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 
32 /*
33  * Balloon device works in 4K page units.  So each page is pointed to by
34  * multiple balloon pages.  All memory counters in this driver are in balloon
35  * page units.
36  */
37 #define VIRTIO_BALLOON_PAGES_PER_PAGE (unsigned)(PAGE_SIZE >> VIRTIO_BALLOON_PFN_SHIFT)
38 #define VIRTIO_BALLOON_ARRAY_PFNS_MAX 256
39 
40 struct virtio_balloon
41 {
42 	struct virtio_device *vdev;
43 	struct virtqueue *inflate_vq, *deflate_vq, *stats_vq;
44 
45 	/* Where the ballooning thread waits for config to change. */
46 	wait_queue_head_t config_change;
47 
48 	/* The thread servicing the balloon. */
49 	struct task_struct *thread;
50 
51 	/* Waiting for host to ack the pages we released. */
52 	wait_queue_head_t acked;
53 
54 	/* Number of balloon pages we've told the Host we're not using. */
55 	unsigned int num_pages;
56 	/*
57 	 * The pages we've told the Host we're not using are enqueued
58 	 * at vb_dev_info->pages list.
59 	 * Each page on this list adds VIRTIO_BALLOON_PAGES_PER_PAGE
60 	 * to num_pages above.
61 	 */
62 	struct balloon_dev_info *vb_dev_info;
63 
64 	/* Synchronize access/update to this struct virtio_balloon elements */
65 	struct mutex balloon_lock;
66 
67 	/* The array of pfns we tell the Host about. */
68 	unsigned int num_pfns;
69 	u32 pfns[VIRTIO_BALLOON_ARRAY_PFNS_MAX];
70 
71 	/* Memory statistics */
72 	int need_stats_update;
73 	struct virtio_balloon_stat stats[VIRTIO_BALLOON_S_NR];
74 };
75 
76 static struct virtio_device_id id_table[] = {
77 	{ VIRTIO_ID_BALLOON, VIRTIO_DEV_ANY_ID },
78 	{ 0 },
79 };
80 
81 static u32 page_to_balloon_pfn(struct page *page)
82 {
83 	unsigned long pfn = page_to_pfn(page);
84 
85 	BUILD_BUG_ON(PAGE_SHIFT < VIRTIO_BALLOON_PFN_SHIFT);
86 	/* Convert pfn from Linux page size to balloon page size. */
87 	return pfn * VIRTIO_BALLOON_PAGES_PER_PAGE;
88 }
89 
90 static struct page *balloon_pfn_to_page(u32 pfn)
91 {
92 	BUG_ON(pfn % VIRTIO_BALLOON_PAGES_PER_PAGE);
93 	return pfn_to_page(pfn / VIRTIO_BALLOON_PAGES_PER_PAGE);
94 }
95 
96 static void balloon_ack(struct virtqueue *vq)
97 {
98 	struct virtio_balloon *vb = vq->vdev->priv;
99 
100 	wake_up(&vb->acked);
101 }
102 
103 static void tell_host(struct virtio_balloon *vb, struct virtqueue *vq)
104 {
105 	struct scatterlist sg;
106 	unsigned int len;
107 
108 	sg_init_one(&sg, vb->pfns, sizeof(vb->pfns[0]) * vb->num_pfns);
109 
110 	/* We should always be able to add one buffer to an empty queue. */
111 	if (virtqueue_add_outbuf(vq, &sg, 1, vb, GFP_KERNEL) < 0)
112 		BUG();
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 
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 
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 
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 		balloon_page_free(page);
168 		adjust_managed_page_count(page, 1);
169 	}
170 }
171 
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 	for (vb->num_pfns = 0; vb->num_pfns < num;
182 	     vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) {
183 		page = balloon_page_dequeue(vb_dev_info);
184 		if (!page)
185 			break;
186 		set_page_pfns(vb->pfns + vb->num_pfns, page);
187 		vb->num_pages -= VIRTIO_BALLOON_PAGES_PER_PAGE;
188 	}
189 
190 	/*
191 	 * Note that if
192 	 * virtio_has_feature(vdev, VIRTIO_BALLOON_F_MUST_TELL_HOST);
193 	 * is true, we *have* to do it in this order
194 	 */
195 	if (vb->num_pfns != 0)
196 		tell_host(vb, vb->deflate_vq);
197 	mutex_unlock(&vb->balloon_lock);
198 	release_pages_by_pfn(vb->pfns, vb->num_pfns);
199 }
200 
201 static inline void update_stat(struct virtio_balloon *vb, int idx,
202 			       u16 tag, u64 val)
203 {
204 	BUG_ON(idx >= VIRTIO_BALLOON_S_NR);
205 	vb->stats[idx].tag = tag;
206 	vb->stats[idx].val = val;
207 }
208 
209 #define pages_to_bytes(x) ((u64)(x) << PAGE_SHIFT)
210 
211 static void update_balloon_stats(struct virtio_balloon *vb)
212 {
213 	unsigned long events[NR_VM_EVENT_ITEMS];
214 	struct sysinfo i;
215 	int idx = 0;
216 
217 	all_vm_events(events);
218 	si_meminfo(&i);
219 
220 	update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_IN,
221 				pages_to_bytes(events[PSWPIN]));
222 	update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_OUT,
223 				pages_to_bytes(events[PSWPOUT]));
224 	update_stat(vb, idx++, VIRTIO_BALLOON_S_MAJFLT, events[PGMAJFAULT]);
225 	update_stat(vb, idx++, VIRTIO_BALLOON_S_MINFLT, events[PGFAULT]);
226 	update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMFREE,
227 				pages_to_bytes(i.freeram));
228 	update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMTOT,
229 				pages_to_bytes(i.totalram));
230 }
231 
232 /*
233  * While most virtqueues communicate guest-initiated requests to the hypervisor,
234  * the stats queue operates in reverse.  The driver initializes the virtqueue
235  * with a single buffer.  From that point forward, all conversations consist of
236  * a hypervisor request (a call to this function) which directs us to refill
237  * the virtqueue with a fresh stats buffer.  Since stats collection can sleep,
238  * we notify our kthread which does the actual work via stats_handle_request().
239  */
240 static void stats_request(struct virtqueue *vq)
241 {
242 	struct virtio_balloon *vb = vq->vdev->priv;
243 
244 	vb->need_stats_update = 1;
245 	wake_up(&vb->config_change);
246 }
247 
248 static void stats_handle_request(struct virtio_balloon *vb)
249 {
250 	struct virtqueue *vq;
251 	struct scatterlist sg;
252 	unsigned int len;
253 
254 	vb->need_stats_update = 0;
255 	update_balloon_stats(vb);
256 
257 	vq = vb->stats_vq;
258 	if (!virtqueue_get_buf(vq, &len))
259 		return;
260 	sg_init_one(&sg, vb->stats, sizeof(vb->stats));
261 	if (virtqueue_add_outbuf(vq, &sg, 1, vb, GFP_KERNEL) < 0)
262 		BUG();
263 	virtqueue_kick(vq);
264 }
265 
266 static void virtballoon_changed(struct virtio_device *vdev)
267 {
268 	struct virtio_balloon *vb = vdev->priv;
269 
270 	wake_up(&vb->config_change);
271 }
272 
273 static inline s64 towards_target(struct virtio_balloon *vb)
274 {
275 	__le32 v;
276 	s64 target;
277 
278 	virtio_cread(vb->vdev, struct virtio_balloon_config, num_pages, &v);
279 
280 	target = le32_to_cpu(v);
281 	return target - vb->num_pages;
282 }
283 
284 static void update_balloon_size(struct virtio_balloon *vb)
285 {
286 	__le32 actual = cpu_to_le32(vb->num_pages);
287 
288 	virtio_cwrite(vb->vdev, struct virtio_balloon_config, actual,
289 		      &actual);
290 }
291 
292 static int balloon(void *_vballoon)
293 {
294 	struct virtio_balloon *vb = _vballoon;
295 
296 	set_freezable();
297 	while (!kthread_should_stop()) {
298 		s64 diff;
299 
300 		try_to_freeze();
301 		wait_event_interruptible(vb->config_change,
302 					 (diff = towards_target(vb)) != 0
303 					 || vb->need_stats_update
304 					 || kthread_should_stop()
305 					 || freezing(current));
306 		if (vb->need_stats_update)
307 			stats_handle_request(vb);
308 		if (diff > 0)
309 			fill_balloon(vb, diff);
310 		else if (diff < 0)
311 			leak_balloon(vb, -diff);
312 		update_balloon_size(vb);
313 	}
314 	return 0;
315 }
316 
317 static int init_vqs(struct virtio_balloon *vb)
318 {
319 	struct virtqueue *vqs[3];
320 	vq_callback_t *callbacks[] = { balloon_ack, balloon_ack, stats_request };
321 	const char *names[] = { "inflate", "deflate", "stats" };
322 	int err, nvqs;
323 
324 	/*
325 	 * We expect two virtqueues: inflate and deflate, and
326 	 * optionally stat.
327 	 */
328 	nvqs = virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ) ? 3 : 2;
329 	err = vb->vdev->config->find_vqs(vb->vdev, nvqs, vqs, callbacks, names);
330 	if (err)
331 		return err;
332 
333 	vb->inflate_vq = vqs[0];
334 	vb->deflate_vq = vqs[1];
335 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ)) {
336 		struct scatterlist sg;
337 		vb->stats_vq = vqs[2];
338 
339 		/*
340 		 * Prime this virtqueue with one buffer so the hypervisor can
341 		 * use it to signal us later.
342 		 */
343 		sg_init_one(&sg, vb->stats, sizeof vb->stats);
344 		if (virtqueue_add_outbuf(vb->stats_vq, &sg, 1, vb, GFP_KERNEL)
345 		    < 0)
346 			BUG();
347 		virtqueue_kick(vb->stats_vq);
348 	}
349 	return 0;
350 }
351 
352 static const struct address_space_operations virtio_balloon_aops;
353 #ifdef CONFIG_BALLOON_COMPACTION
354 /*
355  * virtballoon_migratepage - perform the balloon page migration on behalf of
356  *			     a compation thread.     (called under page lock)
357  * @mapping: the page->mapping which will be assigned to the new migrated page.
358  * @newpage: page that will replace the isolated page after migration finishes.
359  * @page   : the isolated (old) page that is about to be migrated to newpage.
360  * @mode   : compaction mode -- not used for balloon page migration.
361  *
362  * After a ballooned page gets isolated by compaction procedures, this is the
363  * function that performs the page migration on behalf of a compaction thread
364  * The page migration for virtio balloon is done in a simple swap fashion which
365  * follows these two macro steps:
366  *  1) insert newpage into vb->pages list and update the host about it;
367  *  2) update the host about the old page removed from vb->pages list;
368  *
369  * This function preforms the balloon page migration task.
370  * Called through balloon_mapping->a_ops->migratepage
371  */
372 int virtballoon_migratepage(struct address_space *mapping,
373 		struct page *newpage, struct page *page, enum migrate_mode mode)
374 {
375 	struct balloon_dev_info *vb_dev_info = balloon_page_device(page);
376 	struct virtio_balloon *vb;
377 	unsigned long flags;
378 
379 	BUG_ON(!vb_dev_info);
380 
381 	vb = vb_dev_info->balloon_device;
382 
383 	/*
384 	 * In order to avoid lock contention while migrating pages concurrently
385 	 * to leak_balloon() or fill_balloon() we just give up the balloon_lock
386 	 * this turn, as it is easier to retry the page migration later.
387 	 * This also prevents fill_balloon() getting stuck into a mutex
388 	 * recursion in the case it ends up triggering memory compaction
389 	 * while it is attempting to inflate the ballon.
390 	 */
391 	if (!mutex_trylock(&vb->balloon_lock))
392 		return -EAGAIN;
393 
394 	/* balloon's page migration 1st step  -- inflate "newpage" */
395 	spin_lock_irqsave(&vb_dev_info->pages_lock, flags);
396 	balloon_page_insert(newpage, mapping, &vb_dev_info->pages);
397 	vb_dev_info->isolated_pages--;
398 	spin_unlock_irqrestore(&vb_dev_info->pages_lock, flags);
399 	vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE;
400 	set_page_pfns(vb->pfns, newpage);
401 	tell_host(vb, vb->inflate_vq);
402 
403 	/*
404 	 * balloon's page migration 2nd step -- deflate "page"
405 	 *
406 	 * It's safe to delete page->lru here because this page is at
407 	 * an isolated migration list, and this step is expected to happen here
408 	 */
409 	balloon_page_delete(page);
410 	vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE;
411 	set_page_pfns(vb->pfns, page);
412 	tell_host(vb, vb->deflate_vq);
413 
414 	mutex_unlock(&vb->balloon_lock);
415 
416 	return MIGRATEPAGE_BALLOON_SUCCESS;
417 }
418 
419 /* define the balloon_mapping->a_ops callback to allow balloon page migration */
420 static const struct address_space_operations virtio_balloon_aops = {
421 			.migratepage = virtballoon_migratepage,
422 };
423 #endif /* CONFIG_BALLOON_COMPACTION */
424 
425 static int virtballoon_probe(struct virtio_device *vdev)
426 {
427 	struct virtio_balloon *vb;
428 	struct address_space *vb_mapping;
429 	struct balloon_dev_info *vb_devinfo;
430 	int err;
431 
432 	vdev->priv = vb = kmalloc(sizeof(*vb), GFP_KERNEL);
433 	if (!vb) {
434 		err = -ENOMEM;
435 		goto out;
436 	}
437 
438 	vb->num_pages = 0;
439 	mutex_init(&vb->balloon_lock);
440 	init_waitqueue_head(&vb->config_change);
441 	init_waitqueue_head(&vb->acked);
442 	vb->vdev = vdev;
443 	vb->need_stats_update = 0;
444 
445 	vb_devinfo = balloon_devinfo_alloc(vb);
446 	if (IS_ERR(vb_devinfo)) {
447 		err = PTR_ERR(vb_devinfo);
448 		goto out_free_vb;
449 	}
450 
451 	vb_mapping = balloon_mapping_alloc(vb_devinfo,
452 					   (balloon_compaction_check()) ?
453 					   &virtio_balloon_aops : NULL);
454 	if (IS_ERR(vb_mapping)) {
455 		/*
456 		 * IS_ERR(vb_mapping) && PTR_ERR(vb_mapping) == -EOPNOTSUPP
457 		 * This means !CONFIG_BALLOON_COMPACTION, otherwise we get off.
458 		 */
459 		err = PTR_ERR(vb_mapping);
460 		if (err != -EOPNOTSUPP)
461 			goto out_free_vb_devinfo;
462 	}
463 
464 	vb->vb_dev_info = vb_devinfo;
465 
466 	err = init_vqs(vb);
467 	if (err)
468 		goto out_free_vb_mapping;
469 
470 	vb->thread = kthread_run(balloon, vb, "vballoon");
471 	if (IS_ERR(vb->thread)) {
472 		err = PTR_ERR(vb->thread);
473 		goto out_del_vqs;
474 	}
475 
476 	return 0;
477 
478 out_del_vqs:
479 	vdev->config->del_vqs(vdev);
480 out_free_vb_mapping:
481 	balloon_mapping_free(vb_mapping);
482 out_free_vb_devinfo:
483 	balloon_devinfo_free(vb_devinfo);
484 out_free_vb:
485 	kfree(vb);
486 out:
487 	return err;
488 }
489 
490 static void remove_common(struct virtio_balloon *vb)
491 {
492 	/* There might be pages left in the balloon: free them. */
493 	while (vb->num_pages)
494 		leak_balloon(vb, vb->num_pages);
495 	update_balloon_size(vb);
496 
497 	/* Now we reset the device so we can clean up the queues. */
498 	vb->vdev->config->reset(vb->vdev);
499 
500 	vb->vdev->config->del_vqs(vb->vdev);
501 }
502 
503 static void virtballoon_remove(struct virtio_device *vdev)
504 {
505 	struct virtio_balloon *vb = vdev->priv;
506 
507 	kthread_stop(vb->thread);
508 	remove_common(vb);
509 	balloon_mapping_free(vb->vb_dev_info->mapping);
510 	balloon_devinfo_free(vb->vb_dev_info);
511 	kfree(vb);
512 }
513 
514 #ifdef CONFIG_PM_SLEEP
515 static int virtballoon_freeze(struct virtio_device *vdev)
516 {
517 	struct virtio_balloon *vb = vdev->priv;
518 
519 	/*
520 	 * The kthread is already frozen by the PM core before this
521 	 * function is called.
522 	 */
523 
524 	remove_common(vb);
525 	return 0;
526 }
527 
528 static int virtballoon_restore(struct virtio_device *vdev)
529 {
530 	struct virtio_balloon *vb = vdev->priv;
531 	int ret;
532 
533 	ret = init_vqs(vdev->priv);
534 	if (ret)
535 		return ret;
536 
537 	fill_balloon(vb, towards_target(vb));
538 	update_balloon_size(vb);
539 	return 0;
540 }
541 #endif
542 
543 static unsigned int features[] = {
544 	VIRTIO_BALLOON_F_MUST_TELL_HOST,
545 	VIRTIO_BALLOON_F_STATS_VQ,
546 };
547 
548 static struct virtio_driver virtio_balloon_driver = {
549 	.feature_table = features,
550 	.feature_table_size = ARRAY_SIZE(features),
551 	.driver.name =	KBUILD_MODNAME,
552 	.driver.owner =	THIS_MODULE,
553 	.id_table =	id_table,
554 	.probe =	virtballoon_probe,
555 	.remove =	virtballoon_remove,
556 	.config_changed = virtballoon_changed,
557 #ifdef CONFIG_PM_SLEEP
558 	.freeze	=	virtballoon_freeze,
559 	.restore =	virtballoon_restore,
560 #endif
561 };
562 
563 module_virtio_driver(virtio_balloon_driver);
564 MODULE_DEVICE_TABLE(virtio, id_table);
565 MODULE_DESCRIPTION("Virtio balloon driver");
566 MODULE_LICENSE("GPL");
567