Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mason/btrfs...
[linux-ginger.git] / kernel / rcutree.h
blob8e8287a983c2b3714336c2b8eacfc4595323c9df
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 has not been tested, so there is probably some
35 * bug somewhere.
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)
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 */
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)
68 * Dynticks per-CPU state.
70 struct 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. */
79 * Definition for node within the RCU grace-period-detection hierarchy.
81 struct rcu_node {
82 spinlock_t lock;
83 long gpnum; /* Current grace period for this node. */
84 /* This will either be equal to or one */
85 /* behind the root rcu_node's gpnum. */
86 unsigned long qsmask; /* CPUs or groups that need to switch in */
87 /* order for current grace period to proceed.*/
88 unsigned long qsmaskinit;
89 /* Per-GP initialization for qsmask. */
90 unsigned long grpmask; /* Mask to apply to parent qsmask. */
91 int grplo; /* lowest-numbered CPU or group here. */
92 int grphi; /* highest-numbered CPU or group here. */
93 u8 grpnum; /* CPU/group number for next level up. */
94 u8 level; /* root is at level 0. */
95 struct rcu_node *parent;
96 struct list_head blocked_tasks[2];
97 /* Tasks blocked in RCU read-side critsect. */
98 } ____cacheline_internodealigned_in_smp;
100 /* Index values for nxttail array in struct rcu_data. */
101 #define RCU_DONE_TAIL 0 /* Also RCU_WAIT head. */
102 #define RCU_WAIT_TAIL 1 /* Also RCU_NEXT_READY head. */
103 #define RCU_NEXT_READY_TAIL 2 /* Also RCU_NEXT head. */
104 #define RCU_NEXT_TAIL 3
105 #define RCU_NEXT_SIZE 4
107 /* Per-CPU data for read-copy update. */
108 struct rcu_data {
109 /* 1) quiescent-state and grace-period handling : */
110 long completed; /* Track rsp->completed gp number */
111 /* in order to detect GP end. */
112 long gpnum; /* Highest gp number that this CPU */
113 /* is aware of having started. */
114 long passed_quiesc_completed;
115 /* Value of completed at time of qs. */
116 bool passed_quiesc; /* User-mode/idle loop etc. */
117 bool qs_pending; /* Core waits for quiesc state. */
118 bool beenonline; /* CPU online at least once. */
119 bool preemptable; /* Preemptable RCU? */
120 struct rcu_node *mynode; /* This CPU's leaf of hierarchy */
121 unsigned long grpmask; /* Mask to apply to leaf qsmask. */
123 /* 2) batch handling */
125 * If nxtlist is not NULL, it is partitioned as follows.
126 * Any of the partitions might be empty, in which case the
127 * pointer to that partition will be equal to the pointer for
128 * the following partition. When the list is empty, all of
129 * the nxttail elements point to nxtlist, which is NULL.
131 * [*nxttail[RCU_NEXT_READY_TAIL], NULL = *nxttail[RCU_NEXT_TAIL]):
132 * Entries that might have arrived after current GP ended
133 * [*nxttail[RCU_WAIT_TAIL], *nxttail[RCU_NEXT_READY_TAIL]):
134 * Entries known to have arrived before current GP ended
135 * [*nxttail[RCU_DONE_TAIL], *nxttail[RCU_WAIT_TAIL]):
136 * Entries that batch # <= ->completed - 1: waiting for current GP
137 * [nxtlist, *nxttail[RCU_DONE_TAIL]):
138 * Entries that batch # <= ->completed
139 * The grace period for these entries has completed, and
140 * the other grace-period-completed entries may be moved
141 * here temporarily in rcu_process_callbacks().
143 struct rcu_head *nxtlist;
144 struct rcu_head **nxttail[RCU_NEXT_SIZE];
145 long qlen; /* # of queued callbacks */
146 long blimit; /* Upper limit on a processed batch */
148 #ifdef CONFIG_NO_HZ
149 /* 3) dynticks interface. */
150 struct rcu_dynticks *dynticks; /* Shared per-CPU dynticks state. */
151 int dynticks_snap; /* Per-GP tracking for dynticks. */
152 int dynticks_nmi_snap; /* Per-GP tracking for dynticks_nmi. */
153 #endif /* #ifdef CONFIG_NO_HZ */
155 /* 4) reasons this CPU needed to be kicked by force_quiescent_state */
156 #ifdef CONFIG_NO_HZ
157 unsigned long dynticks_fqs; /* Kicked due to dynticks idle. */
158 #endif /* #ifdef CONFIG_NO_HZ */
159 unsigned long offline_fqs; /* Kicked due to being offline. */
160 unsigned long resched_ipi; /* Sent a resched IPI. */
162 /* 5) __rcu_pending() statistics. */
163 long n_rcu_pending; /* rcu_pending() calls since boot. */
164 long n_rp_qs_pending;
165 long n_rp_cb_ready;
166 long n_rp_cpu_needs_gp;
167 long n_rp_gp_completed;
168 long n_rp_gp_started;
169 long n_rp_need_fqs;
170 long n_rp_need_nothing;
172 int cpu;
175 /* Values for signaled field in struct rcu_state. */
176 #define RCU_GP_INIT 0 /* Grace period being initialized. */
177 #define RCU_SAVE_DYNTICK 1 /* Need to scan dyntick state. */
178 #define RCU_FORCE_QS 2 /* Need to force quiescent state. */
179 #ifdef CONFIG_NO_HZ
180 #define RCU_SIGNAL_INIT RCU_SAVE_DYNTICK
181 #else /* #ifdef CONFIG_NO_HZ */
182 #define RCU_SIGNAL_INIT RCU_FORCE_QS
183 #endif /* #else #ifdef CONFIG_NO_HZ */
185 #define RCU_JIFFIES_TILL_FORCE_QS 3 /* for rsp->jiffies_force_qs */
186 #ifdef CONFIG_RCU_CPU_STALL_DETECTOR
187 #define RCU_SECONDS_TILL_STALL_CHECK (10 * HZ) /* for rsp->jiffies_stall */
188 #define RCU_SECONDS_TILL_STALL_RECHECK (30 * HZ) /* for rsp->jiffies_stall */
189 #define RCU_STALL_RAT_DELAY 2 /* Allow other CPUs time */
190 /* to take at least one */
191 /* scheduling clock irq */
192 /* before ratting on them. */
194 #endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
197 * RCU global state, including node hierarchy. This hierarchy is
198 * represented in "heap" form in a dense array. The root (first level)
199 * of the hierarchy is in ->node[0] (referenced by ->level[0]), the second
200 * level in ->node[1] through ->node[m] (->node[1] referenced by ->level[1]),
201 * and the third level in ->node[m+1] and following (->node[m+1] referenced
202 * by ->level[2]). The number of levels is determined by the number of
203 * CPUs and by CONFIG_RCU_FANOUT. Small systems will have a "hierarchy"
204 * consisting of a single rcu_node.
206 struct rcu_state {
207 struct rcu_node node[NUM_RCU_NODES]; /* Hierarchy. */
208 struct rcu_node *level[NUM_RCU_LVLS]; /* Hierarchy levels. */
209 u32 levelcnt[MAX_RCU_LVLS + 1]; /* # nodes in each level. */
210 u8 levelspread[NUM_RCU_LVLS]; /* kids/node in each level. */
211 struct rcu_data *rda[NR_CPUS]; /* array of rdp pointers. */
213 /* The following fields are guarded by the root rcu_node's lock. */
215 u8 signaled ____cacheline_internodealigned_in_smp;
216 /* Force QS state. */
217 long gpnum; /* Current gp number. */
218 long completed; /* # of last completed gp. */
219 spinlock_t onofflock; /* exclude on/offline and */
220 /* starting new GP. */
221 spinlock_t fqslock; /* Only one task forcing */
222 /* quiescent states. */
223 unsigned long jiffies_force_qs; /* Time at which to invoke */
224 /* force_quiescent_state(). */
225 unsigned long n_force_qs; /* Number of calls to */
226 /* force_quiescent_state(). */
227 unsigned long n_force_qs_lh; /* ~Number of calls leaving */
228 /* due to lock unavailable. */
229 unsigned long n_force_qs_ngp; /* Number of calls leaving */
230 /* due to no GP active. */
231 #ifdef CONFIG_RCU_CPU_STALL_DETECTOR
232 unsigned long gp_start; /* Time at which GP started, */
233 /* but in jiffies. */
234 unsigned long jiffies_stall; /* Time at which to check */
235 /* for CPU stalls. */
236 #endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
237 #ifdef CONFIG_NO_HZ
238 long dynticks_completed; /* Value of completed @ snap. */
239 #endif /* #ifdef CONFIG_NO_HZ */
242 #ifdef RCU_TREE_NONCORE
245 * RCU implementation internal declarations:
247 extern struct rcu_state rcu_sched_state;
248 DECLARE_PER_CPU(struct rcu_data, rcu_sched_data);
250 extern struct rcu_state rcu_bh_state;
251 DECLARE_PER_CPU(struct rcu_data, rcu_bh_data);
253 #ifdef CONFIG_TREE_PREEMPT_RCU
254 extern struct rcu_state rcu_preempt_state;
255 DECLARE_PER_CPU(struct rcu_data, rcu_preempt_data);
256 #endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */
258 #endif /* #ifdef RCU_TREE_NONCORE */