1 /*
2 * Virtio balloon implementation, inspired by Dor Laor and Marcelo
3 * Tosatti's implementations.
4 *
5 * Copyright 2008 Rusty Russell IBM Corporation
6 *
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.
11 *
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.
16 *
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
20 */
21
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/wait.h>
32
33 /*
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.
37 */
38 #define VIRTIO_BALLOON_PAGES_PER_PAGE (unsigned)(PAGE_SIZE >> VIRTIO_BALLOON_PFN_SHIFT)
39 #define VIRTIO_BALLOON_ARRAY_PFNS_MAX 256
40
41 struct virtio_balloon
42 {
43 struct virtio_device *vdev;
44 struct virtqueue *inflate_vq, *deflate_vq, *stats_vq;
45
46 /* Where the ballooning thread waits for config to change. */
47 wait_queue_head_t config_change;
48
49 /* The thread servicing the balloon. */
50 struct task_struct *thread;
51
52 /* Waiting for host to ack the pages we released. */
53 wait_queue_head_t acked;
54
55 /* Number of balloon pages we've told the Host we're not using. */
56 unsigned int num_pages;
57 /*
58 * The pages we've told the Host we're not using are enqueued
59 * at vb_dev_info->pages list.
60 * Each page on this list adds VIRTIO_BALLOON_PAGES_PER_PAGE
61 * to num_pages above.
62 */
63 struct balloon_dev_info vb_dev_info;
64
65 /* Synchronize access/update to this struct virtio_balloon elements */
66 struct mutex balloon_lock;
67
68 /* The array of pfns we tell the Host about. */
69 unsigned int num_pfns;
70 u32 pfns[VIRTIO_BALLOON_ARRAY_PFNS_MAX];
71
72 /* Memory statistics */
73 int need_stats_update;
74 struct virtio_balloon_stat stats[VIRTIO_BALLOON_S_NR];
75 };
76
77 static struct virtio_device_id id_table[] = {
78 { VIRTIO_ID_BALLOON, VIRTIO_DEV_ANY_ID },
79 { 0 },
80 };
81
page_to_balloon_pfn(struct page * page)82 static u32 page_to_balloon_pfn(struct page *page)
83 {
84 unsigned long pfn = page_to_pfn(page);
85
86 BUILD_BUG_ON(PAGE_SHIFT < VIRTIO_BALLOON_PFN_SHIFT);
87 /* Convert pfn from Linux page size to balloon page size. */
88 return pfn * VIRTIO_BALLOON_PAGES_PER_PAGE;
89 }
90
balloon_pfn_to_page(u32 pfn)91 static struct page *balloon_pfn_to_page(u32 pfn)
92 {
93 BUG_ON(pfn % VIRTIO_BALLOON_PAGES_PER_PAGE);
94 return pfn_to_page(pfn / VIRTIO_BALLOON_PAGES_PER_PAGE);
95 }
96
balloon_ack(struct virtqueue * vq)97 static void balloon_ack(struct virtqueue *vq)
98 {
99 struct virtio_balloon *vb = vq->vdev->priv;
100
101 wake_up(&vb->acked);
102 }
103
tell_host(struct virtio_balloon * vb,struct virtqueue * vq)104 static void tell_host(struct virtio_balloon *vb, struct virtqueue *vq)
105 {
106 struct scatterlist sg;
107 unsigned int len;
108
109 sg_init_one(&sg, vb->pfns, sizeof(vb->pfns[0]) * vb->num_pfns);
110
111 /* We should always be able to add one buffer to an empty queue. */
112 virtqueue_add_outbuf(vq, &sg, 1, vb, GFP_KERNEL);
113 virtqueue_kick(vq);
114
115 /* When host has read buffer, this completes via balloon_ack */
116 wait_event(vb->acked, virtqueue_get_buf(vq, &len));
117 }
118
set_page_pfns(u32 pfns[],struct page * page)119 static void set_page_pfns(u32 pfns[], struct page *page)
120 {
121 unsigned int i;
122
123 /* Set balloon pfns pointing at this page.
124 * Note that the first pfn points at start of the page. */
125 for (i = 0; i < VIRTIO_BALLOON_PAGES_PER_PAGE; i++)
126 pfns[i] = page_to_balloon_pfn(page) + i;
127 }
128
fill_balloon(struct virtio_balloon * vb,size_t num)129 static void fill_balloon(struct virtio_balloon *vb, size_t num)
130 {
131 struct balloon_dev_info *vb_dev_info = &vb->vb_dev_info;
132
133 /* We can only do one array worth at a time. */
134 num = min(num, ARRAY_SIZE(vb->pfns));
135
136 mutex_lock(&vb->balloon_lock);
137 for (vb->num_pfns = 0; vb->num_pfns < num;
138 vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) {
139 struct page *page = balloon_page_enqueue(vb_dev_info);
140
141 if (!page) {
142 dev_info_ratelimited(&vb->vdev->dev,
143 "Out of puff! Can't get %u pages\n",
144 VIRTIO_BALLOON_PAGES_PER_PAGE);
145 /* Sleep for at least 1/5 of a second before retry. */
146 msleep(200);
147 break;
148 }
149 set_page_pfns(vb->pfns + vb->num_pfns, page);
150 vb->num_pages += VIRTIO_BALLOON_PAGES_PER_PAGE;
151 adjust_managed_page_count(page, -1);
152 }
153
154 /* Did we get any? */
155 if (vb->num_pfns != 0)
156 tell_host(vb, vb->inflate_vq);
157 mutex_unlock(&vb->balloon_lock);
158 }
159
release_pages_by_pfn(const u32 pfns[],unsigned int num)160 static void release_pages_by_pfn(const u32 pfns[], unsigned int num)
161 {
162 unsigned int i;
163
164 /* Find pfns pointing at start of each page, get pages and free them. */
165 for (i = 0; i < num; i += VIRTIO_BALLOON_PAGES_PER_PAGE) {
166 struct page *page = balloon_pfn_to_page(pfns[i]);
167 adjust_managed_page_count(page, 1);
168 put_page(page); /* balloon reference */
169 }
170 }
171
leak_balloon(struct virtio_balloon * vb,size_t num)172 static void leak_balloon(struct virtio_balloon *vb, size_t num)
173 {
174 struct page *page;
175 struct balloon_dev_info *vb_dev_info = &vb->vb_dev_info;
176
177 /* We can only do one array worth at a time. */
178 num = min(num, ARRAY_SIZE(vb->pfns));
179
180 mutex_lock(&vb->balloon_lock);
181 /* We can't release more pages than taken */
182 num = min(num, (size_t)vb->num_pages);
183 for (vb->num_pfns = 0; vb->num_pfns < num;
184 vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) {
185 page = balloon_page_dequeue(vb_dev_info);
186 if (!page)
187 break;
188 set_page_pfns(vb->pfns + vb->num_pfns, page);
189 vb->num_pages -= VIRTIO_BALLOON_PAGES_PER_PAGE;
190 }
191
192 /*
193 * Note that if
194 * virtio_has_feature(vdev, VIRTIO_BALLOON_F_MUST_TELL_HOST);
195 * is true, we *have* to do it in this order
196 */
197 if (vb->num_pfns != 0)
198 tell_host(vb, vb->deflate_vq);
199 mutex_unlock(&vb->balloon_lock);
200 release_pages_by_pfn(vb->pfns, vb->num_pfns);
201 }
202
update_stat(struct virtio_balloon * vb,int idx,u16 tag,u64 val)203 static inline void update_stat(struct virtio_balloon *vb, int idx,
204 u16 tag, u64 val)
205 {
206 BUG_ON(idx >= VIRTIO_BALLOON_S_NR);
207 vb->stats[idx].tag = tag;
208 vb->stats[idx].val = val;
209 }
210
211 #define pages_to_bytes(x) ((u64)(x) << PAGE_SHIFT)
212
update_balloon_stats(struct virtio_balloon * vb)213 static void update_balloon_stats(struct virtio_balloon *vb)
214 {
215 unsigned long events[NR_VM_EVENT_ITEMS];
216 struct sysinfo i;
217 int idx = 0;
218
219 all_vm_events(events);
220 si_meminfo(&i);
221
222 update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_IN,
223 pages_to_bytes(events[PSWPIN]));
224 update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_OUT,
225 pages_to_bytes(events[PSWPOUT]));
226 update_stat(vb, idx++, VIRTIO_BALLOON_S_MAJFLT, events[PGMAJFAULT]);
227 update_stat(vb, idx++, VIRTIO_BALLOON_S_MINFLT, events[PGFAULT]);
228 update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMFREE,
229 pages_to_bytes(i.freeram));
230 update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMTOT,
231 pages_to_bytes(i.totalram));
232 }
233
234 /*
235 * While most virtqueues communicate guest-initiated requests to the hypervisor,
236 * the stats queue operates in reverse. The driver initializes the virtqueue
237 * with a single buffer. From that point forward, all conversations consist of
238 * a hypervisor request (a call to this function) which directs us to refill
239 * the virtqueue with a fresh stats buffer. Since stats collection can sleep,
240 * we notify our kthread which does the actual work via stats_handle_request().
241 */
stats_request(struct virtqueue * vq)242 static void stats_request(struct virtqueue *vq)
243 {
244 struct virtio_balloon *vb = vq->vdev->priv;
245
246 vb->need_stats_update = 1;
247 wake_up(&vb->config_change);
248 }
249
stats_handle_request(struct virtio_balloon * vb)250 static void stats_handle_request(struct virtio_balloon *vb)
251 {
252 struct virtqueue *vq;
253 struct scatterlist sg;
254 unsigned int len;
255
256 vb->need_stats_update = 0;
257 update_balloon_stats(vb);
258
259 vq = vb->stats_vq;
260 if (!virtqueue_get_buf(vq, &len))
261 return;
262 sg_init_one(&sg, vb->stats, sizeof(vb->stats));
263 virtqueue_add_outbuf(vq, &sg, 1, vb, GFP_KERNEL);
264 virtqueue_kick(vq);
265 }
266
virtballoon_changed(struct virtio_device * vdev)267 static void virtballoon_changed(struct virtio_device *vdev)
268 {
269 struct virtio_balloon *vb = vdev->priv;
270
271 wake_up(&vb->config_change);
272 }
273
towards_target(struct virtio_balloon * vb)274 static inline s64 towards_target(struct virtio_balloon *vb)
275 {
276 __le32 v;
277 s64 target;
278
279 virtio_cread(vb->vdev, struct virtio_balloon_config, num_pages, &v);
280
281 target = le32_to_cpu(v);
282 return target - vb->num_pages;
283 }
284
update_balloon_size(struct virtio_balloon * vb)285 static void update_balloon_size(struct virtio_balloon *vb)
286 {
287 __le32 actual = cpu_to_le32(vb->num_pages);
288
289 virtio_cwrite(vb->vdev, struct virtio_balloon_config, actual,
290 &actual);
291 }
292
balloon(void * _vballoon)293 static int balloon(void *_vballoon)
294 {
295 struct virtio_balloon *vb = _vballoon;
296 DEFINE_WAIT_FUNC(wait, woken_wake_function);
297
298 set_freezable();
299 while (!kthread_should_stop()) {
300 s64 diff;
301
302 try_to_freeze();
303
304 add_wait_queue(&vb->config_change, &wait);
305 for (;;) {
306 if ((diff = towards_target(vb)) != 0 ||
307 vb->need_stats_update ||
308 kthread_should_stop() ||
309 freezing(current))
310 break;
311 wait_woken(&wait, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT);
312 }
313 remove_wait_queue(&vb->config_change, &wait);
314
315 if (vb->need_stats_update)
316 stats_handle_request(vb);
317 if (diff > 0)
318 fill_balloon(vb, diff);
319 else if (diff < 0)
320 leak_balloon(vb, -diff);
321 update_balloon_size(vb);
322
323 /*
324 * For large balloon changes, we could spend a lot of time
325 * and always have work to do. Be nice if preempt disabled.
326 */
327 cond_resched();
328 }
329 return 0;
330 }
331
init_vqs(struct virtio_balloon * vb)332 static int init_vqs(struct virtio_balloon *vb)
333 {
334 struct virtqueue *vqs[3];
335 vq_callback_t *callbacks[] = { balloon_ack, balloon_ack, stats_request };
336 const char *names[] = { "inflate", "deflate", "stats" };
337 int err, nvqs;
338
339 /*
340 * We expect two virtqueues: inflate and deflate, and
341 * optionally stat.
342 */
343 nvqs = virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ) ? 3 : 2;
344 err = vb->vdev->config->find_vqs(vb->vdev, nvqs, vqs, callbacks, names);
345 if (err)
346 return err;
347
348 vb->inflate_vq = vqs[0];
349 vb->deflate_vq = vqs[1];
350 if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ)) {
351 struct scatterlist sg;
352 vb->stats_vq = vqs[2];
353
354 /*
355 * Prime this virtqueue with one buffer so the hypervisor can
356 * use it to signal us later (it can't be broken yet!).
357 */
358 update_balloon_stats(vb);
359
360 sg_init_one(&sg, vb->stats, sizeof vb->stats);
361 if (virtqueue_add_outbuf(vb->stats_vq, &sg, 1, vb, GFP_KERNEL)
362 < 0)
363 BUG();
364 virtqueue_kick(vb->stats_vq);
365 }
366 return 0;
367 }
368
369 #ifdef CONFIG_BALLOON_COMPACTION
370 /*
371 * virtballoon_migratepage - perform the balloon page migration on behalf of
372 * a compation thread. (called under page lock)
373 * @vb_dev_info: the balloon device
374 * @newpage: page that will replace the isolated page after migration finishes.
375 * @page : the isolated (old) page that is about to be migrated to newpage.
376 * @mode : compaction mode -- not used for balloon page migration.
377 *
378 * After a ballooned page gets isolated by compaction procedures, this is the
379 * function that performs the page migration on behalf of a compaction thread
380 * The page migration for virtio balloon is done in a simple swap fashion which
381 * follows these two macro steps:
382 * 1) insert newpage into vb->pages list and update the host about it;
383 * 2) update the host about the old page removed from vb->pages list;
384 *
385 * This function preforms the balloon page migration task.
386 * Called through balloon_mapping->a_ops->migratepage
387 */
virtballoon_migratepage(struct balloon_dev_info * vb_dev_info,struct page * newpage,struct page * page,enum migrate_mode mode)388 static int virtballoon_migratepage(struct balloon_dev_info *vb_dev_info,
389 struct page *newpage, struct page *page, enum migrate_mode mode)
390 {
391 struct virtio_balloon *vb = container_of(vb_dev_info,
392 struct virtio_balloon, vb_dev_info);
393 unsigned long flags;
394
395 /*
396 * In order to avoid lock contention while migrating pages concurrently
397 * to leak_balloon() or fill_balloon() we just give up the balloon_lock
398 * this turn, as it is easier to retry the page migration later.
399 * This also prevents fill_balloon() getting stuck into a mutex
400 * recursion in the case it ends up triggering memory compaction
401 * while it is attempting to inflate the ballon.
402 */
403 if (!mutex_trylock(&vb->balloon_lock))
404 return -EAGAIN;
405
406 get_page(newpage); /* balloon reference */
407
408 /* balloon's page migration 1st step -- inflate "newpage" */
409 spin_lock_irqsave(&vb_dev_info->pages_lock, flags);
410 balloon_page_insert(vb_dev_info, newpage);
411 vb_dev_info->isolated_pages--;
412 __count_vm_event(BALLOON_MIGRATE);
413 spin_unlock_irqrestore(&vb_dev_info->pages_lock, flags);
414 vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE;
415 set_page_pfns(vb->pfns, newpage);
416 tell_host(vb, vb->inflate_vq);
417
418 /* balloon's page migration 2nd step -- deflate "page" */
419 balloon_page_delete(page);
420 vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE;
421 set_page_pfns(vb->pfns, page);
422 tell_host(vb, vb->deflate_vq);
423
424 mutex_unlock(&vb->balloon_lock);
425
426 put_page(page); /* balloon reference */
427
428 return MIGRATEPAGE_SUCCESS;
429 }
430 #endif /* CONFIG_BALLOON_COMPACTION */
431
virtballoon_probe(struct virtio_device * vdev)432 static int virtballoon_probe(struct virtio_device *vdev)
433 {
434 struct virtio_balloon *vb;
435 int err;
436
437 vdev->priv = vb = kmalloc(sizeof(*vb), GFP_KERNEL);
438 if (!vb) {
439 err = -ENOMEM;
440 goto out;
441 }
442
443 vb->num_pages = 0;
444 mutex_init(&vb->balloon_lock);
445 init_waitqueue_head(&vb->config_change);
446 init_waitqueue_head(&vb->acked);
447 vb->vdev = vdev;
448 vb->need_stats_update = 0;
449
450 balloon_devinfo_init(&vb->vb_dev_info);
451 #ifdef CONFIG_BALLOON_COMPACTION
452 vb->vb_dev_info.migratepage = virtballoon_migratepage;
453 #endif
454
455 err = init_vqs(vb);
456 if (err)
457 goto out_free_vb;
458
459 virtio_device_ready(vdev);
460
461 vb->thread = kthread_run(balloon, vb, "vballoon");
462 if (IS_ERR(vb->thread)) {
463 err = PTR_ERR(vb->thread);
464 goto out_del_vqs;
465 }
466
467 return 0;
468
469 out_del_vqs:
470 vdev->config->del_vqs(vdev);
471 out_free_vb:
472 kfree(vb);
473 out:
474 return err;
475 }
476
remove_common(struct virtio_balloon * vb)477 static void remove_common(struct virtio_balloon *vb)
478 {
479 /* There might be pages left in the balloon: free them. */
480 while (vb->num_pages)
481 leak_balloon(vb, vb->num_pages);
482 update_balloon_size(vb);
483
484 /* Now we reset the device so we can clean up the queues. */
485 vb->vdev->config->reset(vb->vdev);
486
487 vb->vdev->config->del_vqs(vb->vdev);
488 }
489
virtballoon_remove(struct virtio_device * vdev)490 static void virtballoon_remove(struct virtio_device *vdev)
491 {
492 struct virtio_balloon *vb = vdev->priv;
493
494 kthread_stop(vb->thread);
495 remove_common(vb);
496 kfree(vb);
497 }
498
499 #ifdef CONFIG_PM_SLEEP
virtballoon_freeze(struct virtio_device * vdev)500 static int virtballoon_freeze(struct virtio_device *vdev)
501 {
502 struct virtio_balloon *vb = vdev->priv;
503
504 /*
505 * The kthread is already frozen by the PM core before this
506 * function is called.
507 */
508
509 remove_common(vb);
510 return 0;
511 }
512
virtballoon_restore(struct virtio_device * vdev)513 static int virtballoon_restore(struct virtio_device *vdev)
514 {
515 struct virtio_balloon *vb = vdev->priv;
516 int ret;
517
518 ret = init_vqs(vdev->priv);
519 if (ret)
520 return ret;
521
522 virtio_device_ready(vdev);
523
524 fill_balloon(vb, towards_target(vb));
525 update_balloon_size(vb);
526 return 0;
527 }
528 #endif
529
530 static unsigned int features[] = {
531 VIRTIO_BALLOON_F_MUST_TELL_HOST,
532 VIRTIO_BALLOON_F_STATS_VQ,
533 };
534
535 static struct virtio_driver virtio_balloon_driver = {
536 .feature_table = features,
537 .feature_table_size = ARRAY_SIZE(features),
538 .driver.name = KBUILD_MODNAME,
539 .driver.owner = THIS_MODULE,
540 .id_table = id_table,
541 .probe = virtballoon_probe,
542 .remove = virtballoon_remove,
543 .config_changed = virtballoon_changed,
544 #ifdef CONFIG_PM_SLEEP
545 .freeze = virtballoon_freeze,
546 .restore = virtballoon_restore,
547 #endif
548 };
549
550 module_virtio_driver(virtio_balloon_driver);
551 MODULE_DEVICE_TABLE(virtio, id_table);
552 MODULE_DESCRIPTION("Virtio balloon driver");
553 MODULE_LICENSE("GPL");
554