audio: deprecate -soundhw gus
[qemu/ar7.git] / hw / virtio / virtio-balloon.c
blob10507b2a430ab40f8b8ecd33387ac1c0fd6834fb
1 /*
2 * Virtio Balloon Device
4 * Copyright IBM, Corp. 2008
5 * Copyright (C) 2011 Red Hat, Inc.
6 * Copyright (C) 2011 Amit Shah <amit.shah@redhat.com>
8 * Authors:
9 * Anthony Liguori <aliguori@us.ibm.com>
11 * This work is licensed under the terms of the GNU GPL, version 2. See
12 * the COPYING file in the top-level directory.
16 #include "qemu/osdep.h"
17 #include "qemu/iov.h"
18 #include "qemu/module.h"
19 #include "qemu/timer.h"
20 #include "hw/virtio/virtio.h"
21 #include "hw/mem/pc-dimm.h"
22 #include "hw/qdev-properties.h"
23 #include "sysemu/balloon.h"
24 #include "hw/virtio/virtio-balloon.h"
25 #include "exec/address-spaces.h"
26 #include "qapi/error.h"
27 #include "qapi/qapi-events-misc.h"
28 #include "qapi/visitor.h"
29 #include "trace.h"
30 #include "qemu/error-report.h"
31 #include "migration/misc.h"
33 #include "hw/virtio/virtio-bus.h"
34 #include "hw/virtio/virtio-access.h"
36 #define BALLOON_PAGE_SIZE (1 << VIRTIO_BALLOON_PFN_SHIFT)
38 typedef struct PartiallyBalloonedPage {
39 ram_addr_t base_gpa;
40 unsigned long *bitmap;
41 } PartiallyBalloonedPage;
43 static void virtio_balloon_pbp_free(PartiallyBalloonedPage *pbp)
45 if (!pbp->bitmap) {
46 return;
48 g_free(pbp->bitmap);
49 pbp->bitmap = NULL;
52 static void virtio_balloon_pbp_alloc(PartiallyBalloonedPage *pbp,
53 ram_addr_t base_gpa,
54 long subpages)
56 pbp->base_gpa = base_gpa;
57 pbp->bitmap = bitmap_new(subpages);
60 static bool virtio_balloon_pbp_matches(PartiallyBalloonedPage *pbp,
61 ram_addr_t base_gpa)
63 return pbp->base_gpa == base_gpa;
66 static void balloon_inflate_page(VirtIOBalloon *balloon,
67 MemoryRegion *mr, hwaddr mr_offset,
68 PartiallyBalloonedPage *pbp)
70 void *addr = memory_region_get_ram_ptr(mr) + mr_offset;
71 ram_addr_t rb_offset, rb_aligned_offset, base_gpa;
72 RAMBlock *rb;
73 size_t rb_page_size;
74 int subpages;
76 /* XXX is there a better way to get to the RAMBlock than via a
77 * host address? */
78 rb = qemu_ram_block_from_host(addr, false, &rb_offset);
79 rb_page_size = qemu_ram_pagesize(rb);
81 if (rb_page_size == BALLOON_PAGE_SIZE) {
82 /* Easy case */
84 ram_block_discard_range(rb, rb_offset, rb_page_size);
85 /* We ignore errors from ram_block_discard_range(), because it
86 * has already reported them, and failing to discard a balloon
87 * page is not fatal */
88 return;
91 /* Hard case
93 * We've put a piece of a larger host page into the balloon - we
94 * need to keep track until we have a whole host page to
95 * discard
97 warn_report_once(
98 "Balloon used with backing page size > 4kiB, this may not be reliable");
100 rb_aligned_offset = QEMU_ALIGN_DOWN(rb_offset, rb_page_size);
101 subpages = rb_page_size / BALLOON_PAGE_SIZE;
102 base_gpa = memory_region_get_ram_addr(mr) + mr_offset -
103 (rb_offset - rb_aligned_offset);
105 if (pbp->bitmap && !virtio_balloon_pbp_matches(pbp, base_gpa)) {
106 /* We've partially ballooned part of a host page, but now
107 * we're trying to balloon part of a different one. Too hard,
108 * give up on the old partial page */
109 virtio_balloon_pbp_free(pbp);
112 if (!pbp->bitmap) {
113 virtio_balloon_pbp_alloc(pbp, base_gpa, subpages);
116 set_bit((rb_offset - rb_aligned_offset) / BALLOON_PAGE_SIZE,
117 pbp->bitmap);
119 if (bitmap_full(pbp->bitmap, subpages)) {
120 /* We've accumulated a full host page, we can actually discard
121 * it now */
123 ram_block_discard_range(rb, rb_aligned_offset, rb_page_size);
124 /* We ignore errors from ram_block_discard_range(), because it
125 * has already reported them, and failing to discard a balloon
126 * page is not fatal */
127 virtio_balloon_pbp_free(pbp);
131 static void balloon_deflate_page(VirtIOBalloon *balloon,
132 MemoryRegion *mr, hwaddr mr_offset)
134 void *addr = memory_region_get_ram_ptr(mr) + mr_offset;
135 ram_addr_t rb_offset;
136 RAMBlock *rb;
137 size_t rb_page_size;
138 void *host_addr;
139 int ret;
141 /* XXX is there a better way to get to the RAMBlock than via a
142 * host address? */
143 rb = qemu_ram_block_from_host(addr, false, &rb_offset);
144 rb_page_size = qemu_ram_pagesize(rb);
146 host_addr = (void *)((uintptr_t)addr & ~(rb_page_size - 1));
148 /* When a page is deflated, we hint the whole host page it lives
149 * on, since we can't do anything smaller */
150 ret = qemu_madvise(host_addr, rb_page_size, QEMU_MADV_WILLNEED);
151 if (ret != 0) {
152 warn_report("Couldn't MADV_WILLNEED on balloon deflate: %s",
153 strerror(errno));
154 /* Otherwise ignore, failing to page hint shouldn't be fatal */
158 static const char *balloon_stat_names[] = {
159 [VIRTIO_BALLOON_S_SWAP_IN] = "stat-swap-in",
160 [VIRTIO_BALLOON_S_SWAP_OUT] = "stat-swap-out",
161 [VIRTIO_BALLOON_S_MAJFLT] = "stat-major-faults",
162 [VIRTIO_BALLOON_S_MINFLT] = "stat-minor-faults",
163 [VIRTIO_BALLOON_S_MEMFREE] = "stat-free-memory",
164 [VIRTIO_BALLOON_S_MEMTOT] = "stat-total-memory",
165 [VIRTIO_BALLOON_S_AVAIL] = "stat-available-memory",
166 [VIRTIO_BALLOON_S_CACHES] = "stat-disk-caches",
167 [VIRTIO_BALLOON_S_HTLB_PGALLOC] = "stat-htlb-pgalloc",
168 [VIRTIO_BALLOON_S_HTLB_PGFAIL] = "stat-htlb-pgfail",
169 [VIRTIO_BALLOON_S_NR] = NULL
173 * reset_stats - Mark all items in the stats array as unset
175 * This function needs to be called at device initialization and before
176 * updating to a set of newly-generated stats. This will ensure that no
177 * stale values stick around in case the guest reports a subset of the supported
178 * statistics.
180 static inline void reset_stats(VirtIOBalloon *dev)
182 int i;
183 for (i = 0; i < VIRTIO_BALLOON_S_NR; dev->stats[i++] = -1);
186 static bool balloon_stats_supported(const VirtIOBalloon *s)
188 VirtIODevice *vdev = VIRTIO_DEVICE(s);
189 return virtio_vdev_has_feature(vdev, VIRTIO_BALLOON_F_STATS_VQ);
192 static bool balloon_stats_enabled(const VirtIOBalloon *s)
194 return s->stats_poll_interval > 0;
197 static void balloon_stats_destroy_timer(VirtIOBalloon *s)
199 if (balloon_stats_enabled(s)) {
200 timer_del(s->stats_timer);
201 timer_free(s->stats_timer);
202 s->stats_timer = NULL;
203 s->stats_poll_interval = 0;
207 static void balloon_stats_change_timer(VirtIOBalloon *s, int64_t secs)
209 timer_mod(s->stats_timer, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL) + secs * 1000);
212 static void balloon_stats_poll_cb(void *opaque)
214 VirtIOBalloon *s = opaque;
215 VirtIODevice *vdev = VIRTIO_DEVICE(s);
217 if (s->stats_vq_elem == NULL || !balloon_stats_supported(s)) {
218 /* re-schedule */
219 balloon_stats_change_timer(s, s->stats_poll_interval);
220 return;
223 virtqueue_push(s->svq, s->stats_vq_elem, s->stats_vq_offset);
224 virtio_notify(vdev, s->svq);
225 g_free(s->stats_vq_elem);
226 s->stats_vq_elem = NULL;
229 static void balloon_stats_get_all(Object *obj, Visitor *v, const char *name,
230 void *opaque, Error **errp)
232 Error *err = NULL;
233 VirtIOBalloon *s = opaque;
234 int i;
236 visit_start_struct(v, name, NULL, 0, &err);
237 if (err) {
238 goto out;
240 visit_type_int(v, "last-update", &s->stats_last_update, &err);
241 if (err) {
242 goto out_end;
245 visit_start_struct(v, "stats", NULL, 0, &err);
246 if (err) {
247 goto out_end;
249 for (i = 0; i < VIRTIO_BALLOON_S_NR; i++) {
250 visit_type_uint64(v, balloon_stat_names[i], &s->stats[i], &err);
251 if (err) {
252 goto out_nested;
255 visit_check_struct(v, &err);
256 out_nested:
257 visit_end_struct(v, NULL);
259 if (!err) {
260 visit_check_struct(v, &err);
262 out_end:
263 visit_end_struct(v, NULL);
264 out:
265 error_propagate(errp, err);
268 static void balloon_stats_get_poll_interval(Object *obj, Visitor *v,
269 const char *name, void *opaque,
270 Error **errp)
272 VirtIOBalloon *s = opaque;
273 visit_type_int(v, name, &s->stats_poll_interval, errp);
276 static void balloon_stats_set_poll_interval(Object *obj, Visitor *v,
277 const char *name, void *opaque,
278 Error **errp)
280 VirtIOBalloon *s = opaque;
281 Error *local_err = NULL;
282 int64_t value;
284 visit_type_int(v, name, &value, &local_err);
285 if (local_err) {
286 error_propagate(errp, local_err);
287 return;
290 if (value < 0) {
291 error_setg(errp, "timer value must be greater than zero");
292 return;
295 if (value > UINT32_MAX) {
296 error_setg(errp, "timer value is too big");
297 return;
300 if (value == s->stats_poll_interval) {
301 return;
304 if (value == 0) {
305 /* timer=0 disables the timer */
306 balloon_stats_destroy_timer(s);
307 return;
310 if (balloon_stats_enabled(s)) {
311 /* timer interval change */
312 s->stats_poll_interval = value;
313 balloon_stats_change_timer(s, value);
314 return;
317 /* create a new timer */
318 g_assert(s->stats_timer == NULL);
319 s->stats_timer = timer_new_ms(QEMU_CLOCK_VIRTUAL, balloon_stats_poll_cb, s);
320 s->stats_poll_interval = value;
321 balloon_stats_change_timer(s, 0);
324 static void virtio_balloon_handle_report(VirtIODevice *vdev, VirtQueue *vq)
326 VirtIOBalloon *dev = VIRTIO_BALLOON(vdev);
327 VirtQueueElement *elem;
329 while ((elem = virtqueue_pop(vq, sizeof(VirtQueueElement)))) {
330 unsigned int i;
333 * When we discard the page it has the effect of removing the page
334 * from the hypervisor itself and causing it to be zeroed when it
335 * is returned to us. So we must not discard the page if it is
336 * accessible by another device or process, or if the guest is
337 * expecting it to retain a non-zero value.
339 if (qemu_balloon_is_inhibited() || dev->poison_val) {
340 goto skip_element;
343 for (i = 0; i < elem->in_num; i++) {
344 void *addr = elem->in_sg[i].iov_base;
345 size_t size = elem->in_sg[i].iov_len;
346 ram_addr_t ram_offset;
347 RAMBlock *rb;
350 * There is no need to check the memory section to see if
351 * it is ram/readonly/romd like there is for handle_output
352 * below. If the region is not meant to be written to then
353 * address_space_map will have allocated a bounce buffer
354 * and it will be freed in address_space_unmap and trigger
355 * and unassigned_mem_write before failing to copy over the
356 * buffer. If more than one bad descriptor is provided it
357 * will return NULL after the first bounce buffer and fail
358 * to map any resources.
360 rb = qemu_ram_block_from_host(addr, false, &ram_offset);
361 if (!rb) {
362 trace_virtio_balloon_bad_addr(elem->in_addr[i]);
363 continue;
367 * For now we will simply ignore unaligned memory regions, or
368 * regions that overrun the end of the RAMBlock.
370 if (!QEMU_IS_ALIGNED(ram_offset | size, qemu_ram_pagesize(rb)) ||
371 (ram_offset + size) > qemu_ram_get_used_length(rb)) {
372 continue;
375 ram_block_discard_range(rb, ram_offset, size);
378 skip_element:
379 virtqueue_push(vq, elem, 0);
380 virtio_notify(vdev, vq);
381 g_free(elem);
385 static void virtio_balloon_handle_output(VirtIODevice *vdev, VirtQueue *vq)
387 VirtIOBalloon *s = VIRTIO_BALLOON(vdev);
388 VirtQueueElement *elem;
389 MemoryRegionSection section;
391 for (;;) {
392 PartiallyBalloonedPage pbp = {};
393 size_t offset = 0;
394 uint32_t pfn;
396 elem = virtqueue_pop(vq, sizeof(VirtQueueElement));
397 if (!elem) {
398 break;
401 while (iov_to_buf(elem->out_sg, elem->out_num, offset, &pfn, 4) == 4) {
402 unsigned int p = virtio_ldl_p(vdev, &pfn);
403 hwaddr pa;
405 pa = (hwaddr) p << VIRTIO_BALLOON_PFN_SHIFT;
406 offset += 4;
408 section = memory_region_find(get_system_memory(), pa,
409 BALLOON_PAGE_SIZE);
410 if (!section.mr) {
411 trace_virtio_balloon_bad_addr(pa);
412 continue;
414 if (!memory_region_is_ram(section.mr) ||
415 memory_region_is_rom(section.mr) ||
416 memory_region_is_romd(section.mr)) {
417 trace_virtio_balloon_bad_addr(pa);
418 memory_region_unref(section.mr);
419 continue;
422 trace_virtio_balloon_handle_output(memory_region_name(section.mr),
423 pa);
424 if (!qemu_balloon_is_inhibited()) {
425 if (vq == s->ivq) {
426 balloon_inflate_page(s, section.mr,
427 section.offset_within_region, &pbp);
428 } else if (vq == s->dvq) {
429 balloon_deflate_page(s, section.mr, section.offset_within_region);
430 } else {
431 g_assert_not_reached();
434 memory_region_unref(section.mr);
437 virtqueue_push(vq, elem, offset);
438 virtio_notify(vdev, vq);
439 g_free(elem);
440 virtio_balloon_pbp_free(&pbp);
444 static void virtio_balloon_receive_stats(VirtIODevice *vdev, VirtQueue *vq)
446 VirtIOBalloon *s = VIRTIO_BALLOON(vdev);
447 VirtQueueElement *elem;
448 VirtIOBalloonStat stat;
449 size_t offset = 0;
450 qemu_timeval tv;
452 elem = virtqueue_pop(vq, sizeof(VirtQueueElement));
453 if (!elem) {
454 goto out;
457 if (s->stats_vq_elem != NULL) {
458 /* This should never happen if the driver follows the spec. */
459 virtqueue_push(vq, s->stats_vq_elem, 0);
460 virtio_notify(vdev, vq);
461 g_free(s->stats_vq_elem);
464 s->stats_vq_elem = elem;
466 /* Initialize the stats to get rid of any stale values. This is only
467 * needed to handle the case where a guest supports fewer stats than it
468 * used to (ie. it has booted into an old kernel).
470 reset_stats(s);
472 while (iov_to_buf(elem->out_sg, elem->out_num, offset, &stat, sizeof(stat))
473 == sizeof(stat)) {
474 uint16_t tag = virtio_tswap16(vdev, stat.tag);
475 uint64_t val = virtio_tswap64(vdev, stat.val);
477 offset += sizeof(stat);
478 if (tag < VIRTIO_BALLOON_S_NR)
479 s->stats[tag] = val;
481 s->stats_vq_offset = offset;
483 if (qemu_gettimeofday(&tv) < 0) {
484 warn_report("%s: failed to get time of day", __func__);
485 goto out;
488 s->stats_last_update = tv.tv_sec;
490 out:
491 if (balloon_stats_enabled(s)) {
492 balloon_stats_change_timer(s, s->stats_poll_interval);
496 static void virtio_balloon_handle_free_page_vq(VirtIODevice *vdev,
497 VirtQueue *vq)
499 VirtIOBalloon *s = VIRTIO_BALLOON(vdev);
500 qemu_bh_schedule(s->free_page_bh);
503 static bool get_free_page_hints(VirtIOBalloon *dev)
505 VirtQueueElement *elem;
506 VirtIODevice *vdev = VIRTIO_DEVICE(dev);
507 VirtQueue *vq = dev->free_page_vq;
508 bool ret = true;
510 while (dev->block_iothread) {
511 qemu_cond_wait(&dev->free_page_cond, &dev->free_page_lock);
514 elem = virtqueue_pop(vq, sizeof(VirtQueueElement));
515 if (!elem) {
516 return false;
519 if (elem->out_num) {
520 uint32_t id;
521 size_t size = iov_to_buf(elem->out_sg, elem->out_num, 0,
522 &id, sizeof(id));
524 virtio_tswap32s(vdev, &id);
525 if (unlikely(size != sizeof(id))) {
526 virtio_error(vdev, "received an incorrect cmd id");
527 ret = false;
528 goto out;
530 if (id == dev->free_page_report_cmd_id) {
531 dev->free_page_report_status = FREE_PAGE_REPORT_S_START;
532 } else {
534 * Stop the optimization only when it has started. This
535 * avoids a stale stop sign for the previous command.
537 if (dev->free_page_report_status == FREE_PAGE_REPORT_S_START) {
538 dev->free_page_report_status = FREE_PAGE_REPORT_S_STOP;
543 if (elem->in_num) {
544 if (dev->free_page_report_status == FREE_PAGE_REPORT_S_START) {
545 qemu_guest_free_page_hint(elem->in_sg[0].iov_base,
546 elem->in_sg[0].iov_len);
550 out:
551 virtqueue_push(vq, elem, 1);
552 g_free(elem);
553 return ret;
556 static void virtio_ballloon_get_free_page_hints(void *opaque)
558 VirtIOBalloon *dev = opaque;
559 VirtIODevice *vdev = VIRTIO_DEVICE(dev);
560 VirtQueue *vq = dev->free_page_vq;
561 bool continue_to_get_hints;
563 do {
564 qemu_mutex_lock(&dev->free_page_lock);
565 virtio_queue_set_notification(vq, 0);
566 continue_to_get_hints = get_free_page_hints(dev);
567 qemu_mutex_unlock(&dev->free_page_lock);
568 virtio_notify(vdev, vq);
570 * Start to poll the vq once the reporting started. Otherwise, continue
571 * only when there are entries on the vq, which need to be given back.
573 } while (continue_to_get_hints ||
574 dev->free_page_report_status == FREE_PAGE_REPORT_S_START);
575 virtio_queue_set_notification(vq, 1);
578 static bool virtio_balloon_free_page_support(void *opaque)
580 VirtIOBalloon *s = opaque;
581 VirtIODevice *vdev = VIRTIO_DEVICE(s);
583 return virtio_vdev_has_feature(vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT);
586 static void virtio_balloon_free_page_start(VirtIOBalloon *s)
588 VirtIODevice *vdev = VIRTIO_DEVICE(s);
590 /* For the stop and copy phase, we don't need to start the optimization */
591 if (!vdev->vm_running) {
592 return;
595 if (s->free_page_report_cmd_id == UINT_MAX) {
596 s->free_page_report_cmd_id =
597 VIRTIO_BALLOON_FREE_PAGE_REPORT_CMD_ID_MIN;
598 } else {
599 s->free_page_report_cmd_id++;
602 s->free_page_report_status = FREE_PAGE_REPORT_S_REQUESTED;
603 virtio_notify_config(vdev);
606 static void virtio_balloon_free_page_stop(VirtIOBalloon *s)
608 VirtIODevice *vdev = VIRTIO_DEVICE(s);
610 if (s->free_page_report_status != FREE_PAGE_REPORT_S_STOP) {
612 * The lock also guarantees us that the
613 * virtio_ballloon_get_free_page_hints exits after the
614 * free_page_report_status is set to S_STOP.
616 qemu_mutex_lock(&s->free_page_lock);
618 * The guest hasn't done the reporting, so host sends a notification
619 * to the guest to actively stop the reporting.
621 s->free_page_report_status = FREE_PAGE_REPORT_S_STOP;
622 qemu_mutex_unlock(&s->free_page_lock);
623 virtio_notify_config(vdev);
627 static void virtio_balloon_free_page_done(VirtIOBalloon *s)
629 VirtIODevice *vdev = VIRTIO_DEVICE(s);
631 s->free_page_report_status = FREE_PAGE_REPORT_S_DONE;
632 virtio_notify_config(vdev);
635 static int
636 virtio_balloon_free_page_report_notify(NotifierWithReturn *n, void *data)
638 VirtIOBalloon *dev = container_of(n, VirtIOBalloon,
639 free_page_report_notify);
640 VirtIODevice *vdev = VIRTIO_DEVICE(dev);
641 PrecopyNotifyData *pnd = data;
643 if (!virtio_balloon_free_page_support(dev)) {
645 * This is an optimization provided to migration, so just return 0 to
646 * have the normal migration process not affected when this feature is
647 * not supported.
649 return 0;
652 switch (pnd->reason) {
653 case PRECOPY_NOTIFY_SETUP:
654 precopy_enable_free_page_optimization();
655 break;
656 case PRECOPY_NOTIFY_COMPLETE:
657 case PRECOPY_NOTIFY_CLEANUP:
658 case PRECOPY_NOTIFY_BEFORE_BITMAP_SYNC:
659 virtio_balloon_free_page_stop(dev);
660 break;
661 case PRECOPY_NOTIFY_AFTER_BITMAP_SYNC:
662 if (vdev->vm_running) {
663 virtio_balloon_free_page_start(dev);
664 } else {
665 virtio_balloon_free_page_done(dev);
667 break;
668 default:
669 virtio_error(vdev, "%s: %d reason unknown", __func__, pnd->reason);
672 return 0;
675 static size_t virtio_balloon_config_size(VirtIOBalloon *s)
677 uint64_t features = s->host_features;
679 if (s->qemu_4_0_config_size) {
680 return sizeof(struct virtio_balloon_config);
682 if (virtio_has_feature(features, VIRTIO_BALLOON_F_PAGE_POISON)) {
683 return sizeof(struct virtio_balloon_config);
685 if (virtio_has_feature(features, VIRTIO_BALLOON_F_FREE_PAGE_HINT)) {
686 return offsetof(struct virtio_balloon_config, poison_val);
688 return offsetof(struct virtio_balloon_config, free_page_report_cmd_id);
691 static void virtio_balloon_get_config(VirtIODevice *vdev, uint8_t *config_data)
693 VirtIOBalloon *dev = VIRTIO_BALLOON(vdev);
694 struct virtio_balloon_config config = {};
696 config.num_pages = cpu_to_le32(dev->num_pages);
697 config.actual = cpu_to_le32(dev->actual);
698 config.poison_val = cpu_to_le32(dev->poison_val);
700 if (dev->free_page_report_status == FREE_PAGE_REPORT_S_REQUESTED) {
701 config.free_page_report_cmd_id =
702 cpu_to_le32(dev->free_page_report_cmd_id);
703 } else if (dev->free_page_report_status == FREE_PAGE_REPORT_S_STOP) {
704 config.free_page_report_cmd_id =
705 cpu_to_le32(VIRTIO_BALLOON_CMD_ID_STOP);
706 } else if (dev->free_page_report_status == FREE_PAGE_REPORT_S_DONE) {
707 config.free_page_report_cmd_id =
708 cpu_to_le32(VIRTIO_BALLOON_CMD_ID_DONE);
711 trace_virtio_balloon_get_config(config.num_pages, config.actual);
712 memcpy(config_data, &config, virtio_balloon_config_size(dev));
715 static int build_dimm_list(Object *obj, void *opaque)
717 GSList **list = opaque;
719 if (object_dynamic_cast(obj, TYPE_PC_DIMM)) {
720 DeviceState *dev = DEVICE(obj);
721 if (dev->realized) { /* only realized DIMMs matter */
722 *list = g_slist_prepend(*list, dev);
726 object_child_foreach(obj, build_dimm_list, opaque);
727 return 0;
730 static ram_addr_t get_current_ram_size(void)
732 GSList *list = NULL, *item;
733 ram_addr_t size = ram_size;
735 build_dimm_list(qdev_get_machine(), &list);
736 for (item = list; item; item = g_slist_next(item)) {
737 Object *obj = OBJECT(item->data);
738 if (!strcmp(object_get_typename(obj), TYPE_PC_DIMM)) {
739 size += object_property_get_int(obj, PC_DIMM_SIZE_PROP,
740 &error_abort);
743 g_slist_free(list);
745 return size;
748 static bool virtio_balloon_page_poison_support(void *opaque)
750 VirtIOBalloon *s = opaque;
751 VirtIODevice *vdev = VIRTIO_DEVICE(s);
753 return virtio_vdev_has_feature(vdev, VIRTIO_BALLOON_F_PAGE_POISON);
756 static void virtio_balloon_set_config(VirtIODevice *vdev,
757 const uint8_t *config_data)
759 VirtIOBalloon *dev = VIRTIO_BALLOON(vdev);
760 struct virtio_balloon_config config;
761 uint32_t oldactual = dev->actual;
762 ram_addr_t vm_ram_size = get_current_ram_size();
764 memcpy(&config, config_data, virtio_balloon_config_size(dev));
765 dev->actual = le32_to_cpu(config.actual);
766 if (dev->actual != oldactual) {
767 qapi_event_send_balloon_change(vm_ram_size -
768 ((ram_addr_t) dev->actual << VIRTIO_BALLOON_PFN_SHIFT));
770 dev->poison_val = 0;
771 if (virtio_balloon_page_poison_support(dev)) {
772 dev->poison_val = le32_to_cpu(config.poison_val);
774 trace_virtio_balloon_set_config(dev->actual, oldactual);
777 static uint64_t virtio_balloon_get_features(VirtIODevice *vdev, uint64_t f,
778 Error **errp)
780 VirtIOBalloon *dev = VIRTIO_BALLOON(vdev);
781 f |= dev->host_features;
782 virtio_add_feature(&f, VIRTIO_BALLOON_F_STATS_VQ);
784 return f;
787 static void virtio_balloon_stat(void *opaque, BalloonInfo *info)
789 VirtIOBalloon *dev = opaque;
790 info->actual = get_current_ram_size() - ((uint64_t) dev->actual <<
791 VIRTIO_BALLOON_PFN_SHIFT);
794 static void virtio_balloon_to_target(void *opaque, ram_addr_t target)
796 VirtIOBalloon *dev = VIRTIO_BALLOON(opaque);
797 VirtIODevice *vdev = VIRTIO_DEVICE(dev);
798 ram_addr_t vm_ram_size = get_current_ram_size();
800 if (target > vm_ram_size) {
801 target = vm_ram_size;
803 if (target) {
804 dev->num_pages = (vm_ram_size - target) >> VIRTIO_BALLOON_PFN_SHIFT;
805 virtio_notify_config(vdev);
807 trace_virtio_balloon_to_target(target, dev->num_pages);
810 static int virtio_balloon_post_load_device(void *opaque, int version_id)
812 VirtIOBalloon *s = VIRTIO_BALLOON(opaque);
814 if (balloon_stats_enabled(s)) {
815 balloon_stats_change_timer(s, s->stats_poll_interval);
817 return 0;
820 static const VMStateDescription vmstate_virtio_balloon_free_page_report = {
821 .name = "virtio-balloon-device/free-page-report",
822 .version_id = 1,
823 .minimum_version_id = 1,
824 .needed = virtio_balloon_free_page_support,
825 .fields = (VMStateField[]) {
826 VMSTATE_UINT32(free_page_report_cmd_id, VirtIOBalloon),
827 VMSTATE_UINT32(free_page_report_status, VirtIOBalloon),
828 VMSTATE_END_OF_LIST()
832 static const VMStateDescription vmstate_virtio_balloon_page_poison = {
833 .name = "vitio-balloon-device/page-poison",
834 .version_id = 1,
835 .minimum_version_id = 1,
836 .needed = virtio_balloon_page_poison_support,
837 .fields = (VMStateField[]) {
838 VMSTATE_UINT32(poison_val, VirtIOBalloon),
839 VMSTATE_END_OF_LIST()
843 static const VMStateDescription vmstate_virtio_balloon_device = {
844 .name = "virtio-balloon-device",
845 .version_id = 1,
846 .minimum_version_id = 1,
847 .post_load = virtio_balloon_post_load_device,
848 .fields = (VMStateField[]) {
849 VMSTATE_UINT32(num_pages, VirtIOBalloon),
850 VMSTATE_UINT32(actual, VirtIOBalloon),
851 VMSTATE_END_OF_LIST()
853 .subsections = (const VMStateDescription * []) {
854 &vmstate_virtio_balloon_free_page_report,
855 &vmstate_virtio_balloon_page_poison,
856 NULL
860 static void virtio_balloon_device_realize(DeviceState *dev, Error **errp)
862 VirtIODevice *vdev = VIRTIO_DEVICE(dev);
863 VirtIOBalloon *s = VIRTIO_BALLOON(dev);
864 int ret;
866 virtio_init(vdev, "virtio-balloon", VIRTIO_ID_BALLOON,
867 virtio_balloon_config_size(s));
869 ret = qemu_add_balloon_handler(virtio_balloon_to_target,
870 virtio_balloon_stat, s);
872 if (ret < 0) {
873 error_setg(errp, "Only one balloon device is supported");
874 virtio_cleanup(vdev);
875 return;
878 if (virtio_has_feature(s->host_features, VIRTIO_BALLOON_F_FREE_PAGE_HINT) &&
879 !s->iothread) {
880 error_setg(errp, "'free-page-hint' requires 'iothread' to be set");
881 virtio_cleanup(vdev);
882 return;
885 s->ivq = virtio_add_queue(vdev, 128, virtio_balloon_handle_output);
886 s->dvq = virtio_add_queue(vdev, 128, virtio_balloon_handle_output);
887 s->svq = virtio_add_queue(vdev, 128, virtio_balloon_receive_stats);
889 if (virtio_has_feature(s->host_features,
890 VIRTIO_BALLOON_F_FREE_PAGE_HINT)) {
891 s->free_page_vq = virtio_add_queue(vdev, VIRTQUEUE_MAX_SIZE,
892 virtio_balloon_handle_free_page_vq);
893 precopy_add_notifier(&s->free_page_report_notify);
895 object_ref(OBJECT(s->iothread));
896 s->free_page_bh = aio_bh_new(iothread_get_aio_context(s->iothread),
897 virtio_ballloon_get_free_page_hints, s);
900 if (virtio_has_feature(s->host_features, VIRTIO_BALLOON_F_REPORTING)) {
901 s->reporting_vq = virtio_add_queue(vdev, 32,
902 virtio_balloon_handle_report);
905 reset_stats(s);
908 static void virtio_balloon_device_unrealize(DeviceState *dev)
910 VirtIODevice *vdev = VIRTIO_DEVICE(dev);
911 VirtIOBalloon *s = VIRTIO_BALLOON(dev);
913 if (s->free_page_bh) {
914 qemu_bh_delete(s->free_page_bh);
915 object_unref(OBJECT(s->iothread));
916 virtio_balloon_free_page_stop(s);
917 precopy_remove_notifier(&s->free_page_report_notify);
919 balloon_stats_destroy_timer(s);
920 qemu_remove_balloon_handler(s);
922 virtio_delete_queue(s->ivq);
923 virtio_delete_queue(s->dvq);
924 virtio_delete_queue(s->svq);
925 if (s->free_page_vq) {
926 virtio_delete_queue(s->free_page_vq);
928 if (s->reporting_vq) {
929 virtio_delete_queue(s->reporting_vq);
931 virtio_cleanup(vdev);
934 static void virtio_balloon_device_reset(VirtIODevice *vdev)
936 VirtIOBalloon *s = VIRTIO_BALLOON(vdev);
938 if (virtio_balloon_free_page_support(s)) {
939 virtio_balloon_free_page_stop(s);
942 if (s->stats_vq_elem != NULL) {
943 virtqueue_unpop(s->svq, s->stats_vq_elem, 0);
944 g_free(s->stats_vq_elem);
945 s->stats_vq_elem = NULL;
948 s->poison_val = 0;
951 static void virtio_balloon_set_status(VirtIODevice *vdev, uint8_t status)
953 VirtIOBalloon *s = VIRTIO_BALLOON(vdev);
955 if (!s->stats_vq_elem && vdev->vm_running &&
956 (status & VIRTIO_CONFIG_S_DRIVER_OK) && virtqueue_rewind(s->svq, 1)) {
957 /* poll stats queue for the element we have discarded when the VM
958 * was stopped */
959 virtio_balloon_receive_stats(vdev, s->svq);
962 if (virtio_balloon_free_page_support(s)) {
964 * The VM is woken up and the iothread was blocked, so signal it to
965 * continue.
967 if (vdev->vm_running && s->block_iothread) {
968 qemu_mutex_lock(&s->free_page_lock);
969 s->block_iothread = false;
970 qemu_cond_signal(&s->free_page_cond);
971 qemu_mutex_unlock(&s->free_page_lock);
974 /* The VM is stopped, block the iothread. */
975 if (!vdev->vm_running) {
976 qemu_mutex_lock(&s->free_page_lock);
977 s->block_iothread = true;
978 qemu_mutex_unlock(&s->free_page_lock);
983 static void virtio_balloon_instance_init(Object *obj)
985 VirtIOBalloon *s = VIRTIO_BALLOON(obj);
987 qemu_mutex_init(&s->free_page_lock);
988 qemu_cond_init(&s->free_page_cond);
989 s->free_page_report_cmd_id = VIRTIO_BALLOON_FREE_PAGE_REPORT_CMD_ID_MIN;
990 s->free_page_report_notify.notify = virtio_balloon_free_page_report_notify;
992 object_property_add(obj, "guest-stats", "guest statistics",
993 balloon_stats_get_all, NULL, NULL, s);
995 object_property_add(obj, "guest-stats-polling-interval", "int",
996 balloon_stats_get_poll_interval,
997 balloon_stats_set_poll_interval,
998 NULL, s);
1001 static const VMStateDescription vmstate_virtio_balloon = {
1002 .name = "virtio-balloon",
1003 .minimum_version_id = 1,
1004 .version_id = 1,
1005 .fields = (VMStateField[]) {
1006 VMSTATE_VIRTIO_DEVICE,
1007 VMSTATE_END_OF_LIST()
1011 static Property virtio_balloon_properties[] = {
1012 DEFINE_PROP_BIT("deflate-on-oom", VirtIOBalloon, host_features,
1013 VIRTIO_BALLOON_F_DEFLATE_ON_OOM, false),
1014 DEFINE_PROP_BIT("free-page-hint", VirtIOBalloon, host_features,
1015 VIRTIO_BALLOON_F_FREE_PAGE_HINT, false),
1016 DEFINE_PROP_BIT("page-poison", VirtIOBalloon, host_features,
1017 VIRTIO_BALLOON_F_PAGE_POISON, true),
1018 DEFINE_PROP_BIT("free-page-reporting", VirtIOBalloon, host_features,
1019 VIRTIO_BALLOON_F_REPORTING, false),
1020 /* QEMU 4.0 accidentally changed the config size even when free-page-hint
1021 * is disabled, resulting in QEMU 3.1 migration incompatibility. This
1022 * property retains this quirk for QEMU 4.1 machine types.
1024 DEFINE_PROP_BOOL("qemu-4-0-config-size", VirtIOBalloon,
1025 qemu_4_0_config_size, false),
1026 DEFINE_PROP_LINK("iothread", VirtIOBalloon, iothread, TYPE_IOTHREAD,
1027 IOThread *),
1028 DEFINE_PROP_END_OF_LIST(),
1031 static void virtio_balloon_class_init(ObjectClass *klass, void *data)
1033 DeviceClass *dc = DEVICE_CLASS(klass);
1034 VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
1036 device_class_set_props(dc, virtio_balloon_properties);
1037 dc->vmsd = &vmstate_virtio_balloon;
1038 set_bit(DEVICE_CATEGORY_MISC, dc->categories);
1039 vdc->realize = virtio_balloon_device_realize;
1040 vdc->unrealize = virtio_balloon_device_unrealize;
1041 vdc->reset = virtio_balloon_device_reset;
1042 vdc->get_config = virtio_balloon_get_config;
1043 vdc->set_config = virtio_balloon_set_config;
1044 vdc->get_features = virtio_balloon_get_features;
1045 vdc->set_status = virtio_balloon_set_status;
1046 vdc->vmsd = &vmstate_virtio_balloon_device;
1049 static const TypeInfo virtio_balloon_info = {
1050 .name = TYPE_VIRTIO_BALLOON,
1051 .parent = TYPE_VIRTIO_DEVICE,
1052 .instance_size = sizeof(VirtIOBalloon),
1053 .instance_init = virtio_balloon_instance_init,
1054 .class_init = virtio_balloon_class_init,
1057 static void virtio_register_types(void)
1059 type_register_static(&virtio_balloon_info);
1062 type_init(virtio_register_types)