rcu: Enable fourth level of TREE_RCU hierarchy
[linux-2.6-block.git] / kernel / rcutree.h
CommitLineData
9f77da9f
PM
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 has not been tested, so there is probably some
35 * bug somewhere.
36 */
cf244dc0 37#define MAX_RCU_LVLS 4
9f77da9f
PM
38#define RCU_FANOUT (CONFIG_RCU_FANOUT)
39#define RCU_FANOUT_SQ (RCU_FANOUT * RCU_FANOUT)
40#define RCU_FANOUT_CUBE (RCU_FANOUT_SQ * RCU_FANOUT)
cf244dc0 41#define RCU_FANOUT_FOURTH (RCU_FANOUT_CUBE * RCU_FANOUT)
9f77da9f
PM
42
43#if NR_CPUS <= RCU_FANOUT
44# define NUM_RCU_LVLS 1
45# define NUM_RCU_LVL_0 1
46# define NUM_RCU_LVL_1 (NR_CPUS)
47# define NUM_RCU_LVL_2 0
48# define NUM_RCU_LVL_3 0
cf244dc0 49# define NUM_RCU_LVL_4 0
9f77da9f
PM
50#elif NR_CPUS <= RCU_FANOUT_SQ
51# define NUM_RCU_LVLS 2
52# define NUM_RCU_LVL_0 1
fc2219d4 53# define NUM_RCU_LVL_1 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT)
9f77da9f
PM
54# define NUM_RCU_LVL_2 (NR_CPUS)
55# define NUM_RCU_LVL_3 0
cf244dc0 56# define NUM_RCU_LVL_4 0
9f77da9f
PM
57#elif NR_CPUS <= RCU_FANOUT_CUBE
58# define NUM_RCU_LVLS 3
59# define NUM_RCU_LVL_0 1
fc2219d4
PM
60# define NUM_RCU_LVL_1 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_SQ)
61# define NUM_RCU_LVL_2 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT)
9f77da9f 62# define NUM_RCU_LVL_3 NR_CPUS
cf244dc0
PM
63# define NUM_RCU_LVL_4 0
64#elif NR_CPUS <= RCU_FANOUT_FOURTH
65# define NUM_RCU_LVLS 4
66# define NUM_RCU_LVL_0 1
67# define NUM_RCU_LVL_1 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_CUBE)
68# define NUM_RCU_LVL_2 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT_SQ)
69# define NUM_RCU_LVL_3 DIV_ROUND_UP(NR_CPUS, RCU_FANOUT)
70# define NUM_RCU_LVL_4 NR_CPUS
9f77da9f
PM
71#else
72# error "CONFIG_RCU_FANOUT insufficient for NR_CPUS"
73#endif /* #if (NR_CPUS) <= RCU_FANOUT */
74
cf244dc0 75#define RCU_SUM (NUM_RCU_LVL_0 + NUM_RCU_LVL_1 + NUM_RCU_LVL_2 + NUM_RCU_LVL_3 + NUM_RCU_LVL_4)
9f77da9f
PM
76#define NUM_RCU_NODES (RCU_SUM - NR_CPUS)
77
78/*
79 * Dynticks per-CPU state.
80 */
81struct rcu_dynticks {
82 int dynticks_nesting; /* Track nesting level, sort of. */
83 int dynticks; /* Even value for dynticks-idle, else odd. */
84 int dynticks_nmi; /* Even value for either dynticks-idle or */
85 /* not in nmi handler, else odd. So this */
86 /* remains even for nmi from irq handler. */
87};
88
89/*
90 * Definition for node within the RCU grace-period-detection hierarchy.
91 */
92struct rcu_node {
1eba8f84
PM
93 spinlock_t lock; /* Root rcu_node's lock protects some */
94 /* rcu_state fields as well as following. */
f41d911f 95 long gpnum; /* Current grace period for this node. */
86848966
PM
96 /* This will either be equal to or one */
97 /* behind the root rcu_node's gpnum. */
d09b62df
PM
98 long completed; /* Last grace period completed for this node. */
99 /* This will either be equal to or one */
100 /* behind the root rcu_node's gpnum. */
9f77da9f
PM
101 unsigned long qsmask; /* CPUs or groups that need to switch in */
102 /* order for current grace period to proceed.*/
1eba8f84
PM
103 /* In leaf rcu_node, each bit corresponds to */
104 /* an rcu_data structure, otherwise, each */
105 /* bit corresponds to a child rcu_node */
106 /* structure. */
9f77da9f
PM
107 unsigned long qsmaskinit;
108 /* Per-GP initialization for qsmask. */
109 unsigned long grpmask; /* Mask to apply to parent qsmask. */
1eba8f84 110 /* Only one bit will be set in this mask. */
9f77da9f
PM
111 int grplo; /* lowest-numbered CPU or group here. */
112 int grphi; /* highest-numbered CPU or group here. */
113 u8 grpnum; /* CPU/group number for next level up. */
114 u8 level; /* root is at level 0. */
115 struct rcu_node *parent;
f41d911f
PM
116 struct list_head blocked_tasks[2];
117 /* Tasks blocked in RCU read-side critsect. */
1eba8f84
PM
118 /* Grace period number (->gpnum) x blocked */
119 /* by tasks on the (x & 0x1) element of the */
120 /* blocked_tasks[] array. */
9f77da9f
PM
121} ____cacheline_internodealigned_in_smp;
122
a0b6c9a7
PM
123/*
124 * Do a full breadth-first scan of the rcu_node structures for the
125 * specified rcu_state structure.
126 */
127#define rcu_for_each_node_breadth_first(rsp, rnp) \
128 for ((rnp) = &(rsp)->node[0]; \
129 (rnp) < &(rsp)->node[NUM_RCU_NODES]; (rnp)++)
130
131#define rcu_for_each_leaf_node(rsp, rnp) \
132 for ((rnp) = (rsp)->level[NUM_RCU_LVLS - 1]; \
133 (rnp) < &(rsp)->node[NUM_RCU_NODES]; (rnp)++)
134
9f77da9f
PM
135/* Index values for nxttail array in struct rcu_data. */
136#define RCU_DONE_TAIL 0 /* Also RCU_WAIT head. */
137#define RCU_WAIT_TAIL 1 /* Also RCU_NEXT_READY head. */
138#define RCU_NEXT_READY_TAIL 2 /* Also RCU_NEXT head. */
139#define RCU_NEXT_TAIL 3
140#define RCU_NEXT_SIZE 4
141
142/* Per-CPU data for read-copy update. */
143struct rcu_data {
144 /* 1) quiescent-state and grace-period handling : */
145 long completed; /* Track rsp->completed gp number */
146 /* in order to detect GP end. */
147 long gpnum; /* Highest gp number that this CPU */
148 /* is aware of having started. */
149 long passed_quiesc_completed;
150 /* Value of completed at time of qs. */
151 bool passed_quiesc; /* User-mode/idle loop etc. */
152 bool qs_pending; /* Core waits for quiesc state. */
153 bool beenonline; /* CPU online at least once. */
f41d911f 154 bool preemptable; /* Preemptable RCU? */
9f77da9f
PM
155 struct rcu_node *mynode; /* This CPU's leaf of hierarchy */
156 unsigned long grpmask; /* Mask to apply to leaf qsmask. */
157
158 /* 2) batch handling */
159 /*
160 * If nxtlist is not NULL, it is partitioned as follows.
161 * Any of the partitions might be empty, in which case the
162 * pointer to that partition will be equal to the pointer for
163 * the following partition. When the list is empty, all of
1eba8f84
PM
164 * the nxttail elements point to the ->nxtlist pointer itself,
165 * which in that case is NULL.
9f77da9f 166 *
9f77da9f
PM
167 * [nxtlist, *nxttail[RCU_DONE_TAIL]):
168 * Entries that batch # <= ->completed
169 * The grace period for these entries has completed, and
170 * the other grace-period-completed entries may be moved
171 * here temporarily in rcu_process_callbacks().
1eba8f84
PM
172 * [*nxttail[RCU_DONE_TAIL], *nxttail[RCU_WAIT_TAIL]):
173 * Entries that batch # <= ->completed - 1: waiting for current GP
174 * [*nxttail[RCU_WAIT_TAIL], *nxttail[RCU_NEXT_READY_TAIL]):
175 * Entries known to have arrived before current GP ended
176 * [*nxttail[RCU_NEXT_READY_TAIL], *nxttail[RCU_NEXT_TAIL]):
177 * Entries that might have arrived after current GP ended
178 * Note that the value of *nxttail[RCU_NEXT_TAIL] will
179 * always be NULL, as this is the end of the list.
9f77da9f
PM
180 */
181 struct rcu_head *nxtlist;
182 struct rcu_head **nxttail[RCU_NEXT_SIZE];
a71fca58 183 long qlen; /* # of queued callbacks */
37c72e56
PM
184 long qlen_last_fqs_check;
185 /* qlen at last check for QS forcing */
186 unsigned long n_force_qs_snap;
187 /* did other CPU force QS recently? */
9f77da9f
PM
188 long blimit; /* Upper limit on a processed batch */
189
190#ifdef CONFIG_NO_HZ
191 /* 3) dynticks interface. */
192 struct rcu_dynticks *dynticks; /* Shared per-CPU dynticks state. */
193 int dynticks_snap; /* Per-GP tracking for dynticks. */
194 int dynticks_nmi_snap; /* Per-GP tracking for dynticks_nmi. */
195#endif /* #ifdef CONFIG_NO_HZ */
196
197 /* 4) reasons this CPU needed to be kicked by force_quiescent_state */
198#ifdef CONFIG_NO_HZ
199 unsigned long dynticks_fqs; /* Kicked due to dynticks idle. */
200#endif /* #ifdef CONFIG_NO_HZ */
201 unsigned long offline_fqs; /* Kicked due to being offline. */
202 unsigned long resched_ipi; /* Sent a resched IPI. */
203
204 /* 5) __rcu_pending() statistics. */
205 long n_rcu_pending; /* rcu_pending() calls since boot. */
206 long n_rp_qs_pending;
207 long n_rp_cb_ready;
208 long n_rp_cpu_needs_gp;
209 long n_rp_gp_completed;
210 long n_rp_gp_started;
211 long n_rp_need_fqs;
212 long n_rp_need_nothing;
213
214 int cpu;
215};
216
217/* Values for signaled field in struct rcu_state. */
83f5b01f
PM
218#define RCU_GP_IDLE 0 /* No grace period in progress. */
219#define RCU_GP_INIT 1 /* Grace period being initialized. */
220#define RCU_SAVE_DYNTICK 2 /* Need to scan dyntick state. */
281d150c
PM
221#define RCU_SAVE_COMPLETED 3 /* Need to save rsp->completed. */
222#define RCU_FORCE_QS 4 /* Need to force quiescent state. */
9f77da9f
PM
223#ifdef CONFIG_NO_HZ
224#define RCU_SIGNAL_INIT RCU_SAVE_DYNTICK
225#else /* #ifdef CONFIG_NO_HZ */
281d150c 226#define RCU_SIGNAL_INIT RCU_SAVE_COMPLETED
9f77da9f
PM
227#endif /* #else #ifdef CONFIG_NO_HZ */
228
229#define RCU_JIFFIES_TILL_FORCE_QS 3 /* for rsp->jiffies_force_qs */
230#ifdef CONFIG_RCU_CPU_STALL_DETECTOR
231#define RCU_SECONDS_TILL_STALL_CHECK (10 * HZ) /* for rsp->jiffies_stall */
232#define RCU_SECONDS_TILL_STALL_RECHECK (30 * HZ) /* for rsp->jiffies_stall */
233#define RCU_STALL_RAT_DELAY 2 /* Allow other CPUs time */
234 /* to take at least one */
235 /* scheduling clock irq */
236 /* before ratting on them. */
237
238#endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
239
240/*
241 * RCU global state, including node hierarchy. This hierarchy is
242 * represented in "heap" form in a dense array. The root (first level)
243 * of the hierarchy is in ->node[0] (referenced by ->level[0]), the second
244 * level in ->node[1] through ->node[m] (->node[1] referenced by ->level[1]),
245 * and the third level in ->node[m+1] and following (->node[m+1] referenced
246 * by ->level[2]). The number of levels is determined by the number of
247 * CPUs and by CONFIG_RCU_FANOUT. Small systems will have a "hierarchy"
248 * consisting of a single rcu_node.
249 */
250struct rcu_state {
251 struct rcu_node node[NUM_RCU_NODES]; /* Hierarchy. */
252 struct rcu_node *level[NUM_RCU_LVLS]; /* Hierarchy levels. */
253 u32 levelcnt[MAX_RCU_LVLS + 1]; /* # nodes in each level. */
254 u8 levelspread[NUM_RCU_LVLS]; /* kids/node in each level. */
255 struct rcu_data *rda[NR_CPUS]; /* array of rdp pointers. */
256
257 /* The following fields are guarded by the root rcu_node's lock. */
258
259 u8 signaled ____cacheline_internodealigned_in_smp;
260 /* Force QS state. */
261 long gpnum; /* Current gp number. */
262 long completed; /* # of last completed gp. */
1eba8f84
PM
263
264 /* End of fields guarded by root rcu_node's lock. */
265
9f77da9f 266 spinlock_t onofflock; /* exclude on/offline and */
e74f4c45
PM
267 /* starting new GP. Also */
268 /* protects the following */
269 /* orphan_cbs fields. */
270 struct rcu_head *orphan_cbs_list; /* list of rcu_head structs */
271 /* orphaned by all CPUs in */
272 /* a given leaf rcu_node */
273 /* going offline. */
274 struct rcu_head **orphan_cbs_tail; /* And tail pointer. */
275 long orphan_qlen; /* Number of orphaned cbs. */
9f77da9f
PM
276 spinlock_t fqslock; /* Only one task forcing */
277 /* quiescent states. */
4bcfe055
PM
278 long completed_fqs; /* Value of completed @ snap. */
279 /* Protected by fqslock. */
9f77da9f
PM
280 unsigned long jiffies_force_qs; /* Time at which to invoke */
281 /* force_quiescent_state(). */
282 unsigned long n_force_qs; /* Number of calls to */
283 /* force_quiescent_state(). */
284 unsigned long n_force_qs_lh; /* ~Number of calls leaving */
285 /* due to lock unavailable. */
286 unsigned long n_force_qs_ngp; /* Number of calls leaving */
287 /* due to no GP active. */
288#ifdef CONFIG_RCU_CPU_STALL_DETECTOR
289 unsigned long gp_start; /* Time at which GP started, */
290 /* but in jiffies. */
291 unsigned long jiffies_stall; /* Time at which to check */
292 /* for CPU stalls. */
293#endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
9f77da9f
PM
294};
295
296#ifdef RCU_TREE_NONCORE
6258c4fb
IM
297
298/*
299 * RCU implementation internal declarations:
300 */
d6714c22
PM
301extern struct rcu_state rcu_sched_state;
302DECLARE_PER_CPU(struct rcu_data, rcu_sched_data);
6258c4fb
IM
303
304extern struct rcu_state rcu_bh_state;
305DECLARE_PER_CPU(struct rcu_data, rcu_bh_data);
306
f41d911f
PM
307#ifdef CONFIG_TREE_PREEMPT_RCU
308extern struct rcu_state rcu_preempt_state;
309DECLARE_PER_CPU(struct rcu_data, rcu_preempt_data);
310#endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */
311
9b2619af 312#else /* #ifdef RCU_TREE_NONCORE */
9f77da9f 313
9b2619af 314/* Forward declarations for rcutree_plugin.h */
dbe01350 315static void rcu_bootup_announce(void);
9b2619af
PM
316long rcu_batches_completed(void);
317static void rcu_preempt_note_context_switch(int cpu);
318static int rcu_preempted_readers(struct rcu_node *rnp);
b668c9cf 319#ifdef CONFIG_HOTPLUG_CPU
d3f6bad3
PM
320static void rcu_report_unblock_qs_rnp(struct rcu_node *rnp,
321 unsigned long flags);
b668c9cf 322#endif /* #ifdef CONFIG_HOTPLUG_CPU */
9b2619af
PM
323#ifdef CONFIG_RCU_CPU_STALL_DETECTOR
324static void rcu_print_task_stall(struct rcu_node *rnp);
325#endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
326static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp);
327#ifdef CONFIG_HOTPLUG_CPU
237c80c5
PM
328static int rcu_preempt_offline_tasks(struct rcu_state *rsp,
329 struct rcu_node *rnp,
330 struct rcu_data *rdp);
9b2619af
PM
331static void rcu_preempt_offline_cpu(int cpu);
332#endif /* #ifdef CONFIG_HOTPLUG_CPU */
333static void rcu_preempt_check_callbacks(int cpu);
334static void rcu_preempt_process_callbacks(void);
335void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu));
336static int rcu_preempt_pending(int cpu);
337static int rcu_preempt_needs_cpu(int cpu);
338static void __cpuinit rcu_preempt_init_percpu_data(int cpu);
e74f4c45 339static void rcu_preempt_send_cbs_to_orphanage(void);
9b2619af
PM
340static void __init __rcu_init_preempt(void);
341
342#endif /* #else #ifdef RCU_TREE_NONCORE */