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 "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"
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
{
40 unsigned long *bitmap
;
41 } PartiallyBalloonedPage
;
43 static void virtio_balloon_pbp_free(PartiallyBalloonedPage
*pbp
)
52 static void virtio_balloon_pbp_alloc(PartiallyBalloonedPage
*pbp
,
56 pbp
->base_gpa
= base_gpa
;
57 pbp
->bitmap
= bitmap_new(subpages
);
60 static bool virtio_balloon_pbp_matches(PartiallyBalloonedPage
*pbp
,
63 return pbp
->base_gpa
== base_gpa
;
66 static bool virtio_balloon_inhibited(void)
68 /* Postcopy cannot deal with concurrent discards, so it's special. */
69 return ram_block_discard_is_disabled() || migration_in_incoming_postcopy();
72 static void balloon_inflate_page(VirtIOBalloon
*balloon
,
73 MemoryRegion
*mr
, hwaddr mr_offset
,
74 PartiallyBalloonedPage
*pbp
)
76 void *addr
= memory_region_get_ram_ptr(mr
) + mr_offset
;
77 ram_addr_t rb_offset
, rb_aligned_offset
, base_gpa
;
82 /* XXX is there a better way to get to the RAMBlock than via a
84 rb
= qemu_ram_block_from_host(addr
, false, &rb_offset
);
85 rb_page_size
= qemu_ram_pagesize(rb
);
87 if (rb_page_size
== BALLOON_PAGE_SIZE
) {
90 ram_block_discard_range(rb
, rb_offset
, rb_page_size
);
91 /* We ignore errors from ram_block_discard_range(), because it
92 * has already reported them, and failing to discard a balloon
93 * page is not fatal */
99 * We've put a piece of a larger host page into the balloon - we
100 * need to keep track until we have a whole host page to
104 "Balloon used with backing page size > 4kiB, this may not be reliable");
106 rb_aligned_offset
= QEMU_ALIGN_DOWN(rb_offset
, rb_page_size
);
107 subpages
= rb_page_size
/ BALLOON_PAGE_SIZE
;
108 base_gpa
= memory_region_get_ram_addr(mr
) + mr_offset
-
109 (rb_offset
- rb_aligned_offset
);
111 if (pbp
->bitmap
&& !virtio_balloon_pbp_matches(pbp
, base_gpa
)) {
112 /* We've partially ballooned part of a host page, but now
113 * we're trying to balloon part of a different one. Too hard,
114 * give up on the old partial page */
115 virtio_balloon_pbp_free(pbp
);
119 virtio_balloon_pbp_alloc(pbp
, base_gpa
, subpages
);
122 set_bit((rb_offset
- rb_aligned_offset
) / BALLOON_PAGE_SIZE
,
125 if (bitmap_full(pbp
->bitmap
, subpages
)) {
126 /* We've accumulated a full host page, we can actually discard
129 ram_block_discard_range(rb
, rb_aligned_offset
, rb_page_size
);
130 /* We ignore errors from ram_block_discard_range(), because it
131 * has already reported them, and failing to discard a balloon
132 * page is not fatal */
133 virtio_balloon_pbp_free(pbp
);
137 static void balloon_deflate_page(VirtIOBalloon
*balloon
,
138 MemoryRegion
*mr
, hwaddr mr_offset
)
140 void *addr
= memory_region_get_ram_ptr(mr
) + mr_offset
;
141 ram_addr_t rb_offset
;
147 /* XXX is there a better way to get to the RAMBlock than via a
149 rb
= qemu_ram_block_from_host(addr
, false, &rb_offset
);
150 rb_page_size
= qemu_ram_pagesize(rb
);
152 host_addr
= (void *)((uintptr_t)addr
& ~(rb_page_size
- 1));
154 /* When a page is deflated, we hint the whole host page it lives
155 * on, since we can't do anything smaller */
156 ret
= qemu_madvise(host_addr
, rb_page_size
, QEMU_MADV_WILLNEED
);
158 warn_report("Couldn't MADV_WILLNEED on balloon deflate: %s",
160 /* Otherwise ignore, failing to page hint shouldn't be fatal */
164 static const char *balloon_stat_names
[] = {
165 [VIRTIO_BALLOON_S_SWAP_IN
] = "stat-swap-in",
166 [VIRTIO_BALLOON_S_SWAP_OUT
] = "stat-swap-out",
167 [VIRTIO_BALLOON_S_MAJFLT
] = "stat-major-faults",
168 [VIRTIO_BALLOON_S_MINFLT
] = "stat-minor-faults",
169 [VIRTIO_BALLOON_S_MEMFREE
] = "stat-free-memory",
170 [VIRTIO_BALLOON_S_MEMTOT
] = "stat-total-memory",
171 [VIRTIO_BALLOON_S_AVAIL
] = "stat-available-memory",
172 [VIRTIO_BALLOON_S_CACHES
] = "stat-disk-caches",
173 [VIRTIO_BALLOON_S_HTLB_PGALLOC
] = "stat-htlb-pgalloc",
174 [VIRTIO_BALLOON_S_HTLB_PGFAIL
] = "stat-htlb-pgfail",
175 [VIRTIO_BALLOON_S_NR
] = NULL
179 * reset_stats - Mark all items in the stats array as unset
181 * This function needs to be called at device initialization and before
182 * updating to a set of newly-generated stats. This will ensure that no
183 * stale values stick around in case the guest reports a subset of the supported
186 static inline void reset_stats(VirtIOBalloon
*dev
)
189 for (i
= 0; i
< VIRTIO_BALLOON_S_NR
; dev
->stats
[i
++] = -1);
192 static bool balloon_stats_supported(const VirtIOBalloon
*s
)
194 VirtIODevice
*vdev
= VIRTIO_DEVICE(s
);
195 return virtio_vdev_has_feature(vdev
, VIRTIO_BALLOON_F_STATS_VQ
);
198 static bool balloon_stats_enabled(const VirtIOBalloon
*s
)
200 return s
->stats_poll_interval
> 0;
203 static void balloon_stats_destroy_timer(VirtIOBalloon
*s
)
205 if (balloon_stats_enabled(s
)) {
206 timer_del(s
->stats_timer
);
207 timer_free(s
->stats_timer
);
208 s
->stats_timer
= NULL
;
209 s
->stats_poll_interval
= 0;
213 static void balloon_stats_change_timer(VirtIOBalloon
*s
, int64_t secs
)
215 timer_mod(s
->stats_timer
, qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL
) + secs
* 1000);
218 static void balloon_stats_poll_cb(void *opaque
)
220 VirtIOBalloon
*s
= opaque
;
221 VirtIODevice
*vdev
= VIRTIO_DEVICE(s
);
223 if (s
->stats_vq_elem
== NULL
|| !balloon_stats_supported(s
)) {
225 balloon_stats_change_timer(s
, s
->stats_poll_interval
);
229 virtqueue_push(s
->svq
, s
->stats_vq_elem
, s
->stats_vq_offset
);
230 virtio_notify(vdev
, s
->svq
);
231 g_free(s
->stats_vq_elem
);
232 s
->stats_vq_elem
= NULL
;
235 static void balloon_stats_get_all(Object
*obj
, Visitor
*v
, const char *name
,
236 void *opaque
, Error
**errp
)
239 VirtIOBalloon
*s
= opaque
;
242 visit_start_struct(v
, name
, NULL
, 0, &err
);
246 visit_type_int(v
, "last-update", &s
->stats_last_update
, &err
);
251 visit_start_struct(v
, "stats", NULL
, 0, &err
);
255 for (i
= 0; i
< VIRTIO_BALLOON_S_NR
; i
++) {
256 visit_type_uint64(v
, balloon_stat_names
[i
], &s
->stats
[i
], &err
);
261 visit_check_struct(v
, &err
);
263 visit_end_struct(v
, NULL
);
266 visit_check_struct(v
, &err
);
269 visit_end_struct(v
, NULL
);
271 error_propagate(errp
, err
);
274 static void balloon_stats_get_poll_interval(Object
*obj
, Visitor
*v
,
275 const char *name
, void *opaque
,
278 VirtIOBalloon
*s
= opaque
;
279 visit_type_int(v
, name
, &s
->stats_poll_interval
, errp
);
282 static void balloon_stats_set_poll_interval(Object
*obj
, Visitor
*v
,
283 const char *name
, void *opaque
,
286 VirtIOBalloon
*s
= opaque
;
287 Error
*local_err
= NULL
;
290 visit_type_int(v
, name
, &value
, &local_err
);
292 error_propagate(errp
, local_err
);
297 error_setg(errp
, "timer value must be greater than zero");
301 if (value
> UINT32_MAX
) {
302 error_setg(errp
, "timer value is too big");
306 if (value
== s
->stats_poll_interval
) {
311 /* timer=0 disables the timer */
312 balloon_stats_destroy_timer(s
);
316 if (balloon_stats_enabled(s
)) {
317 /* timer interval change */
318 s
->stats_poll_interval
= value
;
319 balloon_stats_change_timer(s
, value
);
323 /* create a new timer */
324 g_assert(s
->stats_timer
== NULL
);
325 s
->stats_timer
= timer_new_ms(QEMU_CLOCK_VIRTUAL
, balloon_stats_poll_cb
, s
);
326 s
->stats_poll_interval
= value
;
327 balloon_stats_change_timer(s
, 0);
330 static void virtio_balloon_handle_report(VirtIODevice
*vdev
, VirtQueue
*vq
)
332 VirtIOBalloon
*dev
= VIRTIO_BALLOON(vdev
);
333 VirtQueueElement
*elem
;
335 while ((elem
= virtqueue_pop(vq
, sizeof(VirtQueueElement
)))) {
339 * When we discard the page it has the effect of removing the page
340 * from the hypervisor itself and causing it to be zeroed when it
341 * is returned to us. So we must not discard the page if it is
342 * accessible by another device or process, or if the guest is
343 * expecting it to retain a non-zero value.
345 if (virtio_balloon_inhibited() || dev
->poison_val
) {
349 for (i
= 0; i
< elem
->in_num
; i
++) {
350 void *addr
= elem
->in_sg
[i
].iov_base
;
351 size_t size
= elem
->in_sg
[i
].iov_len
;
352 ram_addr_t ram_offset
;
356 * There is no need to check the memory section to see if
357 * it is ram/readonly/romd like there is for handle_output
358 * below. If the region is not meant to be written to then
359 * address_space_map will have allocated a bounce buffer
360 * and it will be freed in address_space_unmap and trigger
361 * and unassigned_mem_write before failing to copy over the
362 * buffer. If more than one bad descriptor is provided it
363 * will return NULL after the first bounce buffer and fail
364 * to map any resources.
366 rb
= qemu_ram_block_from_host(addr
, false, &ram_offset
);
368 trace_virtio_balloon_bad_addr(elem
->in_addr
[i
]);
373 * For now we will simply ignore unaligned memory regions, or
374 * regions that overrun the end of the RAMBlock.
376 if (!QEMU_IS_ALIGNED(ram_offset
| size
, qemu_ram_pagesize(rb
)) ||
377 (ram_offset
+ size
) > qemu_ram_get_used_length(rb
)) {
381 ram_block_discard_range(rb
, ram_offset
, size
);
385 virtqueue_push(vq
, elem
, 0);
386 virtio_notify(vdev
, vq
);
391 static void virtio_balloon_handle_output(VirtIODevice
*vdev
, VirtQueue
*vq
)
393 VirtIOBalloon
*s
= VIRTIO_BALLOON(vdev
);
394 VirtQueueElement
*elem
;
395 MemoryRegionSection section
;
398 PartiallyBalloonedPage pbp
= {};
402 elem
= virtqueue_pop(vq
, sizeof(VirtQueueElement
));
407 while (iov_to_buf(elem
->out_sg
, elem
->out_num
, offset
, &pfn
, 4) == 4) {
408 unsigned int p
= virtio_ldl_p(vdev
, &pfn
);
411 pa
= (hwaddr
) p
<< VIRTIO_BALLOON_PFN_SHIFT
;
414 section
= memory_region_find(get_system_memory(), pa
,
417 trace_virtio_balloon_bad_addr(pa
);
420 if (!memory_region_is_ram(section
.mr
) ||
421 memory_region_is_rom(section
.mr
) ||
422 memory_region_is_romd(section
.mr
)) {
423 trace_virtio_balloon_bad_addr(pa
);
424 memory_region_unref(section
.mr
);
428 trace_virtio_balloon_handle_output(memory_region_name(section
.mr
),
430 if (!virtio_balloon_inhibited()) {
432 balloon_inflate_page(s
, section
.mr
,
433 section
.offset_within_region
, &pbp
);
434 } else if (vq
== s
->dvq
) {
435 balloon_deflate_page(s
, section
.mr
, section
.offset_within_region
);
437 g_assert_not_reached();
440 memory_region_unref(section
.mr
);
443 virtqueue_push(vq
, elem
, offset
);
444 virtio_notify(vdev
, vq
);
446 virtio_balloon_pbp_free(&pbp
);
450 static void virtio_balloon_receive_stats(VirtIODevice
*vdev
, VirtQueue
*vq
)
452 VirtIOBalloon
*s
= VIRTIO_BALLOON(vdev
);
453 VirtQueueElement
*elem
;
454 VirtIOBalloonStat stat
;
458 elem
= virtqueue_pop(vq
, sizeof(VirtQueueElement
));
463 if (s
->stats_vq_elem
!= NULL
) {
464 /* This should never happen if the driver follows the spec. */
465 virtqueue_push(vq
, s
->stats_vq_elem
, 0);
466 virtio_notify(vdev
, vq
);
467 g_free(s
->stats_vq_elem
);
470 s
->stats_vq_elem
= elem
;
472 /* Initialize the stats to get rid of any stale values. This is only
473 * needed to handle the case where a guest supports fewer stats than it
474 * used to (ie. it has booted into an old kernel).
478 while (iov_to_buf(elem
->out_sg
, elem
->out_num
, offset
, &stat
, sizeof(stat
))
480 uint16_t tag
= virtio_tswap16(vdev
, stat
.tag
);
481 uint64_t val
= virtio_tswap64(vdev
, stat
.val
);
483 offset
+= sizeof(stat
);
484 if (tag
< VIRTIO_BALLOON_S_NR
)
487 s
->stats_vq_offset
= offset
;
489 if (qemu_gettimeofday(&tv
) < 0) {
490 warn_report("%s: failed to get time of day", __func__
);
494 s
->stats_last_update
= tv
.tv_sec
;
497 if (balloon_stats_enabled(s
)) {
498 balloon_stats_change_timer(s
, s
->stats_poll_interval
);
502 static void virtio_balloon_handle_free_page_vq(VirtIODevice
*vdev
,
505 VirtIOBalloon
*s
= VIRTIO_BALLOON(vdev
);
506 qemu_bh_schedule(s
->free_page_bh
);
509 static bool get_free_page_hints(VirtIOBalloon
*dev
)
511 VirtQueueElement
*elem
;
512 VirtIODevice
*vdev
= VIRTIO_DEVICE(dev
);
513 VirtQueue
*vq
= dev
->free_page_vq
;
516 while (dev
->block_iothread
) {
517 qemu_cond_wait(&dev
->free_page_cond
, &dev
->free_page_lock
);
520 elem
= virtqueue_pop(vq
, sizeof(VirtQueueElement
));
527 size_t size
= iov_to_buf(elem
->out_sg
, elem
->out_num
, 0,
530 virtio_tswap32s(vdev
, &id
);
531 if (unlikely(size
!= sizeof(id
))) {
532 virtio_error(vdev
, "received an incorrect cmd id");
536 if (id
== dev
->free_page_report_cmd_id
) {
537 dev
->free_page_report_status
= FREE_PAGE_REPORT_S_START
;
540 * Stop the optimization only when it has started. This
541 * avoids a stale stop sign for the previous command.
543 if (dev
->free_page_report_status
== FREE_PAGE_REPORT_S_START
) {
544 dev
->free_page_report_status
= FREE_PAGE_REPORT_S_STOP
;
550 if (dev
->free_page_report_status
== FREE_PAGE_REPORT_S_START
) {
551 qemu_guest_free_page_hint(elem
->in_sg
[0].iov_base
,
552 elem
->in_sg
[0].iov_len
);
557 virtqueue_push(vq
, elem
, 1);
562 static void virtio_ballloon_get_free_page_hints(void *opaque
)
564 VirtIOBalloon
*dev
= opaque
;
565 VirtIODevice
*vdev
= VIRTIO_DEVICE(dev
);
566 VirtQueue
*vq
= dev
->free_page_vq
;
567 bool continue_to_get_hints
;
570 qemu_mutex_lock(&dev
->free_page_lock
);
571 virtio_queue_set_notification(vq
, 0);
572 continue_to_get_hints
= get_free_page_hints(dev
);
573 qemu_mutex_unlock(&dev
->free_page_lock
);
574 virtio_notify(vdev
, vq
);
576 * Start to poll the vq once the reporting started. Otherwise, continue
577 * only when there are entries on the vq, which need to be given back.
579 } while (continue_to_get_hints
||
580 dev
->free_page_report_status
== FREE_PAGE_REPORT_S_START
);
581 virtio_queue_set_notification(vq
, 1);
584 static bool virtio_balloon_free_page_support(void *opaque
)
586 VirtIOBalloon
*s
= opaque
;
587 VirtIODevice
*vdev
= VIRTIO_DEVICE(s
);
589 return virtio_vdev_has_feature(vdev
, VIRTIO_BALLOON_F_FREE_PAGE_HINT
);
592 static void virtio_balloon_free_page_start(VirtIOBalloon
*s
)
594 VirtIODevice
*vdev
= VIRTIO_DEVICE(s
);
596 /* For the stop and copy phase, we don't need to start the optimization */
597 if (!vdev
->vm_running
) {
601 if (s
->free_page_report_cmd_id
== UINT_MAX
) {
602 s
->free_page_report_cmd_id
=
603 VIRTIO_BALLOON_FREE_PAGE_REPORT_CMD_ID_MIN
;
605 s
->free_page_report_cmd_id
++;
608 s
->free_page_report_status
= FREE_PAGE_REPORT_S_REQUESTED
;
609 virtio_notify_config(vdev
);
612 static void virtio_balloon_free_page_stop(VirtIOBalloon
*s
)
614 VirtIODevice
*vdev
= VIRTIO_DEVICE(s
);
616 if (s
->free_page_report_status
!= FREE_PAGE_REPORT_S_STOP
) {
618 * The lock also guarantees us that the
619 * virtio_ballloon_get_free_page_hints exits after the
620 * free_page_report_status is set to S_STOP.
622 qemu_mutex_lock(&s
->free_page_lock
);
624 * The guest hasn't done the reporting, so host sends a notification
625 * to the guest to actively stop the reporting.
627 s
->free_page_report_status
= FREE_PAGE_REPORT_S_STOP
;
628 qemu_mutex_unlock(&s
->free_page_lock
);
629 virtio_notify_config(vdev
);
633 static void virtio_balloon_free_page_done(VirtIOBalloon
*s
)
635 VirtIODevice
*vdev
= VIRTIO_DEVICE(s
);
637 if (s
->free_page_report_status
!= FREE_PAGE_REPORT_S_DONE
) {
638 /* See virtio_balloon_free_page_stop() */
639 qemu_mutex_lock(&s
->free_page_lock
);
640 s
->free_page_report_status
= FREE_PAGE_REPORT_S_DONE
;
641 qemu_mutex_unlock(&s
->free_page_lock
);
642 virtio_notify_config(vdev
);
647 virtio_balloon_free_page_report_notify(NotifierWithReturn
*n
, void *data
)
649 VirtIOBalloon
*dev
= container_of(n
, VirtIOBalloon
,
650 free_page_report_notify
);
651 VirtIODevice
*vdev
= VIRTIO_DEVICE(dev
);
652 PrecopyNotifyData
*pnd
= data
;
654 if (!virtio_balloon_free_page_support(dev
)) {
656 * This is an optimization provided to migration, so just return 0 to
657 * have the normal migration process not affected when this feature is
663 switch (pnd
->reason
) {
664 case PRECOPY_NOTIFY_SETUP
:
665 precopy_enable_free_page_optimization();
667 case PRECOPY_NOTIFY_BEFORE_BITMAP_SYNC
:
668 virtio_balloon_free_page_stop(dev
);
670 case PRECOPY_NOTIFY_AFTER_BITMAP_SYNC
:
671 if (vdev
->vm_running
) {
672 virtio_balloon_free_page_start(dev
);
676 * Set S_DONE before migrating the vmstate, so the guest will reuse
677 * all hinted pages once running on the destination. Fall through.
679 case PRECOPY_NOTIFY_CLEANUP
:
681 * Especially, if something goes wrong during precopy or if migration
682 * is canceled, we have to properly communicate S_DONE to the VM.
684 virtio_balloon_free_page_done(dev
);
686 case PRECOPY_NOTIFY_COMPLETE
:
689 virtio_error(vdev
, "%s: %d reason unknown", __func__
, pnd
->reason
);
695 static size_t virtio_balloon_config_size(VirtIOBalloon
*s
)
697 uint64_t features
= s
->host_features
;
699 if (s
->qemu_4_0_config_size
) {
700 return sizeof(struct virtio_balloon_config
);
702 if (virtio_has_feature(features
, VIRTIO_BALLOON_F_PAGE_POISON
)) {
703 return sizeof(struct virtio_balloon_config
);
705 if (virtio_has_feature(features
, VIRTIO_BALLOON_F_FREE_PAGE_HINT
)) {
706 return offsetof(struct virtio_balloon_config
, poison_val
);
708 return offsetof(struct virtio_balloon_config
, free_page_report_cmd_id
);
711 static void virtio_balloon_get_config(VirtIODevice
*vdev
, uint8_t *config_data
)
713 VirtIOBalloon
*dev
= VIRTIO_BALLOON(vdev
);
714 struct virtio_balloon_config config
= {};
716 config
.num_pages
= cpu_to_le32(dev
->num_pages
);
717 config
.actual
= cpu_to_le32(dev
->actual
);
718 config
.poison_val
= cpu_to_le32(dev
->poison_val
);
720 if (dev
->free_page_report_status
== FREE_PAGE_REPORT_S_REQUESTED
) {
721 config
.free_page_report_cmd_id
=
722 cpu_to_le32(dev
->free_page_report_cmd_id
);
723 } else if (dev
->free_page_report_status
== FREE_PAGE_REPORT_S_STOP
) {
724 config
.free_page_report_cmd_id
=
725 cpu_to_le32(VIRTIO_BALLOON_CMD_ID_STOP
);
726 } else if (dev
->free_page_report_status
== FREE_PAGE_REPORT_S_DONE
) {
727 config
.free_page_report_cmd_id
=
728 cpu_to_le32(VIRTIO_BALLOON_CMD_ID_DONE
);
731 trace_virtio_balloon_get_config(config
.num_pages
, config
.actual
);
732 memcpy(config_data
, &config
, virtio_balloon_config_size(dev
));
735 static int build_dimm_list(Object
*obj
, void *opaque
)
737 GSList
**list
= opaque
;
739 if (object_dynamic_cast(obj
, TYPE_PC_DIMM
)) {
740 DeviceState
*dev
= DEVICE(obj
);
741 if (dev
->realized
) { /* only realized DIMMs matter */
742 *list
= g_slist_prepend(*list
, dev
);
746 object_child_foreach(obj
, build_dimm_list
, opaque
);
750 static ram_addr_t
get_current_ram_size(void)
752 GSList
*list
= NULL
, *item
;
753 ram_addr_t size
= ram_size
;
755 build_dimm_list(qdev_get_machine(), &list
);
756 for (item
= list
; item
; item
= g_slist_next(item
)) {
757 Object
*obj
= OBJECT(item
->data
);
758 if (!strcmp(object_get_typename(obj
), TYPE_PC_DIMM
)) {
759 size
+= object_property_get_int(obj
, PC_DIMM_SIZE_PROP
,
768 static bool virtio_balloon_page_poison_support(void *opaque
)
770 VirtIOBalloon
*s
= opaque
;
771 VirtIODevice
*vdev
= VIRTIO_DEVICE(s
);
773 return virtio_vdev_has_feature(vdev
, VIRTIO_BALLOON_F_PAGE_POISON
);
776 static void virtio_balloon_set_config(VirtIODevice
*vdev
,
777 const uint8_t *config_data
)
779 VirtIOBalloon
*dev
= VIRTIO_BALLOON(vdev
);
780 struct virtio_balloon_config config
;
781 uint32_t oldactual
= dev
->actual
;
782 ram_addr_t vm_ram_size
= get_current_ram_size();
784 memcpy(&config
, config_data
, virtio_balloon_config_size(dev
));
785 dev
->actual
= le32_to_cpu(config
.actual
);
786 if (dev
->actual
!= oldactual
) {
787 qapi_event_send_balloon_change(vm_ram_size
-
788 ((ram_addr_t
) dev
->actual
<< VIRTIO_BALLOON_PFN_SHIFT
));
791 if (virtio_balloon_page_poison_support(dev
)) {
792 dev
->poison_val
= le32_to_cpu(config
.poison_val
);
794 trace_virtio_balloon_set_config(dev
->actual
, oldactual
);
797 static uint64_t virtio_balloon_get_features(VirtIODevice
*vdev
, uint64_t f
,
800 VirtIOBalloon
*dev
= VIRTIO_BALLOON(vdev
);
801 f
|= dev
->host_features
;
802 virtio_add_feature(&f
, VIRTIO_BALLOON_F_STATS_VQ
);
807 static void virtio_balloon_stat(void *opaque
, BalloonInfo
*info
)
809 VirtIOBalloon
*dev
= opaque
;
810 info
->actual
= get_current_ram_size() - ((uint64_t) dev
->actual
<<
811 VIRTIO_BALLOON_PFN_SHIFT
);
814 static void virtio_balloon_to_target(void *opaque
, ram_addr_t target
)
816 VirtIOBalloon
*dev
= VIRTIO_BALLOON(opaque
);
817 VirtIODevice
*vdev
= VIRTIO_DEVICE(dev
);
818 ram_addr_t vm_ram_size
= get_current_ram_size();
820 if (target
> vm_ram_size
) {
821 target
= vm_ram_size
;
824 dev
->num_pages
= (vm_ram_size
- target
) >> VIRTIO_BALLOON_PFN_SHIFT
;
825 virtio_notify_config(vdev
);
827 trace_virtio_balloon_to_target(target
, dev
->num_pages
);
830 static int virtio_balloon_post_load_device(void *opaque
, int version_id
)
832 VirtIOBalloon
*s
= VIRTIO_BALLOON(opaque
);
834 if (balloon_stats_enabled(s
)) {
835 balloon_stats_change_timer(s
, s
->stats_poll_interval
);
840 static const VMStateDescription vmstate_virtio_balloon_free_page_report
= {
841 .name
= "virtio-balloon-device/free-page-report",
843 .minimum_version_id
= 1,
844 .needed
= virtio_balloon_free_page_support
,
845 .fields
= (VMStateField
[]) {
846 VMSTATE_UINT32(free_page_report_cmd_id
, VirtIOBalloon
),
847 VMSTATE_UINT32(free_page_report_status
, VirtIOBalloon
),
848 VMSTATE_END_OF_LIST()
852 static const VMStateDescription vmstate_virtio_balloon_page_poison
= {
853 .name
= "vitio-balloon-device/page-poison",
855 .minimum_version_id
= 1,
856 .needed
= virtio_balloon_page_poison_support
,
857 .fields
= (VMStateField
[]) {
858 VMSTATE_UINT32(poison_val
, VirtIOBalloon
),
859 VMSTATE_END_OF_LIST()
863 static const VMStateDescription vmstate_virtio_balloon_device
= {
864 .name
= "virtio-balloon-device",
866 .minimum_version_id
= 1,
867 .post_load
= virtio_balloon_post_load_device
,
868 .fields
= (VMStateField
[]) {
869 VMSTATE_UINT32(num_pages
, VirtIOBalloon
),
870 VMSTATE_UINT32(actual
, VirtIOBalloon
),
871 VMSTATE_END_OF_LIST()
873 .subsections
= (const VMStateDescription
* []) {
874 &vmstate_virtio_balloon_free_page_report
,
875 &vmstate_virtio_balloon_page_poison
,
880 static void virtio_balloon_device_realize(DeviceState
*dev
, Error
**errp
)
882 VirtIODevice
*vdev
= VIRTIO_DEVICE(dev
);
883 VirtIOBalloon
*s
= VIRTIO_BALLOON(dev
);
886 virtio_init(vdev
, "virtio-balloon", VIRTIO_ID_BALLOON
,
887 virtio_balloon_config_size(s
));
889 ret
= qemu_add_balloon_handler(virtio_balloon_to_target
,
890 virtio_balloon_stat
, s
);
893 error_setg(errp
, "Only one balloon device is supported");
894 virtio_cleanup(vdev
);
898 if (virtio_has_feature(s
->host_features
, VIRTIO_BALLOON_F_FREE_PAGE_HINT
) &&
900 error_setg(errp
, "'free-page-hint' requires 'iothread' to be set");
901 virtio_cleanup(vdev
);
905 s
->ivq
= virtio_add_queue(vdev
, 128, virtio_balloon_handle_output
);
906 s
->dvq
= virtio_add_queue(vdev
, 128, virtio_balloon_handle_output
);
907 s
->svq
= virtio_add_queue(vdev
, 128, virtio_balloon_receive_stats
);
909 if (virtio_has_feature(s
->host_features
,
910 VIRTIO_BALLOON_F_FREE_PAGE_HINT
)) {
911 s
->free_page_vq
= virtio_add_queue(vdev
, VIRTQUEUE_MAX_SIZE
,
912 virtio_balloon_handle_free_page_vq
);
913 precopy_add_notifier(&s
->free_page_report_notify
);
915 object_ref(OBJECT(s
->iothread
));
916 s
->free_page_bh
= aio_bh_new(iothread_get_aio_context(s
->iothread
),
917 virtio_ballloon_get_free_page_hints
, s
);
920 if (virtio_has_feature(s
->host_features
, VIRTIO_BALLOON_F_REPORTING
)) {
921 s
->reporting_vq
= virtio_add_queue(vdev
, 32,
922 virtio_balloon_handle_report
);
928 static void virtio_balloon_device_unrealize(DeviceState
*dev
)
930 VirtIODevice
*vdev
= VIRTIO_DEVICE(dev
);
931 VirtIOBalloon
*s
= VIRTIO_BALLOON(dev
);
933 if (s
->free_page_bh
) {
934 qemu_bh_delete(s
->free_page_bh
);
935 object_unref(OBJECT(s
->iothread
));
936 virtio_balloon_free_page_stop(s
);
937 precopy_remove_notifier(&s
->free_page_report_notify
);
939 balloon_stats_destroy_timer(s
);
940 qemu_remove_balloon_handler(s
);
942 virtio_delete_queue(s
->ivq
);
943 virtio_delete_queue(s
->dvq
);
944 virtio_delete_queue(s
->svq
);
945 if (s
->free_page_vq
) {
946 virtio_delete_queue(s
->free_page_vq
);
948 if (s
->reporting_vq
) {
949 virtio_delete_queue(s
->reporting_vq
);
951 virtio_cleanup(vdev
);
954 static void virtio_balloon_device_reset(VirtIODevice
*vdev
)
956 VirtIOBalloon
*s
= VIRTIO_BALLOON(vdev
);
958 if (virtio_balloon_free_page_support(s
)) {
959 virtio_balloon_free_page_stop(s
);
962 if (s
->stats_vq_elem
!= NULL
) {
963 virtqueue_unpop(s
->svq
, s
->stats_vq_elem
, 0);
964 g_free(s
->stats_vq_elem
);
965 s
->stats_vq_elem
= NULL
;
971 static void virtio_balloon_set_status(VirtIODevice
*vdev
, uint8_t status
)
973 VirtIOBalloon
*s
= VIRTIO_BALLOON(vdev
);
975 if (!s
->stats_vq_elem
&& vdev
->vm_running
&&
976 (status
& VIRTIO_CONFIG_S_DRIVER_OK
) && virtqueue_rewind(s
->svq
, 1)) {
977 /* poll stats queue for the element we have discarded when the VM
979 virtio_balloon_receive_stats(vdev
, s
->svq
);
982 if (virtio_balloon_free_page_support(s
)) {
984 * The VM is woken up and the iothread was blocked, so signal it to
987 if (vdev
->vm_running
&& s
->block_iothread
) {
988 qemu_mutex_lock(&s
->free_page_lock
);
989 s
->block_iothread
= false;
990 qemu_cond_signal(&s
->free_page_cond
);
991 qemu_mutex_unlock(&s
->free_page_lock
);
994 /* The VM is stopped, block the iothread. */
995 if (!vdev
->vm_running
) {
996 qemu_mutex_lock(&s
->free_page_lock
);
997 s
->block_iothread
= true;
998 qemu_mutex_unlock(&s
->free_page_lock
);
1003 static void virtio_balloon_instance_init(Object
*obj
)
1005 VirtIOBalloon
*s
= VIRTIO_BALLOON(obj
);
1007 qemu_mutex_init(&s
->free_page_lock
);
1008 qemu_cond_init(&s
->free_page_cond
);
1009 s
->free_page_report_cmd_id
= VIRTIO_BALLOON_FREE_PAGE_REPORT_CMD_ID_MIN
;
1010 s
->free_page_report_notify
.notify
= virtio_balloon_free_page_report_notify
;
1012 object_property_add(obj
, "guest-stats", "guest statistics",
1013 balloon_stats_get_all
, NULL
, NULL
, s
);
1015 object_property_add(obj
, "guest-stats-polling-interval", "int",
1016 balloon_stats_get_poll_interval
,
1017 balloon_stats_set_poll_interval
,
1021 static const VMStateDescription vmstate_virtio_balloon
= {
1022 .name
= "virtio-balloon",
1023 .minimum_version_id
= 1,
1025 .fields
= (VMStateField
[]) {
1026 VMSTATE_VIRTIO_DEVICE
,
1027 VMSTATE_END_OF_LIST()
1031 static Property virtio_balloon_properties
[] = {
1032 DEFINE_PROP_BIT("deflate-on-oom", VirtIOBalloon
, host_features
,
1033 VIRTIO_BALLOON_F_DEFLATE_ON_OOM
, false),
1034 DEFINE_PROP_BIT("free-page-hint", VirtIOBalloon
, host_features
,
1035 VIRTIO_BALLOON_F_FREE_PAGE_HINT
, false),
1036 DEFINE_PROP_BIT("page-poison", VirtIOBalloon
, host_features
,
1037 VIRTIO_BALLOON_F_PAGE_POISON
, true),
1038 DEFINE_PROP_BIT("free-page-reporting", VirtIOBalloon
, host_features
,
1039 VIRTIO_BALLOON_F_REPORTING
, false),
1040 /* QEMU 4.0 accidentally changed the config size even when free-page-hint
1041 * is disabled, resulting in QEMU 3.1 migration incompatibility. This
1042 * property retains this quirk for QEMU 4.1 machine types.
1044 DEFINE_PROP_BOOL("qemu-4-0-config-size", VirtIOBalloon
,
1045 qemu_4_0_config_size
, false),
1046 DEFINE_PROP_LINK("iothread", VirtIOBalloon
, iothread
, TYPE_IOTHREAD
,
1048 DEFINE_PROP_END_OF_LIST(),
1051 static void virtio_balloon_class_init(ObjectClass
*klass
, void *data
)
1053 DeviceClass
*dc
= DEVICE_CLASS(klass
);
1054 VirtioDeviceClass
*vdc
= VIRTIO_DEVICE_CLASS(klass
);
1056 device_class_set_props(dc
, virtio_balloon_properties
);
1057 dc
->vmsd
= &vmstate_virtio_balloon
;
1058 set_bit(DEVICE_CATEGORY_MISC
, dc
->categories
);
1059 vdc
->realize
= virtio_balloon_device_realize
;
1060 vdc
->unrealize
= virtio_balloon_device_unrealize
;
1061 vdc
->reset
= virtio_balloon_device_reset
;
1062 vdc
->get_config
= virtio_balloon_get_config
;
1063 vdc
->set_config
= virtio_balloon_set_config
;
1064 vdc
->get_features
= virtio_balloon_get_features
;
1065 vdc
->set_status
= virtio_balloon_set_status
;
1066 vdc
->vmsd
= &vmstate_virtio_balloon_device
;
1069 static const TypeInfo virtio_balloon_info
= {
1070 .name
= TYPE_VIRTIO_BALLOON
,
1071 .parent
= TYPE_VIRTIO_DEVICE
,
1072 .instance_size
= sizeof(VirtIOBalloon
),
1073 .instance_init
= virtio_balloon_instance_init
,
1074 .class_init
= virtio_balloon_class_init
,
1077 static void virtio_register_types(void)
1079 type_register_static(&virtio_balloon_info
);
1082 type_init(virtio_register_types
)