Merge branch 'linus' into irq/core
[linux-2.6/btrfs-unstable.git] / drivers / virtio / virtio_balloon.c
blob50c5f42d7a9f3010c5f12275b7fd540161900392
1 /*
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
36 * page units.
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
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
67 * to num_pages above.
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 },
88 { 0 },
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;
110 wake_up(&vb->acked);
113 static void tell_host(struct virtio_balloon *vb, struct virtqueue *vq)
115 struct scatterlist sg;
116 unsigned int len;
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);
122 virtqueue_kick(vq);
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)
130 unsigned int i;
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 void fill_balloon(struct virtio_balloon *vb, size_t num)
140 struct balloon_dev_info *vb_dev_info = &vb->vb_dev_info;
142 /* We can only do one array worth at a time. */
143 num = min(num, ARRAY_SIZE(vb->pfns));
145 mutex_lock(&vb->balloon_lock);
146 for (vb->num_pfns = 0; vb->num_pfns < num;
147 vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) {
148 struct page *page = balloon_page_enqueue(vb_dev_info);
150 if (!page) {
151 dev_info_ratelimited(&vb->vdev->dev,
152 "Out of puff! Can't get %u pages\n",
153 VIRTIO_BALLOON_PAGES_PER_PAGE);
154 /* Sleep for at least 1/5 of a second before retry. */
155 msleep(200);
156 break;
158 set_page_pfns(vb->pfns + vb->num_pfns, page);
159 vb->num_pages += VIRTIO_BALLOON_PAGES_PER_PAGE;
160 adjust_managed_page_count(page, -1);
163 /* Did we get any? */
164 if (vb->num_pfns != 0)
165 tell_host(vb, vb->inflate_vq);
166 mutex_unlock(&vb->balloon_lock);
169 static void release_pages_by_pfn(const u32 pfns[], unsigned int num)
171 unsigned int i;
173 /* Find pfns pointing at start of each page, get pages and free them. */
174 for (i = 0; i < num; i += VIRTIO_BALLOON_PAGES_PER_PAGE) {
175 struct page *page = balloon_pfn_to_page(pfns[i]);
176 adjust_managed_page_count(page, 1);
177 put_page(page); /* balloon reference */
181 static unsigned leak_balloon(struct virtio_balloon *vb, size_t num)
183 unsigned num_freed_pages;
184 struct page *page;
185 struct balloon_dev_info *vb_dev_info = &vb->vb_dev_info;
187 /* We can only do one array worth at a time. */
188 num = min(num, ARRAY_SIZE(vb->pfns));
190 mutex_lock(&vb->balloon_lock);
191 for (vb->num_pfns = 0; vb->num_pfns < num;
192 vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) {
193 page = balloon_page_dequeue(vb_dev_info);
194 if (!page)
195 break;
196 set_page_pfns(vb->pfns + vb->num_pfns, page);
197 vb->num_pages -= VIRTIO_BALLOON_PAGES_PER_PAGE;
200 num_freed_pages = vb->num_pfns;
202 * Note that if
203 * virtio_has_feature(vdev, VIRTIO_BALLOON_F_MUST_TELL_HOST);
204 * is true, we *have* to do it in this order
206 if (vb->num_pfns != 0)
207 tell_host(vb, vb->deflate_vq);
208 mutex_unlock(&vb->balloon_lock);
209 release_pages_by_pfn(vb->pfns, vb->num_pfns);
210 return num_freed_pages;
213 static inline void update_stat(struct virtio_balloon *vb, int idx,
214 u16 tag, u64 val)
216 BUG_ON(idx >= VIRTIO_BALLOON_S_NR);
217 vb->stats[idx].tag = tag;
218 vb->stats[idx].val = val;
221 #define pages_to_bytes(x) ((u64)(x) << PAGE_SHIFT)
223 static void update_balloon_stats(struct virtio_balloon *vb)
225 unsigned long events[NR_VM_EVENT_ITEMS];
226 struct sysinfo i;
227 int idx = 0;
229 all_vm_events(events);
230 si_meminfo(&i);
232 update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_IN,
233 pages_to_bytes(events[PSWPIN]));
234 update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_OUT,
235 pages_to_bytes(events[PSWPOUT]));
236 update_stat(vb, idx++, VIRTIO_BALLOON_S_MAJFLT, events[PGMAJFAULT]);
237 update_stat(vb, idx++, VIRTIO_BALLOON_S_MINFLT, events[PGFAULT]);
238 update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMFREE,
239 pages_to_bytes(i.freeram));
240 update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMTOT,
241 pages_to_bytes(i.totalram));
245 * While most virtqueues communicate guest-initiated requests to the hypervisor,
246 * the stats queue operates in reverse. The driver initializes the virtqueue
247 * with a single buffer. From that point forward, all conversations consist of
248 * a hypervisor request (a call to this function) which directs us to refill
249 * the virtqueue with a fresh stats buffer. Since stats collection can sleep,
250 * we notify our kthread which does the actual work via stats_handle_request().
252 static void stats_request(struct virtqueue *vq)
254 struct virtio_balloon *vb = vq->vdev->priv;
256 vb->need_stats_update = 1;
257 wake_up(&vb->config_change);
260 static void stats_handle_request(struct virtio_balloon *vb)
262 struct virtqueue *vq;
263 struct scatterlist sg;
264 unsigned int len;
266 vb->need_stats_update = 0;
267 update_balloon_stats(vb);
269 vq = vb->stats_vq;
270 if (!virtqueue_get_buf(vq, &len))
271 return;
272 sg_init_one(&sg, vb->stats, sizeof(vb->stats));
273 virtqueue_add_outbuf(vq, &sg, 1, vb, GFP_KERNEL);
274 virtqueue_kick(vq);
277 static void virtballoon_changed(struct virtio_device *vdev)
279 struct virtio_balloon *vb = vdev->priv;
281 wake_up(&vb->config_change);
284 static inline s64 towards_target(struct virtio_balloon *vb)
286 __le32 v;
287 s64 target;
289 virtio_cread(vb->vdev, struct virtio_balloon_config, num_pages, &v);
291 target = le32_to_cpu(v);
292 return target - vb->num_pages;
295 static void update_balloon_size(struct virtio_balloon *vb)
297 __le32 actual = cpu_to_le32(vb->num_pages);
299 virtio_cwrite(vb->vdev, struct virtio_balloon_config, actual,
300 &actual);
304 * virtballoon_oom_notify - release pages when system is under severe
305 * memory pressure (called from out_of_memory())
306 * @self : notifier block struct
307 * @dummy: not used
308 * @parm : returned - number of freed pages
310 * The balancing of memory by use of the virtio balloon should not cause
311 * the termination of processes while there are pages in the balloon.
312 * If virtio balloon manages to release some memory, it will make the
313 * system return and retry the allocation that forced the OOM killer
314 * to run.
316 static int virtballoon_oom_notify(struct notifier_block *self,
317 unsigned long dummy, void *parm)
319 struct virtio_balloon *vb;
320 unsigned long *freed;
321 unsigned num_freed_pages;
323 vb = container_of(self, struct virtio_balloon, nb);
324 if (!virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_DEFLATE_ON_OOM))
325 return NOTIFY_OK;
327 freed = parm;
328 num_freed_pages = leak_balloon(vb, oom_pages);
329 update_balloon_size(vb);
330 *freed += num_freed_pages;
332 return NOTIFY_OK;
335 static int balloon(void *_vballoon)
337 struct virtio_balloon *vb = _vballoon;
339 set_freezable();
340 while (!kthread_should_stop()) {
341 s64 diff;
343 try_to_freeze();
344 wait_event_interruptible(vb->config_change,
345 (diff = towards_target(vb)) != 0
346 || vb->need_stats_update
347 || kthread_should_stop()
348 || freezing(current));
349 if (vb->need_stats_update)
350 stats_handle_request(vb);
351 if (diff > 0)
352 fill_balloon(vb, diff);
353 else if (diff < 0)
354 leak_balloon(vb, -diff);
355 update_balloon_size(vb);
358 * For large balloon changes, we could spend a lot of time
359 * and always have work to do. Be nice if preempt disabled.
361 cond_resched();
363 return 0;
366 static int init_vqs(struct virtio_balloon *vb)
368 struct virtqueue *vqs[3];
369 vq_callback_t *callbacks[] = { balloon_ack, balloon_ack, stats_request };
370 const char *names[] = { "inflate", "deflate", "stats" };
371 int err, nvqs;
374 * We expect two virtqueues: inflate and deflate, and
375 * optionally stat.
377 nvqs = virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ) ? 3 : 2;
378 err = vb->vdev->config->find_vqs(vb->vdev, nvqs, vqs, callbacks, names);
379 if (err)
380 return err;
382 vb->inflate_vq = vqs[0];
383 vb->deflate_vq = vqs[1];
384 if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ)) {
385 struct scatterlist sg;
386 vb->stats_vq = vqs[2];
389 * Prime this virtqueue with one buffer so the hypervisor can
390 * use it to signal us later (it can't be broken yet!).
392 sg_init_one(&sg, vb->stats, sizeof vb->stats);
393 if (virtqueue_add_outbuf(vb->stats_vq, &sg, 1, vb, GFP_KERNEL)
394 < 0)
395 BUG();
396 virtqueue_kick(vb->stats_vq);
398 return 0;
401 #ifdef CONFIG_BALLOON_COMPACTION
403 * virtballoon_migratepage - perform the balloon page migration on behalf of
404 * a compation thread. (called under page lock)
405 * @vb_dev_info: the balloon device
406 * @newpage: page that will replace the isolated page after migration finishes.
407 * @page : the isolated (old) page that is about to be migrated to newpage.
408 * @mode : compaction mode -- not used for balloon page migration.
410 * After a ballooned page gets isolated by compaction procedures, this is the
411 * function that performs the page migration on behalf of a compaction thread
412 * The page migration for virtio balloon is done in a simple swap fashion which
413 * follows these two macro steps:
414 * 1) insert newpage into vb->pages list and update the host about it;
415 * 2) update the host about the old page removed from vb->pages list;
417 * This function preforms the balloon page migration task.
418 * Called through balloon_mapping->a_ops->migratepage
420 static int virtballoon_migratepage(struct balloon_dev_info *vb_dev_info,
421 struct page *newpage, struct page *page, enum migrate_mode mode)
423 struct virtio_balloon *vb = container_of(vb_dev_info,
424 struct virtio_balloon, vb_dev_info);
425 unsigned long flags;
428 * In order to avoid lock contention while migrating pages concurrently
429 * to leak_balloon() or fill_balloon() we just give up the balloon_lock
430 * this turn, as it is easier to retry the page migration later.
431 * This also prevents fill_balloon() getting stuck into a mutex
432 * recursion in the case it ends up triggering memory compaction
433 * while it is attempting to inflate the ballon.
435 if (!mutex_trylock(&vb->balloon_lock))
436 return -EAGAIN;
438 get_page(newpage); /* balloon reference */
440 /* balloon's page migration 1st step -- inflate "newpage" */
441 spin_lock_irqsave(&vb_dev_info->pages_lock, flags);
442 balloon_page_insert(vb_dev_info, newpage);
443 vb_dev_info->isolated_pages--;
444 __count_vm_event(BALLOON_MIGRATE);
445 spin_unlock_irqrestore(&vb_dev_info->pages_lock, flags);
446 vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE;
447 set_page_pfns(vb->pfns, newpage);
448 tell_host(vb, vb->inflate_vq);
450 /* balloon's page migration 2nd step -- deflate "page" */
451 balloon_page_delete(page);
452 vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE;
453 set_page_pfns(vb->pfns, page);
454 tell_host(vb, vb->deflate_vq);
456 mutex_unlock(&vb->balloon_lock);
458 put_page(page); /* balloon reference */
460 return MIGRATEPAGE_SUCCESS;
462 #endif /* CONFIG_BALLOON_COMPACTION */
464 static int virtballoon_probe(struct virtio_device *vdev)
466 struct virtio_balloon *vb;
467 int err;
469 vdev->priv = vb = kmalloc(sizeof(*vb), GFP_KERNEL);
470 if (!vb) {
471 err = -ENOMEM;
472 goto out;
475 vb->num_pages = 0;
476 mutex_init(&vb->balloon_lock);
477 init_waitqueue_head(&vb->config_change);
478 init_waitqueue_head(&vb->acked);
479 vb->vdev = vdev;
480 vb->need_stats_update = 0;
482 balloon_devinfo_init(&vb->vb_dev_info);
483 #ifdef CONFIG_BALLOON_COMPACTION
484 vb->vb_dev_info.migratepage = virtballoon_migratepage;
485 #endif
487 err = init_vqs(vb);
488 if (err)
489 goto out_free_vb;
491 vb->nb.notifier_call = virtballoon_oom_notify;
492 vb->nb.priority = VIRTBALLOON_OOM_NOTIFY_PRIORITY;
493 err = register_oom_notifier(&vb->nb);
494 if (err < 0)
495 goto out_oom_notify;
497 vb->thread = kthread_run(balloon, vb, "vballoon");
498 if (IS_ERR(vb->thread)) {
499 err = PTR_ERR(vb->thread);
500 goto out_del_vqs;
503 return 0;
505 out_del_vqs:
506 unregister_oom_notifier(&vb->nb);
507 out_oom_notify:
508 vdev->config->del_vqs(vdev);
509 out_free_vb:
510 kfree(vb);
511 out:
512 return err;
515 static void remove_common(struct virtio_balloon *vb)
517 /* There might be pages left in the balloon: free them. */
518 while (vb->num_pages)
519 leak_balloon(vb, vb->num_pages);
520 update_balloon_size(vb);
522 /* Now we reset the device so we can clean up the queues. */
523 vb->vdev->config->reset(vb->vdev);
525 vb->vdev->config->del_vqs(vb->vdev);
528 static void virtballoon_remove(struct virtio_device *vdev)
530 struct virtio_balloon *vb = vdev->priv;
532 unregister_oom_notifier(&vb->nb);
533 kthread_stop(vb->thread);
534 remove_common(vb);
535 kfree(vb);
538 #ifdef CONFIG_PM_SLEEP
539 static int virtballoon_freeze(struct virtio_device *vdev)
541 struct virtio_balloon *vb = vdev->priv;
544 * The kthread is already frozen by the PM core before this
545 * function is called.
548 remove_common(vb);
549 return 0;
552 static int virtballoon_restore(struct virtio_device *vdev)
554 struct virtio_balloon *vb = vdev->priv;
555 int ret;
557 ret = init_vqs(vdev->priv);
558 if (ret)
559 return ret;
561 virtio_device_ready(vdev);
563 fill_balloon(vb, towards_target(vb));
564 update_balloon_size(vb);
565 return 0;
567 #endif
569 static unsigned int features[] = {
570 VIRTIO_BALLOON_F_MUST_TELL_HOST,
571 VIRTIO_BALLOON_F_STATS_VQ,
572 VIRTIO_BALLOON_F_DEFLATE_ON_OOM,
575 static struct virtio_driver virtio_balloon_driver = {
576 .feature_table = features,
577 .feature_table_size = ARRAY_SIZE(features),
578 .driver.name = KBUILD_MODNAME,
579 .driver.owner = THIS_MODULE,
580 .id_table = id_table,
581 .probe = virtballoon_probe,
582 .remove = virtballoon_remove,
583 .config_changed = virtballoon_changed,
584 #ifdef CONFIG_PM_SLEEP
585 .freeze = virtballoon_freeze,
586 .restore = virtballoon_restore,
587 #endif
590 module_virtio_driver(virtio_balloon_driver);
591 MODULE_DEVICE_TABLE(virtio, id_table);
592 MODULE_DESCRIPTION("Virtio balloon driver");
593 MODULE_LICENSE("GPL");