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 if (!visit_start_struct(v
, name
, NULL
, 0, &err
)) {
245 if (!visit_type_int(v
, "last-update", &s
->stats_last_update
, &err
)) {
249 if (!visit_start_struct(v
, "stats", NULL
, 0, &err
)) {
252 for (i
= 0; i
< VIRTIO_BALLOON_S_NR
; i
++) {
253 if (!visit_type_uint64(v
, balloon_stat_names
[i
], &s
->stats
[i
], &err
)) {
257 visit_check_struct(v
, &err
);
259 visit_end_struct(v
, NULL
);
262 visit_check_struct(v
, &err
);
265 visit_end_struct(v
, NULL
);
267 error_propagate(errp
, err
);
270 static void balloon_stats_get_poll_interval(Object
*obj
, Visitor
*v
,
271 const char *name
, void *opaque
,
274 VirtIOBalloon
*s
= opaque
;
275 visit_type_int(v
, name
, &s
->stats_poll_interval
, errp
);
278 static void balloon_stats_set_poll_interval(Object
*obj
, Visitor
*v
,
279 const char *name
, void *opaque
,
282 VirtIOBalloon
*s
= opaque
;
283 Error
*local_err
= NULL
;
286 if (!visit_type_int(v
, name
, &value
, &local_err
)) {
287 error_propagate(errp
, local_err
);
292 error_setg(errp
, "timer value must be greater than zero");
296 if (value
> UINT32_MAX
) {
297 error_setg(errp
, "timer value is too big");
301 if (value
== s
->stats_poll_interval
) {
306 /* timer=0 disables the timer */
307 balloon_stats_destroy_timer(s
);
311 if (balloon_stats_enabled(s
)) {
312 /* timer interval change */
313 s
->stats_poll_interval
= value
;
314 balloon_stats_change_timer(s
, value
);
318 /* create a new timer */
319 g_assert(s
->stats_timer
== NULL
);
320 s
->stats_timer
= timer_new_ms(QEMU_CLOCK_VIRTUAL
, balloon_stats_poll_cb
, s
);
321 s
->stats_poll_interval
= value
;
322 balloon_stats_change_timer(s
, 0);
325 static void virtio_balloon_handle_report(VirtIODevice
*vdev
, VirtQueue
*vq
)
327 VirtIOBalloon
*dev
= VIRTIO_BALLOON(vdev
);
328 VirtQueueElement
*elem
;
330 while ((elem
= virtqueue_pop(vq
, sizeof(VirtQueueElement
)))) {
334 * When we discard the page it has the effect of removing the page
335 * from the hypervisor itself and causing it to be zeroed when it
336 * is returned to us. So we must not discard the page if it is
337 * accessible by another device or process, or if the guest is
338 * expecting it to retain a non-zero value.
340 if (virtio_balloon_inhibited() || dev
->poison_val
) {
344 for (i
= 0; i
< elem
->in_num
; i
++) {
345 void *addr
= elem
->in_sg
[i
].iov_base
;
346 size_t size
= elem
->in_sg
[i
].iov_len
;
347 ram_addr_t ram_offset
;
351 * There is no need to check the memory section to see if
352 * it is ram/readonly/romd like there is for handle_output
353 * below. If the region is not meant to be written to then
354 * address_space_map will have allocated a bounce buffer
355 * and it will be freed in address_space_unmap and trigger
356 * and unassigned_mem_write before failing to copy over the
357 * buffer. If more than one bad descriptor is provided it
358 * will return NULL after the first bounce buffer and fail
359 * to map any resources.
361 rb
= qemu_ram_block_from_host(addr
, false, &ram_offset
);
363 trace_virtio_balloon_bad_addr(elem
->in_addr
[i
]);
368 * For now we will simply ignore unaligned memory regions, or
369 * regions that overrun the end of the RAMBlock.
371 if (!QEMU_IS_ALIGNED(ram_offset
| size
, qemu_ram_pagesize(rb
)) ||
372 (ram_offset
+ size
) > qemu_ram_get_used_length(rb
)) {
376 ram_block_discard_range(rb
, ram_offset
, size
);
380 virtqueue_push(vq
, elem
, 0);
381 virtio_notify(vdev
, vq
);
386 static void virtio_balloon_handle_output(VirtIODevice
*vdev
, VirtQueue
*vq
)
388 VirtIOBalloon
*s
= VIRTIO_BALLOON(vdev
);
389 VirtQueueElement
*elem
;
390 MemoryRegionSection section
;
393 PartiallyBalloonedPage pbp
= {};
397 elem
= virtqueue_pop(vq
, sizeof(VirtQueueElement
));
402 while (iov_to_buf(elem
->out_sg
, elem
->out_num
, offset
, &pfn
, 4) == 4) {
403 unsigned int p
= virtio_ldl_p(vdev
, &pfn
);
406 pa
= (hwaddr
) p
<< VIRTIO_BALLOON_PFN_SHIFT
;
409 section
= memory_region_find(get_system_memory(), pa
,
412 trace_virtio_balloon_bad_addr(pa
);
415 if (!memory_region_is_ram(section
.mr
) ||
416 memory_region_is_rom(section
.mr
) ||
417 memory_region_is_romd(section
.mr
)) {
418 trace_virtio_balloon_bad_addr(pa
);
419 memory_region_unref(section
.mr
);
423 trace_virtio_balloon_handle_output(memory_region_name(section
.mr
),
425 if (!virtio_balloon_inhibited()) {
427 balloon_inflate_page(s
, section
.mr
,
428 section
.offset_within_region
, &pbp
);
429 } else if (vq
== s
->dvq
) {
430 balloon_deflate_page(s
, section
.mr
, section
.offset_within_region
);
432 g_assert_not_reached();
435 memory_region_unref(section
.mr
);
438 virtqueue_push(vq
, elem
, offset
);
439 virtio_notify(vdev
, vq
);
441 virtio_balloon_pbp_free(&pbp
);
445 static void virtio_balloon_receive_stats(VirtIODevice
*vdev
, VirtQueue
*vq
)
447 VirtIOBalloon
*s
= VIRTIO_BALLOON(vdev
);
448 VirtQueueElement
*elem
;
449 VirtIOBalloonStat stat
;
453 elem
= virtqueue_pop(vq
, sizeof(VirtQueueElement
));
458 if (s
->stats_vq_elem
!= NULL
) {
459 /* This should never happen if the driver follows the spec. */
460 virtqueue_push(vq
, s
->stats_vq_elem
, 0);
461 virtio_notify(vdev
, vq
);
462 g_free(s
->stats_vq_elem
);
465 s
->stats_vq_elem
= elem
;
467 /* Initialize the stats to get rid of any stale values. This is only
468 * needed to handle the case where a guest supports fewer stats than it
469 * used to (ie. it has booted into an old kernel).
473 while (iov_to_buf(elem
->out_sg
, elem
->out_num
, offset
, &stat
, sizeof(stat
))
475 uint16_t tag
= virtio_tswap16(vdev
, stat
.tag
);
476 uint64_t val
= virtio_tswap64(vdev
, stat
.val
);
478 offset
+= sizeof(stat
);
479 if (tag
< VIRTIO_BALLOON_S_NR
)
482 s
->stats_vq_offset
= offset
;
484 if (qemu_gettimeofday(&tv
) < 0) {
485 warn_report("%s: failed to get time of day", __func__
);
489 s
->stats_last_update
= tv
.tv_sec
;
492 if (balloon_stats_enabled(s
)) {
493 balloon_stats_change_timer(s
, s
->stats_poll_interval
);
497 static void virtio_balloon_handle_free_page_vq(VirtIODevice
*vdev
,
500 VirtIOBalloon
*s
= VIRTIO_BALLOON(vdev
);
501 qemu_bh_schedule(s
->free_page_bh
);
504 static bool get_free_page_hints(VirtIOBalloon
*dev
)
506 VirtQueueElement
*elem
;
507 VirtIODevice
*vdev
= VIRTIO_DEVICE(dev
);
508 VirtQueue
*vq
= dev
->free_page_vq
;
511 while (dev
->block_iothread
) {
512 qemu_cond_wait(&dev
->free_page_cond
, &dev
->free_page_lock
);
515 elem
= virtqueue_pop(vq
, sizeof(VirtQueueElement
));
522 size_t size
= iov_to_buf(elem
->out_sg
, elem
->out_num
, 0,
525 virtio_tswap32s(vdev
, &id
);
526 if (unlikely(size
!= sizeof(id
))) {
527 virtio_error(vdev
, "received an incorrect cmd id");
531 if (id
== dev
->free_page_report_cmd_id
) {
532 dev
->free_page_report_status
= FREE_PAGE_REPORT_S_START
;
535 * Stop the optimization only when it has started. This
536 * avoids a stale stop sign for the previous command.
538 if (dev
->free_page_report_status
== FREE_PAGE_REPORT_S_START
) {
539 dev
->free_page_report_status
= FREE_PAGE_REPORT_S_STOP
;
545 if (dev
->free_page_report_status
== FREE_PAGE_REPORT_S_START
) {
546 qemu_guest_free_page_hint(elem
->in_sg
[0].iov_base
,
547 elem
->in_sg
[0].iov_len
);
552 virtqueue_push(vq
, elem
, 1);
557 static void virtio_ballloon_get_free_page_hints(void *opaque
)
559 VirtIOBalloon
*dev
= opaque
;
560 VirtIODevice
*vdev
= VIRTIO_DEVICE(dev
);
561 VirtQueue
*vq
= dev
->free_page_vq
;
562 bool continue_to_get_hints
;
565 qemu_mutex_lock(&dev
->free_page_lock
);
566 virtio_queue_set_notification(vq
, 0);
567 continue_to_get_hints
= get_free_page_hints(dev
);
568 qemu_mutex_unlock(&dev
->free_page_lock
);
569 virtio_notify(vdev
, vq
);
571 * Start to poll the vq once the reporting started. Otherwise, continue
572 * only when there are entries on the vq, which need to be given back.
574 } while (continue_to_get_hints
||
575 dev
->free_page_report_status
== FREE_PAGE_REPORT_S_START
);
576 virtio_queue_set_notification(vq
, 1);
579 static bool virtio_balloon_free_page_support(void *opaque
)
581 VirtIOBalloon
*s
= opaque
;
582 VirtIODevice
*vdev
= VIRTIO_DEVICE(s
);
584 return virtio_vdev_has_feature(vdev
, VIRTIO_BALLOON_F_FREE_PAGE_HINT
);
587 static void virtio_balloon_free_page_start(VirtIOBalloon
*s
)
589 VirtIODevice
*vdev
= VIRTIO_DEVICE(s
);
591 /* For the stop and copy phase, we don't need to start the optimization */
592 if (!vdev
->vm_running
) {
596 if (s
->free_page_report_cmd_id
== UINT_MAX
) {
597 s
->free_page_report_cmd_id
=
598 VIRTIO_BALLOON_FREE_PAGE_REPORT_CMD_ID_MIN
;
600 s
->free_page_report_cmd_id
++;
603 s
->free_page_report_status
= FREE_PAGE_REPORT_S_REQUESTED
;
604 virtio_notify_config(vdev
);
607 static void virtio_balloon_free_page_stop(VirtIOBalloon
*s
)
609 VirtIODevice
*vdev
= VIRTIO_DEVICE(s
);
611 if (s
->free_page_report_status
!= FREE_PAGE_REPORT_S_STOP
) {
613 * The lock also guarantees us that the
614 * virtio_ballloon_get_free_page_hints exits after the
615 * free_page_report_status is set to S_STOP.
617 qemu_mutex_lock(&s
->free_page_lock
);
619 * The guest hasn't done the reporting, so host sends a notification
620 * to the guest to actively stop the reporting.
622 s
->free_page_report_status
= FREE_PAGE_REPORT_S_STOP
;
623 qemu_mutex_unlock(&s
->free_page_lock
);
624 virtio_notify_config(vdev
);
628 static void virtio_balloon_free_page_done(VirtIOBalloon
*s
)
630 VirtIODevice
*vdev
= VIRTIO_DEVICE(s
);
632 if (s
->free_page_report_status
!= FREE_PAGE_REPORT_S_DONE
) {
633 /* See virtio_balloon_free_page_stop() */
634 qemu_mutex_lock(&s
->free_page_lock
);
635 s
->free_page_report_status
= FREE_PAGE_REPORT_S_DONE
;
636 qemu_mutex_unlock(&s
->free_page_lock
);
637 virtio_notify_config(vdev
);
642 virtio_balloon_free_page_report_notify(NotifierWithReturn
*n
, void *data
)
644 VirtIOBalloon
*dev
= container_of(n
, VirtIOBalloon
,
645 free_page_report_notify
);
646 VirtIODevice
*vdev
= VIRTIO_DEVICE(dev
);
647 PrecopyNotifyData
*pnd
= data
;
649 if (!virtio_balloon_free_page_support(dev
)) {
651 * This is an optimization provided to migration, so just return 0 to
652 * have the normal migration process not affected when this feature is
658 switch (pnd
->reason
) {
659 case PRECOPY_NOTIFY_SETUP
:
660 precopy_enable_free_page_optimization();
662 case PRECOPY_NOTIFY_BEFORE_BITMAP_SYNC
:
663 virtio_balloon_free_page_stop(dev
);
665 case PRECOPY_NOTIFY_AFTER_BITMAP_SYNC
:
666 if (vdev
->vm_running
) {
667 virtio_balloon_free_page_start(dev
);
671 * Set S_DONE before migrating the vmstate, so the guest will reuse
672 * all hinted pages once running on the destination. Fall through.
674 case PRECOPY_NOTIFY_CLEANUP
:
676 * Especially, if something goes wrong during precopy or if migration
677 * is canceled, we have to properly communicate S_DONE to the VM.
679 virtio_balloon_free_page_done(dev
);
681 case PRECOPY_NOTIFY_COMPLETE
:
684 virtio_error(vdev
, "%s: %d reason unknown", __func__
, pnd
->reason
);
690 static size_t virtio_balloon_config_size(VirtIOBalloon
*s
)
692 uint64_t features
= s
->host_features
;
694 if (s
->qemu_4_0_config_size
) {
695 return sizeof(struct virtio_balloon_config
);
697 if (virtio_has_feature(features
, VIRTIO_BALLOON_F_PAGE_POISON
)) {
698 return sizeof(struct virtio_balloon_config
);
700 if (virtio_has_feature(features
, VIRTIO_BALLOON_F_FREE_PAGE_HINT
)) {
701 return offsetof(struct virtio_balloon_config
, poison_val
);
703 return offsetof(struct virtio_balloon_config
, free_page_report_cmd_id
);
706 static void virtio_balloon_get_config(VirtIODevice
*vdev
, uint8_t *config_data
)
708 VirtIOBalloon
*dev
= VIRTIO_BALLOON(vdev
);
709 struct virtio_balloon_config config
= {};
711 config
.num_pages
= cpu_to_le32(dev
->num_pages
);
712 config
.actual
= cpu_to_le32(dev
->actual
);
713 config
.poison_val
= cpu_to_le32(dev
->poison_val
);
715 if (dev
->free_page_report_status
== FREE_PAGE_REPORT_S_REQUESTED
) {
716 config
.free_page_report_cmd_id
=
717 cpu_to_le32(dev
->free_page_report_cmd_id
);
718 } else if (dev
->free_page_report_status
== FREE_PAGE_REPORT_S_STOP
) {
719 config
.free_page_report_cmd_id
=
720 cpu_to_le32(VIRTIO_BALLOON_CMD_ID_STOP
);
721 } else if (dev
->free_page_report_status
== FREE_PAGE_REPORT_S_DONE
) {
722 config
.free_page_report_cmd_id
=
723 cpu_to_le32(VIRTIO_BALLOON_CMD_ID_DONE
);
726 trace_virtio_balloon_get_config(config
.num_pages
, config
.actual
);
727 memcpy(config_data
, &config
, virtio_balloon_config_size(dev
));
730 static int build_dimm_list(Object
*obj
, void *opaque
)
732 GSList
**list
= opaque
;
734 if (object_dynamic_cast(obj
, TYPE_PC_DIMM
)) {
735 DeviceState
*dev
= DEVICE(obj
);
736 if (dev
->realized
) { /* only realized DIMMs matter */
737 *list
= g_slist_prepend(*list
, dev
);
741 object_child_foreach(obj
, build_dimm_list
, opaque
);
745 static ram_addr_t
get_current_ram_size(void)
747 GSList
*list
= NULL
, *item
;
748 ram_addr_t size
= ram_size
;
750 build_dimm_list(qdev_get_machine(), &list
);
751 for (item
= list
; item
; item
= g_slist_next(item
)) {
752 Object
*obj
= OBJECT(item
->data
);
753 if (!strcmp(object_get_typename(obj
), TYPE_PC_DIMM
)) {
754 size
+= object_property_get_int(obj
, PC_DIMM_SIZE_PROP
,
763 static bool virtio_balloon_page_poison_support(void *opaque
)
765 VirtIOBalloon
*s
= opaque
;
766 VirtIODevice
*vdev
= VIRTIO_DEVICE(s
);
768 return virtio_vdev_has_feature(vdev
, VIRTIO_BALLOON_F_PAGE_POISON
);
771 static void virtio_balloon_set_config(VirtIODevice
*vdev
,
772 const uint8_t *config_data
)
774 VirtIOBalloon
*dev
= VIRTIO_BALLOON(vdev
);
775 struct virtio_balloon_config config
;
776 uint32_t oldactual
= dev
->actual
;
777 ram_addr_t vm_ram_size
= get_current_ram_size();
779 memcpy(&config
, config_data
, virtio_balloon_config_size(dev
));
780 dev
->actual
= le32_to_cpu(config
.actual
);
781 if (dev
->actual
!= oldactual
) {
782 qapi_event_send_balloon_change(vm_ram_size
-
783 ((ram_addr_t
) dev
->actual
<< VIRTIO_BALLOON_PFN_SHIFT
));
786 if (virtio_balloon_page_poison_support(dev
)) {
787 dev
->poison_val
= le32_to_cpu(config
.poison_val
);
789 trace_virtio_balloon_set_config(dev
->actual
, oldactual
);
792 static uint64_t virtio_balloon_get_features(VirtIODevice
*vdev
, uint64_t f
,
795 VirtIOBalloon
*dev
= VIRTIO_BALLOON(vdev
);
796 f
|= dev
->host_features
;
797 virtio_add_feature(&f
, VIRTIO_BALLOON_F_STATS_VQ
);
802 static void virtio_balloon_stat(void *opaque
, BalloonInfo
*info
)
804 VirtIOBalloon
*dev
= opaque
;
805 info
->actual
= get_current_ram_size() - ((uint64_t) dev
->actual
<<
806 VIRTIO_BALLOON_PFN_SHIFT
);
809 static void virtio_balloon_to_target(void *opaque
, ram_addr_t target
)
811 VirtIOBalloon
*dev
= VIRTIO_BALLOON(opaque
);
812 VirtIODevice
*vdev
= VIRTIO_DEVICE(dev
);
813 ram_addr_t vm_ram_size
= get_current_ram_size();
815 if (target
> vm_ram_size
) {
816 target
= vm_ram_size
;
819 dev
->num_pages
= (vm_ram_size
- target
) >> VIRTIO_BALLOON_PFN_SHIFT
;
820 virtio_notify_config(vdev
);
822 trace_virtio_balloon_to_target(target
, dev
->num_pages
);
825 static int virtio_balloon_post_load_device(void *opaque
, int version_id
)
827 VirtIOBalloon
*s
= VIRTIO_BALLOON(opaque
);
829 if (balloon_stats_enabled(s
)) {
830 balloon_stats_change_timer(s
, s
->stats_poll_interval
);
835 static const VMStateDescription vmstate_virtio_balloon_free_page_report
= {
836 .name
= "virtio-balloon-device/free-page-report",
838 .minimum_version_id
= 1,
839 .needed
= virtio_balloon_free_page_support
,
840 .fields
= (VMStateField
[]) {
841 VMSTATE_UINT32(free_page_report_cmd_id
, VirtIOBalloon
),
842 VMSTATE_UINT32(free_page_report_status
, VirtIOBalloon
),
843 VMSTATE_END_OF_LIST()
847 static const VMStateDescription vmstate_virtio_balloon_page_poison
= {
848 .name
= "vitio-balloon-device/page-poison",
850 .minimum_version_id
= 1,
851 .needed
= virtio_balloon_page_poison_support
,
852 .fields
= (VMStateField
[]) {
853 VMSTATE_UINT32(poison_val
, VirtIOBalloon
),
854 VMSTATE_END_OF_LIST()
858 static const VMStateDescription vmstate_virtio_balloon_device
= {
859 .name
= "virtio-balloon-device",
861 .minimum_version_id
= 1,
862 .post_load
= virtio_balloon_post_load_device
,
863 .fields
= (VMStateField
[]) {
864 VMSTATE_UINT32(num_pages
, VirtIOBalloon
),
865 VMSTATE_UINT32(actual
, VirtIOBalloon
),
866 VMSTATE_END_OF_LIST()
868 .subsections
= (const VMStateDescription
* []) {
869 &vmstate_virtio_balloon_free_page_report
,
870 &vmstate_virtio_balloon_page_poison
,
875 static void virtio_balloon_device_realize(DeviceState
*dev
, Error
**errp
)
877 VirtIODevice
*vdev
= VIRTIO_DEVICE(dev
);
878 VirtIOBalloon
*s
= VIRTIO_BALLOON(dev
);
881 virtio_init(vdev
, "virtio-balloon", VIRTIO_ID_BALLOON
,
882 virtio_balloon_config_size(s
));
884 ret
= qemu_add_balloon_handler(virtio_balloon_to_target
,
885 virtio_balloon_stat
, s
);
888 error_setg(errp
, "Only one balloon device is supported");
889 virtio_cleanup(vdev
);
893 if (virtio_has_feature(s
->host_features
, VIRTIO_BALLOON_F_FREE_PAGE_HINT
) &&
895 error_setg(errp
, "'free-page-hint' requires 'iothread' to be set");
896 virtio_cleanup(vdev
);
900 s
->ivq
= virtio_add_queue(vdev
, 128, virtio_balloon_handle_output
);
901 s
->dvq
= virtio_add_queue(vdev
, 128, virtio_balloon_handle_output
);
902 s
->svq
= virtio_add_queue(vdev
, 128, virtio_balloon_receive_stats
);
904 if (virtio_has_feature(s
->host_features
,
905 VIRTIO_BALLOON_F_FREE_PAGE_HINT
)) {
906 s
->free_page_vq
= virtio_add_queue(vdev
, VIRTQUEUE_MAX_SIZE
,
907 virtio_balloon_handle_free_page_vq
);
908 precopy_add_notifier(&s
->free_page_report_notify
);
910 object_ref(OBJECT(s
->iothread
));
911 s
->free_page_bh
= aio_bh_new(iothread_get_aio_context(s
->iothread
),
912 virtio_ballloon_get_free_page_hints
, s
);
915 if (virtio_has_feature(s
->host_features
, VIRTIO_BALLOON_F_REPORTING
)) {
916 s
->reporting_vq
= virtio_add_queue(vdev
, 32,
917 virtio_balloon_handle_report
);
923 static void virtio_balloon_device_unrealize(DeviceState
*dev
)
925 VirtIODevice
*vdev
= VIRTIO_DEVICE(dev
);
926 VirtIOBalloon
*s
= VIRTIO_BALLOON(dev
);
928 if (s
->free_page_bh
) {
929 qemu_bh_delete(s
->free_page_bh
);
930 object_unref(OBJECT(s
->iothread
));
931 virtio_balloon_free_page_stop(s
);
932 precopy_remove_notifier(&s
->free_page_report_notify
);
934 balloon_stats_destroy_timer(s
);
935 qemu_remove_balloon_handler(s
);
937 virtio_delete_queue(s
->ivq
);
938 virtio_delete_queue(s
->dvq
);
939 virtio_delete_queue(s
->svq
);
940 if (s
->free_page_vq
) {
941 virtio_delete_queue(s
->free_page_vq
);
943 if (s
->reporting_vq
) {
944 virtio_delete_queue(s
->reporting_vq
);
946 virtio_cleanup(vdev
);
949 static void virtio_balloon_device_reset(VirtIODevice
*vdev
)
951 VirtIOBalloon
*s
= VIRTIO_BALLOON(vdev
);
953 if (virtio_balloon_free_page_support(s
)) {
954 virtio_balloon_free_page_stop(s
);
957 if (s
->stats_vq_elem
!= NULL
) {
958 virtqueue_unpop(s
->svq
, s
->stats_vq_elem
, 0);
959 g_free(s
->stats_vq_elem
);
960 s
->stats_vq_elem
= NULL
;
966 static void virtio_balloon_set_status(VirtIODevice
*vdev
, uint8_t status
)
968 VirtIOBalloon
*s
= VIRTIO_BALLOON(vdev
);
970 if (!s
->stats_vq_elem
&& vdev
->vm_running
&&
971 (status
& VIRTIO_CONFIG_S_DRIVER_OK
) && virtqueue_rewind(s
->svq
, 1)) {
972 /* poll stats queue for the element we have discarded when the VM
974 virtio_balloon_receive_stats(vdev
, s
->svq
);
977 if (virtio_balloon_free_page_support(s
)) {
979 * The VM is woken up and the iothread was blocked, so signal it to
982 if (vdev
->vm_running
&& s
->block_iothread
) {
983 qemu_mutex_lock(&s
->free_page_lock
);
984 s
->block_iothread
= false;
985 qemu_cond_signal(&s
->free_page_cond
);
986 qemu_mutex_unlock(&s
->free_page_lock
);
989 /* The VM is stopped, block the iothread. */
990 if (!vdev
->vm_running
) {
991 qemu_mutex_lock(&s
->free_page_lock
);
992 s
->block_iothread
= true;
993 qemu_mutex_unlock(&s
->free_page_lock
);
998 static void virtio_balloon_instance_init(Object
*obj
)
1000 VirtIOBalloon
*s
= VIRTIO_BALLOON(obj
);
1002 qemu_mutex_init(&s
->free_page_lock
);
1003 qemu_cond_init(&s
->free_page_cond
);
1004 s
->free_page_report_cmd_id
= VIRTIO_BALLOON_FREE_PAGE_REPORT_CMD_ID_MIN
;
1005 s
->free_page_report_notify
.notify
= virtio_balloon_free_page_report_notify
;
1007 object_property_add(obj
, "guest-stats", "guest statistics",
1008 balloon_stats_get_all
, NULL
, NULL
, s
);
1010 object_property_add(obj
, "guest-stats-polling-interval", "int",
1011 balloon_stats_get_poll_interval
,
1012 balloon_stats_set_poll_interval
,
1016 static const VMStateDescription vmstate_virtio_balloon
= {
1017 .name
= "virtio-balloon",
1018 .minimum_version_id
= 1,
1020 .fields
= (VMStateField
[]) {
1021 VMSTATE_VIRTIO_DEVICE
,
1022 VMSTATE_END_OF_LIST()
1026 static Property virtio_balloon_properties
[] = {
1027 DEFINE_PROP_BIT("deflate-on-oom", VirtIOBalloon
, host_features
,
1028 VIRTIO_BALLOON_F_DEFLATE_ON_OOM
, false),
1029 DEFINE_PROP_BIT("free-page-hint", VirtIOBalloon
, host_features
,
1030 VIRTIO_BALLOON_F_FREE_PAGE_HINT
, false),
1031 DEFINE_PROP_BIT("page-poison", VirtIOBalloon
, host_features
,
1032 VIRTIO_BALLOON_F_PAGE_POISON
, true),
1033 DEFINE_PROP_BIT("free-page-reporting", VirtIOBalloon
, host_features
,
1034 VIRTIO_BALLOON_F_REPORTING
, false),
1035 /* QEMU 4.0 accidentally changed the config size even when free-page-hint
1036 * is disabled, resulting in QEMU 3.1 migration incompatibility. This
1037 * property retains this quirk for QEMU 4.1 machine types.
1039 DEFINE_PROP_BOOL("qemu-4-0-config-size", VirtIOBalloon
,
1040 qemu_4_0_config_size
, false),
1041 DEFINE_PROP_LINK("iothread", VirtIOBalloon
, iothread
, TYPE_IOTHREAD
,
1043 DEFINE_PROP_END_OF_LIST(),
1046 static void virtio_balloon_class_init(ObjectClass
*klass
, void *data
)
1048 DeviceClass
*dc
= DEVICE_CLASS(klass
);
1049 VirtioDeviceClass
*vdc
= VIRTIO_DEVICE_CLASS(klass
);
1051 device_class_set_props(dc
, virtio_balloon_properties
);
1052 dc
->vmsd
= &vmstate_virtio_balloon
;
1053 set_bit(DEVICE_CATEGORY_MISC
, dc
->categories
);
1054 vdc
->realize
= virtio_balloon_device_realize
;
1055 vdc
->unrealize
= virtio_balloon_device_unrealize
;
1056 vdc
->reset
= virtio_balloon_device_reset
;
1057 vdc
->get_config
= virtio_balloon_get_config
;
1058 vdc
->set_config
= virtio_balloon_set_config
;
1059 vdc
->get_features
= virtio_balloon_get_features
;
1060 vdc
->set_status
= virtio_balloon_set_status
;
1061 vdc
->vmsd
= &vmstate_virtio_balloon_device
;
1064 static const TypeInfo virtio_balloon_info
= {
1065 .name
= TYPE_VIRTIO_BALLOON
,
1066 .parent
= TYPE_VIRTIO_DEVICE
,
1067 .instance_size
= sizeof(VirtIOBalloon
),
1068 .instance_init
= virtio_balloon_instance_init
,
1069 .class_init
= virtio_balloon_class_init
,
1072 static void virtio_register_types(void)
1074 type_register_static(&virtio_balloon_info
);
1077 type_init(virtio_register_types
)