2 * QEMU Hyper-V Dynamic Memory Protocol driver
4 * Copyright (C) 2020-2023 Oracle and/or its affiliates.
6 * This work is licensed under the terms of the GNU GPL, version 2 or later.
7 * See the COPYING file in the top-level directory.
10 #include "hv-balloon-internal.h"
12 #include "exec/address-spaces.h"
13 #include "exec/cpu-common.h"
14 #include "exec/ramblock.h"
15 #include "hw/boards.h"
16 #include "hw/hyperv/dynmem-proto.h"
17 #include "hw/hyperv/hv-balloon.h"
18 #include "hw/hyperv/vmbus.h"
19 #include "hw/mem/memory-device.h"
20 #include "hw/mem/pc-dimm.h"
21 #include "hw/qdev-core.h"
22 #include "hw/qdev-properties.h"
23 #include "monitor/qdev.h"
24 #include "qapi/error.h"
25 #include "qapi/qapi-commands-machine.h"
26 #include "qapi/qapi-events-machine.h"
27 #include "qapi/qapi-types-machine.h"
28 #include "qapi/qmp/qdict.h"
29 #include "qapi/visitor.h"
30 #include "qemu/error-report.h"
31 #include "qemu/module.h"
32 #include "qemu/units.h"
33 #include "qemu/timer.h"
34 #include "sysemu/balloon.h"
35 #include "sysemu/hostmem.h"
36 #include "sysemu/reset.h"
37 #include "hv-balloon-our_range_memslots.h"
38 #include "hv-balloon-page_range_tree.h"
41 #define HV_BALLOON_ADDR_PROP "addr"
42 #define HV_BALLOON_MEMDEV_PROP "memdev"
43 #define HV_BALLOON_GUID "525074DC-8985-46e2-8057-A307DC18A502"
46 * Some Windows versions (at least Server 2019) will crash with various
47 * error codes when receiving DM protocol requests (at least
48 * DM_MEM_HOT_ADD_REQUEST) immediately after boot.
50 * It looks like Hyper-V from Server 2016 uses a 50-second after-boot
51 * delay, probably to workaround this issue, so we'll use this value, too.
53 #define HV_BALLOON_POST_INIT_WAIT (50 * 1000)
55 #define HV_BALLOON_HA_CHUNK_SIZE (2 * GiB)
56 #define HV_BALLOON_HA_CHUNK_PAGES (HV_BALLOON_HA_CHUNK_SIZE / HV_BALLOON_PAGE_SIZE)
58 #define HV_BALLOON_HA_MEMSLOT_SIZE_ALIGN (128 * MiB)
60 #define HV_BALLOON_HR_CHUNK_PAGES 585728
62 * ^ that's the maximum number of pages
63 * that Windows returns in one hot remove response
65 * If the number requested is too high Windows will no longer honor
69 struct HvBalloonClass
{
70 VMBusDeviceClass parent_class
;
74 /* not a real state */
92 /* unballoon + hot add ops flow */
95 S_UNBALLOON_REPLY_WAIT
,
102 typedef struct StateDesc
{
107 typedef struct HvBalloon
{
111 union dm_version version
;
114 QEMUTimer post_init_timer
;
116 unsigned int trans_id
;
125 /* Guest target size */
129 /* Current (un)balloon / hot-add operation parameters */
131 uint64_t balloon_diff
;
134 uint64_t unballoon_diff
;
135 uint64_t hot_add_diff
;
139 PageRange hot_add_range
;
140 uint64_t ha_current_count
;
144 OurRangeMemslots
*our_range
;
146 /* Count of memslots covering our memory */
147 unsigned int memslot_count
;
149 /* Nominal size of each memslot (the last one might be smaller) */
150 uint64_t memslot_size
;
152 /* Non-ours removed memory */
153 PageRangeTree removed_guest
, removed_both
;
155 /* Grand totals of removed memory (both ours and non-ours) */
156 uint64_t removed_guest_ctr
, removed_both_ctr
;
158 /* MEMORY_DEVICE props */
160 HostMemoryBackend
*hostmem
;
164 OBJECT_DEFINE_TYPE_WITH_INTERFACES(HvBalloon
, hv_balloon
, HV_BALLOON
, VMBUS_DEVICE
, \
165 { TYPE_MEMORY_DEVICE
}, { })
167 #define HV_BALLOON_SET_STATE(hvb, news) \
169 assert(news != S_NO_CHANGE); \
170 hv_balloon_state_set(hvb, news, # news); \
173 #define HV_BALLOON_STATE_DESC_SET(stdesc, news) \
174 _hv_balloon_state_desc_set(stdesc, news, # news)
176 #define HV_BALLOON_STATE_DESC_INIT \
178 .state = S_NO_CHANGE, \
181 typedef struct HvBalloonReq
{
185 /* total our memory includes parts currently removed from the guest */
186 static uint64_t hv_balloon_total_our_ram(HvBalloon
*balloon
)
188 if (!balloon
->our_range
) {
192 return balloon
->our_range
->range
.added
;
195 /* TODO: unify the code below with virtio-balloon and cache the value */
196 static int build_dimm_list(Object
*obj
, void *opaque
)
198 GSList
**list
= opaque
;
200 if (object_dynamic_cast(obj
, TYPE_PC_DIMM
)) {
201 DeviceState
*dev
= DEVICE(obj
);
202 if (dev
->realized
) { /* only realized DIMMs matter */
203 *list
= g_slist_prepend(*list
, dev
);
207 object_child_foreach(obj
, build_dimm_list
, opaque
);
211 static ram_addr_t
get_current_ram_size(void)
213 GSList
*list
= NULL
, *item
;
214 ram_addr_t size
= current_machine
->ram_size
;
216 build_dimm_list(qdev_get_machine(), &list
);
217 for (item
= list
; item
; item
= g_slist_next(item
)) {
218 Object
*obj
= OBJECT(item
->data
);
219 if (!strcmp(object_get_typename(obj
), TYPE_PC_DIMM
))
220 size
+= object_property_get_int(obj
, PC_DIMM_SIZE_PROP
,
228 /* total RAM includes memory currently removed from the guest */
229 static uint64_t hv_balloon_total_ram(HvBalloon
*balloon
)
231 ram_addr_t ram_size
= get_current_ram_size();
232 uint64_t ram_size_pages
= ram_size
>> HV_BALLOON_PFN_SHIFT
;
233 uint64_t our_ram_size_pages
= hv_balloon_total_our_ram(balloon
);
235 assert(ram_size_pages
> 0);
237 return SUM_SATURATE_U64(ram_size_pages
, our_ram_size_pages
);
241 * calculating the total RAM size is a slow operation,
242 * avoid it as much as possible
244 static uint64_t hv_balloon_total_removed_rs(HvBalloon
*balloon
,
245 uint64_t ram_size_pages
)
247 uint64_t total_removed
;
249 total_removed
= SUM_SATURATE_U64(balloon
->removed_guest_ctr
,
250 balloon
->removed_both_ctr
);
252 /* possible if guest returns pages outside actual RAM */
253 if (total_removed
> ram_size_pages
) {
254 total_removed
= ram_size_pages
;
257 return total_removed
;
260 /* Returns whether the state has actually changed */
261 static bool hv_balloon_state_set(HvBalloon
*balloon
,
262 State newst
, const char *newststr
)
264 if (newst
== S_NO_CHANGE
|| balloon
->state
== newst
) {
268 balloon
->state
= newst
;
269 trace_hv_balloon_state_change(newststr
);
273 static void _hv_balloon_state_desc_set(StateDesc
*stdesc
,
274 State newst
, const char *newststr
)
276 /* state setting is only permitted on a freshly init desc */
277 assert(stdesc
->state
== S_NO_CHANGE
);
279 assert(newst
!= S_NO_CHANGE
);
281 stdesc
->state
= newst
;
282 stdesc
->desc
= newststr
;
285 static VMBusChannel
*hv_balloon_get_channel_maybe(HvBalloon
*balloon
)
287 return vmbus_device_channel(&balloon
->parent
, 0);
290 static VMBusChannel
*hv_balloon_get_channel(HvBalloon
*balloon
)
294 chan
= hv_balloon_get_channel_maybe(balloon
);
295 assert(chan
!= NULL
);
299 static ssize_t
hv_balloon_send_packet(VMBusChannel
*chan
,
300 struct dm_message
*msg
)
304 ret
= vmbus_channel_reserve(chan
, 0, msg
->hdr
.size
);
309 return vmbus_channel_send(chan
, VMBUS_PACKET_DATA_INBAND
,
310 NULL
, 0, msg
, msg
->hdr
.size
, false,
314 static bool hv_balloon_unballoon_get_source(HvBalloon
*balloon
,
315 PageRangeTree
*dtree
,
319 OurRange
*our_range
= OUR_RANGE(balloon
->our_range
);
321 /* Try the boot memory first */
322 if (g_tree_nnodes(balloon
->removed_guest
.t
) > 0) {
323 *dtree
= balloon
->removed_guest
;
324 *dctr
= &balloon
->removed_guest_ctr
;
325 *is_our_range
= false;
326 } else if (g_tree_nnodes(balloon
->removed_both
.t
) > 0) {
327 *dtree
= balloon
->removed_both
;
328 *dctr
= &balloon
->removed_both_ctr
;
329 *is_our_range
= false;
330 } else if (!our_range
) {
332 } else if (!our_range_is_removed_tree_empty(our_range
, false)) {
333 *dtree
= our_range_get_removed_tree(our_range
, false);
334 *dctr
= &balloon
->removed_guest_ctr
;
335 *is_our_range
= true;
336 } else if (!our_range_is_removed_tree_empty(our_range
, true)) {
337 *dtree
= our_range_get_removed_tree(our_range
, true);
338 *dctr
= &balloon
->removed_both_ctr
;
339 *is_our_range
= true;
347 static void hv_balloon_unballoon_rb_wait(HvBalloon
*balloon
, StateDesc
*stdesc
)
349 VMBusChannel
*chan
= hv_balloon_get_channel(balloon
);
350 struct dm_unballoon_request
*ur
;
351 size_t ur_size
= sizeof(*ur
) + sizeof(ur
->range_array
[0]);
353 assert(balloon
->state
== S_UNBALLOON_RB_WAIT
);
355 if (vmbus_channel_reserve(chan
, 0, ur_size
) < 0) {
359 HV_BALLOON_STATE_DESC_SET(stdesc
, S_UNBALLOON_POSTING
);
362 static void hv_balloon_unballoon_posting(HvBalloon
*balloon
, StateDesc
*stdesc
)
364 VMBusChannel
*chan
= hv_balloon_get_channel(balloon
);
368 struct dm_unballoon_request
*ur
;
369 size_t ur_size
= sizeof(*ur
) + sizeof(ur
->range_array
[0]);
374 assert(balloon
->state
== S_UNBALLOON_POSTING
);
375 assert(balloon
->unballoon_diff
> 0);
377 if (!hv_balloon_unballoon_get_source(balloon
, &dtree
, &dctr
, &our_range
)) {
378 error_report("trying to unballoon but nothing seems to be ballooned");
380 * there is little we can do as we might have already
381 * sent the guest a partial request we can't cancel
386 assert(balloon
->our_range
|| !our_range
);
390 ur
= alloca(ur_size
);
391 memset(ur
, 0, ur_size
);
392 ur
->hdr
.type
= DM_UNBALLOON_REQUEST
;
393 ur
->hdr
.size
= ur_size
;
394 ur
->hdr
.trans_id
= balloon
->trans_id
;
396 bret
= hvb_page_range_tree_pop(dtree
, &range
, MIN(balloon
->unballoon_diff
,
397 HV_BALLOON_HA_CHUNK_PAGES
));
401 *dctr
-= range
.count
;
402 balloon
->unballoon_diff
-= range
.count
;
405 ur
->range_array
[0].finfo
.start_page
= range
.start
;
406 ur
->range_array
[0].finfo
.page_cnt
= range
.count
;
407 ur
->more_pages
= balloon
->unballoon_diff
> 0;
409 trace_hv_balloon_outgoing_unballoon(ur
->hdr
.trans_id
,
410 range
.count
, range
.start
,
411 balloon
->unballoon_diff
);
413 if (ur
->more_pages
) {
414 HV_BALLOON_STATE_DESC_SET(stdesc
, S_UNBALLOON_RB_WAIT
);
416 HV_BALLOON_STATE_DESC_SET(stdesc
, S_UNBALLOON_REPLY_WAIT
);
419 ret
= vmbus_channel_send(chan
, VMBUS_PACKET_DATA_INBAND
,
420 NULL
, 0, ur
, ur_size
, false,
423 error_report("error %zd when posting unballoon msg, expect problems",
428 static bool hv_balloon_our_range_ensure(HvBalloon
*balloon
)
431 MemoryRegion
*hostmem_mr
;
432 g_autoptr(OurRangeMemslots
) our_range_memslots
= NULL
;
435 if (balloon
->our_range
) {
439 if (!balloon
->hostmem
) {
443 align
= (1 << balloon
->caps
.cap_bits
.hot_add_alignment
) * MiB
;
444 assert(QEMU_IS_ALIGNED(balloon
->addr
, align
));
446 hostmem_mr
= host_memory_backend_get_memory(balloon
->hostmem
);
448 our_range_memslots
= hvb_our_range_memslots_new(balloon
->addr
,
449 balloon
->mr
, hostmem_mr
,
451 balloon
->memslot_count
,
452 balloon
->memslot_size
);
453 our_range
= OUR_RANGE(our_range_memslots
);
455 if (hvb_page_range_tree_intree_any(balloon
->removed_guest
,
456 our_range
->range
.start
,
457 our_range
->range
.count
) ||
458 hvb_page_range_tree_intree_any(balloon
->removed_both
,
459 our_range
->range
.start
,
460 our_range
->range
.count
)) {
461 error_report("some parts of the memory backend were already returned by the guest. this should not happen, please reboot the guest and try again");
465 trace_hv_balloon_our_range_add(our_range
->range
.count
,
466 our_range
->range
.start
);
468 balloon
->our_range
= g_steal_pointer(&our_range_memslots
);
472 static void hv_balloon_hot_add_setup(HvBalloon
*balloon
, StateDesc
*stdesc
)
474 /* need to make copy since it is in union with hot_add_range */
475 uint64_t hot_add_diff
= balloon
->hot_add_diff
;
476 PageRange
*hot_add_range
= &balloon
->hot_add_range
;
477 uint64_t align
, our_range_remaining
;
480 assert(balloon
->state
== S_HOT_ADD_SETUP
);
481 assert(hot_add_diff
> 0);
483 if (!hv_balloon_our_range_ensure(balloon
)) {
487 our_range
= OUR_RANGE(balloon
->our_range
);
489 align
= (1 << balloon
->caps
.cap_bits
.hot_add_alignment
) *
490 (MiB
/ HV_BALLOON_PAGE_SIZE
);
492 /* Absolute GPA in pages */
493 hot_add_range
->start
= our_range_get_remaining_start(our_range
);
494 assert(QEMU_IS_ALIGNED(hot_add_range
->start
, align
));
496 our_range_remaining
= our_range_get_remaining_size(our_range
);
497 hot_add_range
->count
= MIN(our_range_remaining
, hot_add_diff
);
498 hot_add_range
->count
= QEMU_ALIGN_DOWN(hot_add_range
->count
, align
);
499 if (hot_add_range
->count
== 0) {
503 hvb_our_range_memslots_ensure_mapped_additional(balloon
->our_range
,
504 hot_add_range
->count
);
506 HV_BALLOON_STATE_DESC_SET(stdesc
, S_HOT_ADD_RB_WAIT
);
510 HV_BALLOON_STATE_DESC_SET(stdesc
, S_IDLE
);
513 static void hv_balloon_hot_add_rb_wait(HvBalloon
*balloon
, StateDesc
*stdesc
)
515 VMBusChannel
*chan
= hv_balloon_get_channel(balloon
);
516 struct dm_hot_add
*ha
;
517 size_t ha_size
= sizeof(*ha
) + sizeof(ha
->range
);
519 assert(balloon
->state
== S_HOT_ADD_RB_WAIT
);
521 if (vmbus_channel_reserve(chan
, 0, ha_size
) < 0) {
525 HV_BALLOON_STATE_DESC_SET(stdesc
, S_HOT_ADD_POSTING
);
528 static void hv_balloon_hot_add_posting(HvBalloon
*balloon
, StateDesc
*stdesc
)
530 PageRange
*hot_add_range
= &balloon
->hot_add_range
;
531 uint64_t *current_count
= &balloon
->ha_current_count
;
532 VMBusChannel
*chan
= hv_balloon_get_channel(balloon
);
533 struct dm_hot_add
*ha
;
534 size_t ha_size
= sizeof(*ha
) + sizeof(ha
->range
);
535 union dm_mem_page_range
*ha_region
;
536 uint64_t align
, chunk_max_size
;
539 assert(balloon
->state
== S_HOT_ADD_POSTING
);
540 assert(hot_add_range
->count
> 0);
542 align
= (1 << balloon
->caps
.cap_bits
.hot_add_alignment
) *
543 (MiB
/ HV_BALLOON_PAGE_SIZE
);
544 if (align
>= HV_BALLOON_HA_CHUNK_PAGES
) {
546 * If the required alignment is higher than the chunk size we let it
547 * override that size.
549 chunk_max_size
= align
;
551 chunk_max_size
= QEMU_ALIGN_DOWN(HV_BALLOON_HA_CHUNK_PAGES
, align
);
555 * hot_add_range->count starts aligned in hv_balloon_hot_add_setup(),
556 * then it is either reduced by subtracting aligned current_count or
557 * further hot-adds are prevented by marking the whole remaining our range
558 * as unusable in hv_balloon_handle_hot_add_response().
560 *current_count
= MIN(hot_add_range
->count
, chunk_max_size
);
562 ha
= alloca(ha_size
);
563 ha_region
= &(&ha
->range
)[1];
564 memset(ha
, 0, ha_size
);
565 ha
->hdr
.type
= DM_MEM_HOT_ADD_REQUEST
;
566 ha
->hdr
.size
= ha_size
;
567 ha
->hdr
.trans_id
= balloon
->trans_id
;
569 ha
->range
.finfo
.start_page
= hot_add_range
->start
;
570 ha
->range
.finfo
.page_cnt
= *current_count
;
571 ha_region
->finfo
.start_page
= hot_add_range
->start
;
572 ha_region
->finfo
.page_cnt
= ha
->range
.finfo
.page_cnt
;
574 trace_hv_balloon_outgoing_hot_add(ha
->hdr
.trans_id
,
575 *current_count
, hot_add_range
->start
);
577 ret
= vmbus_channel_send(chan
, VMBUS_PACKET_DATA_INBAND
,
578 NULL
, 0, ha
, ha_size
, false,
581 error_report("error %zd when posting hot add msg, expect problems",
585 HV_BALLOON_STATE_DESC_SET(stdesc
, S_HOT_ADD_REPLY_WAIT
);
588 static void hv_balloon_balloon_rb_wait(HvBalloon
*balloon
, StateDesc
*stdesc
)
590 VMBusChannel
*chan
= hv_balloon_get_channel(balloon
);
591 size_t bl_size
= sizeof(struct dm_balloon
);
593 assert(balloon
->state
== S_BALLOON_RB_WAIT
);
595 if (vmbus_channel_reserve(chan
, 0, bl_size
) < 0) {
599 HV_BALLOON_STATE_DESC_SET(stdesc
, S_BALLOON_POSTING
);
602 static void hv_balloon_balloon_posting(HvBalloon
*balloon
, StateDesc
*stdesc
)
604 VMBusChannel
*chan
= hv_balloon_get_channel(balloon
);
605 struct dm_balloon bl
;
606 size_t bl_size
= sizeof(bl
);
609 assert(balloon
->state
== S_BALLOON_POSTING
);
610 assert(balloon
->balloon_diff
> 0);
612 memset(&bl
, 0, sizeof(bl
));
613 bl
.hdr
.type
= DM_BALLOON_REQUEST
;
614 bl
.hdr
.size
= bl_size
;
615 bl
.hdr
.trans_id
= balloon
->trans_id
;
616 bl
.num_pages
= MIN(balloon
->balloon_diff
, HV_BALLOON_HR_CHUNK_PAGES
);
618 trace_hv_balloon_outgoing_balloon(bl
.hdr
.trans_id
, bl
.num_pages
,
619 balloon
->balloon_diff
);
621 ret
= vmbus_channel_send(chan
, VMBUS_PACKET_DATA_INBAND
,
622 NULL
, 0, &bl
, bl_size
, false,
625 error_report("error %zd when posting balloon msg, expect problems",
629 HV_BALLOON_STATE_DESC_SET(stdesc
, S_BALLOON_REPLY_WAIT
);
632 static void hv_balloon_idle_state_process_target(HvBalloon
*balloon
,
635 bool can_balloon
= balloon
->caps
.cap_bits
.balloon
;
636 uint64_t ram_size_pages
, total_removed
;
638 ram_size_pages
= hv_balloon_total_ram(balloon
);
639 total_removed
= hv_balloon_total_removed_rs(balloon
, ram_size_pages
);
642 * we need to cache the values computed from the balloon target value when
643 * starting the adjustment procedure in case someone changes the target when
644 * the procedure is in progress
646 if (balloon
->target
> ram_size_pages
- total_removed
) {
647 bool can_hot_add
= balloon
->caps
.cap_bits
.hot_add
;
648 uint64_t target_diff
= balloon
->target
-
649 (ram_size_pages
- total_removed
);
651 balloon
->unballoon_diff
= MIN(target_diff
, total_removed
);
654 balloon
->hot_add_diff
= target_diff
- balloon
->unballoon_diff
;
656 balloon
->hot_add_diff
= 0;
659 if (balloon
->unballoon_diff
> 0) {
661 HV_BALLOON_STATE_DESC_SET(stdesc
, S_UNBALLOON_RB_WAIT
);
662 } else if (balloon
->hot_add_diff
> 0) {
663 HV_BALLOON_STATE_DESC_SET(stdesc
, S_HOT_ADD_SETUP
);
665 } else if (can_balloon
&&
666 balloon
->target
< ram_size_pages
- total_removed
) {
667 balloon
->balloon_diff
= ram_size_pages
- total_removed
-
669 HV_BALLOON_STATE_DESC_SET(stdesc
, S_BALLOON_RB_WAIT
);
673 static void hv_balloon_idle_state(HvBalloon
*balloon
,
676 assert(balloon
->state
== S_IDLE
);
678 if (balloon
->target_changed
) {
679 balloon
->target_changed
= false;
680 hv_balloon_idle_state_process_target(balloon
, stdesc
);
685 static const struct {
686 void (*handler
)(HvBalloon
*balloon
, StateDesc
*stdesc
);
687 } state_handlers
[] = {
688 [S_IDLE
].handler
= hv_balloon_idle_state
,
689 [S_BALLOON_POSTING
].handler
= hv_balloon_balloon_posting
,
690 [S_BALLOON_RB_WAIT
].handler
= hv_balloon_balloon_rb_wait
,
691 [S_UNBALLOON_POSTING
].handler
= hv_balloon_unballoon_posting
,
692 [S_UNBALLOON_RB_WAIT
].handler
= hv_balloon_unballoon_rb_wait
,
693 [S_HOT_ADD_SETUP
].handler
= hv_balloon_hot_add_setup
,
694 [S_HOT_ADD_RB_WAIT
].handler
= hv_balloon_hot_add_rb_wait
,
695 [S_HOT_ADD_POSTING
].handler
= hv_balloon_hot_add_posting
,
698 static void hv_balloon_handle_state(HvBalloon
*balloon
, StateDesc
*stdesc
)
700 if (balloon
->state
>= ARRAY_SIZE(state_handlers
) ||
701 !state_handlers
[balloon
->state
].handler
) {
705 state_handlers
[balloon
->state
].handler(balloon
, stdesc
);
708 static void hv_balloon_remove_response_insert_range(PageRangeTree tree
,
709 const PageRange
*range
,
714 uint64_t dupcount
, effcount
;
716 if (range
->count
== 0) {
721 hvb_page_range_tree_insert(tree
, range
->start
, range
->count
, &dupcount
);
723 assert(dupcount
<= range
->count
);
724 effcount
= range
->count
- dupcount
;
733 static void hv_balloon_remove_response_handle_range(HvBalloon
*balloon
,
736 uint64_t *removedctr
)
738 OurRange
*our_range
= OUR_RANGE(balloon
->our_range
);
739 PageRangeTree globaltree
=
740 both
? balloon
->removed_both
: balloon
->removed_guest
;
741 uint64_t *globalctr
=
742 both
? &balloon
->removed_both_ctr
: &balloon
->removed_guest_ctr
;
745 if (range
->count
== 0) {
749 trace_hv_balloon_remove_response(range
->count
, range
->start
, both
);
752 /* Includes the not-yet-hot-added and unusable parts. */
753 rangeeff
= our_range
->range
;
755 rangeeff
.start
= rangeeff
.count
= 0;
758 if (page_range_intersection_size(range
, rangeeff
.start
, rangeeff
.count
) > 0) {
759 PageRangeTree ourtree
= our_range_get_removed_tree(our_range
, both
);
760 PageRange rangehole
, rangecommon
;
761 uint64_t ourremoved
= 0;
763 /* process the hole before our range, if it exists */
764 page_range_part_before(range
, rangeeff
.start
, &rangehole
);
765 hv_balloon_remove_response_insert_range(globaltree
, &rangehole
,
766 globalctr
, removedctr
, NULL
);
767 if (rangehole
.count
> 0) {
768 trace_hv_balloon_remove_response_hole(rangehole
.count
,
770 range
->count
, range
->start
,
771 rangeeff
.start
, both
);
774 /* process our part */
775 page_range_intersect(range
, rangeeff
.start
, rangeeff
.count
,
777 hv_balloon_remove_response_insert_range(ourtree
, &rangecommon
,
778 globalctr
, removedctr
,
780 if (rangecommon
.count
> 0) {
781 trace_hv_balloon_remove_response_common(rangecommon
.count
,
783 range
->count
, range
->start
,
785 rangeeff
.start
, ourremoved
,
789 /* calculate what's left after our range */
790 rangecommon
= *range
;
791 page_range_part_after(&rangecommon
, rangeeff
.start
, rangeeff
.count
,
795 /* process the remainder of the range that lies after our range */
796 if (range
->count
> 0) {
797 hv_balloon_remove_response_insert_range(globaltree
, range
,
798 globalctr
, removedctr
, NULL
);
799 trace_hv_balloon_remove_response_remainder(range
->count
, range
->start
,
805 static void hv_balloon_remove_response_handle_pages(HvBalloon
*balloon
,
810 uint64_t *removedctr
)
815 * if there is an existing range that the new range can't be joined to
816 * dump it into tree(s)
818 if (range
->count
> 0 && !page_range_joinable(range
, start
, count
)) {
819 hv_balloon_remove_response_handle_range(balloon
, range
, both
,
823 if (range
->count
== 0) {
824 range
->start
= start
;
825 range
->count
= count
;
826 } else if (page_range_joinable_left(range
, start
, count
)) {
827 range
->start
= start
;
828 range
->count
+= count
;
829 } else { /* page_range_joinable_right() */
830 range
->count
+= count
;
834 static gboolean
hv_balloon_handle_remove_host_addr_node(gpointer key
,
838 PageRange
*range
= value
;
841 for (pageoff
= 0; pageoff
< range
->count
; ) {
842 uint64_t addr_64
= (range
->start
+ pageoff
) * HV_BALLOON_PAGE_SIZE
;
845 ram_addr_t rb_offset
;
849 assert(addr_64
<= UINTPTR_MAX
);
850 addr
= (void *)((uintptr_t)addr_64
);
851 rb
= qemu_ram_block_from_host(addr
, false, &rb_offset
);
852 rb_page_size
= qemu_ram_pagesize(rb
);
854 if (rb_page_size
!= HV_BALLOON_PAGE_SIZE
) {
855 /* TODO: these should end in "removed_guest" */
856 warn_report("guest reported removed page backed by unsupported page size %zu",
862 discard_size
= MIN(range
->count
- pageoff
,
863 (rb
->max_length
- rb_offset
) /
864 HV_BALLOON_PAGE_SIZE
);
865 discard_size
= MAX(discard_size
, 1);
867 if (ram_block_discard_range(rb
, rb_offset
, discard_size
*
868 HV_BALLOON_PAGE_SIZE
) != 0) {
869 warn_report("guest reported removed page failed discard");
872 pageoff
+= discard_size
;
878 static void hv_balloon_handle_remove_host_addr_tree(PageRangeTree tree
)
880 g_tree_foreach(tree
.t
, hv_balloon_handle_remove_host_addr_node
, NULL
);
883 static int hv_balloon_handle_remove_section(PageRangeTree tree
,
884 const MemoryRegionSection
*section
,
887 void *addr
= memory_region_get_ram_ptr(section
->mr
) +
888 section
->offset_within_region
;
893 if ((uintptr_t)addr
% HV_BALLOON_PAGE_SIZE
) {
894 warn_report("guest reported removed pages at an unaligned host addr %p",
899 addr_page
= (uintptr_t)addr
/ HV_BALLOON_PAGE_SIZE
;
900 hvb_page_range_tree_insert(tree
, addr_page
, count
, NULL
);
905 static void hv_balloon_handle_remove_ranges(HvBalloon
*balloon
,
906 union dm_mem_page_range ranges
[],
910 PageRangeTree removed_host_addr
;
911 PageRange range_guest
, range_both
;
913 hvb_page_range_tree_init(&removed_host_addr
);
914 range_guest
.count
= range_both
.count
= removedcnt
= 0;
915 for (unsigned int ctr
= 0; ctr
< count
; ctr
++) {
916 union dm_mem_page_range
*mr
= &ranges
[ctr
];
918 MemoryRegionSection section
;
920 for (unsigned int offset
= 0; offset
< mr
->finfo
.page_cnt
; ) {
922 uint64_t pageno
= mr
->finfo
.start_page
+ offset
;
923 uint64_t pagecnt
= 1;
925 pa
= (hwaddr
)pageno
<< HV_BALLOON_PFN_SHIFT
;
926 section
= memory_region_find(get_system_memory(), pa
,
927 (mr
->finfo
.page_cnt
- offset
) *
928 HV_BALLOON_PAGE_SIZE
);
930 warn_report("guest reported removed page %"PRIu64
" not found in RAM",
936 pagecnt
= int128_get64(section
.size
) / HV_BALLOON_PAGE_SIZE
;
938 warn_report("guest reported removed page %"PRIu64
" in a section smaller than page size",
940 pagecnt
= 1; /* skip the whole page */
945 if (!memory_region_is_ram(section
.mr
) ||
946 memory_region_is_rom(section
.mr
) ||
947 memory_region_is_romd(section
.mr
)) {
948 warn_report("guest reported removed page %"PRIu64
" in a section that is not an ordinary RAM",
954 ret
= hv_balloon_handle_remove_section(removed_host_addr
, §ion
,
959 hv_balloon_remove_response_handle_pages(balloon
,
964 hv_balloon_remove_response_handle_pages(balloon
,
971 memory_region_unref(section
.mr
);
978 hv_balloon_remove_response_handle_range(balloon
, &range_both
, true,
980 hv_balloon_remove_response_handle_range(balloon
, &range_guest
, false,
983 hv_balloon_handle_remove_host_addr_tree(removed_host_addr
);
984 hvb_page_range_tree_destroy(&removed_host_addr
);
986 if (removedcnt
> balloon
->balloon_diff
) {
987 warn_report("guest reported more pages removed than currently pending (%"PRIu64
" vs %"PRIu64
")",
988 removedcnt
, balloon
->balloon_diff
);
989 balloon
->balloon_diff
= 0;
991 balloon
->balloon_diff
-= removedcnt
;
995 static bool hv_balloon_handle_msg_size(HvBalloonReq
*req
, size_t minsize
,
998 VMBusChanReq
*vmreq
= &req
->vmreq
;
999 uint32_t msglen
= vmreq
->msglen
;
1001 if (msglen
>= minsize
) {
1005 warn_report("%s message too short (%u vs %zu), ignoring", msgname
,
1006 (unsigned int)msglen
, minsize
);
1010 static void hv_balloon_handle_version_request(HvBalloon
*balloon
,
1014 VMBusChanReq
*vmreq
= &req
->vmreq
;
1015 struct dm_version_request
*msgVr
= vmreq
->msg
;
1016 struct dm_version_response respVr
;
1018 if (balloon
->state
!= S_VERSION
) {
1019 warn_report("unexpected DM_VERSION_REQUEST in %d state",
1024 if (!hv_balloon_handle_msg_size(req
, sizeof(*msgVr
),
1025 "DM_VERSION_REQUEST")) {
1029 trace_hv_balloon_incoming_version(msgVr
->version
.major_version
,
1030 msgVr
->version
.minor_version
);
1032 memset(&respVr
, 0, sizeof(respVr
));
1033 respVr
.hdr
.type
= DM_VERSION_RESPONSE
;
1034 respVr
.hdr
.size
= sizeof(respVr
);
1035 respVr
.hdr
.trans_id
= msgVr
->hdr
.trans_id
;
1036 respVr
.is_accepted
= msgVr
->version
.version
>= DYNMEM_PROTOCOL_VERSION_1
&&
1037 msgVr
->version
.version
<= DYNMEM_PROTOCOL_VERSION_3
;
1039 hv_balloon_send_packet(vmreq
->chan
, (struct dm_message
*)&respVr
);
1041 if (respVr
.is_accepted
) {
1042 HV_BALLOON_STATE_DESC_SET(stdesc
, S_CAPS
);
1046 static void hv_balloon_handle_caps_report(HvBalloon
*balloon
,
1050 VMBusChanReq
*vmreq
= &req
->vmreq
;
1051 struct dm_capabilities
*msgCap
= vmreq
->msg
;
1052 struct dm_capabilities_resp_msg respCap
;
1054 if (balloon
->state
!= S_CAPS
) {
1055 warn_report("unexpected DM_CAPABILITIES_REPORT in %d state",
1060 if (!hv_balloon_handle_msg_size(req
, sizeof(*msgCap
),
1061 "DM_CAPABILITIES_REPORT")) {
1065 trace_hv_balloon_incoming_caps(msgCap
->caps
.caps
);
1066 balloon
->caps
= msgCap
->caps
;
1068 memset(&respCap
, 0, sizeof(respCap
));
1069 respCap
.hdr
.type
= DM_CAPABILITIES_RESPONSE
;
1070 respCap
.hdr
.size
= sizeof(respCap
);
1071 respCap
.hdr
.trans_id
= msgCap
->hdr
.trans_id
;
1072 respCap
.is_accepted
= 1;
1073 respCap
.hot_remove
= 1;
1074 respCap
.suppress_pressure_reports
= !balloon
->status_report
.enabled
;
1075 hv_balloon_send_packet(vmreq
->chan
, (struct dm_message
*)&respCap
);
1077 timer_mod(&balloon
->post_init_timer
,
1078 qemu_clock_get_ms(QEMU_CLOCK_VIRTUAL
) +
1079 HV_BALLOON_POST_INIT_WAIT
);
1081 HV_BALLOON_STATE_DESC_SET(stdesc
, S_POST_INIT_WAIT
);
1084 static void hv_balloon_handle_status_report(HvBalloon
*balloon
,
1087 VMBusChanReq
*vmreq
= &req
->vmreq
;
1088 struct dm_status
*msgStatus
= vmreq
->msg
;
1090 if (!hv_balloon_handle_msg_size(req
, sizeof(*msgStatus
),
1091 "DM_STATUS_REPORT")) {
1095 if (!balloon
->status_report
.enabled
) {
1099 balloon
->status_report
.committed
= msgStatus
->num_committed
;
1100 balloon
->status_report
.committed
*= HV_BALLOON_PAGE_SIZE
;
1101 balloon
->status_report
.available
= msgStatus
->num_avail
;
1102 balloon
->status_report
.available
*= HV_BALLOON_PAGE_SIZE
;
1103 balloon
->status_report
.received
= true;
1105 qapi_event_send_hv_balloon_status_report(balloon
->status_report
.committed
,
1106 balloon
->status_report
.available
);
1109 HvBalloonInfo
*qmp_query_hv_balloon_status_report(Error
**errp
)
1112 HvBalloonInfo
*info
;
1114 balloon
= HV_BALLOON(object_resolve_path_type("", TYPE_HV_BALLOON
, NULL
));
1116 error_setg(errp
, "no %s device present", TYPE_HV_BALLOON
);
1120 if (!balloon
->status_report
.enabled
) {
1121 error_setg(errp
, "guest memory status reporting not enabled");
1125 if (!balloon
->status_report
.received
) {
1126 error_setg(errp
, "no guest memory status report received yet");
1130 info
= g_malloc0(sizeof(*info
));
1131 info
->committed
= balloon
->status_report
.committed
;
1132 info
->available
= balloon
->status_report
.available
;
1136 static void hv_balloon_handle_unballoon_response(HvBalloon
*balloon
,
1140 VMBusChanReq
*vmreq
= &req
->vmreq
;
1141 struct dm_unballoon_response
*msgUrR
= vmreq
->msg
;
1143 if (balloon
->state
!= S_UNBALLOON_REPLY_WAIT
) {
1144 warn_report("unexpected DM_UNBALLOON_RESPONSE in %d state",
1149 if (!hv_balloon_handle_msg_size(req
, sizeof(*msgUrR
),
1150 "DM_UNBALLOON_RESPONSE"))
1153 trace_hv_balloon_incoming_unballoon(msgUrR
->hdr
.trans_id
);
1155 balloon
->trans_id
++;
1157 if (balloon
->hot_add_diff
> 0) {
1158 bool can_hot_add
= balloon
->caps
.cap_bits
.hot_add
;
1160 assert(can_hot_add
);
1161 HV_BALLOON_STATE_DESC_SET(stdesc
, S_HOT_ADD_SETUP
);
1163 HV_BALLOON_STATE_DESC_SET(stdesc
, S_IDLE
);
1167 static void hv_balloon_handle_hot_add_response(HvBalloon
*balloon
,
1171 PageRange
*hot_add_range
= &balloon
->hot_add_range
;
1172 VMBusChanReq
*vmreq
= &req
->vmreq
;
1173 struct dm_hot_add_response
*msgHaR
= vmreq
->msg
;
1174 OurRange
*our_range
;
1176 if (balloon
->state
!= S_HOT_ADD_REPLY_WAIT
) {
1177 warn_report("unexpected DM_HOT_ADD_RESPONSE in %d state",
1182 assert(balloon
->our_range
);
1183 our_range
= OUR_RANGE(balloon
->our_range
);
1185 if (!hv_balloon_handle_msg_size(req
, sizeof(*msgHaR
),
1186 "DM_HOT_ADD_RESPONSE"))
1189 trace_hv_balloon_incoming_hot_add(msgHaR
->hdr
.trans_id
, msgHaR
->result
,
1190 msgHaR
->page_count
);
1192 balloon
->trans_id
++;
1194 if (msgHaR
->result
) {
1195 if (msgHaR
->page_count
> balloon
->ha_current_count
) {
1196 warn_report("DM_HOT_ADD_RESPONSE page count higher than requested (%"PRIu32
" vs %"PRIu64
")",
1197 msgHaR
->page_count
, balloon
->ha_current_count
);
1198 msgHaR
->page_count
= balloon
->ha_current_count
;
1201 hvb_our_range_mark_added(our_range
, msgHaR
->page_count
);
1202 hot_add_range
->start
+= msgHaR
->page_count
;
1203 hot_add_range
->count
-= msgHaR
->page_count
;
1206 if (!msgHaR
->result
|| msgHaR
->page_count
< balloon
->ha_current_count
) {
1208 * the current planned range was only partially hot-added, take note
1209 * how much of it remains and don't attempt any further hot adds
1211 our_range_mark_remaining_unusable(our_range
);
1216 /* any pages remaining to hot-add in our range? */
1217 if (hot_add_range
->count
> 0) {
1218 HV_BALLOON_STATE_DESC_SET(stdesc
, S_HOT_ADD_RB_WAIT
);
1223 HV_BALLOON_STATE_DESC_SET(stdesc
, S_IDLE
);
1226 static void hv_balloon_handle_balloon_response(HvBalloon
*balloon
,
1230 VMBusChanReq
*vmreq
= &req
->vmreq
;
1231 struct dm_balloon_response
*msgBR
= vmreq
->msg
;
1233 if (balloon
->state
!= S_BALLOON_REPLY_WAIT
) {
1234 warn_report("unexpected DM_BALLOON_RESPONSE in %d state",
1239 if (!hv_balloon_handle_msg_size(req
, sizeof(*msgBR
),
1240 "DM_BALLOON_RESPONSE"))
1243 trace_hv_balloon_incoming_balloon(msgBR
->hdr
.trans_id
, msgBR
->range_count
,
1246 if (vmreq
->msglen
< sizeof(*msgBR
) +
1247 (uint64_t)sizeof(msgBR
->range_array
[0]) * msgBR
->range_count
) {
1248 warn_report("DM_BALLOON_RESPONSE too short for the range count");
1252 if (msgBR
->range_count
== 0) {
1253 /* The guest is already at its minimum size */
1254 balloon
->balloon_diff
= 0;
1257 hv_balloon_handle_remove_ranges(balloon
,
1259 msgBR
->range_count
);
1262 /* More responses expected? */
1263 if (msgBR
->more_pages
) {
1268 balloon
->trans_id
++;
1270 if (balloon
->balloon_diff
> 0) {
1271 HV_BALLOON_STATE_DESC_SET(stdesc
, S_BALLOON_RB_WAIT
);
1273 HV_BALLOON_STATE_DESC_SET(stdesc
, S_IDLE
);
1277 static void hv_balloon_handle_packet(HvBalloon
*balloon
, HvBalloonReq
*req
,
1280 VMBusChanReq
*vmreq
= &req
->vmreq
;
1281 struct dm_message
*msg
= vmreq
->msg
;
1283 if (vmreq
->msglen
< sizeof(msg
->hdr
)) {
1287 switch (msg
->hdr
.type
) {
1288 case DM_VERSION_REQUEST
:
1289 hv_balloon_handle_version_request(balloon
, req
, stdesc
);
1292 case DM_CAPABILITIES_REPORT
:
1293 hv_balloon_handle_caps_report(balloon
, req
, stdesc
);
1296 case DM_STATUS_REPORT
:
1297 hv_balloon_handle_status_report(balloon
, req
);
1300 case DM_MEM_HOT_ADD_RESPONSE
:
1301 hv_balloon_handle_hot_add_response(balloon
, req
, stdesc
);
1304 case DM_UNBALLOON_RESPONSE
:
1305 hv_balloon_handle_unballoon_response(balloon
, req
, stdesc
);
1308 case DM_BALLOON_RESPONSE
:
1309 hv_balloon_handle_balloon_response(balloon
, req
, stdesc
);
1313 warn_report("unknown DM message %u", msg
->hdr
.type
);
1318 static bool hv_balloon_recv_channel(HvBalloon
*balloon
, StateDesc
*stdesc
)
1323 if (balloon
->state
== S_WAIT_RESET
||
1324 balloon
->state
== S_POST_RESET_CLOSED
) {
1328 chan
= hv_balloon_get_channel(balloon
);
1329 if (vmbus_channel_recv_start(chan
)) {
1333 while ((req
= vmbus_channel_recv_peek(chan
, sizeof(*req
)))) {
1334 hv_balloon_handle_packet(balloon
, req
, stdesc
);
1335 vmbus_free_req(req
);
1336 vmbus_channel_recv_pop(chan
);
1338 if (stdesc
->state
!= S_NO_CHANGE
) {
1343 return vmbus_channel_recv_done(chan
) > 0;
1346 /* old state handler -> new state transition (potential) */
1347 static bool hv_balloon_event_loop_state(HvBalloon
*balloon
)
1349 StateDesc state_new
= HV_BALLOON_STATE_DESC_INIT
;
1351 hv_balloon_handle_state(balloon
, &state_new
);
1352 return hv_balloon_state_set(balloon
, state_new
.state
, state_new
.desc
);
1355 /* VMBus message -> new state transition (potential) */
1356 static bool hv_balloon_event_loop_recv(HvBalloon
*balloon
)
1358 StateDesc state_new
= HV_BALLOON_STATE_DESC_INIT
;
1359 bool any_recv
, state_changed
;
1361 any_recv
= hv_balloon_recv_channel(balloon
, &state_new
);
1362 state_changed
= hv_balloon_state_set(balloon
,
1363 state_new
.state
, state_new
.desc
);
1365 return state_changed
|| any_recv
;
1368 static void hv_balloon_event_loop(HvBalloon
*balloon
)
1370 bool state_repeat
, recv_repeat
;
1373 state_repeat
= hv_balloon_event_loop_state(balloon
);
1374 recv_repeat
= hv_balloon_event_loop_recv(balloon
);
1375 } while (state_repeat
|| recv_repeat
);
1378 static void hv_balloon_vmdev_chan_notify(VMBusChannel
*chan
)
1380 HvBalloon
*balloon
= HV_BALLOON(vmbus_channel_device(chan
));
1382 hv_balloon_event_loop(balloon
);
1385 static void hv_balloon_stat(void *opaque
, BalloonInfo
*info
)
1387 HvBalloon
*balloon
= opaque
;
1388 info
->actual
= (hv_balloon_total_ram(balloon
) - balloon
->removed_both_ctr
)
1389 << HV_BALLOON_PFN_SHIFT
;
1392 static void hv_balloon_to_target(void *opaque
, ram_addr_t target
)
1394 HvBalloon
*balloon
= opaque
;
1395 uint64_t target_pages
= target
>> HV_BALLOON_PFN_SHIFT
;
1397 if (!target_pages
) {
1402 * always set target_changed, even with unchanged target, as the user
1403 * might be asking us to try again reaching it
1405 balloon
->target
= target_pages
;
1406 balloon
->target_changed
= true;
1408 hv_balloon_event_loop(balloon
);
1411 static int hv_balloon_vmdev_open_channel(VMBusChannel
*chan
)
1413 HvBalloon
*balloon
= HV_BALLOON(vmbus_channel_device(chan
));
1415 if (balloon
->state
!= S_POST_RESET_CLOSED
) {
1416 warn_report("guest trying to open a DM channel in invalid %d state",
1421 HV_BALLOON_SET_STATE(balloon
, S_VERSION
);
1422 hv_balloon_event_loop(balloon
);
1427 static void hv_balloon_vmdev_close_channel(VMBusChannel
*chan
)
1429 HvBalloon
*balloon
= HV_BALLOON(vmbus_channel_device(chan
));
1431 timer_del(&balloon
->post_init_timer
);
1433 /* Don't report stale data */
1434 balloon
->status_report
.received
= false;
1436 HV_BALLOON_SET_STATE(balloon
, S_WAIT_RESET
);
1437 hv_balloon_event_loop(balloon
);
1440 static void hv_balloon_post_init_timer(void *opaque
)
1442 HvBalloon
*balloon
= opaque
;
1444 if (balloon
->state
!= S_POST_INIT_WAIT
) {
1448 HV_BALLOON_SET_STATE(balloon
, S_IDLE
);
1449 hv_balloon_event_loop(balloon
);
1452 static void hv_balloon_system_reset_unrealize_common(HvBalloon
*balloon
)
1454 g_clear_pointer(&balloon
->our_range
, hvb_our_range_memslots_free
);
1457 static void hv_balloon_system_reset(void *opaque
)
1459 HvBalloon
*balloon
= HV_BALLOON(opaque
);
1461 hv_balloon_system_reset_unrealize_common(balloon
);
1464 static void hv_balloon_ensure_mr(HvBalloon
*balloon
)
1466 MemoryRegion
*hostmem_mr
;
1468 assert(balloon
->hostmem
);
1474 hostmem_mr
= host_memory_backend_get_memory(balloon
->hostmem
);
1476 balloon
->mr
= g_new0(MemoryRegion
, 1);
1477 memory_region_init(balloon
->mr
, OBJECT(balloon
), TYPE_HV_BALLOON
,
1478 memory_region_size(hostmem_mr
));
1481 * The VM can indicate an alignment up to 32 GiB. Memory device core can
1482 * usually only handle/guarantee 1 GiB alignment. The user will have to
1483 * specify a larger maxmem eventually.
1485 * The memory device core will warn the user in case maxmem might have to be
1486 * increased and will fail plugging the device if there is not sufficient
1487 * space after alignment.
1489 * TODO: we could do the alignment ourselves in a slightly bigger region.
1490 * But this feels better, although the warning might be annoying. Maybe
1491 * we can optimize that in the future (e.g., with such a device on the
1492 * cmdline place/size the device memory region differently.
1494 balloon
->mr
->align
= MAX(32 * GiB
, memory_region_get_alignment(hostmem_mr
));
1497 static void hv_balloon_free_mr(HvBalloon
*balloon
)
1503 object_unparent(OBJECT(balloon
->mr
));
1504 g_clear_pointer(&balloon
->mr
, g_free
);
1507 static void hv_balloon_vmdev_realize(VMBusDevice
*vdev
, Error
**errp
)
1510 HvBalloon
*balloon
= HV_BALLOON(vdev
);
1513 balloon
->state
= S_WAIT_RESET
;
1515 ret
= qemu_add_balloon_handler(hv_balloon_to_target
, hv_balloon_stat
,
1518 /* This also protects against having multiple hv-balloon instances */
1519 error_setg(errp
, "Only one balloon device is supported");
1523 if (balloon
->hostmem
) {
1524 if (host_memory_backend_is_mapped(balloon
->hostmem
)) {
1525 Object
*obj
= OBJECT(balloon
->hostmem
);
1527 error_setg(errp
, "'%s' property specifies a busy memdev: %s",
1528 HV_BALLOON_MEMDEV_PROP
,
1529 object_get_canonical_path_component(obj
));
1530 goto out_balloon_handler
;
1533 hv_balloon_ensure_mr(balloon
);
1535 /* This is rather unlikely to happen, but let's still check for it. */
1536 if (!QEMU_IS_ALIGNED(memory_region_size(balloon
->mr
),
1537 HV_BALLOON_PAGE_SIZE
)) {
1538 error_setg(errp
, "'%s' property memdev size has to be a multiple of 0x%" PRIx64
,
1539 HV_BALLOON_MEMDEV_PROP
, (uint64_t)HV_BALLOON_PAGE_SIZE
);
1540 goto out_balloon_handler
;
1543 host_memory_backend_set_mapped(balloon
->hostmem
, true);
1544 vmstate_register_ram(host_memory_backend_get_memory(balloon
->hostmem
),
1546 } else if (balloon
->addr
) {
1547 error_setg(errp
, "'%s' property must not be set without a memdev",
1548 HV_BALLOON_MEMDEV_PROP
);
1549 goto out_balloon_handler
;
1552 timer_init_ms(&balloon
->post_init_timer
, QEMU_CLOCK_VIRTUAL
,
1553 hv_balloon_post_init_timer
, balloon
);
1555 qemu_register_reset(hv_balloon_system_reset
, balloon
);
1559 out_balloon_handler
:
1560 qemu_remove_balloon_handler(balloon
);
1564 * VMBus device reset has to be implemented in case the guest decides to
1565 * disconnect and reconnect to the VMBus without rebooting the whole system.
1567 * However, the hot-added memory can't be removed here as Windows keeps on using
1568 * it until the system is restarted, even after disconnecting from the VMBus.
1570 static void hv_balloon_vmdev_reset(VMBusDevice
*vdev
)
1572 HvBalloon
*balloon
= HV_BALLOON(vdev
);
1574 if (balloon
->state
== S_POST_RESET_CLOSED
) {
1578 if (balloon
->our_range
) {
1579 hvb_our_range_clear_removed_trees(OUR_RANGE(balloon
->our_range
));
1582 hvb_page_range_tree_destroy(&balloon
->removed_guest
);
1583 hvb_page_range_tree_destroy(&balloon
->removed_both
);
1584 hvb_page_range_tree_init(&balloon
->removed_guest
);
1585 hvb_page_range_tree_init(&balloon
->removed_both
);
1587 balloon
->trans_id
= 0;
1588 balloon
->removed_guest_ctr
= 0;
1589 balloon
->removed_both_ctr
= 0;
1591 HV_BALLOON_SET_STATE(balloon
, S_POST_RESET_CLOSED
);
1592 hv_balloon_event_loop(balloon
);
1596 * Clean up things that were (possibly) allocated pre-realization, for example
1597 * from memory_device_pre_plug(), so we don't leak them if the device don't
1598 * actually get realized in the end.
1600 static void hv_balloon_unrealize_finalize_common(HvBalloon
*balloon
)
1602 hv_balloon_free_mr(balloon
);
1605 balloon
->memslot_count
= 0;
1608 static void hv_balloon_vmdev_unrealize(VMBusDevice
*vdev
)
1610 HvBalloon
*balloon
= HV_BALLOON(vdev
);
1612 qemu_unregister_reset(hv_balloon_system_reset
, balloon
);
1614 hv_balloon_system_reset_unrealize_common(balloon
);
1616 qemu_remove_balloon_handler(balloon
);
1618 if (balloon
->hostmem
) {
1619 vmstate_unregister_ram(host_memory_backend_get_memory(balloon
->hostmem
),
1621 host_memory_backend_set_mapped(balloon
->hostmem
, false);
1624 hvb_page_range_tree_destroy(&balloon
->removed_guest
);
1625 hvb_page_range_tree_destroy(&balloon
->removed_both
);
1627 hv_balloon_unrealize_finalize_common(balloon
);
1630 static uint64_t hv_balloon_md_get_addr(const MemoryDeviceState
*md
)
1632 return object_property_get_uint(OBJECT(md
), HV_BALLOON_ADDR_PROP
,
1636 static void hv_balloon_md_set_addr(MemoryDeviceState
*md
, uint64_t addr
,
1639 object_property_set_uint(OBJECT(md
), HV_BALLOON_ADDR_PROP
, addr
, errp
);
1642 static MemoryRegion
*hv_balloon_md_get_memory_region(MemoryDeviceState
*md
,
1645 HvBalloon
*balloon
= HV_BALLOON(md
);
1647 if (!balloon
->hostmem
) {
1651 hv_balloon_ensure_mr(balloon
);
1656 static void hv_balloon_md_fill_device_info(const MemoryDeviceState
*md
,
1657 MemoryDeviceInfo
*info
)
1659 HvBalloonDeviceInfo
*hi
= g_new0(HvBalloonDeviceInfo
, 1);
1660 const HvBalloon
*balloon
= HV_BALLOON(md
);
1661 DeviceState
*dev
= DEVICE(md
);
1664 hi
->id
= g_strdup(dev
->id
);
1667 if (balloon
->hostmem
) {
1668 hi
->memdev
= object_get_canonical_path(OBJECT(balloon
->hostmem
));
1669 hi
->memaddr
= balloon
->addr
;
1670 hi
->has_memaddr
= true;
1671 hi
->max_size
= memory_region_size(balloon
->mr
);
1672 /* TODO: expose current provided size or something else? */
1677 info
->u
.hv_balloon
.data
= hi
;
1678 info
->type
= MEMORY_DEVICE_INFO_KIND_HV_BALLOON
;
1681 static void hv_balloon_decide_memslots(MemoryDeviceState
*md
,
1684 HvBalloon
*balloon
= HV_BALLOON(md
);
1685 MemoryRegion
*hostmem_mr
;
1686 uint64_t region_size
, memslot_size
, memslots
;
1688 /* We're called exactly once, before realizing the device. */
1689 assert(!balloon
->memslot_count
);
1691 /* We should not be called if we don't have a memory backend */
1692 assert(balloon
->hostmem
);
1694 hostmem_mr
= host_memory_backend_get_memory(balloon
->hostmem
);
1695 region_size
= memory_region_size(hostmem_mr
);
1697 assert(region_size
> 0);
1698 memslot_size
= QEMU_ALIGN_UP(region_size
/ limit
,
1699 HV_BALLOON_HA_MEMSLOT_SIZE_ALIGN
);
1700 memslots
= QEMU_ALIGN_UP(region_size
, memslot_size
) / memslot_size
;
1703 balloon
->memslot_size
= memslot_size
;
1705 balloon
->memslot_size
= region_size
;
1708 assert(memslots
<= UINT_MAX
);
1709 balloon
->memslot_count
= memslots
;
1712 static unsigned int hv_balloon_get_memslots(MemoryDeviceState
*md
)
1714 const HvBalloon
*balloon
= HV_BALLOON(md
);
1716 /* We're called after setting the suggested limit. */
1717 assert(balloon
->memslot_count
> 0);
1719 return balloon
->memslot_count
;
1722 static void hv_balloon_init(Object
*obj
)
1726 static void hv_balloon_finalize(Object
*obj
)
1728 HvBalloon
*balloon
= HV_BALLOON(obj
);
1730 hv_balloon_unrealize_finalize_common(balloon
);
1733 static Property hv_balloon_properties
[] = {
1734 DEFINE_PROP_BOOL("status-report", HvBalloon
,
1735 status_report
.enabled
, false),
1737 /* MEMORY_DEVICE props */
1738 DEFINE_PROP_LINK(HV_BALLOON_MEMDEV_PROP
, HvBalloon
, hostmem
,
1739 TYPE_MEMORY_BACKEND
, HostMemoryBackend
*),
1740 DEFINE_PROP_UINT64(HV_BALLOON_ADDR_PROP
, HvBalloon
, addr
, 0),
1742 DEFINE_PROP_END_OF_LIST(),
1745 static void hv_balloon_class_init(ObjectClass
*klass
, void *data
)
1747 DeviceClass
*dc
= DEVICE_CLASS(klass
);
1748 VMBusDeviceClass
*vdc
= VMBUS_DEVICE_CLASS(klass
);
1749 MemoryDeviceClass
*mdc
= MEMORY_DEVICE_CLASS(klass
);
1751 device_class_set_props(dc
, hv_balloon_properties
);
1752 qemu_uuid_parse(HV_BALLOON_GUID
, &vdc
->classid
);
1753 set_bit(DEVICE_CATEGORY_MISC
, dc
->categories
);
1755 vdc
->vmdev_realize
= hv_balloon_vmdev_realize
;
1756 vdc
->vmdev_unrealize
= hv_balloon_vmdev_unrealize
;
1757 vdc
->vmdev_reset
= hv_balloon_vmdev_reset
;
1758 vdc
->open_channel
= hv_balloon_vmdev_open_channel
;
1759 vdc
->close_channel
= hv_balloon_vmdev_close_channel
;
1760 vdc
->chan_notify_cb
= hv_balloon_vmdev_chan_notify
;
1762 mdc
->get_addr
= hv_balloon_md_get_addr
;
1763 mdc
->set_addr
= hv_balloon_md_set_addr
;
1764 mdc
->get_plugged_size
= memory_device_get_region_size
;
1765 mdc
->get_memory_region
= hv_balloon_md_get_memory_region
;
1766 mdc
->decide_memslots
= hv_balloon_decide_memslots
;
1767 mdc
->get_memslots
= hv_balloon_get_memslots
;
1768 mdc
->fill_device_info
= hv_balloon_md_fill_device_info
;