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_HINT_BLOCK_ORDER (MAX_ORDER - 1)
36 /* The size of a free page block in bytes */
37 #define VIRTIO_BALLOON_HINT_BLOCK_BYTES \
38 	(1 << (VIRTIO_BALLOON_HINT_BLOCK_ORDER + PAGE_SHIFT))
39 #define VIRTIO_BALLOON_HINT_BLOCK_PAGES (1 << VIRTIO_BALLOON_HINT_BLOCK_ORDER)
40 
41 #ifdef CONFIG_BALLOON_COMPACTION
42 static struct vfsmount *balloon_mnt;
43 #endif
44 
45 enum virtio_balloon_vq {
46 	VIRTIO_BALLOON_VQ_INFLATE,
47 	VIRTIO_BALLOON_VQ_DEFLATE,
48 	VIRTIO_BALLOON_VQ_STATS,
49 	VIRTIO_BALLOON_VQ_FREE_PAGE,
50 	VIRTIO_BALLOON_VQ_MAX
51 };
52 
53 enum virtio_balloon_config_read {
54 	VIRTIO_BALLOON_CONFIG_READ_CMD_ID = 0,
55 };
56 
57 struct virtio_balloon {
58 	struct virtio_device *vdev;
59 	struct virtqueue *inflate_vq, *deflate_vq, *stats_vq, *free_page_vq;
60 
61 	/* Balloon's own wq for cpu-intensive work items */
62 	struct workqueue_struct *balloon_wq;
63 	/* The free page reporting work item submitted to the balloon wq */
64 	struct work_struct report_free_page_work;
65 
66 	/* The balloon servicing is delegated to a freezable workqueue. */
67 	struct work_struct update_balloon_stats_work;
68 	struct work_struct update_balloon_size_work;
69 
70 	/* Prevent updating balloon when it is being canceled. */
71 	spinlock_t stop_update_lock;
72 	bool stop_update;
73 	/* Bitmap to indicate if reading the related config fields are needed */
74 	unsigned long config_read_bitmap;
75 
76 	/* The list of allocated free pages, waiting to be given back to mm */
77 	struct list_head free_page_list;
78 	spinlock_t free_page_list_lock;
79 	/* The number of free page blocks on the above list */
80 	unsigned long num_free_page_blocks;
81 	/*
82 	 * The cmd id received from host.
83 	 * Read it via virtio_balloon_cmd_id_received to get the latest value
84 	 * sent from host.
85 	 */
86 	u32 cmd_id_received_cache;
87 	/* The cmd id that is actively in use */
88 	__virtio32 cmd_id_active;
89 	/* Buffer to store the stop sign */
90 	__virtio32 cmd_id_stop;
91 
92 	/* Waiting for host to ack the pages we released. */
93 	wait_queue_head_t acked;
94 
95 	/* Number of balloon pages we've told the Host we're not using. */
96 	unsigned int num_pages;
97 	/*
98 	 * The pages we've told the Host we're not using are enqueued
99 	 * at vb_dev_info->pages list.
100 	 * Each page on this list adds VIRTIO_BALLOON_PAGES_PER_PAGE
101 	 * to num_pages above.
102 	 */
103 	struct balloon_dev_info vb_dev_info;
104 
105 	/* Synchronize access/update to this struct virtio_balloon elements */
106 	struct mutex balloon_lock;
107 
108 	/* The array of pfns we tell the Host about. */
109 	unsigned int num_pfns;
110 	__virtio32 pfns[VIRTIO_BALLOON_ARRAY_PFNS_MAX];
111 
112 	/* Memory statistics */
113 	struct virtio_balloon_stat stats[VIRTIO_BALLOON_S_NR];
114 
115 	/* To register a shrinker to shrink memory upon memory pressure */
116 	struct shrinker shrinker;
117 };
118 
119 static struct virtio_device_id id_table[] = {
120 	{ VIRTIO_ID_BALLOON, VIRTIO_DEV_ANY_ID },
121 	{ 0 },
122 };
123 
124 static u32 page_to_balloon_pfn(struct page *page)
125 {
126 	unsigned long pfn = page_to_pfn(page);
127 
128 	BUILD_BUG_ON(PAGE_SHIFT < VIRTIO_BALLOON_PFN_SHIFT);
129 	/* Convert pfn from Linux page size to balloon page size. */
130 	return pfn * VIRTIO_BALLOON_PAGES_PER_PAGE;
131 }
132 
133 static void balloon_ack(struct virtqueue *vq)
134 {
135 	struct virtio_balloon *vb = vq->vdev->priv;
136 
137 	wake_up(&vb->acked);
138 }
139 
140 static void tell_host(struct virtio_balloon *vb, struct virtqueue *vq)
141 {
142 	struct scatterlist sg;
143 	unsigned int len;
144 
145 	sg_init_one(&sg, vb->pfns, sizeof(vb->pfns[0]) * vb->num_pfns);
146 
147 	/* We should always be able to add one buffer to an empty queue. */
148 	virtqueue_add_outbuf(vq, &sg, 1, vb, GFP_KERNEL);
149 	virtqueue_kick(vq);
150 
151 	/* When host has read buffer, this completes via balloon_ack */
152 	wait_event(vb->acked, virtqueue_get_buf(vq, &len));
153 
154 }
155 
156 static void set_page_pfns(struct virtio_balloon *vb,
157 			  __virtio32 pfns[], struct page *page)
158 {
159 	unsigned int i;
160 
161 	/*
162 	 * Set balloon pfns pointing at this page.
163 	 * Note that the first pfn points at start of the page.
164 	 */
165 	for (i = 0; i < VIRTIO_BALLOON_PAGES_PER_PAGE; i++)
166 		pfns[i] = cpu_to_virtio32(vb->vdev,
167 					  page_to_balloon_pfn(page) + i);
168 }
169 
170 static unsigned fill_balloon(struct virtio_balloon *vb, size_t num)
171 {
172 	unsigned num_allocated_pages;
173 	unsigned num_pfns;
174 	struct page *page;
175 	LIST_HEAD(pages);
176 
177 	/* We can only do one array worth at a time. */
178 	num = min(num, ARRAY_SIZE(vb->pfns));
179 
180 	for (num_pfns = 0; num_pfns < num;
181 	     num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) {
182 		struct page *page = balloon_page_alloc();
183 
184 		if (!page) {
185 			dev_info_ratelimited(&vb->vdev->dev,
186 					     "Out of puff! Can't get %u pages\n",
187 					     VIRTIO_BALLOON_PAGES_PER_PAGE);
188 			/* Sleep for at least 1/5 of a second before retry. */
189 			msleep(200);
190 			break;
191 		}
192 
193 		balloon_page_push(&pages, page);
194 	}
195 
196 	mutex_lock(&vb->balloon_lock);
197 
198 	vb->num_pfns = 0;
199 
200 	while ((page = balloon_page_pop(&pages))) {
201 		balloon_page_enqueue(&vb->vb_dev_info, page);
202 
203 		set_page_pfns(vb, vb->pfns + vb->num_pfns, page);
204 		vb->num_pages += VIRTIO_BALLOON_PAGES_PER_PAGE;
205 		if (!virtio_has_feature(vb->vdev,
206 					VIRTIO_BALLOON_F_DEFLATE_ON_OOM))
207 			adjust_managed_page_count(page, -1);
208 		vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE;
209 	}
210 
211 	num_allocated_pages = vb->num_pfns;
212 	/* Did we get any? */
213 	if (vb->num_pfns != 0)
214 		tell_host(vb, vb->inflate_vq);
215 	mutex_unlock(&vb->balloon_lock);
216 
217 	return num_allocated_pages;
218 }
219 
220 static void release_pages_balloon(struct virtio_balloon *vb,
221 				 struct list_head *pages)
222 {
223 	struct page *page, *next;
224 
225 	list_for_each_entry_safe(page, next, pages, lru) {
226 		if (!virtio_has_feature(vb->vdev,
227 					VIRTIO_BALLOON_F_DEFLATE_ON_OOM))
228 			adjust_managed_page_count(page, 1);
229 		list_del(&page->lru);
230 		put_page(page); /* balloon reference */
231 	}
232 }
233 
234 static unsigned leak_balloon(struct virtio_balloon *vb, size_t num)
235 {
236 	unsigned num_freed_pages;
237 	struct page *page;
238 	struct balloon_dev_info *vb_dev_info = &vb->vb_dev_info;
239 	LIST_HEAD(pages);
240 
241 	/* We can only do one array worth at a time. */
242 	num = min(num, ARRAY_SIZE(vb->pfns));
243 
244 	mutex_lock(&vb->balloon_lock);
245 	/* We can't release more pages than taken */
246 	num = min(num, (size_t)vb->num_pages);
247 	for (vb->num_pfns = 0; vb->num_pfns < num;
248 	     vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) {
249 		page = balloon_page_dequeue(vb_dev_info);
250 		if (!page)
251 			break;
252 		set_page_pfns(vb, vb->pfns + vb->num_pfns, page);
253 		list_add(&page->lru, &pages);
254 		vb->num_pages -= VIRTIO_BALLOON_PAGES_PER_PAGE;
255 	}
256 
257 	num_freed_pages = vb->num_pfns;
258 	/*
259 	 * Note that if
260 	 * virtio_has_feature(vdev, VIRTIO_BALLOON_F_MUST_TELL_HOST);
261 	 * is true, we *have* to do it in this order
262 	 */
263 	if (vb->num_pfns != 0)
264 		tell_host(vb, vb->deflate_vq);
265 	release_pages_balloon(vb, &pages);
266 	mutex_unlock(&vb->balloon_lock);
267 	return num_freed_pages;
268 }
269 
270 static inline void update_stat(struct virtio_balloon *vb, int idx,
271 			       u16 tag, u64 val)
272 {
273 	BUG_ON(idx >= VIRTIO_BALLOON_S_NR);
274 	vb->stats[idx].tag = cpu_to_virtio16(vb->vdev, tag);
275 	vb->stats[idx].val = cpu_to_virtio64(vb->vdev, val);
276 }
277 
278 #define pages_to_bytes(x) ((u64)(x) << PAGE_SHIFT)
279 
280 static unsigned int update_balloon_stats(struct virtio_balloon *vb)
281 {
282 	unsigned long events[NR_VM_EVENT_ITEMS];
283 	struct sysinfo i;
284 	unsigned int idx = 0;
285 	long available;
286 	unsigned long caches;
287 
288 	all_vm_events(events);
289 	si_meminfo(&i);
290 
291 	available = si_mem_available();
292 	caches = global_node_page_state(NR_FILE_PAGES);
293 
294 #ifdef CONFIG_VM_EVENT_COUNTERS
295 	update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_IN,
296 				pages_to_bytes(events[PSWPIN]));
297 	update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_OUT,
298 				pages_to_bytes(events[PSWPOUT]));
299 	update_stat(vb, idx++, VIRTIO_BALLOON_S_MAJFLT, events[PGMAJFAULT]);
300 	update_stat(vb, idx++, VIRTIO_BALLOON_S_MINFLT, events[PGFAULT]);
301 #ifdef CONFIG_HUGETLB_PAGE
302 	update_stat(vb, idx++, VIRTIO_BALLOON_S_HTLB_PGALLOC,
303 		    events[HTLB_BUDDY_PGALLOC]);
304 	update_stat(vb, idx++, VIRTIO_BALLOON_S_HTLB_PGFAIL,
305 		    events[HTLB_BUDDY_PGALLOC_FAIL]);
306 #endif
307 #endif
308 	update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMFREE,
309 				pages_to_bytes(i.freeram));
310 	update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMTOT,
311 				pages_to_bytes(i.totalram));
312 	update_stat(vb, idx++, VIRTIO_BALLOON_S_AVAIL,
313 				pages_to_bytes(available));
314 	update_stat(vb, idx++, VIRTIO_BALLOON_S_CACHES,
315 				pages_to_bytes(caches));
316 
317 	return idx;
318 }
319 
320 /*
321  * While most virtqueues communicate guest-initiated requests to the hypervisor,
322  * the stats queue operates in reverse.  The driver initializes the virtqueue
323  * with a single buffer.  From that point forward, all conversations consist of
324  * a hypervisor request (a call to this function) which directs us to refill
325  * the virtqueue with a fresh stats buffer.  Since stats collection can sleep,
326  * we delegate the job to a freezable workqueue that will do the actual work via
327  * stats_handle_request().
328  */
329 static void stats_request(struct virtqueue *vq)
330 {
331 	struct virtio_balloon *vb = vq->vdev->priv;
332 
333 	spin_lock(&vb->stop_update_lock);
334 	if (!vb->stop_update)
335 		queue_work(system_freezable_wq, &vb->update_balloon_stats_work);
336 	spin_unlock(&vb->stop_update_lock);
337 }
338 
339 static void stats_handle_request(struct virtio_balloon *vb)
340 {
341 	struct virtqueue *vq;
342 	struct scatterlist sg;
343 	unsigned int len, num_stats;
344 
345 	num_stats = update_balloon_stats(vb);
346 
347 	vq = vb->stats_vq;
348 	if (!virtqueue_get_buf(vq, &len))
349 		return;
350 	sg_init_one(&sg, vb->stats, sizeof(vb->stats[0]) * num_stats);
351 	virtqueue_add_outbuf(vq, &sg, 1, vb, GFP_KERNEL);
352 	virtqueue_kick(vq);
353 }
354 
355 static inline s64 towards_target(struct virtio_balloon *vb)
356 {
357 	s64 target;
358 	u32 num_pages;
359 
360 	virtio_cread(vb->vdev, struct virtio_balloon_config, num_pages,
361 		     &num_pages);
362 
363 	/* Legacy balloon config space is LE, unlike all other devices. */
364 	if (!virtio_has_feature(vb->vdev, VIRTIO_F_VERSION_1))
365 		num_pages = le32_to_cpu((__force __le32)num_pages);
366 
367 	target = num_pages;
368 	return target - vb->num_pages;
369 }
370 
371 /* Gives back @num_to_return blocks of free pages to mm. */
372 static unsigned long return_free_pages_to_mm(struct virtio_balloon *vb,
373 					     unsigned long num_to_return)
374 {
375 	struct page *page;
376 	unsigned long num_returned;
377 
378 	spin_lock_irq(&vb->free_page_list_lock);
379 	for (num_returned = 0; num_returned < num_to_return; num_returned++) {
380 		page = balloon_page_pop(&vb->free_page_list);
381 		if (!page)
382 			break;
383 		free_pages((unsigned long)page_address(page),
384 			   VIRTIO_BALLOON_HINT_BLOCK_ORDER);
385 	}
386 	vb->num_free_page_blocks -= num_returned;
387 	spin_unlock_irq(&vb->free_page_list_lock);
388 
389 	return num_returned;
390 }
391 
392 static void virtio_balloon_queue_free_page_work(struct virtio_balloon *vb)
393 {
394 	if (!virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT))
395 		return;
396 
397 	/* No need to queue the work if the bit was already set. */
398 	if (test_and_set_bit(VIRTIO_BALLOON_CONFIG_READ_CMD_ID,
399 			     &vb->config_read_bitmap))
400 		return;
401 
402 	queue_work(vb->balloon_wq, &vb->report_free_page_work);
403 }
404 
405 static void virtballoon_changed(struct virtio_device *vdev)
406 {
407 	struct virtio_balloon *vb = vdev->priv;
408 	unsigned long flags;
409 
410 	spin_lock_irqsave(&vb->stop_update_lock, flags);
411 	if (!vb->stop_update) {
412 		queue_work(system_freezable_wq,
413 			   &vb->update_balloon_size_work);
414 		virtio_balloon_queue_free_page_work(vb);
415 	}
416 	spin_unlock_irqrestore(&vb->stop_update_lock, flags);
417 }
418 
419 static void update_balloon_size(struct virtio_balloon *vb)
420 {
421 	u32 actual = vb->num_pages;
422 
423 	/* Legacy balloon config space is LE, unlike all other devices. */
424 	if (!virtio_has_feature(vb->vdev, VIRTIO_F_VERSION_1))
425 		actual = (__force u32)cpu_to_le32(actual);
426 
427 	virtio_cwrite(vb->vdev, struct virtio_balloon_config, actual,
428 		      &actual);
429 }
430 
431 static void update_balloon_stats_func(struct work_struct *work)
432 {
433 	struct virtio_balloon *vb;
434 
435 	vb = container_of(work, struct virtio_balloon,
436 			  update_balloon_stats_work);
437 	stats_handle_request(vb);
438 }
439 
440 static void update_balloon_size_func(struct work_struct *work)
441 {
442 	struct virtio_balloon *vb;
443 	s64 diff;
444 
445 	vb = container_of(work, struct virtio_balloon,
446 			  update_balloon_size_work);
447 	diff = towards_target(vb);
448 
449 	if (!diff)
450 		return;
451 
452 	if (diff > 0)
453 		diff -= fill_balloon(vb, diff);
454 	else
455 		diff += leak_balloon(vb, -diff);
456 	update_balloon_size(vb);
457 
458 	if (diff)
459 		queue_work(system_freezable_wq, work);
460 }
461 
462 static int init_vqs(struct virtio_balloon *vb)
463 {
464 	struct virtqueue *vqs[VIRTIO_BALLOON_VQ_MAX];
465 	vq_callback_t *callbacks[VIRTIO_BALLOON_VQ_MAX];
466 	const char *names[VIRTIO_BALLOON_VQ_MAX];
467 	int err;
468 
469 	/*
470 	 * Inflateq and deflateq are used unconditionally. The names[]
471 	 * will be NULL if the related feature is not enabled, which will
472 	 * cause no allocation for the corresponding virtqueue in find_vqs.
473 	 */
474 	callbacks[VIRTIO_BALLOON_VQ_INFLATE] = balloon_ack;
475 	names[VIRTIO_BALLOON_VQ_INFLATE] = "inflate";
476 	callbacks[VIRTIO_BALLOON_VQ_DEFLATE] = balloon_ack;
477 	names[VIRTIO_BALLOON_VQ_DEFLATE] = "deflate";
478 	names[VIRTIO_BALLOON_VQ_STATS] = NULL;
479 	names[VIRTIO_BALLOON_VQ_FREE_PAGE] = NULL;
480 
481 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ)) {
482 		names[VIRTIO_BALLOON_VQ_STATS] = "stats";
483 		callbacks[VIRTIO_BALLOON_VQ_STATS] = stats_request;
484 	}
485 
486 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT)) {
487 		names[VIRTIO_BALLOON_VQ_FREE_PAGE] = "free_page_vq";
488 		callbacks[VIRTIO_BALLOON_VQ_FREE_PAGE] = NULL;
489 	}
490 
491 	err = vb->vdev->config->find_vqs(vb->vdev, VIRTIO_BALLOON_VQ_MAX,
492 					 vqs, callbacks, names, NULL, NULL);
493 	if (err)
494 		return err;
495 
496 	vb->inflate_vq = vqs[VIRTIO_BALLOON_VQ_INFLATE];
497 	vb->deflate_vq = vqs[VIRTIO_BALLOON_VQ_DEFLATE];
498 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ)) {
499 		struct scatterlist sg;
500 		unsigned int num_stats;
501 		vb->stats_vq = vqs[VIRTIO_BALLOON_VQ_STATS];
502 
503 		/*
504 		 * Prime this virtqueue with one buffer so the hypervisor can
505 		 * use it to signal us later (it can't be broken yet!).
506 		 */
507 		num_stats = update_balloon_stats(vb);
508 
509 		sg_init_one(&sg, vb->stats, sizeof(vb->stats[0]) * num_stats);
510 		err = virtqueue_add_outbuf(vb->stats_vq, &sg, 1, vb,
511 					   GFP_KERNEL);
512 		if (err) {
513 			dev_warn(&vb->vdev->dev, "%s: add stat_vq failed\n",
514 				 __func__);
515 			return err;
516 		}
517 		virtqueue_kick(vb->stats_vq);
518 	}
519 
520 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT))
521 		vb->free_page_vq = vqs[VIRTIO_BALLOON_VQ_FREE_PAGE];
522 
523 	return 0;
524 }
525 
526 static u32 virtio_balloon_cmd_id_received(struct virtio_balloon *vb)
527 {
528 	if (test_and_clear_bit(VIRTIO_BALLOON_CONFIG_READ_CMD_ID,
529 			       &vb->config_read_bitmap))
530 		virtio_cread(vb->vdev, struct virtio_balloon_config,
531 			     free_page_report_cmd_id,
532 			     &vb->cmd_id_received_cache);
533 
534 	return vb->cmd_id_received_cache;
535 }
536 
537 static int send_cmd_id_start(struct virtio_balloon *vb)
538 {
539 	struct scatterlist sg;
540 	struct virtqueue *vq = vb->free_page_vq;
541 	int err, unused;
542 
543 	/* Detach all the used buffers from the vq */
544 	while (virtqueue_get_buf(vq, &unused))
545 		;
546 
547 	vb->cmd_id_active = virtio32_to_cpu(vb->vdev,
548 					virtio_balloon_cmd_id_received(vb));
549 	sg_init_one(&sg, &vb->cmd_id_active, sizeof(vb->cmd_id_active));
550 	err = virtqueue_add_outbuf(vq, &sg, 1, &vb->cmd_id_active, GFP_KERNEL);
551 	if (!err)
552 		virtqueue_kick(vq);
553 	return err;
554 }
555 
556 static int send_cmd_id_stop(struct virtio_balloon *vb)
557 {
558 	struct scatterlist sg;
559 	struct virtqueue *vq = vb->free_page_vq;
560 	int err, unused;
561 
562 	/* Detach all the used buffers from the vq */
563 	while (virtqueue_get_buf(vq, &unused))
564 		;
565 
566 	sg_init_one(&sg, &vb->cmd_id_stop, sizeof(vb->cmd_id_stop));
567 	err = virtqueue_add_outbuf(vq, &sg, 1, &vb->cmd_id_stop, GFP_KERNEL);
568 	if (!err)
569 		virtqueue_kick(vq);
570 	return err;
571 }
572 
573 static int get_free_page_and_send(struct virtio_balloon *vb)
574 {
575 	struct virtqueue *vq = vb->free_page_vq;
576 	struct page *page;
577 	struct scatterlist sg;
578 	int err, unused;
579 	void *p;
580 
581 	/* Detach all the used buffers from the vq */
582 	while (virtqueue_get_buf(vq, &unused))
583 		;
584 
585 	page = alloc_pages(VIRTIO_BALLOON_FREE_PAGE_ALLOC_FLAG,
586 			   VIRTIO_BALLOON_HINT_BLOCK_ORDER);
587 	/*
588 	 * When the allocation returns NULL, it indicates that we have got all
589 	 * the possible free pages, so return -EINTR to stop.
590 	 */
591 	if (!page)
592 		return -EINTR;
593 
594 	p = page_address(page);
595 	sg_init_one(&sg, p, VIRTIO_BALLOON_HINT_BLOCK_BYTES);
596 	/* There is always 1 entry reserved for the cmd id to use. */
597 	if (vq->num_free > 1) {
598 		err = virtqueue_add_inbuf(vq, &sg, 1, p, GFP_KERNEL);
599 		if (unlikely(err)) {
600 			free_pages((unsigned long)p,
601 				   VIRTIO_BALLOON_HINT_BLOCK_ORDER);
602 			return err;
603 		}
604 		virtqueue_kick(vq);
605 		spin_lock_irq(&vb->free_page_list_lock);
606 		balloon_page_push(&vb->free_page_list, page);
607 		vb->num_free_page_blocks++;
608 		spin_unlock_irq(&vb->free_page_list_lock);
609 	} else {
610 		/*
611 		 * The vq has no available entry to add this page block, so
612 		 * just free it.
613 		 */
614 		free_pages((unsigned long)p, VIRTIO_BALLOON_HINT_BLOCK_ORDER);
615 	}
616 
617 	return 0;
618 }
619 
620 static int send_free_pages(struct virtio_balloon *vb)
621 {
622 	int err;
623 	u32 cmd_id_active;
624 
625 	while (1) {
626 		/*
627 		 * If a stop id or a new cmd id was just received from host,
628 		 * stop the reporting.
629 		 */
630 		cmd_id_active = virtio32_to_cpu(vb->vdev, vb->cmd_id_active);
631 		if (unlikely(cmd_id_active !=
632 			     virtio_balloon_cmd_id_received(vb)))
633 			break;
634 
635 		/*
636 		 * The free page blocks are allocated and sent to host one by
637 		 * one.
638 		 */
639 		err = get_free_page_and_send(vb);
640 		if (err == -EINTR)
641 			break;
642 		else if (unlikely(err))
643 			return err;
644 	}
645 
646 	return 0;
647 }
648 
649 static void virtio_balloon_report_free_page(struct virtio_balloon *vb)
650 {
651 	int err;
652 	struct device *dev = &vb->vdev->dev;
653 
654 	/* Start by sending the received cmd id to host with an outbuf. */
655 	err = send_cmd_id_start(vb);
656 	if (unlikely(err))
657 		dev_err(dev, "Failed to send a start id, err = %d\n", err);
658 
659 	err = send_free_pages(vb);
660 	if (unlikely(err))
661 		dev_err(dev, "Failed to send a free page, err = %d\n", err);
662 
663 	/* End by sending a stop id to host with an outbuf. */
664 	err = send_cmd_id_stop(vb);
665 	if (unlikely(err))
666 		dev_err(dev, "Failed to send a stop id, err = %d\n", err);
667 }
668 
669 static void report_free_page_func(struct work_struct *work)
670 {
671 	struct virtio_balloon *vb = container_of(work, struct virtio_balloon,
672 						 report_free_page_work);
673 	u32 cmd_id_received;
674 
675 	cmd_id_received = virtio_balloon_cmd_id_received(vb);
676 	if (cmd_id_received == VIRTIO_BALLOON_CMD_ID_DONE) {
677 		/* Pass ULONG_MAX to give back all the free pages */
678 		return_free_pages_to_mm(vb, ULONG_MAX);
679 	} else if (cmd_id_received != VIRTIO_BALLOON_CMD_ID_STOP &&
680 		   cmd_id_received !=
681 		   virtio32_to_cpu(vb->vdev, vb->cmd_id_active)) {
682 		virtio_balloon_report_free_page(vb);
683 	}
684 }
685 
686 #ifdef CONFIG_BALLOON_COMPACTION
687 /*
688  * virtballoon_migratepage - perform the balloon page migration on behalf of
689  *			     a compation thread.     (called under page lock)
690  * @vb_dev_info: the balloon device
691  * @newpage: page that will replace the isolated page after migration finishes.
692  * @page   : the isolated (old) page that is about to be migrated to newpage.
693  * @mode   : compaction mode -- not used for balloon page migration.
694  *
695  * After a ballooned page gets isolated by compaction procedures, this is the
696  * function that performs the page migration on behalf of a compaction thread
697  * The page migration for virtio balloon is done in a simple swap fashion which
698  * follows these two macro steps:
699  *  1) insert newpage into vb->pages list and update the host about it;
700  *  2) update the host about the old page removed from vb->pages list;
701  *
702  * This function preforms the balloon page migration task.
703  * Called through balloon_mapping->a_ops->migratepage
704  */
705 static int virtballoon_migratepage(struct balloon_dev_info *vb_dev_info,
706 		struct page *newpage, struct page *page, enum migrate_mode mode)
707 {
708 	struct virtio_balloon *vb = container_of(vb_dev_info,
709 			struct virtio_balloon, vb_dev_info);
710 	unsigned long flags;
711 
712 	/*
713 	 * In order to avoid lock contention while migrating pages concurrently
714 	 * to leak_balloon() or fill_balloon() we just give up the balloon_lock
715 	 * this turn, as it is easier to retry the page migration later.
716 	 * This also prevents fill_balloon() getting stuck into a mutex
717 	 * recursion in the case it ends up triggering memory compaction
718 	 * while it is attempting to inflate the ballon.
719 	 */
720 	if (!mutex_trylock(&vb->balloon_lock))
721 		return -EAGAIN;
722 
723 	get_page(newpage); /* balloon reference */
724 
725 	/*
726 	  * When we migrate a page to a different zone and adjusted the
727 	  * managed page count when inflating, we have to fixup the count of
728 	  * both involved zones.
729 	  */
730 	if (!virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_DEFLATE_ON_OOM) &&
731 	    page_zone(page) != page_zone(newpage)) {
732 		adjust_managed_page_count(page, 1);
733 		adjust_managed_page_count(newpage, -1);
734 	}
735 
736 	/* balloon's page migration 1st step  -- inflate "newpage" */
737 	spin_lock_irqsave(&vb_dev_info->pages_lock, flags);
738 	balloon_page_insert(vb_dev_info, newpage);
739 	vb_dev_info->isolated_pages--;
740 	__count_vm_event(BALLOON_MIGRATE);
741 	spin_unlock_irqrestore(&vb_dev_info->pages_lock, flags);
742 	vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE;
743 	set_page_pfns(vb, vb->pfns, newpage);
744 	tell_host(vb, vb->inflate_vq);
745 
746 	/* balloon's page migration 2nd step -- deflate "page" */
747 	spin_lock_irqsave(&vb_dev_info->pages_lock, flags);
748 	balloon_page_delete(page);
749 	spin_unlock_irqrestore(&vb_dev_info->pages_lock, flags);
750 	vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE;
751 	set_page_pfns(vb, vb->pfns, page);
752 	tell_host(vb, vb->deflate_vq);
753 
754 	mutex_unlock(&vb->balloon_lock);
755 
756 	put_page(page); /* balloon reference */
757 
758 	return MIGRATEPAGE_SUCCESS;
759 }
760 
761 static int balloon_init_fs_context(struct fs_context *fc)
762 {
763 	return init_pseudo(fc, BALLOON_KVM_MAGIC) ? 0 : -ENOMEM;
764 }
765 
766 static struct file_system_type balloon_fs = {
767 	.name           = "balloon-kvm",
768 	.init_fs_context = balloon_init_fs_context,
769 	.kill_sb        = kill_anon_super,
770 };
771 
772 #endif /* CONFIG_BALLOON_COMPACTION */
773 
774 static unsigned long shrink_free_pages(struct virtio_balloon *vb,
775 				       unsigned long pages_to_free)
776 {
777 	unsigned long blocks_to_free, blocks_freed;
778 
779 	pages_to_free = round_up(pages_to_free,
780 				 VIRTIO_BALLOON_HINT_BLOCK_PAGES);
781 	blocks_to_free = pages_to_free / VIRTIO_BALLOON_HINT_BLOCK_PAGES;
782 	blocks_freed = return_free_pages_to_mm(vb, blocks_to_free);
783 
784 	return blocks_freed * VIRTIO_BALLOON_HINT_BLOCK_PAGES;
785 }
786 
787 static unsigned long leak_balloon_pages(struct virtio_balloon *vb,
788                                           unsigned long pages_to_free)
789 {
790 	return leak_balloon(vb, pages_to_free * VIRTIO_BALLOON_PAGES_PER_PAGE) /
791 		VIRTIO_BALLOON_PAGES_PER_PAGE;
792 }
793 
794 static unsigned long shrink_balloon_pages(struct virtio_balloon *vb,
795 					  unsigned long pages_to_free)
796 {
797 	unsigned long pages_freed = 0;
798 
799 	/*
800 	 * One invocation of leak_balloon can deflate at most
801 	 * VIRTIO_BALLOON_ARRAY_PFNS_MAX balloon pages, so we call it
802 	 * multiple times to deflate pages till reaching pages_to_free.
803 	 */
804 	while (vb->num_pages && pages_freed < pages_to_free)
805 		pages_freed += leak_balloon_pages(vb,
806 						  pages_to_free - pages_freed);
807 
808 	update_balloon_size(vb);
809 
810 	return pages_freed;
811 }
812 
813 static unsigned long virtio_balloon_shrinker_scan(struct shrinker *shrinker,
814 						  struct shrink_control *sc)
815 {
816 	unsigned long pages_to_free, pages_freed = 0;
817 	struct virtio_balloon *vb = container_of(shrinker,
818 					struct virtio_balloon, shrinker);
819 
820 	pages_to_free = sc->nr_to_scan;
821 
822 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT))
823 		pages_freed = shrink_free_pages(vb, pages_to_free);
824 
825 	if (pages_freed >= pages_to_free)
826 		return pages_freed;
827 
828 	pages_freed += shrink_balloon_pages(vb, pages_to_free - pages_freed);
829 
830 	return pages_freed;
831 }
832 
833 static unsigned long virtio_balloon_shrinker_count(struct shrinker *shrinker,
834 						   struct shrink_control *sc)
835 {
836 	struct virtio_balloon *vb = container_of(shrinker,
837 					struct virtio_balloon, shrinker);
838 	unsigned long count;
839 
840 	count = vb->num_pages / VIRTIO_BALLOON_PAGES_PER_PAGE;
841 	count += vb->num_free_page_blocks * VIRTIO_BALLOON_HINT_BLOCK_PAGES;
842 
843 	return count;
844 }
845 
846 static void virtio_balloon_unregister_shrinker(struct virtio_balloon *vb)
847 {
848 	unregister_shrinker(&vb->shrinker);
849 }
850 
851 static int virtio_balloon_register_shrinker(struct virtio_balloon *vb)
852 {
853 	vb->shrinker.scan_objects = virtio_balloon_shrinker_scan;
854 	vb->shrinker.count_objects = virtio_balloon_shrinker_count;
855 	vb->shrinker.seeks = DEFAULT_SEEKS;
856 
857 	return register_shrinker(&vb->shrinker);
858 }
859 
860 static int virtballoon_probe(struct virtio_device *vdev)
861 {
862 	struct virtio_balloon *vb;
863 	__u32 poison_val;
864 	int err;
865 
866 	if (!vdev->config->get) {
867 		dev_err(&vdev->dev, "%s failure: config access disabled\n",
868 			__func__);
869 		return -EINVAL;
870 	}
871 
872 	vdev->priv = vb = kzalloc(sizeof(*vb), GFP_KERNEL);
873 	if (!vb) {
874 		err = -ENOMEM;
875 		goto out;
876 	}
877 
878 	INIT_WORK(&vb->update_balloon_stats_work, update_balloon_stats_func);
879 	INIT_WORK(&vb->update_balloon_size_work, update_balloon_size_func);
880 	spin_lock_init(&vb->stop_update_lock);
881 	mutex_init(&vb->balloon_lock);
882 	init_waitqueue_head(&vb->acked);
883 	vb->vdev = vdev;
884 
885 	balloon_devinfo_init(&vb->vb_dev_info);
886 
887 	err = init_vqs(vb);
888 	if (err)
889 		goto out_free_vb;
890 
891 #ifdef CONFIG_BALLOON_COMPACTION
892 	balloon_mnt = kern_mount(&balloon_fs);
893 	if (IS_ERR(balloon_mnt)) {
894 		err = PTR_ERR(balloon_mnt);
895 		goto out_del_vqs;
896 	}
897 
898 	vb->vb_dev_info.migratepage = virtballoon_migratepage;
899 	vb->vb_dev_info.inode = alloc_anon_inode(balloon_mnt->mnt_sb);
900 	if (IS_ERR(vb->vb_dev_info.inode)) {
901 		err = PTR_ERR(vb->vb_dev_info.inode);
902 		kern_unmount(balloon_mnt);
903 		goto out_del_vqs;
904 	}
905 	vb->vb_dev_info.inode->i_mapping->a_ops = &balloon_aops;
906 #endif
907 	if (virtio_has_feature(vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT)) {
908 		/*
909 		 * There is always one entry reserved for cmd id, so the ring
910 		 * size needs to be at least two to report free page hints.
911 		 */
912 		if (virtqueue_get_vring_size(vb->free_page_vq) < 2) {
913 			err = -ENOSPC;
914 			goto out_del_vqs;
915 		}
916 		vb->balloon_wq = alloc_workqueue("balloon-wq",
917 					WQ_FREEZABLE | WQ_CPU_INTENSIVE, 0);
918 		if (!vb->balloon_wq) {
919 			err = -ENOMEM;
920 			goto out_del_vqs;
921 		}
922 		INIT_WORK(&vb->report_free_page_work, report_free_page_func);
923 		vb->cmd_id_received_cache = VIRTIO_BALLOON_CMD_ID_STOP;
924 		vb->cmd_id_active = cpu_to_virtio32(vb->vdev,
925 						  VIRTIO_BALLOON_CMD_ID_STOP);
926 		vb->cmd_id_stop = cpu_to_virtio32(vb->vdev,
927 						  VIRTIO_BALLOON_CMD_ID_STOP);
928 		spin_lock_init(&vb->free_page_list_lock);
929 		INIT_LIST_HEAD(&vb->free_page_list);
930 		if (virtio_has_feature(vdev, VIRTIO_BALLOON_F_PAGE_POISON)) {
931 			memset(&poison_val, PAGE_POISON, sizeof(poison_val));
932 			virtio_cwrite(vb->vdev, struct virtio_balloon_config,
933 				      poison_val, &poison_val);
934 		}
935 	}
936 	/*
937 	 * We continue to use VIRTIO_BALLOON_F_DEFLATE_ON_OOM to decide if a
938 	 * shrinker needs to be registered to relieve memory pressure.
939 	 */
940 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_DEFLATE_ON_OOM)) {
941 		err = virtio_balloon_register_shrinker(vb);
942 		if (err)
943 			goto out_del_balloon_wq;
944 	}
945 	virtio_device_ready(vdev);
946 
947 	if (towards_target(vb))
948 		virtballoon_changed(vdev);
949 	return 0;
950 
951 out_del_balloon_wq:
952 	if (virtio_has_feature(vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT))
953 		destroy_workqueue(vb->balloon_wq);
954 out_del_vqs:
955 	vdev->config->del_vqs(vdev);
956 out_free_vb:
957 	kfree(vb);
958 out:
959 	return err;
960 }
961 
962 static void remove_common(struct virtio_balloon *vb)
963 {
964 	/* There might be pages left in the balloon: free them. */
965 	while (vb->num_pages)
966 		leak_balloon(vb, vb->num_pages);
967 	update_balloon_size(vb);
968 
969 	/* Now we reset the device so we can clean up the queues. */
970 	vb->vdev->config->reset(vb->vdev);
971 
972 	vb->vdev->config->del_vqs(vb->vdev);
973 }
974 
975 static void virtballoon_remove(struct virtio_device *vdev)
976 {
977 	struct virtio_balloon *vb = vdev->priv;
978 
979 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_DEFLATE_ON_OOM))
980 		virtio_balloon_unregister_shrinker(vb);
981 	spin_lock_irq(&vb->stop_update_lock);
982 	vb->stop_update = true;
983 	spin_unlock_irq(&vb->stop_update_lock);
984 	cancel_work_sync(&vb->update_balloon_size_work);
985 	cancel_work_sync(&vb->update_balloon_stats_work);
986 
987 	if (virtio_has_feature(vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT)) {
988 		cancel_work_sync(&vb->report_free_page_work);
989 		destroy_workqueue(vb->balloon_wq);
990 	}
991 
992 	remove_common(vb);
993 #ifdef CONFIG_BALLOON_COMPACTION
994 	if (vb->vb_dev_info.inode)
995 		iput(vb->vb_dev_info.inode);
996 
997 	kern_unmount(balloon_mnt);
998 #endif
999 	kfree(vb);
1000 }
1001 
1002 #ifdef CONFIG_PM_SLEEP
1003 static int virtballoon_freeze(struct virtio_device *vdev)
1004 {
1005 	struct virtio_balloon *vb = vdev->priv;
1006 
1007 	/*
1008 	 * The workqueue is already frozen by the PM core before this
1009 	 * function is called.
1010 	 */
1011 	remove_common(vb);
1012 	return 0;
1013 }
1014 
1015 static int virtballoon_restore(struct virtio_device *vdev)
1016 {
1017 	struct virtio_balloon *vb = vdev->priv;
1018 	int ret;
1019 
1020 	ret = init_vqs(vdev->priv);
1021 	if (ret)
1022 		return ret;
1023 
1024 	virtio_device_ready(vdev);
1025 
1026 	if (towards_target(vb))
1027 		virtballoon_changed(vdev);
1028 	update_balloon_size(vb);
1029 	return 0;
1030 }
1031 #endif
1032 
1033 static int virtballoon_validate(struct virtio_device *vdev)
1034 {
1035 	if (!page_poisoning_enabled())
1036 		__virtio_clear_bit(vdev, VIRTIO_BALLOON_F_PAGE_POISON);
1037 
1038 	__virtio_clear_bit(vdev, VIRTIO_F_IOMMU_PLATFORM);
1039 	return 0;
1040 }
1041 
1042 static unsigned int features[] = {
1043 	VIRTIO_BALLOON_F_MUST_TELL_HOST,
1044 	VIRTIO_BALLOON_F_STATS_VQ,
1045 	VIRTIO_BALLOON_F_DEFLATE_ON_OOM,
1046 	VIRTIO_BALLOON_F_FREE_PAGE_HINT,
1047 	VIRTIO_BALLOON_F_PAGE_POISON,
1048 };
1049 
1050 static struct virtio_driver virtio_balloon_driver = {
1051 	.feature_table = features,
1052 	.feature_table_size = ARRAY_SIZE(features),
1053 	.driver.name =	KBUILD_MODNAME,
1054 	.driver.owner =	THIS_MODULE,
1055 	.id_table =	id_table,
1056 	.validate =	virtballoon_validate,
1057 	.probe =	virtballoon_probe,
1058 	.remove =	virtballoon_remove,
1059 	.config_changed = virtballoon_changed,
1060 #ifdef CONFIG_PM_SLEEP
1061 	.freeze	=	virtballoon_freeze,
1062 	.restore =	virtballoon_restore,
1063 #endif
1064 };
1065 
1066 module_virtio_driver(virtio_balloon_driver);
1067 MODULE_DEVICE_TABLE(virtio, id_table);
1068 MODULE_DESCRIPTION("Virtio balloon driver");
1069 MODULE_LICENSE("GPL");
1070