xref: /openbmc/qemu/hw/virtio/virtio-balloon.c (revision 416296a9) 
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 defined(__linux__)
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 #endif
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_NR] = NULL
55 };
56 
57 /*
58  * reset_stats - Mark all items in the stats array as unset
59  *
60  * This function needs to be called at device initialization and before
61  * updating to a set of newly-generated stats.  This will ensure that no
62  * stale values stick around in case the guest reports a subset of the supported
63  * statistics.
64  */
65 static inline void reset_stats(VirtIOBalloon *dev)
66 {
67     int i;
68     for (i = 0; i < VIRTIO_BALLOON_S_NR; dev->stats[i++] = -1);
69 }
70 
71 static bool balloon_stats_supported(const VirtIOBalloon *s)
72 {
73     VirtIODevice *vdev = VIRTIO_DEVICE(s);
74     return virtio_vdev_has_feature(vdev, VIRTIO_BALLOON_F_STATS_VQ);
75 }
76 
77 static bool balloon_stats_enabled(const VirtIOBalloon *s)
78 {
79     return s->stats_poll_interval > 0;
80 }
81 
82 static void balloon_stats_destroy_timer(VirtIOBalloon *s)
83 {
84     if (balloon_stats_enabled(s)) {
85         timer_del(s->stats_timer);
86         timer_free(s->stats_timer);
87         s->stats_timer = NULL;
88         s->stats_poll_interval = 0;
89     }
90 }
91 
92 static void balloon_stats_change_timer(VirtIOBalloon *s, int64_t secs)
93 {
94     timer_mod(s->stats_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + secs * 1000);
95 }
96 
97 static void balloon_stats_poll_cb(void *opaque)
98 {
99     VirtIOBalloon *s = opaque;
100     VirtIODevice *vdev = VIRTIO_DEVICE(s);
101 
102     if (s->stats_vq_elem == NULL || !balloon_stats_supported(s)) {
103         /* re-schedule */
104         balloon_stats_change_timer(s, s->stats_poll_interval);
105         return;
106     }
107 
108     virtqueue_push(s->svq, s->stats_vq_elem, s->stats_vq_offset);
109     virtio_notify(vdev, s->svq);
110     g_free(s->stats_vq_elem);
111     s->stats_vq_elem = NULL;
112 }
113 
114 static void balloon_stats_get_all(Object *obj, Visitor *v, const char *name,
115                                   void *opaque, Error **errp)
116 {
117     Error *err = NULL;
118     VirtIOBalloon *s = opaque;
119     int i;
120 
121     visit_start_struct(v, name, NULL, 0, &err);
122     if (err) {
123         goto out;
124     }
125     visit_type_int(v, "last-update", &s->stats_last_update, &err);
126     if (err) {
127         goto out_end;
128     }
129 
130     visit_start_struct(v, "stats", NULL, 0, &err);
131     if (err) {
132         goto out_end;
133     }
134     for (i = 0; i < VIRTIO_BALLOON_S_NR; i++) {
135         visit_type_uint64(v, balloon_stat_names[i], &s->stats[i], &err);
136         if (err) {
137             goto out_nested;
138         }
139     }
140     visit_check_struct(v, &err);
141 out_nested:
142     visit_end_struct(v, NULL);
143 
144     if (!err) {
145         visit_check_struct(v, &err);
146     }
147 out_end:
148     visit_end_struct(v, NULL);
149 out:
150     error_propagate(errp, err);
151 }
152 
153 static void balloon_stats_get_poll_interval(Object *obj, Visitor *v,
154                                             const char *name, void *opaque,
155                                             Error **errp)
156 {
157     VirtIOBalloon *s = opaque;
158     visit_type_int(v, name, &s->stats_poll_interval, errp);
159 }
160 
161 static void balloon_stats_set_poll_interval(Object *obj, Visitor *v,
162                                             const char *name, void *opaque,
163                                             Error **errp)
164 {
165     VirtIOBalloon *s = opaque;
166     Error *local_err = NULL;
167     int64_t value;
168 
169     visit_type_int(v, name, &value, &local_err);
170     if (local_err) {
171         error_propagate(errp, local_err);
172         return;
173     }
174 
175     if (value < 0) {
176         error_setg(errp, "timer value must be greater than zero");
177         return;
178     }
179 
180     if (value > UINT32_MAX) {
181         error_setg(errp, "timer value is too big");
182         return;
183     }
184 
185     if (value == s->stats_poll_interval) {
186         return;
187     }
188 
189     if (value == 0) {
190         /* timer=0 disables the timer */
191         balloon_stats_destroy_timer(s);
192         return;
193     }
194 
195     if (balloon_stats_enabled(s)) {
196         /* timer interval change */
197         s->stats_poll_interval = value;
198         balloon_stats_change_timer(s, value);
199         return;
200     }
201 
202     /* create a new timer */
203     g_assert(s->stats_timer == NULL);
204     s->stats_timer = timer_new_ms(QEMU_CLOCK_VIRTUAL, balloon_stats_poll_cb, s);
205     s->stats_poll_interval = value;
206     balloon_stats_change_timer(s, 0);
207 }
208 
209 static void virtio_balloon_handle_output(VirtIODevice *vdev, VirtQueue *vq)
210 {
211     VirtIOBalloon *s = VIRTIO_BALLOON(vdev);
212     VirtQueueElement *elem;
213     MemoryRegionSection section;
214 
215     for (;;) {
216         size_t offset = 0;
217         uint32_t pfn;
218         elem = virtqueue_pop(vq, sizeof(VirtQueueElement));
219         if (!elem) {
220             return;
221         }
222 
223         while (iov_to_buf(elem->out_sg, elem->out_num, offset, &pfn, 4) == 4) {
224             ram_addr_t pa;
225             ram_addr_t addr;
226             int p = virtio_ldl_p(vdev, &pfn);
227 
228             pa = (ram_addr_t) p << VIRTIO_BALLOON_PFN_SHIFT;
229             offset += 4;
230 
231             /* FIXME: remove get_system_memory(), but how? */
232             section = memory_region_find(get_system_memory(), pa, 1);
233             if (!int128_nz(section.size) || !memory_region_is_ram(section.mr))
234                 continue;
235 
236             trace_virtio_balloon_handle_output(memory_region_name(section.mr),
237                                                pa);
238             /* Using memory_region_get_ram_ptr is bending the rules a bit, but
239                should be OK because we only want a single page.  */
240             addr = section.offset_within_region;
241             balloon_page(memory_region_get_ram_ptr(section.mr) + addr,
242                          !!(vq == s->dvq));
243             memory_region_unref(section.mr);
244         }
245 
246         virtqueue_push(vq, elem, offset);
247         virtio_notify(vdev, vq);
248         g_free(elem);
249     }
250 }
251 
252 static void virtio_balloon_receive_stats(VirtIODevice *vdev, VirtQueue *vq)
253 {
254     VirtIOBalloon *s = VIRTIO_BALLOON(vdev);
255     VirtQueueElement *elem;
256     VirtIOBalloonStat stat;
257     size_t offset = 0;
258     qemu_timeval tv;
259 
260     elem = virtqueue_pop(vq, sizeof(VirtQueueElement));
261     if (!elem) {
262         goto out;
263     }
264 
265     if (s->stats_vq_elem != NULL) {
266         /* This should never happen if the driver follows the spec. */
267         virtqueue_push(vq, s->stats_vq_elem, 0);
268         virtio_notify(vdev, vq);
269         g_free(s->stats_vq_elem);
270     }
271 
272     s->stats_vq_elem = elem;
273 
274     /* Initialize the stats to get rid of any stale values.  This is only
275      * needed to handle the case where a guest supports fewer stats than it
276      * used to (ie. it has booted into an old kernel).
277      */
278     reset_stats(s);
279 
280     while (iov_to_buf(elem->out_sg, elem->out_num, offset, &stat, sizeof(stat))
281            == sizeof(stat)) {
282         uint16_t tag = virtio_tswap16(vdev, stat.tag);
283         uint64_t val = virtio_tswap64(vdev, stat.val);
284 
285         offset += sizeof(stat);
286         if (tag < VIRTIO_BALLOON_S_NR)
287             s->stats[tag] = val;
288     }
289     s->stats_vq_offset = offset;
290 
291     if (qemu_gettimeofday(&tv) < 0) {
292         fprintf(stderr, "warning: %s: failed to get time of day\n", __func__);
293         goto out;
294     }
295 
296     s->stats_last_update = tv.tv_sec;
297 
298 out:
299     if (balloon_stats_enabled(s)) {
300         balloon_stats_change_timer(s, s->stats_poll_interval);
301     }
302 }
303 
304 static void virtio_balloon_get_config(VirtIODevice *vdev, uint8_t *config_data)
305 {
306     VirtIOBalloon *dev = VIRTIO_BALLOON(vdev);
307     struct virtio_balloon_config config;
308 
309     config.num_pages = cpu_to_le32(dev->num_pages);
310     config.actual = cpu_to_le32(dev->actual);
311 
312     trace_virtio_balloon_get_config(config.num_pages, config.actual);
313     memcpy(config_data, &config, sizeof(struct virtio_balloon_config));
314 }
315 
316 static int build_dimm_list(Object *obj, void *opaque)
317 {
318     GSList **list = opaque;
319 
320     if (object_dynamic_cast(obj, TYPE_PC_DIMM)) {
321         DeviceState *dev = DEVICE(obj);
322         if (dev->realized) { /* only realized DIMMs matter */
323             *list = g_slist_prepend(*list, dev);
324         }
325     }
326 
327     object_child_foreach(obj, build_dimm_list, opaque);
328     return 0;
329 }
330 
331 static ram_addr_t get_current_ram_size(void)
332 {
333     GSList *list = NULL, *item;
334     ram_addr_t size = ram_size;
335 
336     build_dimm_list(qdev_get_machine(), &list);
337     for (item = list; item; item = g_slist_next(item)) {
338         Object *obj = OBJECT(item->data);
339         if (!strcmp(object_get_typename(obj), TYPE_PC_DIMM)) {
340             size += object_property_get_int(obj, PC_DIMM_SIZE_PROP,
341                                             &error_abort);
342         }
343     }
344     g_slist_free(list);
345 
346     return size;
347 }
348 
349 static void virtio_balloon_set_config(VirtIODevice *vdev,
350                                       const uint8_t *config_data)
351 {
352     VirtIOBalloon *dev = VIRTIO_BALLOON(vdev);
353     struct virtio_balloon_config config;
354     uint32_t oldactual = dev->actual;
355     ram_addr_t vm_ram_size = get_current_ram_size();
356 
357     memcpy(&config, config_data, sizeof(struct virtio_balloon_config));
358     dev->actual = le32_to_cpu(config.actual);
359     if (dev->actual != oldactual) {
360         qapi_event_send_balloon_change(vm_ram_size -
361                         ((ram_addr_t) dev->actual << VIRTIO_BALLOON_PFN_SHIFT),
362                         &error_abort);
363     }
364     trace_virtio_balloon_set_config(dev->actual, oldactual);
365 }
366 
367 static uint64_t virtio_balloon_get_features(VirtIODevice *vdev, uint64_t f,
368                                             Error **errp)
369 {
370     VirtIOBalloon *dev = VIRTIO_BALLOON(vdev);
371     f |= dev->host_features;
372     virtio_add_feature(&f, VIRTIO_BALLOON_F_STATS_VQ);
373     return f;
374 }
375 
376 static void virtio_balloon_stat(void *opaque, BalloonInfo *info)
377 {
378     VirtIOBalloon *dev = opaque;
379     info->actual = get_current_ram_size() - ((uint64_t) dev->actual <<
380                                              VIRTIO_BALLOON_PFN_SHIFT);
381 }
382 
383 static void virtio_balloon_to_target(void *opaque, ram_addr_t target)
384 {
385     VirtIOBalloon *dev = VIRTIO_BALLOON(opaque);
386     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
387     ram_addr_t vm_ram_size = get_current_ram_size();
388 
389     if (target > vm_ram_size) {
390         target = vm_ram_size;
391     }
392     if (target) {
393         dev->num_pages = (vm_ram_size - target) >> VIRTIO_BALLOON_PFN_SHIFT;
394         virtio_notify_config(vdev);
395     }
396     trace_virtio_balloon_to_target(target, dev->num_pages);
397 }
398 
399 static void virtio_balloon_save_device(VirtIODevice *vdev, QEMUFile *f)
400 {
401     VirtIOBalloon *s = VIRTIO_BALLOON(vdev);
402 
403     qemu_put_be32(f, s->num_pages);
404     qemu_put_be32(f, s->actual);
405 }
406 
407 static int virtio_balloon_load(QEMUFile *f, void *opaque, size_t size)
408 {
409     return virtio_load(VIRTIO_DEVICE(opaque), f, 1);
410 }
411 
412 static int virtio_balloon_load_device(VirtIODevice *vdev, QEMUFile *f,
413                                       int version_id)
414 {
415     VirtIOBalloon *s = VIRTIO_BALLOON(vdev);
416 
417     s->num_pages = qemu_get_be32(f);
418     s->actual = qemu_get_be32(f);
419 
420     if (balloon_stats_enabled(s)) {
421         balloon_stats_change_timer(s, s->stats_poll_interval);
422     }
423     return 0;
424 }
425 
426 static void virtio_balloon_device_realize(DeviceState *dev, Error **errp)
427 {
428     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
429     VirtIOBalloon *s = VIRTIO_BALLOON(dev);
430     int ret;
431 
432     virtio_init(vdev, "virtio-balloon", VIRTIO_ID_BALLOON,
433                 sizeof(struct virtio_balloon_config));
434 
435     ret = qemu_add_balloon_handler(virtio_balloon_to_target,
436                                    virtio_balloon_stat, s);
437 
438     if (ret < 0) {
439         error_setg(errp, "Only one balloon device is supported");
440         virtio_cleanup(vdev);
441         return;
442     }
443 
444     s->ivq = virtio_add_queue(vdev, 128, virtio_balloon_handle_output);
445     s->dvq = virtio_add_queue(vdev, 128, virtio_balloon_handle_output);
446     s->svq = virtio_add_queue(vdev, 128, virtio_balloon_receive_stats);
447 
448     reset_stats(s);
449 }
450 
451 static void virtio_balloon_device_unrealize(DeviceState *dev, Error **errp)
452 {
453     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
454     VirtIOBalloon *s = VIRTIO_BALLOON(dev);
455 
456     balloon_stats_destroy_timer(s);
457     qemu_remove_balloon_handler(s);
458     virtio_cleanup(vdev);
459 }
460 
461 static void virtio_balloon_device_reset(VirtIODevice *vdev)
462 {
463     VirtIOBalloon *s = VIRTIO_BALLOON(vdev);
464 
465     if (s->stats_vq_elem != NULL) {
466         virtqueue_discard(s->svq, s->stats_vq_elem, 0);
467         g_free(s->stats_vq_elem);
468         s->stats_vq_elem = NULL;
469     }
470 }
471 
472 static void virtio_balloon_set_status(VirtIODevice *vdev, uint8_t status)
473 {
474     VirtIOBalloon *s = VIRTIO_BALLOON(vdev);
475 
476     if (!s->stats_vq_elem && vdev->vm_running &&
477         (status & VIRTIO_CONFIG_S_DRIVER_OK) && virtqueue_rewind(s->svq, 1)) {
478         /* poll stats queue for the element we have discarded when the VM
479          * was stopped */
480         virtio_balloon_receive_stats(vdev, s->svq);
481     }
482 }
483 
484 static void virtio_balloon_instance_init(Object *obj)
485 {
486     VirtIOBalloon *s = VIRTIO_BALLOON(obj);
487 
488     object_property_add(obj, "guest-stats", "guest statistics",
489                         balloon_stats_get_all, NULL, NULL, s, NULL);
490 
491     object_property_add(obj, "guest-stats-polling-interval", "int",
492                         balloon_stats_get_poll_interval,
493                         balloon_stats_set_poll_interval,
494                         NULL, s, NULL);
495 }
496 
497 VMSTATE_VIRTIO_DEVICE(balloon, 1, virtio_balloon_load, virtio_vmstate_save);
498 
499 static Property virtio_balloon_properties[] = {
500     DEFINE_PROP_BIT("deflate-on-oom", VirtIOBalloon, host_features,
501                     VIRTIO_BALLOON_F_DEFLATE_ON_OOM, false),
502     DEFINE_PROP_END_OF_LIST(),
503 };
504 
505 static void virtio_balloon_class_init(ObjectClass *klass, void *data)
506 {
507     DeviceClass *dc = DEVICE_CLASS(klass);
508     VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
509 
510     dc->props = virtio_balloon_properties;
511     dc->vmsd = &vmstate_virtio_balloon;
512     set_bit(DEVICE_CATEGORY_MISC, dc->categories);
513     vdc->realize = virtio_balloon_device_realize;
514     vdc->unrealize = virtio_balloon_device_unrealize;
515     vdc->reset = virtio_balloon_device_reset;
516     vdc->get_config = virtio_balloon_get_config;
517     vdc->set_config = virtio_balloon_set_config;
518     vdc->get_features = virtio_balloon_get_features;
519     vdc->save = virtio_balloon_save_device;
520     vdc->load = virtio_balloon_load_device;
521     vdc->set_status = virtio_balloon_set_status;
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