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>
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"
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 "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"
29 #include "qemu/error-report.h"
30 #include "migration/misc.h"
32 #include "hw/virtio/virtio-bus.h"
33 #include "hw/virtio/virtio-access.h"
35 #define BALLOON_PAGE_SIZE (1 << VIRTIO_BALLOON_PFN_SHIFT)
37 typedef struct PartiallyBalloonedPage
{
39 unsigned long *bitmap
;
40 } PartiallyBalloonedPage
;
42 static void virtio_balloon_pbp_free(PartiallyBalloonedPage
*pbp
)
51 static void virtio_balloon_pbp_alloc(PartiallyBalloonedPage
*pbp
,
55 pbp
->base_gpa
= base_gpa
;
56 pbp
->bitmap
= bitmap_new(subpages
);
59 static bool virtio_balloon_pbp_matches(PartiallyBalloonedPage
*pbp
,
62 return pbp
->base_gpa
== base_gpa
;
65 static void balloon_inflate_page(VirtIOBalloon
*balloon
,
66 MemoryRegion
*mr
, hwaddr mr_offset
,
67 PartiallyBalloonedPage
*pbp
)
69 void *addr
= memory_region_get_ram_ptr(mr
) + mr_offset
;
70 ram_addr_t rb_offset
, rb_aligned_offset
, base_gpa
;
75 /* XXX is there a better way to get to the RAMBlock than via a
77 rb
= qemu_ram_block_from_host(addr
, false, &rb_offset
);
78 rb_page_size
= qemu_ram_pagesize(rb
);
80 if (rb_page_size
== BALLOON_PAGE_SIZE
) {
83 ram_block_discard_range(rb
, rb_offset
, rb_page_size
);
84 /* We ignore errors from ram_block_discard_range(), because it
85 * has already reported them, and failing to discard a balloon
86 * page is not fatal */
92 * We've put a piece of a larger host page into the balloon - we
93 * need to keep track until we have a whole host page to
97 "Balloon used with backing page size > 4kiB, this may not be reliable");
99 rb_aligned_offset
= QEMU_ALIGN_DOWN(rb_offset
, rb_page_size
);
100 subpages
= rb_page_size
/ BALLOON_PAGE_SIZE
;
101 base_gpa
= memory_region_get_ram_addr(mr
) + mr_offset
-
102 (rb_offset
- rb_aligned_offset
);
104 if (pbp
->bitmap
&& !virtio_balloon_pbp_matches(pbp
, base_gpa
)) {
105 /* We've partially ballooned part of a host page, but now
106 * we're trying to balloon part of a different one. Too hard,
107 * give up on the old partial page */
108 virtio_balloon_pbp_free(pbp
);
112 virtio_balloon_pbp_alloc(pbp
, base_gpa
, subpages
);
115 set_bit((rb_offset
- rb_aligned_offset
) / BALLOON_PAGE_SIZE
,
118 if (bitmap_full(pbp
->bitmap
, subpages
)) {
119 /* We've accumulated a full host page, we can actually discard
122 ram_block_discard_range(rb
, rb_aligned_offset
, rb_page_size
);
123 /* We ignore errors from ram_block_discard_range(), because it
124 * has already reported them, and failing to discard a balloon
125 * page is not fatal */
126 virtio_balloon_pbp_free(pbp
);
130 static void balloon_deflate_page(VirtIOBalloon
*balloon
,
131 MemoryRegion
*mr
, hwaddr mr_offset
)
133 void *addr
= memory_region_get_ram_ptr(mr
) + mr_offset
;
134 ram_addr_t rb_offset
;
140 /* XXX is there a better way to get to the RAMBlock than via a
142 rb
= qemu_ram_block_from_host(addr
, false, &rb_offset
);
143 rb_page_size
= qemu_ram_pagesize(rb
);
145 host_addr
= (void *)((uintptr_t)addr
& ~(rb_page_size
- 1));
147 /* When a page is deflated, we hint the whole host page it lives
148 * on, since we can't do anything smaller */
149 ret
= qemu_madvise(host_addr
, rb_page_size
, QEMU_MADV_WILLNEED
);
151 warn_report("Couldn't MADV_WILLNEED on balloon deflate: %s",
153 /* Otherwise ignore, failing to page hint shouldn't be fatal */
157 static const char *balloon_stat_names
[] = {
158 [VIRTIO_BALLOON_S_SWAP_IN
] = "stat-swap-in",
159 [VIRTIO_BALLOON_S_SWAP_OUT
] = "stat-swap-out",
160 [VIRTIO_BALLOON_S_MAJFLT
] = "stat-major-faults",
161 [VIRTIO_BALLOON_S_MINFLT
] = "stat-minor-faults",
162 [VIRTIO_BALLOON_S_MEMFREE
] = "stat-free-memory",
163 [VIRTIO_BALLOON_S_MEMTOT
] = "stat-total-memory",
164 [VIRTIO_BALLOON_S_AVAIL
] = "stat-available-memory",
165 [VIRTIO_BALLOON_S_CACHES
] = "stat-disk-caches",
166 [VIRTIO_BALLOON_S_HTLB_PGALLOC
] = "stat-htlb-pgalloc",
167 [VIRTIO_BALLOON_S_HTLB_PGFAIL
] = "stat-htlb-pgfail",
168 [VIRTIO_BALLOON_S_NR
] = NULL
172 * reset_stats - Mark all items in the stats array as unset
174 * This function needs to be called at device initialization and before
175 * updating to a set of newly-generated stats. This will ensure that no
176 * stale values stick around in case the guest reports a subset of the supported
179 static inline void reset_stats(VirtIOBalloon
*dev
)
182 for (i
= 0; i
< VIRTIO_BALLOON_S_NR
; dev
->stats
[i
++] = -1);
185 static bool balloon_stats_supported(const VirtIOBalloon
*s
)
187 VirtIODevice
*vdev
= VIRTIO_DEVICE(s
);
188 return virtio_vdev_has_feature(vdev
, VIRTIO_BALLOON_F_STATS_VQ
);
191 static bool balloon_stats_enabled(const VirtIOBalloon
*s
)
193 return s
->stats_poll_interval
> 0;
196 static void balloon_stats_destroy_timer(VirtIOBalloon
*s
)
198 if (balloon_stats_enabled(s
)) {
199 timer_del(s
->stats_timer
);
200 timer_free(s
->stats_timer
);
201 s
->stats_timer
= NULL
;
202 s
->stats_poll_interval
= 0;
206 static void balloon_stats_change_timer(VirtIOBalloon
*s
, int64_t secs
)
208 timer_mod(s
->stats_timer
, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL
) + secs
* 1000);
211 static void balloon_stats_poll_cb(void *opaque
)
213 VirtIOBalloon
*s
= opaque
;
214 VirtIODevice
*vdev
= VIRTIO_DEVICE(s
);
216 if (s
->stats_vq_elem
== NULL
|| !balloon_stats_supported(s
)) {
218 balloon_stats_change_timer(s
, s
->stats_poll_interval
);
222 virtqueue_push(s
->svq
, s
->stats_vq_elem
, s
->stats_vq_offset
);
223 virtio_notify(vdev
, s
->svq
);
224 g_free(s
->stats_vq_elem
);
225 s
->stats_vq_elem
= NULL
;
228 static void balloon_stats_get_all(Object
*obj
, Visitor
*v
, const char *name
,
229 void *opaque
, Error
**errp
)
232 VirtIOBalloon
*s
= opaque
;
235 visit_start_struct(v
, name
, NULL
, 0, &err
);
239 visit_type_int(v
, "last-update", &s
->stats_last_update
, &err
);
244 visit_start_struct(v
, "stats", NULL
, 0, &err
);
248 for (i
= 0; i
< VIRTIO_BALLOON_S_NR
; i
++) {
249 visit_type_uint64(v
, balloon_stat_names
[i
], &s
->stats
[i
], &err
);
254 visit_check_struct(v
, &err
);
256 visit_end_struct(v
, NULL
);
259 visit_check_struct(v
, &err
);
262 visit_end_struct(v
, NULL
);
264 error_propagate(errp
, err
);
267 static void balloon_stats_get_poll_interval(Object
*obj
, Visitor
*v
,
268 const char *name
, void *opaque
,
271 VirtIOBalloon
*s
= opaque
;
272 visit_type_int(v
, name
, &s
->stats_poll_interval
, errp
);
275 static void balloon_stats_set_poll_interval(Object
*obj
, Visitor
*v
,
276 const char *name
, void *opaque
,
279 VirtIOBalloon
*s
= opaque
;
280 Error
*local_err
= NULL
;
283 visit_type_int(v
, name
, &value
, &local_err
);
285 error_propagate(errp
, local_err
);
290 error_setg(errp
, "timer value must be greater than zero");
294 if (value
> UINT32_MAX
) {
295 error_setg(errp
, "timer value is too big");
299 if (value
== s
->stats_poll_interval
) {
304 /* timer=0 disables the timer */
305 balloon_stats_destroy_timer(s
);
309 if (balloon_stats_enabled(s
)) {
310 /* timer interval change */
311 s
->stats_poll_interval
= value
;
312 balloon_stats_change_timer(s
, value
);
316 /* create a new timer */
317 g_assert(s
->stats_timer
== NULL
);
318 s
->stats_timer
= timer_new_ms(QEMU_CLOCK_VIRTUAL
, balloon_stats_poll_cb
, s
);
319 s
->stats_poll_interval
= value
;
320 balloon_stats_change_timer(s
, 0);
323 static void virtio_balloon_handle_output(VirtIODevice
*vdev
, VirtQueue
*vq
)
325 VirtIOBalloon
*s
= VIRTIO_BALLOON(vdev
);
326 VirtQueueElement
*elem
;
327 MemoryRegionSection section
;
330 PartiallyBalloonedPage pbp
= {};
334 elem
= virtqueue_pop(vq
, sizeof(VirtQueueElement
));
339 while (iov_to_buf(elem
->out_sg
, elem
->out_num
, offset
, &pfn
, 4) == 4) {
340 unsigned int p
= virtio_ldl_p(vdev
, &pfn
);
343 pa
= (hwaddr
) p
<< VIRTIO_BALLOON_PFN_SHIFT
;
346 section
= memory_region_find(get_system_memory(), pa
,
349 trace_virtio_balloon_bad_addr(pa
);
352 if (!memory_region_is_ram(section
.mr
) ||
353 memory_region_is_rom(section
.mr
) ||
354 memory_region_is_romd(section
.mr
)) {
355 trace_virtio_balloon_bad_addr(pa
);
356 memory_region_unref(section
.mr
);
360 trace_virtio_balloon_handle_output(memory_region_name(section
.mr
),
362 if (!qemu_balloon_is_inhibited()) {
364 balloon_inflate_page(s
, section
.mr
,
365 section
.offset_within_region
, &pbp
);
366 } else if (vq
== s
->dvq
) {
367 balloon_deflate_page(s
, section
.mr
, section
.offset_within_region
);
369 g_assert_not_reached();
372 memory_region_unref(section
.mr
);
375 virtqueue_push(vq
, elem
, offset
);
376 virtio_notify(vdev
, vq
);
378 virtio_balloon_pbp_free(&pbp
);
382 static void virtio_balloon_receive_stats(VirtIODevice
*vdev
, VirtQueue
*vq
)
384 VirtIOBalloon
*s
= VIRTIO_BALLOON(vdev
);
385 VirtQueueElement
*elem
;
386 VirtIOBalloonStat stat
;
390 elem
= virtqueue_pop(vq
, sizeof(VirtQueueElement
));
395 if (s
->stats_vq_elem
!= NULL
) {
396 /* This should never happen if the driver follows the spec. */
397 virtqueue_push(vq
, s
->stats_vq_elem
, 0);
398 virtio_notify(vdev
, vq
);
399 g_free(s
->stats_vq_elem
);
402 s
->stats_vq_elem
= elem
;
404 /* Initialize the stats to get rid of any stale values. This is only
405 * needed to handle the case where a guest supports fewer stats than it
406 * used to (ie. it has booted into an old kernel).
410 while (iov_to_buf(elem
->out_sg
, elem
->out_num
, offset
, &stat
, sizeof(stat
))
412 uint16_t tag
= virtio_tswap16(vdev
, stat
.tag
);
413 uint64_t val
= virtio_tswap64(vdev
, stat
.val
);
415 offset
+= sizeof(stat
);
416 if (tag
< VIRTIO_BALLOON_S_NR
)
419 s
->stats_vq_offset
= offset
;
421 if (qemu_gettimeofday(&tv
) < 0) {
422 warn_report("%s: failed to get time of day", __func__
);
426 s
->stats_last_update
= tv
.tv_sec
;
429 if (balloon_stats_enabled(s
)) {
430 balloon_stats_change_timer(s
, s
->stats_poll_interval
);
434 static void virtio_balloon_handle_free_page_vq(VirtIODevice
*vdev
,
437 VirtIOBalloon
*s
= VIRTIO_BALLOON(vdev
);
438 qemu_bh_schedule(s
->free_page_bh
);
441 static bool get_free_page_hints(VirtIOBalloon
*dev
)
443 VirtQueueElement
*elem
;
444 VirtIODevice
*vdev
= VIRTIO_DEVICE(dev
);
445 VirtQueue
*vq
= dev
->free_page_vq
;
448 while (dev
->block_iothread
) {
449 qemu_cond_wait(&dev
->free_page_cond
, &dev
->free_page_lock
);
452 elem
= virtqueue_pop(vq
, sizeof(VirtQueueElement
));
459 size_t size
= iov_to_buf(elem
->out_sg
, elem
->out_num
, 0,
462 virtio_tswap32s(vdev
, &id
);
463 if (unlikely(size
!= sizeof(id
))) {
464 virtio_error(vdev
, "received an incorrect cmd id");
468 if (id
== dev
->free_page_report_cmd_id
) {
469 dev
->free_page_report_status
= FREE_PAGE_REPORT_S_START
;
472 * Stop the optimization only when it has started. This
473 * avoids a stale stop sign for the previous command.
475 if (dev
->free_page_report_status
== FREE_PAGE_REPORT_S_START
) {
476 dev
->free_page_report_status
= FREE_PAGE_REPORT_S_STOP
;
482 if (dev
->free_page_report_status
== FREE_PAGE_REPORT_S_START
) {
483 qemu_guest_free_page_hint(elem
->in_sg
[0].iov_base
,
484 elem
->in_sg
[0].iov_len
);
489 virtqueue_push(vq
, elem
, 1);
494 static void virtio_ballloon_get_free_page_hints(void *opaque
)
496 VirtIOBalloon
*dev
= opaque
;
497 VirtIODevice
*vdev
= VIRTIO_DEVICE(dev
);
498 VirtQueue
*vq
= dev
->free_page_vq
;
499 bool continue_to_get_hints
;
502 qemu_mutex_lock(&dev
->free_page_lock
);
503 virtio_queue_set_notification(vq
, 0);
504 continue_to_get_hints
= get_free_page_hints(dev
);
505 qemu_mutex_unlock(&dev
->free_page_lock
);
506 virtio_notify(vdev
, vq
);
508 * Start to poll the vq once the reporting started. Otherwise, continue
509 * only when there are entries on the vq, which need to be given back.
511 } while (continue_to_get_hints
||
512 dev
->free_page_report_status
== FREE_PAGE_REPORT_S_START
);
513 virtio_queue_set_notification(vq
, 1);
516 static bool virtio_balloon_free_page_support(void *opaque
)
518 VirtIOBalloon
*s
= opaque
;
519 VirtIODevice
*vdev
= VIRTIO_DEVICE(s
);
521 return virtio_vdev_has_feature(vdev
, VIRTIO_BALLOON_F_FREE_PAGE_HINT
);
524 static void virtio_balloon_free_page_start(VirtIOBalloon
*s
)
526 VirtIODevice
*vdev
= VIRTIO_DEVICE(s
);
528 /* For the stop and copy phase, we don't need to start the optimization */
529 if (!vdev
->vm_running
) {
533 if (s
->free_page_report_cmd_id
== UINT_MAX
) {
534 s
->free_page_report_cmd_id
=
535 VIRTIO_BALLOON_FREE_PAGE_REPORT_CMD_ID_MIN
;
537 s
->free_page_report_cmd_id
++;
540 s
->free_page_report_status
= FREE_PAGE_REPORT_S_REQUESTED
;
541 virtio_notify_config(vdev
);
544 static void virtio_balloon_free_page_stop(VirtIOBalloon
*s
)
546 VirtIODevice
*vdev
= VIRTIO_DEVICE(s
);
548 if (s
->free_page_report_status
!= FREE_PAGE_REPORT_S_STOP
) {
550 * The lock also guarantees us that the
551 * virtio_ballloon_get_free_page_hints exits after the
552 * free_page_report_status is set to S_STOP.
554 qemu_mutex_lock(&s
->free_page_lock
);
556 * The guest hasn't done the reporting, so host sends a notification
557 * to the guest to actively stop the reporting.
559 s
->free_page_report_status
= FREE_PAGE_REPORT_S_STOP
;
560 qemu_mutex_unlock(&s
->free_page_lock
);
561 virtio_notify_config(vdev
);
565 static void virtio_balloon_free_page_done(VirtIOBalloon
*s
)
567 VirtIODevice
*vdev
= VIRTIO_DEVICE(s
);
569 s
->free_page_report_status
= FREE_PAGE_REPORT_S_DONE
;
570 virtio_notify_config(vdev
);
574 virtio_balloon_free_page_report_notify(NotifierWithReturn
*n
, void *data
)
576 VirtIOBalloon
*dev
= container_of(n
, VirtIOBalloon
,
577 free_page_report_notify
);
578 VirtIODevice
*vdev
= VIRTIO_DEVICE(dev
);
579 PrecopyNotifyData
*pnd
= data
;
581 if (!virtio_balloon_free_page_support(dev
)) {
583 * This is an optimization provided to migration, so just return 0 to
584 * have the normal migration process not affected when this feature is
590 switch (pnd
->reason
) {
591 case PRECOPY_NOTIFY_SETUP
:
592 precopy_enable_free_page_optimization();
594 case PRECOPY_NOTIFY_COMPLETE
:
595 case PRECOPY_NOTIFY_CLEANUP
:
596 case PRECOPY_NOTIFY_BEFORE_BITMAP_SYNC
:
597 virtio_balloon_free_page_stop(dev
);
599 case PRECOPY_NOTIFY_AFTER_BITMAP_SYNC
:
600 if (vdev
->vm_running
) {
601 virtio_balloon_free_page_start(dev
);
603 virtio_balloon_free_page_done(dev
);
607 virtio_error(vdev
, "%s: %d reason unknown", __func__
, pnd
->reason
);
613 static size_t virtio_balloon_config_size(VirtIOBalloon
*s
)
615 uint64_t features
= s
->host_features
;
617 if (s
->qemu_4_0_config_size
) {
618 return sizeof(struct virtio_balloon_config
);
620 if (virtio_has_feature(features
, VIRTIO_BALLOON_F_PAGE_POISON
)) {
621 return sizeof(struct virtio_balloon_config
);
623 if (virtio_has_feature(features
, VIRTIO_BALLOON_F_FREE_PAGE_HINT
)) {
624 return offsetof(struct virtio_balloon_config
, poison_val
);
626 return offsetof(struct virtio_balloon_config
, free_page_report_cmd_id
);
629 static void virtio_balloon_get_config(VirtIODevice
*vdev
, uint8_t *config_data
)
631 VirtIOBalloon
*dev
= VIRTIO_BALLOON(vdev
);
632 struct virtio_balloon_config config
= {};
634 config
.num_pages
= cpu_to_le32(dev
->num_pages
);
635 config
.actual
= cpu_to_le32(dev
->actual
);
637 if (dev
->free_page_report_status
== FREE_PAGE_REPORT_S_REQUESTED
) {
638 config
.free_page_report_cmd_id
=
639 cpu_to_le32(dev
->free_page_report_cmd_id
);
640 } else if (dev
->free_page_report_status
== FREE_PAGE_REPORT_S_STOP
) {
641 config
.free_page_report_cmd_id
=
642 cpu_to_le32(VIRTIO_BALLOON_CMD_ID_STOP
);
643 } else if (dev
->free_page_report_status
== FREE_PAGE_REPORT_S_DONE
) {
644 config
.free_page_report_cmd_id
=
645 cpu_to_le32(VIRTIO_BALLOON_CMD_ID_DONE
);
648 trace_virtio_balloon_get_config(config
.num_pages
, config
.actual
);
649 memcpy(config_data
, &config
, virtio_balloon_config_size(dev
));
652 static int build_dimm_list(Object
*obj
, void *opaque
)
654 GSList
**list
= opaque
;
656 if (object_dynamic_cast(obj
, TYPE_PC_DIMM
)) {
657 DeviceState
*dev
= DEVICE(obj
);
658 if (dev
->realized
) { /* only realized DIMMs matter */
659 *list
= g_slist_prepend(*list
, dev
);
663 object_child_foreach(obj
, build_dimm_list
, opaque
);
667 static ram_addr_t
get_current_ram_size(void)
669 GSList
*list
= NULL
, *item
;
670 ram_addr_t size
= ram_size
;
672 build_dimm_list(qdev_get_machine(), &list
);
673 for (item
= list
; item
; item
= g_slist_next(item
)) {
674 Object
*obj
= OBJECT(item
->data
);
675 if (!strcmp(object_get_typename(obj
), TYPE_PC_DIMM
)) {
676 size
+= object_property_get_int(obj
, PC_DIMM_SIZE_PROP
,
685 static void virtio_balloon_set_config(VirtIODevice
*vdev
,
686 const uint8_t *config_data
)
688 VirtIOBalloon
*dev
= VIRTIO_BALLOON(vdev
);
689 struct virtio_balloon_config config
;
690 uint32_t oldactual
= dev
->actual
;
691 ram_addr_t vm_ram_size
= get_current_ram_size();
693 memcpy(&config
, config_data
, virtio_balloon_config_size(dev
));
694 dev
->actual
= le32_to_cpu(config
.actual
);
695 if (dev
->actual
!= oldactual
) {
696 qapi_event_send_balloon_change(vm_ram_size
-
697 ((ram_addr_t
) dev
->actual
<< VIRTIO_BALLOON_PFN_SHIFT
));
699 trace_virtio_balloon_set_config(dev
->actual
, oldactual
);
702 static uint64_t virtio_balloon_get_features(VirtIODevice
*vdev
, uint64_t f
,
705 VirtIOBalloon
*dev
= VIRTIO_BALLOON(vdev
);
706 f
|= dev
->host_features
;
707 virtio_add_feature(&f
, VIRTIO_BALLOON_F_STATS_VQ
);
712 static void virtio_balloon_stat(void *opaque
, BalloonInfo
*info
)
714 VirtIOBalloon
*dev
= opaque
;
715 info
->actual
= get_current_ram_size() - ((uint64_t) dev
->actual
<<
716 VIRTIO_BALLOON_PFN_SHIFT
);
719 static void virtio_balloon_to_target(void *opaque
, ram_addr_t target
)
721 VirtIOBalloon
*dev
= VIRTIO_BALLOON(opaque
);
722 VirtIODevice
*vdev
= VIRTIO_DEVICE(dev
);
723 ram_addr_t vm_ram_size
= get_current_ram_size();
725 if (target
> vm_ram_size
) {
726 target
= vm_ram_size
;
729 dev
->num_pages
= (vm_ram_size
- target
) >> VIRTIO_BALLOON_PFN_SHIFT
;
730 virtio_notify_config(vdev
);
732 trace_virtio_balloon_to_target(target
, dev
->num_pages
);
735 static int virtio_balloon_post_load_device(void *opaque
, int version_id
)
737 VirtIOBalloon
*s
= VIRTIO_BALLOON(opaque
);
739 if (balloon_stats_enabled(s
)) {
740 balloon_stats_change_timer(s
, s
->stats_poll_interval
);
745 static const VMStateDescription vmstate_virtio_balloon_free_page_report
= {
746 .name
= "virtio-balloon-device/free-page-report",
748 .minimum_version_id
= 1,
749 .needed
= virtio_balloon_free_page_support
,
750 .fields
= (VMStateField
[]) {
751 VMSTATE_UINT32(free_page_report_cmd_id
, VirtIOBalloon
),
752 VMSTATE_UINT32(free_page_report_status
, VirtIOBalloon
),
753 VMSTATE_END_OF_LIST()
757 static const VMStateDescription vmstate_virtio_balloon_device
= {
758 .name
= "virtio-balloon-device",
760 .minimum_version_id
= 1,
761 .post_load
= virtio_balloon_post_load_device
,
762 .fields
= (VMStateField
[]) {
763 VMSTATE_UINT32(num_pages
, VirtIOBalloon
),
764 VMSTATE_UINT32(actual
, VirtIOBalloon
),
765 VMSTATE_END_OF_LIST()
767 .subsections
= (const VMStateDescription
* []) {
768 &vmstate_virtio_balloon_free_page_report
,
773 static void virtio_balloon_device_realize(DeviceState
*dev
, Error
**errp
)
775 VirtIODevice
*vdev
= VIRTIO_DEVICE(dev
);
776 VirtIOBalloon
*s
= VIRTIO_BALLOON(dev
);
779 virtio_init(vdev
, "virtio-balloon", VIRTIO_ID_BALLOON
,
780 virtio_balloon_config_size(s
));
782 ret
= qemu_add_balloon_handler(virtio_balloon_to_target
,
783 virtio_balloon_stat
, s
);
786 error_setg(errp
, "Only one balloon device is supported");
787 virtio_cleanup(vdev
);
791 s
->ivq
= virtio_add_queue(vdev
, 128, virtio_balloon_handle_output
);
792 s
->dvq
= virtio_add_queue(vdev
, 128, virtio_balloon_handle_output
);
793 s
->svq
= virtio_add_queue(vdev
, 128, virtio_balloon_receive_stats
);
795 if (virtio_has_feature(s
->host_features
,
796 VIRTIO_BALLOON_F_FREE_PAGE_HINT
)) {
797 s
->free_page_vq
= virtio_add_queue(vdev
, VIRTQUEUE_MAX_SIZE
,
798 virtio_balloon_handle_free_page_vq
);
799 s
->free_page_report_status
= FREE_PAGE_REPORT_S_STOP
;
800 s
->free_page_report_cmd_id
=
801 VIRTIO_BALLOON_FREE_PAGE_REPORT_CMD_ID_MIN
;
802 s
->free_page_report_notify
.notify
=
803 virtio_balloon_free_page_report_notify
;
804 precopy_add_notifier(&s
->free_page_report_notify
);
806 object_ref(OBJECT(s
->iothread
));
807 s
->free_page_bh
= aio_bh_new(iothread_get_aio_context(s
->iothread
),
808 virtio_ballloon_get_free_page_hints
, s
);
809 qemu_mutex_init(&s
->free_page_lock
);
810 qemu_cond_init(&s
->free_page_cond
);
811 s
->block_iothread
= false;
813 /* Simply disable this feature if the iothread wasn't created. */
814 s
->host_features
&= ~(1 << VIRTIO_BALLOON_F_FREE_PAGE_HINT
);
815 virtio_error(vdev
, "iothread is missing");
821 static void virtio_balloon_device_unrealize(DeviceState
*dev
, Error
**errp
)
823 VirtIODevice
*vdev
= VIRTIO_DEVICE(dev
);
824 VirtIOBalloon
*s
= VIRTIO_BALLOON(dev
);
826 if (virtio_balloon_free_page_support(s
)) {
827 qemu_bh_delete(s
->free_page_bh
);
828 virtio_balloon_free_page_stop(s
);
829 precopy_remove_notifier(&s
->free_page_report_notify
);
831 balloon_stats_destroy_timer(s
);
832 qemu_remove_balloon_handler(s
);
833 virtio_cleanup(vdev
);
836 static void virtio_balloon_device_reset(VirtIODevice
*vdev
)
838 VirtIOBalloon
*s
= VIRTIO_BALLOON(vdev
);
840 if (virtio_balloon_free_page_support(s
)) {
841 virtio_balloon_free_page_stop(s
);
844 if (s
->stats_vq_elem
!= NULL
) {
845 virtqueue_unpop(s
->svq
, s
->stats_vq_elem
, 0);
846 g_free(s
->stats_vq_elem
);
847 s
->stats_vq_elem
= NULL
;
851 static void virtio_balloon_set_status(VirtIODevice
*vdev
, uint8_t status
)
853 VirtIOBalloon
*s
= VIRTIO_BALLOON(vdev
);
855 if (!s
->stats_vq_elem
&& vdev
->vm_running
&&
856 (status
& VIRTIO_CONFIG_S_DRIVER_OK
) && virtqueue_rewind(s
->svq
, 1)) {
857 /* poll stats queue for the element we have discarded when the VM
859 virtio_balloon_receive_stats(vdev
, s
->svq
);
862 if (virtio_balloon_free_page_support(s
)) {
864 * The VM is woken up and the iothread was blocked, so signal it to
867 if (vdev
->vm_running
&& s
->block_iothread
) {
868 qemu_mutex_lock(&s
->free_page_lock
);
869 s
->block_iothread
= false;
870 qemu_cond_signal(&s
->free_page_cond
);
871 qemu_mutex_unlock(&s
->free_page_lock
);
874 /* The VM is stopped, block the iothread. */
875 if (!vdev
->vm_running
) {
876 qemu_mutex_lock(&s
->free_page_lock
);
877 s
->block_iothread
= true;
878 qemu_mutex_unlock(&s
->free_page_lock
);
883 static void virtio_balloon_instance_init(Object
*obj
)
885 VirtIOBalloon
*s
= VIRTIO_BALLOON(obj
);
887 object_property_add(obj
, "guest-stats", "guest statistics",
888 balloon_stats_get_all
, NULL
, NULL
, s
, NULL
);
890 object_property_add(obj
, "guest-stats-polling-interval", "int",
891 balloon_stats_get_poll_interval
,
892 balloon_stats_set_poll_interval
,
896 static const VMStateDescription vmstate_virtio_balloon
= {
897 .name
= "virtio-balloon",
898 .minimum_version_id
= 1,
900 .fields
= (VMStateField
[]) {
901 VMSTATE_VIRTIO_DEVICE
,
902 VMSTATE_END_OF_LIST()
906 static Property virtio_balloon_properties
[] = {
907 DEFINE_PROP_BIT("deflate-on-oom", VirtIOBalloon
, host_features
,
908 VIRTIO_BALLOON_F_DEFLATE_ON_OOM
, false),
909 DEFINE_PROP_BIT("free-page-hint", VirtIOBalloon
, host_features
,
910 VIRTIO_BALLOON_F_FREE_PAGE_HINT
, false),
911 /* QEMU 4.0 accidentally changed the config size even when free-page-hint
912 * is disabled, resulting in QEMU 3.1 migration incompatibility. This
913 * property retains this quirk for QEMU 4.1 machine types.
915 DEFINE_PROP_BOOL("qemu-4-0-config-size", VirtIOBalloon
,
916 qemu_4_0_config_size
, false),
917 DEFINE_PROP_LINK("iothread", VirtIOBalloon
, iothread
, TYPE_IOTHREAD
,
919 DEFINE_PROP_END_OF_LIST(),
922 static void virtio_balloon_class_init(ObjectClass
*klass
, void *data
)
924 DeviceClass
*dc
= DEVICE_CLASS(klass
);
925 VirtioDeviceClass
*vdc
= VIRTIO_DEVICE_CLASS(klass
);
927 dc
->props
= virtio_balloon_properties
;
928 dc
->vmsd
= &vmstate_virtio_balloon
;
929 set_bit(DEVICE_CATEGORY_MISC
, dc
->categories
);
930 vdc
->realize
= virtio_balloon_device_realize
;
931 vdc
->unrealize
= virtio_balloon_device_unrealize
;
932 vdc
->reset
= virtio_balloon_device_reset
;
933 vdc
->get_config
= virtio_balloon_get_config
;
934 vdc
->set_config
= virtio_balloon_set_config
;
935 vdc
->get_features
= virtio_balloon_get_features
;
936 vdc
->set_status
= virtio_balloon_set_status
;
937 vdc
->vmsd
= &vmstate_virtio_balloon_device
;
940 static const TypeInfo virtio_balloon_info
= {
941 .name
= TYPE_VIRTIO_BALLOON
,
942 .parent
= TYPE_VIRTIO_DEVICE
,
943 .instance_size
= sizeof(VirtIOBalloon
),
944 .instance_init
= virtio_balloon_instance_init
,
945 .class_init
= virtio_balloon_class_init
,
948 static void virtio_register_types(void)
950 type_register_static(&virtio_balloon_info
);
953 type_init(virtio_register_types
)