xref: /openbmc/qemu/hw/virtio/virtio-balloon.c (revision dfbd2768) 
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/i386/pc.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/visitor.h"
27 #include "qapi-event.h"
28 #include "trace.h"
29 
30 #include "hw/virtio/virtio-bus.h"
31 #include "hw/virtio/virtio-access.h"
32 
33 #define BALLOON_PAGE_SIZE  (1 << VIRTIO_BALLOON_PFN_SHIFT)
34 
35 static void balloon_page(void *addr, int deflate)
36 {
37     if (!qemu_balloon_is_inhibited() && (!kvm_enabled() ||
38                                          kvm_has_sync_mmu())) {
39         qemu_madvise(addr, BALLOON_PAGE_SIZE,
40                 deflate ? QEMU_MADV_WILLNEED : QEMU_MADV_DONTNEED);
41     }
42 }
43 
44 static const char *balloon_stat_names[] = {
45    [VIRTIO_BALLOON_S_SWAP_IN] = "stat-swap-in",
46    [VIRTIO_BALLOON_S_SWAP_OUT] = "stat-swap-out",
47    [VIRTIO_BALLOON_S_MAJFLT] = "stat-major-faults",
48    [VIRTIO_BALLOON_S_MINFLT] = "stat-minor-faults",
49    [VIRTIO_BALLOON_S_MEMFREE] = "stat-free-memory",
50    [VIRTIO_BALLOON_S_MEMTOT] = "stat-total-memory",
51    [VIRTIO_BALLOON_S_AVAIL] = "stat-available-memory",
52    [VIRTIO_BALLOON_S_NR] = NULL
53 };
54 
55 /*
56  * reset_stats - Mark all items in the stats array as unset
57  *
58  * This function needs to be called at device initialization and before
59  * updating to a set of newly-generated stats.  This will ensure that no
60  * stale values stick around in case the guest reports a subset of the supported
61  * statistics.
62  */
63 static inline void reset_stats(VirtIOBalloon *dev)
64 {
65     int i;
66     for (i = 0; i < VIRTIO_BALLOON_S_NR; dev->stats[i++] = -1);
67 }
68 
69 static bool balloon_stats_supported(const VirtIOBalloon *s)
70 {
71     VirtIODevice *vdev = VIRTIO_DEVICE(s);
72     return virtio_vdev_has_feature(vdev, VIRTIO_BALLOON_F_STATS_VQ);
73 }
74 
75 static bool balloon_stats_enabled(const VirtIOBalloon *s)
76 {
77     return s->stats_poll_interval > 0;
78 }
79 
80 static void balloon_stats_destroy_timer(VirtIOBalloon *s)
81 {
82     if (balloon_stats_enabled(s)) {
83         timer_del(s->stats_timer);
84         timer_free(s->stats_timer);
85         s->stats_timer = NULL;
86         s->stats_poll_interval = 0;
87     }
88 }
89 
90 static void balloon_stats_change_timer(VirtIOBalloon *s, int64_t secs)
91 {
92     timer_mod(s->stats_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + secs * 1000);
93 }
94 
95 static void balloon_stats_poll_cb(void *opaque)
96 {
97     VirtIOBalloon *s = opaque;
98     VirtIODevice *vdev = VIRTIO_DEVICE(s);
99 
100     if (s->stats_vq_elem == NULL || !balloon_stats_supported(s)) {
101         /* re-schedule */
102         balloon_stats_change_timer(s, s->stats_poll_interval);
103         return;
104     }
105 
106     virtqueue_push(s->svq, s->stats_vq_elem, s->stats_vq_offset);
107     virtio_notify(vdev, s->svq);
108     g_free(s->stats_vq_elem);
109     s->stats_vq_elem = NULL;
110 }
111 
112 static void balloon_stats_get_all(Object *obj, Visitor *v, const char *name,
113                                   void *opaque, Error **errp)
114 {
115     Error *err = NULL;
116     VirtIOBalloon *s = opaque;
117     int i;
118 
119     visit_start_struct(v, name, NULL, 0, &err);
120     if (err) {
121         goto out;
122     }
123     visit_type_int(v, "last-update", &s->stats_last_update, &err);
124     if (err) {
125         goto out_end;
126     }
127 
128     visit_start_struct(v, "stats", NULL, 0, &err);
129     if (err) {
130         goto out_end;
131     }
132     for (i = 0; i < VIRTIO_BALLOON_S_NR; i++) {
133         visit_type_uint64(v, balloon_stat_names[i], &s->stats[i], &err);
134         if (err) {
135             goto out_nested;
136         }
137     }
138     visit_check_struct(v, &err);
139 out_nested:
140     visit_end_struct(v, NULL);
141 
142     if (!err) {
143         visit_check_struct(v, &err);
144     }
145 out_end:
146     visit_end_struct(v, NULL);
147 out:
148     error_propagate(errp, err);
149 }
150 
151 static void balloon_stats_get_poll_interval(Object *obj, Visitor *v,
152                                             const char *name, void *opaque,
153                                             Error **errp)
154 {
155     VirtIOBalloon *s = opaque;
156     visit_type_int(v, name, &s->stats_poll_interval, errp);
157 }
158 
159 static void balloon_stats_set_poll_interval(Object *obj, Visitor *v,
160                                             const char *name, void *opaque,
161                                             Error **errp)
162 {
163     VirtIOBalloon *s = opaque;
164     Error *local_err = NULL;
165     int64_t value;
166 
167     visit_type_int(v, name, &value, &local_err);
168     if (local_err) {
169         error_propagate(errp, local_err);
170         return;
171     }
172 
173     if (value < 0) {
174         error_setg(errp, "timer value must be greater than zero");
175         return;
176     }
177 
178     if (value > UINT32_MAX) {
179         error_setg(errp, "timer value is too big");
180         return;
181     }
182 
183     if (value == s->stats_poll_interval) {
184         return;
185     }
186 
187     if (value == 0) {
188         /* timer=0 disables the timer */
189         balloon_stats_destroy_timer(s);
190         return;
191     }
192 
193     if (balloon_stats_enabled(s)) {
194         /* timer interval change */
195         s->stats_poll_interval = value;
196         balloon_stats_change_timer(s, value);
197         return;
198     }
199 
200     /* create a new timer */
201     g_assert(s->stats_timer == NULL);
202     s->stats_timer = timer_new_ms(QEMU_CLOCK_VIRTUAL, balloon_stats_poll_cb, s);
203     s->stats_poll_interval = value;
204     balloon_stats_change_timer(s, 0);
205 }
206 
207 static void virtio_balloon_handle_output(VirtIODevice *vdev, VirtQueue *vq)
208 {
209     VirtIOBalloon *s = VIRTIO_BALLOON(vdev);
210     VirtQueueElement *elem;
211     MemoryRegionSection section;
212 
213     for (;;) {
214         size_t offset = 0;
215         uint32_t pfn;
216         elem = virtqueue_pop(vq, sizeof(VirtQueueElement));
217         if (!elem) {
218             return;
219         }
220 
221         while (iov_to_buf(elem->out_sg, elem->out_num, offset, &pfn, 4) == 4) {
222             ram_addr_t pa;
223             ram_addr_t addr;
224             int p = virtio_ldl_p(vdev, &pfn);
225 
226             pa = (ram_addr_t) p << VIRTIO_BALLOON_PFN_SHIFT;
227             offset += 4;
228 
229             /* FIXME: remove get_system_memory(), but how? */
230             section = memory_region_find(get_system_memory(), pa, 1);
231             if (!int128_nz(section.size) || !memory_region_is_ram(section.mr))
232                 continue;
233 
234             trace_virtio_balloon_handle_output(memory_region_name(section.mr),
235                                                pa);
236             /* Using memory_region_get_ram_ptr is bending the rules a bit, but
237                should be OK because we only want a single page.  */
238             addr = section.offset_within_region;
239             balloon_page(memory_region_get_ram_ptr(section.mr) + addr,
240                          !!(vq == s->dvq));
241             memory_region_unref(section.mr);
242         }
243 
244         virtqueue_push(vq, elem, offset);
245         virtio_notify(vdev, vq);
246         g_free(elem);
247     }
248 }
249 
250 static void virtio_balloon_receive_stats(VirtIODevice *vdev, VirtQueue *vq)
251 {
252     VirtIOBalloon *s = VIRTIO_BALLOON(vdev);
253     VirtQueueElement *elem;
254     VirtIOBalloonStat stat;
255     size_t offset = 0;
256     qemu_timeval tv;
257 
258     elem = virtqueue_pop(vq, sizeof(VirtQueueElement));
259     if (!elem) {
260         goto out;
261     }
262 
263     if (s->stats_vq_elem != NULL) {
264         /* This should never happen if the driver follows the spec. */
265         virtqueue_push(vq, s->stats_vq_elem, 0);
266         virtio_notify(vdev, vq);
267         g_free(s->stats_vq_elem);
268     }
269 
270     s->stats_vq_elem = elem;
271 
272     /* Initialize the stats to get rid of any stale values.  This is only
273      * needed to handle the case where a guest supports fewer stats than it
274      * used to (ie. it has booted into an old kernel).
275      */
276     reset_stats(s);
277 
278     while (iov_to_buf(elem->out_sg, elem->out_num, offset, &stat, sizeof(stat))
279            == sizeof(stat)) {
280         uint16_t tag = virtio_tswap16(vdev, stat.tag);
281         uint64_t val = virtio_tswap64(vdev, stat.val);
282 
283         offset += sizeof(stat);
284         if (tag < VIRTIO_BALLOON_S_NR)
285             s->stats[tag] = val;
286     }
287     s->stats_vq_offset = offset;
288 
289     if (qemu_gettimeofday(&tv) < 0) {
290         fprintf(stderr, "warning: %s: failed to get time of day\n", __func__);
291         goto out;
292     }
293 
294     s->stats_last_update = tv.tv_sec;
295 
296 out:
297     if (balloon_stats_enabled(s)) {
298         balloon_stats_change_timer(s, s->stats_poll_interval);
299     }
300 }
301 
302 static void virtio_balloon_get_config(VirtIODevice *vdev, uint8_t *config_data)
303 {
304     VirtIOBalloon *dev = VIRTIO_BALLOON(vdev);
305     struct virtio_balloon_config config;
306 
307     config.num_pages = cpu_to_le32(dev->num_pages);
308     config.actual = cpu_to_le32(dev->actual);
309 
310     trace_virtio_balloon_get_config(config.num_pages, config.actual);
311     memcpy(config_data, &config, sizeof(struct virtio_balloon_config));
312 }
313 
314 static int build_dimm_list(Object *obj, void *opaque)
315 {
316     GSList **list = opaque;
317 
318     if (object_dynamic_cast(obj, TYPE_PC_DIMM)) {
319         DeviceState *dev = DEVICE(obj);
320         if (dev->realized) { /* only realized DIMMs matter */
321             *list = g_slist_prepend(*list, dev);
322         }
323     }
324 
325     object_child_foreach(obj, build_dimm_list, opaque);
326     return 0;
327 }
328 
329 static ram_addr_t get_current_ram_size(void)
330 {
331     GSList *list = NULL, *item;
332     ram_addr_t size = ram_size;
333 
334     build_dimm_list(qdev_get_machine(), &list);
335     for (item = list; item; item = g_slist_next(item)) {
336         Object *obj = OBJECT(item->data);
337         if (!strcmp(object_get_typename(obj), TYPE_PC_DIMM)) {
338             size += object_property_get_int(obj, PC_DIMM_SIZE_PROP,
339                                             &error_abort);
340         }
341     }
342     g_slist_free(list);
343 
344     return size;
345 }
346 
347 static void virtio_balloon_set_config(VirtIODevice *vdev,
348                                       const uint8_t *config_data)
349 {
350     VirtIOBalloon *dev = VIRTIO_BALLOON(vdev);
351     struct virtio_balloon_config config;
352     uint32_t oldactual = dev->actual;
353     ram_addr_t vm_ram_size = get_current_ram_size();
354 
355     memcpy(&config, config_data, sizeof(struct virtio_balloon_config));
356     dev->actual = le32_to_cpu(config.actual);
357     if (dev->actual != oldactual) {
358         qapi_event_send_balloon_change(vm_ram_size -
359                         ((ram_addr_t) dev->actual << VIRTIO_BALLOON_PFN_SHIFT),
360                         &error_abort);
361     }
362     trace_virtio_balloon_set_config(dev->actual, oldactual);
363 }
364 
365 static uint64_t virtio_balloon_get_features(VirtIODevice *vdev, uint64_t f,
366                                             Error **errp)
367 {
368     VirtIOBalloon *dev = VIRTIO_BALLOON(vdev);
369     f |= dev->host_features;
370     virtio_add_feature(&f, VIRTIO_BALLOON_F_STATS_VQ);
371     return f;
372 }
373 
374 static void virtio_balloon_stat(void *opaque, BalloonInfo *info)
375 {
376     VirtIOBalloon *dev = opaque;
377     info->actual = get_current_ram_size() - ((uint64_t) dev->actual <<
378                                              VIRTIO_BALLOON_PFN_SHIFT);
379 }
380 
381 static void virtio_balloon_to_target(void *opaque, ram_addr_t target)
382 {
383     VirtIOBalloon *dev = VIRTIO_BALLOON(opaque);
384     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
385     ram_addr_t vm_ram_size = get_current_ram_size();
386 
387     if (target > vm_ram_size) {
388         target = vm_ram_size;
389     }
390     if (target) {
391         dev->num_pages = (vm_ram_size - target) >> VIRTIO_BALLOON_PFN_SHIFT;
392         virtio_notify_config(vdev);
393     }
394     trace_virtio_balloon_to_target(target, dev->num_pages);
395 }
396 
397 static int virtio_balloon_post_load_device(void *opaque, int version_id)
398 {
399     VirtIOBalloon *s = VIRTIO_BALLOON(opaque);
400 
401     if (balloon_stats_enabled(s)) {
402         balloon_stats_change_timer(s, s->stats_poll_interval);
403     }
404     return 0;
405 }
406 
407 static const VMStateDescription vmstate_virtio_balloon_device = {
408     .name = "virtio-balloon-device",
409     .version_id = 1,
410     .minimum_version_id = 1,
411     .post_load = virtio_balloon_post_load_device,
412     .fields = (VMStateField[]) {
413         VMSTATE_UINT32(num_pages, VirtIOBalloon),
414         VMSTATE_UINT32(actual, VirtIOBalloon),
415         VMSTATE_END_OF_LIST()
416     },
417 };
418 
419 static void virtio_balloon_device_realize(DeviceState *dev, Error **errp)
420 {
421     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
422     VirtIOBalloon *s = VIRTIO_BALLOON(dev);
423     int ret;
424 
425     virtio_init(vdev, "virtio-balloon", VIRTIO_ID_BALLOON,
426                 sizeof(struct virtio_balloon_config));
427 
428     ret = qemu_add_balloon_handler(virtio_balloon_to_target,
429                                    virtio_balloon_stat, s);
430 
431     if (ret < 0) {
432         error_setg(errp, "Only one balloon device is supported");
433         virtio_cleanup(vdev);
434         return;
435     }
436 
437     s->ivq = virtio_add_queue(vdev, 128, virtio_balloon_handle_output);
438     s->dvq = virtio_add_queue(vdev, 128, virtio_balloon_handle_output);
439     s->svq = virtio_add_queue(vdev, 128, virtio_balloon_receive_stats);
440 
441     reset_stats(s);
442 }
443 
444 static void virtio_balloon_device_unrealize(DeviceState *dev, Error **errp)
445 {
446     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
447     VirtIOBalloon *s = VIRTIO_BALLOON(dev);
448 
449     balloon_stats_destroy_timer(s);
450     qemu_remove_balloon_handler(s);
451     virtio_cleanup(vdev);
452 }
453 
454 static void virtio_balloon_device_reset(VirtIODevice *vdev)
455 {
456     VirtIOBalloon *s = VIRTIO_BALLOON(vdev);
457 
458     if (s->stats_vq_elem != NULL) {
459         virtqueue_unpop(s->svq, s->stats_vq_elem, 0);
460         g_free(s->stats_vq_elem);
461         s->stats_vq_elem = NULL;
462     }
463 }
464 
465 static void virtio_balloon_set_status(VirtIODevice *vdev, uint8_t status)
466 {
467     VirtIOBalloon *s = VIRTIO_BALLOON(vdev);
468 
469     if (!s->stats_vq_elem && vdev->vm_running &&
470         (status & VIRTIO_CONFIG_S_DRIVER_OK) && virtqueue_rewind(s->svq, 1)) {
471         /* poll stats queue for the element we have discarded when the VM
472          * was stopped */
473         virtio_balloon_receive_stats(vdev, s->svq);
474     }
475 }
476 
477 static void virtio_balloon_instance_init(Object *obj)
478 {
479     VirtIOBalloon *s = VIRTIO_BALLOON(obj);
480 
481     object_property_add(obj, "guest-stats", "guest statistics",
482                         balloon_stats_get_all, NULL, NULL, s, NULL);
483 
484     object_property_add(obj, "guest-stats-polling-interval", "int",
485                         balloon_stats_get_poll_interval,
486                         balloon_stats_set_poll_interval,
487                         NULL, s, NULL);
488 }
489 
490 static const VMStateDescription vmstate_virtio_balloon = {
491     .name = "virtio-balloon",
492     .minimum_version_id = 1,
493     .version_id = 1,
494     .fields = (VMStateField[]) {
495         VMSTATE_VIRTIO_DEVICE,
496         VMSTATE_END_OF_LIST()
497     },
498 };
499 
500 static Property virtio_balloon_properties[] = {
501     DEFINE_PROP_BIT("deflate-on-oom", VirtIOBalloon, host_features,
502                     VIRTIO_BALLOON_F_DEFLATE_ON_OOM, false),
503     DEFINE_PROP_END_OF_LIST(),
504 };
505 
506 static void virtio_balloon_class_init(ObjectClass *klass, void *data)
507 {
508     DeviceClass *dc = DEVICE_CLASS(klass);
509     VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
510 
511     dc->props = virtio_balloon_properties;
512     dc->vmsd = &vmstate_virtio_balloon;
513     set_bit(DEVICE_CATEGORY_MISC, dc->categories);
514     vdc->realize = virtio_balloon_device_realize;
515     vdc->unrealize = virtio_balloon_device_unrealize;
516     vdc->reset = virtio_balloon_device_reset;
517     vdc->get_config = virtio_balloon_get_config;
518     vdc->set_config = virtio_balloon_set_config;
519     vdc->get_features = virtio_balloon_get_features;
520     vdc->set_status = virtio_balloon_set_status;
521     vdc->vmsd = &vmstate_virtio_balloon_device;
522 }
523 
524 static const TypeInfo virtio_balloon_info = {
525     .name = TYPE_VIRTIO_BALLOON,
526     .parent = TYPE_VIRTIO_DEVICE,
527     .instance_size = sizeof(VirtIOBalloon),
528     .instance_init = virtio_balloon_instance_init,
529     .class_init = virtio_balloon_class_init,
530 };
531 
532 static void virtio_register_types(void)
533 {
534     type_register_static(&virtio_balloon_info);
535 }
536 
537 type_init(virtio_register_types)
538