xref: /openbmc/qemu/hw/virtio/virtio-balloon.c (revision 10df8ff1)
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 "exec/address-spaces.h"
25 #include "qapi/error.h"
26 #include "qapi/qapi-events-misc.h"
27 #include "qapi/visitor.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()) {
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_CACHES] = "stat-disk-caches",
53    [VIRTIO_BALLOON_S_HTLB_PGALLOC] = "stat-htlb-pgalloc",
54    [VIRTIO_BALLOON_S_HTLB_PGFAIL] = "stat-htlb-pgfail",
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     }
370     trace_virtio_balloon_set_config(dev->actual, oldactual);
371 }
372 
373 static uint64_t virtio_balloon_get_features(VirtIODevice *vdev, uint64_t f,
374                                             Error **errp)
375 {
376     VirtIOBalloon *dev = VIRTIO_BALLOON(vdev);
377     f |= dev->host_features;
378     virtio_add_feature(&f, VIRTIO_BALLOON_F_STATS_VQ);
379     return f;
380 }
381 
382 static void virtio_balloon_stat(void *opaque, BalloonInfo *info)
383 {
384     VirtIOBalloon *dev = opaque;
385     info->actual = get_current_ram_size() - ((uint64_t) dev->actual <<
386                                              VIRTIO_BALLOON_PFN_SHIFT);
387 }
388 
389 static void virtio_balloon_to_target(void *opaque, ram_addr_t target)
390 {
391     VirtIOBalloon *dev = VIRTIO_BALLOON(opaque);
392     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
393     ram_addr_t vm_ram_size = get_current_ram_size();
394 
395     if (target > vm_ram_size) {
396         target = vm_ram_size;
397     }
398     if (target) {
399         dev->num_pages = (vm_ram_size - target) >> VIRTIO_BALLOON_PFN_SHIFT;
400         virtio_notify_config(vdev);
401     }
402     trace_virtio_balloon_to_target(target, dev->num_pages);
403 }
404 
405 static int virtio_balloon_post_load_device(void *opaque, int version_id)
406 {
407     VirtIOBalloon *s = VIRTIO_BALLOON(opaque);
408 
409     if (balloon_stats_enabled(s)) {
410         balloon_stats_change_timer(s, s->stats_poll_interval);
411     }
412     return 0;
413 }
414 
415 static const VMStateDescription vmstate_virtio_balloon_device = {
416     .name = "virtio-balloon-device",
417     .version_id = 1,
418     .minimum_version_id = 1,
419     .post_load = virtio_balloon_post_load_device,
420     .fields = (VMStateField[]) {
421         VMSTATE_UINT32(num_pages, VirtIOBalloon),
422         VMSTATE_UINT32(actual, VirtIOBalloon),
423         VMSTATE_END_OF_LIST()
424     },
425 };
426 
427 static void virtio_balloon_device_realize(DeviceState *dev, Error **errp)
428 {
429     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
430     VirtIOBalloon *s = VIRTIO_BALLOON(dev);
431     int ret;
432 
433     virtio_init(vdev, "virtio-balloon", VIRTIO_ID_BALLOON,
434                 sizeof(struct virtio_balloon_config));
435 
436     ret = qemu_add_balloon_handler(virtio_balloon_to_target,
437                                    virtio_balloon_stat, s);
438 
439     if (ret < 0) {
440         error_setg(errp, "Only one balloon device is supported");
441         virtio_cleanup(vdev);
442         return;
443     }
444 
445     s->ivq = virtio_add_queue(vdev, 128, virtio_balloon_handle_output);
446     s->dvq = virtio_add_queue(vdev, 128, virtio_balloon_handle_output);
447     s->svq = virtio_add_queue(vdev, 128, virtio_balloon_receive_stats);
448 
449     reset_stats(s);
450 }
451 
452 static void virtio_balloon_device_unrealize(DeviceState *dev, Error **errp)
453 {
454     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
455     VirtIOBalloon *s = VIRTIO_BALLOON(dev);
456 
457     balloon_stats_destroy_timer(s);
458     qemu_remove_balloon_handler(s);
459     virtio_cleanup(vdev);
460 }
461 
462 static void virtio_balloon_device_reset(VirtIODevice *vdev)
463 {
464     VirtIOBalloon *s = VIRTIO_BALLOON(vdev);
465 
466     if (s->stats_vq_elem != NULL) {
467         virtqueue_unpop(s->svq, s->stats_vq_elem, 0);
468         g_free(s->stats_vq_elem);
469         s->stats_vq_elem = NULL;
470     }
471 }
472 
473 static void virtio_balloon_set_status(VirtIODevice *vdev, uint8_t status)
474 {
475     VirtIOBalloon *s = VIRTIO_BALLOON(vdev);
476 
477     if (!s->stats_vq_elem && vdev->vm_running &&
478         (status & VIRTIO_CONFIG_S_DRIVER_OK) && virtqueue_rewind(s->svq, 1)) {
479         /* poll stats queue for the element we have discarded when the VM
480          * was stopped */
481         virtio_balloon_receive_stats(vdev, s->svq);
482     }
483 }
484 
485 static void virtio_balloon_instance_init(Object *obj)
486 {
487     VirtIOBalloon *s = VIRTIO_BALLOON(obj);
488 
489     object_property_add(obj, "guest-stats", "guest statistics",
490                         balloon_stats_get_all, NULL, NULL, s, NULL);
491 
492     object_property_add(obj, "guest-stats-polling-interval", "int",
493                         balloon_stats_get_poll_interval,
494                         balloon_stats_set_poll_interval,
495                         NULL, s, NULL);
496 }
497 
498 static const VMStateDescription vmstate_virtio_balloon = {
499     .name = "virtio-balloon",
500     .minimum_version_id = 1,
501     .version_id = 1,
502     .fields = (VMStateField[]) {
503         VMSTATE_VIRTIO_DEVICE,
504         VMSTATE_END_OF_LIST()
505     },
506 };
507 
508 static Property virtio_balloon_properties[] = {
509     DEFINE_PROP_BIT("deflate-on-oom", VirtIOBalloon, host_features,
510                     VIRTIO_BALLOON_F_DEFLATE_ON_OOM, false),
511     DEFINE_PROP_END_OF_LIST(),
512 };
513 
514 static void virtio_balloon_class_init(ObjectClass *klass, void *data)
515 {
516     DeviceClass *dc = DEVICE_CLASS(klass);
517     VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
518 
519     dc->props = virtio_balloon_properties;
520     dc->vmsd = &vmstate_virtio_balloon;
521     set_bit(DEVICE_CATEGORY_MISC, dc->categories);
522     vdc->realize = virtio_balloon_device_realize;
523     vdc->unrealize = virtio_balloon_device_unrealize;
524     vdc->reset = virtio_balloon_device_reset;
525     vdc->get_config = virtio_balloon_get_config;
526     vdc->set_config = virtio_balloon_set_config;
527     vdc->get_features = virtio_balloon_get_features;
528     vdc->set_status = virtio_balloon_set_status;
529     vdc->vmsd = &vmstate_virtio_balloon_device;
530 }
531 
532 static const TypeInfo virtio_balloon_info = {
533     .name = TYPE_VIRTIO_BALLOON,
534     .parent = TYPE_VIRTIO_DEVICE,
535     .instance_size = sizeof(VirtIOBalloon),
536     .instance_init = virtio_balloon_instance_init,
537     .class_init = virtio_balloon_class_init,
538 };
539 
540 static void virtio_register_types(void)
541 {
542     type_register_static(&virtio_balloon_info);
543 }
544 
545 type_init(virtio_register_types)
546