2 * Virtio balloon implementation, inspired by Dor Laor and Marcelo
3 * Tosatti's implementations.
5 * Copyright 2008 Rusty Russell IBM Corporation
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
22 #include <linux/virtio.h>
23 #include <linux/virtio_balloon.h>
24 #include <linux/swap.h>
25 #include <linux/workqueue.h>
26 #include <linux/delay.h>
27 #include <linux/slab.h>
28 #include <linux/module.h>
29 #include <linux/balloon_compaction.h>
30 #include <linux/wait.h>
32 #include <linux/mount.h>
33 #include <linux/magic.h>
36 * Balloon device works in 4K page units. So each page is pointed to by
37 * multiple balloon pages. All memory counters in this driver are in balloon
40 #define VIRTIO_BALLOON_PAGES_PER_PAGE (unsigned)(PAGE_SIZE >> VIRTIO_BALLOON_PFN_SHIFT)
41 #define VIRTIO_BALLOON_ARRAY_PFNS_MAX 256
42 #define VIRTBALLOON_OOM_NOTIFY_PRIORITY 80
44 #define VIRTIO_BALLOON_FREE_PAGE_ALLOC_FLAG (__GFP_NORETRY | __GFP_NOWARN | \
46 /* The order of free page blocks to report to host */
47 #define VIRTIO_BALLOON_FREE_PAGE_ORDER (MAX_ORDER - 1)
48 /* The size of a free page block in bytes */
49 #define VIRTIO_BALLOON_FREE_PAGE_SIZE \
50 (1 << (VIRTIO_BALLOON_FREE_PAGE_ORDER + PAGE_SHIFT))
52 #ifdef CONFIG_BALLOON_COMPACTION
53 static struct vfsmount
*balloon_mnt
;
56 enum virtio_balloon_vq
{
57 VIRTIO_BALLOON_VQ_INFLATE
,
58 VIRTIO_BALLOON_VQ_DEFLATE
,
59 VIRTIO_BALLOON_VQ_STATS
,
60 VIRTIO_BALLOON_VQ_FREE_PAGE
,
64 enum virtio_balloon_config_read
{
65 VIRTIO_BALLOON_CONFIG_READ_CMD_ID
= 0,
68 struct virtio_balloon
{
69 struct virtio_device
*vdev
;
70 struct virtqueue
*inflate_vq
, *deflate_vq
, *stats_vq
, *free_page_vq
;
72 /* Balloon's own wq for cpu-intensive work items */
73 struct workqueue_struct
*balloon_wq
;
74 /* The free page reporting work item submitted to the balloon wq */
75 struct work_struct report_free_page_work
;
77 /* The balloon servicing is delegated to a freezable workqueue. */
78 struct work_struct update_balloon_stats_work
;
79 struct work_struct update_balloon_size_work
;
81 /* Prevent updating balloon when it is being canceled. */
82 spinlock_t stop_update_lock
;
84 /* Bitmap to indicate if reading the related config fields are needed */
85 unsigned long config_read_bitmap
;
87 /* The list of allocated free pages, waiting to be given back to mm */
88 struct list_head free_page_list
;
89 spinlock_t free_page_list_lock
;
90 /* The number of free page blocks on the above list */
91 unsigned long num_free_page_blocks
;
93 * The cmd id received from host.
94 * Read it via virtio_balloon_cmd_id_received to get the latest value
97 u32 cmd_id_received_cache
;
98 /* The cmd id that is actively in use */
99 __virtio32 cmd_id_active
;
100 /* Buffer to store the stop sign */
101 __virtio32 cmd_id_stop
;
103 /* Waiting for host to ack the pages we released. */
104 wait_queue_head_t acked
;
106 /* Number of balloon pages we've told the Host we're not using. */
107 unsigned int num_pages
;
109 * The pages we've told the Host we're not using are enqueued
110 * at vb_dev_info->pages list.
111 * Each page on this list adds VIRTIO_BALLOON_PAGES_PER_PAGE
112 * to num_pages above.
114 struct balloon_dev_info vb_dev_info
;
116 /* Synchronize access/update to this struct virtio_balloon elements */
117 struct mutex balloon_lock
;
119 /* The array of pfns we tell the Host about. */
120 unsigned int num_pfns
;
121 __virtio32 pfns
[VIRTIO_BALLOON_ARRAY_PFNS_MAX
];
123 /* Memory statistics */
124 struct virtio_balloon_stat stats
[VIRTIO_BALLOON_S_NR
];
126 /* To register a shrinker to shrink memory upon memory pressure */
127 struct shrinker shrinker
;
130 static struct virtio_device_id id_table
[] = {
131 { VIRTIO_ID_BALLOON
, VIRTIO_DEV_ANY_ID
},
135 static u32
page_to_balloon_pfn(struct page
*page
)
137 unsigned long pfn
= page_to_pfn(page
);
139 BUILD_BUG_ON(PAGE_SHIFT
< VIRTIO_BALLOON_PFN_SHIFT
);
140 /* Convert pfn from Linux page size to balloon page size. */
141 return pfn
* VIRTIO_BALLOON_PAGES_PER_PAGE
;
144 static void balloon_ack(struct virtqueue
*vq
)
146 struct virtio_balloon
*vb
= vq
->vdev
->priv
;
151 static void tell_host(struct virtio_balloon
*vb
, struct virtqueue
*vq
)
153 struct scatterlist sg
;
156 sg_init_one(&sg
, vb
->pfns
, sizeof(vb
->pfns
[0]) * vb
->num_pfns
);
158 /* We should always be able to add one buffer to an empty queue. */
159 virtqueue_add_outbuf(vq
, &sg
, 1, vb
, GFP_KERNEL
);
162 /* When host has read buffer, this completes via balloon_ack */
163 wait_event(vb
->acked
, virtqueue_get_buf(vq
, &len
));
167 static void set_page_pfns(struct virtio_balloon
*vb
,
168 __virtio32 pfns
[], struct page
*page
)
173 * Set balloon pfns pointing at this page.
174 * Note that the first pfn points at start of the page.
176 for (i
= 0; i
< VIRTIO_BALLOON_PAGES_PER_PAGE
; i
++)
177 pfns
[i
] = cpu_to_virtio32(vb
->vdev
,
178 page_to_balloon_pfn(page
) + i
);
181 static unsigned fill_balloon(struct virtio_balloon
*vb
, size_t num
)
183 unsigned num_allocated_pages
;
188 /* We can only do one array worth at a time. */
189 num
= min(num
, ARRAY_SIZE(vb
->pfns
));
191 for (num_pfns
= 0; num_pfns
< num
;
192 num_pfns
+= VIRTIO_BALLOON_PAGES_PER_PAGE
) {
193 struct page
*page
= balloon_page_alloc();
196 dev_info_ratelimited(&vb
->vdev
->dev
,
197 "Out of puff! Can't get %u pages\n",
198 VIRTIO_BALLOON_PAGES_PER_PAGE
);
199 /* Sleep for at least 1/5 of a second before retry. */
204 balloon_page_push(&pages
, page
);
207 mutex_lock(&vb
->balloon_lock
);
211 while ((page
= balloon_page_pop(&pages
))) {
212 balloon_page_enqueue(&vb
->vb_dev_info
, page
);
214 set_page_pfns(vb
, vb
->pfns
+ vb
->num_pfns
, page
);
215 vb
->num_pages
+= VIRTIO_BALLOON_PAGES_PER_PAGE
;
216 if (!virtio_has_feature(vb
->vdev
,
217 VIRTIO_BALLOON_F_DEFLATE_ON_OOM
))
218 adjust_managed_page_count(page
, -1);
219 vb
->num_pfns
+= VIRTIO_BALLOON_PAGES_PER_PAGE
;
222 num_allocated_pages
= vb
->num_pfns
;
223 /* Did we get any? */
224 if (vb
->num_pfns
!= 0)
225 tell_host(vb
, vb
->inflate_vq
);
226 mutex_unlock(&vb
->balloon_lock
);
228 return num_allocated_pages
;
231 static void release_pages_balloon(struct virtio_balloon
*vb
,
232 struct list_head
*pages
)
234 struct page
*page
, *next
;
236 list_for_each_entry_safe(page
, next
, pages
, lru
) {
237 if (!virtio_has_feature(vb
->vdev
,
238 VIRTIO_BALLOON_F_DEFLATE_ON_OOM
))
239 adjust_managed_page_count(page
, 1);
240 list_del(&page
->lru
);
241 put_page(page
); /* balloon reference */
245 static unsigned leak_balloon(struct virtio_balloon
*vb
, size_t num
)
247 unsigned num_freed_pages
;
249 struct balloon_dev_info
*vb_dev_info
= &vb
->vb_dev_info
;
252 /* We can only do one array worth at a time. */
253 num
= min(num
, ARRAY_SIZE(vb
->pfns
));
255 mutex_lock(&vb
->balloon_lock
);
256 /* We can't release more pages than taken */
257 num
= min(num
, (size_t)vb
->num_pages
);
258 for (vb
->num_pfns
= 0; vb
->num_pfns
< num
;
259 vb
->num_pfns
+= VIRTIO_BALLOON_PAGES_PER_PAGE
) {
260 page
= balloon_page_dequeue(vb_dev_info
);
263 set_page_pfns(vb
, vb
->pfns
+ vb
->num_pfns
, page
);
264 list_add(&page
->lru
, &pages
);
265 vb
->num_pages
-= VIRTIO_BALLOON_PAGES_PER_PAGE
;
268 num_freed_pages
= vb
->num_pfns
;
271 * virtio_has_feature(vdev, VIRTIO_BALLOON_F_MUST_TELL_HOST);
272 * is true, we *have* to do it in this order
274 if (vb
->num_pfns
!= 0)
275 tell_host(vb
, vb
->deflate_vq
);
276 release_pages_balloon(vb
, &pages
);
277 mutex_unlock(&vb
->balloon_lock
);
278 return num_freed_pages
;
281 static inline void update_stat(struct virtio_balloon
*vb
, int idx
,
284 BUG_ON(idx
>= VIRTIO_BALLOON_S_NR
);
285 vb
->stats
[idx
].tag
= cpu_to_virtio16(vb
->vdev
, tag
);
286 vb
->stats
[idx
].val
= cpu_to_virtio64(vb
->vdev
, val
);
289 #define pages_to_bytes(x) ((u64)(x) << PAGE_SHIFT)
291 static unsigned int update_balloon_stats(struct virtio_balloon
*vb
)
293 unsigned long events
[NR_VM_EVENT_ITEMS
];
295 unsigned int idx
= 0;
297 unsigned long caches
;
299 all_vm_events(events
);
302 available
= si_mem_available();
303 caches
= global_node_page_state(NR_FILE_PAGES
);
305 #ifdef CONFIG_VM_EVENT_COUNTERS
306 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_SWAP_IN
,
307 pages_to_bytes(events
[PSWPIN
]));
308 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_SWAP_OUT
,
309 pages_to_bytes(events
[PSWPOUT
]));
310 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_MAJFLT
, events
[PGMAJFAULT
]);
311 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_MINFLT
, events
[PGFAULT
]);
312 #ifdef CONFIG_HUGETLB_PAGE
313 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_HTLB_PGALLOC
,
314 events
[HTLB_BUDDY_PGALLOC
]);
315 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_HTLB_PGFAIL
,
316 events
[HTLB_BUDDY_PGALLOC_FAIL
]);
319 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_MEMFREE
,
320 pages_to_bytes(i
.freeram
));
321 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_MEMTOT
,
322 pages_to_bytes(i
.totalram
));
323 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_AVAIL
,
324 pages_to_bytes(available
));
325 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_CACHES
,
326 pages_to_bytes(caches
));
332 * While most virtqueues communicate guest-initiated requests to the hypervisor,
333 * the stats queue operates in reverse. The driver initializes the virtqueue
334 * with a single buffer. From that point forward, all conversations consist of
335 * a hypervisor request (a call to this function) which directs us to refill
336 * the virtqueue with a fresh stats buffer. Since stats collection can sleep,
337 * we delegate the job to a freezable workqueue that will do the actual work via
338 * stats_handle_request().
340 static void stats_request(struct virtqueue
*vq
)
342 struct virtio_balloon
*vb
= vq
->vdev
->priv
;
344 spin_lock(&vb
->stop_update_lock
);
345 if (!vb
->stop_update
)
346 queue_work(system_freezable_wq
, &vb
->update_balloon_stats_work
);
347 spin_unlock(&vb
->stop_update_lock
);
350 static void stats_handle_request(struct virtio_balloon
*vb
)
352 struct virtqueue
*vq
;
353 struct scatterlist sg
;
354 unsigned int len
, num_stats
;
356 num_stats
= update_balloon_stats(vb
);
359 if (!virtqueue_get_buf(vq
, &len
))
361 sg_init_one(&sg
, vb
->stats
, sizeof(vb
->stats
[0]) * num_stats
);
362 virtqueue_add_outbuf(vq
, &sg
, 1, vb
, GFP_KERNEL
);
366 static inline s64
towards_target(struct virtio_balloon
*vb
)
371 virtio_cread(vb
->vdev
, struct virtio_balloon_config
, num_pages
,
374 /* Legacy balloon config space is LE, unlike all other devices. */
375 if (!virtio_has_feature(vb
->vdev
, VIRTIO_F_VERSION_1
))
376 num_pages
= le32_to_cpu((__force __le32
)num_pages
);
379 return target
- vb
->num_pages
;
382 /* Gives back @num_to_return blocks of free pages to mm. */
383 static unsigned long return_free_pages_to_mm(struct virtio_balloon
*vb
,
384 unsigned long num_to_return
)
387 unsigned long num_returned
;
389 spin_lock_irq(&vb
->free_page_list_lock
);
390 for (num_returned
= 0; num_returned
< num_to_return
; num_returned
++) {
391 page
= balloon_page_pop(&vb
->free_page_list
);
394 free_pages((unsigned long)page_address(page
),
395 VIRTIO_BALLOON_FREE_PAGE_ORDER
);
397 vb
->num_free_page_blocks
-= num_returned
;
398 spin_unlock_irq(&vb
->free_page_list_lock
);
403 static void virtio_balloon_queue_free_page_work(struct virtio_balloon
*vb
)
405 if (!virtio_has_feature(vb
->vdev
, VIRTIO_BALLOON_F_FREE_PAGE_HINT
))
408 /* No need to queue the work if the bit was already set. */
409 if (test_and_set_bit(VIRTIO_BALLOON_CONFIG_READ_CMD_ID
,
410 &vb
->config_read_bitmap
))
413 queue_work(vb
->balloon_wq
, &vb
->report_free_page_work
);
416 static void virtballoon_changed(struct virtio_device
*vdev
)
418 struct virtio_balloon
*vb
= vdev
->priv
;
421 spin_lock_irqsave(&vb
->stop_update_lock
, flags
);
422 if (!vb
->stop_update
) {
423 queue_work(system_freezable_wq
,
424 &vb
->update_balloon_size_work
);
425 virtio_balloon_queue_free_page_work(vb
);
427 spin_unlock_irqrestore(&vb
->stop_update_lock
, flags
);
430 static void update_balloon_size(struct virtio_balloon
*vb
)
432 u32 actual
= vb
->num_pages
;
434 /* Legacy balloon config space is LE, unlike all other devices. */
435 if (!virtio_has_feature(vb
->vdev
, VIRTIO_F_VERSION_1
))
436 actual
= (__force u32
)cpu_to_le32(actual
);
438 virtio_cwrite(vb
->vdev
, struct virtio_balloon_config
, actual
,
442 static void update_balloon_stats_func(struct work_struct
*work
)
444 struct virtio_balloon
*vb
;
446 vb
= container_of(work
, struct virtio_balloon
,
447 update_balloon_stats_work
);
448 stats_handle_request(vb
);
451 static void update_balloon_size_func(struct work_struct
*work
)
453 struct virtio_balloon
*vb
;
456 vb
= container_of(work
, struct virtio_balloon
,
457 update_balloon_size_work
);
458 diff
= towards_target(vb
);
461 diff
-= fill_balloon(vb
, diff
);
463 diff
+= leak_balloon(vb
, -diff
);
464 update_balloon_size(vb
);
467 queue_work(system_freezable_wq
, work
);
470 static int init_vqs(struct virtio_balloon
*vb
)
472 struct virtqueue
*vqs
[VIRTIO_BALLOON_VQ_MAX
];
473 vq_callback_t
*callbacks
[VIRTIO_BALLOON_VQ_MAX
];
474 const char *names
[VIRTIO_BALLOON_VQ_MAX
];
478 * Inflateq and deflateq are used unconditionally. The names[]
479 * will be NULL if the related feature is not enabled, which will
480 * cause no allocation for the corresponding virtqueue in find_vqs.
482 callbacks
[VIRTIO_BALLOON_VQ_INFLATE
] = balloon_ack
;
483 names
[VIRTIO_BALLOON_VQ_INFLATE
] = "inflate";
484 callbacks
[VIRTIO_BALLOON_VQ_DEFLATE
] = balloon_ack
;
485 names
[VIRTIO_BALLOON_VQ_DEFLATE
] = "deflate";
486 names
[VIRTIO_BALLOON_VQ_STATS
] = NULL
;
487 names
[VIRTIO_BALLOON_VQ_FREE_PAGE
] = NULL
;
489 if (virtio_has_feature(vb
->vdev
, VIRTIO_BALLOON_F_STATS_VQ
)) {
490 names
[VIRTIO_BALLOON_VQ_STATS
] = "stats";
491 callbacks
[VIRTIO_BALLOON_VQ_STATS
] = stats_request
;
494 if (virtio_has_feature(vb
->vdev
, VIRTIO_BALLOON_F_FREE_PAGE_HINT
)) {
495 names
[VIRTIO_BALLOON_VQ_FREE_PAGE
] = "free_page_vq";
496 callbacks
[VIRTIO_BALLOON_VQ_FREE_PAGE
] = NULL
;
499 err
= vb
->vdev
->config
->find_vqs(vb
->vdev
, VIRTIO_BALLOON_VQ_MAX
,
500 vqs
, callbacks
, names
, NULL
, NULL
);
504 vb
->inflate_vq
= vqs
[VIRTIO_BALLOON_VQ_INFLATE
];
505 vb
->deflate_vq
= vqs
[VIRTIO_BALLOON_VQ_DEFLATE
];
506 if (virtio_has_feature(vb
->vdev
, VIRTIO_BALLOON_F_STATS_VQ
)) {
507 struct scatterlist sg
;
508 unsigned int num_stats
;
509 vb
->stats_vq
= vqs
[VIRTIO_BALLOON_VQ_STATS
];
512 * Prime this virtqueue with one buffer so the hypervisor can
513 * use it to signal us later (it can't be broken yet!).
515 num_stats
= update_balloon_stats(vb
);
517 sg_init_one(&sg
, vb
->stats
, sizeof(vb
->stats
[0]) * num_stats
);
518 err
= virtqueue_add_outbuf(vb
->stats_vq
, &sg
, 1, vb
,
521 dev_warn(&vb
->vdev
->dev
, "%s: add stat_vq failed\n",
525 virtqueue_kick(vb
->stats_vq
);
528 if (virtio_has_feature(vb
->vdev
, VIRTIO_BALLOON_F_FREE_PAGE_HINT
))
529 vb
->free_page_vq
= vqs
[VIRTIO_BALLOON_VQ_FREE_PAGE
];
534 static u32
virtio_balloon_cmd_id_received(struct virtio_balloon
*vb
)
536 if (test_and_clear_bit(VIRTIO_BALLOON_CONFIG_READ_CMD_ID
,
537 &vb
->config_read_bitmap
))
538 virtio_cread(vb
->vdev
, struct virtio_balloon_config
,
539 free_page_report_cmd_id
,
540 &vb
->cmd_id_received_cache
);
542 return vb
->cmd_id_received_cache
;
545 static int send_cmd_id_start(struct virtio_balloon
*vb
)
547 struct scatterlist sg
;
548 struct virtqueue
*vq
= vb
->free_page_vq
;
551 /* Detach all the used buffers from the vq */
552 while (virtqueue_get_buf(vq
, &unused
))
555 vb
->cmd_id_active
= virtio32_to_cpu(vb
->vdev
,
556 virtio_balloon_cmd_id_received(vb
));
557 sg_init_one(&sg
, &vb
->cmd_id_active
, sizeof(vb
->cmd_id_active
));
558 err
= virtqueue_add_outbuf(vq
, &sg
, 1, &vb
->cmd_id_active
, GFP_KERNEL
);
564 static int send_cmd_id_stop(struct virtio_balloon
*vb
)
566 struct scatterlist sg
;
567 struct virtqueue
*vq
= vb
->free_page_vq
;
570 /* Detach all the used buffers from the vq */
571 while (virtqueue_get_buf(vq
, &unused
))
574 sg_init_one(&sg
, &vb
->cmd_id_stop
, sizeof(vb
->cmd_id_stop
));
575 err
= virtqueue_add_outbuf(vq
, &sg
, 1, &vb
->cmd_id_stop
, GFP_KERNEL
);
581 static int get_free_page_and_send(struct virtio_balloon
*vb
)
583 struct virtqueue
*vq
= vb
->free_page_vq
;
585 struct scatterlist sg
;
589 /* Detach all the used buffers from the vq */
590 while (virtqueue_get_buf(vq
, &unused
))
593 page
= alloc_pages(VIRTIO_BALLOON_FREE_PAGE_ALLOC_FLAG
,
594 VIRTIO_BALLOON_FREE_PAGE_ORDER
);
596 * When the allocation returns NULL, it indicates that we have got all
597 * the possible free pages, so return -EINTR to stop.
602 p
= page_address(page
);
603 sg_init_one(&sg
, p
, VIRTIO_BALLOON_FREE_PAGE_SIZE
);
604 /* There is always 1 entry reserved for the cmd id to use. */
605 if (vq
->num_free
> 1) {
606 err
= virtqueue_add_inbuf(vq
, &sg
, 1, p
, GFP_KERNEL
);
608 free_pages((unsigned long)p
,
609 VIRTIO_BALLOON_FREE_PAGE_ORDER
);
613 spin_lock_irq(&vb
->free_page_list_lock
);
614 balloon_page_push(&vb
->free_page_list
, page
);
615 vb
->num_free_page_blocks
++;
616 spin_unlock_irq(&vb
->free_page_list_lock
);
619 * The vq has no available entry to add this page block, so
622 free_pages((unsigned long)p
, VIRTIO_BALLOON_FREE_PAGE_ORDER
);
628 static int send_free_pages(struct virtio_balloon
*vb
)
635 * If a stop id or a new cmd id was just received from host,
636 * stop the reporting.
638 cmd_id_active
= virtio32_to_cpu(vb
->vdev
, vb
->cmd_id_active
);
639 if (unlikely(cmd_id_active
!=
640 virtio_balloon_cmd_id_received(vb
)))
644 * The free page blocks are allocated and sent to host one by
647 err
= get_free_page_and_send(vb
);
650 else if (unlikely(err
))
657 static void virtio_balloon_report_free_page(struct virtio_balloon
*vb
)
660 struct device
*dev
= &vb
->vdev
->dev
;
662 /* Start by sending the received cmd id to host with an outbuf. */
663 err
= send_cmd_id_start(vb
);
665 dev_err(dev
, "Failed to send a start id, err = %d\n", err
);
667 err
= send_free_pages(vb
);
669 dev_err(dev
, "Failed to send a free page, err = %d\n", err
);
671 /* End by sending a stop id to host with an outbuf. */
672 err
= send_cmd_id_stop(vb
);
674 dev_err(dev
, "Failed to send a stop id, err = %d\n", err
);
677 static void report_free_page_func(struct work_struct
*work
)
679 struct virtio_balloon
*vb
= container_of(work
, struct virtio_balloon
,
680 report_free_page_work
);
683 cmd_id_received
= virtio_balloon_cmd_id_received(vb
);
684 if (cmd_id_received
== VIRTIO_BALLOON_CMD_ID_DONE
) {
685 /* Pass ULONG_MAX to give back all the free pages */
686 return_free_pages_to_mm(vb
, ULONG_MAX
);
687 } else if (cmd_id_received
!= VIRTIO_BALLOON_CMD_ID_STOP
&&
689 virtio32_to_cpu(vb
->vdev
, vb
->cmd_id_active
)) {
690 virtio_balloon_report_free_page(vb
);
694 #ifdef CONFIG_BALLOON_COMPACTION
696 * virtballoon_migratepage - perform the balloon page migration on behalf of
697 * a compation thread. (called under page lock)
698 * @vb_dev_info: the balloon device
699 * @newpage: page that will replace the isolated page after migration finishes.
700 * @page : the isolated (old) page that is about to be migrated to newpage.
701 * @mode : compaction mode -- not used for balloon page migration.
703 * After a ballooned page gets isolated by compaction procedures, this is the
704 * function that performs the page migration on behalf of a compaction thread
705 * The page migration for virtio balloon is done in a simple swap fashion which
706 * follows these two macro steps:
707 * 1) insert newpage into vb->pages list and update the host about it;
708 * 2) update the host about the old page removed from vb->pages list;
710 * This function preforms the balloon page migration task.
711 * Called through balloon_mapping->a_ops->migratepage
713 static int virtballoon_migratepage(struct balloon_dev_info
*vb_dev_info
,
714 struct page
*newpage
, struct page
*page
, enum migrate_mode mode
)
716 struct virtio_balloon
*vb
= container_of(vb_dev_info
,
717 struct virtio_balloon
, vb_dev_info
);
721 * In order to avoid lock contention while migrating pages concurrently
722 * to leak_balloon() or fill_balloon() we just give up the balloon_lock
723 * this turn, as it is easier to retry the page migration later.
724 * This also prevents fill_balloon() getting stuck into a mutex
725 * recursion in the case it ends up triggering memory compaction
726 * while it is attempting to inflate the ballon.
728 if (!mutex_trylock(&vb
->balloon_lock
))
731 get_page(newpage
); /* balloon reference */
733 /* balloon's page migration 1st step -- inflate "newpage" */
734 spin_lock_irqsave(&vb_dev_info
->pages_lock
, flags
);
735 balloon_page_insert(vb_dev_info
, newpage
);
736 vb_dev_info
->isolated_pages
--;
737 __count_vm_event(BALLOON_MIGRATE
);
738 spin_unlock_irqrestore(&vb_dev_info
->pages_lock
, flags
);
739 vb
->num_pfns
= VIRTIO_BALLOON_PAGES_PER_PAGE
;
740 set_page_pfns(vb
, vb
->pfns
, newpage
);
741 tell_host(vb
, vb
->inflate_vq
);
743 /* balloon's page migration 2nd step -- deflate "page" */
744 spin_lock_irqsave(&vb_dev_info
->pages_lock
, flags
);
745 balloon_page_delete(page
);
746 spin_unlock_irqrestore(&vb_dev_info
->pages_lock
, flags
);
747 vb
->num_pfns
= VIRTIO_BALLOON_PAGES_PER_PAGE
;
748 set_page_pfns(vb
, vb
->pfns
, page
);
749 tell_host(vb
, vb
->deflate_vq
);
751 mutex_unlock(&vb
->balloon_lock
);
753 put_page(page
); /* balloon reference */
755 return MIGRATEPAGE_SUCCESS
;
758 static struct dentry
*balloon_mount(struct file_system_type
*fs_type
,
759 int flags
, const char *dev_name
, void *data
)
761 static const struct dentry_operations ops
= {
762 .d_dname
= simple_dname
,
765 return mount_pseudo(fs_type
, "balloon-kvm:", NULL
, &ops
,
769 static struct file_system_type balloon_fs
= {
770 .name
= "balloon-kvm",
771 .mount
= balloon_mount
,
772 .kill_sb
= kill_anon_super
,
775 #endif /* CONFIG_BALLOON_COMPACTION */
777 static unsigned long shrink_free_pages(struct virtio_balloon
*vb
,
778 unsigned long pages_to_free
)
780 unsigned long blocks_to_free
, blocks_freed
;
782 pages_to_free
= round_up(pages_to_free
,
783 1 << VIRTIO_BALLOON_FREE_PAGE_ORDER
);
784 blocks_to_free
= pages_to_free
>> VIRTIO_BALLOON_FREE_PAGE_ORDER
;
785 blocks_freed
= return_free_pages_to_mm(vb
, blocks_to_free
);
787 return blocks_freed
<< VIRTIO_BALLOON_FREE_PAGE_ORDER
;
790 static unsigned long shrink_balloon_pages(struct virtio_balloon
*vb
,
791 unsigned long pages_to_free
)
793 unsigned long pages_freed
= 0;
796 * One invocation of leak_balloon can deflate at most
797 * VIRTIO_BALLOON_ARRAY_PFNS_MAX balloon pages, so we call it
798 * multiple times to deflate pages till reaching pages_to_free.
800 while (vb
->num_pages
&& pages_to_free
) {
801 pages_freed
+= leak_balloon(vb
, pages_to_free
) /
802 VIRTIO_BALLOON_PAGES_PER_PAGE
;
803 pages_to_free
-= pages_freed
;
805 update_balloon_size(vb
);
810 static unsigned long virtio_balloon_shrinker_scan(struct shrinker
*shrinker
,
811 struct shrink_control
*sc
)
813 unsigned long pages_to_free
, pages_freed
= 0;
814 struct virtio_balloon
*vb
= container_of(shrinker
,
815 struct virtio_balloon
, shrinker
);
817 pages_to_free
= sc
->nr_to_scan
* VIRTIO_BALLOON_PAGES_PER_PAGE
;
819 if (virtio_has_feature(vb
->vdev
, VIRTIO_BALLOON_F_FREE_PAGE_HINT
))
820 pages_freed
= shrink_free_pages(vb
, pages_to_free
);
822 if (pages_freed
>= pages_to_free
)
825 pages_freed
+= shrink_balloon_pages(vb
, pages_to_free
- pages_freed
);
830 static unsigned long virtio_balloon_shrinker_count(struct shrinker
*shrinker
,
831 struct shrink_control
*sc
)
833 struct virtio_balloon
*vb
= container_of(shrinker
,
834 struct virtio_balloon
, shrinker
);
837 count
= vb
->num_pages
/ VIRTIO_BALLOON_PAGES_PER_PAGE
;
838 count
+= vb
->num_free_page_blocks
>> VIRTIO_BALLOON_FREE_PAGE_ORDER
;
843 static void virtio_balloon_unregister_shrinker(struct virtio_balloon
*vb
)
845 unregister_shrinker(&vb
->shrinker
);
848 static int virtio_balloon_register_shrinker(struct virtio_balloon
*vb
)
850 vb
->shrinker
.scan_objects
= virtio_balloon_shrinker_scan
;
851 vb
->shrinker
.count_objects
= virtio_balloon_shrinker_count
;
852 vb
->shrinker
.seeks
= DEFAULT_SEEKS
;
854 return register_shrinker(&vb
->shrinker
);
857 static int virtballoon_probe(struct virtio_device
*vdev
)
859 struct virtio_balloon
*vb
;
863 if (!vdev
->config
->get
) {
864 dev_err(&vdev
->dev
, "%s failure: config access disabled\n",
869 vdev
->priv
= vb
= kzalloc(sizeof(*vb
), GFP_KERNEL
);
875 INIT_WORK(&vb
->update_balloon_stats_work
, update_balloon_stats_func
);
876 INIT_WORK(&vb
->update_balloon_size_work
, update_balloon_size_func
);
877 spin_lock_init(&vb
->stop_update_lock
);
878 mutex_init(&vb
->balloon_lock
);
879 init_waitqueue_head(&vb
->acked
);
882 balloon_devinfo_init(&vb
->vb_dev_info
);
888 #ifdef CONFIG_BALLOON_COMPACTION
889 balloon_mnt
= kern_mount(&balloon_fs
);
890 if (IS_ERR(balloon_mnt
)) {
891 err
= PTR_ERR(balloon_mnt
);
895 vb
->vb_dev_info
.migratepage
= virtballoon_migratepage
;
896 vb
->vb_dev_info
.inode
= alloc_anon_inode(balloon_mnt
->mnt_sb
);
897 if (IS_ERR(vb
->vb_dev_info
.inode
)) {
898 err
= PTR_ERR(vb
->vb_dev_info
.inode
);
899 kern_unmount(balloon_mnt
);
902 vb
->vb_dev_info
.inode
->i_mapping
->a_ops
= &balloon_aops
;
904 if (virtio_has_feature(vdev
, VIRTIO_BALLOON_F_FREE_PAGE_HINT
)) {
906 * There is always one entry reserved for cmd id, so the ring
907 * size needs to be at least two to report free page hints.
909 if (virtqueue_get_vring_size(vb
->free_page_vq
) < 2) {
913 vb
->balloon_wq
= alloc_workqueue("balloon-wq",
914 WQ_FREEZABLE
| WQ_CPU_INTENSIVE
, 0);
915 if (!vb
->balloon_wq
) {
919 INIT_WORK(&vb
->report_free_page_work
, report_free_page_func
);
920 vb
->cmd_id_received_cache
= VIRTIO_BALLOON_CMD_ID_STOP
;
921 vb
->cmd_id_active
= cpu_to_virtio32(vb
->vdev
,
922 VIRTIO_BALLOON_CMD_ID_STOP
);
923 vb
->cmd_id_stop
= cpu_to_virtio32(vb
->vdev
,
924 VIRTIO_BALLOON_CMD_ID_STOP
);
925 vb
->num_free_page_blocks
= 0;
926 spin_lock_init(&vb
->free_page_list_lock
);
927 INIT_LIST_HEAD(&vb
->free_page_list
);
928 if (virtio_has_feature(vdev
, VIRTIO_BALLOON_F_PAGE_POISON
)) {
929 memset(&poison_val
, PAGE_POISON
, sizeof(poison_val
));
930 virtio_cwrite(vb
->vdev
, struct virtio_balloon_config
,
931 poison_val
, &poison_val
);
935 * We continue to use VIRTIO_BALLOON_F_DEFLATE_ON_OOM to decide if a
936 * shrinker needs to be registered to relieve memory pressure.
938 if (virtio_has_feature(vb
->vdev
, VIRTIO_BALLOON_F_DEFLATE_ON_OOM
)) {
939 err
= virtio_balloon_register_shrinker(vb
);
941 goto out_del_balloon_wq
;
943 virtio_device_ready(vdev
);
945 if (towards_target(vb
))
946 virtballoon_changed(vdev
);
950 if (virtio_has_feature(vdev
, VIRTIO_BALLOON_F_FREE_PAGE_HINT
))
951 destroy_workqueue(vb
->balloon_wq
);
953 vdev
->config
->del_vqs(vdev
);
960 static void remove_common(struct virtio_balloon
*vb
)
962 /* There might be pages left in the balloon: free them. */
963 while (vb
->num_pages
)
964 leak_balloon(vb
, vb
->num_pages
);
965 update_balloon_size(vb
);
967 /* Now we reset the device so we can clean up the queues. */
968 vb
->vdev
->config
->reset(vb
->vdev
);
970 vb
->vdev
->config
->del_vqs(vb
->vdev
);
973 static void virtballoon_remove(struct virtio_device
*vdev
)
975 struct virtio_balloon
*vb
= vdev
->priv
;
977 if (virtio_has_feature(vb
->vdev
, VIRTIO_BALLOON_F_DEFLATE_ON_OOM
))
978 virtio_balloon_unregister_shrinker(vb
);
979 spin_lock_irq(&vb
->stop_update_lock
);
980 vb
->stop_update
= true;
981 spin_unlock_irq(&vb
->stop_update_lock
);
982 cancel_work_sync(&vb
->update_balloon_size_work
);
983 cancel_work_sync(&vb
->update_balloon_stats_work
);
985 if (virtio_has_feature(vdev
, VIRTIO_BALLOON_F_FREE_PAGE_HINT
)) {
986 cancel_work_sync(&vb
->report_free_page_work
);
987 destroy_workqueue(vb
->balloon_wq
);
991 #ifdef CONFIG_BALLOON_COMPACTION
992 if (vb
->vb_dev_info
.inode
)
993 iput(vb
->vb_dev_info
.inode
);
995 kern_unmount(balloon_mnt
);
1000 #ifdef CONFIG_PM_SLEEP
1001 static int virtballoon_freeze(struct virtio_device
*vdev
)
1003 struct virtio_balloon
*vb
= vdev
->priv
;
1006 * The workqueue is already frozen by the PM core before this
1007 * function is called.
1013 static int virtballoon_restore(struct virtio_device
*vdev
)
1015 struct virtio_balloon
*vb
= vdev
->priv
;
1018 ret
= init_vqs(vdev
->priv
);
1022 virtio_device_ready(vdev
);
1024 if (towards_target(vb
))
1025 virtballoon_changed(vdev
);
1026 update_balloon_size(vb
);
1031 static int virtballoon_validate(struct virtio_device
*vdev
)
1033 if (!page_poisoning_enabled())
1034 __virtio_clear_bit(vdev
, VIRTIO_BALLOON_F_PAGE_POISON
);
1036 __virtio_clear_bit(vdev
, VIRTIO_F_IOMMU_PLATFORM
);
1040 static unsigned int features
[] = {
1041 VIRTIO_BALLOON_F_MUST_TELL_HOST
,
1042 VIRTIO_BALLOON_F_STATS_VQ
,
1043 VIRTIO_BALLOON_F_DEFLATE_ON_OOM
,
1044 VIRTIO_BALLOON_F_FREE_PAGE_HINT
,
1045 VIRTIO_BALLOON_F_PAGE_POISON
,
1048 static struct virtio_driver virtio_balloon_driver
= {
1049 .feature_table
= features
,
1050 .feature_table_size
= ARRAY_SIZE(features
),
1051 .driver
.name
= KBUILD_MODNAME
,
1052 .driver
.owner
= THIS_MODULE
,
1053 .id_table
= id_table
,
1054 .validate
= virtballoon_validate
,
1055 .probe
= virtballoon_probe
,
1056 .remove
= virtballoon_remove
,
1057 .config_changed
= virtballoon_changed
,
1058 #ifdef CONFIG_PM_SLEEP
1059 .freeze
= virtballoon_freeze
,
1060 .restore
= virtballoon_restore
,
1064 module_virtio_driver(virtio_balloon_driver
);
1065 MODULE_DEVICE_TABLE(virtio
, id_table
);
1066 MODULE_DESCRIPTION("Virtio balloon driver");
1067 MODULE_LICENSE("GPL");