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 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. */
94 /* RCU's kthread states for tracing. */
95 #define RCU_KTHREAD_STOPPED 0
96 #define RCU_KTHREAD_RUNNING 1
97 #define RCU_KTHREAD_WAITING 2
98 #define RCU_KTHREAD_OFFCPU 3
99 #define RCU_KTHREAD_YIELDING 4
100 #define RCU_KTHREAD_MAX 4
103 * Definition for node within the RCU grace-period-detection hierarchy.
106 raw_spinlock_t lock
; /* Root rcu_node's lock protects some */
107 /* rcu_state fields as well as following. */
108 unsigned long gpnum
; /* Current grace period for this node. */
109 /* This will either be equal to or one */
110 /* behind the root rcu_node's gpnum. */
111 unsigned long completed
; /* Last GP completed for this node. */
112 /* This will either be equal to or one */
113 /* behind the root rcu_node's gpnum. */
114 unsigned long qsmask
; /* CPUs or groups that need to switch in */
115 /* order for current grace period to proceed.*/
116 /* In leaf rcu_node, each bit corresponds to */
117 /* an rcu_data structure, otherwise, each */
118 /* bit corresponds to a child rcu_node */
120 unsigned long expmask
; /* Groups that have ->blkd_tasks */
121 /* elements that need to drain to allow the */
122 /* current expedited grace period to */
123 /* complete (only for TREE_PREEMPT_RCU). */
124 unsigned long wakemask
; /* CPUs whose kthread needs to be awakened. */
125 unsigned long qsmaskinit
;
126 /* Per-GP initial value for qsmask & expmask. */
127 unsigned long grpmask
; /* Mask to apply to parent qsmask. */
128 /* Only one bit will be set in this mask. */
129 int grplo
; /* lowest-numbered CPU or group here. */
130 int grphi
; /* highest-numbered CPU or group here. */
131 u8 grpnum
; /* CPU/group number for next level up. */
132 u8 level
; /* root is at level 0. */
133 struct rcu_node
*parent
;
134 struct list_head blkd_tasks
;
135 /* Tasks blocked in RCU read-side critical */
136 /* section. Tasks are placed at the head */
137 /* of this list and age towards the tail. */
138 struct list_head
*gp_tasks
;
139 /* Pointer to the first task blocking the */
140 /* current grace period, or NULL if there */
141 /* is no such task. */
142 struct list_head
*exp_tasks
;
143 /* Pointer to the first task blocking the */
144 /* current expedited grace period, or NULL */
145 /* if there is no such task. If there */
146 /* is no current expedited grace period, */
147 /* then there can cannot be any such task. */
148 #ifdef CONFIG_RCU_BOOST
149 struct list_head
*boost_tasks
;
150 /* Pointer to first task that needs to be */
151 /* priority boosted, or NULL if no priority */
152 /* boosting is needed for this rcu_node */
153 /* structure. If there are no tasks */
154 /* queued on this rcu_node structure that */
155 /* are blocking the current grace period, */
156 /* there can be no such task. */
157 unsigned long boost_time
;
158 /* When to start boosting (jiffies). */
159 struct task_struct
*boost_kthread_task
;
160 /* kthread that takes care of priority */
161 /* boosting for this rcu_node structure. */
162 wait_queue_head_t boost_wq
;
163 /* Wait queue on which to park the boost */
165 unsigned int boost_kthread_status
;
166 /* State of boost_kthread_task for tracing. */
167 unsigned long n_tasks_boosted
;
168 /* Total number of tasks boosted. */
169 unsigned long n_exp_boosts
;
170 /* Number of tasks boosted for expedited GP. */
171 unsigned long n_normal_boosts
;
172 /* Number of tasks boosted for normal GP. */
173 unsigned long n_balk_blkd_tasks
;
174 /* Refused to boost: no blocked tasks. */
175 unsigned long n_balk_exp_gp_tasks
;
176 /* Refused to boost: nothing blocking GP. */
177 unsigned long n_balk_boost_tasks
;
178 /* Refused to boost: already boosting. */
179 unsigned long n_balk_notblocked
;
180 /* Refused to boost: RCU RS CS still running. */
181 unsigned long n_balk_notyet
;
182 /* Refused to boost: not yet time. */
183 unsigned long n_balk_nos
;
184 /* Refused to boost: not sure why, though. */
185 /* This can happen due to race conditions. */
186 #endif /* #ifdef CONFIG_RCU_BOOST */
187 struct task_struct
*node_kthread_task
;
188 /* kthread that takes care of this rcu_node */
189 /* structure, for example, awakening the */
190 /* per-CPU kthreads as needed. */
191 wait_queue_head_t node_wq
;
192 /* Wait queue on which to park the per-node */
194 unsigned int node_kthread_status
;
195 /* State of node_kthread_task for tracing. */
196 } ____cacheline_internodealigned_in_smp
;
199 * Do a full breadth-first scan of the rcu_node structures for the
200 * specified rcu_state structure.
202 #define rcu_for_each_node_breadth_first(rsp, rnp) \
203 for ((rnp) = &(rsp)->node[0]; \
204 (rnp) < &(rsp)->node[NUM_RCU_NODES]; (rnp)++)
207 * Do a breadth-first scan of the non-leaf rcu_node structures for the
208 * specified rcu_state structure. Note that if there is a singleton
209 * rcu_node tree with but one rcu_node structure, this loop is a no-op.
211 #define rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) \
212 for ((rnp) = &(rsp)->node[0]; \
213 (rnp) < (rsp)->level[NUM_RCU_LVLS - 1]; (rnp)++)
216 * Scan the leaves of the rcu_node hierarchy for the specified rcu_state
217 * structure. Note that if there is a singleton rcu_node tree with but
218 * one rcu_node structure, this loop -will- visit the rcu_node structure.
219 * It is still a leaf node, even if it is also the root node.
221 #define rcu_for_each_leaf_node(rsp, rnp) \
222 for ((rnp) = (rsp)->level[NUM_RCU_LVLS - 1]; \
223 (rnp) < &(rsp)->node[NUM_RCU_NODES]; (rnp)++)
225 /* Index values for nxttail array in struct rcu_data. */
226 #define RCU_DONE_TAIL 0 /* Also RCU_WAIT head. */
227 #define RCU_WAIT_TAIL 1 /* Also RCU_NEXT_READY head. */
228 #define RCU_NEXT_READY_TAIL 2 /* Also RCU_NEXT head. */
229 #define RCU_NEXT_TAIL 3
230 #define RCU_NEXT_SIZE 4
232 /* Per-CPU data for read-copy update. */
234 /* 1) quiescent-state and grace-period handling : */
235 unsigned long completed
; /* Track rsp->completed gp number */
236 /* in order to detect GP end. */
237 unsigned long gpnum
; /* Highest gp number that this CPU */
238 /* is aware of having started. */
239 unsigned long passed_quiesc_completed
;
240 /* Value of completed at time of qs. */
241 bool passed_quiesc
; /* User-mode/idle loop etc. */
242 bool qs_pending
; /* Core waits for quiesc state. */
243 bool beenonline
; /* CPU online at least once. */
244 bool preemptible
; /* Preemptible RCU? */
245 struct rcu_node
*mynode
; /* This CPU's leaf of hierarchy */
246 unsigned long grpmask
; /* Mask to apply to leaf qsmask. */
248 /* 2) batch handling */
250 * If nxtlist is not NULL, it is partitioned as follows.
251 * Any of the partitions might be empty, in which case the
252 * pointer to that partition will be equal to the pointer for
253 * the following partition. When the list is empty, all of
254 * the nxttail elements point to the ->nxtlist pointer itself,
255 * which in that case is NULL.
257 * [nxtlist, *nxttail[RCU_DONE_TAIL]):
258 * Entries that batch # <= ->completed
259 * The grace period for these entries has completed, and
260 * the other grace-period-completed entries may be moved
261 * here temporarily in rcu_process_callbacks().
262 * [*nxttail[RCU_DONE_TAIL], *nxttail[RCU_WAIT_TAIL]):
263 * Entries that batch # <= ->completed - 1: waiting for current GP
264 * [*nxttail[RCU_WAIT_TAIL], *nxttail[RCU_NEXT_READY_TAIL]):
265 * Entries known to have arrived before current GP ended
266 * [*nxttail[RCU_NEXT_READY_TAIL], *nxttail[RCU_NEXT_TAIL]):
267 * Entries that might have arrived after current GP ended
268 * Note that the value of *nxttail[RCU_NEXT_TAIL] will
269 * always be NULL, as this is the end of the list.
271 struct rcu_head
*nxtlist
;
272 struct rcu_head
**nxttail
[RCU_NEXT_SIZE
];
273 long qlen
; /* # of queued callbacks */
274 long qlen_last_fqs_check
;
275 /* qlen at last check for QS forcing */
276 unsigned long n_cbs_invoked
; /* count of RCU cbs invoked. */
277 unsigned long n_cbs_orphaned
; /* RCU cbs orphaned by dying CPU */
278 unsigned long n_cbs_adopted
; /* RCU cbs adopted from dying CPU */
279 unsigned long n_force_qs_snap
;
280 /* did other CPU force QS recently? */
281 long blimit
; /* Upper limit on a processed batch */
284 /* 3) dynticks interface. */
285 struct rcu_dynticks
*dynticks
; /* Shared per-CPU dynticks state. */
286 int dynticks_snap
; /* Per-GP tracking for dynticks. */
287 int dynticks_nmi_snap
; /* Per-GP tracking for dynticks_nmi. */
288 #endif /* #ifdef CONFIG_NO_HZ */
290 /* 4) reasons this CPU needed to be kicked by force_quiescent_state */
292 unsigned long dynticks_fqs
; /* Kicked due to dynticks idle. */
293 #endif /* #ifdef CONFIG_NO_HZ */
294 unsigned long offline_fqs
; /* Kicked due to being offline. */
295 unsigned long resched_ipi
; /* Sent a resched IPI. */
297 /* 5) __rcu_pending() statistics. */
298 unsigned long n_rcu_pending
; /* rcu_pending() calls since boot. */
299 unsigned long n_rp_qs_pending
;
300 unsigned long n_rp_report_qs
;
301 unsigned long n_rp_cb_ready
;
302 unsigned long n_rp_cpu_needs_gp
;
303 unsigned long n_rp_gp_completed
;
304 unsigned long n_rp_gp_started
;
305 unsigned long n_rp_need_fqs
;
306 unsigned long n_rp_need_nothing
;
311 /* Values for signaled field in struct rcu_state. */
312 #define RCU_GP_IDLE 0 /* No grace period in progress. */
313 #define RCU_GP_INIT 1 /* Grace period being initialized. */
314 #define RCU_SAVE_DYNTICK 2 /* Need to scan dyntick state. */
315 #define RCU_FORCE_QS 3 /* Need to force quiescent state. */
317 #define RCU_SIGNAL_INIT RCU_SAVE_DYNTICK
318 #else /* #ifdef CONFIG_NO_HZ */
319 #define RCU_SIGNAL_INIT RCU_FORCE_QS
320 #endif /* #else #ifdef CONFIG_NO_HZ */
322 #define RCU_JIFFIES_TILL_FORCE_QS 3 /* for rsp->jiffies_force_qs */
324 #ifdef CONFIG_PROVE_RCU
325 #define RCU_STALL_DELAY_DELTA (5 * HZ)
327 #define RCU_STALL_DELAY_DELTA 0
330 #define RCU_SECONDS_TILL_STALL_CHECK (CONFIG_RCU_CPU_STALL_TIMEOUT * HZ + \
331 RCU_STALL_DELAY_DELTA)
332 /* for rsp->jiffies_stall */
333 #define RCU_SECONDS_TILL_STALL_RECHECK (3 * RCU_SECONDS_TILL_STALL_CHECK + 30)
334 /* for rsp->jiffies_stall */
335 #define RCU_STALL_RAT_DELAY 2 /* Allow other CPUs time */
336 /* to take at least one */
337 /* scheduling clock irq */
338 /* before ratting on them. */
342 * RCU global state, including node hierarchy. This hierarchy is
343 * represented in "heap" form in a dense array. The root (first level)
344 * of the hierarchy is in ->node[0] (referenced by ->level[0]), the second
345 * level in ->node[1] through ->node[m] (->node[1] referenced by ->level[1]),
346 * and the third level in ->node[m+1] and following (->node[m+1] referenced
347 * by ->level[2]). The number of levels is determined by the number of
348 * CPUs and by CONFIG_RCU_FANOUT. Small systems will have a "hierarchy"
349 * consisting of a single rcu_node.
352 struct rcu_node node
[NUM_RCU_NODES
]; /* Hierarchy. */
353 struct rcu_node
*level
[NUM_RCU_LVLS
]; /* Hierarchy levels. */
354 u32 levelcnt
[MAX_RCU_LVLS
+ 1]; /* # nodes in each level. */
355 u8 levelspread
[NUM_RCU_LVLS
]; /* kids/node in each level. */
356 struct rcu_data __percpu
*rda
; /* pointer of percu rcu_data. */
358 /* The following fields are guarded by the root rcu_node's lock. */
360 u8 signaled ____cacheline_internodealigned_in_smp
;
361 /* Force QS state. */
362 u8 fqs_active
; /* force_quiescent_state() */
364 u8 fqs_need_gp
; /* A CPU was prevented from */
365 /* starting a new grace */
367 /* force_quiescent_state() */
369 unsigned long gpnum
; /* Current gp number. */
370 unsigned long completed
; /* # of last completed gp. */
372 /* End of fields guarded by root rcu_node's lock. */
374 raw_spinlock_t onofflock
; /* exclude on/offline and */
375 /* starting new GP. */
376 raw_spinlock_t fqslock
; /* Only one task forcing */
377 /* quiescent states. */
378 unsigned long jiffies_force_qs
; /* Time at which to invoke */
379 /* force_quiescent_state(). */
380 unsigned long n_force_qs
; /* Number of calls to */
381 /* force_quiescent_state(). */
382 unsigned long n_force_qs_lh
; /* ~Number of calls leaving */
383 /* due to lock unavailable. */
384 unsigned long n_force_qs_ngp
; /* Number of calls leaving */
385 /* due to no GP active. */
386 unsigned long gp_start
; /* Time at which GP started, */
387 /* but in jiffies. */
388 unsigned long jiffies_stall
; /* Time at which to check */
389 /* for CPU stalls. */
390 unsigned long gp_max
; /* Maximum GP duration in */
392 char *name
; /* Name of structure. */
395 /* Return values for rcu_preempt_offline_tasks(). */
397 #define RCU_OFL_TASKS_NORM_GP 0x1 /* Tasks blocking normal */
398 /* GP were moved to root. */
399 #define RCU_OFL_TASKS_EXP_GP 0x2 /* Tasks blocking expedited */
400 /* GP were moved to root. */
403 * RCU implementation internal declarations:
405 extern struct rcu_state rcu_sched_state
;
406 DECLARE_PER_CPU(struct rcu_data
, rcu_sched_data
);
408 extern struct rcu_state rcu_bh_state
;
409 DECLARE_PER_CPU(struct rcu_data
, rcu_bh_data
);
411 #ifdef CONFIG_TREE_PREEMPT_RCU
412 extern struct rcu_state rcu_preempt_state
;
413 DECLARE_PER_CPU(struct rcu_data
, rcu_preempt_data
);
414 #endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */
416 #ifndef RCU_TREE_NONCORE
418 /* Forward declarations for rcutree_plugin.h */
419 static void rcu_bootup_announce(void);
420 long rcu_batches_completed(void);
421 static void rcu_preempt_note_context_switch(int cpu
);
422 static int rcu_preempt_blocked_readers_cgp(struct rcu_node
*rnp
);
423 #ifdef CONFIG_HOTPLUG_CPU
424 static void rcu_report_unblock_qs_rnp(struct rcu_node
*rnp
,
425 unsigned long flags
);
426 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
427 static void rcu_print_detail_task_stall(struct rcu_state
*rsp
);
428 static void rcu_print_task_stall(struct rcu_node
*rnp
);
429 static void rcu_preempt_stall_reset(void);
430 static void rcu_preempt_check_blocked_tasks(struct rcu_node
*rnp
);
431 #ifdef CONFIG_HOTPLUG_CPU
432 static int rcu_preempt_offline_tasks(struct rcu_state
*rsp
,
433 struct rcu_node
*rnp
,
434 struct rcu_data
*rdp
);
435 static void rcu_preempt_offline_cpu(int cpu
);
436 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
437 static void rcu_preempt_check_callbacks(int cpu
);
438 static void rcu_preempt_process_callbacks(void);
439 void call_rcu(struct rcu_head
*head
, void (*func
)(struct rcu_head
*rcu
));
440 #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_TREE_PREEMPT_RCU)
441 static void rcu_report_exp_rnp(struct rcu_state
*rsp
, struct rcu_node
*rnp
);
442 #endif /* #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_TREE_PREEMPT_RCU) */
443 static int rcu_preempt_pending(int cpu
);
444 static int rcu_preempt_needs_cpu(int cpu
);
445 static void __cpuinit
rcu_preempt_init_percpu_data(int cpu
);
446 static void rcu_preempt_send_cbs_to_online(void);
447 static void __init
__rcu_init_preempt(void);
448 static void rcu_needs_cpu_flush(void);
449 static void __init
rcu_init_boost_waitqueue(struct rcu_node
*rnp
);
450 static void rcu_initiate_boost(struct rcu_node
*rnp
, unsigned long flags
);
451 static void rcu_boost_kthread_setaffinity(struct rcu_node
*rnp
,
453 static void rcu_preempt_boost_start_gp(struct rcu_node
*rnp
);
454 static int __cpuinit
rcu_spawn_one_boost_kthread(struct rcu_state
*rsp
,
455 struct rcu_node
*rnp
,
458 #endif /* #ifndef RCU_TREE_NONCORE */