rcu: add grace-period age and more kthread state to tracing
[GitHub/mt8127/android_kernel_alcatel_ttab.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 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. */
90 };
91
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
99
100 /*
101 * Definition for node within the RCU grace-period-detection hierarchy.
102 */
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 unsigned long wakemask; /* CPUs whose kthread needs to be awakened. */
123 unsigned long qsmaskinit;
124 /* Per-GP initial value for qsmask & expmask. */
125 unsigned long grpmask; /* Mask to apply to parent qsmask. */
126 /* Only one bit will be set in this mask. */
127 int grplo; /* lowest-numbered CPU or group here. */
128 int grphi; /* highest-numbered CPU or group here. */
129 u8 grpnum; /* CPU/group number for next level up. */
130 u8 level; /* root is at level 0. */
131 struct rcu_node *parent;
132 struct list_head blkd_tasks;
133 /* Tasks blocked in RCU read-side critical */
134 /* section. Tasks are placed at the head */
135 /* of this list and age towards the tail. */
136 struct list_head *gp_tasks;
137 /* Pointer to the first task blocking the */
138 /* current grace period, or NULL if there */
139 /* is no such task. */
140 struct list_head *exp_tasks;
141 /* Pointer to the first task blocking the */
142 /* current expedited grace period, or NULL */
143 /* if there is no such task. If there */
144 /* is no current expedited grace period, */
145 /* then there can cannot be any such task. */
146 #ifdef CONFIG_RCU_BOOST
147 struct list_head *boost_tasks;
148 /* Pointer to first task that needs to be */
149 /* priority boosted, or NULL if no priority */
150 /* boosting is needed for this rcu_node */
151 /* structure. If there are no tasks */
152 /* queued on this rcu_node structure that */
153 /* are blocking the current grace period, */
154 /* there can be no such task. */
155 unsigned long boost_time;
156 /* When to start boosting (jiffies). */
157 struct task_struct *boost_kthread_task;
158 /* kthread that takes care of priority */
159 /* boosting for this rcu_node structure. */
160 wait_queue_head_t boost_wq;
161 /* Wait queue on which to park the boost */
162 /* kthread. */
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 wait_queue_head_t node_wq;
190 /* Wait queue on which to park the per-node */
191 /* kthread. */
192 unsigned int node_kthread_status;
193 /* State of node_kthread_task for tracing. */
194 } ____cacheline_internodealigned_in_smp;
195
196 /*
197 * Do a full breadth-first scan of the rcu_node structures for the
198 * specified rcu_state structure.
199 */
200 #define rcu_for_each_node_breadth_first(rsp, rnp) \
201 for ((rnp) = &(rsp)->node[0]; \
202 (rnp) < &(rsp)->node[NUM_RCU_NODES]; (rnp)++)
203
204 /*
205 * Do a breadth-first scan of the non-leaf rcu_node structures for the
206 * specified rcu_state structure. Note that if there is a singleton
207 * rcu_node tree with but one rcu_node structure, this loop is a no-op.
208 */
209 #define rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) \
210 for ((rnp) = &(rsp)->node[0]; \
211 (rnp) < (rsp)->level[NUM_RCU_LVLS - 1]; (rnp)++)
212
213 /*
214 * Scan the leaves of the rcu_node hierarchy for the specified rcu_state
215 * structure. Note that if there is a singleton rcu_node tree with but
216 * one rcu_node structure, this loop -will- visit the rcu_node structure.
217 * It is still a leaf node, even if it is also the root node.
218 */
219 #define rcu_for_each_leaf_node(rsp, rnp) \
220 for ((rnp) = (rsp)->level[NUM_RCU_LVLS - 1]; \
221 (rnp) < &(rsp)->node[NUM_RCU_NODES]; (rnp)++)
222
223 /* Index values for nxttail array in struct rcu_data. */
224 #define RCU_DONE_TAIL 0 /* Also RCU_WAIT head. */
225 #define RCU_WAIT_TAIL 1 /* Also RCU_NEXT_READY head. */
226 #define RCU_NEXT_READY_TAIL 2 /* Also RCU_NEXT head. */
227 #define RCU_NEXT_TAIL 3
228 #define RCU_NEXT_SIZE 4
229
230 /* Per-CPU data for read-copy update. */
231 struct rcu_data {
232 /* 1) quiescent-state and grace-period handling : */
233 unsigned long completed; /* Track rsp->completed gp number */
234 /* in order to detect GP end. */
235 unsigned long gpnum; /* Highest gp number that this CPU */
236 /* is aware of having started. */
237 unsigned long passed_quiesc_completed;
238 /* Value of completed at time of qs. */
239 bool passed_quiesc; /* User-mode/idle loop etc. */
240 bool qs_pending; /* Core waits for quiesc state. */
241 bool beenonline; /* CPU online at least once. */
242 bool preemptable; /* Preemptable RCU? */
243 struct rcu_node *mynode; /* This CPU's leaf of hierarchy */
244 unsigned long grpmask; /* Mask to apply to leaf qsmask. */
245
246 /* 2) batch handling */
247 /*
248 * If nxtlist is not NULL, it is partitioned as follows.
249 * Any of the partitions might be empty, in which case the
250 * pointer to that partition will be equal to the pointer for
251 * the following partition. When the list is empty, all of
252 * the nxttail elements point to the ->nxtlist pointer itself,
253 * which in that case is NULL.
254 *
255 * [nxtlist, *nxttail[RCU_DONE_TAIL]):
256 * Entries that batch # <= ->completed
257 * The grace period for these entries has completed, and
258 * the other grace-period-completed entries may be moved
259 * here temporarily in rcu_process_callbacks().
260 * [*nxttail[RCU_DONE_TAIL], *nxttail[RCU_WAIT_TAIL]):
261 * Entries that batch # <= ->completed - 1: waiting for current GP
262 * [*nxttail[RCU_WAIT_TAIL], *nxttail[RCU_NEXT_READY_TAIL]):
263 * Entries known to have arrived before current GP ended
264 * [*nxttail[RCU_NEXT_READY_TAIL], *nxttail[RCU_NEXT_TAIL]):
265 * Entries that might have arrived after current GP ended
266 * Note that the value of *nxttail[RCU_NEXT_TAIL] will
267 * always be NULL, as this is the end of the list.
268 */
269 struct rcu_head *nxtlist;
270 struct rcu_head **nxttail[RCU_NEXT_SIZE];
271 long qlen; /* # of queued callbacks */
272 long qlen_last_fqs_check;
273 /* qlen at last check for QS forcing */
274 unsigned long n_cbs_invoked; /* count of RCU cbs invoked. */
275 unsigned long n_cbs_orphaned; /* RCU cbs orphaned by dying CPU */
276 unsigned long n_cbs_adopted; /* RCU cbs adopted from dying CPU */
277 unsigned long n_force_qs_snap;
278 /* did other CPU force QS recently? */
279 long blimit; /* Upper limit on a processed batch */
280
281 #ifdef CONFIG_NO_HZ
282 /* 3) dynticks interface. */
283 struct rcu_dynticks *dynticks; /* Shared per-CPU dynticks state. */
284 int dynticks_snap; /* Per-GP tracking for dynticks. */
285 #endif /* #ifdef CONFIG_NO_HZ */
286
287 /* 4) reasons this CPU needed to be kicked by force_quiescent_state */
288 #ifdef CONFIG_NO_HZ
289 unsigned long dynticks_fqs; /* Kicked due to dynticks idle. */
290 #endif /* #ifdef CONFIG_NO_HZ */
291 unsigned long offline_fqs; /* Kicked due to being offline. */
292 unsigned long resched_ipi; /* Sent a resched IPI. */
293
294 /* 5) __rcu_pending() statistics. */
295 unsigned long n_rcu_pending; /* rcu_pending() calls since boot. */
296 unsigned long n_rp_qs_pending;
297 unsigned long n_rp_report_qs;
298 unsigned long n_rp_cb_ready;
299 unsigned long n_rp_cpu_needs_gp;
300 unsigned long n_rp_gp_completed;
301 unsigned long n_rp_gp_started;
302 unsigned long n_rp_need_fqs;
303 unsigned long n_rp_need_nothing;
304
305 int cpu;
306 };
307
308 /* Values for signaled 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 #ifdef CONFIG_NO_HZ
314 #define RCU_SIGNAL_INIT RCU_SAVE_DYNTICK
315 #else /* #ifdef CONFIG_NO_HZ */
316 #define RCU_SIGNAL_INIT RCU_FORCE_QS
317 #endif /* #else #ifdef CONFIG_NO_HZ */
318
319 #define RCU_JIFFIES_TILL_FORCE_QS 3 /* for rsp->jiffies_force_qs */
320
321 #ifdef CONFIG_PROVE_RCU
322 #define RCU_STALL_DELAY_DELTA (5 * HZ)
323 #else
324 #define RCU_STALL_DELAY_DELTA 0
325 #endif
326
327 #define RCU_SECONDS_TILL_STALL_CHECK (CONFIG_RCU_CPU_STALL_TIMEOUT * HZ + \
328 RCU_STALL_DELAY_DELTA)
329 /* for rsp->jiffies_stall */
330 #define RCU_SECONDS_TILL_STALL_RECHECK (3 * RCU_SECONDS_TILL_STALL_CHECK + 30)
331 /* for rsp->jiffies_stall */
332 #define RCU_STALL_RAT_DELAY 2 /* Allow other CPUs time */
333 /* to take at least one */
334 /* scheduling clock irq */
335 /* before ratting on them. */
336
337
338 /*
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.
347 */
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. */
354
355 /* The following fields are guarded by the root rcu_node's lock. */
356
357 u8 signaled ____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 unsigned long gpnum; /* Current gp number. */
367 unsigned long completed; /* # of last completed gp. */
368
369 /* End of fields guarded by root rcu_node's lock. */
370
371 raw_spinlock_t onofflock; /* exclude on/offline and */
372 /* starting new GP. */
373 raw_spinlock_t fqslock; /* Only one task forcing */
374 /* quiescent states. */
375 unsigned long jiffies_force_qs; /* Time at which to invoke */
376 /* force_quiescent_state(). */
377 unsigned long n_force_qs; /* Number of calls to */
378 /* force_quiescent_state(). */
379 unsigned long n_force_qs_lh; /* ~Number of calls leaving */
380 /* due to lock unavailable. */
381 unsigned long n_force_qs_ngp; /* Number of calls leaving */
382 /* due to no GP active. */
383 unsigned long gp_start; /* Time at which GP started, */
384 /* but in jiffies. */
385 unsigned long jiffies_stall; /* Time at which to check */
386 /* for CPU stalls. */
387 unsigned long gp_max; /* Maximum GP duration in */
388 /* jiffies. */
389 char *name; /* Name of structure. */
390 };
391
392 /* Return values for rcu_preempt_offline_tasks(). */
393
394 #define RCU_OFL_TASKS_NORM_GP 0x1 /* Tasks blocking normal */
395 /* GP were moved to root. */
396 #define RCU_OFL_TASKS_EXP_GP 0x2 /* Tasks blocking expedited */
397 /* GP were moved to root. */
398
399 /*
400 * RCU implementation internal declarations:
401 */
402 extern struct rcu_state rcu_sched_state;
403 DECLARE_PER_CPU(struct rcu_data, rcu_sched_data);
404
405 extern struct rcu_state rcu_bh_state;
406 DECLARE_PER_CPU(struct rcu_data, rcu_bh_data);
407
408 #ifdef CONFIG_TREE_PREEMPT_RCU
409 extern struct rcu_state rcu_preempt_state;
410 DECLARE_PER_CPU(struct rcu_data, rcu_preempt_data);
411 #endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */
412
413 #ifndef RCU_TREE_NONCORE
414
415 /* Forward declarations for rcutree_plugin.h */
416 static void rcu_bootup_announce(void);
417 long rcu_batches_completed(void);
418 static void rcu_preempt_note_context_switch(int cpu);
419 static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp);
420 #ifdef CONFIG_HOTPLUG_CPU
421 static void rcu_report_unblock_qs_rnp(struct rcu_node *rnp,
422 unsigned long flags);
423 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
424 static void rcu_print_detail_task_stall(struct rcu_state *rsp);
425 static void rcu_print_task_stall(struct rcu_node *rnp);
426 static void rcu_preempt_stall_reset(void);
427 static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp);
428 #ifdef CONFIG_HOTPLUG_CPU
429 static int rcu_preempt_offline_tasks(struct rcu_state *rsp,
430 struct rcu_node *rnp,
431 struct rcu_data *rdp);
432 static void rcu_preempt_offline_cpu(int cpu);
433 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
434 static void rcu_preempt_check_callbacks(int cpu);
435 static void rcu_preempt_process_callbacks(void);
436 void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu));
437 #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_TREE_PREEMPT_RCU)
438 static void rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp);
439 #endif /* #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_TREE_PREEMPT_RCU) */
440 static int rcu_preempt_pending(int cpu);
441 static int rcu_preempt_needs_cpu(int cpu);
442 static void __cpuinit rcu_preempt_init_percpu_data(int cpu);
443 static void rcu_preempt_send_cbs_to_online(void);
444 static void __init __rcu_init_preempt(void);
445 static void rcu_needs_cpu_flush(void);
446 static void __init rcu_init_boost_waitqueue(struct rcu_node *rnp);
447 static void rcu_initiate_boost(struct rcu_node *rnp);
448 static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp,
449 cpumask_var_t cm);
450 static void rcu_preempt_boost_start_gp(struct rcu_node *rnp);
451 static int __cpuinit rcu_spawn_one_boost_kthread(struct rcu_state *rsp,
452 struct rcu_node *rnp,
453 int rnp_index);
454 #ifdef CONFIG_HOTPLUG_CPU
455 static void rcu_stop_boost_kthread(struct rcu_node *rnp);
456 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
457
458 #endif /* #ifndef RCU_TREE_NONCORE */