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>
34 * Balloon device works in 4K page units. So each page is pointed to by
35 * multiple balloon pages. All memory counters in this driver are in balloon
38 #define VIRTIO_BALLOON_PAGES_PER_PAGE (unsigned)(PAGE_SIZE >> VIRTIO_BALLOON_PFN_SHIFT)
39 #define VIRTIO_BALLOON_ARRAY_PFNS_MAX 256
40 #define OOM_VBALLOON_DEFAULT_PAGES 256
41 #define VIRTBALLOON_OOM_NOTIFY_PRIORITY 80
43 static int oom_pages = OOM_VBALLOON_DEFAULT_PAGES;
44 module_param(oom_pages, int, S_IRUSR | S_IWUSR);
45 MODULE_PARM_DESC(oom_pages, "pages to free on OOM");
47 struct virtio_balloon {
48 struct virtio_device *vdev;
49 struct virtqueue *inflate_vq, *deflate_vq, *stats_vq;
51 /* Where the ballooning thread waits for config to change. */
52 wait_queue_head_t config_change;
54 /* The thread servicing the balloon. */
55 struct task_struct *thread;
57 /* Waiting for host to ack the pages we released. */
58 wait_queue_head_t acked;
60 /* Number of balloon pages we've told the Host we're not using. */
61 unsigned int num_pages;
63 * The pages we've told the Host we're not using are enqueued
64 * at vb_dev_info->pages list.
65 * Each page on this list adds VIRTIO_BALLOON_PAGES_PER_PAGE
68 struct balloon_dev_info vb_dev_info;
70 /* Synchronize access/update to this struct virtio_balloon elements */
71 struct mutex balloon_lock;
73 /* The array of pfns we tell the Host about. */
74 unsigned int num_pfns;
75 u32 pfns[VIRTIO_BALLOON_ARRAY_PFNS_MAX];
77 /* Memory statistics */
78 int need_stats_update;
79 struct virtio_balloon_stat stats[VIRTIO_BALLOON_S_NR];
81 /* To register callback in oom notifier call chain */
82 struct notifier_block nb;
85 static struct virtio_device_id id_table[] = {
86 { VIRTIO_ID_BALLOON, VIRTIO_DEV_ANY_ID },
90 static u32 page_to_balloon_pfn(struct page *page)
92 unsigned long pfn = page_to_pfn(page);
94 BUILD_BUG_ON(PAGE_SHIFT < VIRTIO_BALLOON_PFN_SHIFT);
95 /* Convert pfn from Linux page size to balloon page size. */
96 return pfn * VIRTIO_BALLOON_PAGES_PER_PAGE;
99 static struct page *balloon_pfn_to_page(u32 pfn)
101 BUG_ON(pfn % VIRTIO_BALLOON_PAGES_PER_PAGE);
102 return pfn_to_page(pfn / VIRTIO_BALLOON_PAGES_PER_PAGE);
105 static void balloon_ack(struct virtqueue *vq)
107 struct virtio_balloon *vb = vq->vdev->priv;
112 static void tell_host(struct virtio_balloon *vb, struct virtqueue *vq)
114 struct scatterlist sg;
117 sg_init_one(&sg, vb->pfns, sizeof(vb->pfns[0]) * vb->num_pfns);
119 /* We should always be able to add one buffer to an empty queue. */
120 virtqueue_add_outbuf(vq, &sg, 1, vb, GFP_KERNEL);
123 /* When host has read buffer, this completes via balloon_ack */
124 wait_event(vb->acked, virtqueue_get_buf(vq, &len));
127 static void set_page_pfns(u32 pfns[], struct page *page)
131 /* Set balloon pfns pointing at this page.
132 * Note that the first pfn points at start of the page. */
133 for (i = 0; i < VIRTIO_BALLOON_PAGES_PER_PAGE; i++)
134 pfns[i] = page_to_balloon_pfn(page) + i;
137 static void fill_balloon(struct virtio_balloon *vb, size_t num)
139 struct balloon_dev_info *vb_dev_info = &vb->vb_dev_info;
141 /* We can only do one array worth at a time. */
142 num = min(num, ARRAY_SIZE(vb->pfns));
144 mutex_lock(&vb->balloon_lock);
145 for (vb->num_pfns = 0; vb->num_pfns < num;
146 vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) {
147 struct page *page = balloon_page_enqueue(vb_dev_info);
150 dev_info_ratelimited(&vb->vdev->dev,
151 "Out of puff! Can't get %u pages\n",
152 VIRTIO_BALLOON_PAGES_PER_PAGE);
153 /* Sleep for at least 1/5 of a second before retry. */
157 set_page_pfns(vb->pfns + vb->num_pfns, page);
158 vb->num_pages += VIRTIO_BALLOON_PAGES_PER_PAGE;
159 adjust_managed_page_count(page, -1);
162 /* Did we get any? */
163 if (vb->num_pfns != 0)
164 tell_host(vb, vb->inflate_vq);
165 mutex_unlock(&vb->balloon_lock);
168 static void release_pages_by_pfn(const u32 pfns[], unsigned int num)
172 /* Find pfns pointing at start of each page, get pages and free them. */
173 for (i = 0; i < num; i += VIRTIO_BALLOON_PAGES_PER_PAGE) {
174 struct page *page = balloon_pfn_to_page(pfns[i]);
175 adjust_managed_page_count(page, 1);
176 put_page(page); /* balloon reference */
180 static unsigned leak_balloon(struct virtio_balloon *vb, size_t num)
182 unsigned num_freed_pages;
184 struct balloon_dev_info *vb_dev_info = &vb->vb_dev_info;
186 /* We can only do one array worth at a time. */
187 num = min(num, ARRAY_SIZE(vb->pfns));
189 mutex_lock(&vb->balloon_lock);
190 for (vb->num_pfns = 0; vb->num_pfns < num;
191 vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) {
192 page = balloon_page_dequeue(vb_dev_info);
195 set_page_pfns(vb->pfns + vb->num_pfns, page);
196 vb->num_pages -= VIRTIO_BALLOON_PAGES_PER_PAGE;
199 num_freed_pages = vb->num_pfns;
202 * virtio_has_feature(vdev, VIRTIO_BALLOON_F_MUST_TELL_HOST);
203 * is true, we *have* to do it in this order
205 if (vb->num_pfns != 0)
206 tell_host(vb, vb->deflate_vq);
207 mutex_unlock(&vb->balloon_lock);
208 release_pages_by_pfn(vb->pfns, vb->num_pfns);
209 return num_freed_pages;
212 static inline void update_stat(struct virtio_balloon *vb, int idx,
215 BUG_ON(idx >= VIRTIO_BALLOON_S_NR);
216 vb->stats[idx].tag = tag;
217 vb->stats[idx].val = val;
220 #define pages_to_bytes(x) ((u64)(x) << PAGE_SHIFT)
222 static void update_balloon_stats(struct virtio_balloon *vb)
224 unsigned long events[NR_VM_EVENT_ITEMS];
228 all_vm_events(events);
231 update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_IN,
232 pages_to_bytes(events[PSWPIN]));
233 update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_OUT,
234 pages_to_bytes(events[PSWPOUT]));
235 update_stat(vb, idx++, VIRTIO_BALLOON_S_MAJFLT, events[PGMAJFAULT]);
236 update_stat(vb, idx++, VIRTIO_BALLOON_S_MINFLT, events[PGFAULT]);
237 update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMFREE,
238 pages_to_bytes(i.freeram));
239 update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMTOT,
240 pages_to_bytes(i.totalram));
244 * While most virtqueues communicate guest-initiated requests to the hypervisor,
245 * the stats queue operates in reverse. The driver initializes the virtqueue
246 * with a single buffer. From that point forward, all conversations consist of
247 * a hypervisor request (a call to this function) which directs us to refill
248 * the virtqueue with a fresh stats buffer. Since stats collection can sleep,
249 * we notify our kthread which does the actual work via stats_handle_request().
251 static void stats_request(struct virtqueue *vq)
253 struct virtio_balloon *vb = vq->vdev->priv;
255 vb->need_stats_update = 1;
256 wake_up(&vb->config_change);
259 static void stats_handle_request(struct virtio_balloon *vb)
261 struct virtqueue *vq;
262 struct scatterlist sg;
265 vb->need_stats_update = 0;
266 update_balloon_stats(vb);
269 if (!virtqueue_get_buf(vq, &len))
271 sg_init_one(&sg, vb->stats, sizeof(vb->stats));
272 virtqueue_add_outbuf(vq, &sg, 1, vb, GFP_KERNEL);
276 static void virtballoon_changed(struct virtio_device *vdev)
278 struct virtio_balloon *vb = vdev->priv;
280 wake_up(&vb->config_change);
283 static inline s64 towards_target(struct virtio_balloon *vb)
288 virtio_cread(vb->vdev, struct virtio_balloon_config, num_pages, &v);
290 target = le32_to_cpu(v);
291 return target - vb->num_pages;
294 static void update_balloon_size(struct virtio_balloon *vb)
296 __le32 actual = cpu_to_le32(vb->num_pages);
298 virtio_cwrite(vb->vdev, struct virtio_balloon_config, actual,
303 * virtballoon_oom_notify - release pages when system is under severe
304 * memory pressure (called from out_of_memory())
305 * @self : notifier block struct
307 * @parm : returned - number of freed pages
309 * The balancing of memory by use of the virtio balloon should not cause
310 * the termination of processes while there are pages in the balloon.
311 * If virtio balloon manages to release some memory, it will make the
312 * system return and retry the allocation that forced the OOM killer
315 static int virtballoon_oom_notify(struct notifier_block *self,
316 unsigned long dummy, void *parm)
318 struct virtio_balloon *vb;
319 unsigned long *freed;
320 unsigned num_freed_pages;
322 vb = container_of(self, struct virtio_balloon, nb);
323 if (!virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_DEFLATE_ON_OOM))
327 num_freed_pages = leak_balloon(vb, oom_pages);
328 update_balloon_size(vb);
329 *freed += num_freed_pages;
334 static int balloon(void *_vballoon)
336 struct virtio_balloon *vb = _vballoon;
339 while (!kthread_should_stop()) {
343 wait_event_interruptible(vb->config_change,
344 (diff = towards_target(vb)) != 0
345 || vb->need_stats_update
346 || kthread_should_stop()
347 || freezing(current));
348 if (vb->need_stats_update)
349 stats_handle_request(vb);
351 fill_balloon(vb, diff);
353 leak_balloon(vb, -diff);
354 update_balloon_size(vb);
357 * For large balloon changes, we could spend a lot of time
358 * and always have work to do. Be nice if preempt disabled.
365 static int init_vqs(struct virtio_balloon *vb)
367 struct virtqueue *vqs[3];
368 vq_callback_t *callbacks[] = { balloon_ack, balloon_ack, stats_request };
369 const char *names[] = { "inflate", "deflate", "stats" };
373 * We expect two virtqueues: inflate and deflate, and
376 nvqs = virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ) ? 3 : 2;
377 err = vb->vdev->config->find_vqs(vb->vdev, nvqs, vqs, callbacks, names);
381 vb->inflate_vq = vqs[0];
382 vb->deflate_vq = vqs[1];
383 if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ)) {
384 struct scatterlist sg;
385 vb->stats_vq = vqs[2];
388 * Prime this virtqueue with one buffer so the hypervisor can
389 * use it to signal us later (it can't be broken yet!).
391 sg_init_one(&sg, vb->stats, sizeof vb->stats);
392 if (virtqueue_add_outbuf(vb->stats_vq, &sg, 1, vb, GFP_KERNEL)
395 virtqueue_kick(vb->stats_vq);
400 #ifdef CONFIG_BALLOON_COMPACTION
402 * virtballoon_migratepage - perform the balloon page migration on behalf of
403 * a compation thread. (called under page lock)
404 * @vb_dev_info: the balloon device
405 * @newpage: page that will replace the isolated page after migration finishes.
406 * @page : the isolated (old) page that is about to be migrated to newpage.
407 * @mode : compaction mode -- not used for balloon page migration.
409 * After a ballooned page gets isolated by compaction procedures, this is the
410 * function that performs the page migration on behalf of a compaction thread
411 * The page migration for virtio balloon is done in a simple swap fashion which
412 * follows these two macro steps:
413 * 1) insert newpage into vb->pages list and update the host about it;
414 * 2) update the host about the old page removed from vb->pages list;
416 * This function preforms the balloon page migration task.
417 * Called through balloon_mapping->a_ops->migratepage
419 static int virtballoon_migratepage(struct balloon_dev_info *vb_dev_info,
420 struct page *newpage, struct page *page, enum migrate_mode mode)
422 struct virtio_balloon *vb = container_of(vb_dev_info,
423 struct virtio_balloon, vb_dev_info);
427 * In order to avoid lock contention while migrating pages concurrently
428 * to leak_balloon() or fill_balloon() we just give up the balloon_lock
429 * this turn, as it is easier to retry the page migration later.
430 * This also prevents fill_balloon() getting stuck into a mutex
431 * recursion in the case it ends up triggering memory compaction
432 * while it is attempting to inflate the ballon.
434 if (!mutex_trylock(&vb->balloon_lock))
437 get_page(newpage); /* balloon reference */
439 /* balloon's page migration 1st step -- inflate "newpage" */
440 spin_lock_irqsave(&vb_dev_info->pages_lock, flags);
441 balloon_page_insert(vb_dev_info, newpage);
442 vb_dev_info->isolated_pages--;
443 __count_vm_event(BALLOON_MIGRATE);
444 spin_unlock_irqrestore(&vb_dev_info->pages_lock, flags);
445 vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE;
446 set_page_pfns(vb->pfns, newpage);
447 tell_host(vb, vb->inflate_vq);
449 /* balloon's page migration 2nd step -- deflate "page" */
450 balloon_page_delete(page);
451 vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE;
452 set_page_pfns(vb->pfns, page);
453 tell_host(vb, vb->deflate_vq);
455 mutex_unlock(&vb->balloon_lock);
457 put_page(page); /* balloon reference */
459 return MIGRATEPAGE_SUCCESS;
461 #endif /* CONFIG_BALLOON_COMPACTION */
463 static int virtballoon_probe(struct virtio_device *vdev)
465 struct virtio_balloon *vb;
468 if (!vdev->config->get) {
469 dev_err(&vdev->dev, "%s failure: config access disabled\n",
474 vdev->priv = vb = kmalloc(sizeof(*vb), GFP_KERNEL);
481 mutex_init(&vb->balloon_lock);
482 init_waitqueue_head(&vb->config_change);
483 init_waitqueue_head(&vb->acked);
485 vb->need_stats_update = 0;
487 balloon_devinfo_init(&vb->vb_dev_info);
488 #ifdef CONFIG_BALLOON_COMPACTION
489 vb->vb_dev_info.migratepage = virtballoon_migratepage;
496 vb->nb.notifier_call = virtballoon_oom_notify;
497 vb->nb.priority = VIRTBALLOON_OOM_NOTIFY_PRIORITY;
498 err = register_oom_notifier(&vb->nb);
502 vb->thread = kthread_run(balloon, vb, "vballoon");
503 if (IS_ERR(vb->thread)) {
504 err = PTR_ERR(vb->thread);
511 unregister_oom_notifier(&vb->nb);
513 vdev->config->del_vqs(vdev);
520 static void remove_common(struct virtio_balloon *vb)
522 /* There might be pages left in the balloon: free them. */
523 while (vb->num_pages)
524 leak_balloon(vb, vb->num_pages);
525 update_balloon_size(vb);
527 /* Now we reset the device so we can clean up the queues. */
528 vb->vdev->config->reset(vb->vdev);
530 vb->vdev->config->del_vqs(vb->vdev);
533 static void virtballoon_remove(struct virtio_device *vdev)
535 struct virtio_balloon *vb = vdev->priv;
537 unregister_oom_notifier(&vb->nb);
538 kthread_stop(vb->thread);
543 #ifdef CONFIG_PM_SLEEP
544 static int virtballoon_freeze(struct virtio_device *vdev)
546 struct virtio_balloon *vb = vdev->priv;
549 * The kthread is already frozen by the PM core before this
550 * function is called.
557 static int virtballoon_restore(struct virtio_device *vdev)
559 struct virtio_balloon *vb = vdev->priv;
562 ret = init_vqs(vdev->priv);
566 virtio_device_ready(vdev);
568 fill_balloon(vb, towards_target(vb));
569 update_balloon_size(vb);
574 static unsigned int features[] = {
575 VIRTIO_BALLOON_F_MUST_TELL_HOST,
576 VIRTIO_BALLOON_F_STATS_VQ,
577 VIRTIO_BALLOON_F_DEFLATE_ON_OOM,
580 static struct virtio_driver virtio_balloon_driver = {
581 .feature_table = features,
582 .feature_table_size = ARRAY_SIZE(features),
583 .driver.name = KBUILD_MODNAME,
584 .driver.owner = THIS_MODULE,
585 .id_table = id_table,
586 .probe = virtballoon_probe,
587 .remove = virtballoon_remove,
588 .config_changed = virtballoon_changed,
589 #ifdef CONFIG_PM_SLEEP
590 .freeze = virtballoon_freeze,
591 .restore = virtballoon_restore,
595 module_virtio_driver(virtio_balloon_driver);
596 MODULE_DEVICE_TABLE(virtio, id_table);
597 MODULE_DESCRIPTION("Virtio balloon driver");
598 MODULE_LICENSE("GPL");