xref: /openbmc/qemu/hw/virtio/virtio-balloon.c (revision d9d3aaea) 
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 #if defined(__linux__)
31 #include <sys/mman.h>
32 #endif
33 
34 #include "hw/virtio/virtio-bus.h"
35 #include "hw/virtio/virtio-access.h"
36 
37 #define BALLOON_PAGE_SIZE  (1 << VIRTIO_BALLOON_PFN_SHIFT)
38 
39 static void balloon_page(void *addr, int deflate)
40 {
41 #if defined(__linux__)
42     if (!qemu_balloon_is_inhibited() && (!kvm_enabled() ||
43                                          kvm_has_sync_mmu())) {
44         qemu_madvise(addr, BALLOON_PAGE_SIZE,
45                 deflate ? QEMU_MADV_WILLNEED : QEMU_MADV_DONTNEED);
46     }
47 #endif
48 }
49 
50 static const char *balloon_stat_names[] = {
51    [VIRTIO_BALLOON_S_SWAP_IN] = "stat-swap-in",
52    [VIRTIO_BALLOON_S_SWAP_OUT] = "stat-swap-out",
53    [VIRTIO_BALLOON_S_MAJFLT] = "stat-major-faults",
54    [VIRTIO_BALLOON_S_MINFLT] = "stat-minor-faults",
55    [VIRTIO_BALLOON_S_MEMFREE] = "stat-free-memory",
56    [VIRTIO_BALLOON_S_MEMTOT] = "stat-total-memory",
57    [VIRTIO_BALLOON_S_AVAIL] = "stat-available-memory",
58    [VIRTIO_BALLOON_S_NR] = NULL
59 };
60 
61 /*
62  * reset_stats - Mark all items in the stats array as unset
63  *
64  * This function needs to be called at device initialization and before
65  * updating to a set of newly-generated stats.  This will ensure that no
66  * stale values stick around in case the guest reports a subset of the supported
67  * statistics.
68  */
69 static inline void reset_stats(VirtIOBalloon *dev)
70 {
71     int i;
72     for (i = 0; i < VIRTIO_BALLOON_S_NR; dev->stats[i++] = -1);
73 }
74 
75 static bool balloon_stats_supported(const VirtIOBalloon *s)
76 {
77     VirtIODevice *vdev = VIRTIO_DEVICE(s);
78     return virtio_vdev_has_feature(vdev, VIRTIO_BALLOON_F_STATS_VQ);
79 }
80 
81 static bool balloon_stats_enabled(const VirtIOBalloon *s)
82 {
83     return s->stats_poll_interval > 0;
84 }
85 
86 static void balloon_stats_destroy_timer(VirtIOBalloon *s)
87 {
88     if (balloon_stats_enabled(s)) {
89         timer_del(s->stats_timer);
90         timer_free(s->stats_timer);
91         s->stats_timer = NULL;
92         s->stats_poll_interval = 0;
93     }
94 }
95 
96 static void balloon_stats_change_timer(VirtIOBalloon *s, int64_t secs)
97 {
98     timer_mod(s->stats_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + secs * 1000);
99 }
100 
101 static void balloon_stats_poll_cb(void *opaque)
102 {
103     VirtIOBalloon *s = opaque;
104     VirtIODevice *vdev = VIRTIO_DEVICE(s);
105 
106     if (s->stats_vq_elem == NULL || !balloon_stats_supported(s)) {
107         /* re-schedule */
108         balloon_stats_change_timer(s, s->stats_poll_interval);
109         return;
110     }
111 
112     virtqueue_push(s->svq, s->stats_vq_elem, s->stats_vq_offset);
113     virtio_notify(vdev, s->svq);
114     g_free(s->stats_vq_elem);
115     s->stats_vq_elem = NULL;
116 }
117 
118 static void balloon_stats_get_all(Object *obj, Visitor *v, const char *name,
119                                   void *opaque, Error **errp)
120 {
121     Error *err = NULL;
122     VirtIOBalloon *s = opaque;
123     int i;
124 
125     visit_start_struct(v, name, NULL, 0, &err);
126     if (err) {
127         goto out;
128     }
129     visit_type_int(v, "last-update", &s->stats_last_update, &err);
130     if (err) {
131         goto out_end;
132     }
133 
134     visit_start_struct(v, "stats", NULL, 0, &err);
135     if (err) {
136         goto out_end;
137     }
138     for (i = 0; i < VIRTIO_BALLOON_S_NR; i++) {
139         visit_type_uint64(v, balloon_stat_names[i], &s->stats[i], &err);
140         if (err) {
141             break;
142         }
143     }
144     error_propagate(errp, err);
145     err = NULL;
146     visit_end_struct(v, &err);
147 
148 out_end:
149     error_propagate(errp, err);
150     err = NULL;
151     visit_end_struct(v, &err);
152 out:
153     error_propagate(errp, err);
154 }
155 
156 static void balloon_stats_get_poll_interval(Object *obj, Visitor *v,
157                                             const char *name, void *opaque,
158                                             Error **errp)
159 {
160     VirtIOBalloon *s = opaque;
161     visit_type_int(v, name, &s->stats_poll_interval, errp);
162 }
163 
164 static void balloon_stats_set_poll_interval(Object *obj, Visitor *v,
165                                             const char *name, void *opaque,
166                                             Error **errp)
167 {
168     VirtIOBalloon *s = opaque;
169     Error *local_err = NULL;
170     int64_t value;
171 
172     visit_type_int(v, name, &value, &local_err);
173     if (local_err) {
174         error_propagate(errp, local_err);
175         return;
176     }
177 
178     if (value < 0) {
179         error_setg(errp, "timer value must be greater than zero");
180         return;
181     }
182 
183     if (value > UINT32_MAX) {
184         error_setg(errp, "timer value is too big");
185         return;
186     }
187 
188     if (value == s->stats_poll_interval) {
189         return;
190     }
191 
192     if (value == 0) {
193         /* timer=0 disables the timer */
194         balloon_stats_destroy_timer(s);
195         return;
196     }
197 
198     if (balloon_stats_enabled(s)) {
199         /* timer interval change */
200         s->stats_poll_interval = value;
201         balloon_stats_change_timer(s, value);
202         return;
203     }
204 
205     /* create a new timer */
206     g_assert(s->stats_timer == NULL);
207     s->stats_timer = timer_new_ms(QEMU_CLOCK_VIRTUAL, balloon_stats_poll_cb, s);
208     s->stats_poll_interval = value;
209     balloon_stats_change_timer(s, 0);
210 }
211 
212 static void virtio_balloon_handle_output(VirtIODevice *vdev, VirtQueue *vq)
213 {
214     VirtIOBalloon *s = VIRTIO_BALLOON(vdev);
215     VirtQueueElement *elem;
216     MemoryRegionSection section;
217 
218     for (;;) {
219         size_t offset = 0;
220         uint32_t pfn;
221         elem = virtqueue_pop(vq, sizeof(VirtQueueElement));
222         if (!elem) {
223             return;
224         }
225 
226         while (iov_to_buf(elem->out_sg, elem->out_num, offset, &pfn, 4) == 4) {
227             ram_addr_t pa;
228             ram_addr_t addr;
229             int p = virtio_ldl_p(vdev, &pfn);
230 
231             pa = (ram_addr_t) p << VIRTIO_BALLOON_PFN_SHIFT;
232             offset += 4;
233 
234             /* FIXME: remove get_system_memory(), but how? */
235             section = memory_region_find(get_system_memory(), pa, 1);
236             if (!int128_nz(section.size) || !memory_region_is_ram(section.mr))
237                 continue;
238 
239             trace_virtio_balloon_handle_output(memory_region_name(section.mr),
240                                                pa);
241             /* Using memory_region_get_ram_ptr is bending the rules a bit, but
242                should be OK because we only want a single page.  */
243             addr = section.offset_within_region;
244             balloon_page(memory_region_get_ram_ptr(section.mr) + addr,
245                          !!(vq == s->dvq));
246             memory_region_unref(section.mr);
247         }
248 
249         virtqueue_push(vq, elem, offset);
250         virtio_notify(vdev, vq);
251         g_free(elem);
252     }
253 }
254 
255 static void virtio_balloon_receive_stats(VirtIODevice *vdev, VirtQueue *vq)
256 {
257     VirtIOBalloon *s = VIRTIO_BALLOON(vdev);
258     VirtQueueElement *elem;
259     VirtIOBalloonStat stat;
260     size_t offset = 0;
261     qemu_timeval tv;
262 
263     elem = virtqueue_pop(vq, sizeof(VirtQueueElement));
264     if (!elem) {
265         goto out;
266     }
267 
268     if (s->stats_vq_elem != NULL) {
269         /* This should never happen if the driver follows the spec. */
270         virtqueue_push(vq, s->stats_vq_elem, 0);
271         virtio_notify(vdev, vq);
272         g_free(s->stats_vq_elem);
273     }
274 
275     s->stats_vq_elem = elem;
276 
277     /* Initialize the stats to get rid of any stale values.  This is only
278      * needed to handle the case where a guest supports fewer stats than it
279      * used to (ie. it has booted into an old kernel).
280      */
281     reset_stats(s);
282 
283     while (iov_to_buf(elem->out_sg, elem->out_num, offset, &stat, sizeof(stat))
284            == sizeof(stat)) {
285         uint16_t tag = virtio_tswap16(vdev, stat.tag);
286         uint64_t val = virtio_tswap64(vdev, stat.val);
287 
288         offset += sizeof(stat);
289         if (tag < VIRTIO_BALLOON_S_NR)
290             s->stats[tag] = val;
291     }
292     s->stats_vq_offset = offset;
293 
294     if (qemu_gettimeofday(&tv) < 0) {
295         fprintf(stderr, "warning: %s: failed to get time of day\n", __func__);
296         goto out;
297     }
298 
299     s->stats_last_update = tv.tv_sec;
300 
301 out:
302     if (balloon_stats_enabled(s)) {
303         balloon_stats_change_timer(s, s->stats_poll_interval);
304     }
305 }
306 
307 static void virtio_balloon_get_config(VirtIODevice *vdev, uint8_t *config_data)
308 {
309     VirtIOBalloon *dev = VIRTIO_BALLOON(vdev);
310     struct virtio_balloon_config config;
311 
312     config.num_pages = cpu_to_le32(dev->num_pages);
313     config.actual = cpu_to_le32(dev->actual);
314 
315     trace_virtio_balloon_get_config(config.num_pages, config.actual);
316     memcpy(config_data, &config, sizeof(struct virtio_balloon_config));
317 }
318 
319 static int build_dimm_list(Object *obj, void *opaque)
320 {
321     GSList **list = opaque;
322 
323     if (object_dynamic_cast(obj, TYPE_PC_DIMM)) {
324         DeviceState *dev = DEVICE(obj);
325         if (dev->realized) { /* only realized DIMMs matter */
326             *list = g_slist_prepend(*list, dev);
327         }
328     }
329 
330     object_child_foreach(obj, build_dimm_list, opaque);
331     return 0;
332 }
333 
334 static ram_addr_t get_current_ram_size(void)
335 {
336     GSList *list = NULL, *item;
337     ram_addr_t size = ram_size;
338 
339     build_dimm_list(qdev_get_machine(), &list);
340     for (item = list; item; item = g_slist_next(item)) {
341         Object *obj = OBJECT(item->data);
342         if (!strcmp(object_get_typename(obj), TYPE_PC_DIMM)) {
343             size += object_property_get_int(obj, PC_DIMM_SIZE_PROP,
344                                             &error_abort);
345         }
346     }
347     g_slist_free(list);
348 
349     return size;
350 }
351 
352 static void virtio_balloon_set_config(VirtIODevice *vdev,
353                                       const uint8_t *config_data)
354 {
355     VirtIOBalloon *dev = VIRTIO_BALLOON(vdev);
356     struct virtio_balloon_config config;
357     uint32_t oldactual = dev->actual;
358     ram_addr_t vm_ram_size = get_current_ram_size();
359 
360     memcpy(&config, config_data, sizeof(struct virtio_balloon_config));
361     dev->actual = le32_to_cpu(config.actual);
362     if (dev->actual != oldactual) {
363         qapi_event_send_balloon_change(vm_ram_size -
364                         ((ram_addr_t) dev->actual << VIRTIO_BALLOON_PFN_SHIFT),
365                         &error_abort);
366     }
367     trace_virtio_balloon_set_config(dev->actual, oldactual);
368 }
369 
370 static uint64_t virtio_balloon_get_features(VirtIODevice *vdev, uint64_t f,
371                                             Error **errp)
372 {
373     VirtIOBalloon *dev = VIRTIO_BALLOON(vdev);
374     f |= dev->host_features;
375     virtio_add_feature(&f, VIRTIO_BALLOON_F_STATS_VQ);
376     return f;
377 }
378 
379 static void virtio_balloon_stat(void *opaque, BalloonInfo *info)
380 {
381     VirtIOBalloon *dev = opaque;
382     info->actual = get_current_ram_size() - ((uint64_t) dev->actual <<
383                                              VIRTIO_BALLOON_PFN_SHIFT);
384 }
385 
386 static void virtio_balloon_to_target(void *opaque, ram_addr_t target)
387 {
388     VirtIOBalloon *dev = VIRTIO_BALLOON(opaque);
389     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
390     ram_addr_t vm_ram_size = get_current_ram_size();
391 
392     if (target > vm_ram_size) {
393         target = vm_ram_size;
394     }
395     if (target) {
396         dev->num_pages = (vm_ram_size - target) >> VIRTIO_BALLOON_PFN_SHIFT;
397         virtio_notify_config(vdev);
398     }
399     trace_virtio_balloon_to_target(target, dev->num_pages);
400 }
401 
402 static void virtio_balloon_save(QEMUFile *f, void *opaque)
403 {
404     virtio_save(VIRTIO_DEVICE(opaque), f);
405 }
406 
407 static void virtio_balloon_save_device(VirtIODevice *vdev, QEMUFile *f)
408 {
409     VirtIOBalloon *s = VIRTIO_BALLOON(vdev);
410 
411     qemu_put_be32(f, s->num_pages);
412     qemu_put_be32(f, s->actual);
413 }
414 
415 static int virtio_balloon_load(QEMUFile *f, void *opaque, int version_id)
416 {
417     if (version_id != 1)
418         return -EINVAL;
419 
420     return virtio_load(VIRTIO_DEVICE(opaque), f, version_id);
421 }
422 
423 static int virtio_balloon_load_device(VirtIODevice *vdev, QEMUFile *f,
424                                       int version_id)
425 {
426     VirtIOBalloon *s = VIRTIO_BALLOON(vdev);
427 
428     s->num_pages = qemu_get_be32(f);
429     s->actual = qemu_get_be32(f);
430 
431     if (balloon_stats_enabled(s)) {
432         balloon_stats_change_timer(s, s->stats_poll_interval);
433     }
434     return 0;
435 }
436 
437 static void virtio_balloon_device_realize(DeviceState *dev, Error **errp)
438 {
439     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
440     VirtIOBalloon *s = VIRTIO_BALLOON(dev);
441     int ret;
442 
443     virtio_init(vdev, "virtio-balloon", VIRTIO_ID_BALLOON,
444                 sizeof(struct virtio_balloon_config));
445 
446     ret = qemu_add_balloon_handler(virtio_balloon_to_target,
447                                    virtio_balloon_stat, s);
448 
449     if (ret < 0) {
450         error_setg(errp, "Only one balloon device is supported");
451         virtio_cleanup(vdev);
452         return;
453     }
454 
455     s->ivq = virtio_add_queue(vdev, 128, virtio_balloon_handle_output);
456     s->dvq = virtio_add_queue(vdev, 128, virtio_balloon_handle_output);
457     s->svq = virtio_add_queue(vdev, 128, virtio_balloon_receive_stats);
458 
459     reset_stats(s);
460 
461     register_savevm(dev, "virtio-balloon", -1, 1,
462                     virtio_balloon_save, virtio_balloon_load, s);
463 }
464 
465 static void virtio_balloon_device_unrealize(DeviceState *dev, Error **errp)
466 {
467     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
468     VirtIOBalloon *s = VIRTIO_BALLOON(dev);
469 
470     balloon_stats_destroy_timer(s);
471     qemu_remove_balloon_handler(s);
472     unregister_savevm(dev, "virtio-balloon", s);
473     virtio_cleanup(vdev);
474 }
475 
476 static void virtio_balloon_device_reset(VirtIODevice *vdev)
477 {
478     VirtIOBalloon *s = VIRTIO_BALLOON(vdev);
479 
480     if (s->stats_vq_elem != NULL) {
481         g_free(s->stats_vq_elem);
482         s->stats_vq_elem = NULL;
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 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     set_bit(DEVICE_CATEGORY_MISC, dc->categories);
512     vdc->realize = virtio_balloon_device_realize;
513     vdc->unrealize = virtio_balloon_device_unrealize;
514     vdc->reset = virtio_balloon_device_reset;
515     vdc->get_config = virtio_balloon_get_config;
516     vdc->set_config = virtio_balloon_set_config;
517     vdc->get_features = virtio_balloon_get_features;
518     vdc->save = virtio_balloon_save_device;
519     vdc->load = virtio_balloon_load_device;
520 }
521 
522 static const TypeInfo virtio_balloon_info = {
523     .name = TYPE_VIRTIO_BALLOON,
524     .parent = TYPE_VIRTIO_DEVICE,
525     .instance_size = sizeof(VirtIOBalloon),
526     .instance_init = virtio_balloon_instance_init,
527     .class_init = virtio_balloon_class_init,
528 };
529 
530 static void virtio_register_types(void)
531 {
532     type_register_static(&virtio_balloon_info);
533 }
534 
535 type_init(virtio_register_types)
536