jcs.org
qemu with hax to log dma reads & writes
jcs.org/2018/11/12/vfio
1/*
2 * QEMU System Emulator
3 *
4 * Copyright (c) 2003-2008 Fabrice Bellard
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a copy
7 * of this software and associated documentation files (the "Software"), to deal
8 * in the Software without restriction, including without limitation the rights
9 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10 * copies of the Software, and to permit persons to whom the Software is
11 * furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice shall be included in
14 * all copies or substantial portions of the Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22 * THE SOFTWARE.
23 */
24
25#include "qemu/osdep.h"
26#include "qapi/error.h"
27#include "qemu/cutils.h"
28#include "qemu/timer.h"
29#include "sysemu/qtest.h"
30#include "sysemu/cpus.h"
31#include "sysemu/replay.h"
32#include "qemu/main-loop.h"
33#include "block/aio.h"
34#include "qemu/error-report.h"
35#include "qemu/queue.h"
36
37#ifndef _WIN32
38#include <sys/wait.h>
39#endif
40
41#ifndef _WIN32
42
43/* If we have signalfd, we mask out the signals we want to handle and then
44 * use signalfd to listen for them. We rely on whatever the current signal
45 * handler is to dispatch the signals when we receive them.
46 */
47static void sigfd_handler(void *opaque)
48{
49 int fd = (intptr_t)opaque;
50 struct qemu_signalfd_siginfo info;
51 struct sigaction action;
52 ssize_t len;
53
54 while (1) {
55 do {
56 len = read(fd, &info, sizeof(info));
57 } while (len == -1 && errno == EINTR);
58
59 if (len == -1 && errno == EAGAIN) {
60 break;
61 }
62
63 if (len != sizeof(info)) {
64 error_report("read from sigfd returned %zd: %s", len,
65 g_strerror(errno));
66 return;
67 }
68
69 sigaction(info.ssi_signo, NULL, &action);
70 if ((action.sa_flags & SA_SIGINFO) && action.sa_sigaction) {
71 sigaction_invoke(&action, &info);
72 } else if (action.sa_handler) {
73 action.sa_handler(info.ssi_signo);
74 }
75 }
76}
77
78static int qemu_signal_init(Error **errp)
79{
80 int sigfd;
81 sigset_t set;
82
83 /*
84 * SIG_IPI must be blocked in the main thread and must not be caught
85 * by sigwait() in the signal thread. Otherwise, the cpu thread will
86 * not catch it reliably.
87 */
88 sigemptyset(&set);
89 sigaddset(&set, SIG_IPI);
90 sigaddset(&set, SIGIO);
91 sigaddset(&set, SIGALRM);
92 sigaddset(&set, SIGBUS);
93 /* SIGINT cannot be handled via signalfd, so that ^C can be used
94 * to interrupt QEMU when it is being run under gdb. SIGHUP and
95 * SIGTERM are also handled asynchronously, even though it is not
96 * strictly necessary, because they use the same handler as SIGINT.
97 */
98 pthread_sigmask(SIG_BLOCK, &set, NULL);
99
100 sigdelset(&set, SIG_IPI);
101 sigfd = qemu_signalfd(&set);
102 if (sigfd == -1) {
103 error_setg_errno(errp, errno, "failed to create signalfd");
104 return -errno;
105 }
106
107 fcntl_setfl(sigfd, O_NONBLOCK);
108
109 qemu_set_fd_handler(sigfd, sigfd_handler, NULL, (void *)(intptr_t)sigfd);
110
111 return 0;
112}
113
114#else /* _WIN32 */
115
116static int qemu_signal_init(Error **errp)
117{
118 return 0;
119}
120#endif
121
122static AioContext *qemu_aio_context;
123static QEMUBH *qemu_notify_bh;
124
125static void notify_event_cb(void *opaque)
126{
127 /* No need to do anything; this bottom half is only used to
128 * kick the kernel out of ppoll/poll/WaitForMultipleObjects.
129 */
130}
131
132AioContext *qemu_get_aio_context(void)
133{
134 return qemu_aio_context;
135}
136
137void qemu_notify_event(void)
138{
139 if (!qemu_aio_context) {
140 return;
141 }
142 qemu_bh_schedule(qemu_notify_bh);
143}
144
145static GArray *gpollfds;
146
147int qemu_init_main_loop(Error **errp)
148{
149 int ret;
150 GSource *src;
151 Error *local_error = NULL;
152
153 init_clocks(qemu_timer_notify_cb);
154
155 ret = qemu_signal_init(errp);
156 if (ret) {
157 return ret;
158 }
159
160 qemu_aio_context = aio_context_new(&local_error);
161 if (!qemu_aio_context) {
162 error_propagate(errp, local_error);
163 return -EMFILE;
164 }
165 qemu_notify_bh = qemu_bh_new(notify_event_cb, NULL);
166 gpollfds = g_array_new(FALSE, FALSE, sizeof(GPollFD));
167 src = aio_get_g_source(qemu_aio_context);
168 g_source_set_name(src, "aio-context");
169 g_source_attach(src, NULL);
170 g_source_unref(src);
171 src = iohandler_get_g_source();
172 g_source_set_name(src, "io-handler");
173 g_source_attach(src, NULL);
174 g_source_unref(src);
175 return 0;
176}
177
178static int max_priority;
179
180#ifndef _WIN32
181static int glib_pollfds_idx;
182static int glib_n_poll_fds;
183
184static void glib_pollfds_fill(int64_t *cur_timeout)
185{
186 GMainContext *context = g_main_context_default();
187 int timeout = 0;
188 int64_t timeout_ns;
189 int n;
190
191 g_main_context_prepare(context, &max_priority);
192
193 glib_pollfds_idx = gpollfds->len;
194 n = glib_n_poll_fds;
195 do {
196 GPollFD *pfds;
197 glib_n_poll_fds = n;
198 g_array_set_size(gpollfds, glib_pollfds_idx + glib_n_poll_fds);
199 pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
200 n = g_main_context_query(context, max_priority, &timeout, pfds,
201 glib_n_poll_fds);
202 } while (n != glib_n_poll_fds);
203
204 if (timeout < 0) {
205 timeout_ns = -1;
206 } else {
207 timeout_ns = (int64_t)timeout * (int64_t)SCALE_MS;
208 }
209
210 *cur_timeout = qemu_soonest_timeout(timeout_ns, *cur_timeout);
211}
212
213static void glib_pollfds_poll(void)
214{
215 GMainContext *context = g_main_context_default();
216 GPollFD *pfds = &g_array_index(gpollfds, GPollFD, glib_pollfds_idx);
217
218 if (g_main_context_check(context, max_priority, pfds, glib_n_poll_fds)) {
219 g_main_context_dispatch(context);
220 }
221}
222
223#define MAX_MAIN_LOOP_SPIN (1000)
224
225static int os_host_main_loop_wait(int64_t timeout)
226{
227 GMainContext *context = g_main_context_default();
228 int ret;
229
230 g_main_context_acquire(context);
231
232 glib_pollfds_fill(&timeout);
233
234 qemu_mutex_unlock_iothread();
235 replay_mutex_unlock();
236
237 ret = qemu_poll_ns((GPollFD *)gpollfds->data, gpollfds->len, timeout);
238
239 replay_mutex_lock();
240 qemu_mutex_lock_iothread();
241
242 glib_pollfds_poll();
243
244 g_main_context_release(context);
245
246 return ret;
247}
248#else
249/***********************************************************/
250/* Polling handling */
251
252typedef struct PollingEntry {
253 PollingFunc *func;
254 void *opaque;
255 struct PollingEntry *next;
256} PollingEntry;
257
258static PollingEntry *first_polling_entry;
259
260int qemu_add_polling_cb(PollingFunc *func, void *opaque)
261{
262 PollingEntry **ppe, *pe;
263 pe = g_malloc0(sizeof(PollingEntry));
264 pe->func = func;
265 pe->opaque = opaque;
266 for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next);
267 *ppe = pe;
268 return 0;
269}
270
271void qemu_del_polling_cb(PollingFunc *func, void *opaque)
272{
273 PollingEntry **ppe, *pe;
274 for(ppe = &first_polling_entry; *ppe != NULL; ppe = &(*ppe)->next) {
275 pe = *ppe;
276 if (pe->func == func && pe->opaque == opaque) {
277 *ppe = pe->next;
278 g_free(pe);
279 break;
280 }
281 }
282}
283
284/***********************************************************/
285/* Wait objects support */
286typedef struct WaitObjects {
287 int num;
288 int revents[MAXIMUM_WAIT_OBJECTS + 1];
289 HANDLE events[MAXIMUM_WAIT_OBJECTS + 1];
290 WaitObjectFunc *func[MAXIMUM_WAIT_OBJECTS + 1];
291 void *opaque[MAXIMUM_WAIT_OBJECTS + 1];
292} WaitObjects;
293
294static WaitObjects wait_objects = {0};
295
296int qemu_add_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
297{
298 WaitObjects *w = &wait_objects;
299 if (w->num >= MAXIMUM_WAIT_OBJECTS) {
300 return -1;
301 }
302 w->events[w->num] = handle;
303 w->func[w->num] = func;
304 w->opaque[w->num] = opaque;
305 w->revents[w->num] = 0;
306 w->num++;
307 return 0;
308}
309
310void qemu_del_wait_object(HANDLE handle, WaitObjectFunc *func, void *opaque)
311{
312 int i, found;
313 WaitObjects *w = &wait_objects;
314
315 found = 0;
316 for (i = 0; i < w->num; i++) {
317 if (w->events[i] == handle) {
318 found = 1;
319 }
320 if (found) {
321 w->events[i] = w->events[i + 1];
322 w->func[i] = w->func[i + 1];
323 w->opaque[i] = w->opaque[i + 1];
324 w->revents[i] = w->revents[i + 1];
325 }
326 }
327 if (found) {
328 w->num--;
329 }
330}
331
332void qemu_fd_register(int fd)
333{
334 WSAEventSelect(fd, event_notifier_get_handle(&qemu_aio_context->notifier),
335 FD_READ | FD_ACCEPT | FD_CLOSE |
336 FD_CONNECT | FD_WRITE | FD_OOB);
337}
338
339static int pollfds_fill(GArray *pollfds, fd_set *rfds, fd_set *wfds,
340 fd_set *xfds)
341{
342 int nfds = -1;
343 int i;
344
345 for (i = 0; i < pollfds->len; i++) {
346 GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
347 int fd = pfd->fd;
348 int events = pfd->events;
349 if (events & G_IO_IN) {
350 FD_SET(fd, rfds);
351 nfds = MAX(nfds, fd);
352 }
353 if (events & G_IO_OUT) {
354 FD_SET(fd, wfds);
355 nfds = MAX(nfds, fd);
356 }
357 if (events & G_IO_PRI) {
358 FD_SET(fd, xfds);
359 nfds = MAX(nfds, fd);
360 }
361 }
362 return nfds;
363}
364
365static void pollfds_poll(GArray *pollfds, int nfds, fd_set *rfds,
366 fd_set *wfds, fd_set *xfds)
367{
368 int i;
369
370 for (i = 0; i < pollfds->len; i++) {
371 GPollFD *pfd = &g_array_index(pollfds, GPollFD, i);
372 int fd = pfd->fd;
373 int revents = 0;
374
375 if (FD_ISSET(fd, rfds)) {
376 revents |= G_IO_IN;
377 }
378 if (FD_ISSET(fd, wfds)) {
379 revents |= G_IO_OUT;
380 }
381 if (FD_ISSET(fd, xfds)) {
382 revents |= G_IO_PRI;
383 }
384 pfd->revents = revents & pfd->events;
385 }
386}
387
388static int os_host_main_loop_wait(int64_t timeout)
389{
390 GMainContext *context = g_main_context_default();
391 GPollFD poll_fds[1024 * 2]; /* this is probably overkill */
392 int select_ret = 0;
393 int g_poll_ret, ret, i, n_poll_fds;
394 PollingEntry *pe;
395 WaitObjects *w = &wait_objects;
396 gint poll_timeout;
397 int64_t poll_timeout_ns;
398 static struct timeval tv0;
399 fd_set rfds, wfds, xfds;
400 int nfds;
401
402 g_main_context_acquire(context);
403
404 /* XXX: need to suppress polling by better using win32 events */
405 ret = 0;
406 for (pe = first_polling_entry; pe != NULL; pe = pe->next) {
407 ret |= pe->func(pe->opaque);
408 }
409 if (ret != 0) {
410 g_main_context_release(context);
411 return ret;
412 }
413
414 FD_ZERO(&rfds);
415 FD_ZERO(&wfds);
416 FD_ZERO(&xfds);
417 nfds = pollfds_fill(gpollfds, &rfds, &wfds, &xfds);
418 if (nfds >= 0) {
419 select_ret = select(nfds + 1, &rfds, &wfds, &xfds, &tv0);
420 if (select_ret != 0) {
421 timeout = 0;
422 }
423 if (select_ret > 0) {
424 pollfds_poll(gpollfds, nfds, &rfds, &wfds, &xfds);
425 }
426 }
427
428 g_main_context_prepare(context, &max_priority);
429 n_poll_fds = g_main_context_query(context, max_priority, &poll_timeout,
430 poll_fds, ARRAY_SIZE(poll_fds));
431 g_assert(n_poll_fds + w->num <= ARRAY_SIZE(poll_fds));
432
433 for (i = 0; i < w->num; i++) {
434 poll_fds[n_poll_fds + i].fd = (DWORD_PTR)w->events[i];
435 poll_fds[n_poll_fds + i].events = G_IO_IN;
436 }
437
438 if (poll_timeout < 0) {
439 poll_timeout_ns = -1;
440 } else {
441 poll_timeout_ns = (int64_t)poll_timeout * (int64_t)SCALE_MS;
442 }
443
444 poll_timeout_ns = qemu_soonest_timeout(poll_timeout_ns, timeout);
445
446 qemu_mutex_unlock_iothread();
447
448 replay_mutex_unlock();
449
450 g_poll_ret = qemu_poll_ns(poll_fds, n_poll_fds + w->num, poll_timeout_ns);
451
452 replay_mutex_lock();
453
454 qemu_mutex_lock_iothread();
455 if (g_poll_ret > 0) {
456 for (i = 0; i < w->num; i++) {
457 w->revents[i] = poll_fds[n_poll_fds + i].revents;
458 }
459 for (i = 0; i < w->num; i++) {
460 if (w->revents[i] && w->func[i]) {
461 w->func[i](w->opaque[i]);
462 }
463 }
464 }
465
466 if (g_main_context_check(context, max_priority, poll_fds, n_poll_fds)) {
467 g_main_context_dispatch(context);
468 }
469
470 g_main_context_release(context);
471
472 return select_ret || g_poll_ret;
473}
474#endif
475
476static NotifierList main_loop_poll_notifiers =
477 NOTIFIER_LIST_INITIALIZER(main_loop_poll_notifiers);
478
479void main_loop_poll_add_notifier(Notifier *notify)
480{
481 notifier_list_add(&main_loop_poll_notifiers, notify);
482}
483
484void main_loop_poll_remove_notifier(Notifier *notify)
485{
486 notifier_remove(notify);
487}
488
489void main_loop_wait(int nonblocking)
490{
491 MainLoopPoll mlpoll = {
492 .state = MAIN_LOOP_POLL_FILL,
493 .timeout = UINT32_MAX,
494 .pollfds = gpollfds,
495 };
496 int ret;
497 int64_t timeout_ns;
498
499 if (nonblocking) {
500 mlpoll.timeout = 0;
501 }
502
503 /* poll any events */
504 g_array_set_size(gpollfds, 0); /* reset for new iteration */
505 /* XXX: separate device handlers from system ones */
506 notifier_list_notify(&main_loop_poll_notifiers, &mlpoll);
507
508 if (mlpoll.timeout == UINT32_MAX) {
509 timeout_ns = -1;
510 } else {
511 timeout_ns = (uint64_t)mlpoll.timeout * (int64_t)(SCALE_MS);
512 }
513
514 timeout_ns = qemu_soonest_timeout(timeout_ns,
515 timerlistgroup_deadline_ns(
516 &main_loop_tlg));
517
518 ret = os_host_main_loop_wait(timeout_ns);
519 mlpoll.state = ret < 0 ? MAIN_LOOP_POLL_ERR : MAIN_LOOP_POLL_OK;
520 notifier_list_notify(&main_loop_poll_notifiers, &mlpoll);
521
522 /* CPU thread can infinitely wait for event after
523 missing the warp */
524 qemu_start_warp_timer();
525 qemu_clock_run_all_timers();
526}
527
528/* Functions to operate on the main QEMU AioContext. */
529
530QEMUBH *qemu_bh_new(QEMUBHFunc *cb, void *opaque)
531{
532 return aio_bh_new(qemu_aio_context, cb, opaque);
533}
534
535/*
536 * Functions to operate on the I/O handler AioContext.
537 * This context runs on top of main loop. We can't reuse qemu_aio_context
538 * because iohandlers mustn't be polled by aio_poll(qemu_aio_context).
539 */
540static AioContext *iohandler_ctx;
541
542static void iohandler_init(void)
543{
544 if (!iohandler_ctx) {
545 iohandler_ctx = aio_context_new(&error_abort);
546 }
547}
548
549AioContext *iohandler_get_aio_context(void)
550{
551 iohandler_init();
552 return iohandler_ctx;
553}
554
555GSource *iohandler_get_g_source(void)
556{
557 iohandler_init();
558 return aio_get_g_source(iohandler_ctx);
559}
560
561void qemu_set_fd_handler(int fd,
562 IOHandler *fd_read,
563 IOHandler *fd_write,
564 void *opaque)
565{
566 iohandler_init();
567 aio_set_fd_handler(iohandler_ctx, fd, false,
568 fd_read, fd_write, NULL, opaque);
569}
570
571void event_notifier_set_handler(EventNotifier *e,
572 EventNotifierHandler *handler)
573{
574 iohandler_init();
575 aio_set_event_notifier(iohandler_ctx, e, false,
576 handler, NULL);
577}
578
579/* reaping of zombies. right now we're not passing the status to
580 anyone, but it would be possible to add a callback. */
581#ifndef _WIN32
582typedef struct ChildProcessRecord {
583 int pid;
584 QLIST_ENTRY(ChildProcessRecord) next;
585} ChildProcessRecord;
586
587static QLIST_HEAD(, ChildProcessRecord) child_watches =
588 QLIST_HEAD_INITIALIZER(child_watches);
589
590static QEMUBH *sigchld_bh;
591
592static void sigchld_handler(int signal)
593{
594 qemu_bh_schedule(sigchld_bh);
595}
596
597static void sigchld_bh_handler(void *opaque)
598{
599 ChildProcessRecord *rec, *next;
600
601 QLIST_FOREACH_SAFE(rec, &child_watches, next, next) {
602 if (waitpid(rec->pid, NULL, WNOHANG) == rec->pid) {
603 QLIST_REMOVE(rec, next);
604 g_free(rec);
605 }
606 }
607}
608
609static void qemu_init_child_watch(void)
610{
611 struct sigaction act;
612 sigchld_bh = qemu_bh_new(sigchld_bh_handler, NULL);
613
614 memset(&act, 0, sizeof(act));
615 act.sa_handler = sigchld_handler;
616 act.sa_flags = SA_NOCLDSTOP;
617 sigaction(SIGCHLD, &act, NULL);
618}
619
620int qemu_add_child_watch(pid_t pid)
621{
622 ChildProcessRecord *rec;
623
624 if (!sigchld_bh) {
625 qemu_init_child_watch();
626 }
627
628 QLIST_FOREACH(rec, &child_watches, next) {
629 if (rec->pid == pid) {
630 return 1;
631 }
632 }
633 rec = g_malloc0(sizeof(ChildProcessRecord));
634 rec->pid = pid;
635 QLIST_INSERT_HEAD(&child_watches, rec, next);
636 return 0;
637}
638#endif