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/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>
32 #include <linux/mm.h>
33 #include <linux/mount.h>
34 #include <linux/magic.h>
35 
36 /*
37  * Balloon device works in 4K page units.  So each page is pointed to by
38  * multiple balloon pages.  All memory counters in this driver are in balloon
39  * page units.
40  */
41 #define VIRTIO_BALLOON_PAGES_PER_PAGE (unsigned)(PAGE_SIZE >> VIRTIO_BALLOON_PFN_SHIFT)
42 #define VIRTIO_BALLOON_ARRAY_PFNS_MAX 256
43 #define OOM_VBALLOON_DEFAULT_PAGES 256
44 #define VIRTBALLOON_OOM_NOTIFY_PRIORITY 80
45 
46 static int oom_pages = OOM_VBALLOON_DEFAULT_PAGES;
47 module_param(oom_pages, int, S_IRUSR | S_IWUSR);
48 MODULE_PARM_DESC(oom_pages, "pages to free on OOM");
49 
50 #ifdef CONFIG_BALLOON_COMPACTION
51 static struct vfsmount *balloon_mnt;
52 #endif
53 
54 struct virtio_balloon {
55 	struct virtio_device *vdev;
56 	struct virtqueue *inflate_vq, *deflate_vq, *stats_vq;
57 
58 	/* The balloon servicing is delegated to a freezable workqueue. */
59 	struct work_struct update_balloon_stats_work;
60 	struct work_struct update_balloon_size_work;
61 
62 	/* Prevent updating balloon when it is being canceled. */
63 	spinlock_t stop_update_lock;
64 	bool stop_update;
65 
66 	/* Waiting for host to ack the pages we released. */
67 	wait_queue_head_t acked;
68 
69 	/* Number of balloon pages we've told the Host we're not using. */
70 	unsigned int num_pages;
71 	/*
72 	 * The pages we've told the Host we're not using are enqueued
73 	 * at vb_dev_info->pages list.
74 	 * Each page on this list adds VIRTIO_BALLOON_PAGES_PER_PAGE
75 	 * to num_pages above.
76 	 */
77 	struct balloon_dev_info vb_dev_info;
78 
79 	/* Synchronize access/update to this struct virtio_balloon elements */
80 	struct mutex balloon_lock;
81 
82 	/* The array of pfns we tell the Host about. */
83 	unsigned int num_pfns;
84 	__virtio32 pfns[VIRTIO_BALLOON_ARRAY_PFNS_MAX];
85 
86 	/* Memory statistics */
87 	struct virtio_balloon_stat stats[VIRTIO_BALLOON_S_NR];
88 
89 	/* To register callback in oom notifier call chain */
90 	struct notifier_block nb;
91 };
92 
93 static struct virtio_device_id id_table[] = {
94 	{ VIRTIO_ID_BALLOON, VIRTIO_DEV_ANY_ID },
95 	{ 0 },
96 };
97 
98 static u32 page_to_balloon_pfn(struct page *page)
99 {
100 	unsigned long pfn = page_to_pfn(page);
101 
102 	BUILD_BUG_ON(PAGE_SHIFT < VIRTIO_BALLOON_PFN_SHIFT);
103 	/* Convert pfn from Linux page size to balloon page size. */
104 	return pfn * VIRTIO_BALLOON_PAGES_PER_PAGE;
105 }
106 
107 static void balloon_ack(struct virtqueue *vq)
108 {
109 	struct virtio_balloon *vb = vq->vdev->priv;
110 
111 	wake_up(&vb->acked);
112 }
113 
114 static void tell_host(struct virtio_balloon *vb, struct virtqueue *vq)
115 {
116 	struct scatterlist sg;
117 	unsigned int len;
118 
119 	sg_init_one(&sg, vb->pfns, sizeof(vb->pfns[0]) * vb->num_pfns);
120 
121 	/* We should always be able to add one buffer to an empty queue. */
122 	virtqueue_add_outbuf(vq, &sg, 1, vb, GFP_KERNEL);
123 	virtqueue_kick(vq);
124 
125 	/* When host has read buffer, this completes via balloon_ack */
126 	wait_event(vb->acked, virtqueue_get_buf(vq, &len));
127 
128 }
129 
130 static void set_page_pfns(struct virtio_balloon *vb,
131 			  __virtio32 pfns[], struct page *page)
132 {
133 	unsigned int i;
134 
135 	/*
136 	 * Set balloon pfns pointing at this page.
137 	 * Note that the first pfn points at start of the page.
138 	 */
139 	for (i = 0; i < VIRTIO_BALLOON_PAGES_PER_PAGE; i++)
140 		pfns[i] = cpu_to_virtio32(vb->vdev,
141 					  page_to_balloon_pfn(page) + i);
142 }
143 
144 static unsigned fill_balloon(struct virtio_balloon *vb, size_t num)
145 {
146 	unsigned num_allocated_pages;
147 	unsigned num_pfns;
148 	struct page *page;
149 	LIST_HEAD(pages);
150 
151 	/* We can only do one array worth at a time. */
152 	num = min(num, ARRAY_SIZE(vb->pfns));
153 
154 	for (num_pfns = 0; num_pfns < num;
155 	     num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) {
156 		struct page *page = balloon_page_alloc();
157 
158 		if (!page) {
159 			dev_info_ratelimited(&vb->vdev->dev,
160 					     "Out of puff! Can't get %u pages\n",
161 					     VIRTIO_BALLOON_PAGES_PER_PAGE);
162 			/* Sleep for at least 1/5 of a second before retry. */
163 			msleep(200);
164 			break;
165 		}
166 
167 		balloon_page_push(&pages, page);
168 	}
169 
170 	mutex_lock(&vb->balloon_lock);
171 
172 	vb->num_pfns = 0;
173 
174 	while ((page = balloon_page_pop(&pages))) {
175 		balloon_page_enqueue(&vb->vb_dev_info, page);
176 
177 		set_page_pfns(vb, vb->pfns + vb->num_pfns, page);
178 		vb->num_pages += VIRTIO_BALLOON_PAGES_PER_PAGE;
179 		if (!virtio_has_feature(vb->vdev,
180 					VIRTIO_BALLOON_F_DEFLATE_ON_OOM))
181 			adjust_managed_page_count(page, -1);
182 		vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE;
183 	}
184 
185 	num_allocated_pages = vb->num_pfns;
186 	/* Did we get any? */
187 	if (vb->num_pfns != 0)
188 		tell_host(vb, vb->inflate_vq);
189 	mutex_unlock(&vb->balloon_lock);
190 
191 	return num_allocated_pages;
192 }
193 
194 static void release_pages_balloon(struct virtio_balloon *vb,
195 				 struct list_head *pages)
196 {
197 	struct page *page, *next;
198 
199 	list_for_each_entry_safe(page, next, pages, lru) {
200 		if (!virtio_has_feature(vb->vdev,
201 					VIRTIO_BALLOON_F_DEFLATE_ON_OOM))
202 			adjust_managed_page_count(page, 1);
203 		list_del(&page->lru);
204 		put_page(page); /* balloon reference */
205 	}
206 }
207 
208 static unsigned leak_balloon(struct virtio_balloon *vb, size_t num)
209 {
210 	unsigned num_freed_pages;
211 	struct page *page;
212 	struct balloon_dev_info *vb_dev_info = &vb->vb_dev_info;
213 	LIST_HEAD(pages);
214 
215 	/* We can only do one array worth at a time. */
216 	num = min(num, ARRAY_SIZE(vb->pfns));
217 
218 	mutex_lock(&vb->balloon_lock);
219 	/* We can't release more pages than taken */
220 	num = min(num, (size_t)vb->num_pages);
221 	for (vb->num_pfns = 0; vb->num_pfns < num;
222 	     vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) {
223 		page = balloon_page_dequeue(vb_dev_info);
224 		if (!page)
225 			break;
226 		set_page_pfns(vb, vb->pfns + vb->num_pfns, page);
227 		list_add(&page->lru, &pages);
228 		vb->num_pages -= VIRTIO_BALLOON_PAGES_PER_PAGE;
229 	}
230 
231 	num_freed_pages = vb->num_pfns;
232 	/*
233 	 * Note that if
234 	 * virtio_has_feature(vdev, VIRTIO_BALLOON_F_MUST_TELL_HOST);
235 	 * is true, we *have* to do it in this order
236 	 */
237 	if (vb->num_pfns != 0)
238 		tell_host(vb, vb->deflate_vq);
239 	release_pages_balloon(vb, &pages);
240 	mutex_unlock(&vb->balloon_lock);
241 	return num_freed_pages;
242 }
243 
244 static inline void update_stat(struct virtio_balloon *vb, int idx,
245 			       u16 tag, u64 val)
246 {
247 	BUG_ON(idx >= VIRTIO_BALLOON_S_NR);
248 	vb->stats[idx].tag = cpu_to_virtio16(vb->vdev, tag);
249 	vb->stats[idx].val = cpu_to_virtio64(vb->vdev, val);
250 }
251 
252 #define pages_to_bytes(x) ((u64)(x) << PAGE_SHIFT)
253 
254 static unsigned int update_balloon_stats(struct virtio_balloon *vb)
255 {
256 	unsigned long events[NR_VM_EVENT_ITEMS];
257 	struct sysinfo i;
258 	unsigned int idx = 0;
259 	long available;
260 	unsigned long caches;
261 
262 	all_vm_events(events);
263 	si_meminfo(&i);
264 
265 	available = si_mem_available();
266 	caches = global_node_page_state(NR_FILE_PAGES);
267 
268 #ifdef CONFIG_VM_EVENT_COUNTERS
269 	update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_IN,
270 				pages_to_bytes(events[PSWPIN]));
271 	update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_OUT,
272 				pages_to_bytes(events[PSWPOUT]));
273 	update_stat(vb, idx++, VIRTIO_BALLOON_S_MAJFLT, events[PGMAJFAULT]);
274 	update_stat(vb, idx++, VIRTIO_BALLOON_S_MINFLT, events[PGFAULT]);
275 #endif
276 	update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMFREE,
277 				pages_to_bytes(i.freeram));
278 	update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMTOT,
279 				pages_to_bytes(i.totalram));
280 	update_stat(vb, idx++, VIRTIO_BALLOON_S_AVAIL,
281 				pages_to_bytes(available));
282 	update_stat(vb, idx++, VIRTIO_BALLOON_S_CACHES,
283 				pages_to_bytes(caches));
284 
285 	return idx;
286 }
287 
288 /*
289  * While most virtqueues communicate guest-initiated requests to the hypervisor,
290  * the stats queue operates in reverse.  The driver initializes the virtqueue
291  * with a single buffer.  From that point forward, all conversations consist of
292  * a hypervisor request (a call to this function) which directs us to refill
293  * the virtqueue with a fresh stats buffer.  Since stats collection can sleep,
294  * we delegate the job to a freezable workqueue that will do the actual work via
295  * stats_handle_request().
296  */
297 static void stats_request(struct virtqueue *vq)
298 {
299 	struct virtio_balloon *vb = vq->vdev->priv;
300 
301 	spin_lock(&vb->stop_update_lock);
302 	if (!vb->stop_update)
303 		queue_work(system_freezable_wq, &vb->update_balloon_stats_work);
304 	spin_unlock(&vb->stop_update_lock);
305 }
306 
307 static void stats_handle_request(struct virtio_balloon *vb)
308 {
309 	struct virtqueue *vq;
310 	struct scatterlist sg;
311 	unsigned int len, num_stats;
312 
313 	num_stats = update_balloon_stats(vb);
314 
315 	vq = vb->stats_vq;
316 	if (!virtqueue_get_buf(vq, &len))
317 		return;
318 	sg_init_one(&sg, vb->stats, sizeof(vb->stats[0]) * num_stats);
319 	virtqueue_add_outbuf(vq, &sg, 1, vb, GFP_KERNEL);
320 	virtqueue_kick(vq);
321 }
322 
323 static void virtballoon_changed(struct virtio_device *vdev)
324 {
325 	struct virtio_balloon *vb = vdev->priv;
326 	unsigned long flags;
327 
328 	spin_lock_irqsave(&vb->stop_update_lock, flags);
329 	if (!vb->stop_update)
330 		queue_work(system_freezable_wq, &vb->update_balloon_size_work);
331 	spin_unlock_irqrestore(&vb->stop_update_lock, flags);
332 }
333 
334 static inline s64 towards_target(struct virtio_balloon *vb)
335 {
336 	s64 target;
337 	u32 num_pages;
338 
339 	virtio_cread(vb->vdev, struct virtio_balloon_config, num_pages,
340 		     &num_pages);
341 
342 	/* Legacy balloon config space is LE, unlike all other devices. */
343 	if (!virtio_has_feature(vb->vdev, VIRTIO_F_VERSION_1))
344 		num_pages = le32_to_cpu((__force __le32)num_pages);
345 
346 	target = num_pages;
347 	return target - vb->num_pages;
348 }
349 
350 static void update_balloon_size(struct virtio_balloon *vb)
351 {
352 	u32 actual = vb->num_pages;
353 
354 	/* Legacy balloon config space is LE, unlike all other devices. */
355 	if (!virtio_has_feature(vb->vdev, VIRTIO_F_VERSION_1))
356 		actual = (__force u32)cpu_to_le32(actual);
357 
358 	virtio_cwrite(vb->vdev, struct virtio_balloon_config, actual,
359 		      &actual);
360 }
361 
362 /*
363  * virtballoon_oom_notify - release pages when system is under severe
364  *			    memory pressure (called from out_of_memory())
365  * @self : notifier block struct
366  * @dummy: not used
367  * @parm : returned - number of freed pages
368  *
369  * The balancing of memory by use of the virtio balloon should not cause
370  * the termination of processes while there are pages in the balloon.
371  * If virtio balloon manages to release some memory, it will make the
372  * system return and retry the allocation that forced the OOM killer
373  * to run.
374  */
375 static int virtballoon_oom_notify(struct notifier_block *self,
376 				  unsigned long dummy, void *parm)
377 {
378 	struct virtio_balloon *vb;
379 	unsigned long *freed;
380 	unsigned num_freed_pages;
381 
382 	vb = container_of(self, struct virtio_balloon, nb);
383 	if (!virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_DEFLATE_ON_OOM))
384 		return NOTIFY_OK;
385 
386 	freed = parm;
387 	num_freed_pages = leak_balloon(vb, oom_pages);
388 	update_balloon_size(vb);
389 	*freed += num_freed_pages;
390 
391 	return NOTIFY_OK;
392 }
393 
394 static void update_balloon_stats_func(struct work_struct *work)
395 {
396 	struct virtio_balloon *vb;
397 
398 	vb = container_of(work, struct virtio_balloon,
399 			  update_balloon_stats_work);
400 	stats_handle_request(vb);
401 }
402 
403 static void update_balloon_size_func(struct work_struct *work)
404 {
405 	struct virtio_balloon *vb;
406 	s64 diff;
407 
408 	vb = container_of(work, struct virtio_balloon,
409 			  update_balloon_size_work);
410 	diff = towards_target(vb);
411 
412 	if (diff > 0)
413 		diff -= fill_balloon(vb, diff);
414 	else if (diff < 0)
415 		diff += leak_balloon(vb, -diff);
416 	update_balloon_size(vb);
417 
418 	if (diff)
419 		queue_work(system_freezable_wq, work);
420 }
421 
422 static int init_vqs(struct virtio_balloon *vb)
423 {
424 	struct virtqueue *vqs[3];
425 	vq_callback_t *callbacks[] = { balloon_ack, balloon_ack, stats_request };
426 	static const char * const names[] = { "inflate", "deflate", "stats" };
427 	int err, nvqs;
428 
429 	/*
430 	 * We expect two virtqueues: inflate and deflate, and
431 	 * optionally stat.
432 	 */
433 	nvqs = virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ) ? 3 : 2;
434 	err = virtio_find_vqs(vb->vdev, nvqs, vqs, callbacks, names, NULL);
435 	if (err)
436 		return err;
437 
438 	vb->inflate_vq = vqs[0];
439 	vb->deflate_vq = vqs[1];
440 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ)) {
441 		struct scatterlist sg;
442 		unsigned int num_stats;
443 		vb->stats_vq = vqs[2];
444 
445 		/*
446 		 * Prime this virtqueue with one buffer so the hypervisor can
447 		 * use it to signal us later (it can't be broken yet!).
448 		 */
449 		num_stats = update_balloon_stats(vb);
450 
451 		sg_init_one(&sg, vb->stats, sizeof(vb->stats[0]) * num_stats);
452 		if (virtqueue_add_outbuf(vb->stats_vq, &sg, 1, vb, GFP_KERNEL)
453 		    < 0)
454 			BUG();
455 		virtqueue_kick(vb->stats_vq);
456 	}
457 	return 0;
458 }
459 
460 #ifdef CONFIG_BALLOON_COMPACTION
461 /*
462  * virtballoon_migratepage - perform the balloon page migration on behalf of
463  *			     a compation thread.     (called under page lock)
464  * @vb_dev_info: the balloon device
465  * @newpage: page that will replace the isolated page after migration finishes.
466  * @page   : the isolated (old) page that is about to be migrated to newpage.
467  * @mode   : compaction mode -- not used for balloon page migration.
468  *
469  * After a ballooned page gets isolated by compaction procedures, this is the
470  * function that performs the page migration on behalf of a compaction thread
471  * The page migration for virtio balloon is done in a simple swap fashion which
472  * follows these two macro steps:
473  *  1) insert newpage into vb->pages list and update the host about it;
474  *  2) update the host about the old page removed from vb->pages list;
475  *
476  * This function preforms the balloon page migration task.
477  * Called through balloon_mapping->a_ops->migratepage
478  */
479 static int virtballoon_migratepage(struct balloon_dev_info *vb_dev_info,
480 		struct page *newpage, struct page *page, enum migrate_mode mode)
481 {
482 	struct virtio_balloon *vb = container_of(vb_dev_info,
483 			struct virtio_balloon, vb_dev_info);
484 	unsigned long flags;
485 
486 	/*
487 	 * In order to avoid lock contention while migrating pages concurrently
488 	 * to leak_balloon() or fill_balloon() we just give up the balloon_lock
489 	 * this turn, as it is easier to retry the page migration later.
490 	 * This also prevents fill_balloon() getting stuck into a mutex
491 	 * recursion in the case it ends up triggering memory compaction
492 	 * while it is attempting to inflate the ballon.
493 	 */
494 	if (!mutex_trylock(&vb->balloon_lock))
495 		return -EAGAIN;
496 
497 	get_page(newpage); /* balloon reference */
498 
499 	/* balloon's page migration 1st step  -- inflate "newpage" */
500 	spin_lock_irqsave(&vb_dev_info->pages_lock, flags);
501 	balloon_page_insert(vb_dev_info, newpage);
502 	vb_dev_info->isolated_pages--;
503 	__count_vm_event(BALLOON_MIGRATE);
504 	spin_unlock_irqrestore(&vb_dev_info->pages_lock, flags);
505 	vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE;
506 	set_page_pfns(vb, vb->pfns, newpage);
507 	tell_host(vb, vb->inflate_vq);
508 
509 	/* balloon's page migration 2nd step -- deflate "page" */
510 	balloon_page_delete(page);
511 	vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE;
512 	set_page_pfns(vb, vb->pfns, page);
513 	tell_host(vb, vb->deflate_vq);
514 
515 	mutex_unlock(&vb->balloon_lock);
516 
517 	put_page(page); /* balloon reference */
518 
519 	return MIGRATEPAGE_SUCCESS;
520 }
521 
522 static struct dentry *balloon_mount(struct file_system_type *fs_type,
523 		int flags, const char *dev_name, void *data)
524 {
525 	static const struct dentry_operations ops = {
526 		.d_dname = simple_dname,
527 	};
528 
529 	return mount_pseudo(fs_type, "balloon-kvm:", NULL, &ops,
530 				BALLOON_KVM_MAGIC);
531 }
532 
533 static struct file_system_type balloon_fs = {
534 	.name           = "balloon-kvm",
535 	.mount          = balloon_mount,
536 	.kill_sb        = kill_anon_super,
537 };
538 
539 #endif /* CONFIG_BALLOON_COMPACTION */
540 
541 static int virtballoon_probe(struct virtio_device *vdev)
542 {
543 	struct virtio_balloon *vb;
544 	int err;
545 
546 	if (!vdev->config->get) {
547 		dev_err(&vdev->dev, "%s failure: config access disabled\n",
548 			__func__);
549 		return -EINVAL;
550 	}
551 
552 	vdev->priv = vb = kmalloc(sizeof(*vb), GFP_KERNEL);
553 	if (!vb) {
554 		err = -ENOMEM;
555 		goto out;
556 	}
557 
558 	INIT_WORK(&vb->update_balloon_stats_work, update_balloon_stats_func);
559 	INIT_WORK(&vb->update_balloon_size_work, update_balloon_size_func);
560 	spin_lock_init(&vb->stop_update_lock);
561 	vb->stop_update = false;
562 	vb->num_pages = 0;
563 	mutex_init(&vb->balloon_lock);
564 	init_waitqueue_head(&vb->acked);
565 	vb->vdev = vdev;
566 
567 	balloon_devinfo_init(&vb->vb_dev_info);
568 
569 	err = init_vqs(vb);
570 	if (err)
571 		goto out_free_vb;
572 
573 	vb->nb.notifier_call = virtballoon_oom_notify;
574 	vb->nb.priority = VIRTBALLOON_OOM_NOTIFY_PRIORITY;
575 	err = register_oom_notifier(&vb->nb);
576 	if (err < 0)
577 		goto out_del_vqs;
578 
579 #ifdef CONFIG_BALLOON_COMPACTION
580 	balloon_mnt = kern_mount(&balloon_fs);
581 	if (IS_ERR(balloon_mnt)) {
582 		err = PTR_ERR(balloon_mnt);
583 		unregister_oom_notifier(&vb->nb);
584 		goto out_del_vqs;
585 	}
586 
587 	vb->vb_dev_info.migratepage = virtballoon_migratepage;
588 	vb->vb_dev_info.inode = alloc_anon_inode(balloon_mnt->mnt_sb);
589 	if (IS_ERR(vb->vb_dev_info.inode)) {
590 		err = PTR_ERR(vb->vb_dev_info.inode);
591 		kern_unmount(balloon_mnt);
592 		unregister_oom_notifier(&vb->nb);
593 		vb->vb_dev_info.inode = NULL;
594 		goto out_del_vqs;
595 	}
596 	vb->vb_dev_info.inode->i_mapping->a_ops = &balloon_aops;
597 #endif
598 
599 	virtio_device_ready(vdev);
600 
601 	if (towards_target(vb))
602 		virtballoon_changed(vdev);
603 	return 0;
604 
605 out_del_vqs:
606 	vdev->config->del_vqs(vdev);
607 out_free_vb:
608 	kfree(vb);
609 out:
610 	return err;
611 }
612 
613 static void remove_common(struct virtio_balloon *vb)
614 {
615 	/* There might be pages left in the balloon: free them. */
616 	while (vb->num_pages)
617 		leak_balloon(vb, vb->num_pages);
618 	update_balloon_size(vb);
619 
620 	/* Now we reset the device so we can clean up the queues. */
621 	vb->vdev->config->reset(vb->vdev);
622 
623 	vb->vdev->config->del_vqs(vb->vdev);
624 }
625 
626 static void virtballoon_remove(struct virtio_device *vdev)
627 {
628 	struct virtio_balloon *vb = vdev->priv;
629 
630 	unregister_oom_notifier(&vb->nb);
631 
632 	spin_lock_irq(&vb->stop_update_lock);
633 	vb->stop_update = true;
634 	spin_unlock_irq(&vb->stop_update_lock);
635 	cancel_work_sync(&vb->update_balloon_size_work);
636 	cancel_work_sync(&vb->update_balloon_stats_work);
637 
638 	remove_common(vb);
639 #ifdef CONFIG_BALLOON_COMPACTION
640 	if (vb->vb_dev_info.inode)
641 		iput(vb->vb_dev_info.inode);
642 
643 	kern_unmount(balloon_mnt);
644 #endif
645 	kfree(vb);
646 }
647 
648 #ifdef CONFIG_PM_SLEEP
649 static int virtballoon_freeze(struct virtio_device *vdev)
650 {
651 	struct virtio_balloon *vb = vdev->priv;
652 
653 	/*
654 	 * The workqueue is already frozen by the PM core before this
655 	 * function is called.
656 	 */
657 	remove_common(vb);
658 	return 0;
659 }
660 
661 static int virtballoon_restore(struct virtio_device *vdev)
662 {
663 	struct virtio_balloon *vb = vdev->priv;
664 	int ret;
665 
666 	ret = init_vqs(vdev->priv);
667 	if (ret)
668 		return ret;
669 
670 	virtio_device_ready(vdev);
671 
672 	if (towards_target(vb))
673 		virtballoon_changed(vdev);
674 	update_balloon_size(vb);
675 	return 0;
676 }
677 #endif
678 
679 static int virtballoon_validate(struct virtio_device *vdev)
680 {
681 	__virtio_clear_bit(vdev, VIRTIO_F_IOMMU_PLATFORM);
682 	return 0;
683 }
684 
685 static unsigned int features[] = {
686 	VIRTIO_BALLOON_F_MUST_TELL_HOST,
687 	VIRTIO_BALLOON_F_STATS_VQ,
688 	VIRTIO_BALLOON_F_DEFLATE_ON_OOM,
689 };
690 
691 static struct virtio_driver virtio_balloon_driver = {
692 	.feature_table = features,
693 	.feature_table_size = ARRAY_SIZE(features),
694 	.driver.name =	KBUILD_MODNAME,
695 	.driver.owner =	THIS_MODULE,
696 	.id_table =	id_table,
697 	.validate =	virtballoon_validate,
698 	.probe =	virtballoon_probe,
699 	.remove =	virtballoon_remove,
700 	.config_changed = virtballoon_changed,
701 #ifdef CONFIG_PM_SLEEP
702 	.freeze	=	virtballoon_freeze,
703 	.restore =	virtballoon_restore,
704 #endif
705 };
706 
707 module_virtio_driver(virtio_balloon_driver);
708 MODULE_DEVICE_TABLE(virtio, id_table);
709 MODULE_DESCRIPTION("Virtio balloon driver");
710 MODULE_LICENSE("GPL");
711