Merge branch 'for-3.2' of git://linux-nfs.org/~bfields/linux
[linux-btrfs-devel.git] / kernel / rcutree.h
blob01b2ccda26fbf82880cc1d4a50bf9f2877e91bba
1 /*
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)
76 #else
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.
86 struct rcu_dynticks {
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. */
92 /* RCU's kthread states for tracing. */
93 #define RCU_KTHREAD_STOPPED 0
94 #define RCU_KTHREAD_RUNNING 1
95 #define RCU_KTHREAD_WAITING 2
96 #define RCU_KTHREAD_OFFCPU 3
97 #define RCU_KTHREAD_YIELDING 4
98 #define RCU_KTHREAD_MAX 4
101 * Definition for node within the RCU grace-period-detection hierarchy.
103 struct rcu_node {
104 raw_spinlock_t lock; /* Root rcu_node's lock protects some */
105 /* rcu_state fields as well as following. */
106 unsigned long gpnum; /* Current grace period for this node. */
107 /* This will either be equal to or one */
108 /* behind the root rcu_node's gpnum. */
109 unsigned long completed; /* Last GP completed for this node. */
110 /* This will either be equal to or one */
111 /* behind the root rcu_node's gpnum. */
112 unsigned long qsmask; /* CPUs or groups that need to switch in */
113 /* order for current grace period to proceed.*/
114 /* In leaf rcu_node, each bit corresponds to */
115 /* an rcu_data structure, otherwise, each */
116 /* bit corresponds to a child rcu_node */
117 /* structure. */
118 unsigned long expmask; /* Groups that have ->blkd_tasks */
119 /* elements that need to drain to allow the */
120 /* current expedited grace period to */
121 /* complete (only for TREE_PREEMPT_RCU). */
122 atomic_t wakemask; /* CPUs whose kthread needs to be awakened. */
123 /* Since this has meaning only for leaf */
124 /* rcu_node structures, 32 bits suffices. */
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 unsigned int boost_kthread_status;
163 /* State of boost_kthread_task for tracing. */
164 unsigned long n_tasks_boosted;
165 /* Total number of tasks boosted. */
166 unsigned long n_exp_boosts;
167 /* Number of tasks boosted for expedited GP. */
168 unsigned long n_normal_boosts;
169 /* Number of tasks boosted for normal GP. */
170 unsigned long n_balk_blkd_tasks;
171 /* Refused to boost: no blocked tasks. */
172 unsigned long n_balk_exp_gp_tasks;
173 /* Refused to boost: nothing blocking GP. */
174 unsigned long n_balk_boost_tasks;
175 /* Refused to boost: already boosting. */
176 unsigned long n_balk_notblocked;
177 /* Refused to boost: RCU RS CS still running. */
178 unsigned long n_balk_notyet;
179 /* Refused to boost: not yet time. */
180 unsigned long n_balk_nos;
181 /* Refused to boost: not sure why, though. */
182 /* This can happen due to race conditions. */
183 #endif /* #ifdef CONFIG_RCU_BOOST */
184 struct task_struct *node_kthread_task;
185 /* kthread that takes care of this rcu_node */
186 /* structure, for example, awakening the */
187 /* per-CPU kthreads as needed. */
188 unsigned int node_kthread_status;
189 /* State of node_kthread_task for tracing. */
190 } ____cacheline_internodealigned_in_smp;
193 * Do a full breadth-first scan of the rcu_node structures for the
194 * specified rcu_state structure.
196 #define rcu_for_each_node_breadth_first(rsp, rnp) \
197 for ((rnp) = &(rsp)->node[0]; \
198 (rnp) < &(rsp)->node[NUM_RCU_NODES]; (rnp)++)
201 * Do a breadth-first scan of the non-leaf rcu_node structures for the
202 * specified rcu_state structure. Note that if there is a singleton
203 * rcu_node tree with but one rcu_node structure, this loop is a no-op.
205 #define rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) \
206 for ((rnp) = &(rsp)->node[0]; \
207 (rnp) < (rsp)->level[NUM_RCU_LVLS - 1]; (rnp)++)
210 * Scan the leaves of the rcu_node hierarchy for the specified rcu_state
211 * structure. Note that if there is a singleton rcu_node tree with but
212 * one rcu_node structure, this loop -will- visit the rcu_node structure.
213 * It is still a leaf node, even if it is also the root node.
215 #define rcu_for_each_leaf_node(rsp, rnp) \
216 for ((rnp) = (rsp)->level[NUM_RCU_LVLS - 1]; \
217 (rnp) < &(rsp)->node[NUM_RCU_NODES]; (rnp)++)
219 /* Index values for nxttail array in struct rcu_data. */
220 #define RCU_DONE_TAIL 0 /* Also RCU_WAIT head. */
221 #define RCU_WAIT_TAIL 1 /* Also RCU_NEXT_READY head. */
222 #define RCU_NEXT_READY_TAIL 2 /* Also RCU_NEXT head. */
223 #define RCU_NEXT_TAIL 3
224 #define RCU_NEXT_SIZE 4
226 /* Per-CPU data for read-copy update. */
227 struct rcu_data {
228 /* 1) quiescent-state and grace-period handling : */
229 unsigned long completed; /* Track rsp->completed gp number */
230 /* in order to detect GP end. */
231 unsigned long gpnum; /* Highest gp number that this CPU */
232 /* is aware of having started. */
233 unsigned long passed_quiesc_completed;
234 /* Value of completed at time of qs. */
235 bool passed_quiesc; /* User-mode/idle loop etc. */
236 bool qs_pending; /* Core waits for quiesc state. */
237 bool beenonline; /* CPU online at least once. */
238 bool preemptible; /* Preemptible RCU? */
239 struct rcu_node *mynode; /* This CPU's leaf of hierarchy */
240 unsigned long grpmask; /* Mask to apply to leaf qsmask. */
242 /* 2) batch handling */
244 * If nxtlist is not NULL, it is partitioned as follows.
245 * Any of the partitions might be empty, in which case the
246 * pointer to that partition will be equal to the pointer for
247 * the following partition. When the list is empty, all of
248 * the nxttail elements point to the ->nxtlist pointer itself,
249 * which in that case is NULL.
251 * [nxtlist, *nxttail[RCU_DONE_TAIL]):
252 * Entries that batch # <= ->completed
253 * The grace period for these entries has completed, and
254 * the other grace-period-completed entries may be moved
255 * here temporarily in rcu_process_callbacks().
256 * [*nxttail[RCU_DONE_TAIL], *nxttail[RCU_WAIT_TAIL]):
257 * Entries that batch # <= ->completed - 1: waiting for current GP
258 * [*nxttail[RCU_WAIT_TAIL], *nxttail[RCU_NEXT_READY_TAIL]):
259 * Entries known to have arrived before current GP ended
260 * [*nxttail[RCU_NEXT_READY_TAIL], *nxttail[RCU_NEXT_TAIL]):
261 * Entries that might have arrived after current GP ended
262 * Note that the value of *nxttail[RCU_NEXT_TAIL] will
263 * always be NULL, as this is the end of the list.
265 struct rcu_head *nxtlist;
266 struct rcu_head **nxttail[RCU_NEXT_SIZE];
267 long qlen; /* # of queued callbacks */
268 long qlen_last_fqs_check;
269 /* qlen at last check for QS forcing */
270 unsigned long n_cbs_invoked; /* count of RCU cbs invoked. */
271 unsigned long n_cbs_orphaned; /* RCU cbs orphaned by dying CPU */
272 unsigned long n_cbs_adopted; /* RCU cbs adopted from dying CPU */
273 unsigned long n_force_qs_snap;
274 /* did other CPU force QS recently? */
275 long blimit; /* Upper limit on a processed batch */
277 #ifdef CONFIG_NO_HZ
278 /* 3) dynticks interface. */
279 struct rcu_dynticks *dynticks; /* Shared per-CPU dynticks state. */
280 int dynticks_snap; /* Per-GP tracking for dynticks. */
281 #endif /* #ifdef CONFIG_NO_HZ */
283 /* 4) reasons this CPU needed to be kicked by force_quiescent_state */
284 #ifdef CONFIG_NO_HZ
285 unsigned long dynticks_fqs; /* Kicked due to dynticks idle. */
286 #endif /* #ifdef CONFIG_NO_HZ */
287 unsigned long offline_fqs; /* Kicked due to being offline. */
288 unsigned long resched_ipi; /* Sent a resched IPI. */
290 /* 5) __rcu_pending() statistics. */
291 unsigned long n_rcu_pending; /* rcu_pending() calls since boot. */
292 unsigned long n_rp_qs_pending;
293 unsigned long n_rp_report_qs;
294 unsigned long n_rp_cb_ready;
295 unsigned long n_rp_cpu_needs_gp;
296 unsigned long n_rp_gp_completed;
297 unsigned long n_rp_gp_started;
298 unsigned long n_rp_need_fqs;
299 unsigned long n_rp_need_nothing;
301 int cpu;
304 /* Values for signaled field in struct rcu_state. */
305 #define RCU_GP_IDLE 0 /* No grace period in progress. */
306 #define RCU_GP_INIT 1 /* Grace period being initialized. */
307 #define RCU_SAVE_DYNTICK 2 /* Need to scan dyntick state. */
308 #define RCU_FORCE_QS 3 /* Need to force quiescent state. */
309 #ifdef CONFIG_NO_HZ
310 #define RCU_SIGNAL_INIT RCU_SAVE_DYNTICK
311 #else /* #ifdef CONFIG_NO_HZ */
312 #define RCU_SIGNAL_INIT RCU_FORCE_QS
313 #endif /* #else #ifdef CONFIG_NO_HZ */
315 #define RCU_JIFFIES_TILL_FORCE_QS 3 /* for rsp->jiffies_force_qs */
317 #ifdef CONFIG_PROVE_RCU
318 #define RCU_STALL_DELAY_DELTA (5 * HZ)
319 #else
320 #define RCU_STALL_DELAY_DELTA 0
321 #endif
323 #define RCU_SECONDS_TILL_STALL_CHECK (CONFIG_RCU_CPU_STALL_TIMEOUT * HZ + \
324 RCU_STALL_DELAY_DELTA)
325 /* for rsp->jiffies_stall */
326 #define RCU_SECONDS_TILL_STALL_RECHECK (3 * RCU_SECONDS_TILL_STALL_CHECK + 30)
327 /* for rsp->jiffies_stall */
328 #define RCU_STALL_RAT_DELAY 2 /* Allow other CPUs time */
329 /* to take at least one */
330 /* scheduling clock irq */
331 /* before ratting on them. */
333 #define rcu_wait(cond) \
334 do { \
335 for (;;) { \
336 set_current_state(TASK_INTERRUPTIBLE); \
337 if (cond) \
338 break; \
339 schedule(); \
341 __set_current_state(TASK_RUNNING); \
342 } while (0)
345 * RCU global state, including node hierarchy. This hierarchy is
346 * represented in "heap" form in a dense array. The root (first level)
347 * of the hierarchy is in ->node[0] (referenced by ->level[0]), the second
348 * level in ->node[1] through ->node[m] (->node[1] referenced by ->level[1]),
349 * and the third level in ->node[m+1] and following (->node[m+1] referenced
350 * by ->level[2]). The number of levels is determined by the number of
351 * CPUs and by CONFIG_RCU_FANOUT. Small systems will have a "hierarchy"
352 * consisting of a single rcu_node.
354 struct rcu_state {
355 struct rcu_node node[NUM_RCU_NODES]; /* Hierarchy. */
356 struct rcu_node *level[NUM_RCU_LVLS]; /* Hierarchy levels. */
357 u32 levelcnt[MAX_RCU_LVLS + 1]; /* # nodes in each level. */
358 u8 levelspread[NUM_RCU_LVLS]; /* kids/node in each level. */
359 struct rcu_data __percpu *rda; /* pointer of percu rcu_data. */
361 /* The following fields are guarded by the root rcu_node's lock. */
363 u8 signaled ____cacheline_internodealigned_in_smp;
364 /* Force QS state. */
365 u8 fqs_active; /* force_quiescent_state() */
366 /* is running. */
367 u8 fqs_need_gp; /* A CPU was prevented from */
368 /* starting a new grace */
369 /* period because */
370 /* force_quiescent_state() */
371 /* was running. */
372 u8 boost; /* Subject to priority boost. */
373 unsigned long gpnum; /* Current gp number. */
374 unsigned long completed; /* # of last completed gp. */
376 /* End of fields guarded by root rcu_node's lock. */
378 raw_spinlock_t onofflock; /* exclude on/offline and */
379 /* starting new GP. */
380 raw_spinlock_t fqslock; /* Only one task forcing */
381 /* quiescent states. */
382 unsigned long jiffies_force_qs; /* Time at which to invoke */
383 /* force_quiescent_state(). */
384 unsigned long n_force_qs; /* Number of calls to */
385 /* force_quiescent_state(). */
386 unsigned long n_force_qs_lh; /* ~Number of calls leaving */
387 /* due to lock unavailable. */
388 unsigned long n_force_qs_ngp; /* Number of calls leaving */
389 /* due to no GP active. */
390 unsigned long gp_start; /* Time at which GP started, */
391 /* but in jiffies. */
392 unsigned long jiffies_stall; /* Time at which to check */
393 /* for CPU stalls. */
394 unsigned long gp_max; /* Maximum GP duration in */
395 /* jiffies. */
396 char *name; /* Name of structure. */
399 /* Return values for rcu_preempt_offline_tasks(). */
401 #define RCU_OFL_TASKS_NORM_GP 0x1 /* Tasks blocking normal */
402 /* GP were moved to root. */
403 #define RCU_OFL_TASKS_EXP_GP 0x2 /* Tasks blocking expedited */
404 /* GP were moved to root. */
407 * RCU implementation internal declarations:
409 extern struct rcu_state rcu_sched_state;
410 DECLARE_PER_CPU(struct rcu_data, rcu_sched_data);
412 extern struct rcu_state rcu_bh_state;
413 DECLARE_PER_CPU(struct rcu_data, rcu_bh_data);
415 #ifdef CONFIG_TREE_PREEMPT_RCU
416 extern struct rcu_state rcu_preempt_state;
417 DECLARE_PER_CPU(struct rcu_data, rcu_preempt_data);
418 #endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */
420 #ifndef RCU_TREE_NONCORE
422 /* Forward declarations for rcutree_plugin.h */
423 static void rcu_bootup_announce(void);
424 long rcu_batches_completed(void);
425 static void rcu_preempt_note_context_switch(int cpu);
426 static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp);
427 #ifdef CONFIG_HOTPLUG_CPU
428 static void rcu_report_unblock_qs_rnp(struct rcu_node *rnp,
429 unsigned long flags);
430 static void rcu_stop_cpu_kthread(int cpu);
431 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
432 static void rcu_print_detail_task_stall(struct rcu_state *rsp);
433 static void rcu_print_task_stall(struct rcu_node *rnp);
434 static void rcu_preempt_stall_reset(void);
435 static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp);
436 #ifdef CONFIG_HOTPLUG_CPU
437 static int rcu_preempt_offline_tasks(struct rcu_state *rsp,
438 struct rcu_node *rnp,
439 struct rcu_data *rdp);
440 static void rcu_preempt_offline_cpu(int cpu);
441 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
442 static void rcu_preempt_check_callbacks(int cpu);
443 static void rcu_preempt_process_callbacks(void);
444 void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu));
445 #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_TREE_PREEMPT_RCU)
446 static void rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp);
447 #endif /* #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_TREE_PREEMPT_RCU) */
448 static int rcu_preempt_pending(int cpu);
449 static int rcu_preempt_needs_cpu(int cpu);
450 static void __cpuinit rcu_preempt_init_percpu_data(int cpu);
451 static void rcu_preempt_send_cbs_to_online(void);
452 static void __init __rcu_init_preempt(void);
453 static void rcu_needs_cpu_flush(void);
454 static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags);
455 static void rcu_preempt_boost_start_gp(struct rcu_node *rnp);
456 static void invoke_rcu_callbacks_kthread(void);
457 #ifdef CONFIG_RCU_BOOST
458 static void rcu_preempt_do_callbacks(void);
459 static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp,
460 cpumask_var_t cm);
461 static int __cpuinit rcu_spawn_one_boost_kthread(struct rcu_state *rsp,
462 struct rcu_node *rnp,
463 int rnp_index);
464 static void invoke_rcu_node_kthread(struct rcu_node *rnp);
465 static void rcu_yield(void (*f)(unsigned long), unsigned long arg);
466 #endif /* #ifdef CONFIG_RCU_BOOST */
467 static void rcu_cpu_kthread_setrt(int cpu, int to_rt);
468 static void __cpuinit rcu_prepare_kthreads(int cpu);
470 #endif /* #ifndef RCU_TREE_NONCORE */