2 * Read-Copy Update mechanism for mutual exclusion (tree-based version)
3 * Internal non-public definitions.
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.
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.
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.
19 * Copyright IBM Corporation, 2008
21 * Author: Ingo Molnar <mingo@elte.hu>
22 * Paul E. McKenney <paulmck@linux.vnet.ibm.com>
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>
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.
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)
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)
77 # error "CONFIG_RCU_FANOUT insufficient for NR_CPUS"
78 #endif /* #if (NR_CPUS) <= RCU_FANOUT_1 */
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)
84 * Dynticks per-CPU state.
87 int dynticks_nesting; /* Track irq/process nesting level. */
88 int dynticks_nmi_nesting; /* Track NMI nesting level. */
89 atomic_t dynticks; /* Even value for dynticks-idle, else odd. */
93 * Definition for node within the RCU grace-period-detection hierarchy.
96 raw_spinlock_t lock; /* Root rcu_node's lock protects some */
97 /* rcu_state fields as well as following. */
98 unsigned long gpnum; /* Current grace period for this node. */
99 /* This will either be equal to or one */
100 /* behind the root rcu_node's gpnum. */
101 unsigned long completed; /* Last GP completed for this node. */
102 /* This will either be equal to or one */
103 /* behind the root rcu_node's gpnum. */
104 unsigned long qsmask; /* CPUs or groups that need to switch in */
105 /* order for current grace period to proceed.*/
106 /* In leaf rcu_node, each bit corresponds to */
107 /* an rcu_data structure, otherwise, each */
108 /* bit corresponds to a child rcu_node */
110 unsigned long expmask; /* Groups that have ->blkd_tasks */
111 /* elements that need to drain to allow the */
112 /* current expedited grace period to */
113 /* complete (only for TREE_PREEMPT_RCU). */
114 unsigned long wakemask; /* CPUs whose kthread needs to be awakened. */
115 unsigned long qsmaskinit;
116 /* Per-GP initial value for qsmask & expmask. */
117 unsigned long grpmask; /* Mask to apply to parent qsmask. */
118 /* Only one bit will be set in this mask. */
119 int grplo; /* lowest-numbered CPU or group here. */
120 int grphi; /* highest-numbered CPU or group here. */
121 u8 grpnum; /* CPU/group number for next level up. */
122 u8 level; /* root is at level 0. */
123 struct rcu_node *parent;
124 struct list_head blkd_tasks;
125 /* Tasks blocked in RCU read-side critical */
126 /* section. Tasks are placed at the head */
127 /* of this list and age towards the tail. */
128 struct list_head *gp_tasks;
129 /* Pointer to the first task blocking the */
130 /* current grace period, or NULL if there */
131 /* is no such task. */
132 struct list_head *exp_tasks;
133 /* Pointer to the first task blocking the */
134 /* current expedited grace period, or NULL */
135 /* if there is no such task. If there */
136 /* is no current expedited grace period, */
137 /* then there can cannot be any such task. */
138 struct task_struct *node_kthread_task;
139 /* kthread that takes care of this rcu_node */
140 /* structure, for example, awakening the */
141 /* per-CPU kthreads as needed. */
142 wait_queue_head_t node_wq;
143 /* Wait queue on which to park the per-node */
145 } ____cacheline_internodealigned_in_smp;
148 * Do a full breadth-first scan of the rcu_node structures for the
149 * specified rcu_state structure.
151 #define rcu_for_each_node_breadth_first(rsp, rnp) \
152 for ((rnp) = &(rsp)->node[0]; \
153 (rnp) < &(rsp)->node[NUM_RCU_NODES]; (rnp)++)
156 * Do a breadth-first scan of the non-leaf rcu_node structures for the
157 * specified rcu_state structure. Note that if there is a singleton
158 * rcu_node tree with but one rcu_node structure, this loop is a no-op.
160 #define rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) \
161 for ((rnp) = &(rsp)->node[0]; \
162 (rnp) < (rsp)->level[NUM_RCU_LVLS - 1]; (rnp)++)
165 * Scan the leaves of the rcu_node hierarchy for the specified rcu_state
166 * structure. Note that if there is a singleton rcu_node tree with but
167 * one rcu_node structure, this loop -will- visit the rcu_node structure.
168 * It is still a leaf node, even if it is also the root node.
170 #define rcu_for_each_leaf_node(rsp, rnp) \
171 for ((rnp) = (rsp)->level[NUM_RCU_LVLS - 1]; \
172 (rnp) < &(rsp)->node[NUM_RCU_NODES]; (rnp)++)
174 /* Index values for nxttail array in struct rcu_data. */
175 #define RCU_DONE_TAIL 0 /* Also RCU_WAIT head. */
176 #define RCU_WAIT_TAIL 1 /* Also RCU_NEXT_READY head. */
177 #define RCU_NEXT_READY_TAIL 2 /* Also RCU_NEXT head. */
178 #define RCU_NEXT_TAIL 3
179 #define RCU_NEXT_SIZE 4
181 /* Per-CPU data for read-copy update. */
183 /* 1) quiescent-state and grace-period handling : */
184 unsigned long completed; /* Track rsp->completed gp number */
185 /* in order to detect GP end. */
186 unsigned long gpnum; /* Highest gp number that this CPU */
187 /* is aware of having started. */
188 unsigned long passed_quiesc_completed;
189 /* Value of completed at time of qs. */
190 bool passed_quiesc; /* User-mode/idle loop etc. */
191 bool qs_pending; /* Core waits for quiesc state. */
192 bool beenonline; /* CPU online at least once. */
193 bool preemptable; /* Preemptable RCU? */
194 struct rcu_node *mynode; /* This CPU's leaf of hierarchy */
195 unsigned long grpmask; /* Mask to apply to leaf qsmask. */
197 /* 2) batch handling */
199 * If nxtlist is not NULL, it is partitioned as follows.
200 * Any of the partitions might be empty, in which case the
201 * pointer to that partition will be equal to the pointer for
202 * the following partition. When the list is empty, all of
203 * the nxttail elements point to the ->nxtlist pointer itself,
204 * which in that case is NULL.
206 * [nxtlist, *nxttail[RCU_DONE_TAIL]):
207 * Entries that batch # <= ->completed
208 * The grace period for these entries has completed, and
209 * the other grace-period-completed entries may be moved
210 * here temporarily in rcu_process_callbacks().
211 * [*nxttail[RCU_DONE_TAIL], *nxttail[RCU_WAIT_TAIL]):
212 * Entries that batch # <= ->completed - 1: waiting for current GP
213 * [*nxttail[RCU_WAIT_TAIL], *nxttail[RCU_NEXT_READY_TAIL]):
214 * Entries known to have arrived before current GP ended
215 * [*nxttail[RCU_NEXT_READY_TAIL], *nxttail[RCU_NEXT_TAIL]):
216 * Entries that might have arrived after current GP ended
217 * Note that the value of *nxttail[RCU_NEXT_TAIL] will
218 * always be NULL, as this is the end of the list.
220 struct rcu_head *nxtlist;
221 struct rcu_head **nxttail[RCU_NEXT_SIZE];
222 long qlen; /* # of queued callbacks */
223 long qlen_last_fqs_check;
224 /* qlen at last check for QS forcing */
225 unsigned long n_cbs_invoked; /* count of RCU cbs invoked. */
226 unsigned long n_cbs_orphaned; /* RCU cbs orphaned by dying CPU */
227 unsigned long n_cbs_adopted; /* RCU cbs adopted from dying CPU */
228 unsigned long n_force_qs_snap;
229 /* did other CPU force QS recently? */
230 long blimit; /* Upper limit on a processed batch */
233 /* 3) dynticks interface. */
234 struct rcu_dynticks *dynticks; /* Shared per-CPU dynticks state. */
235 int dynticks_snap; /* Per-GP tracking for dynticks. */
236 #endif /* #ifdef CONFIG_NO_HZ */
238 /* 4) reasons this CPU needed to be kicked by force_quiescent_state */
240 unsigned long dynticks_fqs; /* Kicked due to dynticks idle. */
241 #endif /* #ifdef CONFIG_NO_HZ */
242 unsigned long offline_fqs; /* Kicked due to being offline. */
243 unsigned long resched_ipi; /* Sent a resched IPI. */
245 /* 5) __rcu_pending() statistics. */
246 unsigned long n_rcu_pending; /* rcu_pending() calls since boot. */
247 unsigned long n_rp_qs_pending;
248 unsigned long n_rp_report_qs;
249 unsigned long n_rp_cb_ready;
250 unsigned long n_rp_cpu_needs_gp;
251 unsigned long n_rp_gp_completed;
252 unsigned long n_rp_gp_started;
253 unsigned long n_rp_need_fqs;
254 unsigned long n_rp_need_nothing;
259 /* Values for signaled field in struct rcu_state. */
260 #define RCU_GP_IDLE 0 /* No grace period in progress. */
261 #define RCU_GP_INIT 1 /* Grace period being initialized. */
262 #define RCU_SAVE_DYNTICK 2 /* Need to scan dyntick state. */
263 #define RCU_FORCE_QS 3 /* Need to force quiescent state. */
265 #define RCU_SIGNAL_INIT RCU_SAVE_DYNTICK
266 #else /* #ifdef CONFIG_NO_HZ */
267 #define RCU_SIGNAL_INIT RCU_FORCE_QS
268 #endif /* #else #ifdef CONFIG_NO_HZ */
270 #define RCU_JIFFIES_TILL_FORCE_QS 3 /* for rsp->jiffies_force_qs */
272 #ifdef CONFIG_PROVE_RCU
273 #define RCU_STALL_DELAY_DELTA (5 * HZ)
275 #define RCU_STALL_DELAY_DELTA 0
278 #define RCU_SECONDS_TILL_STALL_CHECK (CONFIG_RCU_CPU_STALL_TIMEOUT * HZ + \
279 RCU_STALL_DELAY_DELTA)
280 /* for rsp->jiffies_stall */
281 #define RCU_SECONDS_TILL_STALL_RECHECK (3 * RCU_SECONDS_TILL_STALL_CHECK + 30)
282 /* for rsp->jiffies_stall */
283 #define RCU_STALL_RAT_DELAY 2 /* Allow other CPUs time */
284 /* to take at least one */
285 /* scheduling clock irq */
286 /* before ratting on them. */
290 * RCU global state, including node hierarchy. This hierarchy is
291 * represented in "heap" form in a dense array. The root (first level)
292 * of the hierarchy is in ->node[0] (referenced by ->level[0]), the second
293 * level in ->node[1] through ->node[m] (->node[1] referenced by ->level[1]),
294 * and the third level in ->node[m+1] and following (->node[m+1] referenced
295 * by ->level[2]). The number of levels is determined by the number of
296 * CPUs and by CONFIG_RCU_FANOUT. Small systems will have a "hierarchy"
297 * consisting of a single rcu_node.
300 struct rcu_node node[NUM_RCU_NODES]; /* Hierarchy. */
301 struct rcu_node *level[NUM_RCU_LVLS]; /* Hierarchy levels. */
302 u32 levelcnt[MAX_RCU_LVLS + 1]; /* # nodes in each level. */
303 u8 levelspread[NUM_RCU_LVLS]; /* kids/node in each level. */
304 struct rcu_data __percpu *rda; /* pointer of percu rcu_data. */
306 /* The following fields are guarded by the root rcu_node's lock. */
308 u8 signaled ____cacheline_internodealigned_in_smp;
309 /* Force QS state. */
310 u8 fqs_active; /* force_quiescent_state() */
312 u8 fqs_need_gp; /* A CPU was prevented from */
313 /* starting a new grace */
315 /* force_quiescent_state() */
317 unsigned long gpnum; /* Current gp number. */
318 unsigned long completed; /* # of last completed gp. */
320 /* End of fields guarded by root rcu_node's lock. */
322 raw_spinlock_t onofflock; /* exclude on/offline and */
323 /* starting new GP. */
324 raw_spinlock_t fqslock; /* Only one task forcing */
325 /* quiescent states. */
326 unsigned long jiffies_force_qs; /* Time at which to invoke */
327 /* force_quiescent_state(). */
328 unsigned long n_force_qs; /* Number of calls to */
329 /* force_quiescent_state(). */
330 unsigned long n_force_qs_lh; /* ~Number of calls leaving */
331 /* due to lock unavailable. */
332 unsigned long n_force_qs_ngp; /* Number of calls leaving */
333 /* due to no GP active. */
334 unsigned long gp_start; /* Time at which GP started, */
335 /* but in jiffies. */
336 unsigned long jiffies_stall; /* Time at which to check */
337 /* for CPU stalls. */
338 char *name; /* Name of structure. */
341 /* Return values for rcu_preempt_offline_tasks(). */
343 #define RCU_OFL_TASKS_NORM_GP 0x1 /* Tasks blocking normal */
344 /* GP were moved to root. */
345 #define RCU_OFL_TASKS_EXP_GP 0x2 /* Tasks blocking expedited */
346 /* GP were moved to root. */
349 * RCU implementation internal declarations:
351 extern struct rcu_state rcu_sched_state;
352 DECLARE_PER_CPU(struct rcu_data, rcu_sched_data);
354 extern struct rcu_state rcu_bh_state;
355 DECLARE_PER_CPU(struct rcu_data, rcu_bh_data);
357 #ifdef CONFIG_TREE_PREEMPT_RCU
358 extern struct rcu_state rcu_preempt_state;
359 DECLARE_PER_CPU(struct rcu_data, rcu_preempt_data);
360 #endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */
362 #ifndef RCU_TREE_NONCORE
364 /* Forward declarations for rcutree_plugin.h */
365 static void rcu_bootup_announce(void);
366 long rcu_batches_completed(void);
367 static void rcu_preempt_note_context_switch(int cpu);
368 static int rcu_preempted_readers(struct rcu_node *rnp);
369 #ifdef CONFIG_HOTPLUG_CPU
370 static void rcu_report_unblock_qs_rnp(struct rcu_node *rnp,
371 unsigned long flags);
372 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
373 static void rcu_print_detail_task_stall(struct rcu_state *rsp);
374 static void rcu_print_task_stall(struct rcu_node *rnp);
375 static void rcu_preempt_stall_reset(void);
376 static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp);
377 #ifdef CONFIG_HOTPLUG_CPU
378 static int rcu_preempt_offline_tasks(struct rcu_state *rsp,
379 struct rcu_node *rnp,
380 struct rcu_data *rdp);
381 static void rcu_preempt_offline_cpu(int cpu);
382 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
383 static void rcu_preempt_check_callbacks(int cpu);
384 static void rcu_preempt_process_callbacks(void);
385 void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu));
386 #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_TREE_PREEMPT_RCU)
387 static void rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp);
388 #endif /* #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_TREE_PREEMPT_RCU) */
389 static int rcu_preempt_pending(int cpu);
390 static int rcu_preempt_needs_cpu(int cpu);
391 static void __cpuinit rcu_preempt_init_percpu_data(int cpu);
392 static void rcu_preempt_send_cbs_to_online(void);
393 static void __init __rcu_init_preempt(void);
394 static void rcu_needs_cpu_flush(void);
396 #endif /* #ifndef RCU_TREE_NONCORE */