rcutree.h 9.7 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257
  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 (((NR_CPUS) + RCU_FANOUT - 1) / 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 (((NR_CPUS) + RCU_FANOUT_SQ - 1) / RCU_FANOUT_SQ)
  55. # define NUM_RCU_LVL_2 (((NR_CPUS) + (RCU_FANOUT) - 1) / (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;
  77. long gpnum; /* Current grace period for this node. */
  78. unsigned long qsmask; /* CPUs or groups that need to switch in */
  79. /* order for current grace period to proceed.*/
  80. unsigned long qsmaskinit;
  81. /* Per-GP initialization for qsmask. */
  82. unsigned long grpmask; /* Mask to apply to parent qsmask. */
  83. int grplo; /* lowest-numbered CPU or group here. */
  84. int grphi; /* highest-numbered CPU or group here. */
  85. u8 grpnum; /* CPU/group number for next level up. */
  86. u8 level; /* root is at level 0. */
  87. struct rcu_node *parent;
  88. struct list_head blocked_tasks[2];
  89. /* Tasks blocked in RCU read-side critsect. */
  90. } ____cacheline_internodealigned_in_smp;
  91. /* Index values for nxttail array in struct rcu_data. */
  92. #define RCU_DONE_TAIL 0 /* Also RCU_WAIT head. */
  93. #define RCU_WAIT_TAIL 1 /* Also RCU_NEXT_READY head. */
  94. #define RCU_NEXT_READY_TAIL 2 /* Also RCU_NEXT head. */
  95. #define RCU_NEXT_TAIL 3
  96. #define RCU_NEXT_SIZE 4
  97. /* Per-CPU data for read-copy update. */
  98. struct rcu_data {
  99. /* 1) quiescent-state and grace-period handling : */
  100. long completed; /* Track rsp->completed gp number */
  101. /* in order to detect GP end. */
  102. long gpnum; /* Highest gp number that this CPU */
  103. /* is aware of having started. */
  104. long passed_quiesc_completed;
  105. /* Value of completed at time of qs. */
  106. bool passed_quiesc; /* User-mode/idle loop etc. */
  107. bool qs_pending; /* Core waits for quiesc state. */
  108. bool beenonline; /* CPU online at least once. */
  109. bool preemptable; /* Preemptable RCU? */
  110. struct rcu_node *mynode; /* This CPU's leaf of hierarchy */
  111. unsigned long grpmask; /* Mask to apply to leaf qsmask. */
  112. /* 2) batch handling */
  113. /*
  114. * If nxtlist is not NULL, it is partitioned as follows.
  115. * Any of the partitions might be empty, in which case the
  116. * pointer to that partition will be equal to the pointer for
  117. * the following partition. When the list is empty, all of
  118. * the nxttail elements point to nxtlist, which is NULL.
  119. *
  120. * [*nxttail[RCU_NEXT_READY_TAIL], NULL = *nxttail[RCU_NEXT_TAIL]):
  121. * Entries that might have arrived after current GP ended
  122. * [*nxttail[RCU_WAIT_TAIL], *nxttail[RCU_NEXT_READY_TAIL]):
  123. * Entries known to have arrived before current GP ended
  124. * [*nxttail[RCU_DONE_TAIL], *nxttail[RCU_WAIT_TAIL]):
  125. * Entries that batch # <= ->completed - 1: waiting for current GP
  126. * [nxtlist, *nxttail[RCU_DONE_TAIL]):
  127. * Entries that batch # <= ->completed
  128. * The grace period for these entries has completed, and
  129. * the other grace-period-completed entries may be moved
  130. * here temporarily in rcu_process_callbacks().
  131. */
  132. struct rcu_head *nxtlist;
  133. struct rcu_head **nxttail[RCU_NEXT_SIZE];
  134. long qlen; /* # of queued callbacks */
  135. long blimit; /* Upper limit on a processed batch */
  136. #ifdef CONFIG_NO_HZ
  137. /* 3) dynticks interface. */
  138. struct rcu_dynticks *dynticks; /* Shared per-CPU dynticks state. */
  139. int dynticks_snap; /* Per-GP tracking for dynticks. */
  140. int dynticks_nmi_snap; /* Per-GP tracking for dynticks_nmi. */
  141. #endif /* #ifdef CONFIG_NO_HZ */
  142. /* 4) reasons this CPU needed to be kicked by force_quiescent_state */
  143. #ifdef CONFIG_NO_HZ
  144. unsigned long dynticks_fqs; /* Kicked due to dynticks idle. */
  145. #endif /* #ifdef CONFIG_NO_HZ */
  146. unsigned long offline_fqs; /* Kicked due to being offline. */
  147. unsigned long resched_ipi; /* Sent a resched IPI. */
  148. /* 5) __rcu_pending() statistics. */
  149. long n_rcu_pending; /* rcu_pending() calls since boot. */
  150. long n_rp_qs_pending;
  151. long n_rp_cb_ready;
  152. long n_rp_cpu_needs_gp;
  153. long n_rp_gp_completed;
  154. long n_rp_gp_started;
  155. long n_rp_need_fqs;
  156. long n_rp_need_nothing;
  157. int cpu;
  158. };
  159. /* Values for signaled field in struct rcu_state. */
  160. #define RCU_GP_INIT 0 /* Grace period being initialized. */
  161. #define RCU_SAVE_DYNTICK 1 /* Need to scan dyntick state. */
  162. #define RCU_FORCE_QS 2 /* Need to force quiescent state. */
  163. #ifdef CONFIG_NO_HZ
  164. #define RCU_SIGNAL_INIT RCU_SAVE_DYNTICK
  165. #else /* #ifdef CONFIG_NO_HZ */
  166. #define RCU_SIGNAL_INIT RCU_FORCE_QS
  167. #endif /* #else #ifdef CONFIG_NO_HZ */
  168. #define RCU_JIFFIES_TILL_FORCE_QS 3 /* for rsp->jiffies_force_qs */
  169. #ifdef CONFIG_RCU_CPU_STALL_DETECTOR
  170. #define RCU_SECONDS_TILL_STALL_CHECK (10 * HZ) /* for rsp->jiffies_stall */
  171. #define RCU_SECONDS_TILL_STALL_RECHECK (30 * HZ) /* for rsp->jiffies_stall */
  172. #define RCU_STALL_RAT_DELAY 2 /* Allow other CPUs time */
  173. /* to take at least one */
  174. /* scheduling clock irq */
  175. /* before ratting on them. */
  176. #endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
  177. /*
  178. * RCU global state, including node hierarchy. This hierarchy is
  179. * represented in "heap" form in a dense array. The root (first level)
  180. * of the hierarchy is in ->node[0] (referenced by ->level[0]), the second
  181. * level in ->node[1] through ->node[m] (->node[1] referenced by ->level[1]),
  182. * and the third level in ->node[m+1] and following (->node[m+1] referenced
  183. * by ->level[2]). The number of levels is determined by the number of
  184. * CPUs and by CONFIG_RCU_FANOUT. Small systems will have a "hierarchy"
  185. * consisting of a single rcu_node.
  186. */
  187. struct rcu_state {
  188. struct rcu_node node[NUM_RCU_NODES]; /* Hierarchy. */
  189. struct rcu_node *level[NUM_RCU_LVLS]; /* Hierarchy levels. */
  190. u32 levelcnt[MAX_RCU_LVLS + 1]; /* # nodes in each level. */
  191. u8 levelspread[NUM_RCU_LVLS]; /* kids/node in each level. */
  192. struct rcu_data *rda[NR_CPUS]; /* array of rdp pointers. */
  193. /* The following fields are guarded by the root rcu_node's lock. */
  194. u8 signaled ____cacheline_internodealigned_in_smp;
  195. /* Force QS state. */
  196. long gpnum; /* Current gp number. */
  197. long completed; /* # of last completed gp. */
  198. spinlock_t onofflock; /* exclude on/offline and */
  199. /* starting new GP. */
  200. spinlock_t fqslock; /* Only one task forcing */
  201. /* quiescent states. */
  202. unsigned long jiffies_force_qs; /* Time at which to invoke */
  203. /* force_quiescent_state(). */
  204. unsigned long n_force_qs; /* Number of calls to */
  205. /* force_quiescent_state(). */
  206. unsigned long n_force_qs_lh; /* ~Number of calls leaving */
  207. /* due to lock unavailable. */
  208. unsigned long n_force_qs_ngp; /* Number of calls leaving */
  209. /* due to no GP active. */
  210. #ifdef CONFIG_RCU_CPU_STALL_DETECTOR
  211. unsigned long gp_start; /* Time at which GP started, */
  212. /* but in jiffies. */
  213. unsigned long jiffies_stall; /* Time at which to check */
  214. /* for CPU stalls. */
  215. #endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
  216. #ifdef CONFIG_NO_HZ
  217. long dynticks_completed; /* Value of completed @ snap. */
  218. #endif /* #ifdef CONFIG_NO_HZ */
  219. };
  220. #ifdef RCU_TREE_NONCORE
  221. /*
  222. * RCU implementation internal declarations:
  223. */
  224. extern struct rcu_state rcu_sched_state;
  225. DECLARE_PER_CPU(struct rcu_data, rcu_sched_data);
  226. extern struct rcu_state rcu_bh_state;
  227. DECLARE_PER_CPU(struct rcu_data, rcu_bh_data);
  228. #ifdef CONFIG_TREE_PREEMPT_RCU
  229. extern struct rcu_state rcu_preempt_state;
  230. DECLARE_PER_CPU(struct rcu_data, rcu_preempt_data);
  231. #endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */
  232. #endif /* #ifdef RCU_TREE_NONCORE */