 3
  117. #define RCU_NEXT_SIZE 4
  118. /* Per-CPU data for read-copy update. */
  119. struct rcu_data {
  120. /* 1) quiescent-state and grace-period handling : */
  121. long completed; /* Track rsp->completed gp number */
  122. /* in order to detect GP end. */
  123. long gpnum; /* Highest gp number that this CPU */
  124. /* is aware of having started. */
  125. long passed_quiesc_completed;
  126. /* Value of completed at time of qs. */
  127. bool passed_quiesc; /* User-mode/idle loop etc. */
  128. bool qs_pending; /* Core waits for quiesc state. */
  129. bool beenonline; /* CPU online at least once. */
  130. bool preemptable; /* Preemptable RCU? */
  131. struct rcu_node *mynode; /* This CPU's leaf of hierarchy */
  132. unsigned long grpmask; /* Mask to apply to leaf qsmask. */
  133. /* 2) batch handling */
  134. /*
  135. * If nxtlist is not NULL, it is partitioned as follows.
  136. * Any of the partitions might be empty, in which case the
  137. * pointer to that partition will be equal to the pointer for
  138. * the following partition. When the list is empty, all of
  139. * the nxttail elements point to the ->nxtlist pointer itself,
  140. * which in that case is NULL.
  141. *
  142. * [nxtlist, *nxttail[RCU_DONE_TAIL]):
  143. * Entries that batch # <= ->completed
  144. * The grace period for these entries has completed, and
  145. * the other grace-period-completed entries may be moved
  146. * here temporarily in rcu_process_callbacks().
  147. * [*nxttail[RCU_DONE_TAIL], *nxttail[RCU_WAIT_TAIL]):
  148. * Entries that batch # <= ->completed - 1: waiting for current GP
  149. * [*nxttail[RCU_WAIT_TAIL], *nxttail[RCU_NEXT_READY_TAIL]):
  150. * Entries known to have arrived before current GP ended
  151. * [*nxttail[RCU_NEXT_READY_TAIL], *nxttail[RCU_NEXT_TAIL]):
  152. * Entries that might have arrived after current GP ended
  153. * Note that the value of *nxttail[RCU_NEXT_TAIL] will
  154. * always be NULL, as this is the end of the list.
  155. */
  156. struct rcu_head *nxtlist;
  157. struct rcu_head **nxttail[RCU_NEXT_SIZE];
  158. long qlen; /* # of queued callbacks */
  159. long qlen_last_fqs_check;
  160. /* qlen at last check for QS forcing */
  161. unsigned long n_force_qs_snap;
  162. /* did other CPU force QS recently? */
  163. long blimit; /* Upper limit on a processed batch */
  164. #ifdef CONFIG_NO_HZ
  165. /* 3) dynticks interface. */
  166. struct rcu_dynticks *dynticks; /* Shared per-CPU dynticks state. */
  167. int dynticks_snap; /* Per-GP tracking for dynticks. */
  168. int dynticks_nmi_snap; /* Per-GP tracking for dynticks_nmi. */
  169. #endif /* #ifdef CONFIG_NO_HZ */
  170. /* 4) reasons this CPU needed to be kicked by force_quiescent_state */
  171. #ifdef CONFIG_NO_HZ
  172. unsigned long dynticks_fqs; /* Kicked due to dynticks idle. */
  173. #endif /* #ifdef CONFIG_NO_HZ */
  174. unsigned long offline_fqs; /* Kicked due to being offline. */
  175. unsigned long resched_ipi; /* Sent a resched IPI. */
  176. /* 5) __rcu_pending() statistics. */
  177. long n_rcu_pending; /* rcu_pending() calls since boot. */
  178. long n_rp_qs_pending;
  179. long n_rp_cb_ready;
  180. long n_rp_cpu_needs_gp;
  181. long n_rp_gp_completed;
  182. long n_rp_gp_started;
  183. long n_rp_need_fqs;
  184. long n_rp_need_nothing;
  185. int cpu;
  186. };
  187. /* Values for signaled field in struct rcu_state. */
  188. #define RCU_GP_IDLE 0 /* No grace period in progress. */
  189. #define RCU_GP_INIT 1 /* Grace period being initialized. */
  190. #define RCU_SAVE_DYNTICK 2 /* Need to scan dyntick state. */
  191. #define RCU_FORCE_QS 3 /* Need to force quiescent state. */
  192. #ifdef CONFIG_NO_HZ
  193. #define RCU_SIGNAL_INIT RCU_SAVE_DYNTICK
  194. #else /* #ifdef CONFIG_NO_HZ */
  195. #define RCU_SIGNAL_INIT RCU_FORCE_QS
  196. #endif /* #else #ifdef CONFIG_NO_HZ */
  197. #define RCU_JIFFIES_TILL_FORCE_QS 3 /* for rsp->jiffies_force_qs */
  198. #ifdef CONFIG_RCU_CPU_STALL_DETECTOR
  199. #define RCU_SECONDS_TILL_STALL_CHECK (10 * HZ) /* for rsp->jiffies_stall */
  200. #define RCU_SECONDS_TILL_STALL_RECHECK (30 * HZ) /* for rsp->jiffies_stall */
  201. #define RCU_STALL_RAT_DELAY 2 /* Allow other CPUs time */
  202. /* to take at least one */
  203. /* scheduling clock irq */
  204. /* before ratting on them. */
  205. #endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
  206. /*
  207. * RCU global state, including node hierarchy. This hierarchy is
  208. * represented in "heap" form in a dense array. The root (first level)
  209. * of the hierarchy is in ->node[0] (referenced by ->level[0]), the second
  210. * level in ->node[1] through ->node[m] (->node[1] referenced by ->level[1]),
  211. * and the third level in ->node[m+1] and following (->node[m+1] referenced
  212. * by ->level[2]). The number of levels is determined by the number of
  213. * CPUs and by CONFIG_RCU_FANOUT. Small systems will have a "hierarchy"
  214. * consisting of a single rcu_node.
  215. */
  216. struct rcu_state {
  217. struct rcu_node node[NUM_RCU_NODES]; /* Hierarchy. */
  218. struct rcu_node *level[NUM_RCU_LVLS]; /* Hierarchy levels. */
  219. u32 levelcnt[MAX_RCU_LVLS + 1]; /* # nodes in each level. */
  220. u8 levelspread[NUM_RCU_LVLS]; /* kids/node in each level. */
  221. struct rcu_data *rda[NR_CPUS]; /* array of rdp pointers. */
  222. /* The following fields are guarded by the root rcu_node's lock. */
  223. u8 signaled ____cacheline_internodealigned_in_smp;
  224. /* Force QS state. */
  225. long gpnum; /* Current gp number. */
  226. long completed; /* # of last completed gp. */
  227. /* End of fields guarded by root rcu_node's lock. */
  228. spinlock_t onofflock; /* exclude on/offline and */
  229. /* starting new GP. Also */
  230. /* protects the following */
  231. /* orphan_cbs fields. */
  232. struct rcu_head *orphan_cbs_list; /* list of rcu_head structs */
  233. /* orphaned by all CPUs in */
  234. /* a given leaf rcu_node */
  235. /* going offline. */
  236. struct rcu_head **orphan_cbs_tail; /* And tail pointer. */
  237. long orphan_qlen; /* Number of orphaned cbs. */
  238. spinlock_t fqslock; /* Only one task forcing */
  239. /* quiescent states. */
  240. unsigned long jiffies_force_qs; /* Time at which to invoke */
  241. /* force_quiescent_state(). */
  242. unsigned long n_force_qs; /* Number of calls to */
  243. /* force_quiescent_state(). */
  244. unsigned long n_force_qs_lh; /* ~Number of calls leaving */
  245. /* due to lock unavailable. */
  246. unsigned long n_force_qs_ngp; /* Number of calls leaving */
  247. /* due to no GP active. */
  248. #ifdef CONFIG_RCU_CPU_STALL_DETECTOR
  249. unsigned long gp_start; /* Time at which GP started, */
  250. /* but in jiffies. */
  251. unsigned long jiffies_stall; /* Time at which to check */
  252. /* for CPU stalls. */
  253. #endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
  254. #ifdef CONFIG_NO_HZ
  255. long dynticks_completed; /* Value of completed @ snap. */
  256. #endif /* #ifdef CONFIG_NO_HZ */
  257. };
  258. #ifdef RCU_TREE_NONCORE
  259. /*
  260. * RCU implementation internal declarations:
  261. */
  262. extern struct rcu_state rcu_sched_state;
  263. DECLARE_PER_CPU(struct rcu_data, rcu_sched_data);
  264. extern struct rcu_state rcu_bh_state;
  265. DECLARE_PER_CPU(struct rcu_data, rcu_bh_data);
  266. #ifdef CONFIG_TREE_PREEMPT_RCU
  267. extern struct rcu_state rcu_preempt_state;
  268. DECLARE_PER_CPU(struct rcu_data, rcu_preempt_data);
  269. #endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */
  270. #else /* #ifdef RCU_TREE_NONCORE */
  271. /* Forward declarations for rcutree_plugin.h */
  272. static inline void rcu_bootup_announce(void);
  273. long rcu_batches_completed(void);
  274. static void rcu_preempt_note_context_switch(int cpu);
  275. static int rcu_preempted_readers(struct rcu_node *rnp);
  276. #ifdef CONFIG_RCU_CPU_STALL_DETECTOR
  277. static void rcu_print_task_stall(struct rcu_node *rnp);
  278. #endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
  279. static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp);
  280. #ifdef CONFIG_HOTPLUG_CPU
  281. static int rcu_preempt_offline_tasks(struct rcu_state *rsp,
  282. struct rcu_node *rnp,
  283. struct rcu_data *rdp);
  284. static void rcu_preempt_offline_cpu(int cpu);
  285. #endif /* #ifdef CONFIG_HOTPLUG_CPU */
  286. static void rcu_preempt_check_callbacks(int cpu);
  287. static void rcu_preempt_process_callbacks(void);
  288. void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu));
  289. static int rcu_preempt_pending(int cpu);
  290. static int rcu_preempt_needs_cpu(int cpu);
  291. static void __cpuinit rcu_preempt_init_percpu_data(int cpu);
  292. static void rcu_preempt_send_cbs_to_orphanage(void);
  293. static void __init __rcu_init_preempt(void);
  294. #endif /* #else #ifdef RCU_TREE_NONCORE */