xref: /openbmc/qemu/hw/virtio/virtio-balloon.c (revision b917da4c) 
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 "cpu.h"
23 #include "sysemu/balloon.h"
24 #include "hw/virtio/virtio-balloon.h"
25 #include "sysemu/kvm.h"
26 #include "exec/address-spaces.h"
27 #include "qapi/visitor.h"
28 #include "qapi-event.h"
29 #include "trace.h"
30 
31 #if defined(__linux__)
32 #include <sys/mman.h>
33 #endif
34 
35 #include "hw/virtio/virtio-bus.h"
36 #include "hw/virtio/virtio-access.h"
37 
38 static void balloon_page(void *addr, int deflate)
39 {
40 #if defined(__linux__)
41     if (!qemu_balloon_is_inhibited() && (!kvm_enabled() ||
42                                          kvm_has_sync_mmu())) {
43         qemu_madvise(addr, TARGET_PAGE_SIZE,
44                 deflate ? QEMU_MADV_WILLNEED : QEMU_MADV_DONTNEED);
45     }
46 #endif
47 }
48 
49 static const char *balloon_stat_names[] = {
50    [VIRTIO_BALLOON_S_SWAP_IN] = "stat-swap-in",
51    [VIRTIO_BALLOON_S_SWAP_OUT] = "stat-swap-out",
52    [VIRTIO_BALLOON_S_MAJFLT] = "stat-major-faults",
53    [VIRTIO_BALLOON_S_MINFLT] = "stat-minor-faults",
54    [VIRTIO_BALLOON_S_MEMFREE] = "stat-free-memory",
55    [VIRTIO_BALLOON_S_MEMTOT] = "stat-total-memory",
56    [VIRTIO_BALLOON_S_AVAIL] = "stat-available-memory",
57    [VIRTIO_BALLOON_S_NR] = NULL
58 };
59 
60 /*
61  * reset_stats - Mark all items in the stats array as unset
62  *
63  * This function needs to be called at device initialization and before
64  * updating to a set of newly-generated stats.  This will ensure that no
65  * stale values stick around in case the guest reports a subset of the supported
66  * statistics.
67  */
68 static inline void reset_stats(VirtIOBalloon *dev)
69 {
70     int i;
71     for (i = 0; i < VIRTIO_BALLOON_S_NR; dev->stats[i++] = -1);
72 }
73 
74 static bool balloon_stats_supported(const VirtIOBalloon *s)
75 {
76     VirtIODevice *vdev = VIRTIO_DEVICE(s);
77     return virtio_vdev_has_feature(vdev, VIRTIO_BALLOON_F_STATS_VQ);
78 }
79 
80 static bool balloon_stats_enabled(const VirtIOBalloon *s)
81 {
82     return s->stats_poll_interval > 0;
83 }
84 
85 static void balloon_stats_destroy_timer(VirtIOBalloon *s)
86 {
87     if (balloon_stats_enabled(s)) {
88         timer_del(s->stats_timer);
89         timer_free(s->stats_timer);
90         s->stats_timer = NULL;
91         s->stats_poll_interval = 0;
92     }
93 }
94 
95 static void balloon_stats_change_timer(VirtIOBalloon *s, int64_t secs)
96 {
97     timer_mod(s->stats_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + secs * 1000);
98 }
99 
100 static void balloon_stats_poll_cb(void *opaque)
101 {
102     VirtIOBalloon *s = opaque;
103     VirtIODevice *vdev = VIRTIO_DEVICE(s);
104 
105     if (s->stats_vq_elem == NULL || !balloon_stats_supported(s)) {
106         /* re-schedule */
107         balloon_stats_change_timer(s, s->stats_poll_interval);
108         return;
109     }
110 
111     virtqueue_push(s->svq, s->stats_vq_elem, s->stats_vq_offset);
112     virtio_notify(vdev, s->svq);
113     g_free(s->stats_vq_elem);
114     s->stats_vq_elem = NULL;
115 }
116 
117 static void balloon_stats_get_all(Object *obj, Visitor *v, const char *name,
118                                   void *opaque, Error **errp)
119 {
120     Error *err = NULL;
121     VirtIOBalloon *s = opaque;
122     int i;
123 
124     visit_start_struct(v, name, NULL, 0, &err);
125     if (err) {
126         goto out;
127     }
128     visit_type_int(v, "last-update", &s->stats_last_update, &err);
129     if (err) {
130         goto out_end;
131     }
132 
133     visit_start_struct(v, "stats", NULL, 0, &err);
134     if (err) {
135         goto out_end;
136     }
137     for (i = 0; i < VIRTIO_BALLOON_S_NR; i++) {
138         visit_type_uint64(v, balloon_stat_names[i], &s->stats[i], &err);
139         if (err) {
140             break;
141         }
142     }
143     error_propagate(errp, err);
144     err = NULL;
145     visit_end_struct(v, &err);
146 
147 out_end:
148     error_propagate(errp, err);
149     err = NULL;
150     visit_end_struct(v, &err);
151 out:
152     error_propagate(errp, err);
153 }
154 
155 static void balloon_stats_get_poll_interval(Object *obj, Visitor *v,
156                                             const char *name, void *opaque,
157                                             Error **errp)
158 {
159     VirtIOBalloon *s = opaque;
160     visit_type_int(v, name, &s->stats_poll_interval, errp);
161 }
162 
163 static void balloon_stats_set_poll_interval(Object *obj, Visitor *v,
164                                             const char *name, void *opaque,
165                                             Error **errp)
166 {
167     VirtIOBalloon *s = opaque;
168     Error *local_err = NULL;
169     int64_t value;
170 
171     visit_type_int(v, name, &value, &local_err);
172     if (local_err) {
173         error_propagate(errp, local_err);
174         return;
175     }
176 
177     if (value < 0) {
178         error_setg(errp, "timer value must be greater than zero");
179         return;
180     }
181 
182     if (value > UINT32_MAX) {
183         error_setg(errp, "timer value is too big");
184         return;
185     }
186 
187     if (value == s->stats_poll_interval) {
188         return;
189     }
190 
191     if (value == 0) {
192         /* timer=0 disables the timer */
193         balloon_stats_destroy_timer(s);
194         return;
195     }
196 
197     if (balloon_stats_enabled(s)) {
198         /* timer interval change */
199         s->stats_poll_interval = value;
200         balloon_stats_change_timer(s, value);
201         return;
202     }
203 
204     /* create a new timer */
205     g_assert(s->stats_timer == NULL);
206     s->stats_timer = timer_new_ms(QEMU_CLOCK_VIRTUAL, balloon_stats_poll_cb, s);
207     s->stats_poll_interval = value;
208     balloon_stats_change_timer(s, 0);
209 }
210 
211 static void virtio_balloon_handle_output(VirtIODevice *vdev, VirtQueue *vq)
212 {
213     VirtIOBalloon *s = VIRTIO_BALLOON(vdev);
214     VirtQueueElement *elem;
215     MemoryRegionSection section;
216 
217     for (;;) {
218         size_t offset = 0;
219         uint32_t pfn;
220         elem = virtqueue_pop(vq, sizeof(VirtQueueElement));
221         if (!elem) {
222             return;
223         }
224 
225         while (iov_to_buf(elem->out_sg, elem->out_num, offset, &pfn, 4) == 4) {
226             ram_addr_t pa;
227             ram_addr_t addr;
228             int p = virtio_ldl_p(vdev, &pfn);
229 
230             pa = (ram_addr_t) p << VIRTIO_BALLOON_PFN_SHIFT;
231             offset += 4;
232 
233             /* FIXME: remove get_system_memory(), but how? */
234             section = memory_region_find(get_system_memory(), pa, 1);
235             if (!int128_nz(section.size) || !memory_region_is_ram(section.mr))
236                 continue;
237 
238             trace_virtio_balloon_handle_output(memory_region_name(section.mr),
239                                                pa);
240             /* Using memory_region_get_ram_ptr is bending the rules a bit, but
241                should be OK because we only want a single page.  */
242             addr = section.offset_within_region;
243             balloon_page(memory_region_get_ram_ptr(section.mr) + addr,
244                          !!(vq == s->dvq));
245             memory_region_unref(section.mr);
246         }
247 
248         virtqueue_push(vq, elem, offset);
249         virtio_notify(vdev, vq);
250         g_free(elem);
251     }
252 }
253 
254 static void virtio_balloon_receive_stats(VirtIODevice *vdev, VirtQueue *vq)
255 {
256     VirtIOBalloon *s = VIRTIO_BALLOON(vdev);
257     VirtQueueElement *elem;
258     VirtIOBalloonStat stat;
259     size_t offset = 0;
260     qemu_timeval tv;
261 
262     elem = virtqueue_pop(vq, sizeof(VirtQueueElement));
263     if (!elem) {
264         goto out;
265     }
266 
267     if (s->stats_vq_elem != NULL) {
268         /* This should never happen if the driver follows the spec. */
269         virtqueue_push(vq, s->stats_vq_elem, 0);
270         virtio_notify(vdev, vq);
271         g_free(s->stats_vq_elem);
272     }
273 
274     s->stats_vq_elem = elem;
275 
276     /* Initialize the stats to get rid of any stale values.  This is only
277      * needed to handle the case where a guest supports fewer stats than it
278      * used to (ie. it has booted into an old kernel).
279      */
280     reset_stats(s);
281 
282     while (iov_to_buf(elem->out_sg, elem->out_num, offset, &stat, sizeof(stat))
283            == sizeof(stat)) {
284         uint16_t tag = virtio_tswap16(vdev, stat.tag);
285         uint64_t val = virtio_tswap64(vdev, stat.val);
286 
287         offset += sizeof(stat);
288         if (tag < VIRTIO_BALLOON_S_NR)
289             s->stats[tag] = val;
290     }
291     s->stats_vq_offset = offset;
292 
293     if (qemu_gettimeofday(&tv) < 0) {
294         fprintf(stderr, "warning: %s: failed to get time of day\n", __func__);
295         goto out;
296     }
297 
298     s->stats_last_update = tv.tv_sec;
299 
300 out:
301     if (balloon_stats_enabled(s)) {
302         balloon_stats_change_timer(s, s->stats_poll_interval);
303     }
304 }
305 
306 static void virtio_balloon_get_config(VirtIODevice *vdev, uint8_t *config_data)
307 {
308     VirtIOBalloon *dev = VIRTIO_BALLOON(vdev);
309     struct virtio_balloon_config config;
310 
311     config.num_pages = cpu_to_le32(dev->num_pages);
312     config.actual = cpu_to_le32(dev->actual);
313 
314     trace_virtio_balloon_get_config(config.num_pages, config.actual);
315     memcpy(config_data, &config, sizeof(struct virtio_balloon_config));
316 }
317 
318 static int build_dimm_list(Object *obj, void *opaque)
319 {
320     GSList **list = opaque;
321 
322     if (object_dynamic_cast(obj, TYPE_PC_DIMM)) {
323         DeviceState *dev = DEVICE(obj);
324         if (dev->realized) { /* only realized DIMMs matter */
325             *list = g_slist_prepend(*list, dev);
326         }
327     }
328 
329     object_child_foreach(obj, build_dimm_list, opaque);
330     return 0;
331 }
332 
333 static ram_addr_t get_current_ram_size(void)
334 {
335     GSList *list = NULL, *item;
336     ram_addr_t size = ram_size;
337 
338     build_dimm_list(qdev_get_machine(), &list);
339     for (item = list; item; item = g_slist_next(item)) {
340         Object *obj = OBJECT(item->data);
341         if (!strcmp(object_get_typename(obj), TYPE_PC_DIMM)) {
342             size += object_property_get_int(obj, PC_DIMM_SIZE_PROP,
343                                             &error_abort);
344         }
345     }
346     g_slist_free(list);
347 
348     return size;
349 }
350 
351 static void virtio_balloon_set_config(VirtIODevice *vdev,
352                                       const uint8_t *config_data)
353 {
354     VirtIOBalloon *dev = VIRTIO_BALLOON(vdev);
355     struct virtio_balloon_config config;
356     uint32_t oldactual = dev->actual;
357     ram_addr_t vm_ram_size = get_current_ram_size();
358 
359     memcpy(&config, config_data, sizeof(struct virtio_balloon_config));
360     dev->actual = le32_to_cpu(config.actual);
361     if (dev->actual != oldactual) {
362         qapi_event_send_balloon_change(vm_ram_size -
363                         ((ram_addr_t) dev->actual << VIRTIO_BALLOON_PFN_SHIFT),
364                         &error_abort);
365     }
366     trace_virtio_balloon_set_config(dev->actual, oldactual);
367 }
368 
369 static uint64_t virtio_balloon_get_features(VirtIODevice *vdev, uint64_t f,
370                                             Error **errp)
371 {
372     VirtIOBalloon *dev = VIRTIO_BALLOON(vdev);
373     f |= dev->host_features;
374     virtio_add_feature(&f, VIRTIO_BALLOON_F_STATS_VQ);
375     return f;
376 }
377 
378 static void virtio_balloon_stat(void *opaque, BalloonInfo *info)
379 {
380     VirtIOBalloon *dev = opaque;
381     info->actual = get_current_ram_size() - ((uint64_t) dev->actual <<
382                                              VIRTIO_BALLOON_PFN_SHIFT);
383 }
384 
385 static void virtio_balloon_to_target(void *opaque, ram_addr_t target)
386 {
387     VirtIOBalloon *dev = VIRTIO_BALLOON(opaque);
388     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
389     ram_addr_t vm_ram_size = get_current_ram_size();
390 
391     if (target > vm_ram_size) {
392         target = vm_ram_size;
393     }
394     if (target) {
395         dev->num_pages = (vm_ram_size - target) >> VIRTIO_BALLOON_PFN_SHIFT;
396         virtio_notify_config(vdev);
397     }
398     trace_virtio_balloon_to_target(target, dev->num_pages);
399 }
400 
401 static void virtio_balloon_save(QEMUFile *f, void *opaque)
402 {
403     virtio_save(VIRTIO_DEVICE(opaque), f);
404 }
405 
406 static void virtio_balloon_save_device(VirtIODevice *vdev, QEMUFile *f)
407 {
408     VirtIOBalloon *s = VIRTIO_BALLOON(vdev);
409 
410     qemu_put_be32(f, s->num_pages);
411     qemu_put_be32(f, s->actual);
412 }
413 
414 static int virtio_balloon_load(QEMUFile *f, void *opaque, int version_id)
415 {
416     if (version_id != 1)
417         return -EINVAL;
418 
419     return virtio_load(VIRTIO_DEVICE(opaque), f, version_id);
420 }
421 
422 static int virtio_balloon_load_device(VirtIODevice *vdev, QEMUFile *f,
423                                       int version_id)
424 {
425     VirtIOBalloon *s = VIRTIO_BALLOON(vdev);
426 
427     s->num_pages = qemu_get_be32(f);
428     s->actual = qemu_get_be32(f);
429     return 0;
430 }
431 
432 static void virtio_balloon_device_realize(DeviceState *dev, Error **errp)
433 {
434     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
435     VirtIOBalloon *s = VIRTIO_BALLOON(dev);
436     int ret;
437 
438     virtio_init(vdev, "virtio-balloon", VIRTIO_ID_BALLOON,
439                 sizeof(struct virtio_balloon_config));
440 
441     ret = qemu_add_balloon_handler(virtio_balloon_to_target,
442                                    virtio_balloon_stat, s);
443 
444     if (ret < 0) {
445         error_setg(errp, "Only one balloon device is supported");
446         virtio_cleanup(vdev);
447         return;
448     }
449 
450     s->ivq = virtio_add_queue(vdev, 128, virtio_balloon_handle_output);
451     s->dvq = virtio_add_queue(vdev, 128, virtio_balloon_handle_output);
452     s->svq = virtio_add_queue(vdev, 128, virtio_balloon_receive_stats);
453 
454     reset_stats(s);
455 
456     register_savevm(dev, "virtio-balloon", -1, 1,
457                     virtio_balloon_save, virtio_balloon_load, s);
458 }
459 
460 static void virtio_balloon_device_unrealize(DeviceState *dev, Error **errp)
461 {
462     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
463     VirtIOBalloon *s = VIRTIO_BALLOON(dev);
464 
465     balloon_stats_destroy_timer(s);
466     qemu_remove_balloon_handler(s);
467     unregister_savevm(dev, "virtio-balloon", s);
468     virtio_cleanup(vdev);
469 }
470 
471 static void virtio_balloon_device_reset(VirtIODevice *vdev)
472 {
473     VirtIOBalloon *s = VIRTIO_BALLOON(vdev);
474 
475     if (s->stats_vq_elem != NULL) {
476         g_free(s->stats_vq_elem);
477         s->stats_vq_elem = NULL;
478     }
479 }
480 
481 static void virtio_balloon_instance_init(Object *obj)
482 {
483     VirtIOBalloon *s = VIRTIO_BALLOON(obj);
484 
485     object_property_add(obj, "guest-stats", "guest statistics",
486                         balloon_stats_get_all, NULL, NULL, s, NULL);
487 
488     object_property_add(obj, "guest-stats-polling-interval", "int",
489                         balloon_stats_get_poll_interval,
490                         balloon_stats_set_poll_interval,
491                         NULL, s, NULL);
492 }
493 
494 static Property virtio_balloon_properties[] = {
495     DEFINE_PROP_BIT("deflate-on-oom", VirtIOBalloon, host_features,
496                     VIRTIO_BALLOON_F_DEFLATE_ON_OOM, false),
497     DEFINE_PROP_END_OF_LIST(),
498 };
499 
500 static void virtio_balloon_class_init(ObjectClass *klass, void *data)
501 {
502     DeviceClass *dc = DEVICE_CLASS(klass);
503     VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
504 
505     dc->props = virtio_balloon_properties;
506     set_bit(DEVICE_CATEGORY_MISC, dc->categories);
507     vdc->realize = virtio_balloon_device_realize;
508     vdc->unrealize = virtio_balloon_device_unrealize;
509     vdc->reset = virtio_balloon_device_reset;
510     vdc->get_config = virtio_balloon_get_config;
511     vdc->set_config = virtio_balloon_set_config;
512     vdc->get_features = virtio_balloon_get_features;
513     vdc->save = virtio_balloon_save_device;
514     vdc->load = virtio_balloon_load_device;
515 }
516 
517 static const TypeInfo virtio_balloon_info = {
518     .name = TYPE_VIRTIO_BALLOON,
519     .parent = TYPE_VIRTIO_DEVICE,
520     .instance_size = sizeof(VirtIOBalloon),
521     .instance_init = virtio_balloon_instance_init,
522     .class_init = virtio_balloon_class_init,
523 };
524 
525 static void virtio_register_types(void)
526 {
527     type_register_static(&virtio_balloon_info);
528 }
529 
530 type_init(virtio_register_types)
531