percpu: pcpu_embed_first_chunk() should free unused parts after all allocs are complete
[linux-2.6.git] / kernel / rcutree.h
blobcdd1be0a40720f0a436951035be9f3adc68d1815
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 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. */
93 /* RCU's kthread states for tracing. */
94 #define RCU_KTHREAD_STOPPED 0
95 #define RCU_KTHREAD_RUNNING 1
96 #define RCU_KTHREAD_WAITING 2
97 #define RCU_KTHREAD_OFFCPU 3
98 #define RCU_KTHREAD_YIELDING 4
99 #define RCU_KTHREAD_MAX 4
102 * Definition for node within the RCU grace-period-detection hierarchy.
104 struct rcu_node {
105 raw_spinlock_t lock; /* Root rcu_node's lock protects some */
106 /* rcu_state fields as well as following. */
107 unsigned long gpnum; /* Current grace period for this node. */
108 /* This will either be equal to or one */
109 /* behind the root rcu_node's gpnum. */
110 unsigned long completed; /* Last GP completed for this node. */
111 /* This will either be equal to or one */
112 /* behind the root rcu_node's gpnum. */
113 unsigned long qsmask; /* CPUs or groups that need to switch in */
114 /* order for current grace period to proceed.*/
115 /* In leaf rcu_node, each bit corresponds to */
116 /* an rcu_data structure, otherwise, each */
117 /* bit corresponds to a child rcu_node */
118 /* structure. */
119 unsigned long expmask; /* Groups that have ->blkd_tasks */
120 /* elements that need to drain to allow the */
121 /* current expedited grace period to */
122 /* complete (only for TREE_PREEMPT_RCU). */
123 atomic_t wakemask; /* CPUs whose kthread needs to be awakened. */
124 /* Since this has meaning only for leaf */
125 /* rcu_node structures, 32 bits suffices. */
126 unsigned long qsmaskinit;
127 /* Per-GP initial value for qsmask & expmask. */
128 unsigned long grpmask; /* Mask to apply to parent qsmask. */
129 /* Only one bit will be set in this mask. */
130 int grplo; /* lowest-numbered CPU or group here. */
131 int grphi; /* highest-numbered CPU or group here. */
132 u8 grpnum; /* CPU/group number for next level up. */
133 u8 level; /* root is at level 0. */
134 struct rcu_node *parent;
135 struct list_head blkd_tasks;
136 /* Tasks blocked in RCU read-side critical */
137 /* section. Tasks are placed at the head */
138 /* of this list and age towards the tail. */
139 struct list_head *gp_tasks;
140 /* Pointer to the first task blocking the */
141 /* current grace period, or NULL if there */
142 /* is no such task. */
143 struct list_head *exp_tasks;
144 /* Pointer to the first task blocking the */
145 /* current expedited grace period, or NULL */
146 /* if there is no such task. If there */
147 /* is no current expedited grace period, */
148 /* then there can cannot be any such task. */
149 #ifdef CONFIG_RCU_BOOST
150 struct list_head *boost_tasks;
151 /* Pointer to first task that needs to be */
152 /* priority boosted, or NULL if no priority */
153 /* boosting is needed for this rcu_node */
154 /* structure. If there are no tasks */
155 /* queued on this rcu_node structure that */
156 /* are blocking the current grace period, */
157 /* there can be no such task. */
158 unsigned long boost_time;
159 /* When to start boosting (jiffies). */
160 struct task_struct *boost_kthread_task;
161 /* kthread that takes care of priority */
162 /* boosting for this rcu_node structure. */
163 unsigned int boost_kthread_status;
164 /* State of boost_kthread_task for tracing. */
165 unsigned long n_tasks_boosted;
166 /* Total number of tasks boosted. */
167 unsigned long n_exp_boosts;
168 /* Number of tasks boosted for expedited GP. */
169 unsigned long n_normal_boosts;
170 /* Number of tasks boosted for normal GP. */
171 unsigned long n_balk_blkd_tasks;
172 /* Refused to boost: no blocked tasks. */
173 unsigned long n_balk_exp_gp_tasks;
174 /* Refused to boost: nothing blocking GP. */
175 unsigned long n_balk_boost_tasks;
176 /* Refused to boost: already boosting. */
177 unsigned long n_balk_notblocked;
178 /* Refused to boost: RCU RS CS still running. */
179 unsigned long n_balk_notyet;
180 /* Refused to boost: not yet time. */
181 unsigned long n_balk_nos;
182 /* Refused to boost: not sure why, though. */
183 /* This can happen due to race conditions. */
184 #endif /* #ifdef CONFIG_RCU_BOOST */
185 struct task_struct *node_kthread_task;
186 /* kthread that takes care of this rcu_node */
187 /* structure, for example, awakening the */
188 /* per-CPU kthreads as needed. */
189 unsigned int node_kthread_status;
190 /* State of node_kthread_task for tracing. */
191 } ____cacheline_internodealigned_in_smp;
194 * Do a full breadth-first scan of the rcu_node structures for the
195 * specified rcu_state structure.
197 #define rcu_for_each_node_breadth_first(rsp, rnp) \
198 for ((rnp) = &(rsp)->node[0]; \
199 (rnp) < &(rsp)->node[NUM_RCU_NODES]; (rnp)++)
202 * Do a breadth-first scan of the non-leaf rcu_node structures for the
203 * specified rcu_state structure. Note that if there is a singleton
204 * rcu_node tree with but one rcu_node structure, this loop is a no-op.
206 #define rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) \
207 for ((rnp) = &(rsp)->node[0]; \
208 (rnp) < (rsp)->level[NUM_RCU_LVLS - 1]; (rnp)++)
211 * Scan the leaves of the rcu_node hierarchy for the specified rcu_state
212 * structure. Note that if there is a singleton rcu_node tree with but
213 * one rcu_node structure, this loop -will- visit the rcu_node structure.
214 * It is still a leaf node, even if it is also the root node.
216 #define rcu_for_each_leaf_node(rsp, rnp) \
217 for ((rnp) = (rsp)->level[NUM_RCU_LVLS - 1]; \
218 (rnp) < &(rsp)->node[NUM_RCU_NODES]; (rnp)++)
220 /* Index values for nxttail array in struct rcu_data. */
221 #define RCU_DONE_TAIL 0 /* Also RCU_WAIT head. */
222 #define RCU_WAIT_TAIL 1 /* Also RCU_NEXT_READY head. */
223 #define RCU_NEXT_READY_TAIL 2 /* Also RCU_NEXT head. */
224 #define RCU_NEXT_TAIL 3
225 #define RCU_NEXT_SIZE 4
227 /* Per-CPU data for read-copy update. */
228 struct rcu_data {
229 /* 1) quiescent-state and grace-period handling : */
230 unsigned long completed; /* Track rsp->completed gp number */
231 /* in order to detect GP end. */
232 unsigned long gpnum; /* Highest gp number that this CPU */
233 /* is aware of having started. */
234 unsigned long passed_quiesce_gpnum;
235 /* gpnum at time of quiescent state. */
236 bool passed_quiesce; /* User-mode/idle loop etc. */
237 bool qs_pending; /* Core waits for quiesc state. */
238 bool beenonline; /* CPU online at least once. */
239 bool preemptible; /* Preemptible RCU? */
240 struct rcu_node *mynode; /* This CPU's leaf of hierarchy */
241 unsigned long grpmask; /* Mask to apply to leaf qsmask. */
242 #ifdef CONFIG_RCU_CPU_STALL_INFO
243 unsigned long ticks_this_gp; /* The number of scheduling-clock */
244 /* ticks this CPU has handled */
245 /* during and after the last grace */
246 /* period it is aware of. */
247 #endif /* #ifdef CONFIG_RCU_CPU_STALL_INFO */
249 /* 2) batch handling */
251 * If nxtlist is not NULL, it is partitioned as follows.
252 * Any of the partitions might be empty, in which case the
253 * pointer to that partition will be equal to the pointer for
254 * the following partition. When the list is empty, all of
255 * the nxttail elements point to the ->nxtlist pointer itself,
256 * which in that case is NULL.
258 * [nxtlist, *nxttail[RCU_DONE_TAIL]):
259 * Entries that batch # <= ->completed
260 * The grace period for these entries has completed, and
261 * the other grace-period-completed entries may be moved
262 * here temporarily in rcu_process_callbacks().
263 * [*nxttail[RCU_DONE_TAIL], *nxttail[RCU_WAIT_TAIL]):
264 * Entries that batch # <= ->completed - 1: waiting for current GP
265 * [*nxttail[RCU_WAIT_TAIL], *nxttail[RCU_NEXT_READY_TAIL]):
266 * Entries known to have arrived before current GP ended
267 * [*nxttail[RCU_NEXT_READY_TAIL], *nxttail[RCU_NEXT_TAIL]):
268 * Entries that might have arrived after current GP ended
269 * Note that the value of *nxttail[RCU_NEXT_TAIL] will
270 * always be NULL, as this is the end of the list.
272 struct rcu_head *nxtlist;
273 struct rcu_head **nxttail[RCU_NEXT_SIZE];
274 long qlen_lazy; /* # of lazy queued callbacks */
275 long qlen; /* # of queued callbacks, incl lazy */
276 long qlen_last_fqs_check;
277 /* qlen at last check for QS forcing */
278 unsigned long n_cbs_invoked; /* count of RCU cbs invoked. */
279 unsigned long n_cbs_orphaned; /* RCU cbs orphaned by dying CPU */
280 unsigned long n_cbs_adopted; /* RCU cbs adopted from dying CPU */
281 unsigned long n_force_qs_snap;
282 /* did other CPU force QS recently? */
283 long blimit; /* Upper limit on a processed batch */
285 /* 3) dynticks interface. */
286 struct rcu_dynticks *dynticks; /* Shared per-CPU dynticks state. */
287 int dynticks_snap; /* Per-GP tracking for dynticks. */
289 /* 4) reasons this CPU needed to be kicked by force_quiescent_state */
290 unsigned long dynticks_fqs; /* Kicked due to dynticks idle. */
291 unsigned long offline_fqs; /* Kicked due to being offline. */
293 /* 5) __rcu_pending() statistics. */
294 unsigned long n_rcu_pending; /* rcu_pending() calls since boot. */
295 unsigned long n_rp_qs_pending;
296 unsigned long n_rp_report_qs;
297 unsigned long n_rp_cb_ready;
298 unsigned long n_rp_cpu_needs_gp;
299 unsigned long n_rp_gp_completed;
300 unsigned long n_rp_gp_started;
301 unsigned long n_rp_need_fqs;
302 unsigned long n_rp_need_nothing;
304 int cpu;
305 struct rcu_state *rsp;
308 /* Values for fqs_state field in struct rcu_state. */
309 #define RCU_GP_IDLE 0 /* No grace period in progress. */
310 #define RCU_GP_INIT 1 /* Grace period being initialized. */
311 #define RCU_SAVE_DYNTICK 2 /* Need to scan dyntick state. */
312 #define RCU_FORCE_QS 3 /* Need to force quiescent state. */
313 #define RCU_SIGNAL_INIT RCU_SAVE_DYNTICK
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
322 #define RCU_STALL_RAT_DELAY 2 /* Allow other CPUs time */
323 /* to take at least one */
324 /* scheduling clock irq */
325 /* before ratting on them. */
327 #define rcu_wait(cond) \
328 do { \
329 for (;;) { \
330 set_current_state(TASK_INTERRUPTIBLE); \
331 if (cond) \
332 break; \
333 schedule(); \
335 __set_current_state(TASK_RUNNING); \
336 } while (0)
339 * RCU global state, including node hierarchy. This hierarchy is
340 * represented in "heap" form in a dense array. The root (first level)
341 * of the hierarchy is in ->node[0] (referenced by ->level[0]), the second
342 * level in ->node[1] through ->node[m] (->node[1] referenced by ->level[1]),
343 * and the third level in ->node[m+1] and following (->node[m+1] referenced
344 * by ->level[2]). The number of levels is determined by the number of
345 * CPUs and by CONFIG_RCU_FANOUT. Small systems will have a "hierarchy"
346 * consisting of a single rcu_node.
348 struct rcu_state {
349 struct rcu_node node[NUM_RCU_NODES]; /* Hierarchy. */
350 struct rcu_node *level[NUM_RCU_LVLS]; /* Hierarchy levels. */
351 u32 levelcnt[MAX_RCU_LVLS + 1]; /* # nodes in each level. */
352 u8 levelspread[NUM_RCU_LVLS]; /* kids/node in each level. */
353 struct rcu_data __percpu *rda; /* pointer of percu rcu_data. */
355 /* The following fields are guarded by the root rcu_node's lock. */
357 u8 fqs_state ____cacheline_internodealigned_in_smp;
358 /* Force QS state. */
359 u8 fqs_active; /* force_quiescent_state() */
360 /* is running. */
361 u8 fqs_need_gp; /* A CPU was prevented from */
362 /* starting a new grace */
363 /* period because */
364 /* force_quiescent_state() */
365 /* was running. */
366 u8 boost; /* Subject to priority boost. */
367 unsigned long gpnum; /* Current gp number. */
368 unsigned long completed; /* # of last completed gp. */
370 /* End of fields guarded by root rcu_node's lock. */
372 raw_spinlock_t onofflock; /* exclude on/offline and */
373 /* starting new GP. */
374 raw_spinlock_t fqslock; /* Only one task forcing */
375 /* quiescent states. */
376 unsigned long jiffies_force_qs; /* Time at which to invoke */
377 /* force_quiescent_state(). */
378 unsigned long n_force_qs; /* Number of calls to */
379 /* force_quiescent_state(). */
380 unsigned long n_force_qs_lh; /* ~Number of calls leaving */
381 /* due to lock unavailable. */
382 unsigned long n_force_qs_ngp; /* Number of calls leaving */
383 /* due to no GP active. */
384 unsigned long gp_start; /* Time at which GP started, */
385 /* but in jiffies. */
386 unsigned long jiffies_stall; /* Time at which to check */
387 /* for CPU stalls. */
388 unsigned long gp_max; /* Maximum GP duration in */
389 /* jiffies. */
390 char *name; /* Name of structure. */
393 /* Return values for rcu_preempt_offline_tasks(). */
395 #define RCU_OFL_TASKS_NORM_GP 0x1 /* Tasks blocking normal */
396 /* GP were moved to root. */
397 #define RCU_OFL_TASKS_EXP_GP 0x2 /* Tasks blocking expedited */
398 /* GP were moved to root. */
401 * RCU implementation internal declarations:
403 extern struct rcu_state rcu_sched_state;
404 DECLARE_PER_CPU(struct rcu_data, rcu_sched_data);
406 extern struct rcu_state rcu_bh_state;
407 DECLARE_PER_CPU(struct rcu_data, rcu_bh_data);
409 #ifdef CONFIG_TREE_PREEMPT_RCU
410 extern struct rcu_state rcu_preempt_state;
411 DECLARE_PER_CPU(struct rcu_data, rcu_preempt_data);
412 #endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */
414 #ifdef CONFIG_RCU_BOOST
415 DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_status);
416 DECLARE_PER_CPU(int, rcu_cpu_kthread_cpu);
417 DECLARE_PER_CPU(unsigned int, rcu_cpu_kthread_loops);
418 DECLARE_PER_CPU(char, rcu_cpu_has_work);
419 #endif /* #ifdef CONFIG_RCU_BOOST */
421 #ifndef RCU_TREE_NONCORE
423 /* Forward declarations for rcutree_plugin.h */
424 static void rcu_bootup_announce(void);
425 long rcu_batches_completed(void);
426 static void rcu_preempt_note_context_switch(int cpu);
427 static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp);
428 #ifdef CONFIG_HOTPLUG_CPU
429 static void rcu_report_unblock_qs_rnp(struct rcu_node *rnp,
430 unsigned long flags);
431 static void rcu_stop_cpu_kthread(int cpu);
432 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
433 static void rcu_print_detail_task_stall(struct rcu_state *rsp);
434 static int rcu_print_task_stall(struct rcu_node *rnp);
435 static void rcu_preempt_stall_reset(void);
436 static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp);
437 #ifdef CONFIG_HOTPLUG_CPU
438 static int rcu_preempt_offline_tasks(struct rcu_state *rsp,
439 struct rcu_node *rnp,
440 struct rcu_data *rdp);
441 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
442 static void rcu_preempt_cleanup_dead_cpu(int cpu);
443 static void rcu_preempt_check_callbacks(int cpu);
444 static void rcu_preempt_process_callbacks(void);
445 void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu));
446 #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_TREE_PREEMPT_RCU)
447 static void rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp,
448 bool wake);
449 #endif /* #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_TREE_PREEMPT_RCU) */
450 static int rcu_preempt_pending(int cpu);
451 static int rcu_preempt_cpu_has_callbacks(int cpu);
452 static void __cpuinit rcu_preempt_init_percpu_data(int cpu);
453 static void rcu_preempt_cleanup_dying_cpu(void);
454 static void __init __rcu_init_preempt(void);
455 static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags);
456 static void rcu_preempt_boost_start_gp(struct rcu_node *rnp);
457 static void invoke_rcu_callbacks_kthread(void);
458 static bool rcu_is_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_init(int cpu);
472 static void rcu_cleanup_after_idle(int cpu);
473 static void rcu_prepare_for_idle(int cpu);
474 static void print_cpu_stall_info_begin(void);
475 static void print_cpu_stall_info(struct rcu_state *rsp, int cpu);
476 static void print_cpu_stall_info_end(void);
477 static void zero_cpu_stall_ticks(struct rcu_data *rdp);
478 static void increment_cpu_stall_ticks(void);
480 #endif /* #ifndef RCU_TREE_NONCORE */