2 * Read-Copy Update mechanism for mutual exclusion (tree-based version)
3 * Internal non-public definitions that provide either classic
4 * or preemptable semantics.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
20 * Copyright Red Hat, 2009
21 * Copyright IBM Corporation, 2009
23 * Author: Ingo Molnar <mingo@elte.hu>
24 * Paul E. McKenney <paulmck@linux.vnet.ibm.com>
28 #ifdef CONFIG_TREE_PREEMPT_RCU
30 struct rcu_state rcu_preempt_state
= RCU_STATE_INITIALIZER(rcu_preempt_state
);
31 DEFINE_PER_CPU(struct rcu_data
, rcu_preempt_data
);
34 * Tell them what RCU they are running.
36 static inline void rcu_bootup_announce(void)
39 "Experimental preemptable hierarchical RCU implementation.\n");
43 * Return the number of RCU-preempt batches processed thus far
44 * for debug and statistics.
46 long rcu_batches_completed_preempt(void)
48 return rcu_preempt_state
.completed
;
50 EXPORT_SYMBOL_GPL(rcu_batches_completed_preempt
);
53 * Return the number of RCU batches processed thus far for debug & stats.
55 long rcu_batches_completed(void)
57 return rcu_batches_completed_preempt();
59 EXPORT_SYMBOL_GPL(rcu_batches_completed
);
62 * Record a preemptable-RCU quiescent state for the specified CPU. Note
63 * that this just means that the task currently running on the CPU is
64 * not in a quiescent state. There might be any number of tasks blocked
65 * while in an RCU read-side critical section.
67 static void rcu_preempt_qs(int cpu
)
69 struct rcu_data
*rdp
= &per_cpu(rcu_preempt_data
, cpu
);
70 rdp
->passed_quiesc_completed
= rdp
->completed
;
72 rdp
->passed_quiesc
= 1;
76 * We have entered the scheduler, and the current task might soon be
77 * context-switched away from. If this task is in an RCU read-side
78 * critical section, we will no longer be able to rely on the CPU to
79 * record that fact, so we enqueue the task on the appropriate entry
80 * of the blocked_tasks[] array. The task will dequeue itself when
81 * it exits the outermost enclosing RCU read-side critical section.
82 * Therefore, the current grace period cannot be permitted to complete
83 * until the blocked_tasks[] entry indexed by the low-order bit of
86 * Caller must disable preemption.
88 static void rcu_preempt_note_context_switch(int cpu
)
90 struct task_struct
*t
= current
;
96 if (t
->rcu_read_lock_nesting
&&
97 (t
->rcu_read_unlock_special
& RCU_READ_UNLOCK_BLOCKED
) == 0) {
99 /* Possibly blocking in an RCU read-side critical section. */
100 rdp
= rcu_preempt_state
.rda
[cpu
];
102 spin_lock_irqsave(&rnp
->lock
, flags
);
103 t
->rcu_read_unlock_special
|= RCU_READ_UNLOCK_BLOCKED
;
104 t
->rcu_blocked_node
= rnp
;
107 * If this CPU has already checked in, then this task
108 * will hold up the next grace period rather than the
109 * current grace period. Queue the task accordingly.
110 * If the task is queued for the current grace period
111 * (i.e., this CPU has not yet passed through a quiescent
112 * state for the current grace period), then as long
113 * as that task remains queued, the current grace period
116 * But first, note that the current CPU must still be
119 WARN_ON_ONCE((rdp
->grpmask
& rnp
->qsmaskinit
) == 0);
120 WARN_ON_ONCE(!list_empty(&t
->rcu_node_entry
));
121 phase
= (rnp
->gpnum
+ !(rnp
->qsmask
& rdp
->grpmask
)) & 0x1;
122 list_add(&t
->rcu_node_entry
, &rnp
->blocked_tasks
[phase
]);
123 spin_unlock_irqrestore(&rnp
->lock
, flags
);
127 * Either we were not in an RCU read-side critical section to
128 * begin with, or we have now recorded that critical section
129 * globally. Either way, we can now note a quiescent state
130 * for this CPU. Again, if we were in an RCU read-side critical
131 * section, and if that critical section was blocking the current
132 * grace period, then the fact that the task has been enqueued
133 * means that we continue to block the current grace period.
136 local_irq_save(flags
);
137 t
->rcu_read_unlock_special
&= ~RCU_READ_UNLOCK_NEED_QS
;
138 local_irq_restore(flags
);
142 * Tree-preemptable RCU implementation for rcu_read_lock().
143 * Just increment ->rcu_read_lock_nesting, shared state will be updated
146 void __rcu_read_lock(void)
148 ACCESS_ONCE(current
->rcu_read_lock_nesting
)++;
149 barrier(); /* needed if we ever invoke rcu_read_lock in rcutree.c */
151 EXPORT_SYMBOL_GPL(__rcu_read_lock
);
153 static void rcu_read_unlock_special(struct task_struct
*t
)
158 struct rcu_node
*rnp
;
161 /* NMI handlers cannot block and cannot safely manipulate state. */
165 local_irq_save(flags
);
168 * If RCU core is waiting for this CPU to exit critical section,
169 * let it know that we have done so.
171 special
= t
->rcu_read_unlock_special
;
172 if (special
& RCU_READ_UNLOCK_NEED_QS
) {
173 t
->rcu_read_unlock_special
&= ~RCU_READ_UNLOCK_NEED_QS
;
174 rcu_preempt_qs(smp_processor_id());
177 /* Hardware IRQ handlers cannot block. */
179 local_irq_restore(flags
);
183 /* Clean up if blocked during RCU read-side critical section. */
184 if (special
& RCU_READ_UNLOCK_BLOCKED
) {
185 t
->rcu_read_unlock_special
&= ~RCU_READ_UNLOCK_BLOCKED
;
188 * Remove this task from the list it blocked on. The
189 * task can migrate while we acquire the lock, but at
190 * most one time. So at most two passes through loop.
193 rnp
= t
->rcu_blocked_node
;
194 spin_lock(&rnp
->lock
); /* irqs already disabled. */
195 if (rnp
== t
->rcu_blocked_node
)
197 spin_unlock(&rnp
->lock
); /* irqs remain disabled. */
199 empty
= list_empty(&rnp
->blocked_tasks
[rnp
->gpnum
& 0x1]);
200 list_del_init(&t
->rcu_node_entry
);
201 t
->rcu_blocked_node
= NULL
;
204 * If this was the last task on the current list, and if
205 * we aren't waiting on any CPUs, report the quiescent state.
206 * Note that both cpu_quiet_msk_finish() and cpu_quiet_msk()
207 * drop rnp->lock and restore irq.
209 if (!empty
&& rnp
->qsmask
== 0 &&
210 list_empty(&rnp
->blocked_tasks
[rnp
->gpnum
& 0x1])) {
211 struct rcu_node
*rnp_p
;
213 if (rnp
->parent
== NULL
) {
214 /* Only one rcu_node in the tree. */
215 cpu_quiet_msk_finish(&rcu_preempt_state
, flags
);
218 /* Report up the rest of the hierarchy. */
220 spin_unlock_irqrestore(&rnp
->lock
, flags
);
222 spin_lock_irqsave(&rnp_p
->lock
, flags
);
223 WARN_ON_ONCE(rnp
->qsmask
);
224 cpu_quiet_msk(mask
, &rcu_preempt_state
, rnp_p
, flags
);
227 spin_unlock(&rnp
->lock
);
229 local_irq_restore(flags
);
233 * Tree-preemptable RCU implementation for rcu_read_unlock().
234 * Decrement ->rcu_read_lock_nesting. If the result is zero (outermost
235 * rcu_read_unlock()) and ->rcu_read_unlock_special is non-zero, then
236 * invoke rcu_read_unlock_special() to clean up after a context switch
237 * in an RCU read-side critical section and other special cases.
239 void __rcu_read_unlock(void)
241 struct task_struct
*t
= current
;
243 barrier(); /* needed if we ever invoke rcu_read_unlock in rcutree.c */
244 if (--ACCESS_ONCE(t
->rcu_read_lock_nesting
) == 0 &&
245 unlikely(ACCESS_ONCE(t
->rcu_read_unlock_special
)))
246 rcu_read_unlock_special(t
);
248 EXPORT_SYMBOL_GPL(__rcu_read_unlock
);
250 #ifdef CONFIG_RCU_CPU_STALL_DETECTOR
253 * Scan the current list of tasks blocked within RCU read-side critical
254 * sections, printing out the tid of each.
256 static void rcu_print_task_stall(struct rcu_node
*rnp
)
259 struct list_head
*lp
;
260 int phase
= rnp
->gpnum
& 0x1;
261 struct task_struct
*t
;
263 if (!list_empty(&rnp
->blocked_tasks
[phase
])) {
264 spin_lock_irqsave(&rnp
->lock
, flags
);
265 phase
= rnp
->gpnum
& 0x1; /* re-read under lock. */
266 lp
= &rnp
->blocked_tasks
[phase
];
267 list_for_each_entry(t
, lp
, rcu_node_entry
)
268 printk(" P%d", t
->pid
);
269 spin_unlock_irqrestore(&rnp
->lock
, flags
);
273 #endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
276 * Check that the list of blocked tasks for the newly completed grace
277 * period is in fact empty. It is a serious bug to complete a grace
278 * period that still has RCU readers blocked! This function must be
279 * invoked -before- updating this rnp's ->gpnum, and the rnp's ->lock
280 * must be held by the caller.
282 static void rcu_preempt_check_blocked_tasks(struct rcu_node
*rnp
)
284 WARN_ON_ONCE(!list_empty(&rnp
->blocked_tasks
[rnp
->gpnum
& 0x1]));
285 WARN_ON_ONCE(rnp
->qsmask
);
289 * Check for preempted RCU readers for the specified rcu_node structure.
290 * If the caller needs a reliable answer, it must hold the rcu_node's
293 static int rcu_preempted_readers(struct rcu_node
*rnp
)
295 return !list_empty(&rnp
->blocked_tasks
[rnp
->gpnum
& 0x1]);
298 #ifdef CONFIG_HOTPLUG_CPU
301 * Handle tasklist migration for case in which all CPUs covered by the
302 * specified rcu_node have gone offline. Move them up to the root
303 * rcu_node. The reason for not just moving them to the immediate
304 * parent is to remove the need for rcu_read_unlock_special() to
305 * make more than two attempts to acquire the target rcu_node's lock.
307 * The caller must hold rnp->lock with irqs disabled.
309 static void rcu_preempt_offline_tasks(struct rcu_state
*rsp
,
310 struct rcu_node
*rnp
,
311 struct rcu_data
*rdp
)
314 struct list_head
*lp
;
315 struct list_head
*lp_root
;
316 struct rcu_node
*rnp_root
= rcu_get_root(rsp
);
317 struct task_struct
*tp
;
319 if (rnp
== rnp_root
) {
320 WARN_ONCE(1, "Last CPU thought to be offlined?");
321 return; /* Shouldn't happen: at least one CPU online. */
323 WARN_ON_ONCE(rnp
!= rdp
->mynode
&&
324 (!list_empty(&rnp
->blocked_tasks
[0]) ||
325 !list_empty(&rnp
->blocked_tasks
[1])));
328 * Move tasks up to root rcu_node. Rely on the fact that the
329 * root rcu_node can be at most one ahead of the rest of the
330 * rcu_nodes in terms of gp_num value. This fact allows us to
331 * move the blocked_tasks[] array directly, element by element.
333 for (i
= 0; i
< 2; i
++) {
334 lp
= &rnp
->blocked_tasks
[i
];
335 lp_root
= &rnp_root
->blocked_tasks
[i
];
336 while (!list_empty(lp
)) {
337 tp
= list_entry(lp
->next
, typeof(*tp
), rcu_node_entry
);
338 spin_lock(&rnp_root
->lock
); /* irqs already disabled */
339 list_del(&tp
->rcu_node_entry
);
340 tp
->rcu_blocked_node
= rnp_root
;
341 list_add(&tp
->rcu_node_entry
, lp_root
);
342 spin_unlock(&rnp_root
->lock
); /* irqs remain disabled */
348 * Do CPU-offline processing for preemptable RCU.
350 static void rcu_preempt_offline_cpu(int cpu
)
352 __rcu_offline_cpu(cpu
, &rcu_preempt_state
);
355 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
358 * Check for a quiescent state from the current CPU. When a task blocks,
359 * the task is recorded in the corresponding CPU's rcu_node structure,
360 * which is checked elsewhere.
362 * Caller must disable hard irqs.
364 static void rcu_preempt_check_callbacks(int cpu
)
366 struct task_struct
*t
= current
;
368 if (t
->rcu_read_lock_nesting
== 0) {
369 t
->rcu_read_unlock_special
&= ~RCU_READ_UNLOCK_NEED_QS
;
373 if (per_cpu(rcu_preempt_data
, cpu
).qs_pending
)
374 t
->rcu_read_unlock_special
|= RCU_READ_UNLOCK_NEED_QS
;
378 * Process callbacks for preemptable RCU.
380 static void rcu_preempt_process_callbacks(void)
382 __rcu_process_callbacks(&rcu_preempt_state
,
383 &__get_cpu_var(rcu_preempt_data
));
387 * Queue a preemptable-RCU callback for invocation after a grace period.
389 void call_rcu(struct rcu_head
*head
, void (*func
)(struct rcu_head
*rcu
))
391 __call_rcu(head
, func
, &rcu_preempt_state
);
393 EXPORT_SYMBOL_GPL(call_rcu
);
396 * Check to see if there is any immediate preemptable-RCU-related work
399 static int rcu_preempt_pending(int cpu
)
401 return __rcu_pending(&rcu_preempt_state
,
402 &per_cpu(rcu_preempt_data
, cpu
));
406 * Does preemptable RCU need the CPU to stay out of dynticks mode?
408 static int rcu_preempt_needs_cpu(int cpu
)
410 return !!per_cpu(rcu_preempt_data
, cpu
).nxtlist
;
414 * Initialize preemptable RCU's per-CPU data.
416 static void __cpuinit
rcu_preempt_init_percpu_data(int cpu
)
418 rcu_init_percpu_data(cpu
, &rcu_preempt_state
, 1);
422 * Check for a task exiting while in a preemptable-RCU read-side
423 * critical section, clean up if so. No need to issue warnings,
424 * as debug_check_no_locks_held() already does this if lockdep
429 struct task_struct
*t
= current
;
431 if (t
->rcu_read_lock_nesting
== 0)
433 t
->rcu_read_lock_nesting
= 1;
437 #else /* #ifdef CONFIG_TREE_PREEMPT_RCU */
440 * Tell them what RCU they are running.
442 static inline void rcu_bootup_announce(void)
444 printk(KERN_INFO
"Hierarchical RCU implementation.\n");
448 * Return the number of RCU batches processed thus far for debug & stats.
450 long rcu_batches_completed(void)
452 return rcu_batches_completed_sched();
454 EXPORT_SYMBOL_GPL(rcu_batches_completed
);
457 * Because preemptable RCU does not exist, we never have to check for
458 * CPUs being in quiescent states.
460 static void rcu_preempt_note_context_switch(int cpu
)
464 #ifdef CONFIG_RCU_CPU_STALL_DETECTOR
467 * Because preemptable RCU does not exist, we never have to check for
468 * tasks blocked within RCU read-side critical sections.
470 static void rcu_print_task_stall(struct rcu_node
*rnp
)
474 #endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
477 * Because there is no preemptable RCU, there can be no readers blocked,
478 * so there is no need to check for blocked tasks. So check only for
479 * bogus qsmask values.
481 static void rcu_preempt_check_blocked_tasks(struct rcu_node
*rnp
)
483 WARN_ON_ONCE(rnp
->qsmask
);
487 * Because preemptable RCU does not exist, there are never any preempted
490 static int rcu_preempted_readers(struct rcu_node
*rnp
)
495 #ifdef CONFIG_HOTPLUG_CPU
498 * Because preemptable RCU does not exist, it never needs to migrate
499 * tasks that were blocked within RCU read-side critical sections.
501 static void rcu_preempt_offline_tasks(struct rcu_state
*rsp
,
502 struct rcu_node
*rnp
,
503 struct rcu_data
*rdp
)
508 * Because preemptable RCU does not exist, it never needs CPU-offline
511 static void rcu_preempt_offline_cpu(int cpu
)
515 #endif /* #ifdef CONFIG_HOTPLUG_CPU */
518 * Because preemptable RCU does not exist, it never has any callbacks
521 void rcu_preempt_check_callbacks(int cpu
)
526 * Because preemptable RCU does not exist, it never has any callbacks
529 void rcu_preempt_process_callbacks(void)
534 * In classic RCU, call_rcu() is just call_rcu_sched().
536 void call_rcu(struct rcu_head
*head
, void (*func
)(struct rcu_head
*rcu
))
538 call_rcu_sched(head
, func
);
540 EXPORT_SYMBOL_GPL(call_rcu
);
543 * Because preemptable RCU does not exist, it never has any work to do.
545 static int rcu_preempt_pending(int cpu
)
551 * Because preemptable RCU does not exist, it never needs any CPU.
553 static int rcu_preempt_needs_cpu(int cpu
)
559 * Because preemptable RCU does not exist, there is no per-CPU
560 * data to initialize.
562 static void __cpuinit
rcu_preempt_init_percpu_data(int cpu
)
566 #endif /* #else #ifdef CONFIG_TREE_PREEMPT_RCU */