linux-user: Support for restarting system calls for OpenRISC targets
[qemu.git] / tests / vhost-user-test.c
1 /*
2 * QTest testcase for the vhost-user
3 *
4 * Copyright (c) 2014 Virtual Open Systems Sarl.
5 *
6 * This work is licensed under the terms of the GNU GPL, version 2 or later.
7 * See the COPYING file in the top-level directory.
8 *
9 */
10
11 #include "qemu/osdep.h"
12 #include <glib.h>
13
14 #include "libqtest.h"
15 #include "qemu/option.h"
16 #include "qemu/range.h"
17 #include "sysemu/char.h"
18 #include "sysemu/sysemu.h"
19
20 #include <linux/vhost.h>
21 #include <sys/mman.h>
22 #include <sys/vfs.h>
23 #include <qemu/sockets.h>
24
25 /* GLIB version compatibility flags */
26 #if !GLIB_CHECK_VERSION(2, 26, 0)
27 #define G_TIME_SPAN_SECOND (G_GINT64_CONSTANT(1000000))
28 #endif
29
30 #if GLIB_CHECK_VERSION(2, 28, 0)
31 #define HAVE_MONOTONIC_TIME
32 #endif
33
34 #define QEMU_CMD_ACCEL " -machine accel=tcg"
35 #define QEMU_CMD_MEM " -m %d -object memory-backend-file,id=mem,size=%dM,"\
36 "mem-path=%s,share=on -numa node,memdev=mem"
37 #define QEMU_CMD_CHR " -chardev socket,id=%s,path=%s"
38 #define QEMU_CMD_NETDEV " -netdev vhost-user,id=net0,chardev=%s,vhostforce"
39 #define QEMU_CMD_NET " -device virtio-net-pci,netdev=net0,romfile=./pc-bios/pxe-virtio.rom"
40
41 #define QEMU_CMD QEMU_CMD_ACCEL QEMU_CMD_MEM QEMU_CMD_CHR \
42 QEMU_CMD_NETDEV QEMU_CMD_NET
43
44 #define HUGETLBFS_MAGIC 0x958458f6
45
46 /*********** FROM hw/virtio/vhost-user.c *************************************/
47
48 #define VHOST_MEMORY_MAX_NREGIONS 8
49
50 #define VHOST_USER_F_PROTOCOL_FEATURES 30
51 #define VHOST_USER_PROTOCOL_F_LOG_SHMFD 1
52
53 #define VHOST_LOG_PAGE 0x1000
54
55 typedef enum VhostUserRequest {
56 VHOST_USER_NONE = 0,
57 VHOST_USER_GET_FEATURES = 1,
58 VHOST_USER_SET_FEATURES = 2,
59 VHOST_USER_SET_OWNER = 3,
60 VHOST_USER_RESET_OWNER = 4,
61 VHOST_USER_SET_MEM_TABLE = 5,
62 VHOST_USER_SET_LOG_BASE = 6,
63 VHOST_USER_SET_LOG_FD = 7,
64 VHOST_USER_SET_VRING_NUM = 8,
65 VHOST_USER_SET_VRING_ADDR = 9,
66 VHOST_USER_SET_VRING_BASE = 10,
67 VHOST_USER_GET_VRING_BASE = 11,
68 VHOST_USER_SET_VRING_KICK = 12,
69 VHOST_USER_SET_VRING_CALL = 13,
70 VHOST_USER_SET_VRING_ERR = 14,
71 VHOST_USER_GET_PROTOCOL_FEATURES = 15,
72 VHOST_USER_SET_PROTOCOL_FEATURES = 16,
73 VHOST_USER_SET_VRING_ENABLE = 18,
74 VHOST_USER_MAX
75 } VhostUserRequest;
76
77 typedef struct VhostUserMemoryRegion {
78 uint64_t guest_phys_addr;
79 uint64_t memory_size;
80 uint64_t userspace_addr;
81 uint64_t mmap_offset;
82 } VhostUserMemoryRegion;
83
84 typedef struct VhostUserMemory {
85 uint32_t nregions;
86 uint32_t padding;
87 VhostUserMemoryRegion regions[VHOST_MEMORY_MAX_NREGIONS];
88 } VhostUserMemory;
89
90 typedef struct VhostUserLog {
91 uint64_t mmap_size;
92 uint64_t mmap_offset;
93 } VhostUserLog;
94
95 typedef struct VhostUserMsg {
96 VhostUserRequest request;
97
98 #define VHOST_USER_VERSION_MASK (0x3)
99 #define VHOST_USER_REPLY_MASK (0x1<<2)
100 uint32_t flags;
101 uint32_t size; /* the following payload size */
102 union {
103 #define VHOST_USER_VRING_IDX_MASK (0xff)
104 #define VHOST_USER_VRING_NOFD_MASK (0x1<<8)
105 uint64_t u64;
106 struct vhost_vring_state state;
107 struct vhost_vring_addr addr;
108 VhostUserMemory memory;
109 VhostUserLog log;
110 } payload;
111 } QEMU_PACKED VhostUserMsg;
112
113 static VhostUserMsg m __attribute__ ((unused));
114 #define VHOST_USER_HDR_SIZE (sizeof(m.request) \
115 + sizeof(m.flags) \
116 + sizeof(m.size))
117
118 #define VHOST_USER_PAYLOAD_SIZE (sizeof(m) - VHOST_USER_HDR_SIZE)
119
120 /* The version of the protocol we support */
121 #define VHOST_USER_VERSION (0x1)
122 /*****************************************************************************/
123
124 typedef struct TestServer {
125 gchar *socket_path;
126 gchar *mig_path;
127 gchar *chr_name;
128 CharDriverState *chr;
129 int fds_num;
130 int fds[VHOST_MEMORY_MAX_NREGIONS];
131 VhostUserMemory memory;
132 GMutex data_mutex;
133 GCond data_cond;
134 int log_fd;
135 uint64_t rings;
136 } TestServer;
137
138 #if !GLIB_CHECK_VERSION(2, 32, 0)
139 static gboolean g_cond_wait_until(CompatGCond cond, CompatGMutex mutex,
140 gint64 end_time)
141 {
142 gboolean ret = FALSE;
143 end_time -= g_get_monotonic_time();
144 GTimeVal time = { end_time / G_TIME_SPAN_SECOND,
145 end_time % G_TIME_SPAN_SECOND };
146 ret = g_cond_timed_wait(cond, mutex, &time);
147 return ret;
148 }
149 #endif
150
151 static const char *tmpfs;
152 static const char *root;
153
154 static void wait_for_fds(TestServer *s)
155 {
156 gint64 end_time;
157
158 g_mutex_lock(&s->data_mutex);
159
160 end_time = g_get_monotonic_time() + 5 * G_TIME_SPAN_SECOND;
161 while (!s->fds_num) {
162 if (!g_cond_wait_until(&s->data_cond, &s->data_mutex, end_time)) {
163 /* timeout has passed */
164 g_assert(s->fds_num);
165 break;
166 }
167 }
168
169 /* check for sanity */
170 g_assert_cmpint(s->fds_num, >, 0);
171 g_assert_cmpint(s->fds_num, ==, s->memory.nregions);
172
173 g_mutex_unlock(&s->data_mutex);
174 }
175
176 static void read_guest_mem(const void *data)
177 {
178 TestServer *s = (void *)data;
179 uint32_t *guest_mem;
180 int i, j;
181 size_t size;
182
183 wait_for_fds(s);
184
185 g_mutex_lock(&s->data_mutex);
186
187 /* iterate all regions */
188 for (i = 0; i < s->fds_num; i++) {
189
190 /* We'll check only the region statring at 0x0*/
191 if (s->memory.regions[i].guest_phys_addr != 0x0) {
192 continue;
193 }
194
195 g_assert_cmpint(s->memory.regions[i].memory_size, >, 1024);
196
197 size = s->memory.regions[i].memory_size +
198 s->memory.regions[i].mmap_offset;
199
200 guest_mem = mmap(0, size, PROT_READ | PROT_WRITE,
201 MAP_SHARED, s->fds[i], 0);
202
203 g_assert(guest_mem != MAP_FAILED);
204 guest_mem += (s->memory.regions[i].mmap_offset / sizeof(*guest_mem));
205
206 for (j = 0; j < 256; j++) {
207 uint32_t a = readl(s->memory.regions[i].guest_phys_addr + j*4);
208 uint32_t b = guest_mem[j];
209
210 g_assert_cmpint(a, ==, b);
211 }
212
213 munmap(guest_mem, s->memory.regions[i].memory_size);
214 }
215
216 g_mutex_unlock(&s->data_mutex);
217 }
218
219 static void *thread_function(void *data)
220 {
221 GMainLoop *loop = data;
222 g_main_loop_run(loop);
223 return NULL;
224 }
225
226 static int chr_can_read(void *opaque)
227 {
228 return VHOST_USER_HDR_SIZE;
229 }
230
231 static void chr_read(void *opaque, const uint8_t *buf, int size)
232 {
233 TestServer *s = opaque;
234 CharDriverState *chr = s->chr;
235 VhostUserMsg msg;
236 uint8_t *p = (uint8_t *) &msg;
237 int fd;
238
239 if (size != VHOST_USER_HDR_SIZE) {
240 g_test_message("Wrong message size received %d\n", size);
241 return;
242 }
243
244 g_mutex_lock(&s->data_mutex);
245 memcpy(p, buf, VHOST_USER_HDR_SIZE);
246
247 if (msg.size) {
248 p += VHOST_USER_HDR_SIZE;
249 g_assert_cmpint(qemu_chr_fe_read_all(chr, p, msg.size), ==, msg.size);
250 }
251
252 switch (msg.request) {
253 case VHOST_USER_GET_FEATURES:
254 /* send back features to qemu */
255 msg.flags |= VHOST_USER_REPLY_MASK;
256 msg.size = sizeof(m.payload.u64);
257 msg.payload.u64 = 0x1ULL << VHOST_F_LOG_ALL |
258 0x1ULL << VHOST_USER_F_PROTOCOL_FEATURES;
259 p = (uint8_t *) &msg;
260 qemu_chr_fe_write_all(chr, p, VHOST_USER_HDR_SIZE + msg.size);
261 break;
262
263 case VHOST_USER_SET_FEATURES:
264 g_assert_cmpint(msg.payload.u64 & (0x1ULL << VHOST_USER_F_PROTOCOL_FEATURES),
265 !=, 0ULL);
266 break;
267
268 case VHOST_USER_GET_PROTOCOL_FEATURES:
269 /* send back features to qemu */
270 msg.flags |= VHOST_USER_REPLY_MASK;
271 msg.size = sizeof(m.payload.u64);
272 msg.payload.u64 = 1 << VHOST_USER_PROTOCOL_F_LOG_SHMFD;
273 p = (uint8_t *) &msg;
274 qemu_chr_fe_write_all(chr, p, VHOST_USER_HDR_SIZE + msg.size);
275 break;
276
277 case VHOST_USER_GET_VRING_BASE:
278 /* send back vring base to qemu */
279 msg.flags |= VHOST_USER_REPLY_MASK;
280 msg.size = sizeof(m.payload.state);
281 msg.payload.state.num = 0;
282 p = (uint8_t *) &msg;
283 qemu_chr_fe_write_all(chr, p, VHOST_USER_HDR_SIZE + msg.size);
284
285 assert(msg.payload.state.index < 2);
286 s->rings &= ~(0x1ULL << msg.payload.state.index);
287 break;
288
289 case VHOST_USER_SET_MEM_TABLE:
290 /* received the mem table */
291 memcpy(&s->memory, &msg.payload.memory, sizeof(msg.payload.memory));
292 s->fds_num = qemu_chr_fe_get_msgfds(chr, s->fds, G_N_ELEMENTS(s->fds));
293
294 /* signal the test that it can continue */
295 g_cond_signal(&s->data_cond);
296 break;
297
298 case VHOST_USER_SET_VRING_KICK:
299 case VHOST_USER_SET_VRING_CALL:
300 /* consume the fd */
301 qemu_chr_fe_get_msgfds(chr, &fd, 1);
302 /*
303 * This is a non-blocking eventfd.
304 * The receive function forces it to be blocking,
305 * so revert it back to non-blocking.
306 */
307 qemu_set_nonblock(fd);
308 break;
309
310 case VHOST_USER_SET_LOG_BASE:
311 if (s->log_fd != -1) {
312 close(s->log_fd);
313 s->log_fd = -1;
314 }
315 qemu_chr_fe_get_msgfds(chr, &s->log_fd, 1);
316 msg.flags |= VHOST_USER_REPLY_MASK;
317 msg.size = 0;
318 p = (uint8_t *) &msg;
319 qemu_chr_fe_write_all(chr, p, VHOST_USER_HDR_SIZE);
320
321 g_cond_signal(&s->data_cond);
322 break;
323
324 case VHOST_USER_SET_VRING_BASE:
325 assert(msg.payload.state.index < 2);
326 s->rings |= 0x1ULL << msg.payload.state.index;
327 break;
328
329 default:
330 break;
331 }
332
333 g_mutex_unlock(&s->data_mutex);
334 }
335
336 static const char *init_hugepagefs(const char *path)
337 {
338 struct statfs fs;
339 int ret;
340
341 if (access(path, R_OK | W_OK | X_OK)) {
342 g_test_message("access on path (%s): %s\n", path, strerror(errno));
343 return NULL;
344 }
345
346 do {
347 ret = statfs(path, &fs);
348 } while (ret != 0 && errno == EINTR);
349
350 if (ret != 0) {
351 g_test_message("statfs on path (%s): %s\n", path, strerror(errno));
352 return NULL;
353 }
354
355 if (fs.f_type != HUGETLBFS_MAGIC) {
356 g_test_message("Warning: path not on HugeTLBFS: %s\n", path);
357 return NULL;
358 }
359
360 return path;
361 }
362
363 static TestServer *test_server_new(const gchar *name)
364 {
365 TestServer *server = g_new0(TestServer, 1);
366 gchar *chr_path;
367
368 server->socket_path = g_strdup_printf("%s/%s.sock", tmpfs, name);
369 server->mig_path = g_strdup_printf("%s/%s.mig", tmpfs, name);
370
371 chr_path = g_strdup_printf("unix:%s,server,nowait", server->socket_path);
372 server->chr_name = g_strdup_printf("chr-%s", name);
373 server->chr = qemu_chr_new(server->chr_name, chr_path, NULL);
374 g_free(chr_path);
375
376 qemu_chr_add_handlers(server->chr, chr_can_read, chr_read, NULL, server);
377
378 g_mutex_init(&server->data_mutex);
379 g_cond_init(&server->data_cond);
380
381 server->log_fd = -1;
382
383 return server;
384 }
385
386 #define GET_QEMU_CMD(s) \
387 g_strdup_printf(QEMU_CMD, 512, 512, (root), (s)->chr_name, \
388 (s)->socket_path, (s)->chr_name)
389
390 #define GET_QEMU_CMDE(s, mem, extra, ...) \
391 g_strdup_printf(QEMU_CMD extra, (mem), (mem), (root), (s)->chr_name, \
392 (s)->socket_path, (s)->chr_name, ##__VA_ARGS__)
393
394 static gboolean _test_server_free(TestServer *server)
395 {
396 int i;
397
398 qemu_chr_delete(server->chr);
399
400 for (i = 0; i < server->fds_num; i++) {
401 close(server->fds[i]);
402 }
403
404 if (server->log_fd != -1) {
405 close(server->log_fd);
406 }
407
408 unlink(server->socket_path);
409 g_free(server->socket_path);
410
411 unlink(server->mig_path);
412 g_free(server->mig_path);
413
414 g_free(server->chr_name);
415 g_free(server);
416
417 return FALSE;
418 }
419
420 static void test_server_free(TestServer *server)
421 {
422 g_idle_add((GSourceFunc)_test_server_free, server);
423 }
424
425 static void wait_for_log_fd(TestServer *s)
426 {
427 gint64 end_time;
428
429 g_mutex_lock(&s->data_mutex);
430 end_time = g_get_monotonic_time() + 5 * G_TIME_SPAN_SECOND;
431 while (s->log_fd == -1) {
432 if (!g_cond_wait_until(&s->data_cond, &s->data_mutex, end_time)) {
433 /* timeout has passed */
434 g_assert(s->log_fd != -1);
435 break;
436 }
437 }
438
439 g_mutex_unlock(&s->data_mutex);
440 }
441
442 static void write_guest_mem(TestServer *s, uint32_t seed)
443 {
444 uint32_t *guest_mem;
445 int i, j;
446 size_t size;
447
448 wait_for_fds(s);
449
450 /* iterate all regions */
451 for (i = 0; i < s->fds_num; i++) {
452
453 /* We'll write only the region statring at 0x0 */
454 if (s->memory.regions[i].guest_phys_addr != 0x0) {
455 continue;
456 }
457
458 g_assert_cmpint(s->memory.regions[i].memory_size, >, 1024);
459
460 size = s->memory.regions[i].memory_size +
461 s->memory.regions[i].mmap_offset;
462
463 guest_mem = mmap(0, size, PROT_READ | PROT_WRITE,
464 MAP_SHARED, s->fds[i], 0);
465
466 g_assert(guest_mem != MAP_FAILED);
467 guest_mem += (s->memory.regions[i].mmap_offset / sizeof(*guest_mem));
468
469 for (j = 0; j < 256; j++) {
470 guest_mem[j] = seed + j;
471 }
472
473 munmap(guest_mem, s->memory.regions[i].memory_size);
474 break;
475 }
476 }
477
478 static guint64 get_log_size(TestServer *s)
479 {
480 guint64 log_size = 0;
481 int i;
482
483 for (i = 0; i < s->memory.nregions; ++i) {
484 VhostUserMemoryRegion *reg = &s->memory.regions[i];
485 guint64 last = range_get_last(reg->guest_phys_addr,
486 reg->memory_size);
487 log_size = MAX(log_size, last / (8 * VHOST_LOG_PAGE) + 1);
488 }
489
490 return log_size;
491 }
492
493 typedef struct TestMigrateSource {
494 GSource source;
495 TestServer *src;
496 TestServer *dest;
497 } TestMigrateSource;
498
499 static gboolean
500 test_migrate_source_check(GSource *source)
501 {
502 TestMigrateSource *t = (TestMigrateSource *)source;
503 gboolean overlap = t->src->rings && t->dest->rings;
504
505 g_assert(!overlap);
506
507 return FALSE;
508 }
509
510 #if !GLIB_CHECK_VERSION(2,36,0)
511 /* this callback is unnecessary with glib >2.36, the default
512 * prepare for the source does the same */
513 static gboolean
514 test_migrate_source_prepare(GSource *source, gint *timeout)
515 {
516 *timeout = -1;
517 return FALSE;
518 }
519 #endif
520
521 GSourceFuncs test_migrate_source_funcs = {
522 #if !GLIB_CHECK_VERSION(2,36,0)
523 .prepare = test_migrate_source_prepare,
524 #endif
525 .check = test_migrate_source_check,
526 };
527
528 static void test_migrate(void)
529 {
530 TestServer *s = test_server_new("src");
531 TestServer *dest = test_server_new("dest");
532 char *uri = g_strdup_printf("%s%s", "unix:", dest->mig_path);
533 QTestState *global = global_qtest, *from, *to;
534 GSource *source;
535 gchar *cmd;
536 QDict *rsp;
537 guint8 *log;
538 guint64 size;
539
540 cmd = GET_QEMU_CMDE(s, 2, "");
541 from = qtest_start(cmd);
542 g_free(cmd);
543
544 wait_for_fds(s);
545 size = get_log_size(s);
546 g_assert_cmpint(size, ==, (2 * 1024 * 1024) / (VHOST_LOG_PAGE * 8));
547
548 cmd = GET_QEMU_CMDE(dest, 2, " -incoming %s", uri);
549 to = qtest_init(cmd);
550 g_free(cmd);
551
552 source = g_source_new(&test_migrate_source_funcs,
553 sizeof(TestMigrateSource));
554 ((TestMigrateSource *)source)->src = s;
555 ((TestMigrateSource *)source)->dest = dest;
556 g_source_attach(source, NULL);
557
558 /* slow down migration to have time to fiddle with log */
559 /* TODO: qtest could learn to break on some places */
560 rsp = qmp("{ 'execute': 'migrate_set_speed',"
561 "'arguments': { 'value': 10 } }");
562 g_assert(qdict_haskey(rsp, "return"));
563 QDECREF(rsp);
564
565 cmd = g_strdup_printf("{ 'execute': 'migrate',"
566 "'arguments': { 'uri': '%s' } }",
567 uri);
568 rsp = qmp(cmd);
569 g_free(cmd);
570 g_assert(qdict_haskey(rsp, "return"));
571 QDECREF(rsp);
572
573 wait_for_log_fd(s);
574
575 log = mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, s->log_fd, 0);
576 g_assert(log != MAP_FAILED);
577
578 /* modify first page */
579 write_guest_mem(s, 0x42);
580 log[0] = 1;
581 munmap(log, size);
582
583 /* speed things up */
584 rsp = qmp("{ 'execute': 'migrate_set_speed',"
585 "'arguments': { 'value': 0 } }");
586 g_assert(qdict_haskey(rsp, "return"));
587 QDECREF(rsp);
588
589 qmp_eventwait("STOP");
590
591 global_qtest = to;
592 qmp_eventwait("RESUME");
593
594 read_guest_mem(dest);
595
596 g_source_destroy(source);
597 g_source_unref(source);
598
599 qtest_quit(to);
600 test_server_free(dest);
601 qtest_quit(from);
602 test_server_free(s);
603 g_free(uri);
604
605 global_qtest = global;
606 }
607
608 int main(int argc, char **argv)
609 {
610 QTestState *s = NULL;
611 TestServer *server = NULL;
612 const char *hugefs;
613 char *qemu_cmd = NULL;
614 int ret;
615 char template[] = "/tmp/vhost-test-XXXXXX";
616 GMainLoop *loop;
617 GThread *thread;
618
619 g_test_init(&argc, &argv, NULL);
620
621 module_call_init(MODULE_INIT_QOM);
622 qemu_add_opts(&qemu_chardev_opts);
623
624 tmpfs = mkdtemp(template);
625 if (!tmpfs) {
626 g_test_message("mkdtemp on path (%s): %s\n", template, strerror(errno));
627 }
628 g_assert(tmpfs);
629
630 hugefs = getenv("QTEST_HUGETLBFS_PATH");
631 if (hugefs) {
632 root = init_hugepagefs(hugefs);
633 g_assert(root);
634 } else {
635 root = tmpfs;
636 }
637
638 server = test_server_new("test");
639
640 loop = g_main_loop_new(NULL, FALSE);
641 /* run the main loop thread so the chardev may operate */
642 thread = g_thread_new(NULL, thread_function, loop);
643
644 qemu_cmd = GET_QEMU_CMD(server);
645
646 s = qtest_start(qemu_cmd);
647 g_free(qemu_cmd);
648
649 qtest_add_data_func("/vhost-user/read-guest-mem", server, read_guest_mem);
650 qtest_add_func("/vhost-user/migrate", test_migrate);
651
652 ret = g_test_run();
653
654 if (s) {
655 qtest_quit(s);
656 }
657
658 /* cleanup */
659 test_server_free(server);
660
661 /* finish the helper thread and dispatch pending sources */
662 g_main_loop_quit(loop);
663 g_thread_join(thread);
664 while (g_main_context_pending(NULL)) {
665 g_main_context_iteration (NULL, TRUE);
666 }
667 g_main_loop_unref(loop);
668
669 ret = rmdir(tmpfs);
670 if (ret != 0) {
671 g_test_message("unable to rmdir: path (%s): %s\n",
672 tmpfs, strerror(errno));
673 }
674 g_assert_cmpint(ret, ==, 0);
675
676 return ret;
677 }