]> git.proxmox.com Git - mirror_qemu.git/blob - hw/virtio/virtio-balloon.c
Merge remote-tracking branch 'remotes/mst/tags/for_upstream' into staging
[mirror_qemu.git] / hw / virtio / virtio-balloon.c
1 /*
2 * Virtio Balloon Device
3 *
4 * Copyright IBM, Corp. 2008
5 * Copyright (C) 2011 Red Hat, Inc.
6 * Copyright (C) 2011 Amit Shah <amit.shah@redhat.com>
7 *
8 * Authors:
9 * Anthony Liguori <aliguori@us.ibm.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 */
15
16 #include "qemu/osdep.h"
17 #include "qemu/iov.h"
18 #include "qemu/timer.h"
19 #include "qemu-common.h"
20 #include "hw/virtio/virtio.h"
21 #include "hw/mem/pc-dimm.h"
22 #include "sysemu/balloon.h"
23 #include "hw/virtio/virtio-balloon.h"
24 #include "sysemu/kvm.h"
25 #include "exec/address-spaces.h"
26 #include "qapi/error.h"
27 #include "qapi/visitor.h"
28 #include "qapi-event.h"
29 #include "trace.h"
30 #include "qemu/error-report.h"
31
32 #include "hw/virtio/virtio-bus.h"
33 #include "hw/virtio/virtio-access.h"
34
35 #define BALLOON_PAGE_SIZE (1 << VIRTIO_BALLOON_PFN_SHIFT)
36
37 static void balloon_page(void *addr, int deflate)
38 {
39 if (!qemu_balloon_is_inhibited() && (!kvm_enabled() ||
40 kvm_has_sync_mmu())) {
41 qemu_madvise(addr, BALLOON_PAGE_SIZE,
42 deflate ? QEMU_MADV_WILLNEED : QEMU_MADV_DONTNEED);
43 }
44 }
45
46 static const char *balloon_stat_names[] = {
47 [VIRTIO_BALLOON_S_SWAP_IN] = "stat-swap-in",
48 [VIRTIO_BALLOON_S_SWAP_OUT] = "stat-swap-out",
49 [VIRTIO_BALLOON_S_MAJFLT] = "stat-major-faults",
50 [VIRTIO_BALLOON_S_MINFLT] = "stat-minor-faults",
51 [VIRTIO_BALLOON_S_MEMFREE] = "stat-free-memory",
52 [VIRTIO_BALLOON_S_MEMTOT] = "stat-total-memory",
53 [VIRTIO_BALLOON_S_AVAIL] = "stat-available-memory",
54 [VIRTIO_BALLOON_S_CACHES] = "stat-disk-caches",
55 [VIRTIO_BALLOON_S_NR] = NULL
56 };
57
58 /*
59 * reset_stats - Mark all items in the stats array as unset
60 *
61 * This function needs to be called at device initialization and before
62 * updating to a set of newly-generated stats. This will ensure that no
63 * stale values stick around in case the guest reports a subset of the supported
64 * statistics.
65 */
66 static inline void reset_stats(VirtIOBalloon *dev)
67 {
68 int i;
69 for (i = 0; i < VIRTIO_BALLOON_S_NR; dev->stats[i++] = -1);
70 }
71
72 static bool balloon_stats_supported(const VirtIOBalloon *s)
73 {
74 VirtIODevice *vdev = VIRTIO_DEVICE(s);
75 return virtio_vdev_has_feature(vdev, VIRTIO_BALLOON_F_STATS_VQ);
76 }
77
78 static bool balloon_stats_enabled(const VirtIOBalloon *s)
79 {
80 return s->stats_poll_interval > 0;
81 }
82
83 static void balloon_stats_destroy_timer(VirtIOBalloon *s)
84 {
85 if (balloon_stats_enabled(s)) {
86 timer_del(s->stats_timer);
87 timer_free(s->stats_timer);
88 s->stats_timer = NULL;
89 s->stats_poll_interval = 0;
90 }
91 }
92
93 static void balloon_stats_change_timer(VirtIOBalloon *s, int64_t secs)
94 {
95 timer_mod(s->stats_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + secs * 1000);
96 }
97
98 static void balloon_stats_poll_cb(void *opaque)
99 {
100 VirtIOBalloon *s = opaque;
101 VirtIODevice *vdev = VIRTIO_DEVICE(s);
102
103 if (s->stats_vq_elem == NULL || !balloon_stats_supported(s)) {
104 /* re-schedule */
105 balloon_stats_change_timer(s, s->stats_poll_interval);
106 return;
107 }
108
109 virtqueue_push(s->svq, s->stats_vq_elem, s->stats_vq_offset);
110 virtio_notify(vdev, s->svq);
111 g_free(s->stats_vq_elem);
112 s->stats_vq_elem = NULL;
113 }
114
115 static void balloon_stats_get_all(Object *obj, Visitor *v, const char *name,
116 void *opaque, Error **errp)
117 {
118 Error *err = NULL;
119 VirtIOBalloon *s = opaque;
120 int i;
121
122 visit_start_struct(v, name, NULL, 0, &err);
123 if (err) {
124 goto out;
125 }
126 visit_type_int(v, "last-update", &s->stats_last_update, &err);
127 if (err) {
128 goto out_end;
129 }
130
131 visit_start_struct(v, "stats", NULL, 0, &err);
132 if (err) {
133 goto out_end;
134 }
135 for (i = 0; i < VIRTIO_BALLOON_S_NR; i++) {
136 visit_type_uint64(v, balloon_stat_names[i], &s->stats[i], &err);
137 if (err) {
138 goto out_nested;
139 }
140 }
141 visit_check_struct(v, &err);
142 out_nested:
143 visit_end_struct(v, NULL);
144
145 if (!err) {
146 visit_check_struct(v, &err);
147 }
148 out_end:
149 visit_end_struct(v, NULL);
150 out:
151 error_propagate(errp, err);
152 }
153
154 static void balloon_stats_get_poll_interval(Object *obj, Visitor *v,
155 const char *name, void *opaque,
156 Error **errp)
157 {
158 VirtIOBalloon *s = opaque;
159 visit_type_int(v, name, &s->stats_poll_interval, errp);
160 }
161
162 static void balloon_stats_set_poll_interval(Object *obj, Visitor *v,
163 const char *name, void *opaque,
164 Error **errp)
165 {
166 VirtIOBalloon *s = opaque;
167 Error *local_err = NULL;
168 int64_t value;
169
170 visit_type_int(v, name, &value, &local_err);
171 if (local_err) {
172 error_propagate(errp, local_err);
173 return;
174 }
175
176 if (value < 0) {
177 error_setg(errp, "timer value must be greater than zero");
178 return;
179 }
180
181 if (value > UINT32_MAX) {
182 error_setg(errp, "timer value is too big");
183 return;
184 }
185
186 if (value == s->stats_poll_interval) {
187 return;
188 }
189
190 if (value == 0) {
191 /* timer=0 disables the timer */
192 balloon_stats_destroy_timer(s);
193 return;
194 }
195
196 if (balloon_stats_enabled(s)) {
197 /* timer interval change */
198 s->stats_poll_interval = value;
199 balloon_stats_change_timer(s, value);
200 return;
201 }
202
203 /* create a new timer */
204 g_assert(s->stats_timer == NULL);
205 s->stats_timer = timer_new_ms(QEMU_CLOCK_VIRTUAL, balloon_stats_poll_cb, s);
206 s->stats_poll_interval = value;
207 balloon_stats_change_timer(s, 0);
208 }
209
210 static void virtio_balloon_handle_output(VirtIODevice *vdev, VirtQueue *vq)
211 {
212 VirtIOBalloon *s = VIRTIO_BALLOON(vdev);
213 VirtQueueElement *elem;
214 MemoryRegionSection section;
215
216 for (;;) {
217 size_t offset = 0;
218 uint32_t pfn;
219 elem = virtqueue_pop(vq, sizeof(VirtQueueElement));
220 if (!elem) {
221 return;
222 }
223
224 while (iov_to_buf(elem->out_sg, elem->out_num, offset, &pfn, 4) == 4) {
225 ram_addr_t pa;
226 ram_addr_t addr;
227 int p = virtio_ldl_p(vdev, &pfn);
228
229 pa = (ram_addr_t) p << VIRTIO_BALLOON_PFN_SHIFT;
230 offset += 4;
231
232 /* FIXME: remove get_system_memory(), but how? */
233 section = memory_region_find(get_system_memory(), pa, 1);
234 if (!int128_nz(section.size) ||
235 !memory_region_is_ram(section.mr) ||
236 memory_region_is_rom(section.mr) ||
237 memory_region_is_romd(section.mr)) {
238 trace_virtio_balloon_bad_addr(pa);
239 memory_region_unref(section.mr);
240 continue;
241 }
242
243 trace_virtio_balloon_handle_output(memory_region_name(section.mr),
244 pa);
245 /* Using memory_region_get_ram_ptr is bending the rules a bit, but
246 should be OK because we only want a single page. */
247 addr = section.offset_within_region;
248 balloon_page(memory_region_get_ram_ptr(section.mr) + addr,
249 !!(vq == s->dvq));
250 memory_region_unref(section.mr);
251 }
252
253 virtqueue_push(vq, elem, offset);
254 virtio_notify(vdev, vq);
255 g_free(elem);
256 }
257 }
258
259 static void virtio_balloon_receive_stats(VirtIODevice *vdev, VirtQueue *vq)
260 {
261 VirtIOBalloon *s = VIRTIO_BALLOON(vdev);
262 VirtQueueElement *elem;
263 VirtIOBalloonStat stat;
264 size_t offset = 0;
265 qemu_timeval tv;
266
267 elem = virtqueue_pop(vq, sizeof(VirtQueueElement));
268 if (!elem) {
269 goto out;
270 }
271
272 if (s->stats_vq_elem != NULL) {
273 /* This should never happen if the driver follows the spec. */
274 virtqueue_push(vq, s->stats_vq_elem, 0);
275 virtio_notify(vdev, vq);
276 g_free(s->stats_vq_elem);
277 }
278
279 s->stats_vq_elem = elem;
280
281 /* Initialize the stats to get rid of any stale values. This is only
282 * needed to handle the case where a guest supports fewer stats than it
283 * used to (ie. it has booted into an old kernel).
284 */
285 reset_stats(s);
286
287 while (iov_to_buf(elem->out_sg, elem->out_num, offset, &stat, sizeof(stat))
288 == sizeof(stat)) {
289 uint16_t tag = virtio_tswap16(vdev, stat.tag);
290 uint64_t val = virtio_tswap64(vdev, stat.val);
291
292 offset += sizeof(stat);
293 if (tag < VIRTIO_BALLOON_S_NR)
294 s->stats[tag] = val;
295 }
296 s->stats_vq_offset = offset;
297
298 if (qemu_gettimeofday(&tv) < 0) {
299 warn_report("%s: failed to get time of day", __func__);
300 goto out;
301 }
302
303 s->stats_last_update = tv.tv_sec;
304
305 out:
306 if (balloon_stats_enabled(s)) {
307 balloon_stats_change_timer(s, s->stats_poll_interval);
308 }
309 }
310
311 static void virtio_balloon_get_config(VirtIODevice *vdev, uint8_t *config_data)
312 {
313 VirtIOBalloon *dev = VIRTIO_BALLOON(vdev);
314 struct virtio_balloon_config config;
315
316 config.num_pages = cpu_to_le32(dev->num_pages);
317 config.actual = cpu_to_le32(dev->actual);
318
319 trace_virtio_balloon_get_config(config.num_pages, config.actual);
320 memcpy(config_data, &config, sizeof(struct virtio_balloon_config));
321 }
322
323 static int build_dimm_list(Object *obj, void *opaque)
324 {
325 GSList **list = opaque;
326
327 if (object_dynamic_cast(obj, TYPE_PC_DIMM)) {
328 DeviceState *dev = DEVICE(obj);
329 if (dev->realized) { /* only realized DIMMs matter */
330 *list = g_slist_prepend(*list, dev);
331 }
332 }
333
334 object_child_foreach(obj, build_dimm_list, opaque);
335 return 0;
336 }
337
338 static ram_addr_t get_current_ram_size(void)
339 {
340 GSList *list = NULL, *item;
341 ram_addr_t size = ram_size;
342
343 build_dimm_list(qdev_get_machine(), &list);
344 for (item = list; item; item = g_slist_next(item)) {
345 Object *obj = OBJECT(item->data);
346 if (!strcmp(object_get_typename(obj), TYPE_PC_DIMM)) {
347 size += object_property_get_int(obj, PC_DIMM_SIZE_PROP,
348 &error_abort);
349 }
350 }
351 g_slist_free(list);
352
353 return size;
354 }
355
356 static void virtio_balloon_set_config(VirtIODevice *vdev,
357 const uint8_t *config_data)
358 {
359 VirtIOBalloon *dev = VIRTIO_BALLOON(vdev);
360 struct virtio_balloon_config config;
361 uint32_t oldactual = dev->actual;
362 ram_addr_t vm_ram_size = get_current_ram_size();
363
364 memcpy(&config, config_data, sizeof(struct virtio_balloon_config));
365 dev->actual = le32_to_cpu(config.actual);
366 if (dev->actual != oldactual) {
367 qapi_event_send_balloon_change(vm_ram_size -
368 ((ram_addr_t) dev->actual << VIRTIO_BALLOON_PFN_SHIFT),
369 &error_abort);
370 }
371 trace_virtio_balloon_set_config(dev->actual, oldactual);
372 }
373
374 static uint64_t virtio_balloon_get_features(VirtIODevice *vdev, uint64_t f,
375 Error **errp)
376 {
377 VirtIOBalloon *dev = VIRTIO_BALLOON(vdev);
378 f |= dev->host_features;
379 virtio_add_feature(&f, VIRTIO_BALLOON_F_STATS_VQ);
380 return f;
381 }
382
383 static void virtio_balloon_stat(void *opaque, BalloonInfo *info)
384 {
385 VirtIOBalloon *dev = opaque;
386 info->actual = get_current_ram_size() - ((uint64_t) dev->actual <<
387 VIRTIO_BALLOON_PFN_SHIFT);
388 }
389
390 static void virtio_balloon_to_target(void *opaque, ram_addr_t target)
391 {
392 VirtIOBalloon *dev = VIRTIO_BALLOON(opaque);
393 VirtIODevice *vdev = VIRTIO_DEVICE(dev);
394 ram_addr_t vm_ram_size = get_current_ram_size();
395
396 if (target > vm_ram_size) {
397 target = vm_ram_size;
398 }
399 if (target) {
400 dev->num_pages = (vm_ram_size - target) >> VIRTIO_BALLOON_PFN_SHIFT;
401 virtio_notify_config(vdev);
402 }
403 trace_virtio_balloon_to_target(target, dev->num_pages);
404 }
405
406 static int virtio_balloon_post_load_device(void *opaque, int version_id)
407 {
408 VirtIOBalloon *s = VIRTIO_BALLOON(opaque);
409
410 if (balloon_stats_enabled(s)) {
411 balloon_stats_change_timer(s, s->stats_poll_interval);
412 }
413 return 0;
414 }
415
416 static const VMStateDescription vmstate_virtio_balloon_device = {
417 .name = "virtio-balloon-device",
418 .version_id = 1,
419 .minimum_version_id = 1,
420 .post_load = virtio_balloon_post_load_device,
421 .fields = (VMStateField[]) {
422 VMSTATE_UINT32(num_pages, VirtIOBalloon),
423 VMSTATE_UINT32(actual, VirtIOBalloon),
424 VMSTATE_END_OF_LIST()
425 },
426 };
427
428 static void virtio_balloon_device_realize(DeviceState *dev, Error **errp)
429 {
430 VirtIODevice *vdev = VIRTIO_DEVICE(dev);
431 VirtIOBalloon *s = VIRTIO_BALLOON(dev);
432 int ret;
433
434 virtio_init(vdev, "virtio-balloon", VIRTIO_ID_BALLOON,
435 sizeof(struct virtio_balloon_config));
436
437 ret = qemu_add_balloon_handler(virtio_balloon_to_target,
438 virtio_balloon_stat, s);
439
440 if (ret < 0) {
441 error_setg(errp, "Only one balloon device is supported");
442 virtio_cleanup(vdev);
443 return;
444 }
445
446 s->ivq = virtio_add_queue(vdev, 128, virtio_balloon_handle_output);
447 s->dvq = virtio_add_queue(vdev, 128, virtio_balloon_handle_output);
448 s->svq = virtio_add_queue(vdev, 128, virtio_balloon_receive_stats);
449
450 reset_stats(s);
451 }
452
453 static void virtio_balloon_device_unrealize(DeviceState *dev, Error **errp)
454 {
455 VirtIODevice *vdev = VIRTIO_DEVICE(dev);
456 VirtIOBalloon *s = VIRTIO_BALLOON(dev);
457
458 balloon_stats_destroy_timer(s);
459 qemu_remove_balloon_handler(s);
460 virtio_cleanup(vdev);
461 }
462
463 static void virtio_balloon_device_reset(VirtIODevice *vdev)
464 {
465 VirtIOBalloon *s = VIRTIO_BALLOON(vdev);
466
467 if (s->stats_vq_elem != NULL) {
468 virtqueue_unpop(s->svq, s->stats_vq_elem, 0);
469 g_free(s->stats_vq_elem);
470 s->stats_vq_elem = NULL;
471 }
472 }
473
474 static void virtio_balloon_set_status(VirtIODevice *vdev, uint8_t status)
475 {
476 VirtIOBalloon *s = VIRTIO_BALLOON(vdev);
477
478 if (!s->stats_vq_elem && vdev->vm_running &&
479 (status & VIRTIO_CONFIG_S_DRIVER_OK) && virtqueue_rewind(s->svq, 1)) {
480 /* poll stats queue for the element we have discarded when the VM
481 * was stopped */
482 virtio_balloon_receive_stats(vdev, s->svq);
483 }
484 }
485
486 static void virtio_balloon_instance_init(Object *obj)
487 {
488 VirtIOBalloon *s = VIRTIO_BALLOON(obj);
489
490 object_property_add(obj, "guest-stats", "guest statistics",
491 balloon_stats_get_all, NULL, NULL, s, NULL);
492
493 object_property_add(obj, "guest-stats-polling-interval", "int",
494 balloon_stats_get_poll_interval,
495 balloon_stats_set_poll_interval,
496 NULL, s, NULL);
497 }
498
499 static const VMStateDescription vmstate_virtio_balloon = {
500 .name = "virtio-balloon",
501 .minimum_version_id = 1,
502 .version_id = 1,
503 .fields = (VMStateField[]) {
504 VMSTATE_VIRTIO_DEVICE,
505 VMSTATE_END_OF_LIST()
506 },
507 };
508
509 static Property virtio_balloon_properties[] = {
510 DEFINE_PROP_BIT("deflate-on-oom", VirtIOBalloon, host_features,
511 VIRTIO_BALLOON_F_DEFLATE_ON_OOM, false),
512 DEFINE_PROP_END_OF_LIST(),
513 };
514
515 static void virtio_balloon_class_init(ObjectClass *klass, void *data)
516 {
517 DeviceClass *dc = DEVICE_CLASS(klass);
518 VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
519
520 dc->props = virtio_balloon_properties;
521 dc->vmsd = &vmstate_virtio_balloon;
522 set_bit(DEVICE_CATEGORY_MISC, dc->categories);
523 vdc->realize = virtio_balloon_device_realize;
524 vdc->unrealize = virtio_balloon_device_unrealize;
525 vdc->reset = virtio_balloon_device_reset;
526 vdc->get_config = virtio_balloon_get_config;
527 vdc->set_config = virtio_balloon_set_config;
528 vdc->get_features = virtio_balloon_get_features;
529 vdc->set_status = virtio_balloon_set_status;
530 vdc->vmsd = &vmstate_virtio_balloon_device;
531 }
532
533 static const TypeInfo virtio_balloon_info = {
534 .name = TYPE_VIRTIO_BALLOON,
535 .parent = TYPE_VIRTIO_DEVICE,
536 .instance_size = sizeof(VirtIOBalloon),
537 .instance_init = virtio_balloon_instance_init,
538 .class_init = virtio_balloon_class_init,
539 };
540
541 static void virtio_register_types(void)
542 {
543 type_register_static(&virtio_balloon_info);
544 }
545
546 type_init(virtio_register_types)