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