]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - kernel/rcutree.h
5a439c180e69efe0b2fc7a274e3d918eb4f991d2
[mirror_ubuntu-artful-kernel.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 /*
93 * Definition for node within the RCU grace-period-detection hierarchy.
94 */
95 struct rcu_node {
96 raw_spinlock_t lock; /* Root rcu_node's lock protects some */
97 /* rcu_state fields as well as following. */
98 unsigned long gpnum; /* Current grace period for this node. */
99 /* This will either be equal to or one */
100 /* behind the root rcu_node's gpnum. */
101 unsigned long completed; /* Last GP completed for this node. */
102 /* This will either be equal to or one */
103 /* behind the root rcu_node's gpnum. */
104 unsigned long qsmask; /* CPUs or groups that need to switch in */
105 /* order for current grace period to proceed.*/
106 /* In leaf rcu_node, each bit corresponds to */
107 /* an rcu_data structure, otherwise, each */
108 /* bit corresponds to a child rcu_node */
109 /* structure. */
110 unsigned long expmask; /* Groups that have ->blkd_tasks */
111 /* elements that need to drain to allow the */
112 /* current expedited grace period to */
113 /* complete (only for TREE_PREEMPT_RCU). */
114 unsigned long qsmaskinit;
115 /* Per-GP initial value for qsmask & expmask. */
116 unsigned long grpmask; /* Mask to apply to parent qsmask. */
117 /* Only one bit will be set in this mask. */
118 int grplo; /* lowest-numbered CPU or group here. */
119 int grphi; /* highest-numbered CPU or group here. */
120 u8 grpnum; /* CPU/group number for next level up. */
121 u8 level; /* root is at level 0. */
122 struct rcu_node *parent;
123 struct list_head blkd_tasks;
124 /* Tasks blocked in RCU read-side critical */
125 /* section. Tasks are placed at the head */
126 /* of this list and age towards the tail. */
127 struct list_head *gp_tasks;
128 /* Pointer to the first task blocking the */
129 /* current grace period, or NULL if there */
130 /* is no such task. */
131 struct list_head *exp_tasks;
132 /* Pointer to the first task blocking the */
133 /* current expedited grace period, or NULL */
134 /* if there is no such task. If there */
135 /* is no current expedited grace period, */
136 /* then there can cannot be any such task. */
137 } ____cacheline_internodealigned_in_smp;
138
139 /*
140 * Do a full breadth-first scan of the rcu_node structures for the
141 * specified rcu_state structure.
142 */
143 #define rcu_for_each_node_breadth_first(rsp, rnp) \
144 for ((rnp) = &(rsp)->node[0]; \
145 (rnp) < &(rsp)->node[NUM_RCU_NODES]; (rnp)++)
146
147 /*
148 * Do a breadth-first scan of the non-leaf rcu_node structures for the
149 * specified rcu_state structure. Note that if there is a singleton
150 * rcu_node tree with but one rcu_node structure, this loop is a no-op.
151 */
152 #define rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) \
153 for ((rnp) = &(rsp)->node[0]; \
154 (rnp) < (rsp)->level[NUM_RCU_LVLS - 1]; (rnp)++)
155
156 /*
157 * Scan the leaves of the rcu_node hierarchy for the specified rcu_state
158 * structure. Note that if there is a singleton rcu_node tree with but
159 * one rcu_node structure, this loop -will- visit the rcu_node structure.
160 * It is still a leaf node, even if it is also the root node.
161 */
162 #define rcu_for_each_leaf_node(rsp, rnp) \
163 for ((rnp) = (rsp)->level[NUM_RCU_LVLS - 1]; \
164 (rnp) < &(rsp)->node[NUM_RCU_NODES]; (rnp)++)
165
166 /* Index values for nxttail array in struct rcu_data. */
167 #define RCU_DONE_TAIL 0 /* Also RCU_WAIT head. */
168 #define RCU_WAIT_TAIL 1 /* Also RCU_NEXT_READY head. */
169 #define RCU_NEXT_READY_TAIL 2 /* Also RCU_NEXT head. */
170 #define RCU_NEXT_TAIL 3
171 #define RCU_NEXT_SIZE 4
172
173 /* Per-CPU data for read-copy update. */
174 struct rcu_data {
175 /* 1) quiescent-state and grace-period handling : */
176 unsigned long completed; /* Track rsp->completed gp number */
177 /* in order to detect GP end. */
178 unsigned long gpnum; /* Highest gp number that this CPU */
179 /* is aware of having started. */
180 unsigned long passed_quiesc_completed;
181 /* Value of completed at time of qs. */
182 bool passed_quiesc; /* User-mode/idle loop etc. */
183 bool qs_pending; /* Core waits for quiesc state. */
184 bool beenonline; /* CPU online at least once. */
185 bool preemptable; /* Preemptable RCU? */
186 struct rcu_node *mynode; /* This CPU's leaf of hierarchy */
187 unsigned long grpmask; /* Mask to apply to leaf qsmask. */
188
189 /* 2) batch handling */
190 /*
191 * If nxtlist is not NULL, it is partitioned as follows.
192 * Any of the partitions might be empty, in which case the
193 * pointer to that partition will be equal to the pointer for
194 * the following partition. When the list is empty, all of
195 * the nxttail elements point to the ->nxtlist pointer itself,
196 * which in that case is NULL.
197 *
198 * [nxtlist, *nxttail[RCU_DONE_TAIL]):
199 * Entries that batch # <= ->completed
200 * The grace period for these entries has completed, and
201 * the other grace-period-completed entries may be moved
202 * here temporarily in rcu_process_callbacks().
203 * [*nxttail[RCU_DONE_TAIL], *nxttail[RCU_WAIT_TAIL]):
204 * Entries that batch # <= ->completed - 1: waiting for current GP
205 * [*nxttail[RCU_WAIT_TAIL], *nxttail[RCU_NEXT_READY_TAIL]):
206 * Entries known to have arrived before current GP ended
207 * [*nxttail[RCU_NEXT_READY_TAIL], *nxttail[RCU_NEXT_TAIL]):
208 * Entries that might have arrived after current GP ended
209 * Note that the value of *nxttail[RCU_NEXT_TAIL] will
210 * always be NULL, as this is the end of the list.
211 */
212 struct rcu_head *nxtlist;
213 struct rcu_head **nxttail[RCU_NEXT_SIZE];
214 long qlen; /* # of queued callbacks */
215 long qlen_last_fqs_check;
216 /* qlen at last check for QS forcing */
217 unsigned long n_cbs_invoked; /* count of RCU cbs invoked. */
218 unsigned long n_cbs_orphaned; /* RCU cbs orphaned by dying CPU */
219 unsigned long n_cbs_adopted; /* RCU cbs adopted from dying CPU */
220 unsigned long n_force_qs_snap;
221 /* did other CPU force QS recently? */
222 long blimit; /* Upper limit on a processed batch */
223
224 #ifdef CONFIG_NO_HZ
225 /* 3) dynticks interface. */
226 struct rcu_dynticks *dynticks; /* Shared per-CPU dynticks state. */
227 int dynticks_snap; /* Per-GP tracking for dynticks. */
228 #endif /* #ifdef CONFIG_NO_HZ */
229
230 /* 4) reasons this CPU needed to be kicked by force_quiescent_state */
231 #ifdef CONFIG_NO_HZ
232 unsigned long dynticks_fqs; /* Kicked due to dynticks idle. */
233 #endif /* #ifdef CONFIG_NO_HZ */
234 unsigned long offline_fqs; /* Kicked due to being offline. */
235 unsigned long resched_ipi; /* Sent a resched IPI. */
236
237 /* 5) __rcu_pending() statistics. */
238 unsigned long n_rcu_pending; /* rcu_pending() calls since boot. */
239 unsigned long n_rp_qs_pending;
240 unsigned long n_rp_report_qs;
241 unsigned long n_rp_cb_ready;
242 unsigned long n_rp_cpu_needs_gp;
243 unsigned long n_rp_gp_completed;
244 unsigned long n_rp_gp_started;
245 unsigned long n_rp_need_fqs;
246 unsigned long n_rp_need_nothing;
247
248 int cpu;
249 };
250
251 /* Values for signaled field in struct rcu_state. */
252 #define RCU_GP_IDLE 0 /* No grace period in progress. */
253 #define RCU_GP_INIT 1 /* Grace period being initialized. */
254 #define RCU_SAVE_DYNTICK 2 /* Need to scan dyntick state. */
255 #define RCU_FORCE_QS 3 /* Need to force quiescent state. */
256 #ifdef CONFIG_NO_HZ
257 #define RCU_SIGNAL_INIT RCU_SAVE_DYNTICK
258 #else /* #ifdef CONFIG_NO_HZ */
259 #define RCU_SIGNAL_INIT RCU_FORCE_QS
260 #endif /* #else #ifdef CONFIG_NO_HZ */
261
262 #define RCU_JIFFIES_TILL_FORCE_QS 3 /* for rsp->jiffies_force_qs */
263
264 #ifdef CONFIG_PROVE_RCU
265 #define RCU_STALL_DELAY_DELTA (5 * HZ)
266 #else
267 #define RCU_STALL_DELAY_DELTA 0
268 #endif
269
270 #define RCU_SECONDS_TILL_STALL_CHECK (CONFIG_RCU_CPU_STALL_TIMEOUT * HZ + \
271 RCU_STALL_DELAY_DELTA)
272 /* for rsp->jiffies_stall */
273 #define RCU_SECONDS_TILL_STALL_RECHECK (3 * RCU_SECONDS_TILL_STALL_CHECK + 30)
274 /* for rsp->jiffies_stall */
275 #define RCU_STALL_RAT_DELAY 2 /* Allow other CPUs time */
276 /* to take at least one */
277 /* scheduling clock irq */
278 /* before ratting on them. */
279
280
281 /*
282 * RCU global state, including node hierarchy. This hierarchy is
283 * represented in "heap" form in a dense array. The root (first level)
284 * of the hierarchy is in ->node[0] (referenced by ->level[0]), the second
285 * level in ->node[1] through ->node[m] (->node[1] referenced by ->level[1]),
286 * and the third level in ->node[m+1] and following (->node[m+1] referenced
287 * by ->level[2]). The number of levels is determined by the number of
288 * CPUs and by CONFIG_RCU_FANOUT. Small systems will have a "hierarchy"
289 * consisting of a single rcu_node.
290 */
291 struct rcu_state {
292 struct rcu_node node[NUM_RCU_NODES]; /* Hierarchy. */
293 struct rcu_node *level[NUM_RCU_LVLS]; /* Hierarchy levels. */
294 u32 levelcnt[MAX_RCU_LVLS + 1]; /* # nodes in each level. */
295 u8 levelspread[NUM_RCU_LVLS]; /* kids/node in each level. */
296 struct rcu_data __percpu *rda; /* pointer of percu rcu_data. */
297
298 /* The following fields are guarded by the root rcu_node's lock. */
299
300 u8 signaled ____cacheline_internodealigned_in_smp;
301 /* Force QS state. */
302 u8 fqs_active; /* force_quiescent_state() */
303 /* is running. */
304 u8 fqs_need_gp; /* A CPU was prevented from */
305 /* starting a new grace */
306 /* period because */
307 /* force_quiescent_state() */
308 /* was running. */
309 unsigned long gpnum; /* Current gp number. */
310 unsigned long completed; /* # of last completed gp. */
311
312 /* End of fields guarded by root rcu_node's lock. */
313
314 raw_spinlock_t onofflock; /* exclude on/offline and */
315 /* starting new GP. */
316 raw_spinlock_t fqslock; /* Only one task forcing */
317 /* quiescent states. */
318 unsigned long jiffies_force_qs; /* Time at which to invoke */
319 /* force_quiescent_state(). */
320 unsigned long n_force_qs; /* Number of calls to */
321 /* force_quiescent_state(). */
322 unsigned long n_force_qs_lh; /* ~Number of calls leaving */
323 /* due to lock unavailable. */
324 unsigned long n_force_qs_ngp; /* Number of calls leaving */
325 /* due to no GP active. */
326 unsigned long gp_start; /* Time at which GP started, */
327 /* but in jiffies. */
328 unsigned long jiffies_stall; /* Time at which to check */
329 /* for CPU stalls. */
330 char *name; /* Name of structure. */
331 };
332
333 /* Return values for rcu_preempt_offline_tasks(). */
334
335 #define RCU_OFL_TASKS_NORM_GP 0x1 /* Tasks blocking normal */
336 /* GP were moved to root. */
337 #define RCU_OFL_TASKS_EXP_GP 0x2 /* Tasks blocking expedited */
338 /* GP were moved to root. */
339
340 /*
341 * RCU implementation internal declarations:
342 */
343 extern struct rcu_state rcu_sched_state;
344 DECLARE_PER_CPU(struct rcu_data, rcu_sched_data);
345
346 extern struct rcu_state rcu_bh_state;
347 DECLARE_PER_CPU(struct rcu_data, rcu_bh_data);
348
349 #ifdef CONFIG_TREE_PREEMPT_RCU
350 extern struct rcu_state rcu_preempt_state;
351 DECLARE_PER_CPU(struct rcu_data, rcu_preempt_data);
352 #endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */
353
354 #ifndef RCU_TREE_NONCORE
355
356 /* Forward declarations for rcutree_plugin.h */
357 static void rcu_bootup_announce(void);
358 long rcu_batches_completed(void);
359 static void rcu_preempt_note_context_switch(int cpu);
360 static int rcu_preempted_readers(struct rcu_node *rnp);
361 #ifdef CONFIG_HOTPLUG_CPU
362 static void rcu_report_unblock_qs_rnp(struct rcu_node *rnp,
363 unsigned long flags);
364 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
365 static void rcu_print_detail_task_stall(struct rcu_state *rsp);
366 static void rcu_print_task_stall(struct rcu_node *rnp);
367 static void rcu_preempt_stall_reset(void);
368 static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp);
369 #ifdef CONFIG_HOTPLUG_CPU
370 static int rcu_preempt_offline_tasks(struct rcu_state *rsp,
371 struct rcu_node *rnp,
372 struct rcu_data *rdp);
373 static void rcu_preempt_offline_cpu(int cpu);
374 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
375 static void rcu_preempt_check_callbacks(int cpu);
376 static void rcu_preempt_process_callbacks(void);
377 void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu));
378 #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_TREE_PREEMPT_RCU)
379 static void rcu_report_exp_rnp(struct rcu_state *rsp, struct rcu_node *rnp);
380 #endif /* #if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_TREE_PREEMPT_RCU) */
381 static int rcu_preempt_pending(int cpu);
382 static int rcu_preempt_needs_cpu(int cpu);
383 static void __cpuinit rcu_preempt_init_percpu_data(int cpu);
384 static void rcu_preempt_send_cbs_to_online(void);
385 static void __init __rcu_init_preempt(void);
386 static void rcu_needs_cpu_flush(void);
387
388 #endif /* #ifndef RCU_TREE_NONCORE */