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 struct page *balloon_pfn_to_page(u32 pfn)
109 BUG_ON(pfn % VIRTIO_BALLOON_PAGES_PER_PAGE);
110 return pfn_to_page(pfn / VIRTIO_BALLOON_PAGES_PER_PAGE);
113 static void balloon_ack(struct virtqueue *vq)
115 struct virtio_balloon *vb = vq->vdev->priv;
120 static void tell_host(struct virtio_balloon *vb, struct virtqueue *vq)
122 struct scatterlist sg;
125 sg_init_one(&sg, vb->pfns, sizeof(vb->pfns[0]) * vb->num_pfns);
127 /* We should always be able to add one buffer to an empty queue. */
128 virtqueue_add_outbuf(vq, &sg, 1, vb, GFP_KERNEL);
131 /* When host has read buffer, this completes via balloon_ack */
132 wait_event(vb->acked, virtqueue_get_buf(vq, &len));
136 static void set_page_pfns(struct virtio_balloon *vb,
137 __virtio32 pfns[], struct page *page)
141 /* Set balloon pfns pointing at this page.
142 * Note that the first pfn points at start of the page. */
143 for (i = 0; i < VIRTIO_BALLOON_PAGES_PER_PAGE; i++)
144 pfns[i] = cpu_to_virtio32(vb->vdev,
145 page_to_balloon_pfn(page) + i);
148 static unsigned fill_balloon(struct virtio_balloon *vb, size_t num)
150 struct balloon_dev_info *vb_dev_info = &vb->vb_dev_info;
151 unsigned num_allocated_pages;
153 /* We can only do one array worth at a time. */
154 num = min(num, ARRAY_SIZE(vb->pfns));
156 mutex_lock(&vb->balloon_lock);
157 for (vb->num_pfns = 0; vb->num_pfns < num;
158 vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) {
159 struct page *page = balloon_page_enqueue(vb_dev_info);
162 dev_info_ratelimited(&vb->vdev->dev,
163 "Out of puff! Can't get %u pages\n",
164 VIRTIO_BALLOON_PAGES_PER_PAGE);
165 /* Sleep for at least 1/5 of a second before retry. */
169 set_page_pfns(vb, vb->pfns + vb->num_pfns, page);
170 vb->num_pages += VIRTIO_BALLOON_PAGES_PER_PAGE;
171 if (!virtio_has_feature(vb->vdev,
172 VIRTIO_BALLOON_F_DEFLATE_ON_OOM))
173 adjust_managed_page_count(page, -1);
176 num_allocated_pages = vb->num_pfns;
177 /* Did we get any? */
178 if (vb->num_pfns != 0)
179 tell_host(vb, vb->inflate_vq);
180 mutex_unlock(&vb->balloon_lock);
182 return num_allocated_pages;
185 static void release_pages_balloon(struct virtio_balloon *vb)
190 /* Find pfns pointing at start of each page, get pages and free them. */
191 for (i = 0; i < vb->num_pfns; i += VIRTIO_BALLOON_PAGES_PER_PAGE) {
192 page = balloon_pfn_to_page(virtio32_to_cpu(vb->vdev,
194 if (!virtio_has_feature(vb->vdev,
195 VIRTIO_BALLOON_F_DEFLATE_ON_OOM))
196 adjust_managed_page_count(page, 1);
197 put_page(page); /* balloon reference */
201 static unsigned leak_balloon(struct virtio_balloon *vb, size_t num)
203 unsigned num_freed_pages;
205 struct balloon_dev_info *vb_dev_info = &vb->vb_dev_info;
207 /* We can only do one array worth at a time. */
208 num = min(num, ARRAY_SIZE(vb->pfns));
210 mutex_lock(&vb->balloon_lock);
211 /* We can't release more pages than taken */
212 num = min(num, (size_t)vb->num_pages);
213 for (vb->num_pfns = 0; vb->num_pfns < num;
214 vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) {
215 page = balloon_page_dequeue(vb_dev_info);
218 set_page_pfns(vb, vb->pfns + vb->num_pfns, page);
219 vb->num_pages -= VIRTIO_BALLOON_PAGES_PER_PAGE;
222 num_freed_pages = vb->num_pfns;
225 * virtio_has_feature(vdev, VIRTIO_BALLOON_F_MUST_TELL_HOST);
226 * is true, we *have* to do it in this order
228 if (vb->num_pfns != 0)
229 tell_host(vb, vb->deflate_vq);
230 release_pages_balloon(vb);
231 mutex_unlock(&vb->balloon_lock);
232 return num_freed_pages;
235 static inline void update_stat(struct virtio_balloon *vb, int idx,
238 BUG_ON(idx >= VIRTIO_BALLOON_S_NR);
239 vb->stats[idx].tag = cpu_to_virtio16(vb->vdev, tag);
240 vb->stats[idx].val = cpu_to_virtio64(vb->vdev, val);
243 #define pages_to_bytes(x) ((u64)(x) << PAGE_SHIFT)
245 static unsigned int update_balloon_stats(struct virtio_balloon *vb)
247 unsigned long events[NR_VM_EVENT_ITEMS];
249 unsigned int idx = 0;
252 all_vm_events(events);
255 available = si_mem_available();
257 update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_IN,
258 pages_to_bytes(events[PSWPIN]));
259 update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_OUT,
260 pages_to_bytes(events[PSWPOUT]));
261 update_stat(vb, idx++, VIRTIO_BALLOON_S_MAJFLT, events[PGMAJFAULT]);
262 update_stat(vb, idx++, VIRTIO_BALLOON_S_MINFLT, events[PGFAULT]);
263 update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMFREE,
264 pages_to_bytes(i.freeram));
265 update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMTOT,
266 pages_to_bytes(i.totalram));
267 update_stat(vb, idx++, VIRTIO_BALLOON_S_AVAIL,
268 pages_to_bytes(available));
274 * While most virtqueues communicate guest-initiated requests to the hypervisor,
275 * the stats queue operates in reverse. The driver initializes the virtqueue
276 * with a single buffer. From that point forward, all conversations consist of
277 * a hypervisor request (a call to this function) which directs us to refill
278 * the virtqueue with a fresh stats buffer. Since stats collection can sleep,
279 * we delegate the job to a freezable workqueue that will do the actual work via
280 * stats_handle_request().
282 static void stats_request(struct virtqueue *vq)
284 struct virtio_balloon *vb = vq->vdev->priv;
286 spin_lock(&vb->stop_update_lock);
287 if (!vb->stop_update)
288 queue_work(system_freezable_wq, &vb->update_balloon_stats_work);
289 spin_unlock(&vb->stop_update_lock);
292 static void stats_handle_request(struct virtio_balloon *vb)
294 struct virtqueue *vq;
295 struct scatterlist sg;
296 unsigned int len, num_stats;
298 num_stats = update_balloon_stats(vb);
301 if (!virtqueue_get_buf(vq, &len))
303 sg_init_one(&sg, vb->stats, sizeof(vb->stats[0]) * num_stats);
304 virtqueue_add_outbuf(vq, &sg, 1, vb, GFP_KERNEL);
308 static void virtballoon_changed(struct virtio_device *vdev)
310 struct virtio_balloon *vb = vdev->priv;
313 spin_lock_irqsave(&vb->stop_update_lock, flags);
314 if (!vb->stop_update)
315 queue_work(system_freezable_wq, &vb->update_balloon_size_work);
316 spin_unlock_irqrestore(&vb->stop_update_lock, flags);
319 static inline s64 towards_target(struct virtio_balloon *vb)
324 virtio_cread(vb->vdev, struct virtio_balloon_config, num_pages,
327 /* Legacy balloon config space is LE, unlike all other devices. */
328 if (!virtio_has_feature(vb->vdev, VIRTIO_F_VERSION_1))
329 num_pages = le32_to_cpu((__force __le32)num_pages);
332 return target - vb->num_pages;
335 static void update_balloon_size(struct virtio_balloon *vb)
337 u32 actual = vb->num_pages;
339 /* Legacy balloon config space is LE, unlike all other devices. */
340 if (!virtio_has_feature(vb->vdev, VIRTIO_F_VERSION_1))
341 actual = (__force u32)cpu_to_le32(actual);
343 virtio_cwrite(vb->vdev, struct virtio_balloon_config, actual,
348 * virtballoon_oom_notify - release pages when system is under severe
349 * memory pressure (called from out_of_memory())
350 * @self : notifier block struct
352 * @parm : returned - number of freed pages
354 * The balancing of memory by use of the virtio balloon should not cause
355 * the termination of processes while there are pages in the balloon.
356 * If virtio balloon manages to release some memory, it will make the
357 * system return and retry the allocation that forced the OOM killer
360 static int virtballoon_oom_notify(struct notifier_block *self,
361 unsigned long dummy, void *parm)
363 struct virtio_balloon *vb;
364 unsigned long *freed;
365 unsigned num_freed_pages;
367 vb = container_of(self, struct virtio_balloon, nb);
368 if (!virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_DEFLATE_ON_OOM))
372 num_freed_pages = leak_balloon(vb, oom_pages);
373 update_balloon_size(vb);
374 *freed += num_freed_pages;
379 static void update_balloon_stats_func(struct work_struct *work)
381 struct virtio_balloon *vb;
383 vb = container_of(work, struct virtio_balloon,
384 update_balloon_stats_work);
385 stats_handle_request(vb);
388 static void update_balloon_size_func(struct work_struct *work)
390 struct virtio_balloon *vb;
393 vb = container_of(work, struct virtio_balloon,
394 update_balloon_size_work);
395 diff = towards_target(vb);
398 diff -= fill_balloon(vb, diff);
400 diff += leak_balloon(vb, -diff);
401 update_balloon_size(vb);
404 queue_work(system_freezable_wq, work);
407 static int init_vqs(struct virtio_balloon *vb)
409 struct virtqueue *vqs[3];
410 vq_callback_t *callbacks[] = { balloon_ack, balloon_ack, stats_request };
411 static const char * const names[] = { "inflate", "deflate", "stats" };
415 * We expect two virtqueues: inflate and deflate, and
418 nvqs = virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ) ? 3 : 2;
419 err = vb->vdev->config->find_vqs(vb->vdev, nvqs, vqs, callbacks, names,
424 vb->inflate_vq = vqs[0];
425 vb->deflate_vq = vqs[1];
426 if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ)) {
427 struct scatterlist sg;
428 unsigned int num_stats;
429 vb->stats_vq = vqs[2];
432 * Prime this virtqueue with one buffer so the hypervisor can
433 * use it to signal us later (it can't be broken yet!).
435 num_stats = update_balloon_stats(vb);
437 sg_init_one(&sg, vb->stats, sizeof(vb->stats[0]) * num_stats);
438 if (virtqueue_add_outbuf(vb->stats_vq, &sg, 1, vb, GFP_KERNEL)
441 virtqueue_kick(vb->stats_vq);
446 #ifdef CONFIG_BALLOON_COMPACTION
448 * virtballoon_migratepage - perform the balloon page migration on behalf of
449 * a compation thread. (called under page lock)
450 * @vb_dev_info: the balloon device
451 * @newpage: page that will replace the isolated page after migration finishes.
452 * @page : the isolated (old) page that is about to be migrated to newpage.
453 * @mode : compaction mode -- not used for balloon page migration.
455 * After a ballooned page gets isolated by compaction procedures, this is the
456 * function that performs the page migration on behalf of a compaction thread
457 * The page migration for virtio balloon is done in a simple swap fashion which
458 * follows these two macro steps:
459 * 1) insert newpage into vb->pages list and update the host about it;
460 * 2) update the host about the old page removed from vb->pages list;
462 * This function preforms the balloon page migration task.
463 * Called through balloon_mapping->a_ops->migratepage
465 static int virtballoon_migratepage(struct balloon_dev_info *vb_dev_info,
466 struct page *newpage, struct page *page, enum migrate_mode mode)
468 struct virtio_balloon *vb = container_of(vb_dev_info,
469 struct virtio_balloon, vb_dev_info);
473 * In order to avoid lock contention while migrating pages concurrently
474 * to leak_balloon() or fill_balloon() we just give up the balloon_lock
475 * this turn, as it is easier to retry the page migration later.
476 * This also prevents fill_balloon() getting stuck into a mutex
477 * recursion in the case it ends up triggering memory compaction
478 * while it is attempting to inflate the ballon.
480 if (!mutex_trylock(&vb->balloon_lock))
483 get_page(newpage); /* balloon reference */
485 /* balloon's page migration 1st step -- inflate "newpage" */
486 spin_lock_irqsave(&vb_dev_info->pages_lock, flags);
487 balloon_page_insert(vb_dev_info, newpage);
488 vb_dev_info->isolated_pages--;
489 __count_vm_event(BALLOON_MIGRATE);
490 spin_unlock_irqrestore(&vb_dev_info->pages_lock, flags);
491 vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE;
492 set_page_pfns(vb, vb->pfns, newpage);
493 tell_host(vb, vb->inflate_vq);
495 /* balloon's page migration 2nd step -- deflate "page" */
496 balloon_page_delete(page);
497 vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE;
498 set_page_pfns(vb, vb->pfns, page);
499 tell_host(vb, vb->deflate_vq);
501 mutex_unlock(&vb->balloon_lock);
503 put_page(page); /* balloon reference */
505 return MIGRATEPAGE_SUCCESS;
508 static struct dentry *balloon_mount(struct file_system_type *fs_type,
509 int flags, const char *dev_name, void *data)
511 static const struct dentry_operations ops = {
512 .d_dname = simple_dname,
515 return mount_pseudo(fs_type, "balloon-kvm:", NULL, &ops,
519 static struct file_system_type balloon_fs = {
520 .name = "balloon-kvm",
521 .mount = balloon_mount,
522 .kill_sb = kill_anon_super,
525 #endif /* CONFIG_BALLOON_COMPACTION */
527 static int virtballoon_probe(struct virtio_device *vdev)
529 struct virtio_balloon *vb;
532 if (!vdev->config->get) {
533 dev_err(&vdev->dev, "%s failure: config access disabled\n",
538 vdev->priv = vb = kmalloc(sizeof(*vb), GFP_KERNEL);
544 INIT_WORK(&vb->update_balloon_stats_work, update_balloon_stats_func);
545 INIT_WORK(&vb->update_balloon_size_work, update_balloon_size_func);
546 spin_lock_init(&vb->stop_update_lock);
547 vb->stop_update = false;
549 mutex_init(&vb->balloon_lock);
550 init_waitqueue_head(&vb->acked);
553 balloon_devinfo_init(&vb->vb_dev_info);
559 vb->nb.notifier_call = virtballoon_oom_notify;
560 vb->nb.priority = VIRTBALLOON_OOM_NOTIFY_PRIORITY;
561 err = register_oom_notifier(&vb->nb);
565 #ifdef CONFIG_BALLOON_COMPACTION
566 balloon_mnt = kern_mount(&balloon_fs);
567 if (IS_ERR(balloon_mnt)) {
568 err = PTR_ERR(balloon_mnt);
569 unregister_oom_notifier(&vb->nb);
573 vb->vb_dev_info.migratepage = virtballoon_migratepage;
574 vb->vb_dev_info.inode = alloc_anon_inode(balloon_mnt->mnt_sb);
575 if (IS_ERR(vb->vb_dev_info.inode)) {
576 err = PTR_ERR(vb->vb_dev_info.inode);
577 kern_unmount(balloon_mnt);
578 unregister_oom_notifier(&vb->nb);
579 vb->vb_dev_info.inode = NULL;
582 vb->vb_dev_info.inode->i_mapping->a_ops = &balloon_aops;
585 virtio_device_ready(vdev);
587 if (towards_target(vb))
588 virtballoon_changed(vdev);
592 vdev->config->del_vqs(vdev);
599 static void remove_common(struct virtio_balloon *vb)
601 /* There might be pages left in the balloon: free them. */
602 while (vb->num_pages)
603 leak_balloon(vb, vb->num_pages);
604 update_balloon_size(vb);
606 /* Now we reset the device so we can clean up the queues. */
607 vb->vdev->config->reset(vb->vdev);
609 vb->vdev->config->del_vqs(vb->vdev);
612 static void virtballoon_remove(struct virtio_device *vdev)
614 struct virtio_balloon *vb = vdev->priv;
616 unregister_oom_notifier(&vb->nb);
618 spin_lock_irq(&vb->stop_update_lock);
619 vb->stop_update = true;
620 spin_unlock_irq(&vb->stop_update_lock);
621 cancel_work_sync(&vb->update_balloon_size_work);
622 cancel_work_sync(&vb->update_balloon_stats_work);
625 #ifdef CONFIG_BALLOON_COMPACTION
626 if (vb->vb_dev_info.inode)
627 iput(vb->vb_dev_info.inode);
629 kern_unmount(balloon_mnt);
634 #ifdef CONFIG_PM_SLEEP
635 static int virtballoon_freeze(struct virtio_device *vdev)
637 struct virtio_balloon *vb = vdev->priv;
640 * The workqueue is already frozen by the PM core before this
641 * function is called.
647 static int virtballoon_restore(struct virtio_device *vdev)
649 struct virtio_balloon *vb = vdev->priv;
652 ret = init_vqs(vdev->priv);
656 virtio_device_ready(vdev);
658 if (towards_target(vb))
659 virtballoon_changed(vdev);
660 update_balloon_size(vb);
665 static unsigned int features[] = {
666 VIRTIO_BALLOON_F_MUST_TELL_HOST,
667 VIRTIO_BALLOON_F_STATS_VQ,
668 VIRTIO_BALLOON_F_DEFLATE_ON_OOM,
671 static struct virtio_driver virtio_balloon_driver = {
672 .feature_table = features,
673 .feature_table_size = ARRAY_SIZE(features),
674 .driver.name = KBUILD_MODNAME,
675 .driver.owner = THIS_MODULE,
676 .id_table = id_table,
677 .probe = virtballoon_probe,
678 .remove = virtballoon_remove,
679 .config_changed = virtballoon_changed,
680 #ifdef CONFIG_PM_SLEEP
681 .freeze = virtballoon_freeze,
682 .restore = virtballoon_restore,
686 module_virtio_driver(virtio_balloon_driver);
687 MODULE_DEVICE_TABLE(virtio, id_table);
688 MODULE_DESCRIPTION("Virtio balloon driver");
689 MODULE_LICENSE("GPL");