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>
31 #include <linux/oom.h>
32 #include <linux/wait.h>
35 * Balloon device works in 4K page units. So each page is pointed to by
36 * multiple balloon pages. All memory counters in this driver are in balloon
39 #define VIRTIO_BALLOON_PAGES_PER_PAGE (unsigned)(PAGE_SIZE >> VIRTIO_BALLOON_PFN_SHIFT)
40 #define VIRTIO_BALLOON_ARRAY_PFNS_MAX 256
41 #define OOM_VBALLOON_DEFAULT_PAGES 256
42 #define VIRTBALLOON_OOM_NOTIFY_PRIORITY 80
44 static int oom_pages = OOM_VBALLOON_DEFAULT_PAGES;
45 module_param(oom_pages, int, S_IRUSR | S_IWUSR);
46 MODULE_PARM_DESC(oom_pages, "pages to free on OOM");
48 struct virtio_balloon {
49 struct virtio_device *vdev;
50 struct virtqueue *inflate_vq, *deflate_vq, *stats_vq;
52 /* Where the ballooning thread waits for config to change. */
53 wait_queue_head_t config_change;
55 /* The thread servicing the balloon. */
56 struct task_struct *thread;
58 /* Waiting for host to ack the pages we released. */
59 wait_queue_head_t acked;
61 /* Number of balloon pages we've told the Host we're not using. */
62 unsigned int num_pages;
64 * The pages we've told the Host we're not using are enqueued
65 * at vb_dev_info->pages list.
66 * Each page on this list adds VIRTIO_BALLOON_PAGES_PER_PAGE
69 struct balloon_dev_info vb_dev_info;
71 /* Synchronize access/update to this struct virtio_balloon elements */
72 struct mutex balloon_lock;
74 /* The array of pfns we tell the Host about. */
75 unsigned int num_pfns;
76 __virtio32 pfns[VIRTIO_BALLOON_ARRAY_PFNS_MAX];
78 /* Memory statistics */
79 int need_stats_update;
80 struct virtio_balloon_stat stats[VIRTIO_BALLOON_S_NR];
82 /* To register callback in oom notifier call chain */
83 struct notifier_block nb;
86 static struct virtio_device_id id_table[] = {
87 { VIRTIO_ID_BALLOON, VIRTIO_DEV_ANY_ID },
91 static u32 page_to_balloon_pfn(struct page *page)
93 unsigned long pfn = page_to_pfn(page);
95 BUILD_BUG_ON(PAGE_SHIFT < VIRTIO_BALLOON_PFN_SHIFT);
96 /* Convert pfn from Linux page size to balloon page size. */
97 return pfn * VIRTIO_BALLOON_PAGES_PER_PAGE;
100 static struct page *balloon_pfn_to_page(u32 pfn)
102 BUG_ON(pfn % VIRTIO_BALLOON_PAGES_PER_PAGE);
103 return pfn_to_page(pfn / VIRTIO_BALLOON_PAGES_PER_PAGE);
106 static void balloon_ack(struct virtqueue *vq)
108 struct virtio_balloon *vb = vq->vdev->priv;
113 static void tell_host(struct virtio_balloon *vb, struct virtqueue *vq)
115 struct scatterlist sg;
118 sg_init_one(&sg, vb->pfns, sizeof(vb->pfns[0]) * vb->num_pfns);
120 /* We should always be able to add one buffer to an empty queue. */
121 virtqueue_add_outbuf(vq, &sg, 1, vb, GFP_KERNEL);
124 /* When host has read buffer, this completes via balloon_ack */
125 wait_event(vb->acked, virtqueue_get_buf(vq, &len));
128 static void set_page_pfns(struct virtio_balloon *vb,
129 __virtio32 pfns[], struct page *page)
133 /* Set balloon pfns pointing at this page.
134 * Note that the first pfn points at start of the page. */
135 for (i = 0; i < VIRTIO_BALLOON_PAGES_PER_PAGE; i++)
136 pfns[i] = cpu_to_virtio32(vb->vdev,
137 page_to_balloon_pfn(page) + i);
140 static void fill_balloon(struct virtio_balloon *vb, size_t num)
142 struct balloon_dev_info *vb_dev_info = &vb->vb_dev_info;
144 /* We can only do one array worth at a time. */
145 num = min(num, ARRAY_SIZE(vb->pfns));
147 mutex_lock(&vb->balloon_lock);
148 for (vb->num_pfns = 0; vb->num_pfns < num;
149 vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) {
150 struct page *page = balloon_page_enqueue(vb_dev_info);
153 dev_info_ratelimited(&vb->vdev->dev,
154 "Out of puff! Can't get %u pages\n",
155 VIRTIO_BALLOON_PAGES_PER_PAGE);
156 /* Sleep for at least 1/5 of a second before retry. */
160 set_page_pfns(vb, vb->pfns + vb->num_pfns, page);
161 vb->num_pages += VIRTIO_BALLOON_PAGES_PER_PAGE;
162 if (!virtio_has_feature(vb->vdev,
163 VIRTIO_BALLOON_F_DEFLATE_ON_OOM))
164 adjust_managed_page_count(page, -1);
167 /* Did we get any? */
168 if (vb->num_pfns != 0)
169 tell_host(vb, vb->inflate_vq);
170 mutex_unlock(&vb->balloon_lock);
173 static void release_pages_balloon(struct virtio_balloon *vb)
178 /* Find pfns pointing at start of each page, get pages and free them. */
179 for (i = 0; i < vb->num_pfns; i += VIRTIO_BALLOON_PAGES_PER_PAGE) {
180 page = balloon_pfn_to_page(virtio32_to_cpu(vb->vdev,
182 if (!virtio_has_feature(vb->vdev,
183 VIRTIO_BALLOON_F_DEFLATE_ON_OOM))
184 adjust_managed_page_count(page, 1);
185 put_page(page); /* balloon reference */
189 static unsigned leak_balloon(struct virtio_balloon *vb, size_t num)
191 unsigned num_freed_pages;
193 struct balloon_dev_info *vb_dev_info = &vb->vb_dev_info;
195 /* We can only do one array worth at a time. */
196 num = min(num, ARRAY_SIZE(vb->pfns));
198 mutex_lock(&vb->balloon_lock);
199 /* We can't release more pages than taken */
200 num = min(num, (size_t)vb->num_pages);
201 for (vb->num_pfns = 0; vb->num_pfns < num;
202 vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) {
203 page = balloon_page_dequeue(vb_dev_info);
206 set_page_pfns(vb, vb->pfns + vb->num_pfns, page);
207 vb->num_pages -= VIRTIO_BALLOON_PAGES_PER_PAGE;
210 num_freed_pages = vb->num_pfns;
213 * virtio_has_feature(vdev, VIRTIO_BALLOON_F_MUST_TELL_HOST);
214 * is true, we *have* to do it in this order
216 if (vb->num_pfns != 0)
217 tell_host(vb, vb->deflate_vq);
218 release_pages_balloon(vb);
219 mutex_unlock(&vb->balloon_lock);
220 return num_freed_pages;
223 static inline void update_stat(struct virtio_balloon *vb, int idx,
226 BUG_ON(idx >= VIRTIO_BALLOON_S_NR);
227 vb->stats[idx].tag = cpu_to_virtio16(vb->vdev, tag);
228 vb->stats[idx].val = cpu_to_virtio64(vb->vdev, val);
231 #define pages_to_bytes(x) ((u64)(x) << PAGE_SHIFT)
233 static void update_balloon_stats(struct virtio_balloon *vb)
235 unsigned long events[NR_VM_EVENT_ITEMS];
239 all_vm_events(events);
242 #ifdef CONFIG_VM_EVENT_COUNTERS
243 update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_IN,
244 pages_to_bytes(events[PSWPIN]));
245 update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_OUT,
246 pages_to_bytes(events[PSWPOUT]));
247 update_stat(vb, idx++, VIRTIO_BALLOON_S_MAJFLT, events[PGMAJFAULT]);
248 update_stat(vb, idx++, VIRTIO_BALLOON_S_MINFLT, events[PGFAULT]);
250 update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMFREE,
251 pages_to_bytes(i.freeram));
252 update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMTOT,
253 pages_to_bytes(i.totalram));
257 * While most virtqueues communicate guest-initiated requests to the hypervisor,
258 * the stats queue operates in reverse. The driver initializes the virtqueue
259 * with a single buffer. From that point forward, all conversations consist of
260 * a hypervisor request (a call to this function) which directs us to refill
261 * the virtqueue with a fresh stats buffer. Since stats collection can sleep,
262 * we notify our kthread which does the actual work via stats_handle_request().
264 static void stats_request(struct virtqueue *vq)
266 struct virtio_balloon *vb = vq->vdev->priv;
268 vb->need_stats_update = 1;
269 wake_up(&vb->config_change);
272 static void stats_handle_request(struct virtio_balloon *vb)
274 struct virtqueue *vq;
275 struct scatterlist sg;
278 vb->need_stats_update = 0;
279 update_balloon_stats(vb);
282 if (!virtqueue_get_buf(vq, &len))
284 sg_init_one(&sg, vb->stats, sizeof(vb->stats));
285 virtqueue_add_outbuf(vq, &sg, 1, vb, GFP_KERNEL);
289 static void virtballoon_changed(struct virtio_device *vdev)
291 struct virtio_balloon *vb = vdev->priv;
293 wake_up(&vb->config_change);
296 static inline s64 towards_target(struct virtio_balloon *vb)
301 virtio_cread(vb->vdev, struct virtio_balloon_config, num_pages,
304 /* Legacy balloon config space is LE, unlike all other devices. */
305 if (!virtio_has_feature(vb->vdev, VIRTIO_F_VERSION_1))
306 num_pages = le32_to_cpu((__force __le32)num_pages);
309 return target - vb->num_pages;
312 static void update_balloon_size(struct virtio_balloon *vb)
314 u32 actual = vb->num_pages;
316 /* Legacy balloon config space is LE, unlike all other devices. */
317 if (!virtio_has_feature(vb->vdev, VIRTIO_F_VERSION_1))
318 actual = (__force u32)cpu_to_le32(actual);
320 virtio_cwrite(vb->vdev, struct virtio_balloon_config, actual,
325 * virtballoon_oom_notify - release pages when system is under severe
326 * memory pressure (called from out_of_memory())
327 * @self : notifier block struct
329 * @parm : returned - number of freed pages
331 * The balancing of memory by use of the virtio balloon should not cause
332 * the termination of processes while there are pages in the balloon.
333 * If virtio balloon manages to release some memory, it will make the
334 * system return and retry the allocation that forced the OOM killer
337 static int virtballoon_oom_notify(struct notifier_block *self,
338 unsigned long dummy, void *parm)
340 struct virtio_balloon *vb;
341 unsigned long *freed;
342 unsigned num_freed_pages;
344 vb = container_of(self, struct virtio_balloon, nb);
345 if (!virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_DEFLATE_ON_OOM))
349 num_freed_pages = leak_balloon(vb, oom_pages);
350 update_balloon_size(vb);
351 *freed += num_freed_pages;
356 static int balloon(void *_vballoon)
358 struct virtio_balloon *vb = _vballoon;
359 DEFINE_WAIT_FUNC(wait, woken_wake_function);
362 while (!kthread_should_stop()) {
367 add_wait_queue(&vb->config_change, &wait);
369 if ((diff = towards_target(vb)) != 0 ||
370 vb->need_stats_update ||
371 kthread_should_stop() ||
374 wait_woken(&wait, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
376 remove_wait_queue(&vb->config_change, &wait);
378 if (vb->need_stats_update)
379 stats_handle_request(vb);
381 fill_balloon(vb, diff);
383 leak_balloon(vb, -diff);
384 update_balloon_size(vb);
387 * For large balloon changes, we could spend a lot of time
388 * and always have work to do. Be nice if preempt disabled.
395 static int init_vqs(struct virtio_balloon *vb)
397 struct virtqueue *vqs[3];
398 vq_callback_t *callbacks[] = { balloon_ack, balloon_ack, stats_request };
399 const char *names[] = { "inflate", "deflate", "stats" };
403 * We expect two virtqueues: inflate and deflate, and
406 nvqs = virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ) ? 3 : 2;
407 err = vb->vdev->config->find_vqs(vb->vdev, nvqs, vqs, callbacks, names);
411 vb->inflate_vq = vqs[0];
412 vb->deflate_vq = vqs[1];
413 if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ)) {
414 struct scatterlist sg;
415 vb->stats_vq = vqs[2];
418 * Prime this virtqueue with one buffer so the hypervisor can
419 * use it to signal us later (it can't be broken yet!).
421 update_balloon_stats(vb);
423 sg_init_one(&sg, vb->stats, sizeof vb->stats);
424 if (virtqueue_add_outbuf(vb->stats_vq, &sg, 1, vb, GFP_KERNEL)
427 virtqueue_kick(vb->stats_vq);
432 #ifdef CONFIG_BALLOON_COMPACTION
434 * virtballoon_migratepage - perform the balloon page migration on behalf of
435 * a compation thread. (called under page lock)
436 * @vb_dev_info: the balloon device
437 * @newpage: page that will replace the isolated page after migration finishes.
438 * @page : the isolated (old) page that is about to be migrated to newpage.
439 * @mode : compaction mode -- not used for balloon page migration.
441 * After a ballooned page gets isolated by compaction procedures, this is the
442 * function that performs the page migration on behalf of a compaction thread
443 * The page migration for virtio balloon is done in a simple swap fashion which
444 * follows these two macro steps:
445 * 1) insert newpage into vb->pages list and update the host about it;
446 * 2) update the host about the old page removed from vb->pages list;
448 * This function preforms the balloon page migration task.
449 * Called through balloon_mapping->a_ops->migratepage
451 static int virtballoon_migratepage(struct balloon_dev_info *vb_dev_info,
452 struct page *newpage, struct page *page, enum migrate_mode mode)
454 struct virtio_balloon *vb = container_of(vb_dev_info,
455 struct virtio_balloon, vb_dev_info);
459 * In order to avoid lock contention while migrating pages concurrently
460 * to leak_balloon() or fill_balloon() we just give up the balloon_lock
461 * this turn, as it is easier to retry the page migration later.
462 * This also prevents fill_balloon() getting stuck into a mutex
463 * recursion in the case it ends up triggering memory compaction
464 * while it is attempting to inflate the ballon.
466 if (!mutex_trylock(&vb->balloon_lock))
469 get_page(newpage); /* balloon reference */
472 * When we migrate a page to a different zone and adjusted the
473 * managed page count when inflating, we have to fixup the count of
474 * both involved zones.
476 if (!virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_DEFLATE_ON_OOM) &&
477 page_zone(page) != page_zone(newpage)) {
478 adjust_managed_page_count(page, 1);
479 adjust_managed_page_count(newpage, -1);
482 /* balloon's page migration 1st step -- inflate "newpage" */
483 spin_lock_irqsave(&vb_dev_info->pages_lock, flags);
484 balloon_page_insert(vb_dev_info, newpage);
485 vb_dev_info->isolated_pages--;
486 __count_vm_event(BALLOON_MIGRATE);
487 spin_unlock_irqrestore(&vb_dev_info->pages_lock, flags);
488 vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE;
489 set_page_pfns(vb, vb->pfns, newpage);
490 tell_host(vb, vb->inflate_vq);
492 /* balloon's page migration 2nd step -- deflate "page" */
493 spin_lock_irqsave(&vb_dev_info->pages_lock, flags);
494 balloon_page_delete(page);
495 spin_unlock_irqrestore(&vb_dev_info->pages_lock, flags);
496 vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE;
497 set_page_pfns(vb, vb->pfns, page);
498 tell_host(vb, vb->deflate_vq);
500 mutex_unlock(&vb->balloon_lock);
502 put_page(page); /* balloon reference */
504 return MIGRATEPAGE_SUCCESS;
506 #endif /* CONFIG_BALLOON_COMPACTION */
508 static int virtballoon_probe(struct virtio_device *vdev)
510 struct virtio_balloon *vb;
513 if (!vdev->config->get) {
514 dev_err(&vdev->dev, "%s failure: config access disabled\n",
519 vdev->priv = vb = kmalloc(sizeof(*vb), GFP_KERNEL);
526 mutex_init(&vb->balloon_lock);
527 init_waitqueue_head(&vb->config_change);
528 init_waitqueue_head(&vb->acked);
530 vb->need_stats_update = 0;
532 balloon_devinfo_init(&vb->vb_dev_info);
533 #ifdef CONFIG_BALLOON_COMPACTION
534 vb->vb_dev_info.migratepage = virtballoon_migratepage;
541 vb->nb.notifier_call = virtballoon_oom_notify;
542 vb->nb.priority = VIRTBALLOON_OOM_NOTIFY_PRIORITY;
543 err = register_oom_notifier(&vb->nb);
547 virtio_device_ready(vdev);
549 vb->thread = kthread_run(balloon, vb, "vballoon");
550 if (IS_ERR(vb->thread)) {
551 err = PTR_ERR(vb->thread);
558 unregister_oom_notifier(&vb->nb);
560 vdev->config->del_vqs(vdev);
567 static void remove_common(struct virtio_balloon *vb)
569 /* There might be pages left in the balloon: free them. */
570 while (vb->num_pages)
571 leak_balloon(vb, vb->num_pages);
572 update_balloon_size(vb);
574 /* Now we reset the device so we can clean up the queues. */
575 vb->vdev->config->reset(vb->vdev);
577 vb->vdev->config->del_vqs(vb->vdev);
580 static void virtballoon_remove(struct virtio_device *vdev)
582 struct virtio_balloon *vb = vdev->priv;
584 unregister_oom_notifier(&vb->nb);
585 kthread_stop(vb->thread);
590 #ifdef CONFIG_PM_SLEEP
591 static int virtballoon_freeze(struct virtio_device *vdev)
593 struct virtio_balloon *vb = vdev->priv;
596 * The kthread is already frozen by the PM core before this
597 * function is called.
604 static int virtballoon_restore(struct virtio_device *vdev)
606 struct virtio_balloon *vb = vdev->priv;
609 ret = init_vqs(vdev->priv);
613 virtio_device_ready(vdev);
615 fill_balloon(vb, towards_target(vb));
616 update_balloon_size(vb);
621 static unsigned int features[] = {
622 VIRTIO_BALLOON_F_MUST_TELL_HOST,
623 VIRTIO_BALLOON_F_STATS_VQ,
624 VIRTIO_BALLOON_F_DEFLATE_ON_OOM,
627 static struct virtio_driver virtio_balloon_driver = {
628 .feature_table = features,
629 .feature_table_size = ARRAY_SIZE(features),
630 .driver.name = KBUILD_MODNAME,
631 .driver.owner = THIS_MODULE,
632 .id_table = id_table,
633 .probe = virtballoon_probe,
634 .remove = virtballoon_remove,
635 .config_changed = virtballoon_changed,
636 #ifdef CONFIG_PM_SLEEP
637 .freeze = virtballoon_freeze,
638 .restore = virtballoon_restore,
642 module_virtio_driver(virtio_balloon_driver);
643 MODULE_DEVICE_TABLE(virtio, id_table);
644 MODULE_DESCRIPTION("Virtio balloon driver");
645 MODULE_LICENSE("GPL");