xref: /openbmc/qemu/util/aio-win32.c (revision e8d1e0cd)
1 /*
2  * QEMU aio implementation
3  *
4  * Copyright IBM Corp., 2008
5  * Copyright Red Hat Inc., 2012
6  *
7  * Authors:
8  *  Anthony Liguori   <aliguori@us.ibm.com>
9  *  Paolo Bonzini     <pbonzini@redhat.com>
10  *
11  * This work is licensed under the terms of the GNU GPL, version 2.  See
12  * the COPYING file in the top-level directory.
13  *
14  * Contributions after 2012-01-13 are licensed under the terms of the
15  * GNU GPL, version 2 or (at your option) any later version.
16  */
17 
18 #include "qemu/osdep.h"
19 #include "block/block.h"
20 #include "qemu/main-loop.h"
21 #include "qemu/queue.h"
22 #include "qemu/sockets.h"
23 #include "qapi/error.h"
24 #include "qemu/rcu_queue.h"
25 #include "qemu/error-report.h"
26 
27 struct AioHandler {
28     EventNotifier *e;
29     IOHandler *io_read;
30     IOHandler *io_write;
31     EventNotifierHandler *io_notify;
32     GPollFD pfd;
33     int deleted;
34     void *opaque;
35     bool is_external;
36     QLIST_ENTRY(AioHandler) node;
37 };
38 
39 static void aio_remove_fd_handler(AioContext *ctx, AioHandler *node)
40 {
41     /*
42      * If the GSource is in the process of being destroyed then
43      * g_source_remove_poll() causes an assertion failure.  Skip
44      * removal in that case, because glib cleans up its state during
45      * destruction anyway.
46      */
47     if (!g_source_is_destroyed(&ctx->source)) {
48         g_source_remove_poll(&ctx->source, &node->pfd);
49     }
50 
51     /* If aio_poll is in progress, just mark the node as deleted */
52     if (qemu_lockcnt_count(&ctx->list_lock)) {
53         node->deleted = 1;
54         node->pfd.revents = 0;
55     } else {
56         /* Otherwise, delete it for real.  We can't just mark it as
57          * deleted because deleted nodes are only cleaned up after
58          * releasing the list_lock.
59          */
60         QLIST_REMOVE(node, node);
61         g_free(node);
62     }
63 }
64 
65 void aio_set_fd_handler(AioContext *ctx,
66                         int fd,
67                         bool is_external,
68                         IOHandler *io_read,
69                         IOHandler *io_write,
70                         AioPollFn *io_poll,
71                         IOHandler *io_poll_ready,
72                         void *opaque)
73 {
74     AioHandler *old_node;
75     AioHandler *node = NULL;
76     SOCKET s;
77 
78     if (!fd_is_socket(fd)) {
79         error_report("fd=%d is not a socket, AIO implementation is missing", fd);
80         return;
81     }
82 
83     s = _get_osfhandle(fd);
84 
85     qemu_lockcnt_lock(&ctx->list_lock);
86     QLIST_FOREACH(old_node, &ctx->aio_handlers, node) {
87         if (old_node->pfd.fd == s && !old_node->deleted) {
88             break;
89         }
90     }
91 
92     if (io_read || io_write) {
93         HANDLE event;
94         long bitmask = 0;
95 
96         /* Alloc and insert if it's not already there */
97         node = g_new0(AioHandler, 1);
98         node->pfd.fd = s;
99 
100         node->pfd.events = 0;
101         if (node->io_read) {
102             node->pfd.events |= G_IO_IN;
103         }
104         if (node->io_write) {
105             node->pfd.events |= G_IO_OUT;
106         }
107 
108         node->e = &ctx->notifier;
109 
110         /* Update handler with latest information */
111         node->opaque = opaque;
112         node->io_read = io_read;
113         node->io_write = io_write;
114         node->is_external = is_external;
115 
116         if (io_read) {
117             bitmask |= FD_READ | FD_ACCEPT | FD_CLOSE;
118         }
119 
120         if (io_write) {
121             bitmask |= FD_WRITE | FD_CONNECT;
122         }
123 
124         QLIST_INSERT_HEAD_RCU(&ctx->aio_handlers, node, node);
125         event = event_notifier_get_handle(&ctx->notifier);
126         qemu_socket_select(fd, event, bitmask, NULL);
127     }
128     if (old_node) {
129         aio_remove_fd_handler(ctx, old_node);
130     }
131 
132     qemu_lockcnt_unlock(&ctx->list_lock);
133     aio_notify(ctx);
134 }
135 
136 void aio_set_event_notifier(AioContext *ctx,
137                             EventNotifier *e,
138                             bool is_external,
139                             EventNotifierHandler *io_notify,
140                             AioPollFn *io_poll,
141                             EventNotifierHandler *io_poll_ready)
142 {
143     AioHandler *node;
144 
145     qemu_lockcnt_lock(&ctx->list_lock);
146     QLIST_FOREACH(node, &ctx->aio_handlers, node) {
147         if (node->e == e && !node->deleted) {
148             break;
149         }
150     }
151 
152     /* Are we deleting the fd handler? */
153     if (!io_notify) {
154         if (node) {
155             aio_remove_fd_handler(ctx, node);
156         }
157     } else {
158         if (node == NULL) {
159             /* Alloc and insert if it's not already there */
160             node = g_new0(AioHandler, 1);
161             node->e = e;
162             node->pfd.fd = (uintptr_t)event_notifier_get_handle(e);
163             node->pfd.events = G_IO_IN;
164             node->is_external = is_external;
165             QLIST_INSERT_HEAD_RCU(&ctx->aio_handlers, node, node);
166 
167             g_source_add_poll(&ctx->source, &node->pfd);
168         }
169         /* Update handler with latest information */
170         node->io_notify = io_notify;
171     }
172 
173     qemu_lockcnt_unlock(&ctx->list_lock);
174     aio_notify(ctx);
175 }
176 
177 void aio_set_event_notifier_poll(AioContext *ctx,
178                                  EventNotifier *notifier,
179                                  EventNotifierHandler *io_poll_begin,
180                                  EventNotifierHandler *io_poll_end)
181 {
182     /* Not implemented */
183 }
184 
185 bool aio_prepare(AioContext *ctx)
186 {
187     static struct timeval tv0;
188     AioHandler *node;
189     bool have_select_revents = false;
190     fd_set rfds, wfds;
191 
192     /*
193      * We have to walk very carefully in case aio_set_fd_handler is
194      * called while we're walking.
195      */
196     qemu_lockcnt_inc(&ctx->list_lock);
197 
198     /* fill fd sets */
199     FD_ZERO(&rfds);
200     FD_ZERO(&wfds);
201     QLIST_FOREACH_RCU(node, &ctx->aio_handlers, node) {
202         if (node->io_read) {
203             FD_SET ((SOCKET)node->pfd.fd, &rfds);
204         }
205         if (node->io_write) {
206             FD_SET ((SOCKET)node->pfd.fd, &wfds);
207         }
208     }
209 
210     if (select(0, &rfds, &wfds, NULL, &tv0) > 0) {
211         QLIST_FOREACH_RCU(node, &ctx->aio_handlers, node) {
212             node->pfd.revents = 0;
213             if (FD_ISSET(node->pfd.fd, &rfds)) {
214                 node->pfd.revents |= G_IO_IN;
215                 have_select_revents = true;
216             }
217 
218             if (FD_ISSET(node->pfd.fd, &wfds)) {
219                 node->pfd.revents |= G_IO_OUT;
220                 have_select_revents = true;
221             }
222         }
223     }
224 
225     qemu_lockcnt_dec(&ctx->list_lock);
226     return have_select_revents;
227 }
228 
229 bool aio_pending(AioContext *ctx)
230 {
231     AioHandler *node;
232     bool result = false;
233 
234     /*
235      * We have to walk very carefully in case aio_set_fd_handler is
236      * called while we're walking.
237      */
238     qemu_lockcnt_inc(&ctx->list_lock);
239     QLIST_FOREACH_RCU(node, &ctx->aio_handlers, node) {
240         if (node->pfd.revents && node->io_notify) {
241             result = true;
242             break;
243         }
244 
245         if ((node->pfd.revents & G_IO_IN) && node->io_read) {
246             result = true;
247             break;
248         }
249         if ((node->pfd.revents & G_IO_OUT) && node->io_write) {
250             result = true;
251             break;
252         }
253     }
254 
255     qemu_lockcnt_dec(&ctx->list_lock);
256     return result;
257 }
258 
259 static bool aio_dispatch_handlers(AioContext *ctx, HANDLE event)
260 {
261     AioHandler *node;
262     bool progress = false;
263     AioHandler *tmp;
264 
265     /*
266      * We have to walk very carefully in case aio_set_fd_handler is
267      * called while we're walking.
268      */
269     QLIST_FOREACH_SAFE_RCU(node, &ctx->aio_handlers, node, tmp) {
270         int revents = node->pfd.revents;
271 
272         if (!node->deleted &&
273             (revents || event_notifier_get_handle(node->e) == event) &&
274             node->io_notify) {
275             node->pfd.revents = 0;
276             node->io_notify(node->e);
277 
278             /* aio_notify() does not count as progress */
279             if (node->e != &ctx->notifier) {
280                 progress = true;
281             }
282         }
283 
284         if (!node->deleted &&
285             (node->io_read || node->io_write)) {
286             node->pfd.revents = 0;
287             if ((revents & G_IO_IN) && node->io_read) {
288                 node->io_read(node->opaque);
289                 progress = true;
290             }
291             if ((revents & G_IO_OUT) && node->io_write) {
292                 node->io_write(node->opaque);
293                 progress = true;
294             }
295 
296             /* if the next select() will return an event, we have progressed */
297             if (event == event_notifier_get_handle(&ctx->notifier)) {
298                 WSANETWORKEVENTS ev;
299                 WSAEnumNetworkEvents(node->pfd.fd, event, &ev);
300                 if (ev.lNetworkEvents) {
301                     progress = true;
302                 }
303             }
304         }
305 
306         if (node->deleted) {
307             if (qemu_lockcnt_dec_if_lock(&ctx->list_lock)) {
308                 QLIST_REMOVE(node, node);
309                 g_free(node);
310                 qemu_lockcnt_inc_and_unlock(&ctx->list_lock);
311             }
312         }
313     }
314 
315     return progress;
316 }
317 
318 void aio_dispatch(AioContext *ctx)
319 {
320     qemu_lockcnt_inc(&ctx->list_lock);
321     aio_bh_poll(ctx);
322     aio_dispatch_handlers(ctx, INVALID_HANDLE_VALUE);
323     qemu_lockcnt_dec(&ctx->list_lock);
324     timerlistgroup_run_timers(&ctx->tlg);
325 }
326 
327 bool aio_poll(AioContext *ctx, bool blocking)
328 {
329     AioHandler *node;
330     HANDLE events[MAXIMUM_WAIT_OBJECTS];
331     bool progress, have_select_revents, first;
332     unsigned count;
333     int timeout;
334 
335     /*
336      * There cannot be two concurrent aio_poll calls for the same AioContext (or
337      * an aio_poll concurrent with a GSource prepare/check/dispatch callback).
338      * We rely on this below to avoid slow locked accesses to ctx->notify_me.
339      *
340      * aio_poll() may only be called in the AioContext's thread. iohandler_ctx
341      * is special in that it runs in the main thread, but that thread's context
342      * is qemu_aio_context.
343      */
344     assert(in_aio_context_home_thread(ctx == iohandler_get_aio_context() ?
345                                       qemu_get_aio_context() : ctx));
346     progress = false;
347 
348     /* aio_notify can avoid the expensive event_notifier_set if
349      * everything (file descriptors, bottom halves, timers) will
350      * be re-evaluated before the next blocking poll().  This is
351      * already true when aio_poll is called with blocking == false;
352      * if blocking == true, it is only true after poll() returns,
353      * so disable the optimization now.
354      */
355     if (blocking) {
356         qatomic_set(&ctx->notify_me, qatomic_read(&ctx->notify_me) + 2);
357         /*
358          * Write ctx->notify_me before computing the timeout
359          * (reading bottom half flags, etc.).  Pairs with
360          * smp_mb in aio_notify().
361          */
362         smp_mb();
363     }
364 
365     qemu_lockcnt_inc(&ctx->list_lock);
366     have_select_revents = aio_prepare(ctx);
367 
368     /* fill fd sets */
369     count = 0;
370     QLIST_FOREACH_RCU(node, &ctx->aio_handlers, node) {
371         if (!node->deleted && node->io_notify
372             && aio_node_check(ctx, node->is_external)) {
373             assert(count < MAXIMUM_WAIT_OBJECTS);
374             events[count++] = event_notifier_get_handle(node->e);
375         }
376     }
377 
378     first = true;
379 
380     /* ctx->notifier is always registered.  */
381     assert(count > 0);
382 
383     /* Multiple iterations, all of them non-blocking except the first,
384      * may be necessary to process all pending events.  After the first
385      * WaitForMultipleObjects call ctx->notify_me will be decremented.
386      */
387     do {
388         HANDLE event;
389         int ret;
390 
391         timeout = blocking && !have_select_revents
392             ? qemu_timeout_ns_to_ms(aio_compute_timeout(ctx)) : 0;
393         ret = WaitForMultipleObjects(count, events, FALSE, timeout);
394         if (blocking) {
395             assert(first);
396             qatomic_store_release(&ctx->notify_me,
397                                   qatomic_read(&ctx->notify_me) - 2);
398             aio_notify_accept(ctx);
399         }
400 
401         if (first) {
402             progress |= aio_bh_poll(ctx);
403             first = false;
404         }
405 
406         /* if we have any signaled events, dispatch event */
407         event = NULL;
408         if ((DWORD) (ret - WAIT_OBJECT_0) < count) {
409             event = events[ret - WAIT_OBJECT_0];
410             events[ret - WAIT_OBJECT_0] = events[--count];
411         } else if (!have_select_revents) {
412             break;
413         }
414 
415         have_select_revents = false;
416         blocking = false;
417 
418         progress |= aio_dispatch_handlers(ctx, event);
419     } while (count > 0);
420 
421     qemu_lockcnt_dec(&ctx->list_lock);
422 
423     progress |= timerlistgroup_run_timers(&ctx->tlg);
424     return progress;
425 }
426 
427 void aio_context_setup(AioContext *ctx)
428 {
429 }
430 
431 void aio_context_destroy(AioContext *ctx)
432 {
433 }
434 
435 void aio_context_use_g_source(AioContext *ctx)
436 {
437 }
438 
439 void aio_context_set_poll_params(AioContext *ctx, int64_t max_ns,
440                                  int64_t grow, int64_t shrink, Error **errp)
441 {
442     if (max_ns) {
443         error_setg(errp, "AioContext polling is not implemented on Windows");
444     }
445 }
446 
447 void aio_context_set_aio_params(AioContext *ctx, int64_t max_batch,
448                                 Error **errp)
449 {
450 }
451