xref: /qemu/util/rcu.c (revision b83a80e8)
1 /*
2  * urcu-mb.c
3  *
4  * Userspace RCU library with explicit memory barriers
5  *
6  * Copyright (c) 2009 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
7  * Copyright (c) 2009 Paul E. McKenney, IBM Corporation.
8  * Copyright 2015 Red Hat, Inc.
9  *
10  * Ported to QEMU by Paolo Bonzini  <pbonzini@redhat.com>
11  *
12  * This library is free software; you can redistribute it and/or
13  * modify it under the terms of the GNU Lesser General Public
14  * License as published by the Free Software Foundation; either
15  * version 2.1 of the License, or (at your option) any later version.
16  *
17  * This library is distributed in the hope that it will be useful,
18  * but WITHOUT ANY WARRANTY; without even the implied warranty of
19  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
20  * Lesser General Public License for more details.
21  *
22  * You should have received a copy of the GNU Lesser General Public
23  * License along with this library; if not, write to the Free Software
24  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
25  *
26  * IBM's contributions to this file may be relicensed under LGPLv2 or later.
27  */
28 
29 #include "qemu/osdep.h"
30 #include "qemu/rcu.h"
31 #include "qemu/atomic.h"
32 #include "qemu/thread.h"
33 #include "qemu/main-loop.h"
34 #include "qemu/lockable.h"
35 #if defined(CONFIG_MALLOC_TRIM)
36 #include <malloc.h>
37 #endif
38 
39 /*
40  * Global grace period counter.  Bit 0 is always one in rcu_gp_ctr.
41  * Bits 1 and above are defined in synchronize_rcu.
42  */
43 #define RCU_GP_LOCKED           (1UL << 0)
44 #define RCU_GP_CTR              (1UL << 1)
45 
46 unsigned long rcu_gp_ctr = RCU_GP_LOCKED;
47 
48 QemuEvent rcu_gp_event;
49 static int in_drain_call_rcu;
50 static QemuMutex rcu_registry_lock;
51 static QemuMutex rcu_sync_lock;
52 
53 /*
54  * Check whether a quiescent state was crossed between the beginning of
55  * update_counter_and_wait and now.
56  */
57 static inline int rcu_gp_ongoing(unsigned long *ctr)
58 {
59     unsigned long v;
60 
61     v = qatomic_read(ctr);
62     return v && (v != rcu_gp_ctr);
63 }
64 
65 /* Written to only by each individual reader. Read by both the reader and the
66  * writers.
67  */
68 __thread struct rcu_reader_data rcu_reader;
69 
70 /* Protected by rcu_registry_lock.  */
71 typedef QLIST_HEAD(, rcu_reader_data) ThreadList;
72 static ThreadList registry = QLIST_HEAD_INITIALIZER(registry);
73 
74 /* Wait for previous parity/grace period to be empty of readers.  */
75 static void wait_for_readers(void)
76 {
77     ThreadList qsreaders = QLIST_HEAD_INITIALIZER(qsreaders);
78     struct rcu_reader_data *index, *tmp;
79 
80     for (;;) {
81         /* We want to be notified of changes made to rcu_gp_ongoing
82          * while we walk the list.
83          */
84         qemu_event_reset(&rcu_gp_event);
85 
86         /* Instead of using qatomic_mb_set for index->waiting, and
87          * qatomic_mb_read for index->ctr, memory barriers are placed
88          * manually since writes to different threads are independent.
89          * qemu_event_reset has acquire semantics, so no memory barrier
90          * is needed here.
91          */
92         QLIST_FOREACH(index, &registry, node) {
93             qatomic_set(&index->waiting, true);
94         }
95 
96         /* Here, order the stores to index->waiting before the loads of
97          * index->ctr.  Pairs with smp_mb_placeholder() in rcu_read_unlock(),
98          * ensuring that the loads of index->ctr are sequentially consistent.
99          */
100         smp_mb_global();
101 
102         QLIST_FOREACH_SAFE(index, &registry, node, tmp) {
103             if (!rcu_gp_ongoing(&index->ctr)) {
104                 QLIST_REMOVE(index, node);
105                 QLIST_INSERT_HEAD(&qsreaders, index, node);
106 
107                 /* No need for mb_set here, worst of all we
108                  * get some extra futex wakeups.
109                  */
110                 qatomic_set(&index->waiting, false);
111             } else if (qatomic_read(&in_drain_call_rcu)) {
112                 notifier_list_notify(&index->force_rcu, NULL);
113             }
114         }
115 
116         if (QLIST_EMPTY(&registry)) {
117             break;
118         }
119 
120         /* Wait for one thread to report a quiescent state and try again.
121          * Release rcu_registry_lock, so rcu_(un)register_thread() doesn't
122          * wait too much time.
123          *
124          * rcu_register_thread() may add nodes to &registry; it will not
125          * wake up synchronize_rcu, but that is okay because at least another
126          * thread must exit its RCU read-side critical section before
127          * synchronize_rcu is done.  The next iteration of the loop will
128          * move the new thread's rcu_reader from &registry to &qsreaders,
129          * because rcu_gp_ongoing() will return false.
130          *
131          * rcu_unregister_thread() may remove nodes from &qsreaders instead
132          * of &registry if it runs during qemu_event_wait.  That's okay;
133          * the node then will not be added back to &registry by QLIST_SWAP
134          * below.  The invariant is that the node is part of one list when
135          * rcu_registry_lock is released.
136          */
137         qemu_mutex_unlock(&rcu_registry_lock);
138         qemu_event_wait(&rcu_gp_event);
139         qemu_mutex_lock(&rcu_registry_lock);
140     }
141 
142     /* put back the reader list in the registry */
143     QLIST_SWAP(&registry, &qsreaders, node);
144 }
145 
146 void synchronize_rcu(void)
147 {
148     QEMU_LOCK_GUARD(&rcu_sync_lock);
149 
150     /* Write RCU-protected pointers before reading p_rcu_reader->ctr.
151      * Pairs with smp_mb_placeholder() in rcu_read_lock().
152      */
153     smp_mb_global();
154 
155     QEMU_LOCK_GUARD(&rcu_registry_lock);
156     if (!QLIST_EMPTY(&registry)) {
157         /* In either case, the qatomic_mb_set below blocks stores that free
158          * old RCU-protected pointers.
159          */
160         if (sizeof(rcu_gp_ctr) < 8) {
161             /* For architectures with 32-bit longs, a two-subphases algorithm
162              * ensures we do not encounter overflow bugs.
163              *
164              * Switch parity: 0 -> 1, 1 -> 0.
165              */
166             qatomic_mb_set(&rcu_gp_ctr, rcu_gp_ctr ^ RCU_GP_CTR);
167             wait_for_readers();
168             qatomic_mb_set(&rcu_gp_ctr, rcu_gp_ctr ^ RCU_GP_CTR);
169         } else {
170             /* Increment current grace period.  */
171             qatomic_mb_set(&rcu_gp_ctr, rcu_gp_ctr + RCU_GP_CTR);
172         }
173 
174         wait_for_readers();
175     }
176 }
177 
178 
179 #define RCU_CALL_MIN_SIZE        30
180 
181 /* Multi-producer, single-consumer queue based on urcu/static/wfqueue.h
182  * from liburcu.  Note that head is only used by the consumer.
183  */
184 static struct rcu_head dummy;
185 static struct rcu_head *head = &dummy, **tail = &dummy.next;
186 static int rcu_call_count;
187 static QemuEvent rcu_call_ready_event;
188 
189 static void enqueue(struct rcu_head *node)
190 {
191     struct rcu_head **old_tail;
192 
193     node->next = NULL;
194     old_tail = qatomic_xchg(&tail, &node->next);
195     qatomic_mb_set(old_tail, node);
196 }
197 
198 static struct rcu_head *try_dequeue(void)
199 {
200     struct rcu_head *node, *next;
201 
202 retry:
203     /* Test for an empty list, which we do not expect.  Note that for
204      * the consumer head and tail are always consistent.  The head
205      * is consistent because only the consumer reads/writes it.
206      * The tail, because it is the first step in the enqueuing.
207      * It is only the next pointers that might be inconsistent.
208      */
209     if (head == &dummy && qatomic_mb_read(&tail) == &dummy.next) {
210         abort();
211     }
212 
213     /* If the head node has NULL in its next pointer, the value is
214      * wrong and we need to wait until its enqueuer finishes the update.
215      */
216     node = head;
217     next = qatomic_mb_read(&head->next);
218     if (!next) {
219         return NULL;
220     }
221 
222     /* Since we are the sole consumer, and we excluded the empty case
223      * above, the queue will always have at least two nodes: the
224      * dummy node, and the one being removed.  So we do not need to update
225      * the tail pointer.
226      */
227     head = next;
228 
229     /* If we dequeued the dummy node, add it back at the end and retry.  */
230     if (node == &dummy) {
231         enqueue(node);
232         goto retry;
233     }
234 
235     return node;
236 }
237 
238 static void *call_rcu_thread(void *opaque)
239 {
240     struct rcu_head *node;
241 
242     rcu_register_thread();
243 
244     for (;;) {
245         int tries = 0;
246         int n = qatomic_read(&rcu_call_count);
247 
248         /* Heuristically wait for a decent number of callbacks to pile up.
249          * Fetch rcu_call_count now, we only must process elements that were
250          * added before synchronize_rcu() starts.
251          */
252         while (n == 0 || (n < RCU_CALL_MIN_SIZE && ++tries <= 5)) {
253             g_usleep(10000);
254             if (n == 0) {
255                 qemu_event_reset(&rcu_call_ready_event);
256                 n = qatomic_read(&rcu_call_count);
257                 if (n == 0) {
258 #if defined(CONFIG_MALLOC_TRIM)
259                     malloc_trim(4 * 1024 * 1024);
260 #endif
261                     qemu_event_wait(&rcu_call_ready_event);
262                 }
263             }
264             n = qatomic_read(&rcu_call_count);
265         }
266 
267         qatomic_sub(&rcu_call_count, n);
268         synchronize_rcu();
269         qemu_mutex_lock_iothread();
270         while (n > 0) {
271             node = try_dequeue();
272             while (!node) {
273                 qemu_mutex_unlock_iothread();
274                 qemu_event_reset(&rcu_call_ready_event);
275                 node = try_dequeue();
276                 if (!node) {
277                     qemu_event_wait(&rcu_call_ready_event);
278                     node = try_dequeue();
279                 }
280                 qemu_mutex_lock_iothread();
281             }
282 
283             n--;
284             node->func(node);
285         }
286         qemu_mutex_unlock_iothread();
287     }
288     abort();
289 }
290 
291 void call_rcu1(struct rcu_head *node, void (*func)(struct rcu_head *node))
292 {
293     node->func = func;
294     enqueue(node);
295     qatomic_inc(&rcu_call_count);
296     qemu_event_set(&rcu_call_ready_event);
297 }
298 
299 
300 struct rcu_drain {
301     struct rcu_head rcu;
302     QemuEvent drain_complete_event;
303 };
304 
305 static void drain_rcu_callback(struct rcu_head *node)
306 {
307     struct rcu_drain *event = (struct rcu_drain *)node;
308     qemu_event_set(&event->drain_complete_event);
309 }
310 
311 /*
312  * This function ensures that all pending RCU callbacks
313  * on the current thread are done executing
314 
315  * drops big qemu lock during the wait to allow RCU thread
316  * to process the callbacks
317  *
318  */
319 
320 void drain_call_rcu(void)
321 {
322     struct rcu_drain rcu_drain;
323     bool locked = qemu_mutex_iothread_locked();
324 
325     memset(&rcu_drain, 0, sizeof(struct rcu_drain));
326     qemu_event_init(&rcu_drain.drain_complete_event, false);
327 
328     if (locked) {
329         qemu_mutex_unlock_iothread();
330     }
331 
332 
333     /*
334      * RCU callbacks are invoked in the same order as in which they
335      * are registered, thus we can be sure that when 'drain_rcu_callback'
336      * is called, all RCU callbacks that were registered on this thread
337      * prior to calling this function are completed.
338      *
339      * Note that since we have only one global queue of the RCU callbacks,
340      * we also end up waiting for most of RCU callbacks that were registered
341      * on the other threads, but this is a side effect that shoudn't be
342      * assumed.
343      */
344 
345     qatomic_inc(&in_drain_call_rcu);
346     call_rcu1(&rcu_drain.rcu, drain_rcu_callback);
347     qemu_event_wait(&rcu_drain.drain_complete_event);
348     qatomic_dec(&in_drain_call_rcu);
349 
350     if (locked) {
351         qemu_mutex_lock_iothread();
352     }
353 
354 }
355 
356 void rcu_register_thread(void)
357 {
358     assert(rcu_reader.ctr == 0);
359     qemu_mutex_lock(&rcu_registry_lock);
360     QLIST_INSERT_HEAD(&registry, &rcu_reader, node);
361     qemu_mutex_unlock(&rcu_registry_lock);
362 }
363 
364 void rcu_unregister_thread(void)
365 {
366     qemu_mutex_lock(&rcu_registry_lock);
367     QLIST_REMOVE(&rcu_reader, node);
368     qemu_mutex_unlock(&rcu_registry_lock);
369 }
370 
371 void rcu_add_force_rcu_notifier(Notifier *n)
372 {
373     qemu_mutex_lock(&rcu_registry_lock);
374     notifier_list_add(&rcu_reader.force_rcu, n);
375     qemu_mutex_unlock(&rcu_registry_lock);
376 }
377 
378 void rcu_remove_force_rcu_notifier(Notifier *n)
379 {
380     qemu_mutex_lock(&rcu_registry_lock);
381     notifier_remove(n);
382     qemu_mutex_unlock(&rcu_registry_lock);
383 }
384 
385 static void rcu_init_complete(void)
386 {
387     QemuThread thread;
388 
389     qemu_mutex_init(&rcu_registry_lock);
390     qemu_mutex_init(&rcu_sync_lock);
391     qemu_event_init(&rcu_gp_event, true);
392 
393     qemu_event_init(&rcu_call_ready_event, false);
394 
395     /* The caller is assumed to have iothread lock, so the call_rcu thread
396      * must have been quiescent even after forking, just recreate it.
397      */
398     qemu_thread_create(&thread, "call_rcu", call_rcu_thread,
399                        NULL, QEMU_THREAD_DETACHED);
400 
401     rcu_register_thread();
402 }
403 
404 static int atfork_depth = 1;
405 
406 void rcu_enable_atfork(void)
407 {
408     atfork_depth++;
409 }
410 
411 void rcu_disable_atfork(void)
412 {
413     atfork_depth--;
414 }
415 
416 #ifdef CONFIG_POSIX
417 static void rcu_init_lock(void)
418 {
419     if (atfork_depth < 1) {
420         return;
421     }
422 
423     qemu_mutex_lock(&rcu_sync_lock);
424     qemu_mutex_lock(&rcu_registry_lock);
425 }
426 
427 static void rcu_init_unlock(void)
428 {
429     if (atfork_depth < 1) {
430         return;
431     }
432 
433     qemu_mutex_unlock(&rcu_registry_lock);
434     qemu_mutex_unlock(&rcu_sync_lock);
435 }
436 
437 static void rcu_init_child(void)
438 {
439     if (atfork_depth < 1) {
440         return;
441     }
442 
443     memset(&registry, 0, sizeof(registry));
444     rcu_init_complete();
445 }
446 #endif
447 
448 static void __attribute__((__constructor__)) rcu_init(void)
449 {
450     smp_mb_global_init();
451 #ifdef CONFIG_POSIX
452     pthread_atfork(rcu_init_lock, rcu_init_unlock, rcu_init_child);
453 #endif
454     rcu_init_complete();
455 }
456