xref: /linux/drivers/virtio/virtio_balloon.c (revision 165f2d2858013253042809df082b8df7e34e86d7)
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/oom.h>
18 #include <linux/wait.h>
19 #include <linux/mm.h>
20 #include <linux/mount.h>
21 #include <linux/magic.h>
22 #include <linux/pseudo_fs.h>
23 #include <linux/page_reporting.h>
24 
25 /*
26  * Balloon device works in 4K page units.  So each page is pointed to by
27  * multiple balloon pages.  All memory counters in this driver are in balloon
28  * page units.
29  */
30 #define VIRTIO_BALLOON_PAGES_PER_PAGE (unsigned)(PAGE_SIZE >> VIRTIO_BALLOON_PFN_SHIFT)
31 #define VIRTIO_BALLOON_ARRAY_PFNS_MAX 256
32 /* Maximum number of (4k) pages to deflate on OOM notifications. */
33 #define VIRTIO_BALLOON_OOM_NR_PAGES 256
34 #define VIRTIO_BALLOON_OOM_NOTIFY_PRIORITY 80
35 
36 #define VIRTIO_BALLOON_FREE_PAGE_ALLOC_FLAG (__GFP_NORETRY | __GFP_NOWARN | \
37 					     __GFP_NOMEMALLOC)
38 /* The order of free page blocks to report to host */
39 #define VIRTIO_BALLOON_HINT_BLOCK_ORDER (MAX_ORDER - 1)
40 /* The size of a free page block in bytes */
41 #define VIRTIO_BALLOON_HINT_BLOCK_BYTES \
42 	(1 << (VIRTIO_BALLOON_HINT_BLOCK_ORDER + PAGE_SHIFT))
43 #define VIRTIO_BALLOON_HINT_BLOCK_PAGES (1 << VIRTIO_BALLOON_HINT_BLOCK_ORDER)
44 
45 #ifdef CONFIG_BALLOON_COMPACTION
46 static struct vfsmount *balloon_mnt;
47 #endif
48 
49 enum virtio_balloon_vq {
50 	VIRTIO_BALLOON_VQ_INFLATE,
51 	VIRTIO_BALLOON_VQ_DEFLATE,
52 	VIRTIO_BALLOON_VQ_STATS,
53 	VIRTIO_BALLOON_VQ_FREE_PAGE,
54 	VIRTIO_BALLOON_VQ_REPORTING,
55 	VIRTIO_BALLOON_VQ_MAX
56 };
57 
58 enum virtio_balloon_config_read {
59 	VIRTIO_BALLOON_CONFIG_READ_CMD_ID = 0,
60 };
61 
62 struct virtio_balloon {
63 	struct virtio_device *vdev;
64 	struct virtqueue *inflate_vq, *deflate_vq, *stats_vq, *free_page_vq;
65 
66 	/* Balloon's own wq for cpu-intensive work items */
67 	struct workqueue_struct *balloon_wq;
68 	/* The free page reporting work item submitted to the balloon wq */
69 	struct work_struct report_free_page_work;
70 
71 	/* The balloon servicing is delegated to a freezable workqueue. */
72 	struct work_struct update_balloon_stats_work;
73 	struct work_struct update_balloon_size_work;
74 
75 	/* Prevent updating balloon when it is being canceled. */
76 	spinlock_t stop_update_lock;
77 	bool stop_update;
78 	/* Bitmap to indicate if reading the related config fields are needed */
79 	unsigned long config_read_bitmap;
80 
81 	/* The list of allocated free pages, waiting to be given back to mm */
82 	struct list_head free_page_list;
83 	spinlock_t free_page_list_lock;
84 	/* The number of free page blocks on the above list */
85 	unsigned long num_free_page_blocks;
86 	/*
87 	 * The cmd id received from host.
88 	 * Read it via virtio_balloon_cmd_id_received to get the latest value
89 	 * sent from host.
90 	 */
91 	u32 cmd_id_received_cache;
92 	/* The cmd id that is actively in use */
93 	__virtio32 cmd_id_active;
94 	/* Buffer to store the stop sign */
95 	__virtio32 cmd_id_stop;
96 
97 	/* Waiting for host to ack the pages we released. */
98 	wait_queue_head_t acked;
99 
100 	/* Number of balloon pages we've told the Host we're not using. */
101 	unsigned int num_pages;
102 	/*
103 	 * The pages we've told the Host we're not using are enqueued
104 	 * at vb_dev_info->pages list.
105 	 * Each page on this list adds VIRTIO_BALLOON_PAGES_PER_PAGE
106 	 * to num_pages above.
107 	 */
108 	struct balloon_dev_info vb_dev_info;
109 
110 	/* Synchronize access/update to this struct virtio_balloon elements */
111 	struct mutex balloon_lock;
112 
113 	/* The array of pfns we tell the Host about. */
114 	unsigned int num_pfns;
115 	__virtio32 pfns[VIRTIO_BALLOON_ARRAY_PFNS_MAX];
116 
117 	/* Memory statistics */
118 	struct virtio_balloon_stat stats[VIRTIO_BALLOON_S_NR];
119 
120 	/* Shrinker to return free pages - VIRTIO_BALLOON_F_FREE_PAGE_HINT */
121 	struct shrinker shrinker;
122 
123 	/* OOM notifier to deflate on OOM - VIRTIO_BALLOON_F_DEFLATE_ON_OOM */
124 	struct notifier_block oom_nb;
125 
126 	/* Free page reporting device */
127 	struct virtqueue *reporting_vq;
128 	struct page_reporting_dev_info pr_dev_info;
129 };
130 
131 static struct virtio_device_id id_table[] = {
132 	{ VIRTIO_ID_BALLOON, VIRTIO_DEV_ANY_ID },
133 	{ 0 },
134 };
135 
136 static u32 page_to_balloon_pfn(struct page *page)
137 {
138 	unsigned long pfn = page_to_pfn(page);
139 
140 	BUILD_BUG_ON(PAGE_SHIFT < VIRTIO_BALLOON_PFN_SHIFT);
141 	/* Convert pfn from Linux page size to balloon page size. */
142 	return pfn * VIRTIO_BALLOON_PAGES_PER_PAGE;
143 }
144 
145 static void balloon_ack(struct virtqueue *vq)
146 {
147 	struct virtio_balloon *vb = vq->vdev->priv;
148 
149 	wake_up(&vb->acked);
150 }
151 
152 static void tell_host(struct virtio_balloon *vb, struct virtqueue *vq)
153 {
154 	struct scatterlist sg;
155 	unsigned int len;
156 
157 	sg_init_one(&sg, vb->pfns, sizeof(vb->pfns[0]) * vb->num_pfns);
158 
159 	/* We should always be able to add one buffer to an empty queue. */
160 	virtqueue_add_outbuf(vq, &sg, 1, vb, GFP_KERNEL);
161 	virtqueue_kick(vq);
162 
163 	/* When host has read buffer, this completes via balloon_ack */
164 	wait_event(vb->acked, virtqueue_get_buf(vq, &len));
165 
166 }
167 
168 static int virtballoon_free_page_report(struct page_reporting_dev_info *pr_dev_info,
169 				   struct scatterlist *sg, unsigned int nents)
170 {
171 	struct virtio_balloon *vb =
172 		container_of(pr_dev_info, struct virtio_balloon, pr_dev_info);
173 	struct virtqueue *vq = vb->reporting_vq;
174 	unsigned int unused, err;
175 
176 	/* We should always be able to add these buffers to an empty queue. */
177 	err = virtqueue_add_inbuf(vq, sg, nents, vb, GFP_NOWAIT | __GFP_NOWARN);
178 
179 	/*
180 	 * In the extremely unlikely case that something has occurred and we
181 	 * are able to trigger an error we will simply display a warning
182 	 * and exit without actually processing the pages.
183 	 */
184 	if (WARN_ON_ONCE(err))
185 		return err;
186 
187 	virtqueue_kick(vq);
188 
189 	/* When host has read buffer, this completes via balloon_ack */
190 	wait_event(vb->acked, virtqueue_get_buf(vq, &unused));
191 
192 	return 0;
193 }
194 
195 static void set_page_pfns(struct virtio_balloon *vb,
196 			  __virtio32 pfns[], struct page *page)
197 {
198 	unsigned int i;
199 
200 	BUILD_BUG_ON(VIRTIO_BALLOON_PAGES_PER_PAGE > VIRTIO_BALLOON_ARRAY_PFNS_MAX);
201 
202 	/*
203 	 * Set balloon pfns pointing at this page.
204 	 * Note that the first pfn points at start of the page.
205 	 */
206 	for (i = 0; i < VIRTIO_BALLOON_PAGES_PER_PAGE; i++)
207 		pfns[i] = cpu_to_virtio32(vb->vdev,
208 					  page_to_balloon_pfn(page) + i);
209 }
210 
211 static unsigned fill_balloon(struct virtio_balloon *vb, size_t num)
212 {
213 	unsigned num_allocated_pages;
214 	unsigned num_pfns;
215 	struct page *page;
216 	LIST_HEAD(pages);
217 
218 	/* We can only do one array worth at a time. */
219 	num = min(num, ARRAY_SIZE(vb->pfns));
220 
221 	for (num_pfns = 0; num_pfns < num;
222 	     num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) {
223 		struct page *page = balloon_page_alloc();
224 
225 		if (!page) {
226 			dev_info_ratelimited(&vb->vdev->dev,
227 					     "Out of puff! Can't get %u pages\n",
228 					     VIRTIO_BALLOON_PAGES_PER_PAGE);
229 			/* Sleep for at least 1/5 of a second before retry. */
230 			msleep(200);
231 			break;
232 		}
233 
234 		balloon_page_push(&pages, page);
235 	}
236 
237 	mutex_lock(&vb->balloon_lock);
238 
239 	vb->num_pfns = 0;
240 
241 	while ((page = balloon_page_pop(&pages))) {
242 		balloon_page_enqueue(&vb->vb_dev_info, page);
243 
244 		set_page_pfns(vb, vb->pfns + vb->num_pfns, page);
245 		vb->num_pages += VIRTIO_BALLOON_PAGES_PER_PAGE;
246 		if (!virtio_has_feature(vb->vdev,
247 					VIRTIO_BALLOON_F_DEFLATE_ON_OOM))
248 			adjust_managed_page_count(page, -1);
249 		vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE;
250 	}
251 
252 	num_allocated_pages = vb->num_pfns;
253 	/* Did we get any? */
254 	if (vb->num_pfns != 0)
255 		tell_host(vb, vb->inflate_vq);
256 	mutex_unlock(&vb->balloon_lock);
257 
258 	return num_allocated_pages;
259 }
260 
261 static void release_pages_balloon(struct virtio_balloon *vb,
262 				 struct list_head *pages)
263 {
264 	struct page *page, *next;
265 
266 	list_for_each_entry_safe(page, next, pages, lru) {
267 		if (!virtio_has_feature(vb->vdev,
268 					VIRTIO_BALLOON_F_DEFLATE_ON_OOM))
269 			adjust_managed_page_count(page, 1);
270 		list_del(&page->lru);
271 		put_page(page); /* balloon reference */
272 	}
273 }
274 
275 static unsigned leak_balloon(struct virtio_balloon *vb, size_t num)
276 {
277 	unsigned num_freed_pages;
278 	struct page *page;
279 	struct balloon_dev_info *vb_dev_info = &vb->vb_dev_info;
280 	LIST_HEAD(pages);
281 
282 	/* We can only do one array worth at a time. */
283 	num = min(num, ARRAY_SIZE(vb->pfns));
284 
285 	mutex_lock(&vb->balloon_lock);
286 	/* We can't release more pages than taken */
287 	num = min(num, (size_t)vb->num_pages);
288 	for (vb->num_pfns = 0; vb->num_pfns < num;
289 	     vb->num_pfns += VIRTIO_BALLOON_PAGES_PER_PAGE) {
290 		page = balloon_page_dequeue(vb_dev_info);
291 		if (!page)
292 			break;
293 		set_page_pfns(vb, vb->pfns + vb->num_pfns, page);
294 		list_add(&page->lru, &pages);
295 		vb->num_pages -= VIRTIO_BALLOON_PAGES_PER_PAGE;
296 	}
297 
298 	num_freed_pages = vb->num_pfns;
299 	/*
300 	 * Note that if
301 	 * virtio_has_feature(vdev, VIRTIO_BALLOON_F_MUST_TELL_HOST);
302 	 * is true, we *have* to do it in this order
303 	 */
304 	if (vb->num_pfns != 0)
305 		tell_host(vb, vb->deflate_vq);
306 	release_pages_balloon(vb, &pages);
307 	mutex_unlock(&vb->balloon_lock);
308 	return num_freed_pages;
309 }
310 
311 static inline void update_stat(struct virtio_balloon *vb, int idx,
312 			       u16 tag, u64 val)
313 {
314 	BUG_ON(idx >= VIRTIO_BALLOON_S_NR);
315 	vb->stats[idx].tag = cpu_to_virtio16(vb->vdev, tag);
316 	vb->stats[idx].val = cpu_to_virtio64(vb->vdev, val);
317 }
318 
319 #define pages_to_bytes(x) ((u64)(x) << PAGE_SHIFT)
320 
321 static unsigned int update_balloon_stats(struct virtio_balloon *vb)
322 {
323 	unsigned long events[NR_VM_EVENT_ITEMS];
324 	struct sysinfo i;
325 	unsigned int idx = 0;
326 	long available;
327 	unsigned long caches;
328 
329 	all_vm_events(events);
330 	si_meminfo(&i);
331 
332 	available = si_mem_available();
333 	caches = global_node_page_state(NR_FILE_PAGES);
334 
335 #ifdef CONFIG_VM_EVENT_COUNTERS
336 	update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_IN,
337 				pages_to_bytes(events[PSWPIN]));
338 	update_stat(vb, idx++, VIRTIO_BALLOON_S_SWAP_OUT,
339 				pages_to_bytes(events[PSWPOUT]));
340 	update_stat(vb, idx++, VIRTIO_BALLOON_S_MAJFLT, events[PGMAJFAULT]);
341 	update_stat(vb, idx++, VIRTIO_BALLOON_S_MINFLT, events[PGFAULT]);
342 #ifdef CONFIG_HUGETLB_PAGE
343 	update_stat(vb, idx++, VIRTIO_BALLOON_S_HTLB_PGALLOC,
344 		    events[HTLB_BUDDY_PGALLOC]);
345 	update_stat(vb, idx++, VIRTIO_BALLOON_S_HTLB_PGFAIL,
346 		    events[HTLB_BUDDY_PGALLOC_FAIL]);
347 #endif
348 #endif
349 	update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMFREE,
350 				pages_to_bytes(i.freeram));
351 	update_stat(vb, idx++, VIRTIO_BALLOON_S_MEMTOT,
352 				pages_to_bytes(i.totalram));
353 	update_stat(vb, idx++, VIRTIO_BALLOON_S_AVAIL,
354 				pages_to_bytes(available));
355 	update_stat(vb, idx++, VIRTIO_BALLOON_S_CACHES,
356 				pages_to_bytes(caches));
357 
358 	return idx;
359 }
360 
361 /*
362  * While most virtqueues communicate guest-initiated requests to the hypervisor,
363  * the stats queue operates in reverse.  The driver initializes the virtqueue
364  * with a single buffer.  From that point forward, all conversations consist of
365  * a hypervisor request (a call to this function) which directs us to refill
366  * the virtqueue with a fresh stats buffer.  Since stats collection can sleep,
367  * we delegate the job to a freezable workqueue that will do the actual work via
368  * stats_handle_request().
369  */
370 static void stats_request(struct virtqueue *vq)
371 {
372 	struct virtio_balloon *vb = vq->vdev->priv;
373 
374 	spin_lock(&vb->stop_update_lock);
375 	if (!vb->stop_update)
376 		queue_work(system_freezable_wq, &vb->update_balloon_stats_work);
377 	spin_unlock(&vb->stop_update_lock);
378 }
379 
380 static void stats_handle_request(struct virtio_balloon *vb)
381 {
382 	struct virtqueue *vq;
383 	struct scatterlist sg;
384 	unsigned int len, num_stats;
385 
386 	num_stats = update_balloon_stats(vb);
387 
388 	vq = vb->stats_vq;
389 	if (!virtqueue_get_buf(vq, &len))
390 		return;
391 	sg_init_one(&sg, vb->stats, sizeof(vb->stats[0]) * num_stats);
392 	virtqueue_add_outbuf(vq, &sg, 1, vb, GFP_KERNEL);
393 	virtqueue_kick(vq);
394 }
395 
396 static inline s64 towards_target(struct virtio_balloon *vb)
397 {
398 	s64 target;
399 	u32 num_pages;
400 
401 	virtio_cread(vb->vdev, struct virtio_balloon_config, num_pages,
402 		     &num_pages);
403 
404 	/* Legacy balloon config space is LE, unlike all other devices. */
405 	if (!virtio_has_feature(vb->vdev, VIRTIO_F_VERSION_1))
406 		num_pages = le32_to_cpu((__force __le32)num_pages);
407 
408 	target = num_pages;
409 	return target - vb->num_pages;
410 }
411 
412 /* Gives back @num_to_return blocks of free pages to mm. */
413 static unsigned long return_free_pages_to_mm(struct virtio_balloon *vb,
414 					     unsigned long num_to_return)
415 {
416 	struct page *page;
417 	unsigned long num_returned;
418 
419 	spin_lock_irq(&vb->free_page_list_lock);
420 	for (num_returned = 0; num_returned < num_to_return; num_returned++) {
421 		page = balloon_page_pop(&vb->free_page_list);
422 		if (!page)
423 			break;
424 		free_pages((unsigned long)page_address(page),
425 			   VIRTIO_BALLOON_HINT_BLOCK_ORDER);
426 	}
427 	vb->num_free_page_blocks -= num_returned;
428 	spin_unlock_irq(&vb->free_page_list_lock);
429 
430 	return num_returned;
431 }
432 
433 static void virtio_balloon_queue_free_page_work(struct virtio_balloon *vb)
434 {
435 	if (!virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT))
436 		return;
437 
438 	/* No need to queue the work if the bit was already set. */
439 	if (test_and_set_bit(VIRTIO_BALLOON_CONFIG_READ_CMD_ID,
440 			     &vb->config_read_bitmap))
441 		return;
442 
443 	queue_work(vb->balloon_wq, &vb->report_free_page_work);
444 }
445 
446 static void virtballoon_changed(struct virtio_device *vdev)
447 {
448 	struct virtio_balloon *vb = vdev->priv;
449 	unsigned long flags;
450 
451 	spin_lock_irqsave(&vb->stop_update_lock, flags);
452 	if (!vb->stop_update) {
453 		queue_work(system_freezable_wq,
454 			   &vb->update_balloon_size_work);
455 		virtio_balloon_queue_free_page_work(vb);
456 	}
457 	spin_unlock_irqrestore(&vb->stop_update_lock, flags);
458 }
459 
460 static void update_balloon_size(struct virtio_balloon *vb)
461 {
462 	u32 actual = vb->num_pages;
463 
464 	/* Legacy balloon config space is LE, unlike all other devices. */
465 	if (!virtio_has_feature(vb->vdev, VIRTIO_F_VERSION_1))
466 		actual = (__force u32)cpu_to_le32(actual);
467 
468 	virtio_cwrite(vb->vdev, struct virtio_balloon_config, actual,
469 		      &actual);
470 }
471 
472 static void update_balloon_stats_func(struct work_struct *work)
473 {
474 	struct virtio_balloon *vb;
475 
476 	vb = container_of(work, struct virtio_balloon,
477 			  update_balloon_stats_work);
478 	stats_handle_request(vb);
479 }
480 
481 static void update_balloon_size_func(struct work_struct *work)
482 {
483 	struct virtio_balloon *vb;
484 	s64 diff;
485 
486 	vb = container_of(work, struct virtio_balloon,
487 			  update_balloon_size_work);
488 	diff = towards_target(vb);
489 
490 	if (!diff)
491 		return;
492 
493 	if (diff > 0)
494 		diff -= fill_balloon(vb, diff);
495 	else
496 		diff += leak_balloon(vb, -diff);
497 	update_balloon_size(vb);
498 
499 	if (diff)
500 		queue_work(system_freezable_wq, work);
501 }
502 
503 static int init_vqs(struct virtio_balloon *vb)
504 {
505 	struct virtqueue *vqs[VIRTIO_BALLOON_VQ_MAX];
506 	vq_callback_t *callbacks[VIRTIO_BALLOON_VQ_MAX];
507 	const char *names[VIRTIO_BALLOON_VQ_MAX];
508 	int err;
509 
510 	/*
511 	 * Inflateq and deflateq are used unconditionally. The names[]
512 	 * will be NULL if the related feature is not enabled, which will
513 	 * cause no allocation for the corresponding virtqueue in find_vqs.
514 	 */
515 	callbacks[VIRTIO_BALLOON_VQ_INFLATE] = balloon_ack;
516 	names[VIRTIO_BALLOON_VQ_INFLATE] = "inflate";
517 	callbacks[VIRTIO_BALLOON_VQ_DEFLATE] = balloon_ack;
518 	names[VIRTIO_BALLOON_VQ_DEFLATE] = "deflate";
519 	callbacks[VIRTIO_BALLOON_VQ_STATS] = NULL;
520 	names[VIRTIO_BALLOON_VQ_STATS] = NULL;
521 	callbacks[VIRTIO_BALLOON_VQ_FREE_PAGE] = NULL;
522 	names[VIRTIO_BALLOON_VQ_FREE_PAGE] = NULL;
523 	names[VIRTIO_BALLOON_VQ_REPORTING] = NULL;
524 
525 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ)) {
526 		names[VIRTIO_BALLOON_VQ_STATS] = "stats";
527 		callbacks[VIRTIO_BALLOON_VQ_STATS] = stats_request;
528 	}
529 
530 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT)) {
531 		names[VIRTIO_BALLOON_VQ_FREE_PAGE] = "free_page_vq";
532 		callbacks[VIRTIO_BALLOON_VQ_FREE_PAGE] = NULL;
533 	}
534 
535 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_REPORTING)) {
536 		names[VIRTIO_BALLOON_VQ_REPORTING] = "reporting_vq";
537 		callbacks[VIRTIO_BALLOON_VQ_REPORTING] = balloon_ack;
538 	}
539 
540 	err = vb->vdev->config->find_vqs(vb->vdev, VIRTIO_BALLOON_VQ_MAX,
541 					 vqs, callbacks, names, NULL, NULL);
542 	if (err)
543 		return err;
544 
545 	vb->inflate_vq = vqs[VIRTIO_BALLOON_VQ_INFLATE];
546 	vb->deflate_vq = vqs[VIRTIO_BALLOON_VQ_DEFLATE];
547 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_STATS_VQ)) {
548 		struct scatterlist sg;
549 		unsigned int num_stats;
550 		vb->stats_vq = vqs[VIRTIO_BALLOON_VQ_STATS];
551 
552 		/*
553 		 * Prime this virtqueue with one buffer so the hypervisor can
554 		 * use it to signal us later (it can't be broken yet!).
555 		 */
556 		num_stats = update_balloon_stats(vb);
557 
558 		sg_init_one(&sg, vb->stats, sizeof(vb->stats[0]) * num_stats);
559 		err = virtqueue_add_outbuf(vb->stats_vq, &sg, 1, vb,
560 					   GFP_KERNEL);
561 		if (err) {
562 			dev_warn(&vb->vdev->dev, "%s: add stat_vq failed\n",
563 				 __func__);
564 			return err;
565 		}
566 		virtqueue_kick(vb->stats_vq);
567 	}
568 
569 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT))
570 		vb->free_page_vq = vqs[VIRTIO_BALLOON_VQ_FREE_PAGE];
571 
572 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_REPORTING))
573 		vb->reporting_vq = vqs[VIRTIO_BALLOON_VQ_REPORTING];
574 
575 	return 0;
576 }
577 
578 static u32 virtio_balloon_cmd_id_received(struct virtio_balloon *vb)
579 {
580 	if (test_and_clear_bit(VIRTIO_BALLOON_CONFIG_READ_CMD_ID,
581 			       &vb->config_read_bitmap))
582 		virtio_cread(vb->vdev, struct virtio_balloon_config,
583 			     free_page_hint_cmd_id,
584 			     &vb->cmd_id_received_cache);
585 
586 	return vb->cmd_id_received_cache;
587 }
588 
589 static int send_cmd_id_start(struct virtio_balloon *vb)
590 {
591 	struct scatterlist sg;
592 	struct virtqueue *vq = vb->free_page_vq;
593 	int err, unused;
594 
595 	/* Detach all the used buffers from the vq */
596 	while (virtqueue_get_buf(vq, &unused))
597 		;
598 
599 	vb->cmd_id_active = virtio32_to_cpu(vb->vdev,
600 					virtio_balloon_cmd_id_received(vb));
601 	sg_init_one(&sg, &vb->cmd_id_active, sizeof(vb->cmd_id_active));
602 	err = virtqueue_add_outbuf(vq, &sg, 1, &vb->cmd_id_active, GFP_KERNEL);
603 	if (!err)
604 		virtqueue_kick(vq);
605 	return err;
606 }
607 
608 static int send_cmd_id_stop(struct virtio_balloon *vb)
609 {
610 	struct scatterlist sg;
611 	struct virtqueue *vq = vb->free_page_vq;
612 	int err, unused;
613 
614 	/* Detach all the used buffers from the vq */
615 	while (virtqueue_get_buf(vq, &unused))
616 		;
617 
618 	sg_init_one(&sg, &vb->cmd_id_stop, sizeof(vb->cmd_id_stop));
619 	err = virtqueue_add_outbuf(vq, &sg, 1, &vb->cmd_id_stop, GFP_KERNEL);
620 	if (!err)
621 		virtqueue_kick(vq);
622 	return err;
623 }
624 
625 static int get_free_page_and_send(struct virtio_balloon *vb)
626 {
627 	struct virtqueue *vq = vb->free_page_vq;
628 	struct page *page;
629 	struct scatterlist sg;
630 	int err, unused;
631 	void *p;
632 
633 	/* Detach all the used buffers from the vq */
634 	while (virtqueue_get_buf(vq, &unused))
635 		;
636 
637 	page = alloc_pages(VIRTIO_BALLOON_FREE_PAGE_ALLOC_FLAG,
638 			   VIRTIO_BALLOON_HINT_BLOCK_ORDER);
639 	/*
640 	 * When the allocation returns NULL, it indicates that we have got all
641 	 * the possible free pages, so return -EINTR to stop.
642 	 */
643 	if (!page)
644 		return -EINTR;
645 
646 	p = page_address(page);
647 	sg_init_one(&sg, p, VIRTIO_BALLOON_HINT_BLOCK_BYTES);
648 	/* There is always 1 entry reserved for the cmd id to use. */
649 	if (vq->num_free > 1) {
650 		err = virtqueue_add_inbuf(vq, &sg, 1, p, GFP_KERNEL);
651 		if (unlikely(err)) {
652 			free_pages((unsigned long)p,
653 				   VIRTIO_BALLOON_HINT_BLOCK_ORDER);
654 			return err;
655 		}
656 		virtqueue_kick(vq);
657 		spin_lock_irq(&vb->free_page_list_lock);
658 		balloon_page_push(&vb->free_page_list, page);
659 		vb->num_free_page_blocks++;
660 		spin_unlock_irq(&vb->free_page_list_lock);
661 	} else {
662 		/*
663 		 * The vq has no available entry to add this page block, so
664 		 * just free it.
665 		 */
666 		free_pages((unsigned long)p, VIRTIO_BALLOON_HINT_BLOCK_ORDER);
667 	}
668 
669 	return 0;
670 }
671 
672 static int send_free_pages(struct virtio_balloon *vb)
673 {
674 	int err;
675 	u32 cmd_id_active;
676 
677 	while (1) {
678 		/*
679 		 * If a stop id or a new cmd id was just received from host,
680 		 * stop the reporting.
681 		 */
682 		cmd_id_active = virtio32_to_cpu(vb->vdev, vb->cmd_id_active);
683 		if (unlikely(cmd_id_active !=
684 			     virtio_balloon_cmd_id_received(vb)))
685 			break;
686 
687 		/*
688 		 * The free page blocks are allocated and sent to host one by
689 		 * one.
690 		 */
691 		err = get_free_page_and_send(vb);
692 		if (err == -EINTR)
693 			break;
694 		else if (unlikely(err))
695 			return err;
696 	}
697 
698 	return 0;
699 }
700 
701 static void virtio_balloon_report_free_page(struct virtio_balloon *vb)
702 {
703 	int err;
704 	struct device *dev = &vb->vdev->dev;
705 
706 	/* Start by sending the received cmd id to host with an outbuf. */
707 	err = send_cmd_id_start(vb);
708 	if (unlikely(err))
709 		dev_err(dev, "Failed to send a start id, err = %d\n", err);
710 
711 	err = send_free_pages(vb);
712 	if (unlikely(err))
713 		dev_err(dev, "Failed to send a free page, err = %d\n", err);
714 
715 	/* End by sending a stop id to host with an outbuf. */
716 	err = send_cmd_id_stop(vb);
717 	if (unlikely(err))
718 		dev_err(dev, "Failed to send a stop id, err = %d\n", err);
719 }
720 
721 static void report_free_page_func(struct work_struct *work)
722 {
723 	struct virtio_balloon *vb = container_of(work, struct virtio_balloon,
724 						 report_free_page_work);
725 	u32 cmd_id_received;
726 
727 	cmd_id_received = virtio_balloon_cmd_id_received(vb);
728 	if (cmd_id_received == VIRTIO_BALLOON_CMD_ID_DONE) {
729 		/* Pass ULONG_MAX to give back all the free pages */
730 		return_free_pages_to_mm(vb, ULONG_MAX);
731 	} else if (cmd_id_received != VIRTIO_BALLOON_CMD_ID_STOP &&
732 		   cmd_id_received !=
733 		   virtio32_to_cpu(vb->vdev, vb->cmd_id_active)) {
734 		virtio_balloon_report_free_page(vb);
735 	}
736 }
737 
738 #ifdef CONFIG_BALLOON_COMPACTION
739 /*
740  * virtballoon_migratepage - perform the balloon page migration on behalf of
741  *			     a compation thread.     (called under page lock)
742  * @vb_dev_info: the balloon device
743  * @newpage: page that will replace the isolated page after migration finishes.
744  * @page   : the isolated (old) page that is about to be migrated to newpage.
745  * @mode   : compaction mode -- not used for balloon page migration.
746  *
747  * After a ballooned page gets isolated by compaction procedures, this is the
748  * function that performs the page migration on behalf of a compaction thread
749  * The page migration for virtio balloon is done in a simple swap fashion which
750  * follows these two macro steps:
751  *  1) insert newpage into vb->pages list and update the host about it;
752  *  2) update the host about the old page removed from vb->pages list;
753  *
754  * This function preforms the balloon page migration task.
755  * Called through balloon_mapping->a_ops->migratepage
756  */
757 static int virtballoon_migratepage(struct balloon_dev_info *vb_dev_info,
758 		struct page *newpage, struct page *page, enum migrate_mode mode)
759 {
760 	struct virtio_balloon *vb = container_of(vb_dev_info,
761 			struct virtio_balloon, vb_dev_info);
762 	unsigned long flags;
763 
764 	/*
765 	 * In order to avoid lock contention while migrating pages concurrently
766 	 * to leak_balloon() or fill_balloon() we just give up the balloon_lock
767 	 * this turn, as it is easier to retry the page migration later.
768 	 * This also prevents fill_balloon() getting stuck into a mutex
769 	 * recursion in the case it ends up triggering memory compaction
770 	 * while it is attempting to inflate the ballon.
771 	 */
772 	if (!mutex_trylock(&vb->balloon_lock))
773 		return -EAGAIN;
774 
775 	get_page(newpage); /* balloon reference */
776 
777 	/*
778 	  * When we migrate a page to a different zone and adjusted the
779 	  * managed page count when inflating, we have to fixup the count of
780 	  * both involved zones.
781 	  */
782 	if (!virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_DEFLATE_ON_OOM) &&
783 	    page_zone(page) != page_zone(newpage)) {
784 		adjust_managed_page_count(page, 1);
785 		adjust_managed_page_count(newpage, -1);
786 	}
787 
788 	/* balloon's page migration 1st step  -- inflate "newpage" */
789 	spin_lock_irqsave(&vb_dev_info->pages_lock, flags);
790 	balloon_page_insert(vb_dev_info, newpage);
791 	vb_dev_info->isolated_pages--;
792 	__count_vm_event(BALLOON_MIGRATE);
793 	spin_unlock_irqrestore(&vb_dev_info->pages_lock, flags);
794 	vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE;
795 	set_page_pfns(vb, vb->pfns, newpage);
796 	tell_host(vb, vb->inflate_vq);
797 
798 	/* balloon's page migration 2nd step -- deflate "page" */
799 	spin_lock_irqsave(&vb_dev_info->pages_lock, flags);
800 	balloon_page_delete(page);
801 	spin_unlock_irqrestore(&vb_dev_info->pages_lock, flags);
802 	vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE;
803 	set_page_pfns(vb, vb->pfns, page);
804 	tell_host(vb, vb->deflate_vq);
805 
806 	mutex_unlock(&vb->balloon_lock);
807 
808 	put_page(page); /* balloon reference */
809 
810 	return MIGRATEPAGE_SUCCESS;
811 }
812 
813 static int balloon_init_fs_context(struct fs_context *fc)
814 {
815 	return init_pseudo(fc, BALLOON_KVM_MAGIC) ? 0 : -ENOMEM;
816 }
817 
818 static struct file_system_type balloon_fs = {
819 	.name           = "balloon-kvm",
820 	.init_fs_context = balloon_init_fs_context,
821 	.kill_sb        = kill_anon_super,
822 };
823 
824 #endif /* CONFIG_BALLOON_COMPACTION */
825 
826 static unsigned long shrink_free_pages(struct virtio_balloon *vb,
827 				       unsigned long pages_to_free)
828 {
829 	unsigned long blocks_to_free, blocks_freed;
830 
831 	pages_to_free = round_up(pages_to_free,
832 				 VIRTIO_BALLOON_HINT_BLOCK_PAGES);
833 	blocks_to_free = pages_to_free / VIRTIO_BALLOON_HINT_BLOCK_PAGES;
834 	blocks_freed = return_free_pages_to_mm(vb, blocks_to_free);
835 
836 	return blocks_freed * VIRTIO_BALLOON_HINT_BLOCK_PAGES;
837 }
838 
839 static unsigned long virtio_balloon_shrinker_scan(struct shrinker *shrinker,
840 						  struct shrink_control *sc)
841 {
842 	struct virtio_balloon *vb = container_of(shrinker,
843 					struct virtio_balloon, shrinker);
844 
845 	return shrink_free_pages(vb, sc->nr_to_scan);
846 }
847 
848 static unsigned long virtio_balloon_shrinker_count(struct shrinker *shrinker,
849 						   struct shrink_control *sc)
850 {
851 	struct virtio_balloon *vb = container_of(shrinker,
852 					struct virtio_balloon, shrinker);
853 
854 	return vb->num_free_page_blocks * VIRTIO_BALLOON_HINT_BLOCK_PAGES;
855 }
856 
857 static int virtio_balloon_oom_notify(struct notifier_block *nb,
858 				     unsigned long dummy, void *parm)
859 {
860 	struct virtio_balloon *vb = container_of(nb,
861 						 struct virtio_balloon, oom_nb);
862 	unsigned long *freed = parm;
863 
864 	*freed += leak_balloon(vb, VIRTIO_BALLOON_OOM_NR_PAGES) /
865 		  VIRTIO_BALLOON_PAGES_PER_PAGE;
866 	update_balloon_size(vb);
867 
868 	return NOTIFY_OK;
869 }
870 
871 static void virtio_balloon_unregister_shrinker(struct virtio_balloon *vb)
872 {
873 	unregister_shrinker(&vb->shrinker);
874 }
875 
876 static int virtio_balloon_register_shrinker(struct virtio_balloon *vb)
877 {
878 	vb->shrinker.scan_objects = virtio_balloon_shrinker_scan;
879 	vb->shrinker.count_objects = virtio_balloon_shrinker_count;
880 	vb->shrinker.seeks = DEFAULT_SEEKS;
881 
882 	return register_shrinker(&vb->shrinker);
883 }
884 
885 static int virtballoon_probe(struct virtio_device *vdev)
886 {
887 	struct virtio_balloon *vb;
888 	int err;
889 
890 	if (!vdev->config->get) {
891 		dev_err(&vdev->dev, "%s failure: config access disabled\n",
892 			__func__);
893 		return -EINVAL;
894 	}
895 
896 	vdev->priv = vb = kzalloc(sizeof(*vb), GFP_KERNEL);
897 	if (!vb) {
898 		err = -ENOMEM;
899 		goto out;
900 	}
901 
902 	INIT_WORK(&vb->update_balloon_stats_work, update_balloon_stats_func);
903 	INIT_WORK(&vb->update_balloon_size_work, update_balloon_size_func);
904 	spin_lock_init(&vb->stop_update_lock);
905 	mutex_init(&vb->balloon_lock);
906 	init_waitqueue_head(&vb->acked);
907 	vb->vdev = vdev;
908 
909 	balloon_devinfo_init(&vb->vb_dev_info);
910 
911 	err = init_vqs(vb);
912 	if (err)
913 		goto out_free_vb;
914 
915 #ifdef CONFIG_BALLOON_COMPACTION
916 	balloon_mnt = kern_mount(&balloon_fs);
917 	if (IS_ERR(balloon_mnt)) {
918 		err = PTR_ERR(balloon_mnt);
919 		goto out_del_vqs;
920 	}
921 
922 	vb->vb_dev_info.migratepage = virtballoon_migratepage;
923 	vb->vb_dev_info.inode = alloc_anon_inode(balloon_mnt->mnt_sb);
924 	if (IS_ERR(vb->vb_dev_info.inode)) {
925 		err = PTR_ERR(vb->vb_dev_info.inode);
926 		goto out_kern_unmount;
927 	}
928 	vb->vb_dev_info.inode->i_mapping->a_ops = &balloon_aops;
929 #endif
930 	if (virtio_has_feature(vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT)) {
931 		/*
932 		 * There is always one entry reserved for cmd id, so the ring
933 		 * size needs to be at least two to report free page hints.
934 		 */
935 		if (virtqueue_get_vring_size(vb->free_page_vq) < 2) {
936 			err = -ENOSPC;
937 			goto out_iput;
938 		}
939 		vb->balloon_wq = alloc_workqueue("balloon-wq",
940 					WQ_FREEZABLE | WQ_CPU_INTENSIVE, 0);
941 		if (!vb->balloon_wq) {
942 			err = -ENOMEM;
943 			goto out_iput;
944 		}
945 		INIT_WORK(&vb->report_free_page_work, report_free_page_func);
946 		vb->cmd_id_received_cache = VIRTIO_BALLOON_CMD_ID_STOP;
947 		vb->cmd_id_active = cpu_to_virtio32(vb->vdev,
948 						  VIRTIO_BALLOON_CMD_ID_STOP);
949 		vb->cmd_id_stop = cpu_to_virtio32(vb->vdev,
950 						  VIRTIO_BALLOON_CMD_ID_STOP);
951 		spin_lock_init(&vb->free_page_list_lock);
952 		INIT_LIST_HEAD(&vb->free_page_list);
953 		/*
954 		 * We're allowed to reuse any free pages, even if they are
955 		 * still to be processed by the host.
956 		 */
957 		err = virtio_balloon_register_shrinker(vb);
958 		if (err)
959 			goto out_del_balloon_wq;
960 	}
961 
962 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_DEFLATE_ON_OOM)) {
963 		vb->oom_nb.notifier_call = virtio_balloon_oom_notify;
964 		vb->oom_nb.priority = VIRTIO_BALLOON_OOM_NOTIFY_PRIORITY;
965 		err = register_oom_notifier(&vb->oom_nb);
966 		if (err < 0)
967 			goto out_unregister_shrinker;
968 	}
969 
970 	if (virtio_has_feature(vdev, VIRTIO_BALLOON_F_PAGE_POISON)) {
971 		/* Start with poison val of 0 representing general init */
972 		__u32 poison_val = 0;
973 
974 		/*
975 		 * Let the hypervisor know that we are expecting a
976 		 * specific value to be written back in balloon pages.
977 		 */
978 		if (!want_init_on_free())
979 			memset(&poison_val, PAGE_POISON, sizeof(poison_val));
980 
981 		virtio_cwrite(vb->vdev, struct virtio_balloon_config,
982 			      poison_val, &poison_val);
983 	}
984 
985 	vb->pr_dev_info.report = virtballoon_free_page_report;
986 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_REPORTING)) {
987 		unsigned int capacity;
988 
989 		capacity = virtqueue_get_vring_size(vb->reporting_vq);
990 		if (capacity < PAGE_REPORTING_CAPACITY) {
991 			err = -ENOSPC;
992 			goto out_unregister_oom;
993 		}
994 
995 		err = page_reporting_register(&vb->pr_dev_info);
996 		if (err)
997 			goto out_unregister_oom;
998 	}
999 
1000 	virtio_device_ready(vdev);
1001 
1002 	if (towards_target(vb))
1003 		virtballoon_changed(vdev);
1004 	return 0;
1005 
1006 out_unregister_oom:
1007 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_DEFLATE_ON_OOM))
1008 		unregister_oom_notifier(&vb->oom_nb);
1009 out_unregister_shrinker:
1010 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT))
1011 		virtio_balloon_unregister_shrinker(vb);
1012 out_del_balloon_wq:
1013 	if (virtio_has_feature(vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT))
1014 		destroy_workqueue(vb->balloon_wq);
1015 out_iput:
1016 #ifdef CONFIG_BALLOON_COMPACTION
1017 	iput(vb->vb_dev_info.inode);
1018 out_kern_unmount:
1019 	kern_unmount(balloon_mnt);
1020 out_del_vqs:
1021 #endif
1022 	vdev->config->del_vqs(vdev);
1023 out_free_vb:
1024 	kfree(vb);
1025 out:
1026 	return err;
1027 }
1028 
1029 static void remove_common(struct virtio_balloon *vb)
1030 {
1031 	/* There might be pages left in the balloon: free them. */
1032 	while (vb->num_pages)
1033 		leak_balloon(vb, vb->num_pages);
1034 	update_balloon_size(vb);
1035 
1036 	/* There might be free pages that are being reported: release them. */
1037 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT))
1038 		return_free_pages_to_mm(vb, ULONG_MAX);
1039 
1040 	/* Now we reset the device so we can clean up the queues. */
1041 	vb->vdev->config->reset(vb->vdev);
1042 
1043 	vb->vdev->config->del_vqs(vb->vdev);
1044 }
1045 
1046 static void virtballoon_remove(struct virtio_device *vdev)
1047 {
1048 	struct virtio_balloon *vb = vdev->priv;
1049 
1050 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_REPORTING))
1051 		page_reporting_unregister(&vb->pr_dev_info);
1052 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_DEFLATE_ON_OOM))
1053 		unregister_oom_notifier(&vb->oom_nb);
1054 	if (virtio_has_feature(vb->vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT))
1055 		virtio_balloon_unregister_shrinker(vb);
1056 	spin_lock_irq(&vb->stop_update_lock);
1057 	vb->stop_update = true;
1058 	spin_unlock_irq(&vb->stop_update_lock);
1059 	cancel_work_sync(&vb->update_balloon_size_work);
1060 	cancel_work_sync(&vb->update_balloon_stats_work);
1061 
1062 	if (virtio_has_feature(vdev, VIRTIO_BALLOON_F_FREE_PAGE_HINT)) {
1063 		cancel_work_sync(&vb->report_free_page_work);
1064 		destroy_workqueue(vb->balloon_wq);
1065 	}
1066 
1067 	remove_common(vb);
1068 #ifdef CONFIG_BALLOON_COMPACTION
1069 	if (vb->vb_dev_info.inode)
1070 		iput(vb->vb_dev_info.inode);
1071 
1072 	kern_unmount(balloon_mnt);
1073 #endif
1074 	kfree(vb);
1075 }
1076 
1077 #ifdef CONFIG_PM_SLEEP
1078 static int virtballoon_freeze(struct virtio_device *vdev)
1079 {
1080 	struct virtio_balloon *vb = vdev->priv;
1081 
1082 	/*
1083 	 * The workqueue is already frozen by the PM core before this
1084 	 * function is called.
1085 	 */
1086 	remove_common(vb);
1087 	return 0;
1088 }
1089 
1090 static int virtballoon_restore(struct virtio_device *vdev)
1091 {
1092 	struct virtio_balloon *vb = vdev->priv;
1093 	int ret;
1094 
1095 	ret = init_vqs(vdev->priv);
1096 	if (ret)
1097 		return ret;
1098 
1099 	virtio_device_ready(vdev);
1100 
1101 	if (towards_target(vb))
1102 		virtballoon_changed(vdev);
1103 	update_balloon_size(vb);
1104 	return 0;
1105 }
1106 #endif
1107 
1108 static int virtballoon_validate(struct virtio_device *vdev)
1109 {
1110 	/* Tell the host whether we care about poisoned pages. */
1111 	if (!want_init_on_free() &&
1112 	    (IS_ENABLED(CONFIG_PAGE_POISONING_NO_SANITY) ||
1113 	     !page_poisoning_enabled()))
1114 		__virtio_clear_bit(vdev, VIRTIO_BALLOON_F_PAGE_POISON);
1115 
1116 	__virtio_clear_bit(vdev, VIRTIO_F_IOMMU_PLATFORM);
1117 	return 0;
1118 }
1119 
1120 static unsigned int features[] = {
1121 	VIRTIO_BALLOON_F_MUST_TELL_HOST,
1122 	VIRTIO_BALLOON_F_STATS_VQ,
1123 	VIRTIO_BALLOON_F_DEFLATE_ON_OOM,
1124 	VIRTIO_BALLOON_F_FREE_PAGE_HINT,
1125 	VIRTIO_BALLOON_F_PAGE_POISON,
1126 	VIRTIO_BALLOON_F_REPORTING,
1127 };
1128 
1129 static struct virtio_driver virtio_balloon_driver = {
1130 	.feature_table = features,
1131 	.feature_table_size = ARRAY_SIZE(features),
1132 	.driver.name =	KBUILD_MODNAME,
1133 	.driver.owner =	THIS_MODULE,
1134 	.id_table =	id_table,
1135 	.validate =	virtballoon_validate,
1136 	.probe =	virtballoon_probe,
1137 	.remove =	virtballoon_remove,
1138 	.config_changed = virtballoon_changed,
1139 #ifdef CONFIG_PM_SLEEP
1140 	.freeze	=	virtballoon_freeze,
1141 	.restore =	virtballoon_restore,
1142 #endif
1143 };
1144 
1145 module_virtio_driver(virtio_balloon_driver);
1146 MODULE_DEVICE_TABLE(virtio, id_table);
1147 MODULE_DESCRIPTION("Virtio balloon driver");
1148 MODULE_LICENSE("GPL");
1149