ea32405177c92544402bc296e6541141f707987c
[deliverable/linux.git] / kernel / rcutree.h
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 */
86 struct rcu_dynticks {
87 long long dynticks_nesting; /* Track irq/process nesting level. */
88 /* Process level is worth LLONG_MAX/2. */
89 int dynticks_nmi_nesting; /* Track NMI nesting level. */
90 atomic_t dynticks; /* Even value for idle, else odd. */
91 int wake_gp_end; /* A GP ended, need to wake up CPUs. */
92 };
93
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
101
102 /*
103 * Definition for node within the RCU grace-period-detection hierarchy.
104 */
105 struct rcu_node {
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 */
119 /* structure. */
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 atomic_t wakemask; /* CPUs whose kthread needs to be awakened. */
125 /* Since this has meaning only for leaf */
126 /* rcu_node structures, 32 bits suffices. */
127 unsigned long qsmaskinit;
128 /* Per-GP initial value for qsmask & expmask. */
129 unsigned long grpmask; /* Mask to apply to parent qsmask. */
130 /* Only one bit will be set in this mask. */
131 int grplo; /* lowest-numbered CPU or group here. */
132 int grphi; /* highest-numbered CPU or group here. */
133 u8 grpnum; /* CPU/group number for next level up. */
134 u8 level; /* root is at level 0. */
135 struct rcu_node *parent;
136 struct list_head blkd_tasks;
137 /* Tasks blocked in RCU read-side critical */
138 /* section. Tasks are placed at the head */
139 /* of this list and age towards the tail. */
140 struct list_head *gp_tasks;
141 /* Pointer to the first task blocking the */
142 /* current grace period, or NULL if there */
143 /* is no such task. */
144 struct list_head *exp_tasks;
145 /* Pointer to the first task blocking the */
146 /* current expedited grace period, or NULL */
147 /* if there is no such task. If there */
148 /* is no current expedited grace period, */
149 /* then there can cannot be any such task. */
150 #ifdef CONFIG_RCU_BOOST
151 struct list_head *boost_tasks;
152 /* Pointer to first task that needs to be */
153 /* priority boosted, or NULL if no priority */
154 /* boosting is needed for this rcu_node */
155 /* structure. If there are no tasks */
156 /* queued on this rcu_node structure that */
157 /* are blocking the current grace period, */
158 /* there can be no such task. */
159 unsigned long boost_time;
160 /* When to start boosting (jiffies). */
161 struct task_struct *boost_kthread_task;
162 /* kthread that takes care of priority */
163 /* boosting for this rcu_node structure. */
164 unsigned int boost_kthread_status;
165 /* State of boost_kthread_task for tracing. */
166 unsigned long n_tasks_boosted;
167 /* Total number of tasks boosted. */
168 unsigned long n_exp_boosts;
169 /* Number of tasks boosted for expedited GP. */
170 unsigned long n_normal_boosts;
171 /* Number of tasks boosted for normal GP. */
172 unsigned long n_balk_blkd_tasks;
173 /* Refused to boost: no blocked tasks. */
174 unsigned long n_balk_exp_gp_tasks;
175 /* Refused to boost: nothing blocking GP. */
176 unsigned long n_balk_boost_tasks;
177 /* Refused to boost: already boosting. */
178 unsigned long n_balk_notblocked;
179 /* Refused to boost: RCU RS CS still running. */
180 unsigned long n_balk_notyet;
181 /* Refused to boost: not yet time. */
182 unsigned long n_balk_nos;
183 /* Refused to boost: not sure why, though. */
184 /* This can happen due to race conditions. */
185 #endif /* #ifdef CONFIG_RCU_BOOST */
186 struct task_struct *node_kthread_task;
187 /* kthread that takes care of this rcu_node */
188 /* structure, for example, awakening the */
189 /* per-CPU kthreads as needed. */
190 unsigned int node_kthread_status;
191 /* State of node_kthread_task for tracing. */
192 } ____cacheline_internodealigned_in_smp;
193
194 /*
195 * Do a full breadth-first scan of the rcu_node structures for the
196 * specified rcu_state structure.
197 */
198 #define rcu_for_each_node_breadth_first(rsp, rnp) \
199 for ((rnp) = &(rsp)->node[0]; \
200 (rnp) < &(rsp)->node[NUM_RCU_NODES]; (rnp)++)
201
202 /*
203 * Do a breadth-first scan of the non-leaf rcu_node structures for the
204 * specified rcu_state structure. Note that if there is a singleton
205 * rcu_node tree with but one rcu_node structure, this loop is a no-op.
206 */
207 #define rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) \
208 for ((rnp) = &(rsp)->node[0]; \
209 (rnp) < (rsp)->level[NUM_RCU_LVLS - 1]; (rnp)++)
210
211 /*
212 * Scan the leaves of the rcu_node hierarchy for the specified rcu_state
213 * structure. Note that if there is a singleton rcu_node tree with but
214 * one rcu_node structure, this loop -will- visit the rcu_node structure.
215 * It is still a leaf node, even if it is also the root node.
216 */
217 #define rcu_for_each_leaf_node(rsp, rnp) \
218 for ((rnp) = (rsp)->level[NUM_RCU_LVLS - 1]; \
219 (rnp) < &(rsp)->node[NUM_RCU_NODES]; (rnp)++)
220
221 /* Index values for nxttail array in struct rcu_data. */
222 #define RCU_DONE_TAIL 0 /* Also RCU_WAIT head. */
223 #define RCU_WAIT_TAIL 1 /* Also RCU_NEXT_READY head. */
224 #define RCU_NEXT_READY_TAIL 2 /* Also RCU_NEXT head. */
225 #define RCU_NEXT_TAIL 3
226 #define RCU_NEXT_SIZE 4
227
228 /* Per-CPU data for read-copy update. */
229 struct rcu_data {
230 /* 1) quiescent-state and grace-period handling : */
231 unsigned long completed; /* Track rsp->completed gp number */
232 /* in order to detect GP end. */
233 unsigned long gpnum; /* Highest gp number that this CPU */
234 /* is aware of having started. */
235 unsigned long passed_quiesce_gpnum;
236 /* gpnum at time of quiescent state. */
237 bool passed_quiesce; /* User-mode/idle loop etc. */
238 bool qs_pending; /* Core waits for quiesc state. */
239 bool beenonline; /* CPU online at least once. */
240 bool preemptible; /* Preemptible RCU? */
241 struct rcu_node *mynode; /* This CPU's leaf of hierarchy */
242 unsigned long grpmask; /* Mask to apply to leaf qsmask. */
243
244 /* 2) batch handling */
245 /*
246 * If nxtlist is not NULL, it is partitioned as follows.
247 * Any of the partitions might be empty, in which case the
248 * pointer to that partition will be equal to the pointer for
249 * the following partition. When the list is empty, all of
250 * the nxttail elements point to the ->nxtlist pointer itself,
251 * which in that case is NULL.
252 *
253 * [nxtlist, *nxttail[RCU_DONE_TAIL]):
254 * Entries that batch # <= ->completed
255 * The grace period for these entries has completed, and
256 * the other grace-period-completed entries may be moved
257 * here temporarily in rcu_process_callbacks().
258 * [*nxttail[RCU_DONE_TAIL], *nxttail[RCU_WAIT_TAIL]):
259 * Entries that batch # <= ->completed - 1: waiting for current GP
260 * [*nxttail[RCU_WAIT_TAIL], *nxttail[RCU_NEXT_READY_TAIL]):
261 * Entries known to have arrived before current GP ended
262 * [*nxttail[RCU_NEXT_READY_TAIL], *nxttail[RCU_NEXT_TAIL]):
263 * Entries that might have arrived after current GP ended
264 * Note that the value of *nxttail[RCU_NEXT_TAIL] will
265 * always be NULL, as this is the end of the list.
266 */
267 struct rcu_head *nxtlist;
268 struct rcu_head **nxttail[RCU_NEXT_SIZE];
269 long qlen; /* # of queued callbacks */
270 long qlen_last_fqs_check;
271 /* qlen at last check for QS forcing */
272 unsigned long n_cbs_invoked; /* count of RCU cbs invoked. */
273 unsigned long n_cbs_orphaned; /* RCU cbs orphaned by dying CPU */
274 unsigned long n_cbs_adopted; /* RCU cbs adopted from dying CPU */
275 unsigned long n_force_qs_snap;
276 /* did other CPU force QS recently? */
277 long blimit; /* Upper limit on a processed batch */
278
279 /* 3) dynticks interface. */
280 struct rcu_dynticks *dynticks; /* Shared per-CPU dynticks state. */
281 int dynticks_snap; /* Per-GP tracking for dynticks. */
282
283 /* 4) reasons this CPU needed to be kicked by force_quiescent_state */
284 unsigned long dynticks_fqs; /* Kicked due to dynticks idle. */
285 unsigned long offline_fqs; /* Kicked due to being offline. */
286 unsigned long resched_ipi; /* Sent a resched IPI. */
287
288 /* 5) __rcu_pending() statistics. */
289 unsigned long n_rcu_pending; /* rcu_pending() calls since boot. */
290 unsigned long n_rp_qs_pending;
291 unsigned long n_rp_report_qs;
292 unsigned long n_rp_cb_ready;
293 unsigned long n_rp_cpu_needs_gp;
294 unsigned long n_rp_gp_completed;
295 unsigned long n_rp_gp_started;
296 unsigned long n_rp_need_fqs;
297 unsigned long n_rp_need_nothing;
298
299 int cpu;
300 struct rcu_state *rsp;
301 };
302
303 /* Values for fqs_state field in struct rcu_state. */
304 #define RCU_GP_IDLE 0 /* No grace period in progress. */
305 #define RCU_GP_INIT 1 /* Grace period being initialized. */
306 #define RCU_SAVE_DYNTICK 2 /* Need to scan dyntick state. */
307 #define RCU_FORCE_QS 3 /* Need to force quiescent state. */
308 #define RCU_SIGNAL_INIT RCU_SAVE_DYNTICK
309
310 #define RCU_JIFFIES_TILL_FORCE_QS 3 /* for rsp->jiffies_force_qs */
311
312 #ifdef CONFIG_PROVE_RCU
313 #define RCU_STALL_DELAY_DELTA (5 * HZ)
314 #else
315 #define RCU_STALL_DELAY_DELTA 0
316 #endif
317
318 #define RCU_SECONDS_TILL_STALL_CHECK (CONFIG_RCU_CPU_STALL_TIMEOUT * HZ + \
319 RCU_STALL_DELAY_DELTA)
320 /* for rsp->jiffies_stall */
321 #define RCU_SECONDS_TILL_STALL_RECHECK (3 * RCU_SECONDS_TILL_STALL_CHECK + 30)
322 /* for rsp->jiffies_stall */
323 #define RCU_STALL_RAT_DELAY 2 /* Allow other CPUs time */
324 /* to take at least one */
325 /* scheduling clock irq */
326 /* before ratting on them. */
327
328 #define rcu_wait(cond) \
329 do { \
330 for (;;) { \
331 set_current_state(TASK_INTERRUPTIBLE); \
332 if (cond) \
333 break; \
334 schedule(); \
335 } \
336 __set_current_state(TASK_RUNNING); \
337 } while (0)
338
339 /*
340 * RCU global state, including node hierarchy. This hierarchy is
341 * represented in "heap" form in a dense array. The root (first level)
342 * of the hierarchy is in ->node[0] (referenced by ->level[0]), the second
343 * level in ->node[1] through ->node[m] (->node[1] referenced by ->level[1]),
344 * and the third level in ->node[m+1] and following (->node[m+1] referenced
345 * by ->level[2]). The number of levels is determined by the number of
346 * CPUs and by CONFIG_RCU_FANOUT. Small systems will have a "hierarchy"
347 * consisting of a single rcu_node.
348 */
349 struct rcu_state {
350 struct rcu_node node[NUM_RCU_NODES]; /* Hierarchy. */
351 struct rcu_node *level[NUM_RCU_LVLS]; /* Hierarchy levels. */
352 u32 levelcnt[MAX_RCU_LVLS + 1]; /* # nodes in each level. */
353 u8 levelspread[NUM_RCU_LVLS]; /* kids/node in each level. */
354 struct rcu_data __percpu *rda; /* pointer of percu rcu_data. */
355
356 /* The following fields are guarded by the root rcu_node's lock. */
357
358 u8 fqs_state ____cacheline_internodealigned_in_smp;
359 /* Force QS state. */
360 u8 fqs_active; /* force_quiescent_state() */
361 /* is running. */
362 u8 fqs_need_gp; /* A CPU was prevented from */
363 /* starting a new grace */
364 /* period because */
365 /* force_quiescent_state() */
366 /* was running. */
367 u8 boost; /* Subject to priority boost. */
368 unsigned long gpnum; /* Current gp number. */
369 unsigned long completed; /* # of last completed gp. */
370
371 /* End of fields guarded by root rcu_node's lock. */
372
373 raw_spinlock_t onofflock; /* exclude on/offline and */
374 /* starting new GP. */
375 raw_spinlock_t fqslock; /* Only one task forcing */
376 /* quiescent states. */
377 unsigned long jiffies_force_qs; /* Time at which to invoke */
378 /* force_quiescent_state(). */
379 unsigned long n_force_qs; /* Number of calls to */
380 /* force_quiescent_state(). */
381 unsigned long n_force_qs_lh; /* ~Number of calls leaving */
382 /* due to lock unavailable. */
383 unsigned long n_force_qs_ngp; /* Number of calls leaving */
384 /* due to no GP active. */
385 unsigned long gp_start; /* Time at which GP started, */
386 /* but in jiffies. */
387 unsigned long jiffies_stall; /* Time at which to check */
388 /* for CPU stalls. */
389 unsigned long gp_max; /* Maximum GP duration in */
390 /* jiffies. */
391 char *name; /* Name of structure. */
392 };
393
394 /* Return values for rcu_preempt_offline_tasks(). */
395
396 #define RCU_OFL_TASKS_NORM_GP 0x1 /* Tasks blocking normal */
397 /* GP were moved to root. */
398 #define RCU_OFL_TASKS_EXP_GP 0x2 /* Tasks blocking expedited */
399 /* GP were moved to root. */
400
401 /*
402 * RCU implementation internal declarations:
403 */
404 extern struct rcu_state rcu_sched_state;
405 DECLARE_PER_CPU(struct rcu_data, rcu_sched_data);
406
407 extern struct rcu_state rcu_bh_state;
408 DECLARE_PER_CPU(struct rcu_data, rcu_bh_data);
409
410 #ifdef CONFIG_TREE_PREEMPT_RCU
411 extern struct rcu_state rcu_preempt_state;
412 DECLARE_PER_CPU(struct rcu_data, rcu_preempt_data);
413 #endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */
414
415 #ifdef CONFIG_RCU_BOOST
416 DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_status);
417 DECLARE_PER_CPU(int, rcu_cpu_kthread_cpu);
418 DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_loops);
419 DECLARE_PER_CPU(char, rcu_cpu_has_work);
420 #endif /* #ifdef CONFIG_RCU_BOOST */
421
422 #ifndef RCU_TREE_NONCORE
423
424 /* Forward declarations for rcutree_plugin.h */
425 static void rcu_bootup_announce(void);
426 long rcu_batches_completed(void);
427 static void rcu_preempt_note_context_switch(int cpu);
428 static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp);
429 #ifdef CONFIG_HOTPLUG_CPU
430 static void rcu_report_unblock_qs_rnp(struct rcu_node *rnp,
431 unsigned long flags);
432 static void rcu_stop_cpu_kthread(int cpu);
433 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
434 static void rcu_print_detail_task_stall(struct rcu_state *rsp);
435 static int rcu_print_task_stall(struct rcu_node *rnp);
436 static void rcu_preempt_stall_reset(void);
437 static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp);
438 #ifdef CONFIG_HOTPLUG_CPU
439 static int rcu_preempt_offline_tasks(struct rcu_state *rsp,
440 struct rcu_node *rnp,
441 struct rcu_data *rdp);
442 static void rcu_preempt_offline_cpu(int cpu);
443 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
444 static void rcu_preempt_check_callbacks(int cpu);
445 static void rcu_preempt_process_callbacks(void);
446 void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu));
447 #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_TREE_PREEMPT_RCU)
448 static void rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp,
449 bool wake);
450 #endif /* #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_TREE_PREEMPT_RCU) */
451 static int rcu_preempt_pending(int cpu);
452 static int rcu_preempt_needs_cpu(int cpu);
453 static void __cpuinit rcu_preempt_init_percpu_data(int cpu);
454 static void rcu_preempt_send_cbs_to_online(void);
455 static void __init __rcu_init_preempt(void);
456 static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags);
457 static void rcu_preempt_boost_start_gp(struct rcu_node *rnp);
458 static void invoke_rcu_callbacks_kthread(void);
459 #ifdef CONFIG_RCU_BOOST
460 static void rcu_preempt_do_callbacks(void);
461 static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp,
462 cpumask_var_t cm);
463 static int __cpuinit rcu_spawn_one_boost_kthread(struct rcu_state *rsp,
464 struct rcu_node *rnp,
465 int rnp_index);
466 static void invoke_rcu_node_kthread(struct rcu_node *rnp);
467 static void rcu_yield(void (*f)(unsigned long), unsigned long arg);
468 #endif /* #ifdef CONFIG_RCU_BOOST */
469 static void rcu_cpu_kthread_setrt(int cpu, int to_rt);
470 static void __cpuinit rcu_prepare_kthreads(int cpu);
471 static void rcu_prepare_for_idle(int cpu);
472 static void rcu_wake_cpus_for_gp_end(void);
473 static void rcu_schedule_wake_gp_end(void);
474
475 #endif /* #ifndef RCU_TREE_NONCORE */
This page took 0.043863 seconds and 4 git commands to generate.