rcu: add tracing for RCU's kthread run states.
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / kernel / rcutree.h
blob67341dbebd95961af287a81cc95005292b66bd65
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_YIELDING 3
97 #define RCU_KTHREAD_MAX 3
100 * Definition for node within the RCU grace-period-detection hierarchy.
102 struct rcu_node {
103 raw_spinlock_t lock; /* Root rcu_node's lock protects some */
104 /* rcu_state fields as well as following. */
105 unsigned long gpnum; /* Current grace period for this node. */
106 /* This will either be equal to or one */
107 /* behind the root rcu_node's gpnum. */
108 unsigned long completed; /* Last GP completed for this node. */
109 /* This will either be equal to or one */
110 /* behind the root rcu_node's gpnum. */
111 unsigned long qsmask; /* CPUs or groups that need to switch in */
112 /* order for current grace period to proceed.*/
113 /* In leaf rcu_node, each bit corresponds to */
114 /* an rcu_data structure, otherwise, each */
115 /* bit corresponds to a child rcu_node */
116 /* structure. */
117 unsigned long expmask; /* Groups that have ->blkd_tasks */
118 /* elements that need to drain to allow the */
119 /* current expedited grace period to */
120 /* complete (only for TREE_PREEMPT_RCU). */
121 unsigned long wakemask; /* CPUs whose kthread needs to be awakened. */
122 unsigned long qsmaskinit;
123 /* Per-GP initial value for qsmask & expmask. */
124 unsigned long grpmask; /* Mask to apply to parent qsmask. */
125 /* Only one bit will be set in this mask. */
126 int grplo; /* lowest-numbered CPU or group here. */
127 int grphi; /* highest-numbered CPU or group here. */
128 u8 grpnum; /* CPU/group number for next level up. */
129 u8 level; /* root is at level 0. */
130 struct rcu_node *parent;
131 struct list_head blkd_tasks;
132 /* Tasks blocked in RCU read-side critical */
133 /* section. Tasks are placed at the head */
134 /* of this list and age towards the tail. */
135 struct list_head *gp_tasks;
136 /* Pointer to the first task blocking the */
137 /* current grace period, or NULL if there */
138 /* is no such task. */
139 struct list_head *exp_tasks;
140 /* Pointer to the first task blocking the */
141 /* current expedited grace period, or NULL */
142 /* if there is no such task. If there */
143 /* is no current expedited grace period, */
144 /* then there can cannot be any such task. */
145 #ifdef CONFIG_RCU_BOOST
146 struct list_head *boost_tasks;
147 /* Pointer to first task that needs to be */
148 /* priority boosted, or NULL if no priority */
149 /* boosting is needed for this rcu_node */
150 /* structure. If there are no tasks */
151 /* queued on this rcu_node structure that */
152 /* are blocking the current grace period, */
153 /* there can be no such task. */
154 unsigned long boost_time;
155 /* When to start boosting (jiffies). */
156 struct task_struct *boost_kthread_task;
157 /* kthread that takes care of priority */
158 /* boosting for this rcu_node structure. */
159 wait_queue_head_t boost_wq;
160 /* Wait queue on which to park the boost */
161 /* kthread. */
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 wait_queue_head_t node_wq;
189 /* Wait queue on which to park the per-node */
190 /* kthread. */
191 unsigned int node_kthread_status;
192 /* State of node_kthread_task for tracing. */
193 } ____cacheline_internodealigned_in_smp;
196 * Do a full breadth-first scan of the rcu_node structures for the
197 * specified rcu_state structure.
199 #define rcu_for_each_node_breadth_first(rsp, rnp) \
200 for ((rnp) = &(rsp)->node[0]; \
201 (rnp) < &(rsp)->node[NUM_RCU_NODES]; (rnp)++)
204 * Do a breadth-first scan of the non-leaf rcu_node structures for the
205 * specified rcu_state structure. Note that if there is a singleton
206 * rcu_node tree with but one rcu_node structure, this loop is a no-op.
208 #define rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) \
209 for ((rnp) = &(rsp)->node[0]; \
210 (rnp) < (rsp)->level[NUM_RCU_LVLS - 1]; (rnp)++)
213 * Scan the leaves of the rcu_node hierarchy for the specified rcu_state
214 * structure. Note that if there is a singleton rcu_node tree with but
215 * one rcu_node structure, this loop -will- visit the rcu_node structure.
216 * It is still a leaf node, even if it is also the root node.
218 #define rcu_for_each_leaf_node(rsp, rnp) \
219 for ((rnp) = (rsp)->level[NUM_RCU_LVLS - 1]; \
220 (rnp) < &(rsp)->node[NUM_RCU_NODES]; (rnp)++)
222 /* Index values for nxttail array in struct rcu_data. */
223 #define RCU_DONE_TAIL 0 /* Also RCU_WAIT head. */
224 #define RCU_WAIT_TAIL 1 /* Also RCU_NEXT_READY head. */
225 #define RCU_NEXT_READY_TAIL 2 /* Also RCU_NEXT head. */
226 #define RCU_NEXT_TAIL 3
227 #define RCU_NEXT_SIZE 4
229 /* Per-CPU data for read-copy update. */
230 struct rcu_data {
231 /* 1) quiescent-state and grace-period handling : */
232 unsigned long completed; /* Track rsp->completed gp number */
233 /* in order to detect GP end. */
234 unsigned long gpnum; /* Highest gp number that this CPU */
235 /* is aware of having started. */
236 unsigned long passed_quiesc_completed;
237 /* Value of completed at time of qs. */
238 bool passed_quiesc; /* User-mode/idle loop etc. */
239 bool qs_pending; /* Core waits for quiesc state. */
240 bool beenonline; /* CPU online at least once. */
241 bool preemptable; /* Preemptable RCU? */
242 struct rcu_node *mynode; /* This CPU's leaf of hierarchy */
243 unsigned long grpmask; /* Mask to apply to leaf qsmask. */
245 /* 2) batch handling */
247 * If nxtlist is not NULL, it is partitioned as follows.
248 * Any of the partitions might be empty, in which case the
249 * pointer to that partition will be equal to the pointer for
250 * the following partition. When the list is empty, all of
251 * the nxttail elements point to the ->nxtlist pointer itself,
252 * which in that case is NULL.
254 * [nxtlist, *nxttail[RCU_DONE_TAIL]):
255 * Entries that batch # <= ->completed
256 * The grace period for these entries has completed, and
257 * the other grace-period-completed entries may be moved
258 * here temporarily in rcu_process_callbacks().
259 * [*nxttail[RCU_DONE_TAIL], *nxttail[RCU_WAIT_TAIL]):
260 * Entries that batch # <= ->completed - 1: waiting for current GP
261 * [*nxttail[RCU_WAIT_TAIL], *nxttail[RCU_NEXT_READY_TAIL]):
262 * Entries known to have arrived before current GP ended
263 * [*nxttail[RCU_NEXT_READY_TAIL], *nxttail[RCU_NEXT_TAIL]):
264 * Entries that might have arrived after current GP ended
265 * Note that the value of *nxttail[RCU_NEXT_TAIL] will
266 * always be NULL, as this is the end of the list.
268 struct rcu_head *nxtlist;
269 struct rcu_head **nxttail[RCU_NEXT_SIZE];
270 long qlen; /* # of queued callbacks */
271 long qlen_last_fqs_check;
272 /* qlen at last check for QS forcing */
273 unsigned long n_cbs_invoked; /* count of RCU cbs invoked. */
274 unsigned long n_cbs_orphaned; /* RCU cbs orphaned by dying CPU */
275 unsigned long n_cbs_adopted; /* RCU cbs adopted from dying CPU */
276 unsigned long n_force_qs_snap;
277 /* did other CPU force QS recently? */
278 long blimit; /* Upper limit on a processed batch */
280 #ifdef CONFIG_NO_HZ
281 /* 3) dynticks interface. */
282 struct rcu_dynticks *dynticks; /* Shared per-CPU dynticks state. */
283 int dynticks_snap; /* Per-GP tracking for dynticks. */
284 #endif /* #ifdef CONFIG_NO_HZ */
286 /* 4) reasons this CPU needed to be kicked by force_quiescent_state */
287 #ifdef CONFIG_NO_HZ
288 unsigned long dynticks_fqs; /* Kicked due to dynticks idle. */
289 #endif /* #ifdef CONFIG_NO_HZ */
290 unsigned long offline_fqs; /* Kicked due to being offline. */
291 unsigned long resched_ipi; /* Sent a resched IPI. */
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;
307 /* Values for signaled field in struct rcu_state. */
308 #define RCU_GP_IDLE 0 /* No grace period in progress. */
309 #define RCU_GP_INIT 1 /* Grace period being initialized. */
310 #define RCU_SAVE_DYNTICK 2 /* Need to scan dyntick state. */
311 #define RCU_FORCE_QS 3 /* Need to force quiescent state. */
312 #ifdef CONFIG_NO_HZ
313 #define RCU_SIGNAL_INIT RCU_SAVE_DYNTICK
314 #else /* #ifdef CONFIG_NO_HZ */
315 #define RCU_SIGNAL_INIT RCU_FORCE_QS
316 #endif /* #else #ifdef CONFIG_NO_HZ */
318 #define RCU_JIFFIES_TILL_FORCE_QS 3 /* for rsp->jiffies_force_qs */
320 #ifdef CONFIG_PROVE_RCU
321 #define RCU_STALL_DELAY_DELTA (5 * HZ)
322 #else
323 #define RCU_STALL_DELAY_DELTA 0
324 #endif
326 #define RCU_SECONDS_TILL_STALL_CHECK (CONFIG_RCU_CPU_STALL_TIMEOUT * HZ + \
327 RCU_STALL_DELAY_DELTA)
328 /* for rsp->jiffies_stall */
329 #define RCU_SECONDS_TILL_STALL_RECHECK (3 * RCU_SECONDS_TILL_STALL_CHECK + 30)
330 /* for rsp->jiffies_stall */
331 #define RCU_STALL_RAT_DELAY 2 /* Allow other CPUs time */
332 /* to take at least one */
333 /* scheduling clock irq */
334 /* before ratting on them. */
338 * RCU global state, including node hierarchy. This hierarchy is
339 * represented in "heap" form in a dense array. The root (first level)
340 * of the hierarchy is in ->node[0] (referenced by ->level[0]), the second
341 * level in ->node[1] through ->node[m] (->node[1] referenced by ->level[1]),
342 * and the third level in ->node[m+1] and following (->node[m+1] referenced
343 * by ->level[2]). The number of levels is determined by the number of
344 * CPUs and by CONFIG_RCU_FANOUT. Small systems will have a "hierarchy"
345 * consisting of a single rcu_node.
347 struct rcu_state {
348 struct rcu_node node[NUM_RCU_NODES]; /* Hierarchy. */
349 struct rcu_node *level[NUM_RCU_LVLS]; /* Hierarchy levels. */
350 u32 levelcnt[MAX_RCU_LVLS + 1]; /* # nodes in each level. */
351 u8 levelspread[NUM_RCU_LVLS]; /* kids/node in each level. */
352 struct rcu_data __percpu *rda; /* pointer of percu rcu_data. */
354 /* The following fields are guarded by the root rcu_node's lock. */
356 u8 signaled ____cacheline_internodealigned_in_smp;
357 /* Force QS state. */
358 u8 fqs_active; /* force_quiescent_state() */
359 /* is running. */
360 u8 fqs_need_gp; /* A CPU was prevented from */
361 /* starting a new grace */
362 /* period because */
363 /* force_quiescent_state() */
364 /* was running. */
365 unsigned long gpnum; /* Current gp number. */
366 unsigned long completed; /* # of last completed gp. */
368 /* End of fields guarded by root rcu_node's lock. */
370 raw_spinlock_t onofflock; /* exclude on/offline and */
371 /* starting new GP. */
372 raw_spinlock_t fqslock; /* Only one task forcing */
373 /* quiescent states. */
374 unsigned long jiffies_force_qs; /* Time at which to invoke */
375 /* force_quiescent_state(). */
376 unsigned long n_force_qs; /* Number of calls to */
377 /* force_quiescent_state(). */
378 unsigned long n_force_qs_lh; /* ~Number of calls leaving */
379 /* due to lock unavailable. */
380 unsigned long n_force_qs_ngp; /* Number of calls leaving */
381 /* due to no GP active. */
382 unsigned long gp_start; /* Time at which GP started, */
383 /* but in jiffies. */
384 unsigned long jiffies_stall; /* Time at which to check */
385 /* for CPU stalls. */
386 char *name; /* Name of structure. */
389 /* Return values for rcu_preempt_offline_tasks(). */
391 #define RCU_OFL_TASKS_NORM_GP 0x1 /* Tasks blocking normal */
392 /* GP were moved to root. */
393 #define RCU_OFL_TASKS_EXP_GP 0x2 /* Tasks blocking expedited */
394 /* GP were moved to root. */
397 * RCU implementation internal declarations:
399 extern struct rcu_state rcu_sched_state;
400 DECLARE_PER_CPU(struct rcu_data, rcu_sched_data);
402 extern struct rcu_state rcu_bh_state;
403 DECLARE_PER_CPU(struct rcu_data, rcu_bh_data);
405 #ifdef CONFIG_TREE_PREEMPT_RCU
406 extern struct rcu_state rcu_preempt_state;
407 DECLARE_PER_CPU(struct rcu_data, rcu_preempt_data);
408 #endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */
410 #ifndef RCU_TREE_NONCORE
412 /* Forward declarations for rcutree_plugin.h */
413 static void rcu_bootup_announce(void);
414 long rcu_batches_completed(void);
415 static void rcu_preempt_note_context_switch(int cpu);
416 static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp);
417 #ifdef CONFIG_HOTPLUG_CPU
418 static void rcu_report_unblock_qs_rnp(struct rcu_node *rnp,
419 unsigned long flags);
420 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
421 static void rcu_print_detail_task_stall(struct rcu_state *rsp);
422 static void rcu_print_task_stall(struct rcu_node *rnp);
423 static void rcu_preempt_stall_reset(void);
424 static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp);
425 #ifdef CONFIG_HOTPLUG_CPU
426 static int rcu_preempt_offline_tasks(struct rcu_state *rsp,
427 struct rcu_node *rnp,
428 struct rcu_data *rdp);
429 static void rcu_preempt_offline_cpu(int cpu);
430 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
431 static void rcu_preempt_check_callbacks(int cpu);
432 static void rcu_preempt_process_callbacks(void);
433 void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu));
434 #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_TREE_PREEMPT_RCU)
435 static void rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp);
436 #endif /* #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_TREE_PREEMPT_RCU) */
437 static int rcu_preempt_pending(int cpu);
438 static int rcu_preempt_needs_cpu(int cpu);
439 static void __cpuinit rcu_preempt_init_percpu_data(int cpu);
440 static void rcu_preempt_send_cbs_to_online(void);
441 static void __init __rcu_init_preempt(void);
442 static void rcu_needs_cpu_flush(void);
443 static void __init rcu_init_boost_waitqueue(struct rcu_node *rnp);
444 static void rcu_initiate_boost(struct rcu_node *rnp);
445 static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp,
446 cpumask_var_t cm);
447 static void rcu_preempt_boost_start_gp(struct rcu_node *rnp);
448 static int __cpuinit rcu_spawn_one_boost_kthread(struct rcu_state *rsp,
449 struct rcu_node *rnp,
450 int rnp_index);
451 #ifdef CONFIG_HOTPLUG_CPU
452 static void rcu_stop_boost_kthread(struct rcu_node *rnp);
453 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
455 #endif /* #ifndef RCU_TREE_NONCORE */