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>
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 /* The balloon servicing is delegated to a freezable workqueue. */
52 struct work_struct work;
54 /* Prevent updating balloon when it is being canceled. */
55 spinlock_t stop_update_lock;
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 u32 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(u32 pfns[], struct page *page)
132 /* Set balloon pfns pointing at this page.
133 * Note that the first pfn points at start of the page. */
134 for (i = 0; i < VIRTIO_BALLOON_PAGES_PER_PAGE; i++)
135 pfns[i] = page_to_balloon_pfn(page) + i;
138 static unsigned fill_balloon(struct virtio_balloon *vb, size_t num)
140 struct balloon_dev_info *vb_dev_info = &vb->vb_dev_info;
141 unsigned num_allocated_pages;
143 /* We can only do one array worth at a time. */
144 num = min(num, ARRAY_SIZE(vb->pfns));
146 mutex_lock(&vb->balloon_lock);
147 for (vb->num_pfns = 0; vb->num_pfns < num;
148 vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) {
149 struct page *page = balloon_page_enqueue(vb_dev_info);
152 dev_info_ratelimited(&vb->vdev->dev,
153 "Out of puff! Can't get %u pages\n",
154 VIRTIO_BALLOON_PAGES_PER_PAGE);
155 /* Sleep for at least 1/5 of a second before retry. */
159 set_page_pfns(vb->pfns + vb->num_pfns, page);
160 vb->num_pages += VIRTIO_BALLOON_PAGES_PER_PAGE;
161 if (!virtio_has_feature(vb->vdev,
162 VIRTIO_BALLOON_F_DEFLATE_ON_OOM))
163 adjust_managed_page_count(page, -1);
166 num_allocated_pages = vb->num_pfns;
167 /* Did we get any? */
168 if (vb->num_pfns != 0)
169 tell_host(vb, vb->inflate_vq);
170 mutex_unlock(&vb->balloon_lock);
172 return num_allocated_pages;
175 static void release_pages_balloon(struct virtio_balloon *vb)
179 /* Find pfns pointing at start of each page, get pages and free them. */
180 for (i = 0; i < vb->num_pfns; i += VIRTIO_BALLOON_PAGES_PER_PAGE) {
181 struct page *page = balloon_pfn_to_page(vb->pfns[i]);
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 for (vb->num_pfns = 0; vb->num_pfns < num;
200 vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) {
201 page = balloon_page_dequeue(vb_dev_info);
204 set_page_pfns(vb->pfns + vb->num_pfns, page);
205 vb->num_pages -= VIRTIO_BALLOON_PAGES_PER_PAGE;
208 num_freed_pages = vb->num_pfns;
211 * virtio_has_feature(vdev, VIRTIO_BALLOON_F_MUST_TELL_HOST);
212 * is true, we *have* to do it in this order
214 if (vb->num_pfns != 0)
215 tell_host(vb, vb->deflate_vq);
216 release_pages_balloon(vb);
217 mutex_unlock(&vb->balloon_lock);
218 return num_freed_pages;
221 static inline void update_stat(struct virtio_balloon *vb, int idx,
224 BUG_ON(idx >= VIRTIO_BALLOON_S_NR);
225 vb->stats[idx].tag = cpu_to_virtio16(vb->vdev, tag);
226 vb->stats[idx].val = cpu_to_virtio64(vb->vdev, val);
229 #define pages_to_bytes(x) ((u64)(x) << PAGE_SHIFT)
231 static void update_balloon_stats(struct virtio_balloon *vb)
233 unsigned long events[NR_VM_EVENT_ITEMS];
237 all_vm_events(events);
240 update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_IN,
241 pages_to_bytes(events[PSWPIN]));
242 update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_OUT,
243 pages_to_bytes(events[PSWPOUT]));
244 update_stat(vb, idx++, VIRTIO_BALLOON_S_MAJFLT, events[PGMAJFAULT]);
245 update_stat(vb, idx++, VIRTIO_BALLOON_S_MINFLT, events[PGFAULT]);
246 update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMFREE,
247 pages_to_bytes(i.freeram));
248 update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMTOT,
249 pages_to_bytes(i.totalram));
253 * While most virtqueues communicate guest-initiated requests to the hypervisor,
254 * the stats queue operates in reverse. The driver initializes the virtqueue
255 * with a single buffer. From that point forward, all conversations consist of
256 * a hypervisor request (a call to this function) which directs us to refill
257 * the virtqueue with a fresh stats buffer. Since stats collection can sleep,
258 * we delegate the job to a freezable workqueue that will do the actual work via
259 * stats_handle_request().
261 static void stats_request(struct virtqueue *vq)
263 struct virtio_balloon *vb = vq->vdev->priv;
265 vb->need_stats_update = 1;
267 spin_lock(&vb->stop_update_lock);
268 if (!vb->stop_update)
269 queue_work(system_freezable_wq, &vb->work);
270 spin_unlock(&vb->stop_update_lock);
273 static void stats_handle_request(struct virtio_balloon *vb)
275 struct virtqueue *vq;
276 struct scatterlist sg;
279 vb->need_stats_update = 0;
280 update_balloon_stats(vb);
283 if (!virtqueue_get_buf(vq, &len))
285 sg_init_one(&sg, vb->stats, sizeof(vb->stats));
286 virtqueue_add_outbuf(vq, &sg, 1, vb, GFP_KERNEL);
290 static void virtballoon_changed(struct virtio_device *vdev)
292 struct virtio_balloon *vb = vdev->priv;
295 spin_lock_irqsave(&vb->stop_update_lock, flags);
296 if (!vb->stop_update)
297 queue_work(system_freezable_wq, &vb->work);
298 spin_unlock_irqrestore(&vb->stop_update_lock, flags);
301 static inline s64 towards_target(struct virtio_balloon *vb)
306 virtio_cread(vb->vdev, struct virtio_balloon_config, num_pages,
309 /* Legacy balloon config space is LE, unlike all other devices. */
310 if (!virtio_has_feature(vb->vdev, VIRTIO_F_VERSION_1))
311 num_pages = le32_to_cpu((__force __le32)num_pages);
314 return target - vb->num_pages;
317 static void update_balloon_size(struct virtio_balloon *vb)
319 u32 actual = vb->num_pages;
321 /* Legacy balloon config space is LE, unlike all other devices. */
322 if (!virtio_has_feature(vb->vdev, VIRTIO_F_VERSION_1))
323 actual = (__force u32)cpu_to_le32(actual);
325 virtio_cwrite(vb->vdev, struct virtio_balloon_config, actual,
330 * virtballoon_oom_notify - release pages when system is under severe
331 * memory pressure (called from out_of_memory())
332 * @self : notifier block struct
334 * @parm : returned - number of freed pages
336 * The balancing of memory by use of the virtio balloon should not cause
337 * the termination of processes while there are pages in the balloon.
338 * If virtio balloon manages to release some memory, it will make the
339 * system return and retry the allocation that forced the OOM killer
342 static int virtballoon_oom_notify(struct notifier_block *self,
343 unsigned long dummy, void *parm)
345 struct virtio_balloon *vb;
346 unsigned long *freed;
347 unsigned num_freed_pages;
349 vb = container_of(self, struct virtio_balloon, nb);
350 if (!virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_DEFLATE_ON_OOM))
354 num_freed_pages = leak_balloon(vb, oom_pages);
355 update_balloon_size(vb);
356 *freed += num_freed_pages;
361 static void balloon(struct work_struct *work)
363 struct virtio_balloon *vb;
366 vb = container_of(work, struct virtio_balloon, work);
367 diff = towards_target(vb);
369 if (vb->need_stats_update)
370 stats_handle_request(vb);
373 diff -= fill_balloon(vb, diff);
375 diff += leak_balloon(vb, -diff);
376 update_balloon_size(vb);
379 queue_work(system_freezable_wq, work);
382 static int init_vqs(struct virtio_balloon *vb)
384 struct virtqueue *vqs[3];
385 vq_callback_t *callbacks[] = { balloon_ack, balloon_ack, stats_request };
386 static const char * const names[] = { "inflate", "deflate", "stats" };
390 * We expect two virtqueues: inflate and deflate, and
393 nvqs = virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ) ? 3 : 2;
394 err = vb->vdev->config->find_vqs(vb->vdev, nvqs, vqs, callbacks, names);
398 vb->inflate_vq = vqs[0];
399 vb->deflate_vq = vqs[1];
400 if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ)) {
401 struct scatterlist sg;
402 vb->stats_vq = vqs[2];
405 * Prime this virtqueue with one buffer so the hypervisor can
406 * use it to signal us later (it can't be broken yet!).
408 sg_init_one(&sg, vb->stats, sizeof vb->stats);
409 if (virtqueue_add_outbuf(vb->stats_vq, &sg, 1, vb, GFP_KERNEL)
412 virtqueue_kick(vb->stats_vq);
417 #ifdef CONFIG_BALLOON_COMPACTION
419 * virtballoon_migratepage - perform the balloon page migration on behalf of
420 * a compation thread. (called under page lock)
421 * @vb_dev_info: the balloon device
422 * @newpage: page that will replace the isolated page after migration finishes.
423 * @page : the isolated (old) page that is about to be migrated to newpage.
424 * @mode : compaction mode -- not used for balloon page migration.
426 * After a ballooned page gets isolated by compaction procedures, this is the
427 * function that performs the page migration on behalf of a compaction thread
428 * The page migration for virtio balloon is done in a simple swap fashion which
429 * follows these two macro steps:
430 * 1) insert newpage into vb->pages list and update the host about it;
431 * 2) update the host about the old page removed from vb->pages list;
433 * This function preforms the balloon page migration task.
434 * Called through balloon_mapping->a_ops->migratepage
436 static int virtballoon_migratepage(struct balloon_dev_info *vb_dev_info,
437 struct page *newpage, struct page *page, enum migrate_mode mode)
439 struct virtio_balloon *vb = container_of(vb_dev_info,
440 struct virtio_balloon, vb_dev_info);
444 * In order to avoid lock contention while migrating pages concurrently
445 * to leak_balloon() or fill_balloon() we just give up the balloon_lock
446 * this turn, as it is easier to retry the page migration later.
447 * This also prevents fill_balloon() getting stuck into a mutex
448 * recursion in the case it ends up triggering memory compaction
449 * while it is attempting to inflate the ballon.
451 if (!mutex_trylock(&vb->balloon_lock))
454 get_page(newpage); /* balloon reference */
456 /* balloon's page migration 1st step -- inflate "newpage" */
457 spin_lock_irqsave(&vb_dev_info->pages_lock, flags);
458 balloon_page_insert(vb_dev_info, newpage);
459 vb_dev_info->isolated_pages--;
460 __count_vm_event(BALLOON_MIGRATE);
461 spin_unlock_irqrestore(&vb_dev_info->pages_lock, flags);
462 vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE;
463 set_page_pfns(vb->pfns, newpage);
464 tell_host(vb, vb->inflate_vq);
466 /* balloon's page migration 2nd step -- deflate "page" */
467 balloon_page_delete(page);
468 vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE;
469 set_page_pfns(vb->pfns, page);
470 tell_host(vb, vb->deflate_vq);
472 mutex_unlock(&vb->balloon_lock);
474 put_page(page); /* balloon reference */
476 return MIGRATEPAGE_SUCCESS;
478 #endif /* CONFIG_BALLOON_COMPACTION */
480 static int virtballoon_probe(struct virtio_device *vdev)
482 struct virtio_balloon *vb;
485 if (!vdev->config->get) {
486 dev_err(&vdev->dev, "%s failure: config access disabled\n",
491 vdev->priv = vb = kmalloc(sizeof(*vb), GFP_KERNEL);
497 INIT_WORK(&vb->work, balloon);
498 spin_lock_init(&vb->stop_update_lock);
499 vb->stop_update = false;
501 mutex_init(&vb->balloon_lock);
502 init_waitqueue_head(&vb->acked);
504 vb->need_stats_update = 0;
506 balloon_devinfo_init(&vb->vb_dev_info);
507 #ifdef CONFIG_BALLOON_COMPACTION
508 vb->vb_dev_info.migratepage = virtballoon_migratepage;
515 vb->nb.notifier_call = virtballoon_oom_notify;
516 vb->nb.priority = VIRTBALLOON_OOM_NOTIFY_PRIORITY;
517 err = register_oom_notifier(&vb->nb);
521 virtio_device_ready(vdev);
526 vdev->config->del_vqs(vdev);
533 static void remove_common(struct virtio_balloon *vb)
535 /* There might be pages left in the balloon: free them. */
536 while (vb->num_pages)
537 leak_balloon(vb, vb->num_pages);
538 update_balloon_size(vb);
540 /* Now we reset the device so we can clean up the queues. */
541 vb->vdev->config->reset(vb->vdev);
543 vb->vdev->config->del_vqs(vb->vdev);
546 static void virtballoon_remove(struct virtio_device *vdev)
548 struct virtio_balloon *vb = vdev->priv;
550 unregister_oom_notifier(&vb->nb);
552 spin_lock_irq(&vb->stop_update_lock);
553 vb->stop_update = true;
554 spin_unlock_irq(&vb->stop_update_lock);
555 cancel_work_sync(&vb->work);
561 #ifdef CONFIG_PM_SLEEP
562 static int virtballoon_freeze(struct virtio_device *vdev)
564 struct virtio_balloon *vb = vdev->priv;
567 * The workqueue is already frozen by the PM core before this
568 * function is called.
574 static int virtballoon_restore(struct virtio_device *vdev)
576 struct virtio_balloon *vb = vdev->priv;
579 ret = init_vqs(vdev->priv);
583 virtio_device_ready(vdev);
585 if (towards_target(vb))
586 virtballoon_changed(vdev);
587 update_balloon_size(vb);
592 static unsigned int features[] = {
593 VIRTIO_BALLOON_F_MUST_TELL_HOST,
594 VIRTIO_BALLOON_F_STATS_VQ,
595 VIRTIO_BALLOON_F_DEFLATE_ON_OOM,
598 static struct virtio_driver virtio_balloon_driver = {
599 .feature_table = features,
600 .feature_table_size = ARRAY_SIZE(features),
601 .driver.name = KBUILD_MODNAME,
602 .driver.owner = THIS_MODULE,
603 .id_table = id_table,
604 .probe = virtballoon_probe,
605 .remove = virtballoon_remove,
606 .config_changed = virtballoon_changed,
607 #ifdef CONFIG_PM_SLEEP
608 .freeze = virtballoon_freeze,
609 .restore = virtballoon_restore,
613 module_virtio_driver(virtio_balloon_driver);
614 MODULE_DEVICE_TABLE(virtio, id_table);
615 MODULE_DESCRIPTION("Virtio balloon driver");
616 MODULE_LICENSE("GPL");