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/oom.h>
31 #include <linux/wait.h>
33 #include <linux/mount.h>
34 #include <linux/magic.h>
37 * Balloon device works in 4K page units. So each page is pointed to by
38 * multiple balloon pages. All memory counters in this driver are in balloon
41 #define VIRTIO_BALLOON_PAGES_PER_PAGE (unsigned)(PAGE_SIZE >> VIRTIO_BALLOON_PFN_SHIFT)
42 #define VIRTIO_BALLOON_ARRAY_PFNS_MAX 256
43 #define OOM_VBALLOON_DEFAULT_PAGES 256
44 #define VIRTBALLOON_OOM_NOTIFY_PRIORITY 80
46 static int oom_pages = OOM_VBALLOON_DEFAULT_PAGES;
47 module_param(oom_pages, int, S_IRUSR | S_IWUSR);
48 MODULE_PARM_DESC(oom_pages, "pages to free on OOM");
50 #ifdef CONFIG_BALLOON_COMPACTION
51 static struct vfsmount *balloon_mnt;
54 struct virtio_balloon {
55 struct virtio_device *vdev;
56 struct virtqueue *inflate_vq, *deflate_vq, *stats_vq;
58 /* The balloon servicing is delegated to a freezable workqueue. */
59 struct work_struct update_balloon_stats_work;
60 struct work_struct update_balloon_size_work;
62 /* Prevent updating balloon when it is being canceled. */
63 spinlock_t stop_update_lock;
66 /* Waiting for host to ack the pages we released. */
67 wait_queue_head_t acked;
69 /* Number of balloon pages we've told the Host we're not using. */
70 unsigned int num_pages;
72 * The pages we've told the Host we're not using are enqueued
73 * at vb_dev_info->pages list.
74 * Each page on this list adds VIRTIO_BALLOON_PAGES_PER_PAGE
77 struct balloon_dev_info vb_dev_info;
79 /* Synchronize access/update to this struct virtio_balloon elements */
80 struct mutex balloon_lock;
82 /* The array of pfns we tell the Host about. */
83 unsigned int num_pfns;
84 __virtio32 pfns[VIRTIO_BALLOON_ARRAY_PFNS_MAX];
86 /* Memory statistics */
87 struct virtio_balloon_stat stats[VIRTIO_BALLOON_S_NR];
89 /* To register callback in oom notifier call chain */
90 struct notifier_block nb;
93 static struct virtio_device_id id_table[] = {
94 { VIRTIO_ID_BALLOON, VIRTIO_DEV_ANY_ID },
98 static u32 page_to_balloon_pfn(struct page *page)
100 unsigned long pfn = page_to_pfn(page);
102 BUILD_BUG_ON(PAGE_SHIFT < VIRTIO_BALLOON_PFN_SHIFT);
103 /* Convert pfn from Linux page size to balloon page size. */
104 return pfn * VIRTIO_BALLOON_PAGES_PER_PAGE;
107 static void balloon_ack(struct virtqueue *vq)
109 struct virtio_balloon *vb = vq->vdev->priv;
114 static void tell_host(struct virtio_balloon *vb, struct virtqueue *vq)
116 struct scatterlist sg;
119 sg_init_one(&sg, vb->pfns, sizeof(vb->pfns[0]) * vb->num_pfns);
121 /* We should always be able to add one buffer to an empty queue. */
122 virtqueue_add_outbuf(vq, &sg, 1, vb, GFP_KERNEL);
125 /* When host has read buffer, this completes via balloon_ack */
126 wait_event(vb->acked, virtqueue_get_buf(vq, &len));
130 static void set_page_pfns(struct virtio_balloon *vb,
131 __virtio32 pfns[], struct page *page)
135 BUILD_BUG_ON(VIRTIO_BALLOON_PAGES_PER_PAGE > VIRTIO_BALLOON_ARRAY_PFNS_MAX);
138 * Set balloon pfns pointing at this page.
139 * Note that the first pfn points at start of the page.
141 for (i = 0; i < VIRTIO_BALLOON_PAGES_PER_PAGE; i++)
142 pfns[i] = cpu_to_virtio32(vb->vdev,
143 page_to_balloon_pfn(page) + i);
146 static unsigned fill_balloon(struct virtio_balloon *vb, size_t num)
148 unsigned num_allocated_pages;
153 /* We can only do one array worth at a time. */
154 num = min(num, ARRAY_SIZE(vb->pfns));
156 for (num_pfns = 0; num_pfns < num;
157 num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) {
158 struct page *page = balloon_page_alloc();
161 dev_info_ratelimited(&vb->vdev->dev,
162 "Out of puff! Can't get %u pages\n",
163 VIRTIO_BALLOON_PAGES_PER_PAGE);
164 /* Sleep for at least 1/5 of a second before retry. */
169 balloon_page_push(&pages, page);
172 mutex_lock(&vb->balloon_lock);
176 while ((page = balloon_page_pop(&pages))) {
177 balloon_page_enqueue(&vb->vb_dev_info, page);
179 set_page_pfns(vb, vb->pfns + vb->num_pfns, page);
180 vb->num_pages += VIRTIO_BALLOON_PAGES_PER_PAGE;
181 if (!virtio_has_feature(vb->vdev,
182 VIRTIO_BALLOON_F_DEFLATE_ON_OOM))
183 adjust_managed_page_count(page, -1);
184 vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE;
187 num_allocated_pages = vb->num_pfns;
188 /* Did we get any? */
189 if (vb->num_pfns != 0)
190 tell_host(vb, vb->inflate_vq);
191 mutex_unlock(&vb->balloon_lock);
193 return num_allocated_pages;
196 static void release_pages_balloon(struct virtio_balloon *vb,
197 struct list_head *pages)
199 struct page *page, *next;
201 list_for_each_entry_safe(page, next, pages, lru) {
202 if (!virtio_has_feature(vb->vdev,
203 VIRTIO_BALLOON_F_DEFLATE_ON_OOM))
204 adjust_managed_page_count(page, 1);
205 list_del(&page->lru);
206 put_page(page); /* balloon reference */
210 static unsigned leak_balloon(struct virtio_balloon *vb, size_t num)
212 unsigned num_freed_pages;
214 struct balloon_dev_info *vb_dev_info = &vb->vb_dev_info;
217 /* We can only do one array worth at a time. */
218 num = min(num, ARRAY_SIZE(vb->pfns));
220 mutex_lock(&vb->balloon_lock);
221 /* We can't release more pages than taken */
222 num = min(num, (size_t)vb->num_pages);
223 for (vb->num_pfns = 0; vb->num_pfns < num;
224 vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) {
225 page = balloon_page_dequeue(vb_dev_info);
228 set_page_pfns(vb, vb->pfns + vb->num_pfns, page);
229 list_add(&page->lru, &pages);
230 vb->num_pages -= VIRTIO_BALLOON_PAGES_PER_PAGE;
233 num_freed_pages = vb->num_pfns;
236 * virtio_has_feature(vdev, VIRTIO_BALLOON_F_MUST_TELL_HOST);
237 * is true, we *have* to do it in this order
239 if (vb->num_pfns != 0)
240 tell_host(vb, vb->deflate_vq);
241 release_pages_balloon(vb, &pages);
242 mutex_unlock(&vb->balloon_lock);
243 return num_freed_pages;
246 static inline void update_stat(struct virtio_balloon *vb, int idx,
249 BUG_ON(idx >= VIRTIO_BALLOON_S_NR);
250 vb->stats[idx].tag = cpu_to_virtio16(vb->vdev, tag);
251 vb->stats[idx].val = cpu_to_virtio64(vb->vdev, val);
254 #define pages_to_bytes(x) ((u64)(x) << PAGE_SHIFT)
256 static unsigned int update_balloon_stats(struct virtio_balloon *vb)
258 unsigned long events[NR_VM_EVENT_ITEMS];
260 unsigned int idx = 0;
263 all_vm_events(events);
266 available = si_mem_available();
268 #ifdef CONFIG_VM_EVENT_COUNTERS
269 update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_IN,
270 pages_to_bytes(events[PSWPIN]));
271 update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_OUT,
272 pages_to_bytes(events[PSWPOUT]));
273 update_stat(vb, idx++, VIRTIO_BALLOON_S_MAJFLT, events[PGMAJFAULT]);
274 update_stat(vb, idx++, VIRTIO_BALLOON_S_MINFLT, events[PGFAULT]);
276 update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMFREE,
277 pages_to_bytes(i.freeram));
278 update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMTOT,
279 pages_to_bytes(i.totalram));
280 update_stat(vb, idx++, VIRTIO_BALLOON_S_AVAIL,
281 pages_to_bytes(available));
287 * While most virtqueues communicate guest-initiated requests to the hypervisor,
288 * the stats queue operates in reverse. The driver initializes the virtqueue
289 * with a single buffer. From that point forward, all conversations consist of
290 * a hypervisor request (a call to this function) which directs us to refill
291 * the virtqueue with a fresh stats buffer. Since stats collection can sleep,
292 * we delegate the job to a freezable workqueue that will do the actual work via
293 * stats_handle_request().
295 static void stats_request(struct virtqueue *vq)
297 struct virtio_balloon *vb = vq->vdev->priv;
299 spin_lock(&vb->stop_update_lock);
300 if (!vb->stop_update)
301 queue_work(system_freezable_wq, &vb->update_balloon_stats_work);
302 spin_unlock(&vb->stop_update_lock);
305 static void stats_handle_request(struct virtio_balloon *vb)
307 struct virtqueue *vq;
308 struct scatterlist sg;
309 unsigned int len, num_stats;
311 num_stats = update_balloon_stats(vb);
314 if (!virtqueue_get_buf(vq, &len))
316 sg_init_one(&sg, vb->stats, sizeof(vb->stats[0]) * num_stats);
317 virtqueue_add_outbuf(vq, &sg, 1, vb, GFP_KERNEL);
321 static void virtballoon_changed(struct virtio_device *vdev)
323 struct virtio_balloon *vb = vdev->priv;
326 spin_lock_irqsave(&vb->stop_update_lock, flags);
327 if (!vb->stop_update)
328 queue_work(system_freezable_wq, &vb->update_balloon_size_work);
329 spin_unlock_irqrestore(&vb->stop_update_lock, flags);
332 static inline s64 towards_target(struct virtio_balloon *vb)
337 virtio_cread(vb->vdev, struct virtio_balloon_config, num_pages,
340 /* Legacy balloon config space is LE, unlike all other devices. */
341 if (!virtio_has_feature(vb->vdev, VIRTIO_F_VERSION_1))
342 num_pages = le32_to_cpu((__force __le32)num_pages);
345 return target - vb->num_pages;
348 static void update_balloon_size(struct virtio_balloon *vb)
350 u32 actual = vb->num_pages;
352 /* Legacy balloon config space is LE, unlike all other devices. */
353 if (!virtio_has_feature(vb->vdev, VIRTIO_F_VERSION_1))
354 actual = (__force u32)cpu_to_le32(actual);
356 virtio_cwrite(vb->vdev, struct virtio_balloon_config, actual,
361 * virtballoon_oom_notify - release pages when system is under severe
362 * memory pressure (called from out_of_memory())
363 * @self : notifier block struct
365 * @parm : returned - number of freed pages
367 * The balancing of memory by use of the virtio balloon should not cause
368 * the termination of processes while there are pages in the balloon.
369 * If virtio balloon manages to release some memory, it will make the
370 * system return and retry the allocation that forced the OOM killer
373 static int virtballoon_oom_notify(struct notifier_block *self,
374 unsigned long dummy, void *parm)
376 struct virtio_balloon *vb;
377 unsigned long *freed;
378 unsigned num_freed_pages;
380 vb = container_of(self, struct virtio_balloon, nb);
381 if (!virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_DEFLATE_ON_OOM))
385 num_freed_pages = leak_balloon(vb, oom_pages);
386 update_balloon_size(vb);
387 *freed += num_freed_pages;
392 static void update_balloon_stats_func(struct work_struct *work)
394 struct virtio_balloon *vb;
396 vb = container_of(work, struct virtio_balloon,
397 update_balloon_stats_work);
398 stats_handle_request(vb);
401 static void update_balloon_size_func(struct work_struct *work)
403 struct virtio_balloon *vb;
406 vb = container_of(work, struct virtio_balloon,
407 update_balloon_size_work);
408 diff = towards_target(vb);
411 diff -= fill_balloon(vb, diff);
413 diff += leak_balloon(vb, -diff);
414 update_balloon_size(vb);
417 queue_work(system_freezable_wq, work);
420 static int init_vqs(struct virtio_balloon *vb)
422 struct virtqueue *vqs[3];
423 vq_callback_t *callbacks[] = { balloon_ack, balloon_ack, stats_request };
424 static const char * const names[] = { "inflate", "deflate", "stats" };
428 * We expect two virtqueues: inflate and deflate, and
431 nvqs = virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ) ? 3 : 2;
432 err = virtio_find_vqs(vb->vdev, nvqs, vqs, callbacks, names, NULL);
436 vb->inflate_vq = vqs[0];
437 vb->deflate_vq = vqs[1];
438 if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ)) {
439 struct scatterlist sg;
440 unsigned int num_stats;
441 vb->stats_vq = vqs[2];
444 * Prime this virtqueue with one buffer so the hypervisor can
445 * use it to signal us later (it can't be broken yet!).
447 num_stats = update_balloon_stats(vb);
449 sg_init_one(&sg, vb->stats, sizeof(vb->stats[0]) * num_stats);
450 if (virtqueue_add_outbuf(vb->stats_vq, &sg, 1, vb, GFP_KERNEL)
453 virtqueue_kick(vb->stats_vq);
458 #ifdef CONFIG_BALLOON_COMPACTION
460 * virtballoon_migratepage - perform the balloon page migration on behalf of
461 * a compation thread. (called under page lock)
462 * @vb_dev_info: the balloon device
463 * @newpage: page that will replace the isolated page after migration finishes.
464 * @page : the isolated (old) page that is about to be migrated to newpage.
465 * @mode : compaction mode -- not used for balloon page migration.
467 * After a ballooned page gets isolated by compaction procedures, this is the
468 * function that performs the page migration on behalf of a compaction thread
469 * The page migration for virtio balloon is done in a simple swap fashion which
470 * follows these two macro steps:
471 * 1) insert newpage into vb->pages list and update the host about it;
472 * 2) update the host about the old page removed from vb->pages list;
474 * This function preforms the balloon page migration task.
475 * Called through balloon_mapping->a_ops->migratepage
477 static int virtballoon_migratepage(struct balloon_dev_info *vb_dev_info,
478 struct page *newpage, struct page *page, enum migrate_mode mode)
480 struct virtio_balloon *vb = container_of(vb_dev_info,
481 struct virtio_balloon, vb_dev_info);
485 * In order to avoid lock contention while migrating pages concurrently
486 * to leak_balloon() or fill_balloon() we just give up the balloon_lock
487 * this turn, as it is easier to retry the page migration later.
488 * This also prevents fill_balloon() getting stuck into a mutex
489 * recursion in the case it ends up triggering memory compaction
490 * while it is attempting to inflate the ballon.
492 if (!mutex_trylock(&vb->balloon_lock))
495 get_page(newpage); /* balloon reference */
498 * When we migrate a page to a different zone and adjusted the
499 * managed page count when inflating, we have to fixup the count of
500 * both involved zones.
502 if (!virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_DEFLATE_ON_OOM) &&
503 page_zone(page) != page_zone(newpage)) {
504 adjust_managed_page_count(page, 1);
505 adjust_managed_page_count(newpage, -1);
508 /* balloon's page migration 1st step -- inflate "newpage" */
509 spin_lock_irqsave(&vb_dev_info->pages_lock, flags);
510 balloon_page_insert(vb_dev_info, newpage);
511 vb_dev_info->isolated_pages--;
512 __count_vm_event(BALLOON_MIGRATE);
513 spin_unlock_irqrestore(&vb_dev_info->pages_lock, flags);
514 vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE;
515 set_page_pfns(vb, vb->pfns, newpage);
516 tell_host(vb, vb->inflate_vq);
518 /* balloon's page migration 2nd step -- deflate "page" */
519 spin_lock_irqsave(&vb_dev_info->pages_lock, flags);
520 balloon_page_delete(page);
521 spin_unlock_irqrestore(&vb_dev_info->pages_lock, flags);
522 vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE;
523 set_page_pfns(vb, vb->pfns, page);
524 tell_host(vb, vb->deflate_vq);
526 mutex_unlock(&vb->balloon_lock);
528 put_page(page); /* balloon reference */
530 return MIGRATEPAGE_SUCCESS;
533 static struct dentry *balloon_mount(struct file_system_type *fs_type,
534 int flags, const char *dev_name, void *data)
536 static const struct dentry_operations ops = {
537 .d_dname = simple_dname,
540 return mount_pseudo(fs_type, "balloon-kvm:", NULL, &ops,
544 static struct file_system_type balloon_fs = {
545 .name = "balloon-kvm",
546 .mount = balloon_mount,
547 .kill_sb = kill_anon_super,
550 #endif /* CONFIG_BALLOON_COMPACTION */
552 static int virtballoon_probe(struct virtio_device *vdev)
554 struct virtio_balloon *vb;
557 if (!vdev->config->get) {
558 dev_err(&vdev->dev, "%s failure: config access disabled\n",
563 vdev->priv = vb = kmalloc(sizeof(*vb), GFP_KERNEL);
569 INIT_WORK(&vb->update_balloon_stats_work, update_balloon_stats_func);
570 INIT_WORK(&vb->update_balloon_size_work, update_balloon_size_func);
571 spin_lock_init(&vb->stop_update_lock);
572 vb->stop_update = false;
574 mutex_init(&vb->balloon_lock);
575 init_waitqueue_head(&vb->acked);
578 balloon_devinfo_init(&vb->vb_dev_info);
584 vb->nb.notifier_call = virtballoon_oom_notify;
585 vb->nb.priority = VIRTBALLOON_OOM_NOTIFY_PRIORITY;
586 err = register_oom_notifier(&vb->nb);
590 #ifdef CONFIG_BALLOON_COMPACTION
591 balloon_mnt = kern_mount(&balloon_fs);
592 if (IS_ERR(balloon_mnt)) {
593 err = PTR_ERR(balloon_mnt);
594 unregister_oom_notifier(&vb->nb);
598 vb->vb_dev_info.migratepage = virtballoon_migratepage;
599 vb->vb_dev_info.inode = alloc_anon_inode(balloon_mnt->mnt_sb);
600 if (IS_ERR(vb->vb_dev_info.inode)) {
601 err = PTR_ERR(vb->vb_dev_info.inode);
602 kern_unmount(balloon_mnt);
603 unregister_oom_notifier(&vb->nb);
604 vb->vb_dev_info.inode = NULL;
607 vb->vb_dev_info.inode->i_mapping->a_ops = &balloon_aops;
610 virtio_device_ready(vdev);
612 if (towards_target(vb))
613 virtballoon_changed(vdev);
617 vdev->config->del_vqs(vdev);
624 static void remove_common(struct virtio_balloon *vb)
626 /* There might be pages left in the balloon: free them. */
627 while (vb->num_pages)
628 leak_balloon(vb, vb->num_pages);
629 update_balloon_size(vb);
631 /* Now we reset the device so we can clean up the queues. */
632 vb->vdev->config->reset(vb->vdev);
634 vb->vdev->config->del_vqs(vb->vdev);
637 static void virtballoon_remove(struct virtio_device *vdev)
639 struct virtio_balloon *vb = vdev->priv;
641 unregister_oom_notifier(&vb->nb);
643 spin_lock_irq(&vb->stop_update_lock);
644 vb->stop_update = true;
645 spin_unlock_irq(&vb->stop_update_lock);
646 cancel_work_sync(&vb->update_balloon_size_work);
647 cancel_work_sync(&vb->update_balloon_stats_work);
650 #ifdef CONFIG_BALLOON_COMPACTION
651 if (vb->vb_dev_info.inode)
652 iput(vb->vb_dev_info.inode);
654 kern_unmount(balloon_mnt);
659 #ifdef CONFIG_PM_SLEEP
660 static int virtballoon_freeze(struct virtio_device *vdev)
662 struct virtio_balloon *vb = vdev->priv;
665 * The workqueue is already frozen by the PM core before this
666 * function is called.
672 static int virtballoon_restore(struct virtio_device *vdev)
674 struct virtio_balloon *vb = vdev->priv;
677 ret = init_vqs(vdev->priv);
681 virtio_device_ready(vdev);
683 if (towards_target(vb))
684 virtballoon_changed(vdev);
685 update_balloon_size(vb);
690 static int virtballoon_validate(struct virtio_device *vdev)
692 __virtio_clear_bit(vdev, VIRTIO_F_IOMMU_PLATFORM);
696 static unsigned int features[] = {
697 VIRTIO_BALLOON_F_MUST_TELL_HOST,
698 VIRTIO_BALLOON_F_STATS_VQ,
699 VIRTIO_BALLOON_F_DEFLATE_ON_OOM,
702 static struct virtio_driver virtio_balloon_driver = {
703 .feature_table = features,
704 .feature_table_size = ARRAY_SIZE(features),
705 .driver.name = KBUILD_MODNAME,
706 .driver.owner = THIS_MODULE,
707 .id_table = id_table,
708 .validate = virtballoon_validate,
709 .probe = virtballoon_probe,
710 .remove = virtballoon_remove,
711 .config_changed = virtballoon_changed,
712 #ifdef CONFIG_PM_SLEEP
713 .freeze = virtballoon_freeze,
714 .restore = virtballoon_restore,
718 module_virtio_driver(virtio_balloon_driver);
719 MODULE_DEVICE_TABLE(virtio, id_table);
720 MODULE_DESCRIPTION("Virtio balloon driver");
721 MODULE_LICENSE("GPL");