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_buf(vq
, &sg
, 1, 0, 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 if (printk_ratelimit())
143 dev_printk(KERN_INFO
, &vb
->vdev
->dev
,
144 "Out of puff! Can't get %u pages\n",
145 VIRTIO_BALLOON_PAGES_PER_PAGE
);
146 /* Sleep for at least 1/5 of a second before retry. */
150 set_page_pfns(vb
->pfns
+ vb
->num_pfns
, page
);
151 vb
->num_pages
+= VIRTIO_BALLOON_PAGES_PER_PAGE
;
155 /* Did we get any? */
156 if (vb
->num_pfns
!= 0)
157 tell_host(vb
, vb
->inflate_vq
);
158 mutex_unlock(&vb
->balloon_lock
);
161 static void release_pages_by_pfn(const u32 pfns
[], unsigned int num
)
165 /* Find pfns pointing at start of each page, get pages and free them. */
166 for (i
= 0; i
< num
; i
+= VIRTIO_BALLOON_PAGES_PER_PAGE
) {
167 balloon_page_free(balloon_pfn_to_page(pfns
[i
]));
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 tell_host(vb
, vb
->deflate_vq
);
196 mutex_unlock(&vb
->balloon_lock
);
197 release_pages_by_pfn(vb
->pfns
, vb
->num_pfns
);
200 static inline void update_stat(struct virtio_balloon
*vb
, int idx
,
203 BUG_ON(idx
>= VIRTIO_BALLOON_S_NR
);
204 vb
->stats
[idx
].tag
= tag
;
205 vb
->stats
[idx
].val
= val
;
208 #define pages_to_bytes(x) ((u64)(x) << PAGE_SHIFT)
210 static void update_balloon_stats(struct virtio_balloon
*vb
)
212 unsigned long events
[NR_VM_EVENT_ITEMS
];
216 all_vm_events(events
);
219 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_SWAP_IN
,
220 pages_to_bytes(events
[PSWPIN
]));
221 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_SWAP_OUT
,
222 pages_to_bytes(events
[PSWPOUT
]));
223 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_MAJFLT
, events
[PGMAJFAULT
]);
224 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_MINFLT
, events
[PGFAULT
]);
225 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_MEMFREE
,
226 pages_to_bytes(i
.freeram
));
227 update_stat(vb
, idx
++, VIRTIO_BALLOON_S_MEMTOT
,
228 pages_to_bytes(i
.totalram
));
232 * While most virtqueues communicate guest-initiated requests to the hypervisor,
233 * the stats queue operates in reverse. The driver initializes the virtqueue
234 * with a single buffer. From that point forward, all conversations consist of
235 * a hypervisor request (a call to this function) which directs us to refill
236 * the virtqueue with a fresh stats buffer. Since stats collection can sleep,
237 * we notify our kthread which does the actual work via stats_handle_request().
239 static void stats_request(struct virtqueue
*vq
)
241 struct virtio_balloon
*vb
= vq
->vdev
->priv
;
243 vb
->need_stats_update
= 1;
244 wake_up(&vb
->config_change
);
247 static void stats_handle_request(struct virtio_balloon
*vb
)
249 struct virtqueue
*vq
;
250 struct scatterlist sg
;
253 vb
->need_stats_update
= 0;
254 update_balloon_stats(vb
);
257 if (!virtqueue_get_buf(vq
, &len
))
259 sg_init_one(&sg
, vb
->stats
, sizeof(vb
->stats
));
260 if (virtqueue_add_buf(vq
, &sg
, 1, 0, vb
, GFP_KERNEL
) < 0)
265 static void virtballoon_changed(struct virtio_device
*vdev
)
267 struct virtio_balloon
*vb
= vdev
->priv
;
269 wake_up(&vb
->config_change
);
272 static inline s64
towards_target(struct virtio_balloon
*vb
)
277 vb
->vdev
->config
->get(vb
->vdev
,
278 offsetof(struct virtio_balloon_config
, num_pages
),
280 target
= le32_to_cpu(v
);
281 return target
- vb
->num_pages
;
284 static void update_balloon_size(struct virtio_balloon
*vb
)
286 __le32 actual
= cpu_to_le32(vb
->num_pages
);
288 vb
->vdev
->config
->set(vb
->vdev
,
289 offsetof(struct virtio_balloon_config
, actual
),
290 &actual
, sizeof(actual
));
293 static int balloon(void *_vballoon
)
295 struct virtio_balloon
*vb
= _vballoon
;
298 while (!kthread_should_stop()) {
302 wait_event_interruptible(vb
->config_change
,
303 (diff
= towards_target(vb
)) != 0
304 || vb
->need_stats_update
305 || kthread_should_stop()
306 || freezing(current
));
307 if (vb
->need_stats_update
)
308 stats_handle_request(vb
);
310 fill_balloon(vb
, diff
);
312 leak_balloon(vb
, -diff
);
313 update_balloon_size(vb
);
318 static int init_vqs(struct virtio_balloon
*vb
)
320 struct virtqueue
*vqs
[3];
321 vq_callback_t
*callbacks
[] = { balloon_ack
, balloon_ack
, stats_request
};
322 const char *names
[] = { "inflate", "deflate", "stats" };
326 * We expect two virtqueues: inflate and deflate, and
329 nvqs
= virtio_has_feature(vb
->vdev
, VIRTIO_BALLOON_F_STATS_VQ
) ? 3 : 2;
330 err
= vb
->vdev
->config
->find_vqs(vb
->vdev
, nvqs
, vqs
, callbacks
, names
);
334 vb
->inflate_vq
= vqs
[0];
335 vb
->deflate_vq
= vqs
[1];
336 if (virtio_has_feature(vb
->vdev
, VIRTIO_BALLOON_F_STATS_VQ
)) {
337 struct scatterlist sg
;
338 vb
->stats_vq
= vqs
[2];
341 * Prime this virtqueue with one buffer so the hypervisor can
342 * use it to signal us later.
344 sg_init_one(&sg
, vb
->stats
, sizeof vb
->stats
);
345 if (virtqueue_add_buf(vb
->stats_vq
, &sg
, 1, 0, vb
, GFP_KERNEL
)
348 virtqueue_kick(vb
->stats_vq
);
353 static const struct address_space_operations virtio_balloon_aops
;
354 #ifdef CONFIG_BALLOON_COMPACTION
356 * virtballoon_migratepage - perform the balloon page migration on behalf of
357 * a compation thread. (called under page lock)
358 * @mapping: the page->mapping which will be assigned to the new migrated page.
359 * @newpage: page that will replace the isolated page after migration finishes.
360 * @page : the isolated (old) page that is about to be migrated to newpage.
361 * @mode : compaction mode -- not used for balloon page migration.
363 * After a ballooned page gets isolated by compaction procedures, this is the
364 * function that performs the page migration on behalf of a compaction thread
365 * The page migration for virtio balloon is done in a simple swap fashion which
366 * follows these two macro steps:
367 * 1) insert newpage into vb->pages list and update the host about it;
368 * 2) update the host about the old page removed from vb->pages list;
370 * This function preforms the balloon page migration task.
371 * Called through balloon_mapping->a_ops->migratepage
373 int virtballoon_migratepage(struct address_space
*mapping
,
374 struct page
*newpage
, struct page
*page
, enum migrate_mode mode
)
376 struct balloon_dev_info
*vb_dev_info
= balloon_page_device(page
);
377 struct virtio_balloon
*vb
;
380 BUG_ON(!vb_dev_info
);
382 vb
= vb_dev_info
->balloon_device
;
385 * In order to avoid lock contention while migrating pages concurrently
386 * to leak_balloon() or fill_balloon() we just give up the balloon_lock
387 * this turn, as it is easier to retry the page migration later.
388 * This also prevents fill_balloon() getting stuck into a mutex
389 * recursion in the case it ends up triggering memory compaction
390 * while it is attempting to inflate the ballon.
392 if (!mutex_trylock(&vb
->balloon_lock
))
395 /* balloon's page migration 1st step -- inflate "newpage" */
396 spin_lock_irqsave(&vb_dev_info
->pages_lock
, flags
);
397 balloon_page_insert(newpage
, mapping
, &vb_dev_info
->pages
);
398 vb_dev_info
->isolated_pages
--;
399 spin_unlock_irqrestore(&vb_dev_info
->pages_lock
, flags
);
400 vb
->num_pfns
= VIRTIO_BALLOON_PAGES_PER_PAGE
;
401 set_page_pfns(vb
->pfns
, newpage
);
402 tell_host(vb
, vb
->inflate_vq
);
405 * balloon's page migration 2nd step -- deflate "page"
407 * It's safe to delete page->lru here because this page is at
408 * an isolated migration list, and this step is expected to happen here
410 balloon_page_delete(page
);
411 vb
->num_pfns
= VIRTIO_BALLOON_PAGES_PER_PAGE
;
412 set_page_pfns(vb
->pfns
, page
);
413 tell_host(vb
, vb
->deflate_vq
);
415 mutex_unlock(&vb
->balloon_lock
);
417 return MIGRATEPAGE_BALLOON_SUCCESS
;
420 /* define the balloon_mapping->a_ops callback to allow balloon page migration */
421 static const struct address_space_operations virtio_balloon_aops
= {
422 .migratepage
= virtballoon_migratepage
,
424 #endif /* CONFIG_BALLOON_COMPACTION */
426 static int virtballoon_probe(struct virtio_device
*vdev
)
428 struct virtio_balloon
*vb
;
429 struct address_space
*vb_mapping
;
430 struct balloon_dev_info
*vb_devinfo
;
433 vdev
->priv
= vb
= kmalloc(sizeof(*vb
), GFP_KERNEL
);
440 mutex_init(&vb
->balloon_lock
);
441 init_waitqueue_head(&vb
->config_change
);
442 init_waitqueue_head(&vb
->acked
);
444 vb
->need_stats_update
= 0;
446 vb_devinfo
= balloon_devinfo_alloc(vb
);
447 if (IS_ERR(vb_devinfo
)) {
448 err
= PTR_ERR(vb_devinfo
);
452 vb_mapping
= balloon_mapping_alloc(vb_devinfo
,
453 (balloon_compaction_check()) ?
454 &virtio_balloon_aops
: NULL
);
455 if (IS_ERR(vb_mapping
)) {
457 * IS_ERR(vb_mapping) && PTR_ERR(vb_mapping) == -EOPNOTSUPP
458 * This means !CONFIG_BALLOON_COMPACTION, otherwise we get off.
460 err
= PTR_ERR(vb_mapping
);
461 if (err
!= -EOPNOTSUPP
)
462 goto out_free_vb_devinfo
;
465 vb
->vb_dev_info
= vb_devinfo
;
469 goto out_free_vb_mapping
;
471 vb
->thread
= kthread_run(balloon
, vb
, "vballoon");
472 if (IS_ERR(vb
->thread
)) {
473 err
= PTR_ERR(vb
->thread
);
480 vdev
->config
->del_vqs(vdev
);
482 balloon_mapping_free(vb_mapping
);
484 balloon_devinfo_free(vb_devinfo
);
491 static void remove_common(struct virtio_balloon
*vb
)
493 /* There might be pages left in the balloon: free them. */
494 while (vb
->num_pages
)
495 leak_balloon(vb
, vb
->num_pages
);
496 update_balloon_size(vb
);
498 /* Now we reset the device so we can clean up the queues. */
499 vb
->vdev
->config
->reset(vb
->vdev
);
501 vb
->vdev
->config
->del_vqs(vb
->vdev
);
504 static void __devexit
virtballoon_remove(struct virtio_device
*vdev
)
506 struct virtio_balloon
*vb
= vdev
->priv
;
508 kthread_stop(vb
->thread
);
510 balloon_mapping_free(vb
->vb_dev_info
->mapping
);
511 balloon_devinfo_free(vb
->vb_dev_info
);
516 static int virtballoon_freeze(struct virtio_device
*vdev
)
518 struct virtio_balloon
*vb
= vdev
->priv
;
521 * The kthread is already frozen by the PM core before this
522 * function is called.
529 static int virtballoon_restore(struct virtio_device
*vdev
)
531 struct virtio_balloon
*vb
= vdev
->priv
;
534 ret
= init_vqs(vdev
->priv
);
538 fill_balloon(vb
, towards_target(vb
));
539 update_balloon_size(vb
);
544 static unsigned int features
[] = {
545 VIRTIO_BALLOON_F_MUST_TELL_HOST
,
546 VIRTIO_BALLOON_F_STATS_VQ
,
549 static struct virtio_driver virtio_balloon_driver
= {
550 .feature_table
= features
,
551 .feature_table_size
= ARRAY_SIZE(features
),
552 .driver
.name
= KBUILD_MODNAME
,
553 .driver
.owner
= THIS_MODULE
,
554 .id_table
= id_table
,
555 .probe
= virtballoon_probe
,
556 .remove
= __devexit_p(virtballoon_remove
),
557 .config_changed
= virtballoon_changed
,
559 .freeze
= virtballoon_freeze
,
560 .restore
= virtballoon_restore
,
564 static int __init
init(void)
566 return register_virtio_driver(&virtio_balloon_driver
);
569 static void __exit
fini(void)
571 unregister_virtio_driver(&virtio_balloon_driver
);
576 MODULE_DEVICE_TABLE(virtio
, id_table
);
577 MODULE_DESCRIPTION("Virtio balloon driver");
578 MODULE_LICENSE("GPL");