1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Virtio balloon implementation, inspired by Dor Laor and Marcelo
4  * Tosatti's implementations.
5  *
6  *  Copyright 2008 Rusty Russell IBM Corporation
7  */
8 
9 #include <linux/virtio.h>
10 #include <linux/virtio_balloon.h>
11 #include <linux/swap.h>
12 #include <linux/workqueue.h>
13 #include <linux/delay.h>
14 #include <linux/slab.h>
15 #include <linux/module.h>
16 #include <linux/balloon_compaction.h>
17 #include <linux/wait.h>
18 #include <linux/mm.h>
19 #include <linux/mount.h>
20 #include <linux/magic.h>
21 #include <linux/pseudo_fs.h>
22 
23 /*
24  * Balloon device works in 4K page units.  So each page is pointed to by
25  * multiple balloon pages.  All memory counters in this driver are in balloon
26  * page units.
27  */
28 #define VIRTIO_BALLOON_PAGES_PER_PAGE (unsigned)(PAGE_SIZE >> VIRTIO_BALLOON_PFN_SHIFT)
29 #define VIRTIO_BALLOON_ARRAY_PFNS_MAX 256
30 #define VIRTBALLOON_OOM_NOTIFY_PRIORITY 80
31 
32 #define VIRTIO_BALLOON_FREE_PAGE_ALLOC_FLAG (__GFP_NORETRY | __GFP_NOWARN | \
33 					     __GFP_NOMEMALLOC)
34 /* The order of free page blocks to report to host */
35 #define VIRTIO_BALLOON_FREE_PAGE_ORDER (MAX_ORDER - 1)
36 /* The size of a free page block in bytes */
37 #define VIRTIO_BALLOON_FREE_PAGE_SIZE \
38 	(1 << (VIRTIO_BALLOON_FREE_PAGE_ORDER + PAGE_SHIFT))
39 
40 #ifdef CONFIG_BALLOON_COMPACTION
41 static struct vfsmount *balloon_mnt;
42 #endif
43 
44 enum virtio_balloon_vq {
45 	VIRTIO_BALLOON_VQ_INFLATE,
46 	VIRTIO_BALLOON_VQ_DEFLATE,
47 	VIRTIO_BALLOON_VQ_STATS,
48 	VIRTIO_BALLOON_VQ_FREE_PAGE,
49 	VIRTIO_BALLOON_VQ_MAX
50 };
51 
52 enum virtio_balloon_config_read {
53 	VIRTIO_BALLOON_CONFIG_READ_CMD_ID = 0,
54 };
55 
56 struct virtio_balloon {
57 	struct virtio_device *vdev;
58 	struct virtqueue *inflate_vq, *deflate_vq, *stats_vq, *free_page_vq;
59 
60 	/* Balloon's own wq for cpu-intensive work items */
61 	struct workqueue_struct *balloon_wq;
62 	/* The free page reporting work item submitted to the balloon wq */
63 	struct work_struct report_free_page_work;
64 
65 	/* The balloon servicing is delegated to a freezable workqueue. */
66 	struct work_struct update_balloon_stats_work;
67 	struct work_struct update_balloon_size_work;
68 
69 	/* Prevent updating balloon when it is being canceled. */
70 	spinlock_t stop_update_lock;
71 	bool stop_update;
72 	/* Bitmap to indicate if reading the related config fields are needed */
73 	unsigned long config_read_bitmap;
74 
75 	/* The list of allocated free pages, waiting to be given back to mm */
76 	struct list_head free_page_list;
77 	spinlock_t free_page_list_lock;
78 	/* The number of free page blocks on the above list */
79 	unsigned long num_free_page_blocks;
80 	/*
81 	 * The cmd id received from host.
82 	 * Read it via virtio_balloon_cmd_id_received to get the latest value
83 	 * sent from host.
84 	 */
85 	u32 cmd_id_received_cache;
86 	/* The cmd id that is actively in use */
87 	__virtio32 cmd_id_active;
88 	/* Buffer to store the stop sign */
89 	__virtio32 cmd_id_stop;
90 
91 	/* Waiting for host to ack the pages we released. */
92 	wait_queue_head_t acked;
93 
94 	/* Number of balloon pages we've told the Host we're not using. */
95 	unsigned int num_pages;
96 	/*
97 	 * The pages we've told the Host we're not using are enqueued
98 	 * at vb_dev_info->pages list.
99 	 * Each page on this list adds VIRTIO_BALLOON_PAGES_PER_PAGE
100 	 * to num_pages above.
101 	 */
102 	struct balloon_dev_info vb_dev_info;
103 
104 	/* Synchronize access/update to this struct virtio_balloon elements */
105 	struct mutex balloon_lock;
106 
107 	/* The array of pfns we tell the Host about. */
108 	unsigned int num_pfns;
109 	__virtio32 pfns[VIRTIO_BALLOON_ARRAY_PFNS_MAX];
110 
111 	/* Memory statistics */
112 	struct virtio_balloon_stat stats[VIRTIO_BALLOON_S_NR];
113 
114 	/* To register a shrinker to shrink memory upon memory pressure */
115 	struct shrinker shrinker;
116 };
117 
118 static struct virtio_device_id id_table[] = {
119 	{ VIRTIO_ID_BALLOON, VIRTIO_DEV_ANY_ID },
120 	{ 0 },
121 };
122 
123 static u32 page_to_balloon_pfn(struct page *page)
124 {
125 	unsigned long pfn = page_to_pfn(page);
126 
127 	BUILD_BUG_ON(PAGE_SHIFT < VIRTIO_BALLOON_PFN_SHIFT);
128 	/* Convert pfn from Linux page size to balloon page size. */
129 	return pfn * VIRTIO_BALLOON_PAGES_PER_PAGE;
130 }
131 
132 static void balloon_ack(struct virtqueue *vq)
133 {
134 	struct virtio_balloon *vb = vq->vdev->priv;
135 
136 	wake_up(&vb->acked);
137 }
138 
139 static void tell_host(struct virtio_balloon *vb, struct virtqueue *vq)
140 {
141 	struct scatterlist sg;
142 	unsigned int len;
143 
144 	sg_init_one(&sg, vb->pfns, sizeof(vb->pfns[0]) * vb->num_pfns);
145 
146 	/* We should always be able to add one buffer to an empty queue. */
147 	virtqueue_add_outbuf(vq, &sg, 1, vb, GFP_KERNEL);
148 	virtqueue_kick(vq);
149 
150 	/* When host has read buffer, this completes via balloon_ack */
151 	wait_event(vb->acked, virtqueue_get_buf(vq, &len));
152 
153 }
154 
155 static void set_page_pfns(struct virtio_balloon *vb,
156 			  __virtio32 pfns[], struct page *page)
157 {
158 	unsigned int i;
159 
160 	/*
161 	 * Set balloon pfns pointing at this page.
162 	 * Note that the first pfn points at start of the page.
163 	 */
164 	for (i = 0; i < VIRTIO_BALLOON_PAGES_PER_PAGE; i++)
165 		pfns[i] = cpu_to_virtio32(vb->vdev,
166 					  page_to_balloon_pfn(page) + i);
167 }
168 
169 static unsigned fill_balloon(struct virtio_balloon *vb, size_t num)
170 {
171 	unsigned num_allocated_pages;
172 	unsigned num_pfns;
173 	struct page *page;
174 	LIST_HEAD(pages);
175 
176 	/* We can only do one array worth at a time. */
177 	num = min(num, ARRAY_SIZE(vb->pfns));
178 
179 	for (num_pfns = 0; num_pfns < num;
180 	     num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) {
181 		struct page *page = balloon_page_alloc();
182 
183 		if (!page) {
184 			dev_info_ratelimited(&vb->vdev->dev,
185 					     "Out of puff! Can't get %u pages\n",
186 					     VIRTIO_BALLOON_PAGES_PER_PAGE);
187 			/* Sleep for at least 1/5 of a second before retry. */
188 			msleep(200);
189 			break;
190 		}
191 
192 		balloon_page_push(&pages, page);
193 	}
194 
195 	mutex_lock(&vb->balloon_lock);
196 
197 	vb->num_pfns = 0;
198 
199 	while ((page = balloon_page_pop(&pages))) {
200 		balloon_page_enqueue(&vb->vb_dev_info, page);
201 
202 		set_page_pfns(vb, vb->pfns + vb->num_pfns, page);
203 		vb->num_pages += VIRTIO_BALLOON_PAGES_PER_PAGE;
204 		if (!virtio_has_feature(vb->vdev,
205 					VIRTIO_BALLOON_F_DEFLATE_ON_OOM))
206 			adjust_managed_page_count(page, -1);
207 		vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE;
208 	}
209 
210 	num_allocated_pages = vb->num_pfns;
211 	/* Did we get any? */
212 	if (vb->num_pfns != 0)
213 		tell_host(vb, vb->inflate_vq);
214 	mutex_unlock(&vb->balloon_lock);
215 
216 	return num_allocated_pages;
217 }
218 
219 static void release_pages_balloon(struct virtio_balloon *vb,
220 				 struct list_head *pages)
221 {
222 	struct page *page, *next;
223 
224 	list_for_each_entry_safe(page, next, pages, lru) {
225 		if (!virtio_has_feature(vb->vdev,
226 					VIRTIO_BALLOON_F_DEFLATE_ON_OOM))
227 			adjust_managed_page_count(page, 1);
228 		list_del(&page->lru);
229 		put_page(page); /* balloon reference */
230 	}
231 }
232 
233 static unsigned leak_balloon(struct virtio_balloon *vb, size_t num)
234 {
235 	unsigned num_freed_pages;
236 	struct page *page;
237 	struct balloon_dev_info *vb_dev_info = &vb->vb_dev_info;
238 	LIST_HEAD(pages);
239 
240 	/* We can only do one array worth at a time. */
241 	num = min(num, ARRAY_SIZE(vb->pfns));
242 
243 	mutex_lock(&vb->balloon_lock);
244 	/* We can't release more pages than taken */
245 	num = min(num, (size_t)vb->num_pages);
246 	for (vb->num_pfns = 0; vb->num_pfns < num;
247 	     vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) {
248 		page = balloon_page_dequeue(vb_dev_info);
249 		if (!page)
250 			break;
251 		set_page_pfns(vb, vb->pfns + vb->num_pfns, page);
252 		list_add(&page->lru, &pages);
253 		vb->num_pages -= VIRTIO_BALLOON_PAGES_PER_PAGE;
254 	}
255 
256 	num_freed_pages = vb->num_pfns;
257 	/*
258 	 * Note that if
259 	 * virtio_has_feature(vdev, VIRTIO_BALLOON_F_MUST_TELL_HOST);
260 	 * is true, we *have* to do it in this order
261 	 */
262 	if (vb->num_pfns != 0)
263 		tell_host(vb, vb->deflate_vq);
264 	release_pages_balloon(vb, &pages);
265 	mutex_unlock(&vb->balloon_lock);
266 	return num_freed_pages;
267 }
268 
269 static inline void update_stat(struct virtio_balloon *vb, int idx,
270 			       u16 tag, u64 val)
271 {
272 	BUG_ON(idx >= VIRTIO_BALLOON_S_NR);
273 	vb->stats[idx].tag = cpu_to_virtio16(vb->vdev, tag);
274 	vb->stats[idx].val = cpu_to_virtio64(vb->vdev, val);
275 }
276 
277 #define pages_to_bytes(x) ((u64)(x) << PAGE_SHIFT)
278 
279 static unsigned int update_balloon_stats(struct virtio_balloon *vb)
280 {
281 	unsigned long events[NR_VM_EVENT_ITEMS];
282 	struct sysinfo i;
283 	unsigned int idx = 0;
284 	long available;
285 	unsigned long caches;
286 
287 	all_vm_events(events);
288 	si_meminfo(&i);
289 
290 	available = si_mem_available();
291 	caches = global_node_page_state(NR_FILE_PAGES);
292 
293 #ifdef CONFIG_VM_EVENT_COUNTERS
294 	update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_IN,
295 				pages_to_bytes(events[PSWPIN]));
296 	update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_OUT,
297 				pages_to_bytes(events[PSWPOUT]));
298 	update_stat(vb, idx++, VIRTIO_BALLOON_S_MAJFLT, events[PGMAJFAULT]);
299 	update_stat(vb, idx++, VIRTIO_BALLOON_S_MINFLT, events[PGFAULT]);
300 #ifdef CONFIG_HUGETLB_PAGE
301 	update_stat(vb, idx++, VIRTIO_BALLOON_S_HTLB_PGALLOC,
302 		    events[HTLB_BUDDY_PGALLOC]);
303 	update_stat(vb, idx++, VIRTIO_BALLOON_S_HTLB_PGFAIL,
304 		    events[HTLB_BUDDY_PGALLOC_FAIL]);
305 #endif
306 #endif
307 	update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMFREE,
308 				pages_to_bytes(i.freeram));
309 	update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMTOT,
310 				pages_to_bytes(i.totalram));
311 	update_stat(vb, idx++, VIRTIO_BALLOON_S_AVAIL,
312 				pages_to_bytes(available));
313 	update_stat(vb, idx++, VIRTIO_BALLOON_S_CACHES,
314 				pages_to_bytes(caches));
315 
316 	return idx;
317 }
318 
319 /*
320  * While most virtqueues communicate guest-initiated requests to the hypervisor,
321  * the stats queue operates in reverse.  The driver initializes the virtqueue
322  * with a single buffer.  From that point forward, all conversations consist of
323  * a hypervisor request (a call to this function) which directs us to refill
324  * the virtqueue with a fresh stats buffer.  Since stats collection can sleep,
325  * we delegate the job to a freezable workqueue that will do the actual work via
326  * stats_handle_request().
327  */
328 static void stats_request(struct virtqueue *vq)
329 {
330 	struct virtio_balloon *vb = vq->vdev->priv;
331 
332 	spin_lock(&vb->stop_update_lock);
333 	if (!vb->stop_update)
334 		queue_work(system_freezable_wq, &vb->update_balloon_stats_work);
335 	spin_unlock(&vb->stop_update_lock);
336 }
337 
338 static void stats_handle_request(struct virtio_balloon *vb)
339 {
340 	struct virtqueue *vq;
341 	struct scatterlist sg;
342 	unsigned int len, num_stats;
343 
344 	num_stats = update_balloon_stats(vb);
345 
346 	vq = vb->stats_vq;
347 	if (!virtqueue_get_buf(vq, &len))
348 		return;
349 	sg_init_one(&sg, vb->stats, sizeof(vb->stats[0]) * num_stats);
350 	virtqueue_add_outbuf(vq, &sg, 1, vb, GFP_KERNEL);
351 	virtqueue_kick(vq);
352 }
353 
354 static inline s64 towards_target(struct virtio_balloon *vb)
355 {
356 	s64 target;
357 	u32 num_pages;
358 
359 	virtio_cread(vb->vdev, struct virtio_balloon_config, num_pages,
360 		     &num_pages);
361 
362 	/* Legacy balloon config space is LE, unlike all other devices. */
363 	if (!virtio_has_feature(vb->vdev, VIRTIO_F_VERSION_1))
364 		num_pages = le32_to_cpu((__force __le32)num_pages);
365 
366 	target = num_pages;
367 	return target - vb->num_pages;
368 }
369 
370 /* Gives back @num_to_return blocks of free pages to mm. */
371 static unsigned long return_free_pages_to_mm(struct virtio_balloon *vb,
372 					     unsigned long num_to_return)
373 {
374 	struct page *page;
375 	unsigned long num_returned;
376 
377 	spin_lock_irq(&vb->free_page_list_lock);
378 	for (num_returned = 0; num_returned < num_to_return; num_returned++) {
379 		page = balloon_page_pop(&vb->free_page_list);
380 		if (!page)
381 			break;
382 		free_pages((unsigned long)page_address(page),
383 			   VIRTIO_BALLOON_FREE_PAGE_ORDER);
384 	}
385 	vb->num_free_page_blocks -= num_returned;
386 	spin_unlock_irq(&vb->free_page_list_lock);
387 
388 	return num_returned;
389 }
390 
391 static void virtio_balloon_queue_free_page_work(struct virtio_balloon *vb)
392 {
393 	if (!virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT))
394 		return;
395 
396 	/* No need to queue the work if the bit was already set. */
397 	if (test_and_set_bit(VIRTIO_BALLOON_CONFIG_READ_CMD_ID,
398 			     &vb->config_read_bitmap))
399 		return;
400 
401 	queue_work(vb->balloon_wq, &vb->report_free_page_work);
402 }
403 
404 static void virtballoon_changed(struct virtio_device *vdev)
405 {
406 	struct virtio_balloon *vb = vdev->priv;
407 	unsigned long flags;
408 
409 	spin_lock_irqsave(&vb->stop_update_lock, flags);
410 	if (!vb->stop_update) {
411 		queue_work(system_freezable_wq,
412 			   &vb->update_balloon_size_work);
413 		virtio_balloon_queue_free_page_work(vb);
414 	}
415 	spin_unlock_irqrestore(&vb->stop_update_lock, flags);
416 }
417 
418 static void update_balloon_size(struct virtio_balloon *vb)
419 {
420 	u32 actual = vb->num_pages;
421 
422 	/* Legacy balloon config space is LE, unlike all other devices. */
423 	if (!virtio_has_feature(vb->vdev, VIRTIO_F_VERSION_1))
424 		actual = (__force u32)cpu_to_le32(actual);
425 
426 	virtio_cwrite(vb->vdev, struct virtio_balloon_config, actual,
427 		      &actual);
428 }
429 
430 static void update_balloon_stats_func(struct work_struct *work)
431 {
432 	struct virtio_balloon *vb;
433 
434 	vb = container_of(work, struct virtio_balloon,
435 			  update_balloon_stats_work);
436 	stats_handle_request(vb);
437 }
438 
439 static void update_balloon_size_func(struct work_struct *work)
440 {
441 	struct virtio_balloon *vb;
442 	s64 diff;
443 
444 	vb = container_of(work, struct virtio_balloon,
445 			  update_balloon_size_work);
446 	diff = towards_target(vb);
447 
448 	if (!diff)
449 		return;
450 
451 	if (diff > 0)
452 		diff -= fill_balloon(vb, diff);
453 	else
454 		diff += leak_balloon(vb, -diff);
455 	update_balloon_size(vb);
456 
457 	if (diff)
458 		queue_work(system_freezable_wq, work);
459 }
460 
461 static int init_vqs(struct virtio_balloon *vb)
462 {
463 	struct virtqueue *vqs[VIRTIO_BALLOON_VQ_MAX];
464 	vq_callback_t *callbacks[VIRTIO_BALLOON_VQ_MAX];
465 	const char *names[VIRTIO_BALLOON_VQ_MAX];
466 	int err;
467 
468 	/*
469 	 * Inflateq and deflateq are used unconditionally. The names[]
470 	 * will be NULL if the related feature is not enabled, which will
471 	 * cause no allocation for the corresponding virtqueue in find_vqs.
472 	 */
473 	callbacks[VIRTIO_BALLOON_VQ_INFLATE] = balloon_ack;
474 	names[VIRTIO_BALLOON_VQ_INFLATE] = "inflate";
475 	callbacks[VIRTIO_BALLOON_VQ_DEFLATE] = balloon_ack;
476 	names[VIRTIO_BALLOON_VQ_DEFLATE] = "deflate";
477 	names[VIRTIO_BALLOON_VQ_STATS] = NULL;
478 	names[VIRTIO_BALLOON_VQ_FREE_PAGE] = NULL;
479 
480 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ)) {
481 		names[VIRTIO_BALLOON_VQ_STATS] = "stats";
482 		callbacks[VIRTIO_BALLOON_VQ_STATS] = stats_request;
483 	}
484 
485 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT)) {
486 		names[VIRTIO_BALLOON_VQ_FREE_PAGE] = "free_page_vq";
487 		callbacks[VIRTIO_BALLOON_VQ_FREE_PAGE] = NULL;
488 	}
489 
490 	err = vb->vdev->config->find_vqs(vb->vdev, VIRTIO_BALLOON_VQ_MAX,
491 					 vqs, callbacks, names, NULL, NULL);
492 	if (err)
493 		return err;
494 
495 	vb->inflate_vq = vqs[VIRTIO_BALLOON_VQ_INFLATE];
496 	vb->deflate_vq = vqs[VIRTIO_BALLOON_VQ_DEFLATE];
497 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ)) {
498 		struct scatterlist sg;
499 		unsigned int num_stats;
500 		vb->stats_vq = vqs[VIRTIO_BALLOON_VQ_STATS];
501 
502 		/*
503 		 * Prime this virtqueue with one buffer so the hypervisor can
504 		 * use it to signal us later (it can't be broken yet!).
505 		 */
506 		num_stats = update_balloon_stats(vb);
507 
508 		sg_init_one(&sg, vb->stats, sizeof(vb->stats[0]) * num_stats);
509 		err = virtqueue_add_outbuf(vb->stats_vq, &sg, 1, vb,
510 					   GFP_KERNEL);
511 		if (err) {
512 			dev_warn(&vb->vdev->dev, "%s: add stat_vq failed\n",
513 				 __func__);
514 			return err;
515 		}
516 		virtqueue_kick(vb->stats_vq);
517 	}
518 
519 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT))
520 		vb->free_page_vq = vqs[VIRTIO_BALLOON_VQ_FREE_PAGE];
521 
522 	return 0;
523 }
524 
525 static u32 virtio_balloon_cmd_id_received(struct virtio_balloon *vb)
526 {
527 	if (test_and_clear_bit(VIRTIO_BALLOON_CONFIG_READ_CMD_ID,
528 			       &vb->config_read_bitmap))
529 		virtio_cread(vb->vdev, struct virtio_balloon_config,
530 			     free_page_report_cmd_id,
531 			     &vb->cmd_id_received_cache);
532 
533 	return vb->cmd_id_received_cache;
534 }
535 
536 static int send_cmd_id_start(struct virtio_balloon *vb)
537 {
538 	struct scatterlist sg;
539 	struct virtqueue *vq = vb->free_page_vq;
540 	int err, unused;
541 
542 	/* Detach all the used buffers from the vq */
543 	while (virtqueue_get_buf(vq, &unused))
544 		;
545 
546 	vb->cmd_id_active = virtio32_to_cpu(vb->vdev,
547 					virtio_balloon_cmd_id_received(vb));
548 	sg_init_one(&sg, &vb->cmd_id_active, sizeof(vb->cmd_id_active));
549 	err = virtqueue_add_outbuf(vq, &sg, 1, &vb->cmd_id_active, GFP_KERNEL);
550 	if (!err)
551 		virtqueue_kick(vq);
552 	return err;
553 }
554 
555 static int send_cmd_id_stop(struct virtio_balloon *vb)
556 {
557 	struct scatterlist sg;
558 	struct virtqueue *vq = vb->free_page_vq;
559 	int err, unused;
560 
561 	/* Detach all the used buffers from the vq */
562 	while (virtqueue_get_buf(vq, &unused))
563 		;
564 
565 	sg_init_one(&sg, &vb->cmd_id_stop, sizeof(vb->cmd_id_stop));
566 	err = virtqueue_add_outbuf(vq, &sg, 1, &vb->cmd_id_stop, GFP_KERNEL);
567 	if (!err)
568 		virtqueue_kick(vq);
569 	return err;
570 }
571 
572 static int get_free_page_and_send(struct virtio_balloon *vb)
573 {
574 	struct virtqueue *vq = vb->free_page_vq;
575 	struct page *page;
576 	struct scatterlist sg;
577 	int err, unused;
578 	void *p;
579 
580 	/* Detach all the used buffers from the vq */
581 	while (virtqueue_get_buf(vq, &unused))
582 		;
583 
584 	page = alloc_pages(VIRTIO_BALLOON_FREE_PAGE_ALLOC_FLAG,
585 			   VIRTIO_BALLOON_FREE_PAGE_ORDER);
586 	/*
587 	 * When the allocation returns NULL, it indicates that we have got all
588 	 * the possible free pages, so return -EINTR to stop.
589 	 */
590 	if (!page)
591 		return -EINTR;
592 
593 	p = page_address(page);
594 	sg_init_one(&sg, p, VIRTIO_BALLOON_FREE_PAGE_SIZE);
595 	/* There is always 1 entry reserved for the cmd id to use. */
596 	if (vq->num_free > 1) {
597 		err = virtqueue_add_inbuf(vq, &sg, 1, p, GFP_KERNEL);
598 		if (unlikely(err)) {
599 			free_pages((unsigned long)p,
600 				   VIRTIO_BALLOON_FREE_PAGE_ORDER);
601 			return err;
602 		}
603 		virtqueue_kick(vq);
604 		spin_lock_irq(&vb->free_page_list_lock);
605 		balloon_page_push(&vb->free_page_list, page);
606 		vb->num_free_page_blocks++;
607 		spin_unlock_irq(&vb->free_page_list_lock);
608 	} else {
609 		/*
610 		 * The vq has no available entry to add this page block, so
611 		 * just free it.
612 		 */
613 		free_pages((unsigned long)p, VIRTIO_BALLOON_FREE_PAGE_ORDER);
614 	}
615 
616 	return 0;
617 }
618 
619 static int send_free_pages(struct virtio_balloon *vb)
620 {
621 	int err;
622 	u32 cmd_id_active;
623 
624 	while (1) {
625 		/*
626 		 * If a stop id or a new cmd id was just received from host,
627 		 * stop the reporting.
628 		 */
629 		cmd_id_active = virtio32_to_cpu(vb->vdev, vb->cmd_id_active);
630 		if (unlikely(cmd_id_active !=
631 			     virtio_balloon_cmd_id_received(vb)))
632 			break;
633 
634 		/*
635 		 * The free page blocks are allocated and sent to host one by
636 		 * one.
637 		 */
638 		err = get_free_page_and_send(vb);
639 		if (err == -EINTR)
640 			break;
641 		else if (unlikely(err))
642 			return err;
643 	}
644 
645 	return 0;
646 }
647 
648 static void virtio_balloon_report_free_page(struct virtio_balloon *vb)
649 {
650 	int err;
651 	struct device *dev = &vb->vdev->dev;
652 
653 	/* Start by sending the received cmd id to host with an outbuf. */
654 	err = send_cmd_id_start(vb);
655 	if (unlikely(err))
656 		dev_err(dev, "Failed to send a start id, err = %d\n", err);
657 
658 	err = send_free_pages(vb);
659 	if (unlikely(err))
660 		dev_err(dev, "Failed to send a free page, err = %d\n", err);
661 
662 	/* End by sending a stop id to host with an outbuf. */
663 	err = send_cmd_id_stop(vb);
664 	if (unlikely(err))
665 		dev_err(dev, "Failed to send a stop id, err = %d\n", err);
666 }
667 
668 static void report_free_page_func(struct work_struct *work)
669 {
670 	struct virtio_balloon *vb = container_of(work, struct virtio_balloon,
671 						 report_free_page_work);
672 	u32 cmd_id_received;
673 
674 	cmd_id_received = virtio_balloon_cmd_id_received(vb);
675 	if (cmd_id_received == VIRTIO_BALLOON_CMD_ID_DONE) {
676 		/* Pass ULONG_MAX to give back all the free pages */
677 		return_free_pages_to_mm(vb, ULONG_MAX);
678 	} else if (cmd_id_received != VIRTIO_BALLOON_CMD_ID_STOP &&
679 		   cmd_id_received !=
680 		   virtio32_to_cpu(vb->vdev, vb->cmd_id_active)) {
681 		virtio_balloon_report_free_page(vb);
682 	}
683 }
684 
685 #ifdef CONFIG_BALLOON_COMPACTION
686 /*
687  * virtballoon_migratepage - perform the balloon page migration on behalf of
688  *			     a compation thread.     (called under page lock)
689  * @vb_dev_info: the balloon device
690  * @newpage: page that will replace the isolated page after migration finishes.
691  * @page   : the isolated (old) page that is about to be migrated to newpage.
692  * @mode   : compaction mode -- not used for balloon page migration.
693  *
694  * After a ballooned page gets isolated by compaction procedures, this is the
695  * function that performs the page migration on behalf of a compaction thread
696  * The page migration for virtio balloon is done in a simple swap fashion which
697  * follows these two macro steps:
698  *  1) insert newpage into vb->pages list and update the host about it;
699  *  2) update the host about the old page removed from vb->pages list;
700  *
701  * This function preforms the balloon page migration task.
702  * Called through balloon_mapping->a_ops->migratepage
703  */
704 static int virtballoon_migratepage(struct balloon_dev_info *vb_dev_info,
705 		struct page *newpage, struct page *page, enum migrate_mode mode)
706 {
707 	struct virtio_balloon *vb = container_of(vb_dev_info,
708 			struct virtio_balloon, vb_dev_info);
709 	unsigned long flags;
710 
711 	/*
712 	 * In order to avoid lock contention while migrating pages concurrently
713 	 * to leak_balloon() or fill_balloon() we just give up the balloon_lock
714 	 * this turn, as it is easier to retry the page migration later.
715 	 * This also prevents fill_balloon() getting stuck into a mutex
716 	 * recursion in the case it ends up triggering memory compaction
717 	 * while it is attempting to inflate the ballon.
718 	 */
719 	if (!mutex_trylock(&vb->balloon_lock))
720 		return -EAGAIN;
721 
722 	get_page(newpage); /* balloon reference */
723 
724 	/* balloon's page migration 1st step  -- inflate "newpage" */
725 	spin_lock_irqsave(&vb_dev_info->pages_lock, flags);
726 	balloon_page_insert(vb_dev_info, newpage);
727 	vb_dev_info->isolated_pages--;
728 	__count_vm_event(BALLOON_MIGRATE);
729 	spin_unlock_irqrestore(&vb_dev_info->pages_lock, flags);
730 	vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE;
731 	set_page_pfns(vb, vb->pfns, newpage);
732 	tell_host(vb, vb->inflate_vq);
733 
734 	/* balloon's page migration 2nd step -- deflate "page" */
735 	spin_lock_irqsave(&vb_dev_info->pages_lock, flags);
736 	balloon_page_delete(page);
737 	spin_unlock_irqrestore(&vb_dev_info->pages_lock, flags);
738 	vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE;
739 	set_page_pfns(vb, vb->pfns, page);
740 	tell_host(vb, vb->deflate_vq);
741 
742 	mutex_unlock(&vb->balloon_lock);
743 
744 	put_page(page); /* balloon reference */
745 
746 	return MIGRATEPAGE_SUCCESS;
747 }
748 
749 static int balloon_init_fs_context(struct fs_context *fc)
750 {
751 	return init_pseudo(fc, BALLOON_KVM_MAGIC) ? 0 : -ENOMEM;
752 }
753 
754 static struct file_system_type balloon_fs = {
755 	.name           = "balloon-kvm",
756 	.init_fs_context = balloon_init_fs_context,
757 	.kill_sb        = kill_anon_super,
758 };
759 
760 #endif /* CONFIG_BALLOON_COMPACTION */
761 
762 static unsigned long shrink_free_pages(struct virtio_balloon *vb,
763 				       unsigned long pages_to_free)
764 {
765 	unsigned long blocks_to_free, blocks_freed;
766 
767 	pages_to_free = round_up(pages_to_free,
768 				 1 << VIRTIO_BALLOON_FREE_PAGE_ORDER);
769 	blocks_to_free = pages_to_free >> VIRTIO_BALLOON_FREE_PAGE_ORDER;
770 	blocks_freed = return_free_pages_to_mm(vb, blocks_to_free);
771 
772 	return blocks_freed << VIRTIO_BALLOON_FREE_PAGE_ORDER;
773 }
774 
775 static unsigned long leak_balloon_pages(struct virtio_balloon *vb,
776                                           unsigned long pages_to_free)
777 {
778 	return leak_balloon(vb, pages_to_free * VIRTIO_BALLOON_PAGES_PER_PAGE) /
779 		VIRTIO_BALLOON_PAGES_PER_PAGE;
780 }
781 
782 static unsigned long shrink_balloon_pages(struct virtio_balloon *vb,
783 					  unsigned long pages_to_free)
784 {
785 	unsigned long pages_freed = 0;
786 
787 	/*
788 	 * One invocation of leak_balloon can deflate at most
789 	 * VIRTIO_BALLOON_ARRAY_PFNS_MAX balloon pages, so we call it
790 	 * multiple times to deflate pages till reaching pages_to_free.
791 	 */
792 	while (vb->num_pages && pages_freed < pages_to_free)
793 		pages_freed += leak_balloon_pages(vb,
794 						  pages_to_free - pages_freed);
795 
796 	update_balloon_size(vb);
797 
798 	return pages_freed;
799 }
800 
801 static unsigned long virtio_balloon_shrinker_scan(struct shrinker *shrinker,
802 						  struct shrink_control *sc)
803 {
804 	unsigned long pages_to_free, pages_freed = 0;
805 	struct virtio_balloon *vb = container_of(shrinker,
806 					struct virtio_balloon, shrinker);
807 
808 	pages_to_free = sc->nr_to_scan;
809 
810 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT))
811 		pages_freed = shrink_free_pages(vb, pages_to_free);
812 
813 	if (pages_freed >= pages_to_free)
814 		return pages_freed;
815 
816 	pages_freed += shrink_balloon_pages(vb, pages_to_free - pages_freed);
817 
818 	return pages_freed;
819 }
820 
821 static unsigned long virtio_balloon_shrinker_count(struct shrinker *shrinker,
822 						   struct shrink_control *sc)
823 {
824 	struct virtio_balloon *vb = container_of(shrinker,
825 					struct virtio_balloon, shrinker);
826 	unsigned long count;
827 
828 	count = vb->num_pages / VIRTIO_BALLOON_PAGES_PER_PAGE;
829 	count += vb->num_free_page_blocks << VIRTIO_BALLOON_FREE_PAGE_ORDER;
830 
831 	return count;
832 }
833 
834 static void virtio_balloon_unregister_shrinker(struct virtio_balloon *vb)
835 {
836 	unregister_shrinker(&vb->shrinker);
837 }
838 
839 static int virtio_balloon_register_shrinker(struct virtio_balloon *vb)
840 {
841 	vb->shrinker.scan_objects = virtio_balloon_shrinker_scan;
842 	vb->shrinker.count_objects = virtio_balloon_shrinker_count;
843 	vb->shrinker.seeks = DEFAULT_SEEKS;
844 
845 	return register_shrinker(&vb->shrinker);
846 }
847 
848 static int virtballoon_probe(struct virtio_device *vdev)
849 {
850 	struct virtio_balloon *vb;
851 	__u32 poison_val;
852 	int err;
853 
854 	if (!vdev->config->get) {
855 		dev_err(&vdev->dev, "%s failure: config access disabled\n",
856 			__func__);
857 		return -EINVAL;
858 	}
859 
860 	vdev->priv = vb = kzalloc(sizeof(*vb), GFP_KERNEL);
861 	if (!vb) {
862 		err = -ENOMEM;
863 		goto out;
864 	}
865 
866 	INIT_WORK(&vb->update_balloon_stats_work, update_balloon_stats_func);
867 	INIT_WORK(&vb->update_balloon_size_work, update_balloon_size_func);
868 	spin_lock_init(&vb->stop_update_lock);
869 	mutex_init(&vb->balloon_lock);
870 	init_waitqueue_head(&vb->acked);
871 	vb->vdev = vdev;
872 
873 	balloon_devinfo_init(&vb->vb_dev_info);
874 
875 	err = init_vqs(vb);
876 	if (err)
877 		goto out_free_vb;
878 
879 #ifdef CONFIG_BALLOON_COMPACTION
880 	balloon_mnt = kern_mount(&balloon_fs);
881 	if (IS_ERR(balloon_mnt)) {
882 		err = PTR_ERR(balloon_mnt);
883 		goto out_del_vqs;
884 	}
885 
886 	vb->vb_dev_info.migratepage = virtballoon_migratepage;
887 	vb->vb_dev_info.inode = alloc_anon_inode(balloon_mnt->mnt_sb);
888 	if (IS_ERR(vb->vb_dev_info.inode)) {
889 		err = PTR_ERR(vb->vb_dev_info.inode);
890 		kern_unmount(balloon_mnt);
891 		goto out_del_vqs;
892 	}
893 	vb->vb_dev_info.inode->i_mapping->a_ops = &balloon_aops;
894 #endif
895 	if (virtio_has_feature(vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT)) {
896 		/*
897 		 * There is always one entry reserved for cmd id, so the ring
898 		 * size needs to be at least two to report free page hints.
899 		 */
900 		if (virtqueue_get_vring_size(vb->free_page_vq) < 2) {
901 			err = -ENOSPC;
902 			goto out_del_vqs;
903 		}
904 		vb->balloon_wq = alloc_workqueue("balloon-wq",
905 					WQ_FREEZABLE | WQ_CPU_INTENSIVE, 0);
906 		if (!vb->balloon_wq) {
907 			err = -ENOMEM;
908 			goto out_del_vqs;
909 		}
910 		INIT_WORK(&vb->report_free_page_work, report_free_page_func);
911 		vb->cmd_id_received_cache = VIRTIO_BALLOON_CMD_ID_STOP;
912 		vb->cmd_id_active = cpu_to_virtio32(vb->vdev,
913 						  VIRTIO_BALLOON_CMD_ID_STOP);
914 		vb->cmd_id_stop = cpu_to_virtio32(vb->vdev,
915 						  VIRTIO_BALLOON_CMD_ID_STOP);
916 		spin_lock_init(&vb->free_page_list_lock);
917 		INIT_LIST_HEAD(&vb->free_page_list);
918 		if (virtio_has_feature(vdev, VIRTIO_BALLOON_F_PAGE_POISON)) {
919 			memset(&poison_val, PAGE_POISON, sizeof(poison_val));
920 			virtio_cwrite(vb->vdev, struct virtio_balloon_config,
921 				      poison_val, &poison_val);
922 		}
923 	}
924 	/*
925 	 * We continue to use VIRTIO_BALLOON_F_DEFLATE_ON_OOM to decide if a
926 	 * shrinker needs to be registered to relieve memory pressure.
927 	 */
928 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_DEFLATE_ON_OOM)) {
929 		err = virtio_balloon_register_shrinker(vb);
930 		if (err)
931 			goto out_del_balloon_wq;
932 	}
933 	virtio_device_ready(vdev);
934 
935 	if (towards_target(vb))
936 		virtballoon_changed(vdev);
937 	return 0;
938 
939 out_del_balloon_wq:
940 	if (virtio_has_feature(vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT))
941 		destroy_workqueue(vb->balloon_wq);
942 out_del_vqs:
943 	vdev->config->del_vqs(vdev);
944 out_free_vb:
945 	kfree(vb);
946 out:
947 	return err;
948 }
949 
950 static void remove_common(struct virtio_balloon *vb)
951 {
952 	/* There might be pages left in the balloon: free them. */
953 	while (vb->num_pages)
954 		leak_balloon(vb, vb->num_pages);
955 	update_balloon_size(vb);
956 
957 	/* Now we reset the device so we can clean up the queues. */
958 	vb->vdev->config->reset(vb->vdev);
959 
960 	vb->vdev->config->del_vqs(vb->vdev);
961 }
962 
963 static void virtballoon_remove(struct virtio_device *vdev)
964 {
965 	struct virtio_balloon *vb = vdev->priv;
966 
967 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_DEFLATE_ON_OOM))
968 		virtio_balloon_unregister_shrinker(vb);
969 	spin_lock_irq(&vb->stop_update_lock);
970 	vb->stop_update = true;
971 	spin_unlock_irq(&vb->stop_update_lock);
972 	cancel_work_sync(&vb->update_balloon_size_work);
973 	cancel_work_sync(&vb->update_balloon_stats_work);
974 
975 	if (virtio_has_feature(vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT)) {
976 		cancel_work_sync(&vb->report_free_page_work);
977 		destroy_workqueue(vb->balloon_wq);
978 	}
979 
980 	remove_common(vb);
981 #ifdef CONFIG_BALLOON_COMPACTION
982 	if (vb->vb_dev_info.inode)
983 		iput(vb->vb_dev_info.inode);
984 
985 	kern_unmount(balloon_mnt);
986 #endif
987 	kfree(vb);
988 }
989 
990 #ifdef CONFIG_PM_SLEEP
991 static int virtballoon_freeze(struct virtio_device *vdev)
992 {
993 	struct virtio_balloon *vb = vdev->priv;
994 
995 	/*
996 	 * The workqueue is already frozen by the PM core before this
997 	 * function is called.
998 	 */
999 	remove_common(vb);
1000 	return 0;
1001 }
1002 
1003 static int virtballoon_restore(struct virtio_device *vdev)
1004 {
1005 	struct virtio_balloon *vb = vdev->priv;
1006 	int ret;
1007 
1008 	ret = init_vqs(vdev->priv);
1009 	if (ret)
1010 		return ret;
1011 
1012 	virtio_device_ready(vdev);
1013 
1014 	if (towards_target(vb))
1015 		virtballoon_changed(vdev);
1016 	update_balloon_size(vb);
1017 	return 0;
1018 }
1019 #endif
1020 
1021 static int virtballoon_validate(struct virtio_device *vdev)
1022 {
1023 	if (!page_poisoning_enabled())
1024 		__virtio_clear_bit(vdev, VIRTIO_BALLOON_F_PAGE_POISON);
1025 
1026 	__virtio_clear_bit(vdev, VIRTIO_F_IOMMU_PLATFORM);
1027 	return 0;
1028 }
1029 
1030 static unsigned int features[] = {
1031 	VIRTIO_BALLOON_F_MUST_TELL_HOST,
1032 	VIRTIO_BALLOON_F_STATS_VQ,
1033 	VIRTIO_BALLOON_F_DEFLATE_ON_OOM,
1034 	VIRTIO_BALLOON_F_FREE_PAGE_HINT,
1035 	VIRTIO_BALLOON_F_PAGE_POISON,
1036 };
1037 
1038 static struct virtio_driver virtio_balloon_driver = {
1039 	.feature_table = features,
1040 	.feature_table_size = ARRAY_SIZE(features),
1041 	.driver.name =	KBUILD_MODNAME,
1042 	.driver.owner =	THIS_MODULE,
1043 	.id_table =	id_table,
1044 	.validate =	virtballoon_validate,
1045 	.probe =	virtballoon_probe,
1046 	.remove =	virtballoon_remove,
1047 	.config_changed = virtballoon_changed,
1048 #ifdef CONFIG_PM_SLEEP
1049 	.freeze	=	virtballoon_freeze,
1050 	.restore =	virtballoon_restore,
1051 #endif
1052 };
1053 
1054 module_virtio_driver(virtio_balloon_driver);
1055 MODULE_DEVICE_TABLE(virtio, id_table);
1056 MODULE_DESCRIPTION("Virtio balloon driver");
1057 MODULE_LICENSE("GPL");
1058