xref: /qemu/util/rcu.c (revision 2abf0da2)
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 QEMU_DEFINE_CO_TLS(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         QLIST_FOREACH(index, &registry, node) {
87             qatomic_set(&index->waiting, true);
88         }
89 
90         /* Here, order the stores to index->waiting before the loads of
91          * index->ctr.  Pairs with smp_mb_placeholder() in rcu_read_unlock(),
92          * ensuring that the loads of index->ctr are sequentially consistent.
93          *
94          * If this is the last iteration, this barrier also prevents
95          * frees from seeping upwards, and orders the two wait phases
96          * on architectures with 32-bit longs; see synchronize_rcu().
97          */
98         smp_mb_global();
99 
100         QLIST_FOREACH_SAFE(index, &registry, node, tmp) {
101             if (!rcu_gp_ongoing(&index->ctr)) {
102                 QLIST_REMOVE(index, node);
103                 QLIST_INSERT_HEAD(&qsreaders, index, node);
104 
105                 /* No need for memory barriers here, worst of all we
106                  * get some extra futex wakeups.
107                  */
108                 qatomic_set(&index->waiting, false);
109             } else if (qatomic_read(&in_drain_call_rcu)) {
110                 notifier_list_notify(&index->force_rcu, NULL);
111             }
112         }
113 
114         if (QLIST_EMPTY(&registry)) {
115             break;
116         }
117 
118         /* Wait for one thread to report a quiescent state and try again.
119          * Release rcu_registry_lock, so rcu_(un)register_thread() doesn't
120          * wait too much time.
121          *
122          * rcu_register_thread() may add nodes to &registry; it will not
123          * wake up synchronize_rcu, but that is okay because at least another
124          * thread must exit its RCU read-side critical section before
125          * synchronize_rcu is done.  The next iteration of the loop will
126          * move the new thread's rcu_reader from &registry to &qsreaders,
127          * because rcu_gp_ongoing() will return false.
128          *
129          * rcu_unregister_thread() may remove nodes from &qsreaders instead
130          * of &registry if it runs during qemu_event_wait.  That's okay;
131          * the node then will not be added back to &registry by QLIST_SWAP
132          * below.  The invariant is that the node is part of one list when
133          * rcu_registry_lock is released.
134          */
135         qemu_mutex_unlock(&rcu_registry_lock);
136         qemu_event_wait(&rcu_gp_event);
137         qemu_mutex_lock(&rcu_registry_lock);
138     }
139 
140     /* put back the reader list in the registry */
141     QLIST_SWAP(&registry, &qsreaders, node);
142 }
143 
144 void synchronize_rcu(void)
145 {
146     QEMU_LOCK_GUARD(&rcu_sync_lock);
147 
148     /* Write RCU-protected pointers before reading p_rcu_reader->ctr.
149      * Pairs with smp_mb_placeholder() in rcu_read_lock().
150      *
151      * Also orders write to RCU-protected pointers before
152      * write to rcu_gp_ctr.
153      */
154     smp_mb_global();
155 
156     QEMU_LOCK_GUARD(&rcu_registry_lock);
157     if (!QLIST_EMPTY(&registry)) {
158         if (sizeof(rcu_gp_ctr) < 8) {
159             /* For architectures with 32-bit longs, a two-subphases algorithm
160              * ensures we do not encounter overflow bugs.
161              *
162              * Switch parity: 0 -> 1, 1 -> 0.
163              */
164             qatomic_set(&rcu_gp_ctr, rcu_gp_ctr ^ RCU_GP_CTR);
165             wait_for_readers();
166             qatomic_set(&rcu_gp_ctr, rcu_gp_ctr ^ RCU_GP_CTR);
167         } else {
168             /* Increment current grace period.  */
169             qatomic_set(&rcu_gp_ctr, rcu_gp_ctr + RCU_GP_CTR);
170         }
171 
172         wait_for_readers();
173     }
174 }
175 
176 
177 #define RCU_CALL_MIN_SIZE        30
178 
179 /* Multi-producer, single-consumer queue based on urcu/static/wfqueue.h
180  * from liburcu.  Note that head is only used by the consumer.
181  */
182 static struct rcu_head dummy;
183 static struct rcu_head *head = &dummy, **tail = &dummy.next;
184 static int rcu_call_count;
185 static QemuEvent rcu_call_ready_event;
186 
187 static void enqueue(struct rcu_head *node)
188 {
189     struct rcu_head **old_tail;
190 
191     node->next = NULL;
192 
193     /*
194      * Make this node the tail of the list.  The node will be
195      * used by further enqueue operations, but it will not
196      * be dequeued yet...
197      */
198     old_tail = qatomic_xchg(&tail, &node->next);
199 
200     /*
201      * ... until it is pointed to from another item in the list.
202      * In the meantime, try_dequeue() will find a NULL next pointer
203      * and loop.
204      *
205      * Synchronizes with qatomic_load_acquire() in try_dequeue().
206      */
207     qatomic_store_release(old_tail, node);
208 }
209 
210 static struct rcu_head *try_dequeue(void)
211 {
212     struct rcu_head *node, *next;
213 
214 retry:
215     /* Head is only written by this thread, so no need for barriers.  */
216     node = head;
217 
218     /*
219      * If the head node has NULL in its next pointer, the value is
220      * wrong and we need to wait until its enqueuer finishes the update.
221      */
222     next = qatomic_load_acquire(&node->next);
223     if (!next) {
224         return NULL;
225     }
226 
227     /*
228      * Test for an empty list, which we do not expect.  Note that for
229      * the consumer head and tail are always consistent.  The head
230      * is consistent because only the consumer reads/writes it.
231      * The tail, because it is the first step in the enqueuing.
232      * It is only the next pointers that might be inconsistent.
233      */
234     if (head == &dummy && qatomic_read(&tail) == &dummy.next) {
235         abort();
236     }
237 
238     /*
239      * Since we are the sole consumer, and we excluded the empty case
240      * above, the queue will always have at least two nodes: the
241      * dummy node, and the one being removed.  So we do not need to update
242      * the tail pointer.
243      */
244     head = next;
245 
246     /* If we dequeued the dummy node, add it back at the end and retry.  */
247     if (node == &dummy) {
248         enqueue(node);
249         goto retry;
250     }
251 
252     return node;
253 }
254 
255 static void *call_rcu_thread(void *opaque)
256 {
257     struct rcu_head *node;
258 
259     rcu_register_thread();
260 
261     for (;;) {
262         int tries = 0;
263         int n = qatomic_read(&rcu_call_count);
264 
265         /* Heuristically wait for a decent number of callbacks to pile up.
266          * Fetch rcu_call_count now, we only must process elements that were
267          * added before synchronize_rcu() starts.
268          */
269         while (n == 0 || (n < RCU_CALL_MIN_SIZE && ++tries <= 5)) {
270             g_usleep(10000);
271             if (n == 0) {
272                 qemu_event_reset(&rcu_call_ready_event);
273                 n = qatomic_read(&rcu_call_count);
274                 if (n == 0) {
275 #if defined(CONFIG_MALLOC_TRIM)
276                     malloc_trim(4 * 1024 * 1024);
277 #endif
278                     qemu_event_wait(&rcu_call_ready_event);
279                 }
280             }
281             n = qatomic_read(&rcu_call_count);
282         }
283 
284         qatomic_sub(&rcu_call_count, n);
285         synchronize_rcu();
286         bql_lock();
287         while (n > 0) {
288             node = try_dequeue();
289             while (!node) {
290                 bql_unlock();
291                 qemu_event_reset(&rcu_call_ready_event);
292                 node = try_dequeue();
293                 if (!node) {
294                     qemu_event_wait(&rcu_call_ready_event);
295                     node = try_dequeue();
296                 }
297                 bql_lock();
298             }
299 
300             n--;
301             node->func(node);
302         }
303         bql_unlock();
304     }
305     abort();
306 }
307 
308 void call_rcu1(struct rcu_head *node, void (*func)(struct rcu_head *node))
309 {
310     node->func = func;
311     enqueue(node);
312     qatomic_inc(&rcu_call_count);
313     qemu_event_set(&rcu_call_ready_event);
314 }
315 
316 
317 struct rcu_drain {
318     struct rcu_head rcu;
319     QemuEvent drain_complete_event;
320 };
321 
322 static void drain_rcu_callback(struct rcu_head *node)
323 {
324     struct rcu_drain *event = (struct rcu_drain *)node;
325     qemu_event_set(&event->drain_complete_event);
326 }
327 
328 /*
329  * This function ensures that all pending RCU callbacks
330  * on the current thread are done executing
331 
332  * drops big qemu lock during the wait to allow RCU thread
333  * to process the callbacks
334  *
335  */
336 
337 void drain_call_rcu(void)
338 {
339     struct rcu_drain rcu_drain;
340     bool locked = bql_locked();
341 
342     memset(&rcu_drain, 0, sizeof(struct rcu_drain));
343     qemu_event_init(&rcu_drain.drain_complete_event, false);
344 
345     if (locked) {
346         bql_unlock();
347     }
348 
349 
350     /*
351      * RCU callbacks are invoked in the same order as in which they
352      * are registered, thus we can be sure that when 'drain_rcu_callback'
353      * is called, all RCU callbacks that were registered on this thread
354      * prior to calling this function are completed.
355      *
356      * Note that since we have only one global queue of the RCU callbacks,
357      * we also end up waiting for most of RCU callbacks that were registered
358      * on the other threads, but this is a side effect that shouldn't be
359      * assumed.
360      */
361 
362     qatomic_inc(&in_drain_call_rcu);
363     call_rcu1(&rcu_drain.rcu, drain_rcu_callback);
364     qemu_event_wait(&rcu_drain.drain_complete_event);
365     qatomic_dec(&in_drain_call_rcu);
366 
367     if (locked) {
368         bql_lock();
369     }
370 
371 }
372 
373 void rcu_register_thread(void)
374 {
375     assert(get_ptr_rcu_reader()->ctr == 0);
376     qemu_mutex_lock(&rcu_registry_lock);
377     QLIST_INSERT_HEAD(&registry, get_ptr_rcu_reader(), node);
378     qemu_mutex_unlock(&rcu_registry_lock);
379 }
380 
381 void rcu_unregister_thread(void)
382 {
383     qemu_mutex_lock(&rcu_registry_lock);
384     QLIST_REMOVE(get_ptr_rcu_reader(), node);
385     qemu_mutex_unlock(&rcu_registry_lock);
386 }
387 
388 void rcu_add_force_rcu_notifier(Notifier *n)
389 {
390     qemu_mutex_lock(&rcu_registry_lock);
391     notifier_list_add(&get_ptr_rcu_reader()->force_rcu, n);
392     qemu_mutex_unlock(&rcu_registry_lock);
393 }
394 
395 void rcu_remove_force_rcu_notifier(Notifier *n)
396 {
397     qemu_mutex_lock(&rcu_registry_lock);
398     notifier_remove(n);
399     qemu_mutex_unlock(&rcu_registry_lock);
400 }
401 
402 static void rcu_init_complete(void)
403 {
404     QemuThread thread;
405 
406     qemu_mutex_init(&rcu_registry_lock);
407     qemu_mutex_init(&rcu_sync_lock);
408     qemu_event_init(&rcu_gp_event, true);
409 
410     qemu_event_init(&rcu_call_ready_event, false);
411 
412     /* The caller is assumed to have BQL, so the call_rcu thread
413      * must have been quiescent even after forking, just recreate it.
414      */
415     qemu_thread_create(&thread, "call_rcu", call_rcu_thread,
416                        NULL, QEMU_THREAD_DETACHED);
417 
418     rcu_register_thread();
419 }
420 
421 static int atfork_depth = 1;
422 
423 void rcu_enable_atfork(void)
424 {
425     atfork_depth++;
426 }
427 
428 void rcu_disable_atfork(void)
429 {
430     atfork_depth--;
431 }
432 
433 #ifdef CONFIG_POSIX
434 static void rcu_init_lock(void)
435 {
436     if (atfork_depth < 1) {
437         return;
438     }
439 
440     qemu_mutex_lock(&rcu_sync_lock);
441     qemu_mutex_lock(&rcu_registry_lock);
442 }
443 
444 static void rcu_init_unlock(void)
445 {
446     if (atfork_depth < 1) {
447         return;
448     }
449 
450     qemu_mutex_unlock(&rcu_registry_lock);
451     qemu_mutex_unlock(&rcu_sync_lock);
452 }
453 
454 static void rcu_init_child(void)
455 {
456     if (atfork_depth < 1) {
457         return;
458     }
459 
460     memset(&registry, 0, sizeof(registry));
461     rcu_init_complete();
462 }
463 #endif
464 
465 static void __attribute__((__constructor__)) rcu_init(void)
466 {
467     smp_mb_global_init();
468 #ifdef CONFIG_POSIX
469     pthread_atfork(rcu_init_lock, rcu_init_unlock, rcu_init_child);
470 #endif
471     rcu_init_complete();
472 }
473