1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Virtio balloon implementation, inspired by Dor Laor and Marcelo
4 * Tosatti's implementations.
6 * Copyright 2008 Rusty Russell IBM Corporation
9 #include <linux/virtio.h>
10 #include <linux/virtio_balloon.h>
11 #include <linux/swap.h>
12 #include <linux/workqueue.h>
13 #include <linux/delay.h>
14 #include <linux/slab.h>
15 #include <linux/module.h>
16 #include <linux/balloon_compaction.h>
17 #include <linux/oom.h>
18 #include <linux/wait.h>
20 #include <linux/page_reporting.h>
23 * Balloon device works in 4K page units. So each page is pointed to by
24 * multiple balloon pages. All memory counters in this driver are in balloon
27 #define VIRTIO_BALLOON_PAGES_PER_PAGE (unsigned int)(PAGE_SIZE >> VIRTIO_BALLOON_PFN_SHIFT)
28 #define VIRTIO_BALLOON_ARRAY_PFNS_MAX 256
29 /* Maximum number of (4k) pages to deflate on OOM notifications. */
30 #define VIRTIO_BALLOON_OOM_NR_PAGES 256
31 #define VIRTIO_BALLOON_OOM_NOTIFY_PRIORITY 80
33 #define VIRTIO_BALLOON_FREE_PAGE_ALLOC_FLAG (__GFP_NORETRY | __GFP_NOWARN | \
35 /* The order of free page blocks to report to host */
36 #define VIRTIO_BALLOON_HINT_BLOCK_ORDER MAX_PAGE_ORDER
37 /* The size of a free page block in bytes */
38 #define VIRTIO_BALLOON_HINT_BLOCK_BYTES \
39 (1 << (VIRTIO_BALLOON_HINT_BLOCK_ORDER + PAGE_SHIFT))
40 #define VIRTIO_BALLOON_HINT_BLOCK_PAGES (1 << VIRTIO_BALLOON_HINT_BLOCK_ORDER)
42 enum virtio_balloon_vq
{
43 VIRTIO_BALLOON_VQ_INFLATE
,
44 VIRTIO_BALLOON_VQ_DEFLATE
,
45 VIRTIO_BALLOON_VQ_STATS
,
46 VIRTIO_BALLOON_VQ_FREE_PAGE
,
47 VIRTIO_BALLOON_VQ_REPORTING
,
51 enum virtio_balloon_config_read
{
52 VIRTIO_BALLOON_CONFIG_READ_CMD_ID
= 0,
55 struct virtio_balloon
{
56 struct virtio_device
*vdev
;
57 struct virtqueue
*inflate_vq
, *deflate_vq
, *stats_vq
, *free_page_vq
;
59 /* Balloon's own wq for cpu-intensive work items */
60 struct workqueue_struct
*balloon_wq
;
61 /* The free page reporting work item submitted to the balloon wq */
62 struct work_struct report_free_page_work
;
64 /* The balloon servicing is delegated to a freezable workqueue. */
65 struct work_struct update_balloon_stats_work
;
66 struct work_struct update_balloon_size_work
;
68 /* Prevent updating balloon when it is being canceled. */
69 spinlock_t stop_update_lock
;
71 /* Bitmap to indicate if reading the related config fields are needed */
72 unsigned long config_read_bitmap
;
74 /* The list of allocated free pages, waiting to be given back to mm */
75 struct list_head free_page_list
;
76 spinlock_t free_page_list_lock
;
77 /* The number of free page blocks on the above list */
78 unsigned long num_free_page_blocks
;
80 * The cmd id received from host.
81 * Read it via virtio_balloon_cmd_id_received to get the latest value
84 u32 cmd_id_received_cache
;
85 /* The cmd id that is actively in use */
86 __virtio32 cmd_id_active
;
87 /* Buffer to store the stop sign */
88 __virtio32 cmd_id_stop
;
90 /* Waiting for host to ack the pages we released. */
91 wait_queue_head_t acked
;
93 /* Number of balloon pages we've told the Host we're not using. */
94 unsigned int num_pages
;
96 * The pages we've told the Host we're not using are enqueued
97 * at vb_dev_info->pages list.
98 * Each page on this list adds VIRTIO_BALLOON_PAGES_PER_PAGE
101 struct balloon_dev_info vb_dev_info
;
103 /* Synchronize access/update to this struct virtio_balloon elements */
104 struct mutex balloon_lock
;
106 /* The array of pfns we tell the Host about. */
107 unsigned int num_pfns
;
108 __virtio32 pfns
[VIRTIO_BALLOON_ARRAY_PFNS_MAX
];
110 /* Memory statistics */
111 struct virtio_balloon_stat stats
[VIRTIO_BALLOON_S_NR
];
113 /* Shrinker to return free pages - VIRTIO_BALLOON_F_FREE_PAGE_HINT */
114 struct shrinker
*shrinker
;
116 /* OOM notifier to deflate on OOM - VIRTIO_BALLOON_F_DEFLATE_ON_OOM */
117 struct notifier_block oom_nb
;
119 /* Free page reporting device */
120 struct virtqueue
*reporting_vq
;
121 struct page_reporting_dev_info pr_dev_info
;
123 /* State for keeping the wakeup_source active while adjusting the balloon */
124 spinlock_t adjustment_lock
;
125 bool adjustment_signal_pending
;
126 bool adjustment_in_progress
;
129 static const struct virtio_device_id id_table
[] = {
130 { VIRTIO_ID_BALLOON
, VIRTIO_DEV_ANY_ID
},
134 static u32
page_to_balloon_pfn(struct page
*page
)
136 unsigned long pfn
= page_to_pfn(page
);
138 BUILD_BUG_ON(PAGE_SHIFT
< VIRTIO_BALLOON_PFN_SHIFT
);
139 /* Convert pfn from Linux page size to balloon page size. */
140 return pfn
* VIRTIO_BALLOON_PAGES_PER_PAGE
;
143 static void balloon_ack(struct virtqueue
*vq
)
145 struct virtio_balloon
*vb
= vq
->vdev
->priv
;
150 static void tell_host(struct virtio_balloon
*vb
, struct virtqueue
*vq
)
152 struct scatterlist sg
;
155 sg_init_one(&sg
, vb
->pfns
, sizeof(vb
->pfns
[0]) * vb
->num_pfns
);
157 /* We should always be able to add one buffer to an empty queue. */
158 virtqueue_add_outbuf(vq
, &sg
, 1, vb
, GFP_KERNEL
);
161 /* When host has read buffer, this completes via balloon_ack */
162 wait_event(vb
->acked
, virtqueue_get_buf(vq
, &len
));
166 static int virtballoon_free_page_report(struct page_reporting_dev_info
*pr_dev_info
,
167 struct scatterlist
*sg
, unsigned int nents
)
169 struct virtio_balloon
*vb
=
170 container_of(pr_dev_info
, struct virtio_balloon
, pr_dev_info
);
171 struct virtqueue
*vq
= vb
->reporting_vq
;
172 unsigned int unused
, err
;
174 /* We should always be able to add these buffers to an empty queue. */
175 err
= virtqueue_add_inbuf(vq
, sg
, nents
, vb
, GFP_NOWAIT
| __GFP_NOWARN
);
178 * In the extremely unlikely case that something has occurred and we
179 * are able to trigger an error we will simply display a warning
180 * and exit without actually processing the pages.
182 if (WARN_ON_ONCE(err
))
187 /* When host has read buffer, this completes via balloon_ack */
188 wait_event(vb
->acked
, virtqueue_get_buf(vq
, &unused
));
193 static void set_page_pfns(struct virtio_balloon
*vb
,
194 __virtio32 pfns
[], struct page
*page
)
198 BUILD_BUG_ON(VIRTIO_BALLOON_PAGES_PER_PAGE
> VIRTIO_BALLOON_ARRAY_PFNS_MAX
);
201 * Set balloon pfns pointing at this page.
202 * Note that the first pfn points at start of the page.
204 for (i
= 0; i
< VIRTIO_BALLOON_PAGES_PER_PAGE
; i
++)
205 pfns
[i
] = cpu_to_virtio32(vb
->vdev
,
206 page_to_balloon_pfn(page
) + i
);
209 static unsigned int fill_balloon(struct virtio_balloon
*vb
, size_t num
)
211 unsigned int num_allocated_pages
;
212 unsigned int num_pfns
;
216 /* We can only do one array worth at a time. */
217 num
= min(num
, ARRAY_SIZE(vb
->pfns
));
219 for (num_pfns
= 0; num_pfns
< num
;
220 num_pfns
+= VIRTIO_BALLOON_PAGES_PER_PAGE
) {
221 struct page
*page
= balloon_page_alloc();
224 dev_info_ratelimited(&vb
->vdev
->dev
,
225 "Out of puff! Can't get %u pages\n",
226 VIRTIO_BALLOON_PAGES_PER_PAGE
);
227 /* Sleep for at least 1/5 of a second before retry. */
232 balloon_page_push(&pages
, page
);
235 mutex_lock(&vb
->balloon_lock
);
239 while ((page
= balloon_page_pop(&pages
))) {
240 balloon_page_enqueue(&vb
->vb_dev_info
, page
);
242 set_page_pfns(vb
, vb
->pfns
+ vb
->num_pfns
, page
);
243 vb
->num_pages
+= VIRTIO_BALLOON_PAGES_PER_PAGE
;
244 if (!virtio_has_feature(vb
->vdev
,
245 VIRTIO_BALLOON_F_DEFLATE_ON_OOM
))
246 adjust_managed_page_count(page
, -1);
247 vb
->num_pfns
+= VIRTIO_BALLOON_PAGES_PER_PAGE
;
250 num_allocated_pages
= vb
->num_pfns
;
251 /* Did we get any? */
252 if (vb
->num_pfns
!= 0)
253 tell_host(vb
, vb
->inflate_vq
);
254 mutex_unlock(&vb
->balloon_lock
);
256 return num_allocated_pages
;
259 static void release_pages_balloon(struct virtio_balloon
*vb
,
260 struct list_head
*pages
)
262 struct page
*page
, *next
;
264 list_for_each_entry_safe(page
, next
, pages
, lru
) {
265 if (!virtio_has_feature(vb
->vdev
,
266 VIRTIO_BALLOON_F_DEFLATE_ON_OOM
))
267 adjust_managed_page_count(page
, 1);
268 list_del(&page
->lru
);
269 put_page(page
); /* balloon reference */
273 static unsigned int leak_balloon(struct virtio_balloon
*vb
, size_t num
)
275 unsigned int num_freed_pages
;
277 struct balloon_dev_info
*vb_dev_info
= &vb
->vb_dev_info
;
280 /* We can only do one array worth at a time. */
281 num
= min(num
, ARRAY_SIZE(vb
->pfns
));
283 mutex_lock(&vb
->balloon_lock
);
284 /* We can't release more pages than taken */
285 num
= min(num
, (size_t)vb
->num_pages
);
286 for (vb
->num_pfns
= 0; vb
->num_pfns
< num
;
287 vb
->num_pfns
+= VIRTIO_BALLOON_PAGES_PER_PAGE
) {
288 page
= balloon_page_dequeue(vb_dev_info
);
291 set_page_pfns(vb
, vb
->pfns
+ vb
->num_pfns
, page
);
292 list_add(&page
->lru
, &pages
);
293 vb
->num_pages
-= VIRTIO_BALLOON_PAGES_PER_PAGE
;
296 num_freed_pages
= vb
->num_pfns
;
299 * virtio_has_feature(vdev, VIRTIO_BALLOON_F_MUST_TELL_HOST);
300 * is true, we *have* to do it in this order
302 if (vb
->num_pfns
!= 0)
303 tell_host(vb
, vb
->deflate_vq
);
304 release_pages_balloon(vb
, &pages
);
305 mutex_unlock(&vb
->balloon_lock
);
306 return num_freed_pages
;
309 static inline void update_stat(struct virtio_balloon
*vb
, int idx
,
312 BUG_ON(idx
>= VIRTIO_BALLOON_S_NR
);
313 vb
->stats
[idx
].tag
= cpu_to_virtio16(vb
->vdev
, tag
);
314 vb
->stats
[idx
].val
= cpu_to_virtio64(vb
->vdev
, val
);
317 #define pages_to_bytes(x) ((u64)(x) << PAGE_SHIFT)
319 static unsigned int update_balloon_stats(struct virtio_balloon
*vb
)
321 unsigned long events
[NR_VM_EVENT_ITEMS
];
323 unsigned int idx
= 0;
325 unsigned long caches
;
327 all_vm_events(events
);
330 available
= si_mem_available();
331 caches
= global_node_page_state(NR_FILE_PAGES
);
333 #ifdef CONFIG_VM_EVENT_COUNTERS
334 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_SWAP_IN
,
335 pages_to_bytes(events
[PSWPIN
]));
336 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_SWAP_OUT
,
337 pages_to_bytes(events
[PSWPOUT
]));
338 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_MAJFLT
, events
[PGMAJFAULT
]);
339 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_MINFLT
, events
[PGFAULT
]);
340 #ifdef CONFIG_HUGETLB_PAGE
341 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_HTLB_PGALLOC
,
342 events
[HTLB_BUDDY_PGALLOC
]);
343 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_HTLB_PGFAIL
,
344 events
[HTLB_BUDDY_PGALLOC_FAIL
]);
347 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_MEMFREE
,
348 pages_to_bytes(i
.freeram
));
349 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_MEMTOT
,
350 pages_to_bytes(i
.totalram
));
351 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_AVAIL
,
352 pages_to_bytes(available
));
353 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_CACHES
,
354 pages_to_bytes(caches
));
360 * While most virtqueues communicate guest-initiated requests to the hypervisor,
361 * the stats queue operates in reverse. The driver initializes the virtqueue
362 * with a single buffer. From that point forward, all conversations consist of
363 * a hypervisor request (a call to this function) which directs us to refill
364 * the virtqueue with a fresh stats buffer. Since stats collection can sleep,
365 * we delegate the job to a freezable workqueue that will do the actual work via
366 * stats_handle_request().
368 static void stats_request(struct virtqueue
*vq
)
370 struct virtio_balloon
*vb
= vq
->vdev
->priv
;
372 spin_lock(&vb
->stop_update_lock
);
373 if (!vb
->stop_update
)
374 queue_work(system_freezable_wq
, &vb
->update_balloon_stats_work
);
375 spin_unlock(&vb
->stop_update_lock
);
378 static void stats_handle_request(struct virtio_balloon
*vb
)
380 struct virtqueue
*vq
;
381 struct scatterlist sg
;
382 unsigned int len
, num_stats
;
384 num_stats
= update_balloon_stats(vb
);
387 if (!virtqueue_get_buf(vq
, &len
))
389 sg_init_one(&sg
, vb
->stats
, sizeof(vb
->stats
[0]) * num_stats
);
390 virtqueue_add_outbuf(vq
, &sg
, 1, vb
, GFP_KERNEL
);
394 static inline s64
towards_target(struct virtio_balloon
*vb
)
399 /* Legacy balloon config space is LE, unlike all other devices. */
400 virtio_cread_le(vb
->vdev
, struct virtio_balloon_config
, num_pages
,
404 * Aligned up to guest page size to avoid inflating and deflating
407 target
= ALIGN(num_pages
, VIRTIO_BALLOON_PAGES_PER_PAGE
);
408 return target
- vb
->num_pages
;
411 /* Gives back @num_to_return blocks of free pages to mm. */
412 static unsigned long return_free_pages_to_mm(struct virtio_balloon
*vb
,
413 unsigned long num_to_return
)
416 unsigned long num_returned
;
418 spin_lock_irq(&vb
->free_page_list_lock
);
419 for (num_returned
= 0; num_returned
< num_to_return
; num_returned
++) {
420 page
= balloon_page_pop(&vb
->free_page_list
);
423 free_pages((unsigned long)page_address(page
),
424 VIRTIO_BALLOON_HINT_BLOCK_ORDER
);
426 vb
->num_free_page_blocks
-= num_returned
;
427 spin_unlock_irq(&vb
->free_page_list_lock
);
432 static void virtio_balloon_queue_free_page_work(struct virtio_balloon
*vb
)
434 if (!virtio_has_feature(vb
->vdev
, VIRTIO_BALLOON_F_FREE_PAGE_HINT
))
437 /* No need to queue the work if the bit was already set. */
438 if (test_and_set_bit(VIRTIO_BALLOON_CONFIG_READ_CMD_ID
,
439 &vb
->config_read_bitmap
))
442 queue_work(vb
->balloon_wq
, &vb
->report_free_page_work
);
445 static void start_update_balloon_size(struct virtio_balloon
*vb
)
449 spin_lock_irqsave(&vb
->adjustment_lock
, flags
);
450 vb
->adjustment_signal_pending
= true;
451 if (!vb
->adjustment_in_progress
) {
452 vb
->adjustment_in_progress
= true;
453 pm_stay_awake(vb
->vdev
->dev
.parent
);
455 spin_unlock_irqrestore(&vb
->adjustment_lock
, flags
);
457 queue_work(system_freezable_wq
, &vb
->update_balloon_size_work
);
460 static void end_update_balloon_size(struct virtio_balloon
*vb
)
462 spin_lock_irq(&vb
->adjustment_lock
);
463 if (!vb
->adjustment_signal_pending
&& vb
->adjustment_in_progress
) {
464 vb
->adjustment_in_progress
= false;
465 pm_relax(vb
->vdev
->dev
.parent
);
467 spin_unlock_irq(&vb
->adjustment_lock
);
470 static void virtballoon_changed(struct virtio_device
*vdev
)
472 struct virtio_balloon
*vb
= vdev
->priv
;
475 spin_lock_irqsave(&vb
->stop_update_lock
, flags
);
476 if (!vb
->stop_update
) {
477 start_update_balloon_size(vb
);
478 virtio_balloon_queue_free_page_work(vb
);
480 spin_unlock_irqrestore(&vb
->stop_update_lock
, flags
);
483 static void update_balloon_size(struct virtio_balloon
*vb
)
485 u32 actual
= vb
->num_pages
;
487 /* Legacy balloon config space is LE, unlike all other devices. */
488 virtio_cwrite_le(vb
->vdev
, struct virtio_balloon_config
, actual
,
492 static void update_balloon_stats_func(struct work_struct
*work
)
494 struct virtio_balloon
*vb
;
496 vb
= container_of(work
, struct virtio_balloon
,
497 update_balloon_stats_work
);
498 stats_handle_request(vb
);
501 static void update_balloon_size_func(struct work_struct
*work
)
503 struct virtio_balloon
*vb
;
506 vb
= container_of(work
, struct virtio_balloon
,
507 update_balloon_size_work
);
509 spin_lock_irq(&vb
->adjustment_lock
);
510 vb
->adjustment_signal_pending
= false;
511 spin_unlock_irq(&vb
->adjustment_lock
);
513 diff
= towards_target(vb
);
517 diff
-= fill_balloon(vb
, diff
);
519 diff
+= leak_balloon(vb
, -diff
);
520 update_balloon_size(vb
);
524 queue_work(system_freezable_wq
, work
);
526 end_update_balloon_size(vb
);
529 static int init_vqs(struct virtio_balloon
*vb
)
531 struct virtqueue
*vqs
[VIRTIO_BALLOON_VQ_MAX
];
532 vq_callback_t
*callbacks
[VIRTIO_BALLOON_VQ_MAX
];
533 const char *names
[VIRTIO_BALLOON_VQ_MAX
];
537 * Inflateq and deflateq are used unconditionally. The names[]
538 * will be NULL if the related feature is not enabled, which will
539 * cause no allocation for the corresponding virtqueue in find_vqs.
541 callbacks
[VIRTIO_BALLOON_VQ_INFLATE
] = balloon_ack
;
542 names
[VIRTIO_BALLOON_VQ_INFLATE
] = "inflate";
543 callbacks
[VIRTIO_BALLOON_VQ_DEFLATE
] = balloon_ack
;
544 names
[VIRTIO_BALLOON_VQ_DEFLATE
] = "deflate";
545 callbacks
[VIRTIO_BALLOON_VQ_STATS
] = NULL
;
546 names
[VIRTIO_BALLOON_VQ_STATS
] = NULL
;
547 callbacks
[VIRTIO_BALLOON_VQ_FREE_PAGE
] = NULL
;
548 names
[VIRTIO_BALLOON_VQ_FREE_PAGE
] = NULL
;
549 names
[VIRTIO_BALLOON_VQ_REPORTING
] = NULL
;
551 if (virtio_has_feature(vb
->vdev
, VIRTIO_BALLOON_F_STATS_VQ
)) {
552 names
[VIRTIO_BALLOON_VQ_STATS
] = "stats";
553 callbacks
[VIRTIO_BALLOON_VQ_STATS
] = stats_request
;
556 if (virtio_has_feature(vb
->vdev
, VIRTIO_BALLOON_F_FREE_PAGE_HINT
)) {
557 names
[VIRTIO_BALLOON_VQ_FREE_PAGE
] = "free_page_vq";
558 callbacks
[VIRTIO_BALLOON_VQ_FREE_PAGE
] = NULL
;
561 if (virtio_has_feature(vb
->vdev
, VIRTIO_BALLOON_F_REPORTING
)) {
562 names
[VIRTIO_BALLOON_VQ_REPORTING
] = "reporting_vq";
563 callbacks
[VIRTIO_BALLOON_VQ_REPORTING
] = balloon_ack
;
566 err
= virtio_find_vqs(vb
->vdev
, VIRTIO_BALLOON_VQ_MAX
, vqs
,
567 callbacks
, names
, NULL
);
571 vb
->inflate_vq
= vqs
[VIRTIO_BALLOON_VQ_INFLATE
];
572 vb
->deflate_vq
= vqs
[VIRTIO_BALLOON_VQ_DEFLATE
];
573 if (virtio_has_feature(vb
->vdev
, VIRTIO_BALLOON_F_STATS_VQ
)) {
574 struct scatterlist sg
;
575 unsigned int num_stats
;
576 vb
->stats_vq
= vqs
[VIRTIO_BALLOON_VQ_STATS
];
579 * Prime this virtqueue with one buffer so the hypervisor can
580 * use it to signal us later (it can't be broken yet!).
582 num_stats
= update_balloon_stats(vb
);
584 sg_init_one(&sg
, vb
->stats
, sizeof(vb
->stats
[0]) * num_stats
);
585 err
= virtqueue_add_outbuf(vb
->stats_vq
, &sg
, 1, vb
,
588 dev_warn(&vb
->vdev
->dev
, "%s: add stat_vq failed\n",
592 virtqueue_kick(vb
->stats_vq
);
595 if (virtio_has_feature(vb
->vdev
, VIRTIO_BALLOON_F_FREE_PAGE_HINT
))
596 vb
->free_page_vq
= vqs
[VIRTIO_BALLOON_VQ_FREE_PAGE
];
598 if (virtio_has_feature(vb
->vdev
, VIRTIO_BALLOON_F_REPORTING
))
599 vb
->reporting_vq
= vqs
[VIRTIO_BALLOON_VQ_REPORTING
];
604 static u32
virtio_balloon_cmd_id_received(struct virtio_balloon
*vb
)
606 if (test_and_clear_bit(VIRTIO_BALLOON_CONFIG_READ_CMD_ID
,
607 &vb
->config_read_bitmap
)) {
608 /* Legacy balloon config space is LE, unlike all other devices. */
609 virtio_cread_le(vb
->vdev
, struct virtio_balloon_config
,
610 free_page_hint_cmd_id
,
611 &vb
->cmd_id_received_cache
);
614 return vb
->cmd_id_received_cache
;
617 static int send_cmd_id_start(struct virtio_balloon
*vb
)
619 struct scatterlist sg
;
620 struct virtqueue
*vq
= vb
->free_page_vq
;
623 /* Detach all the used buffers from the vq */
624 while (virtqueue_get_buf(vq
, &unused
))
627 vb
->cmd_id_active
= cpu_to_virtio32(vb
->vdev
,
628 virtio_balloon_cmd_id_received(vb
));
629 sg_init_one(&sg
, &vb
->cmd_id_active
, sizeof(vb
->cmd_id_active
));
630 err
= virtqueue_add_outbuf(vq
, &sg
, 1, &vb
->cmd_id_active
, GFP_KERNEL
);
636 static int send_cmd_id_stop(struct virtio_balloon
*vb
)
638 struct scatterlist sg
;
639 struct virtqueue
*vq
= vb
->free_page_vq
;
642 /* Detach all the used buffers from the vq */
643 while (virtqueue_get_buf(vq
, &unused
))
646 sg_init_one(&sg
, &vb
->cmd_id_stop
, sizeof(vb
->cmd_id_stop
));
647 err
= virtqueue_add_outbuf(vq
, &sg
, 1, &vb
->cmd_id_stop
, GFP_KERNEL
);
653 static int get_free_page_and_send(struct virtio_balloon
*vb
)
655 struct virtqueue
*vq
= vb
->free_page_vq
;
657 struct scatterlist sg
;
661 /* Detach all the used buffers from the vq */
662 while (virtqueue_get_buf(vq
, &unused
))
665 page
= alloc_pages(VIRTIO_BALLOON_FREE_PAGE_ALLOC_FLAG
,
666 VIRTIO_BALLOON_HINT_BLOCK_ORDER
);
668 * When the allocation returns NULL, it indicates that we have got all
669 * the possible free pages, so return -EINTR to stop.
674 p
= page_address(page
);
675 sg_init_one(&sg
, p
, VIRTIO_BALLOON_HINT_BLOCK_BYTES
);
676 /* There is always 1 entry reserved for the cmd id to use. */
677 if (vq
->num_free
> 1) {
678 err
= virtqueue_add_inbuf(vq
, &sg
, 1, p
, GFP_KERNEL
);
680 free_pages((unsigned long)p
,
681 VIRTIO_BALLOON_HINT_BLOCK_ORDER
);
685 spin_lock_irq(&vb
->free_page_list_lock
);
686 balloon_page_push(&vb
->free_page_list
, page
);
687 vb
->num_free_page_blocks
++;
688 spin_unlock_irq(&vb
->free_page_list_lock
);
691 * The vq has no available entry to add this page block, so
694 free_pages((unsigned long)p
, VIRTIO_BALLOON_HINT_BLOCK_ORDER
);
700 static int send_free_pages(struct virtio_balloon
*vb
)
707 * If a stop id or a new cmd id was just received from host,
708 * stop the reporting.
710 cmd_id_active
= virtio32_to_cpu(vb
->vdev
, vb
->cmd_id_active
);
711 if (unlikely(cmd_id_active
!=
712 virtio_balloon_cmd_id_received(vb
)))
716 * The free page blocks are allocated and sent to host one by
719 err
= get_free_page_and_send(vb
);
722 else if (unlikely(err
))
729 static void virtio_balloon_report_free_page(struct virtio_balloon
*vb
)
732 struct device
*dev
= &vb
->vdev
->dev
;
734 /* Start by sending the received cmd id to host with an outbuf. */
735 err
= send_cmd_id_start(vb
);
737 dev_err(dev
, "Failed to send a start id, err = %d\n", err
);
739 err
= send_free_pages(vb
);
741 dev_err(dev
, "Failed to send a free page, err = %d\n", err
);
743 /* End by sending a stop id to host with an outbuf. */
744 err
= send_cmd_id_stop(vb
);
746 dev_err(dev
, "Failed to send a stop id, err = %d\n", err
);
749 static void report_free_page_func(struct work_struct
*work
)
751 struct virtio_balloon
*vb
= container_of(work
, struct virtio_balloon
,
752 report_free_page_work
);
755 cmd_id_received
= virtio_balloon_cmd_id_received(vb
);
756 if (cmd_id_received
== VIRTIO_BALLOON_CMD_ID_DONE
) {
757 /* Pass ULONG_MAX to give back all the free pages */
758 return_free_pages_to_mm(vb
, ULONG_MAX
);
759 } else if (cmd_id_received
!= VIRTIO_BALLOON_CMD_ID_STOP
&&
761 virtio32_to_cpu(vb
->vdev
, vb
->cmd_id_active
)) {
762 virtio_balloon_report_free_page(vb
);
766 #ifdef CONFIG_BALLOON_COMPACTION
768 * virtballoon_migratepage - perform the balloon page migration on behalf of
769 * a compaction thread. (called under page lock)
770 * @vb_dev_info: the balloon device
771 * @newpage: page that will replace the isolated page after migration finishes.
772 * @page : the isolated (old) page that is about to be migrated to newpage.
773 * @mode : compaction mode -- not used for balloon page migration.
775 * After a ballooned page gets isolated by compaction procedures, this is the
776 * function that performs the page migration on behalf of a compaction thread
777 * The page migration for virtio balloon is done in a simple swap fashion which
778 * follows these two macro steps:
779 * 1) insert newpage into vb->pages list and update the host about it;
780 * 2) update the host about the old page removed from vb->pages list;
782 * This function preforms the balloon page migration task.
783 * Called through movable_operations->migrate_page
785 static int virtballoon_migratepage(struct balloon_dev_info
*vb_dev_info
,
786 struct page
*newpage
, struct page
*page
, enum migrate_mode mode
)
788 struct virtio_balloon
*vb
= container_of(vb_dev_info
,
789 struct virtio_balloon
, vb_dev_info
);
793 * In order to avoid lock contention while migrating pages concurrently
794 * to leak_balloon() or fill_balloon() we just give up the balloon_lock
795 * this turn, as it is easier to retry the page migration later.
796 * This also prevents fill_balloon() getting stuck into a mutex
797 * recursion in the case it ends up triggering memory compaction
798 * while it is attempting to inflate the ballon.
800 if (!mutex_trylock(&vb
->balloon_lock
))
803 get_page(newpage
); /* balloon reference */
806 * When we migrate a page to a different zone and adjusted the
807 * managed page count when inflating, we have to fixup the count of
808 * both involved zones.
810 if (!virtio_has_feature(vb
->vdev
, VIRTIO_BALLOON_F_DEFLATE_ON_OOM
) &&
811 page_zone(page
) != page_zone(newpage
)) {
812 adjust_managed_page_count(page
, 1);
813 adjust_managed_page_count(newpage
, -1);
816 /* balloon's page migration 1st step -- inflate "newpage" */
817 spin_lock_irqsave(&vb_dev_info
->pages_lock
, flags
);
818 balloon_page_insert(vb_dev_info
, newpage
);
819 vb_dev_info
->isolated_pages
--;
820 __count_vm_event(BALLOON_MIGRATE
);
821 spin_unlock_irqrestore(&vb_dev_info
->pages_lock
, flags
);
822 vb
->num_pfns
= VIRTIO_BALLOON_PAGES_PER_PAGE
;
823 set_page_pfns(vb
, vb
->pfns
, newpage
);
824 tell_host(vb
, vb
->inflate_vq
);
826 /* balloon's page migration 2nd step -- deflate "page" */
827 spin_lock_irqsave(&vb_dev_info
->pages_lock
, flags
);
828 balloon_page_delete(page
);
829 spin_unlock_irqrestore(&vb_dev_info
->pages_lock
, flags
);
830 vb
->num_pfns
= VIRTIO_BALLOON_PAGES_PER_PAGE
;
831 set_page_pfns(vb
, vb
->pfns
, page
);
832 tell_host(vb
, vb
->deflate_vq
);
834 mutex_unlock(&vb
->balloon_lock
);
836 put_page(page
); /* balloon reference */
838 return MIGRATEPAGE_SUCCESS
;
840 #endif /* CONFIG_BALLOON_COMPACTION */
842 static unsigned long shrink_free_pages(struct virtio_balloon
*vb
,
843 unsigned long pages_to_free
)
845 unsigned long blocks_to_free
, blocks_freed
;
847 pages_to_free
= round_up(pages_to_free
,
848 VIRTIO_BALLOON_HINT_BLOCK_PAGES
);
849 blocks_to_free
= pages_to_free
/ VIRTIO_BALLOON_HINT_BLOCK_PAGES
;
850 blocks_freed
= return_free_pages_to_mm(vb
, blocks_to_free
);
852 return blocks_freed
* VIRTIO_BALLOON_HINT_BLOCK_PAGES
;
855 static unsigned long virtio_balloon_shrinker_scan(struct shrinker
*shrinker
,
856 struct shrink_control
*sc
)
858 struct virtio_balloon
*vb
= shrinker
->private_data
;
860 return shrink_free_pages(vb
, sc
->nr_to_scan
);
863 static unsigned long virtio_balloon_shrinker_count(struct shrinker
*shrinker
,
864 struct shrink_control
*sc
)
866 struct virtio_balloon
*vb
= shrinker
->private_data
;
868 return vb
->num_free_page_blocks
* VIRTIO_BALLOON_HINT_BLOCK_PAGES
;
871 static int virtio_balloon_oom_notify(struct notifier_block
*nb
,
872 unsigned long dummy
, void *parm
)
874 struct virtio_balloon
*vb
= container_of(nb
,
875 struct virtio_balloon
, oom_nb
);
876 unsigned long *freed
= parm
;
878 *freed
+= leak_balloon(vb
, VIRTIO_BALLOON_OOM_NR_PAGES
) /
879 VIRTIO_BALLOON_PAGES_PER_PAGE
;
880 update_balloon_size(vb
);
885 static void virtio_balloon_unregister_shrinker(struct virtio_balloon
*vb
)
887 shrinker_free(vb
->shrinker
);
890 static int virtio_balloon_register_shrinker(struct virtio_balloon
*vb
)
892 vb
->shrinker
= shrinker_alloc(0, "virtio-balloon");
896 vb
->shrinker
->scan_objects
= virtio_balloon_shrinker_scan
;
897 vb
->shrinker
->count_objects
= virtio_balloon_shrinker_count
;
898 vb
->shrinker
->private_data
= vb
;
900 shrinker_register(vb
->shrinker
);
905 static int virtballoon_probe(struct virtio_device
*vdev
)
907 struct virtio_balloon
*vb
;
910 if (!vdev
->config
->get
) {
911 dev_err(&vdev
->dev
, "%s failure: config access disabled\n",
916 vdev
->priv
= vb
= kzalloc(sizeof(*vb
), GFP_KERNEL
);
922 INIT_WORK(&vb
->update_balloon_stats_work
, update_balloon_stats_func
);
923 INIT_WORK(&vb
->update_balloon_size_work
, update_balloon_size_func
);
924 spin_lock_init(&vb
->stop_update_lock
);
925 mutex_init(&vb
->balloon_lock
);
926 init_waitqueue_head(&vb
->acked
);
929 balloon_devinfo_init(&vb
->vb_dev_info
);
935 #ifdef CONFIG_BALLOON_COMPACTION
936 vb
->vb_dev_info
.migratepage
= virtballoon_migratepage
;
938 if (virtio_has_feature(vdev
, VIRTIO_BALLOON_F_FREE_PAGE_HINT
)) {
940 * There is always one entry reserved for cmd id, so the ring
941 * size needs to be at least two to report free page hints.
943 if (virtqueue_get_vring_size(vb
->free_page_vq
) < 2) {
947 vb
->balloon_wq
= alloc_workqueue("balloon-wq",
948 WQ_FREEZABLE
| WQ_CPU_INTENSIVE
, 0);
949 if (!vb
->balloon_wq
) {
953 INIT_WORK(&vb
->report_free_page_work
, report_free_page_func
);
954 vb
->cmd_id_received_cache
= VIRTIO_BALLOON_CMD_ID_STOP
;
955 vb
->cmd_id_active
= cpu_to_virtio32(vb
->vdev
,
956 VIRTIO_BALLOON_CMD_ID_STOP
);
957 vb
->cmd_id_stop
= cpu_to_virtio32(vb
->vdev
,
958 VIRTIO_BALLOON_CMD_ID_STOP
);
959 spin_lock_init(&vb
->free_page_list_lock
);
960 INIT_LIST_HEAD(&vb
->free_page_list
);
962 * We're allowed to reuse any free pages, even if they are
963 * still to be processed by the host.
965 err
= virtio_balloon_register_shrinker(vb
);
967 goto out_del_balloon_wq
;
970 if (virtio_has_feature(vb
->vdev
, VIRTIO_BALLOON_F_DEFLATE_ON_OOM
)) {
971 vb
->oom_nb
.notifier_call
= virtio_balloon_oom_notify
;
972 vb
->oom_nb
.priority
= VIRTIO_BALLOON_OOM_NOTIFY_PRIORITY
;
973 err
= register_oom_notifier(&vb
->oom_nb
);
975 goto out_unregister_shrinker
;
978 if (virtio_has_feature(vdev
, VIRTIO_BALLOON_F_PAGE_POISON
)) {
979 /* Start with poison val of 0 representing general init */
980 __u32 poison_val
= 0;
983 * Let the hypervisor know that we are expecting a
984 * specific value to be written back in balloon pages.
986 * If the PAGE_POISON value was larger than a byte we would
987 * need to byte swap poison_val here to guarantee it is
988 * little-endian. However for now it is a single byte so we
991 if (!want_init_on_free())
992 memset(&poison_val
, PAGE_POISON
, sizeof(poison_val
));
994 virtio_cwrite_le(vb
->vdev
, struct virtio_balloon_config
,
995 poison_val
, &poison_val
);
998 vb
->pr_dev_info
.report
= virtballoon_free_page_report
;
999 if (virtio_has_feature(vb
->vdev
, VIRTIO_BALLOON_F_REPORTING
)) {
1000 unsigned int capacity
;
1002 capacity
= virtqueue_get_vring_size(vb
->reporting_vq
);
1003 if (capacity
< PAGE_REPORTING_CAPACITY
) {
1005 goto out_unregister_oom
;
1009 * The default page reporting order is @pageblock_order, which
1010 * corresponds to 512MB in size on ARM64 when 64KB base page
1011 * size is used. The page reporting won't be triggered if the
1012 * freeing page can't come up with a free area like that huge.
1013 * So we specify the page reporting order to 5, corresponding
1014 * to 2MB. It helps to avoid THP splitting if 4KB base page
1015 * size is used by host.
1017 * Ideally, the page reporting order is selected based on the
1018 * host's base page size. However, it needs more work to report
1019 * that value. The hard-coded order would be fine currently.
1021 #if defined(CONFIG_ARM64) && defined(CONFIG_ARM64_64K_PAGES)
1022 vb
->pr_dev_info
.order
= 5;
1025 err
= page_reporting_register(&vb
->pr_dev_info
);
1027 goto out_unregister_oom
;
1030 spin_lock_init(&vb
->adjustment_lock
);
1032 virtio_device_ready(vdev
);
1034 if (towards_target(vb
))
1035 virtballoon_changed(vdev
);
1039 if (virtio_has_feature(vb
->vdev
, VIRTIO_BALLOON_F_DEFLATE_ON_OOM
))
1040 unregister_oom_notifier(&vb
->oom_nb
);
1041 out_unregister_shrinker
:
1042 if (virtio_has_feature(vb
->vdev
, VIRTIO_BALLOON_F_FREE_PAGE_HINT
))
1043 virtio_balloon_unregister_shrinker(vb
);
1045 if (virtio_has_feature(vdev
, VIRTIO_BALLOON_F_FREE_PAGE_HINT
))
1046 destroy_workqueue(vb
->balloon_wq
);
1048 vdev
->config
->del_vqs(vdev
);
1055 static void remove_common(struct virtio_balloon
*vb
)
1057 /* There might be pages left in the balloon: free them. */
1058 while (vb
->num_pages
)
1059 leak_balloon(vb
, vb
->num_pages
);
1060 update_balloon_size(vb
);
1062 /* There might be free pages that are being reported: release them. */
1063 if (virtio_has_feature(vb
->vdev
, VIRTIO_BALLOON_F_FREE_PAGE_HINT
))
1064 return_free_pages_to_mm(vb
, ULONG_MAX
);
1066 /* Now we reset the device so we can clean up the queues. */
1067 virtio_reset_device(vb
->vdev
);
1069 vb
->vdev
->config
->del_vqs(vb
->vdev
);
1072 static void virtballoon_remove(struct virtio_device
*vdev
)
1074 struct virtio_balloon
*vb
= vdev
->priv
;
1076 if (virtio_has_feature(vb
->vdev
, VIRTIO_BALLOON_F_REPORTING
))
1077 page_reporting_unregister(&vb
->pr_dev_info
);
1078 if (virtio_has_feature(vb
->vdev
, VIRTIO_BALLOON_F_DEFLATE_ON_OOM
))
1079 unregister_oom_notifier(&vb
->oom_nb
);
1080 if (virtio_has_feature(vb
->vdev
, VIRTIO_BALLOON_F_FREE_PAGE_HINT
))
1081 virtio_balloon_unregister_shrinker(vb
);
1082 spin_lock_irq(&vb
->stop_update_lock
);
1083 vb
->stop_update
= true;
1084 spin_unlock_irq(&vb
->stop_update_lock
);
1085 cancel_work_sync(&vb
->update_balloon_size_work
);
1086 cancel_work_sync(&vb
->update_balloon_stats_work
);
1088 if (virtio_has_feature(vdev
, VIRTIO_BALLOON_F_FREE_PAGE_HINT
)) {
1089 cancel_work_sync(&vb
->report_free_page_work
);
1090 destroy_workqueue(vb
->balloon_wq
);
1097 #ifdef CONFIG_PM_SLEEP
1098 static int virtballoon_freeze(struct virtio_device
*vdev
)
1100 struct virtio_balloon
*vb
= vdev
->priv
;
1103 * The workqueue is already frozen by the PM core before this
1104 * function is called.
1110 static int virtballoon_restore(struct virtio_device
*vdev
)
1112 struct virtio_balloon
*vb
= vdev
->priv
;
1115 ret
= init_vqs(vdev
->priv
);
1119 virtio_device_ready(vdev
);
1121 if (towards_target(vb
))
1122 virtballoon_changed(vdev
);
1123 update_balloon_size(vb
);
1128 static int virtballoon_validate(struct virtio_device
*vdev
)
1131 * Inform the hypervisor that our pages are poisoned or
1132 * initialized. If we cannot do that then we should disable
1133 * page reporting as it could potentially change the contents
1134 * of our free pages.
1136 if (!want_init_on_free() && !page_poisoning_enabled_static())
1137 __virtio_clear_bit(vdev
, VIRTIO_BALLOON_F_PAGE_POISON
);
1138 else if (!virtio_has_feature(vdev
, VIRTIO_BALLOON_F_PAGE_POISON
))
1139 __virtio_clear_bit(vdev
, VIRTIO_BALLOON_F_REPORTING
);
1141 __virtio_clear_bit(vdev
, VIRTIO_F_ACCESS_PLATFORM
);
1145 static unsigned int features
[] = {
1146 VIRTIO_BALLOON_F_MUST_TELL_HOST
,
1147 VIRTIO_BALLOON_F_STATS_VQ
,
1148 VIRTIO_BALLOON_F_DEFLATE_ON_OOM
,
1149 VIRTIO_BALLOON_F_FREE_PAGE_HINT
,
1150 VIRTIO_BALLOON_F_PAGE_POISON
,
1151 VIRTIO_BALLOON_F_REPORTING
,
1154 static struct virtio_driver virtio_balloon_driver
= {
1155 .feature_table
= features
,
1156 .feature_table_size
= ARRAY_SIZE(features
),
1157 .driver
.name
= KBUILD_MODNAME
,
1158 .driver
.owner
= THIS_MODULE
,
1159 .id_table
= id_table
,
1160 .validate
= virtballoon_validate
,
1161 .probe
= virtballoon_probe
,
1162 .remove
= virtballoon_remove
,
1163 .config_changed
= virtballoon_changed
,
1164 #ifdef CONFIG_PM_SLEEP
1165 .freeze
= virtballoon_freeze
,
1166 .restore
= virtballoon_restore
,
1170 module_virtio_driver(virtio_balloon_driver
);
1171 MODULE_DEVICE_TABLE(virtio
, id_table
);
1172 MODULE_DESCRIPTION("Virtio balloon driver");
1173 MODULE_LICENSE("GPL");