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/kthread.h>
26 #include <linux/freezer.h>
27 #include <linux/delay.h>
28 #include <linux/slab.h>
29 #include <linux/module.h>
30 #include <linux/balloon_compaction.h>
33 * Balloon device works in 4K page units. So each page is pointed to by
34 * multiple balloon pages. All memory counters in this driver are in balloon
37 #define VIRTIO_BALLOON_PAGES_PER_PAGE (unsigned)(PAGE_SIZE >> VIRTIO_BALLOON_PFN_SHIFT)
38 #define VIRTIO_BALLOON_ARRAY_PFNS_MAX 256
42 struct virtio_device
*vdev
;
43 struct virtqueue
*inflate_vq
, *deflate_vq
, *stats_vq
;
45 /* Where the ballooning thread waits for config to change. */
46 wait_queue_head_t config_change
;
48 /* The thread servicing the balloon. */
49 struct task_struct
*thread
;
51 /* Waiting for host to ack the pages we released. */
52 wait_queue_head_t acked
;
54 /* Number of balloon pages we've told the Host we're not using. */
55 unsigned int num_pages
;
57 * The pages we've told the Host we're not using are enqueued
58 * at vb_dev_info->pages list.
59 * Each page on this list adds VIRTIO_BALLOON_PAGES_PER_PAGE
62 struct balloon_dev_info
*vb_dev_info
;
64 /* Synchronize access/update to this struct virtio_balloon elements */
65 struct mutex balloon_lock
;
67 /* The array of pfns we tell the Host about. */
68 unsigned int num_pfns
;
69 u32 pfns
[VIRTIO_BALLOON_ARRAY_PFNS_MAX
];
71 /* Memory statistics */
72 int need_stats_update
;
73 struct virtio_balloon_stat stats
[VIRTIO_BALLOON_S_NR
];
76 static struct virtio_device_id id_table
[] = {
77 { VIRTIO_ID_BALLOON
, VIRTIO_DEV_ANY_ID
},
81 static u32
page_to_balloon_pfn(struct page
*page
)
83 unsigned long pfn
= page_to_pfn(page
);
85 BUILD_BUG_ON(PAGE_SHIFT
< VIRTIO_BALLOON_PFN_SHIFT
);
86 /* Convert pfn from Linux page size to balloon page size. */
87 return pfn
* VIRTIO_BALLOON_PAGES_PER_PAGE
;
90 static struct page
*balloon_pfn_to_page(u32 pfn
)
92 BUG_ON(pfn
% VIRTIO_BALLOON_PAGES_PER_PAGE
);
93 return pfn_to_page(pfn
/ VIRTIO_BALLOON_PAGES_PER_PAGE
);
96 static void balloon_ack(struct virtqueue
*vq
)
98 struct virtio_balloon
*vb
= vq
->vdev
->priv
;
103 static void tell_host(struct virtio_balloon
*vb
, struct virtqueue
*vq
)
105 struct scatterlist sg
;
108 sg_init_one(&sg
, vb
->pfns
, sizeof(vb
->pfns
[0]) * vb
->num_pfns
);
110 /* We should always be able to add one buffer to an empty queue. */
111 if (virtqueue_add_outbuf(vq
, &sg
, 1, vb
, GFP_KERNEL
) < 0)
115 /* When host has read buffer, this completes via balloon_ack */
116 wait_event(vb
->acked
, virtqueue_get_buf(vq
, &len
));
119 static void set_page_pfns(u32 pfns
[], struct page
*page
)
123 /* Set balloon pfns pointing at this page.
124 * Note that the first pfn points at start of the page. */
125 for (i
= 0; i
< VIRTIO_BALLOON_PAGES_PER_PAGE
; i
++)
126 pfns
[i
] = page_to_balloon_pfn(page
) + i
;
129 static void fill_balloon(struct virtio_balloon
*vb
, size_t num
)
131 struct balloon_dev_info
*vb_dev_info
= vb
->vb_dev_info
;
133 /* We can only do one array worth at a time. */
134 num
= min(num
, ARRAY_SIZE(vb
->pfns
));
136 mutex_lock(&vb
->balloon_lock
);
137 for (vb
->num_pfns
= 0; vb
->num_pfns
< num
;
138 vb
->num_pfns
+= VIRTIO_BALLOON_PAGES_PER_PAGE
) {
139 struct page
*page
= balloon_page_enqueue(vb_dev_info
);
142 dev_info_ratelimited(&vb
->vdev
->dev
,
143 "Out of puff! Can't get %u pages\n",
144 VIRTIO_BALLOON_PAGES_PER_PAGE
);
145 /* Sleep for at least 1/5 of a second before retry. */
149 set_page_pfns(vb
->pfns
+ vb
->num_pfns
, page
);
150 vb
->num_pages
+= VIRTIO_BALLOON_PAGES_PER_PAGE
;
151 adjust_managed_page_count(page
, -1);
154 /* Did we get any? */
155 if (vb
->num_pfns
!= 0)
156 tell_host(vb
, vb
->inflate_vq
);
157 mutex_unlock(&vb
->balloon_lock
);
160 static void release_pages_by_pfn(const u32 pfns
[], unsigned int num
)
164 /* Find pfns pointing at start of each page, get pages and free them. */
165 for (i
= 0; i
< num
; i
+= VIRTIO_BALLOON_PAGES_PER_PAGE
) {
166 struct page
*page
= balloon_pfn_to_page(pfns
[i
]);
167 balloon_page_free(page
);
168 adjust_managed_page_count(page
, 1);
172 static void leak_balloon(struct virtio_balloon
*vb
, size_t num
)
175 struct balloon_dev_info
*vb_dev_info
= vb
->vb_dev_info
;
177 /* We can only do one array worth at a time. */
178 num
= min(num
, ARRAY_SIZE(vb
->pfns
));
180 mutex_lock(&vb
->balloon_lock
);
181 for (vb
->num_pfns
= 0; vb
->num_pfns
< num
;
182 vb
->num_pfns
+= VIRTIO_BALLOON_PAGES_PER_PAGE
) {
183 page
= balloon_page_dequeue(vb_dev_info
);
186 set_page_pfns(vb
->pfns
+ vb
->num_pfns
, page
);
187 vb
->num_pages
-= VIRTIO_BALLOON_PAGES_PER_PAGE
;
192 * virtio_has_feature(vdev, VIRTIO_BALLOON_F_MUST_TELL_HOST);
193 * is true, we *have* to do it in this order
195 if (vb
->num_pfns
!= 0)
196 tell_host(vb
, vb
->deflate_vq
);
197 mutex_unlock(&vb
->balloon_lock
);
198 release_pages_by_pfn(vb
->pfns
, vb
->num_pfns
);
201 static inline void update_stat(struct virtio_balloon
*vb
, int idx
,
204 BUG_ON(idx
>= VIRTIO_BALLOON_S_NR
);
205 vb
->stats
[idx
].tag
= tag
;
206 vb
->stats
[idx
].val
= val
;
209 #define pages_to_bytes(x) ((u64)(x) << PAGE_SHIFT)
211 static void update_balloon_stats(struct virtio_balloon
*vb
)
213 unsigned long events
[NR_VM_EVENT_ITEMS
];
217 all_vm_events(events
);
220 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_SWAP_IN
,
221 pages_to_bytes(events
[PSWPIN
]));
222 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_SWAP_OUT
,
223 pages_to_bytes(events
[PSWPOUT
]));
224 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_MAJFLT
, events
[PGMAJFAULT
]);
225 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_MINFLT
, events
[PGFAULT
]);
226 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_MEMFREE
,
227 pages_to_bytes(i
.freeram
));
228 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_MEMTOT
,
229 pages_to_bytes(i
.totalram
));
233 * While most virtqueues communicate guest-initiated requests to the hypervisor,
234 * the stats queue operates in reverse. The driver initializes the virtqueue
235 * with a single buffer. From that point forward, all conversations consist of
236 * a hypervisor request (a call to this function) which directs us to refill
237 * the virtqueue with a fresh stats buffer. Since stats collection can sleep,
238 * we notify our kthread which does the actual work via stats_handle_request().
240 static void stats_request(struct virtqueue
*vq
)
242 struct virtio_balloon
*vb
= vq
->vdev
->priv
;
244 vb
->need_stats_update
= 1;
245 wake_up(&vb
->config_change
);
248 static void stats_handle_request(struct virtio_balloon
*vb
)
250 struct virtqueue
*vq
;
251 struct scatterlist sg
;
254 vb
->need_stats_update
= 0;
255 update_balloon_stats(vb
);
258 if (!virtqueue_get_buf(vq
, &len
))
260 sg_init_one(&sg
, vb
->stats
, sizeof(vb
->stats
));
261 if (virtqueue_add_outbuf(vq
, &sg
, 1, vb
, GFP_KERNEL
) < 0)
266 static void virtballoon_changed(struct virtio_device
*vdev
)
268 struct virtio_balloon
*vb
= vdev
->priv
;
270 wake_up(&vb
->config_change
);
273 static inline s64
towards_target(struct virtio_balloon
*vb
)
278 vb
->vdev
->config
->get(vb
->vdev
,
279 offsetof(struct virtio_balloon_config
, num_pages
),
281 target
= le32_to_cpu(v
);
282 return target
- vb
->num_pages
;
285 static void update_balloon_size(struct virtio_balloon
*vb
)
287 __le32 actual
= cpu_to_le32(vb
->num_pages
);
289 vb
->vdev
->config
->set(vb
->vdev
,
290 offsetof(struct virtio_balloon_config
, actual
),
291 &actual
, sizeof(actual
));
294 static int balloon(void *_vballoon
)
296 struct virtio_balloon
*vb
= _vballoon
;
299 while (!kthread_should_stop()) {
303 wait_event_interruptible(vb
->config_change
,
304 (diff
= towards_target(vb
)) != 0
305 || vb
->need_stats_update
306 || kthread_should_stop()
307 || freezing(current
));
308 if (vb
->need_stats_update
)
309 stats_handle_request(vb
);
311 fill_balloon(vb
, diff
);
313 leak_balloon(vb
, -diff
);
314 update_balloon_size(vb
);
317 * For large balloon changes, we could spend a lot of time
318 * and always have work to do. Be nice if preempt disabled.
325 static int init_vqs(struct virtio_balloon
*vb
)
327 struct virtqueue
*vqs
[3];
328 vq_callback_t
*callbacks
[] = { balloon_ack
, balloon_ack
, stats_request
};
329 const char *names
[] = { "inflate", "deflate", "stats" };
333 * We expect two virtqueues: inflate and deflate, and
336 nvqs
= virtio_has_feature(vb
->vdev
, VIRTIO_BALLOON_F_STATS_VQ
) ? 3 : 2;
337 err
= vb
->vdev
->config
->find_vqs(vb
->vdev
, nvqs
, vqs
, callbacks
, names
);
341 vb
->inflate_vq
= vqs
[0];
342 vb
->deflate_vq
= vqs
[1];
343 if (virtio_has_feature(vb
->vdev
, VIRTIO_BALLOON_F_STATS_VQ
)) {
344 struct scatterlist sg
;
345 vb
->stats_vq
= vqs
[2];
348 * Prime this virtqueue with one buffer so the hypervisor can
349 * use it to signal us later.
351 sg_init_one(&sg
, vb
->stats
, sizeof vb
->stats
);
352 if (virtqueue_add_outbuf(vb
->stats_vq
, &sg
, 1, vb
, GFP_KERNEL
)
355 virtqueue_kick(vb
->stats_vq
);
360 static const struct address_space_operations virtio_balloon_aops
;
361 #ifdef CONFIG_BALLOON_COMPACTION
363 * virtballoon_migratepage - perform the balloon page migration on behalf of
364 * a compation thread. (called under page lock)
365 * @mapping: the page->mapping which will be assigned to the new migrated page.
366 * @newpage: page that will replace the isolated page after migration finishes.
367 * @page : the isolated (old) page that is about to be migrated to newpage.
368 * @mode : compaction mode -- not used for balloon page migration.
370 * After a ballooned page gets isolated by compaction procedures, this is the
371 * function that performs the page migration on behalf of a compaction thread
372 * The page migration for virtio balloon is done in a simple swap fashion which
373 * follows these two macro steps:
374 * 1) insert newpage into vb->pages list and update the host about it;
375 * 2) update the host about the old page removed from vb->pages list;
377 * This function preforms the balloon page migration task.
378 * Called through balloon_mapping->a_ops->migratepage
380 int virtballoon_migratepage(struct address_space
*mapping
,
381 struct page
*newpage
, struct page
*page
, enum migrate_mode mode
)
383 struct balloon_dev_info
*vb_dev_info
= balloon_page_device(page
);
384 struct virtio_balloon
*vb
;
387 BUG_ON(!vb_dev_info
);
389 vb
= vb_dev_info
->balloon_device
;
392 * In order to avoid lock contention while migrating pages concurrently
393 * to leak_balloon() or fill_balloon() we just give up the balloon_lock
394 * this turn, as it is easier to retry the page migration later.
395 * This also prevents fill_balloon() getting stuck into a mutex
396 * recursion in the case it ends up triggering memory compaction
397 * while it is attempting to inflate the ballon.
399 if (!mutex_trylock(&vb
->balloon_lock
))
402 /* balloon's page migration 1st step -- inflate "newpage" */
403 spin_lock_irqsave(&vb_dev_info
->pages_lock
, flags
);
404 balloon_page_insert(newpage
, mapping
, &vb_dev_info
->pages
);
405 vb_dev_info
->isolated_pages
--;
406 spin_unlock_irqrestore(&vb_dev_info
->pages_lock
, flags
);
407 vb
->num_pfns
= VIRTIO_BALLOON_PAGES_PER_PAGE
;
408 set_page_pfns(vb
->pfns
, newpage
);
409 tell_host(vb
, vb
->inflate_vq
);
412 * balloon's page migration 2nd step -- deflate "page"
414 * It's safe to delete page->lru here because this page is at
415 * an isolated migration list, and this step is expected to happen here
417 balloon_page_delete(page
);
418 vb
->num_pfns
= VIRTIO_BALLOON_PAGES_PER_PAGE
;
419 set_page_pfns(vb
->pfns
, page
);
420 tell_host(vb
, vb
->deflate_vq
);
422 mutex_unlock(&vb
->balloon_lock
);
424 return MIGRATEPAGE_BALLOON_SUCCESS
;
427 /* define the balloon_mapping->a_ops callback to allow balloon page migration */
428 static const struct address_space_operations virtio_balloon_aops
= {
429 .migratepage
= virtballoon_migratepage
,
431 #endif /* CONFIG_BALLOON_COMPACTION */
433 static int virtballoon_probe(struct virtio_device
*vdev
)
435 struct virtio_balloon
*vb
;
436 struct address_space
*vb_mapping
;
437 struct balloon_dev_info
*vb_devinfo
;
440 vdev
->priv
= vb
= kmalloc(sizeof(*vb
), GFP_KERNEL
);
447 mutex_init(&vb
->balloon_lock
);
448 init_waitqueue_head(&vb
->config_change
);
449 init_waitqueue_head(&vb
->acked
);
451 vb
->need_stats_update
= 0;
453 vb_devinfo
= balloon_devinfo_alloc(vb
);
454 if (IS_ERR(vb_devinfo
)) {
455 err
= PTR_ERR(vb_devinfo
);
459 vb_mapping
= balloon_mapping_alloc(vb_devinfo
,
460 (balloon_compaction_check()) ?
461 &virtio_balloon_aops
: NULL
);
462 if (IS_ERR(vb_mapping
)) {
464 * IS_ERR(vb_mapping) && PTR_ERR(vb_mapping) == -EOPNOTSUPP
465 * This means !CONFIG_BALLOON_COMPACTION, otherwise we get off.
467 err
= PTR_ERR(vb_mapping
);
468 if (err
!= -EOPNOTSUPP
)
469 goto out_free_vb_devinfo
;
472 vb
->vb_dev_info
= vb_devinfo
;
476 goto out_free_vb_mapping
;
478 vb
->thread
= kthread_run(balloon
, vb
, "vballoon");
479 if (IS_ERR(vb
->thread
)) {
480 err
= PTR_ERR(vb
->thread
);
487 vdev
->config
->del_vqs(vdev
);
489 balloon_mapping_free(vb_mapping
);
491 balloon_devinfo_free(vb_devinfo
);
498 static void remove_common(struct virtio_balloon
*vb
)
500 /* There might be pages left in the balloon: free them. */
501 while (vb
->num_pages
)
502 leak_balloon(vb
, vb
->num_pages
);
503 update_balloon_size(vb
);
505 /* Now we reset the device so we can clean up the queues. */
506 vb
->vdev
->config
->reset(vb
->vdev
);
508 vb
->vdev
->config
->del_vqs(vb
->vdev
);
511 static void virtballoon_remove(struct virtio_device
*vdev
)
513 struct virtio_balloon
*vb
= vdev
->priv
;
515 kthread_stop(vb
->thread
);
517 balloon_mapping_free(vb
->vb_dev_info
->mapping
);
518 balloon_devinfo_free(vb
->vb_dev_info
);
523 static int virtballoon_freeze(struct virtio_device
*vdev
)
525 struct virtio_balloon
*vb
= vdev
->priv
;
528 * The kthread is already frozen by the PM core before this
529 * function is called.
536 static int virtballoon_restore(struct virtio_device
*vdev
)
538 struct virtio_balloon
*vb
= vdev
->priv
;
541 ret
= init_vqs(vdev
->priv
);
545 fill_balloon(vb
, towards_target(vb
));
546 update_balloon_size(vb
);
551 static unsigned int features
[] = {
552 VIRTIO_BALLOON_F_MUST_TELL_HOST
,
553 VIRTIO_BALLOON_F_STATS_VQ
,
556 static struct virtio_driver virtio_balloon_driver
= {
557 .feature_table
= features
,
558 .feature_table_size
= ARRAY_SIZE(features
),
559 .driver
.name
= KBUILD_MODNAME
,
560 .driver
.owner
= THIS_MODULE
,
561 .id_table
= id_table
,
562 .probe
= virtballoon_probe
,
563 .remove
= virtballoon_remove
,
564 .config_changed
= virtballoon_changed
,
566 .freeze
= virtballoon_freeze
,
567 .restore
= virtballoon_restore
,
571 module_virtio_driver(virtio_balloon_driver
);
572 MODULE_DEVICE_TABLE(virtio
, id_table
);
573 MODULE_DESCRIPTION("Virtio balloon driver");
574 MODULE_LICENSE("GPL");