xref: /openbmc/qemu/util/async.c (revision dd873966)
1 /*
2  * Data plane event loop
3  *
4  * Copyright (c) 2003-2008 Fabrice Bellard
5  * Copyright (c) 2009-2017 QEMU contributors
6  *
7  * Permission is hereby granted, free of charge, to any person obtaining a copy
8  * of this software and associated documentation files (the "Software"), to deal
9  * in the Software without restriction, including without limitation the rights
10  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
11  * copies of the Software, and to permit persons to whom the Software is
12  * furnished to do so, subject to the following conditions:
13  *
14  * The above copyright notice and this permission notice shall be included in
15  * all copies or substantial portions of the Software.
16  *
17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
23  * THE SOFTWARE.
24  */
25 
26 #include "qemu/osdep.h"
27 #include "qapi/error.h"
28 #include "qemu-common.h"
29 #include "block/aio.h"
30 #include "block/thread-pool.h"
31 #include "qemu/main-loop.h"
32 #include "qemu/atomic.h"
33 #include "block/raw-aio.h"
34 #include "qemu/coroutine_int.h"
35 #include "trace.h"
36 
37 /***********************************************************/
38 /* bottom halves (can be seen as timers which expire ASAP) */
39 
40 struct QEMUBH {
41     AioContext *ctx;
42     QEMUBHFunc *cb;
43     void *opaque;
44     QEMUBH *next;
45     bool scheduled;
46     bool idle;
47     bool deleted;
48 };
49 
50 void aio_bh_schedule_oneshot(AioContext *ctx, QEMUBHFunc *cb, void *opaque)
51 {
52     QEMUBH *bh;
53     bh = g_new(QEMUBH, 1);
54     *bh = (QEMUBH){
55         .ctx = ctx,
56         .cb = cb,
57         .opaque = opaque,
58     };
59     qemu_lockcnt_lock(&ctx->list_lock);
60     bh->next = ctx->first_bh;
61     bh->scheduled = 1;
62     bh->deleted = 1;
63     /* Make sure that the members are ready before putting bh into list */
64     smp_wmb();
65     ctx->first_bh = bh;
66     qemu_lockcnt_unlock(&ctx->list_lock);
67     aio_notify(ctx);
68 }
69 
70 QEMUBH *aio_bh_new(AioContext *ctx, QEMUBHFunc *cb, void *opaque)
71 {
72     QEMUBH *bh;
73     bh = g_new(QEMUBH, 1);
74     *bh = (QEMUBH){
75         .ctx = ctx,
76         .cb = cb,
77         .opaque = opaque,
78     };
79     qemu_lockcnt_lock(&ctx->list_lock);
80     bh->next = ctx->first_bh;
81     /* Make sure that the members are ready before putting bh into list */
82     smp_wmb();
83     ctx->first_bh = bh;
84     qemu_lockcnt_unlock(&ctx->list_lock);
85     return bh;
86 }
87 
88 void aio_bh_call(QEMUBH *bh)
89 {
90     bh->cb(bh->opaque);
91 }
92 
93 /* Multiple occurrences of aio_bh_poll cannot be called concurrently.
94  * The count in ctx->list_lock is incremented before the call, and is
95  * not affected by the call.
96  */
97 int aio_bh_poll(AioContext *ctx)
98 {
99     QEMUBH *bh, **bhp, *next;
100     int ret;
101     bool deleted = false;
102 
103     ret = 0;
104     for (bh = atomic_rcu_read(&ctx->first_bh); bh; bh = next) {
105         next = atomic_rcu_read(&bh->next);
106         /* The atomic_xchg is paired with the one in qemu_bh_schedule.  The
107          * implicit memory barrier ensures that the callback sees all writes
108          * done by the scheduling thread.  It also ensures that the scheduling
109          * thread sees the zero before bh->cb has run, and thus will call
110          * aio_notify again if necessary.
111          */
112         if (atomic_xchg(&bh->scheduled, 0)) {
113             /* Idle BHs don't count as progress */
114             if (!bh->idle) {
115                 ret = 1;
116             }
117             bh->idle = 0;
118             aio_bh_call(bh);
119         }
120         if (bh->deleted) {
121             deleted = true;
122         }
123     }
124 
125     /* remove deleted bhs */
126     if (!deleted) {
127         return ret;
128     }
129 
130     if (qemu_lockcnt_dec_if_lock(&ctx->list_lock)) {
131         bhp = &ctx->first_bh;
132         while (*bhp) {
133             bh = *bhp;
134             if (bh->deleted && !bh->scheduled) {
135                 *bhp = bh->next;
136                 g_free(bh);
137             } else {
138                 bhp = &bh->next;
139             }
140         }
141         qemu_lockcnt_inc_and_unlock(&ctx->list_lock);
142     }
143     return ret;
144 }
145 
146 void qemu_bh_schedule_idle(QEMUBH *bh)
147 {
148     bh->idle = 1;
149     /* Make sure that idle & any writes needed by the callback are done
150      * before the locations are read in the aio_bh_poll.
151      */
152     atomic_mb_set(&bh->scheduled, 1);
153 }
154 
155 void qemu_bh_schedule(QEMUBH *bh)
156 {
157     AioContext *ctx;
158 
159     ctx = bh->ctx;
160     bh->idle = 0;
161     /* The memory barrier implicit in atomic_xchg makes sure that:
162      * 1. idle & any writes needed by the callback are done before the
163      *    locations are read in the aio_bh_poll.
164      * 2. ctx is loaded before scheduled is set and the callback has a chance
165      *    to execute.
166      */
167     if (atomic_xchg(&bh->scheduled, 1) == 0) {
168         aio_notify(ctx);
169     }
170 }
171 
172 
173 /* This func is async.
174  */
175 void qemu_bh_cancel(QEMUBH *bh)
176 {
177     atomic_mb_set(&bh->scheduled, 0);
178 }
179 
180 /* This func is async.The bottom half will do the delete action at the finial
181  * end.
182  */
183 void qemu_bh_delete(QEMUBH *bh)
184 {
185     bh->scheduled = 0;
186     bh->deleted = 1;
187 }
188 
189 int64_t
190 aio_compute_timeout(AioContext *ctx)
191 {
192     int64_t deadline;
193     int timeout = -1;
194     QEMUBH *bh;
195 
196     for (bh = atomic_rcu_read(&ctx->first_bh); bh;
197          bh = atomic_rcu_read(&bh->next)) {
198         if (bh->scheduled) {
199             if (bh->idle) {
200                 /* idle bottom halves will be polled at least
201                  * every 10ms */
202                 timeout = 10000000;
203             } else {
204                 /* non-idle bottom halves will be executed
205                  * immediately */
206                 return 0;
207             }
208         }
209     }
210 
211     deadline = timerlistgroup_deadline_ns(&ctx->tlg);
212     if (deadline == 0) {
213         return 0;
214     } else {
215         return qemu_soonest_timeout(timeout, deadline);
216     }
217 }
218 
219 static gboolean
220 aio_ctx_prepare(GSource *source, gint    *timeout)
221 {
222     AioContext *ctx = (AioContext *) source;
223 
224     atomic_or(&ctx->notify_me, 1);
225 
226     /* We assume there is no timeout already supplied */
227     *timeout = qemu_timeout_ns_to_ms(aio_compute_timeout(ctx));
228 
229     if (aio_prepare(ctx)) {
230         *timeout = 0;
231     }
232 
233     return *timeout == 0;
234 }
235 
236 static gboolean
237 aio_ctx_check(GSource *source)
238 {
239     AioContext *ctx = (AioContext *) source;
240     QEMUBH *bh;
241 
242     atomic_and(&ctx->notify_me, ~1);
243     aio_notify_accept(ctx);
244 
245     for (bh = ctx->first_bh; bh; bh = bh->next) {
246         if (bh->scheduled) {
247             return true;
248         }
249     }
250     return aio_pending(ctx) || (timerlistgroup_deadline_ns(&ctx->tlg) == 0);
251 }
252 
253 static gboolean
254 aio_ctx_dispatch(GSource     *source,
255                  GSourceFunc  callback,
256                  gpointer     user_data)
257 {
258     AioContext *ctx = (AioContext *) source;
259 
260     assert(callback == NULL);
261     aio_dispatch(ctx);
262     return true;
263 }
264 
265 static void
266 aio_ctx_finalize(GSource     *source)
267 {
268     AioContext *ctx = (AioContext *) source;
269 
270     thread_pool_free(ctx->thread_pool);
271 
272 #ifdef CONFIG_LINUX_AIO
273     if (ctx->linux_aio) {
274         laio_detach_aio_context(ctx->linux_aio, ctx);
275         laio_cleanup(ctx->linux_aio);
276         ctx->linux_aio = NULL;
277     }
278 #endif
279 
280     assert(QSLIST_EMPTY(&ctx->scheduled_coroutines));
281     qemu_bh_delete(ctx->co_schedule_bh);
282 
283     qemu_lockcnt_lock(&ctx->list_lock);
284     assert(!qemu_lockcnt_count(&ctx->list_lock));
285     while (ctx->first_bh) {
286         QEMUBH *next = ctx->first_bh->next;
287 
288         /* qemu_bh_delete() must have been called on BHs in this AioContext */
289         assert(ctx->first_bh->deleted);
290 
291         g_free(ctx->first_bh);
292         ctx->first_bh = next;
293     }
294     qemu_lockcnt_unlock(&ctx->list_lock);
295 
296     aio_set_event_notifier(ctx, &ctx->notifier, false, NULL, NULL);
297     event_notifier_cleanup(&ctx->notifier);
298     qemu_rec_mutex_destroy(&ctx->lock);
299     qemu_lockcnt_destroy(&ctx->list_lock);
300     timerlistgroup_deinit(&ctx->tlg);
301 }
302 
303 static GSourceFuncs aio_source_funcs = {
304     aio_ctx_prepare,
305     aio_ctx_check,
306     aio_ctx_dispatch,
307     aio_ctx_finalize
308 };
309 
310 GSource *aio_get_g_source(AioContext *ctx)
311 {
312     g_source_ref(&ctx->source);
313     return &ctx->source;
314 }
315 
316 ThreadPool *aio_get_thread_pool(AioContext *ctx)
317 {
318     if (!ctx->thread_pool) {
319         ctx->thread_pool = thread_pool_new(ctx);
320     }
321     return ctx->thread_pool;
322 }
323 
324 #ifdef CONFIG_LINUX_AIO
325 LinuxAioState *aio_get_linux_aio(AioContext *ctx)
326 {
327     if (!ctx->linux_aio) {
328         ctx->linux_aio = laio_init();
329         laio_attach_aio_context(ctx->linux_aio, ctx);
330     }
331     return ctx->linux_aio;
332 }
333 #endif
334 
335 void aio_notify(AioContext *ctx)
336 {
337     /* Write e.g. bh->scheduled before reading ctx->notify_me.  Pairs
338      * with atomic_or in aio_ctx_prepare or atomic_add in aio_poll.
339      */
340     smp_mb();
341     if (ctx->notify_me) {
342         event_notifier_set(&ctx->notifier);
343         atomic_mb_set(&ctx->notified, true);
344     }
345 }
346 
347 void aio_notify_accept(AioContext *ctx)
348 {
349     if (atomic_xchg(&ctx->notified, false)) {
350         event_notifier_test_and_clear(&ctx->notifier);
351     }
352 }
353 
354 static void aio_timerlist_notify(void *opaque, QEMUClockType type)
355 {
356     aio_notify(opaque);
357 }
358 
359 static void event_notifier_dummy_cb(EventNotifier *e)
360 {
361 }
362 
363 /* Returns true if aio_notify() was called (e.g. a BH was scheduled) */
364 static bool event_notifier_poll(void *opaque)
365 {
366     EventNotifier *e = opaque;
367     AioContext *ctx = container_of(e, AioContext, notifier);
368 
369     return atomic_read(&ctx->notified);
370 }
371 
372 static void co_schedule_bh_cb(void *opaque)
373 {
374     AioContext *ctx = opaque;
375     QSLIST_HEAD(, Coroutine) straight, reversed;
376 
377     QSLIST_MOVE_ATOMIC(&reversed, &ctx->scheduled_coroutines);
378     QSLIST_INIT(&straight);
379 
380     while (!QSLIST_EMPTY(&reversed)) {
381         Coroutine *co = QSLIST_FIRST(&reversed);
382         QSLIST_REMOVE_HEAD(&reversed, co_scheduled_next);
383         QSLIST_INSERT_HEAD(&straight, co, co_scheduled_next);
384     }
385 
386     while (!QSLIST_EMPTY(&straight)) {
387         Coroutine *co = QSLIST_FIRST(&straight);
388         QSLIST_REMOVE_HEAD(&straight, co_scheduled_next);
389         trace_aio_co_schedule_bh_cb(ctx, co);
390         aio_context_acquire(ctx);
391 
392         /* Protected by write barrier in qemu_aio_coroutine_enter */
393         atomic_set(&co->scheduled, NULL);
394         qemu_coroutine_enter(co);
395         aio_context_release(ctx);
396     }
397 }
398 
399 AioContext *aio_context_new(Error **errp)
400 {
401     int ret;
402     AioContext *ctx;
403 
404     ctx = (AioContext *) g_source_new(&aio_source_funcs, sizeof(AioContext));
405     aio_context_setup(ctx);
406 
407     ret = event_notifier_init(&ctx->notifier, false);
408     if (ret < 0) {
409         error_setg_errno(errp, -ret, "Failed to initialize event notifier");
410         goto fail;
411     }
412     g_source_set_can_recurse(&ctx->source, true);
413     qemu_lockcnt_init(&ctx->list_lock);
414 
415     ctx->co_schedule_bh = aio_bh_new(ctx, co_schedule_bh_cb, ctx);
416     QSLIST_INIT(&ctx->scheduled_coroutines);
417 
418     aio_set_event_notifier(ctx, &ctx->notifier,
419                            false,
420                            (EventNotifierHandler *)
421                            event_notifier_dummy_cb,
422                            event_notifier_poll);
423 #ifdef CONFIG_LINUX_AIO
424     ctx->linux_aio = NULL;
425 #endif
426     ctx->thread_pool = NULL;
427     qemu_rec_mutex_init(&ctx->lock);
428     timerlistgroup_init(&ctx->tlg, aio_timerlist_notify, ctx);
429 
430     ctx->poll_ns = 0;
431     ctx->poll_max_ns = 0;
432     ctx->poll_grow = 0;
433     ctx->poll_shrink = 0;
434 
435     return ctx;
436 fail:
437     g_source_destroy(&ctx->source);
438     return NULL;
439 }
440 
441 void aio_co_schedule(AioContext *ctx, Coroutine *co)
442 {
443     trace_aio_co_schedule(ctx, co);
444     const char *scheduled = atomic_cmpxchg(&co->scheduled, NULL,
445                                            __func__);
446 
447     if (scheduled) {
448         fprintf(stderr,
449                 "%s: Co-routine was already scheduled in '%s'\n",
450                 __func__, scheduled);
451         abort();
452     }
453 
454     QSLIST_INSERT_HEAD_ATOMIC(&ctx->scheduled_coroutines,
455                               co, co_scheduled_next);
456     qemu_bh_schedule(ctx->co_schedule_bh);
457 }
458 
459 void aio_co_wake(struct Coroutine *co)
460 {
461     AioContext *ctx;
462 
463     /* Read coroutine before co->ctx.  Matches smp_wmb in
464      * qemu_coroutine_enter.
465      */
466     smp_read_barrier_depends();
467     ctx = atomic_read(&co->ctx);
468 
469     aio_co_enter(ctx, co);
470 }
471 
472 void aio_co_enter(AioContext *ctx, struct Coroutine *co)
473 {
474     if (ctx != qemu_get_current_aio_context()) {
475         aio_co_schedule(ctx, co);
476         return;
477     }
478 
479     if (qemu_in_coroutine()) {
480         Coroutine *self = qemu_coroutine_self();
481         assert(self != co);
482         QSIMPLEQ_INSERT_TAIL(&self->co_queue_wakeup, co, co_queue_next);
483     } else {
484         aio_context_acquire(ctx);
485         qemu_aio_coroutine_enter(ctx, co);
486         aio_context_release(ctx);
487     }
488 }
489 
490 void aio_context_ref(AioContext *ctx)
491 {
492     g_source_ref(&ctx->source);
493 }
494 
495 void aio_context_unref(AioContext *ctx)
496 {
497     g_source_unref(&ctx->source);
498 }
499 
500 void aio_context_acquire(AioContext *ctx)
501 {
502     qemu_rec_mutex_lock(&ctx->lock);
503 }
504 
505 void aio_context_release(AioContext *ctx)
506 {
507     qemu_rec_mutex_unlock(&ctx->lock);
508 }
509