rcu: Create rcutree plugins to handle hotplug CPU for multi-level trees
[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 */
37#define MAX_RCU_LVLS 3
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)
41
42#if NR_CPUS <= RCU_FANOUT
43# define NUM_RCU_LVLS 1
44# define NUM_RCU_LVL_0 1
45# define NUM_RCU_LVL_1 (NR_CPUS)
46# define NUM_RCU_LVL_2 0
47# define NUM_RCU_LVL_3 0
48#elif NR_CPUS <= RCU_FANOUT_SQ
49# define NUM_RCU_LVLS 2
50# define NUM_RCU_LVL_0 1
51# define NUM_RCU_LVL_1 (((NR_CPUS) + RCU_FANOUT - 1) / RCU_FANOUT)
52# define NUM_RCU_LVL_2 (NR_CPUS)
53# define NUM_RCU_LVL_3 0
54#elif NR_CPUS <= RCU_FANOUT_CUBE
55# define NUM_RCU_LVLS 3
56# define NUM_RCU_LVL_0 1
57# define NUM_RCU_LVL_1 (((NR_CPUS) + RCU_FANOUT_SQ - 1) / RCU_FANOUT_SQ)
58# define NUM_RCU_LVL_2 (((NR_CPUS) + (RCU_FANOUT) - 1) / (RCU_FANOUT))
59# define NUM_RCU_LVL_3 NR_CPUS
60#else
61# error "CONFIG_RCU_FANOUT insufficient for NR_CPUS"
62#endif /* #if (NR_CPUS) <= RCU_FANOUT */
63
64#define RCU_SUM (NUM_RCU_LVL_0 + NUM_RCU_LVL_1 + NUM_RCU_LVL_2 + NUM_RCU_LVL_3)
65#define NUM_RCU_NODES (RCU_SUM - NR_CPUS)
66
67/*
68 * Dynticks per-CPU state.
69 */
70struct rcu_dynticks {
71 int dynticks_nesting; /* Track nesting level, sort of. */
72 int dynticks; /* Even value for dynticks-idle, else odd. */
73 int dynticks_nmi; /* Even value for either dynticks-idle or */
74 /* not in nmi handler, else odd. So this */
75 /* remains even for nmi from irq handler. */
76};
77
78/*
79 * Definition for node within the RCU grace-period-detection hierarchy.
80 */
81struct rcu_node {
82 spinlock_t lock;
f41d911f 83 long gpnum; /* Current grace period for this node. */
9f77da9f
PM
84 unsigned long qsmask; /* CPUs or groups that need to switch in */
85 /* order for current grace period to proceed.*/
86 unsigned long qsmaskinit;
87 /* Per-GP initialization for qsmask. */
88 unsigned long grpmask; /* Mask to apply to parent qsmask. */
89 int grplo; /* lowest-numbered CPU or group here. */
90 int grphi; /* highest-numbered CPU or group here. */
91 u8 grpnum; /* CPU/group number for next level up. */
92 u8 level; /* root is at level 0. */
93 struct rcu_node *parent;
f41d911f
PM
94 struct list_head blocked_tasks[2];
95 /* Tasks blocked in RCU read-side critsect. */
9f77da9f
PM
96} ____cacheline_internodealigned_in_smp;
97
98/* Index values for nxttail array in struct rcu_data. */
99#define RCU_DONE_TAIL 0 /* Also RCU_WAIT head. */
100#define RCU_WAIT_TAIL 1 /* Also RCU_NEXT_READY head. */
101#define RCU_NEXT_READY_TAIL 2 /* Also RCU_NEXT head. */
102#define RCU_NEXT_TAIL 3
103#define RCU_NEXT_SIZE 4
104
105/* Per-CPU data for read-copy update. */
106struct rcu_data {
107 /* 1) quiescent-state and grace-period handling : */
108 long completed; /* Track rsp->completed gp number */
109 /* in order to detect GP end. */
110 long gpnum; /* Highest gp number that this CPU */
111 /* is aware of having started. */
112 long passed_quiesc_completed;
113 /* Value of completed at time of qs. */
114 bool passed_quiesc; /* User-mode/idle loop etc. */
115 bool qs_pending; /* Core waits for quiesc state. */
116 bool beenonline; /* CPU online at least once. */
f41d911f 117 bool preemptable; /* Preemptable RCU? */
9f77da9f
PM
118 struct rcu_node *mynode; /* This CPU's leaf of hierarchy */
119 unsigned long grpmask; /* Mask to apply to leaf qsmask. */
120
121 /* 2) batch handling */
122 /*
123 * If nxtlist is not NULL, it is partitioned as follows.
124 * Any of the partitions might be empty, in which case the
125 * pointer to that partition will be equal to the pointer for
126 * the following partition. When the list is empty, all of
127 * the nxttail elements point to nxtlist, which is NULL.
128 *
129 * [*nxttail[RCU_NEXT_READY_TAIL], NULL = *nxttail[RCU_NEXT_TAIL]):
130 * Entries that might have arrived after current GP ended
131 * [*nxttail[RCU_WAIT_TAIL], *nxttail[RCU_NEXT_READY_TAIL]):
132 * Entries known to have arrived before current GP ended
133 * [*nxttail[RCU_DONE_TAIL], *nxttail[RCU_WAIT_TAIL]):
134 * Entries that batch # <= ->completed - 1: waiting for current GP
135 * [nxtlist, *nxttail[RCU_DONE_TAIL]):
136 * Entries that batch # <= ->completed
137 * The grace period for these entries has completed, and
138 * the other grace-period-completed entries may be moved
139 * here temporarily in rcu_process_callbacks().
140 */
141 struct rcu_head *nxtlist;
142 struct rcu_head **nxttail[RCU_NEXT_SIZE];
143 long qlen; /* # of queued callbacks */
144 long blimit; /* Upper limit on a processed batch */
145
146#ifdef CONFIG_NO_HZ
147 /* 3) dynticks interface. */
148 struct rcu_dynticks *dynticks; /* Shared per-CPU dynticks state. */
149 int dynticks_snap; /* Per-GP tracking for dynticks. */
150 int dynticks_nmi_snap; /* Per-GP tracking for dynticks_nmi. */
151#endif /* #ifdef CONFIG_NO_HZ */
152
153 /* 4) reasons this CPU needed to be kicked by force_quiescent_state */
154#ifdef CONFIG_NO_HZ
155 unsigned long dynticks_fqs; /* Kicked due to dynticks idle. */
156#endif /* #ifdef CONFIG_NO_HZ */
157 unsigned long offline_fqs; /* Kicked due to being offline. */
158 unsigned long resched_ipi; /* Sent a resched IPI. */
159
160 /* 5) __rcu_pending() statistics. */
161 long n_rcu_pending; /* rcu_pending() calls since boot. */
162 long n_rp_qs_pending;
163 long n_rp_cb_ready;
164 long n_rp_cpu_needs_gp;
165 long n_rp_gp_completed;
166 long n_rp_gp_started;
167 long n_rp_need_fqs;
168 long n_rp_need_nothing;
169
170 int cpu;
171};
172
173/* Values for signaled field in struct rcu_state. */
174#define RCU_GP_INIT 0 /* Grace period being initialized. */
175#define RCU_SAVE_DYNTICK 1 /* Need to scan dyntick state. */
176#define RCU_FORCE_QS 2 /* Need to force quiescent state. */
177#ifdef CONFIG_NO_HZ
178#define RCU_SIGNAL_INIT RCU_SAVE_DYNTICK
179#else /* #ifdef CONFIG_NO_HZ */
180#define RCU_SIGNAL_INIT RCU_FORCE_QS
181#endif /* #else #ifdef CONFIG_NO_HZ */
182
183#define RCU_JIFFIES_TILL_FORCE_QS 3 /* for rsp->jiffies_force_qs */
184#ifdef CONFIG_RCU_CPU_STALL_DETECTOR
185#define RCU_SECONDS_TILL_STALL_CHECK (10 * HZ) /* for rsp->jiffies_stall */
186#define RCU_SECONDS_TILL_STALL_RECHECK (30 * HZ) /* for rsp->jiffies_stall */
187#define RCU_STALL_RAT_DELAY 2 /* Allow other CPUs time */
188 /* to take at least one */
189 /* scheduling clock irq */
190 /* before ratting on them. */
191
192#endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
193
194/*
195 * RCU global state, including node hierarchy. This hierarchy is
196 * represented in "heap" form in a dense array. The root (first level)
197 * of the hierarchy is in ->node[0] (referenced by ->level[0]), the second
198 * level in ->node[1] through ->node[m] (->node[1] referenced by ->level[1]),
199 * and the third level in ->node[m+1] and following (->node[m+1] referenced
200 * by ->level[2]). The number of levels is determined by the number of
201 * CPUs and by CONFIG_RCU_FANOUT. Small systems will have a "hierarchy"
202 * consisting of a single rcu_node.
203 */
204struct rcu_state {
205 struct rcu_node node[NUM_RCU_NODES]; /* Hierarchy. */
206 struct rcu_node *level[NUM_RCU_LVLS]; /* Hierarchy levels. */
207 u32 levelcnt[MAX_RCU_LVLS + 1]; /* # nodes in each level. */
208 u8 levelspread[NUM_RCU_LVLS]; /* kids/node in each level. */
209 struct rcu_data *rda[NR_CPUS]; /* array of rdp pointers. */
210
211 /* The following fields are guarded by the root rcu_node's lock. */
212
213 u8 signaled ____cacheline_internodealigned_in_smp;
214 /* Force QS state. */
215 long gpnum; /* Current gp number. */
216 long completed; /* # of last completed gp. */
217 spinlock_t onofflock; /* exclude on/offline and */
218 /* starting new GP. */
219 spinlock_t fqslock; /* Only one task forcing */
220 /* quiescent states. */
221 unsigned long jiffies_force_qs; /* Time at which to invoke */
222 /* force_quiescent_state(). */
223 unsigned long n_force_qs; /* Number of calls to */
224 /* force_quiescent_state(). */
225 unsigned long n_force_qs_lh; /* ~Number of calls leaving */
226 /* due to lock unavailable. */
227 unsigned long n_force_qs_ngp; /* Number of calls leaving */
228 /* due to no GP active. */
229#ifdef CONFIG_RCU_CPU_STALL_DETECTOR
230 unsigned long gp_start; /* Time at which GP started, */
231 /* but in jiffies. */
232 unsigned long jiffies_stall; /* Time at which to check */
233 /* for CPU stalls. */
234#endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
235#ifdef CONFIG_NO_HZ
236 long dynticks_completed; /* Value of completed @ snap. */
237#endif /* #ifdef CONFIG_NO_HZ */
238};
239
240#ifdef RCU_TREE_NONCORE
6258c4fb
IM
241
242/*
243 * RCU implementation internal declarations:
244 */
d6714c22
PM
245extern struct rcu_state rcu_sched_state;
246DECLARE_PER_CPU(struct rcu_data, rcu_sched_data);
6258c4fb
IM
247
248extern struct rcu_state rcu_bh_state;
249DECLARE_PER_CPU(struct rcu_data, rcu_bh_data);
250
f41d911f
PM
251#ifdef CONFIG_TREE_PREEMPT_RCU
252extern struct rcu_state rcu_preempt_state;
253DECLARE_PER_CPU(struct rcu_data, rcu_preempt_data);
254#endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */
255
9f77da9f
PM
256#endif /* #ifdef RCU_TREE_NONCORE */
257