xref: /openbmc/qemu/hw/virtio/virtio-balloon.c (revision f764718d) 
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/visitor.h"
27 #include "qapi-event.h"
28 #include "trace.h"
29 #include "qemu/error-report.h"
30 
31 #include "hw/virtio/virtio-bus.h"
32 #include "hw/virtio/virtio-access.h"
33 
34 #define BALLOON_PAGE_SIZE  (1 << VIRTIO_BALLOON_PFN_SHIFT)
35 
36 static void balloon_page(void *addr, int deflate)
37 {
38     if (!qemu_balloon_is_inhibited() && (!kvm_enabled() ||
39                                          kvm_has_sync_mmu())) {
40         qemu_madvise(addr, BALLOON_PAGE_SIZE,
41                 deflate ? QEMU_MADV_WILLNEED : QEMU_MADV_DONTNEED);
42     }
43 }
44 
45 static const char *balloon_stat_names[] = {
46    [VIRTIO_BALLOON_S_SWAP_IN] = "stat-swap-in",
47    [VIRTIO_BALLOON_S_SWAP_OUT] = "stat-swap-out",
48    [VIRTIO_BALLOON_S_MAJFLT] = "stat-major-faults",
49    [VIRTIO_BALLOON_S_MINFLT] = "stat-minor-faults",
50    [VIRTIO_BALLOON_S_MEMFREE] = "stat-free-memory",
51    [VIRTIO_BALLOON_S_MEMTOT] = "stat-total-memory",
52    [VIRTIO_BALLOON_S_AVAIL] = "stat-available-memory",
53    [VIRTIO_BALLOON_S_NR] = NULL
54 };
55 
56 /*
57  * reset_stats - Mark all items in the stats array as unset
58  *
59  * This function needs to be called at device initialization and before
60  * updating to a set of newly-generated stats.  This will ensure that no
61  * stale values stick around in case the guest reports a subset of the supported
62  * statistics.
63  */
64 static inline void reset_stats(VirtIOBalloon *dev)
65 {
66     int i;
67     for (i = 0; i < VIRTIO_BALLOON_S_NR; dev->stats[i++] = -1);
68 }
69 
70 static bool balloon_stats_supported(const VirtIOBalloon *s)
71 {
72     VirtIODevice *vdev = VIRTIO_DEVICE(s);
73     return virtio_vdev_has_feature(vdev, VIRTIO_BALLOON_F_STATS_VQ);
74 }
75 
76 static bool balloon_stats_enabled(const VirtIOBalloon *s)
77 {
78     return s->stats_poll_interval > 0;
79 }
80 
81 static void balloon_stats_destroy_timer(VirtIOBalloon *s)
82 {
83     if (balloon_stats_enabled(s)) {
84         timer_del(s->stats_timer);
85         timer_free(s->stats_timer);
86         s->stats_timer = NULL;
87         s->stats_poll_interval = 0;
88     }
89 }
90 
91 static void balloon_stats_change_timer(VirtIOBalloon *s, int64_t secs)
92 {
93     timer_mod(s->stats_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + secs * 1000);
94 }
95 
96 static void balloon_stats_poll_cb(void *opaque)
97 {
98     VirtIOBalloon *s = opaque;
99     VirtIODevice *vdev = VIRTIO_DEVICE(s);
100 
101     if (s->stats_vq_elem == NULL || !balloon_stats_supported(s)) {
102         /* re-schedule */
103         balloon_stats_change_timer(s, s->stats_poll_interval);
104         return;
105     }
106 
107     virtqueue_push(s->svq, s->stats_vq_elem, s->stats_vq_offset);
108     virtio_notify(vdev, s->svq);
109     g_free(s->stats_vq_elem);
110     s->stats_vq_elem = NULL;
111 }
112 
113 static void balloon_stats_get_all(Object *obj, Visitor *v, const char *name,
114                                   void *opaque, Error **errp)
115 {
116     Error *err = NULL;
117     VirtIOBalloon *s = opaque;
118     int i;
119 
120     visit_start_struct(v, name, NULL, 0, &err);
121     if (err) {
122         goto out;
123     }
124     visit_type_int(v, "last-update", &s->stats_last_update, &err);
125     if (err) {
126         goto out_end;
127     }
128 
129     visit_start_struct(v, "stats", NULL, 0, &err);
130     if (err) {
131         goto out_end;
132     }
133     for (i = 0; i < VIRTIO_BALLOON_S_NR; i++) {
134         visit_type_uint64(v, balloon_stat_names[i], &s->stats[i], &err);
135         if (err) {
136             goto out_nested;
137         }
138     }
139     visit_check_struct(v, &err);
140 out_nested:
141     visit_end_struct(v, NULL);
142 
143     if (!err) {
144         visit_check_struct(v, &err);
145     }
146 out_end:
147     visit_end_struct(v, NULL);
148 out:
149     error_propagate(errp, err);
150 }
151 
152 static void balloon_stats_get_poll_interval(Object *obj, Visitor *v,
153                                             const char *name, void *opaque,
154                                             Error **errp)
155 {
156     VirtIOBalloon *s = opaque;
157     visit_type_int(v, name, &s->stats_poll_interval, errp);
158 }
159 
160 static void balloon_stats_set_poll_interval(Object *obj, Visitor *v,
161                                             const char *name, void *opaque,
162                                             Error **errp)
163 {
164     VirtIOBalloon *s = opaque;
165     Error *local_err = NULL;
166     int64_t value;
167 
168     visit_type_int(v, name, &value, &local_err);
169     if (local_err) {
170         error_propagate(errp, local_err);
171         return;
172     }
173 
174     if (value < 0) {
175         error_setg(errp, "timer value must be greater than zero");
176         return;
177     }
178 
179     if (value > UINT32_MAX) {
180         error_setg(errp, "timer value is too big");
181         return;
182     }
183 
184     if (value == s->stats_poll_interval) {
185         return;
186     }
187 
188     if (value == 0) {
189         /* timer=0 disables the timer */
190         balloon_stats_destroy_timer(s);
191         return;
192     }
193 
194     if (balloon_stats_enabled(s)) {
195         /* timer interval change */
196         s->stats_poll_interval = value;
197         balloon_stats_change_timer(s, value);
198         return;
199     }
200 
201     /* create a new timer */
202     g_assert(s->stats_timer == NULL);
203     s->stats_timer = timer_new_ms(QEMU_CLOCK_VIRTUAL, balloon_stats_poll_cb, s);
204     s->stats_poll_interval = value;
205     balloon_stats_change_timer(s, 0);
206 }
207 
208 static void virtio_balloon_handle_output(VirtIODevice *vdev, VirtQueue *vq)
209 {
210     VirtIOBalloon *s = VIRTIO_BALLOON(vdev);
211     VirtQueueElement *elem;
212     MemoryRegionSection section;
213 
214     for (;;) {
215         size_t offset = 0;
216         uint32_t pfn;
217         elem = virtqueue_pop(vq, sizeof(VirtQueueElement));
218         if (!elem) {
219             return;
220         }
221 
222         while (iov_to_buf(elem->out_sg, elem->out_num, offset, &pfn, 4) == 4) {
223             ram_addr_t pa;
224             ram_addr_t addr;
225             int p = virtio_ldl_p(vdev, &pfn);
226 
227             pa = (ram_addr_t) p << VIRTIO_BALLOON_PFN_SHIFT;
228             offset += 4;
229 
230             /* FIXME: remove get_system_memory(), but how? */
231             section = memory_region_find(get_system_memory(), pa, 1);
232             if (!int128_nz(section.size) ||
233                 !memory_region_is_ram(section.mr) ||
234                 memory_region_is_rom(section.mr) ||
235                 memory_region_is_romd(section.mr)) {
236                 trace_virtio_balloon_bad_addr(pa);
237                 continue;
238             }
239 
240             trace_virtio_balloon_handle_output(memory_region_name(section.mr),
241                                                pa);
242             /* Using memory_region_get_ram_ptr is bending the rules a bit, but
243                should be OK because we only want a single page.  */
244             addr = section.offset_within_region;
245             balloon_page(memory_region_get_ram_ptr(section.mr) + addr,
246                          !!(vq == s->dvq));
247             memory_region_unref(section.mr);
248         }
249 
250         virtqueue_push(vq, elem, offset);
251         virtio_notify(vdev, vq);
252         g_free(elem);
253     }
254 }
255 
256 static void virtio_balloon_receive_stats(VirtIODevice *vdev, VirtQueue *vq)
257 {
258     VirtIOBalloon *s = VIRTIO_BALLOON(vdev);
259     VirtQueueElement *elem;
260     VirtIOBalloonStat stat;
261     size_t offset = 0;
262     qemu_timeval tv;
263 
264     elem = virtqueue_pop(vq, sizeof(VirtQueueElement));
265     if (!elem) {
266         goto out;
267     }
268 
269     if (s->stats_vq_elem != NULL) {
270         /* This should never happen if the driver follows the spec. */
271         virtqueue_push(vq, s->stats_vq_elem, 0);
272         virtio_notify(vdev, vq);
273         g_free(s->stats_vq_elem);
274     }
275 
276     s->stats_vq_elem = elem;
277 
278     /* Initialize the stats to get rid of any stale values.  This is only
279      * needed to handle the case where a guest supports fewer stats than it
280      * used to (ie. it has booted into an old kernel).
281      */
282     reset_stats(s);
283 
284     while (iov_to_buf(elem->out_sg, elem->out_num, offset, &stat, sizeof(stat))
285            == sizeof(stat)) {
286         uint16_t tag = virtio_tswap16(vdev, stat.tag);
287         uint64_t val = virtio_tswap64(vdev, stat.val);
288 
289         offset += sizeof(stat);
290         if (tag < VIRTIO_BALLOON_S_NR)
291             s->stats[tag] = val;
292     }
293     s->stats_vq_offset = offset;
294 
295     if (qemu_gettimeofday(&tv) < 0) {
296         warn_report("%s: failed to get time of day", __func__);
297         goto out;
298     }
299 
300     s->stats_last_update = tv.tv_sec;
301 
302 out:
303     if (balloon_stats_enabled(s)) {
304         balloon_stats_change_timer(s, s->stats_poll_interval);
305     }
306 }
307 
308 static void virtio_balloon_get_config(VirtIODevice *vdev, uint8_t *config_data)
309 {
310     VirtIOBalloon *dev = VIRTIO_BALLOON(vdev);
311     struct virtio_balloon_config config;
312 
313     config.num_pages = cpu_to_le32(dev->num_pages);
314     config.actual = cpu_to_le32(dev->actual);
315 
316     trace_virtio_balloon_get_config(config.num_pages, config.actual);
317     memcpy(config_data, &config, sizeof(struct virtio_balloon_config));
318 }
319 
320 static int build_dimm_list(Object *obj, void *opaque)
321 {
322     GSList **list = opaque;
323 
324     if (object_dynamic_cast(obj, TYPE_PC_DIMM)) {
325         DeviceState *dev = DEVICE(obj);
326         if (dev->realized) { /* only realized DIMMs matter */
327             *list = g_slist_prepend(*list, dev);
328         }
329     }
330 
331     object_child_foreach(obj, build_dimm_list, opaque);
332     return 0;
333 }
334 
335 static ram_addr_t get_current_ram_size(void)
336 {
337     GSList *list = NULL, *item;
338     ram_addr_t size = ram_size;
339 
340     build_dimm_list(qdev_get_machine(), &list);
341     for (item = list; item; item = g_slist_next(item)) {
342         Object *obj = OBJECT(item->data);
343         if (!strcmp(object_get_typename(obj), TYPE_PC_DIMM)) {
344             size += object_property_get_int(obj, PC_DIMM_SIZE_PROP,
345                                             &error_abort);
346         }
347     }
348     g_slist_free(list);
349 
350     return size;
351 }
352 
353 static void virtio_balloon_set_config(VirtIODevice *vdev,
354                                       const uint8_t *config_data)
355 {
356     VirtIOBalloon *dev = VIRTIO_BALLOON(vdev);
357     struct virtio_balloon_config config;
358     uint32_t oldactual = dev->actual;
359     ram_addr_t vm_ram_size = get_current_ram_size();
360 
361     memcpy(&config, config_data, sizeof(struct virtio_balloon_config));
362     dev->actual = le32_to_cpu(config.actual);
363     if (dev->actual != oldactual) {
364         qapi_event_send_balloon_change(vm_ram_size -
365                         ((ram_addr_t) dev->actual << VIRTIO_BALLOON_PFN_SHIFT),
366                         &error_abort);
367     }
368     trace_virtio_balloon_set_config(dev->actual, oldactual);
369 }
370 
371 static uint64_t virtio_balloon_get_features(VirtIODevice *vdev, uint64_t f,
372                                             Error **errp)
373 {
374     VirtIOBalloon *dev = VIRTIO_BALLOON(vdev);
375     f |= dev->host_features;
376     virtio_add_feature(&f, VIRTIO_BALLOON_F_STATS_VQ);
377     return f;
378 }
379 
380 static void virtio_balloon_stat(void *opaque, BalloonInfo *info)
381 {
382     VirtIOBalloon *dev = opaque;
383     info->actual = get_current_ram_size() - ((uint64_t) dev->actual <<
384                                              VIRTIO_BALLOON_PFN_SHIFT);
385 }
386 
387 static void virtio_balloon_to_target(void *opaque, ram_addr_t target)
388 {
389     VirtIOBalloon *dev = VIRTIO_BALLOON(opaque);
390     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
391     ram_addr_t vm_ram_size = get_current_ram_size();
392 
393     if (target > vm_ram_size) {
394         target = vm_ram_size;
395     }
396     if (target) {
397         dev->num_pages = (vm_ram_size - target) >> VIRTIO_BALLOON_PFN_SHIFT;
398         virtio_notify_config(vdev);
399     }
400     trace_virtio_balloon_to_target(target, dev->num_pages);
401 }
402 
403 static int virtio_balloon_post_load_device(void *opaque, int version_id)
404 {
405     VirtIOBalloon *s = VIRTIO_BALLOON(opaque);
406 
407     if (balloon_stats_enabled(s)) {
408         balloon_stats_change_timer(s, s->stats_poll_interval);
409     }
410     return 0;
411 }
412 
413 static const VMStateDescription vmstate_virtio_balloon_device = {
414     .name = "virtio-balloon-device",
415     .version_id = 1,
416     .minimum_version_id = 1,
417     .post_load = virtio_balloon_post_load_device,
418     .fields = (VMStateField[]) {
419         VMSTATE_UINT32(num_pages, VirtIOBalloon),
420         VMSTATE_UINT32(actual, VirtIOBalloon),
421         VMSTATE_END_OF_LIST()
422     },
423 };
424 
425 static void virtio_balloon_device_realize(DeviceState *dev, Error **errp)
426 {
427     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
428     VirtIOBalloon *s = VIRTIO_BALLOON(dev);
429     int ret;
430 
431     virtio_init(vdev, "virtio-balloon", VIRTIO_ID_BALLOON,
432                 sizeof(struct virtio_balloon_config));
433 
434     ret = qemu_add_balloon_handler(virtio_balloon_to_target,
435                                    virtio_balloon_stat, s);
436 
437     if (ret < 0) {
438         error_setg(errp, "Only one balloon device is supported");
439         virtio_cleanup(vdev);
440         return;
441     }
442 
443     s->ivq = virtio_add_queue(vdev, 128, virtio_balloon_handle_output);
444     s->dvq = virtio_add_queue(vdev, 128, virtio_balloon_handle_output);
445     s->svq = virtio_add_queue(vdev, 128, virtio_balloon_receive_stats);
446 
447     reset_stats(s);
448 }
449 
450 static void virtio_balloon_device_unrealize(DeviceState *dev, Error **errp)
451 {
452     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
453     VirtIOBalloon *s = VIRTIO_BALLOON(dev);
454 
455     balloon_stats_destroy_timer(s);
456     qemu_remove_balloon_handler(s);
457     virtio_cleanup(vdev);
458 }
459 
460 static void virtio_balloon_device_reset(VirtIODevice *vdev)
461 {
462     VirtIOBalloon *s = VIRTIO_BALLOON(vdev);
463 
464     if (s->stats_vq_elem != NULL) {
465         virtqueue_unpop(s->svq, s->stats_vq_elem, 0);
466         g_free(s->stats_vq_elem);
467         s->stats_vq_elem = NULL;
468     }
469 }
470 
471 static void virtio_balloon_set_status(VirtIODevice *vdev, uint8_t status)
472 {
473     VirtIOBalloon *s = VIRTIO_BALLOON(vdev);
474 
475     if (!s->stats_vq_elem && vdev->vm_running &&
476         (status & VIRTIO_CONFIG_S_DRIVER_OK) && virtqueue_rewind(s->svq, 1)) {
477         /* poll stats queue for the element we have discarded when the VM
478          * was stopped */
479         virtio_balloon_receive_stats(vdev, s->svq);
480     }
481 }
482 
483 static void virtio_balloon_instance_init(Object *obj)
484 {
485     VirtIOBalloon *s = VIRTIO_BALLOON(obj);
486 
487     object_property_add(obj, "guest-stats", "guest statistics",
488                         balloon_stats_get_all, NULL, NULL, s, NULL);
489 
490     object_property_add(obj, "guest-stats-polling-interval", "int",
491                         balloon_stats_get_poll_interval,
492                         balloon_stats_set_poll_interval,
493                         NULL, s, NULL);
494 }
495 
496 static const VMStateDescription vmstate_virtio_balloon = {
497     .name = "virtio-balloon",
498     .minimum_version_id = 1,
499     .version_id = 1,
500     .fields = (VMStateField[]) {
501         VMSTATE_VIRTIO_DEVICE,
502         VMSTATE_END_OF_LIST()
503     },
504 };
505 
506 static Property virtio_balloon_properties[] = {
507     DEFINE_PROP_BIT("deflate-on-oom", VirtIOBalloon, host_features,
508                     VIRTIO_BALLOON_F_DEFLATE_ON_OOM, false),
509     DEFINE_PROP_END_OF_LIST(),
510 };
511 
512 static void virtio_balloon_class_init(ObjectClass *klass, void *data)
513 {
514     DeviceClass *dc = DEVICE_CLASS(klass);
515     VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
516 
517     dc->props = virtio_balloon_properties;
518     dc->vmsd = &vmstate_virtio_balloon;
519     set_bit(DEVICE_CATEGORY_MISC, dc->categories);
520     vdc->realize = virtio_balloon_device_realize;
521     vdc->unrealize = virtio_balloon_device_unrealize;
522     vdc->reset = virtio_balloon_device_reset;
523     vdc->get_config = virtio_balloon_get_config;
524     vdc->set_config = virtio_balloon_set_config;
525     vdc->get_features = virtio_balloon_get_features;
526     vdc->set_status = virtio_balloon_set_status;
527     vdc->vmsd = &vmstate_virtio_balloon_device;
528 }
529 
530 static const TypeInfo virtio_balloon_info = {
531     .name = TYPE_VIRTIO_BALLOON,
532     .parent = TYPE_VIRTIO_DEVICE,
533     .instance_size = sizeof(VirtIOBalloon),
534     .instance_init = virtio_balloon_instance_init,
535     .class_init = virtio_balloon_class_init,
536 };
537 
538 static void virtio_register_types(void)
539 {
540     type_register_static(&virtio_balloon_info);
541 }
542 
543 type_init(virtio_register_types)
544